From 44c11faabfe2468eeb5acad6f02cf9bb1766d9bd Mon Sep 17 00:00:00 2001
From: Ma0 <mateuszb@poczta.onet.pl>
Date: Fri, 24 Nov 2017 17:23:59 +0100
Subject: [PATCH 01/51] nits: remove unnecessary parentheses

---
 source/common/lowres.h | 2 +-
 source/encoder/sao.h   | 7 ++-----
 2 files changed, 3 insertions(+), 6 deletions(-)

diff --git a/source/common/lowres.h b/source/common/lowres.h
index 4cb4d00fcb..072739e80e 100644
--- a/source/common/lowres.h
+++ b/source/common/lowres.h
@@ -129,7 +129,7 @@ struct Lowres : public ReferencePlanes
     uint8_t*  intraMode;
     int64_t   satdCost;
     uint16_t* lowresCostForRc;
-    uint16_t(*lowresCosts[X265_BFRAME_MAX + 2][X265_BFRAME_MAX + 2]);
+    uint16_t* lowresCosts[X265_BFRAME_MAX + 2][X265_BFRAME_MAX + 2];
     int32_t*  lowresMvCosts[2][X265_BFRAME_MAX + 1];
     MV*       lowresMvs[2][X265_BFRAME_MAX + 1];
     uint32_t  maxBlocksInRow;
diff --git a/source/encoder/sao.h b/source/encoder/sao.h
index a62305ec35..258b4dd035 100644
--- a/source/encoder/sao.h
+++ b/source/encoder/sao.h
@@ -55,12 +55,9 @@ class SAO
     enum { NUM_EDGETYPE = 5 };
     enum { NUM_PLANE = 3 };
     enum { SAO_DEPTHRATE_SIZE = 4 };
-
     static const uint32_t s_eoTable[NUM_EDGETYPE];
-
-    typedef int32_t (PerClass[MAX_NUM_SAO_TYPE][MAX_NUM_SAO_CLASS]);
-    typedef int32_t (PerPlane[NUM_PLANE][MAX_NUM_SAO_TYPE][MAX_NUM_SAO_CLASS]);
-
+    typedef int32_t PerClass[MAX_NUM_SAO_TYPE][MAX_NUM_SAO_CLASS];
+    typedef int32_t PerPlane[NUM_PLANE][MAX_NUM_SAO_TYPE][MAX_NUM_SAO_CLASS];
 protected:
 
     /* allocated per part */

From d42a261238880fdd1c2f3ddb9b8333ad8900b6c6 Mon Sep 17 00:00:00 2001
From: Vignesh Vijayakumar <vignesh@multicorewareinc.com>
Date: Thu, 2 Nov 2017 09:39:58 +0530
Subject: [PATCH 02/51] x86: Modify asm codes for NASM compatibility

---
 source/common/x86/blockcopy8.asm  |  8 +--
 source/common/x86/intrapred8.asm  |  2 +-
 source/common/x86/ipfilter16.asm  | 30 +++++------
 source/common/x86/ipfilter8.asm   | 86 +++++++++++++++----------------
 source/common/x86/loopfilter.asm  | 48 ++++++++---------
 source/common/x86/mc-a.asm        | 34 ++++++------
 source/common/x86/pixel-util8.asm |  8 +--
 source/common/x86/sad-a.asm       | 18 +++----
 source/common/x86/seaintegral.asm | 24 ++++-----
 9 files changed, 129 insertions(+), 129 deletions(-)

diff --git a/source/common/x86/blockcopy8.asm b/source/common/x86/blockcopy8.asm
index 8c77f293ee..cb30484a27 100644
--- a/source/common/x86/blockcopy8.asm
+++ b/source/common/x86/blockcopy8.asm
@@ -3850,7 +3850,7 @@ cglobal blockcopy_ss_%1x%2, 4, 5, 6
     mov     r4d, %2/4
     add     r1, r1
     add     r3, r3
-.loop
+.loop:
     movu    m0, [r2]
     movu    m1, [r2 + 16]
     movu    m2, [r2 + 32]
@@ -3905,7 +3905,7 @@ cglobal blockcopy_ss_%1x%2, 4, 7, 2
     lea    r5, [3 * r3]
     lea    r6, [3 * r1]
 
-.loop
+.loop:
     movu    m0, [r2]
     movu    xm1, [r2 + 32]
     movu    [r0], m0
@@ -5085,7 +5085,7 @@ cglobal copy_cnt_16, 3,4,6
      pxor        m4, m4
      pxor        m5, m5
 
-.loop
+.loop:
     ; row 0
     movu        m0, [r1]
     movu        m1, [r1 + 16]
@@ -5196,7 +5196,7 @@ cglobal copy_cnt_32, 3,4,6
     pxor        m4, m4
     pxor        m5, m5
 
-.loop
+.loop:
     ; row 0
     movu        m0, [r1]
     movu        m1, [r1 + 16]
diff --git a/source/common/x86/intrapred8.asm b/source/common/x86/intrapred8.asm
index 34db950575..61a856190d 100644
--- a/source/common/x86/intrapred8.asm
+++ b/source/common/x86/intrapred8.asm
@@ -2148,7 +2148,7 @@ cglobal intra_pred_ang4_26, 3,4,4
     paddw       m0, m1
     packuswb    m0, m0
 
-    movd        r2, m0
+    movd        r2d, m0
     mov         [r0], r2b
     shr         r2, 8
     mov         [r0 + r1], r2b
diff --git a/source/common/x86/ipfilter16.asm b/source/common/x86/ipfilter16.asm
index f9e415d820..d44cfd9edb 100644
--- a/source/common/x86/ipfilter16.asm
+++ b/source/common/x86/ipfilter16.asm
@@ -9103,7 +9103,7 @@ cglobal filterPixelToShort_6x%1, 3, 7, 3
     ; load constant
     mova       m2, [pw_2000]
 
-.loop
+.loop:
     movu       m0, [r0]
     movu       m1, [r0 + r1]
     psllw      m0, (14 - BIT_DEPTH)
@@ -9156,7 +9156,7 @@ cglobal filterPixelToShort_8x%1, 3, 7, 2
     ; load constant
     mova       m1, [pw_2000]
 
-.loop
+.loop:
     movu       m0, [r0]
     psllw      m0, (14 - BIT_DEPTH)
     psubw      m0, m1
@@ -9277,7 +9277,7 @@ cglobal filterPixelToShort_16x%1, 3, 7, 3
     ; load constant
     mova       m2, [pw_2000]
 
-.loop
+.loop:
     movu       m0, [r0]
     movu       m1, [r0 + r1]
     psllw      m0, (14 - BIT_DEPTH)
@@ -9351,7 +9351,7 @@ cglobal filterPixelToShort_16x%1, 3, 7, 3
     ; load constant
     mova       m2, [pw_2000]
 
-.loop
+.loop:
     movu       m0, [r0]
     movu       m1, [r0 + r1]
     psllw      m0, (14 - BIT_DEPTH)
@@ -9405,7 +9405,7 @@ cglobal filterPixelToShort_32x%1, 3, 7, 5
     ; load constant
     mova       m4, [pw_2000]
 
-.loop
+.loop:
     movu       m0, [r0]
     movu       m1, [r0 + r1]
     movu       m2, [r0 + r1 * 2]
@@ -9510,7 +9510,7 @@ cglobal filterPixelToShort_32x%1, 3, 7, 3
     ; load constant
     mova       m2, [pw_2000]
 
-.loop
+.loop:
     movu       m0, [r0]
     movu       m1, [r0 + r1]
     psllw      m0, (14 - BIT_DEPTH)
@@ -9583,7 +9583,7 @@ cglobal filterPixelToShort_64x%1, 3, 7, 5
     ; load constant
     mova       m4, [pw_2000]
 
-.loop
+.loop:
     movu       m0, [r0]
     movu       m1, [r0 + r1]
     movu       m2, [r0 + r1 * 2]
@@ -9758,7 +9758,7 @@ cglobal filterPixelToShort_64x%1, 3, 7, 3
     ; load constant
     mova       m2, [pw_2000]
 
-.loop
+.loop:
     movu       m0, [r0]
     movu       m1, [r0 + r1]
     psllw      m0, (14 - BIT_DEPTH)
@@ -9869,7 +9869,7 @@ cglobal filterPixelToShort_24x%1, 3, 7, 5
     ; load constant
     mova       m4, [pw_2000]
 
-.loop
+.loop:
     movu       m0, [r0]
     movu       m1, [r0 + r1]
     movu       m2, [r0 + r1 * 2]
@@ -9952,7 +9952,7 @@ cglobal filterPixelToShort_24x%1, 3, 7, 3
     ; load constant
     mova       m2, [pw_2000]
 
-.loop
+.loop:
     movu       m0, [r0]
     movu       m1, [r0 + 32]
     psllw      m0, (14 - BIT_DEPTH)
@@ -10017,7 +10017,7 @@ cglobal filterPixelToShort_12x%1, 3, 7, 3
     ; load constant
     mova       m2, [pw_2000]
 
-.loop
+.loop:
     movu       m0, [r0]
     movu       m1, [r0 + r1]
     psllw      m0, (14 - BIT_DEPTH)
@@ -10081,7 +10081,7 @@ cglobal filterPixelToShort_48x64, 3, 7, 5
     ; load constant
     mova       m4, [pw_2000]
 
-.loop
+.loop:
     movu       m0, [r0]
     movu       m1, [r0 + r1]
     movu       m2, [r0 + r1 * 2]
@@ -10214,7 +10214,7 @@ cglobal filterPixelToShort_48x64, 3, 7, 4
     ; load constant
     mova       m3, [pw_2000]
 
-.loop
+.loop:
     movu       m0, [r0]
     movu       m1, [r0 + 32]
     movu       m2, [r0 + 64]
@@ -10314,7 +10314,7 @@ cglobal interp_8tap_horiz_ps_4x%1, 6,8,7
 
 .preloop:
     lea                         r6,                [r3 * 3]
-.loop
+.loop:
     ; Row 0
     movu                        xm3,                [r0]                                 ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
     movu                        xm4,                [r0 + 2]                             ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
@@ -10381,7 +10381,7 @@ cglobal interp_8tap_horiz_ps_4x%1, 6,8,7
     packssdw                    xm4,               xm4
 
     movq                        [r2],              xm3                                   ;row 0
-.end
+.end:
     RET
 %endif
 %endmacro
diff --git a/source/common/x86/ipfilter8.asm b/source/common/x86/ipfilter8.asm
index e4f224f9be..36d8986b5a 100644
--- a/source/common/x86/ipfilter8.asm
+++ b/source/common/x86/ipfilter8.asm
@@ -324,7 +324,7 @@ cextern pw_8192
     paddw       m0, m5
     psraw       m0, 6
     packuswb    m0, m0
-    movd        r4, m0
+    movd        r4d, m0
     mov         [dstq], r4w
     shr         r4, 16
     mov         [dstq + dststrideq], r4w
@@ -3471,7 +3471,7 @@ RET
     phaddw      %2, %2
     pmulhrsw    %2, %3
     packuswb    %2, %2
-    movd        r4, %2
+    movd        r4d, %2
     mov         [dstq], r4w
     shr         r4, 16
     mov         [dstq + dststrideq], r4w
@@ -5336,7 +5336,7 @@ cglobal interp_4tap_horiz_ps_64x%1, 4,7,6
     sub                r0 ,         r1
     add                r6d ,        3
 
-.loop
+.loop:
     ; Row 0
     vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
     pshufb            m3,           m1
@@ -5441,7 +5441,7 @@ cglobal interp_8tap_horiz_ps_4x%1, 6,7,6
 
 .preloop:
     lea                         r6,                [r3 * 3]
-.loop
+.loop:
     ; Row 0-1
     vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
     pshufb                      m3,                m1                           ; shuffled based on the col order tab_Lm
@@ -5502,7 +5502,7 @@ cglobal interp_8tap_horiz_ps_4x%1, 6,7,6
     movq                        [r2],              xm3
     movhps                      [r2 + r3],         xm3
     movq                        [r2 + r3 * 2],     xm4
-.end
+.end:
     RET
 %endif
 %endmacro
@@ -5592,7 +5592,7 @@ cglobal interp_8tap_horiz_ps_8x%1, 4,7,6
     paddw                       xm1,        xm2
     psubw                       xm1,        xm0
     movu                        [r2],       xm1                          ;row 0
-.end
+.end:
     RET
 %endif
 %endmacro ; IPFILTER_LUMA_PS_8xN_AVX2
@@ -5634,7 +5634,7 @@ cglobal interp_8tap_horiz_ps_%1x%2, 6, 10, 7
     sub                         r0,                r8                           ; r0(src)-r8
     add                         r9,                7                            ; blkheight += N - 1  (7 - 1 = 6 ; since the last one row not in loop)
 
-.label
+.label:
     ; Row 0
     vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
     pshufb                      m4,                m3,             m6           ; row 0 (col 4 to 7)
@@ -12374,7 +12374,7 @@ cglobal filterPixelToShort_8x%1, 3, 7, 6
     mova        m4, [pb_128]
     mova        m5, [tab_c_64_n64]
 
-.loop
+.loop:
     movh        m0, [r0]
     punpcklbw   m0, m4
     pmaddubsw   m0, m5
@@ -25491,7 +25491,7 @@ cglobal interp_4tap_horiz_ps_32x32, 4,6,8
     sub             r0, r1
     add             r4d, 3
 
-.loop
+.loop:
     ; Row 0
     movu            m2, [r0]
     movu            m3, [r0 + 1]
@@ -25553,7 +25553,7 @@ cglobal interp_4tap_horiz_ps_16x16, 4,7,6
     sub                r0 ,         r1
     add                r6d ,        3
 
-.loop
+.loop:
     ; Row 0
     vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
     pshufb            m3,           m1
@@ -25607,7 +25607,7 @@ cglobal interp_4tap_horiz_ps_%1x%2, 4,7,6
     sub                    r0 ,        r1
     add                    r6d ,       3
 
-.loop
+.loop:
     ; Row 0
     vbroadcasti128         m3,         [r0]
     pshufb                 m3,         m1
@@ -25670,7 +25670,7 @@ cglobal interp_4tap_horiz_ps_%1x%2, 4,7,6
     sub                r0 ,          r1
     add                r6d ,         3
 
-.loop
+.loop:
     ; Row 0
     vbroadcasti128     m3,           [r0]
     pshufb             m3,           m1
@@ -25743,7 +25743,7 @@ cglobal interp_4tap_horiz_ps_4x4, 4,7,5
     je                 .label
     sub                r0 , r1
 
-.label
+.label:
     ; Row 0-1
     movu              xm3,           [r0]
     vinserti128       m3,           m3,      [r0 + r1],     1
@@ -25795,7 +25795,7 @@ cglobal interp_4tap_horiz_ps_4x4, 4,7,5
     movq              [r2+r3],      xm4
     lea               r2,           [r2 + r3 * 2]
     movhps            [r2],         xm3
-.end
+.end:
     RET
 
 cglobal interp_4tap_horiz_ps_4x2, 4,7,5
@@ -25823,7 +25823,7 @@ cglobal interp_4tap_horiz_ps_4x2, 4,7,5
     je                 .label
     sub                r0 , r1
 
-.label
+.label:
     ; Row 0-1
     movu              xm3,           [r0]
     vinserti128       m3,           m3,      [r0 + r1],     1
@@ -25864,7 +25864,7 @@ cglobal interp_4tap_horiz_ps_4x2, 4,7,5
     movq              [r2+r3],      xm4
     lea               r2,           [r2 + r3 * 2]
     movhps            [r2],         xm3
-.end
+.end:
     RET
 
 ;-----------------------------------------------------------------------------------------------------------------------------
@@ -25899,7 +25899,7 @@ cglobal interp_4tap_horiz_ps_%1x%2, 4,7,5
     sub              r0 ,               r1
 
 
-.loop
+.loop:
     sub              r4d,           4
     ; Row 0-1
     movu              xm3,          [r0]
@@ -25955,7 +25955,7 @@ cglobal interp_4tap_horiz_ps_%1x%2, 4,7,5
     movq              [r2+r3],      xm4
     lea               r2,           [r2 + r3 * 2]
     movhps            [r2],         xm3
-.end
+.end:
     RET
 %endmacro
 
@@ -25993,7 +25993,7 @@ cglobal interp_4tap_horiz_ps_8x8, 4,7,6
     sub                r0 ,      r1
     add                r6d ,     1
 
-.loop
+.loop:
      dec               r6d
     ; Row 0
     vbroadcasti128    m3,           [r0]
@@ -26032,7 +26032,7 @@ cglobal interp_4tap_horiz_ps_8x8, 4,7,6
     psubw             m3,           m5
     vpermq            m3,           m3,          11011000b
     movu             [r2],         xm3
-.end
+.end:
     RET
 
 INIT_YMM avx2
@@ -26237,7 +26237,7 @@ cglobal interp_4tap_horiz_pp_%1x%2, 4,6,6
 
     dec                r0
 
-.loop
+.loop:
     sub               r4d,          4
     ; Row 0-1
     movu              xm3,          [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
@@ -26306,9 +26306,9 @@ cglobal interp_8tap_horiz_ps_%1x%2, 4, 7, 8
     sub                         r0,                r6
     add                         r4d,                7
 
-.label
+.label:
     lea                         r6,                 [pw_2000]
-.loop
+.loop:
     ; Row 0
     vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
     pshufb                      m4,                m3,             m6           ; row 0 (col 4 to 7)
@@ -26405,9 +26405,9 @@ cglobal interp_8tap_horiz_ps_48x64, 4, 7, 8
     sub                         r0,                r6                           ; r0(src)-r6
     add                         r4d,                7                            ; blkheight += N - 1  (7 - 1 = 6 ; since the last one row not in loop)
 
-.label
+.label:
     lea                         r6,                [interp8_hps_shuf]
-.loop
+.loop:
     ; Row 0
     vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
     pshufb                      m4,                m3,             m6           ; row 0 (col 4 to 7)
@@ -26736,9 +26736,9 @@ cglobal interp_8tap_horiz_ps_64x%1, 4, 7, 8
     sub                         r0,                r6                           ; r0(src)-r6
     add                         r4d,               7                            ; blkheight += N - 1
 
-.label
+.label:
     lea                         r6,                [pw_2000]
-.loop
+.loop:
     ; Row 0
     vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
     pshufb                      m4,                m3,             m6           ; row 0 (col 4 to 7)
@@ -26880,7 +26880,7 @@ cglobal interp_4tap_horiz_ps_8x%1, 4,7,6
     sub                r0 ,             r1
     inc                r6d
 
-.loop
+.loop:
     ; Row 0
     vbroadcasti128     m3,              [r0]
     pshufb             m3,              m1
@@ -26915,7 +26915,7 @@ cglobal interp_4tap_horiz_ps_8x%1, 4,7,6
     psubw              m3,              m5
     vpermq             m3,              m3,          11011000b
     movu               [r2],            xm3
-.end
+.end:
     RET
 %endmacro
 
@@ -26945,7 +26945,7 @@ cglobal interp_4tap_horiz_ps_2x4, 4, 7, 3
     jz                 .label
     sub                r0,             r1
 
-.label
+.label:
     lea                r6,             [r1 * 3]
     movq               xm1,            [r0]
     movhps             xm1,            [r0 + r1]
@@ -26985,7 +26985,7 @@ cglobal interp_4tap_horiz_ps_2x4, 4, 7, 3
     movd               [r2],           xm1
     pextrd             [r2 + r3],      xm1,         1
     pextrd             [r2 + r3 * 2],  xm1,         2
-.end
+.end:
     RET
 
 INIT_YMM avx2
@@ -27005,7 +27005,7 @@ cglobal interp_4tap_horiz_ps_2x8, 4, 7, 7
     jz                .label
     sub               r0,             r1
 
-.label
+.label:
     mova              m4,            [interp4_hpp_shuf]
     mova              m5,            [pw_1]
     dec               r0
@@ -27062,7 +27062,7 @@ cglobal interp_4tap_horiz_ps_2x8, 4, 7, 7
     movd              [r2],          xm1
     pextrd            [r2 + r3],     xm1,         1
     movd              [r2 + r3 * 2], xm2
-.end
+.end:
     RET
 
 INIT_YMM avx2
@@ -27217,7 +27217,7 @@ cglobal interp_4tap_horiz_ps_6x8, 4,7,6
     sub               r0 ,             r1
     inc               r6d
 
-.loop
+.loop:
     ; Row 0
     vbroadcasti128    m3,              [r0]
     pshufb            m3,              m1
@@ -27254,7 +27254,7 @@ cglobal interp_4tap_horiz_ps_6x8, 4,7,6
     vextracti128      xm4,             m3,          1
     movq              [r2],            xm3
     movd              [r2+8],          xm4
-.end
+.end:
     RET
 
 INIT_YMM avx2
@@ -27285,7 +27285,7 @@ cglobal interp_8tap_horiz_ps_12x16, 6, 7, 8
     lea                         r6,                [r1 * 3]                     ; r6 = (N / 2 - 1) * srcStride
     sub                         r0,                r6                           ; r0(src)-r6
     add                         r4d,                7
-.loop
+.loop:
 
     ; Row 0
 
@@ -27350,9 +27350,9 @@ cglobal interp_8tap_horiz_ps_24x32, 4, 7, 8
     sub                         r0,                r6                           ; r0(src)-r6
     add                         r4d,               7                            ; blkheight += N - 1  (7 - 1 = 6 ; since the last one row not in loop)
 
-.label
+.label:
     lea                         r6,                [interp8_hps_shuf]
-.loop
+.loop:
     ; Row 0
     vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
     pshufb                      m4,                m3,             m6           ; row 0 (col 4 to 7)
@@ -27430,7 +27430,7 @@ cglobal interp_4tap_horiz_ps_24x32, 4,7,6
     sub                r0 ,            r1
     add                r6d ,           3
 
-.loop
+.loop:
     ; Row 0
     vbroadcasti128     m3,             [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
     pshufb             m3,             m1
@@ -27988,7 +27988,7 @@ cglobal interp_4tap_horiz_ps_48x64, 4,7,6
     sub               r0 ,          r1
     add               r6d ,         3
 
-.loop
+.loop:
     ; Row 0
     vbroadcasti128    m3,           [r0]                           ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
     pshufb            m3,           m1
@@ -28067,7 +28067,7 @@ cglobal interp_4tap_horiz_ps_24x64, 4,7,6
     sub                r0 ,            r1
     add                r6d ,           3
 
-.loop
+.loop:
     ; Row 0
     vbroadcasti128     m3,             [r0]                          ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
     pshufb             m3,             m1
@@ -28114,7 +28114,7 @@ cglobal interp_4tap_horiz_ps_2x16, 4, 7, 7
     jz                .label
     sub               r0,             r1
 
-.label
+.label:
     mova              m4,            [interp4_hps_shuf]
     mova              m5,            [pw_1]
     dec               r0
@@ -28209,7 +28209,7 @@ cglobal interp_4tap_horiz_ps_2x16, 4, 7, 7
     movd              [r2],          xm1
     pextrd            [r2 + r3],     xm1,          1
     movd              [r2 + r3 * 2], xm2
-.end
+.end:
     RET
 
 INIT_YMM avx2
diff --git a/source/common/x86/loopfilter.asm b/source/common/x86/loopfilter.asm
index 7e1ed065e8..590652d130 100644
--- a/source/common/x86/loopfilter.asm
+++ b/source/common/x86/loopfilter.asm
@@ -374,7 +374,7 @@ cglobal saoCuOrgE1, 4,5,8
     pxor        m0, m0                      ; m0 = 0
     mova        m6, [pb_2]                  ; m6 = [2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2]
     shr         r4d, 4
-.loop
+.loop:
     movu        m7, [r0]
     movu        m5, [r0 + 16]
     movu        m3, [r0 + r3]
@@ -430,7 +430,7 @@ cglobal saoCuOrgE1, 3, 5, 8, pRec, m_iUpBuff1, m_iOffsetEo, iStride, iLcuWidth
     mova        m6,    [pb_2]                  ; m6 = [2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2]
     mova        m7,    [pb_128]
     shr         r4d,   4
-.loop
+.loop:
     movu        m1,    [r0]                    ; m1 = pRec[x]
     movu        m2,    [r0 + r3]               ; m2 = pRec[x + iStride]
 
@@ -478,7 +478,7 @@ cglobal saoCuOrgE1, 4,5,6
     mova        m4, [pb_2]
     shr         r4d, 4
     mova        m0, [pw_pixel_max]
-.loop
+.loop:
     movu        m5, [r0]
     movu        m3, [r0 + r3]
 
@@ -523,7 +523,7 @@ cglobal saoCuOrgE1, 3, 5, 8, pRec, m_iUpBuff1, m_iOffsetEo, iStride, iLcuWidth
     mova          xm6,    [pb_2]                  ; xm6 = [2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2]
     mova          xm7,    [pb_128]
     shr           r4d,    4
-.loop
+.loop:
     movu          xm1,    [r0]                    ; xm1 = pRec[x]
     movu          xm2,    [r0 + r3]               ; xm2 = pRec[x + iStride]
 
@@ -572,7 +572,7 @@ cglobal saoCuOrgE1_2Rows, 4,7,8
     mov         r5d, r4d
     shr         r4d, 4
     mov         r6, r0
-.loop
+.loop:
     movu        m7, [r0]
     movu        m5, [r0 + 16]
     movu        m3, [r0 + r3]
@@ -674,7 +674,7 @@ cglobal saoCuOrgE1_2Rows, 3, 5, 8, pRec, m_iUpBuff1, m_iOffsetEo, iStride, iLcuW
     pxor        m0,         m0                      ; m0 = 0
     mova        m7,         [pb_128]
     shr         r4d,        4
-.loop
+.loop:
     movu        m1,         [r0]                    ; m1 = pRec[x]
     movu        m2,         [r0 + r3]               ; m2 = pRec[x + iStride]
 
@@ -748,7 +748,7 @@ cglobal saoCuOrgE1_2Rows, 4,5,8
     mova            m4, [pw_pixel_max]
     vbroadcasti128  m6, [r2]                ; m6 = m_iOffsetEo
     shr             r4d, 4
-.loop
+.loop:
     movu            m7, [r0]
     movu            m5, [r0 + r3]
     movu            m1, [r0 + r3 * 2]
@@ -804,7 +804,7 @@ cglobal saoCuOrgE1_2Rows, 3, 5, 7, pRec, m_iUpBuff1, m_iOffsetEo, iStride, iLcuW
     vbroadcasti128  m5,         [pb_128]
     vbroadcasti128  m6,         [r2]                         ; m6 = m_iOffsetEo
     shr             r4d,        4
-.loop
+.loop:
     movu            xm1,        [r0]                         ; m1 = pRec[x]
     movu            xm2,        [r0 + r3]                    ; m2 = pRec[x + iStride]
     vinserti128     m1,         m1,       xm2,            1
@@ -859,7 +859,7 @@ cglobal saoCuOrgE2, 6,6,8
     movh        m6, [r0 + r4 * 2]
     movhps      m6, [r1 + r4]
 
-.loop
+.loop:
     movu        m7, [r0]
     movu        m5, [r0 + 16]
     movu        m3, [r0 + r5 + 2]
@@ -918,7 +918,7 @@ cglobal saoCuOrgE2, 5, 6, 8, rec, bufft, buff1, offsetEo, lcuWidth
     movh        m5,    [r0 + r4]
     movhps      m5,    [r1 + r4]
 
-.loop
+.loop:
     movu        m1,    [r0]                    ; m1 = rec[x]
     movu        m2,    [r0 + r5 + 1]           ; m2 = rec[x + stride + 1]
     pxor        m3,    m1,    m7
@@ -970,7 +970,7 @@ cglobal saoCuOrgE2, 6,6,7
     movhps          xm4, [r1 + r4]
     vbroadcasti128  m5, [r3]
     mova            m6, [pw_pixel_max]
-.loop
+.loop:
     movu            m1, [r0]
     movu            m3, [r0 + r5 + 2]
 
@@ -1061,7 +1061,7 @@ cglobal saoCuOrgE2_32, 6,6,8
     movhps          xm4, [r1 + r4]
     vbroadcasti128  m5, [r3]
 
-.loop
+.loop:
     movu            m1, [r0]
     movu            m7, [r0 + 32]
     movu            m3, [r0 + r5 + 2]
@@ -1567,11 +1567,11 @@ cglobal saoCuOrgB0, 5,7,8
     movu        m4, [r1 + 16]       ; offset[16-31]
     pxor        m7, m7
 
-.loopH
+.loopH:
     mov         r5d, r2d
     xor         r6,  r6
 
-.loopW
+.loopW:
     movu        m2, [r0 + r6]
     movu        m5, [r0 + r6 + 16]
     psrlw       m0, m2, (BIT_DEPTH - 5)
@@ -1617,11 +1617,11 @@ cglobal saoCuOrgB0, 4, 7, 8
     movu        m3, [r1 + 0]      ; offset[0-15]
     movu        m4, [r1 + 16]     ; offset[16-31]
     pxor        m7, m7            ; m7 =[0]
-.loopH
+.loopH:
     mov         r5d, r2d
     xor         r6,  r6
 
-.loopW
+.loopW:
     movu        m2, [r0 + r6]     ; m0 = [rec]
     psrlw       m1, m2, 3
     pand        m1, [pb_31]       ; m1 = [index]
@@ -1670,9 +1670,9 @@ cglobal saoCuOrgB0, 5,7,8
     mov             r6d, r3d
     shr             r3d, 1
 
-.loopH
+.loopH:
     mov             r5d, r2d
-.loopW
+.loopW:
     movu            m2, [r0]
     movu            m5, [r0 + r4]
     psrlw           m0, m2, (BIT_DEPTH - 5)
@@ -1751,9 +1751,9 @@ cglobal saoCuOrgB0, 4, 7, 8
     shr             r2d,        4
     mov             r1d,        r3d
     shr             r3d,        1
-.loopH
+.loopH:
     mov             r5d,        r2d
-.loopW
+.loopW:
     movu            xm2,        [r0]                ; m2 = [rec]
     vinserti128     m2,         m2,  [r0 + r4],  1
     psrlw           m1,         m2,  3
@@ -1789,7 +1789,7 @@ cglobal saoCuOrgB0, 4, 7, 8
     test            r1b,        1
     jz              .end
     mov             r5d,        r2d
-.loopW1
+.loopW1:
     movu            xm2,        [r0]                ; m2 = [rec]
     psrlw           xm1,        xm2, 3
     pand            xm1,        xm7                 ; m1 = [index]
@@ -1811,7 +1811,7 @@ cglobal saoCuOrgB0, 4, 7, 8
     add             r0,         16
     dec             r5d
     jnz             .loopW1
-.end
+.end:
     RET
 %endif
 
@@ -1827,7 +1827,7 @@ cglobal calSign, 4, 7, 5
     add             r3d, 1
     mov             r5, r0
     movu            m4, [r0 + r4]
-.loop
+.loop:
     movu            m1, [r1]        ; m2 = pRec[x]
     movu            m2, [r2]        ; m3 = pTmpU[x]
 
@@ -1921,7 +1921,7 @@ cglobal calSign, 4, 7, 5
     mov             r5, r0
     movu            m4, [r0 + r4]
 
-.loop
+.loop:
     movu            m1, [r1]        ; m2 = pRec[x]
     movu            m2, [r2]        ; m3 = pTmpU[x]
 
diff --git a/source/common/x86/mc-a.asm b/source/common/x86/mc-a.asm
index fa3c8acd6a..55531e770d 100644
--- a/source/common/x86/mc-a.asm
+++ b/source/common/x86/mc-a.asm
@@ -4115,7 +4115,7 @@ cglobal pixel_avg_8x16, 6,10,4
     lea     r7, [r5 * 3]
     lea     r8, [r1 * 3]
     mov     r9d, 4
-.loop
+.loop:
     pixel_avg_W8
     dec     r9d
     jnz     .loop
@@ -4129,7 +4129,7 @@ cglobal pixel_avg_8x32, 6,10,4
     lea     r7, [r5 * 3]
     lea     r8, [r1 * 3]
     mov     r9d, 8
-.loop
+.loop:
     pixel_avg_W8
     dec     r9d
     jnz     .loop
@@ -4697,7 +4697,7 @@ cglobal pixel_avg_12x16, 6,10,4
     lea     r8, [r1 * 3]
     mov     r9d, 4
 
-.loop
+.loop:
     movu    m0, [r2]
     movu    m1, [r4]
     pavgw   m0, m1
@@ -4834,7 +4834,7 @@ cglobal pixel_avg_16x16, 6,10,4
     lea     r7, [r5 * 3]
     lea     r8, [r1 * 3]
     mov     r9d, 4
-.loop
+.loop:
     pixel_avg_H16
     dec r9d
     jnz .loop
@@ -4848,7 +4848,7 @@ cglobal pixel_avg_16x32, 6,10,4
     lea     r7, [r5 * 3]
     lea     r8, [r1 * 3]
     mov     r9d, 4
-.loop
+.loop:
     pixel_avg_H16
     pixel_avg_H16
     dec r9d
@@ -4863,7 +4863,7 @@ cglobal pixel_avg_16x64, 6,10,4
     lea     r7, [r5 * 3]
     lea     r8, [r1 * 3]
     mov     r9d, 4
-.loop
+.loop:
     pixel_avg_H16
     pixel_avg_H16
     pixel_avg_H16
@@ -4887,7 +4887,7 @@ cglobal pixel_avg_24x32, 6,10,4
     lea     r8, [r1 * 3]
     mov     r9d, 8
 
-.loop
+.loop:
     movu    m0, [r2]
     movu    m1, [r4]
     pavgw   m0, m1
@@ -4987,7 +4987,7 @@ cglobal pixel_avg_32x8, 6,10,4
     lea     r7, [r5 * 3]
     lea     r8, [r1 * 3]
     mov     r9d, 2
-.loop
+.loop:
     pixel_avg_W32
     dec     r9d
     jnz     .loop
@@ -5001,7 +5001,7 @@ cglobal pixel_avg_32x16, 6,10,4
     lea     r7, [r5 * 3]
     lea     r8, [r1 * 3]
     mov     r9d, 4
-.loop
+.loop:
     pixel_avg_W32
     dec     r9d
     jnz     .loop
@@ -5015,7 +5015,7 @@ cglobal pixel_avg_32x24, 6,10,4
     lea     r7, [r5 * 3]
     lea     r8, [r1 * 3]
     mov     r9d, 6
-.loop
+.loop:
     pixel_avg_W32
     dec     r9d
     jnz     .loop
@@ -5029,7 +5029,7 @@ cglobal pixel_avg_32x32, 6,10,4
     lea     r7, [r5 * 3]
     lea     r8, [r1 * 3]
     mov     r9d, 8
-.loop
+.loop:
     pixel_avg_W32
     dec     r9d
     jnz     .loop
@@ -5043,7 +5043,7 @@ cglobal pixel_avg_32x64, 6,10,4
     lea     r7, [r5 * 3]
     lea     r8, [r1 * 3]
     mov     r9d, 16
-.loop
+.loop:
     pixel_avg_W32
     dec     r9d
     jnz     .loop
@@ -5141,7 +5141,7 @@ cglobal pixel_avg_64x16, 6,10,4
     lea     r7, [r5 * 3]
     lea     r8, [r1 * 3]
     mov     r9d, 4
-.loop
+.loop:
     pixel_avg_W64
     dec     r9d
     jnz     .loop
@@ -5155,7 +5155,7 @@ cglobal pixel_avg_64x32, 6,10,4
     lea     r7, [r5 * 3]
     lea     r8, [r1 * 3]
     mov     r9d, 8
-.loop
+.loop:
     pixel_avg_W64
     dec     r9d
     jnz     .loop
@@ -5169,7 +5169,7 @@ cglobal pixel_avg_64x48, 6,10,4
     lea     r7, [r5 * 3]
     lea     r8, [r1 * 3]
     mov     r9d, 12
-.loop
+.loop:
     pixel_avg_W64
     dec     r9d
     jnz     .loop
@@ -5183,7 +5183,7 @@ cglobal pixel_avg_64x64, 6,10,4
     lea     r7, [r5 * 3]
     lea     r8, [r1 * 3]
     mov     r9d, 16
-.loop
+.loop:
     pixel_avg_W64
     dec     r9d
     jnz     .loop
@@ -5204,7 +5204,7 @@ cglobal pixel_avg_48x64, 6,10,4
     lea     r8, [r1 * 3]
     mov     r9d, 16
 
-.loop
+.loop:
     movu    m0, [r2]
     movu    m1, [r4]
     pavgw   m0, m1
diff --git a/source/common/x86/pixel-util8.asm b/source/common/x86/pixel-util8.asm
index 11c2500abf..50264c45c0 100644
--- a/source/common/x86/pixel-util8.asm
+++ b/source/common/x86/pixel-util8.asm
@@ -1785,7 +1785,7 @@ cglobal weight_sp, 6,7,9
     movu                      [r1], xm7
     je                        .nextH
 
-.width6
+.width6:
     cmp                       r6d, 6
     jl                        .width4
     movq                      [r1], xm7
@@ -4937,7 +4937,7 @@ cglobal pixel_sub_ps_16x%2, 6, 10, 4, dest, deststride, src0, src1, srcstride0,
     lea         r9,     [r4 * 3]
     lea         r8,     [r5 * 3]
 
-.loop
+.loop:
     pmovzxbw    m0,     [r2]
     pmovzxbw    m1,     [r3]
     pmovzxbw    m2,     [r2 + r4]
@@ -5150,7 +5150,7 @@ cglobal pixel_sub_ps_32x%1, 6, 10, 4, dest, deststride, src0, src1, srcstride0,
     lea         r7,     [r4 * 3]
     lea         r8,     [r5 * 3]
 
-.loop
+.loop:
     movu        m0,     [r2]
     movu        m1,     [r2 + 32]
     movu        m2,     [r3]
@@ -5557,7 +5557,7 @@ cglobal pixel_sub_ps_64x64, 6, 10, 8, dest, deststride, src0, src1, srcstride0,
     lea         r7,     [r4 * 3]
     lea         r8,     [r5 * 3]
 
-.loop
+.loop:
     movu        m0,     [r2]
     movu        m1,     [r2 + 32]
     movu        m2,     [r2 + 64]
diff --git a/source/common/x86/sad-a.asm b/source/common/x86/sad-a.asm
index 5bd80487aa..62ea494647 100644
--- a/source/common/x86/sad-a.asm
+++ b/source/common/x86/sad-a.asm
@@ -5631,7 +5631,7 @@ cglobal pixel_sad_32x16, 4,5,6
     xorps           m5, m5
     mov             r4d, 4
 
-.loop
+.loop:
     movu           m1, [r0]               ; row 0 of pix0
     movu           m2, [r2]               ; row 0 of pix1
     movu           m3, [r0 + r1]          ; row 1 of pix0
@@ -5676,7 +5676,7 @@ cglobal pixel_sad_32x24, 4,7,6
     mov             r4d, 6
     lea             r5, [r1 * 3]
     lea             r6, [r3 * 3]
-.loop
+.loop:
     movu           m1, [r0]               ; row 0 of pix0
     movu           m2, [r2]               ; row 0 of pix1
     movu           m3, [r0 + r1]          ; row 1 of pix0
@@ -5718,7 +5718,7 @@ cglobal pixel_sad_32x32, 4,7,5
     lea             r5, [r1 * 3]
     lea             r6, [r3 * 3]
 
-.loop
+.loop:
     movu           m1, [r0]               ; row 0 of pix0
     movu           m2, [r2]               ; row 0 of pix1
     movu           m3, [r0 + r1]          ; row 1 of pix0
@@ -5759,7 +5759,7 @@ cglobal pixel_sad_32x64, 4,7,5
     lea             r5, [r1 * 3]
     lea             r6, [r3 * 3]
 
-.loop
+.loop:
     movu           m1, [r0]               ; row 0 of pix0
     movu           m2, [r2]               ; row 0 of pix1
     movu           m3, [r0 + r1]          ; row 1 of pix0
@@ -5822,7 +5822,7 @@ cglobal pixel_sad_48x64, 4,7,7
     mov             r4d, 64/4
     lea             r5, [r1 * 3]
     lea             r6, [r3 * 3]
-.loop
+.loop:
     movu           m1, [r0]               ; row 0 of pix0
     movu           m2, [r2]               ; row 0 of pix1
     movu           m3, [r0 + r1]          ; row 1 of pix0
@@ -5873,7 +5873,7 @@ cglobal pixel_sad_64x16, 4,5,6
     xorps           m0, m0
     xorps           m5, m5
     mov             r4d, 4
-.loop
+.loop:
     movu           m1, [r0]               ; first 32 of row 0 of pix0
     movu           m2, [r2]               ; first 32 of row 0 of pix1
     movu           m3, [r0 + 32]          ; second 32 of row 0 of pix0
@@ -5936,7 +5936,7 @@ cglobal pixel_sad_64x32, 4,5,6
     xorps           m0, m0
     xorps           m5, m5
     mov             r4d, 16
-.loop
+.loop:
     movu           m1, [r0]               ; first 32 of row 0 of pix0
     movu           m2, [r2]               ; first 32 of row 0 of pix1
     movu           m3, [r0 + 32]          ; second 32 of row 0 of pix0
@@ -5978,7 +5978,7 @@ cglobal pixel_sad_64x48, 4,7,6
     mov             r4d, 12
     lea             r5, [r1 * 3]
     lea             r6, [r3 * 3]
-.loop
+.loop:
     movu           m1, [r0]               ; first 32 of row 0 of pix0
     movu           m2, [r2]               ; first 32 of row 0 of pix1
     movu           m3, [r0 + 32]          ; second 32 of row 0 of pix0
@@ -6040,7 +6040,7 @@ cglobal pixel_sad_64x64, 4,7,6
     mov             r4d, 8
     lea             r5, [r1 * 3]
     lea             r6, [r3 * 3]
-.loop
+.loop:
     movu           m1, [r0]               ; first 32 of row 0 of pix0
     movu           m2, [r2]               ; first 32 of row 0 of pix1
     movu           m3, [r0 + 32]          ; second 32 of row 0 of pix0
diff --git a/source/common/x86/seaintegral.asm b/source/common/x86/seaintegral.asm
index cf79ca4478..ea362efd0a 100644
--- a/source/common/x86/seaintegral.asm
+++ b/source/common/x86/seaintegral.asm
@@ -36,7 +36,7 @@ cglobal integral4v, 2, 3, 2
     mov r2, r1
     shl r2, 4
 
-.loop
+.loop:
     movu    m0, [r0]
     movu    m1, [r0 + r2]
     psubd   m1, m0
@@ -54,7 +54,7 @@ cglobal integral8v, 2, 3, 2
     mov r2, r1
     shl r2, 5
 
-.loop
+.loop:
     movu    m0, [r0]
     movu    m1, [r0 + r2]
     psubd   m1, m0
@@ -75,7 +75,7 @@ cglobal integral12v, 2, 4, 2
     shl r3, 4
     add r2, r3
 
-.loop
+.loop:
     movu    m0, [r0]
     movu    m1, [r0 + r2]
     psubd   m1, m0
@@ -93,7 +93,7 @@ cglobal integral16v, 2, 3, 2
     mov r2, r1
     shl r2, 6
 
-.loop
+.loop:
     movu    m0, [r0]
     movu    m1, [r0 + r2]
     psubd   m1, m0
@@ -114,7 +114,7 @@ cglobal integral24v, 2, 4, 2
     shl r3, 5
     add r2, r3
 
-.loop
+.loop:
     movu    m0, [r0]
     movu    m1, [r0 + r2]
     psubd   m1, m0
@@ -132,7 +132,7 @@ cglobal integral32v, 2, 3, 2
     mov r2, r1
     shl r2, 7
 
-.loop
+.loop:
     movu    m0, [r0]
     movu    m1, [r0 + r2]
     psubd   m1, m0
@@ -264,7 +264,7 @@ cglobal integral4h, 3, 5, 3
     movu           [r0 + r3], xm0
     jmp            .end
 
-.end
+.end:
     RET
 %endif
 
@@ -379,7 +379,7 @@ cglobal integral8h, 3, 5, 3
     movu           [r0 + r3], m0
     jmp             .end
 
-.end
+.end:
     RET
 %endif
 
@@ -577,7 +577,7 @@ cglobal integral12h, 3, 5, 3
     movu           [r0 + r3], xm0
     jmp            .end
 
-.end
+.end:
     RET
 %endif
 
@@ -740,7 +740,7 @@ cglobal integral16h, 3, 5, 3
     movu           [r0 + r3], m0
     jmp             .end
 
-.end
+.end:
     RET
 %endif
 
@@ -883,7 +883,7 @@ cglobal integral24h, 3, 5, 3
     movu           [r0 + r3], m0
     jmp             .end
 
-.end
+.end:
     RET
 
 %macro INTEGRAL_THIRTYTWO_HORIZONTAL_16 0
@@ -1058,5 +1058,5 @@ cglobal integral32h, 3, 5, 3
     movu           [r0 + r3], m0
     jmp             .end
 
-.end
+.end:
     RET

From 329cdee9e8c31364f008f4303ec3b08a568fd4a5 Mon Sep 17 00:00:00 2001
From: Vignesh Vijayakumar <vignesh@multicorewareinc.com>
Date: Thu, 2 Nov 2017 09:40:41 +0530
Subject: [PATCH 03/51] x86: Change assembler from YASM to NASM

Supports NASM versions 2.13 and greater
---
 source/CMakeLists.txt                         | 20 ++++----
 ...n.cmake => CMakeASM_NASMInformation.cmake} | 30 +++++------
 ...e => CMakeDetermineASM_NASMCompiler.cmake} |  6 +--
 ....cmake => CMakeTestASM_NASMCompiler.cmake} |  2 +-
 source/cmake/FindNasm.cmake                   | 25 +++++++++
 source/cmake/FindYasm.cmake                   | 25 ---------
 source/common/CMakeLists.txt                  |  6 +--
 source/common/x86/x86inc.asm                  | 51 +++++++++++--------
 source/test/CMakeLists.txt                    | 18 +++----
 source/test/checkasm-a.asm                    | 50 +++++++++---------
 10 files changed, 124 insertions(+), 109 deletions(-)
 rename source/cmake/{CMakeASM_YASMInformation.cmake => CMakeASM_NASMInformation.cmake} (64%)
 rename source/cmake/{CMakeDetermineASM_YASMCompiler.cmake => CMakeDetermineASM_NASMCompiler.cmake} (55%)
 rename source/cmake/{CMakeTestASM_YASMCompiler.cmake => CMakeTestASM_NASMCompiler.cmake} (65%)
 create mode 100644 source/cmake/FindNasm.cmake
 delete mode 100644 source/cmake/FindYasm.cmake

diff --git a/source/CMakeLists.txt b/source/CMakeLists.txt
index dfcef20f49..ca8e4d24dd 100644
--- a/source/CMakeLists.txt
+++ b/source/CMakeLists.txt
@@ -323,15 +323,15 @@ if(GCC)
     execute_process(COMMAND ${CMAKE_CXX_COMPILER} -dumpversion OUTPUT_VARIABLE CC_VERSION)
 endif(GCC)
 
-find_package(Yasm)
+find_package(Nasm)
 if(ARM OR CROSS_COMPILE_ARM)
     option(ENABLE_ASSEMBLY "Enable use of assembly coded primitives" ON)
-elseif(YASM_FOUND AND X86)
-    if (YASM_VERSION_STRING VERSION_LESS "1.2.0")
-        message(STATUS "Yasm version ${YASM_VERSION_STRING} is too old. 1.2.0 or later required")
+elseif(NASM_FOUND AND X86)
+    if (NASM_VERSION_STRING VERSION_LESS "2.13.0")
+        message(STATUS "Nasm version ${NASM_VERSION_STRING} is too old. 2.13.0 or later required")
         option(ENABLE_ASSEMBLY "Enable use of assembly coded primitives" OFF)
     else()
-        message(STATUS "Found Yasm ${YASM_VERSION_STRING} to build assembly primitives")
+        message(STATUS "Found Nasm ${NASM_VERSION_STRING} to build assembly primitives")
         option(ENABLE_ASSEMBLY "Enable use of assembly coded primitives" ON)
     endif()
 else()
@@ -517,18 +517,18 @@ if((MSVC_IDE OR XCODE OR GCC) AND ENABLE_ASSEMBLY)
             list(APPEND ASM_OBJS ${ASM}.${SUFFIX})
             add_custom_command(
                 OUTPUT ${ASM}.${SUFFIX}
-                COMMAND ${YASM_EXECUTABLE} ARGS ${YASM_FLAGS} ${ASM_SRC} -o ${ASM}.${SUFFIX}
+                COMMAND ${NASM_EXECUTABLE} ARGS ${NASM_FLAGS} ${ASM_SRC} -o ${ASM}.${SUFFIX}
                 DEPENDS ${ASM_SRC})
         endforeach()
     endif()
 endif()
 source_group(ASM FILES ${ASM_SRCS})
 if(ENABLE_HDR10_PLUS)
-    add_library(x265-static STATIC $<TARGET_OBJECTS:encoder> $<TARGET_OBJECTS:common> $<TARGET_OBJECTS:dynamicHDR10> ${ASM_OBJS} ${ASM_SRCS})
+    add_library(x265-static STATIC $<TARGET_OBJECTS:encoder> $<TARGET_OBJECTS:common> $<TARGET_OBJECTS:dynamicHDR10> ${ASM_OBJS})
     add_library(hdr10plus-static STATIC $<TARGET_OBJECTS:dynamicHDR10>)
     set_target_properties(hdr10plus-static PROPERTIES OUTPUT_NAME hdr10plus)
 else()
-    add_library(x265-static STATIC $<TARGET_OBJECTS:encoder> $<TARGET_OBJECTS:common> ${ASM_OBJS} ${ASM_SRCS})
+    add_library(x265-static STATIC $<TARGET_OBJECTS:encoder> $<TARGET_OBJECTS:common> ${ASM_OBJS})
 endif()
 if(NOT MSVC)
     set_target_properties(x265-static PROPERTIES OUTPUT_NAME x265)
@@ -686,11 +686,11 @@ if(ENABLE_CLI)
         if(ENABLE_HDR10_PLUS)
         add_executable(cli ../COPYING ${InputFiles} ${OutputFiles} ${GETOPT}
                         x265.cpp x265.h x265cli.h
-                        $<TARGET_OBJECTS:encoder> $<TARGET_OBJECTS:common> $<TARGET_OBJECTS:dynamicHDR10> ${ASM_OBJS} ${ASM_SRCS})
+                        $<TARGET_OBJECTS:encoder> $<TARGET_OBJECTS:common> $<TARGET_OBJECTS:dynamicHDR10> ${ASM_OBJS})
         else()
             add_executable(cli ../COPYING ${InputFiles} ${OutputFiles} ${GETOPT}
                         x265.cpp x265.h x265cli.h
-                        $<TARGET_OBJECTS:encoder> $<TARGET_OBJECTS:common> ${ASM_OBJS} ${ASM_SRCS})
+                        $<TARGET_OBJECTS:encoder> $<TARGET_OBJECTS:common> ${ASM_OBJS})
         endif()
     else()
         add_executable(cli ../COPYING ${InputFiles} ${OutputFiles} ${GETOPT} ${X265_RC_FILE}
diff --git a/source/cmake/CMakeASM_YASMInformation.cmake b/source/cmake/CMakeASM_NASMInformation.cmake
similarity index 64%
rename from source/cmake/CMakeASM_YASMInformation.cmake
rename to source/cmake/CMakeASM_NASMInformation.cmake
index 051a0cd3df..45c0ad1672 100644
--- a/source/cmake/CMakeASM_YASMInformation.cmake
+++ b/source/cmake/CMakeASM_NASMInformation.cmake
@@ -1,22 +1,22 @@
-set(ASM_DIALECT "_YASM")
+set(ASM_DIALECT "_NASM")
 set(CMAKE_ASM${ASM_DIALECT}_SOURCE_FILE_EXTENSIONS asm)
 
 if(X64)
-    list(APPEND ASM_FLAGS -DARCH_X86_64=1)
+    list(APPEND ASM_FLAGS -DARCH_X86_64=1 -I ${CMAKE_CURRENT_SOURCE_DIR}/../common/x86/)
     if(ENABLE_PIC)
         list(APPEND ASM_FLAGS -DPIC)
     endif()
     if(APPLE)
-        set(ARGS -f macho64 -m amd64 -DPREFIX)
+        set(ARGS -f macho64 -DPREFIX)
     elseif(UNIX AND NOT CYGWIN)
-        set(ARGS -f elf64 -m amd64)
+        set(ARGS -f elf64)
     else()
-        set(ARGS -f win64 -m amd64)
+        set(ARGS -f win64)
     endif()
 else()
-    list(APPEND ASM_FLAGS -DARCH_X86_64=0)
+    list(APPEND ASM_FLAGS -DARCH_X86_64=0 -I ${CMAKE_CURRENT_SOURCE_DIR}/../common/x86/)
     if(APPLE)
-        set(ARGS -f macho -DPREFIX)
+        set(ARGS -f macho32 -DPREFIX)
     elseif(UNIX AND NOT CYGWIN)
         set(ARGS -f elf32)
     else()
@@ -40,25 +40,25 @@ else()
     list(APPEND ASM_FLAGS -DHIGH_BIT_DEPTH=0 -DBIT_DEPTH=8 -DX265_NS=${X265_NS})
 endif()
 
-list(APPEND ASM_FLAGS "${CMAKE_ASM_YASM_FLAGS}")
+list(APPEND ASM_FLAGS "${CMAKE_ASM_NASM_FLAGS}")
 
 if(CMAKE_BUILD_TYPE MATCHES Release)
-    list(APPEND ASM_FLAGS "${CMAKE_ASM_YASM_FLAGS_RELEASE}")
+    list(APPEND ASM_FLAGS "${CMAKE_ASM_NASM_FLAGS_RELEASE}")
 elseif(CMAKE_BUILD_TYPE MATCHES Debug)
-    list(APPEND ASM_FLAGS "${CMAKE_ASM_YASM_FLAGS_DEBUG}")
+    list(APPEND ASM_FLAGS "${CMAKE_ASM_NASM_FLAGS_DEBUG}")
 elseif(CMAKE_BUILD_TYPE MATCHES MinSizeRel)
-    list(APPEND ASM_FLAGS "${CMAKE_ASM_YASM_FLAGS_MINSIZEREL}")
+    list(APPEND ASM_FLAGS "${CMAKE_ASM_NASM_FLAGS_MINSIZEREL}")
 elseif(CMAKE_BUILD_TYPE MATCHES RelWithDebInfo)
-    list(APPEND ASM_FLAGS "${CMAKE_ASM_YASM_FLAGS_RELWITHDEBINFO}")
+    list(APPEND ASM_FLAGS "${CMAKE_ASM_NASM_FLAGS_RELWITHDEBINFO}")
 endif()
 
-set(YASM_FLAGS ${ARGS} ${ASM_FLAGS} PARENT_SCOPE)
-string(REPLACE ";" " " CMAKE_ASM_YASM_COMPILER_ARG1 "${ARGS}")
+set(NASM_FLAGS ${ARGS} ${ASM_FLAGS} PARENT_SCOPE)
+string(REPLACE ";" " " CMAKE_ASM_NASM_COMPILER_ARG1 "${ARGS}")
 
 # This section exists to override the one in CMakeASMInformation.cmake
 # (the default Information file). This removes the <FLAGS>
 # thing so that your C compiler flags that have been set via
-# set_target_properties don't get passed to yasm and confuse it.
+# set_target_properties don't get passed to nasm and confuse it.
 if(NOT CMAKE_ASM${ASM_DIALECT}_COMPILE_OBJECT)
     string(REPLACE ";" " " STR_ASM_FLAGS "${ASM_FLAGS}")
     set(CMAKE_ASM${ASM_DIALECT}_COMPILE_OBJECT "<CMAKE_ASM${ASM_DIALECT}_COMPILER> ${STR_ASM_FLAGS} -o <OBJECT> <SOURCE>")
diff --git a/source/cmake/CMakeDetermineASM_YASMCompiler.cmake b/source/cmake/CMakeDetermineASM_NASMCompiler.cmake
similarity index 55%
rename from source/cmake/CMakeDetermineASM_YASMCompiler.cmake
rename to source/cmake/CMakeDetermineASM_NASMCompiler.cmake
index a902ef8cbd..6ee17e4db9 100644
--- a/source/cmake/CMakeDetermineASM_YASMCompiler.cmake
+++ b/source/cmake/CMakeDetermineASM_NASMCompiler.cmake
@@ -1,5 +1,5 @@
-set(ASM_DIALECT "_YASM")
-set(CMAKE_ASM${ASM_DIALECT}_COMPILER ${YASM_EXECUTABLE})
-set(CMAKE_ASM${ASM_DIALECT}_COMPILER_INIT ${_CMAKE_TOOLCHAIN_PREFIX}yasm)
+set(ASM_DIALECT "_NASM")
+set(CMAKE_ASM${ASM_DIALECT}_COMPILER ${NASM_EXECUTABLE})
+set(CMAKE_ASM${ASM_DIALECT}_COMPILER_INIT ${_CMAKE_TOOLCHAIN_PREFIX}nasm)
 include(CMakeDetermineASMCompiler)
 set(ASM_DIALECT)
diff --git a/source/cmake/CMakeTestASM_YASMCompiler.cmake b/source/cmake/CMakeTestASM_NASMCompiler.cmake
similarity index 65%
rename from source/cmake/CMakeTestASM_YASMCompiler.cmake
rename to source/cmake/CMakeTestASM_NASMCompiler.cmake
index c668668a03..f3346c1eec 100644
--- a/source/cmake/CMakeTestASM_YASMCompiler.cmake
+++ b/source/cmake/CMakeTestASM_NASMCompiler.cmake
@@ -1,3 +1,3 @@
-set(ASM_DIALECT "_YASM")
+set(ASM_DIALECT "_NASM")
 include(CMakeTestASMCompiler)
 set(ASM_DIALECT)
diff --git a/source/cmake/FindNasm.cmake b/source/cmake/FindNasm.cmake
new file mode 100644
index 0000000000..ff7eac6227
--- /dev/null
+++ b/source/cmake/FindNasm.cmake
@@ -0,0 +1,25 @@
+include(FindPackageHandleStandardArgs)
+
+# Simple path search with YASM_ROOT environment variable override
+find_program(NASM_EXECUTABLE 
+ NAMES nasm nasm-2.13.0-win32 nasm-2.13.0-win64 nasm nasm-2.13.0-win32 nasm-2.13.0-win64
+ HINTS $ENV{NASM_ROOT} ${NASM_ROOT}
+ PATH_SUFFIXES bin
+)
+
+if(NASM_EXECUTABLE)
+        execute_process(COMMAND ${NASM_EXECUTABLE} -version
+            OUTPUT_VARIABLE nasm_version
+            ERROR_QUIET
+            OUTPUT_STRIP_TRAILING_WHITESPACE
+            )
+    if(nasm_version MATCHES "^NASM version ([0-9\\.]*)")
+        set(NASM_VERSION_STRING "${CMAKE_MATCH_1}")
+    endif()
+    unset(nasm_version)
+endif()
+
+# Provide standardized success/failure messages
+find_package_handle_standard_args(nasm
+    REQUIRED_VARS NASM_EXECUTABLE
+    VERSION_VAR NASM_VERSION_STRING)
diff --git a/source/cmake/FindYasm.cmake b/source/cmake/FindYasm.cmake
deleted file mode 100644
index d93cdece6e..0000000000
--- a/source/cmake/FindYasm.cmake
+++ /dev/null
@@ -1,25 +0,0 @@
-include(FindPackageHandleStandardArgs)
-
-# Simple path search with YASM_ROOT environment variable override
-find_program(YASM_EXECUTABLE 
- NAMES yasm yasm-1.2.0-win32 yasm-1.2.0-win64 yasm yasm-1.3.0-win32 yasm-1.3.0-win64
- HINTS $ENV{YASM_ROOT} ${YASM_ROOT}
- PATH_SUFFIXES bin
-)
-
-if(YASM_EXECUTABLE)
-    execute_process(COMMAND ${YASM_EXECUTABLE} --version
-        OUTPUT_VARIABLE yasm_version
-        ERROR_QUIET
-        OUTPUT_STRIP_TRAILING_WHITESPACE
-        )
-    if(yasm_version MATCHES "^yasm ([0-9\\.]*)")
-        set(YASM_VERSION_STRING "${CMAKE_MATCH_1}")
-    endif()
-    unset(yasm_version)
-endif()
-
-# Provide standardized success/failure messages
-find_package_handle_standard_args(yasm
-    REQUIRED_VARS YASM_EXECUTABLE
-    VERSION_VAR YASM_VERSION_STRING)
diff --git a/source/common/CMakeLists.txt b/source/common/CMakeLists.txt
index 7f2e01b846..65624069c4 100644
--- a/source/common/CMakeLists.txt
+++ b/source/common/CMakeLists.txt
@@ -72,12 +72,12 @@ if(ENABLE_ASSEMBLY AND X86)
     endif()
 
     if(MSVC_IDE OR XCODE)
-        # MSVC requires custom build rules in the main cmake script for yasm
-        set(MSVC_ASMS "${A_SRCS}" CACHE INTERNAL "yasm sources")
+        # MSVC requires custom build rules in the main cmake script for nasm
+        set(MSVC_ASMS "${A_SRCS}" CACHE INTERNAL "nasm sources")
         set(A_SRCS)
     endif()
 
-    enable_language(ASM_YASM)
+    enable_language(ASM_NASM)
 
     foreach(SRC ${A_SRCS} ${C_SRCS})
         set(ASM_PRIMITIVES ${ASM_PRIMITIVES} x86/${SRC})
diff --git a/source/common/x86/x86inc.asm b/source/common/x86/x86inc.asm
index a7d96dd82e..fa5ffcd28b 100644
--- a/source/common/x86/x86inc.asm
+++ b/source/common/x86/x86inc.asm
@@ -66,6 +66,15 @@
     %endif
 %endif
 
+%define FORMAT_ELF 0
+%ifidn __OUTPUT_FORMAT__,elf
+    %define FORMAT_ELF 1
+%elifidn __OUTPUT_FORMAT__,elf32
+    %define FORMAT_ELF 1
+%elifidn __OUTPUT_FORMAT__,elf64
+    %define FORMAT_ELF 1
+%endif
+
 %ifdef PREFIX
     %define mangle(x) _ %+ x
 %else
@@ -88,6 +97,10 @@
     default rel
 %endif
 
+%ifdef __NASM_VER__
+    %use smartalign
+%endif
+
 ; Macros to eliminate most code duplication between x86_32 and x86_64:
 ; Currently this works only for leaf functions which load all their arguments
 ; into registers at the start, and make no other use of the stack. Luckily that
@@ -685,7 +698,7 @@ BRANCH_INSTR jz, je, jnz, jne, jl, jle, jnl, jnle, jg, jge, jng, jnge, ja, jae,
         CAT_XDEFINE cglobaled_, %2, 1
     %endif
     %xdefine current_function %2
-    %ifidn __OUTPUT_FORMAT__,elf
+    %if FORMAT_ELF
         global %2:function %%VISIBILITY
     %else
         global %2
@@ -711,14 +724,16 @@ BRANCH_INSTR jz, je, jnz, jne, jl, jle, jnl, jnle, jg, jge, jng, jnge, ja, jae,
 
 ; like cextern, but without the prefix
 %macro cextern_naked 1
-    %xdefine %1 mangle(%1)
+    %ifdef PREFIX
+        %xdefine %1 mangle(%1)
+    %endif
     CAT_XDEFINE cglobaled_, %1, 1
     extern %1
 %endmacro
 
 %macro const 1-2+
     %xdefine %1 mangle(private_prefix %+ _ %+ %1)
-    %ifidn __OUTPUT_FORMAT__,elf
+    %if FORMAT_ELF
         global %1:data hidden
     %else
         global %1
@@ -727,9 +742,8 @@ BRANCH_INSTR jz, je, jnz, jne, jl, jle, jnl, jnle, jg, jge, jng, jnge, ja, jae,
     %1: %2
 %endmacro
 
-; This is needed for ELF, otherwise the GNU linker assumes the stack is
-; executable by default.
-%ifidn __OUTPUT_FORMAT__,elf
+; This is needed for ELF, otherwise the GNU linker assumes the stack is executable by default.
+%if FORMAT_ELF
     [SECTION .note.GNU-stack noalloc noexec nowrite progbits]
 %endif
 
@@ -801,9 +815,17 @@ BRANCH_INSTR jz, je, jnz, jne, jl, jle, jnl, jnle, jg, jge, jng, jnge, ja, jae,
     %endif
 
     %if ARCH_X86_64 || cpuflag(sse2)
-        CPU amdnop
+        %ifdef __NASM_VER__
+            ALIGNMODE p6
+        %else
+            CPU amdnop
+        %endif
     %else
-        CPU basicnop
+        %ifdef __NASM_VER__
+            ALIGNMODE nop
+        %else
+            CPU basicnop
+        %endif
     %endif
 %endmacro
 
@@ -1467,7 +1489,7 @@ FMA_INSTR pmadcswd, pmaddwd, paddd
                 v%5%6 %1, %2, %3, %4
             %elifidn %1, %2
                 ; If %3 or %4 is a memory operand it needs to be encoded as the last operand.
-                %ifid %3
+                %ifnum sizeof%3
                     v%{5}213%6 %2, %3, %4
                 %else
                     v%{5}132%6 %2, %4, %3
@@ -1491,14 +1513,3 @@ FMA4_INSTR fmsub,    pd, ps, sd, ss
 FMA4_INSTR fmsubadd, pd, ps
 FMA4_INSTR fnmadd,   pd, ps, sd, ss
 FMA4_INSTR fnmsub,   pd, ps, sd, ss
-
-; workaround: vpbroadcastq is broken in x86_32 due to a yasm bug (fixed in 1.3.0)
-%if __YASM_VERSION_ID__ < 0x01030000 && ARCH_X86_64 == 0
-    %macro vpbroadcastq 2
-        %if sizeof%1 == 16
-            movddup %1, %2
-        %else
-            vbroadcastsd %1, %2
-        %endif
-    %endmacro
-%endif
diff --git a/source/test/CMakeLists.txt b/source/test/CMakeLists.txt
index f2f0f47492..260195f53a 100644
--- a/source/test/CMakeLists.txt
+++ b/source/test/CMakeLists.txt
@@ -7,37 +7,37 @@ endif()
 
 # add X86 assembly files
 if(X86)
-enable_language(ASM_YASM)
+enable_language(ASM_NASM)
 
 if(MSVC_IDE)
-    set(YASM_SRC checkasm-a.obj)
+    set(NASM_SRC checkasm-a.obj)
     add_custom_command(
         OUTPUT checkasm-a.obj
-        COMMAND ${YASM_EXECUTABLE}
-        ARGS ${YASM_FLAGS} ${CMAKE_CURRENT_SOURCE_DIR}/checkasm-a.asm -o checkasm-a.obj
+        COMMAND ${NASM_EXECUTABLE}
+        ARGS ${NASM_FLAGS} ${CMAKE_CURRENT_SOURCE_DIR}/checkasm-a.asm -o checkasm-a.obj
         DEPENDS checkasm-a.asm)
 else()
-    set(YASM_SRC checkasm-a.asm)
+    set(NASM_SRC checkasm-a.asm)
 endif()
 endif(X86)
 
 # add ARM assembly files
 if(ARM OR CROSS_COMPILE_ARM)
     enable_language(ASM)
-    set(YASM_SRC checkasm-arm.S)
+    set(NASM_SRC checkasm-arm.S)
     add_custom_command(
         OUTPUT checkasm-arm.obj
         COMMAND ${CMAKE_CXX_COMPILER}
-        ARGS ${YASM_FLAGS} ${CMAKE_CURRENT_SOURCE_DIR}/checkasm-arm.S -o checkasm-arm.obj
+        ARGS ${NASM_FLAGS} ${CMAKE_CURRENT_SOURCE_DIR}/checkasm-arm.S -o checkasm-arm.obj
         DEPENDS checkasm-arm.S)
 endif(ARM OR CROSS_COMPILE_ARM)
 
 # add PowerPC assembly files
 if(POWER)
-    set(YASM_SRC)
+    set(NASM_SRC)
 endif(POWER)
 
-add_executable(TestBench ${YASM_SRC}
+add_executable(TestBench ${NASM_SRC}
     testbench.cpp testharness.h
     pixelharness.cpp pixelharness.h
     mbdstharness.cpp mbdstharness.h
diff --git a/source/test/checkasm-a.asm b/source/test/checkasm-a.asm
index d5b60fdf46..9e57a057cc 100644
--- a/source/test/checkasm-a.asm
+++ b/source/test/checkasm-a.asm
@@ -26,7 +26,7 @@
 ;* For more information, contact us at license @ x265.com.
 ;*****************************************************************************
 
-%include "../common/x86/x86inc.asm"
+%include "x86inc.asm"
 
 SECTION_RODATA
 
@@ -35,24 +35,24 @@ error_message: db "failed to preserve register", 0
 %if ARCH_X86_64
 ; just random numbers to reduce the chance of incidental match
 ALIGN 16
-x6:  ddq 0x79445c159ce790641a1b2550a612b48c
-x7:  ddq 0x86b2536fcd8cf6362eed899d5a28ddcd
-x8:  ddq 0x3f2bf84fc0fcca4eb0856806085e7943
-x9:  ddq 0xd229e1f5b281303facbd382dcf5b8de2
-x10: ddq 0xab63e2e11fa38ed971aeaff20b095fd9
-x11: ddq 0x77d410d5c42c882d89b0c0765892729a
-x12: ddq 0x24b3c1d2a024048bc45ea11a955d8dd5
-x13: ddq 0xdd7b8919edd427862e8ec680de14b47c
-x14: ddq 0x11e53e2b2ac655ef135ce6888fa02cbf
-x15: ddq 0x6de8f4c914c334d5011ff554472a7a10
-n7:   dq 0x21f86d66c8ca00ce
-n8:   dq 0x75b6ba21077c48ad
-n9:   dq 0xed56bb2dcb3c7736
-n10:  dq 0x8bda43d3fd1a7e06
-n11:  dq 0xb64a9c9e5d318408
-n12:  dq 0xdf9a54b303f1d3a3
-n13:  dq 0x4a75479abd64e097
-n14:  dq 0x249214109d5d1c88
+x6:  dq 0x1a1b2550a612b48c,0x79445c159ce79064
+x7:  dq 0x2eed899d5a28ddcd,0x86b2536fcd8cf636
+x8:  dq 0xb0856806085e7943,0x3f2bf84fc0fcca4e
+x9:  dq 0xacbd382dcf5b8de2,0xd229e1f5b281303f
+x10: dq 0x71aeaff20b095fd9,0xab63e2e11fa38ed9
+x11: dq 0x89b0c0765892729a,0x77d410d5c42c882d
+x12: dq 0xc45ea11a955d8dd5,0x24b3c1d2a024048b
+x13: dq 0x2e8ec680de14b47c,0xdd7b8919edd42786
+x14: dq 0x135ce6888fa02cbf,0x11e53e2b2ac655ef
+x15: dq 0x011ff554472a7a10,0x6de8f4c914c334d5
+n7:  dq 0x21f86d66c8ca00ce
+n8:  dq 0x75b6ba21077c48ad
+n9:  dq 0xed56bb2dcb3c7736
+n10: dq 0x8bda43d3fd1a7e06
+n11: dq 0xb64a9c9e5d318408
+n12: dq 0xdf9a54b303f1d3a3
+n13: dq 0x4a75479abd64e097
+n14: dq 0x249214109d5d1c88
 %endif
 
 SECTION .text
@@ -70,14 +70,14 @@ cextern_naked puts
 ;-----------------------------------------------------------------------------
 cglobal checkasm_stack_clobber, 1,2
     ; Clobber the stack with junk below the stack pointer
-    %define size (max_args+6)*8
-    SUB  rsp, size
-    mov   r1, size-8
+    %define argsize (max_args+6)*8
+    SUB  rsp, argsize
+    mov   r1, argsize-8
 .loop:
     mov [rsp+r1], r0
     sub   r1, 8
     jge .loop
-    ADD  rsp, size
+    ADD  rsp, argsize
     RET
 
 %if WIN64
@@ -156,7 +156,11 @@ cglobal checkasm_call, 2,15,16,max_args*8+8
     mov  r9, rax
     mov r10, rdx
     lea  r0, [error_message]
+%if FORMAT_ELF
+    call puts wrt ..plt
+%else
     call puts
+%endif
     mov  r1, [rsp+max_args*8]
     mov  dword [r1], 0
     mov  rdx, r10

From 514e3f824b45469713cf434485d8a552f4ff6eae Mon Sep 17 00:00:00 2001
From: Pradeep Ramachandran <pradeep@multicorewareinc.com>
Date: Thu, 30 Nov 2017 10:06:49 +0530
Subject: [PATCH 04/51] doc: Update build/readme.txt for detailed instructions
 with nasm

---
 build/README.txt | 21 ++++++++++++++++++++-
 1 file changed, 20 insertions(+), 1 deletion(-)

diff --git a/build/README.txt b/build/README.txt
index d13188461b..a555cc1482 100644
--- a/build/README.txt
+++ b/build/README.txt
@@ -9,7 +9,8 @@ Note: MSVC12 requires cmake 2.8.11 or later
 
 = Optional Prerequisites =
 
-1. Yasm 1.2.0 or later, to compile assembly primitives (performance)
+1. To compile assembly primitives (performance) 
+   a) If you are using release 2.6 or older, download and install Yasm 1.2.0 or later, 
 
    For Windows, download the latest yasm executable
    http://yasm.tortall.net/Download.html and copy the EXE into
@@ -33,6 +34,24 @@ Note: MSVC12 requires cmake 2.8.11 or later
    If cpu capabilities line says 'none!', then the encoder was built
    without yasm.
 
+   b) If you are building from the default branch after release 2.6, download and install nasm 2.13 or newer
+   
+   For windows and linux, you can download the nasm installer from http://www.nasm.us/pub/nasm/releasebuilds/?C=M;O=D.
+   Make sure that it is in your PATH environment variable (%PATH% in windows, and $PATH in linux) so that cmake
+   can find it.
+
+   Once NASM is properly installed, run cmake to regenerate projects. If you
+   do not see the below line in the cmake output, NASM is not in the PATH.
+
+   -- Found Nasm 2.13 to build assembly primitives
+
+   Now build the encoder and run x265 -V:
+
+   x265 [info]: using cpu capabilities: MMX, SSE2, ...
+
+   If cpu capabilities line says 'none!', then the encoder was built
+   without nasm and will be considerably slower for performance.
+
 2. VisualLeakDetector (Windows Only)
 
    Download from https://vld.codeplex.com/releases and install. May need

From a38fbf933cd0072de0871b7eb604e698c4a6be17 Mon Sep 17 00:00:00 2001
From: Ma0 <mateuszb@poczta.onet.pl>
Date: Thu, 30 Nov 2017 18:16:40 +0100
Subject: [PATCH 05/51] fix ambiguous call to overloaded function 'sqrt' (in
 old MSVC)

---
 source/encoder/encoder.cpp | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/source/encoder/encoder.cpp b/source/encoder/encoder.cpp
index a8a9eb6db4..a08f7ae87f 100644
--- a/source/encoder/encoder.cpp
+++ b/source/encoder/encoder.cpp
@@ -575,7 +575,7 @@ int Encoder::setAnalysisDataAfterZScan(x265_analysis_data *analysis_data, Frame*
                             {
                                 int mv_x = ((analysis_inter_data *)curFrame->m_analysisData.interData)->mv[k][(mbIndex * 16) + cuOffset].x;
                                 int mv_y = ((analysis_inter_data *)curFrame->m_analysisData.interData)->mv[k][(mbIndex * 16) + cuOffset].y;
-                                double mv = sqrt(mv_x*mv_x + mv_y*mv_y);
+                                double mv = sqrt((double)(mv_x*mv_x + mv_y*mv_y));
                                 if (numPU == PU_2Nx2N && ((srcInterData)->depth[cuPos + cuOffset] == (m_param->maxCUSize >> 5)) && mv <= MVTHRESHOLD)
                                     memset(&curFrame->m_analysisData.modeFlag[k][cuPos + cuOffset], 1, bytes);
                             }
@@ -658,7 +658,7 @@ int Encoder::setAnalysisData(x265_analysis_data *analysis_data, int poc, uint32_
                                     {
                                         int mv_x = ((analysis_inter_data *)curFrame->m_analysisData.interData)->mv[i][count + pu].x;
                                         int mv_y = ((analysis_inter_data *)curFrame->m_analysisData.interData)->mv[i][count + pu].y;
-                                        double mv = sqrt(mv_x*mv_x + mv_y*mv_y);
+                                        double mv = sqrt((double)(mv_x*mv_x + mv_y*mv_y));
                                         if (numPU == PU_2Nx2N && m_param->num4x4Partitions <= 16 && mv <= MVTHRESHOLD)
                                             memset(&curFrame->m_analysisData.modeFlag[i][count + pu], 1, bytes);
                                     }

From d35350163e5d4260cb72cba98afc09bf1c3c316c Mon Sep 17 00:00:00 2001
From: Aruna Matheswaran <aruna@multicorewareinc.com>
Date: Tue, 21 Nov 2017 10:48:24 +0530
Subject: [PATCH 06/51] Extend gop boundary by doing gop lookahead

---
 doc/reST/cli.rst                 | 10 ++++++
 source/CMakeLists.txt            |  2 +-
 source/common/param.cpp          |  5 +++
 source/encoder/slicetype.cpp     | 52 ++++++++++++++++++++++++--------
 source/encoder/slicetype.h       |  1 +
 source/test/regression-tests.txt |  2 +-
 source/x265.h                    |  7 +++--
 source/x265cli.h                 |  2 ++
 8 files changed, 64 insertions(+), 17 deletions(-)

diff --git a/doc/reST/cli.rst b/doc/reST/cli.rst
index 87dc6b68b8..cd6bb41cff 100644
--- a/doc/reST/cli.rst
+++ b/doc/reST/cli.rst
@@ -1373,6 +1373,16 @@ Slice decision options
 	Default 20
 
 	**Range of values:** Between the maximum consecutive bframe count (:option:`--bframes`) and 250
+.. option:: --gop-lookahead <integer>
+
+        Number of frames for GOP boundary decision lookahead. If a scenecut frame is found
+        within this from the gop boundary set by `--keyint`, the GOP will be extented until such a point,
+        otherwise the GOP will be terminated as set by `--keyint`. Default 0.
+
+        **Range of values:** Between 0 and (`--rc-lookahead` - mini-GOP length)
+
+        It is recommended to have `--gop-lookahaed` less than `--min-keyint` as scenecuts beyond
+        `--min-keyint` are already being coded as keyframes.
 
 .. option:: --lookahead-slices <0..16>
 
diff --git a/source/CMakeLists.txt b/source/CMakeLists.txt
index ca8e4d24dd..b6a9a18299 100644
--- a/source/CMakeLists.txt
+++ b/source/CMakeLists.txt
@@ -29,7 +29,7 @@ option(NATIVE_BUILD "Target the build CPU" OFF)
 option(STATIC_LINK_CRT "Statically link C runtime for release builds" OFF)
 mark_as_advanced(FPROFILE_USE FPROFILE_GENERATE NATIVE_BUILD)
 # X265_BUILD must be incremented each time the public API is changed
-set(X265_BUILD 146)
+set(X265_BUILD 147)
 configure_file("${PROJECT_SOURCE_DIR}/x265.def.in"
                "${PROJECT_BINARY_DIR}/x265.def")
 configure_file("${PROJECT_SOURCE_DIR}/x265_config.h.in"
diff --git a/source/common/param.cpp b/source/common/param.cpp
index aad9e27e94..4b037d50f1 100644
--- a/source/common/param.cpp
+++ b/source/common/param.cpp
@@ -144,6 +144,7 @@ void x265_param_default(x265_param* param)
     /* Coding Structure */
     param->keyframeMin = 0;
     param->keyframeMax = 250;
+    param->gopLookahead = 0;
     param->bOpenGOP = 1;
     param->bframes = 4;
     param->lookaheadDepth = 20;
@@ -1004,6 +1005,7 @@ int x265_param_parse(x265_param* p, const char* name, const char* value)
                 bError = true;
             }
          }
+        OPT("gop-lookahead") p->gopLookahead = atoi(value);
         else
             return X265_PARAM_BAD_NAME;
     }
@@ -1314,6 +1316,8 @@ int x265_check_params(x265_param* param)
           "Valid penalty for 32x32 intra TU in non-I slices. 0:disabled 1:RD-penalty 2:maximum");
     CHECK(param->keyframeMax < -1,
           "Invalid max IDR period in frames. value should be greater than -1");
+    CHECK(param->gopLookahead < -1,
+          "GOP lookahead must be greater than -1");
     CHECK(param->decodedPictureHashSEI < 0 || param->decodedPictureHashSEI > 3,
           "Invalid hash option. Decoded Picture Hash SEI 0: disabled, 1: MD5, 2: CRC, 3: Checksum");
     CHECK(param->rc.vbvBufferSize < 0,
@@ -1561,6 +1565,7 @@ char *x265_param2string(x265_param* p, int padx, int pady)
     BOOL(p->bOpenGOP, "open-gop");
     s += sprintf(s, " min-keyint=%d", p->keyframeMin);
     s += sprintf(s, " keyint=%d", p->keyframeMax);
+    s += sprintf(s, " gop-lookahead=%d", p->gopLookahead);
     s += sprintf(s, " bframes=%d", p->bframes);
     s += sprintf(s, " b-adapt=%d", p->bFrameAdaptive);
     BOOL(p->bBPyramid, "b-pyramid");
diff --git a/source/encoder/slicetype.cpp b/source/encoder/slicetype.cpp
index fe2e51aab9..7129db2d2b 100644
--- a/source/encoder/slicetype.cpp
+++ b/source/encoder/slicetype.cpp
@@ -589,7 +589,7 @@ Lookahead::Lookahead(x265_param *param, ThreadPool* pool)
     m_outputSignalRequired = false;
     m_isActive = true;
     m_inputCount = 0;
-
+    m_extendGopBoundary = false;
     m_8x8Height = ((m_param->sourceHeight / 2) + X265_LOWRES_CU_SIZE - 1) >> X265_LOWRES_CU_BITS;
     m_8x8Width = ((m_param->sourceWidth / 2) + X265_LOWRES_CU_SIZE - 1) >> X265_LOWRES_CU_BITS;
     m_cuCount = m_8x8Width * m_8x8Height;
@@ -646,7 +646,11 @@ Lookahead::Lookahead(x265_param *param, ThreadPool* pool)
         m_numRowsPerSlice = m_8x8Height;
         m_numCoopSlices = 1;
     }
-
+    if (param->gopLookahead && (param->gopLookahead > (param->lookaheadDepth - param->bframes - 2)))
+    {
+        param->gopLookahead = X265_MAX(0, param->lookaheadDepth - param->bframes - 2);
+        x265_log(param, X265_LOG_WARNING, "Gop-lookahead cannot be greater than (rc-lookahead - length of the mini-gop); Clipping gop-lookahead to %d\n", param->gopLookahead);
+    }
 #if DETAILED_CU_STATS
     m_slicetypeDecideElapsedTime = 0;
     m_preLookaheadElapsedTime = 0;
@@ -1086,7 +1090,8 @@ void Lookahead::slicetypeDecide()
             x265_log(m_param, X265_LOG_WARNING, "B-ref at frame %d incompatible with B-pyramid and %d reference frames\n",
                      frm.sliceType, m_param->maxNumReferences);
         }
-        if ((!m_param->bIntraRefresh || frm.frameNum == 0) && frm.frameNum - m_lastKeyframe >= m_param->keyframeMax)
+        if ((!m_param->bIntraRefresh || frm.frameNum == 0) && frm.frameNum - m_lastKeyframe >= m_param->keyframeMax &&
+            (!m_extendGopBoundary || frm.frameNum - m_lastKeyframe >= m_param->keyframeMax + m_param->gopLookahead))
         {
             if (frm.sliceType == X265_TYPE_AUTO || frm.sliceType == X265_TYPE_I)
                 frm.sliceType = m_param->bOpenGOP && m_lastKeyframe >= 0 ? X265_TYPE_I : X265_TYPE_IDR;
@@ -1377,12 +1382,14 @@ void Lookahead::slicetypeAnalyse(Lowres **frames, bool bKeyframe)
             cuTree(frames, 0, bKeyframe);
         return;
     }
-
     frames[framecnt + 1] = NULL;
+    int keyFrameLimit = m_param->keyframeMax + m_lastKeyframe - frames[0]->frameNum - 1;
+    if (m_param->gopLookahead && keyFrameLimit <= m_param->bframes + 1)
+        keyintLimit = keyFrameLimit + m_param->gopLookahead;
+    else
+        keyintLimit = keyFrameLimit;
 
-    keyintLimit = m_param->keyframeMax - frames[0]->frameNum + m_lastKeyframe - 1;
     origNumFrames = numFrames = m_param->bIntraRefresh ? framecnt : X265_MIN(framecnt, keyintLimit);
-
     if (bIsVbvLookahead)
         numFrames = framecnt;
     else if (m_param->bOpenGOP && numFrames < framecnt)
@@ -1472,7 +1479,26 @@ void Lookahead::slicetypeAnalyse(Lowres **frames, bool bKeyframe)
         frames[1]->sliceType = X265_TYPE_I;
         return;
     }
-
+    if (m_param->gopLookahead && (keyFrameLimit >= 0) && (keyFrameLimit <= m_param->bframes + 1))
+    {
+        bool sceneTransition = m_isSceneTransition;
+        m_extendGopBoundary = false;
+        for (int i = m_param->bframes + 1; i < origNumFrames; i += m_param->bframes + 1)
+        {
+            scenecut(frames, i, i + 1, true, origNumFrames);
+            for (int j = i + 1; j <= X265_MIN(i + m_param->bframes + 1, origNumFrames); j++)
+            {
+                if (frames[j]->bScenecut && scenecutInternal(frames, j - 1, j, true) )
+                {
+                    m_extendGopBoundary = true;
+                    break;
+                }
+            }
+            if (m_extendGopBoundary)
+                break;
+        }
+        m_isSceneTransition = sceneTransition;
+    }
     if (m_param->bframes)
     {
         if (m_param->bFrameAdaptive == X265_B_ADAPT_TRELLIS)
@@ -1578,6 +1604,8 @@ void Lookahead::slicetypeAnalyse(Lowres **frames, bool bKeyframe)
 
     if (m_param->rc.cuTree)
         cuTree(frames, X265_MIN(numFrames, m_param->keyframeMax), bKeyframe);
+    if (m_param->gopLookahead && (keyFrameLimit >= 0) && (keyFrameLimit <= m_param->bframes + 1) && !m_extendGopBoundary)
+        keyintLimit = keyFrameLimit;
 
     if (!m_param->bIntraRefresh)
         for (int j = keyintLimit + 1; j <= numFrames; j += m_param->keyframeMax)
@@ -1588,8 +1616,8 @@ void Lookahead::slicetypeAnalyse(Lowres **frames, bool bKeyframe)
 
     if (bIsVbvLookahead)
         vbvLookahead(frames, numFrames, bKeyframe);
+    int maxp1 = X265_MIN(m_param->bframes + 1, origNumFrames);
 
-     int maxp1 = X265_MIN(m_param->bframes + 1, origNumFrames);
     /* Restore frame types for all frames that haven't actually been decided yet. */
     for (int j = resetStart; j <= numFrames; j++)
     {
@@ -1613,8 +1641,8 @@ bool Lookahead::scenecut(Lowres **frames, int p0, int p1, bool bRealScenecut, in
         bool fluctuate = false;
         bool noScenecuts = false;
         int64_t avgSatdCost = 0;
-        if (frames[0]->costEst[1][0] > -1)
-            avgSatdCost = frames[0]->costEst[1][0];
+        if (frames[p0]->costEst[p1 - p0][0] > -1)
+            avgSatdCost = frames[p0]->costEst[p1 - p0][0];
         int cnt = 1;
         /* Where A and B are scenes: AAAAAABBBAAAAAA
          * If BBB is shorter than (maxp1-p0), it is detected as a flash
@@ -1700,12 +1728,10 @@ bool Lookahead::scenecutInternal(Lowres **frames, int p0, int p1, bool bRealScen
 
     CostEstimateGroup estGroup(*this, frames);
     estGroup.singleCost(p0, p1, p1);
-
     int64_t icost = frame->costEst[0][0];
     int64_t pcost = frame->costEst[p1 - p0][0];
-    int gopSize = frame->frameNum - m_lastKeyframe;
+    int gopSize = (frame->frameNum - m_lastKeyframe) % m_param->keyframeMax;
     float threshMax = (float)(m_param->scenecutThreshold / 100.0);
-
     /* magic numbers pulled out of thin air */
     float threshMin = (float)(threshMax * 0.25);
     double bias = m_param->scenecutBias;
diff --git a/source/encoder/slicetype.h b/source/encoder/slicetype.h
index a247a8c2c9..f85d4aa128 100644
--- a/source/encoder/slicetype.h
+++ b/source/encoder/slicetype.h
@@ -132,6 +132,7 @@ class Lookahead : public JobProvider
     bool          m_filled;
     bool          m_isSceneTransition;
     int           m_numPools;
+    bool          m_extendGopBoundary;
     Lookahead(x265_param *param, ThreadPool *pool);
 #if DETAILED_CU_STATS
     int64_t       m_slicetypeDecideElapsedTime;
diff --git a/source/test/regression-tests.txt b/source/test/regression-tests.txt
index bdd6935f7c..0dd5dd9323 100644
--- a/source/test/regression-tests.txt
+++ b/source/test/regression-tests.txt
@@ -150,7 +150,7 @@ Kimono1_1920x1080_24_400.yuv,--preset ultrafast --slices 1 --weightp --tu-intra-
 Kimono1_1920x1080_24_400.yuv,--preset medium --rdoq-level 0 --limit-refs 3 --slices 2
 Kimono1_1920x1080_24_400.yuv,--preset veryslow --crf 4 --cu-lossless --slices 2 --limit-refs 3 --limit-modes
 Kimono1_1920x1080_24_400.yuv,--preset placebo --ctu 32 --max-tu-size 8 --limit-tu 2
-
+big_buck_bunny_360p24.y4m, --keyint 60 --min-keyint 40 --gop-lookahead 20
 # Main12 intraCost overflow bug test
 720p50_parkrun_ter.y4m,--preset medium
 
diff --git a/source/x265.h b/source/x265.h
index 05cffd8ab9..8f6d5ff90a 100644
--- a/source/x265.h
+++ b/source/x265.h
@@ -1532,11 +1532,14 @@ typedef struct x265_param
 
     /* Reuse MV information obtained through API */
     int       bMVType;
-
     /* Allow the encoder to have a copy of the planes of x265_picture in Frame */
     int       bCopyPicToFrame;
-} x265_param;
 
+    /*Number of frames for GOP boundary decision lookahead.If a scenecut frame is found
+    * within this from the gop boundary set by keyint, the GOP will be extented until such a point,
+    * otherwise the GOP will be terminated as set by keyint*/
+    int       gopLookahead;
+} x265_param;
 /* x265_param_alloc:
  *  Allocates an x265_param instance. The returned param structure is not
  *  special in any way, but using this method together with x265_param_free()
diff --git a/source/x265cli.h b/source/x265cli.h
index bd5f7c5326..de5afcc1f6 100644
--- a/source/x265cli.h
+++ b/source/x265cli.h
@@ -119,6 +119,7 @@ static const struct option long_options[] =
     { "open-gop",             no_argument, NULL, 0 },
     { "keyint",         required_argument, NULL, 'I' },
     { "min-keyint",     required_argument, NULL, 'i' },
+    { "gop-lookahead",  required_argument, NULL, 0 },
     { "scenecut",       required_argument, NULL, 0 },
     { "no-scenecut",          no_argument, NULL, 0 },
     { "scenecut-bias",  required_argument, NULL, 0 },
@@ -418,6 +419,7 @@ static void showHelp(x265_param *param)
     H0("   --[no-]open-gop               Enable open-GOP, allows I slices to be non-IDR. Default %s\n", OPT(param->bOpenGOP));
     H0("-I/--keyint <integer>            Max IDR period in frames. -1 for infinite-gop. Default %d\n", param->keyframeMax);
     H0("-i/--min-keyint <integer>        Scenecuts closer together than this are coded as I, not IDR. Default: auto\n");
+    H0("   --gop-lookahead <integer>     Extends gop boundary if a scenecut is found within this from keyint boundary. Default 0\n");
     H0("   --no-scenecut                 Disable adaptive I-frame decision\n");
     H0("   --scenecut <integer>          How aggressively to insert extra I-frames. Default %d\n", param->scenecutThreshold);
     H1("   --scenecut-bias <0..100.0>    Bias for scenecut detection. Default %.2f\n", param->scenecutBias);

From 96449185802fc6c21a3e23a366b802e806b796f7 Mon Sep 17 00:00:00 2001
From: Pradeep Ramachandran <pradeep@multicorewareinc.com>
Date: Wed, 29 Nov 2017 08:52:18 +0530
Subject: [PATCH 07/51] Added tag 2.6 for changeset 0e9ea76945c8

---
 .hgtags | 1 +
 1 file changed, 1 insertion(+)

diff --git a/.hgtags b/.hgtags
index 363e6eebe4..ece3fd0d03 100644
--- a/.hgtags
+++ b/.hgtags
@@ -24,3 +24,4 @@ be14a7e9755e54f0fd34911c72bdfa66981220bc 2.2
 3037c1448549ca920967831482c653e5892fa8ed 2.3
 e7a4dd48293b7956d4a20df257d23904cc78e376 2.4
 64b2d0bf45a52511e57a6b7299160b961ca3d51c 2.5
+0e9ea76945c89962cd46cee6537586e2054b2935 2.6

From 371f3ade561880042526e4f954933017fce972a3 Mon Sep 17 00:00:00 2001
From: Santhoshini Sekar <santhoshini@multicorewareinc.com>
Date: Thu, 30 Nov 2017 17:36:32 +0530
Subject: [PATCH 08/51] deblock: set reference frame to NULL if refIdx < 0.
 Dereferencing refFrameList when refIdx < 0 is not valid

---
 source/common/deblock.cpp | 11 ++++-------
 1 file changed, 4 insertions(+), 7 deletions(-)

diff --git a/source/common/deblock.cpp b/source/common/deblock.cpp
index bbdbf7bceb..13ac66f458 100644
--- a/source/common/deblock.cpp
+++ b/source/common/deblock.cpp
@@ -207,21 +207,18 @@ uint8_t Deblock::getBoundaryStrength(const CUData* cuQ, int32_t dir, uint32_t pa
     static const MV zeroMv(0, 0);
     const Slice* const sliceQ = cuQ->m_slice;
     const Slice* const sliceP = cuP->m_slice;
-
-    const Frame* refP0 = sliceP->m_refFrameList[0][cuP->m_refIdx[0][partP]];
-    const Frame* refQ0 = sliceQ->m_refFrameList[0][cuQ->m_refIdx[0][partQ]];
+    const Frame* refP0 = (cuP->m_refIdx[0][partP] >= 0) ? sliceP->m_refFrameList[0][cuP->m_refIdx[0][partP]] : NULL;
+    const Frame* refQ0 = (cuQ->m_refIdx[0][partQ] >= 0) ? sliceQ->m_refFrameList[0][cuQ->m_refIdx[0][partQ]] : NULL;
     const MV& mvP0 = refP0 ? cuP->m_mv[0][partP] : zeroMv;
     const MV& mvQ0 = refQ0 ? cuQ->m_mv[0][partQ] : zeroMv;
-
     if (sliceQ->isInterP() && sliceP->isInterP())
     {
         return ((refP0 != refQ0) ||
                 (abs(mvQ0.x - mvP0.x) >= 4) || (abs(mvQ0.y - mvP0.y) >= 4)) ? 1 : 0;
     }
-
     // (sliceQ->isInterB() || sliceP->isInterB())
-    const Frame* refP1 = sliceP->m_refFrameList[1][cuP->m_refIdx[1][partP]];
-    const Frame* refQ1 = sliceQ->m_refFrameList[1][cuQ->m_refIdx[1][partQ]];
+    const Frame* refP1 = (cuP->m_refIdx[1][partP] >= 0) ? sliceP->m_refFrameList[1][cuP->m_refIdx[1][partP]] : NULL;
+    const Frame* refQ1 = (cuQ->m_refIdx[1][partQ] >= 0) ? sliceQ->m_refFrameList[1][cuQ->m_refIdx[1][partQ]] : NULL;
     const MV& mvP1 = refP1 ? cuP->m_mv[1][partP] : zeroMv;
     const MV& mvQ1 = refQ1 ? cuQ->m_mv[1][partQ] : zeroMv;
 

From 4ba86180bdfc760331dc5bbc279f5fc387c397c8 Mon Sep 17 00:00:00 2001
From: Mahesh Pittala <mahesh@multicorewareinc.com>
Date: Thu, 30 Nov 2017 19:09:45 +0530
Subject: [PATCH 09/51] Install symbol files in MinGW

In debug and RelWithDebInfo configuration, .pdb files are installed.
---
 source/CMakeLists.txt | 17 +++++++++++------
 1 file changed, 11 insertions(+), 6 deletions(-)

diff --git a/source/CMakeLists.txt b/source/CMakeLists.txt
index dfcef20f49..6ab8a32cc4 100644
--- a/source/CMakeLists.txt
+++ b/source/CMakeLists.txt
@@ -546,14 +546,19 @@ if(ENABLE_HDR10_PLUS)
         ARCHIVE DESTINATION ${LIB_INSTALL_DIR})
 endif()
 install(FILES x265.h "${PROJECT_BINARY_DIR}/x265_config.h" DESTINATION include)
-
 if(WIN32)
-    install(FILES "${PROJECT_BINARY_DIR}/Debug/x265.pdb" DESTINATION ${BIN_INSTALL_DIR} CONFIGURATIONS Debug)
-    install(FILES "${PROJECT_BINARY_DIR}/RelWithDebInfo/x265.pdb" DESTINATION ${BIN_INSTALL_DIR} CONFIGURATIONS RelWithDebInfo)
-    install(FILES "${PROJECT_BINARY_DIR}/Debug/libx265.pdb" DESTINATION ${BIN_INSTALL_DIR} CONFIGURATIONS Debug OPTIONAL NAMELINK_ONLY)
-    install(FILES "${PROJECT_BINARY_DIR}/RelWithDebInfo/libx265.pdb" DESTINATION ${BIN_INSTALL_DIR} CONFIGURATIONS RelWithDebInfo OPTIONAL NAMELINK_ONLY)
+    if(MSVC_IDE)
+        install(FILES "${PROJECT_BINARY_DIR}/Debug/x265.pdb" DESTINATION ${BIN_INSTALL_DIR} CONFIGURATIONS Debug)
+        install(FILES "${PROJECT_BINARY_DIR}/RelWithDebInfo/x265.pdb" DESTINATION ${BIN_INSTALL_DIR} CONFIGURATIONS RelWithDebInfo)
+        install(FILES "${PROJECT_BINARY_DIR}/Debug/libx265.pdb" DESTINATION ${BIN_INSTALL_DIR} CONFIGURATIONS Debug OPTIONAL NAMELINK_ONLY)
+        install(FILES "${PROJECT_BINARY_DIR}/RelWithDebInfo/libx265.pdb" DESTINATION ${BIN_INSTALL_DIR} CONFIGURATIONS RelWithDebInfo OPTIONAL NAMELINK_ONLY)
+    else()
+        install(FILES "${PROJECT_BINARY_DIR}/x265.pdb" DESTINATION ${BIN_INSTALL_DIR} CONFIGURATIONS Debug)
+        install(FILES "${PROJECT_BINARY_DIR}/x265.pdb" DESTINATION ${BIN_INSTALL_DIR} CONFIGURATIONS RelWithDebInfo)
+        install(FILES "${PROJECT_BINARY_DIR}/libx265.pdb" DESTINATION ${BIN_INSTALL_DIR} CONFIGURATIONS Debug OPTIONAL NAMELINK_ONLY)
+        install(FILES "${PROJECT_BINARY_DIR}/libx265.pdb" DESTINATION ${BIN_INSTALL_DIR} CONFIGURATIONS RelWithDebInfo OPTIONAL NAMELINK_ONLY)
+    endif()
 endif()
-
 if(CMAKE_RC_COMPILER)
     # The resource compiler does not need CFLAGS or macro defines. It
     # often breaks them

From ea3d0eb69b888a3f79884073c12da076c737afcc Mon Sep 17 00:00:00 2001
From: Ma0 <mateuszb@poczta.onet.pl>
Date: Thu, 30 Nov 2017 18:29:19 +0100
Subject: [PATCH 10/51] cli: add new option '--fullhelp'

---
 source/x265.cpp  | 8 +++++++-
 source/x265cli.h | 5 +++--
 2 files changed, 10 insertions(+), 3 deletions(-)

diff --git a/source/x265.cpp b/source/x265.cpp
index 61e83eaf78..e9d0bf97bb 100644
--- a/source/x265.cpp
+++ b/source/x265.cpp
@@ -301,9 +301,15 @@ bool CLIOptions::parse(int argc, char **argv)
                 if (!this->qpfile)
                     x265_log_file(param, X265_LOG_ERROR, "%s qpfile not found or error in opening qp file\n", optarg);
             }
+            OPT("fullhelp")
+            {
+                param->logLevel = X265_LOG_FULL;
+                printVersion(param, api);
+                showHelp(param);
+                break;
+            }
             else
                 bError |= !!api->param_parse(param, long_options[long_options_index].name, optarg);
-
             if (bError)
             {
                 const char *name = long_options_index > 0 ? long_options[long_options_index].name : argv[optind - 2];
diff --git a/source/x265cli.h b/source/x265cli.h
index de5afcc1f6..3b59cf3a91 100644
--- a/source/x265cli.h
+++ b/source/x265cli.h
@@ -38,6 +38,7 @@ static const char short_options[] = "o:D:P:p:f:F:r:I:i:b:s:t:q:m:hwV?";
 static const struct option long_options[] =
 {
     { "help",                 no_argument, NULL, 'h' },
+    { "fullhelp",             no_argument, NULL, 0 },
     { "version",              no_argument, NULL, 'V' },
     { "asm",            required_argument, NULL, 0 },
     { "no-asm",               no_argument, NULL, 0 },
@@ -315,6 +316,7 @@ static void showHelp(x265_param *param)
     H0("    outfile is raw HEVC bitstream\n");
     H0("\nExecutable Options:\n");
     H0("-h/--help                        Show this help text and exit\n");
+    H0("   --fullhelp                    Show all options and exit\n");
     H0("-V/--version                     Show version info and exit\n");
     H0("\nOutput Options:\n");
     H0("-o/--output <filename>           Bitstream output file name\n");
@@ -565,9 +567,8 @@ static void showHelp(x265_param *param)
 #undef OPT
 #undef H0
 #undef H1
-
     if (level < X265_LOG_DEBUG)
-        printf("\nUse --log-level full --help for a full listing\n");
+        printf("\nUse --fullhelp for a full listing (or --log-level full --help)\n");
     printf("\n\nComplete documentation may be found at http://x265.readthedocs.org/en/default/cli.html\n");
     exit(1);
 }

From f3e9e95b0d20675778aef302cb1f86a29f3abe52 Mon Sep 17 00:00:00 2001
From: Aruna Matheswaran <aruna@multicorewareinc.com>
Date: Mon, 4 Dec 2017 15:00:04 +0530
Subject: [PATCH 11/51] test: Fix gop-lookahead cli error in regression test

---
 source/test/regression-tests.txt | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/source/test/regression-tests.txt b/source/test/regression-tests.txt
index 0dd5dd9323..48e2d43e32 100644
--- a/source/test/regression-tests.txt
+++ b/source/test/regression-tests.txt
@@ -150,7 +150,7 @@ Kimono1_1920x1080_24_400.yuv,--preset ultrafast --slices 1 --weightp --tu-intra-
 Kimono1_1920x1080_24_400.yuv,--preset medium --rdoq-level 0 --limit-refs 3 --slices 2
 Kimono1_1920x1080_24_400.yuv,--preset veryslow --crf 4 --cu-lossless --slices 2 --limit-refs 3 --limit-modes
 Kimono1_1920x1080_24_400.yuv,--preset placebo --ctu 32 --max-tu-size 8 --limit-tu 2
-big_buck_bunny_360p24.y4m, --keyint 60 --min-keyint 40 --gop-lookahead 20
+big_buck_bunny_360p24.y4m, --keyint 60 --min-keyint 40 --gop-lookahead 14
 # Main12 intraCost overflow bug test
 720p50_parkrun_ter.y4m,--preset medium
 

From abb93e663a248e55c56f815861917697a717a559 Mon Sep 17 00:00:00 2001
From: Santhoshini Sekar <santhoshini@multicorewareinc.com>
Date: Mon, 4 Dec 2017 16:08:19 +0530
Subject: [PATCH 12/51] wait until the last row of recon is complete. Waiting
 on m_reconEncoded doesn't ensure full recon generation.

---
 source/encoder/encoder.cpp | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/source/encoder/encoder.cpp b/source/encoder/encoder.cpp
index a8a9eb6db4..0512800fba 100644
--- a/source/encoder/encoder.cpp
+++ b/source/encoder/encoder.cpp
@@ -463,8 +463,8 @@ int Encoder::getRefFrameList(PicYuv** l0, PicYuv** l1, int sliceType, int poc)
                 {
                     int l0POC = framePtr->m_encData->m_slice->m_refFrameList[0][j]->m_poc;
                     Frame* l0Fp = m_dpb->m_picList.getPOC(l0POC);
-                    if (l0Fp->m_reconPic->m_picOrg[0] == NULL)
-                        l0Fp->m_reconEncoded.wait(); /* If recon is not ready, current frame encoder need to wait. */
+                    while (l0Fp->m_reconRowFlag[l0Fp->m_numRows - 1].get() == 0)
+                        l0Fp->m_reconRowFlag[l0Fp->m_numRows - 1].waitForChange(0); /* If recon is not ready, current frame encoder has to wait. */
                     l0[j] = l0Fp->m_reconPic;
                 }
             }
@@ -474,8 +474,8 @@ int Encoder::getRefFrameList(PicYuv** l0, PicYuv** l1, int sliceType, int poc)
                 {
                     int l1POC = framePtr->m_encData->m_slice->m_refFrameList[1][j]->m_poc;
                     Frame* l1Fp = m_dpb->m_picList.getPOC(l1POC);
-                    if (l1Fp->m_reconPic->m_picOrg[0] == NULL)
-                        l1Fp->m_reconEncoded.wait(); /* If recon is not ready, current frame encoder need to wait. */
+                    while (l1Fp->m_reconRowFlag[l1Fp->m_numRows - 1].get() == 0)
+                        l1Fp->m_reconRowFlag[l1Fp->m_numRows - 1].waitForChange(0); /* If recon is not ready, current frame encoder has to wait. */
                     l1[j] = l1Fp->m_reconPic;
                 }
             }

From 305677fdcf70c79719dfdcc6f4f8f230ea32bf68 Mon Sep 17 00:00:00 2001
From: Santhoshini Sekar <santhoshini@multicorewareinc.com>
Date: Mon, 4 Dec 2017 16:13:45 +0530
Subject: [PATCH 13/51] remove unnecessary event m_reconEncoded from Frame
 class

---
 source/common/frame.h           | 1 -
 source/encoder/frameencoder.cpp | 2 --
 2 files changed, 3 deletions(-)

diff --git a/source/common/frame.h b/source/common/frame.h
index 50117a456b..c3b9a89f10 100644
--- a/source/common/frame.h
+++ b/source/common/frame.h
@@ -98,7 +98,6 @@ class Frame
 
     float*                 m_quantOffsets;       // points to quantOffsets in x265_picture
     x265_sei               m_userSEI;
-    Event                  m_reconEncoded;
 
     /* Frame Parallelism - notification between FrameEncoders of available motion reference rows */
     ThreadSafeInteger*     m_reconRowFlag;       // flag of CTU rows completely reconstructed and extended for motion reference
diff --git a/source/encoder/frameencoder.cpp b/source/encoder/frameencoder.cpp
index 2e9cb7550e..ce24227c70 100644
--- a/source/encoder/frameencoder.cpp
+++ b/source/encoder/frameencoder.cpp
@@ -342,8 +342,6 @@ void FrameEncoder::threadMain()
         }
         compressFrame();
         m_done.trigger(); /* FrameEncoder::getEncodedPicture() blocks for this event */
-        if (m_frame != NULL)
-            m_frame->m_reconEncoded.trigger();
         m_enable.wait();
     }
 }

From 0c6817890883462b8e41880a25fbafbc692f1407 Mon Sep 17 00:00:00 2001
From: Aruna Matheswaran <aruna@multicorewareinc.com>
Date: Thu, 7 Dec 2017 10:59:12 +0530
Subject: [PATCH 14/51] rc: Fix inconsistency in --const-vbv (issue #381)

VBV intializations during ABR-reset is removed.
Inconsistent rowTotalBits in rc-update with --const-vbv is fixed.
---
 source/encoder/frameencoder.cpp | 4 ++--
 source/encoder/ratecontrol.cpp  | 4 +++-
 source/encoder/ratecontrol.h    | 1 +
 3 files changed, 6 insertions(+), 3 deletions(-)

diff --git a/source/encoder/frameencoder.cpp b/source/encoder/frameencoder.cpp
index ce24227c70..2eb4447b5e 100644
--- a/source/encoder/frameencoder.cpp
+++ b/source/encoder/frameencoder.cpp
@@ -1746,8 +1746,8 @@ void FrameEncoder::processRowEncoder(int intRow, ThreadLocalData& tld)
         if (rowInSlice == rowCount)
         {
             m_rowSliceTotalBits[sliceId] = 0;
-            if (bIsVbv)
-            {                
+            if (bIsVbv && !(m_param->rc.bEnableConstVbv && m_param->bEnableWavefront))
+            {          
                 for (uint32_t i = m_sliceBaseRow[sliceId]; i < rowCount + m_sliceBaseRow[sliceId]; i++)
                     m_rowSliceTotalBits[sliceId] += curEncData.m_rowStat[i].encodedBits;
             }
diff --git a/source/encoder/ratecontrol.cpp b/source/encoder/ratecontrol.cpp
index 3864f86bfb..5193c1a888 100644
--- a/source/encoder/ratecontrol.cpp
+++ b/source/encoder/ratecontrol.cpp
@@ -219,6 +219,7 @@ RateControl::RateControl(x265_param& p)
     m_param->rc.vbvMaxBitrate = x265_clip3(0, 2000000, m_param->rc.vbvMaxBitrate);
     m_param->rc.vbvBufferInit = x265_clip3(0.0, 2000000.0, m_param->rc.vbvBufferInit);
     m_param->vbvBufferEnd = x265_clip3(0.0, 2000000.0, m_param->vbvBufferEnd);
+    m_initVbv = false;
     m_singleFrameVbv = 0;
     m_rateTolerance = 1.0;
 
@@ -319,7 +320,7 @@ RateControl::RateControl(x265_param& p)
 
 bool RateControl::init(const SPS& sps)
 {
-    if (m_isVbv)
+    if (m_isVbv && !m_initVbv)
     {
         /* We don't support changing the ABR bitrate right now,
          * so if the stream starts as CBR, keep it CBR. */
@@ -353,6 +354,7 @@ bool RateControl::init(const SPS& sps)
         m_bufferFillFinal = m_bufferSize * m_param->rc.vbvBufferInit;
         m_bufferFillActual = m_bufferFillFinal;
         m_bufferExcess = 0;
+        m_initVbv = true;
     }
 
     m_totalBits = 0;
diff --git a/source/encoder/ratecontrol.h b/source/encoder/ratecontrol.h
index 49618da824..4fac01a4bc 100644
--- a/source/encoder/ratecontrol.h
+++ b/source/encoder/ratecontrol.h
@@ -132,6 +132,7 @@ class RateControl
     bool   m_isGrainEnabled;
     bool   m_isAbrReset;
     bool   m_isNextGop;
+    bool   m_initVbv;
     int    m_lastAbrResetPoc;
 
     double m_rateTolerance;

From 317214a3b0d92165116c46f5eb04d65945f24881 Mon Sep 17 00:00:00 2001
From: Ma0 <mateuszb@poczta.onet.pl>
Date: Thu, 7 Dec 2017 09:19:11 +0100
Subject: [PATCH 15/51] remove unnecessary sqrt from MVTHRESHOLD checking

Thanks MonoS Diablo for pointing this.
---
 source/encoder/encoder.cpp | 15 ++++++---------
 1 file changed, 6 insertions(+), 9 deletions(-)

diff --git a/source/encoder/encoder.cpp b/source/encoder/encoder.cpp
index e7d590316d..153b7dce67 100644
--- a/source/encoder/encoder.cpp
+++ b/source/encoder/encoder.cpp
@@ -50,10 +50,8 @@ const char g_sliceTypeToChar[] = {'B', 'P', 'I'};
 
 /* Threshold for motion vection, based on expermental result.
  * TODO: come up an algorithm for adoptive threshold */
-
-#define MVTHRESHOLD 10
+#define MVTHRESHOLD (10*10)
 #define PU_2Nx2N 1
-
 static const char* defaultAnalysisFileName = "x265_analysis.dat";
 
 using namespace X265_NS;
@@ -571,12 +569,12 @@ int Encoder::setAnalysisDataAfterZScan(x265_analysis_data *analysis_data, Frame*
                             (interData)->mvpIdx[k][cuPos + cuOffset] = (srcInterData)->mvpIdx[k][(mbIndex * 16) + cuOffset];
                             (interData)->refIdx[k][cuPos + cuOffset] = (srcInterData)->refIdx[k][(mbIndex * 16) + cuOffset];
                             memcpy(&(interData)->mv[k][cuPos + cuOffset], &(srcInterData)->mv[k][(mbIndex * 16) + cuOffset], sizeof(MV));
-                            if (m_param->analysisReuseLevel == 7)
+                            if (m_param->analysisReuseLevel == 7 && numPU == PU_2Nx2N &&
+                                ((srcInterData)->depth[cuPos + cuOffset] == (m_param->maxCUSize >> 5)))
                             {
                                 int mv_x = ((analysis_inter_data *)curFrame->m_analysisData.interData)->mv[k][(mbIndex * 16) + cuOffset].x;
                                 int mv_y = ((analysis_inter_data *)curFrame->m_analysisData.interData)->mv[k][(mbIndex * 16) + cuOffset].y;
-                                double mv = sqrt((double)(mv_x*mv_x + mv_y*mv_y));
-                                if (numPU == PU_2Nx2N && ((srcInterData)->depth[cuPos + cuOffset] == (m_param->maxCUSize >> 5)) && mv <= MVTHRESHOLD)
+                                if ((mv_x*mv_x + mv_y*mv_y) <= MVTHRESHOLD)
                                     memset(&curFrame->m_analysisData.modeFlag[k][cuPos + cuOffset], 1, bytes);
                             }
                         }
@@ -654,12 +652,11 @@ int Encoder::setAnalysisData(x265_analysis_data *analysis_data, int poc, uint32_
                                     (currInterData)->mvpIdx[i][count + pu] = (interData)->mvpIdx[i][d];
                                     (currInterData)->refIdx[i][count + pu] = (interData)->refIdx[i][d];
                                     memcpy(&(currInterData)->mv[i][count + pu], &(interData)->mv[i][d], sizeof(MV));
-                                    if (m_param->analysisReuseLevel == 7)
+                                    if (m_param->analysisReuseLevel == 7 && numPU == PU_2Nx2N && m_param->num4x4Partitions <= 16)
                                     {
                                         int mv_x = ((analysis_inter_data *)curFrame->m_analysisData.interData)->mv[i][count + pu].x;
                                         int mv_y = ((analysis_inter_data *)curFrame->m_analysisData.interData)->mv[i][count + pu].y;
-                                        double mv = sqrt((double)(mv_x*mv_x + mv_y*mv_y));
-                                        if (numPU == PU_2Nx2N && m_param->num4x4Partitions <= 16 && mv <= MVTHRESHOLD)
+                                        if ((mv_x*mv_x + mv_y*mv_y) <= MVTHRESHOLD)
                                             memset(&curFrame->m_analysisData.modeFlag[i][count + pu], 1, bytes);
                                     }
                                 }

From 026149cf540975964dd9e4976e8d70975b3c722f Mon Sep 17 00:00:00 2001
From: Santhoshini Sekar <santhoshini@multicorewareinc.com>
Date: Tue, 12 Dec 2017 18:04:04 +0530
Subject: [PATCH 16/51] modify api x265_get_ref_frame_list to provide POC lists
 for L0 and L1 references

---
 doc/reST/api.rst           | 2 +-
 source/CMakeLists.txt      | 2 +-
 source/encoder/api.cpp     | 4 ++--
 source/encoder/encoder.cpp | 4 +++-
 source/encoder/encoder.h   | 2 +-
 source/x265.h              | 4 ++--
 6 files changed, 10 insertions(+), 8 deletions(-)

diff --git a/doc/reST/api.rst b/doc/reST/api.rst
index 92d2dab9d4..4c6b71729d 100644
--- a/doc/reST/api.rst
+++ b/doc/reST/api.rst
@@ -206,7 +206,7 @@ changes made to the parameters for auto-detection and other reasons::
     /* x265_get_ref_frame_list:
      *     returns negative on error, 0 when access unit were output.
      *     This API must be called after(poc >= lookaheadDepth + bframes + 2) condition check */
-     int x265_get_ref_frame_list(x265_encoder *encoder, x265_picyuv**, x265_picyuv**, int, int);
+     int x265_get_ref_frame_list(x265_encoder *encoder, x265_picyuv**, x265_picyuv**, int, int, int*, int*);
  
 **x265_encoder_ctu_info** may be used to provide additional CTU-specific information to the encoder::
 
diff --git a/source/CMakeLists.txt b/source/CMakeLists.txt
index 1e1d91d140..c859ff4cc0 100644
--- a/source/CMakeLists.txt
+++ b/source/CMakeLists.txt
@@ -29,7 +29,7 @@ option(NATIVE_BUILD "Target the build CPU" OFF)
 option(STATIC_LINK_CRT "Statically link C runtime for release builds" OFF)
 mark_as_advanced(FPROFILE_USE FPROFILE_GENERATE NATIVE_BUILD)
 # X265_BUILD must be incremented each time the public API is changed
-set(X265_BUILD 147)
+set(X265_BUILD 148)
 configure_file("${PROJECT_SOURCE_DIR}/x265.def.in"
                "${PROJECT_BINARY_DIR}/x265.def")
 configure_file("${PROJECT_SOURCE_DIR}/x265_config.h.in"
diff --git a/source/encoder/api.cpp b/source/encoder/api.cpp
index 2ae2bc82dc..955c941243 100644
--- a/source/encoder/api.cpp
+++ b/source/encoder/api.cpp
@@ -356,13 +356,13 @@ int x265_get_slicetype_poc_and_scenecut(x265_encoder *enc, int *slicetype, int *
     return -1;
 }
 
-int x265_get_ref_frame_list(x265_encoder *enc, x265_picyuv** l0, x265_picyuv** l1, int sliceType, int poc)
+int x265_get_ref_frame_list(x265_encoder *enc, x265_picyuv** l0, x265_picyuv** l1, int sliceType, int poc, int* pocL0, int* pocL1)
 {
     if (!enc)
         return -1;
 
     Encoder *encoder = static_cast<Encoder*>(enc);
-    return encoder->getRefFrameList((PicYuv**)l0, (PicYuv**)l1, sliceType, poc);
+    return encoder->getRefFrameList((PicYuv**)l0, (PicYuv**)l1, sliceType, poc, pocL0, pocL1);
 }
 
 int x265_set_analysis_data(x265_encoder *enc, x265_analysis_data *analysis_data, int poc, uint32_t cuBytes)
diff --git a/source/encoder/encoder.cpp b/source/encoder/encoder.cpp
index 153b7dce67..b8cab1c410 100644
--- a/source/encoder/encoder.cpp
+++ b/source/encoder/encoder.cpp
@@ -448,7 +448,7 @@ int Encoder::copySlicetypePocAndSceneCut(int *slicetype, int *poc, int *sceneCut
     return 0;
 }
 
-int Encoder::getRefFrameList(PicYuv** l0, PicYuv** l1, int sliceType, int poc)
+int Encoder::getRefFrameList(PicYuv** l0, PicYuv** l1, int sliceType, int poc, int* pocL0, int* pocL1)
 {
     if (!(IS_X265_TYPE_I(sliceType)))
     {
@@ -460,6 +460,7 @@ int Encoder::getRefFrameList(PicYuv** l0, PicYuv** l1, int sliceType, int poc)
                 if (framePtr->m_encData->m_slice->m_refFrameList[0][j] && framePtr->m_encData->m_slice->m_refFrameList[0][j]->m_reconPic != NULL)
                 {
                     int l0POC = framePtr->m_encData->m_slice->m_refFrameList[0][j]->m_poc;
+                    pocL0[j] = l0POC;
                     Frame* l0Fp = m_dpb->m_picList.getPOC(l0POC);
                     while (l0Fp->m_reconRowFlag[l0Fp->m_numRows - 1].get() == 0)
                         l0Fp->m_reconRowFlag[l0Fp->m_numRows - 1].waitForChange(0); /* If recon is not ready, current frame encoder has to wait. */
@@ -471,6 +472,7 @@ int Encoder::getRefFrameList(PicYuv** l0, PicYuv** l1, int sliceType, int poc)
                 if (framePtr->m_encData->m_slice->m_refFrameList[1][j] && framePtr->m_encData->m_slice->m_refFrameList[1][j]->m_reconPic != NULL)
                 {
                     int l1POC = framePtr->m_encData->m_slice->m_refFrameList[1][j]->m_poc;
+                    pocL1[j] = l1POC;
                     Frame* l1Fp = m_dpb->m_picList.getPOC(l1POC);
                     while (l1Fp->m_reconRowFlag[l1Fp->m_numRows - 1].get() == 0)
                         l1Fp->m_reconRowFlag[l1Fp->m_numRows - 1].waitForChange(0); /* If recon is not ready, current frame encoder has to wait. */
diff --git a/source/encoder/encoder.h b/source/encoder/encoder.h
index b71b89219a..5948346e96 100644
--- a/source/encoder/encoder.h
+++ b/source/encoder/encoder.h
@@ -208,7 +208,7 @@ class Encoder : public x265_encoder
 
     int copySlicetypePocAndSceneCut(int *slicetype, int *poc, int *sceneCut);
 
-    int getRefFrameList(PicYuv** l0, PicYuv** l1, int sliceType, int poc);
+    int getRefFrameList(PicYuv** l0, PicYuv** l1, int sliceType, int poc, int* pocL0, int* pocL1);
 
     int setAnalysisDataAfterZScan(x265_analysis_data *analysis_data, Frame* curFrame);
 
diff --git a/source/x265.h b/source/x265.h
index 8f6d5ff90a..9d6b5d606f 100644
--- a/source/x265.h
+++ b/source/x265.h
@@ -1746,7 +1746,7 @@ int x265_get_slicetype_poc_and_scenecut(x265_encoder *encoder, int *slicetype, i
 /* x265_get_ref_frame_list:
  *     returns negative on error, 0 when access unit were output.
  *     This API must be called after(poc >= lookaheadDepth + bframes + 2) condition check */
-int x265_get_ref_frame_list(x265_encoder *encoder, x265_picyuv**, x265_picyuv**, int, int);
+int x265_get_ref_frame_list(x265_encoder *encoder, x265_picyuv**, x265_picyuv**, int, int, int*, int*);
 
 /* x265_set_analysis_data:
  *     set the analysis data. The incoming analysis_data structure is assumed to be AVC-sized blocks.
@@ -1823,7 +1823,7 @@ typedef struct x265_api
     int           (*encoder_intra_refresh)(x265_encoder*);
     int           (*encoder_ctu_info)(x265_encoder*, int, x265_ctu_info_t**);
     int           (*get_slicetype_poc_and_scenecut)(x265_encoder*, int*, int*, int*);
-    int           (*get_ref_frame_list)(x265_encoder*, x265_picyuv**, x265_picyuv**, int, int);
+    int           (*get_ref_frame_list)(x265_encoder*, x265_picyuv**, x265_picyuv**, int, int, int*, int*);
     FILE*         (*csvlog_open)(const x265_param*);
     void          (*csvlog_frame)(const x265_param*, const x265_picture*);
     void          (*csvlog_encode)(x265_encoder*, const x265_stats*, int, char**);

From bf494da403ddefc92e2cf5322aaaf776688c3412 Mon Sep 17 00:00:00 2001
From: Divya Manivannan <divya@multicorewareinc.com>
Date: Thu, 14 Dec 2017 15:57:01 +0530
Subject: [PATCH 17/51] analysis: avoid creating analysis file when
 useanalysisFile option is disabled

---
 source/encoder/encoder.cpp | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)

diff --git a/source/encoder/encoder.cpp b/source/encoder/encoder.cpp
index b8cab1c410..072878df59 100644
--- a/source/encoder/encoder.cpp
+++ b/source/encoder/encoder.cpp
@@ -342,10 +342,8 @@ void Encoder::create()
         m_aborted = true;
     if (!m_lookahead->create())
         m_aborted = true;
-
     initRefIdx();
-
-    if (m_param->analysisReuseMode)
+    if (m_param->analysisReuseMode && m_param->bUseAnalysisFile)
     {
         const char* name = m_param->analysisReuseFileName;
         if (!name)
@@ -3248,8 +3246,8 @@ void Encoder::readAnalysisFile(x265_analysis_data* analysis, int curPoc, const x
     static uint64_t consumedBytes = 0;
     static uint64_t totalConsumedBytes = 0;
     uint32_t depthBytes = 0;
-    fseeko(m_analysisFile, totalConsumedBytes, SEEK_SET);
-
+    if (m_param->bUseAnalysisFile)
+        fseeko(m_analysisFile, totalConsumedBytes, SEEK_SET);
     const x265_analysis_data *picData = &(picIn->analysisData);
     analysis_intra_data *intraPic = (analysis_intra_data *)picData->intraData;
     analysis_inter_data *interPic = (analysis_inter_data *)picData->interData;

From b2c261cc10b9864dc52008c5579023603f65817f Mon Sep 17 00:00:00 2001
From: Santhoshini Sekar <santhoshini@multicorewareinc.com>
Date: Tue, 19 Dec 2017 15:33:30 +0530
Subject: [PATCH 18/51] fix bugs in analysis-reuse-level=7 and
 refine-mv-type=AVC

---
 source/encoder/analysis.cpp | 57 +++++++++++++++++++++----------------
 source/encoder/encoder.cpp  | 21 +++++++-------
 2 files changed, 44 insertions(+), 34 deletions(-)

diff --git a/source/encoder/analysis.cpp b/source/encoder/analysis.cpp
index e3d21bf042..6eecc4375e 100644
--- a/source/encoder/analysis.cpp
+++ b/source/encoder/analysis.cpp
@@ -280,7 +280,7 @@ Mode& Analysis::compressCTU(CUData& ctu, Frame& frame, const CUGeom& cuGeom, con
             /* generate residual for entire CTU at once and copy to reconPic */
             encodeResidue(ctu, cuGeom);
         }
-        else if ((m_param->analysisReuseMode == X265_ANALYSIS_LOAD && m_param->analysisReuseLevel == 10) || ((m_param->bMVType == AVC_INFO) && m_param->analysisReuseLevel >= 7))
+        else if ((m_param->analysisReuseMode == X265_ANALYSIS_LOAD && m_param->analysisReuseLevel == 10) || ((m_param->bMVType == AVC_INFO) && m_param->analysisReuseLevel >= 7 && ctu.m_numPartitions <= 16))
         {
             analysis_inter_data* interDataCTU = (analysis_inter_data*)m_frame->m_analysisData.interData;
             int posCTU = ctu.m_cuAddr * numPartition;
@@ -459,11 +459,9 @@ void Analysis::qprdRefine(const CUData& parentCTU, const CUGeom& cuGeom, int32_t
 
     int bestCUQP = qp;
     int lambdaQP = lqp;
-
     bool doQPRefine = (bDecidedDepth && depth <= m_slice->m_pps->maxCuDQPDepth) || (!bDecidedDepth && depth == m_slice->m_pps->maxCuDQPDepth);
-    if (m_param->analysisReuseLevel == 10)
+    if (m_param->analysisReuseLevel >= 7)
         doQPRefine = false;
-
     if (doQPRefine)
     {
         uint64_t bestCUCost, origCUCost, cuCost, cuPrevCost;
@@ -1305,9 +1303,8 @@ SplitData Analysis::compressInterCU_rd0_4(const CUData& parentCTU, const CUGeom&
                 }
             }
         }
-
         /* Step 1. Evaluate Merge/Skip candidates for likely early-outs, if skip mode was not set above */
-        if ((mightNotSplit && depth >= minDepth && !md.bestMode && !bCtuInfoCheck) || (m_param->bMVType && (m_modeFlag[0] || m_modeFlag[1]))) /* TODO: Re-evaluate if analysis load/save still works */
+        if ((mightNotSplit && depth >= minDepth && !md.bestMode && !bCtuInfoCheck) || (m_param->bMVType && m_param->analysisReuseLevel == 7 && (m_modeFlag[0] || m_modeFlag[1]))) /* TODO: Re-evaluate if analysis load/save still works */
         {
             /* Compute Merge Cost */
             md.pred[PRED_MERGE].cu.initSubCU(parentCTU, cuGeom, qp);
@@ -1317,8 +1314,7 @@ SplitData Analysis::compressInterCU_rd0_4(const CUData& parentCTU, const CUGeom&
                 skipModes = (m_param->bEnableEarlySkip || m_param->interRefine == 2)
                 && md.bestMode && md.bestMode->cu.isSkipped(0); // TODO: sa8d threshold per depth
         }
-
-        if (md.bestMode && m_param->bEnableRecursionSkip && !bCtuInfoCheck && !(m_param->bMVType && (m_modeFlag[0] || m_modeFlag[1])))
+        if (md.bestMode && m_param->bEnableRecursionSkip && !bCtuInfoCheck && !(m_param->bMVType && m_param->analysisReuseLevel == 7 && (m_modeFlag[0] || m_modeFlag[1])))
         {
             skipRecursion = md.bestMode->cu.isSkipped(0);
             if (mightSplit && depth >= minDepth && !skipRecursion)
@@ -1329,10 +1325,8 @@ SplitData Analysis::compressInterCU_rd0_4(const CUData& parentCTU, const CUGeom&
                     skipRecursion = complexityCheckCU(*md.bestMode);
             }
         }
-
-        if (m_param->bMVType && md.bestMode && cuGeom.numPartitions <= 16)
+        if (m_param->bMVType && md.bestMode && cuGeom.numPartitions <= 16 && m_param->analysisReuseLevel == 7)
             skipRecursion = true;
-
         /* Step 2. Evaluate each of the 4 split sub-blocks in series */
         if (mightSplit && !skipRecursion)
         {
@@ -1387,11 +1381,20 @@ SplitData Analysis::compressInterCU_rd0_4(const CUData& parentCTU, const CUGeom&
             else
                 splitPred->sa8dCost = m_rdCost.calcRdSADCost((uint32_t)splitPred->distortion, splitPred->sa8dBits);
         }
-
         /* If analysis mode is simple do not Evaluate other modes */
-        if ((m_param->bMVType && cuGeom.numPartitions <= 16) && (m_slice->m_sliceType == P_SLICE || m_slice->m_sliceType == B_SLICE))
-            mightNotSplit = !(m_checkMergeAndSkipOnly[0] || (m_checkMergeAndSkipOnly[0] && m_checkMergeAndSkipOnly[1]));
-
+        if (m_param->bMVType && m_param->analysisReuseLevel == 7)
+        {
+            if (m_slice->m_sliceType == P_SLICE)
+            {
+                if (m_checkMergeAndSkipOnly[0])
+                    skipModes = true;
+            }
+            else
+            {
+                if (m_checkMergeAndSkipOnly[0] && m_checkMergeAndSkipOnly[1])
+                    skipModes = true;
+            }
+        }
         /* Split CUs
          *   0  1
          *   2  3 */
@@ -1998,10 +2001,9 @@ SplitData Analysis::compressInterCU_rd5_6(const CUData& parentCTU, const CUGeom&
                 }
             }
         }
-
         /* Step 1. Evaluate Merge/Skip candidates for likely early-outs */
         if ((mightNotSplit && !md.bestMode && !bCtuInfoCheck) ||
-            (m_param->bMVType && (m_modeFlag[0] || m_modeFlag[1])))
+            (m_param->bMVType && m_param->analysisReuseLevel == 7 && (m_modeFlag[0] || m_modeFlag[1])))
         {
             md.pred[PRED_SKIP].cu.initSubCU(parentCTU, cuGeom, qp);
             md.pred[PRED_MERGE].cu.initSubCU(parentCTU, cuGeom, qp);
@@ -2016,10 +2018,8 @@ SplitData Analysis::compressInterCU_rd5_6(const CUData& parentCTU, const CUGeom&
             if (m_param->bEnableRecursionSkip && depth && m_modeDepth[depth - 1].bestMode)
                 skipRecursion = md.bestMode && !md.bestMode->cu.getQtRootCbf(0);
         }
-
-        if (m_param->bMVType && md.bestMode && cuGeom.numPartitions <= 16)
+        if (m_param->bMVType && md.bestMode && cuGeom.numPartitions <= 16 && m_param->analysisReuseLevel == 7)
             skipRecursion = true;
-
         // estimate split cost
         /* Step 2. Evaluate each of the 4 split sub-blocks in series */
         if (mightSplit && !skipRecursion)
@@ -2071,11 +2071,20 @@ SplitData Analysis::compressInterCU_rd5_6(const CUData& parentCTU, const CUGeom&
 
             checkDQPForSplitPred(*splitPred, cuGeom);
         }
-
         /* If analysis mode is simple do not Evaluate other modes */
-        if ((m_param->bMVType && cuGeom.numPartitions <= 16) && (m_slice->m_sliceType == P_SLICE || m_slice->m_sliceType == B_SLICE))
-            mightNotSplit = !(m_checkMergeAndSkipOnly[0] || (m_checkMergeAndSkipOnly[0] && m_checkMergeAndSkipOnly[1]));
-
+        if (m_param->bMVType && m_param->analysisReuseLevel == 7)
+        {
+            if (m_slice->m_sliceType == P_SLICE)
+            {
+                if (m_checkMergeAndSkipOnly[0])
+                    skipModes = true;
+            }
+            else
+            {
+                if (m_checkMergeAndSkipOnly[0] && m_checkMergeAndSkipOnly[1])
+                    skipModes = true;
+            }
+        }
         /* Split CUs
          *   0  1
          *   2  3 */
diff --git a/source/encoder/encoder.cpp b/source/encoder/encoder.cpp
index 072878df59..d187ca1397 100644
--- a/source/encoder/encoder.cpp
+++ b/source/encoder/encoder.cpp
@@ -562,7 +562,7 @@ int Encoder::setAnalysisDataAfterZScan(x265_analysis_data *analysis_data, Frame*
                     {
                         int cuOffset = cuI * bytes + pu;
                         (interData)->mergeFlag[cuPos + cuOffset] = (srcInterData)->mergeFlag[(mbIndex * 16) + cuOffset];
-
+                        (interData)->sadCost[cuPos + cuOffset] = (srcInterData)->sadCost[(mbIndex * 16) + cuOffset];
                         (interData)->interDir[cuPos + cuOffset] = (srcInterData)->interDir[(mbIndex * 16) + cuOffset];
                         for (uint32_t k = 0; k < numDir; k++)
                         {
@@ -570,10 +570,10 @@ int Encoder::setAnalysisDataAfterZScan(x265_analysis_data *analysis_data, Frame*
                             (interData)->refIdx[k][cuPos + cuOffset] = (srcInterData)->refIdx[k][(mbIndex * 16) + cuOffset];
                             memcpy(&(interData)->mv[k][cuPos + cuOffset], &(srcInterData)->mv[k][(mbIndex * 16) + cuOffset], sizeof(MV));
                             if (m_param->analysisReuseLevel == 7 && numPU == PU_2Nx2N &&
-                                ((srcInterData)->depth[cuPos + cuOffset] == (m_param->maxCUSize >> 5)))
+                                ((interData)->depth[cuPos + cuOffset] == (m_param->maxCUSize >> 5)))
                             {
-                                int mv_x = ((analysis_inter_data *)curFrame->m_analysisData.interData)->mv[k][(mbIndex * 16) + cuOffset].x;
-                                int mv_y = ((analysis_inter_data *)curFrame->m_analysisData.interData)->mv[k][(mbIndex * 16) + cuOffset].y;
+                                int mv_x = (interData)->mv[k][cuPos + cuOffset].x;
+                                int mv_y = (interData)->mv[k][cuPos + cuOffset].y;
                                 if ((mv_x*mv_x + mv_y*mv_y) <= MVTHRESHOLD)
                                     memset(&curFrame->m_analysisData.modeFlag[k][cuPos + cuOffset], 1, bytes);
                             }
@@ -640,9 +640,10 @@ int Encoder::setAnalysisData(x265_analysis_data *analysis_data, int poc, uint32_
                     if (m_param->analysisReuseLevel > 4)
                     {
                         memset(&(currInterData)->partSize[count], (interData)->partSize[d], bytes);
-                        int numPU = nbPartsTable[(currInterData)->partSize[d]];
-                        for (int pu = 0; pu < numPU; pu++, d++)
+                        int numPU = nbPartsTable[(interData)->partSize[d]];
+                        for (int pu = 0; pu < numPU; pu++)
                         {
+                            if (pu) d++;
                             (currInterData)->mergeFlag[count + pu] = (interData)->mergeFlag[d];
                             if (m_param->analysisReuseLevel >= 7)
                             {
@@ -654,8 +655,8 @@ int Encoder::setAnalysisData(x265_analysis_data *analysis_data, int poc, uint32_
                                     memcpy(&(currInterData)->mv[i][count + pu], &(interData)->mv[i][d], sizeof(MV));
                                     if (m_param->analysisReuseLevel == 7 && numPU == PU_2Nx2N && m_param->num4x4Partitions <= 16)
                                     {
-                                        int mv_x = ((analysis_inter_data *)curFrame->m_analysisData.interData)->mv[i][count + pu].x;
-                                        int mv_y = ((analysis_inter_data *)curFrame->m_analysisData.interData)->mv[i][count + pu].y;
+                                        int mv_x = (currInterData)->mv[i][count + pu].x;
+                                        int mv_y = (currInterData)->mv[i][count + pu].y;
                                         if ((mv_x*mv_x + mv_y*mv_y) <= MVTHRESHOLD)
                                             memset(&curFrame->m_analysisData.modeFlag[i][count + pu], 1, bytes);
                                     }
@@ -3061,14 +3062,14 @@ void Encoder::allocAnalysis(x265_analysis_data* analysis)
         if (m_param->analysisReuseLevel >= 7)
         {
             CHECKED_MALLOC(interData->interDir, uint8_t, analysis->numPartitions * analysis->numCUsInFrame);
+            CHECKED_MALLOC(interData->sadCost, int64_t, analysis->numPartitions * analysis->numCUsInFrame);
             for (int dir = 0; dir < numDir; dir++)
             {
                 CHECKED_MALLOC(interData->mvpIdx[dir], uint8_t, analysis->numPartitions * analysis->numCUsInFrame);
                 CHECKED_MALLOC(interData->refIdx[dir], int8_t, analysis->numPartitions * analysis->numCUsInFrame);
                 CHECKED_MALLOC(interData->mv[dir], MV, analysis->numPartitions * analysis->numCUsInFrame);
-                CHECKED_MALLOC(analysis->modeFlag[dir], uint8_t, analysis->numPartitions * analysis->numCUsInFrame);
+                CHECKED_MALLOC_ZERO(analysis->modeFlag[dir], uint8_t, analysis->numPartitions * analysis->numCUsInFrame);
             }
-
             /* Allocate intra in inter */
             if (analysis->sliceType == X265_TYPE_P || m_param->bIntraInBFrames)
             {

From 76f85d1429ab438d8684ffc4c1c4dab91d4b22c6 Mon Sep 17 00:00:00 2001
From: Santhoshini Sekar <santhoshini@multicorewareinc.com>
Date: Thu, 21 Dec 2017 09:09:46 +0530
Subject: [PATCH 19/51] warn out saying that limitTU=3 or 4 with AVCINFO
 produces inconsistent output. Set mincusize to 8  when it is not 8 for
 MVtype=AVCINFO as AVCINFO expects so

---
 source/encoder/encoder.cpp | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/source/encoder/encoder.cpp b/source/encoder/encoder.cpp
index d187ca1397..ddda95bdfd 100644
--- a/source/encoder/encoder.cpp
+++ b/source/encoder/encoder.cpp
@@ -2449,6 +2449,18 @@ void Encoder::configure(x265_param *p)
         this->m_externalFlush = true;
     else 
         this->m_externalFlush = false;
+
+    if (p->bMVType == AVC_INFO && (p->limitTU == 3 || p->limitTU == 4))
+    {
+        x265_log(p, X265_LOG_WARNING, "limit TU = 3 or 4 with MVType AVCINFO produces inconsistent output\n");
+    }
+
+    if (p->bMVType == AVC_INFO && p->minCUSize != 8)
+    {
+        p->minCUSize = 8;
+        x265_log(p, X265_LOG_WARNING, "Setting minCuSize = 8, AVCINFO expects 8x8 blocks\n");
+    }
+
     if (p->keyframeMax < 0)
     {
         /* A negative max GOP size indicates the user wants only one I frame at

From cce6e95d4dfc5e470931013fed8330f5f567120f Mon Sep 17 00:00:00 2001
From: Divya Manivannan <divya@multicorewareinc.com>
Date: Thu, 14 Dec 2017 16:53:44 +0530
Subject: [PATCH 20/51] analysis: avoid copying lowres vbv data when vbv is
 disabled

---
 source/encoder/encoder.cpp | 77 +++++++++++++++++++++-----------------
 1 file changed, 42 insertions(+), 35 deletions(-)

diff --git a/source/encoder/encoder.cpp b/source/encoder/encoder.cpp
index ddda95bdfd..cf7589ba8b 100644
--- a/source/encoder/encoder.cpp
+++ b/source/encoder/encoder.cpp
@@ -1047,11 +1047,14 @@ int Encoder::encode(const x265_picture* pic_in, x265_picture* pic_out)
                 inFrame->m_lowres.sliceType = sliceType;
                 inFrame->m_lowres.bKeyframe = !!inFrame->m_analysisData.lookahead.keyframe;
                 inFrame->m_lowres.bLastMiniGopBFrame = !!inFrame->m_analysisData.lookahead.lastMiniGopBFrame;
-                int vbvCount = m_param->lookaheadDepth + m_param->bframes + 2;
-                for (int index = 0; index < vbvCount; index++)
+                if (m_rateControl->m_isVbv)
                 {
-                    inFrame->m_lowres.plannedSatd[index] = inFrame->m_analysisData.lookahead.plannedSatd[index];
-                    inFrame->m_lowres.plannedType[index] = inFrame->m_analysisData.lookahead.plannedType[index];
+                    int vbvCount = m_param->lookaheadDepth + m_param->bframes + 2;
+                    for (int index = 0; index < vbvCount; index++)
+                    {
+                        inFrame->m_lowres.plannedSatd[index] = inFrame->m_analysisData.lookahead.plannedSatd[index];
+                        inFrame->m_lowres.plannedType[index] = inFrame->m_analysisData.lookahead.plannedType[index];
+                    }
                 }
             }
         }
@@ -1165,26 +1168,29 @@ int Encoder::encode(const x265_picture* pic_in, x265_picture* pic_out)
                         pic_out->analysisData.satdCost *= factor;
                         pic_out->analysisData.lookahead.keyframe = outFrame->m_lowres.bKeyframe;
                         pic_out->analysisData.lookahead.lastMiniGopBFrame = outFrame->m_lowres.bLastMiniGopBFrame;
-                        int vbvCount = m_param->lookaheadDepth + m_param->bframes + 2;
-                        for (int index = 0; index < vbvCount; index++)
-                        {
-                            pic_out->analysisData.lookahead.plannedSatd[index] = outFrame->m_lowres.plannedSatd[index] * factor;
-                            pic_out->analysisData.lookahead.plannedType[index] = outFrame->m_lowres.plannedType[index];
-                        }
-                        for (uint32_t index = 0; index < pic_out->analysisData.numCuInHeight; index++)
-                        {
-                            outFrame->m_analysisData.lookahead.intraSatdForVbv[index] = outFrame->m_encData->m_rowStat[index].intraSatdForVbv * factor;
-                            outFrame->m_analysisData.lookahead.satdForVbv[index] = outFrame->m_encData->m_rowStat[index].satdForVbv * factor;
-                        }
-                        pic_out->analysisData.lookahead.intraSatdForVbv = outFrame->m_analysisData.lookahead.intraSatdForVbv;
-                        pic_out->analysisData.lookahead.satdForVbv = outFrame->m_analysisData.lookahead.satdForVbv;
-                        for (uint32_t index = 0; index < pic_out->analysisData.numCUsInFrame; index++)
+                        if (m_rateControl->m_isVbv)
                         {
-                            outFrame->m_analysisData.lookahead.intraVbvCost[index] = outFrame->m_encData->m_cuStat[index].intraVbvCost * factor;
-                            outFrame->m_analysisData.lookahead.vbvCost[index] = outFrame->m_encData->m_cuStat[index].vbvCost * factor;
+                            int vbvCount = m_param->lookaheadDepth + m_param->bframes + 2;
+                            for (int index = 0; index < vbvCount; index++)
+                            {
+                                pic_out->analysisData.lookahead.plannedSatd[index] = outFrame->m_lowres.plannedSatd[index] * factor;
+                                pic_out->analysisData.lookahead.plannedType[index] = outFrame->m_lowres.plannedType[index];
+                            }
+                            for (uint32_t index = 0; index < pic_out->analysisData.numCuInHeight; index++)
+                            {
+                                outFrame->m_analysisData.lookahead.intraSatdForVbv[index] = outFrame->m_encData->m_rowStat[index].intraSatdForVbv * factor;
+                                outFrame->m_analysisData.lookahead.satdForVbv[index] = outFrame->m_encData->m_rowStat[index].satdForVbv * factor;
+                            }
+                            pic_out->analysisData.lookahead.intraSatdForVbv = outFrame->m_analysisData.lookahead.intraSatdForVbv;
+                            pic_out->analysisData.lookahead.satdForVbv = outFrame->m_analysisData.lookahead.satdForVbv;
+                            for (uint32_t index = 0; index < pic_out->analysisData.numCUsInFrame; index++)
+                            {
+                                outFrame->m_analysisData.lookahead.intraVbvCost[index] = outFrame->m_encData->m_cuStat[index].intraVbvCost * factor;
+                                outFrame->m_analysisData.lookahead.vbvCost[index] = outFrame->m_encData->m_cuStat[index].vbvCost * factor;
+                            }
+                            pic_out->analysisData.lookahead.intraVbvCost = outFrame->m_analysisData.lookahead.intraVbvCost;
+                            pic_out->analysisData.lookahead.vbvCost = outFrame->m_analysisData.lookahead.vbvCost;
                         }
-                        pic_out->analysisData.lookahead.intraVbvCost = outFrame->m_analysisData.lookahead.intraVbvCost;
-                        pic_out->analysisData.lookahead.vbvCost = outFrame->m_analysisData.lookahead.vbvCost;
                     }
                     writeAnalysisFile(&pic_out->analysisData, *outFrame->m_encData);
                     if (m_param->bUseAnalysisFile)
@@ -1351,15 +1357,18 @@ int Encoder::encode(const x265_picture* pic_in, x265_picture* pic_out)
             if (m_param->analysisReuseMode == X265_ANALYSIS_LOAD && m_param->bDisableLookahead)
             {
                 frameEnc->m_dts = frameEnc->m_analysisData.lookahead.dts;
-                for (uint32_t index = 0; index < frameEnc->m_analysisData.numCuInHeight; index++)
+                if (m_rateControl->m_isVbv)
                 {
-                    frameEnc->m_encData->m_rowStat[index].intraSatdForVbv = frameEnc->m_analysisData.lookahead.intraSatdForVbv[index];
-                    frameEnc->m_encData->m_rowStat[index].satdForVbv = frameEnc->m_analysisData.lookahead.satdForVbv[index];
-                }
-                for (uint32_t index = 0; index < frameEnc->m_analysisData.numCUsInFrame; index++)
-                {
-                    frameEnc->m_encData->m_cuStat[index].intraVbvCost = frameEnc->m_analysisData.lookahead.intraVbvCost[index];
-                    frameEnc->m_encData->m_cuStat[index].vbvCost = frameEnc->m_analysisData.lookahead.vbvCost[index];
+                    for (uint32_t index = 0; index < frameEnc->m_analysisData.numCuInHeight; index++)
+                    {
+                        frameEnc->m_encData->m_rowStat[index].intraSatdForVbv = frameEnc->m_analysisData.lookahead.intraSatdForVbv[index];
+                        frameEnc->m_encData->m_rowStat[index].satdForVbv = frameEnc->m_analysisData.lookahead.satdForVbv[index];
+                    }
+                    for (uint32_t index = 0; index < frameEnc->m_analysisData.numCUsInFrame; index++)
+                    {
+                        frameEnc->m_encData->m_cuStat[index].intraVbvCost = frameEnc->m_analysisData.lookahead.intraVbvCost[index];
+                        frameEnc->m_encData->m_cuStat[index].vbvCost = frameEnc->m_analysisData.lookahead.vbvCost[index];
+                    }
                 }
             }
             if (m_param->searchMethod == X265_SEA && frameEnc->m_lowres.sliceType != X265_TYPE_B)
@@ -3032,7 +3041,7 @@ void Encoder::allocAnalysis(x265_analysis_data* analysis)
 {
     X265_CHECK(analysis->sliceType, "invalid slice type\n");
     analysis->interData = analysis->intraData = NULL;
-    if (m_param->bDisableLookahead)
+    if (m_param->bDisableLookahead && m_rateControl->m_isVbv)
     {
         CHECKED_MALLOC_ZERO(analysis->lookahead.intraSatdForVbv, uint32_t, analysis->numCuInHeight);
         CHECKED_MALLOC_ZERO(analysis->lookahead.satdForVbv, uint32_t, analysis->numCuInHeight);
@@ -3103,10 +3112,9 @@ void Encoder::allocAnalysis(x265_analysis_data* analysis)
     freeAnalysis(analysis);
     m_aborted = true;
 }
-
 void Encoder::freeAnalysis(x265_analysis_data* analysis)
 {
-    if (m_param->bDisableLookahead)
+    if (m_param->bDisableLookahead && m_rateControl->m_isVbv)
     {
         X265_FREE(analysis->lookahead.satdForVbv);
         X265_FREE(analysis->lookahead.intraSatdForVbv);
@@ -3309,10 +3317,9 @@ void Encoder::readAnalysisFile(x265_analysis_data* analysis, int curPoc, const x
 
     if (m_param->scaleFactor)
         analysis->numPartitions *= factor;
-
     /* Memory is allocated for inter and intra analysis data based on the slicetype */
     allocAnalysis(analysis);
-    if (m_param->bDisableLookahead)
+    if (m_param->bDisableLookahead && m_rateControl->m_isVbv)
     {
         X265_FREAD(analysis->lookahead.intraVbvCost, sizeof(uint32_t), analysis->numCUsInFrame, m_analysisFile, picData->lookahead.intraVbvCost);
         X265_FREAD(analysis->lookahead.vbvCost, sizeof(uint32_t), analysis->numCUsInFrame, m_analysisFile, picData->lookahead.vbvCost);

From 0d0adbf290dd3945e8f75284d3a3c93c46476b30 Mon Sep 17 00:00:00 2001
From: Aruna Matheswaran <aruna@multicorewareinc.com>
Date: Thu, 21 Dec 2017 15:26:55 +0530
Subject: [PATCH 21/51] analysis: Enable analysis save and load in a single
 run.

This patch does the following,

1. Deprecates the cli option --analysis-reuse-mode and introduces two new cli
options --analysis-save <filename> and --analysis-load <filename>.

2. Deprecates the param analysisReuseMode and introduces two new param options
analysisSave and analysisLoad.
---
 doc/reST/cli.rst                 |  21 ++--
 source/CMakeLists.txt            |   2 +-
 source/common/cudata.cpp         |   4 +-
 source/common/param.cpp          |  15 ++-
 source/encoder/analysis.cpp      |  26 ++--
 source/encoder/api.cpp           |   2 +-
 source/encoder/encoder.cpp       | 199 ++++++++++++++++---------------
 source/encoder/encoder.h         |   1 -
 source/encoder/frameencoder.cpp  |  12 +-
 source/encoder/search.cpp        |   6 +-
 source/encoder/slicetype.cpp     |  12 +-
 source/test/regression-tests.txt |  17 +--
 source/x265.cpp                  |   4 +-
 source/x265.h                    |  36 +++---
 source/x265cli.h                 |  13 +-
 15 files changed, 197 insertions(+), 173 deletions(-)

diff --git a/doc/reST/cli.rst b/doc/reST/cli.rst
index cd6bb41cff..4a60d8aa94 100644
--- a/doc/reST/cli.rst
+++ b/doc/reST/cli.rst
@@ -863,21 +863,22 @@ Analysis re-use options, to improve performance when encoding the same
 sequence multiple times (presumably at varying bitrates). The encoder
 will not reuse analysis if slice type parameters do not match.
 
-.. option:: --analysis-reuse-mode <string|int>
+.. option:: --analysis-save <filename>
 
-	This option allows reuse of analysis information from first pass to second pass.
-	:option:`--analysis-reuse-mode save` specifies that encoder outputs analysis information of each frame.
-	:option:`--analysis-reuse-mode load` specifies that encoder reuses analysis information from first pass.
-	There is no benefit using load mode without running encoder in save mode. Analysis data from save mode is
-	written to a file specified by :option:`--analysis-reuse-file`. The amount of analysis data stored/reused
-	is determined by :option:`--analysis-reuse-level`. By reading the analysis data writen by an earlier encode
-	of the same sequence, substantial redundant work may be avoided. Requires cutree, pmode to be off. Default 0.
+	Encoder outputs analysis information of each frame. Analysis data from save mode is
+	written to the file specified. Requires cutree, pmode to be off. Default disabled.
+	
+.. option:: --analysis-load <filename>
+
+	Encoder reuses analysis information from the file specified. By reading the analysis data writen by
+	an earlier encode of the same sequence, substantial redundant work may be avoided. Requires cutree, pmode
+	to be off. Default disabled.
 
-	**Values:** off(0), save(1): dump analysis data, load(2): read analysis data
+	The amount of analysis data stored/reused is determined by :option:`--analysis-reuse-level`.
 
 .. option:: --analysis-reuse-file <filename>
 
-	Specify a filename for analysis data (see :option:`--analysis-reuse-mode`)
+	Specify a filename for `multi-pass-opt-analysis` and `multi-pass-opt-distortion`.
 	If no filename is specified, x265_analysis.dat is used.
 
 .. option:: --analysis-reuse-level <1..10>
diff --git a/source/CMakeLists.txt b/source/CMakeLists.txt
index c859ff4cc0..d7334e055f 100644
--- a/source/CMakeLists.txt
+++ b/source/CMakeLists.txt
@@ -29,7 +29,7 @@ option(NATIVE_BUILD "Target the build CPU" OFF)
 option(STATIC_LINK_CRT "Statically link C runtime for release builds" OFF)
 mark_as_advanced(FPROFILE_USE FPROFILE_GENERATE NATIVE_BUILD)
 # X265_BUILD must be incremented each time the public API is changed
-set(X265_BUILD 148)
+set(X265_BUILD 149)
 configure_file("${PROJECT_SOURCE_DIR}/x265.def.in"
                "${PROJECT_BINARY_DIR}/x265.def")
 configure_file("${PROJECT_SOURCE_DIR}/x265_config.h.in"
diff --git a/source/common/cudata.cpp b/source/common/cudata.cpp
index a519372e4a..eada66cfe1 100644
--- a/source/common/cudata.cpp
+++ b/source/common/cudata.cpp
@@ -1626,7 +1626,7 @@ uint32_t CUData::getInterMergeCandidates(uint32_t absPartIdx, uint32_t puIdx, MV
                 dir |= (1 << list);
                 candMvField[count][list].mv = colmv;
                 candMvField[count][list].refIdx = refIdx;
-                if (m_encData->m_param->scaleFactor && m_encData->m_param->analysisReuseMode == X265_ANALYSIS_SAVE && m_log2CUSize[0] < 4)
+                if (m_encData->m_param->scaleFactor && m_encData->m_param->analysisSave && m_log2CUSize[0] < 4)
                 {
                     MV dist(MAX_MV, MAX_MV);
                     candMvField[count][list].mv = dist;
@@ -1791,7 +1791,7 @@ int CUData::getPMV(InterNeighbourMV *neighbours, uint32_t picList, uint32_t refI
             int curRefPOC = m_slice->m_refPOCList[picList][refIdx];
             int curPOC = m_slice->m_poc;
 
-            if (m_encData->m_param->scaleFactor && m_encData->m_param->analysisReuseMode == X265_ANALYSIS_SAVE && (m_log2CUSize[0] < 4))
+            if (m_encData->m_param->scaleFactor && m_encData->m_param->analysisSave && (m_log2CUSize[0] < 4))
             {
                 MV dist(MAX_MV, MAX_MV);
                 pmv[numMvc++] = amvpCand[num++] = dist;
diff --git a/source/common/param.cpp b/source/common/param.cpp
index 4b037d50f1..f8c08a084e 100644
--- a/source/common/param.cpp
+++ b/source/common/param.cpp
@@ -197,10 +197,12 @@ void x265_param_default(x265_param* param)
     param->rdPenalty = 0;
     param->psyRd = 2.0;
     param->psyRdoq = 0.0;
-    param->analysisReuseMode = 0;
+    param->analysisReuseMode = 0; /*DEPRECATED*/
     param->analysisMultiPassRefine = 0;
     param->analysisMultiPassDistortion = 0;
     param->analysisReuseFileName = NULL;
+    param->analysisSave = NULL;
+    param->analysisLoad = NULL;
     param->bIntraInBFrames = 0;
     param->bLossless = 0;
     param->bCULossless = 0;
@@ -850,7 +852,7 @@ int x265_param_parse(x265_param* p, const char* name, const char* value)
         p->rc.bStrictCbr = atobool(value);
         p->rc.pbFactor = 1.0;
     }
-    OPT("analysis-reuse-mode") p->analysisReuseMode = parseName(value, x265_analysis_names, bError);
+    OPT("analysis-reuse-mode") p->analysisReuseMode = parseName(value, x265_analysis_names, bError); /*DEPRECATED*/
     OPT("sar")
     {
         p->vui.aspectRatioIdc = parseName(value, x265_sar_names, bError);
@@ -1006,6 +1008,8 @@ int x265_param_parse(x265_param* p, const char* name, const char* value)
             }
          }
         OPT("gop-lookahead") p->gopLookahead = atoi(value);
+        OPT("analysis-save") p->analysisSave = strdup(value);
+        OPT("analysis-load") p->analysisLoad = strdup(value);
         else
             return X265_PARAM_BAD_NAME;
     }
@@ -1344,9 +1348,7 @@ int x265_check_params(x265_param* param)
           "Constant QP is incompatible with 2pass");
     CHECK(param->rc.bStrictCbr && (param->rc.bitrate <= 0 || param->rc.vbvBufferSize <=0),
           "Strict-cbr cannot be applied without specifying target bitrate or vbv bufsize");
-    CHECK(param->analysisReuseMode && (param->analysisReuseMode < X265_ANALYSIS_OFF || param->analysisReuseMode > X265_ANALYSIS_LOAD),
-        "Invalid analysis mode. Analysis mode 0: OFF 1: SAVE : 2 LOAD");
-    CHECK(param->analysisReuseMode && (param->analysisReuseLevel < 1 || param->analysisReuseLevel > 10),
+    CHECK((param->analysisSave || param->analysisLoad) && (param->analysisReuseLevel < 1 || param->analysisReuseLevel > 10),
         "Invalid analysis refine level. Value must be between 1 and 10 (inclusive)");
     CHECK(param->scaleFactor > 2, "Invalid scale-factor. Supports factor <= 2");
     CHECK(param->rc.qpMax < QP_MIN || param->rc.qpMax > QP_MAX_MAX,
@@ -1618,7 +1620,6 @@ char *x265_param2string(x265_param* p, int padx, int pady)
     s += sprintf(s, " psy-rd=%.2f", p->psyRd);
     s += sprintf(s, " psy-rdoq=%.2f", p->psyRdoq);
     BOOL(p->bEnableRdRefine, "rd-refine");
-    s += sprintf(s, " analysis-reuse-mode=%d", p->analysisReuseMode);
     BOOL(p->bLossless, "lossless");
     s += sprintf(s, " cbqpoffs=%d", p->cbQpOffset);
     s += sprintf(s, " crqpoffs=%d", p->crQpOffset);
@@ -1716,6 +1717,8 @@ char *x265_param2string(x265_param* p, int padx, int pady)
     BOOL(p->bEmitHDRSEI, "hdr");
     BOOL(p->bHDROpt, "hdr-opt");
     BOOL(p->bDhdr10opt, "dhdr10-opt");
+    s += sprintf(s, " analysis-save=%s", p->analysisSave);
+    s += sprintf(s, " analysis-load=%s", p->analysisLoad);
     s += sprintf(s, " analysis-reuse-level=%d", p->analysisReuseLevel);
     s += sprintf(s, " scale-factor=%d", p->scaleFactor);
     s += sprintf(s, " refine-intra=%d", p->intraRefine);
diff --git a/source/encoder/analysis.cpp b/source/encoder/analysis.cpp
index 6eecc4375e..76b37cb0d8 100644
--- a/source/encoder/analysis.cpp
+++ b/source/encoder/analysis.cpp
@@ -207,11 +207,11 @@ Mode& Analysis::compressCTU(CUData& ctu, Frame& frame, const CUGeom& cuGeom, con
         }
     }
 
-    if (m_param->analysisReuseMode && m_slice->m_sliceType != I_SLICE && m_param->analysisReuseLevel > 1 && m_param->analysisReuseLevel < 10)
+    if ((m_param->analysisSave || m_param->analysisLoad) && m_slice->m_sliceType != I_SLICE && m_param->analysisReuseLevel > 1 && m_param->analysisReuseLevel < 10)
     {
         int numPredDir = m_slice->isInterP() ? 1 : 2;
         m_reuseInterDataCTU = (analysis_inter_data*)m_frame->m_analysisData.interData;
-        m_reuseRef = &m_reuseInterDataCTU->ref[ctu.m_cuAddr * X265_MAX_PRED_MODE_PER_CTU * numPredDir];
+        m_reuseRef = &m_reuseInterDataCTU->ref [ctu.m_cuAddr * X265_MAX_PRED_MODE_PER_CTU * numPredDir];
         m_reuseDepth = &m_reuseInterDataCTU->depth[ctu.m_cuAddr * ctu.m_numPartitions];
         m_reuseModes = &m_reuseInterDataCTU->modes[ctu.m_cuAddr * ctu.m_numPartitions];
         if (m_param->analysisReuseLevel > 4)
@@ -219,7 +219,7 @@ Mode& Analysis::compressCTU(CUData& ctu, Frame& frame, const CUGeom& cuGeom, con
             m_reusePartSize = &m_reuseInterDataCTU->partSize[ctu.m_cuAddr * ctu.m_numPartitions];
             m_reuseMergeFlag = &m_reuseInterDataCTU->mergeFlag[ctu.m_cuAddr * ctu.m_numPartitions];
         }
-        if (m_param->analysisReuseMode == X265_ANALYSIS_SAVE)
+        if (m_param->analysisSave && !m_param->analysisLoad)
             for (int i = 0; i < X265_MAX_PRED_MODE_PER_CTU * numPredDir; i++)
                 m_reuseRef[i] = -1;
     }
@@ -228,7 +228,7 @@ Mode& Analysis::compressCTU(CUData& ctu, Frame& frame, const CUGeom& cuGeom, con
     if (m_slice->m_sliceType == I_SLICE)
     {
         analysis_intra_data* intraDataCTU = (analysis_intra_data*)m_frame->m_analysisData.intraData;
-        if (m_param->analysisReuseMode == X265_ANALYSIS_LOAD && m_param->analysisReuseLevel > 1)
+        if (m_param->analysisLoad && m_param->analysisReuseLevel > 1)
         {
             memcpy(ctu.m_cuDepth, &intraDataCTU->depth[ctu.m_cuAddr * numPartition], sizeof(uint8_t) * numPartition);
             memcpy(ctu.m_lumaIntraDir, &intraDataCTU->modes[ctu.m_cuAddr * numPartition], sizeof(uint8_t) * numPartition);
@@ -239,7 +239,7 @@ Mode& Analysis::compressCTU(CUData& ctu, Frame& frame, const CUGeom& cuGeom, con
     }
     else
     {
-        bool bCopyAnalysis = ((m_param->analysisReuseMode == X265_ANALYSIS_LOAD && m_param->analysisReuseLevel == 10) || (m_param->bMVType && m_param->analysisReuseLevel >= 7 && ctu.m_numPartitions <= 16));
+        bool bCopyAnalysis = ((m_param->analysisLoad && m_param->analysisReuseLevel == 10) || (m_param->bMVType && m_param->analysisReuseLevel >= 7 && ctu.m_numPartitions <= 16));
         bool BCompressInterCUrd0_4 = (m_param->bMVType && m_param->analysisReuseLevel >= 7 && m_param->rdLevel <= 4);
         bool BCompressInterCUrd5_6 = (m_param->bMVType && m_param->analysisReuseLevel >= 7 && m_param->rdLevel >= 5 && m_param->rdLevel <= 6);
         bCopyAnalysis = bCopyAnalysis || BCompressInterCUrd0_4 || BCompressInterCUrd5_6;
@@ -280,7 +280,7 @@ Mode& Analysis::compressCTU(CUData& ctu, Frame& frame, const CUGeom& cuGeom, con
             /* generate residual for entire CTU at once and copy to reconPic */
             encodeResidue(ctu, cuGeom);
         }
-        else if ((m_param->analysisReuseMode == X265_ANALYSIS_LOAD && m_param->analysisReuseLevel == 10) || ((m_param->bMVType == AVC_INFO) && m_param->analysisReuseLevel >= 7 && ctu.m_numPartitions <= 16))
+        else if ((m_param->analysisLoad && m_param->analysisReuseLevel == 10) || ((m_param->bMVType == AVC_INFO) && m_param->analysisReuseLevel >= 7 && ctu.m_numPartitions <= 16))
         {
             analysis_inter_data* interDataCTU = (analysis_inter_data*)m_frame->m_analysisData.interData;
             int posCTU = ctu.m_cuAddr * numPartition;
@@ -651,7 +651,7 @@ uint64_t Analysis::compressIntraCU(const CUData& parentCTU, const CUGeom& cuGeom
     }
 
     /* Save Intra CUs TU depth only when analysis mode is OFF */
-    if ((m_limitTU & X265_TU_LIMIT_NEIGH) && cuGeom.log2CUSize >= 4 && !m_param->analysisReuseMode)
+    if ((m_limitTU & X265_TU_LIMIT_NEIGH) && cuGeom.log2CUSize >= 4 && (!m_param->analysisSave && !m_param->analysisLoad))
     {
         CUData* ctu = md.bestMode->cu.m_encData->getPicCTU(parentCTU.m_cuAddr);
         int8_t maxTUDepth = -1;
@@ -1263,7 +1263,7 @@ SplitData Analysis::compressInterCU_rd0_4(const CUData& parentCTU, const CUGeom&
                 mightSplit &= !bDecidedDepth;
             }
         }
-        if ((m_param->analysisReuseMode == X265_ANALYSIS_LOAD && m_param->analysisReuseLevel > 1 && m_param->analysisReuseLevel != 10))
+        if ((m_param->analysisLoad && m_param->analysisReuseLevel > 1 && m_param->analysisReuseLevel != 10))
         {
             if (mightNotSplit && depth == m_reuseDepth[cuGeom.absPartIdx])
             {
@@ -1957,7 +1957,7 @@ SplitData Analysis::compressInterCU_rd5_6(const CUData& parentCTU, const CUGeom&
                 mightSplit &= !bDecidedDepth;
             }
         }
-        if (m_param->analysisReuseMode == X265_ANALYSIS_LOAD && m_param->analysisReuseLevel > 1 && m_param->analysisReuseLevel != 10)
+        if (m_param->analysisLoad && m_param->analysisReuseLevel > 1 && m_param->analysisReuseLevel != 10)
         {
             if (mightNotSplit && depth == m_reuseDepth[cuGeom.absPartIdx])
             {
@@ -2890,7 +2890,7 @@ void Analysis::checkInter_rd0_4(Mode& interMode, const CUGeom& cuGeom, PartSize
     interMode.cu.setPredModeSubParts(MODE_INTER);
     int numPredDir = m_slice->isInterP() ? 1 : 2;
 
-    if (m_param->analysisReuseMode == X265_ANALYSIS_LOAD && m_reuseInterDataCTU && m_param->analysisReuseLevel > 1 && m_param->analysisReuseLevel != 10)
+    if (m_param->analysisLoad && m_reuseInterDataCTU && m_param->analysisReuseLevel > 1 && m_param->analysisReuseLevel != 10)
     {
         int refOffset = cuGeom.geomRecurId * 16 * numPredDir + partSize * numPredDir * 2;
         int index = 0;
@@ -2932,7 +2932,7 @@ void Analysis::checkInter_rd0_4(Mode& interMode, const CUGeom& cuGeom, PartSize
     }
     interMode.sa8dCost = m_rdCost.calcRdSADCost((uint32_t)interMode.distortion, interMode.sa8dBits);
 
-    if (m_param->analysisReuseMode == X265_ANALYSIS_SAVE && m_reuseInterDataCTU && m_param->analysisReuseLevel > 1)
+    if (m_param->analysisSave && m_reuseInterDataCTU && m_param->analysisReuseLevel > 1)
     {
         int refOffset = cuGeom.geomRecurId * 16 * numPredDir + partSize * numPredDir * 2;
         int index = 0;
@@ -2954,7 +2954,7 @@ void Analysis::checkInter_rd5_6(Mode& interMode, const CUGeom& cuGeom, PartSize
     interMode.cu.setPredModeSubParts(MODE_INTER);
     int numPredDir = m_slice->isInterP() ? 1 : 2;
 
-    if (m_param->analysisReuseMode == X265_ANALYSIS_LOAD && m_reuseInterDataCTU && m_param->analysisReuseLevel > 1 && m_param->analysisReuseLevel != 10)
+    if (m_param->analysisLoad && m_reuseInterDataCTU && m_param->analysisReuseLevel > 1 && m_param->analysisReuseLevel != 10)
     {
         int refOffset = cuGeom.geomRecurId * 16 * numPredDir + partSize * numPredDir * 2;
         int index = 0;
@@ -2988,7 +2988,7 @@ void Analysis::checkInter_rd5_6(Mode& interMode, const CUGeom& cuGeom, PartSize
     /* predInterSearch sets interMode.sa8dBits, but this is ignored */
     encodeResAndCalcRdInterCU(interMode, cuGeom);
 
-    if (m_param->analysisReuseMode == X265_ANALYSIS_SAVE && m_reuseInterDataCTU && m_param->analysisReuseLevel > 1)
+    if (m_param->analysisSave && m_reuseInterDataCTU && m_param->analysisReuseLevel > 1)
     {
         int refOffset = cuGeom.geomRecurId * 16 * numPredDir + partSize * numPredDir * 2;
         int index = 0;
diff --git a/source/encoder/api.cpp b/source/encoder/api.cpp
index 955c941243..21a58c18f2 100644
--- a/source/encoder/api.cpp
+++ b/source/encoder/api.cpp
@@ -398,7 +398,7 @@ void x265_picture_init(x265_param *param, x265_picture *pic)
     pic->userSEI.payloads = NULL;
     pic->userSEI.numPayloads = 0;
 
-    if (param->analysisReuseMode || (param->bMVType == AVC_INFO))
+    if ((param->analysisSave || param->analysisLoad) || (param->bMVType == AVC_INFO))
     {
         uint32_t widthInCU = (param->sourceWidth + param->maxCUSize - 1) >> param->maxLog2CUSize;
         uint32_t heightInCU = (param->sourceHeight + param->maxCUSize - 1) >> param->maxLog2CUSize;
diff --git a/source/encoder/encoder.cpp b/source/encoder/encoder.cpp
index cf7589ba8b..b060289a64 100644
--- a/source/encoder/encoder.cpp
+++ b/source/encoder/encoder.cpp
@@ -77,7 +77,6 @@ Encoder::Encoder()
     m_param = NULL;
     m_latestParam = NULL;
     m_threadPool = NULL;
-    m_analysisFile = NULL;
     m_analysisFileIn = NULL;
     m_analysisFileOut = NULL;
     m_offsetEmergency = NULL;
@@ -343,16 +342,28 @@ void Encoder::create()
     if (!m_lookahead->create())
         m_aborted = true;
     initRefIdx();
-    if (m_param->analysisReuseMode && m_param->bUseAnalysisFile)
+    if (m_param->analysisSave && m_param->bUseAnalysisFile)
     {
-        const char* name = m_param->analysisReuseFileName;
-        if (!name)
-            name = defaultAnalysisFileName;
-        const char* mode = m_param->analysisReuseMode == X265_ANALYSIS_LOAD ? "rb" : "wb";
-        m_analysisFile = x265_fopen(name, mode);
-        if (!m_analysisFile)
+        char* temp = strcatFilename(m_param->analysisSave, ".temp");
+        if (!temp)
+            m_aborted = true;
+        else
+        {
+            m_analysisFileOut = x265_fopen(temp, "wb");
+            X265_FREE(temp);
+        }
+        if (!m_analysisFileOut)
         {
-            x265_log_file(NULL, X265_LOG_ERROR, "Analysis load/save: failed to open file %s\n", name);
+            x265_log_file(NULL, X265_LOG_ERROR, "Analysis save: failed to open file %s.temp\n", m_param->analysisSave);
+            m_aborted = true;
+        }
+    }
+    if (m_param->analysisLoad && m_param->bUseAnalysisFile)
+    {
+        m_analysisFileIn = x265_fopen(m_param->analysisLoad, "rb");
+        if (!m_analysisFileIn)
+        {
+            x265_log_file(NULL, X265_LOG_ERROR, "Analysis load: failed to open file %s\n", m_param->analysisLoad);
             m_aborted = true;
         }
     }
@@ -718,9 +729,6 @@ void Encoder::destroy()
 
     X265_FREE(m_offsetEmergency);
 
-    if (m_analysisFile)
-        fclose(m_analysisFile);
-
     if (m_latestParam != NULL && m_latestParam != m_param)
     {
         if (m_latestParam->scalingLists != m_param->scalingLists)
@@ -735,7 +743,7 @@ void Encoder::destroy()
     {
         int bError = 1;
         fclose(m_analysisFileOut);
-        const char* name = m_param->analysisReuseFileName;
+        const char* name = m_param->analysisSave ? m_param->analysisSave : m_param->analysisReuseFileName;
         if (!name)
             name = defaultAnalysisFileName;
         char* temp = strcatFilename(name, ".temp");
@@ -763,6 +771,8 @@ void Encoder::destroy()
         free((char*)m_param->numaPools);
         free((char*)m_param->masteringDisplayColorVolume);
         free((char*)m_param->toneMapFile);
+        free((char*)m_param->analysisSave);
+        free((char*)m_param->analysisLoad);
         PARAM_NS::x265_param_free(m_param);
     }
 }
@@ -849,7 +859,7 @@ int Encoder::encode(const x265_picture* pic_in, x265_picture* pic_out)
 
     if (m_exportedPic)
     {
-        if (!m_param->bUseAnalysisFile && m_param->analysisReuseMode == X265_ANALYSIS_SAVE)
+        if (!m_param->bUseAnalysisFile && m_param->analysisSave)
             freeAnalysis(&m_exportedPic->m_analysisData);
         ATOMIC_DEC(&m_exportedPic->m_countRefEncoders);
         m_exportedPic = NULL;
@@ -1034,7 +1044,7 @@ int Encoder::encode(const x265_picture* pic_in, x265_picture* pic_out)
 
         /* In analysisSave mode, x265_analysis_data is allocated in pic_in and inFrame points to this */
         /* Load analysis data before lookahead->addPicture, since sliceType has been decided */
-        if (m_param->analysisReuseMode == X265_ANALYSIS_LOAD)
+        if (m_param->analysisLoad)
         {
             /* readAnalysisFile reads analysis data for the frame and allocates memory based on slicetype */
             readAnalysisFile(&inFrame->m_analysisData, inFrame->m_poc, pic_in);
@@ -1119,7 +1129,7 @@ int Encoder::encode(const x265_picture* pic_in, x265_picture* pic_out)
             x265_frame_stats* frameData = NULL;
 
             /* Free up pic_in->analysisData since it has already been used */
-            if (m_param->analysisReuseMode == X265_ANALYSIS_LOAD || (m_param->bMVType && slice->m_sliceType != I_SLICE))
+            if ((m_param->analysisLoad && !m_param->analysisSave) || (m_param->bMVType && slice->m_sliceType != I_SLICE))
                 freeAnalysis(&outFrame->m_analysisData);
 
             if (pic_out)
@@ -1145,7 +1155,7 @@ int Encoder::encode(const x265_picture* pic_in, x265_picture* pic_out)
                 }
 
                 /* Dump analysis data from pic_out to file in save mode and free */
-                if (m_param->analysisReuseMode == X265_ANALYSIS_SAVE)
+                if (m_param->analysisSave)
                 {
                     pic_out->analysisData.poc = pic_out->poc;
                     pic_out->analysisData.sliceType = pic_out->sliceType;
@@ -1354,7 +1364,7 @@ int Encoder::encode(const x265_picture* pic_in, x265_picture* pic_out)
                 slice->m_maxNumMergeCand = m_param->maxNumMergeCand;
                 slice->m_endCUAddr = slice->realEndAddress(m_sps.numCUsInFrame * m_param->num4x4Partitions);
             }
-            if (m_param->analysisReuseMode == X265_ANALYSIS_LOAD && m_param->bDisableLookahead)
+            if (m_param->analysisLoad && m_param->bDisableLookahead)
             {
                 frameEnc->m_dts = frameEnc->m_analysisData.lookahead.dts;
                 if (m_rateControl->m_isVbv)
@@ -1423,7 +1433,7 @@ int Encoder::encode(const x265_picture* pic_in, x265_picture* pic_out)
             frameEnc->m_encData->m_slice->m_iNumRPSInSPS = m_sps.spsrpsNum;
 
             curEncoder->m_rce.encodeOrder = frameEnc->m_encodeOrder = m_encodedFrameNum++;
-            if (m_param->analysisReuseMode != X265_ANALYSIS_LOAD || !m_param->bDisableLookahead)
+            if (!m_param->analysisLoad || !m_param->bDisableLookahead)
             {
                 if (m_bframeDelay)
                 {
@@ -1438,7 +1448,7 @@ int Encoder::encode(const x265_picture* pic_in, x265_picture* pic_out)
             }
 
             /* Allocate analysis data before encode in save mode. This is allocated in frameEnc */
-            if (m_param->analysisReuseMode == X265_ANALYSIS_SAVE)
+            if (m_param->analysisSave && !m_param->analysisLoad)
             {
                 x265_analysis_data* analysis = &frameEnc->m_analysisData;
                 analysis->poc = frameEnc->m_poc;
@@ -2623,23 +2633,24 @@ void Encoder::configure(x265_param *p)
         p->rc.rfConstantMin = 0;
     }
 
-    if (p->analysisReuseMode && (p->bDistributeModeAnalysis || p->bDistributeMotionEstimation))
+    if ((p->analysisLoad || p->analysisSave) && (p->bDistributeModeAnalysis || p->bDistributeMotionEstimation))
     {
         x265_log(p, X265_LOG_WARNING, "Analysis load/save options incompatible with pmode/pme, Disabling pmode/pme\n");
         p->bDistributeMotionEstimation = p->bDistributeModeAnalysis = 0;
     }
 
-    if (p->analysisReuseMode && p->rc.cuTree)
+    if ((p->analysisLoad || p->analysisSave) && p->rc.cuTree)
     {
         x265_log(p, X265_LOG_WARNING, "Analysis load/save options works only with cu-tree off, Disabling cu-tree\n");
         p->rc.cuTree = 0;
     }
 
-    if (p->analysisReuseMode && (p->analysisMultiPassRefine || p->analysisMultiPassDistortion))
+    if ((p->analysisLoad || p->analysisSave) && (p->analysisMultiPassRefine || p->analysisMultiPassDistortion))
     {
         x265_log(p, X265_LOG_WARNING, "Cannot use Analysis load/save option and multi-pass-opt-analysis/multi-pass-opt-distortion together,"
             "Disabling Analysis load/save and multi-pass-opt-analysis/multi-pass-opt-distortion\n");
-        p->analysisReuseMode = p->analysisMultiPassRefine = p->analysisMultiPassDistortion = 0;
+        p->analysisSave = p->analysisLoad = NULL;
+        p->analysisMultiPassRefine = p->analysisMultiPassDistortion = 0;
     }
     if (p->scaleFactor)
     {
@@ -2647,16 +2658,16 @@ void Encoder::configure(x265_param *p)
         {
             p->scaleFactor = 0;
         }
-        else if (!p->analysisReuseMode || p->analysisReuseLevel < 10)
+        else if ((!p->analysisLoad && !p->analysisSave) || p->analysisReuseLevel < 10)
         {
-            x265_log(p, X265_LOG_WARNING, "Input scaling works with analysis-reuse-mode, analysis-reuse-level 10. Disabling scale-factor.\n");
+            x265_log(p, X265_LOG_WARNING, "Input scaling works with analysis load/save, analysis-reuse-level 10. Disabling scale-factor.\n");
             p->scaleFactor = 0;
         }
     }
 
     if (p->intraRefine)
     {
-        if (p->analysisReuseMode!= X265_ANALYSIS_LOAD || p->analysisReuseLevel < 10 || !p->scaleFactor)
+        if (!p->analysisLoad || p->analysisReuseLevel < 10 || !p->scaleFactor)
         {
             x265_log(p, X265_LOG_WARNING, "Intra refinement requires analysis load, analysis-reuse-level 10, scale factor. Disabling intra refine.\n");
             p->intraRefine = 0;
@@ -2665,7 +2676,7 @@ void Encoder::configure(x265_param *p)
 
     if (p->interRefine)
     {
-        if (p->analysisReuseMode != X265_ANALYSIS_LOAD || p->analysisReuseLevel < 10 || !p->scaleFactor)
+        if (!p->analysisLoad || p->analysisReuseLevel < 10 || !p->scaleFactor)
         {
             x265_log(p, X265_LOG_WARNING, "Inter refinement requires analysis load, analysis-reuse-level 10, scale factor. Disabling inter refine.\n");
             p->interRefine = 0;
@@ -2680,7 +2691,7 @@ void Encoder::configure(x265_param *p)
 
     if (p->mvRefine)
     {
-        if (p->analysisReuseMode != X265_ANALYSIS_LOAD || p->analysisReuseLevel < 10 || !p->scaleFactor)
+        if (!p->analysisLoad || p->analysisReuseLevel < 10 || !p->scaleFactor)
         {
             x265_log(p, X265_LOG_WARNING, "MV refinement requires analysis load, analysis-reuse-level 10, scale factor. Disabling MV refine.\n");
             p->mvRefine = 0;
@@ -2781,7 +2792,7 @@ void Encoder::configure(x265_param *p)
     m_conformanceWindow.bottomOffset = 0;
     m_conformanceWindow.leftOffset = 0;
     /* set pad size if width is not multiple of the minimum CU size */
-    if (p->scaleFactor == 2 && ((p->sourceWidth / 2) & (p->minCUSize - 1)) && p->analysisReuseMode == X265_ANALYSIS_LOAD)
+    if (p->scaleFactor == 2 && ((p->sourceWidth / 2) & (p->minCUSize - 1)) && p->analysisLoad)
     {
         uint32_t rem = (p->sourceWidth / 2) & (p->minCUSize - 1);
         uint32_t padsize = p->minCUSize - rem;
@@ -2970,7 +2981,7 @@ void Encoder::configure(x265_param *p)
         }
     }
     /* set pad size if height is not multiple of the minimum CU size */
-    if (p->scaleFactor == 2 && ((p->sourceHeight / 2) & (p->minCUSize - 1)) && p->analysisReuseMode == X265_ANALYSIS_LOAD)
+    if (p->scaleFactor == 2 && ((p->sourceHeight / 2) & (p->minCUSize - 1)) && p->analysisLoad)
     {
         uint32_t rem = (p->sourceHeight / 2) & (p->minCUSize - 1);
         uint32_t padsize = p->minCUSize - rem;
@@ -3268,30 +3279,30 @@ void Encoder::readAnalysisFile(x265_analysis_data* analysis, int curPoc, const x
     static uint64_t totalConsumedBytes = 0;
     uint32_t depthBytes = 0;
     if (m_param->bUseAnalysisFile)
-        fseeko(m_analysisFile, totalConsumedBytes, SEEK_SET);
+        fseeko(m_analysisFileIn, totalConsumedBytes, SEEK_SET);
     const x265_analysis_data *picData = &(picIn->analysisData);
     analysis_intra_data *intraPic = (analysis_intra_data *)picData->intraData;
     analysis_inter_data *interPic = (analysis_inter_data *)picData->interData;
 
     int poc; uint32_t frameRecordSize;
-    X265_FREAD(&frameRecordSize, sizeof(uint32_t), 1, m_analysisFile, &(picData->frameRecordSize));
-    X265_FREAD(&depthBytes, sizeof(uint32_t), 1, m_analysisFile, &(picData->depthBytes));
-    X265_FREAD(&poc, sizeof(int), 1, m_analysisFile, &(picData->poc));
+    X265_FREAD(&frameRecordSize, sizeof(uint32_t), 1, m_analysisFileIn, &(picData->frameRecordSize));
+    X265_FREAD(&depthBytes, sizeof(uint32_t), 1, m_analysisFileIn, &(picData->depthBytes));
+    X265_FREAD(&poc, sizeof(int), 1, m_analysisFileIn, &(picData->poc));
 
     if (m_param->bUseAnalysisFile)
     {
         uint64_t currentOffset = totalConsumedBytes;
 
         /* Seeking to the right frame Record */
-        while (poc != curPoc && !feof(m_analysisFile))
+        while (poc != curPoc && !feof(m_analysisFileIn))
         {
             currentOffset += frameRecordSize;
-            fseeko(m_analysisFile, currentOffset, SEEK_SET);
-            X265_FREAD(&frameRecordSize, sizeof(uint32_t), 1, m_analysisFile, &(picData->frameRecordSize));
-            X265_FREAD(&depthBytes, sizeof(uint32_t), 1, m_analysisFile, &(picData->depthBytes));
-            X265_FREAD(&poc, sizeof(int), 1, m_analysisFile, &(picData->poc));
+            fseeko(m_analysisFileIn, currentOffset, SEEK_SET);
+            X265_FREAD(&frameRecordSize, sizeof(uint32_t), 1, m_analysisFileIn, &(picData->frameRecordSize));
+            X265_FREAD(&depthBytes, sizeof(uint32_t), 1, m_analysisFileIn, &(picData->depthBytes));
+            X265_FREAD(&poc, sizeof(int), 1, m_analysisFileIn, &(picData->poc));
         }
-        if (poc != curPoc || feof(m_analysisFile))
+        if (poc != curPoc || feof(m_analysisFileIn))
         {
             x265_log(NULL, X265_LOG_WARNING, "Error reading analysis data: Cannot find POC %d\n", curPoc);
             freeAnalysis(analysis);
@@ -3302,15 +3313,15 @@ void Encoder::readAnalysisFile(x265_analysis_data* analysis, int curPoc, const x
     /* Now arrived at the right frame, read the record */
     analysis->poc = poc;
     analysis->frameRecordSize = frameRecordSize;
-    X265_FREAD(&analysis->sliceType, sizeof(int), 1, m_analysisFile, &(picData->sliceType));
-    X265_FREAD(&analysis->bScenecut, sizeof(int), 1, m_analysisFile, &(picData->bScenecut));
-    X265_FREAD(&analysis->satdCost, sizeof(int64_t), 1, m_analysisFile, &(picData->satdCost));
-    X265_FREAD(&analysis->numCUsInFrame, sizeof(int), 1, m_analysisFile, &(picData->numCUsInFrame));
-    X265_FREAD(&analysis->numPartitions, sizeof(int), 1, m_analysisFile, &(picData->numPartitions));
+    X265_FREAD(&analysis->sliceType, sizeof(int), 1, m_analysisFileIn, &(picData->sliceType));
+    X265_FREAD(&analysis->bScenecut, sizeof(int), 1, m_analysisFileIn, &(picData->bScenecut));
+    X265_FREAD(&analysis->satdCost, sizeof(int64_t), 1, m_analysisFileIn, &(picData->satdCost));
+    X265_FREAD(&analysis->numCUsInFrame, sizeof(int), 1, m_analysisFileIn, &(picData->numCUsInFrame));
+    X265_FREAD(&analysis->numPartitions, sizeof(int), 1, m_analysisFileIn, &(picData->numPartitions));
     if (m_param->bDisableLookahead)
     {
-        X265_FREAD(&analysis->numCuInHeight, sizeof(uint32_t), 1, m_analysisFile, &(picData->numCuInHeight));
-        X265_FREAD(&analysis->lookahead, sizeof(x265_lookahead_data), 1, m_analysisFile, &(picData->lookahead));
+        X265_FREAD(&analysis->numCuInHeight, sizeof(uint32_t), 1, m_analysisFileIn, &(picData->numCuInHeight));
+        X265_FREAD(&analysis->lookahead, sizeof(x265_lookahead_data), 1, m_analysisFileIn, &(picData->lookahead));
     }
     int scaledNumPartition = analysis->numPartitions;
     int factor = 1 << m_param->scaleFactor;
@@ -3321,10 +3332,10 @@ void Encoder::readAnalysisFile(x265_analysis_data* analysis, int curPoc, const x
     allocAnalysis(analysis);
     if (m_param->bDisableLookahead && m_rateControl->m_isVbv)
     {
-        X265_FREAD(analysis->lookahead.intraVbvCost, sizeof(uint32_t), analysis->numCUsInFrame, m_analysisFile, picData->lookahead.intraVbvCost);
-        X265_FREAD(analysis->lookahead.vbvCost, sizeof(uint32_t), analysis->numCUsInFrame, m_analysisFile, picData->lookahead.vbvCost);
-        X265_FREAD(analysis->lookahead.satdForVbv, sizeof(uint32_t), analysis->numCuInHeight, m_analysisFile, picData->lookahead.satdForVbv);
-        X265_FREAD(analysis->lookahead.intraSatdForVbv, sizeof(uint32_t), analysis->numCuInHeight, m_analysisFile, picData->lookahead.intraSatdForVbv);
+        X265_FREAD(analysis->lookahead.intraVbvCost, sizeof(uint32_t), analysis->numCUsInFrame, m_analysisFileIn, picData->lookahead.intraVbvCost);
+        X265_FREAD(analysis->lookahead.vbvCost, sizeof(uint32_t), analysis->numCUsInFrame, m_analysisFileIn, picData->lookahead.vbvCost);
+        X265_FREAD(analysis->lookahead.satdForVbv, sizeof(uint32_t), analysis->numCuInHeight, m_analysisFileIn, picData->lookahead.satdForVbv);
+        X265_FREAD(analysis->lookahead.intraSatdForVbv, sizeof(uint32_t), analysis->numCuInHeight, m_analysisFileIn, picData->lookahead.intraSatdForVbv);
     }
     if (analysis->sliceType == X265_TYPE_IDR || analysis->sliceType == X265_TYPE_I)
     {
@@ -3338,9 +3349,9 @@ void Encoder::readAnalysisFile(x265_analysis_data* analysis, int curPoc, const x
         modeBuf = tempBuf + depthBytes;
         partSizes = tempBuf + 2 * depthBytes;
 
-        X265_FREAD(depthBuf, sizeof(uint8_t), depthBytes, m_analysisFile, intraPic->depth);
-        X265_FREAD(modeBuf, sizeof(uint8_t), depthBytes, m_analysisFile, intraPic->chromaModes);
-        X265_FREAD(partSizes, sizeof(uint8_t), depthBytes, m_analysisFile, intraPic->partSizes);
+        X265_FREAD(depthBuf, sizeof(uint8_t), depthBytes, m_analysisFileIn, intraPic->depth);
+        X265_FREAD(modeBuf, sizeof(uint8_t), depthBytes, m_analysisFileIn, intraPic->chromaModes);
+        X265_FREAD(partSizes, sizeof(uint8_t), depthBytes, m_analysisFileIn, intraPic->partSizes);
 
         size_t count = 0;
         for (uint32_t d = 0; d < depthBytes; d++)
@@ -3361,12 +3372,12 @@ void Encoder::readAnalysisFile(x265_analysis_data* analysis, int curPoc, const x
 
         if (!m_param->scaleFactor)
         {
-            X265_FREAD(((analysis_intra_data *)analysis->intraData)->modes, sizeof(uint8_t), analysis->numCUsInFrame * analysis->numPartitions, m_analysisFile, intraPic->modes);
+            X265_FREAD(((analysis_intra_data *)analysis->intraData)->modes, sizeof(uint8_t), analysis->numCUsInFrame * analysis->numPartitions, m_analysisFileIn, intraPic->modes);
         }
         else
         {
             uint8_t *tempLumaBuf = X265_MALLOC(uint8_t, analysis->numCUsInFrame * scaledNumPartition);
-            X265_FREAD(tempLumaBuf, sizeof(uint8_t), analysis->numCUsInFrame * scaledNumPartition, m_analysisFile, intraPic->modes);
+            X265_FREAD(tempLumaBuf, sizeof(uint8_t), analysis->numCUsInFrame * scaledNumPartition, m_analysisFileIn, intraPic->modes);
             for (uint32_t ctu32Idx = 0, cnt = 0; ctu32Idx < analysis->numCUsInFrame * scaledNumPartition; ctu32Idx++, cnt += factor)
                 memset(&((analysis_intra_data *)analysis->intraData)->modes[cnt], tempLumaBuf[ctu32Idx], factor);
             X265_FREE(tempLumaBuf);
@@ -3379,7 +3390,7 @@ void Encoder::readAnalysisFile(x265_analysis_data* analysis, int curPoc, const x
     {
         uint32_t numDir = analysis->sliceType == X265_TYPE_P ? 1 : 2;
         uint32_t numPlanes = m_param->internalCsp == X265_CSP_I400 ? 1 : 3;
-        X265_FREAD((WeightParam*)analysis->wt, sizeof(WeightParam), numPlanes * numDir, m_analysisFile, (picIn->analysisData.wt));
+        X265_FREAD((WeightParam*)analysis->wt, sizeof(WeightParam), numPlanes * numDir, m_analysisFileIn, (picIn->analysisData.wt));
         if (m_param->analysisReuseLevel < 2)
             return;
 
@@ -3401,33 +3412,33 @@ void Encoder::readAnalysisFile(x265_analysis_data* analysis, int curPoc, const x
         depthBuf = tempBuf;
         modeBuf = tempBuf + depthBytes;
 
-        X265_FREAD(depthBuf, sizeof(uint8_t), depthBytes, m_analysisFile, interPic->depth);
-        X265_FREAD(modeBuf, sizeof(uint8_t), depthBytes, m_analysisFile, interPic->modes);
+        X265_FREAD(depthBuf, sizeof(uint8_t), depthBytes, m_analysisFileIn, interPic->depth);
+        X265_FREAD(modeBuf, sizeof(uint8_t), depthBytes, m_analysisFileIn, interPic->modes);
 
         if (m_param->analysisReuseLevel > 4)
         {
             partSize = modeBuf + depthBytes;
             mergeFlag = partSize + depthBytes;
-            X265_FREAD(partSize, sizeof(uint8_t), depthBytes, m_analysisFile, interPic->partSize);
-            X265_FREAD(mergeFlag, sizeof(uint8_t), depthBytes, m_analysisFile, interPic->mergeFlag);
+            X265_FREAD(partSize, sizeof(uint8_t), depthBytes, m_analysisFileIn, interPic->partSize);
+            X265_FREAD(mergeFlag, sizeof(uint8_t), depthBytes, m_analysisFileIn, interPic->mergeFlag);
 
             if (m_param->analysisReuseLevel == 10)
             {
                 interDir = mergeFlag + depthBytes;
-                X265_FREAD(interDir, sizeof(uint8_t), depthBytes, m_analysisFile, interPic->interDir);
+                X265_FREAD(interDir, sizeof(uint8_t), depthBytes, m_analysisFileIn, interPic->interDir);
                 if (bIntraInInter)
                 {
                     chromaDir = interDir + depthBytes;
-                    X265_FREAD(chromaDir, sizeof(uint8_t), depthBytes, m_analysisFile, intraPic->chromaModes);
+                    X265_FREAD(chromaDir, sizeof(uint8_t), depthBytes, m_analysisFileIn, intraPic->chromaModes);
                 }
                 for (uint32_t i = 0; i < numDir; i++)
                 {
                     mvpIdx[i] = X265_MALLOC(uint8_t, depthBytes);
                     refIdx[i] = X265_MALLOC(int8_t, depthBytes);
                     mv[i] = X265_MALLOC(MV, depthBytes);
-                    X265_FREAD(mvpIdx[i], sizeof(uint8_t), depthBytes, m_analysisFile, interPic->mvpIdx[i]);
-                    X265_FREAD(refIdx[i], sizeof(int8_t), depthBytes, m_analysisFile, interPic->refIdx[i]);
-                    X265_FREAD(mv[i], sizeof(MV), depthBytes, m_analysisFile, interPic->mv[i]);
+                    X265_FREAD(mvpIdx[i], sizeof(uint8_t), depthBytes, m_analysisFileIn, interPic->mvpIdx[i]);
+                    X265_FREAD(refIdx[i], sizeof(int8_t), depthBytes, m_analysisFileIn, interPic->refIdx[i]);
+                    X265_FREAD(mv[i], sizeof(MV), depthBytes, m_analysisFileIn, interPic->mv[i]);
                 }
             }
         }
@@ -3486,12 +3497,12 @@ void Encoder::readAnalysisFile(x265_analysis_data* analysis, int curPoc, const x
             {
                 if (!m_param->scaleFactor)
                 {
-                    X265_FREAD(((analysis_intra_data *)analysis->intraData)->modes, sizeof(uint8_t), analysis->numCUsInFrame * analysis->numPartitions, m_analysisFile, intraPic->modes);
+                    X265_FREAD(((analysis_intra_data *)analysis->intraData)->modes, sizeof(uint8_t), analysis->numCUsInFrame * analysis->numPartitions, m_analysisFileIn, intraPic->modes);
                 }
                 else
                 {
                     uint8_t *tempLumaBuf = X265_MALLOC(uint8_t, analysis->numCUsInFrame * scaledNumPartition);
-                    X265_FREAD(tempLumaBuf, sizeof(uint8_t), analysis->numCUsInFrame * scaledNumPartition, m_analysisFile, intraPic->modes);
+                    X265_FREAD(tempLumaBuf, sizeof(uint8_t), analysis->numCUsInFrame * scaledNumPartition, m_analysisFileIn, intraPic->modes);
                     for (uint32_t ctu32Idx = 0, cnt = 0; ctu32Idx < analysis->numCUsInFrame * scaledNumPartition; ctu32Idx++, cnt += factor)
                         memset(&((analysis_intra_data *)analysis->intraData)->modes[cnt], tempLumaBuf[ctu32Idx], factor);
                     X265_FREE(tempLumaBuf);
@@ -3499,7 +3510,7 @@ void Encoder::readAnalysisFile(x265_analysis_data* analysis, int curPoc, const x
             }
         }
         else
-            X265_FREAD(((analysis_inter_data *)analysis->interData)->ref, sizeof(int32_t), analysis->numCUsInFrame * X265_MAX_PRED_MODE_PER_CTU * numDir, m_analysisFile, interPic->ref);
+            X265_FREAD(((analysis_inter_data *)analysis->interData)->ref, sizeof(int32_t), analysis->numCUsInFrame * X265_MAX_PRED_MODE_PER_CTU * numDir, m_analysisFileIn, interPic->ref);
 
         consumedBytes += frameRecordSize;
         if (numDir == 1)
@@ -3774,51 +3785,51 @@ void Encoder::writeAnalysisFile(x265_analysis_data* analysis, FrameData &curEncD
     if (!m_param->bUseAnalysisFile)
         return;
 
-    X265_FWRITE(&analysis->frameRecordSize, sizeof(uint32_t), 1, m_analysisFile);
-    X265_FWRITE(&depthBytes, sizeof(uint32_t), 1, m_analysisFile);
-    X265_FWRITE(&analysis->poc, sizeof(int), 1, m_analysisFile);
-    X265_FWRITE(&analysis->sliceType, sizeof(int), 1, m_analysisFile);
-    X265_FWRITE(&analysis->bScenecut, sizeof(int), 1, m_analysisFile);
-    X265_FWRITE(&analysis->satdCost, sizeof(int64_t), 1, m_analysisFile);
-    X265_FWRITE(&analysis->numCUsInFrame, sizeof(int), 1, m_analysisFile);
-    X265_FWRITE(&analysis->numPartitions, sizeof(int), 1, m_analysisFile);
+    X265_FWRITE(&analysis->frameRecordSize, sizeof(uint32_t), 1, m_analysisFileOut);
+    X265_FWRITE(&depthBytes, sizeof(uint32_t), 1, m_analysisFileOut);
+    X265_FWRITE(&analysis->poc, sizeof(int), 1, m_analysisFileOut);
+    X265_FWRITE(&analysis->sliceType, sizeof(int), 1, m_analysisFileOut);
+    X265_FWRITE(&analysis->bScenecut, sizeof(int), 1, m_analysisFileOut);
+    X265_FWRITE(&analysis->satdCost, sizeof(int64_t), 1, m_analysisFileOut);
+    X265_FWRITE(&analysis->numCUsInFrame, sizeof(int), 1, m_analysisFileOut);
+    X265_FWRITE(&analysis->numPartitions, sizeof(int), 1, m_analysisFileOut);
     if (analysis->sliceType > X265_TYPE_I)
-        X265_FWRITE((WeightParam*)analysis->wt, sizeof(WeightParam), numPlanes * numDir, m_analysisFile);
+        X265_FWRITE((WeightParam*)analysis->wt, sizeof(WeightParam), numPlanes * numDir, m_analysisFileOut);
 
     if (m_param->analysisReuseLevel < 2)
         return;
 
     if (analysis->sliceType == X265_TYPE_IDR || analysis->sliceType == X265_TYPE_I)
     {
-        X265_FWRITE(((analysis_intra_data*)analysis->intraData)->depth, sizeof(uint8_t), depthBytes, m_analysisFile);
-        X265_FWRITE(((analysis_intra_data*)analysis->intraData)->chromaModes, sizeof(uint8_t), depthBytes, m_analysisFile);
-        X265_FWRITE(((analysis_intra_data*)analysis->intraData)->partSizes, sizeof(char), depthBytes, m_analysisFile);
-        X265_FWRITE(((analysis_intra_data*)analysis->intraData)->modes, sizeof(uint8_t), analysis->numCUsInFrame * analysis->numPartitions, m_analysisFile);
+        X265_FWRITE(((analysis_intra_data*)analysis->intraData)->depth, sizeof(uint8_t), depthBytes, m_analysisFileOut);
+        X265_FWRITE(((analysis_intra_data*)analysis->intraData)->chromaModes, sizeof(uint8_t), depthBytes, m_analysisFileOut);
+        X265_FWRITE(((analysis_intra_data*)analysis->intraData)->partSizes, sizeof(char), depthBytes, m_analysisFileOut);
+        X265_FWRITE(((analysis_intra_data*)analysis->intraData)->modes, sizeof(uint8_t), analysis->numCUsInFrame * analysis->numPartitions, m_analysisFileOut);
     }
     else
     {
-        X265_FWRITE(((analysis_inter_data*)analysis->interData)->depth, sizeof(uint8_t), depthBytes, m_analysisFile);
-        X265_FWRITE(((analysis_inter_data*)analysis->interData)->modes, sizeof(uint8_t), depthBytes, m_analysisFile);
+        X265_FWRITE(((analysis_inter_data*)analysis->interData)->depth, sizeof(uint8_t), depthBytes, m_analysisFileOut);
+        X265_FWRITE(((analysis_inter_data*)analysis->interData)->modes, sizeof(uint8_t), depthBytes, m_analysisFileOut);
         if (m_param->analysisReuseLevel > 4)
         {
-            X265_FWRITE(((analysis_inter_data*)analysis->interData)->partSize, sizeof(uint8_t), depthBytes, m_analysisFile);
-            X265_FWRITE(((analysis_inter_data*)analysis->interData)->mergeFlag, sizeof(uint8_t), depthBytes, m_analysisFile);
+            X265_FWRITE(((analysis_inter_data*)analysis->interData)->partSize, sizeof(uint8_t), depthBytes, m_analysisFileOut);
+            X265_FWRITE(((analysis_inter_data*)analysis->interData)->mergeFlag, sizeof(uint8_t), depthBytes, m_analysisFileOut);
             if (m_param->analysisReuseLevel == 10)
             {
-                X265_FWRITE(((analysis_inter_data*)analysis->interData)->interDir, sizeof(uint8_t), depthBytes, m_analysisFile);
-                if (bIntraInInter) X265_FWRITE(((analysis_intra_data*)analysis->intraData)->chromaModes, sizeof(uint8_t), depthBytes, m_analysisFile);
+                X265_FWRITE(((analysis_inter_data*)analysis->interData)->interDir, sizeof(uint8_t), depthBytes, m_analysisFileOut);
+                if (bIntraInInter) X265_FWRITE(((analysis_intra_data*)analysis->intraData)->chromaModes, sizeof(uint8_t), depthBytes, m_analysisFileOut);
                 for (uint32_t dir = 0; dir < numDir; dir++)
                 {
-                    X265_FWRITE(((analysis_inter_data*)analysis->interData)->mvpIdx[dir], sizeof(uint8_t), depthBytes, m_analysisFile);
-                    X265_FWRITE(((analysis_inter_data*)analysis->interData)->refIdx[dir], sizeof(int8_t), depthBytes, m_analysisFile);
-                    X265_FWRITE(((analysis_inter_data*)analysis->interData)->mv[dir], sizeof(MV), depthBytes, m_analysisFile);
+                    X265_FWRITE(((analysis_inter_data*)analysis->interData)->mvpIdx[dir], sizeof(uint8_t), depthBytes, m_analysisFileOut);
+                    X265_FWRITE(((analysis_inter_data*)analysis->interData)->refIdx[dir], sizeof(int8_t), depthBytes, m_analysisFileOut);
+                    X265_FWRITE(((analysis_inter_data*)analysis->interData)->mv[dir], sizeof(MV), depthBytes, m_analysisFileOut);
                 }
                 if (bIntraInInter)
-                    X265_FWRITE(((analysis_intra_data*)analysis->intraData)->modes, sizeof(uint8_t), analysis->numCUsInFrame * analysis->numPartitions, m_analysisFile);
+                    X265_FWRITE(((analysis_intra_data*)analysis->intraData)->modes, sizeof(uint8_t), analysis->numCUsInFrame * analysis->numPartitions, m_analysisFileOut);
             }
         }
         if (m_param->analysisReuseLevel != 10)
-            X265_FWRITE(((analysis_inter_data*)analysis->interData)->ref, sizeof(int32_t), analysis->numCUsInFrame * X265_MAX_PRED_MODE_PER_CTU * numDir, m_analysisFile);
+            X265_FWRITE(((analysis_inter_data*)analysis->interData)->ref, sizeof(int32_t), analysis->numCUsInFrame * X265_MAX_PRED_MODE_PER_CTU * numDir, m_analysisFileOut);
 
     }
 #undef X265_FWRITE
diff --git a/source/encoder/encoder.h b/source/encoder/encoder.h
index 5948346e96..4e6c439f06 100644
--- a/source/encoder/encoder.h
+++ b/source/encoder/encoder.h
@@ -130,7 +130,6 @@ class Encoder : public x265_encoder
     FrameEncoder*      m_frameEncoder[X265_MAX_FRAME_THREADS];
     DPB*               m_dpb;
     Frame*             m_exportedPic;
-    FILE*              m_analysisFile;
     FILE*              m_analysisFileIn;
     FILE*              m_analysisFileOut;
     x265_param*        m_param;
diff --git a/source/encoder/frameencoder.cpp b/source/encoder/frameencoder.cpp
index ce24227c70..44093b9339 100644
--- a/source/encoder/frameencoder.cpp
+++ b/source/encoder/frameencoder.cpp
@@ -335,7 +335,7 @@ void FrameEncoder::threadMain()
             while (!m_frame->m_ctuInfo)
                 m_frame->m_copied.wait();
         }
-        if ((m_param->bMVType == AVC_INFO) && !m_param->analysisReuseMode && !(IS_X265_TYPE_I(m_frame->m_lowres.sliceType)))
+        if ((m_param->bMVType == AVC_INFO) && !m_param->analysisSave && !m_param->analysisLoad && !(IS_X265_TYPE_I(m_frame->m_lowres.sliceType)))
         {
             while (((m_frame->m_analysisData.interData == NULL && m_frame->m_analysisData.intraData == NULL) || (uint32_t)m_frame->m_poc != m_frame->m_analysisData.poc))
                 m_frame->m_copyMVType.wait();
@@ -430,7 +430,7 @@ void FrameEncoder::compressFrame()
     bool bUseWeightB = slice->m_sliceType == B_SLICE && slice->m_pps->bUseWeightedBiPred;
 
     WeightParam* reuseWP = NULL;
-    if (m_param->analysisReuseMode && (bUseWeightP || bUseWeightB))
+    if (m_param->analysisLoad && (bUseWeightP || bUseWeightB))
         reuseWP = (WeightParam*)m_frame->m_analysisData.wt;
 
     if (bUseWeightP || bUseWeightB)
@@ -439,7 +439,7 @@ void FrameEncoder::compressFrame()
         m_cuStats.countWeightAnalyze++;
         ScopedElapsedTime time(m_cuStats.weightAnalyzeTime);
 #endif
-        if (m_param->analysisReuseMode == X265_ANALYSIS_LOAD)
+        if (m_param->analysisLoad)
         {
             for (int list = 0; list < slice->isInterB() + 1; list++) 
             {
@@ -466,6 +466,8 @@ void FrameEncoder::compressFrame()
     else
         slice->disableWeights();
 
+    if (m_param->analysisSave && (bUseWeightP || bUseWeightB))
+        reuseWP = (WeightParam*)m_frame->m_analysisData.wt;
     // Generate motion references
     int numPredDir = slice->isInterP() ? 1 : slice->isInterB() ? 2 : 0;
     for (int l = 0; l < numPredDir; l++)
@@ -478,7 +480,7 @@ void FrameEncoder::compressFrame()
             slice->m_refReconPicList[l][ref] = slice->m_refFrameList[l][ref]->m_reconPic;
             m_mref[l][ref].init(slice->m_refReconPicList[l][ref], w, *m_param);
         }
-        if (m_param->analysisReuseMode == X265_ANALYSIS_SAVE && (bUseWeightP || bUseWeightB))
+        if (m_param->analysisSave && (bUseWeightP || bUseWeightB))
         {
             for (int i = 0; i < (m_param->internalCsp != X265_CSP_I400 ? 3 : 1); i++)
                 *(reuseWP++) = slice->m_weightPredTable[l][0][i];
@@ -1411,7 +1413,7 @@ void FrameEncoder::processRowEncoder(int intRow, ThreadLocalData& tld)
             /* TODO: use defines from slicetype.h for lowres block size */
             uint32_t block_y = (ctu->m_cuPelY >> m_param->maxLog2CUSize) * noOfBlocks;
             uint32_t block_x = (ctu->m_cuPelX >> m_param->maxLog2CUSize) * noOfBlocks;
-            if (m_param->analysisReuseMode != X265_ANALYSIS_LOAD || !m_param->bDisableLookahead)
+            if (!m_param->analysisLoad || !m_param->bDisableLookahead)
             {
                 cuStat.vbvCost = 0;
                 cuStat.intraVbvCost = 0;
diff --git a/source/encoder/search.cpp b/source/encoder/search.cpp
index c890e31967..dd38b157a9 100644
--- a/source/encoder/search.cpp
+++ b/source/encoder/search.cpp
@@ -2073,7 +2073,7 @@ void Search::singleMotionEstimation(Search& master, Mode& interMode, const Predi
     int mvpIdx = selectMVP(interMode.cu, pu, amvp, list, ref);
     MV mvmin, mvmax, outmv, mvp = amvp[mvpIdx];
 
-    if (!m_param->analysisReuseMode) /* Prevents load/save outputs from diverging if lowresMV is not available */
+    if (!m_param->analysisSave && !m_param->analysisLoad) /* Prevents load/save outputs from diverging if lowresMV is not available */
     {
         MV lmv = getLowresMV(interMode.cu, pu, list, ref);
         if (lmv.notZero())
@@ -2161,7 +2161,7 @@ void Search::predInterSearch(Mode& interMode, const CUGeom& cuGeom, bool bChroma
         cu.getNeighbourMV(puIdx, pu.puAbsPartIdx, interMode.interNeighbours);
 
         /* Uni-directional prediction */
-        if ((m_param->analysisReuseMode == X265_ANALYSIS_LOAD && m_param->analysisReuseLevel > 1 && m_param->analysisReuseLevel != 10)
+        if ((m_param->analysisLoad && m_param->analysisReuseLevel > 1 && m_param->analysisReuseLevel != 10)
             || (m_param->analysisMultiPassRefine && m_param->rc.bStatRead) || (m_param->bMVType == AVC_INFO))
         {
             for (int list = 0; list < numPredDir; list++)
@@ -2297,7 +2297,7 @@ void Search::predInterSearch(Mode& interMode, const CUGeom& cuGeom, bool bChroma
                     int mvpIdx = selectMVP(cu, pu, amvp, list, ref);
                     MV mvmin, mvmax, outmv, mvp = amvp[mvpIdx];
 
-                    if (!m_param->analysisReuseMode) /* Prevents load/save outputs from diverging when lowresMV is not available */
+                    if (!m_param->analysisSave && !m_param->analysisLoad) /* Prevents load/save outputs from diverging when lowresMV is not available */
                     {
                         MV lmv = getLowresMV(cu, pu, list, ref);
                         if (lmv.notZero())
diff --git a/source/encoder/slicetype.cpp b/source/encoder/slicetype.cpp
index 7129db2d2b..8513269709 100644
--- a/source/encoder/slicetype.cpp
+++ b/source/encoder/slicetype.cpp
@@ -746,7 +746,7 @@ void Lookahead::destroy()
 /* Called by API thread */
 void Lookahead::addPicture(Frame& curFrame, int sliceType)
 {
-    if (m_param->analysisReuseMode == X265_ANALYSIS_LOAD && m_param->bDisableLookahead)
+    if (m_param->analysisLoad && m_param->bDisableLookahead)
     {
         if (!m_filled)
             m_filled = true;
@@ -847,7 +847,7 @@ Frame* Lookahead::getDecidedPicture()
             return out;
         }
 
-        if (m_param->analysisReuseMode == X265_ANALYSIS_LOAD && m_param->bDisableLookahead)
+        if (m_param->analysisLoad && m_param->bDisableLookahead)
             return NULL;
 
         findJob(-1); /* run slicetypeDecide() if necessary */
@@ -906,13 +906,13 @@ void Lookahead::getEstimatedPictureCost(Frame *curFrame)
     default:
         return;
     }
-    if (m_param->analysisReuseMode != X265_ANALYSIS_LOAD || !m_param->bDisableLookahead)
+    if (!m_param->analysisLoad || !m_param->bDisableLookahead)
     {
         X265_CHECK(curFrame->m_lowres.costEst[b - p0][p1 - b] > 0, "Slice cost not estimated\n")
         if (m_param->rc.cuTree && !m_param->rc.bStatRead)
             /* update row satds based on cutree offsets */
             curFrame->m_lowres.satdCost = frameCostRecalculate(frames, p0, p1, b);
-        else if (m_param->analysisReuseMode != X265_ANALYSIS_LOAD || m_param->scaleFactor)
+        else if (!m_param->analysisLoad || m_param->scaleFactor)
         {
             if (m_param->rc.aqMode)
                 curFrame->m_lowres.satdCost = curFrame->m_lowres.costEstAq[b - p0][p1 - b];
@@ -1056,7 +1056,7 @@ void Lookahead::slicetypeDecide()
     {
         slicetypeAnalyse(frames, false);
         bool bIsVbv = m_param->rc.vbvBufferSize > 0 && m_param->rc.vbvMaxBitrate > 0;
-        if (m_param->analysisReuseMode == X265_ANALYSIS_LOAD && m_param->scaleFactor && bIsVbv)
+        if (m_param->analysisLoad && m_param->scaleFactor && bIsVbv)
         {
             int numFrames;
             for (numFrames = 0; numFrames < maxSearch; numFrames++)
@@ -1252,7 +1252,7 @@ void Lookahead::slicetypeDecide()
         frames[j + 1] = NULL;
         slicetypeAnalyse(frames, true);
         bool bIsVbv = m_param->rc.vbvBufferSize > 0 && m_param->rc.vbvMaxBitrate > 0;
-        if (m_param->analysisReuseMode == X265_ANALYSIS_LOAD && m_param->scaleFactor && bIsVbv)
+        if (m_param->analysisLoad && m_param->scaleFactor && bIsVbv)
         {
             int numFrames;
             for (numFrames = 0; numFrames < maxSearch; numFrames++)
diff --git a/source/test/regression-tests.txt b/source/test/regression-tests.txt
index 48e2d43e32..45cd0f2b1c 100644
--- a/source/test/regression-tests.txt
+++ b/source/test/regression-tests.txt
@@ -18,17 +18,17 @@ BasketballDrive_1920x1080_50.y4m,--preset veryfast --tune zerolatency --no-tempo
 BasketballDrive_1920x1080_50.y4m,--preset faster --aq-strength 2 --merange 190 --slices 3
 BasketballDrive_1920x1080_50.y4m,--preset medium --ctu 16 --max-tu-size 8 --subme 7 --qg-size 16 --cu-lossless --tu-inter-depth 3 --limit-tu 1
 BasketballDrive_1920x1080_50.y4m,--preset medium --keyint -1 --nr-inter 100 -F4 --no-sao
-BasketballDrive_1920x1080_50.y4m,--preset medium --no-cutree --analysis-reuse-mode=save --analysis-reuse-level 2 --bitrate 7000 --limit-modes::--preset medium --no-cutree --analysis-reuse-mode=load --analysis-reuse-level 2 --bitrate 7000 --limit-modes
+BasketballDrive_1920x1080_50.y4m,--preset medium --no-cutree --analysis-save x265_analysis.dat --analysis-reuse-level 2 --bitrate 7000 --limit-modes::--preset medium --no-cutree --analysis-load x265_analysis.dat --analysis-reuse-level 2 --bitrate 7000 --limit-modes
 BasketballDrive_1920x1080_50.y4m,--preset slow --nr-intra 100 -F4 --aq-strength 3 --qg-size 16 --limit-refs 1
 BasketballDrive_1920x1080_50.y4m,--preset slower --lossless --chromaloc 3 --subme 0 --limit-tu 4
-BasketballDrive_1920x1080_50.y4m,--preset slower --no-cutree --analysis-reuse-mode=save --analysis-reuse-level 10 --bitrate 7000 --limit-tu 0::--preset slower --no-cutree --analysis-reuse-mode=load --analysis-reuse-level 10 --bitrate 7000 --limit-tu 0
+BasketballDrive_1920x1080_50.y4m,--preset slower --no-cutree --analysis-save x265_analysis.dat --analysis-reuse-level 10 --bitrate 7000 --limit-tu 0::--preset slower --no-cutree --analysis-load x265_analysis.dat --analysis-reuse-level 10 --bitrate 7000 --limit-tu 0
 BasketballDrive_1920x1080_50.y4m,--preset veryslow --crf 4 --cu-lossless --pmode --limit-refs 1 --aq-mode 3 --limit-tu 3
-BasketballDrive_1920x1080_50.y4m,--preset veryslow --no-cutree --analysis-reuse-mode=save --bitrate 7000 --tskip-fast --limit-tu 4::--preset veryslow --no-cutree --analysis-reuse-mode=load --bitrate 7000  --tskip-fast --limit-tu 4
+BasketballDrive_1920x1080_50.y4m,--preset veryslow --no-cutree --analysis-save x265_analysis.dat --bitrate 7000 --tskip-fast --limit-tu 4::--preset veryslow --no-cutree --analysis-load x265_analysis.dat --bitrate 7000  --tskip-fast --limit-tu 4
 BasketballDrive_1920x1080_50.y4m,--preset veryslow --recon-y4m-exec "ffplay -i pipe:0 -autoexit"
 Coastguard-4k.y4m,--preset ultrafast --recon-y4m-exec "ffplay -i pipe:0 -autoexit"
 Coastguard-4k.y4m,--preset superfast --tune grain --overscan=crop
 Coastguard-4k.y4m,--preset superfast --tune grain --pme --aq-strength 2 --merange 190
-Coastguard-4k.y4m,--preset veryfast --no-cutree --analysis-reuse-mode=save --analysis-reuse-level 1 --bitrate 15000::--preset veryfast --no-cutree --analysis-reuse-mode=load --analysis-reuse-level 1 --bitrate 15000
+Coastguard-4k.y4m,--preset veryfast --no-cutree --analysis-save x265_analysis.dat --analysis-reuse-level 1 --bitrate 15000::--preset veryfast --no-cutree --analysis-load x265_analysis.dat --analysis-reuse-level 1 --bitrate 15000
 Coastguard-4k.y4m,--preset medium --rdoq-level 1 --tune ssim --no-signhide --me umh --slices 2
 Coastguard-4k.y4m,--preset slow --tune psnr --cbqpoffs -1 --crqpoffs 1 --limit-refs 1
 CrowdRun_1920x1080_50_10bit_422.yuv,--preset ultrafast --weightp --tune zerolatency --qg-size 16
@@ -52,7 +52,7 @@ DucksAndLegs_1920x1080_60_10bit_422.yuv,--preset slow --temporal-layers --no-psy
 DucksAndLegs_1920x1080_60_10bit_444.yuv,--preset veryfast --weightp --nr-intra 1000 -F4
 DucksAndLegs_1920x1080_60_10bit_444.yuv,--preset medium --nr-inter 500 -F4 --no-psy-rdoq
 DucksAndLegs_1920x1080_60_10bit_444.yuv,--preset slower --no-weightp --rdoq-level 0 --limit-refs 3 --tu-inter-depth 4 --limit-tu 3
-DucksAndLegs_1920x1080_60_10bit_422.yuv,--preset fast --no-cutree --analysis-reuse-mode=save --bitrate 3000 --early-skip --tu-inter-depth 3 --limit-tu 1::--preset fast --no-cutree --analysis-reuse-mode=load --bitrate 3000 --early-skip --tu-inter-depth 3 --limit-tu 1
+DucksAndLegs_1920x1080_60_10bit_422.yuv,--preset fast --no-cutree --analysis-save x265_analysis.dat --bitrate 3000 --early-skip --tu-inter-depth 3 --limit-tu 1::--preset fast --no-cutree --analysis-load x265_analysis.dat --bitrate 3000 --early-skip --tu-inter-depth 3 --limit-tu 1
 FourPeople_1280x720_60.y4m,--preset superfast --no-wpp --lookahead-slices 2
 FourPeople_1280x720_60.y4m,--preset veryfast --aq-mode 2 --aq-strength 1.5 --qg-size 8
 FourPeople_1280x720_60.y4m,--preset medium --qp 38 --no-psy-rd
@@ -69,8 +69,8 @@ KristenAndSara_1280x720_60.y4m,--preset medium --no-cutree --max-tu-size 16
 KristenAndSara_1280x720_60.y4m,--preset slower --pmode --max-tu-size 8 --limit-refs 0 --limit-modes --limit-tu 1
 NebutaFestival_2560x1600_60_10bit_crop.yuv,--preset superfast --tune psnr
 NebutaFestival_2560x1600_60_10bit_crop.yuv,--preset medium --tune grain --limit-refs 2
-NebutaFestival_2560x1600_60_10bit_crop.yuv,--preset slow --no-cutree --analysis-reuse-mode=save --rd 5 --analysis-reuse-level 10 --bitrate 9000::--preset slow --no-cutree --analysis-reuse-mode=load --rd 5 --analysis-reuse-level 10 --bitrate 9000
-News-4k.y4m,--preset ultrafast --no-cutree --analysis-reuse-mode=save --analysis-reuse-level 2 --bitrate 15000::--preset ultrafast --no-cutree --analysis-reuse-mode=load --analysis-reuse-level 2 --bitrate 15000
+NebutaFestival_2560x1600_60_10bit_crop.yuv,--preset slow --no-cutree --analysis-save x265_analysis.dat --rd 5 --analysis-reuse-level 10 --bitrate 9000::--preset slow --no-cutree --analysis-load x265_analysis.dat --rd 5 --analysis-reuse-level 10 --bitrate 9000
+News-4k.y4m,--preset ultrafast --no-cutree --analysis-save x265_analysis.dat --analysis-reuse-level 2 --bitrate 15000::--preset ultrafast --no-cutree --analysis-load x265_analysis.dat --analysis-reuse-level 2 --bitrate 15000
 News-4k.y4m,--preset superfast --lookahead-slices 6 --aq-mode 0
 News-4k.y4m,--preset superfast --slices 4 --aq-mode 0 
 News-4k.y4m,--preset medium --tune ssim --no-sao --qg-size 16
@@ -125,7 +125,7 @@ old_town_cross_444_720p50.y4m,--preset ultrafast --weightp --min-cu 32
 old_town_cross_444_720p50.y4m,--preset superfast --weightp --min-cu 16 --limit-modes
 old_town_cross_444_720p50.y4m,--preset veryfast --qp 1 --tune ssim
 old_town_cross_444_720p50.y4m,--preset faster --rd 1 --tune zero-latency
-old_town_cross_444_720p50.y4m,--preset fast --no-cutree --analysis-reuse-mode=save --analysis-reuse-level 1 --bitrate 3000 --early-skip::--preset fast --no-cutree --analysis-reuse-mode=load --analysis-reuse-level 1 --bitrate 3000 --early-skip
+old_town_cross_444_720p50.y4m,--preset fast --no-cutree --analysis-save pass1_analysis.dat --analysis-reuse-level 1 --bitrate 3000 --early-skip::--preset fast --no-cutree --analysis-load pass1_analysis.dat --analysis-save pass2_analysis.dat --analysis-reuse-level 1 --bitrate 3000 --early-skip::--preset fast --no-cutree --analysis-load pass2_analysis.dat --analysis-reuse-level 1 --bitrate 3000 --early-skip
 old_town_cross_444_720p50.y4m,--preset medium --keyint -1 --no-weightp --ref 6
 old_town_cross_444_720p50.y4m,--preset slow --rdoq-level 1 --early-skip --ref 7 --no-b-pyramid
 old_town_cross_444_720p50.y4m,--preset slower --crf 4 --cu-lossless
@@ -151,6 +151,7 @@ Kimono1_1920x1080_24_400.yuv,--preset medium --rdoq-level 0 --limit-refs 3 --sli
 Kimono1_1920x1080_24_400.yuv,--preset veryslow --crf 4 --cu-lossless --slices 2 --limit-refs 3 --limit-modes
 Kimono1_1920x1080_24_400.yuv,--preset placebo --ctu 32 --max-tu-size 8 --limit-tu 2
 big_buck_bunny_360p24.y4m, --keyint 60 --min-keyint 40 --gop-lookahead 14
+
 # Main12 intraCost overflow bug test
 720p50_parkrun_ter.y4m,--preset medium
 
diff --git a/source/x265.cpp b/source/x265.cpp
index e9d0bf97bb..b2b31f2ec9 100644
--- a/source/x265.cpp
+++ b/source/x265.cpp
@@ -585,9 +585,9 @@ int main(int argc, char **argv)
 
     x265_picture pic_orig, pic_out;
     x265_picture *pic_in = &pic_orig;
-    /* Allocate recon picture if analysisReuseMode is enabled */
+    /* Allocate recon picture if analysis save/load is enabled */
     std::priority_queue<int64_t>* pts_queue = cliopt.output->needPTS() ? new std::priority_queue<int64_t>() : NULL;
-    x265_picture *pic_recon = (cliopt.recon || !!param->analysisReuseMode || pts_queue || reconPlay || param->csvLogLevel) ? &pic_out : NULL;
+    x265_picture *pic_recon = (cliopt.recon || param->analysisSave || param->analysisLoad || pts_queue || reconPlay || param->csvLogLevel) ? &pic_out : NULL;
     uint32_t inFrameCount = 0;
     uint32_t outFrameCount = 0;
     x265_nal *p_nal;
diff --git a/source/x265.h b/source/x265.h
index 9d6b5d606f..808586d7ea 100644
--- a/source/x265.h
+++ b/source/x265.h
@@ -327,15 +327,15 @@ typedef struct x265_picture
      * to allow the encoder to determine base QP */
     int     forceqp;
 
-    /* If param.analysisReuseMode is X265_ANALYSIS_OFF this field is ignored on input
-     * and output. Else the user must call x265_alloc_analysis_data() to
-     * allocate analysis buffers for every picture passed to the encoder.
+    /* If param.analysisLoad and param.analysisSave are disabled, this field is
+     * ignored on input and output. Else the user must call x265_alloc_analysis_data()
+     * to allocate analysis buffers for every picture passed to the encoder.
      *
-     * On input when param.analysisReuseMode is X265_ANALYSIS_LOAD and analysisData
+     * On input when param.analysisLoad is enabled and analysisData
      * member pointers are valid, the encoder will use the data stored here to
      * reduce encoder work.
      *
-     * On output when param.analysisReuseMode is X265_ANALYSIS_SAVE and analysisData
+     * On output when param.analysisSave is enabled and analysisData
      * member pointers are valid, the encoder will write output analysis into
      * this data structure */
     x265_analysis_data analysisData;
@@ -484,11 +484,6 @@ typedef enum
 
 #define X265_EXTENDED_SAR       255 /* aspect ratio explicitly specified as width:height */
 
-/* Analysis options */
-#define X265_ANALYSIS_OFF  0
-#define X265_ANALYSIS_SAVE 1
-#define X265_ANALYSIS_LOAD 2
-
 typedef struct x265_cli_csp
 {
     int planes;
@@ -1129,13 +1124,13 @@ typedef struct x265_param
      * Default disabled */
     int       bEnableRdRefine;
 
-    /* If X265_ANALYSIS_SAVE, write per-frame analysis information into analysis
-     * buffers.  if X265_ANALYSIS_LOAD, read analysis information into analysis
-     * buffer and use this analysis information to reduce the amount of work
-     * the encoder must perform. Default X265_ANALYSIS_OFF */
+    /* If save, write per-frame analysis information into analysis buffers.
+     * If load, read analysis information into analysis buffer and use this
+     * analysis information to reduce the amount of work the encoder must perform.
+     * Default disabled. Now deprecated*/
     int       analysisReuseMode;
 
-    /* Filename for analysisReuseMode save/load. Default name is "x265_analysis.dat" */
+    /* Filename for multi-pass-opt-analysis/distortion. Default name is "x265_analysis.dat" */
     const char* analysisReuseFileName;
 
     /*== Rate Control ==*/
@@ -1455,7 +1450,7 @@ typedef struct x265_param
     int       bHDROpt;
 
     /* A value between 1 and 10 (both inclusive) determines the level of
-    * information stored/reused in save/load analysis-reuse-mode. Higher the refine
+    * information stored/reused in analysis save/load. Higher the refine
     * level higher the information stored/reused. Default is 5 */
     int       analysisReuseLevel;
 
@@ -1539,7 +1534,16 @@ typedef struct x265_param
     * within this from the gop boundary set by keyint, the GOP will be extented until such a point,
     * otherwise the GOP will be terminated as set by keyint*/
     int       gopLookahead;
+
+    /*Write per-frame analysis information into analysis buffers. Default disabled. */
+    const char* analysisSave;
+
+    /* Read analysis information into analysis buffer and use this analysis information
+     * to reduce the amount of work the encoder must perform. Default disabled. */
+    const char* analysisLoad;
+
 } x265_param;
+
 /* x265_param_alloc:
  *  Allocates an x265_param instance. The returned param structure is not
  *  special in any way, but using this method together with x265_param_free()
diff --git a/source/x265cli.h b/source/x265cli.h
index 3b59cf3a91..5f8a69b9e1 100644
--- a/source/x265cli.h
+++ b/source/x265cli.h
@@ -254,9 +254,11 @@ static const struct option long_options[] =
     { "no-slow-firstpass",    no_argument, NULL, 0 },
     { "multi-pass-opt-rps",   no_argument, NULL, 0 },
     { "no-multi-pass-opt-rps", no_argument, NULL, 0 },
-    { "analysis-reuse-mode",  required_argument, NULL, 0 },
-    { "analysis-reuse-file",  required_argument, NULL, 0 },
+    { "analysis-reuse-mode", required_argument, NULL, 0 }, /* DEPRECATED */
+    { "analysis-reuse-file", required_argument, NULL, 0 },
     { "analysis-reuse-level", required_argument, NULL, 0 },
+    { "analysis-save",  required_argument, NULL, 0 },
+    { "analysis-load",  required_argument, NULL, 0 },
     { "scale-factor",   required_argument, NULL, 0 },
     { "refine-intra",   required_argument, NULL, 0 },
     { "refine-inter",   required_argument, NULL, 0 },
@@ -465,18 +467,19 @@ static void showHelp(x265_param *param)
     H0("   --[no-]analyze-src-pics       Motion estimation uses source frame planes. Default disable\n");
     H0("   --[no-]slow-firstpass         Enable a slow first pass in a multipass rate control mode. Default %s\n", OPT(param->rc.bEnableSlowFirstPass));
     H0("   --[no-]strict-cbr             Enable stricter conditions and tolerance for bitrate deviations in CBR mode. Default %s\n", OPT(param->rc.bStrictCbr));
-    H0("   --analysis-reuse-mode <string|int>  save - Dump analysis info into file, load - Load analysis buffers from the file. Default %d\n", param->analysisReuseMode);
+    H0("   --analysis-save <filename>    Dump analysis info into the specified file. Default Disabled\n");
+    H0("   --analysis-load <filename>    Load analysis buffers from the file specified. Default Disabled\n");
     H0("   --analysis-reuse-file <filename>    Specify file name used for either dumping or reading analysis data. Deault x265_analysis.dat\n");
     H0("   --analysis-reuse-level <1..10>      Level of analysis reuse indicates amount of info stored/reused in save/load mode, 1:least..10:most. Default %d\n", param->analysisReuseLevel);
     H0("   --refine-mv-type <string>     Reuse MV information received through API call. Supported option is avc. Default disabled - %d\n", param->bMVType);
     H0("   --scale-factor <int>          Specify factor by which input video is scaled down for analysis save mode. Default %d\n", param->scaleFactor);
-    H0("   --refine-intra <0..3>         Enable intra refinement for encode that uses analysis-reuse-mode=load.\n"
+    H0("   --refine-intra <0..3>         Enable intra refinement for encode that uses analysis-load.\n"
         "                                    - 0 : Forces both mode and depth from the save encode.\n"
         "                                    - 1 : Functionality of (0) + evaluate all intra modes at min-cu-size's depth when current depth is one smaller than min-cu-size's depth.\n"
         "                                    - 2 : Functionality of (1) + irrespective of size evaluate all angular modes when the save encode decides the best mode as angular.\n"
         "                                    - 3 : Functionality of (1) + irrespective of size evaluate all intra modes.\n"
         "                                Default:%d\n", param->intraRefine);
-    H0("   --refine-inter <0..3>         Enable inter refinement for encode that uses analysis-reuse-mode=load.\n"
+    H0("   --refine-inter <0..3>         Enable inter refinement for encode that uses analysis-load.\n"
         "                                    - 0 : Forces both mode and depth from the save encode.\n"
         "                                    - 1 : Functionality of (0) + evaluate all inter modes at min-cu-size's depth when current depth is one smaller than\n"
         "                                          min-cu-size's depth. When save encode decides the current block as skip(for all sizes) evaluate skip/merge.\n"

From abbe51b956d4c438e8c16bd72738f40c982a364e Mon Sep 17 00:00:00 2001
From: Santhoshini Sekar <santhoshini@multicorewareinc.com>
Date: Fri, 22 Dec 2017 12:48:37 +0530
Subject: [PATCH 22/51] Fix possible NULL pointer dereferencing in cudata init

---
 source/common/framedata.cpp | 11 ++++++-----
 source/encoder/analysis.cpp | 17 +++++++++--------
 2 files changed, 15 insertions(+), 13 deletions(-)

diff --git a/source/common/framedata.cpp b/source/common/framedata.cpp
index 6292b9f628..e0ecd57149 100644
--- a/source/common/framedata.cpp
+++ b/source/common/framedata.cpp
@@ -40,11 +40,12 @@ bool FrameData::create(const x265_param& param, const SPS& sps, int csp)
     m_spsrpsIdx = -1;
     if (param.rc.bStatWrite)
         m_spsrps = const_cast<RPS*>(sps.spsrps);
-
-    m_cuMemPool.create(0, param.internalCsp, sps.numCUsInFrame, param);
-    for (uint32_t ctuAddr = 0; ctuAddr < sps.numCUsInFrame; ctuAddr++)
-        m_picCTU[ctuAddr].initialize(m_cuMemPool, 0, param, ctuAddr);
-
+    bool isallocated = m_cuMemPool.create(0, param.internalCsp, sps.numCUsInFrame, param);
+    if (isallocated)
+        for (uint32_t ctuAddr = 0; ctuAddr < sps.numCUsInFrame; ctuAddr++)
+            m_picCTU[ctuAddr].initialize(m_cuMemPool, 0, param, ctuAddr);
+    else
+        return false;
     CHECKED_MALLOC_ZERO(m_cuStat, RCStatCU, sps.numCUsInFrame);
     CHECKED_MALLOC(m_rowStat, RCStatRow, sps.numCuInHeight);
     reinit(sps);
diff --git a/source/encoder/analysis.cpp b/source/encoder/analysis.cpp
index e3d21bf042..86436c821a 100644
--- a/source/encoder/analysis.cpp
+++ b/source/encoder/analysis.cpp
@@ -100,16 +100,17 @@ bool Analysis::create(ThreadLocalData *tld)
     for (uint32_t depth = 0; depth <= m_param->maxCUDepth; depth++, cuSize >>= 1)
     {
         ModeDepth &md = m_modeDepth[depth];
-
-        md.cuMemPool.create(depth, csp, MAX_PRED_TYPES, *m_param);
+        ok &= md.cuMemPool.create(depth, csp, MAX_PRED_TYPES, *m_param);
         ok &= md.fencYuv.create(cuSize, csp);
-
-        for (int j = 0; j < MAX_PRED_TYPES; j++)
+        if (ok)
         {
-            md.pred[j].cu.initialize(md.cuMemPool, depth, *m_param, j);
-            ok &= md.pred[j].predYuv.create(cuSize, csp);
-            ok &= md.pred[j].reconYuv.create(cuSize, csp);
-            md.pred[j].fencYuv = &md.fencYuv;
+            for (int j = 0; j < MAX_PRED_TYPES; j++)
+            {
+                md.pred[j].cu.initialize(md.cuMemPool, depth, *m_param, j);
+                ok &= md.pred[j].predYuv.create(cuSize, csp);
+                ok &= md.pred[j].reconYuv.create(cuSize, csp);
+                md.pred[j].fencYuv = &md.fencYuv;
+            }
         }
     }
     if (m_param->sourceHeight >= 1080)

From 77c886e3cab825b95a610ebe18cabe61f77d93ed Mon Sep 17 00:00:00 2001
From: Aruna Matheswaran <aruna@multicorewareinc.com>
Date: Fri, 22 Dec 2017 14:50:08 +0530
Subject: [PATCH 23/51] Add support for RADL pictures

---
 doc/reST/cli.rst                    |  9 +++-
 source/CMakeLists.txt               |  2 +-
 source/common/lowres.cpp            |  6 +--
 source/common/lowres.h              |  4 +-
 source/common/param.cpp             |  5 +++
 source/encoder/dpb.cpp              | 18 +++++---
 source/encoder/encoder.cpp          |  6 +++
 source/encoder/search.cpp           |  2 +-
 source/encoder/slicetype.cpp        | 69 ++++++++++++++++++++---------
 source/encoder/weightPrediction.cpp |  2 +-
 source/test/regression-tests.txt    |  1 +
 source/x265.h                       |  3 ++
 source/x265cli.h                    |  2 +
 13 files changed, 92 insertions(+), 37 deletions(-)

diff --git a/doc/reST/cli.rst b/doc/reST/cli.rst
index 4a60d8aa94..74a14d8fa6 100644
--- a/doc/reST/cli.rst
+++ b/doc/reST/cli.rst
@@ -1345,7 +1345,14 @@ Slice decision options
 	This value represents the percentage difference between the inter cost and
 	intra cost of a frame used in scenecut detection. For example, a value of 5 indicates,
 	if the inter cost of a frame is greater than or equal to 95 percent of the intra cost of the frame,
-	then detect this frame as scenecut. Values between 5 and 15 are recommended. Default 5.	
+	then detect this frame as scenecut. Values between 5 and 15 are recommended. Default 5.
+	
+.. option:: --radl <integer>
+	
+	Number of RADL pictures allowed infront of IDR. Requires fixed keyframe interval.
+	Recommended value is 2-3. Default 0 (disabled).
+	
+	**Range of values: Between 0 and `--bframes`
 
 .. option:: --ctu-info <0, 1, 2, 4, 6>
 
diff --git a/source/CMakeLists.txt b/source/CMakeLists.txt
index d7334e055f..e17cca9f56 100644
--- a/source/CMakeLists.txt
+++ b/source/CMakeLists.txt
@@ -29,7 +29,7 @@ option(NATIVE_BUILD "Target the build CPU" OFF)
 option(STATIC_LINK_CRT "Statically link C runtime for release builds" OFF)
 mark_as_advanced(FPROFILE_USE FPROFILE_GENERATE NATIVE_BUILD)
 # X265_BUILD must be incremented each time the public API is changed
-set(X265_BUILD 149)
+set(X265_BUILD 150)
 configure_file("${PROJECT_SOURCE_DIR}/x265.def.in"
                "${PROJECT_BINARY_DIR}/x265.def")
 configure_file("${PROJECT_SOURCE_DIR}/x265_config.h.in"
diff --git a/source/common/lowres.cpp b/source/common/lowres.cpp
index d166f5d25e..e81fc23e72 100644
--- a/source/common/lowres.cpp
+++ b/source/common/lowres.cpp
@@ -89,7 +89,7 @@ bool Lowres::create(PicYuv *origPic, int _bframes, bool bAQEnabled, uint32_t qgS
         }
     }
 
-    for (int i = 0; i < bframes + 1; i++)
+    for (int i = 0; i < bframes + 2; i++)
     {
         CHECKED_MALLOC(lowresMvs[0][i], MV, cuCount);
         CHECKED_MALLOC(lowresMvs[1][i], MV, cuCount);
@@ -118,7 +118,7 @@ void Lowres::destroy()
         }
     }
 
-    for (int i = 0; i < bframes + 1; i++)
+    for (int i = 0; i < bframes + 2; i++)
     {
         X265_FREE(lowresMvs[0][i]);
         X265_FREE(lowresMvs[1][i]);
@@ -152,7 +152,7 @@ void Lowres::init(PicYuv *origPic, int poc)
         for (int x = 0; x < bframes + 2; x++)
             rowSatds[y][x][0] = -1;
 
-    for (int i = 0; i < bframes + 1; i++)
+    for (int i = 0; i < bframes + 2; i++)
     {
         lowresMvs[0][i][0].x = 0x7FFF;
         lowresMvs[1][i][0].x = 0x7FFF;
diff --git a/source/common/lowres.h b/source/common/lowres.h
index 072739e80e..ed7642cfbd 100644
--- a/source/common/lowres.h
+++ b/source/common/lowres.h
@@ -130,8 +130,8 @@ struct Lowres : public ReferencePlanes
     int64_t   satdCost;
     uint16_t* lowresCostForRc;
     uint16_t* lowresCosts[X265_BFRAME_MAX + 2][X265_BFRAME_MAX + 2];
-    int32_t*  lowresMvCosts[2][X265_BFRAME_MAX + 1];
-    MV*       lowresMvs[2][X265_BFRAME_MAX + 1];
+    int32_t*  lowresMvCosts[2][X265_BFRAME_MAX + 2];
+    MV*       lowresMvs[2][X265_BFRAME_MAX + 2];
     uint32_t  maxBlocksInRow;
     uint32_t  maxBlocksInCol;
     uint32_t  maxBlocksInRowFullRes;
diff --git a/source/common/param.cpp b/source/common/param.cpp
index f8c08a084e..ab8f419391 100644
--- a/source/common/param.cpp
+++ b/source/common/param.cpp
@@ -154,6 +154,7 @@ void x265_param_default(x265_param* param)
     param->lookaheadSlices = 8;
     param->lookaheadThreads = 0;
     param->scenecutBias = 5.0;
+    param->radl = 0;
     /* Intra Coding Tools */
     param->bEnableConstrainedIntra = 0;
     param->bEnableStrongIntraSmoothing = 1;
@@ -1010,6 +1011,7 @@ int x265_param_parse(x265_param* p, const char* name, const char* value)
         OPT("gop-lookahead") p->gopLookahead = atoi(value);
         OPT("analysis-save") p->analysisSave = strdup(value);
         OPT("analysis-load") p->analysisLoad = strdup(value);
+        OPT("radl") p->radl = atoi(value);
         else
             return X265_PARAM_BAD_NAME;
     }
@@ -1316,6 +1318,8 @@ int x265_check_params(x265_param* param)
           "scenecutThreshold must be greater than 0");
     CHECK(param->scenecutBias < 0 || 100 < param->scenecutBias,
            "scenecut-bias must be between 0 and 100");
+    CHECK(param->radl < 0 || param->radl > param->bframes,
+          "radl must be between 0 and bframes");
     CHECK(param->rdPenalty < 0 || param->rdPenalty > 2,
           "Valid penalty for 32x32 intra TU in non-I slices. 0:disabled 1:RD-penalty 2:maximum");
     CHECK(param->keyframeMax < -1,
@@ -1575,6 +1579,7 @@ char *x265_param2string(x265_param* p, int padx, int pady)
     s += sprintf(s, " rc-lookahead=%d", p->lookaheadDepth);
     s += sprintf(s, " lookahead-slices=%d", p->lookaheadSlices);
     s += sprintf(s, " scenecut=%d", p->scenecutThreshold);
+    s += sprintf(s, " radl=%d", p->radl);
     BOOL(p->bIntraRefresh, "intra-refresh");
     s += sprintf(s, " ctu=%d", p->maxCUSize);
     s += sprintf(s, " min-cu-size=%d", p->minCUSize);
diff --git a/source/encoder/dpb.cpp b/source/encoder/dpb.cpp
index c225cf3eff..19d41f9cdd 100644
--- a/source/encoder/dpb.cpp
+++ b/source/encoder/dpb.cpp
@@ -181,7 +181,10 @@ void DPB::prepareEncode(Frame *newFrame)
     // Mark pictures in m_piclist as unreferenced if they are not included in RPS
     applyReferencePictureSet(&slice->m_rps, pocCurr);
 
-    slice->m_numRefIdx[0] = X265_MIN(newFrame->m_param->maxNumReferences, slice->m_rps.numberOfNegativePictures); // Ensuring L0 contains just the -ve POC
+    if (slice->m_sliceType != I_SLICE)
+        slice->m_numRefIdx[0] = x265_clip3(1, newFrame->m_param->maxNumReferences, slice->m_rps.numberOfNegativePictures);
+    else
+        slice->m_numRefIdx[0] = X265_MIN(newFrame->m_param->maxNumReferences, slice->m_rps.numberOfNegativePictures); // Ensuring L0 contains just the -ve POC
     slice->m_numRefIdx[1] = X265_MIN(newFrame->m_param->bBPyramid ? 2 : 1, slice->m_rps.numberOfPositivePictures);
     slice->setRefPicList(m_picList);
 
@@ -230,11 +233,14 @@ void DPB::computeRPS(int curPoc, bool isRAP, RPS * rps, unsigned int maxDecPicBu
     {
         if ((iterPic->m_poc != curPoc) && iterPic->m_encData->m_bHasReferences)
         {
-            rps->poc[poci] = iterPic->m_poc;
-            rps->deltaPOC[poci] = rps->poc[poci] - curPoc;
-            (rps->deltaPOC[poci] < 0) ? numNeg++ : numPos++;
-            rps->bUsed[poci] = !isRAP;
-            poci++;
+            if ((m_lastIDR >= curPoc) || (m_lastIDR <= iterPic->m_poc))
+            {
+                    rps->poc[poci] = iterPic->m_poc;
+                    rps->deltaPOC[poci] = rps->poc[poci] - curPoc;
+                    (rps->deltaPOC[poci] < 0) ? numNeg++ : numPos++;
+                    rps->bUsed[poci] = !isRAP;
+                    poci++;
+            }
         }
         iterPic = iterPic->m_next;
     }
diff --git a/source/encoder/encoder.cpp b/source/encoder/encoder.cpp
index b060289a64..825e7b3f7d 100644
--- a/source/encoder/encoder.cpp
+++ b/source/encoder/encoder.cpp
@@ -3046,6 +3046,12 @@ void Encoder::configure(x265_param *p)
     p->maxCUDepth    = p->maxLog2CUSize - g_log2Size[p->minCUSize];
     p->unitSizeDepth = p->maxLog2CUSize - LOG2_UNIT_SIZE;
     p->num4x4Partitions = (1U << (p->unitSizeDepth << 1));
+
+    if (p->radl && (p->keyframeMax != p->keyframeMin))
+    {
+        p->radl = 0;
+        x265_log(p, X265_LOG_WARNING, "Radl requires fixed gop-length (keyint == min-keyint). Disabling radl.\n");
+    }
 }
 
 void Encoder::allocAnalysis(x265_analysis_data* analysis)
diff --git a/source/encoder/search.cpp b/source/encoder/search.cpp
index dd38b157a9..d848c9a60d 100644
--- a/source/encoder/search.cpp
+++ b/source/encoder/search.cpp
@@ -1947,7 +1947,7 @@ MV Search::getLowresMV(const CUData& cu, const PredictionUnit& pu, int list, int
         /* poc difference is out of range for lookahead */
         return 0;
 
-    MV* mvs = m_frame->m_lowres.lowresMvs[list][diffPoc - 1];
+    MV* mvs = m_frame->m_lowres.lowresMvs[list][diffPoc];
     if (mvs[0].x == 0x7FFF)
         /* this motion search was not estimated by lookahead */
         return 0;
diff --git a/source/encoder/slicetype.cpp b/source/encoder/slicetype.cpp
index 8513269709..9f73a3e211 100644
--- a/source/encoder/slicetype.cpp
+++ b/source/encoder/slicetype.cpp
@@ -879,7 +879,7 @@ void Lookahead::getEstimatedPictureCost(Frame *curFrame)
     Slice *slice = curFrame->m_encData->m_slice;
     int p0 = 0, p1, b;
     int poc = slice->m_poc;
-    int l0poc = slice->m_refPOCList[0][0];
+    int l0poc = slice->m_rps.numberOfNegativePictures ? slice->m_refPOCList[0][0] : -1;
     int l1poc = slice->m_refPOCList[1][0];
 
     switch (slice->m_sliceType)
@@ -896,11 +896,22 @@ void Lookahead::getEstimatedPictureCost(Frame *curFrame)
         break;
 
     case B_SLICE:
-        b = poc - l0poc;
-        p1 = b + l1poc - poc;
-        frames[p0] = &slice->m_refFrameList[0][0]->m_lowres;
-        frames[b] = &curFrame->m_lowres;
-        frames[p1] = &slice->m_refFrameList[1][0]->m_lowres;
+        if (l0poc >= 0)
+        {
+            b = poc - l0poc;
+            p1 = b + l1poc - poc;
+            frames[p0] = &slice->m_refFrameList[0][0]->m_lowres;
+            frames[b] = &curFrame->m_lowres;
+            frames[p1] = &slice->m_refFrameList[1][0]->m_lowres;
+        }
+        else 
+        {
+            p0 = b = 0;
+            p1 = b + l1poc - poc;
+            frames[p0] = frames[b] = &curFrame->m_lowres;
+            frames[p1] = &slice->m_refFrameList[1][0]->m_lowres;
+        }
+        
         break;
 
     default:
@@ -1120,12 +1131,20 @@ void Lookahead::slicetypeDecide()
             /* Closed GOP */
             m_lastKeyframe = frm.frameNum;
             frm.bKeyframe = true;
-            if (bframes > 0)
+            if (bframes > 0 && !m_param->radl)
             {
                 list[bframes - 1]->m_lowres.sliceType = X265_TYPE_P;
                 bframes--;
             }
         }
+        if (m_param->radl && !m_param->bOpenGOP && list[bframes + 1])
+        {
+            if ((frm.frameNum - m_lastKeyframe) >  (m_param->keyframeMax - m_param->radl - 1) && (frm.frameNum - m_lastKeyframe) <  m_param->keyframeMax)
+                frm.sliceType = X265_TYPE_B;
+            if ((frm.frameNum - m_lastKeyframe) == (m_param->keyframeMax - m_param->radl - 1))
+                frm.sliceType = X265_TYPE_P;
+        }
+
         if (bframes == m_param->bframes || !list[bframes + 1])
         {
             if (IS_X265_TYPE_B(frm.sliceType))
@@ -1175,8 +1194,13 @@ void Lookahead::slicetypeDecide()
         if (bframes)
         {
             p0 = 0; // last nonb
+            bool isp0available = frames[bframes + 1]->sliceType == X265_TYPE_IDR ? false : true;
+
             for (b = 1; b <= bframes; b++)
             {
+                if (!isp0available)
+                    p0 = b;
+
                 if (frames[b]->sliceType == X265_TYPE_B)
                     for (p1 = b; frames[p1]->sliceType == X265_TYPE_B; p1++)
                         ; // find new nonb or bref
@@ -1186,7 +1210,10 @@ void Lookahead::slicetypeDecide()
                 estGroup.singleCost(p0, p1, b);
 
                 if (frames[b]->sliceType == X265_TYPE_BREF)
+                {
                     p0 = b;
+                    isp0available = true;
+                }
             }
         }
     }
@@ -1413,12 +1440,12 @@ void Lookahead::slicetypeAnalyse(Lowres **frames, bool bKeyframe)
                     continue;
 
                 /* Skip search if already done */
-                if (frames[b]->lowresMvs[0][i - 1][0].x != 0x7FFF)
+                if (frames[b]->lowresMvs[0][i][0].x != 0x7FFF)
                     continue;
 
                 /* perform search to p1 at same distance, if possible */
                 int p1 = b + i;
-                if (p1 >= numFrames || frames[b]->lowresMvs[1][i - 1][0].x != 0x7FFF)
+                if (p1 >= numFrames || frames[b]->lowresMvs[1][i][0].x != 0x7FFF)
                     p1 = b;
 
                 estGroup.add(p0, p1, b);
@@ -1440,7 +1467,7 @@ void Lookahead::slicetypeAnalyse(Lowres **frames, bool bKeyframe)
 
                     /* only measure frame cost in this pass if motion searches
                      * are already done */
-                    if (frames[b]->lowresMvs[0][i - 1][0].x == 0x7FFF)
+                    if (frames[b]->lowresMvs[0][i][0].x == 0x7FFF)
                         continue;
 
                     int p0 = b - i;
@@ -1452,7 +1479,7 @@ void Lookahead::slicetypeAnalyse(Lowres **frames, bool bKeyframe)
                             break;
 
                         /* ensure P1 search is done */
-                        if (j && frames[b]->lowresMvs[1][j - 1][0].x == 0x7FFF)
+                        if (j && frames[b]->lowresMvs[1][j][0].x == 0x7FFF)
                             continue;
 
                         /* ensure frame cost is not done */
@@ -1867,7 +1894,7 @@ void Lookahead::aqMotion(Lowres **frames, bool bIntra)
 
 void Lookahead::calcMotionAdaptiveQuantFrame(Lowres **frames, int p0, int p1, int b)
 {
-    int listDist[2] = { b - p0 - 1, p1 - b - 1 };
+    int listDist[2] = { b - p0, p1 - b };
     int32_t strideInCU = m_8x8Width;
     double qp_adj = 0, avg_adj = 0, avg_adj_pow2 = 0, sd;
     for (uint16_t blocky = 0; blocky < m_8x8Height; blocky++)
@@ -2030,7 +2057,7 @@ void Lookahead::estimateCUPropagate(Lowres **frames, double averageDuration, int
     int32_t distScaleFactor = (((b - p0) << 8) + ((p1 - p0) >> 1)) / (p1 - p0);
     int32_t bipredWeight = m_param->bEnableWeightedBiPred ? 64 - (distScaleFactor >> 2) : 32;
     int32_t bipredWeights[2] = { bipredWeight, 64 - bipredWeight };
-    int listDist[2] = { b - p0 - 1, p1 - b - 1 };
+    int listDist[2] = { b - p0, p1 - b };
 
     memset(m_scratch, 0, m_8x8Width * sizeof(int));
 
@@ -2305,17 +2332,15 @@ int64_t CostEstimateGroup::estimateFrameCost(LookaheadTLD& tld, int p0, int p1,
         score = fenc->costEst[b - p0][p1 - b];
     else
     {
-        X265_CHECK(p0 != b, "I frame estimates should always be pre-calculated\n");
-
         bool bDoSearch[2];
-        bDoSearch[0] = p0 < b && fenc->lowresMvs[0][b - p0 - 1][0].x == 0x7FFF;
-        bDoSearch[1] = p1 > b && fenc->lowresMvs[1][p1 - b - 1][0].x == 0x7FFF;
+        bDoSearch[0] = fenc->lowresMvs[0][b - p0][0].x == 0x7FFF;
+        bDoSearch[1] = p1 > b && fenc->lowresMvs[1][p1 - b][0].x == 0x7FFF;
 
 #if CHECKED_BUILD
-        X265_CHECK(!(p0 < b && fenc->lowresMvs[0][b - p0 - 1][0].x == 0x7FFE), "motion search batch duplication L0\n");
-        X265_CHECK(!(p1 > b && fenc->lowresMvs[1][p1 - b - 1][0].x == 0x7FFE), "motion search batch duplication L1\n");
-        if (bDoSearch[0]) fenc->lowresMvs[0][b - p0 - 1][0].x = 0x7FFE;
-        if (bDoSearch[1]) fenc->lowresMvs[1][p1 - b - 1][0].x = 0x7FFE;
+        X265_CHECK(!(p0 < b && fenc->lowresMvs[0][b - p0][0].x == 0x7FFE), "motion search batch duplication L0\n");
+        X265_CHECK(!(p1 > b && fenc->lowresMvs[1][p1 - b][0].x == 0x7FFE), "motion search batch duplication L1\n");
+        if (bDoSearch[0]) fenc->lowresMvs[0][b - p0][0].x = 0x7FFE;
+        if (bDoSearch[1]) fenc->lowresMvs[1][p1 - b][0].x = 0x7FFE;
 #endif
 
         fenc->weightedRef[b - p0].isWeighted = false;
@@ -2406,7 +2431,7 @@ void CostEstimateGroup::estimateCUCost(LookaheadTLD& tld, int cuX, int cuY, int
 
     /* A small, arbitrary bias to avoid VBV problems caused by zero-residual lookahead blocks. */
     int lowresPenalty = 4;
-    int listDist[2] = { b - p0 - 1, p1 - b - 1 };
+    int listDist[2] = { b - p0, p1 - b};
 
     MV mvmin, mvmax;
     int bcost = tld.me.COST_MAX;
diff --git a/source/encoder/weightPrediction.cpp b/source/encoder/weightPrediction.cpp
index 7c16ac323b..19255f9acf 100644
--- a/source/encoder/weightPrediction.cpp
+++ b/source/encoder/weightPrediction.cpp
@@ -323,7 +323,7 @@ void weightAnalyse(Slice& slice, Frame& frame, x265_param& param)
 
             if (!plane && diffPoc <= param.bframes + 1)
             {
-                mvs = fenc.lowresMvs[list][diffPoc - 1];
+                mvs = fenc.lowresMvs[list][diffPoc];
 
                 /* test whether this motion search was performed by lookahead */
                 if (mvs[0].x != 0x7FFF)
diff --git a/source/test/regression-tests.txt b/source/test/regression-tests.txt
index 45cd0f2b1c..51a1b0f338 100644
--- a/source/test/regression-tests.txt
+++ b/source/test/regression-tests.txt
@@ -151,6 +151,7 @@ Kimono1_1920x1080_24_400.yuv,--preset medium --rdoq-level 0 --limit-refs 3 --sli
 Kimono1_1920x1080_24_400.yuv,--preset veryslow --crf 4 --cu-lossless --slices 2 --limit-refs 3 --limit-modes
 Kimono1_1920x1080_24_400.yuv,--preset placebo --ctu 32 --max-tu-size 8 --limit-tu 2
 big_buck_bunny_360p24.y4m, --keyint 60 --min-keyint 40 --gop-lookahead 14
+BasketballDrive_1920x1080_50.y4m, --preset medium --no-open-gop --keyint 50 --min-keyint 50 --radl 2
 
 # Main12 intraCost overflow bug test
 720p50_parkrun_ter.y4m,--preset medium
diff --git a/source/x265.h b/source/x265.h
index 808586d7ea..0ed077d5ee 100644
--- a/source/x265.h
+++ b/source/x265.h
@@ -1268,6 +1268,7 @@ typedef struct x265_param
 
         /* internally enable if tune grain is set */
         int      bEnableConstVbv;
+
     } rc;
 
     /*== Video Usability Information ==*/
@@ -1542,6 +1543,8 @@ typedef struct x265_param
      * to reduce the amount of work the encoder must perform. Default disabled. */
     const char* analysisLoad;
 
+    /*Number of RADL pictures allowed in front of IDR*/
+    int radl;
 } x265_param;
 
 /* x265_param_alloc:
diff --git a/source/x265cli.h b/source/x265cli.h
index 5f8a69b9e1..0cdaa921da 100644
--- a/source/x265cli.h
+++ b/source/x265cli.h
@@ -124,6 +124,7 @@ static const struct option long_options[] =
     { "scenecut",       required_argument, NULL, 0 },
     { "no-scenecut",          no_argument, NULL, 0 },
     { "scenecut-bias",  required_argument, NULL, 0 },
+    { "radl",           required_argument, NULL, 0 },
     { "ctu-info",       required_argument, NULL, 0 },
     { "intra-refresh",        no_argument, NULL, 0 },
     { "rc-lookahead",   required_argument, NULL, 0 },
@@ -427,6 +428,7 @@ static void showHelp(x265_param *param)
     H0("   --no-scenecut                 Disable adaptive I-frame decision\n");
     H0("   --scenecut <integer>          How aggressively to insert extra I-frames. Default %d\n", param->scenecutThreshold);
     H1("   --scenecut-bias <0..100.0>    Bias for scenecut detection. Default %.2f\n", param->scenecutBias);
+    H0("   --radl <integer>              Number of RADL pictures allowed in front of IDR. Default %d\n", param->radl);
     H0("   --intra-refresh               Use Periodic Intra Refresh instead of IDR frames\n");
     H0("   --rc-lookahead <integer>      Number of frames for frame-type lookahead (determines encoder latency) Default %d\n", param->lookaheadDepth);
     H1("   --lookahead-slices <0..16>    Number of slices to use per lookahead cost estimate. Default %d\n", param->lookaheadSlices);

From acb679fc66b112fd92616f6f972f7d56f5da5142 Mon Sep 17 00:00:00 2001
From: Aruna Matheswaran <aruna@multicorewareinc.com>
Date: Fri, 22 Dec 2017 18:23:24 +0530
Subject: [PATCH 24/51] Restore macros defining analysis options

---
 source/x265.h | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/source/x265.h b/source/x265.h
index 0ed077d5ee..ca6263ef57 100644
--- a/source/x265.h
+++ b/source/x265.h
@@ -481,8 +481,11 @@ typedef enum
 #define X265_CSP_BGRA           7  /* packed bgr 32bits   */
 #define X265_CSP_RGB            8  /* packed rgb 24bits   */
 #define X265_CSP_MAX            9  /* end of list */
-
 #define X265_EXTENDED_SAR       255 /* aspect ratio explicitly specified as width:height */
+/* Analysis options */
+#define X265_ANALYSIS_OFF  0
+#define X265_ANALYSIS_SAVE 1
+#define X265_ANALYSIS_LOAD 2
 
 typedef struct x265_cli_csp
 {

From 5fef3fef3ef4500090891922f0b9a69853300f65 Mon Sep 17 00:00:00 2001
From: Aasaipriya Chandran <aasaipriya@multicorewareinc.com>
Date: Wed, 27 Dec 2017 19:05:56 +0530
Subject: [PATCH 25/51] Reconfigure csvfpt if encoder param is null

---
 source/encoder/api.cpp | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/source/encoder/api.cpp b/source/encoder/api.cpp
index 2ae2bc82dc..985d21ddf0 100644
--- a/source/encoder/api.cpp
+++ b/source/encoder/api.cpp
@@ -192,9 +192,10 @@ int x265_encoder_reconfig(x265_encoder* enc, x265_param* param_in)
 {
     if (!enc || !param_in)
         return -1;
-
     x265_param save;
     Encoder* encoder = static_cast<Encoder*>(enc);
+    if (encoder->m_param->csvfn == NULL && param_in->csvfpt != NULL)
+         encoder->m_param->csvfpt = param_in->csvfpt;
     if (encoder->m_latestParam->forceFlush != param_in->forceFlush)
         return encoder->reconfigureParam(encoder->m_latestParam, param_in);
     bool isReconfigureRc = encoder->isReconfigureRc(encoder->m_latestParam, param_in);

From b82b97087eaca97d10c687f617035bc001eb7405 Mon Sep 17 00:00:00 2001
From: Pradeep Ramachandran <pradeep@multicorewareinc.com>
Date: Thu, 4 Jan 2018 13:56:07 +0530
Subject: [PATCH 26/51] doc: Fix invalid UTF-8 characters in docs to enable
 linux gen

---
 doc/reST/cli.rst | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/doc/reST/cli.rst b/doc/reST/cli.rst
index 74a14d8fa6..83a5eb6f6b 100644
--- a/doc/reST/cli.rst
+++ b/doc/reST/cli.rst
@@ -2058,7 +2058,7 @@ VUI fields must be manually specified.
 	Example for MaxCLL=1000 candela per square meter, MaxFALL=400
 	candela per square meter:
 
-		--max-cll “1000,400”
+		--max-cll "1000,400"
 
 	Note that this string value will need to be escaped or quoted to
 	protect against shell expansion on many platforms. No default.

From e30facce857804794944fd8ad0f72fc9338e0edd Mon Sep 17 00:00:00 2001
From: Aarthi Thirumalai <aarthi@multicorewareinc.com>
Date: Thu, 4 Jan 2018 15:16:06 +0530
Subject: [PATCH 27/51] api: change signature of x265_csvlog_encode()

replace x265_encoder* with x265_param* and int padx, int pady as function arguments.
---
 source/CMakeLists.txt  |   2 +-
 source/encoder/api.cpp | 104 ++++++++++++++++++++---------------------
 source/x265.h          |   7 +--
 3 files changed, 55 insertions(+), 58 deletions(-)

diff --git a/source/CMakeLists.txt b/source/CMakeLists.txt
index e17cca9f56..557814ad40 100644
--- a/source/CMakeLists.txt
+++ b/source/CMakeLists.txt
@@ -29,7 +29,7 @@ option(NATIVE_BUILD "Target the build CPU" OFF)
 option(STATIC_LINK_CRT "Statically link C runtime for release builds" OFF)
 mark_as_advanced(FPROFILE_USE FPROFILE_GENERATE NATIVE_BUILD)
 # X265_BUILD must be incremented each time the public API is changed
-set(X265_BUILD 150)
+set(X265_BUILD 151)
 configure_file("${PROJECT_SOURCE_DIR}/x265.def.in"
                "${PROJECT_BINARY_DIR}/x265.def")
 configure_file("${PROJECT_SOURCE_DIR}/x265_config.h.in"
diff --git a/source/encoder/api.cpp b/source/encoder/api.cpp
index eea76ebd39..85de1d762c 100644
--- a/source/encoder/api.cpp
+++ b/source/encoder/api.cpp
@@ -312,7 +312,9 @@ void x265_encoder_log(x265_encoder* enc, int argc, char **argv)
         Encoder *encoder = static_cast<Encoder*>(enc);
         x265_stats stats;
         encoder->fetchStats(&stats, sizeof(stats));
-        x265_csvlog_encode(enc, &stats, argc, argv);
+        int padx = encoder->m_sps.conformanceWindow.rightOffset;
+        int pady = encoder->m_sps.conformanceWindow.bottomOffset;
+        x265_csvlog_encode(encoder->m_param, &stats, padx, pady, argc, argv);
     }
 }
 
@@ -868,45 +870,38 @@ void x265_csvlog_frame(const x265_param* param, const x265_picture* pic)
     fflush(stderr);
 }
 
-void x265_csvlog_encode(x265_encoder *enc, const x265_stats* stats, int argc, char** argv)
+void x265_csvlog_encode(const x265_param *p, const x265_stats *stats, int padx, int pady, int argc, char** argv)
 {
-    if (enc)
+    if (p && p->csvfpt)
     {
-        Encoder *encoder = static_cast<Encoder*>(enc);
-        int padx = encoder->m_sps.conformanceWindow.rightOffset;
-        int pady = encoder->m_sps.conformanceWindow.bottomOffset;
         const x265_api * api = x265_api_get(0);
 
-        if (!encoder->m_param->csvfpt)
-            return;
-
-        if (encoder->m_param->csvLogLevel)
+        if (p->csvLogLevel)
         {
             // adding summary to a per-frame csv log file, so it needs a summary header
-            fprintf(encoder->m_param->csvfpt, "\nSummary\n");
-            fputs(summaryCSVHeader, encoder->m_param->csvfpt);
+            fprintf(p->csvfpt, "\nSummary\n");
+            fputs(summaryCSVHeader, p->csvfpt);
         }
 
         // CLI arguments or other
         if (argc)
         {
-            fputc('"', encoder->m_param->csvfpt);
+            fputc('"', p->csvfpt);
             for (int i = 1; i < argc; i++)
             {
-                fputc(' ', encoder->m_param->csvfpt);
-                fputs(argv[i], encoder->m_param->csvfpt);
+                fputc(' ', p->csvfpt);
+                fputs(argv[i], p->csvfpt);
             }
-            fputc('"', encoder->m_param->csvfpt);
+            fputc('"', p->csvfpt);
         }
         else
         {
-            const x265_param* paramTemp = encoder->m_param;
-            char *opts = x265_param2string((x265_param*)paramTemp, padx, pady);
+            char *opts = x265_param2string((x265_param*)p, padx, pady);
             if (opts)
             {
-                fputc('"', encoder->m_param->csvfpt);
-                fputs(opts, encoder->m_param->csvfpt);
-                fputc('"', encoder->m_param->csvfpt);
+                fputc('"', p->csvfpt);
+                fputs(opts, p->csvfpt);
+                fputc('"', p->csvfpt);
             }
         }
 
@@ -917,70 +912,71 @@ void x265_csvlog_encode(x265_encoder *enc, const x265_stats* stats, int argc, ch
         timeinfo = localtime(&now);
         char buffer[200];
         strftime(buffer, 128, "%c", timeinfo);
-        fprintf(encoder->m_param->csvfpt, ", %s, ", buffer);
+        fprintf(p->csvfpt, ", %s, ", buffer);
 
         // elapsed time, fps, bitrate
-        fprintf(encoder->m_param->csvfpt, "%.2f, %.2f, %.2f,",
+        fprintf(p->csvfpt, "%.2f, %.2f, %.2f,",
             stats->elapsedEncodeTime, stats->encodedPictureCount / stats->elapsedEncodeTime, stats->bitrate);
 
-        if (encoder->m_param->bEnablePsnr)
-            fprintf(encoder->m_param->csvfpt, " %.3lf, %.3lf, %.3lf, %.3lf,",
+        if (p->bEnablePsnr)
+            fprintf(p->csvfpt, " %.3lf, %.3lf, %.3lf, %.3lf,",
             stats->globalPsnrY / stats->encodedPictureCount, stats->globalPsnrU / stats->encodedPictureCount,
             stats->globalPsnrV / stats->encodedPictureCount, stats->globalPsnr);
         else
-            fprintf(encoder->m_param->csvfpt, " -, -, -, -,");
-        if (encoder->m_param->bEnableSsim)
-            fprintf(encoder->m_param->csvfpt, " %.6f, %6.3f,", stats->globalSsim, x265_ssim2dB(stats->globalSsim));
+            fprintf(p->csvfpt, " -, -, -, -,");
+        if (p->bEnableSsim)
+            fprintf(p->csvfpt, " %.6f, %6.3f,", stats->globalSsim, x265_ssim2dB(stats->globalSsim));
         else
-            fprintf(encoder->m_param->csvfpt, " -, -,");
+            fprintf(p->csvfpt, " -, -,");
 
         if (stats->statsI.numPics)
         {
-            fprintf(encoder->m_param->csvfpt, " %-6u, %2.2lf, %-8.2lf,", stats->statsI.numPics, stats->statsI.avgQp, stats->statsI.bitrate);
-            if (encoder->m_param->bEnablePsnr)
-                fprintf(encoder->m_param->csvfpt, " %.3lf, %.3lf, %.3lf,", stats->statsI.psnrY, stats->statsI.psnrU, stats->statsI.psnrV);
+            fprintf(p->csvfpt, " %-6u, %2.2lf, %-8.2lf,", stats->statsI.numPics, stats->statsI.avgQp, stats->statsI.bitrate);
+            if (p->bEnablePsnr)
+                fprintf(p->csvfpt, " %.3lf, %.3lf, %.3lf,", stats->statsI.psnrY, stats->statsI.psnrU, stats->statsI.psnrV);
             else
-                fprintf(encoder->m_param->csvfpt, " -, -, -,");
-            if (encoder->m_param->bEnableSsim)
-                fprintf(encoder->m_param->csvfpt, " %.3lf,", stats->statsI.ssim);
+                fprintf(p->csvfpt, " -, -, -,");
+            if (p->bEnableSsim)
+                fprintf(p->csvfpt, " %.3lf,", stats->statsI.ssim);
             else
-                fprintf(encoder->m_param->csvfpt, " -,");
+                fprintf(p->csvfpt, " -,");
         }
         else
-            fprintf(encoder->m_param->csvfpt, " -, -, -, -, -, -, -,");
+            fprintf(p->csvfpt, " -, -, -, -, -, -, -,");
 
         if (stats->statsP.numPics)
         {
-            fprintf(encoder->m_param->csvfpt, " %-6u, %2.2lf, %-8.2lf,", stats->statsP.numPics, stats->statsP.avgQp, stats->statsP.bitrate);
-            if (encoder->m_param->bEnablePsnr)
-                fprintf(encoder->m_param->csvfpt, " %.3lf, %.3lf, %.3lf,", stats->statsP.psnrY, stats->statsP.psnrU, stats->statsP.psnrV);
+            fprintf(p->csvfpt, " %-6u, %2.2lf, %-8.2lf,", stats->statsP.numPics, stats->statsP.avgQp, stats->statsP.bitrate);
+            if (p->bEnablePsnr)
+                fprintf(p->csvfpt, " %.3lf, %.3lf, %.3lf,", stats->statsP.psnrY, stats->statsP.psnrU, stats->statsP.psnrV);
             else
-                fprintf(encoder->m_param->csvfpt, " -, -, -,");
-            if (encoder->m_param->bEnableSsim)
-                fprintf(encoder->m_param->csvfpt, " %.3lf,", stats->statsP.ssim);
+                fprintf(p->csvfpt, " -, -, -,");
+            if (p->bEnableSsim)
+                fprintf(p->csvfpt, " %.3lf,", stats->statsP.ssim);
             else
-                fprintf(encoder->m_param->csvfpt, " -,");
+                fprintf(p->csvfpt, " -,");
         }
         else
-            fprintf(encoder->m_param->csvfpt, " -, -, -, -, -, -, -,");
+            fprintf(p->csvfpt, " -, -, -, -, -, -, -,");
 
         if (stats->statsB.numPics)
         {
-            fprintf(encoder->m_param->csvfpt, " %-6u, %2.2lf, %-8.2lf,", stats->statsB.numPics, stats->statsB.avgQp, stats->statsB.bitrate);
-            if (encoder->m_param->bEnablePsnr)
-                fprintf(encoder->m_param->csvfpt, " %.3lf, %.3lf, %.3lf,", stats->statsB.psnrY, stats->statsB.psnrU, stats->statsB.psnrV);
+            fprintf(p->csvfpt, " %-6u, %2.2lf, %-8.2lf,", stats->statsB.numPics, stats->statsB.avgQp, stats->statsB.bitrate);
+            if (p->bEnablePsnr)
+                fprintf(p->csvfpt, " %.3lf, %.3lf, %.3lf,", stats->statsB.psnrY, stats->statsB.psnrU, stats->statsB.psnrV);
             else
-                fprintf(encoder->m_param->csvfpt, " -, -, -,");
-            if (encoder->m_param->bEnableSsim)
-                fprintf(encoder->m_param->csvfpt, " %.3lf,", stats->statsB.ssim);
+                fprintf(p->csvfpt, " -, -, -,");
+            if (p->bEnableSsim)
+                fprintf(p->csvfpt, " %.3lf,", stats->statsB.ssim);
             else
-                fprintf(encoder->m_param->csvfpt, " -,");
+                fprintf(p->csvfpt, " -,");
         }
         else
-            fprintf(encoder->m_param->csvfpt, " -, -, -, -, -, -, -,");
+            fprintf(p->csvfpt, " -, -, -, -, -, -, -,");
 
-        fprintf(encoder->m_param->csvfpt, " %-6u, %-6u, %s\n", stats->maxCLL, stats->maxFALL, api->version_str);
+        fprintf(p->csvfpt, " %-6u, %-6u, %s\n", stats->maxCLL, stats->maxFALL, api->version_str);
     }
+
 }
 
 /* The dithering algorithm is based on Sierra-2-4A error diffusion.
diff --git a/source/x265.h b/source/x265.h
index ca6263ef57..6d783dcbe1 100644
--- a/source/x265.h
+++ b/source/x265.h
@@ -1779,9 +1779,10 @@ FILE* x265_csvlog_open(const x265_param *);
 void x265_csvlog_frame(const x265_param *, const x265_picture *);
 
 /* Log final encode statistics to the CSV file handle. 'argc' and 'argv' are
- * intended to be command line arguments passed to the encoder. Encode
+ * intended to be command line arguments passed to the encoder. padx and pady are
+ * padding offsets for conformance and can be given from sps settings. Encode
  * statistics should be queried from the encoder just prior to closing it. */
-void x265_csvlog_encode(x265_encoder *encoder, const x265_stats *, int argc, char** argv);
+void x265_csvlog_encode(const x265_param*, const x265_stats *, int padx, int pady, int argc, char** argv);
 
 /* In-place downshift from a bit-depth greater than 8 to a bit-depth of 8, using
  * the residual bits to dither each row. */
@@ -1836,7 +1837,7 @@ typedef struct x265_api
     int           (*get_ref_frame_list)(x265_encoder*, x265_picyuv**, x265_picyuv**, int, int, int*, int*);
     FILE*         (*csvlog_open)(const x265_param*);
     void          (*csvlog_frame)(const x265_param*, const x265_picture*);
-    void          (*csvlog_encode)(x265_encoder*, const x265_stats*, int, char**);
+    void          (*csvlog_encode)(const x265_param*, const x265_stats *, int, int, int, char**);
     void          (*dither_image)(x265_picture*, int, int, int16_t*, int);
     int           (*set_analysis_data)(x265_encoder *encoder, x265_analysis_data *analysis_data, int poc, uint32_t cuBytes);
     /* add new pointers to the end, or increment X265_MAJOR_VERSION */

From 1a3e688428002aed95a4933ce87e63d8cb97c4fa Mon Sep 17 00:00:00 2001
From: Vignesh Vijayakumar <unknown>
Date: Thu, 4 Jan 2018 12:37:01 +0530
Subject: [PATCH 28/51] Fix Mac OS build warnings 1. due to SEA motion search
 2. due to condition check for CTUInfo

---
 source/common/framedata.cpp    | 12 +++----
 source/encoder/analysis.cpp    | 58 ++++++++++++++++------------------
 source/encoder/dpb.cpp         | 13 +++-----
 source/encoder/framefilter.cpp |  2 +-
 4 files changed, 36 insertions(+), 49 deletions(-)

diff --git a/source/common/framedata.cpp b/source/common/framedata.cpp
index e0ecd57149..d7d475311b 100644
--- a/source/common/framedata.cpp
+++ b/source/common/framedata.cpp
@@ -77,16 +77,12 @@ void FrameData::destroy()
 
     X265_FREE(m_cuStat);
     X265_FREE(m_rowStat);
-
-    if (m_meBuffer)
+    for (int i = 0; i < INTEGRAL_PLANE_NUM; i++)
     {
-        for (int i = 0; i < INTEGRAL_PLANE_NUM; i++)
+        if (m_meBuffer[i] != NULL)
         {
-            if (m_meBuffer[i] != NULL)
-            {
-                X265_FREE(m_meBuffer[i]);
-                m_meBuffer[i] = NULL;
-            }
+            X265_FREE(m_meBuffer[i]);
+            m_meBuffer[i] = NULL;
         }
     }
 }
diff --git a/source/encoder/analysis.cpp b/source/encoder/analysis.cpp
index 3be581ca6f..3622ed51d6 100644
--- a/source/encoder/analysis.cpp
+++ b/source/encoder/analysis.cpp
@@ -159,38 +159,34 @@ Mode& Analysis::compressCTU(CUData& ctu, Frame& frame, const CUGeom& cuGeom, con
     if (m_param->bCTUInfo && (*m_frame->m_ctuInfo + ctu.m_cuAddr))
     {
         x265_ctu_info_t* ctuTemp = *m_frame->m_ctuInfo + ctu.m_cuAddr;
-        if (ctuTemp->ctuPartitions)
-        {
-            int32_t depthIdx = 0;
-            uint32_t maxNum8x8Partitions = 64;
-            uint8_t* depthInfoPtr = m_frame->m_addOnDepth[ctu.m_cuAddr];
-            uint8_t* contentInfoPtr = m_frame->m_addOnCtuInfo[ctu.m_cuAddr];
-            int* prevCtuInfoChangePtr = m_frame->m_addOnPrevChange[ctu.m_cuAddr];
-            do
-            {
-                uint8_t depth = (uint8_t)ctuTemp->ctuPartitions[depthIdx];
-                uint8_t content = (uint8_t)(*((int32_t *)ctuTemp->ctuInfo + depthIdx));
-                int prevCtuInfoChange = m_frame->m_prevCtuInfoChange[ctu.m_cuAddr * maxNum8x8Partitions + depthIdx];
-                memset(depthInfoPtr, depth, sizeof(uint8_t) * numPartition >> 2 * depth);
-                memset(contentInfoPtr, content, sizeof(uint8_t) * numPartition >> 2 * depth);
-                memset(prevCtuInfoChangePtr, 0, sizeof(int) * numPartition >> 2 * depth);
-                for (uint32_t l = 0; l < numPartition >> 2 * depth; l++)
-                    prevCtuInfoChangePtr[l] = prevCtuInfoChange;
-                depthInfoPtr += ctu.m_numPartitions >> 2 * depth;
-                contentInfoPtr += ctu.m_numPartitions >> 2 * depth;
-                prevCtuInfoChangePtr += ctu.m_numPartitions >> 2 * depth;
-                depthIdx++;
-            } while (ctuTemp->ctuPartitions[depthIdx] != 0);
-
-            m_additionalCtuInfo = m_frame->m_addOnCtuInfo[ctu.m_cuAddr];
-            m_prevCtuInfoChange = m_frame->m_addOnPrevChange[ctu.m_cuAddr];
-            memcpy(ctu.m_cuDepth, m_frame->m_addOnDepth[ctu.m_cuAddr], sizeof(uint8_t) * numPartition);
-            //Calculate log2CUSize from depth
-            for (uint32_t i = 0; i < cuGeom.numPartitions; i++)
-                ctu.m_log2CUSize[i] = (uint8_t)m_param->maxLog2CUSize - ctu.m_cuDepth[i];
-        }
+        int32_t depthIdx = 0;
+        uint32_t maxNum8x8Partitions = 64;
+        uint8_t* depthInfoPtr = m_frame->m_addOnDepth[ctu.m_cuAddr];
+        uint8_t* contentInfoPtr = m_frame->m_addOnCtuInfo[ctu.m_cuAddr];
+        int* prevCtuInfoChangePtr = m_frame->m_addOnPrevChange[ctu.m_cuAddr];
+        do
+        {
+            uint8_t depth = (uint8_t)ctuTemp->ctuPartitions[depthIdx];
+            uint8_t content = (uint8_t)(*((int32_t *)ctuTemp->ctuInfo + depthIdx));
+            int prevCtuInfoChange = m_frame->m_prevCtuInfoChange[ctu.m_cuAddr * maxNum8x8Partitions + depthIdx];
+            memset(depthInfoPtr, depth, sizeof(uint8_t) * numPartition >> 2 * depth);
+            memset(contentInfoPtr, content, sizeof(uint8_t) * numPartition >> 2 * depth);
+            memset(prevCtuInfoChangePtr, 0, sizeof(int) * numPartition >> 2 * depth);
+            for (uint32_t l = 0; l < numPartition >> 2 * depth; l++)
+                prevCtuInfoChangePtr[l] = prevCtuInfoChange;
+            depthInfoPtr += ctu.m_numPartitions >> 2 * depth;
+            contentInfoPtr += ctu.m_numPartitions >> 2 * depth;
+            prevCtuInfoChangePtr += ctu.m_numPartitions >> 2 * depth;
+            depthIdx++;
+        } while (ctuTemp->ctuPartitions[depthIdx] != 0);
+
+        m_additionalCtuInfo = m_frame->m_addOnCtuInfo[ctu.m_cuAddr];
+        m_prevCtuInfoChange = m_frame->m_addOnPrevChange[ctu.m_cuAddr];
+        memcpy(ctu.m_cuDepth, m_frame->m_addOnDepth[ctu.m_cuAddr], sizeof(uint8_t) * numPartition);
+        //Calculate log2CUSize from depth
+        for (uint32_t i = 0; i < cuGeom.numPartitions; i++)
+            ctu.m_log2CUSize[i] = (uint8_t)m_param->maxLog2CUSize - ctu.m_cuDepth[i];
     }
-
     if (m_param->analysisMultiPassRefine && m_param->rc.bStatRead)
     {
         m_multipassAnalysis = (analysis2PassFrameData*)m_frame->m_analysis2Pass.analysisFramedata;
diff --git a/source/encoder/dpb.cpp b/source/encoder/dpb.cpp
index 19d41f9cdd..6c4ce15f3b 100644
--- a/source/encoder/dpb.cpp
+++ b/source/encoder/dpb.cpp
@@ -92,19 +92,14 @@ void DPB::recycleUnreferenced()
             m_freeList.pushBack(*curFrame);
             curFrame->m_encData->m_freeListNext = m_frameDataFreeList;
             m_frameDataFreeList = curFrame->m_encData;
-
-            if (curFrame->m_encData->m_meBuffer)
+            for (int i = 0; i < INTEGRAL_PLANE_NUM; i++)
             {
-                for (int i = 0; i < INTEGRAL_PLANE_NUM; i++)
+                if (curFrame->m_encData->m_meBuffer[i] != NULL)
                 {
-                    if (curFrame->m_encData->m_meBuffer[i] != NULL)
-                    {
-                        X265_FREE(curFrame->m_encData->m_meBuffer[i]);
-                        curFrame->m_encData->m_meBuffer[i] = NULL;
-                    }
+                    X265_FREE(curFrame->m_encData->m_meBuffer[i]);
+                    curFrame->m_encData->m_meBuffer[i] = NULL;
                 }
             }
-
             if (curFrame->m_ctuInfo != NULL)
             {
                 uint32_t widthInCU = (curFrame->m_param->sourceWidth + curFrame->m_param->maxCUSize - 1) >> curFrame->m_param->maxLog2CUSize;
diff --git a/source/encoder/framefilter.cpp b/source/encoder/framefilter.cpp
index ccc7231b70..fd03d24fb8 100644
--- a/source/encoder/framefilter.cpp
+++ b/source/encoder/framefilter.cpp
@@ -795,7 +795,7 @@ void FrameFilter::processPostRow(int row)
 void FrameFilter::computeMEIntegral(int row)
 {
     int lastRow = row == (int)m_frame->m_encData->m_slice->m_sps->numCuInHeight - 1;
-    if (m_frame->m_encData->m_meIntegral && m_frame->m_lowres.sliceType != X265_TYPE_B)
+    if (m_frame->m_lowres.sliceType != X265_TYPE_B)
     {
         /* If WPP, other than first row, integral calculation for current row needs to wait till the
         * integral for the previous row is computed */

From 83e856dc9209809586f73ec4d470343267c4998e Mon Sep 17 00:00:00 2001
From: Ma0 <mateuszb@poczta.onet.pl>
Date: Sat, 13 Jan 2018 00:25:44 +0100
Subject: [PATCH 29/51] param2string: increase buffer size, do not store file
 names

---
 source/common/param.cpp | 16 +++++++++++-----
 source/common/param.h   |  3 ---
 2 files changed, 11 insertions(+), 8 deletions(-)

diff --git a/source/common/param.cpp b/source/common/param.cpp
index ab8f419391..b4b0b83c2f 100644
--- a/source/common/param.cpp
+++ b/source/common/param.cpp
@@ -1530,11 +1530,15 @@ void x265_print_params(x265_param* param)
 char *x265_param2string(x265_param* p, int padx, int pady)
 {
     char *buf, *s;
+    int bufSize = 4000 + p->rc.zoneCount * 64;
+    if (p->numaPools)
+        bufSize += strlen(p->numaPools);
+    if (p->masteringDisplayColorVolume)
+        bufSize += strlen(p->masteringDisplayColorVolume);
 
-    buf = s = X265_MALLOC(char, MAXPARAMSIZE);
+    buf = s = X265_MALLOC(char, bufSize);
     if (!buf)
         return NULL;
-
 #define BOOL(param, cliopt) \
     s += sprintf(s, " %s", (param) ? cliopt : "no-" cliopt);
 
@@ -1549,7 +1553,7 @@ char *x265_param2string(x265_param* p, int padx, int pady)
     BOOL(p->bEnableSsim, "ssim");
     s += sprintf(s, " log-level=%d", p->logLevel);
     if (p->csvfn)
-        s += sprintf(s, " csvfn=%s csv-log-level=%d", p->csvfn, p->csvLogLevel);
+        s += sprintf(s, " csv csv-log-level=%d", p->csvLogLevel);
     s += sprintf(s, " bitdepth=%d", p->internalBitDepth);
     s += sprintf(s, " input-csp=%d", p->internalCsp);
     s += sprintf(s, " fps=%u/%u", p->fpsNum, p->fpsDenom);
@@ -1722,8 +1726,10 @@ char *x265_param2string(x265_param* p, int padx, int pady)
     BOOL(p->bEmitHDRSEI, "hdr");
     BOOL(p->bHDROpt, "hdr-opt");
     BOOL(p->bDhdr10opt, "dhdr10-opt");
-    s += sprintf(s, " analysis-save=%s", p->analysisSave);
-    s += sprintf(s, " analysis-load=%s", p->analysisLoad);
+    if (p->analysisSave)
+        s += sprintf(s, " analysis-save");
+    if (p->analysisLoad)
+        s += sprintf(s, " analysis-load");
     s += sprintf(s, " analysis-reuse-level=%d", p->analysisReuseLevel);
     s += sprintf(s, " scale-factor=%d", p->scaleFactor);
     s += sprintf(s, " refine-intra=%d", p->intraRefine);
diff --git a/source/common/param.h b/source/common/param.h
index 9424b44c41..96775a601c 100644
--- a/source/common/param.h
+++ b/source/common/param.h
@@ -53,8 +53,5 @@ int x265_param_apply_profile(x265_param *, const char *profile);
 int x265_param_parse(x265_param *p, const char *name, const char *value);
 #define PARAM_NS X265_NS
 #endif
-
-#define MAXPARAMSIZE 2000
 }
-
 #endif // ifndef X265_PARAM_H

From 20477d0f17d6e6a4244029905f11807bf97c9f10 Mon Sep 17 00:00:00 2001
From: Ma0 <mateuszb@poczta.onet.pl>
Date: Sat, 13 Jan 2018 00:37:31 +0100
Subject: [PATCH 30/51] change date to 2018 in output file header

---
 source/encoder/encoder.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/source/encoder/encoder.cpp b/source/encoder/encoder.cpp
index 825e7b3f7d..15b6c0c622 100644
--- a/source/encoder/encoder.cpp
+++ b/source/encoder/encoder.cpp
@@ -2302,7 +2302,7 @@ void Encoder::getStreamHeaders(NALList& list, Entropy& sbacCoder, Bitstream& bs)
             if (buffer)
             {
                 sprintf(buffer, "x265 (build %d) - %s:%s - H.265/HEVC codec - "
-                        "Copyright 2013-2017 (c) Multicoreware, Inc - "
+                        "Copyright 2013-2018 (c) Multicoreware, Inc - "
                         "http://x265.org - options: %s",
                         X265_BUILD, PFX(version_str), PFX(build_info_str), opts);
                 

From 94755d3fcd653a8a2380a422b37a170fcb414e8b Mon Sep 17 00:00:00 2001
From: Ashok Kumar Mishra <ashok@multicorewareinc.com>
Date: Tue, 16 Jan 2018 19:25:26 +0530
Subject: [PATCH 31/51] Change type of bufSize variable to size_t to avoid
 warning.

---
 source/common/param.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/source/common/param.cpp b/source/common/param.cpp
index b4b0b83c2f..bae37fb61d 100644
--- a/source/common/param.cpp
+++ b/source/common/param.cpp
@@ -1530,7 +1530,7 @@ void x265_print_params(x265_param* param)
 char *x265_param2string(x265_param* p, int padx, int pady)
 {
     char *buf, *s;
-    int bufSize = 4000 + p->rc.zoneCount * 64;
+    size_t bufSize = 4000 + p->rc.zoneCount * 64;
     if (p->numaPools)
         bufSize += strlen(p->numaPools);
     if (p->masteringDisplayColorVolume)

From e5947ad2f61a401641efd9ea2eec42b502cdf9a2 Mon Sep 17 00:00:00 2001
From: Ma0 <mateuszb@poczta.onet.pl>
Date: Sat, 13 Jan 2018 02:47:30 +0100
Subject: [PATCH 32/51] input: change from ifstream to stdio stream

---
 source/common/common.h |   6 +-
 source/input/y4m.cpp   | 146 +++++++++++++----------------------------
 source/input/y4m.h     |  15 +----
 source/input/yuv.cpp   |  64 +++++-------------
 source/input/yuv.h     |  14 +---
 5 files changed, 71 insertions(+), 174 deletions(-)

diff --git a/source/common/common.h b/source/common/common.h
index 38c2ba07e1..e35d431864 100644
--- a/source/common/common.h
+++ b/source/common/common.h
@@ -75,11 +75,10 @@
 #define ALIGN_VAR_8(T, var)  T var __attribute__((aligned(8)))
 #define ALIGN_VAR_16(T, var) T var __attribute__((aligned(16)))
 #define ALIGN_VAR_32(T, var) T var __attribute__((aligned(32)))
-
 #if defined(__MINGW32__)
 #define fseeko fseeko64
+#define ftello ftello64
 #endif
-
 #elif defined(_MSC_VER)
 
 #define ALIGN_VAR_4(T, var)  __declspec(align(4)) T var
@@ -87,9 +86,8 @@
 #define ALIGN_VAR_16(T, var) __declspec(align(16)) T var
 #define ALIGN_VAR_32(T, var) __declspec(align(32)) T var
 #define fseeko _fseeki64
-
+#define ftello _ftelli64
 #endif // if defined(__GNUC__)
-
 #if HAVE_INT_TYPES_H
 #define __STDC_FORMAT_MACROS
 #include <inttypes.h>
diff --git a/source/input/y4m.cpp b/source/input/y4m.cpp
index 38732643c5..a1f6379f5f 100644
--- a/source/input/y4m.cpp
+++ b/source/input/y4m.cpp
@@ -38,9 +38,7 @@
 
 using namespace X265_NS;
 using namespace std;
-
-static const char header[] = "FRAME";
-
+static const char header[] = {'F','R','A','M','E'};
 Y4MInput::Y4MInput(InputFileInfo& info)
 {
     for (int i = 0; i < QUEUE_SIZE; i++)
@@ -60,15 +58,14 @@ Y4MInput::Y4MInput(InputFileInfo& info)
     ifs = NULL;
     if (!strcmp(info.filename, "-"))
     {
-        ifs = &cin;
+        ifs = stdin;
 #if _WIN32
         setmode(fileno(stdin), O_BINARY);
 #endif
     }
     else
-        ifs = new ifstream(info.filename, ios::binary | ios::in);
-
-    if (ifs && ifs->good() && parseHeader())
+        ifs = x265_fopen(info.filename, "rb");
+    if (ifs && !ferror(ifs) && parseHeader())
     {
         int pixelbytes = depth > 8 ? 2 : 1;
         for (int i = 0; i < x265_cli_csps[colorSpace].planes; i++)
@@ -91,8 +88,8 @@ Y4MInput::Y4MInput(InputFileInfo& info)
     }
     if (!threadActive)
     {
-        if (ifs && ifs != &cin)
-            delete ifs;
+        if (ifs && ifs != stdin)
+            fclose(ifs);
         ifs = NULL;
         return;
     }
@@ -106,61 +103,34 @@ Y4MInput::Y4MInput(InputFileInfo& info)
     info.csp = colorSpace;
     info.depth = depth;
     info.frameCount = -1;
-
-    size_t estFrameSize = framesize + strlen(header) + 1; /* assume basic FRAME\n headers */
-
+    size_t estFrameSize = framesize + sizeof(header) + 1; /* assume basic FRAME\n headers */
     /* try to estimate frame count, if this is not stdin */
-    if (ifs != &cin)
+    if (ifs != stdin)
     {
-        istream::pos_type cur = ifs->tellg();
-
-#if defined(_MSC_VER) && _MSC_VER < 1700
-        /* Older MSVC versions cannot handle 64bit file sizes properly, so go native */
-        HANDLE hFile = CreateFileA(info.filename, GENERIC_READ,
-                                   FILE_SHARE_READ | FILE_SHARE_WRITE, NULL, OPEN_EXISTING,
-                                   FILE_ATTRIBUTE_NORMAL, NULL);
-        if (hFile != INVALID_HANDLE_VALUE)
-        {
-            LARGE_INTEGER size;
-            if (GetFileSizeEx(hFile, &size))
-                info.frameCount = (int)((size.QuadPart - (int64_t)cur) / estFrameSize);
-            CloseHandle(hFile);
-        }
-#else // if defined(_MSC_VER) && _MSC_VER < 1700
+        int64_t cur = ftello(ifs);
         if (cur >= 0)
         {
-            ifs->seekg(0, ios::end);
-            istream::pos_type size = ifs->tellg();
-            ifs->seekg(cur, ios::beg);
+            fseeko(ifs, 0, SEEK_END);
+            int64_t size = ftello(ifs);
+            fseeko(ifs, cur, SEEK_SET);
             if (size > 0)
                 info.frameCount = (int)((size - cur) / estFrameSize);
         }
-#endif // if defined(_MSC_VER) && _MSC_VER < 1700
     }
-
     if (info.skipFrames)
     {
-#if X86_64
-        if (ifs != &cin)
-            ifs->seekg((uint64_t)estFrameSize * info.skipFrames, ios::cur);
+        if (ifs != stdin)
+            fseeko(ifs, (int64_t)estFrameSize * info.skipFrames, SEEK_CUR);
         else
             for (int i = 0; i < info.skipFrames; i++)
-            {
-                ifs->read(buf[0], estFrameSize - framesize);
-                ifs->read(buf[0], framesize);
-            }
-#else
-        for (int i = 0; i < info.skipFrames; i++)
-            ifs->ignore(estFrameSize);
-#endif
+                if (fread(buf[0], estFrameSize - framesize, 1, ifs) + fread(buf[0], framesize, 1, ifs) != 2)
+                    break;
     }
 }
-
 Y4MInput::~Y4MInput()
 {
-    if (ifs && ifs != &cin)
-        delete ifs;
-
+    if (ifs && ifs != stdin)
+        fclose(ifs);
     for (int i = 0; i < QUEUE_SIZE; i++)
         X265_FREE(buf[i]);
 }
@@ -180,37 +150,31 @@ bool Y4MInput::parseHeader()
 
     int csp = 0;
     int d = 0;
-
-    while (ifs->good())
+    int c;
+    while ((c = fgetc(ifs)) != EOF)
     {
         // Skip Y4MPEG string
-        int c = ifs->get();
-        while (ifs->good() && (c != ' ') && (c != '\n'))
-            c = ifs->get();
-
-        while (c == ' ' && ifs->good())
+        while ((c != EOF) && (c != ' ') && (c != '\n'))
+            c = fgetc(ifs);
+        while (c == ' ')
         {
             // read parameter identifier
-            switch (ifs->get())
+            switch (fgetc(ifs))
             {
             case 'W':
                 width = 0;
-                while (ifs->good())
+                while ((c = fgetc(ifs)) != EOF)
                 {
-                    c = ifs->get();
-
                     if (c == ' ' || c == '\n')
                         break;
                     else
                         width = width * 10 + (c - '0');
                 }
                 break;
-
             case 'H':
                 height = 0;
-                while (ifs->good())
+                while ((c = fgetc(ifs)) != EOF)
                 {
-                    c = ifs->get();
                     if (c == ' ' || c == '\n')
                         break;
                     else
@@ -221,15 +185,13 @@ bool Y4MInput::parseHeader()
             case 'F':
                 rateNum = 0;
                 rateDenom = 0;
-                while (ifs->good())
+                while ((c = fgetc(ifs)) != EOF)
                 {
-                    c = ifs->get();
                     if (c == '.')
                     {
                         rateDenom = 1;
-                        while (ifs->good())
+                        while ((c = fgetc(ifs)) != EOF)
                         {
-                            c = ifs->get();
                             if (c == ' ' || c == '\n')
                                 break;
                             else
@@ -242,9 +204,8 @@ bool Y4MInput::parseHeader()
                     }
                     else if (c == ':')
                     {
-                        while (ifs->good())
+                        while ((c = fgetc(ifs)) != EOF)
                         {
-                            c = ifs->get();
                             if (c == ' ' || c == '\n')
                                 break;
                             else
@@ -260,14 +221,12 @@ bool Y4MInput::parseHeader()
             case 'A':
                 sarWidth = 0;
                 sarHeight = 0;
-                while (ifs->good())
+                while ((c = fgetc(ifs)) != EOF)
                 {
-                    c = ifs->get();
                     if (c == ':')
                     {
-                        while (ifs->good())
+                        while ((c = fgetc(ifs)) != EOF)
                         {
-                            c = ifs->get();
                             if (c == ' ' || c == '\n')
                                 break;
                             else
@@ -283,19 +242,15 @@ bool Y4MInput::parseHeader()
             case 'C':
                 csp = 0;
                 d = 0;
-                while (ifs->good())
+                while ((c = fgetc(ifs)) != EOF)
                 {
-                    c = ifs->get();
-
                     if (c <= 'o' && c >= '0')
                         csp = csp * 10 + (c - '0');
                     else if (c == 'p')
                     {
                         // example: C420p16
-                        while (ifs->good())
+                        while ((c = fgetc(ifs)) != EOF)
                         {
-                            c = ifs->get();
-
                             if (c <= '9' && c >= '0')
                                 d = d * 10 + (c - '0');
                             else
@@ -328,12 +283,10 @@ bool Y4MInput::parseHeader()
                 if (d >= 8 && d <= 16)
                     depth = d;
                 break;
-
             default:
-                while (ifs->good())
+                while ((c = fgetc(ifs)) != EOF)
                 {
                     // consume this unsupported configuration word
-                    c = ifs->get();
                     if (c == ' ' || c == '\n')
                         break;
                 }
@@ -375,30 +328,23 @@ void Y4MInput::threadMain()
     threadActive = false;
     writeCount.poke();
 }
-
 bool Y4MInput::populateFrameQueue()
 {
-    if (!ifs || ifs->fail())
+    if (!ifs || ferror(ifs))
         return false;
-
-    /* strip off the FRAME header */
-    char hbuf[sizeof(header)];
-
-    ifs->read(hbuf, strlen(header));
-    if (ifs->eof())
-        return false;
-
-    if (!ifs->good() || memcmp(hbuf, header, strlen(header)))
+    /* strip off the FRAME\n header */
+    char hbuf[sizeof(header) + 1];
+    if (fread(hbuf, sizeof(hbuf), 1, ifs) != 1 || memcmp(hbuf, header, sizeof(header)))
     {
-        x265_log(NULL, X265_LOG_ERROR, "y4m: frame header missing\n");
+        if (!feof(ifs))
+            x265_log(NULL, X265_LOG_ERROR, "y4m: frame header missing\n");
         return false;
     }
-
     /* consume bytes up to line feed */
-    int c = ifs->get();
-    while (c != '\n' && ifs->good())
-        c = ifs->get();
-
+    int c = hbuf[sizeof(header)];
+    while (c != '\n')
+        if ((c = fgetc(ifs)) == EOF)
+            break;
     /* wait for room in the ring buffer */
     int written = writeCount.get();
     int read = readCount.get();
@@ -408,10 +354,8 @@ bool Y4MInput::populateFrameQueue()
         if (!threadActive)
             return false;
     }
-
     ProfileScopeEvent(frameRead);
-    ifs->read(buf[written % QUEUE_SIZE], framesize);
-    if (ifs->good())
+    if (fread(buf[written % QUEUE_SIZE], framesize, 1, ifs) == 1)
     {
         writeCount.incr();
         return true;
diff --git a/source/input/y4m.h b/source/input/y4m.h
index 3ca9cc67b4..14e80c7567 100644
--- a/source/input/y4m.h
+++ b/source/input/y4m.h
@@ -60,13 +60,9 @@ class Y4MInput : public InputFile, public Thread
     ThreadSafeInteger readCount;
 
     ThreadSafeInteger writeCount;
-
     char* buf[QUEUE_SIZE];
-
-    std::istream *ifs;
-
+    FILE *ifs;
     bool parseHeader();
-
     void threadMain();
 
     bool populateFrameQueue();
@@ -76,15 +72,10 @@ class Y4MInput : public InputFile, public Thread
     Y4MInput(InputFileInfo& info);
 
     virtual ~Y4MInput();
-
     void release();
-
-    bool isEof() const            { return ifs && ifs->eof();  }
-
-    bool isFail()                 { return !(ifs && !ifs->fail() && threadActive); }
-
+    bool isEof() const            { return ifs && feof(ifs); }
+    bool isFail()                 { return !(ifs && !ferror(ifs) && threadActive); }
     void startReader();
-
     bool readPicture(x265_picture&);
 
     const char *getName() const   { return "y4m"; }
diff --git a/source/input/yuv.cpp b/source/input/yuv.cpp
index 64a97cd2a2..e53c8b1ef4 100644
--- a/source/input/yuv.cpp
+++ b/source/input/yuv.cpp
@@ -65,23 +65,21 @@ YUVInput::YUVInput(InputFileInfo& info)
         x265_log(NULL, X265_LOG_ERROR, "yuv: width, height, and FPS must be specified\n");
         return;
     }
-
     if (!strcmp(info.filename, "-"))
     {
-        ifs = &cin;
+        ifs = stdin;
 #if _WIN32
         setmode(fileno(stdin), O_BINARY);
 #endif
     }
     else
-        ifs = new ifstream(info.filename, ios::binary | ios::in);
-
-    if (ifs && ifs->good())
+        ifs = x265_fopen(info.filename, "rb");
+    if (ifs && !ferror(ifs))
         threadActive = true;
     else
     {
-        if (ifs && ifs != &cin)
-            delete ifs;
+        if (ifs && ifs != stdin)
+            fclose(ifs);
         ifs = NULL;
         return;
     }
@@ -98,55 +96,33 @@ YUVInput::YUVInput(InputFileInfo& info)
     }
 
     info.frameCount = -1;
-
     /* try to estimate frame count, if this is not stdin */
-    if (ifs != &cin)
+    if (ifs != stdin)
     {
-        istream::pos_type cur = ifs->tellg();
-
-#if defined(_MSC_VER) && _MSC_VER < 1700
-        /* Older MSVC versions cannot handle 64bit file sizes properly, so go native */
-        HANDLE hFile = CreateFileA(info.filename, GENERIC_READ,
-                                   FILE_SHARE_READ | FILE_SHARE_WRITE, NULL, OPEN_EXISTING,
-                                   FILE_ATTRIBUTE_NORMAL, NULL);
-        if (hFile != INVALID_HANDLE_VALUE)
-        {
-            LARGE_INTEGER size;
-            if (GetFileSizeEx(hFile, &size))
-                info.frameCount = (int)((size.QuadPart - (int64_t)cur) / framesize);
-            CloseHandle(hFile);
-        }
-#else // if defined(_MSC_VER) && _MSC_VER < 1700
+        int64_t cur = ftello(ifs);
         if (cur >= 0)
         {
-            ifs->seekg(0, ios::end);
-            istream::pos_type size = ifs->tellg();
-            ifs->seekg(cur, ios::beg);
+            fseeko(ifs, 0, SEEK_END);
+            int64_t size = ftello(ifs);
+            fseeko(ifs, cur, SEEK_SET);
             if (size > 0)
                 info.frameCount = (int)((size - cur) / framesize);
         }
-#endif // if defined(_MSC_VER) && _MSC_VER < 1700
     }
-
     if (info.skipFrames)
     {
-#if X86_64
-        if (ifs != &cin)
-            ifs->seekg((uint64_t)framesize * info.skipFrames, ios::cur);
+        if (ifs != stdin)
+            fseeko(ifs, (int64_t)framesize * info.skipFrames, SEEK_CUR);
         else
             for (int i = 0; i < info.skipFrames; i++)
-                ifs->read(buf[0], framesize);
-#else
-        for (int i = 0; i < info.skipFrames; i++)
-            ifs->ignore(framesize);
-#endif
+                if (fread(buf[0], framesize, 1, ifs) != 1)
+                    break;
     }
 }
-
 YUVInput::~YUVInput()
 {
-    if (ifs && ifs != &cin)
-        delete ifs;
+    if (ifs && ifs != stdin)
+        fclose(ifs);
     for (int i = 0; i < QUEUE_SIZE; i++)
         X265_FREE(buf[i]);
 }
@@ -179,12 +155,10 @@ void YUVInput::threadMain()
     threadActive = false;
     writeCount.poke();
 }
-
 bool YUVInput::populateFrameQueue()
 {
-    if (!ifs || ifs->fail())
+    if (!ifs || ferror(ifs))
         return false;
-
     /* wait for room in the ring buffer */
     int written = writeCount.get();
     int read = readCount.get();
@@ -195,10 +169,8 @@ bool YUVInput::populateFrameQueue()
             // release() has been called
             return false;
     }
-
     ProfileScopeEvent(frameRead);
-    ifs->read(buf[written % QUEUE_SIZE], framesize);
-    if (ifs->good())
+    if (fread(buf[written % QUEUE_SIZE], framesize, 1, ifs) == 1)
     {
         writeCount.incr();
         return true;
diff --git a/source/input/yuv.h b/source/input/yuv.h
index dd33cb3633..a67b22c535 100644
--- a/source/input/yuv.h
+++ b/source/input/yuv.h
@@ -52,13 +52,9 @@ class YUVInput : public InputFile, public Thread
     ThreadSafeInteger readCount;
 
     ThreadSafeInteger writeCount;
-
     char* buf[QUEUE_SIZE];
-
-    std::istream *ifs;
-
+    FILE *ifs;
     int guessFrameCount();
-
     void threadMain();
 
     bool populateFrameQueue();
@@ -68,13 +64,9 @@ class YUVInput : public InputFile, public Thread
     YUVInput(InputFileInfo& info);
 
     virtual ~YUVInput();
-
     void release();
-
-    bool isEof() const                            { return ifs && ifs->eof();  }
-
-    bool isFail()                                 { return !(ifs && !ifs->fail() && threadActive); }
-
+    bool isEof() const                            { return ifs && feof(ifs); }
+    bool isFail()                                 { return !(ifs && !ferror(ifs) && threadActive); }
     void startReader();
 
     bool readPicture(x265_picture&);

From a0fee97c00816aa01cbcd88f8d8ade337cadd453 Mon Sep 17 00:00:00 2001
From: Ashok Kumar Mishra <ashok@multicorewareinc.com>
Date: Fri, 19 Jan 2018 14:46:54 +0530
Subject: [PATCH 33/51] fix for unable to open input file for gcc 32 bit
 compiler

---
 source/input/y4m.cpp | 3 ++-
 source/input/yuv.cpp | 3 ++-
 2 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/source/input/y4m.cpp b/source/input/y4m.cpp
index a1f6379f5f..8a52496679 100644
--- a/source/input/y4m.cpp
+++ b/source/input/y4m.cpp
@@ -20,7 +20,8 @@
  * This program is also available under a commercial proprietary license.
  * For more information, contact us at license @ x265.com.
  *****************************************************************************/
-
+#define _FILE_OFFSET_BITS 64
+#define _LARGEFILE_SOURCE
 #include "y4m.h"
 #include "common.h"
 
diff --git a/source/input/yuv.cpp b/source/input/yuv.cpp
index e53c8b1ef4..ddfff59e44 100644
--- a/source/input/yuv.cpp
+++ b/source/input/yuv.cpp
@@ -20,7 +20,8 @@
  * This program is also available under a commercial proprietary license.
  * For more information, contact us at license @ x265.com.
  *****************************************************************************/
-
+#define _FILE_OFFSET_BITS 64
+#define _LARGEFILE_SOURCE
 #include "yuv.h"
 #include "common.h"
 

From 81039fa83d7855c411b40c422bb58998a1443467 Mon Sep 17 00:00:00 2001
From: Santhoshini Sekar <santhoshini@multicorewareinc.com>
Date: Wed, 24 Jan 2018 13:11:11 +0530
Subject: [PATCH 34/51] fix warning in getRefFrameList

---
 source/encoder/encoder.cpp | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/source/encoder/encoder.cpp b/source/encoder/encoder.cpp
index 15b6c0c622..c47912187f 100644
--- a/source/encoder/encoder.cpp
+++ b/source/encoder/encoder.cpp
@@ -490,7 +490,10 @@ int Encoder::getRefFrameList(PicYuv** l0, PicYuv** l1, int sliceType, int poc, i
             }
         }
         else
-            x265_log(NULL, X265_LOG_WARNING, "Refrence List is not in piclist\n");
+        {
+            x265_log(NULL, X265_LOG_WARNING, "Current frame is not in DPB piclist.\n");
+            return 1;
+        }
     }
     else
     {

From ad6d7e090726f2545765f315caf600929ca81fda Mon Sep 17 00:00:00 2001
From: Ricardo Constantino <wiiaboo@gmail.com>
Date: Tue, 16 Jan 2018 17:39:53 +0000
Subject: [PATCH 35/51] CMake: blacklist mingw implicit link libraries

These also aren't meant to be in pkg-config's Libs.Private.
---
 source/CMakeLists.txt | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/source/CMakeLists.txt b/source/CMakeLists.txt
index 557814ad40..e6d49c69a6 100644
--- a/source/CMakeLists.txt
+++ b/source/CMakeLists.txt
@@ -647,7 +647,9 @@ if(X265_LATEST_TAG)
     endforeach()
     if(PLIBLIST)
         # blacklist of libraries that should not be in Libs.private
-        list(REMOVE_ITEM PLIBLIST "-lc" "-lpthread")
+        list(REMOVE_ITEM PLIBLIST "-lc" "-lpthread" "-lmingwex" "-lmingwthrd"
+            "-lmingw32" "-lmoldname" "-lmsvcrt" "-ladvapi32" "-lshell32"
+            "-luser32" "-lkernel32")
         string(REPLACE ";" " " PRIVATE_LIBS "${PLIBLIST}")
     else()
         set(PRIVATE_LIBS "")

From 8535ec0405b1e7bf1691af7f047e61e28bd3228e Mon Sep 17 00:00:00 2001
From: Aruna Matheswaran <aruna@multicorewareinc.com>
Date: Wed, 31 Jan 2018 18:37:13 +0530
Subject: [PATCH 36/51] slicetype: fix hanging issue due to dropping of BREF
 frames

While forcing slicetypes through qp file, few BREF frames were not copied into
lookahead's output queue due to incorrect check condition.
---
 source/encoder/slicetype.cpp | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/source/encoder/slicetype.cpp b/source/encoder/slicetype.cpp
index eb81529c65..c01beacdd3 100644
--- a/source/encoder/slicetype.cpp
+++ b/source/encoder/slicetype.cpp
@@ -1234,9 +1234,8 @@ void Lookahead::slicetypeDecide()
     int idx = 0;
     list[bframes]->m_reorderedPts = pts[idx++];
     m_outputQueue.pushBack(*list[bframes]);
-
     /* Add B-ref frame next to P frame in output queue, the B-ref encode before non B-ref frame */
-    if (bframes > 1 && m_param->bBPyramid)
+    if (brefs)
     {
         for (int i = 0; i < bframes; i++)
         {

From 7c9bbbcb9af2531529c24c38f6511f0499cc3319 Mon Sep 17 00:00:00 2001
From: Bhavna Hariharan <bhavna@multicorewareinc.com>
Date: Wed, 31 Jan 2018 19:44:19 +0530
Subject: [PATCH 37/51] csv: Bypass luma calculatations when --max-cll is OFF
 or when csv-log-level < 2

The maxFall and maxCll values are calculated from the min, max and average luma
values. The luma values were being calculated even when --max-cll is disabled.
This patch bypasses the luma calculations when --max-cll is OFF or
when csv-log-level < 2
---
 source/common/picyuv.cpp   | 18 ++++++++++--------
 source/encoder/api.cpp     | 19 ++++++++++++-------
 source/encoder/encoder.cpp | 27 +++++++++++++--------------
 3 files changed, 35 insertions(+), 29 deletions(-)

diff --git a/source/common/picyuv.cpp b/source/common/picyuv.cpp
index 311a7bd68d..062f8facff 100644
--- a/source/common/picyuv.cpp
+++ b/source/common/picyuv.cpp
@@ -358,18 +358,20 @@ void PicYuv::copyFromPicture(const x265_picture& pic, const x265_param& param, i
     pixel *uPic = m_picOrg[1];
     pixel *vPic = m_picOrg[2];
 
-    for (int r = 0; r < height; r++)
+    if (param.csvLogLevel >= 2 || param.maxCLL || param.maxFALL)
     {
-        for (int c = 0; c < width; c++)
+        for (int r = 0; r < height; r++)
         {
-            m_maxLumaLevel = X265_MAX(yPic[c], m_maxLumaLevel);
-            m_minLumaLevel = X265_MIN(yPic[c], m_minLumaLevel);
-            lumaSum += yPic[c];
+            for (int c = 0; c < width; c++)
+            {
+                m_maxLumaLevel = X265_MAX(yPic[c], m_maxLumaLevel);
+                m_minLumaLevel = X265_MIN(yPic[c], m_minLumaLevel);
+                lumaSum += yPic[c];
+            }
+            yPic += m_stride;
         }
-        yPic += m_stride;
+        m_avgLumaLevel = (double)lumaSum / (m_picHeight * m_picWidth);
     }
-    m_avgLumaLevel = (double)lumaSum / (m_picHeight * m_picWidth);
-
     if (param.csvLogLevel >= 2)
     {
         if (param.internalCsp != X265_CSP_I400)
diff --git a/source/encoder/api.cpp b/source/encoder/api.cpp
index 85de1d762c..9ea28c2dc6 100644
--- a/source/encoder/api.cpp
+++ b/source/encoder/api.cpp
@@ -67,9 +67,7 @@ static const char* summaryCSVHeader =
     "Y PSNR, U PSNR, V PSNR, Global PSNR, SSIM, SSIM (dB), "
     "I count, I ave-QP, I kbps, I-PSNR Y, I-PSNR U, I-PSNR V, I-SSIM (dB), "
     "P count, P ave-QP, P kbps, P-PSNR Y, P-PSNR U, P-PSNR V, P-SSIM (dB), "
-    "B count, B ave-QP, B kbps, B-PSNR Y, B-PSNR U, B-PSNR V, B-SSIM (dB), "
-    "MaxCLL, MaxFALL, Version\n";
-
+    "B count, B ave-QP, B kbps, B-PSNR Y, B-PSNR U, B-PSNR V, B-SSIM (dB), ";
 x265_encoder *x265_encoder_open(x265_param *p)
 {
     if (!p)
@@ -757,7 +755,12 @@ FILE* x265_csvlog_open(const x265_param* param)
                 fprintf(csvfp, "\n");
             }
             else
+            {
                 fputs(summaryCSVHeader, csvfp);
+                if (param->csvLogLevel >= 2 || param->maxCLL || param->maxFALL)
+                    fputs("MaxCLL, MaxFALL,", csvfp);
+                fputs(" Version\n", csvfp);
+            }
         }
         return csvfp;
     }
@@ -881,8 +884,10 @@ void x265_csvlog_encode(const x265_param *p, const x265_stats *stats, int padx,
             // adding summary to a per-frame csv log file, so it needs a summary header
             fprintf(p->csvfpt, "\nSummary\n");
             fputs(summaryCSVHeader, p->csvfpt);
+            if (p->csvLogLevel >= 2 || p->maxCLL || p->maxFALL)
+                fputs("MaxCLL, MaxFALL,", p->csvfpt);
+            fputs(" Version\n",p->csvfpt);
         }
-
         // CLI arguments or other
         if (argc)
         {
@@ -973,10 +978,10 @@ void x265_csvlog_encode(const x265_param *p, const x265_stats *stats, int padx,
         }
         else
             fprintf(p->csvfpt, " -, -, -, -, -, -, -,");
-
-        fprintf(p->csvfpt, " %-6u, %-6u, %s\n", stats->maxCLL, stats->maxFALL, api->version_str);
+        if (p->csvLogLevel >= 2 || p->maxCLL || p->maxFALL)
+            fprintf(p->csvfpt, " %-6u, %-6u,", stats->maxCLL, stats->maxFALL);
+        fprintf(p->csvfpt, " %s\n", api->version_str);
     }
-
 }
 
 /* The dithering algorithm is based on Sierra-2-4A error diffusion.
diff --git a/source/encoder/encoder.cpp b/source/encoder/encoder.cpp
index c47912187f..9325ed19c9 100644
--- a/source/encoder/encoder.cpp
+++ b/source/encoder/encoder.cpp
@@ -1983,11 +1983,12 @@ void Encoder::fetchStats(x265_stats *stats, size_t statsSizeBytes)
         stats->statsB.psnrU   = m_analyzeB.m_psnrSumU / (double)m_analyzeB.m_numPics;
         stats->statsB.psnrV   = m_analyzeB.m_psnrSumV / (double)m_analyzeB.m_numPics;
         stats->statsB.ssim    = x265_ssim2dB(m_analyzeB.m_globalSsim / (double)m_analyzeB.m_numPics);
-
-        stats->maxCLL         = m_analyzeAll.m_maxCLL;
-        stats->maxFALL        = (uint16_t)(m_analyzeAll.m_maxFALL / m_analyzeAll.m_numPics);
+        if (m_param->csvLogLevel >= 2 || m_param->maxCLL || m_param->maxFALL)
+        {
+            stats->maxCLL = m_analyzeAll.m_maxCLL;
+            stats->maxFALL = (uint16_t)(m_analyzeAll.m_maxFALL / m_analyzeAll.m_numPics);
+        }
     }
-
     /* If new statistics are added to x265_stats, we must check here whether the
      * structure provided by the user is the new structure or an older one (for
      * future safety) */
@@ -2059,10 +2060,11 @@ void Encoder::finishFrameStats(Frame* curFrame, FrameEncoder *curEncoder, x265_f
         if (m_param->bEnableSsim)
             m_analyzeB.addSsim(ssim);
     }
-
-    m_analyzeAll.m_maxFALL += curFrame->m_fencPic->m_avgLumaLevel;
-    m_analyzeAll.m_maxCLL = X265_MAX(m_analyzeAll.m_maxCLL, curFrame->m_fencPic->m_maxLumaLevel);
-
+    if (m_param->csvLogLevel >= 2 || m_param->maxCLL || m_param->maxFALL)
+    {
+        m_analyzeAll.m_maxFALL += curFrame->m_fencPic->m_avgLumaLevel;
+        m_analyzeAll.m_maxCLL = X265_MAX(m_analyzeAll.m_maxCLL, curFrame->m_fencPic->m_maxLumaLevel);
+    }
     char c = (slice->isIntra() ? (curFrame->m_lowres.sliceType == X265_TYPE_IDR ? 'I' : 'i') : slice->isInterP() ? 'P' : 'B');
     int poc = slice->m_poc;
     if (!IS_REFERENCED(curFrame))
@@ -2101,13 +2103,7 @@ void Encoder::finishFrameStats(Frame* curFrame, FrameEncoder *curEncoder, x265_f
                     frameStats->list1POC[ref] = ref < slice->m_numRefIdx[1] ? slice->m_refPOCList[1][ref] - slice->m_lastIDR : -1;
             }
         }
-
 #define ELAPSED_MSEC(start, end) (((double)(end) - (start)) / 1000)
-
-        frameStats->maxLumaLevel = curFrame->m_fencPic->m_maxLumaLevel;
-        frameStats->minLumaLevel = curFrame->m_fencPic->m_minLumaLevel;
-        frameStats->avgLumaLevel = curFrame->m_fencPic->m_avgLumaLevel;
-
         if (m_param->csvLogLevel >= 2)
         {
             frameStats->decideWaitTime = ELAPSED_MSEC(0, curEncoder->m_slicetypeWaitTime);
@@ -2127,6 +2123,9 @@ void Encoder::finishFrameStats(Frame* curFrame, FrameEncoder *curEncoder, x265_f
             frameStats->avgLumaDistortion = curFrame->m_encData->m_frameStats.avgLumaDistortion;
             frameStats->avgPsyEnergy = curFrame->m_encData->m_frameStats.avgPsyEnergy;
             frameStats->avgResEnergy = curFrame->m_encData->m_frameStats.avgResEnergy;
+            frameStats->maxLumaLevel = curFrame->m_fencPic->m_maxLumaLevel;
+            frameStats->minLumaLevel = curFrame->m_fencPic->m_minLumaLevel;
+            frameStats->avgLumaLevel = curFrame->m_fencPic->m_avgLumaLevel;
 
             frameStats->maxChromaULevel = curFrame->m_fencPic->m_maxChromaULevel;
             frameStats->minChromaULevel = curFrame->m_fencPic->m_minChromaULevel;

From 39ed4b6e54448bff16f28d9ba082e5fdbf512ff1 Mon Sep 17 00:00:00 2001
From: David Christenson <mail@noctem.xyz>
Date: Fri, 2 Feb 2018 22:18:18 -0700
Subject: [PATCH 38/51] CMake: fix generation of version info from
 .hg_archival.txt

When building from a commit more recent than the latest tag, version.cmake
wouldn't set X265_LATEST_TAG and X265_TAG_DISTANCE resulting in their defaults
(0.0 and 0) being used. Also truncate the revision ID to 12 characters to match
the length used when parsing from the hg executable.
---
 source/cmake/version.cmake | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/source/cmake/version.cmake b/source/cmake/version.cmake
index c63d9f442e..b1e995f924 100644
--- a/source/cmake/version.cmake
+++ b/source/cmake/version.cmake
@@ -22,12 +22,11 @@ if(EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/../.hg_archival.txt)
         set(hg_${key} ${value})
     endforeach()
     if(DEFINED hg_tag)
-        set(X265_VERSION ${hg_tag})
         set(X265_LATEST_TAG ${hg_tag})
-        set(X265_TAG_DISTANCE "0")
     elseif(DEFINED hg_node)
-        string(SUBSTRING "${hg_node}" 0 16 hg_id)
-        set(X265_VERSION "${hg_latesttag}+${hg_latesttagdistance}-${hg_id}")
+        set(X265_LATEST_TAG ${hg_latesttag})
+        set(X265_TAG_DISTANCE ${hg_latesttagdistance})
+        string(SUBSTRING "${hg_node}" 0 12 X265_REVISION_ID)
     endif()
 elseif(HG_EXECUTABLE AND EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/../.hg)
     if(EXISTS "${HG_EXECUTABLE}.bat")

From 2b8f6dba0b860102f877ee710abc23c5ec8d8b75 Mon Sep 17 00:00:00 2001
From: Ma0 <mateuszb@poczta.onet.pl>
Date: Sat, 10 Feb 2018 06:16:45 +0100
Subject: [PATCH 39/51] fix output to pipe on Windows

---
 source/output/raw.cpp | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/source/output/raw.cpp b/source/output/raw.cpp
index 1529908d50..fe69084829 100644
--- a/source/output/raw.cpp
+++ b/source/output/raw.cpp
@@ -21,18 +21,26 @@
  * This program is also available under a commercial proprietary license.
  * For more information, contact us at license @ x265.com.
  *****************************************************************************/
-
 #include "raw.h"
+#if _WIN32
+#include <io.h>
+#include <fcntl.h>
+#if defined(_MSC_VER)
+#pragma warning(disable: 4996) // POSIX setmode and fileno deprecated
+#endif
+#endif
 
 using namespace X265_NS;
 using namespace std;
-
 RAWOutput::RAWOutput(const char* fname, InputFileInfo&)
 {
     b_fail = false;
     if (!strcmp(fname, "-"))
     {
         ofs = stdout;
+#if _WIN32
+        setmode(fileno(stdout), O_BINARY);
+#endif
         return;
     }
     ofs = x265_fopen(fname, "wb");

From 4c2d4a96ea24af1eb58adaa78c01e27b0e624c89 Mon Sep 17 00:00:00 2001
From: Bhavna Hariharan <bhavna@multicorewareinc.com>
Date: Wed, 14 Feb 2018 15:45:29 +0530
Subject: [PATCH 40/51] limitTU: Save intra CU's TU depth when analysis
 save/load is enabled

This patch will cause a output mismatch between analysis save and load when
limit-tu 3/4 is enabled. This change is expected as the load run will have only
the best mode's TU information. For CUs where the neighbour's TU depth is
unavailable, load encode will evaluate all TU detphs.
---
 doc/reST/cli.rst                 | 8 +++++++-
 source/encoder/analysis.cpp      | 5 ++---
 source/test/regression-tests.txt | 2 +-
 3 files changed, 10 insertions(+), 5 deletions(-)

diff --git a/doc/reST/cli.rst b/doc/reST/cli.rst
index 83a5eb6f6b..ff70169d8a 100644
--- a/doc/reST/cli.rst
+++ b/doc/reST/cli.rst
@@ -1029,7 +1029,13 @@ as the residual quad-tree (RQT).
 	Level 4 - uses the depth of the neighbouring/ co-located CUs TU depth 
 	to limit the 1st subTU depth. The 1st subTU depth is taken as the 
 	limiting depth for the other subTUs.
-
+	
+	Enabling levels 3 or 4 may cause a mismatch in the output bitstreams 
+	between option:`--analysis-save` and option:`--analysis-load`
+	as all neighbouring CUs TU depth may not be available in the 
+	option:`--analysis-load` run as only the best mode's information is 
+	available to it.
+	
 	Default: 0
 
 .. option:: --nr-intra <integer>, --nr-inter <integer>
diff --git a/source/encoder/analysis.cpp b/source/encoder/analysis.cpp
index 3622ed51d6..a8e69c2db9 100644
--- a/source/encoder/analysis.cpp
+++ b/source/encoder/analysis.cpp
@@ -647,13 +647,12 @@ uint64_t Analysis::compressIntraCU(const CUData& parentCTU, const CUGeom& cuGeom
         cacheCost[cuIdx] = md.bestMode->rdCost;
     }
 
-    /* Save Intra CUs TU depth only when analysis mode is OFF */
-    if ((m_limitTU & X265_TU_LIMIT_NEIGH) && cuGeom.log2CUSize >= 4 && (!m_param->analysisSave && !m_param->analysisLoad))
+    if ((m_limitTU & X265_TU_LIMIT_NEIGH) && cuGeom.log2CUSize >= 4)
     {
         CUData* ctu = md.bestMode->cu.m_encData->getPicCTU(parentCTU.m_cuAddr);
         int8_t maxTUDepth = -1;
         for (uint32_t i = 0; i < cuGeom.numPartitions; i++)
-            maxTUDepth = X265_MAX(maxTUDepth, md.pred[PRED_INTRA].cu.m_tuDepth[i]);
+            maxTUDepth = X265_MAX(maxTUDepth, md.bestMode->cu.m_tuDepth[i]);
         ctu->m_refTuDepth[cuGeom.geomRecurId] = maxTUDepth;
     }
 
diff --git a/source/test/regression-tests.txt b/source/test/regression-tests.txt
index 51a1b0f338..d5543dda1c 100644
--- a/source/test/regression-tests.txt
+++ b/source/test/regression-tests.txt
@@ -23,7 +23,7 @@ BasketballDrive_1920x1080_50.y4m,--preset slow --nr-intra 100 -F4 --aq-strength
 BasketballDrive_1920x1080_50.y4m,--preset slower --lossless --chromaloc 3 --subme 0 --limit-tu 4
 BasketballDrive_1920x1080_50.y4m,--preset slower --no-cutree --analysis-save x265_analysis.dat --analysis-reuse-level 10 --bitrate 7000 --limit-tu 0::--preset slower --no-cutree --analysis-load x265_analysis.dat --analysis-reuse-level 10 --bitrate 7000 --limit-tu 0
 BasketballDrive_1920x1080_50.y4m,--preset veryslow --crf 4 --cu-lossless --pmode --limit-refs 1 --aq-mode 3 --limit-tu 3
-BasketballDrive_1920x1080_50.y4m,--preset veryslow --no-cutree --analysis-save x265_analysis.dat --bitrate 7000 --tskip-fast --limit-tu 4::--preset veryslow --no-cutree --analysis-load x265_analysis.dat --bitrate 7000  --tskip-fast --limit-tu 4
+BasketballDrive_1920x1080_50.y4m,--preset veryslow --no-cutree --analysis-save x265_analysis.dat --bitrate 7000 --tskip-fast --limit-tu 2::--preset veryslow --no-cutree --analysis-load x265_analysis.dat --bitrate 7000  --tskip-fast --limit-tu 2
 BasketballDrive_1920x1080_50.y4m,--preset veryslow --recon-y4m-exec "ffplay -i pipe:0 -autoexit"
 Coastguard-4k.y4m,--preset ultrafast --recon-y4m-exec "ffplay -i pipe:0 -autoexit"
 Coastguard-4k.y4m,--preset superfast --tune grain --overscan=crop

From 4617e86ec94d68fab403a4e100fca6afa1d181f2 Mon Sep 17 00:00:00 2001
From: Bhavna Hariharan <bhavna@multicorewareinc.com>
Date: Fri, 16 Feb 2018 11:37:45 +0530
Subject: [PATCH 41/51] dhdr: comply to HDR10+ LLC spec

This patch does the following -
1) Consider Json files without Bezier Curve data as valid and inject them
correctly without errors.
2) The numWindows values is now taken from the numWindows JSON value, instead
of taking it from 1+ellipsesNum. Add numWindows to the dictionary.
3) Changed hardcoded applicationVersion value to 1 due to LLC requirement.
---
 source/dynamicHDR10/SeiMetadataDictionary.cpp |  1 +
 source/dynamicHDR10/SeiMetadataDictionary.h   |  1 +
 source/dynamicHDR10/metadataFromJson.cpp      | 57 +++++++++----------
 3 files changed, 28 insertions(+), 31 deletions(-)

diff --git a/source/dynamicHDR10/SeiMetadataDictionary.cpp b/source/dynamicHDR10/SeiMetadataDictionary.cpp
index 039a4887e2..ad95570926 100644
--- a/source/dynamicHDR10/SeiMetadataDictionary.cpp
+++ b/source/dynamicHDR10/SeiMetadataDictionary.cpp
@@ -28,6 +28,7 @@ using namespace SeiMetadataDictionary;
 
 const std::string JsonDataKeys::LocalParameters = std::string("LocalParameters");
 const std::string JsonDataKeys::TargetDisplayLuminance = std::string("TargetedSystemDisplayMaximumLuminance");
+const std::string JsonDataKeys::NumberOfWindows = std::string("NumberOfWindows");
 
 const std::string BezierCurveNames::TagName = std::string("BezierCurveData");
 const std::string BezierCurveNames::NumberOfAnchors = std::string("NumberOfAnchors");
diff --git a/source/dynamicHDR10/SeiMetadataDictionary.h b/source/dynamicHDR10/SeiMetadataDictionary.h
index 3406652936..727311cc00 100644
--- a/source/dynamicHDR10/SeiMetadataDictionary.h
+++ b/source/dynamicHDR10/SeiMetadataDictionary.h
@@ -37,6 +37,7 @@ namespace SeiMetadataDictionary
         public:
         static const std::string LocalParameters;
         static const std::string TargetDisplayLuminance;
+        static const std::string NumberOfWindows;		
     };
 
     //Bezier Curve Data
diff --git a/source/dynamicHDR10/metadataFromJson.cpp b/source/dynamicHDR10/metadataFromJson.cpp
index 64bf2f8b9b..1e2e1d5ceb 100644
--- a/source/dynamicHDR10/metadataFromJson.cpp
+++ b/source/dynamicHDR10/metadataFromJson.cpp
@@ -372,7 +372,7 @@ void metadataFromJson::fillMetadataArray(const JsonArray &fileData, int frame, u
     const uint16_t terminalProviderCode = 0x003C;
     const uint16_t terminalProviderOrientedCode = 0x0001;
     const uint8_t applicationIdentifier = 4;
-    const uint8_t applicationVersion = 0;
+    const uint8_t applicationVersion = 1;
 
     mPimpl->appendBits(metadata, countryCode, 8);
     mPimpl->appendBits(metadata, terminalProviderCode, 16);
@@ -384,9 +384,7 @@ void metadataFromJson::fillMetadataArray(const JsonArray &fileData, int frame, u
     //Note: Validated only add up to two local selections, ignore the rest
     JsonArray jsonArray = fileData[frame][JsonDataKeys::LocalParameters].array_items();
     int ellipsesNum = static_cast<int>(jsonArray.size() > 2 ? 2 : jsonArray.size());
-
-    uint16_t numWindows = 1 + static_cast<uint16_t>(ellipsesNum);
-
+    uint16_t numWindows = (uint16_t)fileData[frame][JsonDataKeys::NumberOfWindows].int_value();
     mPimpl->appendBits(metadata, numWindows, 2);
     for (int i = 0; i < ellipsesNum; ++i)
     {
@@ -426,16 +424,15 @@ void metadataFromJson::fillMetadataArray(const JsonArray &fileData, int frame, u
 
         mPimpl->appendBits(metadata, semimajorExternalAxis, 16);
         mPimpl->appendBits(metadata, semiminorExternalAxis, 16);
-        /*bool*/ uint8_t overlapProcessOption = static_cast<uint8_t>(ellipseJsonObject[EllipseNames::OverlapProcessOption].int_value()); //1; 
+        uint8_t overlapProcessOption = static_cast<uint8_t>(ellipseJsonObject[EllipseNames::OverlapProcessOption].int_value());
         //TODO: Uses Layering method, the value is "1"
         mPimpl->appendBits(metadata, overlapProcessOption, 1);
     }
     /* Targeted System Display Data */
-    uint32_t TEMPmonitorPeak = fileData[frame][JsonDataKeys::TargetDisplayLuminance].int_value();     //500;
-    mPimpl->appendBits(metadata, TEMPmonitorPeak, 27);
-
+    uint32_t monitorPeak = fileData[frame][JsonDataKeys::TargetDisplayLuminance].int_value();     //500;
+    mPimpl->appendBits(metadata, monitorPeak, 27);
     //NOTE: Set as false for now, as requested
-    /*bool*/uint8_t targetedSystemDisplayActualPeakLuminanceFlag = 0; /*false*/
+    uint8_t targetedSystemDisplayActualPeakLuminanceFlag = 0;
     mPimpl->appendBits(metadata, targetedSystemDisplayActualPeakLuminanceFlag, 1);
     if (targetedSystemDisplayActualPeakLuminanceFlag)
     {
@@ -463,7 +460,6 @@ void metadataFromJson::fillMetadataArray(const JsonArray &fileData, int frame, u
         mPimpl->appendBits(metadata, static_cast<uint16_t>((int)luminanceData.maxGLuminance & 0xFFFF), 16);
         mPimpl->appendBits(metadata, static_cast<uint8_t>(((int)luminanceData.maxBLuminance & 0x10000) >> 16), 1);
         mPimpl->appendBits(metadata, static_cast<uint16_t>((int)luminanceData.maxBLuminance & 0xFFFF), 16);
-        /* changed from maxRGBLuminance to average luminance to match stms implementation */
         mPimpl->appendBits(metadata, static_cast<uint8_t>(((int)luminanceData.averageLuminance & 0x10000) >> 16), 1);
         mPimpl->appendBits(metadata, static_cast<uint16_t>((int)luminanceData.averageLuminance & 0xFFFF), 16);
 
@@ -478,7 +474,7 @@ void metadataFromJson::fillMetadataArray(const JsonArray &fileData, int frame, u
             uint8_t distributionMaxrgbPercentage = static_cast<uint8_t>(percentilPercentages.at(i));
             mPimpl->appendBits(metadata, distributionMaxrgbPercentage, 7);
 
-            // 17bits: 1bit then 16
+            /* 17bits: 1bit then 16 */
             unsigned int ithPercentile = luminanceData.percentiles.at(i);
             uint8_t highValue = static_cast<uint8_t>((ithPercentile & 0x10000) >> 16);
             uint16_t lowValue = static_cast<uint16_t>(ithPercentile & 0xFFFF);
@@ -499,33 +495,32 @@ void metadataFromJson::fillMetadataArray(const JsonArray &fileData, int frame, u
     {
         //TODO
     }
-    // BEZIER CURVE DATA
+    /* Bezier Curve Data */
     for (int w = 0; w < numWindows; ++w)
     {
-        //TODO: 
         uint8_t toneMappingFlag = 1;
-        mPimpl->appendBits(metadata, toneMappingFlag, 1);
-        if (toneMappingFlag)
+		/* Check if the window contains tone mapping bezier curve data and set toneMappingFlag appropriately */
+		//Json bezierData = fileData[frame][BezierCurveNames::TagName];
+        BezierCurveData curveData;
+		/* Select curve data based on global window */
+        if (w == 0)
         {
-            Json bezierData = fileData[frame][BezierCurveNames::TagName];
-            BezierCurveData curveData;
-
-            /* Select curve data based on global window or local window */
-            if (w == 0)
+            if (!mPimpl->bezierCurveFromJson(fileData[frame][BezierCurveNames::TagName], curveData))
             {
-                if (!mPimpl->bezierCurveFromJson(bezierData, curveData))
-                {
-                    std::cout << "error parsing bezierCurve frame: " << w << std::endl;
-                }
+				toneMappingFlag = 0;
             }
-            else
+        }
+	    /* Select curve data based on local window */
+        else
+        {
+            if (!mPimpl->bezierCurveFromJson(jsonArray[w - 1][BezierCurveNames::TagName], curveData))
             {
-                if (!mPimpl->bezierCurveFromJson(jsonArray[w - 1][BezierCurveNames::TagName], curveData))
-                {
-                    std::cout << "error parsing bezierCurve ellipse: " << w - 1 << std::endl;
-                }
+				toneMappingFlag = 0;
             }
-
+        }		
+        mPimpl->appendBits(metadata, toneMappingFlag, 1);
+        if (toneMappingFlag)
+        {
             uint16_t kneePointX = static_cast<uint16_t>(curveData.sPx);
             mPimpl->appendBits(metadata, kneePointX, 12);
             uint16_t kneePointY = static_cast<uint16_t>(curveData.sPy);
@@ -541,7 +536,7 @@ void metadataFromJson::fillMetadataArray(const JsonArray &fileData, int frame, u
                 mPimpl->appendBits(metadata, anchor, 10);
             }
         }
-    }
+	}
     /* Set to false as requested */
     bool colorSaturationMappingFlag = 0;
     mPimpl->appendBits(metadata, colorSaturationMappingFlag, 1);

From 9936a4760c7470637f29bdcbf7cbd8cdcbe942d3 Mon Sep 17 00:00:00 2001
From: Bhavna Hariharan <bhavna@multicorewareinc.com>
Date: Fri, 16 Feb 2018 11:40:59 +0530
Subject: [PATCH 42/51] dhdr: Skip comments in the json file while parsing it.

Comments of the type /* */ OR // are ignored.
---
 source/dynamicHDR10/JsonHelper.cpp | 43 +++++++-----------------------
 1 file changed, 10 insertions(+), 33 deletions(-)

diff --git a/source/dynamicHDR10/JsonHelper.cpp b/source/dynamicHDR10/JsonHelper.cpp
index 4257313faf..b073f3f3ff 100644
--- a/source/dynamicHDR10/JsonHelper.cpp
+++ b/source/dynamicHDR10/JsonHelper.cpp
@@ -139,21 +139,13 @@ JsonObject JsonHelper::readJson(string path)
         return JsonObject();
     }
 
-    ifstream tfile;
-    string json_str;
-    string json_str2;
+    std::ifstream ifs(path);
+    const std::string json_str2((std::istreambuf_iterator<char>(ifs)),
+                                (std::istreambuf_iterator<char>()));
+
     string err = "";
-    tfile.open(path);
-    while(tfile)
-    {
-        std::getline(tfile, json_str);
-        json_str2.append(json_str);
-    }
 
-    tfile.close();
-    size_t beginning = json_str2.find_first_of("{");
-    int fixchar = json_str2[json_str2.size() - 2] == '}' ? 1 : 0;
-    return Json::parse(json_str2.substr(beginning,json_str2.size() - fixchar),err).object_items();
+    return Json::parse(json_str2,err, JsonParse::COMMENTS).object_items();
 }
 
 JsonArray JsonHelper::readJsonArray(const string &path)
@@ -174,28 +166,13 @@ JsonArray JsonHelper::readJsonArray(const string &path)
         return JsonArray();
     }
 
-    ifstream tfile;
-    string json_str;
-    string json_str2;
-    string err = "";
-    tfile.open(path);
-    while(tfile)
-    {
-        std::getline(tfile, json_str);
-        json_str2.append(json_str);
-    }
+    std::ifstream ifs(path);
+    const std::string json_str2((std::istreambuf_iterator<char>(ifs)),
+                                (std::istreambuf_iterator<char>()));
 
-    tfile.close();
+    string err = "";
 
-    vector<Json> data;
-    if (json_str2.size() != 0)
-    {
-        size_t beginning = json_str2.find_first_of("[");
-        int fixchar = json_str2[json_str2.size() - 2] == ']' ? 1 : 0;
-        return Json::parse(json_str2.substr(beginning, json_str2.size() - fixchar), err).array_items();
-    }
-    else
-        return data;
+    return Json::parse(json_str2,err, JsonParse::COMMENTS).array_items();
 }
 
 bool JsonHelper::validatePathExtension(string &path)

From 537f741d89c622fd2889550fc15425ec5909e72a Mon Sep 17 00:00:00 2001
From: Praveen Tiwari <praveen@multicorewareinc.com>
Date: Fri, 19 Jan 2018 12:04:23 +0530
Subject: [PATCH 43/51] x86: split ipfilter8 kernels into two different source
 file

This patch implements infrastructure to split ipfiletr8 asm source file into two
different files in order to avoid longer build time. It moves interp_8tap_horizontal
kernels to the newly created file.
---
 source/common/CMakeLists.txt         |   6 +-
 source/common/x86/asm-primitives.cpp |   2 +-
 source/common/x86/h-ipfilter8.asm    | 267 +++++++++++++++++++++++++++
 source/common/x86/h-ipfilter8.h      |  39 ++++
 source/common/x86/ipfilter8.asm      | 131 -------------
 5 files changed, 309 insertions(+), 136 deletions(-)
 create mode 100644 source/common/x86/h-ipfilter8.asm
 create mode 100644 source/common/x86/h-ipfilter8.h

diff --git a/source/common/CMakeLists.txt b/source/common/CMakeLists.txt
index 65624069c4..021645337c 100644
--- a/source/common/CMakeLists.txt
+++ b/source/common/CMakeLists.txt
@@ -56,17 +56,15 @@ if(ENABLE_ASSEMBLY AND X86)
     endif()
     set(VEC_PRIMITIVES vec/vec-primitives.cpp ${PRIMITIVES})
     source_group(Intrinsics FILES ${VEC_PRIMITIVES})
-
-    set(C_SRCS asm-primitives.cpp pixel.h mc.h ipfilter8.h blockcopy8.h dct8.h loopfilter.h seaintegral.h)
+    set(C_SRCS asm-primitives.cpp pixel.h mc.h ipfilter8.h blockcopy8.h dct8.h h-ipfilter8.h loopfilter.h seaintegral.h)
     set(A_SRCS pixel-a.asm const-a.asm cpu-a.asm ssd-a.asm mc-a.asm
                mc-a2.asm pixel-util8.asm blockcopy8.asm
                pixeladd8.asm dct8.asm seaintegral.asm)
     if(HIGH_BIT_DEPTH)
         set(A_SRCS ${A_SRCS} sad16-a.asm intrapred16.asm ipfilter16.asm loopfilter.asm)
     else()
-        set(A_SRCS ${A_SRCS} sad-a.asm intrapred8.asm intrapred8_allangs.asm ipfilter8.asm loopfilter.asm)
+        set(A_SRCS ${A_SRCS} sad-a.asm intrapred8.asm intrapred8_allangs.asm h-ipfilter8.asm ipfilter8.asm loopfilter.asm)
     endif()
-
     if(NOT X64)
         set(A_SRCS ${A_SRCS} pixel-32.asm)
     endif()
diff --git a/source/common/x86/asm-primitives.cpp b/source/common/x86/asm-primitives.cpp
index 1546734c91..0e54329bc1 100644
--- a/source/common/x86/asm-primitives.cpp
+++ b/source/common/x86/asm-primitives.cpp
@@ -115,8 +115,8 @@ extern "C" {
 #include "intrapred.h"
 #include "dct8.h"
 #include "seaintegral.h"
+#include "h-ipfilter8.h"
 }
-
 #define ALL_LUMA_CU_TYPED(prim, fncdef, fname, cpu) \
     p.cu[BLOCK_8x8].prim   = fncdef PFX(fname ## _8x8_ ## cpu); \
     p.cu[BLOCK_16x16].prim = fncdef PFX(fname ## _16x16_ ## cpu); \
diff --git a/source/common/x86/h-ipfilter8.asm b/source/common/x86/h-ipfilter8.asm
new file mode 100644
index 0000000000..b5ce12c1ea
--- /dev/null
+++ b/source/common/x86/h-ipfilter8.asm
@@ -0,0 +1,267 @@
+;*****************************************************************************
+;* Copyright (C) 2013-2017 MulticoreWare, Inc
+;*
+;* Authors: Min Chen <chenm003@163.com>
+;*          Nabajit Deka <nabajit@multicorewareinc.com>
+;*          Praveen Kumar Tiwari <praveen@multicorewareinc.com>
+;*
+;* This program is free software; you can redistribute it and/or modify
+;* it under the terms of the GNU General Public License as published by
+;* the Free Software Foundation; either version 2 of the License, or
+;* (at your option) any later version.
+;*
+;* This program is distributed in the hope that it will be useful,
+;* but WITHOUT ANY WARRANTY; without even the implied warranty of
+;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+;* GNU General Public License for more details.
+;*
+;* You should have received a copy of the GNU General Public License
+;* along with this program; if not, write to the Free Software
+;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
+;*
+;* This program is also available under a commercial proprietary license.
+;* For more information, contact us at license @ x265.com.
+;*****************************************************************************/
+
+%include "x86inc.asm"
+%include "x86util.asm"
+
+SECTION_RODATA 32
+
+const h_tabw_LumaCoeff,  dw   0, 0,  0,  64,  0,   0,  0,  0
+                       dw  -1, 4, -10, 58,  17, -5,  1,  0
+                       dw  -1, 4, -11, 40,  40, -11, 4, -1
+                       dw   0, 1, -5,  17,  58, -10, 4, -1
+
+SECTION .text
+
+cextern pw_32
+cextern pw_2000
+
+%macro FILTER_H8_W8_sse2 0
+    movh        m1, [r0 + x - 3]
+    movh        m4, [r0 + x - 2]
+    punpcklbw   m1, m6
+    punpcklbw   m4, m6
+    movh        m5, [r0 + x - 1]
+    movh        m0, [r0 + x]
+    punpcklbw   m5, m6
+    punpcklbw   m0, m6
+    pmaddwd     m1, m3
+    pmaddwd     m4, m3
+    pmaddwd     m5, m3
+    pmaddwd     m0, m3
+    packssdw    m1, m4
+    packssdw    m5, m0
+    pshuflw     m4, m1, q2301
+    pshufhw     m4, m4, q2301
+    pshuflw     m0, m5, q2301
+    pshufhw     m0, m0, q2301
+    paddw       m1, m4
+    paddw       m5, m0
+    psrldq      m1, 2
+    psrldq      m5, 2
+    pshufd      m1, m1, q3120
+    pshufd      m5, m5, q3120
+    punpcklqdq  m1, m5
+    movh        m7, [r0 + x + 1]
+    movh        m4, [r0 + x + 2]
+    punpcklbw   m7, m6
+    punpcklbw   m4, m6
+    movh        m5, [r0 + x + 3]
+    movh        m0, [r0 + x + 4]
+    punpcklbw   m5, m6
+    punpcklbw   m0, m6
+    pmaddwd     m7, m3
+    pmaddwd     m4, m3
+    pmaddwd     m5, m3
+    pmaddwd     m0, m3
+    packssdw    m7, m4
+    packssdw    m5, m0
+    pshuflw     m4, m7, q2301
+    pshufhw     m4, m4, q2301
+    pshuflw     m0, m5, q2301
+    pshufhw     m0, m0, q2301
+    paddw       m7, m4
+    paddw       m5, m0
+    psrldq      m7, 2
+    psrldq      m5, 2
+    pshufd      m7, m7, q3120
+    pshufd      m5, m5, q3120
+    punpcklqdq  m7, m5
+    pshuflw     m4, m1, q2301
+    pshufhw     m4, m4, q2301
+    pshuflw     m0, m7, q2301
+    pshufhw     m0, m0, q2301
+    paddw       m1, m4
+    paddw       m7, m0
+    psrldq      m1, 2
+    psrldq      m7, 2
+    pshufd      m1, m1, q3120
+    pshufd      m7, m7, q3120
+    punpcklqdq  m1, m7
+%endmacro
+
+%macro FILTER_H8_W4_sse2 0
+    movh        m1, [r0 + x - 3]
+    movh        m0, [r0 + x - 2]
+    punpcklbw   m1, m6
+    punpcklbw   m0, m6
+    movh        m4, [r0 + x - 1]
+    movh        m5, [r0 + x]
+    punpcklbw   m4, m6
+    punpcklbw   m5, m6
+    pmaddwd     m1, m3
+    pmaddwd     m0, m3
+    pmaddwd     m4, m3
+    pmaddwd     m5, m3
+    packssdw    m1, m0
+    packssdw    m4, m5
+    pshuflw     m0, m1, q2301
+    pshufhw     m0, m0, q2301
+    pshuflw     m5, m4, q2301
+    pshufhw     m5, m5, q2301
+    paddw       m1, m0
+    paddw       m4, m5
+    psrldq      m1, 2
+    psrldq      m4, 2
+    pshufd      m1, m1, q3120
+    pshufd      m4, m4, q3120
+    punpcklqdq  m1, m4
+    pshuflw     m0, m1, q2301
+    pshufhw     m0, m0, q2301
+    paddw       m1, m0
+    psrldq      m1, 2
+    pshufd      m1, m1, q3120
+%endmacro
+
+;----------------------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;----------------------------------------------------------------------------------------------------------------------------
+%macro IPFILTER_LUMA_sse2 3
+INIT_XMM sse2
+cglobal interp_8tap_horiz_%3_%1x%2, 4,6,8
+    mov       r4d, r4m
+    add       r4d, r4d
+    pxor      m6, m6
+
+%ifidn %3, ps
+    add       r3d, r3d
+    cmp       r5m, byte 0
+%endif
+
+%ifdef PIC
+    lea       r5, [h_tabw_LumaCoeff]
+    movu      m3, [r5 + r4 * 8]
+%else
+    movu      m3, [h_tabw_LumaCoeff + r4 * 8]
+%endif
+
+    mov       r4d, %2
+
+%ifidn %3, pp
+    mova      m2, [pw_32]
+%else
+    mova      m2, [pw_2000]
+    je        .loopH
+    lea       r5, [r1 + 2 * r1]
+    sub       r0, r5
+    add       r4d, 7
+%endif
+
+.loopH:
+%assign x 0
+%rep %1 / 8
+    FILTER_H8_W8_sse2
+  %ifidn %3, pp
+    paddw     m1, m2
+    psraw     m1, 6
+    packuswb  m1, m1
+    movh      [r2 + x], m1
+  %else
+    psubw     m1, m2
+    movu      [r2 + 2 * x], m1
+  %endif
+%assign x x+8
+%endrep
+
+%rep (%1 % 8) / 4
+    FILTER_H8_W4_sse2
+  %ifidn %3, pp
+    paddw     m1, m2
+    psraw     m1, 6
+    packuswb  m1, m1
+    movd      [r2 + x], m1
+  %else
+    psubw     m1, m2
+    movh      [r2 + 2 * x], m1
+  %endif
+%endrep
+
+    add       r0, r1
+    add       r2, r3
+
+    dec       r4d
+    jnz       .loopH
+    RET
+
+%endmacro
+
+;--------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+    IPFILTER_LUMA_sse2 4, 4, pp
+    IPFILTER_LUMA_sse2 4, 8, pp
+    IPFILTER_LUMA_sse2 8, 4, pp
+    IPFILTER_LUMA_sse2 8, 8, pp
+    IPFILTER_LUMA_sse2 16, 16, pp
+    IPFILTER_LUMA_sse2 16, 8, pp
+    IPFILTER_LUMA_sse2 8, 16, pp
+    IPFILTER_LUMA_sse2 16, 12, pp
+    IPFILTER_LUMA_sse2 12, 16, pp
+    IPFILTER_LUMA_sse2 16, 4, pp
+    IPFILTER_LUMA_sse2 4, 16, pp
+    IPFILTER_LUMA_sse2 32, 32, pp
+    IPFILTER_LUMA_sse2 32, 16, pp
+    IPFILTER_LUMA_sse2 16, 32, pp
+    IPFILTER_LUMA_sse2 32, 24, pp
+    IPFILTER_LUMA_sse2 24, 32, pp
+    IPFILTER_LUMA_sse2 32, 8, pp
+    IPFILTER_LUMA_sse2 8, 32, pp
+    IPFILTER_LUMA_sse2 64, 64, pp
+    IPFILTER_LUMA_sse2 64, 32, pp
+    IPFILTER_LUMA_sse2 32, 64, pp
+    IPFILTER_LUMA_sse2 64, 48, pp
+    IPFILTER_LUMA_sse2 48, 64, pp
+    IPFILTER_LUMA_sse2 64, 16, pp
+    IPFILTER_LUMA_sse2 16, 64, pp
+
+;----------------------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_ps_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;----------------------------------------------------------------------------------------------------------------------------
+    IPFILTER_LUMA_sse2 4, 4, ps
+    IPFILTER_LUMA_sse2 8, 8, ps
+    IPFILTER_LUMA_sse2 8, 4, ps
+    IPFILTER_LUMA_sse2 4, 8, ps
+    IPFILTER_LUMA_sse2 16, 16, ps
+    IPFILTER_LUMA_sse2 16, 8, ps
+    IPFILTER_LUMA_sse2 8, 16, ps
+    IPFILTER_LUMA_sse2 16, 12, ps
+    IPFILTER_LUMA_sse2 12, 16, ps
+    IPFILTER_LUMA_sse2 16, 4, ps
+    IPFILTER_LUMA_sse2 4, 16, ps
+    IPFILTER_LUMA_sse2 32, 32, ps
+    IPFILTER_LUMA_sse2 32, 16, ps
+    IPFILTER_LUMA_sse2 16, 32, ps
+    IPFILTER_LUMA_sse2 32, 24, ps
+    IPFILTER_LUMA_sse2 24, 32, ps
+    IPFILTER_LUMA_sse2 32, 8, ps
+    IPFILTER_LUMA_sse2 8, 32, ps
+    IPFILTER_LUMA_sse2 64, 64, ps
+    IPFILTER_LUMA_sse2 64, 32, ps
+    IPFILTER_LUMA_sse2 32, 64, ps
+    IPFILTER_LUMA_sse2 64, 48, ps
+    IPFILTER_LUMA_sse2 48, 64, ps
+    IPFILTER_LUMA_sse2 64, 16, ps
+    IPFILTER_LUMA_sse2 16, 64, ps
+
diff --git a/source/common/x86/h-ipfilter8.h b/source/common/x86/h-ipfilter8.h
new file mode 100644
index 0000000000..45ea99293d
--- /dev/null
+++ b/source/common/x86/h-ipfilter8.h
@@ -0,0 +1,39 @@
+/*****************************************************************************
+* Copyright (C) 2013-2017 MulticoreWare, Inc
+*
+* Authors: Steve Borho <steve@borho.org>
+*          Praveen Kuamr Tiwari <praveen@multicorewareinc.com>
+*
+* This program is free software; you can redistribute it and/or modify
+* it under the terms of the GNU General Public License as published by
+* the Free Software Foundation; either version 2 of the License, or
+* (at your option) any later version.
+*
+* This program is distributed in the hope that it will be useful,
+* but WITHOUT ANY WARRANTY; without even the implied warranty of
+* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+* GNU General Public License for more details.
+*
+* You should have received a copy of the GNU General Public License
+* along with this program; if not, write to the Free Software
+* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
+*
+* This program is also available under a commercial proprietary license.
+* For more information, contact us at license @ x265.com.
+*****************************************************************************/
+
+#ifndef X265_H_IPFILTER8_H
+#define X265_H_IPFILTER8_H
+
+
+#define SETUP_H_FUNC_DEF(cpu) \
+    FUNCDEF_PU(void, interp_8tap_horiz_pp, cpu, const pixel* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int coeffIdx); \
+    FUNCDEF_PU(void, interp_8tap_horiz_ps, cpu, const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);
+
+SETUP_H_FUNC_DEF(sse2);
+SETUP_H_FUNC_DEF(ssse3);
+SETUP_H_FUNC_DEF(sse3);
+SETUP_H_FUNC_DEF(sse4);
+SETUP_H_FUNC_DEF(avx2);
+
+#endif // ifndef X265_H_IPFILTER8_H
diff --git a/source/common/x86/ipfilter8.asm b/source/common/x86/ipfilter8.asm
index 36d8986b5a..dc3f04add0 100644
--- a/source/common/x86/ipfilter8.asm
+++ b/source/common/x86/ipfilter8.asm
@@ -855,137 +855,6 @@ cglobal interp_4tap_horiz_ps_%1x%2, 4, 7, 6, src, srcstride, dst, dststride
     psrldq      m1, 2
     pshufd      m1, m1, q3120
 %endmacro
-
-;----------------------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;----------------------------------------------------------------------------------------------------------------------------
-%macro IPFILTER_LUMA_sse2 3
-INIT_XMM sse2
-cglobal interp_8tap_horiz_%3_%1x%2, 4,6,8
-    mov       r4d, r4m
-    add       r4d, r4d
-    pxor      m6, m6
-
-%ifidn %3, ps
-    add       r3d, r3d
-    cmp       r5m, byte 0
-%endif
-
-%ifdef PIC
-    lea       r5, [tabw_LumaCoeff]
-    movu      m3, [r5 + r4 * 8]
-%else
-    movu      m3, [tabw_LumaCoeff + r4 * 8]
-%endif
-
-    mov       r4d, %2
-
-%ifidn %3, pp
-    mova      m2, [pw_32]
-%else
-    mova      m2, [pw_2000]
-    je        .loopH
-    lea       r5, [r1 + 2 * r1]
-    sub       r0, r5
-    add       r4d, 7
-%endif
-
-.loopH:
-%assign x 0
-%rep %1 / 8
-    FILTER_H8_W8_sse2
-  %ifidn %3, pp
-    paddw     m1, m2
-    psraw     m1, 6
-    packuswb  m1, m1
-    movh      [r2 + x], m1
-  %else
-    psubw     m1, m2
-    movu      [r2 + 2 * x], m1
-  %endif
-%assign x x+8
-%endrep
-
-%rep (%1 % 8) / 4
-    FILTER_H8_W4_sse2
-  %ifidn %3, pp
-    paddw     m1, m2
-    psraw     m1, 6
-    packuswb  m1, m1
-    movd      [r2 + x], m1
-  %else
-    psubw     m1, m2
-    movh      [r2 + 2 * x], m1
-  %endif
-%endrep
-
-    add       r0, r1
-    add       r2, r3
-
-    dec       r4d
-    jnz       .loopH
-    RET
-
-%endmacro
-
-;--------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-    IPFILTER_LUMA_sse2 4, 4, pp
-    IPFILTER_LUMA_sse2 4, 8, pp
-    IPFILTER_LUMA_sse2 8, 4, pp
-    IPFILTER_LUMA_sse2 8, 8, pp
-    IPFILTER_LUMA_sse2 16, 16, pp
-    IPFILTER_LUMA_sse2 16, 8, pp
-    IPFILTER_LUMA_sse2 8, 16, pp
-    IPFILTER_LUMA_sse2 16, 12, pp
-    IPFILTER_LUMA_sse2 12, 16, pp
-    IPFILTER_LUMA_sse2 16, 4, pp
-    IPFILTER_LUMA_sse2 4, 16, pp
-    IPFILTER_LUMA_sse2 32, 32, pp
-    IPFILTER_LUMA_sse2 32, 16, pp
-    IPFILTER_LUMA_sse2 16, 32, pp
-    IPFILTER_LUMA_sse2 32, 24, pp
-    IPFILTER_LUMA_sse2 24, 32, pp
-    IPFILTER_LUMA_sse2 32, 8, pp
-    IPFILTER_LUMA_sse2 8, 32, pp
-    IPFILTER_LUMA_sse2 64, 64, pp
-    IPFILTER_LUMA_sse2 64, 32, pp
-    IPFILTER_LUMA_sse2 32, 64, pp
-    IPFILTER_LUMA_sse2 64, 48, pp
-    IPFILTER_LUMA_sse2 48, 64, pp
-    IPFILTER_LUMA_sse2 64, 16, pp
-    IPFILTER_LUMA_sse2 16, 64, pp
-
-;----------------------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_ps_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;----------------------------------------------------------------------------------------------------------------------------
-    IPFILTER_LUMA_sse2 4, 4, ps
-    IPFILTER_LUMA_sse2 8, 8, ps
-    IPFILTER_LUMA_sse2 8, 4, ps
-    IPFILTER_LUMA_sse2 4, 8, ps
-    IPFILTER_LUMA_sse2 16, 16, ps
-    IPFILTER_LUMA_sse2 16, 8, ps
-    IPFILTER_LUMA_sse2 8, 16, ps
-    IPFILTER_LUMA_sse2 16, 12, ps
-    IPFILTER_LUMA_sse2 12, 16, ps
-    IPFILTER_LUMA_sse2 16, 4, ps
-    IPFILTER_LUMA_sse2 4, 16, ps
-    IPFILTER_LUMA_sse2 32, 32, ps
-    IPFILTER_LUMA_sse2 32, 16, ps
-    IPFILTER_LUMA_sse2 16, 32, ps
-    IPFILTER_LUMA_sse2 32, 24, ps
-    IPFILTER_LUMA_sse2 24, 32, ps
-    IPFILTER_LUMA_sse2 32, 8, ps
-    IPFILTER_LUMA_sse2 8, 32, ps
-    IPFILTER_LUMA_sse2 64, 64, ps
-    IPFILTER_LUMA_sse2 64, 32, ps
-    IPFILTER_LUMA_sse2 32, 64, ps
-    IPFILTER_LUMA_sse2 64, 48, ps
-    IPFILTER_LUMA_sse2 48, 64, ps
-    IPFILTER_LUMA_sse2 64, 16, ps
-    IPFILTER_LUMA_sse2 16, 64, ps
-
 %macro PROCESS_LUMA_W4_4R_sse2 0
     movd        m2,     [r0]
     movd        m7,     [r0 + r1]

From c01c61058ff7ed31ee456688c1a8b5e0096b552f Mon Sep 17 00:00:00 2001
From: Radhakrishnan <radhakrishnan@multicorewareinc.com>
Date: Tue, 20 Feb 2018 14:59:18 +0530
Subject: [PATCH 44/51] x86: Remove header file for horizontal ipfilters

Since all ipfilter kernals share the ipfilter8.h file, a separate header file isn't needed for horizontal ipfilter kernals. So this patch removes the duplicate header file
---
 source/common/CMakeLists.txt         |  2 +-
 source/common/x86/asm-primitives.cpp |  1 -
 source/common/x86/h-ipfilter8.h      | 39 ----------------------------
 3 files changed, 1 insertion(+), 41 deletions(-)
 delete mode 100644 source/common/x86/h-ipfilter8.h

diff --git a/source/common/CMakeLists.txt b/source/common/CMakeLists.txt
index 021645337c..ab980a223e 100644
--- a/source/common/CMakeLists.txt
+++ b/source/common/CMakeLists.txt
@@ -56,7 +56,7 @@ if(ENABLE_ASSEMBLY AND X86)
     endif()
     set(VEC_PRIMITIVES vec/vec-primitives.cpp ${PRIMITIVES})
     source_group(Intrinsics FILES ${VEC_PRIMITIVES})
-    set(C_SRCS asm-primitives.cpp pixel.h mc.h ipfilter8.h blockcopy8.h dct8.h h-ipfilter8.h loopfilter.h seaintegral.h)
+    set(C_SRCS asm-primitives.cpp pixel.h mc.h ipfilter8.h blockcopy8.h dct8.h loopfilter.h seaintegral.h)
     set(A_SRCS pixel-a.asm const-a.asm cpu-a.asm ssd-a.asm mc-a.asm
                mc-a2.asm pixel-util8.asm blockcopy8.asm
                pixeladd8.asm dct8.asm seaintegral.asm)
diff --git a/source/common/x86/asm-primitives.cpp b/source/common/x86/asm-primitives.cpp
index 0e54329bc1..cf9c7de59b 100644
--- a/source/common/x86/asm-primitives.cpp
+++ b/source/common/x86/asm-primitives.cpp
@@ -115,7 +115,6 @@ extern "C" {
 #include "intrapred.h"
 #include "dct8.h"
 #include "seaintegral.h"
-#include "h-ipfilter8.h"
 }
 #define ALL_LUMA_CU_TYPED(prim, fncdef, fname, cpu) \
     p.cu[BLOCK_8x8].prim   = fncdef PFX(fname ## _8x8_ ## cpu); \
diff --git a/source/common/x86/h-ipfilter8.h b/source/common/x86/h-ipfilter8.h
deleted file mode 100644
index 45ea99293d..0000000000
--- a/source/common/x86/h-ipfilter8.h
+++ /dev/null
@@ -1,39 +0,0 @@
-/*****************************************************************************
-* Copyright (C) 2013-2017 MulticoreWare, Inc
-*
-* Authors: Steve Borho <steve@borho.org>
-*          Praveen Kuamr Tiwari <praveen@multicorewareinc.com>
-*
-* This program is free software; you can redistribute it and/or modify
-* it under the terms of the GNU General Public License as published by
-* the Free Software Foundation; either version 2 of the License, or
-* (at your option) any later version.
-*
-* This program is distributed in the hope that it will be useful,
-* but WITHOUT ANY WARRANTY; without even the implied warranty of
-* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-* GNU General Public License for more details.
-*
-* You should have received a copy of the GNU General Public License
-* along with this program; if not, write to the Free Software
-* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
-*
-* This program is also available under a commercial proprietary license.
-* For more information, contact us at license @ x265.com.
-*****************************************************************************/
-
-#ifndef X265_H_IPFILTER8_H
-#define X265_H_IPFILTER8_H
-
-
-#define SETUP_H_FUNC_DEF(cpu) \
-    FUNCDEF_PU(void, interp_8tap_horiz_pp, cpu, const pixel* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int coeffIdx); \
-    FUNCDEF_PU(void, interp_8tap_horiz_ps, cpu, const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);
-
-SETUP_H_FUNC_DEF(sse2);
-SETUP_H_FUNC_DEF(ssse3);
-SETUP_H_FUNC_DEF(sse3);
-SETUP_H_FUNC_DEF(sse4);
-SETUP_H_FUNC_DEF(avx2);
-
-#endif // ifndef X265_H_IPFILTER8_H

From 3f7829dd497154ea0b62dd1d47db16f3634a8556 Mon Sep 17 00:00:00 2001
From: Radhakrishnan <radhakrishnan@multicorewareinc.com>
Date: Wed, 24 Jan 2018 16:18:24 +0530
Subject: [PATCH 45/51] x86: Split ipfilter8 kernals into two different source
 files, part2

This patch port the horizontal 4tap and 8tap kernals from ipfilter8.asm to a different source file h-ipfilter8.asm. It improves the build time by 20%.
---
 source/common/x86/h-ipfilter8.asm |  6469 ++++++
 source/common/x86/ipfilter8.asm   | 34217 ++++++++++++----------------
 2 files changed, 20558 insertions(+), 20128 deletions(-)

diff --git a/source/common/x86/h-ipfilter8.asm b/source/common/x86/h-ipfilter8.asm
index b5ce12c1ea..48024c7740 100644
--- a/source/common/x86/h-ipfilter8.asm
+++ b/source/common/x86/h-ipfilter8.asm
@@ -33,10 +33,459 @@ const h_tabw_LumaCoeff,  dw   0, 0,  0,  64,  0,   0,  0,  0
                        dw  -1, 4, -11, 40,  40, -11, 4, -1
                        dw   0, 1, -5,  17,  58, -10, 4, -1
 
+const h_tabw_ChromaCoeff, dw  0, 64,  0,  0
+                        dw -2, 58, 10, -2
+                        dw -4, 54, 16, -2
+                        dw -6, 46, 28, -4
+                        dw -4, 36, 36, -4
+                        dw -4, 28, 46, -6
+                        dw -2, 16, 54, -4
+                        dw -2, 10, 58, -2
+
+const h_tab_ChromaCoeff, db  0, 64,  0,  0
+                       db -2, 58, 10, -2
+                       db -4, 54, 16, -2
+                       db -6, 46, 28, -4
+                       db -4, 36, 36, -4
+                       db -4, 28, 46, -6
+                       db -2, 16, 54, -4
+                       db -2, 10, 58, -2
+
+const h_tab_LumaCoeff,   db   0, 0,  0,  64,  0,   0,  0,  0
+                       db  -1, 4, -10, 58,  17, -5,  1,  0
+                       db  -1, 4, -11, 40,  40, -11, 4, -1
+                       db   0, 1, -5,  17,  58, -10, 4, -1
+
+const h_tab_Tm,    db 0, 1, 2, 3, 1, 2, 3, 4, 2, 3, 4, 5, 3, 4, 5, 6
+                 db 4, 5, 6, 7, 5, 6, 7, 8, 6, 7, 8, 9, 7, 8, 9, 10
+                 db 8, 9,10,11, 9,10,11,12,10,11,12,13,11,12,13, 14
+
+
+const h_tab_Lm,    db 0, 1, 2, 3, 4,  5,  6,  7,  1, 2, 3, 4,  5,  6,  7,  8
+                 db 2, 3, 4, 5, 6,  7,  8,  9,  3, 4, 5, 6,  7,  8,  9,  10
+                 db 4, 5, 6, 7, 8,  9,  10, 11, 5, 6, 7, 8,  9,  10, 11, 12
+                 db 6, 7, 8, 9, 10, 11, 12, 13, 7, 8, 9, 10, 11, 12, 13, 14
+
+const interp4_shuf, times 2 db 0, 1, 8, 9, 4, 5, 12, 13, 2, 3, 10, 11, 6, 7, 14, 15
+
+const h_interp8_hps_shuf,     dd 0, 4, 1, 5, 2, 6, 3, 7
+
+const interp4_hpp_shuf,     times 2 db 0, 1, 2, 3, 1, 2, 3, 4, 8, 9, 10, 11, 9, 10, 11, 12
+
+const interp4_horiz_shuf1,  db 0, 1, 2, 3, 1, 2, 3, 4, 2, 3, 4, 5, 3, 4, 5, 6
+                            db 8, 9, 10, 11, 9, 10, 11, 12, 10, 11, 12, 13, 11, 12, 13, 14
+
+const h_pd_526336, times 8 dd 8192*64+2048
+
+const pb_LumaCoeffVer,  times 16 db 0, 0
+                        times 16 db 0, 64
+                        times 16 db 0, 0
+                        times 16 db 0, 0
+
+                        times 16 db -1, 4
+                        times 16 db -10, 58
+                        times 16 db 17, -5
+                        times 16 db 1, 0
+
+                        times 16 db -1, 4
+                        times 16 db -11, 40
+                        times 16 db 40, -11
+                        times 16 db 4, -1
+
+                        times 16 db 0, 1
+                        times 16 db -5, 17
+                        times 16 db 58, -10
+                        times 16 db 4, -1
+
+const h_pw_LumaCoeffVer,  times 8 dw 0, 0
+                        times 8 dw 0, 64
+                        times 8 dw 0, 0
+                        times 8 dw 0, 0
+
+                        times 8 dw -1, 4
+                        times 8 dw -10, 58
+                        times 8 dw 17, -5
+                        times 8 dw 1, 0
+
+                        times 8 dw -1, 4
+                        times 8 dw -11, 40
+                        times 8 dw 40, -11
+                        times 8 dw 4, -1
+
+                        times 8 dw 0, 1
+                        times 8 dw -5, 17
+                        times 8 dw 58, -10
+                        times 8 dw 4, -1
+
+const pb_8tap_hps_0, times 2 db 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8
+                     times 2 db 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9,10
+                     times 2 db 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9,10,10,11,11,12
+                     times 2 db 6, 7, 7, 8, 8, 9, 9,10,10,11,11,12,12,13,13,14
+
+ALIGN 32
+interp4_hps_shuf: times 2 db 0, 1, 2, 3, 1, 2, 3, 4, 8, 9, 10, 11, 9, 10, 11, 12
+
 SECTION .text
 
+cextern pw_1
 cextern pw_32
 cextern pw_2000
+cextern pw_512
+
+%macro PROCESS_LUMA_AVX2_W8_16R 1
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m2, [r5]
+    lea             r7, [r0 + r1 * 4]
+    movu            xm4, [r7]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    paddd           m1, m5
+    pmaddwd         m3, [r5]
+    movu            xm5, [r7 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 2 * mmsize]
+    paddd           m0, m6
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m2, m6
+    pmaddwd         m4, [r5]
+    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm7, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddwd         m7, m5, [r5 + 2 * mmsize]
+    paddd           m1, m7
+    pmaddwd         m7, m5, [r5 + 1 * mmsize]
+    paddd           m3, m7
+    pmaddwd         m5, [r5]
+    movu            xm7, [r7 + r4]                  ; m7 = row 7
+    punpckhwd       xm8, xm6, xm7
+    punpcklwd       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddwd         m8, m6, [r5 + 3 * mmsize]
+    paddd           m0, m8
+    pmaddwd         m8, m6, [r5 + 2 * mmsize]
+    paddd           m2, m8
+    pmaddwd         m8, m6, [r5 + 1 * mmsize]
+    paddd           m4, m8
+    pmaddwd         m6, [r5]
+    lea             r7, [r7 + r1 * 4]
+    movu            xm8, [r7]                       ; m8 = row 8
+    punpckhwd       xm9, xm7, xm8
+    punpcklwd       xm7, xm8
+    vinserti128     m7, m7, xm9, 1
+    pmaddwd         m9, m7, [r5 + 3 * mmsize]
+    paddd           m1, m9
+    pmaddwd         m9, m7, [r5 + 2 * mmsize]
+    paddd           m3, m9
+    pmaddwd         m9, m7, [r5 + 1 * mmsize]
+    paddd           m5, m9
+    pmaddwd         m7, [r5]
+    movu            xm9, [r7 + r1]                  ; m9 = row 9
+    punpckhwd       xm10, xm8, xm9
+    punpcklwd       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddwd         m10, m8, [r5 + 3 * mmsize]
+    paddd           m2, m10
+    pmaddwd         m10, m8, [r5 + 2 * mmsize]
+    paddd           m4, m10
+    pmaddwd         m10, m8, [r5 + 1 * mmsize]
+    paddd           m6, m10
+    pmaddwd         m8, [r5]
+    movu            xm10, [r7 + r1 * 2]             ; m10 = row 10
+    punpckhwd       xm11, xm9, xm10
+    punpcklwd       xm9, xm10
+    vinserti128     m9, m9, xm11, 1
+    pmaddwd         m11, m9, [r5 + 3 * mmsize]
+    paddd           m3, m11
+    pmaddwd         m11, m9, [r5 + 2 * mmsize]
+    paddd           m5, m11
+    pmaddwd         m11, m9, [r5 + 1 * mmsize]
+    paddd           m7, m11
+    pmaddwd         m9, [r5]
+    movu            xm11, [r7 + r4]                 ; m11 = row 11
+    punpckhwd       xm12, xm10, xm11
+    punpcklwd       xm10, xm11
+    vinserti128     m10, m10, xm12, 1
+    pmaddwd         m12, m10, [r5 + 3 * mmsize]
+    paddd           m4, m12
+    pmaddwd         m12, m10, [r5 + 2 * mmsize]
+    paddd           m6, m12
+    pmaddwd         m12, m10, [r5 + 1 * mmsize]
+    paddd           m8, m12
+    pmaddwd         m10, [r5]
+    lea             r7, [r7 + r1 * 4]
+    movu            xm12, [r7]                      ; m12 = row 12
+    punpckhwd       xm13, xm11, xm12
+    punpcklwd       xm11, xm12
+    vinserti128     m11, m11, xm13, 1
+    pmaddwd         m13, m11, [r5 + 3 * mmsize]
+    paddd           m5, m13
+    pmaddwd         m13, m11, [r5 + 2 * mmsize]
+    paddd           m7, m13
+    pmaddwd         m13, m11, [r5 + 1 * mmsize]
+    paddd           m9, m13
+    pmaddwd         m11, [r5]
+
+%ifidn %1,sp
+    paddd           m0, m14
+    paddd           m1, m14
+    paddd           m2, m14
+    paddd           m3, m14
+    paddd           m4, m14
+    paddd           m5, m14
+    psrad           m0, 12
+    psrad           m1, 12
+    psrad           m2, 12
+    psrad           m3, 12
+    psrad           m4, 12
+    psrad           m5, 12
+%else
+    psrad           m0, 6
+    psrad           m1, 6
+    psrad           m2, 6
+    psrad           m3, 6
+    psrad           m4, 6
+    psrad           m5, 6
+%endif
+    packssdw        m0, m1
+    packssdw        m2, m3
+    packssdw        m4, m5
+%ifidn %1,sp
+    packuswb        m0, m2
+    mova            m5, [h_interp8_hps_shuf]
+    vpermd          m0, m5, m0
+    vextracti128    xm2, m0, 1
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r6], xm2
+%else
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m2, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
+%endif
+
+    movu            xm13, [r7 + r1]                 ; m13 = row 13
+    punpckhwd       xm0, xm12, xm13
+    punpcklwd       xm12, xm13
+    vinserti128     m12, m12, xm0, 1
+    pmaddwd         m0, m12, [r5 + 3 * mmsize]
+    paddd           m6, m0
+    pmaddwd         m0, m12, [r5 + 2 * mmsize]
+    paddd           m8, m0
+    pmaddwd         m0, m12, [r5 + 1 * mmsize]
+    paddd           m10, m0
+    pmaddwd         m12, [r5]
+    movu            xm0, [r7 + r1 * 2]              ; m0 = row 14
+    punpckhwd       xm1, xm13, xm0
+    punpcklwd       xm13, xm0
+    vinserti128     m13, m13, xm1, 1
+    pmaddwd         m1, m13, [r5 + 3 * mmsize]
+    paddd           m7, m1
+    pmaddwd         m1, m13, [r5 + 2 * mmsize]
+    paddd           m9, m1
+    pmaddwd         m1, m13, [r5 + 1 * mmsize]
+    paddd           m11, m1
+    pmaddwd         m13, [r5]
+
+%ifidn %1,sp
+    paddd           m6, m14
+    paddd           m7, m14
+    psrad           m6, 12
+    psrad           m7, 12
+%else
+    psrad           m6, 6
+    psrad           m7, 6
+%endif
+    packssdw        m6, m7
+    lea             r8, [r2 + r3 * 4]
+
+%ifidn %1,sp
+    packuswb        m4, m6
+    vpermd          m4, m5, m4
+    vextracti128    xm6, m4, 1
+    movq            [r8], xm4
+    movhps          [r8 + r3], xm4
+    movq            [r8 + r3 * 2], xm6
+    movhps          [r8 + r6], xm6
+%else
+    vpermq          m4, m4, 11011000b
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm1, m4, 1
+    vextracti128    xm7, m6, 1
+    movu            [r8], xm4
+    movu            [r8 + r3], xm1
+    movu            [r8 + r3 * 2], xm6
+    movu            [r8 + r6], xm7
+%endif
+
+    movu            xm1, [r7 + r4]                  ; m1 = row 15
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m2, m0, [r5 + 3 * mmsize]
+    paddd           m8, m2
+    pmaddwd         m2, m0, [r5 + 2 * mmsize]
+    paddd           m10, m2
+    pmaddwd         m2, m0, [r5 + 1 * mmsize]
+    paddd           m12, m2
+    pmaddwd         m0, [r5]
+    lea             r7, [r7 + r1 * 4]
+    movu            xm2, [r7]                       ; m2 = row 16
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m3, m1, [r5 + 3 * mmsize]
+    paddd           m9, m3
+    pmaddwd         m3, m1, [r5 + 2 * mmsize]
+    paddd           m11, m3
+    pmaddwd         m3, m1, [r5 + 1 * mmsize]
+    paddd           m13, m3
+    pmaddwd         m1, [r5]
+    movu            xm3, [r7 + r1]                  ; m3 = row 17
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 3 * mmsize]
+    paddd           m10, m4
+    pmaddwd         m4, m2, [r5 + 2 * mmsize]
+    paddd           m12, m4
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m0, m2
+    movu            xm4, [r7 + r1 * 2]              ; m4 = row 18
+    punpckhwd       xm2, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm2, 1
+    pmaddwd         m2, m3, [r5 + 3 * mmsize]
+    paddd           m11, m2
+    pmaddwd         m2, m3, [r5 + 2 * mmsize]
+    paddd           m13, m2
+    pmaddwd         m3, [r5 + 1 * mmsize]
+    paddd           m1, m3
+    movu            xm2, [r7 + r4]                  ; m2 = row 19
+    punpckhwd       xm6, xm4, xm2
+    punpcklwd       xm4, xm2
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 3 * mmsize]
+    paddd           m12, m6
+    pmaddwd         m4, [r5 + 2 * mmsize]
+    paddd           m0, m4
+    lea             r7, [r7 + r1 * 4]
+    movu            xm6, [r7]                       ; m6 = row 20
+    punpckhwd       xm7, xm2, xm6
+    punpcklwd       xm2, xm6
+    vinserti128     m2, m2, xm7, 1
+    pmaddwd         m7, m2, [r5 + 3 * mmsize]
+    paddd           m13, m7
+    pmaddwd         m2, [r5 + 2 * mmsize]
+    paddd           m1, m2
+    movu            xm7, [r7 + r1]                  ; m7 = row 21
+    punpckhwd       xm2, xm6, xm7
+    punpcklwd       xm6, xm7
+    vinserti128     m6, m6, xm2, 1
+    pmaddwd         m6, [r5 + 3 * mmsize]
+    paddd           m0, m6
+    movu            xm2, [r7 + r1 * 2]              ; m2 = row 22
+    punpckhwd       xm3, xm7, xm2
+    punpcklwd       xm7, xm2
+    vinserti128     m7, m7, xm3, 1
+    pmaddwd         m7, [r5 + 3 * mmsize]
+    paddd           m1, m7
+
+%ifidn %1,sp
+    paddd           m8, m14
+    paddd           m9, m14
+    paddd           m10, m14
+    paddd           m11, m14
+    paddd           m12, m14
+    paddd           m13, m14
+    paddd           m0, m14
+    paddd           m1, m14
+    psrad           m8, 12
+    psrad           m9, 12
+    psrad           m10, 12
+    psrad           m11, 12
+    psrad           m12, 12
+    psrad           m13, 12
+    psrad           m0, 12
+    psrad           m1, 12
+%else
+    psrad           m8, 6
+    psrad           m9, 6
+    psrad           m10, 6
+    psrad           m11, 6
+    psrad           m12, 6
+    psrad           m13, 6
+    psrad           m0, 6
+    psrad           m1, 6
+%endif
+    packssdw        m8, m9
+    packssdw        m10, m11
+    packssdw        m12, m13
+    packssdw        m0, m1
+    lea             r8, [r8 + r3 * 4]
+
+%ifidn %1,sp
+    packuswb        m8, m10
+    packuswb        m12, m0
+    vpermd          m8, m5, m8
+    vpermd          m12, m5, m12
+    vextracti128    xm10, m8, 1
+    vextracti128    xm0, m12, 1
+    movq            [r8], xm8
+    movhps          [r8 + r3], xm8
+    movq            [r8 + r3 * 2], xm10
+    movhps          [r8 + r6], xm10
+    lea             r8, [r8 + r3 * 4]
+    movq            [r8], xm12
+    movhps          [r8 + r3], xm12
+    movq            [r8 + r3 * 2], xm0
+    movhps          [r8 + r6], xm0
+%else
+    vpermq          m8, m8, 11011000b
+    vpermq          m10, m10, 11011000b
+    vpermq          m12, m12, 11011000b
+    vpermq          m0, m0, 11011000b
+    vextracti128    xm9, m8, 1
+    vextracti128    xm11, m10, 1
+    vextracti128    xm13, m12, 1
+    vextracti128    xm1, m0, 1
+    movu            [r8], xm8
+    movu            [r8 + r3], xm9
+    movu            [r8 + r3 * 2], xm10
+    movu            [r8 + r6], xm11
+    lea             r8, [r8 + r3 * 4]
+    movu            [r8], xm12
+    movu            [r8 + r3], xm13
+    movu            [r8 + r3 * 2], xm0
+    movu            [r8 + r6], xm1
+%endif
+%endmacro
 
 %macro FILTER_H8_W8_sse2 0
     movh        m1, [r0 + x - 3]
@@ -265,3 +714,6023 @@ cglobal interp_8tap_horiz_%3_%1x%2, 4,6,8
     IPFILTER_LUMA_sse2 64, 16, ps
     IPFILTER_LUMA_sse2 16, 64, ps
 
+%macro FILTER_H4_w2_2_sse2 0
+    pxor        m3, m3
+    movd        m0, [srcq - 1]
+    movd        m2, [srcq]
+    punpckldq   m0, m2
+    punpcklbw   m0, m3
+    movd        m1, [srcq + srcstrideq - 1]
+    movd        m2, [srcq + srcstrideq]
+    punpckldq   m1, m2
+    punpcklbw   m1, m3
+    pmaddwd     m0, m4
+    pmaddwd     m1, m4
+    packssdw    m0, m1
+    pshuflw     m1, m0, q2301
+    pshufhw     m1, m1, q2301
+    paddw       m0, m1
+    psrld       m0, 16
+    packssdw    m0, m0
+    paddw       m0, m5
+    psraw       m0, 6
+    packuswb    m0, m0
+    movd        r4d, m0
+    mov         [dstq], r4w
+    shr         r4, 16
+    mov         [dstq + dststrideq], r4w
+%endmacro
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_horiz_pp_2xN(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_H4_W2xN_sse3 1
+INIT_XMM sse3
+cglobal interp_4tap_horiz_pp_2x%1, 4, 6, 6, src, srcstride, dst, dststride
+    mov         r4d,    r4m
+    mova        m5,     [pw_32]
+
+%ifdef PIC
+    lea         r5,     [h_tabw_ChromaCoeff]
+    movddup     m4,     [r5 + r4 * 8]
+%else
+    movddup     m4,     [h_tabw_ChromaCoeff + r4 * 8]
+%endif
+
+%assign x 1
+%rep %1/2
+    FILTER_H4_w2_2_sse2
+%if x < %1/2
+    lea         srcq,   [srcq + srcstrideq * 2]
+    lea         dstq,   [dstq + dststrideq * 2]
+%endif
+%assign x x+1
+%endrep
+
+    RET
+
+%endmacro
+
+    FILTER_H4_W2xN_sse3 4
+    FILTER_H4_W2xN_sse3 8
+    FILTER_H4_W2xN_sse3 16
+
+%macro FILTER_H4_w4_2_sse2 0
+    pxor        m5, m5
+    movd        m0, [srcq - 1]
+    movd        m6, [srcq]
+    punpckldq   m0, m6
+    punpcklbw   m0, m5
+    movd        m1, [srcq + 1]
+    movd        m6, [srcq + 2]
+    punpckldq   m1, m6
+    punpcklbw   m1, m5
+    movd        m2, [srcq + srcstrideq - 1]
+    movd        m6, [srcq + srcstrideq]
+    punpckldq   m2, m6
+    punpcklbw   m2, m5
+    movd        m3, [srcq + srcstrideq + 1]
+    movd        m6, [srcq + srcstrideq + 2]
+    punpckldq   m3, m6
+    punpcklbw   m3, m5
+    pmaddwd     m0, m4
+    pmaddwd     m1, m4
+    pmaddwd     m2, m4
+    pmaddwd     m3, m4
+    packssdw    m0, m1
+    packssdw    m2, m3
+    pshuflw     m1, m0, q2301
+    pshufhw     m1, m1, q2301
+    pshuflw     m3, m2, q2301
+    pshufhw     m3, m3, q2301
+    paddw       m0, m1
+    paddw       m2, m3
+    psrld       m0, 16
+    psrld       m2, 16
+    packssdw    m0, m2
+    paddw       m0, m7
+    psraw       m0, 6
+    packuswb    m0, m2
+    movd        [dstq], m0
+    psrldq      m0, 4
+    movd        [dstq + dststrideq], m0
+%endmacro
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_horiz_pp_4x32(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_H4_W4xN_sse3 1
+INIT_XMM sse3
+cglobal interp_4tap_horiz_pp_4x%1, 4, 6, 8, src, srcstride, dst, dststride
+    mov         r4d,    r4m
+    mova        m7,     [pw_32]
+
+%ifdef PIC
+    lea         r5,     [h_tabw_ChromaCoeff]
+    movddup     m4,     [r5 + r4 * 8]
+%else
+    movddup     m4,     [h_tabw_ChromaCoeff + r4 * 8]
+%endif
+
+%assign x 1
+%rep %1/2
+    FILTER_H4_w4_2_sse2
+%if x < %1/2
+    lea         srcq,   [srcq + srcstrideq * 2]
+    lea         dstq,   [dstq + dststrideq * 2]
+%endif
+%assign x x+1
+%endrep
+
+    RET
+
+%endmacro
+
+    FILTER_H4_W4xN_sse3 2
+    FILTER_H4_W4xN_sse3 4
+    FILTER_H4_W4xN_sse3 8
+    FILTER_H4_W4xN_sse3 16
+    FILTER_H4_W4xN_sse3 32
+
+%macro FILTER_H4_w6_sse2 0
+    pxor        m4, m4
+    movh        m0, [srcq - 1]
+    movh        m5, [srcq]
+    punpckldq   m0, m5
+    movhlps     m2, m0
+    punpcklbw   m0, m4
+    punpcklbw   m2, m4
+    movd        m1, [srcq + 1]
+    movd        m5, [srcq + 2]
+    punpckldq   m1, m5
+    punpcklbw   m1, m4
+    pmaddwd     m0, m6
+    pmaddwd     m1, m6
+    pmaddwd     m2, m6
+    packssdw    m0, m1
+    packssdw    m2, m2
+    pshuflw     m1, m0, q2301
+    pshufhw     m1, m1, q2301
+    pshuflw     m3, m2, q2301
+    paddw       m0, m1
+    paddw       m2, m3
+    psrld       m0, 16
+    psrld       m2, 16
+    packssdw    m0, m2
+    paddw       m0, m7
+    psraw       m0, 6
+    packuswb    m0, m0
+    movd        [dstq], m0
+    pextrw      r4d, m0, 2
+    mov         [dstq + 4], r4w
+%endmacro
+
+%macro FILH4W8_sse2 1
+    movh        m0, [srcq - 1 + %1]
+    movh        m5, [srcq + %1]
+    punpckldq   m0, m5
+    movhlps     m2, m0
+    punpcklbw   m0, m4
+    punpcklbw   m2, m4
+    movh        m1, [srcq + 1 + %1]
+    movh        m5, [srcq + 2 + %1]
+    punpckldq   m1, m5
+    movhlps     m3, m1
+    punpcklbw   m1, m4
+    punpcklbw   m3, m4
+    pmaddwd     m0, m6
+    pmaddwd     m1, m6
+    pmaddwd     m2, m6
+    pmaddwd     m3, m6
+    packssdw    m0, m1
+    packssdw    m2, m3
+    pshuflw     m1, m0, q2301
+    pshufhw     m1, m1, q2301
+    pshuflw     m3, m2, q2301
+    pshufhw     m3, m3, q2301
+    paddw       m0, m1
+    paddw       m2, m3
+    psrld       m0, 16
+    psrld       m2, 16
+    packssdw    m0, m2
+    paddw       m0, m7
+    psraw       m0, 6
+    packuswb    m0, m0
+    movh        [dstq + %1], m0
+%endmacro
+
+%macro FILTER_H4_w8_sse2 0
+    FILH4W8_sse2 0
+%endmacro
+
+%macro FILTER_H4_w12_sse2 0
+    FILH4W8_sse2 0
+    movd        m1, [srcq - 1 + 8]
+    movd        m3, [srcq + 8]
+    punpckldq   m1, m3
+    punpcklbw   m1, m4
+    movd        m2, [srcq + 1 + 8]
+    movd        m3, [srcq + 2 + 8]
+    punpckldq   m2, m3
+    punpcklbw   m2, m4
+    pmaddwd     m1, m6
+    pmaddwd     m2, m6
+    packssdw    m1, m2
+    pshuflw     m2, m1, q2301
+    pshufhw     m2, m2, q2301
+    paddw       m1, m2
+    psrld       m1, 16
+    packssdw    m1, m1
+    paddw       m1, m7
+    psraw       m1, 6
+    packuswb    m1, m1
+    movd        [dstq + 8], m1
+%endmacro
+
+%macro FILTER_H4_w16_sse2 0
+    FILH4W8_sse2 0
+    FILH4W8_sse2 8
+%endmacro
+
+%macro FILTER_H4_w24_sse2 0
+    FILH4W8_sse2 0
+    FILH4W8_sse2 8
+    FILH4W8_sse2 16
+%endmacro
+
+%macro FILTER_H4_w32_sse2 0
+    FILH4W8_sse2 0
+    FILH4W8_sse2 8
+    FILH4W8_sse2 16
+    FILH4W8_sse2 24
+%endmacro
+
+%macro FILTER_H4_w48_sse2 0
+    FILH4W8_sse2 0
+    FILH4W8_sse2 8
+    FILH4W8_sse2 16
+    FILH4W8_sse2 24
+    FILH4W8_sse2 32
+    FILH4W8_sse2 40
+%endmacro
+
+%macro FILTER_H4_w64_sse2 0
+    FILH4W8_sse2 0
+    FILH4W8_sse2 8
+    FILH4W8_sse2 16
+    FILH4W8_sse2 24
+    FILH4W8_sse2 32
+    FILH4W8_sse2 40
+    FILH4W8_sse2 48
+    FILH4W8_sse2 56
+%endmacro
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_horiz_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro IPFILTER_CHROMA_sse3 2
+INIT_XMM sse3
+cglobal interp_4tap_horiz_pp_%1x%2, 4, 6, 8, src, srcstride, dst, dststride
+    mov         r4d,        r4m
+    mova        m7,         [pw_32]
+    pxor        m4,         m4
+
+%ifdef PIC
+    lea         r5,          [h_tabw_ChromaCoeff]
+    movddup     m6,       [r5 + r4 * 8]
+%else
+    movddup     m6,       [h_tabw_ChromaCoeff + r4 * 8]
+%endif
+
+%assign x 1
+%rep %2
+    FILTER_H4_w%1_sse2
+%if x < %2
+    add         srcq,        srcstrideq
+    add         dstq,        dststrideq
+%endif
+%assign x x+1
+%endrep
+
+    RET
+
+%endmacro
+
+    IPFILTER_CHROMA_sse3 6,   8
+    IPFILTER_CHROMA_sse3 8,   2
+    IPFILTER_CHROMA_sse3 8,   4
+    IPFILTER_CHROMA_sse3 8,   6
+    IPFILTER_CHROMA_sse3 8,   8
+    IPFILTER_CHROMA_sse3 8,  16
+    IPFILTER_CHROMA_sse3 8,  32
+    IPFILTER_CHROMA_sse3 12, 16
+
+    IPFILTER_CHROMA_sse3 6,  16
+    IPFILTER_CHROMA_sse3 8,  12
+    IPFILTER_CHROMA_sse3 8,  64
+    IPFILTER_CHROMA_sse3 12, 32
+
+    IPFILTER_CHROMA_sse3 16,  4
+    IPFILTER_CHROMA_sse3 16,  8
+    IPFILTER_CHROMA_sse3 16, 12
+    IPFILTER_CHROMA_sse3 16, 16
+    IPFILTER_CHROMA_sse3 16, 32
+    IPFILTER_CHROMA_sse3 32,  8
+    IPFILTER_CHROMA_sse3 32, 16
+    IPFILTER_CHROMA_sse3 32, 24
+    IPFILTER_CHROMA_sse3 24, 32
+    IPFILTER_CHROMA_sse3 32, 32
+
+    IPFILTER_CHROMA_sse3 16, 24
+    IPFILTER_CHROMA_sse3 16, 64
+    IPFILTER_CHROMA_sse3 32, 48
+    IPFILTER_CHROMA_sse3 24, 64
+    IPFILTER_CHROMA_sse3 32, 64
+
+    IPFILTER_CHROMA_sse3 64, 64
+    IPFILTER_CHROMA_sse3 64, 32
+    IPFILTER_CHROMA_sse3 64, 48
+    IPFILTER_CHROMA_sse3 48, 64
+    IPFILTER_CHROMA_sse3 64, 16
+
+%macro FILTER_2 2
+    movd        m3,     [srcq + %1]
+    movd        m4,     [srcq + 1 + %1]
+    punpckldq   m3,     m4
+    punpcklbw   m3,     m0
+    pmaddwd     m3,     m1
+    packssdw    m3,     m3
+    pshuflw     m4,     m3, q2301
+    paddw       m3,     m4
+    psrldq      m3,     2
+    psubw       m3,     m2
+    movd        [dstq + %2], m3
+%endmacro
+
+%macro FILTER_4 2
+    movd        m3,     [srcq + %1]
+    movd        m4,     [srcq + 1 + %1]
+    punpckldq   m3,     m4
+    punpcklbw   m3,     m0
+    pmaddwd     m3,     m1
+    movd        m4,     [srcq + 2 + %1]
+    movd        m5,     [srcq + 3 + %1]
+    punpckldq   m4,     m5
+    punpcklbw   m4,     m0
+    pmaddwd     m4,     m1
+    packssdw    m3,     m4
+    pshuflw     m4,     m3, q2301
+    pshufhw     m4,     m4, q2301
+    paddw       m3,     m4
+    psrldq      m3,     2
+    pshufd      m3,     m3,     q3120
+    psubw       m3,     m2
+    movh        [dstq + %2], m3
+%endmacro
+
+%macro FILTER_4TAP_HPS_sse3 2
+INIT_XMM sse3
+cglobal interp_4tap_horiz_ps_%1x%2, 4, 7, 6, src, srcstride, dst, dststride
+    mov         r4d,    r4m
+    add         dststrided, dststrided
+    mova        m2,     [pw_2000]
+    pxor        m0,     m0
+
+%ifdef PIC
+    lea         r6,     [h_tabw_ChromaCoeff]
+    movddup     m1,     [r6 + r4 * 8]
+%else
+    movddup     m1,     [h_tabw_ChromaCoeff + r4 * 8]
+%endif
+
+    mov        r4d,     %2
+    cmp        r5m,     byte 0
+    je         .loopH
+    sub        srcq,    srcstrideq
+    add        r4d,     3
+
+.loopH:
+%assign x -1
+%assign y 0
+%rep %1/4
+    FILTER_4 x,y
+%assign x x+4
+%assign y y+8
+%endrep
+%rep (%1 % 4)/2
+    FILTER_2 x,y
+%endrep
+    add         srcq,   srcstrideq
+    add         dstq,   dststrideq
+
+    dec         r4d
+    jnz         .loopH
+    RET
+
+%endmacro
+
+    FILTER_4TAP_HPS_sse3 2, 4
+    FILTER_4TAP_HPS_sse3 2, 8
+    FILTER_4TAP_HPS_sse3 2, 16
+    FILTER_4TAP_HPS_sse3 4, 2
+    FILTER_4TAP_HPS_sse3 4, 4
+    FILTER_4TAP_HPS_sse3 4, 8
+    FILTER_4TAP_HPS_sse3 4, 16
+    FILTER_4TAP_HPS_sse3 4, 32
+    FILTER_4TAP_HPS_sse3 6, 8
+    FILTER_4TAP_HPS_sse3 6, 16
+    FILTER_4TAP_HPS_sse3 8, 2
+    FILTER_4TAP_HPS_sse3 8, 4
+    FILTER_4TAP_HPS_sse3 8, 6
+    FILTER_4TAP_HPS_sse3 8, 8
+    FILTER_4TAP_HPS_sse3 8, 12
+    FILTER_4TAP_HPS_sse3 8, 16
+    FILTER_4TAP_HPS_sse3 8, 32
+    FILTER_4TAP_HPS_sse3 8, 64
+    FILTER_4TAP_HPS_sse3 12, 16
+    FILTER_4TAP_HPS_sse3 12, 32
+    FILTER_4TAP_HPS_sse3 16, 4
+    FILTER_4TAP_HPS_sse3 16, 8
+    FILTER_4TAP_HPS_sse3 16, 12
+    FILTER_4TAP_HPS_sse3 16, 16
+    FILTER_4TAP_HPS_sse3 16, 24
+    FILTER_4TAP_HPS_sse3 16, 32
+    FILTER_4TAP_HPS_sse3 16, 64
+    FILTER_4TAP_HPS_sse3 24, 32
+    FILTER_4TAP_HPS_sse3 24, 64
+    FILTER_4TAP_HPS_sse3 32,  8
+    FILTER_4TAP_HPS_sse3 32, 16
+    FILTER_4TAP_HPS_sse3 32, 24
+    FILTER_4TAP_HPS_sse3 32, 32
+    FILTER_4TAP_HPS_sse3 32, 48
+    FILTER_4TAP_HPS_sse3 32, 64
+    FILTER_4TAP_HPS_sse3 48, 64
+    FILTER_4TAP_HPS_sse3 64, 16
+    FILTER_4TAP_HPS_sse3 64, 32
+    FILTER_4TAP_HPS_sse3 64, 48
+    FILTER_4TAP_HPS_sse3 64, 64
+
+%macro FILTER_H4_w2_2 3
+    movh        %2, [srcq - 1]
+    pshufb      %2, %2, Tm0
+    movh        %1, [srcq + srcstrideq - 1]
+    pshufb      %1, %1, Tm0
+    punpcklqdq  %2, %1
+    pmaddubsw   %2, coef2
+    phaddw      %2, %2
+    pmulhrsw    %2, %3
+    packuswb    %2, %2
+    movd        r4d, %2
+    mov         [dstq], r4w
+    shr         r4, 16
+    mov         [dstq + dststrideq], r4w
+%endmacro
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_horiz_pp_2x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+INIT_XMM sse4
+cglobal interp_4tap_horiz_pp_2x4, 4, 6, 5, src, srcstride, dst, dststride
+%define coef2       m4
+%define Tm0         m3
+%define t2          m2
+%define t1          m1
+%define t0          m0
+
+    mov         r4d,        r4m
+
+%ifdef PIC
+    lea         r5,          [h_tab_ChromaCoeff]
+    movd        coef2,       [r5 + r4 * 4]
+%else
+    movd        coef2,       [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufd      coef2,       coef2,      0
+    mova        t2,          [pw_512]
+    mova        Tm0,         [h_tab_Tm]
+
+%rep 2
+    FILTER_H4_w2_2   t0, t1, t2
+    lea         srcq,       [srcq + srcstrideq * 2]
+    lea         dstq,       [dstq + dststrideq * 2]
+%endrep
+
+    RET
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_horiz_pp_2x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+INIT_XMM sse4
+cglobal interp_4tap_horiz_pp_2x8, 4, 6, 5, src, srcstride, dst, dststride
+%define coef2       m4
+%define Tm0         m3
+%define t2          m2
+%define t1          m1
+%define t0          m0
+
+    mov         r4d,        r4m
+
+%ifdef PIC
+    lea         r5,          [h_tab_ChromaCoeff]
+    movd        coef2,       [r5 + r4 * 4]
+%else
+    movd        coef2,       [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufd      coef2,       coef2,      0
+    mova        t2,          [pw_512]
+    mova        Tm0,         [h_tab_Tm]
+
+%rep 4
+    FILTER_H4_w2_2   t0, t1, t2
+    lea         srcq,       [srcq + srcstrideq * 2]
+    lea         dstq,       [dstq + dststrideq * 2]
+%endrep
+
+    RET
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_horiz_pp_2x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+INIT_XMM sse4
+cglobal interp_4tap_horiz_pp_2x16, 4, 6, 5, src, srcstride, dst, dststride
+%define coef2       m4
+%define Tm0         m3
+%define t2          m2
+%define t1          m1
+%define t0          m0
+
+    mov         r4d,        r4m
+
+%ifdef PIC
+    lea         r5,          [h_tab_ChromaCoeff]
+    movd        coef2,       [r5 + r4 * 4]
+%else
+    movd        coef2,       [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufd      coef2,       coef2,      0
+    mova        t2,          [pw_512]
+    mova        Tm0,         [h_tab_Tm]
+
+    mov         r5d,        16/2
+
+.loop:
+    FILTER_H4_w2_2   t0, t1, t2
+    lea         srcq,       [srcq + srcstrideq * 2]
+    lea         dstq,       [dstq + dststrideq * 2]
+    dec         r5d
+    jnz         .loop
+
+    RET
+
+%macro FILTER_H4_w4_2 3
+    movh        %2, [srcq - 1]
+    pshufb      %2, %2, Tm0
+    pmaddubsw   %2, coef2
+    movh        %1, [srcq + srcstrideq - 1]
+    pshufb      %1, %1, Tm0
+    pmaddubsw   %1, coef2
+    phaddw      %2, %1
+    pmulhrsw    %2, %3
+    packuswb    %2, %2
+    movd        [dstq], %2
+    palignr     %2, %2, 4
+    movd        [dstq + dststrideq], %2
+%endmacro
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_horiz_pp_4x2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+INIT_XMM sse4
+cglobal interp_4tap_horiz_pp_4x2, 4, 6, 5, src, srcstride, dst, dststride
+%define coef2       m4
+%define Tm0         m3
+%define t2          m2
+%define t1          m1
+%define t0          m0
+
+    mov         r4d,        r4m
+
+%ifdef PIC
+    lea         r5,          [h_tab_ChromaCoeff]
+    movd        coef2,       [r5 + r4 * 4]
+%else
+    movd        coef2,       [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufd      coef2,       coef2,      0
+    mova        t2,          [pw_512]
+    mova        Tm0,         [h_tab_Tm]
+
+    FILTER_H4_w4_2   t0, t1, t2
+
+    RET
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_horiz_pp_4x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+INIT_XMM sse4
+cglobal interp_4tap_horiz_pp_4x4, 4, 6, 5, src, srcstride, dst, dststride
+%define coef2       m4
+%define Tm0         m3
+%define t2          m2
+%define t1          m1
+%define t0          m0
+
+    mov         r4d,        r4m
+
+%ifdef PIC
+    lea         r5,          [h_tab_ChromaCoeff]
+    movd        coef2,       [r5 + r4 * 4]
+%else
+    movd        coef2,       [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufd      coef2,       coef2,      0
+    mova        t2,          [pw_512]
+    mova        Tm0,         [h_tab_Tm]
+
+%rep 2
+    FILTER_H4_w4_2   t0, t1, t2
+    lea         srcq,       [srcq + srcstrideq * 2]
+    lea         dstq,       [dstq + dststrideq * 2]
+%endrep
+
+    RET
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_horiz_pp_4x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+INIT_XMM sse4
+cglobal interp_4tap_horiz_pp_4x8, 4, 6, 5, src, srcstride, dst, dststride
+%define coef2       m4
+%define Tm0         m3
+%define t2          m2
+%define t1          m1
+%define t0          m0
+
+    mov         r4d,        r4m
+
+%ifdef PIC
+    lea         r5,          [h_tab_ChromaCoeff]
+    movd        coef2,       [r5 + r4 * 4]
+%else
+    movd        coef2,       [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufd      coef2,       coef2,      0
+    mova        t2,          [pw_512]
+    mova        Tm0,         [h_tab_Tm]
+
+%rep 4
+    FILTER_H4_w4_2   t0, t1, t2
+    lea         srcq,       [srcq + srcstrideq * 2]
+    lea         dstq,       [dstq + dststrideq * 2]
+%endrep
+
+    RET
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_horiz_pp_4x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+INIT_XMM sse4
+cglobal interp_4tap_horiz_pp_4x16, 4, 6, 5, src, srcstride, dst, dststride
+%define coef2       m4
+%define Tm0         m3
+%define t2          m2
+%define t1          m1
+%define t0          m0
+
+    mov         r4d,        r4m
+
+%ifdef PIC
+    lea         r5,          [h_tab_ChromaCoeff]
+    movd        coef2,       [r5 + r4 * 4]
+%else
+    movd        coef2,       [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufd      coef2,       coef2,      0
+    mova        t2,          [pw_512]
+    mova        Tm0,         [h_tab_Tm]
+
+%rep 8
+    FILTER_H4_w4_2   t0, t1, t2
+    lea         srcq,       [srcq + srcstrideq * 2]
+    lea         dstq,       [dstq + dststrideq * 2]
+%endrep
+
+    RET
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_horiz_pp_4x32(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+INIT_XMM sse4
+cglobal interp_4tap_horiz_pp_4x32, 4, 6, 5, src, srcstride, dst, dststride
+%define coef2       m4
+%define Tm0         m3
+%define t2          m2
+%define t1          m1
+%define t0          m0
+
+    mov         r4d,        r4m
+
+%ifdef PIC
+    lea         r5,          [h_tab_ChromaCoeff]
+    movd        coef2,       [r5 + r4 * 4]
+%else
+    movd        coef2,       [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufd      coef2,       coef2,      0
+    mova        t2,          [pw_512]
+    mova        Tm0,         [h_tab_Tm]
+
+    mov         r5d,        32/2
+
+.loop:
+    FILTER_H4_w4_2   t0, t1, t2
+    lea         srcq,       [srcq + srcstrideq * 2]
+    lea         dstq,       [dstq + dststrideq * 2]
+    dec         r5d
+    jnz         .loop
+
+    RET
+
+ALIGN 32
+const interp_4tap_8x8_horiz_shuf,   dd 0, 4, 1, 5, 2, 6, 3, 7
+
+%macro FILTER_H4_w6 3
+    movu        %1, [srcq - 1]
+    pshufb      %2, %1, Tm0
+    pmaddubsw   %2, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %2, %1
+    pmulhrsw    %2, %3
+    packuswb    %2, %2
+    movd        [dstq],      %2
+    pextrw      [dstq + 4], %2, 2
+%endmacro
+
+%macro FILTER_H4_w8 3
+    movu        %1, [srcq - 1]
+    pshufb      %2, %1, Tm0
+    pmaddubsw   %2, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %2, %1
+    pmulhrsw    %2, %3
+    packuswb    %2, %2
+    movh        [dstq],      %2
+%endmacro
+
+%macro FILTER_H4_w12 3
+    movu        %1, [srcq - 1]
+    pshufb      %2, %1, Tm0
+    pmaddubsw   %2, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %2, %1
+    pmulhrsw    %2, %3
+    movu        %1, [srcq - 1 + 8]
+    pshufb      %1, %1, Tm0
+    pmaddubsw   %1, coef2
+    phaddw      %1, %1
+    pmulhrsw    %1, %3
+    packuswb    %2, %1
+    movh        [dstq],      %2
+    pextrd      [dstq + 8], %2, 2
+%endmacro
+
+%macro FILTER_H4_w16 4
+    movu        %1, [srcq - 1]
+    pshufb      %2, %1, Tm0
+    pmaddubsw   %2, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %2, %1
+    movu        %1, [srcq - 1 + 8]
+    pshufb      %4, %1, Tm0
+    pmaddubsw   %4, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %4, %1
+    pmulhrsw    %2, %3
+    pmulhrsw    %4, %3
+    packuswb    %2, %4
+    movu        [dstq],      %2
+%endmacro
+
+%macro FILTER_H4_w24 4
+    movu        %1, [srcq - 1]
+    pshufb      %2, %1, Tm0
+    pmaddubsw   %2, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %2, %1
+    movu        %1, [srcq - 1 + 8]
+    pshufb      %4, %1, Tm0
+    pmaddubsw   %4, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %4, %1
+    pmulhrsw    %2, %3
+    pmulhrsw    %4, %3
+    packuswb    %2, %4
+    movu        [dstq],          %2
+    movu        %1, [srcq - 1 + 16]
+    pshufb      %2, %1, Tm0
+    pmaddubsw   %2, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %2, %1
+    pmulhrsw    %2, %3
+    packuswb    %2, %2
+    movh        [dstq + 16],     %2
+%endmacro
+
+%macro FILTER_H4_w32 4
+    movu        %1, [srcq - 1]
+    pshufb      %2, %1, Tm0
+    pmaddubsw   %2, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %2, %1
+    movu        %1, [srcq - 1 + 8]
+    pshufb      %4, %1, Tm0
+    pmaddubsw   %4, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %4, %1
+    pmulhrsw    %2, %3
+    pmulhrsw    %4, %3
+    packuswb    %2, %4
+    movu        [dstq],      %2
+    movu        %1, [srcq - 1 + 16]
+    pshufb      %2, %1, Tm0
+    pmaddubsw   %2, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %2, %1
+    movu        %1, [srcq - 1 + 24]
+    pshufb      %4, %1, Tm0
+    pmaddubsw   %4, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %4, %1
+    pmulhrsw    %2, %3
+    pmulhrsw    %4, %3
+    packuswb    %2, %4
+    movu        [dstq + 16],      %2
+%endmacro
+
+%macro FILTER_H4_w16o 5
+    movu        %1, [srcq + %5 - 1]
+    pshufb      %2, %1, Tm0
+    pmaddubsw   %2, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %2, %1
+    movu        %1, [srcq + %5 - 1 + 8]
+    pshufb      %4, %1, Tm0
+    pmaddubsw   %4, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %4, %1
+    pmulhrsw    %2, %3
+    pmulhrsw    %4, %3
+    packuswb    %2, %4
+    movu        [dstq + %5],      %2
+%endmacro
+
+%macro FILTER_H4_w48 4
+    FILTER_H4_w16o %1, %2, %3, %4, 0
+    FILTER_H4_w16o %1, %2, %3, %4, 16
+    FILTER_H4_w16o %1, %2, %3, %4, 32
+%endmacro
+
+%macro FILTER_H4_w64 4
+    FILTER_H4_w16o %1, %2, %3, %4, 0
+    FILTER_H4_w16o %1, %2, %3, %4, 16
+    FILTER_H4_w16o %1, %2, %3, %4, 32
+    FILTER_H4_w16o %1, %2, %3, %4, 48
+%endmacro
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_horiz_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro IPFILTER_CHROMA 2
+INIT_XMM sse4
+cglobal interp_4tap_horiz_pp_%1x%2, 4, 6, 6, src, srcstride, dst, dststride
+%define coef2       m5
+%define Tm0         m4
+%define Tm1         m3
+%define t2          m2
+%define t1          m1
+%define t0          m0
+
+    mov         r4d,        r4m
+
+%ifdef PIC
+    lea         r5,          [h_tab_ChromaCoeff]
+    movd        coef2,       [r5 + r4 * 4]
+%else
+    movd        coef2,       [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    mov           r5d,       %2
+
+    pshufd      coef2,       coef2,      0
+    mova        t2,          [pw_512]
+    mova        Tm0,         [h_tab_Tm]
+    mova        Tm1,         [h_tab_Tm + 16]
+
+.loop:
+    FILTER_H4_w%1   t0, t1, t2
+    add         srcq,        srcstrideq
+    add         dstq,        dststrideq
+
+    dec         r5d
+    jnz        .loop
+
+    RET
+%endmacro
+
+
+    IPFILTER_CHROMA 6,   8
+    IPFILTER_CHROMA 8,   2
+    IPFILTER_CHROMA 8,   4
+    IPFILTER_CHROMA 8,   6
+    IPFILTER_CHROMA 8,   8
+    IPFILTER_CHROMA 8,  16
+    IPFILTER_CHROMA 8,  32
+    IPFILTER_CHROMA 12, 16
+
+    IPFILTER_CHROMA 6,  16
+    IPFILTER_CHROMA 8,  12
+    IPFILTER_CHROMA 8,  64
+    IPFILTER_CHROMA 12, 32
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_horiz_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro IPFILTER_CHROMA_W 2
+INIT_XMM sse4
+cglobal interp_4tap_horiz_pp_%1x%2, 4, 6, 7, src, srcstride, dst, dststride
+%define coef2       m6
+%define Tm0         m5
+%define Tm1         m4
+%define t3          m3
+%define t2          m2
+%define t1          m1
+%define t0          m0
+
+    mov         r4d,         r4m
+
+%ifdef PIC
+    lea         r5,          [h_tab_ChromaCoeff]
+    movd        coef2,       [r5 + r4 * 4]
+%else
+    movd        coef2,       [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    mov         r5d,          %2
+
+    pshufd      coef2,       coef2,      0
+    mova        t2,          [pw_512]
+    mova        Tm0,         [h_tab_Tm]
+    mova        Tm1,         [h_tab_Tm + 16]
+
+.loop:
+    FILTER_H4_w%1   t0, t1, t2, t3
+    add         srcq,        srcstrideq
+    add         dstq,        dststrideq
+
+    dec         r5d
+    jnz        .loop
+
+    RET
+%endmacro
+
+    IPFILTER_CHROMA_W 16,  4
+    IPFILTER_CHROMA_W 16,  8
+    IPFILTER_CHROMA_W 16, 12
+    IPFILTER_CHROMA_W 16, 16
+    IPFILTER_CHROMA_W 16, 32
+    IPFILTER_CHROMA_W 32,  8
+    IPFILTER_CHROMA_W 32, 16
+    IPFILTER_CHROMA_W 32, 24
+    IPFILTER_CHROMA_W 24, 32
+    IPFILTER_CHROMA_W 32, 32
+
+    IPFILTER_CHROMA_W 16, 24
+    IPFILTER_CHROMA_W 16, 64
+    IPFILTER_CHROMA_W 32, 48
+    IPFILTER_CHROMA_W 24, 64
+    IPFILTER_CHROMA_W 32, 64
+
+    IPFILTER_CHROMA_W 64, 64
+    IPFILTER_CHROMA_W 64, 32
+    IPFILTER_CHROMA_W 64, 48
+    IPFILTER_CHROMA_W 48, 64
+    IPFILTER_CHROMA_W 64, 16
+
+%macro FILTER_H8_W8 7-8   ; t0, t1, t2, t3, coef, c512, src, dst
+    movu        %1, %7
+    pshufb      %2, %1, [h_tab_Lm +  0]
+    pmaddubsw   %2, %5
+    pshufb      %3, %1, [h_tab_Lm + 16]
+    pmaddubsw   %3, %5
+    phaddw      %2, %3
+    pshufb      %4, %1, [h_tab_Lm + 32]
+    pmaddubsw   %4, %5
+    pshufb      %1, %1, [h_tab_Lm + 48]
+    pmaddubsw   %1, %5
+    phaddw      %4, %1
+    phaddw      %2, %4
+  %if %0 == 8
+    pmulhrsw    %2, %6
+    packuswb    %2, %2
+    movh        %8, %2
+  %endif
+%endmacro
+
+%macro FILTER_H8_W4 2
+    movu        %1, [r0 - 3 + r5]
+    pshufb      %2, %1, [h_tab_Lm]
+    pmaddubsw   %2, m3
+    pshufb      m7, %1, [h_tab_Lm + 16]
+    pmaddubsw   m7, m3
+    phaddw      %2, m7
+    phaddw      %2, %2
+%endmacro
+
+;----------------------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;----------------------------------------------------------------------------------------------------------------------------
+%macro IPFILTER_LUMA 3
+INIT_XMM sse4
+cglobal interp_8tap_horiz_%3_%1x%2, 4,7,8
+
+    mov       r4d, r4m
+
+%ifdef PIC
+    lea       r6, [h_tab_LumaCoeff]
+    movh      m3, [r6 + r4 * 8]
+%else
+    movh      m3, [h_tab_LumaCoeff + r4 * 8]
+%endif
+    punpcklqdq  m3, m3
+
+%ifidn %3, pp
+    mova      m2, [pw_512]
+%else
+    mova      m2, [pw_2000]
+%endif
+
+    mov       r4d, %2
+%ifidn %3, ps
+    add       r3, r3
+    cmp       r5m, byte 0
+    je        .loopH
+    lea       r6, [r1 + 2 * r1]
+    sub       r0, r6
+    add       r4d, 7
+%endif
+
+.loopH:
+    xor       r5, r5
+%rep %1 / 8
+  %ifidn %3, pp
+    FILTER_H8_W8  m0, m1, m4, m5, m3, m2, [r0 - 3 + r5], [r2 + r5]
+  %else
+    FILTER_H8_W8  m0, m1, m4, m5, m3, UNUSED, [r0 - 3 + r5]
+    psubw     m1, m2
+    movu      [r2 + 2 * r5], m1
+  %endif
+    add       r5, 8
+%endrep
+
+%rep (%1 % 8) / 4
+    FILTER_H8_W4  m0, m1
+  %ifidn %3, pp
+    pmulhrsw  m1, m2
+    packuswb  m1, m1
+    movd      [r2 + r5], m1
+  %else
+    psubw     m1, m2
+    movh      [r2 + 2 * r5], m1
+  %endif
+%endrep
+
+    add       r0, r1
+    add       r2, r3
+
+    dec       r4d
+    jnz       .loopH
+    RET
+%endmacro
+
+INIT_YMM avx2
+cglobal interp_8tap_horiz_pp_4x4, 4,6,6
+    mov             r4d, r4m
+
+%ifdef PIC
+    lea             r5, [h_tab_LumaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [h_tab_LumaCoeff + r4 * 8]
+%endif
+
+    mova            m1, [h_tab_Lm]
+    vpbroadcastd    m2, [pw_1]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    sub             r0, 3
+    ; Row 0-1
+    vbroadcasti128  m3, [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb          m3, m1
+    pmaddubsw       m3, m0
+    pmaddwd         m3, m2
+    vbroadcasti128  m4, [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb          m4, m1
+    pmaddubsw       m4, m0
+    pmaddwd         m4, m2
+    phaddd          m3, m4                          ; DWORD [R1D R1C R0D R0C R1B R1A R0B R0A]
+
+    ; Row 2-3
+    lea             r0, [r0 + r1 * 2]
+    vbroadcasti128  m4, [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb          m4, m1
+    pmaddubsw       m4, m0
+    pmaddwd         m4, m2
+    vbroadcasti128  m5, [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb          m5, m1
+    pmaddubsw       m5, m0
+    pmaddwd         m5, m2
+    phaddd          m4, m5                          ; DWORD [R3D R3C R2D R2C R3B R3A R2B R2A]
+
+    packssdw        m3, m4                          ; WORD [R3D R3C R2D R2C R1D R1C R0D R0C R3B R3A R2B R2A R1B R1A R0B R0A]
+    pmulhrsw        m3, [pw_512]
+    vextracti128    xm4, m3, 1
+    packuswb        xm3, xm4                        ; BYTE [R3D R3C R2D R2C R1D R1C R0D R0C R3B R3A R2B R2A R1B R1A R0B R0A]
+    pshufb          xm3, [interp4_shuf]             ; [row3 row1 row2 row0]
+
+    lea             r0, [r3 * 3]
+    movd            [r2], xm3
+    pextrd          [r2+r3], xm3, 2
+    pextrd          [r2+r3*2], xm3, 1
+    pextrd          [r2+r0], xm3, 3
+    RET
+
+%macro FILTER_HORIZ_LUMA_AVX2_4xN 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_horiz_pp_4x%1, 4, 6, 9
+    mov             r4d, r4m
+
+%ifdef PIC
+    lea             r5, [h_tab_LumaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [h_tab_LumaCoeff + r4 * 8]
+%endif
+
+    mova            m1, [h_tab_Lm]
+    mova            m2, [pw_1]
+    mova            m7, [h_interp8_hps_shuf]
+    mova            m8, [pw_512]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+    lea             r4, [r1 * 3]
+    lea             r5, [r3 * 3]
+    sub             r0, 3
+%rep %1 / 8
+    ; Row 0-1
+    vbroadcasti128  m3, [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb          m3, m1
+    pmaddubsw       m3, m0
+    pmaddwd         m3, m2
+    vbroadcasti128  m4, [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb          m4, m1
+    pmaddubsw       m4, m0
+    pmaddwd         m4, m2
+    phaddd          m3, m4                          ; DWORD [R1D R1C R0D R0C R1B R1A R0B R0A]
+
+    ; Row 2-3
+    vbroadcasti128  m4, [r0 + r1 * 2]               ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb          m4, m1
+    pmaddubsw       m4, m0
+    pmaddwd         m4, m2
+    vbroadcasti128  m5, [r0 + r4]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb          m5, m1
+    pmaddubsw       m5, m0
+    pmaddwd         m5, m2
+    phaddd          m4, m5                          ; DWORD [R3D R3C R2D R2C R3B R3A R2B R2A]
+
+    packssdw        m3, m4                          ; WORD [R3D R3C R2D R2C R1D R1C R0D R0C R3B R3A R2B R2A R1B R1A R0B R0A]
+    lea             r0, [r0 + r1 * 4]
+    ; Row 4-5
+    vbroadcasti128  m5, [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb          m5, m1
+    pmaddubsw       m5, m0
+    pmaddwd         m5, m2
+    vbroadcasti128  m4, [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb          m4, m1
+    pmaddubsw       m4, m0
+    pmaddwd         m4, m2
+    phaddd          m5, m4                          ; DWORD [R5D R5C R4D R4C R5B R5A R4B R4A]
+
+    ; Row 6-7
+    vbroadcasti128  m4, [r0 + r1 * 2]               ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb          m4, m1
+    pmaddubsw       m4, m0
+    pmaddwd         m4, m2
+    vbroadcasti128  m6, [r0 + r4]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb          m6, m1
+    pmaddubsw       m6, m0
+    pmaddwd         m6, m2
+    phaddd          m4, m6                          ; DWORD [R7D R7C R6D R6C R7B R7A R6B R6A]
+
+    packssdw        m5, m4                          ; WORD [R7D R7C R6D R6C R5D R5C R4D R4C R7B R7A R6B R6A R5B R5A R4B R4A]
+    vpermd          m3, m7, m3
+    vpermd          m5, m7, m5
+    pmulhrsw        m3, m8
+    pmulhrsw        m5, m8
+    packuswb        m3, m5
+    vextracti128    xm5, m3, 1
+
+    movd            [r2], xm3
+    pextrd          [r2 + r3], xm3, 1
+    movd            [r2 + r3 * 2], xm5
+    pextrd          [r2 + r5], xm5, 1
+    lea             r2, [r2 + r3 * 4]
+    pextrd          [r2], xm3, 2
+    pextrd          [r2 + r3], xm3, 3
+    pextrd          [r2 + r3 * 2], xm5, 2
+    pextrd          [r2 + r5], xm5, 3
+    lea             r0, [r0 + r1 * 4]
+    lea             r2, [r2 + r3 * 4]
+%endrep
+    RET
+%endif
+%endmacro
+
+    FILTER_HORIZ_LUMA_AVX2_4xN 8
+    FILTER_HORIZ_LUMA_AVX2_4xN 16
+
+INIT_YMM avx2
+cglobal interp_8tap_horiz_pp_8x4, 4, 6, 7
+    mov             r4d, r4m
+
+%ifdef PIC
+    lea             r5, [h_tab_LumaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [h_tab_LumaCoeff + r4 * 8]
+%endif
+
+    mova            m1, [h_tab_Lm]
+    mova            m2, [h_tab_Lm + 32]
+
+    ; register map
+    ; m0     - interpolate coeff
+    ; m1, m2 - shuffle order table
+
+    sub             r0, 3
+    lea             r5, [r1 * 3]
+    lea             r4, [r3 * 3]
+
+    ; Row 0
+    vbroadcasti128  m3, [r0]                        ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
+    pshufb          m4, m3, m2
+    pshufb          m3, m1
+    pmaddubsw       m3, m0
+    pmaddubsw       m4, m0
+    phaddw          m3, m4
+    ; Row 1
+    vbroadcasti128  m4, [r0 + r1]                   ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
+    pshufb          m5, m4, m2
+    pshufb          m4, m1
+    pmaddubsw       m4, m0
+    pmaddubsw       m5, m0
+    phaddw          m4, m5
+
+    phaddw          m3, m4                          ; WORD [R1H R1G R1D R1C R0H R0G R0D R0C R1F R1E R1B R1A R0F R0E R0B R0A]
+    pmulhrsw        m3, [pw_512]
+
+    ; Row 2
+    vbroadcasti128  m4, [r0 + r1 * 2]               ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
+    pshufb          m5, m4, m2
+    pshufb          m4, m1
+    pmaddubsw       m4, m0
+    pmaddubsw       m5, m0
+    phaddw          m4, m5
+    ; Row 3
+    vbroadcasti128  m5, [r0 + r5]                   ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
+    pshufb          m6, m5, m2
+    pshufb          m5, m1
+    pmaddubsw       m5, m0
+    pmaddubsw       m6, m0
+    phaddw          m5, m6
+
+    phaddw          m4, m5                          ; WORD [R3H R3G R3D R3C R2H R2G R2D R2C R3F R3E R3B R3A R2F R2E R2B R2A]
+    pmulhrsw        m4, [pw_512]
+
+    packuswb        m3, m4
+    vextracti128    xm4, m3, 1
+    punpcklwd       xm5, xm3, xm4
+
+    movq            [r2], xm5
+    movhps          [r2 + r3], xm5
+
+    punpckhwd       xm5, xm3, xm4
+    movq            [r2 + r3 * 2], xm5
+    movhps          [r2 + r4], xm5
+    RET
+
+%macro IPFILTER_LUMA_AVX2_8xN 2
+INIT_YMM avx2
+cglobal interp_8tap_horiz_pp_%1x%2, 4, 7, 7
+    mov             r4d, r4m
+
+%ifdef PIC
+    lea             r5, [h_tab_LumaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [h_tab_LumaCoeff + r4 * 8]
+%endif
+
+    mova            m1, [h_tab_Lm]
+    mova            m2, [h_tab_Lm + 32]
+
+    ; register map
+    ; m0     - interpolate coeff
+    ; m1, m2 - shuffle order table
+
+    sub             r0, 3
+    lea             r5, [r1 * 3]
+    lea             r6, [r3 * 3]
+    mov             r4d, %2 / 4
+.loop:
+    ; Row 0
+    vbroadcasti128  m3, [r0]                        ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
+    pshufb          m4, m3, m2
+    pshufb          m3, m1
+    pmaddubsw       m3, m0
+    pmaddubsw       m4, m0
+    phaddw          m3, m4
+    ; Row 1
+    vbroadcasti128  m4, [r0 + r1]                   ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
+    pshufb          m5, m4, m2
+    pshufb          m4, m1
+    pmaddubsw       m4, m0
+    pmaddubsw       m5, m0
+    phaddw          m4, m5
+
+    phaddw          m3, m4                          ; WORD [R1H R1G R1D R1C R0H R0G R0D R0C R1F R1E R1B R1A R0F R0E R0B R0A]
+    pmulhrsw        m3, [pw_512]
+
+    ; Row 2
+    vbroadcasti128  m4, [r0 + r1 * 2]               ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
+    pshufb          m5, m4, m2
+    pshufb          m4, m1
+    pmaddubsw       m4, m0
+    pmaddubsw       m5, m0
+    phaddw          m4, m5
+    ; Row 3
+    vbroadcasti128  m5, [r0 + r5]                   ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
+    pshufb          m6, m5, m2
+    pshufb          m5, m1
+    pmaddubsw       m5, m0
+    pmaddubsw       m6, m0
+    phaddw          m5, m6
+
+    phaddw          m4, m5                          ; WORD [R3H R3G R3D R3C R2H R2G R2D R2C R3F R3E R3B R3A R2F R2E R2B R2A]
+    pmulhrsw        m4, [pw_512]
+
+    packuswb        m3, m4
+    vextracti128    xm4, m3, 1
+    punpcklwd       xm5, xm3, xm4
+
+    movq            [r2], xm5
+    movhps          [r2 + r3], xm5
+
+    punpckhwd       xm5, xm3, xm4
+    movq            [r2 + r3 * 2], xm5
+    movhps          [r2 + r6], xm5
+
+    lea             r0, [r0 + r1 * 4]
+    lea             r2, [r2 + r3 * 4]
+    dec             r4d
+    jnz             .loop
+    RET
+%endmacro
+
+    IPFILTER_LUMA_AVX2_8xN 8, 8
+    IPFILTER_LUMA_AVX2_8xN 8, 16
+    IPFILTER_LUMA_AVX2_8xN 8, 32
+
+%macro IPFILTER_LUMA_AVX2 2
+INIT_YMM avx2
+cglobal interp_8tap_horiz_pp_%1x%2, 4,6,8
+    sub               r0,        3
+    mov               r4d,       r4m
+%ifdef PIC
+    lea               r5,        [h_tab_LumaCoeff]
+    vpbroadcastd      m0,        [r5 + r4 * 8]
+    vpbroadcastd      m1,        [r5 + r4 * 8 + 4]
+%else
+    vpbroadcastd      m0,         [h_tab_LumaCoeff + r4 * 8]
+    vpbroadcastd      m1,         [h_tab_LumaCoeff + r4 * 8 + 4]
+%endif
+    movu              m3,         [h_tab_Tm + 16]
+    vpbroadcastd      m7,         [pw_1]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+    ; m2 , m2  shuffle order table
+    ; m7 - pw_1
+
+    mov               r4d,        %2/2
+.loop:
+    ; Row 0
+    vbroadcasti128    m4,         [r0]                        ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m5,         m4,     m3
+    pshufb            m4,         [h_tab_Tm]
+    pmaddubsw         m4,         m0
+    pmaddubsw         m5,         m1
+    paddw             m4,         m5
+    pmaddwd           m4,         m7
+    vbroadcasti128    m5,         [r0 + 8]                    ; second 8 elements in Row0
+    pshufb            m6,         m5,     m3
+    pshufb            m5,         [h_tab_Tm]
+    pmaddubsw         m5,         m0
+    pmaddubsw         m6,         m1
+    paddw             m5,         m6
+    pmaddwd           m5,         m7
+    packssdw          m4,         m5                          ; [17 16 15 14 07 06 05 04 13 12 11 10 03 02 01 00]
+    pmulhrsw          m4,         [pw_512]
+    vbroadcasti128    m2,         [r0 + r1]                        ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m5,         m2,     m3
+    pshufb            m2,         [h_tab_Tm]
+    pmaddubsw         m2,         m0
+    pmaddubsw         m5,         m1
+    paddw             m2,         m5
+    pmaddwd           m2,         m7
+    vbroadcasti128    m5,         [r0 + r1 + 8]                    ; second 8 elements in Row0
+    pshufb            m6,         m5,     m3
+    pshufb            m5,         [h_tab_Tm]
+    pmaddubsw         m5,         m0
+    pmaddubsw         m6,         m1
+    paddw             m5,         m6
+    pmaddwd           m5,         m7
+    packssdw          m2,         m5                          ; [17 16 15 14 07 06 05 04 13 12 11 10 03 02 01 00]
+    pmulhrsw          m2,         [pw_512]
+    packuswb          m4,         m2
+    vpermq            m4,         m4,     11011000b
+    vextracti128      xm5,        m4,     1
+    pshufd            xm4,        xm4,    11011000b
+    pshufd            xm5,        xm5,    11011000b
+    movu              [r2],       xm4
+    movu              [r2+r3],    xm5
+    lea               r0,         [r0 + r1 * 2]
+    lea               r2,         [r2 + r3 * 2]
+    dec               r4d
+    jnz              .loop
+    RET
+%endmacro
+
+%macro IPFILTER_LUMA_32x_avx2 2
+INIT_YMM avx2
+cglobal interp_8tap_horiz_pp_%1x%2, 4,6,8
+    sub               r0,         3
+    mov               r4d,        r4m
+%ifdef PIC
+    lea               r5,         [h_tab_LumaCoeff]
+    vpbroadcastd      m0,         [r5 + r4 * 8]
+    vpbroadcastd      m1,         [r5 + r4 * 8 + 4]
+%else
+    vpbroadcastd      m0,         [h_tab_LumaCoeff + r4 * 8]
+    vpbroadcastd      m1,         [h_tab_LumaCoeff + r4 * 8 + 4]
+%endif
+    movu              m3,         [h_tab_Tm + 16]
+    vpbroadcastd      m7,         [pw_1]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+    ; m2 , m2  shuffle order table
+    ; m7 - pw_1
+
+    mov               r4d,        %2
+.loop:
+    ; Row 0
+    vbroadcasti128    m4,         [r0]                        ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m5,         m4,     m3
+    pshufb            m4,         [h_tab_Tm]
+    pmaddubsw         m4,         m0
+    pmaddubsw         m5,         m1
+    paddw             m4,         m5
+    pmaddwd           m4,         m7
+    vbroadcasti128    m5,         [r0 + 8]
+    pshufb            m6,         m5,     m3
+    pshufb            m5,         [h_tab_Tm]
+    pmaddubsw         m5,         m0
+    pmaddubsw         m6,         m1
+    paddw             m5,         m6
+    pmaddwd           m5,         m7
+    packssdw          m4,         m5                          ; [17 16 15 14 07 06 05 04 13 12 11 10 03 02 01 00]
+    pmulhrsw          m4,         [pw_512]
+    vbroadcasti128    m2,         [r0 + 16]
+    pshufb            m5,         m2,     m3
+    pshufb            m2,         [h_tab_Tm]
+    pmaddubsw         m2,         m0
+    pmaddubsw         m5,         m1
+    paddw             m2,         m5
+    pmaddwd           m2,         m7
+    vbroadcasti128    m5,         [r0 + 24]
+    pshufb            m6,         m5,     m3
+    pshufb            m5,         [h_tab_Tm]
+    pmaddubsw         m5,         m0
+    pmaddubsw         m6,         m1
+    paddw             m5,         m6
+    pmaddwd           m5,         m7
+    packssdw          m2,         m5
+    pmulhrsw          m2,         [pw_512]
+    packuswb          m4,         m2
+    vpermq            m4,         m4,     11011000b
+    vextracti128      xm5,        m4,     1
+    pshufd            xm4,        xm4,    11011000b
+    pshufd            xm5,        xm5,    11011000b
+    movu              [r2],       xm4
+    movu              [r2 + 16],  xm5
+    lea               r0,         [r0 + r1]
+    lea               r2,         [r2 + r3]
+    dec               r4d
+    jnz               .loop
+    RET
+%endmacro
+%macro IPFILTER_LUMA_64x_avx2 2
+INIT_YMM avx2
+cglobal interp_8tap_horiz_pp_%1x%2, 4,6,8
+    sub               r0,    3
+    mov               r4d,   r4m
+%ifdef PIC
+    lea               r5,        [h_tab_LumaCoeff]
+    vpbroadcastd      m0,        [r5 + r4 * 8]
+    vpbroadcastd      m1,        [r5 + r4 * 8 + 4]
+%else
+    vpbroadcastd      m0,        [h_tab_LumaCoeff + r4 * 8]
+    vpbroadcastd      m1,        [h_tab_LumaCoeff + r4 * 8 + 4]
+%endif
+    movu              m3,        [h_tab_Tm + 16]
+    vpbroadcastd      m7,        [pw_1]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+    ; m2 , m2  shuffle order table
+    ; m7 - pw_1
+
+    mov               r4d,   %2
+.loop:
+    ; Row 0
+    vbroadcasti128    m4,        [r0]                        ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m5,        m4,    m3
+    pshufb            m4,        [h_tab_Tm]
+    pmaddubsw         m4,        m0
+    pmaddubsw         m5,        m1
+    paddw             m4,        m5
+    pmaddwd           m4,        m7
+    vbroadcasti128    m5,        [r0 + 8]
+    pshufb            m6,        m5,    m3
+    pshufb            m5,        [h_tab_Tm]
+    pmaddubsw         m5,        m0
+    pmaddubsw         m6,        m1
+    paddw             m5,        m6
+    pmaddwd           m5,        m7
+    packssdw          m4,        m5                          ; [17 16 15 14 07 06 05 04 13 12 11 10 03 02 01 00]
+    pmulhrsw          m4,        [pw_512]
+    vbroadcasti128    m2,        [r0 + 16]
+    pshufb            m5,        m2,    m3
+    pshufb            m2,        [h_tab_Tm]
+    pmaddubsw         m2,        m0
+    pmaddubsw         m5,        m1
+    paddw             m2,        m5
+    pmaddwd           m2,        m7
+    vbroadcasti128    m5,        [r0 + 24]
+    pshufb            m6,        m5,    m3
+    pshufb            m5,        [h_tab_Tm]
+    pmaddubsw         m5,        m0
+    pmaddubsw         m6,        m1
+    paddw             m5,        m6
+    pmaddwd           m5,        m7
+    packssdw          m2,        m5
+    pmulhrsw          m2,        [pw_512]
+    packuswb          m4,        m2
+    vpermq            m4,        m4,    11011000b
+    vextracti128      xm5,       m4,    1
+    pshufd            xm4,       xm4,   11011000b
+    pshufd            xm5,       xm5,   11011000b
+    movu              [r2],      xm4
+    movu              [r2 + 16], xm5
+
+    vbroadcasti128    m4,        [r0 + 32]
+    pshufb            m5,        m4,    m3
+    pshufb            m4,        [h_tab_Tm]
+    pmaddubsw         m4,        m0
+    pmaddubsw         m5,        m1
+    paddw             m4,        m5
+    pmaddwd           m4,        m7
+    vbroadcasti128    m5,        [r0 + 40]
+    pshufb            m6,        m5,    m3
+    pshufb            m5,        [h_tab_Tm]
+    pmaddubsw         m5,        m0
+    pmaddubsw         m6,        m1
+    paddw             m5,        m6
+    pmaddwd           m5,        m7
+    packssdw          m4,        m5
+    pmulhrsw          m4,        [pw_512]
+    vbroadcasti128    m2,        [r0 + 48]
+    pshufb            m5,        m2,    m3
+    pshufb            m2,        [h_tab_Tm]
+    pmaddubsw         m2,        m0
+    pmaddubsw         m5,        m1
+    paddw             m2,        m5
+    pmaddwd           m2,        m7
+    vbroadcasti128    m5,        [r0 + 56]
+    pshufb            m6,        m5,    m3
+    pshufb            m5,        [h_tab_Tm]
+    pmaddubsw         m5,        m0
+    pmaddubsw         m6,        m1
+    paddw             m5,        m6
+    pmaddwd           m5,        m7
+    packssdw          m2,        m5
+    pmulhrsw          m2,        [pw_512]
+    packuswb          m4,        m2
+    vpermq            m4,        m4,    11011000b
+    vextracti128      xm5,       m4,    1
+    pshufd            xm4,       xm4,   11011000b
+    pshufd            xm5,       xm5,   11011000b
+    movu              [r2 +32],  xm4
+    movu              [r2 + 48], xm5
+
+    lea               r0,        [r0 + r1]
+    lea               r2,        [r2 + r3]
+    dec               r4d
+    jnz               .loop
+    RET
+%endmacro
+
+INIT_YMM avx2
+cglobal interp_8tap_horiz_pp_48x64, 4,6,8
+    sub               r0,         3
+    mov               r4d,        r4m
+%ifdef PIC
+    lea               r5,         [h_tab_LumaCoeff]
+    vpbroadcastd      m0,         [r5 + r4 * 8]
+    vpbroadcastd      m1,         [r5 + r4 * 8 + 4]
+%else
+    vpbroadcastd      m0,         [h_tab_LumaCoeff + r4 * 8]
+    vpbroadcastd      m1,         [h_tab_LumaCoeff + r4 * 8 + 4]
+%endif
+    movu              m3,         [h_tab_Tm + 16]
+    vpbroadcastd      m7,         [pw_1]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+    ; m2 , m2  shuffle order table
+    ; m7 - pw_1
+
+    mov               r4d,        64
+.loop:
+    ; Row 0
+    vbroadcasti128    m4,         [r0]                        ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m5,         m4,     m3
+    pshufb            m4,         [h_tab_Tm]
+    pmaddubsw         m4,         m0
+    pmaddubsw         m5,         m1
+    paddw             m4,         m5
+    pmaddwd           m4,         m7
+    vbroadcasti128    m5,         [r0 + 8]
+    pshufb            m6,         m5,     m3
+    pshufb            m5,         [h_tab_Tm]
+    pmaddubsw         m5,         m0
+    pmaddubsw         m6,         m1
+    paddw             m5,         m6
+    pmaddwd           m5,         m7
+    packssdw          m4,         m5                          ; [17 16 15 14 07 06 05 04 13 12 11 10 03 02 01 00]
+    pmulhrsw          m4,         [pw_512]
+
+    vbroadcasti128    m2,         [r0 + 16]
+    pshufb            m5,         m2,     m3
+    pshufb            m2,         [h_tab_Tm]
+    pmaddubsw         m2,         m0
+    pmaddubsw         m5,         m1
+    paddw             m2,         m5
+    pmaddwd           m2,         m7
+    vbroadcasti128    m5,         [r0 + 24]
+    pshufb            m6,         m5,     m3
+    pshufb            m5,         [h_tab_Tm]
+    pmaddubsw         m5,         m0
+    pmaddubsw         m6,         m1
+    paddw             m5,         m6
+    pmaddwd           m5,         m7
+    packssdw          m2,         m5
+    pmulhrsw          m2,         [pw_512]
+    packuswb          m4,         m2
+    vpermq            m4,         m4,     11011000b
+    vextracti128      xm5,        m4,     1
+    pshufd            xm4,        xm4,    11011000b
+    pshufd            xm5,        xm5,    11011000b
+    movu              [r2],       xm4
+    movu              [r2 + 16],  xm5
+
+    vbroadcasti128    m4,         [r0 + 32]
+    pshufb            m5,         m4,     m3
+    pshufb            m4,         [h_tab_Tm]
+    pmaddubsw         m4,         m0
+    pmaddubsw         m5,         m1
+    paddw             m4,         m5
+    pmaddwd           m4,         m7
+    vbroadcasti128    m5,         [r0 + 40]
+    pshufb            m6,         m5,     m3
+    pshufb            m5,         [h_tab_Tm]
+    pmaddubsw         m5,         m0
+    pmaddubsw         m6,         m1
+    paddw             m5,         m6
+    pmaddwd           m5,         m7
+    packssdw          m4,         m5
+    pmulhrsw          m4,         [pw_512]
+    packuswb          m4,         m4
+    vpermq            m4,         m4,     11011000b
+    pshufd            xm4,        xm4,    11011000b
+    movu              [r2 + 32],  xm4
+
+    lea               r0,         [r0 + r1]
+    lea               r2,         [r2 + r3]
+    dec               r4d
+    jnz               .loop
+    RET
+
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_4x4, 4,6,6
+    mov             r4d, r4m
+
+%ifdef PIC
+    lea               r5,           [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,           [r5 + r4 * 4]
+%else
+    vpbroadcastd      m0,           [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    vpbroadcastd      m2,           [pw_1]
+    vbroadcasti128    m1,           [h_tab_Tm]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    dec                r0
+
+    ; Row 0-1
+    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    vinserti128       m3,           m3,      [r0 + r1],     1
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+
+    ; Row 2-3
+    lea               r0,           [r0 + r1 * 2]
+    vbroadcasti128    m4,           [r0]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    vinserti128       m4,           m4,      [r0 + r1],     1
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+
+    packssdw          m3,           m4
+    pmulhrsw          m3,           [pw_512]
+    vextracti128      xm4,          m3,     1
+    packuswb          xm3,          xm4
+
+    lea               r0,           [r3 * 3]
+    movd              [r2],         xm3
+    pextrd            [r2+r3],      xm3,     2
+    pextrd            [r2+r3*2],    xm3,     1
+    pextrd            [r2+r0],      xm3,     3
+    RET
+
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_2x4, 4, 6, 3
+    mov               r4d,           r4m
+
+%ifdef PIC
+    lea               r5,            [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,            [r5 + r4 * 4]
+%else
+    vpbroadcastd      m0,            [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    dec               r0
+    lea               r4,            [r1 * 3]
+    movq              xm1,           [r0]
+    movhps            xm1,           [r0 + r1]
+    movq              xm2,           [r0 + r1 * 2]
+    movhps            xm2,           [r0 + r4]
+    vinserti128       m1,            m1,          xm2,          1
+    pshufb            m1,            [interp4_hpp_shuf]
+    pmaddubsw         m1,            m0
+    pmaddwd           m1,            [pw_1]
+    vextracti128      xm2,           m1,          1
+    packssdw          xm1,           xm2
+    pmulhrsw          xm1,           [pw_512]
+    packuswb          xm1,           xm1
+
+    lea               r4,            [r3 * 3]
+    pextrw            [r2],          xm1,         0
+    pextrw            [r2 + r3],     xm1,         1
+    pextrw            [r2 + r3 * 2], xm1,         2
+    pextrw            [r2 + r4],     xm1,         3
+    RET
+
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_2x8, 4, 6, 6
+    mov               r4d,           r4m
+
+%ifdef PIC
+    lea               r5,            [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,            [r5 + r4 * 4]
+%else
+    vpbroadcastd      m0,            [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    mova              m4,            [interp4_hpp_shuf]
+    mova              m5,            [pw_1]
+    dec               r0
+    lea               r4,            [r1 * 3]
+    movq              xm1,           [r0]
+    movhps            xm1,           [r0 + r1]
+    movq              xm2,           [r0 + r1 * 2]
+    movhps            xm2,           [r0 + r4]
+    vinserti128       m1,            m1,          xm2,          1
+    lea               r0,            [r0 + r1 * 4]
+    movq              xm3,           [r0]
+    movhps            xm3,           [r0 + r1]
+    movq              xm2,           [r0 + r1 * 2]
+    movhps            xm2,           [r0 + r4]
+    vinserti128       m3,            m3,          xm2,          1
+
+    pshufb            m1,            m4
+    pshufb            m3,            m4
+    pmaddubsw         m1,            m0
+    pmaddubsw         m3,            m0
+    pmaddwd           m1,            m5
+    pmaddwd           m3,            m5
+    packssdw          m1,            m3
+    pmulhrsw          m1,            [pw_512]
+    vextracti128      xm2,           m1,          1
+    packuswb          xm1,           xm2
+
+    lea               r4,            [r3 * 3]
+    pextrw            [r2],          xm1,         0
+    pextrw            [r2 + r3],     xm1,         1
+    pextrw            [r2 + r3 * 2], xm1,         4
+    pextrw            [r2 + r4],     xm1,         5
+    lea               r2,            [r2 + r3 * 4]
+    pextrw            [r2],          xm1,         2
+    pextrw            [r2 + r3],     xm1,         3
+    pextrw            [r2 + r3 * 2], xm1,         6
+    pextrw            [r2 + r4],     xm1,         7
+    RET
+
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_32x32, 4,6,7
+    mov             r4d, r4m
+
+%ifdef PIC
+    lea               r5,           [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,           [r5 + r4 * 4]
+%else
+    vpbroadcastd      m0,           [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    mova              m1,           [interp4_horiz_shuf1]
+    vpbroadcastd      m2,           [pw_1]
+    mova              m6,           [pw_512]
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    dec               r0
+    mov               r4d,          32
+
+.loop:
+    ; Row 0
+    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+    vbroadcasti128    m4,           [r0 + 4]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+    packssdw          m3,           m4
+    pmulhrsw          m3,           m6
+
+    vbroadcasti128    m4,           [r0 + 16]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+    vbroadcasti128    m5,           [r0 + 20]
+    pshufb            m5,           m1
+    pmaddubsw         m5,           m0
+    pmaddwd           m5,           m2
+    packssdw          m4,           m5
+    pmulhrsw          m4,           m6
+
+    packuswb          m3,           m4
+    vpermq            m3,           m3,      11011000b
+
+    movu              [r2],         m3
+    lea               r2,           [r2 + r3]
+    lea               r0,           [r0 + r1]
+    dec               r4d
+    jnz               .loop
+    RET
+
+
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_16x16, 4, 6, 7
+    mov               r4d,          r4m
+
+%ifdef PIC
+    lea               r5,           [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,           [r5 + r4 * 4]
+%else
+    vpbroadcastd      m0,           [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    mova              m6,           [pw_512]
+    mova              m1,           [interp4_horiz_shuf1]
+    vpbroadcastd      m2,           [pw_1]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    dec               r0
+    mov               r4d,          8
+
+.loop:
+    ; Row 0
+    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+    vbroadcasti128    m4,           [r0 + 4]                    ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+    packssdw          m3,           m4
+    pmulhrsw          m3,           m6
+
+    ; Row 1
+    vbroadcasti128    m4,           [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+    vbroadcasti128    m5,           [r0 + r1 + 4]               ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m5,           m1
+    pmaddubsw         m5,           m0
+    pmaddwd           m5,           m2
+    packssdw          m4,           m5
+    pmulhrsw          m4,           m6
+
+    packuswb          m3,           m4
+    vpermq            m3,           m3,      11011000b
+
+    vextracti128      xm4,          m3,       1
+    movu              [r2],         xm3
+    movu              [r2 + r3],    xm4
+    lea               r2,           [r2 + r3 * 2]
+    lea               r0,           [r0 + r1 * 2]
+    dec               r4d
+    jnz               .loop
+    RET
+
+;--------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+    IPFILTER_LUMA 4, 4, pp
+    IPFILTER_LUMA 4, 8, pp
+    IPFILTER_LUMA 12, 16, pp
+    IPFILTER_LUMA 4, 16, pp
+
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_8x8, 4,6,6
+    mov               r4d,    r4m
+
+%ifdef PIC
+    lea               r5,           [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,           [r5 + r4 * 4]
+%else
+    vpbroadcastd      m0,           [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    movu              m1,           [h_tab_Tm]
+    vpbroadcastd      m2,           [pw_1]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    sub               r0,           1
+    mov               r4d,          2
+
+.loop:
+    ; Row 0
+    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+
+    ; Row 1
+    vbroadcasti128    m4,           [r0 + r1]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+    packssdw          m3,           m4
+    pmulhrsw          m3,           [pw_512]
+    lea               r0,           [r0 + r1 * 2]
+
+    ; Row 2
+    vbroadcasti128    m4,           [r0 ]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+
+    ; Row 3
+    vbroadcasti128    m5,           [r0 + r1]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m5,           m1
+    pmaddubsw         m5,           m0
+    pmaddwd           m5,           m2
+    packssdw          m4,           m5
+    pmulhrsw          m4,           [pw_512]
+
+    packuswb          m3,           m4
+    mova              m5,           [interp_4tap_8x8_horiz_shuf]
+    vpermd            m3,           m5,     m3
+    vextracti128      xm4,          m3,     1
+    movq              [r2],         xm3
+    movhps            [r2 + r3],    xm3
+    lea               r2,           [r2 + r3 * 2]
+    movq              [r2],         xm4
+    movhps            [r2 + r3],    xm4
+    lea               r2,           [r2 + r3 * 2]
+    lea               r0,           [r0 + r1*2]
+    dec               r4d
+    jnz               .loop
+    RET
+
+    IPFILTER_LUMA_AVX2 16, 4
+    IPFILTER_LUMA_AVX2 16, 8
+    IPFILTER_LUMA_AVX2 16, 12
+    IPFILTER_LUMA_AVX2 16, 16
+    IPFILTER_LUMA_AVX2 16, 32
+    IPFILTER_LUMA_AVX2 16, 64
+
+    IPFILTER_LUMA_32x_avx2 32 , 8
+    IPFILTER_LUMA_32x_avx2 32 , 16
+    IPFILTER_LUMA_32x_avx2 32 , 24
+    IPFILTER_LUMA_32x_avx2 32 , 32
+    IPFILTER_LUMA_32x_avx2 32 , 64
+
+    IPFILTER_LUMA_64x_avx2 64 , 64
+    IPFILTER_LUMA_64x_avx2 64 , 48
+    IPFILTER_LUMA_64x_avx2 64 , 32
+    IPFILTER_LUMA_64x_avx2 64 , 16
+
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_8x2, 4, 6, 5
+    mov               r4d,          r4m
+
+%ifdef PIC
+    lea               r5,           [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,           [r5 + r4 * 4]
+%else
+    vpbroadcastd      m0,           [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    mova              m1,           [h_tab_Tm]
+    mova              m2,           [pw_1]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    dec               r0
+    ; Row 0
+    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+
+    ; Row 1
+    vbroadcasti128    m4,           [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+    packssdw          m3,           m4
+    pmulhrsw          m3,           [pw_512]
+    vextracti128      xm4,          m3,          1
+    packuswb          xm3,          xm4
+    pshufd            xm3,          xm3,         11011000b
+    movq              [r2],         xm3
+    movhps            [r2 + r3],    xm3
+    RET
+
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_8x6, 4, 6, 7
+    mov               r4d,           r4m
+
+%ifdef PIC
+    lea               r5,            [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,            [r5 + r4 * 4]
+%else
+    vpbroadcastd      m0,            [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    mova              m1,            [h_tab_Tm]
+    mova              m2,            [pw_1]
+    mova              m6,            [pw_512]
+    lea               r4,            [r1 * 3]
+    lea               r5,            [r3 * 3]
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    dec               r0
+    ; Row 0
+    vbroadcasti128    m3,            [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,            m1
+    pmaddubsw         m3,            m0
+    pmaddwd           m3,            m2
+
+    ; Row 1
+    vbroadcasti128    m4,            [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,            m1
+    pmaddubsw         m4,            m0
+    pmaddwd           m4,            m2
+    packssdw          m3,            m4
+    pmulhrsw          m3,            m6
+
+    ; Row 2
+    vbroadcasti128    m4,            [r0 + r1 * 2]               ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,            m1
+    pmaddubsw         m4,            m0
+    pmaddwd           m4,            m2
+
+    ; Row 3
+    vbroadcasti128    m5,            [r0 + r4]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m5,            m1
+    pmaddubsw         m5,            m0
+    pmaddwd           m5,            m2
+    packssdw          m4,            m5
+    pmulhrsw          m4,            m6
+
+    packuswb          m3,            m4
+    mova              m5,            [h_interp8_hps_shuf]
+    vpermd            m3,            m5,          m3
+    vextracti128      xm4,           m3,          1
+    movq              [r2],          xm3
+    movhps            [r2 + r3],     xm3
+    movq              [r2 + r3 * 2], xm4
+    movhps            [r2 + r5],     xm4
+    lea               r2,            [r2 + r3 * 4]
+    lea               r0,            [r0 + r1 * 4]
+    ; Row 4
+    vbroadcasti128    m3,            [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,            m1
+    pmaddubsw         m3,            m0
+    pmaddwd           m3,            m2
+
+    ; Row 5
+    vbroadcasti128    m4,            [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,            m1
+    pmaddubsw         m4,            m0
+    pmaddwd           m4,            m2
+    packssdw          m3,            m4
+    pmulhrsw          m3,            m6
+    vextracti128      xm4,           m3,          1
+    packuswb          xm3,           xm4
+    pshufd            xm3,           xm3,         11011000b
+    movq              [r2],          xm3
+    movhps            [r2 + r3],     xm3
+    RET
+
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_6x8, 4, 6, 7
+    mov               r4d,               r4m
+
+%ifdef PIC
+    lea               r5,                [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,                [r5 + r4 * 4]
+%else
+    vpbroadcastd      m0,                [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    mova              m1,                [h_tab_Tm]
+    mova              m2,                [pw_1]
+    mova              m6,                [pw_512]
+    lea               r4,                [r1 * 3]
+    lea               r5,                [r3 * 3]
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    dec               r0
+%rep 2
+    ; Row 0
+    vbroadcasti128    m3,                [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,                m1
+    pmaddubsw         m3,                m0
+    pmaddwd           m3,                m2
+
+    ; Row 1
+    vbroadcasti128    m4,                [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,                m1
+    pmaddubsw         m4,                m0
+    pmaddwd           m4,                m2
+    packssdw          m3,                m4
+    pmulhrsw          m3,                m6
+
+    ; Row 2
+    vbroadcasti128    m4,                [r0 + r1 * 2]               ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,                m1
+    pmaddubsw         m4,                m0
+    pmaddwd           m4,                m2
+
+    ; Row 3
+    vbroadcasti128    m5,                [r0 + r4]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m5,                m1
+    pmaddubsw         m5,                m0
+    pmaddwd           m5,                m2
+    packssdw          m4,                m5
+    pmulhrsw          m4,                m6
+
+    packuswb          m3,                m4
+    vextracti128      xm4,               m3,          1
+    movd              [r2],              xm3
+    pextrw            [r2 + 4],          xm4,         0
+    pextrd            [r2 + r3],         xm3,         1
+    pextrw            [r2 + r3 + 4],     xm4,         2
+    pextrd            [r2 + r3 * 2],     xm3,         2
+    pextrw            [r2 + r3 * 2 + 4], xm4,         4
+    pextrd            [r2 + r5],         xm3,         3
+    pextrw            [r2 + r5 + 4],     xm4,         6
+    lea               r2,                [r2 + r3 * 4]
+    lea               r0,                [r0 + r1 * 4]
+%endrep
+    RET
+
+;-----------------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_ps_64xN(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;-----------------------------------------------------------------------------------------------------------------------------;
+%macro IPFILTER_CHROMA_HPS_64xN 1
+INIT_YMM avx2
+cglobal interp_4tap_horiz_ps_64x%1, 4,7,6
+    mov             r4d, r4m
+    mov             r5d, r5m
+    add             r3d, r3d
+
+%ifdef PIC
+    lea               r6,           [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,           [r6 + r4 * 4]
+%else
+    vpbroadcastd      m0,           [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    vbroadcasti128     m2,           [pw_1]
+    vbroadcasti128     m5,           [pw_2000]
+    mova               m1,           [h_tab_Tm]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+    mov                r6d,         %1
+    dec                r0
+    test                r5d,      r5d
+    je                 .loop
+    sub                r0 ,         r1
+    add                r6d ,        3
+
+.loop:
+    ; Row 0
+    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+    vbroadcasti128    m4,           [r0 + 8]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+
+    packssdw          m3,           m4
+    psubw             m3,           m5
+    vpermq            m3,           m3,          11011000b
+    movu              [r2],         m3
+
+    vbroadcasti128    m3,           [r0 + 16]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+    vbroadcasti128    m4,           [r0 + 24]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+
+    packssdw          m3,           m4
+    psubw             m3,           m5
+    vpermq            m3,           m3,          11011000b
+    movu              [r2 + 32],    m3
+
+    vbroadcasti128    m3,           [r0 + 32]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+    vbroadcasti128    m4,           [r0 + 40]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+
+    packssdw          m3,           m4
+    psubw             m3,           m5
+    vpermq            m3,           m3,          11011000b
+    movu              [r2 + 64],    m3
+
+    vbroadcasti128    m3,           [r0 + 48]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+    vbroadcasti128    m4,           [r0 + 56]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+
+    packssdw          m3,           m4
+    psubw             m3,           m5
+    vpermq            m3,           m3,          11011000b
+    movu              [r2 + 96],    m3
+
+    add                r2,           r3
+    add                r0,           r1
+    dec                r6d
+    jnz                .loop
+    RET
+%endmacro
+
+   IPFILTER_CHROMA_HPS_64xN 64
+   IPFILTER_CHROMA_HPS_64xN 32
+   IPFILTER_CHROMA_HPS_64xN 48
+   IPFILTER_CHROMA_HPS_64xN 16
+
+;-----------------------------------------------------------------------------------------------------------------------------
+;void interp_horiz_ps_c(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;-----------------------------------------------------------------------------------------------------------------------------
+
+%macro IPFILTER_LUMA_PS_4xN_AVX2 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_horiz_ps_4x%1, 6,7,6
+    mov                         r5d,               r5m
+    mov                         r4d,               r4m
+%ifdef PIC
+    lea                         r6,                [h_tab_LumaCoeff]
+    vpbroadcastq                m0,                [r6 + r4 * 8]
+%else
+    vpbroadcastq                m0,                [h_tab_LumaCoeff + r4 * 8]
+%endif
+    mova                        m1,                [h_tab_Lm]
+    add                         r3d,               r3d
+    vbroadcasti128              m2,                [pw_2000]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - pw_2000
+
+    sub                         r0,                3
+    test                        r5d,               r5d
+    mov                         r5d,               %1                           ; loop count variable - height
+    jz                         .preloop
+    lea                         r6,                [r1 * 3]                     ; r8 = (N / 2 - 1) * srcStride
+    sub                         r0,                r6                           ; r0(src) - 3 * srcStride
+    add                         r5d,               7                            ; need extra 7 rows, just set a specially flag here, blkheight += N - 1  (7 - 3 = 4 ; since the last three rows not in loop)
+
+.preloop:
+    lea                         r6,                [r3 * 3]
+.loop:
+    ; Row 0-1
+    vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb                      m3,                m1                           ; shuffled based on the col order tab_Lm
+    pmaddubsw                   m3,                m0
+    vbroadcasti128              m4,                [r0 + r1]                    ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb                      m4,                m1
+    pmaddubsw                   m4,                m0
+    phaddw                      m3,                m4                           ; DWORD [R1D R1C R0D R0C R1B R1A R0B R0A]
+
+    ; Row 2-3
+    lea                         r0,                [r0 + r1 * 2]                ;3rd row(i.e 2nd row)
+    vbroadcasti128              m4,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb                      m4,                m1
+    pmaddubsw                   m4,                m0
+    vbroadcasti128              m5,                [r0 + r1]                    ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb                      m5,                m1
+    pmaddubsw                   m5,                m0
+    phaddw                      m4,                m5                           ; DWORD [R3D R3C R2D R2C R3B R3A R2B R2A]
+    phaddw                      m3,                m4                           ; all rows and col completed.
+
+    mova                        m5,                [h_interp8_hps_shuf]
+    vpermd                      m3,                m5,               m3
+    psubw                       m3,                m2
+
+    vextracti128                xm4,               m3,               1
+    movq                        [r2],              xm3                          ;row 0
+    movhps                      [r2 + r3],         xm3                          ;row 1
+    movq                        [r2 + r3 * 2],     xm4                          ;row 2
+    movhps                      [r2 + r6],         xm4                          ;row 3
+
+    lea                         r0,                [r0 + r1 * 2]                ; first loop src ->5th row(i.e 4)
+    lea                         r2,                [r2 + r3 * 4]                ; first loop dst ->5th row(i.e 4)
+    sub                         r5d,               4
+    jz                         .end
+    cmp                         r5d,               4
+    jge                        .loop
+
+    ; Row 8-9
+    vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb                      m3,                m1
+    pmaddubsw                   m3,                m0
+    vbroadcasti128              m4,                [r0 + r1]                    ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb                      m4,                m1
+    pmaddubsw                   m4,                m0
+    phaddw                      m3,                m4                           ; DWORD [R1D R1C R0D R0C R1B R1A R0B R0A]
+
+    ; Row 10
+    vbroadcasti128              m4,                [r0 + r1 * 2]                ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb                      m4,                m1
+    pmaddubsw                   m4,                m0
+    phaddw                      m4,                m4                           ; DWORD [R3D R3C R2D R2C R3B R3A R2B R2A]
+    phaddw                      m3,                m4
+
+    vpermd                      m3,                m5,            m3            ; m5 don't broken in above
+    psubw                       m3,                m2
+
+    vextracti128                xm4,               m3,            1
+    movq                        [r2],              xm3
+    movhps                      [r2 + r3],         xm3
+    movq                        [r2 + r3 * 2],     xm4
+.end:
+    RET
+%endif
+%endmacro
+
+    IPFILTER_LUMA_PS_4xN_AVX2 4
+    IPFILTER_LUMA_PS_4xN_AVX2 8
+    IPFILTER_LUMA_PS_4xN_AVX2 16
+
+%macro IPFILTER_LUMA_PS_8xN_AVX2 1
+; TODO: verify and enable on X86 mode
+%if ARCH_X86_64 == 1
+; void filter_hps(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+INIT_YMM avx2
+cglobal interp_8tap_horiz_ps_8x%1, 4,7,6
+    mov                         r5d,        r5m
+    mov                         r4d,        r4m
+    shl                         r4d,        7
+%ifdef PIC
+    lea                         r6,         [pb_LumaCoeffVer]
+    add                         r6,         r4
+%else
+    lea                         r6,         [pb_LumaCoeffVer + r4]
+%endif
+    add                         r3d,        r3d
+    vpbroadcastd                m0,         [pw_2000]
+    sub                         r0,         3
+    lea                         r4,         [pb_8tap_hps_0]
+    vbroadcasti128              m5,         [r4 + 0 * mmsize]
+
+    ; check row count extend for interpolateHV
+    test                        r5d,        r5d;
+    mov                         r5d,        %1
+    jz                         .enter_loop
+    lea                         r4,         [r1 * 3]                        ; r8 = (N / 2 - 1) * srcStride
+    sub                         r0,         r4                              ; r0(src)-r8
+    add                         r5d,        8-1-2                           ; blkheight += N - 1  (7 - 3 = 4 ; since the last three rows not in loop)
+
+.enter_loop:
+    lea                         r4,         [pb_8tap_hps_0]
+
+    ; ***** register map *****
+    ; m0 - pw_2000
+    ; r4 - base pointer of shuffle order table
+    ; r5 - count of loop
+    ; r6 - point to LumaCoeff
+.loop:
+
+    ; Row 0-1
+    movu                        xm1,        [r0]
+    movu                        xm2,        [r0 + r1]
+    vinserti128                 m1,         m1,         xm2, 1
+    pshufb                      m2,         m1,         m5                  ; [0 1 1 2 2 3 3 4 ...]
+    pshufb                      m3,         m1,         [r4 + 1 * mmsize]   ; [2 3 3 4 4 5 5 6 ...]
+    pshufb                      m4,         m1,         [r4 + 2 * mmsize]   ; [4 5 5 6 6 7 7 8 ...]
+    pshufb                      m1,         m1,         [r4 + 3 * mmsize]   ; [6 7 7 8 8 9 9 A ...]
+    pmaddubsw                   m2,         [r6 + 0 * mmsize]
+    pmaddubsw                   m3,         [r6 + 1 * mmsize]
+    pmaddubsw                   m4,         [r6 + 2 * mmsize]
+    pmaddubsw                   m1,         [r6 + 3 * mmsize]
+    paddw                       m2,         m3
+    paddw                       m1,         m4
+    paddw                       m1,         m2
+    psubw                       m1,         m0
+
+    vextracti128                xm2,        m1,         1
+    movu                        [r2],       xm1                             ; row 0
+    movu                        [r2 + r3],  xm2                             ; row 1
+
+    lea                         r0,         [r0 + r1 * 2]                   ; first loop src ->5th row(i.e 4)
+    lea                         r2,         [r2 + r3 * 2]                   ; first loop dst ->5th row(i.e 4)
+    sub                         r5d,        2
+    jg                         .loop
+    jz                         .end
+
+    ; last row
+    movu                        xm1,        [r0]
+    pshufb                      xm2,        xm1,         xm5                ; [0 1 1 2 2 3 3 4 ...]
+    pshufb                      xm3,        xm1,         [r4 + 1 * mmsize]  ; [2 3 3 4 4 5 5 6 ...]
+    pshufb                      xm4,        xm1,         [r4 + 2 * mmsize]  ; [4 5 5 6 6 7 7 8 ...]
+    pshufb                      xm1,        xm1,         [r4 + 3 * mmsize]  ; [6 7 7 8 8 9 9 A ...]
+    pmaddubsw                   xm2,        [r6 + 0 * mmsize]
+    pmaddubsw                   xm3,        [r6 + 1 * mmsize]
+    pmaddubsw                   xm4,        [r6 + 2 * mmsize]
+    pmaddubsw                   xm1,        [r6 + 3 * mmsize]
+    paddw                       xm2,        xm3
+    paddw                       xm1,        xm4
+    paddw                       xm1,        xm2
+    psubw                       xm1,        xm0
+    movu                        [r2],       xm1                          ;row 0
+.end:
+    RET
+%endif
+%endmacro ; IPFILTER_LUMA_PS_8xN_AVX2
+
+    IPFILTER_LUMA_PS_8xN_AVX2  4
+    IPFILTER_LUMA_PS_8xN_AVX2  8
+    IPFILTER_LUMA_PS_8xN_AVX2 16
+    IPFILTER_LUMA_PS_8xN_AVX2 32
+
+
+%macro IPFILTER_LUMA_PS_16x_AVX2 2
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_horiz_ps_%1x%2, 6, 10, 7
+    mov                         r5d,               r5m
+    mov                         r4d,               r4m
+%ifdef PIC
+    lea                         r6,                [h_tab_LumaCoeff]
+    vpbroadcastq                m0,                [r6 + r4 * 8]
+%else
+    vpbroadcastq                m0,                [h_tab_LumaCoeff + r4 * 8]
+%endif
+    mova                        m6,                [h_tab_Lm + 32]
+    mova                        m1,                [h_tab_Lm]
+    mov                         r9,                %2                           ;height
+    add                         r3d,               r3d
+    vbroadcasti128              m2,                [pw_2000]
+
+    ; register map
+    ; m0      - interpolate coeff
+    ; m1 , m6 - shuffle order table
+    ; m2      - pw_2000
+
+    xor                         r7,                r7                          ; loop count variable
+    sub                         r0,                3
+    test                        r5d,               r5d
+    jz                          .label
+    lea                         r8,                [r1 * 3]                     ; r8 = (N / 2 - 1) * srcStride
+    sub                         r0,                r8                           ; r0(src)-r8
+    add                         r9,                7                            ; blkheight += N - 1  (7 - 1 = 6 ; since the last one row not in loop)
+
+.label:
+    ; Row 0
+    vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb                      m4,                m3,             m6           ; row 0 (col 4 to 7)
+    pshufb                      m3,                m1                           ; shuffled based on the col order tab_Lm row 0 (col 0 to 3)
+    pmaddubsw                   m3,                m0
+    pmaddubsw                   m4,                m0
+    phaddw                      m3,                m4                           ; DWORD [R1D R1C R0D R0C R1B R1A R0B R0A]
+
+    vbroadcasti128              m4,                [r0 + 8]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb                      m5,                m4,            m6            ;row 1 (col 4 to 7)
+    pshufb                      m4,                m1                           ;row 1 (col 0 to 3)
+    pmaddubsw                   m4,                m0
+    pmaddubsw                   m5,                m0
+    phaddw                      m4,                m5                           ; DWORD [R3D R3C R2D R2C R3B R3A R2B R2A]
+    phaddw                      m3,                m4                           ; all rows and col completed.
+
+    mova                        m5,                [h_interp8_hps_shuf]
+    vpermd                      m3,                m5,               m3
+    psubw                       m3,                m2
+
+    movu                        [r2],              m3                          ;row 0
+
+    lea                         r0,                [r0 + r1]                ; first loop src ->5th row(i.e 4)
+    lea                         r2,                [r2 + r3]                ; first loop dst ->5th row(i.e 4)
+    dec                         r9d
+    jnz                         .label
+
+    RET
+%endif
+%endmacro
+
+
+    IPFILTER_LUMA_PS_16x_AVX2 16 , 16
+    IPFILTER_LUMA_PS_16x_AVX2 16 , 8
+    IPFILTER_LUMA_PS_16x_AVX2 16 , 12
+    IPFILTER_LUMA_PS_16x_AVX2 16 , 4
+    IPFILTER_LUMA_PS_16x_AVX2 16 , 32
+    IPFILTER_LUMA_PS_16x_AVX2 16 , 64
+
+
+;--------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+%macro IPFILTER_LUMA_PP_W8 2
+INIT_XMM sse4
+cglobal interp_8tap_horiz_pp_%1x%2, 4,6,7
+    mov         r4d, r4m
+
+%ifdef PIC
+    lea         r5, [h_tab_LumaCoeff]
+    movh        m3, [r5 + r4 * 8]
+%else
+    movh        m3, [h_tab_LumaCoeff + r4 * 8]
+%endif
+    pshufd      m0, m3, 0                       ; m0 = coeff-L
+    pshufd      m1, m3, 0x55                    ; m1 = coeff-H
+    lea         r5, [h_tab_Tm]                    ; r5 = shuffle
+    mova        m2, [pw_512]                    ; m2 = 512
+
+    mov         r4d, %2
+.loopH:
+%assign x 0
+%rep %1 / 8
+    movu        m3, [r0 - 3 + x]                ; m3 = [F E D C B A 9 8 7 6 5 4 3 2 1 0]
+    pshufb      m4, m3, [r5 + 0*16]             ; m4 = [6 5 4 3 5 4 3 2 4 3 2 1 3 2 1 0]
+    pshufb      m5, m3, [r5 + 1*16]             ; m5 = [A 9 8 7 9 8 7 6 8 7 6 5 7 6 5 4]
+    pshufb          m3, [r5 + 2*16]             ; m3 = [E D C B D C B A C B A 9 B A 9 8]
+    pmaddubsw   m4, m0
+    pmaddubsw   m6, m5, m1
+    pmaddubsw   m5, m0
+    pmaddubsw   m3, m1
+    paddw       m4, m6
+    paddw       m5, m3
+    phaddw      m4, m5
+    pmulhrsw    m4, m2
+    packuswb    m4, m4
+    movh        [r2 + x], m4
+%assign x x+8
+%endrep
+
+    add       r0, r1
+    add       r2, r3
+
+    dec       r4d
+    jnz      .loopH
+    RET
+%endmacro
+
+    IPFILTER_LUMA_PP_W8      8,  4
+    IPFILTER_LUMA_PP_W8      8,  8
+    IPFILTER_LUMA_PP_W8      8, 16
+    IPFILTER_LUMA_PP_W8      8, 32
+    IPFILTER_LUMA_PP_W8     16,  4
+    IPFILTER_LUMA_PP_W8     16,  8
+    IPFILTER_LUMA_PP_W8     16, 12
+    IPFILTER_LUMA_PP_W8     16, 16
+    IPFILTER_LUMA_PP_W8     16, 32
+    IPFILTER_LUMA_PP_W8     16, 64
+    IPFILTER_LUMA_PP_W8     24, 32
+    IPFILTER_LUMA_PP_W8     32,  8
+    IPFILTER_LUMA_PP_W8     32, 16
+    IPFILTER_LUMA_PP_W8     32, 24
+    IPFILTER_LUMA_PP_W8     32, 32
+    IPFILTER_LUMA_PP_W8     32, 64
+    IPFILTER_LUMA_PP_W8     48, 64
+    IPFILTER_LUMA_PP_W8     64, 16
+    IPFILTER_LUMA_PP_W8     64, 32
+    IPFILTER_LUMA_PP_W8     64, 48
+    IPFILTER_LUMA_PP_W8     64, 64
+
+;----------------------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_ps_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;----------------------------------------------------------------------------------------------------------------------------
+    IPFILTER_LUMA 4, 4, ps
+    IPFILTER_LUMA 8, 8, ps
+    IPFILTER_LUMA 8, 4, ps
+    IPFILTER_LUMA 4, 8, ps
+    IPFILTER_LUMA 16, 16, ps
+    IPFILTER_LUMA 16, 8, ps
+    IPFILTER_LUMA 8, 16, ps
+    IPFILTER_LUMA 16, 12, ps
+    IPFILTER_LUMA 12, 16, ps
+    IPFILTER_LUMA 16, 4, ps
+    IPFILTER_LUMA 4, 16, ps
+    IPFILTER_LUMA 32, 32, ps
+    IPFILTER_LUMA 32, 16, ps
+    IPFILTER_LUMA 16, 32, ps
+    IPFILTER_LUMA 32, 24, ps
+    IPFILTER_LUMA 24, 32, ps
+    IPFILTER_LUMA 32, 8, ps
+    IPFILTER_LUMA 8, 32, ps
+    IPFILTER_LUMA 64, 64, ps
+    IPFILTER_LUMA 64, 32, ps
+    IPFILTER_LUMA 32, 64, ps
+    IPFILTER_LUMA 64, 48, ps
+    IPFILTER_LUMA 48, 64, ps
+    IPFILTER_LUMA 64, 16, ps
+    IPFILTER_LUMA 16, 64, ps
+
+;-----------------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_ps_2x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;-----------------------------------------------------------------------------------------------------------------------------
+%macro FILTER_HORIZ_CHROMA_2xN 2
+INIT_XMM sse4
+cglobal interp_4tap_horiz_ps_%1x%2, 4, 7, 4, src, srcstride, dst, dststride
+%define coef2  m3
+%define Tm0    m2
+%define t1     m1
+%define t0     m0
+
+    dec        srcq
+    mov        r4d, r4m
+    add        dststrided, dststrided
+
+%ifdef PIC
+    lea        r6, [h_tab_ChromaCoeff]
+    movd       coef2, [r6 + r4 * 4]
+%else
+    movd       coef2, [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufd     coef2, coef2, 0
+    mova       t1, [pw_2000]
+    mova       Tm0, [h_tab_Tm]
+
+    mov        r4d, %2
+    cmp        r5m, byte 0
+    je         .loopH
+    sub        srcq, srcstrideq
+    add        r4d, 3
+
+.loopH:
+    movh       t0, [srcq]
+    pshufb     t0, t0, Tm0
+    pmaddubsw  t0, coef2
+    phaddw     t0, t0
+    psubw      t0, t1
+    movd       [dstq], t0
+
+    lea        srcq, [srcq + srcstrideq]
+    lea        dstq, [dstq + dststrideq]
+
+    dec        r4d
+    jnz        .loopH
+
+    RET
+%endmacro
+
+    FILTER_HORIZ_CHROMA_2xN 2, 4
+    FILTER_HORIZ_CHROMA_2xN 2, 8
+
+    FILTER_HORIZ_CHROMA_2xN 2, 16
+
+;-----------------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_ps_4x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;-----------------------------------------------------------------------------------------------------------------------------
+%macro FILTER_HORIZ_CHROMA_4xN 2
+INIT_XMM sse4
+cglobal interp_4tap_horiz_ps_%1x%2, 4, 7, 4, src, srcstride, dst, dststride
+%define coef2  m3
+%define Tm0    m2
+%define t1     m1
+%define t0     m0
+
+    dec        srcq
+    mov        r4d, r4m
+    add        dststrided, dststrided
+
+%ifdef PIC
+    lea        r6, [h_tab_ChromaCoeff]
+    movd       coef2, [r6 + r4 * 4]
+%else
+    movd       coef2, [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufd     coef2, coef2, 0
+    mova       t1, [pw_2000]
+    mova       Tm0, [h_tab_Tm]
+
+    mov        r4d, %2
+    cmp        r5m, byte 0
+    je         .loopH
+    sub        srcq, srcstrideq
+    add        r4d, 3
+
+.loopH:
+    movh       t0, [srcq]
+    pshufb     t0, t0, Tm0
+    pmaddubsw  t0, coef2
+    phaddw     t0, t0
+    psubw      t0, t1
+    movlps     [dstq], t0
+
+    lea        srcq, [srcq + srcstrideq]
+    lea        dstq, [dstq + dststrideq]
+
+    dec        r4d
+    jnz        .loopH
+    RET
+%endmacro
+
+    FILTER_HORIZ_CHROMA_4xN 4, 2
+    FILTER_HORIZ_CHROMA_4xN 4, 4
+    FILTER_HORIZ_CHROMA_4xN 4, 8
+    FILTER_HORIZ_CHROMA_4xN 4, 16
+
+    FILTER_HORIZ_CHROMA_4xN 4, 32
+
+%macro PROCESS_CHROMA_W6 3
+    movu       %1, [srcq]
+    pshufb     %2, %1, Tm0
+    pmaddubsw  %2, coef2
+    pshufb     %1, %1, Tm1
+    pmaddubsw  %1, coef2
+    phaddw     %2, %1
+    psubw      %2, %3
+    movh       [dstq], %2
+    pshufd     %2, %2, 2
+    movd       [dstq + 8], %2
+%endmacro
+
+%macro PROCESS_CHROMA_W12 3
+    movu       %1, [srcq]
+    pshufb     %2, %1, Tm0
+    pmaddubsw  %2, coef2
+    pshufb     %1, %1, Tm1
+    pmaddubsw  %1, coef2
+    phaddw     %2, %1
+    psubw      %2, %3
+    movu       [dstq], %2
+    movu       %1, [srcq + 8]
+    pshufb     %1, %1, Tm0
+    pmaddubsw  %1, coef2
+    phaddw     %1, %1
+    psubw      %1, %3
+    movh       [dstq + 16], %1
+%endmacro
+
+;-----------------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_ps_6x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;-----------------------------------------------------------------------------------------------------------------------------
+%macro FILTER_HORIZ_CHROMA 2
+INIT_XMM sse4
+cglobal interp_4tap_horiz_ps_%1x%2, 4, 7, 6, src, srcstride, dst, dststride
+%define coef2    m5
+%define Tm0      m4
+%define Tm1      m3
+%define t2       m2
+%define t1       m1
+%define t0       m0
+
+    dec     srcq
+    mov     r4d, r4m
+    add     dststrided, dststrided
+
+%ifdef PIC
+    lea     r6, [h_tab_ChromaCoeff]
+    movd    coef2, [r6 + r4 * 4]
+%else
+    movd    coef2, [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufd  coef2, coef2, 0
+    mova    t2, [pw_2000]
+    mova    Tm0, [h_tab_Tm]
+    mova    Tm1, [h_tab_Tm + 16]
+
+    mov     r4d, %2
+    cmp     r5m, byte 0
+    je      .loopH
+    sub     srcq, srcstrideq
+    add     r4d, 3
+
+.loopH:
+    PROCESS_CHROMA_W%1  t0, t1, t2
+    add     srcq, srcstrideq
+    add     dstq, dststrideq
+
+    dec     r4d
+    jnz     .loopH
+
+    RET
+%endmacro
+
+    FILTER_HORIZ_CHROMA 6, 8
+    FILTER_HORIZ_CHROMA 12, 16
+
+    FILTER_HORIZ_CHROMA 6, 16
+    FILTER_HORIZ_CHROMA 12, 32
+
+%macro PROCESS_CHROMA_W8 3
+    movu        %1, [srcq]
+    pshufb      %2, %1, Tm0
+    pmaddubsw   %2, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %2, %1
+    psubw       %2, %3
+    movu        [dstq], %2
+%endmacro
+
+;-----------------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_ps_8x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;-----------------------------------------------------------------------------------------------------------------------------
+%macro FILTER_HORIZ_CHROMA_8xN 2
+INIT_XMM sse4
+cglobal interp_4tap_horiz_ps_%1x%2, 4, 7, 6, src, srcstride, dst, dststride
+%define coef2    m5
+%define Tm0      m4
+%define Tm1      m3
+%define t2       m2
+%define t1       m1
+%define t0       m0
+
+    dec     srcq
+    mov     r4d, r4m
+    add     dststrided, dststrided
+
+%ifdef PIC
+    lea     r6, [h_tab_ChromaCoeff]
+    movd    coef2, [r6 + r4 * 4]
+%else
+    movd    coef2, [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufd  coef2, coef2, 0
+    mova    t2, [pw_2000]
+    mova    Tm0, [h_tab_Tm]
+    mova    Tm1, [h_tab_Tm + 16]
+
+    mov     r4d, %2
+    cmp     r5m, byte 0
+    je      .loopH
+    sub     srcq, srcstrideq
+    add     r4d, 3
+
+.loopH:
+    PROCESS_CHROMA_W8  t0, t1, t2
+    add     srcq, srcstrideq
+    add     dstq, dststrideq
+
+    dec     r4d
+    jnz     .loopH
+
+    RET
+%endmacro
+
+    FILTER_HORIZ_CHROMA_8xN 8, 2
+    FILTER_HORIZ_CHROMA_8xN 8, 4
+    FILTER_HORIZ_CHROMA_8xN 8, 6
+    FILTER_HORIZ_CHROMA_8xN 8, 8
+    FILTER_HORIZ_CHROMA_8xN 8, 16
+    FILTER_HORIZ_CHROMA_8xN 8, 32
+
+    FILTER_HORIZ_CHROMA_8xN 8, 12
+    FILTER_HORIZ_CHROMA_8xN 8, 64
+
+%macro PROCESS_CHROMA_W16 4
+    movu        %1, [srcq]
+    pshufb      %2, %1, Tm0
+    pmaddubsw   %2, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %2, %1
+    movu        %1, [srcq + 8]
+    pshufb      %4, %1, Tm0
+    pmaddubsw   %4, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %4, %1
+    psubw       %2, %3
+    psubw       %4, %3
+    movu        [dstq], %2
+    movu        [dstq + 16], %4
+%endmacro
+
+%macro PROCESS_CHROMA_W24 4
+    movu        %1, [srcq]
+    pshufb      %2, %1, Tm0
+    pmaddubsw   %2, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %2, %1
+    movu        %1, [srcq + 8]
+    pshufb      %4, %1, Tm0
+    pmaddubsw   %4, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %4, %1
+    psubw       %2, %3
+    psubw       %4, %3
+    movu        [dstq], %2
+    movu        [dstq + 16], %4
+    movu        %1, [srcq + 16]
+    pshufb      %2, %1, Tm0
+    pmaddubsw   %2, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %2, %1
+    psubw       %2, %3
+    movu        [dstq + 32], %2
+%endmacro
+
+%macro PROCESS_CHROMA_W32 4
+    movu        %1, [srcq]
+    pshufb      %2, %1, Tm0
+    pmaddubsw   %2, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %2, %1
+    movu        %1, [srcq + 8]
+    pshufb      %4, %1, Tm0
+    pmaddubsw   %4, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %4, %1
+    psubw       %2, %3
+    psubw       %4, %3
+    movu        [dstq], %2
+    movu        [dstq + 16], %4
+    movu        %1, [srcq + 16]
+    pshufb      %2, %1, Tm0
+    pmaddubsw   %2, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %2, %1
+    movu        %1, [srcq + 24]
+    pshufb      %4, %1, Tm0
+    pmaddubsw   %4, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %4, %1
+    psubw       %2, %3
+    psubw       %4, %3
+    movu        [dstq + 32], %2
+    movu        [dstq + 48], %4
+%endmacro
+
+%macro PROCESS_CHROMA_W16o 5
+    movu        %1, [srcq + %5]
+    pshufb      %2, %1, Tm0
+    pmaddubsw   %2, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %2, %1
+    movu        %1, [srcq + %5 + 8]
+    pshufb      %4, %1, Tm0
+    pmaddubsw   %4, coef2
+    pshufb      %1, %1, Tm1
+    pmaddubsw   %1, coef2
+    phaddw      %4, %1
+    psubw       %2, %3
+    psubw       %4, %3
+    movu        [dstq + %5 * 2], %2
+    movu        [dstq + %5 * 2 + 16], %4
+%endmacro
+
+%macro PROCESS_CHROMA_W48 4
+    PROCESS_CHROMA_W16o %1, %2, %3, %4, 0
+    PROCESS_CHROMA_W16o %1, %2, %3, %4, 16
+    PROCESS_CHROMA_W16o %1, %2, %3, %4, 32
+%endmacro
+
+%macro PROCESS_CHROMA_W64 4
+    PROCESS_CHROMA_W16o %1, %2, %3, %4, 0
+    PROCESS_CHROMA_W16o %1, %2, %3, %4, 16
+    PROCESS_CHROMA_W16o %1, %2, %3, %4, 32
+    PROCESS_CHROMA_W16o %1, %2, %3, %4, 48
+%endmacro
+
+;------------------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_ps_%1x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;------------------------------------------------------------------------------------------------------------------------------
+%macro FILTER_HORIZ_CHROMA_WxN 2
+INIT_XMM sse4
+cglobal interp_4tap_horiz_ps_%1x%2, 4, 7, 7, src, srcstride, dst, dststride
+%define coef2    m6
+%define Tm0      m5
+%define Tm1      m4
+%define t3       m3
+%define t2       m2
+%define t1       m1
+%define t0       m0
+
+    dec     srcq
+    mov     r4d, r4m
+    add     dststrided, dststrided
+
+%ifdef PIC
+    lea     r6, [h_tab_ChromaCoeff]
+    movd    coef2, [r6 + r4 * 4]
+%else
+    movd    coef2, [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufd  coef2, coef2, 0
+    mova    t2, [pw_2000]
+    mova    Tm0, [h_tab_Tm]
+    mova    Tm1, [h_tab_Tm + 16]
+
+    mov     r4d, %2
+    cmp     r5m, byte 0
+    je      .loopH
+    sub     srcq, srcstrideq
+    add     r4d, 3
+
+.loopH:
+    PROCESS_CHROMA_W%1   t0, t1, t2, t3
+    add     srcq, srcstrideq
+    add     dstq, dststrideq
+
+    dec     r4d
+    jnz     .loopH
+
+    RET
+%endmacro
+
+    FILTER_HORIZ_CHROMA_WxN 16, 4
+    FILTER_HORIZ_CHROMA_WxN 16, 8
+    FILTER_HORIZ_CHROMA_WxN 16, 12
+    FILTER_HORIZ_CHROMA_WxN 16, 16
+    FILTER_HORIZ_CHROMA_WxN 16, 32
+    FILTER_HORIZ_CHROMA_WxN 24, 32
+    FILTER_HORIZ_CHROMA_WxN 32,  8
+    FILTER_HORIZ_CHROMA_WxN 32, 16
+    FILTER_HORIZ_CHROMA_WxN 32, 24
+    FILTER_HORIZ_CHROMA_WxN 32, 32
+
+    FILTER_HORIZ_CHROMA_WxN 16, 24
+    FILTER_HORIZ_CHROMA_WxN 16, 64
+    FILTER_HORIZ_CHROMA_WxN 24, 64
+    FILTER_HORIZ_CHROMA_WxN 32, 48
+    FILTER_HORIZ_CHROMA_WxN 32, 64
+
+    FILTER_HORIZ_CHROMA_WxN 64, 64
+    FILTER_HORIZ_CHROMA_WxN 64, 32
+    FILTER_HORIZ_CHROMA_WxN 64, 48
+    FILTER_HORIZ_CHROMA_WxN 48, 64
+    FILTER_HORIZ_CHROMA_WxN 64, 16
+
+;-----------------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_ps_32x32(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;-----------------------------------------------------------------------------------------------------------------------------;
+INIT_YMM avx2
+cglobal interp_4tap_horiz_ps_32x32, 4,6,8
+    mov             r4d, r4m
+    add             r3d, r3d
+    dec             r0
+
+    ; check isRowExt
+    cmp             r5m, byte 0
+
+    lea             r5, [h_tab_ChromaCoeff]
+    vpbroadcastw    m0, [r5 + r4 * 4 + 0]
+    vpbroadcastw    m1, [r5 + r4 * 4 + 2]
+    mova            m7, [pw_2000]
+
+    ; register map
+    ; m0 - interpolate coeff Low
+    ; m1 - interpolate coeff High
+    ; m7 - constant pw_2000
+    mov             r4d, 32
+    je             .loop
+    sub             r0, r1
+    add             r4d, 3
+
+.loop:
+    ; Row 0
+    movu            m2, [r0]
+    movu            m3, [r0 + 1]
+    punpckhbw       m4, m2, m3
+    punpcklbw       m2, m3
+    pmaddubsw       m4, m0
+    pmaddubsw       m2, m0
+
+    movu            m3, [r0 + 2]
+    movu            m5, [r0 + 3]
+    punpckhbw       m6, m3, m5
+    punpcklbw       m3, m5
+    pmaddubsw       m6, m1
+    pmaddubsw       m3, m1
+
+    paddw           m4, m6
+    paddw           m2, m3
+    psubw           m4, m7
+    psubw           m2, m7
+    vperm2i128      m3, m2, m4, 0x20
+    vperm2i128      m5, m2, m4, 0x31
+    movu            [r2], m3
+    movu            [r2 + mmsize], m5
+
+    add             r2, r3
+    add             r0, r1
+    dec             r4d
+    jnz            .loop
+    RET
+
+;-----------------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_ps_16x16(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;-----------------------------------------------------------------------------------------------------------------------------;
+INIT_YMM avx2
+cglobal interp_4tap_horiz_ps_16x16, 4,7,6
+    mov             r4d, r4m
+    mov             r5d, r5m
+    add             r3d, r3d
+
+%ifdef PIC
+    lea               r6,           [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,           [r6 + r4 * 4]
+%else
+    vpbroadcastd      m0,           [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    vbroadcasti128     m2,           [pw_1]
+    vbroadcasti128     m5,           [pw_2000]
+    mova               m1,           [h_tab_Tm]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+    mov                r6d,         16
+    dec                r0
+    test                r5d,        r5d
+    je                 .loop
+    sub                r0 ,         r1
+    add                r6d ,        3
+
+.loop:
+    ; Row 0
+    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+    vbroadcasti128    m4,           [r0 + 8]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+
+    packssdw          m3,           m4
+    psubw             m3,           m5
+    vpermq            m3,           m3,          11011000b
+    movu              [r2],         m3
+
+    add                r2,          r3
+    add                r0,          r1
+    dec                r6d
+    jnz                .loop
+    RET
+
+;-----------------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_ps_16xN(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;-----------------------------------------------------------------------------------------------------------------------------
+%macro IPFILTER_CHROMA_PS_16xN_AVX2 2
+INIT_YMM avx2
+cglobal interp_4tap_horiz_ps_%1x%2, 4,7,6
+    mov                    r4d,        r4m
+    mov                    r5d,        r5m
+    add                    r3d,        r3d
+
+%ifdef PIC
+    lea                    r6,         [h_tab_ChromaCoeff]
+    vpbroadcastd           m0,         [r6 + r4 * 4]
+%else
+    vpbroadcastd           m0,         [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    vbroadcasti128         m2,         [pw_1]
+    vbroadcasti128         m5,         [pw_2000]
+    mova                   m1,         [h_tab_Tm]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+    mov                    r6d,        %2
+    dec                    r0
+    test                   r5d,        r5d
+    je                     .loop
+    sub                    r0 ,        r1
+    add                    r6d ,       3
+
+.loop:
+    ; Row 0
+    vbroadcasti128         m3,         [r0]
+    pshufb                 m3,         m1
+    pmaddubsw              m3,         m0
+    pmaddwd                m3,         m2
+    vbroadcasti128         m4,         [r0 + 8]
+    pshufb                 m4,         m1
+    pmaddubsw              m4,         m0
+    pmaddwd                m4,         m2
+
+    packssdw               m3,         m4
+    psubw                  m3,         m5
+
+    vpermq                 m3,         m3,          11011000b
+    movu                   [r2],       m3
+
+    add                    r2,         r3
+    add                    r0,         r1
+    dec                    r6d
+    jnz                    .loop
+    RET
+%endmacro
+
+    IPFILTER_CHROMA_PS_16xN_AVX2  16 , 32
+    IPFILTER_CHROMA_PS_16xN_AVX2  16 , 12
+    IPFILTER_CHROMA_PS_16xN_AVX2  16 , 8
+    IPFILTER_CHROMA_PS_16xN_AVX2  16 , 4
+    IPFILTER_CHROMA_PS_16xN_AVX2  16 , 24
+    IPFILTER_CHROMA_PS_16xN_AVX2  16 , 64
+
+;-----------------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_ps_32xN(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;-----------------------------------------------------------------------------------------------------------------------------
+%macro IPFILTER_CHROMA_PS_32xN_AVX2 2
+INIT_YMM avx2
+cglobal interp_4tap_horiz_ps_%1x%2, 4,7,6
+    mov                r4d,          r4m
+    mov                r5d,          r5m
+    add                r3d,          r3d
+
+%ifdef PIC
+    lea                r6,           [h_tab_ChromaCoeff]
+    vpbroadcastd       m0,           [r6 + r4 * 4]
+%else
+    vpbroadcastd       m0,           [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    vbroadcasti128     m2,           [pw_1]
+    vbroadcasti128     m5,           [pw_2000]
+    mova               m1,           [h_tab_Tm]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+    mov                r6d,          %2
+    dec                r0
+    test               r5d,          r5d
+    je                 .loop
+    sub                r0 ,          r1
+    add                r6d ,         3
+
+.loop:
+    ; Row 0
+    vbroadcasti128     m3,           [r0]
+    pshufb             m3,           m1
+    pmaddubsw          m3,           m0
+    pmaddwd            m3,           m2
+    vbroadcasti128     m4,           [r0 + 8]
+    pshufb             m4,           m1
+    pmaddubsw          m4,           m0
+    pmaddwd            m4,           m2
+
+    packssdw           m3,           m4
+    psubw              m3,           m5
+
+    vpermq             m3,           m3,          11011000b
+    movu              [r2],          m3
+
+    vbroadcasti128     m3,           [r0 + 16]
+    pshufb             m3,           m1
+    pmaddubsw          m3,           m0
+    pmaddwd            m3,           m2
+    vbroadcasti128     m4,           [r0 + 24]
+    pshufb             m4,           m1
+    pmaddubsw          m4,           m0
+    pmaddwd            m4,           m2
+
+    packssdw           m3,           m4
+    psubw              m3,           m5
+
+    vpermq             m3,           m3,          11011000b
+    movu               [r2 + 32],    m3
+
+    add                r2,           r3
+    add                r0,           r1
+    dec                r6d
+    jnz                .loop
+    RET
+%endmacro
+
+    IPFILTER_CHROMA_PS_32xN_AVX2  32 , 16
+    IPFILTER_CHROMA_PS_32xN_AVX2  32 , 24
+    IPFILTER_CHROMA_PS_32xN_AVX2  32 , 8
+    IPFILTER_CHROMA_PS_32xN_AVX2  32 , 64
+    IPFILTER_CHROMA_PS_32xN_AVX2  32 , 48
+
+;-----------------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_ps_4x4(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;-----------------------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+cglobal interp_4tap_horiz_ps_4x4, 4,7,5
+    mov             r4d, r4m
+    mov             r5d, r5m
+    add             r3d, r3d
+
+%ifdef PIC
+    lea               r6,           [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,           [r6 + r4 * 4]
+%else
+    vpbroadcastd      m0,           [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    vbroadcasti128     m2,           [pw_1]
+    vbroadcasti128     m1,           [h_tab_Tm]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    dec                r0
+    test                r5d,       r5d
+    je                 .label
+    sub                r0 , r1
+
+.label:
+    ; Row 0-1
+    movu              xm3,           [r0]
+    vinserti128       m3,           m3,      [r0 + r1],     1
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+
+    ; Row 2-3
+    lea               r0,           [r0 + r1 * 2]
+    movu              xm4,           [r0]
+    vinserti128       m4,           m4,      [r0 + r1],     1
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+
+    packssdw          m3,           m4
+    psubw             m3,           [pw_2000]
+    vextracti128      xm4,          m3,     1
+    movq              [r2],         xm3
+    movq              [r2+r3],      xm4
+    lea               r2,           [r2 + r3 * 2]
+    movhps            [r2],         xm3
+    movhps            [r2 + r3],    xm4
+
+    test                r5d,        r5d
+    jz                .end
+    lea               r2,           [r2 + r3 * 2]
+    lea               r0,           [r0 + r1 * 2]
+
+    ;Row 5-6
+    movu              xm3,          [r0]
+    vinserti128       m3,           m3,      [r0 + r1],     1
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+
+    ; Row 7
+    lea               r0,           [r0 + r1 * 2]
+    vbroadcasti128    m4,           [r0]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+
+    packssdw          m3,           m4
+    psubw             m3,           [pw_2000]
+
+    vextracti128      xm4,          m3,     1
+    movq              [r2],         xm3
+    movq              [r2+r3],      xm4
+    lea               r2,           [r2 + r3 * 2]
+    movhps            [r2],         xm3
+.end:
+    RET
+
+cglobal interp_4tap_horiz_ps_4x2, 4,7,5
+    mov             r4d, r4m
+    mov             r5d, r5m
+    add             r3d, r3d
+
+%ifdef PIC
+    lea               r6,           [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,           [r6 + r4 * 4]
+%else
+    vpbroadcastd      m0,           [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    vbroadcasti128     m2,           [pw_1]
+    vbroadcasti128     m1,           [h_tab_Tm]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    dec                r0
+    test                r5d,       r5d
+    je                 .label
+    sub                r0 , r1
+
+.label:
+    ; Row 0-1
+    movu              xm3,           [r0]
+    vinserti128       m3,           m3,      [r0 + r1],     1
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+
+    packssdw          m3,           m3
+    psubw             m3,           [pw_2000]
+    vextracti128      xm4,          m3,     1
+    movq              [r2],         xm3
+    movq              [r2+r3],      xm4
+
+    test              r5d,          r5d
+    jz                .end
+    lea               r2,           [r2 + r3 * 2]
+    lea               r0,           [r0 + r1 * 2]
+
+    ;Row 2-3
+    movu              xm3,          [r0]
+    vinserti128       m3,           m3,      [r0 + r1],     1
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+
+    ; Row 5
+    lea               r0,           [r0 + r1 * 2]
+    vbroadcasti128    m4,           [r0]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+
+    packssdw          m3,           m4
+    psubw             m3,           [pw_2000]
+
+    vextracti128      xm4,          m3,     1
+    movq              [r2],         xm3
+    movq              [r2+r3],      xm4
+    lea               r2,           [r2 + r3 * 2]
+    movhps            [r2],         xm3
+.end:
+    RET
+
+;-----------------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_ps_4xN(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;-----------------------------------------------------------------------------------------------------------------------------;
+%macro IPFILTER_CHROMA_PS_4xN_AVX2 2
+INIT_YMM avx2
+cglobal interp_4tap_horiz_ps_%1x%2, 4,7,5
+    mov             r4d, r4m
+    mov             r5d, r5m
+    add             r3d, r3d
+
+%ifdef PIC
+    lea               r6,           [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,           [r6 + r4 * 4]
+%else
+    vpbroadcastd      m0,           [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    vbroadcasti128     m2,           [pw_1]
+    vbroadcasti128     m1,           [h_tab_Tm]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    mov              r4,                %2
+    dec              r0
+    test             r5d,       r5d
+    je               .loop
+    sub              r0 ,               r1
+
+
+.loop:
+    sub              r4d,           4
+    ; Row 0-1
+    movu              xm3,          [r0]
+    vinserti128       m3,           m3,      [r0 + r1],     1
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+
+    ; Row 2-3
+    lea               r0,           [r0 + r1 * 2]
+    movu              xm4,          [r0]
+    vinserti128       m4,           m4,      [r0 + r1],     1
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+
+    packssdw          m3,           m4
+    psubw             m3,           [pw_2000]
+    vextracti128      xm4,          m3,     1
+    movq              [r2],         xm3
+    movq              [r2+r3],      xm4
+    lea               r2,           [r2 + r3 * 2]
+    movhps            [r2],         xm3
+    movhps            [r2 + r3],    xm4
+
+    lea               r2,           [r2 + r3 * 2]
+    lea               r0,           [r0 + r1 * 2]
+
+    test              r4d,          r4d
+    jnz               .loop
+    test                r5d,        r5d
+    jz                .end
+
+    ;Row 5-6
+    movu              xm3,          [r0]
+    vinserti128       m3,           m3,      [r0 + r1],     1
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+
+    ; Row 7
+    lea               r0,           [r0 + r1 * 2]
+    vbroadcasti128    m4,           [r0]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+
+    packssdw          m3,           m4
+    psubw             m3,           [pw_2000]
+
+    vextracti128      xm4,          m3,     1
+    movq              [r2],         xm3
+    movq              [r2+r3],      xm4
+    lea               r2,           [r2 + r3 * 2]
+    movhps            [r2],         xm3
+.end:
+    RET
+%endmacro
+
+    IPFILTER_CHROMA_PS_4xN_AVX2  4 , 8
+    IPFILTER_CHROMA_PS_4xN_AVX2  4 , 16
+
+;-----------------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_ps_8x8(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;-----------------------------------------------------------------------------------------------------------------------------;
+INIT_YMM avx2
+cglobal interp_4tap_horiz_ps_8x8, 4,7,6
+    mov             r4d, r4m
+    mov             r5d, r5m
+    add             r3d, r3d
+
+%ifdef PIC
+    lea               r6,           [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,           [r6 + r4 * 4]
+%else
+    vpbroadcastd      m0,           [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    vbroadcasti128     m2,           [pw_1]
+    vbroadcasti128     m5,           [pw_2000]
+    mova               m1,           [h_tab_Tm]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    mov                r6d,      4
+    dec                r0
+    test                r5d,     r5d
+    je                 .loop
+    sub                r0 ,      r1
+    add                r6d ,     1
+
+.loop:
+     dec               r6d
+    ; Row 0
+    vbroadcasti128    m3,           [r0]
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+
+    ; Row 1
+    vbroadcasti128    m4,           [r0 + r1]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+
+    packssdw          m3,           m4
+    psubw             m3,           m5
+
+    vpermq            m3,           m3,          11011000b
+    vextracti128      xm4,          m3,     1
+    movu             [r2],         xm3
+    movu             [r2 + r3],    xm4
+
+    lea               r2,           [r2 + r3 * 2]
+    lea               r0,           [r0 + r1 * 2]
+    test               r6d,          r6d
+    jnz               .loop
+    test              r5d,         r5d
+    je                .end
+
+    ;Row 11
+    vbroadcasti128    m3,           [r0]
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+
+    packssdw          m3,           m3
+    psubw             m3,           m5
+    vpermq            m3,           m3,          11011000b
+    movu             [r2],         xm3
+.end:
+    RET
+
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_4x2, 4,6,4
+    mov             r4d, r4m
+%ifdef PIC
+    lea               r5,           [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,           [r5 + r4 * 4]
+%else
+    vpbroadcastd      m0,           [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    vbroadcasti128    m1,           [h_tab_Tm]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+
+    ; Row 0-1
+    movu              xm2,          [r0 - 1]
+    vinserti128       m2,           m2,      [r0 + r1 - 1],     1
+    pshufb            m2,           m1
+    pmaddubsw         m2,           m0
+    pmaddwd           m2,           [pw_1]
+
+    packssdw          m2,           m2
+    pmulhrsw          m2,           [pw_512]
+    vextracti128      xm3,          m2,     1
+    packuswb          xm2,          xm3
+
+    movd              [r2],         xm2
+    pextrd            [r2+r3],      xm2,     2
+    RET
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp_32xN(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+%macro IPFILTER_CHROMA_PP_32xN_AVX2 2
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_%1x%2, 4,6,7
+    mov             r4d, r4m
+
+%ifdef PIC
+    lea               r5,           [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,           [r5 + r4 * 4]
+%else
+    vpbroadcastd      m0,           [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    mova              m1,           [interp4_horiz_shuf1]
+    vpbroadcastd      m2,           [pw_1]
+    mova              m6,           [pw_512]
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    dec               r0
+    mov               r4d,          %2
+
+.loop:
+    ; Row 0
+    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+    vbroadcasti128    m4,           [r0 + 4]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+    packssdw          m3,           m4
+    pmulhrsw          m3,           m6
+
+    vbroadcasti128    m4,           [r0 + 16]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+    vbroadcasti128    m5,           [r0 + 20]
+    pshufb            m5,           m1
+    pmaddubsw         m5,           m0
+    pmaddwd           m5,           m2
+    packssdw          m4,           m5
+    pmulhrsw          m4,           m6
+
+    packuswb          m3,           m4
+    vpermq            m3,           m3,      11011000b
+
+    movu              [r2],         m3
+    add               r2,           r3
+    add               r0,           r1
+    dec               r4d
+    jnz               .loop
+    RET
+%endmacro
+
+    IPFILTER_CHROMA_PP_32xN_AVX2 32, 16
+    IPFILTER_CHROMA_PP_32xN_AVX2 32, 24
+    IPFILTER_CHROMA_PP_32xN_AVX2 32, 8
+    IPFILTER_CHROMA_PP_32xN_AVX2 32, 64
+    IPFILTER_CHROMA_PP_32xN_AVX2 32, 48
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp_8xN(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+%macro IPFILTER_CHROMA_PP_8xN_AVX2 2
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_%1x%2, 4,6,6
+    mov               r4d,    r4m
+
+%ifdef PIC
+    lea               r5,           [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,           [r5 + r4 * 4]
+%else
+    vpbroadcastd      m0,           [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    movu              m1,           [h_tab_Tm]
+    vpbroadcastd      m2,           [pw_1]
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    sub               r0,           1
+    mov               r4d,          %2
+
+.loop:
+    sub               r4d,          4
+    ; Row 0
+    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+
+    ; Row 1
+    vbroadcasti128    m4,           [r0 + r1]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+    packssdw          m3,           m4
+    pmulhrsw          m3,           [pw_512]
+    lea               r0,           [r0 + r1 * 2]
+
+    ; Row 2
+    vbroadcasti128    m4,           [r0 ]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+
+    ; Row 3
+    vbroadcasti128    m5,           [r0 + r1]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m5,           m1
+    pmaddubsw         m5,           m0
+    pmaddwd           m5,           m2
+    packssdw          m4,           m5
+    pmulhrsw          m4,           [pw_512]
+
+    packuswb          m3,           m4
+    mova              m5,           [interp_4tap_8x8_horiz_shuf]
+    vpermd            m3,           m5,     m3
+    vextracti128      xm4,          m3,     1
+    movq              [r2],         xm3
+    movhps            [r2 + r3],    xm3
+    lea               r2,           [r2 + r3 * 2]
+    movq              [r2],         xm4
+    movhps            [r2 + r3],    xm4
+    lea               r2,           [r2 + r3 * 2]
+    lea               r0,           [r0 + r1*2]
+    test              r4d,          r4d
+    jnz               .loop
+    RET
+%endmacro
+
+    IPFILTER_CHROMA_PP_8xN_AVX2   8 , 16
+    IPFILTER_CHROMA_PP_8xN_AVX2   8 , 32
+    IPFILTER_CHROMA_PP_8xN_AVX2   8 , 4
+    IPFILTER_CHROMA_PP_8xN_AVX2   8 , 64
+    IPFILTER_CHROMA_PP_8xN_AVX2   8 , 12
+    
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp_4xN(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+%macro IPFILTER_CHROMA_PP_4xN_AVX2 2
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_%1x%2, 4,6,6
+    mov             r4d, r4m
+
+%ifdef PIC
+    lea               r5,           [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,           [r5 + r4 * 4]
+%else
+    vpbroadcastd      m0,           [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    vpbroadcastd      m2,           [pw_1]
+    vbroadcasti128    m1,           [h_tab_Tm]
+    mov               r4d,          %2
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    dec                r0
+
+.loop:
+    sub               r4d,          4
+    ; Row 0-1
+    movu              xm3,          [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    vinserti128       m3,           m3,      [r0 + r1],     1
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+
+    ; Row 2-3
+    lea               r0,           [r0 + r1 * 2]
+    movu              xm4,          [r0]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    vinserti128       m4,           m4,      [r0 + r1],     1
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+
+    packssdw          m3,           m4
+    pmulhrsw          m3,           [pw_512]
+    vextracti128      xm4,          m3,                     1
+    packuswb          xm3,          xm4
+
+    movd              [r2],         xm3
+    pextrd            [r2+r3],      xm3,                    2
+    lea               r2,           [r2 + r3 * 2]
+    pextrd            [r2],         xm3,                    1
+    pextrd            [r2+r3],      xm3,                    3
+
+    lea               r0,           [r0 + r1 * 2]
+    lea               r2,           [r2 + r3 * 2]
+    test              r4d,          r4d
+    jnz               .loop
+    RET
+%endmacro
+
+    IPFILTER_CHROMA_PP_4xN_AVX2  4 , 8
+    IPFILTER_CHROMA_PP_4xN_AVX2  4 , 16
+
+%macro IPFILTER_LUMA_PS_32xN_AVX2 2
+INIT_YMM avx2
+cglobal interp_8tap_horiz_ps_%1x%2, 4, 7, 8
+    mov                         r5d,               r5m
+    mov                         r4d,               r4m
+%ifdef PIC
+    lea                         r6,                [h_tab_LumaCoeff]
+    vpbroadcastq                m0,                [r6 + r4 * 8]
+%else
+    vpbroadcastq                m0,                [h_tab_LumaCoeff + r4 * 8]
+%endif
+    mova                        m6,                [h_tab_Lm + 32]
+    mova                        m1,                [h_tab_Lm]
+    mov                         r4d,                %2                           ;height
+    add                         r3d,               r3d
+    vbroadcasti128              m2,                [pw_1]
+    mova                        m7,                [h_interp8_hps_shuf]
+
+    ; register map
+    ; m0      - interpolate coeff
+    ; m1 , m6 - shuffle order table
+    ; m2      - pw_1
+
+
+    sub                         r0,                3
+    test                        r5d,               r5d
+    jz                          .label
+    lea                         r6,                [r1 * 3]                     ; r8 = (N / 2 - 1) * srcStride
+    sub                         r0,                r6
+    add                         r4d,                7
+
+.label:
+    lea                         r6,                 [pw_2000]
+.loop:
+    ; Row 0
+    vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb                      m4,                m3,             m6           ; row 0 (col 4 to 7)
+    pshufb                      m3,                m1                           ; shuffled based on the col order tab_Lm row 0 (col 0 to 3)
+    pmaddubsw                   m3,                m0
+    pmaddubsw                   m4,                m0
+    pmaddwd                     m3,                m2
+    pmaddwd                     m4,                m2
+    packssdw                    m3,                m4
+
+
+    vbroadcasti128              m4,                [r0 + 8]
+    pshufb                      m5,                m4,            m6            ;row 0 (col 12 to 15)
+    pshufb                      m4,                m1                           ;row 0 (col 8 to 11)
+    pmaddubsw                   m4,                m0
+    pmaddubsw                   m5,                m0
+    pmaddwd                     m4,                m2
+    pmaddwd                     m5,                m2
+    packssdw                    m4,                m5
+
+    pmaddwd                     m3,                m2
+    pmaddwd                     m4,                m2
+    packssdw                    m3,                m4
+    vpermd                      m3,                m7,               m3
+    psubw                       m3,                [r6]
+
+    movu                        [r2],              m3                          ;row 0
+
+    vbroadcasti128              m3,                [r0 + 16]
+    pshufb                      m4,                m3,             m6           ; row 0 (col 20 to 23)
+    pshufb                      m3,                m1                           ; row 0 (col 16 to 19)
+    pmaddubsw                   m3,                m0
+    pmaddubsw                   m4,                m0
+    pmaddwd                     m3,                m2
+    pmaddwd                     m4,                m2
+    packssdw                    m3,                m4
+
+    vbroadcasti128              m4,                [r0 + 24]
+    pshufb                      m5,                m4,            m6            ;row 0 (col 28 to 31)
+    pshufb                      m4,                m1                           ;row 0 (col 24 to 27)
+    pmaddubsw                   m4,                m0
+    pmaddubsw                   m5,                m0
+    pmaddwd                     m4,                m2
+    pmaddwd                     m5,                m2
+    packssdw                    m4,                m5
+
+    pmaddwd                     m3,                m2
+    pmaddwd                     m4,                m2
+    packssdw                    m3,                m4
+    vpermd                      m3,                m7,               m3
+    psubw                       m3,                [r6]
+
+    movu                        [r2 + 32],         m3                          ;row 0
+
+    add                         r0,                r1
+    add                         r2,                r3
+    dec                         r4d
+    jnz                         .loop
+    RET
+%endmacro
+
+    IPFILTER_LUMA_PS_32xN_AVX2 32 , 32
+    IPFILTER_LUMA_PS_32xN_AVX2 32 , 16
+    IPFILTER_LUMA_PS_32xN_AVX2 32 , 24
+    IPFILTER_LUMA_PS_32xN_AVX2 32 , 8
+    IPFILTER_LUMA_PS_32xN_AVX2 32 , 64
+
+INIT_YMM avx2
+cglobal interp_8tap_horiz_ps_48x64, 4, 7, 8
+    mov                         r5d,               r5m
+    mov                         r4d,               r4m
+%ifdef PIC
+    lea                         r6,                [h_tab_LumaCoeff]
+    vpbroadcastq                m0,                [r6 + r4 * 8]
+%else
+    vpbroadcastq                m0,                [h_tab_LumaCoeff + r4 * 8]
+%endif
+    mova                        m6,                [h_tab_Lm + 32]
+    mova                        m1,                [h_tab_Lm]
+    mov                         r4d,               64                           ;height
+    add                         r3d,               r3d
+    vbroadcasti128              m2,                [pw_2000]
+    mova                        m7,                [pw_1]
+
+    ; register map
+    ; m0      - interpolate coeff
+    ; m1 , m6 - shuffle order table
+    ; m2      - pw_2000
+
+    sub                         r0,                3
+    test                        r5d,               r5d
+    jz                          .label
+    lea                         r6,                [r1 * 3]                     ; r6 = (N / 2 - 1) * srcStride
+    sub                         r0,                r6                           ; r0(src)-r6
+    add                         r4d,                7                            ; blkheight += N - 1  (7 - 1 = 6 ; since the last one row not in loop)
+
+.label:
+    lea                         r6,                [h_interp8_hps_shuf]
+.loop:
+    ; Row 0
+    vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb                      m4,                m3,             m6           ; row 0 (col 4 to 7)
+    pshufb                      m3,                m1                           ; shuffled based on the col order tab_Lm row 0 (col 0 to 3)
+    pmaddubsw                   m3,                m0
+    pmaddubsw                   m4,                m0
+    pmaddwd                     m3,                m7
+    pmaddwd                     m4,                m7
+    packssdw                    m3,                m4
+
+    vbroadcasti128              m4,                [r0 + 8]
+    pshufb                      m5,                m4,             m6            ;row 0 (col 12 to 15)
+    pshufb                      m4,                m1                           ;row 0 (col 8 to 11)
+    pmaddubsw                   m4,                m0
+    pmaddubsw                   m5,                m0
+    pmaddwd                     m4,                m7
+    pmaddwd                     m5,                m7
+    packssdw                    m4,                m5
+    pmaddwd                     m3,                m7
+    pmaddwd                     m4,                m7
+    packssdw                    m3,                m4
+    mova                        m5,                [r6]
+    vpermd                      m3,                m5,             m3
+    psubw                       m3,                m2
+    movu                        [r2],              m3                          ;row 0
+
+    vbroadcasti128              m3,                [r0 + 16]
+    pshufb                      m4,                m3,             m6           ; row 0 (col 20 to 23)
+    pshufb                      m3,                m1                           ; row 0 (col 16 to 19)
+    pmaddubsw                   m3,                m0
+    pmaddubsw                   m4,                m0
+    pmaddwd                     m3,                m7
+    pmaddwd                     m4,                m7
+    packssdw                    m3,                m4
+
+    vbroadcasti128              m4,                [r0 + 24]
+    pshufb                      m5,                m4,             m6            ;row 0 (col 28 to 31)
+    pshufb                      m4,                m1                           ;row 0 (col 24 to 27)
+    pmaddubsw                   m4,                m0
+    pmaddubsw                   m5,                m0
+    pmaddwd                     m4,                m7
+    pmaddwd                     m5,                m7
+    packssdw                    m4,                m5
+    pmaddwd                     m3,                m7
+    pmaddwd                     m4,                m7
+    packssdw                    m3,                m4
+    mova                        m5,                [r6]
+    vpermd                      m3,                m5,               m3
+    psubw                       m3,                m2
+    movu                        [r2 + 32],         m3                          ;row 0
+
+    vbroadcasti128              m3,                [r0 + 32]
+    pshufb                      m4,                m3,             m6           ; row 0 (col 36 to 39)
+    pshufb                      m3,                m1                           ; row 0 (col 32 to 35)
+    pmaddubsw                   m3,                m0
+    pmaddubsw                   m4,                m0
+    pmaddwd                     m3,                m7
+    pmaddwd                     m4,                m7
+    packssdw                    m3,                m4
+
+    vbroadcasti128              m4,                [r0 + 40]
+    pshufb                      m5,                m4,            m6            ;row 0 (col 44 to 47)
+    pshufb                      m4,                m1                           ;row 0 (col 40 to 43)
+    pmaddubsw                   m4,                m0
+    pmaddubsw                   m5,                m0
+    pmaddwd                     m4,                m7
+    pmaddwd                     m5,                m7
+    packssdw                    m4,                m5
+    pmaddwd                     m3,                m7
+    pmaddwd                     m4,                m7
+    packssdw                    m3,                m4
+    mova                        m5,                [r6]
+    vpermd                      m3,                m5,               m3
+    psubw                       m3,                m2
+    movu                        [r2 + 64],         m3                          ;row 0
+
+    add                         r0,                r1
+    add                         r2,                r3
+    dec                         r4d
+    jnz                         .loop
+    RET
+
+INIT_YMM avx2
+cglobal interp_8tap_horiz_pp_24x32, 4,6,8
+    sub               r0,         3
+    mov               r4d,        r4m
+%ifdef PIC
+    lea               r5,         [h_tab_LumaCoeff]
+    vpbroadcastd      m0,         [r5 + r4 * 8]
+    vpbroadcastd      m1,         [r5 + r4 * 8 + 4]
+%else
+    vpbroadcastd      m0,         [h_tab_LumaCoeff + r4 * 8]
+    vpbroadcastd      m1,         [h_tab_LumaCoeff + r4 * 8 + 4]
+%endif
+    movu              m3,         [h_tab_Tm + 16]
+    vpbroadcastd      m7,         [pw_1]
+    lea               r5,         [h_tab_Tm]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+    ; m2 , m2  shuffle order table
+    ; m7 - pw_1
+
+    mov               r4d,        32
+.loop:
+    ; Row 0
+    vbroadcasti128    m4,         [r0]                        ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m5,         m4,     m3
+    pshufb            m4,         [r5]
+    pmaddubsw         m4,         m0
+    pmaddubsw         m5,         m1
+    paddw             m4,         m5
+    pmaddwd           m4,         m7
+
+    vbroadcasti128    m5,         [r0 + 8]
+    pshufb            m6,         m5,     m3
+    pshufb            m5,         [r5]
+    pmaddubsw         m5,         m0
+    pmaddubsw         m6,         m1
+    paddw             m5,         m6
+    pmaddwd           m5,         m7
+    packssdw          m4,         m5                          ; [17 16 15 14 07 06 05 04 13 12 11 10 03 02 01 00]
+    pmulhrsw          m4,         [pw_512]
+
+    vbroadcasti128    m2,         [r0 + 16]
+    pshufb            m5,         m2,     m3
+    pshufb            m2,         [r5]
+    pmaddubsw         m2,         m0
+    pmaddubsw         m5,         m1
+    paddw             m2,         m5
+    pmaddwd           m2,         m7
+
+    packssdw          m2,         m2
+    pmulhrsw          m2,         [pw_512]
+    packuswb          m4,         m2
+    vpermq            m4,         m4,     11011000b
+    vextracti128      xm5,        m4,     1
+    pshufd            xm4,        xm4,    11011000b
+    pshufd            xm5,        xm5,    11011000b
+
+    movu              [r2],       xm4
+    movq              [r2 + 16],  xm5
+    add               r0,         r1
+    add               r2,         r3
+    dec               r4d
+    jnz               .loop
+    RET
+
+INIT_YMM avx2
+cglobal interp_8tap_horiz_pp_12x16, 4,6,8
+    sub               r0,        3
+    mov               r4d,       r4m
+%ifdef PIC
+    lea               r5,        [h_tab_LumaCoeff]
+    vpbroadcastd      m0,        [r5 + r4 * 8]
+    vpbroadcastd      m1,        [r5 + r4 * 8 + 4]
+%else
+    vpbroadcastd      m0,         [h_tab_LumaCoeff + r4 * 8]
+    vpbroadcastd      m1,         [h_tab_LumaCoeff + r4 * 8 + 4]
+%endif
+    movu              m3,         [h_tab_Tm + 16]
+    vpbroadcastd      m7,         [pw_1]
+    lea               r5,         [h_tab_Tm]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+    ; m2 , m2  shuffle order table
+    ; m7 - pw_1
+
+    mov               r4d,        8
+.loop:
+    ; Row 0
+    vbroadcasti128    m4,         [r0]                        ;first 8 element
+    pshufb            m5,         m4,     m3
+    pshufb            m4,         [r5]
+    pmaddubsw         m4,         m0
+    pmaddubsw         m5,         m1
+    paddw             m4,         m5
+    pmaddwd           m4,         m7
+
+    vbroadcasti128    m5,         [r0 + 8]                    ; element 8 to 11
+    pshufb            m6,         m5,     m3
+    pshufb            m5,         [r5]
+    pmaddubsw         m5,         m0
+    pmaddubsw         m6,         m1
+    paddw             m5,         m6
+    pmaddwd           m5,         m7
+
+    packssdw          m4,         m5                          ; [17 16 15 14 07 06 05 04 13 12 11 10 03 02 01 00]
+    pmulhrsw          m4,         [pw_512]
+
+    ;Row 1
+    vbroadcasti128    m2,         [r0 + r1]
+    pshufb            m5,         m2,     m3
+    pshufb            m2,         [r5]
+    pmaddubsw         m2,         m0
+    pmaddubsw         m5,         m1
+    paddw             m2,         m5
+    pmaddwd           m2,         m7
+
+    vbroadcasti128    m5,         [r0 + r1 + 8]
+    pshufb            m6,         m5,     m3
+    pshufb            m5,         [r5]
+    pmaddubsw         m5,         m0
+    pmaddubsw         m6,         m1
+    paddw             m5,         m6
+    pmaddwd           m5,         m7
+
+    packssdw          m2,         m5
+    pmulhrsw          m2,         [pw_512]
+    packuswb          m4,         m2
+    vpermq            m4,         m4,     11011000b
+    vextracti128      xm5,        m4,     1
+    pshufd            xm4,        xm4,    11011000b
+    pshufd            xm5,        xm5,    11011000b
+
+    movq              [r2],       xm4
+    pextrd            [r2+8],     xm4,    2
+    movq              [r2 + r3],  xm5
+    pextrd            [r2+r3+8],  xm5,    2
+    lea               r0,         [r0 + r1 * 2]
+    lea               r2,         [r2 + r3 * 2]
+    dec               r4d
+    jnz              .loop
+    RET
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp_16xN(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+%macro IPFILTER_CHROMA_PP_16xN_AVX2 2
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_%1x%2, 4, 6, 7
+    mov               r4d,          r4m
+
+%ifdef PIC
+    lea               r5,           [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,           [r5 + r4 * 4]
+%else
+    vpbroadcastd      m0,           [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    mova              m6,           [pw_512]
+    mova              m1,           [interp4_horiz_shuf1]
+    vpbroadcastd      m2,           [pw_1]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    dec               r0
+    mov               r4d,          %2/2
+
+.loop:
+    ; Row 0
+    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+    vbroadcasti128    m4,           [r0 + 4]                    ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+    packssdw          m3,           m4
+    pmulhrsw          m3,           m6
+
+    ; Row 1
+    vbroadcasti128    m4,           [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+    vbroadcasti128    m5,           [r0 + r1 + 4]               ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m5,           m1
+    pmaddubsw         m5,           m0
+    pmaddwd           m5,           m2
+    packssdw          m4,           m5
+    pmulhrsw          m4,           m6
+
+    packuswb          m3,           m4
+    vpermq            m3,           m3,      11011000b
+
+    vextracti128      xm4,          m3,       1
+    movu              [r2],         xm3
+    movu              [r2 + r3],    xm4
+    lea               r2,           [r2 + r3 * 2]
+    lea               r0,           [r0 + r1 * 2]
+    dec               r4d
+    jnz               .loop
+    RET
+%endmacro
+
+    IPFILTER_CHROMA_PP_16xN_AVX2 16 , 8
+    IPFILTER_CHROMA_PP_16xN_AVX2 16 , 32
+    IPFILTER_CHROMA_PP_16xN_AVX2 16 , 12
+    IPFILTER_CHROMA_PP_16xN_AVX2 16 , 4
+    IPFILTER_CHROMA_PP_16xN_AVX2 16 , 64
+    IPFILTER_CHROMA_PP_16xN_AVX2 16 , 24
+
+%macro IPFILTER_LUMA_PS_64xN_AVX2 1
+INIT_YMM avx2
+cglobal interp_8tap_horiz_ps_64x%1, 4, 7, 8
+    mov                         r5d,               r5m
+    mov                         r4d,               r4m
+%ifdef PIC
+    lea                         r6,                [h_tab_LumaCoeff]
+    vpbroadcastq                m0,                [r6 + r4 * 8]
+%else
+    vpbroadcastq                m0,                [h_tab_LumaCoeff + r4 * 8]
+%endif
+    mova                        m6,                [h_tab_Lm + 32]
+    mova                        m1,                [h_tab_Lm]
+    mov                         r4d,               %1                           ;height
+    add                         r3d,               r3d
+    vbroadcasti128              m2,                [pw_1]
+    mova                        m7,                [h_interp8_hps_shuf]
+
+    ; register map
+    ; m0      - interpolate coeff
+    ; m1 , m6 - shuffle order table
+    ; m2      - pw_2000
+
+    sub                         r0,                3
+    test                        r5d,               r5d
+    jz                          .label
+    lea                         r6,                [r1 * 3]
+    sub                         r0,                r6                           ; r0(src)-r6
+    add                         r4d,               7                            ; blkheight += N - 1
+
+.label:
+    lea                         r6,                [pw_2000]
+.loop:
+    ; Row 0
+    vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb                      m4,                m3,             m6           ; row 0 (col 4 to 7)
+    pshufb                      m3,                m1                           ; shuffled based on the col order tab_Lm row 0 (col 0 to 3)
+    pmaddubsw                   m3,                m0
+    pmaddubsw                   m4,                m0
+    pmaddwd                     m3,                m2
+    pmaddwd                     m4,                m2
+    packssdw                    m3,                m4
+
+    vbroadcasti128              m4,                [r0 + 8]
+    pshufb                      m5,                m4,            m6            ;row 0 (col 12 to 15)
+    pshufb                      m4,                m1                           ;row 0 (col 8 to 11)
+    pmaddubsw                   m4,                m0
+    pmaddubsw                   m5,                m0
+    pmaddwd                     m4,                m2
+    pmaddwd                     m5,                m2
+    packssdw                    m4,                m5
+    pmaddwd                     m3,                m2
+    pmaddwd                     m4,                m2
+    packssdw                    m3,                m4
+    vpermd                      m3,                m7,               m3
+    psubw                       m3,                [r6]
+    movu                        [r2],              m3                          ;row 0
+
+    vbroadcasti128              m3,                [r0 + 16]
+    pshufb                      m4,                m3,             m6           ; row 0 (col 20 to 23)
+    pshufb                      m3,                m1                           ; row 0 (col 16 to 19)
+    pmaddubsw                   m3,                m0
+    pmaddubsw                   m4,                m0
+    pmaddwd                     m3,                m2
+    pmaddwd                     m4,                m2
+    packssdw                    m3,                m4
+
+    vbroadcasti128              m4,                [r0 + 24]
+    pshufb                      m5,                m4,            m6            ;row 0 (col 28 to 31)
+    pshufb                      m4,                m1                           ;row 0 (col 24 to 27)
+    pmaddubsw                   m4,                m0
+    pmaddubsw                   m5,                m0
+    pmaddwd                     m4,                m2
+    pmaddwd                     m5,                m2
+    packssdw                    m4,                m5
+    pmaddwd                     m3,                m2
+    pmaddwd                     m4,                m2
+    packssdw                    m3,                m4
+    vpermd                      m3,                m7,               m3
+    psubw                       m3,                [r6]
+    movu                        [r2 + 32],         m3                          ;row 0
+
+    vbroadcasti128              m3,                [r0 + 32]
+    pshufb                      m4,                m3,             m6           ; row 0 (col 36 to 39)
+    pshufb                      m3,                m1                           ; row 0 (col 32 to 35)
+    pmaddubsw                   m3,                m0
+    pmaddubsw                   m4,                m0
+    pmaddwd                     m3,                m2
+    pmaddwd                     m4,                m2
+    packssdw                    m3,                m4
+
+    vbroadcasti128              m4,                [r0 + 40]
+    pshufb                      m5,                m4,            m6            ;row 0 (col 44 to 47)
+    pshufb                      m4,                m1                           ;row 0 (col 40 to 43)
+    pmaddubsw                   m4,                m0
+    pmaddubsw                   m5,                m0
+    pmaddwd                     m4,                m2
+    pmaddwd                     m5,                m2
+    packssdw                    m4,                m5
+    pmaddwd                     m3,                m2
+    pmaddwd                     m4,                m2
+    packssdw                    m3,                m4
+    vpermd                      m3,                m7,               m3
+    psubw                       m3,                [r6]
+    movu                        [r2 + 64],         m3                          ;row 0
+    vbroadcasti128              m3,                [r0 + 48]
+    pshufb                      m4,                m3,             m6           ; row 0 (col 52 to 55)
+    pshufb                      m3,                m1                           ; row 0 (col 48 to 51)
+    pmaddubsw                   m3,                m0
+    pmaddubsw                   m4,                m0
+    pmaddwd                     m3,                m2
+    pmaddwd                     m4,                m2
+    packssdw                    m3,                m4
+
+    vbroadcasti128              m4,                [r0 + 56]
+    pshufb                      m5,                m4,            m6            ;row 0 (col 60 to 63)
+    pshufb                      m4,                m1                           ;row 0 (col 56 to 59)
+    pmaddubsw                   m4,                m0
+    pmaddubsw                   m5,                m0
+    pmaddwd                     m4,                m2
+    pmaddwd                     m5,                m2
+    packssdw                    m4,                m5
+    pmaddwd                     m3,                m2
+    pmaddwd                     m4,                m2
+    packssdw                    m3,                m4
+    vpermd                      m3,                m7,               m3
+    psubw                       m3,                [r6]
+    movu                        [r2 + 96],         m3                          ;row 0
+
+    add                          r0,                r1
+    add                          r2,                r3
+    dec                          r4d
+    jnz                         .loop
+    RET
+%endmacro
+
+    IPFILTER_LUMA_PS_64xN_AVX2 64
+    IPFILTER_LUMA_PS_64xN_AVX2 48
+    IPFILTER_LUMA_PS_64xN_AVX2 32
+    IPFILTER_LUMA_PS_64xN_AVX2 16
+
+;-----------------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_ps_8xN(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;-----------------------------------------------------------------------------------------------------------------------------
+%macro IPFILTER_CHROMA_PS_8xN_AVX2 1
+INIT_YMM avx2
+cglobal interp_4tap_horiz_ps_8x%1, 4,7,6
+    mov                r4d,             r4m
+    mov                r5d,             r5m
+    add                r3d,             r3d
+
+%ifdef PIC
+    lea                r6,              [h_tab_ChromaCoeff]
+    vpbroadcastd       m0,              [r6 + r4 * 4]
+%else
+    vpbroadcastd       m0,              [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    vbroadcasti128     m2,              [pw_1]
+    vbroadcasti128     m5,              [pw_2000]
+    mova               m1,              [h_tab_Tm]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    mov                r6d,             %1/2
+    dec                r0
+    test               r5d,             r5d
+    jz                 .loop
+    sub                r0 ,             r1
+    inc                r6d
+
+.loop:
+    ; Row 0
+    vbroadcasti128     m3,              [r0]
+    pshufb             m3,              m1
+    pmaddubsw          m3,              m0
+    pmaddwd            m3,              m2
+
+    ; Row 1
+    vbroadcasti128     m4,              [r0 + r1]
+    pshufb             m4,              m1
+    pmaddubsw          m4,              m0
+    pmaddwd            m4,              m2
+    packssdw           m3,              m4
+    psubw              m3,              m5
+    vpermq             m3,              m3,          11011000b
+    vextracti128       xm4,             m3,          1
+    movu               [r2],            xm3
+    movu               [r2 + r3],       xm4
+
+    lea                r2,              [r2 + r3 * 2]
+    lea                r0,              [r0 + r1 * 2]
+    dec                r6d
+    jnz                .loop
+    test               r5d,             r5d
+    jz                 .end
+
+    ;Row 11
+    vbroadcasti128     m3,              [r0]
+    pshufb             m3,              m1
+    pmaddubsw          m3,              m0
+    pmaddwd            m3,              m2
+    packssdw           m3,              m3
+    psubw              m3,              m5
+    vpermq             m3,              m3,          11011000b
+    movu               [r2],            xm3
+.end:
+    RET
+%endmacro
+
+    IPFILTER_CHROMA_PS_8xN_AVX2  2
+    IPFILTER_CHROMA_PS_8xN_AVX2  32
+    IPFILTER_CHROMA_PS_8xN_AVX2  16
+    IPFILTER_CHROMA_PS_8xN_AVX2  6
+    IPFILTER_CHROMA_PS_8xN_AVX2  4
+    IPFILTER_CHROMA_PS_8xN_AVX2  12
+    IPFILTER_CHROMA_PS_8xN_AVX2  64
+
+INIT_YMM avx2
+cglobal interp_4tap_horiz_ps_2x4, 4, 7, 3
+    mov                r4d,            r4m
+    mov                r5d,            r5m
+    add                r3d,            r3d
+%ifdef PIC
+    lea                r6,             [h_tab_ChromaCoeff]
+    vpbroadcastd       m0,             [r6 + r4 * 4]
+%else
+    vpbroadcastd       m0,             [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    mova               xm3,            [pw_2000]
+    dec                r0
+    test               r5d,            r5d
+    jz                 .label
+    sub                r0,             r1
+
+.label:
+    lea                r6,             [r1 * 3]
+    movq               xm1,            [r0]
+    movhps             xm1,            [r0 + r1]
+    movq               xm2,            [r0 + r1 * 2]
+    movhps             xm2,            [r0 + r6]
+
+    vinserti128        m1,             m1,          xm2,          1
+    pshufb             m1,             [interp4_hpp_shuf]
+    pmaddubsw          m1,             m0
+    pmaddwd            m1,             [pw_1]
+    vextracti128       xm2,            m1,          1
+    packssdw           xm1,            xm2
+    psubw              xm1,            xm3
+
+    lea                r4,             [r3 * 3]
+    movd               [r2],           xm1
+    pextrd             [r2 + r3],      xm1,         1
+    pextrd             [r2 + r3 * 2],  xm1,         2
+    pextrd             [r2 + r4],      xm1,         3
+
+    test               r5d,            r5d
+    jz                .end
+    lea                r2,             [r2 + r3 * 4]
+    lea                r0,             [r0 + r1 * 4]
+
+    movq               xm1,            [r0]
+    movhps             xm1,            [r0 + r1]
+    movq               xm2,            [r0 + r1 * 2]
+    vinserti128        m1,             m1,          xm2,          1
+    pshufb             m1,             [interp4_hpp_shuf]
+    pmaddubsw          m1,             m0
+    pmaddwd            m1,             [pw_1]
+    vextracti128       xm2,            m1,          1
+    packssdw           xm1,            xm2
+    psubw              xm1,            xm3
+
+    movd               [r2],           xm1
+    pextrd             [r2 + r3],      xm1,         1
+    pextrd             [r2 + r3 * 2],  xm1,         2
+.end:
+    RET
+
+INIT_YMM avx2
+cglobal interp_4tap_horiz_ps_2x8, 4, 7, 7
+    mov               r4d,           r4m
+    mov               r5d,           r5m
+    add               r3d,           r3d
+
+%ifdef PIC
+    lea               r6,            [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,            [r6 + r4 * 4]
+%else
+    vpbroadcastd      m0,            [h_tab_ChromaCoeff + r4 * 4]
+%endif
+    vbroadcasti128    m6,            [pw_2000]
+    test              r5d,            r5d
+    jz                .label
+    sub               r0,             r1
+
+.label:
+    mova              m4,            [interp4_hpp_shuf]
+    mova              m5,            [pw_1]
+    dec               r0
+    lea               r4,            [r1 * 3]
+    movq              xm1,           [r0]                                   ;row 0
+    movhps            xm1,           [r0 + r1]
+    movq              xm2,           [r0 + r1 * 2]
+    movhps            xm2,           [r0 + r4]
+    vinserti128       m1,            m1,          xm2,          1
+    lea               r0,            [r0 + r1 * 4]
+    movq              xm3,           [r0]
+    movhps            xm3,           [r0 + r1]
+    movq              xm2,           [r0 + r1 * 2]
+    movhps            xm2,           [r0 + r4]
+    vinserti128       m3,            m3,          xm2,          1
+
+    pshufb            m1,            m4
+    pshufb            m3,            m4
+    pmaddubsw         m1,            m0
+    pmaddubsw         m3,            m0
+    pmaddwd           m1,            m5
+    pmaddwd           m3,            m5
+    packssdw          m1,            m3
+    psubw             m1,            m6
+
+    lea               r4,            [r3 * 3]
+    vextracti128      xm2,           m1,          1
+
+    movd              [r2],          xm1
+    pextrd            [r2 + r3],     xm1,         1
+    movd              [r2 + r3 * 2], xm2
+    pextrd            [r2 + r4],     xm2,         1
+    lea               r2,            [r2 + r3 * 4]
+    pextrd            [r2],          xm1,         2
+    pextrd            [r2 + r3],     xm1,         3
+    pextrd            [r2 + r3 * 2], xm2,         2
+    pextrd            [r2 + r4],     xm2,         3
+    test              r5d,            r5d
+    jz                .end
+
+    lea               r0,            [r0 + r1 * 4]
+    lea               r2,            [r2 + r3 * 4]
+    movq              xm1,           [r0]                                   ;row 0
+    movhps            xm1,           [r0 + r1]
+    movq              xm2,           [r0 + r1 * 2]
+    vinserti128       m1,            m1,          xm2,          1
+    pshufb            m1,            m4
+    pmaddubsw         m1,            m0
+    pmaddwd           m1,            m5
+    packssdw          m1,            m1
+    psubw             m1,            m6
+    vextracti128      xm2,           m1,          1
+
+    movd              [r2],          xm1
+    pextrd            [r2 + r3],     xm1,         1
+    movd              [r2 + r3 * 2], xm2
+.end:
+    RET
+
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_12x16, 4, 6, 7
+    mov               r4d,          r4m
+
+%ifdef PIC
+    lea               r5,           [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,           [r5 + r4 * 4]
+%else
+    vpbroadcastd      m0,           [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    mova              m6,           [pw_512]
+    mova              m1,           [interp4_horiz_shuf1]
+    vpbroadcastd      m2,           [pw_1]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    dec               r0
+    mov               r4d,          8
+
+.loop:
+    ; Row 0
+    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+    vbroadcasti128    m4,           [r0 + 4]                    ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+    packssdw          m3,           m4
+    pmulhrsw          m3,           m6
+
+    ; Row 1
+    vbroadcasti128    m4,           [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+    vbroadcasti128    m5,           [r0 + r1 + 4]               ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m5,           m1
+    pmaddubsw         m5,           m0
+    pmaddwd           m5,           m2
+    packssdw          m4,           m5
+    pmulhrsw          m4,           m6
+
+    packuswb          m3,           m4
+    vpermq            m3,           m3,      11011000b
+
+    vextracti128      xm4,          m3,       1
+    movq              [r2],         xm3
+    pextrd            [r2+8],       xm3,      2
+    movq              [r2 + r3],    xm4
+    pextrd            [r2 + r3 + 8],xm4,      2
+    lea               r2,           [r2 + r3 * 2]
+    lea               r0,           [r0 + r1 * 2]
+    dec               r4d
+    jnz               .loop
+    RET
+
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_24x32, 4,6,7
+    mov              r4d,           r4m
+
+%ifdef PIC
+    lea               r5,           [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,           [r5 + r4 * 4]
+%else
+    vpbroadcastd      m0,           [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    mova              m1,           [interp4_horiz_shuf1]
+    vpbroadcastd      m2,           [pw_1]
+    mova              m6,           [pw_512]
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    dec               r0
+    mov               r4d,          32
+
+.loop:
+    ; Row 0
+    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+    vbroadcasti128    m4,           [r0 + 4]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+    packssdw          m3,           m4
+    pmulhrsw          m3,           m6
+
+    vbroadcasti128    m4,           [r0 + 16]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+    vbroadcasti128    m5,           [r0 + 20]
+    pshufb            m5,           m1
+    pmaddubsw         m5,           m0
+    pmaddwd           m5,           m2
+    packssdw          m4,           m5
+    pmulhrsw          m4,           m6
+
+    packuswb          m3,           m4
+    vpermq            m3,           m3,      11011000b
+
+    vextracti128      xm4,          m3,       1
+    movu              [r2],         xm3
+    movq              [r2 + 16],    xm4
+    add               r2,           r3
+    add               r0,           r1
+    dec               r4d
+    jnz               .loop
+    RET
+
+;-----------------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_ps_6x8(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;-----------------------------------------------------------------------------------------------------------------------------;
+INIT_YMM avx2
+cglobal interp_4tap_horiz_ps_6x8, 4,7,6
+    mov                r4d,            r4m
+    mov                r5d,            r5m
+    add                r3d,            r3d
+
+%ifdef PIC
+    lea                r6,             [h_tab_ChromaCoeff]
+    vpbroadcastd       m0,             [r6 + r4 * 4]
+%else
+    vpbroadcastd       m0,             [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    vbroadcasti128     m2,             [pw_1]
+    vbroadcasti128     m5,             [pw_2000]
+    mova               m1,             [h_tab_Tm]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    mov               r6d,             8/2
+    dec               r0
+    test              r5d,             r5d
+    jz                .loop
+    sub               r0 ,             r1
+    inc               r6d
+
+.loop:
+    ; Row 0
+    vbroadcasti128    m3,              [r0]
+    pshufb            m3,              m1
+    pmaddubsw         m3,              m0
+    pmaddwd           m3,              m2
+
+    ; Row 1
+    vbroadcasti128    m4,              [r0 + r1]
+    pshufb            m4,              m1
+    pmaddubsw         m4,              m0
+    pmaddwd           m4,              m2
+    packssdw          m3,              m4
+    psubw             m3,              m5
+    vpermq            m3,              m3,          11011000b
+    vextracti128      xm4,             m3,          1
+    movq              [r2],            xm3
+    pextrd            [r2 + 8],        xm3,         2
+    movq              [r2 + r3],       xm4
+    pextrd            [r2 + r3 + 8],   xm4,         2
+    lea               r2,              [r2 + r3 * 2]
+    lea               r0,              [r0 + r1 * 2]
+    dec               r6d
+    jnz              .loop
+    test              r5d,             r5d
+    jz               .end
+
+    ;Row 11
+    vbroadcasti128    m3,              [r0]
+    pshufb            m3,              m1
+    pmaddubsw         m3,              m0
+    pmaddwd           m3,              m2
+    packssdw          m3,              m3
+    psubw             m3,              m5
+    vextracti128      xm4,             m3,          1
+    movq              [r2],            xm3
+    movd              [r2+8],          xm4
+.end:
+    RET
+
+INIT_YMM avx2
+cglobal interp_8tap_horiz_ps_12x16, 6, 7, 8
+    mov                         r5d,               r5m
+    mov                         r4d,               r4m
+%ifdef PIC
+    lea                         r6,                [h_tab_LumaCoeff]
+    vpbroadcastq                m0,                [r6 + r4 * 8]
+%else
+    vpbroadcastq                m0,                [h_tab_LumaCoeff + r4 * 8]
+%endif
+    mova                        m6,                [h_tab_Lm + 32]
+    mova                        m1,                [h_tab_Lm]
+    add                         r3d,               r3d
+    vbroadcasti128              m2,                [pw_2000]
+    mov                         r4d,                16
+    vbroadcasti128              m7,                [pw_1]
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - pw_2000
+
+    mova                        m5,                [h_interp8_hps_shuf]
+    sub                         r0,                3
+    test                        r5d,               r5d
+    jz                          .loop
+    lea                         r6,                [r1 * 3]                     ; r6 = (N / 2 - 1) * srcStride
+    sub                         r0,                r6                           ; r0(src)-r6
+    add                         r4d,                7
+.loop:
+
+    ; Row 0
+
+    vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb                      m4,                m3,        m6
+    pshufb                      m3,                m1                           ; shuffled based on the col order tab_Lm
+    pmaddubsw                   m3,                m0
+    pmaddubsw                   m4,                m0
+    pmaddwd                     m3,                m7
+    pmaddwd                     m4,                m7
+    packssdw                    m3,                m4
+
+    vbroadcasti128              m4,                [r0 + 8]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb                      m4,                m1
+    pmaddubsw                   m4,                m0
+    pmaddwd                     m4,                m7
+    packssdw                    m4,                m4
+
+    pmaddwd                     m3,                m7
+    pmaddwd                     m4,                m7
+    packssdw                    m3,                m4
+
+    vpermd                      m3,                m5,               m3
+    psubw                       m3,                m2
+
+    vextracti128                xm4,               m3,               1
+    movu                        [r2],              xm3                          ;row 0
+    movq                        [r2 + 16],         xm4                          ;row 1
+
+    add                         r0,                r1
+    add                         r2,                r3
+    dec                         r4d
+    jnz                         .loop
+    RET
+
+INIT_YMM avx2
+cglobal interp_8tap_horiz_ps_24x32, 4, 7, 8
+    mov                         r5d,               r5m
+    mov                         r4d,               r4m
+%ifdef PIC
+    lea                         r6,                [h_tab_LumaCoeff]
+    vpbroadcastq                m0,                [r6 + r4 * 8]
+%else
+    vpbroadcastq                m0,                [h_tab_LumaCoeff + r4 * 8]
+%endif
+    mova                        m6,                [h_tab_Lm + 32]
+    mova                        m1,                [h_tab_Lm]
+    mov                         r4d,               32                           ;height
+    add                         r3d,               r3d
+    vbroadcasti128              m2,                [pw_2000]
+    vbroadcasti128              m7,                [pw_1]
+
+    ; register map
+    ; m0      - interpolate coeff
+    ; m1 , m6 - shuffle order table
+    ; m2      - pw_2000
+
+    sub                         r0,                3
+    test                        r5d,               r5d
+    jz                          .label
+    lea                         r6,                [r1 * 3]                     ; r6 = (N / 2 - 1) * srcStride
+    sub                         r0,                r6                           ; r0(src)-r6
+    add                         r4d,               7                            ; blkheight += N - 1  (7 - 1 = 6 ; since the last one row not in loop)
+
+.label:
+    lea                         r6,                [h_interp8_hps_shuf]
+.loop:
+    ; Row 0
+    vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb                      m4,                m3,             m6           ; row 0 (col 4 to 7)
+    pshufb                      m3,                m1                           ; shuffled based on the col order tab_Lm row 0 (col 0 to 3)
+    pmaddubsw                   m3,                m0
+    pmaddubsw                   m4,                m0
+    pmaddwd                     m3,                m7
+    pmaddwd                     m4,                m7
+    packssdw                    m3,                m4
+
+    vbroadcasti128              m4,                [r0 + 8]                     ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb                      m5,                m4,            m6            ;row 1 (col 4 to 7)
+    pshufb                      m4,                m1                           ;row 1 (col 0 to 3)
+    pmaddubsw                   m4,                m0
+    pmaddubsw                   m5,                m0
+    pmaddwd                     m4,                m7
+    pmaddwd                     m5,                m7
+    packssdw                    m4,                m5
+    pmaddwd                     m3,                m7
+    pmaddwd                     m4,                m7
+    packssdw                    m3,                m4
+    mova                        m5,                [r6]
+    vpermd                      m3,                m5,               m3
+    psubw                       m3,                m2
+    movu                        [r2],              m3                          ;row 0
+
+    vbroadcasti128              m3,                [r0 + 16]
+    pshufb                      m4,                m3,          m6
+    pshufb                      m3,                m1
+    pmaddubsw                   m3,                m0
+    pmaddubsw                   m4,                m0
+    pmaddwd                     m3,                m7
+    pmaddwd                     m4,                m7
+    packssdw                    m3,                m4
+    pmaddwd                     m3,                m7
+    pmaddwd                     m4,                m7
+    packssdw                    m3,                m4
+    mova                        m4,                [r6]
+    vpermd                      m3,                m4,            m3
+    psubw                       m3,                m2
+    movu                        [r2 + 32],         xm3                          ;row 0
+
+    add                         r0,                r1
+    add                         r2,                r3
+    dec                         r4d
+    jnz                         .loop
+    RET
+
+;-----------------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_ps_24x32(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;-----------------------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+cglobal interp_4tap_horiz_ps_24x32, 4,7,6
+    mov                r4d,            r4m
+    mov                r5d,            r5m
+    add                r3d,            r3d
+%ifdef PIC
+    lea                r6,             [h_tab_ChromaCoeff]
+    vpbroadcastd       m0,             [r6 + r4 * 4]
+%else
+    vpbroadcastd       m0,             [h_tab_ChromaCoeff + r4 * 4]
+%endif
+    vbroadcasti128     m2,             [pw_1]
+    vbroadcasti128     m5,             [pw_2000]
+    mova               m1,             [h_tab_Tm]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+    mov                r6d,            32
+    dec                r0
+    test               r5d,            r5d
+    je                 .loop
+    sub                r0 ,            r1
+    add                r6d ,           3
+
+.loop:
+    ; Row 0
+    vbroadcasti128     m3,             [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb             m3,             m1
+    pmaddubsw          m3,             m0
+    pmaddwd            m3,             m2
+    vbroadcasti128     m4,             [r0 + 8]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb             m4,             m1
+    pmaddubsw          m4,             m0
+    pmaddwd            m4,             m2
+    packssdw           m3,             m4
+    psubw              m3,             m5
+    vpermq             m3,             m3,          11011000b
+    movu               [r2],           m3
+
+    vbroadcasti128     m3,             [r0 + 16]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb             m3,             m1
+    pmaddubsw          m3,             m0
+    pmaddwd            m3,             m2
+    packssdw           m3,             m3
+    psubw              m3,             m5
+    vpermq             m3,             m3,          11011000b
+    movu               [r2 + 32],      xm3
+
+    add                r2,             r3
+    add                r0,             r1
+    dec                r6d
+    jnz                .loop
+    RET
+
+;-----------------------------------------------------------------------------------------------------------------------
+;macro FILTER_H8_W8_16N_AVX2
+;-----------------------------------------------------------------------------------------------------------------------
+%macro  FILTER_H8_W8_16N_AVX2 0
+    vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb                      m4,                m3,             m6           ; row 0 (col 4 to 7)
+    pshufb                      m3,                m1                           ; shuffled based on the col order tab_Lm row 0 (col 0 to 3)
+    pmaddubsw                   m3,                m0
+    pmaddubsw                   m4,                m0
+    pmaddwd                     m3,                m2
+    pmaddwd                     m4,                m2
+    packssdw                    m3,                m4                         ; DWORD [R1D R1C R0D R0C R1B R1A R0B R0A]
+
+    vbroadcasti128              m4,                [r0 + 8]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb                      m5,                m4,            m6            ;row 1 (col 4 to 7)
+    pshufb                      m4,                m1                           ;row 1 (col 0 to 3)
+    pmaddubsw                   m4,                m0
+    pmaddubsw                   m5,                m0
+    pmaddwd                     m4,                m2
+    pmaddwd                     m5,                m2
+    packssdw                    m4,                m5                         ; DWORD [R3D R3C R2D R2C R3B R3A R2B R2A]
+
+    pmaddwd                     m3,                m2
+    pmaddwd                     m4,                m2
+    packssdw                    m3,                m4                         ; all rows and col completed.
+
+    mova                        m5,                [h_interp8_hps_shuf]
+    vpermd                      m3,                m5,               m3
+    psubw                       m3,                m8
+
+    vextracti128                xm4,               m3,               1
+    mova                        [r4],              xm3
+    mova                        [r4 + 16],         xm4
+    %endmacro
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp_64xN(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+%macro IPFILTER_CHROMA_PP_64xN_AVX2 1
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_64x%1, 4,6,7
+    mov             r4d, r4m
+
+%ifdef PIC
+    lea               r5,           [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,           [r5 + r4 * 4]
+%else
+    vpbroadcastd      m0,           [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    mova              m1,           [interp4_horiz_shuf1]
+    vpbroadcastd      m2,           [pw_1]
+    mova              m6,           [pw_512]
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    dec               r0
+    mov               r4d,          %1
+
+.loop:
+    ; Row 0
+    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+    vbroadcasti128    m4,           [r0 + 4]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+    packssdw          m3,           m4
+    pmulhrsw          m3,           m6
+
+    vbroadcasti128    m4,           [r0 + 16]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+    vbroadcasti128    m5,           [r0 + 20]
+    pshufb            m5,           m1
+    pmaddubsw         m5,           m0
+    pmaddwd           m5,           m2
+    packssdw          m4,           m5
+    pmulhrsw          m4,           m6
+    packuswb          m3,           m4
+    vpermq            m3,           m3,      11011000b
+    movu              [r2],         m3
+
+    vbroadcasti128    m3,           [r0 + 32]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+    vbroadcasti128    m4,           [r0 + 36]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+    packssdw          m3,           m4
+    pmulhrsw          m3,           m6
+
+    vbroadcasti128    m4,           [r0 + 48]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+    vbroadcasti128    m5,           [r0 + 52]
+    pshufb            m5,           m1
+    pmaddubsw         m5,           m0
+    pmaddwd           m5,           m2
+    packssdw          m4,           m5
+    pmulhrsw          m4,           m6
+    packuswb          m3,           m4
+    vpermq            m3,           m3,      11011000b
+    movu              [r2 + 32],         m3
+
+    add               r2,           r3
+    add               r0,           r1
+    dec               r4d
+    jnz               .loop
+    RET
+%endmacro
+
+    IPFILTER_CHROMA_PP_64xN_AVX2  64
+    IPFILTER_CHROMA_PP_64xN_AVX2  32
+    IPFILTER_CHROMA_PP_64xN_AVX2  48
+    IPFILTER_CHROMA_PP_64xN_AVX2  16
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp_48x64(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_48x64, 4,6,7
+    mov             r4d, r4m
+
+%ifdef PIC
+    lea               r5,            [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,            [r5 + r4 * 4]
+%else
+    vpbroadcastd      m0,            [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    mova              m1,            [interp4_horiz_shuf1]
+    vpbroadcastd      m2,            [pw_1]
+    mova              m6,            [pw_512]
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    dec               r0
+    mov               r4d,           64
+
+.loop:
+    ; Row 0
+    vbroadcasti128    m3,            [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,            m1
+    pmaddubsw         m3,            m0
+    pmaddwd           m3,            m2
+    vbroadcasti128    m4,            [r0 + 4]
+    pshufb            m4,            m1
+    pmaddubsw         m4,            m0
+    pmaddwd           m4,            m2
+    packssdw          m3,            m4
+    pmulhrsw          m3,            m6
+
+    vbroadcasti128    m4,            [r0 + 16]
+    pshufb            m4,            m1
+    pmaddubsw         m4,            m0
+    pmaddwd           m4,            m2
+    vbroadcasti128    m5,            [r0 + 20]
+    pshufb            m5,            m1
+    pmaddubsw         m5,            m0
+    pmaddwd           m5,            m2
+    packssdw          m4,            m5
+    pmulhrsw          m4,            m6
+
+    packuswb          m3,            m4
+    vpermq            m3,            m3,      q3120
+
+    movu              [r2],          m3
+
+    vbroadcasti128    m3,            [r0 + mmsize]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,            m1
+    pmaddubsw         m3,            m0
+    pmaddwd           m3,            m2
+    vbroadcasti128    m4,            [r0 + mmsize + 4]
+    pshufb            m4,            m1
+    pmaddubsw         m4,            m0
+    pmaddwd           m4,            m2
+    packssdw          m3,            m4
+    pmulhrsw          m3,            m6
+
+    vbroadcasti128    m4,            [r0 + mmsize + 16]
+    pshufb            m4,            m1
+    pmaddubsw         m4,            m0
+    pmaddwd           m4,            m2
+    vbroadcasti128    m5,            [r0 + mmsize + 20]
+    pshufb            m5,            m1
+    pmaddubsw         m5,            m0
+    pmaddwd           m5,            m2
+    packssdw          m4,            m5
+    pmulhrsw          m4,            m6
+
+    packuswb          m3,            m4
+    vpermq            m3,            m3,      q3120
+    movu              [r2 + mmsize], xm3
+
+    add               r2,            r3
+    add               r0,            r1
+    dec               r4d
+    jnz               .loop
+    RET
+
+;-----------------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_ps_48x64(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;-----------------------------------------------------------------------------------------------------------------------------;
+
+INIT_YMM avx2
+cglobal interp_4tap_horiz_ps_48x64, 4,7,6
+    mov             r4d, r4m
+    mov             r5d, r5m
+    add             r3d, r3d
+
+%ifdef PIC
+    lea               r6,           [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,           [r6 + r4 * 4]
+%else
+    vpbroadcastd      m0,           [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    vbroadcasti128     m2,          [pw_1]
+    vbroadcasti128     m5,          [pw_2000]
+    mova               m1,          [h_tab_Tm]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+    mov               r6d,          64
+    dec               r0
+    test              r5d,          r5d
+    je                .loop
+    sub               r0 ,          r1
+    add               r6d ,         3
+
+.loop:
+    ; Row 0
+    vbroadcasti128    m3,           [r0]                           ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+    vbroadcasti128    m4,           [r0 + 8]                       ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+
+    packssdw          m3,           m4
+    psubw             m3,           m5
+    vpermq            m3,           m3,          q3120
+    movu              [r2],         m3
+
+    vbroadcasti128    m3,           [r0 + 16]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+    vbroadcasti128    m4,           [r0 + 24]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+
+    packssdw          m3,           m4
+    psubw             m3,           m5
+    vpermq            m3,           m3,          q3120
+    movu              [r2 + 32],    m3
+
+    vbroadcasti128    m3,           [r0 + 32]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+    vbroadcasti128    m4,           [r0 + 40]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+
+    packssdw          m3,           m4
+    psubw             m3,           m5
+    vpermq            m3,           m3,          q3120
+    movu              [r2 + 64],    m3
+
+    add               r2,          r3
+    add               r0,          r1
+    dec               r6d
+    jnz               .loop
+    RET
+
+;-----------------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_ps_24x64(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;-----------------------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+cglobal interp_4tap_horiz_ps_24x64, 4,7,6
+    mov                r4d,            r4m
+    mov                r5d,            r5m
+    add                r3d,            r3d
+%ifdef PIC
+    lea                r6,             [h_tab_ChromaCoeff]
+    vpbroadcastd       m0,             [r6 + r4 * 4]
+%else
+    vpbroadcastd       m0,             [h_tab_ChromaCoeff + r4 * 4]
+%endif
+    vbroadcasti128     m2,             [pw_1]
+    vbroadcasti128     m5,             [pw_2000]
+    mova               m1,             [h_tab_Tm]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+    mov                r6d,            64
+    dec                r0
+    test               r5d,            r5d
+    je                 .loop
+    sub                r0 ,            r1
+    add                r6d ,           3
+
+.loop:
+    ; Row 0
+    vbroadcasti128     m3,             [r0]                          ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb             m3,             m1
+    pmaddubsw          m3,             m0
+    pmaddwd            m3,             m2
+    vbroadcasti128     m4,             [r0 + 8]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb             m4,             m1
+    pmaddubsw          m4,             m0
+    pmaddwd            m4,             m2
+    packssdw           m3,             m4
+    psubw              m3,             m5
+    vpermq             m3,             m3,          q3120
+    movu               [r2],           m3
+
+    vbroadcasti128     m3,             [r0 + 16]                     ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb             m3,             m1
+    pmaddubsw          m3,             m0
+    pmaddwd            m3,             m2
+    packssdw           m3,             m3
+    psubw              m3,             m5
+    vpermq             m3,             m3,          q3120
+    movu               [r2 + 32],      xm3
+
+    add                r2,             r3
+    add                r0,             r1
+    dec                r6d
+    jnz                .loop
+    RET
+
+INIT_YMM avx2
+cglobal interp_4tap_horiz_ps_2x16, 4, 7, 7
+    mov               r4d,           r4m
+    mov               r5d,           r5m
+    add               r3d,           r3d
+
+%ifdef PIC
+    lea               r6,            [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,            [r6 + r4 * 4]
+%else
+    vpbroadcastd      m0,            [h_tab_ChromaCoeff + r4 * 4]
+%endif
+    vbroadcasti128    m6,            [pw_2000]
+    test              r5d,            r5d
+    jz                .label
+    sub               r0,             r1
+
+.label:
+    mova              m4,            [interp4_hps_shuf]
+    mova              m5,            [pw_1]
+    dec               r0
+    lea               r4,            [r1 * 3]
+    movq              xm1,           [r0]                                   ;row 0
+    movhps            xm1,           [r0 + r1]
+    movq              xm2,           [r0 + r1 * 2]
+    movhps            xm2,           [r0 + r4]
+    vinserti128       m1,            m1,           xm2,          1
+    lea               r0,            [r0 + r1 * 4]
+    movq              xm3,           [r0]
+    movhps            xm3,           [r0 + r1]
+    movq              xm2,           [r0 + r1 * 2]
+    movhps            xm2,           [r0 + r4]
+    vinserti128       m3,            m3,           xm2,          1
+
+    pshufb            m1,            m4
+    pshufb            m3,            m4
+    pmaddubsw         m1,            m0
+    pmaddubsw         m3,            m0
+    pmaddwd           m1,            m5
+    pmaddwd           m3,            m5
+    packssdw          m1,            m3
+    psubw             m1,            m6
+
+    lea               r4,            [r3 * 3]
+    vextracti128      xm2,           m1,           1
+
+    movd              [r2],          xm1
+    pextrd            [r2 + r3],     xm1,          1
+    movd              [r2 + r3 * 2], xm2
+    pextrd            [r2 + r4],     xm2,          1
+    lea               r2,            [r2 + r3 * 4]
+    pextrd            [r2],          xm1,          2
+    pextrd            [r2 + r3],     xm1,          3
+    pextrd            [r2 + r3 * 2], xm2,          2
+    pextrd            [r2 + r4],     xm2,          3
+
+    lea               r0,            [r0 + r1 * 4]
+    lea               r2,            [r2 + r3 * 4]
+    lea               r4,            [r1 * 3]
+    movq              xm1,           [r0]
+    movhps            xm1,           [r0 + r1]
+    movq              xm2,           [r0 + r1 * 2]
+    movhps            xm2,           [r0 + r4]
+    vinserti128       m1,            m1,          xm2,           1
+    lea               r0,            [r0 + r1 * 4]
+    movq              xm3,           [r0]
+    movhps            xm3,           [r0 + r1]
+    movq              xm2,           [r0 + r1 * 2]
+    movhps            xm2,           [r0 + r4]
+    vinserti128       m3,            m3,          xm2,           1
+
+    pshufb            m1,            m4
+    pshufb            m3,            m4
+    pmaddubsw         m1,            m0
+    pmaddubsw         m3,            m0
+    pmaddwd           m1,            m5
+    pmaddwd           m3,            m5
+    packssdw          m1,            m3
+    psubw             m1,            m6
+
+    lea               r4,            [r3 * 3]
+    vextracti128      xm2,           m1,           1
+
+    movd              [r2],          xm1
+    pextrd            [r2 + r3],     xm1,          1
+    movd              [r2 + r3 * 2], xm2
+    pextrd            [r2 + r4],     xm2,          1
+    lea               r2,            [r2 + r3 * 4]
+    pextrd            [r2],          xm1,          2
+    pextrd            [r2 + r3],     xm1,          3
+    pextrd            [r2 + r3 * 2], xm2,          2
+    pextrd            [r2 + r4],     xm2,          3
+
+    test              r5d,            r5d
+    jz                .end
+
+    lea               r0,            [r0 + r1 * 4]
+    lea               r2,            [r2 + r3 * 4]
+    movq              xm1,           [r0]
+    movhps            xm1,           [r0 + r1]
+    movq              xm2,           [r0 + r1 * 2]
+    vinserti128       m1,            m1,          xm2,           1
+    pshufb            m1,            m4
+    pmaddubsw         m1,            m0
+    pmaddwd           m1,            m5
+    packssdw          m1,            m1
+    psubw             m1,            m6
+    vextracti128      xm2,           m1,           1
+
+    movd              [r2],          xm1
+    pextrd            [r2 + r3],     xm1,          1
+    movd              [r2 + r3 * 2], xm2
+.end:
+    RET
+
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_6x16, 4, 6, 7
+    mov               r4d,               r4m
+
+%ifdef PIC
+    lea               r5,                [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,                [r5 + r4 * 4]
+%else
+    vpbroadcastd      m0,                [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    mova              m1,                [h_tab_Tm]
+    mova              m2,                [pw_1]
+    mova              m6,                [pw_512]
+    lea               r4,                [r1 * 3]
+    lea               r5,                [r3 * 3]
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    dec               r0
+%rep 4
+    ; Row 0
+    vbroadcasti128    m3,                [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,                m1
+    pmaddubsw         m3,                m0
+    pmaddwd           m3,                m2
+
+    ; Row 1
+    vbroadcasti128    m4,                [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,                m1
+    pmaddubsw         m4,                m0
+    pmaddwd           m4,                m2
+    packssdw          m3,                m4
+    pmulhrsw          m3,                m6
+
+    ; Row 2
+    vbroadcasti128    m4,                [r0 + r1 * 2]               ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,                m1
+    pmaddubsw         m4,                m0
+    pmaddwd           m4,                m2
+
+    ; Row 3
+    vbroadcasti128    m5,                [r0 + r4]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m5,                m1
+    pmaddubsw         m5,                m0
+    pmaddwd           m5,                m2
+    packssdw          m4,                m5
+    pmulhrsw          m4,                m6
+
+    packuswb          m3,                m4
+    vextracti128      xm4,               m3,          1
+    movd              [r2],              xm3
+    pextrw            [r2 + 4],          xm4,         0
+    pextrd            [r2 + r3],         xm3,         1
+    pextrw            [r2 + r3 + 4],     xm4,         2
+    pextrd            [r2 + r3 * 2],     xm3,         2
+    pextrw            [r2 + r3 * 2 + 4], xm4,         4
+    pextrd            [r2 + r5],         xm3,         3
+    pextrw            [r2 + r5 + 4],     xm4,         6
+    lea               r2,                [r2 + r3 * 4]
+    lea               r0,                [r0 + r1 * 4]
+%endrep
+    RET
+
+;-----------------------------------------------------------------------------
+; void interp_8tap_hv_pp_16x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int idxX, int idxY)
+;-----------------------------------------------------------------------------
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_hv_pp_16x16, 4, 10, 15, 0-31*32
+%define stk_buf1    rsp
+    mov                         r4d,               r4m
+    mov                         r5d,               r5m
+%ifdef PIC
+    lea                         r6,                [h_tab_LumaCoeff]
+    vpbroadcastq                m0,                [r6 + r4 * 8]
+%else
+    vpbroadcastq                m0,                [h_tab_LumaCoeff + r4 * 8]
+%endif
+
+    xor                         r6,                 r6
+    mov                         r4,                 rsp
+    mova                        m6,                [h_tab_Lm + 32]
+    mova                        m1,                [h_tab_Lm]
+    mov                         r8,                16                           ;height
+    vbroadcasti128              m8,                [pw_2000]
+    vbroadcasti128              m2,                [pw_1]
+    sub                         r0,                3
+    lea                         r7,                [r1 * 3]                     ; r7 = (N / 2 - 1) * srcStride
+    sub                         r0,                r7                           ; r0(src)-r7
+    add                         r8,                7
+
+.loopH:
+    FILTER_H8_W8_16N_AVX2
+    add                         r0,                r1
+    add                         r4,                32
+    inc                         r6
+    cmp                         r6,                16+7
+    jnz                        .loopH
+
+; vertical phase
+    xor                         r6,                r6
+    xor                         r1,                r1
+.loopV:
+
+;load necessary variables
+    mov                         r4d,               r5d          ;coeff here for vertical is r5m
+    shl                         r4d,               7
+    mov                         r1d,               16
+    add                         r1d,               r1d
+
+ ; load intermedia buffer
+    mov                         r0,                stk_buf1
+
+    ; register mapping
+    ; r0 - src
+    ; r5 - coeff
+    ; r6 - loop_i
+
+; load coeff table
+%ifdef PIC
+    lea                          r5,                [h_pw_LumaCoeffVer]
+    add                          r5,                r4
+%else
+    lea                          r5,                [h_pw_LumaCoeffVer + r4]
+%endif
+
+    lea                          r4,                [r1*3]
+    mova                         m14,               [h_pd_526336]
+    lea                          r6,                [r3 * 3]
+    mov                          r9d,               16 / 8
+
+.loopW:
+    PROCESS_LUMA_AVX2_W8_16R sp
+    add                          r2,                 8
+    add                          r0,                 16
+    dec                          r9d
+    jnz                          .loopW
+    RET
+%endif
+
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_12x32, 4, 6, 7
+    mov               r4d,          r4m
+
+%ifdef PIC
+    lea               r5,           [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,           [r5 + r4 * 4]
+%else
+    vpbroadcastd      m0,           [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    mova              m6,           [pw_512]
+    mova              m1,           [interp4_horiz_shuf1]
+    vpbroadcastd      m2,           [pw_1]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    dec               r0
+    mov               r4d,          16
+
+.loop:
+    ; Row 0
+    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+    vbroadcasti128    m4,           [r0 + 4]                    ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+    packssdw          m3,           m4
+    pmulhrsw          m3,           m6
+
+    ; Row 1
+    vbroadcasti128    m4,           [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+    vbroadcasti128    m5,           [r0 + r1 + 4]               ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m5,           m1
+    pmaddubsw         m5,           m0
+    pmaddwd           m5,           m2
+    packssdw          m4,           m5
+    pmulhrsw          m4,           m6
+
+    packuswb          m3,           m4
+    vpermq            m3,           m3,      11011000b
+
+    vextracti128      xm4,          m3,       1
+    movq              [r2],         xm3
+    pextrd            [r2+8],       xm3,      2
+    movq              [r2 + r3],    xm4
+    pextrd            [r2 + r3 + 8],xm4,      2
+    lea               r2,           [r2 + r3 * 2]
+    lea               r0,           [r0 + r1 * 2]
+    dec               r4d
+    jnz               .loop
+    RET
+
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_24x64, 4,6,7
+    mov              r4d,           r4m
+
+%ifdef PIC
+    lea               r5,           [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,           [r5 + r4 * 4]
+%else
+    vpbroadcastd      m0,           [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    mova              m1,           [interp4_horiz_shuf1]
+    vpbroadcastd      m2,           [pw_1]
+    mova              m6,           [pw_512]
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - constant word 1
+
+    dec               r0
+    mov               r4d,          64
+
+.loop:
+    ; Row 0
+    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    pshufb            m3,           m1
+    pmaddubsw         m3,           m0
+    pmaddwd           m3,           m2
+    vbroadcasti128    m4,           [r0 + 4]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+    packssdw          m3,           m4
+    pmulhrsw          m3,           m6
+
+    vbroadcasti128    m4,           [r0 + 16]
+    pshufb            m4,           m1
+    pmaddubsw         m4,           m0
+    pmaddwd           m4,           m2
+    vbroadcasti128    m5,           [r0 + 20]
+    pshufb            m5,           m1
+    pmaddubsw         m5,           m0
+    pmaddwd           m5,           m2
+    packssdw          m4,           m5
+    pmulhrsw          m4,           m6
+
+    packuswb          m3,           m4
+    vpermq            m3,           m3,      11011000b
+
+    vextracti128      xm4,          m3,       1
+    movu              [r2],         xm3
+    movq              [r2 + 16],    xm4
+    add               r2,           r3
+    add               r0,           r1
+    dec               r4d
+    jnz               .loop
+    RET
+
+
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_2x16, 4, 6, 6
+    mov               r4d,           r4m
+
+%ifdef PIC
+    lea               r5,            [h_tab_ChromaCoeff]
+    vpbroadcastd      m0,            [r5 + r4 * 4]
+%else
+    vpbroadcastd      m0,            [h_tab_ChromaCoeff + r4 * 4]
+%endif
+
+    mova              m4,            [interp4_hpp_shuf]
+    mova              m5,            [pw_1]
+    dec               r0
+    lea               r4,            [r1 * 3]
+    movq              xm1,           [r0]
+    movhps            xm1,           [r0 + r1]
+    movq              xm2,           [r0 + r1 * 2]
+    movhps            xm2,           [r0 + r4]
+    vinserti128       m1,            m1,          xm2,          1
+    lea               r0,            [r0 + r1 * 4]
+    movq              xm3,           [r0]
+    movhps            xm3,           [r0 + r1]
+    movq              xm2,           [r0 + r1 * 2]
+    movhps            xm2,           [r0 + r4]
+    vinserti128       m3,            m3,          xm2,          1
+
+    pshufb            m1,            m4
+    pshufb            m3,            m4
+    pmaddubsw         m1,            m0
+    pmaddubsw         m3,            m0
+    pmaddwd           m1,            m5
+    pmaddwd           m3,            m5
+    packssdw          m1,            m3
+    pmulhrsw          m1,            [pw_512]
+    vextracti128      xm2,           m1,          1
+    packuswb          xm1,           xm2
+
+    lea               r4,            [r3 * 3]
+    pextrw            [r2],          xm1,         0
+    pextrw            [r2 + r3],     xm1,         1
+    pextrw            [r2 + r3 * 2], xm1,         4
+    pextrw            [r2 + r4],     xm1,         5
+    lea               r2,            [r2 + r3 * 4]
+    pextrw            [r2],          xm1,         2
+    pextrw            [r2 + r3],     xm1,         3
+    pextrw            [r2 + r3 * 2], xm1,         6
+    pextrw            [r2 + r4],     xm1,         7
+    lea               r2,            [r2 + r3 * 4]
+    lea               r0,            [r0 + r1 * 4]
+
+    lea               r4,            [r1 * 3]
+    movq              xm1,           [r0]
+    movhps            xm1,           [r0 + r1]
+    movq              xm2,           [r0 + r1 * 2]
+    movhps            xm2,           [r0 + r4]
+    vinserti128       m1,            m1,          xm2,          1
+    lea               r0,            [r0 + r1 * 4]
+    movq              xm3,           [r0]
+    movhps            xm3,           [r0 + r1]
+    movq              xm2,           [r0 + r1 * 2]
+    movhps            xm2,           [r0 + r4]
+    vinserti128       m3,            m3,          xm2,          1
+
+    pshufb            m1,            m4
+    pshufb            m3,            m4
+    pmaddubsw         m1,            m0
+    pmaddubsw         m3,            m0
+    pmaddwd           m1,            m5
+    pmaddwd           m3,            m5
+    packssdw          m1,            m3
+    pmulhrsw          m1,            [pw_512]
+    vextracti128      xm2,           m1,          1
+    packuswb          xm1,           xm2
+
+    lea               r4,            [r3 * 3]
+    pextrw            [r2],          xm1,         0
+    pextrw            [r2 + r3],     xm1,         1
+    pextrw            [r2 + r3 * 2], xm1,         4
+    pextrw            [r2 + r4],     xm1,         5
+    lea               r2,            [r2 + r3 * 4]
+    pextrw            [r2],          xm1,         2
+    pextrw            [r2 + r3],     xm1,         3
+    pextrw            [r2 + r3 * 2], xm1,         6
+    pextrw            [r2 + r4],     xm1,         7
+    RET
diff --git a/source/common/x86/ipfilter8.asm b/source/common/x86/ipfilter8.asm
index dc3f04add0..223917b0f3 100644
--- a/source/common/x86/ipfilter8.asm
+++ b/source/common/x86/ipfilter8.asm
@@ -39,11 +39,6 @@ const interp_vert_shuf, times 2 db 0, 2, 1, 3, 2, 4, 3, 5, 4, 6, 5, 7, 6, 8, 7,
 const interp4_vpp_shuf1, dd 0, 1, 1, 2, 2, 3, 3, 4
                          dd 2, 3, 3, 4, 4, 5, 5, 6
 
-const pb_8tap_hps_0, times 2 db 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8
-                     times 2 db 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9,10
-                     times 2 db 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9,10,10,11,11,12
-                     times 2 db 6, 7, 7, 8, 8, 9, 9,10,10,11,11,12,12,13,13,14
-
 const tab_Lm,    db 0, 1, 2, 3, 4,  5,  6,  7,  1, 2, 3, 4,  5,  6,  7,  8
                  db 2, 3, 4, 5, 6,  7,  8,  9,  3, 4, 5, 6,  7,  8,  9,  10
                  db 4, 5, 6, 7, 8,  9,  10, 11, 5, 6, 7, 8,  9,  10, 11, 12
@@ -196,26 +191,6 @@ const pw_LumaCoeffVer,  times 8 dw 0, 0
                         times 8 dw 58, -10
                         times 8 dw 4, -1
 
-const pb_LumaCoeffVer,  times 16 db 0, 0
-                        times 16 db 0, 64
-                        times 16 db 0, 0
-                        times 16 db 0, 0
-
-                        times 16 db -1, 4
-                        times 16 db -10, 58
-                        times 16 db 17, -5
-                        times 16 db 1, 0
-
-                        times 16 db -1, 4
-                        times 16 db -11, 40
-                        times 16 db 40, -11
-                        times 16 db 4, -1
-
-                        times 16 db 0, 1
-                        times 16 db -5, 17
-                        times 16 db 58, -10
-                        times 16 db 4, -1
-
 const tab_LumaCoeffVer, times 8 db 0, 0
                         times 8 db 0, 64
                         times 8 db 0, 0
@@ -282,18 +257,8 @@ const tab_ChromaCoeffVer_32,    times 16 db 0, 64
 
 const tab_c_64_n64, times 8 db 64, -64
 
-const interp4_shuf, times 2 db 0, 1, 8, 9, 4, 5, 12, 13, 2, 3, 10, 11, 6, 7, 14, 15
-
-const interp4_horiz_shuf1,  db 0, 1, 2, 3, 1, 2, 3, 4, 2, 3, 4, 5, 3, 4, 5, 6
-                            db 8, 9, 10, 11, 9, 10, 11, 12, 10, 11, 12, 13, 11, 12, 13, 14
-
-const interp4_hpp_shuf,     times 2 db 0, 1, 2, 3, 1, 2, 3, 4, 8, 9, 10, 11, 9, 10, 11, 12
-
 const interp8_hps_shuf,     dd 0, 4, 1, 5, 2, 6, 3, 7
 
-ALIGN 32
-interp4_hps_shuf: times 2 db 0, 1, 2, 3, 1, 2, 3, 4, 8, 9, 10, 11, 9, 10, 11, 12
-
 SECTION .text
 
 cextern pb_128
@@ -303,462 +268,6 @@ cextern pw_512
 cextern pw_2000
 cextern pw_8192
 
-%macro FILTER_H4_w2_2_sse2 0
-    pxor        m3, m3
-    movd        m0, [srcq - 1]
-    movd        m2, [srcq]
-    punpckldq   m0, m2
-    punpcklbw   m0, m3
-    movd        m1, [srcq + srcstrideq - 1]
-    movd        m2, [srcq + srcstrideq]
-    punpckldq   m1, m2
-    punpcklbw   m1, m3
-    pmaddwd     m0, m4
-    pmaddwd     m1, m4
-    packssdw    m0, m1
-    pshuflw     m1, m0, q2301
-    pshufhw     m1, m1, q2301
-    paddw       m0, m1
-    psrld       m0, 16
-    packssdw    m0, m0
-    paddw       m0, m5
-    psraw       m0, 6
-    packuswb    m0, m0
-    movd        r4d, m0
-    mov         [dstq], r4w
-    shr         r4, 16
-    mov         [dstq + dststrideq], r4w
-%endmacro
-
-;-----------------------------------------------------------------------------
-; void interp_4tap_horiz_pp_2xN(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro FILTER_H4_W2xN_sse3 1
-INIT_XMM sse3
-cglobal interp_4tap_horiz_pp_2x%1, 4, 6, 6, src, srcstride, dst, dststride
-    mov         r4d,    r4m
-    mova        m5,     [pw_32]
-
-%ifdef PIC
-    lea         r5,     [tabw_ChromaCoeff]
-    movddup     m4,     [r5 + r4 * 8]
-%else
-    movddup     m4,     [tabw_ChromaCoeff + r4 * 8]
-%endif
-
-%assign x 1
-%rep %1/2
-    FILTER_H4_w2_2_sse2
-%if x < %1/2
-    lea         srcq,   [srcq + srcstrideq * 2]
-    lea         dstq,   [dstq + dststrideq * 2]
-%endif
-%assign x x+1
-%endrep
-
-    RET
-
-%endmacro
-
-    FILTER_H4_W2xN_sse3 4
-    FILTER_H4_W2xN_sse3 8
-    FILTER_H4_W2xN_sse3 16
-
-%macro FILTER_H4_w4_2_sse2 0
-    pxor        m5, m5
-    movd        m0, [srcq - 1]
-    movd        m6, [srcq]
-    punpckldq   m0, m6
-    punpcklbw   m0, m5
-    movd        m1, [srcq + 1]
-    movd        m6, [srcq + 2]
-    punpckldq   m1, m6
-    punpcklbw   m1, m5
-    movd        m2, [srcq + srcstrideq - 1]
-    movd        m6, [srcq + srcstrideq]
-    punpckldq   m2, m6
-    punpcklbw   m2, m5
-    movd        m3, [srcq + srcstrideq + 1]
-    movd        m6, [srcq + srcstrideq + 2]
-    punpckldq   m3, m6
-    punpcklbw   m3, m5
-    pmaddwd     m0, m4
-    pmaddwd     m1, m4
-    pmaddwd     m2, m4
-    pmaddwd     m3, m4
-    packssdw    m0, m1
-    packssdw    m2, m3
-    pshuflw     m1, m0, q2301
-    pshufhw     m1, m1, q2301
-    pshuflw     m3, m2, q2301
-    pshufhw     m3, m3, q2301
-    paddw       m0, m1
-    paddw       m2, m3
-    psrld       m0, 16
-    psrld       m2, 16
-    packssdw    m0, m2
-    paddw       m0, m7
-    psraw       m0, 6
-    packuswb    m0, m2
-    movd        [dstq], m0
-    psrldq      m0, 4
-    movd        [dstq + dststrideq], m0
-%endmacro
-
-;-----------------------------------------------------------------------------
-; void interp_4tap_horiz_pp_4x32(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro FILTER_H4_W4xN_sse3 1
-INIT_XMM sse3
-cglobal interp_4tap_horiz_pp_4x%1, 4, 6, 8, src, srcstride, dst, dststride
-    mov         r4d,    r4m
-    mova        m7,     [pw_32]
-
-%ifdef PIC
-    lea         r5,     [tabw_ChromaCoeff]
-    movddup     m4,     [r5 + r4 * 8]
-%else
-    movddup     m4,     [tabw_ChromaCoeff + r4 * 8]
-%endif
-
-%assign x 1
-%rep %1/2
-    FILTER_H4_w4_2_sse2
-%if x < %1/2
-    lea         srcq,   [srcq + srcstrideq * 2]
-    lea         dstq,   [dstq + dststrideq * 2]
-%endif
-%assign x x+1
-%endrep
-
-    RET
-
-%endmacro
-
-    FILTER_H4_W4xN_sse3 2
-    FILTER_H4_W4xN_sse3 4
-    FILTER_H4_W4xN_sse3 8
-    FILTER_H4_W4xN_sse3 16
-    FILTER_H4_W4xN_sse3 32
-
-%macro FILTER_H4_w6_sse2 0
-    pxor        m4, m4
-    movh        m0, [srcq - 1]
-    movh        m5, [srcq]
-    punpckldq   m0, m5
-    movhlps     m2, m0
-    punpcklbw   m0, m4
-    punpcklbw   m2, m4
-    movd        m1, [srcq + 1]
-    movd        m5, [srcq + 2]
-    punpckldq   m1, m5
-    punpcklbw   m1, m4
-    pmaddwd     m0, m6
-    pmaddwd     m1, m6
-    pmaddwd     m2, m6
-    packssdw    m0, m1
-    packssdw    m2, m2
-    pshuflw     m1, m0, q2301
-    pshufhw     m1, m1, q2301
-    pshuflw     m3, m2, q2301
-    paddw       m0, m1
-    paddw       m2, m3
-    psrld       m0, 16
-    psrld       m2, 16
-    packssdw    m0, m2
-    paddw       m0, m7
-    psraw       m0, 6
-    packuswb    m0, m0
-    movd        [dstq], m0
-    pextrw      r4d, m0, 2
-    mov         [dstq + 4], r4w
-%endmacro
-
-%macro FILH4W8_sse2 1
-    movh        m0, [srcq - 1 + %1]
-    movh        m5, [srcq + %1]
-    punpckldq   m0, m5
-    movhlps     m2, m0
-    punpcklbw   m0, m4
-    punpcklbw   m2, m4
-    movh        m1, [srcq + 1 + %1]
-    movh        m5, [srcq + 2 + %1]
-    punpckldq   m1, m5
-    movhlps     m3, m1
-    punpcklbw   m1, m4
-    punpcklbw   m3, m4
-    pmaddwd     m0, m6
-    pmaddwd     m1, m6
-    pmaddwd     m2, m6
-    pmaddwd     m3, m6
-    packssdw    m0, m1
-    packssdw    m2, m3
-    pshuflw     m1, m0, q2301
-    pshufhw     m1, m1, q2301
-    pshuflw     m3, m2, q2301
-    pshufhw     m3, m3, q2301
-    paddw       m0, m1
-    paddw       m2, m3
-    psrld       m0, 16
-    psrld       m2, 16
-    packssdw    m0, m2
-    paddw       m0, m7
-    psraw       m0, 6
-    packuswb    m0, m0
-    movh        [dstq + %1], m0
-%endmacro
-
-%macro FILTER_H4_w8_sse2 0
-    FILH4W8_sse2 0
-%endmacro
-
-%macro FILTER_H4_w12_sse2 0
-    FILH4W8_sse2 0
-    movd        m1, [srcq - 1 + 8]
-    movd        m3, [srcq + 8]
-    punpckldq   m1, m3
-    punpcklbw   m1, m4
-    movd        m2, [srcq + 1 + 8]
-    movd        m3, [srcq + 2 + 8]
-    punpckldq   m2, m3
-    punpcklbw   m2, m4
-    pmaddwd     m1, m6
-    pmaddwd     m2, m6
-    packssdw    m1, m2
-    pshuflw     m2, m1, q2301
-    pshufhw     m2, m2, q2301
-    paddw       m1, m2
-    psrld       m1, 16
-    packssdw    m1, m1
-    paddw       m1, m7
-    psraw       m1, 6
-    packuswb    m1, m1
-    movd        [dstq + 8], m1
-%endmacro
-
-%macro FILTER_H4_w16_sse2 0
-    FILH4W8_sse2 0
-    FILH4W8_sse2 8
-%endmacro
-
-%macro FILTER_H4_w24_sse2 0
-    FILH4W8_sse2 0
-    FILH4W8_sse2 8
-    FILH4W8_sse2 16
-%endmacro
-
-%macro FILTER_H4_w32_sse2 0
-    FILH4W8_sse2 0
-    FILH4W8_sse2 8
-    FILH4W8_sse2 16
-    FILH4W8_sse2 24
-%endmacro
-
-%macro FILTER_H4_w48_sse2 0
-    FILH4W8_sse2 0
-    FILH4W8_sse2 8
-    FILH4W8_sse2 16
-    FILH4W8_sse2 24
-    FILH4W8_sse2 32
-    FILH4W8_sse2 40
-%endmacro
-
-%macro FILTER_H4_w64_sse2 0
-    FILH4W8_sse2 0
-    FILH4W8_sse2 8
-    FILH4W8_sse2 16
-    FILH4W8_sse2 24
-    FILH4W8_sse2 32
-    FILH4W8_sse2 40
-    FILH4W8_sse2 48
-    FILH4W8_sse2 56
-%endmacro
-
-;-----------------------------------------------------------------------------
-; void interp_4tap_horiz_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro IPFILTER_CHROMA_sse3 2
-INIT_XMM sse3
-cglobal interp_4tap_horiz_pp_%1x%2, 4, 6, 8, src, srcstride, dst, dststride
-    mov         r4d,        r4m
-    mova        m7,         [pw_32]
-    pxor        m4,         m4
-
-%ifdef PIC
-    lea         r5,          [tabw_ChromaCoeff]
-    movddup     m6,       [r5 + r4 * 8]
-%else
-    movddup     m6,       [tabw_ChromaCoeff + r4 * 8]
-%endif
-
-%assign x 1
-%rep %2
-    FILTER_H4_w%1_sse2
-%if x < %2
-    add         srcq,        srcstrideq
-    add         dstq,        dststrideq
-%endif
-%assign x x+1
-%endrep
-
-    RET
-
-%endmacro
-
-    IPFILTER_CHROMA_sse3 6,   8
-    IPFILTER_CHROMA_sse3 8,   2
-    IPFILTER_CHROMA_sse3 8,   4
-    IPFILTER_CHROMA_sse3 8,   6
-    IPFILTER_CHROMA_sse3 8,   8
-    IPFILTER_CHROMA_sse3 8,  16
-    IPFILTER_CHROMA_sse3 8,  32
-    IPFILTER_CHROMA_sse3 12, 16
-
-    IPFILTER_CHROMA_sse3 6,  16
-    IPFILTER_CHROMA_sse3 8,  12
-    IPFILTER_CHROMA_sse3 8,  64
-    IPFILTER_CHROMA_sse3 12, 32
-
-    IPFILTER_CHROMA_sse3 16,  4
-    IPFILTER_CHROMA_sse3 16,  8
-    IPFILTER_CHROMA_sse3 16, 12
-    IPFILTER_CHROMA_sse3 16, 16
-    IPFILTER_CHROMA_sse3 16, 32
-    IPFILTER_CHROMA_sse3 32,  8
-    IPFILTER_CHROMA_sse3 32, 16
-    IPFILTER_CHROMA_sse3 32, 24
-    IPFILTER_CHROMA_sse3 24, 32
-    IPFILTER_CHROMA_sse3 32, 32
-
-    IPFILTER_CHROMA_sse3 16, 24
-    IPFILTER_CHROMA_sse3 16, 64
-    IPFILTER_CHROMA_sse3 32, 48
-    IPFILTER_CHROMA_sse3 24, 64
-    IPFILTER_CHROMA_sse3 32, 64
-
-    IPFILTER_CHROMA_sse3 64, 64
-    IPFILTER_CHROMA_sse3 64, 32
-    IPFILTER_CHROMA_sse3 64, 48
-    IPFILTER_CHROMA_sse3 48, 64
-    IPFILTER_CHROMA_sse3 64, 16
-
-%macro FILTER_2 2
-    movd        m3,     [srcq + %1]
-    movd        m4,     [srcq + 1 + %1]
-    punpckldq   m3,     m4
-    punpcklbw   m3,     m0
-    pmaddwd     m3,     m1
-    packssdw    m3,     m3
-    pshuflw     m4,     m3, q2301
-    paddw       m3,     m4
-    psrldq      m3,     2
-    psubw       m3,     m2
-    movd        [dstq + %2], m3
-%endmacro
-
-%macro FILTER_4 2
-    movd        m3,     [srcq + %1]
-    movd        m4,     [srcq + 1 + %1]
-    punpckldq   m3,     m4
-    punpcklbw   m3,     m0
-    pmaddwd     m3,     m1
-    movd        m4,     [srcq + 2 + %1]
-    movd        m5,     [srcq + 3 + %1]
-    punpckldq   m4,     m5
-    punpcklbw   m4,     m0
-    pmaddwd     m4,     m1
-    packssdw    m3,     m4
-    pshuflw     m4,     m3, q2301
-    pshufhw     m4,     m4, q2301
-    paddw       m3,     m4
-    psrldq      m3,     2
-    pshufd      m3,     m3,     q3120
-    psubw       m3,     m2
-    movh        [dstq + %2], m3
-%endmacro
-
-%macro FILTER_4TAP_HPS_sse3 2
-INIT_XMM sse3
-cglobal interp_4tap_horiz_ps_%1x%2, 4, 7, 6, src, srcstride, dst, dststride
-    mov         r4d,    r4m
-    add         dststrided, dststrided
-    mova        m2,     [pw_2000]
-    pxor        m0,     m0
-
-%ifdef PIC
-    lea         r6,     [tabw_ChromaCoeff]
-    movddup     m1,     [r6 + r4 * 8]
-%else
-    movddup     m1,     [tabw_ChromaCoeff + r4 * 8]
-%endif
-
-    mov        r4d,     %2
-    cmp        r5m,     byte 0
-    je         .loopH
-    sub        srcq,    srcstrideq
-    add        r4d,     3
-
-.loopH:
-%assign x -1
-%assign y 0
-%rep %1/4
-    FILTER_4 x,y
-%assign x x+4
-%assign y y+8
-%endrep
-%rep (%1 % 4)/2
-    FILTER_2 x,y
-%endrep
-    add         srcq,   srcstrideq
-    add         dstq,   dststrideq
-
-    dec         r4d
-    jnz         .loopH
-    RET
-
-%endmacro
-
-    FILTER_4TAP_HPS_sse3 2, 4
-    FILTER_4TAP_HPS_sse3 2, 8
-    FILTER_4TAP_HPS_sse3 2, 16
-    FILTER_4TAP_HPS_sse3 4, 2
-    FILTER_4TAP_HPS_sse3 4, 4
-    FILTER_4TAP_HPS_sse3 4, 8
-    FILTER_4TAP_HPS_sse3 4, 16
-    FILTER_4TAP_HPS_sse3 4, 32
-    FILTER_4TAP_HPS_sse3 6, 8
-    FILTER_4TAP_HPS_sse3 6, 16
-    FILTER_4TAP_HPS_sse3 8, 2
-    FILTER_4TAP_HPS_sse3 8, 4
-    FILTER_4TAP_HPS_sse3 8, 6
-    FILTER_4TAP_HPS_sse3 8, 8
-    FILTER_4TAP_HPS_sse3 8, 12
-    FILTER_4TAP_HPS_sse3 8, 16
-    FILTER_4TAP_HPS_sse3 8, 32
-    FILTER_4TAP_HPS_sse3 8, 64
-    FILTER_4TAP_HPS_sse3 12, 16
-    FILTER_4TAP_HPS_sse3 12, 32
-    FILTER_4TAP_HPS_sse3 16, 4
-    FILTER_4TAP_HPS_sse3 16, 8
-    FILTER_4TAP_HPS_sse3 16, 12
-    FILTER_4TAP_HPS_sse3 16, 16
-    FILTER_4TAP_HPS_sse3 16, 24
-    FILTER_4TAP_HPS_sse3 16, 32
-    FILTER_4TAP_HPS_sse3 16, 64
-    FILTER_4TAP_HPS_sse3 24, 32
-    FILTER_4TAP_HPS_sse3 24, 64
-    FILTER_4TAP_HPS_sse3 32,  8
-    FILTER_4TAP_HPS_sse3 32, 16
-    FILTER_4TAP_HPS_sse3 32, 24
-    FILTER_4TAP_HPS_sse3 32, 32
-    FILTER_4TAP_HPS_sse3 32, 48
-    FILTER_4TAP_HPS_sse3 32, 64
-    FILTER_4TAP_HPS_sse3 48, 64
-    FILTER_4TAP_HPS_sse3 64, 16
-    FILTER_4TAP_HPS_sse3 64, 32
-    FILTER_4TAP_HPS_sse3 64, 48
-    FILTER_4TAP_HPS_sse3 64, 64
-
 %macro FILTER_H8_W8_sse2 0
     movh        m1, [r0 + x - 3]
     movh        m4, [r0 + x - 2]
@@ -3330,2543 +2839,1125 @@ RET
     FILTER_PIX_TO_SHORT_sse2 64, 48
     FILTER_PIX_TO_SHORT_sse2 64, 64
 
-%macro FILTER_H4_w2_2 3
-    movh        %2, [srcq - 1]
-    pshufb      %2, %2, Tm0
-    movh        %1, [srcq + srcstrideq - 1]
-    pshufb      %1, %1, Tm0
-    punpcklqdq  %2, %1
-    pmaddubsw   %2, coef2
-    phaddw      %2, %2
-    pmulhrsw    %2, %3
+%macro FILTER_H8_W8 7-8   ; t0, t1, t2, t3, coef, c512, src, dst
+    movu        %1, %7
+    pshufb      %2, %1, [tab_Lm +  0]
+    pmaddubsw   %2, %5
+    pshufb      %3, %1, [tab_Lm + 16]
+    pmaddubsw   %3, %5
+    phaddw      %2, %3
+    pshufb      %4, %1, [tab_Lm + 32]
+    pmaddubsw   %4, %5
+    pshufb      %1, %1, [tab_Lm + 48]
+    pmaddubsw   %1, %5
+    phaddw      %4, %1
+    phaddw      %2, %4
+  %if %0 == 8
+    pmulhrsw    %2, %6
     packuswb    %2, %2
-    movd        r4d, %2
-    mov         [dstq], r4w
-    shr         r4, 16
-    mov         [dstq + dststrideq], r4w
+    movh        %8, %2
+  %endif
 %endmacro
 
-
 ;-----------------------------------------------------------------------------
-; void interp_4tap_horiz_pp_2x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+; Interpolate HV
 ;-----------------------------------------------------------------------------
-INIT_XMM sse4
-cglobal interp_4tap_horiz_pp_2x4, 4, 6, 5, src, srcstride, dst, dststride
-%define coef2       m4
-%define Tm0         m3
-%define t2          m2
-%define t1          m1
-%define t0          m0
-
-    mov         r4d,        r4m
-
-%ifdef PIC
-    lea         r5,          [tab_ChromaCoeff]
-    movd        coef2,       [r5 + r4 * 4]
-%else
-    movd        coef2,       [tab_ChromaCoeff + r4 * 4]
-%endif
+%macro FILTER_HV8_START 7 ; (t0, t1, t2, t3, t4, off_src, off_coeff) -> (t3, t5), (t4, t1), [2]
+    mova        %5, [r0 +  (%6 + 0) * 16]
+    mova        %1, [r0 +  (%6 + 1) * 16]
+    mova        %2, [r0 +  (%6 + 2) * 16]
+    punpcklwd   %3, %5, %1
+    punpckhwd   %5, %1
+    pmaddwd     %3, [r5 + (%7) * 16]   ; R3 = L[0+1] -- Row 0
+    pmaddwd     %5, [r5 + (%7) * 16]   ; R0 = H[0+1]
+    punpcklwd   %4, %1, %2
+    punpckhwd   %1, %2
+    pmaddwd     %4, [r5 + (%7) * 16]   ; R4 = L[1+2] -- Row 1
+    pmaddwd     %1, [r5 + (%7) * 16]   ; R1 = H[1+2]
+%endmacro ; FILTER_HV8_START
 
-    pshufd      coef2,       coef2,      0
-    mova        t2,          [pw_512]
-    mova        Tm0,         [tab_Tm]
+%macro FILTER_HV8_MID 10 ; (Row3, prevRow, sum0L, sum1L, sum0H, sum1H, t6, t7, off_src, off_coeff) -> [6]
+    mova        %8, [r0 +  (%9 + 0) * 16]
+    mova        %1, [r0 +  (%9 + 1) * 16]
+    punpcklwd   %7, %2, %8
+    punpckhwd   %2, %8
+    pmaddwd     %7, [r5 + %10 * 16]
+    pmaddwd     %2, [r5 + %10 * 16]
+    paddd       %3, %7              ; R3 = L[0+1+2+3] -- Row 0
+    paddd       %5, %2              ; R0 = H[0+1+2+3]
+    punpcklwd   %7, %8, %1
+    punpckhwd   %8, %1
+    pmaddwd     %7, [r5 + %10 * 16]
+    pmaddwd     %8, [r5 + %10 * 16]
+    paddd       %4, %7              ; R4 = L[1+2+3+4] -- Row 1
+    paddd       %6, %8              ; R1 = H[1+2+3+4]
+%endmacro ; FILTER_HV8_MID
 
-%rep 2
-    FILTER_H4_w2_2   t0, t1, t2
-    lea         srcq,       [srcq + srcstrideq * 2]
-    lea         dstq,       [dstq + dststrideq * 2]
-%endrep
+; Round and Saturate
+%macro FILTER_HV8_END 4 ; output in [1, 3]
+    paddd       %1, [pd_526336]
+    paddd       %2, [pd_526336]
+    paddd       %3, [pd_526336]
+    paddd       %4, [pd_526336]
+    psrad       %1, 12
+    psrad       %2, 12
+    psrad       %3, 12
+    psrad       %4, 12
+    packssdw    %1, %2
+    packssdw    %3, %4
 
-    RET
+    ; TODO: is merge better? I think this way is short dependency link
+    packuswb    %1, %3
+%endmacro ; FILTER_HV8_END
 
 ;-----------------------------------------------------------------------------
-; void interp_4tap_horiz_pp_2x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+; void interp_8tap_hv_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int idxX, int idxY)
 ;-----------------------------------------------------------------------------
-INIT_XMM sse4
-cglobal interp_4tap_horiz_pp_2x8, 4, 6, 5, src, srcstride, dst, dststride
-%define coef2       m4
-%define Tm0         m3
-%define t2          m2
-%define t1          m1
-%define t0          m0
+INIT_XMM ssse3
+cglobal interp_8tap_hv_pp_8x8, 4, 7, 8, 0-15*16
+%define coef        m7
+%define stk_buf     rsp
 
     mov         r4d,        r4m
+    mov         r5d,        r5m
 
 %ifdef PIC
-    lea         r5,          [tab_ChromaCoeff]
-    movd        coef2,       [r5 + r4 * 4]
+    lea         r6,         [tab_LumaCoeff]
+    movh        coef,       [r6 + r4 * 8]
 %else
-    movd        coef2,       [tab_ChromaCoeff + r4 * 4]
+    movh        coef,       [tab_LumaCoeff + r4 * 8]
 %endif
+    punpcklqdq  coef,       coef
 
-    pshufd      coef2,       coef2,      0
-    mova        t2,          [pw_512]
-    mova        Tm0,         [tab_Tm]
+    ; move to row -3
+    lea         r6,         [r1 + r1 * 2]
+    sub         r0,         r6
 
-%rep 4
-    FILTER_H4_w2_2   t0, t1, t2
-    lea         srcq,       [srcq + srcstrideq * 2]
-    lea         dstq,       [dstq + dststrideq * 2]
-%endrep
+    xor         r6,         r6
+    mov         r4,         rsp
 
-    RET
+.loopH:
+    FILTER_H8_W8 m0, m1, m2, m3, coef, [pw_512], [r0 - 3]
+    psubw       m1,         [pw_2000]
+    mova        [r4],       m1
 
-;-----------------------------------------------------------------------------
-; void interp_4tap_horiz_pp_2x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-INIT_XMM sse4
-cglobal interp_4tap_horiz_pp_2x16, 4, 6, 5, src, srcstride, dst, dststride
-%define coef2       m4
-%define Tm0         m3
-%define t2          m2
-%define t1          m1
-%define t0          m0
+    add         r0,         r1
+    add         r4,         16
+    inc         r6
+    cmp         r6,         8+7
+    jnz         .loopH
 
-    mov         r4d,        r4m
+    ; ready to phase V
+    ; Here all of mN is free
 
-%ifdef PIC
-    lea         r5,          [tab_ChromaCoeff]
-    movd        coef2,       [r5 + r4 * 4]
-%else
-    movd        coef2,       [tab_ChromaCoeff + r4 * 4]
-%endif
+    ; load coeff table
+    shl         r5,         6
+    lea         r6,         [tab_LumaCoeffV]
+    lea         r5,         [r5 + r6]
 
-    pshufd      coef2,       coef2,      0
-    mova        t2,          [pw_512]
-    mova        Tm0,         [tab_Tm]
+    ; load intermedia buffer
+    mov         r0,         stk_buf
 
-    mov         r5d,        16/2
+    ; register mapping
+    ; r0 - src
+    ; r5 - coeff
+    ; r6 - loop_i
 
-.loop:
-    FILTER_H4_w2_2   t0, t1, t2
-    lea         srcq,       [srcq + srcstrideq * 2]
-    lea         dstq,       [dstq + dststrideq * 2]
-    dec         r5d
-    jnz         .loop
+    ; let's go
+    xor         r6,         r6
 
-    RET
+    ; TODO: this loop have more than 70 instructions, I think it is more than Intel loop decode cache
+.loopV:
 
-%macro FILTER_H4_w4_2 3
-    movh        %2, [srcq - 1]
-    pshufb      %2, %2, Tm0
-    pmaddubsw   %2, coef2
-    movh        %1, [srcq + srcstrideq - 1]
-    pshufb      %1, %1, Tm0
-    pmaddubsw   %1, coef2
-    phaddw      %2, %1
-    pmulhrsw    %2, %3
-    packuswb    %2, %2
-    movd        [dstq], %2
-    palignr     %2, %2, 4
-    movd        [dstq + dststrideq], %2
-%endmacro
-
-;-----------------------------------------------------------------------------
-; void interp_4tap_horiz_pp_4x2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-INIT_XMM sse4
-cglobal interp_4tap_horiz_pp_4x2, 4, 6, 5, src, srcstride, dst, dststride
-%define coef2       m4
-%define Tm0         m3
-%define t2          m2
-%define t1          m1
-%define t0          m0
-
-    mov         r4d,        r4m
+    FILTER_HV8_START    m1, m2, m3, m4, m0,             0, 0
+    FILTER_HV8_MID      m6, m2, m3, m4, m0, m1, m7, m5, 3, 1
+    FILTER_HV8_MID      m5, m6, m3, m4, m0, m1, m7, m2, 5, 2
+    FILTER_HV8_MID      m6, m5, m3, m4, m0, m1, m7, m2, 7, 3
+    FILTER_HV8_END      m3, m0, m4, m1
 
-%ifdef PIC
-    lea         r5,          [tab_ChromaCoeff]
-    movd        coef2,       [r5 + r4 * 4]
-%else
-    movd        coef2,       [tab_ChromaCoeff + r4 * 4]
-%endif
+    movh        [r2],       m3
+    movhps      [r2 + r3],  m3
 
-    pshufd      coef2,       coef2,      0
-    mova        t2,          [pw_512]
-    mova        Tm0,         [tab_Tm]
+    lea         r0,         [r0 + 16 * 2]
+    lea         r2,         [r2 + r3 * 2]
 
-    FILTER_H4_w4_2   t0, t1, t2
+    inc         r6
+    cmp         r6,         8/2
+    jnz         .loopV
 
     RET
 
 ;-----------------------------------------------------------------------------
-; void interp_4tap_horiz_pp_4x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+; void interp_8tap_hv_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int idxX, int idxY)
 ;-----------------------------------------------------------------------------
-INIT_XMM sse4
-cglobal interp_4tap_horiz_pp_4x4, 4, 6, 5, src, srcstride, dst, dststride
-%define coef2       m4
-%define Tm0         m3
-%define t2          m2
-%define t1          m1
-%define t0          m0
-
+INIT_XMM sse3
+cglobal interp_8tap_hv_pp_8x8, 4, 7, 8, 0-15*16
     mov         r4d,        r4m
+    mov         r5d,        r5m
+    add         r4d,        r4d
+    pxor        m6,         m6
 
 %ifdef PIC
-    lea         r5,          [tab_ChromaCoeff]
-    movd        coef2,       [r5 + r4 * 4]
+    lea         r6,         [tabw_LumaCoeff]
+    mova        m3,         [r6 + r4 * 8]
 %else
-    movd        coef2,       [tab_ChromaCoeff + r4 * 4]
+    mova        m3,         [tabw_LumaCoeff + r4 * 8]
 %endif
 
-    pshufd      coef2,       coef2,      0
-    mova        t2,          [pw_512]
-    mova        Tm0,         [tab_Tm]
-
-%rep 2
-    FILTER_H4_w4_2   t0, t1, t2
-    lea         srcq,       [srcq + srcstrideq * 2]
-    lea         dstq,       [dstq + dststrideq * 2]
-%endrep
-
-    RET
+    ; move to row -3
+    lea         r6,         [r1 + r1 * 2]
+    sub         r0,         r6
 
-;-----------------------------------------------------------------------------
-; void interp_4tap_horiz_pp_4x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-INIT_XMM sse4
-cglobal interp_4tap_horiz_pp_4x8, 4, 6, 5, src, srcstride, dst, dststride
-%define coef2       m4
-%define Tm0         m3
-%define t2          m2
-%define t1          m1
-%define t0          m0
+    mov         r4,         rsp
 
-    mov         r4d,        r4m
+%assign x 0     ;needed for FILTER_H8_W8_sse2 macro
+%assign y 1
+%rep 15
+    FILTER_H8_W8_sse2
+    psubw       m1,         [pw_2000]
+    mova        [r4],       m1
 
-%ifdef PIC
-    lea         r5,          [tab_ChromaCoeff]
-    movd        coef2,       [r5 + r4 * 4]
-%else
-    movd        coef2,       [tab_ChromaCoeff + r4 * 4]
+%if y < 15
+    add         r0,         r1
+    add         r4,         16
 %endif
-
-    pshufd      coef2,       coef2,      0
-    mova        t2,          [pw_512]
-    mova        Tm0,         [tab_Tm]
-
-%rep 4
-    FILTER_H4_w4_2   t0, t1, t2
-    lea         srcq,       [srcq + srcstrideq * 2]
-    lea         dstq,       [dstq + dststrideq * 2]
+%assign y y+1
 %endrep
 
-    RET
+    ; ready to phase V
+    ; Here all of mN is free
 
-;-----------------------------------------------------------------------------
-; void interp_4tap_horiz_pp_4x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-INIT_XMM sse4
-cglobal interp_4tap_horiz_pp_4x16, 4, 6, 5, src, srcstride, dst, dststride
-%define coef2       m4
-%define Tm0         m3
-%define t2          m2
-%define t1          m1
-%define t0          m0
+    ; load coeff table
+    shl         r5,         6
+    lea         r6,         [tab_LumaCoeffV]
+    lea         r5,         [r5 + r6]
 
-    mov         r4d,        r4m
+    ; load intermedia buffer
+    mov         r0,         rsp
 
-%ifdef PIC
-    lea         r5,          [tab_ChromaCoeff]
-    movd        coef2,       [r5 + r4 * 4]
-%else
-    movd        coef2,       [tab_ChromaCoeff + r4 * 4]
-%endif
+    ; register mapping
+    ; r0 - src
+    ; r5 - coeff
 
-    pshufd      coef2,       coef2,      0
-    mova        t2,          [pw_512]
-    mova        Tm0,         [tab_Tm]
+    ; let's go
+%assign y 1
+%rep 4
+    FILTER_HV8_START    m1, m2, m3, m4, m0,             0, 0
+    FILTER_HV8_MID      m6, m2, m3, m4, m0, m1, m7, m5, 3, 1
+    FILTER_HV8_MID      m5, m6, m3, m4, m0, m1, m7, m2, 5, 2
+    FILTER_HV8_MID      m6, m5, m3, m4, m0, m1, m7, m2, 7, 3
+    FILTER_HV8_END      m3, m0, m4, m1
 
-%rep 8
-    FILTER_H4_w4_2   t0, t1, t2
-    lea         srcq,       [srcq + srcstrideq * 2]
-    lea         dstq,       [dstq + dststrideq * 2]
-%endrep
+    movh        [r2],       m3
+    movhps      [r2 + r3],  m3
 
+%if y < 4
+    lea         r0,         [r0 + 16 * 2]
+    lea         r2,         [r2 + r3 * 2]
+%endif
+%assign y y+1
+%endrep
     RET
 
 ;-----------------------------------------------------------------------------
-; void interp_4tap_horiz_pp_4x32(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;void interp_4tap_vert_pp_2x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
 ;-----------------------------------------------------------------------------
 INIT_XMM sse4
-cglobal interp_4tap_horiz_pp_4x32, 4, 6, 5, src, srcstride, dst, dststride
-%define coef2       m4
-%define Tm0         m3
-%define t2          m2
-%define t1          m1
-%define t0          m0
+cglobal interp_4tap_vert_pp_2x4, 4, 6, 8
 
-    mov         r4d,        r4m
+    mov         r4d,       r4m
+    sub         r0,        r1
 
 %ifdef PIC
-    lea         r5,          [tab_ChromaCoeff]
-    movd        coef2,       [r5 + r4 * 4]
+    lea         r5,        [tab_ChromaCoeff]
+    movd        m0,        [r5 + r4 * 4]
 %else
-    movd        coef2,       [tab_ChromaCoeff + r4 * 4]
+    movd        m0,        [tab_ChromaCoeff + r4 * 4]
 %endif
+    lea         r4,        [r1 * 3]
+    lea         r5,        [r0 + 4 * r1]
+    pshufb      m0,        [tab_Cm]
+    mova        m1,        [pw_512]
 
-    pshufd      coef2,       coef2,      0
-    mova        t2,          [pw_512]
-    mova        Tm0,         [tab_Tm]
+    movd        m2,        [r0]
+    movd        m3,        [r0 + r1]
+    movd        m4,        [r0 + 2 * r1]
+    movd        m5,        [r0 + r4]
 
-    mov         r5d,        32/2
+    punpcklbw   m2,        m3
+    punpcklbw   m6,        m4,        m5
+    punpcklbw   m2,        m6
 
-.loop:
-    FILTER_H4_w4_2   t0, t1, t2
-    lea         srcq,       [srcq + srcstrideq * 2]
-    lea         dstq,       [dstq + dststrideq * 2]
-    dec         r5d
-    jnz         .loop
+    pmaddubsw   m2,        m0
 
-    RET
+    movd        m6,        [r5]
 
-ALIGN 32
-const interp_4tap_8x8_horiz_shuf,   dd 0, 4, 1, 5, 2, 6, 3, 7
+    punpcklbw   m3,        m4
+    punpcklbw   m7,        m5,        m6
+    punpcklbw   m3,        m7
 
+    pmaddubsw   m3,        m0
 
-%macro FILTER_H4_w6 3
-    movu        %1, [srcq - 1]
-    pshufb      %2, %1, Tm0
-    pmaddubsw   %2, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %2, %1
-    pmulhrsw    %2, %3
-    packuswb    %2, %2
-    movd        [dstq],      %2
-    pextrw      [dstq + 4], %2, 2
-%endmacro
+    phaddw      m2,        m3
 
-%macro FILTER_H4_w8 3
-    movu        %1, [srcq - 1]
-    pshufb      %2, %1, Tm0
-    pmaddubsw   %2, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %2, %1
-    pmulhrsw    %2, %3
-    packuswb    %2, %2
-    movh        [dstq],      %2
-%endmacro
+    pmulhrsw    m2,        m1
 
-%macro FILTER_H4_w12 3
-    movu        %1, [srcq - 1]
-    pshufb      %2, %1, Tm0
-    pmaddubsw   %2, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %2, %1
-    pmulhrsw    %2, %3
-    movu        %1, [srcq - 1 + 8]
-    pshufb      %1, %1, Tm0
-    pmaddubsw   %1, coef2
-    phaddw      %1, %1
-    pmulhrsw    %1, %3
-    packuswb    %2, %1
-    movh        [dstq],      %2
-    pextrd      [dstq + 8], %2, 2
-%endmacro
+    movd        m7,        [r5 + r1]
 
-%macro FILTER_H4_w16 4
-    movu        %1, [srcq - 1]
-    pshufb      %2, %1, Tm0
-    pmaddubsw   %2, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %2, %1
-    movu        %1, [srcq - 1 + 8]
-    pshufb      %4, %1, Tm0
-    pmaddubsw   %4, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %4, %1
-    pmulhrsw    %2, %3
-    pmulhrsw    %4, %3
-    packuswb    %2, %4
-    movu        [dstq],      %2
-%endmacro
+    punpcklbw   m4,        m5
+    punpcklbw   m3,        m6,        m7
+    punpcklbw   m4,        m3
 
-%macro FILTER_H4_w24 4
-    movu        %1, [srcq - 1]
-    pshufb      %2, %1, Tm0
-    pmaddubsw   %2, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %2, %1
-    movu        %1, [srcq - 1 + 8]
-    pshufb      %4, %1, Tm0
-    pmaddubsw   %4, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %4, %1
-    pmulhrsw    %2, %3
-    pmulhrsw    %4, %3
-    packuswb    %2, %4
-    movu        [dstq],          %2
-    movu        %1, [srcq - 1 + 16]
-    pshufb      %2, %1, Tm0
-    pmaddubsw   %2, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %2, %1
-    pmulhrsw    %2, %3
-    packuswb    %2, %2
-    movh        [dstq + 16],     %2
-%endmacro
+    pmaddubsw   m4,        m0
 
-%macro FILTER_H4_w32 4
-    movu        %1, [srcq - 1]
-    pshufb      %2, %1, Tm0
-    pmaddubsw   %2, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %2, %1
-    movu        %1, [srcq - 1 + 8]
-    pshufb      %4, %1, Tm0
-    pmaddubsw   %4, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %4, %1
-    pmulhrsw    %2, %3
-    pmulhrsw    %4, %3
-    packuswb    %2, %4
-    movu        [dstq],      %2
-    movu        %1, [srcq - 1 + 16]
-    pshufb      %2, %1, Tm0
-    pmaddubsw   %2, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %2, %1
-    movu        %1, [srcq - 1 + 24]
-    pshufb      %4, %1, Tm0
-    pmaddubsw   %4, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %4, %1
-    pmulhrsw    %2, %3
-    pmulhrsw    %4, %3
-    packuswb    %2, %4
-    movu        [dstq + 16],      %2
-%endmacro
+    movd        m3,        [r5 + 2 * r1]
 
-%macro FILTER_H4_w16o 5
-    movu        %1, [srcq + %5 - 1]
-    pshufb      %2, %1, Tm0
-    pmaddubsw   %2, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %2, %1
-    movu        %1, [srcq + %5 - 1 + 8]
-    pshufb      %4, %1, Tm0
-    pmaddubsw   %4, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %4, %1
-    pmulhrsw    %2, %3
-    pmulhrsw    %4, %3
-    packuswb    %2, %4
-    movu        [dstq + %5],      %2
-%endmacro
+    punpcklbw   m5,        m6
+    punpcklbw   m7,        m3
+    punpcklbw   m5,        m7
 
-%macro FILTER_H4_w48 4
-    FILTER_H4_w16o %1, %2, %3, %4, 0
-    FILTER_H4_w16o %1, %2, %3, %4, 16
-    FILTER_H4_w16o %1, %2, %3, %4, 32
-%endmacro
+    pmaddubsw   m5,        m0
 
-%macro FILTER_H4_w64 4
-    FILTER_H4_w16o %1, %2, %3, %4, 0
-    FILTER_H4_w16o %1, %2, %3, %4, 16
-    FILTER_H4_w16o %1, %2, %3, %4, 32
-    FILTER_H4_w16o %1, %2, %3, %4, 48
-%endmacro
+    phaddw      m4,        m5
 
-;-----------------------------------------------------------------------------
-; void interp_4tap_horiz_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro IPFILTER_CHROMA 2
-INIT_XMM sse4
-cglobal interp_4tap_horiz_pp_%1x%2, 4, 6, 6, src, srcstride, dst, dststride
-%define coef2       m5
-%define Tm0         m4
-%define Tm1         m3
-%define t2          m2
-%define t1          m1
-%define t0          m0
+    pmulhrsw    m4,        m1
+    packuswb    m2,        m4
 
-    mov         r4d,        r4m
+    pextrw      [r2],      m2, 0
+    pextrw      [r2 + r3], m2, 2
+    lea         r2,        [r2 + 2 * r3]
+    pextrw      [r2],      m2, 4
+    pextrw      [r2 + r3], m2, 6
+
+    RET
+
+%macro FILTER_VER_CHROMA_AVX2_2x4 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_2x4, 4, 6, 2
+    mov             r4d, r4m
+    shl             r4d, 5
+    sub             r0, r1
 
 %ifdef PIC
-    lea         r5,          [tab_ChromaCoeff]
-    movd        coef2,       [r5 + r4 * 4]
+    lea             r5, [tab_ChromaCoeff_V]
+    add             r5, r4
 %else
-    movd        coef2,       [tab_ChromaCoeff + r4 * 4]
+    lea             r5, [tab_ChromaCoeff_V + r4]
 %endif
 
-    mov           r5d,       %2
-
-    pshufd      coef2,       coef2,      0
-    mova        t2,          [pw_512]
-    mova        Tm0,         [tab_Tm]
-    mova        Tm1,         [tab_Tm + 16]
-
-.loop:
-    FILTER_H4_w%1   t0, t1, t2
-    add         srcq,        srcstrideq
-    add         dstq,        dststrideq
+    lea             r4, [r1 * 3]
 
-    dec         r5d
-    jnz        .loop
+    pinsrw          xm1, [r0], 0
+    pinsrw          xm1, [r0 + r1], 1
+    pinsrw          xm1, [r0 + r1 * 2], 2
+    pinsrw          xm1, [r0 + r4], 3
+    lea             r0, [r0 + r1 * 4]
+    pinsrw          xm1, [r0], 4
+    pinsrw          xm1, [r0 + r1], 5
+    pinsrw          xm1, [r0 + r1 * 2], 6
 
+    pshufb          xm0, xm1, [interp_vert_shuf]
+    pshufb          xm1, [interp_vert_shuf + 32]
+    vinserti128     m0, m0, xm1, 1
+    pmaddubsw       m0, [r5]
+    vextracti128    xm1, m0, 1
+    paddw           xm0, xm1
+%ifidn %1,pp
+    pmulhrsw        xm0, [pw_512]
+    packuswb        xm0, xm0
+    lea             r4, [r3 * 3]
+    pextrw          [r2], xm0, 0
+    pextrw          [r2 + r3], xm0, 1
+    pextrw          [r2 + r3 * 2], xm0, 2
+    pextrw          [r2 + r4], xm0, 3
+%else
+    add             r3d, r3d
+    lea             r4, [r3 * 3]
+    psubw           xm0, [pw_2000]
+    movd            [r2], xm0
+    pextrd          [r2 + r3], xm0, 1
+    pextrd          [r2 + r3 * 2], xm0, 2
+    pextrd          [r2 + r4], xm0, 3
+%endif
     RET
 %endmacro
 
+    FILTER_VER_CHROMA_AVX2_2x4 pp
+    FILTER_VER_CHROMA_AVX2_2x4 ps
 
-    IPFILTER_CHROMA 6,   8
-    IPFILTER_CHROMA 8,   2
-    IPFILTER_CHROMA 8,   4
-    IPFILTER_CHROMA 8,   6
-    IPFILTER_CHROMA 8,   8
-    IPFILTER_CHROMA 8,  16
-    IPFILTER_CHROMA 8,  32
-    IPFILTER_CHROMA 12, 16
-
-    IPFILTER_CHROMA 6,  16
-    IPFILTER_CHROMA 8,  12
-    IPFILTER_CHROMA 8,  64
-    IPFILTER_CHROMA 12, 32
-
-;-----------------------------------------------------------------------------
-; void interp_4tap_horiz_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro IPFILTER_CHROMA_W 2
-INIT_XMM sse4
-cglobal interp_4tap_horiz_pp_%1x%2, 4, 6, 7, src, srcstride, dst, dststride
-%define coef2       m6
-%define Tm0         m5
-%define Tm1         m4
-%define t3          m3
-%define t2          m2
-%define t1          m1
-%define t0          m0
-
-    mov         r4d,         r4m
+%macro FILTER_VER_CHROMA_AVX2_2x8 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_2x8, 4, 6, 2
+    mov             r4d, r4m
+    shl             r4d, 6
+    sub             r0, r1
 
 %ifdef PIC
-    lea         r5,          [tab_ChromaCoeff]
-    movd        coef2,       [r5 + r4 * 4]
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
 %else
-    movd        coef2,       [tab_ChromaCoeff + r4 * 4]
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
 %endif
 
-    mov         r5d,          %2
-
-    pshufd      coef2,       coef2,      0
-    mova        t2,          [pw_512]
-    mova        Tm0,         [tab_Tm]
-    mova        Tm1,         [tab_Tm + 16]
-
-.loop:
-    FILTER_H4_w%1   t0, t1, t2, t3
-    add         srcq,        srcstrideq
-    add         dstq,        dststrideq
+    lea             r4, [r1 * 3]
 
-    dec         r5d
-    jnz        .loop
+    pinsrw          xm1, [r0], 0
+    pinsrw          xm1, [r0 + r1], 1
+    pinsrw          xm1, [r0 + r1 * 2], 2
+    pinsrw          xm1, [r0 + r4], 3
+    lea             r0, [r0 + r1 * 4]
+    pinsrw          xm1, [r0], 4
+    pinsrw          xm1, [r0 + r1], 5
+    pinsrw          xm1, [r0 + r1 * 2], 6
+    pinsrw          xm1, [r0 + r4], 7
+    movhlps         xm0, xm1
+    lea             r0, [r0 + r1 * 4]
+    pinsrw          xm0, [r0], 4
+    pinsrw          xm0, [r0 + r1], 5
+    pinsrw          xm0, [r0 + r1 * 2], 6
+    vinserti128     m1, m1, xm0, 1
 
-    RET
-%endmacro
-
-    IPFILTER_CHROMA_W 16,  4
-    IPFILTER_CHROMA_W 16,  8
-    IPFILTER_CHROMA_W 16, 12
-    IPFILTER_CHROMA_W 16, 16
-    IPFILTER_CHROMA_W 16, 32
-    IPFILTER_CHROMA_W 32,  8
-    IPFILTER_CHROMA_W 32, 16
-    IPFILTER_CHROMA_W 32, 24
-    IPFILTER_CHROMA_W 24, 32
-    IPFILTER_CHROMA_W 32, 32
-
-    IPFILTER_CHROMA_W 16, 24
-    IPFILTER_CHROMA_W 16, 64
-    IPFILTER_CHROMA_W 32, 48
-    IPFILTER_CHROMA_W 24, 64
-    IPFILTER_CHROMA_W 32, 64
-
-    IPFILTER_CHROMA_W 64, 64
-    IPFILTER_CHROMA_W 64, 32
-    IPFILTER_CHROMA_W 64, 48
-    IPFILTER_CHROMA_W 48, 64
-    IPFILTER_CHROMA_W 64, 16
-
-
-%macro FILTER_H8_W8 7-8   ; t0, t1, t2, t3, coef, c512, src, dst
-    movu        %1, %7
-    pshufb      %2, %1, [tab_Lm +  0]
-    pmaddubsw   %2, %5
-    pshufb      %3, %1, [tab_Lm + 16]
-    pmaddubsw   %3, %5
-    phaddw      %2, %3
-    pshufb      %4, %1, [tab_Lm + 32]
-    pmaddubsw   %4, %5
-    pshufb      %1, %1, [tab_Lm + 48]
-    pmaddubsw   %1, %5
-    phaddw      %4, %1
-    phaddw      %2, %4
-  %if %0 == 8
-    pmulhrsw    %2, %6
-    packuswb    %2, %2
-    movh        %8, %2
-  %endif
-%endmacro
-
-%macro FILTER_H8_W4 2
-    movu        %1, [r0 - 3 + r5]
-    pshufb      %2, %1, [tab_Lm]
-    pmaddubsw   %2, m3
-    pshufb      m7, %1, [tab_Lm + 16]
-    pmaddubsw   m7, m3
-    phaddw      %2, m7
-    phaddw      %2, %2
-%endmacro
-
-;----------------------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;----------------------------------------------------------------------------------------------------------------------------
-%macro IPFILTER_LUMA 3
-INIT_XMM sse4
-cglobal interp_8tap_horiz_%3_%1x%2, 4,7,8
-
-    mov       r4d, r4m
-
-%ifdef PIC
-    lea       r6, [tab_LumaCoeff]
-    movh      m3, [r6 + r4 * 8]
-%else
-    movh      m3, [tab_LumaCoeff + r4 * 8]
-%endif
-    punpcklqdq  m3, m3
-
-%ifidn %3, pp
-    mova      m2, [pw_512]
+    pshufb          m0, m1, [interp_vert_shuf]
+    pshufb          m1, [interp_vert_shuf + 32]
+    pmaddubsw       m0, [r5]
+    pmaddubsw       m1, [r5 + 1 * mmsize]
+    paddw           m0, m1
+%ifidn %1,pp
+    pmulhrsw        m0, [pw_512]
+    vextracti128    xm1, m0, 1
+    packuswb        xm0, xm1
+    lea             r4, [r3 * 3]
+    pextrw          [r2], xm0, 0
+    pextrw          [r2 + r3], xm0, 1
+    pextrw          [r2 + r3 * 2], xm0, 2
+    pextrw          [r2 + r4], xm0, 3
+    lea             r2, [r2 + r3 * 4]
+    pextrw          [r2], xm0, 4
+    pextrw          [r2 + r3], xm0, 5
+    pextrw          [r2 + r3 * 2], xm0, 6
+    pextrw          [r2 + r4], xm0, 7
 %else
-    mova      m2, [pw_2000]
-%endif
-
-    mov       r4d, %2
-%ifidn %3, ps
-    add       r3, r3
-    cmp       r5m, byte 0
-    je        .loopH
-    lea       r6, [r1 + 2 * r1]
-    sub       r0, r6
-    add       r4d, 7
+    add             r3d, r3d
+    lea             r4, [r3 * 3]
+    psubw           m0, [pw_2000]
+    vextracti128    xm1, m0, 1
+    movd            [r2], xm0
+    pextrd          [r2 + r3], xm0, 1
+    pextrd          [r2 + r3 * 2], xm0, 2
+    pextrd          [r2 + r4], xm0, 3
+    lea             r2, [r2 + r3 * 4]
+    movd            [r2], xm1
+    pextrd          [r2 + r3], xm1, 1
+    pextrd          [r2 + r3 * 2], xm1, 2
+    pextrd          [r2 + r4], xm1, 3
 %endif
-
-.loopH:
-    xor       r5, r5
-%rep %1 / 8
-  %ifidn %3, pp
-    FILTER_H8_W8  m0, m1, m4, m5, m3, m2, [r0 - 3 + r5], [r2 + r5]
-  %else
-    FILTER_H8_W8  m0, m1, m4, m5, m3, UNUSED, [r0 - 3 + r5]
-    psubw     m1, m2
-    movu      [r2 + 2 * r5], m1
-  %endif
-    add       r5, 8
-%endrep
-
-%rep (%1 % 8) / 4
-    FILTER_H8_W4  m0, m1
-  %ifidn %3, pp
-    pmulhrsw  m1, m2
-    packuswb  m1, m1
-    movd      [r2 + r5], m1
-  %else
-    psubw     m1, m2
-    movh      [r2 + 2 * r5], m1
-  %endif
-%endrep
-
-    add       r0, r1
-    add       r2, r3
-
-    dec       r4d
-    jnz       .loopH
     RET
 %endmacro
 
+    FILTER_VER_CHROMA_AVX2_2x8 pp
+    FILTER_VER_CHROMA_AVX2_2x8 ps
 
+%macro FILTER_VER_CHROMA_AVX2_2x16 1
 INIT_YMM avx2
-cglobal interp_8tap_horiz_pp_4x4, 4,6,6
+cglobal interp_4tap_vert_%1_2x16, 4, 6, 3
     mov             r4d, r4m
+    shl             r4d, 6
+    sub             r0,  r1
 
 %ifdef PIC
-    lea             r5, [tab_LumaCoeff]
-    vpbroadcastq    m0, [r5 + r4 * 8]
+    lea             r5,  [tab_ChromaCoeffVer_32]
+    add             r5,  r4
 %else
-    vpbroadcastq    m0, [tab_LumaCoeff + r4 * 8]
+    lea             r5,  [tab_ChromaCoeffVer_32 + r4]
 %endif
 
-    mova            m1, [tab_Lm]
-    vpbroadcastd    m2, [pw_1]
+    lea             r4,  [r1 * 3]
 
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
+    movd            xm1, [r0]
+    pinsrw          xm1, [r0 + r1], 1
+    pinsrw          xm1, [r0 + r1 * 2], 2
+    pinsrw          xm1, [r0 + r4], 3
+    lea             r0,  [r0 + r1 * 4]
+    pinsrw          xm1, [r0], 4
+    pinsrw          xm1, [r0 + r1], 5
+    pinsrw          xm1, [r0 + r1 * 2], 6
+    pinsrw          xm1, [r0 + r4], 7
+    lea             r0,  [r0 + r1 * 4]
+    pinsrw          xm0, [r0], 4
+    pinsrw          xm0, [r0 + r1], 5
+    pinsrw          xm0, [r0 + r1 * 2], 6
+    pinsrw          xm0, [r0 + r4], 7
+    punpckhqdq      xm0, xm1, xm0
+    vinserti128     m1,  m1,  xm0,  1
 
-    sub             r0, 3
-    ; Row 0-1
-    vbroadcasti128  m3, [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb          m3, m1
-    pmaddubsw       m3, m0
-    pmaddwd         m3, m2
-    vbroadcasti128  m4, [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb          m4, m1
-    pmaddubsw       m4, m0
-    pmaddwd         m4, m2
-    phaddd          m3, m4                          ; DWORD [R1D R1C R0D R0C R1B R1A R0B R0A]
-
-    ; Row 2-3
-    lea             r0, [r0 + r1 * 2]
-    vbroadcasti128  m4, [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb          m4, m1
-    pmaddubsw       m4, m0
-    pmaddwd         m4, m2
-    vbroadcasti128  m5, [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb          m5, m1
-    pmaddubsw       m5, m0
-    pmaddwd         m5, m2
-    phaddd          m4, m5                          ; DWORD [R3D R3C R2D R2C R3B R3A R2B R2A]
-
-    packssdw        m3, m4                          ; WORD [R3D R3C R2D R2C R1D R1C R0D R0C R3B R3A R2B R2A R1B R1A R0B R0A]
-    pmulhrsw        m3, [pw_512]
-    vextracti128    xm4, m3, 1
-    packuswb        xm3, xm4                        ; BYTE [R3D R3C R2D R2C R1D R1C R0D R0C R3B R3A R2B R2A R1B R1A R0B R0A]
-    pshufb          xm3, [interp4_shuf]             ; [row3 row1 row2 row0]
-
-    lea             r0, [r3 * 3]
-    movd            [r2], xm3
-    pextrd          [r2+r3], xm3, 2
-    pextrd          [r2+r3*2], xm3, 1
-    pextrd          [r2+r0], xm3, 3
-    RET
+    pshufb          m2,  m1,  [interp_vert_shuf]
+    pshufb          m1,  [interp_vert_shuf + 32]
+    pmaddubsw       m2,  [r5]
+    pmaddubsw       m1,  [r5 + 1 * mmsize]
+    paddw           m2,  m1
 
-%macro FILTER_HORIZ_LUMA_AVX2_4xN 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_horiz_pp_4x%1, 4, 6, 9
-    mov             r4d, r4m
+    lea             r0,  [r0 + r1 * 4]
+    pinsrw          xm1, [r0], 4
+    pinsrw          xm1, [r0 + r1], 5
+    pinsrw          xm1, [r0 + r1 * 2], 6
+    pinsrw          xm1, [r0 + r4], 7
+    punpckhqdq      xm1, xm0, xm1
+    lea             r0,  [r0 + r1 * 4]
+    pinsrw          xm0, [r0], 4
+    pinsrw          xm0, [r0 + r1], 5
+    pinsrw          xm0, [r0 + r1 * 2], 6
+    punpckhqdq      xm0, xm1, xm0
+    vinserti128     m1,  m1,  xm0,  1
 
-%ifdef PIC
-    lea             r5, [tab_LumaCoeff]
-    vpbroadcastq    m0, [r5 + r4 * 8]
+    pshufb          m0,  m1,  [interp_vert_shuf]
+    pshufb          m1,  [interp_vert_shuf + 32]
+    pmaddubsw       m0,  [r5]
+    pmaddubsw       m1,  [r5 + 1 * mmsize]
+    paddw           m0,  m1
+%ifidn %1,pp
+    mova            m1,  [pw_512]
+    pmulhrsw        m2,  m1
+    pmulhrsw        m0,  m1
+    packuswb        m2,  m0
+    lea             r4,  [r3 * 3]
+    pextrw          [r2], xm2, 0
+    pextrw          [r2 + r3], xm2, 1
+    pextrw          [r2 + r3 * 2], xm2, 2
+    pextrw          [r2 + r4], xm2, 3
+    vextracti128    xm0, m2, 1
+    lea             r2,  [r2 + r3 * 4]
+    pextrw          [r2], xm0, 0
+    pextrw          [r2 + r3], xm0, 1
+    pextrw          [r2 + r3 * 2], xm0, 2
+    pextrw          [r2 + r4], xm0, 3
+    lea             r2,  [r2 + r3 * 4]
+    pextrw          [r2], xm2, 4
+    pextrw          [r2 + r3], xm2, 5
+    pextrw          [r2 + r3 * 2], xm2, 6
+    pextrw          [r2 + r4], xm2, 7
+    lea             r2,  [r2 + r3 * 4]
+    pextrw          [r2], xm0, 4
+    pextrw          [r2 + r3], xm0, 5
+    pextrw          [r2 + r3 * 2], xm0, 6
+    pextrw          [r2 + r4], xm0, 7
 %else
-    vpbroadcastq    m0, [tab_LumaCoeff + r4 * 8]
-%endif
-
-    mova            m1, [tab_Lm]
-    mova            m2, [pw_1]
-    mova            m7, [interp8_hps_shuf]
-    mova            m8, [pw_512]
-
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-    lea             r4, [r1 * 3]
-    lea             r5, [r3 * 3]
-    sub             r0, 3
-%rep %1 / 8
-    ; Row 0-1
-    vbroadcasti128  m3, [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb          m3, m1
-    pmaddubsw       m3, m0
-    pmaddwd         m3, m2
-    vbroadcasti128  m4, [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb          m4, m1
-    pmaddubsw       m4, m0
-    pmaddwd         m4, m2
-    phaddd          m3, m4                          ; DWORD [R1D R1C R0D R0C R1B R1A R0B R0A]
-
-    ; Row 2-3
-    vbroadcasti128  m4, [r0 + r1 * 2]               ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb          m4, m1
-    pmaddubsw       m4, m0
-    pmaddwd         m4, m2
-    vbroadcasti128  m5, [r0 + r4]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb          m5, m1
-    pmaddubsw       m5, m0
-    pmaddwd         m5, m2
-    phaddd          m4, m5                          ; DWORD [R3D R3C R2D R2C R3B R3A R2B R2A]
-
-    packssdw        m3, m4                          ; WORD [R3D R3C R2D R2C R1D R1C R0D R0C R3B R3A R2B R2A R1B R1A R0B R0A]
-    lea             r0, [r0 + r1 * 4]
-    ; Row 4-5
-    vbroadcasti128  m5, [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb          m5, m1
-    pmaddubsw       m5, m0
-    pmaddwd         m5, m2
-    vbroadcasti128  m4, [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb          m4, m1
-    pmaddubsw       m4, m0
-    pmaddwd         m4, m2
-    phaddd          m5, m4                          ; DWORD [R5D R5C R4D R4C R5B R5A R4B R4A]
-
-    ; Row 6-7
-    vbroadcasti128  m4, [r0 + r1 * 2]               ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb          m4, m1
-    pmaddubsw       m4, m0
-    pmaddwd         m4, m2
-    vbroadcasti128  m6, [r0 + r4]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb          m6, m1
-    pmaddubsw       m6, m0
-    pmaddwd         m6, m2
-    phaddd          m4, m6                          ; DWORD [R7D R7C R6D R6C R7B R7A R6B R6A]
-
-    packssdw        m5, m4                          ; WORD [R7D R7C R6D R6C R5D R5C R4D R4C R7B R7A R6B R6A R5B R5A R4B R4A]
-    vpermd          m3, m7, m3
-    vpermd          m5, m7, m5
-    pmulhrsw        m3, m8
-    pmulhrsw        m5, m8
-    packuswb        m3, m5
-    vextracti128    xm5, m3, 1
-
-    movd            [r2], xm3
-    pextrd          [r2 + r3], xm3, 1
-    movd            [r2 + r3 * 2], xm5
-    pextrd          [r2 + r5], xm5, 1
-    lea             r2, [r2 + r3 * 4]
-    pextrd          [r2], xm3, 2
-    pextrd          [r2 + r3], xm3, 3
-    pextrd          [r2 + r3 * 2], xm5, 2
-    pextrd          [r2 + r5], xm5, 3
-    lea             r0, [r0 + r1 * 4]
+    add             r3d, r3d
+    lea             r4,  [r3 * 3]
+    vbroadcasti128  m1,  [pw_2000]
+    psubw           m2,  m1
+    psubw           m0,  m1
+    vextracti128    xm1, m2, 1
+    movd            [r2], xm2
+    pextrd          [r2 + r3], xm2, 1
+    pextrd          [r2 + r3 * 2], xm2, 2
+    pextrd          [r2 + r4], xm2, 3
     lea             r2, [r2 + r3 * 4]
-%endrep
-    RET
+    movd            [r2], xm1
+    pextrd          [r2 + r3], xm1, 1
+    pextrd          [r2 + r3 * 2], xm1, 2
+    pextrd          [r2 + r4], xm1, 3
+    vextracti128    xm1, m0, 1
+    lea             r2,  [r2 + r3 * 4]
+    movd            [r2], xm0
+    pextrd          [r2 + r3], xm0, 1
+    pextrd          [r2 + r3 * 2], xm0, 2
+    pextrd          [r2 + r4], xm0, 3
+    lea             r2,  [r2 + r3 * 4]
+    movd            [r2], xm1
+    pextrd          [r2 + r3], xm1, 1
+    pextrd          [r2 + r3 * 2], xm1, 2
+    pextrd          [r2 + r4], xm1, 3
 %endif
+    RET
 %endmacro
 
-    FILTER_HORIZ_LUMA_AVX2_4xN 8
-    FILTER_HORIZ_LUMA_AVX2_4xN 16
+    FILTER_VER_CHROMA_AVX2_2x16 pp
+    FILTER_VER_CHROMA_AVX2_2x16 ps
 
-INIT_YMM avx2
-cglobal interp_8tap_horiz_pp_8x4, 4, 6, 7
-    mov             r4d, r4m
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_pp_2x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W2_H4 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_pp_2x%2, 4, 6, 8
+
+    mov         r4d,       r4m
+    sub         r0,        r1
 
 %ifdef PIC
-    lea             r5, [tab_LumaCoeff]
-    vpbroadcastq    m0, [r5 + r4 * 8]
+    lea         r5,        [tab_ChromaCoeff]
+    movd        m0,        [r5 + r4 * 4]
 %else
-    vpbroadcastq    m0, [tab_LumaCoeff + r4 * 8]
+    movd        m0,        [tab_ChromaCoeff + r4 * 4]
 %endif
 
-    mova            m1, [tab_Lm]
-    mova            m2, [tab_Lm + 32]
-
-    ; register map
-    ; m0     - interpolate coeff
-    ; m1, m2 - shuffle order table
+    pshufb      m0,        [tab_Cm]
 
-    sub             r0, 3
-    lea             r5, [r1 * 3]
-    lea             r4, [r3 * 3]
+    mova        m1,        [pw_512]
 
-    ; Row 0
-    vbroadcasti128  m3, [r0]                        ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
-    pshufb          m4, m3, m2
-    pshufb          m3, m1
-    pmaddubsw       m3, m0
-    pmaddubsw       m4, m0
-    phaddw          m3, m4
-    ; Row 1
-    vbroadcasti128  m4, [r0 + r1]                   ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
-    pshufb          m5, m4, m2
-    pshufb          m4, m1
-    pmaddubsw       m4, m0
-    pmaddubsw       m5, m0
-    phaddw          m4, m5
-
-    phaddw          m3, m4                          ; WORD [R1H R1G R1D R1C R0H R0G R0D R0C R1F R1E R1B R1A R0F R0E R0B R0A]
-    pmulhrsw        m3, [pw_512]
-
-    ; Row 2
-    vbroadcasti128  m4, [r0 + r1 * 2]               ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
-    pshufb          m5, m4, m2
-    pshufb          m4, m1
-    pmaddubsw       m4, m0
-    pmaddubsw       m5, m0
-    phaddw          m4, m5
-    ; Row 3
-    vbroadcasti128  m5, [r0 + r5]                   ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
-    pshufb          m6, m5, m2
-    pshufb          m5, m1
-    pmaddubsw       m5, m0
-    pmaddubsw       m6, m0
-    phaddw          m5, m6
-
-    phaddw          m4, m5                          ; WORD [R3H R3G R3D R3C R2H R2G R2D R2C R3F R3E R3B R3A R2F R2E R2B R2A]
-    pmulhrsw        m4, [pw_512]
+    mov         r4d,       %2
+    lea         r5,        [3 * r1]
 
-    packuswb        m3, m4
-    vextracti128    xm4, m3, 1
-    punpcklwd       xm5, xm3, xm4
+.loop:
+    movd        m2,        [r0]
+    movd        m3,        [r0 + r1]
+    movd        m4,        [r0 + 2 * r1]
+    movd        m5,        [r0 + r5]
 
-    movq            [r2], xm5
-    movhps          [r2 + r3], xm5
+    punpcklbw   m2,        m3
+    punpcklbw   m6,        m4,        m5
+    punpcklbw   m2,        m6
 
-    punpckhwd       xm5, xm3, xm4
-    movq            [r2 + r3 * 2], xm5
-    movhps          [r2 + r4], xm5
-    RET
+    pmaddubsw   m2,        m0
 
-%macro IPFILTER_LUMA_AVX2_8xN 2
-INIT_YMM avx2
-cglobal interp_8tap_horiz_pp_%1x%2, 4, 7, 7
-    mov             r4d, r4m
+    lea         r0,        [r0 + 4 * r1]
+    movd        m6,        [r0]
 
-%ifdef PIC
-    lea             r5, [tab_LumaCoeff]
-    vpbroadcastq    m0, [r5 + r4 * 8]
-%else
-    vpbroadcastq    m0, [tab_LumaCoeff + r4 * 8]
-%endif
+    punpcklbw   m3,        m4
+    punpcklbw   m7,        m5,        m6
+    punpcklbw   m3,        m7
 
-    mova            m1, [tab_Lm]
-    mova            m2, [tab_Lm + 32]
+    pmaddubsw   m3,        m0
 
-    ; register map
-    ; m0     - interpolate coeff
-    ; m1, m2 - shuffle order table
+    phaddw      m2,        m3
 
-    sub             r0, 3
-    lea             r5, [r1 * 3]
-    lea             r6, [r3 * 3]
-    mov             r4d, %2 / 4
-.loop:
-    ; Row 0
-    vbroadcasti128  m3, [r0]                        ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
-    pshufb          m4, m3, m2
-    pshufb          m3, m1
-    pmaddubsw       m3, m0
-    pmaddubsw       m4, m0
-    phaddw          m3, m4
-    ; Row 1
-    vbroadcasti128  m4, [r0 + r1]                   ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
-    pshufb          m5, m4, m2
-    pshufb          m4, m1
-    pmaddubsw       m4, m0
-    pmaddubsw       m5, m0
-    phaddw          m4, m5
-
-    phaddw          m3, m4                          ; WORD [R1H R1G R1D R1C R0H R0G R0D R0C R1F R1E R1B R1A R0F R0E R0B R0A]
-    pmulhrsw        m3, [pw_512]
-
-    ; Row 2
-    vbroadcasti128  m4, [r0 + r1 * 2]               ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
-    pshufb          m5, m4, m2
-    pshufb          m4, m1
-    pmaddubsw       m4, m0
-    pmaddubsw       m5, m0
-    phaddw          m4, m5
-    ; Row 3
-    vbroadcasti128  m5, [r0 + r5]                   ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
-    pshufb          m6, m5, m2
-    pshufb          m5, m1
-    pmaddubsw       m5, m0
-    pmaddubsw       m6, m0
-    phaddw          m5, m6
-
-    phaddw          m4, m5                          ; WORD [R3H R3G R3D R3C R2H R2G R2D R2C R3F R3E R3B R3A R2F R2E R2B R2A]
-    pmulhrsw        m4, [pw_512]
+    pmulhrsw    m2,        m1
 
-    packuswb        m3, m4
-    vextracti128    xm4, m3, 1
-    punpcklwd       xm5, xm3, xm4
+    movd        m7,        [r0 + r1]
 
-    movq            [r2], xm5
-    movhps          [r2 + r3], xm5
+    punpcklbw   m4,        m5
+    punpcklbw   m3,        m6,        m7
+    punpcklbw   m4,        m3
 
-    punpckhwd       xm5, xm3, xm4
-    movq            [r2 + r3 * 2], xm5
-    movhps          [r2 + r6], xm5
+    pmaddubsw   m4,        m0
 
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
-    dec             r4d
-    jnz             .loop
-    RET
-%endmacro
+    movd        m3,        [r0 + 2 * r1]
 
-    IPFILTER_LUMA_AVX2_8xN 8, 8
-    IPFILTER_LUMA_AVX2_8xN 8, 16
-    IPFILTER_LUMA_AVX2_8xN 8, 32
+    punpcklbw   m5,        m6
+    punpcklbw   m7,        m3
+    punpcklbw   m5,        m7
 
-%macro IPFILTER_LUMA_AVX2 2
-INIT_YMM avx2
-cglobal interp_8tap_horiz_pp_%1x%2, 4,6,8
-    sub               r0,        3
-    mov               r4d,       r4m
-%ifdef PIC
-    lea               r5,        [tab_LumaCoeff]
-    vpbroadcastd      m0,        [r5 + r4 * 8]
-    vpbroadcastd      m1,        [r5 + r4 * 8 + 4]
-%else
-    vpbroadcastd      m0,         [tab_LumaCoeff + r4 * 8]
-    vpbroadcastd      m1,         [tab_LumaCoeff + r4 * 8 + 4]
-%endif
-    movu              m3,         [tab_Tm + 16]
-    vpbroadcastd      m7,         [pw_1]
+    pmaddubsw   m5,        m0
 
-    ; register map
-    ; m0 , m1 interpolate coeff
-    ; m2 , m2  shuffle order table
-    ; m7 - pw_1
+    phaddw      m4,        m5
 
-    mov               r4d,        %2/2
-.loop:
-    ; Row 0
-    vbroadcasti128    m4,         [r0]                        ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m5,         m4,     m3
-    pshufb            m4,         [tab_Tm]
-    pmaddubsw         m4,         m0
-    pmaddubsw         m5,         m1
-    paddw             m4,         m5
-    pmaddwd           m4,         m7
-    vbroadcasti128    m5,         [r0 + 8]                    ; second 8 elements in Row0
-    pshufb            m6,         m5,     m3
-    pshufb            m5,         [tab_Tm]
-    pmaddubsw         m5,         m0
-    pmaddubsw         m6,         m1
-    paddw             m5,         m6
-    pmaddwd           m5,         m7
-    packssdw          m4,         m5                          ; [17 16 15 14 07 06 05 04 13 12 11 10 03 02 01 00]
-    pmulhrsw          m4,         [pw_512]
-    vbroadcasti128    m2,         [r0 + r1]                        ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m5,         m2,     m3
-    pshufb            m2,         [tab_Tm]
-    pmaddubsw         m2,         m0
-    pmaddubsw         m5,         m1
-    paddw             m2,         m5
-    pmaddwd           m2,         m7
-    vbroadcasti128    m5,         [r0 + r1 + 8]                    ; second 8 elements in Row0
-    pshufb            m6,         m5,     m3
-    pshufb            m5,         [tab_Tm]
-    pmaddubsw         m5,         m0
-    pmaddubsw         m6,         m1
-    paddw             m5,         m6
-    pmaddwd           m5,         m7
-    packssdw          m2,         m5                          ; [17 16 15 14 07 06 05 04 13 12 11 10 03 02 01 00]
-    pmulhrsw          m2,         [pw_512]
-    packuswb          m4,         m2
-    vpermq            m4,         m4,     11011000b
-    vextracti128      xm5,        m4,     1
-    pshufd            xm4,        xm4,    11011000b
-    pshufd            xm5,        xm5,    11011000b
-    movu              [r2],       xm4
-    movu              [r2+r3],    xm5
-    lea               r0,         [r0 + r1 * 2]
-    lea               r2,         [r2 + r3 * 2]
-    dec               r4d
-    jnz              .loop
-    RET
-%endmacro
+    pmulhrsw    m4,        m1
+    packuswb    m2,        m4
 
-%macro IPFILTER_LUMA_32x_avx2 2
-INIT_YMM avx2
-cglobal interp_8tap_horiz_pp_%1x%2, 4,6,8
-    sub               r0,         3
-    mov               r4d,        r4m
-%ifdef PIC
-    lea               r5,         [tab_LumaCoeff]
-    vpbroadcastd      m0,         [r5 + r4 * 8]
-    vpbroadcastd      m1,         [r5 + r4 * 8 + 4]
-%else
-    vpbroadcastd      m0,         [tab_LumaCoeff + r4 * 8]
-    vpbroadcastd      m1,         [tab_LumaCoeff + r4 * 8 + 4]
-%endif
-    movu              m3,         [tab_Tm + 16]
-    vpbroadcastd      m7,         [pw_1]
+    pextrw      [r2],      m2, 0
+    pextrw      [r2 + r3], m2, 2
+    lea         r2,        [r2 + 2 * r3]
+    pextrw      [r2],      m2, 4
+    pextrw      [r2 + r3], m2, 6
 
-    ; register map
-    ; m0 , m1 interpolate coeff
-    ; m2 , m2  shuffle order table
-    ; m7 - pw_1
+    lea         r2,        [r2 + 2 * r3]
 
-    mov               r4d,        %2
-.loop:
-    ; Row 0
-    vbroadcasti128    m4,         [r0]                        ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m5,         m4,     m3
-    pshufb            m4,         [tab_Tm]
-    pmaddubsw         m4,         m0
-    pmaddubsw         m5,         m1
-    paddw             m4,         m5
-    pmaddwd           m4,         m7
-    vbroadcasti128    m5,         [r0 + 8]
-    pshufb            m6,         m5,     m3
-    pshufb            m5,         [tab_Tm]
-    pmaddubsw         m5,         m0
-    pmaddubsw         m6,         m1
-    paddw             m5,         m6
-    pmaddwd           m5,         m7
-    packssdw          m4,         m5                          ; [17 16 15 14 07 06 05 04 13 12 11 10 03 02 01 00]
-    pmulhrsw          m4,         [pw_512]
-    vbroadcasti128    m2,         [r0 + 16]
-    pshufb            m5,         m2,     m3
-    pshufb            m2,         [tab_Tm]
-    pmaddubsw         m2,         m0
-    pmaddubsw         m5,         m1
-    paddw             m2,         m5
-    pmaddwd           m2,         m7
-    vbroadcasti128    m5,         [r0 + 24]
-    pshufb            m6,         m5,     m3
-    pshufb            m5,         [tab_Tm]
-    pmaddubsw         m5,         m0
-    pmaddubsw         m6,         m1
-    paddw             m5,         m6
-    pmaddwd           m5,         m7
-    packssdw          m2,         m5
-    pmulhrsw          m2,         [pw_512]
-    packuswb          m4,         m2
-    vpermq            m4,         m4,     11011000b
-    vextracti128      xm5,        m4,     1
-    pshufd            xm4,        xm4,    11011000b
-    pshufd            xm5,        xm5,    11011000b
-    movu              [r2],       xm4
-    movu              [r2 + 16],  xm5
-    lea               r0,         [r0 + r1]
-    lea               r2,         [r2 + r3]
-    dec               r4d
-    jnz               .loop
+    sub         r4,        4
+    jnz        .loop
     RET
 %endmacro
 
-%macro IPFILTER_LUMA_64x_avx2 2
-INIT_YMM avx2
-cglobal interp_8tap_horiz_pp_%1x%2, 4,6,8
-    sub               r0,    3
-    mov               r4d,   r4m
-%ifdef PIC
-    lea               r5,        [tab_LumaCoeff]
-    vpbroadcastd      m0,        [r5 + r4 * 8]
-    vpbroadcastd      m1,        [r5 + r4 * 8 + 4]
-%else
-    vpbroadcastd      m0,        [tab_LumaCoeff + r4 * 8]
-    vpbroadcastd      m1,        [tab_LumaCoeff + r4 * 8 + 4]
-%endif
-    movu              m3,        [tab_Tm + 16]
-    vpbroadcastd      m7,        [pw_1]
+    FILTER_V4_W2_H4 2, 8
 
-    ; register map
-    ; m0 , m1 interpolate coeff
-    ; m2 , m2  shuffle order table
-    ; m7 - pw_1
+    FILTER_V4_W2_H4 2, 16
 
-    mov               r4d,   %2
-.loop:
-    ; Row 0
-    vbroadcasti128    m4,        [r0]                        ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m5,        m4,    m3
-    pshufb            m4,        [tab_Tm]
-    pmaddubsw         m4,        m0
-    pmaddubsw         m5,        m1
-    paddw             m4,        m5
-    pmaddwd           m4,        m7
-    vbroadcasti128    m5,        [r0 + 8]
-    pshufb            m6,        m5,    m3
-    pshufb            m5,        [tab_Tm]
-    pmaddubsw         m5,        m0
-    pmaddubsw         m6,        m1
-    paddw             m5,        m6
-    pmaddwd           m5,        m7
-    packssdw          m4,        m5                          ; [17 16 15 14 07 06 05 04 13 12 11 10 03 02 01 00]
-    pmulhrsw          m4,        [pw_512]
-    vbroadcasti128    m2,        [r0 + 16]
-    pshufb            m5,        m2,    m3
-    pshufb            m2,        [tab_Tm]
-    pmaddubsw         m2,        m0
-    pmaddubsw         m5,        m1
-    paddw             m2,        m5
-    pmaddwd           m2,        m7
-    vbroadcasti128    m5,        [r0 + 24]
-    pshufb            m6,        m5,    m3
-    pshufb            m5,        [tab_Tm]
-    pmaddubsw         m5,        m0
-    pmaddubsw         m6,        m1
-    paddw             m5,        m6
-    pmaddwd           m5,        m7
-    packssdw          m2,        m5
-    pmulhrsw          m2,        [pw_512]
-    packuswb          m4,        m2
-    vpermq            m4,        m4,    11011000b
-    vextracti128      xm5,       m4,    1
-    pshufd            xm4,       xm4,   11011000b
-    pshufd            xm5,       xm5,   11011000b
-    movu              [r2],      xm4
-    movu              [r2 + 16], xm5
-
-    vbroadcasti128    m4,        [r0 + 32]
-    pshufb            m5,        m4,    m3
-    pshufb            m4,        [tab_Tm]
-    pmaddubsw         m4,        m0
-    pmaddubsw         m5,        m1
-    paddw             m4,        m5
-    pmaddwd           m4,        m7
-    vbroadcasti128    m5,        [r0 + 40]
-    pshufb            m6,        m5,    m3
-    pshufb            m5,        [tab_Tm]
-    pmaddubsw         m5,        m0
-    pmaddubsw         m6,        m1
-    paddw             m5,        m6
-    pmaddwd           m5,        m7
-    packssdw          m4,        m5
-    pmulhrsw          m4,        [pw_512]
-    vbroadcasti128    m2,        [r0 + 48]
-    pshufb            m5,        m2,    m3
-    pshufb            m2,        [tab_Tm]
-    pmaddubsw         m2,        m0
-    pmaddubsw         m5,        m1
-    paddw             m2,        m5
-    pmaddwd           m2,        m7
-    vbroadcasti128    m5,        [r0 + 56]
-    pshufb            m6,        m5,    m3
-    pshufb            m5,        [tab_Tm]
-    pmaddubsw         m5,        m0
-    pmaddubsw         m6,        m1
-    paddw             m5,        m6
-    pmaddwd           m5,        m7
-    packssdw          m2,        m5
-    pmulhrsw          m2,        [pw_512]
-    packuswb          m4,        m2
-    vpermq            m4,        m4,    11011000b
-    vextracti128      xm5,       m4,    1
-    pshufd            xm4,       xm4,   11011000b
-    pshufd            xm5,       xm5,   11011000b
-    movu              [r2 +32],  xm4
-    movu              [r2 + 48], xm5
-
-    lea               r0,        [r0 + r1]
-    lea               r2,        [r2 + r3]
-    dec               r4d
-    jnz               .loop
-    RET
-%endmacro
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_pp_4x2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+INIT_XMM sse4
+cglobal interp_4tap_vert_pp_4x2, 4, 6, 6
+
+    mov         r4d,       r4m
+    sub         r0,        r1
 
-INIT_YMM avx2
-cglobal interp_8tap_horiz_pp_48x64, 4,6,8
-    sub               r0,         3
-    mov               r4d,        r4m
 %ifdef PIC
-    lea               r5,         [tab_LumaCoeff]
-    vpbroadcastd      m0,         [r5 + r4 * 8]
-    vpbroadcastd      m1,         [r5 + r4 * 8 + 4]
+    lea         r5,        [tab_ChromaCoeff]
+    movd        m0,        [r5 + r4 * 4]
 %else
-    vpbroadcastd      m0,         [tab_LumaCoeff + r4 * 8]
-    vpbroadcastd      m1,         [tab_LumaCoeff + r4 * 8 + 4]
+    movd        m0,        [tab_ChromaCoeff + r4 * 4]
 %endif
-    movu              m3,         [tab_Tm + 16]
-    vpbroadcastd      m7,         [pw_1]
 
-    ; register map
-    ; m0 , m1 interpolate coeff
-    ; m2 , m2  shuffle order table
-    ; m7 - pw_1
+    pshufb      m0,        [tab_Cm]
+    lea         r5,        [r0 + 2 * r1]
+
+    movd        m2,        [r0]
+    movd        m3,        [r0 + r1]
+    movd        m4,        [r5]
+    movd        m5,        [r5 + r1]
 
-    mov               r4d,        64
-.loop:
-    ; Row 0
-    vbroadcasti128    m4,         [r0]                        ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m5,         m4,     m3
-    pshufb            m4,         [tab_Tm]
-    pmaddubsw         m4,         m0
-    pmaddubsw         m5,         m1
-    paddw             m4,         m5
-    pmaddwd           m4,         m7
-    vbroadcasti128    m5,         [r0 + 8]
-    pshufb            m6,         m5,     m3
-    pshufb            m5,         [tab_Tm]
-    pmaddubsw         m5,         m0
-    pmaddubsw         m6,         m1
-    paddw             m5,         m6
-    pmaddwd           m5,         m7
-    packssdw          m4,         m5                          ; [17 16 15 14 07 06 05 04 13 12 11 10 03 02 01 00]
-    pmulhrsw          m4,         [pw_512]
-
-    vbroadcasti128    m2,         [r0 + 16]
-    pshufb            m5,         m2,     m3
-    pshufb            m2,         [tab_Tm]
-    pmaddubsw         m2,         m0
-    pmaddubsw         m5,         m1
-    paddw             m2,         m5
-    pmaddwd           m2,         m7
-    vbroadcasti128    m5,         [r0 + 24]
-    pshufb            m6,         m5,     m3
-    pshufb            m5,         [tab_Tm]
-    pmaddubsw         m5,         m0
-    pmaddubsw         m6,         m1
-    paddw             m5,         m6
-    pmaddwd           m5,         m7
-    packssdw          m2,         m5
-    pmulhrsw          m2,         [pw_512]
-    packuswb          m4,         m2
-    vpermq            m4,         m4,     11011000b
-    vextracti128      xm5,        m4,     1
-    pshufd            xm4,        xm4,    11011000b
-    pshufd            xm5,        xm5,    11011000b
-    movu              [r2],       xm4
-    movu              [r2 + 16],  xm5
-
-    vbroadcasti128    m4,         [r0 + 32]
-    pshufb            m5,         m4,     m3
-    pshufb            m4,         [tab_Tm]
-    pmaddubsw         m4,         m0
-    pmaddubsw         m5,         m1
-    paddw             m4,         m5
-    pmaddwd           m4,         m7
-    vbroadcasti128    m5,         [r0 + 40]
-    pshufb            m6,         m5,     m3
-    pshufb            m5,         [tab_Tm]
-    pmaddubsw         m5,         m0
-    pmaddubsw         m6,         m1
-    paddw             m5,         m6
-    pmaddwd           m5,         m7
-    packssdw          m4,         m5
-    pmulhrsw          m4,         [pw_512]
-    packuswb          m4,         m4
-    vpermq            m4,         m4,     11011000b
-    pshufd            xm4,        xm4,    11011000b
-    movu              [r2 + 32],  xm4
-
-    lea               r0,         [r0 + r1]
-    lea               r2,         [r2 + r3]
-    dec               r4d
-    jnz               .loop
-    RET
+    punpcklbw   m2,        m3
+    punpcklbw   m1,        m4,        m5
+    punpcklbw   m2,        m1
 
-INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_4x4, 4,6,6
-    mov             r4d, r4m
+    pmaddubsw   m2,        m0
 
-%ifdef PIC
-    lea               r5,           [tab_ChromaCoeff]
-    vpbroadcastd      m0,           [r5 + r4 * 4]
-%else
-    vpbroadcastd      m0,           [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    vpbroadcastd      m2,           [pw_1]
-    vbroadcasti128    m1,           [tab_Tm]
-
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-
-    dec                r0
-
-    ; Row 0-1
-    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    vinserti128       m3,           m3,      [r0 + r1],     1
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-
-    ; Row 2-3
-    lea               r0,           [r0 + r1 * 2]
-    vbroadcasti128    m4,           [r0]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    vinserti128       m4,           m4,      [r0 + r1],     1
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-
-    packssdw          m3,           m4
-    pmulhrsw          m3,           [pw_512]
-    vextracti128      xm4,          m3,     1
-    packuswb          xm3,          xm4
-
-    lea               r0,           [r3 * 3]
-    movd              [r2],         xm3
-    pextrd            [r2+r3],      xm3,     2
-    pextrd            [r2+r3*2],    xm3,     1
-    pextrd            [r2+r0],      xm3,     3
-    RET
+    movd        m1,        [r0 + 4 * r1]
 
-INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_2x4, 4, 6, 3
-    mov               r4d,           r4m
+    punpcklbw   m3,        m4
+    punpcklbw   m5,        m1
+    punpcklbw   m3,        m5
 
-%ifdef PIC
-    lea               r5,            [tab_ChromaCoeff]
-    vpbroadcastd      m0,            [r5 + r4 * 4]
-%else
-    vpbroadcastd      m0,            [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    dec               r0
-    lea               r4,            [r1 * 3]
-    movq              xm1,           [r0]
-    movhps            xm1,           [r0 + r1]
-    movq              xm2,           [r0 + r1 * 2]
-    movhps            xm2,           [r0 + r4]
-    vinserti128       m1,            m1,          xm2,          1
-    pshufb            m1,            [interp4_hpp_shuf]
-    pmaddubsw         m1,            m0
-    pmaddwd           m1,            [pw_1]
-    vextracti128      xm2,           m1,          1
-    packssdw          xm1,           xm2
-    pmulhrsw          xm1,           [pw_512]
-    packuswb          xm1,           xm1
-
-    lea               r4,            [r3 * 3]
-    pextrw            [r2],          xm1,         0
-    pextrw            [r2 + r3],     xm1,         1
-    pextrw            [r2 + r3 * 2], xm1,         2
-    pextrw            [r2 + r4],     xm1,         3
-    RET
+    pmaddubsw   m3,        m0
 
-INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_2x8, 4, 6, 6
-    mov               r4d,           r4m
+    phaddw      m2,        m3
+
+    pmulhrsw    m2,        [pw_512]
+    packuswb    m2,        m2
+    movd        [r2],      m2
+    pextrd      [r2 + r3], m2,  1
 
-%ifdef PIC
-    lea               r5,            [tab_ChromaCoeff]
-    vpbroadcastd      m0,            [r5 + r4 * 4]
-%else
-    vpbroadcastd      m0,            [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    mova              m4,            [interp4_hpp_shuf]
-    mova              m5,            [pw_1]
-    dec               r0
-    lea               r4,            [r1 * 3]
-    movq              xm1,           [r0]
-    movhps            xm1,           [r0 + r1]
-    movq              xm2,           [r0 + r1 * 2]
-    movhps            xm2,           [r0 + r4]
-    vinserti128       m1,            m1,          xm2,          1
-    lea               r0,            [r0 + r1 * 4]
-    movq              xm3,           [r0]
-    movhps            xm3,           [r0 + r1]
-    movq              xm2,           [r0 + r1 * 2]
-    movhps            xm2,           [r0 + r4]
-    vinserti128       m3,            m3,          xm2,          1
-
-    pshufb            m1,            m4
-    pshufb            m3,            m4
-    pmaddubsw         m1,            m0
-    pmaddubsw         m3,            m0
-    pmaddwd           m1,            m5
-    pmaddwd           m3,            m5
-    packssdw          m1,            m3
-    pmulhrsw          m1,            [pw_512]
-    vextracti128      xm2,           m1,          1
-    packuswb          xm1,           xm2
-
-    lea               r4,            [r3 * 3]
-    pextrw            [r2],          xm1,         0
-    pextrw            [r2 + r3],     xm1,         1
-    pextrw            [r2 + r3 * 2], xm1,         4
-    pextrw            [r2 + r4],     xm1,         5
-    lea               r2,            [r2 + r3 * 4]
-    pextrw            [r2],          xm1,         2
-    pextrw            [r2 + r3],     xm1,         3
-    pextrw            [r2 + r3 * 2], xm1,         6
-    pextrw            [r2 + r4],     xm1,         7
     RET
 
+%macro FILTER_VER_CHROMA_AVX2_4x2 1
 INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_32x32, 4,6,7
+cglobal interp_4tap_vert_%1_4x2, 4, 6, 4
     mov             r4d, r4m
+    shl             r4d, 5
+    sub             r0, r1
 
 %ifdef PIC
-    lea               r5,           [tab_ChromaCoeff]
-    vpbroadcastd      m0,           [r5 + r4 * 4]
+    lea             r5, [tab_ChromaCoeff_V]
+    add             r5, r4
 %else
-    vpbroadcastd      m0,           [tab_ChromaCoeff + r4 * 4]
+    lea             r5, [tab_ChromaCoeff_V + r4]
 %endif
 
-    mova              m1,           [interp4_horiz_shuf1]
-    vpbroadcastd      m2,           [pw_1]
-    mova              m6,           [pw_512]
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
+    lea             r4, [r1 * 3]
 
-    dec               r0
-    mov               r4d,          32
+    movd            xm1, [r0]
+    movd            xm2, [r0 + r1]
+    punpcklbw       xm1, xm2
+    movd            xm3, [r0 + r1 * 2]
+    punpcklbw       xm2, xm3
+    movlhps         xm1, xm2
+    movd            xm0, [r0 + r4]
+    punpcklbw       xm3, xm0
+    movd            xm2, [r0 + r1 * 4]
+    punpcklbw       xm0, xm2
+    movlhps         xm3, xm0
+    vinserti128     m1, m1, xm3, 1                          ; m1 = row[x x x 4 3 2 1 0]
 
-.loop:
-    ; Row 0
-    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-    vbroadcasti128    m4,           [r0 + 4]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-    packssdw          m3,           m4
-    pmulhrsw          m3,           m6
-
-    vbroadcasti128    m4,           [r0 + 16]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-    vbroadcasti128    m5,           [r0 + 20]
-    pshufb            m5,           m1
-    pmaddubsw         m5,           m0
-    pmaddwd           m5,           m2
-    packssdw          m4,           m5
-    pmulhrsw          m4,           m6
-
-    packuswb          m3,           m4
-    vpermq            m3,           m3,      11011000b
-
-    movu              [r2],         m3
-    lea               r2,           [r2 + r3]
-    lea               r0,           [r0 + r1]
-    dec               r4d
-    jnz               .loop
+    pmaddubsw       m1, [r5]
+    vextracti128    xm3, m1, 1
+    paddw           xm1, xm3
+%ifidn %1,pp
+    pmulhrsw        xm1, [pw_512]
+    packuswb        xm1, xm1
+    movd            [r2], xm1
+    pextrd          [r2 + r3], xm1, 1
+%else
+    add             r3d, r3d
+    psubw           xm1, [pw_2000]
+    movq            [r2], xm1
+    movhps          [r2 + r3], xm1
+%endif
     RET
+%endmacro
 
+    FILTER_VER_CHROMA_AVX2_4x2 pp
+    FILTER_VER_CHROMA_AVX2_4x2 ps
 
-INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_16x16, 4, 6, 7
-    mov               r4d,          r4m
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_pp_4x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+INIT_XMM sse4
+cglobal interp_4tap_vert_pp_4x4, 4, 6, 8
+
+    mov         r4d,       r4m
+    sub         r0,        r1
 
 %ifdef PIC
-    lea               r5,           [tab_ChromaCoeff]
-    vpbroadcastd      m0,           [r5 + r4 * 4]
+    lea         r5,        [tab_ChromaCoeff]
+    movd        m0,        [r5 + r4 * 4]
 %else
-    vpbroadcastd      m0,           [tab_ChromaCoeff + r4 * 4]
+    movd        m0,        [tab_ChromaCoeff + r4 * 4]
 %endif
 
-    mova              m6,           [pw_512]
-    mova              m1,           [interp4_horiz_shuf1]
-    vpbroadcastd      m2,           [pw_1]
-
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-
-    dec               r0
-    mov               r4d,          8
-
-.loop:
-    ; Row 0
-    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-    vbroadcasti128    m4,           [r0 + 4]                    ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-    packssdw          m3,           m4
-    pmulhrsw          m3,           m6
-
-    ; Row 1
-    vbroadcasti128    m4,           [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-    vbroadcasti128    m5,           [r0 + r1 + 4]               ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m5,           m1
-    pmaddubsw         m5,           m0
-    pmaddwd           m5,           m2
-    packssdw          m4,           m5
-    pmulhrsw          m4,           m6
-
-    packuswb          m3,           m4
-    vpermq            m3,           m3,      11011000b
-
-    vextracti128      xm4,          m3,       1
-    movu              [r2],         xm3
-    movu              [r2 + r3],    xm4
-    lea               r2,           [r2 + r3 * 2]
-    lea               r0,           [r0 + r1 * 2]
-    dec               r4d
-    jnz               .loop
-    RET
-;--------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-    IPFILTER_LUMA 4, 4, pp
-    IPFILTER_LUMA 4, 8, pp
-    IPFILTER_LUMA 12, 16, pp
-    IPFILTER_LUMA 4, 16, pp
+    pshufb      m0,        [tab_Cm]
+    mova        m1,        [pw_512]
+    lea         r5,        [r0 + 4 * r1]
+    lea         r4,        [r1 * 3]
 
-INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_8x8, 4,6,6
-    mov               r4d,    r4m
+    movd        m2,        [r0]
+    movd        m3,        [r0 + r1]
+    movd        m4,        [r0 + 2 * r1]
+    movd        m5,        [r0 + r4]
 
-%ifdef PIC
-    lea               r5,           [tab_ChromaCoeff]
-    vpbroadcastd      m0,           [r5 + r4 * 4]
-%else
-    vpbroadcastd      m0,           [tab_ChromaCoeff + r4 * 4]
-%endif
+    punpcklbw   m2,        m3
+    punpcklbw   m6,        m4,        m5
+    punpcklbw   m2,        m6
 
-    movu              m1,           [tab_Tm]
-    vpbroadcastd      m2,           [pw_1]
+    pmaddubsw   m2,        m0
 
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
+    movd        m6,        [r5]
 
-    sub               r0,           1
-    mov               r4d,          2
+    punpcklbw   m3,        m4
+    punpcklbw   m7,        m5,        m6
+    punpcklbw   m3,        m7
 
-.loop:
-    ; Row 0
-    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-
-    ; Row 1
-    vbroadcasti128    m4,           [r0 + r1]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-    packssdw          m3,           m4
-    pmulhrsw          m3,           [pw_512]
-    lea               r0,           [r0 + r1 * 2]
-
-    ; Row 2
-    vbroadcasti128    m4,           [r0 ]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-
-    ; Row 3
-    vbroadcasti128    m5,           [r0 + r1]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m5,           m1
-    pmaddubsw         m5,           m0
-    pmaddwd           m5,           m2
-    packssdw          m4,           m5
-    pmulhrsw          m4,           [pw_512]
-
-    packuswb          m3,           m4
-    mova              m5,           [interp_4tap_8x8_horiz_shuf]
-    vpermd            m3,           m5,     m3
-    vextracti128      xm4,          m3,     1
-    movq              [r2],         xm3
-    movhps            [r2 + r3],    xm3
-    lea               r2,           [r2 + r3 * 2]
-    movq              [r2],         xm4
-    movhps            [r2 + r3],    xm4
-    lea               r2,           [r2 + r3 * 2]
-    lea               r0,           [r0 + r1*2]
-    dec               r4d
-    jnz               .loop
-    RET
+    pmaddubsw   m3,        m0
 
-    IPFILTER_LUMA_AVX2 16, 4
-    IPFILTER_LUMA_AVX2 16, 8
-    IPFILTER_LUMA_AVX2 16, 12
-    IPFILTER_LUMA_AVX2 16, 16
-    IPFILTER_LUMA_AVX2 16, 32
-    IPFILTER_LUMA_AVX2 16, 64
+    phaddw      m2,        m3
 
-    IPFILTER_LUMA_32x_avx2 32 , 8
-    IPFILTER_LUMA_32x_avx2 32 , 16
-    IPFILTER_LUMA_32x_avx2 32 , 24
-    IPFILTER_LUMA_32x_avx2 32 , 32
-    IPFILTER_LUMA_32x_avx2 32 , 64
+    pmulhrsw    m2,        m1
 
-    IPFILTER_LUMA_64x_avx2 64 , 64
-    IPFILTER_LUMA_64x_avx2 64 , 48
-    IPFILTER_LUMA_64x_avx2 64 , 32
-    IPFILTER_LUMA_64x_avx2 64 , 16
+    movd        m7,        [r5 + r1]
 
-INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_8x2, 4, 6, 5
-    mov               r4d,          r4m
+    punpcklbw   m4,        m5
+    punpcklbw   m3,        m6,        m7
+    punpcklbw   m4,        m3
 
-%ifdef PIC
-    lea               r5,           [tab_ChromaCoeff]
-    vpbroadcastd      m0,           [r5 + r4 * 4]
-%else
-    vpbroadcastd      m0,           [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    mova              m1,           [tab_Tm]
-    mova              m2,           [pw_1]
-
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-
-    dec               r0
-    ; Row 0
-    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-
-    ; Row 1
-    vbroadcasti128    m4,           [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-    packssdw          m3,           m4
-    pmulhrsw          m3,           [pw_512]
-    vextracti128      xm4,          m3,          1
-    packuswb          xm3,          xm4
-    pshufd            xm3,          xm3,         11011000b
-    movq              [r2],         xm3
-    movhps            [r2 + r3],    xm3
-    RET
+    pmaddubsw   m4,        m0
 
-INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_8x6, 4, 6, 7
-    mov               r4d,           r4m
+    movd        m3,        [r5 + 2 * r1]
 
-%ifdef PIC
-    lea               r5,            [tab_ChromaCoeff]
-    vpbroadcastd      m0,            [r5 + r4 * 4]
-%else
-    vpbroadcastd      m0,            [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    mova              m1,            [tab_Tm]
-    mova              m2,            [pw_1]
-    mova              m6,            [pw_512]
-    lea               r4,            [r1 * 3]
-    lea               r5,            [r3 * 3]
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-
-    dec               r0
-    ; Row 0
-    vbroadcasti128    m3,            [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,            m1
-    pmaddubsw         m3,            m0
-    pmaddwd           m3,            m2
-
-    ; Row 1
-    vbroadcasti128    m4,            [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,            m1
-    pmaddubsw         m4,            m0
-    pmaddwd           m4,            m2
-    packssdw          m3,            m4
-    pmulhrsw          m3,            m6
-
-    ; Row 2
-    vbroadcasti128    m4,            [r0 + r1 * 2]               ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,            m1
-    pmaddubsw         m4,            m0
-    pmaddwd           m4,            m2
-
-    ; Row 3
-    vbroadcasti128    m5,            [r0 + r4]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m5,            m1
-    pmaddubsw         m5,            m0
-    pmaddwd           m5,            m2
-    packssdw          m4,            m5
-    pmulhrsw          m4,            m6
-
-    packuswb          m3,            m4
-    mova              m5,            [interp8_hps_shuf]
-    vpermd            m3,            m5,          m3
-    vextracti128      xm4,           m3,          1
-    movq              [r2],          xm3
-    movhps            [r2 + r3],     xm3
-    movq              [r2 + r3 * 2], xm4
-    movhps            [r2 + r5],     xm4
-    lea               r2,            [r2 + r3 * 4]
-    lea               r0,            [r0 + r1 * 4]
-    ; Row 4
-    vbroadcasti128    m3,            [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,            m1
-    pmaddubsw         m3,            m0
-    pmaddwd           m3,            m2
-
-    ; Row 5
-    vbroadcasti128    m4,            [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,            m1
-    pmaddubsw         m4,            m0
-    pmaddwd           m4,            m2
-    packssdw          m3,            m4
-    pmulhrsw          m3,            m6
-    vextracti128      xm4,           m3,          1
-    packuswb          xm3,           xm4
-    pshufd            xm3,           xm3,         11011000b
-    movq              [r2],          xm3
-    movhps            [r2 + r3],     xm3
-    RET
+    punpcklbw   m5,        m6
+    punpcklbw   m7,        m3
+    punpcklbw   m5,        m7
 
-INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_6x8, 4, 6, 7
-    mov               r4d,               r4m
+    pmaddubsw   m5,        m0
 
-%ifdef PIC
-    lea               r5,                [tab_ChromaCoeff]
-    vpbroadcastd      m0,                [r5 + r4 * 4]
-%else
-    vpbroadcastd      m0,                [tab_ChromaCoeff + r4 * 4]
-%endif
+    phaddw      m4,        m5
 
-    mova              m1,                [tab_Tm]
-    mova              m2,                [pw_1]
-    mova              m6,                [pw_512]
-    lea               r4,                [r1 * 3]
-    lea               r5,                [r3 * 3]
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
+    pmulhrsw    m4,        m1
 
-    dec               r0
-%rep 2
-    ; Row 0
-    vbroadcasti128    m3,                [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,                m1
-    pmaddubsw         m3,                m0
-    pmaddwd           m3,                m2
-
-    ; Row 1
-    vbroadcasti128    m4,                [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,                m1
-    pmaddubsw         m4,                m0
-    pmaddwd           m4,                m2
-    packssdw          m3,                m4
-    pmulhrsw          m3,                m6
-
-    ; Row 2
-    vbroadcasti128    m4,                [r0 + r1 * 2]               ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,                m1
-    pmaddubsw         m4,                m0
-    pmaddwd           m4,                m2
-
-    ; Row 3
-    vbroadcasti128    m5,                [r0 + r4]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m5,                m1
-    pmaddubsw         m5,                m0
-    pmaddwd           m5,                m2
-    packssdw          m4,                m5
-    pmulhrsw          m4,                m6
-
-    packuswb          m3,                m4
-    vextracti128      xm4,               m3,          1
-    movd              [r2],              xm3
-    pextrw            [r2 + 4],          xm4,         0
-    pextrd            [r2 + r3],         xm3,         1
-    pextrw            [r2 + r3 + 4],     xm4,         2
-    pextrd            [r2 + r3 * 2],     xm3,         2
-    pextrw            [r2 + r3 * 2 + 4], xm4,         4
-    pextrd            [r2 + r5],         xm3,         3
-    pextrw            [r2 + r5 + 4],     xm4,         6
-    lea               r2,                [r2 + r3 * 4]
-    lea               r0,                [r0 + r1 * 4]
-%endrep
+    packuswb    m2,        m4
+    movd        [r2],      m2
+    pextrd      [r2 + r3], m2, 1
+    lea         r2,        [r2 + 2 * r3]
+    pextrd      [r2],      m2, 2
+    pextrd      [r2 + r3], m2, 3
     RET
-
-;-----------------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_ps_64xN(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;-----------------------------------------------------------------------------------------------------------------------------;
-%macro IPFILTER_CHROMA_HPS_64xN 1
+%macro FILTER_VER_CHROMA_AVX2_4x4 1
 INIT_YMM avx2
-cglobal interp_4tap_horiz_ps_64x%1, 4,7,6
+cglobal interp_4tap_vert_%1_4x4, 4, 6, 3
     mov             r4d, r4m
-    mov             r5d, r5m
-    add             r3d, r3d
+    shl             r4d, 6
+    sub             r0, r1
 
 %ifdef PIC
-    lea               r6,           [tab_ChromaCoeff]
-    vpbroadcastd      m0,           [r6 + r4 * 4]
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
 %else
-    vpbroadcastd      m0,           [tab_ChromaCoeff + r4 * 4]
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
 %endif
 
-    vbroadcasti128     m2,           [pw_1]
-    vbroadcasti128     m5,           [pw_2000]
-    mova               m1,           [tab_Tm]
+    lea             r4, [r1 * 3]
 
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-    mov                r6d,         %1
-    dec                r0
-    test                r5d,      r5d
-    je                 .loop
-    sub                r0 ,         r1
-    add                r6d ,        3
+    movd            xm1, [r0]
+    pinsrd          xm1, [r0 + r1], 1
+    pinsrd          xm1, [r0 + r1 * 2], 2
+    pinsrd          xm1, [r0 + r4], 3                       ; m1 = row[3 2 1 0]
+    lea             r0, [r0 + r1 * 4]
+    movd            xm2, [r0]
+    pinsrd          xm2, [r0 + r1], 1
+    pinsrd          xm2, [r0 + r1 * 2], 2                   ; m2 = row[x 6 5 4]
+    vinserti128     m1, m1, xm2, 1                          ; m1 = row[x 6 5 4 3 2 1 0]
+    mova            m2, [interp4_vpp_shuf1]
+    vpermd          m0, m2, m1                              ; m0 = row[4 3 3 2 2 1 1 0]
+    mova            m2, [interp4_vpp_shuf1 + mmsize]
+    vpermd          m1, m2, m1                              ; m1 = row[6 5 5 4 4 3 3 2]
 
-.loop:
-    ; Row 0
-    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-    vbroadcasti128    m4,           [r0 + 8]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-
-    packssdw          m3,           m4
-    psubw             m3,           m5
-    vpermq            m3,           m3,          11011000b
-    movu              [r2],         m3
-
-    vbroadcasti128    m3,           [r0 + 16]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-    vbroadcasti128    m4,           [r0 + 24]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-
-    packssdw          m3,           m4
-    psubw             m3,           m5
-    vpermq            m3,           m3,          11011000b
-    movu              [r2 + 32],    m3
-
-    vbroadcasti128    m3,           [r0 + 32]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-    vbroadcasti128    m4,           [r0 + 40]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-
-    packssdw          m3,           m4
-    psubw             m3,           m5
-    vpermq            m3,           m3,          11011000b
-    movu              [r2 + 64],    m3
-
-    vbroadcasti128    m3,           [r0 + 48]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-    vbroadcasti128    m4,           [r0 + 56]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-
-    packssdw          m3,           m4
-    psubw             m3,           m5
-    vpermq            m3,           m3,          11011000b
-    movu              [r2 + 96],    m3
-
-    add                r2,           r3
-    add                r0,           r1
-    dec                r6d
-    jnz                .loop
+    mova            m2, [interp4_vpp_shuf]
+    pshufb          m0, m0, m2
+    pshufb          m1, m1, m2
+    pmaddubsw       m0, [r5]
+    pmaddubsw       m1, [r5 + mmsize]
+    paddw           m0, m1                                  ; m0 = WORD ROW[3 2 1 0]
+%ifidn %1,pp
+    pmulhrsw        m0, [pw_512]
+    vextracti128    xm1, m0, 1
+    packuswb        xm0, xm1
+    lea             r5, [r3 * 3]
+    movd            [r2], xm0
+    pextrd          [r2 + r3], xm0, 1
+    pextrd          [r2 + r3 * 2], xm0, 2
+    pextrd          [r2 + r5], xm0, 3
+%else
+    add             r3d, r3d
+    psubw           m0, [pw_2000]
+    vextracti128    xm1, m0, 1
+    lea             r5, [r3 * 3]
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm1
+    movhps          [r2 + r5], xm1
+%endif
     RET
 %endmacro
+    FILTER_VER_CHROMA_AVX2_4x4 pp
+    FILTER_VER_CHROMA_AVX2_4x4 ps
 
-   IPFILTER_CHROMA_HPS_64xN 64
-   IPFILTER_CHROMA_HPS_64xN 32
-   IPFILTER_CHROMA_HPS_64xN 48
-   IPFILTER_CHROMA_HPS_64xN 16
-
-;-----------------------------------------------------------------------------------------------------------------------------
-;void interp_horiz_ps_c(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;-----------------------------------------------------------------------------------------------------------------------------
-
-%macro IPFILTER_LUMA_PS_4xN_AVX2 1
+%macro FILTER_VER_CHROMA_AVX2_4x8 1
 INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_horiz_ps_4x%1, 6,7,6
-    mov                         r5d,               r5m
-    mov                         r4d,               r4m
+cglobal interp_4tap_vert_%1_4x8, 4, 6, 5
+    mov             r4d, r4m
+    shl             r4d, 6
+    sub             r0, r1
+
 %ifdef PIC
-    lea                         r6,                [tab_LumaCoeff]
-    vpbroadcastq                m0,                [r6 + r4 * 8]
-%else
-    vpbroadcastq                m0,                [tab_LumaCoeff + r4 * 8]
-%endif
-    mova                        m1,                [tab_Lm]
-    add                         r3d,               r3d
-    vbroadcasti128              m2,                [pw_2000]
-
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - pw_2000
-
-    sub                         r0,                3
-    test                        r5d,               r5d
-    mov                         r5d,               %1                           ; loop count variable - height
-    jz                         .preloop
-    lea                         r6,                [r1 * 3]                     ; r8 = (N / 2 - 1) * srcStride
-    sub                         r0,                r6                           ; r0(src) - 3 * srcStride
-    add                         r5d,               7                            ; need extra 7 rows, just set a specially flag here, blkheight += N - 1  (7 - 3 = 4 ; since the last three rows not in loop)
-
-.preloop:
-    lea                         r6,                [r3 * 3]
-.loop:
-    ; Row 0-1
-    vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb                      m3,                m1                           ; shuffled based on the col order tab_Lm
-    pmaddubsw                   m3,                m0
-    vbroadcasti128              m4,                [r0 + r1]                    ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb                      m4,                m1
-    pmaddubsw                   m4,                m0
-    phaddw                      m3,                m4                           ; DWORD [R1D R1C R0D R0C R1B R1A R0B R0A]
-
-    ; Row 2-3
-    lea                         r0,                [r0 + r1 * 2]                ;3rd row(i.e 2nd row)
-    vbroadcasti128              m4,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb                      m4,                m1
-    pmaddubsw                   m4,                m0
-    vbroadcasti128              m5,                [r0 + r1]                    ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb                      m5,                m1
-    pmaddubsw                   m5,                m0
-    phaddw                      m4,                m5                           ; DWORD [R3D R3C R2D R2C R3B R3A R2B R2A]
-    phaddw                      m3,                m4                           ; all rows and col completed.
-
-    mova                        m5,                [interp8_hps_shuf]
-    vpermd                      m3,                m5,               m3
-    psubw                       m3,                m2
-
-    vextracti128                xm4,               m3,               1
-    movq                        [r2],              xm3                          ;row 0
-    movhps                      [r2 + r3],         xm3                          ;row 1
-    movq                        [r2 + r3 * 2],     xm4                          ;row 2
-    movhps                      [r2 + r6],         xm4                          ;row 3
-
-    lea                         r0,                [r0 + r1 * 2]                ; first loop src ->5th row(i.e 4)
-    lea                         r2,                [r2 + r3 * 4]                ; first loop dst ->5th row(i.e 4)
-    sub                         r5d,               4
-    jz                         .end
-    cmp                         r5d,               4
-    jge                        .loop
-
-    ; Row 8-9
-    vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb                      m3,                m1
-    pmaddubsw                   m3,                m0
-    vbroadcasti128              m4,                [r0 + r1]                    ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb                      m4,                m1
-    pmaddubsw                   m4,                m0
-    phaddw                      m3,                m4                           ; DWORD [R1D R1C R0D R0C R1B R1A R0B R0A]
-
-    ; Row 10
-    vbroadcasti128              m4,                [r0 + r1 * 2]                ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb                      m4,                m1
-    pmaddubsw                   m4,                m0
-    phaddw                      m4,                m4                           ; DWORD [R3D R3C R2D R2C R3B R3A R2B R2A]
-    phaddw                      m3,                m4
-
-    vpermd                      m3,                m5,            m3            ; m5 don't broken in above
-    psubw                       m3,                m2
-
-    vextracti128                xm4,               m3,            1
-    movq                        [r2],              xm3
-    movhps                      [r2 + r3],         xm3
-    movq                        [r2 + r3 * 2],     xm4
-.end:
-    RET
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
 %endif
-%endmacro
 
-    IPFILTER_LUMA_PS_4xN_AVX2 4
-    IPFILTER_LUMA_PS_4xN_AVX2 8
-    IPFILTER_LUMA_PS_4xN_AVX2 16
+    lea             r4, [r1 * 3]
 
-%macro IPFILTER_LUMA_PS_8xN_AVX2 1
-; TODO: verify and enable on X86 mode
-%if ARCH_X86_64 == 1
-; void filter_hps(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-INIT_YMM avx2
-cglobal interp_8tap_horiz_ps_8x%1, 4,7,6
-    mov                         r5d,        r5m
-    mov                         r4d,        r4m
-    shl                         r4d,        7
-%ifdef PIC
-    lea                         r6,         [pb_LumaCoeffVer]
-    add                         r6,         r4
-%else
-    lea                         r6,         [pb_LumaCoeffVer + r4]
-%endif
-    add                         r3d,        r3d
-    vpbroadcastd                m0,         [pw_2000]
-    sub                         r0,         3
-    lea                         r4,         [pb_8tap_hps_0]
-    vbroadcasti128              m5,         [r4 + 0 * mmsize]
-
-    ; check row count extend for interpolateHV
-    test                        r5d,        r5d;
-    mov                         r5d,        %1
-    jz                         .enter_loop
-    lea                         r4,         [r1 * 3]                        ; r8 = (N / 2 - 1) * srcStride
-    sub                         r0,         r4                              ; r0(src)-r8
-    add                         r5d,        8-1-2                           ; blkheight += N - 1  (7 - 3 = 4 ; since the last three rows not in loop)
-
-.enter_loop:
-    lea                         r4,         [pb_8tap_hps_0]
-
-    ; ***** register map *****
-    ; m0 - pw_2000
-    ; r4 - base pointer of shuffle order table
-    ; r5 - count of loop
-    ; r6 - point to LumaCoeff
-.loop:
+    movd            xm1, [r0]
+    pinsrd          xm1, [r0 + r1], 1
+    pinsrd          xm1, [r0 + r1 * 2], 2
+    pinsrd          xm1, [r0 + r4], 3                       ; m1 = row[3 2 1 0]
+    lea             r0, [r0 + r1 * 4]
+    movd            xm2, [r0]
+    pinsrd          xm2, [r0 + r1], 1
+    pinsrd          xm2, [r0 + r1 * 2], 2
+    pinsrd          xm2, [r0 + r4], 3                       ; m2 = row[7 6 5 4]
+    vinserti128     m1, m1, xm2, 1                          ; m1 = row[7 6 5 4 3 2 1 0]
+    lea             r0, [r0 + r1 * 4]
+    movd            xm3, [r0]
+    pinsrd          xm3, [r0 + r1], 1
+    pinsrd          xm3, [r0 + r1 * 2], 2                   ; m3 = row[x 10 9 8]
+    vinserti128     m2, m2, xm3, 1                          ; m2 = row[x 10 9 8 7 6 5 4]
+    mova            m3, [interp4_vpp_shuf1]
+    vpermd          m0, m3, m1                              ; m0 = row[4 3 3 2 2 1 1 0]
+    vpermd          m4, m3, m2                              ; m4 = row[8 7 7 6 6 5 5 4]
+    mova            m3, [interp4_vpp_shuf1 + mmsize]
+    vpermd          m1, m3, m1                              ; m1 = row[6 5 5 4 4 3 3 2]
+    vpermd          m2, m3, m2                              ; m2 = row[10 9 9 8 8 7 7 6]
 
-    ; Row 0-1
-    movu                        xm1,        [r0]
-    movu                        xm2,        [r0 + r1]
-    vinserti128                 m1,         m1,         xm2, 1
-    pshufb                      m2,         m1,         m5                  ; [0 1 1 2 2 3 3 4 ...]
-    pshufb                      m3,         m1,         [r4 + 1 * mmsize]   ; [2 3 3 4 4 5 5 6 ...]
-    pshufb                      m4,         m1,         [r4 + 2 * mmsize]   ; [4 5 5 6 6 7 7 8 ...]
-    pshufb                      m1,         m1,         [r4 + 3 * mmsize]   ; [6 7 7 8 8 9 9 A ...]
-    pmaddubsw                   m2,         [r6 + 0 * mmsize]
-    pmaddubsw                   m3,         [r6 + 1 * mmsize]
-    pmaddubsw                   m4,         [r6 + 2 * mmsize]
-    pmaddubsw                   m1,         [r6 + 3 * mmsize]
-    paddw                       m2,         m3
-    paddw                       m1,         m4
-    paddw                       m1,         m2
-    psubw                       m1,         m0
-
-    vextracti128                xm2,        m1,         1
-    movu                        [r2],       xm1                             ; row 0
-    movu                        [r2 + r3],  xm2                             ; row 1
-
-    lea                         r0,         [r0 + r1 * 2]                   ; first loop src ->5th row(i.e 4)
-    lea                         r2,         [r2 + r3 * 2]                   ; first loop dst ->5th row(i.e 4)
-    sub                         r5d,        2
-    jg                         .loop
-    jz                         .end
-
-    ; last row
-    movu                        xm1,        [r0]
-    pshufb                      xm2,        xm1,         xm5                ; [0 1 1 2 2 3 3 4 ...]
-    pshufb                      xm3,        xm1,         [r4 + 1 * mmsize]  ; [2 3 3 4 4 5 5 6 ...]
-    pshufb                      xm4,        xm1,         [r4 + 2 * mmsize]  ; [4 5 5 6 6 7 7 8 ...]
-    pshufb                      xm1,        xm1,         [r4 + 3 * mmsize]  ; [6 7 7 8 8 9 9 A ...]
-    pmaddubsw                   xm2,        [r6 + 0 * mmsize]
-    pmaddubsw                   xm3,        [r6 + 1 * mmsize]
-    pmaddubsw                   xm4,        [r6 + 2 * mmsize]
-    pmaddubsw                   xm1,        [r6 + 3 * mmsize]
-    paddw                       xm2,        xm3
-    paddw                       xm1,        xm4
-    paddw                       xm1,        xm2
-    psubw                       xm1,        xm0
-    movu                        [r2],       xm1                          ;row 0
-.end:
-    RET
+    mova            m3, [interp4_vpp_shuf]
+    pshufb          m0, m0, m3
+    pshufb          m1, m1, m3
+    pshufb          m2, m2, m3
+    pshufb          m4, m4, m3
+    pmaddubsw       m0, [r5]
+    pmaddubsw       m4, [r5]
+    pmaddubsw       m1, [r5 + mmsize]
+    pmaddubsw       m2, [r5 + mmsize]
+    paddw           m0, m1                                  ; m0 = WORD ROW[3 2 1 0]
+    paddw           m4, m2                                  ; m4 = WORD ROW[7 6 5 4]
+%ifidn %1,pp
+    pmulhrsw        m0, [pw_512]
+    pmulhrsw        m4, [pw_512]
+    packuswb        m0, m4
+    vextracti128    xm1, m0, 1
+    lea             r5, [r3 * 3]
+    movd            [r2], xm0
+    pextrd          [r2 + r3], xm0, 1
+    movd            [r2 + r3 * 2], xm1
+    pextrd          [r2 + r5], xm1, 1
+    lea             r2, [r2 + r3 * 4]
+    pextrd          [r2], xm0, 2
+    pextrd          [r2 + r3], xm0, 3
+    pextrd          [r2 + r3 * 2], xm1, 2
+    pextrd          [r2 + r5], xm1, 3
+%else
+    add             r3d, r3d
+    psubw           m0, [pw_2000]
+    psubw           m4, [pw_2000]
+    vextracti128    xm1, m0, 1
+    vextracti128    xm2, m4, 1
+    lea             r5, [r3 * 3]
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm1
+    movhps          [r2 + r5], xm1
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm4
+    movhps          [r2 + r3], xm4
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r5], xm2
 %endif
-%endmacro ; IPFILTER_LUMA_PS_8xN_AVX2
-
-    IPFILTER_LUMA_PS_8xN_AVX2  4
-    IPFILTER_LUMA_PS_8xN_AVX2  8
-    IPFILTER_LUMA_PS_8xN_AVX2 16
-    IPFILTER_LUMA_PS_8xN_AVX2 32
+    RET
+%endmacro
 
+    FILTER_VER_CHROMA_AVX2_4x8 pp
+    FILTER_VER_CHROMA_AVX2_4x8 ps
 
-%macro IPFILTER_LUMA_PS_16x_AVX2 2
-INIT_YMM avx2
+%macro FILTER_VER_CHROMA_AVX2_4xN 2
 %if ARCH_X86_64 == 1
-cglobal interp_8tap_horiz_ps_%1x%2, 6, 10, 7
-    mov                         r5d,               r5m
-    mov                         r4d,               r4m
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_4x%2, 4, 6, 12
+    mov             r4d, r4m
+    shl             r4d, 6
+    sub             r0, r1
+
 %ifdef PIC
-    lea                         r6,                [tab_LumaCoeff]
-    vpbroadcastq                m0,                [r6 + r4 * 8]
-%else
-    vpbroadcastq                m0,                [tab_LumaCoeff + r4 * 8]
-%endif
-    mova                        m6,                [tab_Lm + 32]
-    mova                        m1,                [tab_Lm]
-    mov                         r9,                %2                           ;height
-    add                         r3d,               r3d
-    vbroadcasti128              m2,                [pw_2000]
-
-    ; register map
-    ; m0      - interpolate coeff
-    ; m1 , m6 - shuffle order table
-    ; m2      - pw_2000
-
-    xor                         r7,                r7                          ; loop count variable
-    sub                         r0,                3
-    test                        r5d,               r5d
-    jz                          .label
-    lea                         r8,                [r1 * 3]                     ; r8 = (N / 2 - 1) * srcStride
-    sub                         r0,                r8                           ; r0(src)-r8
-    add                         r9,                7                            ; blkheight += N - 1  (7 - 1 = 6 ; since the last one row not in loop)
-
-.label:
-    ; Row 0
-    vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb                      m4,                m3,             m6           ; row 0 (col 4 to 7)
-    pshufb                      m3,                m1                           ; shuffled based on the col order tab_Lm row 0 (col 0 to 3)
-    pmaddubsw                   m3,                m0
-    pmaddubsw                   m4,                m0
-    phaddw                      m3,                m4                           ; DWORD [R1D R1C R0D R0C R1B R1A R0B R0A]
-
-    vbroadcasti128              m4,                [r0 + 8]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb                      m5,                m4,            m6            ;row 1 (col 4 to 7)
-    pshufb                      m4,                m1                           ;row 1 (col 0 to 3)
-    pmaddubsw                   m4,                m0
-    pmaddubsw                   m5,                m0
-    phaddw                      m4,                m5                           ; DWORD [R3D R3C R2D R2C R3B R3A R2B R2A]
-    phaddw                      m3,                m4                           ; all rows and col completed.
-
-    mova                        m5,                [interp8_hps_shuf]
-    vpermd                      m3,                m5,               m3
-    psubw                       m3,                m2
-
-    movu                        [r2],              m3                          ;row 0
-
-    lea                         r0,                [r0 + r1]                ; first loop src ->5th row(i.e 4)
-    lea                         r2,                [r2 + r3]                ; first loop dst ->5th row(i.e 4)
-    dec                         r9d
-    jnz                         .label
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    mova            m10, [r5]
+    mova            m11, [r5 + mmsize]
+%ifidn %1,pp
+    mova            m9, [pw_512]
+%else
+    add             r3d, r3d
+    mova            m9, [pw_2000]
+%endif
+    lea             r5, [r3 * 3]
+%rep %2 / 16
+    movd            xm1, [r0]
+    pinsrd          xm1, [r0 + r1], 1
+    pinsrd          xm1, [r0 + r1 * 2], 2
+    pinsrd          xm1, [r0 + r4], 3                       ; m1 = row[3 2 1 0]
+    lea             r0, [r0 + r1 * 4]
+    movd            xm2, [r0]
+    pinsrd          xm2, [r0 + r1], 1
+    pinsrd          xm2, [r0 + r1 * 2], 2
+    pinsrd          xm2, [r0 + r4], 3                       ; m2 = row[7 6 5 4]
+    vinserti128     m1, m1, xm2, 1                          ; m1 = row[7 6 5 4 3 2 1 0]
+    lea             r0, [r0 + r1 * 4]
+    movd            xm3, [r0]
+    pinsrd          xm3, [r0 + r1], 1
+    pinsrd          xm3, [r0 + r1 * 2], 2
+    pinsrd          xm3, [r0 + r4], 3                       ; m3 = row[11 10 9 8]
+    vinserti128     m2, m2, xm3, 1                          ; m2 = row[11 10 9 8 7 6 5 4]
+    lea             r0, [r0 + r1 * 4]
+    movd            xm4, [r0]
+    pinsrd          xm4, [r0 + r1], 1
+    pinsrd          xm4, [r0 + r1 * 2], 2
+    pinsrd          xm4, [r0 + r4], 3                       ; m4 = row[15 14 13 12]
+    vinserti128     m3, m3, xm4, 1                          ; m3 = row[15 14 13 12 11 10 9 8]
+    lea             r0, [r0 + r1 * 4]
+    movd            xm5, [r0]
+    pinsrd          xm5, [r0 + r1], 1
+    pinsrd          xm5, [r0 + r1 * 2], 2                   ; m5 = row[x 18 17 16]
+    vinserti128     m4, m4, xm5, 1                          ; m4 = row[x 18 17 16 15 14 13 12]
+    mova            m5, [interp4_vpp_shuf1]
+    vpermd          m0, m5, m1                              ; m0 = row[4 3 3 2 2 1 1 0]
+    vpermd          m6, m5, m2                              ; m6 = row[8 7 7 6 6 5 5 4]
+    vpermd          m7, m5, m3                              ; m7 = row[12 11 11 10 10 9 9 8]
+    vpermd          m8, m5, m4                              ; m8 = row[16 15 15 14 14 13 13 12]
+    mova            m5, [interp4_vpp_shuf1 + mmsize]
+    vpermd          m1, m5, m1                              ; m1 = row[6 5 5 4 4 3 3 2]
+    vpermd          m2, m5, m2                              ; m2 = row[10 9 9 8 8 7 7 6]
+    vpermd          m3, m5, m3                              ; m3 = row[14 13 13 12 12 11 11 10]
+    vpermd          m4, m5, m4                              ; m4 = row[18 17 17 16 16 15 15 14]
 
+    mova            m5, [interp4_vpp_shuf]
+    pshufb          m0, m0, m5
+    pshufb          m1, m1, m5
+    pshufb          m2, m2, m5
+    pshufb          m4, m4, m5
+    pshufb          m3, m3, m5
+    pshufb          m6, m6, m5
+    pshufb          m7, m7, m5
+    pshufb          m8, m8, m5
+    pmaddubsw       m0, m10
+    pmaddubsw       m6, m10
+    pmaddubsw       m7, m10
+    pmaddubsw       m8, m10
+    pmaddubsw       m1, m11
+    pmaddubsw       m2, m11
+    pmaddubsw       m3, m11
+    pmaddubsw       m4, m11
+    paddw           m0, m1                                  ; m0 = WORD ROW[3 2 1 0]
+    paddw           m6, m2                                  ; m6 = WORD ROW[7 6 5 4]
+    paddw           m7, m3                                  ; m7 = WORD ROW[11 10 9 8]
+    paddw           m8, m4                                  ; m8 = WORD ROW[15 14 13 12]
+%ifidn %1,pp
+    pmulhrsw        m0, m9
+    pmulhrsw        m6, m9
+    pmulhrsw        m7, m9
+    pmulhrsw        m8, m9
+    packuswb        m0, m6
+    packuswb        m7, m8
+    vextracti128    xm1, m0, 1
+    vextracti128    xm2, m7, 1
+    movd            [r2], xm0
+    pextrd          [r2 + r3], xm0, 1
+    movd            [r2 + r3 * 2], xm1
+    pextrd          [r2 + r5], xm1, 1
+    lea             r2, [r2 + r3 * 4]
+    pextrd          [r2], xm0, 2
+    pextrd          [r2 + r3], xm0, 3
+    pextrd          [r2 + r3 * 2], xm1, 2
+    pextrd          [r2 + r5], xm1, 3
+    lea             r2, [r2 + r3 * 4]
+    movd            [r2], xm7
+    pextrd          [r2 + r3], xm7, 1
+    movd            [r2 + r3 * 2], xm2
+    pextrd          [r2 + r5], xm2, 1
+    lea             r2, [r2 + r3 * 4]
+    pextrd          [r2], xm7, 2
+    pextrd          [r2 + r3], xm7, 3
+    pextrd          [r2 + r3 * 2], xm2, 2
+    pextrd          [r2 + r5], xm2, 3
+%else
+    psubw           m0, m9
+    psubw           m6, m9
+    psubw           m7, m9
+    psubw           m8, m9
+    vextracti128    xm1, m0, 1
+    vextracti128    xm2, m6, 1
+    vextracti128    xm3, m7, 1
+    vextracti128    xm4, m8, 1
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm1
+    movhps          [r2 + r5], xm1
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm6
+    movhps          [r2 + r3], xm6
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r5], xm2
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm7
+    movhps          [r2 + r3], xm7
+    movq            [r2 + r3 * 2], xm3
+    movhps          [r2 + r5], xm3
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm8
+    movhps          [r2 + r3], xm8
+    movq            [r2 + r3 * 2], xm4
+    movhps          [r2 + r5], xm4
+%endif
+    lea             r2, [r2 + r3 * 4]
+%endrep
     RET
 %endif
 %endmacro
 
+    FILTER_VER_CHROMA_AVX2_4xN pp, 16
+    FILTER_VER_CHROMA_AVX2_4xN ps, 16
+    FILTER_VER_CHROMA_AVX2_4xN pp, 32
+    FILTER_VER_CHROMA_AVX2_4xN ps, 32
 
-    IPFILTER_LUMA_PS_16x_AVX2 16 , 16
-    IPFILTER_LUMA_PS_16x_AVX2 16 , 8
-    IPFILTER_LUMA_PS_16x_AVX2 16 , 12
-    IPFILTER_LUMA_PS_16x_AVX2 16 , 4
-    IPFILTER_LUMA_PS_16x_AVX2 16 , 32
-    IPFILTER_LUMA_PS_16x_AVX2 16 , 64
-
-
-;--------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-%macro IPFILTER_LUMA_PP_W8 2
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W4_H4 2
 INIT_XMM sse4
-cglobal interp_8tap_horiz_pp_%1x%2, 4,6,7
-    mov         r4d, r4m
+cglobal interp_4tap_vert_pp_%1x%2, 4, 6, 8
+
+    mov         r4d,       r4m
+    sub         r0,        r1
 
 %ifdef PIC
-    lea         r5, [tab_LumaCoeff]
-    movh        m3, [r5 + r4 * 8]
+    lea         r5,        [tab_ChromaCoeff]
+    movd        m0,        [r5 + r4 * 4]
 %else
-    movh        m3, [tab_LumaCoeff + r4 * 8]
+    movd        m0,        [tab_ChromaCoeff + r4 * 4]
 %endif
-    pshufd      m0, m3, 0                       ; m0 = coeff-L
-    pshufd      m1, m3, 0x55                    ; m1 = coeff-H
-    lea         r5, [tab_Tm]                    ; r5 = shuffle
-    mova        m2, [pw_512]                    ; m2 = 512
 
-    mov         r4d, %2
-.loopH:
-%assign x 0
-%rep %1 / 8
-    movu        m3, [r0 - 3 + x]                ; m3 = [F E D C B A 9 8 7 6 5 4 3 2 1 0]
-    pshufb      m4, m3, [r5 + 0*16]             ; m4 = [6 5 4 3 5 4 3 2 4 3 2 1 3 2 1 0]
-    pshufb      m5, m3, [r5 + 1*16]             ; m5 = [A 9 8 7 9 8 7 6 8 7 6 5 7 6 5 4]
-    pshufb          m3, [r5 + 2*16]             ; m3 = [E D C B D C B A C B A 9 B A 9 8]
-    pmaddubsw   m4, m0
-    pmaddubsw   m6, m5, m1
-    pmaddubsw   m5, m0
-    pmaddubsw   m3, m1
-    paddw       m4, m6
-    paddw       m5, m3
-    phaddw      m4, m5
-    pmulhrsw    m4, m2
-    packuswb    m4, m4
-    movh        [r2 + x], m4
-%assign x x+8
-%endrep
-
-    add       r0, r1
-    add       r2, r3
-
-    dec       r4d
-    jnz      .loopH
-    RET
-%endmacro
-
-    IPFILTER_LUMA_PP_W8      8,  4
-    IPFILTER_LUMA_PP_W8      8,  8
-    IPFILTER_LUMA_PP_W8      8, 16
-    IPFILTER_LUMA_PP_W8      8, 32
-    IPFILTER_LUMA_PP_W8     16,  4
-    IPFILTER_LUMA_PP_W8     16,  8
-    IPFILTER_LUMA_PP_W8     16, 12
-    IPFILTER_LUMA_PP_W8     16, 16
-    IPFILTER_LUMA_PP_W8     16, 32
-    IPFILTER_LUMA_PP_W8     16, 64
-    IPFILTER_LUMA_PP_W8     24, 32
-    IPFILTER_LUMA_PP_W8     32,  8
-    IPFILTER_LUMA_PP_W8     32, 16
-    IPFILTER_LUMA_PP_W8     32, 24
-    IPFILTER_LUMA_PP_W8     32, 32
-    IPFILTER_LUMA_PP_W8     32, 64
-    IPFILTER_LUMA_PP_W8     48, 64
-    IPFILTER_LUMA_PP_W8     64, 16
-    IPFILTER_LUMA_PP_W8     64, 32
-    IPFILTER_LUMA_PP_W8     64, 48
-    IPFILTER_LUMA_PP_W8     64, 64
-
-;----------------------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_ps_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;----------------------------------------------------------------------------------------------------------------------------
-    IPFILTER_LUMA 4, 4, ps
-    IPFILTER_LUMA 8, 8, ps
-    IPFILTER_LUMA 8, 4, ps
-    IPFILTER_LUMA 4, 8, ps
-    IPFILTER_LUMA 16, 16, ps
-    IPFILTER_LUMA 16, 8, ps
-    IPFILTER_LUMA 8, 16, ps
-    IPFILTER_LUMA 16, 12, ps
-    IPFILTER_LUMA 12, 16, ps
-    IPFILTER_LUMA 16, 4, ps
-    IPFILTER_LUMA 4, 16, ps
-    IPFILTER_LUMA 32, 32, ps
-    IPFILTER_LUMA 32, 16, ps
-    IPFILTER_LUMA 16, 32, ps
-    IPFILTER_LUMA 32, 24, ps
-    IPFILTER_LUMA 24, 32, ps
-    IPFILTER_LUMA 32, 8, ps
-    IPFILTER_LUMA 8, 32, ps
-    IPFILTER_LUMA 64, 64, ps
-    IPFILTER_LUMA 64, 32, ps
-    IPFILTER_LUMA 32, 64, ps
-    IPFILTER_LUMA 64, 48, ps
-    IPFILTER_LUMA 48, 64, ps
-    IPFILTER_LUMA 64, 16, ps
-    IPFILTER_LUMA 16, 64, ps
-
-;-----------------------------------------------------------------------------
-; Interpolate HV
-;-----------------------------------------------------------------------------
-%macro FILTER_HV8_START 7 ; (t0, t1, t2, t3, t4, off_src, off_coeff) -> (t3, t5), (t4, t1), [2]
-    mova        %5, [r0 +  (%6 + 0) * 16]
-    mova        %1, [r0 +  (%6 + 1) * 16]
-    mova        %2, [r0 +  (%6 + 2) * 16]
-    punpcklwd   %3, %5, %1
-    punpckhwd   %5, %1
-    pmaddwd     %3, [r5 + (%7) * 16]   ; R3 = L[0+1] -- Row 0
-    pmaddwd     %5, [r5 + (%7) * 16]   ; R0 = H[0+1]
-    punpcklwd   %4, %1, %2
-    punpckhwd   %1, %2
-    pmaddwd     %4, [r5 + (%7) * 16]   ; R4 = L[1+2] -- Row 1
-    pmaddwd     %1, [r5 + (%7) * 16]   ; R1 = H[1+2]
-%endmacro ; FILTER_HV8_START
-
-%macro FILTER_HV8_MID 10 ; (Row3, prevRow, sum0L, sum1L, sum0H, sum1H, t6, t7, off_src, off_coeff) -> [6]
-    mova        %8, [r0 +  (%9 + 0) * 16]
-    mova        %1, [r0 +  (%9 + 1) * 16]
-    punpcklwd   %7, %2, %8
-    punpckhwd   %2, %8
-    pmaddwd     %7, [r5 + %10 * 16]
-    pmaddwd     %2, [r5 + %10 * 16]
-    paddd       %3, %7              ; R3 = L[0+1+2+3] -- Row 0
-    paddd       %5, %2              ; R0 = H[0+1+2+3]
-    punpcklwd   %7, %8, %1
-    punpckhwd   %8, %1
-    pmaddwd     %7, [r5 + %10 * 16]
-    pmaddwd     %8, [r5 + %10 * 16]
-    paddd       %4, %7              ; R4 = L[1+2+3+4] -- Row 1
-    paddd       %6, %8              ; R1 = H[1+2+3+4]
-%endmacro ; FILTER_HV8_MID
-
-; Round and Saturate
-%macro FILTER_HV8_END 4 ; output in [1, 3]
-    paddd       %1, [pd_526336]
-    paddd       %2, [pd_526336]
-    paddd       %3, [pd_526336]
-    paddd       %4, [pd_526336]
-    psrad       %1, 12
-    psrad       %2, 12
-    psrad       %3, 12
-    psrad       %4, 12
-    packssdw    %1, %2
-    packssdw    %3, %4
-
-    ; TODO: is merge better? I think this way is short dependency link
-    packuswb    %1, %3
-%endmacro ; FILTER_HV8_END
-
-;-----------------------------------------------------------------------------
-; void interp_8tap_hv_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int idxX, int idxY)
-;-----------------------------------------------------------------------------
-INIT_XMM ssse3
-cglobal interp_8tap_hv_pp_8x8, 4, 7, 8, 0-15*16
-%define coef        m7
-%define stk_buf     rsp
-
-    mov         r4d,        r4m
-    mov         r5d,        r5m
-
-%ifdef PIC
-    lea         r6,         [tab_LumaCoeff]
-    movh        coef,       [r6 + r4 * 8]
-%else
-    movh        coef,       [tab_LumaCoeff + r4 * 8]
-%endif
-    punpcklqdq  coef,       coef
-
-    ; move to row -3
-    lea         r6,         [r1 + r1 * 2]
-    sub         r0,         r6
-
-    xor         r6,         r6
-    mov         r4,         rsp
-
-.loopH:
-    FILTER_H8_W8 m0, m1, m2, m3, coef, [pw_512], [r0 - 3]
-    psubw       m1,         [pw_2000]
-    mova        [r4],       m1
-
-    add         r0,         r1
-    add         r4,         16
-    inc         r6
-    cmp         r6,         8+7
-    jnz         .loopH
-
-    ; ready to phase V
-    ; Here all of mN is free
-
-    ; load coeff table
-    shl         r5,         6
-    lea         r6,         [tab_LumaCoeffV]
-    lea         r5,         [r5 + r6]
-
-    ; load intermedia buffer
-    mov         r0,         stk_buf
-
-    ; register mapping
-    ; r0 - src
-    ; r5 - coeff
-    ; r6 - loop_i
-
-    ; let's go
-    xor         r6,         r6
-
-    ; TODO: this loop have more than 70 instructions, I think it is more than Intel loop decode cache
-.loopV:
-
-    FILTER_HV8_START    m1, m2, m3, m4, m0,             0, 0
-    FILTER_HV8_MID      m6, m2, m3, m4, m0, m1, m7, m5, 3, 1
-    FILTER_HV8_MID      m5, m6, m3, m4, m0, m1, m7, m2, 5, 2
-    FILTER_HV8_MID      m6, m5, m3, m4, m0, m1, m7, m2, 7, 3
-    FILTER_HV8_END      m3, m0, m4, m1
-
-    movh        [r2],       m3
-    movhps      [r2 + r3],  m3
-
-    lea         r0,         [r0 + 16 * 2]
-    lea         r2,         [r2 + r3 * 2]
-
-    inc         r6
-    cmp         r6,         8/2
-    jnz         .loopV
-
-    RET
-
-;-----------------------------------------------------------------------------
-; void interp_8tap_hv_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int idxX, int idxY)
-;-----------------------------------------------------------------------------
-INIT_XMM sse3
-cglobal interp_8tap_hv_pp_8x8, 4, 7, 8, 0-15*16
-    mov         r4d,        r4m
-    mov         r5d,        r5m
-    add         r4d,        r4d
-    pxor        m6,         m6
-
-%ifdef PIC
-    lea         r6,         [tabw_LumaCoeff]
-    mova        m3,         [r6 + r4 * 8]
-%else
-    mova        m3,         [tabw_LumaCoeff + r4 * 8]
-%endif
-
-    ; move to row -3
-    lea         r6,         [r1 + r1 * 2]
-    sub         r0,         r6
-
-    mov         r4,         rsp
-
-%assign x 0     ;needed for FILTER_H8_W8_sse2 macro
-%assign y 1
-%rep 15
-    FILTER_H8_W8_sse2
-    psubw       m1,         [pw_2000]
-    mova        [r4],       m1
-
-%if y < 15
-    add         r0,         r1
-    add         r4,         16
-%endif
-%assign y y+1
-%endrep
-
-    ; ready to phase V
-    ; Here all of mN is free
-
-    ; load coeff table
-    shl         r5,         6
-    lea         r6,         [tab_LumaCoeffV]
-    lea         r5,         [r5 + r6]
-
-    ; load intermedia buffer
-    mov         r0,         rsp
-
-    ; register mapping
-    ; r0 - src
-    ; r5 - coeff
-
-    ; let's go
-%assign y 1
-%rep 4
-    FILTER_HV8_START    m1, m2, m3, m4, m0,             0, 0
-    FILTER_HV8_MID      m6, m2, m3, m4, m0, m1, m7, m5, 3, 1
-    FILTER_HV8_MID      m5, m6, m3, m4, m0, m1, m7, m2, 5, 2
-    FILTER_HV8_MID      m6, m5, m3, m4, m0, m1, m7, m2, 7, 3
-    FILTER_HV8_END      m3, m0, m4, m1
-
-    movh        [r2],       m3
-    movhps      [r2 + r3],  m3
-
-%if y < 4
-    lea         r0,         [r0 + 16 * 2]
-    lea         r2,         [r2 + r3 * 2]
-%endif
-%assign y y+1
-%endrep
-    RET
+    pshufb      m0,        [tab_Cm]
 
-;-----------------------------------------------------------------------------
-;void interp_4tap_vert_pp_2x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-INIT_XMM sse4
-cglobal interp_4tap_vert_pp_2x4, 4, 6, 8
+    mova        m1,        [pw_512]
 
-    mov         r4d,       r4m
-    sub         r0,        r1
+    mov         r4d,       %2
 
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeff]
-    movd        m0,        [r5 + r4 * 4]
-%else
-    movd        m0,        [tab_ChromaCoeff + r4 * 4]
-%endif
-    lea         r4,        [r1 * 3]
-    lea         r5,        [r0 + 4 * r1]
-    pshufb      m0,        [tab_Cm]
-    mova        m1,        [pw_512]
+    lea         r5,        [3 * r1]
 
+.loop:
     movd        m2,        [r0]
     movd        m3,        [r0 + r1]
     movd        m4,        [r0 + 2 * r1]
-    movd        m5,        [r0 + r4]
+    movd        m5,        [r0 + r5]
 
     punpcklbw   m2,        m3
     punpcklbw   m6,        m4,        m5
@@ -5874,7 +3965,8 @@ cglobal interp_4tap_vert_pp_2x4, 4, 6, 8
 
     pmaddubsw   m2,        m0
 
-    movd        m6,        [r5]
+    lea         r0,        [r0 + 4 * r1]
+    movd        m6,        [r0]
 
     punpcklbw   m3,        m4
     punpcklbw   m7,        m5,        m6
@@ -5886,7 +3978,7 @@ cglobal interp_4tap_vert_pp_2x4, 4, 6, 8
 
     pmulhrsw    m2,        m1
 
-    movd        m7,        [r5 + r1]
+    movd        m7,        [r0 + r1]
 
     punpcklbw   m4,        m5
     punpcklbw   m3,        m6,        m7
@@ -5894,7 +3986,7 @@ cglobal interp_4tap_vert_pp_2x4, 4, 6, 8
 
     pmaddubsw   m4,        m0
 
-    movd        m3,        [r5 + 2 * r1]
+    movd        m3,        [r0 + 2 * r1]
 
     punpcklbw   m5,        m6
     punpcklbw   m7,        m3
@@ -5906,1113 +3998,834 @@ cglobal interp_4tap_vert_pp_2x4, 4, 6, 8
 
     pmulhrsw    m4,        m1
     packuswb    m2,        m4
+    movd        [r2],      m2
+    pextrd      [r2 + r3], m2,  1
+    lea         r2,        [r2 + 2 * r3]
+    pextrd      [r2],      m2, 2
+    pextrd      [r2 + r3], m2, 3
 
-    pextrw      [r2],      m2, 0
-    pextrw      [r2 + r3], m2, 2
     lea         r2,        [r2 + 2 * r3]
-    pextrw      [r2],      m2, 4
-    pextrw      [r2 + r3], m2, 6
 
+    sub         r4,        4
+    jnz        .loop
     RET
+%endmacro
 
-%macro FILTER_VER_CHROMA_AVX2_2x4 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_2x4, 4, 6, 2
-    mov             r4d, r4m
-    shl             r4d, 5
-    sub             r0, r1
+    FILTER_V4_W4_H4 4,  8
+    FILTER_V4_W4_H4 4, 16
 
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeff_V]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeff_V + r4]
-%endif
+    FILTER_V4_W4_H4 4, 32
 
-    lea             r4, [r1 * 3]
+%macro FILTER_V4_W8_H2 0
+    punpcklbw   m1,        m2
+    punpcklbw   m7,        m3,        m0
 
-    pinsrw          xm1, [r0], 0
-    pinsrw          xm1, [r0 + r1], 1
-    pinsrw          xm1, [r0 + r1 * 2], 2
-    pinsrw          xm1, [r0 + r4], 3
-    lea             r0, [r0 + r1 * 4]
-    pinsrw          xm1, [r0], 4
-    pinsrw          xm1, [r0 + r1], 5
-    pinsrw          xm1, [r0 + r1 * 2], 6
+    pmaddubsw   m1,        m6
+    pmaddubsw   m7,        m5
 
-    pshufb          xm0, xm1, [interp_vert_shuf]
-    pshufb          xm1, [interp_vert_shuf + 32]
-    vinserti128     m0, m0, xm1, 1
-    pmaddubsw       m0, [r5]
-    vextracti128    xm1, m0, 1
-    paddw           xm0, xm1
-%ifidn %1,pp
-    pmulhrsw        xm0, [pw_512]
-    packuswb        xm0, xm0
-    lea             r4, [r3 * 3]
-    pextrw          [r2], xm0, 0
-    pextrw          [r2 + r3], xm0, 1
-    pextrw          [r2 + r3 * 2], xm0, 2
-    pextrw          [r2 + r4], xm0, 3
-%else
-    add             r3d, r3d
-    lea             r4, [r3 * 3]
-    psubw           xm0, [pw_2000]
-    movd            [r2], xm0
-    pextrd          [r2 + r3], xm0, 1
-    pextrd          [r2 + r3 * 2], xm0, 2
-    pextrd          [r2 + r4], xm0, 3
-%endif
-    RET
+    paddw       m1,        m7
+
+    pmulhrsw    m1,        m4
+    packuswb    m1,        m1
 %endmacro
 
-    FILTER_VER_CHROMA_AVX2_2x4 pp
-    FILTER_VER_CHROMA_AVX2_2x4 ps
+%macro FILTER_V4_W8_H3 0
+    punpcklbw   m2,        m3
+    punpcklbw   m7,        m0,        m1
 
-%macro FILTER_VER_CHROMA_AVX2_2x8 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_2x8, 4, 6, 2
-    mov             r4d, r4m
-    shl             r4d, 6
-    sub             r0, r1
+    pmaddubsw   m2,        m6
+    pmaddubsw   m7,        m5
 
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
+    paddw       m2,        m7
 
-    lea             r4, [r1 * 3]
+    pmulhrsw    m2,        m4
+    packuswb    m2,        m2
+%endmacro
 
-    pinsrw          xm1, [r0], 0
-    pinsrw          xm1, [r0 + r1], 1
-    pinsrw          xm1, [r0 + r1 * 2], 2
-    pinsrw          xm1, [r0 + r4], 3
-    lea             r0, [r0 + r1 * 4]
-    pinsrw          xm1, [r0], 4
-    pinsrw          xm1, [r0 + r1], 5
-    pinsrw          xm1, [r0 + r1 * 2], 6
-    pinsrw          xm1, [r0 + r4], 7
-    movhlps         xm0, xm1
-    lea             r0, [r0 + r1 * 4]
-    pinsrw          xm0, [r0], 4
-    pinsrw          xm0, [r0 + r1], 5
-    pinsrw          xm0, [r0 + r1 * 2], 6
-    vinserti128     m1, m1, xm0, 1
+%macro FILTER_V4_W8_H4 0
+    punpcklbw   m3,        m0
+    punpcklbw   m7,        m1,        m2
 
-    pshufb          m0, m1, [interp_vert_shuf]
-    pshufb          m1, [interp_vert_shuf + 32]
-    pmaddubsw       m0, [r5]
-    pmaddubsw       m1, [r5 + 1 * mmsize]
-    paddw           m0, m1
-%ifidn %1,pp
-    pmulhrsw        m0, [pw_512]
-    vextracti128    xm1, m0, 1
-    packuswb        xm0, xm1
-    lea             r4, [r3 * 3]
-    pextrw          [r2], xm0, 0
-    pextrw          [r2 + r3], xm0, 1
-    pextrw          [r2 + r3 * 2], xm0, 2
-    pextrw          [r2 + r4], xm0, 3
-    lea             r2, [r2 + r3 * 4]
-    pextrw          [r2], xm0, 4
-    pextrw          [r2 + r3], xm0, 5
-    pextrw          [r2 + r3 * 2], xm0, 6
-    pextrw          [r2 + r4], xm0, 7
-%else
-    add             r3d, r3d
-    lea             r4, [r3 * 3]
-    psubw           m0, [pw_2000]
-    vextracti128    xm1, m0, 1
-    movd            [r2], xm0
-    pextrd          [r2 + r3], xm0, 1
-    pextrd          [r2 + r3 * 2], xm0, 2
-    pextrd          [r2 + r4], xm0, 3
-    lea             r2, [r2 + r3 * 4]
-    movd            [r2], xm1
-    pextrd          [r2 + r3], xm1, 1
-    pextrd          [r2 + r3 * 2], xm1, 2
-    pextrd          [r2 + r4], xm1, 3
-%endif
-    RET
+    pmaddubsw   m3,        m6
+    pmaddubsw   m7,        m5
+
+    paddw       m3,        m7
+
+    pmulhrsw    m3,        m4
+    packuswb    m3,        m3
 %endmacro
 
-    FILTER_VER_CHROMA_AVX2_2x8 pp
-    FILTER_VER_CHROMA_AVX2_2x8 ps
+%macro FILTER_V4_W8_H5 0
+    punpcklbw   m0,        m1
+    punpcklbw   m7,        m2,        m3
 
-%macro FILTER_VER_CHROMA_AVX2_2x16 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_2x16, 4, 6, 3
-    mov             r4d, r4m
-    shl             r4d, 6
-    sub             r0,  r1
+    pmaddubsw   m0,        m6
+    pmaddubsw   m7,        m5
+
+    paddw       m0,        m7
+
+    pmulhrsw    m0,        m4
+    packuswb    m0,        m0
+%endmacro
+
+%macro FILTER_V4_W8_8x2 2
+    FILTER_V4_W8 %1, %2
+    movq        m0,        [r0 + 4 * r1]
+
+    FILTER_V4_W8_H2
+
+    movh        [r2 + r3], m1
+%endmacro
+
+%macro FILTER_V4_W8_8x4 2
+    FILTER_V4_W8_8x2 %1, %2
+;8x3
+    lea         r6,        [r0 + 4 * r1]
+    movq        m1,        [r6 + r1]
+
+    FILTER_V4_W8_H3
+
+    movh        [r2 + 2 * r3], m2
+
+;8x4
+    movq        m2,        [r6 + 2 * r1]
+
+    FILTER_V4_W8_H4
+
+    lea         r5,        [r2 + 2 * r3]
+    movh        [r5 + r3], m3
+%endmacro
+
+%macro FILTER_V4_W8_8x6 2
+    FILTER_V4_W8_8x4 %1, %2
+;8x5
+    lea         r6,        [r6 + 2 * r1]
+    movq        m3,        [r6 + r1]
+
+    FILTER_V4_W8_H5
+
+    movh        [r2 + 4 * r3], m0
+
+;8x6
+    movq        m0,        [r0 + 8 * r1]
+
+    FILTER_V4_W8_H2
+
+    lea         r5,        [r2 + 4 * r3]
+    movh        [r5 + r3], m1
+%endmacro
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W8 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_pp_%1x%2, 4, 7, 8
+
+    mov         r4d,       r4m
+
+    sub         r0,        r1
+    movq        m0,        [r0]
+    movq        m1,        [r0 + r1]
+    movq        m2,        [r0 + 2 * r1]
+    lea         r5,        [r0 + 2 * r1]
+    movq        m3,        [r5 + r1]
+
+    punpcklbw   m0,        m1
+    punpcklbw   m4,        m2,          m3
 
 %ifdef PIC
-    lea             r5,  [tab_ChromaCoeffVer_32]
-    add             r5,  r4
+    lea         r6,        [tab_ChromaCoeff]
+    movd        m5,        [r6 + r4 * 4]
 %else
-    lea             r5,  [tab_ChromaCoeffVer_32 + r4]
+    movd        m5,        [tab_ChromaCoeff + r4 * 4]
 %endif
 
-    lea             r4,  [r1 * 3]
+    pshufb      m6,        m5,       [tab_Vm]
+    pmaddubsw   m0,        m6
 
-    movd            xm1, [r0]
-    pinsrw          xm1, [r0 + r1], 1
-    pinsrw          xm1, [r0 + r1 * 2], 2
-    pinsrw          xm1, [r0 + r4], 3
-    lea             r0,  [r0 + r1 * 4]
-    pinsrw          xm1, [r0], 4
-    pinsrw          xm1, [r0 + r1], 5
-    pinsrw          xm1, [r0 + r1 * 2], 6
-    pinsrw          xm1, [r0 + r4], 7
-    lea             r0,  [r0 + r1 * 4]
-    pinsrw          xm0, [r0], 4
-    pinsrw          xm0, [r0 + r1], 5
-    pinsrw          xm0, [r0 + r1 * 2], 6
-    pinsrw          xm0, [r0 + r4], 7
-    punpckhqdq      xm0, xm1, xm0
-    vinserti128     m1,  m1,  xm0,  1
+    pshufb      m5,        [tab_Vm + 16]
+    pmaddubsw   m4,        m5
 
-    pshufb          m2,  m1,  [interp_vert_shuf]
-    pshufb          m1,  [interp_vert_shuf + 32]
-    pmaddubsw       m2,  [r5]
-    pmaddubsw       m1,  [r5 + 1 * mmsize]
-    paddw           m2,  m1
+    paddw       m0,        m4
 
-    lea             r0,  [r0 + r1 * 4]
-    pinsrw          xm1, [r0], 4
-    pinsrw          xm1, [r0 + r1], 5
-    pinsrw          xm1, [r0 + r1 * 2], 6
-    pinsrw          xm1, [r0 + r4], 7
-    punpckhqdq      xm1, xm0, xm1
-    lea             r0,  [r0 + r1 * 4]
-    pinsrw          xm0, [r0], 4
-    pinsrw          xm0, [r0 + r1], 5
-    pinsrw          xm0, [r0 + r1 * 2], 6
-    punpckhqdq      xm0, xm1, xm0
-    vinserti128     m1,  m1,  xm0,  1
+    mova        m4,        [pw_512]
 
-    pshufb          m0,  m1,  [interp_vert_shuf]
-    pshufb          m1,  [interp_vert_shuf + 32]
-    pmaddubsw       m0,  [r5]
-    pmaddubsw       m1,  [r5 + 1 * mmsize]
-    paddw           m0,  m1
-%ifidn %1,pp
-    mova            m1,  [pw_512]
-    pmulhrsw        m2,  m1
-    pmulhrsw        m0,  m1
-    packuswb        m2,  m0
-    lea             r4,  [r3 * 3]
-    pextrw          [r2], xm2, 0
-    pextrw          [r2 + r3], xm2, 1
-    pextrw          [r2 + r3 * 2], xm2, 2
-    pextrw          [r2 + r4], xm2, 3
-    vextracti128    xm0, m2, 1
-    lea             r2,  [r2 + r3 * 4]
-    pextrw          [r2], xm0, 0
-    pextrw          [r2 + r3], xm0, 1
-    pextrw          [r2 + r3 * 2], xm0, 2
-    pextrw          [r2 + r4], xm0, 3
-    lea             r2,  [r2 + r3 * 4]
-    pextrw          [r2], xm2, 4
-    pextrw          [r2 + r3], xm2, 5
-    pextrw          [r2 + r3 * 2], xm2, 6
-    pextrw          [r2 + r4], xm2, 7
-    lea             r2,  [r2 + r3 * 4]
-    pextrw          [r2], xm0, 4
-    pextrw          [r2 + r3], xm0, 5
-    pextrw          [r2 + r3 * 2], xm0, 6
-    pextrw          [r2 + r4], xm0, 7
-%else
-    add             r3d, r3d
-    lea             r4,  [r3 * 3]
-    vbroadcasti128  m1,  [pw_2000]
-    psubw           m2,  m1
-    psubw           m0,  m1
-    vextracti128    xm1, m2, 1
-    movd            [r2], xm2
-    pextrd          [r2 + r3], xm2, 1
-    pextrd          [r2 + r3 * 2], xm2, 2
-    pextrd          [r2 + r4], xm2, 3
-    lea             r2, [r2 + r3 * 4]
-    movd            [r2], xm1
-    pextrd          [r2 + r3], xm1, 1
-    pextrd          [r2 + r3 * 2], xm1, 2
-    pextrd          [r2 + r4], xm1, 3
-    vextracti128    xm1, m0, 1
-    lea             r2,  [r2 + r3 * 4]
-    movd            [r2], xm0
-    pextrd          [r2 + r3], xm0, 1
-    pextrd          [r2 + r3 * 2], xm0, 2
-    pextrd          [r2 + r4], xm0, 3
-    lea             r2,  [r2 + r3 * 4]
-    movd            [r2], xm1
-    pextrd          [r2 + r3], xm1, 1
-    pextrd          [r2 + r3 * 2], xm1, 2
-    pextrd          [r2 + r4], xm1, 3
-%endif
-    RET
+    pmulhrsw    m0,        m4
+    packuswb    m0,        m0
+    movh        [r2],      m0
 %endmacro
 
-    FILTER_VER_CHROMA_AVX2_2x16 pp
-    FILTER_VER_CHROMA_AVX2_2x16 ps
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_pp_8x2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+    FILTER_V4_W8_8x2 8, 2
+
+    RET
 
 ;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_2x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+; void interp_4tap_vert_pp_8x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
 ;-----------------------------------------------------------------------------
-%macro FILTER_V4_W2_H4 2
+    FILTER_V4_W8_8x4 8, 4
+
+    RET
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_pp_8x6(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+    FILTER_V4_W8_8x6 8, 6
+
+    RET
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ps_4x2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
 INIT_XMM sse4
-cglobal interp_4tap_vert_pp_2x%2, 4, 6, 8
+cglobal interp_4tap_vert_ps_4x2, 4, 6, 6
 
-    mov         r4d,       r4m
-    sub         r0,        r1
+    mov         r4d, r4m
+    sub         r0, r1
+    add         r3d, r3d
 
 %ifdef PIC
-    lea         r5,        [tab_ChromaCoeff]
-    movd        m0,        [r5 + r4 * 4]
+    lea         r5, [tab_ChromaCoeff]
+    movd        m0, [r5 + r4 * 4]
 %else
-    movd        m0,        [tab_ChromaCoeff + r4 * 4]
+    movd        m0, [tab_ChromaCoeff + r4 * 4]
 %endif
 
-    pshufb      m0,        [tab_Cm]
+    pshufb      m0, [tab_Cm]
 
-    mova        m1,        [pw_512]
+    movd        m2, [r0]
+    movd        m3, [r0 + r1]
+    lea         r5, [r0 + 2 * r1]
+    movd        m4, [r5]
+    movd        m5, [r5 + r1]
 
-    mov         r4d,       %2
-    lea         r5,        [3 * r1]
+    punpcklbw   m2, m3
+    punpcklbw   m1, m4, m5
+    punpcklbw   m2, m1
 
-.loop:
-    movd        m2,        [r0]
-    movd        m3,        [r0 + r1]
-    movd        m4,        [r0 + 2 * r1]
-    movd        m5,        [r0 + r5]
+    pmaddubsw   m2, m0
 
-    punpcklbw   m2,        m3
-    punpcklbw   m6,        m4,        m5
-    punpcklbw   m2,        m6
+    movd        m1, [r0 + 4 * r1]
 
-    pmaddubsw   m2,        m0
+    punpcklbw   m3, m4
+    punpcklbw   m5, m1
+    punpcklbw   m3, m5
 
-    lea         r0,        [r0 + 4 * r1]
-    movd        m6,        [r0]
+    pmaddubsw   m3, m0
 
-    punpcklbw   m3,        m4
-    punpcklbw   m7,        m5,        m6
-    punpcklbw   m3,        m7
+    phaddw      m2, m3
 
-    pmaddubsw   m3,        m0
+    psubw       m2, [pw_2000]
+    movh        [r2], m2
+    movhps      [r2 + r3], m2
 
-    phaddw      m2,        m3
+    RET
 
-    pmulhrsw    m2,        m1
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ps_4x4(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+INIT_XMM sse4
+cglobal interp_4tap_vert_ps_4x4, 4, 6, 7
 
-    movd        m7,        [r0 + r1]
+    mov        r4d, r4m
+    sub        r0, r1
+    add        r3d, r3d
 
-    punpcklbw   m4,        m5
-    punpcklbw   m3,        m6,        m7
-    punpcklbw   m4,        m3
+%ifdef PIC
+    lea        r5, [tab_ChromaCoeff]
+    movd       m0, [r5 + r4 * 4]
+%else
+    movd       m0, [tab_ChromaCoeff + r4 * 4]
+%endif
 
-    pmaddubsw   m4,        m0
+    pshufb     m0, [tab_Cm]
 
-    movd        m3,        [r0 + 2 * r1]
+    lea        r4, [r1 * 3]
+    lea        r5, [r0 + 4 * r1]
 
-    punpcklbw   m5,        m6
-    punpcklbw   m7,        m3
-    punpcklbw   m5,        m7
+    movd       m2, [r0]
+    movd       m3, [r0 + r1]
+    movd       m4, [r0 + 2 * r1]
+    movd       m5, [r0 + r4]
 
-    pmaddubsw   m5,        m0
+    punpcklbw  m2, m3
+    punpcklbw  m6, m4, m5
+    punpcklbw  m2, m6
 
-    phaddw      m4,        m5
+    pmaddubsw  m2, m0
 
-    pmulhrsw    m4,        m1
-    packuswb    m2,        m4
+    movd       m6, [r5]
 
-    pextrw      [r2],      m2, 0
-    pextrw      [r2 + r3], m2, 2
-    lea         r2,        [r2 + 2 * r3]
-    pextrw      [r2],      m2, 4
-    pextrw      [r2 + r3], m2, 6
+    punpcklbw  m3, m4
+    punpcklbw  m1, m5, m6
+    punpcklbw  m3, m1
 
-    lea         r2,        [r2 + 2 * r3]
+    pmaddubsw  m3, m0
 
-    sub         r4,        4
-    jnz        .loop
-    RET
-%endmacro
+    phaddw     m2, m3
 
-    FILTER_V4_W2_H4 2, 8
+    mova       m1, [pw_2000]
 
-    FILTER_V4_W2_H4 2, 16
+    psubw      m2, m1
+    movh       [r2], m2
+    movhps     [r2 + r3], m2
 
-;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_4x2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-INIT_XMM sse4
-cglobal interp_4tap_vert_pp_4x2, 4, 6, 6
+    movd       m2, [r5 + r1]
 
-    mov         r4d,       r4m
-    sub         r0,        r1
+    punpcklbw  m4, m5
+    punpcklbw  m3, m6, m2
+    punpcklbw  m4, m3
 
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeff]
-    movd        m0,        [r5 + r4 * 4]
-%else
-    movd        m0,        [tab_ChromaCoeff + r4 * 4]
-%endif
+    pmaddubsw  m4, m0
 
-    pshufb      m0,        [tab_Cm]
-    lea         r5,        [r0 + 2 * r1]
+    movd       m3, [r5 + 2 * r1]
 
-    movd        m2,        [r0]
-    movd        m3,        [r0 + r1]
-    movd        m4,        [r5]
-    movd        m5,        [r5 + r1]
+    punpcklbw  m5, m6
+    punpcklbw  m2, m3
+    punpcklbw  m5, m2
 
-    punpcklbw   m2,        m3
-    punpcklbw   m1,        m4,        m5
-    punpcklbw   m2,        m1
+    pmaddubsw  m5, m0
 
-    pmaddubsw   m2,        m0
+    phaddw     m4, m5
 
-    movd        m1,        [r0 + 4 * r1]
-
-    punpcklbw   m3,        m4
-    punpcklbw   m5,        m1
-    punpcklbw   m3,        m5
-
-    pmaddubsw   m3,        m0
-
-    phaddw      m2,        m3
-
-    pmulhrsw    m2,        [pw_512]
-    packuswb    m2,        m2
-    movd        [r2],      m2
-    pextrd      [r2 + r3], m2,  1
+    psubw      m4, m1
+    lea        r2, [r2 + 2 * r3]
+    movh       [r2], m4
+    movhps     [r2 + r3], m4
 
     RET
 
-%macro FILTER_VER_CHROMA_AVX2_4x2 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_4x2, 4, 6, 4
-    mov             r4d, r4m
-    shl             r4d, 5
-    sub             r0, r1
+;---------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ps_%1x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;---------------------------------------------------------------------------------------------------------------
+%macro FILTER_V_PS_W4_H4 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_ps_%1x%2, 4, 6, 8
+
+    mov        r4d, r4m
+    sub        r0, r1
+    add        r3d, r3d
 
 %ifdef PIC
-    lea             r5, [tab_ChromaCoeff_V]
-    add             r5, r4
+    lea        r5, [tab_ChromaCoeff]
+    movd       m0, [r5 + r4 * 4]
 %else
-    lea             r5, [tab_ChromaCoeff_V + r4]
+    movd       m0, [tab_ChromaCoeff + r4 * 4]
 %endif
 
-    lea             r4, [r1 * 3]
+    pshufb     m0, [tab_Cm]
 
-    movd            xm1, [r0]
-    movd            xm2, [r0 + r1]
-    punpcklbw       xm1, xm2
-    movd            xm3, [r0 + r1 * 2]
-    punpcklbw       xm2, xm3
-    movlhps         xm1, xm2
-    movd            xm0, [r0 + r4]
-    punpcklbw       xm3, xm0
-    movd            xm2, [r0 + r1 * 4]
-    punpcklbw       xm0, xm2
-    movlhps         xm3, xm0
-    vinserti128     m1, m1, xm3, 1                          ; m1 = row[x x x 4 3 2 1 0]
+    mova       m1, [pw_2000]
 
-    pmaddubsw       m1, [r5]
-    vextracti128    xm3, m1, 1
-    paddw           xm1, xm3
-%ifidn %1,pp
-    pmulhrsw        xm1, [pw_512]
-    packuswb        xm1, xm1
-    movd            [r2], xm1
-    pextrd          [r2 + r3], xm1, 1
-%else
-    add             r3d, r3d
-    psubw           xm1, [pw_2000]
-    movq            [r2], xm1
-    movhps          [r2 + r3], xm1
-%endif
+    mov        r4d, %2/4
+    lea        r5, [3 * r1]
+
+.loop:
+    movd       m2, [r0]
+    movd       m3, [r0 + r1]
+    movd       m4, [r0 + 2 * r1]
+    movd       m5, [r0 + r5]
+
+    punpcklbw  m2, m3
+    punpcklbw  m6, m4, m5
+    punpcklbw  m2, m6
+
+    pmaddubsw  m2, m0
+
+    lea        r0, [r0 + 4 * r1]
+    movd       m6, [r0]
+
+    punpcklbw  m3, m4
+    punpcklbw  m7, m5, m6
+    punpcklbw  m3, m7
+
+    pmaddubsw  m3, m0
+
+    phaddw     m2, m3
+
+    psubw      m2, m1
+    movh       [r2], m2
+    movhps     [r2 + r3], m2
+
+    movd       m2, [r0 + r1]
+
+    punpcklbw  m4, m5
+    punpcklbw  m3, m6, m2
+    punpcklbw  m4, m3
+
+    pmaddubsw  m4, m0
+
+    movd       m3, [r0 + 2 * r1]
+
+    punpcklbw  m5, m6
+    punpcklbw  m2, m3
+    punpcklbw  m5, m2
+
+    pmaddubsw  m5, m0
+
+    phaddw     m4, m5
+
+    psubw      m4, m1
+    lea        r2, [r2 + 2 * r3]
+    movh       [r2], m4
+    movhps     [r2 + r3], m4
+
+    lea        r2, [r2 + 2 * r3]
+
+    dec        r4d
+    jnz        .loop
     RET
 %endmacro
 
-    FILTER_VER_CHROMA_AVX2_4x2 pp
-    FILTER_VER_CHROMA_AVX2_4x2 ps
+    FILTER_V_PS_W4_H4 4, 8
+    FILTER_V_PS_W4_H4 4, 16
 
-;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_4x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
+    FILTER_V_PS_W4_H4 4, 32
+
+;--------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ps_8x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+%macro FILTER_V_PS_W8_H8_H16_H2 2
 INIT_XMM sse4
-cglobal interp_4tap_vert_pp_4x4, 4, 6, 8
+cglobal interp_4tap_vert_ps_%1x%2, 4, 6, 7
 
-    mov         r4d,       r4m
-    sub         r0,        r1
+    mov        r4d, r4m
+    sub        r0, r1
+    add        r3d, r3d
 
 %ifdef PIC
-    lea         r5,        [tab_ChromaCoeff]
-    movd        m0,        [r5 + r4 * 4]
+    lea        r5, [tab_ChromaCoeff]
+    movd       m5, [r5 + r4 * 4]
 %else
-    movd        m0,        [tab_ChromaCoeff + r4 * 4]
+    movd       m5, [tab_ChromaCoeff + r4 * 4]
 %endif
 
-    pshufb      m0,        [tab_Cm]
-    mova        m1,        [pw_512]
-    lea         r5,        [r0 + 4 * r1]
-    lea         r4,        [r1 * 3]
+    pshufb     m6, m5, [tab_Vm]
+    pshufb     m5, [tab_Vm + 16]
+    mova       m4, [pw_2000]
 
-    movd        m2,        [r0]
-    movd        m3,        [r0 + r1]
-    movd        m4,        [r0 + 2 * r1]
-    movd        m5,        [r0 + r4]
+    mov        r4d, %2/2
+    lea        r5, [3 * r1]
 
-    punpcklbw   m2,        m3
-    punpcklbw   m6,        m4,        m5
-    punpcklbw   m2,        m6
+.loopH:
+    movq       m0, [r0]
+    movq       m1, [r0 + r1]
+    movq       m2, [r0 + 2 * r1]
+    movq       m3, [r0 + r5]
 
-    pmaddubsw   m2,        m0
+    punpcklbw  m0, m1
+    punpcklbw  m1, m2
+    punpcklbw  m2, m3
 
-    movd        m6,        [r5]
+    pmaddubsw  m0, m6
+    pmaddubsw  m2, m5
 
-    punpcklbw   m3,        m4
-    punpcklbw   m7,        m5,        m6
-    punpcklbw   m3,        m7
+    paddw      m0, m2
 
-    pmaddubsw   m3,        m0
+    psubw      m0, m4
+    movu       [r2], m0
 
-    phaddw      m2,        m3
+    movq       m0, [r0 + 4 * r1]
 
-    pmulhrsw    m2,        m1
+    punpcklbw  m3, m0
 
-    movd        m7,        [r5 + r1]
+    pmaddubsw  m1, m6
+    pmaddubsw  m3, m5
 
-    punpcklbw   m4,        m5
-    punpcklbw   m3,        m6,        m7
-    punpcklbw   m4,        m3
+    paddw      m1, m3
+    psubw      m1, m4
 
-    pmaddubsw   m4,        m0
+    movu       [r2 + r3], m1
 
-    movd        m3,        [r5 + 2 * r1]
+    lea        r0, [r0 + 2 * r1]
+    lea        r2, [r2 + 2 * r3]
 
-    punpcklbw   m5,        m6
-    punpcklbw   m7,        m3
-    punpcklbw   m5,        m7
+    dec        r4d
+    jnz       .loopH
 
-    pmaddubsw   m5,        m0
+    RET
+%endmacro
 
-    phaddw      m4,        m5
+    FILTER_V_PS_W8_H8_H16_H2 8, 2
+    FILTER_V_PS_W8_H8_H16_H2 8, 4
+    FILTER_V_PS_W8_H8_H16_H2 8, 6
 
-    pmulhrsw    m4,        m1
+    FILTER_V_PS_W8_H8_H16_H2 8, 12
+    FILTER_V_PS_W8_H8_H16_H2 8, 64
 
-    packuswb    m2,        m4
-    movd        [r2],      m2
-    pextrd      [r2 + r3], m2, 1
-    lea         r2,        [r2 + 2 * r3]
-    pextrd      [r2],      m2, 2
-    pextrd      [r2 + r3], m2, 3
-    RET
-%macro FILTER_VER_CHROMA_AVX2_4x4 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_4x4, 4, 6, 3
-    mov             r4d, r4m
-    shl             r4d, 6
-    sub             r0, r1
+;--------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ps_8x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+%macro FILTER_V_PS_W8_H8_H16_H32 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_ps_%1x%2, 4, 6, 8
+
+    mov        r4d, r4m
+    sub        r0, r1
+    add        r3d, r3d
 
 %ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
+    lea        r5, [tab_ChromaCoeff]
+    movd       m5, [r5 + r4 * 4]
 %else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+    movd       m5, [tab_ChromaCoeff + r4 * 4]
 %endif
 
-    lea             r4, [r1 * 3]
+    pshufb     m6, m5, [tab_Vm]
+    pshufb     m5, [tab_Vm + 16]
+    mova       m4, [pw_2000]
 
-    movd            xm1, [r0]
-    pinsrd          xm1, [r0 + r1], 1
-    pinsrd          xm1, [r0 + r1 * 2], 2
-    pinsrd          xm1, [r0 + r4], 3                       ; m1 = row[3 2 1 0]
-    lea             r0, [r0 + r1 * 4]
-    movd            xm2, [r0]
-    pinsrd          xm2, [r0 + r1], 1
-    pinsrd          xm2, [r0 + r1 * 2], 2                   ; m2 = row[x 6 5 4]
-    vinserti128     m1, m1, xm2, 1                          ; m1 = row[x 6 5 4 3 2 1 0]
-    mova            m2, [interp4_vpp_shuf1]
-    vpermd          m0, m2, m1                              ; m0 = row[4 3 3 2 2 1 1 0]
-    mova            m2, [interp4_vpp_shuf1 + mmsize]
-    vpermd          m1, m2, m1                              ; m1 = row[6 5 5 4 4 3 3 2]
+    mov        r4d, %2/4
+    lea        r5, [3 * r1]
 
-    mova            m2, [interp4_vpp_shuf]
-    pshufb          m0, m0, m2
-    pshufb          m1, m1, m2
-    pmaddubsw       m0, [r5]
-    pmaddubsw       m1, [r5 + mmsize]
-    paddw           m0, m1                                  ; m0 = WORD ROW[3 2 1 0]
-%ifidn %1,pp
-    pmulhrsw        m0, [pw_512]
-    vextracti128    xm1, m0, 1
-    packuswb        xm0, xm1
-    lea             r5, [r3 * 3]
-    movd            [r2], xm0
-    pextrd          [r2 + r3], xm0, 1
-    pextrd          [r2 + r3 * 2], xm0, 2
-    pextrd          [r2 + r5], xm0, 3
-%else
-    add             r3d, r3d
-    psubw           m0, [pw_2000]
-    vextracti128    xm1, m0, 1
-    lea             r5, [r3 * 3]
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm1
-    movhps          [r2 + r5], xm1
-%endif
-    RET
-%endmacro
-    FILTER_VER_CHROMA_AVX2_4x4 pp
-    FILTER_VER_CHROMA_AVX2_4x4 ps
+.loop:
+    movq       m0, [r0]
+    movq       m1, [r0 + r1]
+    movq       m2, [r0 + 2 * r1]
+    movq       m3, [r0 + r5]
 
-%macro FILTER_VER_CHROMA_AVX2_4x8 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_4x8, 4, 6, 5
-    mov             r4d, r4m
-    shl             r4d, 6
-    sub             r0, r1
+    punpcklbw  m0, m1
+    punpcklbw  m1, m2
+    punpcklbw  m2, m3
 
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
+    pmaddubsw  m0, m6
+    pmaddubsw  m7, m2, m5
 
-    lea             r4, [r1 * 3]
+    paddw      m0, m7
 
-    movd            xm1, [r0]
-    pinsrd          xm1, [r0 + r1], 1
-    pinsrd          xm1, [r0 + r1 * 2], 2
-    pinsrd          xm1, [r0 + r4], 3                       ; m1 = row[3 2 1 0]
-    lea             r0, [r0 + r1 * 4]
-    movd            xm2, [r0]
-    pinsrd          xm2, [r0 + r1], 1
-    pinsrd          xm2, [r0 + r1 * 2], 2
-    pinsrd          xm2, [r0 + r4], 3                       ; m2 = row[7 6 5 4]
-    vinserti128     m1, m1, xm2, 1                          ; m1 = row[7 6 5 4 3 2 1 0]
-    lea             r0, [r0 + r1 * 4]
-    movd            xm3, [r0]
-    pinsrd          xm3, [r0 + r1], 1
-    pinsrd          xm3, [r0 + r1 * 2], 2                   ; m3 = row[x 10 9 8]
-    vinserti128     m2, m2, xm3, 1                          ; m2 = row[x 10 9 8 7 6 5 4]
-    mova            m3, [interp4_vpp_shuf1]
-    vpermd          m0, m3, m1                              ; m0 = row[4 3 3 2 2 1 1 0]
-    vpermd          m4, m3, m2                              ; m4 = row[8 7 7 6 6 5 5 4]
-    mova            m3, [interp4_vpp_shuf1 + mmsize]
-    vpermd          m1, m3, m1                              ; m1 = row[6 5 5 4 4 3 3 2]
-    vpermd          m2, m3, m2                              ; m2 = row[10 9 9 8 8 7 7 6]
+    psubw       m0, m4
+    movu       [r2], m0
 
-    mova            m3, [interp4_vpp_shuf]
-    pshufb          m0, m0, m3
-    pshufb          m1, m1, m3
-    pshufb          m2, m2, m3
-    pshufb          m4, m4, m3
-    pmaddubsw       m0, [r5]
-    pmaddubsw       m4, [r5]
-    pmaddubsw       m1, [r5 + mmsize]
-    pmaddubsw       m2, [r5 + mmsize]
-    paddw           m0, m1                                  ; m0 = WORD ROW[3 2 1 0]
-    paddw           m4, m2                                  ; m4 = WORD ROW[7 6 5 4]
-%ifidn %1,pp
-    pmulhrsw        m0, [pw_512]
-    pmulhrsw        m4, [pw_512]
-    packuswb        m0, m4
-    vextracti128    xm1, m0, 1
-    lea             r5, [r3 * 3]
-    movd            [r2], xm0
-    pextrd          [r2 + r3], xm0, 1
-    movd            [r2 + r3 * 2], xm1
-    pextrd          [r2 + r5], xm1, 1
-    lea             r2, [r2 + r3 * 4]
-    pextrd          [r2], xm0, 2
-    pextrd          [r2 + r3], xm0, 3
-    pextrd          [r2 + r3 * 2], xm1, 2
-    pextrd          [r2 + r5], xm1, 3
-%else
-    add             r3d, r3d
-    psubw           m0, [pw_2000]
-    psubw           m4, [pw_2000]
-    vextracti128    xm1, m0, 1
-    vextracti128    xm2, m4, 1
-    lea             r5, [r3 * 3]
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm1
-    movhps          [r2 + r5], xm1
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm4
-    movhps          [r2 + r3], xm4
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r5], xm2
-%endif
+    lea        r0, [r0 + 4 * r1]
+    movq       m0, [r0]
+
+    punpcklbw  m3, m0
+
+    pmaddubsw  m1, m6
+    pmaddubsw  m7, m3, m5
+
+    paddw      m1, m7
+
+    psubw      m1, m4
+    movu       [r2 + r3], m1
+
+    movq       m1, [r0 + r1]
+
+    punpcklbw  m0, m1
+
+    pmaddubsw  m2, m6
+    pmaddubsw  m0, m5
+
+    paddw      m2, m0
+
+    psubw      m2, m4
+    lea        r2, [r2 + 2 * r3]
+    movu       [r2], m2
+
+    movq       m2, [r0 + 2 * r1]
+
+    punpcklbw  m1, m2
+
+    pmaddubsw  m3, m6
+    pmaddubsw  m1, m5
+
+    paddw      m3, m1
+    psubw      m3, m4
+
+    movu       [r2 + r3], m3
+
+    lea        r2, [r2 + 2 * r3]
+
+    dec        r4d
+    jnz        .loop
     RET
 %endmacro
 
-    FILTER_VER_CHROMA_AVX2_4x8 pp
-    FILTER_VER_CHROMA_AVX2_4x8 ps
+    FILTER_V_PS_W8_H8_H16_H32 8,  8
+    FILTER_V_PS_W8_H8_H16_H32 8, 16
+    FILTER_V_PS_W8_H8_H16_H32 8, 32
 
-%macro FILTER_VER_CHROMA_AVX2_4xN 2
-%if ARCH_X86_64 == 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_4x%2, 4, 6, 12
-    mov             r4d, r4m
-    shl             r4d, 6
-    sub             r0, r1
+;------------------------------------------------------------------------------------------------------------
+;void interp_4tap_vert_ps_6x8(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;------------------------------------------------------------------------------------------------------------
+%macro FILTER_V_PS_W6 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_ps_6x%2, 4, 6, 8
+
+    mov        r4d, r4m
+    sub        r0, r1
+    add        r3d, r3d
 
 %ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
+    lea        r5, [tab_ChromaCoeff]
+    movd       m5, [r5 + r4 * 4]
 %else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+    movd       m5, [tab_ChromaCoeff + r4 * 4]
 %endif
 
-    lea             r4, [r1 * 3]
-    mova            m10, [r5]
-    mova            m11, [r5 + mmsize]
-%ifidn %1,pp
-    mova            m9, [pw_512]
-%else
-    add             r3d, r3d
-    mova            m9, [pw_2000]
-%endif
-    lea             r5, [r3 * 3]
-%rep %2 / 16
-    movd            xm1, [r0]
-    pinsrd          xm1, [r0 + r1], 1
-    pinsrd          xm1, [r0 + r1 * 2], 2
-    pinsrd          xm1, [r0 + r4], 3                       ; m1 = row[3 2 1 0]
-    lea             r0, [r0 + r1 * 4]
-    movd            xm2, [r0]
-    pinsrd          xm2, [r0 + r1], 1
-    pinsrd          xm2, [r0 + r1 * 2], 2
-    pinsrd          xm2, [r0 + r4], 3                       ; m2 = row[7 6 5 4]
-    vinserti128     m1, m1, xm2, 1                          ; m1 = row[7 6 5 4 3 2 1 0]
-    lea             r0, [r0 + r1 * 4]
-    movd            xm3, [r0]
-    pinsrd          xm3, [r0 + r1], 1
-    pinsrd          xm3, [r0 + r1 * 2], 2
-    pinsrd          xm3, [r0 + r4], 3                       ; m3 = row[11 10 9 8]
-    vinserti128     m2, m2, xm3, 1                          ; m2 = row[11 10 9 8 7 6 5 4]
-    lea             r0, [r0 + r1 * 4]
-    movd            xm4, [r0]
-    pinsrd          xm4, [r0 + r1], 1
-    pinsrd          xm4, [r0 + r1 * 2], 2
-    pinsrd          xm4, [r0 + r4], 3                       ; m4 = row[15 14 13 12]
-    vinserti128     m3, m3, xm4, 1                          ; m3 = row[15 14 13 12 11 10 9 8]
-    lea             r0, [r0 + r1 * 4]
-    movd            xm5, [r0]
-    pinsrd          xm5, [r0 + r1], 1
-    pinsrd          xm5, [r0 + r1 * 2], 2                   ; m5 = row[x 18 17 16]
-    vinserti128     m4, m4, xm5, 1                          ; m4 = row[x 18 17 16 15 14 13 12]
-    mova            m5, [interp4_vpp_shuf1]
-    vpermd          m0, m5, m1                              ; m0 = row[4 3 3 2 2 1 1 0]
-    vpermd          m6, m5, m2                              ; m6 = row[8 7 7 6 6 5 5 4]
-    vpermd          m7, m5, m3                              ; m7 = row[12 11 11 10 10 9 9 8]
-    vpermd          m8, m5, m4                              ; m8 = row[16 15 15 14 14 13 13 12]
-    mova            m5, [interp4_vpp_shuf1 + mmsize]
-    vpermd          m1, m5, m1                              ; m1 = row[6 5 5 4 4 3 3 2]
-    vpermd          m2, m5, m2                              ; m2 = row[10 9 9 8 8 7 7 6]
-    vpermd          m3, m5, m3                              ; m3 = row[14 13 13 12 12 11 11 10]
-    vpermd          m4, m5, m4                              ; m4 = row[18 17 17 16 16 15 15 14]
+    pshufb     m6, m5, [tab_Vm]
+    pshufb     m5, [tab_Vm + 16]
+    mova       m4, [pw_2000]
+    lea        r5, [3 * r1]
+    mov        r4d, %2/4
 
-    mova            m5, [interp4_vpp_shuf]
-    pshufb          m0, m0, m5
-    pshufb          m1, m1, m5
-    pshufb          m2, m2, m5
-    pshufb          m4, m4, m5
-    pshufb          m3, m3, m5
-    pshufb          m6, m6, m5
-    pshufb          m7, m7, m5
-    pshufb          m8, m8, m5
-    pmaddubsw       m0, m10
-    pmaddubsw       m6, m10
-    pmaddubsw       m7, m10
-    pmaddubsw       m8, m10
-    pmaddubsw       m1, m11
-    pmaddubsw       m2, m11
-    pmaddubsw       m3, m11
-    pmaddubsw       m4, m11
-    paddw           m0, m1                                  ; m0 = WORD ROW[3 2 1 0]
-    paddw           m6, m2                                  ; m6 = WORD ROW[7 6 5 4]
-    paddw           m7, m3                                  ; m7 = WORD ROW[11 10 9 8]
-    paddw           m8, m4                                  ; m8 = WORD ROW[15 14 13 12]
-%ifidn %1,pp
-    pmulhrsw        m0, m9
-    pmulhrsw        m6, m9
-    pmulhrsw        m7, m9
-    pmulhrsw        m8, m9
-    packuswb        m0, m6
-    packuswb        m7, m8
-    vextracti128    xm1, m0, 1
-    vextracti128    xm2, m7, 1
-    movd            [r2], xm0
-    pextrd          [r2 + r3], xm0, 1
-    movd            [r2 + r3 * 2], xm1
-    pextrd          [r2 + r5], xm1, 1
-    lea             r2, [r2 + r3 * 4]
-    pextrd          [r2], xm0, 2
-    pextrd          [r2 + r3], xm0, 3
-    pextrd          [r2 + r3 * 2], xm1, 2
-    pextrd          [r2 + r5], xm1, 3
-    lea             r2, [r2 + r3 * 4]
-    movd            [r2], xm7
-    pextrd          [r2 + r3], xm7, 1
-    movd            [r2 + r3 * 2], xm2
-    pextrd          [r2 + r5], xm2, 1
-    lea             r2, [r2 + r3 * 4]
-    pextrd          [r2], xm7, 2
-    pextrd          [r2 + r3], xm7, 3
-    pextrd          [r2 + r3 * 2], xm2, 2
-    pextrd          [r2 + r5], xm2, 3
-%else
-    psubw           m0, m9
-    psubw           m6, m9
-    psubw           m7, m9
-    psubw           m8, m9
-    vextracti128    xm1, m0, 1
-    vextracti128    xm2, m6, 1
-    vextracti128    xm3, m7, 1
-    vextracti128    xm4, m8, 1
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm1
-    movhps          [r2 + r5], xm1
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm6
-    movhps          [r2 + r3], xm6
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r5], xm2
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm7
-    movhps          [r2 + r3], xm7
-    movq            [r2 + r3 * 2], xm3
-    movhps          [r2 + r5], xm3
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm8
-    movhps          [r2 + r3], xm8
-    movq            [r2 + r3 * 2], xm4
-    movhps          [r2 + r5], xm4
-%endif
-    lea             r2, [r2 + r3 * 4]
-%endrep
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_CHROMA_AVX2_4xN pp, 16
-    FILTER_VER_CHROMA_AVX2_4xN ps, 16
-    FILTER_VER_CHROMA_AVX2_4xN pp, 32
-    FILTER_VER_CHROMA_AVX2_4xN ps, 32
+.loop:
+    movq       m0, [r0]
+    movq       m1, [r0 + r1]
+    movq       m2, [r0 + 2 * r1]
+    movq       m3, [r0 + r5]
 
-;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro FILTER_V4_W4_H4 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_pp_%1x%2, 4, 6, 8
+    punpcklbw  m0, m1
+    punpcklbw  m1, m2
+    punpcklbw  m2, m3
 
-    mov         r4d,       r4m
-    sub         r0,        r1
+    pmaddubsw  m0, m6
+    pmaddubsw  m7, m2, m5
 
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeff]
-    movd        m0,        [r5 + r4 * 4]
-%else
-    movd        m0,        [tab_ChromaCoeff + r4 * 4]
-%endif
+    paddw      m0, m7
+    psubw      m0, m4
 
-    pshufb      m0,        [tab_Cm]
+    movh       [r2], m0
+    pshufd     m0, m0, 2
+    movd       [r2 + 8], m0
 
-    mova        m1,        [pw_512]
+    lea        r0, [r0 + 4 * r1]
+    movq       m0, [r0]
+    punpcklbw  m3, m0
 
-    mov         r4d,       %2
+    pmaddubsw  m1, m6
+    pmaddubsw  m7, m3, m5
 
-    lea         r5,        [3 * r1]
+    paddw      m1, m7
+    psubw      m1, m4
 
-.loop:
-    movd        m2,        [r0]
-    movd        m3,        [r0 + r1]
-    movd        m4,        [r0 + 2 * r1]
-    movd        m5,        [r0 + r5]
+    movh       [r2 + r3], m1
+    pshufd     m1, m1, 2
+    movd       [r2 + r3 + 8], m1
 
-    punpcklbw   m2,        m3
-    punpcklbw   m6,        m4,        m5
-    punpcklbw   m2,        m6
+    movq       m1, [r0 + r1]
+    punpcklbw  m0, m1
 
-    pmaddubsw   m2,        m0
+    pmaddubsw  m2, m6
+    pmaddubsw  m0, m5
 
-    lea         r0,        [r0 + 4 * r1]
-    movd        m6,        [r0]
+    paddw      m2, m0
+    psubw      m2, m4
 
-    punpcklbw   m3,        m4
-    punpcklbw   m7,        m5,        m6
-    punpcklbw   m3,        m7
+    lea        r2,[r2 + 2 * r3]
+    movh       [r2], m2
+    pshufd     m2, m2, 2
+    movd       [r2 + 8], m2
 
-    pmaddubsw   m3,        m0
+    movq       m2,[r0 + 2 * r1]
+    punpcklbw  m1, m2
 
-    phaddw      m2,        m3
+    pmaddubsw  m3, m6
+    pmaddubsw  m1, m5
 
-    pmulhrsw    m2,        m1
+    paddw      m3, m1
+    psubw      m3, m4
 
-    movd        m7,        [r0 + r1]
+    movh       [r2 + r3], m3
+    pshufd     m3, m3, 2
+    movd       [r2 + r3 + 8], m3
 
-    punpcklbw   m4,        m5
-    punpcklbw   m3,        m6,        m7
-    punpcklbw   m4,        m3
+    lea        r2, [r2 + 2 * r3]
 
-    pmaddubsw   m4,        m0
+    dec        r4d
+    jnz        .loop
+    RET
+%endmacro
 
-    movd        m3,        [r0 + 2 * r1]
+    FILTER_V_PS_W6 6, 8
+    FILTER_V_PS_W6 6, 16
 
-    punpcklbw   m5,        m6
-    punpcklbw   m7,        m3
-    punpcklbw   m5,        m7
+;---------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ps_12x16(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;---------------------------------------------------------------------------------------------------------------
+%macro FILTER_V_PS_W12 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_ps_12x%2, 4, 6, 8
 
-    pmaddubsw   m5,        m0
+    mov        r4d, r4m
+    sub        r0, r1
+    add        r3d, r3d
 
-    phaddw      m4,        m5
+%ifdef PIC
+    lea        r5, [tab_ChromaCoeff]
+    movd       m0, [r5 + r4 * 4]
+%else
+    movd       m0, [tab_ChromaCoeff + r4 * 4]
+%endif
 
-    pmulhrsw    m4,        m1
-    packuswb    m2,        m4
-    movd        [r2],      m2
-    pextrd      [r2 + r3], m2,  1
-    lea         r2,        [r2 + 2 * r3]
-    pextrd      [r2],      m2, 2
-    pextrd      [r2 + r3], m2, 3
+    pshufb     m1, m0, [tab_Vm]
+    pshufb     m0, [tab_Vm + 16]
 
-    lea         r2,        [r2 + 2 * r3]
+    mov        r4d, %2/2
 
-    sub         r4,        4
-    jnz        .loop
-    RET
-%endmacro
+.loop:
+    movu       m2, [r0]
+    movu       m3, [r0 + r1]
 
-    FILTER_V4_W4_H4 4,  8
-    FILTER_V4_W4_H4 4, 16
+    punpcklbw  m4, m2, m3
+    punpckhbw  m2, m3
 
-    FILTER_V4_W4_H4 4, 32
+    pmaddubsw  m4, m1
+    pmaddubsw  m2, m1
 
-%macro FILTER_V4_W8_H2 0
-    punpcklbw   m1,        m2
-    punpcklbw   m7,        m3,        m0
+    lea        r0, [r0 + 2 * r1]
+    movu       m5, [r0]
+    movu       m7, [r0 + r1]
 
-    pmaddubsw   m1,        m6
-    pmaddubsw   m7,        m5
+    punpcklbw  m6, m5, m7
+    pmaddubsw  m6, m0
+    paddw      m4, m6
 
-    paddw       m1,        m7
+    punpckhbw  m6, m5, m7
+    pmaddubsw  m6, m0
+    paddw      m2, m6
 
-    pmulhrsw    m1,        m4
-    packuswb    m1,        m1
-%endmacro
+    mova       m6, [pw_2000]
 
-%macro FILTER_V4_W8_H3 0
-    punpcklbw   m2,        m3
-    punpcklbw   m7,        m0,        m1
+    psubw      m4, m6
+    psubw      m2, m6
 
-    pmaddubsw   m2,        m6
-    pmaddubsw   m7,        m5
+    movu       [r2], m4
+    movh       [r2 + 16], m2
 
-    paddw       m2,        m7
+    punpcklbw  m4, m3, m5
+    punpckhbw  m3, m5
 
-    pmulhrsw    m2,        m4
-    packuswb    m2,        m2
-%endmacro
+    pmaddubsw  m4, m1
+    pmaddubsw  m3, m1
 
-%macro FILTER_V4_W8_H4 0
-    punpcklbw   m3,        m0
-    punpcklbw   m7,        m1,        m2
+    movu       m2, [r0 + 2 * r1]
 
-    pmaddubsw   m3,        m6
-    pmaddubsw   m7,        m5
-
-    paddw       m3,        m7
-
-    pmulhrsw    m3,        m4
-    packuswb    m3,        m3
-%endmacro
-
-%macro FILTER_V4_W8_H5 0
-    punpcklbw   m0,        m1
-    punpcklbw   m7,        m2,        m3
-
-    pmaddubsw   m0,        m6
-    pmaddubsw   m7,        m5
-
-    paddw       m0,        m7
-
-    pmulhrsw    m0,        m4
-    packuswb    m0,        m0
-%endmacro
-
-%macro FILTER_V4_W8_8x2 2
-    FILTER_V4_W8 %1, %2
-    movq        m0,        [r0 + 4 * r1]
-
-    FILTER_V4_W8_H2
-
-    movh        [r2 + r3], m1
-%endmacro
+    punpcklbw  m5, m7, m2
+    punpckhbw  m7, m2
 
-%macro FILTER_V4_W8_8x4 2
-    FILTER_V4_W8_8x2 %1, %2
-;8x3
-    lea         r6,        [r0 + 4 * r1]
-    movq        m1,        [r6 + r1]
+    pmaddubsw  m5, m0
+    pmaddubsw  m7, m0
 
-    FILTER_V4_W8_H3
+    paddw      m4, m5
+    paddw      m3, m7
 
-    movh        [r2 + 2 * r3], m2
+    psubw      m4, m6
+    psubw      m3, m6
 
-;8x4
-    movq        m2,        [r6 + 2 * r1]
+    movu       [r2 + r3], m4
+    movh       [r2 + r3 + 16], m3
 
-    FILTER_V4_W8_H4
+    lea        r2, [r2 + 2 * r3]
 
-    lea         r5,        [r2 + 2 * r3]
-    movh        [r5 + r3], m3
+    dec        r4d
+    jnz        .loop
+    RET
 %endmacro
 
-%macro FILTER_V4_W8_8x6 2
-    FILTER_V4_W8_8x4 %1, %2
-;8x5
-    lea         r6,        [r6 + 2 * r1]
-    movq        m3,        [r6 + r1]
-
-    FILTER_V4_W8_H5
-
-    movh        [r2 + 4 * r3], m0
-
-;8x6
-    movq        m0,        [r0 + 8 * r1]
-
-    FILTER_V4_W8_H2
-
-    lea         r5,        [r2 + 4 * r3]
-    movh        [r5 + r3], m1
-%endmacro
+    FILTER_V_PS_W12 12, 16
+    FILTER_V_PS_W12 12, 32
 
-;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro FILTER_V4_W8 2
+;---------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ps_16x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;---------------------------------------------------------------------------------------------------------------
+%macro FILTER_V_PS_W16 2
 INIT_XMM sse4
-cglobal interp_4tap_vert_pp_%1x%2, 4, 7, 8
-
-    mov         r4d,       r4m
-
-    sub         r0,        r1
-    movq        m0,        [r0]
-    movq        m1,        [r0 + r1]
-    movq        m2,        [r0 + 2 * r1]
-    lea         r5,        [r0 + 2 * r1]
-    movq        m3,        [r5 + r1]
+cglobal interp_4tap_vert_ps_%1x%2, 4, 6, 8
 
-    punpcklbw   m0,        m1
-    punpcklbw   m4,        m2,          m3
+    mov        r4d, r4m
+    sub        r0, r1
+    add        r3d, r3d
 
 %ifdef PIC
-    lea         r6,        [tab_ChromaCoeff]
-    movd        m5,        [r6 + r4 * 4]
+    lea        r5, [tab_ChromaCoeff]
+    movd       m0, [r5 + r4 * 4]
 %else
-    movd        m5,        [tab_ChromaCoeff + r4 * 4]
+    movd       m0, [tab_ChromaCoeff + r4 * 4]
 %endif
 
-    pshufb      m6,        m5,       [tab_Vm]
-    pmaddubsw   m0,        m6
-
-    pshufb      m5,        [tab_Vm + 16]
-    pmaddubsw   m4,        m5
-
-    paddw       m0,        m4
+    pshufb     m1, m0, [tab_Vm]
+    pshufb     m0, [tab_Vm + 16]
+    mov        r4d, %2/2
 
-    mova        m4,        [pw_512]
+.loop:
+    movu       m2, [r0]
+    movu       m3, [r0 + r1]
 
-    pmulhrsw    m0,        m4
-    packuswb    m0,        m0
-    movh        [r2],      m0
-%endmacro
+    punpcklbw  m4, m2, m3
+    punpckhbw  m2, m3
 
-;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_8x2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-    FILTER_V4_W8_8x2 8, 2
+    pmaddubsw  m4, m1
+    pmaddubsw  m2, m1
 
-    RET
+    lea        r0, [r0 + 2 * r1]
+    movu       m5, [r0]
+    movu       m7, [r0 + r1]
 
-;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_8x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-    FILTER_V4_W8_8x4 8, 4
+    punpcklbw  m6, m5, m7
+    pmaddubsw  m6, m0
+    paddw      m4, m6
 
-    RET
+    punpckhbw  m6, m5, m7
+    pmaddubsw  m6, m0
+    paddw      m2, m6
 
-;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_8x6(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-    FILTER_V4_W8_8x6 8, 6
+    mova       m6, [pw_2000]
 
-    RET
+    psubw      m4, m6
+    psubw      m2, m6
 
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ps_4x2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-INIT_XMM sse4
-cglobal interp_4tap_vert_ps_4x2, 4, 6, 6
+    movu       [r2], m4
+    movu       [r2 + 16], m2
 
-    mov         r4d, r4m
-    sub         r0, r1
-    add         r3d, r3d
+    punpcklbw  m4, m3, m5
+    punpckhbw  m3, m5
 
-%ifdef PIC
-    lea         r5, [tab_ChromaCoeff]
-    movd        m0, [r5 + r4 * 4]
-%else
-    movd        m0, [tab_ChromaCoeff + r4 * 4]
-%endif
+    pmaddubsw  m4, m1
+    pmaddubsw  m3, m1
 
-    pshufb      m0, [tab_Cm]
+    movu       m5, [r0 + 2 * r1]
 
-    movd        m2, [r0]
-    movd        m3, [r0 + r1]
-    lea         r5, [r0 + 2 * r1]
-    movd        m4, [r5]
-    movd        m5, [r5 + r1]
+    punpcklbw  m2, m7, m5
+    punpckhbw  m7, m5
 
-    punpcklbw   m2, m3
-    punpcklbw   m1, m4, m5
-    punpcklbw   m2, m1
+    pmaddubsw  m2, m0
+    pmaddubsw  m7, m0
 
-    pmaddubsw   m2, m0
+    paddw      m4, m2
+    paddw      m3, m7
 
-    movd        m1, [r0 + 4 * r1]
+    psubw      m4, m6
+    psubw      m3, m6
 
-    punpcklbw   m3, m4
-    punpcklbw   m5, m1
-    punpcklbw   m3, m5
+    movu       [r2 + r3], m4
+    movu       [r2 + r3 + 16], m3
 
-    pmaddubsw   m3, m0
+    lea        r2, [r2 + 2 * r3]
 
-    phaddw      m2, m3
+    dec        r4d
+    jnz        .loop
+    RET
+%endmacro
 
-    psubw       m2, [pw_2000]
-    movh        [r2], m2
-    movhps      [r2 + r3], m2
+    FILTER_V_PS_W16 16,  4
+    FILTER_V_PS_W16 16,  8
+    FILTER_V_PS_W16 16, 12
+    FILTER_V_PS_W16 16, 16
+    FILTER_V_PS_W16 16, 32
 
-    RET
+    FILTER_V_PS_W16 16, 24
+    FILTER_V_PS_W16 16, 64
 
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ps_4x4(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
+;--------------------------------------------------------------------------------------------------------------
+;void interp_4tap_vert_ps_24x32(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+%macro FILTER_V4_PS_W24 2
 INIT_XMM sse4
-cglobal interp_4tap_vert_ps_4x4, 4, 6, 7
+cglobal interp_4tap_vert_ps_24x%2, 4, 6, 8
 
     mov        r4d, r4m
     sub        r0, r1
@@ -7025,67 +4838,110 @@ cglobal interp_4tap_vert_ps_4x4, 4, 6, 7
     movd       m0, [tab_ChromaCoeff + r4 * 4]
 %endif
 
-    pshufb     m0, [tab_Cm]
+    pshufb     m1, m0, [tab_Vm]
+    pshufb     m0, [tab_Vm + 16]
 
-    lea        r4, [r1 * 3]
-    lea        r5, [r0 + 4 * r1]
+    mov        r4d, %2/2
 
-    movd       m2, [r0]
-    movd       m3, [r0 + r1]
-    movd       m4, [r0 + 2 * r1]
-    movd       m5, [r0 + r4]
+.loop:
+    movu       m2, [r0]
+    movu       m3, [r0 + r1]
 
-    punpcklbw  m2, m3
-    punpcklbw  m6, m4, m5
-    punpcklbw  m2, m6
+    punpcklbw  m4, m2, m3
+    punpckhbw  m2, m3
 
-    pmaddubsw  m2, m0
+    pmaddubsw  m4, m1
+    pmaddubsw  m2, m1
 
-    movd       m6, [r5]
+    lea        r5, [r0 + 2 * r1]
 
-    punpcklbw  m3, m4
-    punpcklbw  m1, m5, m6
-    punpcklbw  m3, m1
+    movu       m5, [r5]
+    movu       m7, [r5 + r1]
 
-    pmaddubsw  m3, m0
+    punpcklbw  m6, m5, m7
+    pmaddubsw  m6, m0
+    paddw      m4, m6
 
-    phaddw     m2, m3
+    punpckhbw  m6, m5, m7
+    pmaddubsw  m6, m0
+    paddw      m2, m6
 
-    mova       m1, [pw_2000]
+    mova       m6, [pw_2000]
 
-    psubw      m2, m1
-    movh       [r2], m2
-    movhps     [r2 + r3], m2
+    psubw      m4, m6
+    psubw      m2, m6
 
-    movd       m2, [r5 + r1]
+    movu       [r2], m4
+    movu       [r2 + 16], m2
 
-    punpcklbw  m4, m5
-    punpcklbw  m3, m6, m2
-    punpcklbw  m4, m3
+    punpcklbw  m4, m3, m5
+    punpckhbw  m3, m5
 
-    pmaddubsw  m4, m0
+    pmaddubsw  m4, m1
+    pmaddubsw  m3, m1
 
-    movd       m3, [r5 + 2 * r1]
+    movu       m2, [r5 + 2 * r1]
+
+    punpcklbw  m5, m7, m2
+    punpckhbw  m7, m2
+
+    pmaddubsw  m5, m0
+    pmaddubsw  m7, m0
+
+    paddw      m4, m5
+    paddw      m3, m7
+
+    psubw      m4, m6
+    psubw      m3, m6
+
+    movu       [r2 + r3], m4
+    movu       [r2 + r3 + 16], m3
+
+    movq       m2, [r0 + 16]
+    movq       m3, [r0 + r1 + 16]
+    movq       m4, [r5 + 16]
+    movq       m5, [r5 + r1 + 16]
 
-    punpcklbw  m5, m6
     punpcklbw  m2, m3
+    punpcklbw  m7, m4, m5
+
+    pmaddubsw  m2, m1
+    pmaddubsw  m7, m0
+
+    paddw      m2, m7
+    psubw      m2, m6
+
+    movu       [r2 + 32], m2
+
+    movq       m2, [r5 + 2 * r1 + 16]
+
+    punpcklbw  m3, m4
     punpcklbw  m5, m2
 
+    pmaddubsw  m3, m1
     pmaddubsw  m5, m0
 
-    phaddw     m4, m5
+    paddw      m3, m5
+    psubw      m3,  m6
 
-    psubw      m4, m1
+    movu       [r2 + r3 + 32], m3
+
+    mov        r0, r5
     lea        r2, [r2 + 2 * r3]
-    movh       [r2], m4
-    movhps     [r2 + r3], m4
 
+    dec        r4d
+    jnz        .loop
     RET
+%endmacro
+
+    FILTER_V4_PS_W24 24, 32
+
+    FILTER_V4_PS_W24 24, 64
 
 ;---------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ps_%1x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+; void interp_4tap_vert_ps_32x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
 ;---------------------------------------------------------------------------------------------------------------
-%macro FILTER_V_PS_W4_H4 2
+%macro FILTER_V_PS_W32 2
 INIT_XMM sse4
 cglobal interp_4tap_vert_ps_%1x%2, 4, 6, 8
 
@@ -7100,731 +4956,942 @@ cglobal interp_4tap_vert_ps_%1x%2, 4, 6, 8
     movd       m0, [tab_ChromaCoeff + r4 * 4]
 %endif
 
-    pshufb     m0, [tab_Cm]
+    pshufb     m1, m0, [tab_Vm]
+    pshufb     m0, [tab_Vm + 16]
 
-    mova       m1, [pw_2000]
+    mova       m7, [pw_2000]
 
-    mov        r4d, %2/4
-    lea        r5, [3 * r1]
+    mov        r4d, %2
 
 .loop:
-    movd       m2, [r0]
-    movd       m3, [r0 + r1]
-    movd       m4, [r0 + 2 * r1]
-    movd       m5, [r0 + r5]
+    movu       m2, [r0]
+    movu       m3, [r0 + r1]
 
-    punpcklbw  m2, m3
-    punpcklbw  m6, m4, m5
-    punpcklbw  m2, m6
+    punpcklbw  m4, m2, m3
+    punpckhbw  m2, m3
 
-    pmaddubsw  m2, m0
+    pmaddubsw  m4, m1
+    pmaddubsw  m2, m1
 
-    lea        r0, [r0 + 4 * r1]
-    movd       m6, [r0]
+    lea        r5, [r0 + 2 * r1]
+    movu       m3, [r5]
+    movu       m5, [r5 + r1]
 
-    punpcklbw  m3, m4
-    punpcklbw  m7, m5, m6
-    punpcklbw  m3, m7
+    punpcklbw  m6, m3, m5
+    punpckhbw  m3, m5
 
+    pmaddubsw  m6, m0
     pmaddubsw  m3, m0
 
-    phaddw     m2, m3
+    paddw      m4, m6
+    paddw      m2, m3
 
-    psubw      m2, m1
-    movh       [r2], m2
-    movhps     [r2 + r3], m2
+    psubw      m4, m7
+    psubw      m2, m7
 
-    movd       m2, [r0 + r1]
+    movu       [r2], m4
+    movu       [r2 + 16], m2
 
-    punpcklbw  m4, m5
-    punpcklbw  m3, m6, m2
-    punpcklbw  m4, m3
+    movu       m2, [r0 + 16]
+    movu       m3, [r0 + r1 + 16]
 
-    pmaddubsw  m4, m0
+    punpcklbw  m4, m2, m3
+    punpckhbw  m2, m3
 
-    movd       m3, [r0 + 2 * r1]
+    pmaddubsw  m4, m1
+    pmaddubsw  m2, m1
 
-    punpcklbw  m5, m6
-    punpcklbw  m2, m3
-    punpcklbw  m5, m2
+    movu       m3, [r5 + 16]
+    movu       m5, [r5 + r1 + 16]
 
-    pmaddubsw  m5, m0
+    punpcklbw  m6, m3, m5
+    punpckhbw  m3, m5
 
-    phaddw     m4, m5
+    pmaddubsw  m6, m0
+    pmaddubsw  m3, m0
 
-    psubw      m4, m1
-    lea        r2, [r2 + 2 * r3]
-    movh       [r2], m4
-    movhps     [r2 + r3], m4
+    paddw      m4, m6
+    paddw      m2, m3
 
-    lea        r2, [r2 + 2 * r3]
+    psubw      m4, m7
+    psubw      m2, m7
+
+    movu       [r2 + 32], m4
+    movu       [r2 + 48], m2
+
+    lea        r0, [r0 + r1]
+    lea        r2, [r2 + r3]
 
     dec        r4d
     jnz        .loop
     RET
 %endmacro
 
-    FILTER_V_PS_W4_H4 4, 8
-    FILTER_V_PS_W4_H4 4, 16
+    FILTER_V_PS_W32 32,  8
+    FILTER_V_PS_W32 32, 16
+    FILTER_V_PS_W32 32, 24
+    FILTER_V_PS_W32 32, 32
 
-    FILTER_V_PS_W4_H4 4, 32
+    FILTER_V_PS_W32 32, 48
+    FILTER_V_PS_W32 32, 64
 
-;--------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ps_8x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-%macro FILTER_V_PS_W8_H8_H16_H2 2
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W8_H8_H16_H32 2
 INIT_XMM sse4
-cglobal interp_4tap_vert_ps_%1x%2, 4, 6, 7
+cglobal interp_4tap_vert_pp_%1x%2, 4, 6, 8
 
-    mov        r4d, r4m
-    sub        r0, r1
-    add        r3d, r3d
+    mov         r4d,       r4m
+    sub         r0,        r1
 
 %ifdef PIC
-    lea        r5, [tab_ChromaCoeff]
-    movd       m5, [r5 + r4 * 4]
+    lea         r5,        [tab_ChromaCoeff]
+    movd        m5,        [r5 + r4 * 4]
 %else
-    movd       m5, [tab_ChromaCoeff + r4 * 4]
+    movd        m5,        [tab_ChromaCoeff + r4 * 4]
 %endif
 
-    pshufb     m6, m5, [tab_Vm]
-    pshufb     m5, [tab_Vm + 16]
-    mova       m4, [pw_2000]
+    pshufb      m6,        m5,       [tab_Vm]
+    pshufb      m5,        [tab_Vm + 16]
+    mova        m4,        [pw_512]
+    lea         r5,        [r1 * 3]
 
-    mov        r4d, %2/2
-    lea        r5, [3 * r1]
+    mov         r4d,       %2
 
-.loopH:
-    movq       m0, [r0]
-    movq       m1, [r0 + r1]
-    movq       m2, [r0 + 2 * r1]
-    movq       m3, [r0 + r5]
+.loop:
+    movq        m0,        [r0]
+    movq        m1,        [r0 + r1]
+    movq        m2,        [r0 + 2 * r1]
+    movq        m3,        [r0 + r5]
 
-    punpcklbw  m0, m1
-    punpcklbw  m1, m2
-    punpcklbw  m2, m3
+    punpcklbw   m0,        m1
+    punpcklbw   m1,        m2
+    punpcklbw   m2,        m3
 
-    pmaddubsw  m0, m6
-    pmaddubsw  m2, m5
+    pmaddubsw   m0,        m6
+    pmaddubsw   m7,        m2, m5
 
-    paddw      m0, m2
+    paddw       m0,        m7
 
-    psubw      m0, m4
-    movu       [r2], m0
+    pmulhrsw    m0,        m4
+    packuswb    m0,        m0
+    movh        [r2],      m0
 
-    movq       m0, [r0 + 4 * r1]
+    lea         r0,        [r0 + 4 * r1]
+    movq        m0,        [r0]
 
-    punpcklbw  m3, m0
+    punpcklbw   m3,        m0
 
-    pmaddubsw  m1, m6
-    pmaddubsw  m3, m5
+    pmaddubsw   m1,        m6
+    pmaddubsw   m7,        m3, m5
 
-    paddw      m1, m3
-    psubw      m1, m4
+    paddw       m1,        m7
 
-    movu       [r2 + r3], m1
+    pmulhrsw    m1,        m4
+    packuswb    m1,        m1
+    movh        [r2 + r3], m1
 
-    lea        r0, [r0 + 2 * r1]
-    lea        r2, [r2 + 2 * r3]
+    movq        m1,        [r0 + r1]
 
-    dec        r4d
-    jnz       .loopH
+    punpcklbw   m0,        m1
 
-    RET
-%endmacro
+    pmaddubsw   m2,        m6
+    pmaddubsw   m0,        m5
 
-    FILTER_V_PS_W8_H8_H16_H2 8, 2
-    FILTER_V_PS_W8_H8_H16_H2 8, 4
-    FILTER_V_PS_W8_H8_H16_H2 8, 6
-
-    FILTER_V_PS_W8_H8_H16_H2 8, 12
-    FILTER_V_PS_W8_H8_H16_H2 8, 64
-
-;--------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ps_8x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-%macro FILTER_V_PS_W8_H8_H16_H32 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_ps_%1x%2, 4, 6, 8
-
-    mov        r4d, r4m
-    sub        r0, r1
-    add        r3d, r3d
-
-%ifdef PIC
-    lea        r5, [tab_ChromaCoeff]
-    movd       m5, [r5 + r4 * 4]
-%else
-    movd       m5, [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    pshufb     m6, m5, [tab_Vm]
-    pshufb     m5, [tab_Vm + 16]
-    mova       m4, [pw_2000]
-
-    mov        r4d, %2/4
-    lea        r5, [3 * r1]
-
-.loop:
-    movq       m0, [r0]
-    movq       m1, [r0 + r1]
-    movq       m2, [r0 + 2 * r1]
-    movq       m3, [r0 + r5]
-
-    punpcklbw  m0, m1
-    punpcklbw  m1, m2
-    punpcklbw  m2, m3
-
-    pmaddubsw  m0, m6
-    pmaddubsw  m7, m2, m5
-
-    paddw      m0, m7
-
-    psubw       m0, m4
-    movu       [r2], m0
-
-    lea        r0, [r0 + 4 * r1]
-    movq       m0, [r0]
-
-    punpcklbw  m3, m0
-
-    pmaddubsw  m1, m6
-    pmaddubsw  m7, m3, m5
-
-    paddw      m1, m7
-
-    psubw      m1, m4
-    movu       [r2 + r3], m1
-
-    movq       m1, [r0 + r1]
-
-    punpcklbw  m0, m1
-
-    pmaddubsw  m2, m6
-    pmaddubsw  m0, m5
-
-    paddw      m2, m0
+    paddw       m2,        m0
 
-    psubw      m2, m4
-    lea        r2, [r2 + 2 * r3]
-    movu       [r2], m2
+    pmulhrsw    m2,        m4
 
-    movq       m2, [r0 + 2 * r1]
+    movq        m7,        [r0 + 2 * r1]
+    punpcklbw   m1,        m7
 
-    punpcklbw  m1, m2
+    pmaddubsw   m3,        m6
+    pmaddubsw   m1,        m5
 
-    pmaddubsw  m3, m6
-    pmaddubsw  m1, m5
+    paddw       m3,        m1
 
-    paddw      m3, m1
-    psubw      m3, m4
+    pmulhrsw    m3,        m4
+    packuswb    m2,        m3
 
-    movu       [r2 + r3], m3
+    lea         r2,        [r2 + 2 * r3]
+    movh        [r2],      m2
+    movhps      [r2 + r3], m2
 
-    lea        r2, [r2 + 2 * r3]
+    lea         r2,        [r2 + 2 * r3]
 
-    dec        r4d
+    sub         r4,         4
     jnz        .loop
     RET
 %endmacro
 
-    FILTER_V_PS_W8_H8_H16_H32 8,  8
-    FILTER_V_PS_W8_H8_H16_H32 8, 16
-    FILTER_V_PS_W8_H8_H16_H32 8, 32
+    FILTER_V4_W8_H8_H16_H32 8,  8
+    FILTER_V4_W8_H8_H16_H32 8, 16
+    FILTER_V4_W8_H8_H16_H32 8, 32
 
-;------------------------------------------------------------------------------------------------------------
-;void interp_4tap_vert_ps_6x8(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;------------------------------------------------------------------------------------------------------------
-%macro FILTER_V_PS_W6 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_ps_6x%2, 4, 6, 8
+    FILTER_V4_W8_H8_H16_H32 8, 12
+    FILTER_V4_W8_H8_H16_H32 8, 64
 
-    mov        r4d, r4m
-    sub        r0, r1
-    add        r3d, r3d
+%macro PROCESS_CHROMA_AVX2_W8_8R 0
+    movq            xm1, [r0]                       ; m1 = row 0
+    movq            xm2, [r0 + r1]                  ; m2 = row 1
+    punpcklbw       xm1, xm2                        ; m1 = [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
+    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
+    punpcklbw       xm2, xm3                        ; m2 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10]
+    vinserti128     m5, m1, xm2, 1                  ; m5 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10] - [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
+    pmaddubsw       m5, [r5]
+    movq            xm4, [r0 + r4]                  ; m4 = row 3
+    punpcklbw       xm3, xm4                        ; m3 = [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
+    lea             r0, [r0 + r1 * 4]
+    movq            xm1, [r0]                       ; m1 = row 4
+    punpcklbw       xm4, xm1                        ; m4 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30]
+    vinserti128     m2, m3, xm4, 1                  ; m2 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30] - [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
+    pmaddubsw       m0, m2, [r5 + 1 * mmsize]
+    paddw           m5, m0
+    pmaddubsw       m2, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 5
+    punpcklbw       xm1, xm3                        ; m1 = [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
+    movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
+    punpcklbw       xm3, xm4                        ; m3 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50]
+    vinserti128     m1, m1, xm3, 1                  ; m1 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50] - [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
+    pmaddubsw       m0, m1, [r5 + 1 * mmsize]
+    paddw           m2, m0
+    pmaddubsw       m1, [r5]
+    movq            xm3, [r0 + r4]                  ; m3 = row 7
+    punpcklbw       xm4, xm3                        ; m4 = [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
+    lea             r0, [r0 + r1 * 4]
+    movq            xm0, [r0]                       ; m0 = row 8
+    punpcklbw       xm3, xm0                        ; m3 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70]
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70] - [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
+    pmaddubsw       m3, m4, [r5 + 1 * mmsize]
+    paddw           m1, m3
+    pmaddubsw       m4, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 9
+    punpcklbw       xm0, xm3                        ; m0 = [97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80]
+    movq            xm6, [r0 + r1 * 2]              ; m6 = row 10
+    punpcklbw       xm3, xm6                        ; m3 = [A7 97 A6 96 A5 95 A4 94 A3 93 A2 92 A1 91 A0 90]
+    vinserti128     m0, m0, xm3, 1                  ; m0 = [A7 97 A6 96 A5 95 A4 94 A3 93 A2 92 A1 91 A0 90] - [97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80]
+    pmaddubsw       m0, [r5 + 1 * mmsize]
+    paddw           m4, m0
+%endmacro
+
+%macro FILTER_VER_CHROMA_AVX2_8x8 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_8x8, 4, 6, 7
+    mov             r4d, r4m
+    shl             r4d, 6
 
 %ifdef PIC
-    lea        r5, [tab_ChromaCoeff]
-    movd       m5, [r5 + r4 * 4]
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
 %else
-    movd       m5, [tab_ChromaCoeff + r4 * 4]
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
 %endif
 
-    pshufb     m6, m5, [tab_Vm]
-    pshufb     m5, [tab_Vm + 16]
-    mova       m4, [pw_2000]
-    lea        r5, [3 * r1]
-    mov        r4d, %2/4
-
-.loop:
-    movq       m0, [r0]
-    movq       m1, [r0 + r1]
-    movq       m2, [r0 + 2 * r1]
-    movq       m3, [r0 + r5]
-
-    punpcklbw  m0, m1
-    punpcklbw  m1, m2
-    punpcklbw  m2, m3
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+    PROCESS_CHROMA_AVX2_W8_8R
+%ifidn %1,pp
+    lea             r4, [r3 * 3]
+    mova            m3, [pw_512]
+    pmulhrsw        m5, m3                          ; m5 = word: row 0, row 1
+    pmulhrsw        m2, m3                          ; m2 = word: row 2, row 3
+    pmulhrsw        m1, m3                          ; m1 = word: row 4, row 5
+    pmulhrsw        m4, m3                          ; m4 = word: row 6, row 7
+    packuswb        m5, m2
+    packuswb        m1, m4
+    vextracti128    xm2, m5, 1
+    vextracti128    xm4, m1, 1
+    movq            [r2], xm5
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm5
+    movhps          [r2 + r4], xm2
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm1
+    movq            [r2 + r3], xm4
+    movhps          [r2 + r3 * 2], xm1
+    movhps          [r2 + r4], xm4
+%else
+    add             r3d, r3d
+    vbroadcasti128  m3, [pw_2000]
+    lea             r4, [r3 * 3]
+    psubw           m5, m3                          ; m5 = word: row 0, row 1
+    psubw           m2, m3                          ; m2 = word: row 2, row 3
+    psubw           m1, m3                          ; m1 = word: row 4, row 5
+    psubw           m4, m3                          ; m4 = word: row 6, row 7
+    vextracti128    xm6, m5, 1
+    vextracti128    xm3, m2, 1
+    vextracti128    xm0, m1, 1
+    movu            [r2], xm5
+    movu            [r2 + r3], xm6
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r4], xm3
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm1
+    movu            [r2 + r3], xm0
+    movu            [r2 + r3 * 2], xm4
+    vextracti128    xm4, m4, 1
+    movu            [r2 + r4], xm4
+%endif
+    RET
+%endmacro
 
-    pmaddubsw  m0, m6
-    pmaddubsw  m7, m2, m5
+    FILTER_VER_CHROMA_AVX2_8x8 pp
+    FILTER_VER_CHROMA_AVX2_8x8 ps
 
-    paddw      m0, m7
-    psubw      m0, m4
+%macro FILTER_VER_CHROMA_AVX2_8x6 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_8x6, 4, 6, 6
+    mov             r4d, r4m
+    shl             r4d, 6
 
-    movh       [r2], m0
-    pshufd     m0, m0, 2
-    movd       [r2 + 8], m0
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
 
-    lea        r0, [r0 + 4 * r1]
-    movq       m0, [r0]
-    punpcklbw  m3, m0
+    lea             r4, [r1 * 3]
+    sub             r0, r1
 
-    pmaddubsw  m1, m6
-    pmaddubsw  m7, m3, m5
-
-    paddw      m1, m7
-    psubw      m1, m4
-
-    movh       [r2 + r3], m1
-    pshufd     m1, m1, 2
-    movd       [r2 + r3 + 8], m1
-
-    movq       m1, [r0 + r1]
-    punpcklbw  m0, m1
-
-    pmaddubsw  m2, m6
-    pmaddubsw  m0, m5
-
-    paddw      m2, m0
-    psubw      m2, m4
-
-    lea        r2,[r2 + 2 * r3]
-    movh       [r2], m2
-    pshufd     m2, m2, 2
-    movd       [r2 + 8], m2
-
-    movq       m2,[r0 + 2 * r1]
-    punpcklbw  m1, m2
-
-    pmaddubsw  m3, m6
-    pmaddubsw  m1, m5
-
-    paddw      m3, m1
-    psubw      m3, m4
-
-    movh       [r2 + r3], m3
-    pshufd     m3, m3, 2
-    movd       [r2 + r3 + 8], m3
-
-    lea        r2, [r2 + 2 * r3]
-
-    dec        r4d
-    jnz        .loop
-    RET
-%endmacro
-
-    FILTER_V_PS_W6 6, 8
-    FILTER_V_PS_W6 6, 16
-
-;---------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ps_12x16(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;---------------------------------------------------------------------------------------------------------------
-%macro FILTER_V_PS_W12 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_ps_12x%2, 4, 6, 8
-
-    mov        r4d, r4m
-    sub        r0, r1
-    add        r3d, r3d
-
-%ifdef PIC
-    lea        r5, [tab_ChromaCoeff]
-    movd       m0, [r5 + r4 * 4]
+    movq            xm1, [r0]                       ; m1 = row 0
+    movq            xm2, [r0 + r1]                  ; m2 = row 1
+    punpcklbw       xm1, xm2                        ; m1 = [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
+    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
+    punpcklbw       xm2, xm3                        ; m2 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10]
+    vinserti128     m5, m1, xm2, 1                  ; m5 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10] - [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
+    pmaddubsw       m5, [r5]
+    movq            xm4, [r0 + r4]                  ; m4 = row 3
+    punpcklbw       xm3, xm4                        ; m3 = [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
+    lea             r0, [r0 + r1 * 4]
+    movq            xm1, [r0]                       ; m1 = row 4
+    punpcklbw       xm4, xm1                        ; m4 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30]
+    vinserti128     m2, m3, xm4, 1                  ; m2 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30] - [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
+    pmaddubsw       m0, m2, [r5 + 1 * mmsize]
+    paddw           m5, m0
+    pmaddubsw       m2, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 5
+    punpcklbw       xm1, xm3                        ; m1 = [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
+    movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
+    punpcklbw       xm3, xm4                        ; m3 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50]
+    vinserti128     m1, m1, xm3, 1                  ; m1 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50] - [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
+    pmaddubsw       m0, m1, [r5 + 1 * mmsize]
+    paddw           m2, m0
+    pmaddubsw       m1, [r5]
+    movq            xm3, [r0 + r4]                  ; m3 = row 7
+    punpcklbw       xm4, xm3                        ; m4 = [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
+    lea             r0, [r0 + r1 * 4]
+    movq            xm0, [r0]                       ; m0 = row 8
+    punpcklbw       xm3, xm0                        ; m3 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70]
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70] - [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
+    pmaddubsw       m4, [r5 + 1 * mmsize]
+    paddw           m1, m4
+%ifidn %1,pp
+    lea             r4, [r3 * 3]
+    mova            m3, [pw_512]
+    pmulhrsw        m5, m3                          ; m5 = word: row 0, row 1
+    pmulhrsw        m2, m3                          ; m2 = word: row 2, row 3
+    pmulhrsw        m1, m3                          ; m1 = word: row 4, row 5
+    packuswb        m5, m2
+    packuswb        m1, m1
+    vextracti128    xm2, m5, 1
+    vextracti128    xm4, m1, 1
+    movq            [r2], xm5
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm5
+    movhps          [r2 + r4], xm2
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm1
+    movq            [r2 + r3], xm4
 %else
-    movd       m0, [tab_ChromaCoeff + r4 * 4]
+    add             r3d, r3d
+    mova            m3, [pw_2000]
+    lea             r4, [r3 * 3]
+    psubw           m5, m3                          ; m5 = word: row 0, row 1
+    psubw           m2, m3                          ; m2 = word: row 2, row 3
+    psubw           m1, m3                          ; m1 = word: row 4, row 5
+    vextracti128    xm4, m5, 1
+    vextracti128    xm3, m2, 1
+    vextracti128    xm0, m1, 1
+    movu            [r2], xm5
+    movu            [r2 + r3], xm4
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r4], xm3
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm1
+    movu            [r2 + r3], xm0
 %endif
-
-    pshufb     m1, m0, [tab_Vm]
-    pshufb     m0, [tab_Vm + 16]
-
-    mov        r4d, %2/2
-
-.loop:
-    movu       m2, [r0]
-    movu       m3, [r0 + r1]
-
-    punpcklbw  m4, m2, m3
-    punpckhbw  m2, m3
-
-    pmaddubsw  m4, m1
-    pmaddubsw  m2, m1
-
-    lea        r0, [r0 + 2 * r1]
-    movu       m5, [r0]
-    movu       m7, [r0 + r1]
-
-    punpcklbw  m6, m5, m7
-    pmaddubsw  m6, m0
-    paddw      m4, m6
-
-    punpckhbw  m6, m5, m7
-    pmaddubsw  m6, m0
-    paddw      m2, m6
-
-    mova       m6, [pw_2000]
-
-    psubw      m4, m6
-    psubw      m2, m6
-
-    movu       [r2], m4
-    movh       [r2 + 16], m2
-
-    punpcklbw  m4, m3, m5
-    punpckhbw  m3, m5
-
-    pmaddubsw  m4, m1
-    pmaddubsw  m3, m1
-
-    movu       m2, [r0 + 2 * r1]
-
-    punpcklbw  m5, m7, m2
-    punpckhbw  m7, m2
-
-    pmaddubsw  m5, m0
-    pmaddubsw  m7, m0
-
-    paddw      m4, m5
-    paddw      m3, m7
-
-    psubw      m4, m6
-    psubw      m3, m6
-
-    movu       [r2 + r3], m4
-    movh       [r2 + r3 + 16], m3
-
-    lea        r2, [r2 + 2 * r3]
-
-    dec        r4d
-    jnz        .loop
     RET
 %endmacro
 
-    FILTER_V_PS_W12 12, 16
-    FILTER_V_PS_W12 12, 32
-
-;---------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ps_16x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;---------------------------------------------------------------------------------------------------------------
-%macro FILTER_V_PS_W16 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_ps_%1x%2, 4, 6, 8
-
-    mov        r4d, r4m
-    sub        r0, r1
-    add        r3d, r3d
-
-%ifdef PIC
-    lea        r5, [tab_ChromaCoeff]
-    movd       m0, [r5 + r4 * 4]
-%else
-    movd       m0, [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    pshufb     m1, m0, [tab_Vm]
-    pshufb     m0, [tab_Vm + 16]
-    mov        r4d, %2/2
-
-.loop:
-    movu       m2, [r0]
-    movu       m3, [r0 + r1]
-
-    punpcklbw  m4, m2, m3
-    punpckhbw  m2, m3
-
-    pmaddubsw  m4, m1
-    pmaddubsw  m2, m1
-
-    lea        r0, [r0 + 2 * r1]
-    movu       m5, [r0]
-    movu       m7, [r0 + r1]
-
-    punpcklbw  m6, m5, m7
-    pmaddubsw  m6, m0
-    paddw      m4, m6
-
-    punpckhbw  m6, m5, m7
-    pmaddubsw  m6, m0
-    paddw      m2, m6
-
-    mova       m6, [pw_2000]
-
-    psubw      m4, m6
-    psubw      m2, m6
-
-    movu       [r2], m4
-    movu       [r2 + 16], m2
-
-    punpcklbw  m4, m3, m5
-    punpckhbw  m3, m5
-
-    pmaddubsw  m4, m1
-    pmaddubsw  m3, m1
-
-    movu       m5, [r0 + 2 * r1]
-
-    punpcklbw  m2, m7, m5
-    punpckhbw  m7, m5
+    FILTER_VER_CHROMA_AVX2_8x6 pp
+    FILTER_VER_CHROMA_AVX2_8x6 ps
 
-    pmaddubsw  m2, m0
-    pmaddubsw  m7, m0
-
-    paddw      m4, m2
-    paddw      m3, m7
-
-    psubw      m4, m6
-    psubw      m3, m6
+%macro PROCESS_CHROMA_AVX2_W8_16R 1
+    movq            xm1, [r0]                       ; m1 = row 0
+    movq            xm2, [r0 + r1]                  ; m2 = row 1
+    punpcklbw       xm1, xm2
+    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
+    punpcklbw       xm2, xm3
+    vinserti128     m5, m1, xm2, 1
+    pmaddubsw       m5, [r5]
+    movq            xm4, [r0 + r4]                  ; m4 = row 3
+    punpcklbw       xm3, xm4
+    lea             r0, [r0 + r1 * 4]
+    movq            xm1, [r0]                       ; m1 = row 4
+    punpcklbw       xm4, xm1
+    vinserti128     m2, m3, xm4, 1
+    pmaddubsw       m0, m2, [r5 + 1 * mmsize]
+    paddw           m5, m0
+    pmaddubsw       m2, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 5
+    punpcklbw       xm1, xm3
+    movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
+    punpcklbw       xm3, xm4
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m0, m1, [r5 + 1 * mmsize]
+    paddw           m2, m0
+    pmaddubsw       m1, [r5]
+    movq            xm3, [r0 + r4]                  ; m3 = row 7
+    punpcklbw       xm4, xm3
+    lea             r0, [r0 + r1 * 4]
+    movq            xm0, [r0]                       ; m0 = row 8
+    punpcklbw       xm3, xm0
+    vinserti128     m4, m4, xm3, 1
+    pmaddubsw       m3, m4, [r5 + 1 * mmsize]
+    paddw           m1, m3
+    pmaddubsw       m4, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 9
+    punpcklbw       xm0, xm3
+    movq            xm6, [r0 + r1 * 2]              ; m6 = row 10
+    punpcklbw       xm3, xm6
+    vinserti128     m0, m0, xm3, 1
+    pmaddubsw       m3, m0, [r5 + 1 * mmsize]
+    paddw           m4, m3
+    pmaddubsw       m0, [r5]
+%ifidn %1,pp
+    pmulhrsw        m5, m7                          ; m5 = word: row 0, row 1
+    pmulhrsw        m2, m7                          ; m2 = word: row 2, row 3
+    pmulhrsw        m1, m7                          ; m1 = word: row 4, row 5
+    pmulhrsw        m4, m7                          ; m4 = word: row 6, row 7
+    packuswb        m5, m2
+    packuswb        m1, m4
+    vextracti128    xm2, m5, 1
+    vextracti128    xm4, m1, 1
+    movq            [r2], xm5
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm5
+    movhps          [r2 + r6], xm2
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm1
+    movq            [r2 + r3], xm4
+    movhps          [r2 + r3 * 2], xm1
+    movhps          [r2 + r6], xm4
+%else
+    psubw           m5, m7                          ; m5 = word: row 0, row 1
+    psubw           m2, m7                          ; m2 = word: row 2, row 3
+    psubw           m1, m7                          ; m1 = word: row 4, row 5
+    psubw           m4, m7                          ; m4 = word: row 6, row 7
+    vextracti128    xm3, m5, 1
+    movu            [r2], xm5
+    movu            [r2 + r3], xm3
+    vextracti128    xm3, m2, 1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
+    lea             r2, [r2 + r3 * 4]
+    vextracti128    xm5, m1, 1
+    vextracti128    xm3, m4, 1
+    movu            [r2], xm1
+    movu            [r2 + r3], xm5
+    movu            [r2 + r3 * 2], xm4
+    movu            [r2 + r6], xm3
+%endif
+    movq            xm3, [r0 + r4]                  ; m3 = row 11
+    punpcklbw       xm6, xm3
+    lea             r0, [r0 + r1 * 4]
+    movq            xm5, [r0]                       ; m5 = row 12
+    punpcklbw       xm3, xm5
+    vinserti128     m6, m6, xm3, 1
+    pmaddubsw       m3, m6, [r5 + 1 * mmsize]
+    paddw           m0, m3
+    pmaddubsw       m6, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 13
+    punpcklbw       xm5, xm3
+    movq            xm2, [r0 + r1 * 2]              ; m2 = row 14
+    punpcklbw       xm3, xm2
+    vinserti128     m5, m5, xm3, 1
+    pmaddubsw       m3, m5, [r5 + 1 * mmsize]
+    paddw           m6, m3
+    pmaddubsw       m5, [r5]
+    movq            xm3, [r0 + r4]                  ; m3 = row 15
+    punpcklbw       xm2, xm3
+    lea             r0, [r0 + r1 * 4]
+    movq            xm1, [r0]                       ; m1 = row 16
+    punpcklbw       xm3, xm1
+    vinserti128     m2, m2, xm3, 1
+    pmaddubsw       m3, m2, [r5 + 1 * mmsize]
+    paddw           m5, m3
+    pmaddubsw       m2, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 17
+    punpcklbw       xm1, xm3
+    movq            xm4, [r0 + r1 * 2]              ; m4 = row 18
+    punpcklbw       xm3, xm4
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m1, [r5 + 1 * mmsize]
+    paddw           m2, m1
+    lea             r2, [r2 + r3 * 4]
+%ifidn %1,pp
+    pmulhrsw        m0, m7                          ; m0 = word: row 8, row 9
+    pmulhrsw        m6, m7                          ; m6 = word: row 10, row 11
+    pmulhrsw        m5, m7                          ; m5 = word: row 12, row 13
+    pmulhrsw        m2, m7                          ; m2 = word: row 14, row 15
+    packuswb        m0, m6
+    packuswb        m5, m2
+    vextracti128    xm6, m0, 1
+    vextracti128    xm2, m5, 1
+    movq            [r2], xm0
+    movq            [r2 + r3], xm6
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r6], xm6
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm5
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm5
+    movhps          [r2 + r6], xm2
+%else
+    psubw           m0, m7                          ; m0 = word: row 8, row 9
+    psubw           m6, m7                          ; m6 = word: row 10, row 11
+    psubw           m5, m7                          ; m5 = word: row 12, row 13
+    psubw           m2, m7                          ; m2 = word: row 14, row 15
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m6, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r6], xm3
+    lea             r2, [r2 + r3 * 4]
+    vextracti128    xm1, m5, 1
+    vextracti128    xm3, m2, 1
+    movu            [r2], xm5
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
+%endif
+%endmacro
 
-    movu       [r2 + r3], m4
-    movu       [r2 + r3 + 16], m3
+%macro FILTER_VER_CHROMA_AVX2_8x16 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_8x16, 4, 7, 8
+    mov             r4d, r4m
+    shl             r4d, 6
 
-    lea        r2, [r2 + 2 * r3]
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
 
-    dec        r4d
-    jnz        .loop
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,pp
+    mova            m7, [pw_512]
+%else
+    add             r3d, r3d
+    mova            m7, [pw_2000]
+%endif
+    lea             r6, [r3 * 3]
+    PROCESS_CHROMA_AVX2_W8_16R %1
     RET
 %endmacro
 
-    FILTER_V_PS_W16 16,  4
-    FILTER_V_PS_W16 16,  8
-    FILTER_V_PS_W16 16, 12
-    FILTER_V_PS_W16 16, 16
-    FILTER_V_PS_W16 16, 32
-
-    FILTER_V_PS_W16 16, 24
-    FILTER_V_PS_W16 16, 64
-
-;--------------------------------------------------------------------------------------------------------------
-;void interp_4tap_vert_ps_24x32(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-%macro FILTER_V4_PS_W24 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_ps_24x%2, 4, 6, 8
+    FILTER_VER_CHROMA_AVX2_8x16 pp
+    FILTER_VER_CHROMA_AVX2_8x16 ps
 
-    mov        r4d, r4m
-    sub        r0, r1
-    add        r3d, r3d
+%macro FILTER_VER_CHROMA_AVX2_8x12 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_8x12, 4, 7, 8
+    mov             r4d, r4m
+    shl             r4d, 6
 
 %ifdef PIC
-    lea        r5, [tab_ChromaCoeff]
-    movd       m0, [r5 + r4 * 4]
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
 %else
-    movd       m0, [tab_ChromaCoeff + r4 * 4]
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
 %endif
 
-    pshufb     m1, m0, [tab_Vm]
-    pshufb     m0, [tab_Vm + 16]
-
-    mov        r4d, %2/2
-
-.loop:
-    movu       m2, [r0]
-    movu       m3, [r0 + r1]
-
-    punpcklbw  m4, m2, m3
-    punpckhbw  m2, m3
-
-    pmaddubsw  m4, m1
-    pmaddubsw  m2, m1
-
-    lea        r5, [r0 + 2 * r1]
-
-    movu       m5, [r5]
-    movu       m7, [r5 + r1]
-
-    punpcklbw  m6, m5, m7
-    pmaddubsw  m6, m0
-    paddw      m4, m6
-
-    punpckhbw  m6, m5, m7
-    pmaddubsw  m6, m0
-    paddw      m2, m6
-
-    mova       m6, [pw_2000]
-
-    psubw      m4, m6
-    psubw      m2, m6
-
-    movu       [r2], m4
-    movu       [r2 + 16], m2
-
-    punpcklbw  m4, m3, m5
-    punpckhbw  m3, m5
-
-    pmaddubsw  m4, m1
-    pmaddubsw  m3, m1
-
-    movu       m2, [r5 + 2 * r1]
-
-    punpcklbw  m5, m7, m2
-    punpckhbw  m7, m2
-
-    pmaddubsw  m5, m0
-    pmaddubsw  m7, m0
-
-    paddw      m4, m5
-    paddw      m3, m7
-
-    psubw      m4, m6
-    psubw      m3, m6
-
-    movu       [r2 + r3], m4
-    movu       [r2 + r3 + 16], m3
-
-    movq       m2, [r0 + 16]
-    movq       m3, [r0 + r1 + 16]
-    movq       m4, [r5 + 16]
-    movq       m5, [r5 + r1 + 16]
-
-    punpcklbw  m2, m3
-    punpcklbw  m7, m4, m5
-
-    pmaddubsw  m2, m1
-    pmaddubsw  m7, m0
-
-    paddw      m2, m7
-    psubw      m2, m6
-
-    movu       [r2 + 32], m2
-
-    movq       m2, [r5 + 2 * r1 + 16]
-
-    punpcklbw  m3, m4
-    punpcklbw  m5, m2
-
-    pmaddubsw  m3, m1
-    pmaddubsw  m5, m0
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1, pp
+    mova            m7, [pw_512]
+%else
+    add             r3d, r3d
+    mova            m7, [pw_2000]
+%endif
+    lea             r6, [r3 * 3]
+    movq            xm1, [r0]                       ; m1 = row 0
+    movq            xm2, [r0 + r1]                  ; m2 = row 1
+    punpcklbw       xm1, xm2
+    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
+    punpcklbw       xm2, xm3
+    vinserti128     m5, m1, xm2, 1
+    pmaddubsw       m5, [r5]
+    movq            xm4, [r0 + r4]                  ; m4 = row 3
+    punpcklbw       xm3, xm4
+    lea             r0, [r0 + r1 * 4]
+    movq            xm1, [r0]                       ; m1 = row 4
+    punpcklbw       xm4, xm1
+    vinserti128     m2, m3, xm4, 1
+    pmaddubsw       m0, m2, [r5 + 1 * mmsize]
+    paddw           m5, m0
+    pmaddubsw       m2, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 5
+    punpcklbw       xm1, xm3
+    movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
+    punpcklbw       xm3, xm4
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m0, m1, [r5 + 1 * mmsize]
+    paddw           m2, m0
+    pmaddubsw       m1, [r5]
+    movq            xm3, [r0 + r4]                  ; m3 = row 7
+    punpcklbw       xm4, xm3
+    lea             r0, [r0 + r1 * 4]
+    movq            xm0, [r0]                       ; m0 = row 8
+    punpcklbw       xm3, xm0
+    vinserti128     m4, m4, xm3, 1
+    pmaddubsw       m3, m4, [r5 + 1 * mmsize]
+    paddw           m1, m3
+    pmaddubsw       m4, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 9
+    punpcklbw       xm0, xm3
+    movq            xm6, [r0 + r1 * 2]              ; m6 = row 10
+    punpcklbw       xm3, xm6
+    vinserti128     m0, m0, xm3, 1
+    pmaddubsw       m3, m0, [r5 + 1 * mmsize]
+    paddw           m4, m3
+    pmaddubsw       m0, [r5]
+%ifidn %1, pp
+    pmulhrsw        m5, m7                          ; m5 = word: row 0, row 1
+    pmulhrsw        m2, m7                          ; m2 = word: row 2, row 3
+    pmulhrsw        m1, m7                          ; m1 = word: row 4, row 5
+    pmulhrsw        m4, m7                          ; m4 = word: row 6, row 7
+    packuswb        m5, m2
+    packuswb        m1, m4
+    vextracti128    xm2, m5, 1
+    vextracti128    xm4, m1, 1
+    movq            [r2], xm5
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm5
+    movhps          [r2 + r6], xm2
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm1
+    movq            [r2 + r3], xm4
+    movhps          [r2 + r3 * 2], xm1
+    movhps          [r2 + r6], xm4
+%else
+    psubw           m5, m7                          ; m5 = word: row 0, row 1
+    psubw           m2, m7                          ; m2 = word: row 2, row 3
+    psubw           m1, m7                          ; m1 = word: row 4, row 5
+    psubw           m4, m7                          ; m4 = word: row 6, row 7
+    vextracti128    xm3, m5, 1
+    movu            [r2], xm5
+    movu            [r2 + r3], xm3
+    vextracti128    xm3, m2, 1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
+    lea             r2, [r2 + r3 * 4]
+    vextracti128    xm5, m1, 1
+    vextracti128    xm3, m4, 1
+    movu            [r2], xm1
+    movu            [r2 + r3], xm5
+    movu            [r2 + r3 * 2], xm4
+    movu            [r2 + r6], xm3
+%endif
+    movq            xm3, [r0 + r4]                  ; m3 = row 11
+    punpcklbw       xm6, xm3
+    lea             r0, [r0 + r1 * 4]
+    movq            xm5, [r0]                       ; m5 = row 12
+    punpcklbw       xm3, xm5
+    vinserti128     m6, m6, xm3, 1
+    pmaddubsw       m3, m6, [r5 + 1 * mmsize]
+    paddw           m0, m3
+    pmaddubsw       m6, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 13
+    punpcklbw       xm5, xm3
+    movq            xm2, [r0 + r1 * 2]              ; m2 = row 14
+    punpcklbw       xm3, xm2
+    vinserti128     m5, m5, xm3, 1
+    pmaddubsw       m3, m5, [r5 + 1 * mmsize]
+    paddw           m6, m3
+    lea             r2, [r2 + r3 * 4]
+%ifidn %1, pp
+    pmulhrsw        m0, m7                          ; m0 = word: row 8, row 9
+    pmulhrsw        m6, m7                          ; m6 = word: row 10, row 11
+    packuswb        m0, m6
+    vextracti128    xm6, m0, 1
+    movq            [r2], xm0
+    movq            [r2 + r3], xm6
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r6], xm6
+%else
+    psubw           m0, m7                          ; m0 = word: row 8, row 9
+    psubw           m6, m7                          ; m6 = word: row 10, row 11
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m6, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r6], xm3
+%endif
+    RET
+%endmacro
 
-    paddw      m3, m5
-    psubw      m3,  m6
+    FILTER_VER_CHROMA_AVX2_8x12 pp
+    FILTER_VER_CHROMA_AVX2_8x12 ps
 
-    movu       [r2 + r3 + 32], m3
+%macro FILTER_VER_CHROMA_AVX2_8xN 2
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_8x%2, 4, 7, 8
+    mov             r4d, r4m
+    shl             r4d, 6
 
-    mov        r0, r5
-    lea        r2, [r2 + 2 * r3]
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
 
-    dec        r4d
-    jnz        .loop
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,pp
+    mova            m7, [pw_512]
+%else
+    add             r3d, r3d
+    mova            m7, [pw_2000]
+%endif
+    lea             r6, [r3 * 3]
+%rep %2 / 16
+    PROCESS_CHROMA_AVX2_W8_16R %1
+    lea             r2, [r2 + r3 * 4]
+%endrep
     RET
 %endmacro
 
-    FILTER_V4_PS_W24 24, 32
-
-    FILTER_V4_PS_W24 24, 64
+    FILTER_VER_CHROMA_AVX2_8xN pp, 32
+    FILTER_VER_CHROMA_AVX2_8xN ps, 32
+    FILTER_VER_CHROMA_AVX2_8xN pp, 64
+    FILTER_VER_CHROMA_AVX2_8xN ps, 64
 
-;---------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ps_32x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;---------------------------------------------------------------------------------------------------------------
-%macro FILTER_V_PS_W32 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_ps_%1x%2, 4, 6, 8
+%macro PROCESS_CHROMA_AVX2_W8_4R 0
+    movq            xm1, [r0]                       ; m1 = row 0
+    movq            xm2, [r0 + r1]                  ; m2 = row 1
+    punpcklbw       xm1, xm2                        ; m1 = [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
+    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
+    punpcklbw       xm2, xm3                        ; m2 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10]
+    vinserti128     m0, m1, xm2, 1                  ; m0 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10] - [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
+    pmaddubsw       m0, [r5]
+    movq            xm4, [r0 + r4]                  ; m4 = row 3
+    punpcklbw       xm3, xm4                        ; m3 = [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
+    lea             r0, [r0 + r1 * 4]
+    movq            xm1, [r0]                       ; m1 = row 4
+    punpcklbw       xm4, xm1                        ; m4 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30]
+    vinserti128     m2, m3, xm4, 1                  ; m2 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30] - [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
+    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
+    paddw           m0, m4
+    pmaddubsw       m2, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 5
+    punpcklbw       xm1, xm3                        ; m1 = [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
+    movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
+    punpcklbw       xm3, xm4                        ; m3 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50]
+    vinserti128     m1, m1, xm3, 1                  ; m1 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50] - [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
+    pmaddubsw       m1, [r5 + 1 * mmsize]
+    paddw           m2, m1
+%endmacro
 
-    mov        r4d, r4m
-    sub        r0, r1
-    add        r3d, r3d
+%macro FILTER_VER_CHROMA_AVX2_8x4 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_8x4, 4, 6, 5
+    mov             r4d, r4m
+    shl             r4d, 6
 
 %ifdef PIC
-    lea        r5, [tab_ChromaCoeff]
-    movd       m0, [r5 + r4 * 4]
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
 %else
-    movd       m0, [tab_ChromaCoeff + r4 * 4]
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
 %endif
 
-    pshufb     m1, m0, [tab_Vm]
-    pshufb     m0, [tab_Vm + 16]
-
-    mova       m7, [pw_2000]
-
-    mov        r4d, %2
-
-.loop:
-    movu       m2, [r0]
-    movu       m3, [r0 + r1]
-
-    punpcklbw  m4, m2, m3
-    punpckhbw  m2, m3
-
-    pmaddubsw  m4, m1
-    pmaddubsw  m2, m1
-
-    lea        r5, [r0 + 2 * r1]
-    movu       m3, [r5]
-    movu       m5, [r5 + r1]
-
-    punpcklbw  m6, m3, m5
-    punpckhbw  m3, m5
-
-    pmaddubsw  m6, m0
-    pmaddubsw  m3, m0
-
-    paddw      m4, m6
-    paddw      m2, m3
-
-    psubw      m4, m7
-    psubw      m2, m7
-
-    movu       [r2], m4
-    movu       [r2 + 16], m2
-
-    movu       m2, [r0 + 16]
-    movu       m3, [r0 + r1 + 16]
-
-    punpcklbw  m4, m2, m3
-    punpckhbw  m2, m3
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+    PROCESS_CHROMA_AVX2_W8_4R
+%ifidn %1,pp
+    lea             r4, [r3 * 3]
+    mova            m3, [pw_512]
+    pmulhrsw        m0, m3                          ; m0 = word: row 0, row 1
+    pmulhrsw        m2, m3                          ; m2 = word: row 2, row 3
+    packuswb        m0, m2
+    vextracti128    xm2, m0, 1
+    movq            [r2], xm0
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r4], xm2
+%else
+    add             r3d, r3d
+    vbroadcasti128  m3, [pw_2000]
+    lea             r4, [r3 * 3]
+    psubw           m0, m3                          ; m0 = word: row 0, row 1
+    psubw           m2, m3                          ; m2 = word: row 2, row 3
+    vextracti128    xm1, m0, 1
+    vextracti128    xm4, m2, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r4], xm4
+%endif
+    RET
+%endmacro
 
-    pmaddubsw  m4, m1
-    pmaddubsw  m2, m1
+    FILTER_VER_CHROMA_AVX2_8x4 pp
+    FILTER_VER_CHROMA_AVX2_8x4 ps
 
-    movu       m3, [r5 + 16]
-    movu       m5, [r5 + r1 + 16]
+%macro FILTER_VER_CHROMA_AVX2_8x2 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_8x2, 4, 6, 4
+    mov             r4d, r4m
+    shl             r4d, 6
 
-    punpcklbw  m6, m3, m5
-    punpckhbw  m3, m5
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
 
-    pmaddubsw  m6, m0
-    pmaddubsw  m3, m0
+    lea             r4, [r1 * 3]
+    sub             r0, r1
 
-    paddw      m4, m6
-    paddw      m2, m3
+    movq            xm1, [r0]                       ; m1 = row 0
+    movq            xm2, [r0 + r1]                  ; m2 = row 1
+    punpcklbw       xm1, xm2                        ; m1 = [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
+    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
+    punpcklbw       xm2, xm3                        ; m2 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10]
+    vinserti128     m1, m1, xm2, 1                  ; m1 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10] - [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
+    pmaddubsw       m1, [r5]
+    movq            xm2, [r0 + r4]                  ; m2 = row 3
+    punpcklbw       xm3, xm2                        ; m3 = [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
+    movq            xm0, [r0 + r1 * 4]              ; m0 = row 4
+    punpcklbw       xm2, xm0                        ; m2 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30]
+    vinserti128     m3, m3, xm2, 1                  ; m3 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30] - [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
+    pmaddubsw       m3, [r5 + 1 * mmsize]
+    paddw           m1, m3
+%ifidn %1,pp
+    pmulhrsw        m1, [pw_512]                    ; m1 = word: row 0, row 1
+    packuswb        m1, m1
+    vextracti128    xm0, m1, 1
+    movq            [r2], xm1
+    movq            [r2 + r3], xm0
+%else
+    add             r3d, r3d
+    psubw           m1, [pw_2000]                   ; m1 = word: row 0, row 1
+    vextracti128    xm0, m1, 1
+    movu            [r2], xm1
+    movu            [r2 + r3], xm0
+%endif
+    RET
+%endmacro
 
-    psubw      m4, m7
-    psubw      m2, m7
+    FILTER_VER_CHROMA_AVX2_8x2 pp
+    FILTER_VER_CHROMA_AVX2_8x2 ps
 
-    movu       [r2 + 32], m4
-    movu       [r2 + 48], m2
+%macro FILTER_VER_CHROMA_AVX2_6x8 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_6x8, 4, 6, 7
+    mov             r4d, r4m
+    shl             r4d, 6
 
-    lea        r0, [r0 + r1]
-    lea        r2, [r2 + r3]
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
 
-    dec        r4d
-    jnz        .loop
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+    PROCESS_CHROMA_AVX2_W8_8R
+%ifidn %1,pp
+    lea             r4, [r3 * 3]
+    mova            m3, [pw_512]
+    pmulhrsw        m5, m3                          ; m5 = word: row 0, row 1
+    pmulhrsw        m2, m3                          ; m2 = word: row 2, row 3
+    pmulhrsw        m1, m3                          ; m1 = word: row 4, row 5
+    pmulhrsw        m4, m3                          ; m4 = word: row 6, row 7
+    packuswb        m5, m2
+    packuswb        m1, m4
+    vextracti128    xm2, m5, 1
+    vextracti128    xm4, m1, 1
+    movd            [r2], xm5
+    pextrw          [r2 + 4], xm5, 2
+    movd            [r2 + r3], xm2
+    pextrw          [r2 + r3 + 4], xm2, 2
+    pextrd          [r2 + r3 * 2], xm5, 2
+    pextrw          [r2 + r3 * 2 + 4], xm5, 6
+    pextrd          [r2 + r4], xm2, 2
+    pextrw          [r2 + r4 + 4], xm2, 6
+    lea             r2, [r2 + r3 * 4]
+    movd            [r2], xm1
+    pextrw          [r2 + 4], xm1, 2
+    movd            [r2 + r3], xm4
+    pextrw          [r2 + r3 + 4], xm4, 2
+    pextrd          [r2 + r3 * 2], xm1, 2
+    pextrw          [r2 + r3 * 2 + 4], xm1, 6
+    pextrd          [r2 + r4], xm4, 2
+    pextrw          [r2 + r4 + 4], xm4, 6
+%else
+    add             r3d, r3d
+    vbroadcasti128  m3, [pw_2000]
+    lea             r4, [r3 * 3]
+    psubw           m5, m3                          ; m5 = word: row 0, row 1
+    psubw           m2, m3                          ; m2 = word: row 2, row 3
+    psubw           m1, m3                          ; m1 = word: row 4, row 5
+    psubw           m4, m3                          ; m4 = word: row 6, row 7
+    vextracti128    xm6, m5, 1
+    vextracti128    xm3, m2, 1
+    vextracti128    xm0, m1, 1
+    movq            [r2], xm5
+    pextrd          [r2 + 8], xm5, 2
+    movq            [r2 + r3], xm6
+    pextrd          [r2 + r3 + 8], xm6, 2
+    movq            [r2 + r3 * 2], xm2
+    pextrd          [r2 + r3 * 2 + 8], xm2, 2
+    movq            [r2 + r4], xm3
+    pextrd          [r2 + r4 + 8], xm3, 2
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm1
+    pextrd          [r2 + 8], xm1, 2
+    movq            [r2 + r3], xm0
+    pextrd          [r2 + r3 + 8], xm0, 2
+    movq            [r2 + r3 * 2], xm4
+    pextrd          [r2 + r3 * 2 + 8], xm4, 2
+    vextracti128    xm4, m4, 1
+    movq            [r2 + r4], xm4
+    pextrd          [r2 + r4 + 8], xm4, 2
+%endif
     RET
 %endmacro
 
-    FILTER_V_PS_W32 32,  8
-    FILTER_V_PS_W32 32, 16
-    FILTER_V_PS_W32 32, 24
-    FILTER_V_PS_W32 32, 32
-
-    FILTER_V_PS_W32 32, 48
-    FILTER_V_PS_W32 32, 64
+    FILTER_VER_CHROMA_AVX2_6x8 pp
+    FILTER_VER_CHROMA_AVX2_6x8 ps
 
 ;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;void interp_4tap_vert_pp_6x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
 ;-----------------------------------------------------------------------------
-%macro FILTER_V4_W8_H8_H16_H32 2
+%macro FILTER_V4_W6_H4 2
 INIT_XMM sse4
-cglobal interp_4tap_vert_pp_%1x%2, 4, 6, 8
+cglobal interp_4tap_vert_pp_6x%2, 4, 6, 8
 
     mov         r4d,       r4m
     sub         r0,        r1
@@ -7839,9 +5906,9 @@ cglobal interp_4tap_vert_pp_%1x%2, 4, 6, 8
     pshufb      m6,        m5,       [tab_Vm]
     pshufb      m5,        [tab_Vm + 16]
     mova        m4,        [pw_512]
-    lea         r5,        [r1 * 3]
 
     mov         r4d,       %2
+    lea         r5,        [3 * r1]
 
 .loop:
     movq        m0,        [r0]
@@ -7860,11 +5927,12 @@ cglobal interp_4tap_vert_pp_%1x%2, 4, 6, 8
 
     pmulhrsw    m0,        m4
     packuswb    m0,        m0
-    movh        [r2],      m0
+    movd        [r2],      m0
+    pextrw      [r2 + 4],  m0,    2
 
     lea         r0,        [r0 + 4 * r1]
-    movq        m0,        [r0]
 
+    movq        m0,        [r0]
     punpcklbw   m3,        m0
 
     pmaddubsw   m1,        m6
@@ -7874,21 +5942,25 @@ cglobal interp_4tap_vert_pp_%1x%2, 4, 6, 8
 
     pmulhrsw    m1,        m4
     packuswb    m1,        m1
-    movh        [r2 + r3], m1
+    movd        [r2 + r3],      m1
+    pextrw      [r2 + r3 + 4],  m1,    2
 
     movq        m1,        [r0 + r1]
-
-    punpcklbw   m0,        m1
+    punpcklbw   m7,        m0,        m1
 
     pmaddubsw   m2,        m6
-    pmaddubsw   m0,        m5
+    pmaddubsw   m7,        m5
 
-    paddw       m2,        m0
+    paddw       m2,        m7
 
     pmulhrsw    m2,        m4
+    packuswb    m2,        m2
+    lea         r2,        [r2 + 2 * r3]
+    movd        [r2],      m2
+    pextrw      [r2 + 4],  m2,    2
 
-    movq        m7,        [r0 + 2 * r1]
-    punpcklbw   m1,        m7
+    movq        m2,        [r0 + 2 * r1]
+    punpcklbw   m1,        m2
 
     pmaddubsw   m3,        m6
     pmaddubsw   m1,        m5
@@ -7896,11 +5968,10 @@ cglobal interp_4tap_vert_pp_%1x%2, 4, 6, 8
     paddw       m3,        m1
 
     pmulhrsw    m3,        m4
-    packuswb    m2,        m3
+    packuswb    m3,        m3
 
-    lea         r2,        [r2 + 2 * r3]
-    movh        [r2],      m2
-    movhps      [r2 + r3], m2
+    movd        [r2 + r3],        m3
+    pextrw      [r2 + r3 + 4],    m3,    2
 
     lea         r2,        [r2 + 2 * r3]
 
@@ -7909,59 +5980,197 @@ cglobal interp_4tap_vert_pp_%1x%2, 4, 6, 8
     RET
 %endmacro
 
-    FILTER_V4_W8_H8_H16_H32 8,  8
-    FILTER_V4_W8_H8_H16_H32 8, 16
-    FILTER_V4_W8_H8_H16_H32 8, 32
+    FILTER_V4_W6_H4 6, 8
 
-    FILTER_V4_W8_H8_H16_H32 8, 12
-    FILTER_V4_W8_H8_H16_H32 8, 64
+    FILTER_V4_W6_H4 6, 16
 
-%macro PROCESS_CHROMA_AVX2_W8_8R 0
-    movq            xm1, [r0]                       ; m1 = row 0
-    movq            xm2, [r0 + r1]                  ; m2 = row 1
-    punpcklbw       xm1, xm2                        ; m1 = [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
-    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
-    punpcklbw       xm2, xm3                        ; m2 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10]
-    vinserti128     m5, m1, xm2, 1                  ; m5 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10] - [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
-    pmaddubsw       m5, [r5]
-    movq            xm4, [r0 + r4]                  ; m4 = row 3
-    punpcklbw       xm3, xm4                        ; m3 = [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
-    lea             r0, [r0 + r1 * 4]
-    movq            xm1, [r0]                       ; m1 = row 4
-    punpcklbw       xm4, xm1                        ; m4 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30]
-    vinserti128     m2, m3, xm4, 1                  ; m2 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30] - [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
-    pmaddubsw       m0, m2, [r5 + 1 * mmsize]
-    paddw           m5, m0
-    pmaddubsw       m2, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 5
-    punpcklbw       xm1, xm3                        ; m1 = [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
-    movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
-    punpcklbw       xm3, xm4                        ; m3 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50]
-    vinserti128     m1, m1, xm3, 1                  ; m1 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50] - [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
-    pmaddubsw       m0, m1, [r5 + 1 * mmsize]
-    paddw           m2, m0
-    pmaddubsw       m1, [r5]
-    movq            xm3, [r0 + r4]                  ; m3 = row 7
-    punpcklbw       xm4, xm3                        ; m4 = [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
-    lea             r0, [r0 + r1 * 4]
-    movq            xm0, [r0]                       ; m0 = row 8
-    punpcklbw       xm3, xm0                        ; m3 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70]
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70] - [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
-    pmaddubsw       m3, m4, [r5 + 1 * mmsize]
-    paddw           m1, m3
-    pmaddubsw       m4, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 9
-    punpcklbw       xm0, xm3                        ; m0 = [97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80]
-    movq            xm6, [r0 + r1 * 2]              ; m6 = row 10
-    punpcklbw       xm3, xm6                        ; m3 = [A7 97 A6 96 A5 95 A4 94 A3 93 A2 92 A1 91 A0 90]
-    vinserti128     m0, m0, xm3, 1                  ; m0 = [A7 97 A6 96 A5 95 A4 94 A3 93 A2 92 A1 91 A0 90] - [97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80]
-    pmaddubsw       m0, [r5 + 1 * mmsize]
-    paddw           m4, m0
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_pp_12x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W12_H2 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_pp_12x%2, 4, 6, 8
+
+    mov         r4d,       r4m
+    sub         r0,        r1
+
+%ifdef PIC
+    lea         r5,        [tab_ChromaCoeff]
+    movd        m0,        [r5 + r4 * 4]
+%else
+    movd        m0,        [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb      m1,        m0,       [tab_Vm]
+    pshufb      m0,        [tab_Vm + 16]
+
+    mov         r4d,       %2
+
+.loop:
+    movu        m2,        [r0]
+    movu        m3,        [r0 + r1]
+
+    punpcklbw   m4,        m2,        m3
+    punpckhbw   m2,        m3
+
+    pmaddubsw   m4,        m1
+    pmaddubsw   m2,        m1
+
+    lea         r0,        [r0 + 2 * r1]
+    movu        m5,        [r0]
+    movu        m7,        [r0 + r1]
+
+    punpcklbw   m6,        m5,        m7
+    pmaddubsw   m6,        m0
+    paddw       m4,        m6
+
+    punpckhbw   m6,        m5,        m7
+    pmaddubsw   m6,        m0
+    paddw       m2,        m6
+
+    mova        m6,        [pw_512]
+
+    pmulhrsw    m4,        m6
+    pmulhrsw    m2,        m6
+
+    packuswb    m4,        m2
+
+    movh         [r2],     m4
+    pextrd       [r2 + 8], m4,  2
+
+    punpcklbw   m4,        m3,        m5
+    punpckhbw   m3,        m5
+
+    pmaddubsw   m4,        m1
+    pmaddubsw   m3,        m1
+
+    movu        m5,        [r0 + 2 * r1]
+
+    punpcklbw   m2,        m7,        m5
+    punpckhbw   m7,        m5
+
+    pmaddubsw   m2,        m0
+    pmaddubsw   m7,        m0
+
+    paddw       m4,        m2
+    paddw       m3,        m7
+
+    pmulhrsw    m4,        m6
+    pmulhrsw    m3,        m6
+
+    packuswb    m4,        m3
+
+    movh        [r2 + r3],      m4
+    pextrd      [r2 + r3 + 8],  m4,  2
+
+    lea         r2,        [r2 + 2 * r3]
+
+    sub         r4,        2
+    jnz        .loop
+    RET
 %endmacro
 
-%macro FILTER_VER_CHROMA_AVX2_8x8 1
+    FILTER_V4_W12_H2 12, 16
+
+    FILTER_V4_W12_H2 12, 32
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_pp_16x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W16_H2 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_pp_16x%2, 4, 6, 8
+
+    mov         r4d,       r4m
+    sub         r0,        r1
+
+%ifdef PIC
+    lea         r5,        [tab_ChromaCoeff]
+    movd        m0,        [r5 + r4 * 4]
+%else
+    movd        m0,        [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb      m1,        m0,       [tab_Vm]
+    pshufb      m0,        [tab_Vm + 16]
+
+    mov         r4d,       %2/2
+
+.loop:
+    movu        m2,        [r0]
+    movu        m3,        [r0 + r1]
+
+    punpcklbw   m4,        m2,        m3
+    punpckhbw   m2,        m3
+
+    pmaddubsw   m4,        m1
+    pmaddubsw   m2,        m1
+
+    lea         r0,        [r0 + 2 * r1]
+    movu        m5,        [r0]
+    movu        m6,        [r0 + r1]
+
+    punpckhbw   m7,        m5,        m6
+    pmaddubsw   m7,        m0
+    paddw       m2,        m7
+
+    punpcklbw   m7,        m5,        m6
+    pmaddubsw   m7,        m0
+    paddw       m4,        m7
+
+    mova        m7,        [pw_512]
+
+    pmulhrsw    m4,        m7
+    pmulhrsw    m2,        m7
+
+    packuswb    m4,        m2
+
+    movu        [r2],      m4
+
+    punpcklbw   m4,        m3,        m5
+    punpckhbw   m3,        m5
+
+    pmaddubsw   m4,        m1
+    pmaddubsw   m3,        m1
+
+    movu        m5,        [r0 + 2 * r1]
+
+    punpcklbw   m2,        m6,        m5
+    punpckhbw   m6,        m5
+
+    pmaddubsw   m2,        m0
+    pmaddubsw   m6,        m0
+
+    paddw       m4,        m2
+    paddw       m3,        m6
+
+    pmulhrsw    m4,        m7
+    pmulhrsw    m3,        m7
+
+    packuswb    m4,        m3
+
+    movu        [r2 + r3],      m4
+
+    lea         r2,        [r2 + 2 * r3]
+
+    dec         r4d
+    jnz        .loop
+    RET
+%endmacro
+
+    FILTER_V4_W16_H2 16,  4
+    FILTER_V4_W16_H2 16,  8
+    FILTER_V4_W16_H2 16, 12
+    FILTER_V4_W16_H2 16, 16
+    FILTER_V4_W16_H2 16, 32
+
+    FILTER_V4_W16_H2 16, 24
+    FILTER_V4_W16_H2 16, 64
+
+%macro FILTER_VER_CHROMA_AVX2_16x16 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x8, 4, 6, 7
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_vert_%1_16x16, 4, 6, 15
     mov             r4d, r4m
     shl             r4d, 6
 
@@ -7972,60 +6181,255 @@ cglobal interp_4tap_vert_%1_8x8, 4, 6, 7
     lea             r5, [tab_ChromaCoeffVer_32 + r4]
 %endif
 
+    mova            m12, [r5]
+    mova            m13, [r5 + mmsize]
     lea             r4, [r1 * 3]
     sub             r0, r1
-    PROCESS_CHROMA_AVX2_W8_8R
 %ifidn %1,pp
-    lea             r4, [r3 * 3]
-    mova            m3, [pw_512]
-    pmulhrsw        m5, m3                          ; m5 = word: row 0, row 1
-    pmulhrsw        m2, m3                          ; m2 = word: row 2, row 3
-    pmulhrsw        m1, m3                          ; m1 = word: row 4, row 5
-    pmulhrsw        m4, m3                          ; m4 = word: row 6, row 7
-    packuswb        m5, m2
-    packuswb        m1, m4
-    vextracti128    xm2, m5, 1
-    vextracti128    xm4, m1, 1
-    movq            [r2], xm5
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm5
-    movhps          [r2 + r4], xm2
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm1
-    movq            [r2 + r3], xm4
-    movhps          [r2 + r3 * 2], xm1
-    movhps          [r2 + r4], xm4
+    mova            m14, [pw_512]
 %else
     add             r3d, r3d
-    vbroadcasti128  m3, [pw_2000]
-    lea             r4, [r3 * 3]
-    psubw           m5, m3                          ; m5 = word: row 0, row 1
-    psubw           m2, m3                          ; m2 = word: row 2, row 3
-    psubw           m1, m3                          ; m1 = word: row 4, row 5
-    psubw           m4, m3                          ; m4 = word: row 6, row 7
-    vextracti128    xm6, m5, 1
+    vbroadcasti128  m14, [pw_2000]
+%endif
+    lea             r5, [r3 * 3]
+
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddubsw       m0, m12
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m1, m12
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m4, m2, m13
+    paddw           m0, m4
+    pmaddubsw       m2, m12
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddubsw       m5, m3, m13
+    paddw           m1, m5
+    pmaddubsw       m3, m12
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhbw       xm6, xm4, xm5
+    punpcklbw       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddubsw       m6, m4, m13
+    paddw           m2, m6
+    pmaddubsw       m4, m12
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhbw       xm7, xm5, xm6
+    punpcklbw       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddubsw       m7, m5, m13
+    paddw           m3, m7
+    pmaddubsw       m5, m12
+    movu            xm7, [r0 + r4]                  ; m7 = row 7
+    punpckhbw       xm8, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddubsw       m8, m6, m13
+    paddw           m4, m8
+    pmaddubsw       m6, m12
+    lea             r0, [r0 + r1 * 4]
+    movu            xm8, [r0]                       ; m8 = row 8
+    punpckhbw       xm9, xm7, xm8
+    punpcklbw       xm7, xm8
+    vinserti128     m7, m7, xm9, 1
+    pmaddubsw       m9, m7, m13
+    paddw           m5, m9
+    pmaddubsw       m7, m12
+    movu            xm9, [r0 + r1]                  ; m9 = row 9
+    punpckhbw       xm10, xm8, xm9
+    punpcklbw       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddubsw       m10, m8, m13
+    paddw           m6, m10
+    pmaddubsw       m8, m12
+    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
+    punpckhbw       xm11, xm9, xm10
+    punpcklbw       xm9, xm10
+    vinserti128     m9, m9, xm11, 1
+    pmaddubsw       m11, m9, m13
+    paddw           m7, m11
+    pmaddubsw       m9, m12
+
+%ifidn %1,pp
+    pmulhrsw        m0, m14                         ; m0 = word: row 0
+    pmulhrsw        m1, m14                         ; m1 = word: row 1
+    pmulhrsw        m2, m14                         ; m2 = word: row 2
+    pmulhrsw        m3, m14                         ; m3 = word: row 3
+    pmulhrsw        m4, m14                         ; m4 = word: row 4
+    pmulhrsw        m5, m14                         ; m5 = word: row 5
+    pmulhrsw        m6, m14                         ; m6 = word: row 6
+    pmulhrsw        m7, m14                         ; m7 = word: row 7
+    packuswb        m0, m1
+    packuswb        m2, m3
+    packuswb        m4, m5
+    packuswb        m6, m7
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    vpermq          m4, m4, 11011000b
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm1, m0, 1
     vextracti128    xm3, m2, 1
-    vextracti128    xm0, m1, 1
-    movu            [r2], xm5
-    movu            [r2 + r3], xm6
+    vextracti128    xm5, m4, 1
+    vextracti128    xm7, m6, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
     movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r4], xm3
+    movu            [r2 + r5], xm3
     lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm1
-    movu            [r2 + r3], xm0
-    movu            [r2 + r3 * 2], xm4
-    vextracti128    xm4, m4, 1
-    movu            [r2 + r4], xm4
+    movu            [r2], xm4
+    movu            [r2 + r3], xm5
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r5], xm7
+%else
+    psubw           m0, m14                         ; m0 = word: row 0
+    psubw           m1, m14                         ; m1 = word: row 1
+    psubw           m2, m14                         ; m2 = word: row 2
+    psubw           m3, m14                         ; m3 = word: row 3
+    psubw           m4, m14                         ; m4 = word: row 4
+    psubw           m5, m14                         ; m5 = word: row 5
+    psubw           m6, m14                         ; m6 = word: row 6
+    psubw           m7, m14                         ; m7 = word: row 7
+    movu            [r2], m0
+    movu            [r2 + r3], m1
+    movu            [r2 + r3 * 2], m2
+    movu            [r2 + r5], m3
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], m4
+    movu            [r2 + r3], m5
+    movu            [r2 + r3 * 2], m6
+    movu            [r2 + r5], m7
+%endif
+    lea             r2, [r2 + r3 * 4]
+
+    movu            xm11, [r0 + r4]                 ; m11 = row 11
+    punpckhbw       xm6, xm10, xm11
+    punpcklbw       xm10, xm11
+    vinserti128     m10, m10, xm6, 1
+    pmaddubsw       m6, m10, m13
+    paddw           m8, m6
+    pmaddubsw       m10, m12
+    lea             r0, [r0 + r1 * 4]
+    movu            xm6, [r0]                       ; m6 = row 12
+    punpckhbw       xm7, xm11, xm6
+    punpcklbw       xm11, xm6
+    vinserti128     m11, m11, xm7, 1
+    pmaddubsw       m7, m11, m13
+    paddw           m9, m7
+    pmaddubsw       m11, m12
+
+    movu            xm7, [r0 + r1]                  ; m7 = row 13
+    punpckhbw       xm0, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm0, 1
+    pmaddubsw       m0, m6, m13
+    paddw           m10, m0
+    pmaddubsw       m6, m12
+    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
+    punpckhbw       xm1, xm7, xm0
+    punpcklbw       xm7, xm0
+    vinserti128     m7, m7, xm1, 1
+    pmaddubsw       m1, m7, m13
+    paddw           m11, m1
+    pmaddubsw       m7, m12
+    movu            xm1, [r0 + r4]                  ; m1 = row 15
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddubsw       m2, m0, m13
+    paddw           m6, m2
+    pmaddubsw       m0, m12
+    lea             r0, [r0 + r1 * 4]
+    movu            xm2, [r0]                       ; m2 = row 16
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m3, m1, m13
+    paddw           m7, m3
+    pmaddubsw       m1, m12
+    movu            xm3, [r0 + r1]                  ; m3 = row 17
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m2, m13
+    paddw           m0, m2
+    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddubsw       m3, m13
+    paddw           m1, m3
+
+%ifidn %1,pp
+    pmulhrsw        m8, m14                         ; m8 = word: row 8
+    pmulhrsw        m9, m14                         ; m9 = word: row 9
+    pmulhrsw        m10, m14                        ; m10 = word: row 10
+    pmulhrsw        m11, m14                        ; m11 = word: row 11
+    pmulhrsw        m6, m14                         ; m6 = word: row 12
+    pmulhrsw        m7, m14                         ; m7 = word: row 13
+    pmulhrsw        m0, m14                         ; m0 = word: row 14
+    pmulhrsw        m1, m14                         ; m1 = word: row 15
+    packuswb        m8, m9
+    packuswb        m10, m11
+    packuswb        m6, m7
+    packuswb        m0, m1
+    vpermq          m8, m8, 11011000b
+    vpermq          m10, m10, 11011000b
+    vpermq          m6, m6, 11011000b
+    vpermq          m0, m0, 11011000b
+    vextracti128    xm9, m8, 1
+    vextracti128    xm11, m10, 1
+    vextracti128    xm7, m6, 1
+    vextracti128    xm1, m0, 1
+    movu            [r2], xm8
+    movu            [r2 + r3], xm9
+    movu            [r2 + r3 * 2], xm10
+    movu            [r2 + r5], xm11
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm6
+    movu            [r2 + r3], xm7
+    movu            [r2 + r3 * 2], xm0
+    movu            [r2 + r5], xm1
+%else
+    psubw           m8, m14                         ; m8 = word: row 8
+    psubw           m9, m14                         ; m9 = word: row 9
+    psubw           m10, m14                        ; m10 = word: row 10
+    psubw           m11, m14                        ; m11 = word: row 11
+    psubw           m6, m14                         ; m6 = word: row 12
+    psubw           m7, m14                         ; m7 = word: row 13
+    psubw           m0, m14                         ; m0 = word: row 14
+    psubw           m1, m14                         ; m1 = word: row 15
+    movu            [r2], m8
+    movu            [r2 + r3], m9
+    movu            [r2 + r3 * 2], m10
+    movu            [r2 + r5], m11
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], m6
+    movu            [r2 + r3], m7
+    movu            [r2 + r3 * 2], m0
+    movu            [r2 + r5], m1
 %endif
     RET
+%endif
 %endmacro
 
-    FILTER_VER_CHROMA_AVX2_8x8 pp
-    FILTER_VER_CHROMA_AVX2_8x8 ps
-
-%macro FILTER_VER_CHROMA_AVX2_8x6 1
+    FILTER_VER_CHROMA_AVX2_16x16 pp
+    FILTER_VER_CHROMA_AVX2_16x16 ps
+%macro FILTER_VER_CHROMA_AVX2_16x8 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x6, 4, 6, 6
+cglobal interp_4tap_vert_%1_16x8, 4, 7, 7
     mov             r4d, r4m
     shl             r4d, 6
 
@@ -8038,264 +6442,151 @@ cglobal interp_4tap_vert_%1_8x6, 4, 6, 6
 
     lea             r4, [r1 * 3]
     sub             r0, r1
+%ifidn %1,pp
+    mova            m6, [pw_512]
+%else
+    add             r3d, r3d
+    mova            m6, [pw_2000]
+%endif
+    lea             r6, [r3 * 3]
 
-    movq            xm1, [r0]                       ; m1 = row 0
-    movq            xm2, [r0 + r1]                  ; m2 = row 1
-    punpcklbw       xm1, xm2                        ; m1 = [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
-    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
-    punpcklbw       xm2, xm3                        ; m2 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10]
-    vinserti128     m5, m1, xm2, 1                  ; m5 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10] - [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
-    pmaddubsw       m5, [r5]
-    movq            xm4, [r0 + r4]                  ; m4 = row 3
-    punpcklbw       xm3, xm4                        ; m3 = [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
-    lea             r0, [r0 + r1 * 4]
-    movq            xm1, [r0]                       ; m1 = row 4
-    punpcklbw       xm4, xm1                        ; m4 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30]
-    vinserti128     m2, m3, xm4, 1                  ; m2 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30] - [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
-    pmaddubsw       m0, m2, [r5 + 1 * mmsize]
-    paddw           m5, m0
-    pmaddubsw       m2, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 5
-    punpcklbw       xm1, xm3                        ; m1 = [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
-    movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
-    punpcklbw       xm3, xm4                        ; m3 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50]
-    vinserti128     m1, m1, xm3, 1                  ; m1 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50] - [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
-    pmaddubsw       m0, m1, [r5 + 1 * mmsize]
-    paddw           m2, m0
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddubsw       m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
     pmaddubsw       m1, [r5]
-    movq            xm3, [r0 + r4]                  ; m3 = row 7
-    punpcklbw       xm4, xm3                        ; m4 = [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m4, m2, [r5 + mmsize]
+    paddw           m0, m4
+    pmaddubsw       m2, [r5]
     lea             r0, [r0 + r1 * 4]
-    movq            xm0, [r0]                       ; m0 = row 8
-    punpcklbw       xm3, xm0                        ; m3 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70]
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70] - [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
-    pmaddubsw       m4, [r5 + 1 * mmsize]
-    paddw           m1, m4
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddubsw       m5, m3, [r5 + mmsize]
+    paddw           m1, m5
+    pmaddubsw       m3, [r5]
 %ifidn %1,pp
-    lea             r4, [r3 * 3]
-    mova            m3, [pw_512]
-    pmulhrsw        m5, m3                          ; m5 = word: row 0, row 1
-    pmulhrsw        m2, m3                          ; m2 = word: row 2, row 3
-    pmulhrsw        m1, m3                          ; m1 = word: row 4, row 5
-    packuswb        m5, m2
-    packuswb        m1, m1
-    vextracti128    xm2, m5, 1
-    vextracti128    xm4, m1, 1
-    movq            [r2], xm5
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm5
-    movhps          [r2 + r4], xm2
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm1
-    movq            [r2 + r3], xm4
+    pmulhrsw        m0, m6                          ; m0 = word: row 0
+    pmulhrsw        m1, m6                          ; m1 = word: row 1
+    packuswb        m0, m1
+    vpermq          m0, m0, 11011000b
+    vextracti128    xm1, m0, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
 %else
-    add             r3d, r3d
-    mova            m3, [pw_2000]
-    lea             r4, [r3 * 3]
-    psubw           m5, m3                          ; m5 = word: row 0, row 1
-    psubw           m2, m3                          ; m2 = word: row 2, row 3
-    psubw           m1, m3                          ; m1 = word: row 4, row 5
-    vextracti128    xm4, m5, 1
-    vextracti128    xm3, m2, 1
-    vextracti128    xm0, m1, 1
-    movu            [r2], xm5
-    movu            [r2 + r3], xm4
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r4], xm3
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm1
-    movu            [r2 + r3], xm0
+    psubw           m0, m6                          ; m0 = word: row 0
+    psubw           m1, m6                          ; m1 = word: row 1
+    movu            [r2], m0
+    movu            [r2 + r3], m1
 %endif
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_AVX2_8x6 pp
-    FILTER_VER_CHROMA_AVX2_8x6 ps
 
-%macro PROCESS_CHROMA_AVX2_W8_16R 1
-    movq            xm1, [r0]                       ; m1 = row 0
-    movq            xm2, [r0 + r1]                  ; m2 = row 1
-    punpcklbw       xm1, xm2
-    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
-    punpcklbw       xm2, xm3
-    vinserti128     m5, m1, xm2, 1
-    pmaddubsw       m5, [r5]
-    movq            xm4, [r0 + r4]                  ; m4 = row 3
-    punpcklbw       xm3, xm4
-    lea             r0, [r0 + r1 * 4]
-    movq            xm1, [r0]                       ; m1 = row 4
-    punpcklbw       xm4, xm1
-    vinserti128     m2, m3, xm4, 1
-    pmaddubsw       m0, m2, [r5 + 1 * mmsize]
-    paddw           m5, m0
-    pmaddubsw       m2, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 5
-    punpcklbw       xm1, xm3
-    movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
-    punpcklbw       xm3, xm4
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m0, m1, [r5 + 1 * mmsize]
-    paddw           m2, m0
-    pmaddubsw       m1, [r5]
-    movq            xm3, [r0 + r4]                  ; m3 = row 7
-    punpcklbw       xm4, xm3
-    lea             r0, [r0 + r1 * 4]
-    movq            xm0, [r0]                       ; m0 = row 8
-    punpcklbw       xm3, xm0
-    vinserti128     m4, m4, xm3, 1
-    pmaddubsw       m3, m4, [r5 + 1 * mmsize]
-    paddw           m1, m3
+    movu            xm0, [r0 + r1]                  ; m0 = row 5
+    punpckhbw       xm1, xm4, xm0
+    punpcklbw       xm4, xm0
+    vinserti128     m4, m4, xm1, 1
+    pmaddubsw       m1, m4, [r5 + mmsize]
+    paddw           m2, m1
     pmaddubsw       m4, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 9
-    punpcklbw       xm0, xm3
-    movq            xm6, [r0 + r1 * 2]              ; m6 = row 10
-    punpcklbw       xm3, xm6
-    vinserti128     m0, m0, xm3, 1
-    pmaddubsw       m3, m0, [r5 + 1 * mmsize]
-    paddw           m4, m3
+    movu            xm1, [r0 + r1 * 2]              ; m1 = row 6
+    punpckhbw       xm5, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm5, 1
+    pmaddubsw       m5, m0, [r5 + mmsize]
+    paddw           m3, m5
     pmaddubsw       m0, [r5]
 %ifidn %1,pp
-    pmulhrsw        m5, m7                          ; m5 = word: row 0, row 1
-    pmulhrsw        m2, m7                          ; m2 = word: row 2, row 3
-    pmulhrsw        m1, m7                          ; m1 = word: row 4, row 5
-    pmulhrsw        m4, m7                          ; m4 = word: row 6, row 7
-    packuswb        m5, m2
-    packuswb        m1, m4
-    vextracti128    xm2, m5, 1
-    vextracti128    xm4, m1, 1
-    movq            [r2], xm5
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm5
-    movhps          [r2 + r6], xm2
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm1
-    movq            [r2 + r3], xm4
-    movhps          [r2 + r3 * 2], xm1
-    movhps          [r2 + r6], xm4
-%else
-    psubw           m5, m7                          ; m5 = word: row 0, row 1
-    psubw           m2, m7                          ; m2 = word: row 2, row 3
-    psubw           m1, m7                          ; m1 = word: row 4, row 5
-    psubw           m4, m7                          ; m4 = word: row 6, row 7
-    vextracti128    xm3, m5, 1
-    movu            [r2], xm5
-    movu            [r2 + r3], xm3
+    pmulhrsw        m2, m6                          ; m2 = word: row 2
+    pmulhrsw        m3, m6                          ; m3 = word: row 3
+    packuswb        m2, m3
+    vpermq          m2, m2, 11011000b
     vextracti128    xm3, m2, 1
     movu            [r2 + r3 * 2], xm2
     movu            [r2 + r6], xm3
-    lea             r2, [r2 + r3 * 4]
-    vextracti128    xm5, m1, 1
-    vextracti128    xm3, m4, 1
-    movu            [r2], xm1
-    movu            [r2 + r3], xm5
-    movu            [r2 + r3 * 2], xm4
-    movu            [r2 + r6], xm3
+%else
+    psubw           m2, m6                          ; m2 = word: row 2
+    psubw           m3, m6                          ; m3 = word: row 3
+    movu            [r2 + r3 * 2], m2
+    movu            [r2 + r6], m3
 %endif
-    movq            xm3, [r0 + r4]                  ; m3 = row 11
-    punpcklbw       xm6, xm3
+
+    movu            xm2, [r0 + r4]                  ; m2 = row 7
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m3, m1, [r5 + mmsize]
+    paddw           m4, m3
+    pmaddubsw       m1, [r5]
     lea             r0, [r0 + r1 * 4]
-    movq            xm5, [r0]                       ; m5 = row 12
-    punpcklbw       xm3, xm5
-    vinserti128     m6, m6, xm3, 1
-    pmaddubsw       m3, m6, [r5 + 1 * mmsize]
-    paddw           m0, m3
-    pmaddubsw       m6, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 13
-    punpcklbw       xm5, xm3
-    movq            xm2, [r0 + r1 * 2]              ; m2 = row 14
-    punpcklbw       xm3, xm2
-    vinserti128     m5, m5, xm3, 1
-    pmaddubsw       m3, m5, [r5 + 1 * mmsize]
-    paddw           m6, m3
-    pmaddubsw       m5, [r5]
-    movq            xm3, [r0 + r4]                  ; m3 = row 15
+    movu            xm3, [r0]                       ; m3 = row 8
+    punpckhbw       xm5, xm2, xm3
     punpcklbw       xm2, xm3
-    lea             r0, [r0 + r1 * 4]
-    movq            xm1, [r0]                       ; m1 = row 16
-    punpcklbw       xm3, xm1
-    vinserti128     m2, m2, xm3, 1
-    pmaddubsw       m3, m2, [r5 + 1 * mmsize]
-    paddw           m5, m3
+    vinserti128     m2, m2, xm5, 1
+    pmaddubsw       m5, m2, [r5 + mmsize]
+    paddw           m0, m5
     pmaddubsw       m2, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 17
-    punpcklbw       xm1, xm3
-    movq            xm4, [r0 + r1 * 2]              ; m4 = row 18
-    punpcklbw       xm3, xm4
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m1, [r5 + 1 * mmsize]
-    paddw           m2, m1
     lea             r2, [r2 + r3 * 4]
 %ifidn %1,pp
-    pmulhrsw        m0, m7                          ; m0 = word: row 8, row 9
-    pmulhrsw        m6, m7                          ; m6 = word: row 10, row 11
-    pmulhrsw        m5, m7                          ; m5 = word: row 12, row 13
-    pmulhrsw        m2, m7                          ; m2 = word: row 14, row 15
-    packuswb        m0, m6
-    packuswb        m5, m2
-    vextracti128    xm6, m0, 1
-    vextracti128    xm2, m5, 1
-    movq            [r2], xm0
-    movq            [r2 + r3], xm6
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r6], xm6
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm5
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm5
-    movhps          [r2 + r6], xm2
-%else
-    psubw           m0, m7                          ; m0 = word: row 8, row 9
-    psubw           m6, m7                          ; m6 = word: row 10, row 11
-    psubw           m5, m7                          ; m5 = word: row 12, row 13
-    psubw           m2, m7                          ; m2 = word: row 14, row 15
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m6, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r6], xm3
-    lea             r2, [r2 + r3 * 4]
-    vextracti128    xm1, m5, 1
-    vextracti128    xm3, m2, 1
-    movu            [r2], xm5
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-%endif
-%endmacro
-
-%macro FILTER_VER_CHROMA_AVX2_8x16 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x16, 4, 7, 8
-    mov             r4d, r4m
-    shl             r4d, 6
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
+    pmulhrsw        m4, m6                          ; m4 = word: row 4
+    pmulhrsw        m0, m6                          ; m0 = word: row 5
+    packuswb        m4, m0
+    vpermq          m4, m4, 11011000b
+    vextracti128    xm0, m4, 1
+    movu            [r2], xm4
+    movu            [r2 + r3], xm0
 %else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+    psubw           m4, m6                          ; m4 = word: row 4
+    psubw           m0, m6                          ; m0 = word: row 5
+    movu            [r2], m4
+    movu            [r2 + r3], m0
 %endif
 
-    lea             r4, [r1 * 3]
-    sub             r0, r1
+    movu            xm5, [r0 + r1]                  ; m5 = row 9
+    punpckhbw       xm4, xm3, xm5
+    punpcklbw       xm3, xm5
+    vinserti128     m3, m3, xm4, 1
+    pmaddubsw       m3, [r5 + mmsize]
+    paddw           m1, m3
+    movu            xm4, [r0 + r1 * 2]              ; m4 = row 10
+    punpckhbw       xm0, xm5, xm4
+    punpcklbw       xm5, xm4
+    vinserti128     m5, m5, xm0, 1
+    pmaddubsw       m5, [r5 + mmsize]
+    paddw           m2, m5
 %ifidn %1,pp
-    mova            m7, [pw_512]
+    pmulhrsw        m1, m6                          ; m1 = word: row 6
+    pmulhrsw        m2, m6                          ; m2 = word: row 7
+    packuswb        m1, m2
+    vpermq          m1, m1, 11011000b
+    vextracti128    xm2, m1, 1
+    movu            [r2 + r3 * 2], xm1
+    movu            [r2 + r6], xm2
 %else
-    add             r3d, r3d
-    mova            m7, [pw_2000]
+    psubw           m1, m6                          ; m1 = word: row 6
+    psubw           m2, m6                          ; m2 = word: row 7
+    movu            [r2 + r3 * 2], m1
+    movu            [r2 + r6], m2
 %endif
-    lea             r6, [r3 * 3]
-    PROCESS_CHROMA_AVX2_W8_16R %1
     RET
 %endmacro
 
-    FILTER_VER_CHROMA_AVX2_8x16 pp
-    FILTER_VER_CHROMA_AVX2_8x16 ps
+    FILTER_VER_CHROMA_AVX2_16x8 pp
+    FILTER_VER_CHROMA_AVX2_16x8 ps
 
-%macro FILTER_VER_CHROMA_AVX2_8x12 1
+%macro FILTER_VER_CHROMA_AVX2_16x12 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x12, 4, 7, 8
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_vert_%1_16x12, 4, 6, 10
     mov             r4d, r4m
     shl             r4d, 6
 
@@ -8306,293 +6597,224 @@ cglobal interp_4tap_vert_%1_8x12, 4, 7, 8
     lea             r5, [tab_ChromaCoeffVer_32 + r4]
 %endif
 
+    mova            m8, [r5]
+    mova            m9, [r5 + mmsize]
     lea             r4, [r1 * 3]
     sub             r0, r1
-%ifidn %1, pp
+%ifidn %1,pp
     mova            m7, [pw_512]
 %else
     add             r3d, r3d
-    mova            m7, [pw_2000]
+    vbroadcasti128  m7, [pw_2000]
 %endif
-    lea             r6, [r3 * 3]
-    movq            xm1, [r0]                       ; m1 = row 0
-    movq            xm2, [r0 + r1]                  ; m2 = row 1
-    punpcklbw       xm1, xm2
-    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
-    punpcklbw       xm2, xm3
-    vinserti128     m5, m1, xm2, 1
-    pmaddubsw       m5, [r5]
-    movq            xm4, [r0 + r4]                  ; m4 = row 3
-    punpcklbw       xm3, xm4
-    lea             r0, [r0 + r1 * 4]
-    movq            xm1, [r0]                       ; m1 = row 4
-    punpcklbw       xm4, xm1
-    vinserti128     m2, m3, xm4, 1
-    pmaddubsw       m0, m2, [r5 + 1 * mmsize]
-    paddw           m5, m0
-    pmaddubsw       m2, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 5
-    punpcklbw       xm1, xm3
-    movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
-    punpcklbw       xm3, xm4
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m0, m1, [r5 + 1 * mmsize]
-    paddw           m2, m0
-    pmaddubsw       m1, [r5]
-    movq            xm3, [r0 + r4]                  ; m3 = row 7
-    punpcklbw       xm4, xm3
-    lea             r0, [r0 + r1 * 4]
-    movq            xm0, [r0]                       ; m0 = row 8
-    punpcklbw       xm3, xm0
-    vinserti128     m4, m4, xm3, 1
-    pmaddubsw       m3, m4, [r5 + 1 * mmsize]
-    paddw           m1, m3
-    pmaddubsw       m4, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 9
-    punpcklbw       xm0, xm3
-    movq            xm6, [r0 + r1 * 2]              ; m6 = row 10
-    punpcklbw       xm3, xm6
-    vinserti128     m0, m0, xm3, 1
-    pmaddubsw       m3, m0, [r5 + 1 * mmsize]
+    lea             r5, [r3 * 3]
+
+    movu            xm0, [r0]
+    vinserti128     m0, m0, [r0 + r1 * 2], 1
+    movu            xm1, [r0 + r1]
+    vinserti128     m1, m1, [r0 + r4], 1
+
+    punpcklbw       m2, m0, m1
+    punpckhbw       m3, m0, m1
+    vperm2i128      m4, m2, m3, 0x20
+    vperm2i128      m2, m2, m3, 0x31
+    pmaddubsw       m4, m8
+    pmaddubsw       m3, m2, m9
     paddw           m4, m3
-    pmaddubsw       m0, [r5]
-%ifidn %1, pp
-    pmulhrsw        m5, m7                          ; m5 = word: row 0, row 1
-    pmulhrsw        m2, m7                          ; m2 = word: row 2, row 3
-    pmulhrsw        m1, m7                          ; m1 = word: row 4, row 5
-    pmulhrsw        m4, m7                          ; m4 = word: row 6, row 7
-    packuswb        m5, m2
-    packuswb        m1, m4
-    vextracti128    xm2, m5, 1
-    vextracti128    xm4, m1, 1
-    movq            [r2], xm5
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm5
-    movhps          [r2 + r6], xm2
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm1
-    movq            [r2 + r3], xm4
-    movhps          [r2 + r3 * 2], xm1
-    movhps          [r2 + r6], xm4
-%else
-    psubw           m5, m7                          ; m5 = word: row 0, row 1
-    psubw           m2, m7                          ; m2 = word: row 2, row 3
-    psubw           m1, m7                          ; m1 = word: row 4, row 5
-    psubw           m4, m7                          ; m4 = word: row 6, row 7
-    vextracti128    xm3, m5, 1
-    movu            [r2], xm5
-    movu            [r2 + r3], xm3
-    vextracti128    xm3, m2, 1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-    lea             r2, [r2 + r3 * 4]
-    vextracti128    xm5, m1, 1
-    vextracti128    xm3, m4, 1
-    movu            [r2], xm1
-    movu            [r2 + r3], xm5
-    movu            [r2 + r3 * 2], xm4
-    movu            [r2 + r6], xm3
-%endif
-    movq            xm3, [r0 + r4]                  ; m3 = row 11
-    punpcklbw       xm6, xm3
+    pmaddubsw       m2, m8
+
+    vextracti128    xm0, m0, 1
     lea             r0, [r0 + r1 * 4]
-    movq            xm5, [r0]                       ; m5 = row 12
-    punpcklbw       xm3, xm5
-    vinserti128     m6, m6, xm3, 1
-    pmaddubsw       m3, m6, [r5 + 1 * mmsize]
-    paddw           m0, m3
-    pmaddubsw       m6, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 13
-    punpcklbw       xm5, xm3
-    movq            xm2, [r0 + r1 * 2]              ; m2 = row 14
-    punpcklbw       xm3, xm2
-    vinserti128     m5, m5, xm3, 1
-    pmaddubsw       m3, m5, [r5 + 1 * mmsize]
+    vinserti128     m0, m0, [r0], 1
+
+    punpcklbw       m5, m1, m0
+    punpckhbw       m3, m1, m0
+    vperm2i128      m6, m5, m3, 0x20
+    vperm2i128      m5, m5, m3, 0x31
+    pmaddubsw       m6, m8
+    pmaddubsw       m3, m5, m9
     paddw           m6, m3
-    lea             r2, [r2 + r3 * 4]
-%ifidn %1, pp
-    pmulhrsw        m0, m7                          ; m0 = word: row 8, row 9
-    pmulhrsw        m6, m7                          ; m6 = word: row 10, row 11
-    packuswb        m0, m6
-    vextracti128    xm6, m0, 1
-    movq            [r2], xm0
-    movq            [r2 + r3], xm6
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r6], xm6
+    pmaddubsw       m5, m8
+%ifidn %1,pp
+    pmulhrsw        m4, m7                         ; m4 = word: row 0
+    pmulhrsw        m6, m7                         ; m6 = word: row 1
+    packuswb        m4, m6
+    vpermq          m4, m4, 11011000b
+    vextracti128    xm6, m4, 1
+    movu            [r2], xm4
+    movu            [r2 + r3], xm6
 %else
-    psubw           m0, m7                          ; m0 = word: row 8, row 9
-    psubw           m6, m7                          ; m6 = word: row 10, row 11
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m6, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r6], xm3
+    psubw           m4, m7                         ; m4 = word: row 0
+    psubw           m6, m7                         ; m6 = word: row 1
+    movu            [r2], m4
+    movu            [r2 + r3], m6
 %endif
-    RET
-%endmacro
 
-    FILTER_VER_CHROMA_AVX2_8x12 pp
-    FILTER_VER_CHROMA_AVX2_8x12 ps
+    movu            xm4, [r0 + r1 * 2]
+    vinserti128     m4, m4, [r0 + r1], 1
+    vextracti128    xm1, m4, 1
+    vinserti128     m0, m0, xm1, 0
 
-%macro FILTER_VER_CHROMA_AVX2_8xN 2
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x%2, 4, 7, 8
-    mov             r4d, r4m
-    shl             r4d, 6
+    punpcklbw       m6, m0, m4
+    punpckhbw       m1, m0, m4
+    vperm2i128      m0, m6, m1, 0x20
+    vperm2i128      m6, m6, m1, 0x31
+    pmaddubsw       m1, m0, m9
+    paddw           m5, m1
+    pmaddubsw       m0, m8
+    pmaddubsw       m1, m6, m9
+    paddw           m2, m1
+    pmaddubsw       m6, m8
 
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
 %ifidn %1,pp
-    mova            m7, [pw_512]
+    pmulhrsw        m2, m7                         ; m2 = word: row 2
+    pmulhrsw        m5, m7                         ; m5 = word: row 3
+    packuswb        m2, m5
+    vpermq          m2, m2, 11011000b
+    vextracti128    xm5, m2, 1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r5], xm5
 %else
-    add             r3d, r3d
-    mova            m7, [pw_2000]
+    psubw           m2, m7                         ; m2 = word: row 2
+    psubw           m5, m7                         ; m5 = word: row 3
+    movu            [r2 + r3 * 2], m2
+    movu            [r2 + r5], m5
 %endif
-    lea             r6, [r3 * 3]
-%rep %2 / 16
-    PROCESS_CHROMA_AVX2_W8_16R %1
     lea             r2, [r2 + r3 * 4]
-%endrep
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_AVX2_8xN pp, 32
-    FILTER_VER_CHROMA_AVX2_8xN ps, 32
-    FILTER_VER_CHROMA_AVX2_8xN pp, 64
-    FILTER_VER_CHROMA_AVX2_8xN ps, 64
 
-%macro PROCESS_CHROMA_AVX2_W8_4R 0
-    movq            xm1, [r0]                       ; m1 = row 0
-    movq            xm2, [r0 + r1]                  ; m2 = row 1
-    punpcklbw       xm1, xm2                        ; m1 = [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
-    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
-    punpcklbw       xm2, xm3                        ; m2 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10]
-    vinserti128     m0, m1, xm2, 1                  ; m0 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10] - [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
-    pmaddubsw       m0, [r5]
-    movq            xm4, [r0 + r4]                  ; m4 = row 3
-    punpcklbw       xm3, xm4                        ; m3 = [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
+    movu            xm1, [r0 + r4]
     lea             r0, [r0 + r1 * 4]
-    movq            xm1, [r0]                       ; m1 = row 4
-    punpcklbw       xm4, xm1                        ; m4 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30]
-    vinserti128     m2, m3, xm4, 1                  ; m2 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30] - [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
-    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
-    paddw           m0, m4
-    pmaddubsw       m2, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 5
-    punpcklbw       xm1, xm3                        ; m1 = [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
-    movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
-    punpcklbw       xm3, xm4                        ; m3 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50]
-    vinserti128     m1, m1, xm3, 1                  ; m1 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50] - [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
-    pmaddubsw       m1, [r5 + 1 * mmsize]
-    paddw           m2, m1
-%endmacro
+    vinserti128     m1, m1, [r0], 1
+    vinserti128     m4, m4, xm1, 1
 
-%macro FILTER_VER_CHROMA_AVX2_8x4 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x4, 4, 6, 5
-    mov             r4d, r4m
-    shl             r4d, 6
+    punpcklbw       m2, m4, m1
+    punpckhbw       m5, m4, m1
+    vperm2i128      m3, m2, m5, 0x20
+    vperm2i128      m2, m2, m5, 0x31
+    pmaddubsw       m5, m3, m9
+    paddw           m6, m5
+    pmaddubsw       m3, m8
+    pmaddubsw       m5, m2, m9
+    paddw           m0, m5
+    pmaddubsw       m2, m8
 
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
+%ifidn %1,pp
+    pmulhrsw        m6, m7                         ; m6 = word: row 4
+    pmulhrsw        m0, m7                         ; m0 = word: row 5
+    packuswb        m6, m0
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm0, m6, 1
+    movu            [r2], xm6
+    movu            [r2 + r3], xm0
 %else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+    psubw           m6, m7                         ; m6 = word: row 4
+    psubw           m0, m7                         ; m0 = word: row 5
+    movu            [r2], m6
+    movu            [r2 + r3], m0
 %endif
 
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-    PROCESS_CHROMA_AVX2_W8_4R
+    movu            xm6, [r0 + r1 * 2]
+    vinserti128     m6, m6, [r0 + r1], 1
+    vextracti128    xm0, m6, 1
+    vinserti128     m1, m1, xm0, 0
+
+    punpcklbw       m4, m1, m6
+    punpckhbw       m5, m1, m6
+    vperm2i128      m0, m4, m5, 0x20
+    vperm2i128      m5, m4, m5, 0x31
+    pmaddubsw       m4, m0, m9
+    paddw           m2, m4
+    pmaddubsw       m0, m8
+    pmaddubsw       m4, m5, m9
+    paddw           m3, m4
+    pmaddubsw       m5, m8
+
 %ifidn %1,pp
-    lea             r4, [r3 * 3]
-    mova            m3, [pw_512]
-    pmulhrsw        m0, m3                          ; m0 = word: row 0, row 1
-    pmulhrsw        m2, m3                          ; m2 = word: row 2, row 3
-    packuswb        m0, m2
-    vextracti128    xm2, m0, 1
-    movq            [r2], xm0
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r4], xm2
+    pmulhrsw        m3, m7                         ; m3 = word: row 6
+    pmulhrsw        m2, m7                         ; m2 = word: row 7
+    packuswb        m3, m2
+    vpermq          m3, m3, 11011000b
+    vextracti128    xm2, m3, 1
+    movu            [r2 + r3 * 2], xm3
+    movu            [r2 + r5], xm2
 %else
-    add             r3d, r3d
-    vbroadcasti128  m3, [pw_2000]
-    lea             r4, [r3 * 3]
-    psubw           m0, m3                          ; m0 = word: row 0, row 1
-    psubw           m2, m3                          ; m2 = word: row 2, row 3
-    vextracti128    xm1, m0, 1
-    vextracti128    xm4, m2, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r4], xm4
+    psubw           m3, m7                         ; m3 = word: row 6
+    psubw           m2, m7                         ; m2 = word: row 7
+    movu            [r2 + r3 * 2], m3
+    movu            [r2 + r5], m2
 %endif
-    RET
-%endmacro
+    lea             r2, [r2 + r3 * 4]
 
-    FILTER_VER_CHROMA_AVX2_8x4 pp
-    FILTER_VER_CHROMA_AVX2_8x4 ps
+    movu            xm3, [r0 + r4]
+    lea             r0, [r0 + r1 * 4]
+    vinserti128     m3, m3, [r0], 1
+    vinserti128     m6, m6, xm3, 1
 
-%macro FILTER_VER_CHROMA_AVX2_8x2 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x2, 4, 6, 4
-    mov             r4d, r4m
-    shl             r4d, 6
+    punpcklbw       m2, m6, m3
+    punpckhbw       m1, m6, m3
+    vperm2i128      m4, m2, m1, 0x20
+    vperm2i128      m2, m2, m1, 0x31
+    pmaddubsw       m1, m4, m9
+    paddw           m5, m1
+    pmaddubsw       m4, m8
+    pmaddubsw       m1, m2, m9
+    paddw           m0, m1
+    pmaddubsw       m2, m8
 
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
+%ifidn %1,pp
+    pmulhrsw        m5, m7                         ; m5 = word: row 8
+    pmulhrsw        m0, m7                         ; m0 = word: row 9
+    packuswb        m5, m0
+    vpermq          m5, m5, 11011000b
+    vextracti128    xm0, m5, 1
+    movu            [r2], xm5
+    movu            [r2 + r3], xm0
 %else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+    psubw           m5, m7                         ; m5 = word: row 8
+    psubw           m0, m7                         ; m0 = word: row 9
+    movu            [r2], m5
+    movu            [r2 + r3], m0
 %endif
 
-    lea             r4, [r1 * 3]
-    sub             r0, r1
+    movu            xm5, [r0 + r1 * 2]
+    vinserti128     m5, m5, [r0 + r1], 1
+    vextracti128    xm0, m5, 1
+    vinserti128     m3, m3, xm0, 0
+
+    punpcklbw       m1, m3, m5
+    punpckhbw       m0, m3, m5
+    vperm2i128      m6, m1, m0, 0x20
+    vperm2i128      m0, m1, m0, 0x31
+    pmaddubsw       m1, m6, m9
+    paddw           m2, m1
+    pmaddubsw       m1, m0, m9
+    paddw           m4, m1
 
-    movq            xm1, [r0]                       ; m1 = row 0
-    movq            xm2, [r0 + r1]                  ; m2 = row 1
-    punpcklbw       xm1, xm2                        ; m1 = [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
-    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
-    punpcklbw       xm2, xm3                        ; m2 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10]
-    vinserti128     m1, m1, xm2, 1                  ; m1 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10] - [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
-    pmaddubsw       m1, [r5]
-    movq            xm2, [r0 + r4]                  ; m2 = row 3
-    punpcklbw       xm3, xm2                        ; m3 = [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
-    movq            xm0, [r0 + r1 * 4]              ; m0 = row 4
-    punpcklbw       xm2, xm0                        ; m2 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30]
-    vinserti128     m3, m3, xm2, 1                  ; m3 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30] - [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
-    pmaddubsw       m3, [r5 + 1 * mmsize]
-    paddw           m1, m3
 %ifidn %1,pp
-    pmulhrsw        m1, [pw_512]                    ; m1 = word: row 0, row 1
-    packuswb        m1, m1
-    vextracti128    xm0, m1, 1
-    movq            [r2], xm1
-    movq            [r2 + r3], xm0
+    pmulhrsw        m4, m7                         ; m4 = word: row 10
+    pmulhrsw        m2, m7                         ; m2 = word: row 11
+    packuswb        m4, m2
+    vpermq          m4, m4, 11011000b
+    vextracti128    xm2, m4, 1
+    movu            [r2 + r3 * 2], xm4
+    movu            [r2 + r5], xm2
 %else
-    add             r3d, r3d
-    psubw           m1, [pw_2000]                   ; m1 = word: row 0, row 1
-    vextracti128    xm0, m1, 1
-    movu            [r2], xm1
-    movu            [r2 + r3], xm0
+    psubw           m4, m7                         ; m4 = word: row 10
+    psubw           m2, m7                         ; m2 = word: row 11
+    movu            [r2 + r3 * 2], m4
+    movu            [r2 + r5], m2
 %endif
     RET
+%endif
 %endmacro
 
-    FILTER_VER_CHROMA_AVX2_8x2 pp
-    FILTER_VER_CHROMA_AVX2_8x2 ps
+    FILTER_VER_CHROMA_AVX2_16x12 pp
+    FILTER_VER_CHROMA_AVX2_16x12 ps
 
-%macro FILTER_VER_CHROMA_AVX2_6x8 1
+%macro FILTER_VER_CHROMA_AVX2_16xN 2
+%if ARCH_X86_64 == 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_6x8, 4, 6, 7
+cglobal interp_4tap_vert_%1_16x%2, 4, 8, 8
     mov             r4d, r4m
     shl             r4d, 6
 
@@ -8605,377 +6827,310 @@ cglobal interp_4tap_vert_%1_6x8, 4, 6, 7
 
     lea             r4, [r1 * 3]
     sub             r0, r1
-    PROCESS_CHROMA_AVX2_W8_8R
 %ifidn %1,pp
-    lea             r4, [r3 * 3]
-    mova            m3, [pw_512]
-    pmulhrsw        m5, m3                          ; m5 = word: row 0, row 1
-    pmulhrsw        m2, m3                          ; m2 = word: row 2, row 3
-    pmulhrsw        m1, m3                          ; m1 = word: row 4, row 5
-    pmulhrsw        m4, m3                          ; m4 = word: row 6, row 7
-    packuswb        m5, m2
-    packuswb        m1, m4
-    vextracti128    xm2, m5, 1
-    vextracti128    xm4, m1, 1
-    movd            [r2], xm5
-    pextrw          [r2 + 4], xm5, 2
-    movd            [r2 + r3], xm2
-    pextrw          [r2 + r3 + 4], xm2, 2
-    pextrd          [r2 + r3 * 2], xm5, 2
-    pextrw          [r2 + r3 * 2 + 4], xm5, 6
-    pextrd          [r2 + r4], xm2, 2
-    pextrw          [r2 + r4 + 4], xm2, 6
-    lea             r2, [r2 + r3 * 4]
-    movd            [r2], xm1
-    pextrw          [r2 + 4], xm1, 2
-    movd            [r2 + r3], xm4
-    pextrw          [r2 + r3 + 4], xm4, 2
-    pextrd          [r2 + r3 * 2], xm1, 2
-    pextrw          [r2 + r3 * 2 + 4], xm1, 6
-    pextrd          [r2 + r4], xm4, 2
-    pextrw          [r2 + r4 + 4], xm4, 6
+    mova            m7, [pw_512]
 %else
     add             r3d, r3d
-    vbroadcasti128  m3, [pw_2000]
-    lea             r4, [r3 * 3]
-    psubw           m5, m3                          ; m5 = word: row 0, row 1
-    psubw           m2, m3                          ; m2 = word: row 2, row 3
-    psubw           m1, m3                          ; m1 = word: row 4, row 5
-    psubw           m4, m3                          ; m4 = word: row 6, row 7
-    vextracti128    xm6, m5, 1
-    vextracti128    xm3, m2, 1
-    vextracti128    xm0, m1, 1
-    movq            [r2], xm5
-    pextrd          [r2 + 8], xm5, 2
-    movq            [r2 + r3], xm6
-    pextrd          [r2 + r3 + 8], xm6, 2
-    movq            [r2 + r3 * 2], xm2
-    pextrd          [r2 + r3 * 2 + 8], xm2, 2
-    movq            [r2 + r4], xm3
-    pextrd          [r2 + r4 + 8], xm3, 2
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm1
-    pextrd          [r2 + 8], xm1, 2
-    movq            [r2 + r3], xm0
-    pextrd          [r2 + r3 + 8], xm0, 2
-    movq            [r2 + r3 * 2], xm4
-    pextrd          [r2 + r3 * 2 + 8], xm4, 2
-    vextracti128    xm4, m4, 1
-    movq            [r2 + r4], xm4
-    pextrd          [r2 + r4 + 8], xm4, 2
+    mova            m7, [pw_2000]
 %endif
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_AVX2_6x8 pp
-    FILTER_VER_CHROMA_AVX2_6x8 ps
+    lea             r6, [r3 * 3]
+    mov             r7d, %2 / 16
+.loopH:
+    movu            xm0, [r0]
+    vinserti128     m0, m0, [r0 + r1 * 2], 1
+    movu            xm1, [r0 + r1]
+    vinserti128     m1, m1, [r0 + r4], 1
 
-;-----------------------------------------------------------------------------
-;void interp_4tap_vert_pp_6x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro FILTER_V4_W6_H4 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_pp_6x%2, 4, 6, 8
+    punpcklbw       m2, m0, m1
+    punpckhbw       m3, m0, m1
+    vperm2i128      m4, m2, m3, 0x20
+    vperm2i128      m2, m2, m3, 0x31
+    pmaddubsw       m4, [r5]
+    pmaddubsw       m3, m2, [r5 + mmsize]
+    paddw           m4, m3
+    pmaddubsw       m2, [r5]
 
-    mov         r4d,       r4m
-    sub         r0,        r1
+    vextracti128    xm0, m0, 1
+    lea             r0, [r0 + r1 * 4]
+    vinserti128     m0, m0, [r0], 1
 
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeff]
-    movd        m5,        [r5 + r4 * 4]
+    punpcklbw       m5, m1, m0
+    punpckhbw       m3, m1, m0
+    vperm2i128      m6, m5, m3, 0x20
+    vperm2i128      m5, m5, m3, 0x31
+    pmaddubsw       m6, [r5]
+    pmaddubsw       m3, m5, [r5 + mmsize]
+    paddw           m6, m3
+    pmaddubsw       m5, [r5]
+%ifidn %1,pp
+    pmulhrsw        m4, m7                         ; m4 = word: row 0
+    pmulhrsw        m6, m7                         ; m6 = word: row 1
+    packuswb        m4, m6
+    vpermq          m4, m4, 11011000b
+    vextracti128    xm6, m4, 1
+    movu            [r2], xm4
+    movu            [r2 + r3], xm6
 %else
-    movd        m5,        [tab_ChromaCoeff + r4 * 4]
+    psubw           m4, m7                         ; m4 = word: row 0
+    psubw           m6, m7                         ; m6 = word: row 1
+    movu            [r2], m4
+    movu            [r2 + r3], m6
 %endif
 
-    pshufb      m6,        m5,       [tab_Vm]
-    pshufb      m5,        [tab_Vm + 16]
-    mova        m4,        [pw_512]
-
-    mov         r4d,       %2
-    lea         r5,        [3 * r1]
-
-.loop:
-    movq        m0,        [r0]
-    movq        m1,        [r0 + r1]
-    movq        m2,        [r0 + 2 * r1]
-    movq        m3,        [r0 + r5]
+    movu            xm4, [r0 + r1 * 2]
+    vinserti128     m4, m4, [r0 + r1], 1
+    vextracti128    xm1, m4, 1
+    vinserti128     m0, m0, xm1, 0
 
-    punpcklbw   m0,        m1
-    punpcklbw   m1,        m2
-    punpcklbw   m2,        m3
+    punpcklbw       m6, m0, m4
+    punpckhbw       m1, m0, m4
+    vperm2i128      m0, m6, m1, 0x20
+    vperm2i128      m6, m6, m1, 0x31
+    pmaddubsw       m1, m0, [r5 + mmsize]
+    paddw           m5, m1
+    pmaddubsw       m0, [r5]
+    pmaddubsw       m1, m6, [r5 + mmsize]
+    paddw           m2, m1
+    pmaddubsw       m6, [r5]
 
-    pmaddubsw   m0,        m6
-    pmaddubsw   m7,        m2, m5
+%ifidn %1,pp
+    pmulhrsw        m2, m7                         ; m2 = word: row 2
+    pmulhrsw        m5, m7                         ; m5 = word: row 3
+    packuswb        m2, m5
+    vpermq          m2, m2, 11011000b
+    vextracti128    xm5, m2, 1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm5
+%else
+    psubw           m2, m7                         ; m2 = word: row 2
+    psubw           m5, m7                         ; m5 = word: row 3
+    movu            [r2 + r3 * 2], m2
+    movu            [r2 + r6], m5
+%endif
+    lea             r2, [r2 + r3 * 4]
 
-    paddw       m0,        m7
+    movu            xm1, [r0 + r4]
+    lea             r0, [r0 + r1 * 4]
+    vinserti128     m1, m1, [r0], 1
+    vinserti128     m4, m4, xm1, 1
 
-    pmulhrsw    m0,        m4
-    packuswb    m0,        m0
-    movd        [r2],      m0
-    pextrw      [r2 + 4],  m0,    2
+    punpcklbw       m2, m4, m1
+    punpckhbw       m5, m4, m1
+    vperm2i128      m3, m2, m5, 0x20
+    vperm2i128      m2, m2, m5, 0x31
+    pmaddubsw       m5, m3, [r5 + mmsize]
+    paddw           m6, m5
+    pmaddubsw       m3, [r5]
+    pmaddubsw       m5, m2, [r5 + mmsize]
+    paddw           m0, m5
+    pmaddubsw       m2, [r5]
 
-    lea         r0,        [r0 + 4 * r1]
+%ifidn %1,pp
+    pmulhrsw        m6, m7                         ; m6 = word: row 4
+    pmulhrsw        m0, m7                         ; m0 = word: row 5
+    packuswb        m6, m0
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm0, m6, 1
+    movu            [r2], xm6
+    movu            [r2 + r3], xm0
+%else
+    psubw           m6, m7                         ; m6 = word: row 4
+    psubw           m0, m7                         ; m0 = word: row 5
+    movu            [r2], m6
+    movu            [r2 + r3], m0
+%endif
 
-    movq        m0,        [r0]
-    punpcklbw   m3,        m0
+    movu            xm6, [r0 + r1 * 2]
+    vinserti128     m6, m6, [r0 + r1], 1
+    vextracti128    xm0, m6, 1
+    vinserti128     m1, m1, xm0, 0
 
-    pmaddubsw   m1,        m6
-    pmaddubsw   m7,        m3, m5
+    punpcklbw       m4, m1, m6
+    punpckhbw       m5, m1, m6
+    vperm2i128      m0, m4, m5, 0x20
+    vperm2i128      m5, m4, m5, 0x31
+    pmaddubsw       m4, m0, [r5 + mmsize]
+    paddw           m2, m4
+    pmaddubsw       m0, [r5]
+    pmaddubsw       m4, m5, [r5 + mmsize]
+    paddw           m3, m4
+    pmaddubsw       m5, [r5]
 
-    paddw       m1,        m7
+%ifidn %1,pp
+    pmulhrsw        m3, m7                         ; m3 = word: row 6
+    pmulhrsw        m2, m7                         ; m2 = word: row 7
+    packuswb        m3, m2
+    vpermq          m3, m3, 11011000b
+    vextracti128    xm2, m3, 1
+    movu            [r2 + r3 * 2], xm3
+    movu            [r2 + r6], xm2
+%else
+    psubw           m3, m7                         ; m3 = word: row 6
+    psubw           m2, m7                         ; m2 = word: row 7
+    movu            [r2 + r3 * 2], m3
+    movu            [r2 + r6], m2
+%endif
+    lea             r2, [r2 + r3 * 4]
 
-    pmulhrsw    m1,        m4
-    packuswb    m1,        m1
-    movd        [r2 + r3],      m1
-    pextrw      [r2 + r3 + 4],  m1,    2
+    movu            xm3, [r0 + r4]
+    lea             r0, [r0 + r1 * 4]
+    vinserti128     m3, m3, [r0], 1
+    vinserti128     m6, m6, xm3, 1
 
-    movq        m1,        [r0 + r1]
-    punpcklbw   m7,        m0,        m1
+    punpcklbw       m2, m6, m3
+    punpckhbw       m1, m6, m3
+    vperm2i128      m4, m2, m1, 0x20
+    vperm2i128      m2, m2, m1, 0x31
+    pmaddubsw       m1, m4, [r5 + mmsize]
+    paddw           m5, m1
+    pmaddubsw       m4, [r5]
+    pmaddubsw       m1, m2, [r5 + mmsize]
+    paddw           m0, m1
+    pmaddubsw       m2, [r5]
 
-    pmaddubsw   m2,        m6
-    pmaddubsw   m7,        m5
+%ifidn %1,pp
+    pmulhrsw        m5, m7                         ; m5 = word: row 8
+    pmulhrsw        m0, m7                         ; m0 = word: row 9
+    packuswb        m5, m0
+    vpermq          m5, m5, 11011000b
+    vextracti128    xm0, m5, 1
+    movu            [r2], xm5
+    movu            [r2 + r3], xm0
+%else
+    psubw           m5, m7                         ; m5 = word: row 8
+    psubw           m0, m7                         ; m0 = word: row 9
+    movu            [r2], m5
+    movu            [r2 + r3], m0
+%endif
 
-    paddw       m2,        m7
+    movu            xm5, [r0 + r1 * 2]
+    vinserti128     m5, m5, [r0 + r1], 1
+    vextracti128    xm0, m5, 1
+    vinserti128     m3, m3, xm0, 0
 
-    pmulhrsw    m2,        m4
-    packuswb    m2,        m2
-    lea         r2,        [r2 + 2 * r3]
-    movd        [r2],      m2
-    pextrw      [r2 + 4],  m2,    2
-
-    movq        m2,        [r0 + 2 * r1]
-    punpcklbw   m1,        m2
-
-    pmaddubsw   m3,        m6
-    pmaddubsw   m1,        m5
-
-    paddw       m3,        m1
-
-    pmulhrsw    m3,        m4
-    packuswb    m3,        m3
-
-    movd        [r2 + r3],        m3
-    pextrw      [r2 + r3 + 4],    m3,    2
-
-    lea         r2,        [r2 + 2 * r3]
-
-    sub         r4,         4
-    jnz        .loop
-    RET
-%endmacro
-
-    FILTER_V4_W6_H4 6, 8
-
-    FILTER_V4_W6_H4 6, 16
-
-;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_12x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro FILTER_V4_W12_H2 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_pp_12x%2, 4, 6, 8
-
-    mov         r4d,       r4m
-    sub         r0,        r1
+    punpcklbw       m1, m3, m5
+    punpckhbw       m0, m3, m5
+    vperm2i128      m6, m1, m0, 0x20
+    vperm2i128      m0, m1, m0, 0x31
+    pmaddubsw       m1, m6, [r5 + mmsize]
+    paddw           m2, m1
+    pmaddubsw       m6, [r5]
+    pmaddubsw       m1, m0, [r5 + mmsize]
+    paddw           m4, m1
+    pmaddubsw       m0, [r5]
 
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeff]
-    movd        m0,        [r5 + r4 * 4]
+%ifidn %1,pp
+    pmulhrsw        m4, m7                         ; m4 = word: row 10
+    pmulhrsw        m2, m7                         ; m2 = word: row 11
+    packuswb        m4, m2
+    vpermq          m4, m4, 11011000b
+    vextracti128    xm2, m4, 1
+    movu            [r2 + r3 * 2], xm4
+    movu            [r2 + r6], xm2
 %else
-    movd        m0,        [tab_ChromaCoeff + r4 * 4]
+    psubw           m4, m7                         ; m4 = word: row 10
+    psubw           m2, m7                         ; m2 = word: row 11
+    movu            [r2 + r3 * 2], m4
+    movu            [r2 + r6], m2
 %endif
+    lea             r2, [r2 + r3 * 4]
 
-    pshufb      m1,        m0,       [tab_Vm]
-    pshufb      m0,        [tab_Vm + 16]
-
-    mov         r4d,       %2
-
-.loop:
-    movu        m2,        [r0]
-    movu        m3,        [r0 + r1]
-
-    punpcklbw   m4,        m2,        m3
-    punpckhbw   m2,        m3
-
-    pmaddubsw   m4,        m1
-    pmaddubsw   m2,        m1
-
-    lea         r0,        [r0 + 2 * r1]
-    movu        m5,        [r0]
-    movu        m7,        [r0 + r1]
-
-    punpcklbw   m6,        m5,        m7
-    pmaddubsw   m6,        m0
-    paddw       m4,        m6
-
-    punpckhbw   m6,        m5,        m7
-    pmaddubsw   m6,        m0
-    paddw       m2,        m6
-
-    mova        m6,        [pw_512]
-
-    pmulhrsw    m4,        m6
-    pmulhrsw    m2,        m6
-
-    packuswb    m4,        m2
-
-    movh         [r2],     m4
-    pextrd       [r2 + 8], m4,  2
-
-    punpcklbw   m4,        m3,        m5
-    punpckhbw   m3,        m5
-
-    pmaddubsw   m4,        m1
-    pmaddubsw   m3,        m1
-
-    movu        m5,        [r0 + 2 * r1]
-
-    punpcklbw   m2,        m7,        m5
-    punpckhbw   m7,        m5
-
-    pmaddubsw   m2,        m0
-    pmaddubsw   m7,        m0
-
-    paddw       m4,        m2
-    paddw       m3,        m7
+    movu            xm3, [r0 + r4]
+    lea             r0, [r0 + r1 * 4]
+    vinserti128     m3, m3, [r0], 1
+    vinserti128     m5, m5, xm3, 1
 
-    pmulhrsw    m4,        m6
-    pmulhrsw    m3,        m6
+    punpcklbw       m2, m5, m3
+    punpckhbw       m1, m5, m3
+    vperm2i128      m4, m2, m1, 0x20
+    vperm2i128      m2, m2, m1, 0x31
+    pmaddubsw       m1, m4, [r5 + mmsize]
+    paddw           m0, m1
+    pmaddubsw       m4, [r5]
+    pmaddubsw       m1, m2, [r5 + mmsize]
+    paddw           m6, m1
+    pmaddubsw       m2, [r5]
 
-    packuswb    m4,        m3
+%ifidn %1,pp
+    pmulhrsw        m0, m7                         ; m0 = word: row 12
+    pmulhrsw        m6, m7                         ; m6 = word: row 13
+    packuswb        m0, m6
+    vpermq          m0, m0, 11011000b
+    vextracti128    xm6, m0, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm6
+%else
+    psubw           m0, m7                         ; m0 = word: row 12
+    psubw           m6, m7                         ; m6 = word: row 13
+    movu            [r2], m0
+    movu            [r2 + r3], m6
+%endif
 
-    movh        [r2 + r3],      m4
-    pextrd      [r2 + r3 + 8],  m4,  2
+    movu            xm5, [r0 + r1 * 2]
+    vinserti128     m5, m5, [r0 + r1], 1
+    vextracti128    xm0, m5, 1
+    vinserti128     m3, m3, xm0, 0
 
-    lea         r2,        [r2 + 2 * r3]
+    punpcklbw       m1, m3, m5
+    punpckhbw       m0, m3, m5
+    vperm2i128      m6, m1, m0, 0x20
+    vperm2i128      m0, m1, m0, 0x31
+    pmaddubsw       m6, [r5 + mmsize]
+    paddw           m2, m6
+    pmaddubsw       m0, [r5 + mmsize]
+    paddw           m4, m0
 
-    sub         r4,        2
-    jnz        .loop
+%ifidn %1,pp
+    pmulhrsw        m4, m7                         ; m4 = word: row 14
+    pmulhrsw        m2, m7                         ; m2 = word: row 15
+    packuswb        m4, m2
+    vpermq          m4, m4, 11011000b
+    vextracti128    xm2, m4, 1
+    movu            [r2 + r3 * 2], xm4
+    movu            [r2 + r6], xm2
+%else
+    psubw           m4, m7                         ; m4 = word: row 14
+    psubw           m2, m7                         ; m2 = word: row 15
+    movu            [r2 + r3 * 2], m4
+    movu            [r2 + r6], m2
+%endif
+    lea             r2, [r2 + r3 * 4]
+    dec             r7d
+    jnz             .loopH
     RET
+%endif
 %endmacro
 
-    FILTER_V4_W12_H2 12, 16
-
-    FILTER_V4_W12_H2 12, 32
-
-;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_16x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro FILTER_V4_W16_H2 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_pp_16x%2, 4, 6, 8
+    FILTER_VER_CHROMA_AVX2_16xN pp, 32
+    FILTER_VER_CHROMA_AVX2_16xN ps, 32
+    FILTER_VER_CHROMA_AVX2_16xN pp, 64
+    FILTER_VER_CHROMA_AVX2_16xN ps, 64
 
-    mov         r4d,       r4m
-    sub         r0,        r1
+%macro FILTER_VER_CHROMA_AVX2_16x24 1
+%if ARCH_X86_64 == 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_16x24, 4, 6, 15
+    mov             r4d, r4m
+    shl             r4d, 6
 
 %ifdef PIC
-    lea         r5,        [tab_ChromaCoeff]
-    movd        m0,        [r5 + r4 * 4]
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
 %else
-    movd        m0,        [tab_ChromaCoeff + r4 * 4]
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
 %endif
 
-    pshufb      m1,        m0,       [tab_Vm]
-    pshufb      m0,        [tab_Vm + 16]
-
-    mov         r4d,       %2/2
-
-.loop:
-    movu        m2,        [r0]
-    movu        m3,        [r0 + r1]
-
-    punpcklbw   m4,        m2,        m3
-    punpckhbw   m2,        m3
-
-    pmaddubsw   m4,        m1
-    pmaddubsw   m2,        m1
-
-    lea         r0,        [r0 + 2 * r1]
-    movu        m5,        [r0]
-    movu        m6,        [r0 + r1]
-
-    punpckhbw   m7,        m5,        m6
-    pmaddubsw   m7,        m0
-    paddw       m2,        m7
-
-    punpcklbw   m7,        m5,        m6
-    pmaddubsw   m7,        m0
-    paddw       m4,        m7
-
-    mova        m7,        [pw_512]
-
-    pmulhrsw    m4,        m7
-    pmulhrsw    m2,        m7
-
-    packuswb    m4,        m2
-
-    movu        [r2],      m4
-
-    punpcklbw   m4,        m3,        m5
-    punpckhbw   m3,        m5
-
-    pmaddubsw   m4,        m1
-    pmaddubsw   m3,        m1
-
-    movu        m5,        [r0 + 2 * r1]
-
-    punpcklbw   m2,        m6,        m5
-    punpckhbw   m6,        m5
-
-    pmaddubsw   m2,        m0
-    pmaddubsw   m6,        m0
-
-    paddw       m4,        m2
-    paddw       m3,        m6
-
-    pmulhrsw    m4,        m7
-    pmulhrsw    m3,        m7
-
-    packuswb    m4,        m3
-
-    movu        [r2 + r3],      m4
-
-    lea         r2,        [r2 + 2 * r3]
-
-    dec         r4d
-    jnz        .loop
-    RET
-%endmacro
-
-    FILTER_V4_W16_H2 16,  4
-    FILTER_V4_W16_H2 16,  8
-    FILTER_V4_W16_H2 16, 12
-    FILTER_V4_W16_H2 16, 16
-    FILTER_V4_W16_H2 16, 32
-
-    FILTER_V4_W16_H2 16, 24
-    FILTER_V4_W16_H2 16, 64
-
-%macro FILTER_VER_CHROMA_AVX2_16x16 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_vert_%1_16x16, 4, 6, 15
-    mov             r4d, r4m
-    shl             r4d, 6
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
-
-    mova            m12, [r5]
-    mova            m13, [r5 + mmsize]
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,pp
-    mova            m14, [pw_512]
-%else
-    add             r3d, r3d
-    vbroadcasti128  m14, [pw_2000]
-%endif
-    lea             r5, [r3 * 3]
+    mova            m12, [r5]
+    mova            m13, [r5 + mmsize]
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,pp
+    mova            m14, [pw_512]
+%else
+    add             r3d, r3d
+    vbroadcasti128  m14, [pw_2000]
+%endif
+    lea             r5, [r3 * 3]
 
     movu            xm0, [r0]                       ; m0 = row 0
     movu            xm1, [r0 + r1]                  ; m1 = row 1
@@ -9060,10 +7215,10 @@ cglobal interp_4tap_vert_%1_16x16, 4, 6, 15
     packuswb        m2, m3
     packuswb        m4, m5
     packuswb        m6, m7
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    vpermq          m4, m4, 11011000b
-    vpermq          m6, m6, 11011000b
+    vpermq          m0, m0, q3120
+    vpermq          m2, m2, q3120
+    vpermq          m4, m4, q3120
+    vpermq          m6, m6, q3120
     vextracti128    xm1, m0, 1
     vextracti128    xm3, m2, 1
     vextracti128    xm5, m4, 1
@@ -9147,14 +7302,16 @@ cglobal interp_4tap_vert_%1_16x16, 4, 6, 15
     punpckhbw       xm4, xm2, xm3
     punpcklbw       xm2, xm3
     vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m2, m13
-    paddw           m0, m2
+    pmaddubsw       m4, m2, m13
+    paddw           m0, m4
+    pmaddubsw       m2, m12
     movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
     punpckhbw       xm5, xm3, xm4
     punpcklbw       xm3, xm4
     vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m3, m13
-    paddw           m1, m3
+    pmaddubsw       m5, m3, m13
+    paddw           m1, m5
+    pmaddubsw       m3, m12
 
 %ifidn %1,pp
     pmulhrsw        m8, m14                         ; m8 = word: row 8
@@ -9169,10 +7326,10 @@ cglobal interp_4tap_vert_%1_16x16, 4, 6, 15
     packuswb        m10, m11
     packuswb        m6, m7
     packuswb        m0, m1
-    vpermq          m8, m8, 11011000b
-    vpermq          m10, m10, 11011000b
-    vpermq          m6, m6, 11011000b
-    vpermq          m0, m0, 11011000b
+    vpermq          m8, m8, q3120
+    vpermq          m10, m10, q3120
+    vpermq          m6, m6, q3120
+    vpermq          m0, m0, q3120
     vextracti128    xm9, m8, 1
     vextracti128    xm11, m10, 1
     vextracti128    xm7, m6, 1
@@ -9204,16 +7361,126 @@ cglobal interp_4tap_vert_%1_16x16, 4, 6, 15
     movu            [r2 + r3], m7
     movu            [r2 + r3 * 2], m0
     movu            [r2 + r5], m1
+%endif
+    lea             r2, [r2 + r3 * 4]
+
+    movu            xm5, [r0 + r4]                  ; m5 = row 19
+    punpckhbw       xm6, xm4, xm5
+    punpcklbw       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddubsw       m6, m4, m13
+    paddw           m2, m6
+    pmaddubsw       m4, m12
+    lea             r0, [r0 + r1 * 4]
+    movu            xm6, [r0]                       ; m6 = row 20
+    punpckhbw       xm7, xm5, xm6
+    punpcklbw       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddubsw       m7, m5, m13
+    paddw           m3, m7
+    pmaddubsw       m5, m12
+    movu            xm7, [r0 + r1]                  ; m7 = row 21
+    punpckhbw       xm0, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm0, 1
+    pmaddubsw       m0, m6, m13
+    paddw           m4, m0
+    pmaddubsw       m6, m12
+    movu            xm0, [r0 + r1 * 2]              ; m0 = row 22
+    punpckhbw       xm1, xm7, xm0
+    punpcklbw       xm7, xm0
+    vinserti128     m7, m7, xm1, 1
+    pmaddubsw       m1, m7, m13
+    paddw           m5, m1
+    pmaddubsw       m7, m12
+    movu            xm1, [r0 + r4]                  ; m1 = row 23
+    punpckhbw       xm8, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm8, 1
+    pmaddubsw       m8, m0, m13
+    paddw           m6, m8
+    pmaddubsw       m0, m12
+    lea             r0, [r0 + r1 * 4]
+    movu            xm8, [r0]                       ; m8 = row 24
+    punpckhbw       xm9, xm1, xm8
+    punpcklbw       xm1, xm8
+    vinserti128     m1, m1, xm9, 1
+    pmaddubsw       m9, m1, m13
+    paddw           m7, m9
+    pmaddubsw       m1, m12
+    movu            xm9, [r0 + r1]                  ; m9 = row 25
+    punpckhbw       xm10, xm8, xm9
+    punpcklbw       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddubsw       m8, m13
+    paddw           m0, m8
+    movu            xm10, [r0 + r1 * 2]             ; m10 = row 26
+    punpckhbw       xm11, xm9, xm10
+    punpcklbw       xm9, xm10
+    vinserti128     m9, m9, xm11, 1
+    pmaddubsw       m9, m13
+    paddw           m1, m9
+
+%ifidn %1,pp
+    pmulhrsw        m2, m14                         ; m2 = word: row 16
+    pmulhrsw        m3, m14                         ; m3 = word: row 17
+    pmulhrsw        m4, m14                         ; m4 = word: row 18
+    pmulhrsw        m5, m14                         ; m5 = word: row 19
+    pmulhrsw        m6, m14                         ; m6 = word: row 20
+    pmulhrsw        m7, m14                         ; m7 = word: row 21
+    pmulhrsw        m0, m14                         ; m0 = word: row 22
+    pmulhrsw        m1, m14                         ; m1 = word: row 23
+    packuswb        m2, m3
+    packuswb        m4, m5
+    packuswb        m6, m7
+    packuswb        m0, m1
+    vpermq          m2, m2, q3120
+    vpermq          m4, m4, q3120
+    vpermq          m6, m6, q3120
+    vpermq          m0, m0, q3120
+    vextracti128    xm3, m2, 1
+    vextracti128    xm5, m4, 1
+    vextracti128    xm7, m6, 1
+    vextracti128    xm1, m0, 1
+    movu            [r2], xm2
+    movu            [r2 + r3], xm3
+    movu            [r2 + r3 * 2], xm4
+    movu            [r2 + r5], xm5
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm6
+    movu            [r2 + r3], xm7
+    movu            [r2 + r3 * 2], xm0
+    movu            [r2 + r5], xm1
+%else
+    psubw           m2, m14                         ; m2 = word: row 16
+    psubw           m3, m14                         ; m3 = word: row 17
+    psubw           m4, m14                         ; m4 = word: row 18
+    psubw           m5, m14                         ; m5 = word: row 19
+    psubw           m6, m14                         ; m6 = word: row 20
+    psubw           m7, m14                         ; m7 = word: row 21
+    psubw           m0, m14                         ; m0 = word: row 22
+    psubw           m1, m14                         ; m1 = word: row 23
+    movu            [r2], m2
+    movu            [r2 + r3], m3
+    movu            [r2 + r3 * 2], m4
+    movu            [r2 + r5], m5
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], m6
+    movu            [r2 + r3], m7
+    movu            [r2 + r3 * 2], m0
+    movu            [r2 + r5], m1
 %endif
     RET
 %endif
 %endmacro
 
-    FILTER_VER_CHROMA_AVX2_16x16 pp
-    FILTER_VER_CHROMA_AVX2_16x16 ps
-%macro FILTER_VER_CHROMA_AVX2_16x8 1
+    FILTER_VER_CHROMA_AVX2_16x24 pp
+    FILTER_VER_CHROMA_AVX2_16x24 ps
+
+%macro FILTER_VER_CHROMA_AVX2_24x32 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_16x8, 4, 7, 7
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_vert_%1_24x32, 4, 9, 10
     mov             r4d, r4m
     shl             r4d, 6
 
@@ -9224,179 +7491,23 @@ cglobal interp_4tap_vert_%1_16x8, 4, 7, 7
     lea             r5, [tab_ChromaCoeffVer_32 + r4]
 %endif
 
+    mova            m8, [r5]
+    mova            m9, [r5 + mmsize]
     lea             r4, [r1 * 3]
     sub             r0, r1
 %ifidn %1,pp
-    mova            m6, [pw_512]
+    mova            m7, [pw_512]
 %else
     add             r3d, r3d
-    mova            m6, [pw_2000]
+    vbroadcasti128  m7, [pw_2000]
 %endif
     lea             r6, [r3 * 3]
-
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, [r5 + mmsize]
-    paddw           m0, m4
-    pmaddubsw       m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhbw       xm5, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, [r5 + mmsize]
-    paddw           m1, m5
-    pmaddubsw       m3, [r5]
-%ifidn %1,pp
-    pmulhrsw        m0, m6                          ; m0 = word: row 0
-    pmulhrsw        m1, m6                          ; m1 = word: row 1
-    packuswb        m0, m1
-    vpermq          m0, m0, 11011000b
-    vextracti128    xm1, m0, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-%else
-    psubw           m0, m6                          ; m0 = word: row 0
-    psubw           m1, m6                          ; m1 = word: row 1
-    movu            [r2], m0
-    movu            [r2 + r3], m1
-%endif
-
-    movu            xm0, [r0 + r1]                  ; m0 = row 5
-    punpckhbw       xm1, xm4, xm0
-    punpcklbw       xm4, xm0
-    vinserti128     m4, m4, xm1, 1
-    pmaddubsw       m1, m4, [r5 + mmsize]
-    paddw           m2, m1
-    pmaddubsw       m4, [r5]
-    movu            xm1, [r0 + r1 * 2]              ; m1 = row 6
-    punpckhbw       xm5, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm5, 1
-    pmaddubsw       m5, m0, [r5 + mmsize]
-    paddw           m3, m5
-    pmaddubsw       m0, [r5]
-%ifidn %1,pp
-    pmulhrsw        m2, m6                          ; m2 = word: row 2
-    pmulhrsw        m3, m6                          ; m3 = word: row 3
-    packuswb        m2, m3
-    vpermq          m2, m2, 11011000b
-    vextracti128    xm3, m2, 1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-%else
-    psubw           m2, m6                          ; m2 = word: row 2
-    psubw           m3, m6                          ; m3 = word: row 3
-    movu            [r2 + r3 * 2], m2
-    movu            [r2 + r6], m3
-%endif
-
-    movu            xm2, [r0 + r4]                  ; m2 = row 7
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m3, m1, [r5 + mmsize]
-    paddw           m4, m3
-    pmaddubsw       m1, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm3, [r0]                       ; m3 = row 8
-    punpckhbw       xm5, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm5, 1
-    pmaddubsw       m5, m2, [r5 + mmsize]
-    paddw           m0, m5
-    pmaddubsw       m2, [r5]
-    lea             r2, [r2 + r3 * 4]
-%ifidn %1,pp
-    pmulhrsw        m4, m6                          ; m4 = word: row 4
-    pmulhrsw        m0, m6                          ; m0 = word: row 5
-    packuswb        m4, m0
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm0, m4, 1
-    movu            [r2], xm4
-    movu            [r2 + r3], xm0
-%else
-    psubw           m4, m6                          ; m4 = word: row 4
-    psubw           m0, m6                          ; m0 = word: row 5
-    movu            [r2], m4
-    movu            [r2 + r3], m0
-%endif
-
-    movu            xm5, [r0 + r1]                  ; m5 = row 9
-    punpckhbw       xm4, xm3, xm5
-    punpcklbw       xm3, xm5
-    vinserti128     m3, m3, xm4, 1
-    pmaddubsw       m3, [r5 + mmsize]
-    paddw           m1, m3
-    movu            xm4, [r0 + r1 * 2]              ; m4 = row 10
-    punpckhbw       xm0, xm5, xm4
-    punpcklbw       xm5, xm4
-    vinserti128     m5, m5, xm0, 1
-    pmaddubsw       m5, [r5 + mmsize]
-    paddw           m2, m5
-%ifidn %1,pp
-    pmulhrsw        m1, m6                          ; m1 = word: row 6
-    pmulhrsw        m2, m6                          ; m2 = word: row 7
-    packuswb        m1, m2
-    vpermq          m1, m1, 11011000b
-    vextracti128    xm2, m1, 1
-    movu            [r2 + r3 * 2], xm1
-    movu            [r2 + r6], xm2
-%else
-    psubw           m1, m6                          ; m1 = word: row 6
-    psubw           m2, m6                          ; m2 = word: row 7
-    movu            [r2 + r3 * 2], m1
-    movu            [r2 + r6], m2
-%endif
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_AVX2_16x8 pp
-    FILTER_VER_CHROMA_AVX2_16x8 ps
-
-%macro FILTER_VER_CHROMA_AVX2_16x12 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_vert_%1_16x12, 4, 6, 10
-    mov             r4d, r4m
-    shl             r4d, 6
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
-
-    mova            m8, [r5]
-    mova            m9, [r5 + mmsize]
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,pp
-    mova            m7, [pw_512]
-%else
-    add             r3d, r3d
-    vbroadcasti128  m7, [pw_2000]
-%endif
-    lea             r5, [r3 * 3]
-
-    movu            xm0, [r0]
-    vinserti128     m0, m0, [r0 + r1 * 2], 1
-    movu            xm1, [r0 + r1]
-    vinserti128     m1, m1, [r0 + r4], 1
+    mov             r5d, 2
+.loopH:
+    movu            xm0, [r0]
+    vinserti128     m0, m0, [r0 + r1 * 2], 1
+    movu            xm1, [r0 + r1]
+    vinserti128     m1, m1, [r0 + r4], 1
 
     punpcklbw       m2, m0, m1
     punpckhbw       m3, m0, m1
@@ -9408,8 +7519,8 @@ cglobal interp_4tap_vert_%1_16x12, 4, 6, 10
     pmaddubsw       m2, m8
 
     vextracti128    xm0, m0, 1
-    lea             r0, [r0 + r1 * 4]
-    vinserti128     m0, m0, [r0], 1
+    lea             r7, [r0 + r1 * 4]
+    vinserti128     m0, m0, [r7], 1
 
     punpcklbw       m5, m1, m0
     punpckhbw       m3, m1, m0
@@ -9434,8 +7545,8 @@ cglobal interp_4tap_vert_%1_16x12, 4, 6, 10
     movu            [r2 + r3], m6
 %endif
 
-    movu            xm4, [r0 + r1 * 2]
-    vinserti128     m4, m4, [r0 + r1], 1
+    movu            xm4, [r7 + r1 * 2]
+    vinserti128     m4, m4, [r7 + r1], 1
     vextracti128    xm1, m4, 1
     vinserti128     m0, m0, xm1, 0
 
@@ -9457,18 +7568,18 @@ cglobal interp_4tap_vert_%1_16x12, 4, 6, 10
     vpermq          m2, m2, 11011000b
     vextracti128    xm5, m2, 1
     movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r5], xm5
+    movu            [r2 + r6], xm5
 %else
     psubw           m2, m7                         ; m2 = word: row 2
     psubw           m5, m7                         ; m5 = word: row 3
     movu            [r2 + r3 * 2], m2
-    movu            [r2 + r5], m5
+    movu            [r2 + r6], m5
 %endif
-    lea             r2, [r2 + r3 * 4]
+    lea             r8, [r2 + r3 * 4]
 
-    movu            xm1, [r0 + r4]
-    lea             r0, [r0 + r1 * 4]
-    vinserti128     m1, m1, [r0], 1
+    movu            xm1, [r7 + r4]
+    lea             r7, [r7 + r1 * 4]
+    vinserti128     m1, m1, [r7], 1
     vinserti128     m4, m4, xm1, 1
 
     punpcklbw       m2, m4, m1
@@ -9488,17 +7599,17 @@ cglobal interp_4tap_vert_%1_16x12, 4, 6, 10
     packuswb        m6, m0
     vpermq          m6, m6, 11011000b
     vextracti128    xm0, m6, 1
-    movu            [r2], xm6
-    movu            [r2 + r3], xm0
+    movu            [r8], xm6
+    movu            [r8 + r3], xm0
 %else
     psubw           m6, m7                         ; m6 = word: row 4
     psubw           m0, m7                         ; m0 = word: row 5
-    movu            [r2], m6
-    movu            [r2 + r3], m0
+    movu            [r8], m6
+    movu            [r8 + r3], m0
 %endif
 
-    movu            xm6, [r0 + r1 * 2]
-    vinserti128     m6, m6, [r0 + r1], 1
+    movu            xm6, [r7 + r1 * 2]
+    vinserti128     m6, m6, [r7 + r1], 1
     vextracti128    xm0, m6, 1
     vinserti128     m1, m1, xm0, 0
 
@@ -9519,19 +7630,19 @@ cglobal interp_4tap_vert_%1_16x12, 4, 6, 10
     packuswb        m3, m2
     vpermq          m3, m3, 11011000b
     vextracti128    xm2, m3, 1
-    movu            [r2 + r3 * 2], xm3
-    movu            [r2 + r5], xm2
+    movu            [r8 + r3 * 2], xm3
+    movu            [r8 + r6], xm2
 %else
     psubw           m3, m7                         ; m3 = word: row 6
     psubw           m2, m7                         ; m2 = word: row 7
-    movu            [r2 + r3 * 2], m3
-    movu            [r2 + r5], m2
+    movu            [r8 + r3 * 2], m3
+    movu            [r8 + r6], m2
 %endif
-    lea             r2, [r2 + r3 * 4]
+    lea             r8, [r8 + r3 * 4]
 
-    movu            xm3, [r0 + r4]
-    lea             r0, [r0 + r1 * 4]
-    vinserti128     m3, m3, [r0], 1
+    movu            xm3, [r7 + r4]
+    lea             r7, [r7 + r1 * 4]
+    vinserti128     m3, m3, [r7], 1
     vinserti128     m6, m6, xm3, 1
 
     punpcklbw       m2, m6, m3
@@ -9551,17 +7662,17 @@ cglobal interp_4tap_vert_%1_16x12, 4, 6, 10
     packuswb        m5, m0
     vpermq          m5, m5, 11011000b
     vextracti128    xm0, m5, 1
-    movu            [r2], xm5
-    movu            [r2 + r3], xm0
+    movu            [r8], xm5
+    movu            [r8 + r3], xm0
 %else
     psubw           m5, m7                         ; m5 = word: row 8
     psubw           m0, m7                         ; m0 = word: row 9
-    movu            [r2], m5
-    movu            [r2 + r3], m0
+    movu            [r8], m5
+    movu            [r8 + r3], m0
 %endif
 
-    movu            xm5, [r0 + r1 * 2]
-    vinserti128     m5, m5, [r0 + r1], 1
+    movu            xm5, [r7 + r1 * 2]
+    vinserti128     m5, m5, [r7 + r1], 1
     vextracti128    xm0, m5, 1
     vinserti128     m3, m3, xm0, 0
 
@@ -9571,8 +7682,10 @@ cglobal interp_4tap_vert_%1_16x12, 4, 6, 10
     vperm2i128      m0, m1, m0, 0x31
     pmaddubsw       m1, m6, m9
     paddw           m2, m1
+    pmaddubsw       m6, m8
     pmaddubsw       m1, m0, m9
     paddw           m4, m1
+    pmaddubsw       m0, m8
 
 %ifidn %1,pp
     pmulhrsw        m4, m7                         ; m4 = word: row 10
@@ -9580,278 +7693,49 @@ cglobal interp_4tap_vert_%1_16x12, 4, 6, 10
     packuswb        m4, m2
     vpermq          m4, m4, 11011000b
     vextracti128    xm2, m4, 1
-    movu            [r2 + r3 * 2], xm4
-    movu            [r2 + r5], xm2
+    movu            [r8 + r3 * 2], xm4
+    movu            [r8 + r6], xm2
 %else
     psubw           m4, m7                         ; m4 = word: row 10
     psubw           m2, m7                         ; m2 = word: row 11
-    movu            [r2 + r3 * 2], m4
-    movu            [r2 + r5], m2
-%endif
-    RET
+    movu            [r8 + r3 * 2], m4
+    movu            [r8 + r6], m2
 %endif
-%endmacro
-
-    FILTER_VER_CHROMA_AVX2_16x12 pp
-    FILTER_VER_CHROMA_AVX2_16x12 ps
+    lea             r8, [r8 + r3 * 4]
 
-%macro FILTER_VER_CHROMA_AVX2_16xN 2
-%if ARCH_X86_64 == 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_16x%2, 4, 8, 8
-    mov             r4d, r4m
-    shl             r4d, 6
+    movu            xm3, [r7 + r4]
+    lea             r7, [r7 + r1 * 4]
+    vinserti128     m3, m3, [r7], 1
+    vinserti128     m5, m5, xm3, 1
 
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
+    punpcklbw       m2, m5, m3
+    punpckhbw       m1, m5, m3
+    vperm2i128      m4, m2, m1, 0x20
+    vperm2i128      m2, m2, m1, 0x31
+    pmaddubsw       m1, m4, m9
+    paddw           m0, m1
+    pmaddubsw       m4, m8
+    pmaddubsw       m1, m2, m9
+    paddw           m6, m1
+    pmaddubsw       m2, m8
 
-    lea             r4, [r1 * 3]
-    sub             r0, r1
 %ifidn %1,pp
-    mova            m7, [pw_512]
+    pmulhrsw        m0, m7                         ; m0 = word: row 12
+    pmulhrsw        m6, m7                         ; m6 = word: row 13
+    packuswb        m0, m6
+    vpermq          m0, m0, 11011000b
+    vextracti128    xm6, m0, 1
+    movu            [r8], xm0
+    movu            [r8 + r3], xm6
 %else
-    add             r3d, r3d
-    mova            m7, [pw_2000]
-%endif
-    lea             r6, [r3 * 3]
-    mov             r7d, %2 / 16
-.loopH:
-    movu            xm0, [r0]
-    vinserti128     m0, m0, [r0 + r1 * 2], 1
-    movu            xm1, [r0 + r1]
-    vinserti128     m1, m1, [r0 + r4], 1
-
-    punpcklbw       m2, m0, m1
-    punpckhbw       m3, m0, m1
-    vperm2i128      m4, m2, m3, 0x20
-    vperm2i128      m2, m2, m3, 0x31
-    pmaddubsw       m4, [r5]
-    pmaddubsw       m3, m2, [r5 + mmsize]
-    paddw           m4, m3
-    pmaddubsw       m2, [r5]
-
-    vextracti128    xm0, m0, 1
-    lea             r0, [r0 + r1 * 4]
-    vinserti128     m0, m0, [r0], 1
-
-    punpcklbw       m5, m1, m0
-    punpckhbw       m3, m1, m0
-    vperm2i128      m6, m5, m3, 0x20
-    vperm2i128      m5, m5, m3, 0x31
-    pmaddubsw       m6, [r5]
-    pmaddubsw       m3, m5, [r5 + mmsize]
-    paddw           m6, m3
-    pmaddubsw       m5, [r5]
-%ifidn %1,pp
-    pmulhrsw        m4, m7                         ; m4 = word: row 0
-    pmulhrsw        m6, m7                         ; m6 = word: row 1
-    packuswb        m4, m6
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm6, m4, 1
-    movu            [r2], xm4
-    movu            [r2 + r3], xm6
-%else
-    psubw           m4, m7                         ; m4 = word: row 0
-    psubw           m6, m7                         ; m6 = word: row 1
-    movu            [r2], m4
-    movu            [r2 + r3], m6
-%endif
-
-    movu            xm4, [r0 + r1 * 2]
-    vinserti128     m4, m4, [r0 + r1], 1
-    vextracti128    xm1, m4, 1
-    vinserti128     m0, m0, xm1, 0
-
-    punpcklbw       m6, m0, m4
-    punpckhbw       m1, m0, m4
-    vperm2i128      m0, m6, m1, 0x20
-    vperm2i128      m6, m6, m1, 0x31
-    pmaddubsw       m1, m0, [r5 + mmsize]
-    paddw           m5, m1
-    pmaddubsw       m0, [r5]
-    pmaddubsw       m1, m6, [r5 + mmsize]
-    paddw           m2, m1
-    pmaddubsw       m6, [r5]
-
-%ifidn %1,pp
-    pmulhrsw        m2, m7                         ; m2 = word: row 2
-    pmulhrsw        m5, m7                         ; m5 = word: row 3
-    packuswb        m2, m5
-    vpermq          m2, m2, 11011000b
-    vextracti128    xm5, m2, 1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm5
-%else
-    psubw           m2, m7                         ; m2 = word: row 2
-    psubw           m5, m7                         ; m5 = word: row 3
-    movu            [r2 + r3 * 2], m2
-    movu            [r2 + r6], m5
-%endif
-    lea             r2, [r2 + r3 * 4]
-
-    movu            xm1, [r0 + r4]
-    lea             r0, [r0 + r1 * 4]
-    vinserti128     m1, m1, [r0], 1
-    vinserti128     m4, m4, xm1, 1
-
-    punpcklbw       m2, m4, m1
-    punpckhbw       m5, m4, m1
-    vperm2i128      m3, m2, m5, 0x20
-    vperm2i128      m2, m2, m5, 0x31
-    pmaddubsw       m5, m3, [r5 + mmsize]
-    paddw           m6, m5
-    pmaddubsw       m3, [r5]
-    pmaddubsw       m5, m2, [r5 + mmsize]
-    paddw           m0, m5
-    pmaddubsw       m2, [r5]
-
-%ifidn %1,pp
-    pmulhrsw        m6, m7                         ; m6 = word: row 4
-    pmulhrsw        m0, m7                         ; m0 = word: row 5
-    packuswb        m6, m0
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm0, m6, 1
-    movu            [r2], xm6
-    movu            [r2 + r3], xm0
-%else
-    psubw           m6, m7                         ; m6 = word: row 4
-    psubw           m0, m7                         ; m0 = word: row 5
-    movu            [r2], m6
-    movu            [r2 + r3], m0
-%endif
-
-    movu            xm6, [r0 + r1 * 2]
-    vinserti128     m6, m6, [r0 + r1], 1
-    vextracti128    xm0, m6, 1
-    vinserti128     m1, m1, xm0, 0
-
-    punpcklbw       m4, m1, m6
-    punpckhbw       m5, m1, m6
-    vperm2i128      m0, m4, m5, 0x20
-    vperm2i128      m5, m4, m5, 0x31
-    pmaddubsw       m4, m0, [r5 + mmsize]
-    paddw           m2, m4
-    pmaddubsw       m0, [r5]
-    pmaddubsw       m4, m5, [r5 + mmsize]
-    paddw           m3, m4
-    pmaddubsw       m5, [r5]
-
-%ifidn %1,pp
-    pmulhrsw        m3, m7                         ; m3 = word: row 6
-    pmulhrsw        m2, m7                         ; m2 = word: row 7
-    packuswb        m3, m2
-    vpermq          m3, m3, 11011000b
-    vextracti128    xm2, m3, 1
-    movu            [r2 + r3 * 2], xm3
-    movu            [r2 + r6], xm2
-%else
-    psubw           m3, m7                         ; m3 = word: row 6
-    psubw           m2, m7                         ; m2 = word: row 7
-    movu            [r2 + r3 * 2], m3
-    movu            [r2 + r6], m2
-%endif
-    lea             r2, [r2 + r3 * 4]
-
-    movu            xm3, [r0 + r4]
-    lea             r0, [r0 + r1 * 4]
-    vinserti128     m3, m3, [r0], 1
-    vinserti128     m6, m6, xm3, 1
-
-    punpcklbw       m2, m6, m3
-    punpckhbw       m1, m6, m3
-    vperm2i128      m4, m2, m1, 0x20
-    vperm2i128      m2, m2, m1, 0x31
-    pmaddubsw       m1, m4, [r5 + mmsize]
-    paddw           m5, m1
-    pmaddubsw       m4, [r5]
-    pmaddubsw       m1, m2, [r5 + mmsize]
-    paddw           m0, m1
-    pmaddubsw       m2, [r5]
-
-%ifidn %1,pp
-    pmulhrsw        m5, m7                         ; m5 = word: row 8
-    pmulhrsw        m0, m7                         ; m0 = word: row 9
-    packuswb        m5, m0
-    vpermq          m5, m5, 11011000b
-    vextracti128    xm0, m5, 1
-    movu            [r2], xm5
-    movu            [r2 + r3], xm0
-%else
-    psubw           m5, m7                         ; m5 = word: row 8
-    psubw           m0, m7                         ; m0 = word: row 9
-    movu            [r2], m5
-    movu            [r2 + r3], m0
-%endif
-
-    movu            xm5, [r0 + r1 * 2]
-    vinserti128     m5, m5, [r0 + r1], 1
-    vextracti128    xm0, m5, 1
-    vinserti128     m3, m3, xm0, 0
-
-    punpcklbw       m1, m3, m5
-    punpckhbw       m0, m3, m5
-    vperm2i128      m6, m1, m0, 0x20
-    vperm2i128      m0, m1, m0, 0x31
-    pmaddubsw       m1, m6, [r5 + mmsize]
-    paddw           m2, m1
-    pmaddubsw       m6, [r5]
-    pmaddubsw       m1, m0, [r5 + mmsize]
-    paddw           m4, m1
-    pmaddubsw       m0, [r5]
-
-%ifidn %1,pp
-    pmulhrsw        m4, m7                         ; m4 = word: row 10
-    pmulhrsw        m2, m7                         ; m2 = word: row 11
-    packuswb        m4, m2
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm2, m4, 1
-    movu            [r2 + r3 * 2], xm4
-    movu            [r2 + r6], xm2
-%else
-    psubw           m4, m7                         ; m4 = word: row 10
-    psubw           m2, m7                         ; m2 = word: row 11
-    movu            [r2 + r3 * 2], m4
-    movu            [r2 + r6], m2
-%endif
-    lea             r2, [r2 + r3 * 4]
-
-    movu            xm3, [r0 + r4]
-    lea             r0, [r0 + r1 * 4]
-    vinserti128     m3, m3, [r0], 1
-    vinserti128     m5, m5, xm3, 1
-
-    punpcklbw       m2, m5, m3
-    punpckhbw       m1, m5, m3
-    vperm2i128      m4, m2, m1, 0x20
-    vperm2i128      m2, m2, m1, 0x31
-    pmaddubsw       m1, m4, [r5 + mmsize]
-    paddw           m0, m1
-    pmaddubsw       m4, [r5]
-    pmaddubsw       m1, m2, [r5 + mmsize]
-    paddw           m6, m1
-    pmaddubsw       m2, [r5]
-
-%ifidn %1,pp
-    pmulhrsw        m0, m7                         ; m0 = word: row 12
-    pmulhrsw        m6, m7                         ; m6 = word: row 13
-    packuswb        m0, m6
-    vpermq          m0, m0, 11011000b
-    vextracti128    xm6, m0, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm6
-%else
-    psubw           m0, m7                         ; m0 = word: row 12
-    psubw           m6, m7                         ; m6 = word: row 13
-    movu            [r2], m0
-    movu            [r2 + r3], m6
+    psubw           m0, m7                         ; m0 = word: row 12
+    psubw           m6, m7                         ; m6 = word: row 13
+    movu            [r8], m0
+    movu            [r8 + r3], m6
 %endif
 
-    movu            xm5, [r0 + r1 * 2]
-    vinserti128     m5, m5, [r0 + r1], 1
+    movu            xm5, [r7 + r1 * 2]
+    vinserti128     m5, m5, [r7 + r1], 1
     vextracti128    xm0, m5, 1
     vinserti128     m3, m3, xm0, 0
 
@@ -9859,9 +7743,9 @@ cglobal interp_4tap_vert_%1_16x%2, 4, 8, 8
     punpckhbw       m0, m3, m5
     vperm2i128      m6, m1, m0, 0x20
     vperm2i128      m0, m1, m0, 0x31
-    pmaddubsw       m6, [r5 + mmsize]
+    pmaddubsw       m6, m9
     paddw           m2, m6
-    pmaddubsw       m0, [r5 + mmsize]
+    pmaddubsw       m0, m9
     paddw           m4, m0
 
 %ifidn %1,pp
@@ -9870,30 +7754,184 @@ cglobal interp_4tap_vert_%1_16x%2, 4, 8, 8
     packuswb        m4, m2
     vpermq          m4, m4, 11011000b
     vextracti128    xm2, m4, 1
-    movu            [r2 + r3 * 2], xm4
-    movu            [r2 + r6], xm2
+    movu            [r8 + r3 * 2], xm4
+    movu            [r8 + r6], xm2
+    add             r2, 16
 %else
     psubw           m4, m7                         ; m4 = word: row 14
     psubw           m2, m7                         ; m2 = word: row 15
-    movu            [r2 + r3 * 2], m4
-    movu            [r2 + r6], m2
-%endif
-    lea             r2, [r2 + r3 * 4]
-    dec             r7d
-    jnz             .loopH
-    RET
+    movu            [r8 + r3 * 2], m4
+    movu            [r8 + r6], m2
+    add             r2, 32
 %endif
-%endmacro
-
-    FILTER_VER_CHROMA_AVX2_16xN pp, 32
-    FILTER_VER_CHROMA_AVX2_16xN ps, 32
-    FILTER_VER_CHROMA_AVX2_16xN pp, 64
-    FILTER_VER_CHROMA_AVX2_16xN ps, 64
-
-%macro FILTER_VER_CHROMA_AVX2_16x24 1
+    add             r0, 16
+    movq            xm1, [r0]                       ; m1 = row 0
+    movq            xm2, [r0 + r1]                  ; m2 = row 1
+    punpcklbw       xm1, xm2
+    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
+    punpcklbw       xm2, xm3
+    vinserti128     m5, m1, xm2, 1
+    pmaddubsw       m5, m8
+    movq            xm4, [r0 + r4]                  ; m4 = row 3
+    punpcklbw       xm3, xm4
+    lea             r7, [r0 + r1 * 4]
+    movq            xm1, [r7]                       ; m1 = row 4
+    punpcklbw       xm4, xm1
+    vinserti128     m2, m3, xm4, 1
+    pmaddubsw       m0, m2, m9
+    paddw           m5, m0
+    pmaddubsw       m2, m8
+    movq            xm3, [r7 + r1]                  ; m3 = row 5
+    punpcklbw       xm1, xm3
+    movq            xm4, [r7 + r1 * 2]              ; m4 = row 6
+    punpcklbw       xm3, xm4
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m0, m1, m9
+    paddw           m2, m0
+    pmaddubsw       m1, m8
+    movq            xm3, [r7 + r4]                  ; m3 = row 7
+    punpcklbw       xm4, xm3
+    lea             r7, [r7 + r1 * 4]
+    movq            xm0, [r7]                       ; m0 = row 8
+    punpcklbw       xm3, xm0
+    vinserti128     m4, m4, xm3, 1
+    pmaddubsw       m3, m4, m9
+    paddw           m1, m3
+    pmaddubsw       m4, m8
+    movq            xm3, [r7 + r1]                  ; m3 = row 9
+    punpcklbw       xm0, xm3
+    movq            xm6, [r7 + r1 * 2]              ; m6 = row 10
+    punpcklbw       xm3, xm6
+    vinserti128     m0, m0, xm3, 1
+    pmaddubsw       m3, m0, m9
+    paddw           m4, m3
+    pmaddubsw       m0, m8
+
+%ifidn %1,pp
+    pmulhrsw        m5, m7                          ; m5 = word: row 0, row 1
+    pmulhrsw        m2, m7                          ; m2 = word: row 2, row 3
+    pmulhrsw        m1, m7                          ; m1 = word: row 4, row 5
+    pmulhrsw        m4, m7                          ; m4 = word: row 6, row 7
+    packuswb        m5, m2
+    packuswb        m1, m4
+    vextracti128    xm2, m5, 1
+    vextracti128    xm4, m1, 1
+    movq            [r2], xm5
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm5
+    movhps          [r2 + r6], xm2
+    lea             r8, [r2 + r3 * 4]
+    movq            [r8], xm1
+    movq            [r8 + r3], xm4
+    movhps          [r8 + r3 * 2], xm1
+    movhps          [r8 + r6], xm4
+%else
+    psubw           m5, m7                          ; m5 = word: row 0, row 1
+    psubw           m2, m7                          ; m2 = word: row 2, row 3
+    psubw           m1, m7                          ; m1 = word: row 4, row 5
+    psubw           m4, m7                          ; m4 = word: row 6, row 7
+    vextracti128    xm3, m5, 1
+    movu            [r2], xm5
+    movu            [r2 + r3], xm3
+    vextracti128    xm3, m2, 1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
+    vextracti128    xm3, m1, 1
+    lea             r8, [r2 + r3 * 4]
+    movu            [r8], xm1
+    movu            [r8 + r3], xm3
+    vextracti128    xm3, m4, 1
+    movu            [r8 + r3 * 2], xm4
+    movu            [r8 + r6], xm3
+%endif
+    lea             r8, [r8 + r3 * 4]
+
+    movq            xm3, [r7 + r4]                  ; m3 = row 11
+    punpcklbw       xm6, xm3
+    lea             r7, [r7 + r1 * 4]
+    movq            xm5, [r7]                       ; m5 = row 12
+    punpcklbw       xm3, xm5
+    vinserti128     m6, m6, xm3, 1
+    pmaddubsw       m3, m6, m9
+    paddw           m0, m3
+    pmaddubsw       m6, m8
+    movq            xm3, [r7 + r1]                  ; m3 = row 13
+    punpcklbw       xm5, xm3
+    movq            xm2, [r7 + r1 * 2]              ; m2 = row 14
+    punpcklbw       xm3, xm2
+    vinserti128     m5, m5, xm3, 1
+    pmaddubsw       m3, m5, m9
+    paddw           m6, m3
+    pmaddubsw       m5, m8
+    movq            xm3, [r7 + r4]                  ; m3 = row 15
+    punpcklbw       xm2, xm3
+    lea             r7, [r7 + r1 * 4]
+    movq            xm1, [r7]                       ; m1 = row 16
+    punpcklbw       xm3, xm1
+    vinserti128     m2, m2, xm3, 1
+    pmaddubsw       m3, m2, m9
+    paddw           m5, m3
+    pmaddubsw       m2, m8
+    movq            xm3, [r7 + r1]                  ; m3 = row 17
+    punpcklbw       xm1, xm3
+    movq            xm4, [r7 + r1 * 2]              ; m4 = row 18
+    punpcklbw       xm3, xm4
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m3, m1, m9
+    paddw           m2, m3
+%ifidn %1,pp
+    pmulhrsw        m0, m7                          ; m0 = word: row 8, row 9
+    pmulhrsw        m6, m7                          ; m6 = word: row 10, row 11
+    pmulhrsw        m5, m7                          ; m5 = word: row 12, row 13
+    pmulhrsw        m2, m7                          ; m2 = word: row 14, row 15
+    packuswb        m0, m6
+    packuswb        m5, m2
+    vextracti128    xm6, m0, 1
+    vextracti128    xm2, m5, 1
+    movq            [r8], xm0
+    movq            [r8 + r3], xm6
+    movhps          [r8 + r3 * 2], xm0
+    movhps          [r8 + r6], xm6
+    lea             r8, [r8 + r3 * 4]
+    movq            [r8], xm5
+    movq            [r8 + r3], xm2
+    movhps          [r8 + r3 * 2], xm5
+    movhps          [r8 + r6], xm2
+    lea             r2, [r8 + r3 * 4 - 16]
+%else
+    psubw           m0, m7                          ; m0 = word: row 8, row 9
+    psubw           m6, m7                          ; m6 = word: row 10, row 11
+    psubw           m5, m7                          ; m5 = word: row 12, row 13
+    psubw           m2, m7                          ; m2 = word: row 14, row 15
+    vextracti128    xm3, m0, 1
+    movu            [r8], xm0
+    movu            [r8 + r3], xm3
+    vextracti128    xm3, m6, 1
+    movu            [r8 + r3 * 2], xm6
+    movu            [r8 + r6], xm3
+    vextracti128    xm3, m5, 1
+    lea             r8, [r8 + r3 * 4]
+    movu            [r8], xm5
+    movu            [r8 + r3], xm3
+    vextracti128    xm3, m2, 1
+    movu            [r8 + r3 * 2], xm2
+    movu            [r8 + r6], xm3
+    lea             r2, [r8 + r3 * 4 - 32]
+%endif
+    lea             r0, [r7 - 16]
+    dec             r5d
+    jnz             .loopH
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_24x32 pp
+    FILTER_VER_CHROMA_AVX2_24x32 ps
+
+%macro FILTER_VER_CHROMA_AVX2_24x64 1
 %if ARCH_X86_64 == 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_16x24, 4, 6, 15
+cglobal interp_4tap_vert_%1_24x64, 4, 7, 13
     mov             r4d, r4m
     shl             r4d, 6
 
@@ -9904,367 +7942,138 @@ cglobal interp_4tap_vert_%1_16x24, 4, 6, 15
     lea             r5, [tab_ChromaCoeffVer_32 + r4]
 %endif
 
-    mova            m12, [r5]
-    mova            m13, [r5 + mmsize]
+    mova            m10, [r5]
+    mova            m11, [r5 + mmsize]
     lea             r4, [r1 * 3]
     sub             r0, r1
 %ifidn %1,pp
-    mova            m14, [pw_512]
+    mova            m12, [pw_512]
 %else
     add             r3d, r3d
-    vbroadcasti128  m14, [pw_2000]
+    vbroadcasti128  m12, [pw_2000]
 %endif
     lea             r5, [r3 * 3]
-
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m0, m12
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m1, m12
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, m13
-    paddw           m0, m4
-    pmaddubsw       m2, m12
+    mov             r6d, 16
+.loopH:
+    movu            m0, [r0]                        ; m0 = row 0
+    movu            m1, [r0 + r1]                   ; m1 = row 1
+    punpcklbw       m2, m0, m1
+    punpckhbw       m3, m0, m1
+    pmaddubsw       m2, m10
+    pmaddubsw       m3, m10
+    movu            m0, [r0 + r1 * 2]               ; m0 = row 2
+    punpcklbw       m4, m1, m0
+    punpckhbw       m5, m1, m0
+    pmaddubsw       m4, m10
+    pmaddubsw       m5, m10
+    movu            m1, [r0 + r4]                   ; m1 = row 3
+    punpcklbw       m6, m0, m1
+    punpckhbw       m7, m0, m1
+    pmaddubsw       m8, m6, m11
+    pmaddubsw       m9, m7, m11
+    pmaddubsw       m6, m10
+    pmaddubsw       m7, m10
+    paddw           m2, m8
+    paddw           m3, m9
+%ifidn %1,pp
+    pmulhrsw        m2, m12
+    pmulhrsw        m3, m12
+    packuswb        m2, m3
+    movu            [r2], xm2
+    vextracti128    xm2, m2, 1
+    movq            [r2 + 16], xm2
+%else
+    psubw           m2, m12
+    psubw           m3, m12
+    vperm2i128      m0, m2, m3, 0x20
+    vperm2i128      m2, m2, m3, 0x31
+    movu            [r2], m0
+    movu            [r2 + mmsize], xm2
+%endif
     lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhbw       xm5, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, m13
-    paddw           m1, m5
-    pmaddubsw       m3, m12
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhbw       xm6, xm4, xm5
-    punpcklbw       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, m13
-    paddw           m2, m6
-    pmaddubsw       m4, m12
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhbw       xm7, xm5, xm6
-    punpcklbw       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m7, m5, m13
-    paddw           m3, m7
-    pmaddubsw       m5, m12
-    movu            xm7, [r0 + r4]                  ; m7 = row 7
-    punpckhbw       xm8, xm6, xm7
-    punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm8, 1
-    pmaddubsw       m8, m6, m13
+    movu            m0, [r0]                        ; m0 = row 4
+    punpcklbw       m2, m1, m0
+    punpckhbw       m3, m1, m0
+    pmaddubsw       m8, m2, m11
+    pmaddubsw       m9, m3, m11
+    pmaddubsw       m2, m10
+    pmaddubsw       m3, m10
     paddw           m4, m8
-    pmaddubsw       m6, m12
-    lea             r0, [r0 + r1 * 4]
-    movu            xm8, [r0]                       ; m8 = row 8
-    punpckhbw       xm9, xm7, xm8
-    punpcklbw       xm7, xm8
-    vinserti128     m7, m7, xm9, 1
-    pmaddubsw       m9, m7, m13
     paddw           m5, m9
-    pmaddubsw       m7, m12
-    movu            xm9, [r0 + r1]                  ; m9 = row 9
-    punpckhbw       xm10, xm8, xm9
-    punpcklbw       xm8, xm9
-    vinserti128     m8, m8, xm10, 1
-    pmaddubsw       m10, m8, m13
-    paddw           m6, m10
-    pmaddubsw       m8, m12
-    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
-    punpckhbw       xm11, xm9, xm10
-    punpcklbw       xm9, xm10
-    vinserti128     m9, m9, xm11, 1
-    pmaddubsw       m11, m9, m13
-    paddw           m7, m11
-    pmaddubsw       m9, m12
-
 %ifidn %1,pp
-    pmulhrsw        m0, m14                         ; m0 = word: row 0
-    pmulhrsw        m1, m14                         ; m1 = word: row 1
-    pmulhrsw        m2, m14                         ; m2 = word: row 2
-    pmulhrsw        m3, m14                         ; m3 = word: row 3
-    pmulhrsw        m4, m14                         ; m4 = word: row 4
-    pmulhrsw        m5, m14                         ; m5 = word: row 5
-    pmulhrsw        m6, m14                         ; m6 = word: row 6
-    pmulhrsw        m7, m14                         ; m7 = word: row 7
-    packuswb        m0, m1
-    packuswb        m2, m3
+    pmulhrsw        m4, m12
+    pmulhrsw        m5, m12
     packuswb        m4, m5
-    packuswb        m6, m7
-    vpermq          m0, m0, q3120
-    vpermq          m2, m2, q3120
-    vpermq          m4, m4, q3120
-    vpermq          m6, m6, q3120
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
-    vextracti128    xm5, m4, 1
-    vextracti128    xm7, m6, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r5], xm3
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm4
-    movu            [r2 + r3], xm5
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r5], xm7
+    movu            [r2 + r3], xm4
+    vextracti128    xm4, m4, 1
+    movq            [r2 + r3 + 16], xm4
 %else
-    psubw           m0, m14                         ; m0 = word: row 0
-    psubw           m1, m14                         ; m1 = word: row 1
-    psubw           m2, m14                         ; m2 = word: row 2
-    psubw           m3, m14                         ; m3 = word: row 3
-    psubw           m4, m14                         ; m4 = word: row 4
-    psubw           m5, m14                         ; m5 = word: row 5
-    psubw           m6, m14                         ; m6 = word: row 6
-    psubw           m7, m14                         ; m7 = word: row 7
-    movu            [r2], m0
+    psubw           m4, m12
+    psubw           m5, m12
+    vperm2i128      m1, m4, m5, 0x20
+    vperm2i128      m4, m4, m5, 0x31
     movu            [r2 + r3], m1
-    movu            [r2 + r3 * 2], m2
-    movu            [r2 + r5], m3
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], m4
-    movu            [r2 + r3], m5
-    movu            [r2 + r3 * 2], m6
-    movu            [r2 + r5], m7
+    movu            [r2 + r3 + mmsize], xm4
 %endif
-    lea             r2, [r2 + r3 * 4]
-
-    movu            xm11, [r0 + r4]                 ; m11 = row 11
-    punpckhbw       xm6, xm10, xm11
-    punpcklbw       xm10, xm11
-    vinserti128     m10, m10, xm6, 1
-    pmaddubsw       m6, m10, m13
-    paddw           m8, m6
-    pmaddubsw       m10, m12
-    lea             r0, [r0 + r1 * 4]
-    movu            xm6, [r0]                       ; m6 = row 12
-    punpckhbw       xm7, xm11, xm6
-    punpcklbw       xm11, xm6
-    vinserti128     m11, m11, xm7, 1
-    pmaddubsw       m7, m11, m13
-    paddw           m9, m7
-    pmaddubsw       m11, m12
-
-    movu            xm7, [r0 + r1]                  ; m7 = row 13
-    punpckhbw       xm0, xm6, xm7
-    punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm0, 1
-    pmaddubsw       m0, m6, m13
-    paddw           m10, m0
-    pmaddubsw       m6, m12
-    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
-    punpckhbw       xm1, xm7, xm0
-    punpcklbw       xm7, xm0
-    vinserti128     m7, m7, xm1, 1
-    pmaddubsw       m1, m7, m13
-    paddw           m11, m1
-    pmaddubsw       m7, m12
-    movu            xm1, [r0 + r4]                  ; m1 = row 15
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m2, m0, m13
-    paddw           m6, m2
-    pmaddubsw       m0, m12
-    lea             r0, [r0 + r1 * 4]
-    movu            xm2, [r0]                       ; m2 = row 16
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m3, m1, m13
-    paddw           m7, m3
-    pmaddubsw       m1, m12
-    movu            xm3, [r0 + r1]                  ; m3 = row 17
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, m13
-    paddw           m0, m4
-    pmaddubsw       m2, m12
-    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
-    punpckhbw       xm5, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, m13
-    paddw           m1, m5
-    pmaddubsw       m3, m12
 
+    movu            m1, [r0 + r1]                   ; m1 = row 5
+    punpcklbw       m4, m0, m1
+    punpckhbw       m5, m0, m1
+    pmaddubsw       m4, m11
+    pmaddubsw       m5, m11
+    paddw           m6, m4
+    paddw           m7, m5
 %ifidn %1,pp
-    pmulhrsw        m8, m14                         ; m8 = word: row 8
-    pmulhrsw        m9, m14                         ; m9 = word: row 9
-    pmulhrsw        m10, m14                        ; m10 = word: row 10
-    pmulhrsw        m11, m14                        ; m11 = word: row 11
-    pmulhrsw        m6, m14                         ; m6 = word: row 12
-    pmulhrsw        m7, m14                         ; m7 = word: row 13
-    pmulhrsw        m0, m14                         ; m0 = word: row 14
-    pmulhrsw        m1, m14                         ; m1 = word: row 15
-    packuswb        m8, m9
-    packuswb        m10, m11
+    pmulhrsw        m6, m12
+    pmulhrsw        m7, m12
     packuswb        m6, m7
-    packuswb        m0, m1
-    vpermq          m8, m8, q3120
-    vpermq          m10, m10, q3120
-    vpermq          m6, m6, q3120
-    vpermq          m0, m0, q3120
-    vextracti128    xm9, m8, 1
-    vextracti128    xm11, m10, 1
-    vextracti128    xm7, m6, 1
-    vextracti128    xm1, m0, 1
-    movu            [r2], xm8
-    movu            [r2 + r3], xm9
-    movu            [r2 + r3 * 2], xm10
-    movu            [r2 + r5], xm11
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm6
-    movu            [r2 + r3], xm7
-    movu            [r2 + r3 * 2], xm0
-    movu            [r2 + r5], xm1
+    movu            [r2 + r3 * 2], xm6
+    vextracti128    xm6, m6, 1
+    movq            [r2 + r3 * 2 + 16], xm6
 %else
-    psubw           m8, m14                         ; m8 = word: row 8
-    psubw           m9, m14                         ; m9 = word: row 9
-    psubw           m10, m14                        ; m10 = word: row 10
-    psubw           m11, m14                        ; m11 = word: row 11
-    psubw           m6, m14                         ; m6 = word: row 12
-    psubw           m7, m14                         ; m7 = word: row 13
-    psubw           m0, m14                         ; m0 = word: row 14
-    psubw           m1, m14                         ; m1 = word: row 15
-    movu            [r2], m8
-    movu            [r2 + r3], m9
-    movu            [r2 + r3 * 2], m10
-    movu            [r2 + r5], m11
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], m6
-    movu            [r2 + r3], m7
+    psubw           m6, m12
+    psubw           m7, m12
+    vperm2i128      m0, m6, m7, 0x20
+    vperm2i128      m6, m6, m7, 0x31
     movu            [r2 + r3 * 2], m0
-    movu            [r2 + r5], m1
+    movu            [r2 + r3 * 2 + mmsize], xm6
 %endif
-    lea             r2, [r2 + r3 * 4]
 
-    movu            xm5, [r0 + r4]                  ; m5 = row 19
-    punpckhbw       xm6, xm4, xm5
-    punpcklbw       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, m13
+    movu            m0, [r0 + r1 * 2]               ; m0 = row 6
+    punpcklbw       m6, m1, m0
+    punpckhbw       m7, m1, m0
+    pmaddubsw       m6, m11
+    pmaddubsw       m7, m11
     paddw           m2, m6
-    pmaddubsw       m4, m12
-    lea             r0, [r0 + r1 * 4]
-    movu            xm6, [r0]                       ; m6 = row 20
-    punpckhbw       xm7, xm5, xm6
-    punpcklbw       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m7, m5, m13
     paddw           m3, m7
-    pmaddubsw       m5, m12
-    movu            xm7, [r0 + r1]                  ; m7 = row 21
-    punpckhbw       xm0, xm6, xm7
-    punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm0, 1
-    pmaddubsw       m0, m6, m13
-    paddw           m4, m0
-    pmaddubsw       m6, m12
-    movu            xm0, [r0 + r1 * 2]              ; m0 = row 22
-    punpckhbw       xm1, xm7, xm0
-    punpcklbw       xm7, xm0
-    vinserti128     m7, m7, xm1, 1
-    pmaddubsw       m1, m7, m13
-    paddw           m5, m1
-    pmaddubsw       m7, m12
-    movu            xm1, [r0 + r4]                  ; m1 = row 23
-    punpckhbw       xm8, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm8, 1
-    pmaddubsw       m8, m0, m13
-    paddw           m6, m8
-    pmaddubsw       m0, m12
-    lea             r0, [r0 + r1 * 4]
-    movu            xm8, [r0]                       ; m8 = row 24
-    punpckhbw       xm9, xm1, xm8
-    punpcklbw       xm1, xm8
-    vinserti128     m1, m1, xm9, 1
-    pmaddubsw       m9, m1, m13
-    paddw           m7, m9
-    pmaddubsw       m1, m12
-    movu            xm9, [r0 + r1]                  ; m9 = row 25
-    punpckhbw       xm10, xm8, xm9
-    punpcklbw       xm8, xm9
-    vinserti128     m8, m8, xm10, 1
-    pmaddubsw       m8, m13
-    paddw           m0, m8
-    movu            xm10, [r0 + r1 * 2]             ; m10 = row 26
-    punpckhbw       xm11, xm9, xm10
-    punpcklbw       xm9, xm10
-    vinserti128     m9, m9, xm11, 1
-    pmaddubsw       m9, m13
-    paddw           m1, m9
-
 %ifidn %1,pp
-    pmulhrsw        m2, m14                         ; m2 = word: row 16
-    pmulhrsw        m3, m14                         ; m3 = word: row 17
-    pmulhrsw        m4, m14                         ; m4 = word: row 18
-    pmulhrsw        m5, m14                         ; m5 = word: row 19
-    pmulhrsw        m6, m14                         ; m6 = word: row 20
-    pmulhrsw        m7, m14                         ; m7 = word: row 21
-    pmulhrsw        m0, m14                         ; m0 = word: row 22
-    pmulhrsw        m1, m14                         ; m1 = word: row 23
+    pmulhrsw        m2, m12
+    pmulhrsw        m3, m12
     packuswb        m2, m3
-    packuswb        m4, m5
-    packuswb        m6, m7
-    packuswb        m0, m1
-    vpermq          m2, m2, q3120
-    vpermq          m4, m4, q3120
-    vpermq          m6, m6, q3120
-    vpermq          m0, m0, q3120
-    vextracti128    xm3, m2, 1
-    vextracti128    xm5, m4, 1
-    vextracti128    xm7, m6, 1
-    vextracti128    xm1, m0, 1
-    movu            [r2], xm2
-    movu            [r2 + r3], xm3
-    movu            [r2 + r3 * 2], xm4
-    movu            [r2 + r5], xm5
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm6
-    movu            [r2 + r3], xm7
-    movu            [r2 + r3 * 2], xm0
-    movu            [r2 + r5], xm1
+    movu            [r2 + r5], xm2
+    vextracti128    xm2, m2, 1
+    movq            [r2 + r5 + 16], xm2
 %else
-    psubw           m2, m14                         ; m2 = word: row 16
-    psubw           m3, m14                         ; m3 = word: row 17
-    psubw           m4, m14                         ; m4 = word: row 18
-    psubw           m5, m14                         ; m5 = word: row 19
-    psubw           m6, m14                         ; m6 = word: row 20
-    psubw           m7, m14                         ; m7 = word: row 21
-    psubw           m0, m14                         ; m0 = word: row 22
-    psubw           m1, m14                         ; m1 = word: row 23
-    movu            [r2], m2
-    movu            [r2 + r3], m3
-    movu            [r2 + r3 * 2], m4
-    movu            [r2 + r5], m5
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], m6
-    movu            [r2 + r3], m7
-    movu            [r2 + r3 * 2], m0
-    movu            [r2 + r5], m1
+    psubw           m2, m12
+    psubw           m3, m12
+    vperm2i128      m0, m2, m3, 0x20
+    vperm2i128      m2, m2, m3, 0x31
+    movu            [r2 + r5], m0
+    movu            [r2 + r5 + mmsize], xm2
 %endif
+    lea             r2, [r2 + r3 * 4]
+    dec             r6d
+    jnz             .loopH
     RET
 %endif
 %endmacro
 
-    FILTER_VER_CHROMA_AVX2_16x24 pp
-    FILTER_VER_CHROMA_AVX2_16x24 ps
+    FILTER_VER_CHROMA_AVX2_24x64 pp
+    FILTER_VER_CHROMA_AVX2_24x64 ps
 
-%macro FILTER_VER_CHROMA_AVX2_24x32 1
+%macro FILTER_VER_CHROMA_AVX2_16x4 1
 INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_vert_%1_24x32, 4, 9, 10
+cglobal interp_4tap_vert_%1_16x4, 4, 6, 8
     mov             r4d, r4m
     shl             r4d, 6
 
@@ -10275,19 +8084,15 @@ cglobal interp_4tap_vert_%1_24x32, 4, 9, 10
     lea             r5, [tab_ChromaCoeffVer_32 + r4]
 %endif
 
-    mova            m8, [r5]
-    mova            m9, [r5 + mmsize]
     lea             r4, [r1 * 3]
     sub             r0, r1
 %ifidn %1,pp
     mova            m7, [pw_512]
 %else
     add             r3d, r3d
-    vbroadcasti128  m7, [pw_2000]
+    mova            m7, [pw_2000]
 %endif
-    lea             r6, [r3 * 3]
-    mov             r5d, 2
-.loopH:
+
     movu            xm0, [r0]
     vinserti128     m0, m0, [r0 + r1 * 2], 1
     movu            xm1, [r0 + r1]
@@ -10297,40 +8102,41 @@ cglobal interp_4tap_vert_%1_24x32, 4, 9, 10
     punpckhbw       m3, m0, m1
     vperm2i128      m4, m2, m3, 0x20
     vperm2i128      m2, m2, m3, 0x31
-    pmaddubsw       m4, m8
-    pmaddubsw       m3, m2, m9
+    pmaddubsw       m4, [r5]
+    pmaddubsw       m3, m2, [r5 + mmsize]
     paddw           m4, m3
-    pmaddubsw       m2, m8
+    pmaddubsw       m2, [r5]
 
     vextracti128    xm0, m0, 1
-    lea             r7, [r0 + r1 * 4]
-    vinserti128     m0, m0, [r7], 1
+    lea             r0, [r0 + r1 * 4]
+    vinserti128     m0, m0, [r0], 1
 
     punpcklbw       m5, m1, m0
     punpckhbw       m3, m1, m0
     vperm2i128      m6, m5, m3, 0x20
     vperm2i128      m5, m5, m3, 0x31
-    pmaddubsw       m6, m8
-    pmaddubsw       m3, m5, m9
+    pmaddubsw       m6, [r5]
+    pmaddubsw       m3, m5, [r5 + mmsize]
     paddw           m6, m3
-    pmaddubsw       m5, m8
+    pmaddubsw       m5, [r5]
 %ifidn %1,pp
-    pmulhrsw        m4, m7                         ; m4 = word: row 0
-    pmulhrsw        m6, m7                         ; m6 = word: row 1
+    pmulhrsw        m4, m7                          ; m4 = word: row 0
+    pmulhrsw        m6, m7                          ; m6 = word: row 1
     packuswb        m4, m6
     vpermq          m4, m4, 11011000b
     vextracti128    xm6, m4, 1
     movu            [r2], xm4
     movu            [r2 + r3], xm6
 %else
-    psubw           m4, m7                         ; m4 = word: row 0
-    psubw           m6, m7                         ; m6 = word: row 1
+    psubw           m4, m7                          ; m4 = word: row 0
+    psubw           m6, m7                          ; m6 = word: row 1
     movu            [r2], m4
     movu            [r2 + r3], m6
 %endif
+    lea             r2, [r2 + r3 * 2]
 
-    movu            xm4, [r7 + r1 * 2]
-    vinserti128     m4, m4, [r7 + r1], 1
+    movu            xm4, [r0 + r1 * 2]
+    vinserti128     m4, m4, [r0 + r1], 1
     vextracti128    xm1, m4, 1
     vinserti128     m0, m0, xm1, 0
 
@@ -10338,390 +8144,580 @@ cglobal interp_4tap_vert_%1_24x32, 4, 9, 10
     punpckhbw       m1, m0, m4
     vperm2i128      m0, m6, m1, 0x20
     vperm2i128      m6, m6, m1, 0x31
-    pmaddubsw       m1, m0, m9
-    paddw           m5, m1
-    pmaddubsw       m0, m8
-    pmaddubsw       m1, m6, m9
-    paddw           m2, m1
-    pmaddubsw       m6, m8
+    pmaddubsw       m0, [r5 + mmsize]
+    paddw           m5, m0
+    pmaddubsw       m6, [r5 + mmsize]
+    paddw           m2, m6
 
 %ifidn %1,pp
-    pmulhrsw        m2, m7                         ; m2 = word: row 2
-    pmulhrsw        m5, m7                         ; m5 = word: row 3
+    pmulhrsw        m2, m7                          ; m2 = word: row 2
+    pmulhrsw        m5, m7                          ; m5 = word: row 3
     packuswb        m2, m5
     vpermq          m2, m2, 11011000b
     vextracti128    xm5, m2, 1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm5
-%else
-    psubw           m2, m7                         ; m2 = word: row 2
-    psubw           m5, m7                         ; m5 = word: row 3
-    movu            [r2 + r3 * 2], m2
-    movu            [r2 + r6], m5
-%endif
-    lea             r8, [r2 + r3 * 4]
-
-    movu            xm1, [r7 + r4]
-    lea             r7, [r7 + r1 * 4]
-    vinserti128     m1, m1, [r7], 1
-    vinserti128     m4, m4, xm1, 1
-
-    punpcklbw       m2, m4, m1
-    punpckhbw       m5, m4, m1
-    vperm2i128      m3, m2, m5, 0x20
-    vperm2i128      m2, m2, m5, 0x31
-    pmaddubsw       m5, m3, m9
-    paddw           m6, m5
-    pmaddubsw       m3, m8
-    pmaddubsw       m5, m2, m9
-    paddw           m0, m5
-    pmaddubsw       m2, m8
-
-%ifidn %1,pp
-    pmulhrsw        m6, m7                         ; m6 = word: row 4
-    pmulhrsw        m0, m7                         ; m0 = word: row 5
-    packuswb        m6, m0
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm0, m6, 1
-    movu            [r8], xm6
-    movu            [r8 + r3], xm0
+    movu            [r2], xm2
+    movu            [r2 + r3], xm5
 %else
-    psubw           m6, m7                         ; m6 = word: row 4
-    psubw           m0, m7                         ; m0 = word: row 5
-    movu            [r8], m6
-    movu            [r8 + r3], m0
+    psubw           m2, m7                          ; m2 = word: row 2
+    psubw           m5, m7                          ; m5 = word: row 3
+    movu            [r2], m2
+    movu            [r2 + r3], m5
 %endif
+    RET
+%endmacro
 
-    movu            xm6, [r7 + r1 * 2]
-    vinserti128     m6, m6, [r7 + r1], 1
-    vextracti128    xm0, m6, 1
-    vinserti128     m1, m1, xm0, 0
+    FILTER_VER_CHROMA_AVX2_16x4 pp
+    FILTER_VER_CHROMA_AVX2_16x4 ps
 
-    punpcklbw       m4, m1, m6
-    punpckhbw       m5, m1, m6
-    vperm2i128      m0, m4, m5, 0x20
-    vperm2i128      m5, m4, m5, 0x31
-    pmaddubsw       m4, m0, m9
-    paddw           m2, m4
-    pmaddubsw       m0, m8
-    pmaddubsw       m4, m5, m9
-    paddw           m3, m4
-    pmaddubsw       m5, m8
+%macro FILTER_VER_CHROMA_AVX2_12xN 2
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_12x%2, 4, 7, 8
+    mov             r4d, r4m
+    shl             r4d, 6
 
-%ifidn %1,pp
-    pmulhrsw        m3, m7                         ; m3 = word: row 6
-    pmulhrsw        m2, m7                         ; m2 = word: row 7
-    packuswb        m3, m2
-    vpermq          m3, m3, 11011000b
-    vextracti128    xm2, m3, 1
-    movu            [r8 + r3 * 2], xm3
-    movu            [r8 + r6], xm2
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
 %else
-    psubw           m3, m7                         ; m3 = word: row 6
-    psubw           m2, m7                         ; m2 = word: row 7
-    movu            [r8 + r3 * 2], m3
-    movu            [r8 + r6], m2
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
 %endif
-    lea             r8, [r8 + r3 * 4]
-
-    movu            xm3, [r7 + r4]
-    lea             r7, [r7 + r1 * 4]
-    vinserti128     m3, m3, [r7], 1
-    vinserti128     m6, m6, xm3, 1
-
-    punpcklbw       m2, m6, m3
-    punpckhbw       m1, m6, m3
-    vperm2i128      m4, m2, m1, 0x20
-    vperm2i128      m2, m2, m1, 0x31
-    pmaddubsw       m1, m4, m9
-    paddw           m5, m1
-    pmaddubsw       m4, m8
-    pmaddubsw       m1, m2, m9
-    paddw           m0, m1
-    pmaddubsw       m2, m8
 
+    lea             r4, [r1 * 3]
+    sub             r0, r1
 %ifidn %1,pp
-    pmulhrsw        m5, m7                         ; m5 = word: row 8
-    pmulhrsw        m0, m7                         ; m0 = word: row 9
-    packuswb        m5, m0
-    vpermq          m5, m5, 11011000b
-    vextracti128    xm0, m5, 1
-    movu            [r8], xm5
-    movu            [r8 + r3], xm0
+    mova            m7, [pw_512]
 %else
-    psubw           m5, m7                         ; m5 = word: row 8
-    psubw           m0, m7                         ; m0 = word: row 9
-    movu            [r8], m5
-    movu            [r8 + r3], m0
+    add             r3d, r3d
+    vbroadcasti128  m7, [pw_2000]
 %endif
-
-    movu            xm5, [r7 + r1 * 2]
-    vinserti128     m5, m5, [r7 + r1], 1
-    vextracti128    xm0, m5, 1
-    vinserti128     m3, m3, xm0, 0
-
-    punpcklbw       m1, m3, m5
-    punpckhbw       m0, m3, m5
-    vperm2i128      m6, m1, m0, 0x20
-    vperm2i128      m0, m1, m0, 0x31
-    pmaddubsw       m1, m6, m9
-    paddw           m2, m1
-    pmaddubsw       m6, m8
-    pmaddubsw       m1, m0, m9
-    paddw           m4, m1
-    pmaddubsw       m0, m8
-
+    lea             r6, [r3 * 3]
+%rep %2 / 16
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddubsw       m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
+    paddw           m0, m4
+    pmaddubsw       m2, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
+    paddw           m1, m5
+    pmaddubsw       m3, [r5]
 %ifidn %1,pp
-    pmulhrsw        m4, m7                         ; m4 = word: row 10
-    pmulhrsw        m2, m7                         ; m2 = word: row 11
-    packuswb        m4, m2
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm2, m4, 1
-    movu            [r8 + r3 * 2], xm4
-    movu            [r8 + r6], xm2
+    pmulhrsw        m0, m7                          ; m0 = word: row 0
+    pmulhrsw        m1, m7                          ; m1 = word: row 1
+    packuswb        m0, m1
+    vextracti128    xm1, m0, 1
+    movq            [r2], xm0
+    movd            [r2 + 8], xm1
+    movhps          [r2 + r3], xm0
+    pextrd          [r2 + r3 + 8], xm1, 2
 %else
-    psubw           m4, m7                         ; m4 = word: row 10
-    psubw           m2, m7                         ; m2 = word: row 11
-    movu            [r8 + r3 * 2], m4
-    movu            [r8 + r6], m2
+    psubw           m0, m7                          ; m0 = word: row 0
+    psubw           m1, m7                          ; m1 = word: row 1
+    movu            [r2], xm0
+    vextracti128    xm0, m0, 1
+    movq            [r2 + 16], xm0
+    movu            [r2 + r3], xm1
+    vextracti128    xm1, m1, 1
+    movq            [r2 + r3 + 16], xm1
 %endif
-    lea             r8, [r8 + r3 * 4]
-
-    movu            xm3, [r7 + r4]
-    lea             r7, [r7 + r1 * 4]
-    vinserti128     m3, m3, [r7], 1
-    vinserti128     m5, m5, xm3, 1
-
-    punpcklbw       m2, m5, m3
-    punpckhbw       m1, m5, m3
-    vperm2i128      m4, m2, m1, 0x20
-    vperm2i128      m2, m2, m1, 0x31
-    pmaddubsw       m1, m4, m9
-    paddw           m0, m1
-    pmaddubsw       m4, m8
-    pmaddubsw       m1, m2, m9
-    paddw           m6, m1
-    pmaddubsw       m2, m8
 
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhbw       xm6, xm4, xm5
+    punpcklbw       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
+    paddw           m2, m6
+    pmaddubsw       m4, [r5]
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhbw       xm0, xm5, xm6
+    punpcklbw       xm5, xm6
+    vinserti128     m5, m5, xm0, 1
+    pmaddubsw       m0, m5, [r5 + 1 * mmsize]
+    paddw           m3, m0
+    pmaddubsw       m5, [r5]
 %ifidn %1,pp
-    pmulhrsw        m0, m7                         ; m0 = word: row 12
-    pmulhrsw        m6, m7                         ; m6 = word: row 13
-    packuswb        m0, m6
-    vpermq          m0, m0, 11011000b
-    vextracti128    xm6, m0, 1
-    movu            [r8], xm0
-    movu            [r8 + r3], xm6
+    pmulhrsw        m2, m7                          ; m2 = word: row 2
+    pmulhrsw        m3, m7                          ; m3 = word: row 3
+    packuswb        m2, m3
+    vextracti128    xm3, m2, 1
+    movq            [r2 + r3 * 2], xm2
+    movd            [r2 + r3 * 2 + 8], xm3
+    movhps          [r2 + r6], xm2
+    pextrd          [r2 + r6 + 8], xm3, 2
 %else
-    psubw           m0, m7                         ; m0 = word: row 12
-    psubw           m6, m7                         ; m6 = word: row 13
-    movu            [r8], m0
-    movu            [r8 + r3], m6
+    psubw           m2, m7                          ; m2 = word: row 2
+    psubw           m3, m7                          ; m3 = word: row 3
+    movu            [r2 + r3 * 2], xm2
+    vextracti128    xm2, m2, 1
+    movq            [r2 + r3 * 2 + 16], xm2
+    movu            [r2 + r6], xm3
+    vextracti128    xm3, m3, 1
+    movq            [r2 + r6 + 16], xm3
 %endif
+    lea             r2, [r2 + r3 * 4]
 
-    movu            xm5, [r7 + r1 * 2]
-    vinserti128     m5, m5, [r7 + r1], 1
-    vextracti128    xm0, m5, 1
-    vinserti128     m3, m3, xm0, 0
-
-    punpcklbw       m1, m3, m5
-    punpckhbw       m0, m3, m5
-    vperm2i128      m6, m1, m0, 0x20
-    vperm2i128      m0, m1, m0, 0x31
-    pmaddubsw       m6, m9
-    paddw           m2, m6
-    pmaddubsw       m0, m9
-    paddw           m4, m0
-
+    movu            xm0, [r0 + r4]                  ; m0 = row 7
+    punpckhbw       xm3, xm6, xm0
+    punpcklbw       xm6, xm0
+    vinserti128     m6, m6, xm3, 1
+    pmaddubsw       m3, m6, [r5 + 1 * mmsize]
+    paddw           m4, m3
+    pmaddubsw       m6, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm3, [r0]                       ; m3 = row 8
+    punpckhbw       xm1, xm0, xm3
+    punpcklbw       xm0, xm3
+    vinserti128     m0, m0, xm1, 1
+    pmaddubsw       m1, m0, [r5 + 1 * mmsize]
+    paddw           m5, m1
+    pmaddubsw       m0, [r5]
 %ifidn %1,pp
-    pmulhrsw        m4, m7                         ; m4 = word: row 14
-    pmulhrsw        m2, m7                         ; m2 = word: row 15
-    packuswb        m4, m2
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm2, m4, 1
-    movu            [r8 + r3 * 2], xm4
-    movu            [r8 + r6], xm2
-    add             r2, 16
+    pmulhrsw        m4, m7                          ; m4 = word: row 4
+    pmulhrsw        m5, m7                          ; m5 = word: row 5
+    packuswb        m4, m5
+    vextracti128    xm5, m4, 1
+    movq            [r2], xm4
+    movd            [r2 + 8], xm5
+    movhps          [r2 + r3], xm4
+    pextrd          [r2 + r3 + 8], xm5, 2
 %else
-    psubw           m4, m7                         ; m4 = word: row 14
-    psubw           m2, m7                         ; m2 = word: row 15
-    movu            [r8 + r3 * 2], m4
-    movu            [r8 + r6], m2
-    add             r2, 32
+    psubw           m4, m7                          ; m4 = word: row 4
+    psubw           m5, m7                          ; m5 = word: row 5
+    movu            [r2], xm4
+    vextracti128    xm4, m4, 1
+    movq            [r2 + 16], xm4
+    movu            [r2 + r3], xm5
+    vextracti128    xm5, m5, 1
+    movq            [r2 + r3 + 16], xm5
 %endif
-    add             r0, 16
-    movq            xm1, [r0]                       ; m1 = row 0
-    movq            xm2, [r0 + r1]                  ; m2 = row 1
+
+    movu            xm1, [r0 + r1]                  ; m1 = row 9
+    punpckhbw       xm2, xm3, xm1
+    punpcklbw       xm3, xm1
+    vinserti128     m3, m3, xm2, 1
+    pmaddubsw       m2, m3, [r5 + 1 * mmsize]
+    paddw           m6, m2
+    pmaddubsw       m3, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 10
+    punpckhbw       xm4, xm1, xm2
     punpcklbw       xm1, xm2
-    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
-    punpcklbw       xm2, xm3
-    vinserti128     m5, m1, xm2, 1
-    pmaddubsw       m5, m8
-    movq            xm4, [r0 + r4]                  ; m4 = row 3
-    punpcklbw       xm3, xm4
-    lea             r7, [r0 + r1 * 4]
-    movq            xm1, [r7]                       ; m1 = row 4
-    punpcklbw       xm4, xm1
-    vinserti128     m2, m3, xm4, 1
-    pmaddubsw       m0, m2, m9
-    paddw           m5, m0
-    pmaddubsw       m2, m8
-    movq            xm3, [r7 + r1]                  ; m3 = row 5
-    punpcklbw       xm1, xm3
-    movq            xm4, [r7 + r1 * 2]              ; m4 = row 6
-    punpcklbw       xm3, xm4
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m0, m1, m9
-    paddw           m2, m0
-    pmaddubsw       m1, m8
-    movq            xm3, [r7 + r4]                  ; m3 = row 7
-    punpcklbw       xm4, xm3
-    lea             r7, [r7 + r1 * 4]
-    movq            xm0, [r7]                       ; m0 = row 8
-    punpcklbw       xm3, xm0
-    vinserti128     m4, m4, xm3, 1
-    pmaddubsw       m3, m4, m9
-    paddw           m1, m3
-    pmaddubsw       m4, m8
-    movq            xm3, [r7 + r1]                  ; m3 = row 9
-    punpcklbw       xm0, xm3
-    movq            xm6, [r7 + r1 * 2]              ; m6 = row 10
-    punpcklbw       xm3, xm6
-    vinserti128     m0, m0, xm3, 1
-    pmaddubsw       m3, m0, m9
-    paddw           m4, m3
-    pmaddubsw       m0, m8
+    vinserti128     m1, m1, xm4, 1
+    pmaddubsw       m4, m1, [r5 + 1 * mmsize]
+    paddw           m0, m4
+    pmaddubsw       m1, [r5]
 
 %ifidn %1,pp
-    pmulhrsw        m5, m7                          ; m5 = word: row 0, row 1
-    pmulhrsw        m2, m7                          ; m2 = word: row 2, row 3
-    pmulhrsw        m1, m7                          ; m1 = word: row 4, row 5
-    pmulhrsw        m4, m7                          ; m4 = word: row 6, row 7
-    packuswb        m5, m2
-    packuswb        m1, m4
-    vextracti128    xm2, m5, 1
-    vextracti128    xm4, m1, 1
-    movq            [r2], xm5
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm5
-    movhps          [r2 + r6], xm2
-    lea             r8, [r2 + r3 * 4]
-    movq            [r8], xm1
-    movq            [r8 + r3], xm4
-    movhps          [r8 + r3 * 2], xm1
-    movhps          [r8 + r6], xm4
-%else
-    psubw           m5, m7                          ; m5 = word: row 0, row 1
-    psubw           m2, m7                          ; m2 = word: row 2, row 3
-    psubw           m1, m7                          ; m1 = word: row 4, row 5
-    psubw           m4, m7                          ; m4 = word: row 6, row 7
-    vextracti128    xm3, m5, 1
-    movu            [r2], xm5
-    movu            [r2 + r3], xm3
-    vextracti128    xm3, m2, 1
+    pmulhrsw        m6, m7                          ; m6 = word: row 6
+    pmulhrsw        m0, m7                          ; m0 = word: row 7
+    packuswb        m6, m0
+    vextracti128    xm0, m6, 1
+    movq            [r2 + r3 * 2], xm6
+    movd            [r2 + r3 * 2 + 8], xm0
+    movhps          [r2 + r6], xm6
+    pextrd          [r2 + r6 + 8], xm0, 2
+%else
+    psubw           m6, m7                          ; m6 = word: row 6
+    psubw           m0, m7                          ; m0 = word: row 7
+    movu            [r2 + r3 * 2], xm6
+    vextracti128    xm6, m6, 1
+    movq            [r2 + r3 * 2 + 16], xm6
+    movu            [r2 + r6], xm0
+    vextracti128    xm0, m0, 1
+    movq            [r2 + r6 + 16], xm0
+%endif
+    lea             r2, [r2 + r3 * 4]
+
+    movu            xm4, [r0 + r4]                  ; m4 = row 11
+    punpckhbw       xm6, xm2, xm4
+    punpcklbw       xm2, xm4
+    vinserti128     m2, m2, xm6, 1
+    pmaddubsw       m6, m2, [r5 + 1 * mmsize]
+    paddw           m3, m6
+    pmaddubsw       m2, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm6, [r0]                       ; m6 = row 12
+    punpckhbw       xm0, xm4, xm6
+    punpcklbw       xm4, xm6
+    vinserti128     m4, m4, xm0, 1
+    pmaddubsw       m0, m4, [r5 + 1 * mmsize]
+    paddw           m1, m0
+    pmaddubsw       m4, [r5]
+%ifidn %1,pp
+    pmulhrsw        m3, m7                          ; m3 = word: row 8
+    pmulhrsw        m1, m7                          ; m1 = word: row 9
+    packuswb        m3, m1
+    vextracti128    xm1, m3, 1
+    movq            [r2], xm3
+    movd            [r2 + 8], xm1
+    movhps          [r2 + r3], xm3
+    pextrd          [r2 + r3 + 8], xm1, 2
+%else
+    psubw           m3, m7                          ; m3 = word: row 8
+    psubw           m1, m7                          ; m1 = word: row 9
+    movu            [r2], xm3
+    vextracti128    xm3, m3, 1
+    movq            [r2 + 16], xm3
+    movu            [r2 + r3], xm1
+    vextracti128    xm1, m1, 1
+    movq            [r2 + r3 + 16], xm1
+%endif
+
+    movu            xm0, [r0 + r1]                  ; m0 = row 13
+    punpckhbw       xm1, xm6, xm0
+    punpcklbw       xm6, xm0
+    vinserti128     m6, m6, xm1, 1
+    pmaddubsw       m1, m6, [r5 + 1 * mmsize]
+    paddw           m2, m1
+    pmaddubsw       m6, [r5]
+    movu            xm1, [r0 + r1 * 2]              ; m1 = row 14
+    punpckhbw       xm5, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm5, 1
+    pmaddubsw       m5, m0, [r5 + 1 * mmsize]
+    paddw           m4, m5
+    pmaddubsw       m0, [r5]
+%ifidn %1,pp
+    pmulhrsw        m2, m7                          ; m2 = word: row 10
+    pmulhrsw        m4, m7                          ; m4 = word: row 11
+    packuswb        m2, m4
+    vextracti128    xm4, m2, 1
+    movq            [r2 + r3 * 2], xm2
+    movd            [r2 + r3 * 2 + 8], xm4
+    movhps          [r2 + r6], xm2
+    pextrd          [r2 + r6 + 8], xm4, 2
+%else
+    psubw           m2, m7                          ; m2 = word: row 10
+    psubw           m4, m7                          ; m4 = word: row 11
     movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-    vextracti128    xm3, m1, 1
-    lea             r8, [r2 + r3 * 4]
-    movu            [r8], xm1
-    movu            [r8 + r3], xm3
-    vextracti128    xm3, m4, 1
-    movu            [r8 + r3 * 2], xm4
-    movu            [r8 + r6], xm3
+    vextracti128    xm2, m2, 1
+    movq            [r2 + r3 * 2 + 16], xm2
+    movu            [r2 + r6], xm4
+    vextracti128    xm4, m4, 1
+    movq            [r2 + r6 + 16], xm4
 %endif
-    lea             r8, [r8 + r3 * 4]
+    lea             r2, [r2 + r3 * 4]
 
-    movq            xm3, [r7 + r4]                  ; m3 = row 11
-    punpcklbw       xm6, xm3
-    lea             r7, [r7 + r1 * 4]
-    movq            xm5, [r7]                       ; m5 = row 12
-    punpcklbw       xm3, xm5
-    vinserti128     m6, m6, xm3, 1
-    pmaddubsw       m3, m6, m9
-    paddw           m0, m3
-    pmaddubsw       m6, m8
-    movq            xm3, [r7 + r1]                  ; m3 = row 13
-    punpcklbw       xm5, xm3
-    movq            xm2, [r7 + r1 * 2]              ; m2 = row 14
-    punpcklbw       xm3, xm2
+    movu            xm5, [r0 + r4]                  ; m5 = row 15
+    punpckhbw       xm2, xm1, xm5
+    punpcklbw       xm1, xm5
+    vinserti128     m1, m1, xm2, 1
+    pmaddubsw       m2, m1, [r5 + 1 * mmsize]
+    paddw           m6, m2
+    pmaddubsw       m1, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm2, [r0]                       ; m2 = row 16
+    punpckhbw       xm3, xm5, xm2
+    punpcklbw       xm5, xm2
     vinserti128     m5, m5, xm3, 1
-    pmaddubsw       m3, m5, m9
-    paddw           m6, m3
-    pmaddubsw       m5, m8
-    movq            xm3, [r7 + r4]                  ; m3 = row 15
+    pmaddubsw       m3, m5, [r5 + 1 * mmsize]
+    paddw           m0, m3
+    pmaddubsw       m5, [r5]
+    movu            xm3, [r0 + r1]                  ; m3 = row 17
+    punpckhbw       xm4, xm2, xm3
     punpcklbw       xm2, xm3
-    lea             r7, [r7 + r1 * 4]
-    movq            xm1, [r7]                       ; m1 = row 16
-    punpcklbw       xm3, xm1
-    vinserti128     m2, m2, xm3, 1
-    pmaddubsw       m3, m2, m9
-    paddw           m5, m3
-    pmaddubsw       m2, m8
-    movq            xm3, [r7 + r1]                  ; m3 = row 17
-    punpcklbw       xm1, xm3
-    movq            xm4, [r7 + r1 * 2]              ; m4 = row 18
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m2, [r5 + 1 * mmsize]
+    paddw           m1, m2
+    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
+    punpckhbw       xm2, xm3, xm4
     punpcklbw       xm3, xm4
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m3, m1, m9
-    paddw           m2, m3
+    vinserti128     m3, m3, xm2, 1
+    pmaddubsw       m3, [r5 + 1 * mmsize]
+    paddw           m5, m3
+
 %ifidn %1,pp
-    pmulhrsw        m0, m7                          ; m0 = word: row 8, row 9
-    pmulhrsw        m6, m7                          ; m6 = word: row 10, row 11
-    pmulhrsw        m5, m7                          ; m5 = word: row 12, row 13
-    pmulhrsw        m2, m7                          ; m2 = word: row 14, row 15
-    packuswb        m0, m6
-    packuswb        m5, m2
-    vextracti128    xm6, m0, 1
-    vextracti128    xm2, m5, 1
-    movq            [r8], xm0
-    movq            [r8 + r3], xm6
-    movhps          [r8 + r3 * 2], xm0
-    movhps          [r8 + r6], xm6
-    lea             r8, [r8 + r3 * 4]
-    movq            [r8], xm5
-    movq            [r8 + r3], xm2
-    movhps          [r8 + r3 * 2], xm5
-    movhps          [r8 + r6], xm2
-    lea             r2, [r8 + r3 * 4 - 16]
+    pmulhrsw        m6, m7                          ; m6 = word: row 12
+    pmulhrsw        m0, m7                          ; m0 = word: row 13
+    pmulhrsw        m1, m7                          ; m1 = word: row 14
+    pmulhrsw        m5, m7                          ; m5 = word: row 15
+    packuswb        m6, m0
+    packuswb        m1, m5
+    vextracti128    xm0, m6, 1
+    vextracti128    xm5, m1, 1
+    movq            [r2], xm6
+    movd            [r2 + 8], xm0
+    movhps          [r2 + r3], xm6
+    pextrd          [r2 + r3 + 8], xm0, 2
+    movq            [r2 + r3 * 2], xm1
+    movd            [r2 + r3 * 2 + 8], xm5
+    movhps          [r2 + r6], xm1
+    pextrd          [r2 + r6 + 8], xm5, 2
 %else
-    psubw           m0, m7                          ; m0 = word: row 8, row 9
-    psubw           m6, m7                          ; m6 = word: row 10, row 11
-    psubw           m5, m7                          ; m5 = word: row 12, row 13
-    psubw           m2, m7                          ; m2 = word: row 14, row 15
-    vextracti128    xm3, m0, 1
-    movu            [r8], xm0
-    movu            [r8 + r3], xm3
-    vextracti128    xm3, m6, 1
-    movu            [r8 + r3 * 2], xm6
-    movu            [r8 + r6], xm3
-    vextracti128    xm3, m5, 1
-    lea             r8, [r8 + r3 * 4]
-    movu            [r8], xm5
-    movu            [r8 + r3], xm3
-    vextracti128    xm3, m2, 1
-    movu            [r8 + r3 * 2], xm2
-    movu            [r8 + r6], xm3
-    lea             r2, [r8 + r3 * 4 - 32]
+    psubw           m6, m7                          ; m6 = word: row 12
+    psubw           m0, m7                          ; m0 = word: row 13
+    psubw           m1, m7                          ; m1 = word: row 14
+    psubw           m5, m7                          ; m5 = word: row 15
+    movu            [r2], xm6
+    vextracti128    xm6, m6, 1
+    movq            [r2 + 16], xm6
+    movu            [r2 + r3], xm0
+    vextracti128    xm0, m0, 1
+    movq            [r2 + r3 + 16], xm0
+    movu            [r2 + r3 * 2], xm1
+    vextracti128    xm1, m1, 1
+    movq            [r2 + r3 * 2 + 16], xm1
+    movu            [r2 + r6], xm5
+    vextracti128    xm5, m5, 1
+    movq            [r2 + r6 + 16], xm5
 %endif
-    lea             r0, [r7 - 16]
-    dec             r5d
-    jnz             .loopH
+    lea             r2, [r2 + r3 * 4]
+%endrep
     RET
-%endif
 %endmacro
 
-    FILTER_VER_CHROMA_AVX2_24x32 pp
-    FILTER_VER_CHROMA_AVX2_24x32 ps
+    FILTER_VER_CHROMA_AVX2_12xN pp, 16
+    FILTER_VER_CHROMA_AVX2_12xN ps, 16
+    FILTER_VER_CHROMA_AVX2_12xN pp, 32
+    FILTER_VER_CHROMA_AVX2_12xN ps, 32
 
-%macro FILTER_VER_CHROMA_AVX2_24x64 1
-%if ARCH_X86_64 == 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_24x64, 4, 7, 13
-    mov             r4d, r4m
-    shl             r4d, 6
+;-----------------------------------------------------------------------------
+;void interp_4tap_vert_pp_24x32(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W24 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_pp_24x%2, 4, 6, 8
+
+    mov         r4d,       r4m
+    sub         r0,        r1
 
 %ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
+    lea         r5,        [tab_ChromaCoeff]
+    movd        m0,        [r5 + r4 * 4]
+%else
+    movd        m0,        [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb      m1,        m0,       [tab_Vm]
+    pshufb      m0,        [tab_Vm + 16]
+
+    mov         r4d,       %2
+
+.loop:
+    movu        m2,        [r0]
+    movu        m3,        [r0 + r1]
+
+    punpcklbw   m4,        m2,        m3
+    punpckhbw   m2,        m3
+
+    pmaddubsw   m4,        m1
+    pmaddubsw   m2,        m1
+
+    lea         r5,        [r0 + 2 * r1]
+    movu        m5,        [r5]
+    movu        m7,        [r5 + r1]
+
+    punpcklbw   m6,        m5,        m7
+    pmaddubsw   m6,        m0
+    paddw       m4,        m6
+
+    punpckhbw   m6,        m5,        m7
+    pmaddubsw   m6,        m0
+    paddw       m2,        m6
+
+    mova        m6,        [pw_512]
+
+    pmulhrsw    m4,        m6
+    pmulhrsw    m2,        m6
+
+    packuswb    m4,        m2
+
+    movu        [r2],      m4
+
+    punpcklbw   m4,        m3,        m5
+    punpckhbw   m3,        m5
+
+    pmaddubsw   m4,        m1
+    pmaddubsw   m3,        m1
+
+    movu        m2,        [r5 + 2 * r1]
+
+    punpcklbw   m5,        m7,        m2
+    punpckhbw   m7,        m2
+
+    pmaddubsw   m5,        m0
+    pmaddubsw   m7,        m0
+
+    paddw       m4,        m5
+    paddw       m3,        m7
+
+    pmulhrsw    m4,        m6
+    pmulhrsw    m3,        m6
+
+    packuswb    m4,        m3
+
+    movu        [r2 + r3],      m4
+
+    movq        m2,        [r0 + 16]
+    movq        m3,        [r0 + r1 + 16]
+    movq        m4,        [r5 + 16]
+    movq        m5,        [r5 + r1 + 16]
+
+    punpcklbw   m2,        m3
+    punpcklbw   m4,        m5
+
+    pmaddubsw   m2,        m1
+    pmaddubsw   m4,        m0
+
+    paddw       m2,        m4
+
+    pmulhrsw    m2,        m6
+
+    movq        m3,        [r0 + r1 + 16]
+    movq        m4,        [r5 + 16]
+    movq        m5,        [r5 + r1 + 16]
+    movq        m7,        [r5 + 2 * r1 + 16]
+
+    punpcklbw   m3,        m4
+    punpcklbw   m5,        m7
+
+    pmaddubsw   m3,        m1
+    pmaddubsw   m5,        m0
+
+    paddw       m3,        m5
+
+    pmulhrsw    m3,        m6
+    packuswb    m2,        m3
+
+    movh        [r2 + 16], m2
+    movhps      [r2 + r3 + 16], m2
+
+    mov         r0,        r5
+    lea         r2,        [r2 + 2 * r3]
+
+    sub         r4,        2
+    jnz        .loop
+    RET
+%endmacro
+
+    FILTER_V4_W24 24, 32
+
+    FILTER_V4_W24 24, 64
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_pp_32x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W32 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_pp_%1x%2, 4, 6, 8
+
+    mov         r4d,       r4m
+    sub         r0,        r1
+
+%ifdef PIC
+    lea         r5,        [tab_ChromaCoeff]
+    movd        m0,        [r5 + r4 * 4]
+%else
+    movd        m0,        [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb      m1,        m0,       [tab_Vm]
+    pshufb      m0,        [tab_Vm + 16]
+
+    mova        m7,        [pw_512]
+
+    mov         r4d,       %2
+
+.loop:
+    movu        m2,        [r0]
+    movu        m3,        [r0 + r1]
+
+    punpcklbw   m4,        m2,        m3
+    punpckhbw   m2,        m3
+
+    pmaddubsw   m4,        m1
+    pmaddubsw   m2,        m1
+
+    lea         r5,        [r0 + 2 * r1]
+    movu        m3,        [r5]
+    movu        m5,        [r5 + r1]
+
+    punpcklbw   m6,        m3,        m5
+    punpckhbw   m3,        m5
+
+    pmaddubsw   m6,        m0
+    pmaddubsw   m3,        m0
+
+    paddw       m4,        m6
+    paddw       m2,        m3
+
+    pmulhrsw    m4,        m7
+    pmulhrsw    m2,        m7
+
+    packuswb    m4,        m2
+
+    movu        [r2],      m4
+
+    movu        m2,        [r0 + 16]
+    movu        m3,        [r0 + r1 + 16]
+
+    punpcklbw   m4,        m2,        m3
+    punpckhbw   m2,        m3
+
+    pmaddubsw   m4,        m1
+    pmaddubsw   m2,        m1
+
+    movu        m3,        [r5 + 16]
+    movu        m5,        [r5 + r1 + 16]
+
+    punpcklbw   m6,        m3,        m5
+    punpckhbw   m3,        m5
+
+    pmaddubsw   m6,        m0
+    pmaddubsw   m3,        m0
+
+    paddw       m4,        m6
+    paddw       m2,        m3
+
+    pmulhrsw    m4,        m7
+    pmulhrsw    m2,        m7
+
+    packuswb    m4,        m2
+
+    movu        [r2 + 16], m4
+
+    lea         r0,        [r0 + r1]
+    lea         r2,        [r2 + r3]
+
+    dec         r4
+    jnz        .loop
+    RET
+%endmacro
+
+    FILTER_V4_W32 32,  8
+    FILTER_V4_W32 32, 16
+    FILTER_V4_W32 32, 24
+    FILTER_V4_W32 32, 32
+
+    FILTER_V4_W32 32, 48
+    FILTER_V4_W32 32, 64
+
+%macro FILTER_VER_CHROMA_AVX2_32xN 2
+%if ARCH_X86_64 == 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_32x%2, 4, 7, 13
+    mov             r4d, r4m
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
 %else
     lea             r5, [tab_ChromaCoeffVer_32 + r4]
 %endif
@@ -10737,8 +8733,8 @@ cglobal interp_4tap_vert_%1_24x64, 4, 7, 13
     vbroadcasti128  m12, [pw_2000]
 %endif
     lea             r5, [r3 * 3]
-    mov             r6d, 16
-.loopH:
+    mov             r6d, %2 / 4
+.loopW:
     movu            m0, [r0]                        ; m0 = row 0
     movu            m1, [r0 + r1]                   ; m1 = row 1
     punpcklbw       m2, m0, m1
@@ -10763,16 +8759,14 @@ cglobal interp_4tap_vert_%1_24x64, 4, 7, 13
     pmulhrsw        m2, m12
     pmulhrsw        m3, m12
     packuswb        m2, m3
-    movu            [r2], xm2
-    vextracti128    xm2, m2, 1
-    movq            [r2 + 16], xm2
+    movu            [r2], m2
 %else
     psubw           m2, m12
     psubw           m3, m12
     vperm2i128      m0, m2, m3, 0x20
     vperm2i128      m2, m2, m3, 0x31
     movu            [r2], m0
-    movu            [r2 + mmsize], xm2
+    movu            [r2 + mmsize], m2
 %endif
     lea             r0, [r0 + r1 * 4]
     movu            m0, [r0]                        ; m0 = row 4
@@ -10788,16 +8782,14 @@ cglobal interp_4tap_vert_%1_24x64, 4, 7, 13
     pmulhrsw        m4, m12
     pmulhrsw        m5, m12
     packuswb        m4, m5
-    movu            [r2 + r3], xm4
-    vextracti128    xm4, m4, 1
-    movq            [r2 + r3 + 16], xm4
+    movu            [r2 + r3], m4
 %else
     psubw           m4, m12
     psubw           m5, m12
     vperm2i128      m1, m4, m5, 0x20
     vperm2i128      m4, m4, m5, 0x31
     movu            [r2 + r3], m1
-    movu            [r2 + r3 + mmsize], xm4
+    movu            [r2 + r3 + mmsize], m4
 %endif
 
     movu            m1, [r0 + r1]                   ; m1 = row 5
@@ -10811,16 +8803,14 @@ cglobal interp_4tap_vert_%1_24x64, 4, 7, 13
     pmulhrsw        m6, m12
     pmulhrsw        m7, m12
     packuswb        m6, m7
-    movu            [r2 + r3 * 2], xm6
-    vextracti128    xm6, m6, 1
-    movq            [r2 + r3 * 2 + 16], xm6
+    movu            [r2 + r3 * 2], m6
 %else
     psubw           m6, m12
     psubw           m7, m12
     vperm2i128      m0, m6, m7, 0x20
     vperm2i128      m6, m6, m7, 0x31
     movu            [r2 + r3 * 2], m0
-    movu            [r2 + r3 * 2 + mmsize], xm6
+    movu            [r2 + r3 * 2 + mmsize], m6
 %endif
 
     movu            m0, [r0 + r1 * 2]               ; m0 = row 6
@@ -10834,32 +8824,41 @@ cglobal interp_4tap_vert_%1_24x64, 4, 7, 13
     pmulhrsw        m2, m12
     pmulhrsw        m3, m12
     packuswb        m2, m3
-    movu            [r2 + r5], xm2
-    vextracti128    xm2, m2, 1
-    movq            [r2 + r5 + 16], xm2
+    movu            [r2 + r5], m2
 %else
     psubw           m2, m12
     psubw           m3, m12
     vperm2i128      m0, m2, m3, 0x20
     vperm2i128      m2, m2, m3, 0x31
     movu            [r2 + r5], m0
-    movu            [r2 + r5 + mmsize], xm2
+    movu            [r2 + r5 + mmsize], m2
 %endif
     lea             r2, [r2 + r3 * 4]
     dec             r6d
-    jnz             .loopH
+    jnz             .loopW
     RET
 %endif
 %endmacro
 
-    FILTER_VER_CHROMA_AVX2_24x64 pp
-    FILTER_VER_CHROMA_AVX2_24x64 ps
-
-%macro FILTER_VER_CHROMA_AVX2_16x4 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_16x4, 4, 6, 8
-    mov             r4d, r4m
-    shl             r4d, 6
+    FILTER_VER_CHROMA_AVX2_32xN pp, 64
+    FILTER_VER_CHROMA_AVX2_32xN pp, 48
+    FILTER_VER_CHROMA_AVX2_32xN pp, 32
+    FILTER_VER_CHROMA_AVX2_32xN pp, 24
+    FILTER_VER_CHROMA_AVX2_32xN pp, 16
+    FILTER_VER_CHROMA_AVX2_32xN pp, 8
+    FILTER_VER_CHROMA_AVX2_32xN ps, 64
+    FILTER_VER_CHROMA_AVX2_32xN ps, 48
+    FILTER_VER_CHROMA_AVX2_32xN ps, 32
+    FILTER_VER_CHROMA_AVX2_32xN ps, 24
+    FILTER_VER_CHROMA_AVX2_32xN ps, 16
+    FILTER_VER_CHROMA_AVX2_32xN ps, 8
+
+%macro FILTER_VER_CHROMA_AVX2_48x64 1
+%if ARCH_X86_64 == 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_48x64, 4, 8, 13
+    mov             r4d, r4m
+    shl             r4d, 6
 
 %ifdef PIC
     lea             r5, [tab_ChromaCoeffVer_32]
@@ -10868,94 +8867,200 @@ cglobal interp_4tap_vert_%1_16x4, 4, 6, 8
     lea             r5, [tab_ChromaCoeffVer_32 + r4]
 %endif
 
+    mova            m10, [r5]
+    mova            m11, [r5 + mmsize]
     lea             r4, [r1 * 3]
     sub             r0, r1
 %ifidn %1,pp
-    mova            m7, [pw_512]
+    mova            m12, [pw_512]
 %else
     add             r3d, r3d
-    mova            m7, [pw_2000]
+    vbroadcasti128  m12, [pw_2000]
 %endif
-
-    movu            xm0, [r0]
-    vinserti128     m0, m0, [r0 + r1 * 2], 1
-    movu            xm1, [r0 + r1]
-    vinserti128     m1, m1, [r0 + r4], 1
-
+    lea             r5, [r3 * 3]
+    lea             r7, [r1 * 4]
+    mov             r6d, 16
+.loopH:
+    movu            m0, [r0]                        ; m0 = row 0
+    movu            m1, [r0 + r1]                   ; m1 = row 1
     punpcklbw       m2, m0, m1
     punpckhbw       m3, m0, m1
-    vperm2i128      m4, m2, m3, 0x20
+    pmaddubsw       m2, m10
+    pmaddubsw       m3, m10
+    movu            m0, [r0 + r1 * 2]               ; m0 = row 2
+    punpcklbw       m4, m1, m0
+    punpckhbw       m5, m1, m0
+    pmaddubsw       m4, m10
+    pmaddubsw       m5, m10
+    movu            m1, [r0 + r4]                   ; m1 = row 3
+    punpcklbw       m6, m0, m1
+    punpckhbw       m7, m0, m1
+    pmaddubsw       m8, m6, m11
+    pmaddubsw       m9, m7, m11
+    pmaddubsw       m6, m10
+    pmaddubsw       m7, m10
+    paddw           m2, m8
+    paddw           m3, m9
+%ifidn %1,pp
+    pmulhrsw        m2, m12
+    pmulhrsw        m3, m12
+    packuswb        m2, m3
+    movu            [r2], m2
+%else
+    psubw           m2, m12
+    psubw           m3, m12
+    vperm2i128      m0, m2, m3, 0x20
     vperm2i128      m2, m2, m3, 0x31
-    pmaddubsw       m4, [r5]
-    pmaddubsw       m3, m2, [r5 + mmsize]
-    paddw           m4, m3
-    pmaddubsw       m2, [r5]
-
-    vextracti128    xm0, m0, 1
+    movu            [r2], m0
+    movu            [r2 + mmsize], m2
+%endif
     lea             r0, [r0 + r1 * 4]
-    vinserti128     m0, m0, [r0], 1
-
-    punpcklbw       m5, m1, m0
+    movu            m0, [r0]                        ; m0 = row 4
+    punpcklbw       m2, m1, m0
     punpckhbw       m3, m1, m0
-    vperm2i128      m6, m5, m3, 0x20
-    vperm2i128      m5, m5, m3, 0x31
-    pmaddubsw       m6, [r5]
-    pmaddubsw       m3, m5, [r5 + mmsize]
-    paddw           m6, m3
-    pmaddubsw       m5, [r5]
+    pmaddubsw       m8, m2, m11
+    pmaddubsw       m9, m3, m11
+    pmaddubsw       m2, m10
+    pmaddubsw       m3, m10
+    paddw           m4, m8
+    paddw           m5, m9
 %ifidn %1,pp
-    pmulhrsw        m4, m7                          ; m4 = word: row 0
-    pmulhrsw        m6, m7                          ; m6 = word: row 1
-    packuswb        m4, m6
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm6, m4, 1
-    movu            [r2], xm4
-    movu            [r2 + r3], xm6
+    pmulhrsw        m4, m12
+    pmulhrsw        m5, m12
+    packuswb        m4, m5
+    movu            [r2 + r3], m4
 %else
-    psubw           m4, m7                          ; m4 = word: row 0
-    psubw           m6, m7                          ; m6 = word: row 1
-    movu            [r2], m4
-    movu            [r2 + r3], m6
+    psubw           m4, m12
+    psubw           m5, m12
+    vperm2i128      m1, m4, m5, 0x20
+    vperm2i128      m4, m4, m5, 0x31
+    movu            [r2 + r3], m1
+    movu            [r2 + r3 + mmsize], m4
 %endif
-    lea             r2, [r2 + r3 * 2]
 
-    movu            xm4, [r0 + r1 * 2]
-    vinserti128     m4, m4, [r0 + r1], 1
-    vextracti128    xm1, m4, 1
-    vinserti128     m0, m0, xm1, 0
+    movu            m1, [r0 + r1]                   ; m1 = row 5
+    punpcklbw       m4, m0, m1
+    punpckhbw       m5, m0, m1
+    pmaddubsw       m4, m11
+    pmaddubsw       m5, m11
+    paddw           m6, m4
+    paddw           m7, m5
+%ifidn %1,pp
+    pmulhrsw        m6, m12
+    pmulhrsw        m7, m12
+    packuswb        m6, m7
+    movu            [r2 + r3 * 2], m6
+%else
+    psubw           m6, m12
+    psubw           m7, m12
+    vperm2i128      m0, m6, m7, 0x20
+    vperm2i128      m6, m6, m7, 0x31
+    movu            [r2 + r3 * 2], m0
+    movu            [r2 + r3 * 2 + mmsize], m6
+%endif
 
-    punpcklbw       m6, m0, m4
-    punpckhbw       m1, m0, m4
-    vperm2i128      m0, m6, m1, 0x20
-    vperm2i128      m6, m6, m1, 0x31
-    pmaddubsw       m0, [r5 + mmsize]
-    paddw           m5, m0
-    pmaddubsw       m6, [r5 + mmsize]
+    movu            m0, [r0 + r1 * 2]               ; m0 = row 6
+    punpcklbw       m6, m1, m0
+    punpckhbw       m7, m1, m0
+    pmaddubsw       m6, m11
+    pmaddubsw       m7, m11
     paddw           m2, m6
+    paddw           m3, m7
+%ifidn %1,pp
+    pmulhrsw        m2, m12
+    pmulhrsw        m3, m12
+    packuswb        m2, m3
+    movu            [r2 + r5], m2
+    add             r2, 32
+%else
+    psubw           m2, m12
+    psubw           m3, m12
+    vperm2i128      m0, m2, m3, 0x20
+    vperm2i128      m2, m2, m3, 0x31
+    movu            [r2 + r5], m0
+    movu            [r2 + r5 + mmsize], m2
+    add             r2, 64
+%endif
+    sub             r0, r7
 
+    movu            xm0, [r0 + 32]                  ; m0 = row 0
+    movu            xm1, [r0 + r1 + 32]             ; m1 = row 1
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddubsw       m0, m10
+    movu            xm2, [r0 + r1 * 2 + 32]         ; m2 = row 2
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m1, m10
+    movu            xm3, [r0 + r4 + 32]             ; m3 = row 3
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m4, m2, m11
+    paddw           m0, m4
+    pmaddubsw       m2, m10
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0 + 32]                  ; m4 = row 4
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddubsw       m5, m3, m11
+    paddw           m1, m5
+    pmaddubsw       m3, m10
+    movu            xm5, [r0 + r1 + 32]             ; m5 = row 5
+    punpckhbw       xm6, xm4, xm5
+    punpcklbw       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddubsw       m4, m11
+    paddw           m2, m4
+    movu            xm6, [r0 + r1 * 2 + 32]         ; m6 = row 6
+    punpckhbw       xm7, xm5, xm6
+    punpcklbw       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddubsw       m5, m11
+    paddw           m3, m5
 %ifidn %1,pp
-    pmulhrsw        m2, m7                          ; m2 = word: row 2
-    pmulhrsw        m5, m7                          ; m5 = word: row 3
-    packuswb        m2, m5
+    pmulhrsw        m0, m12                         ; m0 = word: row 0
+    pmulhrsw        m1, m12                         ; m1 = word: row 1
+    pmulhrsw        m2, m12                         ; m2 = word: row 2
+    pmulhrsw        m3, m12                         ; m3 = word: row 3
+    packuswb        m0, m1
+    packuswb        m2, m3
+    vpermq          m0, m0, 11011000b
     vpermq          m2, m2, 11011000b
-    vextracti128    xm5, m2, 1
-    movu            [r2], xm2
-    movu            [r2 + r3], xm5
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m2, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r5], xm3
+    lea             r2, [r2 + r3 * 4 - 32]
 %else
-    psubw           m2, m7                          ; m2 = word: row 2
-    psubw           m5, m7                          ; m5 = word: row 3
-    movu            [r2], m2
-    movu            [r2 + r3], m5
+    psubw           m0, m12                         ; m0 = word: row 0
+    psubw           m1, m12                         ; m1 = word: row 1
+    psubw           m2, m12                         ; m2 = word: row 2
+    psubw           m3, m12                         ; m3 = word: row 3
+    movu            [r2], m0
+    movu            [r2 + r3], m1
+    movu            [r2 + r3 * 2], m2
+    movu            [r2 + r5], m3
+    lea             r2, [r2 + r3 * 4 - 64]
 %endif
+    dec             r6d
+    jnz             .loopH
     RET
+%endif
 %endmacro
 
-    FILTER_VER_CHROMA_AVX2_16x4 pp
-    FILTER_VER_CHROMA_AVX2_16x4 ps
+    FILTER_VER_CHROMA_AVX2_48x64 pp
+    FILTER_VER_CHROMA_AVX2_48x64 ps
 
-%macro FILTER_VER_CHROMA_AVX2_12xN 2
+%macro FILTER_VER_CHROMA_AVX2_64xN 2
+%if ARCH_X86_64 == 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_12x%2, 4, 7, 8
+cglobal interp_4tap_vert_%1_64x%2, 4, 8, 13
     mov             r4d, r4m
     shl             r4d, 6
 
@@ -10966,317 +9071,152 @@ cglobal interp_4tap_vert_%1_12x%2, 4, 7, 8
     lea             r5, [tab_ChromaCoeffVer_32 + r4]
 %endif
 
+    mova            m10, [r5]
+    mova            m11, [r5 + mmsize]
     lea             r4, [r1 * 3]
     sub             r0, r1
 %ifidn %1,pp
-    mova            m7, [pw_512]
+    mova            m12, [pw_512]
 %else
     add             r3d, r3d
-    vbroadcasti128  m7, [pw_2000]
+    vbroadcasti128  m12, [pw_2000]
 %endif
-    lea             r6, [r3 * 3]
-%rep %2 / 16
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
-    paddw           m0, m4
-    pmaddubsw       m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhbw       xm5, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
-    paddw           m1, m5
-    pmaddubsw       m3, [r5]
-%ifidn %1,pp
-    pmulhrsw        m0, m7                          ; m0 = word: row 0
-    pmulhrsw        m1, m7                          ; m1 = word: row 1
-    packuswb        m0, m1
-    vextracti128    xm1, m0, 1
-    movq            [r2], xm0
-    movd            [r2 + 8], xm1
-    movhps          [r2 + r3], xm0
-    pextrd          [r2 + r3 + 8], xm1, 2
-%else
-    psubw           m0, m7                          ; m0 = word: row 0
-    psubw           m1, m7                          ; m1 = word: row 1
-    movu            [r2], xm0
-    vextracti128    xm0, m0, 1
-    movq            [r2 + 16], xm0
-    movu            [r2 + r3], xm1
-    vextracti128    xm1, m1, 1
-    movq            [r2 + r3 + 16], xm1
-%endif
-
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhbw       xm6, xm4, xm5
-    punpcklbw       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
-    paddw           m2, m6
-    pmaddubsw       m4, [r5]
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhbw       xm0, xm5, xm6
-    punpcklbw       xm5, xm6
-    vinserti128     m5, m5, xm0, 1
-    pmaddubsw       m0, m5, [r5 + 1 * mmsize]
-    paddw           m3, m0
-    pmaddubsw       m5, [r5]
+    lea             r5, [r3 * 3]
+    lea             r7, [r1 * 4]
+    mov             r6d, %2 / 4
+.loopH:
+%assign x 0
+%rep 2
+    movu            m0, [r0 + x]                    ; m0 = row 0
+    movu            m1, [r0 + r1 + x]               ; m1 = row 1
+    punpcklbw       m2, m0, m1
+    punpckhbw       m3, m0, m1
+    pmaddubsw       m2, m10
+    pmaddubsw       m3, m10
+    movu            m0, [r0 + r1 * 2 + x]           ; m0 = row 2
+    punpcklbw       m4, m1, m0
+    punpckhbw       m5, m1, m0
+    pmaddubsw       m4, m10
+    pmaddubsw       m5, m10
+    movu            m1, [r0 + r4 + x]               ; m1 = row 3
+    punpcklbw       m6, m0, m1
+    punpckhbw       m7, m0, m1
+    pmaddubsw       m8, m6, m11
+    pmaddubsw       m9, m7, m11
+    pmaddubsw       m6, m10
+    pmaddubsw       m7, m10
+    paddw           m2, m8
+    paddw           m3, m9
 %ifidn %1,pp
-    pmulhrsw        m2, m7                          ; m2 = word: row 2
-    pmulhrsw        m3, m7                          ; m3 = word: row 3
+    pmulhrsw        m2, m12
+    pmulhrsw        m3, m12
     packuswb        m2, m3
-    vextracti128    xm3, m2, 1
-    movq            [r2 + r3 * 2], xm2
-    movd            [r2 + r3 * 2 + 8], xm3
-    movhps          [r2 + r6], xm2
-    pextrd          [r2 + r6 + 8], xm3, 2
+    movu            [r2], m2
 %else
-    psubw           m2, m7                          ; m2 = word: row 2
-    psubw           m3, m7                          ; m3 = word: row 3
-    movu            [r2 + r3 * 2], xm2
-    vextracti128    xm2, m2, 1
-    movq            [r2 + r3 * 2 + 16], xm2
-    movu            [r2 + r6], xm3
-    vextracti128    xm3, m3, 1
-    movq            [r2 + r6 + 16], xm3
+    psubw           m2, m12
+    psubw           m3, m12
+    vperm2i128      m0, m2, m3, 0x20
+    vperm2i128      m2, m2, m3, 0x31
+    movu            [r2], m0
+    movu            [r2 + mmsize], m2
 %endif
-    lea             r2, [r2 + r3 * 4]
-
-    movu            xm0, [r0 + r4]                  ; m0 = row 7
-    punpckhbw       xm3, xm6, xm0
-    punpcklbw       xm6, xm0
-    vinserti128     m6, m6, xm3, 1
-    pmaddubsw       m3, m6, [r5 + 1 * mmsize]
-    paddw           m4, m3
-    pmaddubsw       m6, [r5]
     lea             r0, [r0 + r1 * 4]
-    movu            xm3, [r0]                       ; m3 = row 8
-    punpckhbw       xm1, xm0, xm3
-    punpcklbw       xm0, xm3
-    vinserti128     m0, m0, xm1, 1
-    pmaddubsw       m1, m0, [r5 + 1 * mmsize]
-    paddw           m5, m1
-    pmaddubsw       m0, [r5]
+    movu            m0, [r0 + x]                    ; m0 = row 4
+    punpcklbw       m2, m1, m0
+    punpckhbw       m3, m1, m0
+    pmaddubsw       m8, m2, m11
+    pmaddubsw       m9, m3, m11
+    pmaddubsw       m2, m10
+    pmaddubsw       m3, m10
+    paddw           m4, m8
+    paddw           m5, m9
 %ifidn %1,pp
-    pmulhrsw        m4, m7                          ; m4 = word: row 4
-    pmulhrsw        m5, m7                          ; m5 = word: row 5
+    pmulhrsw        m4, m12
+    pmulhrsw        m5, m12
     packuswb        m4, m5
-    vextracti128    xm5, m4, 1
-    movq            [r2], xm4
-    movd            [r2 + 8], xm5
-    movhps          [r2 + r3], xm4
-    pextrd          [r2 + r3 + 8], xm5, 2
+    movu            [r2 + r3], m4
 %else
-    psubw           m4, m7                          ; m4 = word: row 4
-    psubw           m5, m7                          ; m5 = word: row 5
-    movu            [r2], xm4
-    vextracti128    xm4, m4, 1
-    movq            [r2 + 16], xm4
-    movu            [r2 + r3], xm5
-    vextracti128    xm5, m5, 1
-    movq            [r2 + r3 + 16], xm5
+    psubw           m4, m12
+    psubw           m5, m12
+    vperm2i128      m1, m4, m5, 0x20
+    vperm2i128      m4, m4, m5, 0x31
+    movu            [r2 + r3], m1
+    movu            [r2 + r3 + mmsize], m4
 %endif
 
-    movu            xm1, [r0 + r1]                  ; m1 = row 9
-    punpckhbw       xm2, xm3, xm1
-    punpcklbw       xm3, xm1
-    vinserti128     m3, m3, xm2, 1
-    pmaddubsw       m2, m3, [r5 + 1 * mmsize]
-    paddw           m6, m2
-    pmaddubsw       m3, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 10
-    punpckhbw       xm4, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm4, 1
-    pmaddubsw       m4, m1, [r5 + 1 * mmsize]
-    paddw           m0, m4
-    pmaddubsw       m1, [r5]
-
+    movu            m1, [r0 + r1 + x]               ; m1 = row 5
+    punpcklbw       m4, m0, m1
+    punpckhbw       m5, m0, m1
+    pmaddubsw       m4, m11
+    pmaddubsw       m5, m11
+    paddw           m6, m4
+    paddw           m7, m5
 %ifidn %1,pp
-    pmulhrsw        m6, m7                          ; m6 = word: row 6
-    pmulhrsw        m0, m7                          ; m0 = word: row 7
-    packuswb        m6, m0
-    vextracti128    xm0, m6, 1
-    movq            [r2 + r3 * 2], xm6
-    movd            [r2 + r3 * 2 + 8], xm0
-    movhps          [r2 + r6], xm6
-    pextrd          [r2 + r6 + 8], xm0, 2
+    pmulhrsw        m6, m12
+    pmulhrsw        m7, m12
+    packuswb        m6, m7
+    movu            [r2 + r3 * 2], m6
 %else
-    psubw           m6, m7                          ; m6 = word: row 6
-    psubw           m0, m7                          ; m0 = word: row 7
-    movu            [r2 + r3 * 2], xm6
-    vextracti128    xm6, m6, 1
-    movq            [r2 + r3 * 2 + 16], xm6
-    movu            [r2 + r6], xm0
-    vextracti128    xm0, m0, 1
-    movq            [r2 + r6 + 16], xm0
+    psubw           m6, m12
+    psubw           m7, m12
+    vperm2i128      m0, m6, m7, 0x20
+    vperm2i128      m6, m6, m7, 0x31
+    movu            [r2 + r3 * 2], m0
+    movu            [r2 + r3 * 2 + mmsize], m6
 %endif
-    lea             r2, [r2 + r3 * 4]
 
-    movu            xm4, [r0 + r4]                  ; m4 = row 11
-    punpckhbw       xm6, xm2, xm4
-    punpcklbw       xm2, xm4
-    vinserti128     m2, m2, xm6, 1
-    pmaddubsw       m6, m2, [r5 + 1 * mmsize]
-    paddw           m3, m6
-    pmaddubsw       m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm6, [r0]                       ; m6 = row 12
-    punpckhbw       xm0, xm4, xm6
-    punpcklbw       xm4, xm6
-    vinserti128     m4, m4, xm0, 1
-    pmaddubsw       m0, m4, [r5 + 1 * mmsize]
-    paddw           m1, m0
-    pmaddubsw       m4, [r5]
+    movu            m0, [r0 + r1 * 2 + x]           ; m0 = row 6
+    punpcklbw       m6, m1, m0
+    punpckhbw       m7, m1, m0
+    pmaddubsw       m6, m11
+    pmaddubsw       m7, m11
+    paddw           m2, m6
+    paddw           m3, m7
 %ifidn %1,pp
-    pmulhrsw        m3, m7                          ; m3 = word: row 8
-    pmulhrsw        m1, m7                          ; m1 = word: row 9
-    packuswb        m3, m1
-    vextracti128    xm1, m3, 1
-    movq            [r2], xm3
-    movd            [r2 + 8], xm1
-    movhps          [r2 + r3], xm3
-    pextrd          [r2 + r3 + 8], xm1, 2
+    pmulhrsw        m2, m12
+    pmulhrsw        m3, m12
+    packuswb        m2, m3
+    movu            [r2 + r5], m2
+    add             r2, 32
 %else
-    psubw           m3, m7                          ; m3 = word: row 8
-    psubw           m1, m7                          ; m1 = word: row 9
-    movu            [r2], xm3
-    vextracti128    xm3, m3, 1
-    movq            [r2 + 16], xm3
-    movu            [r2 + r3], xm1
-    vextracti128    xm1, m1, 1
-    movq            [r2 + r3 + 16], xm1
-%endif
-
-    movu            xm0, [r0 + r1]                  ; m0 = row 13
-    punpckhbw       xm1, xm6, xm0
-    punpcklbw       xm6, xm0
-    vinserti128     m6, m6, xm1, 1
-    pmaddubsw       m1, m6, [r5 + 1 * mmsize]
-    paddw           m2, m1
-    pmaddubsw       m6, [r5]
-    movu            xm1, [r0 + r1 * 2]              ; m1 = row 14
-    punpckhbw       xm5, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm5, 1
-    pmaddubsw       m5, m0, [r5 + 1 * mmsize]
-    paddw           m4, m5
-    pmaddubsw       m0, [r5]
-%ifidn %1,pp
-    pmulhrsw        m2, m7                          ; m2 = word: row 10
-    pmulhrsw        m4, m7                          ; m4 = word: row 11
-    packuswb        m2, m4
-    vextracti128    xm4, m2, 1
-    movq            [r2 + r3 * 2], xm2
-    movd            [r2 + r3 * 2 + 8], xm4
-    movhps          [r2 + r6], xm2
-    pextrd          [r2 + r6 + 8], xm4, 2
-%else
-    psubw           m2, m7                          ; m2 = word: row 10
-    psubw           m4, m7                          ; m4 = word: row 11
-    movu            [r2 + r3 * 2], xm2
-    vextracti128    xm2, m2, 1
-    movq            [r2 + r3 * 2 + 16], xm2
-    movu            [r2 + r6], xm4
-    vextracti128    xm4, m4, 1
-    movq            [r2 + r6 + 16], xm4
+    psubw           m2, m12
+    psubw           m3, m12
+    vperm2i128      m0, m2, m3, 0x20
+    vperm2i128      m2, m2, m3, 0x31
+    movu            [r2 + r5], m0
+    movu            [r2 + r5 + mmsize], m2
+    add             r2, 64
 %endif
-    lea             r2, [r2 + r3 * 4]
-
-    movu            xm5, [r0 + r4]                  ; m5 = row 15
-    punpckhbw       xm2, xm1, xm5
-    punpcklbw       xm1, xm5
-    vinserti128     m1, m1, xm2, 1
-    pmaddubsw       m2, m1, [r5 + 1 * mmsize]
-    paddw           m6, m2
-    pmaddubsw       m1, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm2, [r0]                       ; m2 = row 16
-    punpckhbw       xm3, xm5, xm2
-    punpcklbw       xm5, xm2
-    vinserti128     m5, m5, xm3, 1
-    pmaddubsw       m3, m5, [r5 + 1 * mmsize]
-    paddw           m0, m3
-    pmaddubsw       m5, [r5]
-    movu            xm3, [r0 + r1]                  ; m3 = row 17
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m2, [r5 + 1 * mmsize]
-    paddw           m1, m2
-    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
-    punpckhbw       xm2, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm2, 1
-    pmaddubsw       m3, [r5 + 1 * mmsize]
-    paddw           m5, m3
-
+    sub             r0, r7
+%assign x x+32
+%endrep
 %ifidn %1,pp
-    pmulhrsw        m6, m7                          ; m6 = word: row 12
-    pmulhrsw        m0, m7                          ; m0 = word: row 13
-    pmulhrsw        m1, m7                          ; m1 = word: row 14
-    pmulhrsw        m5, m7                          ; m5 = word: row 15
-    packuswb        m6, m0
-    packuswb        m1, m5
-    vextracti128    xm0, m6, 1
-    vextracti128    xm5, m1, 1
-    movq            [r2], xm6
-    movd            [r2 + 8], xm0
-    movhps          [r2 + r3], xm6
-    pextrd          [r2 + r3 + 8], xm0, 2
-    movq            [r2 + r3 * 2], xm1
-    movd            [r2 + r3 * 2 + 8], xm5
-    movhps          [r2 + r6], xm1
-    pextrd          [r2 + r6 + 8], xm5, 2
+    lea             r2, [r2 + r3 * 4 - 64]
 %else
-    psubw           m6, m7                          ; m6 = word: row 12
-    psubw           m0, m7                          ; m0 = word: row 13
-    psubw           m1, m7                          ; m1 = word: row 14
-    psubw           m5, m7                          ; m5 = word: row 15
-    movu            [r2], xm6
-    vextracti128    xm6, m6, 1
-    movq            [r2 + 16], xm6
-    movu            [r2 + r3], xm0
-    vextracti128    xm0, m0, 1
-    movq            [r2 + r3 + 16], xm0
-    movu            [r2 + r3 * 2], xm1
-    vextracti128    xm1, m1, 1
-    movq            [r2 + r3 * 2 + 16], xm1
-    movu            [r2 + r6], xm5
-    vextracti128    xm5, m5, 1
-    movq            [r2 + r6 + 16], xm5
+    lea             r2, [r2 + r3 * 4 - 128]
 %endif
-    lea             r2, [r2 + r3 * 4]
-%endrep
+    add             r0, r7
+    dec             r6d
+    jnz             .loopH
     RET
+%endif
 %endmacro
 
-    FILTER_VER_CHROMA_AVX2_12xN pp, 16
-    FILTER_VER_CHROMA_AVX2_12xN ps, 16
-    FILTER_VER_CHROMA_AVX2_12xN pp, 32
-    FILTER_VER_CHROMA_AVX2_12xN ps, 32
+    FILTER_VER_CHROMA_AVX2_64xN pp, 64
+    FILTER_VER_CHROMA_AVX2_64xN pp, 48
+    FILTER_VER_CHROMA_AVX2_64xN pp, 32
+    FILTER_VER_CHROMA_AVX2_64xN pp, 16
+    FILTER_VER_CHROMA_AVX2_64xN ps, 64
+    FILTER_VER_CHROMA_AVX2_64xN ps, 48
+    FILTER_VER_CHROMA_AVX2_64xN ps, 32
+    FILTER_VER_CHROMA_AVX2_64xN ps, 16
 
 ;-----------------------------------------------------------------------------
-;void interp_4tap_vert_pp_24x32(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+; void interp_4tap_vert_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
 ;-----------------------------------------------------------------------------
-%macro FILTER_V4_W24 2
+%macro FILTER_V4_W16n_H2 2
 INIT_XMM sse4
-cglobal interp_4tap_vert_pp_24x%2, 4, 6, 8
+cglobal interp_4tap_vert_pp_%1x%2, 4, 7, 8
 
     mov         r4d,       r4m
     sub         r0,        r1
@@ -11291,9 +9231,14 @@ cglobal interp_4tap_vert_pp_24x%2, 4, 6, 8
     pshufb      m1,        m0,       [tab_Vm]
     pshufb      m0,        [tab_Vm + 16]
 
-    mov         r4d,       %2
+    mov         r4d,       %2/2
 
 .loop:
+
+    mov         r6d,       %1/16
+
+.loopW:
+
     movu        m2,        [r0]
     movu        m3,        [r0 + r1]
 
@@ -11305,20 +9250,20 @@ cglobal interp_4tap_vert_pp_24x%2, 4, 6, 8
 
     lea         r5,        [r0 + 2 * r1]
     movu        m5,        [r5]
-    movu        m7,        [r5 + r1]
+    movu        m6,        [r5 + r1]
 
-    punpcklbw   m6,        m5,        m7
-    pmaddubsw   m6,        m0
-    paddw       m4,        m6
+    punpckhbw   m7,        m5,        m6
+    pmaddubsw   m7,        m0
+    paddw       m2,        m7
 
-    punpckhbw   m6,        m5,        m7
-    pmaddubsw   m6,        m0
-    paddw       m2,        m6
+    punpcklbw   m7,        m5,        m6
+    pmaddubsw   m7,        m0
+    paddw       m4,        m7
 
-    mova        m6,        [pw_512]
+    mova        m7,        [pw_512]
 
-    pmulhrsw    m4,        m6
-    pmulhrsw    m2,        m6
+    pmulhrsw    m4,        m7
+    pmulhrsw    m2,        m7
 
     packuswb    m4,        m2
 
@@ -11330,857 +9275,817 @@ cglobal interp_4tap_vert_pp_24x%2, 4, 6, 8
     pmaddubsw   m4,        m1
     pmaddubsw   m3,        m1
 
-    movu        m2,        [r5 + 2 * r1]
+    movu        m5,        [r5 + 2 * r1]
 
-    punpcklbw   m5,        m7,        m2
-    punpckhbw   m7,        m2
+    punpcklbw   m2,        m6,        m5
+    punpckhbw   m6,        m5
 
-    pmaddubsw   m5,        m0
-    pmaddubsw   m7,        m0
+    pmaddubsw   m2,        m0
+    pmaddubsw   m6,        m0
 
-    paddw       m4,        m5
-    paddw       m3,        m7
+    paddw       m4,        m2
+    paddw       m3,        m6
 
-    pmulhrsw    m4,        m6
-    pmulhrsw    m3,        m6
+    pmulhrsw    m4,        m7
+    pmulhrsw    m3,        m7
 
     packuswb    m4,        m3
 
     movu        [r2 + r3],      m4
 
-    movq        m2,        [r0 + 16]
-    movq        m3,        [r0 + r1 + 16]
-    movq        m4,        [r5 + 16]
-    movq        m5,        [r5 + r1 + 16]
-
-    punpcklbw   m2,        m3
-    punpcklbw   m4,        m5
-
-    pmaddubsw   m2,        m1
-    pmaddubsw   m4,        m0
-
-    paddw       m2,        m4
-
-    pmulhrsw    m2,        m6
+    add         r0,        16
+    add         r2,        16
+    dec         r6d
+    jnz         .loopW
 
-    movq        m3,        [r0 + r1 + 16]
-    movq        m4,        [r5 + 16]
-    movq        m5,        [r5 + r1 + 16]
-    movq        m7,        [r5 + 2 * r1 + 16]
+    lea         r0,        [r0 + r1 * 2 - %1]
+    lea         r2,        [r2 + r3 * 2 - %1]
 
-    punpcklbw   m3,        m4
-    punpcklbw   m5,        m7
+    dec         r4d
+    jnz        .loop
+    RET
+%endmacro
 
-    pmaddubsw   m3,        m1
-    pmaddubsw   m5,        m0
+    FILTER_V4_W16n_H2 64, 64
+    FILTER_V4_W16n_H2 64, 32
+    FILTER_V4_W16n_H2 64, 48
+    FILTER_V4_W16n_H2 48, 64
+    FILTER_V4_W16n_H2 64, 16
 
-    paddw       m3,        m5
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+%macro P2S_H_2xN 1
+INIT_XMM sse4
+cglobal filterPixelToShort_2x%1, 3, 4, 3
+    mov         r3d, r3m
+    add         r3d, r3d
 
-    pmulhrsw    m3,        m6
-    packuswb    m2,        m3
+    ; load constant
+    mova        m1, [pb_128]
+    mova        m2, [tab_c_64_n64]
 
-    movh        [r2 + 16], m2
-    movhps      [r2 + r3 + 16], m2
+%rep %1/2
+    movd        m0, [r0]
+    pinsrd      m0, [r0 + r1], 1
+    punpcklbw   m0, m1
+    pmaddubsw   m0, m2
 
-    mov         r0,        r5
-    lea         r2,        [r2 + 2 * r3]
+    movd        [r2 + r3 * 0], m0
+    pextrd      [r2 + r3 * 1], m0, 2
 
-    sub         r4,        2
-    jnz        .loop
+    lea         r0, [r0 + r1 * 2]
+    lea         r2, [r2 + r3 * 2]
+%endrep
     RET
 %endmacro
-
-    FILTER_V4_W24 24, 32
-
-    FILTER_V4_W24 24, 64
+    P2S_H_2xN 4
+    P2S_H_2xN 8
+    P2S_H_2xN 16
 
 ;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_32x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
 ;-----------------------------------------------------------------------------
-%macro FILTER_V4_W32 2
+%macro P2S_H_4xN 1
 INIT_XMM sse4
-cglobal interp_4tap_vert_pp_%1x%2, 4, 6, 8
+cglobal filterPixelToShort_4x%1, 3, 6, 4
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r4, [r3 * 3]
+    lea         r5, [r1 * 3]
 
-    mov         r4d,       r4m
-    sub         r0,        r1
+    ; load constant
+    mova        m2, [pb_128]
+    mova        m3, [tab_c_64_n64]
 
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeff]
-    movd        m0,        [r5 + r4 * 4]
-%else
-    movd        m0,        [tab_ChromaCoeff + r4 * 4]
+%assign x 0
+%rep %1/4
+    movd        m0, [r0]
+    pinsrd      m0, [r0 + r1], 1
+    punpcklbw   m0, m2
+    pmaddubsw   m0, m3
+
+    movd        m1, [r0 + r1 * 2]
+    pinsrd      m1, [r0 + r5], 1
+    punpcklbw   m1, m2
+    pmaddubsw   m1, m3
+
+    movq        [r2 + r3 * 0], m0
+    movq        [r2 + r3 * 2], m1
+    movhps      [r2 + r3 * 1], m0
+    movhps      [r2 + r4], m1
+%assign x x+1
+%if (x != %1/4)
+    lea         r0, [r0 + r1 * 4]
+    lea         r2, [r2 + r3 * 4]
 %endif
+%endrep
+    RET
+%endmacro
+    P2S_H_4xN 4
+    P2S_H_4xN 8
+    P2S_H_4xN 16
+    P2S_H_4xN 32
 
-    pshufb      m1,        m0,       [tab_Vm]
-    pshufb      m0,        [tab_Vm + 16]
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+%macro P2S_H_6xN 1
+INIT_XMM sse4
+cglobal filterPixelToShort_6x%1, 3, 7, 6
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r4, [r1 * 3]
+    lea         r5, [r3 * 3]
 
-    mova        m7,        [pw_512]
+    ; load height
+    mov         r6d, %1/4
 
-    mov         r4d,       %2
+    ; load constant
+    mova        m4, [pb_128]
+    mova        m5, [tab_c_64_n64]
 
 .loop:
-    movu        m2,        [r0]
-    movu        m3,        [r0 + r1]
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-    punpcklbw   m4,        m2,        m3
-    punpckhbw   m2,        m3
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-    pmaddubsw   m4,        m1
-    pmaddubsw   m2,        m1
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
 
-    lea         r5,        [r0 + 2 * r1]
-    movu        m3,        [r5]
-    movu        m5,        [r5 + r1]
+    movh        m3, [r0 + r4]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
 
-    punpcklbw   m6,        m3,        m5
-    punpckhbw   m3,        m5
+    movh        [r2 + r3 * 0], m0
+    pextrd      [r2 + r3 * 0 + 8], m0, 2
+    movh        [r2 + r3 * 1], m1
+    pextrd      [r2 + r3 * 1 + 8], m1, 2
+    movh        [r2 + r3 * 2], m2
+    pextrd      [r2 + r3 * 2 + 8], m2, 2
+    movh        [r2 + r5], m3
+    pextrd      [r2 + r5 + 8], m3, 2
 
-    pmaddubsw   m6,        m0
-    pmaddubsw   m3,        m0
+    lea         r0, [r0 + r1 * 4]
+    lea         r2, [r2 + r3 * 4]
 
-    paddw       m4,        m6
-    paddw       m2,        m3
+    dec         r6d
+    jnz         .loop
+    RET
+%endmacro
+    P2S_H_6xN 8
+    P2S_H_6xN 16
 
-    pmulhrsw    m4,        m7
-    pmulhrsw    m2,        m7
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+%macro P2S_H_8xN 1
+INIT_XMM ssse3
+cglobal filterPixelToShort_8x%1, 3, 7, 6
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r5, [r1 * 3]
+    lea         r6, [r3 * 3]
 
-    packuswb    m4,        m2
+    ; load height
+    mov         r4d, %1/4
 
-    movu        [r2],      m4
+    ; load constant
+    mova        m4, [pb_128]
+    mova        m5, [tab_c_64_n64]
 
-    movu        m2,        [r0 + 16]
-    movu        m3,        [r0 + r1 + 16]
+.loop:
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-    punpcklbw   m4,        m2,        m3
-    punpckhbw   m2,        m3
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-    pmaddubsw   m4,        m1
-    pmaddubsw   m2,        m1
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
 
-    movu        m3,        [r5 + 16]
-    movu        m5,        [r5 + r1 + 16]
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
 
-    punpcklbw   m6,        m3,        m5
-    punpckhbw   m3,        m5
+    movu        [r2 + r3 * 0], m0
+    movu        [r2 + r3 * 1], m1
+    movu        [r2 + r3 * 2], m2
+    movu        [r2 + r6 ], m3
 
-    pmaddubsw   m6,        m0
-    pmaddubsw   m3,        m0
+    lea         r0, [r0 + r1 * 4]
+    lea         r2, [r2 + r3 * 4]
 
-    paddw       m4,        m6
-    paddw       m2,        m3
+    dec         r4d
+    jnz         .loop
+    RET
+%endmacro
+    P2S_H_8xN 8
+    P2S_H_8xN 4
+    P2S_H_8xN 16
+    P2S_H_8xN 32
+    P2S_H_8xN 12
+    P2S_H_8xN 64
 
-    pmulhrsw    m4,        m7
-    pmulhrsw    m2,        m7
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+INIT_XMM ssse3
+cglobal filterPixelToShort_8x6, 3, 7, 5
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r4, [r1 * 3]
+    lea         r5, [r1 * 5]
+    lea         r6, [r3 * 3]
 
-    packuswb    m4,        m2
+    ; load constant
+    mova        m3, [pb_128]
+    mova        m4, [tab_c_64_n64]
 
-    movu        [r2 + 16], m4
+    movh        m0, [r0]
+    punpcklbw   m0, m3
+    pmaddubsw   m0, m4
 
-    lea         r0,        [r0 + r1]
-    lea         r2,        [r2 + r3]
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m3
+    pmaddubsw   m1, m4
+
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m3
+    pmaddubsw   m2, m4
+
+    movu        [r2 + r3 * 0], m0
+    movu        [r2 + r3 * 1], m1
+    movu        [r2 + r3 * 2], m2
+
+    movh        m0, [r0 + r4]
+    punpcklbw   m0, m3
+    pmaddubsw   m0, m4
+
+    movh        m1, [r0 + r1 * 4]
+    punpcklbw   m1, m3
+    pmaddubsw   m1, m4
+
+    movh        m2, [r0 + r5]
+    punpcklbw   m2, m3
+    pmaddubsw   m2, m4
+
+    movu        [r2 + r6 ], m0
+    movu        [r2 + r3 * 4], m1
+    lea         r2, [r2 + r3 * 4]
+    movu        [r2 + r3], m2
 
-    dec         r4
-    jnz        .loop
     RET
-%endmacro
 
-    FILTER_V4_W32 32,  8
-    FILTER_V4_W32 32, 16
-    FILTER_V4_W32 32, 24
-    FILTER_V4_W32 32, 32
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+%macro P2S_H_16xN 1
+INIT_XMM ssse3
+cglobal filterPixelToShort_16x%1, 3, 7, 6
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r4, [r3 * 3]
+    lea         r5, [r1 * 3]
 
-    FILTER_V4_W32 32, 48
-    FILTER_V4_W32 32, 64
+   ; load height
+    mov         r6d, %1/4
 
-%macro FILTER_VER_CHROMA_AVX2_32xN 2
-%if ARCH_X86_64 == 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_32x%2, 4, 7, 13
-    mov             r4d, r4m
-    shl             r4d, 6
+    ; load constant
+    mova        m4, [pb_128]
+    mova        m5, [tab_c_64_n64]
 
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
+.loop:
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-    mova            m10, [r5]
-    mova            m11, [r5 + mmsize]
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,pp
-    mova            m12, [pw_512]
-%else
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
+
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
+
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
+
+    movu        [r2 + r3 * 0], m0
+    movu        [r2 + r3 * 1], m1
+    movu        [r2 + r3 * 2], m2
+    movu        [r2 + r4], m3
+
+    lea         r0, [r0 + 8]
+
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
+
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
+
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
+
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
+
+    movu        [r2 + r3 * 0 + 16], m0
+    movu        [r2 + r3 * 1 + 16], m1
+    movu        [r2 + r3 * 2 + 16], m2
+    movu        [r2 + r4 + 16], m3
+
+    lea         r0, [r0 + r1 * 4 - 8]
+    lea         r2, [r2 + r3 * 4]
+
+    dec         r6d
+    jnz         .loop
+    RET
+%endmacro
+    P2S_H_16xN 16
+    P2S_H_16xN 4
+    P2S_H_16xN 8
+    P2S_H_16xN 12
+    P2S_H_16xN 32
+    P2S_H_16xN 64
+    P2S_H_16xN 24
+
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+INIT_YMM avx2
+cglobal filterPixelToShort_16x4, 3, 4, 2
+    mov             r3d, r3m
     add             r3d, r3d
-    vbroadcasti128  m12, [pw_2000]
-%endif
+
+    ; load constant
+    vbroadcasti128  m1, [pw_2000]
+
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
+
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
+
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
+
+    lea             r1, [r1 * 3]
+    lea             r3, [r3 * 3]
+
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
+    RET
+
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+INIT_YMM avx2
+cglobal filterPixelToShort_16x8, 3, 6, 2
+    mov             r3d, r3m
+    add             r3d, r3d
+    lea             r4, [r1 * 3]
     lea             r5, [r3 * 3]
-    mov             r6d, %2 / 4
-.loopW:
-    movu            m0, [r0]                        ; m0 = row 0
-    movu            m1, [r0 + r1]                   ; m1 = row 1
-    punpcklbw       m2, m0, m1
-    punpckhbw       m3, m0, m1
-    pmaddubsw       m2, m10
-    pmaddubsw       m3, m10
-    movu            m0, [r0 + r1 * 2]               ; m0 = row 2
-    punpcklbw       m4, m1, m0
-    punpckhbw       m5, m1, m0
-    pmaddubsw       m4, m10
-    pmaddubsw       m5, m10
-    movu            m1, [r0 + r4]                   ; m1 = row 3
-    punpcklbw       m6, m0, m1
-    punpckhbw       m7, m0, m1
-    pmaddubsw       m8, m6, m11
-    pmaddubsw       m9, m7, m11
-    pmaddubsw       m6, m10
-    pmaddubsw       m7, m10
-    paddw           m2, m8
-    paddw           m3, m9
-%ifidn %1,pp
-    pmulhrsw        m2, m12
-    pmulhrsw        m3, m12
-    packuswb        m2, m3
-    movu            [r2], m2
-%else
-    psubw           m2, m12
-    psubw           m3, m12
-    vperm2i128      m0, m2, m3, 0x20
-    vperm2i128      m2, m2, m3, 0x31
+
+    ; load constant
+    vbroadcasti128  m1, [pw_2000]
+
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
     movu            [r2], m0
-    movu            [r2 + mmsize], m2
-%endif
-    lea             r0, [r0 + r1 * 4]
-    movu            m0, [r0]                        ; m0 = row 4
-    punpcklbw       m2, m1, m0
-    punpckhbw       m3, m1, m0
-    pmaddubsw       m8, m2, m11
-    pmaddubsw       m9, m3, m11
-    pmaddubsw       m2, m10
-    pmaddubsw       m3, m10
-    paddw           m4, m8
-    paddw           m5, m9
-%ifidn %1,pp
-    pmulhrsw        m4, m12
-    pmulhrsw        m5, m12
-    packuswb        m4, m5
-    movu            [r2 + r3], m4
-%else
-    psubw           m4, m12
-    psubw           m5, m12
-    vperm2i128      m1, m4, m5, 0x20
-    vperm2i128      m4, m4, m5, 0x31
-    movu            [r2 + r3], m1
-    movu            [r2 + r3 + mmsize], m4
-%endif
 
-    movu            m1, [r0 + r1]                   ; m1 = row 5
-    punpcklbw       m4, m0, m1
-    punpckhbw       m5, m0, m1
-    pmaddubsw       m4, m11
-    pmaddubsw       m5, m11
-    paddw           m6, m4
-    paddw           m7, m5
-%ifidn %1,pp
-    pmulhrsw        m6, m12
-    pmulhrsw        m7, m12
-    packuswb        m6, m7
-    movu            [r2 + r3 * 2], m6
-%else
-    psubw           m6, m12
-    psubw           m7, m12
-    vperm2i128      m0, m6, m7, 0x20
-    vperm2i128      m6, m6, m7, 0x31
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
+
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
     movu            [r2 + r3 * 2], m0
-    movu            [r2 + r3 * 2 + mmsize], m6
-%endif
 
-    movu            m0, [r0 + r1 * 2]               ; m0 = row 6
-    punpcklbw       m6, m1, m0
-    punpckhbw       m7, m1, m0
-    pmaddubsw       m6, m11
-    pmaddubsw       m7, m11
-    paddw           m2, m6
-    paddw           m3, m7
-%ifidn %1,pp
-    pmulhrsw        m2, m12
-    pmulhrsw        m3, m12
-    packuswb        m2, m3
-    movu            [r2 + r5], m2
-%else
-    psubw           m2, m12
-    psubw           m3, m12
-    vperm2i128      m0, m2, m3, 0x20
-    vperm2i128      m2, m2, m3, 0x31
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
     movu            [r2 + r5], m0
-    movu            [r2 + r5 + mmsize], m2
-%endif
+
+    lea             r0, [r0 + r1 * 4]
     lea             r2, [r2 + r3 * 4]
-    dec             r6d
-    jnz             .loopW
-    RET
-%endif
-%endmacro
 
-    FILTER_VER_CHROMA_AVX2_32xN pp, 64
-    FILTER_VER_CHROMA_AVX2_32xN pp, 48
-    FILTER_VER_CHROMA_AVX2_32xN pp, 32
-    FILTER_VER_CHROMA_AVX2_32xN pp, 24
-    FILTER_VER_CHROMA_AVX2_32xN pp, 16
-    FILTER_VER_CHROMA_AVX2_32xN pp, 8
-    FILTER_VER_CHROMA_AVX2_32xN ps, 64
-    FILTER_VER_CHROMA_AVX2_32xN ps, 48
-    FILTER_VER_CHROMA_AVX2_32xN ps, 32
-    FILTER_VER_CHROMA_AVX2_32xN ps, 24
-    FILTER_VER_CHROMA_AVX2_32xN ps, 16
-    FILTER_VER_CHROMA_AVX2_32xN ps, 8
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-%macro FILTER_VER_CHROMA_AVX2_48x64 1
-%if ARCH_X86_64 == 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_48x64, 4, 8, 13
-    mov             r4d, r4m
-    shl             r4d, 6
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-    mova            m10, [r5]
-    mova            m11, [r5 + mmsize]
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,pp
-    mova            m12, [pw_512]
-%else
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
+    RET
+
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+INIT_YMM avx2
+cglobal filterPixelToShort_16x12, 3, 6, 2
+    mov             r3d, r3m
     add             r3d, r3d
-    vbroadcasti128  m12, [pw_2000]
-%endif
+    lea             r4, [r1 * 3]
     lea             r5, [r3 * 3]
-    lea             r7, [r1 * 4]
-    mov             r6d, 16
-.loopH:
-    movu            m0, [r0]                        ; m0 = row 0
-    movu            m1, [r0 + r1]                   ; m1 = row 1
-    punpcklbw       m2, m0, m1
-    punpckhbw       m3, m0, m1
-    pmaddubsw       m2, m10
-    pmaddubsw       m3, m10
-    movu            m0, [r0 + r1 * 2]               ; m0 = row 2
-    punpcklbw       m4, m1, m0
-    punpckhbw       m5, m1, m0
-    pmaddubsw       m4, m10
-    pmaddubsw       m5, m10
-    movu            m1, [r0 + r4]                   ; m1 = row 3
-    punpcklbw       m6, m0, m1
-    punpckhbw       m7, m0, m1
-    pmaddubsw       m8, m6, m11
-    pmaddubsw       m9, m7, m11
-    pmaddubsw       m6, m10
-    pmaddubsw       m7, m10
-    paddw           m2, m8
-    paddw           m3, m9
-%ifidn %1,pp
-    pmulhrsw        m2, m12
-    pmulhrsw        m3, m12
-    packuswb        m2, m3
-    movu            [r2], m2
-%else
-    psubw           m2, m12
-    psubw           m3, m12
-    vperm2i128      m0, m2, m3, 0x20
-    vperm2i128      m2, m2, m3, 0x31
+
+    ; load constant
+    vbroadcasti128  m1, [pw_2000]
+
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
     movu            [r2], m0
-    movu            [r2 + mmsize], m2
-%endif
-    lea             r0, [r0 + r1 * 4]
-    movu            m0, [r0]                        ; m0 = row 4
-    punpcklbw       m2, m1, m0
-    punpckhbw       m3, m1, m0
-    pmaddubsw       m8, m2, m11
-    pmaddubsw       m9, m3, m11
-    pmaddubsw       m2, m10
-    pmaddubsw       m3, m10
-    paddw           m4, m8
-    paddw           m5, m9
-%ifidn %1,pp
-    pmulhrsw        m4, m12
-    pmulhrsw        m5, m12
-    packuswb        m4, m5
-    movu            [r2 + r3], m4
-%else
-    psubw           m4, m12
-    psubw           m5, m12
-    vperm2i128      m1, m4, m5, 0x20
-    vperm2i128      m4, m4, m5, 0x31
-    movu            [r2 + r3], m1
-    movu            [r2 + r3 + mmsize], m4
-%endif
 
-    movu            m1, [r0 + r1]                   ; m1 = row 5
-    punpcklbw       m4, m0, m1
-    punpckhbw       m5, m0, m1
-    pmaddubsw       m4, m11
-    pmaddubsw       m5, m11
-    paddw           m6, m4
-    paddw           m7, m5
-%ifidn %1,pp
-    pmulhrsw        m6, m12
-    pmulhrsw        m7, m12
-    packuswb        m6, m7
-    movu            [r2 + r3 * 2], m6
-%else
-    psubw           m6, m12
-    psubw           m7, m12
-    vperm2i128      m0, m6, m7, 0x20
-    vperm2i128      m6, m6, m7, 0x31
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
+
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
     movu            [r2 + r3 * 2], m0
-    movu            [r2 + r3 * 2 + mmsize], m6
-%endif
 
-    movu            m0, [r0 + r1 * 2]               ; m0 = row 6
-    punpcklbw       m6, m1, m0
-    punpckhbw       m7, m1, m0
-    pmaddubsw       m6, m11
-    pmaddubsw       m7, m11
-    paddw           m2, m6
-    paddw           m3, m7
-%ifidn %1,pp
-    pmulhrsw        m2, m12
-    pmulhrsw        m3, m12
-    packuswb        m2, m3
-    movu            [r2 + r5], m2
-    add             r2, 32
-%else
-    psubw           m2, m12
-    psubw           m3, m12
-    vperm2i128      m0, m2, m3, 0x20
-    vperm2i128      m2, m2, m3, 0x31
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
     movu            [r2 + r5], m0
-    movu            [r2 + r5 + mmsize], m2
-    add             r2, 64
-%endif
-    sub             r0, r7
 
-    movu            xm0, [r0 + 32]                  ; m0 = row 0
-    movu            xm1, [r0 + r1 + 32]             ; m1 = row 1
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m0, m10
-    movu            xm2, [r0 + r1 * 2 + 32]         ; m2 = row 2
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m1, m10
-    movu            xm3, [r0 + r4 + 32]             ; m3 = row 3
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, m11
-    paddw           m0, m4
-    pmaddubsw       m2, m10
     lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0 + 32]                  ; m4 = row 4
-    punpckhbw       xm5, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, m11
-    paddw           m1, m5
-    pmaddubsw       m3, m10
-    movu            xm5, [r0 + r1 + 32]             ; m5 = row 5
-    punpckhbw       xm6, xm4, xm5
-    punpcklbw       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m4, m11
-    paddw           m2, m4
-    movu            xm6, [r0 + r1 * 2 + 32]         ; m6 = row 6
-    punpckhbw       xm7, xm5, xm6
-    punpcklbw       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m5, m11
-    paddw           m3, m5
-%ifidn %1,pp
-    pmulhrsw        m0, m12                         ; m0 = word: row 0
-    pmulhrsw        m1, m12                         ; m1 = word: row 1
-    pmulhrsw        m2, m12                         ; m2 = word: row 2
-    pmulhrsw        m3, m12                         ; m3 = word: row 3
-    packuswb        m0, m1
-    packuswb        m2, m3
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r5], xm3
-    lea             r2, [r2 + r3 * 4 - 32]
-%else
-    psubw           m0, m12                         ; m0 = word: row 0
-    psubw           m1, m12                         ; m1 = word: row 1
-    psubw           m2, m12                         ; m2 = word: row 2
-    psubw           m3, m12                         ; m3 = word: row 3
+    lea             r2, [r2 + r3 * 4]
+
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
     movu            [r2], m0
-    movu            [r2 + r3], m1
-    movu            [r2 + r3 * 2], m2
-    movu            [r2 + r5], m3
-    lea             r2, [r2 + r3 * 4 - 64]
-%endif
-    dec             r6d
-    jnz             .loopH
-    RET
-%endif
-%endmacro
 
-    FILTER_VER_CHROMA_AVX2_48x64 pp
-    FILTER_VER_CHROMA_AVX2_48x64 ps
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-%macro FILTER_VER_CHROMA_AVX2_64xN 2
-%if ARCH_X86_64 == 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_64x%2, 4, 8, 13
-    mov             r4d, r4m
-    shl             r4d, 6
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
 
-    mova            m10, [r5]
-    mova            m11, [r5 + mmsize]
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,pp
-    mova            m12, [pw_512]
-%else
-    add             r3d, r3d
-    vbroadcasti128  m12, [pw_2000]
-%endif
-    lea             r5, [r3 * 3]
-    lea             r7, [r1 * 4]
-    mov             r6d, %2 / 4
-.loopH:
-%assign x 0
-%rep 2
-    movu            m0, [r0 + x]                    ; m0 = row 0
-    movu            m1, [r0 + r1 + x]               ; m1 = row 1
-    punpcklbw       m2, m0, m1
-    punpckhbw       m3, m0, m1
-    pmaddubsw       m2, m10
-    pmaddubsw       m3, m10
-    movu            m0, [r0 + r1 * 2 + x]           ; m0 = row 2
-    punpcklbw       m4, m1, m0
-    punpckhbw       m5, m1, m0
-    pmaddubsw       m4, m10
-    pmaddubsw       m5, m10
-    movu            m1, [r0 + r4 + x]               ; m1 = row 3
-    punpcklbw       m6, m0, m1
-    punpckhbw       m7, m0, m1
-    pmaddubsw       m8, m6, m11
-    pmaddubsw       m9, m7, m11
-    pmaddubsw       m6, m10
-    pmaddubsw       m7, m10
-    paddw           m2, m8
-    paddw           m3, m9
-%ifidn %1,pp
-    pmulhrsw        m2, m12
-    pmulhrsw        m3, m12
-    packuswb        m2, m3
-    movu            [r2], m2
-%else
-    psubw           m2, m12
-    psubw           m3, m12
-    vperm2i128      m0, m2, m3, 0x20
-    vperm2i128      m2, m2, m3, 0x31
-    movu            [r2], m0
-    movu            [r2 + mmsize], m2
-%endif
     lea             r0, [r0 + r1 * 4]
-    movu            m0, [r0 + x]                    ; m0 = row 4
-    punpcklbw       m2, m1, m0
-    punpckhbw       m3, m1, m0
-    pmaddubsw       m8, m2, m11
-    pmaddubsw       m9, m3, m11
-    pmaddubsw       m2, m10
-    pmaddubsw       m3, m10
-    paddw           m4, m8
-    paddw           m5, m9
-%ifidn %1,pp
-    pmulhrsw        m4, m12
-    pmulhrsw        m5, m12
-    packuswb        m4, m5
-    movu            [r2 + r3], m4
-%else
-    psubw           m4, m12
-    psubw           m5, m12
-    vperm2i128      m1, m4, m5, 0x20
-    vperm2i128      m4, m4, m5, 0x31
-    movu            [r2 + r3], m1
-    movu            [r2 + r3 + mmsize], m4
-%endif
+    lea             r2, [r2 + r3 * 4]
 
-    movu            m1, [r0 + r1 + x]               ; m1 = row 5
-    punpcklbw       m4, m0, m1
-    punpckhbw       m5, m0, m1
-    pmaddubsw       m4, m11
-    pmaddubsw       m5, m11
-    paddw           m6, m4
-    paddw           m7, m5
-%ifidn %1,pp
-    pmulhrsw        m6, m12
-    pmulhrsw        m7, m12
-    packuswb        m6, m7
-    movu            [r2 + r3 * 2], m6
-%else
-    psubw           m6, m12
-    psubw           m7, m12
-    vperm2i128      m0, m6, m7, 0x20
-    vperm2i128      m6, m6, m7, 0x31
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
+
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
+
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
     movu            [r2 + r3 * 2], m0
-    movu            [r2 + r3 * 2 + mmsize], m6
-%endif
 
-    movu            m0, [r0 + r1 * 2 + x]           ; m0 = row 6
-    punpcklbw       m6, m1, m0
-    punpckhbw       m7, m1, m0
-    pmaddubsw       m6, m11
-    pmaddubsw       m7, m11
-    paddw           m2, m6
-    paddw           m3, m7
-%ifidn %1,pp
-    pmulhrsw        m2, m12
-    pmulhrsw        m3, m12
-    packuswb        m2, m3
-    movu            [r2 + r5], m2
-    add             r2, 32
-%else
-    psubw           m2, m12
-    psubw           m3, m12
-    vperm2i128      m0, m2, m3, 0x20
-    vperm2i128      m2, m2, m3, 0x31
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
     movu            [r2 + r5], m0
-    movu            [r2 + r5 + mmsize], m2
-    add             r2, 64
-%endif
-    sub             r0, r7
-%assign x x+32
-%endrep
-%ifidn %1,pp
-    lea             r2, [r2 + r3 * 4 - 64]
-%else
-    lea             r2, [r2 + r3 * 4 - 128]
-%endif
-    add             r0, r7
-    dec             r6d
-    jnz             .loopH
     RET
-%endif
-%endmacro
-
-    FILTER_VER_CHROMA_AVX2_64xN pp, 64
-    FILTER_VER_CHROMA_AVX2_64xN pp, 48
-    FILTER_VER_CHROMA_AVX2_64xN pp, 32
-    FILTER_VER_CHROMA_AVX2_64xN pp, 16
-    FILTER_VER_CHROMA_AVX2_64xN ps, 64
-    FILTER_VER_CHROMA_AVX2_64xN ps, 48
-    FILTER_VER_CHROMA_AVX2_64xN ps, 32
-    FILTER_VER_CHROMA_AVX2_64xN ps, 16
 
 ;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
 ;-----------------------------------------------------------------------------
-%macro FILTER_V4_W16n_H2 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_pp_%1x%2, 4, 7, 8
+INIT_YMM avx2
+cglobal filterPixelToShort_16x16, 3, 6, 2
+    mov             r3d, r3m
+    add             r3d, r3d
+    lea             r4, [r1 * 3]
+    lea             r5, [r3 * 3]
 
-    mov         r4d,       r4m
-    sub         r0,        r1
+    ; load constant
+    vbroadcasti128  m1, [pw_2000]
 
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeff]
-    movd        m0,        [r5 + r4 * 4]
-%else
-    movd        m0,        [tab_ChromaCoeff + r4 * 4]
-%endif
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-    pshufb      m1,        m0,       [tab_Vm]
-    pshufb      m0,        [tab_Vm + 16]
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-    mov         r4d,       %2/2
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-.loop:
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
 
-    mov         r6d,       %1/16
+    lea             r0, [r0 + r1 * 4]
+    lea             r2, [r2 + r3 * 4]
 
-.loopW:
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-    movu        m2,        [r0]
-    movu        m3,        [r0 + r1]
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-    punpcklbw   m4,        m2,        m3
-    punpckhbw   m2,        m3
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-    pmaddubsw   m4,        m1
-    pmaddubsw   m2,        m1
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
 
-    lea         r5,        [r0 + 2 * r1]
-    movu        m5,        [r5]
-    movu        m6,        [r5 + r1]
+    lea             r0, [r0 + r1 * 4]
+    lea             r2, [r2 + r3 * 4]
 
-    punpckhbw   m7,        m5,        m6
-    pmaddubsw   m7,        m0
-    paddw       m2,        m7
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-    punpcklbw   m7,        m5,        m6
-    pmaddubsw   m7,        m0
-    paddw       m4,        m7
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-    mova        m7,        [pw_512]
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-    pmulhrsw    m4,        m7
-    pmulhrsw    m2,        m7
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
 
-    packuswb    m4,        m2
+    lea             r0, [r0 + r1 * 4]
+    lea             r2, [r2 + r3 * 4]
 
-    movu        [r2],      m4
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-    punpcklbw   m4,        m3,        m5
-    punpckhbw   m3,        m5
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-    pmaddubsw   m4,        m1
-    pmaddubsw   m3,        m1
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-    movu        m5,        [r5 + 2 * r1]
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
+    RET
 
-    punpcklbw   m2,        m6,        m5
-    punpckhbw   m6,        m5
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+INIT_YMM avx2
+cglobal filterPixelToShort_16x24, 3, 7, 2
+    mov             r3d, r3m
+    add             r3d, r3d
+    lea             r4, [r1 * 3]
+    lea             r5, [r3 * 3]
+    mov             r6d, 3
 
-    pmaddubsw   m2,        m0
-    pmaddubsw   m6,        m0
+    ; load constant
+    vbroadcasti128  m1, [pw_2000]
+.loop:
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-    paddw       m4,        m2
-    paddw       m3,        m6
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-    pmulhrsw    m4,        m7
-    pmulhrsw    m3,        m7
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-    packuswb    m4,        m3
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
 
-    movu        [r2 + r3],      m4
+    lea             r0, [r0 + r1 * 4]
+    lea             r2, [r2 + r3 * 4]
 
-    add         r0,        16
-    add         r2,        16
-    dec         r6d
-    jnz         .loopW
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-    lea         r0,        [r0 + r1 * 2 - %1]
-    lea         r2,        [r2 + r3 * 2 - %1]
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-    dec         r4d
-    jnz        .loop
-    RET
-%endmacro
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-    FILTER_V4_W16n_H2 64, 64
-    FILTER_V4_W16n_H2 64, 32
-    FILTER_V4_W16n_H2 64, 48
-    FILTER_V4_W16n_H2 48, 64
-    FILTER_V4_W16n_H2 64, 16
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
+
+    lea             r0, [r0 + r1 * 4]
+    lea             r2, [r2 + r3 * 4]
+
+    dec             r6d
+    jnz             .loop
+    RET
 
 ;-----------------------------------------------------------------------------
 ; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
 ;-----------------------------------------------------------------------------
-%macro P2S_H_2xN 1
-INIT_XMM sse4
-cglobal filterPixelToShort_2x%1, 3, 4, 3
-    mov         r3d, r3m
-    add         r3d, r3d
+%macro P2S_H_16xN_avx2 1
+INIT_YMM avx2
+cglobal filterPixelToShort_16x%1, 3, 7, 2
+    mov             r3d, r3m
+    add             r3d, r3d
+    lea             r4, [r1 * 3]
+    lea             r5, [r3 * 3]
+    mov             r6d, %1/16
 
     ; load constant
-    mova        m1, [pb_128]
-    mova        m2, [tab_c_64_n64]
+    vbroadcasti128  m1, [pw_2000]
+.loop:
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-%rep %1/2
-    movd        m0, [r0]
-    pinsrd      m0, [r0 + r1], 1
-    punpcklbw   m0, m1
-    pmaddubsw   m0, m2
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-    movd        [r2 + r3 * 0], m0
-    pextrd      [r2 + r3 * 1], m0, 2
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-    lea         r0, [r0 + r1 * 2]
-    lea         r2, [r2 + r3 * 2]
-%endrep
-    RET
-%endmacro
-    P2S_H_2xN 4
-    P2S_H_2xN 8
-    P2S_H_2xN 16
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
 
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-%macro P2S_H_4xN 1
-INIT_XMM sse4
-cglobal filterPixelToShort_4x%1, 3, 6, 4
-    mov         r3d, r3m
-    add         r3d, r3d
-    lea         r4, [r3 * 3]
-    lea         r5, [r1 * 3]
+    lea             r0, [r0 + r1 * 4]
+    lea             r2, [r2 + r3 * 4]
 
-    ; load constant
-    mova        m2, [pb_128]
-    mova        m3, [tab_c_64_n64]
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-%assign x 0
-%rep %1/4
-    movd        m0, [r0]
-    pinsrd      m0, [r0 + r1], 1
-    punpcklbw   m0, m2
-    pmaddubsw   m0, m3
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-    movd        m1, [r0 + r1 * 2]
-    pinsrd      m1, [r0 + r5], 1
-    punpcklbw   m1, m2
-    pmaddubsw   m1, m3
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-    movq        [r2 + r3 * 0], m0
-    movq        [r2 + r3 * 2], m1
-    movhps      [r2 + r3 * 1], m0
-    movhps      [r2 + r4], m1
-%assign x x+1
-%if (x != %1/4)
-    lea         r0, [r0 + r1 * 4]
-    lea         r2, [r2 + r3 * 4]
-%endif
-%endrep
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
+
+    lea             r0, [r0 + r1 * 4]
+    lea             r2, [r2 + r3 * 4]
+
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
+
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
+
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
+
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
+
+    lea             r0, [r0 + r1 * 4]
+    lea             r2, [r2 + r3 * 4]
+
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
+
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
+
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
+
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
+
+    lea             r0, [r0 + r1 * 4]
+    lea             r2, [r2 + r3 * 4]
+
+    dec             r6d
+    jnz             .loop
     RET
 %endmacro
-    P2S_H_4xN 4
-    P2S_H_4xN 8
-    P2S_H_4xN 16
-    P2S_H_4xN 32
+P2S_H_16xN_avx2 32
+P2S_H_16xN_avx2 64
 
 ;-----------------------------------------------------------------------------
 ; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
 ;-----------------------------------------------------------------------------
-%macro P2S_H_6xN 1
-INIT_XMM sse4
-cglobal filterPixelToShort_6x%1, 3, 7, 6
+%macro P2S_H_32xN 1
+INIT_XMM ssse3
+cglobal filterPixelToShort_32x%1, 3, 7, 6
     mov         r3d, r3m
     add         r3d, r3d
-    lea         r4, [r1 * 3]
-    lea         r5, [r3 * 3]
+    lea         r4, [r3 * 3]
+    lea         r5, [r1 * 3]
 
     ; load height
     mov         r6d, %1/4
@@ -12202,48 +10107,40 @@ cglobal filterPixelToShort_6x%1, 3, 7, 6
     punpcklbw   m2, m4
     pmaddubsw   m2, m5
 
-    movh        m3, [r0 + r4]
+    movh        m3, [r0 + r5]
     punpcklbw   m3, m4
     pmaddubsw   m3, m5
 
-    movh        [r2 + r3 * 0], m0
-    pextrd      [r2 + r3 * 0 + 8], m0, 2
-    movh        [r2 + r3 * 1], m1
-    pextrd      [r2 + r3 * 1 + 8], m1, 2
-    movh        [r2 + r3 * 2], m2
-    pextrd      [r2 + r3 * 2 + 8], m2, 2
-    movh        [r2 + r5], m3
-    pextrd      [r2 + r5 + 8], m3, 2
-
-    lea         r0, [r0 + r1 * 4]
-    lea         r2, [r2 + r3 * 4]
+    movu        [r2 + r3 * 0], m0
+    movu        [r2 + r3 * 1], m1
+    movu        [r2 + r3 * 2], m2
+    movu        [r2 + r4], m3
 
-    dec         r6d
-    jnz         .loop
-    RET
-%endmacro
-    P2S_H_6xN 8
-    P2S_H_6xN 16
+    lea         r0, [r0 + 8]
 
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-%macro P2S_H_8xN 1
-INIT_XMM ssse3
-cglobal filterPixelToShort_8x%1, 3, 7, 6
-    mov         r3d, r3m
-    add         r3d, r3d
-    lea         r5, [r1 * 3]
-    lea         r6, [r3 * 3]
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-    ; load height
-    mov         r4d, %1/4
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-    ; load constant
-    mova        m4, [pb_128]
-    mova        m5, [tab_c_64_n64]
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
+
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
+
+    movu        [r2 + r3 * 0 + 16], m0
+    movu        [r2 + r3 * 1 + 16], m1
+    movu        [r2 + r3 * 2 + 16], m2
+    movu        [r2 + r4 + 16], m3
+
+    lea         r0, [r0 + 8]
 
-.loop:
     movh        m0, [r0]
     punpcklbw   m0, m4
     pmaddubsw   m0, m5
@@ -12260,87 +10157,128 @@ cglobal filterPixelToShort_8x%1, 3, 7, 6
     punpcklbw   m3, m4
     pmaddubsw   m3, m5
 
-    movu        [r2 + r3 * 0], m0
-    movu        [r2 + r3 * 1], m1
-    movu        [r2 + r3 * 2], m2
-    movu        [r2 + r6 ], m3
+    movu        [r2 + r3 * 0 + 32], m0
+    movu        [r2 + r3 * 1 + 32], m1
+    movu        [r2 + r3 * 2 + 32], m2
+    movu        [r2 + r4 + 32], m3
 
-    lea         r0, [r0 + r1 * 4]
+    lea         r0, [r0 + 8]
+
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
+
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
+
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
+
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
+
+    movu        [r2 + r3 * 0 + 48], m0
+    movu        [r2 + r3 * 1 + 48], m1
+    movu        [r2 + r3 * 2 + 48], m2
+    movu        [r2 + r4 + 48], m3
+
+    lea         r0, [r0 + r1 * 4 - 24]
     lea         r2, [r2 + r3 * 4]
 
-    dec         r4d
+    dec         r6d
     jnz         .loop
     RET
 %endmacro
-    P2S_H_8xN 8
-    P2S_H_8xN 4
-    P2S_H_8xN 16
-    P2S_H_8xN 32
-    P2S_H_8xN 12
-    P2S_H_8xN 64
+    P2S_H_32xN 32
+    P2S_H_32xN 8
+    P2S_H_32xN 16
+    P2S_H_32xN 24
+    P2S_H_32xN 64
+    P2S_H_32xN 48
 
 ;-----------------------------------------------------------------------------
 ; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
 ;-----------------------------------------------------------------------------
-INIT_XMM ssse3
-cglobal filterPixelToShort_8x6, 3, 7, 5
+%macro P2S_H_32xN_avx2 1
+INIT_YMM avx2
+cglobal filterPixelToShort_32x%1, 3, 7, 3
     mov         r3d, r3m
     add         r3d, r3d
-    lea         r4, [r1 * 3]
-    lea         r5, [r1 * 5]
+    lea         r5, [r1 * 3]
     lea         r6, [r3 * 3]
 
-    ; load constant
-    mova        m3, [pb_128]
-    mova        m4, [tab_c_64_n64]
-
-    movh        m0, [r0]
-    punpcklbw   m0, m3
-    pmaddubsw   m0, m4
-
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m3
-    pmaddubsw   m1, m4
+    ; load height
+    mov         r4d, %1/4
 
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m3
-    pmaddubsw   m2, m4
+    ; load constant
+    vpbroadcastd m2, [pw_2000]
 
-    movu        [r2 + r3 * 0], m0
-    movu        [r2 + r3 * 1], m1
-    movu        [r2 + r3 * 2], m2
+.loop:
+    pmovzxbw    m0, [r0 + 0 * mmsize/2]
+    pmovzxbw    m1, [r0 + 1 * mmsize/2]
+    psllw       m0, 6
+    psllw       m1, 6
+    psubw       m0, m2
+    psubw       m1, m2
+    movu        [r2 + 0 * mmsize], m0
+    movu        [r2 + 1 * mmsize], m1
 
-    movh        m0, [r0 + r4]
-    punpcklbw   m0, m3
-    pmaddubsw   m0, m4
+    pmovzxbw    m0, [r0 + r1 + 0 * mmsize/2]
+    pmovzxbw    m1, [r0 + r1 + 1 * mmsize/2]
+    psllw       m0, 6
+    psllw       m1, 6
+    psubw       m0, m2
+    psubw       m1, m2
+    movu        [r2 + r3 + 0 * mmsize], m0
+    movu        [r2 + r3 + 1 * mmsize], m1
 
-    movh        m1, [r0 + r1 * 4]
-    punpcklbw   m1, m3
-    pmaddubsw   m1, m4
+    pmovzxbw    m0, [r0 + r1 * 2 + 0 * mmsize/2]
+    pmovzxbw    m1, [r0 + r1 * 2 + 1 * mmsize/2]
+    psllw       m0, 6
+    psllw       m1, 6
+    psubw       m0, m2
+    psubw       m1, m2
+    movu        [r2 + r3 * 2 + 0 * mmsize], m0
+    movu        [r2 + r3 * 2 + 1 * mmsize], m1
 
-    movh        m2, [r0 + r5]
-    punpcklbw   m2, m3
-    pmaddubsw   m2, m4
+    pmovzxbw    m0, [r0 + r5 + 0 * mmsize/2]
+    pmovzxbw    m1, [r0 + r5 + 1 * mmsize/2]
+    psllw       m0, 6
+    psllw       m1, 6
+    psubw       m0, m2
+    psubw       m1, m2
+    movu        [r2 + r6 + 0 * mmsize], m0
+    movu        [r2 + r6 + 1 * mmsize], m1
 
-    movu        [r2 + r6 ], m0
-    movu        [r2 + r3 * 4], m1
+    lea         r0, [r0 + r1 * 4]
     lea         r2, [r2 + r3 * 4]
-    movu        [r2 + r3], m2
 
+    dec         r4d
+    jnz        .loop
     RET
+%endmacro
+    P2S_H_32xN_avx2 32
+    P2S_H_32xN_avx2 8
+    P2S_H_32xN_avx2 16
+    P2S_H_32xN_avx2 24
+    P2S_H_32xN_avx2 64
+    P2S_H_32xN_avx2 48
 
 ;-----------------------------------------------------------------------------
 ; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
 ;-----------------------------------------------------------------------------
-%macro P2S_H_16xN 1
+%macro P2S_H_64xN 1
 INIT_XMM ssse3
-cglobal filterPixelToShort_16x%1, 3, 7, 6
+cglobal filterPixelToShort_64x%1, 3, 7, 6
     mov         r3d, r3m
     add         r3d, r3d
     lea         r4, [r3 * 3]
     lea         r5, [r1 * 3]
 
-   ; load height
+    ; load height
     mov         r6d, %1/4
 
     ; load constant
@@ -12392,650 +10330,671 @@ cglobal filterPixelToShort_16x%1, 3, 7, 6
     movu        [r2 + r3 * 2 + 16], m2
     movu        [r2 + r4 + 16], m3
 
-    lea         r0, [r0 + r1 * 4 - 8]
-    lea         r2, [r2 + r3 * 4]
+    lea         r0, [r0 + 8]
 
-    dec         r6d
-    jnz         .loop
-    RET
-%endmacro
-    P2S_H_16xN 16
-    P2S_H_16xN 4
-    P2S_H_16xN 8
-    P2S_H_16xN 12
-    P2S_H_16xN 32
-    P2S_H_16xN 64
-    P2S_H_16xN 24
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-INIT_YMM avx2
-cglobal filterPixelToShort_16x4, 3, 4, 2
-    mov             r3d, r3m
-    add             r3d, r3d
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-    ; load constant
-    vbroadcasti128  m1, [pw_2000]
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
 
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
 
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
+    movu        [r2 + r3 * 0 + 32], m0
+    movu        [r2 + r3 * 1 + 32], m1
+    movu        [r2 + r3 * 2 + 32], m2
+    movu        [r2 + r4 + 32], m3
 
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
+    lea         r0, [r0 + 8]
 
-    lea             r1, [r1 * 3]
-    lea             r3, [r3 * 3]
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
-    RET
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-INIT_YMM avx2
-cglobal filterPixelToShort_16x8, 3, 6, 2
-    mov             r3d, r3m
-    add             r3d, r3d
-    lea             r4, [r1 * 3]
-    lea             r5, [r3 * 3]
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
 
-    ; load constant
-    vbroadcasti128  m1, [pw_2000]
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
 
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
+    movu        [r2 + r3 * 0 + 48], m0
+    movu        [r2 + r3 * 1 + 48], m1
+    movu        [r2 + r3 * 2 + 48], m2
+    movu        [r2 + r4 + 48], m3
 
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
+    lea         r0, [r0 + 8]
 
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
 
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
 
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
+    movu        [r2 + r3 * 0 + 64], m0
+    movu        [r2 + r3 * 1 + 64], m1
+    movu        [r2 + r3 * 2 + 64], m2
+    movu        [r2 + r4 + 64], m3
 
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
+    lea         r0, [r0 + 8]
 
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
-    RET
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-INIT_YMM avx2
-cglobal filterPixelToShort_16x12, 3, 6, 2
-    mov             r3d, r3m
-    add             r3d, r3d
-    lea             r4, [r1 * 3]
-    lea             r5, [r3 * 3]
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-    ; load constant
-    vbroadcasti128  m1, [pw_2000]
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
 
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
 
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
+    movu        [r2 + r3 * 0 + 80], m0
+    movu        [r2 + r3 * 1 + 80], m1
+    movu        [r2 + r3 * 2 + 80], m2
+    movu        [r2 + r4 + 80], m3
 
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
+    lea         r0, [r0 + 8]
 
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
 
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
 
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
+    movu        [r2 + r3 * 0 + 96], m0
+    movu        [r2 + r3 * 1 + 96], m1
+    movu        [r2 + r3 * 2 + 96], m2
+    movu        [r2 + r4 + 96], m3
 
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
+    lea         r0, [r0 + 8]
 
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
 
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
 
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
+    movu        [r2 + r3 * 0 + 112], m0
+    movu        [r2 + r3 * 1 + 112], m1
+    movu        [r2 + r3 * 2 + 112], m2
+    movu        [r2 + r4 + 112], m3
+
+    lea         r0, [r0 + r1 * 4 - 56]
+    lea         r2, [r2 + r3 * 4]
+
+    dec         r6d
+    jnz         .loop
     RET
+%endmacro
+    P2S_H_64xN 64
+    P2S_H_64xN 16
+    P2S_H_64xN 32
+    P2S_H_64xN 48
 
 ;-----------------------------------------------------------------------------
 ; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
 ;-----------------------------------------------------------------------------
+%macro P2S_H_64xN_avx2 1
 INIT_YMM avx2
-cglobal filterPixelToShort_16x16, 3, 6, 2
-    mov             r3d, r3m
-    add             r3d, r3d
-    lea             r4, [r1 * 3]
-    lea             r5, [r3 * 3]
+cglobal filterPixelToShort_64x%1, 3, 7, 5
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r5, [r1 * 3]
+    lea         r6, [r3 * 3]
 
-    ; load constant
-    vbroadcasti128  m1, [pw_2000]
+    ; load height
+    mov         r4d, %1/4
 
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
+    ; load constant
+    vpbroadcastd m4, [pw_2000]
 
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
+.loop:
+    pmovzxbw    m0, [r0 + 0 * mmsize/2]
+    pmovzxbw    m1, [r0 + 1 * mmsize/2]
+    pmovzxbw    m2, [r0 + 2 * mmsize/2]
+    pmovzxbw    m3, [r0 + 3 * mmsize/2]
+    psllw       m0, 6
+    psllw       m1, 6
+    psllw       m2, 6
+    psllw       m3, 6
+    psubw       m0, m4
+    psubw       m1, m4
+    psubw       m2, m4
+    psubw       m3, m4
 
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
+    movu        [r2 + 0 * mmsize], m0
+    movu        [r2 + 1 * mmsize], m1
+    movu        [r2 + 2 * mmsize], m2
+    movu        [r2 + 3 * mmsize], m3
 
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
+    pmovzxbw    m0, [r0 + r1 + 0 * mmsize/2]
+    pmovzxbw    m1, [r0 + r1 + 1 * mmsize/2]
+    pmovzxbw    m2, [r0 + r1 + 2 * mmsize/2]
+    pmovzxbw    m3, [r0 + r1 + 3 * mmsize/2]
+    psllw       m0, 6
+    psllw       m1, 6
+    psllw       m2, 6
+    psllw       m3, 6
+    psubw       m0, m4
+    psubw       m1, m4
+    psubw       m2, m4
+    psubw       m3, m4
 
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
-
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
-
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
-
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
-
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
-
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
-
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
-
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
-
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
+    movu        [r2 + r3 + 0 * mmsize], m0
+    movu        [r2 + r3 + 1 * mmsize], m1
+    movu        [r2 + r3 + 2 * mmsize], m2
+    movu        [r2 + r3 + 3 * mmsize], m3
 
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
+    pmovzxbw    m0, [r0 + r1 * 2 + 0 * mmsize/2]
+    pmovzxbw    m1, [r0 + r1 * 2 + 1 * mmsize/2]
+    pmovzxbw    m2, [r0 + r1 * 2 + 2 * mmsize/2]
+    pmovzxbw    m3, [r0 + r1 * 2 + 3 * mmsize/2]
+    psllw       m0, 6
+    psllw       m1, 6
+    psllw       m2, 6
+    psllw       m3, 6
+    psubw       m0, m4
+    psubw       m1, m4
+    psubw       m2, m4
+    psubw       m3, m4
 
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
+    movu        [r2 + r3 * 2 + 0 * mmsize], m0
+    movu        [r2 + r3 * 2 + 1 * mmsize], m1
+    movu        [r2 + r3 * 2 + 2 * mmsize], m2
+    movu        [r2 + r3 * 2 + 3 * mmsize], m3
 
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
+    pmovzxbw    m0, [r0 + r5 + 0 * mmsize/2]
+    pmovzxbw    m1, [r0 + r5 + 1 * mmsize/2]
+    pmovzxbw    m2, [r0 + r5 + 2 * mmsize/2]
+    pmovzxbw    m3, [r0 + r5 + 3 * mmsize/2]
+    psllw       m0, 6
+    psllw       m1, 6
+    psllw       m2, 6
+    psllw       m3, 6
+    psubw       m0, m4
+    psubw       m1, m4
+    psubw       m2, m4
+    psubw       m3, m4
 
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
+    movu        [r2 + r6 + 0 * mmsize], m0
+    movu        [r2 + r6 + 1 * mmsize], m1
+    movu        [r2 + r6 + 2 * mmsize], m2
+    movu        [r2 + r6 + 3 * mmsize], m3
 
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
+    lea         r0, [r0 + r1 * 4]
+    lea         r2, [r2 + r3 * 4]
 
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
+    dec         r4d
+    jnz        .loop
     RET
+%endmacro
+    P2S_H_64xN_avx2 64
+    P2S_H_64xN_avx2 16
+    P2S_H_64xN_avx2 32
+    P2S_H_64xN_avx2 48
 
 ;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+; void filterPixelToShort(pixel src, intptr_t srcStride, int16_t dst, int16_t dstStride)
 ;-----------------------------------------------------------------------------
-INIT_YMM avx2
-cglobal filterPixelToShort_16x24, 3, 7, 2
-    mov             r3d, r3m
-    add             r3d, r3d
-    lea             r4, [r1 * 3]
-    lea             r5, [r3 * 3]
-    mov             r6d, 3
+%macro P2S_H_12xN 1
+INIT_XMM ssse3
+cglobal filterPixelToShort_12x%1, 3, 7, 6
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r4, [r1 * 3]
+    lea         r6, [r3 * 3]
+    mov         r5d, %1/4
 
     ; load constant
-    vbroadcasti128  m1, [pw_2000]
-.loop:
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
+    mova        m4, [pb_128]
+    mova        m5, [tab_c_64_n64]
 
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
+.loop:
+    movu        m0, [r0]
+    punpcklbw   m1, m0, m4
+    punpckhbw   m0, m4
+    pmaddubsw   m0, m5
+    pmaddubsw   m1, m5
 
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
+    movu        m2, [r0 + r1]
+    punpcklbw   m3, m2, m4
+    punpckhbw   m2, m4
+    pmaddubsw   m2, m5
+    pmaddubsw   m3, m5
 
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
+    movu        [r2 + r3 * 0], m1
+    movu        [r2 + r3 * 1], m3
 
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
+    movh        [r2 + r3 * 0 + 16], m0
+    movh        [r2 + r3 * 1 + 16], m2
 
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
+    movu        m0, [r0 + r1 * 2]
+    punpcklbw   m1, m0, m4
+    punpckhbw   m0, m4
+    pmaddubsw   m0, m5
+    pmaddubsw   m1, m5
 
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
+    movu        m2, [r0 + r4]
+    punpcklbw   m3, m2, m4
+    punpckhbw   m2, m4
+    pmaddubsw   m2, m5
+    pmaddubsw   m3, m5
 
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
+    movu        [r2 + r3 * 2], m1
+    movu        [r2 + r6], m3
 
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
+    movh        [r2 + r3 * 2 + 16], m0
+    movh        [r2 + r6 + 16], m2
 
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
+    lea         r0, [r0 + r1 * 4]
+    lea         r2, [r2 + r3 * 4]
 
-    dec             r6d
-    jnz             .loop
+    dec         r5d
+    jnz         .loop
     RET
+%endmacro
+    P2S_H_12xN 16
+    P2S_H_12xN 32
 
 ;-----------------------------------------------------------------------------
 ; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
 ;-----------------------------------------------------------------------------
-%macro P2S_H_16xN_avx2 1
-INIT_YMM avx2
-cglobal filterPixelToShort_16x%1, 3, 7, 2
-    mov             r3d, r3m
-    add             r3d, r3d
-    lea             r4, [r1 * 3]
-    lea             r5, [r3 * 3]
-    mov             r6d, %1/16
+%macro P2S_H_24xN 1
+INIT_XMM ssse3
+cglobal filterPixelToShort_24x%1, 3, 7, 5
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r4, [r1 * 3]
+    lea         r5, [r3 * 3]
+    mov         r6d, %1/4
 
     ; load constant
-    vbroadcasti128  m1, [pw_2000]
-.loop:
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
+    mova        m3, [pb_128]
+    mova        m4, [tab_c_64_n64]
 
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
+.loop:
+    movu        m0, [r0]
+    punpcklbw   m1, m0, m3
+    punpckhbw   m0, m3
+    pmaddubsw   m0, m4
+    pmaddubsw   m1, m4
 
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
+    movu        m2, [r0 + 16]
+    punpcklbw   m2, m3
+    pmaddubsw   m2, m4
 
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
+    movu        [r2 +  r3 * 0], m1
+    movu        [r2 +  r3 * 0 + 16], m0
+    movu        [r2 +  r3 * 0 + 32], m2
 
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
+    movu        m0, [r0 + r1]
+    punpcklbw   m1, m0, m3
+    punpckhbw   m0, m3
+    pmaddubsw   m0, m4
+    pmaddubsw   m1, m4
 
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
-
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
-
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
-
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
-
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
-
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
-
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
-
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
+    movu        m2, [r0 + r1 + 16]
+    punpcklbw   m2, m3
+    pmaddubsw   m2, m4
 
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
+    movu        [r2 +  r3 * 1], m1
+    movu        [r2 +  r3 * 1 + 16], m0
+    movu        [r2 +  r3 * 1 + 32], m2
 
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
+    movu        m0, [r0 + r1 * 2]
+    punpcklbw   m1, m0, m3
+    punpckhbw   m0, m3
+    pmaddubsw   m0, m4
+    pmaddubsw   m1, m4
 
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
+    movu        m2, [r0 + r1 * 2 + 16]
+    punpcklbw   m2, m3
+    pmaddubsw   m2, m4
 
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
+    movu        [r2 +  r3 * 2], m1
+    movu        [r2 +  r3 * 2 + 16], m0
+    movu        [r2 +  r3 * 2 + 32], m2
 
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
+    movu        m0, [r0 + r4]
+    punpcklbw   m1, m0, m3
+    punpckhbw   m0, m3
+    pmaddubsw   m0, m4
+    pmaddubsw   m1, m4
 
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
+    movu        m2, [r0 + r4 + 16]
+    punpcklbw   m2, m3
+    pmaddubsw   m2, m4
+    movu        [r2 +  r5], m1
+    movu        [r2 +  r5 + 16], m0
+    movu        [r2 +  r5 + 32], m2
 
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
+    lea         r0, [r0 + r1 * 4]
+    lea         r2, [r2 + r3 * 4]
 
-    dec             r6d
-    jnz             .loop
+    dec         r6d
+    jnz         .loop
     RET
 %endmacro
-P2S_H_16xN_avx2 32
-P2S_H_16xN_avx2 64
+    P2S_H_24xN 32
+    P2S_H_24xN 64
 
 ;-----------------------------------------------------------------------------
 ; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
 ;-----------------------------------------------------------------------------
-%macro P2S_H_32xN 1
-INIT_XMM ssse3
-cglobal filterPixelToShort_32x%1, 3, 7, 6
+%macro P2S_H_24xN_avx2 1
+INIT_YMM avx2
+cglobal filterPixelToShort_24x%1, 3, 7, 4
     mov         r3d, r3m
     add         r3d, r3d
-    lea         r4, [r3 * 3]
-    lea         r5, [r1 * 3]
-
-    ; load height
+    lea         r4, [r1 * 3]
+    lea         r5, [r3 * 3]
     mov         r6d, %1/4
 
     ; load constant
-    mova        m4, [pb_128]
-    mova        m5, [tab_c_64_n64]
+    vpbroadcastd m1, [pw_2000]
+    vpbroadcastd m2, [pb_128]
+    vpbroadcastd m3, [tab_c_64_n64]
 
 .loop:
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
-
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
-
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
-
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
-
-    movu        [r2 + r3 * 0], m0
-    movu        [r2 + r3 * 1], m1
-    movu        [r2 + r3 * 2], m2
-    movu        [r2 + r4], m3
-
-    lea         r0, [r0 + 8]
-
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
-
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
-
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
-
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
-
-    movu        [r2 + r3 * 0 + 16], m0
-    movu        [r2 + r3 * 1 + 16], m1
-    movu        [r2 + r3 * 2 + 16], m2
-    movu        [r2 + r4 + 16], m3
-
-    lea         r0, [r0 + 8]
-
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
-
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
-
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
-
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
+    pmovzxbw    m0, [r0]
+    psllw       m0, 6
+    psubw       m0, m1
+    movu        [r2], m0
 
-    movu        [r2 + r3 * 0 + 32], m0
-    movu        [r2 + r3 * 1 + 32], m1
-    movu        [r2 + r3 * 2 + 32], m2
-    movu        [r2 + r4 + 32], m3
+    movu        m0, [r0 + mmsize/2]
+    punpcklbw   m0, m2
+    pmaddubsw   m0, m3
+    movu        [r2 +  r3 * 0 + mmsize], xm0
 
-    lea         r0, [r0 + 8]
+    pmovzxbw    m0, [r0 + r1]
+    psllw       m0, 6
+    psubw       m0, m1
+    movu        [r2 + r3], m0
 
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
+    movu        m0, [r0 + r1 + mmsize/2]
+    punpcklbw   m0, m2
+    pmaddubsw   m0, m3
+    movu        [r2 +  r3 * 1 + mmsize], xm0
 
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
+    pmovzxbw    m0, [r0 + r1 * 2]
+    psllw       m0, 6
+    psubw       m0, m1
+    movu        [r2 + r3 * 2], m0
 
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
+    movu        m0, [r0 + r1 * 2 + mmsize/2]
+    punpcklbw   m0, m2
+    pmaddubsw   m0, m3
+    movu        [r2 +  r3 * 2 + mmsize], xm0
 
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
+    pmovzxbw    m0, [r0 + r4]
+    psllw       m0, 6
+    psubw       m0, m1
+    movu        [r2 + r5], m0
 
-    movu        [r2 + r3 * 0 + 48], m0
-    movu        [r2 + r3 * 1 + 48], m1
-    movu        [r2 + r3 * 2 + 48], m2
-    movu        [r2 + r4 + 48], m3
+    movu        m0, [r0 + r4 + mmsize/2]
+    punpcklbw   m0, m2
+    pmaddubsw   m0, m3
+    movu        [r2 + r5 + mmsize], xm0
 
-    lea         r0, [r0 + r1 * 4 - 24]
+    lea         r0, [r0 + r1 * 4]
     lea         r2, [r2 + r3 * 4]
 
     dec         r6d
     jnz         .loop
     RET
 %endmacro
-    P2S_H_32xN 32
-    P2S_H_32xN 8
-    P2S_H_32xN 16
-    P2S_H_32xN 24
-    P2S_H_32xN 64
-    P2S_H_32xN 48
+    P2S_H_24xN_avx2 32
+    P2S_H_24xN_avx2 64
 
 ;-----------------------------------------------------------------------------
 ; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
 ;-----------------------------------------------------------------------------
-%macro P2S_H_32xN_avx2 1
-INIT_YMM avx2
-cglobal filterPixelToShort_32x%1, 3, 7, 3
+INIT_XMM ssse3
+cglobal filterPixelToShort_48x64, 3, 7, 4
     mov         r3d, r3m
     add         r3d, r3d
-    lea         r5, [r1 * 3]
-    lea         r6, [r3 * 3]
-
-    ; load height
-    mov         r4d, %1/4
+    lea         r4, [r1 * 3]
+    lea         r5, [r3 * 3]
+    mov         r6d, 16
 
     ; load constant
-    vpbroadcastd m2, [pw_2000]
+    mova        m2, [pb_128]
+    mova        m3, [tab_c_64_n64]
 
+.loop:
+    movu        m0, [r0]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
+
+    movu        [r2 +  r3 * 0], m1
+    movu        [r2 +  r3 * 0 + 16], m0
+
+    movu        m0, [r0 + 16]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
+
+    movu        [r2 +  r3 * 0 + 32], m1
+    movu        [r2 +  r3 * 0 + 48], m0
+
+    movu        m0, [r0 + 32]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
+
+    movu        [r2 +  r3 * 0 + 64], m1
+    movu        [r2 +  r3 * 0 + 80], m0
+
+    movu        m0, [r0 + r1]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
+
+    movu        [r2 +  r3 * 1], m1
+    movu        [r2 +  r3 * 1 + 16], m0
+
+    movu        m0, [r0 + r1 + 16]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
+
+    movu        [r2 +  r3 * 1 + 32], m1
+    movu        [r2 +  r3 * 1 + 48], m0
+
+    movu        m0, [r0 + r1 + 32]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
+
+    movu        [r2 +  r3 * 1 + 64], m1
+    movu        [r2 +  r3 * 1 + 80], m0
+
+    movu        m0, [r0 + r1 * 2]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
+
+    movu        [r2 +  r3 * 2], m1
+    movu        [r2 +  r3 * 2 + 16], m0
+
+    movu        m0, [r0 + r1 * 2 + 16]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
+
+    movu        [r2 +  r3 * 2 + 32], m1
+    movu        [r2 +  r3 * 2 + 48], m0
+
+    movu        m0, [r0 + r1 * 2 + 32]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
+
+    movu        [r2 +  r3 * 2 + 64], m1
+    movu        [r2 +  r3 * 2 + 80], m0
+
+    movu        m0, [r0 + r4]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
+
+    movu        [r2 +  r5], m1
+    movu        [r2 +  r5 + 16], m0
+
+    movu        m0, [r0 + r4 + 16]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
+
+    movu        [r2 +  r5 + 32], m1
+    movu        [r2 +  r5 + 48], m0
+
+    movu        m0, [r0 + r4 + 32]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
+
+    movu        [r2 +  r5 + 64], m1
+    movu        [r2 +  r5 + 80], m0
+
+    lea         r0, [r0 + r1 * 4]
+    lea         r2, [r2 + r3 * 4]
+
+    dec         r6d
+    jnz         .loop
+    RET
+
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+INIT_YMM avx2
+cglobal filterPixelToShort_48x64, 3,7,4
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r5, [r1 * 3]
+    lea         r6, [r3 * 3]
+
+    ; load height
+    mov         r4d, 64/4
+
+    ; load constant
+    vpbroadcastd m3, [pw_2000]
+
+    ; just unroll(1) because it is best choice for 48x64
 .loop:
     pmovzxbw    m0, [r0 + 0 * mmsize/2]
     pmovzxbw    m1, [r0 + 1 * mmsize/2]
+    pmovzxbw    m2, [r0 + 2 * mmsize/2]
     psllw       m0, 6
     psllw       m1, 6
-    psubw       m0, m2
-    psubw       m1, m2
+    psllw       m2, 6
+    psubw       m0, m3
+    psubw       m1, m3
+    psubw       m2, m3
     movu        [r2 + 0 * mmsize], m0
     movu        [r2 + 1 * mmsize], m1
+    movu        [r2 + 2 * mmsize], m2
 
     pmovzxbw    m0, [r0 + r1 + 0 * mmsize/2]
     pmovzxbw    m1, [r0 + r1 + 1 * mmsize/2]
+    pmovzxbw    m2, [r0 + r1 + 2 * mmsize/2]
     psllw       m0, 6
     psllw       m1, 6
-    psubw       m0, m2
-    psubw       m1, m2
+    psllw       m2, 6
+    psubw       m0, m3
+    psubw       m1, m3
+    psubw       m2, m3
     movu        [r2 + r3 + 0 * mmsize], m0
     movu        [r2 + r3 + 1 * mmsize], m1
+    movu        [r2 + r3 + 2 * mmsize], m2
 
     pmovzxbw    m0, [r0 + r1 * 2 + 0 * mmsize/2]
     pmovzxbw    m1, [r0 + r1 * 2 + 1 * mmsize/2]
+    pmovzxbw    m2, [r0 + r1 * 2 + 2 * mmsize/2]
     psllw       m0, 6
     psllw       m1, 6
-    psubw       m0, m2
-    psubw       m1, m2
+    psllw       m2, 6
+    psubw       m0, m3
+    psubw       m1, m3
+    psubw       m2, m3
     movu        [r2 + r3 * 2 + 0 * mmsize], m0
     movu        [r2 + r3 * 2 + 1 * mmsize], m1
+    movu        [r2 + r3 * 2 + 2 * mmsize], m2
 
     pmovzxbw    m0, [r0 + r5 + 0 * mmsize/2]
     pmovzxbw    m1, [r0 + r5 + 1 * mmsize/2]
+    pmovzxbw    m2, [r0 + r5 + 2 * mmsize/2]
     psllw       m0, 6
     psllw       m1, 6
-    psubw       m0, m2
-    psubw       m1, m2
+    psllw       m2, 6
+    psubw       m0, m3
+    psubw       m1, m3
+    psubw       m2, m3
     movu        [r2 + r6 + 0 * mmsize], m0
     movu        [r2 + r6 + 1 * mmsize], m1
+    movu        [r2 + r6 + 2 * mmsize], m2
 
     lea         r0, [r0 + r1 * 4]
     lea         r2, [r2 + r3 * 4]
@@ -13043,1014 +11002,1403 @@ cglobal filterPixelToShort_32x%1, 3, 7, 3
     dec         r4d
     jnz        .loop
     RET
-%endmacro
-    P2S_H_32xN_avx2 32
-    P2S_H_32xN_avx2 8
-    P2S_H_32xN_avx2 16
-    P2S_H_32xN_avx2 24
-    P2S_H_32xN_avx2 64
-    P2S_H_32xN_avx2 48
 
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-%macro P2S_H_64xN 1
-INIT_XMM ssse3
-cglobal filterPixelToShort_64x%1, 3, 7, 6
-    mov         r3d, r3m
-    add         r3d, r3d
-    lea         r4, [r3 * 3]
-    lea         r5, [r1 * 3]
 
-    ; load height
-    mov         r6d, %1/4
+%macro PROCESS_LUMA_W4_4R 0
+    movd        m0, [r0]
+    movd        m1, [r0 + r1]
+    punpcklbw   m2, m0, m1                     ; m2=[0 1]
 
-    ; load constant
-    mova        m4, [pb_128]
-    mova        m5, [tab_c_64_n64]
+    lea         r0, [r0 + 2 * r1]
+    movd        m0, [r0]
+    punpcklbw   m1, m0                         ; m1=[1 2]
+    punpcklqdq  m2, m1                         ; m2=[0 1 1 2]
+    pmaddubsw   m4, m2, [r6 + 0 * 16]          ; m4=[0+1 1+2]
 
-.loop:
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
+    movd        m1, [r0 + r1]
+    punpcklbw   m5, m0, m1                     ; m2=[2 3]
+    lea         r0, [r0 + 2 * r1]
+    movd        m0, [r0]
+    punpcklbw   m1, m0                         ; m1=[3 4]
+    punpcklqdq  m5, m1                         ; m5=[2 3 3 4]
+    pmaddubsw   m2, m5, [r6 + 1 * 16]          ; m2=[2+3 3+4]
+    paddw       m4, m2                         ; m4=[0+1+2+3 1+2+3+4]                   Row1-2
+    pmaddubsw   m5, [r6 + 0 * 16]              ; m5=[2+3 3+4]                           Row3-4
 
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
+    movd        m1, [r0 + r1]
+    punpcklbw   m2, m0, m1                     ; m2=[4 5]
+    lea         r0, [r0 + 2 * r1]
+    movd        m0, [r0]
+    punpcklbw   m1, m0                         ; m1=[5 6]
+    punpcklqdq  m2, m1                         ; m2=[4 5 5 6]
+    pmaddubsw   m1, m2, [r6 + 2 * 16]          ; m1=[4+5 5+6]
+    paddw       m4, m1                         ; m4=[0+1+2+3+4+5 1+2+3+4+5+6]           Row1-2
+    pmaddubsw   m2, [r6 + 1 * 16]              ; m2=[4+5 5+6]
+    paddw       m5, m2                         ; m5=[2+3+4+5 3+4+5+6]                   Row3-4
 
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
+    movd        m1, [r0 + r1]
+    punpcklbw   m2, m0, m1                     ; m2=[6 7]
+    lea         r0, [r0 + 2 * r1]
+    movd        m0, [r0]
+    punpcklbw   m1, m0                         ; m1=[7 8]
+    punpcklqdq  m2, m1                         ; m2=[6 7 7 8]
+    pmaddubsw   m1, m2, [r6 + 3 * 16]          ; m1=[6+7 7+8]
+    paddw       m4, m1                         ; m4=[0+1+2+3+4+5+6+7 1+2+3+4+5+6+7+8]   Row1-2 end
+    pmaddubsw   m2, [r6 + 2 * 16]              ; m2=[6+7 7+8]
+    paddw       m5, m2                         ; m5=[2+3+4+5+6+7 3+4+5+6+7+8]           Row3-4
 
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
+    movd        m1, [r0 + r1]
+    punpcklbw   m2, m0, m1                     ; m2=[8 9]
+    movd        m0, [r0 + 2 * r1]
+    punpcklbw   m1, m0                         ; m1=[9 10]
+    punpcklqdq  m2, m1                         ; m2=[8 9 9 10]
+    pmaddubsw   m2, [r6 + 3 * 16]              ; m2=[8+9 9+10]
+    paddw       m5, m2                         ; m5=[2+3+4+5+6+7+8+9 3+4+5+6+7+8+9+10]  Row3-4 end
+%endmacro
 
-    movu        [r2 + r3 * 0], m0
-    movu        [r2 + r3 * 1], m1
-    movu        [r2 + r3 * 2], m2
-    movu        [r2 + r4], m3
+%macro PROCESS_LUMA_W8_4R 0
+    movq       m0, [r0]
+    movq       m1, [r0 + r1]
+    punpcklbw  m0, m1
+    pmaddubsw  m7, m0, [r6 + 0 *16]            ;m7=[0+1]               Row1
 
-    lea         r0, [r0 + 8]
-
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
-
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
-
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
+    lea        r0, [r0 + 2 * r1]
+    movq       m0, [r0]
+    punpcklbw  m1, m0
+    pmaddubsw  m6, m1, [r6 + 0 *16]            ;m6=[1+2]               Row2
 
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
+    movq       m1, [r0 + r1]
+    punpcklbw  m0, m1
+    pmaddubsw  m5, m0, [r6 + 0 *16]            ;m5=[2+3]               Row3
+    pmaddubsw  m0, [r6 + 1 * 16]
+    paddw      m7, m0                          ;m7=[0+1+2+3]           Row1
 
-    movu        [r2 + r3 * 0 + 16], m0
-    movu        [r2 + r3 * 1 + 16], m1
-    movu        [r2 + r3 * 2 + 16], m2
-    movu        [r2 + r4 + 16], m3
+    lea        r0, [r0 + 2 * r1]
+    movq       m0, [r0]
+    punpcklbw  m1, m0
+    pmaddubsw  m4, m1, [r6 + 0 *16]            ;m4=[3+4]               Row4
+    pmaddubsw  m1, [r6 + 1 * 16]
+    paddw      m6, m1                          ;m6 = [1+2+3+4]         Row2
 
-    lea         r0, [r0 + 8]
+    movq       m1, [r0 + r1]
+    punpcklbw  m0, m1
+    pmaddubsw  m2, m0, [r6 + 1 * 16]
+    pmaddubsw  m0, [r6 + 2 * 16]
+    paddw      m7, m0                          ;m7=[0+1+2+3+4+5]       Row1
+    paddw      m5, m2                          ;m5=[2+3+4+5]           Row3
 
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
+    lea        r0, [r0 + 2 * r1]
+    movq       m0, [r0]
+    punpcklbw  m1, m0
+    pmaddubsw  m2, m1, [r6 + 1 * 16]
+    pmaddubsw  m1, [r6 + 2 * 16]
+    paddw      m6, m1                          ;m6=[1+2+3+4+5+6]       Row2
+    paddw      m4, m2                          ;m4=[3+4+5+6]           Row4
 
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
+    movq       m1, [r0 + r1]
+    punpcklbw  m0, m1
+    pmaddubsw  m2, m0, [r6 + 2 * 16]
+    pmaddubsw  m0, [r6 + 3 * 16]
+    paddw      m7, m0                          ;m7=[0+1+2+3+4+5+6+7]   Row1 end
+    paddw      m5, m2                          ;m5=[2+3+4+5+6+7]       Row3
 
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
+    lea        r0, [r0 + 2 * r1]
+    movq       m0, [r0]
+    punpcklbw  m1, m0
+    pmaddubsw  m2, m1, [r6 + 2 * 16]
+    pmaddubsw  m1, [r6 + 3 * 16]
+    paddw      m6, m1                          ;m6=[1+2+3+4+5+6+7+8]   Row2 end
+    paddw      m4, m2                          ;m4=[3+4+5+6+7+8]       Row4
 
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
+    movq       m1, [r0 + r1]
+    punpcklbw  m0, m1
+    pmaddubsw  m0, [r6 + 3 * 16]
+    paddw      m5, m0                          ;m5=[2+3+4+5+6+7+8+9]   Row3 end
 
-    movu        [r2 + r3 * 0 + 32], m0
-    movu        [r2 + r3 * 1 + 32], m1
-    movu        [r2 + r3 * 2 + 32], m2
-    movu        [r2 + r4 + 32], m3
+    movq       m0, [r0 + 2 * r1]
+    punpcklbw  m1, m0
+    pmaddubsw  m1, [r6 + 3 * 16]
+    paddw      m4, m1                          ;m4=[3+4+5+6+7+8+9+10]  Row4 end
+%endmacro
 
-    lea         r0, [r0 + 8]
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_%3_4x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_LUMA_4xN 3
+INIT_XMM sse4
+cglobal interp_8tap_vert_%3_%1x%2, 5, 7, 6
+    lea       r5, [3 * r1]
+    sub       r0, r5
+    shl       r4d, 6
+%ifidn %3,ps
+    add       r3d, r3d
+%endif
 
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
+%ifdef PIC
+    lea       r5, [tab_LumaCoeffVer]
+    lea       r6, [r5 + r4]
+%else
+    lea       r6, [tab_LumaCoeffVer + r4]
+%endif
 
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
+%ifidn %3,pp
+    mova      m3, [pw_512]
+%else
+    mova      m3, [pw_2000]
+%endif
 
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
+    mov       r4d, %2/4
+    lea       r5, [4 * r1]
 
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
+.loopH:
+    PROCESS_LUMA_W4_4R
 
-    movu        [r2 + r3 * 0 + 48], m0
-    movu        [r2 + r3 * 1 + 48], m1
-    movu        [r2 + r3 * 2 + 48], m2
-    movu        [r2 + r4 + 48], m3
+%ifidn %3,pp
+    pmulhrsw  m4, m3
+    pmulhrsw  m5, m3
 
-    lea         r0, [r0 + 8]
+    packuswb  m4, m5
 
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
+    movd      [r2], m4
+    pextrd    [r2 + r3], m4, 1
+    lea       r2, [r2 + 2 * r3]
+    pextrd    [r2], m4, 2
+    pextrd    [r2 + r3], m4, 3
+%else
+    psubw     m4, m3
+    psubw     m5, m3
 
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
+    movlps    [r2], m4
+    movhps    [r2 + r3], m4
+    lea       r2, [r2 + 2 * r3]
+    movlps    [r2], m5
+    movhps    [r2 + r3], m5
+%endif
 
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
+    sub       r0, r5
+    lea       r2, [r2 + 2 * r3]
 
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
+    dec       r4d
+    jnz       .loopH
 
-    movu        [r2 + r3 * 0 + 64], m0
-    movu        [r2 + r3 * 1 + 64], m1
-    movu        [r2 + r3 * 2 + 64], m2
-    movu        [r2 + r4 + 64], m3
+    RET
+%endmacro
 
-    lea         r0, [r0 + 8]
 
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
+INIT_YMM avx2
+cglobal interp_8tap_vert_pp_4x4, 4,6,8
+    mov             r4d, r4m
+    lea             r5, [r1 * 3]
+    sub             r0, r5
 
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
+    ; TODO: VPGATHERDD
+    movd            xm1, [r0]                       ; m1 = row0
+    movd            xm2, [r0 + r1]                  ; m2 = row1
+    punpcklbw       xm1, xm2                        ; m1 = [13 03 12 02 11 01 10 00]
 
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
+    movd            xm3, [r0 + r1 * 2]              ; m3 = row2
+    punpcklbw       xm2, xm3                        ; m2 = [23 13 22 12 21 11 20 10]
+    movd            xm4, [r0 + r5]
+    punpcklbw       xm3, xm4                        ; m3 = [33 23 32 22 31 21 30 20]
+    punpcklwd       xm1, xm3                        ; m1 = [33 23 13 03 32 22 12 02 31 21 11 01 30 20 10 00]
 
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
+    lea             r0, [r0 + r1 * 4]
+    movd            xm5, [r0]                       ; m5 = row4
+    punpcklbw       xm4, xm5                        ; m4 = [43 33 42 32 41 31 40 30]
+    punpcklwd       xm2, xm4                        ; m2 = [43 33 21 13 42 32 22 12 41 31 21 11 40 30 20 10]
+    vinserti128     m1, m1, xm2, 1                  ; m1 = [43 33 21 13 42 32 22 12 41 31 21 11 40 30 20 10] - [33 23 13 03 32 22 12 02 31 21 11 01 30 20 10 00]
+    movd            xm2, [r0 + r1]                  ; m2 = row5
+    punpcklbw       xm5, xm2                        ; m5 = [53 43 52 42 51 41 50 40]
+    punpcklwd       xm3, xm5                        ; m3 = [53 43 44 23 52 42 32 22 51 41 31 21 50 40 30 20]
+    movd            xm6, [r0 + r1 * 2]              ; m6 = row6
+    punpcklbw       xm2, xm6                        ; m2 = [63 53 62 52 61 51 60 50]
+    punpcklwd       xm4, xm2                        ; m4 = [63 53 43 33 62 52 42 32 61 51 41 31 60 50 40 30]
+    vinserti128     m3, m3, xm4, 1                  ; m3 = [63 53 43 33 62 52 42 32 61 51 41 31 60 50 40 30] - [53 43 44 23 52 42 32 22 51 41 31 21 50 40 30 20]
+    movd            xm4, [r0 + r5]                  ; m4 = row7
+    punpcklbw       xm6, xm4                        ; m6 = [73 63 72 62 71 61 70 60]
+    punpcklwd       xm5, xm6                        ; m5 = [73 63 53 43 72 62 52 42 71 61 51 41 70 60 50 40]
 
-    movu        [r2 + r3 * 0 + 80], m0
-    movu        [r2 + r3 * 1 + 80], m1
-    movu        [r2 + r3 * 2 + 80], m2
-    movu        [r2 + r4 + 80], m3
+    lea             r0, [r0 + r1 * 4]
+    movd            xm7, [r0]                       ; m7 = row8
+    punpcklbw       xm4, xm7                        ; m4 = [83 73 82 72 81 71 80 70]
+    punpcklwd       xm2, xm4                        ; m2 = [83 73 63 53 82 72 62 52 81 71 61 51 80 70 60 50]
+    vinserti128     m5, m5, xm2, 1                  ; m5 = [83 73 63 53 82 72 62 52 81 71 61 51 80 70 60 50] - [73 63 53 43 72 62 52 42 71 61 51 41 70 60 50 40]
+    movd            xm2, [r0 + r1]                  ; m2 = row9
+    punpcklbw       xm7, xm2                        ; m7 = [93 83 92 82 91 81 90 80]
+    punpcklwd       xm6, xm7                        ; m6 = [93 83 73 63 92 82 72 62 91 81 71 61 90 80 70 60]
+    movd            xm7, [r0 + r1 * 2]              ; m7 = rowA
+    punpcklbw       xm2, xm7                        ; m2 = [A3 93 A2 92 A1 91 A0 90]
+    punpcklwd       xm4, xm2                        ; m4 = [A3 93 83 73 A2 92 82 72 A1 91 81 71 A0 90 80 70]
+    vinserti128     m6, m6, xm4, 1                  ; m6 = [A3 93 83 73 A2 92 82 72 A1 91 81 71 A0 90 80 70] - [93 83 73 63 92 82 72 62 91 81 71 61 90 80 70 60]
 
-    lea         r0, [r0 + 8]
+    ; load filter coeff
+%ifdef PIC
+    lea             r5, [tab_LumaCoeff]
+    vpbroadcastd    m0, [r5 + r4 * 8 + 0]
+    vpbroadcastd    m2, [r5 + r4 * 8 + 4]
+%else
+    vpbroadcastd    m0, [tab_LumaCoeff + r4 * 8 + 0]
+    vpbroadcastd    m2, [tab_LumaCoeff + r4 * 8 + 4]
+%endif
 
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
+    pmaddubsw       m1, m0
+    pmaddubsw       m3, m0
+    pmaddubsw       m5, m2
+    pmaddubsw       m6, m2
+    vbroadcasti128  m0, [pw_1]
+    pmaddwd         m1, m0
+    pmaddwd         m3, m0
+    pmaddwd         m5, m0
+    pmaddwd         m6, m0
+    paddd           m1, m5                          ; m1 = DQWORD ROW[1 0]
+    paddd           m3, m6                          ; m3 = DQWORD ROW[3 2]
+    packssdw        m1, m3                          ; m1 =  QWORD ROW[3 1 2 0]
 
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
+    ; TODO: does it overflow?
+    pmulhrsw        m1, [pw_512]
+    vextracti128    xm2, m1, 1
+    packuswb        xm1, xm2                        ; m1 =  DWORD ROW[3 1 2 0]
+    movd            [r2], xm1
+    pextrd          [r2 + r3], xm1, 2
+    pextrd          [r2 + r3 * 2], xm1, 1
+    lea             r4, [r3 * 3]
+    pextrd          [r2 + r4], xm1, 3
+    RET
 
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
+INIT_YMM avx2
+cglobal interp_8tap_vert_ps_4x4, 4, 6, 5
+    mov             r4d, r4m
+    shl             r4d, 7
 
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
+%ifdef PIC
+    lea             r5, [tab_LumaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
+%endif
 
-    movu        [r2 + r3 * 0 + 96], m0
-    movu        [r2 + r3 * 1 + 96], m1
-    movu        [r2 + r3 * 2 + 96], m2
-    movu        [r2 + r4 + 96], m3
+    lea             r4, [r1 * 3]
+    sub             r0, r4
 
-    lea         r0, [r0 + 8]
+    add             r3d, r3d
 
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
+    movd            xm1, [r0]
+    pinsrd          xm1, [r0 + r1], 1
+    pinsrd          xm1, [r0 + r1 * 2], 2
+    pinsrd          xm1, [r0 + r4], 3                       ; m1 = row[3 2 1 0]
+    lea             r0, [r0 + r1 * 4]
+    movd            xm2, [r0]
+    pinsrd          xm2, [r0 + r1], 1
+    pinsrd          xm2, [r0 + r1 * 2], 2
+    pinsrd          xm2, [r0 + r4], 3                       ; m2 = row[7 6 5 4]
+    vinserti128     m1, m1, xm2, 1                          ; m1 = row[7 6 5 4 3 2 1 0]
+    lea             r0, [r0 + r1 * 4]
+    movd            xm3, [r0]
+    pinsrd          xm3, [r0 + r1], 1
+    pinsrd          xm3, [r0 + r1 * 2], 2                   ; m3 = row[x 10 9 8]
+    vinserti128     m2, m2, xm3, 1                          ; m2 = row[x 10 9 8 7 6 5 4]
+    mova            m3, [interp4_vpp_shuf1]
+    vpermd          m0, m3, m1                              ; m0 = row[4 3 3 2 2 1 1 0]
+    vpermd          m4, m3, m2                              ; m4 = row[8 7 7 6 6 5 5 4]
+    mova            m3, [interp4_vpp_shuf1 + mmsize]
+    vpermd          m1, m3, m1                              ; m1 = row[6 5 5 4 4 3 3 2]
+    vpermd          m2, m3, m2                              ; m2 = row[10 9 9 8 8 7 7 6]
 
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
+    mova            m3, [interp4_vpp_shuf]
+    pshufb          m0, m0, m3
+    pshufb          m1, m1, m3
+    pshufb          m4, m4, m3
+    pshufb          m2, m2, m3
+    pmaddubsw       m0, [r5]
+    pmaddubsw       m1, [r5 + mmsize]
+    pmaddubsw       m4, [r5 + 2 * mmsize]
+    pmaddubsw       m2, [r5 + 3 * mmsize]
+    paddw           m0, m1
+    paddw           m0, m4
+    paddw           m0, m2                                  ; m0 = WORD ROW[3 2 1 0]
 
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
+    psubw           m0, [pw_2000]
+    vextracti128    xm2, m0, 1
+    lea             r5, [r3 * 3]
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r5], xm2
+    RET
 
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
+%macro FILTER_VER_LUMA_AVX2_4xN 3
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%3_%1x%2, 4, 9, 10
+    mov             r4d, r4m
+    shl             r4d, 7
 
-    movu        [r2 + r3 * 0 + 112], m0
-    movu        [r2 + r3 * 1 + 112], m1
-    movu        [r2 + r3 * 2 + 112], m2
-    movu        [r2 + r4 + 112], m3
+%ifdef PIC
+    lea             r5, [tab_LumaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
+%endif
 
-    lea         r0, [r0 + r1 * 4 - 56]
-    lea         r2, [r2 + r3 * 4]
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+    lea             r6, [r1 * 4]
+%ifidn %3,pp
+    mova            m6, [pw_512]
+%else
+    add             r3d, r3d
+    vbroadcasti128  m6, [pw_2000]
+%endif
+    lea             r8, [r3 * 3]
+    mova            m5, [interp4_vpp_shuf]
+    mova            m0, [interp4_vpp_shuf1]
+    mova            m7, [interp4_vpp_shuf1 + mmsize]
+    mov             r7d, %2 / 8
+.loop:
+    movd            xm1, [r0]
+    pinsrd          xm1, [r0 + r1], 1
+    pinsrd          xm1, [r0 + r1 * 2], 2
+    pinsrd          xm1, [r0 + r4], 3                       ; m1 = row[3 2 1 0]
+    lea             r0, [r0 + r1 * 4]
+    movd            xm2, [r0]
+    pinsrd          xm2, [r0 + r1], 1
+    pinsrd          xm2, [r0 + r1 * 2], 2
+    pinsrd          xm2, [r0 + r4], 3                       ; m2 = row[7 6 5 4]
+    vinserti128     m1, m1, xm2, 1                          ; m1 = row[7 6 5 4 3 2 1 0]
+    lea             r0, [r0 + r1 * 4]
+    movd            xm3, [r0]
+    pinsrd          xm3, [r0 + r1], 1
+    pinsrd          xm3, [r0 + r1 * 2], 2
+    pinsrd          xm3, [r0 + r4], 3                       ; m3 = row[11 10 9 8]
+    vinserti128     m2, m2, xm3, 1                          ; m2 = row[11 10 9 8 7 6 5 4]
+    lea             r0, [r0 + r1 * 4]
+    movd            xm4, [r0]
+    pinsrd          xm4, [r0 + r1], 1
+    pinsrd          xm4, [r0 + r1 * 2], 2                   ; m4 = row[x 14 13 12]
+    vinserti128     m3, m3, xm4, 1                          ; m3 = row[x 14 13 12 11 10 9 8]
+    vpermd          m8, m0, m1                              ; m8 = row[4 3 3 2 2 1 1 0]
+    vpermd          m4, m0, m2                              ; m4 = row[8 7 7 6 6 5 5 4]
+    vpermd          m1, m7, m1                              ; m1 = row[6 5 5 4 4 3 3 2]
+    vpermd          m2, m7, m2                              ; m2 = row[10 9 9 8 8 7 7 6]
+    vpermd          m9, m0, m3                              ; m9 = row[12 11 11 10 10 9 9 8]
+    vpermd          m3, m7, m3                              ; m3 = row[14 13 13 12 12 11 11 10]
 
-    dec         r6d
-    jnz         .loop
+    pshufb          m8, m8, m5
+    pshufb          m1, m1, m5
+    pshufb          m4, m4, m5
+    pshufb          m9, m9, m5
+    pshufb          m2, m2, m5
+    pshufb          m3, m3, m5
+    pmaddubsw       m8, [r5]
+    pmaddubsw       m1, [r5 + mmsize]
+    pmaddubsw       m9, [r5 + 2 * mmsize]
+    pmaddubsw       m3, [r5 + 3 * mmsize]
+    paddw           m8, m1
+    paddw           m9, m3
+    pmaddubsw       m1, m4, [r5 + 2 * mmsize]
+    pmaddubsw       m3, m2, [r5 + 3 * mmsize]
+    pmaddubsw       m4, [r5]
+    pmaddubsw       m2, [r5 + mmsize]
+    paddw           m3, m1
+    paddw           m2, m4
+    paddw           m8, m3                                  ; m8 = WORD ROW[3 2 1 0]
+    paddw           m9, m2                                  ; m9 = WORD ROW[7 6 5 4]
+
+%ifidn %3,pp
+    pmulhrsw        m8, m6
+    pmulhrsw        m9, m6
+    packuswb        m8, m9
+    vextracti128    xm1, m8, 1
+    movd            [r2], xm8
+    pextrd          [r2 + r3], xm8, 1
+    movd            [r2 + r3 * 2], xm1
+    pextrd          [r2 + r8], xm1, 1
+    lea             r2, [r2 + r3 * 4]
+    pextrd          [r2], xm8, 2
+    pextrd          [r2 + r3], xm8, 3
+    pextrd          [r2 + r3 * 2], xm1, 2
+    pextrd          [r2 + r8], xm1, 3
+%else
+    psubw           m8, m6
+    psubw           m9, m6
+    vextracti128    xm1, m8, 1
+    vextracti128    xm2, m9, 1
+    movq            [r2], xm8
+    movhps          [r2 + r3], xm8
+    movq            [r2 + r3 * 2], xm1
+    movhps          [r2 + r8], xm1
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm9
+    movhps          [r2 + r3], xm9
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r8], xm2
+%endif
+    lea             r2, [r2 + r3 * 4]
+    sub             r0, r6
+    dec             r7d
+    jnz             .loop
     RET
+%endif
 %endmacro
-    P2S_H_64xN 64
-    P2S_H_64xN 16
-    P2S_H_64xN 32
-    P2S_H_64xN 48
 
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-%macro P2S_H_64xN_avx2 1
-INIT_YMM avx2
-cglobal filterPixelToShort_64x%1, 3, 7, 5
-    mov         r3d, r3m
-    add         r3d, r3d
-    lea         r5, [r1 * 3]
-    lea         r6, [r3 * 3]
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_pp_4x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_4xN 4, 4, pp
 
-    ; load height
-    mov         r4d, %1/4
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_pp_4x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_4xN 4, 8, pp
+    FILTER_VER_LUMA_AVX2_4xN 4, 8, pp
 
-    ; load constant
-    vpbroadcastd m4, [pw_2000]
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_pp_4x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_4xN 4, 16, pp
+    FILTER_VER_LUMA_AVX2_4xN 4, 16, pp
 
-.loop:
-    pmovzxbw    m0, [r0 + 0 * mmsize/2]
-    pmovzxbw    m1, [r0 + 1 * mmsize/2]
-    pmovzxbw    m2, [r0 + 2 * mmsize/2]
-    pmovzxbw    m3, [r0 + 3 * mmsize/2]
-    psllw       m0, 6
-    psllw       m1, 6
-    psllw       m2, 6
-    psllw       m3, 6
-    psubw       m0, m4
-    psubw       m1, m4
-    psubw       m2, m4
-    psubw       m3, m4
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_ps_4x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_4xN 4, 4, ps
 
-    movu        [r2 + 0 * mmsize], m0
-    movu        [r2 + 1 * mmsize], m1
-    movu        [r2 + 2 * mmsize], m2
-    movu        [r2 + 3 * mmsize], m3
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_ps_4x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_4xN 4, 8, ps
+    FILTER_VER_LUMA_AVX2_4xN 4, 8, ps
 
-    pmovzxbw    m0, [r0 + r1 + 0 * mmsize/2]
-    pmovzxbw    m1, [r0 + r1 + 1 * mmsize/2]
-    pmovzxbw    m2, [r0 + r1 + 2 * mmsize/2]
-    pmovzxbw    m3, [r0 + r1 + 3 * mmsize/2]
-    psllw       m0, 6
-    psllw       m1, 6
-    psllw       m2, 6
-    psllw       m3, 6
-    psubw       m0, m4
-    psubw       m1, m4
-    psubw       m2, m4
-    psubw       m3, m4
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_ps_4x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_4xN 4, 16, ps
+    FILTER_VER_LUMA_AVX2_4xN 4, 16, ps
 
-    movu        [r2 + r3 + 0 * mmsize], m0
-    movu        [r2 + r3 + 1 * mmsize], m1
-    movu        [r2 + r3 + 2 * mmsize], m2
-    movu        [r2 + r3 + 3 * mmsize], m3
+%macro PROCESS_LUMA_AVX2_W8_8R 0
+    movq            xm1, [r0]                       ; m1 = row 0
+    movq            xm2, [r0 + r1]                  ; m2 = row 1
+    punpcklbw       xm1, xm2                        ; m1 = [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
+    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
+    punpcklbw       xm2, xm3                        ; m2 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10]
+    vinserti128     m5, m1, xm2, 1                  ; m5 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10] - [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
+    pmaddubsw       m5, [r5]
+    movq            xm4, [r0 + r4]                  ; m4 = row 3
+    punpcklbw       xm3, xm4                        ; m3 = [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
+    lea             r0, [r0 + r1 * 4]
+    movq            xm1, [r0]                       ; m1 = row 4
+    punpcklbw       xm4, xm1                        ; m4 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30]
+    vinserti128     m2, m3, xm4, 1                  ; m2 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30] - [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
+    pmaddubsw       m0, m2, [r5 + 1 * mmsize]
+    paddw           m5, m0
+    pmaddubsw       m2, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 5
+    punpcklbw       xm1, xm3                        ; m1 = [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
+    movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
+    punpcklbw       xm3, xm4                        ; m3 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50]
+    vinserti128     m1, m1, xm3, 1                  ; m1 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50] - [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
+    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
+    paddw           m5, m3
+    pmaddubsw       m0, m1, [r5 + 1 * mmsize]
+    paddw           m2, m0
+    pmaddubsw       m1, [r5]
+    movq            xm3, [r0 + r4]                  ; m3 = row 7
+    punpcklbw       xm4, xm3                        ; m4 = [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
+    lea             r0, [r0 + r1 * 4]
+    movq            xm0, [r0]                       ; m0 = row 8
+    punpcklbw       xm3, xm0                        ; m3 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70]
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70] - [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
+    pmaddubsw       m3, m4, [r5 + 3 * mmsize]
+    paddw           m5, m3
+    pmaddubsw       m3, m4, [r5 + 2 * mmsize]
+    paddw           m2, m3
+    pmaddubsw       m3, m4, [r5 + 1 * mmsize]
+    paddw           m1, m3
+    pmaddubsw       m4, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 9
+    punpcklbw       xm0, xm3                        ; m0 = [97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80]
+    movq            xm6, [r0 + r1 * 2]              ; m6 = row 10
+    punpcklbw       xm3, xm6                        ; m3 = [A7 97 A6 96 A5 95 A4 94 A3 93 A2 92 A1 91 A0 90]
+    vinserti128     m0, m0, xm3, 1                  ; m0 = [A7 97 A6 96 A5 95 A4 94 A3 93 A2 92 A1 91 A0 90] - [97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80]
+    pmaddubsw       m3, m0, [r5 + 3 * mmsize]
+    paddw           m2, m3
+    pmaddubsw       m3, m0, [r5 + 2 * mmsize]
+    paddw           m1, m3
+    pmaddubsw       m0, [r5 + 1 * mmsize]
+    paddw           m4, m0
 
-    pmovzxbw    m0, [r0 + r1 * 2 + 0 * mmsize/2]
-    pmovzxbw    m1, [r0 + r1 * 2 + 1 * mmsize/2]
-    pmovzxbw    m2, [r0 + r1 * 2 + 2 * mmsize/2]
-    pmovzxbw    m3, [r0 + r1 * 2 + 3 * mmsize/2]
-    psllw       m0, 6
-    psllw       m1, 6
-    psllw       m2, 6
-    psllw       m3, 6
-    psubw       m0, m4
-    psubw       m1, m4
-    psubw       m2, m4
-    psubw       m3, m4
+    movq            xm3, [r0 + r4]                  ; m3 = row 11
+    punpcklbw       xm6, xm3                        ; m6 = [B7 A7 B6 A6 B5 A5 B4 A4 B3 A3 B2 A2 B1 A1 B0 A0]
+    lea             r0, [r0 + r1 * 4]
+    movq            xm0, [r0]                       ; m0 = row 12
+    punpcklbw       xm3, xm0                        ; m3 = [C7 B7 C6 B6 C5 B5 C4 B4 C3 B3 C2 B2 C1 B1 C0 B0]
+    vinserti128     m6, m6, xm3, 1                  ; m6 = [C7 B7 C6 B6 C5 B5 C4 B4 C3 B3 C2 B2 C1 B1 C0 B0] - [B7 A7 B6 A6 B5 A5 B4 A4 B3 A3 B2 A2 B1 A1 B0 A0]
+    pmaddubsw       m3, m6, [r5 + 3 * mmsize]
+    paddw           m1, m3
+    pmaddubsw       m6, [r5 + 2 * mmsize]
+    paddw           m4, m6
+    movq            xm3, [r0 + r1]                  ; m3 = row 13
+    punpcklbw       xm0, xm3                        ; m0 = [D7 C7 D6 C6 D5 C5 D4 C4 D3 C3 D2 C2 D1 C1 D0 C0]
+    movq            xm6, [r0 + r1 * 2]              ; m6 = row 14
+    punpcklbw       xm3, xm6                        ; m3 = [E7 D7 E6 D6 E5 D5 E4 D4 E3 D3 E2 D2 E1 D1 E0 D0]
+    vinserti128     m0, m0, xm3, 1                  ; m0 = [E7 D7 E6 D6 E5 D5 E4 D4 E3 D3 E2 D2 E1 D1 E0 D0] - [D7 C7 D6 C6 D5 C5 D4 C4 D3 C3 D2 C2 D1 C1 D0 C0]
+    pmaddubsw       m0, [r5 + 3 * mmsize]
+    paddw           m4, m0
+%endmacro
 
-    movu        [r2 + r3 * 2 + 0 * mmsize], m0
-    movu        [r2 + r3 * 2 + 1 * mmsize], m1
-    movu        [r2 + r3 * 2 + 2 * mmsize], m2
-    movu        [r2 + r3 * 2 + 3 * mmsize], m3
+%macro PROCESS_LUMA_AVX2_W8_4R 0
+    movq            xm1, [r0]                       ; m1 = row 0
+    movq            xm2, [r0 + r1]                  ; m2 = row 1
+    punpcklbw       xm1, xm2                        ; m1 = [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
+    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
+    punpcklbw       xm2, xm3                        ; m2 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10]
+    vinserti128     m5, m1, xm2, 1                  ; m5 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10] - [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
+    pmaddubsw       m5, [r5]
+    movq            xm4, [r0 + r4]                  ; m4 = row 3
+    punpcklbw       xm3, xm4                        ; m3 = [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
+    lea             r0, [r0 + r1 * 4]
+    movq            xm1, [r0]                       ; m1 = row 4
+    punpcklbw       xm4, xm1                        ; m4 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30]
+    vinserti128     m2, m3, xm4, 1                  ; m2 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30] - [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
+    pmaddubsw       m0, m2, [r5 + 1 * mmsize]
+    paddw           m5, m0
+    pmaddubsw       m2, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 5
+    punpcklbw       xm1, xm3                        ; m1 = [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
+    movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
+    punpcklbw       xm3, xm4                        ; m3 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50]
+    vinserti128     m1, m1, xm3, 1                  ; m1 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50] - [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
+    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
+    paddw           m5, m3
+    pmaddubsw       m0, m1, [r5 + 1 * mmsize]
+    paddw           m2, m0
+    movq            xm3, [r0 + r4]                  ; m3 = row 7
+    punpcklbw       xm4, xm3                        ; m4 = [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
+    lea             r0, [r0 + r1 * 4]
+    movq            xm0, [r0]                       ; m0 = row 8
+    punpcklbw       xm3, xm0                        ; m3 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70]
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70] - [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
+    pmaddubsw       m3, m4, [r5 + 3 * mmsize]
+    paddw           m5, m3
+    pmaddubsw       m3, m4, [r5 + 2 * mmsize]
+    paddw           m2, m3
+    movq            xm3, [r0 + r1]                  ; m3 = row 9
+    punpcklbw       xm0, xm3                        ; m0 = [97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80]
+    movq            xm6, [r0 + r1 * 2]              ; m6 = row 10
+    punpcklbw       xm3, xm6                        ; m3 = [A7 97 A6 96 A5 95 A4 94 A3 93 A2 92 A1 91 A0 90]
+    vinserti128     m0, m0, xm3, 1                  ; m0 = [A7 97 A6 96 A5 95 A4 94 A3 93 A2 92 A1 91 A0 90] - [97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80]
+    pmaddubsw       m3, m0, [r5 + 3 * mmsize]
+    paddw           m2, m3
+%endmacro
 
-    pmovzxbw    m0, [r0 + r5 + 0 * mmsize/2]
-    pmovzxbw    m1, [r0 + r5 + 1 * mmsize/2]
-    pmovzxbw    m2, [r0 + r5 + 2 * mmsize/2]
-    pmovzxbw    m3, [r0 + r5 + 3 * mmsize/2]
-    psllw       m0, 6
-    psllw       m1, 6
-    psllw       m2, 6
-    psllw       m3, 6
-    psubw       m0, m4
-    psubw       m1, m4
-    psubw       m2, m4
-    psubw       m3, m4
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_%3_8x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_LUMA_8xN 3
+INIT_XMM sse4
+cglobal interp_8tap_vert_%3_%1x%2, 5, 7, 8
+    lea       r5, [3 * r1]
+    sub       r0, r5
+    shl       r4d, 6
 
-    movu        [r2 + r6 + 0 * mmsize], m0
-    movu        [r2 + r6 + 1 * mmsize], m1
-    movu        [r2 + r6 + 2 * mmsize], m2
-    movu        [r2 + r6 + 3 * mmsize], m3
+%ifidn %3,ps
+    add       r3d, r3d
+%endif
 
-    lea         r0, [r0 + r1 * 4]
-    lea         r2, [r2 + r3 * 4]
+%ifdef PIC
+    lea       r5, [tab_LumaCoeffVer]
+    lea       r6, [r5 + r4]
+%else
+    lea       r6, [tab_LumaCoeffVer + r4]
+%endif
 
-    dec         r4d
-    jnz        .loop
-    RET
-%endmacro
-    P2S_H_64xN_avx2 64
-    P2S_H_64xN_avx2 16
-    P2S_H_64xN_avx2 32
-    P2S_H_64xN_avx2 48
+ %ifidn %3,pp
+    mova      m3, [pw_512]
+%else
+    mova      m3, [pw_2000]
+%endif
 
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel src, intptr_t srcStride, int16_t dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-%macro P2S_H_12xN 1
-INIT_XMM ssse3
-cglobal filterPixelToShort_12x%1, 3, 7, 6
-    mov         r3d, r3m
-    add         r3d, r3d
-    lea         r4, [r1 * 3]
-    lea         r6, [r3 * 3]
-    mov         r5d, %1/4
+    mov       r4d, %2/4
+    lea       r5, [4 * r1]
 
-    ; load constant
-    mova        m4, [pb_128]
-    mova        m5, [tab_c_64_n64]
+.loopH:
+    PROCESS_LUMA_W8_4R
 
-.loop:
-    movu        m0, [r0]
-    punpcklbw   m1, m0, m4
-    punpckhbw   m0, m4
-    pmaddubsw   m0, m5
-    pmaddubsw   m1, m5
+%ifidn %3,pp
+    pmulhrsw  m7, m3
+    pmulhrsw  m6, m3
+    pmulhrsw  m5, m3
+    pmulhrsw  m4, m3
 
-    movu        m2, [r0 + r1]
-    punpcklbw   m3, m2, m4
-    punpckhbw   m2, m4
-    pmaddubsw   m2, m5
-    pmaddubsw   m3, m5
+    packuswb  m7, m6
+    packuswb  m5, m4
 
-    movu        [r2 + r3 * 0], m1
-    movu        [r2 + r3 * 1], m3
+    movlps    [r2], m7
+    movhps    [r2 + r3], m7
+    lea       r2, [r2 + 2 * r3]
+    movlps    [r2], m5
+    movhps    [r2 + r3], m5
+%else
+    psubw     m7, m3
+    psubw     m6, m3
+    psubw     m5, m3
+    psubw     m4, m3
 
-    movh        [r2 + r3 * 0 + 16], m0
-    movh        [r2 + r3 * 1 + 16], m2
-
-    movu        m0, [r0 + r1 * 2]
-    punpcklbw   m1, m0, m4
-    punpckhbw   m0, m4
-    pmaddubsw   m0, m5
-    pmaddubsw   m1, m5
-
-    movu        m2, [r0 + r4]
-    punpcklbw   m3, m2, m4
-    punpckhbw   m2, m4
-    pmaddubsw   m2, m5
-    pmaddubsw   m3, m5
-
-    movu        [r2 + r3 * 2], m1
-    movu        [r2 + r6], m3
+    movu      [r2], m7
+    movu      [r2 + r3], m6
+    lea       r2, [r2 + 2 * r3]
+    movu      [r2], m5
+    movu      [r2 + r3], m4
+%endif
 
-    movh        [r2 + r3 * 2 + 16], m0
-    movh        [r2 + r6 + 16], m2
+    sub       r0, r5
+    lea       r2, [r2 + 2 * r3]
 
-    lea         r0, [r0 + r1 * 4]
-    lea         r2, [r2 + r3 * 4]
+    dec       r4d
+    jnz       .loopH
 
-    dec         r5d
-    jnz         .loop
     RET
 %endmacro
-    P2S_H_12xN 16
-    P2S_H_12xN 32
 
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-%macro P2S_H_24xN 1
-INIT_XMM ssse3
-cglobal filterPixelToShort_24x%1, 3, 7, 5
-    mov         r3d, r3m
-    add         r3d, r3d
-    lea         r4, [r1 * 3]
-    lea         r5, [r3 * 3]
-    mov         r6d, %1/4
+%macro FILTER_VER_LUMA_AVX2_8xN 3
+INIT_YMM avx2
+cglobal interp_8tap_vert_%3_%1x%2, 4, 7, 8, 0-gprsize
+    mov             r4d, r4m
+    shl             r4d, 7
 
-    ; load constant
-    mova        m3, [pb_128]
-    mova        m4, [tab_c_64_n64]
+%ifdef PIC
+    lea             r5, [tab_LumaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
+%endif
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+    lea             r6, [r1 * 4]
+%ifidn %3,pp
+    mova            m7, [pw_512]
+%else
+    add             r3d, r3d
+    vbroadcasti128  m7, [pw_2000]
+%endif
+    mov             word [rsp], %2 / 8
 
 .loop:
-    movu        m0, [r0]
-    punpcklbw   m1, m0, m3
-    punpckhbw   m0, m3
-    pmaddubsw   m0, m4
-    pmaddubsw   m1, m4
-
-    movu        m2, [r0 + 16]
-    punpcklbw   m2, m3
-    pmaddubsw   m2, m4
-
-    movu        [r2 +  r3 * 0], m1
-    movu        [r2 +  r3 * 0 + 16], m0
-    movu        [r2 +  r3 * 0 + 32], m2
-
-    movu        m0, [r0 + r1]
-    punpcklbw   m1, m0, m3
-    punpckhbw   m0, m3
-    pmaddubsw   m0, m4
-    pmaddubsw   m1, m4
-
-    movu        m2, [r0 + r1 + 16]
-    punpcklbw   m2, m3
-    pmaddubsw   m2, m4
-
-    movu        [r2 +  r3 * 1], m1
-    movu        [r2 +  r3 * 1 + 16], m0
-    movu        [r2 +  r3 * 1 + 32], m2
-
-    movu        m0, [r0 + r1 * 2]
-    punpcklbw   m1, m0, m3
-    punpckhbw   m0, m3
-    pmaddubsw   m0, m4
-    pmaddubsw   m1, m4
-
-    movu        m2, [r0 + r1 * 2 + 16]
-    punpcklbw   m2, m3
-    pmaddubsw   m2, m4
-
-    movu        [r2 +  r3 * 2], m1
-    movu        [r2 +  r3 * 2 + 16], m0
-    movu        [r2 +  r3 * 2 + 32], m2
-
-    movu        m0, [r0 + r4]
-    punpcklbw   m1, m0, m3
-    punpckhbw   m0, m3
-    pmaddubsw   m0, m4
-    pmaddubsw   m1, m4
+    PROCESS_LUMA_AVX2_W8_8R
+%ifidn %3,pp
+    pmulhrsw        m5, m7                          ; m5 = word: row 0, row 1
+    pmulhrsw        m2, m7                          ; m2 = word: row 2, row 3
+    pmulhrsw        m1, m7                          ; m1 = word: row 4, row 5
+    pmulhrsw        m4, m7                          ; m4 = word: row 6, row 7
+    packuswb        m5, m2
+    packuswb        m1, m4
+    vextracti128    xm2, m5, 1
+    vextracti128    xm4, m1, 1
+    movq            [r2], xm5
+    movq            [r2 + r3], xm2
+    lea             r2, [r2 + r3 * 2]
+    movhps          [r2], xm5
+    movhps          [r2 + r3], xm2
+    lea             r2, [r2 + r3 * 2]
+    movq            [r2], xm1
+    movq            [r2 + r3], xm4
+    lea             r2, [r2 + r3 * 2]
+    movhps          [r2], xm1
+    movhps          [r2 + r3], xm4
+%else
+    psubw           m5, m7                          ; m5 = word: row 0, row 1
+    psubw           m2, m7                          ; m2 = word: row 2, row 3
+    psubw           m1, m7                          ; m1 = word: row 4, row 5
+    psubw           m4, m7                          ; m4 = word: row 6, row 7
+    vextracti128    xm6, m5, 1
+    vextracti128    xm3, m2, 1
+    vextracti128    xm0, m1, 1
+    movu            [r2], xm5
+    movu            [r2 + r3], xm6
+    lea             r2, [r2 + r3 * 2]
+    movu            [r2], xm2
+    movu            [r2 + r3], xm3
+    lea             r2, [r2 + r3 * 2]
+    movu            [r2], xm1
+    movu            [r2 + r3], xm0
+    lea             r2, [r2 + r3 * 2]
+    movu            [r2], xm4
+    vextracti128    xm4, m4, 1
+    movu            [r2 + r3], xm4
+%endif
+    lea             r2, [r2 + r3 * 2]
+    sub             r0, r6
+    dec             word [rsp]
+    jnz             .loop
+    RET
+%endmacro
 
-    movu        m2, [r0 + r4 + 16]
-    punpcklbw   m2, m3
-    pmaddubsw   m2, m4
-    movu        [r2 +  r5], m1
-    movu        [r2 +  r5 + 16], m0
-    movu        [r2 +  r5 + 32], m2
+%macro FILTER_VER_LUMA_AVX2_8x8 1
+INIT_YMM avx2
+cglobal interp_8tap_vert_%1_8x8, 4, 6, 7
+    mov             r4d, r4m
+    shl             r4d, 7
 
-    lea         r0, [r0 + r1 * 4]
-    lea         r2, [r2 + r3 * 4]
+%ifdef PIC
+    lea             r5, [tab_LumaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
+%endif
 
-    dec         r6d
-    jnz         .loop
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+    PROCESS_LUMA_AVX2_W8_8R
+%ifidn %1,pp
+    mova            m3, [pw_512]
+%else
+    add             r3d, r3d
+    vbroadcasti128  m3, [pw_2000]
+%endif
+    lea             r4, [r3 * 3]
+%ifidn %1,pp
+    pmulhrsw        m5, m3                          ; m5 = word: row 0, row 1
+    pmulhrsw        m2, m3                          ; m2 = word: row 2, row 3
+    pmulhrsw        m1, m3                          ; m1 = word: row 4, row 5
+    pmulhrsw        m4, m3                          ; m4 = word: row 6, row 7
+    packuswb        m5, m2
+    packuswb        m1, m4
+    vextracti128    xm2, m5, 1
+    vextracti128    xm4, m1, 1
+    movq            [r2], xm5
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm5
+    movhps          [r2 + r4], xm2
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm1
+    movq            [r2 + r3], xm4
+    movhps          [r2 + r3 * 2], xm1
+    movhps          [r2 + r4], xm4
+%else
+    psubw           m5, m3                          ; m5 = word: row 0, row 1
+    psubw           m2, m3                          ; m2 = word: row 2, row 3
+    psubw           m1, m3                          ; m1 = word: row 4, row 5
+    psubw           m4, m3                          ; m4 = word: row 6, row 7
+    vextracti128    xm6, m5, 1
+    vextracti128    xm3, m2, 1
+    vextracti128    xm0, m1, 1
+    movu            [r2], xm5
+    movu            [r2 + r3], xm6
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r4], xm3
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm1
+    movu            [r2 + r3], xm0
+    movu            [r2 + r3 * 2], xm4
+    vextracti128    xm4, m4, 1
+    movu            [r2 + r4], xm4
+%endif
     RET
 %endmacro
-    P2S_H_24xN 32
-    P2S_H_24xN 64
 
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-%macro P2S_H_24xN_avx2 1
+%macro FILTER_VER_LUMA_AVX2_8x4 1
 INIT_YMM avx2
-cglobal filterPixelToShort_24x%1, 3, 7, 4
-    mov         r3d, r3m
-    add         r3d, r3d
-    lea         r4, [r1 * 3]
-    lea         r5, [r3 * 3]
-    mov         r6d, %1/4
-
-    ; load constant
-    vpbroadcastd m1, [pw_2000]
-    vpbroadcastd m2, [pb_128]
-    vpbroadcastd m3, [tab_c_64_n64]
+cglobal interp_8tap_vert_%1_8x4, 4, 6, 7
+    mov             r4d, r4m
+    shl             r4d, 7
 
-.loop:
-    pmovzxbw    m0, [r0]
-    psllw       m0, 6
-    psubw       m0, m1
-    movu        [r2], m0
-
-    movu        m0, [r0 + mmsize/2]
-    punpcklbw   m0, m2
-    pmaddubsw   m0, m3
-    movu        [r2 +  r3 * 0 + mmsize], xm0
+%ifdef PIC
+    lea             r5, [tab_LumaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
+%endif
 
-    pmovzxbw    m0, [r0 + r1]
-    psllw       m0, 6
-    psubw       m0, m1
-    movu        [r2 + r3], m0
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+    PROCESS_LUMA_AVX2_W8_4R
+%ifidn %1,pp
+    mova            m3, [pw_512]
+%else
+    add             r3d, r3d
+    vbroadcasti128  m3, [pw_2000]
+%endif
+    lea             r4, [r3 * 3]
+%ifidn %1,pp
+    pmulhrsw        m5, m3                          ; m5 = word: row 0, row 1
+    pmulhrsw        m2, m3                          ; m2 = word: row 2, row 3
+    packuswb        m5, m2
+    vextracti128    xm2, m5, 1
+    movq            [r2], xm5
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm5
+    movhps          [r2 + r4], xm2
+%else
+    psubw           m5, m3                          ; m5 = word: row 0, row 1
+    psubw           m2, m3                          ; m2 = word: row 2, row 3
+    movu            [r2], xm5
+    vextracti128    xm5, m5, 1
+    movu            [r2 + r3], xm5
+    movu            [r2 + r3 * 2], xm2
+    vextracti128    xm2, m2, 1
+    movu            [r2 + r4], xm2
+%endif
+    RET
+%endmacro
 
-    movu        m0, [r0 + r1 + mmsize/2]
-    punpcklbw   m0, m2
-    pmaddubsw   m0, m3
-    movu        [r2 +  r3 * 1 + mmsize], xm0
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_pp_8x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_8xN 8, 4, pp
+    FILTER_VER_LUMA_AVX2_8x4 pp
 
-    pmovzxbw    m0, [r0 + r1 * 2]
-    psllw       m0, 6
-    psubw       m0, m1
-    movu        [r2 + r3 * 2], m0
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_pp_8x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_8xN 8, 8, pp
+    FILTER_VER_LUMA_AVX2_8x8 pp
 
-    movu        m0, [r0 + r1 * 2 + mmsize/2]
-    punpcklbw   m0, m2
-    pmaddubsw   m0, m3
-    movu        [r2 +  r3 * 2 + mmsize], xm0
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_pp_8x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_8xN 8, 16, pp
+    FILTER_VER_LUMA_AVX2_8xN 8, 16, pp
 
-    pmovzxbw    m0, [r0 + r4]
-    psllw       m0, 6
-    psubw       m0, m1
-    movu        [r2 + r5], m0
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_pp_8x32(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_8xN 8, 32, pp
+    FILTER_VER_LUMA_AVX2_8xN 8, 32, pp
 
-    movu        m0, [r0 + r4 + mmsize/2]
-    punpcklbw   m0, m2
-    pmaddubsw   m0, m3
-    movu        [r2 + r5 + mmsize], xm0
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_ps_8x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_8xN 8, 4, ps
+    FILTER_VER_LUMA_AVX2_8x4 ps
 
-    lea         r0, [r0 + r1 * 4]
-    lea         r2, [r2 + r3 * 4]
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_ps_8x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_8xN 8, 8, ps
+    FILTER_VER_LUMA_AVX2_8x8 ps
 
-    dec         r6d
-    jnz         .loop
-    RET
-%endmacro
-    P2S_H_24xN_avx2 32
-    P2S_H_24xN_avx2 64
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_ps_8x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_8xN 8, 16, ps
+    FILTER_VER_LUMA_AVX2_8xN 8, 16, ps
 
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-INIT_XMM ssse3
-cglobal filterPixelToShort_48x64, 3, 7, 4
-    mov         r3d, r3m
-    add         r3d, r3d
-    lea         r4, [r1 * 3]
-    lea         r5, [r3 * 3]
-    mov         r6d, 16
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_ps_8x32(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_8xN 8, 32, ps
+    FILTER_VER_LUMA_AVX2_8xN 8, 32, ps
 
-    ; load constant
-    mova        m2, [pb_128]
-    mova        m3, [tab_c_64_n64]
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_%3_12x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_LUMA_12xN 3
+INIT_XMM sse4
+cglobal interp_8tap_vert_%3_%1x%2, 5, 7, 8
+    lea       r5, [3 * r1]
+    sub       r0, r5
+    shl       r4d, 6
+%ifidn %3,ps
+    add       r3d, r3d
+%endif
 
-.loop:
-    movu        m0, [r0]
-    punpcklbw   m1, m0, m2
-    punpckhbw   m0, m2
-    pmaddubsw   m0, m3
-    pmaddubsw   m1, m3
+%ifdef PIC
+    lea       r5, [tab_LumaCoeffVer]
+    lea       r6, [r5 + r4]
+%else
+    lea       r6, [tab_LumaCoeffVer + r4]
+%endif
 
-    movu        [r2 +  r3 * 0], m1
-    movu        [r2 +  r3 * 0 + 16], m0
+ %ifidn %3,pp
+    mova      m3, [pw_512]
+%else
+    mova      m3, [pw_2000]
+%endif
 
-    movu        m0, [r0 + 16]
-    punpcklbw   m1, m0, m2
-    punpckhbw   m0, m2
-    pmaddubsw   m0, m3
-    pmaddubsw   m1, m3
+    mov       r4d, %2/4
 
-    movu        [r2 +  r3 * 0 + 32], m1
-    movu        [r2 +  r3 * 0 + 48], m0
+.loopH:
+    PROCESS_LUMA_W8_4R
 
-    movu        m0, [r0 + 32]
-    punpcklbw   m1, m0, m2
-    punpckhbw   m0, m2
-    pmaddubsw   m0, m3
-    pmaddubsw   m1, m3
+%ifidn %3,pp
+    pmulhrsw  m7, m3
+    pmulhrsw  m6, m3
+    pmulhrsw  m5, m3
+    pmulhrsw  m4, m3
 
-    movu        [r2 +  r3 * 0 + 64], m1
-    movu        [r2 +  r3 * 0 + 80], m0
+    packuswb  m7, m6
+    packuswb  m5, m4
 
-    movu        m0, [r0 + r1]
-    punpcklbw   m1, m0, m2
-    punpckhbw   m0, m2
-    pmaddubsw   m0, m3
-    pmaddubsw   m1, m3
+    movlps    [r2], m7
+    movhps    [r2 + r3], m7
+    lea       r5, [r2 + 2 * r3]
+    movlps    [r5], m5
+    movhps    [r5 + r3], m5
+%else
+    psubw     m7, m3
+    psubw     m6, m3
+    psubw     m5, m3
+    psubw     m4, m3
 
-    movu        [r2 +  r3 * 1], m1
-    movu        [r2 +  r3 * 1 + 16], m0
+    movu      [r2], m7
+    movu      [r2 + r3], m6
+    lea       r5, [r2 + 2 * r3]
+    movu      [r5], m5
+    movu      [r5 + r3], m4
+%endif
 
-    movu        m0, [r0 + r1 + 16]
-    punpcklbw   m1, m0, m2
-    punpckhbw   m0, m2
-    pmaddubsw   m0, m3
-    pmaddubsw   m1, m3
+    lea       r5, [8 * r1 - 8]
+    sub       r0, r5
+%ifidn %3,pp
+    add       r2, 8
+%else
+    add       r2, 16
+%endif
 
-    movu        [r2 +  r3 * 1 + 32], m1
-    movu        [r2 +  r3 * 1 + 48], m0
+    PROCESS_LUMA_W4_4R
 
-    movu        m0, [r0 + r1 + 32]
-    punpcklbw   m1, m0, m2
-    punpckhbw   m0, m2
-    pmaddubsw   m0, m3
-    pmaddubsw   m1, m3
+%ifidn %3,pp
+    pmulhrsw  m4, m3
+    pmulhrsw  m5, m3
 
-    movu        [r2 +  r3 * 1 + 64], m1
-    movu        [r2 +  r3 * 1 + 80], m0
+    packuswb  m4, m5
 
-    movu        m0, [r0 + r1 * 2]
-    punpcklbw   m1, m0, m2
-    punpckhbw   m0, m2
-    pmaddubsw   m0, m3
-    pmaddubsw   m1, m3
+    movd      [r2], m4
+    pextrd    [r2 + r3], m4, 1
+    lea       r5, [r2 + 2 * r3]
+    pextrd    [r5], m4, 2
+    pextrd    [r5 + r3], m4, 3
+%else
+    psubw     m4, m3
+    psubw     m5, m3
 
-    movu        [r2 +  r3 * 2], m1
-    movu        [r2 +  r3 * 2 + 16], m0
+    movlps    [r2], m4
+    movhps    [r2 + r3], m4
+    lea       r5, [r2 + 2 * r3]
+    movlps    [r5], m5
+    movhps    [r5 + r3], m5
+%endif
 
-    movu        m0, [r0 + r1 * 2 + 16]
-    punpcklbw   m1, m0, m2
-    punpckhbw   m0, m2
-    pmaddubsw   m0, m3
-    pmaddubsw   m1, m3
-
-    movu        [r2 +  r3 * 2 + 32], m1
-    movu        [r2 +  r3 * 2 + 48], m0
-
-    movu        m0, [r0 + r1 * 2 + 32]
-    punpcklbw   m1, m0, m2
-    punpckhbw   m0, m2
-    pmaddubsw   m0, m3
-    pmaddubsw   m1, m3
-
-    movu        [r2 +  r3 * 2 + 64], m1
-    movu        [r2 +  r3 * 2 + 80], m0
-
-    movu        m0, [r0 + r4]
-    punpcklbw   m1, m0, m2
-    punpckhbw   m0, m2
-    pmaddubsw   m0, m3
-    pmaddubsw   m1, m3
-
-    movu        [r2 +  r5], m1
-    movu        [r2 +  r5 + 16], m0
-
-    movu        m0, [r0 + r4 + 16]
-    punpcklbw   m1, m0, m2
-    punpckhbw   m0, m2
-    pmaddubsw   m0, m3
-    pmaddubsw   m1, m3
-
-    movu        [r2 +  r5 + 32], m1
-    movu        [r2 +  r5 + 48], m0
-
-    movu        m0, [r0 + r4 + 32]
-    punpcklbw   m1, m0, m2
-    punpckhbw   m0, m2
-    pmaddubsw   m0, m3
-    pmaddubsw   m1, m3
-
-    movu        [r2 +  r5 + 64], m1
-    movu        [r2 +  r5 + 80], m0
-
-    lea         r0, [r0 + r1 * 4]
-    lea         r2, [r2 + r3 * 4]
-
-    dec         r6d
-    jnz         .loop
-    RET
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-INIT_YMM avx2
-cglobal filterPixelToShort_48x64, 3,7,4
-    mov         r3d, r3m
-    add         r3d, r3d
-    lea         r5, [r1 * 3]
-    lea         r6, [r3 * 3]
-
-    ; load height
-    mov         r4d, 64/4
-
-    ; load constant
-    vpbroadcastd m3, [pw_2000]
-
-    ; just unroll(1) because it is best choice for 48x64
-.loop:
-    pmovzxbw    m0, [r0 + 0 * mmsize/2]
-    pmovzxbw    m1, [r0 + 1 * mmsize/2]
-    pmovzxbw    m2, [r0 + 2 * mmsize/2]
-    psllw       m0, 6
-    psllw       m1, 6
-    psllw       m2, 6
-    psubw       m0, m3
-    psubw       m1, m3
-    psubw       m2, m3
-    movu        [r2 + 0 * mmsize], m0
-    movu        [r2 + 1 * mmsize], m1
-    movu        [r2 + 2 * mmsize], m2
-
-    pmovzxbw    m0, [r0 + r1 + 0 * mmsize/2]
-    pmovzxbw    m1, [r0 + r1 + 1 * mmsize/2]
-    pmovzxbw    m2, [r0 + r1 + 2 * mmsize/2]
-    psllw       m0, 6
-    psllw       m1, 6
-    psllw       m2, 6
-    psubw       m0, m3
-    psubw       m1, m3
-    psubw       m2, m3
-    movu        [r2 + r3 + 0 * mmsize], m0
-    movu        [r2 + r3 + 1 * mmsize], m1
-    movu        [r2 + r3 + 2 * mmsize], m2
-
-    pmovzxbw    m0, [r0 + r1 * 2 + 0 * mmsize/2]
-    pmovzxbw    m1, [r0 + r1 * 2 + 1 * mmsize/2]
-    pmovzxbw    m2, [r0 + r1 * 2 + 2 * mmsize/2]
-    psllw       m0, 6
-    psllw       m1, 6
-    psllw       m2, 6
-    psubw       m0, m3
-    psubw       m1, m3
-    psubw       m2, m3
-    movu        [r2 + r3 * 2 + 0 * mmsize], m0
-    movu        [r2 + r3 * 2 + 1 * mmsize], m1
-    movu        [r2 + r3 * 2 + 2 * mmsize], m2
-
-    pmovzxbw    m0, [r0 + r5 + 0 * mmsize/2]
-    pmovzxbw    m1, [r0 + r5 + 1 * mmsize/2]
-    pmovzxbw    m2, [r0 + r5 + 2 * mmsize/2]
-    psllw       m0, 6
-    psllw       m1, 6
-    psllw       m2, 6
-    psubw       m0, m3
-    psubw       m1, m3
-    psubw       m2, m3
-    movu        [r2 + r6 + 0 * mmsize], m0
-    movu        [r2 + r6 + 1 * mmsize], m1
-    movu        [r2 + r6 + 2 * mmsize], m2
+    lea       r5, [4 * r1 + 8]
+    sub       r0, r5
+%ifidn %3,pp
+    lea       r2, [r2 + 4 * r3 - 8]
+%else
+    lea       r2, [r2 + 4 * r3 - 16]
+%endif
 
-    lea         r0, [r0 + r1 * 4]
-    lea         r2, [r2 + r3 * 4]
+    dec       r4d
+    jnz       .loopH
 
-    dec         r4d
-    jnz        .loop
     RET
-
-
-%macro PROCESS_LUMA_W4_4R 0
-    movd        m0, [r0]
-    movd        m1, [r0 + r1]
-    punpcklbw   m2, m0, m1                     ; m2=[0 1]
-
-    lea         r0, [r0 + 2 * r1]
-    movd        m0, [r0]
-    punpcklbw   m1, m0                         ; m1=[1 2]
-    punpcklqdq  m2, m1                         ; m2=[0 1 1 2]
-    pmaddubsw   m4, m2, [r6 + 0 * 16]          ; m4=[0+1 1+2]
-
-    movd        m1, [r0 + r1]
-    punpcklbw   m5, m0, m1                     ; m2=[2 3]
-    lea         r0, [r0 + 2 * r1]
-    movd        m0, [r0]
-    punpcklbw   m1, m0                         ; m1=[3 4]
-    punpcklqdq  m5, m1                         ; m5=[2 3 3 4]
-    pmaddubsw   m2, m5, [r6 + 1 * 16]          ; m2=[2+3 3+4]
-    paddw       m4, m2                         ; m4=[0+1+2+3 1+2+3+4]                   Row1-2
-    pmaddubsw   m5, [r6 + 0 * 16]              ; m5=[2+3 3+4]                           Row3-4
-
-    movd        m1, [r0 + r1]
-    punpcklbw   m2, m0, m1                     ; m2=[4 5]
-    lea         r0, [r0 + 2 * r1]
-    movd        m0, [r0]
-    punpcklbw   m1, m0                         ; m1=[5 6]
-    punpcklqdq  m2, m1                         ; m2=[4 5 5 6]
-    pmaddubsw   m1, m2, [r6 + 2 * 16]          ; m1=[4+5 5+6]
-    paddw       m4, m1                         ; m4=[0+1+2+3+4+5 1+2+3+4+5+6]           Row1-2
-    pmaddubsw   m2, [r6 + 1 * 16]              ; m2=[4+5 5+6]
-    paddw       m5, m2                         ; m5=[2+3+4+5 3+4+5+6]                   Row3-4
-
-    movd        m1, [r0 + r1]
-    punpcklbw   m2, m0, m1                     ; m2=[6 7]
-    lea         r0, [r0 + 2 * r1]
-    movd        m0, [r0]
-    punpcklbw   m1, m0                         ; m1=[7 8]
-    punpcklqdq  m2, m1                         ; m2=[6 7 7 8]
-    pmaddubsw   m1, m2, [r6 + 3 * 16]          ; m1=[6+7 7+8]
-    paddw       m4, m1                         ; m4=[0+1+2+3+4+5+6+7 1+2+3+4+5+6+7+8]   Row1-2 end
-    pmaddubsw   m2, [r6 + 2 * 16]              ; m2=[6+7 7+8]
-    paddw       m5, m2                         ; m5=[2+3+4+5+6+7 3+4+5+6+7+8]           Row3-4
-
-    movd        m1, [r0 + r1]
-    punpcklbw   m2, m0, m1                     ; m2=[8 9]
-    movd        m0, [r0 + 2 * r1]
-    punpcklbw   m1, m0                         ; m1=[9 10]
-    punpcklqdq  m2, m1                         ; m2=[8 9 9 10]
-    pmaddubsw   m2, [r6 + 3 * 16]              ; m2=[8+9 9+10]
-    paddw       m5, m2                         ; m5=[2+3+4+5+6+7+8+9 3+4+5+6+7+8+9+10]  Row3-4 end
 %endmacro
 
-%macro PROCESS_LUMA_W8_4R 0
-    movq       m0, [r0]
-    movq       m1, [r0 + r1]
-    punpcklbw  m0, m1
-    pmaddubsw  m7, m0, [r6 + 0 *16]            ;m7=[0+1]               Row1
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m0, [r0]
-    punpcklbw  m1, m0
-    pmaddubsw  m6, m1, [r6 + 0 *16]            ;m6=[1+2]               Row2
-
-    movq       m1, [r0 + r1]
-    punpcklbw  m0, m1
-    pmaddubsw  m5, m0, [r6 + 0 *16]            ;m5=[2+3]               Row3
-    pmaddubsw  m0, [r6 + 1 * 16]
-    paddw      m7, m0                          ;m7=[0+1+2+3]           Row1
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_pp_12x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_12xN 12, 16, pp
 
-    lea        r0, [r0 + 2 * r1]
-    movq       m0, [r0]
-    punpcklbw  m1, m0
-    pmaddubsw  m4, m1, [r6 + 0 *16]            ;m4=[3+4]               Row4
-    pmaddubsw  m1, [r6 + 1 * 16]
-    paddw      m6, m1                          ;m6 = [1+2+3+4]         Row2
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_ps_12x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_12xN 12, 16, ps
 
-    movq       m1, [r0 + r1]
-    punpcklbw  m0, m1
-    pmaddubsw  m2, m0, [r6 + 1 * 16]
-    pmaddubsw  m0, [r6 + 2 * 16]
-    paddw      m7, m0                          ;m7=[0+1+2+3+4+5]       Row1
-    paddw      m5, m2                          ;m5=[2+3+4+5]           Row3
+%macro FILTER_VER_LUMA_AVX2_12x16 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_12x16, 4, 7, 15
+    mov             r4d, r4m
+    shl             r4d, 7
 
-    lea        r0, [r0 + 2 * r1]
-    movq       m0, [r0]
-    punpcklbw  m1, m0
-    pmaddubsw  m2, m1, [r6 + 1 * 16]
-    pmaddubsw  m1, [r6 + 2 * 16]
-    paddw      m6, m1                          ;m6=[1+2+3+4+5+6]       Row2
-    paddw      m4, m2                          ;m4=[3+4+5+6]           Row4
+%ifdef PIC
+    lea             r5, [tab_LumaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
+%endif
 
-    movq       m1, [r0 + r1]
-    punpcklbw  m0, m1
-    pmaddubsw  m2, m0, [r6 + 2 * 16]
-    pmaddubsw  m0, [r6 + 3 * 16]
-    paddw      m7, m0                          ;m7=[0+1+2+3+4+5+6+7]   Row1 end
-    paddw      m5, m2                          ;m5=[2+3+4+5+6+7]       Row3
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+%ifidn %1,pp
+    mova            m14, [pw_512]
+%else
+    add             r3d, r3d
+    vbroadcasti128  m14, [pw_2000]
+%endif
+    lea             r6, [r3 * 3]
 
-    lea        r0, [r0 + 2 * r1]
-    movq       m0, [r0]
-    punpcklbw  m1, m0
-    pmaddubsw  m2, m1, [r6 + 2 * 16]
-    pmaddubsw  m1, [r6 + 3 * 16]
-    paddw      m6, m1                          ;m6=[1+2+3+4+5+6+7+8]   Row2 end
-    paddw      m4, m2                          ;m4=[3+4+5+6+7+8]       Row4
-
-    movq       m1, [r0 + r1]
-    punpcklbw  m0, m1
-    pmaddubsw  m0, [r6 + 3 * 16]
-    paddw      m5, m0                          ;m5=[2+3+4+5+6+7+8+9]   Row3 end
-
-    movq       m0, [r0 + 2 * r1]
-    punpcklbw  m1, m0
-    pmaddubsw  m1, [r6 + 3 * 16]
-    paddw      m4, m1                          ;m4=[3+4+5+6+7+8+9+10]  Row4 end
-%endmacro
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_%3_4x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_LUMA_4xN 3
-INIT_XMM sse4
-cglobal interp_8tap_vert_%3_%1x%2, 5, 7, 6
-    lea       r5, [3 * r1]
-    sub       r0, r5
-    shl       r4d, 6
-%ifidn %3,ps
-    add       r3d, r3d
-%endif
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddubsw       m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
+    paddw           m0, m4
+    pmaddubsw       m2, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
+    paddw           m1, m5
+    pmaddubsw       m3, [r5]
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhbw       xm6, xm4, xm5
+    punpcklbw       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
+    paddw           m0, m6
+    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
+    paddw           m2, m6
+    pmaddubsw       m4, [r5]
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhbw       xm7, xm5, xm6
+    punpcklbw       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
+    paddw           m1, m7
+    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
+    paddw           m3, m7
+    pmaddubsw       m5, [r5]
+    movu            xm7, [r0 + r4]                  ; m7 = row 7
+    punpckhbw       xm8, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
+    paddw           m0, m8
+    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
+    paddw           m2, m8
+    pmaddubsw       m8, m6, [r5 + 1 * mmsize]
+    paddw           m4, m8
+    pmaddubsw       m6, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm8, [r0]                       ; m8 = row 8
+    punpckhbw       xm9, xm7, xm8
+    punpcklbw       xm7, xm8
+    vinserti128     m7, m7, xm9, 1
+    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
+    paddw           m1, m9
+    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
+    paddw           m3, m9
+    pmaddubsw       m9, m7, [r5 + 1 * mmsize]
+    paddw           m5, m9
+    pmaddubsw       m7, [r5]
+    movu            xm9, [r0 + r1]                  ; m9 = row 9
+    punpckhbw       xm10, xm8, xm9
+    punpcklbw       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
+    paddw           m2, m10
+    pmaddubsw       m10, m8, [r5 + 2 * mmsize]
+    paddw           m4, m10
+    pmaddubsw       m10, m8, [r5 + 1 * mmsize]
+    paddw           m6, m10
+    pmaddubsw       m8, [r5]
+    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
+    punpckhbw       xm11, xm9, xm10
+    punpcklbw       xm9, xm10
+    vinserti128     m9, m9, xm11, 1
+    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
+    paddw           m3, m11
+    pmaddubsw       m11, m9, [r5 + 2 * mmsize]
+    paddw           m5, m11
+    pmaddubsw       m11, m9, [r5 + 1 * mmsize]
+    paddw           m7, m11
+    pmaddubsw       m9, [r5]
+    movu            xm11, [r0 + r4]                 ; m11 = row 11
+    punpckhbw       xm12, xm10, xm11
+    punpcklbw       xm10, xm11
+    vinserti128     m10, m10, xm12, 1
+    pmaddubsw       m12, m10, [r5 + 3 * mmsize]
+    paddw           m4, m12
+    pmaddubsw       m12, m10, [r5 + 2 * mmsize]
+    paddw           m6, m12
+    pmaddubsw       m12, m10, [r5 + 1 * mmsize]
+    paddw           m8, m12
+    pmaddubsw       m10, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm12, [r0]                      ; m12 = row 12
+    punpckhbw       xm13, xm11, xm12
+    punpcklbw       xm11, xm12
+    vinserti128     m11, m11, xm13, 1
+    pmaddubsw       m13, m11, [r5 + 3 * mmsize]
+    paddw           m5, m13
+    pmaddubsw       m13, m11, [r5 + 2 * mmsize]
+    paddw           m7, m13
+    pmaddubsw       m13, m11, [r5 + 1 * mmsize]
+    paddw           m9, m13
+    pmaddubsw       m11, [r5]
 
-%ifdef PIC
-    lea       r5, [tab_LumaCoeffVer]
-    lea       r6, [r5 + r4]
+%ifidn %1,pp
+    pmulhrsw        m0, m14                         ; m0 = word: row 0
+    pmulhrsw        m1, m14                         ; m1 = word: row 1
+    pmulhrsw        m2, m14                         ; m2 = word: row 2
+    pmulhrsw        m3, m14                         ; m3 = word: row 3
+    pmulhrsw        m4, m14                         ; m4 = word: row 4
+    pmulhrsw        m5, m14                         ; m5 = word: row 5
+    packuswb        m0, m1
+    packuswb        m2, m3
+    packuswb        m4, m5
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    vpermq          m4, m4, 11011000b
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m2, 1
+    vextracti128    xm5, m4, 1
+    movq            [r2], xm0
+    pextrd          [r2 + 8], xm0, 2
+    movq            [r2 + r3], xm1
+    pextrd          [r2 + r3 + 8], xm1, 2
+    movq            [r2 + r3 * 2], xm2
+    pextrd          [r2 + r3 * 2 + 8], xm2, 2
+    movq            [r2 + r6], xm3
+    pextrd          [r2 + r6 + 8], xm3, 2
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm4
+    pextrd          [r2 + 8], xm4, 2
+    movq            [r2 + r3], xm5
+    pextrd          [r2 + r3 + 8], xm5, 2
 %else
-    lea       r6, [tab_LumaCoeffVer + r4]
+    psubw           m0, m14                         ; m0 = word: row 0
+    psubw           m1, m14                         ; m1 = word: row 1
+    psubw           m2, m14                         ; m2 = word: row 2
+    psubw           m3, m14                         ; m3 = word: row 3
+    psubw           m4, m14                         ; m4 = word: row 4
+    psubw           m5, m14                         ; m5 = word: row 5
+    movu            [r2], xm0
+    vextracti128    xm0, m0, 1
+    movq            [r2 + 16], xm0
+    movu            [r2 + r3], xm1
+    vextracti128    xm1, m1, 1
+    movq            [r2 + r3 + 16], xm1
+    movu            [r2 + r3 * 2], xm2
+    vextracti128    xm2, m2, 1
+    movq            [r2 + r3 * 2 + 16], xm2
+    movu            [r2 + r6], xm3
+    vextracti128    xm3, m3, 1
+    movq            [r2 + r6 + 16], xm3
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm4
+    vextracti128    xm4, m4, 1
+    movq            [r2 + 16], xm4
+    movu            [r2 + r3], xm5
+    vextracti128    xm5, m5, 1
+    movq            [r2 + r3 + 16], xm5
 %endif
 
-%ifidn %3,pp
-    mova      m3, [pw_512]
+    movu            xm13, [r0 + r1]                 ; m13 = row 13
+    punpckhbw       xm0, xm12, xm13
+    punpcklbw       xm12, xm13
+    vinserti128     m12, m12, xm0, 1
+    pmaddubsw       m0, m12, [r5 + 3 * mmsize]
+    paddw           m6, m0
+    pmaddubsw       m0, m12, [r5 + 2 * mmsize]
+    paddw           m8, m0
+    pmaddubsw       m0, m12, [r5 + 1 * mmsize]
+    paddw           m10, m0
+    pmaddubsw       m12, [r5]
+    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
+    punpckhbw       xm1, xm13, xm0
+    punpcklbw       xm13, xm0
+    vinserti128     m13, m13, xm1, 1
+    pmaddubsw       m1, m13, [r5 + 3 * mmsize]
+    paddw           m7, m1
+    pmaddubsw       m1, m13, [r5 + 2 * mmsize]
+    paddw           m9, m1
+    pmaddubsw       m1, m13, [r5 + 1 * mmsize]
+    paddw           m11, m1
+    pmaddubsw       m13, [r5]
+
+%ifidn %1,pp
+    pmulhrsw        m6, m14                         ; m6 = word: row 6
+    pmulhrsw        m7, m14                         ; m7 = word: row 7
+    packuswb        m6, m7
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm7, m6, 1
+    movq            [r2 + r3 * 2], xm6
+    pextrd          [r2 + r3 * 2 + 8], xm6, 2
+    movq            [r2 + r6], xm7
+    pextrd          [r2 + r6 + 8], xm7, 2
 %else
-    mova      m3, [pw_2000]
+    psubw           m6, m14                         ; m6 = word: row 6
+    psubw           m7, m14                         ; m7 = word: row 7
+    movu            [r2 + r3 * 2], xm6
+    vextracti128    xm6, m6, 1
+    movq            [r2 + r3 * 2 + 16], xm6
+    movu            [r2 + r6], xm7
+    vextracti128    xm7, m7, 1
+    movq            [r2 + r6 + 16], xm7
 %endif
+    lea             r2, [r2 + r3 * 4]
 
-    mov       r4d, %2/4
-    lea       r5, [4 * r1]
-
-.loopH:
-    PROCESS_LUMA_W4_4R
-
-%ifidn %3,pp
-    pmulhrsw  m4, m3
-    pmulhrsw  m5, m3
-
-    packuswb  m4, m5
-
-    movd      [r2], m4
-    pextrd    [r2 + r3], m4, 1
-    lea       r2, [r2 + 2 * r3]
-    pextrd    [r2], m4, 2
-    pextrd    [r2 + r3], m4, 3
-%else
-    psubw     m4, m3
-    psubw     m5, m3
-
-    movlps    [r2], m4
-    movhps    [r2 + r3], m4
-    lea       r2, [r2 + 2 * r3]
-    movlps    [r2], m5
-    movhps    [r2 + r3], m5
-%endif
-
-    sub       r0, r5
-    lea       r2, [r2 + 2 * r3]
-
-    dec       r4d
-    jnz       .loopH
-
-    RET
-%endmacro
-
-
-INIT_YMM avx2
-cglobal interp_8tap_vert_pp_4x4, 4,6,8
-    mov             r4d, r4m
-    lea             r5, [r1 * 3]
-    sub             r0, r5
-
-    ; TODO: VPGATHERDD
-    movd            xm1, [r0]                       ; m1 = row0
-    movd            xm2, [r0 + r1]                  ; m2 = row1
-    punpcklbw       xm1, xm2                        ; m1 = [13 03 12 02 11 01 10 00]
-
-    movd            xm3, [r0 + r1 * 2]              ; m3 = row2
-    punpcklbw       xm2, xm3                        ; m2 = [23 13 22 12 21 11 20 10]
-    movd            xm4, [r0 + r5]
-    punpcklbw       xm3, xm4                        ; m3 = [33 23 32 22 31 21 30 20]
-    punpcklwd       xm1, xm3                        ; m1 = [33 23 13 03 32 22 12 02 31 21 11 01 30 20 10 00]
-
+    movu            xm1, [r0 + r4]                  ; m1 = row 15
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddubsw       m2, m0, [r5 + 3 * mmsize]
+    paddw           m8, m2
+    pmaddubsw       m2, m0, [r5 + 2 * mmsize]
+    paddw           m10, m2
+    pmaddubsw       m2, m0, [r5 + 1 * mmsize]
+    paddw           m12, m2
+    pmaddubsw       m0, [r5]
     lea             r0, [r0 + r1 * 4]
-    movd            xm5, [r0]                       ; m5 = row4
-    punpcklbw       xm4, xm5                        ; m4 = [43 33 42 32 41 31 40 30]
-    punpcklwd       xm2, xm4                        ; m2 = [43 33 21 13 42 32 22 12 41 31 21 11 40 30 20 10]
-    vinserti128     m1, m1, xm2, 1                  ; m1 = [43 33 21 13 42 32 22 12 41 31 21 11 40 30 20 10] - [33 23 13 03 32 22 12 02 31 21 11 01 30 20 10 00]
-    movd            xm2, [r0 + r1]                  ; m2 = row5
-    punpcklbw       xm5, xm2                        ; m5 = [53 43 52 42 51 41 50 40]
-    punpcklwd       xm3, xm5                        ; m3 = [53 43 44 23 52 42 32 22 51 41 31 21 50 40 30 20]
-    movd            xm6, [r0 + r1 * 2]              ; m6 = row6
-    punpcklbw       xm2, xm6                        ; m2 = [63 53 62 52 61 51 60 50]
-    punpcklwd       xm4, xm2                        ; m4 = [63 53 43 33 62 52 42 32 61 51 41 31 60 50 40 30]
-    vinserti128     m3, m3, xm4, 1                  ; m3 = [63 53 43 33 62 52 42 32 61 51 41 31 60 50 40 30] - [53 43 44 23 52 42 32 22 51 41 31 21 50 40 30 20]
-    movd            xm4, [r0 + r5]                  ; m4 = row7
-    punpcklbw       xm6, xm4                        ; m6 = [73 63 72 62 71 61 70 60]
-    punpcklwd       xm5, xm6                        ; m5 = [73 63 53 43 72 62 52 42 71 61 51 41 70 60 50 40]
-
+    movu            xm2, [r0]                       ; m2 = row 16
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m3, m1, [r5 + 3 * mmsize]
+    paddw           m9, m3
+    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
+    paddw           m11, m3
+    pmaddubsw       m3, m1, [r5 + 1 * mmsize]
+    paddw           m13, m3
+    pmaddubsw       m1, [r5]
+    movu            xm3, [r0 + r1]                  ; m3 = row 17
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m4, m2, [r5 + 3 * mmsize]
+    paddw           m10, m4
+    pmaddubsw       m4, m2, [r5 + 2 * mmsize]
+    paddw           m12, m4
+    pmaddubsw       m2, [r5 + 1 * mmsize]
+    paddw           m0, m2
+    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddubsw       m5, m3, [r5 + 3 * mmsize]
+    paddw           m11, m5
+    pmaddubsw       m5, m3, [r5 + 2 * mmsize]
+    paddw           m13, m5
+    pmaddubsw       m3, [r5 + 1 * mmsize]
+    paddw           m1, m3
+    movu            xm5, [r0 + r4]                  ; m5 = row 19
+    punpckhbw       xm6, xm4, xm5
+    punpcklbw       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddubsw       m6, m4, [r5 + 3 * mmsize]
+    paddw           m12, m6
+    pmaddubsw       m4, [r5 + 2 * mmsize]
+    paddw           m0, m4
     lea             r0, [r0 + r1 * 4]
-    movd            xm7, [r0]                       ; m7 = row8
-    punpcklbw       xm4, xm7                        ; m4 = [83 73 82 72 81 71 80 70]
-    punpcklwd       xm2, xm4                        ; m2 = [83 73 63 53 82 72 62 52 81 71 61 51 80 70 60 50]
-    vinserti128     m5, m5, xm2, 1                  ; m5 = [83 73 63 53 82 72 62 52 81 71 61 51 80 70 60 50] - [73 63 53 43 72 62 52 42 71 61 51 41 70 60 50 40]
-    movd            xm2, [r0 + r1]                  ; m2 = row9
-    punpcklbw       xm7, xm2                        ; m7 = [93 83 92 82 91 81 90 80]
-    punpcklwd       xm6, xm7                        ; m6 = [93 83 73 63 92 82 72 62 91 81 71 61 90 80 70 60]
-    movd            xm7, [r0 + r1 * 2]              ; m7 = rowA
-    punpcklbw       xm2, xm7                        ; m2 = [A3 93 A2 92 A1 91 A0 90]
-    punpcklwd       xm4, xm2                        ; m4 = [A3 93 83 73 A2 92 82 72 A1 91 81 71 A0 90 80 70]
-    vinserti128     m6, m6, xm4, 1                  ; m6 = [A3 93 83 73 A2 92 82 72 A1 91 81 71 A0 90 80 70] - [93 83 73 63 92 82 72 62 91 81 71 61 90 80 70 60]
+    movu            xm6, [r0]                       ; m6 = row 20
+    punpckhbw       xm7, xm5, xm6
+    punpcklbw       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddubsw       m7, m5, [r5 + 3 * mmsize]
+    paddw           m13, m7
+    pmaddubsw       m5, [r5 + 2 * mmsize]
+    paddw           m1, m5
+    movu            xm7, [r0 + r1]                  ; m7 = row 21
+    punpckhbw       xm2, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm2, 1
+    pmaddubsw       m6, [r5 + 3 * mmsize]
+    paddw           m0, m6
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 22
+    punpckhbw       xm3, xm7, xm2
+    punpcklbw       xm7, xm2
+    vinserti128     m7, m7, xm3, 1
+    pmaddubsw       m7, [r5 + 3 * mmsize]
+    paddw           m1, m7
 
-    ; load filter coeff
-%ifdef PIC
-    lea             r5, [tab_LumaCoeff]
-    vpbroadcastd    m0, [r5 + r4 * 8 + 0]
-    vpbroadcastd    m2, [r5 + r4 * 8 + 4]
+%ifidn %1,pp
+    pmulhrsw        m8, m14                         ; m8 = word: row 8
+    pmulhrsw        m9, m14                         ; m9 = word: row 9
+    pmulhrsw        m10, m14                        ; m10 = word: row 10
+    pmulhrsw        m11, m14                        ; m11 = word: row 11
+    pmulhrsw        m12, m14                        ; m12 = word: row 12
+    pmulhrsw        m13, m14                        ; m13 = word: row 13
+    pmulhrsw        m0, m14                         ; m0 = word: row 14
+    pmulhrsw        m1, m14                         ; m1 = word: row 15
+    packuswb        m8, m9
+    packuswb        m10, m11
+    packuswb        m12, m13
+    packuswb        m0, m1
+    vpermq          m8, m8, 11011000b
+    vpermq          m10, m10, 11011000b
+    vpermq          m12, m12, 11011000b
+    vpermq          m0, m0, 11011000b
+    vextracti128    xm9, m8, 1
+    vextracti128    xm11, m10, 1
+    vextracti128    xm13, m12, 1
+    vextracti128    xm1, m0, 1
+    movq            [r2], xm8
+    pextrd          [r2 + 8], xm8, 2
+    movq            [r2 + r3], xm9
+    pextrd          [r2 + r3 + 8], xm9, 2
+    movq            [r2 + r3 * 2], xm10
+    pextrd          [r2 + r3 * 2 + 8], xm10, 2
+    movq            [r2 + r6], xm11
+    pextrd          [r2 + r6 + 8], xm11, 2
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm12
+    pextrd          [r2 + 8], xm12, 2
+    movq            [r2 + r3], xm13
+    pextrd          [r2 + r3 + 8], xm13, 2
+    movq            [r2 + r3 * 2], xm0
+    pextrd          [r2 + r3 * 2 + 8], xm0, 2
+    movq            [r2 + r6], xm1
+    pextrd          [r2 + r6 + 8], xm1, 2
 %else
-    vpbroadcastd    m0, [tab_LumaCoeff + r4 * 8 + 0]
-    vpbroadcastd    m2, [tab_LumaCoeff + r4 * 8 + 4]
+    psubw           m8, m14                         ; m8 = word: row 8
+    psubw           m9, m14                         ; m9 = word: row 9
+    psubw           m10, m14                        ; m10 = word: row 10
+    psubw           m11, m14                        ; m11 = word: row 11
+    psubw           m12, m14                        ; m12 = word: row 12
+    psubw           m13, m14                        ; m13 = word: row 13
+    psubw           m0, m14                         ; m0 = word: row 14
+    psubw           m1, m14                         ; m1 = word: row 15
+    movu            [r2], xm8
+    vextracti128    xm8, m8, 1
+    movq            [r2 + 16], xm8
+    movu            [r2 + r3], xm9
+    vextracti128    xm9, m9, 1
+    movq            [r2 + r3 + 16], xm9
+    movu            [r2 + r3 * 2], xm10
+    vextracti128    xm10, m10, 1
+    movq            [r2 + r3 * 2 + 16], xm10
+    movu            [r2 + r6], xm11
+    vextracti128    xm11, m11, 1
+    movq            [r2 + r6 + 16], xm11
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm12
+    vextracti128    xm12, m12, 1
+    movq            [r2 + 16], xm12
+    movu            [r2 + r3], xm13
+    vextracti128    xm13, m13, 1
+    movq            [r2 + r3 + 16], xm13
+    movu            [r2 + r3 * 2], xm0
+    vextracti128    xm0, m0, 1
+    movq            [r2 + r3 * 2 + 16], xm0
+    movu            [r2 + r6], xm1
+    vextracti128    xm1, m1, 1
+    movq            [r2 + r6 + 16], xm1
 %endif
-
-    pmaddubsw       m1, m0
-    pmaddubsw       m3, m0
-    pmaddubsw       m5, m2
-    pmaddubsw       m6, m2
-    vbroadcasti128  m0, [pw_1]
-    pmaddwd         m1, m0
-    pmaddwd         m3, m0
-    pmaddwd         m5, m0
-    pmaddwd         m6, m0
-    paddd           m1, m5                          ; m1 = DQWORD ROW[1 0]
-    paddd           m3, m6                          ; m3 = DQWORD ROW[3 2]
-    packssdw        m1, m3                          ; m1 =  QWORD ROW[3 1 2 0]
-
-    ; TODO: does it overflow?
-    pmulhrsw        m1, [pw_512]
-    vextracti128    xm2, m1, 1
-    packuswb        xm1, xm2                        ; m1 =  DWORD ROW[3 1 2 0]
-    movd            [r2], xm1
-    pextrd          [r2 + r3], xm1, 2
-    pextrd          [r2 + r3 * 2], xm1, 1
-    lea             r4, [r3 * 3]
-    pextrd          [r2 + r4], xm1, 3
     RET
+%endif
+%endmacro
+
+    FILTER_VER_LUMA_AVX2_12x16 pp
+    FILTER_VER_LUMA_AVX2_12x16 ps
 
+%macro FILTER_VER_LUMA_AVX2_16x16 1
 INIT_YMM avx2
-cglobal interp_8tap_vert_ps_4x4, 4, 6, 5
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_16x16, 4, 7, 15
     mov             r4d, r4m
     shl             r4d, 7
 
@@ -14063,754 +12411,629 @@ cglobal interp_8tap_vert_ps_4x4, 4, 6, 5
 
     lea             r4, [r1 * 3]
     sub             r0, r4
-
+%ifidn %1,pp
+    mova            m14, [pw_512]
+%else
     add             r3d, r3d
+    vbroadcasti128  m14, [pw_2000]
+%endif
+    lea             r6, [r3 * 3]
 
-    movd            xm1, [r0]
-    pinsrd          xm1, [r0 + r1], 1
-    pinsrd          xm1, [r0 + r1 * 2], 2
-    pinsrd          xm1, [r0 + r4], 3                       ; m1 = row[3 2 1 0]
-    lea             r0, [r0 + r1 * 4]
-    movd            xm2, [r0]
-    pinsrd          xm2, [r0 + r1], 1
-    pinsrd          xm2, [r0 + r1 * 2], 2
-    pinsrd          xm2, [r0 + r4], 3                       ; m2 = row[7 6 5 4]
-    vinserti128     m1, m1, xm2, 1                          ; m1 = row[7 6 5 4 3 2 1 0]
-    lea             r0, [r0 + r1 * 4]
-    movd            xm3, [r0]
-    pinsrd          xm3, [r0 + r1], 1
-    pinsrd          xm3, [r0 + r1 * 2], 2                   ; m3 = row[x 10 9 8]
-    vinserti128     m2, m2, xm3, 1                          ; m2 = row[x 10 9 8 7 6 5 4]
-    mova            m3, [interp4_vpp_shuf1]
-    vpermd          m0, m3, m1                              ; m0 = row[4 3 3 2 2 1 1 0]
-    vpermd          m4, m3, m2                              ; m4 = row[8 7 7 6 6 5 5 4]
-    mova            m3, [interp4_vpp_shuf1 + mmsize]
-    vpermd          m1, m3, m1                              ; m1 = row[6 5 5 4 4 3 3 2]
-    vpermd          m2, m3, m2                              ; m2 = row[10 9 9 8 8 7 7 6]
-
-    mova            m3, [interp4_vpp_shuf]
-    pshufb          m0, m0, m3
-    pshufb          m1, m1, m3
-    pshufb          m4, m4, m3
-    pshufb          m2, m2, m3
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
     pmaddubsw       m0, [r5]
-    pmaddubsw       m1, [r5 + mmsize]
-    pmaddubsw       m4, [r5 + 2 * mmsize]
-    pmaddubsw       m2, [r5 + 3 * mmsize]
-    paddw           m0, m1
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
     paddw           m0, m4
-    paddw           m0, m2                                  ; m0 = WORD ROW[3 2 1 0]
-
-    psubw           m0, [pw_2000]
-    vextracti128    xm2, m0, 1
-    lea             r5, [r3 * 3]
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r5], xm2
-    RET
-
-%macro FILTER_VER_LUMA_AVX2_4xN 3
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%3_%1x%2, 4, 9, 10
-    mov             r4d, r4m
-    shl             r4d, 7
+    pmaddubsw       m2, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
+    paddw           m1, m5
+    pmaddubsw       m3, [r5]
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhbw       xm6, xm4, xm5
+    punpcklbw       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
+    paddw           m0, m6
+    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
+    paddw           m2, m6
+    pmaddubsw       m4, [r5]
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhbw       xm7, xm5, xm6
+    punpcklbw       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
+    paddw           m1, m7
+    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
+    paddw           m3, m7
+    pmaddubsw       m5, [r5]
+    movu            xm7, [r0 + r4]                  ; m7 = row 7
+    punpckhbw       xm8, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
+    paddw           m0, m8
+    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
+    paddw           m2, m8
+    pmaddubsw       m8, m6, [r5 + 1 * mmsize]
+    paddw           m4, m8
+    pmaddubsw       m6, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm8, [r0]                       ; m8 = row 8
+    punpckhbw       xm9, xm7, xm8
+    punpcklbw       xm7, xm8
+    vinserti128     m7, m7, xm9, 1
+    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
+    paddw           m1, m9
+    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
+    paddw           m3, m9
+    pmaddubsw       m9, m7, [r5 + 1 * mmsize]
+    paddw           m5, m9
+    pmaddubsw       m7, [r5]
+    movu            xm9, [r0 + r1]                  ; m9 = row 9
+    punpckhbw       xm10, xm8, xm9
+    punpcklbw       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
+    paddw           m2, m10
+    pmaddubsw       m10, m8, [r5 + 2 * mmsize]
+    paddw           m4, m10
+    pmaddubsw       m10, m8, [r5 + 1 * mmsize]
+    paddw           m6, m10
+    pmaddubsw       m8, [r5]
+    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
+    punpckhbw       xm11, xm9, xm10
+    punpcklbw       xm9, xm10
+    vinserti128     m9, m9, xm11, 1
+    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
+    paddw           m3, m11
+    pmaddubsw       m11, m9, [r5 + 2 * mmsize]
+    paddw           m5, m11
+    pmaddubsw       m11, m9, [r5 + 1 * mmsize]
+    paddw           m7, m11
+    pmaddubsw       m9, [r5]
+    movu            xm11, [r0 + r4]                 ; m11 = row 11
+    punpckhbw       xm12, xm10, xm11
+    punpcklbw       xm10, xm11
+    vinserti128     m10, m10, xm12, 1
+    pmaddubsw       m12, m10, [r5 + 3 * mmsize]
+    paddw           m4, m12
+    pmaddubsw       m12, m10, [r5 + 2 * mmsize]
+    paddw           m6, m12
+    pmaddubsw       m12, m10, [r5 + 1 * mmsize]
+    paddw           m8, m12
+    pmaddubsw       m10, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm12, [r0]                      ; m12 = row 12
+    punpckhbw       xm13, xm11, xm12
+    punpcklbw       xm11, xm12
+    vinserti128     m11, m11, xm13, 1
+    pmaddubsw       m13, m11, [r5 + 3 * mmsize]
+    paddw           m5, m13
+    pmaddubsw       m13, m11, [r5 + 2 * mmsize]
+    paddw           m7, m13
+    pmaddubsw       m13, m11, [r5 + 1 * mmsize]
+    paddw           m9, m13
+    pmaddubsw       m11, [r5]
 
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
+%ifidn %1,pp
+    pmulhrsw        m0, m14                         ; m0 = word: row 0
+    pmulhrsw        m1, m14                         ; m1 = word: row 1
+    pmulhrsw        m2, m14                         ; m2 = word: row 2
+    pmulhrsw        m3, m14                         ; m3 = word: row 3
+    pmulhrsw        m4, m14                         ; m4 = word: row 4
+    pmulhrsw        m5, m14                         ; m5 = word: row 5
+    packuswb        m0, m1
+    packuswb        m2, m3
+    packuswb        m4, m5
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    vpermq          m4, m4, 11011000b
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m2, 1
+    vextracti128    xm5, m4, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm4
+    movu            [r2 + r3], xm5
 %else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
+    psubw           m0, m14                         ; m0 = word: row 0
+    psubw           m1, m14                         ; m1 = word: row 1
+    psubw           m2, m14                         ; m2 = word: row 2
+    psubw           m3, m14                         ; m3 = word: row 3
+    psubw           m4, m14                         ; m4 = word: row 4
+    psubw           m5, m14                         ; m5 = word: row 5
+    movu            [r2], m0
+    movu            [r2 + r3], m1
+    movu            [r2 + r3 * 2], m2
+    movu            [r2 + r6], m3
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], m4
+    movu            [r2 + r3], m5
 %endif
 
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-    lea             r6, [r1 * 4]
-%ifidn %3,pp
-    mova            m6, [pw_512]
+    movu            xm13, [r0 + r1]                 ; m13 = row 13
+    punpckhbw       xm0, xm12, xm13
+    punpcklbw       xm12, xm13
+    vinserti128     m12, m12, xm0, 1
+    pmaddubsw       m0, m12, [r5 + 3 * mmsize]
+    paddw           m6, m0
+    pmaddubsw       m0, m12, [r5 + 2 * mmsize]
+    paddw           m8, m0
+    pmaddubsw       m0, m12, [r5 + 1 * mmsize]
+    paddw           m10, m0
+    pmaddubsw       m12, [r5]
+    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
+    punpckhbw       xm1, xm13, xm0
+    punpcklbw       xm13, xm0
+    vinserti128     m13, m13, xm1, 1
+    pmaddubsw       m1, m13, [r5 + 3 * mmsize]
+    paddw           m7, m1
+    pmaddubsw       m1, m13, [r5 + 2 * mmsize]
+    paddw           m9, m1
+    pmaddubsw       m1, m13, [r5 + 1 * mmsize]
+    paddw           m11, m1
+    pmaddubsw       m13, [r5]
+
+%ifidn %1,pp
+    pmulhrsw        m6, m14                         ; m6 = word: row 6
+    pmulhrsw        m7, m14                         ; m7 = word: row 7
+    packuswb        m6, m7
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm7, m6, 1
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r6], xm7
 %else
-    add             r3d, r3d
-    vbroadcasti128  m6, [pw_2000]
+    psubw           m6, m14                         ; m6 = word: row 6
+    psubw           m7, m14                         ; m7 = word: row 7
+    movu            [r2 + r3 * 2], m6
+    movu            [r2 + r6], m7
 %endif
-    lea             r8, [r3 * 3]
-    mova            m5, [interp4_vpp_shuf]
-    mova            m0, [interp4_vpp_shuf1]
-    mova            m7, [interp4_vpp_shuf1 + mmsize]
-    mov             r7d, %2 / 8
-.loop:
-    movd            xm1, [r0]
-    pinsrd          xm1, [r0 + r1], 1
-    pinsrd          xm1, [r0 + r1 * 2], 2
-    pinsrd          xm1, [r0 + r4], 3                       ; m1 = row[3 2 1 0]
-    lea             r0, [r0 + r1 * 4]
-    movd            xm2, [r0]
-    pinsrd          xm2, [r0 + r1], 1
-    pinsrd          xm2, [r0 + r1 * 2], 2
-    pinsrd          xm2, [r0 + r4], 3                       ; m2 = row[7 6 5 4]
-    vinserti128     m1, m1, xm2, 1                          ; m1 = row[7 6 5 4 3 2 1 0]
-    lea             r0, [r0 + r1 * 4]
-    movd            xm3, [r0]
-    pinsrd          xm3, [r0 + r1], 1
-    pinsrd          xm3, [r0 + r1 * 2], 2
-    pinsrd          xm3, [r0 + r4], 3                       ; m3 = row[11 10 9 8]
-    vinserti128     m2, m2, xm3, 1                          ; m2 = row[11 10 9 8 7 6 5 4]
-    lea             r0, [r0 + r1 * 4]
-    movd            xm4, [r0]
-    pinsrd          xm4, [r0 + r1], 1
-    pinsrd          xm4, [r0 + r1 * 2], 2                   ; m4 = row[x 14 13 12]
-    vinserti128     m3, m3, xm4, 1                          ; m3 = row[x 14 13 12 11 10 9 8]
-    vpermd          m8, m0, m1                              ; m8 = row[4 3 3 2 2 1 1 0]
-    vpermd          m4, m0, m2                              ; m4 = row[8 7 7 6 6 5 5 4]
-    vpermd          m1, m7, m1                              ; m1 = row[6 5 5 4 4 3 3 2]
-    vpermd          m2, m7, m2                              ; m2 = row[10 9 9 8 8 7 7 6]
-    vpermd          m9, m0, m3                              ; m9 = row[12 11 11 10 10 9 9 8]
-    vpermd          m3, m7, m3                              ; m3 = row[14 13 13 12 12 11 11 10]
+    lea             r2, [r2 + r3 * 4]
 
-    pshufb          m8, m8, m5
-    pshufb          m1, m1, m5
-    pshufb          m4, m4, m5
-    pshufb          m9, m9, m5
-    pshufb          m2, m2, m5
-    pshufb          m3, m3, m5
-    pmaddubsw       m8, [r5]
-    pmaddubsw       m1, [r5 + mmsize]
-    pmaddubsw       m9, [r5 + 2 * mmsize]
-    pmaddubsw       m3, [r5 + 3 * mmsize]
-    paddw           m8, m1
+    movu            xm1, [r0 + r4]                  ; m1 = row 15
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddubsw       m2, m0, [r5 + 3 * mmsize]
+    paddw           m8, m2
+    pmaddubsw       m2, m0, [r5 + 2 * mmsize]
+    paddw           m10, m2
+    pmaddubsw       m2, m0, [r5 + 1 * mmsize]
+    paddw           m12, m2
+    pmaddubsw       m0, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm2, [r0]                       ; m2 = row 16
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m3, m1, [r5 + 3 * mmsize]
     paddw           m9, m3
-    pmaddubsw       m1, m4, [r5 + 2 * mmsize]
-    pmaddubsw       m3, m2, [r5 + 3 * mmsize]
-    pmaddubsw       m4, [r5]
-    pmaddubsw       m2, [r5 + mmsize]
-    paddw           m3, m1
-    paddw           m2, m4
-    paddw           m8, m3                                  ; m8 = WORD ROW[3 2 1 0]
-    paddw           m9, m2                                  ; m9 = WORD ROW[7 6 5 4]
+    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
+    paddw           m11, m3
+    pmaddubsw       m3, m1, [r5 + 1 * mmsize]
+    paddw           m13, m3
+    pmaddubsw       m1, [r5]
+    movu            xm3, [r0 + r1]                  ; m3 = row 17
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m4, m2, [r5 + 3 * mmsize]
+    paddw           m10, m4
+    pmaddubsw       m4, m2, [r5 + 2 * mmsize]
+    paddw           m12, m4
+    pmaddubsw       m2, [r5 + 1 * mmsize]
+    paddw           m0, m2
+    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddubsw       m5, m3, [r5 + 3 * mmsize]
+    paddw           m11, m5
+    pmaddubsw       m5, m3, [r5 + 2 * mmsize]
+    paddw           m13, m5
+    pmaddubsw       m3, [r5 + 1 * mmsize]
+    paddw           m1, m3
+    movu            xm5, [r0 + r4]                  ; m5 = row 19
+    punpckhbw       xm6, xm4, xm5
+    punpcklbw       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddubsw       m6, m4, [r5 + 3 * mmsize]
+    paddw           m12, m6
+    pmaddubsw       m4, [r5 + 2 * mmsize]
+    paddw           m0, m4
+    lea             r0, [r0 + r1 * 4]
+    movu            xm6, [r0]                       ; m6 = row 20
+    punpckhbw       xm7, xm5, xm6
+    punpcklbw       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddubsw       m7, m5, [r5 + 3 * mmsize]
+    paddw           m13, m7
+    pmaddubsw       m5, [r5 + 2 * mmsize]
+    paddw           m1, m5
+    movu            xm7, [r0 + r1]                  ; m7 = row 21
+    punpckhbw       xm2, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm2, 1
+    pmaddubsw       m6, [r5 + 3 * mmsize]
+    paddw           m0, m6
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 22
+    punpckhbw       xm3, xm7, xm2
+    punpcklbw       xm7, xm2
+    vinserti128     m7, m7, xm3, 1
+    pmaddubsw       m7, [r5 + 3 * mmsize]
+    paddw           m1, m7
 
-%ifidn %3,pp
-    pmulhrsw        m8, m6
-    pmulhrsw        m9, m6
+%ifidn %1,pp
+    pmulhrsw        m8, m14                         ; m8 = word: row 8
+    pmulhrsw        m9, m14                         ; m9 = word: row 9
+    pmulhrsw        m10, m14                        ; m10 = word: row 10
+    pmulhrsw        m11, m14                        ; m11 = word: row 11
+    pmulhrsw        m12, m14                        ; m12 = word: row 12
+    pmulhrsw        m13, m14                        ; m13 = word: row 13
+    pmulhrsw        m0, m14                         ; m0 = word: row 14
+    pmulhrsw        m1, m14                         ; m1 = word: row 15
     packuswb        m8, m9
-    vextracti128    xm1, m8, 1
-    movd            [r2], xm8
-    pextrd          [r2 + r3], xm8, 1
-    movd            [r2 + r3 * 2], xm1
-    pextrd          [r2 + r8], xm1, 1
+    packuswb        m10, m11
+    packuswb        m12, m13
+    packuswb        m0, m1
+    vpermq          m8, m8, 11011000b
+    vpermq          m10, m10, 11011000b
+    vpermq          m12, m12, 11011000b
+    vpermq          m0, m0, 11011000b
+    vextracti128    xm9, m8, 1
+    vextracti128    xm11, m10, 1
+    vextracti128    xm13, m12, 1
+    vextracti128    xm1, m0, 1
+    movu            [r2], xm8
+    movu            [r2 + r3], xm9
+    movu            [r2 + r3 * 2], xm10
+    movu            [r2 + r6], xm11
     lea             r2, [r2 + r3 * 4]
-    pextrd          [r2], xm8, 2
-    pextrd          [r2 + r3], xm8, 3
-    pextrd          [r2 + r3 * 2], xm1, 2
-    pextrd          [r2 + r8], xm1, 3
+    movu            [r2], xm12
+    movu            [r2 + r3], xm13
+    movu            [r2 + r3 * 2], xm0
+    movu            [r2 + r6], xm1
 %else
-    psubw           m8, m6
-    psubw           m9, m6
-    vextracti128    xm1, m8, 1
-    vextracti128    xm2, m9, 1
-    movq            [r2], xm8
-    movhps          [r2 + r3], xm8
-    movq            [r2 + r3 * 2], xm1
-    movhps          [r2 + r8], xm1
+    psubw           m8, m14                         ; m8 = word: row 8
+    psubw           m9, m14                         ; m9 = word: row 9
+    psubw           m10, m14                        ; m10 = word: row 10
+    psubw           m11, m14                        ; m11 = word: row 11
+    psubw           m12, m14                        ; m12 = word: row 12
+    psubw           m13, m14                        ; m13 = word: row 13
+    psubw           m0, m14                         ; m0 = word: row 14
+    psubw           m1, m14                         ; m1 = word: row 15
+    movu            [r2], m8
+    movu            [r2 + r3], m9
+    movu            [r2 + r3 * 2], m10
+    movu            [r2 + r6], m11
     lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm9
-    movhps          [r2 + r3], xm9
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r8], xm2
+    movu            [r2], m12
+    movu            [r2 + r3], m13
+    movu            [r2 + r3 * 2], m0
+    movu            [r2 + r6], m1
 %endif
-    lea             r2, [r2 + r3 * 4]
-    sub             r0, r6
-    dec             r7d
-    jnz             .loop
     RET
 %endif
 %endmacro
 
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_pp_4x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_4xN 4, 4, pp
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_pp_4x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_4xN 4, 8, pp
-    FILTER_VER_LUMA_AVX2_4xN 4, 8, pp
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_pp_4x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_4xN 4, 16, pp
-    FILTER_VER_LUMA_AVX2_4xN 4, 16, pp
+    FILTER_VER_LUMA_AVX2_16x16 pp
+    FILTER_VER_LUMA_AVX2_16x16 ps
 
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_ps_4x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_4xN 4, 4, ps
+%macro FILTER_VER_LUMA_AVX2_16x12 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_16x12, 4, 7, 15
+    mov             r4d, r4m
+    shl             r4d, 7
 
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_ps_4x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_4xN 4, 8, ps
-    FILTER_VER_LUMA_AVX2_4xN 4, 8, ps
+%ifdef PIC
+    lea             r5, [tab_LumaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
+%endif
 
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_ps_4x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_4xN 4, 16, ps
-    FILTER_VER_LUMA_AVX2_4xN 4, 16, ps
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+%ifidn %1,pp
+    mova            m14, [pw_512]
+%else
+    add             r3d, r3d
+    vbroadcasti128  m14, [pw_2000]
+%endif
+    lea             r6, [r3 * 3]
 
-%macro PROCESS_LUMA_AVX2_W8_8R 0
-    movq            xm1, [r0]                       ; m1 = row 0
-    movq            xm2, [r0 + r1]                  ; m2 = row 1
-    punpcklbw       xm1, xm2                        ; m1 = [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
-    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
-    punpcklbw       xm2, xm3                        ; m2 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10]
-    vinserti128     m5, m1, xm2, 1                  ; m5 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10] - [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
-    pmaddubsw       m5, [r5]
-    movq            xm4, [r0 + r4]                  ; m4 = row 3
-    punpcklbw       xm3, xm4                        ; m3 = [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
-    lea             r0, [r0 + r1 * 4]
-    movq            xm1, [r0]                       ; m1 = row 4
-    punpcklbw       xm4, xm1                        ; m4 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30]
-    vinserti128     m2, m3, xm4, 1                  ; m2 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30] - [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
-    pmaddubsw       m0, m2, [r5 + 1 * mmsize]
-    paddw           m5, m0
-    pmaddubsw       m2, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 5
-    punpcklbw       xm1, xm3                        ; m1 = [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
-    movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
-    punpcklbw       xm3, xm4                        ; m3 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50]
-    vinserti128     m1, m1, xm3, 1                  ; m1 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50] - [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
-    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
-    paddw           m5, m3
-    pmaddubsw       m0, m1, [r5 + 1 * mmsize]
-    paddw           m2, m0
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddubsw       m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
     pmaddubsw       m1, [r5]
-    movq            xm3, [r0 + r4]                  ; m3 = row 7
-    punpcklbw       xm4, xm3                        ; m4 = [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
+    paddw           m0, m4
+    pmaddubsw       m2, [r5]
     lea             r0, [r0 + r1 * 4]
-    movq            xm0, [r0]                       ; m0 = row 8
-    punpcklbw       xm3, xm0                        ; m3 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70]
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70] - [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
-    pmaddubsw       m3, m4, [r5 + 3 * mmsize]
-    paddw           m5, m3
-    pmaddubsw       m3, m4, [r5 + 2 * mmsize]
-    paddw           m2, m3
-    pmaddubsw       m3, m4, [r5 + 1 * mmsize]
-    paddw           m1, m3
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
+    paddw           m1, m5
+    pmaddubsw       m3, [r5]
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhbw       xm6, xm4, xm5
+    punpcklbw       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
+    paddw           m0, m6
+    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
+    paddw           m2, m6
     pmaddubsw       m4, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 9
-    punpcklbw       xm0, xm3                        ; m0 = [97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80]
-    movq            xm6, [r0 + r1 * 2]              ; m6 = row 10
-    punpcklbw       xm3, xm6                        ; m3 = [A7 97 A6 96 A5 95 A4 94 A3 93 A2 92 A1 91 A0 90]
-    vinserti128     m0, m0, xm3, 1                  ; m0 = [A7 97 A6 96 A5 95 A4 94 A3 93 A2 92 A1 91 A0 90] - [97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80]
-    pmaddubsw       m3, m0, [r5 + 3 * mmsize]
-    paddw           m2, m3
-    pmaddubsw       m3, m0, [r5 + 2 * mmsize]
-    paddw           m1, m3
-    pmaddubsw       m0, [r5 + 1 * mmsize]
-    paddw           m4, m0
-
-    movq            xm3, [r0 + r4]                  ; m3 = row 11
-    punpcklbw       xm6, xm3                        ; m6 = [B7 A7 B6 A6 B5 A5 B4 A4 B3 A3 B2 A2 B1 A1 B0 A0]
-    lea             r0, [r0 + r1 * 4]
-    movq            xm0, [r0]                       ; m0 = row 12
-    punpcklbw       xm3, xm0                        ; m3 = [C7 B7 C6 B6 C5 B5 C4 B4 C3 B3 C2 B2 C1 B1 C0 B0]
-    vinserti128     m6, m6, xm3, 1                  ; m6 = [C7 B7 C6 B6 C5 B5 C4 B4 C3 B3 C2 B2 C1 B1 C0 B0] - [B7 A7 B6 A6 B5 A5 B4 A4 B3 A3 B2 A2 B1 A1 B0 A0]
-    pmaddubsw       m3, m6, [r5 + 3 * mmsize]
-    paddw           m1, m3
-    pmaddubsw       m6, [r5 + 2 * mmsize]
-    paddw           m4, m6
-    movq            xm3, [r0 + r1]                  ; m3 = row 13
-    punpcklbw       xm0, xm3                        ; m0 = [D7 C7 D6 C6 D5 C5 D4 C4 D3 C3 D2 C2 D1 C1 D0 C0]
-    movq            xm6, [r0 + r1 * 2]              ; m6 = row 14
-    punpcklbw       xm3, xm6                        ; m3 = [E7 D7 E6 D6 E5 D5 E4 D4 E3 D3 E2 D2 E1 D1 E0 D0]
-    vinserti128     m0, m0, xm3, 1                  ; m0 = [E7 D7 E6 D6 E5 D5 E4 D4 E3 D3 E2 D2 E1 D1 E0 D0] - [D7 C7 D6 C6 D5 C5 D4 C4 D3 C3 D2 C2 D1 C1 D0 C0]
-    pmaddubsw       m0, [r5 + 3 * mmsize]
-    paddw           m4, m0
-%endmacro
-
-%macro PROCESS_LUMA_AVX2_W8_4R 0
-    movq            xm1, [r0]                       ; m1 = row 0
-    movq            xm2, [r0 + r1]                  ; m2 = row 1
-    punpcklbw       xm1, xm2                        ; m1 = [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
-    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
-    punpcklbw       xm2, xm3                        ; m2 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10]
-    vinserti128     m5, m1, xm2, 1                  ; m5 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10] - [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhbw       xm7, xm5, xm6
+    punpcklbw       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
+    paddw           m1, m7
+    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
+    paddw           m3, m7
     pmaddubsw       m5, [r5]
-    movq            xm4, [r0 + r4]                  ; m4 = row 3
-    punpcklbw       xm3, xm4                        ; m3 = [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
+    movu            xm7, [r0 + r4]                  ; m7 = row 7
+    punpckhbw       xm8, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
+    paddw           m0, m8
+    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
+    paddw           m2, m8
+    pmaddubsw       m8, m6, [r5 + 1 * mmsize]
+    paddw           m4, m8
+    pmaddubsw       m6, [r5]
     lea             r0, [r0 + r1 * 4]
-    movq            xm1, [r0]                       ; m1 = row 4
-    punpcklbw       xm4, xm1                        ; m4 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30]
-    vinserti128     m2, m3, xm4, 1                  ; m2 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30] - [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
-    pmaddubsw       m0, m2, [r5 + 1 * mmsize]
-    paddw           m5, m0
-    pmaddubsw       m2, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 5
-    punpcklbw       xm1, xm3                        ; m1 = [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
-    movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
-    punpcklbw       xm3, xm4                        ; m3 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50]
-    vinserti128     m1, m1, xm3, 1                  ; m1 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50] - [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
-    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
-    paddw           m5, m3
-    pmaddubsw       m0, m1, [r5 + 1 * mmsize]
-    paddw           m2, m0
-    movq            xm3, [r0 + r4]                  ; m3 = row 7
-    punpcklbw       xm4, xm3                        ; m4 = [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
+    movu            xm8, [r0]                       ; m8 = row 8
+    punpckhbw       xm9, xm7, xm8
+    punpcklbw       xm7, xm8
+    vinserti128     m7, m7, xm9, 1
+    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
+    paddw           m1, m9
+    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
+    paddw           m3, m9
+    pmaddubsw       m9, m7, [r5 + 1 * mmsize]
+    paddw           m5, m9
+    pmaddubsw       m7, [r5]
+    movu            xm9, [r0 + r1]                  ; m9 = row 9
+    punpckhbw       xm10, xm8, xm9
+    punpcklbw       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
+    paddw           m2, m10
+    pmaddubsw       m10, m8, [r5 + 2 * mmsize]
+    paddw           m4, m10
+    pmaddubsw       m10, m8, [r5 + 1 * mmsize]
+    paddw           m6, m10
+    pmaddubsw       m8, [r5]
+    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
+    punpckhbw       xm11, xm9, xm10
+    punpcklbw       xm9, xm10
+    vinserti128     m9, m9, xm11, 1
+    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
+    paddw           m3, m11
+    pmaddubsw       m11, m9, [r5 + 2 * mmsize]
+    paddw           m5, m11
+    pmaddubsw       m11, m9, [r5 + 1 * mmsize]
+    paddw           m7, m11
+    pmaddubsw       m9, [r5]
+    movu            xm11, [r0 + r4]                 ; m11 = row 11
+    punpckhbw       xm12, xm10, xm11
+    punpcklbw       xm10, xm11
+    vinserti128     m10, m10, xm12, 1
+    pmaddubsw       m12, m10, [r5 + 3 * mmsize]
+    paddw           m4, m12
+    pmaddubsw       m12, m10, [r5 + 2 * mmsize]
+    paddw           m6, m12
+    pmaddubsw       m12, m10, [r5 + 1 * mmsize]
+    paddw           m8, m12
+    pmaddubsw       m10, [r5]
     lea             r0, [r0 + r1 * 4]
-    movq            xm0, [r0]                       ; m0 = row 8
-    punpcklbw       xm3, xm0                        ; m3 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70]
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70] - [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
-    pmaddubsw       m3, m4, [r5 + 3 * mmsize]
-    paddw           m5, m3
-    pmaddubsw       m3, m4, [r5 + 2 * mmsize]
-    paddw           m2, m3
-    movq            xm3, [r0 + r1]                  ; m3 = row 9
-    punpcklbw       xm0, xm3                        ; m0 = [97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80]
-    movq            xm6, [r0 + r1 * 2]              ; m6 = row 10
-    punpcklbw       xm3, xm6                        ; m3 = [A7 97 A6 96 A5 95 A4 94 A3 93 A2 92 A1 91 A0 90]
-    vinserti128     m0, m0, xm3, 1                  ; m0 = [A7 97 A6 96 A5 95 A4 94 A3 93 A2 92 A1 91 A0 90] - [97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80]
-    pmaddubsw       m3, m0, [r5 + 3 * mmsize]
-    paddw           m2, m3
-%endmacro
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_%3_8x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_LUMA_8xN 3
-INIT_XMM sse4
-cglobal interp_8tap_vert_%3_%1x%2, 5, 7, 8
-    lea       r5, [3 * r1]
-    sub       r0, r5
-    shl       r4d, 6
-
-%ifidn %3,ps
-    add       r3d, r3d
-%endif
-
-%ifdef PIC
-    lea       r5, [tab_LumaCoeffVer]
-    lea       r6, [r5 + r4]
-%else
-    lea       r6, [tab_LumaCoeffVer + r4]
-%endif
+    movu            xm12, [r0]                      ; m12 = row 12
+    punpckhbw       xm13, xm11, xm12
+    punpcklbw       xm11, xm12
+    vinserti128     m11, m11, xm13, 1
+    pmaddubsw       m13, m11, [r5 + 3 * mmsize]
+    paddw           m5, m13
+    pmaddubsw       m13, m11, [r5 + 2 * mmsize]
+    paddw           m7, m13
+    pmaddubsw       m13, m11, [r5 + 1 * mmsize]
+    paddw           m9, m13
+    pmaddubsw       m11, [r5]
 
- %ifidn %3,pp
-    mova      m3, [pw_512]
+%ifidn %1,pp
+    pmulhrsw        m0, m14                         ; m0 = word: row 0
+    pmulhrsw        m1, m14                         ; m1 = word: row 1
+    pmulhrsw        m2, m14                         ; m2 = word: row 2
+    pmulhrsw        m3, m14                         ; m3 = word: row 3
+    pmulhrsw        m4, m14                         ; m4 = word: row 4
+    pmulhrsw        m5, m14                         ; m5 = word: row 5
+    packuswb        m0, m1
+    packuswb        m2, m3
+    packuswb        m4, m5
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    vpermq          m4, m4, 11011000b
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m2, 1
+    vextracti128    xm5, m4, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm4
+    movu            [r2 + r3], xm5
 %else
-    mova      m3, [pw_2000]
+    psubw           m0, m14                         ; m0 = word: row 0
+    psubw           m1, m14                         ; m1 = word: row 1
+    psubw           m2, m14                         ; m2 = word: row 2
+    psubw           m3, m14                         ; m3 = word: row 3
+    psubw           m4, m14                         ; m4 = word: row 4
+    psubw           m5, m14                         ; m5 = word: row 5
+    movu            [r2], m0
+    movu            [r2 + r3], m1
+    movu            [r2 + r3 * 2], m2
+    movu            [r2 + r6], m3
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], m4
+    movu            [r2 + r3], m5
 %endif
 
-    mov       r4d, %2/4
-    lea       r5, [4 * r1]
-
-.loopH:
-    PROCESS_LUMA_W8_4R
-
-%ifidn %3,pp
-    pmulhrsw  m7, m3
-    pmulhrsw  m6, m3
-    pmulhrsw  m5, m3
-    pmulhrsw  m4, m3
-
-    packuswb  m7, m6
-    packuswb  m5, m4
+    movu            xm13, [r0 + r1]                 ; m13 = row 13
+    punpckhbw       xm0, xm12, xm13
+    punpcklbw       xm12, xm13
+    vinserti128     m12, m12, xm0, 1
+    pmaddubsw       m0, m12, [r5 + 3 * mmsize]
+    paddw           m6, m0
+    pmaddubsw       m0, m12, [r5 + 2 * mmsize]
+    paddw           m8, m0
+    pmaddubsw       m0, m12, [r5 + 1 * mmsize]
+    paddw           m10, m0
+    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
+    punpckhbw       xm1, xm13, xm0
+    punpcklbw       xm13, xm0
+    vinserti128     m13, m13, xm1, 1
+    pmaddubsw       m1, m13, [r5 + 3 * mmsize]
+    paddw           m7, m1
+    pmaddubsw       m1, m13, [r5 + 2 * mmsize]
+    paddw           m9, m1
+    pmaddubsw       m1, m13, [r5 + 1 * mmsize]
+    paddw           m11, m1
 
-    movlps    [r2], m7
-    movhps    [r2 + r3], m7
-    lea       r2, [r2 + 2 * r3]
-    movlps    [r2], m5
-    movhps    [r2 + r3], m5
+%ifidn %1,pp
+    pmulhrsw        m6, m14                         ; m6 = word: row 6
+    pmulhrsw        m7, m14                         ; m7 = word: row 7
+    packuswb        m6, m7
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm7, m6, 1
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r6], xm7
 %else
-    psubw     m7, m3
-    psubw     m6, m3
-    psubw     m5, m3
-    psubw     m4, m3
-
-    movu      [r2], m7
-    movu      [r2 + r3], m6
-    lea       r2, [r2 + 2 * r3]
-    movu      [r2], m5
-    movu      [r2 + r3], m4
+    psubw           m6, m14                         ; m6 = word: row 6
+    psubw           m7, m14                         ; m7 = word: row 7
+    movu            [r2 + r3 * 2], m6
+    movu            [r2 + r6], m7
 %endif
+    lea             r2, [r2 + r3 * 4]
 
-    sub       r0, r5
-    lea       r2, [r2 + 2 * r3]
-
-    dec       r4d
-    jnz       .loopH
-
-    RET
-%endmacro
-
-%macro FILTER_VER_LUMA_AVX2_8xN 3
-INIT_YMM avx2
-cglobal interp_8tap_vert_%3_%1x%2, 4, 7, 8, 0-gprsize
-    mov             r4d, r4m
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-    lea             r6, [r1 * 4]
-%ifidn %3,pp
-    mova            m7, [pw_512]
-%else
-    add             r3d, r3d
-    vbroadcasti128  m7, [pw_2000]
-%endif
-    mov             word [rsp], %2 / 8
+    movu            xm1, [r0 + r4]                  ; m1 = row 15
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddubsw       m2, m0, [r5 + 3 * mmsize]
+    paddw           m8, m2
+    pmaddubsw       m2, m0, [r5 + 2 * mmsize]
+    paddw           m10, m2
+    lea             r0, [r0 + r1 * 4]
+    movu            xm2, [r0]                       ; m2 = row 16
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m3, m1, [r5 + 3 * mmsize]
+    paddw           m9, m3
+    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
+    paddw           m11, m3
+    movu            xm3, [r0 + r1]                  ; m3 = row 17
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m4, m2, [r5 + 3 * mmsize]
+    paddw           m10, m4
+    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddubsw       m5, m3, [r5 + 3 * mmsize]
+    paddw           m11, m5
 
-.loop:
-    PROCESS_LUMA_AVX2_W8_8R
-%ifidn %3,pp
-    pmulhrsw        m5, m7                          ; m5 = word: row 0, row 1
-    pmulhrsw        m2, m7                          ; m2 = word: row 2, row 3
-    pmulhrsw        m1, m7                          ; m1 = word: row 4, row 5
-    pmulhrsw        m4, m7                          ; m4 = word: row 6, row 7
-    packuswb        m5, m2
-    packuswb        m1, m4
-    vextracti128    xm2, m5, 1
-    vextracti128    xm4, m1, 1
-    movq            [r2], xm5
-    movq            [r2 + r3], xm2
-    lea             r2, [r2 + r3 * 2]
-    movhps          [r2], xm5
-    movhps          [r2 + r3], xm2
-    lea             r2, [r2 + r3 * 2]
-    movq            [r2], xm1
-    movq            [r2 + r3], xm4
-    lea             r2, [r2 + r3 * 2]
-    movhps          [r2], xm1
-    movhps          [r2 + r3], xm4
+%ifidn %1,pp
+    pmulhrsw        m8, m14                         ; m8 = word: row 8
+    pmulhrsw        m9, m14                         ; m9 = word: row 9
+    pmulhrsw        m10, m14                        ; m10 = word: row 10
+    pmulhrsw        m11, m14                        ; m11 = word: row 11
+    packuswb        m8, m9
+    packuswb        m10, m11
+    vpermq          m8, m8, 11011000b
+    vpermq          m10, m10, 11011000b
+    vextracti128    xm9, m8, 1
+    vextracti128    xm11, m10, 1
+    movu            [r2], xm8
+    movu            [r2 + r3], xm9
+    movu            [r2 + r3 * 2], xm10
+    movu            [r2 + r6], xm11
 %else
-    psubw           m5, m7                          ; m5 = word: row 0, row 1
-    psubw           m2, m7                          ; m2 = word: row 2, row 3
-    psubw           m1, m7                          ; m1 = word: row 4, row 5
-    psubw           m4, m7                          ; m4 = word: row 6, row 7
-    vextracti128    xm6, m5, 1
-    vextracti128    xm3, m2, 1
-    vextracti128    xm0, m1, 1
-    movu            [r2], xm5
-    movu            [r2 + r3], xm6
-    lea             r2, [r2 + r3 * 2]
-    movu            [r2], xm2
-    movu            [r2 + r3], xm3
-    lea             r2, [r2 + r3 * 2]
-    movu            [r2], xm1
-    movu            [r2 + r3], xm0
-    lea             r2, [r2 + r3 * 2]
-    movu            [r2], xm4
-    vextracti128    xm4, m4, 1
-    movu            [r2 + r3], xm4
+    psubw           m8, m14                         ; m8 = word: row 8
+    psubw           m9, m14                         ; m9 = word: row 9
+    psubw           m10, m14                        ; m10 = word: row 10
+    psubw           m11, m14                        ; m11 = word: row 11
+    movu            [r2], m8
+    movu            [r2 + r3], m9
+    movu            [r2 + r3 * 2], m10
+    movu            [r2 + r6], m11
 %endif
-    lea             r2, [r2 + r3 * 2]
-    sub             r0, r6
-    dec             word [rsp]
-    jnz             .loop
     RET
+%endif
 %endmacro
 
-%macro FILTER_VER_LUMA_AVX2_8x8 1
+    FILTER_VER_LUMA_AVX2_16x12 pp
+    FILTER_VER_LUMA_AVX2_16x12 ps
+
+%macro FILTER_VER_LUMA_AVX2_16x8 1
 INIT_YMM avx2
-cglobal interp_8tap_vert_%1_8x8, 4, 6, 7
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_16x8, 4, 6, 15
     mov             r4d, r4m
     shl             r4d, 7
-
 %ifdef PIC
     lea             r5, [tab_LumaCoeffVer_32]
     add             r5, r4
 %else
     lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
-
     lea             r4, [r1 * 3]
     sub             r0, r4
-    PROCESS_LUMA_AVX2_W8_8R
 %ifidn %1,pp
-    mova            m3, [pw_512]
+    mova            m14, [pw_512]
 %else
     add             r3d, r3d
-    vbroadcasti128  m3, [pw_2000]
-%endif
-    lea             r4, [r3 * 3]
-%ifidn %1,pp
-    pmulhrsw        m5, m3                          ; m5 = word: row 0, row 1
-    pmulhrsw        m2, m3                          ; m2 = word: row 2, row 3
-    pmulhrsw        m1, m3                          ; m1 = word: row 4, row 5
-    pmulhrsw        m4, m3                          ; m4 = word: row 6, row 7
-    packuswb        m5, m2
-    packuswb        m1, m4
-    vextracti128    xm2, m5, 1
-    vextracti128    xm4, m1, 1
-    movq            [r2], xm5
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm5
-    movhps          [r2 + r4], xm2
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm1
-    movq            [r2 + r3], xm4
-    movhps          [r2 + r3 * 2], xm1
-    movhps          [r2 + r4], xm4
-%else
-    psubw           m5, m3                          ; m5 = word: row 0, row 1
-    psubw           m2, m3                          ; m2 = word: row 2, row 3
-    psubw           m1, m3                          ; m1 = word: row 4, row 5
-    psubw           m4, m3                          ; m4 = word: row 6, row 7
-    vextracti128    xm6, m5, 1
-    vextracti128    xm3, m2, 1
-    vextracti128    xm0, m1, 1
-    movu            [r2], xm5
-    movu            [r2 + r3], xm6
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r4], xm3
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm1
-    movu            [r2 + r3], xm0
-    movu            [r2 + r3 * 2], xm4
-    vextracti128    xm4, m4, 1
-    movu            [r2 + r4], xm4
-%endif
-    RET
-%endmacro
-
-%macro FILTER_VER_LUMA_AVX2_8x4 1
-INIT_YMM avx2
-cglobal interp_8tap_vert_%1_8x4, 4, 6, 7
-    mov             r4d, r4m
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
+    vbroadcasti128  m14, [pw_2000]
 %endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-    PROCESS_LUMA_AVX2_W8_4R
-%ifidn %1,pp
-    mova            m3, [pw_512]
-%else
-    add             r3d, r3d
-    vbroadcasti128  m3, [pw_2000]
-%endif
-    lea             r4, [r3 * 3]
-%ifidn %1,pp
-    pmulhrsw        m5, m3                          ; m5 = word: row 0, row 1
-    pmulhrsw        m2, m3                          ; m2 = word: row 2, row 3
-    packuswb        m5, m2
-    vextracti128    xm2, m5, 1
-    movq            [r2], xm5
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm5
-    movhps          [r2 + r4], xm2
-%else
-    psubw           m5, m3                          ; m5 = word: row 0, row 1
-    psubw           m2, m3                          ; m2 = word: row 2, row 3
-    movu            [r2], xm5
-    vextracti128    xm5, m5, 1
-    movu            [r2 + r3], xm5
-    movu            [r2 + r3 * 2], xm2
-    vextracti128    xm2, m2, 1
-    movu            [r2 + r4], xm2
-%endif
-    RET
-%endmacro
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_pp_8x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_8xN 8, 4, pp
-    FILTER_VER_LUMA_AVX2_8x4 pp
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_pp_8x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_8xN 8, 8, pp
-    FILTER_VER_LUMA_AVX2_8x8 pp
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_pp_8x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_8xN 8, 16, pp
-    FILTER_VER_LUMA_AVX2_8xN 8, 16, pp
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_pp_8x32(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_8xN 8, 32, pp
-    FILTER_VER_LUMA_AVX2_8xN 8, 32, pp
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_ps_8x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_8xN 8, 4, ps
-    FILTER_VER_LUMA_AVX2_8x4 ps
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_ps_8x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_8xN 8, 8, ps
-    FILTER_VER_LUMA_AVX2_8x8 ps
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_ps_8x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_8xN 8, 16, ps
-    FILTER_VER_LUMA_AVX2_8xN 8, 16, ps
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_ps_8x32(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_8xN 8, 32, ps
-    FILTER_VER_LUMA_AVX2_8xN 8, 32, ps
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_%3_12x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_LUMA_12xN 3
-INIT_XMM sse4
-cglobal interp_8tap_vert_%3_%1x%2, 5, 7, 8
-    lea       r5, [3 * r1]
-    sub       r0, r5
-    shl       r4d, 6
-%ifidn %3,ps
-    add       r3d, r3d
-%endif
-
-%ifdef PIC
-    lea       r5, [tab_LumaCoeffVer]
-    lea       r6, [r5 + r4]
-%else
-    lea       r6, [tab_LumaCoeffVer + r4]
-%endif
-
- %ifidn %3,pp
-    mova      m3, [pw_512]
-%else
-    mova      m3, [pw_2000]
-%endif
-
-    mov       r4d, %2/4
-
-.loopH:
-    PROCESS_LUMA_W8_4R
-
-%ifidn %3,pp
-    pmulhrsw  m7, m3
-    pmulhrsw  m6, m3
-    pmulhrsw  m5, m3
-    pmulhrsw  m4, m3
-
-    packuswb  m7, m6
-    packuswb  m5, m4
-
-    movlps    [r2], m7
-    movhps    [r2 + r3], m7
-    lea       r5, [r2 + 2 * r3]
-    movlps    [r5], m5
-    movhps    [r5 + r3], m5
-%else
-    psubw     m7, m3
-    psubw     m6, m3
-    psubw     m5, m3
-    psubw     m4, m3
-
-    movu      [r2], m7
-    movu      [r2 + r3], m6
-    lea       r5, [r2 + 2 * r3]
-    movu      [r5], m5
-    movu      [r5 + r3], m4
-%endif
-
-    lea       r5, [8 * r1 - 8]
-    sub       r0, r5
-%ifidn %3,pp
-    add       r2, 8
-%else
-    add       r2, 16
-%endif
-
-    PROCESS_LUMA_W4_4R
-
-%ifidn %3,pp
-    pmulhrsw  m4, m3
-    pmulhrsw  m5, m3
-
-    packuswb  m4, m5
-
-    movd      [r2], m4
-    pextrd    [r2 + r3], m4, 1
-    lea       r5, [r2 + 2 * r3]
-    pextrd    [r5], m4, 2
-    pextrd    [r5 + r3], m4, 3
-%else
-    psubw     m4, m3
-    psubw     m5, m3
-
-    movlps    [r2], m4
-    movhps    [r2 + r3], m4
-    lea       r5, [r2 + 2 * r3]
-    movlps    [r5], m5
-    movhps    [r5 + r3], m5
-%endif
-
-    lea       r5, [4 * r1 + 8]
-    sub       r0, r5
-%ifidn %3,pp
-    lea       r2, [r2 + 4 * r3 - 8]
-%else
-    lea       r2, [r2 + 4 * r3 - 16]
-%endif
-
-    dec       r4d
-    jnz       .loopH
-
-    RET
-%endmacro
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_pp_12x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_12xN 12, 16, pp
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_ps_12x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_12xN 12, 16, ps
-
-%macro FILTER_VER_LUMA_AVX2_12x16 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_12x16, 4, 7, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-%ifidn %1,pp
-    mova            m14, [pw_512]
-%else
-    add             r3d, r3d
-    vbroadcasti128  m14, [pw_2000]
-%endif
-    lea             r6, [r3 * 3]
-
     movu            xm0, [r0]                       ; m0 = row 0
     movu            xm1, [r0 + r1]                  ; m1 = row 1
     punpckhbw       xm2, xm0, xm1
@@ -14888,7 +13111,6 @@ cglobal interp_8tap_vert_%1_12x16, 4, 7, 15
     paddw           m4, m10
     pmaddubsw       m10, m8, [r5 + 1 * mmsize]
     paddw           m6, m10
-    pmaddubsw       m8, [r5]
     movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
     punpckhbw       xm11, xm9, xm10
     punpcklbw       xm9, xm10
@@ -14899,7 +13121,6 @@ cglobal interp_8tap_vert_%1_12x16, 4, 7, 15
     paddw           m5, m11
     pmaddubsw       m11, m9, [r5 + 1 * mmsize]
     paddw           m7, m11
-    pmaddubsw       m9, [r5]
     movu            xm11, [r0 + r4]                 ; m11 = row 11
     punpckhbw       xm12, xm10, xm11
     punpcklbw       xm10, xm11
@@ -14908,9 +13129,6 @@ cglobal interp_8tap_vert_%1_12x16, 4, 7, 15
     paddw           m4, m12
     pmaddubsw       m12, m10, [r5 + 2 * mmsize]
     paddw           m6, m12
-    pmaddubsw       m12, m10, [r5 + 1 * mmsize]
-    paddw           m8, m12
-    pmaddubsw       m10, [r5]
     lea             r0, [r0 + r1 * 4]
     movu            xm12, [r0]                      ; m12 = row 12
     punpckhbw       xm13, xm11, xm12
@@ -14920,10 +13138,7 @@ cglobal interp_8tap_vert_%1_12x16, 4, 7, 15
     paddw           m5, m13
     pmaddubsw       m13, m11, [r5 + 2 * mmsize]
     paddw           m7, m13
-    pmaddubsw       m13, m11, [r5 + 1 * mmsize]
-    paddw           m9, m13
-    pmaddubsw       m11, [r5]
-
+    lea             r4, [r3 * 3]
 %ifidn %1,pp
     pmulhrsw        m0, m14                         ; m0 = word: row 0
     pmulhrsw        m1, m14                         ; m1 = word: row 1
@@ -14940,19 +13155,13 @@ cglobal interp_8tap_vert_%1_12x16, 4, 7, 15
     vextracti128    xm1, m0, 1
     vextracti128    xm3, m2, 1
     vextracti128    xm5, m4, 1
-    movq            [r2], xm0
-    pextrd          [r2 + 8], xm0, 2
-    movq            [r2 + r3], xm1
-    pextrd          [r2 + r3 + 8], xm1, 2
-    movq            [r2 + r3 * 2], xm2
-    pextrd          [r2 + r3 * 2 + 8], xm2, 2
-    movq            [r2 + r6], xm3
-    pextrd          [r2 + r6 + 8], xm3, 2
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r4], xm3
     lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm4
-    pextrd          [r2 + 8], xm4, 2
-    movq            [r2 + r3], xm5
-    pextrd          [r2 + r3 + 8], xm5, 2
+    movu            [r2], xm4
+    movu            [r2 + r3], xm5
 %else
     psubw           m0, m14                         ; m0 = word: row 0
     psubw           m1, m14                         ; m1 = word: row 1
@@ -14960,249 +13169,67 @@ cglobal interp_8tap_vert_%1_12x16, 4, 7, 15
     psubw           m3, m14                         ; m3 = word: row 3
     psubw           m4, m14                         ; m4 = word: row 4
     psubw           m5, m14                         ; m5 = word: row 5
-    movu            [r2], xm0
-    vextracti128    xm0, m0, 1
-    movq            [r2 + 16], xm0
-    movu            [r2 + r3], xm1
-    vextracti128    xm1, m1, 1
-    movq            [r2 + r3 + 16], xm1
-    movu            [r2 + r3 * 2], xm2
-    vextracti128    xm2, m2, 1
-    movq            [r2 + r3 * 2 + 16], xm2
-    movu            [r2 + r6], xm3
-    vextracti128    xm3, m3, 1
-    movq            [r2 + r6 + 16], xm3
+    movu            [r2], m0
+    movu            [r2 + r3], m1
+    movu            [r2 + r3 * 2], m2
+    movu            [r2 + r4], m3
     lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm4
-    vextracti128    xm4, m4, 1
-    movq            [r2 + 16], xm4
-    movu            [r2 + r3], xm5
-    vextracti128    xm5, m5, 1
-    movq            [r2 + r3 + 16], xm5
+    movu            [r2], m4
+    movu            [r2 + r3], m5
 %endif
-
     movu            xm13, [r0 + r1]                 ; m13 = row 13
     punpckhbw       xm0, xm12, xm13
     punpcklbw       xm12, xm13
     vinserti128     m12, m12, xm0, 1
     pmaddubsw       m0, m12, [r5 + 3 * mmsize]
     paddw           m6, m0
-    pmaddubsw       m0, m12, [r5 + 2 * mmsize]
-    paddw           m8, m0
-    pmaddubsw       m0, m12, [r5 + 1 * mmsize]
-    paddw           m10, m0
-    pmaddubsw       m12, [r5]
     movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
     punpckhbw       xm1, xm13, xm0
     punpcklbw       xm13, xm0
     vinserti128     m13, m13, xm1, 1
     pmaddubsw       m1, m13, [r5 + 3 * mmsize]
     paddw           m7, m1
-    pmaddubsw       m1, m13, [r5 + 2 * mmsize]
-    paddw           m9, m1
-    pmaddubsw       m1, m13, [r5 + 1 * mmsize]
-    paddw           m11, m1
-    pmaddubsw       m13, [r5]
-
 %ifidn %1,pp
     pmulhrsw        m6, m14                         ; m6 = word: row 6
     pmulhrsw        m7, m14                         ; m7 = word: row 7
     packuswb        m6, m7
     vpermq          m6, m6, 11011000b
     vextracti128    xm7, m6, 1
-    movq            [r2 + r3 * 2], xm6
-    pextrd          [r2 + r3 * 2 + 8], xm6, 2
-    movq            [r2 + r6], xm7
-    pextrd          [r2 + r6 + 8], xm7, 2
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r4], xm7
 %else
     psubw           m6, m14                         ; m6 = word: row 6
     psubw           m7, m14                         ; m7 = word: row 7
-    movu            [r2 + r3 * 2], xm6
-    vextracti128    xm6, m6, 1
-    movq            [r2 + r3 * 2 + 16], xm6
-    movu            [r2 + r6], xm7
-    vextracti128    xm7, m7, 1
-    movq            [r2 + r6 + 16], xm7
-%endif
-    lea             r2, [r2 + r3 * 4]
-
-    movu            xm1, [r0 + r4]                  ; m1 = row 15
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m2, m0, [r5 + 3 * mmsize]
-    paddw           m8, m2
-    pmaddubsw       m2, m0, [r5 + 2 * mmsize]
-    paddw           m10, m2
-    pmaddubsw       m2, m0, [r5 + 1 * mmsize]
-    paddw           m12, m2
-    pmaddubsw       m0, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm2, [r0]                       ; m2 = row 16
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m3, m1, [r5 + 3 * mmsize]
-    paddw           m9, m3
-    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
-    paddw           m11, m3
-    pmaddubsw       m3, m1, [r5 + 1 * mmsize]
-    paddw           m13, m3
-    pmaddubsw       m1, [r5]
-    movu            xm3, [r0 + r1]                  ; m3 = row 17
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, [r5 + 3 * mmsize]
-    paddw           m10, m4
-    pmaddubsw       m4, m2, [r5 + 2 * mmsize]
-    paddw           m12, m4
-    pmaddubsw       m2, [r5 + 1 * mmsize]
-    paddw           m0, m2
-    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
-    punpckhbw       xm5, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, [r5 + 3 * mmsize]
-    paddw           m11, m5
-    pmaddubsw       m5, m3, [r5 + 2 * mmsize]
-    paddw           m13, m5
-    pmaddubsw       m3, [r5 + 1 * mmsize]
-    paddw           m1, m3
-    movu            xm5, [r0 + r4]                  ; m5 = row 19
-    punpckhbw       xm6, xm4, xm5
-    punpcklbw       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, [r5 + 3 * mmsize]
-    paddw           m12, m6
-    pmaddubsw       m4, [r5 + 2 * mmsize]
-    paddw           m0, m4
-    lea             r0, [r0 + r1 * 4]
-    movu            xm6, [r0]                       ; m6 = row 20
-    punpckhbw       xm7, xm5, xm6
-    punpcklbw       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m7, m5, [r5 + 3 * mmsize]
-    paddw           m13, m7
-    pmaddubsw       m5, [r5 + 2 * mmsize]
-    paddw           m1, m5
-    movu            xm7, [r0 + r1]                  ; m7 = row 21
-    punpckhbw       xm2, xm6, xm7
-    punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm2, 1
-    pmaddubsw       m6, [r5 + 3 * mmsize]
-    paddw           m0, m6
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 22
-    punpckhbw       xm3, xm7, xm2
-    punpcklbw       xm7, xm2
-    vinserti128     m7, m7, xm3, 1
-    pmaddubsw       m7, [r5 + 3 * mmsize]
-    paddw           m1, m7
-
-%ifidn %1,pp
-    pmulhrsw        m8, m14                         ; m8 = word: row 8
-    pmulhrsw        m9, m14                         ; m9 = word: row 9
-    pmulhrsw        m10, m14                        ; m10 = word: row 10
-    pmulhrsw        m11, m14                        ; m11 = word: row 11
-    pmulhrsw        m12, m14                        ; m12 = word: row 12
-    pmulhrsw        m13, m14                        ; m13 = word: row 13
-    pmulhrsw        m0, m14                         ; m0 = word: row 14
-    pmulhrsw        m1, m14                         ; m1 = word: row 15
-    packuswb        m8, m9
-    packuswb        m10, m11
-    packuswb        m12, m13
-    packuswb        m0, m1
-    vpermq          m8, m8, 11011000b
-    vpermq          m10, m10, 11011000b
-    vpermq          m12, m12, 11011000b
-    vpermq          m0, m0, 11011000b
-    vextracti128    xm9, m8, 1
-    vextracti128    xm11, m10, 1
-    vextracti128    xm13, m12, 1
-    vextracti128    xm1, m0, 1
-    movq            [r2], xm8
-    pextrd          [r2 + 8], xm8, 2
-    movq            [r2 + r3], xm9
-    pextrd          [r2 + r3 + 8], xm9, 2
-    movq            [r2 + r3 * 2], xm10
-    pextrd          [r2 + r3 * 2 + 8], xm10, 2
-    movq            [r2 + r6], xm11
-    pextrd          [r2 + r6 + 8], xm11, 2
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm12
-    pextrd          [r2 + 8], xm12, 2
-    movq            [r2 + r3], xm13
-    pextrd          [r2 + r3 + 8], xm13, 2
-    movq            [r2 + r3 * 2], xm0
-    pextrd          [r2 + r3 * 2 + 8], xm0, 2
-    movq            [r2 + r6], xm1
-    pextrd          [r2 + r6 + 8], xm1, 2
-%else
-    psubw           m8, m14                         ; m8 = word: row 8
-    psubw           m9, m14                         ; m9 = word: row 9
-    psubw           m10, m14                        ; m10 = word: row 10
-    psubw           m11, m14                        ; m11 = word: row 11
-    psubw           m12, m14                        ; m12 = word: row 12
-    psubw           m13, m14                        ; m13 = word: row 13
-    psubw           m0, m14                         ; m0 = word: row 14
-    psubw           m1, m14                         ; m1 = word: row 15
-    movu            [r2], xm8
-    vextracti128    xm8, m8, 1
-    movq            [r2 + 16], xm8
-    movu            [r2 + r3], xm9
-    vextracti128    xm9, m9, 1
-    movq            [r2 + r3 + 16], xm9
-    movu            [r2 + r3 * 2], xm10
-    vextracti128    xm10, m10, 1
-    movq            [r2 + r3 * 2 + 16], xm10
-    movu            [r2 + r6], xm11
-    vextracti128    xm11, m11, 1
-    movq            [r2 + r6 + 16], xm11
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm12
-    vextracti128    xm12, m12, 1
-    movq            [r2 + 16], xm12
-    movu            [r2 + r3], xm13
-    vextracti128    xm13, m13, 1
-    movq            [r2 + r3 + 16], xm13
-    movu            [r2 + r3 * 2], xm0
-    vextracti128    xm0, m0, 1
-    movq            [r2 + r3 * 2 + 16], xm0
-    movu            [r2 + r6], xm1
-    vextracti128    xm1, m1, 1
-    movq            [r2 + r6 + 16], xm1
+    movu            [r2 + r3 * 2], m6
+    movu            [r2 + r4], m7
 %endif
     RET
 %endif
 %endmacro
 
-    FILTER_VER_LUMA_AVX2_12x16 pp
-    FILTER_VER_LUMA_AVX2_12x16 ps
+    FILTER_VER_LUMA_AVX2_16x8 pp
+    FILTER_VER_LUMA_AVX2_16x8 ps
 
-%macro FILTER_VER_LUMA_AVX2_16x16 1
+%macro FILTER_VER_LUMA_AVX2_16x4 1
 INIT_YMM avx2
 %if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_16x16, 4, 7, 15
+cglobal interp_8tap_vert_%1_16x4, 4, 6, 13
     mov             r4d, r4m
     shl             r4d, 7
-
 %ifdef PIC
     lea             r5, [tab_LumaCoeffVer_32]
     add             r5, r4
 %else
     lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
-
     lea             r4, [r1 * 3]
     sub             r0, r4
 %ifidn %1,pp
-    mova            m14, [pw_512]
+    mova            m12, [pw_512]
 %else
     add             r3d, r3d
-    vbroadcasti128  m14, [pw_2000]
+    vbroadcasti128  m12, [pw_2000]
 %endif
-    lea             r6, [r3 * 3]
-
     movu            xm0, [r0]                       ; m0 = row 0
     movu            xm1, [r0 + r1]                  ; m1 = row 1
     punpckhbw       xm2, xm0, xm1
@@ -15237,7 +13264,6 @@ cglobal interp_8tap_vert_%1_16x16, 4, 7, 15
     paddw           m0, m6
     pmaddubsw       m6, m4, [r5 + 1 * mmsize]
     paddw           m2, m6
-    pmaddubsw       m4, [r5]
     movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
     punpckhbw       xm7, xm5, xm6
     punpcklbw       xm5, xm6
@@ -15246,7 +13272,6 @@ cglobal interp_8tap_vert_%1_16x16, 4, 7, 15
     paddw           m1, m7
     pmaddubsw       m7, m5, [r5 + 1 * mmsize]
     paddw           m3, m7
-    pmaddubsw       m5, [r5]
     movu            xm7, [r0 + r4]                  ; m7 = row 7
     punpckhbw       xm8, xm6, xm7
     punpcklbw       xm6, xm7
@@ -15255,9 +13280,6 @@ cglobal interp_8tap_vert_%1_16x16, 4, 7, 15
     paddw           m0, m8
     pmaddubsw       m8, m6, [r5 + 2 * mmsize]
     paddw           m2, m8
-    pmaddubsw       m8, m6, [r5 + 1 * mmsize]
-    paddw           m4, m8
-    pmaddubsw       m6, [r5]
     lea             r0, [r0 + r1 * 4]
     movu            xm8, [r0]                       ; m8 = row 8
     punpckhbw       xm9, xm7, xm8
@@ -15267,65 +13289,201 @@ cglobal interp_8tap_vert_%1_16x16, 4, 7, 15
     paddw           m1, m9
     pmaddubsw       m9, m7, [r5 + 2 * mmsize]
     paddw           m3, m9
-    pmaddubsw       m9, m7, [r5 + 1 * mmsize]
-    paddw           m5, m9
-    pmaddubsw       m7, [r5]
     movu            xm9, [r0 + r1]                  ; m9 = row 9
     punpckhbw       xm10, xm8, xm9
     punpcklbw       xm8, xm9
     vinserti128     m8, m8, xm10, 1
     pmaddubsw       m10, m8, [r5 + 3 * mmsize]
     paddw           m2, m10
-    pmaddubsw       m10, m8, [r5 + 2 * mmsize]
-    paddw           m4, m10
-    pmaddubsw       m10, m8, [r5 + 1 * mmsize]
-    paddw           m6, m10
-    pmaddubsw       m8, [r5]
     movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
     punpckhbw       xm11, xm9, xm10
     punpcklbw       xm9, xm10
     vinserti128     m9, m9, xm11, 1
     pmaddubsw       m11, m9, [r5 + 3 * mmsize]
     paddw           m3, m11
-    pmaddubsw       m11, m9, [r5 + 2 * mmsize]
-    paddw           m5, m11
-    pmaddubsw       m11, m9, [r5 + 1 * mmsize]
-    paddw           m7, m11
-    pmaddubsw       m9, [r5]
-    movu            xm11, [r0 + r4]                 ; m11 = row 11
-    punpckhbw       xm12, xm10, xm11
-    punpcklbw       xm10, xm11
-    vinserti128     m10, m10, xm12, 1
-    pmaddubsw       m12, m10, [r5 + 3 * mmsize]
-    paddw           m4, m12
-    pmaddubsw       m12, m10, [r5 + 2 * mmsize]
-    paddw           m6, m12
-    pmaddubsw       m12, m10, [r5 + 1 * mmsize]
-    paddw           m8, m12
-    pmaddubsw       m10, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm12, [r0]                      ; m12 = row 12
-    punpckhbw       xm13, xm11, xm12
-    punpcklbw       xm11, xm12
-    vinserti128     m11, m11, xm13, 1
-    pmaddubsw       m13, m11, [r5 + 3 * mmsize]
-    paddw           m5, m13
-    pmaddubsw       m13, m11, [r5 + 2 * mmsize]
-    paddw           m7, m13
-    pmaddubsw       m13, m11, [r5 + 1 * mmsize]
-    paddw           m9, m13
-    pmaddubsw       m11, [r5]
-
 %ifidn %1,pp
-    pmulhrsw        m0, m14                         ; m0 = word: row 0
-    pmulhrsw        m1, m14                         ; m1 = word: row 1
-    pmulhrsw        m2, m14                         ; m2 = word: row 2
-    pmulhrsw        m3, m14                         ; m3 = word: row 3
-    pmulhrsw        m4, m14                         ; m4 = word: row 4
-    pmulhrsw        m5, m14                         ; m5 = word: row 5
+    pmulhrsw        m0, m12                         ; m0 = word: row 0
+    pmulhrsw        m1, m12                         ; m1 = word: row 1
+    pmulhrsw        m2, m12                         ; m2 = word: row 2
+    pmulhrsw        m3, m12                         ; m3 = word: row 3
     packuswb        m0, m1
     packuswb        m2, m3
-    packuswb        m4, m5
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m2, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    lea             r4, [r3 * 3]
+    movu            [r2 + r4], xm3
+%else
+    psubw           m0, m12                         ; m0 = word: row 0
+    psubw           m1, m12                         ; m1 = word: row 1
+    psubw           m2, m12                         ; m2 = word: row 2
+    psubw           m3, m12                         ; m3 = word: row 3
+    movu            [r2], m0
+    movu            [r2 + r3], m1
+    movu            [r2 + r3 * 2], m2
+    lea             r4, [r3 * 3]
+    movu            [r2 + r4], m3
+%endif
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_LUMA_AVX2_16x4 pp
+    FILTER_VER_LUMA_AVX2_16x4 ps
+%macro FILTER_VER_LUMA_AVX2_16xN 3
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%3_%1x%2, 4, 9, 15
+    mov             r4d, r4m
+    shl             r4d, 7
+
+%ifdef PIC
+    lea             r5, [tab_LumaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+%ifidn %3,ps
+    add             r3d, r3d
+    vbroadcasti128  m14, [pw_2000]
+%else
+    mova            m14, [pw_512]
+%endif
+    lea             r6, [r3 * 3]
+    lea             r7, [r1 * 4]
+    mov             r8d, %2 / 16
+
+.loop:
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddubsw       m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
+    paddw           m0, m4
+    pmaddubsw       m2, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
+    paddw           m1, m5
+    pmaddubsw       m3, [r5]
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhbw       xm6, xm4, xm5
+    punpcklbw       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
+    paddw           m0, m6
+    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
+    paddw           m2, m6
+    pmaddubsw       m4, [r5]
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhbw       xm7, xm5, xm6
+    punpcklbw       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
+    paddw           m1, m7
+    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
+    paddw           m3, m7
+    pmaddubsw       m5, [r5]
+    movu            xm7, [r0 + r4]                  ; m7 = row 7
+    punpckhbw       xm8, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
+    paddw           m0, m8
+    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
+    paddw           m2, m8
+    pmaddubsw       m8, m6, [r5 + 1 * mmsize]
+    paddw           m4, m8
+    pmaddubsw       m6, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm8, [r0]                       ; m8 = row 8
+    punpckhbw       xm9, xm7, xm8
+    punpcklbw       xm7, xm8
+    vinserti128     m7, m7, xm9, 1
+    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
+    paddw           m1, m9
+    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
+    paddw           m3, m9
+    pmaddubsw       m9, m7, [r5 + 1 * mmsize]
+    paddw           m5, m9
+    pmaddubsw       m7, [r5]
+    movu            xm9, [r0 + r1]                  ; m9 = row 9
+    punpckhbw       xm10, xm8, xm9
+    punpcklbw       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
+    paddw           m2, m10
+    pmaddubsw       m10, m8, [r5 + 2 * mmsize]
+    paddw           m4, m10
+    pmaddubsw       m10, m8, [r5 + 1 * mmsize]
+    paddw           m6, m10
+    pmaddubsw       m8, [r5]
+    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
+    punpckhbw       xm11, xm9, xm10
+    punpcklbw       xm9, xm10
+    vinserti128     m9, m9, xm11, 1
+    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
+    paddw           m3, m11
+    pmaddubsw       m11, m9, [r5 + 2 * mmsize]
+    paddw           m5, m11
+    pmaddubsw       m11, m9, [r5 + 1 * mmsize]
+    paddw           m7, m11
+    pmaddubsw       m9, [r5]
+    movu            xm11, [r0 + r4]                 ; m11 = row 11
+    punpckhbw       xm12, xm10, xm11
+    punpcklbw       xm10, xm11
+    vinserti128     m10, m10, xm12, 1
+    pmaddubsw       m12, m10, [r5 + 3 * mmsize]
+    paddw           m4, m12
+    pmaddubsw       m12, m10, [r5 + 2 * mmsize]
+    paddw           m6, m12
+    pmaddubsw       m12, m10, [r5 + 1 * mmsize]
+    paddw           m8, m12
+    pmaddubsw       m10, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm12, [r0]                      ; m12 = row 12
+    punpckhbw       xm13, xm11, xm12
+    punpcklbw       xm11, xm12
+    vinserti128     m11, m11, xm13, 1
+    pmaddubsw       m13, m11, [r5 + 3 * mmsize]
+    paddw           m5, m13
+    pmaddubsw       m13, m11, [r5 + 2 * mmsize]
+    paddw           m7, m13
+    pmaddubsw       m13, m11, [r5 + 1 * mmsize]
+    paddw           m9, m13
+    pmaddubsw       m11, [r5]
+
+%ifidn %3,pp
+    pmulhrsw        m0, m14                         ; m0 = word: row 0
+    pmulhrsw        m1, m14                         ; m1 = word: row 1
+    pmulhrsw        m2, m14                         ; m2 = word: row 2
+    pmulhrsw        m3, m14                         ; m3 = word: row 3
+    pmulhrsw        m4, m14                         ; m4 = word: row 4
+    pmulhrsw        m5, m14                         ; m5 = word: row 5
+    packuswb        m0, m1
+    packuswb        m2, m3
+    packuswb        m4, m5
     vpermq          m0, m0, 11011000b
     vpermq          m2, m2, 11011000b
     vpermq          m4, m4, 11011000b
@@ -15378,7 +13536,7 @@ cglobal interp_8tap_vert_%1_16x16, 4, 7, 15
     paddw           m11, m1
     pmaddubsw       m13, [r5]
 
-%ifidn %1,pp
+%ifidn %3,pp
     pmulhrsw        m6, m14                         ; m6 = word: row 6
     pmulhrsw        m7, m14                         ; m7 = word: row 7
     packuswb        m6, m7
@@ -15392,6 +13550,7 @@ cglobal interp_8tap_vert_%1_16x16, 4, 7, 15
     movu            [r2 + r3 * 2], m6
     movu            [r2 + r6], m7
 %endif
+
     lea             r2, [r2 + r3 * 4]
 
     movu            xm1, [r0 + r4]                  ; m1 = row 15
@@ -15467,7 +13626,7 @@ cglobal interp_8tap_vert_%1_16x16, 4, 7, 15
     pmaddubsw       m7, [r5 + 3 * mmsize]
     paddw           m1, m7
 
-%ifidn %1,pp
+%ifidn %3,pp
     pmulhrsw        m8, m14                         ; m8 = word: row 8
     pmulhrsw        m9, m14                         ; m9 = word: row 9
     pmulhrsw        m10, m14                        ; m10 = word: row 10
@@ -15516,37 +13675,21 @@ cglobal interp_8tap_vert_%1_16x16, 4, 7, 15
     movu            [r2 + r3 * 2], m0
     movu            [r2 + r6], m1
 %endif
+
+    lea             r2, [r2 + r3 * 4]
+    sub             r0, r7
+    dec             r8d
+    jnz             .loop
     RET
 %endif
 %endmacro
 
-    FILTER_VER_LUMA_AVX2_16x16 pp
-    FILTER_VER_LUMA_AVX2_16x16 ps
-
-%macro FILTER_VER_LUMA_AVX2_16x12 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_16x12, 4, 7, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-%ifidn %1,pp
-    mova            m14, [pw_512]
-%else
-    add             r3d, r3d
-    vbroadcasti128  m14, [pw_2000]
-%endif
-    lea             r6, [r3 * 3]
+    FILTER_VER_LUMA_AVX2_16xN 16, 32, pp
+    FILTER_VER_LUMA_AVX2_16xN 16, 64, pp
+    FILTER_VER_LUMA_AVX2_16xN 16, 32, ps
+    FILTER_VER_LUMA_AVX2_16xN 16, 64, ps
 
+%macro PROCESS_LUMA_AVX2_W16_16R 1
     movu            xm0, [r0]                       ; m0 = row 0
     movu            xm1, [r0 + r1]                  ; m1 = row 1
     punpckhbw       xm2, xm0, xm1
@@ -15565,15 +13708,15 @@ cglobal interp_8tap_vert_%1_16x12, 4, 7, 15
     pmaddubsw       m4, m2, [r5 + 1 * mmsize]
     paddw           m0, m4
     pmaddubsw       m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
+    lea             r7, [r0 + r1 * 4]
+    movu            xm4, [r7]                       ; m4 = row 4
     punpckhbw       xm5, xm3, xm4
     punpcklbw       xm3, xm4
     vinserti128     m3, m3, xm5, 1
     pmaddubsw       m5, m3, [r5 + 1 * mmsize]
     paddw           m1, m5
     pmaddubsw       m3, [r5]
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    movu            xm5, [r7 + r1]                  ; m5 = row 5
     punpckhbw       xm6, xm4, xm5
     punpcklbw       xm4, xm5
     vinserti128     m4, m4, xm6, 1
@@ -15582,7 +13725,7 @@ cglobal interp_8tap_vert_%1_16x12, 4, 7, 15
     pmaddubsw       m6, m4, [r5 + 1 * mmsize]
     paddw           m2, m6
     pmaddubsw       m4, [r5]
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
     punpckhbw       xm7, xm5, xm6
     punpcklbw       xm5, xm6
     vinserti128     m5, m5, xm7, 1
@@ -15591,7 +13734,7 @@ cglobal interp_8tap_vert_%1_16x12, 4, 7, 15
     pmaddubsw       m7, m5, [r5 + 1 * mmsize]
     paddw           m3, m7
     pmaddubsw       m5, [r5]
-    movu            xm7, [r0 + r4]                  ; m7 = row 7
+    movu            xm7, [r7 + r4]                  ; m7 = row 7
     punpckhbw       xm8, xm6, xm7
     punpcklbw       xm6, xm7
     vinserti128     m6, m6, xm8, 1
@@ -15602,8 +13745,8 @@ cglobal interp_8tap_vert_%1_16x12, 4, 7, 15
     pmaddubsw       m8, m6, [r5 + 1 * mmsize]
     paddw           m4, m8
     pmaddubsw       m6, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm8, [r0]                       ; m8 = row 8
+    lea             r7, [r7 + r1 * 4]
+    movu            xm8, [r7]                       ; m8 = row 8
     punpckhbw       xm9, xm7, xm8
     punpcklbw       xm7, xm8
     vinserti128     m7, m7, xm9, 1
@@ -15614,7 +13757,7 @@ cglobal interp_8tap_vert_%1_16x12, 4, 7, 15
     pmaddubsw       m9, m7, [r5 + 1 * mmsize]
     paddw           m5, m9
     pmaddubsw       m7, [r5]
-    movu            xm9, [r0 + r1]                  ; m9 = row 9
+    movu            xm9, [r7 + r1]                  ; m9 = row 9
     punpckhbw       xm10, xm8, xm9
     punpcklbw       xm8, xm9
     vinserti128     m8, m8, xm10, 1
@@ -15625,7 +13768,7 @@ cglobal interp_8tap_vert_%1_16x12, 4, 7, 15
     pmaddubsw       m10, m8, [r5 + 1 * mmsize]
     paddw           m6, m10
     pmaddubsw       m8, [r5]
-    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
+    movu            xm10, [r7 + r1 * 2]             ; m10 = row 10
     punpckhbw       xm11, xm9, xm10
     punpcklbw       xm9, xm10
     vinserti128     m9, m9, xm11, 1
@@ -15636,7 +13779,7 @@ cglobal interp_8tap_vert_%1_16x12, 4, 7, 15
     pmaddubsw       m11, m9, [r5 + 1 * mmsize]
     paddw           m7, m11
     pmaddubsw       m9, [r5]
-    movu            xm11, [r0 + r4]                 ; m11 = row 11
+    movu            xm11, [r7 + r4]                 ; m11 = row 11
     punpckhbw       xm12, xm10, xm11
     punpcklbw       xm10, xm11
     vinserti128     m10, m10, xm12, 1
@@ -15647,8 +13790,8 @@ cglobal interp_8tap_vert_%1_16x12, 4, 7, 15
     pmaddubsw       m12, m10, [r5 + 1 * mmsize]
     paddw           m8, m12
     pmaddubsw       m10, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm12, [r0]                      ; m12 = row 12
+    lea             r7, [r7 + r1 * 4]
+    movu            xm12, [r7]                      ; m12 = row 12
     punpckhbw       xm13, xm11, xm12
     punpcklbw       xm11, xm12
     vinserti128     m11, m11, xm13, 1
@@ -15680,9 +13823,9 @@ cglobal interp_8tap_vert_%1_16x12, 4, 7, 15
     movu            [r2 + r3], xm1
     movu            [r2 + r3 * 2], xm2
     movu            [r2 + r6], xm3
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm4
-    movu            [r2 + r3], xm5
+    lea             r8, [r2 + r3 * 4]
+    movu            [r8], xm4
+    movu            [r8 + r3], xm5
 %else
     psubw           m0, m14                         ; m0 = word: row 0
     psubw           m1, m14                         ; m1 = word: row 1
@@ -15694,12 +13837,12 @@ cglobal interp_8tap_vert_%1_16x12, 4, 7, 15
     movu            [r2 + r3], m1
     movu            [r2 + r3 * 2], m2
     movu            [r2 + r6], m3
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], m4
-    movu            [r2 + r3], m5
+    lea             r8, [r2 + r3 * 4]
+    movu            [r8], m4
+    movu            [r8 + r3], m5
 %endif
 
-    movu            xm13, [r0 + r1]                 ; m13 = row 13
+    movu            xm13, [r7 + r1]                 ; m13 = row 13
     punpckhbw       xm0, xm12, xm13
     punpcklbw       xm12, xm13
     vinserti128     m12, m12, xm0, 1
@@ -15709,7 +13852,8 @@ cglobal interp_8tap_vert_%1_16x12, 4, 7, 15
     paddw           m8, m0
     pmaddubsw       m0, m12, [r5 + 1 * mmsize]
     paddw           m10, m0
-    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
+    pmaddubsw       m12, [r5]
+    movu            xm0, [r7 + r1 * 2]              ; m0 = row 14
     punpckhbw       xm1, xm13, xm0
     punpcklbw       xm13, xm0
     vinserti128     m13, m13, xm1, 1
@@ -15719,6 +13863,7 @@ cglobal interp_8tap_vert_%1_16x12, 4, 7, 15
     paddw           m9, m1
     pmaddubsw       m1, m13, [r5 + 1 * mmsize]
     paddw           m11, m1
+    pmaddubsw       m13, [r5]
 
 %ifidn %1,pp
     pmulhrsw        m6, m14                         ; m6 = word: row 6
@@ -15726,17 +13871,18 @@ cglobal interp_8tap_vert_%1_16x12, 4, 7, 15
     packuswb        m6, m7
     vpermq          m6, m6, 11011000b
     vextracti128    xm7, m6, 1
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r6], xm7
+    movu            [r8 + r3 * 2], xm6
+    movu            [r8 + r6], xm7
 %else
     psubw           m6, m14                         ; m6 = word: row 6
     psubw           m7, m14                         ; m7 = word: row 7
-    movu            [r2 + r3 * 2], m6
-    movu            [r2 + r6], m7
+    movu            [r8 + r3 * 2], m6
+    movu            [r8 + r6], m7
 %endif
-    lea             r2, [r2 + r3 * 4]
 
-    movu            xm1, [r0 + r4]                  ; m1 = row 15
+    lea             r8, [r8 + r3 * 4]
+
+    movu            xm1, [r7 + r4]                  ; m1 = row 15
     punpckhbw       xm2, xm0, xm1
     punpcklbw       xm0, xm1
     vinserti128     m0, m0, xm2, 1
@@ -15744,8 +13890,11 @@ cglobal interp_8tap_vert_%1_16x12, 4, 7, 15
     paddw           m8, m2
     pmaddubsw       m2, m0, [r5 + 2 * mmsize]
     paddw           m10, m2
-    lea             r0, [r0 + r1 * 4]
-    movu            xm2, [r0]                       ; m2 = row 16
+    pmaddubsw       m2, m0, [r5 + 1 * mmsize]
+    paddw           m12, m2
+    pmaddubsw       m0, [r5]
+    lea             r7, [r7 + r1 * 4]
+    movu            xm2, [r7]                       ; m2 = row 16
     punpckhbw       xm3, xm1, xm2
     punpcklbw       xm1, xm2
     vinserti128     m1, m1, xm3, 1
@@ -15753,71 +13902,111 @@ cglobal interp_8tap_vert_%1_16x12, 4, 7, 15
     paddw           m9, m3
     pmaddubsw       m3, m1, [r5 + 2 * mmsize]
     paddw           m11, m3
-    movu            xm3, [r0 + r1]                  ; m3 = row 17
+    pmaddubsw       m3, m1, [r5 + 1 * mmsize]
+    paddw           m13, m3
+    pmaddubsw       m1, [r5]
+    movu            xm3, [r7 + r1]                  ; m3 = row 17
     punpckhbw       xm4, xm2, xm3
     punpcklbw       xm2, xm3
     vinserti128     m2, m2, xm4, 1
     pmaddubsw       m4, m2, [r5 + 3 * mmsize]
     paddw           m10, m4
-    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
+    pmaddubsw       m4, m2, [r5 + 2 * mmsize]
+    paddw           m12, m4
+    pmaddubsw       m2, [r5 + 1 * mmsize]
+    paddw           m0, m2
+    movu            xm4, [r7 + r1 * 2]              ; m4 = row 18
     punpckhbw       xm5, xm3, xm4
     punpcklbw       xm3, xm4
     vinserti128     m3, m3, xm5, 1
     pmaddubsw       m5, m3, [r5 + 3 * mmsize]
     paddw           m11, m5
+    pmaddubsw       m5, m3, [r5 + 2 * mmsize]
+    paddw           m13, m5
+    pmaddubsw       m3, [r5 + 1 * mmsize]
+    paddw           m1, m3
+    movu            xm5, [r7 + r4]                  ; m5 = row 19
+    punpckhbw       xm6, xm4, xm5
+    punpcklbw       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddubsw       m6, m4, [r5 + 3 * mmsize]
+    paddw           m12, m6
+    pmaddubsw       m4, [r5 + 2 * mmsize]
+    paddw           m0, m4
+    lea             r7, [r7 + r1 * 4]
+    movu            xm6, [r7]                       ; m6 = row 20
+    punpckhbw       xm7, xm5, xm6
+    punpcklbw       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddubsw       m7, m5, [r5 + 3 * mmsize]
+    paddw           m13, m7
+    pmaddubsw       m5, [r5 + 2 * mmsize]
+    paddw           m1, m5
+    movu            xm7, [r7 + r1]                  ; m7 = row 21
+    punpckhbw       xm2, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm2, 1
+    pmaddubsw       m6, [r5 + 3 * mmsize]
+    paddw           m0, m6
+    movu            xm2, [r7 + r1 * 2]              ; m2 = row 22
+    punpckhbw       xm3, xm7, xm2
+    punpcklbw       xm7, xm2
+    vinserti128     m7, m7, xm3, 1
+    pmaddubsw       m7, [r5 + 3 * mmsize]
+    paddw           m1, m7
 
 %ifidn %1,pp
     pmulhrsw        m8, m14                         ; m8 = word: row 8
     pmulhrsw        m9, m14                         ; m9 = word: row 9
     pmulhrsw        m10, m14                        ; m10 = word: row 10
     pmulhrsw        m11, m14                        ; m11 = word: row 11
+    pmulhrsw        m12, m14                        ; m12 = word: row 12
+    pmulhrsw        m13, m14                        ; m13 = word: row 13
+    pmulhrsw        m0, m14                         ; m0 = word: row 14
+    pmulhrsw        m1, m14                         ; m1 = word: row 15
     packuswb        m8, m9
     packuswb        m10, m11
+    packuswb        m12, m13
+    packuswb        m0, m1
     vpermq          m8, m8, 11011000b
     vpermq          m10, m10, 11011000b
+    vpermq          m12, m12, 11011000b
+    vpermq          m0, m0, 11011000b
     vextracti128    xm9, m8, 1
     vextracti128    xm11, m10, 1
-    movu            [r2], xm8
-    movu            [r2 + r3], xm9
-    movu            [r2 + r3 * 2], xm10
-    movu            [r2 + r6], xm11
+    vextracti128    xm13, m12, 1
+    vextracti128    xm1, m0, 1
+    movu            [r8], xm8
+    movu            [r8 + r3], xm9
+    movu            [r8 + r3 * 2], xm10
+    movu            [r8 + r6], xm11
+    lea             r8, [r8 + r3 * 4]
+    movu            [r8], xm12
+    movu            [r8 + r3], xm13
+    movu            [r8 + r3 * 2], xm0
+    movu            [r8 + r6], xm1
 %else
     psubw           m8, m14                         ; m8 = word: row 8
     psubw           m9, m14                         ; m9 = word: row 9
     psubw           m10, m14                        ; m10 = word: row 10
     psubw           m11, m14                        ; m11 = word: row 11
-    movu            [r2], m8
-    movu            [r2 + r3], m9
-    movu            [r2 + r3 * 2], m10
-    movu            [r2 + r6], m11
-%endif
-    RET
+    psubw           m12, m14                        ; m12 = word: row 12
+    psubw           m13, m14                        ; m13 = word: row 13
+    psubw           m0, m14                         ; m0 = word: row 14
+    psubw           m1, m14                         ; m1 = word: row 15
+    movu            [r8], m8
+    movu            [r8 + r3], m9
+    movu            [r8 + r3 * 2], m10
+    movu            [r8 + r6], m11
+    lea             r8, [r8 + r3 * 4]
+    movu            [r8], m12
+    movu            [r8 + r3], m13
+    movu            [r8 + r3 * 2], m0
+    movu            [r8 + r6], m1
 %endif
 %endmacro
 
-    FILTER_VER_LUMA_AVX2_16x12 pp
-    FILTER_VER_LUMA_AVX2_16x12 ps
-
-%macro FILTER_VER_LUMA_AVX2_16x8 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_16x8, 4, 6, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-%ifidn %1,pp
-    mova            m14, [pw_512]
-%else
-    add             r3d, r3d
-    vbroadcasti128  m14, [pw_2000]
-%endif
+%macro PROCESS_LUMA_AVX2_W16_8R 1
     movu            xm0, [r0]                       ; m0 = row 0
     movu            xm1, [r0 + r1]                  ; m1 = row 1
     punpckhbw       xm2, xm0, xm1
@@ -15836,15 +14025,15 @@ cglobal interp_8tap_vert_%1_16x8, 4, 6, 15
     pmaddubsw       m4, m2, [r5 + 1 * mmsize]
     paddw           m0, m4
     pmaddubsw       m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
+    lea             r7, [r0 + r1 * 4]
+    movu            xm4, [r7]                       ; m4 = row 4
     punpckhbw       xm5, xm3, xm4
     punpcklbw       xm3, xm4
     vinserti128     m3, m3, xm5, 1
     pmaddubsw       m5, m3, [r5 + 1 * mmsize]
     paddw           m1, m5
     pmaddubsw       m3, [r5]
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    movu            xm5, [r7 + r1]                  ; m5 = row 5
     punpckhbw       xm6, xm4, xm5
     punpcklbw       xm4, xm5
     vinserti128     m4, m4, xm6, 1
@@ -15853,7 +14042,7 @@ cglobal interp_8tap_vert_%1_16x8, 4, 6, 15
     pmaddubsw       m6, m4, [r5 + 1 * mmsize]
     paddw           m2, m6
     pmaddubsw       m4, [r5]
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
     punpckhbw       xm7, xm5, xm6
     punpcklbw       xm5, xm6
     vinserti128     m5, m5, xm7, 1
@@ -15862,7 +14051,7 @@ cglobal interp_8tap_vert_%1_16x8, 4, 6, 15
     pmaddubsw       m7, m5, [r5 + 1 * mmsize]
     paddw           m3, m7
     pmaddubsw       m5, [r5]
-    movu            xm7, [r0 + r4]                  ; m7 = row 7
+    movu            xm7, [r7 + r4]                  ; m7 = row 7
     punpckhbw       xm8, xm6, xm7
     punpcklbw       xm6, xm7
     vinserti128     m6, m6, xm8, 1
@@ -15873,8 +14062,8 @@ cglobal interp_8tap_vert_%1_16x8, 4, 6, 15
     pmaddubsw       m8, m6, [r5 + 1 * mmsize]
     paddw           m4, m8
     pmaddubsw       m6, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm8, [r0]                       ; m8 = row 8
+    lea             r7, [r7 + r1 * 4]
+    movu            xm8, [r7]                       ; m8 = row 8
     punpckhbw       xm9, xm7, xm8
     punpcklbw       xm7, xm8
     vinserti128     m7, m7, xm9, 1
@@ -15885,7 +14074,7 @@ cglobal interp_8tap_vert_%1_16x8, 4, 6, 15
     pmaddubsw       m9, m7, [r5 + 1 * mmsize]
     paddw           m5, m9
     pmaddubsw       m7, [r5]
-    movu            xm9, [r0 + r1]                  ; m9 = row 9
+    movu            xm9, [r7 + r1]                  ; m9 = row 9
     punpckhbw       xm10, xm8, xm9
     punpcklbw       xm8, xm9
     vinserti128     m8, m8, xm10, 1
@@ -15895,7 +14084,7 @@ cglobal interp_8tap_vert_%1_16x8, 4, 6, 15
     paddw           m4, m10
     pmaddubsw       m10, m8, [r5 + 1 * mmsize]
     paddw           m6, m10
-    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
+    movu            xm10, [r7 + r1 * 2]             ; m10 = row 10
     punpckhbw       xm11, xm9, xm10
     punpcklbw       xm9, xm10
     vinserti128     m9, m9, xm11, 1
@@ -15905,7 +14094,7 @@ cglobal interp_8tap_vert_%1_16x8, 4, 6, 15
     paddw           m5, m11
     pmaddubsw       m11, m9, [r5 + 1 * mmsize]
     paddw           m7, m11
-    movu            xm11, [r0 + r4]                 ; m11 = row 11
+    movu            xm11, [r7 + r4]                 ; m11 = row 11
     punpckhbw       xm12, xm10, xm11
     punpcklbw       xm10, xm11
     vinserti128     m10, m10, xm12, 1
@@ -15913,8 +14102,8 @@ cglobal interp_8tap_vert_%1_16x8, 4, 6, 15
     paddw           m4, m12
     pmaddubsw       m12, m10, [r5 + 2 * mmsize]
     paddw           m6, m12
-    lea             r0, [r0 + r1 * 4]
-    movu            xm12, [r0]                      ; m12 = row 12
+    lea             r7, [r7 + r1 * 4]
+    movu            xm12, [r7]                      ; m12 = row 12
     punpckhbw       xm13, xm11, xm12
     punpcklbw       xm11, xm12
     vinserti128     m11, m11, xm13, 1
@@ -15922,7 +14111,7 @@ cglobal interp_8tap_vert_%1_16x8, 4, 6, 15
     paddw           m5, m13
     pmaddubsw       m13, m11, [r5 + 2 * mmsize]
     paddw           m7, m13
-    lea             r4, [r3 * 3]
+
 %ifidn %1,pp
     pmulhrsw        m0, m14                         ; m0 = word: row 0
     pmulhrsw        m1, m14                         ; m1 = word: row 1
@@ -15942,10 +14131,10 @@ cglobal interp_8tap_vert_%1_16x8, 4, 6, 15
     movu            [r2], xm0
     movu            [r2 + r3], xm1
     movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r4], xm3
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm4
-    movu            [r2 + r3], xm5
+    movu            [r2 + r6], xm3
+    lea             r8, [r2 + r3 * 4]
+    movu            [r8], xm4
+    movu            [r8 + r3], xm5
 %else
     psubw           m0, m14                         ; m0 = word: row 0
     psubw           m1, m14                         ; m1 = word: row 1
@@ -15956,48 +14145,45 @@ cglobal interp_8tap_vert_%1_16x8, 4, 6, 15
     movu            [r2], m0
     movu            [r2 + r3], m1
     movu            [r2 + r3 * 2], m2
-    movu            [r2 + r4], m3
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], m4
-    movu            [r2 + r3], m5
+    movu            [r2 + r6], m3
+    lea             r8, [r2 + r3 * 4]
+    movu            [r8], m4
+    movu            [r8 + r3], m5
 %endif
-    movu            xm13, [r0 + r1]                 ; m13 = row 13
+
+    movu            xm13, [r7 + r1]                 ; m13 = row 13
     punpckhbw       xm0, xm12, xm13
     punpcklbw       xm12, xm13
     vinserti128     m12, m12, xm0, 1
     pmaddubsw       m0, m12, [r5 + 3 * mmsize]
     paddw           m6, m0
-    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
+    movu            xm0, [r7 + r1 * 2]              ; m0 = row 14
     punpckhbw       xm1, xm13, xm0
     punpcklbw       xm13, xm0
     vinserti128     m13, m13, xm1, 1
     pmaddubsw       m1, m13, [r5 + 3 * mmsize]
     paddw           m7, m1
+
 %ifidn %1,pp
     pmulhrsw        m6, m14                         ; m6 = word: row 6
     pmulhrsw        m7, m14                         ; m7 = word: row 7
     packuswb        m6, m7
     vpermq          m6, m6, 11011000b
     vextracti128    xm7, m6, 1
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r4], xm7
+    movu            [r8 + r3 * 2], xm6
+    movu            [r8 + r6], xm7
 %else
     psubw           m6, m14                         ; m6 = word: row 6
     psubw           m7, m14                         ; m7 = word: row 7
-    movu            [r2 + r3 * 2], m6
-    movu            [r2 + r4], m7
-%endif
-    RET
+    movu            [r8 + r3 * 2], m6
+    movu            [r8 + r6], m7
 %endif
 %endmacro
 
-    FILTER_VER_LUMA_AVX2_16x8 pp
-    FILTER_VER_LUMA_AVX2_16x8 ps
-
-%macro FILTER_VER_LUMA_AVX2_16x4 1
+%macro FILTER_VER_LUMA_AVX2_24x32 1
 INIT_YMM avx2
 %if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_16x4, 4, 6, 13
+cglobal interp_8tap_vert_%1_24x32, 4, 11, 15
     mov             r4d, r4m
     shl             r4d, 7
 %ifdef PIC
@@ -16008,4684 +14194,1710 @@ cglobal interp_8tap_vert_%1_16x4, 4, 6, 13
 %endif
     lea             r4, [r1 * 3]
     sub             r0, r4
+%ifidn %1,ps
+    add             r3d, r3d
+    vbroadcasti128  m14, [pw_2000]
+%else
+    mova            m14, [pw_512]
+%endif
+    lea             r6, [r3 * 3]
+    lea             r10, [r1 * 4]
+    mov             r9d, 2
+.loopH:
+    PROCESS_LUMA_AVX2_W16_16R %1
 %ifidn %1,pp
-    mova            m12, [pw_512]
+    add             r2, 16
 %else
-    add             r3d, r3d
-    vbroadcasti128  m12, [pw_2000]
+    add             r2, 32
 %endif
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhbw       xm3, xm1, xm2
+    add             r0, 16
+    movq            xm1, [r0]                       ; m1 = row 0
+    movq            xm2, [r0 + r1]                  ; m2 = row 1
     punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhbw       xm4, xm2, xm3
+    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
     punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
-    paddw           m0, m4
+    vinserti128     m5, m1, xm2, 1
+    pmaddubsw       m5, [r5]
+    movq            xm4, [r0 + r4]                  ; m4 = row 3
+    punpcklbw       xm3, xm4
+    lea             r7, [r0 + r1 * 4]
+    movq            xm1, [r7]                       ; m1 = row 4
+    punpcklbw       xm4, xm1
+    vinserti128     m2, m3, xm4, 1
+    pmaddubsw       m0, m2, [r5 + 1 * mmsize]
+    paddw           m5, m0
     pmaddubsw       m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhbw       xm5, xm3, xm4
+    movq            xm3, [r7 + r1]                  ; m3 = row 5
+    punpcklbw       xm1, xm3
+    movq            xm4, [r7 + r1 * 2]              ; m4 = row 6
     punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
-    paddw           m1, m5
-    pmaddubsw       m3, [r5]
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhbw       xm6, xm4, xm5
-    punpcklbw       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
-    paddw           m0, m6
-    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
-    paddw           m2, m6
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhbw       xm7, xm5, xm6
-    punpcklbw       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
-    paddw           m1, m7
-    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
-    paddw           m3, m7
-    movu            xm7, [r0 + r4]                  ; m7 = row 7
-    punpckhbw       xm8, xm6, xm7
-    punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm8, 1
-    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
-    paddw           m0, m8
-    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
-    paddw           m2, m8
-    lea             r0, [r0 + r1 * 4]
-    movu            xm8, [r0]                       ; m8 = row 8
-    punpckhbw       xm9, xm7, xm8
-    punpcklbw       xm7, xm8
-    vinserti128     m7, m7, xm9, 1
-    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
-    paddw           m1, m9
-    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
-    paddw           m3, m9
-    movu            xm9, [r0 + r1]                  ; m9 = row 9
-    punpckhbw       xm10, xm8, xm9
-    punpcklbw       xm8, xm9
-    vinserti128     m8, m8, xm10, 1
-    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
-    paddw           m2, m10
-    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
-    punpckhbw       xm11, xm9, xm10
-    punpcklbw       xm9, xm10
-    vinserti128     m9, m9, xm11, 1
-    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
-    paddw           m3, m11
-%ifidn %1,pp
-    pmulhrsw        m0, m12                         ; m0 = word: row 0
-    pmulhrsw        m1, m12                         ; m1 = word: row 1
-    pmulhrsw        m2, m12                         ; m2 = word: row 2
-    pmulhrsw        m3, m12                         ; m3 = word: row 3
-    packuswb        m0, m1
-    packuswb        m2, m3
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    lea             r4, [r3 * 3]
-    movu            [r2 + r4], xm3
-%else
-    psubw           m0, m12                         ; m0 = word: row 0
-    psubw           m1, m12                         ; m1 = word: row 1
-    psubw           m2, m12                         ; m2 = word: row 2
-    psubw           m3, m12                         ; m3 = word: row 3
-    movu            [r2], m0
-    movu            [r2 + r3], m1
-    movu            [r2 + r3 * 2], m2
-    lea             r4, [r3 * 3]
-    movu            [r2 + r4], m3
-%endif
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_16x4 pp
-    FILTER_VER_LUMA_AVX2_16x4 ps
-%macro FILTER_VER_LUMA_AVX2_16xN 3
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%3_%1x%2, 4, 9, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-%ifidn %3,ps
-    add             r3d, r3d
-    vbroadcasti128  m14, [pw_2000]
-%else
-    mova            m14, [pw_512]
-%endif
-    lea             r6, [r3 * 3]
-    lea             r7, [r1 * 4]
-    mov             r8d, %2 / 16
-
-.loop:
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
     vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
+    paddw           m5, m3
+    pmaddubsw       m0, m1, [r5 + 1 * mmsize]
+    paddw           m2, m0
     pmaddubsw       m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
-    paddw           m0, m4
-    pmaddubsw       m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhbw       xm5, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
-    paddw           m1, m5
-    pmaddubsw       m3, [r5]
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhbw       xm6, xm4, xm5
-    punpcklbw       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
-    paddw           m0, m6
-    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
-    paddw           m2, m6
+    movq            xm3, [r7 + r4]                  ; m3 = row 7
+    punpcklbw       xm4, xm3
+    lea             r7, [r7 + r1 * 4]
+    movq            xm0, [r7]                       ; m0 = row 8
+    punpcklbw       xm3, xm0
+    vinserti128     m4, m4, xm3, 1
+    pmaddubsw       m3, m4, [r5 + 3 * mmsize]
+    paddw           m5, m3
+    pmaddubsw       m3, m4, [r5 + 2 * mmsize]
+    paddw           m2, m3
+    pmaddubsw       m3, m4, [r5 + 1 * mmsize]
+    paddw           m1, m3
     pmaddubsw       m4, [r5]
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhbw       xm7, xm5, xm6
-    punpcklbw       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
-    paddw           m1, m7
-    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
-    paddw           m3, m7
-    pmaddubsw       m5, [r5]
-    movu            xm7, [r0 + r4]                  ; m7 = row 7
-    punpckhbw       xm8, xm6, xm7
-    punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm8, 1
-    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
-    paddw           m0, m8
-    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
-    paddw           m2, m8
-    pmaddubsw       m8, m6, [r5 + 1 * mmsize]
-    paddw           m4, m8
+    movq            xm3, [r7 + r1]                  ; m3 = row 9
+    punpcklbw       xm0, xm3
+    movq            xm6, [r7 + r1 * 2]              ; m6 = row 10
+    punpcklbw       xm3, xm6
+    vinserti128     m0, m0, xm3, 1
+    pmaddubsw       m3, m0, [r5 + 3 * mmsize]
+    paddw           m2, m3
+    pmaddubsw       m3, m0, [r5 + 2 * mmsize]
+    paddw           m1, m3
+    pmaddubsw       m3, m0, [r5 + 1 * mmsize]
+    paddw           m4, m3
+    pmaddubsw       m0, [r5]
+
+    movq            xm3, [r7 + r4]                  ; m3 = row 11
+    punpcklbw       xm6, xm3
+    lea             r7, [r7 + r1 * 4]
+    movq            xm7, [r7]                       ; m7 = row 12
+    punpcklbw       xm3, xm7
+    vinserti128     m6, m6, xm3, 1
+    pmaddubsw       m3, m6, [r5 + 3 * mmsize]
+    paddw           m1, m3
+    pmaddubsw       m3, m6, [r5 + 2 * mmsize]
+    paddw           m4, m3
+    pmaddubsw       m3, m6, [r5 + 1 * mmsize]
+    paddw           m0, m3
     pmaddubsw       m6, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm8, [r0]                       ; m8 = row 8
-    punpckhbw       xm9, xm7, xm8
-    punpcklbw       xm7, xm8
-    vinserti128     m7, m7, xm9, 1
-    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
-    paddw           m1, m9
-    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
-    paddw           m3, m9
-    pmaddubsw       m9, m7, [r5 + 1 * mmsize]
-    paddw           m5, m9
+    movq            xm3, [r7 + r1]                  ; m3 = row 13
+    punpcklbw       xm7, xm3
+    movq            xm8, [r7 + r1 * 2]              ; m8 = row 14
+    punpcklbw       xm3, xm8
+    vinserti128     m7, m7, xm3, 1
+    pmaddubsw       m3, m7, [r5 + 3 * mmsize]
+    paddw           m4, m3
+    pmaddubsw       m3, m7, [r5 + 2 * mmsize]
+    paddw           m0, m3
+    pmaddubsw       m3, m7, [r5 + 1 * mmsize]
+    paddw           m6, m3
     pmaddubsw       m7, [r5]
-    movu            xm9, [r0 + r1]                  ; m9 = row 9
-    punpckhbw       xm10, xm8, xm9
-    punpcklbw       xm8, xm9
-    vinserti128     m8, m8, xm10, 1
-    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
-    paddw           m2, m10
-    pmaddubsw       m10, m8, [r5 + 2 * mmsize]
-    paddw           m4, m10
-    pmaddubsw       m10, m8, [r5 + 1 * mmsize]
-    paddw           m6, m10
+    movq            xm3, [r7 + r4]                  ; m3 = row 15
+    punpcklbw       xm8, xm3
+    lea             r7, [r7 + r1 * 4]
+    movq            xm9, [r7]                       ; m9 = row 16
+    punpcklbw       xm3, xm9
+    vinserti128     m8, m8, xm3, 1
+    pmaddubsw       m3, m8, [r5 + 3 * mmsize]
+    paddw           m0, m3
+    pmaddubsw       m3, m8, [r5 + 2 * mmsize]
+    paddw           m6, m3
+    pmaddubsw       m3, m8, [r5 + 1 * mmsize]
+    paddw           m7, m3
     pmaddubsw       m8, [r5]
-    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
-    punpckhbw       xm11, xm9, xm10
-    punpcklbw       xm9, xm10
-    vinserti128     m9, m9, xm11, 1
-    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
-    paddw           m3, m11
-    pmaddubsw       m11, m9, [r5 + 2 * mmsize]
-    paddw           m5, m11
-    pmaddubsw       m11, m9, [r5 + 1 * mmsize]
-    paddw           m7, m11
-    pmaddubsw       m9, [r5]
-    movu            xm11, [r0 + r4]                 ; m11 = row 11
-    punpckhbw       xm12, xm10, xm11
-    punpcklbw       xm10, xm11
-    vinserti128     m10, m10, xm12, 1
-    pmaddubsw       m12, m10, [r5 + 3 * mmsize]
-    paddw           m4, m12
-    pmaddubsw       m12, m10, [r5 + 2 * mmsize]
-    paddw           m6, m12
-    pmaddubsw       m12, m10, [r5 + 1 * mmsize]
-    paddw           m8, m12
-    pmaddubsw       m10, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm12, [r0]                      ; m12 = row 12
-    punpckhbw       xm13, xm11, xm12
-    punpcklbw       xm11, xm12
-    vinserti128     m11, m11, xm13, 1
-    pmaddubsw       m13, m11, [r5 + 3 * mmsize]
-    paddw           m5, m13
-    pmaddubsw       m13, m11, [r5 + 2 * mmsize]
-    paddw           m7, m13
-    pmaddubsw       m13, m11, [r5 + 1 * mmsize]
-    paddw           m9, m13
-    pmaddubsw       m11, [r5]
-
-%ifidn %3,pp
-    pmulhrsw        m0, m14                         ; m0 = word: row 0
-    pmulhrsw        m1, m14                         ; m1 = word: row 1
-    pmulhrsw        m2, m14                         ; m2 = word: row 2
-    pmulhrsw        m3, m14                         ; m3 = word: row 3
-    pmulhrsw        m4, m14                         ; m4 = word: row 4
-    pmulhrsw        m5, m14                         ; m5 = word: row 5
-    packuswb        m0, m1
-    packuswb        m2, m3
-    packuswb        m4, m5
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm1, m0, 1
+    movq            xm3, [r7 + r1]                  ; m3 = row 17
+    punpcklbw       xm9, xm3
+    movq            xm10, [r7 + r1 * 2]             ; m10 = row 18
+    punpcklbw       xm3, xm10
+    vinserti128     m9, m9, xm3, 1
+    pmaddubsw       m3, m9, [r5 + 3 * mmsize]
+    paddw           m6, m3
+    pmaddubsw       m3, m9, [r5 + 2 * mmsize]
+    paddw           m7, m3
+    pmaddubsw       m3, m9, [r5 + 1 * mmsize]
+    paddw           m8, m3
+    movq            xm3, [r7 + r4]                  ; m3 = row 19
+    punpcklbw       xm10, xm3
+    lea             r7, [r7 + r1 * 4]
+    movq            xm9, [r7]                       ; m9 = row 20
+    punpcklbw       xm3, xm9
+    vinserti128     m10, m10, xm3, 1
+    pmaddubsw       m3, m10, [r5 + 3 * mmsize]
+    paddw           m7, m3
+    pmaddubsw       m3, m10, [r5 + 2 * mmsize]
+    paddw           m8, m3
+    movq            xm3, [r7 + r1]                  ; m3 = row 21
+    punpcklbw       xm9, xm3
+    movq            xm10, [r7 + r1 * 2]             ; m10 = row 22
+    punpcklbw       xm3, xm10
+    vinserti128     m9, m9, xm3, 1
+    pmaddubsw       m3, m9, [r5 + 3 * mmsize]
+    paddw           m8, m3
+%ifidn %1,pp
+    pmulhrsw        m5, m14                         ; m5 = word: row 0, row 1
+    pmulhrsw        m2, m14                         ; m2 = word: row 2, row 3
+    pmulhrsw        m1, m14                         ; m1 = word: row 4, row 5
+    pmulhrsw        m4, m14                         ; m4 = word: row 6, row 7
+    pmulhrsw        m0, m14                         ; m0 = word: row 8, row 9
+    pmulhrsw        m6, m14                         ; m6 = word: row 10, row 11
+    pmulhrsw        m7, m14                         ; m7 = word: row 12, row 13
+    pmulhrsw        m8, m14                         ; m8 = word: row 14, row 15
+    packuswb        m5, m2
+    packuswb        m1, m4
+    packuswb        m0, m6
+    packuswb        m7, m8
+    vextracti128    xm2, m5, 1
+    vextracti128    xm4, m1, 1
+    vextracti128    xm6, m0, 1
+    vextracti128    xm8, m7, 1
+    movq            [r2], xm5
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm5
+    movhps          [r2 + r6], xm2
+    lea             r8, [r2 + r3 * 4]
+    movq            [r8], xm1
+    movq            [r8 + r3], xm4
+    movhps          [r8 + r3 * 2], xm1
+    movhps          [r8 + r6], xm4
+    lea             r8, [r8 + r3 * 4]
+    movq            [r8], xm0
+    movq            [r8 + r3], xm6
+    movhps          [r8 + r3 * 2], xm0
+    movhps          [r8 + r6], xm6
+    lea             r8, [r8 + r3 * 4]
+    movq            [r8], xm7
+    movq            [r8 + r3], xm8
+    movhps          [r8 + r3 * 2], xm7
+    movhps          [r8 + r6], xm8
+%else
+    psubw           m5, m14                         ; m5 = word: row 0, row 1
+    psubw           m2, m14                         ; m2 = word: row 2, row 3
+    psubw           m1, m14                         ; m1 = word: row 4, row 5
+    psubw           m4, m14                         ; m4 = word: row 6, row 7
+    psubw           m0, m14                         ; m0 = word: row 8, row 9
+    psubw           m6, m14                         ; m6 = word: row 10, row 11
+    psubw           m7, m14                         ; m7 = word: row 12, row 13
+    psubw           m8, m14                         ; m8 = word: row 14, row 15
+    vextracti128    xm3, m5, 1
+    movu            [r2], xm5
+    movu            [r2 + r3], xm3
     vextracti128    xm3, m2, 1
-    vextracti128    xm5, m4, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
     movu            [r2 + r3 * 2], xm2
     movu            [r2 + r6], xm3
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm4
-    movu            [r2 + r3], xm5
-%else
-    psubw           m0, m14                         ; m0 = word: row 0
-    psubw           m1, m14                         ; m1 = word: row 1
-    psubw           m2, m14                         ; m2 = word: row 2
-    psubw           m3, m14                         ; m3 = word: row 3
-    psubw           m4, m14                         ; m4 = word: row 4
-    psubw           m5, m14                         ; m5 = word: row 5
-    movu            [r2], m0
-    movu            [r2 + r3], m1
-    movu            [r2 + r3 * 2], m2
-    movu            [r2 + r6], m3
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], m4
-    movu            [r2 + r3], m5
-%endif
-
-    movu            xm13, [r0 + r1]                 ; m13 = row 13
-    punpckhbw       xm0, xm12, xm13
-    punpcklbw       xm12, xm13
-    vinserti128     m12, m12, xm0, 1
-    pmaddubsw       m0, m12, [r5 + 3 * mmsize]
-    paddw           m6, m0
-    pmaddubsw       m0, m12, [r5 + 2 * mmsize]
-    paddw           m8, m0
-    pmaddubsw       m0, m12, [r5 + 1 * mmsize]
-    paddw           m10, m0
-    pmaddubsw       m12, [r5]
-    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
-    punpckhbw       xm1, xm13, xm0
-    punpcklbw       xm13, xm0
-    vinserti128     m13, m13, xm1, 1
-    pmaddubsw       m1, m13, [r5 + 3 * mmsize]
-    paddw           m7, m1
-    pmaddubsw       m1, m13, [r5 + 2 * mmsize]
-    paddw           m9, m1
-    pmaddubsw       m1, m13, [r5 + 1 * mmsize]
-    paddw           m11, m1
-    pmaddubsw       m13, [r5]
-
-%ifidn %3,pp
-    pmulhrsw        m6, m14                         ; m6 = word: row 6
-    pmulhrsw        m7, m14                         ; m7 = word: row 7
-    packuswb        m6, m7
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm7, m6, 1
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r6], xm7
-%else
-    psubw           m6, m14                         ; m6 = word: row 6
-    psubw           m7, m14                         ; m7 = word: row 7
-    movu            [r2 + r3 * 2], m6
-    movu            [r2 + r6], m7
-%endif
-
-    lea             r2, [r2 + r3 * 4]
-
-    movu            xm1, [r0 + r4]                  ; m1 = row 15
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m2, m0, [r5 + 3 * mmsize]
-    paddw           m8, m2
-    pmaddubsw       m2, m0, [r5 + 2 * mmsize]
-    paddw           m10, m2
-    pmaddubsw       m2, m0, [r5 + 1 * mmsize]
-    paddw           m12, m2
-    pmaddubsw       m0, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm2, [r0]                       ; m2 = row 16
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m3, m1, [r5 + 3 * mmsize]
-    paddw           m9, m3
-    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
-    paddw           m11, m3
-    pmaddubsw       m3, m1, [r5 + 1 * mmsize]
-    paddw           m13, m3
-    pmaddubsw       m1, [r5]
-    movu            xm3, [r0 + r1]                  ; m3 = row 17
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, [r5 + 3 * mmsize]
-    paddw           m10, m4
-    pmaddubsw       m4, m2, [r5 + 2 * mmsize]
-    paddw           m12, m4
-    pmaddubsw       m2, [r5 + 1 * mmsize]
-    paddw           m0, m2
-    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
-    punpckhbw       xm5, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, [r5 + 3 * mmsize]
-    paddw           m11, m5
-    pmaddubsw       m5, m3, [r5 + 2 * mmsize]
-    paddw           m13, m5
-    pmaddubsw       m3, [r5 + 1 * mmsize]
-    paddw           m1, m3
-    movu            xm5, [r0 + r4]                  ; m5 = row 19
-    punpckhbw       xm6, xm4, xm5
-    punpcklbw       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, [r5 + 3 * mmsize]
-    paddw           m12, m6
-    pmaddubsw       m4, [r5 + 2 * mmsize]
-    paddw           m0, m4
-    lea             r0, [r0 + r1 * 4]
-    movu            xm6, [r0]                       ; m6 = row 20
-    punpckhbw       xm7, xm5, xm6
-    punpcklbw       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m7, m5, [r5 + 3 * mmsize]
-    paddw           m13, m7
-    pmaddubsw       m5, [r5 + 2 * mmsize]
-    paddw           m1, m5
-    movu            xm7, [r0 + r1]                  ; m7 = row 21
-    punpckhbw       xm2, xm6, xm7
-    punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm2, 1
-    pmaddubsw       m6, [r5 + 3 * mmsize]
-    paddw           m0, m6
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 22
-    punpckhbw       xm3, xm7, xm2
-    punpcklbw       xm7, xm2
-    vinserti128     m7, m7, xm3, 1
-    pmaddubsw       m7, [r5 + 3 * mmsize]
-    paddw           m1, m7
-
-%ifidn %3,pp
-    pmulhrsw        m8, m14                         ; m8 = word: row 8
-    pmulhrsw        m9, m14                         ; m9 = word: row 9
-    pmulhrsw        m10, m14                        ; m10 = word: row 10
-    pmulhrsw        m11, m14                        ; m11 = word: row 11
-    pmulhrsw        m12, m14                        ; m12 = word: row 12
-    pmulhrsw        m13, m14                        ; m13 = word: row 13
-    pmulhrsw        m0, m14                         ; m0 = word: row 14
-    pmulhrsw        m1, m14                         ; m1 = word: row 15
-    packuswb        m8, m9
-    packuswb        m10, m11
-    packuswb        m12, m13
-    packuswb        m0, m1
-    vpermq          m8, m8, 11011000b
-    vpermq          m10, m10, 11011000b
-    vpermq          m12, m12, 11011000b
-    vpermq          m0, m0, 11011000b
-    vextracti128    xm9, m8, 1
-    vextracti128    xm11, m10, 1
-    vextracti128    xm13, m12, 1
-    vextracti128    xm1, m0, 1
-    movu            [r2], xm8
-    movu            [r2 + r3], xm9
-    movu            [r2 + r3 * 2], xm10
-    movu            [r2 + r6], xm11
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm12
-    movu            [r2 + r3], xm13
-    movu            [r2 + r3 * 2], xm0
-    movu            [r2 + r6], xm1
-%else
-    psubw           m8, m14                         ; m8 = word: row 8
-    psubw           m9, m14                         ; m9 = word: row 9
-    psubw           m10, m14                        ; m10 = word: row 10
-    psubw           m11, m14                        ; m11 = word: row 11
-    psubw           m12, m14                        ; m12 = word: row 12
-    psubw           m13, m14                        ; m13 = word: row 13
-    psubw           m0, m14                         ; m0 = word: row 14
-    psubw           m1, m14                         ; m1 = word: row 15
-    movu            [r2], m8
-    movu            [r2 + r3], m9
-    movu            [r2 + r3 * 2], m10
-    movu            [r2 + r6], m11
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], m12
-    movu            [r2 + r3], m13
-    movu            [r2 + r3 * 2], m0
-    movu            [r2 + r6], m1
-%endif
-
-    lea             r2, [r2 + r3 * 4]
-    sub             r0, r7
-    dec             r8d
-    jnz             .loop
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_16xN 16, 32, pp
-    FILTER_VER_LUMA_AVX2_16xN 16, 64, pp
-    FILTER_VER_LUMA_AVX2_16xN 16, 32, ps
-    FILTER_VER_LUMA_AVX2_16xN 16, 64, ps
-
-%macro PROCESS_LUMA_AVX2_W16_16R 1
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
-    paddw           m0, m4
-    pmaddubsw       m2, [r5]
-    lea             r7, [r0 + r1 * 4]
-    movu            xm4, [r7]                       ; m4 = row 4
-    punpckhbw       xm5, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
-    paddw           m1, m5
-    pmaddubsw       m3, [r5]
-    movu            xm5, [r7 + r1]                  ; m5 = row 5
-    punpckhbw       xm6, xm4, xm5
-    punpcklbw       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
-    paddw           m0, m6
-    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
-    paddw           m2, m6
-    pmaddubsw       m4, [r5]
-    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
-    punpckhbw       xm7, xm5, xm6
-    punpcklbw       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
-    paddw           m1, m7
-    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
-    paddw           m3, m7
-    pmaddubsw       m5, [r5]
-    movu            xm7, [r7 + r4]                  ; m7 = row 7
-    punpckhbw       xm8, xm6, xm7
-    punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm8, 1
-    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
-    paddw           m0, m8
-    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
-    paddw           m2, m8
-    pmaddubsw       m8, m6, [r5 + 1 * mmsize]
-    paddw           m4, m8
-    pmaddubsw       m6, [r5]
-    lea             r7, [r7 + r1 * 4]
-    movu            xm8, [r7]                       ; m8 = row 8
-    punpckhbw       xm9, xm7, xm8
-    punpcklbw       xm7, xm8
-    vinserti128     m7, m7, xm9, 1
-    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
-    paddw           m1, m9
-    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
-    paddw           m3, m9
-    pmaddubsw       m9, m7, [r5 + 1 * mmsize]
-    paddw           m5, m9
-    pmaddubsw       m7, [r5]
-    movu            xm9, [r7 + r1]                  ; m9 = row 9
-    punpckhbw       xm10, xm8, xm9
-    punpcklbw       xm8, xm9
-    vinserti128     m8, m8, xm10, 1
-    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
-    paddw           m2, m10
-    pmaddubsw       m10, m8, [r5 + 2 * mmsize]
-    paddw           m4, m10
-    pmaddubsw       m10, m8, [r5 + 1 * mmsize]
-    paddw           m6, m10
-    pmaddubsw       m8, [r5]
-    movu            xm10, [r7 + r1 * 2]             ; m10 = row 10
-    punpckhbw       xm11, xm9, xm10
-    punpcklbw       xm9, xm10
-    vinserti128     m9, m9, xm11, 1
-    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
-    paddw           m3, m11
-    pmaddubsw       m11, m9, [r5 + 2 * mmsize]
-    paddw           m5, m11
-    pmaddubsw       m11, m9, [r5 + 1 * mmsize]
-    paddw           m7, m11
-    pmaddubsw       m9, [r5]
-    movu            xm11, [r7 + r4]                 ; m11 = row 11
-    punpckhbw       xm12, xm10, xm11
-    punpcklbw       xm10, xm11
-    vinserti128     m10, m10, xm12, 1
-    pmaddubsw       m12, m10, [r5 + 3 * mmsize]
-    paddw           m4, m12
-    pmaddubsw       m12, m10, [r5 + 2 * mmsize]
-    paddw           m6, m12
-    pmaddubsw       m12, m10, [r5 + 1 * mmsize]
-    paddw           m8, m12
-    pmaddubsw       m10, [r5]
-    lea             r7, [r7 + r1 * 4]
-    movu            xm12, [r7]                      ; m12 = row 12
-    punpckhbw       xm13, xm11, xm12
-    punpcklbw       xm11, xm12
-    vinserti128     m11, m11, xm13, 1
-    pmaddubsw       m13, m11, [r5 + 3 * mmsize]
-    paddw           m5, m13
-    pmaddubsw       m13, m11, [r5 + 2 * mmsize]
-    paddw           m7, m13
-    pmaddubsw       m13, m11, [r5 + 1 * mmsize]
-    paddw           m9, m13
-    pmaddubsw       m11, [r5]
-
-%ifidn %1,pp
-    pmulhrsw        m0, m14                         ; m0 = word: row 0
-    pmulhrsw        m1, m14                         ; m1 = word: row 1
-    pmulhrsw        m2, m14                         ; m2 = word: row 2
-    pmulhrsw        m3, m14                         ; m3 = word: row 3
-    pmulhrsw        m4, m14                         ; m4 = word: row 4
-    pmulhrsw        m5, m14                         ; m5 = word: row 5
-    packuswb        m0, m1
-    packuswb        m2, m3
-    packuswb        m4, m5
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
-    vextracti128    xm5, m4, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-    lea             r8, [r2 + r3 * 4]
-    movu            [r8], xm4
-    movu            [r8 + r3], xm5
-%else
-    psubw           m0, m14                         ; m0 = word: row 0
-    psubw           m1, m14                         ; m1 = word: row 1
-    psubw           m2, m14                         ; m2 = word: row 2
-    psubw           m3, m14                         ; m3 = word: row 3
-    psubw           m4, m14                         ; m4 = word: row 4
-    psubw           m5, m14                         ; m5 = word: row 5
-    movu            [r2], m0
-    movu            [r2 + r3], m1
-    movu            [r2 + r3 * 2], m2
-    movu            [r2 + r6], m3
-    lea             r8, [r2 + r3 * 4]
-    movu            [r8], m4
-    movu            [r8 + r3], m5
-%endif
-
-    movu            xm13, [r7 + r1]                 ; m13 = row 13
-    punpckhbw       xm0, xm12, xm13
-    punpcklbw       xm12, xm13
-    vinserti128     m12, m12, xm0, 1
-    pmaddubsw       m0, m12, [r5 + 3 * mmsize]
-    paddw           m6, m0
-    pmaddubsw       m0, m12, [r5 + 2 * mmsize]
-    paddw           m8, m0
-    pmaddubsw       m0, m12, [r5 + 1 * mmsize]
-    paddw           m10, m0
-    pmaddubsw       m12, [r5]
-    movu            xm0, [r7 + r1 * 2]              ; m0 = row 14
-    punpckhbw       xm1, xm13, xm0
-    punpcklbw       xm13, xm0
-    vinserti128     m13, m13, xm1, 1
-    pmaddubsw       m1, m13, [r5 + 3 * mmsize]
-    paddw           m7, m1
-    pmaddubsw       m1, m13, [r5 + 2 * mmsize]
-    paddw           m9, m1
-    pmaddubsw       m1, m13, [r5 + 1 * mmsize]
-    paddw           m11, m1
-    pmaddubsw       m13, [r5]
-
-%ifidn %1,pp
-    pmulhrsw        m6, m14                         ; m6 = word: row 6
-    pmulhrsw        m7, m14                         ; m7 = word: row 7
-    packuswb        m6, m7
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm7, m6, 1
-    movu            [r8 + r3 * 2], xm6
-    movu            [r8 + r6], xm7
-%else
-    psubw           m6, m14                         ; m6 = word: row 6
-    psubw           m7, m14                         ; m7 = word: row 7
-    movu            [r8 + r3 * 2], m6
-    movu            [r8 + r6], m7
-%endif
-
-    lea             r8, [r8 + r3 * 4]
-
-    movu            xm1, [r7 + r4]                  ; m1 = row 15
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m2, m0, [r5 + 3 * mmsize]
-    paddw           m8, m2
-    pmaddubsw       m2, m0, [r5 + 2 * mmsize]
-    paddw           m10, m2
-    pmaddubsw       m2, m0, [r5 + 1 * mmsize]
-    paddw           m12, m2
-    pmaddubsw       m0, [r5]
-    lea             r7, [r7 + r1 * 4]
-    movu            xm2, [r7]                       ; m2 = row 16
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m3, m1, [r5 + 3 * mmsize]
-    paddw           m9, m3
-    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
-    paddw           m11, m3
-    pmaddubsw       m3, m1, [r5 + 1 * mmsize]
-    paddw           m13, m3
-    pmaddubsw       m1, [r5]
-    movu            xm3, [r7 + r1]                  ; m3 = row 17
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, [r5 + 3 * mmsize]
-    paddw           m10, m4
-    pmaddubsw       m4, m2, [r5 + 2 * mmsize]
-    paddw           m12, m4
-    pmaddubsw       m2, [r5 + 1 * mmsize]
-    paddw           m0, m2
-    movu            xm4, [r7 + r1 * 2]              ; m4 = row 18
-    punpckhbw       xm5, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, [r5 + 3 * mmsize]
-    paddw           m11, m5
-    pmaddubsw       m5, m3, [r5 + 2 * mmsize]
-    paddw           m13, m5
-    pmaddubsw       m3, [r5 + 1 * mmsize]
-    paddw           m1, m3
-    movu            xm5, [r7 + r4]                  ; m5 = row 19
-    punpckhbw       xm6, xm4, xm5
-    punpcklbw       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, [r5 + 3 * mmsize]
-    paddw           m12, m6
-    pmaddubsw       m4, [r5 + 2 * mmsize]
-    paddw           m0, m4
-    lea             r7, [r7 + r1 * 4]
-    movu            xm6, [r7]                       ; m6 = row 20
-    punpckhbw       xm7, xm5, xm6
-    punpcklbw       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m7, m5, [r5 + 3 * mmsize]
-    paddw           m13, m7
-    pmaddubsw       m5, [r5 + 2 * mmsize]
-    paddw           m1, m5
-    movu            xm7, [r7 + r1]                  ; m7 = row 21
-    punpckhbw       xm2, xm6, xm7
-    punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm2, 1
-    pmaddubsw       m6, [r5 + 3 * mmsize]
-    paddw           m0, m6
-    movu            xm2, [r7 + r1 * 2]              ; m2 = row 22
-    punpckhbw       xm3, xm7, xm2
-    punpcklbw       xm7, xm2
-    vinserti128     m7, m7, xm3, 1
-    pmaddubsw       m7, [r5 + 3 * mmsize]
-    paddw           m1, m7
-
-%ifidn %1,pp
-    pmulhrsw        m8, m14                         ; m8 = word: row 8
-    pmulhrsw        m9, m14                         ; m9 = word: row 9
-    pmulhrsw        m10, m14                        ; m10 = word: row 10
-    pmulhrsw        m11, m14                        ; m11 = word: row 11
-    pmulhrsw        m12, m14                        ; m12 = word: row 12
-    pmulhrsw        m13, m14                        ; m13 = word: row 13
-    pmulhrsw        m0, m14                         ; m0 = word: row 14
-    pmulhrsw        m1, m14                         ; m1 = word: row 15
-    packuswb        m8, m9
-    packuswb        m10, m11
-    packuswb        m12, m13
-    packuswb        m0, m1
-    vpermq          m8, m8, 11011000b
-    vpermq          m10, m10, 11011000b
-    vpermq          m12, m12, 11011000b
-    vpermq          m0, m0, 11011000b
-    vextracti128    xm9, m8, 1
-    vextracti128    xm11, m10, 1
-    vextracti128    xm13, m12, 1
-    vextracti128    xm1, m0, 1
-    movu            [r8], xm8
-    movu            [r8 + r3], xm9
-    movu            [r8 + r3 * 2], xm10
-    movu            [r8 + r6], xm11
-    lea             r8, [r8 + r3 * 4]
-    movu            [r8], xm12
-    movu            [r8 + r3], xm13
-    movu            [r8 + r3 * 2], xm0
-    movu            [r8 + r6], xm1
-%else
-    psubw           m8, m14                         ; m8 = word: row 8
-    psubw           m9, m14                         ; m9 = word: row 9
-    psubw           m10, m14                        ; m10 = word: row 10
-    psubw           m11, m14                        ; m11 = word: row 11
-    psubw           m12, m14                        ; m12 = word: row 12
-    psubw           m13, m14                        ; m13 = word: row 13
-    psubw           m0, m14                         ; m0 = word: row 14
-    psubw           m1, m14                         ; m1 = word: row 15
-    movu            [r8], m8
-    movu            [r8 + r3], m9
-    movu            [r8 + r3 * 2], m10
-    movu            [r8 + r6], m11
-    lea             r8, [r8 + r3 * 4]
-    movu            [r8], m12
-    movu            [r8 + r3], m13
-    movu            [r8 + r3 * 2], m0
-    movu            [r8 + r6], m1
-%endif
-%endmacro
-
-%macro PROCESS_LUMA_AVX2_W16_8R 1
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
-    paddw           m0, m4
-    pmaddubsw       m2, [r5]
-    lea             r7, [r0 + r1 * 4]
-    movu            xm4, [r7]                       ; m4 = row 4
-    punpckhbw       xm5, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
-    paddw           m1, m5
-    pmaddubsw       m3, [r5]
-    movu            xm5, [r7 + r1]                  ; m5 = row 5
-    punpckhbw       xm6, xm4, xm5
-    punpcklbw       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
-    paddw           m0, m6
-    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
-    paddw           m2, m6
-    pmaddubsw       m4, [r5]
-    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
-    punpckhbw       xm7, xm5, xm6
-    punpcklbw       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
-    paddw           m1, m7
-    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
-    paddw           m3, m7
-    pmaddubsw       m5, [r5]
-    movu            xm7, [r7 + r4]                  ; m7 = row 7
-    punpckhbw       xm8, xm6, xm7
-    punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm8, 1
-    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
-    paddw           m0, m8
-    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
-    paddw           m2, m8
-    pmaddubsw       m8, m6, [r5 + 1 * mmsize]
-    paddw           m4, m8
-    pmaddubsw       m6, [r5]
-    lea             r7, [r7 + r1 * 4]
-    movu            xm8, [r7]                       ; m8 = row 8
-    punpckhbw       xm9, xm7, xm8
-    punpcklbw       xm7, xm8
-    vinserti128     m7, m7, xm9, 1
-    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
-    paddw           m1, m9
-    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
-    paddw           m3, m9
-    pmaddubsw       m9, m7, [r5 + 1 * mmsize]
-    paddw           m5, m9
-    pmaddubsw       m7, [r5]
-    movu            xm9, [r7 + r1]                  ; m9 = row 9
-    punpckhbw       xm10, xm8, xm9
-    punpcklbw       xm8, xm9
-    vinserti128     m8, m8, xm10, 1
-    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
-    paddw           m2, m10
-    pmaddubsw       m10, m8, [r5 + 2 * mmsize]
-    paddw           m4, m10
-    pmaddubsw       m10, m8, [r5 + 1 * mmsize]
-    paddw           m6, m10
-    movu            xm10, [r7 + r1 * 2]             ; m10 = row 10
-    punpckhbw       xm11, xm9, xm10
-    punpcklbw       xm9, xm10
-    vinserti128     m9, m9, xm11, 1
-    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
-    paddw           m3, m11
-    pmaddubsw       m11, m9, [r5 + 2 * mmsize]
-    paddw           m5, m11
-    pmaddubsw       m11, m9, [r5 + 1 * mmsize]
-    paddw           m7, m11
-    movu            xm11, [r7 + r4]                 ; m11 = row 11
-    punpckhbw       xm12, xm10, xm11
-    punpcklbw       xm10, xm11
-    vinserti128     m10, m10, xm12, 1
-    pmaddubsw       m12, m10, [r5 + 3 * mmsize]
-    paddw           m4, m12
-    pmaddubsw       m12, m10, [r5 + 2 * mmsize]
-    paddw           m6, m12
-    lea             r7, [r7 + r1 * 4]
-    movu            xm12, [r7]                      ; m12 = row 12
-    punpckhbw       xm13, xm11, xm12
-    punpcklbw       xm11, xm12
-    vinserti128     m11, m11, xm13, 1
-    pmaddubsw       m13, m11, [r5 + 3 * mmsize]
-    paddw           m5, m13
-    pmaddubsw       m13, m11, [r5 + 2 * mmsize]
-    paddw           m7, m13
-
-%ifidn %1,pp
-    pmulhrsw        m0, m14                         ; m0 = word: row 0
-    pmulhrsw        m1, m14                         ; m1 = word: row 1
-    pmulhrsw        m2, m14                         ; m2 = word: row 2
-    pmulhrsw        m3, m14                         ; m3 = word: row 3
-    pmulhrsw        m4, m14                         ; m4 = word: row 4
-    pmulhrsw        m5, m14                         ; m5 = word: row 5
-    packuswb        m0, m1
-    packuswb        m2, m3
-    packuswb        m4, m5
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
-    vextracti128    xm5, m4, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-    lea             r8, [r2 + r3 * 4]
-    movu            [r8], xm4
-    movu            [r8 + r3], xm5
-%else
-    psubw           m0, m14                         ; m0 = word: row 0
-    psubw           m1, m14                         ; m1 = word: row 1
-    psubw           m2, m14                         ; m2 = word: row 2
-    psubw           m3, m14                         ; m3 = word: row 3
-    psubw           m4, m14                         ; m4 = word: row 4
-    psubw           m5, m14                         ; m5 = word: row 5
-    movu            [r2], m0
-    movu            [r2 + r3], m1
-    movu            [r2 + r3 * 2], m2
-    movu            [r2 + r6], m3
-    lea             r8, [r2 + r3 * 4]
-    movu            [r8], m4
-    movu            [r8 + r3], m5
-%endif
-
-    movu            xm13, [r7 + r1]                 ; m13 = row 13
-    punpckhbw       xm0, xm12, xm13
-    punpcklbw       xm12, xm13
-    vinserti128     m12, m12, xm0, 1
-    pmaddubsw       m0, m12, [r5 + 3 * mmsize]
-    paddw           m6, m0
-    movu            xm0, [r7 + r1 * 2]              ; m0 = row 14
-    punpckhbw       xm1, xm13, xm0
-    punpcklbw       xm13, xm0
-    vinserti128     m13, m13, xm1, 1
-    pmaddubsw       m1, m13, [r5 + 3 * mmsize]
-    paddw           m7, m1
-
-%ifidn %1,pp
-    pmulhrsw        m6, m14                         ; m6 = word: row 6
-    pmulhrsw        m7, m14                         ; m7 = word: row 7
-    packuswb        m6, m7
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm7, m6, 1
-    movu            [r8 + r3 * 2], xm6
-    movu            [r8 + r6], xm7
-%else
-    psubw           m6, m14                         ; m6 = word: row 6
-    psubw           m7, m14                         ; m7 = word: row 7
-    movu            [r8 + r3 * 2], m6
-    movu            [r8 + r6], m7
-%endif
-%endmacro
-
-%macro FILTER_VER_LUMA_AVX2_24x32 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_24x32, 4, 11, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-%ifidn %1,ps
-    add             r3d, r3d
-    vbroadcasti128  m14, [pw_2000]
-%else
-    mova            m14, [pw_512]
-%endif
-    lea             r6, [r3 * 3]
-    lea             r10, [r1 * 4]
-    mov             r9d, 2
-.loopH:
-    PROCESS_LUMA_AVX2_W16_16R %1
-%ifidn %1,pp
-    add             r2, 16
-%else
-    add             r2, 32
-%endif
-    add             r0, 16
-    movq            xm1, [r0]                       ; m1 = row 0
-    movq            xm2, [r0 + r1]                  ; m2 = row 1
-    punpcklbw       xm1, xm2
-    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
-    punpcklbw       xm2, xm3
-    vinserti128     m5, m1, xm2, 1
-    pmaddubsw       m5, [r5]
-    movq            xm4, [r0 + r4]                  ; m4 = row 3
-    punpcklbw       xm3, xm4
-    lea             r7, [r0 + r1 * 4]
-    movq            xm1, [r7]                       ; m1 = row 4
-    punpcklbw       xm4, xm1
-    vinserti128     m2, m3, xm4, 1
-    pmaddubsw       m0, m2, [r5 + 1 * mmsize]
-    paddw           m5, m0
-    pmaddubsw       m2, [r5]
-    movq            xm3, [r7 + r1]                  ; m3 = row 5
-    punpcklbw       xm1, xm3
-    movq            xm4, [r7 + r1 * 2]              ; m4 = row 6
-    punpcklbw       xm3, xm4
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
-    paddw           m5, m3
-    pmaddubsw       m0, m1, [r5 + 1 * mmsize]
-    paddw           m2, m0
-    pmaddubsw       m1, [r5]
-    movq            xm3, [r7 + r4]                  ; m3 = row 7
-    punpcklbw       xm4, xm3
-    lea             r7, [r7 + r1 * 4]
-    movq            xm0, [r7]                       ; m0 = row 8
-    punpcklbw       xm3, xm0
-    vinserti128     m4, m4, xm3, 1
-    pmaddubsw       m3, m4, [r5 + 3 * mmsize]
-    paddw           m5, m3
-    pmaddubsw       m3, m4, [r5 + 2 * mmsize]
-    paddw           m2, m3
-    pmaddubsw       m3, m4, [r5 + 1 * mmsize]
-    paddw           m1, m3
-    pmaddubsw       m4, [r5]
-    movq            xm3, [r7 + r1]                  ; m3 = row 9
-    punpcklbw       xm0, xm3
-    movq            xm6, [r7 + r1 * 2]              ; m6 = row 10
-    punpcklbw       xm3, xm6
-    vinserti128     m0, m0, xm3, 1
-    pmaddubsw       m3, m0, [r5 + 3 * mmsize]
-    paddw           m2, m3
-    pmaddubsw       m3, m0, [r5 + 2 * mmsize]
-    paddw           m1, m3
-    pmaddubsw       m3, m0, [r5 + 1 * mmsize]
-    paddw           m4, m3
-    pmaddubsw       m0, [r5]
-
-    movq            xm3, [r7 + r4]                  ; m3 = row 11
-    punpcklbw       xm6, xm3
-    lea             r7, [r7 + r1 * 4]
-    movq            xm7, [r7]                       ; m7 = row 12
-    punpcklbw       xm3, xm7
-    vinserti128     m6, m6, xm3, 1
-    pmaddubsw       m3, m6, [r5 + 3 * mmsize]
-    paddw           m1, m3
-    pmaddubsw       m3, m6, [r5 + 2 * mmsize]
-    paddw           m4, m3
-    pmaddubsw       m3, m6, [r5 + 1 * mmsize]
-    paddw           m0, m3
-    pmaddubsw       m6, [r5]
-    movq            xm3, [r7 + r1]                  ; m3 = row 13
-    punpcklbw       xm7, xm3
-    movq            xm8, [r7 + r1 * 2]              ; m8 = row 14
-    punpcklbw       xm3, xm8
-    vinserti128     m7, m7, xm3, 1
-    pmaddubsw       m3, m7, [r5 + 3 * mmsize]
-    paddw           m4, m3
-    pmaddubsw       m3, m7, [r5 + 2 * mmsize]
-    paddw           m0, m3
-    pmaddubsw       m3, m7, [r5 + 1 * mmsize]
-    paddw           m6, m3
-    pmaddubsw       m7, [r5]
-    movq            xm3, [r7 + r4]                  ; m3 = row 15
-    punpcklbw       xm8, xm3
-    lea             r7, [r7 + r1 * 4]
-    movq            xm9, [r7]                       ; m9 = row 16
-    punpcklbw       xm3, xm9
-    vinserti128     m8, m8, xm3, 1
-    pmaddubsw       m3, m8, [r5 + 3 * mmsize]
-    paddw           m0, m3
-    pmaddubsw       m3, m8, [r5 + 2 * mmsize]
-    paddw           m6, m3
-    pmaddubsw       m3, m8, [r5 + 1 * mmsize]
-    paddw           m7, m3
-    pmaddubsw       m8, [r5]
-    movq            xm3, [r7 + r1]                  ; m3 = row 17
-    punpcklbw       xm9, xm3
-    movq            xm10, [r7 + r1 * 2]             ; m10 = row 18
-    punpcklbw       xm3, xm10
-    vinserti128     m9, m9, xm3, 1
-    pmaddubsw       m3, m9, [r5 + 3 * mmsize]
-    paddw           m6, m3
-    pmaddubsw       m3, m9, [r5 + 2 * mmsize]
-    paddw           m7, m3
-    pmaddubsw       m3, m9, [r5 + 1 * mmsize]
-    paddw           m8, m3
-    movq            xm3, [r7 + r4]                  ; m3 = row 19
-    punpcklbw       xm10, xm3
-    lea             r7, [r7 + r1 * 4]
-    movq            xm9, [r7]                       ; m9 = row 20
-    punpcklbw       xm3, xm9
-    vinserti128     m10, m10, xm3, 1
-    pmaddubsw       m3, m10, [r5 + 3 * mmsize]
-    paddw           m7, m3
-    pmaddubsw       m3, m10, [r5 + 2 * mmsize]
-    paddw           m8, m3
-    movq            xm3, [r7 + r1]                  ; m3 = row 21
-    punpcklbw       xm9, xm3
-    movq            xm10, [r7 + r1 * 2]             ; m10 = row 22
-    punpcklbw       xm3, xm10
-    vinserti128     m9, m9, xm3, 1
-    pmaddubsw       m3, m9, [r5 + 3 * mmsize]
-    paddw           m8, m3
-%ifidn %1,pp
-    pmulhrsw        m5, m14                         ; m5 = word: row 0, row 1
-    pmulhrsw        m2, m14                         ; m2 = word: row 2, row 3
-    pmulhrsw        m1, m14                         ; m1 = word: row 4, row 5
-    pmulhrsw        m4, m14                         ; m4 = word: row 6, row 7
-    pmulhrsw        m0, m14                         ; m0 = word: row 8, row 9
-    pmulhrsw        m6, m14                         ; m6 = word: row 10, row 11
-    pmulhrsw        m7, m14                         ; m7 = word: row 12, row 13
-    pmulhrsw        m8, m14                         ; m8 = word: row 14, row 15
-    packuswb        m5, m2
-    packuswb        m1, m4
-    packuswb        m0, m6
-    packuswb        m7, m8
-    vextracti128    xm2, m5, 1
-    vextracti128    xm4, m1, 1
-    vextracti128    xm6, m0, 1
-    vextracti128    xm8, m7, 1
-    movq            [r2], xm5
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm5
-    movhps          [r2 + r6], xm2
-    lea             r8, [r2 + r3 * 4]
-    movq            [r8], xm1
-    movq            [r8 + r3], xm4
-    movhps          [r8 + r3 * 2], xm1
-    movhps          [r8 + r6], xm4
-    lea             r8, [r8 + r3 * 4]
-    movq            [r8], xm0
-    movq            [r8 + r3], xm6
-    movhps          [r8 + r3 * 2], xm0
-    movhps          [r8 + r6], xm6
-    lea             r8, [r8 + r3 * 4]
-    movq            [r8], xm7
-    movq            [r8 + r3], xm8
-    movhps          [r8 + r3 * 2], xm7
-    movhps          [r8 + r6], xm8
-%else
-    psubw           m5, m14                         ; m5 = word: row 0, row 1
-    psubw           m2, m14                         ; m2 = word: row 2, row 3
-    psubw           m1, m14                         ; m1 = word: row 4, row 5
-    psubw           m4, m14                         ; m4 = word: row 6, row 7
-    psubw           m0, m14                         ; m0 = word: row 8, row 9
-    psubw           m6, m14                         ; m6 = word: row 10, row 11
-    psubw           m7, m14                         ; m7 = word: row 12, row 13
-    psubw           m8, m14                         ; m8 = word: row 14, row 15
-    vextracti128    xm3, m5, 1
-    movu            [r2], xm5
-    movu            [r2 + r3], xm3
-    vextracti128    xm3, m2, 1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-    vextracti128    xm3, m1, 1
-    lea             r8, [r2 + r3 * 4]
-    movu            [r8], xm1
-    movu            [r8 + r3], xm3
-    vextracti128    xm3, m4, 1
-    movu            [r8 + r3 * 2], xm4
-    movu            [r8 + r6], xm3
-    vextracti128    xm3, m0, 1
-    lea             r8, [r8 + r3 * 4]
-    movu            [r8], xm0
-    movu            [r8 + r3], xm3
-    vextracti128    xm3, m6, 1
-    movu            [r8 + r3 * 2], xm6
-    movu            [r8 + r6], xm3
-    vextracti128    xm3, m7, 1
-    lea             r8, [r8 + r3 * 4]
-    movu            [r8], xm7
-    movu            [r8 + r3], xm3
-    vextracti128    xm3, m8, 1
-    movu            [r8 + r3 * 2], xm8
-    movu            [r8 + r6], xm3
-%endif
-    sub             r7, r10
-    lea             r0, [r7 - 16]
-%ifidn %1,pp
-    lea             r2, [r8 + r3 * 4 - 16]
-%else
-    lea             r2, [r8 + r3 * 4 - 32]
-%endif
-    dec             r9d
-    jnz             .loopH
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_24x32 pp
-    FILTER_VER_LUMA_AVX2_24x32 ps
-
-%macro FILTER_VER_LUMA_AVX2_32xN 3
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%3_%1x%2, 4, 12, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-%ifidn %3,ps
-    add             r3d, r3d
-    vbroadcasti128  m14, [pw_2000]
-%else
-    mova            m14, [pw_512]
-%endif
-    lea             r6, [r3 * 3]
-    lea             r11, [r1 * 4]
-    mov             r9d, %2 / 16
-.loopH:
-    mov             r10d, %1 / 16
-.loopW:
-    PROCESS_LUMA_AVX2_W16_16R %3
-%ifidn %3,pp
-    add             r2, 16
-%else
-    add             r2, 32
-%endif
-    add             r0, 16
-    dec             r10d
-    jnz             .loopW
-    sub             r7, r11
-    lea             r0, [r7 - 16]
-%ifidn %3,pp
-    lea             r2, [r8 + r3 * 4 - 16]
-%else
-    lea             r2, [r8 + r3 * 4 - 32]
-%endif
-    dec             r9d
-    jnz             .loopH
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_32xN 32, 32, pp
-    FILTER_VER_LUMA_AVX2_32xN 32, 64, pp
-    FILTER_VER_LUMA_AVX2_32xN 32, 32, ps
-    FILTER_VER_LUMA_AVX2_32xN 32, 64, ps
-
-%macro FILTER_VER_LUMA_AVX2_32x16 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_32x16, 4, 10, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-%ifidn %1,ps
-    add             r3d, r3d
-    vbroadcasti128  m14, [pw_2000]
-%else
-    mova            m14, [pw_512]
-%endif
-    lea             r6, [r3 * 3]
-    mov             r9d, 2
-.loopW:
-    PROCESS_LUMA_AVX2_W16_16R %1
-%ifidn %1,pp
-    add             r2, 16
-%else
-    add             r2, 32
-%endif
-    add             r0, 16
-    dec             r9d
-    jnz             .loopW
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_32x16 pp
-    FILTER_VER_LUMA_AVX2_32x16 ps
-
-%macro FILTER_VER_LUMA_AVX2_32x24 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_32x24, 4, 10, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-%ifidn %1,ps
-    add             r3d, r3d
-%endif
-    lea             r6, [r3 * 3]
-%ifidn %1,pp
-    mova            m14, [pw_512]
-%else
-    vbroadcasti128  m14, [pw_2000]
-%endif
-    mov             r9d, 2
-.loopW:
-    PROCESS_LUMA_AVX2_W16_16R %1
-%ifidn %1,pp
-    add             r2, 16
-%else
-    add             r2, 32
-%endif
-    add             r0, 16
-    dec             r9d
-    jnz             .loopW
-    lea             r9, [r1 * 4]
-    sub             r7, r9
-    lea             r0, [r7 - 16]
-%ifidn %1,pp
-    lea             r2, [r8 + r3 * 4 - 16]
-%else
-    lea             r2, [r8 + r3 * 4 - 32]
-%endif
-    mov             r9d, 2
-.loop:
-    PROCESS_LUMA_AVX2_W16_8R %1
-%ifidn %1,pp
-    add             r2, 16
-%else
-    add             r2, 32
-%endif
-    add             r0, 16
-    dec             r9d
-    jnz             .loop
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_32x24 pp
-    FILTER_VER_LUMA_AVX2_32x24 ps
-
-%macro FILTER_VER_LUMA_AVX2_32x8 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_32x8, 4, 10, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-%ifidn %1,ps
-    add             r3d, r3d
-%endif
-    lea             r6, [r3 * 3]
-%ifidn %1,pp
-    mova            m14, [pw_512]
-%else
-    vbroadcasti128  m14, [pw_2000]
-%endif
-    mov             r9d, 2
-.loopW:
-    PROCESS_LUMA_AVX2_W16_8R %1
-%ifidn %1,pp
-    add             r2, 16
-%else
-    add             r2, 32
-%endif
-    add             r0, 16
-    dec             r9d
-    jnz             .loopW
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_32x8 pp
-    FILTER_VER_LUMA_AVX2_32x8 ps
-
-%macro FILTER_VER_LUMA_AVX2_48x64 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_48x64, 4, 12, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-
-%ifidn %1,ps
-    add             r3d, r3d
-%endif
-
-    lea             r6, [r3 * 3]
-    lea             r11, [r1 * 4]
-
-%ifidn %1,pp
-    mova            m14, [pw_512]
-%else
-    vbroadcasti128  m14, [pw_2000]
-%endif
-
-    mov             r9d, 4
-.loopH:
-    mov             r10d, 3
-.loopW:
-    PROCESS_LUMA_AVX2_W16_16R %1
-%ifidn %1,pp
-    add             r2, 16
-%else
-    add             r2, 32
-%endif
-    add             r0, 16
-    dec             r10d
-    jnz             .loopW
-    sub             r7, r11
-    lea             r0, [r7 - 32]
-%ifidn %1,pp
-    lea             r2, [r8 + r3 * 4 - 32]
-%else
-    lea             r2, [r8 + r3 * 4 - 64]
-%endif
-    dec             r9d
-    jnz             .loopH
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_48x64 pp
-    FILTER_VER_LUMA_AVX2_48x64 ps
-
-%macro FILTER_VER_LUMA_AVX2_64xN 3
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%3_%1x%2, 4, 12, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-
-%ifidn %3,ps
-    add             r3d, r3d
-%endif
-
-    lea             r6, [r3 * 3]
-    lea             r11, [r1 * 4]
-
-%ifidn %3,pp
-    mova            m14, [pw_512]
-%else
-    vbroadcasti128  m14, [pw_2000]
-%endif
-
-    mov             r9d, %2 / 16
-.loopH:
-    mov             r10d, %1 / 16
-.loopW:
-    PROCESS_LUMA_AVX2_W16_16R %3
-%ifidn %3,pp
-    add             r2, 16
-%else
-    add             r2, 32
-%endif
-    add             r0, 16
-    dec             r10d
-    jnz             .loopW
-    sub             r7, r11
-    lea             r0, [r7 - 48]
-%ifidn %3,pp
-    lea             r2, [r8 + r3 * 4 - 48]
-%else
-    lea             r2, [r8 + r3 * 4 - 96]
-%endif
-    dec             r9d
-    jnz             .loopH
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_64xN 64, 32, pp
-    FILTER_VER_LUMA_AVX2_64xN 64, 48, pp
-    FILTER_VER_LUMA_AVX2_64xN 64, 64, pp
-    FILTER_VER_LUMA_AVX2_64xN 64, 32, ps
-    FILTER_VER_LUMA_AVX2_64xN 64, 48, ps
-    FILTER_VER_LUMA_AVX2_64xN 64, 64, ps
-
-%macro FILTER_VER_LUMA_AVX2_64x16 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_64x16, 4, 10, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-
-%ifidn %1,ps
-    add             r3d, r3d
-%endif
-
-    lea             r6, [r3 * 3]
-
-%ifidn %1,pp
-    mova            m14, [pw_512]
-%else
-    vbroadcasti128  m14, [pw_2000]
-%endif
-
-    mov             r9d, 4
-.loopW:
-    PROCESS_LUMA_AVX2_W16_16R %1
-%ifidn %1,pp
-    add             r2, 16
-%else
-    add             r2, 32
-%endif
-    add             r0, 16
-    dec             r9d
-    jnz             .loopW
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_64x16 pp
-    FILTER_VER_LUMA_AVX2_64x16 ps
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_LUMA 3
-INIT_XMM sse4
-cglobal interp_8tap_vert_%3_%1x%2, 5, 7, 8 ,0-gprsize
-    lea       r5, [3 * r1]
-    sub       r0, r5
-    shl       r4d, 6
-%ifidn %3,ps
-    add       r3d, r3d
-%endif
-
-%ifdef PIC
-    lea       r5, [tab_LumaCoeffVer]
-    lea       r6, [r5 + r4]
-%else
-    lea       r6, [tab_LumaCoeffVer + r4]
-%endif
-
-%ifidn %3,pp
-    mova      m3, [pw_512]
-%else
-    mova      m3, [pw_2000]
-%endif
-    mov       dword [rsp], %2/4
-
-.loopH:
-    mov       r4d, (%1/8)
-.loopW:
-    PROCESS_LUMA_W8_4R
-%ifidn %3,pp
-    pmulhrsw  m7, m3
-    pmulhrsw  m6, m3
-    pmulhrsw  m5, m3
-    pmulhrsw  m4, m3
-
-    packuswb  m7, m6
-    packuswb  m5, m4
-
-    movlps    [r2], m7
-    movhps    [r2 + r3], m7
-    lea       r5, [r2 + 2 * r3]
-    movlps    [r5], m5
-    movhps    [r5 + r3], m5
-%else
-    psubw     m7, m3
-    psubw     m6, m3
-    psubw     m5, m3
-    psubw     m4, m3
-
-    movu      [r2], m7
-    movu      [r2 + r3], m6
-    lea       r5, [r2 + 2 * r3]
-    movu      [r5], m5
-    movu      [r5 + r3], m4
-%endif
-
-    lea       r5, [8 * r1 - 8]
-    sub       r0, r5
-%ifidn %3,pp
-    add       r2, 8
-%else
-    add       r2, 16
-%endif
-    dec       r4d
-    jnz       .loopW
-
-    lea       r0, [r0 + 4 * r1 - %1]
-%ifidn %3,pp
-    lea       r2, [r2 + 4 * r3 - %1]
-%else
-    lea       r2, [r2 + 4 * r3 - 2 * %1]
-%endif
-
-    dec       dword [rsp]
-    jnz       .loopH
-
-    RET
-%endmacro
-
-    FILTER_VER_LUMA 16, 4, pp
-    FILTER_VER_LUMA 16, 8, pp
-    FILTER_VER_LUMA 16, 12, pp
-    FILTER_VER_LUMA 16, 16, pp
-    FILTER_VER_LUMA 16, 32, pp
-    FILTER_VER_LUMA 16, 64, pp
-    FILTER_VER_LUMA 24, 32, pp
-    FILTER_VER_LUMA 32, 8, pp
-    FILTER_VER_LUMA 32, 16, pp
-    FILTER_VER_LUMA 32, 24, pp
-    FILTER_VER_LUMA 32, 32, pp
-    FILTER_VER_LUMA 32, 64, pp
-    FILTER_VER_LUMA 48, 64, pp
-    FILTER_VER_LUMA 64, 16, pp
-    FILTER_VER_LUMA 64, 32, pp
-    FILTER_VER_LUMA 64, 48, pp
-    FILTER_VER_LUMA 64, 64, pp
-
-    FILTER_VER_LUMA 16, 4, ps
-    FILTER_VER_LUMA 16, 8, ps
-    FILTER_VER_LUMA 16, 12, ps
-    FILTER_VER_LUMA 16, 16, ps
-    FILTER_VER_LUMA 16, 32, ps
-    FILTER_VER_LUMA 16, 64, ps
-    FILTER_VER_LUMA 24, 32, ps
-    FILTER_VER_LUMA 32, 8, ps
-    FILTER_VER_LUMA 32, 16, ps
-    FILTER_VER_LUMA 32, 24, ps
-    FILTER_VER_LUMA 32, 32, ps
-    FILTER_VER_LUMA 32, 64, ps
-    FILTER_VER_LUMA 48, 64, ps
-    FILTER_VER_LUMA 64, 16, ps
-    FILTER_VER_LUMA 64, 32, ps
-    FILTER_VER_LUMA 64, 48, ps
-    FILTER_VER_LUMA 64, 64, ps
-
-%macro PROCESS_LUMA_SP_W4_4R 0
-    movq       m0, [r0]
-    movq       m1, [r0 + r1]
-    punpcklwd  m0, m1                          ;m0=[0 1]
-    pmaddwd    m0, [r6 + 0 *16]                ;m0=[0+1]  Row1
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m4, [r0]
-    punpcklwd  m1, m4                          ;m1=[1 2]
-    pmaddwd    m1, [r6 + 0 *16]                ;m1=[1+2]  Row2
-
-    movq       m5, [r0 + r1]
-    punpcklwd  m4, m5                          ;m4=[2 3]
-    pmaddwd    m2, m4, [r6 + 0 *16]            ;m2=[2+3]  Row3
-    pmaddwd    m4, [r6 + 1 * 16]
-    paddd      m0, m4                          ;m0=[0+1+2+3]  Row1
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m4, [r0]
-    punpcklwd  m5, m4                          ;m5=[3 4]
-    pmaddwd    m3, m5, [r6 + 0 *16]            ;m3=[3+4]  Row4
-    pmaddwd    m5, [r6 + 1 * 16]
-    paddd      m1, m5                          ;m1 = [1+2+3+4]  Row2
-
-    movq       m5, [r0 + r1]
-    punpcklwd  m4, m5                          ;m4=[4 5]
-    pmaddwd    m6, m4, [r6 + 1 * 16]
-    paddd      m2, m6                          ;m2=[2+3+4+5]  Row3
-    pmaddwd    m4, [r6 + 2 * 16]
-    paddd      m0, m4                          ;m0=[0+1+2+3+4+5]  Row1
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m4, [r0]
-    punpcklwd  m5, m4                          ;m5=[5 6]
-    pmaddwd    m6, m5, [r6 + 1 * 16]
-    paddd      m3, m6                          ;m3=[3+4+5+6]  Row4
-    pmaddwd    m5, [r6 + 2 * 16]
-    paddd      m1, m5                          ;m1=[1+2+3+4+5+6]  Row2
-
-    movq       m5, [r0 + r1]
-    punpcklwd  m4, m5                          ;m4=[6 7]
-    pmaddwd    m6, m4, [r6 + 2 * 16]
-    paddd      m2, m6                          ;m2=[2+3+4+5+6+7]  Row3
-    pmaddwd    m4, [r6 + 3 * 16]
-    paddd      m0, m4                          ;m0=[0+1+2+3+4+5+6+7]  Row1 end
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m4, [r0]
-    punpcklwd  m5, m4                          ;m5=[7 8]
-    pmaddwd    m6, m5, [r6 + 2 * 16]
-    paddd      m3, m6                          ;m3=[3+4+5+6+7+8]  Row4
-    pmaddwd    m5, [r6 + 3 * 16]
-    paddd      m1, m5                          ;m1=[1+2+3+4+5+6+7+8]  Row2 end
-
-    movq       m5, [r0 + r1]
-    punpcklwd  m4, m5                          ;m4=[8 9]
-    pmaddwd    m4, [r6 + 3 * 16]
-    paddd      m2, m4                          ;m2=[2+3+4+5+6+7+8+9]  Row3 end
-
-    movq       m4, [r0 + 2 * r1]
-    punpcklwd  m5, m4                          ;m5=[9 10]
-    pmaddwd    m5, [r6 + 3 * 16]
-    paddd      m3, m5                          ;m3=[3+4+5+6+7+8+9+10]  Row4 end
-%endmacro
-
-;--------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_sp_%1x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_LUMA_SP 2
-INIT_XMM sse4
-cglobal interp_8tap_vert_sp_%1x%2, 5, 7, 8 ,0-gprsize
-
-    add       r1d, r1d
-    lea       r5, [r1 + 2 * r1]
-    sub       r0, r5
-    shl       r4d, 6
-
-%ifdef PIC
-    lea       r5, [tab_LumaCoeffV]
-    lea       r6, [r5 + r4]
-%else
-    lea       r6, [tab_LumaCoeffV + r4]
-%endif
-
-    mova      m7, [pd_526336]
-
-    mov       dword [rsp], %2/4
-.loopH:
-    mov       r4d, (%1/4)
-.loopW:
-    PROCESS_LUMA_SP_W4_4R
-
-    paddd     m0, m7
-    paddd     m1, m7
-    paddd     m2, m7
-    paddd     m3, m7
-
-    psrad     m0, 12
-    psrad     m1, 12
-    psrad     m2, 12
-    psrad     m3, 12
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-
-    packuswb  m0, m2
-
-    movd      [r2], m0
-    pextrd    [r2 + r3], m0, 1
-    lea       r5, [r2 + 2 * r3]
-    pextrd    [r5], m0, 2
-    pextrd    [r5 + r3], m0, 3
-
-    lea       r5, [8 * r1 - 2 * 4]
-    sub       r0, r5
-    add       r2, 4
-
-    dec       r4d
-    jnz       .loopW
-
-    lea       r0, [r0 + 4 * r1 - 2 * %1]
-    lea       r2, [r2 + 4 * r3 - %1]
-
-    dec       dword [rsp]
-    jnz       .loopH
-
-    RET
-%endmacro
-
-;--------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_sp_%1x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_SP 4, 4
-    FILTER_VER_LUMA_SP 8, 8
-    FILTER_VER_LUMA_SP 8, 4
-    FILTER_VER_LUMA_SP 4, 8
-    FILTER_VER_LUMA_SP 16, 16
-    FILTER_VER_LUMA_SP 16, 8
-    FILTER_VER_LUMA_SP 8, 16
-    FILTER_VER_LUMA_SP 16, 12
-    FILTER_VER_LUMA_SP 12, 16
-    FILTER_VER_LUMA_SP 16, 4
-    FILTER_VER_LUMA_SP 4, 16
-    FILTER_VER_LUMA_SP 32, 32
-    FILTER_VER_LUMA_SP 32, 16
-    FILTER_VER_LUMA_SP 16, 32
-    FILTER_VER_LUMA_SP 32, 24
-    FILTER_VER_LUMA_SP 24, 32
-    FILTER_VER_LUMA_SP 32, 8
-    FILTER_VER_LUMA_SP 8, 32
-    FILTER_VER_LUMA_SP 64, 64
-    FILTER_VER_LUMA_SP 64, 32
-    FILTER_VER_LUMA_SP 32, 64
-    FILTER_VER_LUMA_SP 64, 48
-    FILTER_VER_LUMA_SP 48, 64
-    FILTER_VER_LUMA_SP 64, 16
-    FILTER_VER_LUMA_SP 16, 64
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-INIT_XMM sse4
-cglobal filterPixelToShort_4x2, 3, 4, 3
-    mov         r3d, r3m
-    add         r3d, r3d
-
-    ; load constant
-    mova        m1, [pb_128]
-    mova        m2, [tab_c_64_n64]
-
-    movd        m0, [r0]
-    pinsrd      m0, [r0 + r1], 1
-    punpcklbw   m0, m1
-    pmaddubsw   m0, m2
-
-    movq        [r2 + r3 * 0], m0
-    movhps      [r2 + r3 * 1], m0
-
-    RET
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-INIT_XMM ssse3
-cglobal filterPixelToShort_8x2, 3, 4, 3
-    mov         r3d, r3m
-    add         r3d, r3d
-
-    ; load constant
-    mova        m1, [pb_128]
-    mova        m2, [tab_c_64_n64]
-
-    movh        m0, [r0]
-    punpcklbw   m0, m1
-    pmaddubsw   m0, m2
-    movu        [r2 + r3 * 0], m0
-
-    movh        m0, [r0 + r1]
-    punpcklbw   m0, m1
-    pmaddubsw   m0, m2
-    movu        [r2 + r3 * 1], m0
-
-    RET
-
-%macro PROCESS_CHROMA_SP_W4_4R 0
-    movq       m0, [r0]
-    movq       m1, [r0 + r1]
-    punpcklwd  m0, m1                          ;m0=[0 1]
-    pmaddwd    m0, [r6 + 0 *16]                ;m0=[0+1]         Row1
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m4, [r0]
-    punpcklwd  m1, m4                          ;m1=[1 2]
-    pmaddwd    m1, [r6 + 0 *16]                ;m1=[1+2]         Row2
-
-    movq       m5, [r0 + r1]
-    punpcklwd  m4, m5                          ;m4=[2 3]
-    pmaddwd    m2, m4, [r6 + 0 *16]            ;m2=[2+3]         Row3
-    pmaddwd    m4, [r6 + 1 * 16]
-    paddd      m0, m4                          ;m0=[0+1+2+3]     Row1 done
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m4, [r0]
-    punpcklwd  m5, m4                          ;m5=[3 4]
-    pmaddwd    m3, m5, [r6 + 0 *16]            ;m3=[3+4]         Row4
-    pmaddwd    m5, [r6 + 1 * 16]
-    paddd      m1, m5                          ;m1 = [1+2+3+4]   Row2
-
-    movq       m5, [r0 + r1]
-    punpcklwd  m4, m5                          ;m4=[4 5]
-    pmaddwd    m4, [r6 + 1 * 16]
-    paddd      m2, m4                          ;m2=[2+3+4+5]     Row3
-
-    movq       m4, [r0 + 2 * r1]
-    punpcklwd  m5, m4                          ;m5=[5 6]
-    pmaddwd    m5, [r6 + 1 * 16]
-    paddd      m3, m5                          ;m3=[3+4+5+6]     Row4
-%endmacro
-
-;--------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_sp_%1x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_SP 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_sp_%1x%2, 5, 7, 7 ,0-gprsize
-
-    add       r1d, r1d
-    sub       r0, r1
-    shl       r4d, 5
-
-%ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
-    lea       r6, [r5 + r4]
-%else
-    lea       r6, [tab_ChromaCoeffV + r4]
-%endif
-
-    mova      m6, [pd_526336]
-
-    mov       dword [rsp], %2/4
-
-.loopH:
-    mov       r4d, (%1/4)
-.loopW:
-    PROCESS_CHROMA_SP_W4_4R
-
-    paddd     m0, m6
-    paddd     m1, m6
-    paddd     m2, m6
-    paddd     m3, m6
-
-    psrad     m0, 12
-    psrad     m1, 12
-    psrad     m2, 12
-    psrad     m3, 12
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-
-    packuswb  m0, m2
-
-    movd      [r2], m0
-    pextrd    [r2 + r3], m0, 1
-    lea       r5, [r2 + 2 * r3]
-    pextrd    [r5], m0, 2
-    pextrd    [r5 + r3], m0, 3
-
-    lea       r5, [4 * r1 - 2 * 4]
-    sub       r0, r5
-    add       r2, 4
-
-    dec       r4d
-    jnz       .loopW
-
-    lea       r0, [r0 + 4 * r1 - 2 * %1]
-    lea       r2, [r2 + 4 * r3 - %1]
-
-    dec       dword [rsp]
-    jnz       .loopH
-
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_SP 4, 4
-    FILTER_VER_CHROMA_SP 4, 8
-    FILTER_VER_CHROMA_SP 16, 16
-    FILTER_VER_CHROMA_SP 16, 8
-    FILTER_VER_CHROMA_SP 16, 12
-    FILTER_VER_CHROMA_SP 12, 16
-    FILTER_VER_CHROMA_SP 16, 4
-    FILTER_VER_CHROMA_SP 4, 16
-    FILTER_VER_CHROMA_SP 32, 32
-    FILTER_VER_CHROMA_SP 32, 16
-    FILTER_VER_CHROMA_SP 16, 32
-    FILTER_VER_CHROMA_SP 32, 24
-    FILTER_VER_CHROMA_SP 24, 32
-    FILTER_VER_CHROMA_SP 32, 8
-
-    FILTER_VER_CHROMA_SP 16, 24
-    FILTER_VER_CHROMA_SP 16, 64
-    FILTER_VER_CHROMA_SP 12, 32
-    FILTER_VER_CHROMA_SP 4, 32
-    FILTER_VER_CHROMA_SP 32, 64
-    FILTER_VER_CHROMA_SP 32, 48
-    FILTER_VER_CHROMA_SP 24, 64
-
-    FILTER_VER_CHROMA_SP 64, 64
-    FILTER_VER_CHROMA_SP 64, 32
-    FILTER_VER_CHROMA_SP 64, 48
-    FILTER_VER_CHROMA_SP 48, 64
-    FILTER_VER_CHROMA_SP 64, 16
-
-
-%macro PROCESS_CHROMA_SP_W2_4R 1
-    movd       m0, [r0]
-    movd       m1, [r0 + r1]
-    punpcklwd  m0, m1                          ;m0=[0 1]
-
-    lea        r0, [r0 + 2 * r1]
-    movd       m2, [r0]
-    punpcklwd  m1, m2                          ;m1=[1 2]
-    punpcklqdq m0, m1                          ;m0=[0 1 1 2]
-    pmaddwd    m0, [%1 + 0 *16]                ;m0=[0+1 1+2] Row 1-2
-
-    movd       m1, [r0 + r1]
-    punpcklwd  m2, m1                          ;m2=[2 3]
-
-    lea        r0, [r0 + 2 * r1]
-    movd       m3, [r0]
-    punpcklwd  m1, m3                          ;m2=[3 4]
-    punpcklqdq m2, m1                          ;m2=[2 3 3 4]
-
-    pmaddwd    m4, m2, [%1 + 1 * 16]           ;m4=[2+3 3+4] Row 1-2
-    pmaddwd    m2, [%1 + 0 * 16]               ;m2=[2+3 3+4] Row 3-4
-    paddd      m0, m4                          ;m0=[0+1+2+3 1+2+3+4] Row 1-2
-
-    movd       m1, [r0 + r1]
-    punpcklwd  m3, m1                          ;m3=[4 5]
-
-    movd       m4, [r0 + 2 * r1]
-    punpcklwd  m1, m4                          ;m1=[5 6]
-    punpcklqdq m3, m1                          ;m2=[4 5 5 6]
-    pmaddwd    m3, [%1 + 1 * 16]               ;m3=[4+5 5+6] Row 3-4
-    paddd      m2, m3                          ;m2=[2+3+4+5 3+4+5+6] Row 3-4
-%endmacro
-
-;-------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vertical_sp_%1x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_SP_W2_4R 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_sp_%1x%2, 5, 6, 6
-
-    add       r1d, r1d
-    sub       r0, r1
-    shl       r4d, 5
-
-%ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
-    lea       r5, [r5 + r4]
-%else
-    lea       r5, [tab_ChromaCoeffV + r4]
-%endif
-
-    mova      m5, [pd_526336]
-
-    mov       r4d, (%2/4)
-
-.loopH:
-    PROCESS_CHROMA_SP_W2_4R r5
-
-    paddd     m0, m5
-    paddd     m2, m5
-
-    psrad     m0, 12
-    psrad     m2, 12
-
-    packssdw  m0, m2
-    packuswb  m0, m0
-
-    pextrw    [r2], m0, 0
-    pextrw    [r2 + r3], m0, 1
-    lea       r2, [r2 + 2 * r3]
-    pextrw    [r2], m0, 2
-    pextrw    [r2 + r3], m0, 3
-
-    lea       r2, [r2 + 2 * r3]
-
-    dec       r4d
-    jnz       .loopH
-
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_SP_W2_4R 2, 4
-    FILTER_VER_CHROMA_SP_W2_4R 2, 8
-
-    FILTER_VER_CHROMA_SP_W2_4R 2, 16
-
-;--------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_sp_4x2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-INIT_XMM sse4
-cglobal interp_4tap_vert_sp_4x2, 5, 6, 5
-
-    add        r1d, r1d
-    sub        r0, r1
-    shl        r4d, 5
-
-%ifdef PIC
-    lea        r5, [tab_ChromaCoeffV]
-    lea        r5, [r5 + r4]
-%else
-    lea        r5, [tab_ChromaCoeffV + r4]
-%endif
-
-    mova       m4, [pd_526336]
-
-    movq       m0, [r0]
-    movq       m1, [r0 + r1]
-    punpcklwd  m0, m1                          ;m0=[0 1]
-    pmaddwd    m0, [r5 + 0 *16]                ;m0=[0+1]  Row1
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m2, [r0]
-    punpcklwd  m1, m2                          ;m1=[1 2]
-    pmaddwd    m1, [r5 + 0 *16]                ;m1=[1+2]  Row2
-
-    movq       m3, [r0 + r1]
-    punpcklwd  m2, m3                          ;m4=[2 3]
-    pmaddwd    m2, [r5 + 1 * 16]
-    paddd      m0, m2                          ;m0=[0+1+2+3]  Row1 done
-    paddd      m0, m4
-    psrad      m0, 12
-
-    movq       m2, [r0 + 2 * r1]
-    punpcklwd  m3, m2                          ;m5=[3 4]
-    pmaddwd    m3, [r5 + 1 * 16]
-    paddd      m1, m3                          ;m1 = [1+2+3+4]  Row2 done
-    paddd      m1, m4
-    psrad      m1, 12
-
-    packssdw   m0, m1
-    packuswb   m0, m0
-
-    movd       [r2], m0
-    pextrd     [r2 + r3], m0, 1
-
-    RET
-
-;-------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vertical_sp_6x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_SP_W6_H4 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_sp_6x%2, 5, 7, 7
-
-    add       r1d, r1d
-    sub       r0, r1
-    shl       r4d, 5
-
-%ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
-    lea       r6, [r5 + r4]
-%else
-    lea       r6, [tab_ChromaCoeffV + r4]
-%endif
-
-    mova      m6, [pd_526336]
-
-    mov       r4d, %2/4
-
-.loopH:
-    PROCESS_CHROMA_SP_W4_4R
-
-    paddd     m0, m6
-    paddd     m1, m6
-    paddd     m2, m6
-    paddd     m3, m6
-
-    psrad     m0, 12
-    psrad     m1, 12
-    psrad     m2, 12
-    psrad     m3, 12
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-
-    packuswb  m0, m2
-
-    movd      [r2], m0
-    pextrd    [r2 + r3], m0, 1
-    lea       r5, [r2 + 2 * r3]
-    pextrd    [r5], m0, 2
-    pextrd    [r5 + r3], m0, 3
-
-    lea       r5, [4 * r1 - 2 * 4]
-    sub       r0, r5
-    add       r2, 4
-
-    PROCESS_CHROMA_SP_W2_4R r6
-
-    paddd     m0, m6
-    paddd     m2, m6
-
-    psrad     m0, 12
-    psrad     m2, 12
-
-    packssdw  m0, m2
-    packuswb  m0, m0
-
-    pextrw    [r2], m0, 0
-    pextrw    [r2 + r3], m0, 1
-    lea       r2, [r2 + 2 * r3]
-    pextrw    [r2], m0, 2
-    pextrw    [r2 + r3], m0, 3
-
-    sub       r0, 2 * 4
-    lea       r2, [r2 + 2 * r3 - 4]
-
-    dec       r4d
-    jnz       .loopH
-
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_SP_W6_H4 6, 8
-
-    FILTER_VER_CHROMA_SP_W6_H4 6, 16
-
-%macro PROCESS_CHROMA_SP_W8_2R 0
-    movu       m1, [r0]
-    movu       m3, [r0 + r1]
-    punpcklwd  m0, m1, m3
-    pmaddwd    m0, [r5 + 0 * 16]                ;m0 = [0l+1l]  Row1l
-    punpckhwd  m1, m3
-    pmaddwd    m1, [r5 + 0 * 16]                ;m1 = [0h+1h]  Row1h
-
-    movu       m4, [r0 + 2 * r1]
-    punpcklwd  m2, m3, m4
-    pmaddwd    m2, [r5 + 0 * 16]                ;m2 = [1l+2l]  Row2l
-    punpckhwd  m3, m4
-    pmaddwd    m3, [r5 + 0 * 16]                ;m3 = [1h+2h]  Row2h
-
-    lea        r0, [r0 + 2 * r1]
-    movu       m5, [r0 + r1]
-    punpcklwd  m6, m4, m5
-    pmaddwd    m6, [r5 + 1 * 16]                ;m6 = [2l+3l]  Row1l
-    paddd      m0, m6                           ;m0 = [0l+1l+2l+3l]  Row1l sum
-    punpckhwd  m4, m5
-    pmaddwd    m4, [r5 + 1 * 16]                ;m6 = [2h+3h]  Row1h
-    paddd      m1, m4                           ;m1 = [0h+1h+2h+3h]  Row1h sum
-
-    movu       m4, [r0 + 2 * r1]
-    punpcklwd  m6, m5, m4
-    pmaddwd    m6, [r5 + 1 * 16]                ;m6 = [3l+4l]  Row2l
-    paddd      m2, m6                           ;m2 = [1l+2l+3l+4l]  Row2l sum
-    punpckhwd  m5, m4
-    pmaddwd    m5, [r5 + 1 * 16]                ;m1 = [3h+4h]  Row2h
-    paddd      m3, m5                           ;m3 = [1h+2h+3h+4h]  Row2h sum
-%endmacro
-
-;--------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_sp_8x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_SP_W8_H2 2
-INIT_XMM sse2
-cglobal interp_4tap_vert_sp_%1x%2, 5, 6, 8
-
-    add       r1d, r1d
-    sub       r0, r1
-    shl       r4d, 5
-
-%ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
-    lea       r5, [r5 + r4]
-%else
-    lea       r5, [tab_ChromaCoeffV + r4]
-%endif
-
-    mova      m7, [pd_526336]
-
-    mov       r4d, %2/2
-.loopH:
-    PROCESS_CHROMA_SP_W8_2R
-
-    paddd     m0, m7
-    paddd     m1, m7
-    paddd     m2, m7
-    paddd     m3, m7
-
-    psrad     m0, 12
-    psrad     m1, 12
-    psrad     m2, 12
-    psrad     m3, 12
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-
-    packuswb  m0, m2
-
-    movlps    [r2], m0
-    movhps    [r2 + r3], m0
-
-    lea       r2, [r2 + 2 * r3]
-
-    dec r4d
-    jnz .loopH
-
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_SP_W8_H2 8, 2
-    FILTER_VER_CHROMA_SP_W8_H2 8, 4
-    FILTER_VER_CHROMA_SP_W8_H2 8, 6
-    FILTER_VER_CHROMA_SP_W8_H2 8, 8
-    FILTER_VER_CHROMA_SP_W8_H2 8, 16
-    FILTER_VER_CHROMA_SP_W8_H2 8, 32
-
-    FILTER_VER_CHROMA_SP_W8_H2 8, 12
-    FILTER_VER_CHROMA_SP_W8_H2 8, 64
-
-
-;-----------------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_ps_2x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;-----------------------------------------------------------------------------------------------------------------------------
-%macro FILTER_HORIZ_CHROMA_2xN 2
-INIT_XMM sse4
-cglobal interp_4tap_horiz_ps_%1x%2, 4, 7, 4, src, srcstride, dst, dststride
-%define coef2  m3
-%define Tm0    m2
-%define t1     m1
-%define t0     m0
-
-    dec        srcq
-    mov        r4d, r4m
-    add        dststrided, dststrided
-
-%ifdef PIC
-    lea        r6, [tab_ChromaCoeff]
-    movd       coef2, [r6 + r4 * 4]
-%else
-    movd       coef2, [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    pshufd     coef2, coef2, 0
-    mova       t1, [pw_2000]
-    mova       Tm0, [tab_Tm]
-
-    mov        r4d, %2
-    cmp        r5m, byte 0
-    je         .loopH
-    sub        srcq, srcstrideq
-    add        r4d, 3
-
-.loopH:
-    movh       t0, [srcq]
-    pshufb     t0, t0, Tm0
-    pmaddubsw  t0, coef2
-    phaddw     t0, t0
-    psubw      t0, t1
-    movd       [dstq], t0
-
-    lea        srcq, [srcq + srcstrideq]
-    lea        dstq, [dstq + dststrideq]
-
-    dec        r4d
-    jnz        .loopH
-
-    RET
-%endmacro
-
-    FILTER_HORIZ_CHROMA_2xN 2, 4
-    FILTER_HORIZ_CHROMA_2xN 2, 8
-
-    FILTER_HORIZ_CHROMA_2xN 2, 16
-
-;-----------------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_ps_4x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;-----------------------------------------------------------------------------------------------------------------------------
-%macro FILTER_HORIZ_CHROMA_4xN 2
-INIT_XMM sse4
-cglobal interp_4tap_horiz_ps_%1x%2, 4, 7, 4, src, srcstride, dst, dststride
-%define coef2  m3
-%define Tm0    m2
-%define t1     m1
-%define t0     m0
-
-    dec        srcq
-    mov        r4d, r4m
-    add        dststrided, dststrided
-
-%ifdef PIC
-    lea        r6, [tab_ChromaCoeff]
-    movd       coef2, [r6 + r4 * 4]
-%else
-    movd       coef2, [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    pshufd     coef2, coef2, 0
-    mova       t1, [pw_2000]
-    mova       Tm0, [tab_Tm]
-
-    mov        r4d, %2
-    cmp        r5m, byte 0
-    je         .loopH
-    sub        srcq, srcstrideq
-    add        r4d, 3
-
-.loopH:
-    movh       t0, [srcq]
-    pshufb     t0, t0, Tm0
-    pmaddubsw  t0, coef2
-    phaddw     t0, t0
-    psubw      t0, t1
-    movlps     [dstq], t0
-
-    lea        srcq, [srcq + srcstrideq]
-    lea        dstq, [dstq + dststrideq]
-
-    dec        r4d
-    jnz        .loopH
-    RET
-%endmacro
-
-    FILTER_HORIZ_CHROMA_4xN 4, 2
-    FILTER_HORIZ_CHROMA_4xN 4, 4
-    FILTER_HORIZ_CHROMA_4xN 4, 8
-    FILTER_HORIZ_CHROMA_4xN 4, 16
-
-    FILTER_HORIZ_CHROMA_4xN 4, 32
-
-%macro PROCESS_CHROMA_W6 3
-    movu       %1, [srcq]
-    pshufb     %2, %1, Tm0
-    pmaddubsw  %2, coef2
-    pshufb     %1, %1, Tm1
-    pmaddubsw  %1, coef2
-    phaddw     %2, %1
-    psubw      %2, %3
-    movh       [dstq], %2
-    pshufd     %2, %2, 2
-    movd       [dstq + 8], %2
-%endmacro
-
-%macro PROCESS_CHROMA_W12 3
-    movu       %1, [srcq]
-    pshufb     %2, %1, Tm0
-    pmaddubsw  %2, coef2
-    pshufb     %1, %1, Tm1
-    pmaddubsw  %1, coef2
-    phaddw     %2, %1
-    psubw      %2, %3
-    movu       [dstq], %2
-    movu       %1, [srcq + 8]
-    pshufb     %1, %1, Tm0
-    pmaddubsw  %1, coef2
-    phaddw     %1, %1
-    psubw      %1, %3
-    movh       [dstq + 16], %1
-%endmacro
-
-;-----------------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_ps_6x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;-----------------------------------------------------------------------------------------------------------------------------
-%macro FILTER_HORIZ_CHROMA 2
-INIT_XMM sse4
-cglobal interp_4tap_horiz_ps_%1x%2, 4, 7, 6, src, srcstride, dst, dststride
-%define coef2    m5
-%define Tm0      m4
-%define Tm1      m3
-%define t2       m2
-%define t1       m1
-%define t0       m0
-
-    dec     srcq
-    mov     r4d, r4m
-    add     dststrided, dststrided
-
-%ifdef PIC
-    lea     r6, [tab_ChromaCoeff]
-    movd    coef2, [r6 + r4 * 4]
-%else
-    movd    coef2, [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    pshufd  coef2, coef2, 0
-    mova    t2, [pw_2000]
-    mova    Tm0, [tab_Tm]
-    mova    Tm1, [tab_Tm + 16]
-
-    mov     r4d, %2
-    cmp     r5m, byte 0
-    je      .loopH
-    sub     srcq, srcstrideq
-    add     r4d, 3
-
-.loopH:
-    PROCESS_CHROMA_W%1  t0, t1, t2
-    add     srcq, srcstrideq
-    add     dstq, dststrideq
-
-    dec     r4d
-    jnz     .loopH
-
-    RET
-%endmacro
-
-    FILTER_HORIZ_CHROMA 6, 8
-    FILTER_HORIZ_CHROMA 12, 16
-
-    FILTER_HORIZ_CHROMA 6, 16
-    FILTER_HORIZ_CHROMA 12, 32
-
-%macro PROCESS_CHROMA_W8 3
-    movu        %1, [srcq]
-    pshufb      %2, %1, Tm0
-    pmaddubsw   %2, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %2, %1
-    psubw       %2, %3
-    movu        [dstq], %2
-%endmacro
-
-;-----------------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_ps_8x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;-----------------------------------------------------------------------------------------------------------------------------
-%macro FILTER_HORIZ_CHROMA_8xN 2
-INIT_XMM sse4
-cglobal interp_4tap_horiz_ps_%1x%2, 4, 7, 6, src, srcstride, dst, dststride
-%define coef2    m5
-%define Tm0      m4
-%define Tm1      m3
-%define t2       m2
-%define t1       m1
-%define t0       m0
-
-    dec     srcq
-    mov     r4d, r4m
-    add     dststrided, dststrided
-
-%ifdef PIC
-    lea     r6, [tab_ChromaCoeff]
-    movd    coef2, [r6 + r4 * 4]
-%else
-    movd    coef2, [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    pshufd  coef2, coef2, 0
-    mova    t2, [pw_2000]
-    mova    Tm0, [tab_Tm]
-    mova    Tm1, [tab_Tm + 16]
-
-    mov     r4d, %2
-    cmp     r5m, byte 0
-    je      .loopH
-    sub     srcq, srcstrideq
-    add     r4d, 3
-
-.loopH:
-    PROCESS_CHROMA_W8  t0, t1, t2
-    add     srcq, srcstrideq
-    add     dstq, dststrideq
-
-    dec     r4d
-    jnz     .loopH
-
-    RET
-%endmacro
-
-    FILTER_HORIZ_CHROMA_8xN 8, 2
-    FILTER_HORIZ_CHROMA_8xN 8, 4
-    FILTER_HORIZ_CHROMA_8xN 8, 6
-    FILTER_HORIZ_CHROMA_8xN 8, 8
-    FILTER_HORIZ_CHROMA_8xN 8, 16
-    FILTER_HORIZ_CHROMA_8xN 8, 32
-
-    FILTER_HORIZ_CHROMA_8xN 8, 12
-    FILTER_HORIZ_CHROMA_8xN 8, 64
-
-%macro PROCESS_CHROMA_W16 4
-    movu        %1, [srcq]
-    pshufb      %2, %1, Tm0
-    pmaddubsw   %2, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %2, %1
-    movu        %1, [srcq + 8]
-    pshufb      %4, %1, Tm0
-    pmaddubsw   %4, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %4, %1
-    psubw       %2, %3
-    psubw       %4, %3
-    movu        [dstq], %2
-    movu        [dstq + 16], %4
-%endmacro
-
-%macro PROCESS_CHROMA_W24 4
-    movu        %1, [srcq]
-    pshufb      %2, %1, Tm0
-    pmaddubsw   %2, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %2, %1
-    movu        %1, [srcq + 8]
-    pshufb      %4, %1, Tm0
-    pmaddubsw   %4, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %4, %1
-    psubw       %2, %3
-    psubw       %4, %3
-    movu        [dstq], %2
-    movu        [dstq + 16], %4
-    movu        %1, [srcq + 16]
-    pshufb      %2, %1, Tm0
-    pmaddubsw   %2, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %2, %1
-    psubw       %2, %3
-    movu        [dstq + 32], %2
-%endmacro
-
-%macro PROCESS_CHROMA_W32 4
-    movu        %1, [srcq]
-    pshufb      %2, %1, Tm0
-    pmaddubsw   %2, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %2, %1
-    movu        %1, [srcq + 8]
-    pshufb      %4, %1, Tm0
-    pmaddubsw   %4, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %4, %1
-    psubw       %2, %3
-    psubw       %4, %3
-    movu        [dstq], %2
-    movu        [dstq + 16], %4
-    movu        %1, [srcq + 16]
-    pshufb      %2, %1, Tm0
-    pmaddubsw   %2, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %2, %1
-    movu        %1, [srcq + 24]
-    pshufb      %4, %1, Tm0
-    pmaddubsw   %4, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %4, %1
-    psubw       %2, %3
-    psubw       %4, %3
-    movu        [dstq + 32], %2
-    movu        [dstq + 48], %4
-%endmacro
-
-%macro PROCESS_CHROMA_W16o 5
-    movu        %1, [srcq + %5]
-    pshufb      %2, %1, Tm0
-    pmaddubsw   %2, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %2, %1
-    movu        %1, [srcq + %5 + 8]
-    pshufb      %4, %1, Tm0
-    pmaddubsw   %4, coef2
-    pshufb      %1, %1, Tm1
-    pmaddubsw   %1, coef2
-    phaddw      %4, %1
-    psubw       %2, %3
-    psubw       %4, %3
-    movu        [dstq + %5 * 2], %2
-    movu        [dstq + %5 * 2 + 16], %4
-%endmacro
-
-%macro PROCESS_CHROMA_W48 4
-    PROCESS_CHROMA_W16o %1, %2, %3, %4, 0
-    PROCESS_CHROMA_W16o %1, %2, %3, %4, 16
-    PROCESS_CHROMA_W16o %1, %2, %3, %4, 32
-%endmacro
-
-%macro PROCESS_CHROMA_W64 4
-    PROCESS_CHROMA_W16o %1, %2, %3, %4, 0
-    PROCESS_CHROMA_W16o %1, %2, %3, %4, 16
-    PROCESS_CHROMA_W16o %1, %2, %3, %4, 32
-    PROCESS_CHROMA_W16o %1, %2, %3, %4, 48
-%endmacro
-
-;------------------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_ps_%1x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;------------------------------------------------------------------------------------------------------------------------------
-%macro FILTER_HORIZ_CHROMA_WxN 2
-INIT_XMM sse4
-cglobal interp_4tap_horiz_ps_%1x%2, 4, 7, 7, src, srcstride, dst, dststride
-%define coef2    m6
-%define Tm0      m5
-%define Tm1      m4
-%define t3       m3
-%define t2       m2
-%define t1       m1
-%define t0       m0
-
-    dec     srcq
-    mov     r4d, r4m
-    add     dststrided, dststrided
-
-%ifdef PIC
-    lea     r6, [tab_ChromaCoeff]
-    movd    coef2, [r6 + r4 * 4]
-%else
-    movd    coef2, [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    pshufd  coef2, coef2, 0
-    mova    t2, [pw_2000]
-    mova    Tm0, [tab_Tm]
-    mova    Tm1, [tab_Tm + 16]
-
-    mov     r4d, %2
-    cmp     r5m, byte 0
-    je      .loopH
-    sub     srcq, srcstrideq
-    add     r4d, 3
-
-.loopH:
-    PROCESS_CHROMA_W%1   t0, t1, t2, t3
-    add     srcq, srcstrideq
-    add     dstq, dststrideq
-
-    dec     r4d
-    jnz     .loopH
-
-    RET
-%endmacro
-
-    FILTER_HORIZ_CHROMA_WxN 16, 4
-    FILTER_HORIZ_CHROMA_WxN 16, 8
-    FILTER_HORIZ_CHROMA_WxN 16, 12
-    FILTER_HORIZ_CHROMA_WxN 16, 16
-    FILTER_HORIZ_CHROMA_WxN 16, 32
-    FILTER_HORIZ_CHROMA_WxN 24, 32
-    FILTER_HORIZ_CHROMA_WxN 32,  8
-    FILTER_HORIZ_CHROMA_WxN 32, 16
-    FILTER_HORIZ_CHROMA_WxN 32, 24
-    FILTER_HORIZ_CHROMA_WxN 32, 32
-
-    FILTER_HORIZ_CHROMA_WxN 16, 24
-    FILTER_HORIZ_CHROMA_WxN 16, 64
-    FILTER_HORIZ_CHROMA_WxN 24, 64
-    FILTER_HORIZ_CHROMA_WxN 32, 48
-    FILTER_HORIZ_CHROMA_WxN 32, 64
-
-    FILTER_HORIZ_CHROMA_WxN 64, 64
-    FILTER_HORIZ_CHROMA_WxN 64, 32
-    FILTER_HORIZ_CHROMA_WxN 64, 48
-    FILTER_HORIZ_CHROMA_WxN 48, 64
-    FILTER_HORIZ_CHROMA_WxN 64, 16
-
-
-;---------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ps_%1x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;---------------------------------------------------------------------------------------------------------------
-%macro FILTER_V_PS_W16n 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_ps_%1x%2, 4, 7, 8
-
-    mov        r4d, r4m
-    sub        r0, r1
-    add        r3d, r3d
-
-%ifdef PIC
-    lea        r5, [tab_ChromaCoeff]
-    movd       m0, [r5 + r4 * 4]
-%else
-    movd       m0, [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    pshufb     m1, m0, [tab_Vm]
-    pshufb     m0, [tab_Vm + 16]
-    mov        r4d, %2/2
-
-.loop:
-
-    mov         r6d,       %1/16
-
-.loopW:
-
-    movu       m2, [r0]
-    movu       m3, [r0 + r1]
-
-    punpcklbw  m4, m2, m3
-    punpckhbw  m2, m3
-
-    pmaddubsw  m4, m1
-    pmaddubsw  m2, m1
-
-    lea        r5, [r0 + 2 * r1]
-    movu       m5, [r5]
-    movu       m7, [r5 + r1]
-
-    punpcklbw  m6, m5, m7
-    pmaddubsw  m6, m0
-    paddw      m4, m6
-
-    punpckhbw  m6, m5, m7
-    pmaddubsw  m6, m0
-    paddw      m2, m6
-
-    mova       m6, [pw_2000]
-
-    psubw      m4, m6
-    psubw      m2, m6
-
-    movu       [r2], m4
-    movu       [r2 + 16], m2
-
-    punpcklbw  m4, m3, m5
-    punpckhbw  m3, m5
-
-    pmaddubsw  m4, m1
-    pmaddubsw  m3, m1
-
-    movu       m5, [r5 + 2 * r1]
-
-    punpcklbw  m2, m7, m5
-    punpckhbw  m7, m5
-
-    pmaddubsw  m2, m0
-    pmaddubsw  m7, m0
-
-    paddw      m4, m2
-    paddw      m3, m7
-
-    psubw      m4, m6
-    psubw      m3, m6
-
-    movu       [r2 + r3], m4
-    movu       [r2 + r3 + 16], m3
-
-    add         r0,        16
-    add         r2,        32
-    dec         r6d
-    jnz         .loopW
-
-    lea         r0,        [r0 + r1 * 2 - %1]
-    lea         r2,        [r2 + r3 * 2 - %1 * 2]
-
-    dec        r4d
-    jnz        .loop
-    RET
-%endmacro
-
-    FILTER_V_PS_W16n 64, 64
-    FILTER_V_PS_W16n 64, 32
-    FILTER_V_PS_W16n 64, 48
-    FILTER_V_PS_W16n 48, 64
-    FILTER_V_PS_W16n 64, 16
-
-
-;------------------------------------------------------------------------------------------------------------
-;void interp_4tap_vert_ps_2x4(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;------------------------------------------------------------------------------------------------------------
-INIT_XMM sse4
-cglobal interp_4tap_vert_ps_2x4, 4, 6, 7
-
-    mov         r4d, r4m
-    sub         r0, r1
-    add         r3d, r3d
-
-%ifdef PIC
-    lea         r5, [tab_ChromaCoeff]
-    movd        m0, [r5 + r4 * 4]
-%else
-    movd        m0, [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    pshufb      m0, [tab_Cm]
-
-    lea         r5, [3 * r1]
-
-    movd        m2, [r0]
-    movd        m3, [r0 + r1]
-    movd        m4, [r0 + 2 * r1]
-    movd        m5, [r0 + r5]
-
-    punpcklbw   m2, m3
-    punpcklbw   m6, m4, m5
-    punpcklbw   m2, m6
-
-    pmaddubsw   m2, m0
-
-    lea         r0, [r0 + 4 * r1]
-    movd        m6, [r0]
-
-    punpcklbw   m3, m4
-    punpcklbw   m1, m5, m6
-    punpcklbw   m3, m1
-
-    pmaddubsw   m3, m0
-    phaddw      m2, m3
-
-    mova        m1, [pw_2000]
-
-    psubw       m2, m1
-
-    movd        [r2], m2
-    pextrd      [r2 + r3], m2, 2
-
-    movd        m2, [r0 + r1]
-
-    punpcklbw   m4, m5
-    punpcklbw   m3, m6, m2
-    punpcklbw   m4, m3
-
-    pmaddubsw   m4, m0
-
-    movd        m3, [r0 + 2 * r1]
-
-    punpcklbw   m5, m6
-    punpcklbw   m2, m3
-    punpcklbw   m5, m2
-
-    pmaddubsw   m5, m0
-    phaddw      m4, m5
-    psubw       m4, m1
-
-    lea         r2, [r2 + 2 * r3]
-    movd        [r2], m4
-    pextrd      [r2 + r3], m4, 2
-
-    RET
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ps_2x8(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-%macro FILTER_V_PS_W2 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_ps_2x%2, 4, 6, 8
-
-    mov        r4d, r4m
-    sub        r0, r1
-    add        r3d, r3d
-
-%ifdef PIC
-    lea        r5, [tab_ChromaCoeff]
-    movd       m0, [r5 + r4 * 4]
-%else
-    movd       m0, [tab_ChromaCoeff + r4 * 4]
+    vextracti128    xm3, m1, 1
+    lea             r8, [r2 + r3 * 4]
+    movu            [r8], xm1
+    movu            [r8 + r3], xm3
+    vextracti128    xm3, m4, 1
+    movu            [r8 + r3 * 2], xm4
+    movu            [r8 + r6], xm3
+    vextracti128    xm3, m0, 1
+    lea             r8, [r8 + r3 * 4]
+    movu            [r8], xm0
+    movu            [r8 + r3], xm3
+    vextracti128    xm3, m6, 1
+    movu            [r8 + r3 * 2], xm6
+    movu            [r8 + r6], xm3
+    vextracti128    xm3, m7, 1
+    lea             r8, [r8 + r3 * 4]
+    movu            [r8], xm7
+    movu            [r8 + r3], xm3
+    vextracti128    xm3, m8, 1
+    movu            [r8 + r3 * 2], xm8
+    movu            [r8 + r6], xm3
 %endif
-
-    pshufb     m0, [tab_Cm]
-
-    mova       m1, [pw_2000]
-    lea        r5, [3 * r1]
-    mov        r4d, %2/4
-.loop:
-    movd       m2, [r0]
-    movd       m3, [r0 + r1]
-    movd       m4, [r0 + 2 * r1]
-    movd       m5, [r0 + r5]
-
-    punpcklbw  m2, m3
-    punpcklbw  m6, m4, m5
-    punpcklbw  m2, m6
-
-    pmaddubsw  m2, m0
-
-    lea        r0, [r0 + 4 * r1]
-    movd       m6, [r0]
-
-    punpcklbw  m3, m4
-    punpcklbw  m7, m5, m6
-    punpcklbw  m3, m7
-
-    pmaddubsw  m3, m0
-
-    phaddw     m2, m3
-    psubw      m2, m1
-
-
-    movd       [r2], m2
-    pshufd     m2, m2, 2
-    movd       [r2 + r3], m2
-
-    movd       m2, [r0 + r1]
-
-    punpcklbw  m4, m5
-    punpcklbw  m3, m6, m2
-    punpcklbw  m4, m3
-
-    pmaddubsw  m4, m0
-
-    movd       m3, [r0 + 2 * r1]
-
-    punpcklbw  m5, m6
-    punpcklbw  m2, m3
-    punpcklbw  m5, m2
-
-    pmaddubsw  m5, m0
-
-    phaddw     m4, m5
-
-    psubw      m4, m1
-
-    lea        r2, [r2 + 2 * r3]
-    movd       [r2], m4
-    pshufd     m4 , m4 ,2
-    movd       [r2 + r3], m4
-
-    lea        r2, [r2 + 2 * r3]
-
-    dec        r4d
-    jnz        .loop
-
-    RET
-%endmacro
-
-    FILTER_V_PS_W2 2, 8
-
-    FILTER_V_PS_W2 2, 16
-
-;-----------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ss_%1x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_SS 2
-INIT_XMM sse2
-cglobal interp_4tap_vert_ss_%1x%2, 5, 7, 6 ,0-gprsize
-
-    add       r1d, r1d
-    add       r3d, r3d
-    sub       r0, r1
-    shl       r4d, 5
-
-%ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
-    lea       r6, [r5 + r4]
+    sub             r7, r10
+    lea             r0, [r7 - 16]
+%ifidn %1,pp
+    lea             r2, [r8 + r3 * 4 - 16]
 %else
-    lea       r6, [tab_ChromaCoeffV + r4]
+    lea             r2, [r8 + r3 * 4 - 32]
 %endif
-
-    mov       dword [rsp], %2/4
-
-.loopH:
-    mov       r4d, (%1/4)
-.loopW:
-    PROCESS_CHROMA_SP_W4_4R
-
-    psrad     m0, 6
-    psrad     m1, 6
-    psrad     m2, 6
-    psrad     m3, 6
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-
-    movlps    [r2], m0
-    movhps    [r2 + r3], m0
-    lea       r5, [r2 + 2 * r3]
-    movlps    [r5], m2
-    movhps    [r5 + r3], m2
-
-    lea       r5, [4 * r1 - 2 * 4]
-    sub       r0, r5
-    add       r2, 2 * 4
-
-    dec       r4d
-    jnz       .loopW
-
-    lea       r0, [r0 + 4 * r1 - 2 * %1]
-    lea       r2, [r2 + 4 * r3 - 2 * %1]
-
-    dec       dword [rsp]
-    jnz       .loopH
-
+    dec             r9d
+    jnz             .loopH
     RET
+%endif
 %endmacro
 
-    FILTER_VER_CHROMA_SS 4, 4
-    FILTER_VER_CHROMA_SS 4, 8
-    FILTER_VER_CHROMA_SS 16, 16
-    FILTER_VER_CHROMA_SS 16, 8
-    FILTER_VER_CHROMA_SS 16, 12
-    FILTER_VER_CHROMA_SS 12, 16
-    FILTER_VER_CHROMA_SS 16, 4
-    FILTER_VER_CHROMA_SS 4, 16
-    FILTER_VER_CHROMA_SS 32, 32
-    FILTER_VER_CHROMA_SS 32, 16
-    FILTER_VER_CHROMA_SS 16, 32
-    FILTER_VER_CHROMA_SS 32, 24
-    FILTER_VER_CHROMA_SS 24, 32
-    FILTER_VER_CHROMA_SS 32, 8
-
-    FILTER_VER_CHROMA_SS 16, 24
-    FILTER_VER_CHROMA_SS 12, 32
-    FILTER_VER_CHROMA_SS 4, 32
-    FILTER_VER_CHROMA_SS 32, 64
-    FILTER_VER_CHROMA_SS 16, 64
-    FILTER_VER_CHROMA_SS 32, 48
-    FILTER_VER_CHROMA_SS 24, 64
-
-    FILTER_VER_CHROMA_SS 64, 64
-    FILTER_VER_CHROMA_SS 64, 32
-    FILTER_VER_CHROMA_SS 64, 48
-    FILTER_VER_CHROMA_SS 48, 64
-    FILTER_VER_CHROMA_SS 64, 16
+    FILTER_VER_LUMA_AVX2_24x32 pp
+    FILTER_VER_LUMA_AVX2_24x32 ps
 
-%macro FILTER_VER_CHROMA_S_AVX2_4x4 1
+%macro FILTER_VER_LUMA_AVX2_32xN 3
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_4x4, 4, 6, 7
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%3_%1x%2, 4, 12, 15
     mov             r4d, r4m
-    add             r1d, r1d
-    shl             r4d, 6
-    sub             r0, r1
-
+    shl             r4d, 7
 %ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
+    lea             r5, [tab_LumaCoeffVer_32]
     add             r5, r4
 %else
-    lea             r5, [pw_ChromaCoeffV + r4]
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
-
     lea             r4, [r1 * 3]
-%ifidn %1,sp
-    mova            m6, [pd_526336]
-%else
+    sub             r0, r4
+%ifidn %3,ps
     add             r3d, r3d
+    vbroadcasti128  m14, [pw_2000]
+%else
+    mova            m14, [pw_512]
 %endif
-
-    movq            xm0, [r0]
-    movq            xm1, [r0 + r1]
-    punpcklwd       xm0, xm1
-    movq            xm2, [r0 + r1 * 2]
-    punpcklwd       xm1, xm2
-    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
-    pmaddwd         m0, [r5]
-    movq            xm3, [r0 + r4]
-    punpcklwd       xm2, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm4, [r0]
-    punpcklwd       xm3, xm4
-    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
-    pmaddwd         m5, m2, [r5 + 1 * mmsize]
-    pmaddwd         m2, [r5]
-    paddd           m0, m5
-    movq            xm3, [r0 + r1]
-    punpcklwd       xm4, xm3
-    movq            xm1, [r0 + r1 * 2]
-    punpcklwd       xm3, xm1
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
-    pmaddwd         m4, [r5 + 1 * mmsize]
-    paddd           m2, m4
-
-%ifidn %1,sp
-    paddd           m0, m6
-    paddd           m2, m6
-    psrad           m0, 12
-    psrad           m2, 12
+    lea             r6, [r3 * 3]
+    lea             r11, [r1 * 4]
+    mov             r9d, %2 / 16
+.loopH:
+    mov             r10d, %1 / 16
+.loopW:
+    PROCESS_LUMA_AVX2_W16_16R %3
+%ifidn %3,pp
+    add             r2, 16
 %else
-    psrad           m0, 6
-    psrad           m2, 6
+    add             r2, 32
 %endif
-    packssdw        m0, m2
-    vextracti128    xm2, m0, 1
-    lea             r4, [r3 * 3]
-
-%ifidn %1,sp
-    packuswb        xm0, xm2
-    movd            [r2], xm0
-    pextrd          [r2 + r3], xm0, 2
-    pextrd          [r2 + r3 * 2], xm0, 1
-    pextrd          [r2 + r4], xm0, 3
+    add             r0, 16
+    dec             r10d
+    jnz             .loopW
+    sub             r7, r11
+    lea             r0, [r7 - 16]
+%ifidn %3,pp
+    lea             r2, [r8 + r3 * 4 - 16]
 %else
-    movq            [r2], xm0
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r4], xm2
+    lea             r2, [r8 + r3 * 4 - 32]
 %endif
+    dec             r9d
+    jnz             .loopH
     RET
+%endif
 %endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_4x4 sp
-    FILTER_VER_CHROMA_S_AVX2_4x4 ss
+    FILTER_VER_LUMA_AVX2_32xN 32, 32, pp
+    FILTER_VER_LUMA_AVX2_32xN 32, 64, pp
+    FILTER_VER_LUMA_AVX2_32xN 32, 32, ps
+    FILTER_VER_LUMA_AVX2_32xN 32, 64, ps
 
-%macro FILTER_VER_CHROMA_S_AVX2_4x8 1
+%macro FILTER_VER_LUMA_AVX2_32x16 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_4x8, 4, 6, 8
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_32x16, 4, 10, 15
     mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-    sub             r0, r1
+    shl             r4d, 7
 
 %ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
+    lea             r5, [tab_LumaCoeffVer_32]
     add             r5, r4
 %else
-    lea             r5, [pw_ChromaCoeffV + r4]
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
 
     lea             r4, [r1 * 3]
-%ifidn %1,sp
-    mova            m7, [pd_526336]
-%else
+    sub             r0, r4
+%ifidn %1,ps
     add             r3d, r3d
-%endif
-
-    movq            xm0, [r0]
-    movq            xm1, [r0 + r1]
-    punpcklwd       xm0, xm1
-    movq            xm2, [r0 + r1 * 2]
-    punpcklwd       xm1, xm2
-    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
-    pmaddwd         m0, [r5]
-    movq            xm3, [r0 + r4]
-    punpcklwd       xm2, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm4, [r0]
-    punpcklwd       xm3, xm4
-    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
-    pmaddwd         m5, m2, [r5 + 1 * mmsize]
-    pmaddwd         m2, [r5]
-    paddd           m0, m5
-    movq            xm3, [r0 + r1]
-    punpcklwd       xm4, xm3
-    movq            xm1, [r0 + r1 * 2]
-    punpcklwd       xm3, xm1
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
-    pmaddwd         m5, m4, [r5 + 1 * mmsize]
-    paddd           m2, m5
-    pmaddwd         m4, [r5]
-    movq            xm3, [r0 + r4]
-    punpcklwd       xm1, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm6, [r0]
-    punpcklwd       xm3, xm6
-    vinserti128     m1, m1, xm3, 1                  ; m1 = [8 7 7 6]
-    pmaddwd         m5, m1, [r5 + 1 * mmsize]
-    paddd           m4, m5
-    pmaddwd         m1, [r5]
-    movq            xm3, [r0 + r1]
-    punpcklwd       xm6, xm3
-    movq            xm5, [r0 + 2 * r1]
-    punpcklwd       xm3, xm5
-    vinserti128     m6, m6, xm3, 1                  ; m6 = [A 9 9 8]
-    pmaddwd         m6, [r5 + 1 * mmsize]
-    paddd           m1, m6
-    lea             r4, [r3 * 3]
-
-%ifidn %1,sp
-    paddd           m0, m7
-    paddd           m2, m7
-    paddd           m4, m7
-    paddd           m1, m7
-    psrad           m0, 12
-    psrad           m2, 12
-    psrad           m4, 12
-    psrad           m1, 12
+    vbroadcasti128  m14, [pw_2000]
 %else
-    psrad           m0, 6
-    psrad           m2, 6
-    psrad           m4, 6
-    psrad           m1, 6
+    mova            m14, [pw_512]
 %endif
-    packssdw        m0, m2
-    packssdw        m4, m1
-%ifidn %1,sp
-    packuswb        m0, m4
-    vextracti128    xm2, m0, 1
-    movd            [r2], xm0
-    movd            [r2 + r3], xm2
-    pextrd          [r2 + r3 * 2], xm0, 1
-    pextrd          [r2 + r4], xm2, 1
-    lea             r2, [r2 + r3 * 4]
-    pextrd          [r2], xm0, 2
-    pextrd          [r2 + r3], xm2, 2
-    pextrd          [r2 + r3 * 2], xm0, 3
-    pextrd          [r2 + r4], xm2, 3
+    lea             r6, [r3 * 3]
+    mov             r9d, 2
+.loopW:
+    PROCESS_LUMA_AVX2_W16_16R %1
+%ifidn %1,pp
+    add             r2, 16
 %else
-    vextracti128    xm2, m0, 1
-    vextracti128    xm1, m4, 1
-    movq            [r2], xm0
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r4], xm2
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm4
-    movq            [r2 + r3], xm1
-    movhps          [r2 + r3 * 2], xm4
-    movhps          [r2 + r4], xm1
+    add             r2, 32
 %endif
+    add             r0, 16
+    dec             r9d
+    jnz             .loopW
     RET
+%endif
 %endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_4x8 sp
-    FILTER_VER_CHROMA_S_AVX2_4x8 ss
-
-%macro PROCESS_CHROMA_AVX2_W4_16R 1
-    movq            xm0, [r0]
-    movq            xm1, [r0 + r1]
-    punpcklwd       xm0, xm1
-    movq            xm2, [r0 + r1 * 2]
-    punpcklwd       xm1, xm2
-    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
-    pmaddwd         m0, [r5]
-    movq            xm3, [r0 + r4]
-    punpcklwd       xm2, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm4, [r0]
-    punpcklwd       xm3, xm4
-    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
-    pmaddwd         m5, m2, [r5 + 1 * mmsize]
-    pmaddwd         m2, [r5]
-    paddd           m0, m5
-    movq            xm3, [r0 + r1]
-    punpcklwd       xm4, xm3
-    movq            xm1, [r0 + r1 * 2]
-    punpcklwd       xm3, xm1
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
-    pmaddwd         m5, m4, [r5 + 1 * mmsize]
-    paddd           m2, m5
-    pmaddwd         m4, [r5]
-    movq            xm3, [r0 + r4]
-    punpcklwd       xm1, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm6, [r0]
-    punpcklwd       xm3, xm6
-    vinserti128     m1, m1, xm3, 1                  ; m1 = [8 7 7 6]
-    pmaddwd         m5, m1, [r5 + 1 * mmsize]
-    paddd           m4, m5
-    pmaddwd         m1, [r5]
-    movq            xm3, [r0 + r1]
-    punpcklwd       xm6, xm3
-    movq            xm5, [r0 + 2 * r1]
-    punpcklwd       xm3, xm5
-    vinserti128     m6, m6, xm3, 1                  ; m6 = [10 9 9 8]
-    pmaddwd         m3, m6, [r5 + 1 * mmsize]
-    paddd           m1, m3
-    pmaddwd         m6, [r5]
+    FILTER_VER_LUMA_AVX2_32x16 pp
+    FILTER_VER_LUMA_AVX2_32x16 ps
 
-%ifidn %1,sp
-    paddd           m0, m7
-    paddd           m2, m7
-    paddd           m4, m7
-    paddd           m1, m7
-    psrad           m4, 12
-    psrad           m1, 12
-    psrad           m0, 12
-    psrad           m2, 12
+%macro FILTER_VER_LUMA_AVX2_32x24 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_32x24, 4, 10, 15
+    mov             r4d, r4m
+    shl             r4d, 7
+%ifdef PIC
+    lea             r5, [tab_LumaCoeffVer_32]
+    add             r5, r4
 %else
-    psrad           m0, 6
-    psrad           m2, 6
-    psrad           m4, 6
-    psrad           m1, 6
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
-    packssdw        m0, m2
-    packssdw        m4, m1
-%ifidn %1,sp
-    packuswb        m0, m4
-    vextracti128    xm4, m0, 1
-    movd            [r2], xm0
-    movd            [r2 + r3], xm4
-    pextrd          [r2 + r3 * 2], xm0, 1
-    pextrd          [r2 + r6], xm4, 1
-    lea             r2, [r2 + r3 * 4]
-    pextrd          [r2], xm0, 2
-    pextrd          [r2 + r3], xm4, 2
-    pextrd          [r2 + r3 * 2], xm0, 3
-    pextrd          [r2 + r6], xm4, 3
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+%ifidn %1,ps
+    add             r3d, r3d
+%endif
+    lea             r6, [r3 * 3]
+%ifidn %1,pp
+    mova            m14, [pw_512]
 %else
-    vextracti128    xm2, m0, 1
-    vextracti128    xm1, m4, 1
-    movq            [r2], xm0
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r6], xm2
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm4
-    movq            [r2 + r3], xm1
-    movhps          [r2 + r3 * 2], xm4
-    movhps          [r2 + r6], xm1
+    vbroadcasti128  m14, [pw_2000]
+%endif
+    mov             r9d, 2
+.loopW:
+    PROCESS_LUMA_AVX2_W16_16R %1
+%ifidn %1,pp
+    add             r2, 16
+%else
+    add             r2, 32
+%endif
+    add             r0, 16
+    dec             r9d
+    jnz             .loopW
+    lea             r9, [r1 * 4]
+    sub             r7, r9
+    lea             r0, [r7 - 16]
+%ifidn %1,pp
+    lea             r2, [r8 + r3 * 4 - 16]
+%else
+    lea             r2, [r8 + r3 * 4 - 32]
+%endif
+    mov             r9d, 2
+.loop:
+    PROCESS_LUMA_AVX2_W16_8R %1
+%ifidn %1,pp
+    add             r2, 16
+%else
+    add             r2, 32
+%endif
+    add             r0, 16
+    dec             r9d
+    jnz             .loop
+    RET
 %endif
+%endmacro
 
-    movq            xm2, [r0 + r4]
-    punpcklwd       xm5, xm2
-    lea             r0, [r0 + 4 * r1]
-    movq            xm0, [r0]
-    punpcklwd       xm2, xm0
-    vinserti128     m5, m5, xm2, 1                  ; m5 = [12 11 11 10]
-    pmaddwd         m2, m5, [r5 + 1 * mmsize]
-    paddd           m6, m2
-    pmaddwd         m5, [r5]
-    movq            xm2, [r0 + r1]
-    punpcklwd       xm0, xm2
-    movq            xm3, [r0 + 2 * r1]
-    punpcklwd       xm2, xm3
-    vinserti128     m0, m0, xm2, 1                  ; m0 = [14 13 13 12]
-    pmaddwd         m2, m0, [r5 + 1 * mmsize]
-    paddd           m5, m2
-    pmaddwd         m0, [r5]
-    movq            xm4, [r0 + r4]
-    punpcklwd       xm3, xm4
-    lea             r0, [r0 + 4 * r1]
-    movq            xm1, [r0]
-    punpcklwd       xm4, xm1
-    vinserti128     m3, m3, xm4, 1                  ; m3 = [16 15 15 14]
-    pmaddwd         m4, m3, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m3, [r5]
-    movq            xm4, [r0 + r1]
-    punpcklwd       xm1, xm4
-    movq            xm2, [r0 + 2 * r1]
-    punpcklwd       xm4, xm2
-    vinserti128     m1, m1, xm4, 1                  ; m1 = [18 17 17 16]
-    pmaddwd         m1, [r5 + 1 * mmsize]
-    paddd           m3, m1
+    FILTER_VER_LUMA_AVX2_32x24 pp
+    FILTER_VER_LUMA_AVX2_32x24 ps
 
-%ifidn %1,sp
-    paddd           m6, m7
-    paddd           m5, m7
-    paddd           m0, m7
-    paddd           m3, m7
-    psrad           m6, 12
-    psrad           m5, 12
-    psrad           m0, 12
-    psrad           m3, 12
+%macro FILTER_VER_LUMA_AVX2_32x8 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_32x8, 4, 10, 15
+    mov             r4d, r4m
+    shl             r4d, 7
+
+%ifdef PIC
+    lea             r5, [tab_LumaCoeffVer_32]
+    add             r5, r4
 %else
-    psrad           m6, 6
-    psrad           m5, 6
-    psrad           m0, 6
-    psrad           m3, 6
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
-    packssdw        m6, m5
-    packssdw        m0, m3
-    lea             r2, [r2 + r3 * 4]
 
-%ifidn %1,sp
-    packuswb        m6, m0
-    vextracti128    xm0, m6, 1
-    movd            [r2], xm6
-    movd            [r2 + r3], xm0
-    pextrd          [r2 + r3 * 2], xm6, 1
-    pextrd          [r2 + r6], xm0, 1
-    lea             r2, [r2 + r3 * 4]
-    pextrd          [r2], xm6, 2
-    pextrd          [r2 + r3], xm0, 2
-    pextrd          [r2 + r3 * 2], xm6, 3
-    pextrd          [r2 + r6], xm0, 3
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+%ifidn %1,ps
+    add             r3d, r3d
+%endif
+    lea             r6, [r3 * 3]
+%ifidn %1,pp
+    mova            m14, [pw_512]
 %else
-    vextracti128    xm5, m6, 1
-    vextracti128    xm3, m0, 1
-    movq            [r2], xm6
-    movq            [r2 + r3], xm5
-    movhps          [r2 + r3 * 2], xm6
-    movhps          [r2 + r6], xm5
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm0
-    movq            [r2 + r3], xm3
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r6], xm3
+    vbroadcasti128  m14, [pw_2000]
+%endif
+    mov             r9d, 2
+.loopW:
+    PROCESS_LUMA_AVX2_W16_8R %1
+%ifidn %1,pp
+    add             r2, 16
+%else
+    add             r2, 32
+%endif
+    add             r0, 16
+    dec             r9d
+    jnz             .loopW
+    RET
 %endif
 %endmacro
 
-%macro FILTER_VER_CHROMA_S_AVX2_4x16 1
+    FILTER_VER_LUMA_AVX2_32x8 pp
+    FILTER_VER_LUMA_AVX2_32x8 ps
+
+%macro FILTER_VER_LUMA_AVX2_48x64 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_4x16, 4, 7, 8
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_48x64, 4, 12, 15
     mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-    sub             r0, r1
+    shl             r4d, 7
 
 %ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
+    lea             r5, [tab_LumaCoeffVer_32]
     add             r5, r4
 %else
-    lea             r5, [pw_ChromaCoeffV + r4]
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
 
     lea             r4, [r1 * 3]
-%ifidn %1,sp
-    mova            m7, [pd_526336]
-%else
+    sub             r0, r4
+
+%ifidn %1,ps
     add             r3d, r3d
 %endif
+
     lea             r6, [r3 * 3]
-    PROCESS_CHROMA_AVX2_W4_16R %1
+    lea             r11, [r1 * 4]
+
+%ifidn %1,pp
+    mova            m14, [pw_512]
+%else
+    vbroadcasti128  m14, [pw_2000]
+%endif
+
+    mov             r9d, 4
+.loopH:
+    mov             r10d, 3
+.loopW:
+    PROCESS_LUMA_AVX2_W16_16R %1
+%ifidn %1,pp
+    add             r2, 16
+%else
+    add             r2, 32
+%endif
+    add             r0, 16
+    dec             r10d
+    jnz             .loopW
+    sub             r7, r11
+    lea             r0, [r7 - 32]
+%ifidn %1,pp
+    lea             r2, [r8 + r3 * 4 - 32]
+%else
+    lea             r2, [r8 + r3 * 4 - 64]
+%endif
+    dec             r9d
+    jnz             .loopH
     RET
+%endif
 %endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_4x16 sp
-    FILTER_VER_CHROMA_S_AVX2_4x16 ss
+    FILTER_VER_LUMA_AVX2_48x64 pp
+    FILTER_VER_LUMA_AVX2_48x64 ps
 
-%macro FILTER_VER_CHROMA_S_AVX2_4x32 1
+%macro FILTER_VER_LUMA_AVX2_64xN 3
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_4x32, 4, 7, 8
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%3_%1x%2, 4, 12, 15
     mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-    sub             r0, r1
+    shl             r4d, 7
 
 %ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
+    lea             r5, [tab_LumaCoeffVer_32]
     add             r5, r4
 %else
-    lea             r5, [pw_ChromaCoeffV + r4]
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
 
     lea             r4, [r1 * 3]
-%ifidn %1,sp
-    mova            m7, [pd_526336]
-%else
+    sub             r0, r4
+
+%ifidn %3,ps
     add             r3d, r3d
 %endif
+
     lea             r6, [r3 * 3]
-%rep 2
-    PROCESS_CHROMA_AVX2_W4_16R %1
-    lea             r2, [r2 + r3 * 4]
-%endrep
+    lea             r11, [r1 * 4]
+
+%ifidn %3,pp
+    mova            m14, [pw_512]
+%else
+    vbroadcasti128  m14, [pw_2000]
+%endif
+
+    mov             r9d, %2 / 16
+.loopH:
+    mov             r10d, %1 / 16
+.loopW:
+    PROCESS_LUMA_AVX2_W16_16R %3
+%ifidn %3,pp
+    add             r2, 16
+%else
+    add             r2, 32
+%endif
+    add             r0, 16
+    dec             r10d
+    jnz             .loopW
+    sub             r7, r11
+    lea             r0, [r7 - 48]
+%ifidn %3,pp
+    lea             r2, [r8 + r3 * 4 - 48]
+%else
+    lea             r2, [r8 + r3 * 4 - 96]
+%endif
+    dec             r9d
+    jnz             .loopH
     RET
+%endif
 %endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_4x32 sp
-    FILTER_VER_CHROMA_S_AVX2_4x32 ss
+    FILTER_VER_LUMA_AVX2_64xN 64, 32, pp
+    FILTER_VER_LUMA_AVX2_64xN 64, 48, pp
+    FILTER_VER_LUMA_AVX2_64xN 64, 64, pp
+    FILTER_VER_LUMA_AVX2_64xN 64, 32, ps
+    FILTER_VER_LUMA_AVX2_64xN 64, 48, ps
+    FILTER_VER_LUMA_AVX2_64xN 64, 64, ps
 
-%macro FILTER_VER_CHROMA_S_AVX2_4x2 1
+%macro FILTER_VER_LUMA_AVX2_64x16 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_4x2, 4, 6, 6
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_64x16, 4, 10, 15
     mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-    sub             r0, r1
+    shl             r4d, 7
 
 %ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
+    lea             r5, [tab_LumaCoeffVer_32]
     add             r5, r4
 %else
-    lea             r5, [pw_ChromaCoeffV + r4]
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
 
     lea             r4, [r1 * 3]
-%ifidn %1,sp
-    mova            m5, [pd_526336]
-%else
+    sub             r0, r4
+
+%ifidn %1,ps
     add             r3d, r3d
 %endif
-    movq            xm0, [r0]
-    movq            xm1, [r0 + r1]
-    punpcklwd       xm0, xm1
-    movq            xm2, [r0 + r1 * 2]
-    punpcklwd       xm1, xm2
-    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
-    pmaddwd         m0, [r5]
-    movq            xm3, [r0 + r4]
-    punpcklwd       xm2, xm3
-    movq            xm4, [r0 + 4 * r1]
-    punpcklwd       xm3, xm4
-    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
-    pmaddwd         m2, [r5 + 1 * mmsize]
-    paddd           m0, m2
-%ifidn %1,sp
-    paddd           m0, m5
-    psrad           m0, 12
+
+    lea             r6, [r3 * 3]
+
+%ifidn %1,pp
+    mova            m14, [pw_512]
 %else
-    psrad           m0, 6
+    vbroadcasti128  m14, [pw_2000]
 %endif
-    vextracti128    xm1, m0, 1
-    packssdw        xm0, xm1
-%ifidn %1,sp
-    packuswb        xm0, xm0
-    movd            [r2], xm0
-    pextrd          [r2 + r3], xm0, 1
+
+    mov             r9d, 4
+.loopW:
+    PROCESS_LUMA_AVX2_W16_16R %1
+%ifidn %1,pp
+    add             r2, 16
 %else
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
+    add             r2, 32
 %endif
+    add             r0, 16
+    dec             r9d
+    jnz             .loopW
     RET
+%endif
 %endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_4x2 sp
-    FILTER_VER_CHROMA_S_AVX2_4x2 ss
+    FILTER_VER_LUMA_AVX2_64x16 pp
+    FILTER_VER_LUMA_AVX2_64x16 ps
 
-%macro FILTER_VER_CHROMA_S_AVX2_2x4 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_2x4, 4, 6, 6
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-    sub             r0, r1
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_LUMA 3
+INIT_XMM sse4
+cglobal interp_8tap_vert_%3_%1x%2, 5, 7, 8 ,0-gprsize
+    lea       r5, [3 * r1]
+    sub       r0, r5
+    shl       r4d, 6
+%ifidn %3,ps
+    add       r3d, r3d
+%endif
 
 %ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
+    lea       r5, [tab_LumaCoeffVer]
+    lea       r6, [r5 + r4]
 %else
-    lea             r5, [pw_ChromaCoeffV + r4]
+    lea       r6, [tab_LumaCoeffVer + r4]
 %endif
 
-    lea             r4, [r1 * 3]
-%ifidn %1,sp
-    mova            m5, [pd_526336]
+%ifidn %3,pp
+    mova      m3, [pw_512]
 %else
-    add             r3d, r3d
+    mova      m3, [pw_2000]
 %endif
-    movd            xm0, [r0]
-    movd            xm1, [r0 + r1]
-    punpcklwd       xm0, xm1
-    movd            xm2, [r0 + r1 * 2]
-    punpcklwd       xm1, xm2
-    punpcklqdq      xm0, xm1                        ; m0 = [2 1 1 0]
-    movd            xm3, [r0 + r4]
-    punpcklwd       xm2, xm3
-    lea             r0, [r0 + 4 * r1]
-    movd            xm4, [r0]
-    punpcklwd       xm3, xm4
-    punpcklqdq      xm2, xm3                        ; m2 = [4 3 3 2]
-    vinserti128     m0, m0, xm2, 1                  ; m0 = [4 3 3 2 2 1 1 0]
-    movd            xm1, [r0 + r1]
-    punpcklwd       xm4, xm1
-    movd            xm3, [r0 + r1 * 2]
-    punpcklwd       xm1, xm3
-    punpcklqdq      xm4, xm1                        ; m4 = [6 5 5 4]
-    vinserti128     m2, m2, xm4, 1                  ; m2 = [6 5 5 4 4 3 3 2]
-    pmaddwd         m0, [r5]
-    pmaddwd         m2, [r5 + 1 * mmsize]
-    paddd           m0, m2
-%ifidn %1,sp
-    paddd           m0, m5
-    psrad           m0, 12
+    mov       dword [rsp], %2/4
+
+.loopH:
+    mov       r4d, (%1/8)
+.loopW:
+    PROCESS_LUMA_W8_4R
+%ifidn %3,pp
+    pmulhrsw  m7, m3
+    pmulhrsw  m6, m3
+    pmulhrsw  m5, m3
+    pmulhrsw  m4, m3
+
+    packuswb  m7, m6
+    packuswb  m5, m4
+
+    movlps    [r2], m7
+    movhps    [r2 + r3], m7
+    lea       r5, [r2 + 2 * r3]
+    movlps    [r5], m5
+    movhps    [r5 + r3], m5
 %else
-    psrad           m0, 6
+    psubw     m7, m3
+    psubw     m6, m3
+    psubw     m5, m3
+    psubw     m4, m3
+
+    movu      [r2], m7
+    movu      [r2 + r3], m6
+    lea       r5, [r2 + 2 * r3]
+    movu      [r5], m5
+    movu      [r5 + r3], m4
 %endif
-    vextracti128    xm1, m0, 1
-    packssdw        xm0, xm1
-    lea             r4, [r3 * 3]
-%ifidn %1,sp
-    packuswb        xm0, xm0
-    pextrw          [r2], xm0, 0
-    pextrw          [r2 + r3], xm0, 1
-    pextrw          [r2 + 2 * r3], xm0, 2
-    pextrw          [r2 + r4], xm0, 3
+
+    lea       r5, [8 * r1 - 8]
+    sub       r0, r5
+%ifidn %3,pp
+    add       r2, 8
 %else
-    movd            [r2], xm0
-    pextrd          [r2 + r3], xm0, 1
-    pextrd          [r2 + 2 * r3], xm0, 2
-    pextrd          [r2 + r4], xm0, 3
+    add       r2, 16
+%endif
+    dec       r4d
+    jnz       .loopW
+
+    lea       r0, [r0 + 4 * r1 - %1]
+%ifidn %3,pp
+    lea       r2, [r2 + 4 * r3 - %1]
+%else
+    lea       r2, [r2 + 4 * r3 - 2 * %1]
 %endif
+
+    dec       dword [rsp]
+    jnz       .loopH
+
     RET
 %endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_2x4 sp
-    FILTER_VER_CHROMA_S_AVX2_2x4 ss
+    FILTER_VER_LUMA 16, 4, pp
+    FILTER_VER_LUMA 16, 8, pp
+    FILTER_VER_LUMA 16, 12, pp
+    FILTER_VER_LUMA 16, 16, pp
+    FILTER_VER_LUMA 16, 32, pp
+    FILTER_VER_LUMA 16, 64, pp
+    FILTER_VER_LUMA 24, 32, pp
+    FILTER_VER_LUMA 32, 8, pp
+    FILTER_VER_LUMA 32, 16, pp
+    FILTER_VER_LUMA 32, 24, pp
+    FILTER_VER_LUMA 32, 32, pp
+    FILTER_VER_LUMA 32, 64, pp
+    FILTER_VER_LUMA 48, 64, pp
+    FILTER_VER_LUMA 64, 16, pp
+    FILTER_VER_LUMA 64, 32, pp
+    FILTER_VER_LUMA 64, 48, pp
+    FILTER_VER_LUMA 64, 64, pp
+
+    FILTER_VER_LUMA 16, 4, ps
+    FILTER_VER_LUMA 16, 8, ps
+    FILTER_VER_LUMA 16, 12, ps
+    FILTER_VER_LUMA 16, 16, ps
+    FILTER_VER_LUMA 16, 32, ps
+    FILTER_VER_LUMA 16, 64, ps
+    FILTER_VER_LUMA 24, 32, ps
+    FILTER_VER_LUMA 32, 8, ps
+    FILTER_VER_LUMA 32, 16, ps
+    FILTER_VER_LUMA 32, 24, ps
+    FILTER_VER_LUMA 32, 32, ps
+    FILTER_VER_LUMA 32, 64, ps
+    FILTER_VER_LUMA 48, 64, ps
+    FILTER_VER_LUMA 64, 16, ps
+    FILTER_VER_LUMA 64, 32, ps
+    FILTER_VER_LUMA 64, 48, ps
+    FILTER_VER_LUMA 64, 64, ps
+
+%macro PROCESS_LUMA_SP_W4_4R 0
+    movq       m0, [r0]
+    movq       m1, [r0 + r1]
+    punpcklwd  m0, m1                          ;m0=[0 1]
+    pmaddwd    m0, [r6 + 0 *16]                ;m0=[0+1]  Row1
+
+    lea        r0, [r0 + 2 * r1]
+    movq       m4, [r0]
+    punpcklwd  m1, m4                          ;m1=[1 2]
+    pmaddwd    m1, [r6 + 0 *16]                ;m1=[1+2]  Row2
+
+    movq       m5, [r0 + r1]
+    punpcklwd  m4, m5                          ;m4=[2 3]
+    pmaddwd    m2, m4, [r6 + 0 *16]            ;m2=[2+3]  Row3
+    pmaddwd    m4, [r6 + 1 * 16]
+    paddd      m0, m4                          ;m0=[0+1+2+3]  Row1
+
+    lea        r0, [r0 + 2 * r1]
+    movq       m4, [r0]
+    punpcklwd  m5, m4                          ;m5=[3 4]
+    pmaddwd    m3, m5, [r6 + 0 *16]            ;m3=[3+4]  Row4
+    pmaddwd    m5, [r6 + 1 * 16]
+    paddd      m1, m5                          ;m1 = [1+2+3+4]  Row2
+
+    movq       m5, [r0 + r1]
+    punpcklwd  m4, m5                          ;m4=[4 5]
+    pmaddwd    m6, m4, [r6 + 1 * 16]
+    paddd      m2, m6                          ;m2=[2+3+4+5]  Row3
+    pmaddwd    m4, [r6 + 2 * 16]
+    paddd      m0, m4                          ;m0=[0+1+2+3+4+5]  Row1
+
+    lea        r0, [r0 + 2 * r1]
+    movq       m4, [r0]
+    punpcklwd  m5, m4                          ;m5=[5 6]
+    pmaddwd    m6, m5, [r6 + 1 * 16]
+    paddd      m3, m6                          ;m3=[3+4+5+6]  Row4
+    pmaddwd    m5, [r6 + 2 * 16]
+    paddd      m1, m5                          ;m1=[1+2+3+4+5+6]  Row2
+
+    movq       m5, [r0 + r1]
+    punpcklwd  m4, m5                          ;m4=[6 7]
+    pmaddwd    m6, m4, [r6 + 2 * 16]
+    paddd      m2, m6                          ;m2=[2+3+4+5+6+7]  Row3
+    pmaddwd    m4, [r6 + 3 * 16]
+    paddd      m0, m4                          ;m0=[0+1+2+3+4+5+6+7]  Row1 end
+
+    lea        r0, [r0 + 2 * r1]
+    movq       m4, [r0]
+    punpcklwd  m5, m4                          ;m5=[7 8]
+    pmaddwd    m6, m5, [r6 + 2 * 16]
+    paddd      m3, m6                          ;m3=[3+4+5+6+7+8]  Row4
+    pmaddwd    m5, [r6 + 3 * 16]
+    paddd      m1, m5                          ;m1=[1+2+3+4+5+6+7+8]  Row2 end
+
+    movq       m5, [r0 + r1]
+    punpcklwd  m4, m5                          ;m4=[8 9]
+    pmaddwd    m4, [r6 + 3 * 16]
+    paddd      m2, m4                          ;m2=[2+3+4+5+6+7+8+9]  Row3 end
+
+    movq       m4, [r0 + 2 * r1]
+    punpcklwd  m5, m4                          ;m5=[9 10]
+    pmaddwd    m5, [r6 + 3 * 16]
+    paddd      m3, m5                          ;m3=[3+4+5+6+7+8+9+10]  Row4 end
+%endmacro
+
+;--------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_sp_%1x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_LUMA_SP 2
+INIT_XMM sse4
+cglobal interp_8tap_vert_sp_%1x%2, 5, 7, 8 ,0-gprsize
 
-%macro FILTER_VER_CHROMA_S_AVX2_8x8 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x8, 4, 6, 8
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
+    add       r1d, r1d
+    lea       r5, [r1 + 2 * r1]
+    sub       r0, r5
+    shl       r4d, 6
 
 %ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
+    lea       r5, [tab_LumaCoeffV]
+    lea       r6, [r5 + r4]
 %else
-    lea             r5, [pw_ChromaCoeffV + r4]
+    lea       r6, [tab_LumaCoeffV + r4]
 %endif
 
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,sp
-    mova            m7, [pd_526336]
-%else
-    add             r3d, r3d
-%endif
+    mova      m7, [pd_526336]
 
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    pmaddwd         m2, [r5]
-    paddd           m0, m4
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    pmaddwd         m3, [r5]
-    paddd           m1, m5
-%ifidn %1,sp
-    paddd           m0, m7
-    paddd           m1, m7
-    psrad           m0, 12
-    psrad           m1, 12
-%else
-    psrad           m0, 6
-    psrad           m1, 6
-%endif
-    packssdw        m0, m1
+    mov       dword [rsp], %2/4
+.loopH:
+    mov       r4d, (%1/4)
+.loopW:
+    PROCESS_LUMA_SP_W4_4R
 
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m2, m6
-    pmaddwd         m4, [r5]
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm1, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm1, 1
-    pmaddwd         m1, m5, [r5 + 1 * mmsize]
-    pmaddwd         m5, [r5]
-    paddd           m3, m1
-%ifidn %1,sp
-    paddd           m2, m7
-    paddd           m3, m7
-    psrad           m2, 12
-    psrad           m3, 12
-%else
-    psrad           m2, 6
-    psrad           m3, 6
-%endif
-    packssdw        m2, m3
+    paddd     m0, m7
+    paddd     m1, m7
+    paddd     m2, m7
+    paddd     m3, m7
 
-    movu            xm1, [r0 + r4]                  ; m1 = row 7
-    punpckhwd       xm3, xm6, xm1
-    punpcklwd       xm6, xm1
-    vinserti128     m6, m6, xm3, 1
-    pmaddwd         m3, m6, [r5 + 1 * mmsize]
-    pmaddwd         m6, [r5]
-    paddd           m4, m3
+    psrad     m0, 12
+    psrad     m1, 12
+    psrad     m2, 12
+    psrad     m3, 12
 
-    lea             r4, [r3 * 3]
-%ifidn %1,sp
-    packuswb        m0, m2
-    mova            m3, [interp8_hps_shuf]
-    vpermd          m0, m3, m0
-    vextracti128    xm2, m0, 1
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r4], xm2
-%else
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    movu            [r2], xm0
-    vextracti128    xm0, m0, 1
-    vextracti128    xm3, m2, 1
-    movu            [r2 + r3], xm0
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r4], xm3
-%endif
-    lea             r2, [r2 + r3 * 4]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm0, [r0]                       ; m0 = row 8
-    punpckhwd       xm2, xm1, xm0
-    punpcklwd       xm1, xm0
-    vinserti128     m1, m1, xm2, 1
-    pmaddwd         m2, m1, [r5 + 1 * mmsize]
-    pmaddwd         m1, [r5]
-    paddd           m5, m2
-%ifidn %1,sp
-    paddd           m4, m7
-    paddd           m5, m7
-    psrad           m4, 12
-    psrad           m5, 12
-%else
-    psrad           m4, 6
-    psrad           m5, 6
-%endif
-    packssdw        m4, m5
+    packssdw  m0, m1
+    packssdw  m2, m3
 
-    movu            xm2, [r0 + r1]                  ; m2 = row 9
-    punpckhwd       xm5, xm0, xm2
-    punpcklwd       xm0, xm2
-    vinserti128     m0, m0, xm5, 1
-    pmaddwd         m0, [r5 + 1 * mmsize]
-    paddd           m6, m0
-    movu            xm5, [r0 + r1 * 2]              ; m5 = row 10
-    punpckhwd       xm0, xm2, xm5
-    punpcklwd       xm2, xm5
-    vinserti128     m2, m2, xm0, 1
-    pmaddwd         m2, [r5 + 1 * mmsize]
-    paddd           m1, m2
+    packuswb  m0, m2
+
+    movd      [r2], m0
+    pextrd    [r2 + r3], m0, 1
+    lea       r5, [r2 + 2 * r3]
+    pextrd    [r5], m0, 2
+    pextrd    [r5 + r3], m0, 3
+
+    lea       r5, [8 * r1 - 2 * 4]
+    sub       r0, r5
+    add       r2, 4
+
+    dec       r4d
+    jnz       .loopW
+
+    lea       r0, [r0 + 4 * r1 - 2 * %1]
+    lea       r2, [r2 + 4 * r3 - %1]
+
+    dec       dword [rsp]
+    jnz       .loopH
+
+    RET
+%endmacro
+
+;--------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_sp_%1x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_SP 4, 4
+    FILTER_VER_LUMA_SP 8, 8
+    FILTER_VER_LUMA_SP 8, 4
+    FILTER_VER_LUMA_SP 4, 8
+    FILTER_VER_LUMA_SP 16, 16
+    FILTER_VER_LUMA_SP 16, 8
+    FILTER_VER_LUMA_SP 8, 16
+    FILTER_VER_LUMA_SP 16, 12
+    FILTER_VER_LUMA_SP 12, 16
+    FILTER_VER_LUMA_SP 16, 4
+    FILTER_VER_LUMA_SP 4, 16
+    FILTER_VER_LUMA_SP 32, 32
+    FILTER_VER_LUMA_SP 32, 16
+    FILTER_VER_LUMA_SP 16, 32
+    FILTER_VER_LUMA_SP 32, 24
+    FILTER_VER_LUMA_SP 24, 32
+    FILTER_VER_LUMA_SP 32, 8
+    FILTER_VER_LUMA_SP 8, 32
+    FILTER_VER_LUMA_SP 64, 64
+    FILTER_VER_LUMA_SP 64, 32
+    FILTER_VER_LUMA_SP 32, 64
+    FILTER_VER_LUMA_SP 64, 48
+    FILTER_VER_LUMA_SP 48, 64
+    FILTER_VER_LUMA_SP 64, 16
+    FILTER_VER_LUMA_SP 16, 64
+
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+INIT_XMM sse4
+cglobal filterPixelToShort_4x2, 3, 4, 3
+    mov         r3d, r3m
+    add         r3d, r3d
+
+    ; load constant
+    mova        m1, [pb_128]
+    mova        m2, [tab_c_64_n64]
+
+    movd        m0, [r0]
+    pinsrd      m0, [r0 + r1], 1
+    punpcklbw   m0, m1
+    pmaddubsw   m0, m2
+
+    movq        [r2 + r3 * 0], m0
+    movhps      [r2 + r3 * 1], m0
 
-%ifidn %1,sp
-    paddd           m6, m7
-    paddd           m1, m7
-    psrad           m6, 12
-    psrad           m1, 12
-%else
-    psrad           m6, 6
-    psrad           m1, 6
-%endif
-    packssdw        m6, m1
-%ifidn %1,sp
-    packuswb        m4, m6
-    vpermd          m4, m3, m4
-    vextracti128    xm6, m4, 1
-    movq            [r2], xm4
-    movhps          [r2 + r3], xm4
-    movq            [r2 + r3 * 2], xm6
-    movhps          [r2 + r4], xm6
-%else
-    vpermq          m4, m4, 11011000b
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm5, m4, 1
-    vextracti128    xm1, m6, 1
-    movu            [r2], xm4
-    movu            [r2 + r3], xm5
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r4], xm1
-%endif
     RET
-%endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_8x8 sp
-    FILTER_VER_CHROMA_S_AVX2_8x8 ss
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+INIT_XMM ssse3
+cglobal filterPixelToShort_8x2, 3, 4, 3
+    mov         r3d, r3m
+    add         r3d, r3d
 
-%macro PROCESS_CHROMA_S_AVX2_W8_16R 1
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m2, [r5]
-    lea             r7, [r0 + r1 * 4]
-    movu            xm4, [r7]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    paddd           m1, m5
-    pmaddwd         m3, [r5]
-    movu            xm5, [r7 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m2, m6
-    pmaddwd         m4, [r5]
-    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm7, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddwd         m7, m5, [r5 + 1 * mmsize]
-    paddd           m3, m7
-    pmaddwd         m5, [r5]
-%ifidn %1,sp
-    paddd           m0, m9
-    paddd           m1, m9
-    paddd           m2, m9
-    paddd           m3, m9
-    psrad           m0, 12
-    psrad           m1, 12
-    psrad           m2, 12
-    psrad           m3, 12
-%else
-    psrad           m0, 6
-    psrad           m1, 6
-    psrad           m2, 6
-    psrad           m3, 6
-%endif
-    packssdw        m0, m1
-    packssdw        m2, m3
-%ifidn %1,sp
-    packuswb        m0, m2
-    mova            m3, [interp8_hps_shuf]
-    vpermd          m0, m3, m0
-    vextracti128    xm2, m0, 1
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r6], xm2
-%else
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-%endif
+    ; load constant
+    mova        m1, [pb_128]
+    mova        m2, [tab_c_64_n64]
 
-    movu            xm7, [r7 + r4]                  ; m7 = row 7
-    punpckhwd       xm8, xm6, xm7
-    punpcklwd       xm6, xm7
-    vinserti128     m6, m6, xm8, 1
-    pmaddwd         m8, m6, [r5 + 1 * mmsize]
-    paddd           m4, m8
-    pmaddwd         m6, [r5]
-    lea             r7, [r7 + r1 * 4]
-    movu            xm8, [r7]                       ; m8 = row 8
-    punpckhwd       xm0, xm7, xm8
-    punpcklwd       xm7, xm8
-    vinserti128     m7, m7, xm0, 1
-    pmaddwd         m0, m7, [r5 + 1 * mmsize]
-    paddd           m5, m0
-    pmaddwd         m7, [r5]
-    movu            xm0, [r7 + r1]                  ; m0 = row 9
-    punpckhwd       xm1, xm8, xm0
-    punpcklwd       xm8, xm0
-    vinserti128     m8, m8, xm1, 1
-    pmaddwd         m1, m8, [r5 + 1 * mmsize]
-    paddd           m6, m1
-    pmaddwd         m8, [r5]
-    movu            xm1, [r7 + r1 * 2]              ; m1 = row 10
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m2, m0, [r5 + 1 * mmsize]
-    paddd           m7, m2
-    pmaddwd         m0, [r5]
-%ifidn %1,sp
-    paddd           m4, m9
-    paddd           m5, m9
-    psrad           m4, 12
-    psrad           m5, 12
-    paddd           m6, m9
-    paddd           m7, m9
-    psrad           m6, 12
-    psrad           m7, 12
-%else
-    psrad           m4, 6
-    psrad           m5, 6
-    psrad           m6, 6
-    psrad           m7, 6
-%endif
-    packssdw        m4, m5
-    packssdw        m6, m7
-    lea             r8, [r2 + r3 * 4]
-%ifidn %1,sp
-    packuswb        m4, m6
-    vpermd          m4, m3, m4
-    vextracti128    xm6, m4, 1
-    movq            [r8], xm4
-    movhps          [r8 + r3], xm4
-    movq            [r8 + r3 * 2], xm6
-    movhps          [r8 + r6], xm6
-%else
-    vpermq          m4, m4, 11011000b
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm5, m4, 1
-    vextracti128    xm7, m6, 1
-    movu            [r8], xm4
-    movu            [r8 + r3], xm5
-    movu            [r8 + r3 * 2], xm6
-    movu            [r8 + r6], xm7
-%endif
+    movh        m0, [r0]
+    punpcklbw   m0, m1
+    pmaddubsw   m0, m2
+    movu        [r2 + r3 * 0], m0
 
-    movu            xm2, [r7 + r4]                  ; m2 = row 11
-    punpckhwd       xm4, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm4, 1
-    pmaddwd         m4, m1, [r5 + 1 * mmsize]
-    paddd           m8, m4
-    pmaddwd         m1, [r5]
-    lea             r7, [r7 + r1 * 4]
-    movu            xm4, [r7]                       ; m4 = row 12
-    punpckhwd       xm5, xm2, xm4
-    punpcklwd       xm2, xm4
-    vinserti128     m2, m2, xm5, 1
-    pmaddwd         m5, m2, [r5 + 1 * mmsize]
-    paddd           m0, m5
-    pmaddwd         m2, [r5]
-    movu            xm5, [r7 + r1]                  ; m5 = row 13
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m1, m6
-    pmaddwd         m4, [r5]
-    movu            xm6, [r7 + r1 * 2]              ; m6 = row 14
-    punpckhwd       xm7, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddwd         m7, m5, [r5 + 1 * mmsize]
-    paddd           m2, m7
-    pmaddwd         m5, [r5]
-%ifidn %1,sp
-    paddd           m8, m9
-    paddd           m0, m9
-    paddd           m1, m9
-    paddd           m2, m9
-    psrad           m8, 12
-    psrad           m0, 12
-    psrad           m1, 12
-    psrad           m2, 12
-%else
-    psrad           m8, 6
-    psrad           m0, 6
-    psrad           m1, 6
-    psrad           m2, 6
-%endif
-    packssdw        m8, m0
-    packssdw        m1, m2
-    lea             r8, [r8 + r3 * 4]
-%ifidn %1,sp
-    packuswb        m8, m1
-    vpermd          m8, m3, m8
-    vextracti128    xm1, m8, 1
-    movq            [r8], xm8
-    movhps          [r8 + r3], xm8
-    movq            [r8 + r3 * 2], xm1
-    movhps          [r8 + r6], xm1
-%else
-    vpermq          m8, m8, 11011000b
-    vpermq          m1, m1, 11011000b
-    vextracti128    xm0, m8, 1
-    vextracti128    xm2, m1, 1
-    movu            [r8], xm8
-    movu            [r8 + r3], xm0
-    movu            [r8 + r3 * 2], xm1
-    movu            [r8 + r6], xm2
-%endif
-    lea             r8, [r8 + r3 * 4]
+    movh        m0, [r0 + r1]
+    punpcklbw   m0, m1
+    pmaddubsw   m0, m2
+    movu        [r2 + r3 * 1], m0
 
-    movu            xm7, [r7 + r4]                  ; m7 = row 15
-    punpckhwd       xm2, xm6, xm7
-    punpcklwd       xm6, xm7
-    vinserti128     m6, m6, xm2, 1
-    pmaddwd         m2, m6, [r5 + 1 * mmsize]
-    paddd           m4, m2
-    pmaddwd         m6, [r5]
-    lea             r7, [r7 + r1 * 4]
-    movu            xm2, [r7]                       ; m2 = row 16
-    punpckhwd       xm1, xm7, xm2
-    punpcklwd       xm7, xm2
-    vinserti128     m7, m7, xm1, 1
-    pmaddwd         m1, m7, [r5 + 1 * mmsize]
-    paddd           m5, m1
-    pmaddwd         m7, [r5]
-    movu            xm1, [r7 + r1]                  ; m1 = row 17
-    punpckhwd       xm0, xm2, xm1
-    punpcklwd       xm2, xm1
-    vinserti128     m2, m2, xm0, 1
-    pmaddwd         m2, [r5 + 1 * mmsize]
-    paddd           m6, m2
-    movu            xm0, [r7 + r1 * 2]              ; m0 = row 18
-    punpckhwd       xm2, xm1, xm0
-    punpcklwd       xm1, xm0
-    vinserti128     m1, m1, xm2, 1
-    pmaddwd         m1, [r5 + 1 * mmsize]
-    paddd           m7, m1
+    RET
 
-%ifidn %1,sp
-    paddd           m4, m9
-    paddd           m5, m9
-    paddd           m6, m9
-    paddd           m7, m9
-    psrad           m4, 12
-    psrad           m5, 12
-    psrad           m6, 12
-    psrad           m7, 12
+%macro PROCESS_CHROMA_SP_W4_4R 0
+    movq       m0, [r0]
+    movq       m1, [r0 + r1]
+    punpcklwd  m0, m1                          ;m0=[0 1]
+    pmaddwd    m0, [r6 + 0 *16]                ;m0=[0+1]         Row1
+
+    lea        r0, [r0 + 2 * r1]
+    movq       m4, [r0]
+    punpcklwd  m1, m4                          ;m1=[1 2]
+    pmaddwd    m1, [r6 + 0 *16]                ;m1=[1+2]         Row2
+
+    movq       m5, [r0 + r1]
+    punpcklwd  m4, m5                          ;m4=[2 3]
+    pmaddwd    m2, m4, [r6 + 0 *16]            ;m2=[2+3]         Row3
+    pmaddwd    m4, [r6 + 1 * 16]
+    paddd      m0, m4                          ;m0=[0+1+2+3]     Row1 done
+
+    lea        r0, [r0 + 2 * r1]
+    movq       m4, [r0]
+    punpcklwd  m5, m4                          ;m5=[3 4]
+    pmaddwd    m3, m5, [r6 + 0 *16]            ;m3=[3+4]         Row4
+    pmaddwd    m5, [r6 + 1 * 16]
+    paddd      m1, m5                          ;m1 = [1+2+3+4]   Row2
+
+    movq       m5, [r0 + r1]
+    punpcklwd  m4, m5                          ;m4=[4 5]
+    pmaddwd    m4, [r6 + 1 * 16]
+    paddd      m2, m4                          ;m2=[2+3+4+5]     Row3
+
+    movq       m4, [r0 + 2 * r1]
+    punpcklwd  m5, m4                          ;m5=[5 6]
+    pmaddwd    m5, [r6 + 1 * 16]
+    paddd      m3, m5                          ;m3=[3+4+5+6]     Row4
+%endmacro
+
+;--------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_sp_%1x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_SP 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_sp_%1x%2, 5, 7, 7 ,0-gprsize
+
+    add       r1d, r1d
+    sub       r0, r1
+    shl       r4d, 5
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r6, [r5 + r4]
 %else
-    psrad           m4, 6
-    psrad           m5, 6
-    psrad           m6, 6
-    psrad           m7, 6
+    lea       r6, [tab_ChromaCoeffV + r4]
 %endif
-    packssdw        m4, m5
-    packssdw        m6, m7
-%ifidn %1,sp
-    packuswb        m4, m6
-    vpermd          m4, m3, m4
-    vextracti128    xm6, m4, 1
-    movq            [r8], xm4
-    movhps          [r8 + r3], xm4
-    movq            [r8 + r3 * 2], xm6
-    movhps          [r8 + r6], xm6
+
+    mova      m6, [pd_526336]
+
+    mov       dword [rsp], %2/4
+
+.loopH:
+    mov       r4d, (%1/4)
+.loopW:
+    PROCESS_CHROMA_SP_W4_4R
+
+    paddd     m0, m6
+    paddd     m1, m6
+    paddd     m2, m6
+    paddd     m3, m6
+
+    psrad     m0, 12
+    psrad     m1, 12
+    psrad     m2, 12
+    psrad     m3, 12
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+
+    packuswb  m0, m2
+
+    movd      [r2], m0
+    pextrd    [r2 + r3], m0, 1
+    lea       r5, [r2 + 2 * r3]
+    pextrd    [r5], m0, 2
+    pextrd    [r5 + r3], m0, 3
+
+    lea       r5, [4 * r1 - 2 * 4]
+    sub       r0, r5
+    add       r2, 4
+
+    dec       r4d
+    jnz       .loopW
+
+    lea       r0, [r0 + 4 * r1 - 2 * %1]
+    lea       r2, [r2 + 4 * r3 - %1]
+
+    dec       dword [rsp]
+    jnz       .loopH
+
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_SP 4, 4
+    FILTER_VER_CHROMA_SP 4, 8
+    FILTER_VER_CHROMA_SP 16, 16
+    FILTER_VER_CHROMA_SP 16, 8
+    FILTER_VER_CHROMA_SP 16, 12
+    FILTER_VER_CHROMA_SP 12, 16
+    FILTER_VER_CHROMA_SP 16, 4
+    FILTER_VER_CHROMA_SP 4, 16
+    FILTER_VER_CHROMA_SP 32, 32
+    FILTER_VER_CHROMA_SP 32, 16
+    FILTER_VER_CHROMA_SP 16, 32
+    FILTER_VER_CHROMA_SP 32, 24
+    FILTER_VER_CHROMA_SP 24, 32
+    FILTER_VER_CHROMA_SP 32, 8
+
+    FILTER_VER_CHROMA_SP 16, 24
+    FILTER_VER_CHROMA_SP 16, 64
+    FILTER_VER_CHROMA_SP 12, 32
+    FILTER_VER_CHROMA_SP 4, 32
+    FILTER_VER_CHROMA_SP 32, 64
+    FILTER_VER_CHROMA_SP 32, 48
+    FILTER_VER_CHROMA_SP 24, 64
+
+    FILTER_VER_CHROMA_SP 64, 64
+    FILTER_VER_CHROMA_SP 64, 32
+    FILTER_VER_CHROMA_SP 64, 48
+    FILTER_VER_CHROMA_SP 48, 64
+    FILTER_VER_CHROMA_SP 64, 16
+
+
+%macro PROCESS_CHROMA_SP_W2_4R 1
+    movd       m0, [r0]
+    movd       m1, [r0 + r1]
+    punpcklwd  m0, m1                          ;m0=[0 1]
+
+    lea        r0, [r0 + 2 * r1]
+    movd       m2, [r0]
+    punpcklwd  m1, m2                          ;m1=[1 2]
+    punpcklqdq m0, m1                          ;m0=[0 1 1 2]
+    pmaddwd    m0, [%1 + 0 *16]                ;m0=[0+1 1+2] Row 1-2
+
+    movd       m1, [r0 + r1]
+    punpcklwd  m2, m1                          ;m2=[2 3]
+
+    lea        r0, [r0 + 2 * r1]
+    movd       m3, [r0]
+    punpcklwd  m1, m3                          ;m2=[3 4]
+    punpcklqdq m2, m1                          ;m2=[2 3 3 4]
+
+    pmaddwd    m4, m2, [%1 + 1 * 16]           ;m4=[2+3 3+4] Row 1-2
+    pmaddwd    m2, [%1 + 0 * 16]               ;m2=[2+3 3+4] Row 3-4
+    paddd      m0, m4                          ;m0=[0+1+2+3 1+2+3+4] Row 1-2
+
+    movd       m1, [r0 + r1]
+    punpcklwd  m3, m1                          ;m3=[4 5]
+
+    movd       m4, [r0 + 2 * r1]
+    punpcklwd  m1, m4                          ;m1=[5 6]
+    punpcklqdq m3, m1                          ;m2=[4 5 5 6]
+    pmaddwd    m3, [%1 + 1 * 16]               ;m3=[4+5 5+6] Row 3-4
+    paddd      m2, m3                          ;m2=[2+3+4+5 3+4+5+6] Row 3-4
+%endmacro
+
+;-------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vertical_sp_%1x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_SP_W2_4R 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_sp_%1x%2, 5, 6, 6
+
+    add       r1d, r1d
+    sub       r0, r1
+    shl       r4d, 5
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r5, [r5 + r4]
 %else
-    vpermq          m4, m4, 11011000b
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm5, m4, 1
-    vextracti128    xm7, m6, 1
-    movu            [r8], xm4
-    movu            [r8 + r3], xm5
-    movu            [r8 + r3 * 2], xm6
-    movu            [r8 + r6], xm7
+    lea       r5, [tab_ChromaCoeffV + r4]
 %endif
+
+    mova      m5, [pd_526336]
+
+    mov       r4d, (%2/4)
+
+.loopH:
+    PROCESS_CHROMA_SP_W2_4R r5
+
+    paddd     m0, m5
+    paddd     m2, m5
+
+    psrad     m0, 12
+    psrad     m2, 12
+
+    packssdw  m0, m2
+    packuswb  m0, m0
+
+    pextrw    [r2], m0, 0
+    pextrw    [r2 + r3], m0, 1
+    lea       r2, [r2 + 2 * r3]
+    pextrw    [r2], m0, 2
+    pextrw    [r2 + r3], m0, 3
+
+    lea       r2, [r2 + 2 * r3]
+
+    dec       r4d
+    jnz       .loopH
+
+    RET
 %endmacro
 
-%macro FILTER_VER_CHROMA_S_AVX2_Nx16 2
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_vert_%1_%2x16, 4, 10, 10
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
+    FILTER_VER_CHROMA_SP_W2_4R 2, 4
+    FILTER_VER_CHROMA_SP_W2_4R 2, 8
+
+    FILTER_VER_CHROMA_SP_W2_4R 2, 16
+
+;--------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_sp_4x2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+INIT_XMM sse4
+cglobal interp_4tap_vert_sp_4x2, 5, 6, 5
+
+    add        r1d, r1d
+    sub        r0, r1
+    shl        r4d, 5
 
 %ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
+    lea        r5, [tab_ChromaCoeffV]
+    lea        r5, [r5 + r4]
 %else
-    lea             r5, [pw_ChromaCoeffV + r4]
+    lea        r5, [tab_ChromaCoeffV + r4]
 %endif
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,sp
-    mova            m9, [pd_526336]
+
+    mova       m4, [pd_526336]
+
+    movq       m0, [r0]
+    movq       m1, [r0 + r1]
+    punpcklwd  m0, m1                          ;m0=[0 1]
+    pmaddwd    m0, [r5 + 0 *16]                ;m0=[0+1]  Row1
+
+    lea        r0, [r0 + 2 * r1]
+    movq       m2, [r0]
+    punpcklwd  m1, m2                          ;m1=[1 2]
+    pmaddwd    m1, [r5 + 0 *16]                ;m1=[1+2]  Row2
+
+    movq       m3, [r0 + r1]
+    punpcklwd  m2, m3                          ;m4=[2 3]
+    pmaddwd    m2, [r5 + 1 * 16]
+    paddd      m0, m2                          ;m0=[0+1+2+3]  Row1 done
+    paddd      m0, m4
+    psrad      m0, 12
+
+    movq       m2, [r0 + 2 * r1]
+    punpcklwd  m3, m2                          ;m5=[3 4]
+    pmaddwd    m3, [r5 + 1 * 16]
+    paddd      m1, m3                          ;m1 = [1+2+3+4]  Row2 done
+    paddd      m1, m4
+    psrad      m1, 12
+
+    packssdw   m0, m1
+    packuswb   m0, m0
+
+    movd       [r2], m0
+    pextrd     [r2 + r3], m0, 1
+
+    RET
+
+;-------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vertical_sp_6x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_SP_W6_H4 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_sp_6x%2, 5, 7, 7
+
+    add       r1d, r1d
+    sub       r0, r1
+    shl       r4d, 5
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r6, [r5 + r4]
 %else
-    add             r3d, r3d
+    lea       r6, [tab_ChromaCoeffV + r4]
 %endif
-    lea             r6, [r3 * 3]
-    mov             r9d, %2 / 8
-.loopW:
-    PROCESS_CHROMA_S_AVX2_W8_16R %1
-%ifidn %1,sp
-    add             r2, 8
+
+    mova      m6, [pd_526336]
+
+    mov       r4d, %2/4
+
+.loopH:
+    PROCESS_CHROMA_SP_W4_4R
+
+    paddd     m0, m6
+    paddd     m1, m6
+    paddd     m2, m6
+    paddd     m3, m6
+
+    psrad     m0, 12
+    psrad     m1, 12
+    psrad     m2, 12
+    psrad     m3, 12
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+
+    packuswb  m0, m2
+
+    movd      [r2], m0
+    pextrd    [r2 + r3], m0, 1
+    lea       r5, [r2 + 2 * r3]
+    pextrd    [r5], m0, 2
+    pextrd    [r5 + r3], m0, 3
+
+    lea       r5, [4 * r1 - 2 * 4]
+    sub       r0, r5
+    add       r2, 4
+
+    PROCESS_CHROMA_SP_W2_4R r6
+
+    paddd     m0, m6
+    paddd     m2, m6
+
+    psrad     m0, 12
+    psrad     m2, 12
+
+    packssdw  m0, m2
+    packuswb  m0, m0
+
+    pextrw    [r2], m0, 0
+    pextrw    [r2 + r3], m0, 1
+    lea       r2, [r2 + 2 * r3]
+    pextrw    [r2], m0, 2
+    pextrw    [r2 + r3], m0, 3
+
+    sub       r0, 2 * 4
+    lea       r2, [r2 + 2 * r3 - 4]
+
+    dec       r4d
+    jnz       .loopH
+
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_SP_W6_H4 6, 8
+
+    FILTER_VER_CHROMA_SP_W6_H4 6, 16
+
+%macro PROCESS_CHROMA_SP_W8_2R 0
+    movu       m1, [r0]
+    movu       m3, [r0 + r1]
+    punpcklwd  m0, m1, m3
+    pmaddwd    m0, [r5 + 0 * 16]                ;m0 = [0l+1l]  Row1l
+    punpckhwd  m1, m3
+    pmaddwd    m1, [r5 + 0 * 16]                ;m1 = [0h+1h]  Row1h
+
+    movu       m4, [r0 + 2 * r1]
+    punpcklwd  m2, m3, m4
+    pmaddwd    m2, [r5 + 0 * 16]                ;m2 = [1l+2l]  Row2l
+    punpckhwd  m3, m4
+    pmaddwd    m3, [r5 + 0 * 16]                ;m3 = [1h+2h]  Row2h
+
+    lea        r0, [r0 + 2 * r1]
+    movu       m5, [r0 + r1]
+    punpcklwd  m6, m4, m5
+    pmaddwd    m6, [r5 + 1 * 16]                ;m6 = [2l+3l]  Row1l
+    paddd      m0, m6                           ;m0 = [0l+1l+2l+3l]  Row1l sum
+    punpckhwd  m4, m5
+    pmaddwd    m4, [r5 + 1 * 16]                ;m6 = [2h+3h]  Row1h
+    paddd      m1, m4                           ;m1 = [0h+1h+2h+3h]  Row1h sum
+
+    movu       m4, [r0 + 2 * r1]
+    punpcklwd  m6, m5, m4
+    pmaddwd    m6, [r5 + 1 * 16]                ;m6 = [3l+4l]  Row2l
+    paddd      m2, m6                           ;m2 = [1l+2l+3l+4l]  Row2l sum
+    punpckhwd  m5, m4
+    pmaddwd    m5, [r5 + 1 * 16]                ;m1 = [3h+4h]  Row2h
+    paddd      m3, m5                           ;m3 = [1h+2h+3h+4h]  Row2h sum
+%endmacro
+
+;--------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_sp_8x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_SP_W8_H2 2
+INIT_XMM sse2
+cglobal interp_4tap_vert_sp_%1x%2, 5, 6, 8
+
+    add       r1d, r1d
+    sub       r0, r1
+    shl       r4d, 5
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r5, [r5 + r4]
 %else
-    add             r2, 16
+    lea       r5, [tab_ChromaCoeffV + r4]
 %endif
-    add             r0, 16
-    dec             r9d
-    jnz             .loopW
+
+    mova      m7, [pd_526336]
+
+    mov       r4d, %2/2
+.loopH:
+    PROCESS_CHROMA_SP_W8_2R
+
+    paddd     m0, m7
+    paddd     m1, m7
+    paddd     m2, m7
+    paddd     m3, m7
+
+    psrad     m0, 12
+    psrad     m1, 12
+    psrad     m2, 12
+    psrad     m3, 12
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+
+    packuswb  m0, m2
+
+    movlps    [r2], m0
+    movhps    [r2 + r3], m0
+
+    lea       r2, [r2 + 2 * r3]
+
+    dec r4d
+    jnz .loopH
+
     RET
-%endif
 %endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_Nx16 sp, 16
-    FILTER_VER_CHROMA_S_AVX2_Nx16 sp, 32
-    FILTER_VER_CHROMA_S_AVX2_Nx16 sp, 64
-    FILTER_VER_CHROMA_S_AVX2_Nx16 ss, 16
-    FILTER_VER_CHROMA_S_AVX2_Nx16 ss, 32
-    FILTER_VER_CHROMA_S_AVX2_Nx16 ss, 64
+    FILTER_VER_CHROMA_SP_W8_H2 8, 2
+    FILTER_VER_CHROMA_SP_W8_H2 8, 4
+    FILTER_VER_CHROMA_SP_W8_H2 8, 6
+    FILTER_VER_CHROMA_SP_W8_H2 8, 8
+    FILTER_VER_CHROMA_SP_W8_H2 8, 16
+    FILTER_VER_CHROMA_SP_W8_H2 8, 32
+
+    FILTER_VER_CHROMA_SP_W8_H2 8, 12
+    FILTER_VER_CHROMA_SP_W8_H2 8, 64
 
-%macro FILTER_VER_CHROMA_S_AVX2_NxN 3
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_vert_%3_%1x%2, 4, 11, 10
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
+
+;---------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ps_%1x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;---------------------------------------------------------------------------------------------------------------
+%macro FILTER_V_PS_W16n 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_ps_%1x%2, 4, 7, 8
+
+    mov        r4d, r4m
+    sub        r0, r1
+    add        r3d, r3d
 
 %ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
-%else
-    lea             r5, [pw_ChromaCoeffV + r4]
-%endif
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %3,sp
-    mova            m9, [pd_526336]
+    lea        r5, [tab_ChromaCoeff]
+    movd       m0, [r5 + r4 * 4]
 %else
-    add             r3d, r3d
+    movd       m0, [tab_ChromaCoeff + r4 * 4]
 %endif
-    lea             r6, [r3 * 3]
-    mov             r9d, %2 / 16
-.loopH:
-    mov             r10d, %1 / 8
+
+    pshufb     m1, m0, [tab_Vm]
+    pshufb     m0, [tab_Vm + 16]
+    mov        r4d, %2/2
+
+.loop:
+
+    mov         r6d,       %1/16
+
 .loopW:
-    PROCESS_CHROMA_S_AVX2_W8_16R %3
-%ifidn %3,sp
-    add             r2, 8
-%else
-    add             r2, 16
-%endif
-    add             r0, 16
-    dec             r10d
-    jnz             .loopW
-    lea             r0, [r7 - 2 * %1 + 16]
-%ifidn %3,sp
-    lea             r2, [r8 + r3 * 4 - %1 + 8]
-%else
-    lea             r2, [r8 + r3 * 4 - 2 * %1 + 16]
-%endif
-    dec             r9d
-    jnz             .loopH
-    RET
-%endif
-%endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_NxN 16, 32, sp
-    FILTER_VER_CHROMA_S_AVX2_NxN 24, 32, sp
-    FILTER_VER_CHROMA_S_AVX2_NxN 32, 32, sp
-    FILTER_VER_CHROMA_S_AVX2_NxN 16, 32, ss
-    FILTER_VER_CHROMA_S_AVX2_NxN 24, 32, ss
-    FILTER_VER_CHROMA_S_AVX2_NxN 32, 32, ss
-    FILTER_VER_CHROMA_S_AVX2_NxN 16, 64, sp
-    FILTER_VER_CHROMA_S_AVX2_NxN 24, 64, sp
-    FILTER_VER_CHROMA_S_AVX2_NxN 32, 64, sp
-    FILTER_VER_CHROMA_S_AVX2_NxN 32, 48, sp
-    FILTER_VER_CHROMA_S_AVX2_NxN 32, 48, ss
-    FILTER_VER_CHROMA_S_AVX2_NxN 16, 64, ss
-    FILTER_VER_CHROMA_S_AVX2_NxN 24, 64, ss
-    FILTER_VER_CHROMA_S_AVX2_NxN 32, 64, ss
-    FILTER_VER_CHROMA_S_AVX2_NxN 64, 64, sp
-    FILTER_VER_CHROMA_S_AVX2_NxN 64, 32, sp
-    FILTER_VER_CHROMA_S_AVX2_NxN 64, 48, sp
-    FILTER_VER_CHROMA_S_AVX2_NxN 48, 64, sp
-    FILTER_VER_CHROMA_S_AVX2_NxN 64, 64, ss
-    FILTER_VER_CHROMA_S_AVX2_NxN 64, 32, ss
-    FILTER_VER_CHROMA_S_AVX2_NxN 64, 48, ss
-    FILTER_VER_CHROMA_S_AVX2_NxN 48, 64, ss
+    movu       m2, [r0]
+    movu       m3, [r0 + r1]
 
-%macro PROCESS_CHROMA_S_AVX2_W8_4R 1
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    paddd           m1, m5
-    pmaddwd         m3, [r5]
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m4, [r5 + 1 * mmsize]
-    paddd           m2, m4
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm4, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm4, 1
-    pmaddwd         m5, [r5 + 1 * mmsize]
-    paddd           m3, m5
-%ifidn %1,sp
-    paddd           m0, m7
-    paddd           m1, m7
-    paddd           m2, m7
-    paddd           m3, m7
-    psrad           m0, 12
-    psrad           m1, 12
-    psrad           m2, 12
-    psrad           m3, 12
-%else
-    psrad           m0, 6
-    psrad           m1, 6
-    psrad           m2, 6
-    psrad           m3, 6
-%endif
-    packssdw        m0, m1
-    packssdw        m2, m3
-%ifidn %1,sp
-    packuswb        m0, m2
-    mova            m3, [interp8_hps_shuf]
-    vpermd          m0, m3, m0
-    vextracti128    xm2, m0, 1
-%else
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
-%endif
-%endmacro
+    punpcklbw  m4, m2, m3
+    punpckhbw  m2, m3
 
-%macro FILTER_VER_CHROMA_S_AVX2_8x4 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x4, 4, 6, 8
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
+    pmaddubsw  m4, m1
+    pmaddubsw  m2, m1
 
-%ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
-%else
-    lea             r5, [pw_ChromaCoeffV + r4]
-%endif
+    lea        r5, [r0 + 2 * r1]
+    movu       m5, [r5]
+    movu       m7, [r5 + r1]
 
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,sp
-    mova            m7, [pd_526336]
-%else
-    add             r3d, r3d
-%endif
+    punpcklbw  m6, m5, m7
+    pmaddubsw  m6, m0
+    paddw      m4, m6
 
-    PROCESS_CHROMA_S_AVX2_W8_4R %1
-    lea             r4, [r3 * 3]
-%ifidn %1,sp
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r4], xm2
-%else
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r4], xm3
-%endif
+    punpckhbw  m6, m5, m7
+    pmaddubsw  m6, m0
+    paddw      m2, m6
+
+    mova       m6, [pw_2000]
+
+    psubw      m4, m6
+    psubw      m2, m6
+
+    movu       [r2], m4
+    movu       [r2 + 16], m2
+
+    punpcklbw  m4, m3, m5
+    punpckhbw  m3, m5
+
+    pmaddubsw  m4, m1
+    pmaddubsw  m3, m1
+
+    movu       m5, [r5 + 2 * r1]
+
+    punpcklbw  m2, m7, m5
+    punpckhbw  m7, m5
+
+    pmaddubsw  m2, m0
+    pmaddubsw  m7, m0
+
+    paddw      m4, m2
+    paddw      m3, m7
+
+    psubw      m4, m6
+    psubw      m3, m6
+
+    movu       [r2 + r3], m4
+    movu       [r2 + r3 + 16], m3
+
+    add         r0,        16
+    add         r2,        32
+    dec         r6d
+    jnz         .loopW
+
+    lea         r0,        [r0 + r1 * 2 - %1]
+    lea         r2,        [r2 + r3 * 2 - %1 * 2]
+
+    dec        r4d
+    jnz        .loop
     RET
 %endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_8x4 sp
-    FILTER_VER_CHROMA_S_AVX2_8x4 ss
+    FILTER_V_PS_W16n 64, 64
+    FILTER_V_PS_W16n 64, 32
+    FILTER_V_PS_W16n 64, 48
+    FILTER_V_PS_W16n 48, 64
+    FILTER_V_PS_W16n 64, 16
+
+
+;------------------------------------------------------------------------------------------------------------
+;void interp_4tap_vert_ps_2x4(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;------------------------------------------------------------------------------------------------------------
+INIT_XMM sse4
+cglobal interp_4tap_vert_ps_2x4, 4, 6, 7
 
-%macro FILTER_VER_CHROMA_S_AVX2_12x16 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_vert_%1_12x16, 4, 9, 10
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
+    mov         r4d, r4m
+    sub         r0, r1
+    add         r3d, r3d
 
 %ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
+    lea         r5, [tab_ChromaCoeff]
+    movd        m0, [r5 + r4 * 4]
 %else
-    lea             r5, [pw_ChromaCoeffV + r4]
+    movd        m0, [tab_ChromaCoeff + r4 * 4]
 %endif
 
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,sp
-    mova            m9, [pd_526336]
-%else
-    add             r3d, r3d
-%endif
-    lea             r6, [r3 * 3]
-    PROCESS_CHROMA_S_AVX2_W8_16R %1
-%ifidn %1,sp
-    add             r2, 8
-%else
-    add             r2, 16
-%endif
-    add             r0, 16
-    mova            m7, m9
-    PROCESS_CHROMA_AVX2_W4_16R %1
-    RET
-%endif
-%endmacro
+    pshufb      m0, [tab_Cm]
 
-    FILTER_VER_CHROMA_S_AVX2_12x16 sp
-    FILTER_VER_CHROMA_S_AVX2_12x16 ss
+    lea         r5, [3 * r1]
 
-%macro FILTER_VER_CHROMA_S_AVX2_12x32 1
-%if ARCH_X86_64 == 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_12x32, 4, 9, 10
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
+    movd        m2, [r0]
+    movd        m3, [r0 + r1]
+    movd        m4, [r0 + 2 * r1]
+    movd        m5, [r0 + r5]
 
-%ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
-%else
-    lea             r5, [pw_ChromaCoeffV + r4]
-%endif
+    punpcklbw   m2, m3
+    punpcklbw   m6, m4, m5
+    punpcklbw   m2, m6
 
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1, sp
-    mova            m9, [pd_526336]
-%else
-    add             r3d, r3d
-%endif
-    lea             r6, [r3 * 3]
-%rep 2
-    PROCESS_CHROMA_S_AVX2_W8_16R %1
-%ifidn %1, sp
-    add             r2, 8
-%else
-    add             r2, 16
-%endif
-    add             r0, 16
-    mova            m7, m9
-    PROCESS_CHROMA_AVX2_W4_16R %1
-    sub             r0, 16
-%ifidn %1, sp
-    lea             r2, [r2 + r3 * 4 - 8]
-%else
-    lea             r2, [r2 + r3 * 4 - 16]
-%endif
-%endrep
-    RET
-%endif
-%endmacro
+    pmaddubsw   m2, m0
 
-    FILTER_VER_CHROMA_S_AVX2_12x32 sp
-    FILTER_VER_CHROMA_S_AVX2_12x32 ss
+    lea         r0, [r0 + 4 * r1]
+    movd        m6, [r0]
 
-%macro FILTER_VER_CHROMA_S_AVX2_16x12 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_vert_%1_16x12, 4, 9, 9
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
+    punpcklbw   m3, m4
+    punpcklbw   m1, m5, m6
+    punpcklbw   m3, m1
 
-%ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
-%else
-    lea             r5, [pw_ChromaCoeffV + r4]
-%endif
+    pmaddubsw   m3, m0
+    phaddw      m2, m3
 
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,sp
-    mova            m8, [pd_526336]
-%else
-    add             r3d, r3d
-%endif
-    lea             r6, [r3 * 3]
-%rep 2
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m2, [r5]
-    lea             r7, [r0 + r1 * 4]
-    movu            xm4, [r7]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    paddd           m1, m5
-    pmaddwd         m3, [r5]
-%ifidn %1,sp
-    paddd           m0, m8
-    paddd           m1, m8
-    psrad           m0, 12
-    psrad           m1, 12
-%else
-    psrad           m0, 6
-    psrad           m1, 6
-%endif
-    packssdw        m0, m1
+    mova        m1, [pw_2000]
 
-    movu            xm5, [r7 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m2, m6
-    pmaddwd         m4, [r5]
-    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm1, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm1, 1
-    pmaddwd         m1, m5, [r5 + 1 * mmsize]
-    pmaddwd         m5, [r5]
-    paddd           m3, m1
-%ifidn %1,sp
-    paddd           m2, m8
-    paddd           m3, m8
-    psrad           m2, 12
-    psrad           m3, 12
-%else
-    psrad           m2, 6
-    psrad           m3, 6
-%endif
-    packssdw        m2, m3
-%ifidn %1,sp
-    packuswb        m0, m2
-    mova            m3, [interp8_hps_shuf]
-    vpermd          m0, m3, m0
-    vextracti128    xm2, m0, 1
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r6], xm2
-%else
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    movu            [r2], xm0
-    vextracti128    xm0, m0, 1
-    vextracti128    xm3, m2, 1
-    movu            [r2 + r3], xm0
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-%endif
-    lea             r8, [r2 + r3 * 4]
+    psubw       m2, m1
 
-    movu            xm1, [r7 + r4]                  ; m1 = row 7
-    punpckhwd       xm0, xm6, xm1
-    punpcklwd       xm6, xm1
-    vinserti128     m6, m6, xm0, 1
-    pmaddwd         m0, m6, [r5 + 1 * mmsize]
-    pmaddwd         m6, [r5]
-    paddd           m4, m0
-    lea             r7, [r7 + r1 * 4]
-    movu            xm0, [r7]                       ; m0 = row 8
-    punpckhwd       xm2, xm1, xm0
-    punpcklwd       xm1, xm0
-    vinserti128     m1, m1, xm2, 1
-    pmaddwd         m2, m1, [r5 + 1 * mmsize]
-    pmaddwd         m1, [r5]
-    paddd           m5, m2
-%ifidn %1,sp
-    paddd           m4, m8
-    paddd           m5, m8
-    psrad           m4, 12
-    psrad           m5, 12
-%else
-    psrad           m4, 6
-    psrad           m5, 6
-%endif
-    packssdw        m4, m5
+    movd        [r2], m2
+    pextrd      [r2 + r3], m2, 2
+
+    movd        m2, [r0 + r1]
+
+    punpcklbw   m4, m5
+    punpcklbw   m3, m6, m2
+    punpcklbw   m4, m3
+
+    pmaddubsw   m4, m0
+
+    movd        m3, [r0 + 2 * r1]
+
+    punpcklbw   m5, m6
+    punpcklbw   m2, m3
+    punpcklbw   m5, m2
+
+    pmaddubsw   m5, m0
+    phaddw      m4, m5
+    psubw       m4, m1
+
+    lea         r2, [r2 + 2 * r3]
+    movd        [r2], m4
+    pextrd      [r2 + r3], m4, 2
+
+    RET
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ps_2x8(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+%macro FILTER_V_PS_W2 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_ps_2x%2, 4, 6, 8
 
-    movu            xm2, [r7 + r1]                  ; m2 = row 9
-    punpckhwd       xm5, xm0, xm2
-    punpcklwd       xm0, xm2
-    vinserti128     m0, m0, xm5, 1
-    pmaddwd         m5, m0, [r5 + 1 * mmsize]
-    paddd           m6, m5
-    pmaddwd         m0, [r5]
-    movu            xm5, [r7 + r1 * 2]              ; m5 = row 10
-    punpckhwd       xm7, xm2, xm5
-    punpcklwd       xm2, xm5
-    vinserti128     m2, m2, xm7, 1
-    pmaddwd         m7, m2, [r5 + 1 * mmsize]
-    paddd           m1, m7
-    pmaddwd         m2, [r5]
+    mov        r4d, r4m
+    sub        r0, r1
+    add        r3d, r3d
 
-%ifidn %1,sp
-    paddd           m6, m8
-    paddd           m1, m8
-    psrad           m6, 12
-    psrad           m1, 12
-%else
-    psrad           m6, 6
-    psrad           m1, 6
-%endif
-    packssdw        m6, m1
-%ifidn %1,sp
-    packuswb        m4, m6
-    vpermd          m4, m3, m4
-    vextracti128    xm6, m4, 1
-    movq            [r8], xm4
-    movhps          [r8 + r3], xm4
-    movq            [r8 + r3 * 2], xm6
-    movhps          [r8 + r6], xm6
+%ifdef PIC
+    lea        r5, [tab_ChromaCoeff]
+    movd       m0, [r5 + r4 * 4]
 %else
-    vpermq          m4, m4, 11011000b
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm7, m4, 1
-    vextracti128    xm1, m6, 1
-    movu            [r8], xm4
-    movu            [r8 + r3], xm7
-    movu            [r8 + r3 * 2], xm6
-    movu            [r8 + r6], xm1
+    movd       m0, [tab_ChromaCoeff + r4 * 4]
 %endif
-    lea             r8, [r8 + r3 * 4]
 
-    movu            xm7, [r7 + r4]                  ; m7 = row 11
-    punpckhwd       xm1, xm5, xm7
-    punpcklwd       xm5, xm7
-    vinserti128     m5, m5, xm1, 1
-    pmaddwd         m1, m5, [r5 + 1 * mmsize]
-    paddd           m0, m1
-    pmaddwd         m5, [r5]
-    lea             r7, [r7 + r1 * 4]
-    movu            xm1, [r7]                       ; m1 = row 12
-    punpckhwd       xm4, xm7, xm1
-    punpcklwd       xm7, xm1
-    vinserti128     m7, m7, xm4, 1
-    pmaddwd         m4, m7, [r5 + 1 * mmsize]
-    paddd           m2, m4
-    pmaddwd         m7, [r5]
-%ifidn %1,sp
-    paddd           m0, m8
-    paddd           m2, m8
-    psrad           m0, 12
-    psrad           m2, 12
-%else
-    psrad           m0, 6
-    psrad           m2, 6
-%endif
-    packssdw        m0, m2
+    pshufb     m0, [tab_Cm]
 
-    movu            xm4, [r7 + r1]                  ; m4 = row 13
-    punpckhwd       xm2, xm1, xm4
-    punpcklwd       xm1, xm4
-    vinserti128     m1, m1, xm2, 1
-    pmaddwd         m1, [r5 + 1 * mmsize]
-    paddd           m5, m1
-    movu            xm2, [r7 + r1 * 2]              ; m2 = row 14
-    punpckhwd       xm6, xm4, xm2
-    punpcklwd       xm4, xm2
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m4, [r5 + 1 * mmsize]
-    paddd           m7, m4
-%ifidn %1,sp
-    paddd           m5, m8
-    paddd           m7, m8
-    psrad           m5, 12
-    psrad           m7, 12
-%else
-    psrad           m5, 6
-    psrad           m7, 6
-%endif
-    packssdw        m5, m7
-%ifidn %1,sp
-    packuswb        m0, m5
-    vpermd          m0, m3, m0
-    vextracti128    xm5, m0, 1
-    movq            [r8], xm0
-    movhps          [r8 + r3], xm0
-    movq            [r8 + r3 * 2], xm5
-    movhps          [r8 + r6], xm5
-    add             r2, 8
+    mova       m1, [pw_2000]
+    lea        r5, [3 * r1]
+    mov        r4d, %2/4
+.loop:
+    movd       m2, [r0]
+    movd       m3, [r0 + r1]
+    movd       m4, [r0 + 2 * r1]
+    movd       m5, [r0 + r5]
+
+    punpcklbw  m2, m3
+    punpcklbw  m6, m4, m5
+    punpcklbw  m2, m6
+
+    pmaddubsw  m2, m0
+
+    lea        r0, [r0 + 4 * r1]
+    movd       m6, [r0]
+
+    punpcklbw  m3, m4
+    punpcklbw  m7, m5, m6
+    punpcklbw  m3, m7
+
+    pmaddubsw  m3, m0
+
+    phaddw     m2, m3
+    psubw      m2, m1
+
+
+    movd       [r2], m2
+    pshufd     m2, m2, 2
+    movd       [r2 + r3], m2
+
+    movd       m2, [r0 + r1]
+
+    punpcklbw  m4, m5
+    punpcklbw  m3, m6, m2
+    punpcklbw  m4, m3
+
+    pmaddubsw  m4, m0
+
+    movd       m3, [r0 + 2 * r1]
+
+    punpcklbw  m5, m6
+    punpcklbw  m2, m3
+    punpcklbw  m5, m2
+
+    pmaddubsw  m5, m0
+
+    phaddw     m4, m5
+
+    psubw      m4, m1
+
+    lea        r2, [r2 + 2 * r3]
+    movd       [r2], m4
+    pshufd     m4 , m4 ,2
+    movd       [r2 + r3], m4
+
+    lea        r2, [r2 + 2 * r3]
+
+    dec        r4d
+    jnz        .loop
+
+    RET
+%endmacro
+
+    FILTER_V_PS_W2 2, 8
+
+    FILTER_V_PS_W2 2, 16
+
+;-----------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ss_%1x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_SS 2
+INIT_XMM sse2
+cglobal interp_4tap_vert_ss_%1x%2, 5, 7, 6 ,0-gprsize
+
+    add       r1d, r1d
+    add       r3d, r3d
+    sub       r0, r1
+    shl       r4d, 5
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r6, [r5 + r4]
 %else
-    vpermq          m0, m0, 11011000b
-    vpermq          m5, m5, 11011000b
-    vextracti128    xm7, m0, 1
-    vextracti128    xm6, m5, 1
-    movu            [r8], xm0
-    movu            [r8 + r3], xm7
-    movu            [r8 + r3 * 2], xm5
-    movu            [r8 + r6], xm6
-    add             r2, 16
+    lea       r6, [tab_ChromaCoeffV + r4]
 %endif
-    add             r0, 16
-%endrep
+
+    mov       dword [rsp], %2/4
+
+.loopH:
+    mov       r4d, (%1/4)
+.loopW:
+    PROCESS_CHROMA_SP_W4_4R
+
+    psrad     m0, 6
+    psrad     m1, 6
+    psrad     m2, 6
+    psrad     m3, 6
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+
+    movlps    [r2], m0
+    movhps    [r2 + r3], m0
+    lea       r5, [r2 + 2 * r3]
+    movlps    [r5], m2
+    movhps    [r5 + r3], m2
+
+    lea       r5, [4 * r1 - 2 * 4]
+    sub       r0, r5
+    add       r2, 2 * 4
+
+    dec       r4d
+    jnz       .loopW
+
+    lea       r0, [r0 + 4 * r1 - 2 * %1]
+    lea       r2, [r2 + 4 * r3 - 2 * %1]
+
+    dec       dword [rsp]
+    jnz       .loopH
+
     RET
-%endif
 %endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_16x12 sp
-    FILTER_VER_CHROMA_S_AVX2_16x12 ss
+    FILTER_VER_CHROMA_SS 4, 4
+    FILTER_VER_CHROMA_SS 4, 8
+    FILTER_VER_CHROMA_SS 16, 16
+    FILTER_VER_CHROMA_SS 16, 8
+    FILTER_VER_CHROMA_SS 16, 12
+    FILTER_VER_CHROMA_SS 12, 16
+    FILTER_VER_CHROMA_SS 16, 4
+    FILTER_VER_CHROMA_SS 4, 16
+    FILTER_VER_CHROMA_SS 32, 32
+    FILTER_VER_CHROMA_SS 32, 16
+    FILTER_VER_CHROMA_SS 16, 32
+    FILTER_VER_CHROMA_SS 32, 24
+    FILTER_VER_CHROMA_SS 24, 32
+    FILTER_VER_CHROMA_SS 32, 8
+
+    FILTER_VER_CHROMA_SS 16, 24
+    FILTER_VER_CHROMA_SS 12, 32
+    FILTER_VER_CHROMA_SS 4, 32
+    FILTER_VER_CHROMA_SS 32, 64
+    FILTER_VER_CHROMA_SS 16, 64
+    FILTER_VER_CHROMA_SS 32, 48
+    FILTER_VER_CHROMA_SS 24, 64
 
-%macro FILTER_VER_CHROMA_S_AVX2_8x12 1
-%if ARCH_X86_64 == 1
+    FILTER_VER_CHROMA_SS 64, 64
+    FILTER_VER_CHROMA_SS 64, 32
+    FILTER_VER_CHROMA_SS 64, 48
+    FILTER_VER_CHROMA_SS 48, 64
+    FILTER_VER_CHROMA_SS 64, 16
+
+%macro FILTER_VER_CHROMA_S_AVX2_4x4 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x12, 4, 7, 9
+cglobal interp_4tap_vert_%1_4x4, 4, 6, 7
     mov             r4d, r4m
-    shl             r4d, 6
     add             r1d, r1d
+    shl             r4d, 6
+    sub             r0, r1
 
 %ifdef PIC
     lea             r5, [pw_ChromaCoeffV]
@@ -20695,184 +15907,39 @@ cglobal interp_4tap_vert_%1_8x12, 4, 7, 9
 %endif
 
     lea             r4, [r1 * 3]
-    sub             r0, r1
 %ifidn %1,sp
-    mova            m8, [pd_526336]
+    mova            m6, [pd_526336]
 %else
     add             r3d, r3d
 %endif
-    lea             r6, [r3 * 3]
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
+
+    movq            xm0, [r0]
+    movq            xm1, [r0 + r1]
     punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
+    movq            xm2, [r0 + r1 * 2]
     punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
+    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
+    pmaddwd         m0, [r5]
+    movq            xm3, [r0 + r4]
     punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
+    lea             r0, [r0 + 4 * r1]
+    movq            xm4, [r0]
     punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    paddd           m1, m5
-    pmaddwd         m3, [r5]
-%ifidn %1,sp
-    paddd           m0, m8
-    paddd           m1, m8
-    psrad           m0, 12
-    psrad           m1, 12
-%else
-    psrad           m0, 6
-    psrad           m1, 6
-%endif
-    packssdw        m0, m1
-
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m2, m6
-    pmaddwd         m4, [r5]
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm1, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm1, 1
-    pmaddwd         m1, m5, [r5 + 1 * mmsize]
-    pmaddwd         m5, [r5]
-    paddd           m3, m1
-%ifidn %1,sp
-    paddd           m2, m8
-    paddd           m3, m8
-    psrad           m2, 12
-    psrad           m3, 12
-%else
-    psrad           m2, 6
-    psrad           m3, 6
-%endif
-    packssdw        m2, m3
-%ifidn %1,sp
-    packuswb        m0, m2
-    mova            m3, [interp8_hps_shuf]
-    vpermd          m0, m3, m0
-    vextracti128    xm2, m0, 1
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r6], xm2
-%else
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    movu            [r2], xm0
-    vextracti128    xm0, m0, 1
-    vextracti128    xm3, m2, 1
-    movu            [r2 + r3], xm0
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-%endif
-    lea             r2, [r2 + r3 * 4]
-
-    movu            xm1, [r0 + r4]                  ; m1 = row 7
-    punpckhwd       xm0, xm6, xm1
-    punpcklwd       xm6, xm1
-    vinserti128     m6, m6, xm0, 1
-    pmaddwd         m0, m6, [r5 + 1 * mmsize]
-    pmaddwd         m6, [r5]
-    paddd           m4, m0
-    lea             r0, [r0 + r1 * 4]
-    movu            xm0, [r0]                       ; m0 = row 8
-    punpckhwd       xm2, xm1, xm0
-    punpcklwd       xm1, xm0
-    vinserti128     m1, m1, xm2, 1
-    pmaddwd         m2, m1, [r5 + 1 * mmsize]
-    pmaddwd         m1, [r5]
-    paddd           m5, m2
-%ifidn %1,sp
-    paddd           m4, m8
-    paddd           m5, m8
-    psrad           m4, 12
-    psrad           m5, 12
-%else
-    psrad           m4, 6
-    psrad           m5, 6
-%endif
-    packssdw        m4, m5
-
-    movu            xm2, [r0 + r1]                  ; m2 = row 9
-    punpckhwd       xm5, xm0, xm2
-    punpcklwd       xm0, xm2
-    vinserti128     m0, m0, xm5, 1
-    pmaddwd         m5, m0, [r5 + 1 * mmsize]
-    paddd           m6, m5
-    pmaddwd         m0, [r5]
-    movu            xm5, [r0 + r1 * 2]              ; m5 = row 10
-    punpckhwd       xm7, xm2, xm5
-    punpcklwd       xm2, xm5
-    vinserti128     m2, m2, xm7, 1
-    pmaddwd         m7, m2, [r5 + 1 * mmsize]
-    paddd           m1, m7
+    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
+    pmaddwd         m5, m2, [r5 + 1 * mmsize]
     pmaddwd         m2, [r5]
-
-%ifidn %1,sp
-    paddd           m6, m8
-    paddd           m1, m8
-    psrad           m6, 12
-    psrad           m1, 12
-%else
-    psrad           m6, 6
-    psrad           m1, 6
-%endif
-    packssdw        m6, m1
-%ifidn %1,sp
-    packuswb        m4, m6
-    vpermd          m4, m3, m4
-    vextracti128    xm6, m4, 1
-    movq            [r2], xm4
-    movhps          [r2 + r3], xm4
-    movq            [r2 + r3 * 2], xm6
-    movhps          [r2 + r6], xm6
-%else
-    vpermq          m4, m4, 11011000b
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm7, m4, 1
-    vextracti128    xm1, m6, 1
-    movu            [r2], xm4
-    movu            [r2 + r3], xm7
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r6], xm1
-%endif
-    lea             r2, [r2 + r3 * 4]
-
-    movu            xm7, [r0 + r4]                  ; m7 = row 11
-    punpckhwd       xm1, xm5, xm7
-    punpcklwd       xm5, xm7
-    vinserti128     m5, m5, xm1, 1
-    pmaddwd         m1, m5, [r5 + 1 * mmsize]
-    paddd           m0, m1
-    pmaddwd         m5, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm1, [r0]                       ; m1 = row 12
-    punpckhwd       xm4, xm7, xm1
-    punpcklwd       xm7, xm1
-    vinserti128     m7, m7, xm4, 1
-    pmaddwd         m4, m7, [r5 + 1 * mmsize]
+    paddd           m0, m5
+    movq            xm3, [r0 + r1]
+    punpcklwd       xm4, xm3
+    movq            xm1, [r0 + r1 * 2]
+    punpcklwd       xm3, xm1
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
+    pmaddwd         m4, [r5 + 1 * mmsize]
     paddd           m2, m4
-    pmaddwd         m7, [r5]
+
 %ifidn %1,sp
-    paddd           m0, m8
-    paddd           m2, m8
+    paddd           m0, m6
+    paddd           m2, m6
     psrad           m0, 12
     psrad           m2, 12
 %else
@@ -20880,60 +15947,34 @@ cglobal interp_4tap_vert_%1_8x12, 4, 7, 9
     psrad           m2, 6
 %endif
     packssdw        m0, m2
+    vextracti128    xm2, m0, 1
+    lea             r4, [r3 * 3]
 
-    movu            xm4, [r0 + r1]                  ; m4 = row 13
-    punpckhwd       xm2, xm1, xm4
-    punpcklwd       xm1, xm4
-    vinserti128     m1, m1, xm2, 1
-    pmaddwd         m1, [r5 + 1 * mmsize]
-    paddd           m5, m1
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 14
-    punpckhwd       xm6, xm4, xm2
-    punpcklwd       xm4, xm2
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m4, [r5 + 1 * mmsize]
-    paddd           m7, m4
 %ifidn %1,sp
-    paddd           m5, m8
-    paddd           m7, m8
-    psrad           m5, 12
-    psrad           m7, 12
+    packuswb        xm0, xm2
+    movd            [r2], xm0
+    pextrd          [r2 + r3], xm0, 2
+    pextrd          [r2 + r3 * 2], xm0, 1
+    pextrd          [r2 + r4], xm0, 3
 %else
-    psrad           m5, 6
-    psrad           m7, 6
-%endif
-    packssdw        m5, m7
-%ifidn %1,sp
-    packuswb        m0, m5
-    vpermd          m0, m3, m0
-    vextracti128    xm5, m0, 1
     movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm5
-    movhps          [r2 + r6], xm5
-%else
-    vpermq          m0, m0, 11011000b
-    vpermq          m5, m5, 11011000b
-    vextracti128    xm7, m0, 1
-    vextracti128    xm6, m5, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm7
-    movu            [r2 + r3 * 2], xm5
-    movu            [r2 + r6], xm6
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r4], xm2
 %endif
     RET
-%endif
 %endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_8x12 sp
-    FILTER_VER_CHROMA_S_AVX2_8x12 ss
+    FILTER_VER_CHROMA_S_AVX2_4x4 sp
+    FILTER_VER_CHROMA_S_AVX2_4x4 ss
 
-%macro FILTER_VER_CHROMA_S_AVX2_16x4 1
+%macro FILTER_VER_CHROMA_S_AVX2_4x8 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_16x4, 4, 7, 8
+cglobal interp_4tap_vert_%1_4x8, 4, 6, 8
     mov             r4d, r4m
     shl             r4d, 6
     add             r1d, r1d
+    sub             r0, r1
 
 %ifdef PIC
     lea             r5, [pw_ChromaCoeffV]
@@ -20943,196 +15984,275 @@ cglobal interp_4tap_vert_%1_16x4, 4, 7, 8
 %endif
 
     lea             r4, [r1 * 3]
-    sub             r0, r1
 %ifidn %1,sp
     mova            m7, [pd_526336]
 %else
     add             r3d, r3d
 %endif
-%rep 2
-    PROCESS_CHROMA_S_AVX2_W8_4R %1
-    lea             r6, [r3 * 3]
-%ifidn %1,sp
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r6], xm2
-    add             r2, 8
-%else
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-    add             r2, 16
-%endif
-    lea             r6, [4 * r1 - 16]
-    sub             r0, r6
-%endrep
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_S_AVX2_16x4 sp
-    FILTER_VER_CHROMA_S_AVX2_16x4 ss
 
-%macro PROCESS_CHROMA_S_AVX2_W8_8R 1
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
+    movq            xm0, [r0]
+    movq            xm1, [r0 + r1]
     punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
+    movq            xm2, [r0 + r1 * 2]
     punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
+    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
+    pmaddwd         m0, [r5]
+    movq            xm3, [r0 + r4]
     punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m2, [r5]
-    lea             r7, [r0 + r1 * 4]
-    movu            xm4, [r7]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
+    lea             r0, [r0 + 4 * r1]
+    movq            xm4, [r0]
     punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    paddd           m1, m5
-    pmaddwd         m3, [r5]
+    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
+    pmaddwd         m5, m2, [r5 + 1 * mmsize]
+    pmaddwd         m2, [r5]
+    paddd           m0, m5
+    movq            xm3, [r0 + r1]
+    punpcklwd       xm4, xm3
+    movq            xm1, [r0 + r1 * 2]
+    punpcklwd       xm3, xm1
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
+    pmaddwd         m5, m4, [r5 + 1 * mmsize]
+    paddd           m2, m5
+    pmaddwd         m4, [r5]
+    movq            xm3, [r0 + r4]
+    punpcklwd       xm1, xm3
+    lea             r0, [r0 + 4 * r1]
+    movq            xm6, [r0]
+    punpcklwd       xm3, xm6
+    vinserti128     m1, m1, xm3, 1                  ; m1 = [8 7 7 6]
+    pmaddwd         m5, m1, [r5 + 1 * mmsize]
+    paddd           m4, m5
+    pmaddwd         m1, [r5]
+    movq            xm3, [r0 + r1]
+    punpcklwd       xm6, xm3
+    movq            xm5, [r0 + 2 * r1]
+    punpcklwd       xm3, xm5
+    vinserti128     m6, m6, xm3, 1                  ; m6 = [A 9 9 8]
+    pmaddwd         m6, [r5 + 1 * mmsize]
+    paddd           m1, m6
+    lea             r4, [r3 * 3]
+
 %ifidn %1,sp
     paddd           m0, m7
+    paddd           m2, m7
+    paddd           m4, m7
     paddd           m1, m7
     psrad           m0, 12
+    psrad           m2, 12
+    psrad           m4, 12
     psrad           m1, 12
 %else
     psrad           m0, 6
+    psrad           m2, 6
+    psrad           m4, 6
     psrad           m1, 6
 %endif
-    packssdw        m0, m1
-
-    movu            xm5, [r7 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m2, m6
-    pmaddwd         m4, [r5]
-    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm1, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm1, 1
-    pmaddwd         m1, m5, [r5 + 1 * mmsize]
-    pmaddwd         m5, [r5]
-    paddd           m3, m1
+    packssdw        m0, m2
+    packssdw        m4, m1
 %ifidn %1,sp
-    paddd           m2, m7
-    paddd           m3, m7
-    psrad           m2, 12
-    psrad           m3, 12
+    packuswb        m0, m4
+    vextracti128    xm2, m0, 1
+    movd            [r2], xm0
+    movd            [r2 + r3], xm2
+    pextrd          [r2 + r3 * 2], xm0, 1
+    pextrd          [r2 + r4], xm2, 1
+    lea             r2, [r2 + r3 * 4]
+    pextrd          [r2], xm0, 2
+    pextrd          [r2 + r3], xm2, 2
+    pextrd          [r2 + r3 * 2], xm0, 3
+    pextrd          [r2 + r4], xm2, 3
 %else
-    psrad           m2, 6
-    psrad           m3, 6
-%endif
-    packssdw        m2, m3
-%ifidn %1,sp
-    packuswb        m0, m2
-    mova            m3, [interp8_hps_shuf]
-    vpermd          m0, m3, m0
     vextracti128    xm2, m0, 1
+    vextracti128    xm1, m4, 1
     movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r6], xm2
-%else
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    movu            [r2], xm0
-    vextracti128    xm0, m0, 1
-    vextracti128    xm3, m2, 1
-    movu            [r2 + r3], xm0
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r4], xm2
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm4
+    movq            [r2 + r3], xm1
+    movhps          [r2 + r3 * 2], xm4
+    movhps          [r2 + r4], xm1
 %endif
-    lea             r8, [r2 + r3 * 4]
+    RET
+%endmacro
 
-    movu            xm1, [r7 + r4]                  ; m1 = row 7
-    punpckhwd       xm0, xm6, xm1
-    punpcklwd       xm6, xm1
-    vinserti128     m6, m6, xm0, 1
-    pmaddwd         m0, m6, [r5 + 1 * mmsize]
-    pmaddwd         m6, [r5]
-    paddd           m4, m0
-    lea             r7, [r7 + r1 * 4]
-    movu            xm0, [r7]                       ; m0 = row 8
-    punpckhwd       xm2, xm1, xm0
-    punpcklwd       xm1, xm0
-    vinserti128     m1, m1, xm2, 1
-    pmaddwd         m2, m1, [r5 + 1 * mmsize]
+    FILTER_VER_CHROMA_S_AVX2_4x8 sp
+    FILTER_VER_CHROMA_S_AVX2_4x8 ss
+
+%macro PROCESS_CHROMA_AVX2_W4_16R 1
+    movq            xm0, [r0]
+    movq            xm1, [r0 + r1]
+    punpcklwd       xm0, xm1
+    movq            xm2, [r0 + r1 * 2]
+    punpcklwd       xm1, xm2
+    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
+    pmaddwd         m0, [r5]
+    movq            xm3, [r0 + r4]
+    punpcklwd       xm2, xm3
+    lea             r0, [r0 + 4 * r1]
+    movq            xm4, [r0]
+    punpcklwd       xm3, xm4
+    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
+    pmaddwd         m5, m2, [r5 + 1 * mmsize]
+    pmaddwd         m2, [r5]
+    paddd           m0, m5
+    movq            xm3, [r0 + r1]
+    punpcklwd       xm4, xm3
+    movq            xm1, [r0 + r1 * 2]
+    punpcklwd       xm3, xm1
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
+    pmaddwd         m5, m4, [r5 + 1 * mmsize]
+    paddd           m2, m5
+    pmaddwd         m4, [r5]
+    movq            xm3, [r0 + r4]
+    punpcklwd       xm1, xm3
+    lea             r0, [r0 + 4 * r1]
+    movq            xm6, [r0]
+    punpcklwd       xm3, xm6
+    vinserti128     m1, m1, xm3, 1                  ; m1 = [8 7 7 6]
+    pmaddwd         m5, m1, [r5 + 1 * mmsize]
+    paddd           m4, m5
     pmaddwd         m1, [r5]
-    paddd           m5, m2
+    movq            xm3, [r0 + r1]
+    punpcklwd       xm6, xm3
+    movq            xm5, [r0 + 2 * r1]
+    punpcklwd       xm3, xm5
+    vinserti128     m6, m6, xm3, 1                  ; m6 = [10 9 9 8]
+    pmaddwd         m3, m6, [r5 + 1 * mmsize]
+    paddd           m1, m3
+    pmaddwd         m6, [r5]
+
 %ifidn %1,sp
+    paddd           m0, m7
+    paddd           m2, m7
     paddd           m4, m7
-    paddd           m5, m7
+    paddd           m1, m7
     psrad           m4, 12
-    psrad           m5, 12
+    psrad           m1, 12
+    psrad           m0, 12
+    psrad           m2, 12
 %else
+    psrad           m0, 6
+    psrad           m2, 6
     psrad           m4, 6
-    psrad           m5, 6
+    psrad           m1, 6
+%endif
+    packssdw        m0, m2
+    packssdw        m4, m1
+%ifidn %1,sp
+    packuswb        m0, m4
+    vextracti128    xm4, m0, 1
+    movd            [r2], xm0
+    movd            [r2 + r3], xm4
+    pextrd          [r2 + r3 * 2], xm0, 1
+    pextrd          [r2 + r6], xm4, 1
+    lea             r2, [r2 + r3 * 4]
+    pextrd          [r2], xm0, 2
+    pextrd          [r2 + r3], xm4, 2
+    pextrd          [r2 + r3 * 2], xm0, 3
+    pextrd          [r2 + r6], xm4, 3
+%else
+    vextracti128    xm2, m0, 1
+    vextracti128    xm1, m4, 1
+    movq            [r2], xm0
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r6], xm2
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm4
+    movq            [r2 + r3], xm1
+    movhps          [r2 + r3 * 2], xm4
+    movhps          [r2 + r6], xm1
 %endif
-    packssdw        m4, m5
 
-    movu            xm2, [r7 + r1]                  ; m2 = row 9
-    punpckhwd       xm5, xm0, xm2
+    movq            xm2, [r0 + r4]
+    punpcklwd       xm5, xm2
+    lea             r0, [r0 + 4 * r1]
+    movq            xm0, [r0]
+    punpcklwd       xm2, xm0
+    vinserti128     m5, m5, xm2, 1                  ; m5 = [12 11 11 10]
+    pmaddwd         m2, m5, [r5 + 1 * mmsize]
+    paddd           m6, m2
+    pmaddwd         m5, [r5]
+    movq            xm2, [r0 + r1]
     punpcklwd       xm0, xm2
-    vinserti128     m0, m0, xm5, 1
-    pmaddwd         m0, [r5 + 1 * mmsize]
-    paddd           m6, m0
-    movu            xm5, [r7 + r1 * 2]              ; m5 = row 10
-    punpckhwd       xm0, xm2, xm5
-    punpcklwd       xm2, xm5
-    vinserti128     m2, m2, xm0, 1
-    pmaddwd         m2, [r5 + 1 * mmsize]
-    paddd           m1, m2
+    movq            xm3, [r0 + 2 * r1]
+    punpcklwd       xm2, xm3
+    vinserti128     m0, m0, xm2, 1                  ; m0 = [14 13 13 12]
+    pmaddwd         m2, m0, [r5 + 1 * mmsize]
+    paddd           m5, m2
+    pmaddwd         m0, [r5]
+    movq            xm4, [r0 + r4]
+    punpcklwd       xm3, xm4
+    lea             r0, [r0 + 4 * r1]
+    movq            xm1, [r0]
+    punpcklwd       xm4, xm1
+    vinserti128     m3, m3, xm4, 1                  ; m3 = [16 15 15 14]
+    pmaddwd         m4, m3, [r5 + 1 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m3, [r5]
+    movq            xm4, [r0 + r1]
+    punpcklwd       xm1, xm4
+    movq            xm2, [r0 + 2 * r1]
+    punpcklwd       xm4, xm2
+    vinserti128     m1, m1, xm4, 1                  ; m1 = [18 17 17 16]
+    pmaddwd         m1, [r5 + 1 * mmsize]
+    paddd           m3, m1
 
 %ifidn %1,sp
     paddd           m6, m7
-    paddd           m1, m7
+    paddd           m5, m7
+    paddd           m0, m7
+    paddd           m3, m7
     psrad           m6, 12
-    psrad           m1, 12
+    psrad           m5, 12
+    psrad           m0, 12
+    psrad           m3, 12
 %else
     psrad           m6, 6
-    psrad           m1, 6
+    psrad           m5, 6
+    psrad           m0, 6
+    psrad           m3, 6
 %endif
-    packssdw        m6, m1
+    packssdw        m6, m5
+    packssdw        m0, m3
+    lea             r2, [r2 + r3 * 4]
+
 %ifidn %1,sp
-    packuswb        m4, m6
-    vpermd          m4, m3, m4
-    vextracti128    xm6, m4, 1
-    movq            [r8], xm4
-    movhps          [r8 + r3], xm4
-    movq            [r8 + r3 * 2], xm6
-    movhps          [r8 + r6], xm6
+    packuswb        m6, m0
+    vextracti128    xm0, m6, 1
+    movd            [r2], xm6
+    movd            [r2 + r3], xm0
+    pextrd          [r2 + r3 * 2], xm6, 1
+    pextrd          [r2 + r6], xm0, 1
+    lea             r2, [r2 + r3 * 4]
+    pextrd          [r2], xm6, 2
+    pextrd          [r2 + r3], xm0, 2
+    pextrd          [r2 + r3 * 2], xm6, 3
+    pextrd          [r2 + r6], xm0, 3
 %else
-    vpermq          m4, m4, 11011000b
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm7, m4, 1
-    vextracti128    xm1, m6, 1
-    movu            [r8], xm4
-    movu            [r8 + r3], xm7
-    movu            [r8 + r3 * 2], xm6
-    movu            [r8 + r6], xm1
+    vextracti128    xm5, m6, 1
+    vextracti128    xm3, m0, 1
+    movq            [r2], xm6
+    movq            [r2 + r3], xm5
+    movhps          [r2 + r3 * 2], xm6
+    movhps          [r2 + r6], xm5
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm0
+    movq            [r2 + r3], xm3
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r6], xm3
 %endif
 %endmacro
 
-%macro FILTER_VER_CHROMA_S_AVX2_Nx8 2
+%macro FILTER_VER_CHROMA_S_AVX2_4x16 1
 INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_vert_%1_%2x8, 4, 9, 8
+cglobal interp_4tap_vert_%1_4x16, 4, 7, 8
     mov             r4d, r4m
     shl             r4d, 6
     add             r1d, r1d
+    sub             r0, r1
 
 %ifdef PIC
     lea             r5, [pw_ChromaCoeffV]
@@ -21142,37 +16262,58 @@ cglobal interp_4tap_vert_%1_%2x8, 4, 9, 8
 %endif
 
     lea             r4, [r1 * 3]
-    sub             r0, r1
 %ifidn %1,sp
     mova            m7, [pd_526336]
 %else
     add             r3d, r3d
 %endif
     lea             r6, [r3 * 3]
-%rep %2 / 8
-    PROCESS_CHROMA_S_AVX2_W8_8R %1
+    PROCESS_CHROMA_AVX2_W4_16R %1
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_4x16 sp
+    FILTER_VER_CHROMA_S_AVX2_4x16 ss
+
+%macro FILTER_VER_CHROMA_S_AVX2_4x32 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_4x32, 4, 7, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+    sub             r0, r1
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
 %ifidn %1,sp
-    add             r2, 8
+    mova            m7, [pd_526336]
 %else
-    add             r2, 16
+    add             r3d, r3d
 %endif
-    add             r0, 16
+    lea             r6, [r3 * 3]
+%rep 2
+    PROCESS_CHROMA_AVX2_W4_16R %1
+    lea             r2, [r2 + r3 * 4]
 %endrep
     RET
-%endif
 %endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_Nx8 sp, 32
-    FILTER_VER_CHROMA_S_AVX2_Nx8 sp, 16
-    FILTER_VER_CHROMA_S_AVX2_Nx8 ss, 32
-    FILTER_VER_CHROMA_S_AVX2_Nx8 ss, 16
+    FILTER_VER_CHROMA_S_AVX2_4x32 sp
+    FILTER_VER_CHROMA_S_AVX2_4x32 ss
 
-%macro FILTER_VER_CHROMA_S_AVX2_8x2 1
+%macro FILTER_VER_CHROMA_S_AVX2_4x2 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x2, 4, 6, 6
+cglobal interp_4tap_vert_%1_4x2, 4, 6, 6
     mov             r4d, r4m
     shl             r4d, 6
     add             r1d, r1d
+    sub             r0, r1
 
 %ifdef PIC
     lea             r5, [pw_ChromaCoeffV]
@@ -21182,67 +16323,120 @@ cglobal interp_4tap_vert_%1_8x2, 4, 6, 6
 %endif
 
     lea             r4, [r1 * 3]
-    sub             r0, r1
 %ifidn %1,sp
     mova            m5, [pd_526336]
 %else
     add             r3d, r3d
 %endif
-
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
+    movq            xm0, [r0]
+    movq            xm1, [r0 + r1]
     punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
+    movq            xm2, [r0 + r1 * 2]
     punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
+    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
+    pmaddwd         m0, [r5]
+    movq            xm3, [r0 + r4]
     punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
+    movq            xm4, [r0 + 4 * r1]
+    punpcklwd       xm3, xm4
+    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
     pmaddwd         m2, [r5 + 1 * mmsize]
     paddd           m0, m2
-    movu            xm4, [r0 + r1 * 4]              ; m4 = row 4
-    punpckhwd       xm2, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm2, 1
-    pmaddwd         m3, [r5 + 1 * mmsize]
-    paddd           m1, m3
 %ifidn %1,sp
     paddd           m0, m5
-    paddd           m1, m5
     psrad           m0, 12
-    psrad           m1, 12
 %else
     psrad           m0, 6
-    psrad           m1, 6
 %endif
-    packssdw        m0, m1
-%ifidn %1,sp
     vextracti128    xm1, m0, 1
-    packuswb        xm0, xm1
-    pshufd          xm0, xm0, 11011000b
+    packssdw        xm0, xm1
+%ifidn %1,sp
+    packuswb        xm0, xm0
+    movd            [r2], xm0
+    pextrd          [r2 + r3], xm0, 1
+%else
     movq            [r2], xm0
     movhps          [r2 + r3], xm0
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_4x2 sp
+    FILTER_VER_CHROMA_S_AVX2_4x2 ss
+
+%macro FILTER_VER_CHROMA_S_AVX2_2x4 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_2x4, 4, 6, 6
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+    sub             r0, r1
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+%ifidn %1,sp
+    mova            m5, [pd_526336]
+%else
+    add             r3d, r3d
+%endif
+    movd            xm0, [r0]
+    movd            xm1, [r0 + r1]
+    punpcklwd       xm0, xm1
+    movd            xm2, [r0 + r1 * 2]
+    punpcklwd       xm1, xm2
+    punpcklqdq      xm0, xm1                        ; m0 = [2 1 1 0]
+    movd            xm3, [r0 + r4]
+    punpcklwd       xm2, xm3
+    lea             r0, [r0 + 4 * r1]
+    movd            xm4, [r0]
+    punpcklwd       xm3, xm4
+    punpcklqdq      xm2, xm3                        ; m2 = [4 3 3 2]
+    vinserti128     m0, m0, xm2, 1                  ; m0 = [4 3 3 2 2 1 1 0]
+    movd            xm1, [r0 + r1]
+    punpcklwd       xm4, xm1
+    movd            xm3, [r0 + r1 * 2]
+    punpcklwd       xm1, xm3
+    punpcklqdq      xm4, xm1                        ; m4 = [6 5 5 4]
+    vinserti128     m2, m2, xm4, 1                  ; m2 = [6 5 5 4 4 3 3 2]
+    pmaddwd         m0, [r5]
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m0, m2
+%ifidn %1,sp
+    paddd           m0, m5
+    psrad           m0, 12
 %else
-    vpermq          m0, m0, 11011000b
+    psrad           m0, 6
+%endif
     vextracti128    xm1, m0, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
+    packssdw        xm0, xm1
+    lea             r4, [r3 * 3]
+%ifidn %1,sp
+    packuswb        xm0, xm0
+    pextrw          [r2], xm0, 0
+    pextrw          [r2 + r3], xm0, 1
+    pextrw          [r2 + 2 * r3], xm0, 2
+    pextrw          [r2 + r4], xm0, 3
+%else
+    movd            [r2], xm0
+    pextrd          [r2 + r3], xm0, 1
+    pextrd          [r2 + 2 * r3], xm0, 2
+    pextrd          [r2 + r4], xm0, 3
 %endif
     RET
 %endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_8x2 sp
-    FILTER_VER_CHROMA_S_AVX2_8x2 ss
+    FILTER_VER_CHROMA_S_AVX2_2x4 sp
+    FILTER_VER_CHROMA_S_AVX2_2x4 ss
 
-%macro FILTER_VER_CHROMA_S_AVX2_8x6 1
+%macro FILTER_VER_CHROMA_S_AVX2_8x8 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x6, 4, 6, 8
+cglobal interp_4tap_vert_%1_8x8, 4, 6, 8
     mov             r4d, r4m
     shl             r4d, 6
     add             r1d, r1d
@@ -21328,85 +16522,98 @@ cglobal interp_4tap_vert_%1_8x6, 4, 6, 8
     punpckhwd       xm3, xm6, xm1
     punpcklwd       xm6, xm1
     vinserti128     m6, m6, xm3, 1
-    pmaddwd         m6, [r5 + 1 * mmsize]
-    paddd           m4, m6
-    movu            xm6, [r0 + r1 * 4]              ; m6 = row 8
-    punpckhwd       xm3, xm1, xm6
-    punpcklwd       xm1, xm6
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5 + 1 * mmsize]
-    paddd           m5, m1
-%ifidn %1,sp
-    paddd           m4, m7
-    paddd           m5, m7
-    psrad           m4, 12
-    psrad           m5, 12
-%else
-    psrad           m4, 6
-    psrad           m5, 6
-%endif
-    packssdw        m4, m5
+    pmaddwd         m3, m6, [r5 + 1 * mmsize]
+    pmaddwd         m6, [r5]
+    paddd           m4, m3
+
     lea             r4, [r3 * 3]
 %ifidn %1,sp
     packuswb        m0, m2
     mova            m3, [interp8_hps_shuf]
     vpermd          m0, m3, m0
     vextracti128    xm2, m0, 1
-    vextracti128    xm5, m4, 1
-    packuswb        xm4, xm5
-    pshufd          xm4, xm4, 11011000b
     movq            [r2], xm0
     movhps          [r2 + r3], xm0
     movq            [r2 + r3 * 2], xm2
     movhps          [r2 + r4], xm2
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm4
-    movhps          [r2 + r3], xm4
 %else
     vpermq          m0, m0, 11011000b
     vpermq          m2, m2, 11011000b
-    vpermq          m4, m4, 11011000b
     movu            [r2], xm0
     vextracti128    xm0, m0, 1
     vextracti128    xm3, m2, 1
-    vextracti128    xm5, m4, 1
     movu            [r2 + r3], xm0
     movu            [r2 + r3 * 2], xm2
     movu            [r2 + r4], xm3
+%endif
     lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm4
-    movu            [r2 + r3], xm5
+    lea             r0, [r0 + r1 * 4]
+    movu            xm0, [r0]                       ; m0 = row 8
+    punpckhwd       xm2, xm1, xm0
+    punpcklwd       xm1, xm0
+    vinserti128     m1, m1, xm2, 1
+    pmaddwd         m2, m1, [r5 + 1 * mmsize]
+    pmaddwd         m1, [r5]
+    paddd           m5, m2
+%ifidn %1,sp
+    paddd           m4, m7
+    paddd           m5, m7
+    psrad           m4, 12
+    psrad           m5, 12
+%else
+    psrad           m4, 6
+    psrad           m5, 6
 %endif
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_S_AVX2_8x6 sp
-    FILTER_VER_CHROMA_S_AVX2_8x6 ss
+    packssdw        m4, m5
 
-%macro FILTER_VER_CHROMA_S_AVX2_8xN 2
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_vert_%1_8x%2, 4, 7, 9
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
+    movu            xm2, [r0 + r1]                  ; m2 = row 9
+    punpckhwd       xm5, xm0, xm2
+    punpcklwd       xm0, xm2
+    vinserti128     m0, m0, xm5, 1
+    pmaddwd         m0, [r5 + 1 * mmsize]
+    paddd           m6, m0
+    movu            xm5, [r0 + r1 * 2]              ; m5 = row 10
+    punpckhwd       xm0, xm2, xm5
+    punpcklwd       xm2, xm5
+    vinserti128     m2, m2, xm0, 1
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m1, m2
 
-%ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
+%ifidn %1,sp
+    paddd           m6, m7
+    paddd           m1, m7
+    psrad           m6, 12
+    psrad           m1, 12
 %else
-    lea             r5, [pw_ChromaCoeffV + r4]
+    psrad           m6, 6
+    psrad           m1, 6
 %endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
+    packssdw        m6, m1
 %ifidn %1,sp
-    mova            m8, [pd_526336]
+    packuswb        m4, m6
+    vpermd          m4, m3, m4
+    vextracti128    xm6, m4, 1
+    movq            [r2], xm4
+    movhps          [r2 + r3], xm4
+    movq            [r2 + r3 * 2], xm6
+    movhps          [r2 + r4], xm6
 %else
-    add             r3d, r3d
+    vpermq          m4, m4, 11011000b
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm5, m4, 1
+    vextracti128    xm1, m6, 1
+    movu            [r2], xm4
+    movu            [r2 + r3], xm5
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r4], xm1
 %endif
-    lea             r6, [r3 * 3]
-%rep %2 / 16
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_8x8 sp
+    FILTER_VER_CHROMA_S_AVX2_8x8 ss
+
+%macro PROCESS_CHROMA_S_AVX2_W8_16R 1
     movu            xm0, [r0]                       ; m0 = row 0
     movu            xm1, [r0 + r1]                  ; m1 = row 1
     punpckhwd       xm2, xm0, xm1
@@ -21425,48 +16632,44 @@ cglobal interp_4tap_vert_%1_8x%2, 4, 7, 9
     pmaddwd         m4, m2, [r5 + 1 * mmsize]
     paddd           m0, m4
     pmaddwd         m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
+    lea             r7, [r0 + r1 * 4]
+    movu            xm4, [r7]                       ; m4 = row 4
     punpckhwd       xm5, xm3, xm4
     punpcklwd       xm3, xm4
     vinserti128     m3, m3, xm5, 1
     pmaddwd         m5, m3, [r5 + 1 * mmsize]
     paddd           m1, m5
     pmaddwd         m3, [r5]
-%ifidn %1,sp
-    paddd           m0, m8
-    paddd           m1, m8
-    psrad           m0, 12
-    psrad           m1, 12
-%else
-    psrad           m0, 6
-    psrad           m1, 6
-%endif
-    packssdw        m0, m1
-
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    movu            xm5, [r7 + r1]                  ; m5 = row 5
     punpckhwd       xm6, xm4, xm5
     punpcklwd       xm4, xm5
     vinserti128     m4, m4, xm6, 1
     pmaddwd         m6, m4, [r5 + 1 * mmsize]
     paddd           m2, m6
     pmaddwd         m4, [r5]
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm1, xm5, xm6
+    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm7, xm5, xm6
     punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm1, 1
-    pmaddwd         m1, m5, [r5 + 1 * mmsize]
+    vinserti128     m5, m5, xm7, 1
+    pmaddwd         m7, m5, [r5 + 1 * mmsize]
+    paddd           m3, m7
     pmaddwd         m5, [r5]
-    paddd           m3, m1
 %ifidn %1,sp
-    paddd           m2, m8
-    paddd           m3, m8
+    paddd           m0, m9
+    paddd           m1, m9
+    paddd           m2, m9
+    paddd           m3, m9
+    psrad           m0, 12
+    psrad           m1, 12
     psrad           m2, 12
     psrad           m3, 12
 %else
+    psrad           m0, 6
+    psrad           m1, 6
     psrad           m2, 6
     psrad           m3, 6
 %endif
+    packssdw        m0, m1
     packssdw        m2, m3
 %ifidn %1,sp
     packuswb        m0, m2
@@ -21480,239 +16683,216 @@ cglobal interp_4tap_vert_%1_8x%2, 4, 7, 9
 %else
     vpermq          m0, m0, 11011000b
     vpermq          m2, m2, 11011000b
-    movu            [r2], xm0
-    vextracti128    xm0, m0, 1
+    vextracti128    xm1, m0, 1
     vextracti128    xm3, m2, 1
-    movu            [r2 + r3], xm0
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
     movu            [r2 + r3 * 2], xm2
     movu            [r2 + r6], xm3
 %endif
-    lea             r2, [r2 + r3 * 4]
 
-    movu            xm1, [r0 + r4]                  ; m1 = row 7
-    punpckhwd       xm0, xm6, xm1
-    punpcklwd       xm6, xm1
-    vinserti128     m6, m6, xm0, 1
-    pmaddwd         m0, m6, [r5 + 1 * mmsize]
-    pmaddwd         m6, [r5]
-    paddd           m4, m0
-    lea             r0, [r0 + r1 * 4]
-    movu            xm0, [r0]                       ; m0 = row 8
-    punpckhwd       xm2, xm1, xm0
-    punpcklwd       xm1, xm0
-    vinserti128     m1, m1, xm2, 1
-    pmaddwd         m2, m1, [r5 + 1 * mmsize]
-    pmaddwd         m1, [r5]
-    paddd           m5, m2
-%ifidn %1,sp
+    movu            xm7, [r7 + r4]                  ; m7 = row 7
+    punpckhwd       xm8, xm6, xm7
+    punpcklwd       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddwd         m8, m6, [r5 + 1 * mmsize]
     paddd           m4, m8
-    paddd           m5, m8
-    psrad           m4, 12
-    psrad           m5, 12
-%else
-    psrad           m4, 6
-    psrad           m5, 6
-%endif
-    packssdw        m4, m5
-
-    movu            xm2, [r0 + r1]                  ; m2 = row 9
-    punpckhwd       xm5, xm0, xm2
-    punpcklwd       xm0, xm2
-    vinserti128     m0, m0, xm5, 1
-    pmaddwd         m5, m0, [r5 + 1 * mmsize]
-    paddd           m6, m5
+    pmaddwd         m6, [r5]
+    lea             r7, [r7 + r1 * 4]
+    movu            xm8, [r7]                       ; m8 = row 8
+    punpckhwd       xm0, xm7, xm8
+    punpcklwd       xm7, xm8
+    vinserti128     m7, m7, xm0, 1
+    pmaddwd         m0, m7, [r5 + 1 * mmsize]
+    paddd           m5, m0
+    pmaddwd         m7, [r5]
+    movu            xm0, [r7 + r1]                  ; m0 = row 9
+    punpckhwd       xm1, xm8, xm0
+    punpcklwd       xm8, xm0
+    vinserti128     m8, m8, xm1, 1
+    pmaddwd         m1, m8, [r5 + 1 * mmsize]
+    paddd           m6, m1
+    pmaddwd         m8, [r5]
+    movu            xm1, [r7 + r1 * 2]              ; m1 = row 10
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m2, m0, [r5 + 1 * mmsize]
+    paddd           m7, m2
     pmaddwd         m0, [r5]
-    movu            xm5, [r0 + r1 * 2]              ; m5 = row 10
-    punpckhwd       xm7, xm2, xm5
-    punpcklwd       xm2, xm5
-    vinserti128     m2, m2, xm7, 1
-    pmaddwd         m7, m2, [r5 + 1 * mmsize]
-    paddd           m1, m7
-    pmaddwd         m2, [r5]
-
 %ifidn %1,sp
-    paddd           m6, m8
-    paddd           m1, m8
+    paddd           m4, m9
+    paddd           m5, m9
+    psrad           m4, 12
+    psrad           m5, 12
+    paddd           m6, m9
+    paddd           m7, m9
     psrad           m6, 12
-    psrad           m1, 12
+    psrad           m7, 12
 %else
+    psrad           m4, 6
+    psrad           m5, 6
     psrad           m6, 6
-    psrad           m1, 6
+    psrad           m7, 6
 %endif
-    packssdw        m6, m1
+    packssdw        m4, m5
+    packssdw        m6, m7
+    lea             r8, [r2 + r3 * 4]
 %ifidn %1,sp
     packuswb        m4, m6
     vpermd          m4, m3, m4
     vextracti128    xm6, m4, 1
-    movq            [r2], xm4
-    movhps          [r2 + r3], xm4
-    movq            [r2 + r3 * 2], xm6
-    movhps          [r2 + r6], xm6
+    movq            [r8], xm4
+    movhps          [r8 + r3], xm4
+    movq            [r8 + r3 * 2], xm6
+    movhps          [r8 + r6], xm6
 %else
     vpermq          m4, m4, 11011000b
     vpermq          m6, m6, 11011000b
-    vextracti128    xm7, m4, 1
-    vextracti128    xm1, m6, 1
-    movu            [r2], xm4
-    movu            [r2 + r3], xm7
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r6], xm1
+    vextracti128    xm5, m4, 1
+    vextracti128    xm7, m6, 1
+    movu            [r8], xm4
+    movu            [r8 + r3], xm5
+    movu            [r8 + r3 * 2], xm6
+    movu            [r8 + r6], xm7
 %endif
-    lea             r2, [r2 + r3 * 4]
 
-    movu            xm7, [r0 + r4]                  ; m7 = row 11
-    punpckhwd       xm1, xm5, xm7
-    punpcklwd       xm5, xm7
-    vinserti128     m5, m5, xm1, 1
-    pmaddwd         m1, m5, [r5 + 1 * mmsize]
-    paddd           m0, m1
+    movu            xm2, [r7 + r4]                  ; m2 = row 11
+    punpckhwd       xm4, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm4, 1
+    pmaddwd         m4, m1, [r5 + 1 * mmsize]
+    paddd           m8, m4
+    pmaddwd         m1, [r5]
+    lea             r7, [r7 + r1 * 4]
+    movu            xm4, [r7]                       ; m4 = row 12
+    punpckhwd       xm5, xm2, xm4
+    punpcklwd       xm2, xm4
+    vinserti128     m2, m2, xm5, 1
+    pmaddwd         m5, m2, [r5 + 1 * mmsize]
+    paddd           m0, m5
+    pmaddwd         m2, [r5]
+    movu            xm5, [r7 + r1]                  ; m5 = row 13
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m1, m6
+    pmaddwd         m4, [r5]
+    movu            xm6, [r7 + r1 * 2]              ; m6 = row 14
+    punpckhwd       xm7, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddwd         m7, m5, [r5 + 1 * mmsize]
+    paddd           m2, m7
     pmaddwd         m5, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm1, [r0]                       ; m1 = row 12
-    punpckhwd       xm4, xm7, xm1
-    punpcklwd       xm7, xm1
-    vinserti128     m7, m7, xm4, 1
-    pmaddwd         m4, m7, [r5 + 1 * mmsize]
-    paddd           m2, m4
-    pmaddwd         m7, [r5]
 %ifidn %1,sp
-    paddd           m0, m8
-    paddd           m2, m8
+    paddd           m8, m9
+    paddd           m0, m9
+    paddd           m1, m9
+    paddd           m2, m9
+    psrad           m8, 12
     psrad           m0, 12
+    psrad           m1, 12
     psrad           m2, 12
 %else
+    psrad           m8, 6
     psrad           m0, 6
+    psrad           m1, 6
     psrad           m2, 6
 %endif
-    packssdw        m0, m2
-
-    movu            xm4, [r0 + r1]                  ; m4 = row 13
-    punpckhwd       xm2, xm1, xm4
-    punpcklwd       xm1, xm4
-    vinserti128     m1, m1, xm2, 1
-    pmaddwd         m2, m1, [r5 + 1 * mmsize]
-    paddd           m5, m2
-    pmaddwd         m1, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 14
-    punpckhwd       xm6, xm4, xm2
-    punpcklwd       xm4, xm2
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m7, m6
-    pmaddwd         m4, [r5]
-%ifidn %1,sp
-    paddd           m5, m8
-    paddd           m7, m8
-    psrad           m5, 12
-    psrad           m7, 12
-%else
-    psrad           m5, 6
-    psrad           m7, 6
-%endif
-    packssdw        m5, m7
+    packssdw        m8, m0
+    packssdw        m1, m2
+    lea             r8, [r8 + r3 * 4]
 %ifidn %1,sp
-    packuswb        m0, m5
-    vpermd          m0, m3, m0
-    vextracti128    xm5, m0, 1
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm5
-    movhps          [r2 + r6], xm5
+    packuswb        m8, m1
+    vpermd          m8, m3, m8
+    vextracti128    xm1, m8, 1
+    movq            [r8], xm8
+    movhps          [r8 + r3], xm8
+    movq            [r8 + r3 * 2], xm1
+    movhps          [r8 + r6], xm1
 %else
-    vpermq          m0, m0, 11011000b
-    vpermq          m5, m5, 11011000b
-    vextracti128    xm7, m0, 1
-    vextracti128    xm6, m5, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm7
-    movu            [r2 + r3 * 2], xm5
-    movu            [r2 + r6], xm6
+    vpermq          m8, m8, 11011000b
+    vpermq          m1, m1, 11011000b
+    vextracti128    xm0, m8, 1
+    vextracti128    xm2, m1, 1
+    movu            [r8], xm8
+    movu            [r8 + r3], xm0
+    movu            [r8 + r3 * 2], xm1
+    movu            [r8 + r6], xm2
 %endif
-    lea             r2, [r2 + r3 * 4]
+    lea             r8, [r8 + r3 * 4]
 
-    movu            xm6, [r0 + r4]                  ; m6 = row 15
-    punpckhwd       xm5, xm2, xm6
-    punpcklwd       xm2, xm6
-    vinserti128     m2, m2, xm5, 1
-    pmaddwd         m5, m2, [r5 + 1 * mmsize]
-    paddd           m1, m5
-    pmaddwd         m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm0, [r0]                       ; m0 = row 16
-    punpckhwd       xm5, xm6, xm0
-    punpcklwd       xm6, xm0
-    vinserti128     m6, m6, xm5, 1
-    pmaddwd         m5, m6, [r5 + 1 * mmsize]
-    paddd           m4, m5
+    movu            xm7, [r7 + r4]                  ; m7 = row 15
+    punpckhwd       xm2, xm6, xm7
+    punpcklwd       xm6, xm7
+    vinserti128     m6, m6, xm2, 1
+    pmaddwd         m2, m6, [r5 + 1 * mmsize]
+    paddd           m4, m2
     pmaddwd         m6, [r5]
-%ifidn %1,sp
-    paddd           m1, m8
-    paddd           m4, m8
-    psrad           m1, 12
-    psrad           m4, 12
-%else
-    psrad           m1, 6
-    psrad           m4, 6
-%endif
-    packssdw        m1, m4
+    lea             r7, [r7 + r1 * 4]
+    movu            xm2, [r7]                       ; m2 = row 16
+    punpckhwd       xm1, xm7, xm2
+    punpcklwd       xm7, xm2
+    vinserti128     m7, m7, xm1, 1
+    pmaddwd         m1, m7, [r5 + 1 * mmsize]
+    paddd           m5, m1
+    pmaddwd         m7, [r5]
+    movu            xm1, [r7 + r1]                  ; m1 = row 17
+    punpckhwd       xm0, xm2, xm1
+    punpcklwd       xm2, xm1
+    vinserti128     m2, m2, xm0, 1
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m6, m2
+    movu            xm0, [r7 + r1 * 2]              ; m0 = row 18
+    punpckhwd       xm2, xm1, xm0
+    punpcklwd       xm1, xm0
+    vinserti128     m1, m1, xm2, 1
+    pmaddwd         m1, [r5 + 1 * mmsize]
+    paddd           m7, m1
 
-    movu            xm5, [r0 + r1]                  ; m5 = row 17
-    punpckhwd       xm4, xm0, xm5
-    punpcklwd       xm0, xm5
-    vinserti128     m0, m0, xm4, 1
-    pmaddwd         m0, [r5 + 1 * mmsize]
-    paddd           m2, m0
-    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
-    punpckhwd       xm0, xm5, xm4
-    punpcklwd       xm5, xm4
-    vinserti128     m5, m5, xm0, 1
-    pmaddwd         m5, [r5 + 1 * mmsize]
-    paddd           m6, m5
 %ifidn %1,sp
-    paddd           m2, m8
-    paddd           m6, m8
-    psrad           m2, 12
+    paddd           m4, m9
+    paddd           m5, m9
+    paddd           m6, m9
+    paddd           m7, m9
+    psrad           m4, 12
+    psrad           m5, 12
     psrad           m6, 12
+    psrad           m7, 12
 %else
-    psrad           m2, 6
+    psrad           m4, 6
+    psrad           m5, 6
     psrad           m6, 6
+    psrad           m7, 6
 %endif
-    packssdw        m2, m6
+    packssdw        m4, m5
+    packssdw        m6, m7
 %ifidn %1,sp
-    packuswb        m1, m2
-    vpermd          m1, m3, m1
-    vextracti128    xm2, m1, 1
-    movq            [r2], xm1
-    movhps          [r2 + r3], xm1
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r6], xm2
-%else
-    vpermq          m1, m1, 11011000b
-    vpermq          m2, m2, 11011000b
-    vextracti128    xm6, m1, 1
-    vextracti128    xm4, m2, 1
-    movu            [r2], xm1
-    movu            [r2 + r3], xm6
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm4
-%endif
-    lea             r2, [r2 + r3 * 4]
-%endrep
-    RET
+    packuswb        m4, m6
+    vpermd          m4, m3, m4
+    vextracti128    xm6, m4, 1
+    movq            [r8], xm4
+    movhps          [r8 + r3], xm4
+    movq            [r8 + r3 * 2], xm6
+    movhps          [r8 + r6], xm6
+%else
+    vpermq          m4, m4, 11011000b
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm5, m4, 1
+    vextracti128    xm7, m6, 1
+    movu            [r8], xm4
+    movu            [r8 + r3], xm5
+    movu            [r8 + r3 * 2], xm6
+    movu            [r8 + r6], xm7
 %endif
 %endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_8xN sp, 16
-    FILTER_VER_CHROMA_S_AVX2_8xN sp, 32
-    FILTER_VER_CHROMA_S_AVX2_8xN sp, 64
-    FILTER_VER_CHROMA_S_AVX2_8xN ss, 16
-    FILTER_VER_CHROMA_S_AVX2_8xN ss, 32
-    FILTER_VER_CHROMA_S_AVX2_8xN ss, 64
-
-%macro FILTER_VER_CHROMA_S_AVX2_Nx24 2
-%if ARCH_X86_64 == 1
+%macro FILTER_VER_CHROMA_S_AVX2_Nx16 2
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_%2x24, 4, 10, 10
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_vert_%1_%2x16, 4, 10, 10
     mov             r4d, r4m
     shl             r4d, 6
     add             r1d, r1d
@@ -21723,7 +16903,6 @@ cglobal interp_4tap_vert_%1_%2x24, 4, 10, 10
 %else
     lea             r5, [pw_ChromaCoeffV + r4]
 %endif
-
     lea             r4, [r1 * 3]
     sub             r0, r1
 %ifidn %1,sp
@@ -21743,40 +16922,162 @@ cglobal interp_4tap_vert_%1_%2x24, 4, 10, 10
     add             r0, 16
     dec             r9d
     jnz             .loopW
-%ifidn %1,sp
-    lea             r2, [r8 + r3 * 4 - %2 + 8]
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_Nx16 sp, 16
+    FILTER_VER_CHROMA_S_AVX2_Nx16 sp, 32
+    FILTER_VER_CHROMA_S_AVX2_Nx16 sp, 64
+    FILTER_VER_CHROMA_S_AVX2_Nx16 ss, 16
+    FILTER_VER_CHROMA_S_AVX2_Nx16 ss, 32
+    FILTER_VER_CHROMA_S_AVX2_Nx16 ss, 64
+
+%macro FILTER_VER_CHROMA_S_AVX2_NxN 3
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_vert_%3_%1x%2, 4, 11, 10
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
 %else
-    lea             r2, [r8 + r3 * 4 - 2 * %2 + 16]
+    lea             r5, [pw_ChromaCoeffV + r4]
 %endif
-    lea             r0, [r7 - 2 * %2 + 16]
-    mova            m7, m9
-    mov             r9d, %2 / 8
-.loop:
-    PROCESS_CHROMA_S_AVX2_W8_8R %1
-%ifidn %1,sp
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %3,sp
+    mova            m9, [pd_526336]
+%else
+    add             r3d, r3d
+%endif
+    lea             r6, [r3 * 3]
+    mov             r9d, %2 / 16
+.loopH:
+    mov             r10d, %1 / 8
+.loopW:
+    PROCESS_CHROMA_S_AVX2_W8_16R %3
+%ifidn %3,sp
     add             r2, 8
 %else
     add             r2, 16
 %endif
     add             r0, 16
+    dec             r10d
+    jnz             .loopW
+    lea             r0, [r7 - 2 * %1 + 16]
+%ifidn %3,sp
+    lea             r2, [r8 + r3 * 4 - %1 + 8]
+%else
+    lea             r2, [r8 + r3 * 4 - 2 * %1 + 16]
+%endif
     dec             r9d
-    jnz             .loop
+    jnz             .loopH
     RET
 %endif
 %endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_Nx24 sp, 32
-    FILTER_VER_CHROMA_S_AVX2_Nx24 sp, 16
-    FILTER_VER_CHROMA_S_AVX2_Nx24 ss, 32
-    FILTER_VER_CHROMA_S_AVX2_Nx24 ss, 16
+    FILTER_VER_CHROMA_S_AVX2_NxN 16, 32, sp
+    FILTER_VER_CHROMA_S_AVX2_NxN 24, 32, sp
+    FILTER_VER_CHROMA_S_AVX2_NxN 32, 32, sp
+    FILTER_VER_CHROMA_S_AVX2_NxN 16, 32, ss
+    FILTER_VER_CHROMA_S_AVX2_NxN 24, 32, ss
+    FILTER_VER_CHROMA_S_AVX2_NxN 32, 32, ss
+    FILTER_VER_CHROMA_S_AVX2_NxN 16, 64, sp
+    FILTER_VER_CHROMA_S_AVX2_NxN 24, 64, sp
+    FILTER_VER_CHROMA_S_AVX2_NxN 32, 64, sp
+    FILTER_VER_CHROMA_S_AVX2_NxN 32, 48, sp
+    FILTER_VER_CHROMA_S_AVX2_NxN 32, 48, ss
+    FILTER_VER_CHROMA_S_AVX2_NxN 16, 64, ss
+    FILTER_VER_CHROMA_S_AVX2_NxN 24, 64, ss
+    FILTER_VER_CHROMA_S_AVX2_NxN 32, 64, ss
+    FILTER_VER_CHROMA_S_AVX2_NxN 64, 64, sp
+    FILTER_VER_CHROMA_S_AVX2_NxN 64, 32, sp
+    FILTER_VER_CHROMA_S_AVX2_NxN 64, 48, sp
+    FILTER_VER_CHROMA_S_AVX2_NxN 48, 64, sp
+    FILTER_VER_CHROMA_S_AVX2_NxN 64, 64, ss
+    FILTER_VER_CHROMA_S_AVX2_NxN 64, 32, ss
+    FILTER_VER_CHROMA_S_AVX2_NxN 64, 48, ss
+    FILTER_VER_CHROMA_S_AVX2_NxN 48, 64, ss
 
-%macro FILTER_VER_CHROMA_S_AVX2_2x8 1
+%macro PROCESS_CHROMA_S_AVX2_W8_4R 1
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m2, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    paddd           m1, m5
+    pmaddwd         m3, [r5]
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m4, [r5 + 1 * mmsize]
+    paddd           m2, m4
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm4, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm4, 1
+    pmaddwd         m5, [r5 + 1 * mmsize]
+    paddd           m3, m5
+%ifidn %1,sp
+    paddd           m0, m7
+    paddd           m1, m7
+    paddd           m2, m7
+    paddd           m3, m7
+    psrad           m0, 12
+    psrad           m1, 12
+    psrad           m2, 12
+    psrad           m3, 12
+%else
+    psrad           m0, 6
+    psrad           m1, 6
+    psrad           m2, 6
+    psrad           m3, 6
+%endif
+    packssdw        m0, m1
+    packssdw        m2, m3
+%ifidn %1,sp
+    packuswb        m0, m2
+    mova            m3, [interp8_hps_shuf]
+    vpermd          m0, m3, m0
+    vextracti128    xm2, m0, 1
+%else
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m2, 1
+%endif
+%endmacro
+
+%macro FILTER_VER_CHROMA_S_AVX2_8x4 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_2x8, 4, 6, 7
+cglobal interp_4tap_vert_%1_8x4, 4, 6, 8
     mov             r4d, r4m
     shl             r4d, 6
     add             r1d, r1d
-    sub             r0, r1
 
 %ifdef PIC
     lea             r5, [pw_ChromaCoeffV]
@@ -21786,97 +17087,78 @@ cglobal interp_4tap_vert_%1_2x8, 4, 6, 7
 %endif
 
     lea             r4, [r1 * 3]
+    sub             r0, r1
 %ifidn %1,sp
-    mova            m6, [pd_526336]
+    mova            m7, [pd_526336]
 %else
     add             r3d, r3d
 %endif
-    movd            xm0, [r0]
-    movd            xm1, [r0 + r1]
-    punpcklwd       xm0, xm1
-    movd            xm2, [r0 + r1 * 2]
-    punpcklwd       xm1, xm2
-    punpcklqdq      xm0, xm1                        ; m0 = [2 1 1 0]
-    movd            xm3, [r0 + r4]
-    punpcklwd       xm2, xm3
-    lea             r0, [r0 + 4 * r1]
-    movd            xm4, [r0]
-    punpcklwd       xm3, xm4
-    punpcklqdq      xm2, xm3                        ; m2 = [4 3 3 2]
-    vinserti128     m0, m0, xm2, 1                  ; m0 = [4 3 3 2 2 1 1 0]
-    movd            xm1, [r0 + r1]
-    punpcklwd       xm4, xm1
-    movd            xm3, [r0 + r1 * 2]
-    punpcklwd       xm1, xm3
-    punpcklqdq      xm4, xm1                        ; m4 = [6 5 5 4]
-    vinserti128     m2, m2, xm4, 1                  ; m2 = [6 5 5 4 4 3 3 2]
-    pmaddwd         m0, [r5]
-    pmaddwd         m2, [r5 + 1 * mmsize]
-    paddd           m0, m2
-    movd            xm1, [r0 + r4]
-    punpcklwd       xm3, xm1
-    lea             r0, [r0 + 4 * r1]
-    movd            xm2, [r0]
-    punpcklwd       xm1, xm2
-    punpcklqdq      xm3, xm1                        ; m3 = [8 7 7 6]
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [8 7 7 6 6 5 5 4]
-    movd            xm1, [r0 + r1]
-    punpcklwd       xm2, xm1
-    movd            xm5, [r0 + r1 * 2]
-    punpcklwd       xm1, xm5
-    punpcklqdq      xm2, xm1                        ; m2 = [10 9 9 8]
-    vinserti128     m3, m3, xm2, 1                  ; m3 = [10 9 9 8 8 7 7 6]
-    pmaddwd         m4, [r5]
-    pmaddwd         m3, [r5 + 1 * mmsize]
-    paddd           m4, m3
+
+    PROCESS_CHROMA_S_AVX2_W8_4R %1
+    lea             r4, [r3 * 3]
+%ifidn %1,sp
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r4], xm2
+%else
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r4], xm3
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_8x4 sp
+    FILTER_VER_CHROMA_S_AVX2_8x4 ss
+
+%macro FILTER_VER_CHROMA_S_AVX2_12x16 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_vert_%1_12x16, 4, 9, 10
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
 %ifidn %1,sp
-    paddd           m0, m6
-    paddd           m4, m6
-    psrad           m0, 12
-    psrad           m4, 12
+    mova            m9, [pd_526336]
 %else
-    psrad           m0, 6
-    psrad           m4, 6
+    add             r3d, r3d
 %endif
-    packssdw        m0, m4
-    vextracti128    xm4, m0, 1
-    lea             r4, [r3 * 3]
+    lea             r6, [r3 * 3]
+    PROCESS_CHROMA_S_AVX2_W8_16R %1
 %ifidn %1,sp
-    packuswb        xm0, xm4
-    pextrw          [r2], xm0, 0
-    pextrw          [r2 + r3], xm0, 1
-    pextrw          [r2 + 2 * r3], xm0, 4
-    pextrw          [r2 + r4], xm0, 5
-    lea             r2, [r2 + r3 * 4]
-    pextrw          [r2], xm0, 2
-    pextrw          [r2 + r3], xm0, 3
-    pextrw          [r2 + 2 * r3], xm0, 6
-    pextrw          [r2 + r4], xm0, 7
+    add             r2, 8
 %else
-    movd            [r2], xm0
-    pextrd          [r2 + r3], xm0, 1
-    movd            [r2 + 2 * r3], xm4
-    pextrd          [r2 + r4], xm4, 1
-    lea             r2, [r2 + r3 * 4]
-    pextrd          [r2], xm0, 2
-    pextrd          [r2 + r3], xm0, 3
-    pextrd          [r2 + 2 * r3], xm4, 2
-    pextrd          [r2 + r4], xm4, 3
+    add             r2, 16
 %endif
+    add             r0, 16
+    mova            m7, m9
+    PROCESS_CHROMA_AVX2_W4_16R %1
     RET
+%endif
 %endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_2x8 sp
-    FILTER_VER_CHROMA_S_AVX2_2x8 ss
+    FILTER_VER_CHROMA_S_AVX2_12x16 sp
+    FILTER_VER_CHROMA_S_AVX2_12x16 ss
 
-%macro FILTER_VER_CHROMA_S_AVX2_2x16 1
+%macro FILTER_VER_CHROMA_S_AVX2_12x32 1
 %if ARCH_X86_64 == 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_2x16, 4, 6, 9
+cglobal interp_4tap_vert_%1_12x32, 4, 9, 10
     mov             r4d, r4m
     shl             r4d, 6
     add             r1d, r1d
-    sub             r0, r1
 
 %ifdef PIC
     lea             r5, [pw_ChromaCoeffV]
@@ -21886,154 +17168,41 @@ cglobal interp_4tap_vert_%1_2x16, 4, 6, 9
 %endif
 
     lea             r4, [r1 * 3]
-%ifidn %1,sp
-    mova            m6, [pd_526336]
+    sub             r0, r1
+%ifidn %1, sp
+    mova            m9, [pd_526336]
 %else
     add             r3d, r3d
 %endif
-    movd            xm0, [r0]
-    movd            xm1, [r0 + r1]
-    punpcklwd       xm0, xm1
-    movd            xm2, [r0 + r1 * 2]
-    punpcklwd       xm1, xm2
-    punpcklqdq      xm0, xm1                        ; m0 = [2 1 1 0]
-    movd            xm3, [r0 + r4]
-    punpcklwd       xm2, xm3
-    lea             r0, [r0 + 4 * r1]
-    movd            xm4, [r0]
-    punpcklwd       xm3, xm4
-    punpcklqdq      xm2, xm3                        ; m2 = [4 3 3 2]
-    vinserti128     m0, m0, xm2, 1                  ; m0 = [4 3 3 2 2 1 1 0]
-    movd            xm1, [r0 + r1]
-    punpcklwd       xm4, xm1
-    movd            xm3, [r0 + r1 * 2]
-    punpcklwd       xm1, xm3
-    punpcklqdq      xm4, xm1                        ; m4 = [6 5 5 4]
-    vinserti128     m2, m2, xm4, 1                  ; m2 = [6 5 5 4 4 3 3 2]
-    pmaddwd         m0, [r5]
-    pmaddwd         m2, [r5 + 1 * mmsize]
-    paddd           m0, m2
-    movd            xm1, [r0 + r4]
-    punpcklwd       xm3, xm1
-    lea             r0, [r0 + 4 * r1]
-    movd            xm2, [r0]
-    punpcklwd       xm1, xm2
-    punpcklqdq      xm3, xm1                        ; m3 = [8 7 7 6]
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [8 7 7 6 6 5 5 4]
-    movd            xm1, [r0 + r1]
-    punpcklwd       xm2, xm1
-    movd            xm5, [r0 + r1 * 2]
-    punpcklwd       xm1, xm5
-    punpcklqdq      xm2, xm1                        ; m2 = [10 9 9 8]
-    vinserti128     m3, m3, xm2, 1                  ; m3 = [10 9 9 8 8 7 7 6]
-    pmaddwd         m4, [r5]
-    pmaddwd         m3, [r5 + 1 * mmsize]
-    paddd           m4, m3
-    movd            xm1, [r0 + r4]
-    punpcklwd       xm5, xm1
-    lea             r0, [r0 + 4 * r1]
-    movd            xm3, [r0]
-    punpcklwd       xm1, xm3
-    punpcklqdq      xm5, xm1                        ; m5 = [12 11 11 10]
-    vinserti128     m2, m2, xm5, 1                  ; m2 = [12 11 11 10 10 9 9 8]
-    movd            xm1, [r0 + r1]
-    punpcklwd       xm3, xm1
-    movd            xm7, [r0 + r1 * 2]
-    punpcklwd       xm1, xm7
-    punpcklqdq      xm3, xm1                        ; m3 = [14 13 13 12]
-    vinserti128     m5, m5, xm3, 1                  ; m5 = [14 13 13 12 12 11 11 10]
-    pmaddwd         m2, [r5]
-    pmaddwd         m5, [r5 + 1 * mmsize]
-    paddd           m2, m5
-    movd            xm5, [r0 + r4]
-    punpcklwd       xm7, xm5
-    lea             r0, [r0 + 4 * r1]
-    movd            xm1, [r0]
-    punpcklwd       xm5, xm1
-    punpcklqdq      xm7, xm5                        ; m7 = [16 15 15 14]
-    vinserti128     m3, m3, xm7, 1                  ; m3 = [16 15 15 14 14 13 13 12]
-    movd            xm5, [r0 + r1]
-    punpcklwd       xm1, xm5
-    movd            xm8, [r0 + r1 * 2]
-    punpcklwd       xm5, xm8
-    punpcklqdq      xm1, xm5                        ; m1 = [18 17 17 16]
-    vinserti128     m7, m7, xm1, 1                  ; m7 = [18 17 17 16 16 15 15 14]
-    pmaddwd         m3, [r5]
-    pmaddwd         m7, [r5 + 1 * mmsize]
-    paddd           m3, m7
-%ifidn %1,sp
-    paddd           m0, m6
-    paddd           m4, m6
-    paddd           m2, m6
-    paddd           m3, m6
-    psrad           m0, 12
-    psrad           m4, 12
-    psrad           m2, 12
-    psrad           m3, 12
+    lea             r6, [r3 * 3]
+%rep 2
+    PROCESS_CHROMA_S_AVX2_W8_16R %1
+%ifidn %1, sp
+    add             r2, 8
 %else
-    psrad           m0, 6
-    psrad           m4, 6
-    psrad           m2, 6
-    psrad           m3, 6
+    add             r2, 16
 %endif
-    packssdw        m0, m4
-    packssdw        m2, m3
-    lea             r4, [r3 * 3]
-%ifidn %1,sp
-    packuswb        m0, m2
-    vextracti128    xm2, m0, 1
-    pextrw          [r2], xm0, 0
-    pextrw          [r2 + r3], xm0, 1
-    pextrw          [r2 + 2 * r3], xm2, 0
-    pextrw          [r2 + r4], xm2, 1
-    lea             r2, [r2 + r3 * 4]
-    pextrw          [r2], xm0, 2
-    pextrw          [r2 + r3], xm0, 3
-    pextrw          [r2 + 2 * r3], xm2, 2
-    pextrw          [r2 + r4], xm2, 3
-    lea             r2, [r2 + r3 * 4]
-    pextrw          [r2], xm0, 4
-    pextrw          [r2 + r3], xm0, 5
-    pextrw          [r2 + 2 * r3], xm2, 4
-    pextrw          [r2 + r4], xm2, 5
-    lea             r2, [r2 + r3 * 4]
-    pextrw          [r2], xm0, 6
-    pextrw          [r2 + r3], xm0, 7
-    pextrw          [r2 + 2 * r3], xm2, 6
-    pextrw          [r2 + r4], xm2, 7
+    add             r0, 16
+    mova            m7, m9
+    PROCESS_CHROMA_AVX2_W4_16R %1
+    sub             r0, 16
+%ifidn %1, sp
+    lea             r2, [r2 + r3 * 4 - 8]
 %else
-    vextracti128    xm4, m0, 1
-    vextracti128    xm3, m2, 1
-    movd            [r2], xm0
-    pextrd          [r2 + r3], xm0, 1
-    movd            [r2 + 2 * r3], xm4
-    pextrd          [r2 + r4], xm4, 1
-    lea             r2, [r2 + r3 * 4]
-    pextrd          [r2], xm0, 2
-    pextrd          [r2 + r3], xm0, 3
-    pextrd          [r2 + 2 * r3], xm4, 2
-    pextrd          [r2 + r4], xm4, 3
-    lea             r2, [r2 + r3 * 4]
-    movd            [r2], xm2
-    pextrd          [r2 + r3], xm2, 1
-    movd            [r2 + 2 * r3], xm3
-    pextrd          [r2 + r4], xm3, 1
-    lea             r2, [r2 + r3 * 4]
-    pextrd          [r2], xm2, 2
-    pextrd          [r2 + r3], xm2, 3
-    pextrd          [r2 + 2 * r3], xm3, 2
-    pextrd          [r2 + r4], xm3, 3
+    lea             r2, [r2 + r3 * 4 - 16]
 %endif
+%endrep
     RET
 %endif
 %endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_2x16 sp
-    FILTER_VER_CHROMA_S_AVX2_2x16 ss
+    FILTER_VER_CHROMA_S_AVX2_12x32 sp
+    FILTER_VER_CHROMA_S_AVX2_12x32 ss
 
-%macro FILTER_VER_CHROMA_S_AVX2_6x8 1
+%macro FILTER_VER_CHROMA_S_AVX2_16x12 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_6x8, 4, 6, 8
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_vert_%1_16x12, 4, 9, 9
     mov             r4d, r4m
     shl             r4d, 6
     add             r1d, r1d
@@ -22048,11 +17217,12 @@ cglobal interp_4tap_vert_%1_6x8, 4, 6, 8
     lea             r4, [r1 * 3]
     sub             r0, r1
 %ifidn %1,sp
-    mova            m7, [pd_526336]
+    mova            m8, [pd_526336]
 %else
     add             r3d, r3d
 %endif
-
+    lea             r6, [r3 * 3]
+%rep 2
     movu            xm0, [r0]                       ; m0 = row 0
     movu            xm1, [r0 + r1]                  ; m1 = row 1
     punpckhwd       xm2, xm0, xm1
@@ -22069,19 +17239,19 @@ cglobal interp_4tap_vert_%1_6x8, 4, 6, 8
     punpcklwd       xm2, xm3
     vinserti128     m2, m2, xm4, 1
     pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    pmaddwd         m2, [r5]
     paddd           m0, m4
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
+    pmaddwd         m2, [r5]
+    lea             r7, [r0 + r1 * 4]
+    movu            xm4, [r7]                       ; m4 = row 4
     punpckhwd       xm5, xm3, xm4
     punpcklwd       xm3, xm4
     vinserti128     m3, m3, xm5, 1
     pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    pmaddwd         m3, [r5]
     paddd           m1, m5
+    pmaddwd         m3, [r5]
 %ifidn %1,sp
-    paddd           m0, m7
-    paddd           m1, m7
+    paddd           m0, m8
+    paddd           m1, m8
     psrad           m0, 12
     psrad           m1, 12
 %else
@@ -22090,14 +17260,14 @@ cglobal interp_4tap_vert_%1_6x8, 4, 6, 8
 %endif
     packssdw        m0, m1
 
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    movu            xm5, [r7 + r1]                  ; m5 = row 5
     punpckhwd       xm6, xm4, xm5
     punpcklwd       xm4, xm5
     vinserti128     m4, m4, xm6, 1
     pmaddwd         m6, m4, [r5 + 1 * mmsize]
     paddd           m2, m6
     pmaddwd         m4, [r5]
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
     punpckhwd       xm1, xm5, xm6
     punpcklwd       xm5, xm6
     vinserti128     m5, m5, xm1, 1
@@ -22105,8 +17275,8 @@ cglobal interp_4tap_vert_%1_6x8, 4, 6, 8
     pmaddwd         m5, [r5]
     paddd           m3, m1
 %ifidn %1,sp
-    paddd           m2, m7
-    paddd           m3, m7
+    paddd           m2, m8
+    paddd           m3, m8
     psrad           m2, 12
     psrad           m3, 12
 %else
@@ -22114,42 +17284,36 @@ cglobal interp_4tap_vert_%1_6x8, 4, 6, 8
     psrad           m3, 6
 %endif
     packssdw        m2, m3
-
-    movu            xm1, [r0 + r4]                  ; m1 = row 7
-    punpckhwd       xm3, xm6, xm1
-    punpcklwd       xm6, xm1
-    vinserti128     m6, m6, xm3, 1
-    pmaddwd         m3, m6, [r5 + 1 * mmsize]
-    pmaddwd         m6, [r5]
-    paddd           m4, m3
-
-    lea             r4, [r3 * 3]
 %ifidn %1,sp
     packuswb        m0, m2
+    mova            m3, [interp8_hps_shuf]
+    vpermd          m0, m3, m0
     vextracti128    xm2, m0, 1
-    movd            [r2], xm0
-    pextrw          [r2 + 4], xm2, 0
-    pextrd          [r2 + r3], xm0, 1
-    pextrw          [r2 + r3 + 4], xm2, 2
-    pextrd          [r2 + r3 * 2], xm0, 2
-    pextrw          [r2 + r3 * 2 + 4], xm2, 4
-    pextrd          [r2 + r4], xm0, 3
-    pextrw          [r2 + r4 + 4], xm2, 6
-%else
     movq            [r2], xm0
     movhps          [r2 + r3], xm0
     movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r4], xm2
+    movhps          [r2 + r6], xm2
+%else
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    movu            [r2], xm0
     vextracti128    xm0, m0, 1
     vextracti128    xm3, m2, 1
-    movd            [r2 + 8], xm0
-    pextrd          [r2 + r3 + 8], xm0, 2
-    movd            [r2 + r3 * 2 + 8], xm3
-    pextrd          [r2 + r4 + 8], xm3, 2
+    movu            [r2 + r3], xm0
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
 %endif
-    lea             r2, [r2 + r3 * 4]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm0, [r0]                       ; m0 = row 8
+    lea             r8, [r2 + r3 * 4]
+
+    movu            xm1, [r7 + r4]                  ; m1 = row 7
+    punpckhwd       xm0, xm6, xm1
+    punpcklwd       xm6, xm1
+    vinserti128     m6, m6, xm0, 1
+    pmaddwd         m0, m6, [r5 + 1 * mmsize]
+    pmaddwd         m6, [r5]
+    paddd           m4, m0
+    lea             r7, [r7 + r1 * 4]
+    movu            xm0, [r7]                       ; m0 = row 8
     punpckhwd       xm2, xm1, xm0
     punpcklwd       xm1, xm0
     vinserti128     m1, m1, xm2, 1
@@ -22157,8 +17321,8 @@ cglobal interp_4tap_vert_%1_6x8, 4, 6, 8
     pmaddwd         m1, [r5]
     paddd           m5, m2
 %ifidn %1,sp
-    paddd           m4, m7
-    paddd           m5, m7
+    paddd           m4, m8
+    paddd           m5, m8
     psrad           m4, 12
     psrad           m5, 12
 %else
@@ -22167,22 +17331,24 @@ cglobal interp_4tap_vert_%1_6x8, 4, 6, 8
 %endif
     packssdw        m4, m5
 
-    movu            xm2, [r0 + r1]                  ; m2 = row 9
+    movu            xm2, [r7 + r1]                  ; m2 = row 9
     punpckhwd       xm5, xm0, xm2
     punpcklwd       xm0, xm2
     vinserti128     m0, m0, xm5, 1
-    pmaddwd         m0, [r5 + 1 * mmsize]
-    paddd           m6, m0
-    movu            xm5, [r0 + r1 * 2]              ; m5 = row 10
-    punpckhwd       xm0, xm2, xm5
+    pmaddwd         m5, m0, [r5 + 1 * mmsize]
+    paddd           m6, m5
+    pmaddwd         m0, [r5]
+    movu            xm5, [r7 + r1 * 2]              ; m5 = row 10
+    punpckhwd       xm7, xm2, xm5
     punpcklwd       xm2, xm5
-    vinserti128     m2, m2, xm0, 1
-    pmaddwd         m2, [r5 + 1 * mmsize]
-    paddd           m1, m2
+    vinserti128     m2, m2, xm7, 1
+    pmaddwd         m7, m2, [r5 + 1 * mmsize]
+    paddd           m1, m7
+    pmaddwd         m2, [r5]
 
 %ifidn %1,sp
-    paddd           m6, m7
-    paddd           m1, m7
+    paddd           m6, m8
+    paddd           m1, m8
     psrad           m6, 12
     psrad           m1, 12
 %else
@@ -22192,37 +17358,105 @@ cglobal interp_4tap_vert_%1_6x8, 4, 6, 8
     packssdw        m6, m1
 %ifidn %1,sp
     packuswb        m4, m6
+    vpermd          m4, m3, m4
     vextracti128    xm6, m4, 1
-    movd            [r2], xm4
-    pextrw          [r2 + 4], xm6, 0
-    pextrd          [r2 + r3], xm4, 1
-    pextrw          [r2 + r3 + 4], xm6, 2
-    pextrd          [r2 + r3 * 2], xm4, 2
-    pextrw          [r2 + r3 * 2 + 4], xm6, 4
-    pextrd          [r2 + r4], xm4, 3
-    pextrw          [r2 + r4 + 4], xm6, 6
+    movq            [r8], xm4
+    movhps          [r8 + r3], xm4
+    movq            [r8 + r3 * 2], xm6
+    movhps          [r8 + r6], xm6
 %else
-    movq            [r2], xm4
-    movhps          [r2 + r3], xm4
-    movq            [r2 + r3 * 2], xm6
-    movhps          [r2 + r4], xm6
-    vextracti128    xm5, m4, 1
+    vpermq          m4, m4, 11011000b
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm7, m4, 1
     vextracti128    xm1, m6, 1
-    movd            [r2 + 8], xm5
-    pextrd          [r2 + r3 + 8], xm5, 2
-    movd            [r2 + r3 * 2 + 8], xm1
-    pextrd          [r2 + r4 + 8], xm1, 2
+    movu            [r8], xm4
+    movu            [r8 + r3], xm7
+    movu            [r8 + r3 * 2], xm6
+    movu            [r8 + r6], xm1
+%endif
+    lea             r8, [r8 + r3 * 4]
+
+    movu            xm7, [r7 + r4]                  ; m7 = row 11
+    punpckhwd       xm1, xm5, xm7
+    punpcklwd       xm5, xm7
+    vinserti128     m5, m5, xm1, 1
+    pmaddwd         m1, m5, [r5 + 1 * mmsize]
+    paddd           m0, m1
+    pmaddwd         m5, [r5]
+    lea             r7, [r7 + r1 * 4]
+    movu            xm1, [r7]                       ; m1 = row 12
+    punpckhwd       xm4, xm7, xm1
+    punpcklwd       xm7, xm1
+    vinserti128     m7, m7, xm4, 1
+    pmaddwd         m4, m7, [r5 + 1 * mmsize]
+    paddd           m2, m4
+    pmaddwd         m7, [r5]
+%ifidn %1,sp
+    paddd           m0, m8
+    paddd           m2, m8
+    psrad           m0, 12
+    psrad           m2, 12
+%else
+    psrad           m0, 6
+    psrad           m2, 6
+%endif
+    packssdw        m0, m2
+
+    movu            xm4, [r7 + r1]                  ; m4 = row 13
+    punpckhwd       xm2, xm1, xm4
+    punpcklwd       xm1, xm4
+    vinserti128     m1, m1, xm2, 1
+    pmaddwd         m1, [r5 + 1 * mmsize]
+    paddd           m5, m1
+    movu            xm2, [r7 + r1 * 2]              ; m2 = row 14
+    punpckhwd       xm6, xm4, xm2
+    punpcklwd       xm4, xm2
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m4, [r5 + 1 * mmsize]
+    paddd           m7, m4
+%ifidn %1,sp
+    paddd           m5, m8
+    paddd           m7, m8
+    psrad           m5, 12
+    psrad           m7, 12
+%else
+    psrad           m5, 6
+    psrad           m7, 6
+%endif
+    packssdw        m5, m7
+%ifidn %1,sp
+    packuswb        m0, m5
+    vpermd          m0, m3, m0
+    vextracti128    xm5, m0, 1
+    movq            [r8], xm0
+    movhps          [r8 + r3], xm0
+    movq            [r8 + r3 * 2], xm5
+    movhps          [r8 + r6], xm5
+    add             r2, 8
+%else
+    vpermq          m0, m0, 11011000b
+    vpermq          m5, m5, 11011000b
+    vextracti128    xm7, m0, 1
+    vextracti128    xm6, m5, 1
+    movu            [r8], xm0
+    movu            [r8 + r3], xm7
+    movu            [r8 + r3 * 2], xm5
+    movu            [r8 + r6], xm6
+    add             r2, 16
 %endif
+    add             r0, 16
+%endrep
     RET
+%endif
 %endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_6x8 sp
-    FILTER_VER_CHROMA_S_AVX2_6x8 ss
+    FILTER_VER_CHROMA_S_AVX2_16x12 sp
+    FILTER_VER_CHROMA_S_AVX2_16x12 ss
 
-%macro FILTER_VER_CHROMA_S_AVX2_6x16 1
+%macro FILTER_VER_CHROMA_S_AVX2_8x12 1
 %if ARCH_X86_64 == 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_6x16, 4, 7, 9
+cglobal interp_4tap_vert_%1_8x12, 4, 7, 9
     mov             r4d, r4m
     shl             r4d, 6
     add             r1d, r1d
@@ -22305,28 +17539,25 @@ cglobal interp_4tap_vert_%1_6x16, 4, 7, 9
     packssdw        m2, m3
 %ifidn %1,sp
     packuswb        m0, m2
+    mova            m3, [interp8_hps_shuf]
+    vpermd          m0, m3, m0
     vextracti128    xm2, m0, 1
-    movd            [r2], xm0
-    pextrw          [r2 + 4], xm2, 0
-    pextrd          [r2 + r3], xm0, 1
-    pextrw          [r2 + r3 + 4], xm2, 2
-    pextrd          [r2 + r3 * 2], xm0, 2
-    pextrw          [r2 + r3 * 2 + 4], xm2, 4
-    pextrd          [r2 + r6], xm0, 3
-    pextrw          [r2 + r6 + 4], xm2, 6
-%else
     movq            [r2], xm0
     movhps          [r2 + r3], xm0
     movq            [r2 + r3 * 2], xm2
     movhps          [r2 + r6], xm2
+%else
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    movu            [r2], xm0
     vextracti128    xm0, m0, 1
     vextracti128    xm3, m2, 1
-    movd            [r2 + 8], xm0
-    pextrd          [r2 + r3 + 8], xm0, 2
-    movd            [r2 + r3 * 2 + 8], xm3
-    pextrd          [r2 + r6 + 8], xm3, 2
+    movu            [r2 + r3], xm0
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
 %endif
     lea             r2, [r2 + r3 * 4]
+
     movu            xm1, [r0 + r4]                  ; m1 = row 7
     punpckhwd       xm0, xm6, xm1
     punpcklwd       xm6, xm1
@@ -22380,28 +17611,24 @@ cglobal interp_4tap_vert_%1_6x16, 4, 7, 9
     packssdw        m6, m1
 %ifidn %1,sp
     packuswb        m4, m6
+    vpermd          m4, m3, m4
     vextracti128    xm6, m4, 1
-    movd            [r2], xm4
-    pextrw          [r2 + 4], xm6, 0
-    pextrd          [r2 + r3], xm4, 1
-    pextrw          [r2 + r3 + 4], xm6, 2
-    pextrd          [r2 + r3 * 2], xm4, 2
-    pextrw          [r2 + r3 * 2 + 4], xm6, 4
-    pextrd          [r2 + r6], xm4, 3
-    pextrw          [r2 + r6 + 4], xm6, 6
-%else
     movq            [r2], xm4
     movhps          [r2 + r3], xm4
     movq            [r2 + r3 * 2], xm6
     movhps          [r2 + r6], xm6
-    vextracti128    xm4, m4, 1
+%else
+    vpermq          m4, m4, 11011000b
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm7, m4, 1
     vextracti128    xm1, m6, 1
-    movd            [r2 + 8], xm4
-    pextrd          [r2 + r3 + 8], xm4, 2
-    movd            [r2 + r3 * 2 + 8], xm1
-    pextrd          [r2 + r6 + 8], xm1, 2
+    movu            [r2], xm4
+    movu            [r2 + r3], xm7
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r6], xm1
 %endif
     lea             r2, [r2 + r3 * 4]
+
     movu            xm7, [r0 + r4]                  ; m7 = row 11
     punpckhwd       xm1, xm5, xm7
     punpcklwd       xm5, xm7
@@ -22432,16 +17659,14 @@ cglobal interp_4tap_vert_%1_6x16, 4, 7, 9
     punpckhwd       xm2, xm1, xm4
     punpcklwd       xm1, xm4
     vinserti128     m1, m1, xm2, 1
-    pmaddwd         m2, m1, [r5 + 1 * mmsize]
-    paddd           m5, m2
-    pmaddwd         m1, [r5]
+    pmaddwd         m1, [r5 + 1 * mmsize]
+    paddd           m5, m1
     movu            xm2, [r0 + r1 * 2]              ; m2 = row 14
     punpckhwd       xm6, xm4, xm2
     punpcklwd       xm4, xm2
     vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m7, m6
-    pmaddwd         m4, [r5]
+    pmaddwd         m4, [r5 + 1 * mmsize]
+    paddd           m7, m4
 %ifidn %1,sp
     paddd           m5, m8
     paddd           m7, m8
@@ -22453,980 +17678,1344 @@ cglobal interp_4tap_vert_%1_6x16, 4, 7, 9
 %endif
     packssdw        m5, m7
 %ifidn %1,sp
-    packuswb        m0, m5
-    vextracti128    xm5, m0, 1
-    movd            [r2], xm0
-    pextrw          [r2 + 4], xm5, 0
-    pextrd          [r2 + r3], xm0, 1
-    pextrw          [r2 + r3 + 4], xm5, 2
-    pextrd          [r2 + r3 * 2], xm0, 2
-    pextrw          [r2 + r3 * 2 + 4], xm5, 4
-    pextrd          [r2 + r6], xm0, 3
-    pextrw          [r2 + r6 + 4], xm5, 6
+    packuswb        m0, m5
+    vpermd          m0, m3, m0
+    vextracti128    xm5, m0, 1
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm5
+    movhps          [r2 + r6], xm5
+%else
+    vpermq          m0, m0, 11011000b
+    vpermq          m5, m5, 11011000b
+    vextracti128    xm7, m0, 1
+    vextracti128    xm6, m5, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm7
+    movu            [r2 + r3 * 2], xm5
+    movu            [r2 + r6], xm6
+%endif
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_8x12 sp
+    FILTER_VER_CHROMA_S_AVX2_8x12 ss
+
+%macro FILTER_VER_CHROMA_S_AVX2_16x4 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_16x4, 4, 7, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,sp
+    mova            m7, [pd_526336]
 %else
+    add             r3d, r3d
+%endif
+%rep 2
+    PROCESS_CHROMA_S_AVX2_W8_4R %1
+    lea             r6, [r3 * 3]
+%ifidn %1,sp
     movq            [r2], xm0
     movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm5
-    movhps          [r2 + r6], xm5
-    vextracti128    xm0, m0, 1
-    vextracti128    xm7, m5, 1
-    movd            [r2 + 8], xm0
-    pextrd          [r2 + r3 + 8], xm0, 2
-    movd            [r2 + r3 * 2 + 8], xm7
-    pextrd          [r2 + r6 + 8], xm7, 2
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r6], xm2
+    add             r2, 8
+%else
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
+    add             r2, 16
 %endif
-    lea             r2, [r2 + r3 * 4]
+    lea             r6, [4 * r1 - 16]
+    sub             r0, r6
+%endrep
+    RET
+%endmacro
 
-    movu            xm6, [r0 + r4]                  ; m6 = row 15
-    punpckhwd       xm5, xm2, xm6
-    punpcklwd       xm2, xm6
-    vinserti128     m2, m2, xm5, 1
-    pmaddwd         m5, m2, [r5 + 1 * mmsize]
-    paddd           m1, m5
+    FILTER_VER_CHROMA_S_AVX2_16x4 sp
+    FILTER_VER_CHROMA_S_AVX2_16x4 ss
+
+%macro PROCESS_CHROMA_S_AVX2_W8_8R 1
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    paddd           m0, m4
     pmaddwd         m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm0, [r0]                       ; m0 = row 16
-    punpckhwd       xm5, xm6, xm0
-    punpcklwd       xm6, xm0
-    vinserti128     m6, m6, xm5, 1
-    pmaddwd         m5, m6, [r5 + 1 * mmsize]
-    paddd           m4, m5
-    pmaddwd         m6, [r5]
+    lea             r7, [r0 + r1 * 4]
+    movu            xm4, [r7]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    paddd           m1, m5
+    pmaddwd         m3, [r5]
 %ifidn %1,sp
-    paddd           m1, m8
-    paddd           m4, m8
+    paddd           m0, m7
+    paddd           m1, m7
+    psrad           m0, 12
     psrad           m1, 12
-    psrad           m4, 12
 %else
+    psrad           m0, 6
     psrad           m1, 6
-    psrad           m4, 6
 %endif
-    packssdw        m1, m4
+    packssdw        m0, m1
 
-    movu            xm5, [r0 + r1]                  ; m5 = row 17
-    punpckhwd       xm4, xm0, xm5
-    punpcklwd       xm0, xm5
-    vinserti128     m0, m0, xm4, 1
-    pmaddwd         m0, [r5 + 1 * mmsize]
-    paddd           m2, m0
-    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
-    punpckhwd       xm0, xm5, xm4
-    punpcklwd       xm5, xm4
-    vinserti128     m5, m5, xm0, 1
-    pmaddwd         m5, [r5 + 1 * mmsize]
-    paddd           m6, m5
+    movu            xm5, [r7 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m2, m6
+    pmaddwd         m4, [r5]
+    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm1, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm1, 1
+    pmaddwd         m1, m5, [r5 + 1 * mmsize]
+    pmaddwd         m5, [r5]
+    paddd           m3, m1
 %ifidn %1,sp
-    paddd           m2, m8
-    paddd           m6, m8
+    paddd           m2, m7
+    paddd           m3, m7
     psrad           m2, 12
-    psrad           m6, 12
+    psrad           m3, 12
 %else
     psrad           m2, 6
-    psrad           m6, 6
+    psrad           m3, 6
 %endif
-    packssdw        m2, m6
+    packssdw        m2, m3
 %ifidn %1,sp
-    packuswb        m1, m2
-    vextracti128    xm2, m1, 1
-    movd            [r2], xm1
-    pextrw          [r2 + 4], xm2, 0
-    pextrd          [r2 + r3], xm1, 1
-    pextrw          [r2 + r3 + 4], xm2, 2
-    pextrd          [r2 + r3 * 2], xm1, 2
-    pextrw          [r2 + r3 * 2 + 4], xm2, 4
-    pextrd          [r2 + r6], xm1, 3
-    pextrw          [r2 + r6 + 4], xm2, 6
-%else
-    movq            [r2], xm1
-    movhps          [r2 + r3], xm1
+    packuswb        m0, m2
+    mova            m3, [interp8_hps_shuf]
+    vpermd          m0, m3, m0
+    vextracti128    xm2, m0, 1
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
     movq            [r2 + r3 * 2], xm2
     movhps          [r2 + r6], xm2
-    vextracti128    xm4, m1, 1
-    vextracti128    xm6, m2, 1
-    movd            [r2 + 8], xm4
-    pextrd          [r2 + r3 + 8], xm4, 2
-    movd            [r2 + r3 * 2 + 8], xm6
-    pextrd          [r2 + r6 + 8], xm6, 2
-%endif
-    RET
+%else
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    movu            [r2], xm0
+    vextracti128    xm0, m0, 1
+    vextracti128    xm3, m2, 1
+    movu            [r2 + r3], xm0
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
 %endif
-%endmacro
-
-    FILTER_VER_CHROMA_S_AVX2_6x16 sp
-    FILTER_VER_CHROMA_S_AVX2_6x16 ss
-
-;---------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vertical_ss_%1x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;---------------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_SS_W2_4R 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_ss_%1x%2, 5, 6, 5
-
-    add       r1d, r1d
-    add       r3d, r3d
-    sub       r0, r1
-    shl       r4d, 5
+    lea             r8, [r2 + r3 * 4]
 
-%ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
-    lea       r5, [r5 + r4]
+    movu            xm1, [r7 + r4]                  ; m1 = row 7
+    punpckhwd       xm0, xm6, xm1
+    punpcklwd       xm6, xm1
+    vinserti128     m6, m6, xm0, 1
+    pmaddwd         m0, m6, [r5 + 1 * mmsize]
+    pmaddwd         m6, [r5]
+    paddd           m4, m0
+    lea             r7, [r7 + r1 * 4]
+    movu            xm0, [r7]                       ; m0 = row 8
+    punpckhwd       xm2, xm1, xm0
+    punpcklwd       xm1, xm0
+    vinserti128     m1, m1, xm2, 1
+    pmaddwd         m2, m1, [r5 + 1 * mmsize]
+    pmaddwd         m1, [r5]
+    paddd           m5, m2
+%ifidn %1,sp
+    paddd           m4, m7
+    paddd           m5, m7
+    psrad           m4, 12
+    psrad           m5, 12
 %else
-    lea       r5, [tab_ChromaCoeffV + r4]
+    psrad           m4, 6
+    psrad           m5, 6
 %endif
+    packssdw        m4, m5
 
-    mov       r4d, (%2/4)
-
-.loopH:
-    PROCESS_CHROMA_SP_W2_4R r5
-
-    psrad     m0, 6
-    psrad     m2, 6
-
-    packssdw  m0, m2
-
-    movd      [r2], m0
-    pextrd    [r2 + r3], m0, 1
-    lea       r2, [r2 + 2 * r3]
-    pextrd    [r2], m0, 2
-    pextrd    [r2 + r3], m0, 3
-
-    lea       r2, [r2 + 2 * r3]
-
-    dec       r4d
-    jnz       .loopH
-
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_SS_W2_4R 2, 4
-    FILTER_VER_CHROMA_SS_W2_4R 2, 8
-
-    FILTER_VER_CHROMA_SS_W2_4R 2, 16
-
-;---------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ss_4x2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;---------------------------------------------------------------------------------------------------------------
-INIT_XMM sse2
-cglobal interp_4tap_vert_ss_4x2, 5, 6, 4
-
-    add        r1d, r1d
-    add        r3d, r3d
-    sub        r0, r1
-    shl        r4d, 5
+    movu            xm2, [r7 + r1]                  ; m2 = row 9
+    punpckhwd       xm5, xm0, xm2
+    punpcklwd       xm0, xm2
+    vinserti128     m0, m0, xm5, 1
+    pmaddwd         m0, [r5 + 1 * mmsize]
+    paddd           m6, m0
+    movu            xm5, [r7 + r1 * 2]              ; m5 = row 10
+    punpckhwd       xm0, xm2, xm5
+    punpcklwd       xm2, xm5
+    vinserti128     m2, m2, xm0, 1
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m1, m2
 
-%ifdef PIC
-    lea        r5, [tab_ChromaCoeffV]
-    lea        r5, [r5 + r4]
+%ifidn %1,sp
+    paddd           m6, m7
+    paddd           m1, m7
+    psrad           m6, 12
+    psrad           m1, 12
 %else
-    lea        r5, [tab_ChromaCoeffV + r4]
+    psrad           m6, 6
+    psrad           m1, 6
 %endif
+    packssdw        m6, m1
+%ifidn %1,sp
+    packuswb        m4, m6
+    vpermd          m4, m3, m4
+    vextracti128    xm6, m4, 1
+    movq            [r8], xm4
+    movhps          [r8 + r3], xm4
+    movq            [r8 + r3 * 2], xm6
+    movhps          [r8 + r6], xm6
+%else
+    vpermq          m4, m4, 11011000b
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm7, m4, 1
+    vextracti128    xm1, m6, 1
+    movu            [r8], xm4
+    movu            [r8 + r3], xm7
+    movu            [r8 + r3 * 2], xm6
+    movu            [r8 + r6], xm1
+%endif
+%endmacro
 
-    movq       m0, [r0]
-    movq       m1, [r0 + r1]
-    punpcklwd  m0, m1                          ;m0=[0 1]
-    pmaddwd    m0, [r5 + 0 *16]                ;m0=[0+1]  Row1
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m2, [r0]
-    punpcklwd  m1, m2                          ;m1=[1 2]
-    pmaddwd    m1, [r5 + 0 *16]                ;m1=[1+2]  Row2
-
-    movq       m3, [r0 + r1]
-    punpcklwd  m2, m3                          ;m4=[2 3]
-    pmaddwd    m2, [r5 + 1 * 16]
-    paddd      m0, m2                          ;m0=[0+1+2+3]  Row1 done
-    psrad      m0, 6
-
-    movq       m2, [r0 + 2 * r1]
-    punpcklwd  m3, m2                          ;m5=[3 4]
-    pmaddwd    m3, [r5 + 1 * 16]
-    paddd      m1, m3                          ;m1=[1+2+3+4]  Row2 done
-    psrad      m1, 6
-
-    packssdw   m0, m1
-
-    movlps     [r2], m0
-    movhps     [r2 + r3], m0
-
-    RET
-
-;-------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vertical_ss_6x8(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_SS_W6_H4 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_ss_6x%2, 5, 7, 6
-
-    add       r1d, r1d
-    add       r3d, r3d
-    sub       r0, r1
-    shl       r4d, 5
+%macro FILTER_VER_CHROMA_S_AVX2_Nx8 2
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_vert_%1_%2x8, 4, 9, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
 
 %ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
-    lea       r6, [r5 + r4]
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
 %else
-    lea       r6, [tab_ChromaCoeffV + r4]
+    lea             r5, [pw_ChromaCoeffV + r4]
 %endif
 
-    mov       r4d, %2/4
-
-.loopH:
-    PROCESS_CHROMA_SP_W4_4R
-
-    psrad     m0, 6
-    psrad     m1, 6
-    psrad     m2, 6
-    psrad     m3, 6
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-
-    movlps    [r2], m0
-    movhps    [r2 + r3], m0
-    lea       r5, [r2 + 2 * r3]
-    movlps    [r5], m2
-    movhps    [r5 + r3], m2
-
-    lea       r5, [4 * r1 - 2 * 4]
-    sub       r0, r5
-    add       r2, 2 * 4
-
-    PROCESS_CHROMA_SP_W2_4R r6
-
-    psrad     m0, 6
-    psrad     m2, 6
-
-    packssdw  m0, m2
-
-    movd      [r2], m0
-    pextrd    [r2 + r3], m0, 1
-    lea       r2, [r2 + 2 * r3]
-    pextrd    [r2], m0, 2
-    pextrd    [r2 + r3], m0, 3
-
-    sub       r0, 2 * 4
-    lea       r2, [r2 + 2 * r3 - 2 * 4]
-
-    dec       r4d
-    jnz       .loopH
-
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,sp
+    mova            m7, [pd_526336]
+%else
+    add             r3d, r3d
+%endif
+    lea             r6, [r3 * 3]
+%rep %2 / 8
+    PROCESS_CHROMA_S_AVX2_W8_8R %1
+%ifidn %1,sp
+    add             r2, 8
+%else
+    add             r2, 16
+%endif
+    add             r0, 16
+%endrep
     RET
+%endif
 %endmacro
 
-    FILTER_VER_CHROMA_SS_W6_H4 6, 8
-
-    FILTER_VER_CHROMA_SS_W6_H4 6, 16
-
-
-;----------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ss_8x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;----------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_SS_W8_H2 2
-INIT_XMM sse2
-cglobal interp_4tap_vert_ss_%1x%2, 5, 6, 7
+    FILTER_VER_CHROMA_S_AVX2_Nx8 sp, 32
+    FILTER_VER_CHROMA_S_AVX2_Nx8 sp, 16
+    FILTER_VER_CHROMA_S_AVX2_Nx8 ss, 32
+    FILTER_VER_CHROMA_S_AVX2_Nx8 ss, 16
 
-    add       r1d, r1d
-    add       r3d, r3d
-    sub       r0, r1
-    shl       r4d, 5
+%macro FILTER_VER_CHROMA_S_AVX2_8x2 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_8x2, 4, 6, 6
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
 
 %ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
-    lea       r5, [r5 + r4]
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
 %else
-    lea       r5, [tab_ChromaCoeffV + r4]
+    lea             r5, [pw_ChromaCoeffV + r4]
 %endif
 
-    mov       r4d, %2/2
-.loopH:
-    PROCESS_CHROMA_SP_W8_2R
-
-    psrad     m0, 6
-    psrad     m1, 6
-    psrad     m2, 6
-    psrad     m3, 6
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-
-    movu      [r2], m0
-    movu      [r2 + r3], m2
-
-    lea       r2, [r2 + 2 * r3]
-
-    dec       r4d
-    jnz       .loopH
-
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_SS_W8_H2 8, 2
-    FILTER_VER_CHROMA_SS_W8_H2 8, 4
-    FILTER_VER_CHROMA_SS_W8_H2 8, 6
-    FILTER_VER_CHROMA_SS_W8_H2 8, 8
-    FILTER_VER_CHROMA_SS_W8_H2 8, 16
-    FILTER_VER_CHROMA_SS_W8_H2 8, 32
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,sp
+    mova            m5, [pd_526336]
+%else
+    add             r3d, r3d
+%endif
 
-    FILTER_VER_CHROMA_SS_W8_H2 8, 12
-    FILTER_VER_CHROMA_SS_W8_H2 8, 64
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m0, m2
+    movu            xm4, [r0 + r1 * 4]              ; m4 = row 4
+    punpckhwd       xm2, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm2, 1
+    pmaddwd         m3, [r5 + 1 * mmsize]
+    paddd           m1, m3
+%ifidn %1,sp
+    paddd           m0, m5
+    paddd           m1, m5
+    psrad           m0, 12
+    psrad           m1, 12
+%else
+    psrad           m0, 6
+    psrad           m1, 6
+%endif
+    packssdw        m0, m1
+%ifidn %1,sp
+    vextracti128    xm1, m0, 1
+    packuswb        xm0, xm1
+    pshufd          xm0, xm0, 11011000b
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+%else
+    vpermq          m0, m0, 11011000b
+    vextracti128    xm1, m0, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+%endif
+    RET
+%endmacro
 
-;-----------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_ss_%1x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_LUMA_SS 2
-INIT_XMM sse2
-cglobal interp_8tap_vert_ss_%1x%2, 5, 7, 7 ,0-gprsize
+    FILTER_VER_CHROMA_S_AVX2_8x2 sp
+    FILTER_VER_CHROMA_S_AVX2_8x2 ss
 
-    add        r1d, r1d
-    add        r3d, r3d
-    lea        r5, [3 * r1]
-    sub        r0, r5
-    shl        r4d, 6
+%macro FILTER_VER_CHROMA_S_AVX2_8x6 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_8x6, 4, 6, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
 
 %ifdef PIC
-    lea        r5, [tab_LumaCoeffV]
-    lea        r6, [r5 + r4]
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
 %else
-    lea        r6, [tab_LumaCoeffV + r4]
+    lea             r5, [pw_ChromaCoeffV + r4]
 %endif
 
-    mov        dword [rsp], %2/4
-.loopH:
-    mov        r4d, (%1/4)
-.loopW:
-    movq       m0, [r0]
-    movq       m1, [r0 + r1]
-    punpcklwd  m0, m1                          ;m0=[0 1]
-    pmaddwd    m0, [r6 + 0 *16]                ;m0=[0+1]  Row1
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,sp
+    mova            m7, [pd_526336]
+%else
+    add             r3d, r3d
+%endif
 
-    lea        r0, [r0 + 2 * r1]
-    movq       m4, [r0]
-    punpcklwd  m1, m4                          ;m1=[1 2]
-    pmaddwd    m1, [r6 + 0 *16]                ;m1=[1+2]  Row2
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    pmaddwd         m2, [r5]
+    paddd           m0, m4
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    pmaddwd         m3, [r5]
+    paddd           m1, m5
+%ifidn %1,sp
+    paddd           m0, m7
+    paddd           m1, m7
+    psrad           m0, 12
+    psrad           m1, 12
+%else
+    psrad           m0, 6
+    psrad           m1, 6
+%endif
+    packssdw        m0, m1
 
-    movq       m5, [r0 + r1]
-    punpcklwd  m4, m5                          ;m4=[2 3]
-    pmaddwd    m2, m4, [r6 + 0 *16]            ;m2=[2+3]  Row3
-    pmaddwd    m4, [r6 + 1 * 16]
-    paddd      m0, m4                          ;m0=[0+1+2+3]  Row1
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m2, m6
+    pmaddwd         m4, [r5]
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm1, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm1, 1
+    pmaddwd         m1, m5, [r5 + 1 * mmsize]
+    pmaddwd         m5, [r5]
+    paddd           m3, m1
+%ifidn %1,sp
+    paddd           m2, m7
+    paddd           m3, m7
+    psrad           m2, 12
+    psrad           m3, 12
+%else
+    psrad           m2, 6
+    psrad           m3, 6
+%endif
+    packssdw        m2, m3
 
-    lea        r0, [r0 + 2 * r1]
-    movq       m4, [r0]
-    punpcklwd  m5, m4                          ;m5=[3 4]
-    pmaddwd    m3, m5, [r6 + 0 *16]            ;m3=[3+4]  Row4
-    pmaddwd    m5, [r6 + 1 * 16]
-    paddd      m1, m5                          ;m1 = [1+2+3+4]  Row2
+    movu            xm1, [r0 + r4]                  ; m1 = row 7
+    punpckhwd       xm3, xm6, xm1
+    punpcklwd       xm6, xm1
+    vinserti128     m6, m6, xm3, 1
+    pmaddwd         m6, [r5 + 1 * mmsize]
+    paddd           m4, m6
+    movu            xm6, [r0 + r1 * 4]              ; m6 = row 8
+    punpckhwd       xm3, xm1, xm6
+    punpcklwd       xm1, xm6
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5 + 1 * mmsize]
+    paddd           m5, m1
+%ifidn %1,sp
+    paddd           m4, m7
+    paddd           m5, m7
+    psrad           m4, 12
+    psrad           m5, 12
+%else
+    psrad           m4, 6
+    psrad           m5, 6
+%endif
+    packssdw        m4, m5
+    lea             r4, [r3 * 3]
+%ifidn %1,sp
+    packuswb        m0, m2
+    mova            m3, [interp8_hps_shuf]
+    vpermd          m0, m3, m0
+    vextracti128    xm2, m0, 1
+    vextracti128    xm5, m4, 1
+    packuswb        xm4, xm5
+    pshufd          xm4, xm4, 11011000b
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r4], xm2
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm4
+    movhps          [r2 + r3], xm4
+%else
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    vpermq          m4, m4, 11011000b
+    movu            [r2], xm0
+    vextracti128    xm0, m0, 1
+    vextracti128    xm3, m2, 1
+    vextracti128    xm5, m4, 1
+    movu            [r2 + r3], xm0
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r4], xm3
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm4
+    movu            [r2 + r3], xm5
+%endif
+    RET
+%endmacro
 
-    movq       m5, [r0 + r1]
-    punpcklwd  m4, m5                          ;m4=[4 5]
-    pmaddwd    m6, m4, [r6 + 1 * 16]
-    paddd      m2, m6                          ;m2=[2+3+4+5]  Row3
-    pmaddwd    m4, [r6 + 2 * 16]
-    paddd      m0, m4                          ;m0=[0+1+2+3+4+5]  Row1
+    FILTER_VER_CHROMA_S_AVX2_8x6 sp
+    FILTER_VER_CHROMA_S_AVX2_8x6 ss
 
-    lea        r0, [r0 + 2 * r1]
-    movq       m4, [r0]
-    punpcklwd  m5, m4                          ;m5=[5 6]
-    pmaddwd    m6, m5, [r6 + 1 * 16]
-    paddd      m3, m6                          ;m3=[3+4+5+6]  Row4
-    pmaddwd    m5, [r6 + 2 * 16]
-    paddd      m1, m5                          ;m1=[1+2+3+4+5+6]  Row2
+%macro FILTER_VER_CHROMA_S_AVX2_8xN 2
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_vert_%1_8x%2, 4, 7, 9
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
 
-    movq       m5, [r0 + r1]
-    punpcklwd  m4, m5                          ;m4=[6 7]
-    pmaddwd    m6, m4, [r6 + 2 * 16]
-    paddd      m2, m6                          ;m2=[2+3+4+5+6+7]  Row3
-    pmaddwd    m4, [r6 + 3 * 16]
-    paddd      m0, m4                          ;m0=[0+1+2+3+4+5+6+7]  Row1 end
-    psrad      m0, 6
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,sp
+    mova            m8, [pd_526336]
+%else
+    add             r3d, r3d
+%endif
+    lea             r6, [r3 * 3]
+%rep %2 / 16
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m2, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    paddd           m1, m5
+    pmaddwd         m3, [r5]
+%ifidn %1,sp
+    paddd           m0, m8
+    paddd           m1, m8
+    psrad           m0, 12
+    psrad           m1, 12
+%else
+    psrad           m0, 6
+    psrad           m1, 6
+%endif
+    packssdw        m0, m1
 
-    lea        r0, [r0 + 2 * r1]
-    movq       m4, [r0]
-    punpcklwd  m5, m4                          ;m5=[7 8]
-    pmaddwd    m6, m5, [r6 + 2 * 16]
-    paddd      m3, m6                          ;m3=[3+4+5+6+7+8]  Row4
-    pmaddwd    m5, [r6 + 3 * 16]
-    paddd      m1, m5                          ;m1=[1+2+3+4+5+6+7+8]  Row2 end
-    psrad      m1, 6
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m2, m6
+    pmaddwd         m4, [r5]
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm1, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm1, 1
+    pmaddwd         m1, m5, [r5 + 1 * mmsize]
+    pmaddwd         m5, [r5]
+    paddd           m3, m1
+%ifidn %1,sp
+    paddd           m2, m8
+    paddd           m3, m8
+    psrad           m2, 12
+    psrad           m3, 12
+%else
+    psrad           m2, 6
+    psrad           m3, 6
+%endif
+    packssdw        m2, m3
+%ifidn %1,sp
+    packuswb        m0, m2
+    mova            m3, [interp8_hps_shuf]
+    vpermd          m0, m3, m0
+    vextracti128    xm2, m0, 1
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r6], xm2
+%else
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    movu            [r2], xm0
+    vextracti128    xm0, m0, 1
+    vextracti128    xm3, m2, 1
+    movu            [r2 + r3], xm0
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
+%endif
+    lea             r2, [r2 + r3 * 4]
 
-    packssdw   m0, m1
+    movu            xm1, [r0 + r4]                  ; m1 = row 7
+    punpckhwd       xm0, xm6, xm1
+    punpcklwd       xm6, xm1
+    vinserti128     m6, m6, xm0, 1
+    pmaddwd         m0, m6, [r5 + 1 * mmsize]
+    pmaddwd         m6, [r5]
+    paddd           m4, m0
+    lea             r0, [r0 + r1 * 4]
+    movu            xm0, [r0]                       ; m0 = row 8
+    punpckhwd       xm2, xm1, xm0
+    punpcklwd       xm1, xm0
+    vinserti128     m1, m1, xm2, 1
+    pmaddwd         m2, m1, [r5 + 1 * mmsize]
+    pmaddwd         m1, [r5]
+    paddd           m5, m2
+%ifidn %1,sp
+    paddd           m4, m8
+    paddd           m5, m8
+    psrad           m4, 12
+    psrad           m5, 12
+%else
+    psrad           m4, 6
+    psrad           m5, 6
+%endif
+    packssdw        m4, m5
 
-    movlps     [r2], m0
-    movhps     [r2 + r3], m0
+    movu            xm2, [r0 + r1]                  ; m2 = row 9
+    punpckhwd       xm5, xm0, xm2
+    punpcklwd       xm0, xm2
+    vinserti128     m0, m0, xm5, 1
+    pmaddwd         m5, m0, [r5 + 1 * mmsize]
+    paddd           m6, m5
+    pmaddwd         m0, [r5]
+    movu            xm5, [r0 + r1 * 2]              ; m5 = row 10
+    punpckhwd       xm7, xm2, xm5
+    punpcklwd       xm2, xm5
+    vinserti128     m2, m2, xm7, 1
+    pmaddwd         m7, m2, [r5 + 1 * mmsize]
+    paddd           m1, m7
+    pmaddwd         m2, [r5]
 
-    movq       m5, [r0 + r1]
-    punpcklwd  m4, m5                          ;m4=[8 9]
-    pmaddwd    m4, [r6 + 3 * 16]
-    paddd      m2, m4                          ;m2=[2+3+4+5+6+7+8+9]  Row3 end
-    psrad      m2, 6
+%ifidn %1,sp
+    paddd           m6, m8
+    paddd           m1, m8
+    psrad           m6, 12
+    psrad           m1, 12
+%else
+    psrad           m6, 6
+    psrad           m1, 6
+%endif
+    packssdw        m6, m1
+%ifidn %1,sp
+    packuswb        m4, m6
+    vpermd          m4, m3, m4
+    vextracti128    xm6, m4, 1
+    movq            [r2], xm4
+    movhps          [r2 + r3], xm4
+    movq            [r2 + r3 * 2], xm6
+    movhps          [r2 + r6], xm6
+%else
+    vpermq          m4, m4, 11011000b
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm7, m4, 1
+    vextracti128    xm1, m6, 1
+    movu            [r2], xm4
+    movu            [r2 + r3], xm7
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r6], xm1
+%endif
+    lea             r2, [r2 + r3 * 4]
 
-    movq       m4, [r0 + 2 * r1]
-    punpcklwd  m5, m4                          ;m5=[9 10]
-    pmaddwd    m5, [r6 + 3 * 16]
-    paddd      m3, m5                          ;m3=[3+4+5+6+7+8+9+10]  Row4 end
-    psrad      m3, 6
+    movu            xm7, [r0 + r4]                  ; m7 = row 11
+    punpckhwd       xm1, xm5, xm7
+    punpcklwd       xm5, xm7
+    vinserti128     m5, m5, xm1, 1
+    pmaddwd         m1, m5, [r5 + 1 * mmsize]
+    paddd           m0, m1
+    pmaddwd         m5, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm1, [r0]                       ; m1 = row 12
+    punpckhwd       xm4, xm7, xm1
+    punpcklwd       xm7, xm1
+    vinserti128     m7, m7, xm4, 1
+    pmaddwd         m4, m7, [r5 + 1 * mmsize]
+    paddd           m2, m4
+    pmaddwd         m7, [r5]
+%ifidn %1,sp
+    paddd           m0, m8
+    paddd           m2, m8
+    psrad           m0, 12
+    psrad           m2, 12
+%else
+    psrad           m0, 6
+    psrad           m2, 6
+%endif
+    packssdw        m0, m2
 
-    packssdw   m2, m3
+    movu            xm4, [r0 + r1]                  ; m4 = row 13
+    punpckhwd       xm2, xm1, xm4
+    punpcklwd       xm1, xm4
+    vinserti128     m1, m1, xm2, 1
+    pmaddwd         m2, m1, [r5 + 1 * mmsize]
+    paddd           m5, m2
+    pmaddwd         m1, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 14
+    punpckhwd       xm6, xm4, xm2
+    punpcklwd       xm4, xm2
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m7, m6
+    pmaddwd         m4, [r5]
+%ifidn %1,sp
+    paddd           m5, m8
+    paddd           m7, m8
+    psrad           m5, 12
+    psrad           m7, 12
+%else
+    psrad           m5, 6
+    psrad           m7, 6
+%endif
+    packssdw        m5, m7
+%ifidn %1,sp
+    packuswb        m0, m5
+    vpermd          m0, m3, m0
+    vextracti128    xm5, m0, 1
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm5
+    movhps          [r2 + r6], xm5
+%else
+    vpermq          m0, m0, 11011000b
+    vpermq          m5, m5, 11011000b
+    vextracti128    xm7, m0, 1
+    vextracti128    xm6, m5, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm7
+    movu            [r2 + r3 * 2], xm5
+    movu            [r2 + r6], xm6
+%endif
+    lea             r2, [r2 + r3 * 4]
 
-    movlps     [r2 + 2 * r3], m2
-    lea        r5, [3 * r3]
-    movhps     [r2 + r5], m2
+    movu            xm6, [r0 + r4]                  ; m6 = row 15
+    punpckhwd       xm5, xm2, xm6
+    punpcklwd       xm2, xm6
+    vinserti128     m2, m2, xm5, 1
+    pmaddwd         m5, m2, [r5 + 1 * mmsize]
+    paddd           m1, m5
+    pmaddwd         m2, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm0, [r0]                       ; m0 = row 16
+    punpckhwd       xm5, xm6, xm0
+    punpcklwd       xm6, xm0
+    vinserti128     m6, m6, xm5, 1
+    pmaddwd         m5, m6, [r5 + 1 * mmsize]
+    paddd           m4, m5
+    pmaddwd         m6, [r5]
+%ifidn %1,sp
+    paddd           m1, m8
+    paddd           m4, m8
+    psrad           m1, 12
+    psrad           m4, 12
+%else
+    psrad           m1, 6
+    psrad           m4, 6
+%endif
+    packssdw        m1, m4
 
-    lea        r5, [8 * r1 - 2 * 4]
-    sub        r0, r5
-    add        r2, 2 * 4
+    movu            xm5, [r0 + r1]                  ; m5 = row 17
+    punpckhwd       xm4, xm0, xm5
+    punpcklwd       xm0, xm5
+    vinserti128     m0, m0, xm4, 1
+    pmaddwd         m0, [r5 + 1 * mmsize]
+    paddd           m2, m0
+    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
+    punpckhwd       xm0, xm5, xm4
+    punpcklwd       xm5, xm4
+    vinserti128     m5, m5, xm0, 1
+    pmaddwd         m5, [r5 + 1 * mmsize]
+    paddd           m6, m5
+%ifidn %1,sp
+    paddd           m2, m8
+    paddd           m6, m8
+    psrad           m2, 12
+    psrad           m6, 12
+%else
+    psrad           m2, 6
+    psrad           m6, 6
+%endif
+    packssdw        m2, m6
+%ifidn %1,sp
+    packuswb        m1, m2
+    vpermd          m1, m3, m1
+    vextracti128    xm2, m1, 1
+    movq            [r2], xm1
+    movhps          [r2 + r3], xm1
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r6], xm2
+%else
+    vpermq          m1, m1, 11011000b
+    vpermq          m2, m2, 11011000b
+    vextracti128    xm6, m1, 1
+    vextracti128    xm4, m2, 1
+    movu            [r2], xm1
+    movu            [r2 + r3], xm6
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm4
+%endif
+    lea             r2, [r2 + r3 * 4]
+%endrep
+    RET
+%endif
+%endmacro
 
-    dec        r4d
-    jnz        .loopW
+    FILTER_VER_CHROMA_S_AVX2_8xN sp, 16
+    FILTER_VER_CHROMA_S_AVX2_8xN sp, 32
+    FILTER_VER_CHROMA_S_AVX2_8xN sp, 64
+    FILTER_VER_CHROMA_S_AVX2_8xN ss, 16
+    FILTER_VER_CHROMA_S_AVX2_8xN ss, 32
+    FILTER_VER_CHROMA_S_AVX2_8xN ss, 64
 
-    lea        r0, [r0 + 4 * r1 - 2 * %1]
-    lea        r2, [r2 + 4 * r3 - 2 * %1]
+%macro FILTER_VER_CHROMA_S_AVX2_Nx24 2
+%if ARCH_X86_64 == 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_%2x24, 4, 10, 10
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
 
-    dec        dword [rsp]
-    jnz        .loopH
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
 
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,sp
+    mova            m9, [pd_526336]
+%else
+    add             r3d, r3d
+%endif
+    lea             r6, [r3 * 3]
+    mov             r9d, %2 / 8
+.loopW:
+    PROCESS_CHROMA_S_AVX2_W8_16R %1
+%ifidn %1,sp
+    add             r2, 8
+%else
+    add             r2, 16
+%endif
+    add             r0, 16
+    dec             r9d
+    jnz             .loopW
+%ifidn %1,sp
+    lea             r2, [r8 + r3 * 4 - %2 + 8]
+%else
+    lea             r2, [r8 + r3 * 4 - 2 * %2 + 16]
+%endif
+    lea             r0, [r7 - 2 * %2 + 16]
+    mova            m7, m9
+    mov             r9d, %2 / 8
+.loop:
+    PROCESS_CHROMA_S_AVX2_W8_8R %1
+%ifidn %1,sp
+    add             r2, 8
+%else
+    add             r2, 16
+%endif
+    add             r0, 16
+    dec             r9d
+    jnz             .loop
     RET
+%endif
 %endmacro
 
-    FILTER_VER_LUMA_SS 4, 4
-    FILTER_VER_LUMA_SS 8, 8
-    FILTER_VER_LUMA_SS 8, 4
-    FILTER_VER_LUMA_SS 4, 8
-    FILTER_VER_LUMA_SS 16, 16
-    FILTER_VER_LUMA_SS 16, 8
-    FILTER_VER_LUMA_SS 8, 16
-    FILTER_VER_LUMA_SS 16, 12
-    FILTER_VER_LUMA_SS 12, 16
-    FILTER_VER_LUMA_SS 16, 4
-    FILTER_VER_LUMA_SS 4, 16
-    FILTER_VER_LUMA_SS 32, 32
-    FILTER_VER_LUMA_SS 32, 16
-    FILTER_VER_LUMA_SS 16, 32
-    FILTER_VER_LUMA_SS 32, 24
-    FILTER_VER_LUMA_SS 24, 32
-    FILTER_VER_LUMA_SS 32, 8
-    FILTER_VER_LUMA_SS 8, 32
-    FILTER_VER_LUMA_SS 64, 64
-    FILTER_VER_LUMA_SS 64, 32
-    FILTER_VER_LUMA_SS 32, 64
-    FILTER_VER_LUMA_SS 64, 48
-    FILTER_VER_LUMA_SS 48, 64
-    FILTER_VER_LUMA_SS 64, 16
-    FILTER_VER_LUMA_SS 16, 64
+    FILTER_VER_CHROMA_S_AVX2_Nx24 sp, 32
+    FILTER_VER_CHROMA_S_AVX2_Nx24 sp, 16
+    FILTER_VER_CHROMA_S_AVX2_Nx24 ss, 32
+    FILTER_VER_CHROMA_S_AVX2_Nx24 ss, 16
 
-%macro FILTER_VER_LUMA_AVX2_4x4 1
+%macro FILTER_VER_CHROMA_S_AVX2_2x8 1
 INIT_YMM avx2
-cglobal interp_8tap_vert_%1_4x4, 4, 6, 7
+cglobal interp_4tap_vert_%1_2x8, 4, 6, 7
     mov             r4d, r4m
+    shl             r4d, 6
     add             r1d, r1d
-    shl             r4d, 7
+    sub             r0, r1
 
 %ifdef PIC
-    lea             r5, [pw_LumaCoeffVer]
+    lea             r5, [pw_ChromaCoeffV]
     add             r5, r4
 %else
-    lea             r5, [pw_LumaCoeffVer + r4]
+    lea             r5, [pw_ChromaCoeffV + r4]
 %endif
 
     lea             r4, [r1 * 3]
-    sub             r0, r4
-
 %ifidn %1,sp
     mova            m6, [pd_526336]
 %else
     add             r3d, r3d
 %endif
-
-    movq            xm0, [r0]
-    movq            xm1, [r0 + r1]
+    movd            xm0, [r0]
+    movd            xm1, [r0 + r1]
     punpcklwd       xm0, xm1
-    movq            xm2, [r0 + r1 * 2]
+    movd            xm2, [r0 + r1 * 2]
     punpcklwd       xm1, xm2
-    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
-    pmaddwd         m0, [r5]
-    movq            xm3, [r0 + r4]
+    punpcklqdq      xm0, xm1                        ; m0 = [2 1 1 0]
+    movd            xm3, [r0 + r4]
     punpcklwd       xm2, xm3
     lea             r0, [r0 + 4 * r1]
-    movq            xm4, [r0]
+    movd            xm4, [r0]
     punpcklwd       xm3, xm4
-    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
-    pmaddwd         m5, m2, [r5 + 1 * mmsize]
-    pmaddwd         m2, [r5]
-    paddd           m0, m5
-    movq            xm3, [r0 + r1]
-    punpcklwd       xm4, xm3
-    movq            xm1, [r0 + r1 * 2]
-    punpcklwd       xm3, xm1
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
-    pmaddwd         m5, m4, [r5 + 2 * mmsize]
-    pmaddwd         m4, [r5 + 1 * mmsize]
-    paddd           m0, m5
-    paddd           m2, m4
-    movq            xm3, [r0 + r4]
+    punpcklqdq      xm2, xm3                        ; m2 = [4 3 3 2]
+    vinserti128     m0, m0, xm2, 1                  ; m0 = [4 3 3 2 2 1 1 0]
+    movd            xm1, [r0 + r1]
+    punpcklwd       xm4, xm1
+    movd            xm3, [r0 + r1 * 2]
     punpcklwd       xm1, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm4, [r0]
-    punpcklwd       xm3, xm4
-    vinserti128     m1, m1, xm3, 1                  ; m1 = [8 7 7 6]
-    pmaddwd         m5, m1, [r5 + 3 * mmsize]
-    pmaddwd         m1, [r5 + 2 * mmsize]
-    paddd           m0, m5
-    paddd           m2, m1
-    movq            xm3, [r0 + r1]
-    punpcklwd       xm4, xm3
-    movq            xm1, [r0 + 2 * r1]
+    punpcklqdq      xm4, xm1                        ; m4 = [6 5 5 4]
+    vinserti128     m2, m2, xm4, 1                  ; m2 = [6 5 5 4 4 3 3 2]
+    pmaddwd         m0, [r5]
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m0, m2
+    movd            xm1, [r0 + r4]
     punpcklwd       xm3, xm1
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [A 9 9 8]
-    pmaddwd         m4, [r5 + 3 * mmsize]
-    paddd           m2, m4
-
+    lea             r0, [r0 + 4 * r1]
+    movd            xm2, [r0]
+    punpcklwd       xm1, xm2
+    punpcklqdq      xm3, xm1                        ; m3 = [8 7 7 6]
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [8 7 7 6 6 5 5 4]
+    movd            xm1, [r0 + r1]
+    punpcklwd       xm2, xm1
+    movd            xm5, [r0 + r1 * 2]
+    punpcklwd       xm1, xm5
+    punpcklqdq      xm2, xm1                        ; m2 = [10 9 9 8]
+    vinserti128     m3, m3, xm2, 1                  ; m3 = [10 9 9 8 8 7 7 6]
+    pmaddwd         m4, [r5]
+    pmaddwd         m3, [r5 + 1 * mmsize]
+    paddd           m4, m3
 %ifidn %1,sp
     paddd           m0, m6
-    paddd           m2, m6
+    paddd           m4, m6
     psrad           m0, 12
-    psrad           m2, 12
+    psrad           m4, 12
 %else
     psrad           m0, 6
-    psrad           m2, 6
+    psrad           m4, 6
 %endif
-    packssdw        m0, m2
-    vextracti128    xm2, m0, 1
+    packssdw        m0, m4
+    vextracti128    xm4, m0, 1
     lea             r4, [r3 * 3]
-
 %ifidn %1,sp
-    packuswb        xm0, xm2
-    movd            [r2], xm0
-    pextrd          [r2 + r3], xm0, 2
-    pextrd          [r2 + r3 * 2], xm0, 1
-    pextrd          [r2 + r4], xm0, 3
+    packuswb        xm0, xm4
+    pextrw          [r2], xm0, 0
+    pextrw          [r2 + r3], xm0, 1
+    pextrw          [r2 + 2 * r3], xm0, 4
+    pextrw          [r2 + r4], xm0, 5
+    lea             r2, [r2 + r3 * 4]
+    pextrw          [r2], xm0, 2
+    pextrw          [r2 + r3], xm0, 3
+    pextrw          [r2 + 2 * r3], xm0, 6
+    pextrw          [r2 + r4], xm0, 7
 %else
-    movq            [r2], xm0
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r4], xm2
+    movd            [r2], xm0
+    pextrd          [r2 + r3], xm0, 1
+    movd            [r2 + 2 * r3], xm4
+    pextrd          [r2 + r4], xm4, 1
+    lea             r2, [r2 + r3 * 4]
+    pextrd          [r2], xm0, 2
+    pextrd          [r2 + r3], xm0, 3
+    pextrd          [r2 + 2 * r3], xm4, 2
+    pextrd          [r2 + r4], xm4, 3
 %endif
     RET
 %endmacro
 
-    FILTER_VER_LUMA_AVX2_4x4 sp
-    FILTER_VER_LUMA_AVX2_4x4 ss
+    FILTER_VER_CHROMA_S_AVX2_2x8 sp
+    FILTER_VER_CHROMA_S_AVX2_2x8 ss
 
-%macro FILTER_VER_LUMA_AVX2_4x8 1
+%macro FILTER_VER_CHROMA_S_AVX2_2x16 1
+%if ARCH_X86_64 == 1
 INIT_YMM avx2
-cglobal interp_8tap_vert_%1_4x8, 4, 7, 8
+cglobal interp_4tap_vert_%1_2x16, 4, 6, 9
     mov             r4d, r4m
-    shl             r4d, 7
+    shl             r4d, 6
     add             r1d, r1d
+    sub             r0, r1
 
 %ifdef PIC
-    lea             r5, [pw_LumaCoeffVer]
+    lea             r5, [pw_ChromaCoeffV]
     add             r5, r4
 %else
-    lea             r5, [pw_LumaCoeffVer + r4]
+    lea             r5, [pw_ChromaCoeffV + r4]
 %endif
 
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-
+    lea             r4, [r1 * 3]
 %ifidn %1,sp
-    mova            m7, [pd_526336]
+    mova            m6, [pd_526336]
 %else
     add             r3d, r3d
 %endif
-    lea             r6, [r3 * 3]
-
-    movq            xm0, [r0]
-    movq            xm1, [r0 + r1]
+    movd            xm0, [r0]
+    movd            xm1, [r0 + r1]
     punpcklwd       xm0, xm1
-    movq            xm2, [r0 + r1 * 2]
+    movd            xm2, [r0 + r1 * 2]
     punpcklwd       xm1, xm2
-    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
-    pmaddwd         m0, [r5]
-    movq            xm3, [r0 + r4]
+    punpcklqdq      xm0, xm1                        ; m0 = [2 1 1 0]
+    movd            xm3, [r0 + r4]
     punpcklwd       xm2, xm3
     lea             r0, [r0 + 4 * r1]
-    movq            xm4, [r0]
+    movd            xm4, [r0]
     punpcklwd       xm3, xm4
-    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
-    pmaddwd         m5, m2, [r5 + 1 * mmsize]
-    pmaddwd         m2, [r5]
-    paddd           m0, m5
-    movq            xm3, [r0 + r1]
-    punpcklwd       xm4, xm3
-    movq            xm1, [r0 + r1 * 2]
+    punpcklqdq      xm2, xm3                        ; m2 = [4 3 3 2]
+    vinserti128     m0, m0, xm2, 1                  ; m0 = [4 3 3 2 2 1 1 0]
+    movd            xm1, [r0 + r1]
+    punpcklwd       xm4, xm1
+    movd            xm3, [r0 + r1 * 2]
+    punpcklwd       xm1, xm3
+    punpcklqdq      xm4, xm1                        ; m4 = [6 5 5 4]
+    vinserti128     m2, m2, xm4, 1                  ; m2 = [6 5 5 4 4 3 3 2]
+    pmaddwd         m0, [r5]
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m0, m2
+    movd            xm1, [r0 + r4]
     punpcklwd       xm3, xm1
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
-    pmaddwd         m5, m4, [r5 + 2 * mmsize]
-    paddd           m0, m5
-    pmaddwd         m5, m4, [r5 + 1 * mmsize]
-    paddd           m2, m5
+    lea             r0, [r0 + 4 * r1]
+    movd            xm2, [r0]
+    punpcklwd       xm1, xm2
+    punpcklqdq      xm3, xm1                        ; m3 = [8 7 7 6]
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [8 7 7 6 6 5 5 4]
+    movd            xm1, [r0 + r1]
+    punpcklwd       xm2, xm1
+    movd            xm5, [r0 + r1 * 2]
+    punpcklwd       xm1, xm5
+    punpcklqdq      xm2, xm1                        ; m2 = [10 9 9 8]
+    vinserti128     m3, m3, xm2, 1                  ; m3 = [10 9 9 8 8 7 7 6]
     pmaddwd         m4, [r5]
-    movq            xm3, [r0 + r4]
-    punpcklwd       xm1, xm3
+    pmaddwd         m3, [r5 + 1 * mmsize]
+    paddd           m4, m3
+    movd            xm1, [r0 + r4]
+    punpcklwd       xm5, xm1
     lea             r0, [r0 + 4 * r1]
-    movq            xm6, [r0]
-    punpcklwd       xm3, xm6
-    vinserti128     m1, m1, xm3, 1                  ; m1 = [8 7 7 6]
-    pmaddwd         m5, m1, [r5 + 3 * mmsize]
-    paddd           m0, m5
-    pmaddwd         m5, m1, [r5 + 2 * mmsize]
+    movd            xm3, [r0]
+    punpcklwd       xm1, xm3
+    punpcklqdq      xm5, xm1                        ; m5 = [12 11 11 10]
+    vinserti128     m2, m2, xm5, 1                  ; m2 = [12 11 11 10 10 9 9 8]
+    movd            xm1, [r0 + r1]
+    punpcklwd       xm3, xm1
+    movd            xm7, [r0 + r1 * 2]
+    punpcklwd       xm1, xm7
+    punpcklqdq      xm3, xm1                        ; m3 = [14 13 13 12]
+    vinserti128     m5, m5, xm3, 1                  ; m5 = [14 13 13 12 12 11 11 10]
+    pmaddwd         m2, [r5]
+    pmaddwd         m5, [r5 + 1 * mmsize]
     paddd           m2, m5
-    pmaddwd         m5, m1, [r5 + 1 * mmsize]
-    paddd           m4, m5
-    pmaddwd         m1, [r5]
-    movq            xm3, [r0 + r1]
-    punpcklwd       xm6, xm3
-    movq            xm5, [r0 + 2 * r1]
-    punpcklwd       xm3, xm5
-    vinserti128     m6, m6, xm3, 1                  ; m6 = [A 9 9 8]
-    pmaddwd         m3, m6, [r5 + 3 * mmsize]
-    paddd           m2, m3
-    pmaddwd         m3, m6, [r5 + 2 * mmsize]
-    paddd           m4, m3
-    pmaddwd         m6, [r5 + 1 * mmsize]
-    paddd           m1, m6
-
+    movd            xm5, [r0 + r4]
+    punpcklwd       xm7, xm5
+    lea             r0, [r0 + 4 * r1]
+    movd            xm1, [r0]
+    punpcklwd       xm5, xm1
+    punpcklqdq      xm7, xm5                        ; m7 = [16 15 15 14]
+    vinserti128     m3, m3, xm7, 1                  ; m3 = [16 15 15 14 14 13 13 12]
+    movd            xm5, [r0 + r1]
+    punpcklwd       xm1, xm5
+    movd            xm8, [r0 + r1 * 2]
+    punpcklwd       xm5, xm8
+    punpcklqdq      xm1, xm5                        ; m1 = [18 17 17 16]
+    vinserti128     m7, m7, xm1, 1                  ; m7 = [18 17 17 16 16 15 15 14]
+    pmaddwd         m3, [r5]
+    pmaddwd         m7, [r5 + 1 * mmsize]
+    paddd           m3, m7
 %ifidn %1,sp
-    paddd           m0, m7
-    paddd           m2, m7
+    paddd           m0, m6
+    paddd           m4, m6
+    paddd           m2, m6
+    paddd           m3, m6
     psrad           m0, 12
+    psrad           m4, 12
     psrad           m2, 12
+    psrad           m3, 12
 %else
     psrad           m0, 6
-    psrad           m2, 6
-%endif
-    packssdw        m0, m2
-
-    movq            xm3, [r0 + r4]
-    punpcklwd       xm5, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm2, [r0]
-    punpcklwd       xm3, xm2
-    vinserti128     m5, m5, xm3, 1                  ; m5 = [C B B A]
-    pmaddwd         m3, m5, [r5 + 3 * mmsize]
-    paddd           m4, m3
-    pmaddwd         m5, [r5 + 2 * mmsize]
-    paddd           m1, m5
-    movq            xm3, [r0 + r1]
-    punpcklwd       xm2, xm3
-    movq            xm5, [r0 + 2 * r1]
-    punpcklwd       xm3, xm5
-    vinserti128     m2, m2, xm3, 1                  ; m2 = [E D D C]
-    pmaddwd         m2, [r5 + 3 * mmsize]
-    paddd           m1, m2
-
-%ifidn %1,sp
-    paddd           m4, m7
-    paddd           m1, m7
-    psrad           m4, 12
-    psrad           m1, 12
-%else
     psrad           m4, 6
-    psrad           m1, 6
+    psrad           m2, 6
+    psrad           m3, 6
 %endif
-    packssdw        m4, m1
-
+    packssdw        m0, m4
+    packssdw        m2, m3
+    lea             r4, [r3 * 3]
 %ifidn %1,sp
-    packuswb        m0, m4
+    packuswb        m0, m2
     vextracti128    xm2, m0, 1
-    movd            [r2], xm0
-    movd            [r2 + r3], xm2
-    pextrd          [r2 + r3 * 2], xm0, 1
-    pextrd          [r2 + r6], xm2, 1
+    pextrw          [r2], xm0, 0
+    pextrw          [r2 + r3], xm0, 1
+    pextrw          [r2 + 2 * r3], xm2, 0
+    pextrw          [r2 + r4], xm2, 1
     lea             r2, [r2 + r3 * 4]
-    pextrd          [r2], xm0, 2
-    pextrd          [r2 + r3], xm2, 2
-    pextrd          [r2 + r3 * 2], xm0, 3
-    pextrd          [r2 + r6], xm2, 3
+    pextrw          [r2], xm0, 2
+    pextrw          [r2 + r3], xm0, 3
+    pextrw          [r2 + 2 * r3], xm2, 2
+    pextrw          [r2 + r4], xm2, 3
+    lea             r2, [r2 + r3 * 4]
+    pextrw          [r2], xm0, 4
+    pextrw          [r2 + r3], xm0, 5
+    pextrw          [r2 + 2 * r3], xm2, 4
+    pextrw          [r2 + r4], xm2, 5
+    lea             r2, [r2 + r3 * 4]
+    pextrw          [r2], xm0, 6
+    pextrw          [r2 + r3], xm0, 7
+    pextrw          [r2 + 2 * r3], xm2, 6
+    pextrw          [r2 + r4], xm2, 7
 %else
-    vextracti128    xm2, m0, 1
-    vextracti128    xm1, m4, 1
-    movq            [r2], xm0
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r6], xm2
+    vextracti128    xm4, m0, 1
+    vextracti128    xm3, m2, 1
+    movd            [r2], xm0
+    pextrd          [r2 + r3], xm0, 1
+    movd            [r2 + 2 * r3], xm4
+    pextrd          [r2 + r4], xm4, 1
     lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm4
-    movq            [r2 + r3], xm1
-    movhps          [r2 + r3 * 2], xm4
-    movhps          [r2 + r6], xm1
-%endif
-    RET
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_4x8 sp
-    FILTER_VER_LUMA_AVX2_4x8 ss
-
-%macro PROCESS_LUMA_AVX2_W4_16R 1
-    movq            xm0, [r0]
-    movq            xm1, [r0 + r1]
-    punpcklwd       xm0, xm1
-    movq            xm2, [r0 + r1 * 2]
-    punpcklwd       xm1, xm2
-    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
-    pmaddwd         m0, [r5]
-    movq            xm3, [r0 + r4]
-    punpcklwd       xm2, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm4, [r0]
-    punpcklwd       xm3, xm4
-    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
-    pmaddwd         m5, m2, [r5 + 1 * mmsize]
-    pmaddwd         m2, [r5]
-    paddd           m0, m5
-    movq            xm3, [r0 + r1]
-    punpcklwd       xm4, xm3
-    movq            xm1, [r0 + r1 * 2]
-    punpcklwd       xm3, xm1
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
-    pmaddwd         m5, m4, [r5 + 2 * mmsize]
-    paddd           m0, m5
-    pmaddwd         m5, m4, [r5 + 1 * mmsize]
-    paddd           m2, m5
-    pmaddwd         m4, [r5]
-    movq            xm3, [r0 + r4]
-    punpcklwd       xm1, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm6, [r0]
-    punpcklwd       xm3, xm6
-    vinserti128     m1, m1, xm3, 1                  ; m1 = [8 7 7 6]
-    pmaddwd         m5, m1, [r5 + 3 * mmsize]
-    paddd           m0, m5
-    pmaddwd         m5, m1, [r5 + 2 * mmsize]
-    paddd           m2, m5
-    pmaddwd         m5, m1, [r5 + 1 * mmsize]
-    paddd           m4, m5
-    pmaddwd         m1, [r5]
-    movq            xm3, [r0 + r1]
-    punpcklwd       xm6, xm3
-    movq            xm5, [r0 + 2 * r1]
-    punpcklwd       xm3, xm5
-    vinserti128     m6, m6, xm3, 1                  ; m6 = [10 9 9 8]
-    pmaddwd         m3, m6, [r5 + 3 * mmsize]
-    paddd           m2, m3
-    pmaddwd         m3, m6, [r5 + 2 * mmsize]
-    paddd           m4, m3
-    pmaddwd         m3, m6, [r5 + 1 * mmsize]
-    paddd           m1, m3
-    pmaddwd         m6, [r5]
+    pextrd          [r2], xm0, 2
+    pextrd          [r2 + r3], xm0, 3
+    pextrd          [r2 + 2 * r3], xm4, 2
+    pextrd          [r2 + r4], xm4, 3
+    lea             r2, [r2 + r3 * 4]
+    movd            [r2], xm2
+    pextrd          [r2 + r3], xm2, 1
+    movd            [r2 + 2 * r3], xm3
+    pextrd          [r2 + r4], xm3, 1
+    lea             r2, [r2 + r3 * 4]
+    pextrd          [r2], xm2, 2
+    pextrd          [r2 + r3], xm2, 3
+    pextrd          [r2 + 2 * r3], xm3, 2
+    pextrd          [r2 + r4], xm3, 3
+%endif
+    RET
+%endif
+%endmacro
 
-%ifidn %1,sp
-    paddd           m0, m7
-    paddd           m2, m7
-    psrad           m0, 12
-    psrad           m2, 12
+    FILTER_VER_CHROMA_S_AVX2_2x16 sp
+    FILTER_VER_CHROMA_S_AVX2_2x16 ss
+
+%macro FILTER_VER_CHROMA_S_AVX2_6x8 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_6x8, 4, 6, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
 %else
-    psrad           m0, 6
-    psrad           m2, 6
+    lea             r5, [pw_ChromaCoeffV + r4]
 %endif
-    packssdw        m0, m2
-    vextracti128    xm2, m0, 1
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
 %ifidn %1,sp
-    packuswb        xm0, xm2
-    movd            [r2], xm0
-    pextrd          [r2 + r3], xm0, 2
-    pextrd          [r2 + r3 * 2], xm0, 1
-    pextrd          [r2 + r6], xm0, 3
+    mova            m7, [pd_526336]
 %else
-    movq            [r2], xm0
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r6], xm2
+    add             r3d, r3d
 %endif
 
-    movq            xm2, [r0 + r4]
-    punpcklwd       xm5, xm2
-    lea             r0, [r0 + 4 * r1]
-    movq            xm0, [r0]
-    punpcklwd       xm2, xm0
-    vinserti128     m5, m5, xm2, 1                  ; m5 = [12 11 11 10]
-    pmaddwd         m2, m5, [r5 + 3 * mmsize]
-    paddd           m4, m2
-    pmaddwd         m2, m5, [r5 + 2 * mmsize]
-    paddd           m1, m2
-    pmaddwd         m2, m5, [r5 + 1 * mmsize]
-    paddd           m6, m2
-    pmaddwd         m5, [r5]
-    movq            xm2, [r0 + r1]
-    punpcklwd       xm0, xm2
-    movq            xm3, [r0 + 2 * r1]
-    punpcklwd       xm2, xm3
-    vinserti128     m0, m0, xm2, 1                  ; m0 = [14 13 13 12]
-    pmaddwd         m2, m0, [r5 + 3 * mmsize]
-    paddd           m1, m2
-    pmaddwd         m2, m0, [r5 + 2 * mmsize]
-    paddd           m6, m2
-    pmaddwd         m2, m0, [r5 + 1 * mmsize]
-    paddd           m5, m2
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
     pmaddwd         m0, [r5]
-
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    pmaddwd         m2, [r5]
+    paddd           m0, m4
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    pmaddwd         m3, [r5]
+    paddd           m1, m5
 %ifidn %1,sp
-    paddd           m4, m7
+    paddd           m0, m7
     paddd           m1, m7
-    psrad           m4, 12
+    psrad           m0, 12
     psrad           m1, 12
 %else
-    psrad           m4, 6
+    psrad           m0, 6
     psrad           m1, 6
 %endif
-    packssdw        m4, m1
-    vextracti128    xm1, m4, 1
-    lea             r2, [r2 + r3 * 4]
-%ifidn %1,sp
-    packuswb        xm4, xm1
-    movd            [r2], xm4
-    pextrd          [r2 + r3], xm4, 2
-    pextrd          [r2 + r3 * 2], xm4, 1
-    pextrd          [r2 + r6], xm4, 3
-%else
-    movq            [r2], xm4
-    movq            [r2 + r3], xm1
-    movhps          [r2 + r3 * 2], xm4
-    movhps          [r2 + r6], xm1
-%endif
+    packssdw        m0, m1
 
-    movq            xm4, [r0 + r4]
-    punpcklwd       xm3, xm4
-    lea             r0, [r0 + 4 * r1]
-    movq            xm1, [r0]
-    punpcklwd       xm4, xm1
-    vinserti128     m3, m3, xm4, 1                  ; m3 = [16 15 15 14]
-    pmaddwd         m4, m3, [r5 + 3 * mmsize]
-    paddd           m6, m4
-    pmaddwd         m4, m3, [r5 + 2 * mmsize]
-    paddd           m5, m4
-    pmaddwd         m4, m3, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m3, [r5]
-    movq            xm4, [r0 + r1]
-    punpcklwd       xm1, xm4
-    movq            xm2, [r0 + 2 * r1]
-    punpcklwd       xm4, xm2
-    vinserti128     m1, m1, xm4, 1                  ; m1 = [18 17 17 16]
-    pmaddwd         m4, m1, [r5 + 3 * mmsize]
-    paddd           m5, m4
-    pmaddwd         m4, m1, [r5 + 2 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m1, [r5 + 1 * mmsize]
-    paddd           m3, m1
-    movq            xm4, [r0 + r4]
-    punpcklwd       xm2, xm4
-    lea             r0, [r0 + 4 * r1]
-    movq            xm1, [r0]
-    punpcklwd       xm4, xm1
-    vinserti128     m2, m2, xm4, 1                  ; m2 = [20 19 19 18]
-    pmaddwd         m4, m2, [r5 + 3 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m2, [r5 + 2 * mmsize]
-    paddd           m3, m2
-    movq            xm4, [r0 + r1]
-    punpcklwd       xm1, xm4
-    movq            xm2, [r0 + 2 * r1]
-    punpcklwd       xm4, xm2
-    vinserti128     m1, m1, xm4, 1                  ; m1 = [22 21 21 20]
-    pmaddwd         m1, [r5 + 3 * mmsize]
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m2, m6
+    pmaddwd         m4, [r5]
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm1, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm1, 1
+    pmaddwd         m1, m5, [r5 + 1 * mmsize]
+    pmaddwd         m5, [r5]
     paddd           m3, m1
-
 %ifidn %1,sp
-    paddd           m6, m7
-    paddd           m5, m7
-    paddd           m0, m7
+    paddd           m2, m7
     paddd           m3, m7
-    psrad           m6, 12
-    psrad           m5, 12
-    psrad           m0, 12
+    psrad           m2, 12
     psrad           m3, 12
 %else
-    psrad           m6, 6
-    psrad           m5, 6
-    psrad           m0, 6
+    psrad           m2, 6
     psrad           m3, 6
 %endif
-    packssdw        m6, m5
-    packssdw        m0, m3
-    lea             r2, [r2 + r3 * 4]
+    packssdw        m2, m3
+
+    movu            xm1, [r0 + r4]                  ; m1 = row 7
+    punpckhwd       xm3, xm6, xm1
+    punpcklwd       xm6, xm1
+    vinserti128     m6, m6, xm3, 1
+    pmaddwd         m3, m6, [r5 + 1 * mmsize]
+    pmaddwd         m6, [r5]
+    paddd           m4, m3
 
+    lea             r4, [r3 * 3]
 %ifidn %1,sp
-    packuswb        m6, m0
-    vextracti128    xm0, m6, 1
-    movd            [r2], xm6
-    movd            [r2 + r3], xm0
-    pextrd          [r2 + r3 * 2], xm6, 1
-    pextrd          [r2 + r6], xm0, 1
-    lea             r2, [r2 + r3 * 4]
-    pextrd          [r2], xm6, 2
-    pextrd          [r2 + r3], xm0, 2
-    pextrd          [r2 + r3 * 2], xm6, 3
-    pextrd          [r2 + r6], xm0, 3
+    packuswb        m0, m2
+    vextracti128    xm2, m0, 1
+    movd            [r2], xm0
+    pextrw          [r2 + 4], xm2, 0
+    pextrd          [r2 + r3], xm0, 1
+    pextrw          [r2 + r3 + 4], xm2, 2
+    pextrd          [r2 + r3 * 2], xm0, 2
+    pextrw          [r2 + r3 * 2 + 4], xm2, 4
+    pextrd          [r2 + r4], xm0, 3
+    pextrw          [r2 + r4 + 4], xm2, 6
 %else
-    vextracti128    xm5, m6, 1
-    vextracti128    xm3, m0, 1
-    movq            [r2], xm6
-    movq            [r2 + r3], xm5
-    movhps          [r2 + r3 * 2], xm6
-    movhps          [r2 + r6], xm5
-    lea             r2, [r2 + r3 * 4]
     movq            [r2], xm0
-    movq            [r2 + r3], xm3
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r6], xm3
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r4], xm2
+    vextracti128    xm0, m0, 1
+    vextracti128    xm3, m2, 1
+    movd            [r2 + 8], xm0
+    pextrd          [r2 + r3 + 8], xm0, 2
+    movd            [r2 + r3 * 2 + 8], xm3
+    pextrd          [r2 + r4 + 8], xm3, 2
+%endif
+    lea             r2, [r2 + r3 * 4]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm0, [r0]                       ; m0 = row 8
+    punpckhwd       xm2, xm1, xm0
+    punpcklwd       xm1, xm0
+    vinserti128     m1, m1, xm2, 1
+    pmaddwd         m2, m1, [r5 + 1 * mmsize]
+    pmaddwd         m1, [r5]
+    paddd           m5, m2
+%ifidn %1,sp
+    paddd           m4, m7
+    paddd           m5, m7
+    psrad           m4, 12
+    psrad           m5, 12
+%else
+    psrad           m4, 6
+    psrad           m5, 6
 %endif
-%endmacro
+    packssdw        m4, m5
 
-%macro FILTER_VER_LUMA_AVX2_4x16 1
-INIT_YMM avx2
-cglobal interp_8tap_vert_%1_4x16, 4, 7, 8
-    mov             r4d, r4m
-    shl             r4d, 7
-    add             r1d, r1d
+    movu            xm2, [r0 + r1]                  ; m2 = row 9
+    punpckhwd       xm5, xm0, xm2
+    punpcklwd       xm0, xm2
+    vinserti128     m0, m0, xm5, 1
+    pmaddwd         m0, [r5 + 1 * mmsize]
+    paddd           m6, m0
+    movu            xm5, [r0 + r1 * 2]              ; m5 = row 10
+    punpckhwd       xm0, xm2, xm5
+    punpcklwd       xm2, xm5
+    vinserti128     m2, m2, xm0, 1
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m1, m2
 
-%ifdef PIC
-    lea             r5, [pw_LumaCoeffVer]
-    add             r5, r4
+%ifidn %1,sp
+    paddd           m6, m7
+    paddd           m1, m7
+    psrad           m6, 12
+    psrad           m1, 12
 %else
-    lea             r5, [pw_LumaCoeffVer + r4]
+    psrad           m6, 6
+    psrad           m1, 6
 %endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
+    packssdw        m6, m1
 %ifidn %1,sp
-    mova            m7, [pd_526336]
+    packuswb        m4, m6
+    vextracti128    xm6, m4, 1
+    movd            [r2], xm4
+    pextrw          [r2 + 4], xm6, 0
+    pextrd          [r2 + r3], xm4, 1
+    pextrw          [r2 + r3 + 4], xm6, 2
+    pextrd          [r2 + r3 * 2], xm4, 2
+    pextrw          [r2 + r3 * 2 + 4], xm6, 4
+    pextrd          [r2 + r4], xm4, 3
+    pextrw          [r2 + r4 + 4], xm6, 6
 %else
-    add             r3d, r3d
+    movq            [r2], xm4
+    movhps          [r2 + r3], xm4
+    movq            [r2 + r3 * 2], xm6
+    movhps          [r2 + r4], xm6
+    vextracti128    xm5, m4, 1
+    vextracti128    xm1, m6, 1
+    movd            [r2 + 8], xm5
+    pextrd          [r2 + r3 + 8], xm5, 2
+    movd            [r2 + r3 * 2 + 8], xm1
+    pextrd          [r2 + r4 + 8], xm1, 2
 %endif
-    lea             r6, [r3 * 3]
-    PROCESS_LUMA_AVX2_W4_16R %1
     RET
 %endmacro
 
-    FILTER_VER_LUMA_AVX2_4x16 sp
-    FILTER_VER_LUMA_AVX2_4x16 ss
+    FILTER_VER_CHROMA_S_AVX2_6x8 sp
+    FILTER_VER_CHROMA_S_AVX2_6x8 ss
 
-%macro FILTER_VER_LUMA_S_AVX2_8x8 1
-INIT_YMM avx2
+%macro FILTER_VER_CHROMA_S_AVX2_6x16 1
 %if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_8x8, 4, 6, 12
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_6x16, 4, 7, 9
     mov             r4d, r4m
-    shl             r4d, 7
+    shl             r4d, 6
     add             r1d, r1d
 
 %ifdef PIC
-    lea             r5, [pw_LumaCoeffVer]
+    lea             r5, [pw_ChromaCoeffV]
     add             r5, r4
 %else
-    lea             r5, [pw_LumaCoeffVer + r4]
+    lea             r5, [pw_ChromaCoeffV + r4]
 %endif
 
     lea             r4, [r1 * 3]
-    sub             r0, r4
-
+    sub             r0, r1
 %ifidn %1,sp
-    mova            m11, [pd_526336]
+    mova            m8, [pd_526336]
 %else
     add             r3d, r3d
 %endif
-
+    lea             r6, [r3 * 3]
     movu            xm0, [r0]                       ; m0 = row 0
     movu            xm1, [r0 + r1]                  ; m1 = row 1
     punpckhwd       xm2, xm0, xm1
@@ -23443,687 +19032,763 @@ cglobal interp_8tap_vert_%1_8x8, 4, 6, 12
     punpcklwd       xm2, xm3
     vinserti128     m2, m2, xm4, 1
     pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    pmaddwd         m2, [r5]
     paddd           m0, m4
+    pmaddwd         m2, [r5]
     lea             r0, [r0 + r1 * 4]
     movu            xm4, [r0]                       ; m4 = row 4
     punpckhwd       xm5, xm3, xm4
     punpcklwd       xm3, xm4
     vinserti128     m3, m3, xm5, 1
     pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    pmaddwd         m3, [r5]
     paddd           m1, m5
+    pmaddwd         m3, [r5]
+%ifidn %1,sp
+    paddd           m0, m8
+    paddd           m1, m8
+    psrad           m0, 12
+    psrad           m1, 12
+%else
+    psrad           m0, 6
+    psrad           m1, 6
+%endif
+    packssdw        m0, m1
+
     movu            xm5, [r0 + r1]                  ; m5 = row 5
     punpckhwd       xm6, xm4, xm5
     punpcklwd       xm4, xm5
     vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 2 * mmsize]
-    paddd           m0, m6
     pmaddwd         m6, m4, [r5 + 1 * mmsize]
     paddd           m2, m6
     pmaddwd         m4, [r5]
     movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm7, xm5, xm6
+    punpckhwd       xm1, xm5, xm6
     punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddwd         m7, m5, [r5 + 2 * mmsize]
-    paddd           m1, m7
-    pmaddwd         m7, m5, [r5 + 1 * mmsize]
+    vinserti128     m5, m5, xm1, 1
+    pmaddwd         m1, m5, [r5 + 1 * mmsize]
     pmaddwd         m5, [r5]
-    paddd           m3, m7
-    movu            xm7, [r0 + r4]                  ; m7 = row 7
-    punpckhwd       xm8, xm6, xm7
-    punpcklwd       xm6, xm7
-    vinserti128     m6, m6, xm8, 1
-    pmaddwd         m8, m6, [r5 + 3 * mmsize]
-    paddd           m0, m8
-    pmaddwd         m8, m6, [r5 + 2 * mmsize]
+    paddd           m3, m1
+%ifidn %1,sp
     paddd           m2, m8
-    pmaddwd         m8, m6, [r5 + 1 * mmsize]
-    pmaddwd         m6, [r5]
-    paddd           m4, m8
-    lea             r0, [r0 + r1 * 4]
-    movu            xm8, [r0]                       ; m8 = row 8
-    punpckhwd       xm9, xm7, xm8
-    punpcklwd       xm7, xm8
-    vinserti128     m7, m7, xm9, 1
-    pmaddwd         m9, m7, [r5 + 3 * mmsize]
-    paddd           m1, m9
-    pmaddwd         m9, m7, [r5 + 2 * mmsize]
-    paddd           m3, m9
-    pmaddwd         m9, m7, [r5 + 1 * mmsize]
-    pmaddwd         m7, [r5]
-    paddd           m5, m9
-    movu            xm9, [r0 + r1]                  ; m9 = row 9
-    punpckhwd       xm10, xm8, xm9
-    punpcklwd       xm8, xm9
-    vinserti128     m8, m8, xm10, 1
-    pmaddwd         m10, m8, [r5 + 3 * mmsize]
-    paddd           m2, m10
-    pmaddwd         m10, m8, [r5 + 2 * mmsize]
-    pmaddwd         m8, [r5 + 1 * mmsize]
-    paddd           m4, m10
-    paddd           m6, m8
-    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
-    punpckhwd       xm8, xm9, xm10
-    punpcklwd       xm9, xm10
-    vinserti128     m9, m9, xm8, 1
-    pmaddwd         m8, m9, [r5 + 3 * mmsize]
     paddd           m3, m8
-    pmaddwd         m8, m9, [r5 + 2 * mmsize]
-    pmaddwd         m9, [r5 + 1 * mmsize]
-    paddd           m5, m8
-    paddd           m7, m9
-    movu            xm8, [r0 + r4]                  ; m8 = row 11
-    punpckhwd       xm9, xm10, xm8
-    punpcklwd       xm10, xm8
-    vinserti128     m10, m10, xm9, 1
-    pmaddwd         m9, m10, [r5 + 3 * mmsize]
-    pmaddwd         m10, [r5 + 2 * mmsize]
-    paddd           m4, m9
-    paddd           m6, m10
-
-    lea             r4, [r3 * 3]
-%ifidn %1,sp
-    paddd           m0, m11
-    paddd           m1, m11
-    paddd           m2, m11
-    paddd           m3, m11
-    psrad           m0, 12
-    psrad           m1, 12
     psrad           m2, 12
     psrad           m3, 12
 %else
-    psrad           m0, 6
-    psrad           m1, 6
     psrad           m2, 6
     psrad           m3, 6
 %endif
-    packssdw        m0, m1
     packssdw        m2, m3
 %ifidn %1,sp
     packuswb        m0, m2
-    mova            m1, [interp8_hps_shuf]
-    vpermd          m0, m1, m0
     vextracti128    xm2, m0, 1
+    movd            [r2], xm0
+    pextrw          [r2 + 4], xm2, 0
+    pextrd          [r2 + r3], xm0, 1
+    pextrw          [r2 + r3 + 4], xm2, 2
+    pextrd          [r2 + r3 * 2], xm0, 2
+    pextrw          [r2 + r3 * 2 + 4], xm2, 4
+    pextrd          [r2 + r6], xm0, 3
+    pextrw          [r2 + r6 + 4], xm2, 6
+%else
     movq            [r2], xm0
     movhps          [r2 + r3], xm0
     movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r4], xm2
-%else
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    vextracti128    xm1, m0, 1
+    movhps          [r2 + r6], xm2
+    vextracti128    xm0, m0, 1
     vextracti128    xm3, m2, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r4], xm3
+    movd            [r2 + 8], xm0
+    pextrd          [r2 + r3 + 8], xm0, 2
+    movd            [r2 + r3 * 2 + 8], xm3
+    pextrd          [r2 + r6 + 8], xm3, 2
 %endif
-
+    lea             r2, [r2 + r3 * 4]
+    movu            xm1, [r0 + r4]                  ; m1 = row 7
+    punpckhwd       xm0, xm6, xm1
+    punpcklwd       xm6, xm1
+    vinserti128     m6, m6, xm0, 1
+    pmaddwd         m0, m6, [r5 + 1 * mmsize]
+    pmaddwd         m6, [r5]
+    paddd           m4, m0
     lea             r0, [r0 + r1 * 4]
-    movu            xm9, [r0]                       ; m9 = row 12
-    punpckhwd       xm3, xm8, xm9
-    punpcklwd       xm8, xm9
-    vinserti128     m8, m8, xm3, 1
-    pmaddwd         m3, m8, [r5 + 3 * mmsize]
-    pmaddwd         m8, [r5 + 2 * mmsize]
-    paddd           m5, m3
-    paddd           m7, m8
-    movu            xm3, [r0 + r1]                  ; m3 = row 13
-    punpckhwd       xm0, xm9, xm3
-    punpcklwd       xm9, xm3
-    vinserti128     m9, m9, xm0, 1
-    pmaddwd         m9, [r5 + 3 * mmsize]
-    paddd           m6, m9
-    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
-    punpckhwd       xm9, xm3, xm0
-    punpcklwd       xm3, xm0
-    vinserti128     m3, m3, xm9, 1
-    pmaddwd         m3, [r5 + 3 * mmsize]
-    paddd           m7, m3
+    movu            xm0, [r0]                       ; m0 = row 8
+    punpckhwd       xm2, xm1, xm0
+    punpcklwd       xm1, xm0
+    vinserti128     m1, m1, xm2, 1
+    pmaddwd         m2, m1, [r5 + 1 * mmsize]
+    pmaddwd         m1, [r5]
+    paddd           m5, m2
+%ifidn %1,sp
+    paddd           m4, m8
+    paddd           m5, m8
+    psrad           m4, 12
+    psrad           m5, 12
+%else
+    psrad           m4, 6
+    psrad           m5, 6
+%endif
+    packssdw        m4, m5
+
+    movu            xm2, [r0 + r1]                  ; m2 = row 9
+    punpckhwd       xm5, xm0, xm2
+    punpcklwd       xm0, xm2
+    vinserti128     m0, m0, xm5, 1
+    pmaddwd         m5, m0, [r5 + 1 * mmsize]
+    paddd           m6, m5
+    pmaddwd         m0, [r5]
+    movu            xm5, [r0 + r1 * 2]              ; m5 = row 10
+    punpckhwd       xm7, xm2, xm5
+    punpcklwd       xm2, xm5
+    vinserti128     m2, m2, xm7, 1
+    pmaddwd         m7, m2, [r5 + 1 * mmsize]
+    paddd           m1, m7
+    pmaddwd         m2, [r5]
 
 %ifidn %1,sp
-    paddd           m4, m11
-    paddd           m5, m11
-    paddd           m6, m11
-    paddd           m7, m11
-    psrad           m4, 12
-    psrad           m5, 12
+    paddd           m6, m8
+    paddd           m1, m8
     psrad           m6, 12
-    psrad           m7, 12
+    psrad           m1, 12
 %else
-    psrad           m4, 6
-    psrad           m5, 6
     psrad           m6, 6
-    psrad           m7, 6
+    psrad           m1, 6
 %endif
-    packssdw        m4, m5
-    packssdw        m6, m7
-    lea             r2, [r2 + r3 * 4]
+    packssdw        m6, m1
 %ifidn %1,sp
     packuswb        m4, m6
-    vpermd          m4, m1, m4
     vextracti128    xm6, m4, 1
+    movd            [r2], xm4
+    pextrw          [r2 + 4], xm6, 0
+    pextrd          [r2 + r3], xm4, 1
+    pextrw          [r2 + r3 + 4], xm6, 2
+    pextrd          [r2 + r3 * 2], xm4, 2
+    pextrw          [r2 + r3 * 2 + 4], xm6, 4
+    pextrd          [r2 + r6], xm4, 3
+    pextrw          [r2 + r6 + 4], xm6, 6
+%else
     movq            [r2], xm4
     movhps          [r2 + r3], xm4
     movq            [r2 + r3 * 2], xm6
-    movhps          [r2 + r4], xm6
-%else
-    vpermq          m4, m4, 11011000b
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm5, m4, 1
-    vextracti128    xm7, m6, 1
-    movu            [r2], xm4
-    movu            [r2 + r3], xm5
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r4], xm7
-%endif
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_S_AVX2_8x8 sp
-    FILTER_VER_LUMA_S_AVX2_8x8 ss
-
-%macro FILTER_VER_LUMA_S_AVX2_8xN 2
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_8x%2, 4, 9, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-    add             r1d, r1d
-
-%ifdef PIC
-    lea             r5, [pw_LumaCoeffVer]
-    add             r5, r4
-%else
-    lea             r5, [pw_LumaCoeffVer + r4]
+    movhps          [r2 + r6], xm6
+    vextracti128    xm4, m4, 1
+    vextracti128    xm1, m6, 1
+    movd            [r2 + 8], xm4
+    pextrd          [r2 + r3 + 8], xm4, 2
+    movd            [r2 + r3 * 2 + 8], xm1
+    pextrd          [r2 + r6 + 8], xm1, 2
 %endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
+    lea             r2, [r2 + r3 * 4]
+    movu            xm7, [r0 + r4]                  ; m7 = row 11
+    punpckhwd       xm1, xm5, xm7
+    punpcklwd       xm5, xm7
+    vinserti128     m5, m5, xm1, 1
+    pmaddwd         m1, m5, [r5 + 1 * mmsize]
+    paddd           m0, m1
+    pmaddwd         m5, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm1, [r0]                       ; m1 = row 12
+    punpckhwd       xm4, xm7, xm1
+    punpcklwd       xm7, xm1
+    vinserti128     m7, m7, xm4, 1
+    pmaddwd         m4, m7, [r5 + 1 * mmsize]
+    paddd           m2, m4
+    pmaddwd         m7, [r5]
 %ifidn %1,sp
-    mova            m14, [pd_526336]
+    paddd           m0, m8
+    paddd           m2, m8
+    psrad           m0, 12
+    psrad           m2, 12
 %else
-    add             r3d, r3d
+    psrad           m0, 6
+    psrad           m2, 6
 %endif
-    lea             r6, [r3 * 3]
-    lea             r7, [r1 * 4]
-    mov             r8d, %2 / 16
-.loopH:
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
+    packssdw        m0, m2
+
+    movu            xm4, [r0 + r1]                  ; m4 = row 13
+    punpckhwd       xm2, xm1, xm4
+    punpcklwd       xm1, xm4
+    vinserti128     m1, m1, xm2, 1
+    pmaddwd         m2, m1, [r5 + 1 * mmsize]
+    paddd           m5, m2
     pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    paddd           m1, m5
-    pmaddwd         m3, [r5]
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 14
+    punpckhwd       xm6, xm4, xm2
+    punpcklwd       xm4, xm2
     vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 2 * mmsize]
-    paddd           m0, m6
     pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m2, m6
+    paddd           m7, m6
     pmaddwd         m4, [r5]
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm7, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddwd         m7, m5, [r5 + 2 * mmsize]
-    paddd           m1, m7
-    pmaddwd         m7, m5, [r5 + 1 * mmsize]
-    paddd           m3, m7
-    pmaddwd         m5, [r5]
-    movu            xm7, [r0 + r4]                  ; m7 = row 7
-    punpckhwd       xm8, xm6, xm7
-    punpcklwd       xm6, xm7
-    vinserti128     m6, m6, xm8, 1
-    pmaddwd         m8, m6, [r5 + 3 * mmsize]
-    paddd           m0, m8
-    pmaddwd         m8, m6, [r5 + 2 * mmsize]
-    paddd           m2, m8
-    pmaddwd         m8, m6, [r5 + 1 * mmsize]
-    paddd           m4, m8
-    pmaddwd         m6, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm8, [r0]                       ; m8 = row 8
-    punpckhwd       xm9, xm7, xm8
-    punpcklwd       xm7, xm8
-    vinserti128     m7, m7, xm9, 1
-    pmaddwd         m9, m7, [r5 + 3 * mmsize]
-    paddd           m1, m9
-    pmaddwd         m9, m7, [r5 + 2 * mmsize]
-    paddd           m3, m9
-    pmaddwd         m9, m7, [r5 + 1 * mmsize]
-    paddd           m5, m9
-    pmaddwd         m7, [r5]
-    movu            xm9, [r0 + r1]                  ; m9 = row 9
-    punpckhwd       xm10, xm8, xm9
-    punpcklwd       xm8, xm9
-    vinserti128     m8, m8, xm10, 1
-    pmaddwd         m10, m8, [r5 + 3 * mmsize]
-    paddd           m2, m10
-    pmaddwd         m10, m8, [r5 + 2 * mmsize]
-    paddd           m4, m10
-    pmaddwd         m10, m8, [r5 + 1 * mmsize]
-    paddd           m6, m10
-    pmaddwd         m8, [r5]
-    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
-    punpckhwd       xm11, xm9, xm10
-    punpcklwd       xm9, xm10
-    vinserti128     m9, m9, xm11, 1
-    pmaddwd         m11, m9, [r5 + 3 * mmsize]
-    paddd           m3, m11
-    pmaddwd         m11, m9, [r5 + 2 * mmsize]
-    paddd           m5, m11
-    pmaddwd         m11, m9, [r5 + 1 * mmsize]
-    paddd           m7, m11
-    pmaddwd         m9, [r5]
-    movu            xm11, [r0 + r4]                 ; m11 = row 11
-    punpckhwd       xm12, xm10, xm11
-    punpcklwd       xm10, xm11
-    vinserti128     m10, m10, xm12, 1
-    pmaddwd         m12, m10, [r5 + 3 * mmsize]
-    paddd           m4, m12
-    pmaddwd         m12, m10, [r5 + 2 * mmsize]
-    paddd           m6, m12
-    pmaddwd         m12, m10, [r5 + 1 * mmsize]
-    paddd           m8, m12
-    pmaddwd         m10, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm12, [r0]                      ; m12 = row 12
-    punpckhwd       xm13, xm11, xm12
-    punpcklwd       xm11, xm12
-    vinserti128     m11, m11, xm13, 1
-    pmaddwd         m13, m11, [r5 + 3 * mmsize]
-    paddd           m5, m13
-    pmaddwd         m13, m11, [r5 + 2 * mmsize]
-    paddd           m7, m13
-    pmaddwd         m13, m11, [r5 + 1 * mmsize]
-    paddd           m9, m13
-    pmaddwd         m11, [r5]
+%ifidn %1,sp
+    paddd           m5, m8
+    paddd           m7, m8
+    psrad           m5, 12
+    psrad           m7, 12
+%else
+    psrad           m5, 6
+    psrad           m7, 6
+%endif
+    packssdw        m5, m7
+%ifidn %1,sp
+    packuswb        m0, m5
+    vextracti128    xm5, m0, 1
+    movd            [r2], xm0
+    pextrw          [r2 + 4], xm5, 0
+    pextrd          [r2 + r3], xm0, 1
+    pextrw          [r2 + r3 + 4], xm5, 2
+    pextrd          [r2 + r3 * 2], xm0, 2
+    pextrw          [r2 + r3 * 2 + 4], xm5, 4
+    pextrd          [r2 + r6], xm0, 3
+    pextrw          [r2 + r6 + 4], xm5, 6
+%else
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm5
+    movhps          [r2 + r6], xm5
+    vextracti128    xm0, m0, 1
+    vextracti128    xm7, m5, 1
+    movd            [r2 + 8], xm0
+    pextrd          [r2 + r3 + 8], xm0, 2
+    movd            [r2 + r3 * 2 + 8], xm7
+    pextrd          [r2 + r6 + 8], xm7, 2
+%endif
+    lea             r2, [r2 + r3 * 4]
 
+    movu            xm6, [r0 + r4]                  ; m6 = row 15
+    punpckhwd       xm5, xm2, xm6
+    punpcklwd       xm2, xm6
+    vinserti128     m2, m2, xm5, 1
+    pmaddwd         m5, m2, [r5 + 1 * mmsize]
+    paddd           m1, m5
+    pmaddwd         m2, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm0, [r0]                       ; m0 = row 16
+    punpckhwd       xm5, xm6, xm0
+    punpcklwd       xm6, xm0
+    vinserti128     m6, m6, xm5, 1
+    pmaddwd         m5, m6, [r5 + 1 * mmsize]
+    paddd           m4, m5
+    pmaddwd         m6, [r5]
 %ifidn %1,sp
-    paddd           m0, m14
-    paddd           m1, m14
-    paddd           m2, m14
-    paddd           m3, m14
-    paddd           m4, m14
-    paddd           m5, m14
-    psrad           m0, 12
+    paddd           m1, m8
+    paddd           m4, m8
     psrad           m1, 12
-    psrad           m2, 12
-    psrad           m3, 12
     psrad           m4, 12
-    psrad           m5, 12
 %else
-    psrad           m0, 6
     psrad           m1, 6
-    psrad           m2, 6
-    psrad           m3, 6
     psrad           m4, 6
-    psrad           m5, 6
 %endif
-    packssdw        m0, m1
-    packssdw        m2, m3
-    packssdw        m4, m5
+    packssdw        m1, m4
+
+    movu            xm5, [r0 + r1]                  ; m5 = row 17
+    punpckhwd       xm4, xm0, xm5
+    punpcklwd       xm0, xm5
+    vinserti128     m0, m0, xm4, 1
+    pmaddwd         m0, [r5 + 1 * mmsize]
+    paddd           m2, m0
+    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
+    punpckhwd       xm0, xm5, xm4
+    punpcklwd       xm5, xm4
+    vinserti128     m5, m5, xm0, 1
+    pmaddwd         m5, [r5 + 1 * mmsize]
+    paddd           m6, m5
 %ifidn %1,sp
-    packuswb        m0, m2
-    mova            m1, [interp8_hps_shuf]
-    vpermd          m0, m1, m0
-    vextracti128    xm2, m0, 1
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
+    paddd           m2, m8
+    paddd           m6, m8
+    psrad           m2, 12
+    psrad           m6, 12
+%else
+    psrad           m2, 6
+    psrad           m6, 6
+%endif
+    packssdw        m2, m6
+%ifidn %1,sp
+    packuswb        m1, m2
+    vextracti128    xm2, m1, 1
+    movd            [r2], xm1
+    pextrw          [r2 + 4], xm2, 0
+    pextrd          [r2 + r3], xm1, 1
+    pextrw          [r2 + r3 + 4], xm2, 2
+    pextrd          [r2 + r3 * 2], xm1, 2
+    pextrw          [r2 + r3 * 2 + 4], xm2, 4
+    pextrd          [r2 + r6], xm1, 3
+    pextrw          [r2 + r6 + 4], xm2, 6
+%else
+    movq            [r2], xm1
+    movhps          [r2 + r3], xm1
     movq            [r2 + r3 * 2], xm2
     movhps          [r2 + r6], xm2
+    vextracti128    xm4, m1, 1
+    vextracti128    xm6, m2, 1
+    movd            [r2 + 8], xm4
+    pextrd          [r2 + r3 + 8], xm4, 2
+    movd            [r2 + r3 * 2 + 8], xm6
+    pextrd          [r2 + r6 + 8], xm6, 2
+%endif
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_6x16 sp
+    FILTER_VER_CHROMA_S_AVX2_6x16 ss
+
+;---------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vertical_ss_%1x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;---------------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_SS_W2_4R 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_ss_%1x%2, 5, 6, 5
+
+    add       r1d, r1d
+    add       r3d, r3d
+    sub       r0, r1
+    shl       r4d, 5
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r5, [r5 + r4]
 %else
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
+    lea       r5, [tab_ChromaCoeffV + r4]
+%endif
+
+    mov       r4d, (%2/4)
+
+.loopH:
+    PROCESS_CHROMA_SP_W2_4R r5
+
+    psrad     m0, 6
+    psrad     m2, 6
+
+    packssdw  m0, m2
+
+    movd      [r2], m0
+    pextrd    [r2 + r3], m0, 1
+    lea       r2, [r2 + 2 * r3]
+    pextrd    [r2], m0, 2
+    pextrd    [r2 + r3], m0, 3
+
+    lea       r2, [r2 + 2 * r3]
+
+    dec       r4d
+    jnz       .loopH
+
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_SS_W2_4R 2, 4
+    FILTER_VER_CHROMA_SS_W2_4R 2, 8
+
+    FILTER_VER_CHROMA_SS_W2_4R 2, 16
+
+;---------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ss_4x2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;---------------------------------------------------------------------------------------------------------------
+INIT_XMM sse2
+cglobal interp_4tap_vert_ss_4x2, 5, 6, 4
+
+    add        r1d, r1d
+    add        r3d, r3d
+    sub        r0, r1
+    shl        r4d, 5
+
+%ifdef PIC
+    lea        r5, [tab_ChromaCoeffV]
+    lea        r5, [r5 + r4]
+%else
+    lea        r5, [tab_ChromaCoeffV + r4]
+%endif
+
+    movq       m0, [r0]
+    movq       m1, [r0 + r1]
+    punpcklwd  m0, m1                          ;m0=[0 1]
+    pmaddwd    m0, [r5 + 0 *16]                ;m0=[0+1]  Row1
+
+    lea        r0, [r0 + 2 * r1]
+    movq       m2, [r0]
+    punpcklwd  m1, m2                          ;m1=[1 2]
+    pmaddwd    m1, [r5 + 0 *16]                ;m1=[1+2]  Row2
+
+    movq       m3, [r0 + r1]
+    punpcklwd  m2, m3                          ;m4=[2 3]
+    pmaddwd    m2, [r5 + 1 * 16]
+    paddd      m0, m2                          ;m0=[0+1+2+3]  Row1 done
+    psrad      m0, 6
+
+    movq       m2, [r0 + 2 * r1]
+    punpcklwd  m3, m2                          ;m5=[3 4]
+    pmaddwd    m3, [r5 + 1 * 16]
+    paddd      m1, m3                          ;m1=[1+2+3+4]  Row2 done
+    psrad      m1, 6
+
+    packssdw   m0, m1
+
+    movlps     [r2], m0
+    movhps     [r2 + r3], m0
+
+    RET
+
+;-------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vertical_ss_6x8(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_SS_W6_H4 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_ss_6x%2, 5, 7, 6
+
+    add       r1d, r1d
+    add       r3d, r3d
+    sub       r0, r1
+    shl       r4d, 5
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r6, [r5 + r4]
+%else
+    lea       r6, [tab_ChromaCoeffV + r4]
 %endif
 
-    movu            xm13, [r0 + r1]                 ; m13 = row 13
-    punpckhwd       xm0, xm12, xm13
-    punpcklwd       xm12, xm13
-    vinserti128     m12, m12, xm0, 1
-    pmaddwd         m0, m12, [r5 + 3 * mmsize]
-    paddd           m6, m0
-    pmaddwd         m0, m12, [r5 + 2 * mmsize]
-    paddd           m8, m0
-    pmaddwd         m0, m12, [r5 + 1 * mmsize]
-    paddd           m10, m0
-    pmaddwd         m12, [r5]
-    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
-    punpckhwd       xm2, xm13, xm0
-    punpcklwd       xm13, xm0
-    vinserti128     m13, m13, xm2, 1
-    pmaddwd         m2, m13, [r5 + 3 * mmsize]
-    paddd           m7, m2
-    pmaddwd         m2, m13, [r5 + 2 * mmsize]
-    paddd           m9, m2
-    pmaddwd         m2, m13, [r5 + 1 * mmsize]
-    paddd           m11, m2
-    pmaddwd         m13, [r5]
+    mov       r4d, %2/4
+
+.loopH:
+    PROCESS_CHROMA_SP_W4_4R
+
+    psrad     m0, 6
+    psrad     m1, 6
+    psrad     m2, 6
+    psrad     m3, 6
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+
+    movlps    [r2], m0
+    movhps    [r2 + r3], m0
+    lea       r5, [r2 + 2 * r3]
+    movlps    [r5], m2
+    movhps    [r5 + r3], m2
+
+    lea       r5, [4 * r1 - 2 * 4]
+    sub       r0, r5
+    add       r2, 2 * 4
+
+    PROCESS_CHROMA_SP_W2_4R r6
+
+    psrad     m0, 6
+    psrad     m2, 6
+
+    packssdw  m0, m2
+
+    movd      [r2], m0
+    pextrd    [r2 + r3], m0, 1
+    lea       r2, [r2 + 2 * r3]
+    pextrd    [r2], m0, 2
+    pextrd    [r2 + r3], m0, 3
+
+    sub       r0, 2 * 4
+    lea       r2, [r2 + 2 * r3 - 2 * 4]
+
+    dec       r4d
+    jnz       .loopH
+
+    RET
+%endmacro
 
-%ifidn %1,sp
-    paddd           m6, m14
-    paddd           m7, m14
-    psrad           m6, 12
-    psrad           m7, 12
+    FILTER_VER_CHROMA_SS_W6_H4 6, 8
+
+    FILTER_VER_CHROMA_SS_W6_H4 6, 16
+
+
+;----------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ss_8x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;----------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_SS_W8_H2 2
+INIT_XMM sse2
+cglobal interp_4tap_vert_ss_%1x%2, 5, 6, 7
+
+    add       r1d, r1d
+    add       r3d, r3d
+    sub       r0, r1
+    shl       r4d, 5
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r5, [r5 + r4]
 %else
-    psrad           m6, 6
-    psrad           m7, 6
+    lea       r5, [tab_ChromaCoeffV + r4]
 %endif
-    packssdw        m6, m7
-    lea             r2, [r2 + r3 * 4]
 
-%ifidn %1,sp
-    packuswb        m4, m6
-    vpermd          m4, m1, m4
-    vextracti128    xm6, m4, 1
-    movq            [r2], xm4
-    movhps          [r2 + r3], xm4
-    movq            [r2 + r3 * 2], xm6
-    movhps          [r2 + r6], xm6
+    mov       r4d, %2/2
+.loopH:
+    PROCESS_CHROMA_SP_W8_2R
+
+    psrad     m0, 6
+    psrad     m1, 6
+    psrad     m2, 6
+    psrad     m3, 6
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+
+    movu      [r2], m0
+    movu      [r2 + r3], m2
+
+    lea       r2, [r2 + 2 * r3]
+
+    dec       r4d
+    jnz       .loopH
+
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_SS_W8_H2 8, 2
+    FILTER_VER_CHROMA_SS_W8_H2 8, 4
+    FILTER_VER_CHROMA_SS_W8_H2 8, 6
+    FILTER_VER_CHROMA_SS_W8_H2 8, 8
+    FILTER_VER_CHROMA_SS_W8_H2 8, 16
+    FILTER_VER_CHROMA_SS_W8_H2 8, 32
+
+    FILTER_VER_CHROMA_SS_W8_H2 8, 12
+    FILTER_VER_CHROMA_SS_W8_H2 8, 64
+
+;-----------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_ss_%1x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_LUMA_SS 2
+INIT_XMM sse2
+cglobal interp_8tap_vert_ss_%1x%2, 5, 7, 7 ,0-gprsize
+
+    add        r1d, r1d
+    add        r3d, r3d
+    lea        r5, [3 * r1]
+    sub        r0, r5
+    shl        r4d, 6
+
+%ifdef PIC
+    lea        r5, [tab_LumaCoeffV]
+    lea        r6, [r5 + r4]
 %else
-    vpermq          m6, m6, 11011000b
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm1, m4, 1
-    vextracti128    xm7, m6, 1
-    movu            [r2], xm4
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r6], xm7
+    lea        r6, [tab_LumaCoeffV + r4]
 %endif
 
-    movu            xm6, [r0 + r4]                  ; m6 = row 15
-    punpckhwd       xm5, xm0, xm6
-    punpcklwd       xm0, xm6
-    vinserti128     m0, m0, xm5, 1
-    pmaddwd         m5, m0, [r5 + 3 * mmsize]
-    paddd           m8, m5
-    pmaddwd         m5, m0, [r5 + 2 * mmsize]
-    paddd           m10, m5
-    pmaddwd         m5, m0, [r5 + 1 * mmsize]
-    paddd           m12, m5
-    pmaddwd         m0, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm2, [r0]                       ; m2 = row 16
-    punpckhwd       xm3, xm6, xm2
-    punpcklwd       xm6, xm2
-    vinserti128     m6, m6, xm3, 1
-    pmaddwd         m3, m6, [r5 + 3 * mmsize]
-    paddd           m9, m3
-    pmaddwd         m3, m6, [r5 + 2 * mmsize]
-    paddd           m11, m3
-    pmaddwd         m3, m6, [r5 + 1 * mmsize]
-    paddd           m13, m3
-    pmaddwd         m6, [r5]
-    movu            xm3, [r0 + r1]                  ; m3 = row 17
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 3 * mmsize]
-    paddd           m10, m4
-    pmaddwd         m4, m2, [r5 + 2 * mmsize]
-    paddd           m12, m4
-    pmaddwd         m2, [r5 + 1 * mmsize]
-    paddd           m0, m2
-    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
-    punpckhwd       xm2, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm2, 1
-    pmaddwd         m2, m3, [r5 + 3 * mmsize]
-    paddd           m11, m2
-    pmaddwd         m2, m3, [r5 + 2 * mmsize]
-    paddd           m13, m2
-    pmaddwd         m3, [r5 + 1 * mmsize]
-    paddd           m6, m3
-    movu            xm2, [r0 + r4]                  ; m2 = row 19
-    punpckhwd       xm7, xm4, xm2
-    punpcklwd       xm4, xm2
-    vinserti128     m4, m4, xm7, 1
-    pmaddwd         m7, m4, [r5 + 3 * mmsize]
-    paddd           m12, m7
-    pmaddwd         m4, [r5 + 2 * mmsize]
-    paddd           m0, m4
-    lea             r0, [r0 + r1 * 4]
-    movu            xm7, [r0]                       ; m7 = row 20
-    punpckhwd       xm3, xm2, xm7
-    punpcklwd       xm2, xm7
-    vinserti128     m2, m2, xm3, 1
-    pmaddwd         m3, m2, [r5 + 3 * mmsize]
-    paddd           m13, m3
-    pmaddwd         m2, [r5 + 2 * mmsize]
-    paddd           m6, m2
-    movu            xm3, [r0 + r1]                  ; m3 = row 21
-    punpckhwd       xm2, xm7, xm3
-    punpcklwd       xm7, xm3
-    vinserti128     m7, m7, xm2, 1
-    pmaddwd         m7, [r5 + 3 * mmsize]
-    paddd           m0, m7
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 22
-    punpckhwd       xm7, xm3, xm2
-    punpcklwd       xm3, xm2
-    vinserti128     m3, m3, xm7, 1
-    pmaddwd         m3, [r5 + 3 * mmsize]
-    paddd           m6, m3
+    mov        dword [rsp], %2/4
+.loopH:
+    mov        r4d, (%1/4)
+.loopW:
+    movq       m0, [r0]
+    movq       m1, [r0 + r1]
+    punpcklwd  m0, m1                          ;m0=[0 1]
+    pmaddwd    m0, [r6 + 0 *16]                ;m0=[0+1]  Row1
+
+    lea        r0, [r0 + 2 * r1]
+    movq       m4, [r0]
+    punpcklwd  m1, m4                          ;m1=[1 2]
+    pmaddwd    m1, [r6 + 0 *16]                ;m1=[1+2]  Row2
+
+    movq       m5, [r0 + r1]
+    punpcklwd  m4, m5                          ;m4=[2 3]
+    pmaddwd    m2, m4, [r6 + 0 *16]            ;m2=[2+3]  Row3
+    pmaddwd    m4, [r6 + 1 * 16]
+    paddd      m0, m4                          ;m0=[0+1+2+3]  Row1
+
+    lea        r0, [r0 + 2 * r1]
+    movq       m4, [r0]
+    punpcklwd  m5, m4                          ;m5=[3 4]
+    pmaddwd    m3, m5, [r6 + 0 *16]            ;m3=[3+4]  Row4
+    pmaddwd    m5, [r6 + 1 * 16]
+    paddd      m1, m5                          ;m1 = [1+2+3+4]  Row2
+
+    movq       m5, [r0 + r1]
+    punpcklwd  m4, m5                          ;m4=[4 5]
+    pmaddwd    m6, m4, [r6 + 1 * 16]
+    paddd      m2, m6                          ;m2=[2+3+4+5]  Row3
+    pmaddwd    m4, [r6 + 2 * 16]
+    paddd      m0, m4                          ;m0=[0+1+2+3+4+5]  Row1
+
+    lea        r0, [r0 + 2 * r1]
+    movq       m4, [r0]
+    punpcklwd  m5, m4                          ;m5=[5 6]
+    pmaddwd    m6, m5, [r6 + 1 * 16]
+    paddd      m3, m6                          ;m3=[3+4+5+6]  Row4
+    pmaddwd    m5, [r6 + 2 * 16]
+    paddd      m1, m5                          ;m1=[1+2+3+4+5+6]  Row2
+
+    movq       m5, [r0 + r1]
+    punpcklwd  m4, m5                          ;m4=[6 7]
+    pmaddwd    m6, m4, [r6 + 2 * 16]
+    paddd      m2, m6                          ;m2=[2+3+4+5+6+7]  Row3
+    pmaddwd    m4, [r6 + 3 * 16]
+    paddd      m0, m4                          ;m0=[0+1+2+3+4+5+6+7]  Row1 end
+    psrad      m0, 6
+
+    lea        r0, [r0 + 2 * r1]
+    movq       m4, [r0]
+    punpcklwd  m5, m4                          ;m5=[7 8]
+    pmaddwd    m6, m5, [r6 + 2 * 16]
+    paddd      m3, m6                          ;m3=[3+4+5+6+7+8]  Row4
+    pmaddwd    m5, [r6 + 3 * 16]
+    paddd      m1, m5                          ;m1=[1+2+3+4+5+6+7+8]  Row2 end
+    psrad      m1, 6
+
+    packssdw   m0, m1
+
+    movlps     [r2], m0
+    movhps     [r2 + r3], m0
+
+    movq       m5, [r0 + r1]
+    punpcklwd  m4, m5                          ;m4=[8 9]
+    pmaddwd    m4, [r6 + 3 * 16]
+    paddd      m2, m4                          ;m2=[2+3+4+5+6+7+8+9]  Row3 end
+    psrad      m2, 6
 
-%ifidn %1,sp
-    paddd           m8, m14
-    paddd           m9, m14
-    paddd           m10, m14
-    paddd           m11, m14
-    paddd           m12, m14
-    paddd           m13, m14
-    paddd           m0, m14
-    paddd           m6, m14
-    psrad           m8, 12
-    psrad           m9, 12
-    psrad           m10, 12
-    psrad           m11, 12
-    psrad           m12, 12
-    psrad           m13, 12
-    psrad           m0, 12
-    psrad           m6, 12
+    movq       m4, [r0 + 2 * r1]
+    punpcklwd  m5, m4                          ;m5=[9 10]
+    pmaddwd    m5, [r6 + 3 * 16]
+    paddd      m3, m5                          ;m3=[3+4+5+6+7+8+9+10]  Row4 end
+    psrad      m3, 6
+
+    packssdw   m2, m3
+
+    movlps     [r2 + 2 * r3], m2
+    lea        r5, [3 * r3]
+    movhps     [r2 + r5], m2
+
+    lea        r5, [8 * r1 - 2 * 4]
+    sub        r0, r5
+    add        r2, 2 * 4
+
+    dec        r4d
+    jnz        .loopW
+
+    lea        r0, [r0 + 4 * r1 - 2 * %1]
+    lea        r2, [r2 + 4 * r3 - 2 * %1]
+
+    dec        dword [rsp]
+    jnz        .loopH
+
+    RET
+%endmacro
+
+    FILTER_VER_LUMA_SS 4, 4
+    FILTER_VER_LUMA_SS 8, 8
+    FILTER_VER_LUMA_SS 8, 4
+    FILTER_VER_LUMA_SS 4, 8
+    FILTER_VER_LUMA_SS 16, 16
+    FILTER_VER_LUMA_SS 16, 8
+    FILTER_VER_LUMA_SS 8, 16
+    FILTER_VER_LUMA_SS 16, 12
+    FILTER_VER_LUMA_SS 12, 16
+    FILTER_VER_LUMA_SS 16, 4
+    FILTER_VER_LUMA_SS 4, 16
+    FILTER_VER_LUMA_SS 32, 32
+    FILTER_VER_LUMA_SS 32, 16
+    FILTER_VER_LUMA_SS 16, 32
+    FILTER_VER_LUMA_SS 32, 24
+    FILTER_VER_LUMA_SS 24, 32
+    FILTER_VER_LUMA_SS 32, 8
+    FILTER_VER_LUMA_SS 8, 32
+    FILTER_VER_LUMA_SS 64, 64
+    FILTER_VER_LUMA_SS 64, 32
+    FILTER_VER_LUMA_SS 32, 64
+    FILTER_VER_LUMA_SS 64, 48
+    FILTER_VER_LUMA_SS 48, 64
+    FILTER_VER_LUMA_SS 64, 16
+    FILTER_VER_LUMA_SS 16, 64
+
+%macro FILTER_VER_LUMA_AVX2_4x4 1
+INIT_YMM avx2
+cglobal interp_8tap_vert_%1_4x4, 4, 6, 7
+    mov             r4d, r4m
+    add             r1d, r1d
+    shl             r4d, 7
+
+%ifdef PIC
+    lea             r5, [pw_LumaCoeffVer]
+    add             r5, r4
 %else
-    psrad           m8, 6
-    psrad           m9, 6
-    psrad           m10, 6
-    psrad           m11, 6
-    psrad           m12, 6
-    psrad           m13, 6
-    psrad           m0, 6
-    psrad           m6, 6
+    lea             r5, [pw_LumaCoeffVer + r4]
 %endif
-    packssdw        m8, m9
-    packssdw        m10, m11
-    packssdw        m12, m13
-    packssdw        m0, m6
-    lea             r2, [r2 + r3 * 4]
+
+    lea             r4, [r1 * 3]
+    sub             r0, r4
 
 %ifidn %1,sp
-    packuswb        m8, m10
-    packuswb        m12, m0
-    vpermd          m8, m1, m8
-    vpermd          m12, m1, m12
-    vextracti128    xm10, m8, 1
-    vextracti128    xm0, m12, 1
-    movq            [r2], xm8
-    movhps          [r2 + r3], xm8
-    movq            [r2 + r3 * 2], xm10
-    movhps          [r2 + r6], xm10
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm12
-    movhps          [r2 + r3], xm12
-    movq            [r2 + r3 * 2], xm0
-    movhps          [r2 + r6], xm0
+    mova            m6, [pd_526336]
 %else
-    vpermq          m8, m8, 11011000b
-    vpermq          m10, m10, 11011000b
-    vpermq          m12, m12, 11011000b
-    vpermq          m0, m0, 11011000b
-    vextracti128    xm9, m8, 1
-    vextracti128    xm11, m10, 1
-    vextracti128    xm13, m12, 1
-    vextracti128    xm6, m0, 1
-    movu            [r2], xm8
-    movu            [r2 + r3], xm9
-    movu            [r2 + r3 * 2], xm10
-    movu            [r2 + r6], xm11
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm12
-    movu            [r2 + r3], xm13
-    movu            [r2 + r3 * 2], xm0
-    movu            [r2 + r6], xm6
-%endif
-
-    lea             r2, [r2 + r3 * 4]
-    sub             r0, r7
-    dec             r8d
-    jnz             .loopH
-    RET
+    add             r3d, r3d
 %endif
-%endmacro
-
-    FILTER_VER_LUMA_S_AVX2_8xN sp, 16
-    FILTER_VER_LUMA_S_AVX2_8xN sp, 32
-    FILTER_VER_LUMA_S_AVX2_8xN ss, 16
-    FILTER_VER_LUMA_S_AVX2_8xN ss, 32
 
-%macro PROCESS_LUMA_S_AVX2_W8_4R 1
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
+    movq            xm0, [r0]
+    movq            xm1, [r0 + r1]
     punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
+    movq            xm2, [r0 + r1 * 2]
     punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
+    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
+    pmaddwd         m0, [r5]
+    movq            xm3, [r0 + r4]
     punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
+    lea             r0, [r0 + 4 * r1]
+    movq            xm4, [r0]
     punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    paddd           m1, m5
-    pmaddwd         m3, [r5]
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 2 * mmsize]
-    paddd           m0, m6
+    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
+    pmaddwd         m5, m2, [r5 + 1 * mmsize]
+    pmaddwd         m2, [r5]
+    paddd           m0, m5
+    movq            xm3, [r0 + r1]
+    punpcklwd       xm4, xm3
+    movq            xm1, [r0 + r1 * 2]
+    punpcklwd       xm3, xm1
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
+    pmaddwd         m5, m4, [r5 + 2 * mmsize]
     pmaddwd         m4, [r5 + 1 * mmsize]
+    paddd           m0, m5
     paddd           m2, m4
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm4, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm4, 1
-    pmaddwd         m4, m5, [r5 + 2 * mmsize]
-    paddd           m1, m4
-    pmaddwd         m5, [r5 + 1 * mmsize]
-    paddd           m3, m5
-    movu            xm4, [r0 + r4]                  ; m4 = row 7
-    punpckhwd       xm5, xm6, xm4
-    punpcklwd       xm6, xm4
-    vinserti128     m6, m6, xm5, 1
-    pmaddwd         m5, m6, [r5 + 3 * mmsize]
+    movq            xm3, [r0 + r4]
+    punpcklwd       xm1, xm3
+    lea             r0, [r0 + 4 * r1]
+    movq            xm4, [r0]
+    punpcklwd       xm3, xm4
+    vinserti128     m1, m1, xm3, 1                  ; m1 = [8 7 7 6]
+    pmaddwd         m5, m1, [r5 + 3 * mmsize]
+    pmaddwd         m1, [r5 + 2 * mmsize]
     paddd           m0, m5
-    pmaddwd         m6, [r5 + 2 * mmsize]
-    paddd           m2, m6
-    lea             r0, [r0 + r1 * 4]
-    movu            xm5, [r0]                       ; m5 = row 8
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 3 * mmsize]
-    paddd           m1, m6
-    pmaddwd         m4, [r5 + 2 * mmsize]
-    paddd           m3, m4
-    movu            xm6, [r0 + r1]                  ; m6 = row 9
-    punpckhwd       xm4, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm4, 1
-    pmaddwd         m5, [r5 + 3 * mmsize]
-    paddd           m2, m5
-    movu            xm4, [r0 + r1 * 2]              ; m4 = row 10
-    punpckhwd       xm5, xm6, xm4
-    punpcklwd       xm6, xm4
-    vinserti128     m6, m6, xm5, 1
-    pmaddwd         m6, [r5 + 3 * mmsize]
-    paddd           m3, m6
-
-%ifidn %1,sp
-    paddd           m0, m7
-    paddd           m1, m7
-    paddd           m2, m7
-    paddd           m3, m7
+    paddd           m2, m1
+    movq            xm3, [r0 + r1]
+    punpcklwd       xm4, xm3
+    movq            xm1, [r0 + 2 * r1]
+    punpcklwd       xm3, xm1
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [A 9 9 8]
+    pmaddwd         m4, [r5 + 3 * mmsize]
+    paddd           m2, m4
+
+%ifidn %1,sp
+    paddd           m0, m6
+    paddd           m2, m6
     psrad           m0, 12
-    psrad           m1, 12
     psrad           m2, 12
-    psrad           m3, 12
 %else
     psrad           m0, 6
-    psrad           m1, 6
     psrad           m2, 6
-    psrad           m3, 6
 %endif
-    packssdw        m0, m1
-    packssdw        m2, m3
-%ifidn %1,sp
-    packuswb        m0, m2
-    mova            m4, [interp8_hps_shuf]
-    vpermd          m0, m4, m0
+    packssdw        m0, m2
     vextracti128    xm2, m0, 1
+    lea             r4, [r3 * 3]
+
+%ifidn %1,sp
+    packuswb        xm0, xm2
+    movd            [r2], xm0
+    pextrd          [r2 + r3], xm0, 2
+    pextrd          [r2 + r3 * 2], xm0, 1
+    pextrd          [r2 + r4], xm0, 3
 %else
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
+    movq            [r2], xm0
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r4], xm2
 %endif
+    RET
 %endmacro
 
-%macro FILTER_VER_LUMA_S_AVX2_8x4 1
+    FILTER_VER_LUMA_AVX2_4x4 sp
+    FILTER_VER_LUMA_AVX2_4x4 ss
+
+%macro FILTER_VER_LUMA_AVX2_4x8 1
 INIT_YMM avx2
-cglobal interp_8tap_vert_%1_8x4, 4, 6, 8
+cglobal interp_8tap_vert_%1_4x8, 4, 7, 8
     mov             r4d, r4m
     shl             r4d, 7
     add             r1d, r1d
@@ -24137,390 +19802,355 @@ cglobal interp_8tap_vert_%1_8x4, 4, 6, 8
 
     lea             r4, [r1 * 3]
     sub             r0, r4
+
 %ifidn %1,sp
     mova            m7, [pd_526336]
 %else
     add             r3d, r3d
 %endif
+    lea             r6, [r3 * 3]
+
+    movq            xm0, [r0]
+    movq            xm1, [r0 + r1]
+    punpcklwd       xm0, xm1
+    movq            xm2, [r0 + r1 * 2]
+    punpcklwd       xm1, xm2
+    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
+    pmaddwd         m0, [r5]
+    movq            xm3, [r0 + r4]
+    punpcklwd       xm2, xm3
+    lea             r0, [r0 + 4 * r1]
+    movq            xm4, [r0]
+    punpcklwd       xm3, xm4
+    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
+    pmaddwd         m5, m2, [r5 + 1 * mmsize]
+    pmaddwd         m2, [r5]
+    paddd           m0, m5
+    movq            xm3, [r0 + r1]
+    punpcklwd       xm4, xm3
+    movq            xm1, [r0 + r1 * 2]
+    punpcklwd       xm3, xm1
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
+    pmaddwd         m5, m4, [r5 + 2 * mmsize]
+    paddd           m0, m5
+    pmaddwd         m5, m4, [r5 + 1 * mmsize]
+    paddd           m2, m5
+    pmaddwd         m4, [r5]
+    movq            xm3, [r0 + r4]
+    punpcklwd       xm1, xm3
+    lea             r0, [r0 + 4 * r1]
+    movq            xm6, [r0]
+    punpcklwd       xm3, xm6
+    vinserti128     m1, m1, xm3, 1                  ; m1 = [8 7 7 6]
+    pmaddwd         m5, m1, [r5 + 3 * mmsize]
+    paddd           m0, m5
+    pmaddwd         m5, m1, [r5 + 2 * mmsize]
+    paddd           m2, m5
+    pmaddwd         m5, m1, [r5 + 1 * mmsize]
+    paddd           m4, m5
+    pmaddwd         m1, [r5]
+    movq            xm3, [r0 + r1]
+    punpcklwd       xm6, xm3
+    movq            xm5, [r0 + 2 * r1]
+    punpcklwd       xm3, xm5
+    vinserti128     m6, m6, xm3, 1                  ; m6 = [A 9 9 8]
+    pmaddwd         m3, m6, [r5 + 3 * mmsize]
+    paddd           m2, m3
+    pmaddwd         m3, m6, [r5 + 2 * mmsize]
+    paddd           m4, m3
+    pmaddwd         m6, [r5 + 1 * mmsize]
+    paddd           m1, m6
 
-    PROCESS_LUMA_S_AVX2_W8_4R %1
-    lea             r4, [r3 * 3]
 %ifidn %1,sp
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r4], xm2
+    paddd           m0, m7
+    paddd           m2, m7
+    psrad           m0, 12
+    psrad           m2, 12
 %else
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r4], xm3
+    psrad           m0, 6
+    psrad           m2, 6
+%endif
+    packssdw        m0, m2
+
+    movq            xm3, [r0 + r4]
+    punpcklwd       xm5, xm3
+    lea             r0, [r0 + 4 * r1]
+    movq            xm2, [r0]
+    punpcklwd       xm3, xm2
+    vinserti128     m5, m5, xm3, 1                  ; m5 = [C B B A]
+    pmaddwd         m3, m5, [r5 + 3 * mmsize]
+    paddd           m4, m3
+    pmaddwd         m5, [r5 + 2 * mmsize]
+    paddd           m1, m5
+    movq            xm3, [r0 + r1]
+    punpcklwd       xm2, xm3
+    movq            xm5, [r0 + 2 * r1]
+    punpcklwd       xm3, xm5
+    vinserti128     m2, m2, xm3, 1                  ; m2 = [E D D C]
+    pmaddwd         m2, [r5 + 3 * mmsize]
+    paddd           m1, m2
+
+%ifidn %1,sp
+    paddd           m4, m7
+    paddd           m1, m7
+    psrad           m4, 12
+    psrad           m1, 12
+%else
+    psrad           m4, 6
+    psrad           m1, 6
+%endif
+    packssdw        m4, m1
+
+%ifidn %1,sp
+    packuswb        m0, m4
+    vextracti128    xm2, m0, 1
+    movd            [r2], xm0
+    movd            [r2 + r3], xm2
+    pextrd          [r2 + r3 * 2], xm0, 1
+    pextrd          [r2 + r6], xm2, 1
+    lea             r2, [r2 + r3 * 4]
+    pextrd          [r2], xm0, 2
+    pextrd          [r2 + r3], xm2, 2
+    pextrd          [r2 + r3 * 2], xm0, 3
+    pextrd          [r2 + r6], xm2, 3
+%else
+    vextracti128    xm2, m0, 1
+    vextracti128    xm1, m4, 1
+    movq            [r2], xm0
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r6], xm2
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm4
+    movq            [r2 + r3], xm1
+    movhps          [r2 + r3 * 2], xm4
+    movhps          [r2 + r6], xm1
 %endif
     RET
 %endmacro
 
-    FILTER_VER_LUMA_S_AVX2_8x4 sp
-    FILTER_VER_LUMA_S_AVX2_8x4 ss
+    FILTER_VER_LUMA_AVX2_4x8 sp
+    FILTER_VER_LUMA_AVX2_4x8 ss
 
-%macro PROCESS_LUMA_AVX2_W8_16R 1
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
+%macro PROCESS_LUMA_AVX2_W4_16R 1
+    movq            xm0, [r0]
+    movq            xm1, [r0 + r1]
     punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
+    movq            xm2, [r0 + r1 * 2]
     punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
+    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
+    pmaddwd         m0, [r5]
+    movq            xm3, [r0 + r4]
     punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m2, [r5]
-    lea             r7, [r0 + r1 * 4]
-    movu            xm4, [r7]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
+    lea             r0, [r0 + 4 * r1]
+    movq            xm4, [r0]
     punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    paddd           m1, m5
-    pmaddwd         m3, [r5]
-    movu            xm5, [r7 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 2 * mmsize]
-    paddd           m0, m6
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m2, m6
+    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
+    pmaddwd         m5, m2, [r5 + 1 * mmsize]
+    pmaddwd         m2, [r5]
+    paddd           m0, m5
+    movq            xm3, [r0 + r1]
+    punpcklwd       xm4, xm3
+    movq            xm1, [r0 + r1 * 2]
+    punpcklwd       xm3, xm1
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
+    pmaddwd         m5, m4, [r5 + 2 * mmsize]
+    paddd           m0, m5
+    pmaddwd         m5, m4, [r5 + 1 * mmsize]
+    paddd           m2, m5
     pmaddwd         m4, [r5]
-    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm7, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddwd         m7, m5, [r5 + 2 * mmsize]
-    paddd           m1, m7
-    pmaddwd         m7, m5, [r5 + 1 * mmsize]
-    paddd           m3, m7
-    pmaddwd         m5, [r5]
-    movu            xm7, [r7 + r4]                  ; m7 = row 7
-    punpckhwd       xm8, xm6, xm7
-    punpcklwd       xm6, xm7
-    vinserti128     m6, m6, xm8, 1
-    pmaddwd         m8, m6, [r5 + 3 * mmsize]
-    paddd           m0, m8
-    pmaddwd         m8, m6, [r5 + 2 * mmsize]
-    paddd           m2, m8
-    pmaddwd         m8, m6, [r5 + 1 * mmsize]
-    paddd           m4, m8
+    movq            xm3, [r0 + r4]
+    punpcklwd       xm1, xm3
+    lea             r0, [r0 + 4 * r1]
+    movq            xm6, [r0]
+    punpcklwd       xm3, xm6
+    vinserti128     m1, m1, xm3, 1                  ; m1 = [8 7 7 6]
+    pmaddwd         m5, m1, [r5 + 3 * mmsize]
+    paddd           m0, m5
+    pmaddwd         m5, m1, [r5 + 2 * mmsize]
+    paddd           m2, m5
+    pmaddwd         m5, m1, [r5 + 1 * mmsize]
+    paddd           m4, m5
+    pmaddwd         m1, [r5]
+    movq            xm3, [r0 + r1]
+    punpcklwd       xm6, xm3
+    movq            xm5, [r0 + 2 * r1]
+    punpcklwd       xm3, xm5
+    vinserti128     m6, m6, xm3, 1                  ; m6 = [10 9 9 8]
+    pmaddwd         m3, m6, [r5 + 3 * mmsize]
+    paddd           m2, m3
+    pmaddwd         m3, m6, [r5 + 2 * mmsize]
+    paddd           m4, m3
+    pmaddwd         m3, m6, [r5 + 1 * mmsize]
+    paddd           m1, m3
     pmaddwd         m6, [r5]
-    lea             r7, [r7 + r1 * 4]
-    movu            xm8, [r7]                       ; m8 = row 8
-    punpckhwd       xm9, xm7, xm8
-    punpcklwd       xm7, xm8
-    vinserti128     m7, m7, xm9, 1
-    pmaddwd         m9, m7, [r5 + 3 * mmsize]
-    paddd           m1, m9
-    pmaddwd         m9, m7, [r5 + 2 * mmsize]
-    paddd           m3, m9
-    pmaddwd         m9, m7, [r5 + 1 * mmsize]
-    paddd           m5, m9
-    pmaddwd         m7, [r5]
-    movu            xm9, [r7 + r1]                  ; m9 = row 9
-    punpckhwd       xm10, xm8, xm9
-    punpcklwd       xm8, xm9
-    vinserti128     m8, m8, xm10, 1
-    pmaddwd         m10, m8, [r5 + 3 * mmsize]
-    paddd           m2, m10
-    pmaddwd         m10, m8, [r5 + 2 * mmsize]
-    paddd           m4, m10
-    pmaddwd         m10, m8, [r5 + 1 * mmsize]
-    paddd           m6, m10
-    pmaddwd         m8, [r5]
-    movu            xm10, [r7 + r1 * 2]             ; m10 = row 10
-    punpckhwd       xm11, xm9, xm10
-    punpcklwd       xm9, xm10
-    vinserti128     m9, m9, xm11, 1
-    pmaddwd         m11, m9, [r5 + 3 * mmsize]
-    paddd           m3, m11
-    pmaddwd         m11, m9, [r5 + 2 * mmsize]
-    paddd           m5, m11
-    pmaddwd         m11, m9, [r5 + 1 * mmsize]
-    paddd           m7, m11
-    pmaddwd         m9, [r5]
-    movu            xm11, [r7 + r4]                 ; m11 = row 11
-    punpckhwd       xm12, xm10, xm11
-    punpcklwd       xm10, xm11
-    vinserti128     m10, m10, xm12, 1
-    pmaddwd         m12, m10, [r5 + 3 * mmsize]
-    paddd           m4, m12
-    pmaddwd         m12, m10, [r5 + 2 * mmsize]
-    paddd           m6, m12
-    pmaddwd         m12, m10, [r5 + 1 * mmsize]
-    paddd           m8, m12
-    pmaddwd         m10, [r5]
-    lea             r7, [r7 + r1 * 4]
-    movu            xm12, [r7]                      ; m12 = row 12
-    punpckhwd       xm13, xm11, xm12
-    punpcklwd       xm11, xm12
-    vinserti128     m11, m11, xm13, 1
-    pmaddwd         m13, m11, [r5 + 3 * mmsize]
-    paddd           m5, m13
-    pmaddwd         m13, m11, [r5 + 2 * mmsize]
-    paddd           m7, m13
-    pmaddwd         m13, m11, [r5 + 1 * mmsize]
-    paddd           m9, m13
-    pmaddwd         m11, [r5]
 
 %ifidn %1,sp
-    paddd           m0, m14
-    paddd           m1, m14
-    paddd           m2, m14
-    paddd           m3, m14
-    paddd           m4, m14
-    paddd           m5, m14
+    paddd           m0, m7
+    paddd           m2, m7
     psrad           m0, 12
-    psrad           m1, 12
     psrad           m2, 12
-    psrad           m3, 12
-    psrad           m4, 12
-    psrad           m5, 12
 %else
     psrad           m0, 6
-    psrad           m1, 6
     psrad           m2, 6
-    psrad           m3, 6
-    psrad           m4, 6
-    psrad           m5, 6
 %endif
-    packssdw        m0, m1
-    packssdw        m2, m3
-    packssdw        m4, m5
-%ifidn %1,sp
-    packuswb        m0, m2
-    mova            m5, [interp8_hps_shuf]
-    vpermd          m0, m5, m0
+    packssdw        m0, m2
     vextracti128    xm2, m0, 1
+%ifidn %1,sp
+    packuswb        xm0, xm2
+    movd            [r2], xm0
+    pextrd          [r2 + r3], xm0, 2
+    pextrd          [r2 + r3 * 2], xm0, 1
+    pextrd          [r2 + r6], xm0, 3
+%else
     movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm2
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm0
     movhps          [r2 + r6], xm2
-%else
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
 %endif
 
-    movu            xm13, [r7 + r1]                 ; m13 = row 13
-    punpckhwd       xm0, xm12, xm13
-    punpcklwd       xm12, xm13
-    vinserti128     m12, m12, xm0, 1
-    pmaddwd         m0, m12, [r5 + 3 * mmsize]
-    paddd           m6, m0
-    pmaddwd         m0, m12, [r5 + 2 * mmsize]
-    paddd           m8, m0
-    pmaddwd         m0, m12, [r5 + 1 * mmsize]
-    paddd           m10, m0
-    pmaddwd         m12, [r5]
-    movu            xm0, [r7 + r1 * 2]              ; m0 = row 14
-    punpckhwd       xm1, xm13, xm0
-    punpcklwd       xm13, xm0
-    vinserti128     m13, m13, xm1, 1
-    pmaddwd         m1, m13, [r5 + 3 * mmsize]
-    paddd           m7, m1
-    pmaddwd         m1, m13, [r5 + 2 * mmsize]
-    paddd           m9, m1
-    pmaddwd         m1, m13, [r5 + 1 * mmsize]
-    paddd           m11, m1
-    pmaddwd         m13, [r5]
+    movq            xm2, [r0 + r4]
+    punpcklwd       xm5, xm2
+    lea             r0, [r0 + 4 * r1]
+    movq            xm0, [r0]
+    punpcklwd       xm2, xm0
+    vinserti128     m5, m5, xm2, 1                  ; m5 = [12 11 11 10]
+    pmaddwd         m2, m5, [r5 + 3 * mmsize]
+    paddd           m4, m2
+    pmaddwd         m2, m5, [r5 + 2 * mmsize]
+    paddd           m1, m2
+    pmaddwd         m2, m5, [r5 + 1 * mmsize]
+    paddd           m6, m2
+    pmaddwd         m5, [r5]
+    movq            xm2, [r0 + r1]
+    punpcklwd       xm0, xm2
+    movq            xm3, [r0 + 2 * r1]
+    punpcklwd       xm2, xm3
+    vinserti128     m0, m0, xm2, 1                  ; m0 = [14 13 13 12]
+    pmaddwd         m2, m0, [r5 + 3 * mmsize]
+    paddd           m1, m2
+    pmaddwd         m2, m0, [r5 + 2 * mmsize]
+    paddd           m6, m2
+    pmaddwd         m2, m0, [r5 + 1 * mmsize]
+    paddd           m5, m2
+    pmaddwd         m0, [r5]
 
 %ifidn %1,sp
-    paddd           m6, m14
-    paddd           m7, m14
-    psrad           m6, 12
-    psrad           m7, 12
+    paddd           m4, m7
+    paddd           m1, m7
+    psrad           m4, 12
+    psrad           m1, 12
 %else
-    psrad           m6, 6
-    psrad           m7, 6
+    psrad           m4, 6
+    psrad           m1, 6
 %endif
-    packssdw        m6, m7
-    lea             r8, [r2 + r3 * 4]
-
+    packssdw        m4, m1
+    vextracti128    xm1, m4, 1
+    lea             r2, [r2 + r3 * 4]
 %ifidn %1,sp
-    packuswb        m4, m6
-    vpermd          m4, m5, m4
-    vextracti128    xm6, m4, 1
-    movq            [r8], xm4
-    movhps          [r8 + r3], xm4
-    movq            [r8 + r3 * 2], xm6
-    movhps          [r8 + r6], xm6
+    packuswb        xm4, xm1
+    movd            [r2], xm4
+    pextrd          [r2 + r3], xm4, 2
+    pextrd          [r2 + r3 * 2], xm4, 1
+    pextrd          [r2 + r6], xm4, 3
 %else
-    vpermq          m4, m4, 11011000b
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm1, m4, 1
-    vextracti128    xm7, m6, 1
-    movu            [r8], xm4
-    movu            [r8 + r3], xm1
-    movu            [r8 + r3 * 2], xm6
-    movu            [r8 + r6], xm7
+    movq            [r2], xm4
+    movq            [r2 + r3], xm1
+    movhps          [r2 + r3 * 2], xm4
+    movhps          [r2 + r6], xm1
 %endif
 
-    movu            xm1, [r7 + r4]                  ; m1 = row 15
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m2, m0, [r5 + 3 * mmsize]
-    paddd           m8, m2
-    pmaddwd         m2, m0, [r5 + 2 * mmsize]
-    paddd           m10, m2
-    pmaddwd         m2, m0, [r5 + 1 * mmsize]
-    paddd           m12, m2
-    pmaddwd         m0, [r5]
-    lea             r7, [r7 + r1 * 4]
-    movu            xm2, [r7]                       ; m2 = row 16
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m3, m1, [r5 + 3 * mmsize]
-    paddd           m9, m3
-    pmaddwd         m3, m1, [r5 + 2 * mmsize]
-    paddd           m11, m3
-    pmaddwd         m3, m1, [r5 + 1 * mmsize]
-    paddd           m13, m3
-    pmaddwd         m1, [r5]
-    movu            xm3, [r7 + r1]                  ; m3 = row 17
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 3 * mmsize]
-    paddd           m10, m4
-    pmaddwd         m4, m2, [r5 + 2 * mmsize]
-    paddd           m12, m4
-    pmaddwd         m2, [r5 + 1 * mmsize]
-    paddd           m0, m2
-    movu            xm4, [r7 + r1 * 2]              ; m4 = row 18
-    punpckhwd       xm2, xm3, xm4
+    movq            xm4, [r0 + r4]
     punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm2, 1
-    pmaddwd         m2, m3, [r5 + 3 * mmsize]
-    paddd           m11, m2
-    pmaddwd         m2, m3, [r5 + 2 * mmsize]
-    paddd           m13, m2
-    pmaddwd         m3, [r5 + 1 * mmsize]
-    paddd           m1, m3
-    movu            xm2, [r7 + r4]                  ; m2 = row 19
-    punpckhwd       xm6, xm4, xm2
+    lea             r0, [r0 + 4 * r1]
+    movq            xm1, [r0]
+    punpcklwd       xm4, xm1
+    vinserti128     m3, m3, xm4, 1                  ; m3 = [16 15 15 14]
+    pmaddwd         m4, m3, [r5 + 3 * mmsize]
+    paddd           m6, m4
+    pmaddwd         m4, m3, [r5 + 2 * mmsize]
+    paddd           m5, m4
+    pmaddwd         m4, m3, [r5 + 1 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m3, [r5]
+    movq            xm4, [r0 + r1]
+    punpcklwd       xm1, xm4
+    movq            xm2, [r0 + 2 * r1]
     punpcklwd       xm4, xm2
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 3 * mmsize]
-    paddd           m12, m6
-    pmaddwd         m4, [r5 + 2 * mmsize]
+    vinserti128     m1, m1, xm4, 1                  ; m1 = [18 17 17 16]
+    pmaddwd         m4, m1, [r5 + 3 * mmsize]
+    paddd           m5, m4
+    pmaddwd         m4, m1, [r5 + 2 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m1, [r5 + 1 * mmsize]
+    paddd           m3, m1
+    movq            xm4, [r0 + r4]
+    punpcklwd       xm2, xm4
+    lea             r0, [r0 + 4 * r1]
+    movq            xm1, [r0]
+    punpcklwd       xm4, xm1
+    vinserti128     m2, m2, xm4, 1                  ; m2 = [20 19 19 18]
+    pmaddwd         m4, m2, [r5 + 3 * mmsize]
     paddd           m0, m4
-    lea             r7, [r7 + r1 * 4]
-    movu            xm6, [r7]                       ; m6 = row 20
-    punpckhwd       xm7, xm2, xm6
-    punpcklwd       xm2, xm6
-    vinserti128     m2, m2, xm7, 1
-    pmaddwd         m7, m2, [r5 + 3 * mmsize]
-    paddd           m13, m7
     pmaddwd         m2, [r5 + 2 * mmsize]
-    paddd           m1, m2
-    movu            xm7, [r7 + r1]                  ; m7 = row 21
-    punpckhwd       xm2, xm6, xm7
-    punpcklwd       xm6, xm7
-    vinserti128     m6, m6, xm2, 1
-    pmaddwd         m6, [r5 + 3 * mmsize]
-    paddd           m0, m6
-    movu            xm2, [r7 + r1 * 2]              ; m2 = row 22
-    punpckhwd       xm3, xm7, xm2
-    punpcklwd       xm7, xm2
-    vinserti128     m7, m7, xm3, 1
-    pmaddwd         m7, [r5 + 3 * mmsize]
-    paddd           m1, m7
+    paddd           m3, m2
+    movq            xm4, [r0 + r1]
+    punpcklwd       xm1, xm4
+    movq            xm2, [r0 + 2 * r1]
+    punpcklwd       xm4, xm2
+    vinserti128     m1, m1, xm4, 1                  ; m1 = [22 21 21 20]
+    pmaddwd         m1, [r5 + 3 * mmsize]
+    paddd           m3, m1
 
 %ifidn %1,sp
-    paddd           m8, m14
-    paddd           m9, m14
-    paddd           m10, m14
-    paddd           m11, m14
-    paddd           m12, m14
-    paddd           m13, m14
-    paddd           m0, m14
-    paddd           m1, m14
-    psrad           m8, 12
-    psrad           m9, 12
-    psrad           m10, 12
-    psrad           m11, 12
-    psrad           m12, 12
-    psrad           m13, 12
+    paddd           m6, m7
+    paddd           m5, m7
+    paddd           m0, m7
+    paddd           m3, m7
+    psrad           m6, 12
+    psrad           m5, 12
     psrad           m0, 12
-    psrad           m1, 12
+    psrad           m3, 12
 %else
-    psrad           m8, 6
-    psrad           m9, 6
-    psrad           m10, 6
-    psrad           m11, 6
-    psrad           m12, 6
-    psrad           m13, 6
+    psrad           m6, 6
+    psrad           m5, 6
     psrad           m0, 6
-    psrad           m1, 6
+    psrad           m3, 6
 %endif
-    packssdw        m8, m9
-    packssdw        m10, m11
-    packssdw        m12, m13
-    packssdw        m0, m1
-    lea             r8, [r8 + r3 * 4]
+    packssdw        m6, m5
+    packssdw        m0, m3
+    lea             r2, [r2 + r3 * 4]
 
 %ifidn %1,sp
-    packuswb        m8, m10
-    packuswb        m12, m0
-    vpermd          m8, m5, m8
-    vpermd          m12, m5, m12
-    vextracti128    xm10, m8, 1
-    vextracti128    xm0, m12, 1
-    movq            [r8], xm8
-    movhps          [r8 + r3], xm8
-    movq            [r8 + r3 * 2], xm10
-    movhps          [r8 + r6], xm10
-    lea             r8, [r8 + r3 * 4]
-    movq            [r8], xm12
-    movhps          [r8 + r3], xm12
-    movq            [r8 + r3 * 2], xm0
-    movhps          [r8 + r6], xm0
+    packuswb        m6, m0
+    vextracti128    xm0, m6, 1
+    movd            [r2], xm6
+    movd            [r2 + r3], xm0
+    pextrd          [r2 + r3 * 2], xm6, 1
+    pextrd          [r2 + r6], xm0, 1
+    lea             r2, [r2 + r3 * 4]
+    pextrd          [r2], xm6, 2
+    pextrd          [r2 + r3], xm0, 2
+    pextrd          [r2 + r3 * 2], xm6, 3
+    pextrd          [r2 + r6], xm0, 3
 %else
-    vpermq          m8, m8, 11011000b
-    vpermq          m10, m10, 11011000b
-    vpermq          m12, m12, 11011000b
-    vpermq          m0, m0, 11011000b
-    vextracti128    xm9, m8, 1
-    vextracti128    xm11, m10, 1
-    vextracti128    xm13, m12, 1
-    vextracti128    xm1, m0, 1
-    movu            [r8], xm8
-    movu            [r8 + r3], xm9
-    movu            [r8 + r3 * 2], xm10
-    movu            [r8 + r6], xm11
-    lea             r8, [r8 + r3 * 4]
-    movu            [r8], xm12
-    movu            [r8 + r3], xm13
-    movu            [r8 + r3 * 2], xm0
-    movu            [r8 + r6], xm1
+    vextracti128    xm5, m6, 1
+    vextracti128    xm3, m0, 1
+    movq            [r2], xm6
+    movq            [r2 + r3], xm5
+    movhps          [r2 + r3 * 2], xm6
+    movhps          [r2 + r6], xm5
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm0
+    movq            [r2 + r3], xm3
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r6], xm3
 %endif
 %endmacro
 
-%macro FILTER_VER_LUMA_AVX2_Nx16 2
+%macro FILTER_VER_LUMA_AVX2_4x16 1
 INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_%2x16, 4, 10, 15
+cglobal interp_8tap_vert_%1_4x16, 4, 7, 8
     mov             r4d, r4m
     shl             r4d, 7
     add             r1d, r1d
@@ -24535,37 +20165,22 @@ cglobal interp_8tap_vert_%1_%2x16, 4, 10, 15
     lea             r4, [r1 * 3]
     sub             r0, r4
 %ifidn %1,sp
-    mova            m14, [pd_526336]
+    mova            m7, [pd_526336]
 %else
     add             r3d, r3d
 %endif
     lea             r6, [r3 * 3]
-    mov             r9d, %2 / 8
-.loopW:
-    PROCESS_LUMA_AVX2_W8_16R %1
-%ifidn %1,sp
-    add             r2, 8
-%else
-    add             r2, 16
-%endif
-    add             r0, 16
-    dec             r9d
-    jnz             .loopW
+    PROCESS_LUMA_AVX2_W4_16R %1
     RET
-%endif
 %endmacro
 
-    FILTER_VER_LUMA_AVX2_Nx16 sp, 16
-    FILTER_VER_LUMA_AVX2_Nx16 sp, 32
-    FILTER_VER_LUMA_AVX2_Nx16 sp, 64
-    FILTER_VER_LUMA_AVX2_Nx16 ss, 16
-    FILTER_VER_LUMA_AVX2_Nx16 ss, 32
-    FILTER_VER_LUMA_AVX2_Nx16 ss, 64
+    FILTER_VER_LUMA_AVX2_4x16 sp
+    FILTER_VER_LUMA_AVX2_4x16 ss
 
-%macro FILTER_VER_LUMA_AVX2_NxN 3
+%macro FILTER_VER_LUMA_S_AVX2_8x8 1
 INIT_YMM avx2
 %if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%3_%1x%2, 4, 12, 15
+cglobal interp_8tap_vert_%1_8x8, 4, 6, 12
     mov             r4d, r4m
     shl             r4d, 7
     add             r1d, r1d
@@ -24580,102 +20195,215 @@ cglobal interp_8tap_vert_%3_%1x%2, 4, 12, 15
     lea             r4, [r1 * 3]
     sub             r0, r4
 
-%ifidn %3,sp
-    mova            m14, [pd_526336]
+%ifidn %1,sp
+    mova            m11, [pd_526336]
 %else
     add             r3d, r3d
 %endif
 
-    lea             r6, [r3 * 3]
-    lea             r11, [r1 * 4]
-    mov             r9d, %2 / 16
-.loopH:
-    mov             r10d, %1 / 8
-.loopW:
-    PROCESS_LUMA_AVX2_W8_16R %3
-%ifidn %3,sp
-    add             r2, 8
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    pmaddwd         m2, [r5]
+    paddd           m0, m4
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    pmaddwd         m3, [r5]
+    paddd           m1, m5
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 2 * mmsize]
+    paddd           m0, m6
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m2, m6
+    pmaddwd         m4, [r5]
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm7, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddwd         m7, m5, [r5 + 2 * mmsize]
+    paddd           m1, m7
+    pmaddwd         m7, m5, [r5 + 1 * mmsize]
+    pmaddwd         m5, [r5]
+    paddd           m3, m7
+    movu            xm7, [r0 + r4]                  ; m7 = row 7
+    punpckhwd       xm8, xm6, xm7
+    punpcklwd       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddwd         m8, m6, [r5 + 3 * mmsize]
+    paddd           m0, m8
+    pmaddwd         m8, m6, [r5 + 2 * mmsize]
+    paddd           m2, m8
+    pmaddwd         m8, m6, [r5 + 1 * mmsize]
+    pmaddwd         m6, [r5]
+    paddd           m4, m8
+    lea             r0, [r0 + r1 * 4]
+    movu            xm8, [r0]                       ; m8 = row 8
+    punpckhwd       xm9, xm7, xm8
+    punpcklwd       xm7, xm8
+    vinserti128     m7, m7, xm9, 1
+    pmaddwd         m9, m7, [r5 + 3 * mmsize]
+    paddd           m1, m9
+    pmaddwd         m9, m7, [r5 + 2 * mmsize]
+    paddd           m3, m9
+    pmaddwd         m9, m7, [r5 + 1 * mmsize]
+    pmaddwd         m7, [r5]
+    paddd           m5, m9
+    movu            xm9, [r0 + r1]                  ; m9 = row 9
+    punpckhwd       xm10, xm8, xm9
+    punpcklwd       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddwd         m10, m8, [r5 + 3 * mmsize]
+    paddd           m2, m10
+    pmaddwd         m10, m8, [r5 + 2 * mmsize]
+    pmaddwd         m8, [r5 + 1 * mmsize]
+    paddd           m4, m10
+    paddd           m6, m8
+    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
+    punpckhwd       xm8, xm9, xm10
+    punpcklwd       xm9, xm10
+    vinserti128     m9, m9, xm8, 1
+    pmaddwd         m8, m9, [r5 + 3 * mmsize]
+    paddd           m3, m8
+    pmaddwd         m8, m9, [r5 + 2 * mmsize]
+    pmaddwd         m9, [r5 + 1 * mmsize]
+    paddd           m5, m8
+    paddd           m7, m9
+    movu            xm8, [r0 + r4]                  ; m8 = row 11
+    punpckhwd       xm9, xm10, xm8
+    punpcklwd       xm10, xm8
+    vinserti128     m10, m10, xm9, 1
+    pmaddwd         m9, m10, [r5 + 3 * mmsize]
+    pmaddwd         m10, [r5 + 2 * mmsize]
+    paddd           m4, m9
+    paddd           m6, m10
+
+    lea             r4, [r3 * 3]
+%ifidn %1,sp
+    paddd           m0, m11
+    paddd           m1, m11
+    paddd           m2, m11
+    paddd           m3, m11
+    psrad           m0, 12
+    psrad           m1, 12
+    psrad           m2, 12
+    psrad           m3, 12
 %else
-    add             r2, 16
+    psrad           m0, 6
+    psrad           m1, 6
+    psrad           m2, 6
+    psrad           m3, 6
 %endif
-    add             r0, 16
-    dec             r10d
-    jnz             .loopW
-    sub             r7, r11
-    lea             r0, [r7 - 2 * %1 + 16]
-%ifidn %3,sp
-    lea             r2, [r8 + r3 * 4 - %1 + 8]
+    packssdw        m0, m1
+    packssdw        m2, m3
+%ifidn %1,sp
+    packuswb        m0, m2
+    mova            m1, [interp8_hps_shuf]
+    vpermd          m0, m1, m0
+    vextracti128    xm2, m0, 1
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r4], xm2
 %else
-    lea             r2, [r8 + r3 * 4 - 2 * %1 + 16]
-%endif
-    dec             r9d
-    jnz             .loopH
-    RET
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m2, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r4], xm3
 %endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_NxN 16, 32, sp
-    FILTER_VER_LUMA_AVX2_NxN 16, 64, sp
-    FILTER_VER_LUMA_AVX2_NxN 24, 32, sp
-    FILTER_VER_LUMA_AVX2_NxN 32, 32, sp
-    FILTER_VER_LUMA_AVX2_NxN 32, 64, sp
-    FILTER_VER_LUMA_AVX2_NxN 48, 64, sp
-    FILTER_VER_LUMA_AVX2_NxN 64, 32, sp
-    FILTER_VER_LUMA_AVX2_NxN 64, 48, sp
-    FILTER_VER_LUMA_AVX2_NxN 64, 64, sp
-    FILTER_VER_LUMA_AVX2_NxN 16, 32, ss
-    FILTER_VER_LUMA_AVX2_NxN 16, 64, ss
-    FILTER_VER_LUMA_AVX2_NxN 24, 32, ss
-    FILTER_VER_LUMA_AVX2_NxN 32, 32, ss
-    FILTER_VER_LUMA_AVX2_NxN 32, 64, ss
-    FILTER_VER_LUMA_AVX2_NxN 48, 64, ss
-    FILTER_VER_LUMA_AVX2_NxN 64, 32, ss
-    FILTER_VER_LUMA_AVX2_NxN 64, 48, ss
-    FILTER_VER_LUMA_AVX2_NxN 64, 64, ss
-
-%macro FILTER_VER_LUMA_S_AVX2_12x16 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_12x16, 4, 9, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-    add             r1d, r1d
 
-%ifdef PIC
-    lea             r5, [pw_LumaCoeffVer]
-    add             r5, r4
-%else
-    lea             r5, [pw_LumaCoeffVer + r4]
-%endif
+    lea             r0, [r0 + r1 * 4]
+    movu            xm9, [r0]                       ; m9 = row 12
+    punpckhwd       xm3, xm8, xm9
+    punpcklwd       xm8, xm9
+    vinserti128     m8, m8, xm3, 1
+    pmaddwd         m3, m8, [r5 + 3 * mmsize]
+    pmaddwd         m8, [r5 + 2 * mmsize]
+    paddd           m5, m3
+    paddd           m7, m8
+    movu            xm3, [r0 + r1]                  ; m3 = row 13
+    punpckhwd       xm0, xm9, xm3
+    punpcklwd       xm9, xm3
+    vinserti128     m9, m9, xm0, 1
+    pmaddwd         m9, [r5 + 3 * mmsize]
+    paddd           m6, m9
+    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
+    punpckhwd       xm9, xm3, xm0
+    punpcklwd       xm3, xm0
+    vinserti128     m3, m3, xm9, 1
+    pmaddwd         m3, [r5 + 3 * mmsize]
+    paddd           m7, m3
 
-    lea             r4, [r1 * 3]
-    sub             r0, r4
 %ifidn %1,sp
-    mova            m14, [pd_526336]
+    paddd           m4, m11
+    paddd           m5, m11
+    paddd           m6, m11
+    paddd           m7, m11
+    psrad           m4, 12
+    psrad           m5, 12
+    psrad           m6, 12
+    psrad           m7, 12
 %else
-    add             r3d, r3d
+    psrad           m4, 6
+    psrad           m5, 6
+    psrad           m6, 6
+    psrad           m7, 6
 %endif
-    lea             r6, [r3 * 3]
-    PROCESS_LUMA_AVX2_W8_16R %1
+    packssdw        m4, m5
+    packssdw        m6, m7
+    lea             r2, [r2 + r3 * 4]
 %ifidn %1,sp
-    add             r2, 8
+    packuswb        m4, m6
+    vpermd          m4, m1, m4
+    vextracti128    xm6, m4, 1
+    movq            [r2], xm4
+    movhps          [r2 + r3], xm4
+    movq            [r2 + r3 * 2], xm6
+    movhps          [r2 + r4], xm6
 %else
-    add             r2, 16
+    vpermq          m4, m4, 11011000b
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm5, m4, 1
+    vextracti128    xm7, m6, 1
+    movu            [r2], xm4
+    movu            [r2 + r3], xm5
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r4], xm7
 %endif
-    add             r0, 16
-    mova            m7, m14
-    PROCESS_LUMA_AVX2_W4_16R %1
     RET
 %endif
 %endmacro
 
-    FILTER_VER_LUMA_S_AVX2_12x16 sp
-    FILTER_VER_LUMA_S_AVX2_12x16 ss
+    FILTER_VER_LUMA_S_AVX2_8x8 sp
+    FILTER_VER_LUMA_S_AVX2_8x8 ss
 
-%macro FILTER_VER_LUMA_S_AVX2_16x12 1
+%macro FILTER_VER_LUMA_S_AVX2_8xN 2
 INIT_YMM avx2
 %if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_16x12, 4, 10, 15
+cglobal interp_8tap_vert_%1_8x%2, 4, 9, 15
     mov             r4d, r4m
     shl             r4d, 7
     add             r1d, r1d
@@ -24695,8 +20423,9 @@ cglobal interp_8tap_vert_%1_16x12, 4, 10, 15
     add             r3d, r3d
 %endif
     lea             r6, [r3 * 3]
-    mov             r9d, 2
-.loopW:
+    lea             r7, [r1 * 4]
+    mov             r8d, %2 / 16
+.loopH:
     movu            xm0, [r0]                       ; m0 = row 0
     movu            xm1, [r0 + r1]                  ; m1 = row 1
     punpckhwd       xm2, xm0, xm1
@@ -24715,15 +20444,15 @@ cglobal interp_8tap_vert_%1_16x12, 4, 10, 15
     pmaddwd         m4, m2, [r5 + 1 * mmsize]
     paddd           m0, m4
     pmaddwd         m2, [r5]
-    lea             r7, [r0 + r1 * 4]
-    movu            xm4, [r7]                       ; m4 = row 4
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
     punpckhwd       xm5, xm3, xm4
     punpcklwd       xm3, xm4
     vinserti128     m3, m3, xm5, 1
     pmaddwd         m5, m3, [r5 + 1 * mmsize]
     paddd           m1, m5
     pmaddwd         m3, [r5]
-    movu            xm5, [r7 + r1]                  ; m5 = row 5
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
     punpckhwd       xm6, xm4, xm5
     punpcklwd       xm4, xm5
     vinserti128     m4, m4, xm6, 1
@@ -24732,7 +20461,7 @@ cglobal interp_8tap_vert_%1_16x12, 4, 10, 15
     pmaddwd         m6, m4, [r5 + 1 * mmsize]
     paddd           m2, m6
     pmaddwd         m4, [r5]
-    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
     punpckhwd       xm7, xm5, xm6
     punpcklwd       xm5, xm6
     vinserti128     m5, m5, xm7, 1
@@ -24741,7 +20470,7 @@ cglobal interp_8tap_vert_%1_16x12, 4, 10, 15
     pmaddwd         m7, m5, [r5 + 1 * mmsize]
     paddd           m3, m7
     pmaddwd         m5, [r5]
-    movu            xm7, [r7 + r4]                  ; m7 = row 7
+    movu            xm7, [r0 + r4]                  ; m7 = row 7
     punpckhwd       xm8, xm6, xm7
     punpcklwd       xm6, xm7
     vinserti128     m6, m6, xm8, 1
@@ -24752,8 +20481,8 @@ cglobal interp_8tap_vert_%1_16x12, 4, 10, 15
     pmaddwd         m8, m6, [r5 + 1 * mmsize]
     paddd           m4, m8
     pmaddwd         m6, [r5]
-    lea             r7, [r7 + r1 * 4]
-    movu            xm8, [r7]                       ; m8 = row 8
+    lea             r0, [r0 + r1 * 4]
+    movu            xm8, [r0]                       ; m8 = row 8
     punpckhwd       xm9, xm7, xm8
     punpcklwd       xm7, xm8
     vinserti128     m7, m7, xm9, 1
@@ -24764,7 +20493,7 @@ cglobal interp_8tap_vert_%1_16x12, 4, 10, 15
     pmaddwd         m9, m7, [r5 + 1 * mmsize]
     paddd           m5, m9
     pmaddwd         m7, [r5]
-    movu            xm9, [r7 + r1]                  ; m9 = row 9
+    movu            xm9, [r0 + r1]                  ; m9 = row 9
     punpckhwd       xm10, xm8, xm9
     punpcklwd       xm8, xm9
     vinserti128     m8, m8, xm10, 1
@@ -24775,7 +20504,7 @@ cglobal interp_8tap_vert_%1_16x12, 4, 10, 15
     pmaddwd         m10, m8, [r5 + 1 * mmsize]
     paddd           m6, m10
     pmaddwd         m8, [r5]
-    movu            xm10, [r7 + r1 * 2]             ; m10 = row 10
+    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
     punpckhwd       xm11, xm9, xm10
     punpcklwd       xm9, xm10
     vinserti128     m9, m9, xm11, 1
@@ -24786,7 +20515,7 @@ cglobal interp_8tap_vert_%1_16x12, 4, 10, 15
     pmaddwd         m11, m9, [r5 + 1 * mmsize]
     paddd           m7, m11
     pmaddwd         m9, [r5]
-    movu            xm11, [r7 + r4]                 ; m11 = row 11
+    movu            xm11, [r0 + r4]                 ; m11 = row 11
     punpckhwd       xm12, xm10, xm11
     punpcklwd       xm10, xm11
     vinserti128     m10, m10, xm12, 1
@@ -24797,8 +20526,8 @@ cglobal interp_8tap_vert_%1_16x12, 4, 10, 15
     pmaddwd         m12, m10, [r5 + 1 * mmsize]
     paddd           m8, m12
     pmaddwd         m10, [r5]
-    lea             r7, [r7 + r1 * 4]
-    movu            xm12, [r7]                      ; m12 = row 12
+    lea             r0, [r0 + r1 * 4]
+    movu            xm12, [r0]                      ; m12 = row 12
     punpckhwd       xm13, xm11, xm12
     punpcklwd       xm11, xm12
     vinserti128     m11, m11, xm13, 1
@@ -24834,11 +20563,10 @@ cglobal interp_8tap_vert_%1_16x12, 4, 10, 15
     packssdw        m0, m1
     packssdw        m2, m3
     packssdw        m4, m5
-
 %ifidn %1,sp
     packuswb        m0, m2
-    mova            m5, [interp8_hps_shuf]
-    vpermd          m0, m5, m0
+    mova            m1, [interp8_hps_shuf]
+    vpermd          m0, m1, m0
     vextracti128    xm2, m0, 1
     movq            [r2], xm0
     movhps          [r2 + r3], xm0
@@ -24855,7 +20583,7 @@ cglobal interp_8tap_vert_%1_16x12, 4, 10, 15
     movu            [r2 + r6], xm3
 %endif
 
-    movu            xm13, [r7 + r1]                 ; m13 = row 13
+    movu            xm13, [r0 + r1]                 ; m13 = row 13
     punpckhwd       xm0, xm12, xm13
     punpcklwd       xm12, xm13
     vinserti128     m12, m12, xm0, 1
@@ -24863,18 +20591,20 @@ cglobal interp_8tap_vert_%1_16x12, 4, 10, 15
     paddd           m6, m0
     pmaddwd         m0, m12, [r5 + 2 * mmsize]
     paddd           m8, m0
-    pmaddwd         m12, [r5 + 1 * mmsize]
-    paddd           m10, m12
-    movu            xm0, [r7 + r1 * 2]              ; m0 = row 14
-    punpckhwd       xm1, xm13, xm0
+    pmaddwd         m0, m12, [r5 + 1 * mmsize]
+    paddd           m10, m0
+    pmaddwd         m12, [r5]
+    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
+    punpckhwd       xm2, xm13, xm0
     punpcklwd       xm13, xm0
-    vinserti128     m13, m13, xm1, 1
-    pmaddwd         m1, m13, [r5 + 3 * mmsize]
-    paddd           m7, m1
-    pmaddwd         m1, m13, [r5 + 2 * mmsize]
-    paddd           m9, m1
-    pmaddwd         m13, [r5 + 1 * mmsize]
-    paddd           m11, m13
+    vinserti128     m13, m13, xm2, 1
+    pmaddwd         m2, m13, [r5 + 3 * mmsize]
+    paddd           m7, m2
+    pmaddwd         m2, m13, [r5 + 2 * mmsize]
+    paddd           m9, m2
+    pmaddwd         m2, m13, [r5 + 1 * mmsize]
+    paddd           m11, m2
+    pmaddwd         m13, [r5]
 
 %ifidn %1,sp
     paddd           m6, m14
@@ -24886,109 +20616,288 @@ cglobal interp_8tap_vert_%1_16x12, 4, 10, 15
     psrad           m7, 6
 %endif
     packssdw        m6, m7
-    lea             r8, [r2 + r3 * 4]
+    lea             r2, [r2 + r3 * 4]
 
 %ifidn %1,sp
     packuswb        m4, m6
-    vpermd          m4, m5, m4
+    vpermd          m4, m1, m4
     vextracti128    xm6, m4, 1
-    movq            [r8], xm4
-    movhps          [r8 + r3], xm4
-    movq            [r8 + r3 * 2], xm6
-    movhps          [r8 + r6], xm6
+    movq            [r2], xm4
+    movhps          [r2 + r3], xm4
+    movq            [r2 + r3 * 2], xm6
+    movhps          [r2 + r6], xm6
 %else
-    vpermq          m4, m4, 11011000b
     vpermq          m6, m6, 11011000b
+    vpermq          m4, m4, 11011000b
     vextracti128    xm1, m4, 1
     vextracti128    xm7, m6, 1
-    movu            [r8], xm4
-    movu            [r8 + r3], xm1
-    movu            [r8 + r3 * 2], xm6
-    movu            [r8 + r6], xm7
+    movu            [r2], xm4
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r6], xm7
 %endif
 
-    movu            xm1, [r7 + r4]                  ; m1 = row 15
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m2, m0, [r5 + 3 * mmsize]
-    paddd           m8, m2
-    pmaddwd         m0, [r5 + 2 * mmsize]
-    paddd           m10, m0
-    lea             r7, [r7 + r1 * 4]
-    movu            xm2, [r7]                       ; m2 = row 16
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m3, m1, [r5 + 3 * mmsize]
+    movu            xm6, [r0 + r4]                  ; m6 = row 15
+    punpckhwd       xm5, xm0, xm6
+    punpcklwd       xm0, xm6
+    vinserti128     m0, m0, xm5, 1
+    pmaddwd         m5, m0, [r5 + 3 * mmsize]
+    paddd           m8, m5
+    pmaddwd         m5, m0, [r5 + 2 * mmsize]
+    paddd           m10, m5
+    pmaddwd         m5, m0, [r5 + 1 * mmsize]
+    paddd           m12, m5
+    pmaddwd         m0, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm2, [r0]                       ; m2 = row 16
+    punpckhwd       xm3, xm6, xm2
+    punpcklwd       xm6, xm2
+    vinserti128     m6, m6, xm3, 1
+    pmaddwd         m3, m6, [r5 + 3 * mmsize]
     paddd           m9, m3
-    pmaddwd         m1, [r5 + 2 * mmsize]
-    paddd           m11, m1
-    movu            xm3, [r7 + r1]                  ; m3 = row 17
+    pmaddwd         m3, m6, [r5 + 2 * mmsize]
+    paddd           m11, m3
+    pmaddwd         m3, m6, [r5 + 1 * mmsize]
+    paddd           m13, m3
+    pmaddwd         m6, [r5]
+    movu            xm3, [r0 + r1]                  ; m3 = row 17
     punpckhwd       xm4, xm2, xm3
     punpcklwd       xm2, xm3
     vinserti128     m2, m2, xm4, 1
-    pmaddwd         m2, [r5 + 3 * mmsize]
-    paddd           m10, m2
-    movu            xm4, [r7 + r1 * 2]              ; m4 = row 18
+    pmaddwd         m4, m2, [r5 + 3 * mmsize]
+    paddd           m10, m4
+    pmaddwd         m4, m2, [r5 + 2 * mmsize]
+    paddd           m12, m4
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m0, m2
+    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
     punpckhwd       xm2, xm3, xm4
     punpcklwd       xm3, xm4
     vinserti128     m3, m3, xm2, 1
+    pmaddwd         m2, m3, [r5 + 3 * mmsize]
+    paddd           m11, m2
+    pmaddwd         m2, m3, [r5 + 2 * mmsize]
+    paddd           m13, m2
+    pmaddwd         m3, [r5 + 1 * mmsize]
+    paddd           m6, m3
+    movu            xm2, [r0 + r4]                  ; m2 = row 19
+    punpckhwd       xm7, xm4, xm2
+    punpcklwd       xm4, xm2
+    vinserti128     m4, m4, xm7, 1
+    pmaddwd         m7, m4, [r5 + 3 * mmsize]
+    paddd           m12, m7
+    pmaddwd         m4, [r5 + 2 * mmsize]
+    paddd           m0, m4
+    lea             r0, [r0 + r1 * 4]
+    movu            xm7, [r0]                       ; m7 = row 20
+    punpckhwd       xm3, xm2, xm7
+    punpcklwd       xm2, xm7
+    vinserti128     m2, m2, xm3, 1
+    pmaddwd         m3, m2, [r5 + 3 * mmsize]
+    paddd           m13, m3
+    pmaddwd         m2, [r5 + 2 * mmsize]
+    paddd           m6, m2
+    movu            xm3, [r0 + r1]                  ; m3 = row 21
+    punpckhwd       xm2, xm7, xm3
+    punpcklwd       xm7, xm3
+    vinserti128     m7, m7, xm2, 1
+    pmaddwd         m7, [r5 + 3 * mmsize]
+    paddd           m0, m7
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 22
+    punpckhwd       xm7, xm3, xm2
+    punpcklwd       xm3, xm2
+    vinserti128     m3, m3, xm7, 1
     pmaddwd         m3, [r5 + 3 * mmsize]
-    paddd           m11, m3
+    paddd           m6, m3
 
 %ifidn %1,sp
     paddd           m8, m14
     paddd           m9, m14
     paddd           m10, m14
     paddd           m11, m14
+    paddd           m12, m14
+    paddd           m13, m14
+    paddd           m0, m14
+    paddd           m6, m14
     psrad           m8, 12
     psrad           m9, 12
     psrad           m10, 12
     psrad           m11, 12
+    psrad           m12, 12
+    psrad           m13, 12
+    psrad           m0, 12
+    psrad           m6, 12
 %else
     psrad           m8, 6
     psrad           m9, 6
     psrad           m10, 6
     psrad           m11, 6
+    psrad           m12, 6
+    psrad           m13, 6
+    psrad           m0, 6
+    psrad           m6, 6
 %endif
     packssdw        m8, m9
     packssdw        m10, m11
-    lea             r8, [r8 + r3 * 4]
+    packssdw        m12, m13
+    packssdw        m0, m6
+    lea             r2, [r2 + r3 * 4]
 
 %ifidn %1,sp
     packuswb        m8, m10
-    vpermd          m8, m5, m8
+    packuswb        m12, m0
+    vpermd          m8, m1, m8
+    vpermd          m12, m1, m12
     vextracti128    xm10, m8, 1
-    movq            [r8], xm8
-    movhps          [r8 + r3], xm8
-    movq            [r8 + r3 * 2], xm10
-    movhps          [r8 + r6], xm10
-    add             r2, 8
+    vextracti128    xm0, m12, 1
+    movq            [r2], xm8
+    movhps          [r2 + r3], xm8
+    movq            [r2 + r3 * 2], xm10
+    movhps          [r2 + r6], xm10
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm12
+    movhps          [r2 + r3], xm12
+    movq            [r2 + r3 * 2], xm0
+    movhps          [r2 + r6], xm0
 %else
     vpermq          m8, m8, 11011000b
     vpermq          m10, m10, 11011000b
+    vpermq          m12, m12, 11011000b
+    vpermq          m0, m0, 11011000b
     vextracti128    xm9, m8, 1
     vextracti128    xm11, m10, 1
-    movu            [r8], xm8
-    movu            [r8 + r3], xm9
-    movu            [r8 + r3 * 2], xm10
-    movu            [r8 + r6], xm11
-    add             r2, 16
+    vextracti128    xm13, m12, 1
+    vextracti128    xm6, m0, 1
+    movu            [r2], xm8
+    movu            [r2 + r3], xm9
+    movu            [r2 + r3 * 2], xm10
+    movu            [r2 + r6], xm11
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm12
+    movu            [r2 + r3], xm13
+    movu            [r2 + r3 * 2], xm0
+    movu            [r2 + r6], xm6
 %endif
-    add             r0, 16
-    dec             r9d
-    jnz             .loopW
+
+    lea             r2, [r2 + r3 * 4]
+    sub             r0, r7
+    dec             r8d
+    jnz             .loopH
     RET
 %endif
 %endmacro
 
-    FILTER_VER_LUMA_S_AVX2_16x12 sp
-    FILTER_VER_LUMA_S_AVX2_16x12 ss
+    FILTER_VER_LUMA_S_AVX2_8xN sp, 16
+    FILTER_VER_LUMA_S_AVX2_8xN sp, 32
+    FILTER_VER_LUMA_S_AVX2_8xN ss, 16
+    FILTER_VER_LUMA_S_AVX2_8xN ss, 32
+
+%macro PROCESS_LUMA_S_AVX2_W8_4R 1
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m2, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    paddd           m1, m5
+    pmaddwd         m3, [r5]
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 2 * mmsize]
+    paddd           m0, m6
+    pmaddwd         m4, [r5 + 1 * mmsize]
+    paddd           m2, m4
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm4, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm4, 1
+    pmaddwd         m4, m5, [r5 + 2 * mmsize]
+    paddd           m1, m4
+    pmaddwd         m5, [r5 + 1 * mmsize]
+    paddd           m3, m5
+    movu            xm4, [r0 + r4]                  ; m4 = row 7
+    punpckhwd       xm5, xm6, xm4
+    punpcklwd       xm6, xm4
+    vinserti128     m6, m6, xm5, 1
+    pmaddwd         m5, m6, [r5 + 3 * mmsize]
+    paddd           m0, m5
+    pmaddwd         m6, [r5 + 2 * mmsize]
+    paddd           m2, m6
+    lea             r0, [r0 + r1 * 4]
+    movu            xm5, [r0]                       ; m5 = row 8
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 3 * mmsize]
+    paddd           m1, m6
+    pmaddwd         m4, [r5 + 2 * mmsize]
+    paddd           m3, m4
+    movu            xm6, [r0 + r1]                  ; m6 = row 9
+    punpckhwd       xm4, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm4, 1
+    pmaddwd         m5, [r5 + 3 * mmsize]
+    paddd           m2, m5
+    movu            xm4, [r0 + r1 * 2]              ; m4 = row 10
+    punpckhwd       xm5, xm6, xm4
+    punpcklwd       xm6, xm4
+    vinserti128     m6, m6, xm5, 1
+    pmaddwd         m6, [r5 + 3 * mmsize]
+    paddd           m3, m6
+
+%ifidn %1,sp
+    paddd           m0, m7
+    paddd           m1, m7
+    paddd           m2, m7
+    paddd           m3, m7
+    psrad           m0, 12
+    psrad           m1, 12
+    psrad           m2, 12
+    psrad           m3, 12
+%else
+    psrad           m0, 6
+    psrad           m1, 6
+    psrad           m2, 6
+    psrad           m3, 6
+%endif
+    packssdw        m0, m1
+    packssdw        m2, m3
+%ifidn %1,sp
+    packuswb        m0, m2
+    mova            m4, [interp8_hps_shuf]
+    vpermd          m0, m4, m0
+    vextracti128    xm2, m0, 1
+%else
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m2, 1
+%endif
+%endmacro
 
-%macro FILTER_VER_LUMA_S_AVX2_16x4 1
+%macro FILTER_VER_LUMA_S_AVX2_8x4 1
 INIT_YMM avx2
-cglobal interp_8tap_vert_%1_16x4, 4, 7, 8, 0 - gprsize
+cglobal interp_8tap_vert_%1_8x4, 4, 6, 8
     mov             r4d, r4m
     shl             r4d, 7
     add             r1d, r1d
@@ -25007,34 +20916,27 @@ cglobal interp_8tap_vert_%1_16x4, 4, 7, 8, 0 - gprsize
 %else
     add             r3d, r3d
 %endif
-    mov             dword [rsp], 2
-.loopW:
+
     PROCESS_LUMA_S_AVX2_W8_4R %1
-    lea             r6, [r3 * 3]
+    lea             r4, [r3 * 3]
 %ifidn %1,sp
     movq            [r2], xm0
     movhps          [r2 + r3], xm0
     movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r6], xm2
-    add             r2, 8
+    movhps          [r2 + r4], xm2
 %else
     movu            [r2], xm0
     movu            [r2 + r3], xm1
     movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-    add             r2, 16
+    movu            [r2 + r4], xm3
 %endif
-    lea             r6, [8 * r1 - 16]
-    sub             r0, r6
-    dec             dword [rsp]
-    jnz             .loopW
     RET
 %endmacro
 
-    FILTER_VER_LUMA_S_AVX2_16x4 sp
-    FILTER_VER_LUMA_S_AVX2_16x4 ss
+    FILTER_VER_LUMA_S_AVX2_8x4 sp
+    FILTER_VER_LUMA_S_AVX2_8x4 ss
 
-%macro PROCESS_LUMA_S_AVX2_W8_8R 1
+%macro PROCESS_LUMA_AVX2_W8_16R 1
     movu            xm0, [r0]                       ; m0 = row 0
     movu            xm1, [r0 + r1]                  ; m1 = row 1
     punpckhwd       xm2, xm0, xm1
@@ -25110,43 +21012,51 @@ cglobal interp_8tap_vert_%1_16x4, 4, 7, 8, 0 - gprsize
     paddd           m2, m10
     pmaddwd         m10, m8, [r5 + 2 * mmsize]
     paddd           m4, m10
-    pmaddwd         m8, [r5 + 1 * mmsize]
-    paddd           m6, m8
+    pmaddwd         m10, m8, [r5 + 1 * mmsize]
+    paddd           m6, m10
+    pmaddwd         m8, [r5]
     movu            xm10, [r7 + r1 * 2]             ; m10 = row 10
-    punpckhwd       xm8, xm9, xm10
+    punpckhwd       xm11, xm9, xm10
     punpcklwd       xm9, xm10
-    vinserti128     m9, m9, xm8, 1
-    pmaddwd         m8, m9, [r5 + 3 * mmsize]
-    paddd           m3, m8
-    pmaddwd         m8, m9, [r5 + 2 * mmsize]
-    paddd           m5, m8
-    pmaddwd         m9, [r5 + 1 * mmsize]
-    paddd           m7, m9
-    movu            xm8, [r7 + r4]                  ; m8 = row 11
-    punpckhwd       xm9, xm10, xm8
-    punpcklwd       xm10, xm8
-    vinserti128     m10, m10, xm9, 1
-    pmaddwd         m9, m10, [r5 + 3 * mmsize]
-    paddd           m4, m9
-    pmaddwd         m10, [r5 + 2 * mmsize]
-    paddd           m6, m10
+    vinserti128     m9, m9, xm11, 1
+    pmaddwd         m11, m9, [r5 + 3 * mmsize]
+    paddd           m3, m11
+    pmaddwd         m11, m9, [r5 + 2 * mmsize]
+    paddd           m5, m11
+    pmaddwd         m11, m9, [r5 + 1 * mmsize]
+    paddd           m7, m11
+    pmaddwd         m9, [r5]
+    movu            xm11, [r7 + r4]                 ; m11 = row 11
+    punpckhwd       xm12, xm10, xm11
+    punpcklwd       xm10, xm11
+    vinserti128     m10, m10, xm12, 1
+    pmaddwd         m12, m10, [r5 + 3 * mmsize]
+    paddd           m4, m12
+    pmaddwd         m12, m10, [r5 + 2 * mmsize]
+    paddd           m6, m12
+    pmaddwd         m12, m10, [r5 + 1 * mmsize]
+    paddd           m8, m12
+    pmaddwd         m10, [r5]
     lea             r7, [r7 + r1 * 4]
-    movu            xm9, [r7]                       ; m9 = row 12
-    punpckhwd       xm10, xm8, xm9
-    punpcklwd       xm8, xm9
-    vinserti128     m8, m8, xm10, 1
-    pmaddwd         m10, m8, [r5 + 3 * mmsize]
-    paddd           m5, m10
-    pmaddwd         m8, [r5 + 2 * mmsize]
-    paddd           m7, m8
+    movu            xm12, [r7]                      ; m12 = row 12
+    punpckhwd       xm13, xm11, xm12
+    punpcklwd       xm11, xm12
+    vinserti128     m11, m11, xm13, 1
+    pmaddwd         m13, m11, [r5 + 3 * mmsize]
+    paddd           m5, m13
+    pmaddwd         m13, m11, [r5 + 2 * mmsize]
+    paddd           m7, m13
+    pmaddwd         m13, m11, [r5 + 1 * mmsize]
+    paddd           m9, m13
+    pmaddwd         m11, [r5]
 
 %ifidn %1,sp
-    paddd           m0, m11
-    paddd           m1, m11
-    paddd           m2, m11
-    paddd           m3, m11
-    paddd           m4, m11
-    paddd           m5, m11
+    paddd           m0, m14
+    paddd           m1, m14
+    paddd           m2, m14
+    paddd           m3, m14
+    paddd           m4, m14
+    paddd           m5, m14
     psrad           m0, 12
     psrad           m1, 12
     psrad           m2, 12
@@ -25164,7 +21074,6 @@ cglobal interp_8tap_vert_%1_16x4, 4, 7, 8, 0 - gprsize
     packssdw        m0, m1
     packssdw        m2, m3
     packssdw        m4, m5
-
 %ifidn %1,sp
     packuswb        m0, m2
     mova            m5, [interp8_hps_shuf]
@@ -25185,22 +21094,32 @@ cglobal interp_8tap_vert_%1_16x4, 4, 7, 8, 0 - gprsize
     movu            [r2 + r6], xm3
 %endif
 
-    movu            xm10, [r7 + r1]                 ; m10 = row 13
-    punpckhwd       xm0, xm9, xm10
-    punpcklwd       xm9, xm10
-    vinserti128     m9, m9, xm0, 1
-    pmaddwd         m9, [r5 + 3 * mmsize]
-    paddd           m6, m9
+    movu            xm13, [r7 + r1]                 ; m13 = row 13
+    punpckhwd       xm0, xm12, xm13
+    punpcklwd       xm12, xm13
+    vinserti128     m12, m12, xm0, 1
+    pmaddwd         m0, m12, [r5 + 3 * mmsize]
+    paddd           m6, m0
+    pmaddwd         m0, m12, [r5 + 2 * mmsize]
+    paddd           m8, m0
+    pmaddwd         m0, m12, [r5 + 1 * mmsize]
+    paddd           m10, m0
+    pmaddwd         m12, [r5]
     movu            xm0, [r7 + r1 * 2]              ; m0 = row 14
-    punpckhwd       xm1, xm10, xm0
-    punpcklwd       xm10, xm0
-    vinserti128     m10, m10, xm1, 1
-    pmaddwd         m10, [r5 + 3 * mmsize]
-    paddd           m7, m10
+    punpckhwd       xm1, xm13, xm0
+    punpcklwd       xm13, xm0
+    vinserti128     m13, m13, xm1, 1
+    pmaddwd         m1, m13, [r5 + 3 * mmsize]
+    paddd           m7, m1
+    pmaddwd         m1, m13, [r5 + 2 * mmsize]
+    paddd           m9, m1
+    pmaddwd         m1, m13, [r5 + 1 * mmsize]
+    paddd           m11, m1
+    pmaddwd         m13, [r5]
 
 %ifidn %1,sp
-    paddd           m6, m11
-    paddd           m7, m11
+    paddd           m6, m14
+    paddd           m7, m14
     psrad           m6, 12
     psrad           m7, 12
 %else
@@ -25219,21 +21138,163 @@ cglobal interp_8tap_vert_%1_16x4, 4, 7, 8, 0 - gprsize
     movq            [r8 + r3 * 2], xm6
     movhps          [r8 + r6], xm6
 %else
-    vpermq          m4, m4, 11011000b
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm5, m4, 1
-    vextracti128    xm7, m6, 1
-    movu            [r8], xm4
-    movu            [r8 + r3], xm5
-    movu            [r8 + r3 * 2], xm6
-    movu            [r8 + r6], xm7
+    vpermq          m4, m4, 11011000b
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm1, m4, 1
+    vextracti128    xm7, m6, 1
+    movu            [r8], xm4
+    movu            [r8 + r3], xm1
+    movu            [r8 + r3 * 2], xm6
+    movu            [r8 + r6], xm7
+%endif
+
+    movu            xm1, [r7 + r4]                  ; m1 = row 15
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m2, m0, [r5 + 3 * mmsize]
+    paddd           m8, m2
+    pmaddwd         m2, m0, [r5 + 2 * mmsize]
+    paddd           m10, m2
+    pmaddwd         m2, m0, [r5 + 1 * mmsize]
+    paddd           m12, m2
+    pmaddwd         m0, [r5]
+    lea             r7, [r7 + r1 * 4]
+    movu            xm2, [r7]                       ; m2 = row 16
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m3, m1, [r5 + 3 * mmsize]
+    paddd           m9, m3
+    pmaddwd         m3, m1, [r5 + 2 * mmsize]
+    paddd           m11, m3
+    pmaddwd         m3, m1, [r5 + 1 * mmsize]
+    paddd           m13, m3
+    pmaddwd         m1, [r5]
+    movu            xm3, [r7 + r1]                  ; m3 = row 17
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 3 * mmsize]
+    paddd           m10, m4
+    pmaddwd         m4, m2, [r5 + 2 * mmsize]
+    paddd           m12, m4
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m0, m2
+    movu            xm4, [r7 + r1 * 2]              ; m4 = row 18
+    punpckhwd       xm2, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm2, 1
+    pmaddwd         m2, m3, [r5 + 3 * mmsize]
+    paddd           m11, m2
+    pmaddwd         m2, m3, [r5 + 2 * mmsize]
+    paddd           m13, m2
+    pmaddwd         m3, [r5 + 1 * mmsize]
+    paddd           m1, m3
+    movu            xm2, [r7 + r4]                  ; m2 = row 19
+    punpckhwd       xm6, xm4, xm2
+    punpcklwd       xm4, xm2
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 3 * mmsize]
+    paddd           m12, m6
+    pmaddwd         m4, [r5 + 2 * mmsize]
+    paddd           m0, m4
+    lea             r7, [r7 + r1 * 4]
+    movu            xm6, [r7]                       ; m6 = row 20
+    punpckhwd       xm7, xm2, xm6
+    punpcklwd       xm2, xm6
+    vinserti128     m2, m2, xm7, 1
+    pmaddwd         m7, m2, [r5 + 3 * mmsize]
+    paddd           m13, m7
+    pmaddwd         m2, [r5 + 2 * mmsize]
+    paddd           m1, m2
+    movu            xm7, [r7 + r1]                  ; m7 = row 21
+    punpckhwd       xm2, xm6, xm7
+    punpcklwd       xm6, xm7
+    vinserti128     m6, m6, xm2, 1
+    pmaddwd         m6, [r5 + 3 * mmsize]
+    paddd           m0, m6
+    movu            xm2, [r7 + r1 * 2]              ; m2 = row 22
+    punpckhwd       xm3, xm7, xm2
+    punpcklwd       xm7, xm2
+    vinserti128     m7, m7, xm3, 1
+    pmaddwd         m7, [r5 + 3 * mmsize]
+    paddd           m1, m7
+
+%ifidn %1,sp
+    paddd           m8, m14
+    paddd           m9, m14
+    paddd           m10, m14
+    paddd           m11, m14
+    paddd           m12, m14
+    paddd           m13, m14
+    paddd           m0, m14
+    paddd           m1, m14
+    psrad           m8, 12
+    psrad           m9, 12
+    psrad           m10, 12
+    psrad           m11, 12
+    psrad           m12, 12
+    psrad           m13, 12
+    psrad           m0, 12
+    psrad           m1, 12
+%else
+    psrad           m8, 6
+    psrad           m9, 6
+    psrad           m10, 6
+    psrad           m11, 6
+    psrad           m12, 6
+    psrad           m13, 6
+    psrad           m0, 6
+    psrad           m1, 6
+%endif
+    packssdw        m8, m9
+    packssdw        m10, m11
+    packssdw        m12, m13
+    packssdw        m0, m1
+    lea             r8, [r8 + r3 * 4]
+
+%ifidn %1,sp
+    packuswb        m8, m10
+    packuswb        m12, m0
+    vpermd          m8, m5, m8
+    vpermd          m12, m5, m12
+    vextracti128    xm10, m8, 1
+    vextracti128    xm0, m12, 1
+    movq            [r8], xm8
+    movhps          [r8 + r3], xm8
+    movq            [r8 + r3 * 2], xm10
+    movhps          [r8 + r6], xm10
+    lea             r8, [r8 + r3 * 4]
+    movq            [r8], xm12
+    movhps          [r8 + r3], xm12
+    movq            [r8 + r3 * 2], xm0
+    movhps          [r8 + r6], xm0
+%else
+    vpermq          m8, m8, 11011000b
+    vpermq          m10, m10, 11011000b
+    vpermq          m12, m12, 11011000b
+    vpermq          m0, m0, 11011000b
+    vextracti128    xm9, m8, 1
+    vextracti128    xm11, m10, 1
+    vextracti128    xm13, m12, 1
+    vextracti128    xm1, m0, 1
+    movu            [r8], xm8
+    movu            [r8 + r3], xm9
+    movu            [r8 + r3 * 2], xm10
+    movu            [r8 + r6], xm11
+    lea             r8, [r8 + r3 * 4]
+    movu            [r8], xm12
+    movu            [r8 + r3], xm13
+    movu            [r8 + r3 * 2], xm0
+    movu            [r8 + r6], xm1
 %endif
 %endmacro
 
-%macro FILTER_VER_LUMA_AVX2_Nx8 2
+%macro FILTER_VER_LUMA_AVX2_Nx16 2
 INIT_YMM avx2
 %if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_%2x8, 4, 10, 12
+cglobal interp_8tap_vert_%1_%2x16, 4, 10, 15
     mov             r4d, r4m
     shl             r4d, 7
     add             r1d, r1d
@@ -25248,14 +21309,14 @@ cglobal interp_8tap_vert_%1_%2x8, 4, 10, 12
     lea             r4, [r1 * 3]
     sub             r0, r4
 %ifidn %1,sp
-    mova            m11, [pd_526336]
+    mova            m14, [pd_526336]
 %else
     add             r3d, r3d
 %endif
     lea             r6, [r3 * 3]
     mov             r9d, %2 / 8
 .loopW:
-    PROCESS_LUMA_S_AVX2_W8_8R %1
+    PROCESS_LUMA_AVX2_W8_16R %1
 %ifidn %1,sp
     add             r2, 8
 %else
@@ -25268,15 +21329,17 @@ cglobal interp_8tap_vert_%1_%2x8, 4, 10, 12
 %endif
 %endmacro
 
-    FILTER_VER_LUMA_AVX2_Nx8 sp, 32
-    FILTER_VER_LUMA_AVX2_Nx8 sp, 16
-    FILTER_VER_LUMA_AVX2_Nx8 ss, 32
-    FILTER_VER_LUMA_AVX2_Nx8 ss, 16
+    FILTER_VER_LUMA_AVX2_Nx16 sp, 16
+    FILTER_VER_LUMA_AVX2_Nx16 sp, 32
+    FILTER_VER_LUMA_AVX2_Nx16 sp, 64
+    FILTER_VER_LUMA_AVX2_Nx16 ss, 16
+    FILTER_VER_LUMA_AVX2_Nx16 ss, 32
+    FILTER_VER_LUMA_AVX2_Nx16 ss, 64
 
-%macro FILTER_VER_LUMA_S_AVX2_32x24 1
+%macro FILTER_VER_LUMA_AVX2_NxN 3
 INIT_YMM avx2
 %if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_32x24, 4, 10, 15
+cglobal interp_8tap_vert_%3_%1x%2, 4, 12, 15
     mov             r4d, r4m
     shl             r4d, 7
     add             r1d, r1d
@@ -25290,2859 +21353,757 @@ cglobal interp_8tap_vert_%1_32x24, 4, 10, 15
 
     lea             r4, [r1 * 3]
     sub             r0, r4
-%ifidn %1,sp
+
+%ifidn %3,sp
     mova            m14, [pd_526336]
 %else
     add             r3d, r3d
 %endif
+
     lea             r6, [r3 * 3]
-    mov             r9d, 4
+    lea             r11, [r1 * 4]
+    mov             r9d, %2 / 16
+.loopH:
+    mov             r10d, %1 / 8
 .loopW:
-    PROCESS_LUMA_AVX2_W8_16R %1
-%ifidn %1,sp
+    PROCESS_LUMA_AVX2_W8_16R %3
+%ifidn %3,sp
     add             r2, 8
 %else
     add             r2, 16
 %endif
     add             r0, 16
-    dec             r9d
+    dec             r10d
     jnz             .loopW
-    lea             r9, [r1 * 4]
-    sub             r7, r9
-    lea             r0, [r7 - 48]
-%ifidn %1,sp
-    lea             r2, [r8 + r3 * 4 - 24]
-%else
-    lea             r2, [r8 + r3 * 4 - 48]
-%endif
-    mova            m11, m14
-    mov             r9d, 4
-.loop:
-    PROCESS_LUMA_S_AVX2_W8_8R %1
-%ifidn %1,sp
-    add             r2, 8
+    sub             r7, r11
+    lea             r0, [r7 - 2 * %1 + 16]
+%ifidn %3,sp
+    lea             r2, [r8 + r3 * 4 - %1 + 8]
 %else
-    add             r2, 16
+    lea             r2, [r8 + r3 * 4 - 2 * %1 + 16]
 %endif
-    add             r0, 16
     dec             r9d
-    jnz             .loop
+    jnz             .loopH
     RET
 %endif
 %endmacro
 
-    FILTER_VER_LUMA_S_AVX2_32x24 sp
-    FILTER_VER_LUMA_S_AVX2_32x24 ss
-
-;-----------------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_ps_32x32(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;-----------------------------------------------------------------------------------------------------------------------------;
-INIT_YMM avx2
-cglobal interp_4tap_horiz_ps_32x32, 4,6,8
-    mov             r4d, r4m
-    add             r3d, r3d
-    dec             r0
-
-    ; check isRowExt
-    cmp             r5m, byte 0
-
-    lea             r5, [tab_ChromaCoeff]
-    vpbroadcastw    m0, [r5 + r4 * 4 + 0]
-    vpbroadcastw    m1, [r5 + r4 * 4 + 2]
-    mova            m7, [pw_2000]
-
-    ; register map
-    ; m0 - interpolate coeff Low
-    ; m1 - interpolate coeff High
-    ; m7 - constant pw_2000
-    mov             r4d, 32
-    je             .loop
-    sub             r0, r1
-    add             r4d, 3
-
-.loop:
-    ; Row 0
-    movu            m2, [r0]
-    movu            m3, [r0 + 1]
-    punpckhbw       m4, m2, m3
-    punpcklbw       m2, m3
-    pmaddubsw       m4, m0
-    pmaddubsw       m2, m0
-
-    movu            m3, [r0 + 2]
-    movu            m5, [r0 + 3]
-    punpckhbw       m6, m3, m5
-    punpcklbw       m3, m5
-    pmaddubsw       m6, m1
-    pmaddubsw       m3, m1
-
-    paddw           m4, m6
-    paddw           m2, m3
-    psubw           m4, m7
-    psubw           m2, m7
-    vperm2i128      m3, m2, m4, 0x20
-    vperm2i128      m5, m2, m4, 0x31
-    movu            [r2], m3
-    movu            [r2 + mmsize], m5
-
-    add             r2, r3
-    add             r0, r1
-    dec             r4d
-    jnz            .loop
-    RET
+    FILTER_VER_LUMA_AVX2_NxN 16, 32, sp
+    FILTER_VER_LUMA_AVX2_NxN 16, 64, sp
+    FILTER_VER_LUMA_AVX2_NxN 24, 32, sp
+    FILTER_VER_LUMA_AVX2_NxN 32, 32, sp
+    FILTER_VER_LUMA_AVX2_NxN 32, 64, sp
+    FILTER_VER_LUMA_AVX2_NxN 48, 64, sp
+    FILTER_VER_LUMA_AVX2_NxN 64, 32, sp
+    FILTER_VER_LUMA_AVX2_NxN 64, 48, sp
+    FILTER_VER_LUMA_AVX2_NxN 64, 64, sp
+    FILTER_VER_LUMA_AVX2_NxN 16, 32, ss
+    FILTER_VER_LUMA_AVX2_NxN 16, 64, ss
+    FILTER_VER_LUMA_AVX2_NxN 24, 32, ss
+    FILTER_VER_LUMA_AVX2_NxN 32, 32, ss
+    FILTER_VER_LUMA_AVX2_NxN 32, 64, ss
+    FILTER_VER_LUMA_AVX2_NxN 48, 64, ss
+    FILTER_VER_LUMA_AVX2_NxN 64, 32, ss
+    FILTER_VER_LUMA_AVX2_NxN 64, 48, ss
+    FILTER_VER_LUMA_AVX2_NxN 64, 64, ss
 
-;-----------------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_ps_16x16(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;-----------------------------------------------------------------------------------------------------------------------------;
+%macro FILTER_VER_LUMA_S_AVX2_12x16 1
 INIT_YMM avx2
-cglobal interp_4tap_horiz_ps_16x16, 4,7,6
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_12x16, 4, 9, 15
     mov             r4d, r4m
-    mov             r5d, r5m
-    add             r3d, r3d
+    shl             r4d, 7
+    add             r1d, r1d
 
 %ifdef PIC
-    lea               r6,           [tab_ChromaCoeff]
-    vpbroadcastd      m0,           [r6 + r4 * 4]
+    lea             r5, [pw_LumaCoeffVer]
+    add             r5, r4
 %else
-    vpbroadcastd      m0,           [tab_ChromaCoeff + r4 * 4]
+    lea             r5, [pw_LumaCoeffVer + r4]
 %endif
 
-    vbroadcasti128     m2,           [pw_1]
-    vbroadcasti128     m5,           [pw_2000]
-    mova               m1,           [tab_Tm]
-
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-    mov                r6d,         16
-    dec                r0
-    test                r5d,        r5d
-    je                 .loop
-    sub                r0 ,         r1
-    add                r6d ,        3
-
-.loop:
-    ; Row 0
-    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-    vbroadcasti128    m4,           [r0 + 8]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-
-    packssdw          m3,           m4
-    psubw             m3,           m5
-    vpermq            m3,           m3,          11011000b
-    movu              [r2],         m3
-
-    add                r2,          r3
-    add                r0,          r1
-    dec                r6d
-    jnz                .loop
-    RET
-
-;-----------------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_ps_16xN(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;-----------------------------------------------------------------------------------------------------------------------------
-%macro IPFILTER_CHROMA_PS_16xN_AVX2 2
-INIT_YMM avx2
-cglobal interp_4tap_horiz_ps_%1x%2, 4,7,6
-    mov                    r4d,        r4m
-    mov                    r5d,        r5m
-    add                    r3d,        r3d
-
-%ifdef PIC
-    lea                    r6,         [tab_ChromaCoeff]
-    vpbroadcastd           m0,         [r6 + r4 * 4]
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+%ifidn %1,sp
+    mova            m14, [pd_526336]
 %else
-    vpbroadcastd           m0,         [tab_ChromaCoeff + r4 * 4]
+    add             r3d, r3d
 %endif
-
-    vbroadcasti128         m2,         [pw_1]
-    vbroadcasti128         m5,         [pw_2000]
-    mova                   m1,         [tab_Tm]
-
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-    mov                    r6d,        %2
-    dec                    r0
-    test                   r5d,        r5d
-    je                     .loop
-    sub                    r0 ,        r1
-    add                    r6d ,       3
-
-.loop:
-    ; Row 0
-    vbroadcasti128         m3,         [r0]
-    pshufb                 m3,         m1
-    pmaddubsw              m3,         m0
-    pmaddwd                m3,         m2
-    vbroadcasti128         m4,         [r0 + 8]
-    pshufb                 m4,         m1
-    pmaddubsw              m4,         m0
-    pmaddwd                m4,         m2
-
-    packssdw               m3,         m4
-    psubw                  m3,         m5
-
-    vpermq                 m3,         m3,          11011000b
-    movu                   [r2],       m3
-
-    add                    r2,         r3
-    add                    r0,         r1
-    dec                    r6d
-    jnz                    .loop
-    RET
-%endmacro
-
-    IPFILTER_CHROMA_PS_16xN_AVX2  16 , 32
-    IPFILTER_CHROMA_PS_16xN_AVX2  16 , 12
-    IPFILTER_CHROMA_PS_16xN_AVX2  16 , 8
-    IPFILTER_CHROMA_PS_16xN_AVX2  16 , 4
-    IPFILTER_CHROMA_PS_16xN_AVX2  16 , 24
-    IPFILTER_CHROMA_PS_16xN_AVX2  16 , 64
-
-;-----------------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_ps_32xN(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;-----------------------------------------------------------------------------------------------------------------------------
-%macro IPFILTER_CHROMA_PS_32xN_AVX2 2
-INIT_YMM avx2
-cglobal interp_4tap_horiz_ps_%1x%2, 4,7,6
-    mov                r4d,          r4m
-    mov                r5d,          r5m
-    add                r3d,          r3d
-
-%ifdef PIC
-    lea                r6,           [tab_ChromaCoeff]
-    vpbroadcastd       m0,           [r6 + r4 * 4]
+    lea             r6, [r3 * 3]
+    PROCESS_LUMA_AVX2_W8_16R %1
+%ifidn %1,sp
+    add             r2, 8
 %else
-    vpbroadcastd       m0,           [tab_ChromaCoeff + r4 * 4]
+    add             r2, 16
 %endif
-
-    vbroadcasti128     m2,           [pw_1]
-    vbroadcasti128     m5,           [pw_2000]
-    mova               m1,           [tab_Tm]
-
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-    mov                r6d,          %2
-    dec                r0
-    test               r5d,          r5d
-    je                 .loop
-    sub                r0 ,          r1
-    add                r6d ,         3
-
-.loop:
-    ; Row 0
-    vbroadcasti128     m3,           [r0]
-    pshufb             m3,           m1
-    pmaddubsw          m3,           m0
-    pmaddwd            m3,           m2
-    vbroadcasti128     m4,           [r0 + 8]
-    pshufb             m4,           m1
-    pmaddubsw          m4,           m0
-    pmaddwd            m4,           m2
-
-    packssdw           m3,           m4
-    psubw              m3,           m5
-
-    vpermq             m3,           m3,          11011000b
-    movu              [r2],          m3
-
-    vbroadcasti128     m3,           [r0 + 16]
-    pshufb             m3,           m1
-    pmaddubsw          m3,           m0
-    pmaddwd            m3,           m2
-    vbroadcasti128     m4,           [r0 + 24]
-    pshufb             m4,           m1
-    pmaddubsw          m4,           m0
-    pmaddwd            m4,           m2
-
-    packssdw           m3,           m4
-    psubw              m3,           m5
-
-    vpermq             m3,           m3,          11011000b
-    movu               [r2 + 32],    m3
-
-    add                r2,           r3
-    add                r0,           r1
-    dec                r6d
-    jnz                .loop
-    RET
-%endmacro
-
-    IPFILTER_CHROMA_PS_32xN_AVX2  32 , 16
-    IPFILTER_CHROMA_PS_32xN_AVX2  32 , 24
-    IPFILTER_CHROMA_PS_32xN_AVX2  32 , 8
-    IPFILTER_CHROMA_PS_32xN_AVX2  32 , 64
-    IPFILTER_CHROMA_PS_32xN_AVX2  32 , 48
-;-----------------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_ps_4x4(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;-----------------------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-cglobal interp_4tap_horiz_ps_4x4, 4,7,5
-    mov             r4d, r4m
-    mov             r5d, r5m
-    add             r3d, r3d
-
-%ifdef PIC
-    lea               r6,           [tab_ChromaCoeff]
-    vpbroadcastd      m0,           [r6 + r4 * 4]
-%else
-    vpbroadcastd      m0,           [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    vbroadcasti128     m2,           [pw_1]
-    vbroadcasti128     m1,           [tab_Tm]
-
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-
-    dec                r0
-    test                r5d,       r5d
-    je                 .label
-    sub                r0 , r1
-
-.label:
-    ; Row 0-1
-    movu              xm3,           [r0]
-    vinserti128       m3,           m3,      [r0 + r1],     1
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-
-    ; Row 2-3
-    lea               r0,           [r0 + r1 * 2]
-    movu              xm4,           [r0]
-    vinserti128       m4,           m4,      [r0 + r1],     1
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-
-    packssdw          m3,           m4
-    psubw             m3,           [pw_2000]
-    vextracti128      xm4,          m3,     1
-    movq              [r2],         xm3
-    movq              [r2+r3],      xm4
-    lea               r2,           [r2 + r3 * 2]
-    movhps            [r2],         xm3
-    movhps            [r2 + r3],    xm4
-
-    test                r5d,        r5d
-    jz                .end
-    lea               r2,           [r2 + r3 * 2]
-    lea               r0,           [r0 + r1 * 2]
-
-    ;Row 5-6
-    movu              xm3,          [r0]
-    vinserti128       m3,           m3,      [r0 + r1],     1
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-
-    ; Row 7
-    lea               r0,           [r0 + r1 * 2]
-    vbroadcasti128    m4,           [r0]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-
-    packssdw          m3,           m4
-    psubw             m3,           [pw_2000]
-
-    vextracti128      xm4,          m3,     1
-    movq              [r2],         xm3
-    movq              [r2+r3],      xm4
-    lea               r2,           [r2 + r3 * 2]
-    movhps            [r2],         xm3
-.end:
-    RET
-
-cglobal interp_4tap_horiz_ps_4x2, 4,7,5
-    mov             r4d, r4m
-    mov             r5d, r5m
-    add             r3d, r3d
-
-%ifdef PIC
-    lea               r6,           [tab_ChromaCoeff]
-    vpbroadcastd      m0,           [r6 + r4 * 4]
-%else
-    vpbroadcastd      m0,           [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    vbroadcasti128     m2,           [pw_1]
-    vbroadcasti128     m1,           [tab_Tm]
-
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-
-    dec                r0
-    test                r5d,       r5d
-    je                 .label
-    sub                r0 , r1
-
-.label:
-    ; Row 0-1
-    movu              xm3,           [r0]
-    vinserti128       m3,           m3,      [r0 + r1],     1
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-
-    packssdw          m3,           m3
-    psubw             m3,           [pw_2000]
-    vextracti128      xm4,          m3,     1
-    movq              [r2],         xm3
-    movq              [r2+r3],      xm4
-
-    test              r5d,          r5d
-    jz                .end
-    lea               r2,           [r2 + r3 * 2]
-    lea               r0,           [r0 + r1 * 2]
-
-    ;Row 2-3
-    movu              xm3,          [r0]
-    vinserti128       m3,           m3,      [r0 + r1],     1
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-
-    ; Row 5
-    lea               r0,           [r0 + r1 * 2]
-    vbroadcasti128    m4,           [r0]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-
-    packssdw          m3,           m4
-    psubw             m3,           [pw_2000]
-
-    vextracti128      xm4,          m3,     1
-    movq              [r2],         xm3
-    movq              [r2+r3],      xm4
-    lea               r2,           [r2 + r3 * 2]
-    movhps            [r2],         xm3
-.end:
+    add             r0, 16
+    mova            m7, m14
+    PROCESS_LUMA_AVX2_W4_16R %1
     RET
-
-;-----------------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_ps_4xN(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;-----------------------------------------------------------------------------------------------------------------------------;
-%macro IPFILTER_CHROMA_PS_4xN_AVX2 2
-INIT_YMM avx2
-cglobal interp_4tap_horiz_ps_%1x%2, 4,7,5
-    mov             r4d, r4m
-    mov             r5d, r5m
-    add             r3d, r3d
-
-%ifdef PIC
-    lea               r6,           [tab_ChromaCoeff]
-    vpbroadcastd      m0,           [r6 + r4 * 4]
-%else
-    vpbroadcastd      m0,           [tab_ChromaCoeff + r4 * 4]
 %endif
-
-    vbroadcasti128     m2,           [pw_1]
-    vbroadcasti128     m1,           [tab_Tm]
-
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-
-    mov              r4,                %2
-    dec              r0
-    test             r5d,       r5d
-    je               .loop
-    sub              r0 ,               r1
-
-
-.loop:
-    sub              r4d,           4
-    ; Row 0-1
-    movu              xm3,          [r0]
-    vinserti128       m3,           m3,      [r0 + r1],     1
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-
-    ; Row 2-3
-    lea               r0,           [r0 + r1 * 2]
-    movu              xm4,          [r0]
-    vinserti128       m4,           m4,      [r0 + r1],     1
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-
-    packssdw          m3,           m4
-    psubw             m3,           [pw_2000]
-    vextracti128      xm4,          m3,     1
-    movq              [r2],         xm3
-    movq              [r2+r3],      xm4
-    lea               r2,           [r2 + r3 * 2]
-    movhps            [r2],         xm3
-    movhps            [r2 + r3],    xm4
-
-    lea               r2,           [r2 + r3 * 2]
-    lea               r0,           [r0 + r1 * 2]
-
-    test              r4d,          r4d
-    jnz               .loop
-    test                r5d,        r5d
-    jz                .end
-
-    ;Row 5-6
-    movu              xm3,          [r0]
-    vinserti128       m3,           m3,      [r0 + r1],     1
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-
-    ; Row 7
-    lea               r0,           [r0 + r1 * 2]
-    vbroadcasti128    m4,           [r0]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-
-    packssdw          m3,           m4
-    psubw             m3,           [pw_2000]
-
-    vextracti128      xm4,          m3,     1
-    movq              [r2],         xm3
-    movq              [r2+r3],      xm4
-    lea               r2,           [r2 + r3 * 2]
-    movhps            [r2],         xm3
-.end:
-    RET
 %endmacro
 
-    IPFILTER_CHROMA_PS_4xN_AVX2  4 , 8
-    IPFILTER_CHROMA_PS_4xN_AVX2  4 , 16
-;-----------------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_ps_8x8(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;-----------------------------------------------------------------------------------------------------------------------------;
+    FILTER_VER_LUMA_S_AVX2_12x16 sp
+    FILTER_VER_LUMA_S_AVX2_12x16 ss
+
+%macro FILTER_VER_LUMA_S_AVX2_16x12 1
 INIT_YMM avx2
-cglobal interp_4tap_horiz_ps_8x8, 4,7,6
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_16x12, 4, 10, 15
     mov             r4d, r4m
-    mov             r5d, r5m
-    add             r3d, r3d
+    shl             r4d, 7
+    add             r1d, r1d
 
 %ifdef PIC
-    lea               r6,           [tab_ChromaCoeff]
-    vpbroadcastd      m0,           [r6 + r4 * 4]
+    lea             r5, [pw_LumaCoeffVer]
+    add             r5, r4
 %else
-    vpbroadcastd      m0,           [tab_ChromaCoeff + r4 * 4]
+    lea             r5, [pw_LumaCoeffVer + r4]
 %endif
 
-    vbroadcasti128     m2,           [pw_1]
-    vbroadcasti128     m5,           [pw_2000]
-    mova               m1,           [tab_Tm]
-
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-
-    mov                r6d,      4
-    dec                r0
-    test                r5d,     r5d
-    je                 .loop
-    sub                r0 ,      r1
-    add                r6d ,     1
-
-.loop:
-     dec               r6d
-    ; Row 0
-    vbroadcasti128    m3,           [r0]
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-
-    ; Row 1
-    vbroadcasti128    m4,           [r0 + r1]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-
-    packssdw          m3,           m4
-    psubw             m3,           m5
-
-    vpermq            m3,           m3,          11011000b
-    vextracti128      xm4,          m3,     1
-    movu             [r2],         xm3
-    movu             [r2 + r3],    xm4
-
-    lea               r2,           [r2 + r3 * 2]
-    lea               r0,           [r0 + r1 * 2]
-    test               r6d,          r6d
-    jnz               .loop
-    test              r5d,         r5d
-    je                .end
-
-    ;Row 11
-    vbroadcasti128    m3,           [r0]
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-
-    packssdw          m3,           m3
-    psubw             m3,           m5
-    vpermq            m3,           m3,          11011000b
-    movu             [r2],         xm3
-.end:
-    RET
-
-INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_4x2, 4,6,4
-    mov             r4d, r4m
-%ifdef PIC
-    lea               r5,           [tab_ChromaCoeff]
-    vpbroadcastd      m0,           [r5 + r4 * 4]
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+%ifidn %1,sp
+    mova            m14, [pd_526336]
 %else
-    vpbroadcastd      m0,           [tab_ChromaCoeff + r4 * 4]
+    add             r3d, r3d
 %endif
+    lea             r6, [r3 * 3]
+    mov             r9d, 2
+.loopW:
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m2, [r5]
+    lea             r7, [r0 + r1 * 4]
+    movu            xm4, [r7]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    paddd           m1, m5
+    pmaddwd         m3, [r5]
+    movu            xm5, [r7 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 2 * mmsize]
+    paddd           m0, m6
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m2, m6
+    pmaddwd         m4, [r5]
+    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm7, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddwd         m7, m5, [r5 + 2 * mmsize]
+    paddd           m1, m7
+    pmaddwd         m7, m5, [r5 + 1 * mmsize]
+    paddd           m3, m7
+    pmaddwd         m5, [r5]
+    movu            xm7, [r7 + r4]                  ; m7 = row 7
+    punpckhwd       xm8, xm6, xm7
+    punpcklwd       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddwd         m8, m6, [r5 + 3 * mmsize]
+    paddd           m0, m8
+    pmaddwd         m8, m6, [r5 + 2 * mmsize]
+    paddd           m2, m8
+    pmaddwd         m8, m6, [r5 + 1 * mmsize]
+    paddd           m4, m8
+    pmaddwd         m6, [r5]
+    lea             r7, [r7 + r1 * 4]
+    movu            xm8, [r7]                       ; m8 = row 8
+    punpckhwd       xm9, xm7, xm8
+    punpcklwd       xm7, xm8
+    vinserti128     m7, m7, xm9, 1
+    pmaddwd         m9, m7, [r5 + 3 * mmsize]
+    paddd           m1, m9
+    pmaddwd         m9, m7, [r5 + 2 * mmsize]
+    paddd           m3, m9
+    pmaddwd         m9, m7, [r5 + 1 * mmsize]
+    paddd           m5, m9
+    pmaddwd         m7, [r5]
+    movu            xm9, [r7 + r1]                  ; m9 = row 9
+    punpckhwd       xm10, xm8, xm9
+    punpcklwd       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddwd         m10, m8, [r5 + 3 * mmsize]
+    paddd           m2, m10
+    pmaddwd         m10, m8, [r5 + 2 * mmsize]
+    paddd           m4, m10
+    pmaddwd         m10, m8, [r5 + 1 * mmsize]
+    paddd           m6, m10
+    pmaddwd         m8, [r5]
+    movu            xm10, [r7 + r1 * 2]             ; m10 = row 10
+    punpckhwd       xm11, xm9, xm10
+    punpcklwd       xm9, xm10
+    vinserti128     m9, m9, xm11, 1
+    pmaddwd         m11, m9, [r5 + 3 * mmsize]
+    paddd           m3, m11
+    pmaddwd         m11, m9, [r5 + 2 * mmsize]
+    paddd           m5, m11
+    pmaddwd         m11, m9, [r5 + 1 * mmsize]
+    paddd           m7, m11
+    pmaddwd         m9, [r5]
+    movu            xm11, [r7 + r4]                 ; m11 = row 11
+    punpckhwd       xm12, xm10, xm11
+    punpcklwd       xm10, xm11
+    vinserti128     m10, m10, xm12, 1
+    pmaddwd         m12, m10, [r5 + 3 * mmsize]
+    paddd           m4, m12
+    pmaddwd         m12, m10, [r5 + 2 * mmsize]
+    paddd           m6, m12
+    pmaddwd         m12, m10, [r5 + 1 * mmsize]
+    paddd           m8, m12
+    pmaddwd         m10, [r5]
+    lea             r7, [r7 + r1 * 4]
+    movu            xm12, [r7]                      ; m12 = row 12
+    punpckhwd       xm13, xm11, xm12
+    punpcklwd       xm11, xm12
+    vinserti128     m11, m11, xm13, 1
+    pmaddwd         m13, m11, [r5 + 3 * mmsize]
+    paddd           m5, m13
+    pmaddwd         m13, m11, [r5 + 2 * mmsize]
+    paddd           m7, m13
+    pmaddwd         m13, m11, [r5 + 1 * mmsize]
+    paddd           m9, m13
+    pmaddwd         m11, [r5]
 
-    vbroadcasti128    m1,           [tab_Tm]
-
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-
-    ; Row 0-1
-    movu              xm2,          [r0 - 1]
-    vinserti128       m2,           m2,      [r0 + r1 - 1],     1
-    pshufb            m2,           m1
-    pmaddubsw         m2,           m0
-    pmaddwd           m2,           [pw_1]
-
-    packssdw          m2,           m2
-    pmulhrsw          m2,           [pw_512]
-    vextracti128      xm3,          m2,     1
-    packuswb          xm2,          xm3
-
-    movd              [r2],         xm2
-    pextrd            [r2+r3],      xm2,     2
-    RET
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp_32xN(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-%macro IPFILTER_CHROMA_PP_32xN_AVX2 2
-INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_%1x%2, 4,6,7
-    mov             r4d, r4m
-
-%ifdef PIC
-    lea               r5,           [tab_ChromaCoeff]
-    vpbroadcastd      m0,           [r5 + r4 * 4]
+%ifidn %1,sp
+    paddd           m0, m14
+    paddd           m1, m14
+    paddd           m2, m14
+    paddd           m3, m14
+    paddd           m4, m14
+    paddd           m5, m14
+    psrad           m0, 12
+    psrad           m1, 12
+    psrad           m2, 12
+    psrad           m3, 12
+    psrad           m4, 12
+    psrad           m5, 12
 %else
-    vpbroadcastd      m0,           [tab_ChromaCoeff + r4 * 4]
+    psrad           m0, 6
+    psrad           m1, 6
+    psrad           m2, 6
+    psrad           m3, 6
+    psrad           m4, 6
+    psrad           m5, 6
 %endif
+    packssdw        m0, m1
+    packssdw        m2, m3
+    packssdw        m4, m5
 
-    mova              m1,           [interp4_horiz_shuf1]
-    vpbroadcastd      m2,           [pw_1]
-    mova              m6,           [pw_512]
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-
-    dec               r0
-    mov               r4d,          %2
-
-.loop:
-    ; Row 0
-    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-    vbroadcasti128    m4,           [r0 + 4]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-    packssdw          m3,           m4
-    pmulhrsw          m3,           m6
-
-    vbroadcasti128    m4,           [r0 + 16]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-    vbroadcasti128    m5,           [r0 + 20]
-    pshufb            m5,           m1
-    pmaddubsw         m5,           m0
-    pmaddwd           m5,           m2
-    packssdw          m4,           m5
-    pmulhrsw          m4,           m6
-
-    packuswb          m3,           m4
-    vpermq            m3,           m3,      11011000b
-
-    movu              [r2],         m3
-    add               r2,           r3
-    add               r0,           r1
-    dec               r4d
-    jnz               .loop
-    RET
-%endmacro
-
-    IPFILTER_CHROMA_PP_32xN_AVX2 32, 16
-    IPFILTER_CHROMA_PP_32xN_AVX2 32, 24
-    IPFILTER_CHROMA_PP_32xN_AVX2 32, 8
-    IPFILTER_CHROMA_PP_32xN_AVX2 32, 64
-    IPFILTER_CHROMA_PP_32xN_AVX2 32, 48
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp_8xN(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-%macro IPFILTER_CHROMA_PP_8xN_AVX2 2
-INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_%1x%2, 4,6,6
-    mov               r4d,    r4m
-
-%ifdef PIC
-    lea               r5,           [tab_ChromaCoeff]
-    vpbroadcastd      m0,           [r5 + r4 * 4]
+%ifidn %1,sp
+    packuswb        m0, m2
+    mova            m5, [interp8_hps_shuf]
+    vpermd          m0, m5, m0
+    vextracti128    xm2, m0, 1
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r6], xm2
 %else
-    vpbroadcastd      m0,           [tab_ChromaCoeff + r4 * 4]
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m2, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
 %endif
 
-    movu              m1,           [tab_Tm]
-    vpbroadcastd      m2,           [pw_1]
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-
-    sub               r0,           1
-    mov               r4d,          %2
-
-.loop:
-    sub               r4d,          4
-    ; Row 0
-    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-
-    ; Row 1
-    vbroadcasti128    m4,           [r0 + r1]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-    packssdw          m3,           m4
-    pmulhrsw          m3,           [pw_512]
-    lea               r0,           [r0 + r1 * 2]
-
-    ; Row 2
-    vbroadcasti128    m4,           [r0 ]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-
-    ; Row 3
-    vbroadcasti128    m5,           [r0 + r1]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m5,           m1
-    pmaddubsw         m5,           m0
-    pmaddwd           m5,           m2
-    packssdw          m4,           m5
-    pmulhrsw          m4,           [pw_512]
-
-    packuswb          m3,           m4
-    mova              m5,           [interp_4tap_8x8_horiz_shuf]
-    vpermd            m3,           m5,     m3
-    vextracti128      xm4,          m3,     1
-    movq              [r2],         xm3
-    movhps            [r2 + r3],    xm3
-    lea               r2,           [r2 + r3 * 2]
-    movq              [r2],         xm4
-    movhps            [r2 + r3],    xm4
-    lea               r2,           [r2 + r3 * 2]
-    lea               r0,           [r0 + r1*2]
-    test              r4d,          r4d
-    jnz               .loop
-    RET
-%endmacro
-
-    IPFILTER_CHROMA_PP_8xN_AVX2   8 , 16
-    IPFILTER_CHROMA_PP_8xN_AVX2   8 , 32
-    IPFILTER_CHROMA_PP_8xN_AVX2   8 , 4
-    IPFILTER_CHROMA_PP_8xN_AVX2   8 , 64
-    IPFILTER_CHROMA_PP_8xN_AVX2   8 , 12
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp_4xN(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-%macro IPFILTER_CHROMA_PP_4xN_AVX2 2
-INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_%1x%2, 4,6,6
-    mov             r4d, r4m
+    movu            xm13, [r7 + r1]                 ; m13 = row 13
+    punpckhwd       xm0, xm12, xm13
+    punpcklwd       xm12, xm13
+    vinserti128     m12, m12, xm0, 1
+    pmaddwd         m0, m12, [r5 + 3 * mmsize]
+    paddd           m6, m0
+    pmaddwd         m0, m12, [r5 + 2 * mmsize]
+    paddd           m8, m0
+    pmaddwd         m12, [r5 + 1 * mmsize]
+    paddd           m10, m12
+    movu            xm0, [r7 + r1 * 2]              ; m0 = row 14
+    punpckhwd       xm1, xm13, xm0
+    punpcklwd       xm13, xm0
+    vinserti128     m13, m13, xm1, 1
+    pmaddwd         m1, m13, [r5 + 3 * mmsize]
+    paddd           m7, m1
+    pmaddwd         m1, m13, [r5 + 2 * mmsize]
+    paddd           m9, m1
+    pmaddwd         m13, [r5 + 1 * mmsize]
+    paddd           m11, m13
 
-%ifdef PIC
-    lea               r5,           [tab_ChromaCoeff]
-    vpbroadcastd      m0,           [r5 + r4 * 4]
+%ifidn %1,sp
+    paddd           m6, m14
+    paddd           m7, m14
+    psrad           m6, 12
+    psrad           m7, 12
 %else
-    vpbroadcastd      m0,           [tab_ChromaCoeff + r4 * 4]
+    psrad           m6, 6
+    psrad           m7, 6
 %endif
+    packssdw        m6, m7
+    lea             r8, [r2 + r3 * 4]
 
-    vpbroadcastd      m2,           [pw_1]
-    vbroadcasti128    m1,           [tab_Tm]
-    mov               r4d,          %2
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-
-    dec                r0
-
-.loop:
-    sub               r4d,          4
-    ; Row 0-1
-    movu              xm3,          [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    vinserti128       m3,           m3,      [r0 + r1],     1
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-
-    ; Row 2-3
-    lea               r0,           [r0 + r1 * 2]
-    movu              xm4,          [r0]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    vinserti128       m4,           m4,      [r0 + r1],     1
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-
-    packssdw          m3,           m4
-    pmulhrsw          m3,           [pw_512]
-    vextracti128      xm4,          m3,                     1
-    packuswb          xm3,          xm4
-
-    movd              [r2],         xm3
-    pextrd            [r2+r3],      xm3,                    2
-    lea               r2,           [r2 + r3 * 2]
-    pextrd            [r2],         xm3,                    1
-    pextrd            [r2+r3],      xm3,                    3
-
-    lea               r0,           [r0 + r1 * 2]
-    lea               r2,           [r2 + r3 * 2]
-    test              r4d,          r4d
-    jnz               .loop
-    RET
-%endmacro
-
-    IPFILTER_CHROMA_PP_4xN_AVX2  4 , 8
-    IPFILTER_CHROMA_PP_4xN_AVX2  4 , 16
-
-%macro IPFILTER_LUMA_PS_32xN_AVX2 2
-INIT_YMM avx2
-cglobal interp_8tap_horiz_ps_%1x%2, 4, 7, 8
-    mov                         r5d,               r5m
-    mov                         r4d,               r4m
-%ifdef PIC
-    lea                         r6,                [tab_LumaCoeff]
-    vpbroadcastq                m0,                [r6 + r4 * 8]
+%ifidn %1,sp
+    packuswb        m4, m6
+    vpermd          m4, m5, m4
+    vextracti128    xm6, m4, 1
+    movq            [r8], xm4
+    movhps          [r8 + r3], xm4
+    movq            [r8 + r3 * 2], xm6
+    movhps          [r8 + r6], xm6
 %else
-    vpbroadcastq                m0,                [tab_LumaCoeff + r4 * 8]
+    vpermq          m4, m4, 11011000b
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm1, m4, 1
+    vextracti128    xm7, m6, 1
+    movu            [r8], xm4
+    movu            [r8 + r3], xm1
+    movu            [r8 + r3 * 2], xm6
+    movu            [r8 + r6], xm7
 %endif
-    mova                        m6,                [tab_Lm + 32]
-    mova                        m1,                [tab_Lm]
-    mov                         r4d,                %2                           ;height
-    add                         r3d,               r3d
-    vbroadcasti128              m2,                [pw_1]
-    mova                        m7,                [interp8_hps_shuf]
-
-    ; register map
-    ; m0      - interpolate coeff
-    ; m1 , m6 - shuffle order table
-    ; m2      - pw_1
-
-
-    sub                         r0,                3
-    test                        r5d,               r5d
-    jz                          .label
-    lea                         r6,                [r1 * 3]                     ; r8 = (N / 2 - 1) * srcStride
-    sub                         r0,                r6
-    add                         r4d,                7
-
-.label:
-    lea                         r6,                 [pw_2000]
-.loop:
-    ; Row 0
-    vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb                      m4,                m3,             m6           ; row 0 (col 4 to 7)
-    pshufb                      m3,                m1                           ; shuffled based on the col order tab_Lm row 0 (col 0 to 3)
-    pmaddubsw                   m3,                m0
-    pmaddubsw                   m4,                m0
-    pmaddwd                     m3,                m2
-    pmaddwd                     m4,                m2
-    packssdw                    m3,                m4
-
-
-    vbroadcasti128              m4,                [r0 + 8]
-    pshufb                      m5,                m4,            m6            ;row 0 (col 12 to 15)
-    pshufb                      m4,                m1                           ;row 0 (col 8 to 11)
-    pmaddubsw                   m4,                m0
-    pmaddubsw                   m5,                m0
-    pmaddwd                     m4,                m2
-    pmaddwd                     m5,                m2
-    packssdw                    m4,                m5
-
-    pmaddwd                     m3,                m2
-    pmaddwd                     m4,                m2
-    packssdw                    m3,                m4
-    vpermd                      m3,                m7,               m3
-    psubw                       m3,                [r6]
-
-    movu                        [r2],              m3                          ;row 0
-
-    vbroadcasti128              m3,                [r0 + 16]
-    pshufb                      m4,                m3,             m6           ; row 0 (col 20 to 23)
-    pshufb                      m3,                m1                           ; row 0 (col 16 to 19)
-    pmaddubsw                   m3,                m0
-    pmaddubsw                   m4,                m0
-    pmaddwd                     m3,                m2
-    pmaddwd                     m4,                m2
-    packssdw                    m3,                m4
-
-    vbroadcasti128              m4,                [r0 + 24]
-    pshufb                      m5,                m4,            m6            ;row 0 (col 28 to 31)
-    pshufb                      m4,                m1                           ;row 0 (col 24 to 27)
-    pmaddubsw                   m4,                m0
-    pmaddubsw                   m5,                m0
-    pmaddwd                     m4,                m2
-    pmaddwd                     m5,                m2
-    packssdw                    m4,                m5
-
-    pmaddwd                     m3,                m2
-    pmaddwd                     m4,                m2
-    packssdw                    m3,                m4
-    vpermd                      m3,                m7,               m3
-    psubw                       m3,                [r6]
-
-    movu                        [r2 + 32],         m3                          ;row 0
-
-    add                         r0,                r1
-    add                         r2,                r3
-    dec                         r4d
-    jnz                         .loop
-    RET
-%endmacro
-
-    IPFILTER_LUMA_PS_32xN_AVX2 32 , 32
-    IPFILTER_LUMA_PS_32xN_AVX2 32 , 16
-    IPFILTER_LUMA_PS_32xN_AVX2 32 , 24
-    IPFILTER_LUMA_PS_32xN_AVX2 32 , 8
-    IPFILTER_LUMA_PS_32xN_AVX2 32 , 64
 
-INIT_YMM avx2
-cglobal interp_8tap_horiz_ps_48x64, 4, 7, 8
-    mov                         r5d,               r5m
-    mov                         r4d,               r4m
-%ifdef PIC
-    lea                         r6,                [tab_LumaCoeff]
-    vpbroadcastq                m0,                [r6 + r4 * 8]
-%else
-    vpbroadcastq                m0,                [tab_LumaCoeff + r4 * 8]
-%endif
-    mova                        m6,                [tab_Lm + 32]
-    mova                        m1,                [tab_Lm]
-    mov                         r4d,               64                           ;height
-    add                         r3d,               r3d
-    vbroadcasti128              m2,                [pw_2000]
-    mova                        m7,                [pw_1]
-
-    ; register map
-    ; m0      - interpolate coeff
-    ; m1 , m6 - shuffle order table
-    ; m2      - pw_2000
-
-    sub                         r0,                3
-    test                        r5d,               r5d
-    jz                          .label
-    lea                         r6,                [r1 * 3]                     ; r6 = (N / 2 - 1) * srcStride
-    sub                         r0,                r6                           ; r0(src)-r6
-    add                         r4d,                7                            ; blkheight += N - 1  (7 - 1 = 6 ; since the last one row not in loop)
-
-.label:
-    lea                         r6,                [interp8_hps_shuf]
-.loop:
-    ; Row 0
-    vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb                      m4,                m3,             m6           ; row 0 (col 4 to 7)
-    pshufb                      m3,                m1                           ; shuffled based on the col order tab_Lm row 0 (col 0 to 3)
-    pmaddubsw                   m3,                m0
-    pmaddubsw                   m4,                m0
-    pmaddwd                     m3,                m7
-    pmaddwd                     m4,                m7
-    packssdw                    m3,                m4
-
-    vbroadcasti128              m4,                [r0 + 8]
-    pshufb                      m5,                m4,             m6            ;row 0 (col 12 to 15)
-    pshufb                      m4,                m1                           ;row 0 (col 8 to 11)
-    pmaddubsw                   m4,                m0
-    pmaddubsw                   m5,                m0
-    pmaddwd                     m4,                m7
-    pmaddwd                     m5,                m7
-    packssdw                    m4,                m5
-    pmaddwd                     m3,                m7
-    pmaddwd                     m4,                m7
-    packssdw                    m3,                m4
-    mova                        m5,                [r6]
-    vpermd                      m3,                m5,             m3
-    psubw                       m3,                m2
-    movu                        [r2],              m3                          ;row 0
-
-    vbroadcasti128              m3,                [r0 + 16]
-    pshufb                      m4,                m3,             m6           ; row 0 (col 20 to 23)
-    pshufb                      m3,                m1                           ; row 0 (col 16 to 19)
-    pmaddubsw                   m3,                m0
-    pmaddubsw                   m4,                m0
-    pmaddwd                     m3,                m7
-    pmaddwd                     m4,                m7
-    packssdw                    m3,                m4
-
-    vbroadcasti128              m4,                [r0 + 24]
-    pshufb                      m5,                m4,             m6            ;row 0 (col 28 to 31)
-    pshufb                      m4,                m1                           ;row 0 (col 24 to 27)
-    pmaddubsw                   m4,                m0
-    pmaddubsw                   m5,                m0
-    pmaddwd                     m4,                m7
-    pmaddwd                     m5,                m7
-    packssdw                    m4,                m5
-    pmaddwd                     m3,                m7
-    pmaddwd                     m4,                m7
-    packssdw                    m3,                m4
-    mova                        m5,                [r6]
-    vpermd                      m3,                m5,               m3
-    psubw                       m3,                m2
-    movu                        [r2 + 32],         m3                          ;row 0
-
-    vbroadcasti128              m3,                [r0 + 32]
-    pshufb                      m4,                m3,             m6           ; row 0 (col 36 to 39)
-    pshufb                      m3,                m1                           ; row 0 (col 32 to 35)
-    pmaddubsw                   m3,                m0
-    pmaddubsw                   m4,                m0
-    pmaddwd                     m3,                m7
-    pmaddwd                     m4,                m7
-    packssdw                    m3,                m4
-
-    vbroadcasti128              m4,                [r0 + 40]
-    pshufb                      m5,                m4,            m6            ;row 0 (col 44 to 47)
-    pshufb                      m4,                m1                           ;row 0 (col 40 to 43)
-    pmaddubsw                   m4,                m0
-    pmaddubsw                   m5,                m0
-    pmaddwd                     m4,                m7
-    pmaddwd                     m5,                m7
-    packssdw                    m4,                m5
-    pmaddwd                     m3,                m7
-    pmaddwd                     m4,                m7
-    packssdw                    m3,                m4
-    mova                        m5,                [r6]
-    vpermd                      m3,                m5,               m3
-    psubw                       m3,                m2
-    movu                        [r2 + 64],         m3                          ;row 0
-
-    add                         r0,                r1
-    add                         r2,                r3
-    dec                         r4d
-    jnz                         .loop
-    RET
+    movu            xm1, [r7 + r4]                  ; m1 = row 15
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m2, m0, [r5 + 3 * mmsize]
+    paddd           m8, m2
+    pmaddwd         m0, [r5 + 2 * mmsize]
+    paddd           m10, m0
+    lea             r7, [r7 + r1 * 4]
+    movu            xm2, [r7]                       ; m2 = row 16
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m3, m1, [r5 + 3 * mmsize]
+    paddd           m9, m3
+    pmaddwd         m1, [r5 + 2 * mmsize]
+    paddd           m11, m1
+    movu            xm3, [r7 + r1]                  ; m3 = row 17
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m2, [r5 + 3 * mmsize]
+    paddd           m10, m2
+    movu            xm4, [r7 + r1 * 2]              ; m4 = row 18
+    punpckhwd       xm2, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm2, 1
+    pmaddwd         m3, [r5 + 3 * mmsize]
+    paddd           m11, m3
 
-INIT_YMM avx2
-cglobal interp_8tap_horiz_pp_24x32, 4,6,8
-    sub               r0,         3
-    mov               r4d,        r4m
-%ifdef PIC
-    lea               r5,         [tab_LumaCoeff]
-    vpbroadcastd      m0,         [r5 + r4 * 8]
-    vpbroadcastd      m1,         [r5 + r4 * 8 + 4]
+%ifidn %1,sp
+    paddd           m8, m14
+    paddd           m9, m14
+    paddd           m10, m14
+    paddd           m11, m14
+    psrad           m8, 12
+    psrad           m9, 12
+    psrad           m10, 12
+    psrad           m11, 12
 %else
-    vpbroadcastd      m0,         [tab_LumaCoeff + r4 * 8]
-    vpbroadcastd      m1,         [tab_LumaCoeff + r4 * 8 + 4]
+    psrad           m8, 6
+    psrad           m9, 6
+    psrad           m10, 6
+    psrad           m11, 6
 %endif
-    movu              m3,         [tab_Tm + 16]
-    vpbroadcastd      m7,         [pw_1]
-    lea               r5,         [tab_Tm]
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-    ; m2 , m2  shuffle order table
-    ; m7 - pw_1
-
-    mov               r4d,        32
-.loop:
-    ; Row 0
-    vbroadcasti128    m4,         [r0]                        ; [x E D C B A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m5,         m4,     m3
-    pshufb            m4,         [r5]
-    pmaddubsw         m4,         m0
-    pmaddubsw         m5,         m1
-    paddw             m4,         m5
-    pmaddwd           m4,         m7
-
-    vbroadcasti128    m5,         [r0 + 8]
-    pshufb            m6,         m5,     m3
-    pshufb            m5,         [r5]
-    pmaddubsw         m5,         m0
-    pmaddubsw         m6,         m1
-    paddw             m5,         m6
-    pmaddwd           m5,         m7
-    packssdw          m4,         m5                          ; [17 16 15 14 07 06 05 04 13 12 11 10 03 02 01 00]
-    pmulhrsw          m4,         [pw_512]
-
-    vbroadcasti128    m2,         [r0 + 16]
-    pshufb            m5,         m2,     m3
-    pshufb            m2,         [r5]
-    pmaddubsw         m2,         m0
-    pmaddubsw         m5,         m1
-    paddw             m2,         m5
-    pmaddwd           m2,         m7
-
-    packssdw          m2,         m2
-    pmulhrsw          m2,         [pw_512]
-    packuswb          m4,         m2
-    vpermq            m4,         m4,     11011000b
-    vextracti128      xm5,        m4,     1
-    pshufd            xm4,        xm4,    11011000b
-    pshufd            xm5,        xm5,    11011000b
-
-    movu              [r2],       xm4
-    movq              [r2 + 16],  xm5
-    add               r0,         r1
-    add               r2,         r3
-    dec               r4d
-    jnz               .loop
-    RET
+    packssdw        m8, m9
+    packssdw        m10, m11
+    lea             r8, [r8 + r3 * 4]
 
-INIT_YMM avx2
-cglobal interp_8tap_horiz_pp_12x16, 4,6,8
-    sub               r0,        3
-    mov               r4d,       r4m
-%ifdef PIC
-    lea               r5,        [tab_LumaCoeff]
-    vpbroadcastd      m0,        [r5 + r4 * 8]
-    vpbroadcastd      m1,        [r5 + r4 * 8 + 4]
+%ifidn %1,sp
+    packuswb        m8, m10
+    vpermd          m8, m5, m8
+    vextracti128    xm10, m8, 1
+    movq            [r8], xm8
+    movhps          [r8 + r3], xm8
+    movq            [r8 + r3 * 2], xm10
+    movhps          [r8 + r6], xm10
+    add             r2, 8
 %else
-    vpbroadcastd      m0,         [tab_LumaCoeff + r4 * 8]
-    vpbroadcastd      m1,         [tab_LumaCoeff + r4 * 8 + 4]
+    vpermq          m8, m8, 11011000b
+    vpermq          m10, m10, 11011000b
+    vextracti128    xm9, m8, 1
+    vextracti128    xm11, m10, 1
+    movu            [r8], xm8
+    movu            [r8 + r3], xm9
+    movu            [r8 + r3 * 2], xm10
+    movu            [r8 + r6], xm11
+    add             r2, 16
 %endif
-    movu              m3,         [tab_Tm + 16]
-    vpbroadcastd      m7,         [pw_1]
-    lea               r5,         [tab_Tm]
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-    ; m2 , m2  shuffle order table
-    ; m7 - pw_1
-
-    mov               r4d,        8
-.loop:
-    ; Row 0
-    vbroadcasti128    m4,         [r0]                        ;first 8 element
-    pshufb            m5,         m4,     m3
-    pshufb            m4,         [r5]
-    pmaddubsw         m4,         m0
-    pmaddubsw         m5,         m1
-    paddw             m4,         m5
-    pmaddwd           m4,         m7
-
-    vbroadcasti128    m5,         [r0 + 8]                    ; element 8 to 11
-    pshufb            m6,         m5,     m3
-    pshufb            m5,         [r5]
-    pmaddubsw         m5,         m0
-    pmaddubsw         m6,         m1
-    paddw             m5,         m6
-    pmaddwd           m5,         m7
-
-    packssdw          m4,         m5                          ; [17 16 15 14 07 06 05 04 13 12 11 10 03 02 01 00]
-    pmulhrsw          m4,         [pw_512]
-
-    ;Row 1
-    vbroadcasti128    m2,         [r0 + r1]
-    pshufb            m5,         m2,     m3
-    pshufb            m2,         [r5]
-    pmaddubsw         m2,         m0
-    pmaddubsw         m5,         m1
-    paddw             m2,         m5
-    pmaddwd           m2,         m7
-
-    vbroadcasti128    m5,         [r0 + r1 + 8]
-    pshufb            m6,         m5,     m3
-    pshufb            m5,         [r5]
-    pmaddubsw         m5,         m0
-    pmaddubsw         m6,         m1
-    paddw             m5,         m6
-    pmaddwd           m5,         m7
-
-    packssdw          m2,         m5
-    pmulhrsw          m2,         [pw_512]
-    packuswb          m4,         m2
-    vpermq            m4,         m4,     11011000b
-    vextracti128      xm5,        m4,     1
-    pshufd            xm4,        xm4,    11011000b
-    pshufd            xm5,        xm5,    11011000b
-
-    movq              [r2],       xm4
-    pextrd            [r2+8],     xm4,    2
-    movq              [r2 + r3],  xm5
-    pextrd            [r2+r3+8],  xm5,    2
-    lea               r0,         [r0 + r1 * 2]
-    lea               r2,         [r2 + r3 * 2]
-    dec               r4d
-    jnz              .loop
+    add             r0, 16
+    dec             r9d
+    jnz             .loopW
     RET
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp_16xN(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-%macro IPFILTER_CHROMA_PP_16xN_AVX2 2
-INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_%1x%2, 4, 6, 7
-    mov               r4d,          r4m
-
-%ifdef PIC
-    lea               r5,           [tab_ChromaCoeff]
-    vpbroadcastd      m0,           [r5 + r4 * 4]
-%else
-    vpbroadcastd      m0,           [tab_ChromaCoeff + r4 * 4]
 %endif
-
-    mova              m6,           [pw_512]
-    mova              m1,           [interp4_horiz_shuf1]
-    vpbroadcastd      m2,           [pw_1]
-
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-
-    dec               r0
-    mov               r4d,          %2/2
-
-.loop:
-    ; Row 0
-    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-    vbroadcasti128    m4,           [r0 + 4]                    ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-    packssdw          m3,           m4
-    pmulhrsw          m3,           m6
-
-    ; Row 1
-    vbroadcasti128    m4,           [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-    vbroadcasti128    m5,           [r0 + r1 + 4]               ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m5,           m1
-    pmaddubsw         m5,           m0
-    pmaddwd           m5,           m2
-    packssdw          m4,           m5
-    pmulhrsw          m4,           m6
-
-    packuswb          m3,           m4
-    vpermq            m3,           m3,      11011000b
-
-    vextracti128      xm4,          m3,       1
-    movu              [r2],         xm3
-    movu              [r2 + r3],    xm4
-    lea               r2,           [r2 + r3 * 2]
-    lea               r0,           [r0 + r1 * 2]
-    dec               r4d
-    jnz               .loop
-    RET
 %endmacro
 
-    IPFILTER_CHROMA_PP_16xN_AVX2 16 , 8
-    IPFILTER_CHROMA_PP_16xN_AVX2 16 , 32
-    IPFILTER_CHROMA_PP_16xN_AVX2 16 , 12
-    IPFILTER_CHROMA_PP_16xN_AVX2 16 , 4
-    IPFILTER_CHROMA_PP_16xN_AVX2 16 , 64
-    IPFILTER_CHROMA_PP_16xN_AVX2 16 , 24
-
-%macro IPFILTER_LUMA_PS_64xN_AVX2 1
-INIT_YMM avx2
-cglobal interp_8tap_horiz_ps_64x%1, 4, 7, 8
-    mov                         r5d,               r5m
-    mov                         r4d,               r4m
-%ifdef PIC
-    lea                         r6,                [tab_LumaCoeff]
-    vpbroadcastq                m0,                [r6 + r4 * 8]
-%else
-    vpbroadcastq                m0,                [tab_LumaCoeff + r4 * 8]
-%endif
-    mova                        m6,                [tab_Lm + 32]
-    mova                        m1,                [tab_Lm]
-    mov                         r4d,               %1                           ;height
-    add                         r3d,               r3d
-    vbroadcasti128              m2,                [pw_1]
-    mova                        m7,                [interp8_hps_shuf]
-
-    ; register map
-    ; m0      - interpolate coeff
-    ; m1 , m6 - shuffle order table
-    ; m2      - pw_2000
-
-    sub                         r0,                3
-    test                        r5d,               r5d
-    jz                          .label
-    lea                         r6,                [r1 * 3]
-    sub                         r0,                r6                           ; r0(src)-r6
-    add                         r4d,               7                            ; blkheight += N - 1
-
-.label:
-    lea                         r6,                [pw_2000]
-.loop:
-    ; Row 0
-    vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb                      m4,                m3,             m6           ; row 0 (col 4 to 7)
-    pshufb                      m3,                m1                           ; shuffled based on the col order tab_Lm row 0 (col 0 to 3)
-    pmaddubsw                   m3,                m0
-    pmaddubsw                   m4,                m0
-    pmaddwd                     m3,                m2
-    pmaddwd                     m4,                m2
-    packssdw                    m3,                m4
-
-    vbroadcasti128              m4,                [r0 + 8]
-    pshufb                      m5,                m4,            m6            ;row 0 (col 12 to 15)
-    pshufb                      m4,                m1                           ;row 0 (col 8 to 11)
-    pmaddubsw                   m4,                m0
-    pmaddubsw                   m5,                m0
-    pmaddwd                     m4,                m2
-    pmaddwd                     m5,                m2
-    packssdw                    m4,                m5
-    pmaddwd                     m3,                m2
-    pmaddwd                     m4,                m2
-    packssdw                    m3,                m4
-    vpermd                      m3,                m7,               m3
-    psubw                       m3,                [r6]
-    movu                        [r2],              m3                          ;row 0
-
-    vbroadcasti128              m3,                [r0 + 16]
-    pshufb                      m4,                m3,             m6           ; row 0 (col 20 to 23)
-    pshufb                      m3,                m1                           ; row 0 (col 16 to 19)
-    pmaddubsw                   m3,                m0
-    pmaddubsw                   m4,                m0
-    pmaddwd                     m3,                m2
-    pmaddwd                     m4,                m2
-    packssdw                    m3,                m4
-
-    vbroadcasti128              m4,                [r0 + 24]
-    pshufb                      m5,                m4,            m6            ;row 0 (col 28 to 31)
-    pshufb                      m4,                m1                           ;row 0 (col 24 to 27)
-    pmaddubsw                   m4,                m0
-    pmaddubsw                   m5,                m0
-    pmaddwd                     m4,                m2
-    pmaddwd                     m5,                m2
-    packssdw                    m4,                m5
-    pmaddwd                     m3,                m2
-    pmaddwd                     m4,                m2
-    packssdw                    m3,                m4
-    vpermd                      m3,                m7,               m3
-    psubw                       m3,                [r6]
-    movu                        [r2 + 32],         m3                          ;row 0
-
-    vbroadcasti128              m3,                [r0 + 32]
-    pshufb                      m4,                m3,             m6           ; row 0 (col 36 to 39)
-    pshufb                      m3,                m1                           ; row 0 (col 32 to 35)
-    pmaddubsw                   m3,                m0
-    pmaddubsw                   m4,                m0
-    pmaddwd                     m3,                m2
-    pmaddwd                     m4,                m2
-    packssdw                    m3,                m4
-
-    vbroadcasti128              m4,                [r0 + 40]
-    pshufb                      m5,                m4,            m6            ;row 0 (col 44 to 47)
-    pshufb                      m4,                m1                           ;row 0 (col 40 to 43)
-    pmaddubsw                   m4,                m0
-    pmaddubsw                   m5,                m0
-    pmaddwd                     m4,                m2
-    pmaddwd                     m5,                m2
-    packssdw                    m4,                m5
-    pmaddwd                     m3,                m2
-    pmaddwd                     m4,                m2
-    packssdw                    m3,                m4
-    vpermd                      m3,                m7,               m3
-    psubw                       m3,                [r6]
-    movu                        [r2 + 64],         m3                          ;row 0
-    vbroadcasti128              m3,                [r0 + 48]
-    pshufb                      m4,                m3,             m6           ; row 0 (col 52 to 55)
-    pshufb                      m3,                m1                           ; row 0 (col 48 to 51)
-    pmaddubsw                   m3,                m0
-    pmaddubsw                   m4,                m0
-    pmaddwd                     m3,                m2
-    pmaddwd                     m4,                m2
-    packssdw                    m3,                m4
-
-    vbroadcasti128              m4,                [r0 + 56]
-    pshufb                      m5,                m4,            m6            ;row 0 (col 60 to 63)
-    pshufb                      m4,                m1                           ;row 0 (col 56 to 59)
-    pmaddubsw                   m4,                m0
-    pmaddubsw                   m5,                m0
-    pmaddwd                     m4,                m2
-    pmaddwd                     m5,                m2
-    packssdw                    m4,                m5
-    pmaddwd                     m3,                m2
-    pmaddwd                     m4,                m2
-    packssdw                    m3,                m4
-    vpermd                      m3,                m7,               m3
-    psubw                       m3,                [r6]
-    movu                        [r2 + 96],         m3                          ;row 0
-
-    add                          r0,                r1
-    add                          r2,                r3
-    dec                          r4d
-    jnz                         .loop
-    RET
-%endmacro
-
-    IPFILTER_LUMA_PS_64xN_AVX2 64
-    IPFILTER_LUMA_PS_64xN_AVX2 48
-    IPFILTER_LUMA_PS_64xN_AVX2 32
-    IPFILTER_LUMA_PS_64xN_AVX2 16
+    FILTER_VER_LUMA_S_AVX2_16x12 sp
+    FILTER_VER_LUMA_S_AVX2_16x12 ss
 
-;-----------------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_ps_8xN(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;-----------------------------------------------------------------------------------------------------------------------------
-%macro IPFILTER_CHROMA_PS_8xN_AVX2 1
+%macro FILTER_VER_LUMA_S_AVX2_16x4 1
 INIT_YMM avx2
-cglobal interp_4tap_horiz_ps_8x%1, 4,7,6
-    mov                r4d,             r4m
-    mov                r5d,             r5m
-    add                r3d,             r3d
+cglobal interp_8tap_vert_%1_16x4, 4, 7, 8, 0 - gprsize
+    mov             r4d, r4m
+    shl             r4d, 7
+    add             r1d, r1d
 
 %ifdef PIC
-    lea                r6,              [tab_ChromaCoeff]
-    vpbroadcastd       m0,              [r6 + r4 * 4]
+    lea             r5, [pw_LumaCoeffVer]
+    add             r5, r4
 %else
-    vpbroadcastd       m0,              [tab_ChromaCoeff + r4 * 4]
+    lea             r5, [pw_LumaCoeffVer + r4]
 %endif
 
-    vbroadcasti128     m2,              [pw_1]
-    vbroadcasti128     m5,              [pw_2000]
-    mova               m1,              [tab_Tm]
-
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-
-    mov                r6d,             %1/2
-    dec                r0
-    test               r5d,             r5d
-    jz                 .loop
-    sub                r0 ,             r1
-    inc                r6d
-
-.loop:
-    ; Row 0
-    vbroadcasti128     m3,              [r0]
-    pshufb             m3,              m1
-    pmaddubsw          m3,              m0
-    pmaddwd            m3,              m2
-
-    ; Row 1
-    vbroadcasti128     m4,              [r0 + r1]
-    pshufb             m4,              m1
-    pmaddubsw          m4,              m0
-    pmaddwd            m4,              m2
-    packssdw           m3,              m4
-    psubw              m3,              m5
-    vpermq             m3,              m3,          11011000b
-    vextracti128       xm4,             m3,          1
-    movu               [r2],            xm3
-    movu               [r2 + r3],       xm4
-
-    lea                r2,              [r2 + r3 * 2]
-    lea                r0,              [r0 + r1 * 2]
-    dec                r6d
-    jnz                .loop
-    test               r5d,             r5d
-    jz                 .end
-
-    ;Row 11
-    vbroadcasti128     m3,              [r0]
-    pshufb             m3,              m1
-    pmaddubsw          m3,              m0
-    pmaddwd            m3,              m2
-    packssdw           m3,              m3
-    psubw              m3,              m5
-    vpermq             m3,              m3,          11011000b
-    movu               [r2],            xm3
-.end:
-    RET
-%endmacro
-
-    IPFILTER_CHROMA_PS_8xN_AVX2  2
-    IPFILTER_CHROMA_PS_8xN_AVX2  32
-    IPFILTER_CHROMA_PS_8xN_AVX2  16
-    IPFILTER_CHROMA_PS_8xN_AVX2  6
-    IPFILTER_CHROMA_PS_8xN_AVX2  4
-    IPFILTER_CHROMA_PS_8xN_AVX2  12
-    IPFILTER_CHROMA_PS_8xN_AVX2  64
-
-INIT_YMM avx2
-cglobal interp_4tap_horiz_ps_2x4, 4, 7, 3
-    mov                r4d,            r4m
-    mov                r5d,            r5m
-    add                r3d,            r3d
-%ifdef PIC
-    lea                r6,             [tab_ChromaCoeff]
-    vpbroadcastd       m0,             [r6 + r4 * 4]
-%else
-    vpbroadcastd       m0,             [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    mova               xm3,            [pw_2000]
-    dec                r0
-    test               r5d,            r5d
-    jz                 .label
-    sub                r0,             r1
-
-.label:
-    lea                r6,             [r1 * 3]
-    movq               xm1,            [r0]
-    movhps             xm1,            [r0 + r1]
-    movq               xm2,            [r0 + r1 * 2]
-    movhps             xm2,            [r0 + r6]
-
-    vinserti128        m1,             m1,          xm2,          1
-    pshufb             m1,             [interp4_hpp_shuf]
-    pmaddubsw          m1,             m0
-    pmaddwd            m1,             [pw_1]
-    vextracti128       xm2,            m1,          1
-    packssdw           xm1,            xm2
-    psubw              xm1,            xm3
-
-    lea                r4,             [r3 * 3]
-    movd               [r2],           xm1
-    pextrd             [r2 + r3],      xm1,         1
-    pextrd             [r2 + r3 * 2],  xm1,         2
-    pextrd             [r2 + r4],      xm1,         3
-
-    test               r5d,            r5d
-    jz                .end
-    lea                r2,             [r2 + r3 * 4]
-    lea                r0,             [r0 + r1 * 4]
-
-    movq               xm1,            [r0]
-    movhps             xm1,            [r0 + r1]
-    movq               xm2,            [r0 + r1 * 2]
-    vinserti128        m1,             m1,          xm2,          1
-    pshufb             m1,             [interp4_hpp_shuf]
-    pmaddubsw          m1,             m0
-    pmaddwd            m1,             [pw_1]
-    vextracti128       xm2,            m1,          1
-    packssdw           xm1,            xm2
-    psubw              xm1,            xm3
-
-    movd               [r2],           xm1
-    pextrd             [r2 + r3],      xm1,         1
-    pextrd             [r2 + r3 * 2],  xm1,         2
-.end:
-    RET
-
-INIT_YMM avx2
-cglobal interp_4tap_horiz_ps_2x8, 4, 7, 7
-    mov               r4d,           r4m
-    mov               r5d,           r5m
-    add               r3d,           r3d
-
-%ifdef PIC
-    lea               r6,            [tab_ChromaCoeff]
-    vpbroadcastd      m0,            [r6 + r4 * 4]
-%else
-    vpbroadcastd      m0,            [tab_ChromaCoeff + r4 * 4]
-%endif
-    vbroadcasti128    m6,            [pw_2000]
-    test              r5d,            r5d
-    jz                .label
-    sub               r0,             r1
-
-.label:
-    mova              m4,            [interp4_hpp_shuf]
-    mova              m5,            [pw_1]
-    dec               r0
-    lea               r4,            [r1 * 3]
-    movq              xm1,           [r0]                                   ;row 0
-    movhps            xm1,           [r0 + r1]
-    movq              xm2,           [r0 + r1 * 2]
-    movhps            xm2,           [r0 + r4]
-    vinserti128       m1,            m1,          xm2,          1
-    lea               r0,            [r0 + r1 * 4]
-    movq              xm3,           [r0]
-    movhps            xm3,           [r0 + r1]
-    movq              xm2,           [r0 + r1 * 2]
-    movhps            xm2,           [r0 + r4]
-    vinserti128       m3,            m3,          xm2,          1
-
-    pshufb            m1,            m4
-    pshufb            m3,            m4
-    pmaddubsw         m1,            m0
-    pmaddubsw         m3,            m0
-    pmaddwd           m1,            m5
-    pmaddwd           m3,            m5
-    packssdw          m1,            m3
-    psubw             m1,            m6
-
-    lea               r4,            [r3 * 3]
-    vextracti128      xm2,           m1,          1
-
-    movd              [r2],          xm1
-    pextrd            [r2 + r3],     xm1,         1
-    movd              [r2 + r3 * 2], xm2
-    pextrd            [r2 + r4],     xm2,         1
-    lea               r2,            [r2 + r3 * 4]
-    pextrd            [r2],          xm1,         2
-    pextrd            [r2 + r3],     xm1,         3
-    pextrd            [r2 + r3 * 2], xm2,         2
-    pextrd            [r2 + r4],     xm2,         3
-    test              r5d,            r5d
-    jz                .end
-
-    lea               r0,            [r0 + r1 * 4]
-    lea               r2,            [r2 + r3 * 4]
-    movq              xm1,           [r0]                                   ;row 0
-    movhps            xm1,           [r0 + r1]
-    movq              xm2,           [r0 + r1 * 2]
-    vinserti128       m1,            m1,          xm2,          1
-    pshufb            m1,            m4
-    pmaddubsw         m1,            m0
-    pmaddwd           m1,            m5
-    packssdw          m1,            m1
-    psubw             m1,            m6
-    vextracti128      xm2,           m1,          1
-
-    movd              [r2],          xm1
-    pextrd            [r2 + r3],     xm1,         1
-    movd              [r2 + r3 * 2], xm2
-.end:
-    RET
-
-INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_12x16, 4, 6, 7
-    mov               r4d,          r4m
-
-%ifdef PIC
-    lea               r5,           [tab_ChromaCoeff]
-    vpbroadcastd      m0,           [r5 + r4 * 4]
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+%ifidn %1,sp
+    mova            m7, [pd_526336]
 %else
-    vpbroadcastd      m0,           [tab_ChromaCoeff + r4 * 4]
+    add             r3d, r3d
 %endif
-
-    mova              m6,           [pw_512]
-    mova              m1,           [interp4_horiz_shuf1]
-    vpbroadcastd      m2,           [pw_1]
-
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-
-    dec               r0
-    mov               r4d,          8
-
-.loop:
-    ; Row 0
-    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-    vbroadcasti128    m4,           [r0 + 4]                    ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-    packssdw          m3,           m4
-    pmulhrsw          m3,           m6
-
-    ; Row 1
-    vbroadcasti128    m4,           [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-    vbroadcasti128    m5,           [r0 + r1 + 4]               ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m5,           m1
-    pmaddubsw         m5,           m0
-    pmaddwd           m5,           m2
-    packssdw          m4,           m5
-    pmulhrsw          m4,           m6
-
-    packuswb          m3,           m4
-    vpermq            m3,           m3,      11011000b
-
-    vextracti128      xm4,          m3,       1
-    movq              [r2],         xm3
-    pextrd            [r2+8],       xm3,      2
-    movq              [r2 + r3],    xm4
-    pextrd            [r2 + r3 + 8],xm4,      2
-    lea               r2,           [r2 + r3 * 2]
-    lea               r0,           [r0 + r1 * 2]
-    dec               r4d
-    jnz               .loop
-    RET
-
-INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_24x32, 4,6,7
-    mov              r4d,           r4m
-
-%ifdef PIC
-    lea               r5,           [tab_ChromaCoeff]
-    vpbroadcastd      m0,           [r5 + r4 * 4]
+    mov             dword [rsp], 2
+.loopW:
+    PROCESS_LUMA_S_AVX2_W8_4R %1
+    lea             r6, [r3 * 3]
+%ifidn %1,sp
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r6], xm2
+    add             r2, 8
 %else
-    vpbroadcastd      m0,           [tab_ChromaCoeff + r4 * 4]
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
+    add             r2, 16
 %endif
-
-    mova              m1,           [interp4_horiz_shuf1]
-    vpbroadcastd      m2,           [pw_1]
-    mova              m6,           [pw_512]
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-
-    dec               r0
-    mov               r4d,          32
-
-.loop:
-    ; Row 0
-    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-    vbroadcasti128    m4,           [r0 + 4]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-    packssdw          m3,           m4
-    pmulhrsw          m3,           m6
-
-    vbroadcasti128    m4,           [r0 + 16]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-    vbroadcasti128    m5,           [r0 + 20]
-    pshufb            m5,           m1
-    pmaddubsw         m5,           m0
-    pmaddwd           m5,           m2
-    packssdw          m4,           m5
-    pmulhrsw          m4,           m6
-
-    packuswb          m3,           m4
-    vpermq            m3,           m3,      11011000b
-
-    vextracti128      xm4,          m3,       1
-    movu              [r2],         xm3
-    movq              [r2 + 16],    xm4
-    add               r2,           r3
-    add               r0,           r1
-    dec               r4d
-    jnz               .loop
+    lea             r6, [8 * r1 - 16]
+    sub             r0, r6
+    dec             dword [rsp]
+    jnz             .loopW
     RET
+%endmacro
 
-;-----------------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_ps_6x8(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;-----------------------------------------------------------------------------------------------------------------------------;
-INIT_YMM avx2
-cglobal interp_4tap_horiz_ps_6x8, 4,7,6
-    mov                r4d,            r4m
-    mov                r5d,            r5m
-    add                r3d,            r3d
+    FILTER_VER_LUMA_S_AVX2_16x4 sp
+    FILTER_VER_LUMA_S_AVX2_16x4 ss
 
-%ifdef PIC
-    lea                r6,             [tab_ChromaCoeff]
-    vpbroadcastd       m0,             [r6 + r4 * 4]
+%macro PROCESS_LUMA_S_AVX2_W8_8R 1
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m2, [r5]
+    lea             r7, [r0 + r1 * 4]
+    movu            xm4, [r7]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    paddd           m1, m5
+    pmaddwd         m3, [r5]
+    movu            xm5, [r7 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 2 * mmsize]
+    paddd           m0, m6
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m2, m6
+    pmaddwd         m4, [r5]
+    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm7, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddwd         m7, m5, [r5 + 2 * mmsize]
+    paddd           m1, m7
+    pmaddwd         m7, m5, [r5 + 1 * mmsize]
+    paddd           m3, m7
+    pmaddwd         m5, [r5]
+    movu            xm7, [r7 + r4]                  ; m7 = row 7
+    punpckhwd       xm8, xm6, xm7
+    punpcklwd       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddwd         m8, m6, [r5 + 3 * mmsize]
+    paddd           m0, m8
+    pmaddwd         m8, m6, [r5 + 2 * mmsize]
+    paddd           m2, m8
+    pmaddwd         m8, m6, [r5 + 1 * mmsize]
+    paddd           m4, m8
+    pmaddwd         m6, [r5]
+    lea             r7, [r7 + r1 * 4]
+    movu            xm8, [r7]                       ; m8 = row 8
+    punpckhwd       xm9, xm7, xm8
+    punpcklwd       xm7, xm8
+    vinserti128     m7, m7, xm9, 1
+    pmaddwd         m9, m7, [r5 + 3 * mmsize]
+    paddd           m1, m9
+    pmaddwd         m9, m7, [r5 + 2 * mmsize]
+    paddd           m3, m9
+    pmaddwd         m9, m7, [r5 + 1 * mmsize]
+    paddd           m5, m9
+    pmaddwd         m7, [r5]
+    movu            xm9, [r7 + r1]                  ; m9 = row 9
+    punpckhwd       xm10, xm8, xm9
+    punpcklwd       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddwd         m10, m8, [r5 + 3 * mmsize]
+    paddd           m2, m10
+    pmaddwd         m10, m8, [r5 + 2 * mmsize]
+    paddd           m4, m10
+    pmaddwd         m8, [r5 + 1 * mmsize]
+    paddd           m6, m8
+    movu            xm10, [r7 + r1 * 2]             ; m10 = row 10
+    punpckhwd       xm8, xm9, xm10
+    punpcklwd       xm9, xm10
+    vinserti128     m9, m9, xm8, 1
+    pmaddwd         m8, m9, [r5 + 3 * mmsize]
+    paddd           m3, m8
+    pmaddwd         m8, m9, [r5 + 2 * mmsize]
+    paddd           m5, m8
+    pmaddwd         m9, [r5 + 1 * mmsize]
+    paddd           m7, m9
+    movu            xm8, [r7 + r4]                  ; m8 = row 11
+    punpckhwd       xm9, xm10, xm8
+    punpcklwd       xm10, xm8
+    vinserti128     m10, m10, xm9, 1
+    pmaddwd         m9, m10, [r5 + 3 * mmsize]
+    paddd           m4, m9
+    pmaddwd         m10, [r5 + 2 * mmsize]
+    paddd           m6, m10
+    lea             r7, [r7 + r1 * 4]
+    movu            xm9, [r7]                       ; m9 = row 12
+    punpckhwd       xm10, xm8, xm9
+    punpcklwd       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddwd         m10, m8, [r5 + 3 * mmsize]
+    paddd           m5, m10
+    pmaddwd         m8, [r5 + 2 * mmsize]
+    paddd           m7, m8
+
+%ifidn %1,sp
+    paddd           m0, m11
+    paddd           m1, m11
+    paddd           m2, m11
+    paddd           m3, m11
+    paddd           m4, m11
+    paddd           m5, m11
+    psrad           m0, 12
+    psrad           m1, 12
+    psrad           m2, 12
+    psrad           m3, 12
+    psrad           m4, 12
+    psrad           m5, 12
 %else
-    vpbroadcastd       m0,             [tab_ChromaCoeff + r4 * 4]
+    psrad           m0, 6
+    psrad           m1, 6
+    psrad           m2, 6
+    psrad           m3, 6
+    psrad           m4, 6
+    psrad           m5, 6
 %endif
+    packssdw        m0, m1
+    packssdw        m2, m3
+    packssdw        m4, m5
 
-    vbroadcasti128     m2,             [pw_1]
-    vbroadcasti128     m5,             [pw_2000]
-    mova               m1,             [tab_Tm]
-
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-
-    mov               r6d,             8/2
-    dec               r0
-    test              r5d,             r5d
-    jz                .loop
-    sub               r0 ,             r1
-    inc               r6d
-
-.loop:
-    ; Row 0
-    vbroadcasti128    m3,              [r0]
-    pshufb            m3,              m1
-    pmaddubsw         m3,              m0
-    pmaddwd           m3,              m2
-
-    ; Row 1
-    vbroadcasti128    m4,              [r0 + r1]
-    pshufb            m4,              m1
-    pmaddubsw         m4,              m0
-    pmaddwd           m4,              m2
-    packssdw          m3,              m4
-    psubw             m3,              m5
-    vpermq            m3,              m3,          11011000b
-    vextracti128      xm4,             m3,          1
-    movq              [r2],            xm3
-    pextrd            [r2 + 8],        xm3,         2
-    movq              [r2 + r3],       xm4
-    pextrd            [r2 + r3 + 8],   xm4,         2
-    lea               r2,              [r2 + r3 * 2]
-    lea               r0,              [r0 + r1 * 2]
-    dec               r6d
-    jnz              .loop
-    test              r5d,             r5d
-    jz               .end
-
-    ;Row 11
-    vbroadcasti128    m3,              [r0]
-    pshufb            m3,              m1
-    pmaddubsw         m3,              m0
-    pmaddwd           m3,              m2
-    packssdw          m3,              m3
-    psubw             m3,              m5
-    vextracti128      xm4,             m3,          1
-    movq              [r2],            xm3
-    movd              [r2+8],          xm4
-.end:
-    RET
-
-INIT_YMM avx2
-cglobal interp_8tap_horiz_ps_12x16, 6, 7, 8
-    mov                         r5d,               r5m
-    mov                         r4d,               r4m
-%ifdef PIC
-    lea                         r6,                [tab_LumaCoeff]
-    vpbroadcastq                m0,                [r6 + r4 * 8]
-%else
-    vpbroadcastq                m0,                [tab_LumaCoeff + r4 * 8]
-%endif
-    mova                        m6,                [tab_Lm + 32]
-    mova                        m1,                [tab_Lm]
-    add                         r3d,               r3d
-    vbroadcasti128              m2,                [pw_2000]
-    mov                         r4d,                16
-    vbroadcasti128              m7,                [pw_1]
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - pw_2000
-
-    mova                        m5,                [interp8_hps_shuf]
-    sub                         r0,                3
-    test                        r5d,               r5d
-    jz                          .loop
-    lea                         r6,                [r1 * 3]                     ; r6 = (N / 2 - 1) * srcStride
-    sub                         r0,                r6                           ; r0(src)-r6
-    add                         r4d,                7
-.loop:
-
-    ; Row 0
-
-    vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb                      m4,                m3,        m6
-    pshufb                      m3,                m1                           ; shuffled based on the col order tab_Lm
-    pmaddubsw                   m3,                m0
-    pmaddubsw                   m4,                m0
-    pmaddwd                     m3,                m7
-    pmaddwd                     m4,                m7
-    packssdw                    m3,                m4
-
-    vbroadcasti128              m4,                [r0 + 8]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb                      m4,                m1
-    pmaddubsw                   m4,                m0
-    pmaddwd                     m4,                m7
-    packssdw                    m4,                m4
-
-    pmaddwd                     m3,                m7
-    pmaddwd                     m4,                m7
-    packssdw                    m3,                m4
-
-    vpermd                      m3,                m5,               m3
-    psubw                       m3,                m2
-
-    vextracti128                xm4,               m3,               1
-    movu                        [r2],              xm3                          ;row 0
-    movq                        [r2 + 16],         xm4                          ;row 1
-
-    add                         r0,                r1
-    add                         r2,                r3
-    dec                         r4d
-    jnz                         .loop
-    RET
-
-INIT_YMM avx2
-cglobal interp_8tap_horiz_ps_24x32, 4, 7, 8
-    mov                         r5d,               r5m
-    mov                         r4d,               r4m
-%ifdef PIC
-    lea                         r6,                [tab_LumaCoeff]
-    vpbroadcastq                m0,                [r6 + r4 * 8]
-%else
-    vpbroadcastq                m0,                [tab_LumaCoeff + r4 * 8]
-%endif
-    mova                        m6,                [tab_Lm + 32]
-    mova                        m1,                [tab_Lm]
-    mov                         r4d,               32                           ;height
-    add                         r3d,               r3d
-    vbroadcasti128              m2,                [pw_2000]
-    vbroadcasti128              m7,                [pw_1]
-
-    ; register map
-    ; m0      - interpolate coeff
-    ; m1 , m6 - shuffle order table
-    ; m2      - pw_2000
-
-    sub                         r0,                3
-    test                        r5d,               r5d
-    jz                          .label
-    lea                         r6,                [r1 * 3]                     ; r6 = (N / 2 - 1) * srcStride
-    sub                         r0,                r6                           ; r0(src)-r6
-    add                         r4d,               7                            ; blkheight += N - 1  (7 - 1 = 6 ; since the last one row not in loop)
-
-.label:
-    lea                         r6,                [interp8_hps_shuf]
-.loop:
-    ; Row 0
-    vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb                      m4,                m3,             m6           ; row 0 (col 4 to 7)
-    pshufb                      m3,                m1                           ; shuffled based on the col order tab_Lm row 0 (col 0 to 3)
-    pmaddubsw                   m3,                m0
-    pmaddubsw                   m4,                m0
-    pmaddwd                     m3,                m7
-    pmaddwd                     m4,                m7
-    packssdw                    m3,                m4
-
-    vbroadcasti128              m4,                [r0 + 8]                     ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb                      m5,                m4,            m6            ;row 1 (col 4 to 7)
-    pshufb                      m4,                m1                           ;row 1 (col 0 to 3)
-    pmaddubsw                   m4,                m0
-    pmaddubsw                   m5,                m0
-    pmaddwd                     m4,                m7
-    pmaddwd                     m5,                m7
-    packssdw                    m4,                m5
-    pmaddwd                     m3,                m7
-    pmaddwd                     m4,                m7
-    packssdw                    m3,                m4
-    mova                        m5,                [r6]
-    vpermd                      m3,                m5,               m3
-    psubw                       m3,                m2
-    movu                        [r2],              m3                          ;row 0
-
-    vbroadcasti128              m3,                [r0 + 16]
-    pshufb                      m4,                m3,          m6
-    pshufb                      m3,                m1
-    pmaddubsw                   m3,                m0
-    pmaddubsw                   m4,                m0
-    pmaddwd                     m3,                m7
-    pmaddwd                     m4,                m7
-    packssdw                    m3,                m4
-    pmaddwd                     m3,                m7
-    pmaddwd                     m4,                m7
-    packssdw                    m3,                m4
-    mova                        m4,                [r6]
-    vpermd                      m3,                m4,            m3
-    psubw                       m3,                m2
-    movu                        [r2 + 32],         xm3                          ;row 0
-
-    add                         r0,                r1
-    add                         r2,                r3
-    dec                         r4d
-    jnz                         .loop
-    RET
-
-;-----------------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_ps_24x32(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;-----------------------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-cglobal interp_4tap_horiz_ps_24x32, 4,7,6
-    mov                r4d,            r4m
-    mov                r5d,            r5m
-    add                r3d,            r3d
-%ifdef PIC
-    lea                r6,             [tab_ChromaCoeff]
-    vpbroadcastd       m0,             [r6 + r4 * 4]
-%else
-    vpbroadcastd       m0,             [tab_ChromaCoeff + r4 * 4]
-%endif
-    vbroadcasti128     m2,             [pw_1]
-    vbroadcasti128     m5,             [pw_2000]
-    mova               m1,             [tab_Tm]
-
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-    mov                r6d,            32
-    dec                r0
-    test               r5d,            r5d
-    je                 .loop
-    sub                r0 ,            r1
-    add                r6d ,           3
-
-.loop:
-    ; Row 0
-    vbroadcasti128     m3,             [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb             m3,             m1
-    pmaddubsw          m3,             m0
-    pmaddwd            m3,             m2
-    vbroadcasti128     m4,             [r0 + 8]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb             m4,             m1
-    pmaddubsw          m4,             m0
-    pmaddwd            m4,             m2
-    packssdw           m3,             m4
-    psubw              m3,             m5
-    vpermq             m3,             m3,          11011000b
-    movu               [r2],           m3
-
-    vbroadcasti128     m3,             [r0 + 16]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb             m3,             m1
-    pmaddubsw          m3,             m0
-    pmaddwd            m3,             m2
-    packssdw           m3,             m3
-    psubw              m3,             m5
-    vpermq             m3,             m3,          11011000b
-    movu               [r2 + 32],      xm3
-
-    add                r2,             r3
-    add                r0,             r1
-    dec                r6d
-    jnz                .loop
-    RET
-
-;-----------------------------------------------------------------------------------------------------------------------
-;macro FILTER_H8_W8_16N_AVX2
-;-----------------------------------------------------------------------------------------------------------------------
-%macro  FILTER_H8_W8_16N_AVX2 0
-    vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb                      m4,                m3,             m6           ; row 0 (col 4 to 7)
-    pshufb                      m3,                m1                           ; shuffled based on the col order tab_Lm row 0 (col 0 to 3)
-    pmaddubsw                   m3,                m0
-    pmaddubsw                   m4,                m0
-    pmaddwd                     m3,                m2
-    pmaddwd                     m4,                m2
-    packssdw                    m3,                m4                         ; DWORD [R1D R1C R0D R0C R1B R1A R0B R0A]
-
-    vbroadcasti128              m4,                [r0 + 8]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb                      m5,                m4,            m6            ;row 1 (col 4 to 7)
-    pshufb                      m4,                m1                           ;row 1 (col 0 to 3)
-    pmaddubsw                   m4,                m0
-    pmaddubsw                   m5,                m0
-    pmaddwd                     m4,                m2
-    pmaddwd                     m5,                m2
-    packssdw                    m4,                m5                         ; DWORD [R3D R3C R2D R2C R3B R3A R2B R2A]
-
-    pmaddwd                     m3,                m2
-    pmaddwd                     m4,                m2
-    packssdw                    m3,                m4                         ; all rows and col completed.
-
-    mova                        m5,                [interp8_hps_shuf]
-    vpermd                      m3,                m5,               m3
-    psubw                       m3,                m8
-
-    vextracti128                xm4,               m3,               1
-    mova                        [r4],              xm3
-    mova                        [r4 + 16],         xm4
-    %endmacro
-
-;-----------------------------------------------------------------------------
-; void interp_8tap_hv_pp_16x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int idxX, int idxY)
-;-----------------------------------------------------------------------------
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_hv_pp_16x16, 4, 10, 15, 0-31*32
-%define stk_buf1    rsp
-    mov                         r4d,               r4m
-    mov                         r5d,               r5m
-%ifdef PIC
-    lea                         r6,                [tab_LumaCoeff]
-    vpbroadcastq                m0,                [r6 + r4 * 8]
+%ifidn %1,sp
+    packuswb        m0, m2
+    mova            m5, [interp8_hps_shuf]
+    vpermd          m0, m5, m0
+    vextracti128    xm2, m0, 1
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r6], xm2
 %else
-    vpbroadcastq                m0,                [tab_LumaCoeff + r4 * 8]
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m2, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
 %endif
 
-    xor                         r6,                 r6
-    mov                         r4,                 rsp
-    mova                        m6,                [tab_Lm + 32]
-    mova                        m1,                [tab_Lm]
-    mov                         r8,                16                           ;height
-    vbroadcasti128              m8,                [pw_2000]
-    vbroadcasti128              m2,                [pw_1]
-    sub                         r0,                3
-    lea                         r7,                [r1 * 3]                     ; r7 = (N / 2 - 1) * srcStride
-    sub                         r0,                r7                           ; r0(src)-r7
-    add                         r8,                7
-
-.loopH:
-    FILTER_H8_W8_16N_AVX2
-    add                         r0,                r1
-    add                         r4,                32
-    inc                         r6
-    cmp                         r6,                16+7
-    jnz                        .loopH
-
-; vertical phase
-    xor                         r6,                r6
-    xor                         r1,                r1
-.loopV:
-
-;load necessary variables
-    mov                         r4d,               r5d          ;coeff here for vertical is r5m
-    shl                         r4d,               7
-    mov                         r1d,               16
-    add                         r1d,               r1d
-
- ; load intermedia buffer
-    mov                         r0,                stk_buf1
-
-    ; register mapping
-    ; r0 - src
-    ; r5 - coeff
-    ; r6 - loop_i
+    movu            xm10, [r7 + r1]                 ; m10 = row 13
+    punpckhwd       xm0, xm9, xm10
+    punpcklwd       xm9, xm10
+    vinserti128     m9, m9, xm0, 1
+    pmaddwd         m9, [r5 + 3 * mmsize]
+    paddd           m6, m9
+    movu            xm0, [r7 + r1 * 2]              ; m0 = row 14
+    punpckhwd       xm1, xm10, xm0
+    punpcklwd       xm10, xm0
+    vinserti128     m10, m10, xm1, 1
+    pmaddwd         m10, [r5 + 3 * mmsize]
+    paddd           m7, m10
 
-; load coeff table
-%ifdef PIC
-    lea                          r5,                [pw_LumaCoeffVer]
-    add                          r5,                r4
+%ifidn %1,sp
+    paddd           m6, m11
+    paddd           m7, m11
+    psrad           m6, 12
+    psrad           m7, 12
 %else
-    lea                          r5,                [pw_LumaCoeffVer + r4]
+    psrad           m6, 6
+    psrad           m7, 6
 %endif
+    packssdw        m6, m7
+    lea             r8, [r2 + r3 * 4]
 
-    lea                          r4,                [r1*3]
-    mova                         m14,               [pd_526336]
-    lea                          r6,                [r3 * 3]
-    mov                          r9d,               16 / 8
-
-.loopW:
-    PROCESS_LUMA_AVX2_W8_16R sp
-    add                          r2,                 8
-    add                          r0,                 16
-    dec                          r9d
-    jnz                          .loopW
-    RET
+%ifidn %1,sp
+    packuswb        m4, m6
+    vpermd          m4, m5, m4
+    vextracti128    xm6, m4, 1
+    movq            [r8], xm4
+    movhps          [r8 + r3], xm4
+    movq            [r8 + r3 * 2], xm6
+    movhps          [r8 + r6], xm6
+%else
+    vpermq          m4, m4, 11011000b
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm5, m4, 1
+    vextracti128    xm7, m6, 1
+    movu            [r8], xm4
+    movu            [r8 + r3], xm5
+    movu            [r8 + r3 * 2], xm6
+    movu            [r8 + r6], xm7
 %endif
+%endmacro
 
+%macro FILTER_VER_LUMA_AVX2_Nx8 2
 INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_12x32, 4, 6, 7
-    mov               r4d,          r4m
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_%2x8, 4, 10, 12
+    mov             r4d, r4m
+    shl             r4d, 7
+    add             r1d, r1d
 
 %ifdef PIC
-    lea               r5,           [tab_ChromaCoeff]
-    vpbroadcastd      m0,           [r5 + r4 * 4]
+    lea             r5, [pw_LumaCoeffVer]
+    add             r5, r4
 %else
-    vpbroadcastd      m0,           [tab_ChromaCoeff + r4 * 4]
+    lea             r5, [pw_LumaCoeffVer + r4]
 %endif
 
-    mova              m6,           [pw_512]
-    mova              m1,           [interp4_horiz_shuf1]
-    vpbroadcastd      m2,           [pw_1]
-
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-
-    dec               r0
-    mov               r4d,          16
-
-.loop:
-    ; Row 0
-    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-    vbroadcasti128    m4,           [r0 + 4]                    ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-    packssdw          m3,           m4
-    pmulhrsw          m3,           m6
-
-    ; Row 1
-    vbroadcasti128    m4,           [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-    vbroadcasti128    m5,           [r0 + r1 + 4]               ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m5,           m1
-    pmaddubsw         m5,           m0
-    pmaddwd           m5,           m2
-    packssdw          m4,           m5
-    pmulhrsw          m4,           m6
-
-    packuswb          m3,           m4
-    vpermq            m3,           m3,      11011000b
-
-    vextracti128      xm4,          m3,       1
-    movq              [r2],         xm3
-    pextrd            [r2+8],       xm3,      2
-    movq              [r2 + r3],    xm4
-    pextrd            [r2 + r3 + 8],xm4,      2
-    lea               r2,           [r2 + r3 * 2]
-    lea               r0,           [r0 + r1 * 2]
-    dec               r4d
-    jnz               .loop
-    RET
-
-INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_24x64, 4,6,7
-    mov              r4d,           r4m
-
-%ifdef PIC
-    lea               r5,           [tab_ChromaCoeff]
-    vpbroadcastd      m0,           [r5 + r4 * 4]
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+%ifidn %1,sp
+    mova            m11, [pd_526336]
 %else
-    vpbroadcastd      m0,           [tab_ChromaCoeff + r4 * 4]
+    add             r3d, r3d
 %endif
-
-    mova              m1,           [interp4_horiz_shuf1]
-    vpbroadcastd      m2,           [pw_1]
-    mova              m6,           [pw_512]
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-
-    dec               r0
-    mov               r4d,          64
-
-.loop:
-    ; Row 0
-    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-    vbroadcasti128    m4,           [r0 + 4]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-    packssdw          m3,           m4
-    pmulhrsw          m3,           m6
-
-    vbroadcasti128    m4,           [r0 + 16]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-    vbroadcasti128    m5,           [r0 + 20]
-    pshufb            m5,           m1
-    pmaddubsw         m5,           m0
-    pmaddwd           m5,           m2
-    packssdw          m4,           m5
-    pmulhrsw          m4,           m6
-
-    packuswb          m3,           m4
-    vpermq            m3,           m3,      11011000b
-
-    vextracti128      xm4,          m3,       1
-    movu              [r2],         xm3
-    movq              [r2 + 16],    xm4
-    add               r2,           r3
-    add               r0,           r1
-    dec               r4d
-    jnz               .loop
-    RET
-
-
-INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_2x16, 4, 6, 6
-    mov               r4d,           r4m
-
-%ifdef PIC
-    lea               r5,            [tab_ChromaCoeff]
-    vpbroadcastd      m0,            [r5 + r4 * 4]
-%else
-    vpbroadcastd      m0,            [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    mova              m4,            [interp4_hpp_shuf]
-    mova              m5,            [pw_1]
-    dec               r0
-    lea               r4,            [r1 * 3]
-    movq              xm1,           [r0]
-    movhps            xm1,           [r0 + r1]
-    movq              xm2,           [r0 + r1 * 2]
-    movhps            xm2,           [r0 + r4]
-    vinserti128       m1,            m1,          xm2,          1
-    lea               r0,            [r0 + r1 * 4]
-    movq              xm3,           [r0]
-    movhps            xm3,           [r0 + r1]
-    movq              xm2,           [r0 + r1 * 2]
-    movhps            xm2,           [r0 + r4]
-    vinserti128       m3,            m3,          xm2,          1
-
-    pshufb            m1,            m4
-    pshufb            m3,            m4
-    pmaddubsw         m1,            m0
-    pmaddubsw         m3,            m0
-    pmaddwd           m1,            m5
-    pmaddwd           m3,            m5
-    packssdw          m1,            m3
-    pmulhrsw          m1,            [pw_512]
-    vextracti128      xm2,           m1,          1
-    packuswb          xm1,           xm2
-
-    lea               r4,            [r3 * 3]
-    pextrw            [r2],          xm1,         0
-    pextrw            [r2 + r3],     xm1,         1
-    pextrw            [r2 + r3 * 2], xm1,         4
-    pextrw            [r2 + r4],     xm1,         5
-    lea               r2,            [r2 + r3 * 4]
-    pextrw            [r2],          xm1,         2
-    pextrw            [r2 + r3],     xm1,         3
-    pextrw            [r2 + r3 * 2], xm1,         6
-    pextrw            [r2 + r4],     xm1,         7
-    lea               r2,            [r2 + r3 * 4]
-    lea               r0,            [r0 + r1 * 4]
-
-    lea               r4,            [r1 * 3]
-    movq              xm1,           [r0]
-    movhps            xm1,           [r0 + r1]
-    movq              xm2,           [r0 + r1 * 2]
-    movhps            xm2,           [r0 + r4]
-    vinserti128       m1,            m1,          xm2,          1
-    lea               r0,            [r0 + r1 * 4]
-    movq              xm3,           [r0]
-    movhps            xm3,           [r0 + r1]
-    movq              xm2,           [r0 + r1 * 2]
-    movhps            xm2,           [r0 + r4]
-    vinserti128       m3,            m3,          xm2,          1
-
-    pshufb            m1,            m4
-    pshufb            m3,            m4
-    pmaddubsw         m1,            m0
-    pmaddubsw         m3,            m0
-    pmaddwd           m1,            m5
-    pmaddwd           m3,            m5
-    packssdw          m1,            m3
-    pmulhrsw          m1,            [pw_512]
-    vextracti128      xm2,           m1,          1
-    packuswb          xm1,           xm2
-
-    lea               r4,            [r3 * 3]
-    pextrw            [r2],          xm1,         0
-    pextrw            [r2 + r3],     xm1,         1
-    pextrw            [r2 + r3 * 2], xm1,         4
-    pextrw            [r2 + r4],     xm1,         5
-    lea               r2,            [r2 + r3 * 4]
-    pextrw            [r2],          xm1,         2
-    pextrw            [r2 + r3],     xm1,         3
-    pextrw            [r2 + r3 * 2], xm1,         6
-    pextrw            [r2 + r4],     xm1,         7
-    RET
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp_64xN(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-%macro IPFILTER_CHROMA_PP_64xN_AVX2 1
-INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_64x%1, 4,6,7
-    mov             r4d, r4m
-
-%ifdef PIC
-    lea               r5,           [tab_ChromaCoeff]
-    vpbroadcastd      m0,           [r5 + r4 * 4]
+    lea             r6, [r3 * 3]
+    mov             r9d, %2 / 8
+.loopW:
+    PROCESS_LUMA_S_AVX2_W8_8R %1
+%ifidn %1,sp
+    add             r2, 8
 %else
-    vpbroadcastd      m0,           [tab_ChromaCoeff + r4 * 4]
+    add             r2, 16
 %endif
-
-    mova              m1,           [interp4_horiz_shuf1]
-    vpbroadcastd      m2,           [pw_1]
-    mova              m6,           [pw_512]
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-
-    dec               r0
-    mov               r4d,          %1
-
-.loop:
-    ; Row 0
-    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-    vbroadcasti128    m4,           [r0 + 4]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-    packssdw          m3,           m4
-    pmulhrsw          m3,           m6
-
-    vbroadcasti128    m4,           [r0 + 16]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-    vbroadcasti128    m5,           [r0 + 20]
-    pshufb            m5,           m1
-    pmaddubsw         m5,           m0
-    pmaddwd           m5,           m2
-    packssdw          m4,           m5
-    pmulhrsw          m4,           m6
-    packuswb          m3,           m4
-    vpermq            m3,           m3,      11011000b
-    movu              [r2],         m3
-
-    vbroadcasti128    m3,           [r0 + 32]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-    vbroadcasti128    m4,           [r0 + 36]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-    packssdw          m3,           m4
-    pmulhrsw          m3,           m6
-
-    vbroadcasti128    m4,           [r0 + 48]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-    vbroadcasti128    m5,           [r0 + 52]
-    pshufb            m5,           m1
-    pmaddubsw         m5,           m0
-    pmaddwd           m5,           m2
-    packssdw          m4,           m5
-    pmulhrsw          m4,           m6
-    packuswb          m3,           m4
-    vpermq            m3,           m3,      11011000b
-    movu              [r2 + 32],         m3
-
-    add               r2,           r3
-    add               r0,           r1
-    dec               r4d
-    jnz               .loop
+    add             r0, 16
+    dec             r9d
+    jnz             .loopW
     RET
+%endif
 %endmacro
 
-    IPFILTER_CHROMA_PP_64xN_AVX2  64
-    IPFILTER_CHROMA_PP_64xN_AVX2  32
-    IPFILTER_CHROMA_PP_64xN_AVX2  48
-    IPFILTER_CHROMA_PP_64xN_AVX2  16
+    FILTER_VER_LUMA_AVX2_Nx8 sp, 32
+    FILTER_VER_LUMA_AVX2_Nx8 sp, 16
+    FILTER_VER_LUMA_AVX2_Nx8 ss, 32
+    FILTER_VER_LUMA_AVX2_Nx8 ss, 16
 
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp_48x64(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_LUMA_S_AVX2_32x24 1
 INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_48x64, 4,6,7
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_32x24, 4, 10, 15
     mov             r4d, r4m
+    shl             r4d, 7
+    add             r1d, r1d
 
 %ifdef PIC
-    lea               r5,            [tab_ChromaCoeff]
-    vpbroadcastd      m0,            [r5 + r4 * 4]
+    lea             r5, [pw_LumaCoeffVer]
+    add             r5, r4
 %else
-    vpbroadcastd      m0,            [tab_ChromaCoeff + r4 * 4]
+    lea             r5, [pw_LumaCoeffVer + r4]
 %endif
 
-    mova              m1,            [interp4_horiz_shuf1]
-    vpbroadcastd      m2,            [pw_1]
-    mova              m6,            [pw_512]
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-
-    dec               r0
-    mov               r4d,           64
-
-.loop:
-    ; Row 0
-    vbroadcasti128    m3,            [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,            m1
-    pmaddubsw         m3,            m0
-    pmaddwd           m3,            m2
-    vbroadcasti128    m4,            [r0 + 4]
-    pshufb            m4,            m1
-    pmaddubsw         m4,            m0
-    pmaddwd           m4,            m2
-    packssdw          m3,            m4
-    pmulhrsw          m3,            m6
-
-    vbroadcasti128    m4,            [r0 + 16]
-    pshufb            m4,            m1
-    pmaddubsw         m4,            m0
-    pmaddwd           m4,            m2
-    vbroadcasti128    m5,            [r0 + 20]
-    pshufb            m5,            m1
-    pmaddubsw         m5,            m0
-    pmaddwd           m5,            m2
-    packssdw          m4,            m5
-    pmulhrsw          m4,            m6
-
-    packuswb          m3,            m4
-    vpermq            m3,            m3,      q3120
-
-    movu              [r2],          m3
-
-    vbroadcasti128    m3,            [r0 + mmsize]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,            m1
-    pmaddubsw         m3,            m0
-    pmaddwd           m3,            m2
-    vbroadcasti128    m4,            [r0 + mmsize + 4]
-    pshufb            m4,            m1
-    pmaddubsw         m4,            m0
-    pmaddwd           m4,            m2
-    packssdw          m3,            m4
-    pmulhrsw          m3,            m6
-
-    vbroadcasti128    m4,            [r0 + mmsize + 16]
-    pshufb            m4,            m1
-    pmaddubsw         m4,            m0
-    pmaddwd           m4,            m2
-    vbroadcasti128    m5,            [r0 + mmsize + 20]
-    pshufb            m5,            m1
-    pmaddubsw         m5,            m0
-    pmaddwd           m5,            m2
-    packssdw          m4,            m5
-    pmulhrsw          m4,            m6
-
-    packuswb          m3,            m4
-    vpermq            m3,            m3,      q3120
-    movu              [r2 + mmsize], xm3
-
-    add               r2,            r3
-    add               r0,            r1
-    dec               r4d
-    jnz               .loop
-    RET
-
-;-----------------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_ps_48x64(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;-----------------------------------------------------------------------------------------------------------------------------;
-
-INIT_YMM avx2
-cglobal interp_4tap_horiz_ps_48x64, 4,7,6
-    mov             r4d, r4m
-    mov             r5d, r5m
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+%ifidn %1,sp
+    mova            m14, [pd_526336]
+%else
     add             r3d, r3d
-
-%ifdef PIC
-    lea               r6,           [tab_ChromaCoeff]
-    vpbroadcastd      m0,           [r6 + r4 * 4]
+%endif
+    lea             r6, [r3 * 3]
+    mov             r9d, 4
+.loopW:
+    PROCESS_LUMA_AVX2_W8_16R %1
+%ifidn %1,sp
+    add             r2, 8
 %else
-    vpbroadcastd      m0,           [tab_ChromaCoeff + r4 * 4]
+    add             r2, 16
 %endif
-
-    vbroadcasti128     m2,          [pw_1]
-    vbroadcasti128     m5,          [pw_2000]
-    mova               m1,          [tab_Tm]
-
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-    mov               r6d,          64
-    dec               r0
-    test              r5d,          r5d
-    je                .loop
-    sub               r0 ,          r1
-    add               r6d ,         3
-
-.loop:
-    ; Row 0
-    vbroadcasti128    m3,           [r0]                           ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-    vbroadcasti128    m4,           [r0 + 8]                       ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-
-    packssdw          m3,           m4
-    psubw             m3,           m5
-    vpermq            m3,           m3,          q3120
-    movu              [r2],         m3
-
-    vbroadcasti128    m3,           [r0 + 16]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-    vbroadcasti128    m4,           [r0 + 24]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-
-    packssdw          m3,           m4
-    psubw             m3,           m5
-    vpermq            m3,           m3,          q3120
-    movu              [r2 + 32],    m3
-
-    vbroadcasti128    m3,           [r0 + 32]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,           m1
-    pmaddubsw         m3,           m0
-    pmaddwd           m3,           m2
-    vbroadcasti128    m4,           [r0 + 40]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,           m1
-    pmaddubsw         m4,           m0
-    pmaddwd           m4,           m2
-
-    packssdw          m3,           m4
-    psubw             m3,           m5
-    vpermq            m3,           m3,          q3120
-    movu              [r2 + 64],    m3
-
-    add               r2,          r3
-    add               r0,          r1
-    dec               r6d
-    jnz               .loop
-    RET
-
-;-----------------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_ps_24x64(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;-----------------------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-cglobal interp_4tap_horiz_ps_24x64, 4,7,6
-    mov                r4d,            r4m
-    mov                r5d,            r5m
-    add                r3d,            r3d
-%ifdef PIC
-    lea                r6,             [tab_ChromaCoeff]
-    vpbroadcastd       m0,             [r6 + r4 * 4]
-%else
-    vpbroadcastd       m0,             [tab_ChromaCoeff + r4 * 4]
-%endif
-    vbroadcasti128     m2,             [pw_1]
-    vbroadcasti128     m5,             [pw_2000]
-    mova               m1,             [tab_Tm]
-
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-    mov                r6d,            64
-    dec                r0
-    test               r5d,            r5d
-    je                 .loop
-    sub                r0 ,            r1
-    add                r6d ,           3
-
+    add             r0, 16
+    dec             r9d
+    jnz             .loopW
+    lea             r9, [r1 * 4]
+    sub             r7, r9
+    lea             r0, [r7 - 48]
+%ifidn %1,sp
+    lea             r2, [r8 + r3 * 4 - 24]
+%else
+    lea             r2, [r8 + r3 * 4 - 48]
+%endif
+    mova            m11, m14
+    mov             r9d, 4
 .loop:
-    ; Row 0
-    vbroadcasti128     m3,             [r0]                          ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb             m3,             m1
-    pmaddubsw          m3,             m0
-    pmaddwd            m3,             m2
-    vbroadcasti128     m4,             [r0 + 8]                      ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb             m4,             m1
-    pmaddubsw          m4,             m0
-    pmaddwd            m4,             m2
-    packssdw           m3,             m4
-    psubw              m3,             m5
-    vpermq             m3,             m3,          q3120
-    movu               [r2],           m3
-
-    vbroadcasti128     m3,             [r0 + 16]                     ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb             m3,             m1
-    pmaddubsw          m3,             m0
-    pmaddwd            m3,             m2
-    packssdw           m3,             m3
-    psubw              m3,             m5
-    vpermq             m3,             m3,          q3120
-    movu               [r2 + 32],      xm3
-
-    add                r2,             r3
-    add                r0,             r1
-    dec                r6d
-    jnz                .loop
-    RET
-
-INIT_YMM avx2
-cglobal interp_4tap_horiz_ps_2x16, 4, 7, 7
-    mov               r4d,           r4m
-    mov               r5d,           r5m
-    add               r3d,           r3d
-
-%ifdef PIC
-    lea               r6,            [tab_ChromaCoeff]
-    vpbroadcastd      m0,            [r6 + r4 * 4]
-%else
-    vpbroadcastd      m0,            [tab_ChromaCoeff + r4 * 4]
-%endif
-    vbroadcasti128    m6,            [pw_2000]
-    test              r5d,            r5d
-    jz                .label
-    sub               r0,             r1
-
-.label:
-    mova              m4,            [interp4_hps_shuf]
-    mova              m5,            [pw_1]
-    dec               r0
-    lea               r4,            [r1 * 3]
-    movq              xm1,           [r0]                                   ;row 0
-    movhps            xm1,           [r0 + r1]
-    movq              xm2,           [r0 + r1 * 2]
-    movhps            xm2,           [r0 + r4]
-    vinserti128       m1,            m1,           xm2,          1
-    lea               r0,            [r0 + r1 * 4]
-    movq              xm3,           [r0]
-    movhps            xm3,           [r0 + r1]
-    movq              xm2,           [r0 + r1 * 2]
-    movhps            xm2,           [r0 + r4]
-    vinserti128       m3,            m3,           xm2,          1
-
-    pshufb            m1,            m4
-    pshufb            m3,            m4
-    pmaddubsw         m1,            m0
-    pmaddubsw         m3,            m0
-    pmaddwd           m1,            m5
-    pmaddwd           m3,            m5
-    packssdw          m1,            m3
-    psubw             m1,            m6
-
-    lea               r4,            [r3 * 3]
-    vextracti128      xm2,           m1,           1
-
-    movd              [r2],          xm1
-    pextrd            [r2 + r3],     xm1,          1
-    movd              [r2 + r3 * 2], xm2
-    pextrd            [r2 + r4],     xm2,          1
-    lea               r2,            [r2 + r3 * 4]
-    pextrd            [r2],          xm1,          2
-    pextrd            [r2 + r3],     xm1,          3
-    pextrd            [r2 + r3 * 2], xm2,          2
-    pextrd            [r2 + r4],     xm2,          3
-
-    lea               r0,            [r0 + r1 * 4]
-    lea               r2,            [r2 + r3 * 4]
-    lea               r4,            [r1 * 3]
-    movq              xm1,           [r0]
-    movhps            xm1,           [r0 + r1]
-    movq              xm2,           [r0 + r1 * 2]
-    movhps            xm2,           [r0 + r4]
-    vinserti128       m1,            m1,          xm2,           1
-    lea               r0,            [r0 + r1 * 4]
-    movq              xm3,           [r0]
-    movhps            xm3,           [r0 + r1]
-    movq              xm2,           [r0 + r1 * 2]
-    movhps            xm2,           [r0 + r4]
-    vinserti128       m3,            m3,          xm2,           1
-
-    pshufb            m1,            m4
-    pshufb            m3,            m4
-    pmaddubsw         m1,            m0
-    pmaddubsw         m3,            m0
-    pmaddwd           m1,            m5
-    pmaddwd           m3,            m5
-    packssdw          m1,            m3
-    psubw             m1,            m6
-
-    lea               r4,            [r3 * 3]
-    vextracti128      xm2,           m1,           1
-
-    movd              [r2],          xm1
-    pextrd            [r2 + r3],     xm1,          1
-    movd              [r2 + r3 * 2], xm2
-    pextrd            [r2 + r4],     xm2,          1
-    lea               r2,            [r2 + r3 * 4]
-    pextrd            [r2],          xm1,          2
-    pextrd            [r2 + r3],     xm1,          3
-    pextrd            [r2 + r3 * 2], xm2,          2
-    pextrd            [r2 + r4],     xm2,          3
-
-    test              r5d,            r5d
-    jz                .end
-
-    lea               r0,            [r0 + r1 * 4]
-    lea               r2,            [r2 + r3 * 4]
-    movq              xm1,           [r0]
-    movhps            xm1,           [r0 + r1]
-    movq              xm2,           [r0 + r1 * 2]
-    vinserti128       m1,            m1,          xm2,           1
-    pshufb            m1,            m4
-    pmaddubsw         m1,            m0
-    pmaddwd           m1,            m5
-    packssdw          m1,            m1
-    psubw             m1,            m6
-    vextracti128      xm2,           m1,           1
-
-    movd              [r2],          xm1
-    pextrd            [r2 + r3],     xm1,          1
-    movd              [r2 + r3 * 2], xm2
-.end:
-    RET
-
-INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_6x16, 4, 6, 7
-    mov               r4d,               r4m
-
-%ifdef PIC
-    lea               r5,                [tab_ChromaCoeff]
-    vpbroadcastd      m0,                [r5 + r4 * 4]
+    PROCESS_LUMA_S_AVX2_W8_8R %1
+%ifidn %1,sp
+    add             r2, 8
 %else
-    vpbroadcastd      m0,                [tab_ChromaCoeff + r4 * 4]
+    add             r2, 16
 %endif
-
-    mova              m1,                [tab_Tm]
-    mova              m2,                [pw_1]
-    mova              m6,                [pw_512]
-    lea               r4,                [r1 * 3]
-    lea               r5,                [r3 * 3]
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - constant word 1
-
-    dec               r0
-%rep 4
-    ; Row 0
-    vbroadcasti128    m3,                [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m3,                m1
-    pmaddubsw         m3,                m0
-    pmaddwd           m3,                m2
-
-    ; Row 1
-    vbroadcasti128    m4,                [r0 + r1]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,                m1
-    pmaddubsw         m4,                m0
-    pmaddwd           m4,                m2
-    packssdw          m3,                m4
-    pmulhrsw          m3,                m6
-
-    ; Row 2
-    vbroadcasti128    m4,                [r0 + r1 * 2]               ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m4,                m1
-    pmaddubsw         m4,                m0
-    pmaddwd           m4,                m2
-
-    ; Row 3
-    vbroadcasti128    m5,                [r0 + r4]                   ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    pshufb            m5,                m1
-    pmaddubsw         m5,                m0
-    pmaddwd           m5,                m2
-    packssdw          m4,                m5
-    pmulhrsw          m4,                m6
-
-    packuswb          m3,                m4
-    vextracti128      xm4,               m3,          1
-    movd              [r2],              xm3
-    pextrw            [r2 + 4],          xm4,         0
-    pextrd            [r2 + r3],         xm3,         1
-    pextrw            [r2 + r3 + 4],     xm4,         2
-    pextrd            [r2 + r3 * 2],     xm3,         2
-    pextrw            [r2 + r3 * 2 + 4], xm4,         4
-    pextrd            [r2 + r5],         xm3,         3
-    pextrw            [r2 + r5 + 4],     xm4,         6
-    lea               r2,                [r2 + r3 * 4]
-    lea               r0,                [r0 + r1 * 4]
-%endrep
+    add             r0, 16
+    dec             r9d
+    jnz             .loop
     RET
+%endif
+%endmacro
+
+    FILTER_VER_LUMA_S_AVX2_32x24 sp
+    FILTER_VER_LUMA_S_AVX2_32x24 ss

From 0e13e211e62c12a67a3c3b2365b555742ff012fc Mon Sep 17 00:00:00 2001
From: Aruna Matheswaran <aruna@multicorewareinc.com>
Date: Sat, 27 Jan 2018 18:45:21 +0530
Subject: [PATCH 46/51] rc: fix inconsistency in pass 2 when weightp and cutree
 are enabled.

---
 source/encoder/slicetype.cpp | 7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)

diff --git a/source/encoder/slicetype.cpp b/source/encoder/slicetype.cpp
index fe2e51aab9..544e6c645f 100644
--- a/source/encoder/slicetype.cpp
+++ b/source/encoder/slicetype.cpp
@@ -154,7 +154,7 @@ void LookaheadTLD::calcAdaptiveQuantFrame(Frame *curFrame, x265_param* param)
     int blockXY = 0;
     int blockX = 0, blockY = 0;
     double strength = 0.f;
-    if (param->rc.aqMode == X265_AQ_NONE || param->rc.aqStrength == 0)
+    if ((param->rc.aqMode == X265_AQ_NONE || param->rc.aqStrength == 0) || (param->rc.bStatRead && param->rc.cuTree && IS_REFERENCED(curFrame)))
     {
         /* Need to init it anyways for CU tree */
         int cuCount = blockCount;
@@ -982,11 +982,8 @@ void PreLookaheadGroup::processTasks(int workerThreadID)
         ProfileLookaheadTime(m_lookahead.m_preLookaheadElapsedTime, m_lookahead.m_countPreLookahead);
         ProfileScopeEvent(prelookahead);
         m_lock.release();
-
         preFrame->m_lowres.init(preFrame->m_fencPic, preFrame->m_poc);
-        if (m_lookahead.m_param->rc.bStatRead && m_lookahead.m_param->rc.cuTree && IS_REFERENCED(preFrame))
-            /* cu-tree offsets were read from stats file */;
-        else if (m_lookahead.m_bAdaptiveQuant)
+        if (m_lookahead.m_bAdaptiveQuant)
             tld.calcAdaptiveQuantFrame(preFrame, m_lookahead.m_param);
         tld.lowresIntraEstimate(preFrame->m_lowres, m_lookahead.m_param->rc.qgSize);
         preFrame->m_lowresInit = true;

From c56fd20ff467e29d4b3e4e2a97cb854a469054d0 Mon Sep 17 00:00:00 2001
From: Radhakrishnan <radhakrishnan@multicorewareinc.com>
Date: Wed, 31 Jan 2018 15:13:30 +0530
Subject: [PATCH 47/51] x86: Split ipfilter8 kernals part3

This patch adds infra and moves the vertical 4tap ipfilter kernals from ipfilter8.asm to new source file v4-ipfilter8.asm to reduce compile time of x265.
---
 source/common/CMakeLists.txt       |     2 +-
 source/common/x86/ipfilter8.asm    | 22286 ++++++---------------------
 source/common/x86/v4-ipfilter8.asm | 12799 +++++++++++++++
 3 files changed, 17566 insertions(+), 17521 deletions(-)
 create mode 100644 source/common/x86/v4-ipfilter8.asm

diff --git a/source/common/CMakeLists.txt b/source/common/CMakeLists.txt
index ab980a223e..8261fac0c3 100644
--- a/source/common/CMakeLists.txt
+++ b/source/common/CMakeLists.txt
@@ -63,7 +63,7 @@ if(ENABLE_ASSEMBLY AND X86)
     if(HIGH_BIT_DEPTH)
         set(A_SRCS ${A_SRCS} sad16-a.asm intrapred16.asm ipfilter16.asm loopfilter.asm)
     else()
-        set(A_SRCS ${A_SRCS} sad-a.asm intrapred8.asm intrapred8_allangs.asm h-ipfilter8.asm ipfilter8.asm loopfilter.asm)
+        set(A_SRCS ${A_SRCS} sad-a.asm intrapred8.asm intrapred8_allangs.asm v4-ipfilter8.asm h-ipfilter8.asm ipfilter8.asm loopfilter.asm)
     endif()
     if(NOT X64)
         set(A_SRCS ${A_SRCS} pixel-32.asm)
diff --git a/source/common/x86/ipfilter8.asm b/source/common/x86/ipfilter8.asm
index 223917b0f3..28ef118ee3 100644
--- a/source/common/x86/ipfilter8.asm
+++ b/source/common/x86/ipfilter8.asm
@@ -33,9 +33,6 @@ const tab_Tm,    db 0, 1, 2, 3, 1, 2, 3, 4, 2, 3, 4, 5, 3, 4, 5, 6
 
 const interp4_vpp_shuf, times 2 db 0, 4, 1, 5, 2, 6, 3, 7, 8, 12, 9, 13, 10, 14, 11, 15
 
-const interp_vert_shuf, times 2 db 0, 2, 1, 3, 2, 4, 3, 5, 4, 6, 5, 7, 6, 8, 7, 9
-                        times 2 db 4, 6, 5, 7, 6, 8, 7, 9, 8, 10, 9, 11, 10, 12, 11, 13
-
 const interp4_vpp_shuf1, dd 0, 1, 1, 2, 2, 3, 3, 4
                          dd 2, 3, 3, 4, 4, 5, 5, 6
 
@@ -44,103 +41,8 @@ const tab_Lm,    db 0, 1, 2, 3, 4,  5,  6,  7,  1, 2, 3, 4,  5,  6,  7,  8
                  db 4, 5, 6, 7, 8,  9,  10, 11, 5, 6, 7, 8,  9,  10, 11, 12
                  db 6, 7, 8, 9, 10, 11, 12, 13, 7, 8, 9, 10, 11, 12, 13, 14
 
-const tab_Vm,    db 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1
-                 db 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3
-
-const tab_Cm,    db 0, 2, 1, 3, 0, 2, 1, 3, 0, 2, 1, 3, 0, 2, 1, 3
-
 const pd_526336, times 8 dd 8192*64+2048
 
-const tab_ChromaCoeff, db  0, 64,  0,  0
-                       db -2, 58, 10, -2
-                       db -4, 54, 16, -2
-                       db -6, 46, 28, -4
-                       db -4, 36, 36, -4
-                       db -4, 28, 46, -6
-                       db -2, 16, 54, -4
-                       db -2, 10, 58, -2
-
-const tabw_ChromaCoeff, dw  0, 64,  0,  0
-                        dw -2, 58, 10, -2
-                        dw -4, 54, 16, -2
-                        dw -6, 46, 28, -4
-                        dw -4, 36, 36, -4
-                        dw -4, 28, 46, -6
-                        dw -2, 16, 54, -4
-                        dw -2, 10, 58, -2
-
-const tab_ChromaCoeff_V, times 8 db 0, 64
-                         times 8 db 0,  0
-
-                         times 8 db -2, 58
-                         times 8 db 10, -2
-
-                         times 8 db -4, 54
-                         times 8 db 16, -2
-
-                         times 8 db -6, 46
-                         times 8 db 28, -4
-
-                         times 8 db -4, 36
-                         times 8 db 36, -4
-
-                         times 8 db -4, 28
-                         times 8 db 46, -6
-
-                         times 8 db -2, 16
-                         times 8 db 54, -4
-
-                         times 8 db -2, 10
-                         times 8 db 58, -2
-
-const tab_ChromaCoeffV, times 4 dw 0, 64
-                        times 4 dw 0, 0
-
-                        times 4 dw -2, 58
-                        times 4 dw 10, -2
-
-                        times 4 dw -4, 54
-                        times 4 dw 16, -2
-
-                        times 4 dw -6, 46
-                        times 4 dw 28, -4
-
-                        times 4 dw -4, 36
-                        times 4 dw 36, -4
-
-                        times 4 dw -4, 28
-                        times 4 dw 46, -6
-
-                        times 4 dw -2, 16
-                        times 4 dw 54, -4
-
-                        times 4 dw -2, 10
-                        times 4 dw 58, -2
-
-const pw_ChromaCoeffV,  times 8 dw 0, 64
-                        times 8 dw 0, 0
-
-                        times 8 dw -2, 58
-                        times 8 dw 10, -2
-
-                        times 8 dw -4, 54
-                        times 8 dw 16, -2
-
-                        times 8 dw -6, 46
-                        times 8 dw 28, -4
-
-                        times 8 dw -4, 36
-                        times 8 dw 36, -4
-
-                        times 8 dw -4, 28
-                        times 8 dw 46, -6
-
-                        times 8 dw -2, 16
-                        times 8 dw 54, -4
-
-                        times 8 dw -2, 10
-                        times 8 dw 58, -2
-
 const tab_LumaCoeff,   db   0, 0,  0,  64,  0,   0,  0,  0
                        db  -1, 4, -10, 58,  17, -5,  1,  0
                        db  -1, 4, -11, 40,  40, -11, 4, -1
@@ -231,30 +133,6 @@ const tab_LumaCoeffVer_32,  times 16 db 0, 0
                             times 16 db 58, -10
                             times 16 db 4, -1
 
-const tab_ChromaCoeffVer_32,    times 16 db 0, 64
-                                times 16 db 0, 0
-
-                                times 16 db -2, 58
-                                times 16 db 10, -2
-
-                                times 16 db -4, 54
-                                times 16 db 16, -2
-
-                                times 16 db -6, 46
-                                times 16 db 28, -4
-
-                                times 16 db -4, 36
-                                times 16 db 36, -4
-
-                                times 16 db -4, 28
-                                times 16 db 46, -6
-
-                                times 16 db -2, 16
-                                times 16 db 54, -4
-
-                                times 16 db -2, 10
-                                times 16 db 58, -2
-
 const tab_c_64_n64, times 8 db 64, -64
 
 const interp8_hps_shuf,     dd 0, 4, 1, 5, 2, 6, 3, 7
@@ -723,4416 +601,2567 @@ cglobal interp_8tap_vert_%3_%1x%2, 5, 8, 11
     FILTER_VER_LUMA_sse2 64, 64, ps
 %endif
 
-%macro  WORD_TO_DOUBLE 1
-%if ARCH_X86_64
-    punpcklbw   %1,     m8
-%else
-    punpcklbw   %1,     %1
-    psrlw       %1,     8
-%endif
-%endmacro
-
-;-----------------------------------------------------------------------------
-; void interp_4tap_vert_%1_2x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro FILTER_V4_W2_H4_sse2 2
-INIT_XMM sse2
-%if ARCH_X86_64
-cglobal interp_4tap_vert_%1_2x%2, 4, 6, 9
-    pxor        m8,        m8
-%else
-cglobal interp_4tap_vert_%1_2x%2, 4, 6, 8
-%endif
-    mov         r4d,       r4m
-    sub         r0,        r1
-
-%ifidn %1,pp
-    mova        m1,        [pw_32]
-%elifidn %1,ps
-    mova        m1,        [pw_2000]
-    add         r3d,       r3d
-%endif
-
-%ifdef PIC
-    lea         r5,        [tabw_ChromaCoeff]
-    movh        m0,        [r5 + r4 * 8]
-%else
-    movh        m0,        [tabw_ChromaCoeff + r4 * 8]
-%endif
-
-    punpcklqdq  m0,        m0
-    lea         r5,        [3 * r1]
-
-%assign x 1
-%rep %2/4
-    movd        m2,        [r0]
-    movd        m3,        [r0 + r1]
-    movd        m4,        [r0 + 2 * r1]
-    movd        m5,        [r0 + r5]
-
-    punpcklbw   m2,        m3
-    punpcklbw   m6,        m4,        m5
-    punpcklwd   m2,        m6
-
-    WORD_TO_DOUBLE         m2
-    pmaddwd     m2,        m0
-
-    lea         r0,        [r0 + 4 * r1]
-    movd        m6,        [r0]
-
-    punpcklbw   m3,        m4
-    punpcklbw   m7,        m5,        m6
-    punpcklwd   m3,        m7
-
-    WORD_TO_DOUBLE         m3
-    pmaddwd     m3,        m0
-
-    packssdw    m2,        m3
-    pshuflw     m3,        m2,          q2301
-    pshufhw     m3,        m3,          q2301
-    paddw       m2,        m3
-
-    movd        m7,        [r0 + r1]
-
-    punpcklbw   m4,        m5
-    punpcklbw   m3,        m6,        m7
-    punpcklwd   m4,        m3
-
-    WORD_TO_DOUBLE         m4
-    pmaddwd     m4,        m0
-
-    movd        m3,        [r0 + 2 * r1]
-
-    punpcklbw   m5,        m6
-    punpcklbw   m7,        m3
-    punpcklwd   m5,        m7
-
-    WORD_TO_DOUBLE         m5
-    pmaddwd     m5,        m0
-
-    packssdw    m4,        m5
-    pshuflw     m5,        m4,          q2301
-    pshufhw     m5,        m5,          q2301
-    paddw       m4,        m5
-
-%ifidn %1,pp
-    psrld       m2,        16
-    psrld       m4,        16
-    packssdw    m2,        m4
-    paddw       m2,        m1
-    psraw       m2,        6
-    packuswb    m2,        m2
-
-%if ARCH_X86_64
-    movq        r4,        m2
-    mov         [r2],      r4w
-    shr         r4,        16
-    mov         [r2 + r3], r4w
-    lea         r2,        [r2 + 2 * r3]
-    shr         r4,        16
-    mov         [r2],      r4w
-    shr         r4,        16
-    mov         [r2 + r3], r4w
-%else
-    movd        r4,        m2
-    mov         [r2],      r4w
-    shr         r4,        16
-    mov         [r2 + r3], r4w
-    lea         r2,        [r2 + 2 * r3]
-    psrldq      m2,        4
-    movd        r4,        m2
-    mov         [r2],      r4w
-    shr         r4,        16
-    mov         [r2 + r3], r4w
-%endif
-%elifidn %1,ps
-    psrldq      m2,        2
-    psrldq      m4,        2
-    pshufd      m2,        m2, q3120
-    pshufd      m4,        m4, q3120
-    psubw       m4,        m1
-    psubw       m2,        m1
-
-    movd        [r2],      m2
-    psrldq      m2,        4
-    movd        [r2 + r3], m2
-    lea         r2,        [r2 + 2 * r3]
-    movd        [r2],      m4
-    psrldq      m4,        4
-    movd        [r2 + r3], m4
-%endif
-
-%if x < %2/4
-    lea         r2,        [r2 + 2 * r3]
-%endif
-%assign x x+1
-%endrep
-    RET
+%macro FILTER_P2S_2_4_sse2 1
+    movd        m2,     [r0 + %1]
+    movd        m3,     [r0 + r1 + %1]
+    punpcklwd   m2,     m3
+    movd        m3,     [r0 + r1 * 2 + %1]
+    movd        m4,     [r0 + r4 + %1]
+    punpcklwd   m3,     m4
+    punpckldq   m2,     m3
+    punpcklbw   m2,     m0
+    psllw       m2,     6
+    psubw       m2,     m1
 
+    movd        [r2 + r3 * 0 + %1 * 2], m2
+    psrldq      m2,     4
+    movd        [r2 + r3 * 1 + %1 * 2], m2
+    psrldq      m2,     4
+    movd        [r2 + r3 * 2 + %1 * 2], m2
+    psrldq      m2,     4
+    movd        [r2 + r5 + %1 * 2], m2
 %endmacro
 
-    FILTER_V4_W2_H4_sse2 pp, 4
-    FILTER_V4_W2_H4_sse2 pp, 8
-    FILTER_V4_W2_H4_sse2 pp, 16
-
-    FILTER_V4_W2_H4_sse2 ps, 4
-    FILTER_V4_W2_H4_sse2 ps, 8
-    FILTER_V4_W2_H4_sse2 ps, 16
-
-;-----------------------------------------------------------------------------
-; void interp_4tap_vert_%1_4x2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro  FILTER_V2_W4_H4_sse2 1
-INIT_XMM sse2
-cglobal interp_4tap_vert_%1_4x2, 4, 6, 8
-    mov         r4d,       r4m
-    sub         r0,        r1
-    pxor        m7,        m7
-
-%ifdef PIC
-    lea         r5,        [tabw_ChromaCoeff]
-    movh        m0,        [r5 + r4 * 8]
-%else
-    movh        m0,        [tabw_ChromaCoeff + r4 * 8]
-%endif
+%macro FILTER_P2S_4_4_sse2 1
+    movd        m2,     [r0 + %1]
+    movd        m3,     [r0 + r1 + %1]
+    movd        m4,     [r0 + r1 * 2 + %1]
+    movd        m5,     [r0 + r4 + %1]
+    punpckldq   m2,     m3
+    punpcklbw   m2,     m0
+    punpckldq   m4,     m5
+    punpcklbw   m4,     m0
+    psllw       m2,     6
+    psllw       m4,     6
+    psubw       m2,     m1
+    psubw       m4,     m1
+    movh        [r2 + r3 * 0 + %1 * 2], m2
+    movh        [r2 + r3 * 2 + %1 * 2], m4
+    movhps      [r2 + r3 * 1 + %1 * 2], m2
+    movhps      [r2 + r5 + %1 * 2], m4
+%endmacro
 
-    lea         r5,        [r0 + 2 * r1]
-    punpcklqdq  m0,        m0
-    movd        m2,        [r0]
-    movd        m3,        [r0 + r1]
-    movd        m4,        [r5]
-    movd        m5,        [r5 + r1]
-
-    punpcklbw   m2,        m3
-    punpcklbw   m1,        m4,        m5
-    punpcklwd   m2,        m1
-
-    movhlps     m6,        m2
-    punpcklbw   m2,        m7
-    punpcklbw   m6,        m7
-    pmaddwd     m2,        m0
-    pmaddwd     m6,        m0
-    packssdw    m2,        m6
-
-    movd        m1,        [r0 + 4 * r1]
-
-    punpcklbw   m3,        m4
-    punpcklbw   m5,        m1
-    punpcklwd   m3,        m5
-
-    movhlps     m6,        m3
-    punpcklbw   m3,        m7
-    punpcklbw   m6,        m7
-    pmaddwd     m3,        m0
-    pmaddwd     m6,        m0
-    packssdw    m3,        m6
-
-    pshuflw     m4,        m2,        q2301
-    pshufhw     m4,        m4,        q2301
-    paddw       m2,        m4
-    pshuflw     m5,        m3,        q2301
-    pshufhw     m5,        m5,        q2301
-    paddw       m3,        m5
-
-%ifidn %1, pp
-    psrld       m2,        16
-    psrld       m3,        16
-    packssdw    m2,        m3
-
-    paddw       m2,        [pw_32]
-    psraw       m2,        6
-    packuswb    m2,        m2
-
-    movd        [r2],      m2
-    psrldq      m2,        4
-    movd        [r2 + r3], m2
-%elifidn %1, ps
-    psrldq      m2,        2
-    psrldq      m3,        2
-    pshufd      m2,        m2, q3120
-    pshufd      m3,        m3, q3120
-    punpcklqdq  m2, m3
-
-    add         r3d,       r3d
-    psubw       m2,        [pw_2000]
-    movh        [r2],      m2
-    movhps      [r2 + r3], m2
-%endif
-    RET
+%macro FILTER_P2S_4_2_sse2 0
+    movd        m2,     [r0]
+    movd        m3,     [r0 + r1]
+    punpckldq   m2,     m3
+    punpcklbw   m2,     m0
+    psllw       m2,     6
+    psubw       m2,     [pw_8192]
+    movh        [r2],   m2
+    movhps      [r2 + r3 * 2], m2
+%endmacro
 
+%macro FILTER_P2S_8_4_sse2 1
+    movh        m2,     [r0 + %1]
+    movh        m3,     [r0 + r1 + %1]
+    movh        m4,     [r0 + r1 * 2 + %1]
+    movh        m5,     [r0 + r4 + %1]
+    punpcklbw   m2,     m0
+    punpcklbw   m3,     m0
+    punpcklbw   m5,     m0
+    punpcklbw   m4,     m0
+    psllw       m2,     6
+    psllw       m3,     6
+    psllw       m5,     6
+    psllw       m4,     6
+    psubw       m2,     m1
+    psubw       m3,     m1
+    psubw       m4,     m1
+    psubw       m5,     m1
+    movu        [r2 + r3 * 0 + %1 * 2], m2
+    movu        [r2 + r3 * 1 + %1 * 2], m3
+    movu        [r2 + r3 * 2 + %1 * 2], m4
+    movu        [r2 + r5 + %1 * 2], m5
 %endmacro
 
-    FILTER_V2_W4_H4_sse2 pp
-    FILTER_V2_W4_H4_sse2 ps
+%macro FILTER_P2S_8_2_sse2 1
+    movh        m2,     [r0 + %1]
+    movh        m3,     [r0 + r1 + %1]
+    punpcklbw   m2,     m0
+    punpcklbw   m3,     m0
+    psllw       m2,     6
+    psllw       m3,     6
+    psubw       m2,     m1
+    psubw       m3,     m1
+    movu        [r2 + r3 * 0 + %1 * 2], m2
+    movu        [r2 + r3 * 1 + %1 * 2], m3
+%endmacro
 
 ;-----------------------------------------------------------------------------
-; void interp_4tap_vert_%1_4x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride)
 ;-----------------------------------------------------------------------------
-%macro FILTER_V4_W4_H4_sse2 2
+%macro FILTER_PIX_TO_SHORT_sse2 2
 INIT_XMM sse2
-%if ARCH_X86_64
-cglobal interp_4tap_vert_%1_4x%2, 4, 6, 9
-    pxor        m8,        m8
-%else
-cglobal interp_4tap_vert_%1_4x%2, 4, 6, 8
+cglobal filterPixelToShort_%1x%2, 4, 6, 6
+    pxor        m0,     m0
+%if %2 == 2
+%if %1 == 4
+    FILTER_P2S_4_2_sse2
+%elif %1 == 8
+    add        r3d, r3d
+    mova       m1, [pw_8192]
+    FILTER_P2S_8_2_sse2 0
 %endif
-
-    mov         r4d,       r4m
-    sub         r0,        r1
-
-%ifdef PIC
-    lea         r5,        [tabw_ChromaCoeff]
-    movh        m0,        [r5 + r4 * 8]
 %else
-    movh        m0,        [tabw_ChromaCoeff + r4 * 8]
-%endif
-
-%ifidn %1,pp
-    mova        m1,        [pw_32]
-%elifidn %1,ps
-    add         r3d,       r3d
-    mova        m1,        [pw_2000]
-%endif
-
-    lea         r5,        [3 * r1]
-    lea         r4,        [3 * r3]
-    punpcklqdq  m0,        m0
-
-%assign x 1
+    add        r3d, r3d
+    mova       m1, [pw_8192]
+    lea        r4, [r1 * 3]
+    lea        r5, [r3 * 3]
+%assign y 1
 %rep %2/4
-    movd        m2,        [r0]
-    movd        m3,        [r0 + r1]
-    movd        m4,        [r0 + 2 * r1]
-    movd        m5,        [r0 + r5]
-
-    punpcklbw   m2,        m3
-    punpcklbw   m6,        m4,        m5
-    punpcklwd   m2,        m6
-
-    movhlps     m6,        m2
-    WORD_TO_DOUBLE         m2
-    WORD_TO_DOUBLE         m6
-    pmaddwd     m2,        m0
-    pmaddwd     m6,        m0
-    packssdw    m2,        m6
-
-    lea         r0,        [r0 + 4 * r1]
-    movd        m6,        [r0]
-
-    punpcklbw   m3,        m4
-    punpcklbw   m7,        m5,        m6
-    punpcklwd   m3,        m7
-
-    movhlps     m7,        m3
-    WORD_TO_DOUBLE         m3
-    WORD_TO_DOUBLE         m7
-    pmaddwd     m3,        m0
-    pmaddwd     m7,        m0
-    packssdw    m3,        m7
-
-    pshuflw     m7,        m2,        q2301
-    pshufhw     m7,        m7,        q2301
-    paddw       m2,        m7
-    pshuflw     m7,        m3,        q2301
-    pshufhw     m7,        m7,        q2301
-    paddw       m3,        m7
-
-%ifidn %1,pp
-    psrld       m2,        16
-    psrld       m3,        16
-    packssdw    m2,        m3
-    paddw       m2,        m1
-    psraw       m2,        6
-%elifidn %1,ps
-    psrldq      m2,        2
-    psrldq      m3,        2
-    pshufd      m2,        m2, q3120
-    pshufd      m3,        m3, q3120
-    punpcklqdq  m2,        m3
-
-    psubw       m2,        m1
-    movh        [r2],      m2
-    movhps      [r2 + r3], m2
-%endif
-
-    movd        m7,        [r0 + r1]
-
-    punpcklbw   m4,        m5
-    punpcklbw   m3,        m6,        m7
-    punpcklwd   m4,        m3
-
-    movhlps     m3,        m4
-    WORD_TO_DOUBLE         m4
-    WORD_TO_DOUBLE         m3
-    pmaddwd     m4,        m0
-    pmaddwd     m3,        m0
-    packssdw    m4,        m3
-
-    movd        m3,        [r0 + 2 * r1]
-
-    punpcklbw   m5,        m6
-    punpcklbw   m7,        m3
-    punpcklwd   m5,        m7
-
-    movhlps     m3,        m5
-    WORD_TO_DOUBLE         m5
-    WORD_TO_DOUBLE         m3
-    pmaddwd     m5,        m0
-    pmaddwd     m3,        m0
-    packssdw    m5,        m3
-
-    pshuflw     m7,        m4,        q2301
-    pshufhw     m7,        m7,        q2301
-    paddw       m4,        m7
-    pshuflw     m7,        m5,        q2301
-    pshufhw     m7,        m7,        q2301
-    paddw       m5,        m7
-
-%ifidn %1,pp
-    psrld       m4,        16
-    psrld       m5,        16
-    packssdw    m4,        m5
-
-    paddw       m4,        m1
-    psraw       m4,        6
-    packuswb    m2,        m4
-
-    movd        [r2],      m2
-    psrldq      m2,        4
-    movd        [r2 + r3], m2
-    psrldq      m2,        4
-    movd        [r2 + 2 * r3],      m2
-    psrldq      m2,        4
-    movd        [r2 + r4], m2
-%elifidn %1,ps
-    psrldq      m4,        2
-    psrldq      m5,        2
-    pshufd      m4,        m4, q3120
-    pshufd      m5,        m5, q3120
-    punpcklqdq  m4,        m5
-    psubw       m4,        m1
-    movh        [r2 + 2 * r3],      m4
-    movhps      [r2 + r4], m4
-%endif
-
-%if x < %2/4
-    lea         r2,        [r2 + 4 * r3]
+%assign x 0
+%rep %1/8
+    FILTER_P2S_8_4_sse2 x
+%if %2 == 6
+    lea         r0,     [r0 + 4 * r1]
+    lea         r2,     [r2 + 4 * r3]
+    FILTER_P2S_8_2_sse2 x
 %endif
-
-%assign x x+1
+%assign x x+8
 %endrep
-    RET
-
-%endmacro
-
-    FILTER_V4_W4_H4_sse2 pp, 4
-    FILTER_V4_W4_H4_sse2 pp, 8
-    FILTER_V4_W4_H4_sse2 pp, 16
-    FILTER_V4_W4_H4_sse2 pp, 32
-
-    FILTER_V4_W4_H4_sse2 ps, 4
-    FILTER_V4_W4_H4_sse2 ps, 8
-    FILTER_V4_W4_H4_sse2 ps, 16
-    FILTER_V4_W4_H4_sse2 ps, 32
-
-;-----------------------------------------------------------------------------
-;void interp_4tap_vert_%1_6x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro FILTER_V4_W6_H4_sse2 2
-INIT_XMM sse2
-cglobal interp_4tap_vert_%1_6x%2, 4, 7, 10
-    mov         r4d,       r4m
-    sub         r0,        r1
-    shl         r4d,       5
-    pxor        m9,        m9
-
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeffV]
-    mova        m6,        [r5 + r4]
-    mova        m5,        [r5 + r4 + 16]
-%else
-    mova        m6,        [tab_ChromaCoeffV + r4]
-    mova        m5,        [tab_ChromaCoeffV + r4 + 16]
+%rep (%1 % 8)/4
+    FILTER_P2S_4_4_sse2 x
+%assign x x+4
+%endrep
+%rep (%1 % 4)/2
+    FILTER_P2S_2_4_sse2 x
+%endrep
+%if y < %2/4
+    lea         r0,     [r0 + 4 * r1]
+    lea         r2,     [r2 + 4 * r3]
+%assign y y+1
 %endif
-
-%ifidn %1,pp
-    mova        m4,        [pw_32]
-%elifidn %1,ps
-    mova        m4,        [pw_2000]
-    add         r3d,       r3d
+%endrep
 %endif
-    lea         r5,        [3 * r1]
-
-%assign x 1
-%rep %2/4
-    movq        m0,        [r0]
-    movq        m1,        [r0 + r1]
-    movq        m2,        [r0 + 2 * r1]
-    movq        m3,        [r0 + r5]
-
-    punpcklbw   m0,        m1
-    punpcklbw   m1,        m2
-    punpcklbw   m2,        m3
-
-    movhlps     m7,        m0
-    punpcklbw   m0,        m9
-    punpcklbw   m7,        m9
-    pmaddwd     m0,        m6
-    pmaddwd     m7,        m6
-    packssdw    m0,        m7
-
-    movhlps     m8,        m2
-    movq        m7,        m2
-    punpcklbw   m8,        m9
-    punpcklbw   m7,        m9
-    pmaddwd     m8,        m5
-    pmaddwd     m7,        m5
-    packssdw    m7,        m8
-
-    paddw       m0,        m7
+RET
+%endmacro
 
-%ifidn %1,pp
-    paddw       m0,        m4
-    psraw       m0,        6
-    packuswb    m0,        m0
-
-    movd        [r2],      m0
-    pextrw      r6d,       m0,        2
-    mov         [r2 + 4],  r6w
-%elifidn %1,ps
-    psubw       m0,        m4
-    movh        [r2],      m0
-    pshufd      m0,        m0,        2
-    movd        [r2 + 8],  m0
-%endif
+    FILTER_PIX_TO_SHORT_sse2 2, 4
+    FILTER_PIX_TO_SHORT_sse2 2, 8
+    FILTER_PIX_TO_SHORT_sse2 2, 16
+    FILTER_PIX_TO_SHORT_sse2 4, 2
+    FILTER_PIX_TO_SHORT_sse2 4, 4
+    FILTER_PIX_TO_SHORT_sse2 4, 8
+    FILTER_PIX_TO_SHORT_sse2 4, 16
+    FILTER_PIX_TO_SHORT_sse2 4, 32
+    FILTER_PIX_TO_SHORT_sse2 6, 8
+    FILTER_PIX_TO_SHORT_sse2 6, 16
+    FILTER_PIX_TO_SHORT_sse2 8, 2
+    FILTER_PIX_TO_SHORT_sse2 8, 4
+    FILTER_PIX_TO_SHORT_sse2 8, 6
+    FILTER_PIX_TO_SHORT_sse2 8, 8
+    FILTER_PIX_TO_SHORT_sse2 8, 12
+    FILTER_PIX_TO_SHORT_sse2 8, 16
+    FILTER_PIX_TO_SHORT_sse2 8, 32
+    FILTER_PIX_TO_SHORT_sse2 8, 64
+    FILTER_PIX_TO_SHORT_sse2 12, 16
+    FILTER_PIX_TO_SHORT_sse2 12, 32
+    FILTER_PIX_TO_SHORT_sse2 16, 4
+    FILTER_PIX_TO_SHORT_sse2 16, 8
+    FILTER_PIX_TO_SHORT_sse2 16, 12
+    FILTER_PIX_TO_SHORT_sse2 16, 16
+    FILTER_PIX_TO_SHORT_sse2 16, 24
+    FILTER_PIX_TO_SHORT_sse2 16, 32
+    FILTER_PIX_TO_SHORT_sse2 16, 64
+    FILTER_PIX_TO_SHORT_sse2 24, 32
+    FILTER_PIX_TO_SHORT_sse2 24, 64
+    FILTER_PIX_TO_SHORT_sse2 32, 8
+    FILTER_PIX_TO_SHORT_sse2 32, 16
+    FILTER_PIX_TO_SHORT_sse2 32, 24
+    FILTER_PIX_TO_SHORT_sse2 32, 32
+    FILTER_PIX_TO_SHORT_sse2 32, 48
+    FILTER_PIX_TO_SHORT_sse2 32, 64
+    FILTER_PIX_TO_SHORT_sse2 48, 64
+    FILTER_PIX_TO_SHORT_sse2 64, 16
+    FILTER_PIX_TO_SHORT_sse2 64, 32
+    FILTER_PIX_TO_SHORT_sse2 64, 48
+    FILTER_PIX_TO_SHORT_sse2 64, 64
+
+%macro FILTER_H8_W8 7-8   ; t0, t1, t2, t3, coef, c512, src, dst
+    movu        %1, %7
+    pshufb      %2, %1, [tab_Lm +  0]
+    pmaddubsw   %2, %5
+    pshufb      %3, %1, [tab_Lm + 16]
+    pmaddubsw   %3, %5
+    phaddw      %2, %3
+    pshufb      %4, %1, [tab_Lm + 32]
+    pmaddubsw   %4, %5
+    pshufb      %1, %1, [tab_Lm + 48]
+    pmaddubsw   %1, %5
+    phaddw      %4, %1
+    phaddw      %2, %4
+  %if %0 == 8
+    pmulhrsw    %2, %6
+    packuswb    %2, %2
+    movh        %8, %2
+  %endif
+%endmacro
+
+;-----------------------------------------------------------------------------
+; Interpolate HV
+;-----------------------------------------------------------------------------
+%macro FILTER_HV8_START 7 ; (t0, t1, t2, t3, t4, off_src, off_coeff) -> (t3, t5), (t4, t1), [2]
+    mova        %5, [r0 +  (%6 + 0) * 16]
+    mova        %1, [r0 +  (%6 + 1) * 16]
+    mova        %2, [r0 +  (%6 + 2) * 16]
+    punpcklwd   %3, %5, %1
+    punpckhwd   %5, %1
+    pmaddwd     %3, [r5 + (%7) * 16]   ; R3 = L[0+1] -- Row 0
+    pmaddwd     %5, [r5 + (%7) * 16]   ; R0 = H[0+1]
+    punpcklwd   %4, %1, %2
+    punpckhwd   %1, %2
+    pmaddwd     %4, [r5 + (%7) * 16]   ; R4 = L[1+2] -- Row 1
+    pmaddwd     %1, [r5 + (%7) * 16]   ; R1 = H[1+2]
+%endmacro ; FILTER_HV8_START
 
-    lea         r0,        [r0 + 4 * r1]
+%macro FILTER_HV8_MID 10 ; (Row3, prevRow, sum0L, sum1L, sum0H, sum1H, t6, t7, off_src, off_coeff) -> [6]
+    mova        %8, [r0 +  (%9 + 0) * 16]
+    mova        %1, [r0 +  (%9 + 1) * 16]
+    punpcklwd   %7, %2, %8
+    punpckhwd   %2, %8
+    pmaddwd     %7, [r5 + %10 * 16]
+    pmaddwd     %2, [r5 + %10 * 16]
+    paddd       %3, %7              ; R3 = L[0+1+2+3] -- Row 0
+    paddd       %5, %2              ; R0 = H[0+1+2+3]
+    punpcklwd   %7, %8, %1
+    punpckhwd   %8, %1
+    pmaddwd     %7, [r5 + %10 * 16]
+    pmaddwd     %8, [r5 + %10 * 16]
+    paddd       %4, %7              ; R4 = L[1+2+3+4] -- Row 1
+    paddd       %6, %8              ; R1 = H[1+2+3+4]
+%endmacro ; FILTER_HV8_MID
 
-    movq        m0,        [r0]
-    punpcklbw   m3,        m0
+; Round and Saturate
+%macro FILTER_HV8_END 4 ; output in [1, 3]
+    paddd       %1, [pd_526336]
+    paddd       %2, [pd_526336]
+    paddd       %3, [pd_526336]
+    paddd       %4, [pd_526336]
+    psrad       %1, 12
+    psrad       %2, 12
+    psrad       %3, 12
+    psrad       %4, 12
+    packssdw    %1, %2
+    packssdw    %3, %4
 
-    movhlps     m8,        m1
-    punpcklbw   m1,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m1,        m6
-    pmaddwd     m8,        m6
-    packssdw    m1,        m8
+    ; TODO: is merge better? I think this way is short dependency link
+    packuswb    %1, %3
+%endmacro ; FILTER_HV8_END
 
-    movhlps     m8,        m3
-    movq        m7,        m3
-    punpcklbw   m8,        m9
-    punpcklbw   m7,        m9
-    pmaddwd     m8,        m5
-    pmaddwd     m7,        m5
-    packssdw    m7,        m8
+;-----------------------------------------------------------------------------
+; void interp_8tap_hv_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int idxX, int idxY)
+;-----------------------------------------------------------------------------
+INIT_XMM ssse3
+cglobal interp_8tap_hv_pp_8x8, 4, 7, 8, 0-15*16
+%define coef        m7
+%define stk_buf     rsp
 
-    paddw       m1,        m7
+    mov         r4d,        r4m
+    mov         r5d,        r5m
 
-%ifidn %1,pp
-    paddw       m1,        m4
-    psraw       m1,        6
-    packuswb    m1,        m1
-
-    movd        [r2 + r3], m1
-    pextrw      r6d,       m1,        2
-    mov         [r2 + r3 + 4], r6w
-%elifidn %1,ps
-    psubw       m1,        m4
-    movh        [r2 + r3], m1
-    pshufd      m1,        m1,        2
-    movd        [r2 + r3 + 8],  m1
+%ifdef PIC
+    lea         r6,         [tab_LumaCoeff]
+    movh        coef,       [r6 + r4 * 8]
+%else
+    movh        coef,       [tab_LumaCoeff + r4 * 8]
 %endif
+    punpcklqdq  coef,       coef
+
+    ; move to row -3
+    lea         r6,         [r1 + r1 * 2]
+    sub         r0,         r6
 
-    movq        m1,        [r0 + r1]
-    punpcklbw   m7,        m0,        m1
+    xor         r6,         r6
+    mov         r4,         rsp
 
-    movhlps     m8,        m2
-    punpcklbw   m2,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m2,        m6
-    pmaddwd     m8,        m6
-    packssdw    m2,        m8
+.loopH:
+    FILTER_H8_W8 m0, m1, m2, m3, coef, [pw_512], [r0 - 3]
+    psubw       m1,         [pw_2000]
+    mova        [r4],       m1
 
-    movhlps     m8,        m7
-    punpcklbw   m7,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m7,        m5
-    pmaddwd     m8,        m5
-    packssdw    m7,        m8
+    add         r0,         r1
+    add         r4,         16
+    inc         r6
+    cmp         r6,         8+7
+    jnz         .loopH
 
-    paddw       m2,        m7
-    lea         r2,        [r2 + 2 * r3]
+    ; ready to phase V
+    ; Here all of mN is free
 
-%ifidn %1,pp
-    paddw       m2,        m4
-    psraw       m2,        6
-    packuswb    m2,        m2
-    movd        [r2],      m2
-    pextrw      r6d,       m2,    2
-    mov         [r2 + 4],  r6w
-%elifidn %1,ps
-    psubw       m2,        m4
-    movh        [r2],      m2
-    pshufd      m2,        m2,        2
-    movd        [r2 + 8],  m2
-%endif
+    ; load coeff table
+    shl         r5,         6
+    lea         r6,         [tab_LumaCoeffV]
+    lea         r5,         [r5 + r6]
 
-    movq        m2,        [r0 + 2 * r1]
-    punpcklbw   m1,        m2
+    ; load intermedia buffer
+    mov         r0,         stk_buf
 
-    movhlps     m8,        m3
-    punpcklbw   m3,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m3,        m6
-    pmaddwd     m8,        m6
-    packssdw    m3,        m8
+    ; register mapping
+    ; r0 - src
+    ; r5 - coeff
+    ; r6 - loop_i
 
-    movhlps     m8,        m1
-    punpcklbw   m1,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m1,        m5
-    pmaddwd     m8,        m5
-    packssdw    m1,        m8
+    ; let's go
+    xor         r6,         r6
 
-    paddw       m3,        m1
+    ; TODO: this loop have more than 70 instructions, I think it is more than Intel loop decode cache
+.loopV:
 
-%ifidn %1,pp
-    paddw       m3,        m4
-    psraw       m3,        6
-    packuswb    m3,        m3
-
-    movd        [r2 + r3], m3
-    pextrw      r6d,       m3,    2
-    mov         [r2 + r3 + 4], r6w
-%elifidn %1,ps
-    psubw       m3,        m4
-    movh        [r2 + r3], m3
-    pshufd      m3,        m3,        2
-    movd        [r2 + r3 + 8],  m3
-%endif
+    FILTER_HV8_START    m1, m2, m3, m4, m0,             0, 0
+    FILTER_HV8_MID      m6, m2, m3, m4, m0, m1, m7, m5, 3, 1
+    FILTER_HV8_MID      m5, m6, m3, m4, m0, m1, m7, m2, 5, 2
+    FILTER_HV8_MID      m6, m5, m3, m4, m0, m1, m7, m2, 7, 3
+    FILTER_HV8_END      m3, m0, m4, m1
 
-%if x < %2/4
-    lea         r2,        [r2 + 2 * r3]
-%endif
+    movh        [r2],       m3
+    movhps      [r2 + r3],  m3
 
-%assign x x+1
-%endrep
-    RET
+    lea         r0,         [r0 + 16 * 2]
+    lea         r2,         [r2 + r3 * 2]
 
-%endmacro
+    inc         r6
+    cmp         r6,         8/2
+    jnz         .loopV
 
-%if ARCH_X86_64
-    FILTER_V4_W6_H4_sse2 pp, 8
-    FILTER_V4_W6_H4_sse2 pp, 16
-    FILTER_V4_W6_H4_sse2 ps, 8
-    FILTER_V4_W6_H4_sse2 ps, 16
-%endif
+    RET
 
 ;-----------------------------------------------------------------------------
-; void interp_4tap_vert_%1_8x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+; void interp_8tap_hv_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int idxX, int idxY)
 ;-----------------------------------------------------------------------------
-%macro FILTER_V4_W8_sse2 2
-INIT_XMM sse2
-cglobal interp_4tap_vert_%1_8x%2, 4, 7, 12
-    mov         r4d,       r4m
-    sub         r0,        r1
-    shl         r4d,       5
-    pxor        m9,        m9
-
-%ifidn %1,pp
-    mova        m4,        [pw_32]
-%elifidn %1,ps
-    mova        m4,        [pw_2000]
-    add         r3d,       r3d
-%endif
+INIT_XMM sse3
+cglobal interp_8tap_hv_pp_8x8, 4, 7, 8, 0-15*16
+    mov         r4d,        r4m
+    mov         r5d,        r5m
+    add         r4d,        r4d
+    pxor        m6,         m6
 
 %ifdef PIC
-    lea         r6,        [tab_ChromaCoeffV]
-    mova        m6,        [r6 + r4]
-    mova        m5,        [r6 + r4 + 16]
+    lea         r6,         [tabw_LumaCoeff]
+    mova        m3,         [r6 + r4 * 8]
 %else
-    mova        m6,        [tab_ChromaCoeffV + r4]
-    mova        m5,        [tab_ChromaCoeffV + r4 + 16]
+    mova        m3,         [tabw_LumaCoeff + r4 * 8]
 %endif
 
-    movq        m0,        [r0]
-    movq        m1,        [r0 + r1]
-    movq        m2,        [r0 + 2 * r1]
-    lea         r5,        [r0 + 2 * r1]
-    movq        m3,        [r5 + r1]
-
-    punpcklbw   m0,        m1
-    punpcklbw   m7,        m2,          m3
-
-    movhlps     m8,        m0
-    punpcklbw   m0,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m0,        m6
-    pmaddwd     m8,        m6
-    packssdw    m0,        m8
+    ; move to row -3
+    lea         r6,         [r1 + r1 * 2]
+    sub         r0,         r6
 
-    movhlps     m8,        m7
-    punpcklbw   m7,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m7,        m5
-    pmaddwd     m8,        m5
-    packssdw    m7,        m8
+    mov         r4,         rsp
 
-    paddw       m0,        m7
+%assign x 0     ;needed for FILTER_H8_W8_sse2 macro
+%assign y 1
+%rep 15
+    FILTER_H8_W8_sse2
+    psubw       m1,         [pw_2000]
+    mova        [r4],       m1
 
-%ifidn %1,pp
-    paddw       m0,        m4
-    psraw       m0,        6
-%elifidn %1,ps
-    psubw       m0,        m4
-    movu        [r2],      m0
+%if y < 15
+    add         r0,         r1
+    add         r4,         16
 %endif
+%assign y y+1
+%endrep
+
+    ; ready to phase V
+    ; Here all of mN is free
 
-    movq        m11,        [r0 + 4 * r1]
+    ; load coeff table
+    shl         r5,         6
+    lea         r6,         [tab_LumaCoeffV]
+    lea         r5,         [r5 + r6]
 
-    punpcklbw   m1,        m2
-    punpcklbw   m7,        m3,        m11
+    ; load intermedia buffer
+    mov         r0,         rsp
 
-    movhlps     m8,        m1
-    punpcklbw   m1,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m1,        m6
-    pmaddwd     m8,        m6
-    packssdw    m1,        m8
+    ; register mapping
+    ; r0 - src
+    ; r5 - coeff
 
-    movhlps     m8,        m7
-    punpcklbw   m7,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m7,        m5
-    pmaddwd     m8,        m5
-    packssdw    m7,        m8
+    ; let's go
+%assign y 1
+%rep 4
+    FILTER_HV8_START    m1, m2, m3, m4, m0,             0, 0
+    FILTER_HV8_MID      m6, m2, m3, m4, m0, m1, m7, m5, 3, 1
+    FILTER_HV8_MID      m5, m6, m3, m4, m0, m1, m7, m2, 5, 2
+    FILTER_HV8_MID      m6, m5, m3, m4, m0, m1, m7, m2, 7, 3
+    FILTER_HV8_END      m3, m0, m4, m1
 
-    paddw       m1,        m7
+    movh        [r2],       m3
+    movhps      [r2 + r3],  m3
 
-%ifidn %1,pp
-    paddw       m1,        m4
-    psraw       m1,        6
-    packuswb    m1,        m0
-
-    movhps      [r2],      m1
-    movh        [r2 + r3], m1
-%elifidn %1,ps
-    psubw       m1,        m4
-    movu        [r2 + r3], m1
+%if y < 4
+    lea         r0,         [r0 + 16 * 2]
+    lea         r2,         [r2 + r3 * 2]
 %endif
-%if %2 == 2     ;end of 8x2
+%assign y y+1
+%endrep
     RET
 
-%else
-    lea         r6,        [r0 + 4 * r1]
-    movq        m1,        [r6 + r1]
-
-    punpcklbw   m2,        m3
-    punpcklbw   m7,        m11,        m1
-
-    movhlps     m8,        m2
-    punpcklbw   m2,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m2,        m6
-    pmaddwd     m8,        m6
-    packssdw    m2,        m8
-
-    movhlps     m8,        m7
-    punpcklbw   m7,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m7,        m5
-    pmaddwd     m8,        m5
-    packssdw    m7,        m8
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+%macro P2S_H_2xN 1
+INIT_XMM sse4
+cglobal filterPixelToShort_2x%1, 3, 4, 3
+    mov         r3d, r3m
+    add         r3d, r3d
 
-    paddw       m2,        m7
+    ; load constant
+    mova        m1, [pb_128]
+    mova        m2, [tab_c_64_n64]
 
-%ifidn %1,pp
-    paddw       m2,        m4
-    psraw       m2,        6
-%elifidn %1,ps
-    psubw       m2,        m4
-    movu        [r2 + 2 * r3], m2
-%endif
-
-    movq        m10,        [r6 + 2 * r1]
-
-    punpcklbw   m3,        m11
-    punpcklbw   m7,        m1,        m10
-
-    movhlps     m8,        m3
-    punpcklbw   m3,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m3,        m6
-    pmaddwd     m8,        m6
-    packssdw    m3,        m8
-
-    movhlps     m8,        m7
-    punpcklbw   m7,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m7,        m5
-    pmaddwd     m8,        m5
-    packssdw    m7,        m8
+%rep %1/2
+    movd        m0, [r0]
+    pinsrd      m0, [r0 + r1], 1
+    punpcklbw   m0, m1
+    pmaddubsw   m0, m2
 
-    paddw       m3,        m7
-    lea         r5,        [r2 + 2 * r3]
+    movd        [r2 + r3 * 0], m0
+    pextrd      [r2 + r3 * 1], m0, 2
 
-%ifidn %1,pp
-    paddw       m3,        m4
-    psraw       m3,        6
-    packuswb    m3,        m2
-
-    movhps      [r2 + 2 * r3], m3
-    movh        [r5 + r3], m3
-%elifidn %1,ps
-    psubw       m3,        m4
-    movu        [r5 + r3], m3
-%endif
-%if %2 == 4     ;end of 8x4
+    lea         r0, [r0 + r1 * 2]
+    lea         r2, [r2 + r3 * 2]
+%endrep
     RET
+%endmacro
+    P2S_H_2xN 4
+    P2S_H_2xN 8
+    P2S_H_2xN 16
 
-%else
-    lea         r6,        [r6 + 2 * r1]
-    movq        m3,        [r6 + r1]
-
-    punpcklbw   m11,        m1
-    punpcklbw   m7,        m10,        m3
-
-    movhlps     m8,        m11
-    punpcklbw   m11,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m11,        m6
-    pmaddwd     m8,        m6
-    packssdw    m11,        m8
-
-    movhlps     m8,        m7
-    punpcklbw   m7,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m7,        m5
-    pmaddwd     m8,        m5
-    packssdw    m7,        m8
-
-    paddw       m11,       m7
-
-%ifidn %1, pp
-    paddw       m11,       m4
-    psraw       m11,       6
-%elifidn %1,ps
-    psubw       m11,       m4
-    movu        [r2 + 4 * r3], m11
-%endif
-
-    movq        m7,        [r0 + 8 * r1]
-
-    punpcklbw   m1,        m10
-    punpcklbw   m3,        m7
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+%macro P2S_H_4xN 1
+INIT_XMM sse4
+cglobal filterPixelToShort_4x%1, 3, 6, 4
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r4, [r3 * 3]
+    lea         r5, [r1 * 3]
 
-    movhlps     m8,        m1
-    punpcklbw   m1,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m1,        m6
-    pmaddwd     m8,        m6
-    packssdw    m1,        m8
+    ; load constant
+    mova        m2, [pb_128]
+    mova        m3, [tab_c_64_n64]
 
-    movhlps     m8,        m3
-    punpcklbw   m3,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m3,        m5
-    pmaddwd     m8,        m5
-    packssdw    m3,        m8
+%assign x 0
+%rep %1/4
+    movd        m0, [r0]
+    pinsrd      m0, [r0 + r1], 1
+    punpcklbw   m0, m2
+    pmaddubsw   m0, m3
 
-    paddw       m1,        m3
-    lea         r5,        [r2 + 4 * r3]
+    movd        m1, [r0 + r1 * 2]
+    pinsrd      m1, [r0 + r5], 1
+    punpcklbw   m1, m2
+    pmaddubsw   m1, m3
 
-%ifidn %1,pp
-    paddw       m1,        m4
-    psraw       m1,        6
-    packuswb    m1,        m11
-
-    movhps      [r2 + 4 * r3], m1
-    movh        [r5 + r3], m1
-%elifidn %1,ps
-    psubw       m1,        m4
-    movu        [r5 + r3], m1
+    movq        [r2 + r3 * 0], m0
+    movq        [r2 + r3 * 2], m1
+    movhps      [r2 + r3 * 1], m0
+    movhps      [r2 + r4], m1
+%assign x x+1
+%if (x != %1/4)
+    lea         r0, [r0 + r1 * 4]
+    lea         r2, [r2 + r3 * 4]
 %endif
-%if %2 == 6
+%endrep
     RET
-
-%else
-  %error INVALID macro argument, only 2, 4 or 6!
-%endif
-%endif
-%endif
 %endmacro
-
-%if ARCH_X86_64
-    FILTER_V4_W8_sse2 pp, 2
-    FILTER_V4_W8_sse2 pp, 4
-    FILTER_V4_W8_sse2 pp, 6
-    FILTER_V4_W8_sse2 ps, 2
-    FILTER_V4_W8_sse2 ps, 4
-    FILTER_V4_W8_sse2 ps, 6
-%endif
+    P2S_H_4xN 4
+    P2S_H_4xN 8
+    P2S_H_4xN 16
+    P2S_H_4xN 32
 
 ;-----------------------------------------------------------------------------
-; void interp_4tap_vert_%1_8x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
 ;-----------------------------------------------------------------------------
-%macro FILTER_V4_W8_H8_H16_H32_sse2 2
-INIT_XMM sse2
-cglobal interp_4tap_vert_%1_8x%2, 4, 6, 11
-    mov         r4d,       r4m
-    sub         r0,        r1
-    shl         r4d,       5
-    pxor        m9,        m9
-
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeffV]
-    mova        m6,        [r5 + r4]
-    mova        m5,        [r5 + r4 + 16]
-%else
-    mova        m6,        [tab_ChromaCoeff + r4]
-    mova        m5,        [tab_ChromaCoeff + r4 + 16]
-%endif
-
-%ifidn %1,pp
-    mova        m4,        [pw_32]
-%elifidn %1,ps
-    mova        m4,        [pw_2000]
-    add         r3d,       r3d
-%endif
+%macro P2S_H_6xN 1
+INIT_XMM sse4
+cglobal filterPixelToShort_6x%1, 3, 7, 6
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r4, [r1 * 3]
+    lea         r5, [r3 * 3]
 
-    lea         r5,        [r1 * 3]
+    ; load height
+    mov         r6d, %1/4
 
-%assign x 1
-%rep %2/4
-    movq        m0,        [r0]
-    movq        m1,        [r0 + r1]
-    movq        m2,        [r0 + 2 * r1]
-    movq        m3,        [r0 + r5]
-
-    punpcklbw   m0,        m1
-    punpcklbw   m1,        m2
-    punpcklbw   m2,        m3
-
-    movhlps     m7,        m0
-    punpcklbw   m0,        m9
-    punpcklbw   m7,        m9
-    pmaddwd     m0,        m6
-    pmaddwd     m7,        m6
-    packssdw    m0,        m7
-
-    movhlps     m8,        m2
-    movq        m7,        m2
-    punpcklbw   m8,        m9
-    punpcklbw   m7,        m9
-    pmaddwd     m8,        m5
-    pmaddwd     m7,        m5
-    packssdw    m7,        m8
-
-    paddw       m0,        m7
+    ; load constant
+    mova        m4, [pb_128]
+    mova        m5, [tab_c_64_n64]
 
-%ifidn %1,pp
-    paddw       m0,        m4
-    psraw       m0,        6
-%elifidn %1,ps
-    psubw       m0,        m4
-    movu        [r2],      m0
-%endif
+.loop:
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-    lea         r0,        [r0 + 4 * r1]
-    movq        m10,       [r0]
-    punpcklbw   m3,        m10
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-    movhlps     m8,        m1
-    punpcklbw   m1,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m1,        m6
-    pmaddwd     m8,        m6
-    packssdw    m1,        m8
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
 
-    movhlps     m8,        m3
-    movq        m7,        m3
-    punpcklbw   m8,        m9
-    punpcklbw   m7,        m9
-    pmaddwd     m8,        m5
-    pmaddwd     m7,        m5
-    packssdw    m7,        m8
+    movh        m3, [r0 + r4]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
 
-    paddw       m1,        m7
+    movh        [r2 + r3 * 0], m0
+    pextrd      [r2 + r3 * 0 + 8], m0, 2
+    movh        [r2 + r3 * 1], m1
+    pextrd      [r2 + r3 * 1 + 8], m1, 2
+    movh        [r2 + r3 * 2], m2
+    pextrd      [r2 + r3 * 2 + 8], m2, 2
+    movh        [r2 + r5], m3
+    pextrd      [r2 + r5 + 8], m3, 2
 
-%ifidn %1,pp
-    paddw       m1,        m4
-    psraw       m1,        6
-
-    packuswb    m0,        m1
-    movh        [r2],      m0
-    movhps      [r2 + r3], m0
-%elifidn %1,ps
-    psubw       m1,        m4
-    movu        [r2 + r3], m1
-%endif
+    lea         r0, [r0 + r1 * 4]
+    lea         r2, [r2 + r3 * 4]
 
-    movq        m1,        [r0 + r1]
-    punpcklbw   m10,       m1
+    dec         r6d
+    jnz         .loop
+    RET
+%endmacro
+    P2S_H_6xN 8
+    P2S_H_6xN 16
 
-    movhlps     m8,        m2
-    punpcklbw   m2,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m2,        m6
-    pmaddwd     m8,        m6
-    packssdw    m2,        m8
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+%macro P2S_H_8xN 1
+INIT_XMM ssse3
+cglobal filterPixelToShort_8x%1, 3, 7, 6
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r5, [r1 * 3]
+    lea         r6, [r3 * 3]
 
-    movhlps     m8,        m10
-    punpcklbw   m10,       m9
-    punpcklbw   m8,        m9
-    pmaddwd     m10,       m5
-    pmaddwd     m8,        m5
-    packssdw    m10,       m8
+    ; load height
+    mov         r4d, %1/4
 
-    paddw       m2,        m10
-    lea         r2,        [r2 + 2 * r3]
+    ; load constant
+    mova        m4, [pb_128]
+    mova        m5, [tab_c_64_n64]
 
-%ifidn %1,pp
-    paddw       m2,        m4
-    psraw       m2,        6
-%elifidn %1,ps
-    psubw       m2,        m4
-    movu        [r2],      m2
-%endif
+.loop:
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-    movq        m7,        [r0 + 2 * r1]
-    punpcklbw   m1,        m7
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-    movhlps     m8,        m3
-    punpcklbw   m3,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m3,        m6
-    pmaddwd     m8,        m6
-    packssdw    m3,        m8
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
 
-    movhlps     m8,        m1
-    punpcklbw   m1,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m1,        m5
-    pmaddwd     m8,        m5
-    packssdw    m1,        m8
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
 
-    paddw       m3,        m1
+    movu        [r2 + r3 * 0], m0
+    movu        [r2 + r3 * 1], m1
+    movu        [r2 + r3 * 2], m2
+    movu        [r2 + r6 ], m3
 
-%ifidn %1,pp
-    paddw       m3,        m4
-    psraw       m3,        6
-
-    packuswb    m2,        m3
-    movh        [r2],      m2
-    movhps      [r2 + r3], m2
-%elifidn %1,ps
-    psubw       m3,        m4
-    movu        [r2 + r3], m3
-%endif
+    lea         r0, [r0 + r1 * 4]
+    lea         r2, [r2 + r3 * 4]
 
-%if x < %2/4
-    lea         r2,        [r2 + 2 * r3]
-%endif
-%endrep
+    dec         r4d
+    jnz         .loop
     RET
 %endmacro
-
-%if ARCH_X86_64
-    FILTER_V4_W8_H8_H16_H32_sse2 pp,  8
-    FILTER_V4_W8_H8_H16_H32_sse2 pp, 16
-    FILTER_V4_W8_H8_H16_H32_sse2 pp, 32
-
-    FILTER_V4_W8_H8_H16_H32_sse2 pp, 12
-    FILTER_V4_W8_H8_H16_H32_sse2 pp, 64
-
-    FILTER_V4_W8_H8_H16_H32_sse2 ps,  8
-    FILTER_V4_W8_H8_H16_H32_sse2 ps, 16
-    FILTER_V4_W8_H8_H16_H32_sse2 ps, 32
-
-    FILTER_V4_W8_H8_H16_H32_sse2 ps, 12
-    FILTER_V4_W8_H8_H16_H32_sse2 ps, 64
-%endif
+    P2S_H_8xN 8
+    P2S_H_8xN 4
+    P2S_H_8xN 16
+    P2S_H_8xN 32
+    P2S_H_8xN 12
+    P2S_H_8xN 64
 
 ;-----------------------------------------------------------------------------
-; void interp_4tap_vert_%1_12x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
 ;-----------------------------------------------------------------------------
-%macro FILTER_V4_W12_H2_sse2 2
-INIT_XMM sse2
-cglobal interp_4tap_vert_%1_12x%2, 4, 6, 11
-    mov         r4d,       r4m
-    sub         r0,        r1
-    shl         r4d,       5
-    pxor        m9,        m9
+INIT_XMM ssse3
+cglobal filterPixelToShort_8x6, 3, 7, 5
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r4, [r1 * 3]
+    lea         r5, [r1 * 5]
+    lea         r6, [r3 * 3]
 
-%ifidn %1,pp
-    mova        m6,        [pw_32]
-%elifidn %1,ps
-    mova        m6,        [pw_2000]
-    add         r3d,       r3d
-%endif
+    ; load constant
+    mova        m3, [pb_128]
+    mova        m4, [tab_c_64_n64]
 
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeffV]
-    mova        m1,        [r5 + r4]
-    mova        m0,        [r5 + r4 + 16]
-%else
-    mova        m1,        [tab_ChromaCoeffV + r4]
-    mova        m0,        [tab_ChromaCoeffV + r4 + 16]
-%endif
+    movh        m0, [r0]
+    punpcklbw   m0, m3
+    pmaddubsw   m0, m4
 
-%assign x 1
-%rep %2/2
-    movu        m2,        [r0]
-    movu        m3,        [r0 + r1]
-
-    punpcklbw   m4,        m2,        m3
-    punpckhbw   m2,        m3
-
-    movhlps     m8,        m4
-    punpcklbw   m4,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m4,        m1
-    pmaddwd     m8,        m1
-    packssdw    m4,        m8
-
-    movhlps     m8,        m2
-    punpcklbw   m2,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m2,        m1
-    pmaddwd     m8,        m1
-    packssdw    m2,        m8
-
-    lea         r0,        [r0 + 2 * r1]
-    movu        m5,        [r0]
-    movu        m7,        [r0 + r1]
-
-    punpcklbw   m10,       m5,        m7
-    movhlps     m8,        m10
-    punpcklbw   m10,       m9
-    punpcklbw   m8,        m9
-    pmaddwd     m10,       m0
-    pmaddwd     m8,        m0
-    packssdw    m10,       m8
-
-    paddw       m4,        m10
-
-    punpckhbw   m10,       m5,        m7
-    movhlps     m8,        m10
-    punpcklbw   m10,       m9
-    punpcklbw   m8,        m9
-    pmaddwd     m10,       m0
-    pmaddwd     m8,        m0
-    packssdw    m10,       m8
-
-    paddw       m2,        m10
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m3
+    pmaddubsw   m1, m4
 
-%ifidn %1,pp
-    paddw       m4,        m6
-    psraw       m4,        6
-    paddw       m2,        m6
-    psraw       m2,        6
-
-    packuswb    m4,        m2
-    movh        [r2],      m4
-    psrldq      m4,        8
-    movd        [r2 + 8],  m4
-%elifidn %1,ps
-    psubw       m4,        m6
-    psubw       m2,        m6
-    movu        [r2],      m4
-    movh        [r2 + 16], m2
-%endif
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m3
+    pmaddubsw   m2, m4
 
-    punpcklbw   m4,        m3,        m5
-    punpckhbw   m3,        m5
-
-    movhlps     m8,        m4
-    punpcklbw   m4,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m4,        m1
-    pmaddwd     m8,        m1
-    packssdw    m4,        m8
-
-    movhlps     m8,        m4
-    punpcklbw   m3,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m3,        m1
-    pmaddwd     m8,        m1
-    packssdw    m3,        m8
-
-    movu        m5,        [r0 + 2 * r1]
-    punpcklbw   m2,        m7,        m5
-    punpckhbw   m7,        m5
-
-    movhlps     m8,        m2
-    punpcklbw   m2,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m2,        m0
-    pmaddwd     m8,        m0
-    packssdw    m2,        m8
-
-    movhlps     m8,        m7
-    punpcklbw   m7,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m7,        m0
-    pmaddwd     m8,        m0
-    packssdw    m7,        m8
-
-    paddw       m4,        m2
-    paddw       m3,        m7
+    movu        [r2 + r3 * 0], m0
+    movu        [r2 + r3 * 1], m1
+    movu        [r2 + r3 * 2], m2
 
-%ifidn %1,pp
-    paddw       m4,        m6
-    psraw       m4,        6
-    paddw       m3,        m6
-    psraw       m3,        6
-
-    packuswb    m4,        m3
-    movh        [r2 + r3], m4
-    psrldq      m4,        8
-    movd        [r2 + r3 + 8], m4
-%elifidn %1,ps
-    psubw       m4,        m6
-    psubw       m3,        m6
-    movu        [r2 + r3], m4
-    movh        [r2 + r3 + 16], m3
-%endif
+    movh        m0, [r0 + r4]
+    punpcklbw   m0, m3
+    pmaddubsw   m0, m4
 
-%if x < %2/2
-    lea         r2,        [r2 + 2 * r3]
-%endif
-%assign x x+1
-%endrep
-    RET
+    movh        m1, [r0 + r1 * 4]
+    punpcklbw   m1, m3
+    pmaddubsw   m1, m4
 
-%endmacro
+    movh        m2, [r0 + r5]
+    punpcklbw   m2, m3
+    pmaddubsw   m2, m4
 
-%if ARCH_X86_64
-    FILTER_V4_W12_H2_sse2 pp, 16
-    FILTER_V4_W12_H2_sse2 pp, 32
-    FILTER_V4_W12_H2_sse2 ps, 16
-    FILTER_V4_W12_H2_sse2 ps, 32
-%endif
+    movu        [r2 + r6 ], m0
+    movu        [r2 + r3 * 4], m1
+    lea         r2, [r2 + r3 * 4]
+    movu        [r2 + r3], m2
+
+    RET
 
 ;-----------------------------------------------------------------------------
-; void interp_4tap_vert_%1_16x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
 ;-----------------------------------------------------------------------------
-%macro FILTER_V4_W16_H2_sse2 2
-INIT_XMM sse2
-cglobal interp_4tap_vert_%1_16x%2, 4, 6, 11
-    mov         r4d,       r4m
-    sub         r0,        r1
-    shl         r4d,       5
-    pxor        m9,        m9
-
-%ifidn %1,pp
-    mova        m6,        [pw_32]
-%elifidn %1,ps
-    mova        m6,        [pw_2000]
-    add         r3d,       r3d
-%endif
+%macro P2S_H_16xN 1
+INIT_XMM ssse3
+cglobal filterPixelToShort_16x%1, 3, 7, 6
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r4, [r3 * 3]
+    lea         r5, [r1 * 3]
 
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeffV]
-    mova        m1,        [r5 + r4]
-    mova        m0,        [r5 + r4 + 16]
-%else
-    mova        m1,        [tab_ChromaCoeffV + r4]
-    mova        m0,        [tab_ChromaCoeffV + r4 + 16]
-%endif
+   ; load height
+    mov         r6d, %1/4
 
-%assign x 1
-%rep %2/2
-    movu        m2,        [r0]
-    movu        m3,        [r0 + r1]
-
-    punpcklbw   m4,        m2,        m3
-    punpckhbw   m2,        m3
-
-    movhlps     m8,        m4
-    punpcklbw   m4,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m4,        m1
-    pmaddwd     m8,        m1
-    packssdw    m4,        m8
-
-    movhlps     m8,        m2
-    punpcklbw   m2,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m2,        m1
-    pmaddwd     m8,        m1
-    packssdw    m2,        m8
-
-    lea         r0,        [r0 + 2 * r1]
-    movu        m5,        [r0]
-    movu        m10,       [r0 + r1]
-
-    punpckhbw   m7,        m5,        m10
-    movhlps     m8,        m7
-    punpcklbw   m7,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m7,        m0
-    pmaddwd     m8,        m0
-    packssdw    m7,        m8
-    paddw       m2,        m7
-
-    punpcklbw   m7,        m5,        m10
-    movhlps     m8,        m7
-    punpcklbw   m7,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m7,        m0
-    pmaddwd     m8,        m0
-    packssdw    m7,        m8
-    paddw       m4,        m7
+    ; load constant
+    mova        m4, [pb_128]
+    mova        m5, [tab_c_64_n64]
 
-%ifidn %1,pp
-    paddw       m4,        m6
-    psraw       m4,        6
-    paddw       m2,        m6
-    psraw       m2,        6
-
-    packuswb    m4,        m2
-    movu        [r2],      m4
-%elifidn %1,ps
-    psubw       m4,        m6
-    psubw       m2,        m6
-    movu        [r2],      m4
-    movu        [r2 + 16], m2
-%endif
-
-    punpcklbw   m4,        m3,        m5
-    punpckhbw   m3,        m5
-
-    movhlps     m8,        m4
-    punpcklbw   m4,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m4,        m1
-    pmaddwd     m8,        m1
-    packssdw    m4,        m8
-
-    movhlps     m8,        m3
-    punpcklbw   m3,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m3,        m1
-    pmaddwd     m8,        m1
-    packssdw    m3,        m8
-
-    movu        m5,        [r0 + 2 * r1]
-
-    punpcklbw   m2,        m10,       m5
-    punpckhbw   m10,       m5
-
-    movhlps     m8,        m2
-    punpcklbw   m2,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m2,        m0
-    pmaddwd     m8,        m0
-    packssdw    m2,        m8
-
-    movhlps     m8,        m10
-    punpcklbw   m10,       m9
-    punpcklbw   m8,        m9
-    pmaddwd     m10,       m0
-    pmaddwd     m8,        m0
-    packssdw    m10,       m8
-
-    paddw       m4,        m2
-    paddw       m3,        m10
-
-%ifidn %1,pp
-    paddw       m4,        m6
-    psraw       m4,        6
-    paddw       m3,        m6
-    psraw       m3,        6
-
-    packuswb    m4,        m3
-    movu        [r2 + r3], m4
-%elifidn %1,ps
-    psubw       m4,        m6
-    psubw       m3,        m6
-    movu        [r2 + r3], m4
-    movu        [r2 + r3 + 16], m3
-%endif
-
-%if x < %2/2
-    lea         r2,        [r2 + 2 * r3]
-%endif
-%assign x x+1
-%endrep
-    RET
-
-%endmacro
-
-%if ARCH_X86_64
-    FILTER_V4_W16_H2_sse2 pp, 4
-    FILTER_V4_W16_H2_sse2 pp, 8
-    FILTER_V4_W16_H2_sse2 pp, 12
-    FILTER_V4_W16_H2_sse2 pp, 16
-    FILTER_V4_W16_H2_sse2 pp, 32
-
-    FILTER_V4_W16_H2_sse2 pp, 24
-    FILTER_V4_W16_H2_sse2 pp, 64
-
-    FILTER_V4_W16_H2_sse2 ps, 4
-    FILTER_V4_W16_H2_sse2 ps, 8
-    FILTER_V4_W16_H2_sse2 ps, 12
-    FILTER_V4_W16_H2_sse2 ps, 16
-    FILTER_V4_W16_H2_sse2 ps, 32
-
-    FILTER_V4_W16_H2_sse2 ps, 24
-    FILTER_V4_W16_H2_sse2 ps, 64
-%endif
-
-;-----------------------------------------------------------------------------
-;void interp_4tap_vert_%1_24%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro FILTER_V4_W24_sse2 2
-INIT_XMM sse2
-cglobal interp_4tap_vert_%1_24x%2, 4, 6, 11
-    mov         r4d,       r4m
-    sub         r0,        r1
-    shl         r4d,       5
-    pxor        m9,        m9
+.loop:
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-%ifidn %1,pp
-    mova        m6,        [pw_32]
-%elifidn %1,ps
-    mova        m6,        [pw_2000]
-    add         r3d,       r3d
-%endif
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeffV]
-    mova        m1,        [r5 + r4]
-    mova        m0,        [r5 + r4 + 16]
-%else
-    mova        m1,        [tab_ChromaCoeffV + r4]
-    mova        m0,        [tab_ChromaCoeffV + r4 + 16]
-%endif
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
 
-%assign x 1
-%rep %2/2
-    movu        m2,        [r0]
-    movu        m3,        [r0 + r1]
-
-    punpcklbw   m4,        m2,        m3
-    punpckhbw   m2,        m3
-
-    movhlps     m8,        m4
-    punpcklbw   m4,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m4,        m1
-    pmaddwd     m8,        m1
-    packssdw    m4,        m8
-
-    movhlps     m8,        m2
-    punpcklbw   m2,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m2,        m1
-    pmaddwd     m8,        m1
-    packssdw    m2,        m8
-
-    lea         r5,        [r0 + 2 * r1]
-    movu        m5,        [r5]
-    movu        m10,       [r5 + r1]
-    punpcklbw   m7,        m5,        m10
-
-    movhlps     m8,        m7
-    punpcklbw   m7,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m7,        m0
-    pmaddwd     m8,        m0
-    packssdw    m7,        m8
-    paddw       m4,        m7
-
-    punpckhbw   m7,        m5,        m10
-
-    movhlps     m8,        m7
-    punpcklbw   m7,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m7,        m0
-    pmaddwd     m8,        m0
-    packssdw    m7,        m8
-
-    paddw       m2,        m7
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
 
-%ifidn %1,pp
-    paddw       m4,        m6
-    psraw       m4,        6
-    paddw       m2,        m6
-    psraw       m2,        6
-
-    packuswb    m4,        m2
-    movu        [r2],      m4
-%elifidn %1,ps
-    psubw       m4,        m6
-    psubw       m2,        m6
-    movu        [r2],      m4
-    movu        [r2 + 16], m2
-%endif
+    movu        [r2 + r3 * 0], m0
+    movu        [r2 + r3 * 1], m1
+    movu        [r2 + r3 * 2], m2
+    movu        [r2 + r4], m3
 
-    punpcklbw   m4,        m3,        m5
-    punpckhbw   m3,        m5
-
-    movhlps     m8,        m4
-    punpcklbw   m4,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m4,        m1
-    pmaddwd     m8,        m1
-    packssdw    m4,        m8
-
-    movhlps     m8,        m3
-    punpcklbw   m3,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m3,        m1
-    pmaddwd     m8,        m1
-    packssdw    m3,        m8
-
-    movu        m2,        [r5 + 2 * r1]
-
-    punpcklbw   m5,        m10,        m2
-    punpckhbw   m10,       m2
-
-    movhlps     m8,        m5
-    punpcklbw   m5,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m5,        m0
-    pmaddwd     m8,        m0
-    packssdw    m5,        m8
-
-    movhlps     m8,        m10
-    punpcklbw   m10,       m9
-    punpcklbw   m8,        m9
-    pmaddwd     m10,       m0
-    pmaddwd     m8,        m0
-    packssdw    m10,       m8
-
-    paddw       m4,        m5
-    paddw       m3,        m10
+    lea         r0, [r0 + 8]
 
-%ifidn %1,pp
-    paddw       m4,        m6
-    psraw       m4,        6
-    paddw       m3,        m6
-    psraw       m3,        6
-
-    packuswb    m4,        m3
-    movu        [r2 + r3], m4
-%elifidn %1,ps
-    psubw       m4,        m6
-    psubw       m3,        m6
-    movu        [r2 + r3], m4
-    movu        [r2 + r3 + 16], m3
-%endif
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-    movq        m2,        [r0 + 16]
-    movq        m3,        [r0 + r1 + 16]
-    movq        m4,        [r5 + 16]
-    movq        m5,        [r5 + r1 + 16]
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-    punpcklbw   m2,        m3
-    punpcklbw   m4,        m5
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
 
-    movhlps     m8,        m4
-    punpcklbw   m4,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m4,        m0
-    pmaddwd     m8,        m0
-    packssdw    m4,        m8
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
 
-    movhlps     m8,        m2
-    punpcklbw   m2,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m2,        m1
-    pmaddwd     m8,        m1
-    packssdw    m2,        m8
+    movu        [r2 + r3 * 0 + 16], m0
+    movu        [r2 + r3 * 1 + 16], m1
+    movu        [r2 + r3 * 2 + 16], m2
+    movu        [r2 + r4 + 16], m3
 
-    paddw       m2,        m4
+    lea         r0, [r0 + r1 * 4 - 8]
+    lea         r2, [r2 + r3 * 4]
 
-%ifidn %1,pp
-    paddw       m2,        m6
-    psraw       m2,        6
-%elifidn %1,ps
-    psubw       m2,        m6
-    movu        [r2 + 32], m2
-%endif
+    dec         r6d
+    jnz         .loop
+    RET
+%endmacro
+    P2S_H_16xN 16
+    P2S_H_16xN 4
+    P2S_H_16xN 8
+    P2S_H_16xN 12
+    P2S_H_16xN 32
+    P2S_H_16xN 64
+    P2S_H_16xN 24
 
-    movq        m3,        [r0 + r1 + 16]
-    movq        m4,        [r5 + 16]
-    movq        m5,        [r5 + r1 + 16]
-    movq        m7,        [r5 + 2 * r1 + 16]
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+INIT_YMM avx2
+cglobal filterPixelToShort_16x4, 3, 4, 2
+    mov             r3d, r3m
+    add             r3d, r3d
 
-    punpcklbw   m3,        m4
-    punpcklbw   m5,        m7
+    ; load constant
+    vbroadcasti128  m1, [pw_2000]
 
-    movhlps     m8,        m5
-    punpcklbw   m5,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m5,        m0
-    pmaddwd     m8,        m0
-    packssdw    m5,        m8
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-    movhlps     m8,        m3
-    punpcklbw   m3,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m3,        m1
-    pmaddwd     m8,        m1
-    packssdw    m3,        m8
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-    paddw       m3,        m5
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-%ifidn %1,pp
-    paddw       m3,        m6
-    psraw       m3,        6
-
-    packuswb    m2,        m3
-    movh        [r2 + 16], m2
-    movhps      [r2 + r3 + 16], m2
-%elifidn %1,ps
-    psubw       m3,        m6
-    movu        [r2 + r3 + 32], m3
-%endif
+    lea             r1, [r1 * 3]
+    lea             r3, [r3 * 3]
 
-%if x < %2/2
-    mov         r0,        r5
-    lea         r2,        [r2 + 2 * r3]
-%endif
-%assign x x+1
-%endrep
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
     RET
 
-%endmacro
-
-%if ARCH_X86_64
-    FILTER_V4_W24_sse2 pp, 32
-    FILTER_V4_W24_sse2 pp, 64
-    FILTER_V4_W24_sse2 ps, 32
-    FILTER_V4_W24_sse2 ps, 64
-%endif
-
 ;-----------------------------------------------------------------------------
-; void interp_4tap_vert_%1_32x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
 ;-----------------------------------------------------------------------------
-%macro FILTER_V4_W32_sse2 2
-INIT_XMM sse2
-cglobal interp_4tap_vert_%1_32x%2, 4, 6, 10
-    mov         r4d,       r4m
-    sub         r0,        r1
-    shl         r4d,       5
-    pxor        m9,        m9
-
-%ifidn %1,pp
-    mova        m6,        [pw_32]
-%elifidn %1,ps
-    mova        m6,        [pw_2000]
-    add         r3d,       r3d
-%endif
-
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeffV]
-    mova        m1,        [r5 + r4]
-    mova        m0,        [r5 + r4 + 16]
-%else
-    mova        m1,        [tab_ChromaCoeffV + r4]
-    mova        m0,        [tab_ChromaCoeffV + r4 + 16]
-%endif
-
-    mov         r4d,       %2
+INIT_YMM avx2
+cglobal filterPixelToShort_16x8, 3, 6, 2
+    mov             r3d, r3m
+    add             r3d, r3d
+    lea             r4, [r1 * 3]
+    lea             r5, [r3 * 3]
 
-.loop:
-    movu        m2,        [r0]
-    movu        m3,        [r0 + r1]
-
-    punpcklbw   m4,        m2,        m3
-    punpckhbw   m2,        m3
-
-    movhlps     m8,        m4
-    punpcklbw   m4,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m4,        m1
-    pmaddwd     m8,        m1
-    packssdw    m4,        m8
-
-    movhlps     m8,        m2
-    punpcklbw   m2,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m2,        m1
-    pmaddwd     m8,        m1
-    packssdw    m2,        m8
-
-    lea         r5,        [r0 + 2 * r1]
-    movu        m3,        [r5]
-    movu        m5,        [r5 + r1]
-
-    punpcklbw   m7,        m3,        m5
-    punpckhbw   m3,        m5
-
-    movhlps     m8,        m7
-    punpcklbw   m7,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m7,        m0
-    pmaddwd     m8,        m0
-    packssdw    m7,        m8
-
-    movhlps     m8,        m3
-    punpcklbw   m3,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m3,        m0
-    pmaddwd     m8,        m0
-    packssdw    m3,        m8
-
-    paddw       m4,        m7
-    paddw       m2,        m3
+    ; load constant
+    vbroadcasti128  m1, [pw_2000]
 
-%ifidn %1,pp
-    paddw       m4,        m6
-    psraw       m4,        6
-    paddw       m2,        m6
-    psraw       m2,        6
-
-    packuswb    m4,        m2
-    movu        [r2],      m4
-%elifidn %1,ps
-    psubw       m4,        m6
-    psubw       m2,        m6
-    movu        [r2],      m4
-    movu        [r2 + 16], m2
-%endif
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-    movu        m2,        [r0 + 16]
-    movu        m3,        [r0 + r1 + 16]
-
-    punpcklbw   m4,        m2,        m3
-    punpckhbw   m2,        m3
-
-    movhlps     m8,        m4
-    punpcklbw   m4,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m4,        m1
-    pmaddwd     m8,        m1
-    packssdw    m4,        m8
-
-    movhlps     m8,        m2
-    punpcklbw   m2,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m2,        m1
-    pmaddwd     m8,        m1
-    packssdw    m2,        m8
-
-    movu        m3,        [r5 + 16]
-    movu        m5,        [r5 + r1 + 16]
-
-    punpcklbw   m7,        m3,        m5
-    punpckhbw   m3,        m5
-
-    movhlps     m8,        m7
-    punpcklbw   m7,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m7,        m0
-    pmaddwd     m8,        m0
-    packssdw    m7,        m8
-
-    movhlps     m8,        m3
-    punpcklbw   m3,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m3,        m0
-    pmaddwd     m8,        m0
-    packssdw    m3,        m8
-
-    paddw       m4,        m7
-    paddw       m2,        m3
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-%ifidn %1,pp
-    paddw       m4,        m6
-    psraw       m4,        6
-    paddw       m2,        m6
-    psraw       m2,        6
-
-    packuswb    m4,        m2
-    movu        [r2 + 16], m4
-%elifidn %1,ps
-    psubw       m4,        m6
-    psubw       m2,        m6
-    movu        [r2 + 32], m4
-    movu        [r2 + 48], m2
-%endif
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-    lea         r0,        [r0 + r1]
-    lea         r2,        [r2 + r3]
-    dec         r4
-    jnz        .loop
-    RET
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
 
-%endmacro
+    lea             r0, [r0 + r1 * 4]
+    lea             r2, [r2 + r3 * 4]
 
-%if ARCH_X86_64
-    FILTER_V4_W32_sse2 pp, 8
-    FILTER_V4_W32_sse2 pp, 16
-    FILTER_V4_W32_sse2 pp, 24
-    FILTER_V4_W32_sse2 pp, 32
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-    FILTER_V4_W32_sse2 pp, 48
-    FILTER_V4_W32_sse2 pp, 64
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-    FILTER_V4_W32_sse2 ps, 8
-    FILTER_V4_W32_sse2 ps, 16
-    FILTER_V4_W32_sse2 ps, 24
-    FILTER_V4_W32_sse2 ps, 32
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-    FILTER_V4_W32_sse2 ps, 48
-    FILTER_V4_W32_sse2 ps, 64
-%endif
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
+    RET
 
 ;-----------------------------------------------------------------------------
-; void interp_4tap_vert_%1_%2x%3(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
 ;-----------------------------------------------------------------------------
-%macro FILTER_V4_W16n_H2_sse2 3
-INIT_XMM sse2
-cglobal interp_4tap_vert_%1_%2x%3, 4, 7, 11
-    mov         r4d,       r4m
-    sub         r0,        r1
-    shl         r4d,       5
-    pxor        m9,        m9
-
-%ifidn %1,pp
-    mova        m7,        [pw_32]
-%elifidn %1,ps
-    mova        m7,        [pw_2000]
-    add         r3d,       r3d
-%endif
+INIT_YMM avx2
+cglobal filterPixelToShort_16x12, 3, 6, 2
+    mov             r3d, r3m
+    add             r3d, r3d
+    lea             r4, [r1 * 3]
+    lea             r5, [r3 * 3]
 
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeffV]
-    mova        m1,        [r5 + r4]
-    mova        m0,        [r5 + r4 + 16]
-%else
-    mova        m1,        [tab_ChromaCoeffV + r4]
-    mova        m0,        [tab_ChromaCoeffV + r4 + 16]
-%endif
+    ; load constant
+    vbroadcasti128  m1, [pw_2000]
 
-    mov         r4d,       %3/2
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-.loop:
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-    mov         r6d,       %2/16
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-.loopW:
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
 
-    movu        m2,        [r0]
-    movu        m3,        [r0 + r1]
-
-    punpcklbw   m4,        m2,        m3
-    punpckhbw   m2,        m3
-
-    movhlps     m8,        m4
-    punpcklbw   m4,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m4,        m1
-    pmaddwd     m8,        m1
-    packssdw    m4,        m8
-
-    movhlps     m8,        m2
-    punpcklbw   m2,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m2,        m1
-    pmaddwd     m8,        m1
-    packssdw    m2,        m8
-
-    lea         r5,        [r0 + 2 * r1]
-    movu        m5,        [r5]
-    movu        m6,        [r5 + r1]
-
-    punpckhbw   m10,        m5,        m6
-    movhlps     m8,        m10
-    punpcklbw   m10,       m9
-    punpcklbw   m8,        m9
-    pmaddwd     m10,       m0
-    pmaddwd     m8,        m0
-    packssdw    m10,       m8
-    paddw       m2,        m10
-
-    punpcklbw   m10,        m5,        m6
-    movhlps     m8,        m10
-    punpcklbw   m10,       m9
-    punpcklbw   m8,        m9
-    pmaddwd     m10,       m0
-    pmaddwd     m8,        m0
-    packssdw    m10,       m8
-    paddw       m4,        m10
+    lea             r0, [r0 + r1 * 4]
+    lea             r2, [r2 + r3 * 4]
 
-%ifidn %1,pp
-    paddw       m4,        m7
-    psraw       m4,        6
-    paddw       m2,        m7
-    psraw       m2,        6
-
-    packuswb    m4,        m2
-    movu        [r2],      m4
-%elifidn %1,ps
-    psubw       m4,        m7
-    psubw       m2,        m7
-    movu        [r2],      m4
-    movu        [r2 + 16], m2
-%endif
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-    punpcklbw   m4,        m3,        m5
-    punpckhbw   m3,        m5
-
-    movhlps     m8,        m4
-    punpcklbw   m4,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m4,        m1
-    pmaddwd     m8,        m1
-    packssdw    m4,        m8
-
-    movhlps     m8,        m3
-    punpcklbw   m3,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m3,        m1
-    pmaddwd     m8,        m1
-    packssdw    m3,        m8
-
-    movu        m5,        [r5 + 2 * r1]
-
-    punpcklbw   m2,        m6,        m5
-    punpckhbw   m6,        m5
-
-    movhlps     m8,        m2
-    punpcklbw   m2,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m2,        m0
-    pmaddwd     m8,        m0
-    packssdw    m2,        m8
-
-    movhlps     m8,        m6
-    punpcklbw   m6,        m9
-    punpcklbw   m8,        m9
-    pmaddwd     m6,        m0
-    pmaddwd     m8,        m0
-    packssdw    m6,        m8
-
-    paddw       m4,        m2
-    paddw       m3,        m6
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-%ifidn %1,pp
-    paddw       m4,        m7
-    psraw       m4,        6
-    paddw       m3,        m7
-    psraw       m3,        6
-
-    packuswb    m4,        m3
-    movu        [r2 + r3], m4
-    add         r2,        16
-%elifidn %1,ps
-    psubw       m4,        m7
-    psubw       m3,        m7
-    movu        [r2 + r3], m4
-    movu        [r2 + r3 + 16], m3
-    add         r2,        32
-%endif
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-    add         r0,        16
-    dec         r6d
-    jnz         .loopW
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
 
-    lea         r0,        [r0 + r1 * 2 - %2]
+    lea             r0, [r0 + r1 * 4]
+    lea             r2, [r2 + r3 * 4]
 
-%ifidn %1,pp
-    lea         r2,        [r2 + r3 * 2 - %2]
-%elifidn %1,ps
-    lea         r2,        [r2 + r3 * 2 - (%2 * 2)]
-%endif
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-    dec         r4d
-    jnz        .loop
-    RET
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-%endmacro
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-%if ARCH_X86_64
-    FILTER_V4_W16n_H2_sse2 pp, 64, 64
-    FILTER_V4_W16n_H2_sse2 pp, 64, 32
-    FILTER_V4_W16n_H2_sse2 pp, 64, 48
-    FILTER_V4_W16n_H2_sse2 pp, 48, 64
-    FILTER_V4_W16n_H2_sse2 pp, 64, 16
-    FILTER_V4_W16n_H2_sse2 ps, 64, 64
-    FILTER_V4_W16n_H2_sse2 ps, 64, 32
-    FILTER_V4_W16n_H2_sse2 ps, 64, 48
-    FILTER_V4_W16n_H2_sse2 ps, 48, 64
-    FILTER_V4_W16n_H2_sse2 ps, 64, 16
-%endif
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
+    RET
 
-%macro FILTER_P2S_2_4_sse2 1
-    movd        m2,     [r0 + %1]
-    movd        m3,     [r0 + r1 + %1]
-    punpcklwd   m2,     m3
-    movd        m3,     [r0 + r1 * 2 + %1]
-    movd        m4,     [r0 + r4 + %1]
-    punpcklwd   m3,     m4
-    punpckldq   m2,     m3
-    punpcklbw   m2,     m0
-    psllw       m2,     6
-    psubw       m2,     m1
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+INIT_YMM avx2
+cglobal filterPixelToShort_16x16, 3, 6, 2
+    mov             r3d, r3m
+    add             r3d, r3d
+    lea             r4, [r1 * 3]
+    lea             r5, [r3 * 3]
 
-    movd        [r2 + r3 * 0 + %1 * 2], m2
-    psrldq      m2,     4
-    movd        [r2 + r3 * 1 + %1 * 2], m2
-    psrldq      m2,     4
-    movd        [r2 + r3 * 2 + %1 * 2], m2
-    psrldq      m2,     4
-    movd        [r2 + r5 + %1 * 2], m2
-%endmacro
+    ; load constant
+    vbroadcasti128  m1, [pw_2000]
 
-%macro FILTER_P2S_4_4_sse2 1
-    movd        m2,     [r0 + %1]
-    movd        m3,     [r0 + r1 + %1]
-    movd        m4,     [r0 + r1 * 2 + %1]
-    movd        m5,     [r0 + r4 + %1]
-    punpckldq   m2,     m3
-    punpcklbw   m2,     m0
-    punpckldq   m4,     m5
-    punpcklbw   m4,     m0
-    psllw       m2,     6
-    psllw       m4,     6
-    psubw       m2,     m1
-    psubw       m4,     m1
-    movh        [r2 + r3 * 0 + %1 * 2], m2
-    movh        [r2 + r3 * 2 + %1 * 2], m4
-    movhps      [r2 + r3 * 1 + %1 * 2], m2
-    movhps      [r2 + r5 + %1 * 2], m4
-%endmacro
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-%macro FILTER_P2S_4_2_sse2 0
-    movd        m2,     [r0]
-    movd        m3,     [r0 + r1]
-    punpckldq   m2,     m3
-    punpcklbw   m2,     m0
-    psllw       m2,     6
-    psubw       m2,     [pw_8192]
-    movh        [r2],   m2
-    movhps      [r2 + r3 * 2], m2
-%endmacro
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-%macro FILTER_P2S_8_4_sse2 1
-    movh        m2,     [r0 + %1]
-    movh        m3,     [r0 + r1 + %1]
-    movh        m4,     [r0 + r1 * 2 + %1]
-    movh        m5,     [r0 + r4 + %1]
-    punpcklbw   m2,     m0
-    punpcklbw   m3,     m0
-    punpcklbw   m5,     m0
-    punpcklbw   m4,     m0
-    psllw       m2,     6
-    psllw       m3,     6
-    psllw       m5,     6
-    psllw       m4,     6
-    psubw       m2,     m1
-    psubw       m3,     m1
-    psubw       m4,     m1
-    psubw       m5,     m1
-    movu        [r2 + r3 * 0 + %1 * 2], m2
-    movu        [r2 + r3 * 1 + %1 * 2], m3
-    movu        [r2 + r3 * 2 + %1 * 2], m4
-    movu        [r2 + r5 + %1 * 2], m5
-%endmacro
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-%macro FILTER_P2S_8_2_sse2 1
-    movh        m2,     [r0 + %1]
-    movh        m3,     [r0 + r1 + %1]
-    punpcklbw   m2,     m0
-    punpcklbw   m3,     m0
-    psllw       m2,     6
-    psllw       m3,     6
-    psubw       m2,     m1
-    psubw       m3,     m1
-    movu        [r2 + r3 * 0 + %1 * 2], m2
-    movu        [r2 + r3 * 1 + %1 * 2], m3
-%endmacro
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
 
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride)
-;-----------------------------------------------------------------------------
-%macro FILTER_PIX_TO_SHORT_sse2 2
-INIT_XMM sse2
-cglobal filterPixelToShort_%1x%2, 4, 6, 6
-    pxor        m0,     m0
-%if %2 == 2
-%if %1 == 4
-    FILTER_P2S_4_2_sse2
-%elif %1 == 8
-    add        r3d, r3d
-    mova       m1, [pw_8192]
-    FILTER_P2S_8_2_sse2 0
-%endif
-%else
-    add        r3d, r3d
-    mova       m1, [pw_8192]
-    lea        r4, [r1 * 3]
-    lea        r5, [r3 * 3]
-%assign y 1
-%rep %2/4
-%assign x 0
-%rep %1/8
-    FILTER_P2S_8_4_sse2 x
-%if %2 == 6
-    lea         r0,     [r0 + 4 * r1]
-    lea         r2,     [r2 + 4 * r3]
-    FILTER_P2S_8_2_sse2 x
-%endif
-%assign x x+8
-%endrep
-%rep (%1 % 8)/4
-    FILTER_P2S_4_4_sse2 x
-%assign x x+4
-%endrep
-%rep (%1 % 4)/2
-    FILTER_P2S_2_4_sse2 x
-%endrep
-%if y < %2/4
-    lea         r0,     [r0 + 4 * r1]
-    lea         r2,     [r2 + 4 * r3]
-%assign y y+1
-%endif
-%endrep
-%endif
-RET
-%endmacro
+    lea             r0, [r0 + r1 * 4]
+    lea             r2, [r2 + r3 * 4]
 
-    FILTER_PIX_TO_SHORT_sse2 2, 4
-    FILTER_PIX_TO_SHORT_sse2 2, 8
-    FILTER_PIX_TO_SHORT_sse2 2, 16
-    FILTER_PIX_TO_SHORT_sse2 4, 2
-    FILTER_PIX_TO_SHORT_sse2 4, 4
-    FILTER_PIX_TO_SHORT_sse2 4, 8
-    FILTER_PIX_TO_SHORT_sse2 4, 16
-    FILTER_PIX_TO_SHORT_sse2 4, 32
-    FILTER_PIX_TO_SHORT_sse2 6, 8
-    FILTER_PIX_TO_SHORT_sse2 6, 16
-    FILTER_PIX_TO_SHORT_sse2 8, 2
-    FILTER_PIX_TO_SHORT_sse2 8, 4
-    FILTER_PIX_TO_SHORT_sse2 8, 6
-    FILTER_PIX_TO_SHORT_sse2 8, 8
-    FILTER_PIX_TO_SHORT_sse2 8, 12
-    FILTER_PIX_TO_SHORT_sse2 8, 16
-    FILTER_PIX_TO_SHORT_sse2 8, 32
-    FILTER_PIX_TO_SHORT_sse2 8, 64
-    FILTER_PIX_TO_SHORT_sse2 12, 16
-    FILTER_PIX_TO_SHORT_sse2 12, 32
-    FILTER_PIX_TO_SHORT_sse2 16, 4
-    FILTER_PIX_TO_SHORT_sse2 16, 8
-    FILTER_PIX_TO_SHORT_sse2 16, 12
-    FILTER_PIX_TO_SHORT_sse2 16, 16
-    FILTER_PIX_TO_SHORT_sse2 16, 24
-    FILTER_PIX_TO_SHORT_sse2 16, 32
-    FILTER_PIX_TO_SHORT_sse2 16, 64
-    FILTER_PIX_TO_SHORT_sse2 24, 32
-    FILTER_PIX_TO_SHORT_sse2 24, 64
-    FILTER_PIX_TO_SHORT_sse2 32, 8
-    FILTER_PIX_TO_SHORT_sse2 32, 16
-    FILTER_PIX_TO_SHORT_sse2 32, 24
-    FILTER_PIX_TO_SHORT_sse2 32, 32
-    FILTER_PIX_TO_SHORT_sse2 32, 48
-    FILTER_PIX_TO_SHORT_sse2 32, 64
-    FILTER_PIX_TO_SHORT_sse2 48, 64
-    FILTER_PIX_TO_SHORT_sse2 64, 16
-    FILTER_PIX_TO_SHORT_sse2 64, 32
-    FILTER_PIX_TO_SHORT_sse2 64, 48
-    FILTER_PIX_TO_SHORT_sse2 64, 64
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-%macro FILTER_H8_W8 7-8   ; t0, t1, t2, t3, coef, c512, src, dst
-    movu        %1, %7
-    pshufb      %2, %1, [tab_Lm +  0]
-    pmaddubsw   %2, %5
-    pshufb      %3, %1, [tab_Lm + 16]
-    pmaddubsw   %3, %5
-    phaddw      %2, %3
-    pshufb      %4, %1, [tab_Lm + 32]
-    pmaddubsw   %4, %5
-    pshufb      %1, %1, [tab_Lm + 48]
-    pmaddubsw   %1, %5
-    phaddw      %4, %1
-    phaddw      %2, %4
-  %if %0 == 8
-    pmulhrsw    %2, %6
-    packuswb    %2, %2
-    movh        %8, %2
-  %endif
-%endmacro
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-;-----------------------------------------------------------------------------
-; Interpolate HV
-;-----------------------------------------------------------------------------
-%macro FILTER_HV8_START 7 ; (t0, t1, t2, t3, t4, off_src, off_coeff) -> (t3, t5), (t4, t1), [2]
-    mova        %5, [r0 +  (%6 + 0) * 16]
-    mova        %1, [r0 +  (%6 + 1) * 16]
-    mova        %2, [r0 +  (%6 + 2) * 16]
-    punpcklwd   %3, %5, %1
-    punpckhwd   %5, %1
-    pmaddwd     %3, [r5 + (%7) * 16]   ; R3 = L[0+1] -- Row 0
-    pmaddwd     %5, [r5 + (%7) * 16]   ; R0 = H[0+1]
-    punpcklwd   %4, %1, %2
-    punpckhwd   %1, %2
-    pmaddwd     %4, [r5 + (%7) * 16]   ; R4 = L[1+2] -- Row 1
-    pmaddwd     %1, [r5 + (%7) * 16]   ; R1 = H[1+2]
-%endmacro ; FILTER_HV8_START
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-%macro FILTER_HV8_MID 10 ; (Row3, prevRow, sum0L, sum1L, sum0H, sum1H, t6, t7, off_src, off_coeff) -> [6]
-    mova        %8, [r0 +  (%9 + 0) * 16]
-    mova        %1, [r0 +  (%9 + 1) * 16]
-    punpcklwd   %7, %2, %8
-    punpckhwd   %2, %8
-    pmaddwd     %7, [r5 + %10 * 16]
-    pmaddwd     %2, [r5 + %10 * 16]
-    paddd       %3, %7              ; R3 = L[0+1+2+3] -- Row 0
-    paddd       %5, %2              ; R0 = H[0+1+2+3]
-    punpcklwd   %7, %8, %1
-    punpckhwd   %8, %1
-    pmaddwd     %7, [r5 + %10 * 16]
-    pmaddwd     %8, [r5 + %10 * 16]
-    paddd       %4, %7              ; R4 = L[1+2+3+4] -- Row 1
-    paddd       %6, %8              ; R1 = H[1+2+3+4]
-%endmacro ; FILTER_HV8_MID
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
 
-; Round and Saturate
-%macro FILTER_HV8_END 4 ; output in [1, 3]
-    paddd       %1, [pd_526336]
-    paddd       %2, [pd_526336]
-    paddd       %3, [pd_526336]
-    paddd       %4, [pd_526336]
-    psrad       %1, 12
-    psrad       %2, 12
-    psrad       %3, 12
-    psrad       %4, 12
-    packssdw    %1, %2
-    packssdw    %3, %4
+    lea             r0, [r0 + r1 * 4]
+    lea             r2, [r2 + r3 * 4]
 
-    ; TODO: is merge better? I think this way is short dependency link
-    packuswb    %1, %3
-%endmacro ; FILTER_HV8_END
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-;-----------------------------------------------------------------------------
-; void interp_8tap_hv_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int idxX, int idxY)
-;-----------------------------------------------------------------------------
-INIT_XMM ssse3
-cglobal interp_8tap_hv_pp_8x8, 4, 7, 8, 0-15*16
-%define coef        m7
-%define stk_buf     rsp
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-    mov         r4d,        r4m
-    mov         r5d,        r5m
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-%ifdef PIC
-    lea         r6,         [tab_LumaCoeff]
-    movh        coef,       [r6 + r4 * 8]
-%else
-    movh        coef,       [tab_LumaCoeff + r4 * 8]
-%endif
-    punpcklqdq  coef,       coef
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
 
-    ; move to row -3
-    lea         r6,         [r1 + r1 * 2]
-    sub         r0,         r6
+    lea             r0, [r0 + r1 * 4]
+    lea             r2, [r2 + r3 * 4]
 
-    xor         r6,         r6
-    mov         r4,         rsp
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-.loopH:
-    FILTER_H8_W8 m0, m1, m2, m3, coef, [pw_512], [r0 - 3]
-    psubw       m1,         [pw_2000]
-    mova        [r4],       m1
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-    add         r0,         r1
-    add         r4,         16
-    inc         r6
-    cmp         r6,         8+7
-    jnz         .loopH
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-    ; ready to phase V
-    ; Here all of mN is free
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
+    RET
 
-    ; load coeff table
-    shl         r5,         6
-    lea         r6,         [tab_LumaCoeffV]
-    lea         r5,         [r5 + r6]
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+INIT_YMM avx2
+cglobal filterPixelToShort_16x24, 3, 7, 2
+    mov             r3d, r3m
+    add             r3d, r3d
+    lea             r4, [r1 * 3]
+    lea             r5, [r3 * 3]
+    mov             r6d, 3
 
-    ; load intermedia buffer
-    mov         r0,         stk_buf
+    ; load constant
+    vbroadcasti128  m1, [pw_2000]
+.loop:
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-    ; register mapping
-    ; r0 - src
-    ; r5 - coeff
-    ; r6 - loop_i
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-    ; let's go
-    xor         r6,         r6
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-    ; TODO: this loop have more than 70 instructions, I think it is more than Intel loop decode cache
-.loopV:
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
 
-    FILTER_HV8_START    m1, m2, m3, m4, m0,             0, 0
-    FILTER_HV8_MID      m6, m2, m3, m4, m0, m1, m7, m5, 3, 1
-    FILTER_HV8_MID      m5, m6, m3, m4, m0, m1, m7, m2, 5, 2
-    FILTER_HV8_MID      m6, m5, m3, m4, m0, m1, m7, m2, 7, 3
-    FILTER_HV8_END      m3, m0, m4, m1
+    lea             r0, [r0 + r1 * 4]
+    lea             r2, [r2 + r3 * 4]
 
-    movh        [r2],       m3
-    movhps      [r2 + r3],  m3
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-    lea         r0,         [r0 + 16 * 2]
-    lea         r2,         [r2 + r3 * 2]
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-    inc         r6
-    cmp         r6,         8/2
-    jnz         .loopV
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
+
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
+
+    lea             r0, [r0 + r1 * 4]
+    lea             r2, [r2 + r3 * 4]
 
+    dec             r6d
+    jnz             .loop
     RET
 
 ;-----------------------------------------------------------------------------
-; void interp_8tap_hv_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int idxX, int idxY)
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
 ;-----------------------------------------------------------------------------
-INIT_XMM sse3
-cglobal interp_8tap_hv_pp_8x8, 4, 7, 8, 0-15*16
-    mov         r4d,        r4m
-    mov         r5d,        r5m
-    add         r4d,        r4d
-    pxor        m6,         m6
-
-%ifdef PIC
-    lea         r6,         [tabw_LumaCoeff]
-    mova        m3,         [r6 + r4 * 8]
-%else
-    mova        m3,         [tabw_LumaCoeff + r4 * 8]
-%endif
+%macro P2S_H_16xN_avx2 1
+INIT_YMM avx2
+cglobal filterPixelToShort_16x%1, 3, 7, 2
+    mov             r3d, r3m
+    add             r3d, r3d
+    lea             r4, [r1 * 3]
+    lea             r5, [r3 * 3]
+    mov             r6d, %1/16
 
-    ; move to row -3
-    lea         r6,         [r1 + r1 * 2]
-    sub         r0,         r6
+    ; load constant
+    vbroadcasti128  m1, [pw_2000]
+.loop:
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-    mov         r4,         rsp
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-%assign x 0     ;needed for FILTER_H8_W8_sse2 macro
-%assign y 1
-%rep 15
-    FILTER_H8_W8_sse2
-    psubw       m1,         [pw_2000]
-    mova        [r4],       m1
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-%if y < 15
-    add         r0,         r1
-    add         r4,         16
-%endif
-%assign y y+1
-%endrep
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
 
-    ; ready to phase V
-    ; Here all of mN is free
+    lea             r0, [r0 + r1 * 4]
+    lea             r2, [r2 + r3 * 4]
 
-    ; load coeff table
-    shl         r5,         6
-    lea         r6,         [tab_LumaCoeffV]
-    lea         r5,         [r5 + r6]
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-    ; load intermedia buffer
-    mov         r0,         rsp
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-    ; register mapping
-    ; r0 - src
-    ; r5 - coeff
-
-    ; let's go
-%assign y 1
-%rep 4
-    FILTER_HV8_START    m1, m2, m3, m4, m0,             0, 0
-    FILTER_HV8_MID      m6, m2, m3, m4, m0, m1, m7, m5, 3, 1
-    FILTER_HV8_MID      m5, m6, m3, m4, m0, m1, m7, m2, 5, 2
-    FILTER_HV8_MID      m6, m5, m3, m4, m0, m1, m7, m2, 7, 3
-    FILTER_HV8_END      m3, m0, m4, m1
-
-    movh        [r2],       m3
-    movhps      [r2 + r3],  m3
-
-%if y < 4
-    lea         r0,         [r0 + 16 * 2]
-    lea         r2,         [r2 + r3 * 2]
-%endif
-%assign y y+1
-%endrep
-    RET
-
-;-----------------------------------------------------------------------------
-;void interp_4tap_vert_pp_2x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-INIT_XMM sse4
-cglobal interp_4tap_vert_pp_2x4, 4, 6, 8
-
-    mov         r4d,       r4m
-    sub         r0,        r1
-
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeff]
-    movd        m0,        [r5 + r4 * 4]
-%else
-    movd        m0,        [tab_ChromaCoeff + r4 * 4]
-%endif
-    lea         r4,        [r1 * 3]
-    lea         r5,        [r0 + 4 * r1]
-    pshufb      m0,        [tab_Cm]
-    mova        m1,        [pw_512]
-
-    movd        m2,        [r0]
-    movd        m3,        [r0 + r1]
-    movd        m4,        [r0 + 2 * r1]
-    movd        m5,        [r0 + r4]
-
-    punpcklbw   m2,        m3
-    punpcklbw   m6,        m4,        m5
-    punpcklbw   m2,        m6
-
-    pmaddubsw   m2,        m0
-
-    movd        m6,        [r5]
-
-    punpcklbw   m3,        m4
-    punpcklbw   m7,        m5,        m6
-    punpcklbw   m3,        m7
-
-    pmaddubsw   m3,        m0
-
-    phaddw      m2,        m3
-
-    pmulhrsw    m2,        m1
-
-    movd        m7,        [r5 + r1]
-
-    punpcklbw   m4,        m5
-    punpcklbw   m3,        m6,        m7
-    punpcklbw   m4,        m3
-
-    pmaddubsw   m4,        m0
-
-    movd        m3,        [r5 + 2 * r1]
-
-    punpcklbw   m5,        m6
-    punpcklbw   m7,        m3
-    punpcklbw   m5,        m7
-
-    pmaddubsw   m5,        m0
-
-    phaddw      m4,        m5
-
-    pmulhrsw    m4,        m1
-    packuswb    m2,        m4
-
-    pextrw      [r2],      m2, 0
-    pextrw      [r2 + r3], m2, 2
-    lea         r2,        [r2 + 2 * r3]
-    pextrw      [r2],      m2, 4
-    pextrw      [r2 + r3], m2, 6
-
-    RET
-
-%macro FILTER_VER_CHROMA_AVX2_2x4 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_2x4, 4, 6, 2
-    mov             r4d, r4m
-    shl             r4d, 5
-    sub             r0, r1
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeff_V]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeff_V + r4]
-%endif
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-    lea             r4, [r1 * 3]
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
 
-    pinsrw          xm1, [r0], 0
-    pinsrw          xm1, [r0 + r1], 1
-    pinsrw          xm1, [r0 + r1 * 2], 2
-    pinsrw          xm1, [r0 + r4], 3
     lea             r0, [r0 + r1 * 4]
-    pinsrw          xm1, [r0], 4
-    pinsrw          xm1, [r0 + r1], 5
-    pinsrw          xm1, [r0 + r1 * 2], 6
-
-    pshufb          xm0, xm1, [interp_vert_shuf]
-    pshufb          xm1, [interp_vert_shuf + 32]
-    vinserti128     m0, m0, xm1, 1
-    pmaddubsw       m0, [r5]
-    vextracti128    xm1, m0, 1
-    paddw           xm0, xm1
-%ifidn %1,pp
-    pmulhrsw        xm0, [pw_512]
-    packuswb        xm0, xm0
-    lea             r4, [r3 * 3]
-    pextrw          [r2], xm0, 0
-    pextrw          [r2 + r3], xm0, 1
-    pextrw          [r2 + r3 * 2], xm0, 2
-    pextrw          [r2 + r4], xm0, 3
-%else
-    add             r3d, r3d
-    lea             r4, [r3 * 3]
-    psubw           xm0, [pw_2000]
-    movd            [r2], xm0
-    pextrd          [r2 + r3], xm0, 1
-    pextrd          [r2 + r3 * 2], xm0, 2
-    pextrd          [r2 + r4], xm0, 3
-%endif
-    RET
-%endmacro
+    lea             r2, [r2 + r3 * 4]
 
-    FILTER_VER_CHROMA_AVX2_2x4 pp
-    FILTER_VER_CHROMA_AVX2_2x4 ps
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-%macro FILTER_VER_CHROMA_AVX2_2x8 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_2x8, 4, 6, 2
-    mov             r4d, r4m
-    shl             r4d, 6
-    sub             r0, r1
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-    lea             r4, [r1 * 3]
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
 
-    pinsrw          xm1, [r0], 0
-    pinsrw          xm1, [r0 + r1], 1
-    pinsrw          xm1, [r0 + r1 * 2], 2
-    pinsrw          xm1, [r0 + r4], 3
-    lea             r0, [r0 + r1 * 4]
-    pinsrw          xm1, [r0], 4
-    pinsrw          xm1, [r0 + r1], 5
-    pinsrw          xm1, [r0 + r1 * 2], 6
-    pinsrw          xm1, [r0 + r4], 7
-    movhlps         xm0, xm1
     lea             r0, [r0 + r1 * 4]
-    pinsrw          xm0, [r0], 4
-    pinsrw          xm0, [r0 + r1], 5
-    pinsrw          xm0, [r0 + r1 * 2], 6
-    vinserti128     m1, m1, xm0, 1
-
-    pshufb          m0, m1, [interp_vert_shuf]
-    pshufb          m1, [interp_vert_shuf + 32]
-    pmaddubsw       m0, [r5]
-    pmaddubsw       m1, [r5 + 1 * mmsize]
-    paddw           m0, m1
-%ifidn %1,pp
-    pmulhrsw        m0, [pw_512]
-    vextracti128    xm1, m0, 1
-    packuswb        xm0, xm1
-    lea             r4, [r3 * 3]
-    pextrw          [r2], xm0, 0
-    pextrw          [r2 + r3], xm0, 1
-    pextrw          [r2 + r3 * 2], xm0, 2
-    pextrw          [r2 + r4], xm0, 3
-    lea             r2, [r2 + r3 * 4]
-    pextrw          [r2], xm0, 4
-    pextrw          [r2 + r3], xm0, 5
-    pextrw          [r2 + r3 * 2], xm0, 6
-    pextrw          [r2 + r4], xm0, 7
-%else
-    add             r3d, r3d
-    lea             r4, [r3 * 3]
-    psubw           m0, [pw_2000]
-    vextracti128    xm1, m0, 1
-    movd            [r2], xm0
-    pextrd          [r2 + r3], xm0, 1
-    pextrd          [r2 + r3 * 2], xm0, 2
-    pextrd          [r2 + r4], xm0, 3
     lea             r2, [r2 + r3 * 4]
-    movd            [r2], xm1
-    pextrd          [r2 + r3], xm1, 1
-    pextrd          [r2 + r3 * 2], xm1, 2
-    pextrd          [r2 + r4], xm1, 3
-%endif
-    RET
-%endmacro
 
-    FILTER_VER_CHROMA_AVX2_2x8 pp
-    FILTER_VER_CHROMA_AVX2_2x8 ps
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
 
-%macro FILTER_VER_CHROMA_AVX2_2x16 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_2x16, 4, 6, 3
-    mov             r4d, r4m
-    shl             r4d, 6
-    sub             r0,  r1
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
 
-%ifdef PIC
-    lea             r5,  [tab_ChromaCoeffVer_32]
-    add             r5,  r4
-%else
-    lea             r5,  [tab_ChromaCoeffVer_32 + r4]
-%endif
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
 
-    lea             r4,  [r1 * 3]
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
 
-    movd            xm1, [r0]
-    pinsrw          xm1, [r0 + r1], 1
-    pinsrw          xm1, [r0 + r1 * 2], 2
-    pinsrw          xm1, [r0 + r4], 3
-    lea             r0,  [r0 + r1 * 4]
-    pinsrw          xm1, [r0], 4
-    pinsrw          xm1, [r0 + r1], 5
-    pinsrw          xm1, [r0 + r1 * 2], 6
-    pinsrw          xm1, [r0 + r4], 7
-    lea             r0,  [r0 + r1 * 4]
-    pinsrw          xm0, [r0], 4
-    pinsrw          xm0, [r0 + r1], 5
-    pinsrw          xm0, [r0 + r1 * 2], 6
-    pinsrw          xm0, [r0 + r4], 7
-    punpckhqdq      xm0, xm1, xm0
-    vinserti128     m1,  m1,  xm0,  1
-
-    pshufb          m2,  m1,  [interp_vert_shuf]
-    pshufb          m1,  [interp_vert_shuf + 32]
-    pmaddubsw       m2,  [r5]
-    pmaddubsw       m1,  [r5 + 1 * mmsize]
-    paddw           m2,  m1
-
-    lea             r0,  [r0 + r1 * 4]
-    pinsrw          xm1, [r0], 4
-    pinsrw          xm1, [r0 + r1], 5
-    pinsrw          xm1, [r0 + r1 * 2], 6
-    pinsrw          xm1, [r0 + r4], 7
-    punpckhqdq      xm1, xm0, xm1
-    lea             r0,  [r0 + r1 * 4]
-    pinsrw          xm0, [r0], 4
-    pinsrw          xm0, [r0 + r1], 5
-    pinsrw          xm0, [r0 + r1 * 2], 6
-    punpckhqdq      xm0, xm1, xm0
-    vinserti128     m1,  m1,  xm0,  1
-
-    pshufb          m0,  m1,  [interp_vert_shuf]
-    pshufb          m1,  [interp_vert_shuf + 32]
-    pmaddubsw       m0,  [r5]
-    pmaddubsw       m1,  [r5 + 1 * mmsize]
-    paddw           m0,  m1
-%ifidn %1,pp
-    mova            m1,  [pw_512]
-    pmulhrsw        m2,  m1
-    pmulhrsw        m0,  m1
-    packuswb        m2,  m0
-    lea             r4,  [r3 * 3]
-    pextrw          [r2], xm2, 0
-    pextrw          [r2 + r3], xm2, 1
-    pextrw          [r2 + r3 * 2], xm2, 2
-    pextrw          [r2 + r4], xm2, 3
-    vextracti128    xm0, m2, 1
-    lea             r2,  [r2 + r3 * 4]
-    pextrw          [r2], xm0, 0
-    pextrw          [r2 + r3], xm0, 1
-    pextrw          [r2 + r3 * 2], xm0, 2
-    pextrw          [r2 + r4], xm0, 3
-    lea             r2,  [r2 + r3 * 4]
-    pextrw          [r2], xm2, 4
-    pextrw          [r2 + r3], xm2, 5
-    pextrw          [r2 + r3 * 2], xm2, 6
-    pextrw          [r2 + r4], xm2, 7
-    lea             r2,  [r2 + r3 * 4]
-    pextrw          [r2], xm0, 4
-    pextrw          [r2 + r3], xm0, 5
-    pextrw          [r2 + r3 * 2], xm0, 6
-    pextrw          [r2 + r4], xm0, 7
-%else
-    add             r3d, r3d
-    lea             r4,  [r3 * 3]
-    vbroadcasti128  m1,  [pw_2000]
-    psubw           m2,  m1
-    psubw           m0,  m1
-    vextracti128    xm1, m2, 1
-    movd            [r2], xm2
-    pextrd          [r2 + r3], xm2, 1
-    pextrd          [r2 + r3 * 2], xm2, 2
-    pextrd          [r2 + r4], xm2, 3
+    lea             r0, [r0 + r1 * 4]
     lea             r2, [r2 + r3 * 4]
-    movd            [r2], xm1
-    pextrd          [r2 + r3], xm1, 1
-    pextrd          [r2 + r3 * 2], xm1, 2
-    pextrd          [r2 + r4], xm1, 3
-    vextracti128    xm1, m0, 1
-    lea             r2,  [r2 + r3 * 4]
-    movd            [r2], xm0
-    pextrd          [r2 + r3], xm0, 1
-    pextrd          [r2 + r3 * 2], xm0, 2
-    pextrd          [r2 + r4], xm0, 3
-    lea             r2,  [r2 + r3 * 4]
-    movd            [r2], xm1
-    pextrd          [r2 + r3], xm1, 1
-    pextrd          [r2 + r3 * 2], xm1, 2
-    pextrd          [r2 + r4], xm1, 3
-%endif
+
+    dec             r6d
+    jnz             .loop
     RET
 %endmacro
-
-    FILTER_VER_CHROMA_AVX2_2x16 pp
-    FILTER_VER_CHROMA_AVX2_2x16 ps
+P2S_H_16xN_avx2 32
+P2S_H_16xN_avx2 64
 
 ;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_2x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
 ;-----------------------------------------------------------------------------
-%macro FILTER_V4_W2_H4 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_pp_2x%2, 4, 6, 8
-
-    mov         r4d,       r4m
-    sub         r0,        r1
-
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeff]
-    movd        m0,        [r5 + r4 * 4]
-%else
-    movd        m0,        [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    pshufb      m0,        [tab_Cm]
+%macro P2S_H_32xN 1
+INIT_XMM ssse3
+cglobal filterPixelToShort_32x%1, 3, 7, 6
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r4, [r3 * 3]
+    lea         r5, [r1 * 3]
 
-    mova        m1,        [pw_512]
+    ; load height
+    mov         r6d, %1/4
 
-    mov         r4d,       %2
-    lea         r5,        [3 * r1]
+    ; load constant
+    mova        m4, [pb_128]
+    mova        m5, [tab_c_64_n64]
 
 .loop:
-    movd        m2,        [r0]
-    movd        m3,        [r0 + r1]
-    movd        m4,        [r0 + 2 * r1]
-    movd        m5,        [r0 + r5]
-
-    punpcklbw   m2,        m3
-    punpcklbw   m6,        m4,        m5
-    punpcklbw   m2,        m6
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-    pmaddubsw   m2,        m0
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-    lea         r0,        [r0 + 4 * r1]
-    movd        m6,        [r0]
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
 
-    punpcklbw   m3,        m4
-    punpcklbw   m7,        m5,        m6
-    punpcklbw   m3,        m7
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
 
-    pmaddubsw   m3,        m0
+    movu        [r2 + r3 * 0], m0
+    movu        [r2 + r3 * 1], m1
+    movu        [r2 + r3 * 2], m2
+    movu        [r2 + r4], m3
 
-    phaddw      m2,        m3
+    lea         r0, [r0 + 8]
 
-    pmulhrsw    m2,        m1
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-    movd        m7,        [r0 + r1]
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-    punpcklbw   m4,        m5
-    punpcklbw   m3,        m6,        m7
-    punpcklbw   m4,        m3
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
 
-    pmaddubsw   m4,        m0
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
 
-    movd        m3,        [r0 + 2 * r1]
+    movu        [r2 + r3 * 0 + 16], m0
+    movu        [r2 + r3 * 1 + 16], m1
+    movu        [r2 + r3 * 2 + 16], m2
+    movu        [r2 + r4 + 16], m3
 
-    punpcklbw   m5,        m6
-    punpcklbw   m7,        m3
-    punpcklbw   m5,        m7
+    lea         r0, [r0 + 8]
 
-    pmaddubsw   m5,        m0
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-    phaddw      m4,        m5
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-    pmulhrsw    m4,        m1
-    packuswb    m2,        m4
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
 
-    pextrw      [r2],      m2, 0
-    pextrw      [r2 + r3], m2, 2
-    lea         r2,        [r2 + 2 * r3]
-    pextrw      [r2],      m2, 4
-    pextrw      [r2 + r3], m2, 6
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
 
-    lea         r2,        [r2 + 2 * r3]
+    movu        [r2 + r3 * 0 + 32], m0
+    movu        [r2 + r3 * 1 + 32], m1
+    movu        [r2 + r3 * 2 + 32], m2
+    movu        [r2 + r4 + 32], m3
 
-    sub         r4,        4
-    jnz        .loop
-    RET
-%endmacro
+    lea         r0, [r0 + 8]
 
-    FILTER_V4_W2_H4 2, 8
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-    FILTER_V4_W2_H4 2, 16
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_4x2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-INIT_XMM sse4
-cglobal interp_4tap_vert_pp_4x2, 4, 6, 6
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
 
-    mov         r4d,       r4m
-    sub         r0,        r1
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
 
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeff]
-    movd        m0,        [r5 + r4 * 4]
-%else
-    movd        m0,        [tab_ChromaCoeff + r4 * 4]
-%endif
+    movu        [r2 + r3 * 0 + 48], m0
+    movu        [r2 + r3 * 1 + 48], m1
+    movu        [r2 + r3 * 2 + 48], m2
+    movu        [r2 + r4 + 48], m3
+
+    lea         r0, [r0 + r1 * 4 - 24]
+    lea         r2, [r2 + r3 * 4]
 
-    pshufb      m0,        [tab_Cm]
-    lea         r5,        [r0 + 2 * r1]
+    dec         r6d
+    jnz         .loop
+    RET
+%endmacro
+    P2S_H_32xN 32
+    P2S_H_32xN 8
+    P2S_H_32xN 16
+    P2S_H_32xN 24
+    P2S_H_32xN 64
+    P2S_H_32xN 48
 
-    movd        m2,        [r0]
-    movd        m3,        [r0 + r1]
-    movd        m4,        [r5]
-    movd        m5,        [r5 + r1]
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+%macro P2S_H_32xN_avx2 1
+INIT_YMM avx2
+cglobal filterPixelToShort_32x%1, 3, 7, 3
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r5, [r1 * 3]
+    lea         r6, [r3 * 3]
 
-    punpcklbw   m2,        m3
-    punpcklbw   m1,        m4,        m5
-    punpcklbw   m2,        m1
+    ; load height
+    mov         r4d, %1/4
 
-    pmaddubsw   m2,        m0
+    ; load constant
+    vpbroadcastd m2, [pw_2000]
 
-    movd        m1,        [r0 + 4 * r1]
+.loop:
+    pmovzxbw    m0, [r0 + 0 * mmsize/2]
+    pmovzxbw    m1, [r0 + 1 * mmsize/2]
+    psllw       m0, 6
+    psllw       m1, 6
+    psubw       m0, m2
+    psubw       m1, m2
+    movu        [r2 + 0 * mmsize], m0
+    movu        [r2 + 1 * mmsize], m1
 
-    punpcklbw   m3,        m4
-    punpcklbw   m5,        m1
-    punpcklbw   m3,        m5
+    pmovzxbw    m0, [r0 + r1 + 0 * mmsize/2]
+    pmovzxbw    m1, [r0 + r1 + 1 * mmsize/2]
+    psllw       m0, 6
+    psllw       m1, 6
+    psubw       m0, m2
+    psubw       m1, m2
+    movu        [r2 + r3 + 0 * mmsize], m0
+    movu        [r2 + r3 + 1 * mmsize], m1
 
-    pmaddubsw   m3,        m0
+    pmovzxbw    m0, [r0 + r1 * 2 + 0 * mmsize/2]
+    pmovzxbw    m1, [r0 + r1 * 2 + 1 * mmsize/2]
+    psllw       m0, 6
+    psllw       m1, 6
+    psubw       m0, m2
+    psubw       m1, m2
+    movu        [r2 + r3 * 2 + 0 * mmsize], m0
+    movu        [r2 + r3 * 2 + 1 * mmsize], m1
 
-    phaddw      m2,        m3
+    pmovzxbw    m0, [r0 + r5 + 0 * mmsize/2]
+    pmovzxbw    m1, [r0 + r5 + 1 * mmsize/2]
+    psllw       m0, 6
+    psllw       m1, 6
+    psubw       m0, m2
+    psubw       m1, m2
+    movu        [r2 + r6 + 0 * mmsize], m0
+    movu        [r2 + r6 + 1 * mmsize], m1
 
-    pmulhrsw    m2,        [pw_512]
-    packuswb    m2,        m2
-    movd        [r2],      m2
-    pextrd      [r2 + r3], m2,  1
+    lea         r0, [r0 + r1 * 4]
+    lea         r2, [r2 + r3 * 4]
 
+    dec         r4d
+    jnz        .loop
     RET
+%endmacro
+    P2S_H_32xN_avx2 32
+    P2S_H_32xN_avx2 8
+    P2S_H_32xN_avx2 16
+    P2S_H_32xN_avx2 24
+    P2S_H_32xN_avx2 64
+    P2S_H_32xN_avx2 48
 
-%macro FILTER_VER_CHROMA_AVX2_4x2 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_4x2, 4, 6, 4
-    mov             r4d, r4m
-    shl             r4d, 5
-    sub             r0, r1
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+%macro P2S_H_64xN 1
+INIT_XMM ssse3
+cglobal filterPixelToShort_64x%1, 3, 7, 6
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r4, [r3 * 3]
+    lea         r5, [r1 * 3]
 
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeff_V]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeff_V + r4]
-%endif
+    ; load height
+    mov         r6d, %1/4
 
-    lea             r4, [r1 * 3]
+    ; load constant
+    mova        m4, [pb_128]
+    mova        m5, [tab_c_64_n64]
 
-    movd            xm1, [r0]
-    movd            xm2, [r0 + r1]
-    punpcklbw       xm1, xm2
-    movd            xm3, [r0 + r1 * 2]
-    punpcklbw       xm2, xm3
-    movlhps         xm1, xm2
-    movd            xm0, [r0 + r4]
-    punpcklbw       xm3, xm0
-    movd            xm2, [r0 + r1 * 4]
-    punpcklbw       xm0, xm2
-    movlhps         xm3, xm0
-    vinserti128     m1, m1, xm3, 1                          ; m1 = row[x x x 4 3 2 1 0]
+.loop:
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-    pmaddubsw       m1, [r5]
-    vextracti128    xm3, m1, 1
-    paddw           xm1, xm3
-%ifidn %1,pp
-    pmulhrsw        xm1, [pw_512]
-    packuswb        xm1, xm1
-    movd            [r2], xm1
-    pextrd          [r2 + r3], xm1, 1
-%else
-    add             r3d, r3d
-    psubw           xm1, [pw_2000]
-    movq            [r2], xm1
-    movhps          [r2 + r3], xm1
-%endif
-    RET
-%endmacro
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-    FILTER_VER_CHROMA_AVX2_4x2 pp
-    FILTER_VER_CHROMA_AVX2_4x2 ps
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
 
-;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_4x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-INIT_XMM sse4
-cglobal interp_4tap_vert_pp_4x4, 4, 6, 8
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
 
-    mov         r4d,       r4m
-    sub         r0,        r1
+    movu        [r2 + r3 * 0], m0
+    movu        [r2 + r3 * 1], m1
+    movu        [r2 + r3 * 2], m2
+    movu        [r2 + r4], m3
 
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeff]
-    movd        m0,        [r5 + r4 * 4]
-%else
-    movd        m0,        [tab_ChromaCoeff + r4 * 4]
-%endif
+    lea         r0, [r0 + 8]
 
-    pshufb      m0,        [tab_Cm]
-    mova        m1,        [pw_512]
-    lea         r5,        [r0 + 4 * r1]
-    lea         r4,        [r1 * 3]
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-    movd        m2,        [r0]
-    movd        m3,        [r0 + r1]
-    movd        m4,        [r0 + 2 * r1]
-    movd        m5,        [r0 + r4]
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-    punpcklbw   m2,        m3
-    punpcklbw   m6,        m4,        m5
-    punpcklbw   m2,        m6
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
 
-    pmaddubsw   m2,        m0
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
 
-    movd        m6,        [r5]
+    movu        [r2 + r3 * 0 + 16], m0
+    movu        [r2 + r3 * 1 + 16], m1
+    movu        [r2 + r3 * 2 + 16], m2
+    movu        [r2 + r4 + 16], m3
 
-    punpcklbw   m3,        m4
-    punpcklbw   m7,        m5,        m6
-    punpcklbw   m3,        m7
+    lea         r0, [r0 + 8]
 
-    pmaddubsw   m3,        m0
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-    phaddw      m2,        m3
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-    pmulhrsw    m2,        m1
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
 
-    movd        m7,        [r5 + r1]
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
 
-    punpcklbw   m4,        m5
-    punpcklbw   m3,        m6,        m7
-    punpcklbw   m4,        m3
+    movu        [r2 + r3 * 0 + 32], m0
+    movu        [r2 + r3 * 1 + 32], m1
+    movu        [r2 + r3 * 2 + 32], m2
+    movu        [r2 + r4 + 32], m3
 
-    pmaddubsw   m4,        m0
+    lea         r0, [r0 + 8]
 
-    movd        m3,        [r5 + 2 * r1]
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-    punpcklbw   m5,        m6
-    punpcklbw   m7,        m3
-    punpcklbw   m5,        m7
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-    pmaddubsw   m5,        m0
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
 
-    phaddw      m4,        m5
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
 
-    pmulhrsw    m4,        m1
+    movu        [r2 + r3 * 0 + 48], m0
+    movu        [r2 + r3 * 1 + 48], m1
+    movu        [r2 + r3 * 2 + 48], m2
+    movu        [r2 + r4 + 48], m3
 
-    packuswb    m2,        m4
-    movd        [r2],      m2
-    pextrd      [r2 + r3], m2, 1
-    lea         r2,        [r2 + 2 * r3]
-    pextrd      [r2],      m2, 2
-    pextrd      [r2 + r3], m2, 3
-    RET
-%macro FILTER_VER_CHROMA_AVX2_4x4 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_4x4, 4, 6, 3
-    mov             r4d, r4m
-    shl             r4d, 6
-    sub             r0, r1
+    lea         r0, [r0 + 8]
 
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-    lea             r4, [r1 * 3]
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-    movd            xm1, [r0]
-    pinsrd          xm1, [r0 + r1], 1
-    pinsrd          xm1, [r0 + r1 * 2], 2
-    pinsrd          xm1, [r0 + r4], 3                       ; m1 = row[3 2 1 0]
-    lea             r0, [r0 + r1 * 4]
-    movd            xm2, [r0]
-    pinsrd          xm2, [r0 + r1], 1
-    pinsrd          xm2, [r0 + r1 * 2], 2                   ; m2 = row[x 6 5 4]
-    vinserti128     m1, m1, xm2, 1                          ; m1 = row[x 6 5 4 3 2 1 0]
-    mova            m2, [interp4_vpp_shuf1]
-    vpermd          m0, m2, m1                              ; m0 = row[4 3 3 2 2 1 1 0]
-    mova            m2, [interp4_vpp_shuf1 + mmsize]
-    vpermd          m1, m2, m1                              ; m1 = row[6 5 5 4 4 3 3 2]
-
-    mova            m2, [interp4_vpp_shuf]
-    pshufb          m0, m0, m2
-    pshufb          m1, m1, m2
-    pmaddubsw       m0, [r5]
-    pmaddubsw       m1, [r5 + mmsize]
-    paddw           m0, m1                                  ; m0 = WORD ROW[3 2 1 0]
-%ifidn %1,pp
-    pmulhrsw        m0, [pw_512]
-    vextracti128    xm1, m0, 1
-    packuswb        xm0, xm1
-    lea             r5, [r3 * 3]
-    movd            [r2], xm0
-    pextrd          [r2 + r3], xm0, 1
-    pextrd          [r2 + r3 * 2], xm0, 2
-    pextrd          [r2 + r5], xm0, 3
-%else
-    add             r3d, r3d
-    psubw           m0, [pw_2000]
-    vextracti128    xm1, m0, 1
-    lea             r5, [r3 * 3]
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm1
-    movhps          [r2 + r5], xm1
-%endif
-    RET
-%endmacro
-    FILTER_VER_CHROMA_AVX2_4x4 pp
-    FILTER_VER_CHROMA_AVX2_4x4 ps
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
 
-%macro FILTER_VER_CHROMA_AVX2_4x8 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_4x8, 4, 6, 5
-    mov             r4d, r4m
-    shl             r4d, 6
-    sub             r0, r1
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
 
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
+    movu        [r2 + r3 * 0 + 64], m0
+    movu        [r2 + r3 * 1 + 64], m1
+    movu        [r2 + r3 * 2 + 64], m2
+    movu        [r2 + r4 + 64], m3
 
-    lea             r4, [r1 * 3]
+    lea         r0, [r0 + 8]
 
-    movd            xm1, [r0]
-    pinsrd          xm1, [r0 + r1], 1
-    pinsrd          xm1, [r0 + r1 * 2], 2
-    pinsrd          xm1, [r0 + r4], 3                       ; m1 = row[3 2 1 0]
-    lea             r0, [r0 + r1 * 4]
-    movd            xm2, [r0]
-    pinsrd          xm2, [r0 + r1], 1
-    pinsrd          xm2, [r0 + r1 * 2], 2
-    pinsrd          xm2, [r0 + r4], 3                       ; m2 = row[7 6 5 4]
-    vinserti128     m1, m1, xm2, 1                          ; m1 = row[7 6 5 4 3 2 1 0]
-    lea             r0, [r0 + r1 * 4]
-    movd            xm3, [r0]
-    pinsrd          xm3, [r0 + r1], 1
-    pinsrd          xm3, [r0 + r1 * 2], 2                   ; m3 = row[x 10 9 8]
-    vinserti128     m2, m2, xm3, 1                          ; m2 = row[x 10 9 8 7 6 5 4]
-    mova            m3, [interp4_vpp_shuf1]
-    vpermd          m0, m3, m1                              ; m0 = row[4 3 3 2 2 1 1 0]
-    vpermd          m4, m3, m2                              ; m4 = row[8 7 7 6 6 5 5 4]
-    mova            m3, [interp4_vpp_shuf1 + mmsize]
-    vpermd          m1, m3, m1                              ; m1 = row[6 5 5 4 4 3 3 2]
-    vpermd          m2, m3, m2                              ; m2 = row[10 9 9 8 8 7 7 6]
-
-    mova            m3, [interp4_vpp_shuf]
-    pshufb          m0, m0, m3
-    pshufb          m1, m1, m3
-    pshufb          m2, m2, m3
-    pshufb          m4, m4, m3
-    pmaddubsw       m0, [r5]
-    pmaddubsw       m4, [r5]
-    pmaddubsw       m1, [r5 + mmsize]
-    pmaddubsw       m2, [r5 + mmsize]
-    paddw           m0, m1                                  ; m0 = WORD ROW[3 2 1 0]
-    paddw           m4, m2                                  ; m4 = WORD ROW[7 6 5 4]
-%ifidn %1,pp
-    pmulhrsw        m0, [pw_512]
-    pmulhrsw        m4, [pw_512]
-    packuswb        m0, m4
-    vextracti128    xm1, m0, 1
-    lea             r5, [r3 * 3]
-    movd            [r2], xm0
-    pextrd          [r2 + r3], xm0, 1
-    movd            [r2 + r3 * 2], xm1
-    pextrd          [r2 + r5], xm1, 1
-    lea             r2, [r2 + r3 * 4]
-    pextrd          [r2], xm0, 2
-    pextrd          [r2 + r3], xm0, 3
-    pextrd          [r2 + r3 * 2], xm1, 2
-    pextrd          [r2 + r5], xm1, 3
-%else
-    add             r3d, r3d
-    psubw           m0, [pw_2000]
-    psubw           m4, [pw_2000]
-    vextracti128    xm1, m0, 1
-    vextracti128    xm2, m4, 1
-    lea             r5, [r3 * 3]
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm1
-    movhps          [r2 + r5], xm1
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm4
-    movhps          [r2 + r3], xm4
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r5], xm2
-%endif
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_AVX2_4x8 pp
-    FILTER_VER_CHROMA_AVX2_4x8 ps
-
-%macro FILTER_VER_CHROMA_AVX2_4xN 2
-%if ARCH_X86_64 == 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_4x%2, 4, 6, 12
-    mov             r4d, r4m
-    shl             r4d, 6
-    sub             r0, r1
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    mova            m10, [r5]
-    mova            m11, [r5 + mmsize]
-%ifidn %1,pp
-    mova            m9, [pw_512]
-%else
-    add             r3d, r3d
-    mova            m9, [pw_2000]
-%endif
-    lea             r5, [r3 * 3]
-%rep %2 / 16
-    movd            xm1, [r0]
-    pinsrd          xm1, [r0 + r1], 1
-    pinsrd          xm1, [r0 + r1 * 2], 2
-    pinsrd          xm1, [r0 + r4], 3                       ; m1 = row[3 2 1 0]
-    lea             r0, [r0 + r1 * 4]
-    movd            xm2, [r0]
-    pinsrd          xm2, [r0 + r1], 1
-    pinsrd          xm2, [r0 + r1 * 2], 2
-    pinsrd          xm2, [r0 + r4], 3                       ; m2 = row[7 6 5 4]
-    vinserti128     m1, m1, xm2, 1                          ; m1 = row[7 6 5 4 3 2 1 0]
-    lea             r0, [r0 + r1 * 4]
-    movd            xm3, [r0]
-    pinsrd          xm3, [r0 + r1], 1
-    pinsrd          xm3, [r0 + r1 * 2], 2
-    pinsrd          xm3, [r0 + r4], 3                       ; m3 = row[11 10 9 8]
-    vinserti128     m2, m2, xm3, 1                          ; m2 = row[11 10 9 8 7 6 5 4]
-    lea             r0, [r0 + r1 * 4]
-    movd            xm4, [r0]
-    pinsrd          xm4, [r0 + r1], 1
-    pinsrd          xm4, [r0 + r1 * 2], 2
-    pinsrd          xm4, [r0 + r4], 3                       ; m4 = row[15 14 13 12]
-    vinserti128     m3, m3, xm4, 1                          ; m3 = row[15 14 13 12 11 10 9 8]
-    lea             r0, [r0 + r1 * 4]
-    movd            xm5, [r0]
-    pinsrd          xm5, [r0 + r1], 1
-    pinsrd          xm5, [r0 + r1 * 2], 2                   ; m5 = row[x 18 17 16]
-    vinserti128     m4, m4, xm5, 1                          ; m4 = row[x 18 17 16 15 14 13 12]
-    mova            m5, [interp4_vpp_shuf1]
-    vpermd          m0, m5, m1                              ; m0 = row[4 3 3 2 2 1 1 0]
-    vpermd          m6, m5, m2                              ; m6 = row[8 7 7 6 6 5 5 4]
-    vpermd          m7, m5, m3                              ; m7 = row[12 11 11 10 10 9 9 8]
-    vpermd          m8, m5, m4                              ; m8 = row[16 15 15 14 14 13 13 12]
-    mova            m5, [interp4_vpp_shuf1 + mmsize]
-    vpermd          m1, m5, m1                              ; m1 = row[6 5 5 4 4 3 3 2]
-    vpermd          m2, m5, m2                              ; m2 = row[10 9 9 8 8 7 7 6]
-    vpermd          m3, m5, m3                              ; m3 = row[14 13 13 12 12 11 11 10]
-    vpermd          m4, m5, m4                              ; m4 = row[18 17 17 16 16 15 15 14]
-
-    mova            m5, [interp4_vpp_shuf]
-    pshufb          m0, m0, m5
-    pshufb          m1, m1, m5
-    pshufb          m2, m2, m5
-    pshufb          m4, m4, m5
-    pshufb          m3, m3, m5
-    pshufb          m6, m6, m5
-    pshufb          m7, m7, m5
-    pshufb          m8, m8, m5
-    pmaddubsw       m0, m10
-    pmaddubsw       m6, m10
-    pmaddubsw       m7, m10
-    pmaddubsw       m8, m10
-    pmaddubsw       m1, m11
-    pmaddubsw       m2, m11
-    pmaddubsw       m3, m11
-    pmaddubsw       m4, m11
-    paddw           m0, m1                                  ; m0 = WORD ROW[3 2 1 0]
-    paddw           m6, m2                                  ; m6 = WORD ROW[7 6 5 4]
-    paddw           m7, m3                                  ; m7 = WORD ROW[11 10 9 8]
-    paddw           m8, m4                                  ; m8 = WORD ROW[15 14 13 12]
-%ifidn %1,pp
-    pmulhrsw        m0, m9
-    pmulhrsw        m6, m9
-    pmulhrsw        m7, m9
-    pmulhrsw        m8, m9
-    packuswb        m0, m6
-    packuswb        m7, m8
-    vextracti128    xm1, m0, 1
-    vextracti128    xm2, m7, 1
-    movd            [r2], xm0
-    pextrd          [r2 + r3], xm0, 1
-    movd            [r2 + r3 * 2], xm1
-    pextrd          [r2 + r5], xm1, 1
-    lea             r2, [r2 + r3 * 4]
-    pextrd          [r2], xm0, 2
-    pextrd          [r2 + r3], xm0, 3
-    pextrd          [r2 + r3 * 2], xm1, 2
-    pextrd          [r2 + r5], xm1, 3
-    lea             r2, [r2 + r3 * 4]
-    movd            [r2], xm7
-    pextrd          [r2 + r3], xm7, 1
-    movd            [r2 + r3 * 2], xm2
-    pextrd          [r2 + r5], xm2, 1
-    lea             r2, [r2 + r3 * 4]
-    pextrd          [r2], xm7, 2
-    pextrd          [r2 + r3], xm7, 3
-    pextrd          [r2 + r3 * 2], xm2, 2
-    pextrd          [r2 + r5], xm2, 3
-%else
-    psubw           m0, m9
-    psubw           m6, m9
-    psubw           m7, m9
-    psubw           m8, m9
-    vextracti128    xm1, m0, 1
-    vextracti128    xm2, m6, 1
-    vextracti128    xm3, m7, 1
-    vextracti128    xm4, m8, 1
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm1
-    movhps          [r2 + r5], xm1
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm6
-    movhps          [r2 + r3], xm6
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r5], xm2
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm7
-    movhps          [r2 + r3], xm7
-    movq            [r2 + r3 * 2], xm3
-    movhps          [r2 + r5], xm3
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm8
-    movhps          [r2 + r3], xm8
-    movq            [r2 + r3 * 2], xm4
-    movhps          [r2 + r5], xm4
-%endif
-    lea             r2, [r2 + r3 * 4]
-%endrep
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_CHROMA_AVX2_4xN pp, 16
-    FILTER_VER_CHROMA_AVX2_4xN ps, 16
-    FILTER_VER_CHROMA_AVX2_4xN pp, 32
-    FILTER_VER_CHROMA_AVX2_4xN ps, 32
-
-;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro FILTER_V4_W4_H4 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_pp_%1x%2, 4, 6, 8
-
-    mov         r4d,       r4m
-    sub         r0,        r1
-
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeff]
-    movd        m0,        [r5 + r4 * 4]
-%else
-    movd        m0,        [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    pshufb      m0,        [tab_Cm]
-
-    mova        m1,        [pw_512]
-
-    mov         r4d,       %2
-
-    lea         r5,        [3 * r1]
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-.loop:
-    movd        m2,        [r0]
-    movd        m3,        [r0 + r1]
-    movd        m4,        [r0 + 2 * r1]
-    movd        m5,        [r0 + r5]
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-    punpcklbw   m2,        m3
-    punpcklbw   m6,        m4,        m5
-    punpcklbw   m2,        m6
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
 
-    pmaddubsw   m2,        m0
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
 
-    lea         r0,        [r0 + 4 * r1]
-    movd        m6,        [r0]
+    movu        [r2 + r3 * 0 + 80], m0
+    movu        [r2 + r3 * 1 + 80], m1
+    movu        [r2 + r3 * 2 + 80], m2
+    movu        [r2 + r4 + 80], m3
 
-    punpcklbw   m3,        m4
-    punpcklbw   m7,        m5,        m6
-    punpcklbw   m3,        m7
+    lea         r0, [r0 + 8]
 
-    pmaddubsw   m3,        m0
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-    phaddw      m2,        m3
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-    pmulhrsw    m2,        m1
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
 
-    movd        m7,        [r0 + r1]
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
 
-    punpcklbw   m4,        m5
-    punpcklbw   m3,        m6,        m7
-    punpcklbw   m4,        m3
+    movu        [r2 + r3 * 0 + 96], m0
+    movu        [r2 + r3 * 1 + 96], m1
+    movu        [r2 + r3 * 2 + 96], m2
+    movu        [r2 + r4 + 96], m3
 
-    pmaddubsw   m4,        m0
+    lea         r0, [r0 + 8]
 
-    movd        m3,        [r0 + 2 * r1]
+    movh        m0, [r0]
+    punpcklbw   m0, m4
+    pmaddubsw   m0, m5
 
-    punpcklbw   m5,        m6
-    punpcklbw   m7,        m3
-    punpcklbw   m5,        m7
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m4
+    pmaddubsw   m1, m5
 
-    pmaddubsw   m5,        m0
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m4
+    pmaddubsw   m2, m5
 
-    phaddw      m4,        m5
+    movh        m3, [r0 + r5]
+    punpcklbw   m3, m4
+    pmaddubsw   m3, m5
 
-    pmulhrsw    m4,        m1
-    packuswb    m2,        m4
-    movd        [r2],      m2
-    pextrd      [r2 + r3], m2,  1
-    lea         r2,        [r2 + 2 * r3]
-    pextrd      [r2],      m2, 2
-    pextrd      [r2 + r3], m2, 3
+    movu        [r2 + r3 * 0 + 112], m0
+    movu        [r2 + r3 * 1 + 112], m1
+    movu        [r2 + r3 * 2 + 112], m2
+    movu        [r2 + r4 + 112], m3
 
-    lea         r2,        [r2 + 2 * r3]
+    lea         r0, [r0 + r1 * 4 - 56]
+    lea         r2, [r2 + r3 * 4]
 
-    sub         r4,        4
-    jnz        .loop
+    dec         r6d
+    jnz         .loop
     RET
 %endmacro
+    P2S_H_64xN 64
+    P2S_H_64xN 16
+    P2S_H_64xN 32
+    P2S_H_64xN 48
 
-    FILTER_V4_W4_H4 4,  8
-    FILTER_V4_W4_H4 4, 16
-
-    FILTER_V4_W4_H4 4, 32
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+%macro P2S_H_64xN_avx2 1
+INIT_YMM avx2
+cglobal filterPixelToShort_64x%1, 3, 7, 5
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r5, [r1 * 3]
+    lea         r6, [r3 * 3]
 
-%macro FILTER_V4_W8_H2 0
-    punpcklbw   m1,        m2
-    punpcklbw   m7,        m3,        m0
+    ; load height
+    mov         r4d, %1/4
 
-    pmaddubsw   m1,        m6
-    pmaddubsw   m7,        m5
+    ; load constant
+    vpbroadcastd m4, [pw_2000]
 
-    paddw       m1,        m7
+.loop:
+    pmovzxbw    m0, [r0 + 0 * mmsize/2]
+    pmovzxbw    m1, [r0 + 1 * mmsize/2]
+    pmovzxbw    m2, [r0 + 2 * mmsize/2]
+    pmovzxbw    m3, [r0 + 3 * mmsize/2]
+    psllw       m0, 6
+    psllw       m1, 6
+    psllw       m2, 6
+    psllw       m3, 6
+    psubw       m0, m4
+    psubw       m1, m4
+    psubw       m2, m4
+    psubw       m3, m4
 
-    pmulhrsw    m1,        m4
-    packuswb    m1,        m1
-%endmacro
+    movu        [r2 + 0 * mmsize], m0
+    movu        [r2 + 1 * mmsize], m1
+    movu        [r2 + 2 * mmsize], m2
+    movu        [r2 + 3 * mmsize], m3
 
-%macro FILTER_V4_W8_H3 0
-    punpcklbw   m2,        m3
-    punpcklbw   m7,        m0,        m1
+    pmovzxbw    m0, [r0 + r1 + 0 * mmsize/2]
+    pmovzxbw    m1, [r0 + r1 + 1 * mmsize/2]
+    pmovzxbw    m2, [r0 + r1 + 2 * mmsize/2]
+    pmovzxbw    m3, [r0 + r1 + 3 * mmsize/2]
+    psllw       m0, 6
+    psllw       m1, 6
+    psllw       m2, 6
+    psllw       m3, 6
+    psubw       m0, m4
+    psubw       m1, m4
+    psubw       m2, m4
+    psubw       m3, m4
 
-    pmaddubsw   m2,        m6
-    pmaddubsw   m7,        m5
+    movu        [r2 + r3 + 0 * mmsize], m0
+    movu        [r2 + r3 + 1 * mmsize], m1
+    movu        [r2 + r3 + 2 * mmsize], m2
+    movu        [r2 + r3 + 3 * mmsize], m3
 
-    paddw       m2,        m7
+    pmovzxbw    m0, [r0 + r1 * 2 + 0 * mmsize/2]
+    pmovzxbw    m1, [r0 + r1 * 2 + 1 * mmsize/2]
+    pmovzxbw    m2, [r0 + r1 * 2 + 2 * mmsize/2]
+    pmovzxbw    m3, [r0 + r1 * 2 + 3 * mmsize/2]
+    psllw       m0, 6
+    psllw       m1, 6
+    psllw       m2, 6
+    psllw       m3, 6
+    psubw       m0, m4
+    psubw       m1, m4
+    psubw       m2, m4
+    psubw       m3, m4
 
-    pmulhrsw    m2,        m4
-    packuswb    m2,        m2
-%endmacro
-
-%macro FILTER_V4_W8_H4 0
-    punpcklbw   m3,        m0
-    punpcklbw   m7,        m1,        m2
-
-    pmaddubsw   m3,        m6
-    pmaddubsw   m7,        m5
-
-    paddw       m3,        m7
-
-    pmulhrsw    m3,        m4
-    packuswb    m3,        m3
-%endmacro
-
-%macro FILTER_V4_W8_H5 0
-    punpcklbw   m0,        m1
-    punpcklbw   m7,        m2,        m3
-
-    pmaddubsw   m0,        m6
-    pmaddubsw   m7,        m5
-
-    paddw       m0,        m7
+    movu        [r2 + r3 * 2 + 0 * mmsize], m0
+    movu        [r2 + r3 * 2 + 1 * mmsize], m1
+    movu        [r2 + r3 * 2 + 2 * mmsize], m2
+    movu        [r2 + r3 * 2 + 3 * mmsize], m3
 
-    pmulhrsw    m0,        m4
-    packuswb    m0,        m0
-%endmacro
+    pmovzxbw    m0, [r0 + r5 + 0 * mmsize/2]
+    pmovzxbw    m1, [r0 + r5 + 1 * mmsize/2]
+    pmovzxbw    m2, [r0 + r5 + 2 * mmsize/2]
+    pmovzxbw    m3, [r0 + r5 + 3 * mmsize/2]
+    psllw       m0, 6
+    psllw       m1, 6
+    psllw       m2, 6
+    psllw       m3, 6
+    psubw       m0, m4
+    psubw       m1, m4
+    psubw       m2, m4
+    psubw       m3, m4
 
-%macro FILTER_V4_W8_8x2 2
-    FILTER_V4_W8 %1, %2
-    movq        m0,        [r0 + 4 * r1]
+    movu        [r2 + r6 + 0 * mmsize], m0
+    movu        [r2 + r6 + 1 * mmsize], m1
+    movu        [r2 + r6 + 2 * mmsize], m2
+    movu        [r2 + r6 + 3 * mmsize], m3
 
-    FILTER_V4_W8_H2
+    lea         r0, [r0 + r1 * 4]
+    lea         r2, [r2 + r3 * 4]
 
-    movh        [r2 + r3], m1
+    dec         r4d
+    jnz        .loop
+    RET
 %endmacro
+    P2S_H_64xN_avx2 64
+    P2S_H_64xN_avx2 16
+    P2S_H_64xN_avx2 32
+    P2S_H_64xN_avx2 48
 
-%macro FILTER_V4_W8_8x4 2
-    FILTER_V4_W8_8x2 %1, %2
-;8x3
-    lea         r6,        [r0 + 4 * r1]
-    movq        m1,        [r6 + r1]
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel src, intptr_t srcStride, int16_t dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+%macro P2S_H_12xN 1
+INIT_XMM ssse3
+cglobal filterPixelToShort_12x%1, 3, 7, 6
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r4, [r1 * 3]
+    lea         r6, [r3 * 3]
+    mov         r5d, %1/4
 
-    FILTER_V4_W8_H3
+    ; load constant
+    mova        m4, [pb_128]
+    mova        m5, [tab_c_64_n64]
 
-    movh        [r2 + 2 * r3], m2
+.loop:
+    movu        m0, [r0]
+    punpcklbw   m1, m0, m4
+    punpckhbw   m0, m4
+    pmaddubsw   m0, m5
+    pmaddubsw   m1, m5
 
-;8x4
-    movq        m2,        [r6 + 2 * r1]
+    movu        m2, [r0 + r1]
+    punpcklbw   m3, m2, m4
+    punpckhbw   m2, m4
+    pmaddubsw   m2, m5
+    pmaddubsw   m3, m5
 
-    FILTER_V4_W8_H4
+    movu        [r2 + r3 * 0], m1
+    movu        [r2 + r3 * 1], m3
 
-    lea         r5,        [r2 + 2 * r3]
-    movh        [r5 + r3], m3
-%endmacro
+    movh        [r2 + r3 * 0 + 16], m0
+    movh        [r2 + r3 * 1 + 16], m2
 
-%macro FILTER_V4_W8_8x6 2
-    FILTER_V4_W8_8x4 %1, %2
-;8x5
-    lea         r6,        [r6 + 2 * r1]
-    movq        m3,        [r6 + r1]
+    movu        m0, [r0 + r1 * 2]
+    punpcklbw   m1, m0, m4
+    punpckhbw   m0, m4
+    pmaddubsw   m0, m5
+    pmaddubsw   m1, m5
 
-    FILTER_V4_W8_H5
+    movu        m2, [r0 + r4]
+    punpcklbw   m3, m2, m4
+    punpckhbw   m2, m4
+    pmaddubsw   m2, m5
+    pmaddubsw   m3, m5
 
-    movh        [r2 + 4 * r3], m0
+    movu        [r2 + r3 * 2], m1
+    movu        [r2 + r6], m3
 
-;8x6
-    movq        m0,        [r0 + 8 * r1]
+    movh        [r2 + r3 * 2 + 16], m0
+    movh        [r2 + r6 + 16], m2
 
-    FILTER_V4_W8_H2
+    lea         r0, [r0 + r1 * 4]
+    lea         r2, [r2 + r3 * 4]
 
-    lea         r5,        [r2 + 4 * r3]
-    movh        [r5 + r3], m1
+    dec         r5d
+    jnz         .loop
+    RET
 %endmacro
+    P2S_H_12xN 16
+    P2S_H_12xN 32
 
 ;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
 ;-----------------------------------------------------------------------------
-%macro FILTER_V4_W8 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_pp_%1x%2, 4, 7, 8
+%macro P2S_H_24xN 1
+INIT_XMM ssse3
+cglobal filterPixelToShort_24x%1, 3, 7, 5
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r4, [r1 * 3]
+    lea         r5, [r3 * 3]
+    mov         r6d, %1/4
 
-    mov         r4d,       r4m
+    ; load constant
+    mova        m3, [pb_128]
+    mova        m4, [tab_c_64_n64]
 
-    sub         r0,        r1
-    movq        m0,        [r0]
-    movq        m1,        [r0 + r1]
-    movq        m2,        [r0 + 2 * r1]
-    lea         r5,        [r0 + 2 * r1]
-    movq        m3,        [r5 + r1]
+.loop:
+    movu        m0, [r0]
+    punpcklbw   m1, m0, m3
+    punpckhbw   m0, m3
+    pmaddubsw   m0, m4
+    pmaddubsw   m1, m4
 
-    punpcklbw   m0,        m1
-    punpcklbw   m4,        m2,          m3
+    movu        m2, [r0 + 16]
+    punpcklbw   m2, m3
+    pmaddubsw   m2, m4
 
-%ifdef PIC
-    lea         r6,        [tab_ChromaCoeff]
-    movd        m5,        [r6 + r4 * 4]
-%else
-    movd        m5,        [tab_ChromaCoeff + r4 * 4]
-%endif
+    movu        [r2 +  r3 * 0], m1
+    movu        [r2 +  r3 * 0 + 16], m0
+    movu        [r2 +  r3 * 0 + 32], m2
+
+    movu        m0, [r0 + r1]
+    punpcklbw   m1, m0, m3
+    punpckhbw   m0, m3
+    pmaddubsw   m0, m4
+    pmaddubsw   m1, m4
 
-    pshufb      m6,        m5,       [tab_Vm]
-    pmaddubsw   m0,        m6
+    movu        m2, [r0 + r1 + 16]
+    punpcklbw   m2, m3
+    pmaddubsw   m2, m4
 
-    pshufb      m5,        [tab_Vm + 16]
-    pmaddubsw   m4,        m5
+    movu        [r2 +  r3 * 1], m1
+    movu        [r2 +  r3 * 1 + 16], m0
+    movu        [r2 +  r3 * 1 + 32], m2
 
-    paddw       m0,        m4
+    movu        m0, [r0 + r1 * 2]
+    punpcklbw   m1, m0, m3
+    punpckhbw   m0, m3
+    pmaddubsw   m0, m4
+    pmaddubsw   m1, m4
 
-    mova        m4,        [pw_512]
+    movu        m2, [r0 + r1 * 2 + 16]
+    punpcklbw   m2, m3
+    pmaddubsw   m2, m4
 
-    pmulhrsw    m0,        m4
-    packuswb    m0,        m0
-    movh        [r2],      m0
-%endmacro
+    movu        [r2 +  r3 * 2], m1
+    movu        [r2 +  r3 * 2 + 16], m0
+    movu        [r2 +  r3 * 2 + 32], m2
 
-;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_8x2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-    FILTER_V4_W8_8x2 8, 2
+    movu        m0, [r0 + r4]
+    punpcklbw   m1, m0, m3
+    punpckhbw   m0, m3
+    pmaddubsw   m0, m4
+    pmaddubsw   m1, m4
 
-    RET
+    movu        m2, [r0 + r4 + 16]
+    punpcklbw   m2, m3
+    pmaddubsw   m2, m4
+    movu        [r2 +  r5], m1
+    movu        [r2 +  r5 + 16], m0
+    movu        [r2 +  r5 + 32], m2
 
-;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_8x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-    FILTER_V4_W8_8x4 8, 4
+    lea         r0, [r0 + r1 * 4]
+    lea         r2, [r2 + r3 * 4]
 
+    dec         r6d
+    jnz         .loop
     RET
+%endmacro
+    P2S_H_24xN 32
+    P2S_H_24xN 64
 
 ;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_8x6(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
 ;-----------------------------------------------------------------------------
-    FILTER_V4_W8_8x6 8, 6
-
-    RET
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ps_4x2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-INIT_XMM sse4
-cglobal interp_4tap_vert_ps_4x2, 4, 6, 6
-
-    mov         r4d, r4m
-    sub         r0, r1
+%macro P2S_H_24xN_avx2 1
+INIT_YMM avx2
+cglobal filterPixelToShort_24x%1, 3, 7, 4
+    mov         r3d, r3m
     add         r3d, r3d
+    lea         r4, [r1 * 3]
+    lea         r5, [r3 * 3]
+    mov         r6d, %1/4
 
-%ifdef PIC
-    lea         r5, [tab_ChromaCoeff]
-    movd        m0, [r5 + r4 * 4]
-%else
-    movd        m0, [tab_ChromaCoeff + r4 * 4]
-%endif
+    ; load constant
+    vpbroadcastd m1, [pw_2000]
+    vpbroadcastd m2, [pb_128]
+    vpbroadcastd m3, [tab_c_64_n64]
 
-    pshufb      m0, [tab_Cm]
+.loop:
+    pmovzxbw    m0, [r0]
+    psllw       m0, 6
+    psubw       m0, m1
+    movu        [r2], m0
 
-    movd        m2, [r0]
-    movd        m3, [r0 + r1]
-    lea         r5, [r0 + 2 * r1]
-    movd        m4, [r5]
-    movd        m5, [r5 + r1]
+    movu        m0, [r0 + mmsize/2]
+    punpcklbw   m0, m2
+    pmaddubsw   m0, m3
+    movu        [r2 +  r3 * 0 + mmsize], xm0
 
-    punpcklbw   m2, m3
-    punpcklbw   m1, m4, m5
-    punpcklbw   m2, m1
+    pmovzxbw    m0, [r0 + r1]
+    psllw       m0, 6
+    psubw       m0, m1
+    movu        [r2 + r3], m0
 
-    pmaddubsw   m2, m0
+    movu        m0, [r0 + r1 + mmsize/2]
+    punpcklbw   m0, m2
+    pmaddubsw   m0, m3
+    movu        [r2 +  r3 * 1 + mmsize], xm0
 
-    movd        m1, [r0 + 4 * r1]
+    pmovzxbw    m0, [r0 + r1 * 2]
+    psllw       m0, 6
+    psubw       m0, m1
+    movu        [r2 + r3 * 2], m0
 
-    punpcklbw   m3, m4
-    punpcklbw   m5, m1
-    punpcklbw   m3, m5
+    movu        m0, [r0 + r1 * 2 + mmsize/2]
+    punpcklbw   m0, m2
+    pmaddubsw   m0, m3
+    movu        [r2 +  r3 * 2 + mmsize], xm0
 
-    pmaddubsw   m3, m0
+    pmovzxbw    m0, [r0 + r4]
+    psllw       m0, 6
+    psubw       m0, m1
+    movu        [r2 + r5], m0
 
-    phaddw      m2, m3
+    movu        m0, [r0 + r4 + mmsize/2]
+    punpcklbw   m0, m2
+    pmaddubsw   m0, m3
+    movu        [r2 + r5 + mmsize], xm0
 
-    psubw       m2, [pw_2000]
-    movh        [r2], m2
-    movhps      [r2 + r3], m2
+    lea         r0, [r0 + r1 * 4]
+    lea         r2, [r2 + r3 * 4]
 
+    dec         r6d
+    jnz         .loop
     RET
+%endmacro
+    P2S_H_24xN_avx2 32
+    P2S_H_24xN_avx2 64
 
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ps_4x4(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-INIT_XMM sse4
-cglobal interp_4tap_vert_ps_4x4, 4, 6, 7
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+INIT_XMM ssse3
+cglobal filterPixelToShort_48x64, 3, 7, 4
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r4, [r1 * 3]
+    lea         r5, [r3 * 3]
+    mov         r6d, 16
 
-    mov        r4d, r4m
-    sub        r0, r1
-    add        r3d, r3d
+    ; load constant
+    mova        m2, [pb_128]
+    mova        m3, [tab_c_64_n64]
 
-%ifdef PIC
-    lea        r5, [tab_ChromaCoeff]
-    movd       m0, [r5 + r4 * 4]
-%else
-    movd       m0, [tab_ChromaCoeff + r4 * 4]
-%endif
+.loop:
+    movu        m0, [r0]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
 
-    pshufb     m0, [tab_Cm]
+    movu        [r2 +  r3 * 0], m1
+    movu        [r2 +  r3 * 0 + 16], m0
 
-    lea        r4, [r1 * 3]
-    lea        r5, [r0 + 4 * r1]
-
-    movd       m2, [r0]
-    movd       m3, [r0 + r1]
-    movd       m4, [r0 + 2 * r1]
-    movd       m5, [r0 + r4]
-
-    punpcklbw  m2, m3
-    punpcklbw  m6, m4, m5
-    punpcklbw  m2, m6
-
-    pmaddubsw  m2, m0
-
-    movd       m6, [r5]
-
-    punpcklbw  m3, m4
-    punpcklbw  m1, m5, m6
-    punpcklbw  m3, m1
-
-    pmaddubsw  m3, m0
-
-    phaddw     m2, m3
-
-    mova       m1, [pw_2000]
-
-    psubw      m2, m1
-    movh       [r2], m2
-    movhps     [r2 + r3], m2
-
-    movd       m2, [r5 + r1]
+    movu        m0, [r0 + 16]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
 
-    punpcklbw  m4, m5
-    punpcklbw  m3, m6, m2
-    punpcklbw  m4, m3
+    movu        [r2 +  r3 * 0 + 32], m1
+    movu        [r2 +  r3 * 0 + 48], m0
 
-    pmaddubsw  m4, m0
+    movu        m0, [r0 + 32]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
 
-    movd       m3, [r5 + 2 * r1]
+    movu        [r2 +  r3 * 0 + 64], m1
+    movu        [r2 +  r3 * 0 + 80], m0
 
-    punpcklbw  m5, m6
-    punpcklbw  m2, m3
-    punpcklbw  m5, m2
+    movu        m0, [r0 + r1]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
 
-    pmaddubsw  m5, m0
+    movu        [r2 +  r3 * 1], m1
+    movu        [r2 +  r3 * 1 + 16], m0
 
-    phaddw     m4, m5
+    movu        m0, [r0 + r1 + 16]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
 
-    psubw      m4, m1
-    lea        r2, [r2 + 2 * r3]
-    movh       [r2], m4
-    movhps     [r2 + r3], m4
+    movu        [r2 +  r3 * 1 + 32], m1
+    movu        [r2 +  r3 * 1 + 48], m0
 
-    RET
+    movu        m0, [r0 + r1 + 32]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
 
-;---------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ps_%1x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;---------------------------------------------------------------------------------------------------------------
-%macro FILTER_V_PS_W4_H4 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_ps_%1x%2, 4, 6, 8
+    movu        [r2 +  r3 * 1 + 64], m1
+    movu        [r2 +  r3 * 1 + 80], m0
 
-    mov        r4d, r4m
-    sub        r0, r1
-    add        r3d, r3d
+    movu        m0, [r0 + r1 * 2]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
 
-%ifdef PIC
-    lea        r5, [tab_ChromaCoeff]
-    movd       m0, [r5 + r4 * 4]
-%else
-    movd       m0, [tab_ChromaCoeff + r4 * 4]
-%endif
+    movu        [r2 +  r3 * 2], m1
+    movu        [r2 +  r3 * 2 + 16], m0
 
-    pshufb     m0, [tab_Cm]
+    movu        m0, [r0 + r1 * 2 + 16]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
 
-    mova       m1, [pw_2000]
+    movu        [r2 +  r3 * 2 + 32], m1
+    movu        [r2 +  r3 * 2 + 48], m0
 
-    mov        r4d, %2/4
-    lea        r5, [3 * r1]
+    movu        m0, [r0 + r1 * 2 + 32]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
 
-.loop:
-    movd       m2, [r0]
-    movd       m3, [r0 + r1]
-    movd       m4, [r0 + 2 * r1]
-    movd       m5, [r0 + r5]
+    movu        [r2 +  r3 * 2 + 64], m1
+    movu        [r2 +  r3 * 2 + 80], m0
 
-    punpcklbw  m2, m3
-    punpcklbw  m6, m4, m5
-    punpcklbw  m2, m6
+    movu        m0, [r0 + r4]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
 
-    pmaddubsw  m2, m0
+    movu        [r2 +  r5], m1
+    movu        [r2 +  r5 + 16], m0
 
-    lea        r0, [r0 + 4 * r1]
-    movd       m6, [r0]
+    movu        m0, [r0 + r4 + 16]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
 
-    punpcklbw  m3, m4
-    punpcklbw  m7, m5, m6
-    punpcklbw  m3, m7
+    movu        [r2 +  r5 + 32], m1
+    movu        [r2 +  r5 + 48], m0
 
-    pmaddubsw  m3, m0
+    movu        m0, [r0 + r4 + 32]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
 
-    phaddw     m2, m3
+    movu        [r2 +  r5 + 64], m1
+    movu        [r2 +  r5 + 80], m0
 
-    psubw      m2, m1
-    movh       [r2], m2
-    movhps     [r2 + r3], m2
+    lea         r0, [r0 + r1 * 4]
+    lea         r2, [r2 + r3 * 4]
 
-    movd       m2, [r0 + r1]
+    dec         r6d
+    jnz         .loop
+    RET
 
-    punpcklbw  m4, m5
-    punpcklbw  m3, m6, m2
-    punpcklbw  m4, m3
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+INIT_YMM avx2
+cglobal filterPixelToShort_48x64, 3,7,4
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r5, [r1 * 3]
+    lea         r6, [r3 * 3]
 
-    pmaddubsw  m4, m0
+    ; load height
+    mov         r4d, 64/4
 
-    movd       m3, [r0 + 2 * r1]
+    ; load constant
+    vpbroadcastd m3, [pw_2000]
 
-    punpcklbw  m5, m6
-    punpcklbw  m2, m3
-    punpcklbw  m5, m2
+    ; just unroll(1) because it is best choice for 48x64
+.loop:
+    pmovzxbw    m0, [r0 + 0 * mmsize/2]
+    pmovzxbw    m1, [r0 + 1 * mmsize/2]
+    pmovzxbw    m2, [r0 + 2 * mmsize/2]
+    psllw       m0, 6
+    psllw       m1, 6
+    psllw       m2, 6
+    psubw       m0, m3
+    psubw       m1, m3
+    psubw       m2, m3
+    movu        [r2 + 0 * mmsize], m0
+    movu        [r2 + 1 * mmsize], m1
+    movu        [r2 + 2 * mmsize], m2
 
-    pmaddubsw  m5, m0
+    pmovzxbw    m0, [r0 + r1 + 0 * mmsize/2]
+    pmovzxbw    m1, [r0 + r1 + 1 * mmsize/2]
+    pmovzxbw    m2, [r0 + r1 + 2 * mmsize/2]
+    psllw       m0, 6
+    psllw       m1, 6
+    psllw       m2, 6
+    psubw       m0, m3
+    psubw       m1, m3
+    psubw       m2, m3
+    movu        [r2 + r3 + 0 * mmsize], m0
+    movu        [r2 + r3 + 1 * mmsize], m1
+    movu        [r2 + r3 + 2 * mmsize], m2
 
-    phaddw     m4, m5
+    pmovzxbw    m0, [r0 + r1 * 2 + 0 * mmsize/2]
+    pmovzxbw    m1, [r0 + r1 * 2 + 1 * mmsize/2]
+    pmovzxbw    m2, [r0 + r1 * 2 + 2 * mmsize/2]
+    psllw       m0, 6
+    psllw       m1, 6
+    psllw       m2, 6
+    psubw       m0, m3
+    psubw       m1, m3
+    psubw       m2, m3
+    movu        [r2 + r3 * 2 + 0 * mmsize], m0
+    movu        [r2 + r3 * 2 + 1 * mmsize], m1
+    movu        [r2 + r3 * 2 + 2 * mmsize], m2
 
-    psubw      m4, m1
-    lea        r2, [r2 + 2 * r3]
-    movh       [r2], m4
-    movhps     [r2 + r3], m4
+    pmovzxbw    m0, [r0 + r5 + 0 * mmsize/2]
+    pmovzxbw    m1, [r0 + r5 + 1 * mmsize/2]
+    pmovzxbw    m2, [r0 + r5 + 2 * mmsize/2]
+    psllw       m0, 6
+    psllw       m1, 6
+    psllw       m2, 6
+    psubw       m0, m3
+    psubw       m1, m3
+    psubw       m2, m3
+    movu        [r2 + r6 + 0 * mmsize], m0
+    movu        [r2 + r6 + 1 * mmsize], m1
+    movu        [r2 + r6 + 2 * mmsize], m2
 
-    lea        r2, [r2 + 2 * r3]
+    lea         r0, [r0 + r1 * 4]
+    lea         r2, [r2 + r3 * 4]
 
-    dec        r4d
+    dec         r4d
     jnz        .loop
     RET
-%endmacro
 
-    FILTER_V_PS_W4_H4 4, 8
-    FILTER_V_PS_W4_H4 4, 16
 
-    FILTER_V_PS_W4_H4 4, 32
+%macro PROCESS_LUMA_W4_4R 0
+    movd        m0, [r0]
+    movd        m1, [r0 + r1]
+    punpcklbw   m2, m0, m1                     ; m2=[0 1]
 
-;--------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ps_8x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-%macro FILTER_V_PS_W8_H8_H16_H2 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_ps_%1x%2, 4, 6, 7
+    lea         r0, [r0 + 2 * r1]
+    movd        m0, [r0]
+    punpcklbw   m1, m0                         ; m1=[1 2]
+    punpcklqdq  m2, m1                         ; m2=[0 1 1 2]
+    pmaddubsw   m4, m2, [r6 + 0 * 16]          ; m4=[0+1 1+2]
 
-    mov        r4d, r4m
-    sub        r0, r1
-    add        r3d, r3d
+    movd        m1, [r0 + r1]
+    punpcklbw   m5, m0, m1                     ; m2=[2 3]
+    lea         r0, [r0 + 2 * r1]
+    movd        m0, [r0]
+    punpcklbw   m1, m0                         ; m1=[3 4]
+    punpcklqdq  m5, m1                         ; m5=[2 3 3 4]
+    pmaddubsw   m2, m5, [r6 + 1 * 16]          ; m2=[2+3 3+4]
+    paddw       m4, m2                         ; m4=[0+1+2+3 1+2+3+4]                   Row1-2
+    pmaddubsw   m5, [r6 + 0 * 16]              ; m5=[2+3 3+4]                           Row3-4
 
-%ifdef PIC
-    lea        r5, [tab_ChromaCoeff]
-    movd       m5, [r5 + r4 * 4]
-%else
-    movd       m5, [tab_ChromaCoeff + r4 * 4]
-%endif
+    movd        m1, [r0 + r1]
+    punpcklbw   m2, m0, m1                     ; m2=[4 5]
+    lea         r0, [r0 + 2 * r1]
+    movd        m0, [r0]
+    punpcklbw   m1, m0                         ; m1=[5 6]
+    punpcklqdq  m2, m1                         ; m2=[4 5 5 6]
+    pmaddubsw   m1, m2, [r6 + 2 * 16]          ; m1=[4+5 5+6]
+    paddw       m4, m1                         ; m4=[0+1+2+3+4+5 1+2+3+4+5+6]           Row1-2
+    pmaddubsw   m2, [r6 + 1 * 16]              ; m2=[4+5 5+6]
+    paddw       m5, m2                         ; m5=[2+3+4+5 3+4+5+6]                   Row3-4
 
-    pshufb     m6, m5, [tab_Vm]
-    pshufb     m5, [tab_Vm + 16]
-    mova       m4, [pw_2000]
+    movd        m1, [r0 + r1]
+    punpcklbw   m2, m0, m1                     ; m2=[6 7]
+    lea         r0, [r0 + 2 * r1]
+    movd        m0, [r0]
+    punpcklbw   m1, m0                         ; m1=[7 8]
+    punpcklqdq  m2, m1                         ; m2=[6 7 7 8]
+    pmaddubsw   m1, m2, [r6 + 3 * 16]          ; m1=[6+7 7+8]
+    paddw       m4, m1                         ; m4=[0+1+2+3+4+5+6+7 1+2+3+4+5+6+7+8]   Row1-2 end
+    pmaddubsw   m2, [r6 + 2 * 16]              ; m2=[6+7 7+8]
+    paddw       m5, m2                         ; m5=[2+3+4+5+6+7 3+4+5+6+7+8]           Row3-4
 
-    mov        r4d, %2/2
-    lea        r5, [3 * r1]
+    movd        m1, [r0 + r1]
+    punpcklbw   m2, m0, m1                     ; m2=[8 9]
+    movd        m0, [r0 + 2 * r1]
+    punpcklbw   m1, m0                         ; m1=[9 10]
+    punpcklqdq  m2, m1                         ; m2=[8 9 9 10]
+    pmaddubsw   m2, [r6 + 3 * 16]              ; m2=[8+9 9+10]
+    paddw       m5, m2                         ; m5=[2+3+4+5+6+7+8+9 3+4+5+6+7+8+9+10]  Row3-4 end
+%endmacro
 
-.loopH:
+%macro PROCESS_LUMA_W8_4R 0
     movq       m0, [r0]
     movq       m1, [r0 + r1]
-    movq       m2, [r0 + 2 * r1]
-    movq       m3, [r0 + r5]
-
     punpcklbw  m0, m1
-    punpcklbw  m1, m2
-    punpcklbw  m2, m3
-
-    pmaddubsw  m0, m6
-    pmaddubsw  m2, m5
-
-    paddw      m0, m2
-
-    psubw      m0, m4
-    movu       [r2], m0
-
-    movq       m0, [r0 + 4 * r1]
-
-    punpcklbw  m3, m0
-
-    pmaddubsw  m1, m6
-    pmaddubsw  m3, m5
+    pmaddubsw  m7, m0, [r6 + 0 *16]            ;m7=[0+1]               Row1
 
-    paddw      m1, m3
-    psubw      m1, m4
+    lea        r0, [r0 + 2 * r1]
+    movq       m0, [r0]
+    punpcklbw  m1, m0
+    pmaddubsw  m6, m1, [r6 + 0 *16]            ;m6=[1+2]               Row2
 
-    movu       [r2 + r3], m1
+    movq       m1, [r0 + r1]
+    punpcklbw  m0, m1
+    pmaddubsw  m5, m0, [r6 + 0 *16]            ;m5=[2+3]               Row3
+    pmaddubsw  m0, [r6 + 1 * 16]
+    paddw      m7, m0                          ;m7=[0+1+2+3]           Row1
 
     lea        r0, [r0 + 2 * r1]
-    lea        r2, [r2 + 2 * r3]
+    movq       m0, [r0]
+    punpcklbw  m1, m0
+    pmaddubsw  m4, m1, [r6 + 0 *16]            ;m4=[3+4]               Row4
+    pmaddubsw  m1, [r6 + 1 * 16]
+    paddw      m6, m1                          ;m6 = [1+2+3+4]         Row2
 
-    dec        r4d
-    jnz       .loopH
+    movq       m1, [r0 + r1]
+    punpcklbw  m0, m1
+    pmaddubsw  m2, m0, [r6 + 1 * 16]
+    pmaddubsw  m0, [r6 + 2 * 16]
+    paddw      m7, m0                          ;m7=[0+1+2+3+4+5]       Row1
+    paddw      m5, m2                          ;m5=[2+3+4+5]           Row3
 
-    RET
-%endmacro
-
-    FILTER_V_PS_W8_H8_H16_H2 8, 2
-    FILTER_V_PS_W8_H8_H16_H2 8, 4
-    FILTER_V_PS_W8_H8_H16_H2 8, 6
-
-    FILTER_V_PS_W8_H8_H16_H2 8, 12
-    FILTER_V_PS_W8_H8_H16_H2 8, 64
-
-;--------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ps_8x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-%macro FILTER_V_PS_W8_H8_H16_H32 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_ps_%1x%2, 4, 6, 8
-
-    mov        r4d, r4m
-    sub        r0, r1
-    add        r3d, r3d
-
-%ifdef PIC
-    lea        r5, [tab_ChromaCoeff]
-    movd       m5, [r5 + r4 * 4]
-%else
-    movd       m5, [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    pshufb     m6, m5, [tab_Vm]
-    pshufb     m5, [tab_Vm + 16]
-    mova       m4, [pw_2000]
-
-    mov        r4d, %2/4
-    lea        r5, [3 * r1]
-
-.loop:
+    lea        r0, [r0 + 2 * r1]
     movq       m0, [r0]
-    movq       m1, [r0 + r1]
-    movq       m2, [r0 + 2 * r1]
-    movq       m3, [r0 + r5]
+    punpcklbw  m1, m0
+    pmaddubsw  m2, m1, [r6 + 1 * 16]
+    pmaddubsw  m1, [r6 + 2 * 16]
+    paddw      m6, m1                          ;m6=[1+2+3+4+5+6]       Row2
+    paddw      m4, m2                          ;m4=[3+4+5+6]           Row4
 
+    movq       m1, [r0 + r1]
     punpcklbw  m0, m1
-    punpcklbw  m1, m2
-    punpcklbw  m2, m3
-
-    pmaddubsw  m0, m6
-    pmaddubsw  m7, m2, m5
-
-    paddw      m0, m7
-
-    psubw       m0, m4
-    movu       [r2], m0
+    pmaddubsw  m2, m0, [r6 + 2 * 16]
+    pmaddubsw  m0, [r6 + 3 * 16]
+    paddw      m7, m0                          ;m7=[0+1+2+3+4+5+6+7]   Row1 end
+    paddw      m5, m2                          ;m5=[2+3+4+5+6+7]       Row3
 
-    lea        r0, [r0 + 4 * r1]
+    lea        r0, [r0 + 2 * r1]
     movq       m0, [r0]
-
-    punpcklbw  m3, m0
-
-    pmaddubsw  m1, m6
-    pmaddubsw  m7, m3, m5
-
-    paddw      m1, m7
-
-    psubw      m1, m4
-    movu       [r2 + r3], m1
+    punpcklbw  m1, m0
+    pmaddubsw  m2, m1, [r6 + 2 * 16]
+    pmaddubsw  m1, [r6 + 3 * 16]
+    paddw      m6, m1                          ;m6=[1+2+3+4+5+6+7+8]   Row2 end
+    paddw      m4, m2                          ;m4=[3+4+5+6+7+8]       Row4
 
     movq       m1, [r0 + r1]
-
     punpcklbw  m0, m1
+    pmaddubsw  m0, [r6 + 3 * 16]
+    paddw      m5, m0                          ;m5=[2+3+4+5+6+7+8+9]   Row3 end
 
-    pmaddubsw  m2, m6
-    pmaddubsw  m0, m5
-
-    paddw      m2, m0
-
-    psubw      m2, m4
-    lea        r2, [r2 + 2 * r3]
-    movu       [r2], m2
-
-    movq       m2, [r0 + 2 * r1]
-
-    punpcklbw  m1, m2
-
-    pmaddubsw  m3, m6
-    pmaddubsw  m1, m5
-
-    paddw      m3, m1
-    psubw      m3, m4
-
-    movu       [r2 + r3], m3
-
-    lea        r2, [r2 + 2 * r3]
-
-    dec        r4d
-    jnz        .loop
-    RET
+    movq       m0, [r0 + 2 * r1]
+    punpcklbw  m1, m0
+    pmaddubsw  m1, [r6 + 3 * 16]
+    paddw      m4, m1                          ;m4=[3+4+5+6+7+8+9+10]  Row4 end
 %endmacro
 
-    FILTER_V_PS_W8_H8_H16_H32 8,  8
-    FILTER_V_PS_W8_H8_H16_H32 8, 16
-    FILTER_V_PS_W8_H8_H16_H32 8, 32
-
-;------------------------------------------------------------------------------------------------------------
-;void interp_4tap_vert_ps_6x8(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;------------------------------------------------------------------------------------------------------------
-%macro FILTER_V_PS_W6 2
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_%3_4x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_LUMA_4xN 3
 INIT_XMM sse4
-cglobal interp_4tap_vert_ps_6x%2, 4, 6, 8
-
-    mov        r4d, r4m
-    sub        r0, r1
-    add        r3d, r3d
+cglobal interp_8tap_vert_%3_%1x%2, 5, 7, 6
+    lea       r5, [3 * r1]
+    sub       r0, r5
+    shl       r4d, 6
+%ifidn %3,ps
+    add       r3d, r3d
+%endif
 
 %ifdef PIC
-    lea        r5, [tab_ChromaCoeff]
-    movd       m5, [r5 + r4 * 4]
+    lea       r5, [tab_LumaCoeffVer]
+    lea       r6, [r5 + r4]
 %else
-    movd       m5, [tab_ChromaCoeff + r4 * 4]
+    lea       r6, [tab_LumaCoeffVer + r4]
 %endif
 
-    pshufb     m6, m5, [tab_Vm]
-    pshufb     m5, [tab_Vm + 16]
-    mova       m4, [pw_2000]
-    lea        r5, [3 * r1]
-    mov        r4d, %2/4
-
-.loop:
-    movq       m0, [r0]
-    movq       m1, [r0 + r1]
-    movq       m2, [r0 + 2 * r1]
-    movq       m3, [r0 + r5]
-
-    punpcklbw  m0, m1
-    punpcklbw  m1, m2
-    punpcklbw  m2, m3
-
-    pmaddubsw  m0, m6
-    pmaddubsw  m7, m2, m5
-
-    paddw      m0, m7
-    psubw      m0, m4
-
-    movh       [r2], m0
-    pshufd     m0, m0, 2
-    movd       [r2 + 8], m0
-
-    lea        r0, [r0 + 4 * r1]
-    movq       m0, [r0]
-    punpcklbw  m3, m0
-
-    pmaddubsw  m1, m6
-    pmaddubsw  m7, m3, m5
+%ifidn %3,pp
+    mova      m3, [pw_512]
+%else
+    mova      m3, [pw_2000]
+%endif
 
-    paddw      m1, m7
-    psubw      m1, m4
+    mov       r4d, %2/4
+    lea       r5, [4 * r1]
 
-    movh       [r2 + r3], m1
-    pshufd     m1, m1, 2
-    movd       [r2 + r3 + 8], m1
+.loopH:
+    PROCESS_LUMA_W4_4R
 
-    movq       m1, [r0 + r1]
-    punpcklbw  m0, m1
+%ifidn %3,pp
+    pmulhrsw  m4, m3
+    pmulhrsw  m5, m3
 
-    pmaddubsw  m2, m6
-    pmaddubsw  m0, m5
+    packuswb  m4, m5
 
-    paddw      m2, m0
-    psubw      m2, m4
+    movd      [r2], m4
+    pextrd    [r2 + r3], m4, 1
+    lea       r2, [r2 + 2 * r3]
+    pextrd    [r2], m4, 2
+    pextrd    [r2 + r3], m4, 3
+%else
+    psubw     m4, m3
+    psubw     m5, m3
 
-    lea        r2,[r2 + 2 * r3]
-    movh       [r2], m2
-    pshufd     m2, m2, 2
-    movd       [r2 + 8], m2
+    movlps    [r2], m4
+    movhps    [r2 + r3], m4
+    lea       r2, [r2 + 2 * r3]
+    movlps    [r2], m5
+    movhps    [r2 + r3], m5
+%endif
 
-    movq       m2,[r0 + 2 * r1]
-    punpcklbw  m1, m2
+    sub       r0, r5
+    lea       r2, [r2 + 2 * r3]
 
-    pmaddubsw  m3, m6
-    pmaddubsw  m1, m5
+    dec       r4d
+    jnz       .loopH
 
-    paddw      m3, m1
-    psubw      m3, m4
+    RET
+%endmacro
 
-    movh       [r2 + r3], m3
-    pshufd     m3, m3, 2
-    movd       [r2 + r3 + 8], m3
 
-    lea        r2, [r2 + 2 * r3]
+INIT_YMM avx2
+cglobal interp_8tap_vert_pp_4x4, 4,6,8
+    mov             r4d, r4m
+    lea             r5, [r1 * 3]
+    sub             r0, r5
 
-    dec        r4d
-    jnz        .loop
-    RET
-%endmacro
+    ; TODO: VPGATHERDD
+    movd            xm1, [r0]                       ; m1 = row0
+    movd            xm2, [r0 + r1]                  ; m2 = row1
+    punpcklbw       xm1, xm2                        ; m1 = [13 03 12 02 11 01 10 00]
 
-    FILTER_V_PS_W6 6, 8
-    FILTER_V_PS_W6 6, 16
+    movd            xm3, [r0 + r1 * 2]              ; m3 = row2
+    punpcklbw       xm2, xm3                        ; m2 = [23 13 22 12 21 11 20 10]
+    movd            xm4, [r0 + r5]
+    punpcklbw       xm3, xm4                        ; m3 = [33 23 32 22 31 21 30 20]
+    punpcklwd       xm1, xm3                        ; m1 = [33 23 13 03 32 22 12 02 31 21 11 01 30 20 10 00]
 
-;---------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ps_12x16(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;---------------------------------------------------------------------------------------------------------------
-%macro FILTER_V_PS_W12 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_ps_12x%2, 4, 6, 8
+    lea             r0, [r0 + r1 * 4]
+    movd            xm5, [r0]                       ; m5 = row4
+    punpcklbw       xm4, xm5                        ; m4 = [43 33 42 32 41 31 40 30]
+    punpcklwd       xm2, xm4                        ; m2 = [43 33 21 13 42 32 22 12 41 31 21 11 40 30 20 10]
+    vinserti128     m1, m1, xm2, 1                  ; m1 = [43 33 21 13 42 32 22 12 41 31 21 11 40 30 20 10] - [33 23 13 03 32 22 12 02 31 21 11 01 30 20 10 00]
+    movd            xm2, [r0 + r1]                  ; m2 = row5
+    punpcklbw       xm5, xm2                        ; m5 = [53 43 52 42 51 41 50 40]
+    punpcklwd       xm3, xm5                        ; m3 = [53 43 44 23 52 42 32 22 51 41 31 21 50 40 30 20]
+    movd            xm6, [r0 + r1 * 2]              ; m6 = row6
+    punpcklbw       xm2, xm6                        ; m2 = [63 53 62 52 61 51 60 50]
+    punpcklwd       xm4, xm2                        ; m4 = [63 53 43 33 62 52 42 32 61 51 41 31 60 50 40 30]
+    vinserti128     m3, m3, xm4, 1                  ; m3 = [63 53 43 33 62 52 42 32 61 51 41 31 60 50 40 30] - [53 43 44 23 52 42 32 22 51 41 31 21 50 40 30 20]
+    movd            xm4, [r0 + r5]                  ; m4 = row7
+    punpcklbw       xm6, xm4                        ; m6 = [73 63 72 62 71 61 70 60]
+    punpcklwd       xm5, xm6                        ; m5 = [73 63 53 43 72 62 52 42 71 61 51 41 70 60 50 40]
 
-    mov        r4d, r4m
-    sub        r0, r1
-    add        r3d, r3d
+    lea             r0, [r0 + r1 * 4]
+    movd            xm7, [r0]                       ; m7 = row8
+    punpcklbw       xm4, xm7                        ; m4 = [83 73 82 72 81 71 80 70]
+    punpcklwd       xm2, xm4                        ; m2 = [83 73 63 53 82 72 62 52 81 71 61 51 80 70 60 50]
+    vinserti128     m5, m5, xm2, 1                  ; m5 = [83 73 63 53 82 72 62 52 81 71 61 51 80 70 60 50] - [73 63 53 43 72 62 52 42 71 61 51 41 70 60 50 40]
+    movd            xm2, [r0 + r1]                  ; m2 = row9
+    punpcklbw       xm7, xm2                        ; m7 = [93 83 92 82 91 81 90 80]
+    punpcklwd       xm6, xm7                        ; m6 = [93 83 73 63 92 82 72 62 91 81 71 61 90 80 70 60]
+    movd            xm7, [r0 + r1 * 2]              ; m7 = rowA
+    punpcklbw       xm2, xm7                        ; m2 = [A3 93 A2 92 A1 91 A0 90]
+    punpcklwd       xm4, xm2                        ; m4 = [A3 93 83 73 A2 92 82 72 A1 91 81 71 A0 90 80 70]
+    vinserti128     m6, m6, xm4, 1                  ; m6 = [A3 93 83 73 A2 92 82 72 A1 91 81 71 A0 90 80 70] - [93 83 73 63 92 82 72 62 91 81 71 61 90 80 70 60]
 
+    ; load filter coeff
 %ifdef PIC
-    lea        r5, [tab_ChromaCoeff]
-    movd       m0, [r5 + r4 * 4]
+    lea             r5, [tab_LumaCoeff]
+    vpbroadcastd    m0, [r5 + r4 * 8 + 0]
+    vpbroadcastd    m2, [r5 + r4 * 8 + 4]
 %else
-    movd       m0, [tab_ChromaCoeff + r4 * 4]
+    vpbroadcastd    m0, [tab_LumaCoeff + r4 * 8 + 0]
+    vpbroadcastd    m2, [tab_LumaCoeff + r4 * 8 + 4]
 %endif
 
-    pshufb     m1, m0, [tab_Vm]
-    pshufb     m0, [tab_Vm + 16]
-
-    mov        r4d, %2/2
-
-.loop:
-    movu       m2, [r0]
-    movu       m3, [r0 + r1]
-
-    punpcklbw  m4, m2, m3
-    punpckhbw  m2, m3
-
-    pmaddubsw  m4, m1
-    pmaddubsw  m2, m1
-
-    lea        r0, [r0 + 2 * r1]
-    movu       m5, [r0]
-    movu       m7, [r0 + r1]
-
-    punpcklbw  m6, m5, m7
-    pmaddubsw  m6, m0
-    paddw      m4, m6
-
-    punpckhbw  m6, m5, m7
-    pmaddubsw  m6, m0
-    paddw      m2, m6
-
-    mova       m6, [pw_2000]
-
-    psubw      m4, m6
-    psubw      m2, m6
-
-    movu       [r2], m4
-    movh       [r2 + 16], m2
-
-    punpcklbw  m4, m3, m5
-    punpckhbw  m3, m5
-
-    pmaddubsw  m4, m1
-    pmaddubsw  m3, m1
-
-    movu       m2, [r0 + 2 * r1]
-
-    punpcklbw  m5, m7, m2
-    punpckhbw  m7, m2
-
-    pmaddubsw  m5, m0
-    pmaddubsw  m7, m0
-
-    paddw      m4, m5
-    paddw      m3, m7
-
-    psubw      m4, m6
-    psubw      m3, m6
-
-    movu       [r2 + r3], m4
-    movh       [r2 + r3 + 16], m3
-
-    lea        r2, [r2 + 2 * r3]
+    pmaddubsw       m1, m0
+    pmaddubsw       m3, m0
+    pmaddubsw       m5, m2
+    pmaddubsw       m6, m2
+    vbroadcasti128  m0, [pw_1]
+    pmaddwd         m1, m0
+    pmaddwd         m3, m0
+    pmaddwd         m5, m0
+    pmaddwd         m6, m0
+    paddd           m1, m5                          ; m1 = DQWORD ROW[1 0]
+    paddd           m3, m6                          ; m3 = DQWORD ROW[3 2]
+    packssdw        m1, m3                          ; m1 =  QWORD ROW[3 1 2 0]
 
-    dec        r4d
-    jnz        .loop
+    ; TODO: does it overflow?
+    pmulhrsw        m1, [pw_512]
+    vextracti128    xm2, m1, 1
+    packuswb        xm1, xm2                        ; m1 =  DWORD ROW[3 1 2 0]
+    movd            [r2], xm1
+    pextrd          [r2 + r3], xm1, 2
+    pextrd          [r2 + r3 * 2], xm1, 1
+    lea             r4, [r3 * 3]
+    pextrd          [r2 + r4], xm1, 3
     RET
-%endmacro
-
-    FILTER_V_PS_W12 12, 16
-    FILTER_V_PS_W12 12, 32
-
-;---------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ps_16x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;---------------------------------------------------------------------------------------------------------------
-%macro FILTER_V_PS_W16 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_ps_%1x%2, 4, 6, 8
 
-    mov        r4d, r4m
-    sub        r0, r1
-    add        r3d, r3d
+INIT_YMM avx2
+cglobal interp_8tap_vert_ps_4x4, 4, 6, 5
+    mov             r4d, r4m
+    shl             r4d, 7
 
 %ifdef PIC
-    lea        r5, [tab_ChromaCoeff]
-    movd       m0, [r5 + r4 * 4]
+    lea             r5, [tab_LumaCoeffVer_32]
+    add             r5, r4
 %else
-    movd       m0, [tab_ChromaCoeff + r4 * 4]
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
 
-    pshufb     m1, m0, [tab_Vm]
-    pshufb     m0, [tab_Vm + 16]
-    mov        r4d, %2/2
-
-.loop:
-    movu       m2, [r0]
-    movu       m3, [r0 + r1]
-
-    punpcklbw  m4, m2, m3
-    punpckhbw  m2, m3
-
-    pmaddubsw  m4, m1
-    pmaddubsw  m2, m1
-
-    lea        r0, [r0 + 2 * r1]
-    movu       m5, [r0]
-    movu       m7, [r0 + r1]
-
-    punpcklbw  m6, m5, m7
-    pmaddubsw  m6, m0
-    paddw      m4, m6
-
-    punpckhbw  m6, m5, m7
-    pmaddubsw  m6, m0
-    paddw      m2, m6
-
-    mova       m6, [pw_2000]
-
-    psubw      m4, m6
-    psubw      m2, m6
-
-    movu       [r2], m4
-    movu       [r2 + 16], m2
-
-    punpcklbw  m4, m3, m5
-    punpckhbw  m3, m5
-
-    pmaddubsw  m4, m1
-    pmaddubsw  m3, m1
-
-    movu       m5, [r0 + 2 * r1]
-
-    punpcklbw  m2, m7, m5
-    punpckhbw  m7, m5
-
-    pmaddubsw  m2, m0
-    pmaddubsw  m7, m0
+    lea             r4, [r1 * 3]
+    sub             r0, r4
 
-    paddw      m4, m2
-    paddw      m3, m7
+    add             r3d, r3d
 
-    psubw      m4, m6
-    psubw      m3, m6
-
-    movu       [r2 + r3], m4
-    movu       [r2 + r3 + 16], m3
+    movd            xm1, [r0]
+    pinsrd          xm1, [r0 + r1], 1
+    pinsrd          xm1, [r0 + r1 * 2], 2
+    pinsrd          xm1, [r0 + r4], 3                       ; m1 = row[3 2 1 0]
+    lea             r0, [r0 + r1 * 4]
+    movd            xm2, [r0]
+    pinsrd          xm2, [r0 + r1], 1
+    pinsrd          xm2, [r0 + r1 * 2], 2
+    pinsrd          xm2, [r0 + r4], 3                       ; m2 = row[7 6 5 4]
+    vinserti128     m1, m1, xm2, 1                          ; m1 = row[7 6 5 4 3 2 1 0]
+    lea             r0, [r0 + r1 * 4]
+    movd            xm3, [r0]
+    pinsrd          xm3, [r0 + r1], 1
+    pinsrd          xm3, [r0 + r1 * 2], 2                   ; m3 = row[x 10 9 8]
+    vinserti128     m2, m2, xm3, 1                          ; m2 = row[x 10 9 8 7 6 5 4]
+    mova            m3, [interp4_vpp_shuf1]
+    vpermd          m0, m3, m1                              ; m0 = row[4 3 3 2 2 1 1 0]
+    vpermd          m4, m3, m2                              ; m4 = row[8 7 7 6 6 5 5 4]
+    mova            m3, [interp4_vpp_shuf1 + mmsize]
+    vpermd          m1, m3, m1                              ; m1 = row[6 5 5 4 4 3 3 2]
+    vpermd          m2, m3, m2                              ; m2 = row[10 9 9 8 8 7 7 6]
 
-    lea        r2, [r2 + 2 * r3]
+    mova            m3, [interp4_vpp_shuf]
+    pshufb          m0, m0, m3
+    pshufb          m1, m1, m3
+    pshufb          m4, m4, m3
+    pshufb          m2, m2, m3
+    pmaddubsw       m0, [r5]
+    pmaddubsw       m1, [r5 + mmsize]
+    pmaddubsw       m4, [r5 + 2 * mmsize]
+    pmaddubsw       m2, [r5 + 3 * mmsize]
+    paddw           m0, m1
+    paddw           m0, m4
+    paddw           m0, m2                                  ; m0 = WORD ROW[3 2 1 0]
 
-    dec        r4d
-    jnz        .loop
+    psubw           m0, [pw_2000]
+    vextracti128    xm2, m0, 1
+    lea             r5, [r3 * 3]
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r5], xm2
     RET
-%endmacro
-
-    FILTER_V_PS_W16 16,  4
-    FILTER_V_PS_W16 16,  8
-    FILTER_V_PS_W16 16, 12
-    FILTER_V_PS_W16 16, 16
-    FILTER_V_PS_W16 16, 32
-
-    FILTER_V_PS_W16 16, 24
-    FILTER_V_PS_W16 16, 64
-
-;--------------------------------------------------------------------------------------------------------------
-;void interp_4tap_vert_ps_24x32(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-%macro FILTER_V4_PS_W24 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_ps_24x%2, 4, 6, 8
 
-    mov        r4d, r4m
-    sub        r0, r1
-    add        r3d, r3d
+%macro FILTER_VER_LUMA_AVX2_4xN 3
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%3_%1x%2, 4, 9, 10
+    mov             r4d, r4m
+    shl             r4d, 7
 
 %ifdef PIC
-    lea        r5, [tab_ChromaCoeff]
-    movd       m0, [r5 + r4 * 4]
+    lea             r5, [tab_LumaCoeffVer_32]
+    add             r5, r4
 %else
-    movd       m0, [tab_ChromaCoeff + r4 * 4]
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
 
-    pshufb     m1, m0, [tab_Vm]
-    pshufb     m0, [tab_Vm + 16]
-
-    mov        r4d, %2/2
-
-.loop:
-    movu       m2, [r0]
-    movu       m3, [r0 + r1]
-
-    punpcklbw  m4, m2, m3
-    punpckhbw  m2, m3
-
-    pmaddubsw  m4, m1
-    pmaddubsw  m2, m1
-
-    lea        r5, [r0 + 2 * r1]
-
-    movu       m5, [r5]
-    movu       m7, [r5 + r1]
-
-    punpcklbw  m6, m5, m7
-    pmaddubsw  m6, m0
-    paddw      m4, m6
-
-    punpckhbw  m6, m5, m7
-    pmaddubsw  m6, m0
-    paddw      m2, m6
-
-    mova       m6, [pw_2000]
-
-    psubw      m4, m6
-    psubw      m2, m6
-
-    movu       [r2], m4
-    movu       [r2 + 16], m2
-
-    punpcklbw  m4, m3, m5
-    punpckhbw  m3, m5
-
-    pmaddubsw  m4, m1
-    pmaddubsw  m3, m1
-
-    movu       m2, [r5 + 2 * r1]
-
-    punpcklbw  m5, m7, m2
-    punpckhbw  m7, m2
-
-    pmaddubsw  m5, m0
-    pmaddubsw  m7, m0
-
-    paddw      m4, m5
-    paddw      m3, m7
-
-    psubw      m4, m6
-    psubw      m3, m6
-
-    movu       [r2 + r3], m4
-    movu       [r2 + r3 + 16], m3
-
-    movq       m2, [r0 + 16]
-    movq       m3, [r0 + r1 + 16]
-    movq       m4, [r5 + 16]
-    movq       m5, [r5 + r1 + 16]
-
-    punpcklbw  m2, m3
-    punpcklbw  m7, m4, m5
-
-    pmaddubsw  m2, m1
-    pmaddubsw  m7, m0
-
-    paddw      m2, m7
-    psubw      m2, m6
-
-    movu       [r2 + 32], m2
-
-    movq       m2, [r5 + 2 * r1 + 16]
-
-    punpcklbw  m3, m4
-    punpcklbw  m5, m2
-
-    pmaddubsw  m3, m1
-    pmaddubsw  m5, m0
-
-    paddw      m3, m5
-    psubw      m3,  m6
-
-    movu       [r2 + r3 + 32], m3
-
-    mov        r0, r5
-    lea        r2, [r2 + 2 * r3]
-
-    dec        r4d
-    jnz        .loop
-    RET
-%endmacro
-
-    FILTER_V4_PS_W24 24, 32
-
-    FILTER_V4_PS_W24 24, 64
-
-;---------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ps_32x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;---------------------------------------------------------------------------------------------------------------
-%macro FILTER_V_PS_W32 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_ps_%1x%2, 4, 6, 8
-
-    mov        r4d, r4m
-    sub        r0, r1
-    add        r3d, r3d
-
-%ifdef PIC
-    lea        r5, [tab_ChromaCoeff]
-    movd       m0, [r5 + r4 * 4]
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+    lea             r6, [r1 * 4]
+%ifidn %3,pp
+    mova            m6, [pw_512]
 %else
-    movd       m0, [tab_ChromaCoeff + r4 * 4]
+    add             r3d, r3d
+    vbroadcasti128  m6, [pw_2000]
 %endif
-
-    pshufb     m1, m0, [tab_Vm]
-    pshufb     m0, [tab_Vm + 16]
-
-    mova       m7, [pw_2000]
-
-    mov        r4d, %2
-
+    lea             r8, [r3 * 3]
+    mova            m5, [interp4_vpp_shuf]
+    mova            m0, [interp4_vpp_shuf1]
+    mova            m7, [interp4_vpp_shuf1 + mmsize]
+    mov             r7d, %2 / 8
 .loop:
-    movu       m2, [r0]
-    movu       m3, [r0 + r1]
-
-    punpcklbw  m4, m2, m3
-    punpckhbw  m2, m3
-
-    pmaddubsw  m4, m1
-    pmaddubsw  m2, m1
-
-    lea        r5, [r0 + 2 * r1]
-    movu       m3, [r5]
-    movu       m5, [r5 + r1]
-
-    punpcklbw  m6, m3, m5
-    punpckhbw  m3, m5
-
-    pmaddubsw  m6, m0
-    pmaddubsw  m3, m0
-
-    paddw      m4, m6
-    paddw      m2, m3
-
-    psubw      m4, m7
-    psubw      m2, m7
-
-    movu       [r2], m4
-    movu       [r2 + 16], m2
-
-    movu       m2, [r0 + 16]
-    movu       m3, [r0 + r1 + 16]
-
-    punpcklbw  m4, m2, m3
-    punpckhbw  m2, m3
-
-    pmaddubsw  m4, m1
-    pmaddubsw  m2, m1
-
-    movu       m3, [r5 + 16]
-    movu       m5, [r5 + r1 + 16]
-
-    punpcklbw  m6, m3, m5
-    punpckhbw  m3, m5
-
-    pmaddubsw  m6, m0
-    pmaddubsw  m3, m0
-
-    paddw      m4, m6
-    paddw      m2, m3
-
-    psubw      m4, m7
-    psubw      m2, m7
-
-    movu       [r2 + 32], m4
-    movu       [r2 + 48], m2
+    movd            xm1, [r0]
+    pinsrd          xm1, [r0 + r1], 1
+    pinsrd          xm1, [r0 + r1 * 2], 2
+    pinsrd          xm1, [r0 + r4], 3                       ; m1 = row[3 2 1 0]
+    lea             r0, [r0 + r1 * 4]
+    movd            xm2, [r0]
+    pinsrd          xm2, [r0 + r1], 1
+    pinsrd          xm2, [r0 + r1 * 2], 2
+    pinsrd          xm2, [r0 + r4], 3                       ; m2 = row[7 6 5 4]
+    vinserti128     m1, m1, xm2, 1                          ; m1 = row[7 6 5 4 3 2 1 0]
+    lea             r0, [r0 + r1 * 4]
+    movd            xm3, [r0]
+    pinsrd          xm3, [r0 + r1], 1
+    pinsrd          xm3, [r0 + r1 * 2], 2
+    pinsrd          xm3, [r0 + r4], 3                       ; m3 = row[11 10 9 8]
+    vinserti128     m2, m2, xm3, 1                          ; m2 = row[11 10 9 8 7 6 5 4]
+    lea             r0, [r0 + r1 * 4]
+    movd            xm4, [r0]
+    pinsrd          xm4, [r0 + r1], 1
+    pinsrd          xm4, [r0 + r1 * 2], 2                   ; m4 = row[x 14 13 12]
+    vinserti128     m3, m3, xm4, 1                          ; m3 = row[x 14 13 12 11 10 9 8]
+    vpermd          m8, m0, m1                              ; m8 = row[4 3 3 2 2 1 1 0]
+    vpermd          m4, m0, m2                              ; m4 = row[8 7 7 6 6 5 5 4]
+    vpermd          m1, m7, m1                              ; m1 = row[6 5 5 4 4 3 3 2]
+    vpermd          m2, m7, m2                              ; m2 = row[10 9 9 8 8 7 7 6]
+    vpermd          m9, m0, m3                              ; m9 = row[12 11 11 10 10 9 9 8]
+    vpermd          m3, m7, m3                              ; m3 = row[14 13 13 12 12 11 11 10]
 
-    lea        r0, [r0 + r1]
-    lea        r2, [r2 + r3]
+    pshufb          m8, m8, m5
+    pshufb          m1, m1, m5
+    pshufb          m4, m4, m5
+    pshufb          m9, m9, m5
+    pshufb          m2, m2, m5
+    pshufb          m3, m3, m5
+    pmaddubsw       m8, [r5]
+    pmaddubsw       m1, [r5 + mmsize]
+    pmaddubsw       m9, [r5 + 2 * mmsize]
+    pmaddubsw       m3, [r5 + 3 * mmsize]
+    paddw           m8, m1
+    paddw           m9, m3
+    pmaddubsw       m1, m4, [r5 + 2 * mmsize]
+    pmaddubsw       m3, m2, [r5 + 3 * mmsize]
+    pmaddubsw       m4, [r5]
+    pmaddubsw       m2, [r5 + mmsize]
+    paddw           m3, m1
+    paddw           m2, m4
+    paddw           m8, m3                                  ; m8 = WORD ROW[3 2 1 0]
+    paddw           m9, m2                                  ; m9 = WORD ROW[7 6 5 4]
 
-    dec        r4d
-    jnz        .loop
+%ifidn %3,pp
+    pmulhrsw        m8, m6
+    pmulhrsw        m9, m6
+    packuswb        m8, m9
+    vextracti128    xm1, m8, 1
+    movd            [r2], xm8
+    pextrd          [r2 + r3], xm8, 1
+    movd            [r2 + r3 * 2], xm1
+    pextrd          [r2 + r8], xm1, 1
+    lea             r2, [r2 + r3 * 4]
+    pextrd          [r2], xm8, 2
+    pextrd          [r2 + r3], xm8, 3
+    pextrd          [r2 + r3 * 2], xm1, 2
+    pextrd          [r2 + r8], xm1, 3
+%else
+    psubw           m8, m6
+    psubw           m9, m6
+    vextracti128    xm1, m8, 1
+    vextracti128    xm2, m9, 1
+    movq            [r2], xm8
+    movhps          [r2 + r3], xm8
+    movq            [r2 + r3 * 2], xm1
+    movhps          [r2 + r8], xm1
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm9
+    movhps          [r2 + r3], xm9
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r8], xm2
+%endif
+    lea             r2, [r2 + r3 * 4]
+    sub             r0, r6
+    dec             r7d
+    jnz             .loop
     RET
+%endif
 %endmacro
 
-    FILTER_V_PS_W32 32,  8
-    FILTER_V_PS_W32 32, 16
-    FILTER_V_PS_W32 32, 24
-    FILTER_V_PS_W32 32, 32
-
-    FILTER_V_PS_W32 32, 48
-    FILTER_V_PS_W32 32, 64
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_pp_4x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_4xN 4, 4, pp
 
-;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro FILTER_V4_W8_H8_H16_H32 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_pp_%1x%2, 4, 6, 8
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_pp_4x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_4xN 4, 8, pp
+    FILTER_VER_LUMA_AVX2_4xN 4, 8, pp
 
-    mov         r4d,       r4m
-    sub         r0,        r1
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_pp_4x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_4xN 4, 16, pp
+    FILTER_VER_LUMA_AVX2_4xN 4, 16, pp
 
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeff]
-    movd        m5,        [r5 + r4 * 4]
-%else
-    movd        m5,        [tab_ChromaCoeff + r4 * 4]
-%endif
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_ps_4x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_4xN 4, 4, ps
 
-    pshufb      m6,        m5,       [tab_Vm]
-    pshufb      m5,        [tab_Vm + 16]
-    mova        m4,        [pw_512]
-    lea         r5,        [r1 * 3]
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_ps_4x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_4xN 4, 8, ps
+    FILTER_VER_LUMA_AVX2_4xN 4, 8, ps
 
-    mov         r4d,       %2
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_ps_4x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_4xN 4, 16, ps
+    FILTER_VER_LUMA_AVX2_4xN 4, 16, ps
 
-.loop:
-    movq        m0,        [r0]
-    movq        m1,        [r0 + r1]
-    movq        m2,        [r0 + 2 * r1]
-    movq        m3,        [r0 + r5]
-
-    punpcklbw   m0,        m1
-    punpcklbw   m1,        m2
-    punpcklbw   m2,        m3
-
-    pmaddubsw   m0,        m6
-    pmaddubsw   m7,        m2, m5
-
-    paddw       m0,        m7
-
-    pmulhrsw    m0,        m4
-    packuswb    m0,        m0
-    movh        [r2],      m0
-
-    lea         r0,        [r0 + 4 * r1]
-    movq        m0,        [r0]
-
-    punpcklbw   m3,        m0
-
-    pmaddubsw   m1,        m6
-    pmaddubsw   m7,        m3, m5
-
-    paddw       m1,        m7
-
-    pmulhrsw    m1,        m4
-    packuswb    m1,        m1
-    movh        [r2 + r3], m1
-
-    movq        m1,        [r0 + r1]
-
-    punpcklbw   m0,        m1
-
-    pmaddubsw   m2,        m6
-    pmaddubsw   m0,        m5
-
-    paddw       m2,        m0
-
-    pmulhrsw    m2,        m4
-
-    movq        m7,        [r0 + 2 * r1]
-    punpcklbw   m1,        m7
-
-    pmaddubsw   m3,        m6
-    pmaddubsw   m1,        m5
-
-    paddw       m3,        m1
-
-    pmulhrsw    m3,        m4
-    packuswb    m2,        m3
-
-    lea         r2,        [r2 + 2 * r3]
-    movh        [r2],      m2
-    movhps      [r2 + r3], m2
-
-    lea         r2,        [r2 + 2 * r3]
-
-    sub         r4,         4
-    jnz        .loop
-    RET
-%endmacro
-
-    FILTER_V4_W8_H8_H16_H32 8,  8
-    FILTER_V4_W8_H8_H16_H32 8, 16
-    FILTER_V4_W8_H8_H16_H32 8, 32
-
-    FILTER_V4_W8_H8_H16_H32 8, 12
-    FILTER_V4_W8_H8_H16_H32 8, 64
-
-%macro PROCESS_CHROMA_AVX2_W8_8R 0
+%macro PROCESS_LUMA_AVX2_W8_8R 0
     movq            xm1, [r0]                       ; m1 = row 0
     movq            xm2, [r0 + r1]                  ; m2 = row 1
     punpcklbw       xm1, xm2                        ; m1 = [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
@@ -5154,6 +3183,8 @@ cglobal interp_4tap_vert_pp_%1x%2, 4, 6, 8
     movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
     punpcklbw       xm3, xm4                        ; m3 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50]
     vinserti128     m1, m1, xm3, 1                  ; m1 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50] - [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
+    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
+    paddw           m5, m3
     pmaddubsw       m0, m1, [r5 + 1 * mmsize]
     paddw           m2, m0
     pmaddubsw       m1, [r5]
@@ -5163,6 +3194,10 @@ cglobal interp_4tap_vert_pp_%1x%2, 4, 6, 8
     movq            xm0, [r0]                       ; m0 = row 8
     punpcklbw       xm3, xm0                        ; m3 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70]
     vinserti128     m4, m4, xm3, 1                  ; m4 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70] - [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
+    pmaddubsw       m3, m4, [r5 + 3 * mmsize]
+    paddw           m5, m3
+    pmaddubsw       m3, m4, [r5 + 2 * mmsize]
+    paddw           m2, m3
     pmaddubsw       m3, m4, [r5 + 1 * mmsize]
     paddw           m1, m3
     pmaddubsw       m4, [r5]
@@ -5171,90 +3206,33 @@ cglobal interp_4tap_vert_pp_%1x%2, 4, 6, 8
     movq            xm6, [r0 + r1 * 2]              ; m6 = row 10
     punpcklbw       xm3, xm6                        ; m3 = [A7 97 A6 96 A5 95 A4 94 A3 93 A2 92 A1 91 A0 90]
     vinserti128     m0, m0, xm3, 1                  ; m0 = [A7 97 A6 96 A5 95 A4 94 A3 93 A2 92 A1 91 A0 90] - [97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80]
+    pmaddubsw       m3, m0, [r5 + 3 * mmsize]
+    paddw           m2, m3
+    pmaddubsw       m3, m0, [r5 + 2 * mmsize]
+    paddw           m1, m3
     pmaddubsw       m0, [r5 + 1 * mmsize]
     paddw           m4, m0
-%endmacro
-
-%macro FILTER_VER_CHROMA_AVX2_8x8 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x8, 4, 6, 7
-    mov             r4d, r4m
-    shl             r4d, 6
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
 
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-    PROCESS_CHROMA_AVX2_W8_8R
-%ifidn %1,pp
-    lea             r4, [r3 * 3]
-    mova            m3, [pw_512]
-    pmulhrsw        m5, m3                          ; m5 = word: row 0, row 1
-    pmulhrsw        m2, m3                          ; m2 = word: row 2, row 3
-    pmulhrsw        m1, m3                          ; m1 = word: row 4, row 5
-    pmulhrsw        m4, m3                          ; m4 = word: row 6, row 7
-    packuswb        m5, m2
-    packuswb        m1, m4
-    vextracti128    xm2, m5, 1
-    vextracti128    xm4, m1, 1
-    movq            [r2], xm5
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm5
-    movhps          [r2 + r4], xm2
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm1
-    movq            [r2 + r3], xm4
-    movhps          [r2 + r3 * 2], xm1
-    movhps          [r2 + r4], xm4
-%else
-    add             r3d, r3d
-    vbroadcasti128  m3, [pw_2000]
-    lea             r4, [r3 * 3]
-    psubw           m5, m3                          ; m5 = word: row 0, row 1
-    psubw           m2, m3                          ; m2 = word: row 2, row 3
-    psubw           m1, m3                          ; m1 = word: row 4, row 5
-    psubw           m4, m3                          ; m4 = word: row 6, row 7
-    vextracti128    xm6, m5, 1
-    vextracti128    xm3, m2, 1
-    vextracti128    xm0, m1, 1
-    movu            [r2], xm5
-    movu            [r2 + r3], xm6
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r4], xm3
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm1
-    movu            [r2 + r3], xm0
-    movu            [r2 + r3 * 2], xm4
-    vextracti128    xm4, m4, 1
-    movu            [r2 + r4], xm4
-%endif
-    RET
+    movq            xm3, [r0 + r4]                  ; m3 = row 11
+    punpcklbw       xm6, xm3                        ; m6 = [B7 A7 B6 A6 B5 A5 B4 A4 B3 A3 B2 A2 B1 A1 B0 A0]
+    lea             r0, [r0 + r1 * 4]
+    movq            xm0, [r0]                       ; m0 = row 12
+    punpcklbw       xm3, xm0                        ; m3 = [C7 B7 C6 B6 C5 B5 C4 B4 C3 B3 C2 B2 C1 B1 C0 B0]
+    vinserti128     m6, m6, xm3, 1                  ; m6 = [C7 B7 C6 B6 C5 B5 C4 B4 C3 B3 C2 B2 C1 B1 C0 B0] - [B7 A7 B6 A6 B5 A5 B4 A4 B3 A3 B2 A2 B1 A1 B0 A0]
+    pmaddubsw       m3, m6, [r5 + 3 * mmsize]
+    paddw           m1, m3
+    pmaddubsw       m6, [r5 + 2 * mmsize]
+    paddw           m4, m6
+    movq            xm3, [r0 + r1]                  ; m3 = row 13
+    punpcklbw       xm0, xm3                        ; m0 = [D7 C7 D6 C6 D5 C5 D4 C4 D3 C3 D2 C2 D1 C1 D0 C0]
+    movq            xm6, [r0 + r1 * 2]              ; m6 = row 14
+    punpcklbw       xm3, xm6                        ; m3 = [E7 D7 E6 D6 E5 D5 E4 D4 E3 D3 E2 D2 E1 D1 E0 D0]
+    vinserti128     m0, m0, xm3, 1                  ; m0 = [E7 D7 E6 D6 E5 D5 E4 D4 E3 D3 E2 D2 E1 D1 E0 D0] - [D7 C7 D6 C6 D5 C5 D4 C4 D3 C3 D2 C2 D1 C1 D0 C0]
+    pmaddubsw       m0, [r5 + 3 * mmsize]
+    paddw           m4, m0
 %endmacro
 
-    FILTER_VER_CHROMA_AVX2_8x8 pp
-    FILTER_VER_CHROMA_AVX2_8x8 ps
-
-%macro FILTER_VER_CHROMA_AVX2_8x6 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x6, 4, 6, 6
-    mov             r4d, r4m
-    shl             r4d, 6
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-
+%macro PROCESS_LUMA_AVX2_W8_4R 0
     movq            xm1, [r0]                       ; m1 = row 0
     movq            xm2, [r0 + r1]                  ; m2 = row 1
     punpcklbw       xm1, xm2                        ; m1 = [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
@@ -5276,105 +3254,199 @@ cglobal interp_4tap_vert_%1_8x6, 4, 6, 6
     movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
     punpcklbw       xm3, xm4                        ; m3 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50]
     vinserti128     m1, m1, xm3, 1                  ; m1 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50] - [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
+    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
+    paddw           m5, m3
     pmaddubsw       m0, m1, [r5 + 1 * mmsize]
     paddw           m2, m0
-    pmaddubsw       m1, [r5]
     movq            xm3, [r0 + r4]                  ; m3 = row 7
     punpcklbw       xm4, xm3                        ; m4 = [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
     lea             r0, [r0 + r1 * 4]
     movq            xm0, [r0]                       ; m0 = row 8
     punpcklbw       xm3, xm0                        ; m3 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70]
     vinserti128     m4, m4, xm3, 1                  ; m4 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70] - [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
-    pmaddubsw       m4, [r5 + 1 * mmsize]
-    paddw           m1, m4
-%ifidn %1,pp
-    lea             r4, [r3 * 3]
-    mova            m3, [pw_512]
-    pmulhrsw        m5, m3                          ; m5 = word: row 0, row 1
-    pmulhrsw        m2, m3                          ; m2 = word: row 2, row 3
-    pmulhrsw        m1, m3                          ; m1 = word: row 4, row 5
+    pmaddubsw       m3, m4, [r5 + 3 * mmsize]
+    paddw           m5, m3
+    pmaddubsw       m3, m4, [r5 + 2 * mmsize]
+    paddw           m2, m3
+    movq            xm3, [r0 + r1]                  ; m3 = row 9
+    punpcklbw       xm0, xm3                        ; m0 = [97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80]
+    movq            xm6, [r0 + r1 * 2]              ; m6 = row 10
+    punpcklbw       xm3, xm6                        ; m3 = [A7 97 A6 96 A5 95 A4 94 A3 93 A2 92 A1 91 A0 90]
+    vinserti128     m0, m0, xm3, 1                  ; m0 = [A7 97 A6 96 A5 95 A4 94 A3 93 A2 92 A1 91 A0 90] - [97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80]
+    pmaddubsw       m3, m0, [r5 + 3 * mmsize]
+    paddw           m2, m3
+%endmacro
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_%3_8x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_LUMA_8xN 3
+INIT_XMM sse4
+cglobal interp_8tap_vert_%3_%1x%2, 5, 7, 8
+    lea       r5, [3 * r1]
+    sub       r0, r5
+    shl       r4d, 6
+
+%ifidn %3,ps
+    add       r3d, r3d
+%endif
+
+%ifdef PIC
+    lea       r5, [tab_LumaCoeffVer]
+    lea       r6, [r5 + r4]
+%else
+    lea       r6, [tab_LumaCoeffVer + r4]
+%endif
+
+ %ifidn %3,pp
+    mova      m3, [pw_512]
+%else
+    mova      m3, [pw_2000]
+%endif
+
+    mov       r4d, %2/4
+    lea       r5, [4 * r1]
+
+.loopH:
+    PROCESS_LUMA_W8_4R
+
+%ifidn %3,pp
+    pmulhrsw  m7, m3
+    pmulhrsw  m6, m3
+    pmulhrsw  m5, m3
+    pmulhrsw  m4, m3
+
+    packuswb  m7, m6
+    packuswb  m5, m4
+
+    movlps    [r2], m7
+    movhps    [r2 + r3], m7
+    lea       r2, [r2 + 2 * r3]
+    movlps    [r2], m5
+    movhps    [r2 + r3], m5
+%else
+    psubw     m7, m3
+    psubw     m6, m3
+    psubw     m5, m3
+    psubw     m4, m3
+
+    movu      [r2], m7
+    movu      [r2 + r3], m6
+    lea       r2, [r2 + 2 * r3]
+    movu      [r2], m5
+    movu      [r2 + r3], m4
+%endif
+
+    sub       r0, r5
+    lea       r2, [r2 + 2 * r3]
+
+    dec       r4d
+    jnz       .loopH
+
+    RET
+%endmacro
+
+%macro FILTER_VER_LUMA_AVX2_8xN 3
+INIT_YMM avx2
+cglobal interp_8tap_vert_%3_%1x%2, 4, 7, 8, 0-gprsize
+    mov             r4d, r4m
+    shl             r4d, 7
+
+%ifdef PIC
+    lea             r5, [tab_LumaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
+%endif
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+    lea             r6, [r1 * 4]
+%ifidn %3,pp
+    mova            m7, [pw_512]
+%else
+    add             r3d, r3d
+    vbroadcasti128  m7, [pw_2000]
+%endif
+    mov             word [rsp], %2 / 8
+
+.loop:
+    PROCESS_LUMA_AVX2_W8_8R
+%ifidn %3,pp
+    pmulhrsw        m5, m7                          ; m5 = word: row 0, row 1
+    pmulhrsw        m2, m7                          ; m2 = word: row 2, row 3
+    pmulhrsw        m1, m7                          ; m1 = word: row 4, row 5
+    pmulhrsw        m4, m7                          ; m4 = word: row 6, row 7
     packuswb        m5, m2
-    packuswb        m1, m1
+    packuswb        m1, m4
     vextracti128    xm2, m5, 1
     vextracti128    xm4, m1, 1
     movq            [r2], xm5
     movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm5
-    movhps          [r2 + r4], xm2
-    lea             r2, [r2 + r3 * 4]
+    lea             r2, [r2 + r3 * 2]
+    movhps          [r2], xm5
+    movhps          [r2 + r3], xm2
+    lea             r2, [r2 + r3 * 2]
     movq            [r2], xm1
     movq            [r2 + r3], xm4
+    lea             r2, [r2 + r3 * 2]
+    movhps          [r2], xm1
+    movhps          [r2 + r3], xm4
 %else
-    add             r3d, r3d
-    mova            m3, [pw_2000]
-    lea             r4, [r3 * 3]
-    psubw           m5, m3                          ; m5 = word: row 0, row 1
-    psubw           m2, m3                          ; m2 = word: row 2, row 3
-    psubw           m1, m3                          ; m1 = word: row 4, row 5
-    vextracti128    xm4, m5, 1
+    psubw           m5, m7                          ; m5 = word: row 0, row 1
+    psubw           m2, m7                          ; m2 = word: row 2, row 3
+    psubw           m1, m7                          ; m1 = word: row 4, row 5
+    psubw           m4, m7                          ; m4 = word: row 6, row 7
+    vextracti128    xm6, m5, 1
     vextracti128    xm3, m2, 1
     vextracti128    xm0, m1, 1
     movu            [r2], xm5
-    movu            [r2 + r3], xm4
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r4], xm3
-    lea             r2, [r2 + r3 * 4]
+    movu            [r2 + r3], xm6
+    lea             r2, [r2 + r3 * 2]
+    movu            [r2], xm2
+    movu            [r2 + r3], xm3
+    lea             r2, [r2 + r3 * 2]
     movu            [r2], xm1
     movu            [r2 + r3], xm0
+    lea             r2, [r2 + r3 * 2]
+    movu            [r2], xm4
+    vextracti128    xm4, m4, 1
+    movu            [r2 + r3], xm4
 %endif
+    lea             r2, [r2 + r3 * 2]
+    sub             r0, r6
+    dec             word [rsp]
+    jnz             .loop
     RET
 %endmacro
 
-    FILTER_VER_CHROMA_AVX2_8x6 pp
-    FILTER_VER_CHROMA_AVX2_8x6 ps
+%macro FILTER_VER_LUMA_AVX2_8x8 1
+INIT_YMM avx2
+cglobal interp_8tap_vert_%1_8x8, 4, 6, 7
+    mov             r4d, r4m
+    shl             r4d, 7
 
-%macro PROCESS_CHROMA_AVX2_W8_16R 1
-    movq            xm1, [r0]                       ; m1 = row 0
-    movq            xm2, [r0 + r1]                  ; m2 = row 1
-    punpcklbw       xm1, xm2
-    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
-    punpcklbw       xm2, xm3
-    vinserti128     m5, m1, xm2, 1
-    pmaddubsw       m5, [r5]
-    movq            xm4, [r0 + r4]                  ; m4 = row 3
-    punpcklbw       xm3, xm4
-    lea             r0, [r0 + r1 * 4]
-    movq            xm1, [r0]                       ; m1 = row 4
-    punpcklbw       xm4, xm1
-    vinserti128     m2, m3, xm4, 1
-    pmaddubsw       m0, m2, [r5 + 1 * mmsize]
-    paddw           m5, m0
-    pmaddubsw       m2, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 5
-    punpcklbw       xm1, xm3
-    movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
-    punpcklbw       xm3, xm4
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m0, m1, [r5 + 1 * mmsize]
-    paddw           m2, m0
-    pmaddubsw       m1, [r5]
-    movq            xm3, [r0 + r4]                  ; m3 = row 7
-    punpcklbw       xm4, xm3
-    lea             r0, [r0 + r1 * 4]
-    movq            xm0, [r0]                       ; m0 = row 8
-    punpcklbw       xm3, xm0
-    vinserti128     m4, m4, xm3, 1
-    pmaddubsw       m3, m4, [r5 + 1 * mmsize]
-    paddw           m1, m3
-    pmaddubsw       m4, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 9
-    punpcklbw       xm0, xm3
-    movq            xm6, [r0 + r1 * 2]              ; m6 = row 10
-    punpcklbw       xm3, xm6
-    vinserti128     m0, m0, xm3, 1
-    pmaddubsw       m3, m0, [r5 + 1 * mmsize]
-    paddw           m4, m3
-    pmaddubsw       m0, [r5]
+%ifdef PIC
+    lea             r5, [tab_LumaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+    PROCESS_LUMA_AVX2_W8_8R
 %ifidn %1,pp
-    pmulhrsw        m5, m7                          ; m5 = word: row 0, row 1
-    pmulhrsw        m2, m7                          ; m2 = word: row 2, row 3
-    pmulhrsw        m1, m7                          ; m1 = word: row 4, row 5
-    pmulhrsw        m4, m7                          ; m4 = word: row 6, row 7
+    mova            m3, [pw_512]
+%else
+    add             r3d, r3d
+    vbroadcasti128  m3, [pw_2000]
+%endif
+    lea             r4, [r3 * 3]
+%ifidn %1,pp
+    pmulhrsw        m5, m3                          ; m5 = word: row 0, row 1
+    pmulhrsw        m2, m3                          ; m2 = word: row 2, row 3
+    pmulhrsw        m1, m3                          ; m1 = word: row 4, row 5
+    pmulhrsw        m4, m3                          ; m4 = word: row 6, row 7
     packuswb        m5, m2
     packuswb        m1, m4
     vextracti128    xm2, m5, 1
@@ -5382,886 +3454,377 @@ cglobal interp_4tap_vert_%1_8x6, 4, 6, 6
     movq            [r2], xm5
     movq            [r2 + r3], xm2
     movhps          [r2 + r3 * 2], xm5
-    movhps          [r2 + r6], xm2
+    movhps          [r2 + r4], xm2
     lea             r2, [r2 + r3 * 4]
     movq            [r2], xm1
     movq            [r2 + r3], xm4
     movhps          [r2 + r3 * 2], xm1
-    movhps          [r2 + r6], xm4
+    movhps          [r2 + r4], xm4
 %else
-    psubw           m5, m7                          ; m5 = word: row 0, row 1
-    psubw           m2, m7                          ; m2 = word: row 2, row 3
-    psubw           m1, m7                          ; m1 = word: row 4, row 5
-    psubw           m4, m7                          ; m4 = word: row 6, row 7
-    vextracti128    xm3, m5, 1
-    movu            [r2], xm5
-    movu            [r2 + r3], xm3
+    psubw           m5, m3                          ; m5 = word: row 0, row 1
+    psubw           m2, m3                          ; m2 = word: row 2, row 3
+    psubw           m1, m3                          ; m1 = word: row 4, row 5
+    psubw           m4, m3                          ; m4 = word: row 6, row 7
+    vextracti128    xm6, m5, 1
     vextracti128    xm3, m2, 1
+    vextracti128    xm0, m1, 1
+    movu            [r2], xm5
+    movu            [r2 + r3], xm6
     movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
+    movu            [r2 + r4], xm3
     lea             r2, [r2 + r3 * 4]
-    vextracti128    xm5, m1, 1
-    vextracti128    xm3, m4, 1
     movu            [r2], xm1
-    movu            [r2 + r3], xm5
+    movu            [r2 + r3], xm0
     movu            [r2 + r3 * 2], xm4
-    movu            [r2 + r6], xm3
-%endif
-    movq            xm3, [r0 + r4]                  ; m3 = row 11
-    punpcklbw       xm6, xm3
-    lea             r0, [r0 + r1 * 4]
-    movq            xm5, [r0]                       ; m5 = row 12
-    punpcklbw       xm3, xm5
-    vinserti128     m6, m6, xm3, 1
-    pmaddubsw       m3, m6, [r5 + 1 * mmsize]
-    paddw           m0, m3
-    pmaddubsw       m6, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 13
-    punpcklbw       xm5, xm3
-    movq            xm2, [r0 + r1 * 2]              ; m2 = row 14
-    punpcklbw       xm3, xm2
-    vinserti128     m5, m5, xm3, 1
-    pmaddubsw       m3, m5, [r5 + 1 * mmsize]
-    paddw           m6, m3
-    pmaddubsw       m5, [r5]
-    movq            xm3, [r0 + r4]                  ; m3 = row 15
-    punpcklbw       xm2, xm3
-    lea             r0, [r0 + r1 * 4]
-    movq            xm1, [r0]                       ; m1 = row 16
-    punpcklbw       xm3, xm1
-    vinserti128     m2, m2, xm3, 1
-    pmaddubsw       m3, m2, [r5 + 1 * mmsize]
-    paddw           m5, m3
-    pmaddubsw       m2, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 17
-    punpcklbw       xm1, xm3
-    movq            xm4, [r0 + r1 * 2]              ; m4 = row 18
-    punpcklbw       xm3, xm4
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m1, [r5 + 1 * mmsize]
-    paddw           m2, m1
-    lea             r2, [r2 + r3 * 4]
-%ifidn %1,pp
-    pmulhrsw        m0, m7                          ; m0 = word: row 8, row 9
-    pmulhrsw        m6, m7                          ; m6 = word: row 10, row 11
-    pmulhrsw        m5, m7                          ; m5 = word: row 12, row 13
-    pmulhrsw        m2, m7                          ; m2 = word: row 14, row 15
-    packuswb        m0, m6
-    packuswb        m5, m2
-    vextracti128    xm6, m0, 1
-    vextracti128    xm2, m5, 1
-    movq            [r2], xm0
-    movq            [r2 + r3], xm6
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r6], xm6
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm5
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm5
-    movhps          [r2 + r6], xm2
-%else
-    psubw           m0, m7                          ; m0 = word: row 8, row 9
-    psubw           m6, m7                          ; m6 = word: row 10, row 11
-    psubw           m5, m7                          ; m5 = word: row 12, row 13
-    psubw           m2, m7                          ; m2 = word: row 14, row 15
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m6, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r6], xm3
-    lea             r2, [r2 + r3 * 4]
-    vextracti128    xm1, m5, 1
-    vextracti128    xm3, m2, 1
-    movu            [r2], xm5
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-%endif
-%endmacro
-
-%macro FILTER_VER_CHROMA_AVX2_8x16 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x16, 4, 7, 8
-    mov             r4d, r4m
-    shl             r4d, 6
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,pp
-    mova            m7, [pw_512]
-%else
-    add             r3d, r3d
-    mova            m7, [pw_2000]
+    vextracti128    xm4, m4, 1
+    movu            [r2 + r4], xm4
 %endif
-    lea             r6, [r3 * 3]
-    PROCESS_CHROMA_AVX2_W8_16R %1
     RET
 %endmacro
 
-    FILTER_VER_CHROMA_AVX2_8x16 pp
-    FILTER_VER_CHROMA_AVX2_8x16 ps
-
-%macro FILTER_VER_CHROMA_AVX2_8x12 1
+%macro FILTER_VER_LUMA_AVX2_8x4 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x12, 4, 7, 8
+cglobal interp_8tap_vert_%1_8x4, 4, 6, 7
     mov             r4d, r4m
-    shl             r4d, 6
+    shl             r4d, 7
 
 %ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
+    lea             r5, [tab_LumaCoeffVer_32]
     add             r5, r4
 %else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
 
     lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1, pp
-    mova            m7, [pw_512]
+    sub             r0, r4
+    PROCESS_LUMA_AVX2_W8_4R
+%ifidn %1,pp
+    mova            m3, [pw_512]
 %else
     add             r3d, r3d
-    mova            m7, [pw_2000]
+    vbroadcasti128  m3, [pw_2000]
 %endif
-    lea             r6, [r3 * 3]
-    movq            xm1, [r0]                       ; m1 = row 0
-    movq            xm2, [r0 + r1]                  ; m2 = row 1
-    punpcklbw       xm1, xm2
-    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
-    punpcklbw       xm2, xm3
-    vinserti128     m5, m1, xm2, 1
-    pmaddubsw       m5, [r5]
-    movq            xm4, [r0 + r4]                  ; m4 = row 3
-    punpcklbw       xm3, xm4
-    lea             r0, [r0 + r1 * 4]
-    movq            xm1, [r0]                       ; m1 = row 4
-    punpcklbw       xm4, xm1
-    vinserti128     m2, m3, xm4, 1
-    pmaddubsw       m0, m2, [r5 + 1 * mmsize]
-    paddw           m5, m0
-    pmaddubsw       m2, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 5
-    punpcklbw       xm1, xm3
-    movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
-    punpcklbw       xm3, xm4
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m0, m1, [r5 + 1 * mmsize]
-    paddw           m2, m0
-    pmaddubsw       m1, [r5]
-    movq            xm3, [r0 + r4]                  ; m3 = row 7
-    punpcklbw       xm4, xm3
-    lea             r0, [r0 + r1 * 4]
-    movq            xm0, [r0]                       ; m0 = row 8
-    punpcklbw       xm3, xm0
-    vinserti128     m4, m4, xm3, 1
-    pmaddubsw       m3, m4, [r5 + 1 * mmsize]
-    paddw           m1, m3
-    pmaddubsw       m4, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 9
-    punpcklbw       xm0, xm3
-    movq            xm6, [r0 + r1 * 2]              ; m6 = row 10
-    punpcklbw       xm3, xm6
-    vinserti128     m0, m0, xm3, 1
-    pmaddubsw       m3, m0, [r5 + 1 * mmsize]
-    paddw           m4, m3
-    pmaddubsw       m0, [r5]
-%ifidn %1, pp
-    pmulhrsw        m5, m7                          ; m5 = word: row 0, row 1
-    pmulhrsw        m2, m7                          ; m2 = word: row 2, row 3
-    pmulhrsw        m1, m7                          ; m1 = word: row 4, row 5
-    pmulhrsw        m4, m7                          ; m4 = word: row 6, row 7
+    lea             r4, [r3 * 3]
+%ifidn %1,pp
+    pmulhrsw        m5, m3                          ; m5 = word: row 0, row 1
+    pmulhrsw        m2, m3                          ; m2 = word: row 2, row 3
     packuswb        m5, m2
-    packuswb        m1, m4
     vextracti128    xm2, m5, 1
-    vextracti128    xm4, m1, 1
     movq            [r2], xm5
     movq            [r2 + r3], xm2
     movhps          [r2 + r3 * 2], xm5
-    movhps          [r2 + r6], xm2
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm1
-    movq            [r2 + r3], xm4
-    movhps          [r2 + r3 * 2], xm1
-    movhps          [r2 + r6], xm4
+    movhps          [r2 + r4], xm2
 %else
-    psubw           m5, m7                          ; m5 = word: row 0, row 1
-    psubw           m2, m7                          ; m2 = word: row 2, row 3
-    psubw           m1, m7                          ; m1 = word: row 4, row 5
-    psubw           m4, m7                          ; m4 = word: row 6, row 7
-    vextracti128    xm3, m5, 1
+    psubw           m5, m3                          ; m5 = word: row 0, row 1
+    psubw           m2, m3                          ; m2 = word: row 2, row 3
     movu            [r2], xm5
-    movu            [r2 + r3], xm3
-    vextracti128    xm3, m2, 1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-    lea             r2, [r2 + r3 * 4]
-    vextracti128    xm5, m1, 1
-    vextracti128    xm3, m4, 1
-    movu            [r2], xm1
+    vextracti128    xm5, m5, 1
     movu            [r2 + r3], xm5
-    movu            [r2 + r3 * 2], xm4
-    movu            [r2 + r6], xm3
-%endif
-    movq            xm3, [r0 + r4]                  ; m3 = row 11
-    punpcklbw       xm6, xm3
-    lea             r0, [r0 + r1 * 4]
-    movq            xm5, [r0]                       ; m5 = row 12
-    punpcklbw       xm3, xm5
-    vinserti128     m6, m6, xm3, 1
-    pmaddubsw       m3, m6, [r5 + 1 * mmsize]
-    paddw           m0, m3
-    pmaddubsw       m6, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 13
-    punpcklbw       xm5, xm3
-    movq            xm2, [r0 + r1 * 2]              ; m2 = row 14
-    punpcklbw       xm3, xm2
-    vinserti128     m5, m5, xm3, 1
-    pmaddubsw       m3, m5, [r5 + 1 * mmsize]
-    paddw           m6, m3
-    lea             r2, [r2 + r3 * 4]
-%ifidn %1, pp
-    pmulhrsw        m0, m7                          ; m0 = word: row 8, row 9
-    pmulhrsw        m6, m7                          ; m6 = word: row 10, row 11
-    packuswb        m0, m6
-    vextracti128    xm6, m0, 1
-    movq            [r2], xm0
-    movq            [r2 + r3], xm6
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r6], xm6
-%else
-    psubw           m0, m7                          ; m0 = word: row 8, row 9
-    psubw           m6, m7                          ; m6 = word: row 10, row 11
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m6, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r6], xm3
+    movu            [r2 + r3 * 2], xm2
+    vextracti128    xm2, m2, 1
+    movu            [r2 + r4], xm2
 %endif
     RET
 %endmacro
 
-    FILTER_VER_CHROMA_AVX2_8x12 pp
-    FILTER_VER_CHROMA_AVX2_8x12 ps
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_pp_8x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_8xN 8, 4, pp
+    FILTER_VER_LUMA_AVX2_8x4 pp
 
-%macro FILTER_VER_CHROMA_AVX2_8xN 2
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x%2, 4, 7, 8
-    mov             r4d, r4m
-    shl             r4d, 6
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_pp_8x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_8xN 8, 8, pp
+    FILTER_VER_LUMA_AVX2_8x8 pp
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_pp_8x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_8xN 8, 16, pp
+    FILTER_VER_LUMA_AVX2_8xN 8, 16, pp
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_pp_8x32(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_8xN 8, 32, pp
+    FILTER_VER_LUMA_AVX2_8xN 8, 32, pp
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_ps_8x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_8xN 8, 4, ps
+    FILTER_VER_LUMA_AVX2_8x4 ps
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_ps_8x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_8xN 8, 8, ps
+    FILTER_VER_LUMA_AVX2_8x8 ps
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_ps_8x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_8xN 8, 16, ps
+    FILTER_VER_LUMA_AVX2_8xN 8, 16, ps
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_ps_8x32(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_8xN 8, 32, ps
+    FILTER_VER_LUMA_AVX2_8xN 8, 32, ps
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_%3_12x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_LUMA_12xN 3
+INIT_XMM sse4
+cglobal interp_8tap_vert_%3_%1x%2, 5, 7, 8
+    lea       r5, [3 * r1]
+    sub       r0, r5
+    shl       r4d, 6
+%ifidn %3,ps
+    add       r3d, r3d
+%endif
 
 %ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
+    lea       r5, [tab_LumaCoeffVer]
+    lea       r6, [r5 + r4]
 %else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+    lea       r6, [tab_LumaCoeffVer + r4]
 %endif
 
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,pp
-    mova            m7, [pw_512]
+ %ifidn %3,pp
+    mova      m3, [pw_512]
 %else
-    add             r3d, r3d
-    mova            m7, [pw_2000]
+    mova      m3, [pw_2000]
 %endif
-    lea             r6, [r3 * 3]
-%rep %2 / 16
-    PROCESS_CHROMA_AVX2_W8_16R %1
-    lea             r2, [r2 + r3 * 4]
-%endrep
-    RET
-%endmacro
 
-    FILTER_VER_CHROMA_AVX2_8xN pp, 32
-    FILTER_VER_CHROMA_AVX2_8xN ps, 32
-    FILTER_VER_CHROMA_AVX2_8xN pp, 64
-    FILTER_VER_CHROMA_AVX2_8xN ps, 64
+    mov       r4d, %2/4
 
-%macro PROCESS_CHROMA_AVX2_W8_4R 0
-    movq            xm1, [r0]                       ; m1 = row 0
-    movq            xm2, [r0 + r1]                  ; m2 = row 1
-    punpcklbw       xm1, xm2                        ; m1 = [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
-    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
-    punpcklbw       xm2, xm3                        ; m2 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10]
-    vinserti128     m0, m1, xm2, 1                  ; m0 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10] - [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
-    pmaddubsw       m0, [r5]
-    movq            xm4, [r0 + r4]                  ; m4 = row 3
-    punpcklbw       xm3, xm4                        ; m3 = [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
-    lea             r0, [r0 + r1 * 4]
-    movq            xm1, [r0]                       ; m1 = row 4
-    punpcklbw       xm4, xm1                        ; m4 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30]
-    vinserti128     m2, m3, xm4, 1                  ; m2 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30] - [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
-    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
-    paddw           m0, m4
-    pmaddubsw       m2, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 5
-    punpcklbw       xm1, xm3                        ; m1 = [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
-    movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
-    punpcklbw       xm3, xm4                        ; m3 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50]
-    vinserti128     m1, m1, xm3, 1                  ; m1 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50] - [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
-    pmaddubsw       m1, [r5 + 1 * mmsize]
-    paddw           m2, m1
-%endmacro
+.loopH:
+    PROCESS_LUMA_W8_4R
 
-%macro FILTER_VER_CHROMA_AVX2_8x4 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x4, 4, 6, 5
-    mov             r4d, r4m
-    shl             r4d, 6
+%ifidn %3,pp
+    pmulhrsw  m7, m3
+    pmulhrsw  m6, m3
+    pmulhrsw  m5, m3
+    pmulhrsw  m4, m3
 
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
+    packuswb  m7, m6
+    packuswb  m5, m4
+
+    movlps    [r2], m7
+    movhps    [r2 + r3], m7
+    lea       r5, [r2 + 2 * r3]
+    movlps    [r5], m5
+    movhps    [r5 + r3], m5
 %else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+    psubw     m7, m3
+    psubw     m6, m3
+    psubw     m5, m3
+    psubw     m4, m3
+
+    movu      [r2], m7
+    movu      [r2 + r3], m6
+    lea       r5, [r2 + 2 * r3]
+    movu      [r5], m5
+    movu      [r5 + r3], m4
 %endif
 
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-    PROCESS_CHROMA_AVX2_W8_4R
-%ifidn %1,pp
-    lea             r4, [r3 * 3]
-    mova            m3, [pw_512]
-    pmulhrsw        m0, m3                          ; m0 = word: row 0, row 1
-    pmulhrsw        m2, m3                          ; m2 = word: row 2, row 3
-    packuswb        m0, m2
-    vextracti128    xm2, m0, 1
-    movq            [r2], xm0
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r4], xm2
+    lea       r5, [8 * r1 - 8]
+    sub       r0, r5
+%ifidn %3,pp
+    add       r2, 8
 %else
-    add             r3d, r3d
-    vbroadcasti128  m3, [pw_2000]
-    lea             r4, [r3 * 3]
-    psubw           m0, m3                          ; m0 = word: row 0, row 1
-    psubw           m2, m3                          ; m2 = word: row 2, row 3
-    vextracti128    xm1, m0, 1
-    vextracti128    xm4, m2, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r4], xm4
+    add       r2, 16
 %endif
-    RET
-%endmacro
 
-    FILTER_VER_CHROMA_AVX2_8x4 pp
-    FILTER_VER_CHROMA_AVX2_8x4 ps
+    PROCESS_LUMA_W4_4R
 
-%macro FILTER_VER_CHROMA_AVX2_8x2 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x2, 4, 6, 4
-    mov             r4d, r4m
-    shl             r4d, 6
+%ifidn %3,pp
+    pmulhrsw  m4, m3
+    pmulhrsw  m5, m3
 
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
+    packuswb  m4, m5
+
+    movd      [r2], m4
+    pextrd    [r2 + r3], m4, 1
+    lea       r5, [r2 + 2 * r3]
+    pextrd    [r5], m4, 2
+    pextrd    [r5 + r3], m4, 3
 %else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
+    psubw     m4, m3
+    psubw     m5, m3
 
-    lea             r4, [r1 * 3]
-    sub             r0, r1
+    movlps    [r2], m4
+    movhps    [r2 + r3], m4
+    lea       r5, [r2 + 2 * r3]
+    movlps    [r5], m5
+    movhps    [r5 + r3], m5
+%endif
 
-    movq            xm1, [r0]                       ; m1 = row 0
-    movq            xm2, [r0 + r1]                  ; m2 = row 1
-    punpcklbw       xm1, xm2                        ; m1 = [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
-    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
-    punpcklbw       xm2, xm3                        ; m2 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10]
-    vinserti128     m1, m1, xm2, 1                  ; m1 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10] - [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
-    pmaddubsw       m1, [r5]
-    movq            xm2, [r0 + r4]                  ; m2 = row 3
-    punpcklbw       xm3, xm2                        ; m3 = [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
-    movq            xm0, [r0 + r1 * 4]              ; m0 = row 4
-    punpcklbw       xm2, xm0                        ; m2 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30]
-    vinserti128     m3, m3, xm2, 1                  ; m3 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30] - [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
-    pmaddubsw       m3, [r5 + 1 * mmsize]
-    paddw           m1, m3
-%ifidn %1,pp
-    pmulhrsw        m1, [pw_512]                    ; m1 = word: row 0, row 1
-    packuswb        m1, m1
-    vextracti128    xm0, m1, 1
-    movq            [r2], xm1
-    movq            [r2 + r3], xm0
+    lea       r5, [4 * r1 + 8]
+    sub       r0, r5
+%ifidn %3,pp
+    lea       r2, [r2 + 4 * r3 - 8]
 %else
-    add             r3d, r3d
-    psubw           m1, [pw_2000]                   ; m1 = word: row 0, row 1
-    vextracti128    xm0, m1, 1
-    movu            [r2], xm1
-    movu            [r2 + r3], xm0
+    lea       r2, [r2 + 4 * r3 - 16]
 %endif
+
+    dec       r4d
+    jnz       .loopH
+
     RET
 %endmacro
 
-    FILTER_VER_CHROMA_AVX2_8x2 pp
-    FILTER_VER_CHROMA_AVX2_8x2 ps
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_pp_12x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_12xN 12, 16, pp
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_ps_12x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_12xN 12, 16, ps
 
-%macro FILTER_VER_CHROMA_AVX2_6x8 1
+%macro FILTER_VER_LUMA_AVX2_12x16 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_6x8, 4, 6, 7
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_12x16, 4, 7, 15
     mov             r4d, r4m
-    shl             r4d, 6
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-    PROCESS_CHROMA_AVX2_W8_8R
-%ifidn %1,pp
-    lea             r4, [r3 * 3]
-    mova            m3, [pw_512]
-    pmulhrsw        m5, m3                          ; m5 = word: row 0, row 1
-    pmulhrsw        m2, m3                          ; m2 = word: row 2, row 3
-    pmulhrsw        m1, m3                          ; m1 = word: row 4, row 5
-    pmulhrsw        m4, m3                          ; m4 = word: row 6, row 7
-    packuswb        m5, m2
-    packuswb        m1, m4
-    vextracti128    xm2, m5, 1
-    vextracti128    xm4, m1, 1
-    movd            [r2], xm5
-    pextrw          [r2 + 4], xm5, 2
-    movd            [r2 + r3], xm2
-    pextrw          [r2 + r3 + 4], xm2, 2
-    pextrd          [r2 + r3 * 2], xm5, 2
-    pextrw          [r2 + r3 * 2 + 4], xm5, 6
-    pextrd          [r2 + r4], xm2, 2
-    pextrw          [r2 + r4 + 4], xm2, 6
-    lea             r2, [r2 + r3 * 4]
-    movd            [r2], xm1
-    pextrw          [r2 + 4], xm1, 2
-    movd            [r2 + r3], xm4
-    pextrw          [r2 + r3 + 4], xm4, 2
-    pextrd          [r2 + r3 * 2], xm1, 2
-    pextrw          [r2 + r3 * 2 + 4], xm1, 6
-    pextrd          [r2 + r4], xm4, 2
-    pextrw          [r2 + r4 + 4], xm4, 6
-%else
-    add             r3d, r3d
-    vbroadcasti128  m3, [pw_2000]
-    lea             r4, [r3 * 3]
-    psubw           m5, m3                          ; m5 = word: row 0, row 1
-    psubw           m2, m3                          ; m2 = word: row 2, row 3
-    psubw           m1, m3                          ; m1 = word: row 4, row 5
-    psubw           m4, m3                          ; m4 = word: row 6, row 7
-    vextracti128    xm6, m5, 1
-    vextracti128    xm3, m2, 1
-    vextracti128    xm0, m1, 1
-    movq            [r2], xm5
-    pextrd          [r2 + 8], xm5, 2
-    movq            [r2 + r3], xm6
-    pextrd          [r2 + r3 + 8], xm6, 2
-    movq            [r2 + r3 * 2], xm2
-    pextrd          [r2 + r3 * 2 + 8], xm2, 2
-    movq            [r2 + r4], xm3
-    pextrd          [r2 + r4 + 8], xm3, 2
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm1
-    pextrd          [r2 + 8], xm1, 2
-    movq            [r2 + r3], xm0
-    pextrd          [r2 + r3 + 8], xm0, 2
-    movq            [r2 + r3 * 2], xm4
-    pextrd          [r2 + r3 * 2 + 8], xm4, 2
-    vextracti128    xm4, m4, 1
-    movq            [r2 + r4], xm4
-    pextrd          [r2 + r4 + 8], xm4, 2
-%endif
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_AVX2_6x8 pp
-    FILTER_VER_CHROMA_AVX2_6x8 ps
-
-;-----------------------------------------------------------------------------
-;void interp_4tap_vert_pp_6x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro FILTER_V4_W6_H4 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_pp_6x%2, 4, 6, 8
-
-    mov         r4d,       r4m
-    sub         r0,        r1
-
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeff]
-    movd        m5,        [r5 + r4 * 4]
-%else
-    movd        m5,        [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    pshufb      m6,        m5,       [tab_Vm]
-    pshufb      m5,        [tab_Vm + 16]
-    mova        m4,        [pw_512]
-
-    mov         r4d,       %2
-    lea         r5,        [3 * r1]
-
-.loop:
-    movq        m0,        [r0]
-    movq        m1,        [r0 + r1]
-    movq        m2,        [r0 + 2 * r1]
-    movq        m3,        [r0 + r5]
-
-    punpcklbw   m0,        m1
-    punpcklbw   m1,        m2
-    punpcklbw   m2,        m3
-
-    pmaddubsw   m0,        m6
-    pmaddubsw   m7,        m2, m5
-
-    paddw       m0,        m7
-
-    pmulhrsw    m0,        m4
-    packuswb    m0,        m0
-    movd        [r2],      m0
-    pextrw      [r2 + 4],  m0,    2
-
-    lea         r0,        [r0 + 4 * r1]
-
-    movq        m0,        [r0]
-    punpcklbw   m3,        m0
-
-    pmaddubsw   m1,        m6
-    pmaddubsw   m7,        m3, m5
-
-    paddw       m1,        m7
-
-    pmulhrsw    m1,        m4
-    packuswb    m1,        m1
-    movd        [r2 + r3],      m1
-    pextrw      [r2 + r3 + 4],  m1,    2
-
-    movq        m1,        [r0 + r1]
-    punpcklbw   m7,        m0,        m1
-
-    pmaddubsw   m2,        m6
-    pmaddubsw   m7,        m5
-
-    paddw       m2,        m7
-
-    pmulhrsw    m2,        m4
-    packuswb    m2,        m2
-    lea         r2,        [r2 + 2 * r3]
-    movd        [r2],      m2
-    pextrw      [r2 + 4],  m2,    2
-
-    movq        m2,        [r0 + 2 * r1]
-    punpcklbw   m1,        m2
-
-    pmaddubsw   m3,        m6
-    pmaddubsw   m1,        m5
-
-    paddw       m3,        m1
-
-    pmulhrsw    m3,        m4
-    packuswb    m3,        m3
-
-    movd        [r2 + r3],        m3
-    pextrw      [r2 + r3 + 4],    m3,    2
-
-    lea         r2,        [r2 + 2 * r3]
-
-    sub         r4,         4
-    jnz        .loop
-    RET
-%endmacro
-
-    FILTER_V4_W6_H4 6, 8
-
-    FILTER_V4_W6_H4 6, 16
-
-;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_12x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro FILTER_V4_W12_H2 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_pp_12x%2, 4, 6, 8
-
-    mov         r4d,       r4m
-    sub         r0,        r1
-
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeff]
-    movd        m0,        [r5 + r4 * 4]
-%else
-    movd        m0,        [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    pshufb      m1,        m0,       [tab_Vm]
-    pshufb      m0,        [tab_Vm + 16]
-
-    mov         r4d,       %2
-
-.loop:
-    movu        m2,        [r0]
-    movu        m3,        [r0 + r1]
-
-    punpcklbw   m4,        m2,        m3
-    punpckhbw   m2,        m3
-
-    pmaddubsw   m4,        m1
-    pmaddubsw   m2,        m1
-
-    lea         r0,        [r0 + 2 * r1]
-    movu        m5,        [r0]
-    movu        m7,        [r0 + r1]
-
-    punpcklbw   m6,        m5,        m7
-    pmaddubsw   m6,        m0
-    paddw       m4,        m6
-
-    punpckhbw   m6,        m5,        m7
-    pmaddubsw   m6,        m0
-    paddw       m2,        m6
-
-    mova        m6,        [pw_512]
-
-    pmulhrsw    m4,        m6
-    pmulhrsw    m2,        m6
-
-    packuswb    m4,        m2
-
-    movh         [r2],     m4
-    pextrd       [r2 + 8], m4,  2
-
-    punpcklbw   m4,        m3,        m5
-    punpckhbw   m3,        m5
-
-    pmaddubsw   m4,        m1
-    pmaddubsw   m3,        m1
-
-    movu        m5,        [r0 + 2 * r1]
-
-    punpcklbw   m2,        m7,        m5
-    punpckhbw   m7,        m5
-
-    pmaddubsw   m2,        m0
-    pmaddubsw   m7,        m0
-
-    paddw       m4,        m2
-    paddw       m3,        m7
-
-    pmulhrsw    m4,        m6
-    pmulhrsw    m3,        m6
-
-    packuswb    m4,        m3
-
-    movh        [r2 + r3],      m4
-    pextrd      [r2 + r3 + 8],  m4,  2
-
-    lea         r2,        [r2 + 2 * r3]
-
-    sub         r4,        2
-    jnz        .loop
-    RET
-%endmacro
-
-    FILTER_V4_W12_H2 12, 16
-
-    FILTER_V4_W12_H2 12, 32
-
-;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_16x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro FILTER_V4_W16_H2 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_pp_16x%2, 4, 6, 8
-
-    mov         r4d,       r4m
-    sub         r0,        r1
-
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeff]
-    movd        m0,        [r5 + r4 * 4]
-%else
-    movd        m0,        [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    pshufb      m1,        m0,       [tab_Vm]
-    pshufb      m0,        [tab_Vm + 16]
-
-    mov         r4d,       %2/2
-
-.loop:
-    movu        m2,        [r0]
-    movu        m3,        [r0 + r1]
-
-    punpcklbw   m4,        m2,        m3
-    punpckhbw   m2,        m3
-
-    pmaddubsw   m4,        m1
-    pmaddubsw   m2,        m1
-
-    lea         r0,        [r0 + 2 * r1]
-    movu        m5,        [r0]
-    movu        m6,        [r0 + r1]
-
-    punpckhbw   m7,        m5,        m6
-    pmaddubsw   m7,        m0
-    paddw       m2,        m7
-
-    punpcklbw   m7,        m5,        m6
-    pmaddubsw   m7,        m0
-    paddw       m4,        m7
-
-    mova        m7,        [pw_512]
-
-    pmulhrsw    m4,        m7
-    pmulhrsw    m2,        m7
-
-    packuswb    m4,        m2
-
-    movu        [r2],      m4
-
-    punpcklbw   m4,        m3,        m5
-    punpckhbw   m3,        m5
-
-    pmaddubsw   m4,        m1
-    pmaddubsw   m3,        m1
-
-    movu        m5,        [r0 + 2 * r1]
-
-    punpcklbw   m2,        m6,        m5
-    punpckhbw   m6,        m5
-
-    pmaddubsw   m2,        m0
-    pmaddubsw   m6,        m0
-
-    paddw       m4,        m2
-    paddw       m3,        m6
-
-    pmulhrsw    m4,        m7
-    pmulhrsw    m3,        m7
-
-    packuswb    m4,        m3
-
-    movu        [r2 + r3],      m4
-
-    lea         r2,        [r2 + 2 * r3]
-
-    dec         r4d
-    jnz        .loop
-    RET
-%endmacro
-
-    FILTER_V4_W16_H2 16,  4
-    FILTER_V4_W16_H2 16,  8
-    FILTER_V4_W16_H2 16, 12
-    FILTER_V4_W16_H2 16, 16
-    FILTER_V4_W16_H2 16, 32
-
-    FILTER_V4_W16_H2 16, 24
-    FILTER_V4_W16_H2 16, 64
-
-%macro FILTER_VER_CHROMA_AVX2_16x16 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_vert_%1_16x16, 4, 6, 15
-    mov             r4d, r4m
-    shl             r4d, 6
+    shl             r4d, 7
 
 %ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
+    lea             r5, [tab_LumaCoeffVer_32]
     add             r5, r4
 %else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
 
-    mova            m12, [r5]
-    mova            m13, [r5 + mmsize]
     lea             r4, [r1 * 3]
-    sub             r0, r1
+    sub             r0, r4
 %ifidn %1,pp
     mova            m14, [pw_512]
 %else
     add             r3d, r3d
     vbroadcasti128  m14, [pw_2000]
 %endif
-    lea             r5, [r3 * 3]
+    lea             r6, [r3 * 3]
 
     movu            xm0, [r0]                       ; m0 = row 0
     movu            xm1, [r0 + r1]                  ; m1 = row 1
     punpckhbw       xm2, xm0, xm1
     punpcklbw       xm0, xm1
     vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m0, m12
+    pmaddubsw       m0, [r5]
     movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
     punpckhbw       xm3, xm1, xm2
     punpcklbw       xm1, xm2
     vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m1, m12
+    pmaddubsw       m1, [r5]
     movu            xm3, [r0 + r4]                  ; m3 = row 3
     punpckhbw       xm4, xm2, xm3
     punpcklbw       xm2, xm3
     vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, m13
+    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
     paddw           m0, m4
-    pmaddubsw       m2, m12
+    pmaddubsw       m2, [r5]
     lea             r0, [r0 + r1 * 4]
     movu            xm4, [r0]                       ; m4 = row 4
     punpckhbw       xm5, xm3, xm4
     punpcklbw       xm3, xm4
     vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, m13
+    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
     paddw           m1, m5
-    pmaddubsw       m3, m12
+    pmaddubsw       m3, [r5]
     movu            xm5, [r0 + r1]                  ; m5 = row 5
     punpckhbw       xm6, xm4, xm5
     punpcklbw       xm4, xm5
     vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, m13
+    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
+    paddw           m0, m6
+    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
     paddw           m2, m6
-    pmaddubsw       m4, m12
+    pmaddubsw       m4, [r5]
     movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
     punpckhbw       xm7, xm5, xm6
     punpcklbw       xm5, xm6
     vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m7, m5, m13
+    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
+    paddw           m1, m7
+    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
     paddw           m3, m7
-    pmaddubsw       m5, m12
+    pmaddubsw       m5, [r5]
     movu            xm7, [r0 + r4]                  ; m7 = row 7
     punpckhbw       xm8, xm6, xm7
     punpcklbw       xm6, xm7
     vinserti128     m6, m6, xm8, 1
-    pmaddubsw       m8, m6, m13
+    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
+    paddw           m0, m8
+    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
+    paddw           m2, m8
+    pmaddubsw       m8, m6, [r5 + 1 * mmsize]
     paddw           m4, m8
-    pmaddubsw       m6, m12
+    pmaddubsw       m6, [r5]
     lea             r0, [r0 + r1 * 4]
     movu            xm8, [r0]                       ; m8 = row 8
     punpckhbw       xm9, xm7, xm8
     punpcklbw       xm7, xm8
     vinserti128     m7, m7, xm9, 1
-    pmaddubsw       m9, m7, m13
+    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
+    paddw           m1, m9
+    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
+    paddw           m3, m9
+    pmaddubsw       m9, m7, [r5 + 1 * mmsize]
     paddw           m5, m9
-    pmaddubsw       m7, m12
+    pmaddubsw       m7, [r5]
     movu            xm9, [r0 + r1]                  ; m9 = row 9
     punpckhbw       xm10, xm8, xm9
     punpcklbw       xm8, xm9
     vinserti128     m8, m8, xm10, 1
-    pmaddubsw       m10, m8, m13
+    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
+    paddw           m2, m10
+    pmaddubsw       m10, m8, [r5 + 2 * mmsize]
+    paddw           m4, m10
+    pmaddubsw       m10, m8, [r5 + 1 * mmsize]
     paddw           m6, m10
-    pmaddubsw       m8, m12
+    pmaddubsw       m8, [r5]
     movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
     punpckhbw       xm11, xm9, xm10
     punpcklbw       xm9, xm10
     vinserti128     m9, m9, xm11, 1
-    pmaddubsw       m11, m9, m13
+    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
+    paddw           m3, m11
+    pmaddubsw       m11, m9, [r5 + 2 * mmsize]
+    paddw           m5, m11
+    pmaddubsw       m11, m9, [r5 + 1 * mmsize]
     paddw           m7, m11
-    pmaddubsw       m9, m12
+    pmaddubsw       m9, [r5]
+    movu            xm11, [r0 + r4]                 ; m11 = row 11
+    punpckhbw       xm12, xm10, xm11
+    punpcklbw       xm10, xm11
+    vinserti128     m10, m10, xm12, 1
+    pmaddubsw       m12, m10, [r5 + 3 * mmsize]
+    paddw           m4, m12
+    pmaddubsw       m12, m10, [r5 + 2 * mmsize]
+    paddw           m6, m12
+    pmaddubsw       m12, m10, [r5 + 1 * mmsize]
+    paddw           m8, m12
+    pmaddubsw       m10, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm12, [r0]                      ; m12 = row 12
+    punpckhbw       xm13, xm11, xm12
+    punpcklbw       xm11, xm12
+    vinserti128     m11, m11, xm13, 1
+    pmaddubsw       m13, m11, [r5 + 3 * mmsize]
+    paddw           m5, m13
+    pmaddubsw       m13, m11, [r5 + 2 * mmsize]
+    paddw           m7, m13
+    pmaddubsw       m13, m11, [r5 + 1 * mmsize]
+    paddw           m9, m13
+    pmaddubsw       m11, [r5]
 
 %ifidn %1,pp
     pmulhrsw        m0, m14                         ; m0 = word: row 0
@@ -6270,29 +3833,28 @@ cglobal interp_4tap_vert_%1_16x16, 4, 6, 15
     pmulhrsw        m3, m14                         ; m3 = word: row 3
     pmulhrsw        m4, m14                         ; m4 = word: row 4
     pmulhrsw        m5, m14                         ; m5 = word: row 5
-    pmulhrsw        m6, m14                         ; m6 = word: row 6
-    pmulhrsw        m7, m14                         ; m7 = word: row 7
     packuswb        m0, m1
     packuswb        m2, m3
     packuswb        m4, m5
-    packuswb        m6, m7
     vpermq          m0, m0, 11011000b
     vpermq          m2, m2, 11011000b
     vpermq          m4, m4, 11011000b
-    vpermq          m6, m6, 11011000b
     vextracti128    xm1, m0, 1
     vextracti128    xm3, m2, 1
     vextracti128    xm5, m4, 1
-    vextracti128    xm7, m6, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r5], xm3
+    movq            [r2], xm0
+    pextrd          [r2 + 8], xm0, 2
+    movq            [r2 + r3], xm1
+    pextrd          [r2 + r3 + 8], xm1, 2
+    movq            [r2 + r3 * 2], xm2
+    pextrd          [r2 + r3 * 2 + 8], xm2, 2
+    movq            [r2 + r6], xm3
+    pextrd          [r2 + r6 + 8], xm3, 2
     lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm4
-    movu            [r2 + r3], xm5
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r5], xm7
+    movq            [r2], xm4
+    pextrd          [r2 + 8], xm4, 2
+    movq            [r2 + r3], xm5
+    pextrd          [r2 + r3 + 8], xm5, 2
 %else
     psubw           m0, m14                         ; m0 = word: row 0
     psubw           m1, m14                         ; m1 = word: row 1
@@ -6300,153 +3862,246 @@ cglobal interp_4tap_vert_%1_16x16, 4, 6, 15
     psubw           m3, m14                         ; m3 = word: row 3
     psubw           m4, m14                         ; m4 = word: row 4
     psubw           m5, m14                         ; m5 = word: row 5
-    psubw           m6, m14                         ; m6 = word: row 6
-    psubw           m7, m14                         ; m7 = word: row 7
-    movu            [r2], m0
-    movu            [r2 + r3], m1
-    movu            [r2 + r3 * 2], m2
-    movu            [r2 + r5], m3
+    movu            [r2], xm0
+    vextracti128    xm0, m0, 1
+    movq            [r2 + 16], xm0
+    movu            [r2 + r3], xm1
+    vextracti128    xm1, m1, 1
+    movq            [r2 + r3 + 16], xm1
+    movu            [r2 + r3 * 2], xm2
+    vextracti128    xm2, m2, 1
+    movq            [r2 + r3 * 2 + 16], xm2
+    movu            [r2 + r6], xm3
+    vextracti128    xm3, m3, 1
+    movq            [r2 + r6 + 16], xm3
     lea             r2, [r2 + r3 * 4]
-    movu            [r2], m4
-    movu            [r2 + r3], m5
-    movu            [r2 + r3 * 2], m6
-    movu            [r2 + r5], m7
+    movu            [r2], xm4
+    vextracti128    xm4, m4, 1
+    movq            [r2 + 16], xm4
+    movu            [r2 + r3], xm5
+    vextracti128    xm5, m5, 1
+    movq            [r2 + r3 + 16], xm5
 %endif
-    lea             r2, [r2 + r3 * 4]
 
-    movu            xm11, [r0 + r4]                 ; m11 = row 11
-    punpckhbw       xm6, xm10, xm11
-    punpcklbw       xm10, xm11
-    vinserti128     m10, m10, xm6, 1
-    pmaddubsw       m6, m10, m13
-    paddw           m8, m6
-    pmaddubsw       m10, m12
-    lea             r0, [r0 + r1 * 4]
-    movu            xm6, [r0]                       ; m6 = row 12
-    punpckhbw       xm7, xm11, xm6
-    punpcklbw       xm11, xm6
-    vinserti128     m11, m11, xm7, 1
-    pmaddubsw       m7, m11, m13
-    paddw           m9, m7
-    pmaddubsw       m11, m12
-
-    movu            xm7, [r0 + r1]                  ; m7 = row 13
-    punpckhbw       xm0, xm6, xm7
-    punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm0, 1
-    pmaddubsw       m0, m6, m13
+    movu            xm13, [r0 + r1]                 ; m13 = row 13
+    punpckhbw       xm0, xm12, xm13
+    punpcklbw       xm12, xm13
+    vinserti128     m12, m12, xm0, 1
+    pmaddubsw       m0, m12, [r5 + 3 * mmsize]
+    paddw           m6, m0
+    pmaddubsw       m0, m12, [r5 + 2 * mmsize]
+    paddw           m8, m0
+    pmaddubsw       m0, m12, [r5 + 1 * mmsize]
     paddw           m10, m0
-    pmaddubsw       m6, m12
+    pmaddubsw       m12, [r5]
     movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
-    punpckhbw       xm1, xm7, xm0
-    punpcklbw       xm7, xm0
-    vinserti128     m7, m7, xm1, 1
-    pmaddubsw       m1, m7, m13
+    punpckhbw       xm1, xm13, xm0
+    punpcklbw       xm13, xm0
+    vinserti128     m13, m13, xm1, 1
+    pmaddubsw       m1, m13, [r5 + 3 * mmsize]
+    paddw           m7, m1
+    pmaddubsw       m1, m13, [r5 + 2 * mmsize]
+    paddw           m9, m1
+    pmaddubsw       m1, m13, [r5 + 1 * mmsize]
     paddw           m11, m1
-    pmaddubsw       m7, m12
+    pmaddubsw       m13, [r5]
+
+%ifidn %1,pp
+    pmulhrsw        m6, m14                         ; m6 = word: row 6
+    pmulhrsw        m7, m14                         ; m7 = word: row 7
+    packuswb        m6, m7
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm7, m6, 1
+    movq            [r2 + r3 * 2], xm6
+    pextrd          [r2 + r3 * 2 + 8], xm6, 2
+    movq            [r2 + r6], xm7
+    pextrd          [r2 + r6 + 8], xm7, 2
+%else
+    psubw           m6, m14                         ; m6 = word: row 6
+    psubw           m7, m14                         ; m7 = word: row 7
+    movu            [r2 + r3 * 2], xm6
+    vextracti128    xm6, m6, 1
+    movq            [r2 + r3 * 2 + 16], xm6
+    movu            [r2 + r6], xm7
+    vextracti128    xm7, m7, 1
+    movq            [r2 + r6 + 16], xm7
+%endif
+    lea             r2, [r2 + r3 * 4]
+
     movu            xm1, [r0 + r4]                  ; m1 = row 15
     punpckhbw       xm2, xm0, xm1
     punpcklbw       xm0, xm1
     vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m2, m0, m13
-    paddw           m6, m2
-    pmaddubsw       m0, m12
+    pmaddubsw       m2, m0, [r5 + 3 * mmsize]
+    paddw           m8, m2
+    pmaddubsw       m2, m0, [r5 + 2 * mmsize]
+    paddw           m10, m2
+    pmaddubsw       m2, m0, [r5 + 1 * mmsize]
+    paddw           m12, m2
+    pmaddubsw       m0, [r5]
     lea             r0, [r0 + r1 * 4]
     movu            xm2, [r0]                       ; m2 = row 16
     punpckhbw       xm3, xm1, xm2
     punpcklbw       xm1, xm2
     vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m3, m1, m13
-    paddw           m7, m3
-    pmaddubsw       m1, m12
+    pmaddubsw       m3, m1, [r5 + 3 * mmsize]
+    paddw           m9, m3
+    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
+    paddw           m11, m3
+    pmaddubsw       m3, m1, [r5 + 1 * mmsize]
+    paddw           m13, m3
+    pmaddubsw       m1, [r5]
     movu            xm3, [r0 + r1]                  ; m3 = row 17
     punpckhbw       xm4, xm2, xm3
     punpcklbw       xm2, xm3
     vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m2, m13
+    pmaddubsw       m4, m2, [r5 + 3 * mmsize]
+    paddw           m10, m4
+    pmaddubsw       m4, m2, [r5 + 2 * mmsize]
+    paddw           m12, m4
+    pmaddubsw       m2, [r5 + 1 * mmsize]
     paddw           m0, m2
     movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
     punpckhbw       xm5, xm3, xm4
     punpcklbw       xm3, xm4
     vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m3, m13
+    pmaddubsw       m5, m3, [r5 + 3 * mmsize]
+    paddw           m11, m5
+    pmaddubsw       m5, m3, [r5 + 2 * mmsize]
+    paddw           m13, m5
+    pmaddubsw       m3, [r5 + 1 * mmsize]
     paddw           m1, m3
+    movu            xm5, [r0 + r4]                  ; m5 = row 19
+    punpckhbw       xm6, xm4, xm5
+    punpcklbw       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddubsw       m6, m4, [r5 + 3 * mmsize]
+    paddw           m12, m6
+    pmaddubsw       m4, [r5 + 2 * mmsize]
+    paddw           m0, m4
+    lea             r0, [r0 + r1 * 4]
+    movu            xm6, [r0]                       ; m6 = row 20
+    punpckhbw       xm7, xm5, xm6
+    punpcklbw       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddubsw       m7, m5, [r5 + 3 * mmsize]
+    paddw           m13, m7
+    pmaddubsw       m5, [r5 + 2 * mmsize]
+    paddw           m1, m5
+    movu            xm7, [r0 + r1]                  ; m7 = row 21
+    punpckhbw       xm2, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm2, 1
+    pmaddubsw       m6, [r5 + 3 * mmsize]
+    paddw           m0, m6
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 22
+    punpckhbw       xm3, xm7, xm2
+    punpcklbw       xm7, xm2
+    vinserti128     m7, m7, xm3, 1
+    pmaddubsw       m7, [r5 + 3 * mmsize]
+    paddw           m1, m7
 
 %ifidn %1,pp
     pmulhrsw        m8, m14                         ; m8 = word: row 8
     pmulhrsw        m9, m14                         ; m9 = word: row 9
     pmulhrsw        m10, m14                        ; m10 = word: row 10
     pmulhrsw        m11, m14                        ; m11 = word: row 11
-    pmulhrsw        m6, m14                         ; m6 = word: row 12
-    pmulhrsw        m7, m14                         ; m7 = word: row 13
+    pmulhrsw        m12, m14                        ; m12 = word: row 12
+    pmulhrsw        m13, m14                        ; m13 = word: row 13
     pmulhrsw        m0, m14                         ; m0 = word: row 14
     pmulhrsw        m1, m14                         ; m1 = word: row 15
     packuswb        m8, m9
     packuswb        m10, m11
-    packuswb        m6, m7
+    packuswb        m12, m13
     packuswb        m0, m1
     vpermq          m8, m8, 11011000b
     vpermq          m10, m10, 11011000b
-    vpermq          m6, m6, 11011000b
+    vpermq          m12, m12, 11011000b
     vpermq          m0, m0, 11011000b
     vextracti128    xm9, m8, 1
     vextracti128    xm11, m10, 1
-    vextracti128    xm7, m6, 1
+    vextracti128    xm13, m12, 1
     vextracti128    xm1, m0, 1
-    movu            [r2], xm8
-    movu            [r2 + r3], xm9
-    movu            [r2 + r3 * 2], xm10
-    movu            [r2 + r5], xm11
+    movq            [r2], xm8
+    pextrd          [r2 + 8], xm8, 2
+    movq            [r2 + r3], xm9
+    pextrd          [r2 + r3 + 8], xm9, 2
+    movq            [r2 + r3 * 2], xm10
+    pextrd          [r2 + r3 * 2 + 8], xm10, 2
+    movq            [r2 + r6], xm11
+    pextrd          [r2 + r6 + 8], xm11, 2
     lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm6
-    movu            [r2 + r3], xm7
-    movu            [r2 + r3 * 2], xm0
-    movu            [r2 + r5], xm1
+    movq            [r2], xm12
+    pextrd          [r2 + 8], xm12, 2
+    movq            [r2 + r3], xm13
+    pextrd          [r2 + r3 + 8], xm13, 2
+    movq            [r2 + r3 * 2], xm0
+    pextrd          [r2 + r3 * 2 + 8], xm0, 2
+    movq            [r2 + r6], xm1
+    pextrd          [r2 + r6 + 8], xm1, 2
 %else
     psubw           m8, m14                         ; m8 = word: row 8
     psubw           m9, m14                         ; m9 = word: row 9
     psubw           m10, m14                        ; m10 = word: row 10
     psubw           m11, m14                        ; m11 = word: row 11
-    psubw           m6, m14                         ; m6 = word: row 12
-    psubw           m7, m14                         ; m7 = word: row 13
+    psubw           m12, m14                        ; m12 = word: row 12
+    psubw           m13, m14                        ; m13 = word: row 13
     psubw           m0, m14                         ; m0 = word: row 14
     psubw           m1, m14                         ; m1 = word: row 15
-    movu            [r2], m8
-    movu            [r2 + r3], m9
-    movu            [r2 + r3 * 2], m10
-    movu            [r2 + r5], m11
+    movu            [r2], xm8
+    vextracti128    xm8, m8, 1
+    movq            [r2 + 16], xm8
+    movu            [r2 + r3], xm9
+    vextracti128    xm9, m9, 1
+    movq            [r2 + r3 + 16], xm9
+    movu            [r2 + r3 * 2], xm10
+    vextracti128    xm10, m10, 1
+    movq            [r2 + r3 * 2 + 16], xm10
+    movu            [r2 + r6], xm11
+    vextracti128    xm11, m11, 1
+    movq            [r2 + r6 + 16], xm11
     lea             r2, [r2 + r3 * 4]
-    movu            [r2], m6
-    movu            [r2 + r3], m7
-    movu            [r2 + r3 * 2], m0
-    movu            [r2 + r5], m1
+    movu            [r2], xm12
+    vextracti128    xm12, m12, 1
+    movq            [r2 + 16], xm12
+    movu            [r2 + r3], xm13
+    vextracti128    xm13, m13, 1
+    movq            [r2 + r3 + 16], xm13
+    movu            [r2 + r3 * 2], xm0
+    vextracti128    xm0, m0, 1
+    movq            [r2 + r3 * 2 + 16], xm0
+    movu            [r2 + r6], xm1
+    vextracti128    xm1, m1, 1
+    movq            [r2 + r6 + 16], xm1
 %endif
     RET
 %endif
 %endmacro
 
-    FILTER_VER_CHROMA_AVX2_16x16 pp
-    FILTER_VER_CHROMA_AVX2_16x16 ps
-%macro FILTER_VER_CHROMA_AVX2_16x8 1
+    FILTER_VER_LUMA_AVX2_12x16 pp
+    FILTER_VER_LUMA_AVX2_12x16 ps
+
+%macro FILTER_VER_LUMA_AVX2_16x16 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_16x8, 4, 7, 7
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_16x16, 4, 7, 15
     mov             r4d, r4m
-    shl             r4d, 6
+    shl             r4d, 7
 
 %ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
+    lea             r5, [tab_LumaCoeffVer_32]
     add             r5, r4
 %else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
 
     lea             r4, [r1 * 3]
-    sub             r0, r1
+    sub             r0, r4
 %ifidn %1,pp
-    mova            m6, [pw_512]
+    mova            m14, [pw_512]
 %else
     add             r3d, r3d
-    mova            m6, [pw_2000]
+    vbroadcasti128  m14, [pw_2000]
 %endif
     lea             r6, [r3 * 3]
 
@@ -6465,7 +4120,7 @@ cglobal interp_4tap_vert_%1_16x8, 4, 7, 7
     punpckhbw       xm4, xm2, xm3
     punpcklbw       xm2, xm3
     vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, [r5 + mmsize]
+    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
     paddw           m0, m4
     pmaddubsw       m2, [r5]
     lea             r0, [r0 + r1 * 4]
@@ -6473,735 +4128,440 @@ cglobal interp_4tap_vert_%1_16x8, 4, 7, 7
     punpckhbw       xm5, xm3, xm4
     punpcklbw       xm3, xm4
     vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, [r5 + mmsize]
+    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
     paddw           m1, m5
     pmaddubsw       m3, [r5]
-%ifidn %1,pp
-    pmulhrsw        m0, m6                          ; m0 = word: row 0
-    pmulhrsw        m1, m6                          ; m1 = word: row 1
-    packuswb        m0, m1
-    vpermq          m0, m0, 11011000b
-    vextracti128    xm1, m0, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-%else
-    psubw           m0, m6                          ; m0 = word: row 0
-    psubw           m1, m6                          ; m1 = word: row 1
-    movu            [r2], m0
-    movu            [r2 + r3], m1
-%endif
-
-    movu            xm0, [r0 + r1]                  ; m0 = row 5
-    punpckhbw       xm1, xm4, xm0
-    punpcklbw       xm4, xm0
-    vinserti128     m4, m4, xm1, 1
-    pmaddubsw       m1, m4, [r5 + mmsize]
-    paddw           m2, m1
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhbw       xm6, xm4, xm5
+    punpcklbw       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
+    paddw           m0, m6
+    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
+    paddw           m2, m6
     pmaddubsw       m4, [r5]
-    movu            xm1, [r0 + r1 * 2]              ; m1 = row 6
-    punpckhbw       xm5, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm5, 1
-    pmaddubsw       m5, m0, [r5 + mmsize]
-    paddw           m3, m5
-    pmaddubsw       m0, [r5]
-%ifidn %1,pp
-    pmulhrsw        m2, m6                          ; m2 = word: row 2
-    pmulhrsw        m3, m6                          ; m3 = word: row 3
-    packuswb        m2, m3
-    vpermq          m2, m2, 11011000b
-    vextracti128    xm3, m2, 1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-%else
-    psubw           m2, m6                          ; m2 = word: row 2
-    psubw           m3, m6                          ; m3 = word: row 3
-    movu            [r2 + r3 * 2], m2
-    movu            [r2 + r6], m3
-%endif
-
-    movu            xm2, [r0 + r4]                  ; m2 = row 7
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m3, m1, [r5 + mmsize]
-    paddw           m4, m3
-    pmaddubsw       m1, [r5]
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhbw       xm7, xm5, xm6
+    punpcklbw       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
+    paddw           m1, m7
+    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
+    paddw           m3, m7
+    pmaddubsw       m5, [r5]
+    movu            xm7, [r0 + r4]                  ; m7 = row 7
+    punpckhbw       xm8, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
+    paddw           m0, m8
+    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
+    paddw           m2, m8
+    pmaddubsw       m8, m6, [r5 + 1 * mmsize]
+    paddw           m4, m8
+    pmaddubsw       m6, [r5]
     lea             r0, [r0 + r1 * 4]
-    movu            xm3, [r0]                       ; m3 = row 8
-    punpckhbw       xm5, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm5, 1
-    pmaddubsw       m5, m2, [r5 + mmsize]
-    paddw           m0, m5
-    pmaddubsw       m2, [r5]
-    lea             r2, [r2 + r3 * 4]
+    movu            xm8, [r0]                       ; m8 = row 8
+    punpckhbw       xm9, xm7, xm8
+    punpcklbw       xm7, xm8
+    vinserti128     m7, m7, xm9, 1
+    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
+    paddw           m1, m9
+    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
+    paddw           m3, m9
+    pmaddubsw       m9, m7, [r5 + 1 * mmsize]
+    paddw           m5, m9
+    pmaddubsw       m7, [r5]
+    movu            xm9, [r0 + r1]                  ; m9 = row 9
+    punpckhbw       xm10, xm8, xm9
+    punpcklbw       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
+    paddw           m2, m10
+    pmaddubsw       m10, m8, [r5 + 2 * mmsize]
+    paddw           m4, m10
+    pmaddubsw       m10, m8, [r5 + 1 * mmsize]
+    paddw           m6, m10
+    pmaddubsw       m8, [r5]
+    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
+    punpckhbw       xm11, xm9, xm10
+    punpcklbw       xm9, xm10
+    vinserti128     m9, m9, xm11, 1
+    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
+    paddw           m3, m11
+    pmaddubsw       m11, m9, [r5 + 2 * mmsize]
+    paddw           m5, m11
+    pmaddubsw       m11, m9, [r5 + 1 * mmsize]
+    paddw           m7, m11
+    pmaddubsw       m9, [r5]
+    movu            xm11, [r0 + r4]                 ; m11 = row 11
+    punpckhbw       xm12, xm10, xm11
+    punpcklbw       xm10, xm11
+    vinserti128     m10, m10, xm12, 1
+    pmaddubsw       m12, m10, [r5 + 3 * mmsize]
+    paddw           m4, m12
+    pmaddubsw       m12, m10, [r5 + 2 * mmsize]
+    paddw           m6, m12
+    pmaddubsw       m12, m10, [r5 + 1 * mmsize]
+    paddw           m8, m12
+    pmaddubsw       m10, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm12, [r0]                      ; m12 = row 12
+    punpckhbw       xm13, xm11, xm12
+    punpcklbw       xm11, xm12
+    vinserti128     m11, m11, xm13, 1
+    pmaddubsw       m13, m11, [r5 + 3 * mmsize]
+    paddw           m5, m13
+    pmaddubsw       m13, m11, [r5 + 2 * mmsize]
+    paddw           m7, m13
+    pmaddubsw       m13, m11, [r5 + 1 * mmsize]
+    paddw           m9, m13
+    pmaddubsw       m11, [r5]
+
 %ifidn %1,pp
-    pmulhrsw        m4, m6                          ; m4 = word: row 4
-    pmulhrsw        m0, m6                          ; m0 = word: row 5
-    packuswb        m4, m0
+    pmulhrsw        m0, m14                         ; m0 = word: row 0
+    pmulhrsw        m1, m14                         ; m1 = word: row 1
+    pmulhrsw        m2, m14                         ; m2 = word: row 2
+    pmulhrsw        m3, m14                         ; m3 = word: row 3
+    pmulhrsw        m4, m14                         ; m4 = word: row 4
+    pmulhrsw        m5, m14                         ; m5 = word: row 5
+    packuswb        m0, m1
+    packuswb        m2, m3
+    packuswb        m4, m5
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
     vpermq          m4, m4, 11011000b
-    vextracti128    xm0, m4, 1
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m2, 1
+    vextracti128    xm5, m4, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
+    lea             r2, [r2 + r3 * 4]
     movu            [r2], xm4
-    movu            [r2 + r3], xm0
+    movu            [r2 + r3], xm5
 %else
-    psubw           m4, m6                          ; m4 = word: row 4
-    psubw           m0, m6                          ; m0 = word: row 5
+    psubw           m0, m14                         ; m0 = word: row 0
+    psubw           m1, m14                         ; m1 = word: row 1
+    psubw           m2, m14                         ; m2 = word: row 2
+    psubw           m3, m14                         ; m3 = word: row 3
+    psubw           m4, m14                         ; m4 = word: row 4
+    psubw           m5, m14                         ; m5 = word: row 5
+    movu            [r2], m0
+    movu            [r2 + r3], m1
+    movu            [r2 + r3 * 2], m2
+    movu            [r2 + r6], m3
+    lea             r2, [r2 + r3 * 4]
     movu            [r2], m4
-    movu            [r2 + r3], m0
+    movu            [r2 + r3], m5
+%endif
+
+    movu            xm13, [r0 + r1]                 ; m13 = row 13
+    punpckhbw       xm0, xm12, xm13
+    punpcklbw       xm12, xm13
+    vinserti128     m12, m12, xm0, 1
+    pmaddubsw       m0, m12, [r5 + 3 * mmsize]
+    paddw           m6, m0
+    pmaddubsw       m0, m12, [r5 + 2 * mmsize]
+    paddw           m8, m0
+    pmaddubsw       m0, m12, [r5 + 1 * mmsize]
+    paddw           m10, m0
+    pmaddubsw       m12, [r5]
+    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
+    punpckhbw       xm1, xm13, xm0
+    punpcklbw       xm13, xm0
+    vinserti128     m13, m13, xm1, 1
+    pmaddubsw       m1, m13, [r5 + 3 * mmsize]
+    paddw           m7, m1
+    pmaddubsw       m1, m13, [r5 + 2 * mmsize]
+    paddw           m9, m1
+    pmaddubsw       m1, m13, [r5 + 1 * mmsize]
+    paddw           m11, m1
+    pmaddubsw       m13, [r5]
+
+%ifidn %1,pp
+    pmulhrsw        m6, m14                         ; m6 = word: row 6
+    pmulhrsw        m7, m14                         ; m7 = word: row 7
+    packuswb        m6, m7
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm7, m6, 1
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r6], xm7
+%else
+    psubw           m6, m14                         ; m6 = word: row 6
+    psubw           m7, m14                         ; m7 = word: row 7
+    movu            [r2 + r3 * 2], m6
+    movu            [r2 + r6], m7
 %endif
+    lea             r2, [r2 + r3 * 4]
 
-    movu            xm5, [r0 + r1]                  ; m5 = row 9
-    punpckhbw       xm4, xm3, xm5
-    punpcklbw       xm3, xm5
-    vinserti128     m3, m3, xm4, 1
-    pmaddubsw       m3, [r5 + mmsize]
+    movu            xm1, [r0 + r4]                  ; m1 = row 15
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddubsw       m2, m0, [r5 + 3 * mmsize]
+    paddw           m8, m2
+    pmaddubsw       m2, m0, [r5 + 2 * mmsize]
+    paddw           m10, m2
+    pmaddubsw       m2, m0, [r5 + 1 * mmsize]
+    paddw           m12, m2
+    pmaddubsw       m0, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm2, [r0]                       ; m2 = row 16
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m3, m1, [r5 + 3 * mmsize]
+    paddw           m9, m3
+    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
+    paddw           m11, m3
+    pmaddubsw       m3, m1, [r5 + 1 * mmsize]
+    paddw           m13, m3
+    pmaddubsw       m1, [r5]
+    movu            xm3, [r0 + r1]                  ; m3 = row 17
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m4, m2, [r5 + 3 * mmsize]
+    paddw           m10, m4
+    pmaddubsw       m4, m2, [r5 + 2 * mmsize]
+    paddw           m12, m4
+    pmaddubsw       m2, [r5 + 1 * mmsize]
+    paddw           m0, m2
+    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddubsw       m5, m3, [r5 + 3 * mmsize]
+    paddw           m11, m5
+    pmaddubsw       m5, m3, [r5 + 2 * mmsize]
+    paddw           m13, m5
+    pmaddubsw       m3, [r5 + 1 * mmsize]
     paddw           m1, m3
-    movu            xm4, [r0 + r1 * 2]              ; m4 = row 10
-    punpckhbw       xm0, xm5, xm4
-    punpcklbw       xm5, xm4
-    vinserti128     m5, m5, xm0, 1
-    pmaddubsw       m5, [r5 + mmsize]
-    paddw           m2, m5
+    movu            xm5, [r0 + r4]                  ; m5 = row 19
+    punpckhbw       xm6, xm4, xm5
+    punpcklbw       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddubsw       m6, m4, [r5 + 3 * mmsize]
+    paddw           m12, m6
+    pmaddubsw       m4, [r5 + 2 * mmsize]
+    paddw           m0, m4
+    lea             r0, [r0 + r1 * 4]
+    movu            xm6, [r0]                       ; m6 = row 20
+    punpckhbw       xm7, xm5, xm6
+    punpcklbw       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddubsw       m7, m5, [r5 + 3 * mmsize]
+    paddw           m13, m7
+    pmaddubsw       m5, [r5 + 2 * mmsize]
+    paddw           m1, m5
+    movu            xm7, [r0 + r1]                  ; m7 = row 21
+    punpckhbw       xm2, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm2, 1
+    pmaddubsw       m6, [r5 + 3 * mmsize]
+    paddw           m0, m6
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 22
+    punpckhbw       xm3, xm7, xm2
+    punpcklbw       xm7, xm2
+    vinserti128     m7, m7, xm3, 1
+    pmaddubsw       m7, [r5 + 3 * mmsize]
+    paddw           m1, m7
+
 %ifidn %1,pp
-    pmulhrsw        m1, m6                          ; m1 = word: row 6
-    pmulhrsw        m2, m6                          ; m2 = word: row 7
-    packuswb        m1, m2
-    vpermq          m1, m1, 11011000b
-    vextracti128    xm2, m1, 1
-    movu            [r2 + r3 * 2], xm1
-    movu            [r2 + r6], xm2
+    pmulhrsw        m8, m14                         ; m8 = word: row 8
+    pmulhrsw        m9, m14                         ; m9 = word: row 9
+    pmulhrsw        m10, m14                        ; m10 = word: row 10
+    pmulhrsw        m11, m14                        ; m11 = word: row 11
+    pmulhrsw        m12, m14                        ; m12 = word: row 12
+    pmulhrsw        m13, m14                        ; m13 = word: row 13
+    pmulhrsw        m0, m14                         ; m0 = word: row 14
+    pmulhrsw        m1, m14                         ; m1 = word: row 15
+    packuswb        m8, m9
+    packuswb        m10, m11
+    packuswb        m12, m13
+    packuswb        m0, m1
+    vpermq          m8, m8, 11011000b
+    vpermq          m10, m10, 11011000b
+    vpermq          m12, m12, 11011000b
+    vpermq          m0, m0, 11011000b
+    vextracti128    xm9, m8, 1
+    vextracti128    xm11, m10, 1
+    vextracti128    xm13, m12, 1
+    vextracti128    xm1, m0, 1
+    movu            [r2], xm8
+    movu            [r2 + r3], xm9
+    movu            [r2 + r3 * 2], xm10
+    movu            [r2 + r6], xm11
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm12
+    movu            [r2 + r3], xm13
+    movu            [r2 + r3 * 2], xm0
+    movu            [r2 + r6], xm1
 %else
-    psubw           m1, m6                          ; m1 = word: row 6
-    psubw           m2, m6                          ; m2 = word: row 7
-    movu            [r2 + r3 * 2], m1
-    movu            [r2 + r6], m2
+    psubw           m8, m14                         ; m8 = word: row 8
+    psubw           m9, m14                         ; m9 = word: row 9
+    psubw           m10, m14                        ; m10 = word: row 10
+    psubw           m11, m14                        ; m11 = word: row 11
+    psubw           m12, m14                        ; m12 = word: row 12
+    psubw           m13, m14                        ; m13 = word: row 13
+    psubw           m0, m14                         ; m0 = word: row 14
+    psubw           m1, m14                         ; m1 = word: row 15
+    movu            [r2], m8
+    movu            [r2 + r3], m9
+    movu            [r2 + r3 * 2], m10
+    movu            [r2 + r6], m11
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], m12
+    movu            [r2 + r3], m13
+    movu            [r2 + r3 * 2], m0
+    movu            [r2 + r6], m1
 %endif
     RET
+%endif
 %endmacro
 
-    FILTER_VER_CHROMA_AVX2_16x8 pp
-    FILTER_VER_CHROMA_AVX2_16x8 ps
+    FILTER_VER_LUMA_AVX2_16x16 pp
+    FILTER_VER_LUMA_AVX2_16x16 ps
 
-%macro FILTER_VER_CHROMA_AVX2_16x12 1
+%macro FILTER_VER_LUMA_AVX2_16x12 1
 INIT_YMM avx2
 %if ARCH_X86_64 == 1
-cglobal interp_4tap_vert_%1_16x12, 4, 6, 10
+cglobal interp_8tap_vert_%1_16x12, 4, 7, 15
     mov             r4d, r4m
-    shl             r4d, 6
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
-
-    mova            m8, [r5]
-    mova            m9, [r5 + mmsize]
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,pp
-    mova            m7, [pw_512]
-%else
-    add             r3d, r3d
-    vbroadcasti128  m7, [pw_2000]
-%endif
-    lea             r5, [r3 * 3]
-
-    movu            xm0, [r0]
-    vinserti128     m0, m0, [r0 + r1 * 2], 1
-    movu            xm1, [r0 + r1]
-    vinserti128     m1, m1, [r0 + r4], 1
-
-    punpcklbw       m2, m0, m1
-    punpckhbw       m3, m0, m1
-    vperm2i128      m4, m2, m3, 0x20
-    vperm2i128      m2, m2, m3, 0x31
-    pmaddubsw       m4, m8
-    pmaddubsw       m3, m2, m9
-    paddw           m4, m3
-    pmaddubsw       m2, m8
-
-    vextracti128    xm0, m0, 1
-    lea             r0, [r0 + r1 * 4]
-    vinserti128     m0, m0, [r0], 1
-
-    punpcklbw       m5, m1, m0
-    punpckhbw       m3, m1, m0
-    vperm2i128      m6, m5, m3, 0x20
-    vperm2i128      m5, m5, m3, 0x31
-    pmaddubsw       m6, m8
-    pmaddubsw       m3, m5, m9
-    paddw           m6, m3
-    pmaddubsw       m5, m8
-%ifidn %1,pp
-    pmulhrsw        m4, m7                         ; m4 = word: row 0
-    pmulhrsw        m6, m7                         ; m6 = word: row 1
-    packuswb        m4, m6
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm6, m4, 1
-    movu            [r2], xm4
-    movu            [r2 + r3], xm6
-%else
-    psubw           m4, m7                         ; m4 = word: row 0
-    psubw           m6, m7                         ; m6 = word: row 1
-    movu            [r2], m4
-    movu            [r2 + r3], m6
-%endif
-
-    movu            xm4, [r0 + r1 * 2]
-    vinserti128     m4, m4, [r0 + r1], 1
-    vextracti128    xm1, m4, 1
-    vinserti128     m0, m0, xm1, 0
-
-    punpcklbw       m6, m0, m4
-    punpckhbw       m1, m0, m4
-    vperm2i128      m0, m6, m1, 0x20
-    vperm2i128      m6, m6, m1, 0x31
-    pmaddubsw       m1, m0, m9
-    paddw           m5, m1
-    pmaddubsw       m0, m8
-    pmaddubsw       m1, m6, m9
-    paddw           m2, m1
-    pmaddubsw       m6, m8
-
-%ifidn %1,pp
-    pmulhrsw        m2, m7                         ; m2 = word: row 2
-    pmulhrsw        m5, m7                         ; m5 = word: row 3
-    packuswb        m2, m5
-    vpermq          m2, m2, 11011000b
-    vextracti128    xm5, m2, 1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r5], xm5
-%else
-    psubw           m2, m7                         ; m2 = word: row 2
-    psubw           m5, m7                         ; m5 = word: row 3
-    movu            [r2 + r3 * 2], m2
-    movu            [r2 + r5], m5
-%endif
-    lea             r2, [r2 + r3 * 4]
-
-    movu            xm1, [r0 + r4]
-    lea             r0, [r0 + r1 * 4]
-    vinserti128     m1, m1, [r0], 1
-    vinserti128     m4, m4, xm1, 1
-
-    punpcklbw       m2, m4, m1
-    punpckhbw       m5, m4, m1
-    vperm2i128      m3, m2, m5, 0x20
-    vperm2i128      m2, m2, m5, 0x31
-    pmaddubsw       m5, m3, m9
-    paddw           m6, m5
-    pmaddubsw       m3, m8
-    pmaddubsw       m5, m2, m9
-    paddw           m0, m5
-    pmaddubsw       m2, m8
-
-%ifidn %1,pp
-    pmulhrsw        m6, m7                         ; m6 = word: row 4
-    pmulhrsw        m0, m7                         ; m0 = word: row 5
-    packuswb        m6, m0
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm0, m6, 1
-    movu            [r2], xm6
-    movu            [r2 + r3], xm0
-%else
-    psubw           m6, m7                         ; m6 = word: row 4
-    psubw           m0, m7                         ; m0 = word: row 5
-    movu            [r2], m6
-    movu            [r2 + r3], m0
-%endif
-
-    movu            xm6, [r0 + r1 * 2]
-    vinserti128     m6, m6, [r0 + r1], 1
-    vextracti128    xm0, m6, 1
-    vinserti128     m1, m1, xm0, 0
-
-    punpcklbw       m4, m1, m6
-    punpckhbw       m5, m1, m6
-    vperm2i128      m0, m4, m5, 0x20
-    vperm2i128      m5, m4, m5, 0x31
-    pmaddubsw       m4, m0, m9
-    paddw           m2, m4
-    pmaddubsw       m0, m8
-    pmaddubsw       m4, m5, m9
-    paddw           m3, m4
-    pmaddubsw       m5, m8
-
-%ifidn %1,pp
-    pmulhrsw        m3, m7                         ; m3 = word: row 6
-    pmulhrsw        m2, m7                         ; m2 = word: row 7
-    packuswb        m3, m2
-    vpermq          m3, m3, 11011000b
-    vextracti128    xm2, m3, 1
-    movu            [r2 + r3 * 2], xm3
-    movu            [r2 + r5], xm2
-%else
-    psubw           m3, m7                         ; m3 = word: row 6
-    psubw           m2, m7                         ; m2 = word: row 7
-    movu            [r2 + r3 * 2], m3
-    movu            [r2 + r5], m2
-%endif
-    lea             r2, [r2 + r3 * 4]
-
-    movu            xm3, [r0 + r4]
-    lea             r0, [r0 + r1 * 4]
-    vinserti128     m3, m3, [r0], 1
-    vinserti128     m6, m6, xm3, 1
-
-    punpcklbw       m2, m6, m3
-    punpckhbw       m1, m6, m3
-    vperm2i128      m4, m2, m1, 0x20
-    vperm2i128      m2, m2, m1, 0x31
-    pmaddubsw       m1, m4, m9
-    paddw           m5, m1
-    pmaddubsw       m4, m8
-    pmaddubsw       m1, m2, m9
-    paddw           m0, m1
-    pmaddubsw       m2, m8
-
-%ifidn %1,pp
-    pmulhrsw        m5, m7                         ; m5 = word: row 8
-    pmulhrsw        m0, m7                         ; m0 = word: row 9
-    packuswb        m5, m0
-    vpermq          m5, m5, 11011000b
-    vextracti128    xm0, m5, 1
-    movu            [r2], xm5
-    movu            [r2 + r3], xm0
-%else
-    psubw           m5, m7                         ; m5 = word: row 8
-    psubw           m0, m7                         ; m0 = word: row 9
-    movu            [r2], m5
-    movu            [r2 + r3], m0
-%endif
-
-    movu            xm5, [r0 + r1 * 2]
-    vinserti128     m5, m5, [r0 + r1], 1
-    vextracti128    xm0, m5, 1
-    vinserti128     m3, m3, xm0, 0
-
-    punpcklbw       m1, m3, m5
-    punpckhbw       m0, m3, m5
-    vperm2i128      m6, m1, m0, 0x20
-    vperm2i128      m0, m1, m0, 0x31
-    pmaddubsw       m1, m6, m9
-    paddw           m2, m1
-    pmaddubsw       m1, m0, m9
-    paddw           m4, m1
-
-%ifidn %1,pp
-    pmulhrsw        m4, m7                         ; m4 = word: row 10
-    pmulhrsw        m2, m7                         ; m2 = word: row 11
-    packuswb        m4, m2
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm2, m4, 1
-    movu            [r2 + r3 * 2], xm4
-    movu            [r2 + r5], xm2
-%else
-    psubw           m4, m7                         ; m4 = word: row 10
-    psubw           m2, m7                         ; m2 = word: row 11
-    movu            [r2 + r3 * 2], m4
-    movu            [r2 + r5], m2
-%endif
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_CHROMA_AVX2_16x12 pp
-    FILTER_VER_CHROMA_AVX2_16x12 ps
-
-%macro FILTER_VER_CHROMA_AVX2_16xN 2
-%if ARCH_X86_64 == 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_16x%2, 4, 8, 8
-    mov             r4d, r4m
-    shl             r4d, 6
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,pp
-    mova            m7, [pw_512]
-%else
-    add             r3d, r3d
-    mova            m7, [pw_2000]
-%endif
-    lea             r6, [r3 * 3]
-    mov             r7d, %2 / 16
-.loopH:
-    movu            xm0, [r0]
-    vinserti128     m0, m0, [r0 + r1 * 2], 1
-    movu            xm1, [r0 + r1]
-    vinserti128     m1, m1, [r0 + r4], 1
-
-    punpcklbw       m2, m0, m1
-    punpckhbw       m3, m0, m1
-    vperm2i128      m4, m2, m3, 0x20
-    vperm2i128      m2, m2, m3, 0x31
-    pmaddubsw       m4, [r5]
-    pmaddubsw       m3, m2, [r5 + mmsize]
-    paddw           m4, m3
-    pmaddubsw       m2, [r5]
-
-    vextracti128    xm0, m0, 1
-    lea             r0, [r0 + r1 * 4]
-    vinserti128     m0, m0, [r0], 1
-
-    punpcklbw       m5, m1, m0
-    punpckhbw       m3, m1, m0
-    vperm2i128      m6, m5, m3, 0x20
-    vperm2i128      m5, m5, m3, 0x31
-    pmaddubsw       m6, [r5]
-    pmaddubsw       m3, m5, [r5 + mmsize]
-    paddw           m6, m3
-    pmaddubsw       m5, [r5]
-%ifidn %1,pp
-    pmulhrsw        m4, m7                         ; m4 = word: row 0
-    pmulhrsw        m6, m7                         ; m6 = word: row 1
-    packuswb        m4, m6
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm6, m4, 1
-    movu            [r2], xm4
-    movu            [r2 + r3], xm6
-%else
-    psubw           m4, m7                         ; m4 = word: row 0
-    psubw           m6, m7                         ; m6 = word: row 1
-    movu            [r2], m4
-    movu            [r2 + r3], m6
-%endif
-
-    movu            xm4, [r0 + r1 * 2]
-    vinserti128     m4, m4, [r0 + r1], 1
-    vextracti128    xm1, m4, 1
-    vinserti128     m0, m0, xm1, 0
-
-    punpcklbw       m6, m0, m4
-    punpckhbw       m1, m0, m4
-    vperm2i128      m0, m6, m1, 0x20
-    vperm2i128      m6, m6, m1, 0x31
-    pmaddubsw       m1, m0, [r5 + mmsize]
-    paddw           m5, m1
-    pmaddubsw       m0, [r5]
-    pmaddubsw       m1, m6, [r5 + mmsize]
-    paddw           m2, m1
-    pmaddubsw       m6, [r5]
-
-%ifidn %1,pp
-    pmulhrsw        m2, m7                         ; m2 = word: row 2
-    pmulhrsw        m5, m7                         ; m5 = word: row 3
-    packuswb        m2, m5
-    vpermq          m2, m2, 11011000b
-    vextracti128    xm5, m2, 1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm5
-%else
-    psubw           m2, m7                         ; m2 = word: row 2
-    psubw           m5, m7                         ; m5 = word: row 3
-    movu            [r2 + r3 * 2], m2
-    movu            [r2 + r6], m5
-%endif
-    lea             r2, [r2 + r3 * 4]
-
-    movu            xm1, [r0 + r4]
-    lea             r0, [r0 + r1 * 4]
-    vinserti128     m1, m1, [r0], 1
-    vinserti128     m4, m4, xm1, 1
-
-    punpcklbw       m2, m4, m1
-    punpckhbw       m5, m4, m1
-    vperm2i128      m3, m2, m5, 0x20
-    vperm2i128      m2, m2, m5, 0x31
-    pmaddubsw       m5, m3, [r5 + mmsize]
-    paddw           m6, m5
-    pmaddubsw       m3, [r5]
-    pmaddubsw       m5, m2, [r5 + mmsize]
-    paddw           m0, m5
-    pmaddubsw       m2, [r5]
-
-%ifidn %1,pp
-    pmulhrsw        m6, m7                         ; m6 = word: row 4
-    pmulhrsw        m0, m7                         ; m0 = word: row 5
-    packuswb        m6, m0
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm0, m6, 1
-    movu            [r2], xm6
-    movu            [r2 + r3], xm0
-%else
-    psubw           m6, m7                         ; m6 = word: row 4
-    psubw           m0, m7                         ; m0 = word: row 5
-    movu            [r2], m6
-    movu            [r2 + r3], m0
-%endif
-
-    movu            xm6, [r0 + r1 * 2]
-    vinserti128     m6, m6, [r0 + r1], 1
-    vextracti128    xm0, m6, 1
-    vinserti128     m1, m1, xm0, 0
-
-    punpcklbw       m4, m1, m6
-    punpckhbw       m5, m1, m6
-    vperm2i128      m0, m4, m5, 0x20
-    vperm2i128      m5, m4, m5, 0x31
-    pmaddubsw       m4, m0, [r5 + mmsize]
-    paddw           m2, m4
-    pmaddubsw       m0, [r5]
-    pmaddubsw       m4, m5, [r5 + mmsize]
-    paddw           m3, m4
-    pmaddubsw       m5, [r5]
-
-%ifidn %1,pp
-    pmulhrsw        m3, m7                         ; m3 = word: row 6
-    pmulhrsw        m2, m7                         ; m2 = word: row 7
-    packuswb        m3, m2
-    vpermq          m3, m3, 11011000b
-    vextracti128    xm2, m3, 1
-    movu            [r2 + r3 * 2], xm3
-    movu            [r2 + r6], xm2
-%else
-    psubw           m3, m7                         ; m3 = word: row 6
-    psubw           m2, m7                         ; m2 = word: row 7
-    movu            [r2 + r3 * 2], m3
-    movu            [r2 + r6], m2
-%endif
-    lea             r2, [r2 + r3 * 4]
-
-    movu            xm3, [r0 + r4]
-    lea             r0, [r0 + r1 * 4]
-    vinserti128     m3, m3, [r0], 1
-    vinserti128     m6, m6, xm3, 1
-
-    punpcklbw       m2, m6, m3
-    punpckhbw       m1, m6, m3
-    vperm2i128      m4, m2, m1, 0x20
-    vperm2i128      m2, m2, m1, 0x31
-    pmaddubsw       m1, m4, [r5 + mmsize]
-    paddw           m5, m1
-    pmaddubsw       m4, [r5]
-    pmaddubsw       m1, m2, [r5 + mmsize]
-    paddw           m0, m1
-    pmaddubsw       m2, [r5]
-
-%ifidn %1,pp
-    pmulhrsw        m5, m7                         ; m5 = word: row 8
-    pmulhrsw        m0, m7                         ; m0 = word: row 9
-    packuswb        m5, m0
-    vpermq          m5, m5, 11011000b
-    vextracti128    xm0, m5, 1
-    movu            [r2], xm5
-    movu            [r2 + r3], xm0
-%else
-    psubw           m5, m7                         ; m5 = word: row 8
-    psubw           m0, m7                         ; m0 = word: row 9
-    movu            [r2], m5
-    movu            [r2 + r3], m0
-%endif
-
-    movu            xm5, [r0 + r1 * 2]
-    vinserti128     m5, m5, [r0 + r1], 1
-    vextracti128    xm0, m5, 1
-    vinserti128     m3, m3, xm0, 0
-
-    punpcklbw       m1, m3, m5
-    punpckhbw       m0, m3, m5
-    vperm2i128      m6, m1, m0, 0x20
-    vperm2i128      m0, m1, m0, 0x31
-    pmaddubsw       m1, m6, [r5 + mmsize]
-    paddw           m2, m1
-    pmaddubsw       m6, [r5]
-    pmaddubsw       m1, m0, [r5 + mmsize]
-    paddw           m4, m1
-    pmaddubsw       m0, [r5]
-
-%ifidn %1,pp
-    pmulhrsw        m4, m7                         ; m4 = word: row 10
-    pmulhrsw        m2, m7                         ; m2 = word: row 11
-    packuswb        m4, m2
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm2, m4, 1
-    movu            [r2 + r3 * 2], xm4
-    movu            [r2 + r6], xm2
-%else
-    psubw           m4, m7                         ; m4 = word: row 10
-    psubw           m2, m7                         ; m2 = word: row 11
-    movu            [r2 + r3 * 2], m4
-    movu            [r2 + r6], m2
-%endif
-    lea             r2, [r2 + r3 * 4]
-
-    movu            xm3, [r0 + r4]
-    lea             r0, [r0 + r1 * 4]
-    vinserti128     m3, m3, [r0], 1
-    vinserti128     m5, m5, xm3, 1
-
-    punpcklbw       m2, m5, m3
-    punpckhbw       m1, m5, m3
-    vperm2i128      m4, m2, m1, 0x20
-    vperm2i128      m2, m2, m1, 0x31
-    pmaddubsw       m1, m4, [r5 + mmsize]
-    paddw           m0, m1
-    pmaddubsw       m4, [r5]
-    pmaddubsw       m1, m2, [r5 + mmsize]
-    paddw           m6, m1
-    pmaddubsw       m2, [r5]
-
-%ifidn %1,pp
-    pmulhrsw        m0, m7                         ; m0 = word: row 12
-    pmulhrsw        m6, m7                         ; m6 = word: row 13
-    packuswb        m0, m6
-    vpermq          m0, m0, 11011000b
-    vextracti128    xm6, m0, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm6
-%else
-    psubw           m0, m7                         ; m0 = word: row 12
-    psubw           m6, m7                         ; m6 = word: row 13
-    movu            [r2], m0
-    movu            [r2 + r3], m6
-%endif
-
-    movu            xm5, [r0 + r1 * 2]
-    vinserti128     m5, m5, [r0 + r1], 1
-    vextracti128    xm0, m5, 1
-    vinserti128     m3, m3, xm0, 0
-
-    punpcklbw       m1, m3, m5
-    punpckhbw       m0, m3, m5
-    vperm2i128      m6, m1, m0, 0x20
-    vperm2i128      m0, m1, m0, 0x31
-    pmaddubsw       m6, [r5 + mmsize]
-    paddw           m2, m6
-    pmaddubsw       m0, [r5 + mmsize]
-    paddw           m4, m0
-
-%ifidn %1,pp
-    pmulhrsw        m4, m7                         ; m4 = word: row 14
-    pmulhrsw        m2, m7                         ; m2 = word: row 15
-    packuswb        m4, m2
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm2, m4, 1
-    movu            [r2 + r3 * 2], xm4
-    movu            [r2 + r6], xm2
-%else
-    psubw           m4, m7                         ; m4 = word: row 14
-    psubw           m2, m7                         ; m2 = word: row 15
-    movu            [r2 + r3 * 2], m4
-    movu            [r2 + r6], m2
-%endif
-    lea             r2, [r2 + r3 * 4]
-    dec             r7d
-    jnz             .loopH
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_CHROMA_AVX2_16xN pp, 32
-    FILTER_VER_CHROMA_AVX2_16xN ps, 32
-    FILTER_VER_CHROMA_AVX2_16xN pp, 64
-    FILTER_VER_CHROMA_AVX2_16xN ps, 64
-
-%macro FILTER_VER_CHROMA_AVX2_16x24 1
-%if ARCH_X86_64 == 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_16x24, 4, 6, 15
-    mov             r4d, r4m
-    shl             r4d, 6
+    shl             r4d, 7
 
 %ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
+    lea             r5, [tab_LumaCoeffVer_32]
     add             r5, r4
 %else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
 
-    mova            m12, [r5]
-    mova            m13, [r5 + mmsize]
     lea             r4, [r1 * 3]
-    sub             r0, r1
+    sub             r0, r4
 %ifidn %1,pp
     mova            m14, [pw_512]
 %else
     add             r3d, r3d
     vbroadcasti128  m14, [pw_2000]
 %endif
-    lea             r5, [r3 * 3]
+    lea             r6, [r3 * 3]
 
     movu            xm0, [r0]                       ; m0 = row 0
     movu            xm1, [r0 + r1]                  ; m1 = row 1
     punpckhbw       xm2, xm0, xm1
     punpcklbw       xm0, xm1
     vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m0, m12
+    pmaddubsw       m0, [r5]
     movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
     punpckhbw       xm3, xm1, xm2
     punpcklbw       xm1, xm2
     vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m1, m12
+    pmaddubsw       m1, [r5]
     movu            xm3, [r0 + r4]                  ; m3 = row 3
     punpckhbw       xm4, xm2, xm3
     punpcklbw       xm2, xm3
     vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, m13
+    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
     paddw           m0, m4
-    pmaddubsw       m2, m12
+    pmaddubsw       m2, [r5]
     lea             r0, [r0 + r1 * 4]
     movu            xm4, [r0]                       ; m4 = row 4
     punpckhbw       xm5, xm3, xm4
     punpcklbw       xm3, xm4
     vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, m13
+    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
     paddw           m1, m5
-    pmaddubsw       m3, m12
+    pmaddubsw       m3, [r5]
     movu            xm5, [r0 + r1]                  ; m5 = row 5
     punpckhbw       xm6, xm4, xm5
     punpcklbw       xm4, xm5
     vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, m13
+    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
+    paddw           m0, m6
+    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
     paddw           m2, m6
-    pmaddubsw       m4, m12
+    pmaddubsw       m4, [r5]
     movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
     punpckhbw       xm7, xm5, xm6
     punpcklbw       xm5, xm6
     vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m7, m5, m13
+    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
+    paddw           m1, m7
+    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
     paddw           m3, m7
-    pmaddubsw       m5, m12
+    pmaddubsw       m5, [r5]
     movu            xm7, [r0 + r4]                  ; m7 = row 7
     punpckhbw       xm8, xm6, xm7
     punpcklbw       xm6, xm7
     vinserti128     m6, m6, xm8, 1
-    pmaddubsw       m8, m6, m13
+    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
+    paddw           m0, m8
+    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
+    paddw           m2, m8
+    pmaddubsw       m8, m6, [r5 + 1 * mmsize]
     paddw           m4, m8
-    pmaddubsw       m6, m12
+    pmaddubsw       m6, [r5]
     lea             r0, [r0 + r1 * 4]
     movu            xm8, [r0]                       ; m8 = row 8
     punpckhbw       xm9, xm7, xm8
     punpcklbw       xm7, xm8
     vinserti128     m7, m7, xm9, 1
-    pmaddubsw       m9, m7, m13
+    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
+    paddw           m1, m9
+    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
+    paddw           m3, m9
+    pmaddubsw       m9, m7, [r5 + 1 * mmsize]
     paddw           m5, m9
-    pmaddubsw       m7, m12
+    pmaddubsw       m7, [r5]
     movu            xm9, [r0 + r1]                  ; m9 = row 9
     punpckhbw       xm10, xm8, xm9
     punpcklbw       xm8, xm9
     vinserti128     m8, m8, xm10, 1
-    pmaddubsw       m10, m8, m13
+    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
+    paddw           m2, m10
+    pmaddubsw       m10, m8, [r5 + 2 * mmsize]
+    paddw           m4, m10
+    pmaddubsw       m10, m8, [r5 + 1 * mmsize]
     paddw           m6, m10
-    pmaddubsw       m8, m12
+    pmaddubsw       m8, [r5]
     movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
     punpckhbw       xm11, xm9, xm10
     punpcklbw       xm9, xm10
     vinserti128     m9, m9, xm11, 1
-    pmaddubsw       m11, m9, m13
-    paddw           m7, m11
-    pmaddubsw       m9, m12
-
+    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
+    paddw           m3, m11
+    pmaddubsw       m11, m9, [r5 + 2 * mmsize]
+    paddw           m5, m11
+    pmaddubsw       m11, m9, [r5 + 1 * mmsize]
+    paddw           m7, m11
+    pmaddubsw       m9, [r5]
+    movu            xm11, [r0 + r4]                 ; m11 = row 11
+    punpckhbw       xm12, xm10, xm11
+    punpcklbw       xm10, xm11
+    vinserti128     m10, m10, xm12, 1
+    pmaddubsw       m12, m10, [r5 + 3 * mmsize]
+    paddw           m4, m12
+    pmaddubsw       m12, m10, [r5 + 2 * mmsize]
+    paddw           m6, m12
+    pmaddubsw       m12, m10, [r5 + 1 * mmsize]
+    paddw           m8, m12
+    pmaddubsw       m10, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm12, [r0]                      ; m12 = row 12
+    punpckhbw       xm13, xm11, xm12
+    punpcklbw       xm11, xm12
+    vinserti128     m11, m11, xm13, 1
+    pmaddubsw       m13, m11, [r5 + 3 * mmsize]
+    paddw           m5, m13
+    pmaddubsw       m13, m11, [r5 + 2 * mmsize]
+    paddw           m7, m13
+    pmaddubsw       m13, m11, [r5 + 1 * mmsize]
+    paddw           m9, m13
+    pmaddubsw       m11, [r5]
+
 %ifidn %1,pp
     pmulhrsw        m0, m14                         ; m0 = word: row 0
     pmulhrsw        m1, m14                         ; m1 = word: row 1
@@ -7209,29 +4569,22 @@ cglobal interp_4tap_vert_%1_16x24, 4, 6, 15
     pmulhrsw        m3, m14                         ; m3 = word: row 3
     pmulhrsw        m4, m14                         ; m4 = word: row 4
     pmulhrsw        m5, m14                         ; m5 = word: row 5
-    pmulhrsw        m6, m14                         ; m6 = word: row 6
-    pmulhrsw        m7, m14                         ; m7 = word: row 7
     packuswb        m0, m1
     packuswb        m2, m3
     packuswb        m4, m5
-    packuswb        m6, m7
-    vpermq          m0, m0, q3120
-    vpermq          m2, m2, q3120
-    vpermq          m4, m4, q3120
-    vpermq          m6, m6, q3120
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    vpermq          m4, m4, 11011000b
     vextracti128    xm1, m0, 1
     vextracti128    xm3, m2, 1
     vextracti128    xm5, m4, 1
-    vextracti128    xm7, m6, 1
     movu            [r2], xm0
     movu            [r2 + r3], xm1
     movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r5], xm3
+    movu            [r2 + r6], xm3
     lea             r2, [r2 + r3 * 4]
     movu            [r2], xm4
     movu            [r2 + r3], xm5
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r5], xm7
 %else
     psubw           m0, m14                         ; m0 = word: row 0
     psubw           m1, m14                         ; m1 = word: row 1
@@ -7239,959 +4592,461 @@ cglobal interp_4tap_vert_%1_16x24, 4, 6, 15
     psubw           m3, m14                         ; m3 = word: row 3
     psubw           m4, m14                         ; m4 = word: row 4
     psubw           m5, m14                         ; m5 = word: row 5
-    psubw           m6, m14                         ; m6 = word: row 6
-    psubw           m7, m14                         ; m7 = word: row 7
     movu            [r2], m0
     movu            [r2 + r3], m1
     movu            [r2 + r3 * 2], m2
-    movu            [r2 + r5], m3
+    movu            [r2 + r6], m3
     lea             r2, [r2 + r3 * 4]
     movu            [r2], m4
     movu            [r2 + r3], m5
-    movu            [r2 + r3 * 2], m6
-    movu            [r2 + r5], m7
 %endif
-    lea             r2, [r2 + r3 * 4]
 
-    movu            xm11, [r0 + r4]                 ; m11 = row 11
-    punpckhbw       xm6, xm10, xm11
-    punpcklbw       xm10, xm11
-    vinserti128     m10, m10, xm6, 1
-    pmaddubsw       m6, m10, m13
-    paddw           m8, m6
-    pmaddubsw       m10, m12
-    lea             r0, [r0 + r1 * 4]
-    movu            xm6, [r0]                       ; m6 = row 12
-    punpckhbw       xm7, xm11, xm6
-    punpcklbw       xm11, xm6
-    vinserti128     m11, m11, xm7, 1
-    pmaddubsw       m7, m11, m13
-    paddw           m9, m7
-    pmaddubsw       m11, m12
-
-    movu            xm7, [r0 + r1]                  ; m7 = row 13
-    punpckhbw       xm0, xm6, xm7
-    punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm0, 1
-    pmaddubsw       m0, m6, m13
+    movu            xm13, [r0 + r1]                 ; m13 = row 13
+    punpckhbw       xm0, xm12, xm13
+    punpcklbw       xm12, xm13
+    vinserti128     m12, m12, xm0, 1
+    pmaddubsw       m0, m12, [r5 + 3 * mmsize]
+    paddw           m6, m0
+    pmaddubsw       m0, m12, [r5 + 2 * mmsize]
+    paddw           m8, m0
+    pmaddubsw       m0, m12, [r5 + 1 * mmsize]
     paddw           m10, m0
-    pmaddubsw       m6, m12
     movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
-    punpckhbw       xm1, xm7, xm0
-    punpcklbw       xm7, xm0
-    vinserti128     m7, m7, xm1, 1
-    pmaddubsw       m1, m7, m13
+    punpckhbw       xm1, xm13, xm0
+    punpcklbw       xm13, xm0
+    vinserti128     m13, m13, xm1, 1
+    pmaddubsw       m1, m13, [r5 + 3 * mmsize]
+    paddw           m7, m1
+    pmaddubsw       m1, m13, [r5 + 2 * mmsize]
+    paddw           m9, m1
+    pmaddubsw       m1, m13, [r5 + 1 * mmsize]
     paddw           m11, m1
-    pmaddubsw       m7, m12
+
+%ifidn %1,pp
+    pmulhrsw        m6, m14                         ; m6 = word: row 6
+    pmulhrsw        m7, m14                         ; m7 = word: row 7
+    packuswb        m6, m7
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm7, m6, 1
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r6], xm7
+%else
+    psubw           m6, m14                         ; m6 = word: row 6
+    psubw           m7, m14                         ; m7 = word: row 7
+    movu            [r2 + r3 * 2], m6
+    movu            [r2 + r6], m7
+%endif
+    lea             r2, [r2 + r3 * 4]
+
     movu            xm1, [r0 + r4]                  ; m1 = row 15
     punpckhbw       xm2, xm0, xm1
     punpcklbw       xm0, xm1
     vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m2, m0, m13
-    paddw           m6, m2
-    pmaddubsw       m0, m12
+    pmaddubsw       m2, m0, [r5 + 3 * mmsize]
+    paddw           m8, m2
+    pmaddubsw       m2, m0, [r5 + 2 * mmsize]
+    paddw           m10, m2
     lea             r0, [r0 + r1 * 4]
     movu            xm2, [r0]                       ; m2 = row 16
     punpckhbw       xm3, xm1, xm2
     punpcklbw       xm1, xm2
     vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m3, m1, m13
-    paddw           m7, m3
-    pmaddubsw       m1, m12
+    pmaddubsw       m3, m1, [r5 + 3 * mmsize]
+    paddw           m9, m3
+    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
+    paddw           m11, m3
     movu            xm3, [r0 + r1]                  ; m3 = row 17
     punpckhbw       xm4, xm2, xm3
     punpcklbw       xm2, xm3
     vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, m13
-    paddw           m0, m4
-    pmaddubsw       m2, m12
+    pmaddubsw       m4, m2, [r5 + 3 * mmsize]
+    paddw           m10, m4
     movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
     punpckhbw       xm5, xm3, xm4
     punpcklbw       xm3, xm4
     vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, m13
-    paddw           m1, m5
-    pmaddubsw       m3, m12
+    pmaddubsw       m5, m3, [r5 + 3 * mmsize]
+    paddw           m11, m5
 
 %ifidn %1,pp
     pmulhrsw        m8, m14                         ; m8 = word: row 8
     pmulhrsw        m9, m14                         ; m9 = word: row 9
     pmulhrsw        m10, m14                        ; m10 = word: row 10
     pmulhrsw        m11, m14                        ; m11 = word: row 11
-    pmulhrsw        m6, m14                         ; m6 = word: row 12
-    pmulhrsw        m7, m14                         ; m7 = word: row 13
-    pmulhrsw        m0, m14                         ; m0 = word: row 14
-    pmulhrsw        m1, m14                         ; m1 = word: row 15
     packuswb        m8, m9
     packuswb        m10, m11
-    packuswb        m6, m7
-    packuswb        m0, m1
-    vpermq          m8, m8, q3120
-    vpermq          m10, m10, q3120
-    vpermq          m6, m6, q3120
-    vpermq          m0, m0, q3120
+    vpermq          m8, m8, 11011000b
+    vpermq          m10, m10, 11011000b
     vextracti128    xm9, m8, 1
     vextracti128    xm11, m10, 1
-    vextracti128    xm7, m6, 1
-    vextracti128    xm1, m0, 1
     movu            [r2], xm8
     movu            [r2 + r3], xm9
     movu            [r2 + r3 * 2], xm10
-    movu            [r2 + r5], xm11
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm6
-    movu            [r2 + r3], xm7
-    movu            [r2 + r3 * 2], xm0
-    movu            [r2 + r5], xm1
+    movu            [r2 + r6], xm11
 %else
     psubw           m8, m14                         ; m8 = word: row 8
     psubw           m9, m14                         ; m9 = word: row 9
     psubw           m10, m14                        ; m10 = word: row 10
     psubw           m11, m14                        ; m11 = word: row 11
-    psubw           m6, m14                         ; m6 = word: row 12
-    psubw           m7, m14                         ; m7 = word: row 13
-    psubw           m0, m14                         ; m0 = word: row 14
-    psubw           m1, m14                         ; m1 = word: row 15
     movu            [r2], m8
     movu            [r2 + r3], m9
     movu            [r2 + r3 * 2], m10
-    movu            [r2 + r5], m11
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], m6
-    movu            [r2 + r3], m7
-    movu            [r2 + r3 * 2], m0
-    movu            [r2 + r5], m1
+    movu            [r2 + r6], m11
 %endif
-    lea             r2, [r2 + r3 * 4]
+    RET
+%endif
+%endmacro
 
-    movu            xm5, [r0 + r4]                  ; m5 = row 19
+    FILTER_VER_LUMA_AVX2_16x12 pp
+    FILTER_VER_LUMA_AVX2_16x12 ps
+
+%macro FILTER_VER_LUMA_AVX2_16x8 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_16x8, 4, 6, 15
+    mov             r4d, r4m
+    shl             r4d, 7
+%ifdef PIC
+    lea             r5, [tab_LumaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
+%endif
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+%ifidn %1,pp
+    mova            m14, [pw_512]
+%else
+    add             r3d, r3d
+    vbroadcasti128  m14, [pw_2000]
+%endif
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddubsw       m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
+    paddw           m0, m4
+    pmaddubsw       m2, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
+    paddw           m1, m5
+    pmaddubsw       m3, [r5]
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
     punpckhbw       xm6, xm4, xm5
     punpcklbw       xm4, xm5
     vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, m13
+    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
+    paddw           m0, m6
+    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
     paddw           m2, m6
-    pmaddubsw       m4, m12
-    lea             r0, [r0 + r1 * 4]
-    movu            xm6, [r0]                       ; m6 = row 20
+    pmaddubsw       m4, [r5]
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
     punpckhbw       xm7, xm5, xm6
     punpcklbw       xm5, xm6
     vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m7, m5, m13
+    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
+    paddw           m1, m7
+    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
     paddw           m3, m7
-    pmaddubsw       m5, m12
-    movu            xm7, [r0 + r1]                  ; m7 = row 21
-    punpckhbw       xm0, xm6, xm7
+    pmaddubsw       m5, [r5]
+    movu            xm7, [r0 + r4]                  ; m7 = row 7
+    punpckhbw       xm8, xm6, xm7
     punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm0, 1
-    pmaddubsw       m0, m6, m13
-    paddw           m4, m0
-    pmaddubsw       m6, m12
-    movu            xm0, [r0 + r1 * 2]              ; m0 = row 22
-    punpckhbw       xm1, xm7, xm0
-    punpcklbw       xm7, xm0
-    vinserti128     m7, m7, xm1, 1
-    pmaddubsw       m1, m7, m13
-    paddw           m5, m1
-    pmaddubsw       m7, m12
-    movu            xm1, [r0 + r4]                  ; m1 = row 23
-    punpckhbw       xm8, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm8, 1
-    pmaddubsw       m8, m0, m13
-    paddw           m6, m8
-    pmaddubsw       m0, m12
+    vinserti128     m6, m6, xm8, 1
+    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
+    paddw           m0, m8
+    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
+    paddw           m2, m8
+    pmaddubsw       m8, m6, [r5 + 1 * mmsize]
+    paddw           m4, m8
+    pmaddubsw       m6, [r5]
     lea             r0, [r0 + r1 * 4]
-    movu            xm8, [r0]                       ; m8 = row 24
-    punpckhbw       xm9, xm1, xm8
-    punpcklbw       xm1, xm8
-    vinserti128     m1, m1, xm9, 1
-    pmaddubsw       m9, m1, m13
-    paddw           m7, m9
-    pmaddubsw       m1, m12
-    movu            xm9, [r0 + r1]                  ; m9 = row 25
+    movu            xm8, [r0]                       ; m8 = row 8
+    punpckhbw       xm9, xm7, xm8
+    punpcklbw       xm7, xm8
+    vinserti128     m7, m7, xm9, 1
+    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
+    paddw           m1, m9
+    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
+    paddw           m3, m9
+    pmaddubsw       m9, m7, [r5 + 1 * mmsize]
+    paddw           m5, m9
+    pmaddubsw       m7, [r5]
+    movu            xm9, [r0 + r1]                  ; m9 = row 9
     punpckhbw       xm10, xm8, xm9
     punpcklbw       xm8, xm9
     vinserti128     m8, m8, xm10, 1
-    pmaddubsw       m8, m13
-    paddw           m0, m8
-    movu            xm10, [r0 + r1 * 2]             ; m10 = row 26
+    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
+    paddw           m2, m10
+    pmaddubsw       m10, m8, [r5 + 2 * mmsize]
+    paddw           m4, m10
+    pmaddubsw       m10, m8, [r5 + 1 * mmsize]
+    paddw           m6, m10
+    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
     punpckhbw       xm11, xm9, xm10
     punpcklbw       xm9, xm10
     vinserti128     m9, m9, xm11, 1
-    pmaddubsw       m9, m13
-    paddw           m1, m9
-
+    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
+    paddw           m3, m11
+    pmaddubsw       m11, m9, [r5 + 2 * mmsize]
+    paddw           m5, m11
+    pmaddubsw       m11, m9, [r5 + 1 * mmsize]
+    paddw           m7, m11
+    movu            xm11, [r0 + r4]                 ; m11 = row 11
+    punpckhbw       xm12, xm10, xm11
+    punpcklbw       xm10, xm11
+    vinserti128     m10, m10, xm12, 1
+    pmaddubsw       m12, m10, [r5 + 3 * mmsize]
+    paddw           m4, m12
+    pmaddubsw       m12, m10, [r5 + 2 * mmsize]
+    paddw           m6, m12
+    lea             r0, [r0 + r1 * 4]
+    movu            xm12, [r0]                      ; m12 = row 12
+    punpckhbw       xm13, xm11, xm12
+    punpcklbw       xm11, xm12
+    vinserti128     m11, m11, xm13, 1
+    pmaddubsw       m13, m11, [r5 + 3 * mmsize]
+    paddw           m5, m13
+    pmaddubsw       m13, m11, [r5 + 2 * mmsize]
+    paddw           m7, m13
+    lea             r4, [r3 * 3]
 %ifidn %1,pp
-    pmulhrsw        m2, m14                         ; m2 = word: row 16
-    pmulhrsw        m3, m14                         ; m3 = word: row 17
-    pmulhrsw        m4, m14                         ; m4 = word: row 18
-    pmulhrsw        m5, m14                         ; m5 = word: row 19
-    pmulhrsw        m6, m14                         ; m6 = word: row 20
-    pmulhrsw        m7, m14                         ; m7 = word: row 21
-    pmulhrsw        m0, m14                         ; m0 = word: row 22
-    pmulhrsw        m1, m14                         ; m1 = word: row 23
+    pmulhrsw        m0, m14                         ; m0 = word: row 0
+    pmulhrsw        m1, m14                         ; m1 = word: row 1
+    pmulhrsw        m2, m14                         ; m2 = word: row 2
+    pmulhrsw        m3, m14                         ; m3 = word: row 3
+    pmulhrsw        m4, m14                         ; m4 = word: row 4
+    pmulhrsw        m5, m14                         ; m5 = word: row 5
+    packuswb        m0, m1
     packuswb        m2, m3
     packuswb        m4, m5
-    packuswb        m6, m7
-    packuswb        m0, m1
-    vpermq          m2, m2, q3120
-    vpermq          m4, m4, q3120
-    vpermq          m6, m6, q3120
-    vpermq          m0, m0, q3120
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    vpermq          m4, m4, 11011000b
+    vextracti128    xm1, m0, 1
     vextracti128    xm3, m2, 1
     vextracti128    xm5, m4, 1
-    vextracti128    xm7, m6, 1
-    vextracti128    xm1, m0, 1
-    movu            [r2], xm2
-    movu            [r2 + r3], xm3
-    movu            [r2 + r3 * 2], xm4
-    movu            [r2 + r5], xm5
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r4], xm3
     lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm6
-    movu            [r2 + r3], xm7
-    movu            [r2 + r3 * 2], xm0
-    movu            [r2 + r5], xm1
+    movu            [r2], xm4
+    movu            [r2 + r3], xm5
 %else
-    psubw           m2, m14                         ; m2 = word: row 16
-    psubw           m3, m14                         ; m3 = word: row 17
-    psubw           m4, m14                         ; m4 = word: row 18
-    psubw           m5, m14                         ; m5 = word: row 19
-    psubw           m6, m14                         ; m6 = word: row 20
-    psubw           m7, m14                         ; m7 = word: row 21
-    psubw           m0, m14                         ; m0 = word: row 22
-    psubw           m1, m14                         ; m1 = word: row 23
-    movu            [r2], m2
-    movu            [r2 + r3], m3
-    movu            [r2 + r3 * 2], m4
-    movu            [r2 + r5], m5
+    psubw           m0, m14                         ; m0 = word: row 0
+    psubw           m1, m14                         ; m1 = word: row 1
+    psubw           m2, m14                         ; m2 = word: row 2
+    psubw           m3, m14                         ; m3 = word: row 3
+    psubw           m4, m14                         ; m4 = word: row 4
+    psubw           m5, m14                         ; m5 = word: row 5
+    movu            [r2], m0
+    movu            [r2 + r3], m1
+    movu            [r2 + r3 * 2], m2
+    movu            [r2 + r4], m3
     lea             r2, [r2 + r3 * 4]
-    movu            [r2], m6
-    movu            [r2 + r3], m7
-    movu            [r2 + r3 * 2], m0
-    movu            [r2 + r5], m1
+    movu            [r2], m4
+    movu            [r2 + r3], m5
+%endif
+    movu            xm13, [r0 + r1]                 ; m13 = row 13
+    punpckhbw       xm0, xm12, xm13
+    punpcklbw       xm12, xm13
+    vinserti128     m12, m12, xm0, 1
+    pmaddubsw       m0, m12, [r5 + 3 * mmsize]
+    paddw           m6, m0
+    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
+    punpckhbw       xm1, xm13, xm0
+    punpcklbw       xm13, xm0
+    vinserti128     m13, m13, xm1, 1
+    pmaddubsw       m1, m13, [r5 + 3 * mmsize]
+    paddw           m7, m1
+%ifidn %1,pp
+    pmulhrsw        m6, m14                         ; m6 = word: row 6
+    pmulhrsw        m7, m14                         ; m7 = word: row 7
+    packuswb        m6, m7
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm7, m6, 1
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r4], xm7
+%else
+    psubw           m6, m14                         ; m6 = word: row 6
+    psubw           m7, m14                         ; m7 = word: row 7
+    movu            [r2 + r3 * 2], m6
+    movu            [r2 + r4], m7
 %endif
     RET
 %endif
 %endmacro
 
-    FILTER_VER_CHROMA_AVX2_16x24 pp
-    FILTER_VER_CHROMA_AVX2_16x24 ps
+    FILTER_VER_LUMA_AVX2_16x8 pp
+    FILTER_VER_LUMA_AVX2_16x8 ps
 
-%macro FILTER_VER_CHROMA_AVX2_24x32 1
+%macro FILTER_VER_LUMA_AVX2_16x4 1
 INIT_YMM avx2
 %if ARCH_X86_64 == 1
-cglobal interp_4tap_vert_%1_24x32, 4, 9, 10
+cglobal interp_8tap_vert_%1_16x4, 4, 6, 13
     mov             r4d, r4m
-    shl             r4d, 6
-
+    shl             r4d, 7
 %ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
+    lea             r5, [tab_LumaCoeffVer_32]
     add             r5, r4
 %else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
-
-    mova            m8, [r5]
-    mova            m9, [r5 + mmsize]
     lea             r4, [r1 * 3]
-    sub             r0, r1
+    sub             r0, r4
 %ifidn %1,pp
-    mova            m7, [pw_512]
+    mova            m12, [pw_512]
 %else
     add             r3d, r3d
-    vbroadcasti128  m7, [pw_2000]
-%endif
-    lea             r6, [r3 * 3]
-    mov             r5d, 2
-.loopH:
-    movu            xm0, [r0]
-    vinserti128     m0, m0, [r0 + r1 * 2], 1
-    movu            xm1, [r0 + r1]
-    vinserti128     m1, m1, [r0 + r4], 1
-
-    punpcklbw       m2, m0, m1
-    punpckhbw       m3, m0, m1
-    vperm2i128      m4, m2, m3, 0x20
-    vperm2i128      m2, m2, m3, 0x31
-    pmaddubsw       m4, m8
-    pmaddubsw       m3, m2, m9
-    paddw           m4, m3
-    pmaddubsw       m2, m8
-
-    vextracti128    xm0, m0, 1
-    lea             r7, [r0 + r1 * 4]
-    vinserti128     m0, m0, [r7], 1
-
-    punpcklbw       m5, m1, m0
-    punpckhbw       m3, m1, m0
-    vperm2i128      m6, m5, m3, 0x20
-    vperm2i128      m5, m5, m3, 0x31
-    pmaddubsw       m6, m8
-    pmaddubsw       m3, m5, m9
-    paddw           m6, m3
-    pmaddubsw       m5, m8
-%ifidn %1,pp
-    pmulhrsw        m4, m7                         ; m4 = word: row 0
-    pmulhrsw        m6, m7                         ; m6 = word: row 1
-    packuswb        m4, m6
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm6, m4, 1
-    movu            [r2], xm4
-    movu            [r2 + r3], xm6
-%else
-    psubw           m4, m7                         ; m4 = word: row 0
-    psubw           m6, m7                         ; m6 = word: row 1
-    movu            [r2], m4
-    movu            [r2 + r3], m6
+    vbroadcasti128  m12, [pw_2000]
 %endif
-
-    movu            xm4, [r7 + r1 * 2]
-    vinserti128     m4, m4, [r7 + r1], 1
-    vextracti128    xm1, m4, 1
-    vinserti128     m0, m0, xm1, 0
-
-    punpcklbw       m6, m0, m4
-    punpckhbw       m1, m0, m4
-    vperm2i128      m0, m6, m1, 0x20
-    vperm2i128      m6, m6, m1, 0x31
-    pmaddubsw       m1, m0, m9
-    paddw           m5, m1
-    pmaddubsw       m0, m8
-    pmaddubsw       m1, m6, m9
-    paddw           m2, m1
-    pmaddubsw       m6, m8
-
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddubsw       m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
+    paddw           m0, m4
+    pmaddubsw       m2, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
+    paddw           m1, m5
+    pmaddubsw       m3, [r5]
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhbw       xm6, xm4, xm5
+    punpcklbw       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
+    paddw           m0, m6
+    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
+    paddw           m2, m6
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhbw       xm7, xm5, xm6
+    punpcklbw       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
+    paddw           m1, m7
+    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
+    paddw           m3, m7
+    movu            xm7, [r0 + r4]                  ; m7 = row 7
+    punpckhbw       xm8, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
+    paddw           m0, m8
+    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
+    paddw           m2, m8
+    lea             r0, [r0 + r1 * 4]
+    movu            xm8, [r0]                       ; m8 = row 8
+    punpckhbw       xm9, xm7, xm8
+    punpcklbw       xm7, xm8
+    vinserti128     m7, m7, xm9, 1
+    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
+    paddw           m1, m9
+    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
+    paddw           m3, m9
+    movu            xm9, [r0 + r1]                  ; m9 = row 9
+    punpckhbw       xm10, xm8, xm9
+    punpcklbw       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
+    paddw           m2, m10
+    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
+    punpckhbw       xm11, xm9, xm10
+    punpcklbw       xm9, xm10
+    vinserti128     m9, m9, xm11, 1
+    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
+    paddw           m3, m11
 %ifidn %1,pp
-    pmulhrsw        m2, m7                         ; m2 = word: row 2
-    pmulhrsw        m5, m7                         ; m5 = word: row 3
-    packuswb        m2, m5
+    pmulhrsw        m0, m12                         ; m0 = word: row 0
+    pmulhrsw        m1, m12                         ; m1 = word: row 1
+    pmulhrsw        m2, m12                         ; m2 = word: row 2
+    pmulhrsw        m3, m12                         ; m3 = word: row 3
+    packuswb        m0, m1
+    packuswb        m2, m3
+    vpermq          m0, m0, 11011000b
     vpermq          m2, m2, 11011000b
-    vextracti128    xm5, m2, 1
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m2, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
     movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm5
+    lea             r4, [r3 * 3]
+    movu            [r2 + r4], xm3
 %else
-    psubw           m2, m7                         ; m2 = word: row 2
-    psubw           m5, m7                         ; m5 = word: row 3
+    psubw           m0, m12                         ; m0 = word: row 0
+    psubw           m1, m12                         ; m1 = word: row 1
+    psubw           m2, m12                         ; m2 = word: row 2
+    psubw           m3, m12                         ; m3 = word: row 3
+    movu            [r2], m0
+    movu            [r2 + r3], m1
     movu            [r2 + r3 * 2], m2
-    movu            [r2 + r6], m5
-%endif
-    lea             r8, [r2 + r3 * 4]
-
-    movu            xm1, [r7 + r4]
-    lea             r7, [r7 + r1 * 4]
-    vinserti128     m1, m1, [r7], 1
-    vinserti128     m4, m4, xm1, 1
-
-    punpcklbw       m2, m4, m1
-    punpckhbw       m5, m4, m1
-    vperm2i128      m3, m2, m5, 0x20
-    vperm2i128      m2, m2, m5, 0x31
-    pmaddubsw       m5, m3, m9
-    paddw           m6, m5
-    pmaddubsw       m3, m8
-    pmaddubsw       m5, m2, m9
-    paddw           m0, m5
-    pmaddubsw       m2, m8
-
-%ifidn %1,pp
-    pmulhrsw        m6, m7                         ; m6 = word: row 4
-    pmulhrsw        m0, m7                         ; m0 = word: row 5
-    packuswb        m6, m0
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm0, m6, 1
-    movu            [r8], xm6
-    movu            [r8 + r3], xm0
-%else
-    psubw           m6, m7                         ; m6 = word: row 4
-    psubw           m0, m7                         ; m0 = word: row 5
-    movu            [r8], m6
-    movu            [r8 + r3], m0
-%endif
-
-    movu            xm6, [r7 + r1 * 2]
-    vinserti128     m6, m6, [r7 + r1], 1
-    vextracti128    xm0, m6, 1
-    vinserti128     m1, m1, xm0, 0
-
-    punpcklbw       m4, m1, m6
-    punpckhbw       m5, m1, m6
-    vperm2i128      m0, m4, m5, 0x20
-    vperm2i128      m5, m4, m5, 0x31
-    pmaddubsw       m4, m0, m9
-    paddw           m2, m4
-    pmaddubsw       m0, m8
-    pmaddubsw       m4, m5, m9
-    paddw           m3, m4
-    pmaddubsw       m5, m8
-
-%ifidn %1,pp
-    pmulhrsw        m3, m7                         ; m3 = word: row 6
-    pmulhrsw        m2, m7                         ; m2 = word: row 7
-    packuswb        m3, m2
-    vpermq          m3, m3, 11011000b
-    vextracti128    xm2, m3, 1
-    movu            [r8 + r3 * 2], xm3
-    movu            [r8 + r6], xm2
-%else
-    psubw           m3, m7                         ; m3 = word: row 6
-    psubw           m2, m7                         ; m2 = word: row 7
-    movu            [r8 + r3 * 2], m3
-    movu            [r8 + r6], m2
+    lea             r4, [r3 * 3]
+    movu            [r2 + r4], m3
 %endif
-    lea             r8, [r8 + r3 * 4]
-
-    movu            xm3, [r7 + r4]
-    lea             r7, [r7 + r1 * 4]
-    vinserti128     m3, m3, [r7], 1
-    vinserti128     m6, m6, xm3, 1
-
-    punpcklbw       m2, m6, m3
-    punpckhbw       m1, m6, m3
-    vperm2i128      m4, m2, m1, 0x20
-    vperm2i128      m2, m2, m1, 0x31
-    pmaddubsw       m1, m4, m9
-    paddw           m5, m1
-    pmaddubsw       m4, m8
-    pmaddubsw       m1, m2, m9
-    paddw           m0, m1
-    pmaddubsw       m2, m8
-
-%ifidn %1,pp
-    pmulhrsw        m5, m7                         ; m5 = word: row 8
-    pmulhrsw        m0, m7                         ; m0 = word: row 9
-    packuswb        m5, m0
-    vpermq          m5, m5, 11011000b
-    vextracti128    xm0, m5, 1
-    movu            [r8], xm5
-    movu            [r8 + r3], xm0
-%else
-    psubw           m5, m7                         ; m5 = word: row 8
-    psubw           m0, m7                         ; m0 = word: row 9
-    movu            [r8], m5
-    movu            [r8 + r3], m0
+    RET
 %endif
+%endmacro
 
-    movu            xm5, [r7 + r1 * 2]
-    vinserti128     m5, m5, [r7 + r1], 1
-    vextracti128    xm0, m5, 1
-    vinserti128     m3, m3, xm0, 0
-
-    punpcklbw       m1, m3, m5
-    punpckhbw       m0, m3, m5
-    vperm2i128      m6, m1, m0, 0x20
-    vperm2i128      m0, m1, m0, 0x31
-    pmaddubsw       m1, m6, m9
-    paddw           m2, m1
-    pmaddubsw       m6, m8
-    pmaddubsw       m1, m0, m9
-    paddw           m4, m1
-    pmaddubsw       m0, m8
+    FILTER_VER_LUMA_AVX2_16x4 pp
+    FILTER_VER_LUMA_AVX2_16x4 ps
+%macro FILTER_VER_LUMA_AVX2_16xN 3
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%3_%1x%2, 4, 9, 15
+    mov             r4d, r4m
+    shl             r4d, 7
 
-%ifidn %1,pp
-    pmulhrsw        m4, m7                         ; m4 = word: row 10
-    pmulhrsw        m2, m7                         ; m2 = word: row 11
-    packuswb        m4, m2
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm2, m4, 1
-    movu            [r8 + r3 * 2], xm4
-    movu            [r8 + r6], xm2
+%ifdef PIC
+    lea             r5, [tab_LumaCoeffVer_32]
+    add             r5, r4
 %else
-    psubw           m4, m7                         ; m4 = word: row 10
-    psubw           m2, m7                         ; m2 = word: row 11
-    movu            [r8 + r3 * 2], m4
-    movu            [r8 + r6], m2
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
-    lea             r8, [r8 + r3 * 4]
-
-    movu            xm3, [r7 + r4]
-    lea             r7, [r7 + r1 * 4]
-    vinserti128     m3, m3, [r7], 1
-    vinserti128     m5, m5, xm3, 1
-
-    punpcklbw       m2, m5, m3
-    punpckhbw       m1, m5, m3
-    vperm2i128      m4, m2, m1, 0x20
-    vperm2i128      m2, m2, m1, 0x31
-    pmaddubsw       m1, m4, m9
-    paddw           m0, m1
-    pmaddubsw       m4, m8
-    pmaddubsw       m1, m2, m9
-    paddw           m6, m1
-    pmaddubsw       m2, m8
 
-%ifidn %1,pp
-    pmulhrsw        m0, m7                         ; m0 = word: row 12
-    pmulhrsw        m6, m7                         ; m6 = word: row 13
-    packuswb        m0, m6
-    vpermq          m0, m0, 11011000b
-    vextracti128    xm6, m0, 1
-    movu            [r8], xm0
-    movu            [r8 + r3], xm6
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+%ifidn %3,ps
+    add             r3d, r3d
+    vbroadcasti128  m14, [pw_2000]
 %else
-    psubw           m0, m7                         ; m0 = word: row 12
-    psubw           m6, m7                         ; m6 = word: row 13
-    movu            [r8], m0
-    movu            [r8 + r3], m6
+    mova            m14, [pw_512]
 %endif
+    lea             r6, [r3 * 3]
+    lea             r7, [r1 * 4]
+    mov             r8d, %2 / 16
 
-    movu            xm5, [r7 + r1 * 2]
-    vinserti128     m5, m5, [r7 + r1], 1
-    vextracti128    xm0, m5, 1
-    vinserti128     m3, m3, xm0, 0
-
-    punpcklbw       m1, m3, m5
-    punpckhbw       m0, m3, m5
-    vperm2i128      m6, m1, m0, 0x20
-    vperm2i128      m0, m1, m0, 0x31
-    pmaddubsw       m6, m9
-    paddw           m2, m6
-    pmaddubsw       m0, m9
-    paddw           m4, m0
-
-%ifidn %1,pp
-    pmulhrsw        m4, m7                         ; m4 = word: row 14
-    pmulhrsw        m2, m7                         ; m2 = word: row 15
-    packuswb        m4, m2
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm2, m4, 1
-    movu            [r8 + r3 * 2], xm4
-    movu            [r8 + r6], xm2
-    add             r2, 16
-%else
-    psubw           m4, m7                         ; m4 = word: row 14
-    psubw           m2, m7                         ; m2 = word: row 15
-    movu            [r8 + r3 * 2], m4
-    movu            [r8 + r6], m2
-    add             r2, 32
-%endif
-    add             r0, 16
-    movq            xm1, [r0]                       ; m1 = row 0
-    movq            xm2, [r0 + r1]                  ; m2 = row 1
-    punpcklbw       xm1, xm2
-    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
-    punpcklbw       xm2, xm3
-    vinserti128     m5, m1, xm2, 1
-    pmaddubsw       m5, m8
-    movq            xm4, [r0 + r4]                  ; m4 = row 3
-    punpcklbw       xm3, xm4
-    lea             r7, [r0 + r1 * 4]
-    movq            xm1, [r7]                       ; m1 = row 4
-    punpcklbw       xm4, xm1
-    vinserti128     m2, m3, xm4, 1
-    pmaddubsw       m0, m2, m9
-    paddw           m5, m0
-    pmaddubsw       m2, m8
-    movq            xm3, [r7 + r1]                  ; m3 = row 5
-    punpcklbw       xm1, xm3
-    movq            xm4, [r7 + r1 * 2]              ; m4 = row 6
-    punpcklbw       xm3, xm4
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m0, m1, m9
-    paddw           m2, m0
-    pmaddubsw       m1, m8
-    movq            xm3, [r7 + r4]                  ; m3 = row 7
-    punpcklbw       xm4, xm3
-    lea             r7, [r7 + r1 * 4]
-    movq            xm0, [r7]                       ; m0 = row 8
-    punpcklbw       xm3, xm0
-    vinserti128     m4, m4, xm3, 1
-    pmaddubsw       m3, m4, m9
-    paddw           m1, m3
-    pmaddubsw       m4, m8
-    movq            xm3, [r7 + r1]                  ; m3 = row 9
-    punpcklbw       xm0, xm3
-    movq            xm6, [r7 + r1 * 2]              ; m6 = row 10
-    punpcklbw       xm3, xm6
-    vinserti128     m0, m0, xm3, 1
-    pmaddubsw       m3, m0, m9
-    paddw           m4, m3
-    pmaddubsw       m0, m8
-
-%ifidn %1,pp
-    pmulhrsw        m5, m7                          ; m5 = word: row 0, row 1
-    pmulhrsw        m2, m7                          ; m2 = word: row 2, row 3
-    pmulhrsw        m1, m7                          ; m1 = word: row 4, row 5
-    pmulhrsw        m4, m7                          ; m4 = word: row 6, row 7
-    packuswb        m5, m2
-    packuswb        m1, m4
-    vextracti128    xm2, m5, 1
-    vextracti128    xm4, m1, 1
-    movq            [r2], xm5
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm5
-    movhps          [r2 + r6], xm2
-    lea             r8, [r2 + r3 * 4]
-    movq            [r8], xm1
-    movq            [r8 + r3], xm4
-    movhps          [r8 + r3 * 2], xm1
-    movhps          [r8 + r6], xm4
-%else
-    psubw           m5, m7                          ; m5 = word: row 0, row 1
-    psubw           m2, m7                          ; m2 = word: row 2, row 3
-    psubw           m1, m7                          ; m1 = word: row 4, row 5
-    psubw           m4, m7                          ; m4 = word: row 6, row 7
-    vextracti128    xm3, m5, 1
-    movu            [r2], xm5
-    movu            [r2 + r3], xm3
-    vextracti128    xm3, m2, 1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-    vextracti128    xm3, m1, 1
-    lea             r8, [r2 + r3 * 4]
-    movu            [r8], xm1
-    movu            [r8 + r3], xm3
-    vextracti128    xm3, m4, 1
-    movu            [r8 + r3 * 2], xm4
-    movu            [r8 + r6], xm3
-%endif
-    lea             r8, [r8 + r3 * 4]
-
-    movq            xm3, [r7 + r4]                  ; m3 = row 11
-    punpcklbw       xm6, xm3
-    lea             r7, [r7 + r1 * 4]
-    movq            xm5, [r7]                       ; m5 = row 12
-    punpcklbw       xm3, xm5
-    vinserti128     m6, m6, xm3, 1
-    pmaddubsw       m3, m6, m9
-    paddw           m0, m3
-    pmaddubsw       m6, m8
-    movq            xm3, [r7 + r1]                  ; m3 = row 13
-    punpcklbw       xm5, xm3
-    movq            xm2, [r7 + r1 * 2]              ; m2 = row 14
-    punpcklbw       xm3, xm2
-    vinserti128     m5, m5, xm3, 1
-    pmaddubsw       m3, m5, m9
-    paddw           m6, m3
-    pmaddubsw       m5, m8
-    movq            xm3, [r7 + r4]                  ; m3 = row 15
-    punpcklbw       xm2, xm3
-    lea             r7, [r7 + r1 * 4]
-    movq            xm1, [r7]                       ; m1 = row 16
-    punpcklbw       xm3, xm1
-    vinserti128     m2, m2, xm3, 1
-    pmaddubsw       m3, m2, m9
-    paddw           m5, m3
-    pmaddubsw       m2, m8
-    movq            xm3, [r7 + r1]                  ; m3 = row 17
-    punpcklbw       xm1, xm3
-    movq            xm4, [r7 + r1 * 2]              ; m4 = row 18
-    punpcklbw       xm3, xm4
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m3, m1, m9
-    paddw           m2, m3
-%ifidn %1,pp
-    pmulhrsw        m0, m7                          ; m0 = word: row 8, row 9
-    pmulhrsw        m6, m7                          ; m6 = word: row 10, row 11
-    pmulhrsw        m5, m7                          ; m5 = word: row 12, row 13
-    pmulhrsw        m2, m7                          ; m2 = word: row 14, row 15
-    packuswb        m0, m6
-    packuswb        m5, m2
-    vextracti128    xm6, m0, 1
-    vextracti128    xm2, m5, 1
-    movq            [r8], xm0
-    movq            [r8 + r3], xm6
-    movhps          [r8 + r3 * 2], xm0
-    movhps          [r8 + r6], xm6
-    lea             r8, [r8 + r3 * 4]
-    movq            [r8], xm5
-    movq            [r8 + r3], xm2
-    movhps          [r8 + r3 * 2], xm5
-    movhps          [r8 + r6], xm2
-    lea             r2, [r8 + r3 * 4 - 16]
-%else
-    psubw           m0, m7                          ; m0 = word: row 8, row 9
-    psubw           m6, m7                          ; m6 = word: row 10, row 11
-    psubw           m5, m7                          ; m5 = word: row 12, row 13
-    psubw           m2, m7                          ; m2 = word: row 14, row 15
-    vextracti128    xm3, m0, 1
-    movu            [r8], xm0
-    movu            [r8 + r3], xm3
-    vextracti128    xm3, m6, 1
-    movu            [r8 + r3 * 2], xm6
-    movu            [r8 + r6], xm3
-    vextracti128    xm3, m5, 1
-    lea             r8, [r8 + r3 * 4]
-    movu            [r8], xm5
-    movu            [r8 + r3], xm3
-    vextracti128    xm3, m2, 1
-    movu            [r8 + r3 * 2], xm2
-    movu            [r8 + r6], xm3
-    lea             r2, [r8 + r3 * 4 - 32]
-%endif
-    lea             r0, [r7 - 16]
-    dec             r5d
-    jnz             .loopH
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_CHROMA_AVX2_24x32 pp
-    FILTER_VER_CHROMA_AVX2_24x32 ps
-
-%macro FILTER_VER_CHROMA_AVX2_24x64 1
-%if ARCH_X86_64 == 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_24x64, 4, 7, 13
-    mov             r4d, r4m
-    shl             r4d, 6
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
-
-    mova            m10, [r5]
-    mova            m11, [r5 + mmsize]
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,pp
-    mova            m12, [pw_512]
-%else
-    add             r3d, r3d
-    vbroadcasti128  m12, [pw_2000]
-%endif
-    lea             r5, [r3 * 3]
-    mov             r6d, 16
-.loopH:
-    movu            m0, [r0]                        ; m0 = row 0
-    movu            m1, [r0 + r1]                   ; m1 = row 1
-    punpcklbw       m2, m0, m1
-    punpckhbw       m3, m0, m1
-    pmaddubsw       m2, m10
-    pmaddubsw       m3, m10
-    movu            m0, [r0 + r1 * 2]               ; m0 = row 2
-    punpcklbw       m4, m1, m0
-    punpckhbw       m5, m1, m0
-    pmaddubsw       m4, m10
-    pmaddubsw       m5, m10
-    movu            m1, [r0 + r4]                   ; m1 = row 3
-    punpcklbw       m6, m0, m1
-    punpckhbw       m7, m0, m1
-    pmaddubsw       m8, m6, m11
-    pmaddubsw       m9, m7, m11
-    pmaddubsw       m6, m10
-    pmaddubsw       m7, m10
-    paddw           m2, m8
-    paddw           m3, m9
-%ifidn %1,pp
-    pmulhrsw        m2, m12
-    pmulhrsw        m3, m12
-    packuswb        m2, m3
-    movu            [r2], xm2
-    vextracti128    xm2, m2, 1
-    movq            [r2 + 16], xm2
-%else
-    psubw           m2, m12
-    psubw           m3, m12
-    vperm2i128      m0, m2, m3, 0x20
-    vperm2i128      m2, m2, m3, 0x31
-    movu            [r2], m0
-    movu            [r2 + mmsize], xm2
-%endif
-    lea             r0, [r0 + r1 * 4]
-    movu            m0, [r0]                        ; m0 = row 4
-    punpcklbw       m2, m1, m0
-    punpckhbw       m3, m1, m0
-    pmaddubsw       m8, m2, m11
-    pmaddubsw       m9, m3, m11
-    pmaddubsw       m2, m10
-    pmaddubsw       m3, m10
-    paddw           m4, m8
-    paddw           m5, m9
-%ifidn %1,pp
-    pmulhrsw        m4, m12
-    pmulhrsw        m5, m12
-    packuswb        m4, m5
-    movu            [r2 + r3], xm4
-    vextracti128    xm4, m4, 1
-    movq            [r2 + r3 + 16], xm4
-%else
-    psubw           m4, m12
-    psubw           m5, m12
-    vperm2i128      m1, m4, m5, 0x20
-    vperm2i128      m4, m4, m5, 0x31
-    movu            [r2 + r3], m1
-    movu            [r2 + r3 + mmsize], xm4
-%endif
-
-    movu            m1, [r0 + r1]                   ; m1 = row 5
-    punpcklbw       m4, m0, m1
-    punpckhbw       m5, m0, m1
-    pmaddubsw       m4, m11
-    pmaddubsw       m5, m11
-    paddw           m6, m4
-    paddw           m7, m5
-%ifidn %1,pp
-    pmulhrsw        m6, m12
-    pmulhrsw        m7, m12
-    packuswb        m6, m7
-    movu            [r2 + r3 * 2], xm6
-    vextracti128    xm6, m6, 1
-    movq            [r2 + r3 * 2 + 16], xm6
-%else
-    psubw           m6, m12
-    psubw           m7, m12
-    vperm2i128      m0, m6, m7, 0x20
-    vperm2i128      m6, m6, m7, 0x31
-    movu            [r2 + r3 * 2], m0
-    movu            [r2 + r3 * 2 + mmsize], xm6
-%endif
-
-    movu            m0, [r0 + r1 * 2]               ; m0 = row 6
-    punpcklbw       m6, m1, m0
-    punpckhbw       m7, m1, m0
-    pmaddubsw       m6, m11
-    pmaddubsw       m7, m11
-    paddw           m2, m6
-    paddw           m3, m7
-%ifidn %1,pp
-    pmulhrsw        m2, m12
-    pmulhrsw        m3, m12
-    packuswb        m2, m3
-    movu            [r2 + r5], xm2
-    vextracti128    xm2, m2, 1
-    movq            [r2 + r5 + 16], xm2
-%else
-    psubw           m2, m12
-    psubw           m3, m12
-    vperm2i128      m0, m2, m3, 0x20
-    vperm2i128      m2, m2, m3, 0x31
-    movu            [r2 + r5], m0
-    movu            [r2 + r5 + mmsize], xm2
-%endif
-    lea             r2, [r2 + r3 * 4]
-    dec             r6d
-    jnz             .loopH
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_CHROMA_AVX2_24x64 pp
-    FILTER_VER_CHROMA_AVX2_24x64 ps
-
-%macro FILTER_VER_CHROMA_AVX2_16x4 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_16x4, 4, 6, 8
-    mov             r4d, r4m
-    shl             r4d, 6
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,pp
-    mova            m7, [pw_512]
-%else
-    add             r3d, r3d
-    mova            m7, [pw_2000]
-%endif
-
-    movu            xm0, [r0]
-    vinserti128     m0, m0, [r0 + r1 * 2], 1
-    movu            xm1, [r0 + r1]
-    vinserti128     m1, m1, [r0 + r4], 1
-
-    punpcklbw       m2, m0, m1
-    punpckhbw       m3, m0, m1
-    vperm2i128      m4, m2, m3, 0x20
-    vperm2i128      m2, m2, m3, 0x31
-    pmaddubsw       m4, [r5]
-    pmaddubsw       m3, m2, [r5 + mmsize]
-    paddw           m4, m3
-    pmaddubsw       m2, [r5]
-
-    vextracti128    xm0, m0, 1
-    lea             r0, [r0 + r1 * 4]
-    vinserti128     m0, m0, [r0], 1
-
-    punpcklbw       m5, m1, m0
-    punpckhbw       m3, m1, m0
-    vperm2i128      m6, m5, m3, 0x20
-    vperm2i128      m5, m5, m3, 0x31
-    pmaddubsw       m6, [r5]
-    pmaddubsw       m3, m5, [r5 + mmsize]
-    paddw           m6, m3
-    pmaddubsw       m5, [r5]
-%ifidn %1,pp
-    pmulhrsw        m4, m7                          ; m4 = word: row 0
-    pmulhrsw        m6, m7                          ; m6 = word: row 1
-    packuswb        m4, m6
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm6, m4, 1
-    movu            [r2], xm4
-    movu            [r2 + r3], xm6
-%else
-    psubw           m4, m7                          ; m4 = word: row 0
-    psubw           m6, m7                          ; m6 = word: row 1
-    movu            [r2], m4
-    movu            [r2 + r3], m6
-%endif
-    lea             r2, [r2 + r3 * 2]
-
-    movu            xm4, [r0 + r1 * 2]
-    vinserti128     m4, m4, [r0 + r1], 1
-    vextracti128    xm1, m4, 1
-    vinserti128     m0, m0, xm1, 0
-
-    punpcklbw       m6, m0, m4
-    punpckhbw       m1, m0, m4
-    vperm2i128      m0, m6, m1, 0x20
-    vperm2i128      m6, m6, m1, 0x31
-    pmaddubsw       m0, [r5 + mmsize]
-    paddw           m5, m0
-    pmaddubsw       m6, [r5 + mmsize]
-    paddw           m2, m6
-
-%ifidn %1,pp
-    pmulhrsw        m2, m7                          ; m2 = word: row 2
-    pmulhrsw        m5, m7                          ; m5 = word: row 3
-    packuswb        m2, m5
-    vpermq          m2, m2, 11011000b
-    vextracti128    xm5, m2, 1
-    movu            [r2], xm2
-    movu            [r2 + r3], xm5
-%else
-    psubw           m2, m7                          ; m2 = word: row 2
-    psubw           m5, m7                          ; m5 = word: row 3
-    movu            [r2], m2
-    movu            [r2 + r3], m5
-%endif
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_AVX2_16x4 pp
-    FILTER_VER_CHROMA_AVX2_16x4 ps
-
-%macro FILTER_VER_CHROMA_AVX2_12xN 2
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_12x%2, 4, 7, 8
-    mov             r4d, r4m
-    shl             r4d, 6
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,pp
-    mova            m7, [pw_512]
-%else
-    add             r3d, r3d
-    vbroadcasti128  m7, [pw_2000]
-%endif
-    lea             r6, [r3 * 3]
-%rep %2 / 16
+.loop:
     movu            xm0, [r0]                       ; m0 = row 0
     movu            xm1, [r0 + r1]                  ; m1 = row 1
     punpckhbw       xm2, xm0, xm1
@@ -8218,11332 +5073,1723 @@ cglobal interp_4tap_vert_%1_12x%2, 4, 7, 8
     pmaddubsw       m5, m3, [r5 + 1 * mmsize]
     paddw           m1, m5
     pmaddubsw       m3, [r5]
-%ifidn %1,pp
-    pmulhrsw        m0, m7                          ; m0 = word: row 0
-    pmulhrsw        m1, m7                          ; m1 = word: row 1
-    packuswb        m0, m1
-    vextracti128    xm1, m0, 1
-    movq            [r2], xm0
-    movd            [r2 + 8], xm1
-    movhps          [r2 + r3], xm0
-    pextrd          [r2 + r3 + 8], xm1, 2
-%else
-    psubw           m0, m7                          ; m0 = word: row 0
-    psubw           m1, m7                          ; m1 = word: row 1
-    movu            [r2], xm0
-    vextracti128    xm0, m0, 1
-    movq            [r2 + 16], xm0
-    movu            [r2 + r3], xm1
-    vextracti128    xm1, m1, 1
-    movq            [r2 + r3 + 16], xm1
-%endif
-
     movu            xm5, [r0 + r1]                  ; m5 = row 5
     punpckhbw       xm6, xm4, xm5
     punpcklbw       xm4, xm5
     vinserti128     m4, m4, xm6, 1
+    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
+    paddw           m0, m6
     pmaddubsw       m6, m4, [r5 + 1 * mmsize]
     paddw           m2, m6
     pmaddubsw       m4, [r5]
     movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhbw       xm0, xm5, xm6
+    punpckhbw       xm7, xm5, xm6
     punpcklbw       xm5, xm6
-    vinserti128     m5, m5, xm0, 1
-    pmaddubsw       m0, m5, [r5 + 1 * mmsize]
-    paddw           m3, m0
+    vinserti128     m5, m5, xm7, 1
+    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
+    paddw           m1, m7
+    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
+    paddw           m3, m7
     pmaddubsw       m5, [r5]
-%ifidn %1,pp
-    pmulhrsw        m2, m7                          ; m2 = word: row 2
-    pmulhrsw        m3, m7                          ; m3 = word: row 3
+    movu            xm7, [r0 + r4]                  ; m7 = row 7
+    punpckhbw       xm8, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
+    paddw           m0, m8
+    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
+    paddw           m2, m8
+    pmaddubsw       m8, m6, [r5 + 1 * mmsize]
+    paddw           m4, m8
+    pmaddubsw       m6, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm8, [r0]                       ; m8 = row 8
+    punpckhbw       xm9, xm7, xm8
+    punpcklbw       xm7, xm8
+    vinserti128     m7, m7, xm9, 1
+    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
+    paddw           m1, m9
+    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
+    paddw           m3, m9
+    pmaddubsw       m9, m7, [r5 + 1 * mmsize]
+    paddw           m5, m9
+    pmaddubsw       m7, [r5]
+    movu            xm9, [r0 + r1]                  ; m9 = row 9
+    punpckhbw       xm10, xm8, xm9
+    punpcklbw       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
+    paddw           m2, m10
+    pmaddubsw       m10, m8, [r5 + 2 * mmsize]
+    paddw           m4, m10
+    pmaddubsw       m10, m8, [r5 + 1 * mmsize]
+    paddw           m6, m10
+    pmaddubsw       m8, [r5]
+    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
+    punpckhbw       xm11, xm9, xm10
+    punpcklbw       xm9, xm10
+    vinserti128     m9, m9, xm11, 1
+    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
+    paddw           m3, m11
+    pmaddubsw       m11, m9, [r5 + 2 * mmsize]
+    paddw           m5, m11
+    pmaddubsw       m11, m9, [r5 + 1 * mmsize]
+    paddw           m7, m11
+    pmaddubsw       m9, [r5]
+    movu            xm11, [r0 + r4]                 ; m11 = row 11
+    punpckhbw       xm12, xm10, xm11
+    punpcklbw       xm10, xm11
+    vinserti128     m10, m10, xm12, 1
+    pmaddubsw       m12, m10, [r5 + 3 * mmsize]
+    paddw           m4, m12
+    pmaddubsw       m12, m10, [r5 + 2 * mmsize]
+    paddw           m6, m12
+    pmaddubsw       m12, m10, [r5 + 1 * mmsize]
+    paddw           m8, m12
+    pmaddubsw       m10, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm12, [r0]                      ; m12 = row 12
+    punpckhbw       xm13, xm11, xm12
+    punpcklbw       xm11, xm12
+    vinserti128     m11, m11, xm13, 1
+    pmaddubsw       m13, m11, [r5 + 3 * mmsize]
+    paddw           m5, m13
+    pmaddubsw       m13, m11, [r5 + 2 * mmsize]
+    paddw           m7, m13
+    pmaddubsw       m13, m11, [r5 + 1 * mmsize]
+    paddw           m9, m13
+    pmaddubsw       m11, [r5]
+
+%ifidn %3,pp
+    pmulhrsw        m0, m14                         ; m0 = word: row 0
+    pmulhrsw        m1, m14                         ; m1 = word: row 1
+    pmulhrsw        m2, m14                         ; m2 = word: row 2
+    pmulhrsw        m3, m14                         ; m3 = word: row 3
+    pmulhrsw        m4, m14                         ; m4 = word: row 4
+    pmulhrsw        m5, m14                         ; m5 = word: row 5
+    packuswb        m0, m1
     packuswb        m2, m3
+    packuswb        m4, m5
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    vpermq          m4, m4, 11011000b
+    vextracti128    xm1, m0, 1
     vextracti128    xm3, m2, 1
-    movq            [r2 + r3 * 2], xm2
-    movd            [r2 + r3 * 2 + 8], xm3
-    movhps          [r2 + r6], xm2
-    pextrd          [r2 + r6 + 8], xm3, 2
-%else
-    psubw           m2, m7                          ; m2 = word: row 2
-    psubw           m3, m7                          ; m3 = word: row 3
+    vextracti128    xm5, m4, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
     movu            [r2 + r3 * 2], xm2
-    vextracti128    xm2, m2, 1
-    movq            [r2 + r3 * 2 + 16], xm2
     movu            [r2 + r6], xm3
-    vextracti128    xm3, m3, 1
-    movq            [r2 + r6 + 16], xm3
-%endif
     lea             r2, [r2 + r3 * 4]
-
-    movu            xm0, [r0 + r4]                  ; m0 = row 7
-    punpckhbw       xm3, xm6, xm0
-    punpcklbw       xm6, xm0
-    vinserti128     m6, m6, xm3, 1
-    pmaddubsw       m3, m6, [r5 + 1 * mmsize]
-    paddw           m4, m3
-    pmaddubsw       m6, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm3, [r0]                       ; m3 = row 8
-    punpckhbw       xm1, xm0, xm3
-    punpcklbw       xm0, xm3
-    vinserti128     m0, m0, xm1, 1
-    pmaddubsw       m1, m0, [r5 + 1 * mmsize]
-    paddw           m5, m1
-    pmaddubsw       m0, [r5]
-%ifidn %1,pp
-    pmulhrsw        m4, m7                          ; m4 = word: row 4
-    pmulhrsw        m5, m7                          ; m5 = word: row 5
-    packuswb        m4, m5
-    vextracti128    xm5, m4, 1
-    movq            [r2], xm4
-    movd            [r2 + 8], xm5
-    movhps          [r2 + r3], xm4
-    pextrd          [r2 + r3 + 8], xm5, 2
-%else
-    psubw           m4, m7                          ; m4 = word: row 4
-    psubw           m5, m7                          ; m5 = word: row 5
     movu            [r2], xm4
-    vextracti128    xm4, m4, 1
-    movq            [r2 + 16], xm4
     movu            [r2 + r3], xm5
-    vextracti128    xm5, m5, 1
-    movq            [r2 + r3 + 16], xm5
+%else
+    psubw           m0, m14                         ; m0 = word: row 0
+    psubw           m1, m14                         ; m1 = word: row 1
+    psubw           m2, m14                         ; m2 = word: row 2
+    psubw           m3, m14                         ; m3 = word: row 3
+    psubw           m4, m14                         ; m4 = word: row 4
+    psubw           m5, m14                         ; m5 = word: row 5
+    movu            [r2], m0
+    movu            [r2 + r3], m1
+    movu            [r2 + r3 * 2], m2
+    movu            [r2 + r6], m3
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], m4
+    movu            [r2 + r3], m5
 %endif
 
-    movu            xm1, [r0 + r1]                  ; m1 = row 9
-    punpckhbw       xm2, xm3, xm1
-    punpcklbw       xm3, xm1
-    vinserti128     m3, m3, xm2, 1
-    pmaddubsw       m2, m3, [r5 + 1 * mmsize]
-    paddw           m6, m2
-    pmaddubsw       m3, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 10
-    punpckhbw       xm4, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm4, 1
-    pmaddubsw       m4, m1, [r5 + 1 * mmsize]
-    paddw           m0, m4
-    pmaddubsw       m1, [r5]
+    movu            xm13, [r0 + r1]                 ; m13 = row 13
+    punpckhbw       xm0, xm12, xm13
+    punpcklbw       xm12, xm13
+    vinserti128     m12, m12, xm0, 1
+    pmaddubsw       m0, m12, [r5 + 3 * mmsize]
+    paddw           m6, m0
+    pmaddubsw       m0, m12, [r5 + 2 * mmsize]
+    paddw           m8, m0
+    pmaddubsw       m0, m12, [r5 + 1 * mmsize]
+    paddw           m10, m0
+    pmaddubsw       m12, [r5]
+    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
+    punpckhbw       xm1, xm13, xm0
+    punpcklbw       xm13, xm0
+    vinserti128     m13, m13, xm1, 1
+    pmaddubsw       m1, m13, [r5 + 3 * mmsize]
+    paddw           m7, m1
+    pmaddubsw       m1, m13, [r5 + 2 * mmsize]
+    paddw           m9, m1
+    pmaddubsw       m1, m13, [r5 + 1 * mmsize]
+    paddw           m11, m1
+    pmaddubsw       m13, [r5]
 
-%ifidn %1,pp
-    pmulhrsw        m6, m7                          ; m6 = word: row 6
-    pmulhrsw        m0, m7                          ; m0 = word: row 7
-    packuswb        m6, m0
-    vextracti128    xm0, m6, 1
-    movq            [r2 + r3 * 2], xm6
-    movd            [r2 + r3 * 2 + 8], xm0
-    movhps          [r2 + r6], xm6
-    pextrd          [r2 + r6 + 8], xm0, 2
-%else
-    psubw           m6, m7                          ; m6 = word: row 6
-    psubw           m0, m7                          ; m0 = word: row 7
+%ifidn %3,pp
+    pmulhrsw        m6, m14                         ; m6 = word: row 6
+    pmulhrsw        m7, m14                         ; m7 = word: row 7
+    packuswb        m6, m7
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm7, m6, 1
     movu            [r2 + r3 * 2], xm6
-    vextracti128    xm6, m6, 1
-    movq            [r2 + r3 * 2 + 16], xm6
-    movu            [r2 + r6], xm0
-    vextracti128    xm0, m0, 1
-    movq            [r2 + r6 + 16], xm0
-%endif
-    lea             r2, [r2 + r3 * 4]
-
-    movu            xm4, [r0 + r4]                  ; m4 = row 11
-    punpckhbw       xm6, xm2, xm4
-    punpcklbw       xm2, xm4
-    vinserti128     m2, m2, xm6, 1
-    pmaddubsw       m6, m2, [r5 + 1 * mmsize]
-    paddw           m3, m6
-    pmaddubsw       m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm6, [r0]                       ; m6 = row 12
-    punpckhbw       xm0, xm4, xm6
-    punpcklbw       xm4, xm6
-    vinserti128     m4, m4, xm0, 1
-    pmaddubsw       m0, m4, [r5 + 1 * mmsize]
-    paddw           m1, m0
-    pmaddubsw       m4, [r5]
-%ifidn %1,pp
-    pmulhrsw        m3, m7                          ; m3 = word: row 8
-    pmulhrsw        m1, m7                          ; m1 = word: row 9
-    packuswb        m3, m1
-    vextracti128    xm1, m3, 1
-    movq            [r2], xm3
-    movd            [r2 + 8], xm1
-    movhps          [r2 + r3], xm3
-    pextrd          [r2 + r3 + 8], xm1, 2
+    movu            [r2 + r6], xm7
 %else
-    psubw           m3, m7                          ; m3 = word: row 8
-    psubw           m1, m7                          ; m1 = word: row 9
-    movu            [r2], xm3
-    vextracti128    xm3, m3, 1
-    movq            [r2 + 16], xm3
-    movu            [r2 + r3], xm1
-    vextracti128    xm1, m1, 1
-    movq            [r2 + r3 + 16], xm1
+    psubw           m6, m14                         ; m6 = word: row 6
+    psubw           m7, m14                         ; m7 = word: row 7
+    movu            [r2 + r3 * 2], m6
+    movu            [r2 + r6], m7
 %endif
 
-    movu            xm0, [r0 + r1]                  ; m0 = row 13
-    punpckhbw       xm1, xm6, xm0
-    punpcklbw       xm6, xm0
-    vinserti128     m6, m6, xm1, 1
-    pmaddubsw       m1, m6, [r5 + 1 * mmsize]
-    paddw           m2, m1
-    pmaddubsw       m6, [r5]
-    movu            xm1, [r0 + r1 * 2]              ; m1 = row 14
-    punpckhbw       xm5, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm5, 1
-    pmaddubsw       m5, m0, [r5 + 1 * mmsize]
-    paddw           m4, m5
-    pmaddubsw       m0, [r5]
-%ifidn %1,pp
-    pmulhrsw        m2, m7                          ; m2 = word: row 10
-    pmulhrsw        m4, m7                          ; m4 = word: row 11
-    packuswb        m2, m4
-    vextracti128    xm4, m2, 1
-    movq            [r2 + r3 * 2], xm2
-    movd            [r2 + r3 * 2 + 8], xm4
-    movhps          [r2 + r6], xm2
-    pextrd          [r2 + r6 + 8], xm4, 2
-%else
-    psubw           m2, m7                          ; m2 = word: row 10
-    psubw           m4, m7                          ; m4 = word: row 11
-    movu            [r2 + r3 * 2], xm2
-    vextracti128    xm2, m2, 1
-    movq            [r2 + r3 * 2 + 16], xm2
-    movu            [r2 + r6], xm4
-    vextracti128    xm4, m4, 1
-    movq            [r2 + r6 + 16], xm4
-%endif
     lea             r2, [r2 + r3 * 4]
 
-    movu            xm5, [r0 + r4]                  ; m5 = row 15
-    punpckhbw       xm2, xm1, xm5
-    punpcklbw       xm1, xm5
-    vinserti128     m1, m1, xm2, 1
-    pmaddubsw       m2, m1, [r5 + 1 * mmsize]
-    paddw           m6, m2
-    pmaddubsw       m1, [r5]
+    movu            xm1, [r0 + r4]                  ; m1 = row 15
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddubsw       m2, m0, [r5 + 3 * mmsize]
+    paddw           m8, m2
+    pmaddubsw       m2, m0, [r5 + 2 * mmsize]
+    paddw           m10, m2
+    pmaddubsw       m2, m0, [r5 + 1 * mmsize]
+    paddw           m12, m2
+    pmaddubsw       m0, [r5]
     lea             r0, [r0 + r1 * 4]
     movu            xm2, [r0]                       ; m2 = row 16
-    punpckhbw       xm3, xm5, xm2
-    punpcklbw       xm5, xm2
-    vinserti128     m5, m5, xm3, 1
-    pmaddubsw       m3, m5, [r5 + 1 * mmsize]
-    paddw           m0, m3
-    pmaddubsw       m5, [r5]
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m3, m1, [r5 + 3 * mmsize]
+    paddw           m9, m3
+    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
+    paddw           m11, m3
+    pmaddubsw       m3, m1, [r5 + 1 * mmsize]
+    paddw           m13, m3
+    pmaddubsw       m1, [r5]
     movu            xm3, [r0 + r1]                  ; m3 = row 17
     punpckhbw       xm4, xm2, xm3
     punpcklbw       xm2, xm3
     vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m4, m2, [r5 + 3 * mmsize]
+    paddw           m10, m4
+    pmaddubsw       m4, m2, [r5 + 2 * mmsize]
+    paddw           m12, m4
     pmaddubsw       m2, [r5 + 1 * mmsize]
-    paddw           m1, m2
+    paddw           m0, m2
     movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
-    punpckhbw       xm2, xm3, xm4
+    punpckhbw       xm5, xm3, xm4
     punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm2, 1
+    vinserti128     m3, m3, xm5, 1
+    pmaddubsw       m5, m3, [r5 + 3 * mmsize]
+    paddw           m11, m5
+    pmaddubsw       m5, m3, [r5 + 2 * mmsize]
+    paddw           m13, m5
     pmaddubsw       m3, [r5 + 1 * mmsize]
-    paddw           m5, m3
-
-%ifidn %1,pp
-    pmulhrsw        m6, m7                          ; m6 = word: row 12
-    pmulhrsw        m0, m7                          ; m0 = word: row 13
-    pmulhrsw        m1, m7                          ; m1 = word: row 14
-    pmulhrsw        m5, m7                          ; m5 = word: row 15
-    packuswb        m6, m0
-    packuswb        m1, m5
-    vextracti128    xm0, m6, 1
-    vextracti128    xm5, m1, 1
-    movq            [r2], xm6
-    movd            [r2 + 8], xm0
-    movhps          [r2 + r3], xm6
-    pextrd          [r2 + r3 + 8], xm0, 2
-    movq            [r2 + r3 * 2], xm1
-    movd            [r2 + r3 * 2 + 8], xm5
-    movhps          [r2 + r6], xm1
-    pextrd          [r2 + r6 + 8], xm5, 2
-%else
-    psubw           m6, m7                          ; m6 = word: row 12
-    psubw           m0, m7                          ; m0 = word: row 13
-    psubw           m1, m7                          ; m1 = word: row 14
-    psubw           m5, m7                          ; m5 = word: row 15
-    movu            [r2], xm6
-    vextracti128    xm6, m6, 1
-    movq            [r2 + 16], xm6
-    movu            [r2 + r3], xm0
-    vextracti128    xm0, m0, 1
-    movq            [r2 + r3 + 16], xm0
-    movu            [r2 + r3 * 2], xm1
-    vextracti128    xm1, m1, 1
-    movq            [r2 + r3 * 2 + 16], xm1
-    movu            [r2 + r6], xm5
-    vextracti128    xm5, m5, 1
-    movq            [r2 + r6 + 16], xm5
-%endif
-    lea             r2, [r2 + r3 * 4]
-%endrep
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_AVX2_12xN pp, 16
-    FILTER_VER_CHROMA_AVX2_12xN ps, 16
-    FILTER_VER_CHROMA_AVX2_12xN pp, 32
-    FILTER_VER_CHROMA_AVX2_12xN ps, 32
-
-;-----------------------------------------------------------------------------
-;void interp_4tap_vert_pp_24x32(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro FILTER_V4_W24 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_pp_24x%2, 4, 6, 8
-
-    mov         r4d,       r4m
-    sub         r0,        r1
-
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeff]
-    movd        m0,        [r5 + r4 * 4]
-%else
-    movd        m0,        [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    pshufb      m1,        m0,       [tab_Vm]
-    pshufb      m0,        [tab_Vm + 16]
-
-    mov         r4d,       %2
-
-.loop:
-    movu        m2,        [r0]
-    movu        m3,        [r0 + r1]
-
-    punpcklbw   m4,        m2,        m3
-    punpckhbw   m2,        m3
-
-    pmaddubsw   m4,        m1
-    pmaddubsw   m2,        m1
-
-    lea         r5,        [r0 + 2 * r1]
-    movu        m5,        [r5]
-    movu        m7,        [r5 + r1]
-
-    punpcklbw   m6,        m5,        m7
-    pmaddubsw   m6,        m0
-    paddw       m4,        m6
-
-    punpckhbw   m6,        m5,        m7
-    pmaddubsw   m6,        m0
-    paddw       m2,        m6
-
-    mova        m6,        [pw_512]
-
-    pmulhrsw    m4,        m6
-    pmulhrsw    m2,        m6
-
-    packuswb    m4,        m2
-
-    movu        [r2],      m4
-
-    punpcklbw   m4,        m3,        m5
-    punpckhbw   m3,        m5
-
-    pmaddubsw   m4,        m1
-    pmaddubsw   m3,        m1
-
-    movu        m2,        [r5 + 2 * r1]
-
-    punpcklbw   m5,        m7,        m2
-    punpckhbw   m7,        m2
-
-    pmaddubsw   m5,        m0
-    pmaddubsw   m7,        m0
-
-    paddw       m4,        m5
-    paddw       m3,        m7
-
-    pmulhrsw    m4,        m6
-    pmulhrsw    m3,        m6
-
-    packuswb    m4,        m3
-
-    movu        [r2 + r3],      m4
-
-    movq        m2,        [r0 + 16]
-    movq        m3,        [r0 + r1 + 16]
-    movq        m4,        [r5 + 16]
-    movq        m5,        [r5 + r1 + 16]
-
-    punpcklbw   m2,        m3
-    punpcklbw   m4,        m5
-
-    pmaddubsw   m2,        m1
-    pmaddubsw   m4,        m0
-
-    paddw       m2,        m4
-
-    pmulhrsw    m2,        m6
-
-    movq        m3,        [r0 + r1 + 16]
-    movq        m4,        [r5 + 16]
-    movq        m5,        [r5 + r1 + 16]
-    movq        m7,        [r5 + 2 * r1 + 16]
-
-    punpcklbw   m3,        m4
-    punpcklbw   m5,        m7
-
-    pmaddubsw   m3,        m1
-    pmaddubsw   m5,        m0
-
-    paddw       m3,        m5
-
-    pmulhrsw    m3,        m6
-    packuswb    m2,        m3
-
-    movh        [r2 + 16], m2
-    movhps      [r2 + r3 + 16], m2
-
-    mov         r0,        r5
-    lea         r2,        [r2 + 2 * r3]
-
-    sub         r4,        2
-    jnz        .loop
-    RET
-%endmacro
-
-    FILTER_V4_W24 24, 32
-
-    FILTER_V4_W24 24, 64
-
-;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_32x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro FILTER_V4_W32 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_pp_%1x%2, 4, 6, 8
-
-    mov         r4d,       r4m
-    sub         r0,        r1
-
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeff]
-    movd        m0,        [r5 + r4 * 4]
-%else
-    movd        m0,        [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    pshufb      m1,        m0,       [tab_Vm]
-    pshufb      m0,        [tab_Vm + 16]
-
-    mova        m7,        [pw_512]
-
-    mov         r4d,       %2
-
-.loop:
-    movu        m2,        [r0]
-    movu        m3,        [r0 + r1]
-
-    punpcklbw   m4,        m2,        m3
-    punpckhbw   m2,        m3
-
-    pmaddubsw   m4,        m1
-    pmaddubsw   m2,        m1
-
-    lea         r5,        [r0 + 2 * r1]
-    movu        m3,        [r5]
-    movu        m5,        [r5 + r1]
-
-    punpcklbw   m6,        m3,        m5
-    punpckhbw   m3,        m5
-
-    pmaddubsw   m6,        m0
-    pmaddubsw   m3,        m0
-
-    paddw       m4,        m6
-    paddw       m2,        m3
-
-    pmulhrsw    m4,        m7
-    pmulhrsw    m2,        m7
-
-    packuswb    m4,        m2
-
-    movu        [r2],      m4
-
-    movu        m2,        [r0 + 16]
-    movu        m3,        [r0 + r1 + 16]
-
-    punpcklbw   m4,        m2,        m3
-    punpckhbw   m2,        m3
-
-    pmaddubsw   m4,        m1
-    pmaddubsw   m2,        m1
-
-    movu        m3,        [r5 + 16]
-    movu        m5,        [r5 + r1 + 16]
-
-    punpcklbw   m6,        m3,        m5
-    punpckhbw   m3,        m5
-
-    pmaddubsw   m6,        m0
-    pmaddubsw   m3,        m0
-
-    paddw       m4,        m6
-    paddw       m2,        m3
-
-    pmulhrsw    m4,        m7
-    pmulhrsw    m2,        m7
-
-    packuswb    m4,        m2
-
-    movu        [r2 + 16], m4
-
-    lea         r0,        [r0 + r1]
-    lea         r2,        [r2 + r3]
-
-    dec         r4
-    jnz        .loop
-    RET
-%endmacro
-
-    FILTER_V4_W32 32,  8
-    FILTER_V4_W32 32, 16
-    FILTER_V4_W32 32, 24
-    FILTER_V4_W32 32, 32
-
-    FILTER_V4_W32 32, 48
-    FILTER_V4_W32 32, 64
-
-%macro FILTER_VER_CHROMA_AVX2_32xN 2
-%if ARCH_X86_64 == 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_32x%2, 4, 7, 13
-    mov             r4d, r4m
-    shl             r4d, 6
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
-
-    mova            m10, [r5]
-    mova            m11, [r5 + mmsize]
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,pp
-    mova            m12, [pw_512]
-%else
-    add             r3d, r3d
-    vbroadcasti128  m12, [pw_2000]
-%endif
-    lea             r5, [r3 * 3]
-    mov             r6d, %2 / 4
-.loopW:
-    movu            m0, [r0]                        ; m0 = row 0
-    movu            m1, [r0 + r1]                   ; m1 = row 1
-    punpcklbw       m2, m0, m1
-    punpckhbw       m3, m0, m1
-    pmaddubsw       m2, m10
-    pmaddubsw       m3, m10
-    movu            m0, [r0 + r1 * 2]               ; m0 = row 2
-    punpcklbw       m4, m1, m0
-    punpckhbw       m5, m1, m0
-    pmaddubsw       m4, m10
-    pmaddubsw       m5, m10
-    movu            m1, [r0 + r4]                   ; m1 = row 3
-    punpcklbw       m6, m0, m1
-    punpckhbw       m7, m0, m1
-    pmaddubsw       m8, m6, m11
-    pmaddubsw       m9, m7, m11
-    pmaddubsw       m6, m10
-    pmaddubsw       m7, m10
-    paddw           m2, m8
-    paddw           m3, m9
-%ifidn %1,pp
-    pmulhrsw        m2, m12
-    pmulhrsw        m3, m12
-    packuswb        m2, m3
-    movu            [r2], m2
-%else
-    psubw           m2, m12
-    psubw           m3, m12
-    vperm2i128      m0, m2, m3, 0x20
-    vperm2i128      m2, m2, m3, 0x31
-    movu            [r2], m0
-    movu            [r2 + mmsize], m2
-%endif
+    paddw           m1, m3
+    movu            xm5, [r0 + r4]                  ; m5 = row 19
+    punpckhbw       xm6, xm4, xm5
+    punpcklbw       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddubsw       m6, m4, [r5 + 3 * mmsize]
+    paddw           m12, m6
+    pmaddubsw       m4, [r5 + 2 * mmsize]
+    paddw           m0, m4
     lea             r0, [r0 + r1 * 4]
-    movu            m0, [r0]                        ; m0 = row 4
-    punpcklbw       m2, m1, m0
-    punpckhbw       m3, m1, m0
-    pmaddubsw       m8, m2, m11
-    pmaddubsw       m9, m3, m11
-    pmaddubsw       m2, m10
-    pmaddubsw       m3, m10
-    paddw           m4, m8
-    paddw           m5, m9
-%ifidn %1,pp
-    pmulhrsw        m4, m12
-    pmulhrsw        m5, m12
-    packuswb        m4, m5
-    movu            [r2 + r3], m4
-%else
-    psubw           m4, m12
-    psubw           m5, m12
-    vperm2i128      m1, m4, m5, 0x20
-    vperm2i128      m4, m4, m5, 0x31
-    movu            [r2 + r3], m1
-    movu            [r2 + r3 + mmsize], m4
-%endif
+    movu            xm6, [r0]                       ; m6 = row 20
+    punpckhbw       xm7, xm5, xm6
+    punpcklbw       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddubsw       m7, m5, [r5 + 3 * mmsize]
+    paddw           m13, m7
+    pmaddubsw       m5, [r5 + 2 * mmsize]
+    paddw           m1, m5
+    movu            xm7, [r0 + r1]                  ; m7 = row 21
+    punpckhbw       xm2, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm2, 1
+    pmaddubsw       m6, [r5 + 3 * mmsize]
+    paddw           m0, m6
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 22
+    punpckhbw       xm3, xm7, xm2
+    punpcklbw       xm7, xm2
+    vinserti128     m7, m7, xm3, 1
+    pmaddubsw       m7, [r5 + 3 * mmsize]
+    paddw           m1, m7
 
-    movu            m1, [r0 + r1]                   ; m1 = row 5
-    punpcklbw       m4, m0, m1
-    punpckhbw       m5, m0, m1
-    pmaddubsw       m4, m11
-    pmaddubsw       m5, m11
-    paddw           m6, m4
-    paddw           m7, m5
-%ifidn %1,pp
-    pmulhrsw        m6, m12
-    pmulhrsw        m7, m12
-    packuswb        m6, m7
-    movu            [r2 + r3 * 2], m6
+%ifidn %3,pp
+    pmulhrsw        m8, m14                         ; m8 = word: row 8
+    pmulhrsw        m9, m14                         ; m9 = word: row 9
+    pmulhrsw        m10, m14                        ; m10 = word: row 10
+    pmulhrsw        m11, m14                        ; m11 = word: row 11
+    pmulhrsw        m12, m14                        ; m12 = word: row 12
+    pmulhrsw        m13, m14                        ; m13 = word: row 13
+    pmulhrsw        m0, m14                         ; m0 = word: row 14
+    pmulhrsw        m1, m14                         ; m1 = word: row 15
+    packuswb        m8, m9
+    packuswb        m10, m11
+    packuswb        m12, m13
+    packuswb        m0, m1
+    vpermq          m8, m8, 11011000b
+    vpermq          m10, m10, 11011000b
+    vpermq          m12, m12, 11011000b
+    vpermq          m0, m0, 11011000b
+    vextracti128    xm9, m8, 1
+    vextracti128    xm11, m10, 1
+    vextracti128    xm13, m12, 1
+    vextracti128    xm1, m0, 1
+    movu            [r2], xm8
+    movu            [r2 + r3], xm9
+    movu            [r2 + r3 * 2], xm10
+    movu            [r2 + r6], xm11
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm12
+    movu            [r2 + r3], xm13
+    movu            [r2 + r3 * 2], xm0
+    movu            [r2 + r6], xm1
 %else
-    psubw           m6, m12
-    psubw           m7, m12
-    vperm2i128      m0, m6, m7, 0x20
-    vperm2i128      m6, m6, m7, 0x31
+    psubw           m8, m14                         ; m8 = word: row 8
+    psubw           m9, m14                         ; m9 = word: row 9
+    psubw           m10, m14                        ; m10 = word: row 10
+    psubw           m11, m14                        ; m11 = word: row 11
+    psubw           m12, m14                        ; m12 = word: row 12
+    psubw           m13, m14                        ; m13 = word: row 13
+    psubw           m0, m14                         ; m0 = word: row 14
+    psubw           m1, m14                         ; m1 = word: row 15
+    movu            [r2], m8
+    movu            [r2 + r3], m9
+    movu            [r2 + r3 * 2], m10
+    movu            [r2 + r6], m11
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], m12
+    movu            [r2 + r3], m13
     movu            [r2 + r3 * 2], m0
-    movu            [r2 + r3 * 2 + mmsize], m6
+    movu            [r2 + r6], m1
 %endif
 
-    movu            m0, [r0 + r1 * 2]               ; m0 = row 6
-    punpcklbw       m6, m1, m0
-    punpckhbw       m7, m1, m0
-    pmaddubsw       m6, m11
-    pmaddubsw       m7, m11
-    paddw           m2, m6
-    paddw           m3, m7
-%ifidn %1,pp
-    pmulhrsw        m2, m12
-    pmulhrsw        m3, m12
-    packuswb        m2, m3
-    movu            [r2 + r5], m2
-%else
-    psubw           m2, m12
-    psubw           m3, m12
-    vperm2i128      m0, m2, m3, 0x20
-    vperm2i128      m2, m2, m3, 0x31
-    movu            [r2 + r5], m0
-    movu            [r2 + r5 + mmsize], m2
-%endif
     lea             r2, [r2 + r3 * 4]
-    dec             r6d
-    jnz             .loopW
+    sub             r0, r7
+    dec             r8d
+    jnz             .loop
     RET
 %endif
 %endmacro
 
-    FILTER_VER_CHROMA_AVX2_32xN pp, 64
-    FILTER_VER_CHROMA_AVX2_32xN pp, 48
-    FILTER_VER_CHROMA_AVX2_32xN pp, 32
-    FILTER_VER_CHROMA_AVX2_32xN pp, 24
-    FILTER_VER_CHROMA_AVX2_32xN pp, 16
-    FILTER_VER_CHROMA_AVX2_32xN pp, 8
-    FILTER_VER_CHROMA_AVX2_32xN ps, 64
-    FILTER_VER_CHROMA_AVX2_32xN ps, 48
-    FILTER_VER_CHROMA_AVX2_32xN ps, 32
-    FILTER_VER_CHROMA_AVX2_32xN ps, 24
-    FILTER_VER_CHROMA_AVX2_32xN ps, 16
-    FILTER_VER_CHROMA_AVX2_32xN ps, 8
-
-%macro FILTER_VER_CHROMA_AVX2_48x64 1
-%if ARCH_X86_64 == 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_48x64, 4, 8, 13
-    mov             r4d, r4m
-    shl             r4d, 6
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
-%endif
+    FILTER_VER_LUMA_AVX2_16xN 16, 32, pp
+    FILTER_VER_LUMA_AVX2_16xN 16, 64, pp
+    FILTER_VER_LUMA_AVX2_16xN 16, 32, ps
+    FILTER_VER_LUMA_AVX2_16xN 16, 64, ps
 
-    mova            m10, [r5]
-    mova            m11, [r5 + mmsize]
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,pp
-    mova            m12, [pw_512]
-%else
-    add             r3d, r3d
-    vbroadcasti128  m12, [pw_2000]
-%endif
-    lea             r5, [r3 * 3]
-    lea             r7, [r1 * 4]
-    mov             r6d, 16
-.loopH:
-    movu            m0, [r0]                        ; m0 = row 0
-    movu            m1, [r0 + r1]                   ; m1 = row 1
-    punpcklbw       m2, m0, m1
-    punpckhbw       m3, m0, m1
-    pmaddubsw       m2, m10
-    pmaddubsw       m3, m10
-    movu            m0, [r0 + r1 * 2]               ; m0 = row 2
-    punpcklbw       m4, m1, m0
-    punpckhbw       m5, m1, m0
-    pmaddubsw       m4, m10
-    pmaddubsw       m5, m10
-    movu            m1, [r0 + r4]                   ; m1 = row 3
-    punpcklbw       m6, m0, m1
-    punpckhbw       m7, m0, m1
-    pmaddubsw       m8, m6, m11
-    pmaddubsw       m9, m7, m11
-    pmaddubsw       m6, m10
-    pmaddubsw       m7, m10
-    paddw           m2, m8
-    paddw           m3, m9
-%ifidn %1,pp
-    pmulhrsw        m2, m12
-    pmulhrsw        m3, m12
-    packuswb        m2, m3
-    movu            [r2], m2
-%else
-    psubw           m2, m12
-    psubw           m3, m12
-    vperm2i128      m0, m2, m3, 0x20
-    vperm2i128      m2, m2, m3, 0x31
-    movu            [r2], m0
-    movu            [r2 + mmsize], m2
-%endif
-    lea             r0, [r0 + r1 * 4]
-    movu            m0, [r0]                        ; m0 = row 4
-    punpcklbw       m2, m1, m0
-    punpckhbw       m3, m1, m0
-    pmaddubsw       m8, m2, m11
-    pmaddubsw       m9, m3, m11
-    pmaddubsw       m2, m10
-    pmaddubsw       m3, m10
-    paddw           m4, m8
-    paddw           m5, m9
-%ifidn %1,pp
-    pmulhrsw        m4, m12
-    pmulhrsw        m5, m12
-    packuswb        m4, m5
-    movu            [r2 + r3], m4
-%else
-    psubw           m4, m12
-    psubw           m5, m12
-    vperm2i128      m1, m4, m5, 0x20
-    vperm2i128      m4, m4, m5, 0x31
-    movu            [r2 + r3], m1
-    movu            [r2 + r3 + mmsize], m4
-%endif
-
-    movu            m1, [r0 + r1]                   ; m1 = row 5
-    punpcklbw       m4, m0, m1
-    punpckhbw       m5, m0, m1
-    pmaddubsw       m4, m11
-    pmaddubsw       m5, m11
-    paddw           m6, m4
-    paddw           m7, m5
-%ifidn %1,pp
-    pmulhrsw        m6, m12
-    pmulhrsw        m7, m12
-    packuswb        m6, m7
-    movu            [r2 + r3 * 2], m6
-%else
-    psubw           m6, m12
-    psubw           m7, m12
-    vperm2i128      m0, m6, m7, 0x20
-    vperm2i128      m6, m6, m7, 0x31
-    movu            [r2 + r3 * 2], m0
-    movu            [r2 + r3 * 2 + mmsize], m6
-%endif
-
-    movu            m0, [r0 + r1 * 2]               ; m0 = row 6
-    punpcklbw       m6, m1, m0
-    punpckhbw       m7, m1, m0
-    pmaddubsw       m6, m11
-    pmaddubsw       m7, m11
-    paddw           m2, m6
-    paddw           m3, m7
-%ifidn %1,pp
-    pmulhrsw        m2, m12
-    pmulhrsw        m3, m12
-    packuswb        m2, m3
-    movu            [r2 + r5], m2
-    add             r2, 32
-%else
-    psubw           m2, m12
-    psubw           m3, m12
-    vperm2i128      m0, m2, m3, 0x20
-    vperm2i128      m2, m2, m3, 0x31
-    movu            [r2 + r5], m0
-    movu            [r2 + r5 + mmsize], m2
-    add             r2, 64
-%endif
-    sub             r0, r7
-
-    movu            xm0, [r0 + 32]                  ; m0 = row 0
-    movu            xm1, [r0 + r1 + 32]             ; m1 = row 1
+%macro PROCESS_LUMA_AVX2_W16_16R 1
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
     punpckhbw       xm2, xm0, xm1
     punpcklbw       xm0, xm1
     vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m0, m10
-    movu            xm2, [r0 + r1 * 2 + 32]         ; m2 = row 2
+    pmaddubsw       m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
     punpckhbw       xm3, xm1, xm2
     punpcklbw       xm1, xm2
     vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m1, m10
-    movu            xm3, [r0 + r4 + 32]             ; m3 = row 3
+    pmaddubsw       m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
     punpckhbw       xm4, xm2, xm3
     punpcklbw       xm2, xm3
     vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, m11
+    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
     paddw           m0, m4
-    pmaddubsw       m2, m10
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0 + 32]                  ; m4 = row 4
+    pmaddubsw       m2, [r5]
+    lea             r7, [r0 + r1 * 4]
+    movu            xm4, [r7]                       ; m4 = row 4
     punpckhbw       xm5, xm3, xm4
     punpcklbw       xm3, xm4
     vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, m11
+    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
     paddw           m1, m5
-    pmaddubsw       m3, m10
-    movu            xm5, [r0 + r1 + 32]             ; m5 = row 5
+    pmaddubsw       m3, [r5]
+    movu            xm5, [r7 + r1]                  ; m5 = row 5
     punpckhbw       xm6, xm4, xm5
     punpcklbw       xm4, xm5
     vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m4, m11
-    paddw           m2, m4
-    movu            xm6, [r0 + r1 * 2 + 32]         ; m6 = row 6
+    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
+    paddw           m0, m6
+    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
+    paddw           m2, m6
+    pmaddubsw       m4, [r5]
+    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
     punpckhbw       xm7, xm5, xm6
     punpcklbw       xm5, xm6
     vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m5, m11
-    paddw           m3, m5
+    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
+    paddw           m1, m7
+    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
+    paddw           m3, m7
+    pmaddubsw       m5, [r5]
+    movu            xm7, [r7 + r4]                  ; m7 = row 7
+    punpckhbw       xm8, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
+    paddw           m0, m8
+    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
+    paddw           m2, m8
+    pmaddubsw       m8, m6, [r5 + 1 * mmsize]
+    paddw           m4, m8
+    pmaddubsw       m6, [r5]
+    lea             r7, [r7 + r1 * 4]
+    movu            xm8, [r7]                       ; m8 = row 8
+    punpckhbw       xm9, xm7, xm8
+    punpcklbw       xm7, xm8
+    vinserti128     m7, m7, xm9, 1
+    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
+    paddw           m1, m9
+    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
+    paddw           m3, m9
+    pmaddubsw       m9, m7, [r5 + 1 * mmsize]
+    paddw           m5, m9
+    pmaddubsw       m7, [r5]
+    movu            xm9, [r7 + r1]                  ; m9 = row 9
+    punpckhbw       xm10, xm8, xm9
+    punpcklbw       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
+    paddw           m2, m10
+    pmaddubsw       m10, m8, [r5 + 2 * mmsize]
+    paddw           m4, m10
+    pmaddubsw       m10, m8, [r5 + 1 * mmsize]
+    paddw           m6, m10
+    pmaddubsw       m8, [r5]
+    movu            xm10, [r7 + r1 * 2]             ; m10 = row 10
+    punpckhbw       xm11, xm9, xm10
+    punpcklbw       xm9, xm10
+    vinserti128     m9, m9, xm11, 1
+    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
+    paddw           m3, m11
+    pmaddubsw       m11, m9, [r5 + 2 * mmsize]
+    paddw           m5, m11
+    pmaddubsw       m11, m9, [r5 + 1 * mmsize]
+    paddw           m7, m11
+    pmaddubsw       m9, [r5]
+    movu            xm11, [r7 + r4]                 ; m11 = row 11
+    punpckhbw       xm12, xm10, xm11
+    punpcklbw       xm10, xm11
+    vinserti128     m10, m10, xm12, 1
+    pmaddubsw       m12, m10, [r5 + 3 * mmsize]
+    paddw           m4, m12
+    pmaddubsw       m12, m10, [r5 + 2 * mmsize]
+    paddw           m6, m12
+    pmaddubsw       m12, m10, [r5 + 1 * mmsize]
+    paddw           m8, m12
+    pmaddubsw       m10, [r5]
+    lea             r7, [r7 + r1 * 4]
+    movu            xm12, [r7]                      ; m12 = row 12
+    punpckhbw       xm13, xm11, xm12
+    punpcklbw       xm11, xm12
+    vinserti128     m11, m11, xm13, 1
+    pmaddubsw       m13, m11, [r5 + 3 * mmsize]
+    paddw           m5, m13
+    pmaddubsw       m13, m11, [r5 + 2 * mmsize]
+    paddw           m7, m13
+    pmaddubsw       m13, m11, [r5 + 1 * mmsize]
+    paddw           m9, m13
+    pmaddubsw       m11, [r5]
+
 %ifidn %1,pp
-    pmulhrsw        m0, m12                         ; m0 = word: row 0
-    pmulhrsw        m1, m12                         ; m1 = word: row 1
-    pmulhrsw        m2, m12                         ; m2 = word: row 2
-    pmulhrsw        m3, m12                         ; m3 = word: row 3
+    pmulhrsw        m0, m14                         ; m0 = word: row 0
+    pmulhrsw        m1, m14                         ; m1 = word: row 1
+    pmulhrsw        m2, m14                         ; m2 = word: row 2
+    pmulhrsw        m3, m14                         ; m3 = word: row 3
+    pmulhrsw        m4, m14                         ; m4 = word: row 4
+    pmulhrsw        m5, m14                         ; m5 = word: row 5
     packuswb        m0, m1
     packuswb        m2, m3
+    packuswb        m4, m5
     vpermq          m0, m0, 11011000b
     vpermq          m2, m2, 11011000b
+    vpermq          m4, m4, 11011000b
     vextracti128    xm1, m0, 1
     vextracti128    xm3, m2, 1
+    vextracti128    xm5, m4, 1
     movu            [r2], xm0
     movu            [r2 + r3], xm1
     movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r5], xm3
-    lea             r2, [r2 + r3 * 4 - 32]
+    movu            [r2 + r6], xm3
+    lea             r8, [r2 + r3 * 4]
+    movu            [r8], xm4
+    movu            [r8 + r3], xm5
 %else
-    psubw           m0, m12                         ; m0 = word: row 0
-    psubw           m1, m12                         ; m1 = word: row 1
-    psubw           m2, m12                         ; m2 = word: row 2
-    psubw           m3, m12                         ; m3 = word: row 3
+    psubw           m0, m14                         ; m0 = word: row 0
+    psubw           m1, m14                         ; m1 = word: row 1
+    psubw           m2, m14                         ; m2 = word: row 2
+    psubw           m3, m14                         ; m3 = word: row 3
+    psubw           m4, m14                         ; m4 = word: row 4
+    psubw           m5, m14                         ; m5 = word: row 5
     movu            [r2], m0
     movu            [r2 + r3], m1
     movu            [r2 + r3 * 2], m2
-    movu            [r2 + r5], m3
-    lea             r2, [r2 + r3 * 4 - 64]
-%endif
-    dec             r6d
-    jnz             .loopH
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_CHROMA_AVX2_48x64 pp
-    FILTER_VER_CHROMA_AVX2_48x64 ps
-
-%macro FILTER_VER_CHROMA_AVX2_64xN 2
-%if ARCH_X86_64 == 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_64x%2, 4, 8, 13
-    mov             r4d, r4m
-    shl             r4d, 6
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+    movu            [r2 + r6], m3
+    lea             r8, [r2 + r3 * 4]
+    movu            [r8], m4
+    movu            [r8 + r3], m5
 %endif
 
-    mova            m10, [r5]
-    mova            m11, [r5 + mmsize]
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,pp
-    mova            m12, [pw_512]
-%else
-    add             r3d, r3d
-    vbroadcasti128  m12, [pw_2000]
-%endif
-    lea             r5, [r3 * 3]
-    lea             r7, [r1 * 4]
-    mov             r6d, %2 / 4
-.loopH:
-%assign x 0
-%rep 2
-    movu            m0, [r0 + x]                    ; m0 = row 0
-    movu            m1, [r0 + r1 + x]               ; m1 = row 1
-    punpcklbw       m2, m0, m1
-    punpckhbw       m3, m0, m1
-    pmaddubsw       m2, m10
-    pmaddubsw       m3, m10
-    movu            m0, [r0 + r1 * 2 + x]           ; m0 = row 2
-    punpcklbw       m4, m1, m0
-    punpckhbw       m5, m1, m0
-    pmaddubsw       m4, m10
-    pmaddubsw       m5, m10
-    movu            m1, [r0 + r4 + x]               ; m1 = row 3
-    punpcklbw       m6, m0, m1
-    punpckhbw       m7, m0, m1
-    pmaddubsw       m8, m6, m11
-    pmaddubsw       m9, m7, m11
-    pmaddubsw       m6, m10
-    pmaddubsw       m7, m10
-    paddw           m2, m8
-    paddw           m3, m9
-%ifidn %1,pp
-    pmulhrsw        m2, m12
-    pmulhrsw        m3, m12
-    packuswb        m2, m3
-    movu            [r2], m2
-%else
-    psubw           m2, m12
-    psubw           m3, m12
-    vperm2i128      m0, m2, m3, 0x20
-    vperm2i128      m2, m2, m3, 0x31
-    movu            [r2], m0
-    movu            [r2 + mmsize], m2
-%endif
-    lea             r0, [r0 + r1 * 4]
-    movu            m0, [r0 + x]                    ; m0 = row 4
-    punpcklbw       m2, m1, m0
-    punpckhbw       m3, m1, m0
-    pmaddubsw       m8, m2, m11
-    pmaddubsw       m9, m3, m11
-    pmaddubsw       m2, m10
-    pmaddubsw       m3, m10
-    paddw           m4, m8
-    paddw           m5, m9
-%ifidn %1,pp
-    pmulhrsw        m4, m12
-    pmulhrsw        m5, m12
-    packuswb        m4, m5
-    movu            [r2 + r3], m4
-%else
-    psubw           m4, m12
-    psubw           m5, m12
-    vperm2i128      m1, m4, m5, 0x20
-    vperm2i128      m4, m4, m5, 0x31
-    movu            [r2 + r3], m1
-    movu            [r2 + r3 + mmsize], m4
-%endif
+    movu            xm13, [r7 + r1]                 ; m13 = row 13
+    punpckhbw       xm0, xm12, xm13
+    punpcklbw       xm12, xm13
+    vinserti128     m12, m12, xm0, 1
+    pmaddubsw       m0, m12, [r5 + 3 * mmsize]
+    paddw           m6, m0
+    pmaddubsw       m0, m12, [r5 + 2 * mmsize]
+    paddw           m8, m0
+    pmaddubsw       m0, m12, [r5 + 1 * mmsize]
+    paddw           m10, m0
+    pmaddubsw       m12, [r5]
+    movu            xm0, [r7 + r1 * 2]              ; m0 = row 14
+    punpckhbw       xm1, xm13, xm0
+    punpcklbw       xm13, xm0
+    vinserti128     m13, m13, xm1, 1
+    pmaddubsw       m1, m13, [r5 + 3 * mmsize]
+    paddw           m7, m1
+    pmaddubsw       m1, m13, [r5 + 2 * mmsize]
+    paddw           m9, m1
+    pmaddubsw       m1, m13, [r5 + 1 * mmsize]
+    paddw           m11, m1
+    pmaddubsw       m13, [r5]
 
-    movu            m1, [r0 + r1 + x]               ; m1 = row 5
-    punpcklbw       m4, m0, m1
-    punpckhbw       m5, m0, m1
-    pmaddubsw       m4, m11
-    pmaddubsw       m5, m11
-    paddw           m6, m4
-    paddw           m7, m5
 %ifidn %1,pp
-    pmulhrsw        m6, m12
-    pmulhrsw        m7, m12
+    pmulhrsw        m6, m14                         ; m6 = word: row 6
+    pmulhrsw        m7, m14                         ; m7 = word: row 7
     packuswb        m6, m7
-    movu            [r2 + r3 * 2], m6
-%else
-    psubw           m6, m12
-    psubw           m7, m12
-    vperm2i128      m0, m6, m7, 0x20
-    vperm2i128      m6, m6, m7, 0x31
-    movu            [r2 + r3 * 2], m0
-    movu            [r2 + r3 * 2 + mmsize], m6
-%endif
-
-    movu            m0, [r0 + r1 * 2 + x]           ; m0 = row 6
-    punpcklbw       m6, m1, m0
-    punpckhbw       m7, m1, m0
-    pmaddubsw       m6, m11
-    pmaddubsw       m7, m11
-    paddw           m2, m6
-    paddw           m3, m7
-%ifidn %1,pp
-    pmulhrsw        m2, m12
-    pmulhrsw        m3, m12
-    packuswb        m2, m3
-    movu            [r2 + r5], m2
-    add             r2, 32
-%else
-    psubw           m2, m12
-    psubw           m3, m12
-    vperm2i128      m0, m2, m3, 0x20
-    vperm2i128      m2, m2, m3, 0x31
-    movu            [r2 + r5], m0
-    movu            [r2 + r5 + mmsize], m2
-    add             r2, 64
-%endif
-    sub             r0, r7
-%assign x x+32
-%endrep
-%ifidn %1,pp
-    lea             r2, [r2 + r3 * 4 - 64]
-%else
-    lea             r2, [r2 + r3 * 4 - 128]
-%endif
-    add             r0, r7
-    dec             r6d
-    jnz             .loopH
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_CHROMA_AVX2_64xN pp, 64
-    FILTER_VER_CHROMA_AVX2_64xN pp, 48
-    FILTER_VER_CHROMA_AVX2_64xN pp, 32
-    FILTER_VER_CHROMA_AVX2_64xN pp, 16
-    FILTER_VER_CHROMA_AVX2_64xN ps, 64
-    FILTER_VER_CHROMA_AVX2_64xN ps, 48
-    FILTER_VER_CHROMA_AVX2_64xN ps, 32
-    FILTER_VER_CHROMA_AVX2_64xN ps, 16
-
-;-----------------------------------------------------------------------------
-; void interp_4tap_vert_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro FILTER_V4_W16n_H2 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_pp_%1x%2, 4, 7, 8
-
-    mov         r4d,       r4m
-    sub         r0,        r1
-
-%ifdef PIC
-    lea         r5,        [tab_ChromaCoeff]
-    movd        m0,        [r5 + r4 * 4]
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm7, m6, 1
+    movu            [r8 + r3 * 2], xm6
+    movu            [r8 + r6], xm7
 %else
-    movd        m0,        [tab_ChromaCoeff + r4 * 4]
+    psubw           m6, m14                         ; m6 = word: row 6
+    psubw           m7, m14                         ; m7 = word: row 7
+    movu            [r8 + r3 * 2], m6
+    movu            [r8 + r6], m7
 %endif
 
-    pshufb      m1,        m0,       [tab_Vm]
-    pshufb      m0,        [tab_Vm + 16]
-
-    mov         r4d,       %2/2
+    lea             r8, [r8 + r3 * 4]
 
-.loop:
-
-    mov         r6d,       %1/16
-
-.loopW:
-
-    movu        m2,        [r0]
-    movu        m3,        [r0 + r1]
-
-    punpcklbw   m4,        m2,        m3
-    punpckhbw   m2,        m3
-
-    pmaddubsw   m4,        m1
-    pmaddubsw   m2,        m1
-
-    lea         r5,        [r0 + 2 * r1]
-    movu        m5,        [r5]
-    movu        m6,        [r5 + r1]
-
-    punpckhbw   m7,        m5,        m6
-    pmaddubsw   m7,        m0
-    paddw       m2,        m7
-
-    punpcklbw   m7,        m5,        m6
-    pmaddubsw   m7,        m0
-    paddw       m4,        m7
-
-    mova        m7,        [pw_512]
-
-    pmulhrsw    m4,        m7
-    pmulhrsw    m2,        m7
-
-    packuswb    m4,        m2
-
-    movu        [r2],      m4
-
-    punpcklbw   m4,        m3,        m5
-    punpckhbw   m3,        m5
-
-    pmaddubsw   m4,        m1
-    pmaddubsw   m3,        m1
-
-    movu        m5,        [r5 + 2 * r1]
-
-    punpcklbw   m2,        m6,        m5
-    punpckhbw   m6,        m5
-
-    pmaddubsw   m2,        m0
-    pmaddubsw   m6,        m0
-
-    paddw       m4,        m2
-    paddw       m3,        m6
-
-    pmulhrsw    m4,        m7
-    pmulhrsw    m3,        m7
-
-    packuswb    m4,        m3
-
-    movu        [r2 + r3],      m4
-
-    add         r0,        16
-    add         r2,        16
-    dec         r6d
-    jnz         .loopW
-
-    lea         r0,        [r0 + r1 * 2 - %1]
-    lea         r2,        [r2 + r3 * 2 - %1]
-
-    dec         r4d
-    jnz        .loop
-    RET
-%endmacro
-
-    FILTER_V4_W16n_H2 64, 64
-    FILTER_V4_W16n_H2 64, 32
-    FILTER_V4_W16n_H2 64, 48
-    FILTER_V4_W16n_H2 48, 64
-    FILTER_V4_W16n_H2 64, 16
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-%macro P2S_H_2xN 1
-INIT_XMM sse4
-cglobal filterPixelToShort_2x%1, 3, 4, 3
-    mov         r3d, r3m
-    add         r3d, r3d
-
-    ; load constant
-    mova        m1, [pb_128]
-    mova        m2, [tab_c_64_n64]
-
-%rep %1/2
-    movd        m0, [r0]
-    pinsrd      m0, [r0 + r1], 1
-    punpcklbw   m0, m1
-    pmaddubsw   m0, m2
-
-    movd        [r2 + r3 * 0], m0
-    pextrd      [r2 + r3 * 1], m0, 2
-
-    lea         r0, [r0 + r1 * 2]
-    lea         r2, [r2 + r3 * 2]
-%endrep
-    RET
-%endmacro
-    P2S_H_2xN 4
-    P2S_H_2xN 8
-    P2S_H_2xN 16
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-%macro P2S_H_4xN 1
-INIT_XMM sse4
-cglobal filterPixelToShort_4x%1, 3, 6, 4
-    mov         r3d, r3m
-    add         r3d, r3d
-    lea         r4, [r3 * 3]
-    lea         r5, [r1 * 3]
-
-    ; load constant
-    mova        m2, [pb_128]
-    mova        m3, [tab_c_64_n64]
-
-%assign x 0
-%rep %1/4
-    movd        m0, [r0]
-    pinsrd      m0, [r0 + r1], 1
-    punpcklbw   m0, m2
-    pmaddubsw   m0, m3
-
-    movd        m1, [r0 + r1 * 2]
-    pinsrd      m1, [r0 + r5], 1
-    punpcklbw   m1, m2
-    pmaddubsw   m1, m3
-
-    movq        [r2 + r3 * 0], m0
-    movq        [r2 + r3 * 2], m1
-    movhps      [r2 + r3 * 1], m0
-    movhps      [r2 + r4], m1
-%assign x x+1
-%if (x != %1/4)
-    lea         r0, [r0 + r1 * 4]
-    lea         r2, [r2 + r3 * 4]
-%endif
-%endrep
-    RET
-%endmacro
-    P2S_H_4xN 4
-    P2S_H_4xN 8
-    P2S_H_4xN 16
-    P2S_H_4xN 32
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-%macro P2S_H_6xN 1
-INIT_XMM sse4
-cglobal filterPixelToShort_6x%1, 3, 7, 6
-    mov         r3d, r3m
-    add         r3d, r3d
-    lea         r4, [r1 * 3]
-    lea         r5, [r3 * 3]
-
-    ; load height
-    mov         r6d, %1/4
-
-    ; load constant
-    mova        m4, [pb_128]
-    mova        m5, [tab_c_64_n64]
-
-.loop:
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
-
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
-
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
-
-    movh        m3, [r0 + r4]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
-
-    movh        [r2 + r3 * 0], m0
-    pextrd      [r2 + r3 * 0 + 8], m0, 2
-    movh        [r2 + r3 * 1], m1
-    pextrd      [r2 + r3 * 1 + 8], m1, 2
-    movh        [r2 + r3 * 2], m2
-    pextrd      [r2 + r3 * 2 + 8], m2, 2
-    movh        [r2 + r5], m3
-    pextrd      [r2 + r5 + 8], m3, 2
-
-    lea         r0, [r0 + r1 * 4]
-    lea         r2, [r2 + r3 * 4]
-
-    dec         r6d
-    jnz         .loop
-    RET
-%endmacro
-    P2S_H_6xN 8
-    P2S_H_6xN 16
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-%macro P2S_H_8xN 1
-INIT_XMM ssse3
-cglobal filterPixelToShort_8x%1, 3, 7, 6
-    mov         r3d, r3m
-    add         r3d, r3d
-    lea         r5, [r1 * 3]
-    lea         r6, [r3 * 3]
-
-    ; load height
-    mov         r4d, %1/4
-
-    ; load constant
-    mova        m4, [pb_128]
-    mova        m5, [tab_c_64_n64]
-
-.loop:
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
-
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
-
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
-
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
-
-    movu        [r2 + r3 * 0], m0
-    movu        [r2 + r3 * 1], m1
-    movu        [r2 + r3 * 2], m2
-    movu        [r2 + r6 ], m3
-
-    lea         r0, [r0 + r1 * 4]
-    lea         r2, [r2 + r3 * 4]
-
-    dec         r4d
-    jnz         .loop
-    RET
-%endmacro
-    P2S_H_8xN 8
-    P2S_H_8xN 4
-    P2S_H_8xN 16
-    P2S_H_8xN 32
-    P2S_H_8xN 12
-    P2S_H_8xN 64
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-INIT_XMM ssse3
-cglobal filterPixelToShort_8x6, 3, 7, 5
-    mov         r3d, r3m
-    add         r3d, r3d
-    lea         r4, [r1 * 3]
-    lea         r5, [r1 * 5]
-    lea         r6, [r3 * 3]
-
-    ; load constant
-    mova        m3, [pb_128]
-    mova        m4, [tab_c_64_n64]
-
-    movh        m0, [r0]
-    punpcklbw   m0, m3
-    pmaddubsw   m0, m4
-
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m3
-    pmaddubsw   m1, m4
-
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m3
-    pmaddubsw   m2, m4
-
-    movu        [r2 + r3 * 0], m0
-    movu        [r2 + r3 * 1], m1
-    movu        [r2 + r3 * 2], m2
-
-    movh        m0, [r0 + r4]
-    punpcklbw   m0, m3
-    pmaddubsw   m0, m4
-
-    movh        m1, [r0 + r1 * 4]
-    punpcklbw   m1, m3
-    pmaddubsw   m1, m4
-
-    movh        m2, [r0 + r5]
-    punpcklbw   m2, m3
-    pmaddubsw   m2, m4
-
-    movu        [r2 + r6 ], m0
-    movu        [r2 + r3 * 4], m1
-    lea         r2, [r2 + r3 * 4]
-    movu        [r2 + r3], m2
-
-    RET
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-%macro P2S_H_16xN 1
-INIT_XMM ssse3
-cglobal filterPixelToShort_16x%1, 3, 7, 6
-    mov         r3d, r3m
-    add         r3d, r3d
-    lea         r4, [r3 * 3]
-    lea         r5, [r1 * 3]
-
-   ; load height
-    mov         r6d, %1/4
-
-    ; load constant
-    mova        m4, [pb_128]
-    mova        m5, [tab_c_64_n64]
-
-.loop:
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
-
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
-
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
-
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
-
-    movu        [r2 + r3 * 0], m0
-    movu        [r2 + r3 * 1], m1
-    movu        [r2 + r3 * 2], m2
-    movu        [r2 + r4], m3
-
-    lea         r0, [r0 + 8]
-
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
-
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
-
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
-
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
-
-    movu        [r2 + r3 * 0 + 16], m0
-    movu        [r2 + r3 * 1 + 16], m1
-    movu        [r2 + r3 * 2 + 16], m2
-    movu        [r2 + r4 + 16], m3
-
-    lea         r0, [r0 + r1 * 4 - 8]
-    lea         r2, [r2 + r3 * 4]
-
-    dec         r6d
-    jnz         .loop
-    RET
-%endmacro
-    P2S_H_16xN 16
-    P2S_H_16xN 4
-    P2S_H_16xN 8
-    P2S_H_16xN 12
-    P2S_H_16xN 32
-    P2S_H_16xN 64
-    P2S_H_16xN 24
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-INIT_YMM avx2
-cglobal filterPixelToShort_16x4, 3, 4, 2
-    mov             r3d, r3m
-    add             r3d, r3d
-
-    ; load constant
-    vbroadcasti128  m1, [pw_2000]
-
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
-
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
-
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
-
-    lea             r1, [r1 * 3]
-    lea             r3, [r3 * 3]
-
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
-    RET
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-INIT_YMM avx2
-cglobal filterPixelToShort_16x8, 3, 6, 2
-    mov             r3d, r3m
-    add             r3d, r3d
-    lea             r4, [r1 * 3]
-    lea             r5, [r3 * 3]
-
-    ; load constant
-    vbroadcasti128  m1, [pw_2000]
-
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
-
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
-
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
-
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
-
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
-
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
-
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
-
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
-
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
-    RET
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-INIT_YMM avx2
-cglobal filterPixelToShort_16x12, 3, 6, 2
-    mov             r3d, r3m
-    add             r3d, r3d
-    lea             r4, [r1 * 3]
-    lea             r5, [r3 * 3]
-
-    ; load constant
-    vbroadcasti128  m1, [pw_2000]
-
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
-
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
-
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
-
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
-
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
-
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
-
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
-
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
-
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
-
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
-
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
-
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
-
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
-
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
-    RET
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-INIT_YMM avx2
-cglobal filterPixelToShort_16x16, 3, 6, 2
-    mov             r3d, r3m
-    add             r3d, r3d
-    lea             r4, [r1 * 3]
-    lea             r5, [r3 * 3]
-
-    ; load constant
-    vbroadcasti128  m1, [pw_2000]
-
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
-
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
-
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
-
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
-
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
-
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
-
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
-
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
-
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
-
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
-
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
-
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
-
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
-
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
-
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
-
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
-
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
-
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
-
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
-    RET
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-INIT_YMM avx2
-cglobal filterPixelToShort_16x24, 3, 7, 2
-    mov             r3d, r3m
-    add             r3d, r3d
-    lea             r4, [r1 * 3]
-    lea             r5, [r3 * 3]
-    mov             r6d, 3
-
-    ; load constant
-    vbroadcasti128  m1, [pw_2000]
-.loop:
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
-
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
-
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
-
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
-
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
-
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
-
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
-
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
-
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
-
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
-
-    dec             r6d
-    jnz             .loop
-    RET
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-%macro P2S_H_16xN_avx2 1
-INIT_YMM avx2
-cglobal filterPixelToShort_16x%1, 3, 7, 2
-    mov             r3d, r3m
-    add             r3d, r3d
-    lea             r4, [r1 * 3]
-    lea             r5, [r3 * 3]
-    mov             r6d, %1/16
-
-    ; load constant
-    vbroadcasti128  m1, [pw_2000]
-.loop:
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
-
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
-
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
-
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
-
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
-
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
-
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
-
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
-
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
-
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
-
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
-
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
-
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
-
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
-
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
-
-    pmovzxbw        m0, [r0]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2], m0
-
-    pmovzxbw        m0, [r0 + r1]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3], m0
-
-    pmovzxbw        m0, [r0 + r1 * 2]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r3 * 2], m0
-
-    pmovzxbw        m0, [r0 + r4]
-    psllw           m0, 6
-    psubw           m0, m1
-    movu            [r2 + r5], m0
-
-    lea             r0, [r0 + r1 * 4]
-    lea             r2, [r2 + r3 * 4]
-
-    dec             r6d
-    jnz             .loop
-    RET
-%endmacro
-P2S_H_16xN_avx2 32
-P2S_H_16xN_avx2 64
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-%macro P2S_H_32xN 1
-INIT_XMM ssse3
-cglobal filterPixelToShort_32x%1, 3, 7, 6
-    mov         r3d, r3m
-    add         r3d, r3d
-    lea         r4, [r3 * 3]
-    lea         r5, [r1 * 3]
-
-    ; load height
-    mov         r6d, %1/4
-
-    ; load constant
-    mova        m4, [pb_128]
-    mova        m5, [tab_c_64_n64]
-
-.loop:
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
-
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
-
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
-
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
-
-    movu        [r2 + r3 * 0], m0
-    movu        [r2 + r3 * 1], m1
-    movu        [r2 + r3 * 2], m2
-    movu        [r2 + r4], m3
-
-    lea         r0, [r0 + 8]
-
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
-
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
-
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
-
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
-
-    movu        [r2 + r3 * 0 + 16], m0
-    movu        [r2 + r3 * 1 + 16], m1
-    movu        [r2 + r3 * 2 + 16], m2
-    movu        [r2 + r4 + 16], m3
-
-    lea         r0, [r0 + 8]
-
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
-
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
-
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
-
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
-
-    movu        [r2 + r3 * 0 + 32], m0
-    movu        [r2 + r3 * 1 + 32], m1
-    movu        [r2 + r3 * 2 + 32], m2
-    movu        [r2 + r4 + 32], m3
-
-    lea         r0, [r0 + 8]
-
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
-
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
-
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
-
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
-
-    movu        [r2 + r3 * 0 + 48], m0
-    movu        [r2 + r3 * 1 + 48], m1
-    movu        [r2 + r3 * 2 + 48], m2
-    movu        [r2 + r4 + 48], m3
-
-    lea         r0, [r0 + r1 * 4 - 24]
-    lea         r2, [r2 + r3 * 4]
-
-    dec         r6d
-    jnz         .loop
-    RET
-%endmacro
-    P2S_H_32xN 32
-    P2S_H_32xN 8
-    P2S_H_32xN 16
-    P2S_H_32xN 24
-    P2S_H_32xN 64
-    P2S_H_32xN 48
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-%macro P2S_H_32xN_avx2 1
-INIT_YMM avx2
-cglobal filterPixelToShort_32x%1, 3, 7, 3
-    mov         r3d, r3m
-    add         r3d, r3d
-    lea         r5, [r1 * 3]
-    lea         r6, [r3 * 3]
-
-    ; load height
-    mov         r4d, %1/4
-
-    ; load constant
-    vpbroadcastd m2, [pw_2000]
-
-.loop:
-    pmovzxbw    m0, [r0 + 0 * mmsize/2]
-    pmovzxbw    m1, [r0 + 1 * mmsize/2]
-    psllw       m0, 6
-    psllw       m1, 6
-    psubw       m0, m2
-    psubw       m1, m2
-    movu        [r2 + 0 * mmsize], m0
-    movu        [r2 + 1 * mmsize], m1
-
-    pmovzxbw    m0, [r0 + r1 + 0 * mmsize/2]
-    pmovzxbw    m1, [r0 + r1 + 1 * mmsize/2]
-    psllw       m0, 6
-    psllw       m1, 6
-    psubw       m0, m2
-    psubw       m1, m2
-    movu        [r2 + r3 + 0 * mmsize], m0
-    movu        [r2 + r3 + 1 * mmsize], m1
-
-    pmovzxbw    m0, [r0 + r1 * 2 + 0 * mmsize/2]
-    pmovzxbw    m1, [r0 + r1 * 2 + 1 * mmsize/2]
-    psllw       m0, 6
-    psllw       m1, 6
-    psubw       m0, m2
-    psubw       m1, m2
-    movu        [r2 + r3 * 2 + 0 * mmsize], m0
-    movu        [r2 + r3 * 2 + 1 * mmsize], m1
-
-    pmovzxbw    m0, [r0 + r5 + 0 * mmsize/2]
-    pmovzxbw    m1, [r0 + r5 + 1 * mmsize/2]
-    psllw       m0, 6
-    psllw       m1, 6
-    psubw       m0, m2
-    psubw       m1, m2
-    movu        [r2 + r6 + 0 * mmsize], m0
-    movu        [r2 + r6 + 1 * mmsize], m1
-
-    lea         r0, [r0 + r1 * 4]
-    lea         r2, [r2 + r3 * 4]
-
-    dec         r4d
-    jnz        .loop
-    RET
-%endmacro
-    P2S_H_32xN_avx2 32
-    P2S_H_32xN_avx2 8
-    P2S_H_32xN_avx2 16
-    P2S_H_32xN_avx2 24
-    P2S_H_32xN_avx2 64
-    P2S_H_32xN_avx2 48
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-%macro P2S_H_64xN 1
-INIT_XMM ssse3
-cglobal filterPixelToShort_64x%1, 3, 7, 6
-    mov         r3d, r3m
-    add         r3d, r3d
-    lea         r4, [r3 * 3]
-    lea         r5, [r1 * 3]
-
-    ; load height
-    mov         r6d, %1/4
-
-    ; load constant
-    mova        m4, [pb_128]
-    mova        m5, [tab_c_64_n64]
-
-.loop:
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
-
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
-
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
-
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
-
-    movu        [r2 + r3 * 0], m0
-    movu        [r2 + r3 * 1], m1
-    movu        [r2 + r3 * 2], m2
-    movu        [r2 + r4], m3
-
-    lea         r0, [r0 + 8]
-
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
-
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
-
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
-
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
-
-    movu        [r2 + r3 * 0 + 16], m0
-    movu        [r2 + r3 * 1 + 16], m1
-    movu        [r2 + r3 * 2 + 16], m2
-    movu        [r2 + r4 + 16], m3
-
-    lea         r0, [r0 + 8]
-
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
-
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
-
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
-
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
-
-    movu        [r2 + r3 * 0 + 32], m0
-    movu        [r2 + r3 * 1 + 32], m1
-    movu        [r2 + r3 * 2 + 32], m2
-    movu        [r2 + r4 + 32], m3
-
-    lea         r0, [r0 + 8]
-
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
-
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
-
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
-
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
-
-    movu        [r2 + r3 * 0 + 48], m0
-    movu        [r2 + r3 * 1 + 48], m1
-    movu        [r2 + r3 * 2 + 48], m2
-    movu        [r2 + r4 + 48], m3
-
-    lea         r0, [r0 + 8]
-
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
-
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
-
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
-
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
-
-    movu        [r2 + r3 * 0 + 64], m0
-    movu        [r2 + r3 * 1 + 64], m1
-    movu        [r2 + r3 * 2 + 64], m2
-    movu        [r2 + r4 + 64], m3
-
-    lea         r0, [r0 + 8]
-
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
-
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
-
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
-
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
-
-    movu        [r2 + r3 * 0 + 80], m0
-    movu        [r2 + r3 * 1 + 80], m1
-    movu        [r2 + r3 * 2 + 80], m2
-    movu        [r2 + r4 + 80], m3
-
-    lea         r0, [r0 + 8]
-
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
-
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
-
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
-
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
-
-    movu        [r2 + r3 * 0 + 96], m0
-    movu        [r2 + r3 * 1 + 96], m1
-    movu        [r2 + r3 * 2 + 96], m2
-    movu        [r2 + r4 + 96], m3
-
-    lea         r0, [r0 + 8]
-
-    movh        m0, [r0]
-    punpcklbw   m0, m4
-    pmaddubsw   m0, m5
-
-    movh        m1, [r0 + r1]
-    punpcklbw   m1, m4
-    pmaddubsw   m1, m5
-
-    movh        m2, [r0 + r1 * 2]
-    punpcklbw   m2, m4
-    pmaddubsw   m2, m5
-
-    movh        m3, [r0 + r5]
-    punpcklbw   m3, m4
-    pmaddubsw   m3, m5
-
-    movu        [r2 + r3 * 0 + 112], m0
-    movu        [r2 + r3 * 1 + 112], m1
-    movu        [r2 + r3 * 2 + 112], m2
-    movu        [r2 + r4 + 112], m3
-
-    lea         r0, [r0 + r1 * 4 - 56]
-    lea         r2, [r2 + r3 * 4]
-
-    dec         r6d
-    jnz         .loop
-    RET
-%endmacro
-    P2S_H_64xN 64
-    P2S_H_64xN 16
-    P2S_H_64xN 32
-    P2S_H_64xN 48
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-%macro P2S_H_64xN_avx2 1
-INIT_YMM avx2
-cglobal filterPixelToShort_64x%1, 3, 7, 5
-    mov         r3d, r3m
-    add         r3d, r3d
-    lea         r5, [r1 * 3]
-    lea         r6, [r3 * 3]
-
-    ; load height
-    mov         r4d, %1/4
-
-    ; load constant
-    vpbroadcastd m4, [pw_2000]
-
-.loop:
-    pmovzxbw    m0, [r0 + 0 * mmsize/2]
-    pmovzxbw    m1, [r0 + 1 * mmsize/2]
-    pmovzxbw    m2, [r0 + 2 * mmsize/2]
-    pmovzxbw    m3, [r0 + 3 * mmsize/2]
-    psllw       m0, 6
-    psllw       m1, 6
-    psllw       m2, 6
-    psllw       m3, 6
-    psubw       m0, m4
-    psubw       m1, m4
-    psubw       m2, m4
-    psubw       m3, m4
-
-    movu        [r2 + 0 * mmsize], m0
-    movu        [r2 + 1 * mmsize], m1
-    movu        [r2 + 2 * mmsize], m2
-    movu        [r2 + 3 * mmsize], m3
-
-    pmovzxbw    m0, [r0 + r1 + 0 * mmsize/2]
-    pmovzxbw    m1, [r0 + r1 + 1 * mmsize/2]
-    pmovzxbw    m2, [r0 + r1 + 2 * mmsize/2]
-    pmovzxbw    m3, [r0 + r1 + 3 * mmsize/2]
-    psllw       m0, 6
-    psllw       m1, 6
-    psllw       m2, 6
-    psllw       m3, 6
-    psubw       m0, m4
-    psubw       m1, m4
-    psubw       m2, m4
-    psubw       m3, m4
-
-    movu        [r2 + r3 + 0 * mmsize], m0
-    movu        [r2 + r3 + 1 * mmsize], m1
-    movu        [r2 + r3 + 2 * mmsize], m2
-    movu        [r2 + r3 + 3 * mmsize], m3
-
-    pmovzxbw    m0, [r0 + r1 * 2 + 0 * mmsize/2]
-    pmovzxbw    m1, [r0 + r1 * 2 + 1 * mmsize/2]
-    pmovzxbw    m2, [r0 + r1 * 2 + 2 * mmsize/2]
-    pmovzxbw    m3, [r0 + r1 * 2 + 3 * mmsize/2]
-    psllw       m0, 6
-    psllw       m1, 6
-    psllw       m2, 6
-    psllw       m3, 6
-    psubw       m0, m4
-    psubw       m1, m4
-    psubw       m2, m4
-    psubw       m3, m4
-
-    movu        [r2 + r3 * 2 + 0 * mmsize], m0
-    movu        [r2 + r3 * 2 + 1 * mmsize], m1
-    movu        [r2 + r3 * 2 + 2 * mmsize], m2
-    movu        [r2 + r3 * 2 + 3 * mmsize], m3
-
-    pmovzxbw    m0, [r0 + r5 + 0 * mmsize/2]
-    pmovzxbw    m1, [r0 + r5 + 1 * mmsize/2]
-    pmovzxbw    m2, [r0 + r5 + 2 * mmsize/2]
-    pmovzxbw    m3, [r0 + r5 + 3 * mmsize/2]
-    psllw       m0, 6
-    psllw       m1, 6
-    psllw       m2, 6
-    psllw       m3, 6
-    psubw       m0, m4
-    psubw       m1, m4
-    psubw       m2, m4
-    psubw       m3, m4
-
-    movu        [r2 + r6 + 0 * mmsize], m0
-    movu        [r2 + r6 + 1 * mmsize], m1
-    movu        [r2 + r6 + 2 * mmsize], m2
-    movu        [r2 + r6 + 3 * mmsize], m3
-
-    lea         r0, [r0 + r1 * 4]
-    lea         r2, [r2 + r3 * 4]
-
-    dec         r4d
-    jnz        .loop
-    RET
-%endmacro
-    P2S_H_64xN_avx2 64
-    P2S_H_64xN_avx2 16
-    P2S_H_64xN_avx2 32
-    P2S_H_64xN_avx2 48
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel src, intptr_t srcStride, int16_t dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-%macro P2S_H_12xN 1
-INIT_XMM ssse3
-cglobal filterPixelToShort_12x%1, 3, 7, 6
-    mov         r3d, r3m
-    add         r3d, r3d
-    lea         r4, [r1 * 3]
-    lea         r6, [r3 * 3]
-    mov         r5d, %1/4
-
-    ; load constant
-    mova        m4, [pb_128]
-    mova        m5, [tab_c_64_n64]
-
-.loop:
-    movu        m0, [r0]
-    punpcklbw   m1, m0, m4
-    punpckhbw   m0, m4
-    pmaddubsw   m0, m5
-    pmaddubsw   m1, m5
-
-    movu        m2, [r0 + r1]
-    punpcklbw   m3, m2, m4
-    punpckhbw   m2, m4
-    pmaddubsw   m2, m5
-    pmaddubsw   m3, m5
-
-    movu        [r2 + r3 * 0], m1
-    movu        [r2 + r3 * 1], m3
-
-    movh        [r2 + r3 * 0 + 16], m0
-    movh        [r2 + r3 * 1 + 16], m2
-
-    movu        m0, [r0 + r1 * 2]
-    punpcklbw   m1, m0, m4
-    punpckhbw   m0, m4
-    pmaddubsw   m0, m5
-    pmaddubsw   m1, m5
-
-    movu        m2, [r0 + r4]
-    punpcklbw   m3, m2, m4
-    punpckhbw   m2, m4
-    pmaddubsw   m2, m5
-    pmaddubsw   m3, m5
-
-    movu        [r2 + r3 * 2], m1
-    movu        [r2 + r6], m3
-
-    movh        [r2 + r3 * 2 + 16], m0
-    movh        [r2 + r6 + 16], m2
-
-    lea         r0, [r0 + r1 * 4]
-    lea         r2, [r2 + r3 * 4]
-
-    dec         r5d
-    jnz         .loop
-    RET
-%endmacro
-    P2S_H_12xN 16
-    P2S_H_12xN 32
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-%macro P2S_H_24xN 1
-INIT_XMM ssse3
-cglobal filterPixelToShort_24x%1, 3, 7, 5
-    mov         r3d, r3m
-    add         r3d, r3d
-    lea         r4, [r1 * 3]
-    lea         r5, [r3 * 3]
-    mov         r6d, %1/4
-
-    ; load constant
-    mova        m3, [pb_128]
-    mova        m4, [tab_c_64_n64]
-
-.loop:
-    movu        m0, [r0]
-    punpcklbw   m1, m0, m3
-    punpckhbw   m0, m3
-    pmaddubsw   m0, m4
-    pmaddubsw   m1, m4
-
-    movu        m2, [r0 + 16]
-    punpcklbw   m2, m3
-    pmaddubsw   m2, m4
-
-    movu        [r2 +  r3 * 0], m1
-    movu        [r2 +  r3 * 0 + 16], m0
-    movu        [r2 +  r3 * 0 + 32], m2
-
-    movu        m0, [r0 + r1]
-    punpcklbw   m1, m0, m3
-    punpckhbw   m0, m3
-    pmaddubsw   m0, m4
-    pmaddubsw   m1, m4
-
-    movu        m2, [r0 + r1 + 16]
-    punpcklbw   m2, m3
-    pmaddubsw   m2, m4
-
-    movu        [r2 +  r3 * 1], m1
-    movu        [r2 +  r3 * 1 + 16], m0
-    movu        [r2 +  r3 * 1 + 32], m2
-
-    movu        m0, [r0 + r1 * 2]
-    punpcklbw   m1, m0, m3
-    punpckhbw   m0, m3
-    pmaddubsw   m0, m4
-    pmaddubsw   m1, m4
-
-    movu        m2, [r0 + r1 * 2 + 16]
-    punpcklbw   m2, m3
-    pmaddubsw   m2, m4
-
-    movu        [r2 +  r3 * 2], m1
-    movu        [r2 +  r3 * 2 + 16], m0
-    movu        [r2 +  r3 * 2 + 32], m2
-
-    movu        m0, [r0 + r4]
-    punpcklbw   m1, m0, m3
-    punpckhbw   m0, m3
-    pmaddubsw   m0, m4
-    pmaddubsw   m1, m4
-
-    movu        m2, [r0 + r4 + 16]
-    punpcklbw   m2, m3
-    pmaddubsw   m2, m4
-    movu        [r2 +  r5], m1
-    movu        [r2 +  r5 + 16], m0
-    movu        [r2 +  r5 + 32], m2
-
-    lea         r0, [r0 + r1 * 4]
-    lea         r2, [r2 + r3 * 4]
-
-    dec         r6d
-    jnz         .loop
-    RET
-%endmacro
-    P2S_H_24xN 32
-    P2S_H_24xN 64
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-%macro P2S_H_24xN_avx2 1
-INIT_YMM avx2
-cglobal filterPixelToShort_24x%1, 3, 7, 4
-    mov         r3d, r3m
-    add         r3d, r3d
-    lea         r4, [r1 * 3]
-    lea         r5, [r3 * 3]
-    mov         r6d, %1/4
-
-    ; load constant
-    vpbroadcastd m1, [pw_2000]
-    vpbroadcastd m2, [pb_128]
-    vpbroadcastd m3, [tab_c_64_n64]
-
-.loop:
-    pmovzxbw    m0, [r0]
-    psllw       m0, 6
-    psubw       m0, m1
-    movu        [r2], m0
-
-    movu        m0, [r0 + mmsize/2]
-    punpcklbw   m0, m2
-    pmaddubsw   m0, m3
-    movu        [r2 +  r3 * 0 + mmsize], xm0
-
-    pmovzxbw    m0, [r0 + r1]
-    psllw       m0, 6
-    psubw       m0, m1
-    movu        [r2 + r3], m0
-
-    movu        m0, [r0 + r1 + mmsize/2]
-    punpcklbw   m0, m2
-    pmaddubsw   m0, m3
-    movu        [r2 +  r3 * 1 + mmsize], xm0
-
-    pmovzxbw    m0, [r0 + r1 * 2]
-    psllw       m0, 6
-    psubw       m0, m1
-    movu        [r2 + r3 * 2], m0
-
-    movu        m0, [r0 + r1 * 2 + mmsize/2]
-    punpcklbw   m0, m2
-    pmaddubsw   m0, m3
-    movu        [r2 +  r3 * 2 + mmsize], xm0
-
-    pmovzxbw    m0, [r0 + r4]
-    psllw       m0, 6
-    psubw       m0, m1
-    movu        [r2 + r5], m0
-
-    movu        m0, [r0 + r4 + mmsize/2]
-    punpcklbw   m0, m2
-    pmaddubsw   m0, m3
-    movu        [r2 + r5 + mmsize], xm0
-
-    lea         r0, [r0 + r1 * 4]
-    lea         r2, [r2 + r3 * 4]
-
-    dec         r6d
-    jnz         .loop
-    RET
-%endmacro
-    P2S_H_24xN_avx2 32
-    P2S_H_24xN_avx2 64
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-INIT_XMM ssse3
-cglobal filterPixelToShort_48x64, 3, 7, 4
-    mov         r3d, r3m
-    add         r3d, r3d
-    lea         r4, [r1 * 3]
-    lea         r5, [r3 * 3]
-    mov         r6d, 16
-
-    ; load constant
-    mova        m2, [pb_128]
-    mova        m3, [tab_c_64_n64]
-
-.loop:
-    movu        m0, [r0]
-    punpcklbw   m1, m0, m2
-    punpckhbw   m0, m2
-    pmaddubsw   m0, m3
-    pmaddubsw   m1, m3
-
-    movu        [r2 +  r3 * 0], m1
-    movu        [r2 +  r3 * 0 + 16], m0
-
-    movu        m0, [r0 + 16]
-    punpcklbw   m1, m0, m2
-    punpckhbw   m0, m2
-    pmaddubsw   m0, m3
-    pmaddubsw   m1, m3
-
-    movu        [r2 +  r3 * 0 + 32], m1
-    movu        [r2 +  r3 * 0 + 48], m0
-
-    movu        m0, [r0 + 32]
-    punpcklbw   m1, m0, m2
-    punpckhbw   m0, m2
-    pmaddubsw   m0, m3
-    pmaddubsw   m1, m3
-
-    movu        [r2 +  r3 * 0 + 64], m1
-    movu        [r2 +  r3 * 0 + 80], m0
-
-    movu        m0, [r0 + r1]
-    punpcklbw   m1, m0, m2
-    punpckhbw   m0, m2
-    pmaddubsw   m0, m3
-    pmaddubsw   m1, m3
-
-    movu        [r2 +  r3 * 1], m1
-    movu        [r2 +  r3 * 1 + 16], m0
-
-    movu        m0, [r0 + r1 + 16]
-    punpcklbw   m1, m0, m2
-    punpckhbw   m0, m2
-    pmaddubsw   m0, m3
-    pmaddubsw   m1, m3
-
-    movu        [r2 +  r3 * 1 + 32], m1
-    movu        [r2 +  r3 * 1 + 48], m0
-
-    movu        m0, [r0 + r1 + 32]
-    punpcklbw   m1, m0, m2
-    punpckhbw   m0, m2
-    pmaddubsw   m0, m3
-    pmaddubsw   m1, m3
-
-    movu        [r2 +  r3 * 1 + 64], m1
-    movu        [r2 +  r3 * 1 + 80], m0
-
-    movu        m0, [r0 + r1 * 2]
-    punpcklbw   m1, m0, m2
-    punpckhbw   m0, m2
-    pmaddubsw   m0, m3
-    pmaddubsw   m1, m3
-
-    movu        [r2 +  r3 * 2], m1
-    movu        [r2 +  r3 * 2 + 16], m0
-
-    movu        m0, [r0 + r1 * 2 + 16]
-    punpcklbw   m1, m0, m2
-    punpckhbw   m0, m2
-    pmaddubsw   m0, m3
-    pmaddubsw   m1, m3
-
-    movu        [r2 +  r3 * 2 + 32], m1
-    movu        [r2 +  r3 * 2 + 48], m0
-
-    movu        m0, [r0 + r1 * 2 + 32]
-    punpcklbw   m1, m0, m2
-    punpckhbw   m0, m2
-    pmaddubsw   m0, m3
-    pmaddubsw   m1, m3
-
-    movu        [r2 +  r3 * 2 + 64], m1
-    movu        [r2 +  r3 * 2 + 80], m0
-
-    movu        m0, [r0 + r4]
-    punpcklbw   m1, m0, m2
-    punpckhbw   m0, m2
-    pmaddubsw   m0, m3
-    pmaddubsw   m1, m3
-
-    movu        [r2 +  r5], m1
-    movu        [r2 +  r5 + 16], m0
-
-    movu        m0, [r0 + r4 + 16]
-    punpcklbw   m1, m0, m2
-    punpckhbw   m0, m2
-    pmaddubsw   m0, m3
-    pmaddubsw   m1, m3
-
-    movu        [r2 +  r5 + 32], m1
-    movu        [r2 +  r5 + 48], m0
-
-    movu        m0, [r0 + r4 + 32]
-    punpcklbw   m1, m0, m2
-    punpckhbw   m0, m2
-    pmaddubsw   m0, m3
-    pmaddubsw   m1, m3
-
-    movu        [r2 +  r5 + 64], m1
-    movu        [r2 +  r5 + 80], m0
-
-    lea         r0, [r0 + r1 * 4]
-    lea         r2, [r2 + r3 * 4]
-
-    dec         r6d
-    jnz         .loop
-    RET
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-INIT_YMM avx2
-cglobal filterPixelToShort_48x64, 3,7,4
-    mov         r3d, r3m
-    add         r3d, r3d
-    lea         r5, [r1 * 3]
-    lea         r6, [r3 * 3]
-
-    ; load height
-    mov         r4d, 64/4
-
-    ; load constant
-    vpbroadcastd m3, [pw_2000]
-
-    ; just unroll(1) because it is best choice for 48x64
-.loop:
-    pmovzxbw    m0, [r0 + 0 * mmsize/2]
-    pmovzxbw    m1, [r0 + 1 * mmsize/2]
-    pmovzxbw    m2, [r0 + 2 * mmsize/2]
-    psllw       m0, 6
-    psllw       m1, 6
-    psllw       m2, 6
-    psubw       m0, m3
-    psubw       m1, m3
-    psubw       m2, m3
-    movu        [r2 + 0 * mmsize], m0
-    movu        [r2 + 1 * mmsize], m1
-    movu        [r2 + 2 * mmsize], m2
-
-    pmovzxbw    m0, [r0 + r1 + 0 * mmsize/2]
-    pmovzxbw    m1, [r0 + r1 + 1 * mmsize/2]
-    pmovzxbw    m2, [r0 + r1 + 2 * mmsize/2]
-    psllw       m0, 6
-    psllw       m1, 6
-    psllw       m2, 6
-    psubw       m0, m3
-    psubw       m1, m3
-    psubw       m2, m3
-    movu        [r2 + r3 + 0 * mmsize], m0
-    movu        [r2 + r3 + 1 * mmsize], m1
-    movu        [r2 + r3 + 2 * mmsize], m2
-
-    pmovzxbw    m0, [r0 + r1 * 2 + 0 * mmsize/2]
-    pmovzxbw    m1, [r0 + r1 * 2 + 1 * mmsize/2]
-    pmovzxbw    m2, [r0 + r1 * 2 + 2 * mmsize/2]
-    psllw       m0, 6
-    psllw       m1, 6
-    psllw       m2, 6
-    psubw       m0, m3
-    psubw       m1, m3
-    psubw       m2, m3
-    movu        [r2 + r3 * 2 + 0 * mmsize], m0
-    movu        [r2 + r3 * 2 + 1 * mmsize], m1
-    movu        [r2 + r3 * 2 + 2 * mmsize], m2
-
-    pmovzxbw    m0, [r0 + r5 + 0 * mmsize/2]
-    pmovzxbw    m1, [r0 + r5 + 1 * mmsize/2]
-    pmovzxbw    m2, [r0 + r5 + 2 * mmsize/2]
-    psllw       m0, 6
-    psllw       m1, 6
-    psllw       m2, 6
-    psubw       m0, m3
-    psubw       m1, m3
-    psubw       m2, m3
-    movu        [r2 + r6 + 0 * mmsize], m0
-    movu        [r2 + r6 + 1 * mmsize], m1
-    movu        [r2 + r6 + 2 * mmsize], m2
-
-    lea         r0, [r0 + r1 * 4]
-    lea         r2, [r2 + r3 * 4]
-
-    dec         r4d
-    jnz        .loop
-    RET
-
-
-%macro PROCESS_LUMA_W4_4R 0
-    movd        m0, [r0]
-    movd        m1, [r0 + r1]
-    punpcklbw   m2, m0, m1                     ; m2=[0 1]
-
-    lea         r0, [r0 + 2 * r1]
-    movd        m0, [r0]
-    punpcklbw   m1, m0                         ; m1=[1 2]
-    punpcklqdq  m2, m1                         ; m2=[0 1 1 2]
-    pmaddubsw   m4, m2, [r6 + 0 * 16]          ; m4=[0+1 1+2]
-
-    movd        m1, [r0 + r1]
-    punpcklbw   m5, m0, m1                     ; m2=[2 3]
-    lea         r0, [r0 + 2 * r1]
-    movd        m0, [r0]
-    punpcklbw   m1, m0                         ; m1=[3 4]
-    punpcklqdq  m5, m1                         ; m5=[2 3 3 4]
-    pmaddubsw   m2, m5, [r6 + 1 * 16]          ; m2=[2+3 3+4]
-    paddw       m4, m2                         ; m4=[0+1+2+3 1+2+3+4]                   Row1-2
-    pmaddubsw   m5, [r6 + 0 * 16]              ; m5=[2+3 3+4]                           Row3-4
-
-    movd        m1, [r0 + r1]
-    punpcklbw   m2, m0, m1                     ; m2=[4 5]
-    lea         r0, [r0 + 2 * r1]
-    movd        m0, [r0]
-    punpcklbw   m1, m0                         ; m1=[5 6]
-    punpcklqdq  m2, m1                         ; m2=[4 5 5 6]
-    pmaddubsw   m1, m2, [r6 + 2 * 16]          ; m1=[4+5 5+6]
-    paddw       m4, m1                         ; m4=[0+1+2+3+4+5 1+2+3+4+5+6]           Row1-2
-    pmaddubsw   m2, [r6 + 1 * 16]              ; m2=[4+5 5+6]
-    paddw       m5, m2                         ; m5=[2+3+4+5 3+4+5+6]                   Row3-4
-
-    movd        m1, [r0 + r1]
-    punpcklbw   m2, m0, m1                     ; m2=[6 7]
-    lea         r0, [r0 + 2 * r1]
-    movd        m0, [r0]
-    punpcklbw   m1, m0                         ; m1=[7 8]
-    punpcklqdq  m2, m1                         ; m2=[6 7 7 8]
-    pmaddubsw   m1, m2, [r6 + 3 * 16]          ; m1=[6+7 7+8]
-    paddw       m4, m1                         ; m4=[0+1+2+3+4+5+6+7 1+2+3+4+5+6+7+8]   Row1-2 end
-    pmaddubsw   m2, [r6 + 2 * 16]              ; m2=[6+7 7+8]
-    paddw       m5, m2                         ; m5=[2+3+4+5+6+7 3+4+5+6+7+8]           Row3-4
-
-    movd        m1, [r0 + r1]
-    punpcklbw   m2, m0, m1                     ; m2=[8 9]
-    movd        m0, [r0 + 2 * r1]
-    punpcklbw   m1, m0                         ; m1=[9 10]
-    punpcklqdq  m2, m1                         ; m2=[8 9 9 10]
-    pmaddubsw   m2, [r6 + 3 * 16]              ; m2=[8+9 9+10]
-    paddw       m5, m2                         ; m5=[2+3+4+5+6+7+8+9 3+4+5+6+7+8+9+10]  Row3-4 end
-%endmacro
-
-%macro PROCESS_LUMA_W8_4R 0
-    movq       m0, [r0]
-    movq       m1, [r0 + r1]
-    punpcklbw  m0, m1
-    pmaddubsw  m7, m0, [r6 + 0 *16]            ;m7=[0+1]               Row1
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m0, [r0]
-    punpcklbw  m1, m0
-    pmaddubsw  m6, m1, [r6 + 0 *16]            ;m6=[1+2]               Row2
-
-    movq       m1, [r0 + r1]
-    punpcklbw  m0, m1
-    pmaddubsw  m5, m0, [r6 + 0 *16]            ;m5=[2+3]               Row3
-    pmaddubsw  m0, [r6 + 1 * 16]
-    paddw      m7, m0                          ;m7=[0+1+2+3]           Row1
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m0, [r0]
-    punpcklbw  m1, m0
-    pmaddubsw  m4, m1, [r6 + 0 *16]            ;m4=[3+4]               Row4
-    pmaddubsw  m1, [r6 + 1 * 16]
-    paddw      m6, m1                          ;m6 = [1+2+3+4]         Row2
-
-    movq       m1, [r0 + r1]
-    punpcklbw  m0, m1
-    pmaddubsw  m2, m0, [r6 + 1 * 16]
-    pmaddubsw  m0, [r6 + 2 * 16]
-    paddw      m7, m0                          ;m7=[0+1+2+3+4+5]       Row1
-    paddw      m5, m2                          ;m5=[2+3+4+5]           Row3
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m0, [r0]
-    punpcklbw  m1, m0
-    pmaddubsw  m2, m1, [r6 + 1 * 16]
-    pmaddubsw  m1, [r6 + 2 * 16]
-    paddw      m6, m1                          ;m6=[1+2+3+4+5+6]       Row2
-    paddw      m4, m2                          ;m4=[3+4+5+6]           Row4
-
-    movq       m1, [r0 + r1]
-    punpcklbw  m0, m1
-    pmaddubsw  m2, m0, [r6 + 2 * 16]
-    pmaddubsw  m0, [r6 + 3 * 16]
-    paddw      m7, m0                          ;m7=[0+1+2+3+4+5+6+7]   Row1 end
-    paddw      m5, m2                          ;m5=[2+3+4+5+6+7]       Row3
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m0, [r0]
-    punpcklbw  m1, m0
-    pmaddubsw  m2, m1, [r6 + 2 * 16]
-    pmaddubsw  m1, [r6 + 3 * 16]
-    paddw      m6, m1                          ;m6=[1+2+3+4+5+6+7+8]   Row2 end
-    paddw      m4, m2                          ;m4=[3+4+5+6+7+8]       Row4
-
-    movq       m1, [r0 + r1]
-    punpcklbw  m0, m1
-    pmaddubsw  m0, [r6 + 3 * 16]
-    paddw      m5, m0                          ;m5=[2+3+4+5+6+7+8+9]   Row3 end
-
-    movq       m0, [r0 + 2 * r1]
-    punpcklbw  m1, m0
-    pmaddubsw  m1, [r6 + 3 * 16]
-    paddw      m4, m1                          ;m4=[3+4+5+6+7+8+9+10]  Row4 end
-%endmacro
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_%3_4x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_LUMA_4xN 3
-INIT_XMM sse4
-cglobal interp_8tap_vert_%3_%1x%2, 5, 7, 6
-    lea       r5, [3 * r1]
-    sub       r0, r5
-    shl       r4d, 6
-%ifidn %3,ps
-    add       r3d, r3d
-%endif
-
-%ifdef PIC
-    lea       r5, [tab_LumaCoeffVer]
-    lea       r6, [r5 + r4]
-%else
-    lea       r6, [tab_LumaCoeffVer + r4]
-%endif
-
-%ifidn %3,pp
-    mova      m3, [pw_512]
-%else
-    mova      m3, [pw_2000]
-%endif
-
-    mov       r4d, %2/4
-    lea       r5, [4 * r1]
-
-.loopH:
-    PROCESS_LUMA_W4_4R
-
-%ifidn %3,pp
-    pmulhrsw  m4, m3
-    pmulhrsw  m5, m3
-
-    packuswb  m4, m5
-
-    movd      [r2], m4
-    pextrd    [r2 + r3], m4, 1
-    lea       r2, [r2 + 2 * r3]
-    pextrd    [r2], m4, 2
-    pextrd    [r2 + r3], m4, 3
-%else
-    psubw     m4, m3
-    psubw     m5, m3
-
-    movlps    [r2], m4
-    movhps    [r2 + r3], m4
-    lea       r2, [r2 + 2 * r3]
-    movlps    [r2], m5
-    movhps    [r2 + r3], m5
-%endif
-
-    sub       r0, r5
-    lea       r2, [r2 + 2 * r3]
-
-    dec       r4d
-    jnz       .loopH
-
-    RET
-%endmacro
-
-
-INIT_YMM avx2
-cglobal interp_8tap_vert_pp_4x4, 4,6,8
-    mov             r4d, r4m
-    lea             r5, [r1 * 3]
-    sub             r0, r5
-
-    ; TODO: VPGATHERDD
-    movd            xm1, [r0]                       ; m1 = row0
-    movd            xm2, [r0 + r1]                  ; m2 = row1
-    punpcklbw       xm1, xm2                        ; m1 = [13 03 12 02 11 01 10 00]
-
-    movd            xm3, [r0 + r1 * 2]              ; m3 = row2
-    punpcklbw       xm2, xm3                        ; m2 = [23 13 22 12 21 11 20 10]
-    movd            xm4, [r0 + r5]
-    punpcklbw       xm3, xm4                        ; m3 = [33 23 32 22 31 21 30 20]
-    punpcklwd       xm1, xm3                        ; m1 = [33 23 13 03 32 22 12 02 31 21 11 01 30 20 10 00]
-
-    lea             r0, [r0 + r1 * 4]
-    movd            xm5, [r0]                       ; m5 = row4
-    punpcklbw       xm4, xm5                        ; m4 = [43 33 42 32 41 31 40 30]
-    punpcklwd       xm2, xm4                        ; m2 = [43 33 21 13 42 32 22 12 41 31 21 11 40 30 20 10]
-    vinserti128     m1, m1, xm2, 1                  ; m1 = [43 33 21 13 42 32 22 12 41 31 21 11 40 30 20 10] - [33 23 13 03 32 22 12 02 31 21 11 01 30 20 10 00]
-    movd            xm2, [r0 + r1]                  ; m2 = row5
-    punpcklbw       xm5, xm2                        ; m5 = [53 43 52 42 51 41 50 40]
-    punpcklwd       xm3, xm5                        ; m3 = [53 43 44 23 52 42 32 22 51 41 31 21 50 40 30 20]
-    movd            xm6, [r0 + r1 * 2]              ; m6 = row6
-    punpcklbw       xm2, xm6                        ; m2 = [63 53 62 52 61 51 60 50]
-    punpcklwd       xm4, xm2                        ; m4 = [63 53 43 33 62 52 42 32 61 51 41 31 60 50 40 30]
-    vinserti128     m3, m3, xm4, 1                  ; m3 = [63 53 43 33 62 52 42 32 61 51 41 31 60 50 40 30] - [53 43 44 23 52 42 32 22 51 41 31 21 50 40 30 20]
-    movd            xm4, [r0 + r5]                  ; m4 = row7
-    punpcklbw       xm6, xm4                        ; m6 = [73 63 72 62 71 61 70 60]
-    punpcklwd       xm5, xm6                        ; m5 = [73 63 53 43 72 62 52 42 71 61 51 41 70 60 50 40]
-
-    lea             r0, [r0 + r1 * 4]
-    movd            xm7, [r0]                       ; m7 = row8
-    punpcklbw       xm4, xm7                        ; m4 = [83 73 82 72 81 71 80 70]
-    punpcklwd       xm2, xm4                        ; m2 = [83 73 63 53 82 72 62 52 81 71 61 51 80 70 60 50]
-    vinserti128     m5, m5, xm2, 1                  ; m5 = [83 73 63 53 82 72 62 52 81 71 61 51 80 70 60 50] - [73 63 53 43 72 62 52 42 71 61 51 41 70 60 50 40]
-    movd            xm2, [r0 + r1]                  ; m2 = row9
-    punpcklbw       xm7, xm2                        ; m7 = [93 83 92 82 91 81 90 80]
-    punpcklwd       xm6, xm7                        ; m6 = [93 83 73 63 92 82 72 62 91 81 71 61 90 80 70 60]
-    movd            xm7, [r0 + r1 * 2]              ; m7 = rowA
-    punpcklbw       xm2, xm7                        ; m2 = [A3 93 A2 92 A1 91 A0 90]
-    punpcklwd       xm4, xm2                        ; m4 = [A3 93 83 73 A2 92 82 72 A1 91 81 71 A0 90 80 70]
-    vinserti128     m6, m6, xm4, 1                  ; m6 = [A3 93 83 73 A2 92 82 72 A1 91 81 71 A0 90 80 70] - [93 83 73 63 92 82 72 62 91 81 71 61 90 80 70 60]
-
-    ; load filter coeff
-%ifdef PIC
-    lea             r5, [tab_LumaCoeff]
-    vpbroadcastd    m0, [r5 + r4 * 8 + 0]
-    vpbroadcastd    m2, [r5 + r4 * 8 + 4]
-%else
-    vpbroadcastd    m0, [tab_LumaCoeff + r4 * 8 + 0]
-    vpbroadcastd    m2, [tab_LumaCoeff + r4 * 8 + 4]
-%endif
-
-    pmaddubsw       m1, m0
-    pmaddubsw       m3, m0
-    pmaddubsw       m5, m2
-    pmaddubsw       m6, m2
-    vbroadcasti128  m0, [pw_1]
-    pmaddwd         m1, m0
-    pmaddwd         m3, m0
-    pmaddwd         m5, m0
-    pmaddwd         m6, m0
-    paddd           m1, m5                          ; m1 = DQWORD ROW[1 0]
-    paddd           m3, m6                          ; m3 = DQWORD ROW[3 2]
-    packssdw        m1, m3                          ; m1 =  QWORD ROW[3 1 2 0]
-
-    ; TODO: does it overflow?
-    pmulhrsw        m1, [pw_512]
-    vextracti128    xm2, m1, 1
-    packuswb        xm1, xm2                        ; m1 =  DWORD ROW[3 1 2 0]
-    movd            [r2], xm1
-    pextrd          [r2 + r3], xm1, 2
-    pextrd          [r2 + r3 * 2], xm1, 1
-    lea             r4, [r3 * 3]
-    pextrd          [r2 + r4], xm1, 3
-    RET
-
-INIT_YMM avx2
-cglobal interp_8tap_vert_ps_4x4, 4, 6, 5
-    mov             r4d, r4m
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-
-    add             r3d, r3d
-
-    movd            xm1, [r0]
-    pinsrd          xm1, [r0 + r1], 1
-    pinsrd          xm1, [r0 + r1 * 2], 2
-    pinsrd          xm1, [r0 + r4], 3                       ; m1 = row[3 2 1 0]
-    lea             r0, [r0 + r1 * 4]
-    movd            xm2, [r0]
-    pinsrd          xm2, [r0 + r1], 1
-    pinsrd          xm2, [r0 + r1 * 2], 2
-    pinsrd          xm2, [r0 + r4], 3                       ; m2 = row[7 6 5 4]
-    vinserti128     m1, m1, xm2, 1                          ; m1 = row[7 6 5 4 3 2 1 0]
-    lea             r0, [r0 + r1 * 4]
-    movd            xm3, [r0]
-    pinsrd          xm3, [r0 + r1], 1
-    pinsrd          xm3, [r0 + r1 * 2], 2                   ; m3 = row[x 10 9 8]
-    vinserti128     m2, m2, xm3, 1                          ; m2 = row[x 10 9 8 7 6 5 4]
-    mova            m3, [interp4_vpp_shuf1]
-    vpermd          m0, m3, m1                              ; m0 = row[4 3 3 2 2 1 1 0]
-    vpermd          m4, m3, m2                              ; m4 = row[8 7 7 6 6 5 5 4]
-    mova            m3, [interp4_vpp_shuf1 + mmsize]
-    vpermd          m1, m3, m1                              ; m1 = row[6 5 5 4 4 3 3 2]
-    vpermd          m2, m3, m2                              ; m2 = row[10 9 9 8 8 7 7 6]
-
-    mova            m3, [interp4_vpp_shuf]
-    pshufb          m0, m0, m3
-    pshufb          m1, m1, m3
-    pshufb          m4, m4, m3
-    pshufb          m2, m2, m3
-    pmaddubsw       m0, [r5]
-    pmaddubsw       m1, [r5 + mmsize]
-    pmaddubsw       m4, [r5 + 2 * mmsize]
-    pmaddubsw       m2, [r5 + 3 * mmsize]
-    paddw           m0, m1
-    paddw           m0, m4
-    paddw           m0, m2                                  ; m0 = WORD ROW[3 2 1 0]
-
-    psubw           m0, [pw_2000]
-    vextracti128    xm2, m0, 1
-    lea             r5, [r3 * 3]
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r5], xm2
-    RET
-
-%macro FILTER_VER_LUMA_AVX2_4xN 3
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%3_%1x%2, 4, 9, 10
-    mov             r4d, r4m
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-    lea             r6, [r1 * 4]
-%ifidn %3,pp
-    mova            m6, [pw_512]
-%else
-    add             r3d, r3d
-    vbroadcasti128  m6, [pw_2000]
-%endif
-    lea             r8, [r3 * 3]
-    mova            m5, [interp4_vpp_shuf]
-    mova            m0, [interp4_vpp_shuf1]
-    mova            m7, [interp4_vpp_shuf1 + mmsize]
-    mov             r7d, %2 / 8
-.loop:
-    movd            xm1, [r0]
-    pinsrd          xm1, [r0 + r1], 1
-    pinsrd          xm1, [r0 + r1 * 2], 2
-    pinsrd          xm1, [r0 + r4], 3                       ; m1 = row[3 2 1 0]
-    lea             r0, [r0 + r1 * 4]
-    movd            xm2, [r0]
-    pinsrd          xm2, [r0 + r1], 1
-    pinsrd          xm2, [r0 + r1 * 2], 2
-    pinsrd          xm2, [r0 + r4], 3                       ; m2 = row[7 6 5 4]
-    vinserti128     m1, m1, xm2, 1                          ; m1 = row[7 6 5 4 3 2 1 0]
-    lea             r0, [r0 + r1 * 4]
-    movd            xm3, [r0]
-    pinsrd          xm3, [r0 + r1], 1
-    pinsrd          xm3, [r0 + r1 * 2], 2
-    pinsrd          xm3, [r0 + r4], 3                       ; m3 = row[11 10 9 8]
-    vinserti128     m2, m2, xm3, 1                          ; m2 = row[11 10 9 8 7 6 5 4]
-    lea             r0, [r0 + r1 * 4]
-    movd            xm4, [r0]
-    pinsrd          xm4, [r0 + r1], 1
-    pinsrd          xm4, [r0 + r1 * 2], 2                   ; m4 = row[x 14 13 12]
-    vinserti128     m3, m3, xm4, 1                          ; m3 = row[x 14 13 12 11 10 9 8]
-    vpermd          m8, m0, m1                              ; m8 = row[4 3 3 2 2 1 1 0]
-    vpermd          m4, m0, m2                              ; m4 = row[8 7 7 6 6 5 5 4]
-    vpermd          m1, m7, m1                              ; m1 = row[6 5 5 4 4 3 3 2]
-    vpermd          m2, m7, m2                              ; m2 = row[10 9 9 8 8 7 7 6]
-    vpermd          m9, m0, m3                              ; m9 = row[12 11 11 10 10 9 9 8]
-    vpermd          m3, m7, m3                              ; m3 = row[14 13 13 12 12 11 11 10]
-
-    pshufb          m8, m8, m5
-    pshufb          m1, m1, m5
-    pshufb          m4, m4, m5
-    pshufb          m9, m9, m5
-    pshufb          m2, m2, m5
-    pshufb          m3, m3, m5
-    pmaddubsw       m8, [r5]
-    pmaddubsw       m1, [r5 + mmsize]
-    pmaddubsw       m9, [r5 + 2 * mmsize]
-    pmaddubsw       m3, [r5 + 3 * mmsize]
-    paddw           m8, m1
-    paddw           m9, m3
-    pmaddubsw       m1, m4, [r5 + 2 * mmsize]
-    pmaddubsw       m3, m2, [r5 + 3 * mmsize]
-    pmaddubsw       m4, [r5]
-    pmaddubsw       m2, [r5 + mmsize]
-    paddw           m3, m1
-    paddw           m2, m4
-    paddw           m8, m3                                  ; m8 = WORD ROW[3 2 1 0]
-    paddw           m9, m2                                  ; m9 = WORD ROW[7 6 5 4]
-
-%ifidn %3,pp
-    pmulhrsw        m8, m6
-    pmulhrsw        m9, m6
-    packuswb        m8, m9
-    vextracti128    xm1, m8, 1
-    movd            [r2], xm8
-    pextrd          [r2 + r3], xm8, 1
-    movd            [r2 + r3 * 2], xm1
-    pextrd          [r2 + r8], xm1, 1
-    lea             r2, [r2 + r3 * 4]
-    pextrd          [r2], xm8, 2
-    pextrd          [r2 + r3], xm8, 3
-    pextrd          [r2 + r3 * 2], xm1, 2
-    pextrd          [r2 + r8], xm1, 3
-%else
-    psubw           m8, m6
-    psubw           m9, m6
-    vextracti128    xm1, m8, 1
-    vextracti128    xm2, m9, 1
-    movq            [r2], xm8
-    movhps          [r2 + r3], xm8
-    movq            [r2 + r3 * 2], xm1
-    movhps          [r2 + r8], xm1
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm9
-    movhps          [r2 + r3], xm9
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r8], xm2
-%endif
-    lea             r2, [r2 + r3 * 4]
-    sub             r0, r6
-    dec             r7d
-    jnz             .loop
-    RET
-%endif
-%endmacro
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_pp_4x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_4xN 4, 4, pp
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_pp_4x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_4xN 4, 8, pp
-    FILTER_VER_LUMA_AVX2_4xN 4, 8, pp
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_pp_4x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_4xN 4, 16, pp
-    FILTER_VER_LUMA_AVX2_4xN 4, 16, pp
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_ps_4x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_4xN 4, 4, ps
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_ps_4x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_4xN 4, 8, ps
-    FILTER_VER_LUMA_AVX2_4xN 4, 8, ps
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_ps_4x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_4xN 4, 16, ps
-    FILTER_VER_LUMA_AVX2_4xN 4, 16, ps
-
-%macro PROCESS_LUMA_AVX2_W8_8R 0
-    movq            xm1, [r0]                       ; m1 = row 0
-    movq            xm2, [r0 + r1]                  ; m2 = row 1
-    punpcklbw       xm1, xm2                        ; m1 = [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
-    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
-    punpcklbw       xm2, xm3                        ; m2 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10]
-    vinserti128     m5, m1, xm2, 1                  ; m5 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10] - [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
-    pmaddubsw       m5, [r5]
-    movq            xm4, [r0 + r4]                  ; m4 = row 3
-    punpcklbw       xm3, xm4                        ; m3 = [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
-    lea             r0, [r0 + r1 * 4]
-    movq            xm1, [r0]                       ; m1 = row 4
-    punpcklbw       xm4, xm1                        ; m4 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30]
-    vinserti128     m2, m3, xm4, 1                  ; m2 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30] - [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
-    pmaddubsw       m0, m2, [r5 + 1 * mmsize]
-    paddw           m5, m0
-    pmaddubsw       m2, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 5
-    punpcklbw       xm1, xm3                        ; m1 = [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
-    movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
-    punpcklbw       xm3, xm4                        ; m3 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50]
-    vinserti128     m1, m1, xm3, 1                  ; m1 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50] - [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
-    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
-    paddw           m5, m3
-    pmaddubsw       m0, m1, [r5 + 1 * mmsize]
-    paddw           m2, m0
-    pmaddubsw       m1, [r5]
-    movq            xm3, [r0 + r4]                  ; m3 = row 7
-    punpcklbw       xm4, xm3                        ; m4 = [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
-    lea             r0, [r0 + r1 * 4]
-    movq            xm0, [r0]                       ; m0 = row 8
-    punpcklbw       xm3, xm0                        ; m3 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70]
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70] - [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
-    pmaddubsw       m3, m4, [r5 + 3 * mmsize]
-    paddw           m5, m3
-    pmaddubsw       m3, m4, [r5 + 2 * mmsize]
-    paddw           m2, m3
-    pmaddubsw       m3, m4, [r5 + 1 * mmsize]
-    paddw           m1, m3
-    pmaddubsw       m4, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 9
-    punpcklbw       xm0, xm3                        ; m0 = [97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80]
-    movq            xm6, [r0 + r1 * 2]              ; m6 = row 10
-    punpcklbw       xm3, xm6                        ; m3 = [A7 97 A6 96 A5 95 A4 94 A3 93 A2 92 A1 91 A0 90]
-    vinserti128     m0, m0, xm3, 1                  ; m0 = [A7 97 A6 96 A5 95 A4 94 A3 93 A2 92 A1 91 A0 90] - [97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80]
-    pmaddubsw       m3, m0, [r5 + 3 * mmsize]
-    paddw           m2, m3
-    pmaddubsw       m3, m0, [r5 + 2 * mmsize]
-    paddw           m1, m3
-    pmaddubsw       m0, [r5 + 1 * mmsize]
-    paddw           m4, m0
-
-    movq            xm3, [r0 + r4]                  ; m3 = row 11
-    punpcklbw       xm6, xm3                        ; m6 = [B7 A7 B6 A6 B5 A5 B4 A4 B3 A3 B2 A2 B1 A1 B0 A0]
-    lea             r0, [r0 + r1 * 4]
-    movq            xm0, [r0]                       ; m0 = row 12
-    punpcklbw       xm3, xm0                        ; m3 = [C7 B7 C6 B6 C5 B5 C4 B4 C3 B3 C2 B2 C1 B1 C0 B0]
-    vinserti128     m6, m6, xm3, 1                  ; m6 = [C7 B7 C6 B6 C5 B5 C4 B4 C3 B3 C2 B2 C1 B1 C0 B0] - [B7 A7 B6 A6 B5 A5 B4 A4 B3 A3 B2 A2 B1 A1 B0 A0]
-    pmaddubsw       m3, m6, [r5 + 3 * mmsize]
-    paddw           m1, m3
-    pmaddubsw       m6, [r5 + 2 * mmsize]
-    paddw           m4, m6
-    movq            xm3, [r0 + r1]                  ; m3 = row 13
-    punpcklbw       xm0, xm3                        ; m0 = [D7 C7 D6 C6 D5 C5 D4 C4 D3 C3 D2 C2 D1 C1 D0 C0]
-    movq            xm6, [r0 + r1 * 2]              ; m6 = row 14
-    punpcklbw       xm3, xm6                        ; m3 = [E7 D7 E6 D6 E5 D5 E4 D4 E3 D3 E2 D2 E1 D1 E0 D0]
-    vinserti128     m0, m0, xm3, 1                  ; m0 = [E7 D7 E6 D6 E5 D5 E4 D4 E3 D3 E2 D2 E1 D1 E0 D0] - [D7 C7 D6 C6 D5 C5 D4 C4 D3 C3 D2 C2 D1 C1 D0 C0]
-    pmaddubsw       m0, [r5 + 3 * mmsize]
-    paddw           m4, m0
-%endmacro
-
-%macro PROCESS_LUMA_AVX2_W8_4R 0
-    movq            xm1, [r0]                       ; m1 = row 0
-    movq            xm2, [r0 + r1]                  ; m2 = row 1
-    punpcklbw       xm1, xm2                        ; m1 = [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
-    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
-    punpcklbw       xm2, xm3                        ; m2 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10]
-    vinserti128     m5, m1, xm2, 1                  ; m5 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10] - [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
-    pmaddubsw       m5, [r5]
-    movq            xm4, [r0 + r4]                  ; m4 = row 3
-    punpcklbw       xm3, xm4                        ; m3 = [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
-    lea             r0, [r0 + r1 * 4]
-    movq            xm1, [r0]                       ; m1 = row 4
-    punpcklbw       xm4, xm1                        ; m4 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30]
-    vinserti128     m2, m3, xm4, 1                  ; m2 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30] - [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
-    pmaddubsw       m0, m2, [r5 + 1 * mmsize]
-    paddw           m5, m0
-    pmaddubsw       m2, [r5]
-    movq            xm3, [r0 + r1]                  ; m3 = row 5
-    punpcklbw       xm1, xm3                        ; m1 = [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
-    movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
-    punpcklbw       xm3, xm4                        ; m3 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50]
-    vinserti128     m1, m1, xm3, 1                  ; m1 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50] - [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
-    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
-    paddw           m5, m3
-    pmaddubsw       m0, m1, [r5 + 1 * mmsize]
-    paddw           m2, m0
-    movq            xm3, [r0 + r4]                  ; m3 = row 7
-    punpcklbw       xm4, xm3                        ; m4 = [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
-    lea             r0, [r0 + r1 * 4]
-    movq            xm0, [r0]                       ; m0 = row 8
-    punpcklbw       xm3, xm0                        ; m3 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70]
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70] - [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
-    pmaddubsw       m3, m4, [r5 + 3 * mmsize]
-    paddw           m5, m3
-    pmaddubsw       m3, m4, [r5 + 2 * mmsize]
-    paddw           m2, m3
-    movq            xm3, [r0 + r1]                  ; m3 = row 9
-    punpcklbw       xm0, xm3                        ; m0 = [97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80]
-    movq            xm6, [r0 + r1 * 2]              ; m6 = row 10
-    punpcklbw       xm3, xm6                        ; m3 = [A7 97 A6 96 A5 95 A4 94 A3 93 A2 92 A1 91 A0 90]
-    vinserti128     m0, m0, xm3, 1                  ; m0 = [A7 97 A6 96 A5 95 A4 94 A3 93 A2 92 A1 91 A0 90] - [97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80]
-    pmaddubsw       m3, m0, [r5 + 3 * mmsize]
-    paddw           m2, m3
-%endmacro
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_%3_8x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_LUMA_8xN 3
-INIT_XMM sse4
-cglobal interp_8tap_vert_%3_%1x%2, 5, 7, 8
-    lea       r5, [3 * r1]
-    sub       r0, r5
-    shl       r4d, 6
-
-%ifidn %3,ps
-    add       r3d, r3d
-%endif
-
-%ifdef PIC
-    lea       r5, [tab_LumaCoeffVer]
-    lea       r6, [r5 + r4]
-%else
-    lea       r6, [tab_LumaCoeffVer + r4]
-%endif
-
- %ifidn %3,pp
-    mova      m3, [pw_512]
-%else
-    mova      m3, [pw_2000]
-%endif
-
-    mov       r4d, %2/4
-    lea       r5, [4 * r1]
-
-.loopH:
-    PROCESS_LUMA_W8_4R
-
-%ifidn %3,pp
-    pmulhrsw  m7, m3
-    pmulhrsw  m6, m3
-    pmulhrsw  m5, m3
-    pmulhrsw  m4, m3
-
-    packuswb  m7, m6
-    packuswb  m5, m4
-
-    movlps    [r2], m7
-    movhps    [r2 + r3], m7
-    lea       r2, [r2 + 2 * r3]
-    movlps    [r2], m5
-    movhps    [r2 + r3], m5
-%else
-    psubw     m7, m3
-    psubw     m6, m3
-    psubw     m5, m3
-    psubw     m4, m3
-
-    movu      [r2], m7
-    movu      [r2 + r3], m6
-    lea       r2, [r2 + 2 * r3]
-    movu      [r2], m5
-    movu      [r2 + r3], m4
-%endif
-
-    sub       r0, r5
-    lea       r2, [r2 + 2 * r3]
-
-    dec       r4d
-    jnz       .loopH
-
-    RET
-%endmacro
-
-%macro FILTER_VER_LUMA_AVX2_8xN 3
-INIT_YMM avx2
-cglobal interp_8tap_vert_%3_%1x%2, 4, 7, 8, 0-gprsize
-    mov             r4d, r4m
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-    lea             r6, [r1 * 4]
-%ifidn %3,pp
-    mova            m7, [pw_512]
-%else
-    add             r3d, r3d
-    vbroadcasti128  m7, [pw_2000]
-%endif
-    mov             word [rsp], %2 / 8
-
-.loop:
-    PROCESS_LUMA_AVX2_W8_8R
-%ifidn %3,pp
-    pmulhrsw        m5, m7                          ; m5 = word: row 0, row 1
-    pmulhrsw        m2, m7                          ; m2 = word: row 2, row 3
-    pmulhrsw        m1, m7                          ; m1 = word: row 4, row 5
-    pmulhrsw        m4, m7                          ; m4 = word: row 6, row 7
-    packuswb        m5, m2
-    packuswb        m1, m4
-    vextracti128    xm2, m5, 1
-    vextracti128    xm4, m1, 1
-    movq            [r2], xm5
-    movq            [r2 + r3], xm2
-    lea             r2, [r2 + r3 * 2]
-    movhps          [r2], xm5
-    movhps          [r2 + r3], xm2
-    lea             r2, [r2 + r3 * 2]
-    movq            [r2], xm1
-    movq            [r2 + r3], xm4
-    lea             r2, [r2 + r3 * 2]
-    movhps          [r2], xm1
-    movhps          [r2 + r3], xm4
-%else
-    psubw           m5, m7                          ; m5 = word: row 0, row 1
-    psubw           m2, m7                          ; m2 = word: row 2, row 3
-    psubw           m1, m7                          ; m1 = word: row 4, row 5
-    psubw           m4, m7                          ; m4 = word: row 6, row 7
-    vextracti128    xm6, m5, 1
-    vextracti128    xm3, m2, 1
-    vextracti128    xm0, m1, 1
-    movu            [r2], xm5
-    movu            [r2 + r3], xm6
-    lea             r2, [r2 + r3 * 2]
-    movu            [r2], xm2
-    movu            [r2 + r3], xm3
-    lea             r2, [r2 + r3 * 2]
-    movu            [r2], xm1
-    movu            [r2 + r3], xm0
-    lea             r2, [r2 + r3 * 2]
-    movu            [r2], xm4
-    vextracti128    xm4, m4, 1
-    movu            [r2 + r3], xm4
-%endif
-    lea             r2, [r2 + r3 * 2]
-    sub             r0, r6
-    dec             word [rsp]
-    jnz             .loop
-    RET
-%endmacro
-
-%macro FILTER_VER_LUMA_AVX2_8x8 1
-INIT_YMM avx2
-cglobal interp_8tap_vert_%1_8x8, 4, 6, 7
-    mov             r4d, r4m
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-    PROCESS_LUMA_AVX2_W8_8R
-%ifidn %1,pp
-    mova            m3, [pw_512]
-%else
-    add             r3d, r3d
-    vbroadcasti128  m3, [pw_2000]
-%endif
-    lea             r4, [r3 * 3]
-%ifidn %1,pp
-    pmulhrsw        m5, m3                          ; m5 = word: row 0, row 1
-    pmulhrsw        m2, m3                          ; m2 = word: row 2, row 3
-    pmulhrsw        m1, m3                          ; m1 = word: row 4, row 5
-    pmulhrsw        m4, m3                          ; m4 = word: row 6, row 7
-    packuswb        m5, m2
-    packuswb        m1, m4
-    vextracti128    xm2, m5, 1
-    vextracti128    xm4, m1, 1
-    movq            [r2], xm5
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm5
-    movhps          [r2 + r4], xm2
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm1
-    movq            [r2 + r3], xm4
-    movhps          [r2 + r3 * 2], xm1
-    movhps          [r2 + r4], xm4
-%else
-    psubw           m5, m3                          ; m5 = word: row 0, row 1
-    psubw           m2, m3                          ; m2 = word: row 2, row 3
-    psubw           m1, m3                          ; m1 = word: row 4, row 5
-    psubw           m4, m3                          ; m4 = word: row 6, row 7
-    vextracti128    xm6, m5, 1
-    vextracti128    xm3, m2, 1
-    vextracti128    xm0, m1, 1
-    movu            [r2], xm5
-    movu            [r2 + r3], xm6
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r4], xm3
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm1
-    movu            [r2 + r3], xm0
-    movu            [r2 + r3 * 2], xm4
-    vextracti128    xm4, m4, 1
-    movu            [r2 + r4], xm4
-%endif
-    RET
-%endmacro
-
-%macro FILTER_VER_LUMA_AVX2_8x4 1
-INIT_YMM avx2
-cglobal interp_8tap_vert_%1_8x4, 4, 6, 7
-    mov             r4d, r4m
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-    PROCESS_LUMA_AVX2_W8_4R
-%ifidn %1,pp
-    mova            m3, [pw_512]
-%else
-    add             r3d, r3d
-    vbroadcasti128  m3, [pw_2000]
-%endif
-    lea             r4, [r3 * 3]
-%ifidn %1,pp
-    pmulhrsw        m5, m3                          ; m5 = word: row 0, row 1
-    pmulhrsw        m2, m3                          ; m2 = word: row 2, row 3
-    packuswb        m5, m2
-    vextracti128    xm2, m5, 1
-    movq            [r2], xm5
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm5
-    movhps          [r2 + r4], xm2
-%else
-    psubw           m5, m3                          ; m5 = word: row 0, row 1
-    psubw           m2, m3                          ; m2 = word: row 2, row 3
-    movu            [r2], xm5
-    vextracti128    xm5, m5, 1
-    movu            [r2 + r3], xm5
-    movu            [r2 + r3 * 2], xm2
-    vextracti128    xm2, m2, 1
-    movu            [r2 + r4], xm2
-%endif
-    RET
-%endmacro
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_pp_8x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_8xN 8, 4, pp
-    FILTER_VER_LUMA_AVX2_8x4 pp
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_pp_8x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_8xN 8, 8, pp
-    FILTER_VER_LUMA_AVX2_8x8 pp
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_pp_8x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_8xN 8, 16, pp
-    FILTER_VER_LUMA_AVX2_8xN 8, 16, pp
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_pp_8x32(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_8xN 8, 32, pp
-    FILTER_VER_LUMA_AVX2_8xN 8, 32, pp
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_ps_8x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_8xN 8, 4, ps
-    FILTER_VER_LUMA_AVX2_8x4 ps
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_ps_8x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_8xN 8, 8, ps
-    FILTER_VER_LUMA_AVX2_8x8 ps
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_ps_8x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_8xN 8, 16, ps
-    FILTER_VER_LUMA_AVX2_8xN 8, 16, ps
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_ps_8x32(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_8xN 8, 32, ps
-    FILTER_VER_LUMA_AVX2_8xN 8, 32, ps
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_%3_12x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_LUMA_12xN 3
-INIT_XMM sse4
-cglobal interp_8tap_vert_%3_%1x%2, 5, 7, 8
-    lea       r5, [3 * r1]
-    sub       r0, r5
-    shl       r4d, 6
-%ifidn %3,ps
-    add       r3d, r3d
-%endif
-
-%ifdef PIC
-    lea       r5, [tab_LumaCoeffVer]
-    lea       r6, [r5 + r4]
-%else
-    lea       r6, [tab_LumaCoeffVer + r4]
-%endif
-
- %ifidn %3,pp
-    mova      m3, [pw_512]
-%else
-    mova      m3, [pw_2000]
-%endif
-
-    mov       r4d, %2/4
-
-.loopH:
-    PROCESS_LUMA_W8_4R
-
-%ifidn %3,pp
-    pmulhrsw  m7, m3
-    pmulhrsw  m6, m3
-    pmulhrsw  m5, m3
-    pmulhrsw  m4, m3
-
-    packuswb  m7, m6
-    packuswb  m5, m4
-
-    movlps    [r2], m7
-    movhps    [r2 + r3], m7
-    lea       r5, [r2 + 2 * r3]
-    movlps    [r5], m5
-    movhps    [r5 + r3], m5
-%else
-    psubw     m7, m3
-    psubw     m6, m3
-    psubw     m5, m3
-    psubw     m4, m3
-
-    movu      [r2], m7
-    movu      [r2 + r3], m6
-    lea       r5, [r2 + 2 * r3]
-    movu      [r5], m5
-    movu      [r5 + r3], m4
-%endif
-
-    lea       r5, [8 * r1 - 8]
-    sub       r0, r5
-%ifidn %3,pp
-    add       r2, 8
-%else
-    add       r2, 16
-%endif
-
-    PROCESS_LUMA_W4_4R
-
-%ifidn %3,pp
-    pmulhrsw  m4, m3
-    pmulhrsw  m5, m3
-
-    packuswb  m4, m5
-
-    movd      [r2], m4
-    pextrd    [r2 + r3], m4, 1
-    lea       r5, [r2 + 2 * r3]
-    pextrd    [r5], m4, 2
-    pextrd    [r5 + r3], m4, 3
-%else
-    psubw     m4, m3
-    psubw     m5, m3
-
-    movlps    [r2], m4
-    movhps    [r2 + r3], m4
-    lea       r5, [r2 + 2 * r3]
-    movlps    [r5], m5
-    movhps    [r5 + r3], m5
-%endif
-
-    lea       r5, [4 * r1 + 8]
-    sub       r0, r5
-%ifidn %3,pp
-    lea       r2, [r2 + 4 * r3 - 8]
-%else
-    lea       r2, [r2 + 4 * r3 - 16]
-%endif
-
-    dec       r4d
-    jnz       .loopH
-
-    RET
-%endmacro
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_pp_12x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_12xN 12, 16, pp
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_ps_12x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_12xN 12, 16, ps
-
-%macro FILTER_VER_LUMA_AVX2_12x16 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_12x16, 4, 7, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-%ifidn %1,pp
-    mova            m14, [pw_512]
-%else
-    add             r3d, r3d
-    vbroadcasti128  m14, [pw_2000]
-%endif
-    lea             r6, [r3 * 3]
-
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
-    paddw           m0, m4
-    pmaddubsw       m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhbw       xm5, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
-    paddw           m1, m5
-    pmaddubsw       m3, [r5]
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhbw       xm6, xm4, xm5
-    punpcklbw       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
-    paddw           m0, m6
-    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
-    paddw           m2, m6
-    pmaddubsw       m4, [r5]
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhbw       xm7, xm5, xm6
-    punpcklbw       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
-    paddw           m1, m7
-    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
-    paddw           m3, m7
-    pmaddubsw       m5, [r5]
-    movu            xm7, [r0 + r4]                  ; m7 = row 7
-    punpckhbw       xm8, xm6, xm7
-    punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm8, 1
-    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
-    paddw           m0, m8
-    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
-    paddw           m2, m8
-    pmaddubsw       m8, m6, [r5 + 1 * mmsize]
-    paddw           m4, m8
-    pmaddubsw       m6, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm8, [r0]                       ; m8 = row 8
-    punpckhbw       xm9, xm7, xm8
-    punpcklbw       xm7, xm8
-    vinserti128     m7, m7, xm9, 1
-    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
-    paddw           m1, m9
-    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
-    paddw           m3, m9
-    pmaddubsw       m9, m7, [r5 + 1 * mmsize]
-    paddw           m5, m9
-    pmaddubsw       m7, [r5]
-    movu            xm9, [r0 + r1]                  ; m9 = row 9
-    punpckhbw       xm10, xm8, xm9
-    punpcklbw       xm8, xm9
-    vinserti128     m8, m8, xm10, 1
-    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
-    paddw           m2, m10
-    pmaddubsw       m10, m8, [r5 + 2 * mmsize]
-    paddw           m4, m10
-    pmaddubsw       m10, m8, [r5 + 1 * mmsize]
-    paddw           m6, m10
-    pmaddubsw       m8, [r5]
-    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
-    punpckhbw       xm11, xm9, xm10
-    punpcklbw       xm9, xm10
-    vinserti128     m9, m9, xm11, 1
-    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
-    paddw           m3, m11
-    pmaddubsw       m11, m9, [r5 + 2 * mmsize]
-    paddw           m5, m11
-    pmaddubsw       m11, m9, [r5 + 1 * mmsize]
-    paddw           m7, m11
-    pmaddubsw       m9, [r5]
-    movu            xm11, [r0 + r4]                 ; m11 = row 11
-    punpckhbw       xm12, xm10, xm11
-    punpcklbw       xm10, xm11
-    vinserti128     m10, m10, xm12, 1
-    pmaddubsw       m12, m10, [r5 + 3 * mmsize]
-    paddw           m4, m12
-    pmaddubsw       m12, m10, [r5 + 2 * mmsize]
-    paddw           m6, m12
-    pmaddubsw       m12, m10, [r5 + 1 * mmsize]
-    paddw           m8, m12
-    pmaddubsw       m10, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm12, [r0]                      ; m12 = row 12
-    punpckhbw       xm13, xm11, xm12
-    punpcklbw       xm11, xm12
-    vinserti128     m11, m11, xm13, 1
-    pmaddubsw       m13, m11, [r5 + 3 * mmsize]
-    paddw           m5, m13
-    pmaddubsw       m13, m11, [r5 + 2 * mmsize]
-    paddw           m7, m13
-    pmaddubsw       m13, m11, [r5 + 1 * mmsize]
-    paddw           m9, m13
-    pmaddubsw       m11, [r5]
-
-%ifidn %1,pp
-    pmulhrsw        m0, m14                         ; m0 = word: row 0
-    pmulhrsw        m1, m14                         ; m1 = word: row 1
-    pmulhrsw        m2, m14                         ; m2 = word: row 2
-    pmulhrsw        m3, m14                         ; m3 = word: row 3
-    pmulhrsw        m4, m14                         ; m4 = word: row 4
-    pmulhrsw        m5, m14                         ; m5 = word: row 5
-    packuswb        m0, m1
-    packuswb        m2, m3
-    packuswb        m4, m5
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
-    vextracti128    xm5, m4, 1
-    movq            [r2], xm0
-    pextrd          [r2 + 8], xm0, 2
-    movq            [r2 + r3], xm1
-    pextrd          [r2 + r3 + 8], xm1, 2
-    movq            [r2 + r3 * 2], xm2
-    pextrd          [r2 + r3 * 2 + 8], xm2, 2
-    movq            [r2 + r6], xm3
-    pextrd          [r2 + r6 + 8], xm3, 2
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm4
-    pextrd          [r2 + 8], xm4, 2
-    movq            [r2 + r3], xm5
-    pextrd          [r2 + r3 + 8], xm5, 2
-%else
-    psubw           m0, m14                         ; m0 = word: row 0
-    psubw           m1, m14                         ; m1 = word: row 1
-    psubw           m2, m14                         ; m2 = word: row 2
-    psubw           m3, m14                         ; m3 = word: row 3
-    psubw           m4, m14                         ; m4 = word: row 4
-    psubw           m5, m14                         ; m5 = word: row 5
-    movu            [r2], xm0
-    vextracti128    xm0, m0, 1
-    movq            [r2 + 16], xm0
-    movu            [r2 + r3], xm1
-    vextracti128    xm1, m1, 1
-    movq            [r2 + r3 + 16], xm1
-    movu            [r2 + r3 * 2], xm2
-    vextracti128    xm2, m2, 1
-    movq            [r2 + r3 * 2 + 16], xm2
-    movu            [r2 + r6], xm3
-    vextracti128    xm3, m3, 1
-    movq            [r2 + r6 + 16], xm3
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm4
-    vextracti128    xm4, m4, 1
-    movq            [r2 + 16], xm4
-    movu            [r2 + r3], xm5
-    vextracti128    xm5, m5, 1
-    movq            [r2 + r3 + 16], xm5
-%endif
-
-    movu            xm13, [r0 + r1]                 ; m13 = row 13
-    punpckhbw       xm0, xm12, xm13
-    punpcklbw       xm12, xm13
-    vinserti128     m12, m12, xm0, 1
-    pmaddubsw       m0, m12, [r5 + 3 * mmsize]
-    paddw           m6, m0
-    pmaddubsw       m0, m12, [r5 + 2 * mmsize]
-    paddw           m8, m0
-    pmaddubsw       m0, m12, [r5 + 1 * mmsize]
-    paddw           m10, m0
-    pmaddubsw       m12, [r5]
-    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
-    punpckhbw       xm1, xm13, xm0
-    punpcklbw       xm13, xm0
-    vinserti128     m13, m13, xm1, 1
-    pmaddubsw       m1, m13, [r5 + 3 * mmsize]
-    paddw           m7, m1
-    pmaddubsw       m1, m13, [r5 + 2 * mmsize]
-    paddw           m9, m1
-    pmaddubsw       m1, m13, [r5 + 1 * mmsize]
-    paddw           m11, m1
-    pmaddubsw       m13, [r5]
-
-%ifidn %1,pp
-    pmulhrsw        m6, m14                         ; m6 = word: row 6
-    pmulhrsw        m7, m14                         ; m7 = word: row 7
-    packuswb        m6, m7
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm7, m6, 1
-    movq            [r2 + r3 * 2], xm6
-    pextrd          [r2 + r3 * 2 + 8], xm6, 2
-    movq            [r2 + r6], xm7
-    pextrd          [r2 + r6 + 8], xm7, 2
-%else
-    psubw           m6, m14                         ; m6 = word: row 6
-    psubw           m7, m14                         ; m7 = word: row 7
-    movu            [r2 + r3 * 2], xm6
-    vextracti128    xm6, m6, 1
-    movq            [r2 + r3 * 2 + 16], xm6
-    movu            [r2 + r6], xm7
-    vextracti128    xm7, m7, 1
-    movq            [r2 + r6 + 16], xm7
-%endif
-    lea             r2, [r2 + r3 * 4]
-
-    movu            xm1, [r0 + r4]                  ; m1 = row 15
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m2, m0, [r5 + 3 * mmsize]
-    paddw           m8, m2
-    pmaddubsw       m2, m0, [r5 + 2 * mmsize]
-    paddw           m10, m2
-    pmaddubsw       m2, m0, [r5 + 1 * mmsize]
-    paddw           m12, m2
-    pmaddubsw       m0, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm2, [r0]                       ; m2 = row 16
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m3, m1, [r5 + 3 * mmsize]
-    paddw           m9, m3
-    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
-    paddw           m11, m3
-    pmaddubsw       m3, m1, [r5 + 1 * mmsize]
-    paddw           m13, m3
-    pmaddubsw       m1, [r5]
-    movu            xm3, [r0 + r1]                  ; m3 = row 17
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, [r5 + 3 * mmsize]
-    paddw           m10, m4
-    pmaddubsw       m4, m2, [r5 + 2 * mmsize]
-    paddw           m12, m4
-    pmaddubsw       m2, [r5 + 1 * mmsize]
-    paddw           m0, m2
-    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
-    punpckhbw       xm5, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, [r5 + 3 * mmsize]
-    paddw           m11, m5
-    pmaddubsw       m5, m3, [r5 + 2 * mmsize]
-    paddw           m13, m5
-    pmaddubsw       m3, [r5 + 1 * mmsize]
-    paddw           m1, m3
-    movu            xm5, [r0 + r4]                  ; m5 = row 19
-    punpckhbw       xm6, xm4, xm5
-    punpcklbw       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, [r5 + 3 * mmsize]
-    paddw           m12, m6
-    pmaddubsw       m4, [r5 + 2 * mmsize]
-    paddw           m0, m4
-    lea             r0, [r0 + r1 * 4]
-    movu            xm6, [r0]                       ; m6 = row 20
-    punpckhbw       xm7, xm5, xm6
-    punpcklbw       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m7, m5, [r5 + 3 * mmsize]
-    paddw           m13, m7
-    pmaddubsw       m5, [r5 + 2 * mmsize]
-    paddw           m1, m5
-    movu            xm7, [r0 + r1]                  ; m7 = row 21
-    punpckhbw       xm2, xm6, xm7
-    punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm2, 1
-    pmaddubsw       m6, [r5 + 3 * mmsize]
-    paddw           m0, m6
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 22
-    punpckhbw       xm3, xm7, xm2
-    punpcklbw       xm7, xm2
-    vinserti128     m7, m7, xm3, 1
-    pmaddubsw       m7, [r5 + 3 * mmsize]
-    paddw           m1, m7
-
-%ifidn %1,pp
-    pmulhrsw        m8, m14                         ; m8 = word: row 8
-    pmulhrsw        m9, m14                         ; m9 = word: row 9
-    pmulhrsw        m10, m14                        ; m10 = word: row 10
-    pmulhrsw        m11, m14                        ; m11 = word: row 11
-    pmulhrsw        m12, m14                        ; m12 = word: row 12
-    pmulhrsw        m13, m14                        ; m13 = word: row 13
-    pmulhrsw        m0, m14                         ; m0 = word: row 14
-    pmulhrsw        m1, m14                         ; m1 = word: row 15
-    packuswb        m8, m9
-    packuswb        m10, m11
-    packuswb        m12, m13
-    packuswb        m0, m1
-    vpermq          m8, m8, 11011000b
-    vpermq          m10, m10, 11011000b
-    vpermq          m12, m12, 11011000b
-    vpermq          m0, m0, 11011000b
-    vextracti128    xm9, m8, 1
-    vextracti128    xm11, m10, 1
-    vextracti128    xm13, m12, 1
-    vextracti128    xm1, m0, 1
-    movq            [r2], xm8
-    pextrd          [r2 + 8], xm8, 2
-    movq            [r2 + r3], xm9
-    pextrd          [r2 + r3 + 8], xm9, 2
-    movq            [r2 + r3 * 2], xm10
-    pextrd          [r2 + r3 * 2 + 8], xm10, 2
-    movq            [r2 + r6], xm11
-    pextrd          [r2 + r6 + 8], xm11, 2
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm12
-    pextrd          [r2 + 8], xm12, 2
-    movq            [r2 + r3], xm13
-    pextrd          [r2 + r3 + 8], xm13, 2
-    movq            [r2 + r3 * 2], xm0
-    pextrd          [r2 + r3 * 2 + 8], xm0, 2
-    movq            [r2 + r6], xm1
-    pextrd          [r2 + r6 + 8], xm1, 2
-%else
-    psubw           m8, m14                         ; m8 = word: row 8
-    psubw           m9, m14                         ; m9 = word: row 9
-    psubw           m10, m14                        ; m10 = word: row 10
-    psubw           m11, m14                        ; m11 = word: row 11
-    psubw           m12, m14                        ; m12 = word: row 12
-    psubw           m13, m14                        ; m13 = word: row 13
-    psubw           m0, m14                         ; m0 = word: row 14
-    psubw           m1, m14                         ; m1 = word: row 15
-    movu            [r2], xm8
-    vextracti128    xm8, m8, 1
-    movq            [r2 + 16], xm8
-    movu            [r2 + r3], xm9
-    vextracti128    xm9, m9, 1
-    movq            [r2 + r3 + 16], xm9
-    movu            [r2 + r3 * 2], xm10
-    vextracti128    xm10, m10, 1
-    movq            [r2 + r3 * 2 + 16], xm10
-    movu            [r2 + r6], xm11
-    vextracti128    xm11, m11, 1
-    movq            [r2 + r6 + 16], xm11
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm12
-    vextracti128    xm12, m12, 1
-    movq            [r2 + 16], xm12
-    movu            [r2 + r3], xm13
-    vextracti128    xm13, m13, 1
-    movq            [r2 + r3 + 16], xm13
-    movu            [r2 + r3 * 2], xm0
-    vextracti128    xm0, m0, 1
-    movq            [r2 + r3 * 2 + 16], xm0
-    movu            [r2 + r6], xm1
-    vextracti128    xm1, m1, 1
-    movq            [r2 + r6 + 16], xm1
-%endif
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_12x16 pp
-    FILTER_VER_LUMA_AVX2_12x16 ps
-
-%macro FILTER_VER_LUMA_AVX2_16x16 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_16x16, 4, 7, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-%ifidn %1,pp
-    mova            m14, [pw_512]
-%else
-    add             r3d, r3d
-    vbroadcasti128  m14, [pw_2000]
-%endif
-    lea             r6, [r3 * 3]
-
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
-    paddw           m0, m4
-    pmaddubsw       m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhbw       xm5, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
-    paddw           m1, m5
-    pmaddubsw       m3, [r5]
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhbw       xm6, xm4, xm5
-    punpcklbw       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
-    paddw           m0, m6
-    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
-    paddw           m2, m6
-    pmaddubsw       m4, [r5]
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhbw       xm7, xm5, xm6
-    punpcklbw       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
-    paddw           m1, m7
-    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
-    paddw           m3, m7
-    pmaddubsw       m5, [r5]
-    movu            xm7, [r0 + r4]                  ; m7 = row 7
-    punpckhbw       xm8, xm6, xm7
-    punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm8, 1
-    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
-    paddw           m0, m8
-    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
-    paddw           m2, m8
-    pmaddubsw       m8, m6, [r5 + 1 * mmsize]
-    paddw           m4, m8
-    pmaddubsw       m6, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm8, [r0]                       ; m8 = row 8
-    punpckhbw       xm9, xm7, xm8
-    punpcklbw       xm7, xm8
-    vinserti128     m7, m7, xm9, 1
-    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
-    paddw           m1, m9
-    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
-    paddw           m3, m9
-    pmaddubsw       m9, m7, [r5 + 1 * mmsize]
-    paddw           m5, m9
-    pmaddubsw       m7, [r5]
-    movu            xm9, [r0 + r1]                  ; m9 = row 9
-    punpckhbw       xm10, xm8, xm9
-    punpcklbw       xm8, xm9
-    vinserti128     m8, m8, xm10, 1
-    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
-    paddw           m2, m10
-    pmaddubsw       m10, m8, [r5 + 2 * mmsize]
-    paddw           m4, m10
-    pmaddubsw       m10, m8, [r5 + 1 * mmsize]
-    paddw           m6, m10
-    pmaddubsw       m8, [r5]
-    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
-    punpckhbw       xm11, xm9, xm10
-    punpcklbw       xm9, xm10
-    vinserti128     m9, m9, xm11, 1
-    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
-    paddw           m3, m11
-    pmaddubsw       m11, m9, [r5 + 2 * mmsize]
-    paddw           m5, m11
-    pmaddubsw       m11, m9, [r5 + 1 * mmsize]
-    paddw           m7, m11
-    pmaddubsw       m9, [r5]
-    movu            xm11, [r0 + r4]                 ; m11 = row 11
-    punpckhbw       xm12, xm10, xm11
-    punpcklbw       xm10, xm11
-    vinserti128     m10, m10, xm12, 1
-    pmaddubsw       m12, m10, [r5 + 3 * mmsize]
-    paddw           m4, m12
-    pmaddubsw       m12, m10, [r5 + 2 * mmsize]
-    paddw           m6, m12
-    pmaddubsw       m12, m10, [r5 + 1 * mmsize]
-    paddw           m8, m12
-    pmaddubsw       m10, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm12, [r0]                      ; m12 = row 12
-    punpckhbw       xm13, xm11, xm12
-    punpcklbw       xm11, xm12
-    vinserti128     m11, m11, xm13, 1
-    pmaddubsw       m13, m11, [r5 + 3 * mmsize]
-    paddw           m5, m13
-    pmaddubsw       m13, m11, [r5 + 2 * mmsize]
-    paddw           m7, m13
-    pmaddubsw       m13, m11, [r5 + 1 * mmsize]
-    paddw           m9, m13
-    pmaddubsw       m11, [r5]
-
-%ifidn %1,pp
-    pmulhrsw        m0, m14                         ; m0 = word: row 0
-    pmulhrsw        m1, m14                         ; m1 = word: row 1
-    pmulhrsw        m2, m14                         ; m2 = word: row 2
-    pmulhrsw        m3, m14                         ; m3 = word: row 3
-    pmulhrsw        m4, m14                         ; m4 = word: row 4
-    pmulhrsw        m5, m14                         ; m5 = word: row 5
-    packuswb        m0, m1
-    packuswb        m2, m3
-    packuswb        m4, m5
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
-    vextracti128    xm5, m4, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm4
-    movu            [r2 + r3], xm5
-%else
-    psubw           m0, m14                         ; m0 = word: row 0
-    psubw           m1, m14                         ; m1 = word: row 1
-    psubw           m2, m14                         ; m2 = word: row 2
-    psubw           m3, m14                         ; m3 = word: row 3
-    psubw           m4, m14                         ; m4 = word: row 4
-    psubw           m5, m14                         ; m5 = word: row 5
-    movu            [r2], m0
-    movu            [r2 + r3], m1
-    movu            [r2 + r3 * 2], m2
-    movu            [r2 + r6], m3
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], m4
-    movu            [r2 + r3], m5
-%endif
-
-    movu            xm13, [r0 + r1]                 ; m13 = row 13
-    punpckhbw       xm0, xm12, xm13
-    punpcklbw       xm12, xm13
-    vinserti128     m12, m12, xm0, 1
-    pmaddubsw       m0, m12, [r5 + 3 * mmsize]
-    paddw           m6, m0
-    pmaddubsw       m0, m12, [r5 + 2 * mmsize]
-    paddw           m8, m0
-    pmaddubsw       m0, m12, [r5 + 1 * mmsize]
-    paddw           m10, m0
-    pmaddubsw       m12, [r5]
-    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
-    punpckhbw       xm1, xm13, xm0
-    punpcklbw       xm13, xm0
-    vinserti128     m13, m13, xm1, 1
-    pmaddubsw       m1, m13, [r5 + 3 * mmsize]
-    paddw           m7, m1
-    pmaddubsw       m1, m13, [r5 + 2 * mmsize]
-    paddw           m9, m1
-    pmaddubsw       m1, m13, [r5 + 1 * mmsize]
-    paddw           m11, m1
-    pmaddubsw       m13, [r5]
-
-%ifidn %1,pp
-    pmulhrsw        m6, m14                         ; m6 = word: row 6
-    pmulhrsw        m7, m14                         ; m7 = word: row 7
-    packuswb        m6, m7
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm7, m6, 1
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r6], xm7
-%else
-    psubw           m6, m14                         ; m6 = word: row 6
-    psubw           m7, m14                         ; m7 = word: row 7
-    movu            [r2 + r3 * 2], m6
-    movu            [r2 + r6], m7
-%endif
-    lea             r2, [r2 + r3 * 4]
-
-    movu            xm1, [r0 + r4]                  ; m1 = row 15
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m2, m0, [r5 + 3 * mmsize]
-    paddw           m8, m2
-    pmaddubsw       m2, m0, [r5 + 2 * mmsize]
-    paddw           m10, m2
-    pmaddubsw       m2, m0, [r5 + 1 * mmsize]
-    paddw           m12, m2
-    pmaddubsw       m0, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm2, [r0]                       ; m2 = row 16
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m3, m1, [r5 + 3 * mmsize]
-    paddw           m9, m3
-    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
-    paddw           m11, m3
-    pmaddubsw       m3, m1, [r5 + 1 * mmsize]
-    paddw           m13, m3
-    pmaddubsw       m1, [r5]
-    movu            xm3, [r0 + r1]                  ; m3 = row 17
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, [r5 + 3 * mmsize]
-    paddw           m10, m4
-    pmaddubsw       m4, m2, [r5 + 2 * mmsize]
-    paddw           m12, m4
-    pmaddubsw       m2, [r5 + 1 * mmsize]
-    paddw           m0, m2
-    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
-    punpckhbw       xm5, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, [r5 + 3 * mmsize]
-    paddw           m11, m5
-    pmaddubsw       m5, m3, [r5 + 2 * mmsize]
-    paddw           m13, m5
-    pmaddubsw       m3, [r5 + 1 * mmsize]
-    paddw           m1, m3
-    movu            xm5, [r0 + r4]                  ; m5 = row 19
-    punpckhbw       xm6, xm4, xm5
-    punpcklbw       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, [r5 + 3 * mmsize]
-    paddw           m12, m6
-    pmaddubsw       m4, [r5 + 2 * mmsize]
-    paddw           m0, m4
-    lea             r0, [r0 + r1 * 4]
-    movu            xm6, [r0]                       ; m6 = row 20
-    punpckhbw       xm7, xm5, xm6
-    punpcklbw       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m7, m5, [r5 + 3 * mmsize]
-    paddw           m13, m7
-    pmaddubsw       m5, [r5 + 2 * mmsize]
-    paddw           m1, m5
-    movu            xm7, [r0 + r1]                  ; m7 = row 21
-    punpckhbw       xm2, xm6, xm7
-    punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm2, 1
-    pmaddubsw       m6, [r5 + 3 * mmsize]
-    paddw           m0, m6
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 22
-    punpckhbw       xm3, xm7, xm2
-    punpcklbw       xm7, xm2
-    vinserti128     m7, m7, xm3, 1
-    pmaddubsw       m7, [r5 + 3 * mmsize]
-    paddw           m1, m7
-
-%ifidn %1,pp
-    pmulhrsw        m8, m14                         ; m8 = word: row 8
-    pmulhrsw        m9, m14                         ; m9 = word: row 9
-    pmulhrsw        m10, m14                        ; m10 = word: row 10
-    pmulhrsw        m11, m14                        ; m11 = word: row 11
-    pmulhrsw        m12, m14                        ; m12 = word: row 12
-    pmulhrsw        m13, m14                        ; m13 = word: row 13
-    pmulhrsw        m0, m14                         ; m0 = word: row 14
-    pmulhrsw        m1, m14                         ; m1 = word: row 15
-    packuswb        m8, m9
-    packuswb        m10, m11
-    packuswb        m12, m13
-    packuswb        m0, m1
-    vpermq          m8, m8, 11011000b
-    vpermq          m10, m10, 11011000b
-    vpermq          m12, m12, 11011000b
-    vpermq          m0, m0, 11011000b
-    vextracti128    xm9, m8, 1
-    vextracti128    xm11, m10, 1
-    vextracti128    xm13, m12, 1
-    vextracti128    xm1, m0, 1
-    movu            [r2], xm8
-    movu            [r2 + r3], xm9
-    movu            [r2 + r3 * 2], xm10
-    movu            [r2 + r6], xm11
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm12
-    movu            [r2 + r3], xm13
-    movu            [r2 + r3 * 2], xm0
-    movu            [r2 + r6], xm1
-%else
-    psubw           m8, m14                         ; m8 = word: row 8
-    psubw           m9, m14                         ; m9 = word: row 9
-    psubw           m10, m14                        ; m10 = word: row 10
-    psubw           m11, m14                        ; m11 = word: row 11
-    psubw           m12, m14                        ; m12 = word: row 12
-    psubw           m13, m14                        ; m13 = word: row 13
-    psubw           m0, m14                         ; m0 = word: row 14
-    psubw           m1, m14                         ; m1 = word: row 15
-    movu            [r2], m8
-    movu            [r2 + r3], m9
-    movu            [r2 + r3 * 2], m10
-    movu            [r2 + r6], m11
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], m12
-    movu            [r2 + r3], m13
-    movu            [r2 + r3 * 2], m0
-    movu            [r2 + r6], m1
-%endif
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_16x16 pp
-    FILTER_VER_LUMA_AVX2_16x16 ps
-
-%macro FILTER_VER_LUMA_AVX2_16x12 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_16x12, 4, 7, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-%ifidn %1,pp
-    mova            m14, [pw_512]
-%else
-    add             r3d, r3d
-    vbroadcasti128  m14, [pw_2000]
-%endif
-    lea             r6, [r3 * 3]
-
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
-    paddw           m0, m4
-    pmaddubsw       m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhbw       xm5, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
-    paddw           m1, m5
-    pmaddubsw       m3, [r5]
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhbw       xm6, xm4, xm5
-    punpcklbw       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
-    paddw           m0, m6
-    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
-    paddw           m2, m6
-    pmaddubsw       m4, [r5]
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhbw       xm7, xm5, xm6
-    punpcklbw       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
-    paddw           m1, m7
-    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
-    paddw           m3, m7
-    pmaddubsw       m5, [r5]
-    movu            xm7, [r0 + r4]                  ; m7 = row 7
-    punpckhbw       xm8, xm6, xm7
-    punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm8, 1
-    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
-    paddw           m0, m8
-    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
-    paddw           m2, m8
-    pmaddubsw       m8, m6, [r5 + 1 * mmsize]
-    paddw           m4, m8
-    pmaddubsw       m6, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm8, [r0]                       ; m8 = row 8
-    punpckhbw       xm9, xm7, xm8
-    punpcklbw       xm7, xm8
-    vinserti128     m7, m7, xm9, 1
-    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
-    paddw           m1, m9
-    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
-    paddw           m3, m9
-    pmaddubsw       m9, m7, [r5 + 1 * mmsize]
-    paddw           m5, m9
-    pmaddubsw       m7, [r5]
-    movu            xm9, [r0 + r1]                  ; m9 = row 9
-    punpckhbw       xm10, xm8, xm9
-    punpcklbw       xm8, xm9
-    vinserti128     m8, m8, xm10, 1
-    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
-    paddw           m2, m10
-    pmaddubsw       m10, m8, [r5 + 2 * mmsize]
-    paddw           m4, m10
-    pmaddubsw       m10, m8, [r5 + 1 * mmsize]
-    paddw           m6, m10
-    pmaddubsw       m8, [r5]
-    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
-    punpckhbw       xm11, xm9, xm10
-    punpcklbw       xm9, xm10
-    vinserti128     m9, m9, xm11, 1
-    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
-    paddw           m3, m11
-    pmaddubsw       m11, m9, [r5 + 2 * mmsize]
-    paddw           m5, m11
-    pmaddubsw       m11, m9, [r5 + 1 * mmsize]
-    paddw           m7, m11
-    pmaddubsw       m9, [r5]
-    movu            xm11, [r0 + r4]                 ; m11 = row 11
-    punpckhbw       xm12, xm10, xm11
-    punpcklbw       xm10, xm11
-    vinserti128     m10, m10, xm12, 1
-    pmaddubsw       m12, m10, [r5 + 3 * mmsize]
-    paddw           m4, m12
-    pmaddubsw       m12, m10, [r5 + 2 * mmsize]
-    paddw           m6, m12
-    pmaddubsw       m12, m10, [r5 + 1 * mmsize]
-    paddw           m8, m12
-    pmaddubsw       m10, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm12, [r0]                      ; m12 = row 12
-    punpckhbw       xm13, xm11, xm12
-    punpcklbw       xm11, xm12
-    vinserti128     m11, m11, xm13, 1
-    pmaddubsw       m13, m11, [r5 + 3 * mmsize]
-    paddw           m5, m13
-    pmaddubsw       m13, m11, [r5 + 2 * mmsize]
-    paddw           m7, m13
-    pmaddubsw       m13, m11, [r5 + 1 * mmsize]
-    paddw           m9, m13
-    pmaddubsw       m11, [r5]
-
-%ifidn %1,pp
-    pmulhrsw        m0, m14                         ; m0 = word: row 0
-    pmulhrsw        m1, m14                         ; m1 = word: row 1
-    pmulhrsw        m2, m14                         ; m2 = word: row 2
-    pmulhrsw        m3, m14                         ; m3 = word: row 3
-    pmulhrsw        m4, m14                         ; m4 = word: row 4
-    pmulhrsw        m5, m14                         ; m5 = word: row 5
-    packuswb        m0, m1
-    packuswb        m2, m3
-    packuswb        m4, m5
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
-    vextracti128    xm5, m4, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm4
-    movu            [r2 + r3], xm5
-%else
-    psubw           m0, m14                         ; m0 = word: row 0
-    psubw           m1, m14                         ; m1 = word: row 1
-    psubw           m2, m14                         ; m2 = word: row 2
-    psubw           m3, m14                         ; m3 = word: row 3
-    psubw           m4, m14                         ; m4 = word: row 4
-    psubw           m5, m14                         ; m5 = word: row 5
-    movu            [r2], m0
-    movu            [r2 + r3], m1
-    movu            [r2 + r3 * 2], m2
-    movu            [r2 + r6], m3
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], m4
-    movu            [r2 + r3], m5
-%endif
-
-    movu            xm13, [r0 + r1]                 ; m13 = row 13
-    punpckhbw       xm0, xm12, xm13
-    punpcklbw       xm12, xm13
-    vinserti128     m12, m12, xm0, 1
-    pmaddubsw       m0, m12, [r5 + 3 * mmsize]
-    paddw           m6, m0
-    pmaddubsw       m0, m12, [r5 + 2 * mmsize]
-    paddw           m8, m0
-    pmaddubsw       m0, m12, [r5 + 1 * mmsize]
-    paddw           m10, m0
-    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
-    punpckhbw       xm1, xm13, xm0
-    punpcklbw       xm13, xm0
-    vinserti128     m13, m13, xm1, 1
-    pmaddubsw       m1, m13, [r5 + 3 * mmsize]
-    paddw           m7, m1
-    pmaddubsw       m1, m13, [r5 + 2 * mmsize]
-    paddw           m9, m1
-    pmaddubsw       m1, m13, [r5 + 1 * mmsize]
-    paddw           m11, m1
-
-%ifidn %1,pp
-    pmulhrsw        m6, m14                         ; m6 = word: row 6
-    pmulhrsw        m7, m14                         ; m7 = word: row 7
-    packuswb        m6, m7
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm7, m6, 1
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r6], xm7
-%else
-    psubw           m6, m14                         ; m6 = word: row 6
-    psubw           m7, m14                         ; m7 = word: row 7
-    movu            [r2 + r3 * 2], m6
-    movu            [r2 + r6], m7
-%endif
-    lea             r2, [r2 + r3 * 4]
-
-    movu            xm1, [r0 + r4]                  ; m1 = row 15
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m2, m0, [r5 + 3 * mmsize]
-    paddw           m8, m2
-    pmaddubsw       m2, m0, [r5 + 2 * mmsize]
-    paddw           m10, m2
-    lea             r0, [r0 + r1 * 4]
-    movu            xm2, [r0]                       ; m2 = row 16
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m3, m1, [r5 + 3 * mmsize]
-    paddw           m9, m3
-    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
-    paddw           m11, m3
-    movu            xm3, [r0 + r1]                  ; m3 = row 17
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, [r5 + 3 * mmsize]
-    paddw           m10, m4
-    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
-    punpckhbw       xm5, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, [r5 + 3 * mmsize]
-    paddw           m11, m5
-
-%ifidn %1,pp
-    pmulhrsw        m8, m14                         ; m8 = word: row 8
-    pmulhrsw        m9, m14                         ; m9 = word: row 9
-    pmulhrsw        m10, m14                        ; m10 = word: row 10
-    pmulhrsw        m11, m14                        ; m11 = word: row 11
-    packuswb        m8, m9
-    packuswb        m10, m11
-    vpermq          m8, m8, 11011000b
-    vpermq          m10, m10, 11011000b
-    vextracti128    xm9, m8, 1
-    vextracti128    xm11, m10, 1
-    movu            [r2], xm8
-    movu            [r2 + r3], xm9
-    movu            [r2 + r3 * 2], xm10
-    movu            [r2 + r6], xm11
-%else
-    psubw           m8, m14                         ; m8 = word: row 8
-    psubw           m9, m14                         ; m9 = word: row 9
-    psubw           m10, m14                        ; m10 = word: row 10
-    psubw           m11, m14                        ; m11 = word: row 11
-    movu            [r2], m8
-    movu            [r2 + r3], m9
-    movu            [r2 + r3 * 2], m10
-    movu            [r2 + r6], m11
-%endif
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_16x12 pp
-    FILTER_VER_LUMA_AVX2_16x12 ps
-
-%macro FILTER_VER_LUMA_AVX2_16x8 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_16x8, 4, 6, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-%ifidn %1,pp
-    mova            m14, [pw_512]
-%else
-    add             r3d, r3d
-    vbroadcasti128  m14, [pw_2000]
-%endif
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
-    paddw           m0, m4
-    pmaddubsw       m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhbw       xm5, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
-    paddw           m1, m5
-    pmaddubsw       m3, [r5]
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhbw       xm6, xm4, xm5
-    punpcklbw       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
-    paddw           m0, m6
-    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
-    paddw           m2, m6
-    pmaddubsw       m4, [r5]
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhbw       xm7, xm5, xm6
-    punpcklbw       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
-    paddw           m1, m7
-    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
-    paddw           m3, m7
-    pmaddubsw       m5, [r5]
-    movu            xm7, [r0 + r4]                  ; m7 = row 7
-    punpckhbw       xm8, xm6, xm7
-    punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm8, 1
-    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
-    paddw           m0, m8
-    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
-    paddw           m2, m8
-    pmaddubsw       m8, m6, [r5 + 1 * mmsize]
-    paddw           m4, m8
-    pmaddubsw       m6, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm8, [r0]                       ; m8 = row 8
-    punpckhbw       xm9, xm7, xm8
-    punpcklbw       xm7, xm8
-    vinserti128     m7, m7, xm9, 1
-    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
-    paddw           m1, m9
-    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
-    paddw           m3, m9
-    pmaddubsw       m9, m7, [r5 + 1 * mmsize]
-    paddw           m5, m9
-    pmaddubsw       m7, [r5]
-    movu            xm9, [r0 + r1]                  ; m9 = row 9
-    punpckhbw       xm10, xm8, xm9
-    punpcklbw       xm8, xm9
-    vinserti128     m8, m8, xm10, 1
-    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
-    paddw           m2, m10
-    pmaddubsw       m10, m8, [r5 + 2 * mmsize]
-    paddw           m4, m10
-    pmaddubsw       m10, m8, [r5 + 1 * mmsize]
-    paddw           m6, m10
-    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
-    punpckhbw       xm11, xm9, xm10
-    punpcklbw       xm9, xm10
-    vinserti128     m9, m9, xm11, 1
-    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
-    paddw           m3, m11
-    pmaddubsw       m11, m9, [r5 + 2 * mmsize]
-    paddw           m5, m11
-    pmaddubsw       m11, m9, [r5 + 1 * mmsize]
-    paddw           m7, m11
-    movu            xm11, [r0 + r4]                 ; m11 = row 11
-    punpckhbw       xm12, xm10, xm11
-    punpcklbw       xm10, xm11
-    vinserti128     m10, m10, xm12, 1
-    pmaddubsw       m12, m10, [r5 + 3 * mmsize]
-    paddw           m4, m12
-    pmaddubsw       m12, m10, [r5 + 2 * mmsize]
-    paddw           m6, m12
-    lea             r0, [r0 + r1 * 4]
-    movu            xm12, [r0]                      ; m12 = row 12
-    punpckhbw       xm13, xm11, xm12
-    punpcklbw       xm11, xm12
-    vinserti128     m11, m11, xm13, 1
-    pmaddubsw       m13, m11, [r5 + 3 * mmsize]
-    paddw           m5, m13
-    pmaddubsw       m13, m11, [r5 + 2 * mmsize]
-    paddw           m7, m13
-    lea             r4, [r3 * 3]
-%ifidn %1,pp
-    pmulhrsw        m0, m14                         ; m0 = word: row 0
-    pmulhrsw        m1, m14                         ; m1 = word: row 1
-    pmulhrsw        m2, m14                         ; m2 = word: row 2
-    pmulhrsw        m3, m14                         ; m3 = word: row 3
-    pmulhrsw        m4, m14                         ; m4 = word: row 4
-    pmulhrsw        m5, m14                         ; m5 = word: row 5
-    packuswb        m0, m1
-    packuswb        m2, m3
-    packuswb        m4, m5
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
-    vextracti128    xm5, m4, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r4], xm3
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm4
-    movu            [r2 + r3], xm5
-%else
-    psubw           m0, m14                         ; m0 = word: row 0
-    psubw           m1, m14                         ; m1 = word: row 1
-    psubw           m2, m14                         ; m2 = word: row 2
-    psubw           m3, m14                         ; m3 = word: row 3
-    psubw           m4, m14                         ; m4 = word: row 4
-    psubw           m5, m14                         ; m5 = word: row 5
-    movu            [r2], m0
-    movu            [r2 + r3], m1
-    movu            [r2 + r3 * 2], m2
-    movu            [r2 + r4], m3
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], m4
-    movu            [r2 + r3], m5
-%endif
-    movu            xm13, [r0 + r1]                 ; m13 = row 13
-    punpckhbw       xm0, xm12, xm13
-    punpcklbw       xm12, xm13
-    vinserti128     m12, m12, xm0, 1
-    pmaddubsw       m0, m12, [r5 + 3 * mmsize]
-    paddw           m6, m0
-    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
-    punpckhbw       xm1, xm13, xm0
-    punpcklbw       xm13, xm0
-    vinserti128     m13, m13, xm1, 1
-    pmaddubsw       m1, m13, [r5 + 3 * mmsize]
-    paddw           m7, m1
-%ifidn %1,pp
-    pmulhrsw        m6, m14                         ; m6 = word: row 6
-    pmulhrsw        m7, m14                         ; m7 = word: row 7
-    packuswb        m6, m7
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm7, m6, 1
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r4], xm7
-%else
-    psubw           m6, m14                         ; m6 = word: row 6
-    psubw           m7, m14                         ; m7 = word: row 7
-    movu            [r2 + r3 * 2], m6
-    movu            [r2 + r4], m7
-%endif
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_16x8 pp
-    FILTER_VER_LUMA_AVX2_16x8 ps
-
-%macro FILTER_VER_LUMA_AVX2_16x4 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_16x4, 4, 6, 13
-    mov             r4d, r4m
-    shl             r4d, 7
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-%ifidn %1,pp
-    mova            m12, [pw_512]
-%else
-    add             r3d, r3d
-    vbroadcasti128  m12, [pw_2000]
-%endif
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
-    paddw           m0, m4
-    pmaddubsw       m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhbw       xm5, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
-    paddw           m1, m5
-    pmaddubsw       m3, [r5]
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhbw       xm6, xm4, xm5
-    punpcklbw       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
-    paddw           m0, m6
-    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
-    paddw           m2, m6
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhbw       xm7, xm5, xm6
-    punpcklbw       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
-    paddw           m1, m7
-    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
-    paddw           m3, m7
-    movu            xm7, [r0 + r4]                  ; m7 = row 7
-    punpckhbw       xm8, xm6, xm7
-    punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm8, 1
-    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
-    paddw           m0, m8
-    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
-    paddw           m2, m8
-    lea             r0, [r0 + r1 * 4]
-    movu            xm8, [r0]                       ; m8 = row 8
-    punpckhbw       xm9, xm7, xm8
-    punpcklbw       xm7, xm8
-    vinserti128     m7, m7, xm9, 1
-    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
-    paddw           m1, m9
-    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
-    paddw           m3, m9
-    movu            xm9, [r0 + r1]                  ; m9 = row 9
-    punpckhbw       xm10, xm8, xm9
-    punpcklbw       xm8, xm9
-    vinserti128     m8, m8, xm10, 1
-    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
-    paddw           m2, m10
-    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
-    punpckhbw       xm11, xm9, xm10
-    punpcklbw       xm9, xm10
-    vinserti128     m9, m9, xm11, 1
-    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
-    paddw           m3, m11
-%ifidn %1,pp
-    pmulhrsw        m0, m12                         ; m0 = word: row 0
-    pmulhrsw        m1, m12                         ; m1 = word: row 1
-    pmulhrsw        m2, m12                         ; m2 = word: row 2
-    pmulhrsw        m3, m12                         ; m3 = word: row 3
-    packuswb        m0, m1
-    packuswb        m2, m3
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    lea             r4, [r3 * 3]
-    movu            [r2 + r4], xm3
-%else
-    psubw           m0, m12                         ; m0 = word: row 0
-    psubw           m1, m12                         ; m1 = word: row 1
-    psubw           m2, m12                         ; m2 = word: row 2
-    psubw           m3, m12                         ; m3 = word: row 3
-    movu            [r2], m0
-    movu            [r2 + r3], m1
-    movu            [r2 + r3 * 2], m2
-    lea             r4, [r3 * 3]
-    movu            [r2 + r4], m3
-%endif
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_16x4 pp
-    FILTER_VER_LUMA_AVX2_16x4 ps
-%macro FILTER_VER_LUMA_AVX2_16xN 3
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%3_%1x%2, 4, 9, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-%ifidn %3,ps
-    add             r3d, r3d
-    vbroadcasti128  m14, [pw_2000]
-%else
-    mova            m14, [pw_512]
-%endif
-    lea             r6, [r3 * 3]
-    lea             r7, [r1 * 4]
-    mov             r8d, %2 / 16
-
-.loop:
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
-    paddw           m0, m4
-    pmaddubsw       m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhbw       xm5, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
-    paddw           m1, m5
-    pmaddubsw       m3, [r5]
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhbw       xm6, xm4, xm5
-    punpcklbw       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
-    paddw           m0, m6
-    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
-    paddw           m2, m6
-    pmaddubsw       m4, [r5]
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhbw       xm7, xm5, xm6
-    punpcklbw       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
-    paddw           m1, m7
-    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
-    paddw           m3, m7
-    pmaddubsw       m5, [r5]
-    movu            xm7, [r0 + r4]                  ; m7 = row 7
-    punpckhbw       xm8, xm6, xm7
-    punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm8, 1
-    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
-    paddw           m0, m8
-    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
-    paddw           m2, m8
-    pmaddubsw       m8, m6, [r5 + 1 * mmsize]
-    paddw           m4, m8
-    pmaddubsw       m6, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm8, [r0]                       ; m8 = row 8
-    punpckhbw       xm9, xm7, xm8
-    punpcklbw       xm7, xm8
-    vinserti128     m7, m7, xm9, 1
-    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
-    paddw           m1, m9
-    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
-    paddw           m3, m9
-    pmaddubsw       m9, m7, [r5 + 1 * mmsize]
-    paddw           m5, m9
-    pmaddubsw       m7, [r5]
-    movu            xm9, [r0 + r1]                  ; m9 = row 9
-    punpckhbw       xm10, xm8, xm9
-    punpcklbw       xm8, xm9
-    vinserti128     m8, m8, xm10, 1
-    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
-    paddw           m2, m10
-    pmaddubsw       m10, m8, [r5 + 2 * mmsize]
-    paddw           m4, m10
-    pmaddubsw       m10, m8, [r5 + 1 * mmsize]
-    paddw           m6, m10
-    pmaddubsw       m8, [r5]
-    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
-    punpckhbw       xm11, xm9, xm10
-    punpcklbw       xm9, xm10
-    vinserti128     m9, m9, xm11, 1
-    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
-    paddw           m3, m11
-    pmaddubsw       m11, m9, [r5 + 2 * mmsize]
-    paddw           m5, m11
-    pmaddubsw       m11, m9, [r5 + 1 * mmsize]
-    paddw           m7, m11
-    pmaddubsw       m9, [r5]
-    movu            xm11, [r0 + r4]                 ; m11 = row 11
-    punpckhbw       xm12, xm10, xm11
-    punpcklbw       xm10, xm11
-    vinserti128     m10, m10, xm12, 1
-    pmaddubsw       m12, m10, [r5 + 3 * mmsize]
-    paddw           m4, m12
-    pmaddubsw       m12, m10, [r5 + 2 * mmsize]
-    paddw           m6, m12
-    pmaddubsw       m12, m10, [r5 + 1 * mmsize]
-    paddw           m8, m12
-    pmaddubsw       m10, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm12, [r0]                      ; m12 = row 12
-    punpckhbw       xm13, xm11, xm12
-    punpcklbw       xm11, xm12
-    vinserti128     m11, m11, xm13, 1
-    pmaddubsw       m13, m11, [r5 + 3 * mmsize]
-    paddw           m5, m13
-    pmaddubsw       m13, m11, [r5 + 2 * mmsize]
-    paddw           m7, m13
-    pmaddubsw       m13, m11, [r5 + 1 * mmsize]
-    paddw           m9, m13
-    pmaddubsw       m11, [r5]
-
-%ifidn %3,pp
-    pmulhrsw        m0, m14                         ; m0 = word: row 0
-    pmulhrsw        m1, m14                         ; m1 = word: row 1
-    pmulhrsw        m2, m14                         ; m2 = word: row 2
-    pmulhrsw        m3, m14                         ; m3 = word: row 3
-    pmulhrsw        m4, m14                         ; m4 = word: row 4
-    pmulhrsw        m5, m14                         ; m5 = word: row 5
-    packuswb        m0, m1
-    packuswb        m2, m3
-    packuswb        m4, m5
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
-    vextracti128    xm5, m4, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm4
-    movu            [r2 + r3], xm5
-%else
-    psubw           m0, m14                         ; m0 = word: row 0
-    psubw           m1, m14                         ; m1 = word: row 1
-    psubw           m2, m14                         ; m2 = word: row 2
-    psubw           m3, m14                         ; m3 = word: row 3
-    psubw           m4, m14                         ; m4 = word: row 4
-    psubw           m5, m14                         ; m5 = word: row 5
-    movu            [r2], m0
-    movu            [r2 + r3], m1
-    movu            [r2 + r3 * 2], m2
-    movu            [r2 + r6], m3
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], m4
-    movu            [r2 + r3], m5
-%endif
-
-    movu            xm13, [r0 + r1]                 ; m13 = row 13
-    punpckhbw       xm0, xm12, xm13
-    punpcklbw       xm12, xm13
-    vinserti128     m12, m12, xm0, 1
-    pmaddubsw       m0, m12, [r5 + 3 * mmsize]
-    paddw           m6, m0
-    pmaddubsw       m0, m12, [r5 + 2 * mmsize]
-    paddw           m8, m0
-    pmaddubsw       m0, m12, [r5 + 1 * mmsize]
-    paddw           m10, m0
-    pmaddubsw       m12, [r5]
-    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
-    punpckhbw       xm1, xm13, xm0
-    punpcklbw       xm13, xm0
-    vinserti128     m13, m13, xm1, 1
-    pmaddubsw       m1, m13, [r5 + 3 * mmsize]
-    paddw           m7, m1
-    pmaddubsw       m1, m13, [r5 + 2 * mmsize]
-    paddw           m9, m1
-    pmaddubsw       m1, m13, [r5 + 1 * mmsize]
-    paddw           m11, m1
-    pmaddubsw       m13, [r5]
-
-%ifidn %3,pp
-    pmulhrsw        m6, m14                         ; m6 = word: row 6
-    pmulhrsw        m7, m14                         ; m7 = word: row 7
-    packuswb        m6, m7
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm7, m6, 1
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r6], xm7
-%else
-    psubw           m6, m14                         ; m6 = word: row 6
-    psubw           m7, m14                         ; m7 = word: row 7
-    movu            [r2 + r3 * 2], m6
-    movu            [r2 + r6], m7
-%endif
-
-    lea             r2, [r2 + r3 * 4]
-
-    movu            xm1, [r0 + r4]                  ; m1 = row 15
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m2, m0, [r5 + 3 * mmsize]
-    paddw           m8, m2
-    pmaddubsw       m2, m0, [r5 + 2 * mmsize]
-    paddw           m10, m2
-    pmaddubsw       m2, m0, [r5 + 1 * mmsize]
-    paddw           m12, m2
-    pmaddubsw       m0, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm2, [r0]                       ; m2 = row 16
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m3, m1, [r5 + 3 * mmsize]
-    paddw           m9, m3
-    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
-    paddw           m11, m3
-    pmaddubsw       m3, m1, [r5 + 1 * mmsize]
-    paddw           m13, m3
-    pmaddubsw       m1, [r5]
-    movu            xm3, [r0 + r1]                  ; m3 = row 17
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, [r5 + 3 * mmsize]
-    paddw           m10, m4
-    pmaddubsw       m4, m2, [r5 + 2 * mmsize]
-    paddw           m12, m4
-    pmaddubsw       m2, [r5 + 1 * mmsize]
-    paddw           m0, m2
-    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
-    punpckhbw       xm5, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, [r5 + 3 * mmsize]
-    paddw           m11, m5
-    pmaddubsw       m5, m3, [r5 + 2 * mmsize]
-    paddw           m13, m5
-    pmaddubsw       m3, [r5 + 1 * mmsize]
-    paddw           m1, m3
-    movu            xm5, [r0 + r4]                  ; m5 = row 19
-    punpckhbw       xm6, xm4, xm5
-    punpcklbw       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, [r5 + 3 * mmsize]
-    paddw           m12, m6
-    pmaddubsw       m4, [r5 + 2 * mmsize]
-    paddw           m0, m4
-    lea             r0, [r0 + r1 * 4]
-    movu            xm6, [r0]                       ; m6 = row 20
-    punpckhbw       xm7, xm5, xm6
-    punpcklbw       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m7, m5, [r5 + 3 * mmsize]
-    paddw           m13, m7
-    pmaddubsw       m5, [r5 + 2 * mmsize]
-    paddw           m1, m5
-    movu            xm7, [r0 + r1]                  ; m7 = row 21
-    punpckhbw       xm2, xm6, xm7
-    punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm2, 1
-    pmaddubsw       m6, [r5 + 3 * mmsize]
-    paddw           m0, m6
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 22
-    punpckhbw       xm3, xm7, xm2
-    punpcklbw       xm7, xm2
-    vinserti128     m7, m7, xm3, 1
-    pmaddubsw       m7, [r5 + 3 * mmsize]
-    paddw           m1, m7
-
-%ifidn %3,pp
-    pmulhrsw        m8, m14                         ; m8 = word: row 8
-    pmulhrsw        m9, m14                         ; m9 = word: row 9
-    pmulhrsw        m10, m14                        ; m10 = word: row 10
-    pmulhrsw        m11, m14                        ; m11 = word: row 11
-    pmulhrsw        m12, m14                        ; m12 = word: row 12
-    pmulhrsw        m13, m14                        ; m13 = word: row 13
-    pmulhrsw        m0, m14                         ; m0 = word: row 14
-    pmulhrsw        m1, m14                         ; m1 = word: row 15
-    packuswb        m8, m9
-    packuswb        m10, m11
-    packuswb        m12, m13
-    packuswb        m0, m1
-    vpermq          m8, m8, 11011000b
-    vpermq          m10, m10, 11011000b
-    vpermq          m12, m12, 11011000b
-    vpermq          m0, m0, 11011000b
-    vextracti128    xm9, m8, 1
-    vextracti128    xm11, m10, 1
-    vextracti128    xm13, m12, 1
-    vextracti128    xm1, m0, 1
-    movu            [r2], xm8
-    movu            [r2 + r3], xm9
-    movu            [r2 + r3 * 2], xm10
-    movu            [r2 + r6], xm11
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm12
-    movu            [r2 + r3], xm13
-    movu            [r2 + r3 * 2], xm0
-    movu            [r2 + r6], xm1
-%else
-    psubw           m8, m14                         ; m8 = word: row 8
-    psubw           m9, m14                         ; m9 = word: row 9
-    psubw           m10, m14                        ; m10 = word: row 10
-    psubw           m11, m14                        ; m11 = word: row 11
-    psubw           m12, m14                        ; m12 = word: row 12
-    psubw           m13, m14                        ; m13 = word: row 13
-    psubw           m0, m14                         ; m0 = word: row 14
-    psubw           m1, m14                         ; m1 = word: row 15
-    movu            [r2], m8
-    movu            [r2 + r3], m9
-    movu            [r2 + r3 * 2], m10
-    movu            [r2 + r6], m11
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], m12
-    movu            [r2 + r3], m13
-    movu            [r2 + r3 * 2], m0
-    movu            [r2 + r6], m1
-%endif
-
-    lea             r2, [r2 + r3 * 4]
-    sub             r0, r7
-    dec             r8d
-    jnz             .loop
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_16xN 16, 32, pp
-    FILTER_VER_LUMA_AVX2_16xN 16, 64, pp
-    FILTER_VER_LUMA_AVX2_16xN 16, 32, ps
-    FILTER_VER_LUMA_AVX2_16xN 16, 64, ps
-
-%macro PROCESS_LUMA_AVX2_W16_16R 1
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
-    paddw           m0, m4
-    pmaddubsw       m2, [r5]
-    lea             r7, [r0 + r1 * 4]
-    movu            xm4, [r7]                       ; m4 = row 4
-    punpckhbw       xm5, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
-    paddw           m1, m5
-    pmaddubsw       m3, [r5]
-    movu            xm5, [r7 + r1]                  ; m5 = row 5
-    punpckhbw       xm6, xm4, xm5
-    punpcklbw       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
-    paddw           m0, m6
-    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
-    paddw           m2, m6
-    pmaddubsw       m4, [r5]
-    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
-    punpckhbw       xm7, xm5, xm6
-    punpcklbw       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
-    paddw           m1, m7
-    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
-    paddw           m3, m7
-    pmaddubsw       m5, [r5]
-    movu            xm7, [r7 + r4]                  ; m7 = row 7
-    punpckhbw       xm8, xm6, xm7
-    punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm8, 1
-    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
-    paddw           m0, m8
-    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
-    paddw           m2, m8
-    pmaddubsw       m8, m6, [r5 + 1 * mmsize]
-    paddw           m4, m8
-    pmaddubsw       m6, [r5]
-    lea             r7, [r7 + r1 * 4]
-    movu            xm8, [r7]                       ; m8 = row 8
-    punpckhbw       xm9, xm7, xm8
-    punpcklbw       xm7, xm8
-    vinserti128     m7, m7, xm9, 1
-    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
-    paddw           m1, m9
-    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
-    paddw           m3, m9
-    pmaddubsw       m9, m7, [r5 + 1 * mmsize]
-    paddw           m5, m9
-    pmaddubsw       m7, [r5]
-    movu            xm9, [r7 + r1]                  ; m9 = row 9
-    punpckhbw       xm10, xm8, xm9
-    punpcklbw       xm8, xm9
-    vinserti128     m8, m8, xm10, 1
-    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
-    paddw           m2, m10
-    pmaddubsw       m10, m8, [r5 + 2 * mmsize]
-    paddw           m4, m10
-    pmaddubsw       m10, m8, [r5 + 1 * mmsize]
-    paddw           m6, m10
-    pmaddubsw       m8, [r5]
-    movu            xm10, [r7 + r1 * 2]             ; m10 = row 10
-    punpckhbw       xm11, xm9, xm10
-    punpcklbw       xm9, xm10
-    vinserti128     m9, m9, xm11, 1
-    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
-    paddw           m3, m11
-    pmaddubsw       m11, m9, [r5 + 2 * mmsize]
-    paddw           m5, m11
-    pmaddubsw       m11, m9, [r5 + 1 * mmsize]
-    paddw           m7, m11
-    pmaddubsw       m9, [r5]
-    movu            xm11, [r7 + r4]                 ; m11 = row 11
-    punpckhbw       xm12, xm10, xm11
-    punpcklbw       xm10, xm11
-    vinserti128     m10, m10, xm12, 1
-    pmaddubsw       m12, m10, [r5 + 3 * mmsize]
-    paddw           m4, m12
-    pmaddubsw       m12, m10, [r5 + 2 * mmsize]
-    paddw           m6, m12
-    pmaddubsw       m12, m10, [r5 + 1 * mmsize]
-    paddw           m8, m12
-    pmaddubsw       m10, [r5]
-    lea             r7, [r7 + r1 * 4]
-    movu            xm12, [r7]                      ; m12 = row 12
-    punpckhbw       xm13, xm11, xm12
-    punpcklbw       xm11, xm12
-    vinserti128     m11, m11, xm13, 1
-    pmaddubsw       m13, m11, [r5 + 3 * mmsize]
-    paddw           m5, m13
-    pmaddubsw       m13, m11, [r5 + 2 * mmsize]
-    paddw           m7, m13
-    pmaddubsw       m13, m11, [r5 + 1 * mmsize]
-    paddw           m9, m13
-    pmaddubsw       m11, [r5]
-
-%ifidn %1,pp
-    pmulhrsw        m0, m14                         ; m0 = word: row 0
-    pmulhrsw        m1, m14                         ; m1 = word: row 1
-    pmulhrsw        m2, m14                         ; m2 = word: row 2
-    pmulhrsw        m3, m14                         ; m3 = word: row 3
-    pmulhrsw        m4, m14                         ; m4 = word: row 4
-    pmulhrsw        m5, m14                         ; m5 = word: row 5
-    packuswb        m0, m1
-    packuswb        m2, m3
-    packuswb        m4, m5
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
-    vextracti128    xm5, m4, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-    lea             r8, [r2 + r3 * 4]
-    movu            [r8], xm4
-    movu            [r8 + r3], xm5
-%else
-    psubw           m0, m14                         ; m0 = word: row 0
-    psubw           m1, m14                         ; m1 = word: row 1
-    psubw           m2, m14                         ; m2 = word: row 2
-    psubw           m3, m14                         ; m3 = word: row 3
-    psubw           m4, m14                         ; m4 = word: row 4
-    psubw           m5, m14                         ; m5 = word: row 5
-    movu            [r2], m0
-    movu            [r2 + r3], m1
-    movu            [r2 + r3 * 2], m2
-    movu            [r2 + r6], m3
-    lea             r8, [r2 + r3 * 4]
-    movu            [r8], m4
-    movu            [r8 + r3], m5
-%endif
-
-    movu            xm13, [r7 + r1]                 ; m13 = row 13
-    punpckhbw       xm0, xm12, xm13
-    punpcklbw       xm12, xm13
-    vinserti128     m12, m12, xm0, 1
-    pmaddubsw       m0, m12, [r5 + 3 * mmsize]
-    paddw           m6, m0
-    pmaddubsw       m0, m12, [r5 + 2 * mmsize]
-    paddw           m8, m0
-    pmaddubsw       m0, m12, [r5 + 1 * mmsize]
-    paddw           m10, m0
-    pmaddubsw       m12, [r5]
-    movu            xm0, [r7 + r1 * 2]              ; m0 = row 14
-    punpckhbw       xm1, xm13, xm0
-    punpcklbw       xm13, xm0
-    vinserti128     m13, m13, xm1, 1
-    pmaddubsw       m1, m13, [r5 + 3 * mmsize]
-    paddw           m7, m1
-    pmaddubsw       m1, m13, [r5 + 2 * mmsize]
-    paddw           m9, m1
-    pmaddubsw       m1, m13, [r5 + 1 * mmsize]
-    paddw           m11, m1
-    pmaddubsw       m13, [r5]
-
-%ifidn %1,pp
-    pmulhrsw        m6, m14                         ; m6 = word: row 6
-    pmulhrsw        m7, m14                         ; m7 = word: row 7
-    packuswb        m6, m7
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm7, m6, 1
-    movu            [r8 + r3 * 2], xm6
-    movu            [r8 + r6], xm7
-%else
-    psubw           m6, m14                         ; m6 = word: row 6
-    psubw           m7, m14                         ; m7 = word: row 7
-    movu            [r8 + r3 * 2], m6
-    movu            [r8 + r6], m7
-%endif
-
-    lea             r8, [r8 + r3 * 4]
-
-    movu            xm1, [r7 + r4]                  ; m1 = row 15
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m2, m0, [r5 + 3 * mmsize]
-    paddw           m8, m2
-    pmaddubsw       m2, m0, [r5 + 2 * mmsize]
-    paddw           m10, m2
-    pmaddubsw       m2, m0, [r5 + 1 * mmsize]
-    paddw           m12, m2
-    pmaddubsw       m0, [r5]
-    lea             r7, [r7 + r1 * 4]
-    movu            xm2, [r7]                       ; m2 = row 16
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m3, m1, [r5 + 3 * mmsize]
-    paddw           m9, m3
-    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
-    paddw           m11, m3
-    pmaddubsw       m3, m1, [r5 + 1 * mmsize]
-    paddw           m13, m3
-    pmaddubsw       m1, [r5]
-    movu            xm3, [r7 + r1]                  ; m3 = row 17
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, [r5 + 3 * mmsize]
-    paddw           m10, m4
-    pmaddubsw       m4, m2, [r5 + 2 * mmsize]
-    paddw           m12, m4
-    pmaddubsw       m2, [r5 + 1 * mmsize]
-    paddw           m0, m2
-    movu            xm4, [r7 + r1 * 2]              ; m4 = row 18
-    punpckhbw       xm5, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, [r5 + 3 * mmsize]
-    paddw           m11, m5
-    pmaddubsw       m5, m3, [r5 + 2 * mmsize]
-    paddw           m13, m5
-    pmaddubsw       m3, [r5 + 1 * mmsize]
-    paddw           m1, m3
-    movu            xm5, [r7 + r4]                  ; m5 = row 19
-    punpckhbw       xm6, xm4, xm5
-    punpcklbw       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, [r5 + 3 * mmsize]
-    paddw           m12, m6
-    pmaddubsw       m4, [r5 + 2 * mmsize]
-    paddw           m0, m4
-    lea             r7, [r7 + r1 * 4]
-    movu            xm6, [r7]                       ; m6 = row 20
-    punpckhbw       xm7, xm5, xm6
-    punpcklbw       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m7, m5, [r5 + 3 * mmsize]
-    paddw           m13, m7
-    pmaddubsw       m5, [r5 + 2 * mmsize]
-    paddw           m1, m5
-    movu            xm7, [r7 + r1]                  ; m7 = row 21
-    punpckhbw       xm2, xm6, xm7
-    punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm2, 1
-    pmaddubsw       m6, [r5 + 3 * mmsize]
-    paddw           m0, m6
-    movu            xm2, [r7 + r1 * 2]              ; m2 = row 22
-    punpckhbw       xm3, xm7, xm2
-    punpcklbw       xm7, xm2
-    vinserti128     m7, m7, xm3, 1
-    pmaddubsw       m7, [r5 + 3 * mmsize]
-    paddw           m1, m7
-
-%ifidn %1,pp
-    pmulhrsw        m8, m14                         ; m8 = word: row 8
-    pmulhrsw        m9, m14                         ; m9 = word: row 9
-    pmulhrsw        m10, m14                        ; m10 = word: row 10
-    pmulhrsw        m11, m14                        ; m11 = word: row 11
-    pmulhrsw        m12, m14                        ; m12 = word: row 12
-    pmulhrsw        m13, m14                        ; m13 = word: row 13
-    pmulhrsw        m0, m14                         ; m0 = word: row 14
-    pmulhrsw        m1, m14                         ; m1 = word: row 15
-    packuswb        m8, m9
-    packuswb        m10, m11
-    packuswb        m12, m13
-    packuswb        m0, m1
-    vpermq          m8, m8, 11011000b
-    vpermq          m10, m10, 11011000b
-    vpermq          m12, m12, 11011000b
-    vpermq          m0, m0, 11011000b
-    vextracti128    xm9, m8, 1
-    vextracti128    xm11, m10, 1
-    vextracti128    xm13, m12, 1
-    vextracti128    xm1, m0, 1
-    movu            [r8], xm8
-    movu            [r8 + r3], xm9
-    movu            [r8 + r3 * 2], xm10
-    movu            [r8 + r6], xm11
-    lea             r8, [r8 + r3 * 4]
-    movu            [r8], xm12
-    movu            [r8 + r3], xm13
-    movu            [r8 + r3 * 2], xm0
-    movu            [r8 + r6], xm1
-%else
-    psubw           m8, m14                         ; m8 = word: row 8
-    psubw           m9, m14                         ; m9 = word: row 9
-    psubw           m10, m14                        ; m10 = word: row 10
-    psubw           m11, m14                        ; m11 = word: row 11
-    psubw           m12, m14                        ; m12 = word: row 12
-    psubw           m13, m14                        ; m13 = word: row 13
-    psubw           m0, m14                         ; m0 = word: row 14
-    psubw           m1, m14                         ; m1 = word: row 15
-    movu            [r8], m8
-    movu            [r8 + r3], m9
-    movu            [r8 + r3 * 2], m10
-    movu            [r8 + r6], m11
-    lea             r8, [r8 + r3 * 4]
-    movu            [r8], m12
-    movu            [r8 + r3], m13
-    movu            [r8 + r3 * 2], m0
-    movu            [r8 + r6], m1
-%endif
-%endmacro
-
-%macro PROCESS_LUMA_AVX2_W16_8R 1
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhbw       xm2, xm0, xm1
-    punpcklbw       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddubsw       m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhbw       xm3, xm1, xm2
-    punpcklbw       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhbw       xm4, xm2, xm3
-    punpcklbw       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
-    paddw           m0, m4
-    pmaddubsw       m2, [r5]
-    lea             r7, [r0 + r1 * 4]
-    movu            xm4, [r7]                       ; m4 = row 4
-    punpckhbw       xm5, xm3, xm4
-    punpcklbw       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
-    paddw           m1, m5
-    pmaddubsw       m3, [r5]
-    movu            xm5, [r7 + r1]                  ; m5 = row 5
-    punpckhbw       xm6, xm4, xm5
-    punpcklbw       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
-    paddw           m0, m6
-    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
-    paddw           m2, m6
-    pmaddubsw       m4, [r5]
-    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
-    punpckhbw       xm7, xm5, xm6
-    punpcklbw       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
-    paddw           m1, m7
-    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
-    paddw           m3, m7
-    pmaddubsw       m5, [r5]
-    movu            xm7, [r7 + r4]                  ; m7 = row 7
-    punpckhbw       xm8, xm6, xm7
-    punpcklbw       xm6, xm7
-    vinserti128     m6, m6, xm8, 1
-    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
-    paddw           m0, m8
-    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
-    paddw           m2, m8
-    pmaddubsw       m8, m6, [r5 + 1 * mmsize]
-    paddw           m4, m8
-    pmaddubsw       m6, [r5]
-    lea             r7, [r7 + r1 * 4]
-    movu            xm8, [r7]                       ; m8 = row 8
-    punpckhbw       xm9, xm7, xm8
-    punpcklbw       xm7, xm8
-    vinserti128     m7, m7, xm9, 1
-    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
-    paddw           m1, m9
-    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
-    paddw           m3, m9
-    pmaddubsw       m9, m7, [r5 + 1 * mmsize]
-    paddw           m5, m9
-    pmaddubsw       m7, [r5]
-    movu            xm9, [r7 + r1]                  ; m9 = row 9
-    punpckhbw       xm10, xm8, xm9
-    punpcklbw       xm8, xm9
-    vinserti128     m8, m8, xm10, 1
-    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
-    paddw           m2, m10
-    pmaddubsw       m10, m8, [r5 + 2 * mmsize]
-    paddw           m4, m10
-    pmaddubsw       m10, m8, [r5 + 1 * mmsize]
-    paddw           m6, m10
-    movu            xm10, [r7 + r1 * 2]             ; m10 = row 10
-    punpckhbw       xm11, xm9, xm10
-    punpcklbw       xm9, xm10
-    vinserti128     m9, m9, xm11, 1
-    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
-    paddw           m3, m11
-    pmaddubsw       m11, m9, [r5 + 2 * mmsize]
-    paddw           m5, m11
-    pmaddubsw       m11, m9, [r5 + 1 * mmsize]
-    paddw           m7, m11
-    movu            xm11, [r7 + r4]                 ; m11 = row 11
-    punpckhbw       xm12, xm10, xm11
-    punpcklbw       xm10, xm11
-    vinserti128     m10, m10, xm12, 1
-    pmaddubsw       m12, m10, [r5 + 3 * mmsize]
-    paddw           m4, m12
-    pmaddubsw       m12, m10, [r5 + 2 * mmsize]
-    paddw           m6, m12
-    lea             r7, [r7 + r1 * 4]
-    movu            xm12, [r7]                      ; m12 = row 12
-    punpckhbw       xm13, xm11, xm12
-    punpcklbw       xm11, xm12
-    vinserti128     m11, m11, xm13, 1
-    pmaddubsw       m13, m11, [r5 + 3 * mmsize]
-    paddw           m5, m13
-    pmaddubsw       m13, m11, [r5 + 2 * mmsize]
-    paddw           m7, m13
-
-%ifidn %1,pp
-    pmulhrsw        m0, m14                         ; m0 = word: row 0
-    pmulhrsw        m1, m14                         ; m1 = word: row 1
-    pmulhrsw        m2, m14                         ; m2 = word: row 2
-    pmulhrsw        m3, m14                         ; m3 = word: row 3
-    pmulhrsw        m4, m14                         ; m4 = word: row 4
-    pmulhrsw        m5, m14                         ; m5 = word: row 5
-    packuswb        m0, m1
-    packuswb        m2, m3
-    packuswb        m4, m5
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    vpermq          m4, m4, 11011000b
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
-    vextracti128    xm5, m4, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-    lea             r8, [r2 + r3 * 4]
-    movu            [r8], xm4
-    movu            [r8 + r3], xm5
-%else
-    psubw           m0, m14                         ; m0 = word: row 0
-    psubw           m1, m14                         ; m1 = word: row 1
-    psubw           m2, m14                         ; m2 = word: row 2
-    psubw           m3, m14                         ; m3 = word: row 3
-    psubw           m4, m14                         ; m4 = word: row 4
-    psubw           m5, m14                         ; m5 = word: row 5
-    movu            [r2], m0
-    movu            [r2 + r3], m1
-    movu            [r2 + r3 * 2], m2
-    movu            [r2 + r6], m3
-    lea             r8, [r2 + r3 * 4]
-    movu            [r8], m4
-    movu            [r8 + r3], m5
-%endif
-
-    movu            xm13, [r7 + r1]                 ; m13 = row 13
-    punpckhbw       xm0, xm12, xm13
-    punpcklbw       xm12, xm13
-    vinserti128     m12, m12, xm0, 1
-    pmaddubsw       m0, m12, [r5 + 3 * mmsize]
-    paddw           m6, m0
-    movu            xm0, [r7 + r1 * 2]              ; m0 = row 14
-    punpckhbw       xm1, xm13, xm0
-    punpcklbw       xm13, xm0
-    vinserti128     m13, m13, xm1, 1
-    pmaddubsw       m1, m13, [r5 + 3 * mmsize]
-    paddw           m7, m1
-
-%ifidn %1,pp
-    pmulhrsw        m6, m14                         ; m6 = word: row 6
-    pmulhrsw        m7, m14                         ; m7 = word: row 7
-    packuswb        m6, m7
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm7, m6, 1
-    movu            [r8 + r3 * 2], xm6
-    movu            [r8 + r6], xm7
-%else
-    psubw           m6, m14                         ; m6 = word: row 6
-    psubw           m7, m14                         ; m7 = word: row 7
-    movu            [r8 + r3 * 2], m6
-    movu            [r8 + r6], m7
-%endif
-%endmacro
-
-%macro FILTER_VER_LUMA_AVX2_24x32 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_24x32, 4, 11, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-%ifidn %1,ps
-    add             r3d, r3d
-    vbroadcasti128  m14, [pw_2000]
-%else
-    mova            m14, [pw_512]
-%endif
-    lea             r6, [r3 * 3]
-    lea             r10, [r1 * 4]
-    mov             r9d, 2
-.loopH:
-    PROCESS_LUMA_AVX2_W16_16R %1
-%ifidn %1,pp
-    add             r2, 16
-%else
-    add             r2, 32
-%endif
-    add             r0, 16
-    movq            xm1, [r0]                       ; m1 = row 0
-    movq            xm2, [r0 + r1]                  ; m2 = row 1
-    punpcklbw       xm1, xm2
-    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
-    punpcklbw       xm2, xm3
-    vinserti128     m5, m1, xm2, 1
-    pmaddubsw       m5, [r5]
-    movq            xm4, [r0 + r4]                  ; m4 = row 3
-    punpcklbw       xm3, xm4
-    lea             r7, [r0 + r1 * 4]
-    movq            xm1, [r7]                       ; m1 = row 4
-    punpcklbw       xm4, xm1
-    vinserti128     m2, m3, xm4, 1
-    pmaddubsw       m0, m2, [r5 + 1 * mmsize]
-    paddw           m5, m0
-    pmaddubsw       m2, [r5]
-    movq            xm3, [r7 + r1]                  ; m3 = row 5
-    punpcklbw       xm1, xm3
-    movq            xm4, [r7 + r1 * 2]              ; m4 = row 6
-    punpcklbw       xm3, xm4
-    vinserti128     m1, m1, xm3, 1
-    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
-    paddw           m5, m3
-    pmaddubsw       m0, m1, [r5 + 1 * mmsize]
-    paddw           m2, m0
-    pmaddubsw       m1, [r5]
-    movq            xm3, [r7 + r4]                  ; m3 = row 7
-    punpcklbw       xm4, xm3
-    lea             r7, [r7 + r1 * 4]
-    movq            xm0, [r7]                       ; m0 = row 8
-    punpcklbw       xm3, xm0
-    vinserti128     m4, m4, xm3, 1
-    pmaddubsw       m3, m4, [r5 + 3 * mmsize]
-    paddw           m5, m3
-    pmaddubsw       m3, m4, [r5 + 2 * mmsize]
-    paddw           m2, m3
-    pmaddubsw       m3, m4, [r5 + 1 * mmsize]
-    paddw           m1, m3
-    pmaddubsw       m4, [r5]
-    movq            xm3, [r7 + r1]                  ; m3 = row 9
-    punpcklbw       xm0, xm3
-    movq            xm6, [r7 + r1 * 2]              ; m6 = row 10
-    punpcklbw       xm3, xm6
-    vinserti128     m0, m0, xm3, 1
-    pmaddubsw       m3, m0, [r5 + 3 * mmsize]
-    paddw           m2, m3
-    pmaddubsw       m3, m0, [r5 + 2 * mmsize]
-    paddw           m1, m3
-    pmaddubsw       m3, m0, [r5 + 1 * mmsize]
-    paddw           m4, m3
-    pmaddubsw       m0, [r5]
-
-    movq            xm3, [r7 + r4]                  ; m3 = row 11
-    punpcklbw       xm6, xm3
-    lea             r7, [r7 + r1 * 4]
-    movq            xm7, [r7]                       ; m7 = row 12
-    punpcklbw       xm3, xm7
-    vinserti128     m6, m6, xm3, 1
-    pmaddubsw       m3, m6, [r5 + 3 * mmsize]
-    paddw           m1, m3
-    pmaddubsw       m3, m6, [r5 + 2 * mmsize]
-    paddw           m4, m3
-    pmaddubsw       m3, m6, [r5 + 1 * mmsize]
-    paddw           m0, m3
-    pmaddubsw       m6, [r5]
-    movq            xm3, [r7 + r1]                  ; m3 = row 13
-    punpcklbw       xm7, xm3
-    movq            xm8, [r7 + r1 * 2]              ; m8 = row 14
-    punpcklbw       xm3, xm8
-    vinserti128     m7, m7, xm3, 1
-    pmaddubsw       m3, m7, [r5 + 3 * mmsize]
-    paddw           m4, m3
-    pmaddubsw       m3, m7, [r5 + 2 * mmsize]
-    paddw           m0, m3
-    pmaddubsw       m3, m7, [r5 + 1 * mmsize]
-    paddw           m6, m3
-    pmaddubsw       m7, [r5]
-    movq            xm3, [r7 + r4]                  ; m3 = row 15
-    punpcklbw       xm8, xm3
-    lea             r7, [r7 + r1 * 4]
-    movq            xm9, [r7]                       ; m9 = row 16
-    punpcklbw       xm3, xm9
-    vinserti128     m8, m8, xm3, 1
-    pmaddubsw       m3, m8, [r5 + 3 * mmsize]
-    paddw           m0, m3
-    pmaddubsw       m3, m8, [r5 + 2 * mmsize]
-    paddw           m6, m3
-    pmaddubsw       m3, m8, [r5 + 1 * mmsize]
-    paddw           m7, m3
-    pmaddubsw       m8, [r5]
-    movq            xm3, [r7 + r1]                  ; m3 = row 17
-    punpcklbw       xm9, xm3
-    movq            xm10, [r7 + r1 * 2]             ; m10 = row 18
-    punpcklbw       xm3, xm10
-    vinserti128     m9, m9, xm3, 1
-    pmaddubsw       m3, m9, [r5 + 3 * mmsize]
-    paddw           m6, m3
-    pmaddubsw       m3, m9, [r5 + 2 * mmsize]
-    paddw           m7, m3
-    pmaddubsw       m3, m9, [r5 + 1 * mmsize]
-    paddw           m8, m3
-    movq            xm3, [r7 + r4]                  ; m3 = row 19
-    punpcklbw       xm10, xm3
-    lea             r7, [r7 + r1 * 4]
-    movq            xm9, [r7]                       ; m9 = row 20
-    punpcklbw       xm3, xm9
-    vinserti128     m10, m10, xm3, 1
-    pmaddubsw       m3, m10, [r5 + 3 * mmsize]
-    paddw           m7, m3
-    pmaddubsw       m3, m10, [r5 + 2 * mmsize]
-    paddw           m8, m3
-    movq            xm3, [r7 + r1]                  ; m3 = row 21
-    punpcklbw       xm9, xm3
-    movq            xm10, [r7 + r1 * 2]             ; m10 = row 22
-    punpcklbw       xm3, xm10
-    vinserti128     m9, m9, xm3, 1
-    pmaddubsw       m3, m9, [r5 + 3 * mmsize]
-    paddw           m8, m3
-%ifidn %1,pp
-    pmulhrsw        m5, m14                         ; m5 = word: row 0, row 1
-    pmulhrsw        m2, m14                         ; m2 = word: row 2, row 3
-    pmulhrsw        m1, m14                         ; m1 = word: row 4, row 5
-    pmulhrsw        m4, m14                         ; m4 = word: row 6, row 7
-    pmulhrsw        m0, m14                         ; m0 = word: row 8, row 9
-    pmulhrsw        m6, m14                         ; m6 = word: row 10, row 11
-    pmulhrsw        m7, m14                         ; m7 = word: row 12, row 13
-    pmulhrsw        m8, m14                         ; m8 = word: row 14, row 15
-    packuswb        m5, m2
-    packuswb        m1, m4
-    packuswb        m0, m6
-    packuswb        m7, m8
-    vextracti128    xm2, m5, 1
-    vextracti128    xm4, m1, 1
-    vextracti128    xm6, m0, 1
-    vextracti128    xm8, m7, 1
-    movq            [r2], xm5
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm5
-    movhps          [r2 + r6], xm2
-    lea             r8, [r2 + r3 * 4]
-    movq            [r8], xm1
-    movq            [r8 + r3], xm4
-    movhps          [r8 + r3 * 2], xm1
-    movhps          [r8 + r6], xm4
-    lea             r8, [r8 + r3 * 4]
-    movq            [r8], xm0
-    movq            [r8 + r3], xm6
-    movhps          [r8 + r3 * 2], xm0
-    movhps          [r8 + r6], xm6
-    lea             r8, [r8 + r3 * 4]
-    movq            [r8], xm7
-    movq            [r8 + r3], xm8
-    movhps          [r8 + r3 * 2], xm7
-    movhps          [r8 + r6], xm8
-%else
-    psubw           m5, m14                         ; m5 = word: row 0, row 1
-    psubw           m2, m14                         ; m2 = word: row 2, row 3
-    psubw           m1, m14                         ; m1 = word: row 4, row 5
-    psubw           m4, m14                         ; m4 = word: row 6, row 7
-    psubw           m0, m14                         ; m0 = word: row 8, row 9
-    psubw           m6, m14                         ; m6 = word: row 10, row 11
-    psubw           m7, m14                         ; m7 = word: row 12, row 13
-    psubw           m8, m14                         ; m8 = word: row 14, row 15
-    vextracti128    xm3, m5, 1
-    movu            [r2], xm5
-    movu            [r2 + r3], xm3
-    vextracti128    xm3, m2, 1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-    vextracti128    xm3, m1, 1
-    lea             r8, [r2 + r3 * 4]
-    movu            [r8], xm1
-    movu            [r8 + r3], xm3
-    vextracti128    xm3, m4, 1
-    movu            [r8 + r3 * 2], xm4
-    movu            [r8 + r6], xm3
-    vextracti128    xm3, m0, 1
-    lea             r8, [r8 + r3 * 4]
-    movu            [r8], xm0
-    movu            [r8 + r3], xm3
-    vextracti128    xm3, m6, 1
-    movu            [r8 + r3 * 2], xm6
-    movu            [r8 + r6], xm3
-    vextracti128    xm3, m7, 1
-    lea             r8, [r8 + r3 * 4]
-    movu            [r8], xm7
-    movu            [r8 + r3], xm3
-    vextracti128    xm3, m8, 1
-    movu            [r8 + r3 * 2], xm8
-    movu            [r8 + r6], xm3
-%endif
-    sub             r7, r10
-    lea             r0, [r7 - 16]
-%ifidn %1,pp
-    lea             r2, [r8 + r3 * 4 - 16]
-%else
-    lea             r2, [r8 + r3 * 4 - 32]
-%endif
-    dec             r9d
-    jnz             .loopH
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_24x32 pp
-    FILTER_VER_LUMA_AVX2_24x32 ps
-
-%macro FILTER_VER_LUMA_AVX2_32xN 3
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%3_%1x%2, 4, 12, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-%ifidn %3,ps
-    add             r3d, r3d
-    vbroadcasti128  m14, [pw_2000]
-%else
-    mova            m14, [pw_512]
-%endif
-    lea             r6, [r3 * 3]
-    lea             r11, [r1 * 4]
-    mov             r9d, %2 / 16
-.loopH:
-    mov             r10d, %1 / 16
-.loopW:
-    PROCESS_LUMA_AVX2_W16_16R %3
-%ifidn %3,pp
-    add             r2, 16
-%else
-    add             r2, 32
-%endif
-    add             r0, 16
-    dec             r10d
-    jnz             .loopW
-    sub             r7, r11
-    lea             r0, [r7 - 16]
-%ifidn %3,pp
-    lea             r2, [r8 + r3 * 4 - 16]
-%else
-    lea             r2, [r8 + r3 * 4 - 32]
-%endif
-    dec             r9d
-    jnz             .loopH
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_32xN 32, 32, pp
-    FILTER_VER_LUMA_AVX2_32xN 32, 64, pp
-    FILTER_VER_LUMA_AVX2_32xN 32, 32, ps
-    FILTER_VER_LUMA_AVX2_32xN 32, 64, ps
-
-%macro FILTER_VER_LUMA_AVX2_32x16 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_32x16, 4, 10, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-%ifidn %1,ps
-    add             r3d, r3d
-    vbroadcasti128  m14, [pw_2000]
-%else
-    mova            m14, [pw_512]
-%endif
-    lea             r6, [r3 * 3]
-    mov             r9d, 2
-.loopW:
-    PROCESS_LUMA_AVX2_W16_16R %1
-%ifidn %1,pp
-    add             r2, 16
-%else
-    add             r2, 32
-%endif
-    add             r0, 16
-    dec             r9d
-    jnz             .loopW
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_32x16 pp
-    FILTER_VER_LUMA_AVX2_32x16 ps
-
-%macro FILTER_VER_LUMA_AVX2_32x24 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_32x24, 4, 10, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-%ifidn %1,ps
-    add             r3d, r3d
-%endif
-    lea             r6, [r3 * 3]
-%ifidn %1,pp
-    mova            m14, [pw_512]
-%else
-    vbroadcasti128  m14, [pw_2000]
-%endif
-    mov             r9d, 2
-.loopW:
-    PROCESS_LUMA_AVX2_W16_16R %1
-%ifidn %1,pp
-    add             r2, 16
-%else
-    add             r2, 32
-%endif
-    add             r0, 16
-    dec             r9d
-    jnz             .loopW
-    lea             r9, [r1 * 4]
-    sub             r7, r9
-    lea             r0, [r7 - 16]
-%ifidn %1,pp
-    lea             r2, [r8 + r3 * 4 - 16]
-%else
-    lea             r2, [r8 + r3 * 4 - 32]
-%endif
-    mov             r9d, 2
-.loop:
-    PROCESS_LUMA_AVX2_W16_8R %1
-%ifidn %1,pp
-    add             r2, 16
-%else
-    add             r2, 32
-%endif
-    add             r0, 16
-    dec             r9d
-    jnz             .loop
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_32x24 pp
-    FILTER_VER_LUMA_AVX2_32x24 ps
-
-%macro FILTER_VER_LUMA_AVX2_32x8 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_32x8, 4, 10, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-%ifidn %1,ps
-    add             r3d, r3d
-%endif
-    lea             r6, [r3 * 3]
-%ifidn %1,pp
-    mova            m14, [pw_512]
-%else
-    vbroadcasti128  m14, [pw_2000]
-%endif
-    mov             r9d, 2
-.loopW:
-    PROCESS_LUMA_AVX2_W16_8R %1
-%ifidn %1,pp
-    add             r2, 16
-%else
-    add             r2, 32
-%endif
-    add             r0, 16
-    dec             r9d
-    jnz             .loopW
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_32x8 pp
-    FILTER_VER_LUMA_AVX2_32x8 ps
-
-%macro FILTER_VER_LUMA_AVX2_48x64 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_48x64, 4, 12, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-
-%ifidn %1,ps
-    add             r3d, r3d
-%endif
-
-    lea             r6, [r3 * 3]
-    lea             r11, [r1 * 4]
-
-%ifidn %1,pp
-    mova            m14, [pw_512]
-%else
-    vbroadcasti128  m14, [pw_2000]
-%endif
-
-    mov             r9d, 4
-.loopH:
-    mov             r10d, 3
-.loopW:
-    PROCESS_LUMA_AVX2_W16_16R %1
-%ifidn %1,pp
-    add             r2, 16
-%else
-    add             r2, 32
-%endif
-    add             r0, 16
-    dec             r10d
-    jnz             .loopW
-    sub             r7, r11
-    lea             r0, [r7 - 32]
-%ifidn %1,pp
-    lea             r2, [r8 + r3 * 4 - 32]
-%else
-    lea             r2, [r8 + r3 * 4 - 64]
-%endif
-    dec             r9d
-    jnz             .loopH
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_48x64 pp
-    FILTER_VER_LUMA_AVX2_48x64 ps
-
-%macro FILTER_VER_LUMA_AVX2_64xN 3
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%3_%1x%2, 4, 12, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-
-%ifidn %3,ps
-    add             r3d, r3d
-%endif
-
-    lea             r6, [r3 * 3]
-    lea             r11, [r1 * 4]
-
-%ifidn %3,pp
-    mova            m14, [pw_512]
-%else
-    vbroadcasti128  m14, [pw_2000]
-%endif
-
-    mov             r9d, %2 / 16
-.loopH:
-    mov             r10d, %1 / 16
-.loopW:
-    PROCESS_LUMA_AVX2_W16_16R %3
-%ifidn %3,pp
-    add             r2, 16
-%else
-    add             r2, 32
-%endif
-    add             r0, 16
-    dec             r10d
-    jnz             .loopW
-    sub             r7, r11
-    lea             r0, [r7 - 48]
-%ifidn %3,pp
-    lea             r2, [r8 + r3 * 4 - 48]
-%else
-    lea             r2, [r8 + r3 * 4 - 96]
-%endif
-    dec             r9d
-    jnz             .loopH
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_64xN 64, 32, pp
-    FILTER_VER_LUMA_AVX2_64xN 64, 48, pp
-    FILTER_VER_LUMA_AVX2_64xN 64, 64, pp
-    FILTER_VER_LUMA_AVX2_64xN 64, 32, ps
-    FILTER_VER_LUMA_AVX2_64xN 64, 48, ps
-    FILTER_VER_LUMA_AVX2_64xN 64, 64, ps
-
-%macro FILTER_VER_LUMA_AVX2_64x16 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_64x16, 4, 10, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer_32]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer_32 + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-
-%ifidn %1,ps
-    add             r3d, r3d
-%endif
-
-    lea             r6, [r3 * 3]
-
-%ifidn %1,pp
-    mova            m14, [pw_512]
-%else
-    vbroadcasti128  m14, [pw_2000]
-%endif
-
-    mov             r9d, 4
-.loopW:
-    PROCESS_LUMA_AVX2_W16_16R %1
-%ifidn %1,pp
-    add             r2, 16
-%else
-    add             r2, 32
-%endif
-    add             r0, 16
-    dec             r9d
-    jnz             .loopW
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_LUMA_AVX2_64x16 pp
-    FILTER_VER_LUMA_AVX2_64x16 ps
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_LUMA 3
-INIT_XMM sse4
-cglobal interp_8tap_vert_%3_%1x%2, 5, 7, 8 ,0-gprsize
-    lea       r5, [3 * r1]
-    sub       r0, r5
-    shl       r4d, 6
-%ifidn %3,ps
-    add       r3d, r3d
-%endif
-
-%ifdef PIC
-    lea       r5, [tab_LumaCoeffVer]
-    lea       r6, [r5 + r4]
-%else
-    lea       r6, [tab_LumaCoeffVer + r4]
-%endif
-
-%ifidn %3,pp
-    mova      m3, [pw_512]
-%else
-    mova      m3, [pw_2000]
-%endif
-    mov       dword [rsp], %2/4
-
-.loopH:
-    mov       r4d, (%1/8)
-.loopW:
-    PROCESS_LUMA_W8_4R
-%ifidn %3,pp
-    pmulhrsw  m7, m3
-    pmulhrsw  m6, m3
-    pmulhrsw  m5, m3
-    pmulhrsw  m4, m3
-
-    packuswb  m7, m6
-    packuswb  m5, m4
-
-    movlps    [r2], m7
-    movhps    [r2 + r3], m7
-    lea       r5, [r2 + 2 * r3]
-    movlps    [r5], m5
-    movhps    [r5 + r3], m5
-%else
-    psubw     m7, m3
-    psubw     m6, m3
-    psubw     m5, m3
-    psubw     m4, m3
-
-    movu      [r2], m7
-    movu      [r2 + r3], m6
-    lea       r5, [r2 + 2 * r3]
-    movu      [r5], m5
-    movu      [r5 + r3], m4
-%endif
-
-    lea       r5, [8 * r1 - 8]
-    sub       r0, r5
-%ifidn %3,pp
-    add       r2, 8
-%else
-    add       r2, 16
-%endif
-    dec       r4d
-    jnz       .loopW
-
-    lea       r0, [r0 + 4 * r1 - %1]
-%ifidn %3,pp
-    lea       r2, [r2 + 4 * r3 - %1]
-%else
-    lea       r2, [r2 + 4 * r3 - 2 * %1]
-%endif
-
-    dec       dword [rsp]
-    jnz       .loopH
-
-    RET
-%endmacro
-
-    FILTER_VER_LUMA 16, 4, pp
-    FILTER_VER_LUMA 16, 8, pp
-    FILTER_VER_LUMA 16, 12, pp
-    FILTER_VER_LUMA 16, 16, pp
-    FILTER_VER_LUMA 16, 32, pp
-    FILTER_VER_LUMA 16, 64, pp
-    FILTER_VER_LUMA 24, 32, pp
-    FILTER_VER_LUMA 32, 8, pp
-    FILTER_VER_LUMA 32, 16, pp
-    FILTER_VER_LUMA 32, 24, pp
-    FILTER_VER_LUMA 32, 32, pp
-    FILTER_VER_LUMA 32, 64, pp
-    FILTER_VER_LUMA 48, 64, pp
-    FILTER_VER_LUMA 64, 16, pp
-    FILTER_VER_LUMA 64, 32, pp
-    FILTER_VER_LUMA 64, 48, pp
-    FILTER_VER_LUMA 64, 64, pp
-
-    FILTER_VER_LUMA 16, 4, ps
-    FILTER_VER_LUMA 16, 8, ps
-    FILTER_VER_LUMA 16, 12, ps
-    FILTER_VER_LUMA 16, 16, ps
-    FILTER_VER_LUMA 16, 32, ps
-    FILTER_VER_LUMA 16, 64, ps
-    FILTER_VER_LUMA 24, 32, ps
-    FILTER_VER_LUMA 32, 8, ps
-    FILTER_VER_LUMA 32, 16, ps
-    FILTER_VER_LUMA 32, 24, ps
-    FILTER_VER_LUMA 32, 32, ps
-    FILTER_VER_LUMA 32, 64, ps
-    FILTER_VER_LUMA 48, 64, ps
-    FILTER_VER_LUMA 64, 16, ps
-    FILTER_VER_LUMA 64, 32, ps
-    FILTER_VER_LUMA 64, 48, ps
-    FILTER_VER_LUMA 64, 64, ps
-
-%macro PROCESS_LUMA_SP_W4_4R 0
-    movq       m0, [r0]
-    movq       m1, [r0 + r1]
-    punpcklwd  m0, m1                          ;m0=[0 1]
-    pmaddwd    m0, [r6 + 0 *16]                ;m0=[0+1]  Row1
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m4, [r0]
-    punpcklwd  m1, m4                          ;m1=[1 2]
-    pmaddwd    m1, [r6 + 0 *16]                ;m1=[1+2]  Row2
-
-    movq       m5, [r0 + r1]
-    punpcklwd  m4, m5                          ;m4=[2 3]
-    pmaddwd    m2, m4, [r6 + 0 *16]            ;m2=[2+3]  Row3
-    pmaddwd    m4, [r6 + 1 * 16]
-    paddd      m0, m4                          ;m0=[0+1+2+3]  Row1
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m4, [r0]
-    punpcklwd  m5, m4                          ;m5=[3 4]
-    pmaddwd    m3, m5, [r6 + 0 *16]            ;m3=[3+4]  Row4
-    pmaddwd    m5, [r6 + 1 * 16]
-    paddd      m1, m5                          ;m1 = [1+2+3+4]  Row2
-
-    movq       m5, [r0 + r1]
-    punpcklwd  m4, m5                          ;m4=[4 5]
-    pmaddwd    m6, m4, [r6 + 1 * 16]
-    paddd      m2, m6                          ;m2=[2+3+4+5]  Row3
-    pmaddwd    m4, [r6 + 2 * 16]
-    paddd      m0, m4                          ;m0=[0+1+2+3+4+5]  Row1
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m4, [r0]
-    punpcklwd  m5, m4                          ;m5=[5 6]
-    pmaddwd    m6, m5, [r6 + 1 * 16]
-    paddd      m3, m6                          ;m3=[3+4+5+6]  Row4
-    pmaddwd    m5, [r6 + 2 * 16]
-    paddd      m1, m5                          ;m1=[1+2+3+4+5+6]  Row2
-
-    movq       m5, [r0 + r1]
-    punpcklwd  m4, m5                          ;m4=[6 7]
-    pmaddwd    m6, m4, [r6 + 2 * 16]
-    paddd      m2, m6                          ;m2=[2+3+4+5+6+7]  Row3
-    pmaddwd    m4, [r6 + 3 * 16]
-    paddd      m0, m4                          ;m0=[0+1+2+3+4+5+6+7]  Row1 end
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m4, [r0]
-    punpcklwd  m5, m4                          ;m5=[7 8]
-    pmaddwd    m6, m5, [r6 + 2 * 16]
-    paddd      m3, m6                          ;m3=[3+4+5+6+7+8]  Row4
-    pmaddwd    m5, [r6 + 3 * 16]
-    paddd      m1, m5                          ;m1=[1+2+3+4+5+6+7+8]  Row2 end
-
-    movq       m5, [r0 + r1]
-    punpcklwd  m4, m5                          ;m4=[8 9]
-    pmaddwd    m4, [r6 + 3 * 16]
-    paddd      m2, m4                          ;m2=[2+3+4+5+6+7+8+9]  Row3 end
-
-    movq       m4, [r0 + 2 * r1]
-    punpcklwd  m5, m4                          ;m5=[9 10]
-    pmaddwd    m5, [r6 + 3 * 16]
-    paddd      m3, m5                          ;m3=[3+4+5+6+7+8+9+10]  Row4 end
-%endmacro
-
-;--------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_sp_%1x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_LUMA_SP 2
-INIT_XMM sse4
-cglobal interp_8tap_vert_sp_%1x%2, 5, 7, 8 ,0-gprsize
-
-    add       r1d, r1d
-    lea       r5, [r1 + 2 * r1]
-    sub       r0, r5
-    shl       r4d, 6
-
-%ifdef PIC
-    lea       r5, [tab_LumaCoeffV]
-    lea       r6, [r5 + r4]
-%else
-    lea       r6, [tab_LumaCoeffV + r4]
-%endif
-
-    mova      m7, [pd_526336]
-
-    mov       dword [rsp], %2/4
-.loopH:
-    mov       r4d, (%1/4)
-.loopW:
-    PROCESS_LUMA_SP_W4_4R
-
-    paddd     m0, m7
-    paddd     m1, m7
-    paddd     m2, m7
-    paddd     m3, m7
-
-    psrad     m0, 12
-    psrad     m1, 12
-    psrad     m2, 12
-    psrad     m3, 12
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-
-    packuswb  m0, m2
-
-    movd      [r2], m0
-    pextrd    [r2 + r3], m0, 1
-    lea       r5, [r2 + 2 * r3]
-    pextrd    [r5], m0, 2
-    pextrd    [r5 + r3], m0, 3
-
-    lea       r5, [8 * r1 - 2 * 4]
-    sub       r0, r5
-    add       r2, 4
-
-    dec       r4d
-    jnz       .loopW
-
-    lea       r0, [r0 + 4 * r1 - 2 * %1]
-    lea       r2, [r2 + 4 * r3 - %1]
-
-    dec       dword [rsp]
-    jnz       .loopH
-
-    RET
-%endmacro
-
-;--------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_sp_%1x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_SP 4, 4
-    FILTER_VER_LUMA_SP 8, 8
-    FILTER_VER_LUMA_SP 8, 4
-    FILTER_VER_LUMA_SP 4, 8
-    FILTER_VER_LUMA_SP 16, 16
-    FILTER_VER_LUMA_SP 16, 8
-    FILTER_VER_LUMA_SP 8, 16
-    FILTER_VER_LUMA_SP 16, 12
-    FILTER_VER_LUMA_SP 12, 16
-    FILTER_VER_LUMA_SP 16, 4
-    FILTER_VER_LUMA_SP 4, 16
-    FILTER_VER_LUMA_SP 32, 32
-    FILTER_VER_LUMA_SP 32, 16
-    FILTER_VER_LUMA_SP 16, 32
-    FILTER_VER_LUMA_SP 32, 24
-    FILTER_VER_LUMA_SP 24, 32
-    FILTER_VER_LUMA_SP 32, 8
-    FILTER_VER_LUMA_SP 8, 32
-    FILTER_VER_LUMA_SP 64, 64
-    FILTER_VER_LUMA_SP 64, 32
-    FILTER_VER_LUMA_SP 32, 64
-    FILTER_VER_LUMA_SP 64, 48
-    FILTER_VER_LUMA_SP 48, 64
-    FILTER_VER_LUMA_SP 64, 16
-    FILTER_VER_LUMA_SP 16, 64
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-INIT_XMM sse4
-cglobal filterPixelToShort_4x2, 3, 4, 3
-    mov         r3d, r3m
-    add         r3d, r3d
-
-    ; load constant
-    mova        m1, [pb_128]
-    mova        m2, [tab_c_64_n64]
-
-    movd        m0, [r0]
-    pinsrd      m0, [r0 + r1], 1
-    punpcklbw   m0, m1
-    pmaddubsw   m0, m2
-
-    movq        [r2 + r3 * 0], m0
-    movhps      [r2 + r3 * 1], m0
-
-    RET
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
-;-----------------------------------------------------------------------------
-INIT_XMM ssse3
-cglobal filterPixelToShort_8x2, 3, 4, 3
-    mov         r3d, r3m
-    add         r3d, r3d
-
-    ; load constant
-    mova        m1, [pb_128]
-    mova        m2, [tab_c_64_n64]
-
-    movh        m0, [r0]
-    punpcklbw   m0, m1
-    pmaddubsw   m0, m2
-    movu        [r2 + r3 * 0], m0
-
-    movh        m0, [r0 + r1]
-    punpcklbw   m0, m1
-    pmaddubsw   m0, m2
-    movu        [r2 + r3 * 1], m0
-
-    RET
-
-%macro PROCESS_CHROMA_SP_W4_4R 0
-    movq       m0, [r0]
-    movq       m1, [r0 + r1]
-    punpcklwd  m0, m1                          ;m0=[0 1]
-    pmaddwd    m0, [r6 + 0 *16]                ;m0=[0+1]         Row1
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m4, [r0]
-    punpcklwd  m1, m4                          ;m1=[1 2]
-    pmaddwd    m1, [r6 + 0 *16]                ;m1=[1+2]         Row2
-
-    movq       m5, [r0 + r1]
-    punpcklwd  m4, m5                          ;m4=[2 3]
-    pmaddwd    m2, m4, [r6 + 0 *16]            ;m2=[2+3]         Row3
-    pmaddwd    m4, [r6 + 1 * 16]
-    paddd      m0, m4                          ;m0=[0+1+2+3]     Row1 done
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m4, [r0]
-    punpcklwd  m5, m4                          ;m5=[3 4]
-    pmaddwd    m3, m5, [r6 + 0 *16]            ;m3=[3+4]         Row4
-    pmaddwd    m5, [r6 + 1 * 16]
-    paddd      m1, m5                          ;m1 = [1+2+3+4]   Row2
-
-    movq       m5, [r0 + r1]
-    punpcklwd  m4, m5                          ;m4=[4 5]
-    pmaddwd    m4, [r6 + 1 * 16]
-    paddd      m2, m4                          ;m2=[2+3+4+5]     Row3
-
-    movq       m4, [r0 + 2 * r1]
-    punpcklwd  m5, m4                          ;m5=[5 6]
-    pmaddwd    m5, [r6 + 1 * 16]
-    paddd      m3, m5                          ;m3=[3+4+5+6]     Row4
-%endmacro
-
-;--------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_sp_%1x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_SP 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_sp_%1x%2, 5, 7, 7 ,0-gprsize
-
-    add       r1d, r1d
-    sub       r0, r1
-    shl       r4d, 5
-
-%ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
-    lea       r6, [r5 + r4]
-%else
-    lea       r6, [tab_ChromaCoeffV + r4]
-%endif
-
-    mova      m6, [pd_526336]
-
-    mov       dword [rsp], %2/4
-
-.loopH:
-    mov       r4d, (%1/4)
-.loopW:
-    PROCESS_CHROMA_SP_W4_4R
-
-    paddd     m0, m6
-    paddd     m1, m6
-    paddd     m2, m6
-    paddd     m3, m6
-
-    psrad     m0, 12
-    psrad     m1, 12
-    psrad     m2, 12
-    psrad     m3, 12
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-
-    packuswb  m0, m2
-
-    movd      [r2], m0
-    pextrd    [r2 + r3], m0, 1
-    lea       r5, [r2 + 2 * r3]
-    pextrd    [r5], m0, 2
-    pextrd    [r5 + r3], m0, 3
-
-    lea       r5, [4 * r1 - 2 * 4]
-    sub       r0, r5
-    add       r2, 4
-
-    dec       r4d
-    jnz       .loopW
-
-    lea       r0, [r0 + 4 * r1 - 2 * %1]
-    lea       r2, [r2 + 4 * r3 - %1]
-
-    dec       dword [rsp]
-    jnz       .loopH
-
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_SP 4, 4
-    FILTER_VER_CHROMA_SP 4, 8
-    FILTER_VER_CHROMA_SP 16, 16
-    FILTER_VER_CHROMA_SP 16, 8
-    FILTER_VER_CHROMA_SP 16, 12
-    FILTER_VER_CHROMA_SP 12, 16
-    FILTER_VER_CHROMA_SP 16, 4
-    FILTER_VER_CHROMA_SP 4, 16
-    FILTER_VER_CHROMA_SP 32, 32
-    FILTER_VER_CHROMA_SP 32, 16
-    FILTER_VER_CHROMA_SP 16, 32
-    FILTER_VER_CHROMA_SP 32, 24
-    FILTER_VER_CHROMA_SP 24, 32
-    FILTER_VER_CHROMA_SP 32, 8
-
-    FILTER_VER_CHROMA_SP 16, 24
-    FILTER_VER_CHROMA_SP 16, 64
-    FILTER_VER_CHROMA_SP 12, 32
-    FILTER_VER_CHROMA_SP 4, 32
-    FILTER_VER_CHROMA_SP 32, 64
-    FILTER_VER_CHROMA_SP 32, 48
-    FILTER_VER_CHROMA_SP 24, 64
-
-    FILTER_VER_CHROMA_SP 64, 64
-    FILTER_VER_CHROMA_SP 64, 32
-    FILTER_VER_CHROMA_SP 64, 48
-    FILTER_VER_CHROMA_SP 48, 64
-    FILTER_VER_CHROMA_SP 64, 16
-
-
-%macro PROCESS_CHROMA_SP_W2_4R 1
-    movd       m0, [r0]
-    movd       m1, [r0 + r1]
-    punpcklwd  m0, m1                          ;m0=[0 1]
-
-    lea        r0, [r0 + 2 * r1]
-    movd       m2, [r0]
-    punpcklwd  m1, m2                          ;m1=[1 2]
-    punpcklqdq m0, m1                          ;m0=[0 1 1 2]
-    pmaddwd    m0, [%1 + 0 *16]                ;m0=[0+1 1+2] Row 1-2
-
-    movd       m1, [r0 + r1]
-    punpcklwd  m2, m1                          ;m2=[2 3]
-
-    lea        r0, [r0 + 2 * r1]
-    movd       m3, [r0]
-    punpcklwd  m1, m3                          ;m2=[3 4]
-    punpcklqdq m2, m1                          ;m2=[2 3 3 4]
-
-    pmaddwd    m4, m2, [%1 + 1 * 16]           ;m4=[2+3 3+4] Row 1-2
-    pmaddwd    m2, [%1 + 0 * 16]               ;m2=[2+3 3+4] Row 3-4
-    paddd      m0, m4                          ;m0=[0+1+2+3 1+2+3+4] Row 1-2
-
-    movd       m1, [r0 + r1]
-    punpcklwd  m3, m1                          ;m3=[4 5]
-
-    movd       m4, [r0 + 2 * r1]
-    punpcklwd  m1, m4                          ;m1=[5 6]
-    punpcklqdq m3, m1                          ;m2=[4 5 5 6]
-    pmaddwd    m3, [%1 + 1 * 16]               ;m3=[4+5 5+6] Row 3-4
-    paddd      m2, m3                          ;m2=[2+3+4+5 3+4+5+6] Row 3-4
-%endmacro
-
-;-------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vertical_sp_%1x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_SP_W2_4R 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_sp_%1x%2, 5, 6, 6
-
-    add       r1d, r1d
-    sub       r0, r1
-    shl       r4d, 5
-
-%ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
-    lea       r5, [r5 + r4]
-%else
-    lea       r5, [tab_ChromaCoeffV + r4]
-%endif
-
-    mova      m5, [pd_526336]
-
-    mov       r4d, (%2/4)
-
-.loopH:
-    PROCESS_CHROMA_SP_W2_4R r5
-
-    paddd     m0, m5
-    paddd     m2, m5
-
-    psrad     m0, 12
-    psrad     m2, 12
-
-    packssdw  m0, m2
-    packuswb  m0, m0
-
-    pextrw    [r2], m0, 0
-    pextrw    [r2 + r3], m0, 1
-    lea       r2, [r2 + 2 * r3]
-    pextrw    [r2], m0, 2
-    pextrw    [r2 + r3], m0, 3
-
-    lea       r2, [r2 + 2 * r3]
-
-    dec       r4d
-    jnz       .loopH
-
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_SP_W2_4R 2, 4
-    FILTER_VER_CHROMA_SP_W2_4R 2, 8
-
-    FILTER_VER_CHROMA_SP_W2_4R 2, 16
-
-;--------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_sp_4x2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-INIT_XMM sse4
-cglobal interp_4tap_vert_sp_4x2, 5, 6, 5
-
-    add        r1d, r1d
-    sub        r0, r1
-    shl        r4d, 5
-
-%ifdef PIC
-    lea        r5, [tab_ChromaCoeffV]
-    lea        r5, [r5 + r4]
-%else
-    lea        r5, [tab_ChromaCoeffV + r4]
-%endif
-
-    mova       m4, [pd_526336]
-
-    movq       m0, [r0]
-    movq       m1, [r0 + r1]
-    punpcklwd  m0, m1                          ;m0=[0 1]
-    pmaddwd    m0, [r5 + 0 *16]                ;m0=[0+1]  Row1
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m2, [r0]
-    punpcklwd  m1, m2                          ;m1=[1 2]
-    pmaddwd    m1, [r5 + 0 *16]                ;m1=[1+2]  Row2
-
-    movq       m3, [r0 + r1]
-    punpcklwd  m2, m3                          ;m4=[2 3]
-    pmaddwd    m2, [r5 + 1 * 16]
-    paddd      m0, m2                          ;m0=[0+1+2+3]  Row1 done
-    paddd      m0, m4
-    psrad      m0, 12
-
-    movq       m2, [r0 + 2 * r1]
-    punpcklwd  m3, m2                          ;m5=[3 4]
-    pmaddwd    m3, [r5 + 1 * 16]
-    paddd      m1, m3                          ;m1 = [1+2+3+4]  Row2 done
-    paddd      m1, m4
-    psrad      m1, 12
-
-    packssdw   m0, m1
-    packuswb   m0, m0
-
-    movd       [r2], m0
-    pextrd     [r2 + r3], m0, 1
-
-    RET
-
-;-------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vertical_sp_6x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_SP_W6_H4 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_sp_6x%2, 5, 7, 7
-
-    add       r1d, r1d
-    sub       r0, r1
-    shl       r4d, 5
-
-%ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
-    lea       r6, [r5 + r4]
-%else
-    lea       r6, [tab_ChromaCoeffV + r4]
-%endif
-
-    mova      m6, [pd_526336]
-
-    mov       r4d, %2/4
-
-.loopH:
-    PROCESS_CHROMA_SP_W4_4R
-
-    paddd     m0, m6
-    paddd     m1, m6
-    paddd     m2, m6
-    paddd     m3, m6
-
-    psrad     m0, 12
-    psrad     m1, 12
-    psrad     m2, 12
-    psrad     m3, 12
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-
-    packuswb  m0, m2
-
-    movd      [r2], m0
-    pextrd    [r2 + r3], m0, 1
-    lea       r5, [r2 + 2 * r3]
-    pextrd    [r5], m0, 2
-    pextrd    [r5 + r3], m0, 3
-
-    lea       r5, [4 * r1 - 2 * 4]
-    sub       r0, r5
-    add       r2, 4
-
-    PROCESS_CHROMA_SP_W2_4R r6
-
-    paddd     m0, m6
-    paddd     m2, m6
-
-    psrad     m0, 12
-    psrad     m2, 12
-
-    packssdw  m0, m2
-    packuswb  m0, m0
-
-    pextrw    [r2], m0, 0
-    pextrw    [r2 + r3], m0, 1
-    lea       r2, [r2 + 2 * r3]
-    pextrw    [r2], m0, 2
-    pextrw    [r2 + r3], m0, 3
-
-    sub       r0, 2 * 4
-    lea       r2, [r2 + 2 * r3 - 4]
-
-    dec       r4d
-    jnz       .loopH
-
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_SP_W6_H4 6, 8
-
-    FILTER_VER_CHROMA_SP_W6_H4 6, 16
-
-%macro PROCESS_CHROMA_SP_W8_2R 0
-    movu       m1, [r0]
-    movu       m3, [r0 + r1]
-    punpcklwd  m0, m1, m3
-    pmaddwd    m0, [r5 + 0 * 16]                ;m0 = [0l+1l]  Row1l
-    punpckhwd  m1, m3
-    pmaddwd    m1, [r5 + 0 * 16]                ;m1 = [0h+1h]  Row1h
-
-    movu       m4, [r0 + 2 * r1]
-    punpcklwd  m2, m3, m4
-    pmaddwd    m2, [r5 + 0 * 16]                ;m2 = [1l+2l]  Row2l
-    punpckhwd  m3, m4
-    pmaddwd    m3, [r5 + 0 * 16]                ;m3 = [1h+2h]  Row2h
-
-    lea        r0, [r0 + 2 * r1]
-    movu       m5, [r0 + r1]
-    punpcklwd  m6, m4, m5
-    pmaddwd    m6, [r5 + 1 * 16]                ;m6 = [2l+3l]  Row1l
-    paddd      m0, m6                           ;m0 = [0l+1l+2l+3l]  Row1l sum
-    punpckhwd  m4, m5
-    pmaddwd    m4, [r5 + 1 * 16]                ;m6 = [2h+3h]  Row1h
-    paddd      m1, m4                           ;m1 = [0h+1h+2h+3h]  Row1h sum
-
-    movu       m4, [r0 + 2 * r1]
-    punpcklwd  m6, m5, m4
-    pmaddwd    m6, [r5 + 1 * 16]                ;m6 = [3l+4l]  Row2l
-    paddd      m2, m6                           ;m2 = [1l+2l+3l+4l]  Row2l sum
-    punpckhwd  m5, m4
-    pmaddwd    m5, [r5 + 1 * 16]                ;m1 = [3h+4h]  Row2h
-    paddd      m3, m5                           ;m3 = [1h+2h+3h+4h]  Row2h sum
-%endmacro
-
-;--------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_sp_8x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_SP_W8_H2 2
-INIT_XMM sse2
-cglobal interp_4tap_vert_sp_%1x%2, 5, 6, 8
-
-    add       r1d, r1d
-    sub       r0, r1
-    shl       r4d, 5
-
-%ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
-    lea       r5, [r5 + r4]
-%else
-    lea       r5, [tab_ChromaCoeffV + r4]
-%endif
-
-    mova      m7, [pd_526336]
-
-    mov       r4d, %2/2
-.loopH:
-    PROCESS_CHROMA_SP_W8_2R
-
-    paddd     m0, m7
-    paddd     m1, m7
-    paddd     m2, m7
-    paddd     m3, m7
-
-    psrad     m0, 12
-    psrad     m1, 12
-    psrad     m2, 12
-    psrad     m3, 12
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-
-    packuswb  m0, m2
-
-    movlps    [r2], m0
-    movhps    [r2 + r3], m0
-
-    lea       r2, [r2 + 2 * r3]
-
-    dec r4d
-    jnz .loopH
-
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_SP_W8_H2 8, 2
-    FILTER_VER_CHROMA_SP_W8_H2 8, 4
-    FILTER_VER_CHROMA_SP_W8_H2 8, 6
-    FILTER_VER_CHROMA_SP_W8_H2 8, 8
-    FILTER_VER_CHROMA_SP_W8_H2 8, 16
-    FILTER_VER_CHROMA_SP_W8_H2 8, 32
-
-    FILTER_VER_CHROMA_SP_W8_H2 8, 12
-    FILTER_VER_CHROMA_SP_W8_H2 8, 64
-
-
-;---------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ps_%1x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;---------------------------------------------------------------------------------------------------------------
-%macro FILTER_V_PS_W16n 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_ps_%1x%2, 4, 7, 8
-
-    mov        r4d, r4m
-    sub        r0, r1
-    add        r3d, r3d
-
-%ifdef PIC
-    lea        r5, [tab_ChromaCoeff]
-    movd       m0, [r5 + r4 * 4]
-%else
-    movd       m0, [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    pshufb     m1, m0, [tab_Vm]
-    pshufb     m0, [tab_Vm + 16]
-    mov        r4d, %2/2
-
-.loop:
-
-    mov         r6d,       %1/16
-
-.loopW:
-
-    movu       m2, [r0]
-    movu       m3, [r0 + r1]
-
-    punpcklbw  m4, m2, m3
-    punpckhbw  m2, m3
-
-    pmaddubsw  m4, m1
-    pmaddubsw  m2, m1
-
-    lea        r5, [r0 + 2 * r1]
-    movu       m5, [r5]
-    movu       m7, [r5 + r1]
-
-    punpcklbw  m6, m5, m7
-    pmaddubsw  m6, m0
-    paddw      m4, m6
-
-    punpckhbw  m6, m5, m7
-    pmaddubsw  m6, m0
-    paddw      m2, m6
-
-    mova       m6, [pw_2000]
-
-    psubw      m4, m6
-    psubw      m2, m6
-
-    movu       [r2], m4
-    movu       [r2 + 16], m2
-
-    punpcklbw  m4, m3, m5
-    punpckhbw  m3, m5
-
-    pmaddubsw  m4, m1
-    pmaddubsw  m3, m1
-
-    movu       m5, [r5 + 2 * r1]
-
-    punpcklbw  m2, m7, m5
-    punpckhbw  m7, m5
-
-    pmaddubsw  m2, m0
-    pmaddubsw  m7, m0
-
-    paddw      m4, m2
-    paddw      m3, m7
-
-    psubw      m4, m6
-    psubw      m3, m6
-
-    movu       [r2 + r3], m4
-    movu       [r2 + r3 + 16], m3
-
-    add         r0,        16
-    add         r2,        32
-    dec         r6d
-    jnz         .loopW
-
-    lea         r0,        [r0 + r1 * 2 - %1]
-    lea         r2,        [r2 + r3 * 2 - %1 * 2]
-
-    dec        r4d
-    jnz        .loop
-    RET
-%endmacro
-
-    FILTER_V_PS_W16n 64, 64
-    FILTER_V_PS_W16n 64, 32
-    FILTER_V_PS_W16n 64, 48
-    FILTER_V_PS_W16n 48, 64
-    FILTER_V_PS_W16n 64, 16
-
-
-;------------------------------------------------------------------------------------------------------------
-;void interp_4tap_vert_ps_2x4(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;------------------------------------------------------------------------------------------------------------
-INIT_XMM sse4
-cglobal interp_4tap_vert_ps_2x4, 4, 6, 7
-
-    mov         r4d, r4m
-    sub         r0, r1
-    add         r3d, r3d
-
-%ifdef PIC
-    lea         r5, [tab_ChromaCoeff]
-    movd        m0, [r5 + r4 * 4]
-%else
-    movd        m0, [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    pshufb      m0, [tab_Cm]
-
-    lea         r5, [3 * r1]
-
-    movd        m2, [r0]
-    movd        m3, [r0 + r1]
-    movd        m4, [r0 + 2 * r1]
-    movd        m5, [r0 + r5]
-
-    punpcklbw   m2, m3
-    punpcklbw   m6, m4, m5
-    punpcklbw   m2, m6
-
-    pmaddubsw   m2, m0
-
-    lea         r0, [r0 + 4 * r1]
-    movd        m6, [r0]
-
-    punpcklbw   m3, m4
-    punpcklbw   m1, m5, m6
-    punpcklbw   m3, m1
-
-    pmaddubsw   m3, m0
-    phaddw      m2, m3
-
-    mova        m1, [pw_2000]
-
-    psubw       m2, m1
-
-    movd        [r2], m2
-    pextrd      [r2 + r3], m2, 2
-
-    movd        m2, [r0 + r1]
-
-    punpcklbw   m4, m5
-    punpcklbw   m3, m6, m2
-    punpcklbw   m4, m3
-
-    pmaddubsw   m4, m0
-
-    movd        m3, [r0 + 2 * r1]
-
-    punpcklbw   m5, m6
-    punpcklbw   m2, m3
-    punpcklbw   m5, m2
-
-    pmaddubsw   m5, m0
-    phaddw      m4, m5
-    psubw       m4, m1
-
-    lea         r2, [r2 + 2 * r3]
-    movd        [r2], m4
-    pextrd      [r2 + r3], m4, 2
-
-    RET
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ps_2x8(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-%macro FILTER_V_PS_W2 2
-INIT_XMM sse4
-cglobal interp_4tap_vert_ps_2x%2, 4, 6, 8
-
-    mov        r4d, r4m
-    sub        r0, r1
-    add        r3d, r3d
-
-%ifdef PIC
-    lea        r5, [tab_ChromaCoeff]
-    movd       m0, [r5 + r4 * 4]
-%else
-    movd       m0, [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    pshufb     m0, [tab_Cm]
-
-    mova       m1, [pw_2000]
-    lea        r5, [3 * r1]
-    mov        r4d, %2/4
-.loop:
-    movd       m2, [r0]
-    movd       m3, [r0 + r1]
-    movd       m4, [r0 + 2 * r1]
-    movd       m5, [r0 + r5]
-
-    punpcklbw  m2, m3
-    punpcklbw  m6, m4, m5
-    punpcklbw  m2, m6
-
-    pmaddubsw  m2, m0
-
-    lea        r0, [r0 + 4 * r1]
-    movd       m6, [r0]
-
-    punpcklbw  m3, m4
-    punpcklbw  m7, m5, m6
-    punpcklbw  m3, m7
-
-    pmaddubsw  m3, m0
-
-    phaddw     m2, m3
-    psubw      m2, m1
-
-
-    movd       [r2], m2
-    pshufd     m2, m2, 2
-    movd       [r2 + r3], m2
-
-    movd       m2, [r0 + r1]
-
-    punpcklbw  m4, m5
-    punpcklbw  m3, m6, m2
-    punpcklbw  m4, m3
-
-    pmaddubsw  m4, m0
-
-    movd       m3, [r0 + 2 * r1]
-
-    punpcklbw  m5, m6
-    punpcklbw  m2, m3
-    punpcklbw  m5, m2
-
-    pmaddubsw  m5, m0
-
-    phaddw     m4, m5
-
-    psubw      m4, m1
-
-    lea        r2, [r2 + 2 * r3]
-    movd       [r2], m4
-    pshufd     m4 , m4 ,2
-    movd       [r2 + r3], m4
-
-    lea        r2, [r2 + 2 * r3]
-
-    dec        r4d
-    jnz        .loop
-
-    RET
-%endmacro
-
-    FILTER_V_PS_W2 2, 8
-
-    FILTER_V_PS_W2 2, 16
-
-;-----------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ss_%1x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_SS 2
-INIT_XMM sse2
-cglobal interp_4tap_vert_ss_%1x%2, 5, 7, 6 ,0-gprsize
-
-    add       r1d, r1d
-    add       r3d, r3d
-    sub       r0, r1
-    shl       r4d, 5
-
-%ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
-    lea       r6, [r5 + r4]
-%else
-    lea       r6, [tab_ChromaCoeffV + r4]
-%endif
-
-    mov       dword [rsp], %2/4
-
-.loopH:
-    mov       r4d, (%1/4)
-.loopW:
-    PROCESS_CHROMA_SP_W4_4R
-
-    psrad     m0, 6
-    psrad     m1, 6
-    psrad     m2, 6
-    psrad     m3, 6
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-
-    movlps    [r2], m0
-    movhps    [r2 + r3], m0
-    lea       r5, [r2 + 2 * r3]
-    movlps    [r5], m2
-    movhps    [r5 + r3], m2
-
-    lea       r5, [4 * r1 - 2 * 4]
-    sub       r0, r5
-    add       r2, 2 * 4
-
-    dec       r4d
-    jnz       .loopW
-
-    lea       r0, [r0 + 4 * r1 - 2 * %1]
-    lea       r2, [r2 + 4 * r3 - 2 * %1]
-
-    dec       dword [rsp]
-    jnz       .loopH
-
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_SS 4, 4
-    FILTER_VER_CHROMA_SS 4, 8
-    FILTER_VER_CHROMA_SS 16, 16
-    FILTER_VER_CHROMA_SS 16, 8
-    FILTER_VER_CHROMA_SS 16, 12
-    FILTER_VER_CHROMA_SS 12, 16
-    FILTER_VER_CHROMA_SS 16, 4
-    FILTER_VER_CHROMA_SS 4, 16
-    FILTER_VER_CHROMA_SS 32, 32
-    FILTER_VER_CHROMA_SS 32, 16
-    FILTER_VER_CHROMA_SS 16, 32
-    FILTER_VER_CHROMA_SS 32, 24
-    FILTER_VER_CHROMA_SS 24, 32
-    FILTER_VER_CHROMA_SS 32, 8
-
-    FILTER_VER_CHROMA_SS 16, 24
-    FILTER_VER_CHROMA_SS 12, 32
-    FILTER_VER_CHROMA_SS 4, 32
-    FILTER_VER_CHROMA_SS 32, 64
-    FILTER_VER_CHROMA_SS 16, 64
-    FILTER_VER_CHROMA_SS 32, 48
-    FILTER_VER_CHROMA_SS 24, 64
-
-    FILTER_VER_CHROMA_SS 64, 64
-    FILTER_VER_CHROMA_SS 64, 32
-    FILTER_VER_CHROMA_SS 64, 48
-    FILTER_VER_CHROMA_SS 48, 64
-    FILTER_VER_CHROMA_SS 64, 16
-
-%macro FILTER_VER_CHROMA_S_AVX2_4x4 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_4x4, 4, 6, 7
-    mov             r4d, r4m
-    add             r1d, r1d
-    shl             r4d, 6
-    sub             r0, r1
-
-%ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
-%else
-    lea             r5, [pw_ChromaCoeffV + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-%ifidn %1,sp
-    mova            m6, [pd_526336]
-%else
-    add             r3d, r3d
-%endif
-
-    movq            xm0, [r0]
-    movq            xm1, [r0 + r1]
-    punpcklwd       xm0, xm1
-    movq            xm2, [r0 + r1 * 2]
-    punpcklwd       xm1, xm2
-    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
-    pmaddwd         m0, [r5]
-    movq            xm3, [r0 + r4]
-    punpcklwd       xm2, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm4, [r0]
-    punpcklwd       xm3, xm4
-    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
-    pmaddwd         m5, m2, [r5 + 1 * mmsize]
-    pmaddwd         m2, [r5]
-    paddd           m0, m5
-    movq            xm3, [r0 + r1]
-    punpcklwd       xm4, xm3
-    movq            xm1, [r0 + r1 * 2]
-    punpcklwd       xm3, xm1
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
-    pmaddwd         m4, [r5 + 1 * mmsize]
-    paddd           m2, m4
-
-%ifidn %1,sp
-    paddd           m0, m6
-    paddd           m2, m6
-    psrad           m0, 12
-    psrad           m2, 12
-%else
-    psrad           m0, 6
-    psrad           m2, 6
-%endif
-    packssdw        m0, m2
-    vextracti128    xm2, m0, 1
-    lea             r4, [r3 * 3]
-
-%ifidn %1,sp
-    packuswb        xm0, xm2
-    movd            [r2], xm0
-    pextrd          [r2 + r3], xm0, 2
-    pextrd          [r2 + r3 * 2], xm0, 1
-    pextrd          [r2 + r4], xm0, 3
-%else
-    movq            [r2], xm0
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r4], xm2
-%endif
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_S_AVX2_4x4 sp
-    FILTER_VER_CHROMA_S_AVX2_4x4 ss
-
-%macro FILTER_VER_CHROMA_S_AVX2_4x8 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_4x8, 4, 6, 8
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-    sub             r0, r1
-
-%ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
-%else
-    lea             r5, [pw_ChromaCoeffV + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-%ifidn %1,sp
-    mova            m7, [pd_526336]
-%else
-    add             r3d, r3d
-%endif
-
-    movq            xm0, [r0]
-    movq            xm1, [r0 + r1]
-    punpcklwd       xm0, xm1
-    movq            xm2, [r0 + r1 * 2]
-    punpcklwd       xm1, xm2
-    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
-    pmaddwd         m0, [r5]
-    movq            xm3, [r0 + r4]
-    punpcklwd       xm2, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm4, [r0]
-    punpcklwd       xm3, xm4
-    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
-    pmaddwd         m5, m2, [r5 + 1 * mmsize]
-    pmaddwd         m2, [r5]
-    paddd           m0, m5
-    movq            xm3, [r0 + r1]
-    punpcklwd       xm4, xm3
-    movq            xm1, [r0 + r1 * 2]
-    punpcklwd       xm3, xm1
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
-    pmaddwd         m5, m4, [r5 + 1 * mmsize]
-    paddd           m2, m5
-    pmaddwd         m4, [r5]
-    movq            xm3, [r0 + r4]
-    punpcklwd       xm1, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm6, [r0]
-    punpcklwd       xm3, xm6
-    vinserti128     m1, m1, xm3, 1                  ; m1 = [8 7 7 6]
-    pmaddwd         m5, m1, [r5 + 1 * mmsize]
-    paddd           m4, m5
-    pmaddwd         m1, [r5]
-    movq            xm3, [r0 + r1]
-    punpcklwd       xm6, xm3
-    movq            xm5, [r0 + 2 * r1]
-    punpcklwd       xm3, xm5
-    vinserti128     m6, m6, xm3, 1                  ; m6 = [A 9 9 8]
-    pmaddwd         m6, [r5 + 1 * mmsize]
-    paddd           m1, m6
-    lea             r4, [r3 * 3]
-
-%ifidn %1,sp
-    paddd           m0, m7
-    paddd           m2, m7
-    paddd           m4, m7
-    paddd           m1, m7
-    psrad           m0, 12
-    psrad           m2, 12
-    psrad           m4, 12
-    psrad           m1, 12
-%else
-    psrad           m0, 6
-    psrad           m2, 6
-    psrad           m4, 6
-    psrad           m1, 6
-%endif
-    packssdw        m0, m2
-    packssdw        m4, m1
-%ifidn %1,sp
-    packuswb        m0, m4
-    vextracti128    xm2, m0, 1
-    movd            [r2], xm0
-    movd            [r2 + r3], xm2
-    pextrd          [r2 + r3 * 2], xm0, 1
-    pextrd          [r2 + r4], xm2, 1
-    lea             r2, [r2 + r3 * 4]
-    pextrd          [r2], xm0, 2
-    pextrd          [r2 + r3], xm2, 2
-    pextrd          [r2 + r3 * 2], xm0, 3
-    pextrd          [r2 + r4], xm2, 3
-%else
-    vextracti128    xm2, m0, 1
-    vextracti128    xm1, m4, 1
-    movq            [r2], xm0
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r4], xm2
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm4
-    movq            [r2 + r3], xm1
-    movhps          [r2 + r3 * 2], xm4
-    movhps          [r2 + r4], xm1
-%endif
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_S_AVX2_4x8 sp
-    FILTER_VER_CHROMA_S_AVX2_4x8 ss
-
-%macro PROCESS_CHROMA_AVX2_W4_16R 1
-    movq            xm0, [r0]
-    movq            xm1, [r0 + r1]
-    punpcklwd       xm0, xm1
-    movq            xm2, [r0 + r1 * 2]
-    punpcklwd       xm1, xm2
-    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
-    pmaddwd         m0, [r5]
-    movq            xm3, [r0 + r4]
-    punpcklwd       xm2, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm4, [r0]
-    punpcklwd       xm3, xm4
-    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
-    pmaddwd         m5, m2, [r5 + 1 * mmsize]
-    pmaddwd         m2, [r5]
-    paddd           m0, m5
-    movq            xm3, [r0 + r1]
-    punpcklwd       xm4, xm3
-    movq            xm1, [r0 + r1 * 2]
-    punpcklwd       xm3, xm1
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
-    pmaddwd         m5, m4, [r5 + 1 * mmsize]
-    paddd           m2, m5
-    pmaddwd         m4, [r5]
-    movq            xm3, [r0 + r4]
-    punpcklwd       xm1, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm6, [r0]
-    punpcklwd       xm3, xm6
-    vinserti128     m1, m1, xm3, 1                  ; m1 = [8 7 7 6]
-    pmaddwd         m5, m1, [r5 + 1 * mmsize]
-    paddd           m4, m5
-    pmaddwd         m1, [r5]
-    movq            xm3, [r0 + r1]
-    punpcklwd       xm6, xm3
-    movq            xm5, [r0 + 2 * r1]
-    punpcklwd       xm3, xm5
-    vinserti128     m6, m6, xm3, 1                  ; m6 = [10 9 9 8]
-    pmaddwd         m3, m6, [r5 + 1 * mmsize]
-    paddd           m1, m3
-    pmaddwd         m6, [r5]
-
-%ifidn %1,sp
-    paddd           m0, m7
-    paddd           m2, m7
-    paddd           m4, m7
-    paddd           m1, m7
-    psrad           m4, 12
-    psrad           m1, 12
-    psrad           m0, 12
-    psrad           m2, 12
-%else
-    psrad           m0, 6
-    psrad           m2, 6
-    psrad           m4, 6
-    psrad           m1, 6
-%endif
-    packssdw        m0, m2
-    packssdw        m4, m1
-%ifidn %1,sp
-    packuswb        m0, m4
-    vextracti128    xm4, m0, 1
-    movd            [r2], xm0
-    movd            [r2 + r3], xm4
-    pextrd          [r2 + r3 * 2], xm0, 1
-    pextrd          [r2 + r6], xm4, 1
-    lea             r2, [r2 + r3 * 4]
-    pextrd          [r2], xm0, 2
-    pextrd          [r2 + r3], xm4, 2
-    pextrd          [r2 + r3 * 2], xm0, 3
-    pextrd          [r2 + r6], xm4, 3
-%else
-    vextracti128    xm2, m0, 1
-    vextracti128    xm1, m4, 1
-    movq            [r2], xm0
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r6], xm2
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm4
-    movq            [r2 + r3], xm1
-    movhps          [r2 + r3 * 2], xm4
-    movhps          [r2 + r6], xm1
-%endif
-
-    movq            xm2, [r0 + r4]
-    punpcklwd       xm5, xm2
-    lea             r0, [r0 + 4 * r1]
-    movq            xm0, [r0]
-    punpcklwd       xm2, xm0
-    vinserti128     m5, m5, xm2, 1                  ; m5 = [12 11 11 10]
-    pmaddwd         m2, m5, [r5 + 1 * mmsize]
-    paddd           m6, m2
-    pmaddwd         m5, [r5]
-    movq            xm2, [r0 + r1]
-    punpcklwd       xm0, xm2
-    movq            xm3, [r0 + 2 * r1]
-    punpcklwd       xm2, xm3
-    vinserti128     m0, m0, xm2, 1                  ; m0 = [14 13 13 12]
-    pmaddwd         m2, m0, [r5 + 1 * mmsize]
-    paddd           m5, m2
-    pmaddwd         m0, [r5]
-    movq            xm4, [r0 + r4]
-    punpcklwd       xm3, xm4
-    lea             r0, [r0 + 4 * r1]
-    movq            xm1, [r0]
-    punpcklwd       xm4, xm1
-    vinserti128     m3, m3, xm4, 1                  ; m3 = [16 15 15 14]
-    pmaddwd         m4, m3, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m3, [r5]
-    movq            xm4, [r0 + r1]
-    punpcklwd       xm1, xm4
-    movq            xm2, [r0 + 2 * r1]
-    punpcklwd       xm4, xm2
-    vinserti128     m1, m1, xm4, 1                  ; m1 = [18 17 17 16]
-    pmaddwd         m1, [r5 + 1 * mmsize]
-    paddd           m3, m1
-
-%ifidn %1,sp
-    paddd           m6, m7
-    paddd           m5, m7
-    paddd           m0, m7
-    paddd           m3, m7
-    psrad           m6, 12
-    psrad           m5, 12
-    psrad           m0, 12
-    psrad           m3, 12
-%else
-    psrad           m6, 6
-    psrad           m5, 6
-    psrad           m0, 6
-    psrad           m3, 6
-%endif
-    packssdw        m6, m5
-    packssdw        m0, m3
-    lea             r2, [r2 + r3 * 4]
-
-%ifidn %1,sp
-    packuswb        m6, m0
-    vextracti128    xm0, m6, 1
-    movd            [r2], xm6
-    movd            [r2 + r3], xm0
-    pextrd          [r2 + r3 * 2], xm6, 1
-    pextrd          [r2 + r6], xm0, 1
-    lea             r2, [r2 + r3 * 4]
-    pextrd          [r2], xm6, 2
-    pextrd          [r2 + r3], xm0, 2
-    pextrd          [r2 + r3 * 2], xm6, 3
-    pextrd          [r2 + r6], xm0, 3
-%else
-    vextracti128    xm5, m6, 1
-    vextracti128    xm3, m0, 1
-    movq            [r2], xm6
-    movq            [r2 + r3], xm5
-    movhps          [r2 + r3 * 2], xm6
-    movhps          [r2 + r6], xm5
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm0
-    movq            [r2 + r3], xm3
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r6], xm3
-%endif
-%endmacro
-
-%macro FILTER_VER_CHROMA_S_AVX2_4x16 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_4x16, 4, 7, 8
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-    sub             r0, r1
-
-%ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
-%else
-    lea             r5, [pw_ChromaCoeffV + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-%ifidn %1,sp
-    mova            m7, [pd_526336]
-%else
-    add             r3d, r3d
-%endif
-    lea             r6, [r3 * 3]
-    PROCESS_CHROMA_AVX2_W4_16R %1
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_S_AVX2_4x16 sp
-    FILTER_VER_CHROMA_S_AVX2_4x16 ss
-
-%macro FILTER_VER_CHROMA_S_AVX2_4x32 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_4x32, 4, 7, 8
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-    sub             r0, r1
-
-%ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
-%else
-    lea             r5, [pw_ChromaCoeffV + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-%ifidn %1,sp
-    mova            m7, [pd_526336]
-%else
-    add             r3d, r3d
-%endif
-    lea             r6, [r3 * 3]
-%rep 2
-    PROCESS_CHROMA_AVX2_W4_16R %1
-    lea             r2, [r2 + r3 * 4]
-%endrep
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_S_AVX2_4x32 sp
-    FILTER_VER_CHROMA_S_AVX2_4x32 ss
-
-%macro FILTER_VER_CHROMA_S_AVX2_4x2 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_4x2, 4, 6, 6
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-    sub             r0, r1
-
-%ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
-%else
-    lea             r5, [pw_ChromaCoeffV + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-%ifidn %1,sp
-    mova            m5, [pd_526336]
-%else
-    add             r3d, r3d
-%endif
-    movq            xm0, [r0]
-    movq            xm1, [r0 + r1]
-    punpcklwd       xm0, xm1
-    movq            xm2, [r0 + r1 * 2]
-    punpcklwd       xm1, xm2
-    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
-    pmaddwd         m0, [r5]
-    movq            xm3, [r0 + r4]
-    punpcklwd       xm2, xm3
-    movq            xm4, [r0 + 4 * r1]
-    punpcklwd       xm3, xm4
-    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
-    pmaddwd         m2, [r5 + 1 * mmsize]
-    paddd           m0, m2
-%ifidn %1,sp
-    paddd           m0, m5
-    psrad           m0, 12
-%else
-    psrad           m0, 6
-%endif
-    vextracti128    xm1, m0, 1
-    packssdw        xm0, xm1
-%ifidn %1,sp
-    packuswb        xm0, xm0
-    movd            [r2], xm0
-    pextrd          [r2 + r3], xm0, 1
-%else
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-%endif
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_S_AVX2_4x2 sp
-    FILTER_VER_CHROMA_S_AVX2_4x2 ss
-
-%macro FILTER_VER_CHROMA_S_AVX2_2x4 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_2x4, 4, 6, 6
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-    sub             r0, r1
-
-%ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
-%else
-    lea             r5, [pw_ChromaCoeffV + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-%ifidn %1,sp
-    mova            m5, [pd_526336]
-%else
-    add             r3d, r3d
-%endif
-    movd            xm0, [r0]
-    movd            xm1, [r0 + r1]
-    punpcklwd       xm0, xm1
-    movd            xm2, [r0 + r1 * 2]
-    punpcklwd       xm1, xm2
-    punpcklqdq      xm0, xm1                        ; m0 = [2 1 1 0]
-    movd            xm3, [r0 + r4]
-    punpcklwd       xm2, xm3
-    lea             r0, [r0 + 4 * r1]
-    movd            xm4, [r0]
-    punpcklwd       xm3, xm4
-    punpcklqdq      xm2, xm3                        ; m2 = [4 3 3 2]
-    vinserti128     m0, m0, xm2, 1                  ; m0 = [4 3 3 2 2 1 1 0]
-    movd            xm1, [r0 + r1]
-    punpcklwd       xm4, xm1
-    movd            xm3, [r0 + r1 * 2]
-    punpcklwd       xm1, xm3
-    punpcklqdq      xm4, xm1                        ; m4 = [6 5 5 4]
-    vinserti128     m2, m2, xm4, 1                  ; m2 = [6 5 5 4 4 3 3 2]
-    pmaddwd         m0, [r5]
-    pmaddwd         m2, [r5 + 1 * mmsize]
-    paddd           m0, m2
-%ifidn %1,sp
-    paddd           m0, m5
-    psrad           m0, 12
-%else
-    psrad           m0, 6
-%endif
-    vextracti128    xm1, m0, 1
-    packssdw        xm0, xm1
-    lea             r4, [r3 * 3]
-%ifidn %1,sp
-    packuswb        xm0, xm0
-    pextrw          [r2], xm0, 0
-    pextrw          [r2 + r3], xm0, 1
-    pextrw          [r2 + 2 * r3], xm0, 2
-    pextrw          [r2 + r4], xm0, 3
-%else
-    movd            [r2], xm0
-    pextrd          [r2 + r3], xm0, 1
-    pextrd          [r2 + 2 * r3], xm0, 2
-    pextrd          [r2 + r4], xm0, 3
-%endif
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_S_AVX2_2x4 sp
-    FILTER_VER_CHROMA_S_AVX2_2x4 ss
-
-%macro FILTER_VER_CHROMA_S_AVX2_8x8 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x8, 4, 6, 8
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-
-%ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
-%else
-    lea             r5, [pw_ChromaCoeffV + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,sp
-    mova            m7, [pd_526336]
-%else
-    add             r3d, r3d
-%endif
-
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    pmaddwd         m2, [r5]
-    paddd           m0, m4
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    pmaddwd         m3, [r5]
-    paddd           m1, m5
-%ifidn %1,sp
-    paddd           m0, m7
-    paddd           m1, m7
-    psrad           m0, 12
-    psrad           m1, 12
-%else
-    psrad           m0, 6
-    psrad           m1, 6
-%endif
-    packssdw        m0, m1
-
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m2, m6
-    pmaddwd         m4, [r5]
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm1, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm1, 1
-    pmaddwd         m1, m5, [r5 + 1 * mmsize]
-    pmaddwd         m5, [r5]
-    paddd           m3, m1
-%ifidn %1,sp
-    paddd           m2, m7
-    paddd           m3, m7
-    psrad           m2, 12
-    psrad           m3, 12
-%else
-    psrad           m2, 6
-    psrad           m3, 6
-%endif
-    packssdw        m2, m3
-
-    movu            xm1, [r0 + r4]                  ; m1 = row 7
-    punpckhwd       xm3, xm6, xm1
-    punpcklwd       xm6, xm1
-    vinserti128     m6, m6, xm3, 1
-    pmaddwd         m3, m6, [r5 + 1 * mmsize]
-    pmaddwd         m6, [r5]
-    paddd           m4, m3
-
-    lea             r4, [r3 * 3]
-%ifidn %1,sp
-    packuswb        m0, m2
-    mova            m3, [interp8_hps_shuf]
-    vpermd          m0, m3, m0
-    vextracti128    xm2, m0, 1
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r4], xm2
-%else
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    movu            [r2], xm0
-    vextracti128    xm0, m0, 1
-    vextracti128    xm3, m2, 1
-    movu            [r2 + r3], xm0
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r4], xm3
-%endif
-    lea             r2, [r2 + r3 * 4]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm0, [r0]                       ; m0 = row 8
-    punpckhwd       xm2, xm1, xm0
-    punpcklwd       xm1, xm0
-    vinserti128     m1, m1, xm2, 1
-    pmaddwd         m2, m1, [r5 + 1 * mmsize]
-    pmaddwd         m1, [r5]
-    paddd           m5, m2
-%ifidn %1,sp
-    paddd           m4, m7
-    paddd           m5, m7
-    psrad           m4, 12
-    psrad           m5, 12
-%else
-    psrad           m4, 6
-    psrad           m5, 6
-%endif
-    packssdw        m4, m5
-
-    movu            xm2, [r0 + r1]                  ; m2 = row 9
-    punpckhwd       xm5, xm0, xm2
-    punpcklwd       xm0, xm2
-    vinserti128     m0, m0, xm5, 1
-    pmaddwd         m0, [r5 + 1 * mmsize]
-    paddd           m6, m0
-    movu            xm5, [r0 + r1 * 2]              ; m5 = row 10
-    punpckhwd       xm0, xm2, xm5
-    punpcklwd       xm2, xm5
-    vinserti128     m2, m2, xm0, 1
-    pmaddwd         m2, [r5 + 1 * mmsize]
-    paddd           m1, m2
-
-%ifidn %1,sp
-    paddd           m6, m7
-    paddd           m1, m7
-    psrad           m6, 12
-    psrad           m1, 12
-%else
-    psrad           m6, 6
-    psrad           m1, 6
-%endif
-    packssdw        m6, m1
-%ifidn %1,sp
-    packuswb        m4, m6
-    vpermd          m4, m3, m4
-    vextracti128    xm6, m4, 1
-    movq            [r2], xm4
-    movhps          [r2 + r3], xm4
-    movq            [r2 + r3 * 2], xm6
-    movhps          [r2 + r4], xm6
-%else
-    vpermq          m4, m4, 11011000b
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm5, m4, 1
-    vextracti128    xm1, m6, 1
-    movu            [r2], xm4
-    movu            [r2 + r3], xm5
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r4], xm1
-%endif
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_S_AVX2_8x8 sp
-    FILTER_VER_CHROMA_S_AVX2_8x8 ss
-
-%macro PROCESS_CHROMA_S_AVX2_W8_16R 1
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m2, [r5]
-    lea             r7, [r0 + r1 * 4]
-    movu            xm4, [r7]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    paddd           m1, m5
-    pmaddwd         m3, [r5]
-    movu            xm5, [r7 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m2, m6
-    pmaddwd         m4, [r5]
-    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm7, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddwd         m7, m5, [r5 + 1 * mmsize]
-    paddd           m3, m7
-    pmaddwd         m5, [r5]
-%ifidn %1,sp
-    paddd           m0, m9
-    paddd           m1, m9
-    paddd           m2, m9
-    paddd           m3, m9
-    psrad           m0, 12
-    psrad           m1, 12
-    psrad           m2, 12
-    psrad           m3, 12
-%else
-    psrad           m0, 6
-    psrad           m1, 6
-    psrad           m2, 6
-    psrad           m3, 6
-%endif
-    packssdw        m0, m1
-    packssdw        m2, m3
-%ifidn %1,sp
-    packuswb        m0, m2
-    mova            m3, [interp8_hps_shuf]
-    vpermd          m0, m3, m0
-    vextracti128    xm2, m0, 1
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r6], xm2
-%else
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-%endif
-
-    movu            xm7, [r7 + r4]                  ; m7 = row 7
-    punpckhwd       xm8, xm6, xm7
-    punpcklwd       xm6, xm7
-    vinserti128     m6, m6, xm8, 1
-    pmaddwd         m8, m6, [r5 + 1 * mmsize]
-    paddd           m4, m8
-    pmaddwd         m6, [r5]
-    lea             r7, [r7 + r1 * 4]
-    movu            xm8, [r7]                       ; m8 = row 8
-    punpckhwd       xm0, xm7, xm8
-    punpcklwd       xm7, xm8
-    vinserti128     m7, m7, xm0, 1
-    pmaddwd         m0, m7, [r5 + 1 * mmsize]
-    paddd           m5, m0
-    pmaddwd         m7, [r5]
-    movu            xm0, [r7 + r1]                  ; m0 = row 9
-    punpckhwd       xm1, xm8, xm0
-    punpcklwd       xm8, xm0
-    vinserti128     m8, m8, xm1, 1
-    pmaddwd         m1, m8, [r5 + 1 * mmsize]
-    paddd           m6, m1
-    pmaddwd         m8, [r5]
-    movu            xm1, [r7 + r1 * 2]              ; m1 = row 10
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m2, m0, [r5 + 1 * mmsize]
-    paddd           m7, m2
-    pmaddwd         m0, [r5]
-%ifidn %1,sp
-    paddd           m4, m9
-    paddd           m5, m9
-    psrad           m4, 12
-    psrad           m5, 12
-    paddd           m6, m9
-    paddd           m7, m9
-    psrad           m6, 12
-    psrad           m7, 12
-%else
-    psrad           m4, 6
-    psrad           m5, 6
-    psrad           m6, 6
-    psrad           m7, 6
-%endif
-    packssdw        m4, m5
-    packssdw        m6, m7
-    lea             r8, [r2 + r3 * 4]
-%ifidn %1,sp
-    packuswb        m4, m6
-    vpermd          m4, m3, m4
-    vextracti128    xm6, m4, 1
-    movq            [r8], xm4
-    movhps          [r8 + r3], xm4
-    movq            [r8 + r3 * 2], xm6
-    movhps          [r8 + r6], xm6
-%else
-    vpermq          m4, m4, 11011000b
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm5, m4, 1
-    vextracti128    xm7, m6, 1
-    movu            [r8], xm4
-    movu            [r8 + r3], xm5
-    movu            [r8 + r3 * 2], xm6
-    movu            [r8 + r6], xm7
-%endif
-
-    movu            xm2, [r7 + r4]                  ; m2 = row 11
-    punpckhwd       xm4, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm4, 1
-    pmaddwd         m4, m1, [r5 + 1 * mmsize]
-    paddd           m8, m4
-    pmaddwd         m1, [r5]
-    lea             r7, [r7 + r1 * 4]
-    movu            xm4, [r7]                       ; m4 = row 12
-    punpckhwd       xm5, xm2, xm4
-    punpcklwd       xm2, xm4
-    vinserti128     m2, m2, xm5, 1
-    pmaddwd         m5, m2, [r5 + 1 * mmsize]
-    paddd           m0, m5
-    pmaddwd         m2, [r5]
-    movu            xm5, [r7 + r1]                  ; m5 = row 13
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m1, m6
-    pmaddwd         m4, [r5]
-    movu            xm6, [r7 + r1 * 2]              ; m6 = row 14
-    punpckhwd       xm7, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddwd         m7, m5, [r5 + 1 * mmsize]
-    paddd           m2, m7
-    pmaddwd         m5, [r5]
-%ifidn %1,sp
-    paddd           m8, m9
-    paddd           m0, m9
-    paddd           m1, m9
-    paddd           m2, m9
-    psrad           m8, 12
-    psrad           m0, 12
-    psrad           m1, 12
-    psrad           m2, 12
-%else
-    psrad           m8, 6
-    psrad           m0, 6
-    psrad           m1, 6
-    psrad           m2, 6
-%endif
-    packssdw        m8, m0
-    packssdw        m1, m2
-    lea             r8, [r8 + r3 * 4]
-%ifidn %1,sp
-    packuswb        m8, m1
-    vpermd          m8, m3, m8
-    vextracti128    xm1, m8, 1
-    movq            [r8], xm8
-    movhps          [r8 + r3], xm8
-    movq            [r8 + r3 * 2], xm1
-    movhps          [r8 + r6], xm1
-%else
-    vpermq          m8, m8, 11011000b
-    vpermq          m1, m1, 11011000b
-    vextracti128    xm0, m8, 1
-    vextracti128    xm2, m1, 1
-    movu            [r8], xm8
-    movu            [r8 + r3], xm0
-    movu            [r8 + r3 * 2], xm1
-    movu            [r8 + r6], xm2
-%endif
-    lea             r8, [r8 + r3 * 4]
-
-    movu            xm7, [r7 + r4]                  ; m7 = row 15
-    punpckhwd       xm2, xm6, xm7
-    punpcklwd       xm6, xm7
-    vinserti128     m6, m6, xm2, 1
-    pmaddwd         m2, m6, [r5 + 1 * mmsize]
-    paddd           m4, m2
-    pmaddwd         m6, [r5]
-    lea             r7, [r7 + r1 * 4]
-    movu            xm2, [r7]                       ; m2 = row 16
-    punpckhwd       xm1, xm7, xm2
-    punpcklwd       xm7, xm2
-    vinserti128     m7, m7, xm1, 1
-    pmaddwd         m1, m7, [r5 + 1 * mmsize]
-    paddd           m5, m1
-    pmaddwd         m7, [r5]
-    movu            xm1, [r7 + r1]                  ; m1 = row 17
-    punpckhwd       xm0, xm2, xm1
-    punpcklwd       xm2, xm1
-    vinserti128     m2, m2, xm0, 1
-    pmaddwd         m2, [r5 + 1 * mmsize]
-    paddd           m6, m2
-    movu            xm0, [r7 + r1 * 2]              ; m0 = row 18
-    punpckhwd       xm2, xm1, xm0
-    punpcklwd       xm1, xm0
-    vinserti128     m1, m1, xm2, 1
-    pmaddwd         m1, [r5 + 1 * mmsize]
-    paddd           m7, m1
-
-%ifidn %1,sp
-    paddd           m4, m9
-    paddd           m5, m9
-    paddd           m6, m9
-    paddd           m7, m9
-    psrad           m4, 12
-    psrad           m5, 12
-    psrad           m6, 12
-    psrad           m7, 12
-%else
-    psrad           m4, 6
-    psrad           m5, 6
-    psrad           m6, 6
-    psrad           m7, 6
-%endif
-    packssdw        m4, m5
-    packssdw        m6, m7
-%ifidn %1,sp
-    packuswb        m4, m6
-    vpermd          m4, m3, m4
-    vextracti128    xm6, m4, 1
-    movq            [r8], xm4
-    movhps          [r8 + r3], xm4
-    movq            [r8 + r3 * 2], xm6
-    movhps          [r8 + r6], xm6
-%else
-    vpermq          m4, m4, 11011000b
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm5, m4, 1
-    vextracti128    xm7, m6, 1
-    movu            [r8], xm4
-    movu            [r8 + r3], xm5
-    movu            [r8 + r3 * 2], xm6
-    movu            [r8 + r6], xm7
-%endif
-%endmacro
-
-%macro FILTER_VER_CHROMA_S_AVX2_Nx16 2
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_vert_%1_%2x16, 4, 10, 10
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-
-%ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
-%else
-    lea             r5, [pw_ChromaCoeffV + r4]
-%endif
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,sp
-    mova            m9, [pd_526336]
-%else
-    add             r3d, r3d
-%endif
-    lea             r6, [r3 * 3]
-    mov             r9d, %2 / 8
-.loopW:
-    PROCESS_CHROMA_S_AVX2_W8_16R %1
-%ifidn %1,sp
-    add             r2, 8
-%else
-    add             r2, 16
-%endif
-    add             r0, 16
-    dec             r9d
-    jnz             .loopW
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_CHROMA_S_AVX2_Nx16 sp, 16
-    FILTER_VER_CHROMA_S_AVX2_Nx16 sp, 32
-    FILTER_VER_CHROMA_S_AVX2_Nx16 sp, 64
-    FILTER_VER_CHROMA_S_AVX2_Nx16 ss, 16
-    FILTER_VER_CHROMA_S_AVX2_Nx16 ss, 32
-    FILTER_VER_CHROMA_S_AVX2_Nx16 ss, 64
-
-%macro FILTER_VER_CHROMA_S_AVX2_NxN 3
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_vert_%3_%1x%2, 4, 11, 10
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-
-%ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
-%else
-    lea             r5, [pw_ChromaCoeffV + r4]
-%endif
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %3,sp
-    mova            m9, [pd_526336]
-%else
-    add             r3d, r3d
-%endif
-    lea             r6, [r3 * 3]
-    mov             r9d, %2 / 16
-.loopH:
-    mov             r10d, %1 / 8
-.loopW:
-    PROCESS_CHROMA_S_AVX2_W8_16R %3
-%ifidn %3,sp
-    add             r2, 8
-%else
-    add             r2, 16
-%endif
-    add             r0, 16
-    dec             r10d
-    jnz             .loopW
-    lea             r0, [r7 - 2 * %1 + 16]
-%ifidn %3,sp
-    lea             r2, [r8 + r3 * 4 - %1 + 8]
-%else
-    lea             r2, [r8 + r3 * 4 - 2 * %1 + 16]
-%endif
-    dec             r9d
-    jnz             .loopH
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_CHROMA_S_AVX2_NxN 16, 32, sp
-    FILTER_VER_CHROMA_S_AVX2_NxN 24, 32, sp
-    FILTER_VER_CHROMA_S_AVX2_NxN 32, 32, sp
-    FILTER_VER_CHROMA_S_AVX2_NxN 16, 32, ss
-    FILTER_VER_CHROMA_S_AVX2_NxN 24, 32, ss
-    FILTER_VER_CHROMA_S_AVX2_NxN 32, 32, ss
-    FILTER_VER_CHROMA_S_AVX2_NxN 16, 64, sp
-    FILTER_VER_CHROMA_S_AVX2_NxN 24, 64, sp
-    FILTER_VER_CHROMA_S_AVX2_NxN 32, 64, sp
-    FILTER_VER_CHROMA_S_AVX2_NxN 32, 48, sp
-    FILTER_VER_CHROMA_S_AVX2_NxN 32, 48, ss
-    FILTER_VER_CHROMA_S_AVX2_NxN 16, 64, ss
-    FILTER_VER_CHROMA_S_AVX2_NxN 24, 64, ss
-    FILTER_VER_CHROMA_S_AVX2_NxN 32, 64, ss
-    FILTER_VER_CHROMA_S_AVX2_NxN 64, 64, sp
-    FILTER_VER_CHROMA_S_AVX2_NxN 64, 32, sp
-    FILTER_VER_CHROMA_S_AVX2_NxN 64, 48, sp
-    FILTER_VER_CHROMA_S_AVX2_NxN 48, 64, sp
-    FILTER_VER_CHROMA_S_AVX2_NxN 64, 64, ss
-    FILTER_VER_CHROMA_S_AVX2_NxN 64, 32, ss
-    FILTER_VER_CHROMA_S_AVX2_NxN 64, 48, ss
-    FILTER_VER_CHROMA_S_AVX2_NxN 48, 64, ss
-
-%macro PROCESS_CHROMA_S_AVX2_W8_4R 1
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    paddd           m1, m5
-    pmaddwd         m3, [r5]
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m4, [r5 + 1 * mmsize]
-    paddd           m2, m4
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm4, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm4, 1
-    pmaddwd         m5, [r5 + 1 * mmsize]
-    paddd           m3, m5
-%ifidn %1,sp
-    paddd           m0, m7
-    paddd           m1, m7
-    paddd           m2, m7
-    paddd           m3, m7
-    psrad           m0, 12
-    psrad           m1, 12
-    psrad           m2, 12
-    psrad           m3, 12
-%else
-    psrad           m0, 6
-    psrad           m1, 6
-    psrad           m2, 6
-    psrad           m3, 6
-%endif
-    packssdw        m0, m1
-    packssdw        m2, m3
-%ifidn %1,sp
-    packuswb        m0, m2
-    mova            m3, [interp8_hps_shuf]
-    vpermd          m0, m3, m0
-    vextracti128    xm2, m0, 1
-%else
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
-%endif
-%endmacro
-
-%macro FILTER_VER_CHROMA_S_AVX2_8x4 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x4, 4, 6, 8
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-
-%ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
-%else
-    lea             r5, [pw_ChromaCoeffV + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,sp
-    mova            m7, [pd_526336]
-%else
-    add             r3d, r3d
-%endif
-
-    PROCESS_CHROMA_S_AVX2_W8_4R %1
-    lea             r4, [r3 * 3]
-%ifidn %1,sp
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r4], xm2
-%else
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r4], xm3
-%endif
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_S_AVX2_8x4 sp
-    FILTER_VER_CHROMA_S_AVX2_8x4 ss
-
-%macro FILTER_VER_CHROMA_S_AVX2_12x16 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_vert_%1_12x16, 4, 9, 10
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-
-%ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
-%else
-    lea             r5, [pw_ChromaCoeffV + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,sp
-    mova            m9, [pd_526336]
-%else
-    add             r3d, r3d
-%endif
-    lea             r6, [r3 * 3]
-    PROCESS_CHROMA_S_AVX2_W8_16R %1
-%ifidn %1,sp
-    add             r2, 8
-%else
-    add             r2, 16
-%endif
-    add             r0, 16
-    mova            m7, m9
-    PROCESS_CHROMA_AVX2_W4_16R %1
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_CHROMA_S_AVX2_12x16 sp
-    FILTER_VER_CHROMA_S_AVX2_12x16 ss
-
-%macro FILTER_VER_CHROMA_S_AVX2_12x32 1
-%if ARCH_X86_64 == 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_12x32, 4, 9, 10
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-
-%ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
-%else
-    lea             r5, [pw_ChromaCoeffV + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1, sp
-    mova            m9, [pd_526336]
-%else
-    add             r3d, r3d
-%endif
-    lea             r6, [r3 * 3]
-%rep 2
-    PROCESS_CHROMA_S_AVX2_W8_16R %1
-%ifidn %1, sp
-    add             r2, 8
-%else
-    add             r2, 16
-%endif
-    add             r0, 16
-    mova            m7, m9
-    PROCESS_CHROMA_AVX2_W4_16R %1
-    sub             r0, 16
-%ifidn %1, sp
-    lea             r2, [r2 + r3 * 4 - 8]
-%else
-    lea             r2, [r2 + r3 * 4 - 16]
-%endif
-%endrep
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_CHROMA_S_AVX2_12x32 sp
-    FILTER_VER_CHROMA_S_AVX2_12x32 ss
-
-%macro FILTER_VER_CHROMA_S_AVX2_16x12 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_vert_%1_16x12, 4, 9, 9
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-
-%ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
-%else
-    lea             r5, [pw_ChromaCoeffV + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,sp
-    mova            m8, [pd_526336]
-%else
-    add             r3d, r3d
-%endif
-    lea             r6, [r3 * 3]
-%rep 2
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m2, [r5]
-    lea             r7, [r0 + r1 * 4]
-    movu            xm4, [r7]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    paddd           m1, m5
-    pmaddwd         m3, [r5]
-%ifidn %1,sp
-    paddd           m0, m8
-    paddd           m1, m8
-    psrad           m0, 12
-    psrad           m1, 12
-%else
-    psrad           m0, 6
-    psrad           m1, 6
-%endif
-    packssdw        m0, m1
-
-    movu            xm5, [r7 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m2, m6
-    pmaddwd         m4, [r5]
-    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm1, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm1, 1
-    pmaddwd         m1, m5, [r5 + 1 * mmsize]
-    pmaddwd         m5, [r5]
-    paddd           m3, m1
-%ifidn %1,sp
-    paddd           m2, m8
-    paddd           m3, m8
-    psrad           m2, 12
-    psrad           m3, 12
-%else
-    psrad           m2, 6
-    psrad           m3, 6
-%endif
-    packssdw        m2, m3
-%ifidn %1,sp
-    packuswb        m0, m2
-    mova            m3, [interp8_hps_shuf]
-    vpermd          m0, m3, m0
-    vextracti128    xm2, m0, 1
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r6], xm2
-%else
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    movu            [r2], xm0
-    vextracti128    xm0, m0, 1
-    vextracti128    xm3, m2, 1
-    movu            [r2 + r3], xm0
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-%endif
-    lea             r8, [r2 + r3 * 4]
-
-    movu            xm1, [r7 + r4]                  ; m1 = row 7
-    punpckhwd       xm0, xm6, xm1
-    punpcklwd       xm6, xm1
-    vinserti128     m6, m6, xm0, 1
-    pmaddwd         m0, m6, [r5 + 1 * mmsize]
-    pmaddwd         m6, [r5]
-    paddd           m4, m0
-    lea             r7, [r7 + r1 * 4]
-    movu            xm0, [r7]                       ; m0 = row 8
-    punpckhwd       xm2, xm1, xm0
-    punpcklwd       xm1, xm0
-    vinserti128     m1, m1, xm2, 1
-    pmaddwd         m2, m1, [r5 + 1 * mmsize]
-    pmaddwd         m1, [r5]
-    paddd           m5, m2
-%ifidn %1,sp
-    paddd           m4, m8
-    paddd           m5, m8
-    psrad           m4, 12
-    psrad           m5, 12
-%else
-    psrad           m4, 6
-    psrad           m5, 6
-%endif
-    packssdw        m4, m5
-
-    movu            xm2, [r7 + r1]                  ; m2 = row 9
-    punpckhwd       xm5, xm0, xm2
-    punpcklwd       xm0, xm2
-    vinserti128     m0, m0, xm5, 1
-    pmaddwd         m5, m0, [r5 + 1 * mmsize]
-    paddd           m6, m5
-    pmaddwd         m0, [r5]
-    movu            xm5, [r7 + r1 * 2]              ; m5 = row 10
-    punpckhwd       xm7, xm2, xm5
-    punpcklwd       xm2, xm5
-    vinserti128     m2, m2, xm7, 1
-    pmaddwd         m7, m2, [r5 + 1 * mmsize]
-    paddd           m1, m7
-    pmaddwd         m2, [r5]
-
-%ifidn %1,sp
-    paddd           m6, m8
-    paddd           m1, m8
-    psrad           m6, 12
-    psrad           m1, 12
-%else
-    psrad           m6, 6
-    psrad           m1, 6
-%endif
-    packssdw        m6, m1
-%ifidn %1,sp
-    packuswb        m4, m6
-    vpermd          m4, m3, m4
-    vextracti128    xm6, m4, 1
-    movq            [r8], xm4
-    movhps          [r8 + r3], xm4
-    movq            [r8 + r3 * 2], xm6
-    movhps          [r8 + r6], xm6
-%else
-    vpermq          m4, m4, 11011000b
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm7, m4, 1
-    vextracti128    xm1, m6, 1
-    movu            [r8], xm4
-    movu            [r8 + r3], xm7
-    movu            [r8 + r3 * 2], xm6
-    movu            [r8 + r6], xm1
-%endif
-    lea             r8, [r8 + r3 * 4]
-
-    movu            xm7, [r7 + r4]                  ; m7 = row 11
-    punpckhwd       xm1, xm5, xm7
-    punpcklwd       xm5, xm7
-    vinserti128     m5, m5, xm1, 1
-    pmaddwd         m1, m5, [r5 + 1 * mmsize]
-    paddd           m0, m1
-    pmaddwd         m5, [r5]
-    lea             r7, [r7 + r1 * 4]
-    movu            xm1, [r7]                       ; m1 = row 12
-    punpckhwd       xm4, xm7, xm1
-    punpcklwd       xm7, xm1
-    vinserti128     m7, m7, xm4, 1
-    pmaddwd         m4, m7, [r5 + 1 * mmsize]
-    paddd           m2, m4
-    pmaddwd         m7, [r5]
-%ifidn %1,sp
-    paddd           m0, m8
-    paddd           m2, m8
-    psrad           m0, 12
-    psrad           m2, 12
-%else
-    psrad           m0, 6
-    psrad           m2, 6
-%endif
-    packssdw        m0, m2
-
-    movu            xm4, [r7 + r1]                  ; m4 = row 13
-    punpckhwd       xm2, xm1, xm4
-    punpcklwd       xm1, xm4
-    vinserti128     m1, m1, xm2, 1
-    pmaddwd         m1, [r5 + 1 * mmsize]
-    paddd           m5, m1
-    movu            xm2, [r7 + r1 * 2]              ; m2 = row 14
-    punpckhwd       xm6, xm4, xm2
-    punpcklwd       xm4, xm2
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m4, [r5 + 1 * mmsize]
-    paddd           m7, m4
-%ifidn %1,sp
-    paddd           m5, m8
-    paddd           m7, m8
-    psrad           m5, 12
-    psrad           m7, 12
-%else
-    psrad           m5, 6
-    psrad           m7, 6
-%endif
-    packssdw        m5, m7
-%ifidn %1,sp
-    packuswb        m0, m5
-    vpermd          m0, m3, m0
-    vextracti128    xm5, m0, 1
-    movq            [r8], xm0
-    movhps          [r8 + r3], xm0
-    movq            [r8 + r3 * 2], xm5
-    movhps          [r8 + r6], xm5
-    add             r2, 8
-%else
-    vpermq          m0, m0, 11011000b
-    vpermq          m5, m5, 11011000b
-    vextracti128    xm7, m0, 1
-    vextracti128    xm6, m5, 1
-    movu            [r8], xm0
-    movu            [r8 + r3], xm7
-    movu            [r8 + r3 * 2], xm5
-    movu            [r8 + r6], xm6
-    add             r2, 16
-%endif
-    add             r0, 16
-%endrep
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_CHROMA_S_AVX2_16x12 sp
-    FILTER_VER_CHROMA_S_AVX2_16x12 ss
-
-%macro FILTER_VER_CHROMA_S_AVX2_8x12 1
-%if ARCH_X86_64 == 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x12, 4, 7, 9
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-
-%ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
-%else
-    lea             r5, [pw_ChromaCoeffV + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,sp
-    mova            m8, [pd_526336]
-%else
-    add             r3d, r3d
-%endif
-    lea             r6, [r3 * 3]
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    paddd           m1, m5
-    pmaddwd         m3, [r5]
-%ifidn %1,sp
-    paddd           m0, m8
-    paddd           m1, m8
-    psrad           m0, 12
-    psrad           m1, 12
-%else
-    psrad           m0, 6
-    psrad           m1, 6
-%endif
-    packssdw        m0, m1
-
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m2, m6
-    pmaddwd         m4, [r5]
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm1, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm1, 1
-    pmaddwd         m1, m5, [r5 + 1 * mmsize]
-    pmaddwd         m5, [r5]
-    paddd           m3, m1
-%ifidn %1,sp
-    paddd           m2, m8
-    paddd           m3, m8
-    psrad           m2, 12
-    psrad           m3, 12
-%else
-    psrad           m2, 6
-    psrad           m3, 6
-%endif
-    packssdw        m2, m3
-%ifidn %1,sp
-    packuswb        m0, m2
-    mova            m3, [interp8_hps_shuf]
-    vpermd          m0, m3, m0
-    vextracti128    xm2, m0, 1
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r6], xm2
-%else
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    movu            [r2], xm0
-    vextracti128    xm0, m0, 1
-    vextracti128    xm3, m2, 1
-    movu            [r2 + r3], xm0
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-%endif
-    lea             r2, [r2 + r3 * 4]
-
-    movu            xm1, [r0 + r4]                  ; m1 = row 7
-    punpckhwd       xm0, xm6, xm1
-    punpcklwd       xm6, xm1
-    vinserti128     m6, m6, xm0, 1
-    pmaddwd         m0, m6, [r5 + 1 * mmsize]
-    pmaddwd         m6, [r5]
-    paddd           m4, m0
-    lea             r0, [r0 + r1 * 4]
-    movu            xm0, [r0]                       ; m0 = row 8
-    punpckhwd       xm2, xm1, xm0
-    punpcklwd       xm1, xm0
-    vinserti128     m1, m1, xm2, 1
-    pmaddwd         m2, m1, [r5 + 1 * mmsize]
-    pmaddwd         m1, [r5]
-    paddd           m5, m2
-%ifidn %1,sp
-    paddd           m4, m8
-    paddd           m5, m8
-    psrad           m4, 12
-    psrad           m5, 12
-%else
-    psrad           m4, 6
-    psrad           m5, 6
-%endif
-    packssdw        m4, m5
-
-    movu            xm2, [r0 + r1]                  ; m2 = row 9
-    punpckhwd       xm5, xm0, xm2
-    punpcklwd       xm0, xm2
-    vinserti128     m0, m0, xm5, 1
-    pmaddwd         m5, m0, [r5 + 1 * mmsize]
-    paddd           m6, m5
-    pmaddwd         m0, [r5]
-    movu            xm5, [r0 + r1 * 2]              ; m5 = row 10
-    punpckhwd       xm7, xm2, xm5
-    punpcklwd       xm2, xm5
-    vinserti128     m2, m2, xm7, 1
-    pmaddwd         m7, m2, [r5 + 1 * mmsize]
-    paddd           m1, m7
-    pmaddwd         m2, [r5]
-
-%ifidn %1,sp
-    paddd           m6, m8
-    paddd           m1, m8
-    psrad           m6, 12
-    psrad           m1, 12
-%else
-    psrad           m6, 6
-    psrad           m1, 6
-%endif
-    packssdw        m6, m1
-%ifidn %1,sp
-    packuswb        m4, m6
-    vpermd          m4, m3, m4
-    vextracti128    xm6, m4, 1
-    movq            [r2], xm4
-    movhps          [r2 + r3], xm4
-    movq            [r2 + r3 * 2], xm6
-    movhps          [r2 + r6], xm6
-%else
-    vpermq          m4, m4, 11011000b
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm7, m4, 1
-    vextracti128    xm1, m6, 1
-    movu            [r2], xm4
-    movu            [r2 + r3], xm7
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r6], xm1
-%endif
-    lea             r2, [r2 + r3 * 4]
-
-    movu            xm7, [r0 + r4]                  ; m7 = row 11
-    punpckhwd       xm1, xm5, xm7
-    punpcklwd       xm5, xm7
-    vinserti128     m5, m5, xm1, 1
-    pmaddwd         m1, m5, [r5 + 1 * mmsize]
-    paddd           m0, m1
-    pmaddwd         m5, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm1, [r0]                       ; m1 = row 12
-    punpckhwd       xm4, xm7, xm1
-    punpcklwd       xm7, xm1
-    vinserti128     m7, m7, xm4, 1
-    pmaddwd         m4, m7, [r5 + 1 * mmsize]
-    paddd           m2, m4
-    pmaddwd         m7, [r5]
-%ifidn %1,sp
-    paddd           m0, m8
-    paddd           m2, m8
-    psrad           m0, 12
-    psrad           m2, 12
-%else
-    psrad           m0, 6
-    psrad           m2, 6
-%endif
-    packssdw        m0, m2
-
-    movu            xm4, [r0 + r1]                  ; m4 = row 13
-    punpckhwd       xm2, xm1, xm4
-    punpcklwd       xm1, xm4
-    vinserti128     m1, m1, xm2, 1
-    pmaddwd         m1, [r5 + 1 * mmsize]
-    paddd           m5, m1
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 14
-    punpckhwd       xm6, xm4, xm2
-    punpcklwd       xm4, xm2
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m4, [r5 + 1 * mmsize]
-    paddd           m7, m4
-%ifidn %1,sp
-    paddd           m5, m8
-    paddd           m7, m8
-    psrad           m5, 12
-    psrad           m7, 12
-%else
-    psrad           m5, 6
-    psrad           m7, 6
-%endif
-    packssdw        m5, m7
-%ifidn %1,sp
-    packuswb        m0, m5
-    vpermd          m0, m3, m0
-    vextracti128    xm5, m0, 1
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm5
-    movhps          [r2 + r6], xm5
-%else
-    vpermq          m0, m0, 11011000b
-    vpermq          m5, m5, 11011000b
-    vextracti128    xm7, m0, 1
-    vextracti128    xm6, m5, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm7
-    movu            [r2 + r3 * 2], xm5
-    movu            [r2 + r6], xm6
-%endif
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_CHROMA_S_AVX2_8x12 sp
-    FILTER_VER_CHROMA_S_AVX2_8x12 ss
-
-%macro FILTER_VER_CHROMA_S_AVX2_16x4 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_16x4, 4, 7, 8
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-
-%ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
-%else
-    lea             r5, [pw_ChromaCoeffV + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,sp
-    mova            m7, [pd_526336]
-%else
-    add             r3d, r3d
-%endif
-%rep 2
-    PROCESS_CHROMA_S_AVX2_W8_4R %1
-    lea             r6, [r3 * 3]
-%ifidn %1,sp
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r6], xm2
-    add             r2, 8
-%else
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-    add             r2, 16
-%endif
-    lea             r6, [4 * r1 - 16]
-    sub             r0, r6
-%endrep
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_S_AVX2_16x4 sp
-    FILTER_VER_CHROMA_S_AVX2_16x4 ss
-
-%macro PROCESS_CHROMA_S_AVX2_W8_8R 1
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m2, [r5]
-    lea             r7, [r0 + r1 * 4]
-    movu            xm4, [r7]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    paddd           m1, m5
-    pmaddwd         m3, [r5]
-%ifidn %1,sp
-    paddd           m0, m7
-    paddd           m1, m7
-    psrad           m0, 12
-    psrad           m1, 12
-%else
-    psrad           m0, 6
-    psrad           m1, 6
-%endif
-    packssdw        m0, m1
-
-    movu            xm5, [r7 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m2, m6
-    pmaddwd         m4, [r5]
-    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm1, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm1, 1
-    pmaddwd         m1, m5, [r5 + 1 * mmsize]
-    pmaddwd         m5, [r5]
-    paddd           m3, m1
-%ifidn %1,sp
-    paddd           m2, m7
-    paddd           m3, m7
-    psrad           m2, 12
-    psrad           m3, 12
-%else
-    psrad           m2, 6
-    psrad           m3, 6
-%endif
-    packssdw        m2, m3
-%ifidn %1,sp
-    packuswb        m0, m2
-    mova            m3, [interp8_hps_shuf]
-    vpermd          m0, m3, m0
-    vextracti128    xm2, m0, 1
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r6], xm2
-%else
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    movu            [r2], xm0
-    vextracti128    xm0, m0, 1
-    vextracti128    xm3, m2, 1
-    movu            [r2 + r3], xm0
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
-%endif
-    lea             r8, [r2 + r3 * 4]
-
-    movu            xm1, [r7 + r4]                  ; m1 = row 7
-    punpckhwd       xm0, xm6, xm1
-    punpcklwd       xm6, xm1
-    vinserti128     m6, m6, xm0, 1
-    pmaddwd         m0, m6, [r5 + 1 * mmsize]
-    pmaddwd         m6, [r5]
-    paddd           m4, m0
-    lea             r7, [r7 + r1 * 4]
-    movu            xm0, [r7]                       ; m0 = row 8
-    punpckhwd       xm2, xm1, xm0
-    punpcklwd       xm1, xm0
-    vinserti128     m1, m1, xm2, 1
-    pmaddwd         m2, m1, [r5 + 1 * mmsize]
-    pmaddwd         m1, [r5]
-    paddd           m5, m2
-%ifidn %1,sp
-    paddd           m4, m7
-    paddd           m5, m7
-    psrad           m4, 12
-    psrad           m5, 12
-%else
-    psrad           m4, 6
-    psrad           m5, 6
-%endif
-    packssdw        m4, m5
-
-    movu            xm2, [r7 + r1]                  ; m2 = row 9
-    punpckhwd       xm5, xm0, xm2
-    punpcklwd       xm0, xm2
-    vinserti128     m0, m0, xm5, 1
-    pmaddwd         m0, [r5 + 1 * mmsize]
-    paddd           m6, m0
-    movu            xm5, [r7 + r1 * 2]              ; m5 = row 10
-    punpckhwd       xm0, xm2, xm5
-    punpcklwd       xm2, xm5
-    vinserti128     m2, m2, xm0, 1
-    pmaddwd         m2, [r5 + 1 * mmsize]
-    paddd           m1, m2
-
-%ifidn %1,sp
-    paddd           m6, m7
-    paddd           m1, m7
-    psrad           m6, 12
-    psrad           m1, 12
-%else
-    psrad           m6, 6
-    psrad           m1, 6
-%endif
-    packssdw        m6, m1
-%ifidn %1,sp
-    packuswb        m4, m6
-    vpermd          m4, m3, m4
-    vextracti128    xm6, m4, 1
-    movq            [r8], xm4
-    movhps          [r8 + r3], xm4
-    movq            [r8 + r3 * 2], xm6
-    movhps          [r8 + r6], xm6
-%else
-    vpermq          m4, m4, 11011000b
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm7, m4, 1
-    vextracti128    xm1, m6, 1
-    movu            [r8], xm4
-    movu            [r8 + r3], xm7
-    movu            [r8 + r3 * 2], xm6
-    movu            [r8 + r6], xm1
-%endif
-%endmacro
-
-%macro FILTER_VER_CHROMA_S_AVX2_Nx8 2
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_vert_%1_%2x8, 4, 9, 8
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-
-%ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
-%else
-    lea             r5, [pw_ChromaCoeffV + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,sp
-    mova            m7, [pd_526336]
-%else
-    add             r3d, r3d
-%endif
-    lea             r6, [r3 * 3]
-%rep %2 / 8
-    PROCESS_CHROMA_S_AVX2_W8_8R %1
-%ifidn %1,sp
-    add             r2, 8
-%else
-    add             r2, 16
-%endif
-    add             r0, 16
-%endrep
-    RET
-%endif
-%endmacro
-
-    FILTER_VER_CHROMA_S_AVX2_Nx8 sp, 32
-    FILTER_VER_CHROMA_S_AVX2_Nx8 sp, 16
-    FILTER_VER_CHROMA_S_AVX2_Nx8 ss, 32
-    FILTER_VER_CHROMA_S_AVX2_Nx8 ss, 16
-
-%macro FILTER_VER_CHROMA_S_AVX2_8x2 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x2, 4, 6, 6
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-
-%ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
-%else
-    lea             r5, [pw_ChromaCoeffV + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,sp
-    mova            m5, [pd_526336]
-%else
-    add             r3d, r3d
-%endif
-
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m2, [r5 + 1 * mmsize]
-    paddd           m0, m2
-    movu            xm4, [r0 + r1 * 4]              ; m4 = row 4
-    punpckhwd       xm2, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm2, 1
-    pmaddwd         m3, [r5 + 1 * mmsize]
-    paddd           m1, m3
-%ifidn %1,sp
-    paddd           m0, m5
-    paddd           m1, m5
-    psrad           m0, 12
-    psrad           m1, 12
-%else
-    psrad           m0, 6
-    psrad           m1, 6
-%endif
-    packssdw        m0, m1
-%ifidn %1,sp
-    vextracti128    xm1, m0, 1
-    packuswb        xm0, xm1
-    pshufd          xm0, xm0, 11011000b
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-%else
-    vpermq          m0, m0, 11011000b
-    vextracti128    xm1, m0, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-%endif
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_S_AVX2_8x2 sp
-    FILTER_VER_CHROMA_S_AVX2_8x2 ss
-
-%macro FILTER_VER_CHROMA_S_AVX2_8x6 1
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x6, 4, 6, 8
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-
-%ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
-%else
-    lea             r5, [pw_ChromaCoeffV + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,sp
-    mova            m7, [pd_526336]
-%else
-    add             r3d, r3d
-%endif
-
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
+    movu            xm1, [r7 + r4]                  ; m1 = row 15
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
     vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
+    pmaddubsw       m2, m0, [r5 + 3 * mmsize]
+    paddw           m8, m2
+    pmaddubsw       m2, m0, [r5 + 2 * mmsize]
+    paddw           m10, m2
+    pmaddubsw       m2, m0, [r5 + 1 * mmsize]
+    paddw           m12, m2
+    pmaddubsw       m0, [r5]
+    lea             r7, [r7 + r1 * 4]
+    movu            xm2, [r7]                       ; m2 = row 16
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
     vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
+    pmaddubsw       m3, m1, [r5 + 3 * mmsize]
+    paddw           m9, m3
+    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
+    paddw           m11, m3
+    pmaddubsw       m3, m1, [r5 + 1 * mmsize]
+    paddw           m13, m3
+    pmaddubsw       m1, [r5]
+    movu            xm3, [r7 + r1]                  ; m3 = row 17
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
     vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    pmaddwd         m2, [r5]
-    paddd           m0, m4
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
-    punpcklwd       xm3, xm4
+    pmaddubsw       m4, m2, [r5 + 3 * mmsize]
+    paddw           m10, m4
+    pmaddubsw       m4, m2, [r5 + 2 * mmsize]
+    paddw           m12, m4
+    pmaddubsw       m2, [r5 + 1 * mmsize]
+    paddw           m0, m2
+    movu            xm4, [r7 + r1 * 2]              ; m4 = row 18
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
     vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    pmaddwd         m3, [r5]
-    paddd           m1, m5
-%ifidn %1,sp
-    paddd           m0, m7
-    paddd           m1, m7
-    psrad           m0, 12
-    psrad           m1, 12
-%else
-    psrad           m0, 6
-    psrad           m1, 6
-%endif
-    packssdw        m0, m1
-
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
+    pmaddubsw       m5, m3, [r5 + 3 * mmsize]
+    paddw           m11, m5
+    pmaddubsw       m5, m3, [r5 + 2 * mmsize]
+    paddw           m13, m5
+    pmaddubsw       m3, [r5 + 1 * mmsize]
+    paddw           m1, m3
+    movu            xm5, [r7 + r4]                  ; m5 = row 19
+    punpckhbw       xm6, xm4, xm5
+    punpcklbw       xm4, xm5
     vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m2, m6
-    pmaddwd         m4, [r5]
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm1, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm1, 1
-    pmaddwd         m1, m5, [r5 + 1 * mmsize]
-    pmaddwd         m5, [r5]
-    paddd           m3, m1
-%ifidn %1,sp
-    paddd           m2, m7
-    paddd           m3, m7
-    psrad           m2, 12
-    psrad           m3, 12
-%else
-    psrad           m2, 6
-    psrad           m3, 6
-%endif
-    packssdw        m2, m3
+    pmaddubsw       m6, m4, [r5 + 3 * mmsize]
+    paddw           m12, m6
+    pmaddubsw       m4, [r5 + 2 * mmsize]
+    paddw           m0, m4
+    lea             r7, [r7 + r1 * 4]
+    movu            xm6, [r7]                       ; m6 = row 20
+    punpckhbw       xm7, xm5, xm6
+    punpcklbw       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddubsw       m7, m5, [r5 + 3 * mmsize]
+    paddw           m13, m7
+    pmaddubsw       m5, [r5 + 2 * mmsize]
+    paddw           m1, m5
+    movu            xm7, [r7 + r1]                  ; m7 = row 21
+    punpckhbw       xm2, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm2, 1
+    pmaddubsw       m6, [r5 + 3 * mmsize]
+    paddw           m0, m6
+    movu            xm2, [r7 + r1 * 2]              ; m2 = row 22
+    punpckhbw       xm3, xm7, xm2
+    punpcklbw       xm7, xm2
+    vinserti128     m7, m7, xm3, 1
+    pmaddubsw       m7, [r5 + 3 * mmsize]
+    paddw           m1, m7
 
-    movu            xm1, [r0 + r4]                  ; m1 = row 7
-    punpckhwd       xm3, xm6, xm1
-    punpcklwd       xm6, xm1
-    vinserti128     m6, m6, xm3, 1
-    pmaddwd         m6, [r5 + 1 * mmsize]
-    paddd           m4, m6
-    movu            xm6, [r0 + r1 * 4]              ; m6 = row 8
-    punpckhwd       xm3, xm1, xm6
-    punpcklwd       xm1, xm6
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5 + 1 * mmsize]
-    paddd           m5, m1
-%ifidn %1,sp
-    paddd           m4, m7
-    paddd           m5, m7
-    psrad           m4, 12
-    psrad           m5, 12
-%else
-    psrad           m4, 6
-    psrad           m5, 6
-%endif
-    packssdw        m4, m5
-    lea             r4, [r3 * 3]
-%ifidn %1,sp
-    packuswb        m0, m2
-    mova            m3, [interp8_hps_shuf]
-    vpermd          m0, m3, m0
-    vextracti128    xm2, m0, 1
-    vextracti128    xm5, m4, 1
-    packuswb        xm4, xm5
-    pshufd          xm4, xm4, 11011000b
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r4], xm2
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm4
-    movhps          [r2 + r3], xm4
-%else
+%ifidn %1,pp
+    pmulhrsw        m8, m14                         ; m8 = word: row 8
+    pmulhrsw        m9, m14                         ; m9 = word: row 9
+    pmulhrsw        m10, m14                        ; m10 = word: row 10
+    pmulhrsw        m11, m14                        ; m11 = word: row 11
+    pmulhrsw        m12, m14                        ; m12 = word: row 12
+    pmulhrsw        m13, m14                        ; m13 = word: row 13
+    pmulhrsw        m0, m14                         ; m0 = word: row 14
+    pmulhrsw        m1, m14                         ; m1 = word: row 15
+    packuswb        m8, m9
+    packuswb        m10, m11
+    packuswb        m12, m13
+    packuswb        m0, m1
+    vpermq          m8, m8, 11011000b
+    vpermq          m10, m10, 11011000b
+    vpermq          m12, m12, 11011000b
     vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    vpermq          m4, m4, 11011000b
-    movu            [r2], xm0
-    vextracti128    xm0, m0, 1
-    vextracti128    xm3, m2, 1
-    vextracti128    xm5, m4, 1
-    movu            [r2 + r3], xm0
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r4], xm3
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm4
-    movu            [r2 + r3], xm5
-%endif
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_S_AVX2_8x6 sp
-    FILTER_VER_CHROMA_S_AVX2_8x6 ss
-
-%macro FILTER_VER_CHROMA_S_AVX2_8xN 2
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_vert_%1_8x%2, 4, 7, 9
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-
-%ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
-    add             r5, r4
+    vextracti128    xm9, m8, 1
+    vextracti128    xm11, m10, 1
+    vextracti128    xm13, m12, 1
+    vextracti128    xm1, m0, 1
+    movu            [r8], xm8
+    movu            [r8 + r3], xm9
+    movu            [r8 + r3 * 2], xm10
+    movu            [r8 + r6], xm11
+    lea             r8, [r8 + r3 * 4]
+    movu            [r8], xm12
+    movu            [r8 + r3], xm13
+    movu            [r8 + r3 * 2], xm0
+    movu            [r8 + r6], xm1
 %else
-    lea             r5, [pw_ChromaCoeffV + r4]
+    psubw           m8, m14                         ; m8 = word: row 8
+    psubw           m9, m14                         ; m9 = word: row 9
+    psubw           m10, m14                        ; m10 = word: row 10
+    psubw           m11, m14                        ; m11 = word: row 11
+    psubw           m12, m14                        ; m12 = word: row 12
+    psubw           m13, m14                        ; m13 = word: row 13
+    psubw           m0, m14                         ; m0 = word: row 14
+    psubw           m1, m14                         ; m1 = word: row 15
+    movu            [r8], m8
+    movu            [r8 + r3], m9
+    movu            [r8 + r3 * 2], m10
+    movu            [r8 + r6], m11
+    lea             r8, [r8 + r3 * 4]
+    movu            [r8], m12
+    movu            [r8 + r3], m13
+    movu            [r8 + r3 * 2], m0
+    movu            [r8 + r6], m1
 %endif
+%endmacro
 
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,sp
-    mova            m8, [pd_526336]
-%else
-    add             r3d, r3d
-%endif
-    lea             r6, [r3 * 3]
-%rep %2 / 16
+%macro PROCESS_LUMA_AVX2_W16_8R 1
     movu            xm0, [r0]                       ; m0 = row 0
     movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
     vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
+    pmaddubsw       m0, [r5]
     movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
     vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
+    pmaddubsw       m1, [r5]
     movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
     vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
-    punpcklwd       xm3, xm4
+    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
+    paddw           m0, m4
+    pmaddubsw       m2, [r5]
+    lea             r7, [r0 + r1 * 4]
+    movu            xm4, [r7]                       ; m4 = row 4
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
     vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    paddd           m1, m5
-    pmaddwd         m3, [r5]
-%ifidn %1,sp
-    paddd           m0, m8
-    paddd           m1, m8
-    psrad           m0, 12
-    psrad           m1, 12
-%else
-    psrad           m0, 6
-    psrad           m1, 6
-%endif
-    packssdw        m0, m1
-
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
+    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
+    paddw           m1, m5
+    pmaddubsw       m3, [r5]
+    movu            xm5, [r7 + r1]                  ; m5 = row 5
+    punpckhbw       xm6, xm4, xm5
+    punpcklbw       xm4, xm5
     vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m2, m6
-    pmaddwd         m4, [r5]
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm1, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm1, 1
-    pmaddwd         m1, m5, [r5 + 1 * mmsize]
-    pmaddwd         m5, [r5]
-    paddd           m3, m1
-%ifidn %1,sp
-    paddd           m2, m8
-    paddd           m3, m8
-    psrad           m2, 12
-    psrad           m3, 12
-%else
-    psrad           m2, 6
-    psrad           m3, 6
-%endif
-    packssdw        m2, m3
-%ifidn %1,sp
-    packuswb        m0, m2
-    mova            m3, [interp8_hps_shuf]
-    vpermd          m0, m3, m0
-    vextracti128    xm2, m0, 1
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r6], xm2
-%else
+    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
+    paddw           m0, m6
+    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
+    paddw           m2, m6
+    pmaddubsw       m4, [r5]
+    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
+    punpckhbw       xm7, xm5, xm6
+    punpcklbw       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
+    paddw           m1, m7
+    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
+    paddw           m3, m7
+    pmaddubsw       m5, [r5]
+    movu            xm7, [r7 + r4]                  ; m7 = row 7
+    punpckhbw       xm8, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
+    paddw           m0, m8
+    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
+    paddw           m2, m8
+    pmaddubsw       m8, m6, [r5 + 1 * mmsize]
+    paddw           m4, m8
+    pmaddubsw       m6, [r5]
+    lea             r7, [r7 + r1 * 4]
+    movu            xm8, [r7]                       ; m8 = row 8
+    punpckhbw       xm9, xm7, xm8
+    punpcklbw       xm7, xm8
+    vinserti128     m7, m7, xm9, 1
+    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
+    paddw           m1, m9
+    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
+    paddw           m3, m9
+    pmaddubsw       m9, m7, [r5 + 1 * mmsize]
+    paddw           m5, m9
+    pmaddubsw       m7, [r5]
+    movu            xm9, [r7 + r1]                  ; m9 = row 9
+    punpckhbw       xm10, xm8, xm9
+    punpcklbw       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
+    paddw           m2, m10
+    pmaddubsw       m10, m8, [r5 + 2 * mmsize]
+    paddw           m4, m10
+    pmaddubsw       m10, m8, [r5 + 1 * mmsize]
+    paddw           m6, m10
+    movu            xm10, [r7 + r1 * 2]             ; m10 = row 10
+    punpckhbw       xm11, xm9, xm10
+    punpcklbw       xm9, xm10
+    vinserti128     m9, m9, xm11, 1
+    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
+    paddw           m3, m11
+    pmaddubsw       m11, m9, [r5 + 2 * mmsize]
+    paddw           m5, m11
+    pmaddubsw       m11, m9, [r5 + 1 * mmsize]
+    paddw           m7, m11
+    movu            xm11, [r7 + r4]                 ; m11 = row 11
+    punpckhbw       xm12, xm10, xm11
+    punpcklbw       xm10, xm11
+    vinserti128     m10, m10, xm12, 1
+    pmaddubsw       m12, m10, [r5 + 3 * mmsize]
+    paddw           m4, m12
+    pmaddubsw       m12, m10, [r5 + 2 * mmsize]
+    paddw           m6, m12
+    lea             r7, [r7 + r1 * 4]
+    movu            xm12, [r7]                      ; m12 = row 12
+    punpckhbw       xm13, xm11, xm12
+    punpcklbw       xm11, xm12
+    vinserti128     m11, m11, xm13, 1
+    pmaddubsw       m13, m11, [r5 + 3 * mmsize]
+    paddw           m5, m13
+    pmaddubsw       m13, m11, [r5 + 2 * mmsize]
+    paddw           m7, m13
+
+%ifidn %1,pp
+    pmulhrsw        m0, m14                         ; m0 = word: row 0
+    pmulhrsw        m1, m14                         ; m1 = word: row 1
+    pmulhrsw        m2, m14                         ; m2 = word: row 2
+    pmulhrsw        m3, m14                         ; m3 = word: row 3
+    pmulhrsw        m4, m14                         ; m4 = word: row 4
+    pmulhrsw        m5, m14                         ; m5 = word: row 5
+    packuswb        m0, m1
+    packuswb        m2, m3
+    packuswb        m4, m5
     vpermq          m0, m0, 11011000b
     vpermq          m2, m2, 11011000b
-    movu            [r2], xm0
-    vextracti128    xm0, m0, 1
+    vpermq          m4, m4, 11011000b
+    vextracti128    xm1, m0, 1
     vextracti128    xm3, m2, 1
-    movu            [r2 + r3], xm0
+    vextracti128    xm5, m4, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
     movu            [r2 + r3 * 2], xm2
     movu            [r2 + r6], xm3
-%endif
-    lea             r2, [r2 + r3 * 4]
-
-    movu            xm1, [r0 + r4]                  ; m1 = row 7
-    punpckhwd       xm0, xm6, xm1
-    punpcklwd       xm6, xm1
-    vinserti128     m6, m6, xm0, 1
-    pmaddwd         m0, m6, [r5 + 1 * mmsize]
-    pmaddwd         m6, [r5]
-    paddd           m4, m0
-    lea             r0, [r0 + r1 * 4]
-    movu            xm0, [r0]                       ; m0 = row 8
-    punpckhwd       xm2, xm1, xm0
-    punpcklwd       xm1, xm0
-    vinserti128     m1, m1, xm2, 1
-    pmaddwd         m2, m1, [r5 + 1 * mmsize]
-    pmaddwd         m1, [r5]
-    paddd           m5, m2
-%ifidn %1,sp
-    paddd           m4, m8
-    paddd           m5, m8
-    psrad           m4, 12
-    psrad           m5, 12
-%else
-    psrad           m4, 6
-    psrad           m5, 6
-%endif
-    packssdw        m4, m5
-
-    movu            xm2, [r0 + r1]                  ; m2 = row 9
-    punpckhwd       xm5, xm0, xm2
-    punpcklwd       xm0, xm2
-    vinserti128     m0, m0, xm5, 1
-    pmaddwd         m5, m0, [r5 + 1 * mmsize]
-    paddd           m6, m5
-    pmaddwd         m0, [r5]
-    movu            xm5, [r0 + r1 * 2]              ; m5 = row 10
-    punpckhwd       xm7, xm2, xm5
-    punpcklwd       xm2, xm5
-    vinserti128     m2, m2, xm7, 1
-    pmaddwd         m7, m2, [r5 + 1 * mmsize]
-    paddd           m1, m7
-    pmaddwd         m2, [r5]
-
-%ifidn %1,sp
-    paddd           m6, m8
-    paddd           m1, m8
-    psrad           m6, 12
-    psrad           m1, 12
-%else
-    psrad           m6, 6
-    psrad           m1, 6
-%endif
-    packssdw        m6, m1
-%ifidn %1,sp
-    packuswb        m4, m6
-    vpermd          m4, m3, m4
-    vextracti128    xm6, m4, 1
-    movq            [r2], xm4
-    movhps          [r2 + r3], xm4
-    movq            [r2 + r3 * 2], xm6
-    movhps          [r2 + r6], xm6
+    lea             r8, [r2 + r3 * 4]
+    movu            [r8], xm4
+    movu            [r8 + r3], xm5
 %else
-    vpermq          m4, m4, 11011000b
-    vpermq          m6, m6, 11011000b
-    vextracti128    xm7, m4, 1
-    vextracti128    xm1, m6, 1
-    movu            [r2], xm4
-    movu            [r2 + r3], xm7
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r6], xm1
+    psubw           m0, m14                         ; m0 = word: row 0
+    psubw           m1, m14                         ; m1 = word: row 1
+    psubw           m2, m14                         ; m2 = word: row 2
+    psubw           m3, m14                         ; m3 = word: row 3
+    psubw           m4, m14                         ; m4 = word: row 4
+    psubw           m5, m14                         ; m5 = word: row 5
+    movu            [r2], m0
+    movu            [r2 + r3], m1
+    movu            [r2 + r3 * 2], m2
+    movu            [r2 + r6], m3
+    lea             r8, [r2 + r3 * 4]
+    movu            [r8], m4
+    movu            [r8 + r3], m5
 %endif
-    lea             r2, [r2 + r3 * 4]
 
-    movu            xm7, [r0 + r4]                  ; m7 = row 11
-    punpckhwd       xm1, xm5, xm7
-    punpcklwd       xm5, xm7
-    vinserti128     m5, m5, xm1, 1
-    pmaddwd         m1, m5, [r5 + 1 * mmsize]
-    paddd           m0, m1
-    pmaddwd         m5, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm1, [r0]                       ; m1 = row 12
-    punpckhwd       xm4, xm7, xm1
-    punpcklwd       xm7, xm1
-    vinserti128     m7, m7, xm4, 1
-    pmaddwd         m4, m7, [r5 + 1 * mmsize]
-    paddd           m2, m4
-    pmaddwd         m7, [r5]
-%ifidn %1,sp
-    paddd           m0, m8
-    paddd           m2, m8
-    psrad           m0, 12
-    psrad           m2, 12
-%else
-    psrad           m0, 6
-    psrad           m2, 6
-%endif
-    packssdw        m0, m2
+    movu            xm13, [r7 + r1]                 ; m13 = row 13
+    punpckhbw       xm0, xm12, xm13
+    punpcklbw       xm12, xm13
+    vinserti128     m12, m12, xm0, 1
+    pmaddubsw       m0, m12, [r5 + 3 * mmsize]
+    paddw           m6, m0
+    movu            xm0, [r7 + r1 * 2]              ; m0 = row 14
+    punpckhbw       xm1, xm13, xm0
+    punpcklbw       xm13, xm0
+    vinserti128     m13, m13, xm1, 1
+    pmaddubsw       m1, m13, [r5 + 3 * mmsize]
+    paddw           m7, m1
 
-    movu            xm4, [r0 + r1]                  ; m4 = row 13
-    punpckhwd       xm2, xm1, xm4
-    punpcklwd       xm1, xm4
-    vinserti128     m1, m1, xm2, 1
-    pmaddwd         m2, m1, [r5 + 1 * mmsize]
-    paddd           m5, m2
-    pmaddwd         m1, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 14
-    punpckhwd       xm6, xm4, xm2
-    punpcklwd       xm4, xm2
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m7, m6
-    pmaddwd         m4, [r5]
-%ifidn %1,sp
-    paddd           m5, m8
-    paddd           m7, m8
-    psrad           m5, 12
-    psrad           m7, 12
+%ifidn %1,pp
+    pmulhrsw        m6, m14                         ; m6 = word: row 6
+    pmulhrsw        m7, m14                         ; m7 = word: row 7
+    packuswb        m6, m7
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm7, m6, 1
+    movu            [r8 + r3 * 2], xm6
+    movu            [r8 + r6], xm7
 %else
-    psrad           m5, 6
-    psrad           m7, 6
+    psubw           m6, m14                         ; m6 = word: row 6
+    psubw           m7, m14                         ; m7 = word: row 7
+    movu            [r8 + r3 * 2], m6
+    movu            [r8 + r6], m7
 %endif
-    packssdw        m5, m7
-%ifidn %1,sp
-    packuswb        m0, m5
-    vpermd          m0, m3, m0
-    vextracti128    xm5, m0, 1
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm5
-    movhps          [r2 + r6], xm5
+%endmacro
+
+%macro FILTER_VER_LUMA_AVX2_24x32 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_24x32, 4, 11, 15
+    mov             r4d, r4m
+    shl             r4d, 7
+%ifdef PIC
+    lea             r5, [tab_LumaCoeffVer_32]
+    add             r5, r4
 %else
-    vpermq          m0, m0, 11011000b
-    vpermq          m5, m5, 11011000b
-    vextracti128    xm7, m0, 1
-    vextracti128    xm6, m5, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm7
-    movu            [r2 + r3 * 2], xm5
-    movu            [r2 + r6], xm6
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
-    lea             r2, [r2 + r3 * 4]
-
-    movu            xm6, [r0 + r4]                  ; m6 = row 15
-    punpckhwd       xm5, xm2, xm6
-    punpcklwd       xm2, xm6
-    vinserti128     m2, m2, xm5, 1
-    pmaddwd         m5, m2, [r5 + 1 * mmsize]
-    paddd           m1, m5
-    pmaddwd         m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm0, [r0]                       ; m0 = row 16
-    punpckhwd       xm5, xm6, xm0
-    punpcklwd       xm6, xm0
-    vinserti128     m6, m6, xm5, 1
-    pmaddwd         m5, m6, [r5 + 1 * mmsize]
-    paddd           m4, m5
-    pmaddwd         m6, [r5]
-%ifidn %1,sp
-    paddd           m1, m8
-    paddd           m4, m8
-    psrad           m1, 12
-    psrad           m4, 12
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+%ifidn %1,ps
+    add             r3d, r3d
+    vbroadcasti128  m14, [pw_2000]
 %else
-    psrad           m1, 6
-    psrad           m4, 6
+    mova            m14, [pw_512]
 %endif
-    packssdw        m1, m4
-
-    movu            xm5, [r0 + r1]                  ; m5 = row 17
-    punpckhwd       xm4, xm0, xm5
-    punpcklwd       xm0, xm5
-    vinserti128     m0, m0, xm4, 1
-    pmaddwd         m0, [r5 + 1 * mmsize]
-    paddd           m2, m0
-    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
-    punpckhwd       xm0, xm5, xm4
-    punpcklwd       xm5, xm4
-    vinserti128     m5, m5, xm0, 1
-    pmaddwd         m5, [r5 + 1 * mmsize]
-    paddd           m6, m5
-%ifidn %1,sp
-    paddd           m2, m8
-    paddd           m6, m8
-    psrad           m2, 12
-    psrad           m6, 12
+    lea             r6, [r3 * 3]
+    lea             r10, [r1 * 4]
+    mov             r9d, 2
+.loopH:
+    PROCESS_LUMA_AVX2_W16_16R %1
+%ifidn %1,pp
+    add             r2, 16
 %else
-    psrad           m2, 6
-    psrad           m6, 6
+    add             r2, 32
 %endif
-    packssdw        m2, m6
-%ifidn %1,sp
-    packuswb        m1, m2
-    vpermd          m1, m3, m1
-    vextracti128    xm2, m1, 1
-    movq            [r2], xm1
-    movhps          [r2 + r3], xm1
-    movq            [r2 + r3 * 2], xm2
+    add             r0, 16
+    movq            xm1, [r0]                       ; m1 = row 0
+    movq            xm2, [r0 + r1]                  ; m2 = row 1
+    punpcklbw       xm1, xm2
+    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
+    punpcklbw       xm2, xm3
+    vinserti128     m5, m1, xm2, 1
+    pmaddubsw       m5, [r5]
+    movq            xm4, [r0 + r4]                  ; m4 = row 3
+    punpcklbw       xm3, xm4
+    lea             r7, [r0 + r1 * 4]
+    movq            xm1, [r7]                       ; m1 = row 4
+    punpcklbw       xm4, xm1
+    vinserti128     m2, m3, xm4, 1
+    pmaddubsw       m0, m2, [r5 + 1 * mmsize]
+    paddw           m5, m0
+    pmaddubsw       m2, [r5]
+    movq            xm3, [r7 + r1]                  ; m3 = row 5
+    punpcklbw       xm1, xm3
+    movq            xm4, [r7 + r1 * 2]              ; m4 = row 6
+    punpcklbw       xm3, xm4
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m3, m1, [r5 + 2 * mmsize]
+    paddw           m5, m3
+    pmaddubsw       m0, m1, [r5 + 1 * mmsize]
+    paddw           m2, m0
+    pmaddubsw       m1, [r5]
+    movq            xm3, [r7 + r4]                  ; m3 = row 7
+    punpcklbw       xm4, xm3
+    lea             r7, [r7 + r1 * 4]
+    movq            xm0, [r7]                       ; m0 = row 8
+    punpcklbw       xm3, xm0
+    vinserti128     m4, m4, xm3, 1
+    pmaddubsw       m3, m4, [r5 + 3 * mmsize]
+    paddw           m5, m3
+    pmaddubsw       m3, m4, [r5 + 2 * mmsize]
+    paddw           m2, m3
+    pmaddubsw       m3, m4, [r5 + 1 * mmsize]
+    paddw           m1, m3
+    pmaddubsw       m4, [r5]
+    movq            xm3, [r7 + r1]                  ; m3 = row 9
+    punpcklbw       xm0, xm3
+    movq            xm6, [r7 + r1 * 2]              ; m6 = row 10
+    punpcklbw       xm3, xm6
+    vinserti128     m0, m0, xm3, 1
+    pmaddubsw       m3, m0, [r5 + 3 * mmsize]
+    paddw           m2, m3
+    pmaddubsw       m3, m0, [r5 + 2 * mmsize]
+    paddw           m1, m3
+    pmaddubsw       m3, m0, [r5 + 1 * mmsize]
+    paddw           m4, m3
+    pmaddubsw       m0, [r5]
+
+    movq            xm3, [r7 + r4]                  ; m3 = row 11
+    punpcklbw       xm6, xm3
+    lea             r7, [r7 + r1 * 4]
+    movq            xm7, [r7]                       ; m7 = row 12
+    punpcklbw       xm3, xm7
+    vinserti128     m6, m6, xm3, 1
+    pmaddubsw       m3, m6, [r5 + 3 * mmsize]
+    paddw           m1, m3
+    pmaddubsw       m3, m6, [r5 + 2 * mmsize]
+    paddw           m4, m3
+    pmaddubsw       m3, m6, [r5 + 1 * mmsize]
+    paddw           m0, m3
+    pmaddubsw       m6, [r5]
+    movq            xm3, [r7 + r1]                  ; m3 = row 13
+    punpcklbw       xm7, xm3
+    movq            xm8, [r7 + r1 * 2]              ; m8 = row 14
+    punpcklbw       xm3, xm8
+    vinserti128     m7, m7, xm3, 1
+    pmaddubsw       m3, m7, [r5 + 3 * mmsize]
+    paddw           m4, m3
+    pmaddubsw       m3, m7, [r5 + 2 * mmsize]
+    paddw           m0, m3
+    pmaddubsw       m3, m7, [r5 + 1 * mmsize]
+    paddw           m6, m3
+    pmaddubsw       m7, [r5]
+    movq            xm3, [r7 + r4]                  ; m3 = row 15
+    punpcklbw       xm8, xm3
+    lea             r7, [r7 + r1 * 4]
+    movq            xm9, [r7]                       ; m9 = row 16
+    punpcklbw       xm3, xm9
+    vinserti128     m8, m8, xm3, 1
+    pmaddubsw       m3, m8, [r5 + 3 * mmsize]
+    paddw           m0, m3
+    pmaddubsw       m3, m8, [r5 + 2 * mmsize]
+    paddw           m6, m3
+    pmaddubsw       m3, m8, [r5 + 1 * mmsize]
+    paddw           m7, m3
+    pmaddubsw       m8, [r5]
+    movq            xm3, [r7 + r1]                  ; m3 = row 17
+    punpcklbw       xm9, xm3
+    movq            xm10, [r7 + r1 * 2]             ; m10 = row 18
+    punpcklbw       xm3, xm10
+    vinserti128     m9, m9, xm3, 1
+    pmaddubsw       m3, m9, [r5 + 3 * mmsize]
+    paddw           m6, m3
+    pmaddubsw       m3, m9, [r5 + 2 * mmsize]
+    paddw           m7, m3
+    pmaddubsw       m3, m9, [r5 + 1 * mmsize]
+    paddw           m8, m3
+    movq            xm3, [r7 + r4]                  ; m3 = row 19
+    punpcklbw       xm10, xm3
+    lea             r7, [r7 + r1 * 4]
+    movq            xm9, [r7]                       ; m9 = row 20
+    punpcklbw       xm3, xm9
+    vinserti128     m10, m10, xm3, 1
+    pmaddubsw       m3, m10, [r5 + 3 * mmsize]
+    paddw           m7, m3
+    pmaddubsw       m3, m10, [r5 + 2 * mmsize]
+    paddw           m8, m3
+    movq            xm3, [r7 + r1]                  ; m3 = row 21
+    punpcklbw       xm9, xm3
+    movq            xm10, [r7 + r1 * 2]             ; m10 = row 22
+    punpcklbw       xm3, xm10
+    vinserti128     m9, m9, xm3, 1
+    pmaddubsw       m3, m9, [r5 + 3 * mmsize]
+    paddw           m8, m3
+%ifidn %1,pp
+    pmulhrsw        m5, m14                         ; m5 = word: row 0, row 1
+    pmulhrsw        m2, m14                         ; m2 = word: row 2, row 3
+    pmulhrsw        m1, m14                         ; m1 = word: row 4, row 5
+    pmulhrsw        m4, m14                         ; m4 = word: row 6, row 7
+    pmulhrsw        m0, m14                         ; m0 = word: row 8, row 9
+    pmulhrsw        m6, m14                         ; m6 = word: row 10, row 11
+    pmulhrsw        m7, m14                         ; m7 = word: row 12, row 13
+    pmulhrsw        m8, m14                         ; m8 = word: row 14, row 15
+    packuswb        m5, m2
+    packuswb        m1, m4
+    packuswb        m0, m6
+    packuswb        m7, m8
+    vextracti128    xm2, m5, 1
+    vextracti128    xm4, m1, 1
+    vextracti128    xm6, m0, 1
+    vextracti128    xm8, m7, 1
+    movq            [r2], xm5
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm5
     movhps          [r2 + r6], xm2
+    lea             r8, [r2 + r3 * 4]
+    movq            [r8], xm1
+    movq            [r8 + r3], xm4
+    movhps          [r8 + r3 * 2], xm1
+    movhps          [r8 + r6], xm4
+    lea             r8, [r8 + r3 * 4]
+    movq            [r8], xm0
+    movq            [r8 + r3], xm6
+    movhps          [r8 + r3 * 2], xm0
+    movhps          [r8 + r6], xm6
+    lea             r8, [r8 + r3 * 4]
+    movq            [r8], xm7
+    movq            [r8 + r3], xm8
+    movhps          [r8 + r3 * 2], xm7
+    movhps          [r8 + r6], xm8
 %else
-    vpermq          m1, m1, 11011000b
-    vpermq          m2, m2, 11011000b
-    vextracti128    xm6, m1, 1
-    vextracti128    xm4, m2, 1
-    movu            [r2], xm1
-    movu            [r2 + r3], xm6
+    psubw           m5, m14                         ; m5 = word: row 0, row 1
+    psubw           m2, m14                         ; m2 = word: row 2, row 3
+    psubw           m1, m14                         ; m1 = word: row 4, row 5
+    psubw           m4, m14                         ; m4 = word: row 6, row 7
+    psubw           m0, m14                         ; m0 = word: row 8, row 9
+    psubw           m6, m14                         ; m6 = word: row 10, row 11
+    psubw           m7, m14                         ; m7 = word: row 12, row 13
+    psubw           m8, m14                         ; m8 = word: row 14, row 15
+    vextracti128    xm3, m5, 1
+    movu            [r2], xm5
+    movu            [r2 + r3], xm3
+    vextracti128    xm3, m2, 1
     movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm4
+    movu            [r2 + r6], xm3
+    vextracti128    xm3, m1, 1
+    lea             r8, [r2 + r3 * 4]
+    movu            [r8], xm1
+    movu            [r8 + r3], xm3
+    vextracti128    xm3, m4, 1
+    movu            [r8 + r3 * 2], xm4
+    movu            [r8 + r6], xm3
+    vextracti128    xm3, m0, 1
+    lea             r8, [r8 + r3 * 4]
+    movu            [r8], xm0
+    movu            [r8 + r3], xm3
+    vextracti128    xm3, m6, 1
+    movu            [r8 + r3 * 2], xm6
+    movu            [r8 + r6], xm3
+    vextracti128    xm3, m7, 1
+    lea             r8, [r8 + r3 * 4]
+    movu            [r8], xm7
+    movu            [r8 + r3], xm3
+    vextracti128    xm3, m8, 1
+    movu            [r8 + r3 * 2], xm8
+    movu            [r8 + r6], xm3
 %endif
-    lea             r2, [r2 + r3 * 4]
-%endrep
+    sub             r7, r10
+    lea             r0, [r7 - 16]
+%ifidn %1,pp
+    lea             r2, [r8 + r3 * 4 - 16]
+%else
+    lea             r2, [r8 + r3 * 4 - 32]
+%endif
+    dec             r9d
+    jnz             .loopH
     RET
 %endif
 %endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_8xN sp, 16
-    FILTER_VER_CHROMA_S_AVX2_8xN sp, 32
-    FILTER_VER_CHROMA_S_AVX2_8xN sp, 64
-    FILTER_VER_CHROMA_S_AVX2_8xN ss, 16
-    FILTER_VER_CHROMA_S_AVX2_8xN ss, 32
-    FILTER_VER_CHROMA_S_AVX2_8xN ss, 64
+    FILTER_VER_LUMA_AVX2_24x32 pp
+    FILTER_VER_LUMA_AVX2_24x32 ps
 
-%macro FILTER_VER_CHROMA_S_AVX2_Nx24 2
-%if ARCH_X86_64 == 1
+%macro FILTER_VER_LUMA_AVX2_32xN 3
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_%2x24, 4, 10, 10
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%3_%1x%2, 4, 12, 15
     mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-
+    shl             r4d, 7
 %ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
+    lea             r5, [tab_LumaCoeffVer_32]
     add             r5, r4
 %else
-    lea             r5, [pw_ChromaCoeffV + r4]
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
-
     lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,sp
-    mova            m9, [pd_526336]
-%else
+    sub             r0, r4
+%ifidn %3,ps
     add             r3d, r3d
+    vbroadcasti128  m14, [pw_2000]
+%else
+    mova            m14, [pw_512]
 %endif
     lea             r6, [r3 * 3]
-    mov             r9d, %2 / 8
+    lea             r11, [r1 * 4]
+    mov             r9d, %2 / 16
+.loopH:
+    mov             r10d, %1 / 16
 .loopW:
-    PROCESS_CHROMA_S_AVX2_W8_16R %1
-%ifidn %1,sp
-    add             r2, 8
-%else
+    PROCESS_LUMA_AVX2_W16_16R %3
+%ifidn %3,pp
     add             r2, 16
+%else
+    add             r2, 32
 %endif
     add             r0, 16
-    dec             r9d
+    dec             r10d
     jnz             .loopW
-%ifidn %1,sp
-    lea             r2, [r8 + r3 * 4 - %2 + 8]
-%else
-    lea             r2, [r8 + r3 * 4 - 2 * %2 + 16]
-%endif
-    lea             r0, [r7 - 2 * %2 + 16]
-    mova            m7, m9
-    mov             r9d, %2 / 8
-.loop:
-    PROCESS_CHROMA_S_AVX2_W8_8R %1
-%ifidn %1,sp
-    add             r2, 8
+    sub             r7, r11
+    lea             r0, [r7 - 16]
+%ifidn %3,pp
+    lea             r2, [r8 + r3 * 4 - 16]
 %else
-    add             r2, 16
+    lea             r2, [r8 + r3 * 4 - 32]
 %endif
-    add             r0, 16
     dec             r9d
-    jnz             .loop
+    jnz             .loopH
     RET
 %endif
 %endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_Nx24 sp, 32
-    FILTER_VER_CHROMA_S_AVX2_Nx24 sp, 16
-    FILTER_VER_CHROMA_S_AVX2_Nx24 ss, 32
-    FILTER_VER_CHROMA_S_AVX2_Nx24 ss, 16
+    FILTER_VER_LUMA_AVX2_32xN 32, 32, pp
+    FILTER_VER_LUMA_AVX2_32xN 32, 64, pp
+    FILTER_VER_LUMA_AVX2_32xN 32, 32, ps
+    FILTER_VER_LUMA_AVX2_32xN 32, 64, ps
 
-%macro FILTER_VER_CHROMA_S_AVX2_2x8 1
+%macro FILTER_VER_LUMA_AVX2_32x16 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_2x8, 4, 6, 7
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_32x16, 4, 10, 15
     mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-    sub             r0, r1
+    shl             r4d, 7
 
 %ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
+    lea             r5, [tab_LumaCoeffVer_32]
     add             r5, r4
 %else
-    lea             r5, [pw_ChromaCoeffV + r4]
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
 
     lea             r4, [r1 * 3]
-%ifidn %1,sp
-    mova            m6, [pd_526336]
-%else
+    sub             r0, r4
+%ifidn %1,ps
     add             r3d, r3d
-%endif
-    movd            xm0, [r0]
-    movd            xm1, [r0 + r1]
-    punpcklwd       xm0, xm1
-    movd            xm2, [r0 + r1 * 2]
-    punpcklwd       xm1, xm2
-    punpcklqdq      xm0, xm1                        ; m0 = [2 1 1 0]
-    movd            xm3, [r0 + r4]
-    punpcklwd       xm2, xm3
-    lea             r0, [r0 + 4 * r1]
-    movd            xm4, [r0]
-    punpcklwd       xm3, xm4
-    punpcklqdq      xm2, xm3                        ; m2 = [4 3 3 2]
-    vinserti128     m0, m0, xm2, 1                  ; m0 = [4 3 3 2 2 1 1 0]
-    movd            xm1, [r0 + r1]
-    punpcklwd       xm4, xm1
-    movd            xm3, [r0 + r1 * 2]
-    punpcklwd       xm1, xm3
-    punpcklqdq      xm4, xm1                        ; m4 = [6 5 5 4]
-    vinserti128     m2, m2, xm4, 1                  ; m2 = [6 5 5 4 4 3 3 2]
-    pmaddwd         m0, [r5]
-    pmaddwd         m2, [r5 + 1 * mmsize]
-    paddd           m0, m2
-    movd            xm1, [r0 + r4]
-    punpcklwd       xm3, xm1
-    lea             r0, [r0 + 4 * r1]
-    movd            xm2, [r0]
-    punpcklwd       xm1, xm2
-    punpcklqdq      xm3, xm1                        ; m3 = [8 7 7 6]
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [8 7 7 6 6 5 5 4]
-    movd            xm1, [r0 + r1]
-    punpcklwd       xm2, xm1
-    movd            xm5, [r0 + r1 * 2]
-    punpcklwd       xm1, xm5
-    punpcklqdq      xm2, xm1                        ; m2 = [10 9 9 8]
-    vinserti128     m3, m3, xm2, 1                  ; m3 = [10 9 9 8 8 7 7 6]
-    pmaddwd         m4, [r5]
-    pmaddwd         m3, [r5 + 1 * mmsize]
-    paddd           m4, m3
-%ifidn %1,sp
-    paddd           m0, m6
-    paddd           m4, m6
-    psrad           m0, 12
-    psrad           m4, 12
+    vbroadcasti128  m14, [pw_2000]
 %else
-    psrad           m0, 6
-    psrad           m4, 6
+    mova            m14, [pw_512]
 %endif
-    packssdw        m0, m4
-    vextracti128    xm4, m0, 1
-    lea             r4, [r3 * 3]
-%ifidn %1,sp
-    packuswb        xm0, xm4
-    pextrw          [r2], xm0, 0
-    pextrw          [r2 + r3], xm0, 1
-    pextrw          [r2 + 2 * r3], xm0, 4
-    pextrw          [r2 + r4], xm0, 5
-    lea             r2, [r2 + r3 * 4]
-    pextrw          [r2], xm0, 2
-    pextrw          [r2 + r3], xm0, 3
-    pextrw          [r2 + 2 * r3], xm0, 6
-    pextrw          [r2 + r4], xm0, 7
+    lea             r6, [r3 * 3]
+    mov             r9d, 2
+.loopW:
+    PROCESS_LUMA_AVX2_W16_16R %1
+%ifidn %1,pp
+    add             r2, 16
 %else
-    movd            [r2], xm0
-    pextrd          [r2 + r3], xm0, 1
-    movd            [r2 + 2 * r3], xm4
-    pextrd          [r2 + r4], xm4, 1
-    lea             r2, [r2 + r3 * 4]
-    pextrd          [r2], xm0, 2
-    pextrd          [r2 + r3], xm0, 3
-    pextrd          [r2 + 2 * r3], xm4, 2
-    pextrd          [r2 + r4], xm4, 3
+    add             r2, 32
 %endif
+    add             r0, 16
+    dec             r9d
+    jnz             .loopW
     RET
+%endif
 %endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_2x8 sp
-    FILTER_VER_CHROMA_S_AVX2_2x8 ss
+    FILTER_VER_LUMA_AVX2_32x16 pp
+    FILTER_VER_LUMA_AVX2_32x16 ps
 
-%macro FILTER_VER_CHROMA_S_AVX2_2x16 1
-%if ARCH_X86_64 == 1
+%macro FILTER_VER_LUMA_AVX2_32x24 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_2x16, 4, 6, 9
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_32x24, 4, 10, 15
     mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-    sub             r0, r1
-
+    shl             r4d, 7
 %ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
+    lea             r5, [tab_LumaCoeffVer_32]
     add             r5, r4
 %else
-    lea             r5, [pw_ChromaCoeffV + r4]
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
-
     lea             r4, [r1 * 3]
-%ifidn %1,sp
-    mova            m6, [pd_526336]
-%else
+    sub             r0, r4
+%ifidn %1,ps
     add             r3d, r3d
 %endif
-    movd            xm0, [r0]
-    movd            xm1, [r0 + r1]
-    punpcklwd       xm0, xm1
-    movd            xm2, [r0 + r1 * 2]
-    punpcklwd       xm1, xm2
-    punpcklqdq      xm0, xm1                        ; m0 = [2 1 1 0]
-    movd            xm3, [r0 + r4]
-    punpcklwd       xm2, xm3
-    lea             r0, [r0 + 4 * r1]
-    movd            xm4, [r0]
-    punpcklwd       xm3, xm4
-    punpcklqdq      xm2, xm3                        ; m2 = [4 3 3 2]
-    vinserti128     m0, m0, xm2, 1                  ; m0 = [4 3 3 2 2 1 1 0]
-    movd            xm1, [r0 + r1]
-    punpcklwd       xm4, xm1
-    movd            xm3, [r0 + r1 * 2]
-    punpcklwd       xm1, xm3
-    punpcklqdq      xm4, xm1                        ; m4 = [6 5 5 4]
-    vinserti128     m2, m2, xm4, 1                  ; m2 = [6 5 5 4 4 3 3 2]
-    pmaddwd         m0, [r5]
-    pmaddwd         m2, [r5 + 1 * mmsize]
-    paddd           m0, m2
-    movd            xm1, [r0 + r4]
-    punpcklwd       xm3, xm1
-    lea             r0, [r0 + 4 * r1]
-    movd            xm2, [r0]
-    punpcklwd       xm1, xm2
-    punpcklqdq      xm3, xm1                        ; m3 = [8 7 7 6]
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [8 7 7 6 6 5 5 4]
-    movd            xm1, [r0 + r1]
-    punpcklwd       xm2, xm1
-    movd            xm5, [r0 + r1 * 2]
-    punpcklwd       xm1, xm5
-    punpcklqdq      xm2, xm1                        ; m2 = [10 9 9 8]
-    vinserti128     m3, m3, xm2, 1                  ; m3 = [10 9 9 8 8 7 7 6]
-    pmaddwd         m4, [r5]
-    pmaddwd         m3, [r5 + 1 * mmsize]
-    paddd           m4, m3
-    movd            xm1, [r0 + r4]
-    punpcklwd       xm5, xm1
-    lea             r0, [r0 + 4 * r1]
-    movd            xm3, [r0]
-    punpcklwd       xm1, xm3
-    punpcklqdq      xm5, xm1                        ; m5 = [12 11 11 10]
-    vinserti128     m2, m2, xm5, 1                  ; m2 = [12 11 11 10 10 9 9 8]
-    movd            xm1, [r0 + r1]
-    punpcklwd       xm3, xm1
-    movd            xm7, [r0 + r1 * 2]
-    punpcklwd       xm1, xm7
-    punpcklqdq      xm3, xm1                        ; m3 = [14 13 13 12]
-    vinserti128     m5, m5, xm3, 1                  ; m5 = [14 13 13 12 12 11 11 10]
-    pmaddwd         m2, [r5]
-    pmaddwd         m5, [r5 + 1 * mmsize]
-    paddd           m2, m5
-    movd            xm5, [r0 + r4]
-    punpcklwd       xm7, xm5
-    lea             r0, [r0 + 4 * r1]
-    movd            xm1, [r0]
-    punpcklwd       xm5, xm1
-    punpcklqdq      xm7, xm5                        ; m7 = [16 15 15 14]
-    vinserti128     m3, m3, xm7, 1                  ; m3 = [16 15 15 14 14 13 13 12]
-    movd            xm5, [r0 + r1]
-    punpcklwd       xm1, xm5
-    movd            xm8, [r0 + r1 * 2]
-    punpcklwd       xm5, xm8
-    punpcklqdq      xm1, xm5                        ; m1 = [18 17 17 16]
-    vinserti128     m7, m7, xm1, 1                  ; m7 = [18 17 17 16 16 15 15 14]
-    pmaddwd         m3, [r5]
-    pmaddwd         m7, [r5 + 1 * mmsize]
-    paddd           m3, m7
-%ifidn %1,sp
-    paddd           m0, m6
-    paddd           m4, m6
-    paddd           m2, m6
-    paddd           m3, m6
-    psrad           m0, 12
-    psrad           m4, 12
-    psrad           m2, 12
-    psrad           m3, 12
+    lea             r6, [r3 * 3]
+%ifidn %1,pp
+    mova            m14, [pw_512]
 %else
-    psrad           m0, 6
-    psrad           m4, 6
-    psrad           m2, 6
-    psrad           m3, 6
+    vbroadcasti128  m14, [pw_2000]
 %endif
-    packssdw        m0, m4
-    packssdw        m2, m3
-    lea             r4, [r3 * 3]
-%ifidn %1,sp
-    packuswb        m0, m2
-    vextracti128    xm2, m0, 1
-    pextrw          [r2], xm0, 0
-    pextrw          [r2 + r3], xm0, 1
-    pextrw          [r2 + 2 * r3], xm2, 0
-    pextrw          [r2 + r4], xm2, 1
-    lea             r2, [r2 + r3 * 4]
-    pextrw          [r2], xm0, 2
-    pextrw          [r2 + r3], xm0, 3
-    pextrw          [r2 + 2 * r3], xm2, 2
-    pextrw          [r2 + r4], xm2, 3
-    lea             r2, [r2 + r3 * 4]
-    pextrw          [r2], xm0, 4
-    pextrw          [r2 + r3], xm0, 5
-    pextrw          [r2 + 2 * r3], xm2, 4
-    pextrw          [r2 + r4], xm2, 5
-    lea             r2, [r2 + r3 * 4]
-    pextrw          [r2], xm0, 6
-    pextrw          [r2 + r3], xm0, 7
-    pextrw          [r2 + 2 * r3], xm2, 6
-    pextrw          [r2 + r4], xm2, 7
+    mov             r9d, 2
+.loopW:
+    PROCESS_LUMA_AVX2_W16_16R %1
+%ifidn %1,pp
+    add             r2, 16
 %else
-    vextracti128    xm4, m0, 1
-    vextracti128    xm3, m2, 1
-    movd            [r2], xm0
-    pextrd          [r2 + r3], xm0, 1
-    movd            [r2 + 2 * r3], xm4
-    pextrd          [r2 + r4], xm4, 1
-    lea             r2, [r2 + r3 * 4]
-    pextrd          [r2], xm0, 2
-    pextrd          [r2 + r3], xm0, 3
-    pextrd          [r2 + 2 * r3], xm4, 2
-    pextrd          [r2 + r4], xm4, 3
-    lea             r2, [r2 + r3 * 4]
-    movd            [r2], xm2
-    pextrd          [r2 + r3], xm2, 1
-    movd            [r2 + 2 * r3], xm3
-    pextrd          [r2 + r4], xm3, 1
-    lea             r2, [r2 + r3 * 4]
-    pextrd          [r2], xm2, 2
-    pextrd          [r2 + r3], xm2, 3
-    pextrd          [r2 + 2 * r3], xm3, 2
-    pextrd          [r2 + r4], xm3, 3
+    add             r2, 32
+%endif
+    add             r0, 16
+    dec             r9d
+    jnz             .loopW
+    lea             r9, [r1 * 4]
+    sub             r7, r9
+    lea             r0, [r7 - 16]
+%ifidn %1,pp
+    lea             r2, [r8 + r3 * 4 - 16]
+%else
+    lea             r2, [r8 + r3 * 4 - 32]
+%endif
+    mov             r9d, 2
+.loop:
+    PROCESS_LUMA_AVX2_W16_8R %1
+%ifidn %1,pp
+    add             r2, 16
+%else
+    add             r2, 32
 %endif
+    add             r0, 16
+    dec             r9d
+    jnz             .loop
     RET
 %endif
 %endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_2x16 sp
-    FILTER_VER_CHROMA_S_AVX2_2x16 ss
+    FILTER_VER_LUMA_AVX2_32x24 pp
+    FILTER_VER_LUMA_AVX2_32x24 ps
 
-%macro FILTER_VER_CHROMA_S_AVX2_6x8 1
+%macro FILTER_VER_LUMA_AVX2_32x8 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_6x8, 4, 6, 8
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_32x8, 4, 10, 15
     mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
+    shl             r4d, 7
 
 %ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
+    lea             r5, [tab_LumaCoeffVer_32]
     add             r5, r4
 %else
-    lea             r5, [pw_ChromaCoeffV + r4]
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
 
     lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,sp
-    mova            m7, [pd_526336]
-%else
+    sub             r0, r4
+%ifidn %1,ps
     add             r3d, r3d
 %endif
-
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    pmaddwd         m2, [r5]
-    paddd           m0, m4
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    pmaddwd         m3, [r5]
-    paddd           m1, m5
-%ifidn %1,sp
-    paddd           m0, m7
-    paddd           m1, m7
-    psrad           m0, 12
-    psrad           m1, 12
-%else
-    psrad           m0, 6
-    psrad           m1, 6
-%endif
-    packssdw        m0, m1
-
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m2, m6
-    pmaddwd         m4, [r5]
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm1, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm1, 1
-    pmaddwd         m1, m5, [r5 + 1 * mmsize]
-    pmaddwd         m5, [r5]
-    paddd           m3, m1
-%ifidn %1,sp
-    paddd           m2, m7
-    paddd           m3, m7
-    psrad           m2, 12
-    psrad           m3, 12
-%else
-    psrad           m2, 6
-    psrad           m3, 6
-%endif
-    packssdw        m2, m3
-
-    movu            xm1, [r0 + r4]                  ; m1 = row 7
-    punpckhwd       xm3, xm6, xm1
-    punpcklwd       xm6, xm1
-    vinserti128     m6, m6, xm3, 1
-    pmaddwd         m3, m6, [r5 + 1 * mmsize]
-    pmaddwd         m6, [r5]
-    paddd           m4, m3
-
-    lea             r4, [r3 * 3]
-%ifidn %1,sp
-    packuswb        m0, m2
-    vextracti128    xm2, m0, 1
-    movd            [r2], xm0
-    pextrw          [r2 + 4], xm2, 0
-    pextrd          [r2 + r3], xm0, 1
-    pextrw          [r2 + r3 + 4], xm2, 2
-    pextrd          [r2 + r3 * 2], xm0, 2
-    pextrw          [r2 + r3 * 2 + 4], xm2, 4
-    pextrd          [r2 + r4], xm0, 3
-    pextrw          [r2 + r4 + 4], xm2, 6
-%else
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r4], xm2
-    vextracti128    xm0, m0, 1
-    vextracti128    xm3, m2, 1
-    movd            [r2 + 8], xm0
-    pextrd          [r2 + r3 + 8], xm0, 2
-    movd            [r2 + r3 * 2 + 8], xm3
-    pextrd          [r2 + r4 + 8], xm3, 2
-%endif
-    lea             r2, [r2 + r3 * 4]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm0, [r0]                       ; m0 = row 8
-    punpckhwd       xm2, xm1, xm0
-    punpcklwd       xm1, xm0
-    vinserti128     m1, m1, xm2, 1
-    pmaddwd         m2, m1, [r5 + 1 * mmsize]
-    pmaddwd         m1, [r5]
-    paddd           m5, m2
-%ifidn %1,sp
-    paddd           m4, m7
-    paddd           m5, m7
-    psrad           m4, 12
-    psrad           m5, 12
-%else
-    psrad           m4, 6
-    psrad           m5, 6
-%endif
-    packssdw        m4, m5
-
-    movu            xm2, [r0 + r1]                  ; m2 = row 9
-    punpckhwd       xm5, xm0, xm2
-    punpcklwd       xm0, xm2
-    vinserti128     m0, m0, xm5, 1
-    pmaddwd         m0, [r5 + 1 * mmsize]
-    paddd           m6, m0
-    movu            xm5, [r0 + r1 * 2]              ; m5 = row 10
-    punpckhwd       xm0, xm2, xm5
-    punpcklwd       xm2, xm5
-    vinserti128     m2, m2, xm0, 1
-    pmaddwd         m2, [r5 + 1 * mmsize]
-    paddd           m1, m2
-
-%ifidn %1,sp
-    paddd           m6, m7
-    paddd           m1, m7
-    psrad           m6, 12
-    psrad           m1, 12
+    lea             r6, [r3 * 3]
+%ifidn %1,pp
+    mova            m14, [pw_512]
 %else
-    psrad           m6, 6
-    psrad           m1, 6
+    vbroadcasti128  m14, [pw_2000]
 %endif
-    packssdw        m6, m1
-%ifidn %1,sp
-    packuswb        m4, m6
-    vextracti128    xm6, m4, 1
-    movd            [r2], xm4
-    pextrw          [r2 + 4], xm6, 0
-    pextrd          [r2 + r3], xm4, 1
-    pextrw          [r2 + r3 + 4], xm6, 2
-    pextrd          [r2 + r3 * 2], xm4, 2
-    pextrw          [r2 + r3 * 2 + 4], xm6, 4
-    pextrd          [r2 + r4], xm4, 3
-    pextrw          [r2 + r4 + 4], xm6, 6
+    mov             r9d, 2
+.loopW:
+    PROCESS_LUMA_AVX2_W16_8R %1
+%ifidn %1,pp
+    add             r2, 16
 %else
-    movq            [r2], xm4
-    movhps          [r2 + r3], xm4
-    movq            [r2 + r3 * 2], xm6
-    movhps          [r2 + r4], xm6
-    vextracti128    xm5, m4, 1
-    vextracti128    xm1, m6, 1
-    movd            [r2 + 8], xm5
-    pextrd          [r2 + r3 + 8], xm5, 2
-    movd            [r2 + r3 * 2 + 8], xm1
-    pextrd          [r2 + r4 + 8], xm1, 2
+    add             r2, 32
 %endif
+    add             r0, 16
+    dec             r9d
+    jnz             .loopW
     RET
+%endif
 %endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_6x8 sp
-    FILTER_VER_CHROMA_S_AVX2_6x8 ss
+    FILTER_VER_LUMA_AVX2_32x8 pp
+    FILTER_VER_LUMA_AVX2_32x8 ps
 
-%macro FILTER_VER_CHROMA_S_AVX2_6x16 1
-%if ARCH_X86_64 == 1
+%macro FILTER_VER_LUMA_AVX2_48x64 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_%1_6x16, 4, 7, 9
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_48x64, 4, 12, 15
     mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
+    shl             r4d, 7
 
 %ifdef PIC
-    lea             r5, [pw_ChromaCoeffV]
+    lea             r5, [tab_LumaCoeffVer_32]
     add             r5, r4
 %else
-    lea             r5, [pw_ChromaCoeffV + r4]
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
 
     lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,sp
-    mova            m8, [pd_526336]
-%else
+    sub             r0, r4
+
+%ifidn %1,ps
     add             r3d, r3d
 %endif
+
     lea             r6, [r3 * 3]
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    paddd           m1, m5
-    pmaddwd         m3, [r5]
-%ifidn %1,sp
-    paddd           m0, m8
-    paddd           m1, m8
-    psrad           m0, 12
-    psrad           m1, 12
-%else
-    psrad           m0, 6
-    psrad           m1, 6
-%endif
-    packssdw        m0, m1
+    lea             r11, [r1 * 4]
 
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m2, m6
-    pmaddwd         m4, [r5]
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm1, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm1, 1
-    pmaddwd         m1, m5, [r5 + 1 * mmsize]
-    pmaddwd         m5, [r5]
-    paddd           m3, m1
-%ifidn %1,sp
-    paddd           m2, m8
-    paddd           m3, m8
-    psrad           m2, 12
-    psrad           m3, 12
+%ifidn %1,pp
+    mova            m14, [pw_512]
 %else
-    psrad           m2, 6
-    psrad           m3, 6
+    vbroadcasti128  m14, [pw_2000]
 %endif
-    packssdw        m2, m3
-%ifidn %1,sp
-    packuswb        m0, m2
-    vextracti128    xm2, m0, 1
-    movd            [r2], xm0
-    pextrw          [r2 + 4], xm2, 0
-    pextrd          [r2 + r3], xm0, 1
-    pextrw          [r2 + r3 + 4], xm2, 2
-    pextrd          [r2 + r3 * 2], xm0, 2
-    pextrw          [r2 + r3 * 2 + 4], xm2, 4
-    pextrd          [r2 + r6], xm0, 3
-    pextrw          [r2 + r6 + 4], xm2, 6
+
+    mov             r9d, 4
+.loopH:
+    mov             r10d, 3
+.loopW:
+    PROCESS_LUMA_AVX2_W16_16R %1
+%ifidn %1,pp
+    add             r2, 16
 %else
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r6], xm2
-    vextracti128    xm0, m0, 1
-    vextracti128    xm3, m2, 1
-    movd            [r2 + 8], xm0
-    pextrd          [r2 + r3 + 8], xm0, 2
-    movd            [r2 + r3 * 2 + 8], xm3
-    pextrd          [r2 + r6 + 8], xm3, 2
+    add             r2, 32
 %endif
-    lea             r2, [r2 + r3 * 4]
-    movu            xm1, [r0 + r4]                  ; m1 = row 7
-    punpckhwd       xm0, xm6, xm1
-    punpcklwd       xm6, xm1
-    vinserti128     m6, m6, xm0, 1
-    pmaddwd         m0, m6, [r5 + 1 * mmsize]
-    pmaddwd         m6, [r5]
-    paddd           m4, m0
-    lea             r0, [r0 + r1 * 4]
-    movu            xm0, [r0]                       ; m0 = row 8
-    punpckhwd       xm2, xm1, xm0
-    punpcklwd       xm1, xm0
-    vinserti128     m1, m1, xm2, 1
-    pmaddwd         m2, m1, [r5 + 1 * mmsize]
-    pmaddwd         m1, [r5]
-    paddd           m5, m2
-%ifidn %1,sp
-    paddd           m4, m8
-    paddd           m5, m8
-    psrad           m4, 12
-    psrad           m5, 12
+    add             r0, 16
+    dec             r10d
+    jnz             .loopW
+    sub             r7, r11
+    lea             r0, [r7 - 32]
+%ifidn %1,pp
+    lea             r2, [r8 + r3 * 4 - 32]
 %else
-    psrad           m4, 6
-    psrad           m5, 6
+    lea             r2, [r8 + r3 * 4 - 64]
 %endif
-    packssdw        m4, m5
+    dec             r9d
+    jnz             .loopH
+    RET
+%endif
+%endmacro
 
-    movu            xm2, [r0 + r1]                  ; m2 = row 9
-    punpckhwd       xm5, xm0, xm2
-    punpcklwd       xm0, xm2
-    vinserti128     m0, m0, xm5, 1
-    pmaddwd         m5, m0, [r5 + 1 * mmsize]
-    paddd           m6, m5
-    pmaddwd         m0, [r5]
-    movu            xm5, [r0 + r1 * 2]              ; m5 = row 10
-    punpckhwd       xm7, xm2, xm5
-    punpcklwd       xm2, xm5
-    vinserti128     m2, m2, xm7, 1
-    pmaddwd         m7, m2, [r5 + 1 * mmsize]
-    paddd           m1, m7
-    pmaddwd         m2, [r5]
+    FILTER_VER_LUMA_AVX2_48x64 pp
+    FILTER_VER_LUMA_AVX2_48x64 ps
 
-%ifidn %1,sp
-    paddd           m6, m8
-    paddd           m1, m8
-    psrad           m6, 12
-    psrad           m1, 12
+%macro FILTER_VER_LUMA_AVX2_64xN 3
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%3_%1x%2, 4, 12, 15
+    mov             r4d, r4m
+    shl             r4d, 7
+
+%ifdef PIC
+    lea             r5, [tab_LumaCoeffVer_32]
+    add             r5, r4
 %else
-    psrad           m6, 6
-    psrad           m1, 6
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
-    packssdw        m6, m1
-%ifidn %1,sp
-    packuswb        m4, m6
-    vextracti128    xm6, m4, 1
-    movd            [r2], xm4
-    pextrw          [r2 + 4], xm6, 0
-    pextrd          [r2 + r3], xm4, 1
-    pextrw          [r2 + r3 + 4], xm6, 2
-    pextrd          [r2 + r3 * 2], xm4, 2
-    pextrw          [r2 + r3 * 2 + 4], xm6, 4
-    pextrd          [r2 + r6], xm4, 3
-    pextrw          [r2 + r6 + 4], xm6, 6
-%else
-    movq            [r2], xm4
-    movhps          [r2 + r3], xm4
-    movq            [r2 + r3 * 2], xm6
-    movhps          [r2 + r6], xm6
-    vextracti128    xm4, m4, 1
-    vextracti128    xm1, m6, 1
-    movd            [r2 + 8], xm4
-    pextrd          [r2 + r3 + 8], xm4, 2
-    movd            [r2 + r3 * 2 + 8], xm1
-    pextrd          [r2 + r6 + 8], xm1, 2
+
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+
+%ifidn %3,ps
+    add             r3d, r3d
 %endif
-    lea             r2, [r2 + r3 * 4]
-    movu            xm7, [r0 + r4]                  ; m7 = row 11
-    punpckhwd       xm1, xm5, xm7
-    punpcklwd       xm5, xm7
-    vinserti128     m5, m5, xm1, 1
-    pmaddwd         m1, m5, [r5 + 1 * mmsize]
-    paddd           m0, m1
-    pmaddwd         m5, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm1, [r0]                       ; m1 = row 12
-    punpckhwd       xm4, xm7, xm1
-    punpcklwd       xm7, xm1
-    vinserti128     m7, m7, xm4, 1
-    pmaddwd         m4, m7, [r5 + 1 * mmsize]
-    paddd           m2, m4
-    pmaddwd         m7, [r5]
-%ifidn %1,sp
-    paddd           m0, m8
-    paddd           m2, m8
-    psrad           m0, 12
-    psrad           m2, 12
+
+    lea             r6, [r3 * 3]
+    lea             r11, [r1 * 4]
+
+%ifidn %3,pp
+    mova            m14, [pw_512]
 %else
-    psrad           m0, 6
-    psrad           m2, 6
+    vbroadcasti128  m14, [pw_2000]
 %endif
-    packssdw        m0, m2
 
-    movu            xm4, [r0 + r1]                  ; m4 = row 13
-    punpckhwd       xm2, xm1, xm4
-    punpcklwd       xm1, xm4
-    vinserti128     m1, m1, xm2, 1
-    pmaddwd         m2, m1, [r5 + 1 * mmsize]
-    paddd           m5, m2
-    pmaddwd         m1, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 14
-    punpckhwd       xm6, xm4, xm2
-    punpcklwd       xm4, xm2
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m7, m6
-    pmaddwd         m4, [r5]
-%ifidn %1,sp
-    paddd           m5, m8
-    paddd           m7, m8
-    psrad           m5, 12
-    psrad           m7, 12
+    mov             r9d, %2 / 16
+.loopH:
+    mov             r10d, %1 / 16
+.loopW:
+    PROCESS_LUMA_AVX2_W16_16R %3
+%ifidn %3,pp
+    add             r2, 16
 %else
-    psrad           m5, 6
-    psrad           m7, 6
+    add             r2, 32
 %endif
-    packssdw        m5, m7
-%ifidn %1,sp
-    packuswb        m0, m5
-    vextracti128    xm5, m0, 1
-    movd            [r2], xm0
-    pextrw          [r2 + 4], xm5, 0
-    pextrd          [r2 + r3], xm0, 1
-    pextrw          [r2 + r3 + 4], xm5, 2
-    pextrd          [r2 + r3 * 2], xm0, 2
-    pextrw          [r2 + r3 * 2 + 4], xm5, 4
-    pextrd          [r2 + r6], xm0, 3
-    pextrw          [r2 + r6 + 4], xm5, 6
+    add             r0, 16
+    dec             r10d
+    jnz             .loopW
+    sub             r7, r11
+    lea             r0, [r7 - 48]
+%ifidn %3,pp
+    lea             r2, [r8 + r3 * 4 - 48]
 %else
-    movq            [r2], xm0
-    movhps          [r2 + r3], xm0
-    movq            [r2 + r3 * 2], xm5
-    movhps          [r2 + r6], xm5
-    vextracti128    xm0, m0, 1
-    vextracti128    xm7, m5, 1
-    movd            [r2 + 8], xm0
-    pextrd          [r2 + r3 + 8], xm0, 2
-    movd            [r2 + r3 * 2 + 8], xm7
-    pextrd          [r2 + r6 + 8], xm7, 2
+    lea             r2, [r8 + r3 * 4 - 96]
 %endif
-    lea             r2, [r2 + r3 * 4]
+    dec             r9d
+    jnz             .loopH
+    RET
+%endif
+%endmacro
 
-    movu            xm6, [r0 + r4]                  ; m6 = row 15
-    punpckhwd       xm5, xm2, xm6
-    punpcklwd       xm2, xm6
-    vinserti128     m2, m2, xm5, 1
-    pmaddwd         m5, m2, [r5 + 1 * mmsize]
-    paddd           m1, m5
-    pmaddwd         m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm0, [r0]                       ; m0 = row 16
-    punpckhwd       xm5, xm6, xm0
-    punpcklwd       xm6, xm0
-    vinserti128     m6, m6, xm5, 1
-    pmaddwd         m5, m6, [r5 + 1 * mmsize]
-    paddd           m4, m5
-    pmaddwd         m6, [r5]
-%ifidn %1,sp
-    paddd           m1, m8
-    paddd           m4, m8
-    psrad           m1, 12
-    psrad           m4, 12
+    FILTER_VER_LUMA_AVX2_64xN 64, 32, pp
+    FILTER_VER_LUMA_AVX2_64xN 64, 48, pp
+    FILTER_VER_LUMA_AVX2_64xN 64, 64, pp
+    FILTER_VER_LUMA_AVX2_64xN 64, 32, ps
+    FILTER_VER_LUMA_AVX2_64xN 64, 48, ps
+    FILTER_VER_LUMA_AVX2_64xN 64, 64, ps
+
+%macro FILTER_VER_LUMA_AVX2_64x16 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_64x16, 4, 10, 15
+    mov             r4d, r4m
+    shl             r4d, 7
+
+%ifdef PIC
+    lea             r5, [tab_LumaCoeffVer_32]
+    add             r5, r4
 %else
-    psrad           m1, 6
-    psrad           m4, 6
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
 %endif
-    packssdw        m1, m4
-
-    movu            xm5, [r0 + r1]                  ; m5 = row 17
-    punpckhwd       xm4, xm0, xm5
-    punpcklwd       xm0, xm5
-    vinserti128     m0, m0, xm4, 1
-    pmaddwd         m0, [r5 + 1 * mmsize]
-    paddd           m2, m0
-    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
-    punpckhwd       xm0, xm5, xm4
-    punpcklwd       xm5, xm4
-    vinserti128     m5, m5, xm0, 1
-    pmaddwd         m5, [r5 + 1 * mmsize]
-    paddd           m6, m5
-%ifidn %1,sp
-    paddd           m2, m8
-    paddd           m6, m8
-    psrad           m2, 12
-    psrad           m6, 12
+
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+
+%ifidn %1,ps
+    add             r3d, r3d
+%endif
+
+    lea             r6, [r3 * 3]
+
+%ifidn %1,pp
+    mova            m14, [pw_512]
 %else
-    psrad           m2, 6
-    psrad           m6, 6
+    vbroadcasti128  m14, [pw_2000]
 %endif
-    packssdw        m2, m6
-%ifidn %1,sp
-    packuswb        m1, m2
-    vextracti128    xm2, m1, 1
-    movd            [r2], xm1
-    pextrw          [r2 + 4], xm2, 0
-    pextrd          [r2 + r3], xm1, 1
-    pextrw          [r2 + r3 + 4], xm2, 2
-    pextrd          [r2 + r3 * 2], xm1, 2
-    pextrw          [r2 + r3 * 2 + 4], xm2, 4
-    pextrd          [r2 + r6], xm1, 3
-    pextrw          [r2 + r6 + 4], xm2, 6
+
+    mov             r9d, 4
+.loopW:
+    PROCESS_LUMA_AVX2_W16_16R %1
+%ifidn %1,pp
+    add             r2, 16
 %else
-    movq            [r2], xm1
-    movhps          [r2 + r3], xm1
-    movq            [r2 + r3 * 2], xm2
-    movhps          [r2 + r6], xm2
-    vextracti128    xm4, m1, 1
-    vextracti128    xm6, m2, 1
-    movd            [r2 + 8], xm4
-    pextrd          [r2 + r3 + 8], xm4, 2
-    movd            [r2 + r3 * 2 + 8], xm6
-    pextrd          [r2 + r6 + 8], xm6, 2
+    add             r2, 32
 %endif
+    add             r0, 16
+    dec             r9d
+    jnz             .loopW
     RET
 %endif
 %endmacro
 
-    FILTER_VER_CHROMA_S_AVX2_6x16 sp
-    FILTER_VER_CHROMA_S_AVX2_6x16 ss
+    FILTER_VER_LUMA_AVX2_64x16 pp
+    FILTER_VER_LUMA_AVX2_64x16 ps
 
-;---------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vertical_ss_%1x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;---------------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_SS_W2_4R 2
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_LUMA 3
 INIT_XMM sse4
-cglobal interp_4tap_vert_ss_%1x%2, 5, 6, 5
-
-    add       r1d, r1d
+cglobal interp_8tap_vert_%3_%1x%2, 5, 7, 8 ,0-gprsize
+    lea       r5, [3 * r1]
+    sub       r0, r5
+    shl       r4d, 6
+%ifidn %3,ps
     add       r3d, r3d
-    sub       r0, r1
-    shl       r4d, 5
+%endif
+
+%ifdef PIC
+    lea       r5, [tab_LumaCoeffVer]
+    lea       r6, [r5 + r4]
+%else
+    lea       r6, [tab_LumaCoeffVer + r4]
+%endif
+
+%ifidn %3,pp
+    mova      m3, [pw_512]
+%else
+    mova      m3, [pw_2000]
+%endif
+    mov       dword [rsp], %2/4
+
+.loopH:
+    mov       r4d, (%1/8)
+.loopW:
+    PROCESS_LUMA_W8_4R
+%ifidn %3,pp
+    pmulhrsw  m7, m3
+    pmulhrsw  m6, m3
+    pmulhrsw  m5, m3
+    pmulhrsw  m4, m3
+
+    packuswb  m7, m6
+    packuswb  m5, m4
+
+    movlps    [r2], m7
+    movhps    [r2 + r3], m7
+    lea       r5, [r2 + 2 * r3]
+    movlps    [r5], m5
+    movhps    [r5 + r3], m5
+%else
+    psubw     m7, m3
+    psubw     m6, m3
+    psubw     m5, m3
+    psubw     m4, m3
+
+    movu      [r2], m7
+    movu      [r2 + r3], m6
+    lea       r5, [r2 + 2 * r3]
+    movu      [r5], m5
+    movu      [r5 + r3], m4
+%endif
 
-%ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
-    lea       r5, [r5 + r4]
+    lea       r5, [8 * r1 - 8]
+    sub       r0, r5
+%ifidn %3,pp
+    add       r2, 8
 %else
-    lea       r5, [tab_ChromaCoeffV + r4]
+    add       r2, 16
 %endif
+    dec       r4d
+    jnz       .loopW
 
-    mov       r4d, (%2/4)
-
-.loopH:
-    PROCESS_CHROMA_SP_W2_4R r5
-
-    psrad     m0, 6
-    psrad     m2, 6
-
-    packssdw  m0, m2
-
-    movd      [r2], m0
-    pextrd    [r2 + r3], m0, 1
-    lea       r2, [r2 + 2 * r3]
-    pextrd    [r2], m0, 2
-    pextrd    [r2 + r3], m0, 3
-
-    lea       r2, [r2 + 2 * r3]
+    lea       r0, [r0 + 4 * r1 - %1]
+%ifidn %3,pp
+    lea       r2, [r2 + 4 * r3 - %1]
+%else
+    lea       r2, [r2 + 4 * r3 - 2 * %1]
+%endif
 
-    dec       r4d
+    dec       dword [rsp]
     jnz       .loopH
 
     RET
 %endmacro
 
-    FILTER_VER_CHROMA_SS_W2_4R 2, 4
-    FILTER_VER_CHROMA_SS_W2_4R 2, 8
-
-    FILTER_VER_CHROMA_SS_W2_4R 2, 16
-
-;---------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ss_4x2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;---------------------------------------------------------------------------------------------------------------
-INIT_XMM sse2
-cglobal interp_4tap_vert_ss_4x2, 5, 6, 4
-
-    add        r1d, r1d
-    add        r3d, r3d
-    sub        r0, r1
-    shl        r4d, 5
+    FILTER_VER_LUMA 16, 4, pp
+    FILTER_VER_LUMA 16, 8, pp
+    FILTER_VER_LUMA 16, 12, pp
+    FILTER_VER_LUMA 16, 16, pp
+    FILTER_VER_LUMA 16, 32, pp
+    FILTER_VER_LUMA 16, 64, pp
+    FILTER_VER_LUMA 24, 32, pp
+    FILTER_VER_LUMA 32, 8, pp
+    FILTER_VER_LUMA 32, 16, pp
+    FILTER_VER_LUMA 32, 24, pp
+    FILTER_VER_LUMA 32, 32, pp
+    FILTER_VER_LUMA 32, 64, pp
+    FILTER_VER_LUMA 48, 64, pp
+    FILTER_VER_LUMA 64, 16, pp
+    FILTER_VER_LUMA 64, 32, pp
+    FILTER_VER_LUMA 64, 48, pp
+    FILTER_VER_LUMA 64, 64, pp
 
-%ifdef PIC
-    lea        r5, [tab_ChromaCoeffV]
-    lea        r5, [r5 + r4]
-%else
-    lea        r5, [tab_ChromaCoeffV + r4]
-%endif
+    FILTER_VER_LUMA 16, 4, ps
+    FILTER_VER_LUMA 16, 8, ps
+    FILTER_VER_LUMA 16, 12, ps
+    FILTER_VER_LUMA 16, 16, ps
+    FILTER_VER_LUMA 16, 32, ps
+    FILTER_VER_LUMA 16, 64, ps
+    FILTER_VER_LUMA 24, 32, ps
+    FILTER_VER_LUMA 32, 8, ps
+    FILTER_VER_LUMA 32, 16, ps
+    FILTER_VER_LUMA 32, 24, ps
+    FILTER_VER_LUMA 32, 32, ps
+    FILTER_VER_LUMA 32, 64, ps
+    FILTER_VER_LUMA 48, 64, ps
+    FILTER_VER_LUMA 64, 16, ps
+    FILTER_VER_LUMA 64, 32, ps
+    FILTER_VER_LUMA 64, 48, ps
+    FILTER_VER_LUMA 64, 64, ps
 
+%macro PROCESS_LUMA_SP_W4_4R 0
     movq       m0, [r0]
     movq       m1, [r0 + r1]
     punpcklwd  m0, m1                          ;m0=[0 1]
-    pmaddwd    m0, [r5 + 0 *16]                ;m0=[0+1]  Row1
+    pmaddwd    m0, [r6 + 0 *16]                ;m0=[0+1]  Row1
 
     lea        r0, [r0 + 2 * r1]
-    movq       m2, [r0]
-    punpcklwd  m1, m2                          ;m1=[1 2]
-    pmaddwd    m1, [r5 + 0 *16]                ;m1=[1+2]  Row2
-
-    movq       m3, [r0 + r1]
-    punpcklwd  m2, m3                          ;m4=[2 3]
-    pmaddwd    m2, [r5 + 1 * 16]
-    paddd      m0, m2                          ;m0=[0+1+2+3]  Row1 done
-    psrad      m0, 6
+    movq       m4, [r0]
+    punpcklwd  m1, m4                          ;m1=[1 2]
+    pmaddwd    m1, [r6 + 0 *16]                ;m1=[1+2]  Row2
 
-    movq       m2, [r0 + 2 * r1]
-    punpcklwd  m3, m2                          ;m5=[3 4]
-    pmaddwd    m3, [r5 + 1 * 16]
-    paddd      m1, m3                          ;m1=[1+2+3+4]  Row2 done
-    psrad      m1, 6
+    movq       m5, [r0 + r1]
+    punpcklwd  m4, m5                          ;m4=[2 3]
+    pmaddwd    m2, m4, [r6 + 0 *16]            ;m2=[2+3]  Row3
+    pmaddwd    m4, [r6 + 1 * 16]
+    paddd      m0, m4                          ;m0=[0+1+2+3]  Row1
 
-    packssdw   m0, m1
+    lea        r0, [r0 + 2 * r1]
+    movq       m4, [r0]
+    punpcklwd  m5, m4                          ;m5=[3 4]
+    pmaddwd    m3, m5, [r6 + 0 *16]            ;m3=[3+4]  Row4
+    pmaddwd    m5, [r6 + 1 * 16]
+    paddd      m1, m5                          ;m1 = [1+2+3+4]  Row2
 
-    movlps     [r2], m0
-    movhps     [r2 + r3], m0
+    movq       m5, [r0 + r1]
+    punpcklwd  m4, m5                          ;m4=[4 5]
+    pmaddwd    m6, m4, [r6 + 1 * 16]
+    paddd      m2, m6                          ;m2=[2+3+4+5]  Row3
+    pmaddwd    m4, [r6 + 2 * 16]
+    paddd      m0, m4                          ;m0=[0+1+2+3+4+5]  Row1
 
-    RET
+    lea        r0, [r0 + 2 * r1]
+    movq       m4, [r0]
+    punpcklwd  m5, m4                          ;m5=[5 6]
+    pmaddwd    m6, m5, [r6 + 1 * 16]
+    paddd      m3, m6                          ;m3=[3+4+5+6]  Row4
+    pmaddwd    m5, [r6 + 2 * 16]
+    paddd      m1, m5                          ;m1=[1+2+3+4+5+6]  Row2
+
+    movq       m5, [r0 + r1]
+    punpcklwd  m4, m5                          ;m4=[6 7]
+    pmaddwd    m6, m4, [r6 + 2 * 16]
+    paddd      m2, m6                          ;m2=[2+3+4+5+6+7]  Row3
+    pmaddwd    m4, [r6 + 3 * 16]
+    paddd      m0, m4                          ;m0=[0+1+2+3+4+5+6+7]  Row1 end
+
+    lea        r0, [r0 + 2 * r1]
+    movq       m4, [r0]
+    punpcklwd  m5, m4                          ;m5=[7 8]
+    pmaddwd    m6, m5, [r6 + 2 * 16]
+    paddd      m3, m6                          ;m3=[3+4+5+6+7+8]  Row4
+    pmaddwd    m5, [r6 + 3 * 16]
+    paddd      m1, m5                          ;m1=[1+2+3+4+5+6+7+8]  Row2 end
+
+    movq       m5, [r0 + r1]
+    punpcklwd  m4, m5                          ;m4=[8 9]
+    pmaddwd    m4, [r6 + 3 * 16]
+    paddd      m2, m4                          ;m2=[2+3+4+5+6+7+8+9]  Row3 end
+
+    movq       m4, [r0 + 2 * r1]
+    punpcklwd  m5, m4                          ;m5=[9 10]
+    pmaddwd    m5, [r6 + 3 * 16]
+    paddd      m3, m5                          ;m3=[3+4+5+6+7+8+9+10]  Row4 end
+%endmacro
 
-;-------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vertical_ss_6x8(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_SS_W6_H4 2
+;--------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_sp_%1x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_LUMA_SP 2
 INIT_XMM sse4
-cglobal interp_4tap_vert_ss_6x%2, 5, 7, 6
+cglobal interp_8tap_vert_sp_%1x%2, 5, 7, 8 ,0-gprsize
 
     add       r1d, r1d
-    add       r3d, r3d
-    sub       r0, r1
-    shl       r4d, 5
+    lea       r5, [r1 + 2 * r1]
+    sub       r0, r5
+    shl       r4d, 6
 
 %ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
+    lea       r5, [tab_LumaCoeffV]
     lea       r6, [r5 + r4]
 %else
-    lea       r6, [tab_ChromaCoeffV + r4]
+    lea       r6, [tab_LumaCoeffV + r4]
 %endif
 
-    mov       r4d, %2/4
+    mova      m7, [pd_526336]
 
+    mov       dword [rsp], %2/4
 .loopH:
-    PROCESS_CHROMA_SP_W4_4R
+    mov       r4d, (%1/4)
+.loopW:
+    PROCESS_LUMA_SP_W4_4R
+
+    paddd     m0, m7
+    paddd     m1, m7
+    paddd     m2, m7
+    paddd     m3, m7
 
-    psrad     m0, 6
-    psrad     m1, 6
-    psrad     m2, 6
-    psrad     m3, 6
+    psrad     m0, 12
+    psrad     m1, 12
+    psrad     m2, 12
+    psrad     m3, 12
 
     packssdw  m0, m1
     packssdw  m2, m3
 
-    movlps    [r2], m0
-    movhps    [r2 + r3], m0
-    lea       r5, [r2 + 2 * r3]
-    movlps    [r5], m2
-    movhps    [r5 + r3], m2
-
-    lea       r5, [4 * r1 - 2 * 4]
-    sub       r0, r5
-    add       r2, 2 * 4
-
-    PROCESS_CHROMA_SP_W2_4R r6
-
-    psrad     m0, 6
-    psrad     m2, 6
-
-    packssdw  m0, m2
+    packuswb  m0, m2
 
     movd      [r2], m0
     pextrd    [r2 + r3], m0, 1
-    lea       r2, [r2 + 2 * r3]
-    pextrd    [r2], m0, 2
-    pextrd    [r2 + r3], m0, 3
+    lea       r5, [r2 + 2 * r3]
+    pextrd    [r5], m0, 2
+    pextrd    [r5 + r3], m0, 3
 
-    sub       r0, 2 * 4
-    lea       r2, [r2 + 2 * r3 - 2 * 4]
+    lea       r5, [8 * r1 - 2 * 4]
+    sub       r0, r5
+    add       r2, 4
 
     dec       r4d
+    jnz       .loopW
+
+    lea       r0, [r0 + 4 * r1 - 2 * %1]
+    lea       r2, [r2 + 4 * r3 - %1]
+
+    dec       dword [rsp]
     jnz       .loopH
 
     RET
 %endmacro
 
-    FILTER_VER_CHROMA_SS_W6_H4 6, 8
-
-    FILTER_VER_CHROMA_SS_W6_H4 6, 16
-
+;--------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_sp_%1x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_SP 4, 4
+    FILTER_VER_LUMA_SP 8, 8
+    FILTER_VER_LUMA_SP 8, 4
+    FILTER_VER_LUMA_SP 4, 8
+    FILTER_VER_LUMA_SP 16, 16
+    FILTER_VER_LUMA_SP 16, 8
+    FILTER_VER_LUMA_SP 8, 16
+    FILTER_VER_LUMA_SP 16, 12
+    FILTER_VER_LUMA_SP 12, 16
+    FILTER_VER_LUMA_SP 16, 4
+    FILTER_VER_LUMA_SP 4, 16
+    FILTER_VER_LUMA_SP 32, 32
+    FILTER_VER_LUMA_SP 32, 16
+    FILTER_VER_LUMA_SP 16, 32
+    FILTER_VER_LUMA_SP 32, 24
+    FILTER_VER_LUMA_SP 24, 32
+    FILTER_VER_LUMA_SP 32, 8
+    FILTER_VER_LUMA_SP 8, 32
+    FILTER_VER_LUMA_SP 64, 64
+    FILTER_VER_LUMA_SP 64, 32
+    FILTER_VER_LUMA_SP 32, 64
+    FILTER_VER_LUMA_SP 64, 48
+    FILTER_VER_LUMA_SP 48, 64
+    FILTER_VER_LUMA_SP 64, 16
+    FILTER_VER_LUMA_SP 16, 64
 
-;----------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_ss_8x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;----------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_SS_W8_H2 2
-INIT_XMM sse2
-cglobal interp_4tap_vert_ss_%1x%2, 5, 6, 7
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+INIT_XMM sse4
+cglobal filterPixelToShort_4x2, 3, 4, 3
+    mov         r3d, r3m
+    add         r3d, r3d
 
-    add       r1d, r1d
-    add       r3d, r3d
-    sub       r0, r1
-    shl       r4d, 5
+    ; load constant
+    mova        m1, [pb_128]
+    mova        m2, [tab_c_64_n64]
 
-%ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
-    lea       r5, [r5 + r4]
-%else
-    lea       r5, [tab_ChromaCoeffV + r4]
-%endif
+    movd        m0, [r0]
+    pinsrd      m0, [r0 + r1], 1
+    punpcklbw   m0, m1
+    pmaddubsw   m0, m2
 
-    mov       r4d, %2/2
-.loopH:
-    PROCESS_CHROMA_SP_W8_2R
+    movq        [r2 + r3 * 0], m0
+    movhps      [r2 + r3 * 1], m0
 
-    psrad     m0, 6
-    psrad     m1, 6
-    psrad     m2, 6
-    psrad     m3, 6
+    RET
 
-    packssdw  m0, m1
-    packssdw  m2, m3
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+INIT_XMM ssse3
+cglobal filterPixelToShort_8x2, 3, 4, 3
+    mov         r3d, r3m
+    add         r3d, r3d
 
-    movu      [r2], m0
-    movu      [r2 + r3], m2
+    ; load constant
+    mova        m1, [pb_128]
+    mova        m2, [tab_c_64_n64]
 
-    lea       r2, [r2 + 2 * r3]
+    movh        m0, [r0]
+    punpcklbw   m0, m1
+    pmaddubsw   m0, m2
+    movu        [r2 + r3 * 0], m0
 
-    dec       r4d
-    jnz       .loopH
+    movh        m0, [r0 + r1]
+    punpcklbw   m0, m1
+    pmaddubsw   m0, m2
+    movu        [r2 + r3 * 1], m0
 
     RET
-%endmacro
-
-    FILTER_VER_CHROMA_SS_W8_H2 8, 2
-    FILTER_VER_CHROMA_SS_W8_H2 8, 4
-    FILTER_VER_CHROMA_SS_W8_H2 8, 6
-    FILTER_VER_CHROMA_SS_W8_H2 8, 8
-    FILTER_VER_CHROMA_SS_W8_H2 8, 16
-    FILTER_VER_CHROMA_SS_W8_H2 8, 32
-
-    FILTER_VER_CHROMA_SS_W8_H2 8, 12
-    FILTER_VER_CHROMA_SS_W8_H2 8, 64
 
 ;-----------------------------------------------------------------------------------------------------------------
 ; void interp_8tap_vert_ss_%1x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
diff --git a/source/common/x86/v4-ipfilter8.asm b/source/common/x86/v4-ipfilter8.asm
new file mode 100644
index 0000000000..bfdc0fada1
--- /dev/null
+++ b/source/common/x86/v4-ipfilter8.asm
@@ -0,0 +1,12799 @@
+;*****************************************************************************
+;* Copyright (C) 2013-2017 MulticoreWare, Inc
+;*
+;* Authors: Min Chen <chenm003@163.com>
+;*          Nabajit Deka <nabajit@multicorewareinc.com>
+;*          Praveen Kumar Tiwari <praveen@multicorewareinc.com>
+;*
+;* This program is free software; you can redistribute it and/or modify
+;* it under the terms of the GNU General Public License as published by
+;* the Free Software Foundation; either version 2 of the License, or
+;* (at your option) any later version.
+;*
+;* This program is distributed in the hope that it will be useful,
+;* but WITHOUT ANY WARRANTY; without even the implied warranty of
+;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+;* GNU General Public License for more details.
+;*
+;* You should have received a copy of the GNU General Public License
+;* along with this program; if not, write to the Free Software
+;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
+;*
+;* This program is also available under a commercial proprietary license.
+;* For more information, contact us at license @ x265.com.
+;*****************************************************************************/
+
+%include "x86inc.asm"
+%include "x86util.asm"
+
+SECTION_RODATA 32
+
+const v4_pd_526336, times 8 dd 8192*64+2048
+
+const tab_Vm,    db 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1
+                 db 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3
+
+const tab_Cm,    db 0, 2, 1, 3, 0, 2, 1, 3, 0, 2, 1, 3, 0, 2, 1, 3
+
+const interp_vert_shuf, times 2 db 0, 2, 1, 3, 2, 4, 3, 5, 4, 6, 5, 7, 6, 8, 7, 9
+                        times 2 db 4, 6, 5, 7, 6, 8, 7, 9, 8, 10, 9, 11, 10, 12, 11, 13
+
+const v4_interp4_vpp_shuf, times 2 db 0, 4, 1, 5, 2, 6, 3, 7, 8, 12, 9, 13, 10, 14, 11, 15
+
+const v4_interp4_vpp_shuf1, dd 0, 1, 1, 2, 2, 3, 3, 4
+                         dd 2, 3, 3, 4, 4, 5, 5, 6
+
+const tab_ChromaCoeff, db  0, 64,  0,  0
+                       db -2, 58, 10, -2
+                       db -4, 54, 16, -2
+                       db -6, 46, 28, -4
+                       db -4, 36, 36, -4
+                       db -4, 28, 46, -6
+                       db -2, 16, 54, -4
+                       db -2, 10, 58, -2
+
+const tabw_ChromaCoeff, dw  0, 64,  0,  0
+                        dw -2, 58, 10, -2
+                        dw -4, 54, 16, -2
+                        dw -6, 46, 28, -4
+                        dw -4, 36, 36, -4
+                        dw -4, 28, 46, -6
+                        dw -2, 16, 54, -4
+                        dw -2, 10, 58, -2
+
+const tab_ChromaCoeffV, times 4 dw 0, 64
+                        times 4 dw 0, 0
+
+                        times 4 dw -2, 58
+                        times 4 dw 10, -2
+
+                        times 4 dw -4, 54
+                        times 4 dw 16, -2
+
+                        times 4 dw -6, 46
+                        times 4 dw 28, -4
+
+                        times 4 dw -4, 36
+                        times 4 dw 36, -4
+
+                        times 4 dw -4, 28
+                        times 4 dw 46, -6
+
+                        times 4 dw -2, 16
+                        times 4 dw 54, -4
+
+                        times 4 dw -2, 10
+                        times 4 dw 58, -2
+
+const tab_ChromaCoeff_V, times 8 db 0, 64
+                         times 8 db 0,  0
+
+                         times 8 db -2, 58
+                         times 8 db 10, -2
+
+                         times 8 db -4, 54
+                         times 8 db 16, -2
+
+                         times 8 db -6, 46
+                         times 8 db 28, -4
+
+                         times 8 db -4, 36
+                         times 8 db 36, -4
+
+                         times 8 db -4, 28
+                         times 8 db 46, -6
+
+                         times 8 db -2, 16
+                         times 8 db 54, -4
+
+                         times 8 db -2, 10
+                         times 8 db 58, -2
+
+const tab_ChromaCoeffVer_32,    times 16 db 0, 64
+                                times 16 db 0, 0
+
+                                times 16 db -2, 58
+                                times 16 db 10, -2
+
+                                times 16 db -4, 54
+                                times 16 db 16, -2
+
+                                times 16 db -6, 46
+                                times 16 db 28, -4
+
+                                times 16 db -4, 36
+                                times 16 db 36, -4
+
+                                times 16 db -4, 28
+                                times 16 db 46, -6
+
+                                times 16 db -2, 16
+                                times 16 db 54, -4
+
+                                times 16 db -2, 10
+                                times 16 db 58, -2
+
+const pw_ChromaCoeffV,  times 8 dw 0, 64
+                        times 8 dw 0, 0
+
+                        times 8 dw -2, 58
+                        times 8 dw 10, -2
+
+                        times 8 dw -4, 54
+                        times 8 dw 16, -2
+
+                        times 8 dw -6, 46
+                        times 8 dw 28, -4
+
+                        times 8 dw -4, 36
+                        times 8 dw 36, -4
+
+                        times 8 dw -4, 28
+                        times 8 dw 46, -6
+
+                        times 8 dw -2, 16
+                        times 8 dw 54, -4
+
+                        times 8 dw -2, 10
+                        times 8 dw 58, -2
+
+const v4_interp8_hps_shuf,     dd 0, 4, 1, 5, 2, 6, 3, 7
+
+SECTION .text
+
+cextern pw_32
+cextern pw_512
+cextern pw_2000
+
+%macro  WORD_TO_DOUBLE 1
+%if ARCH_X86_64
+    punpcklbw   %1,     m8
+%else
+    punpcklbw   %1,     %1
+    psrlw       %1,     8
+%endif
+%endmacro
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_%1_2x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W2_H4_sse2 2
+INIT_XMM sse2
+%if ARCH_X86_64
+cglobal interp_4tap_vert_%1_2x%2, 4, 6, 9
+    pxor        m8,        m8
+%else
+cglobal interp_4tap_vert_%1_2x%2, 4, 6, 8
+%endif
+    mov         r4d,       r4m
+    sub         r0,        r1
+
+%ifidn %1,pp
+    mova        m1,        [pw_32]
+%elifidn %1,ps
+    mova        m1,        [pw_2000]
+    add         r3d,       r3d
+%endif
+
+%ifdef PIC
+    lea         r5,        [tabw_ChromaCoeff]
+    movh        m0,        [r5 + r4 * 8]
+%else
+    movh        m0,        [tabw_ChromaCoeff + r4 * 8]
+%endif
+
+    punpcklqdq  m0,        m0
+    lea         r5,        [3 * r1]
+
+%assign x 1
+%rep %2/4
+    movd        m2,        [r0]
+    movd        m3,        [r0 + r1]
+    movd        m4,        [r0 + 2 * r1]
+    movd        m5,        [r0 + r5]
+
+    punpcklbw   m2,        m3
+    punpcklbw   m6,        m4,        m5
+    punpcklwd   m2,        m6
+
+    WORD_TO_DOUBLE         m2
+    pmaddwd     m2,        m0
+
+    lea         r0,        [r0 + 4 * r1]
+    movd        m6,        [r0]
+
+    punpcklbw   m3,        m4
+    punpcklbw   m7,        m5,        m6
+    punpcklwd   m3,        m7
+
+    WORD_TO_DOUBLE         m3
+    pmaddwd     m3,        m0
+
+    packssdw    m2,        m3
+    pshuflw     m3,        m2,          q2301
+    pshufhw     m3,        m3,          q2301
+    paddw       m2,        m3
+
+    movd        m7,        [r0 + r1]
+
+    punpcklbw   m4,        m5
+    punpcklbw   m3,        m6,        m7
+    punpcklwd   m4,        m3
+
+    WORD_TO_DOUBLE         m4
+    pmaddwd     m4,        m0
+
+    movd        m3,        [r0 + 2 * r1]
+
+    punpcklbw   m5,        m6
+    punpcklbw   m7,        m3
+    punpcklwd   m5,        m7
+
+    WORD_TO_DOUBLE         m5
+    pmaddwd     m5,        m0
+
+    packssdw    m4,        m5
+    pshuflw     m5,        m4,          q2301
+    pshufhw     m5,        m5,          q2301
+    paddw       m4,        m5
+
+%ifidn %1,pp
+    psrld       m2,        16
+    psrld       m4,        16
+    packssdw    m2,        m4
+    paddw       m2,        m1
+    psraw       m2,        6
+    packuswb    m2,        m2
+
+%if ARCH_X86_64
+    movq        r4,        m2
+    mov         [r2],      r4w
+    shr         r4,        16
+    mov         [r2 + r3], r4w
+    lea         r2,        [r2 + 2 * r3]
+    shr         r4,        16
+    mov         [r2],      r4w
+    shr         r4,        16
+    mov         [r2 + r3], r4w
+%else
+    movd        r4,        m2
+    mov         [r2],      r4w
+    shr         r4,        16
+    mov         [r2 + r3], r4w
+    lea         r2,        [r2 + 2 * r3]
+    psrldq      m2,        4
+    movd        r4,        m2
+    mov         [r2],      r4w
+    shr         r4,        16
+    mov         [r2 + r3], r4w
+%endif
+%elifidn %1,ps
+    psrldq      m2,        2
+    psrldq      m4,        2
+    pshufd      m2,        m2, q3120
+    pshufd      m4,        m4, q3120
+    psubw       m4,        m1
+    psubw       m2,        m1
+
+    movd        [r2],      m2
+    psrldq      m2,        4
+    movd        [r2 + r3], m2
+    lea         r2,        [r2 + 2 * r3]
+    movd        [r2],      m4
+    psrldq      m4,        4
+    movd        [r2 + r3], m4
+%endif
+
+%if x < %2/4
+    lea         r2,        [r2 + 2 * r3]
+%endif
+%assign x x+1
+%endrep
+    RET
+
+%endmacro
+
+    FILTER_V4_W2_H4_sse2 pp, 4
+    FILTER_V4_W2_H4_sse2 pp, 8
+    FILTER_V4_W2_H4_sse2 pp, 16
+
+    FILTER_V4_W2_H4_sse2 ps, 4
+    FILTER_V4_W2_H4_sse2 ps, 8
+    FILTER_V4_W2_H4_sse2 ps, 16
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_%1_4x2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro  FILTER_V2_W4_H4_sse2 1
+INIT_XMM sse2
+cglobal interp_4tap_vert_%1_4x2, 4, 6, 8
+    mov         r4d,       r4m
+    sub         r0,        r1
+    pxor        m7,        m7
+
+%ifdef PIC
+    lea         r5,        [tabw_ChromaCoeff]
+    movh        m0,        [r5 + r4 * 8]
+%else
+    movh        m0,        [tabw_ChromaCoeff + r4 * 8]
+%endif
+
+    lea         r5,        [r0 + 2 * r1]
+    punpcklqdq  m0,        m0
+    movd        m2,        [r0]
+    movd        m3,        [r0 + r1]
+    movd        m4,        [r5]
+    movd        m5,        [r5 + r1]
+
+    punpcklbw   m2,        m3
+    punpcklbw   m1,        m4,        m5
+    punpcklwd   m2,        m1
+
+    movhlps     m6,        m2
+    punpcklbw   m2,        m7
+    punpcklbw   m6,        m7
+    pmaddwd     m2,        m0
+    pmaddwd     m6,        m0
+    packssdw    m2,        m6
+
+    movd        m1,        [r0 + 4 * r1]
+
+    punpcklbw   m3,        m4
+    punpcklbw   m5,        m1
+    punpcklwd   m3,        m5
+
+    movhlps     m6,        m3
+    punpcklbw   m3,        m7
+    punpcklbw   m6,        m7
+    pmaddwd     m3,        m0
+    pmaddwd     m6,        m0
+    packssdw    m3,        m6
+
+    pshuflw     m4,        m2,        q2301
+    pshufhw     m4,        m4,        q2301
+    paddw       m2,        m4
+    pshuflw     m5,        m3,        q2301
+    pshufhw     m5,        m5,        q2301
+    paddw       m3,        m5
+
+%ifidn %1, pp
+    psrld       m2,        16
+    psrld       m3,        16
+    packssdw    m2,        m3
+
+    paddw       m2,        [pw_32]
+    psraw       m2,        6
+    packuswb    m2,        m2
+
+    movd        [r2],      m2
+    psrldq      m2,        4
+    movd        [r2 + r3], m2
+%elifidn %1, ps
+    psrldq      m2,        2
+    psrldq      m3,        2
+    pshufd      m2,        m2, q3120
+    pshufd      m3,        m3, q3120
+    punpcklqdq  m2, m3
+
+    add         r3d,       r3d
+    psubw       m2,        [pw_2000]
+    movh        [r2],      m2
+    movhps      [r2 + r3], m2
+%endif
+    RET
+
+%endmacro
+
+    FILTER_V2_W4_H4_sse2 pp
+    FILTER_V2_W4_H4_sse2 ps
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_%1_4x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W4_H4_sse2 2
+INIT_XMM sse2
+%if ARCH_X86_64
+cglobal interp_4tap_vert_%1_4x%2, 4, 6, 9
+    pxor        m8,        m8
+%else
+cglobal interp_4tap_vert_%1_4x%2, 4, 6, 8
+%endif
+
+    mov         r4d,       r4m
+    sub         r0,        r1
+
+%ifdef PIC
+    lea         r5,        [tabw_ChromaCoeff]
+    movh        m0,        [r5 + r4 * 8]
+%else
+    movh        m0,        [tabw_ChromaCoeff + r4 * 8]
+%endif
+
+%ifidn %1,pp
+    mova        m1,        [pw_32]
+%elifidn %1,ps
+    add         r3d,       r3d
+    mova        m1,        [pw_2000]
+%endif
+
+    lea         r5,        [3 * r1]
+    lea         r4,        [3 * r3]
+    punpcklqdq  m0,        m0
+
+%assign x 1
+%rep %2/4
+    movd        m2,        [r0]
+    movd        m3,        [r0 + r1]
+    movd        m4,        [r0 + 2 * r1]
+    movd        m5,        [r0 + r5]
+
+    punpcklbw   m2,        m3
+    punpcklbw   m6,        m4,        m5
+    punpcklwd   m2,        m6
+
+    movhlps     m6,        m2
+    WORD_TO_DOUBLE         m2
+    WORD_TO_DOUBLE         m6
+    pmaddwd     m2,        m0
+    pmaddwd     m6,        m0
+    packssdw    m2,        m6
+
+    lea         r0,        [r0 + 4 * r1]
+    movd        m6,        [r0]
+
+    punpcklbw   m3,        m4
+    punpcklbw   m7,        m5,        m6
+    punpcklwd   m3,        m7
+
+    movhlps     m7,        m3
+    WORD_TO_DOUBLE         m3
+    WORD_TO_DOUBLE         m7
+    pmaddwd     m3,        m0
+    pmaddwd     m7,        m0
+    packssdw    m3,        m7
+
+    pshuflw     m7,        m2,        q2301
+    pshufhw     m7,        m7,        q2301
+    paddw       m2,        m7
+    pshuflw     m7,        m3,        q2301
+    pshufhw     m7,        m7,        q2301
+    paddw       m3,        m7
+
+%ifidn %1,pp
+    psrld       m2,        16
+    psrld       m3,        16
+    packssdw    m2,        m3
+    paddw       m2,        m1
+    psraw       m2,        6
+%elifidn %1,ps
+    psrldq      m2,        2
+    psrldq      m3,        2
+    pshufd      m2,        m2, q3120
+    pshufd      m3,        m3, q3120
+    punpcklqdq  m2,        m3
+
+    psubw       m2,        m1
+    movh        [r2],      m2
+    movhps      [r2 + r3], m2
+%endif
+
+    movd        m7,        [r0 + r1]
+
+    punpcklbw   m4,        m5
+    punpcklbw   m3,        m6,        m7
+    punpcklwd   m4,        m3
+
+    movhlps     m3,        m4
+    WORD_TO_DOUBLE         m4
+    WORD_TO_DOUBLE         m3
+    pmaddwd     m4,        m0
+    pmaddwd     m3,        m0
+    packssdw    m4,        m3
+
+    movd        m3,        [r0 + 2 * r1]
+
+    punpcklbw   m5,        m6
+    punpcklbw   m7,        m3
+    punpcklwd   m5,        m7
+
+    movhlps     m3,        m5
+    WORD_TO_DOUBLE         m5
+    WORD_TO_DOUBLE         m3
+    pmaddwd     m5,        m0
+    pmaddwd     m3,        m0
+    packssdw    m5,        m3
+
+    pshuflw     m7,        m4,        q2301
+    pshufhw     m7,        m7,        q2301
+    paddw       m4,        m7
+    pshuflw     m7,        m5,        q2301
+    pshufhw     m7,        m7,        q2301
+    paddw       m5,        m7
+
+%ifidn %1,pp
+    psrld       m4,        16
+    psrld       m5,        16
+    packssdw    m4,        m5
+
+    paddw       m4,        m1
+    psraw       m4,        6
+    packuswb    m2,        m4
+
+    movd        [r2],      m2
+    psrldq      m2,        4
+    movd        [r2 + r3], m2
+    psrldq      m2,        4
+    movd        [r2 + 2 * r3],      m2
+    psrldq      m2,        4
+    movd        [r2 + r4], m2
+%elifidn %1,ps
+    psrldq      m4,        2
+    psrldq      m5,        2
+    pshufd      m4,        m4, q3120
+    pshufd      m5,        m5, q3120
+    punpcklqdq  m4,        m5
+    psubw       m4,        m1
+    movh        [r2 + 2 * r3],      m4
+    movhps      [r2 + r4], m4
+%endif
+
+%if x < %2/4
+    lea         r2,        [r2 + 4 * r3]
+%endif
+
+%assign x x+1
+%endrep
+    RET
+
+%endmacro
+
+    FILTER_V4_W4_H4_sse2 pp, 4
+    FILTER_V4_W4_H4_sse2 pp, 8
+    FILTER_V4_W4_H4_sse2 pp, 16
+    FILTER_V4_W4_H4_sse2 pp, 32
+
+    FILTER_V4_W4_H4_sse2 ps, 4
+    FILTER_V4_W4_H4_sse2 ps, 8
+    FILTER_V4_W4_H4_sse2 ps, 16
+    FILTER_V4_W4_H4_sse2 ps, 32
+
+;-----------------------------------------------------------------------------
+;void interp_4tap_vert_%1_6x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W6_H4_sse2 2
+INIT_XMM sse2
+cglobal interp_4tap_vert_%1_6x%2, 4, 7, 10
+    mov         r4d,       r4m
+    sub         r0,        r1
+    shl         r4d,       5
+    pxor        m9,        m9
+
+%ifdef PIC
+    lea         r5,        [tab_ChromaCoeffV]
+    mova        m6,        [r5 + r4]
+    mova        m5,        [r5 + r4 + 16]
+%else
+    mova        m6,        [tab_ChromaCoeffV + r4]
+    mova        m5,        [tab_ChromaCoeffV + r4 + 16]
+%endif
+
+%ifidn %1,pp
+    mova        m4,        [pw_32]
+%elifidn %1,ps
+    mova        m4,        [pw_2000]
+    add         r3d,       r3d
+%endif
+    lea         r5,        [3 * r1]
+
+%assign x 1
+%rep %2/4
+    movq        m0,        [r0]
+    movq        m1,        [r0 + r1]
+    movq        m2,        [r0 + 2 * r1]
+    movq        m3,        [r0 + r5]
+
+    punpcklbw   m0,        m1
+    punpcklbw   m1,        m2
+    punpcklbw   m2,        m3
+
+    movhlps     m7,        m0
+    punpcklbw   m0,        m9
+    punpcklbw   m7,        m9
+    pmaddwd     m0,        m6
+    pmaddwd     m7,        m6
+    packssdw    m0,        m7
+
+    movhlps     m8,        m2
+    movq        m7,        m2
+    punpcklbw   m8,        m9
+    punpcklbw   m7,        m9
+    pmaddwd     m8,        m5
+    pmaddwd     m7,        m5
+    packssdw    m7,        m8
+
+    paddw       m0,        m7
+
+%ifidn %1,pp
+    paddw       m0,        m4
+    psraw       m0,        6
+    packuswb    m0,        m0
+
+    movd        [r2],      m0
+    pextrw      r6d,       m0,        2
+    mov         [r2 + 4],  r6w
+%elifidn %1,ps
+    psubw       m0,        m4
+    movh        [r2],      m0
+    pshufd      m0,        m0,        2
+    movd        [r2 + 8],  m0
+%endif
+
+    lea         r0,        [r0 + 4 * r1]
+
+    movq        m0,        [r0]
+    punpcklbw   m3,        m0
+
+    movhlps     m8,        m1
+    punpcklbw   m1,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m1,        m6
+    pmaddwd     m8,        m6
+    packssdw    m1,        m8
+
+    movhlps     m8,        m3
+    movq        m7,        m3
+    punpcklbw   m8,        m9
+    punpcklbw   m7,        m9
+    pmaddwd     m8,        m5
+    pmaddwd     m7,        m5
+    packssdw    m7,        m8
+
+    paddw       m1,        m7
+
+%ifidn %1,pp
+    paddw       m1,        m4
+    psraw       m1,        6
+    packuswb    m1,        m1
+
+    movd        [r2 + r3], m1
+    pextrw      r6d,       m1,        2
+    mov         [r2 + r3 + 4], r6w
+%elifidn %1,ps
+    psubw       m1,        m4
+    movh        [r2 + r3], m1
+    pshufd      m1,        m1,        2
+    movd        [r2 + r3 + 8],  m1
+%endif
+
+    movq        m1,        [r0 + r1]
+    punpcklbw   m7,        m0,        m1
+
+    movhlps     m8,        m2
+    punpcklbw   m2,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m2,        m6
+    pmaddwd     m8,        m6
+    packssdw    m2,        m8
+
+    movhlps     m8,        m7
+    punpcklbw   m7,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m7,        m5
+    pmaddwd     m8,        m5
+    packssdw    m7,        m8
+
+    paddw       m2,        m7
+    lea         r2,        [r2 + 2 * r3]
+
+%ifidn %1,pp
+    paddw       m2,        m4
+    psraw       m2,        6
+    packuswb    m2,        m2
+    movd        [r2],      m2
+    pextrw      r6d,       m2,    2
+    mov         [r2 + 4],  r6w
+%elifidn %1,ps
+    psubw       m2,        m4
+    movh        [r2],      m2
+    pshufd      m2,        m2,        2
+    movd        [r2 + 8],  m2
+%endif
+
+    movq        m2,        [r0 + 2 * r1]
+    punpcklbw   m1,        m2
+
+    movhlps     m8,        m3
+    punpcklbw   m3,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m3,        m6
+    pmaddwd     m8,        m6
+    packssdw    m3,        m8
+
+    movhlps     m8,        m1
+    punpcklbw   m1,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m1,        m5
+    pmaddwd     m8,        m5
+    packssdw    m1,        m8
+
+    paddw       m3,        m1
+
+%ifidn %1,pp
+    paddw       m3,        m4
+    psraw       m3,        6
+    packuswb    m3,        m3
+
+    movd        [r2 + r3], m3
+    pextrw      r6d,       m3,    2
+    mov         [r2 + r3 + 4], r6w
+%elifidn %1,ps
+    psubw       m3,        m4
+    movh        [r2 + r3], m3
+    pshufd      m3,        m3,        2
+    movd        [r2 + r3 + 8],  m3
+%endif
+
+%if x < %2/4
+    lea         r2,        [r2 + 2 * r3]
+%endif
+
+%assign x x+1
+%endrep
+    RET
+
+%endmacro
+
+%if ARCH_X86_64
+    FILTER_V4_W6_H4_sse2 pp, 8
+    FILTER_V4_W6_H4_sse2 pp, 16
+    FILTER_V4_W6_H4_sse2 ps, 8
+    FILTER_V4_W6_H4_sse2 ps, 16
+%endif
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_%1_8x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W8_sse2 2
+INIT_XMM sse2
+cglobal interp_4tap_vert_%1_8x%2, 4, 7, 12
+    mov         r4d,       r4m
+    sub         r0,        r1
+    shl         r4d,       5
+    pxor        m9,        m9
+
+%ifidn %1,pp
+    mova        m4,        [pw_32]
+%elifidn %1,ps
+    mova        m4,        [pw_2000]
+    add         r3d,       r3d
+%endif
+
+%ifdef PIC
+    lea         r6,        [tab_ChromaCoeffV]
+    mova        m6,        [r6 + r4]
+    mova        m5,        [r6 + r4 + 16]
+%else
+    mova        m6,        [tab_ChromaCoeffV + r4]
+    mova        m5,        [tab_ChromaCoeffV + r4 + 16]
+%endif
+
+    movq        m0,        [r0]
+    movq        m1,        [r0 + r1]
+    movq        m2,        [r0 + 2 * r1]
+    lea         r5,        [r0 + 2 * r1]
+    movq        m3,        [r5 + r1]
+
+    punpcklbw   m0,        m1
+    punpcklbw   m7,        m2,          m3
+
+    movhlps     m8,        m0
+    punpcklbw   m0,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m0,        m6
+    pmaddwd     m8,        m6
+    packssdw    m0,        m8
+
+    movhlps     m8,        m7
+    punpcklbw   m7,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m7,        m5
+    pmaddwd     m8,        m5
+    packssdw    m7,        m8
+
+    paddw       m0,        m7
+
+%ifidn %1,pp
+    paddw       m0,        m4
+    psraw       m0,        6
+%elifidn %1,ps
+    psubw       m0,        m4
+    movu        [r2],      m0
+%endif
+
+    movq        m11,        [r0 + 4 * r1]
+
+    punpcklbw   m1,        m2
+    punpcklbw   m7,        m3,        m11
+
+    movhlps     m8,        m1
+    punpcklbw   m1,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m1,        m6
+    pmaddwd     m8,        m6
+    packssdw    m1,        m8
+
+    movhlps     m8,        m7
+    punpcklbw   m7,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m7,        m5
+    pmaddwd     m8,        m5
+    packssdw    m7,        m8
+
+    paddw       m1,        m7
+
+%ifidn %1,pp
+    paddw       m1,        m4
+    psraw       m1,        6
+    packuswb    m1,        m0
+
+    movhps      [r2],      m1
+    movh        [r2 + r3], m1
+%elifidn %1,ps
+    psubw       m1,        m4
+    movu        [r2 + r3], m1
+%endif
+%if %2 == 2     ;end of 8x2
+    RET
+
+%else
+    lea         r6,        [r0 + 4 * r1]
+    movq        m1,        [r6 + r1]
+
+    punpcklbw   m2,        m3
+    punpcklbw   m7,        m11,        m1
+
+    movhlps     m8,        m2
+    punpcklbw   m2,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m2,        m6
+    pmaddwd     m8,        m6
+    packssdw    m2,        m8
+
+    movhlps     m8,        m7
+    punpcklbw   m7,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m7,        m5
+    pmaddwd     m8,        m5
+    packssdw    m7,        m8
+
+    paddw       m2,        m7
+
+%ifidn %1,pp
+    paddw       m2,        m4
+    psraw       m2,        6
+%elifidn %1,ps
+    psubw       m2,        m4
+    movu        [r2 + 2 * r3], m2
+%endif
+
+    movq        m10,        [r6 + 2 * r1]
+
+    punpcklbw   m3,        m11
+    punpcklbw   m7,        m1,        m10
+
+    movhlps     m8,        m3
+    punpcklbw   m3,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m3,        m6
+    pmaddwd     m8,        m6
+    packssdw    m3,        m8
+
+    movhlps     m8,        m7
+    punpcklbw   m7,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m7,        m5
+    pmaddwd     m8,        m5
+    packssdw    m7,        m8
+
+    paddw       m3,        m7
+    lea         r5,        [r2 + 2 * r3]
+
+%ifidn %1,pp
+    paddw       m3,        m4
+    psraw       m3,        6
+    packuswb    m3,        m2
+
+    movhps      [r2 + 2 * r3], m3
+    movh        [r5 + r3], m3
+%elifidn %1,ps
+    psubw       m3,        m4
+    movu        [r5 + r3], m3
+%endif
+%if %2 == 4     ;end of 8x4
+    RET
+
+%else
+    lea         r6,        [r6 + 2 * r1]
+    movq        m3,        [r6 + r1]
+
+    punpcklbw   m11,        m1
+    punpcklbw   m7,        m10,        m3
+
+    movhlps     m8,        m11
+    punpcklbw   m11,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m11,        m6
+    pmaddwd     m8,        m6
+    packssdw    m11,        m8
+
+    movhlps     m8,        m7
+    punpcklbw   m7,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m7,        m5
+    pmaddwd     m8,        m5
+    packssdw    m7,        m8
+
+    paddw       m11,       m7
+
+%ifidn %1, pp
+    paddw       m11,       m4
+    psraw       m11,       6
+%elifidn %1,ps
+    psubw       m11,       m4
+    movu        [r2 + 4 * r3], m11
+%endif
+
+    movq        m7,        [r0 + 8 * r1]
+
+    punpcklbw   m1,        m10
+    punpcklbw   m3,        m7
+
+    movhlps     m8,        m1
+    punpcklbw   m1,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m1,        m6
+    pmaddwd     m8,        m6
+    packssdw    m1,        m8
+
+    movhlps     m8,        m3
+    punpcklbw   m3,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m3,        m5
+    pmaddwd     m8,        m5
+    packssdw    m3,        m8
+
+    paddw       m1,        m3
+    lea         r5,        [r2 + 4 * r3]
+
+%ifidn %1,pp
+    paddw       m1,        m4
+    psraw       m1,        6
+    packuswb    m1,        m11
+
+    movhps      [r2 + 4 * r3], m1
+    movh        [r5 + r3], m1
+%elifidn %1,ps
+    psubw       m1,        m4
+    movu        [r5 + r3], m1
+%endif
+%if %2 == 6
+    RET
+
+%else
+  %error INVALID macro argument, only 2, 4 or 6!
+%endif
+%endif
+%endif
+%endmacro
+
+%if ARCH_X86_64
+    FILTER_V4_W8_sse2 pp, 2
+    FILTER_V4_W8_sse2 pp, 4
+    FILTER_V4_W8_sse2 pp, 6
+    FILTER_V4_W8_sse2 ps, 2
+    FILTER_V4_W8_sse2 ps, 4
+    FILTER_V4_W8_sse2 ps, 6
+%endif
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_%1_8x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W8_H8_H16_H32_sse2 2
+INIT_XMM sse2
+cglobal interp_4tap_vert_%1_8x%2, 4, 6, 11
+    mov         r4d,       r4m
+    sub         r0,        r1
+    shl         r4d,       5
+    pxor        m9,        m9
+
+%ifdef PIC
+    lea         r5,        [tab_ChromaCoeffV]
+    mova        m6,        [r5 + r4]
+    mova        m5,        [r5 + r4 + 16]
+%else
+    mova        m6,        [tab_ChromaCoeff + r4]
+    mova        m5,        [tab_ChromaCoeff + r4 + 16]
+%endif
+
+%ifidn %1,pp
+    mova        m4,        [pw_32]
+%elifidn %1,ps
+    mova        m4,        [pw_2000]
+    add         r3d,       r3d
+%endif
+
+    lea         r5,        [r1 * 3]
+
+%assign x 1
+%rep %2/4
+    movq        m0,        [r0]
+    movq        m1,        [r0 + r1]
+    movq        m2,        [r0 + 2 * r1]
+    movq        m3,        [r0 + r5]
+
+    punpcklbw   m0,        m1
+    punpcklbw   m1,        m2
+    punpcklbw   m2,        m3
+
+    movhlps     m7,        m0
+    punpcklbw   m0,        m9
+    punpcklbw   m7,        m9
+    pmaddwd     m0,        m6
+    pmaddwd     m7,        m6
+    packssdw    m0,        m7
+
+    movhlps     m8,        m2
+    movq        m7,        m2
+    punpcklbw   m8,        m9
+    punpcklbw   m7,        m9
+    pmaddwd     m8,        m5
+    pmaddwd     m7,        m5
+    packssdw    m7,        m8
+
+    paddw       m0,        m7
+
+%ifidn %1,pp
+    paddw       m0,        m4
+    psraw       m0,        6
+%elifidn %1,ps
+    psubw       m0,        m4
+    movu        [r2],      m0
+%endif
+
+    lea         r0,        [r0 + 4 * r1]
+    movq        m10,       [r0]
+    punpcklbw   m3,        m10
+
+    movhlps     m8,        m1
+    punpcklbw   m1,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m1,        m6
+    pmaddwd     m8,        m6
+    packssdw    m1,        m8
+
+    movhlps     m8,        m3
+    movq        m7,        m3
+    punpcklbw   m8,        m9
+    punpcklbw   m7,        m9
+    pmaddwd     m8,        m5
+    pmaddwd     m7,        m5
+    packssdw    m7,        m8
+
+    paddw       m1,        m7
+
+%ifidn %1,pp
+    paddw       m1,        m4
+    psraw       m1,        6
+
+    packuswb    m0,        m1
+    movh        [r2],      m0
+    movhps      [r2 + r3], m0
+%elifidn %1,ps
+    psubw       m1,        m4
+    movu        [r2 + r3], m1
+%endif
+
+    movq        m1,        [r0 + r1]
+    punpcklbw   m10,       m1
+
+    movhlps     m8,        m2
+    punpcklbw   m2,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m2,        m6
+    pmaddwd     m8,        m6
+    packssdw    m2,        m8
+
+    movhlps     m8,        m10
+    punpcklbw   m10,       m9
+    punpcklbw   m8,        m9
+    pmaddwd     m10,       m5
+    pmaddwd     m8,        m5
+    packssdw    m10,       m8
+
+    paddw       m2,        m10
+    lea         r2,        [r2 + 2 * r3]
+
+%ifidn %1,pp
+    paddw       m2,        m4
+    psraw       m2,        6
+%elifidn %1,ps
+    psubw       m2,        m4
+    movu        [r2],      m2
+%endif
+
+    movq        m7,        [r0 + 2 * r1]
+    punpcklbw   m1,        m7
+
+    movhlps     m8,        m3
+    punpcklbw   m3,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m3,        m6
+    pmaddwd     m8,        m6
+    packssdw    m3,        m8
+
+    movhlps     m8,        m1
+    punpcklbw   m1,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m1,        m5
+    pmaddwd     m8,        m5
+    packssdw    m1,        m8
+
+    paddw       m3,        m1
+
+%ifidn %1,pp
+    paddw       m3,        m4
+    psraw       m3,        6
+
+    packuswb    m2,        m3
+    movh        [r2],      m2
+    movhps      [r2 + r3], m2
+%elifidn %1,ps
+    psubw       m3,        m4
+    movu        [r2 + r3], m3
+%endif
+
+%if x < %2/4
+    lea         r2,        [r2 + 2 * r3]
+%endif
+%endrep
+    RET
+%endmacro
+
+%if ARCH_X86_64
+    FILTER_V4_W8_H8_H16_H32_sse2 pp,  8
+    FILTER_V4_W8_H8_H16_H32_sse2 pp, 16
+    FILTER_V4_W8_H8_H16_H32_sse2 pp, 32
+
+    FILTER_V4_W8_H8_H16_H32_sse2 pp, 12
+    FILTER_V4_W8_H8_H16_H32_sse2 pp, 64
+
+    FILTER_V4_W8_H8_H16_H32_sse2 ps,  8
+    FILTER_V4_W8_H8_H16_H32_sse2 ps, 16
+    FILTER_V4_W8_H8_H16_H32_sse2 ps, 32
+
+    FILTER_V4_W8_H8_H16_H32_sse2 ps, 12
+    FILTER_V4_W8_H8_H16_H32_sse2 ps, 64
+%endif
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_%1_12x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W12_H2_sse2 2
+INIT_XMM sse2
+cglobal interp_4tap_vert_%1_12x%2, 4, 6, 11
+    mov         r4d,       r4m
+    sub         r0,        r1
+    shl         r4d,       5
+    pxor        m9,        m9
+
+%ifidn %1,pp
+    mova        m6,        [pw_32]
+%elifidn %1,ps
+    mova        m6,        [pw_2000]
+    add         r3d,       r3d
+%endif
+
+%ifdef PIC
+    lea         r5,        [tab_ChromaCoeffV]
+    mova        m1,        [r5 + r4]
+    mova        m0,        [r5 + r4 + 16]
+%else
+    mova        m1,        [tab_ChromaCoeffV + r4]
+    mova        m0,        [tab_ChromaCoeffV + r4 + 16]
+%endif
+
+%assign x 1
+%rep %2/2
+    movu        m2,        [r0]
+    movu        m3,        [r0 + r1]
+
+    punpcklbw   m4,        m2,        m3
+    punpckhbw   m2,        m3
+
+    movhlps     m8,        m4
+    punpcklbw   m4,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m4,        m1
+    pmaddwd     m8,        m1
+    packssdw    m4,        m8
+
+    movhlps     m8,        m2
+    punpcklbw   m2,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m2,        m1
+    pmaddwd     m8,        m1
+    packssdw    m2,        m8
+
+    lea         r0,        [r0 + 2 * r1]
+    movu        m5,        [r0]
+    movu        m7,        [r0 + r1]
+
+    punpcklbw   m10,       m5,        m7
+    movhlps     m8,        m10
+    punpcklbw   m10,       m9
+    punpcklbw   m8,        m9
+    pmaddwd     m10,       m0
+    pmaddwd     m8,        m0
+    packssdw    m10,       m8
+
+    paddw       m4,        m10
+
+    punpckhbw   m10,       m5,        m7
+    movhlps     m8,        m10
+    punpcklbw   m10,       m9
+    punpcklbw   m8,        m9
+    pmaddwd     m10,       m0
+    pmaddwd     m8,        m0
+    packssdw    m10,       m8
+
+    paddw       m2,        m10
+
+%ifidn %1,pp
+    paddw       m4,        m6
+    psraw       m4,        6
+    paddw       m2,        m6
+    psraw       m2,        6
+
+    packuswb    m4,        m2
+    movh        [r2],      m4
+    psrldq      m4,        8
+    movd        [r2 + 8],  m4
+%elifidn %1,ps
+    psubw       m4,        m6
+    psubw       m2,        m6
+    movu        [r2],      m4
+    movh        [r2 + 16], m2
+%endif
+
+    punpcklbw   m4,        m3,        m5
+    punpckhbw   m3,        m5
+
+    movhlps     m8,        m4
+    punpcklbw   m4,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m4,        m1
+    pmaddwd     m8,        m1
+    packssdw    m4,        m8
+
+    movhlps     m8,        m4
+    punpcklbw   m3,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m3,        m1
+    pmaddwd     m8,        m1
+    packssdw    m3,        m8
+
+    movu        m5,        [r0 + 2 * r1]
+    punpcklbw   m2,        m7,        m5
+    punpckhbw   m7,        m5
+
+    movhlps     m8,        m2
+    punpcklbw   m2,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m2,        m0
+    pmaddwd     m8,        m0
+    packssdw    m2,        m8
+
+    movhlps     m8,        m7
+    punpcklbw   m7,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m7,        m0
+    pmaddwd     m8,        m0
+    packssdw    m7,        m8
+
+    paddw       m4,        m2
+    paddw       m3,        m7
+
+%ifidn %1,pp
+    paddw       m4,        m6
+    psraw       m4,        6
+    paddw       m3,        m6
+    psraw       m3,        6
+
+    packuswb    m4,        m3
+    movh        [r2 + r3], m4
+    psrldq      m4,        8
+    movd        [r2 + r3 + 8], m4
+%elifidn %1,ps
+    psubw       m4,        m6
+    psubw       m3,        m6
+    movu        [r2 + r3], m4
+    movh        [r2 + r3 + 16], m3
+%endif
+
+%if x < %2/2
+    lea         r2,        [r2 + 2 * r3]
+%endif
+%assign x x+1
+%endrep
+    RET
+
+%endmacro
+
+%if ARCH_X86_64
+    FILTER_V4_W12_H2_sse2 pp, 16
+    FILTER_V4_W12_H2_sse2 pp, 32
+    FILTER_V4_W12_H2_sse2 ps, 16
+    FILTER_V4_W12_H2_sse2 ps, 32
+%endif
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_%1_16x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W16_H2_sse2 2
+INIT_XMM sse2
+cglobal interp_4tap_vert_%1_16x%2, 4, 6, 11
+    mov         r4d,       r4m
+    sub         r0,        r1
+    shl         r4d,       5
+    pxor        m9,        m9
+
+%ifidn %1,pp
+    mova        m6,        [pw_32]
+%elifidn %1,ps
+    mova        m6,        [pw_2000]
+    add         r3d,       r3d
+%endif
+
+%ifdef PIC
+    lea         r5,        [tab_ChromaCoeffV]
+    mova        m1,        [r5 + r4]
+    mova        m0,        [r5 + r4 + 16]
+%else
+    mova        m1,        [tab_ChromaCoeffV + r4]
+    mova        m0,        [tab_ChromaCoeffV + r4 + 16]
+%endif
+
+%assign x 1
+%rep %2/2
+    movu        m2,        [r0]
+    movu        m3,        [r0 + r1]
+
+    punpcklbw   m4,        m2,        m3
+    punpckhbw   m2,        m3
+
+    movhlps     m8,        m4
+    punpcklbw   m4,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m4,        m1
+    pmaddwd     m8,        m1
+    packssdw    m4,        m8
+
+    movhlps     m8,        m2
+    punpcklbw   m2,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m2,        m1
+    pmaddwd     m8,        m1
+    packssdw    m2,        m8
+
+    lea         r0,        [r0 + 2 * r1]
+    movu        m5,        [r0]
+    movu        m10,       [r0 + r1]
+
+    punpckhbw   m7,        m5,        m10
+    movhlps     m8,        m7
+    punpcklbw   m7,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m7,        m0
+    pmaddwd     m8,        m0
+    packssdw    m7,        m8
+    paddw       m2,        m7
+
+    punpcklbw   m7,        m5,        m10
+    movhlps     m8,        m7
+    punpcklbw   m7,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m7,        m0
+    pmaddwd     m8,        m0
+    packssdw    m7,        m8
+    paddw       m4,        m7
+
+%ifidn %1,pp
+    paddw       m4,        m6
+    psraw       m4,        6
+    paddw       m2,        m6
+    psraw       m2,        6
+
+    packuswb    m4,        m2
+    movu        [r2],      m4
+%elifidn %1,ps
+    psubw       m4,        m6
+    psubw       m2,        m6
+    movu        [r2],      m4
+    movu        [r2 + 16], m2
+%endif
+
+    punpcklbw   m4,        m3,        m5
+    punpckhbw   m3,        m5
+
+    movhlps     m8,        m4
+    punpcklbw   m4,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m4,        m1
+    pmaddwd     m8,        m1
+    packssdw    m4,        m8
+
+    movhlps     m8,        m3
+    punpcklbw   m3,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m3,        m1
+    pmaddwd     m8,        m1
+    packssdw    m3,        m8
+
+    movu        m5,        [r0 + 2 * r1]
+
+    punpcklbw   m2,        m10,       m5
+    punpckhbw   m10,       m5
+
+    movhlps     m8,        m2
+    punpcklbw   m2,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m2,        m0
+    pmaddwd     m8,        m0
+    packssdw    m2,        m8
+
+    movhlps     m8,        m10
+    punpcklbw   m10,       m9
+    punpcklbw   m8,        m9
+    pmaddwd     m10,       m0
+    pmaddwd     m8,        m0
+    packssdw    m10,       m8
+
+    paddw       m4,        m2
+    paddw       m3,        m10
+
+%ifidn %1,pp
+    paddw       m4,        m6
+    psraw       m4,        6
+    paddw       m3,        m6
+    psraw       m3,        6
+
+    packuswb    m4,        m3
+    movu        [r2 + r3], m4
+%elifidn %1,ps
+    psubw       m4,        m6
+    psubw       m3,        m6
+    movu        [r2 + r3], m4
+    movu        [r2 + r3 + 16], m3
+%endif
+
+%if x < %2/2
+    lea         r2,        [r2 + 2 * r3]
+%endif
+%assign x x+1
+%endrep
+    RET
+
+%endmacro
+
+%if ARCH_X86_64
+    FILTER_V4_W16_H2_sse2 pp, 4
+    FILTER_V4_W16_H2_sse2 pp, 8
+    FILTER_V4_W16_H2_sse2 pp, 12
+    FILTER_V4_W16_H2_sse2 pp, 16
+    FILTER_V4_W16_H2_sse2 pp, 32
+
+    FILTER_V4_W16_H2_sse2 pp, 24
+    FILTER_V4_W16_H2_sse2 pp, 64
+
+    FILTER_V4_W16_H2_sse2 ps, 4
+    FILTER_V4_W16_H2_sse2 ps, 8
+    FILTER_V4_W16_H2_sse2 ps, 12
+    FILTER_V4_W16_H2_sse2 ps, 16
+    FILTER_V4_W16_H2_sse2 ps, 32
+
+    FILTER_V4_W16_H2_sse2 ps, 24
+    FILTER_V4_W16_H2_sse2 ps, 64
+%endif
+
+;-----------------------------------------------------------------------------
+;void interp_4tap_vert_%1_24%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W24_sse2 2
+INIT_XMM sse2
+cglobal interp_4tap_vert_%1_24x%2, 4, 6, 11
+    mov         r4d,       r4m
+    sub         r0,        r1
+    shl         r4d,       5
+    pxor        m9,        m9
+
+%ifidn %1,pp
+    mova        m6,        [pw_32]
+%elifidn %1,ps
+    mova        m6,        [pw_2000]
+    add         r3d,       r3d
+%endif
+
+%ifdef PIC
+    lea         r5,        [tab_ChromaCoeffV]
+    mova        m1,        [r5 + r4]
+    mova        m0,        [r5 + r4 + 16]
+%else
+    mova        m1,        [tab_ChromaCoeffV + r4]
+    mova        m0,        [tab_ChromaCoeffV + r4 + 16]
+%endif
+
+%assign x 1
+%rep %2/2
+    movu        m2,        [r0]
+    movu        m3,        [r0 + r1]
+
+    punpcklbw   m4,        m2,        m3
+    punpckhbw   m2,        m3
+
+    movhlps     m8,        m4
+    punpcklbw   m4,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m4,        m1
+    pmaddwd     m8,        m1
+    packssdw    m4,        m8
+
+    movhlps     m8,        m2
+    punpcklbw   m2,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m2,        m1
+    pmaddwd     m8,        m1
+    packssdw    m2,        m8
+
+    lea         r5,        [r0 + 2 * r1]
+    movu        m5,        [r5]
+    movu        m10,       [r5 + r1]
+    punpcklbw   m7,        m5,        m10
+
+    movhlps     m8,        m7
+    punpcklbw   m7,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m7,        m0
+    pmaddwd     m8,        m0
+    packssdw    m7,        m8
+    paddw       m4,        m7
+
+    punpckhbw   m7,        m5,        m10
+
+    movhlps     m8,        m7
+    punpcklbw   m7,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m7,        m0
+    pmaddwd     m8,        m0
+    packssdw    m7,        m8
+
+    paddw       m2,        m7
+
+%ifidn %1,pp
+    paddw       m4,        m6
+    psraw       m4,        6
+    paddw       m2,        m6
+    psraw       m2,        6
+
+    packuswb    m4,        m2
+    movu        [r2],      m4
+%elifidn %1,ps
+    psubw       m4,        m6
+    psubw       m2,        m6
+    movu        [r2],      m4
+    movu        [r2 + 16], m2
+%endif
+
+    punpcklbw   m4,        m3,        m5
+    punpckhbw   m3,        m5
+
+    movhlps     m8,        m4
+    punpcklbw   m4,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m4,        m1
+    pmaddwd     m8,        m1
+    packssdw    m4,        m8
+
+    movhlps     m8,        m3
+    punpcklbw   m3,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m3,        m1
+    pmaddwd     m8,        m1
+    packssdw    m3,        m8
+
+    movu        m2,        [r5 + 2 * r1]
+
+    punpcklbw   m5,        m10,        m2
+    punpckhbw   m10,       m2
+
+    movhlps     m8,        m5
+    punpcklbw   m5,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m5,        m0
+    pmaddwd     m8,        m0
+    packssdw    m5,        m8
+
+    movhlps     m8,        m10
+    punpcklbw   m10,       m9
+    punpcklbw   m8,        m9
+    pmaddwd     m10,       m0
+    pmaddwd     m8,        m0
+    packssdw    m10,       m8
+
+    paddw       m4,        m5
+    paddw       m3,        m10
+
+%ifidn %1,pp
+    paddw       m4,        m6
+    psraw       m4,        6
+    paddw       m3,        m6
+    psraw       m3,        6
+
+    packuswb    m4,        m3
+    movu        [r2 + r3], m4
+%elifidn %1,ps
+    psubw       m4,        m6
+    psubw       m3,        m6
+    movu        [r2 + r3], m4
+    movu        [r2 + r3 + 16], m3
+%endif
+
+    movq        m2,        [r0 + 16]
+    movq        m3,        [r0 + r1 + 16]
+    movq        m4,        [r5 + 16]
+    movq        m5,        [r5 + r1 + 16]
+
+    punpcklbw   m2,        m3
+    punpcklbw   m4,        m5
+
+    movhlps     m8,        m4
+    punpcklbw   m4,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m4,        m0
+    pmaddwd     m8,        m0
+    packssdw    m4,        m8
+
+    movhlps     m8,        m2
+    punpcklbw   m2,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m2,        m1
+    pmaddwd     m8,        m1
+    packssdw    m2,        m8
+
+    paddw       m2,        m4
+
+%ifidn %1,pp
+    paddw       m2,        m6
+    psraw       m2,        6
+%elifidn %1,ps
+    psubw       m2,        m6
+    movu        [r2 + 32], m2
+%endif
+
+    movq        m3,        [r0 + r1 + 16]
+    movq        m4,        [r5 + 16]
+    movq        m5,        [r5 + r1 + 16]
+    movq        m7,        [r5 + 2 * r1 + 16]
+
+    punpcklbw   m3,        m4
+    punpcklbw   m5,        m7
+
+    movhlps     m8,        m5
+    punpcklbw   m5,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m5,        m0
+    pmaddwd     m8,        m0
+    packssdw    m5,        m8
+
+    movhlps     m8,        m3
+    punpcklbw   m3,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m3,        m1
+    pmaddwd     m8,        m1
+    packssdw    m3,        m8
+
+    paddw       m3,        m5
+
+%ifidn %1,pp
+    paddw       m3,        m6
+    psraw       m3,        6
+
+    packuswb    m2,        m3
+    movh        [r2 + 16], m2
+    movhps      [r2 + r3 + 16], m2
+%elifidn %1,ps
+    psubw       m3,        m6
+    movu        [r2 + r3 + 32], m3
+%endif
+
+%if x < %2/2
+    mov         r0,        r5
+    lea         r2,        [r2 + 2 * r3]
+%endif
+%assign x x+1
+%endrep
+    RET
+
+%endmacro
+
+%if ARCH_X86_64
+    FILTER_V4_W24_sse2 pp, 32
+    FILTER_V4_W24_sse2 pp, 64
+    FILTER_V4_W24_sse2 ps, 32
+    FILTER_V4_W24_sse2 ps, 64
+%endif
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_%1_32x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W32_sse2 2
+INIT_XMM sse2
+cglobal interp_4tap_vert_%1_32x%2, 4, 6, 10
+    mov         r4d,       r4m
+    sub         r0,        r1
+    shl         r4d,       5
+    pxor        m9,        m9
+
+%ifidn %1,pp
+    mova        m6,        [pw_32]
+%elifidn %1,ps
+    mova        m6,        [pw_2000]
+    add         r3d,       r3d
+%endif
+
+%ifdef PIC
+    lea         r5,        [tab_ChromaCoeffV]
+    mova        m1,        [r5 + r4]
+    mova        m0,        [r5 + r4 + 16]
+%else
+    mova        m1,        [tab_ChromaCoeffV + r4]
+    mova        m0,        [tab_ChromaCoeffV + r4 + 16]
+%endif
+
+    mov         r4d,       %2
+
+.loop:
+    movu        m2,        [r0]
+    movu        m3,        [r0 + r1]
+
+    punpcklbw   m4,        m2,        m3
+    punpckhbw   m2,        m3
+
+    movhlps     m8,        m4
+    punpcklbw   m4,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m4,        m1
+    pmaddwd     m8,        m1
+    packssdw    m4,        m8
+
+    movhlps     m8,        m2
+    punpcklbw   m2,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m2,        m1
+    pmaddwd     m8,        m1
+    packssdw    m2,        m8
+
+    lea         r5,        [r0 + 2 * r1]
+    movu        m3,        [r5]
+    movu        m5,        [r5 + r1]
+
+    punpcklbw   m7,        m3,        m5
+    punpckhbw   m3,        m5
+
+    movhlps     m8,        m7
+    punpcklbw   m7,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m7,        m0
+    pmaddwd     m8,        m0
+    packssdw    m7,        m8
+
+    movhlps     m8,        m3
+    punpcklbw   m3,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m3,        m0
+    pmaddwd     m8,        m0
+    packssdw    m3,        m8
+
+    paddw       m4,        m7
+    paddw       m2,        m3
+
+%ifidn %1,pp
+    paddw       m4,        m6
+    psraw       m4,        6
+    paddw       m2,        m6
+    psraw       m2,        6
+
+    packuswb    m4,        m2
+    movu        [r2],      m4
+%elifidn %1,ps
+    psubw       m4,        m6
+    psubw       m2,        m6
+    movu        [r2],      m4
+    movu        [r2 + 16], m2
+%endif
+
+    movu        m2,        [r0 + 16]
+    movu        m3,        [r0 + r1 + 16]
+
+    punpcklbw   m4,        m2,        m3
+    punpckhbw   m2,        m3
+
+    movhlps     m8,        m4
+    punpcklbw   m4,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m4,        m1
+    pmaddwd     m8,        m1
+    packssdw    m4,        m8
+
+    movhlps     m8,        m2
+    punpcklbw   m2,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m2,        m1
+    pmaddwd     m8,        m1
+    packssdw    m2,        m8
+
+    movu        m3,        [r5 + 16]
+    movu        m5,        [r5 + r1 + 16]
+
+    punpcklbw   m7,        m3,        m5
+    punpckhbw   m3,        m5
+
+    movhlps     m8,        m7
+    punpcklbw   m7,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m7,        m0
+    pmaddwd     m8,        m0
+    packssdw    m7,        m8
+
+    movhlps     m8,        m3
+    punpcklbw   m3,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m3,        m0
+    pmaddwd     m8,        m0
+    packssdw    m3,        m8
+
+    paddw       m4,        m7
+    paddw       m2,        m3
+
+%ifidn %1,pp
+    paddw       m4,        m6
+    psraw       m4,        6
+    paddw       m2,        m6
+    psraw       m2,        6
+
+    packuswb    m4,        m2
+    movu        [r2 + 16], m4
+%elifidn %1,ps
+    psubw       m4,        m6
+    psubw       m2,        m6
+    movu        [r2 + 32], m4
+    movu        [r2 + 48], m2
+%endif
+
+    lea         r0,        [r0 + r1]
+    lea         r2,        [r2 + r3]
+    dec         r4
+    jnz        .loop
+    RET
+
+%endmacro
+
+%if ARCH_X86_64
+    FILTER_V4_W32_sse2 pp, 8
+    FILTER_V4_W32_sse2 pp, 16
+    FILTER_V4_W32_sse2 pp, 24
+    FILTER_V4_W32_sse2 pp, 32
+
+    FILTER_V4_W32_sse2 pp, 48
+    FILTER_V4_W32_sse2 pp, 64
+
+    FILTER_V4_W32_sse2 ps, 8
+    FILTER_V4_W32_sse2 ps, 16
+    FILTER_V4_W32_sse2 ps, 24
+    FILTER_V4_W32_sse2 ps, 32
+
+    FILTER_V4_W32_sse2 ps, 48
+    FILTER_V4_W32_sse2 ps, 64
+%endif
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_%1_%2x%3(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W16n_H2_sse2 3
+INIT_XMM sse2
+cglobal interp_4tap_vert_%1_%2x%3, 4, 7, 11
+    mov         r4d,       r4m
+    sub         r0,        r1
+    shl         r4d,       5
+    pxor        m9,        m9
+
+%ifidn %1,pp
+    mova        m7,        [pw_32]
+%elifidn %1,ps
+    mova        m7,        [pw_2000]
+    add         r3d,       r3d
+%endif
+
+%ifdef PIC
+    lea         r5,        [tab_ChromaCoeffV]
+    mova        m1,        [r5 + r4]
+    mova        m0,        [r5 + r4 + 16]
+%else
+    mova        m1,        [tab_ChromaCoeffV + r4]
+    mova        m0,        [tab_ChromaCoeffV + r4 + 16]
+%endif
+
+    mov         r4d,       %3/2
+
+.loop:
+
+    mov         r6d,       %2/16
+
+.loopW:
+
+    movu        m2,        [r0]
+    movu        m3,        [r0 + r1]
+
+    punpcklbw   m4,        m2,        m3
+    punpckhbw   m2,        m3
+
+    movhlps     m8,        m4
+    punpcklbw   m4,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m4,        m1
+    pmaddwd     m8,        m1
+    packssdw    m4,        m8
+
+    movhlps     m8,        m2
+    punpcklbw   m2,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m2,        m1
+    pmaddwd     m8,        m1
+    packssdw    m2,        m8
+
+    lea         r5,        [r0 + 2 * r1]
+    movu        m5,        [r5]
+    movu        m6,        [r5 + r1]
+
+    punpckhbw   m10,        m5,        m6
+    movhlps     m8,        m10
+    punpcklbw   m10,       m9
+    punpcklbw   m8,        m9
+    pmaddwd     m10,       m0
+    pmaddwd     m8,        m0
+    packssdw    m10,       m8
+    paddw       m2,        m10
+
+    punpcklbw   m10,        m5,        m6
+    movhlps     m8,        m10
+    punpcklbw   m10,       m9
+    punpcklbw   m8,        m9
+    pmaddwd     m10,       m0
+    pmaddwd     m8,        m0
+    packssdw    m10,       m8
+    paddw       m4,        m10
+
+%ifidn %1,pp
+    paddw       m4,        m7
+    psraw       m4,        6
+    paddw       m2,        m7
+    psraw       m2,        6
+
+    packuswb    m4,        m2
+    movu        [r2],      m4
+%elifidn %1,ps
+    psubw       m4,        m7
+    psubw       m2,        m7
+    movu        [r2],      m4
+    movu        [r2 + 16], m2
+%endif
+
+    punpcklbw   m4,        m3,        m5
+    punpckhbw   m3,        m5
+
+    movhlps     m8,        m4
+    punpcklbw   m4,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m4,        m1
+    pmaddwd     m8,        m1
+    packssdw    m4,        m8
+
+    movhlps     m8,        m3
+    punpcklbw   m3,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m3,        m1
+    pmaddwd     m8,        m1
+    packssdw    m3,        m8
+
+    movu        m5,        [r5 + 2 * r1]
+
+    punpcklbw   m2,        m6,        m5
+    punpckhbw   m6,        m5
+
+    movhlps     m8,        m2
+    punpcklbw   m2,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m2,        m0
+    pmaddwd     m8,        m0
+    packssdw    m2,        m8
+
+    movhlps     m8,        m6
+    punpcklbw   m6,        m9
+    punpcklbw   m8,        m9
+    pmaddwd     m6,        m0
+    pmaddwd     m8,        m0
+    packssdw    m6,        m8
+
+    paddw       m4,        m2
+    paddw       m3,        m6
+
+%ifidn %1,pp
+    paddw       m4,        m7
+    psraw       m4,        6
+    paddw       m3,        m7
+    psraw       m3,        6
+
+    packuswb    m4,        m3
+    movu        [r2 + r3], m4
+    add         r2,        16
+%elifidn %1,ps
+    psubw       m4,        m7
+    psubw       m3,        m7
+    movu        [r2 + r3], m4
+    movu        [r2 + r3 + 16], m3
+    add         r2,        32
+%endif
+
+    add         r0,        16
+    dec         r6d
+    jnz         .loopW
+
+    lea         r0,        [r0 + r1 * 2 - %2]
+
+%ifidn %1,pp
+    lea         r2,        [r2 + r3 * 2 - %2]
+%elifidn %1,ps
+    lea         r2,        [r2 + r3 * 2 - (%2 * 2)]
+%endif
+
+    dec         r4d
+    jnz        .loop
+    RET
+
+%endmacro
+
+%if ARCH_X86_64
+    FILTER_V4_W16n_H2_sse2 pp, 64, 64
+    FILTER_V4_W16n_H2_sse2 pp, 64, 32
+    FILTER_V4_W16n_H2_sse2 pp, 64, 48
+    FILTER_V4_W16n_H2_sse2 pp, 48, 64
+    FILTER_V4_W16n_H2_sse2 pp, 64, 16
+    FILTER_V4_W16n_H2_sse2 ps, 64, 64
+    FILTER_V4_W16n_H2_sse2 ps, 64, 32
+    FILTER_V4_W16n_H2_sse2 ps, 64, 48
+    FILTER_V4_W16n_H2_sse2 ps, 48, 64
+    FILTER_V4_W16n_H2_sse2 ps, 64, 16
+%endif
+
+;-----------------------------------------------------------------------------
+;void interp_4tap_vert_pp_2x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+INIT_XMM sse4
+cglobal interp_4tap_vert_pp_2x4, 4, 6, 8
+
+    mov         r4d,       r4m
+    sub         r0,        r1
+
+%ifdef PIC
+    lea         r5,        [tab_ChromaCoeff]
+    movd        m0,        [r5 + r4 * 4]
+%else
+    movd        m0,        [tab_ChromaCoeff + r4 * 4]
+%endif
+    lea         r4,        [r1 * 3]
+    lea         r5,        [r0 + 4 * r1]
+    pshufb      m0,        [tab_Cm]
+    mova        m1,        [pw_512]
+
+    movd        m2,        [r0]
+    movd        m3,        [r0 + r1]
+    movd        m4,        [r0 + 2 * r1]
+    movd        m5,        [r0 + r4]
+
+    punpcklbw   m2,        m3
+    punpcklbw   m6,        m4,        m5
+    punpcklbw   m2,        m6
+
+    pmaddubsw   m2,        m0
+
+    movd        m6,        [r5]
+
+    punpcklbw   m3,        m4
+    punpcklbw   m7,        m5,        m6
+    punpcklbw   m3,        m7
+
+    pmaddubsw   m3,        m0
+
+    phaddw      m2,        m3
+
+    pmulhrsw    m2,        m1
+
+    movd        m7,        [r5 + r1]
+
+    punpcklbw   m4,        m5
+    punpcklbw   m3,        m6,        m7
+    punpcklbw   m4,        m3
+
+    pmaddubsw   m4,        m0
+
+    movd        m3,        [r5 + 2 * r1]
+
+    punpcklbw   m5,        m6
+    punpcklbw   m7,        m3
+    punpcklbw   m5,        m7
+
+    pmaddubsw   m5,        m0
+
+    phaddw      m4,        m5
+
+    pmulhrsw    m4,        m1
+    packuswb    m2,        m4
+
+    pextrw      [r2],      m2, 0
+    pextrw      [r2 + r3], m2, 2
+    lea         r2,        [r2 + 2 * r3]
+    pextrw      [r2],      m2, 4
+    pextrw      [r2 + r3], m2, 6
+
+    RET
+
+%macro FILTER_VER_CHROMA_AVX2_2x4 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_2x4, 4, 6, 2
+    mov             r4d, r4m
+    shl             r4d, 5
+    sub             r0, r1
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeff_V]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeff_V + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+
+    pinsrw          xm1, [r0], 0
+    pinsrw          xm1, [r0 + r1], 1
+    pinsrw          xm1, [r0 + r1 * 2], 2
+    pinsrw          xm1, [r0 + r4], 3
+    lea             r0, [r0 + r1 * 4]
+    pinsrw          xm1, [r0], 4
+    pinsrw          xm1, [r0 + r1], 5
+    pinsrw          xm1, [r0 + r1 * 2], 6
+
+    pshufb          xm0, xm1, [interp_vert_shuf]
+    pshufb          xm1, [interp_vert_shuf + 32]
+    vinserti128     m0, m0, xm1, 1
+    pmaddubsw       m0, [r5]
+    vextracti128    xm1, m0, 1
+    paddw           xm0, xm1
+%ifidn %1,pp
+    pmulhrsw        xm0, [pw_512]
+    packuswb        xm0, xm0
+    lea             r4, [r3 * 3]
+    pextrw          [r2], xm0, 0
+    pextrw          [r2 + r3], xm0, 1
+    pextrw          [r2 + r3 * 2], xm0, 2
+    pextrw          [r2 + r4], xm0, 3
+%else
+    add             r3d, r3d
+    lea             r4, [r3 * 3]
+    psubw           xm0, [pw_2000]
+    movd            [r2], xm0
+    pextrd          [r2 + r3], xm0, 1
+    pextrd          [r2 + r3 * 2], xm0, 2
+    pextrd          [r2 + r4], xm0, 3
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_2x4 pp
+    FILTER_VER_CHROMA_AVX2_2x4 ps
+
+%macro FILTER_VER_CHROMA_AVX2_2x8 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_2x8, 4, 6, 2
+    mov             r4d, r4m
+    shl             r4d, 6
+    sub             r0, r1
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+
+    pinsrw          xm1, [r0], 0
+    pinsrw          xm1, [r0 + r1], 1
+    pinsrw          xm1, [r0 + r1 * 2], 2
+    pinsrw          xm1, [r0 + r4], 3
+    lea             r0, [r0 + r1 * 4]
+    pinsrw          xm1, [r0], 4
+    pinsrw          xm1, [r0 + r1], 5
+    pinsrw          xm1, [r0 + r1 * 2], 6
+    pinsrw          xm1, [r0 + r4], 7
+    movhlps         xm0, xm1
+    lea             r0, [r0 + r1 * 4]
+    pinsrw          xm0, [r0], 4
+    pinsrw          xm0, [r0 + r1], 5
+    pinsrw          xm0, [r0 + r1 * 2], 6
+    vinserti128     m1, m1, xm0, 1
+
+    pshufb          m0, m1, [interp_vert_shuf]
+    pshufb          m1, [interp_vert_shuf + 32]
+    pmaddubsw       m0, [r5]
+    pmaddubsw       m1, [r5 + 1 * mmsize]
+    paddw           m0, m1
+%ifidn %1,pp
+    pmulhrsw        m0, [pw_512]
+    vextracti128    xm1, m0, 1
+    packuswb        xm0, xm1
+    lea             r4, [r3 * 3]
+    pextrw          [r2], xm0, 0
+    pextrw          [r2 + r3], xm0, 1
+    pextrw          [r2 + r3 * 2], xm0, 2
+    pextrw          [r2 + r4], xm0, 3
+    lea             r2, [r2 + r3 * 4]
+    pextrw          [r2], xm0, 4
+    pextrw          [r2 + r3], xm0, 5
+    pextrw          [r2 + r3 * 2], xm0, 6
+    pextrw          [r2 + r4], xm0, 7
+%else
+    add             r3d, r3d
+    lea             r4, [r3 * 3]
+    psubw           m0, [pw_2000]
+    vextracti128    xm1, m0, 1
+    movd            [r2], xm0
+    pextrd          [r2 + r3], xm0, 1
+    pextrd          [r2 + r3 * 2], xm0, 2
+    pextrd          [r2 + r4], xm0, 3
+    lea             r2, [r2 + r3 * 4]
+    movd            [r2], xm1
+    pextrd          [r2 + r3], xm1, 1
+    pextrd          [r2 + r3 * 2], xm1, 2
+    pextrd          [r2 + r4], xm1, 3
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_2x8 pp
+    FILTER_VER_CHROMA_AVX2_2x8 ps
+
+%macro FILTER_VER_CHROMA_AVX2_2x16 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_2x16, 4, 6, 3
+    mov             r4d, r4m
+    shl             r4d, 6
+    sub             r0,  r1
+
+%ifdef PIC
+    lea             r5,  [tab_ChromaCoeffVer_32]
+    add             r5,  r4
+%else
+    lea             r5,  [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    lea             r4,  [r1 * 3]
+
+    movd            xm1, [r0]
+    pinsrw          xm1, [r0 + r1], 1
+    pinsrw          xm1, [r0 + r1 * 2], 2
+    pinsrw          xm1, [r0 + r4], 3
+    lea             r0,  [r0 + r1 * 4]
+    pinsrw          xm1, [r0], 4
+    pinsrw          xm1, [r0 + r1], 5
+    pinsrw          xm1, [r0 + r1 * 2], 6
+    pinsrw          xm1, [r0 + r4], 7
+    lea             r0,  [r0 + r1 * 4]
+    pinsrw          xm0, [r0], 4
+    pinsrw          xm0, [r0 + r1], 5
+    pinsrw          xm0, [r0 + r1 * 2], 6
+    pinsrw          xm0, [r0 + r4], 7
+    punpckhqdq      xm0, xm1, xm0
+    vinserti128     m1,  m1,  xm0,  1
+
+    pshufb          m2,  m1,  [interp_vert_shuf]
+    pshufb          m1,  [interp_vert_shuf + 32]
+    pmaddubsw       m2,  [r5]
+    pmaddubsw       m1,  [r5 + 1 * mmsize]
+    paddw           m2,  m1
+
+    lea             r0,  [r0 + r1 * 4]
+    pinsrw          xm1, [r0], 4
+    pinsrw          xm1, [r0 + r1], 5
+    pinsrw          xm1, [r0 + r1 * 2], 6
+    pinsrw          xm1, [r0 + r4], 7
+    punpckhqdq      xm1, xm0, xm1
+    lea             r0,  [r0 + r1 * 4]
+    pinsrw          xm0, [r0], 4
+    pinsrw          xm0, [r0 + r1], 5
+    pinsrw          xm0, [r0 + r1 * 2], 6
+    punpckhqdq      xm0, xm1, xm0
+    vinserti128     m1,  m1,  xm0,  1
+
+    pshufb          m0,  m1,  [interp_vert_shuf]
+    pshufb          m1,  [interp_vert_shuf + 32]
+    pmaddubsw       m0,  [r5]
+    pmaddubsw       m1,  [r5 + 1 * mmsize]
+    paddw           m0,  m1
+%ifidn %1,pp
+    mova            m1,  [pw_512]
+    pmulhrsw        m2,  m1
+    pmulhrsw        m0,  m1
+    packuswb        m2,  m0
+    lea             r4,  [r3 * 3]
+    pextrw          [r2], xm2, 0
+    pextrw          [r2 + r3], xm2, 1
+    pextrw          [r2 + r3 * 2], xm2, 2
+    pextrw          [r2 + r4], xm2, 3
+    vextracti128    xm0, m2, 1
+    lea             r2,  [r2 + r3 * 4]
+    pextrw          [r2], xm0, 0
+    pextrw          [r2 + r3], xm0, 1
+    pextrw          [r2 + r3 * 2], xm0, 2
+    pextrw          [r2 + r4], xm0, 3
+    lea             r2,  [r2 + r3 * 4]
+    pextrw          [r2], xm2, 4
+    pextrw          [r2 + r3], xm2, 5
+    pextrw          [r2 + r3 * 2], xm2, 6
+    pextrw          [r2 + r4], xm2, 7
+    lea             r2,  [r2 + r3 * 4]
+    pextrw          [r2], xm0, 4
+    pextrw          [r2 + r3], xm0, 5
+    pextrw          [r2 + r3 * 2], xm0, 6
+    pextrw          [r2 + r4], xm0, 7
+%else
+    add             r3d, r3d
+    lea             r4,  [r3 * 3]
+    vbroadcasti128  m1,  [pw_2000]
+    psubw           m2,  m1
+    psubw           m0,  m1
+    vextracti128    xm1, m2, 1
+    movd            [r2], xm2
+    pextrd          [r2 + r3], xm2, 1
+    pextrd          [r2 + r3 * 2], xm2, 2
+    pextrd          [r2 + r4], xm2, 3
+    lea             r2, [r2 + r3 * 4]
+    movd            [r2], xm1
+    pextrd          [r2 + r3], xm1, 1
+    pextrd          [r2 + r3 * 2], xm1, 2
+    pextrd          [r2 + r4], xm1, 3
+    vextracti128    xm1, m0, 1
+    lea             r2,  [r2 + r3 * 4]
+    movd            [r2], xm0
+    pextrd          [r2 + r3], xm0, 1
+    pextrd          [r2 + r3 * 2], xm0, 2
+    pextrd          [r2 + r4], xm0, 3
+    lea             r2,  [r2 + r3 * 4]
+    movd            [r2], xm1
+    pextrd          [r2 + r3], xm1, 1
+    pextrd          [r2 + r3 * 2], xm1, 2
+    pextrd          [r2 + r4], xm1, 3
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_2x16 pp
+    FILTER_VER_CHROMA_AVX2_2x16 ps
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_pp_2x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W2_H4 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_pp_2x%2, 4, 6, 8
+
+    mov         r4d,       r4m
+    sub         r0,        r1
+
+%ifdef PIC
+    lea         r5,        [tab_ChromaCoeff]
+    movd        m0,        [r5 + r4 * 4]
+%else
+    movd        m0,        [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb      m0,        [tab_Cm]
+
+    mova        m1,        [pw_512]
+
+    mov         r4d,       %2
+    lea         r5,        [3 * r1]
+
+.loop:
+    movd        m2,        [r0]
+    movd        m3,        [r0 + r1]
+    movd        m4,        [r0 + 2 * r1]
+    movd        m5,        [r0 + r5]
+
+    punpcklbw   m2,        m3
+    punpcklbw   m6,        m4,        m5
+    punpcklbw   m2,        m6
+
+    pmaddubsw   m2,        m0
+
+    lea         r0,        [r0 + 4 * r1]
+    movd        m6,        [r0]
+
+    punpcklbw   m3,        m4
+    punpcklbw   m7,        m5,        m6
+    punpcklbw   m3,        m7
+
+    pmaddubsw   m3,        m0
+
+    phaddw      m2,        m3
+
+    pmulhrsw    m2,        m1
+
+    movd        m7,        [r0 + r1]
+
+    punpcklbw   m4,        m5
+    punpcklbw   m3,        m6,        m7
+    punpcklbw   m4,        m3
+
+    pmaddubsw   m4,        m0
+
+    movd        m3,        [r0 + 2 * r1]
+
+    punpcklbw   m5,        m6
+    punpcklbw   m7,        m3
+    punpcklbw   m5,        m7
+
+    pmaddubsw   m5,        m0
+
+    phaddw      m4,        m5
+
+    pmulhrsw    m4,        m1
+    packuswb    m2,        m4
+
+    pextrw      [r2],      m2, 0
+    pextrw      [r2 + r3], m2, 2
+    lea         r2,        [r2 + 2 * r3]
+    pextrw      [r2],      m2, 4
+    pextrw      [r2 + r3], m2, 6
+
+    lea         r2,        [r2 + 2 * r3]
+
+    sub         r4,        4
+    jnz        .loop
+    RET
+%endmacro
+
+    FILTER_V4_W2_H4 2, 8
+    FILTER_V4_W2_H4 2, 16
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_pp_4x2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+INIT_XMM sse4
+cglobal interp_4tap_vert_pp_4x2, 4, 6, 6
+
+    mov         r4d,       r4m
+    sub         r0,        r1
+
+%ifdef PIC
+    lea         r5,        [tab_ChromaCoeff]
+    movd        m0,        [r5 + r4 * 4]
+%else
+    movd        m0,        [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb      m0,        [tab_Cm]
+    lea         r5,        [r0 + 2 * r1]
+
+    movd        m2,        [r0]
+    movd        m3,        [r0 + r1]
+    movd        m4,        [r5]
+    movd        m5,        [r5 + r1]
+
+    punpcklbw   m2,        m3
+    punpcklbw   m1,        m4,        m5
+    punpcklbw   m2,        m1
+
+    pmaddubsw   m2,        m0
+
+    movd        m1,        [r0 + 4 * r1]
+
+    punpcklbw   m3,        m4
+    punpcklbw   m5,        m1
+    punpcklbw   m3,        m5
+
+    pmaddubsw   m3,        m0
+
+    phaddw      m2,        m3
+
+    pmulhrsw    m2,        [pw_512]
+    packuswb    m2,        m2
+    movd        [r2],      m2
+    pextrd      [r2 + r3], m2,  1
+
+    RET
+
+%macro FILTER_VER_CHROMA_AVX2_4x2 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_4x2, 4, 6, 4
+    mov             r4d, r4m
+    shl             r4d, 5
+    sub             r0, r1
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeff_V]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeff_V + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+
+    movd            xm1, [r0]
+    movd            xm2, [r0 + r1]
+    punpcklbw       xm1, xm2
+    movd            xm3, [r0 + r1 * 2]
+    punpcklbw       xm2, xm3
+    movlhps         xm1, xm2
+    movd            xm0, [r0 + r4]
+    punpcklbw       xm3, xm0
+    movd            xm2, [r0 + r1 * 4]
+    punpcklbw       xm0, xm2
+    movlhps         xm3, xm0
+    vinserti128     m1, m1, xm3, 1                          ; m1 = row[x x x 4 3 2 1 0]
+
+    pmaddubsw       m1, [r5]
+    vextracti128    xm3, m1, 1
+    paddw           xm1, xm3
+%ifidn %1,pp
+    pmulhrsw        xm1, [pw_512]
+    packuswb        xm1, xm1
+    movd            [r2], xm1
+    pextrd          [r2 + r3], xm1, 1
+%else
+    add             r3d, r3d
+    psubw           xm1, [pw_2000]
+    movq            [r2], xm1
+    movhps          [r2 + r3], xm1
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_4x2 pp
+    FILTER_VER_CHROMA_AVX2_4x2 ps
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_pp_4x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+INIT_XMM sse4
+cglobal interp_4tap_vert_pp_4x4, 4, 6, 8
+
+    mov         r4d,       r4m
+    sub         r0,        r1
+
+%ifdef PIC
+    lea         r5,        [tab_ChromaCoeff]
+    movd        m0,        [r5 + r4 * 4]
+%else
+    movd        m0,        [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb      m0,        [tab_Cm]
+    mova        m1,        [pw_512]
+    lea         r5,        [r0 + 4 * r1]
+    lea         r4,        [r1 * 3]
+
+    movd        m2,        [r0]
+    movd        m3,        [r0 + r1]
+    movd        m4,        [r0 + 2 * r1]
+    movd        m5,        [r0 + r4]
+
+    punpcklbw   m2,        m3
+    punpcklbw   m6,        m4,        m5
+    punpcklbw   m2,        m6
+
+    pmaddubsw   m2,        m0
+
+    movd        m6,        [r5]
+
+    punpcklbw   m3,        m4
+    punpcklbw   m7,        m5,        m6
+    punpcklbw   m3,        m7
+
+    pmaddubsw   m3,        m0
+
+    phaddw      m2,        m3
+
+    pmulhrsw    m2,        m1
+
+    movd        m7,        [r5 + r1]
+
+    punpcklbw   m4,        m5
+    punpcklbw   m3,        m6,        m7
+    punpcklbw   m4,        m3
+
+    pmaddubsw   m4,        m0
+
+    movd        m3,        [r5 + 2 * r1]
+
+    punpcklbw   m5,        m6
+    punpcklbw   m7,        m3
+    punpcklbw   m5,        m7
+
+    pmaddubsw   m5,        m0
+
+    phaddw      m4,        m5
+
+    pmulhrsw    m4,        m1
+
+    packuswb    m2,        m4
+    movd        [r2],      m2
+    pextrd      [r2 + r3], m2, 1
+    lea         r2,        [r2 + 2 * r3]
+    pextrd      [r2],      m2, 2
+    pextrd      [r2 + r3], m2, 3
+    RET
+
+%macro FILTER_VER_CHROMA_AVX2_4x4 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_4x4, 4, 6, 3
+    mov             r4d, r4m
+    shl             r4d, 6
+    sub             r0, r1
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+
+    movd            xm1, [r0]
+    pinsrd          xm1, [r0 + r1], 1
+    pinsrd          xm1, [r0 + r1 * 2], 2
+    pinsrd          xm1, [r0 + r4], 3                       ; m1 = row[3 2 1 0]
+    lea             r0, [r0 + r1 * 4]
+    movd            xm2, [r0]
+    pinsrd          xm2, [r0 + r1], 1
+    pinsrd          xm2, [r0 + r1 * 2], 2                   ; m2 = row[x 6 5 4]
+    vinserti128     m1, m1, xm2, 1                          ; m1 = row[x 6 5 4 3 2 1 0]
+    mova            m2, [v4_interp4_vpp_shuf1]
+    vpermd          m0, m2, m1                              ; m0 = row[4 3 3 2 2 1 1 0]
+    mova            m2, [v4_interp4_vpp_shuf1 + mmsize]
+    vpermd          m1, m2, m1                              ; m1 = row[6 5 5 4 4 3 3 2]
+
+    mova            m2, [v4_interp4_vpp_shuf]
+    pshufb          m0, m0, m2
+    pshufb          m1, m1, m2
+    pmaddubsw       m0, [r5]
+    pmaddubsw       m1, [r5 + mmsize]
+    paddw           m0, m1                                  ; m0 = WORD ROW[3 2 1 0]
+%ifidn %1,pp
+    pmulhrsw        m0, [pw_512]
+    vextracti128    xm1, m0, 1
+    packuswb        xm0, xm1
+    lea             r5, [r3 * 3]
+    movd            [r2], xm0
+    pextrd          [r2 + r3], xm0, 1
+    pextrd          [r2 + r3 * 2], xm0, 2
+    pextrd          [r2 + r5], xm0, 3
+%else
+    add             r3d, r3d
+    psubw           m0, [pw_2000]
+    vextracti128    xm1, m0, 1
+    lea             r5, [r3 * 3]
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm1
+    movhps          [r2 + r5], xm1
+%endif
+    RET
+%endmacro
+    FILTER_VER_CHROMA_AVX2_4x4 pp
+    FILTER_VER_CHROMA_AVX2_4x4 ps
+
+%macro FILTER_VER_CHROMA_AVX2_4x8 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_4x8, 4, 6, 5
+    mov             r4d, r4m
+    shl             r4d, 6
+    sub             r0, r1
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+
+    movd            xm1, [r0]
+    pinsrd          xm1, [r0 + r1], 1
+    pinsrd          xm1, [r0 + r1 * 2], 2
+    pinsrd          xm1, [r0 + r4], 3                       ; m1 = row[3 2 1 0]
+    lea             r0, [r0 + r1 * 4]
+    movd            xm2, [r0]
+    pinsrd          xm2, [r0 + r1], 1
+    pinsrd          xm2, [r0 + r1 * 2], 2
+    pinsrd          xm2, [r0 + r4], 3                       ; m2 = row[7 6 5 4]
+    vinserti128     m1, m1, xm2, 1                          ; m1 = row[7 6 5 4 3 2 1 0]
+    lea             r0, [r0 + r1 * 4]
+    movd            xm3, [r0]
+    pinsrd          xm3, [r0 + r1], 1
+    pinsrd          xm3, [r0 + r1 * 2], 2                   ; m3 = row[x 10 9 8]
+    vinserti128     m2, m2, xm3, 1                          ; m2 = row[x 10 9 8 7 6 5 4]
+    mova            m3, [v4_interp4_vpp_shuf1]
+    vpermd          m0, m3, m1                              ; m0 = row[4 3 3 2 2 1 1 0]
+    vpermd          m4, m3, m2                              ; m4 = row[8 7 7 6 6 5 5 4]
+    mova            m3, [v4_interp4_vpp_shuf1 + mmsize]
+    vpermd          m1, m3, m1                              ; m1 = row[6 5 5 4 4 3 3 2]
+    vpermd          m2, m3, m2                              ; m2 = row[10 9 9 8 8 7 7 6]
+
+    mova            m3, [v4_interp4_vpp_shuf]
+    pshufb          m0, m0, m3
+    pshufb          m1, m1, m3
+    pshufb          m2, m2, m3
+    pshufb          m4, m4, m3
+    pmaddubsw       m0, [r5]
+    pmaddubsw       m4, [r5]
+    pmaddubsw       m1, [r5 + mmsize]
+    pmaddubsw       m2, [r5 + mmsize]
+    paddw           m0, m1                                  ; m0 = WORD ROW[3 2 1 0]
+    paddw           m4, m2                                  ; m4 = WORD ROW[7 6 5 4]
+%ifidn %1,pp
+    pmulhrsw        m0, [pw_512]
+    pmulhrsw        m4, [pw_512]
+    packuswb        m0, m4
+    vextracti128    xm1, m0, 1
+    lea             r5, [r3 * 3]
+    movd            [r2], xm0
+    pextrd          [r2 + r3], xm0, 1
+    movd            [r2 + r3 * 2], xm1
+    pextrd          [r2 + r5], xm1, 1
+    lea             r2, [r2 + r3 * 4]
+    pextrd          [r2], xm0, 2
+    pextrd          [r2 + r3], xm0, 3
+    pextrd          [r2 + r3 * 2], xm1, 2
+    pextrd          [r2 + r5], xm1, 3
+%else
+    add             r3d, r3d
+    psubw           m0, [pw_2000]
+    psubw           m4, [pw_2000]
+    vextracti128    xm1, m0, 1
+    vextracti128    xm2, m4, 1
+    lea             r5, [r3 * 3]
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm1
+    movhps          [r2 + r5], xm1
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm4
+    movhps          [r2 + r3], xm4
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r5], xm2
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_4x8 pp
+    FILTER_VER_CHROMA_AVX2_4x8 ps
+
+%macro FILTER_VER_CHROMA_AVX2_4xN 2
+%if ARCH_X86_64 == 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_4x%2, 4, 6, 12
+    mov             r4d, r4m
+    shl             r4d, 6
+    sub             r0, r1
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    mova            m10, [r5]
+    mova            m11, [r5 + mmsize]
+%ifidn %1,pp
+    mova            m9, [pw_512]
+%else
+    add             r3d, r3d
+    mova            m9, [pw_2000]
+%endif
+    lea             r5, [r3 * 3]
+%rep %2 / 16
+    movd            xm1, [r0]
+    pinsrd          xm1, [r0 + r1], 1
+    pinsrd          xm1, [r0 + r1 * 2], 2
+    pinsrd          xm1, [r0 + r4], 3                       ; m1 = row[3 2 1 0]
+    lea             r0, [r0 + r1 * 4]
+    movd            xm2, [r0]
+    pinsrd          xm2, [r0 + r1], 1
+    pinsrd          xm2, [r0 + r1 * 2], 2
+    pinsrd          xm2, [r0 + r4], 3                       ; m2 = row[7 6 5 4]
+    vinserti128     m1, m1, xm2, 1                          ; m1 = row[7 6 5 4 3 2 1 0]
+    lea             r0, [r0 + r1 * 4]
+    movd            xm3, [r0]
+    pinsrd          xm3, [r0 + r1], 1
+    pinsrd          xm3, [r0 + r1 * 2], 2
+    pinsrd          xm3, [r0 + r4], 3                       ; m3 = row[11 10 9 8]
+    vinserti128     m2, m2, xm3, 1                          ; m2 = row[11 10 9 8 7 6 5 4]
+    lea             r0, [r0 + r1 * 4]
+    movd            xm4, [r0]
+    pinsrd          xm4, [r0 + r1], 1
+    pinsrd          xm4, [r0 + r1 * 2], 2
+    pinsrd          xm4, [r0 + r4], 3                       ; m4 = row[15 14 13 12]
+    vinserti128     m3, m3, xm4, 1                          ; m3 = row[15 14 13 12 11 10 9 8]
+    lea             r0, [r0 + r1 * 4]
+    movd            xm5, [r0]
+    pinsrd          xm5, [r0 + r1], 1
+    pinsrd          xm5, [r0 + r1 * 2], 2                   ; m5 = row[x 18 17 16]
+    vinserti128     m4, m4, xm5, 1                          ; m4 = row[x 18 17 16 15 14 13 12]
+    mova            m5, [v4_interp4_vpp_shuf1]
+    vpermd          m0, m5, m1                              ; m0 = row[4 3 3 2 2 1 1 0]
+    vpermd          m6, m5, m2                              ; m6 = row[8 7 7 6 6 5 5 4]
+    vpermd          m7, m5, m3                              ; m7 = row[12 11 11 10 10 9 9 8]
+    vpermd          m8, m5, m4                              ; m8 = row[16 15 15 14 14 13 13 12]
+    mova            m5, [v4_interp4_vpp_shuf1 + mmsize]
+    vpermd          m1, m5, m1                              ; m1 = row[6 5 5 4 4 3 3 2]
+    vpermd          m2, m5, m2                              ; m2 = row[10 9 9 8 8 7 7 6]
+    vpermd          m3, m5, m3                              ; m3 = row[14 13 13 12 12 11 11 10]
+    vpermd          m4, m5, m4                              ; m4 = row[18 17 17 16 16 15 15 14]
+
+    mova            m5, [v4_interp4_vpp_shuf]
+    pshufb          m0, m0, m5
+    pshufb          m1, m1, m5
+    pshufb          m2, m2, m5
+    pshufb          m4, m4, m5
+    pshufb          m3, m3, m5
+    pshufb          m6, m6, m5
+    pshufb          m7, m7, m5
+    pshufb          m8, m8, m5
+    pmaddubsw       m0, m10
+    pmaddubsw       m6, m10
+    pmaddubsw       m7, m10
+    pmaddubsw       m8, m10
+    pmaddubsw       m1, m11
+    pmaddubsw       m2, m11
+    pmaddubsw       m3, m11
+    pmaddubsw       m4, m11
+    paddw           m0, m1                                  ; m0 = WORD ROW[3 2 1 0]
+    paddw           m6, m2                                  ; m6 = WORD ROW[7 6 5 4]
+    paddw           m7, m3                                  ; m7 = WORD ROW[11 10 9 8]
+    paddw           m8, m4                                  ; m8 = WORD ROW[15 14 13 12]
+%ifidn %1,pp
+    pmulhrsw        m0, m9
+    pmulhrsw        m6, m9
+    pmulhrsw        m7, m9
+    pmulhrsw        m8, m9
+    packuswb        m0, m6
+    packuswb        m7, m8
+    vextracti128    xm1, m0, 1
+    vextracti128    xm2, m7, 1
+    movd            [r2], xm0
+    pextrd          [r2 + r3], xm0, 1
+    movd            [r2 + r3 * 2], xm1
+    pextrd          [r2 + r5], xm1, 1
+    lea             r2, [r2 + r3 * 4]
+    pextrd          [r2], xm0, 2
+    pextrd          [r2 + r3], xm0, 3
+    pextrd          [r2 + r3 * 2], xm1, 2
+    pextrd          [r2 + r5], xm1, 3
+    lea             r2, [r2 + r3 * 4]
+    movd            [r2], xm7
+    pextrd          [r2 + r3], xm7, 1
+    movd            [r2 + r3 * 2], xm2
+    pextrd          [r2 + r5], xm2, 1
+    lea             r2, [r2 + r3 * 4]
+    pextrd          [r2], xm7, 2
+    pextrd          [r2 + r3], xm7, 3
+    pextrd          [r2 + r3 * 2], xm2, 2
+    pextrd          [r2 + r5], xm2, 3
+%else
+    psubw           m0, m9
+    psubw           m6, m9
+    psubw           m7, m9
+    psubw           m8, m9
+    vextracti128    xm1, m0, 1
+    vextracti128    xm2, m6, 1
+    vextracti128    xm3, m7, 1
+    vextracti128    xm4, m8, 1
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm1
+    movhps          [r2 + r5], xm1
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm6
+    movhps          [r2 + r3], xm6
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r5], xm2
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm7
+    movhps          [r2 + r3], xm7
+    movq            [r2 + r3 * 2], xm3
+    movhps          [r2 + r5], xm3
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm8
+    movhps          [r2 + r3], xm8
+    movq            [r2 + r3 * 2], xm4
+    movhps          [r2 + r5], xm4
+%endif
+    lea             r2, [r2 + r3 * 4]
+%endrep
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_4xN pp, 16
+    FILTER_VER_CHROMA_AVX2_4xN ps, 16
+    FILTER_VER_CHROMA_AVX2_4xN pp, 32
+    FILTER_VER_CHROMA_AVX2_4xN ps, 32
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W4_H4 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_pp_%1x%2, 4, 6, 8
+
+    mov         r4d,       r4m
+    sub         r0,        r1
+
+%ifdef PIC
+    lea         r5,        [tab_ChromaCoeff]
+    movd        m0,        [r5 + r4 * 4]
+%else
+    movd        m0,        [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb      m0,        [tab_Cm]
+
+    mova        m1,        [pw_512]
+
+    mov         r4d,       %2
+
+    lea         r5,        [3 * r1]
+
+.loop:
+    movd        m2,        [r0]
+    movd        m3,        [r0 + r1]
+    movd        m4,        [r0 + 2 * r1]
+    movd        m5,        [r0 + r5]
+
+    punpcklbw   m2,        m3
+    punpcklbw   m6,        m4,        m5
+    punpcklbw   m2,        m6
+
+    pmaddubsw   m2,        m0
+
+    lea         r0,        [r0 + 4 * r1]
+    movd        m6,        [r0]
+
+    punpcklbw   m3,        m4
+    punpcklbw   m7,        m5,        m6
+    punpcklbw   m3,        m7
+
+    pmaddubsw   m3,        m0
+
+    phaddw      m2,        m3
+
+    pmulhrsw    m2,        m1
+
+    movd        m7,        [r0 + r1]
+
+    punpcklbw   m4,        m5
+    punpcklbw   m3,        m6,        m7
+    punpcklbw   m4,        m3
+
+    pmaddubsw   m4,        m0
+
+    movd        m3,        [r0 + 2 * r1]
+
+    punpcklbw   m5,        m6
+    punpcklbw   m7,        m3
+    punpcklbw   m5,        m7
+
+    pmaddubsw   m5,        m0
+
+    phaddw      m4,        m5
+
+    pmulhrsw    m4,        m1
+    packuswb    m2,        m4
+    movd        [r2],      m2
+    pextrd      [r2 + r3], m2,  1
+    lea         r2,        [r2 + 2 * r3]
+    pextrd      [r2],      m2, 2
+    pextrd      [r2 + r3], m2, 3
+
+    lea         r2,        [r2 + 2 * r3]
+
+    sub         r4,        4
+    jnz        .loop
+    RET
+%endmacro
+
+    FILTER_V4_W4_H4 4,  8
+    FILTER_V4_W4_H4 4, 16
+
+    FILTER_V4_W4_H4 4, 32
+
+%macro FILTER_V4_W8_H2 0
+    punpcklbw   m1,        m2
+    punpcklbw   m7,        m3,        m0
+
+    pmaddubsw   m1,        m6
+    pmaddubsw   m7,        m5
+
+    paddw       m1,        m7
+
+    pmulhrsw    m1,        m4
+    packuswb    m1,        m1
+%endmacro
+
+%macro FILTER_V4_W8_H3 0
+    punpcklbw   m2,        m3
+    punpcklbw   m7,        m0,        m1
+
+    pmaddubsw   m2,        m6
+    pmaddubsw   m7,        m5
+
+    paddw       m2,        m7
+
+    pmulhrsw    m2,        m4
+    packuswb    m2,        m2
+%endmacro
+
+%macro FILTER_V4_W8_H4 0
+    punpcklbw   m3,        m0
+    punpcklbw   m7,        m1,        m2
+
+    pmaddubsw   m3,        m6
+    pmaddubsw   m7,        m5
+
+    paddw       m3,        m7
+
+    pmulhrsw    m3,        m4
+    packuswb    m3,        m3
+%endmacro
+
+%macro FILTER_V4_W8_H5 0
+    punpcklbw   m0,        m1
+    punpcklbw   m7,        m2,        m3
+
+    pmaddubsw   m0,        m6
+    pmaddubsw   m7,        m5
+
+    paddw       m0,        m7
+
+    pmulhrsw    m0,        m4
+    packuswb    m0,        m0
+%endmacro
+
+%macro FILTER_V4_W8_8x2 2
+    FILTER_V4_W8 %1, %2
+    movq        m0,        [r0 + 4 * r1]
+
+    FILTER_V4_W8_H2
+
+    movh        [r2 + r3], m1
+%endmacro
+
+%macro FILTER_V4_W8_8x4 2
+    FILTER_V4_W8_8x2 %1, %2
+;8x3
+    lea         r6,        [r0 + 4 * r1]
+    movq        m1,        [r6 + r1]
+
+    FILTER_V4_W8_H3
+
+    movh        [r2 + 2 * r3], m2
+
+;8x4
+    movq        m2,        [r6 + 2 * r1]
+
+    FILTER_V4_W8_H4
+
+    lea         r5,        [r2 + 2 * r3]
+    movh        [r5 + r3], m3
+%endmacro
+
+%macro FILTER_V4_W8_8x6 2
+    FILTER_V4_W8_8x4 %1, %2
+;8x5
+    lea         r6,        [r6 + 2 * r1]
+    movq        m3,        [r6 + r1]
+
+    FILTER_V4_W8_H5
+
+    movh        [r2 + 4 * r3], m0
+
+;8x6
+    movq        m0,        [r0 + 8 * r1]
+
+    FILTER_V4_W8_H2
+
+    lea         r5,        [r2 + 4 * r3]
+    movh        [r5 + r3], m1
+%endmacro
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W8 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_pp_%1x%2, 4, 7, 8
+
+    mov         r4d,       r4m
+
+    sub         r0,        r1
+    movq        m0,        [r0]
+    movq        m1,        [r0 + r1]
+    movq        m2,        [r0 + 2 * r1]
+    lea         r5,        [r0 + 2 * r1]
+    movq        m3,        [r5 + r1]
+
+    punpcklbw   m0,        m1
+    punpcklbw   m4,        m2,          m3
+
+%ifdef PIC
+    lea         r6,        [tab_ChromaCoeff]
+    movd        m5,        [r6 + r4 * 4]
+%else
+    movd        m5,        [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb      m6,        m5,       [tab_Vm]
+    pmaddubsw   m0,        m6
+
+    pshufb      m5,        [tab_Vm + 16]
+    pmaddubsw   m4,        m5
+
+    paddw       m0,        m4
+
+    mova        m4,        [pw_512]
+
+    pmulhrsw    m0,        m4
+    packuswb    m0,        m0
+    movh        [r2],      m0
+%endmacro
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_pp_8x2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+    FILTER_V4_W8_8x2 8, 2
+
+    RET
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_pp_8x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+    FILTER_V4_W8_8x4 8, 4
+
+    RET
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_pp_8x6(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+    FILTER_V4_W8_8x6 8, 6
+
+    RET
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ps_4x2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+INIT_XMM sse4
+cglobal interp_4tap_vert_ps_4x2, 4, 6, 6
+
+    mov         r4d, r4m
+    sub         r0, r1
+    add         r3d, r3d
+
+%ifdef PIC
+    lea         r5, [tab_ChromaCoeff]
+    movd        m0, [r5 + r4 * 4]
+%else
+    movd        m0, [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb      m0, [tab_Cm]
+
+    movd        m2, [r0]
+    movd        m3, [r0 + r1]
+    lea         r5, [r0 + 2 * r1]
+    movd        m4, [r5]
+    movd        m5, [r5 + r1]
+
+    punpcklbw   m2, m3
+    punpcklbw   m1, m4, m5
+    punpcklbw   m2, m1
+
+    pmaddubsw   m2, m0
+
+    movd        m1, [r0 + 4 * r1]
+
+    punpcklbw   m3, m4
+    punpcklbw   m5, m1
+    punpcklbw   m3, m5
+
+    pmaddubsw   m3, m0
+
+    phaddw      m2, m3
+
+    psubw       m2, [pw_2000]
+    movh        [r2], m2
+    movhps      [r2 + r3], m2
+
+    RET
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ps_4x4(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+INIT_XMM sse4
+cglobal interp_4tap_vert_ps_4x4, 4, 6, 7
+
+    mov        r4d, r4m
+    sub        r0, r1
+    add        r3d, r3d
+
+%ifdef PIC
+    lea        r5, [tab_ChromaCoeff]
+    movd       m0, [r5 + r4 * 4]
+%else
+    movd       m0, [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb     m0, [tab_Cm]
+
+    lea        r4, [r1 * 3]
+    lea        r5, [r0 + 4 * r1]
+
+    movd       m2, [r0]
+    movd       m3, [r0 + r1]
+    movd       m4, [r0 + 2 * r1]
+    movd       m5, [r0 + r4]
+
+    punpcklbw  m2, m3
+    punpcklbw  m6, m4, m5
+    punpcklbw  m2, m6
+
+    pmaddubsw  m2, m0
+
+    movd       m6, [r5]
+
+    punpcklbw  m3, m4
+    punpcklbw  m1, m5, m6
+    punpcklbw  m3, m1
+
+    pmaddubsw  m3, m0
+
+    phaddw     m2, m3
+
+    mova       m1, [pw_2000]
+
+    psubw      m2, m1
+    movh       [r2], m2
+    movhps     [r2 + r3], m2
+
+    movd       m2, [r5 + r1]
+
+    punpcklbw  m4, m5
+    punpcklbw  m3, m6, m2
+    punpcklbw  m4, m3
+
+    pmaddubsw  m4, m0
+
+    movd       m3, [r5 + 2 * r1]
+
+    punpcklbw  m5, m6
+    punpcklbw  m2, m3
+    punpcklbw  m5, m2
+
+    pmaddubsw  m5, m0
+
+    phaddw     m4, m5
+
+    psubw      m4, m1
+    lea        r2, [r2 + 2 * r3]
+    movh       [r2], m4
+    movhps     [r2 + r3], m4
+
+    RET
+
+;---------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ps_%1x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;---------------------------------------------------------------------------------------------------------------
+%macro FILTER_V_PS_W4_H4 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_ps_%1x%2, 4, 6, 8
+
+    mov        r4d, r4m
+    sub        r0, r1
+    add        r3d, r3d
+
+%ifdef PIC
+    lea        r5, [tab_ChromaCoeff]
+    movd       m0, [r5 + r4 * 4]
+%else
+    movd       m0, [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb     m0, [tab_Cm]
+
+    mova       m1, [pw_2000]
+
+    mov        r4d, %2/4
+    lea        r5, [3 * r1]
+
+.loop:
+    movd       m2, [r0]
+    movd       m3, [r0 + r1]
+    movd       m4, [r0 + 2 * r1]
+    movd       m5, [r0 + r5]
+
+    punpcklbw  m2, m3
+    punpcklbw  m6, m4, m5
+    punpcklbw  m2, m6
+
+    pmaddubsw  m2, m0
+
+    lea        r0, [r0 + 4 * r1]
+    movd       m6, [r0]
+
+    punpcklbw  m3, m4
+    punpcklbw  m7, m5, m6
+    punpcklbw  m3, m7
+
+    pmaddubsw  m3, m0
+
+    phaddw     m2, m3
+
+    psubw      m2, m1
+    movh       [r2], m2
+    movhps     [r2 + r3], m2
+
+    movd       m2, [r0 + r1]
+
+    punpcklbw  m4, m5
+    punpcklbw  m3, m6, m2
+    punpcklbw  m4, m3
+
+    pmaddubsw  m4, m0
+
+    movd       m3, [r0 + 2 * r1]
+
+    punpcklbw  m5, m6
+    punpcklbw  m2, m3
+    punpcklbw  m5, m2
+
+    pmaddubsw  m5, m0
+
+    phaddw     m4, m5
+
+    psubw      m4, m1
+    lea        r2, [r2 + 2 * r3]
+    movh       [r2], m4
+    movhps     [r2 + r3], m4
+
+    lea        r2, [r2 + 2 * r3]
+
+    dec        r4d
+    jnz        .loop
+    RET
+%endmacro
+
+    FILTER_V_PS_W4_H4 4, 8
+    FILTER_V_PS_W4_H4 4, 16
+
+    FILTER_V_PS_W4_H4 4, 32
+
+;--------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ps_8x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+%macro FILTER_V_PS_W8_H8_H16_H2 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_ps_%1x%2, 4, 6, 7
+
+    mov        r4d, r4m
+    sub        r0, r1
+    add        r3d, r3d
+
+%ifdef PIC
+    lea        r5, [tab_ChromaCoeff]
+    movd       m5, [r5 + r4 * 4]
+%else
+    movd       m5, [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb     m6, m5, [tab_Vm]
+    pshufb     m5, [tab_Vm + 16]
+    mova       m4, [pw_2000]
+
+    mov        r4d, %2/2
+    lea        r5, [3 * r1]
+
+.loopH:
+    movq       m0, [r0]
+    movq       m1, [r0 + r1]
+    movq       m2, [r0 + 2 * r1]
+    movq       m3, [r0 + r5]
+
+    punpcklbw  m0, m1
+    punpcklbw  m1, m2
+    punpcklbw  m2, m3
+
+    pmaddubsw  m0, m6
+    pmaddubsw  m2, m5
+
+    paddw      m0, m2
+
+    psubw      m0, m4
+    movu       [r2], m0
+
+    movq       m0, [r0 + 4 * r1]
+
+    punpcklbw  m3, m0
+
+    pmaddubsw  m1, m6
+    pmaddubsw  m3, m5
+
+    paddw      m1, m3
+    psubw      m1, m4
+
+    movu       [r2 + r3], m1
+
+    lea        r0, [r0 + 2 * r1]
+    lea        r2, [r2 + 2 * r3]
+
+    dec        r4d
+    jnz       .loopH
+
+    RET
+%endmacro
+
+    FILTER_V_PS_W8_H8_H16_H2 8, 2
+    FILTER_V_PS_W8_H8_H16_H2 8, 4
+    FILTER_V_PS_W8_H8_H16_H2 8, 6
+
+    FILTER_V_PS_W8_H8_H16_H2 8, 12
+    FILTER_V_PS_W8_H8_H16_H2 8, 64
+
+;--------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ps_8x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+%macro FILTER_V_PS_W8_H8_H16_H32 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_ps_%1x%2, 4, 6, 8
+
+    mov        r4d, r4m
+    sub        r0, r1
+    add        r3d, r3d
+
+%ifdef PIC
+    lea        r5, [tab_ChromaCoeff]
+    movd       m5, [r5 + r4 * 4]
+%else
+    movd       m5, [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb     m6, m5, [tab_Vm]
+    pshufb     m5, [tab_Vm + 16]
+    mova       m4, [pw_2000]
+
+    mov        r4d, %2/4
+    lea        r5, [3 * r1]
+
+.loop:
+    movq       m0, [r0]
+    movq       m1, [r0 + r1]
+    movq       m2, [r0 + 2 * r1]
+    movq       m3, [r0 + r5]
+
+    punpcklbw  m0, m1
+    punpcklbw  m1, m2
+    punpcklbw  m2, m3
+
+    pmaddubsw  m0, m6
+    pmaddubsw  m7, m2, m5
+
+    paddw      m0, m7
+
+    psubw       m0, m4
+    movu       [r2], m0
+
+    lea        r0, [r0 + 4 * r1]
+    movq       m0, [r0]
+
+    punpcklbw  m3, m0
+
+    pmaddubsw  m1, m6
+    pmaddubsw  m7, m3, m5
+
+    paddw      m1, m7
+
+    psubw      m1, m4
+    movu       [r2 + r3], m1
+
+    movq       m1, [r0 + r1]
+
+    punpcklbw  m0, m1
+
+    pmaddubsw  m2, m6
+    pmaddubsw  m0, m5
+
+    paddw      m2, m0
+
+    psubw      m2, m4
+    lea        r2, [r2 + 2 * r3]
+    movu       [r2], m2
+
+    movq       m2, [r0 + 2 * r1]
+
+    punpcklbw  m1, m2
+
+    pmaddubsw  m3, m6
+    pmaddubsw  m1, m5
+
+    paddw      m3, m1
+    psubw      m3, m4
+
+    movu       [r2 + r3], m3
+
+    lea        r2, [r2 + 2 * r3]
+
+    dec        r4d
+    jnz        .loop
+    RET
+%endmacro
+
+    FILTER_V_PS_W8_H8_H16_H32 8,  8
+    FILTER_V_PS_W8_H8_H16_H32 8, 16
+    FILTER_V_PS_W8_H8_H16_H32 8, 32
+
+;------------------------------------------------------------------------------------------------------------
+;void interp_4tap_vert_ps_6x8(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;------------------------------------------------------------------------------------------------------------
+%macro FILTER_V_PS_W6 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_ps_6x%2, 4, 6, 8
+
+    mov        r4d, r4m
+    sub        r0, r1
+    add        r3d, r3d
+
+%ifdef PIC
+    lea        r5, [tab_ChromaCoeff]
+    movd       m5, [r5 + r4 * 4]
+%else
+    movd       m5, [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb     m6, m5, [tab_Vm]
+    pshufb     m5, [tab_Vm + 16]
+    mova       m4, [pw_2000]
+    lea        r5, [3 * r1]
+    mov        r4d, %2/4
+
+.loop:
+    movq       m0, [r0]
+    movq       m1, [r0 + r1]
+    movq       m2, [r0 + 2 * r1]
+    movq       m3, [r0 + r5]
+
+    punpcklbw  m0, m1
+    punpcklbw  m1, m2
+    punpcklbw  m2, m3
+
+    pmaddubsw  m0, m6
+    pmaddubsw  m7, m2, m5
+
+    paddw      m0, m7
+    psubw      m0, m4
+
+    movh       [r2], m0
+    pshufd     m0, m0, 2
+    movd       [r2 + 8], m0
+
+    lea        r0, [r0 + 4 * r1]
+    movq       m0, [r0]
+    punpcklbw  m3, m0
+
+    pmaddubsw  m1, m6
+    pmaddubsw  m7, m3, m5
+
+    paddw      m1, m7
+    psubw      m1, m4
+
+    movh       [r2 + r3], m1
+    pshufd     m1, m1, 2
+    movd       [r2 + r3 + 8], m1
+
+    movq       m1, [r0 + r1]
+    punpcklbw  m0, m1
+
+    pmaddubsw  m2, m6
+    pmaddubsw  m0, m5
+
+    paddw      m2, m0
+    psubw      m2, m4
+
+    lea        r2,[r2 + 2 * r3]
+    movh       [r2], m2
+    pshufd     m2, m2, 2
+    movd       [r2 + 8], m2
+
+    movq       m2,[r0 + 2 * r1]
+    punpcklbw  m1, m2
+
+    pmaddubsw  m3, m6
+    pmaddubsw  m1, m5
+
+    paddw      m3, m1
+    psubw      m3, m4
+
+    movh       [r2 + r3], m3
+    pshufd     m3, m3, 2
+    movd       [r2 + r3 + 8], m3
+
+    lea        r2, [r2 + 2 * r3]
+
+    dec        r4d
+    jnz        .loop
+    RET
+%endmacro
+
+    FILTER_V_PS_W6 6, 8
+    FILTER_V_PS_W6 6, 16
+
+;---------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ps_12x16(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;---------------------------------------------------------------------------------------------------------------
+%macro FILTER_V_PS_W12 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_ps_12x%2, 4, 6, 8
+
+    mov        r4d, r4m
+    sub        r0, r1
+    add        r3d, r3d
+
+%ifdef PIC
+    lea        r5, [tab_ChromaCoeff]
+    movd       m0, [r5 + r4 * 4]
+%else
+    movd       m0, [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb     m1, m0, [tab_Vm]
+    pshufb     m0, [tab_Vm + 16]
+
+    mov        r4d, %2/2
+
+.loop:
+    movu       m2, [r0]
+    movu       m3, [r0 + r1]
+
+    punpcklbw  m4, m2, m3
+    punpckhbw  m2, m3
+
+    pmaddubsw  m4, m1
+    pmaddubsw  m2, m1
+
+    lea        r0, [r0 + 2 * r1]
+    movu       m5, [r0]
+    movu       m7, [r0 + r1]
+
+    punpcklbw  m6, m5, m7
+    pmaddubsw  m6, m0
+    paddw      m4, m6
+
+    punpckhbw  m6, m5, m7
+    pmaddubsw  m6, m0
+    paddw      m2, m6
+
+    mova       m6, [pw_2000]
+
+    psubw      m4, m6
+    psubw      m2, m6
+
+    movu       [r2], m4
+    movh       [r2 + 16], m2
+
+    punpcklbw  m4, m3, m5
+    punpckhbw  m3, m5
+
+    pmaddubsw  m4, m1
+    pmaddubsw  m3, m1
+
+    movu       m2, [r0 + 2 * r1]
+
+    punpcklbw  m5, m7, m2
+    punpckhbw  m7, m2
+
+    pmaddubsw  m5, m0
+    pmaddubsw  m7, m0
+
+    paddw      m4, m5
+    paddw      m3, m7
+
+    psubw      m4, m6
+    psubw      m3, m6
+
+    movu       [r2 + r3], m4
+    movh       [r2 + r3 + 16], m3
+
+    lea        r2, [r2 + 2 * r3]
+
+    dec        r4d
+    jnz        .loop
+    RET
+%endmacro
+
+    FILTER_V_PS_W12 12, 16
+    FILTER_V_PS_W12 12, 32
+
+;---------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ps_16x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;---------------------------------------------------------------------------------------------------------------
+%macro FILTER_V_PS_W16 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_ps_%1x%2, 4, 6, 8
+
+    mov        r4d, r4m
+    sub        r0, r1
+    add        r3d, r3d
+
+%ifdef PIC
+    lea        r5, [tab_ChromaCoeff]
+    movd       m0, [r5 + r4 * 4]
+%else
+    movd       m0, [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb     m1, m0, [tab_Vm]
+    pshufb     m0, [tab_Vm + 16]
+    mov        r4d, %2/2
+
+.loop:
+    movu       m2, [r0]
+    movu       m3, [r0 + r1]
+
+    punpcklbw  m4, m2, m3
+    punpckhbw  m2, m3
+
+    pmaddubsw  m4, m1
+    pmaddubsw  m2, m1
+
+    lea        r0, [r0 + 2 * r1]
+    movu       m5, [r0]
+    movu       m7, [r0 + r1]
+
+    punpcklbw  m6, m5, m7
+    pmaddubsw  m6, m0
+    paddw      m4, m6
+
+    punpckhbw  m6, m5, m7
+    pmaddubsw  m6, m0
+    paddw      m2, m6
+
+    mova       m6, [pw_2000]
+
+    psubw      m4, m6
+    psubw      m2, m6
+
+    movu       [r2], m4
+    movu       [r2 + 16], m2
+
+    punpcklbw  m4, m3, m5
+    punpckhbw  m3, m5
+
+    pmaddubsw  m4, m1
+    pmaddubsw  m3, m1
+
+    movu       m5, [r0 + 2 * r1]
+
+    punpcklbw  m2, m7, m5
+    punpckhbw  m7, m5
+
+    pmaddubsw  m2, m0
+    pmaddubsw  m7, m0
+
+    paddw      m4, m2
+    paddw      m3, m7
+
+    psubw      m4, m6
+    psubw      m3, m6
+
+    movu       [r2 + r3], m4
+    movu       [r2 + r3 + 16], m3
+
+    lea        r2, [r2 + 2 * r3]
+
+    dec        r4d
+    jnz        .loop
+    RET
+%endmacro
+
+    FILTER_V_PS_W16 16,  4
+    FILTER_V_PS_W16 16,  8
+    FILTER_V_PS_W16 16, 12
+    FILTER_V_PS_W16 16, 16
+    FILTER_V_PS_W16 16, 32
+
+    FILTER_V_PS_W16 16, 24
+    FILTER_V_PS_W16 16, 64
+
+;--------------------------------------------------------------------------------------------------------------
+;void interp_4tap_vert_ps_24x32(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+%macro FILTER_V4_PS_W24 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_ps_24x%2, 4, 6, 8
+
+    mov        r4d, r4m
+    sub        r0, r1
+    add        r3d, r3d
+
+%ifdef PIC
+    lea        r5, [tab_ChromaCoeff]
+    movd       m0, [r5 + r4 * 4]
+%else
+    movd       m0, [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb     m1, m0, [tab_Vm]
+    pshufb     m0, [tab_Vm + 16]
+
+    mov        r4d, %2/2
+
+.loop:
+    movu       m2, [r0]
+    movu       m3, [r0 + r1]
+
+    punpcklbw  m4, m2, m3
+    punpckhbw  m2, m3
+
+    pmaddubsw  m4, m1
+    pmaddubsw  m2, m1
+
+    lea        r5, [r0 + 2 * r1]
+
+    movu       m5, [r5]
+    movu       m7, [r5 + r1]
+
+    punpcklbw  m6, m5, m7
+    pmaddubsw  m6, m0
+    paddw      m4, m6
+
+    punpckhbw  m6, m5, m7
+    pmaddubsw  m6, m0
+    paddw      m2, m6
+
+    mova       m6, [pw_2000]
+
+    psubw      m4, m6
+    psubw      m2, m6
+
+    movu       [r2], m4
+    movu       [r2 + 16], m2
+
+    punpcklbw  m4, m3, m5
+    punpckhbw  m3, m5
+
+    pmaddubsw  m4, m1
+    pmaddubsw  m3, m1
+
+    movu       m2, [r5 + 2 * r1]
+
+    punpcklbw  m5, m7, m2
+    punpckhbw  m7, m2
+
+    pmaddubsw  m5, m0
+    pmaddubsw  m7, m0
+
+    paddw      m4, m5
+    paddw      m3, m7
+
+    psubw      m4, m6
+    psubw      m3, m6
+
+    movu       [r2 + r3], m4
+    movu       [r2 + r3 + 16], m3
+
+    movq       m2, [r0 + 16]
+    movq       m3, [r0 + r1 + 16]
+    movq       m4, [r5 + 16]
+    movq       m5, [r5 + r1 + 16]
+
+    punpcklbw  m2, m3
+    punpcklbw  m7, m4, m5
+
+    pmaddubsw  m2, m1
+    pmaddubsw  m7, m0
+
+    paddw      m2, m7
+    psubw      m2, m6
+
+    movu       [r2 + 32], m2
+
+    movq       m2, [r5 + 2 * r1 + 16]
+
+    punpcklbw  m3, m4
+    punpcklbw  m5, m2
+
+    pmaddubsw  m3, m1
+    pmaddubsw  m5, m0
+
+    paddw      m3, m5
+    psubw      m3,  m6
+
+    movu       [r2 + r3 + 32], m3
+
+    mov        r0, r5
+    lea        r2, [r2 + 2 * r3]
+
+    dec        r4d
+    jnz        .loop
+    RET
+%endmacro
+
+    FILTER_V4_PS_W24 24, 32
+
+    FILTER_V4_PS_W24 24, 64
+
+;---------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ps_32x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;---------------------------------------------------------------------------------------------------------------
+%macro FILTER_V_PS_W32 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_ps_%1x%2, 4, 6, 8
+
+    mov        r4d, r4m
+    sub        r0, r1
+    add        r3d, r3d
+
+%ifdef PIC
+    lea        r5, [tab_ChromaCoeff]
+    movd       m0, [r5 + r4 * 4]
+%else
+    movd       m0, [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb     m1, m0, [tab_Vm]
+    pshufb     m0, [tab_Vm + 16]
+
+    mova       m7, [pw_2000]
+
+    mov        r4d, %2
+
+.loop:
+    movu       m2, [r0]
+    movu       m3, [r0 + r1]
+
+    punpcklbw  m4, m2, m3
+    punpckhbw  m2, m3
+
+    pmaddubsw  m4, m1
+    pmaddubsw  m2, m1
+
+    lea        r5, [r0 + 2 * r1]
+    movu       m3, [r5]
+    movu       m5, [r5 + r1]
+
+    punpcklbw  m6, m3, m5
+    punpckhbw  m3, m5
+
+    pmaddubsw  m6, m0
+    pmaddubsw  m3, m0
+
+    paddw      m4, m6
+    paddw      m2, m3
+
+    psubw      m4, m7
+    psubw      m2, m7
+
+    movu       [r2], m4
+    movu       [r2 + 16], m2
+
+    movu       m2, [r0 + 16]
+    movu       m3, [r0 + r1 + 16]
+
+    punpcklbw  m4, m2, m3
+    punpckhbw  m2, m3
+
+    pmaddubsw  m4, m1
+    pmaddubsw  m2, m1
+
+    movu       m3, [r5 + 16]
+    movu       m5, [r5 + r1 + 16]
+
+    punpcklbw  m6, m3, m5
+    punpckhbw  m3, m5
+
+    pmaddubsw  m6, m0
+    pmaddubsw  m3, m0
+
+    paddw      m4, m6
+    paddw      m2, m3
+
+    psubw      m4, m7
+    psubw      m2, m7
+
+    movu       [r2 + 32], m4
+    movu       [r2 + 48], m2
+
+    lea        r0, [r0 + r1]
+    lea        r2, [r2 + r3]
+
+    dec        r4d
+    jnz        .loop
+    RET
+%endmacro
+
+    FILTER_V_PS_W32 32,  8
+    FILTER_V_PS_W32 32, 16
+    FILTER_V_PS_W32 32, 24
+    FILTER_V_PS_W32 32, 32
+
+    FILTER_V_PS_W32 32, 48
+    FILTER_V_PS_W32 32, 64
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W8_H8_H16_H32 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_pp_%1x%2, 4, 6, 8
+
+    mov         r4d,       r4m
+    sub         r0,        r1
+
+%ifdef PIC
+    lea         r5,        [tab_ChromaCoeff]
+    movd        m5,        [r5 + r4 * 4]
+%else
+    movd        m5,        [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb      m6,        m5,       [tab_Vm]
+    pshufb      m5,        [tab_Vm + 16]
+    mova        m4,        [pw_512]
+    lea         r5,        [r1 * 3]
+
+    mov         r4d,       %2
+
+.loop:
+    movq        m0,        [r0]
+    movq        m1,        [r0 + r1]
+    movq        m2,        [r0 + 2 * r1]
+    movq        m3,        [r0 + r5]
+
+    punpcklbw   m0,        m1
+    punpcklbw   m1,        m2
+    punpcklbw   m2,        m3
+
+    pmaddubsw   m0,        m6
+    pmaddubsw   m7,        m2, m5
+
+    paddw       m0,        m7
+
+    pmulhrsw    m0,        m4
+    packuswb    m0,        m0
+    movh        [r2],      m0
+
+    lea         r0,        [r0 + 4 * r1]
+    movq        m0,        [r0]
+
+    punpcklbw   m3,        m0
+
+    pmaddubsw   m1,        m6
+    pmaddubsw   m7,        m3, m5
+
+    paddw       m1,        m7
+
+    pmulhrsw    m1,        m4
+    packuswb    m1,        m1
+    movh        [r2 + r3], m1
+
+    movq        m1,        [r0 + r1]
+
+    punpcklbw   m0,        m1
+
+    pmaddubsw   m2,        m6
+    pmaddubsw   m0,        m5
+
+    paddw       m2,        m0
+
+    pmulhrsw    m2,        m4
+
+    movq        m7,        [r0 + 2 * r1]
+    punpcklbw   m1,        m7
+
+    pmaddubsw   m3,        m6
+    pmaddubsw   m1,        m5
+
+    paddw       m3,        m1
+
+    pmulhrsw    m3,        m4
+    packuswb    m2,        m3
+
+    lea         r2,        [r2 + 2 * r3]
+    movh        [r2],      m2
+    movhps      [r2 + r3], m2
+
+    lea         r2,        [r2 + 2 * r3]
+
+    sub         r4,         4
+    jnz        .loop
+    RET
+%endmacro
+
+    FILTER_V4_W8_H8_H16_H32 8,  8
+    FILTER_V4_W8_H8_H16_H32 8, 16
+    FILTER_V4_W8_H8_H16_H32 8, 32
+
+    FILTER_V4_W8_H8_H16_H32 8, 12
+    FILTER_V4_W8_H8_H16_H32 8, 64
+
+%macro PROCESS_CHROMA_AVX2_W8_8R 0
+    movq            xm1, [r0]                       ; m1 = row 0
+    movq            xm2, [r0 + r1]                  ; m2 = row 1
+    punpcklbw       xm1, xm2                        ; m1 = [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
+    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
+    punpcklbw       xm2, xm3                        ; m2 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10]
+    vinserti128     m5, m1, xm2, 1                  ; m5 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10] - [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
+    pmaddubsw       m5, [r5]
+    movq            xm4, [r0 + r4]                  ; m4 = row 3
+    punpcklbw       xm3, xm4                        ; m3 = [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
+    lea             r0, [r0 + r1 * 4]
+    movq            xm1, [r0]                       ; m1 = row 4
+    punpcklbw       xm4, xm1                        ; m4 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30]
+    vinserti128     m2, m3, xm4, 1                  ; m2 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30] - [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
+    pmaddubsw       m0, m2, [r5 + 1 * mmsize]
+    paddw           m5, m0
+    pmaddubsw       m2, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 5
+    punpcklbw       xm1, xm3                        ; m1 = [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
+    movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
+    punpcklbw       xm3, xm4                        ; m3 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50]
+    vinserti128     m1, m1, xm3, 1                  ; m1 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50] - [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
+    pmaddubsw       m0, m1, [r5 + 1 * mmsize]
+    paddw           m2, m0
+    pmaddubsw       m1, [r5]
+    movq            xm3, [r0 + r4]                  ; m3 = row 7
+    punpcklbw       xm4, xm3                        ; m4 = [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
+    lea             r0, [r0 + r1 * 4]
+    movq            xm0, [r0]                       ; m0 = row 8
+    punpcklbw       xm3, xm0                        ; m3 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70]
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70] - [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
+    pmaddubsw       m3, m4, [r5 + 1 * mmsize]
+    paddw           m1, m3
+    pmaddubsw       m4, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 9
+    punpcklbw       xm0, xm3                        ; m0 = [97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80]
+    movq            xm6, [r0 + r1 * 2]              ; m6 = row 10
+    punpcklbw       xm3, xm6                        ; m3 = [A7 97 A6 96 A5 95 A4 94 A3 93 A2 92 A1 91 A0 90]
+    vinserti128     m0, m0, xm3, 1                  ; m0 = [A7 97 A6 96 A5 95 A4 94 A3 93 A2 92 A1 91 A0 90] - [97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80]
+    pmaddubsw       m0, [r5 + 1 * mmsize]
+    paddw           m4, m0
+%endmacro
+
+%macro FILTER_VER_CHROMA_AVX2_8x8 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_8x8, 4, 6, 7
+    mov             r4d, r4m
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+    PROCESS_CHROMA_AVX2_W8_8R
+%ifidn %1,pp
+    lea             r4, [r3 * 3]
+    mova            m3, [pw_512]
+    pmulhrsw        m5, m3                          ; m5 = word: row 0, row 1
+    pmulhrsw        m2, m3                          ; m2 = word: row 2, row 3
+    pmulhrsw        m1, m3                          ; m1 = word: row 4, row 5
+    pmulhrsw        m4, m3                          ; m4 = word: row 6, row 7
+    packuswb        m5, m2
+    packuswb        m1, m4
+    vextracti128    xm2, m5, 1
+    vextracti128    xm4, m1, 1
+    movq            [r2], xm5
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm5
+    movhps          [r2 + r4], xm2
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm1
+    movq            [r2 + r3], xm4
+    movhps          [r2 + r3 * 2], xm1
+    movhps          [r2 + r4], xm4
+%else
+    add             r3d, r3d
+    vbroadcasti128  m3, [pw_2000]
+    lea             r4, [r3 * 3]
+    psubw           m5, m3                          ; m5 = word: row 0, row 1
+    psubw           m2, m3                          ; m2 = word: row 2, row 3
+    psubw           m1, m3                          ; m1 = word: row 4, row 5
+    psubw           m4, m3                          ; m4 = word: row 6, row 7
+    vextracti128    xm6, m5, 1
+    vextracti128    xm3, m2, 1
+    vextracti128    xm0, m1, 1
+    movu            [r2], xm5
+    movu            [r2 + r3], xm6
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r4], xm3
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm1
+    movu            [r2 + r3], xm0
+    movu            [r2 + r3 * 2], xm4
+    vextracti128    xm4, m4, 1
+    movu            [r2 + r4], xm4
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_8x8 pp
+    FILTER_VER_CHROMA_AVX2_8x8 ps
+
+%macro FILTER_VER_CHROMA_AVX2_8x6 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_8x6, 4, 6, 6
+    mov             r4d, r4m
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+
+    movq            xm1, [r0]                       ; m1 = row 0
+    movq            xm2, [r0 + r1]                  ; m2 = row 1
+    punpcklbw       xm1, xm2                        ; m1 = [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
+    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
+    punpcklbw       xm2, xm3                        ; m2 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10]
+    vinserti128     m5, m1, xm2, 1                  ; m5 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10] - [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
+    pmaddubsw       m5, [r5]
+    movq            xm4, [r0 + r4]                  ; m4 = row 3
+    punpcklbw       xm3, xm4                        ; m3 = [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
+    lea             r0, [r0 + r1 * 4]
+    movq            xm1, [r0]                       ; m1 = row 4
+    punpcklbw       xm4, xm1                        ; m4 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30]
+    vinserti128     m2, m3, xm4, 1                  ; m2 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30] - [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
+    pmaddubsw       m0, m2, [r5 + 1 * mmsize]
+    paddw           m5, m0
+    pmaddubsw       m2, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 5
+    punpcklbw       xm1, xm3                        ; m1 = [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
+    movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
+    punpcklbw       xm3, xm4                        ; m3 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50]
+    vinserti128     m1, m1, xm3, 1                  ; m1 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50] - [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
+    pmaddubsw       m0, m1, [r5 + 1 * mmsize]
+    paddw           m2, m0
+    pmaddubsw       m1, [r5]
+    movq            xm3, [r0 + r4]                  ; m3 = row 7
+    punpcklbw       xm4, xm3                        ; m4 = [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
+    lea             r0, [r0 + r1 * 4]
+    movq            xm0, [r0]                       ; m0 = row 8
+    punpcklbw       xm3, xm0                        ; m3 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70]
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70] - [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
+    pmaddubsw       m4, [r5 + 1 * mmsize]
+    paddw           m1, m4
+%ifidn %1,pp
+    lea             r4, [r3 * 3]
+    mova            m3, [pw_512]
+    pmulhrsw        m5, m3                          ; m5 = word: row 0, row 1
+    pmulhrsw        m2, m3                          ; m2 = word: row 2, row 3
+    pmulhrsw        m1, m3                          ; m1 = word: row 4, row 5
+    packuswb        m5, m2
+    packuswb        m1, m1
+    vextracti128    xm2, m5, 1
+    vextracti128    xm4, m1, 1
+    movq            [r2], xm5
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm5
+    movhps          [r2 + r4], xm2
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm1
+    movq            [r2 + r3], xm4
+%else
+    add             r3d, r3d
+    mova            m3, [pw_2000]
+    lea             r4, [r3 * 3]
+    psubw           m5, m3                          ; m5 = word: row 0, row 1
+    psubw           m2, m3                          ; m2 = word: row 2, row 3
+    psubw           m1, m3                          ; m1 = word: row 4, row 5
+    vextracti128    xm4, m5, 1
+    vextracti128    xm3, m2, 1
+    vextracti128    xm0, m1, 1
+    movu            [r2], xm5
+    movu            [r2 + r3], xm4
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r4], xm3
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm1
+    movu            [r2 + r3], xm0
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_8x6 pp
+    FILTER_VER_CHROMA_AVX2_8x6 ps
+
+%macro PROCESS_CHROMA_AVX2_W8_16R 1
+    movq            xm1, [r0]                       ; m1 = row 0
+    movq            xm2, [r0 + r1]                  ; m2 = row 1
+    punpcklbw       xm1, xm2
+    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
+    punpcklbw       xm2, xm3
+    vinserti128     m5, m1, xm2, 1
+    pmaddubsw       m5, [r5]
+    movq            xm4, [r0 + r4]                  ; m4 = row 3
+    punpcklbw       xm3, xm4
+    lea             r0, [r0 + r1 * 4]
+    movq            xm1, [r0]                       ; m1 = row 4
+    punpcklbw       xm4, xm1
+    vinserti128     m2, m3, xm4, 1
+    pmaddubsw       m0, m2, [r5 + 1 * mmsize]
+    paddw           m5, m0
+    pmaddubsw       m2, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 5
+    punpcklbw       xm1, xm3
+    movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
+    punpcklbw       xm3, xm4
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m0, m1, [r5 + 1 * mmsize]
+    paddw           m2, m0
+    pmaddubsw       m1, [r5]
+    movq            xm3, [r0 + r4]                  ; m3 = row 7
+    punpcklbw       xm4, xm3
+    lea             r0, [r0 + r1 * 4]
+    movq            xm0, [r0]                       ; m0 = row 8
+    punpcklbw       xm3, xm0
+    vinserti128     m4, m4, xm3, 1
+    pmaddubsw       m3, m4, [r5 + 1 * mmsize]
+    paddw           m1, m3
+    pmaddubsw       m4, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 9
+    punpcklbw       xm0, xm3
+    movq            xm6, [r0 + r1 * 2]              ; m6 = row 10
+    punpcklbw       xm3, xm6
+    vinserti128     m0, m0, xm3, 1
+    pmaddubsw       m3, m0, [r5 + 1 * mmsize]
+    paddw           m4, m3
+    pmaddubsw       m0, [r5]
+%ifidn %1,pp
+    pmulhrsw        m5, m7                          ; m5 = word: row 0, row 1
+    pmulhrsw        m2, m7                          ; m2 = word: row 2, row 3
+    pmulhrsw        m1, m7                          ; m1 = word: row 4, row 5
+    pmulhrsw        m4, m7                          ; m4 = word: row 6, row 7
+    packuswb        m5, m2
+    packuswb        m1, m4
+    vextracti128    xm2, m5, 1
+    vextracti128    xm4, m1, 1
+    movq            [r2], xm5
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm5
+    movhps          [r2 + r6], xm2
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm1
+    movq            [r2 + r3], xm4
+    movhps          [r2 + r3 * 2], xm1
+    movhps          [r2 + r6], xm4
+%else
+    psubw           m5, m7                          ; m5 = word: row 0, row 1
+    psubw           m2, m7                          ; m2 = word: row 2, row 3
+    psubw           m1, m7                          ; m1 = word: row 4, row 5
+    psubw           m4, m7                          ; m4 = word: row 6, row 7
+    vextracti128    xm3, m5, 1
+    movu            [r2], xm5
+    movu            [r2 + r3], xm3
+    vextracti128    xm3, m2, 1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
+    lea             r2, [r2 + r3 * 4]
+    vextracti128    xm5, m1, 1
+    vextracti128    xm3, m4, 1
+    movu            [r2], xm1
+    movu            [r2 + r3], xm5
+    movu            [r2 + r3 * 2], xm4
+    movu            [r2 + r6], xm3
+%endif
+    movq            xm3, [r0 + r4]                  ; m3 = row 11
+    punpcklbw       xm6, xm3
+    lea             r0, [r0 + r1 * 4]
+    movq            xm5, [r0]                       ; m5 = row 12
+    punpcklbw       xm3, xm5
+    vinserti128     m6, m6, xm3, 1
+    pmaddubsw       m3, m6, [r5 + 1 * mmsize]
+    paddw           m0, m3
+    pmaddubsw       m6, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 13
+    punpcklbw       xm5, xm3
+    movq            xm2, [r0 + r1 * 2]              ; m2 = row 14
+    punpcklbw       xm3, xm2
+    vinserti128     m5, m5, xm3, 1
+    pmaddubsw       m3, m5, [r5 + 1 * mmsize]
+    paddw           m6, m3
+    pmaddubsw       m5, [r5]
+    movq            xm3, [r0 + r4]                  ; m3 = row 15
+    punpcklbw       xm2, xm3
+    lea             r0, [r0 + r1 * 4]
+    movq            xm1, [r0]                       ; m1 = row 16
+    punpcklbw       xm3, xm1
+    vinserti128     m2, m2, xm3, 1
+    pmaddubsw       m3, m2, [r5 + 1 * mmsize]
+    paddw           m5, m3
+    pmaddubsw       m2, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 17
+    punpcklbw       xm1, xm3
+    movq            xm4, [r0 + r1 * 2]              ; m4 = row 18
+    punpcklbw       xm3, xm4
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m1, [r5 + 1 * mmsize]
+    paddw           m2, m1
+    lea             r2, [r2 + r3 * 4]
+%ifidn %1,pp
+    pmulhrsw        m0, m7                          ; m0 = word: row 8, row 9
+    pmulhrsw        m6, m7                          ; m6 = word: row 10, row 11
+    pmulhrsw        m5, m7                          ; m5 = word: row 12, row 13
+    pmulhrsw        m2, m7                          ; m2 = word: row 14, row 15
+    packuswb        m0, m6
+    packuswb        m5, m2
+    vextracti128    xm6, m0, 1
+    vextracti128    xm2, m5, 1
+    movq            [r2], xm0
+    movq            [r2 + r3], xm6
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r6], xm6
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm5
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm5
+    movhps          [r2 + r6], xm2
+%else
+    psubw           m0, m7                          ; m0 = word: row 8, row 9
+    psubw           m6, m7                          ; m6 = word: row 10, row 11
+    psubw           m5, m7                          ; m5 = word: row 12, row 13
+    psubw           m2, m7                          ; m2 = word: row 14, row 15
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m6, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r6], xm3
+    lea             r2, [r2 + r3 * 4]
+    vextracti128    xm1, m5, 1
+    vextracti128    xm3, m2, 1
+    movu            [r2], xm5
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
+%endif
+%endmacro
+
+%macro FILTER_VER_CHROMA_AVX2_8x16 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_8x16, 4, 7, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,pp
+    mova            m7, [pw_512]
+%else
+    add             r3d, r3d
+    mova            m7, [pw_2000]
+%endif
+    lea             r6, [r3 * 3]
+    PROCESS_CHROMA_AVX2_W8_16R %1
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_8x16 pp
+    FILTER_VER_CHROMA_AVX2_8x16 ps
+
+%macro FILTER_VER_CHROMA_AVX2_8x12 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_8x12, 4, 7, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1, pp
+    mova            m7, [pw_512]
+%else
+    add             r3d, r3d
+    mova            m7, [pw_2000]
+%endif
+    lea             r6, [r3 * 3]
+    movq            xm1, [r0]                       ; m1 = row 0
+    movq            xm2, [r0 + r1]                  ; m2 = row 1
+    punpcklbw       xm1, xm2
+    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
+    punpcklbw       xm2, xm3
+    vinserti128     m5, m1, xm2, 1
+    pmaddubsw       m5, [r5]
+    movq            xm4, [r0 + r4]                  ; m4 = row 3
+    punpcklbw       xm3, xm4
+    lea             r0, [r0 + r1 * 4]
+    movq            xm1, [r0]                       ; m1 = row 4
+    punpcklbw       xm4, xm1
+    vinserti128     m2, m3, xm4, 1
+    pmaddubsw       m0, m2, [r5 + 1 * mmsize]
+    paddw           m5, m0
+    pmaddubsw       m2, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 5
+    punpcklbw       xm1, xm3
+    movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
+    punpcklbw       xm3, xm4
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m0, m1, [r5 + 1 * mmsize]
+    paddw           m2, m0
+    pmaddubsw       m1, [r5]
+    movq            xm3, [r0 + r4]                  ; m3 = row 7
+    punpcklbw       xm4, xm3
+    lea             r0, [r0 + r1 * 4]
+    movq            xm0, [r0]                       ; m0 = row 8
+    punpcklbw       xm3, xm0
+    vinserti128     m4, m4, xm3, 1
+    pmaddubsw       m3, m4, [r5 + 1 * mmsize]
+    paddw           m1, m3
+    pmaddubsw       m4, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 9
+    punpcklbw       xm0, xm3
+    movq            xm6, [r0 + r1 * 2]              ; m6 = row 10
+    punpcklbw       xm3, xm6
+    vinserti128     m0, m0, xm3, 1
+    pmaddubsw       m3, m0, [r5 + 1 * mmsize]
+    paddw           m4, m3
+    pmaddubsw       m0, [r5]
+%ifidn %1, pp
+    pmulhrsw        m5, m7                          ; m5 = word: row 0, row 1
+    pmulhrsw        m2, m7                          ; m2 = word: row 2, row 3
+    pmulhrsw        m1, m7                          ; m1 = word: row 4, row 5
+    pmulhrsw        m4, m7                          ; m4 = word: row 6, row 7
+    packuswb        m5, m2
+    packuswb        m1, m4
+    vextracti128    xm2, m5, 1
+    vextracti128    xm4, m1, 1
+    movq            [r2], xm5
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm5
+    movhps          [r2 + r6], xm2
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm1
+    movq            [r2 + r3], xm4
+    movhps          [r2 + r3 * 2], xm1
+    movhps          [r2 + r6], xm4
+%else
+    psubw           m5, m7                          ; m5 = word: row 0, row 1
+    psubw           m2, m7                          ; m2 = word: row 2, row 3
+    psubw           m1, m7                          ; m1 = word: row 4, row 5
+    psubw           m4, m7                          ; m4 = word: row 6, row 7
+    vextracti128    xm3, m5, 1
+    movu            [r2], xm5
+    movu            [r2 + r3], xm3
+    vextracti128    xm3, m2, 1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
+    lea             r2, [r2 + r3 * 4]
+    vextracti128    xm5, m1, 1
+    vextracti128    xm3, m4, 1
+    movu            [r2], xm1
+    movu            [r2 + r3], xm5
+    movu            [r2 + r3 * 2], xm4
+    movu            [r2 + r6], xm3
+%endif
+    movq            xm3, [r0 + r4]                  ; m3 = row 11
+    punpcklbw       xm6, xm3
+    lea             r0, [r0 + r1 * 4]
+    movq            xm5, [r0]                       ; m5 = row 12
+    punpcklbw       xm3, xm5
+    vinserti128     m6, m6, xm3, 1
+    pmaddubsw       m3, m6, [r5 + 1 * mmsize]
+    paddw           m0, m3
+    pmaddubsw       m6, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 13
+    punpcklbw       xm5, xm3
+    movq            xm2, [r0 + r1 * 2]              ; m2 = row 14
+    punpcklbw       xm3, xm2
+    vinserti128     m5, m5, xm3, 1
+    pmaddubsw       m3, m5, [r5 + 1 * mmsize]
+    paddw           m6, m3
+    lea             r2, [r2 + r3 * 4]
+%ifidn %1, pp
+    pmulhrsw        m0, m7                          ; m0 = word: row 8, row 9
+    pmulhrsw        m6, m7                          ; m6 = word: row 10, row 11
+    packuswb        m0, m6
+    vextracti128    xm6, m0, 1
+    movq            [r2], xm0
+    movq            [r2 + r3], xm6
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r6], xm6
+%else
+    psubw           m0, m7                          ; m0 = word: row 8, row 9
+    psubw           m6, m7                          ; m6 = word: row 10, row 11
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m6, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r6], xm3
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_8x12 pp
+    FILTER_VER_CHROMA_AVX2_8x12 ps
+
+%macro FILTER_VER_CHROMA_AVX2_8xN 2
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_8x%2, 4, 7, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,pp
+    mova            m7, [pw_512]
+%else
+    add             r3d, r3d
+    mova            m7, [pw_2000]
+%endif
+    lea             r6, [r3 * 3]
+%rep %2 / 16
+    PROCESS_CHROMA_AVX2_W8_16R %1
+    lea             r2, [r2 + r3 * 4]
+%endrep
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_8xN pp, 32
+    FILTER_VER_CHROMA_AVX2_8xN ps, 32
+    FILTER_VER_CHROMA_AVX2_8xN pp, 64
+    FILTER_VER_CHROMA_AVX2_8xN ps, 64
+
+%macro PROCESS_CHROMA_AVX2_W8_4R 0
+    movq            xm1, [r0]                       ; m1 = row 0
+    movq            xm2, [r0 + r1]                  ; m2 = row 1
+    punpcklbw       xm1, xm2                        ; m1 = [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
+    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
+    punpcklbw       xm2, xm3                        ; m2 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10]
+    vinserti128     m0, m1, xm2, 1                  ; m0 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10] - [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
+    pmaddubsw       m0, [r5]
+    movq            xm4, [r0 + r4]                  ; m4 = row 3
+    punpcklbw       xm3, xm4                        ; m3 = [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
+    lea             r0, [r0 + r1 * 4]
+    movq            xm1, [r0]                       ; m1 = row 4
+    punpcklbw       xm4, xm1                        ; m4 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30]
+    vinserti128     m2, m3, xm4, 1                  ; m2 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30] - [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
+    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
+    paddw           m0, m4
+    pmaddubsw       m2, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 5
+    punpcklbw       xm1, xm3                        ; m1 = [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
+    movq            xm4, [r0 + r1 * 2]              ; m4 = row 6
+    punpcklbw       xm3, xm4                        ; m3 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50]
+    vinserti128     m1, m1, xm3, 1                  ; m1 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50] - [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
+    pmaddubsw       m1, [r5 + 1 * mmsize]
+    paddw           m2, m1
+%endmacro
+
+%macro FILTER_VER_CHROMA_AVX2_8x4 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_8x4, 4, 6, 5
+    mov             r4d, r4m
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+    PROCESS_CHROMA_AVX2_W8_4R
+%ifidn %1,pp
+    lea             r4, [r3 * 3]
+    mova            m3, [pw_512]
+    pmulhrsw        m0, m3                          ; m0 = word: row 0, row 1
+    pmulhrsw        m2, m3                          ; m2 = word: row 2, row 3
+    packuswb        m0, m2
+    vextracti128    xm2, m0, 1
+    movq            [r2], xm0
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r4], xm2
+%else
+    add             r3d, r3d
+    vbroadcasti128  m3, [pw_2000]
+    lea             r4, [r3 * 3]
+    psubw           m0, m3                          ; m0 = word: row 0, row 1
+    psubw           m2, m3                          ; m2 = word: row 2, row 3
+    vextracti128    xm1, m0, 1
+    vextracti128    xm4, m2, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r4], xm4
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_8x4 pp
+    FILTER_VER_CHROMA_AVX2_8x4 ps
+
+%macro FILTER_VER_CHROMA_AVX2_8x2 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_8x2, 4, 6, 4
+    mov             r4d, r4m
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+
+    movq            xm1, [r0]                       ; m1 = row 0
+    movq            xm2, [r0 + r1]                  ; m2 = row 1
+    punpcklbw       xm1, xm2                        ; m1 = [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
+    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
+    punpcklbw       xm2, xm3                        ; m2 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10]
+    vinserti128     m1, m1, xm2, 1                  ; m1 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10] - [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
+    pmaddubsw       m1, [r5]
+    movq            xm2, [r0 + r4]                  ; m2 = row 3
+    punpcklbw       xm3, xm2                        ; m3 = [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
+    movq            xm0, [r0 + r1 * 4]              ; m0 = row 4
+    punpcklbw       xm2, xm0                        ; m2 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30]
+    vinserti128     m3, m3, xm2, 1                  ; m3 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30] - [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
+    pmaddubsw       m3, [r5 + 1 * mmsize]
+    paddw           m1, m3
+%ifidn %1,pp
+    pmulhrsw        m1, [pw_512]                    ; m1 = word: row 0, row 1
+    packuswb        m1, m1
+    vextracti128    xm0, m1, 1
+    movq            [r2], xm1
+    movq            [r2 + r3], xm0
+%else
+    add             r3d, r3d
+    psubw           m1, [pw_2000]                   ; m1 = word: row 0, row 1
+    vextracti128    xm0, m1, 1
+    movu            [r2], xm1
+    movu            [r2 + r3], xm0
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_8x2 pp
+    FILTER_VER_CHROMA_AVX2_8x2 ps
+
+%macro FILTER_VER_CHROMA_AVX2_6x8 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_6x8, 4, 6, 7
+    mov             r4d, r4m
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+    PROCESS_CHROMA_AVX2_W8_8R
+%ifidn %1,pp
+    lea             r4, [r3 * 3]
+    mova            m3, [pw_512]
+    pmulhrsw        m5, m3                          ; m5 = word: row 0, row 1
+    pmulhrsw        m2, m3                          ; m2 = word: row 2, row 3
+    pmulhrsw        m1, m3                          ; m1 = word: row 4, row 5
+    pmulhrsw        m4, m3                          ; m4 = word: row 6, row 7
+    packuswb        m5, m2
+    packuswb        m1, m4
+    vextracti128    xm2, m5, 1
+    vextracti128    xm4, m1, 1
+    movd            [r2], xm5
+    pextrw          [r2 + 4], xm5, 2
+    movd            [r2 + r3], xm2
+    pextrw          [r2 + r3 + 4], xm2, 2
+    pextrd          [r2 + r3 * 2], xm5, 2
+    pextrw          [r2 + r3 * 2 + 4], xm5, 6
+    pextrd          [r2 + r4], xm2, 2
+    pextrw          [r2 + r4 + 4], xm2, 6
+    lea             r2, [r2 + r3 * 4]
+    movd            [r2], xm1
+    pextrw          [r2 + 4], xm1, 2
+    movd            [r2 + r3], xm4
+    pextrw          [r2 + r3 + 4], xm4, 2
+    pextrd          [r2 + r3 * 2], xm1, 2
+    pextrw          [r2 + r3 * 2 + 4], xm1, 6
+    pextrd          [r2 + r4], xm4, 2
+    pextrw          [r2 + r4 + 4], xm4, 6
+%else
+    add             r3d, r3d
+    vbroadcasti128  m3, [pw_2000]
+    lea             r4, [r3 * 3]
+    psubw           m5, m3                          ; m5 = word: row 0, row 1
+    psubw           m2, m3                          ; m2 = word: row 2, row 3
+    psubw           m1, m3                          ; m1 = word: row 4, row 5
+    psubw           m4, m3                          ; m4 = word: row 6, row 7
+    vextracti128    xm6, m5, 1
+    vextracti128    xm3, m2, 1
+    vextracti128    xm0, m1, 1
+    movq            [r2], xm5
+    pextrd          [r2 + 8], xm5, 2
+    movq            [r2 + r3], xm6
+    pextrd          [r2 + r3 + 8], xm6, 2
+    movq            [r2 + r3 * 2], xm2
+    pextrd          [r2 + r3 * 2 + 8], xm2, 2
+    movq            [r2 + r4], xm3
+    pextrd          [r2 + r4 + 8], xm3, 2
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm1
+    pextrd          [r2 + 8], xm1, 2
+    movq            [r2 + r3], xm0
+    pextrd          [r2 + r3 + 8], xm0, 2
+    movq            [r2 + r3 * 2], xm4
+    pextrd          [r2 + r3 * 2 + 8], xm4, 2
+    vextracti128    xm4, m4, 1
+    movq            [r2 + r4], xm4
+    pextrd          [r2 + r4 + 8], xm4, 2
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_6x8 pp
+    FILTER_VER_CHROMA_AVX2_6x8 ps
+
+;-----------------------------------------------------------------------------
+;void interp_4tap_vert_pp_6x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W6_H4 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_pp_6x%2, 4, 6, 8
+
+    mov         r4d,       r4m
+    sub         r0,        r1
+
+%ifdef PIC
+    lea         r5,        [tab_ChromaCoeff]
+    movd        m5,        [r5 + r4 * 4]
+%else
+    movd        m5,        [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb      m6,        m5,       [tab_Vm]
+    pshufb      m5,        [tab_Vm + 16]
+    mova        m4,        [pw_512]
+
+    mov         r4d,       %2
+    lea         r5,        [3 * r1]
+
+.loop:
+    movq        m0,        [r0]
+    movq        m1,        [r0 + r1]
+    movq        m2,        [r0 + 2 * r1]
+    movq        m3,        [r0 + r5]
+
+    punpcklbw   m0,        m1
+    punpcklbw   m1,        m2
+    punpcklbw   m2,        m3
+
+    pmaddubsw   m0,        m6
+    pmaddubsw   m7,        m2, m5
+
+    paddw       m0,        m7
+
+    pmulhrsw    m0,        m4
+    packuswb    m0,        m0
+    movd        [r2],      m0
+    pextrw      [r2 + 4],  m0,    2
+
+    lea         r0,        [r0 + 4 * r1]
+
+    movq        m0,        [r0]
+    punpcklbw   m3,        m0
+
+    pmaddubsw   m1,        m6
+    pmaddubsw   m7,        m3, m5
+
+    paddw       m1,        m7
+
+    pmulhrsw    m1,        m4
+    packuswb    m1,        m1
+    movd        [r2 + r3],      m1
+    pextrw      [r2 + r3 + 4],  m1,    2
+
+    movq        m1,        [r0 + r1]
+    punpcklbw   m7,        m0,        m1
+
+    pmaddubsw   m2,        m6
+    pmaddubsw   m7,        m5
+
+    paddw       m2,        m7
+
+    pmulhrsw    m2,        m4
+    packuswb    m2,        m2
+    lea         r2,        [r2 + 2 * r3]
+    movd        [r2],      m2
+    pextrw      [r2 + 4],  m2,    2
+
+    movq        m2,        [r0 + 2 * r1]
+    punpcklbw   m1,        m2
+
+    pmaddubsw   m3,        m6
+    pmaddubsw   m1,        m5
+
+    paddw       m3,        m1
+
+    pmulhrsw    m3,        m4
+    packuswb    m3,        m3
+
+    movd        [r2 + r3],        m3
+    pextrw      [r2 + r3 + 4],    m3,    2
+
+    lea         r2,        [r2 + 2 * r3]
+
+    sub         r4,         4
+    jnz        .loop
+    RET
+%endmacro
+
+    FILTER_V4_W6_H4 6, 8
+
+    FILTER_V4_W6_H4 6, 16
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_pp_12x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W12_H2 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_pp_12x%2, 4, 6, 8
+
+    mov         r4d,       r4m
+    sub         r0,        r1
+
+%ifdef PIC
+    lea         r5,        [tab_ChromaCoeff]
+    movd        m0,        [r5 + r4 * 4]
+%else
+    movd        m0,        [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb      m1,        m0,       [tab_Vm]
+    pshufb      m0,        [tab_Vm + 16]
+
+    mov         r4d,       %2
+
+.loop:
+    movu        m2,        [r0]
+    movu        m3,        [r0 + r1]
+
+    punpcklbw   m4,        m2,        m3
+    punpckhbw   m2,        m3
+
+    pmaddubsw   m4,        m1
+    pmaddubsw   m2,        m1
+
+    lea         r0,        [r0 + 2 * r1]
+    movu        m5,        [r0]
+    movu        m7,        [r0 + r1]
+
+    punpcklbw   m6,        m5,        m7
+    pmaddubsw   m6,        m0
+    paddw       m4,        m6
+
+    punpckhbw   m6,        m5,        m7
+    pmaddubsw   m6,        m0
+    paddw       m2,        m6
+
+    mova        m6,        [pw_512]
+
+    pmulhrsw    m4,        m6
+    pmulhrsw    m2,        m6
+
+    packuswb    m4,        m2
+
+    movh         [r2],     m4
+    pextrd       [r2 + 8], m4,  2
+
+    punpcklbw   m4,        m3,        m5
+    punpckhbw   m3,        m5
+
+    pmaddubsw   m4,        m1
+    pmaddubsw   m3,        m1
+
+    movu        m5,        [r0 + 2 * r1]
+
+    punpcklbw   m2,        m7,        m5
+    punpckhbw   m7,        m5
+
+    pmaddubsw   m2,        m0
+    pmaddubsw   m7,        m0
+
+    paddw       m4,        m2
+    paddw       m3,        m7
+
+    pmulhrsw    m4,        m6
+    pmulhrsw    m3,        m6
+
+    packuswb    m4,        m3
+
+    movh        [r2 + r3],      m4
+    pextrd      [r2 + r3 + 8],  m4,  2
+
+    lea         r2,        [r2 + 2 * r3]
+
+    sub         r4,        2
+    jnz        .loop
+    RET
+%endmacro
+
+    FILTER_V4_W12_H2 12, 16
+
+    FILTER_V4_W12_H2 12, 32
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_pp_16x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W16_H2 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_pp_16x%2, 4, 6, 8
+
+    mov         r4d,       r4m
+    sub         r0,        r1
+
+%ifdef PIC
+    lea         r5,        [tab_ChromaCoeff]
+    movd        m0,        [r5 + r4 * 4]
+%else
+    movd        m0,        [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb      m1,        m0,       [tab_Vm]
+    pshufb      m0,        [tab_Vm + 16]
+
+    mov         r4d,       %2/2
+
+.loop:
+    movu        m2,        [r0]
+    movu        m3,        [r0 + r1]
+
+    punpcklbw   m4,        m2,        m3
+    punpckhbw   m2,        m3
+
+    pmaddubsw   m4,        m1
+    pmaddubsw   m2,        m1
+
+    lea         r0,        [r0 + 2 * r1]
+    movu        m5,        [r0]
+    movu        m6,        [r0 + r1]
+
+    punpckhbw   m7,        m5,        m6
+    pmaddubsw   m7,        m0
+    paddw       m2,        m7
+
+    punpcklbw   m7,        m5,        m6
+    pmaddubsw   m7,        m0
+    paddw       m4,        m7
+
+    mova        m7,        [pw_512]
+
+    pmulhrsw    m4,        m7
+    pmulhrsw    m2,        m7
+
+    packuswb    m4,        m2
+
+    movu        [r2],      m4
+
+    punpcklbw   m4,        m3,        m5
+    punpckhbw   m3,        m5
+
+    pmaddubsw   m4,        m1
+    pmaddubsw   m3,        m1
+
+    movu        m5,        [r0 + 2 * r1]
+
+    punpcklbw   m2,        m6,        m5
+    punpckhbw   m6,        m5
+
+    pmaddubsw   m2,        m0
+    pmaddubsw   m6,        m0
+
+    paddw       m4,        m2
+    paddw       m3,        m6
+
+    pmulhrsw    m4,        m7
+    pmulhrsw    m3,        m7
+
+    packuswb    m4,        m3
+
+    movu        [r2 + r3],      m4
+
+    lea         r2,        [r2 + 2 * r3]
+
+    dec         r4d
+    jnz        .loop
+    RET
+%endmacro
+
+    FILTER_V4_W16_H2 16,  4
+    FILTER_V4_W16_H2 16,  8
+    FILTER_V4_W16_H2 16, 12
+    FILTER_V4_W16_H2 16, 16
+    FILTER_V4_W16_H2 16, 32
+
+    FILTER_V4_W16_H2 16, 24
+    FILTER_V4_W16_H2 16, 64
+
+%macro FILTER_VER_CHROMA_AVX2_16x16 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_vert_%1_16x16, 4, 6, 15
+    mov             r4d, r4m
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    mova            m12, [r5]
+    mova            m13, [r5 + mmsize]
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,pp
+    mova            m14, [pw_512]
+%else
+    add             r3d, r3d
+    vbroadcasti128  m14, [pw_2000]
+%endif
+    lea             r5, [r3 * 3]
+
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddubsw       m0, m12
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m1, m12
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m4, m2, m13
+    paddw           m0, m4
+    pmaddubsw       m2, m12
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddubsw       m5, m3, m13
+    paddw           m1, m5
+    pmaddubsw       m3, m12
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhbw       xm6, xm4, xm5
+    punpcklbw       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddubsw       m6, m4, m13
+    paddw           m2, m6
+    pmaddubsw       m4, m12
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhbw       xm7, xm5, xm6
+    punpcklbw       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddubsw       m7, m5, m13
+    paddw           m3, m7
+    pmaddubsw       m5, m12
+    movu            xm7, [r0 + r4]                  ; m7 = row 7
+    punpckhbw       xm8, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddubsw       m8, m6, m13
+    paddw           m4, m8
+    pmaddubsw       m6, m12
+    lea             r0, [r0 + r1 * 4]
+    movu            xm8, [r0]                       ; m8 = row 8
+    punpckhbw       xm9, xm7, xm8
+    punpcklbw       xm7, xm8
+    vinserti128     m7, m7, xm9, 1
+    pmaddubsw       m9, m7, m13
+    paddw           m5, m9
+    pmaddubsw       m7, m12
+    movu            xm9, [r0 + r1]                  ; m9 = row 9
+    punpckhbw       xm10, xm8, xm9
+    punpcklbw       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddubsw       m10, m8, m13
+    paddw           m6, m10
+    pmaddubsw       m8, m12
+    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
+    punpckhbw       xm11, xm9, xm10
+    punpcklbw       xm9, xm10
+    vinserti128     m9, m9, xm11, 1
+    pmaddubsw       m11, m9, m13
+    paddw           m7, m11
+    pmaddubsw       m9, m12
+
+%ifidn %1,pp
+    pmulhrsw        m0, m14                         ; m0 = word: row 0
+    pmulhrsw        m1, m14                         ; m1 = word: row 1
+    pmulhrsw        m2, m14                         ; m2 = word: row 2
+    pmulhrsw        m3, m14                         ; m3 = word: row 3
+    pmulhrsw        m4, m14                         ; m4 = word: row 4
+    pmulhrsw        m5, m14                         ; m5 = word: row 5
+    pmulhrsw        m6, m14                         ; m6 = word: row 6
+    pmulhrsw        m7, m14                         ; m7 = word: row 7
+    packuswb        m0, m1
+    packuswb        m2, m3
+    packuswb        m4, m5
+    packuswb        m6, m7
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    vpermq          m4, m4, 11011000b
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m2, 1
+    vextracti128    xm5, m4, 1
+    vextracti128    xm7, m6, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r5], xm3
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm4
+    movu            [r2 + r3], xm5
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r5], xm7
+%else
+    psubw           m0, m14                         ; m0 = word: row 0
+    psubw           m1, m14                         ; m1 = word: row 1
+    psubw           m2, m14                         ; m2 = word: row 2
+    psubw           m3, m14                         ; m3 = word: row 3
+    psubw           m4, m14                         ; m4 = word: row 4
+    psubw           m5, m14                         ; m5 = word: row 5
+    psubw           m6, m14                         ; m6 = word: row 6
+    psubw           m7, m14                         ; m7 = word: row 7
+    movu            [r2], m0
+    movu            [r2 + r3], m1
+    movu            [r2 + r3 * 2], m2
+    movu            [r2 + r5], m3
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], m4
+    movu            [r2 + r3], m5
+    movu            [r2 + r3 * 2], m6
+    movu            [r2 + r5], m7
+%endif
+    lea             r2, [r2 + r3 * 4]
+
+    movu            xm11, [r0 + r4]                 ; m11 = row 11
+    punpckhbw       xm6, xm10, xm11
+    punpcklbw       xm10, xm11
+    vinserti128     m10, m10, xm6, 1
+    pmaddubsw       m6, m10, m13
+    paddw           m8, m6
+    pmaddubsw       m10, m12
+    lea             r0, [r0 + r1 * 4]
+    movu            xm6, [r0]                       ; m6 = row 12
+    punpckhbw       xm7, xm11, xm6
+    punpcklbw       xm11, xm6
+    vinserti128     m11, m11, xm7, 1
+    pmaddubsw       m7, m11, m13
+    paddw           m9, m7
+    pmaddubsw       m11, m12
+
+    movu            xm7, [r0 + r1]                  ; m7 = row 13
+    punpckhbw       xm0, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm0, 1
+    pmaddubsw       m0, m6, m13
+    paddw           m10, m0
+    pmaddubsw       m6, m12
+    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
+    punpckhbw       xm1, xm7, xm0
+    punpcklbw       xm7, xm0
+    vinserti128     m7, m7, xm1, 1
+    pmaddubsw       m1, m7, m13
+    paddw           m11, m1
+    pmaddubsw       m7, m12
+    movu            xm1, [r0 + r4]                  ; m1 = row 15
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddubsw       m2, m0, m13
+    paddw           m6, m2
+    pmaddubsw       m0, m12
+    lea             r0, [r0 + r1 * 4]
+    movu            xm2, [r0]                       ; m2 = row 16
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m3, m1, m13
+    paddw           m7, m3
+    pmaddubsw       m1, m12
+    movu            xm3, [r0 + r1]                  ; m3 = row 17
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m2, m13
+    paddw           m0, m2
+    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddubsw       m3, m13
+    paddw           m1, m3
+
+%ifidn %1,pp
+    pmulhrsw        m8, m14                         ; m8 = word: row 8
+    pmulhrsw        m9, m14                         ; m9 = word: row 9
+    pmulhrsw        m10, m14                        ; m10 = word: row 10
+    pmulhrsw        m11, m14                        ; m11 = word: row 11
+    pmulhrsw        m6, m14                         ; m6 = word: row 12
+    pmulhrsw        m7, m14                         ; m7 = word: row 13
+    pmulhrsw        m0, m14                         ; m0 = word: row 14
+    pmulhrsw        m1, m14                         ; m1 = word: row 15
+    packuswb        m8, m9
+    packuswb        m10, m11
+    packuswb        m6, m7
+    packuswb        m0, m1
+    vpermq          m8, m8, 11011000b
+    vpermq          m10, m10, 11011000b
+    vpermq          m6, m6, 11011000b
+    vpermq          m0, m0, 11011000b
+    vextracti128    xm9, m8, 1
+    vextracti128    xm11, m10, 1
+    vextracti128    xm7, m6, 1
+    vextracti128    xm1, m0, 1
+    movu            [r2], xm8
+    movu            [r2 + r3], xm9
+    movu            [r2 + r3 * 2], xm10
+    movu            [r2 + r5], xm11
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm6
+    movu            [r2 + r3], xm7
+    movu            [r2 + r3 * 2], xm0
+    movu            [r2 + r5], xm1
+%else
+    psubw           m8, m14                         ; m8 = word: row 8
+    psubw           m9, m14                         ; m9 = word: row 9
+    psubw           m10, m14                        ; m10 = word: row 10
+    psubw           m11, m14                        ; m11 = word: row 11
+    psubw           m6, m14                         ; m6 = word: row 12
+    psubw           m7, m14                         ; m7 = word: row 13
+    psubw           m0, m14                         ; m0 = word: row 14
+    psubw           m1, m14                         ; m1 = word: row 15
+    movu            [r2], m8
+    movu            [r2 + r3], m9
+    movu            [r2 + r3 * 2], m10
+    movu            [r2 + r5], m11
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], m6
+    movu            [r2 + r3], m7
+    movu            [r2 + r3 * 2], m0
+    movu            [r2 + r5], m1
+%endif
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_16x16 pp
+    FILTER_VER_CHROMA_AVX2_16x16 ps
+%macro FILTER_VER_CHROMA_AVX2_16x8 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_16x8, 4, 7, 7
+    mov             r4d, r4m
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,pp
+    mova            m6, [pw_512]
+%else
+    add             r3d, r3d
+    mova            m6, [pw_2000]
+%endif
+    lea             r6, [r3 * 3]
+
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddubsw       m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m4, m2, [r5 + mmsize]
+    paddw           m0, m4
+    pmaddubsw       m2, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddubsw       m5, m3, [r5 + mmsize]
+    paddw           m1, m5
+    pmaddubsw       m3, [r5]
+%ifidn %1,pp
+    pmulhrsw        m0, m6                          ; m0 = word: row 0
+    pmulhrsw        m1, m6                          ; m1 = word: row 1
+    packuswb        m0, m1
+    vpermq          m0, m0, 11011000b
+    vextracti128    xm1, m0, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+%else
+    psubw           m0, m6                          ; m0 = word: row 0
+    psubw           m1, m6                          ; m1 = word: row 1
+    movu            [r2], m0
+    movu            [r2 + r3], m1
+%endif
+
+    movu            xm0, [r0 + r1]                  ; m0 = row 5
+    punpckhbw       xm1, xm4, xm0
+    punpcklbw       xm4, xm0
+    vinserti128     m4, m4, xm1, 1
+    pmaddubsw       m1, m4, [r5 + mmsize]
+    paddw           m2, m1
+    pmaddubsw       m4, [r5]
+    movu            xm1, [r0 + r1 * 2]              ; m1 = row 6
+    punpckhbw       xm5, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm5, 1
+    pmaddubsw       m5, m0, [r5 + mmsize]
+    paddw           m3, m5
+    pmaddubsw       m0, [r5]
+%ifidn %1,pp
+    pmulhrsw        m2, m6                          ; m2 = word: row 2
+    pmulhrsw        m3, m6                          ; m3 = word: row 3
+    packuswb        m2, m3
+    vpermq          m2, m2, 11011000b
+    vextracti128    xm3, m2, 1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
+%else
+    psubw           m2, m6                          ; m2 = word: row 2
+    psubw           m3, m6                          ; m3 = word: row 3
+    movu            [r2 + r3 * 2], m2
+    movu            [r2 + r6], m3
+%endif
+
+    movu            xm2, [r0 + r4]                  ; m2 = row 7
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m3, m1, [r5 + mmsize]
+    paddw           m4, m3
+    pmaddubsw       m1, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm3, [r0]                       ; m3 = row 8
+    punpckhbw       xm5, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm5, 1
+    pmaddubsw       m5, m2, [r5 + mmsize]
+    paddw           m0, m5
+    pmaddubsw       m2, [r5]
+    lea             r2, [r2 + r3 * 4]
+%ifidn %1,pp
+    pmulhrsw        m4, m6                          ; m4 = word: row 4
+    pmulhrsw        m0, m6                          ; m0 = word: row 5
+    packuswb        m4, m0
+    vpermq          m4, m4, 11011000b
+    vextracti128    xm0, m4, 1
+    movu            [r2], xm4
+    movu            [r2 + r3], xm0
+%else
+    psubw           m4, m6                          ; m4 = word: row 4
+    psubw           m0, m6                          ; m0 = word: row 5
+    movu            [r2], m4
+    movu            [r2 + r3], m0
+%endif
+
+    movu            xm5, [r0 + r1]                  ; m5 = row 9
+    punpckhbw       xm4, xm3, xm5
+    punpcklbw       xm3, xm5
+    vinserti128     m3, m3, xm4, 1
+    pmaddubsw       m3, [r5 + mmsize]
+    paddw           m1, m3
+    movu            xm4, [r0 + r1 * 2]              ; m4 = row 10
+    punpckhbw       xm0, xm5, xm4
+    punpcklbw       xm5, xm4
+    vinserti128     m5, m5, xm0, 1
+    pmaddubsw       m5, [r5 + mmsize]
+    paddw           m2, m5
+%ifidn %1,pp
+    pmulhrsw        m1, m6                          ; m1 = word: row 6
+    pmulhrsw        m2, m6                          ; m2 = word: row 7
+    packuswb        m1, m2
+    vpermq          m1, m1, 11011000b
+    vextracti128    xm2, m1, 1
+    movu            [r2 + r3 * 2], xm1
+    movu            [r2 + r6], xm2
+%else
+    psubw           m1, m6                          ; m1 = word: row 6
+    psubw           m2, m6                          ; m2 = word: row 7
+    movu            [r2 + r3 * 2], m1
+    movu            [r2 + r6], m2
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_16x8 pp
+    FILTER_VER_CHROMA_AVX2_16x8 ps
+
+%macro FILTER_VER_CHROMA_AVX2_16x12 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_vert_%1_16x12, 4, 6, 10
+    mov             r4d, r4m
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    mova            m8, [r5]
+    mova            m9, [r5 + mmsize]
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,pp
+    mova            m7, [pw_512]
+%else
+    add             r3d, r3d
+    vbroadcasti128  m7, [pw_2000]
+%endif
+    lea             r5, [r3 * 3]
+
+    movu            xm0, [r0]
+    vinserti128     m0, m0, [r0 + r1 * 2], 1
+    movu            xm1, [r0 + r1]
+    vinserti128     m1, m1, [r0 + r4], 1
+
+    punpcklbw       m2, m0, m1
+    punpckhbw       m3, m0, m1
+    vperm2i128      m4, m2, m3, 0x20
+    vperm2i128      m2, m2, m3, 0x31
+    pmaddubsw       m4, m8
+    pmaddubsw       m3, m2, m9
+    paddw           m4, m3
+    pmaddubsw       m2, m8
+
+    vextracti128    xm0, m0, 1
+    lea             r0, [r0 + r1 * 4]
+    vinserti128     m0, m0, [r0], 1
+
+    punpcklbw       m5, m1, m0
+    punpckhbw       m3, m1, m0
+    vperm2i128      m6, m5, m3, 0x20
+    vperm2i128      m5, m5, m3, 0x31
+    pmaddubsw       m6, m8
+    pmaddubsw       m3, m5, m9
+    paddw           m6, m3
+    pmaddubsw       m5, m8
+%ifidn %1,pp
+    pmulhrsw        m4, m7                         ; m4 = word: row 0
+    pmulhrsw        m6, m7                         ; m6 = word: row 1
+    packuswb        m4, m6
+    vpermq          m4, m4, 11011000b
+    vextracti128    xm6, m4, 1
+    movu            [r2], xm4
+    movu            [r2 + r3], xm6
+%else
+    psubw           m4, m7                         ; m4 = word: row 0
+    psubw           m6, m7                         ; m6 = word: row 1
+    movu            [r2], m4
+    movu            [r2 + r3], m6
+%endif
+
+    movu            xm4, [r0 + r1 * 2]
+    vinserti128     m4, m4, [r0 + r1], 1
+    vextracti128    xm1, m4, 1
+    vinserti128     m0, m0, xm1, 0
+
+    punpcklbw       m6, m0, m4
+    punpckhbw       m1, m0, m4
+    vperm2i128      m0, m6, m1, 0x20
+    vperm2i128      m6, m6, m1, 0x31
+    pmaddubsw       m1, m0, m9
+    paddw           m5, m1
+    pmaddubsw       m0, m8
+    pmaddubsw       m1, m6, m9
+    paddw           m2, m1
+    pmaddubsw       m6, m8
+
+%ifidn %1,pp
+    pmulhrsw        m2, m7                         ; m2 = word: row 2
+    pmulhrsw        m5, m7                         ; m5 = word: row 3
+    packuswb        m2, m5
+    vpermq          m2, m2, 11011000b
+    vextracti128    xm5, m2, 1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r5], xm5
+%else
+    psubw           m2, m7                         ; m2 = word: row 2
+    psubw           m5, m7                         ; m5 = word: row 3
+    movu            [r2 + r3 * 2], m2
+    movu            [r2 + r5], m5
+%endif
+    lea             r2, [r2 + r3 * 4]
+
+    movu            xm1, [r0 + r4]
+    lea             r0, [r0 + r1 * 4]
+    vinserti128     m1, m1, [r0], 1
+    vinserti128     m4, m4, xm1, 1
+
+    punpcklbw       m2, m4, m1
+    punpckhbw       m5, m4, m1
+    vperm2i128      m3, m2, m5, 0x20
+    vperm2i128      m2, m2, m5, 0x31
+    pmaddubsw       m5, m3, m9
+    paddw           m6, m5
+    pmaddubsw       m3, m8
+    pmaddubsw       m5, m2, m9
+    paddw           m0, m5
+    pmaddubsw       m2, m8
+
+%ifidn %1,pp
+    pmulhrsw        m6, m7                         ; m6 = word: row 4
+    pmulhrsw        m0, m7                         ; m0 = word: row 5
+    packuswb        m6, m0
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm0, m6, 1
+    movu            [r2], xm6
+    movu            [r2 + r3], xm0
+%else
+    psubw           m6, m7                         ; m6 = word: row 4
+    psubw           m0, m7                         ; m0 = word: row 5
+    movu            [r2], m6
+    movu            [r2 + r3], m0
+%endif
+
+    movu            xm6, [r0 + r1 * 2]
+    vinserti128     m6, m6, [r0 + r1], 1
+    vextracti128    xm0, m6, 1
+    vinserti128     m1, m1, xm0, 0
+
+    punpcklbw       m4, m1, m6
+    punpckhbw       m5, m1, m6
+    vperm2i128      m0, m4, m5, 0x20
+    vperm2i128      m5, m4, m5, 0x31
+    pmaddubsw       m4, m0, m9
+    paddw           m2, m4
+    pmaddubsw       m0, m8
+    pmaddubsw       m4, m5, m9
+    paddw           m3, m4
+    pmaddubsw       m5, m8
+
+%ifidn %1,pp
+    pmulhrsw        m3, m7                         ; m3 = word: row 6
+    pmulhrsw        m2, m7                         ; m2 = word: row 7
+    packuswb        m3, m2
+    vpermq          m3, m3, 11011000b
+    vextracti128    xm2, m3, 1
+    movu            [r2 + r3 * 2], xm3
+    movu            [r2 + r5], xm2
+%else
+    psubw           m3, m7                         ; m3 = word: row 6
+    psubw           m2, m7                         ; m2 = word: row 7
+    movu            [r2 + r3 * 2], m3
+    movu            [r2 + r5], m2
+%endif
+    lea             r2, [r2 + r3 * 4]
+
+    movu            xm3, [r0 + r4]
+    lea             r0, [r0 + r1 * 4]
+    vinserti128     m3, m3, [r0], 1
+    vinserti128     m6, m6, xm3, 1
+
+    punpcklbw       m2, m6, m3
+    punpckhbw       m1, m6, m3
+    vperm2i128      m4, m2, m1, 0x20
+    vperm2i128      m2, m2, m1, 0x31
+    pmaddubsw       m1, m4, m9
+    paddw           m5, m1
+    pmaddubsw       m4, m8
+    pmaddubsw       m1, m2, m9
+    paddw           m0, m1
+    pmaddubsw       m2, m8
+
+%ifidn %1,pp
+    pmulhrsw        m5, m7                         ; m5 = word: row 8
+    pmulhrsw        m0, m7                         ; m0 = word: row 9
+    packuswb        m5, m0
+    vpermq          m5, m5, 11011000b
+    vextracti128    xm0, m5, 1
+    movu            [r2], xm5
+    movu            [r2 + r3], xm0
+%else
+    psubw           m5, m7                         ; m5 = word: row 8
+    psubw           m0, m7                         ; m0 = word: row 9
+    movu            [r2], m5
+    movu            [r2 + r3], m0
+%endif
+
+    movu            xm5, [r0 + r1 * 2]
+    vinserti128     m5, m5, [r0 + r1], 1
+    vextracti128    xm0, m5, 1
+    vinserti128     m3, m3, xm0, 0
+
+    punpcklbw       m1, m3, m5
+    punpckhbw       m0, m3, m5
+    vperm2i128      m6, m1, m0, 0x20
+    vperm2i128      m0, m1, m0, 0x31
+    pmaddubsw       m1, m6, m9
+    paddw           m2, m1
+    pmaddubsw       m1, m0, m9
+    paddw           m4, m1
+
+%ifidn %1,pp
+    pmulhrsw        m4, m7                         ; m4 = word: row 10
+    pmulhrsw        m2, m7                         ; m2 = word: row 11
+    packuswb        m4, m2
+    vpermq          m4, m4, 11011000b
+    vextracti128    xm2, m4, 1
+    movu            [r2 + r3 * 2], xm4
+    movu            [r2 + r5], xm2
+%else
+    psubw           m4, m7                         ; m4 = word: row 10
+    psubw           m2, m7                         ; m2 = word: row 11
+    movu            [r2 + r3 * 2], m4
+    movu            [r2 + r5], m2
+%endif
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_16x12 pp
+    FILTER_VER_CHROMA_AVX2_16x12 ps
+
+%macro FILTER_VER_CHROMA_AVX2_16xN 2
+%if ARCH_X86_64 == 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_16x%2, 4, 8, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,pp
+    mova            m7, [pw_512]
+%else
+    add             r3d, r3d
+    mova            m7, [pw_2000]
+%endif
+    lea             r6, [r3 * 3]
+    mov             r7d, %2 / 16
+.loopH:
+    movu            xm0, [r0]
+    vinserti128     m0, m0, [r0 + r1 * 2], 1
+    movu            xm1, [r0 + r1]
+    vinserti128     m1, m1, [r0 + r4], 1
+
+    punpcklbw       m2, m0, m1
+    punpckhbw       m3, m0, m1
+    vperm2i128      m4, m2, m3, 0x20
+    vperm2i128      m2, m2, m3, 0x31
+    pmaddubsw       m4, [r5]
+    pmaddubsw       m3, m2, [r5 + mmsize]
+    paddw           m4, m3
+    pmaddubsw       m2, [r5]
+
+    vextracti128    xm0, m0, 1
+    lea             r0, [r0 + r1 * 4]
+    vinserti128     m0, m0, [r0], 1
+
+    punpcklbw       m5, m1, m0
+    punpckhbw       m3, m1, m0
+    vperm2i128      m6, m5, m3, 0x20
+    vperm2i128      m5, m5, m3, 0x31
+    pmaddubsw       m6, [r5]
+    pmaddubsw       m3, m5, [r5 + mmsize]
+    paddw           m6, m3
+    pmaddubsw       m5, [r5]
+%ifidn %1,pp
+    pmulhrsw        m4, m7                         ; m4 = word: row 0
+    pmulhrsw        m6, m7                         ; m6 = word: row 1
+    packuswb        m4, m6
+    vpermq          m4, m4, 11011000b
+    vextracti128    xm6, m4, 1
+    movu            [r2], xm4
+    movu            [r2 + r3], xm6
+%else
+    psubw           m4, m7                         ; m4 = word: row 0
+    psubw           m6, m7                         ; m6 = word: row 1
+    movu            [r2], m4
+    movu            [r2 + r3], m6
+%endif
+
+    movu            xm4, [r0 + r1 * 2]
+    vinserti128     m4, m4, [r0 + r1], 1
+    vextracti128    xm1, m4, 1
+    vinserti128     m0, m0, xm1, 0
+
+    punpcklbw       m6, m0, m4
+    punpckhbw       m1, m0, m4
+    vperm2i128      m0, m6, m1, 0x20
+    vperm2i128      m6, m6, m1, 0x31
+    pmaddubsw       m1, m0, [r5 + mmsize]
+    paddw           m5, m1
+    pmaddubsw       m0, [r5]
+    pmaddubsw       m1, m6, [r5 + mmsize]
+    paddw           m2, m1
+    pmaddubsw       m6, [r5]
+
+%ifidn %1,pp
+    pmulhrsw        m2, m7                         ; m2 = word: row 2
+    pmulhrsw        m5, m7                         ; m5 = word: row 3
+    packuswb        m2, m5
+    vpermq          m2, m2, 11011000b
+    vextracti128    xm5, m2, 1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm5
+%else
+    psubw           m2, m7                         ; m2 = word: row 2
+    psubw           m5, m7                         ; m5 = word: row 3
+    movu            [r2 + r3 * 2], m2
+    movu            [r2 + r6], m5
+%endif
+    lea             r2, [r2 + r3 * 4]
+
+    movu            xm1, [r0 + r4]
+    lea             r0, [r0 + r1 * 4]
+    vinserti128     m1, m1, [r0], 1
+    vinserti128     m4, m4, xm1, 1
+
+    punpcklbw       m2, m4, m1
+    punpckhbw       m5, m4, m1
+    vperm2i128      m3, m2, m5, 0x20
+    vperm2i128      m2, m2, m5, 0x31
+    pmaddubsw       m5, m3, [r5 + mmsize]
+    paddw           m6, m5
+    pmaddubsw       m3, [r5]
+    pmaddubsw       m5, m2, [r5 + mmsize]
+    paddw           m0, m5
+    pmaddubsw       m2, [r5]
+
+%ifidn %1,pp
+    pmulhrsw        m6, m7                         ; m6 = word: row 4
+    pmulhrsw        m0, m7                         ; m0 = word: row 5
+    packuswb        m6, m0
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm0, m6, 1
+    movu            [r2], xm6
+    movu            [r2 + r3], xm0
+%else
+    psubw           m6, m7                         ; m6 = word: row 4
+    psubw           m0, m7                         ; m0 = word: row 5
+    movu            [r2], m6
+    movu            [r2 + r3], m0
+%endif
+
+    movu            xm6, [r0 + r1 * 2]
+    vinserti128     m6, m6, [r0 + r1], 1
+    vextracti128    xm0, m6, 1
+    vinserti128     m1, m1, xm0, 0
+
+    punpcklbw       m4, m1, m6
+    punpckhbw       m5, m1, m6
+    vperm2i128      m0, m4, m5, 0x20
+    vperm2i128      m5, m4, m5, 0x31
+    pmaddubsw       m4, m0, [r5 + mmsize]
+    paddw           m2, m4
+    pmaddubsw       m0, [r5]
+    pmaddubsw       m4, m5, [r5 + mmsize]
+    paddw           m3, m4
+    pmaddubsw       m5, [r5]
+
+%ifidn %1,pp
+    pmulhrsw        m3, m7                         ; m3 = word: row 6
+    pmulhrsw        m2, m7                         ; m2 = word: row 7
+    packuswb        m3, m2
+    vpermq          m3, m3, 11011000b
+    vextracti128    xm2, m3, 1
+    movu            [r2 + r3 * 2], xm3
+    movu            [r2 + r6], xm2
+%else
+    psubw           m3, m7                         ; m3 = word: row 6
+    psubw           m2, m7                         ; m2 = word: row 7
+    movu            [r2 + r3 * 2], m3
+    movu            [r2 + r6], m2
+%endif
+    lea             r2, [r2 + r3 * 4]
+
+    movu            xm3, [r0 + r4]
+    lea             r0, [r0 + r1 * 4]
+    vinserti128     m3, m3, [r0], 1
+    vinserti128     m6, m6, xm3, 1
+
+    punpcklbw       m2, m6, m3
+    punpckhbw       m1, m6, m3
+    vperm2i128      m4, m2, m1, 0x20
+    vperm2i128      m2, m2, m1, 0x31
+    pmaddubsw       m1, m4, [r5 + mmsize]
+    paddw           m5, m1
+    pmaddubsw       m4, [r5]
+    pmaddubsw       m1, m2, [r5 + mmsize]
+    paddw           m0, m1
+    pmaddubsw       m2, [r5]
+
+%ifidn %1,pp
+    pmulhrsw        m5, m7                         ; m5 = word: row 8
+    pmulhrsw        m0, m7                         ; m0 = word: row 9
+    packuswb        m5, m0
+    vpermq          m5, m5, 11011000b
+    vextracti128    xm0, m5, 1
+    movu            [r2], xm5
+    movu            [r2 + r3], xm0
+%else
+    psubw           m5, m7                         ; m5 = word: row 8
+    psubw           m0, m7                         ; m0 = word: row 9
+    movu            [r2], m5
+    movu            [r2 + r3], m0
+%endif
+
+    movu            xm5, [r0 + r1 * 2]
+    vinserti128     m5, m5, [r0 + r1], 1
+    vextracti128    xm0, m5, 1
+    vinserti128     m3, m3, xm0, 0
+
+    punpcklbw       m1, m3, m5
+    punpckhbw       m0, m3, m5
+    vperm2i128      m6, m1, m0, 0x20
+    vperm2i128      m0, m1, m0, 0x31
+    pmaddubsw       m1, m6, [r5 + mmsize]
+    paddw           m2, m1
+    pmaddubsw       m6, [r5]
+    pmaddubsw       m1, m0, [r5 + mmsize]
+    paddw           m4, m1
+    pmaddubsw       m0, [r5]
+
+%ifidn %1,pp
+    pmulhrsw        m4, m7                         ; m4 = word: row 10
+    pmulhrsw        m2, m7                         ; m2 = word: row 11
+    packuswb        m4, m2
+    vpermq          m4, m4, 11011000b
+    vextracti128    xm2, m4, 1
+    movu            [r2 + r3 * 2], xm4
+    movu            [r2 + r6], xm2
+%else
+    psubw           m4, m7                         ; m4 = word: row 10
+    psubw           m2, m7                         ; m2 = word: row 11
+    movu            [r2 + r3 * 2], m4
+    movu            [r2 + r6], m2
+%endif
+    lea             r2, [r2 + r3 * 4]
+
+    movu            xm3, [r0 + r4]
+    lea             r0, [r0 + r1 * 4]
+    vinserti128     m3, m3, [r0], 1
+    vinserti128     m5, m5, xm3, 1
+
+    punpcklbw       m2, m5, m3
+    punpckhbw       m1, m5, m3
+    vperm2i128      m4, m2, m1, 0x20
+    vperm2i128      m2, m2, m1, 0x31
+    pmaddubsw       m1, m4, [r5 + mmsize]
+    paddw           m0, m1
+    pmaddubsw       m4, [r5]
+    pmaddubsw       m1, m2, [r5 + mmsize]
+    paddw           m6, m1
+    pmaddubsw       m2, [r5]
+
+%ifidn %1,pp
+    pmulhrsw        m0, m7                         ; m0 = word: row 12
+    pmulhrsw        m6, m7                         ; m6 = word: row 13
+    packuswb        m0, m6
+    vpermq          m0, m0, 11011000b
+    vextracti128    xm6, m0, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm6
+%else
+    psubw           m0, m7                         ; m0 = word: row 12
+    psubw           m6, m7                         ; m6 = word: row 13
+    movu            [r2], m0
+    movu            [r2 + r3], m6
+%endif
+
+    movu            xm5, [r0 + r1 * 2]
+    vinserti128     m5, m5, [r0 + r1], 1
+    vextracti128    xm0, m5, 1
+    vinserti128     m3, m3, xm0, 0
+
+    punpcklbw       m1, m3, m5
+    punpckhbw       m0, m3, m5
+    vperm2i128      m6, m1, m0, 0x20
+    vperm2i128      m0, m1, m0, 0x31
+    pmaddubsw       m6, [r5 + mmsize]
+    paddw           m2, m6
+    pmaddubsw       m0, [r5 + mmsize]
+    paddw           m4, m0
+
+%ifidn %1,pp
+    pmulhrsw        m4, m7                         ; m4 = word: row 14
+    pmulhrsw        m2, m7                         ; m2 = word: row 15
+    packuswb        m4, m2
+    vpermq          m4, m4, 11011000b
+    vextracti128    xm2, m4, 1
+    movu            [r2 + r3 * 2], xm4
+    movu            [r2 + r6], xm2
+%else
+    psubw           m4, m7                         ; m4 = word: row 14
+    psubw           m2, m7                         ; m2 = word: row 15
+    movu            [r2 + r3 * 2], m4
+    movu            [r2 + r6], m2
+%endif
+    lea             r2, [r2 + r3 * 4]
+    dec             r7d
+    jnz             .loopH
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_16xN pp, 32
+    FILTER_VER_CHROMA_AVX2_16xN ps, 32
+    FILTER_VER_CHROMA_AVX2_16xN pp, 64
+    FILTER_VER_CHROMA_AVX2_16xN ps, 64
+
+%macro FILTER_VER_CHROMA_AVX2_16x24 1
+%if ARCH_X86_64 == 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_16x24, 4, 6, 15
+    mov             r4d, r4m
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    mova            m12, [r5]
+    mova            m13, [r5 + mmsize]
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,pp
+    mova            m14, [pw_512]
+%else
+    add             r3d, r3d
+    vbroadcasti128  m14, [pw_2000]
+%endif
+    lea             r5, [r3 * 3]
+
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddubsw       m0, m12
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m1, m12
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m4, m2, m13
+    paddw           m0, m4
+    pmaddubsw       m2, m12
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddubsw       m5, m3, m13
+    paddw           m1, m5
+    pmaddubsw       m3, m12
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhbw       xm6, xm4, xm5
+    punpcklbw       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddubsw       m6, m4, m13
+    paddw           m2, m6
+    pmaddubsw       m4, m12
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhbw       xm7, xm5, xm6
+    punpcklbw       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddubsw       m7, m5, m13
+    paddw           m3, m7
+    pmaddubsw       m5, m12
+    movu            xm7, [r0 + r4]                  ; m7 = row 7
+    punpckhbw       xm8, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddubsw       m8, m6, m13
+    paddw           m4, m8
+    pmaddubsw       m6, m12
+    lea             r0, [r0 + r1 * 4]
+    movu            xm8, [r0]                       ; m8 = row 8
+    punpckhbw       xm9, xm7, xm8
+    punpcklbw       xm7, xm8
+    vinserti128     m7, m7, xm9, 1
+    pmaddubsw       m9, m7, m13
+    paddw           m5, m9
+    pmaddubsw       m7, m12
+    movu            xm9, [r0 + r1]                  ; m9 = row 9
+    punpckhbw       xm10, xm8, xm9
+    punpcklbw       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddubsw       m10, m8, m13
+    paddw           m6, m10
+    pmaddubsw       m8, m12
+    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
+    punpckhbw       xm11, xm9, xm10
+    punpcklbw       xm9, xm10
+    vinserti128     m9, m9, xm11, 1
+    pmaddubsw       m11, m9, m13
+    paddw           m7, m11
+    pmaddubsw       m9, m12
+
+%ifidn %1,pp
+    pmulhrsw        m0, m14                         ; m0 = word: row 0
+    pmulhrsw        m1, m14                         ; m1 = word: row 1
+    pmulhrsw        m2, m14                         ; m2 = word: row 2
+    pmulhrsw        m3, m14                         ; m3 = word: row 3
+    pmulhrsw        m4, m14                         ; m4 = word: row 4
+    pmulhrsw        m5, m14                         ; m5 = word: row 5
+    pmulhrsw        m6, m14                         ; m6 = word: row 6
+    pmulhrsw        m7, m14                         ; m7 = word: row 7
+    packuswb        m0, m1
+    packuswb        m2, m3
+    packuswb        m4, m5
+    packuswb        m6, m7
+    vpermq          m0, m0, q3120
+    vpermq          m2, m2, q3120
+    vpermq          m4, m4, q3120
+    vpermq          m6, m6, q3120
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m2, 1
+    vextracti128    xm5, m4, 1
+    vextracti128    xm7, m6, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r5], xm3
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm4
+    movu            [r2 + r3], xm5
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r5], xm7
+%else
+    psubw           m0, m14                         ; m0 = word: row 0
+    psubw           m1, m14                         ; m1 = word: row 1
+    psubw           m2, m14                         ; m2 = word: row 2
+    psubw           m3, m14                         ; m3 = word: row 3
+    psubw           m4, m14                         ; m4 = word: row 4
+    psubw           m5, m14                         ; m5 = word: row 5
+    psubw           m6, m14                         ; m6 = word: row 6
+    psubw           m7, m14                         ; m7 = word: row 7
+    movu            [r2], m0
+    movu            [r2 + r3], m1
+    movu            [r2 + r3 * 2], m2
+    movu            [r2 + r5], m3
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], m4
+    movu            [r2 + r3], m5
+    movu            [r2 + r3 * 2], m6
+    movu            [r2 + r5], m7
+%endif
+    lea             r2, [r2 + r3 * 4]
+
+    movu            xm11, [r0 + r4]                 ; m11 = row 11
+    punpckhbw       xm6, xm10, xm11
+    punpcklbw       xm10, xm11
+    vinserti128     m10, m10, xm6, 1
+    pmaddubsw       m6, m10, m13
+    paddw           m8, m6
+    pmaddubsw       m10, m12
+    lea             r0, [r0 + r1 * 4]
+    movu            xm6, [r0]                       ; m6 = row 12
+    punpckhbw       xm7, xm11, xm6
+    punpcklbw       xm11, xm6
+    vinserti128     m11, m11, xm7, 1
+    pmaddubsw       m7, m11, m13
+    paddw           m9, m7
+    pmaddubsw       m11, m12
+
+    movu            xm7, [r0 + r1]                  ; m7 = row 13
+    punpckhbw       xm0, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm0, 1
+    pmaddubsw       m0, m6, m13
+    paddw           m10, m0
+    pmaddubsw       m6, m12
+    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
+    punpckhbw       xm1, xm7, xm0
+    punpcklbw       xm7, xm0
+    vinserti128     m7, m7, xm1, 1
+    pmaddubsw       m1, m7, m13
+    paddw           m11, m1
+    pmaddubsw       m7, m12
+    movu            xm1, [r0 + r4]                  ; m1 = row 15
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddubsw       m2, m0, m13
+    paddw           m6, m2
+    pmaddubsw       m0, m12
+    lea             r0, [r0 + r1 * 4]
+    movu            xm2, [r0]                       ; m2 = row 16
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m3, m1, m13
+    paddw           m7, m3
+    pmaddubsw       m1, m12
+    movu            xm3, [r0 + r1]                  ; m3 = row 17
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m4, m2, m13
+    paddw           m0, m4
+    pmaddubsw       m2, m12
+    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddubsw       m5, m3, m13
+    paddw           m1, m5
+    pmaddubsw       m3, m12
+
+%ifidn %1,pp
+    pmulhrsw        m8, m14                         ; m8 = word: row 8
+    pmulhrsw        m9, m14                         ; m9 = word: row 9
+    pmulhrsw        m10, m14                        ; m10 = word: row 10
+    pmulhrsw        m11, m14                        ; m11 = word: row 11
+    pmulhrsw        m6, m14                         ; m6 = word: row 12
+    pmulhrsw        m7, m14                         ; m7 = word: row 13
+    pmulhrsw        m0, m14                         ; m0 = word: row 14
+    pmulhrsw        m1, m14                         ; m1 = word: row 15
+    packuswb        m8, m9
+    packuswb        m10, m11
+    packuswb        m6, m7
+    packuswb        m0, m1
+    vpermq          m8, m8, q3120
+    vpermq          m10, m10, q3120
+    vpermq          m6, m6, q3120
+    vpermq          m0, m0, q3120
+    vextracti128    xm9, m8, 1
+    vextracti128    xm11, m10, 1
+    vextracti128    xm7, m6, 1
+    vextracti128    xm1, m0, 1
+    movu            [r2], xm8
+    movu            [r2 + r3], xm9
+    movu            [r2 + r3 * 2], xm10
+    movu            [r2 + r5], xm11
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm6
+    movu            [r2 + r3], xm7
+    movu            [r2 + r3 * 2], xm0
+    movu            [r2 + r5], xm1
+%else
+    psubw           m8, m14                         ; m8 = word: row 8
+    psubw           m9, m14                         ; m9 = word: row 9
+    psubw           m10, m14                        ; m10 = word: row 10
+    psubw           m11, m14                        ; m11 = word: row 11
+    psubw           m6, m14                         ; m6 = word: row 12
+    psubw           m7, m14                         ; m7 = word: row 13
+    psubw           m0, m14                         ; m0 = word: row 14
+    psubw           m1, m14                         ; m1 = word: row 15
+    movu            [r2], m8
+    movu            [r2 + r3], m9
+    movu            [r2 + r3 * 2], m10
+    movu            [r2 + r5], m11
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], m6
+    movu            [r2 + r3], m7
+    movu            [r2 + r3 * 2], m0
+    movu            [r2 + r5], m1
+%endif
+    lea             r2, [r2 + r3 * 4]
+
+    movu            xm5, [r0 + r4]                  ; m5 = row 19
+    punpckhbw       xm6, xm4, xm5
+    punpcklbw       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddubsw       m6, m4, m13
+    paddw           m2, m6
+    pmaddubsw       m4, m12
+    lea             r0, [r0 + r1 * 4]
+    movu            xm6, [r0]                       ; m6 = row 20
+    punpckhbw       xm7, xm5, xm6
+    punpcklbw       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddubsw       m7, m5, m13
+    paddw           m3, m7
+    pmaddubsw       m5, m12
+    movu            xm7, [r0 + r1]                  ; m7 = row 21
+    punpckhbw       xm0, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm0, 1
+    pmaddubsw       m0, m6, m13
+    paddw           m4, m0
+    pmaddubsw       m6, m12
+    movu            xm0, [r0 + r1 * 2]              ; m0 = row 22
+    punpckhbw       xm1, xm7, xm0
+    punpcklbw       xm7, xm0
+    vinserti128     m7, m7, xm1, 1
+    pmaddubsw       m1, m7, m13
+    paddw           m5, m1
+    pmaddubsw       m7, m12
+    movu            xm1, [r0 + r4]                  ; m1 = row 23
+    punpckhbw       xm8, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm8, 1
+    pmaddubsw       m8, m0, m13
+    paddw           m6, m8
+    pmaddubsw       m0, m12
+    lea             r0, [r0 + r1 * 4]
+    movu            xm8, [r0]                       ; m8 = row 24
+    punpckhbw       xm9, xm1, xm8
+    punpcklbw       xm1, xm8
+    vinserti128     m1, m1, xm9, 1
+    pmaddubsw       m9, m1, m13
+    paddw           m7, m9
+    pmaddubsw       m1, m12
+    movu            xm9, [r0 + r1]                  ; m9 = row 25
+    punpckhbw       xm10, xm8, xm9
+    punpcklbw       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddubsw       m8, m13
+    paddw           m0, m8
+    movu            xm10, [r0 + r1 * 2]             ; m10 = row 26
+    punpckhbw       xm11, xm9, xm10
+    punpcklbw       xm9, xm10
+    vinserti128     m9, m9, xm11, 1
+    pmaddubsw       m9, m13
+    paddw           m1, m9
+
+%ifidn %1,pp
+    pmulhrsw        m2, m14                         ; m2 = word: row 16
+    pmulhrsw        m3, m14                         ; m3 = word: row 17
+    pmulhrsw        m4, m14                         ; m4 = word: row 18
+    pmulhrsw        m5, m14                         ; m5 = word: row 19
+    pmulhrsw        m6, m14                         ; m6 = word: row 20
+    pmulhrsw        m7, m14                         ; m7 = word: row 21
+    pmulhrsw        m0, m14                         ; m0 = word: row 22
+    pmulhrsw        m1, m14                         ; m1 = word: row 23
+    packuswb        m2, m3
+    packuswb        m4, m5
+    packuswb        m6, m7
+    packuswb        m0, m1
+    vpermq          m2, m2, q3120
+    vpermq          m4, m4, q3120
+    vpermq          m6, m6, q3120
+    vpermq          m0, m0, q3120
+    vextracti128    xm3, m2, 1
+    vextracti128    xm5, m4, 1
+    vextracti128    xm7, m6, 1
+    vextracti128    xm1, m0, 1
+    movu            [r2], xm2
+    movu            [r2 + r3], xm3
+    movu            [r2 + r3 * 2], xm4
+    movu            [r2 + r5], xm5
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm6
+    movu            [r2 + r3], xm7
+    movu            [r2 + r3 * 2], xm0
+    movu            [r2 + r5], xm1
+%else
+    psubw           m2, m14                         ; m2 = word: row 16
+    psubw           m3, m14                         ; m3 = word: row 17
+    psubw           m4, m14                         ; m4 = word: row 18
+    psubw           m5, m14                         ; m5 = word: row 19
+    psubw           m6, m14                         ; m6 = word: row 20
+    psubw           m7, m14                         ; m7 = word: row 21
+    psubw           m0, m14                         ; m0 = word: row 22
+    psubw           m1, m14                         ; m1 = word: row 23
+    movu            [r2], m2
+    movu            [r2 + r3], m3
+    movu            [r2 + r3 * 2], m4
+    movu            [r2 + r5], m5
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], m6
+    movu            [r2 + r3], m7
+    movu            [r2 + r3 * 2], m0
+    movu            [r2 + r5], m1
+%endif
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_16x24 pp
+    FILTER_VER_CHROMA_AVX2_16x24 ps
+
+%macro FILTER_VER_CHROMA_AVX2_24x32 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_vert_%1_24x32, 4, 9, 10
+    mov             r4d, r4m
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    mova            m8, [r5]
+    mova            m9, [r5 + mmsize]
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,pp
+    mova            m7, [pw_512]
+%else
+    add             r3d, r3d
+    vbroadcasti128  m7, [pw_2000]
+%endif
+    lea             r6, [r3 * 3]
+    mov             r5d, 2
+.loopH:
+    movu            xm0, [r0]
+    vinserti128     m0, m0, [r0 + r1 * 2], 1
+    movu            xm1, [r0 + r1]
+    vinserti128     m1, m1, [r0 + r4], 1
+
+    punpcklbw       m2, m0, m1
+    punpckhbw       m3, m0, m1
+    vperm2i128      m4, m2, m3, 0x20
+    vperm2i128      m2, m2, m3, 0x31
+    pmaddubsw       m4, m8
+    pmaddubsw       m3, m2, m9
+    paddw           m4, m3
+    pmaddubsw       m2, m8
+
+    vextracti128    xm0, m0, 1
+    lea             r7, [r0 + r1 * 4]
+    vinserti128     m0, m0, [r7], 1
+
+    punpcklbw       m5, m1, m0
+    punpckhbw       m3, m1, m0
+    vperm2i128      m6, m5, m3, 0x20
+    vperm2i128      m5, m5, m3, 0x31
+    pmaddubsw       m6, m8
+    pmaddubsw       m3, m5, m9
+    paddw           m6, m3
+    pmaddubsw       m5, m8
+%ifidn %1,pp
+    pmulhrsw        m4, m7                         ; m4 = word: row 0
+    pmulhrsw        m6, m7                         ; m6 = word: row 1
+    packuswb        m4, m6
+    vpermq          m4, m4, 11011000b
+    vextracti128    xm6, m4, 1
+    movu            [r2], xm4
+    movu            [r2 + r3], xm6
+%else
+    psubw           m4, m7                         ; m4 = word: row 0
+    psubw           m6, m7                         ; m6 = word: row 1
+    movu            [r2], m4
+    movu            [r2 + r3], m6
+%endif
+
+    movu            xm4, [r7 + r1 * 2]
+    vinserti128     m4, m4, [r7 + r1], 1
+    vextracti128    xm1, m4, 1
+    vinserti128     m0, m0, xm1, 0
+
+    punpcklbw       m6, m0, m4
+    punpckhbw       m1, m0, m4
+    vperm2i128      m0, m6, m1, 0x20
+    vperm2i128      m6, m6, m1, 0x31
+    pmaddubsw       m1, m0, m9
+    paddw           m5, m1
+    pmaddubsw       m0, m8
+    pmaddubsw       m1, m6, m9
+    paddw           m2, m1
+    pmaddubsw       m6, m8
+
+%ifidn %1,pp
+    pmulhrsw        m2, m7                         ; m2 = word: row 2
+    pmulhrsw        m5, m7                         ; m5 = word: row 3
+    packuswb        m2, m5
+    vpermq          m2, m2, 11011000b
+    vextracti128    xm5, m2, 1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm5
+%else
+    psubw           m2, m7                         ; m2 = word: row 2
+    psubw           m5, m7                         ; m5 = word: row 3
+    movu            [r2 + r3 * 2], m2
+    movu            [r2 + r6], m5
+%endif
+    lea             r8, [r2 + r3 * 4]
+
+    movu            xm1, [r7 + r4]
+    lea             r7, [r7 + r1 * 4]
+    vinserti128     m1, m1, [r7], 1
+    vinserti128     m4, m4, xm1, 1
+
+    punpcklbw       m2, m4, m1
+    punpckhbw       m5, m4, m1
+    vperm2i128      m3, m2, m5, 0x20
+    vperm2i128      m2, m2, m5, 0x31
+    pmaddubsw       m5, m3, m9
+    paddw           m6, m5
+    pmaddubsw       m3, m8
+    pmaddubsw       m5, m2, m9
+    paddw           m0, m5
+    pmaddubsw       m2, m8
+
+%ifidn %1,pp
+    pmulhrsw        m6, m7                         ; m6 = word: row 4
+    pmulhrsw        m0, m7                         ; m0 = word: row 5
+    packuswb        m6, m0
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm0, m6, 1
+    movu            [r8], xm6
+    movu            [r8 + r3], xm0
+%else
+    psubw           m6, m7                         ; m6 = word: row 4
+    psubw           m0, m7                         ; m0 = word: row 5
+    movu            [r8], m6
+    movu            [r8 + r3], m0
+%endif
+
+    movu            xm6, [r7 + r1 * 2]
+    vinserti128     m6, m6, [r7 + r1], 1
+    vextracti128    xm0, m6, 1
+    vinserti128     m1, m1, xm0, 0
+
+    punpcklbw       m4, m1, m6
+    punpckhbw       m5, m1, m6
+    vperm2i128      m0, m4, m5, 0x20
+    vperm2i128      m5, m4, m5, 0x31
+    pmaddubsw       m4, m0, m9
+    paddw           m2, m4
+    pmaddubsw       m0, m8
+    pmaddubsw       m4, m5, m9
+    paddw           m3, m4
+    pmaddubsw       m5, m8
+
+%ifidn %1,pp
+    pmulhrsw        m3, m7                         ; m3 = word: row 6
+    pmulhrsw        m2, m7                         ; m2 = word: row 7
+    packuswb        m3, m2
+    vpermq          m3, m3, 11011000b
+    vextracti128    xm2, m3, 1
+    movu            [r8 + r3 * 2], xm3
+    movu            [r8 + r6], xm2
+%else
+    psubw           m3, m7                         ; m3 = word: row 6
+    psubw           m2, m7                         ; m2 = word: row 7
+    movu            [r8 + r3 * 2], m3
+    movu            [r8 + r6], m2
+%endif
+    lea             r8, [r8 + r3 * 4]
+
+    movu            xm3, [r7 + r4]
+    lea             r7, [r7 + r1 * 4]
+    vinserti128     m3, m3, [r7], 1
+    vinserti128     m6, m6, xm3, 1
+
+    punpcklbw       m2, m6, m3
+    punpckhbw       m1, m6, m3
+    vperm2i128      m4, m2, m1, 0x20
+    vperm2i128      m2, m2, m1, 0x31
+    pmaddubsw       m1, m4, m9
+    paddw           m5, m1
+    pmaddubsw       m4, m8
+    pmaddubsw       m1, m2, m9
+    paddw           m0, m1
+    pmaddubsw       m2, m8
+
+%ifidn %1,pp
+    pmulhrsw        m5, m7                         ; m5 = word: row 8
+    pmulhrsw        m0, m7                         ; m0 = word: row 9
+    packuswb        m5, m0
+    vpermq          m5, m5, 11011000b
+    vextracti128    xm0, m5, 1
+    movu            [r8], xm5
+    movu            [r8 + r3], xm0
+%else
+    psubw           m5, m7                         ; m5 = word: row 8
+    psubw           m0, m7                         ; m0 = word: row 9
+    movu            [r8], m5
+    movu            [r8 + r3], m0
+%endif
+
+    movu            xm5, [r7 + r1 * 2]
+    vinserti128     m5, m5, [r7 + r1], 1
+    vextracti128    xm0, m5, 1
+    vinserti128     m3, m3, xm0, 0
+
+    punpcklbw       m1, m3, m5
+    punpckhbw       m0, m3, m5
+    vperm2i128      m6, m1, m0, 0x20
+    vperm2i128      m0, m1, m0, 0x31
+    pmaddubsw       m1, m6, m9
+    paddw           m2, m1
+    pmaddubsw       m6, m8
+    pmaddubsw       m1, m0, m9
+    paddw           m4, m1
+    pmaddubsw       m0, m8
+
+%ifidn %1,pp
+    pmulhrsw        m4, m7                         ; m4 = word: row 10
+    pmulhrsw        m2, m7                         ; m2 = word: row 11
+    packuswb        m4, m2
+    vpermq          m4, m4, 11011000b
+    vextracti128    xm2, m4, 1
+    movu            [r8 + r3 * 2], xm4
+    movu            [r8 + r6], xm2
+%else
+    psubw           m4, m7                         ; m4 = word: row 10
+    psubw           m2, m7                         ; m2 = word: row 11
+    movu            [r8 + r3 * 2], m4
+    movu            [r8 + r6], m2
+%endif
+    lea             r8, [r8 + r3 * 4]
+
+    movu            xm3, [r7 + r4]
+    lea             r7, [r7 + r1 * 4]
+    vinserti128     m3, m3, [r7], 1
+    vinserti128     m5, m5, xm3, 1
+
+    punpcklbw       m2, m5, m3
+    punpckhbw       m1, m5, m3
+    vperm2i128      m4, m2, m1, 0x20
+    vperm2i128      m2, m2, m1, 0x31
+    pmaddubsw       m1, m4, m9
+    paddw           m0, m1
+    pmaddubsw       m4, m8
+    pmaddubsw       m1, m2, m9
+    paddw           m6, m1
+    pmaddubsw       m2, m8
+
+%ifidn %1,pp
+    pmulhrsw        m0, m7                         ; m0 = word: row 12
+    pmulhrsw        m6, m7                         ; m6 = word: row 13
+    packuswb        m0, m6
+    vpermq          m0, m0, 11011000b
+    vextracti128    xm6, m0, 1
+    movu            [r8], xm0
+    movu            [r8 + r3], xm6
+%else
+    psubw           m0, m7                         ; m0 = word: row 12
+    psubw           m6, m7                         ; m6 = word: row 13
+    movu            [r8], m0
+    movu            [r8 + r3], m6
+%endif
+
+    movu            xm5, [r7 + r1 * 2]
+    vinserti128     m5, m5, [r7 + r1], 1
+    vextracti128    xm0, m5, 1
+    vinserti128     m3, m3, xm0, 0
+
+    punpcklbw       m1, m3, m5
+    punpckhbw       m0, m3, m5
+    vperm2i128      m6, m1, m0, 0x20
+    vperm2i128      m0, m1, m0, 0x31
+    pmaddubsw       m6, m9
+    paddw           m2, m6
+    pmaddubsw       m0, m9
+    paddw           m4, m0
+
+%ifidn %1,pp
+    pmulhrsw        m4, m7                         ; m4 = word: row 14
+    pmulhrsw        m2, m7                         ; m2 = word: row 15
+    packuswb        m4, m2
+    vpermq          m4, m4, 11011000b
+    vextracti128    xm2, m4, 1
+    movu            [r8 + r3 * 2], xm4
+    movu            [r8 + r6], xm2
+    add             r2, 16
+%else
+    psubw           m4, m7                         ; m4 = word: row 14
+    psubw           m2, m7                         ; m2 = word: row 15
+    movu            [r8 + r3 * 2], m4
+    movu            [r8 + r6], m2
+    add             r2, 32
+%endif
+    add             r0, 16
+    movq            xm1, [r0]                       ; m1 = row 0
+    movq            xm2, [r0 + r1]                  ; m2 = row 1
+    punpcklbw       xm1, xm2
+    movq            xm3, [r0 + r1 * 2]              ; m3 = row 2
+    punpcklbw       xm2, xm3
+    vinserti128     m5, m1, xm2, 1
+    pmaddubsw       m5, m8
+    movq            xm4, [r0 + r4]                  ; m4 = row 3
+    punpcklbw       xm3, xm4
+    lea             r7, [r0 + r1 * 4]
+    movq            xm1, [r7]                       ; m1 = row 4
+    punpcklbw       xm4, xm1
+    vinserti128     m2, m3, xm4, 1
+    pmaddubsw       m0, m2, m9
+    paddw           m5, m0
+    pmaddubsw       m2, m8
+    movq            xm3, [r7 + r1]                  ; m3 = row 5
+    punpcklbw       xm1, xm3
+    movq            xm4, [r7 + r1 * 2]              ; m4 = row 6
+    punpcklbw       xm3, xm4
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m0, m1, m9
+    paddw           m2, m0
+    pmaddubsw       m1, m8
+    movq            xm3, [r7 + r4]                  ; m3 = row 7
+    punpcklbw       xm4, xm3
+    lea             r7, [r7 + r1 * 4]
+    movq            xm0, [r7]                       ; m0 = row 8
+    punpcklbw       xm3, xm0
+    vinserti128     m4, m4, xm3, 1
+    pmaddubsw       m3, m4, m9
+    paddw           m1, m3
+    pmaddubsw       m4, m8
+    movq            xm3, [r7 + r1]                  ; m3 = row 9
+    punpcklbw       xm0, xm3
+    movq            xm6, [r7 + r1 * 2]              ; m6 = row 10
+    punpcklbw       xm3, xm6
+    vinserti128     m0, m0, xm3, 1
+    pmaddubsw       m3, m0, m9
+    paddw           m4, m3
+    pmaddubsw       m0, m8
+
+%ifidn %1,pp
+    pmulhrsw        m5, m7                          ; m5 = word: row 0, row 1
+    pmulhrsw        m2, m7                          ; m2 = word: row 2, row 3
+    pmulhrsw        m1, m7                          ; m1 = word: row 4, row 5
+    pmulhrsw        m4, m7                          ; m4 = word: row 6, row 7
+    packuswb        m5, m2
+    packuswb        m1, m4
+    vextracti128    xm2, m5, 1
+    vextracti128    xm4, m1, 1
+    movq            [r2], xm5
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm5
+    movhps          [r2 + r6], xm2
+    lea             r8, [r2 + r3 * 4]
+    movq            [r8], xm1
+    movq            [r8 + r3], xm4
+    movhps          [r8 + r3 * 2], xm1
+    movhps          [r8 + r6], xm4
+%else
+    psubw           m5, m7                          ; m5 = word: row 0, row 1
+    psubw           m2, m7                          ; m2 = word: row 2, row 3
+    psubw           m1, m7                          ; m1 = word: row 4, row 5
+    psubw           m4, m7                          ; m4 = word: row 6, row 7
+    vextracti128    xm3, m5, 1
+    movu            [r2], xm5
+    movu            [r2 + r3], xm3
+    vextracti128    xm3, m2, 1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
+    vextracti128    xm3, m1, 1
+    lea             r8, [r2 + r3 * 4]
+    movu            [r8], xm1
+    movu            [r8 + r3], xm3
+    vextracti128    xm3, m4, 1
+    movu            [r8 + r3 * 2], xm4
+    movu            [r8 + r6], xm3
+%endif
+    lea             r8, [r8 + r3 * 4]
+
+    movq            xm3, [r7 + r4]                  ; m3 = row 11
+    punpcklbw       xm6, xm3
+    lea             r7, [r7 + r1 * 4]
+    movq            xm5, [r7]                       ; m5 = row 12
+    punpcklbw       xm3, xm5
+    vinserti128     m6, m6, xm3, 1
+    pmaddubsw       m3, m6, m9
+    paddw           m0, m3
+    pmaddubsw       m6, m8
+    movq            xm3, [r7 + r1]                  ; m3 = row 13
+    punpcklbw       xm5, xm3
+    movq            xm2, [r7 + r1 * 2]              ; m2 = row 14
+    punpcklbw       xm3, xm2
+    vinserti128     m5, m5, xm3, 1
+    pmaddubsw       m3, m5, m9
+    paddw           m6, m3
+    pmaddubsw       m5, m8
+    movq            xm3, [r7 + r4]                  ; m3 = row 15
+    punpcklbw       xm2, xm3
+    lea             r7, [r7 + r1 * 4]
+    movq            xm1, [r7]                       ; m1 = row 16
+    punpcklbw       xm3, xm1
+    vinserti128     m2, m2, xm3, 1
+    pmaddubsw       m3, m2, m9
+    paddw           m5, m3
+    pmaddubsw       m2, m8
+    movq            xm3, [r7 + r1]                  ; m3 = row 17
+    punpcklbw       xm1, xm3
+    movq            xm4, [r7 + r1 * 2]              ; m4 = row 18
+    punpcklbw       xm3, xm4
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m3, m1, m9
+    paddw           m2, m3
+%ifidn %1,pp
+    pmulhrsw        m0, m7                          ; m0 = word: row 8, row 9
+    pmulhrsw        m6, m7                          ; m6 = word: row 10, row 11
+    pmulhrsw        m5, m7                          ; m5 = word: row 12, row 13
+    pmulhrsw        m2, m7                          ; m2 = word: row 14, row 15
+    packuswb        m0, m6
+    packuswb        m5, m2
+    vextracti128    xm6, m0, 1
+    vextracti128    xm2, m5, 1
+    movq            [r8], xm0
+    movq            [r8 + r3], xm6
+    movhps          [r8 + r3 * 2], xm0
+    movhps          [r8 + r6], xm6
+    lea             r8, [r8 + r3 * 4]
+    movq            [r8], xm5
+    movq            [r8 + r3], xm2
+    movhps          [r8 + r3 * 2], xm5
+    movhps          [r8 + r6], xm2
+    lea             r2, [r8 + r3 * 4 - 16]
+%else
+    psubw           m0, m7                          ; m0 = word: row 8, row 9
+    psubw           m6, m7                          ; m6 = word: row 10, row 11
+    psubw           m5, m7                          ; m5 = word: row 12, row 13
+    psubw           m2, m7                          ; m2 = word: row 14, row 15
+    vextracti128    xm3, m0, 1
+    movu            [r8], xm0
+    movu            [r8 + r3], xm3
+    vextracti128    xm3, m6, 1
+    movu            [r8 + r3 * 2], xm6
+    movu            [r8 + r6], xm3
+    vextracti128    xm3, m5, 1
+    lea             r8, [r8 + r3 * 4]
+    movu            [r8], xm5
+    movu            [r8 + r3], xm3
+    vextracti128    xm3, m2, 1
+    movu            [r8 + r3 * 2], xm2
+    movu            [r8 + r6], xm3
+    lea             r2, [r8 + r3 * 4 - 32]
+%endif
+    lea             r0, [r7 - 16]
+    dec             r5d
+    jnz             .loopH
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_24x32 pp
+    FILTER_VER_CHROMA_AVX2_24x32 ps
+
+%macro FILTER_VER_CHROMA_AVX2_24x64 1
+%if ARCH_X86_64 == 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_24x64, 4, 7, 13
+    mov             r4d, r4m
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    mova            m10, [r5]
+    mova            m11, [r5 + mmsize]
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,pp
+    mova            m12, [pw_512]
+%else
+    add             r3d, r3d
+    vbroadcasti128  m12, [pw_2000]
+%endif
+    lea             r5, [r3 * 3]
+    mov             r6d, 16
+.loopH:
+    movu            m0, [r0]                        ; m0 = row 0
+    movu            m1, [r0 + r1]                   ; m1 = row 1
+    punpcklbw       m2, m0, m1
+    punpckhbw       m3, m0, m1
+    pmaddubsw       m2, m10
+    pmaddubsw       m3, m10
+    movu            m0, [r0 + r1 * 2]               ; m0 = row 2
+    punpcklbw       m4, m1, m0
+    punpckhbw       m5, m1, m0
+    pmaddubsw       m4, m10
+    pmaddubsw       m5, m10
+    movu            m1, [r0 + r4]                   ; m1 = row 3
+    punpcklbw       m6, m0, m1
+    punpckhbw       m7, m0, m1
+    pmaddubsw       m8, m6, m11
+    pmaddubsw       m9, m7, m11
+    pmaddubsw       m6, m10
+    pmaddubsw       m7, m10
+    paddw           m2, m8
+    paddw           m3, m9
+%ifidn %1,pp
+    pmulhrsw        m2, m12
+    pmulhrsw        m3, m12
+    packuswb        m2, m3
+    movu            [r2], xm2
+    vextracti128    xm2, m2, 1
+    movq            [r2 + 16], xm2
+%else
+    psubw           m2, m12
+    psubw           m3, m12
+    vperm2i128      m0, m2, m3, 0x20
+    vperm2i128      m2, m2, m3, 0x31
+    movu            [r2], m0
+    movu            [r2 + mmsize], xm2
+%endif
+    lea             r0, [r0 + r1 * 4]
+    movu            m0, [r0]                        ; m0 = row 4
+    punpcklbw       m2, m1, m0
+    punpckhbw       m3, m1, m0
+    pmaddubsw       m8, m2, m11
+    pmaddubsw       m9, m3, m11
+    pmaddubsw       m2, m10
+    pmaddubsw       m3, m10
+    paddw           m4, m8
+    paddw           m5, m9
+%ifidn %1,pp
+    pmulhrsw        m4, m12
+    pmulhrsw        m5, m12
+    packuswb        m4, m5
+    movu            [r2 + r3], xm4
+    vextracti128    xm4, m4, 1
+    movq            [r2 + r3 + 16], xm4
+%else
+    psubw           m4, m12
+    psubw           m5, m12
+    vperm2i128      m1, m4, m5, 0x20
+    vperm2i128      m4, m4, m5, 0x31
+    movu            [r2 + r3], m1
+    movu            [r2 + r3 + mmsize], xm4
+%endif
+
+    movu            m1, [r0 + r1]                   ; m1 = row 5
+    punpcklbw       m4, m0, m1
+    punpckhbw       m5, m0, m1
+    pmaddubsw       m4, m11
+    pmaddubsw       m5, m11
+    paddw           m6, m4
+    paddw           m7, m5
+%ifidn %1,pp
+    pmulhrsw        m6, m12
+    pmulhrsw        m7, m12
+    packuswb        m6, m7
+    movu            [r2 + r3 * 2], xm6
+    vextracti128    xm6, m6, 1
+    movq            [r2 + r3 * 2 + 16], xm6
+%else
+    psubw           m6, m12
+    psubw           m7, m12
+    vperm2i128      m0, m6, m7, 0x20
+    vperm2i128      m6, m6, m7, 0x31
+    movu            [r2 + r3 * 2], m0
+    movu            [r2 + r3 * 2 + mmsize], xm6
+%endif
+
+    movu            m0, [r0 + r1 * 2]               ; m0 = row 6
+    punpcklbw       m6, m1, m0
+    punpckhbw       m7, m1, m0
+    pmaddubsw       m6, m11
+    pmaddubsw       m7, m11
+    paddw           m2, m6
+    paddw           m3, m7
+%ifidn %1,pp
+    pmulhrsw        m2, m12
+    pmulhrsw        m3, m12
+    packuswb        m2, m3
+    movu            [r2 + r5], xm2
+    vextracti128    xm2, m2, 1
+    movq            [r2 + r5 + 16], xm2
+%else
+    psubw           m2, m12
+    psubw           m3, m12
+    vperm2i128      m0, m2, m3, 0x20
+    vperm2i128      m2, m2, m3, 0x31
+    movu            [r2 + r5], m0
+    movu            [r2 + r5 + mmsize], xm2
+%endif
+    lea             r2, [r2 + r3 * 4]
+    dec             r6d
+    jnz             .loopH
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_24x64 pp
+    FILTER_VER_CHROMA_AVX2_24x64 ps
+
+%macro FILTER_VER_CHROMA_AVX2_16x4 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_16x4, 4, 6, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,pp
+    mova            m7, [pw_512]
+%else
+    add             r3d, r3d
+    mova            m7, [pw_2000]
+%endif
+
+    movu            xm0, [r0]
+    vinserti128     m0, m0, [r0 + r1 * 2], 1
+    movu            xm1, [r0 + r1]
+    vinserti128     m1, m1, [r0 + r4], 1
+
+    punpcklbw       m2, m0, m1
+    punpckhbw       m3, m0, m1
+    vperm2i128      m4, m2, m3, 0x20
+    vperm2i128      m2, m2, m3, 0x31
+    pmaddubsw       m4, [r5]
+    pmaddubsw       m3, m2, [r5 + mmsize]
+    paddw           m4, m3
+    pmaddubsw       m2, [r5]
+
+    vextracti128    xm0, m0, 1
+    lea             r0, [r0 + r1 * 4]
+    vinserti128     m0, m0, [r0], 1
+
+    punpcklbw       m5, m1, m0
+    punpckhbw       m3, m1, m0
+    vperm2i128      m6, m5, m3, 0x20
+    vperm2i128      m5, m5, m3, 0x31
+    pmaddubsw       m6, [r5]
+    pmaddubsw       m3, m5, [r5 + mmsize]
+    paddw           m6, m3
+    pmaddubsw       m5, [r5]
+%ifidn %1,pp
+    pmulhrsw        m4, m7                          ; m4 = word: row 0
+    pmulhrsw        m6, m7                          ; m6 = word: row 1
+    packuswb        m4, m6
+    vpermq          m4, m4, 11011000b
+    vextracti128    xm6, m4, 1
+    movu            [r2], xm4
+    movu            [r2 + r3], xm6
+%else
+    psubw           m4, m7                          ; m4 = word: row 0
+    psubw           m6, m7                          ; m6 = word: row 1
+    movu            [r2], m4
+    movu            [r2 + r3], m6
+%endif
+    lea             r2, [r2 + r3 * 2]
+
+    movu            xm4, [r0 + r1 * 2]
+    vinserti128     m4, m4, [r0 + r1], 1
+    vextracti128    xm1, m4, 1
+    vinserti128     m0, m0, xm1, 0
+
+    punpcklbw       m6, m0, m4
+    punpckhbw       m1, m0, m4
+    vperm2i128      m0, m6, m1, 0x20
+    vperm2i128      m6, m6, m1, 0x31
+    pmaddubsw       m0, [r5 + mmsize]
+    paddw           m5, m0
+    pmaddubsw       m6, [r5 + mmsize]
+    paddw           m2, m6
+
+%ifidn %1,pp
+    pmulhrsw        m2, m7                          ; m2 = word: row 2
+    pmulhrsw        m5, m7                          ; m5 = word: row 3
+    packuswb        m2, m5
+    vpermq          m2, m2, 11011000b
+    vextracti128    xm5, m2, 1
+    movu            [r2], xm2
+    movu            [r2 + r3], xm5
+%else
+    psubw           m2, m7                          ; m2 = word: row 2
+    psubw           m5, m7                          ; m5 = word: row 3
+    movu            [r2], m2
+    movu            [r2 + r3], m5
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_16x4 pp
+    FILTER_VER_CHROMA_AVX2_16x4 ps
+
+%macro FILTER_VER_CHROMA_AVX2_12xN 2
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_12x%2, 4, 7, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,pp
+    mova            m7, [pw_512]
+%else
+    add             r3d, r3d
+    vbroadcasti128  m7, [pw_2000]
+%endif
+    lea             r6, [r3 * 3]
+%rep %2 / 16
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddubsw       m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
+    paddw           m0, m4
+    pmaddubsw       m2, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
+    paddw           m1, m5
+    pmaddubsw       m3, [r5]
+%ifidn %1,pp
+    pmulhrsw        m0, m7                          ; m0 = word: row 0
+    pmulhrsw        m1, m7                          ; m1 = word: row 1
+    packuswb        m0, m1
+    vextracti128    xm1, m0, 1
+    movq            [r2], xm0
+    movd            [r2 + 8], xm1
+    movhps          [r2 + r3], xm0
+    pextrd          [r2 + r3 + 8], xm1, 2
+%else
+    psubw           m0, m7                          ; m0 = word: row 0
+    psubw           m1, m7                          ; m1 = word: row 1
+    movu            [r2], xm0
+    vextracti128    xm0, m0, 1
+    movq            [r2 + 16], xm0
+    movu            [r2 + r3], xm1
+    vextracti128    xm1, m1, 1
+    movq            [r2 + r3 + 16], xm1
+%endif
+
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhbw       xm6, xm4, xm5
+    punpcklbw       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
+    paddw           m2, m6
+    pmaddubsw       m4, [r5]
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhbw       xm0, xm5, xm6
+    punpcklbw       xm5, xm6
+    vinserti128     m5, m5, xm0, 1
+    pmaddubsw       m0, m5, [r5 + 1 * mmsize]
+    paddw           m3, m0
+    pmaddubsw       m5, [r5]
+%ifidn %1,pp
+    pmulhrsw        m2, m7                          ; m2 = word: row 2
+    pmulhrsw        m3, m7                          ; m3 = word: row 3
+    packuswb        m2, m3
+    vextracti128    xm3, m2, 1
+    movq            [r2 + r3 * 2], xm2
+    movd            [r2 + r3 * 2 + 8], xm3
+    movhps          [r2 + r6], xm2
+    pextrd          [r2 + r6 + 8], xm3, 2
+%else
+    psubw           m2, m7                          ; m2 = word: row 2
+    psubw           m3, m7                          ; m3 = word: row 3
+    movu            [r2 + r3 * 2], xm2
+    vextracti128    xm2, m2, 1
+    movq            [r2 + r3 * 2 + 16], xm2
+    movu            [r2 + r6], xm3
+    vextracti128    xm3, m3, 1
+    movq            [r2 + r6 + 16], xm3
+%endif
+    lea             r2, [r2 + r3 * 4]
+
+    movu            xm0, [r0 + r4]                  ; m0 = row 7
+    punpckhbw       xm3, xm6, xm0
+    punpcklbw       xm6, xm0
+    vinserti128     m6, m6, xm3, 1
+    pmaddubsw       m3, m6, [r5 + 1 * mmsize]
+    paddw           m4, m3
+    pmaddubsw       m6, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm3, [r0]                       ; m3 = row 8
+    punpckhbw       xm1, xm0, xm3
+    punpcklbw       xm0, xm3
+    vinserti128     m0, m0, xm1, 1
+    pmaddubsw       m1, m0, [r5 + 1 * mmsize]
+    paddw           m5, m1
+    pmaddubsw       m0, [r5]
+%ifidn %1,pp
+    pmulhrsw        m4, m7                          ; m4 = word: row 4
+    pmulhrsw        m5, m7                          ; m5 = word: row 5
+    packuswb        m4, m5
+    vextracti128    xm5, m4, 1
+    movq            [r2], xm4
+    movd            [r2 + 8], xm5
+    movhps          [r2 + r3], xm4
+    pextrd          [r2 + r3 + 8], xm5, 2
+%else
+    psubw           m4, m7                          ; m4 = word: row 4
+    psubw           m5, m7                          ; m5 = word: row 5
+    movu            [r2], xm4
+    vextracti128    xm4, m4, 1
+    movq            [r2 + 16], xm4
+    movu            [r2 + r3], xm5
+    vextracti128    xm5, m5, 1
+    movq            [r2 + r3 + 16], xm5
+%endif
+
+    movu            xm1, [r0 + r1]                  ; m1 = row 9
+    punpckhbw       xm2, xm3, xm1
+    punpcklbw       xm3, xm1
+    vinserti128     m3, m3, xm2, 1
+    pmaddubsw       m2, m3, [r5 + 1 * mmsize]
+    paddw           m6, m2
+    pmaddubsw       m3, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 10
+    punpckhbw       xm4, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm4, 1
+    pmaddubsw       m4, m1, [r5 + 1 * mmsize]
+    paddw           m0, m4
+    pmaddubsw       m1, [r5]
+
+%ifidn %1,pp
+    pmulhrsw        m6, m7                          ; m6 = word: row 6
+    pmulhrsw        m0, m7                          ; m0 = word: row 7
+    packuswb        m6, m0
+    vextracti128    xm0, m6, 1
+    movq            [r2 + r3 * 2], xm6
+    movd            [r2 + r3 * 2 + 8], xm0
+    movhps          [r2 + r6], xm6
+    pextrd          [r2 + r6 + 8], xm0, 2
+%else
+    psubw           m6, m7                          ; m6 = word: row 6
+    psubw           m0, m7                          ; m0 = word: row 7
+    movu            [r2 + r3 * 2], xm6
+    vextracti128    xm6, m6, 1
+    movq            [r2 + r3 * 2 + 16], xm6
+    movu            [r2 + r6], xm0
+    vextracti128    xm0, m0, 1
+    movq            [r2 + r6 + 16], xm0
+%endif
+    lea             r2, [r2 + r3 * 4]
+
+    movu            xm4, [r0 + r4]                  ; m4 = row 11
+    punpckhbw       xm6, xm2, xm4
+    punpcklbw       xm2, xm4
+    vinserti128     m2, m2, xm6, 1
+    pmaddubsw       m6, m2, [r5 + 1 * mmsize]
+    paddw           m3, m6
+    pmaddubsw       m2, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm6, [r0]                       ; m6 = row 12
+    punpckhbw       xm0, xm4, xm6
+    punpcklbw       xm4, xm6
+    vinserti128     m4, m4, xm0, 1
+    pmaddubsw       m0, m4, [r5 + 1 * mmsize]
+    paddw           m1, m0
+    pmaddubsw       m4, [r5]
+%ifidn %1,pp
+    pmulhrsw        m3, m7                          ; m3 = word: row 8
+    pmulhrsw        m1, m7                          ; m1 = word: row 9
+    packuswb        m3, m1
+    vextracti128    xm1, m3, 1
+    movq            [r2], xm3
+    movd            [r2 + 8], xm1
+    movhps          [r2 + r3], xm3
+    pextrd          [r2 + r3 + 8], xm1, 2
+%else
+    psubw           m3, m7                          ; m3 = word: row 8
+    psubw           m1, m7                          ; m1 = word: row 9
+    movu            [r2], xm3
+    vextracti128    xm3, m3, 1
+    movq            [r2 + 16], xm3
+    movu            [r2 + r3], xm1
+    vextracti128    xm1, m1, 1
+    movq            [r2 + r3 + 16], xm1
+%endif
+
+    movu            xm0, [r0 + r1]                  ; m0 = row 13
+    punpckhbw       xm1, xm6, xm0
+    punpcklbw       xm6, xm0
+    vinserti128     m6, m6, xm1, 1
+    pmaddubsw       m1, m6, [r5 + 1 * mmsize]
+    paddw           m2, m1
+    pmaddubsw       m6, [r5]
+    movu            xm1, [r0 + r1 * 2]              ; m1 = row 14
+    punpckhbw       xm5, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm5, 1
+    pmaddubsw       m5, m0, [r5 + 1 * mmsize]
+    paddw           m4, m5
+    pmaddubsw       m0, [r5]
+%ifidn %1,pp
+    pmulhrsw        m2, m7                          ; m2 = word: row 10
+    pmulhrsw        m4, m7                          ; m4 = word: row 11
+    packuswb        m2, m4
+    vextracti128    xm4, m2, 1
+    movq            [r2 + r3 * 2], xm2
+    movd            [r2 + r3 * 2 + 8], xm4
+    movhps          [r2 + r6], xm2
+    pextrd          [r2 + r6 + 8], xm4, 2
+%else
+    psubw           m2, m7                          ; m2 = word: row 10
+    psubw           m4, m7                          ; m4 = word: row 11
+    movu            [r2 + r3 * 2], xm2
+    vextracti128    xm2, m2, 1
+    movq            [r2 + r3 * 2 + 16], xm2
+    movu            [r2 + r6], xm4
+    vextracti128    xm4, m4, 1
+    movq            [r2 + r6 + 16], xm4
+%endif
+    lea             r2, [r2 + r3 * 4]
+
+    movu            xm5, [r0 + r4]                  ; m5 = row 15
+    punpckhbw       xm2, xm1, xm5
+    punpcklbw       xm1, xm5
+    vinserti128     m1, m1, xm2, 1
+    pmaddubsw       m2, m1, [r5 + 1 * mmsize]
+    paddw           m6, m2
+    pmaddubsw       m1, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm2, [r0]                       ; m2 = row 16
+    punpckhbw       xm3, xm5, xm2
+    punpcklbw       xm5, xm2
+    vinserti128     m5, m5, xm3, 1
+    pmaddubsw       m3, m5, [r5 + 1 * mmsize]
+    paddw           m0, m3
+    pmaddubsw       m5, [r5]
+    movu            xm3, [r0 + r1]                  ; m3 = row 17
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m2, [r5 + 1 * mmsize]
+    paddw           m1, m2
+    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
+    punpckhbw       xm2, xm3, xm4
+    punpcklbw       xm3, xm4
+    vinserti128     m3, m3, xm2, 1
+    pmaddubsw       m3, [r5 + 1 * mmsize]
+    paddw           m5, m3
+
+%ifidn %1,pp
+    pmulhrsw        m6, m7                          ; m6 = word: row 12
+    pmulhrsw        m0, m7                          ; m0 = word: row 13
+    pmulhrsw        m1, m7                          ; m1 = word: row 14
+    pmulhrsw        m5, m7                          ; m5 = word: row 15
+    packuswb        m6, m0
+    packuswb        m1, m5
+    vextracti128    xm0, m6, 1
+    vextracti128    xm5, m1, 1
+    movq            [r2], xm6
+    movd            [r2 + 8], xm0
+    movhps          [r2 + r3], xm6
+    pextrd          [r2 + r3 + 8], xm0, 2
+    movq            [r2 + r3 * 2], xm1
+    movd            [r2 + r3 * 2 + 8], xm5
+    movhps          [r2 + r6], xm1
+    pextrd          [r2 + r6 + 8], xm5, 2
+%else
+    psubw           m6, m7                          ; m6 = word: row 12
+    psubw           m0, m7                          ; m0 = word: row 13
+    psubw           m1, m7                          ; m1 = word: row 14
+    psubw           m5, m7                          ; m5 = word: row 15
+    movu            [r2], xm6
+    vextracti128    xm6, m6, 1
+    movq            [r2 + 16], xm6
+    movu            [r2 + r3], xm0
+    vextracti128    xm0, m0, 1
+    movq            [r2 + r3 + 16], xm0
+    movu            [r2 + r3 * 2], xm1
+    vextracti128    xm1, m1, 1
+    movq            [r2 + r3 * 2 + 16], xm1
+    movu            [r2 + r6], xm5
+    vextracti128    xm5, m5, 1
+    movq            [r2 + r6 + 16], xm5
+%endif
+    lea             r2, [r2 + r3 * 4]
+%endrep
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_12xN pp, 16
+    FILTER_VER_CHROMA_AVX2_12xN ps, 16
+    FILTER_VER_CHROMA_AVX2_12xN pp, 32
+    FILTER_VER_CHROMA_AVX2_12xN ps, 32
+
+;-----------------------------------------------------------------------------
+;void interp_4tap_vert_pp_24x32(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W24 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_pp_24x%2, 4, 6, 8
+
+    mov         r4d,       r4m
+    sub         r0,        r1
+
+%ifdef PIC
+    lea         r5,        [tab_ChromaCoeff]
+    movd        m0,        [r5 + r4 * 4]
+%else
+    movd        m0,        [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb      m1,        m0,       [tab_Vm]
+    pshufb      m0,        [tab_Vm + 16]
+
+    mov         r4d,       %2
+
+.loop:
+    movu        m2,        [r0]
+    movu        m3,        [r0 + r1]
+
+    punpcklbw   m4,        m2,        m3
+    punpckhbw   m2,        m3
+
+    pmaddubsw   m4,        m1
+    pmaddubsw   m2,        m1
+
+    lea         r5,        [r0 + 2 * r1]
+    movu        m5,        [r5]
+    movu        m7,        [r5 + r1]
+
+    punpcklbw   m6,        m5,        m7
+    pmaddubsw   m6,        m0
+    paddw       m4,        m6
+
+    punpckhbw   m6,        m5,        m7
+    pmaddubsw   m6,        m0
+    paddw       m2,        m6
+
+    mova        m6,        [pw_512]
+
+    pmulhrsw    m4,        m6
+    pmulhrsw    m2,        m6
+
+    packuswb    m4,        m2
+
+    movu        [r2],      m4
+
+    punpcklbw   m4,        m3,        m5
+    punpckhbw   m3,        m5
+
+    pmaddubsw   m4,        m1
+    pmaddubsw   m3,        m1
+
+    movu        m2,        [r5 + 2 * r1]
+
+    punpcklbw   m5,        m7,        m2
+    punpckhbw   m7,        m2
+
+    pmaddubsw   m5,        m0
+    pmaddubsw   m7,        m0
+
+    paddw       m4,        m5
+    paddw       m3,        m7
+
+    pmulhrsw    m4,        m6
+    pmulhrsw    m3,        m6
+
+    packuswb    m4,        m3
+
+    movu        [r2 + r3],      m4
+
+    movq        m2,        [r0 + 16]
+    movq        m3,        [r0 + r1 + 16]
+    movq        m4,        [r5 + 16]
+    movq        m5,        [r5 + r1 + 16]
+
+    punpcklbw   m2,        m3
+    punpcklbw   m4,        m5
+
+    pmaddubsw   m2,        m1
+    pmaddubsw   m4,        m0
+
+    paddw       m2,        m4
+
+    pmulhrsw    m2,        m6
+
+    movq        m3,        [r0 + r1 + 16]
+    movq        m4,        [r5 + 16]
+    movq        m5,        [r5 + r1 + 16]
+    movq        m7,        [r5 + 2 * r1 + 16]
+
+    punpcklbw   m3,        m4
+    punpcklbw   m5,        m7
+
+    pmaddubsw   m3,        m1
+    pmaddubsw   m5,        m0
+
+    paddw       m3,        m5
+
+    pmulhrsw    m3,        m6
+    packuswb    m2,        m3
+
+    movh        [r2 + 16], m2
+    movhps      [r2 + r3 + 16], m2
+
+    mov         r0,        r5
+    lea         r2,        [r2 + 2 * r3]
+
+    sub         r4,        2
+    jnz        .loop
+    RET
+%endmacro
+
+    FILTER_V4_W24 24, 32
+
+    FILTER_V4_W24 24, 64
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_pp_32x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W32 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_pp_%1x%2, 4, 6, 8
+
+    mov         r4d,       r4m
+    sub         r0,        r1
+
+%ifdef PIC
+    lea         r5,        [tab_ChromaCoeff]
+    movd        m0,        [r5 + r4 * 4]
+%else
+    movd        m0,        [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb      m1,        m0,       [tab_Vm]
+    pshufb      m0,        [tab_Vm + 16]
+
+    mova        m7,        [pw_512]
+
+    mov         r4d,       %2
+
+.loop:
+    movu        m2,        [r0]
+    movu        m3,        [r0 + r1]
+
+    punpcklbw   m4,        m2,        m3
+    punpckhbw   m2,        m3
+
+    pmaddubsw   m4,        m1
+    pmaddubsw   m2,        m1
+
+    lea         r5,        [r0 + 2 * r1]
+    movu        m3,        [r5]
+    movu        m5,        [r5 + r1]
+
+    punpcklbw   m6,        m3,        m5
+    punpckhbw   m3,        m5
+
+    pmaddubsw   m6,        m0
+    pmaddubsw   m3,        m0
+
+    paddw       m4,        m6
+    paddw       m2,        m3
+
+    pmulhrsw    m4,        m7
+    pmulhrsw    m2,        m7
+
+    packuswb    m4,        m2
+
+    movu        [r2],      m4
+
+    movu        m2,        [r0 + 16]
+    movu        m3,        [r0 + r1 + 16]
+
+    punpcklbw   m4,        m2,        m3
+    punpckhbw   m2,        m3
+
+    pmaddubsw   m4,        m1
+    pmaddubsw   m2,        m1
+
+    movu        m3,        [r5 + 16]
+    movu        m5,        [r5 + r1 + 16]
+
+    punpcklbw   m6,        m3,        m5
+    punpckhbw   m3,        m5
+
+    pmaddubsw   m6,        m0
+    pmaddubsw   m3,        m0
+
+    paddw       m4,        m6
+    paddw       m2,        m3
+
+    pmulhrsw    m4,        m7
+    pmulhrsw    m2,        m7
+
+    packuswb    m4,        m2
+
+    movu        [r2 + 16], m4
+
+    lea         r0,        [r0 + r1]
+    lea         r2,        [r2 + r3]
+
+    dec         r4
+    jnz        .loop
+    RET
+%endmacro
+
+    FILTER_V4_W32 32,  8
+    FILTER_V4_W32 32, 16
+    FILTER_V4_W32 32, 24
+    FILTER_V4_W32 32, 32
+
+    FILTER_V4_W32 32, 48
+    FILTER_V4_W32 32, 64
+
+%macro FILTER_VER_CHROMA_AVX2_32xN 2
+%if ARCH_X86_64 == 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_32x%2, 4, 7, 13
+    mov             r4d, r4m
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    mova            m10, [r5]
+    mova            m11, [r5 + mmsize]
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,pp
+    mova            m12, [pw_512]
+%else
+    add             r3d, r3d
+    vbroadcasti128  m12, [pw_2000]
+%endif
+    lea             r5, [r3 * 3]
+    mov             r6d, %2 / 4
+.loopW:
+    movu            m0, [r0]                        ; m0 = row 0
+    movu            m1, [r0 + r1]                   ; m1 = row 1
+    punpcklbw       m2, m0, m1
+    punpckhbw       m3, m0, m1
+    pmaddubsw       m2, m10
+    pmaddubsw       m3, m10
+    movu            m0, [r0 + r1 * 2]               ; m0 = row 2
+    punpcklbw       m4, m1, m0
+    punpckhbw       m5, m1, m0
+    pmaddubsw       m4, m10
+    pmaddubsw       m5, m10
+    movu            m1, [r0 + r4]                   ; m1 = row 3
+    punpcklbw       m6, m0, m1
+    punpckhbw       m7, m0, m1
+    pmaddubsw       m8, m6, m11
+    pmaddubsw       m9, m7, m11
+    pmaddubsw       m6, m10
+    pmaddubsw       m7, m10
+    paddw           m2, m8
+    paddw           m3, m9
+%ifidn %1,pp
+    pmulhrsw        m2, m12
+    pmulhrsw        m3, m12
+    packuswb        m2, m3
+    movu            [r2], m2
+%else
+    psubw           m2, m12
+    psubw           m3, m12
+    vperm2i128      m0, m2, m3, 0x20
+    vperm2i128      m2, m2, m3, 0x31
+    movu            [r2], m0
+    movu            [r2 + mmsize], m2
+%endif
+    lea             r0, [r0 + r1 * 4]
+    movu            m0, [r0]                        ; m0 = row 4
+    punpcklbw       m2, m1, m0
+    punpckhbw       m3, m1, m0
+    pmaddubsw       m8, m2, m11
+    pmaddubsw       m9, m3, m11
+    pmaddubsw       m2, m10
+    pmaddubsw       m3, m10
+    paddw           m4, m8
+    paddw           m5, m9
+%ifidn %1,pp
+    pmulhrsw        m4, m12
+    pmulhrsw        m5, m12
+    packuswb        m4, m5
+    movu            [r2 + r3], m4
+%else
+    psubw           m4, m12
+    psubw           m5, m12
+    vperm2i128      m1, m4, m5, 0x20
+    vperm2i128      m4, m4, m5, 0x31
+    movu            [r2 + r3], m1
+    movu            [r2 + r3 + mmsize], m4
+%endif
+
+    movu            m1, [r0 + r1]                   ; m1 = row 5
+    punpcklbw       m4, m0, m1
+    punpckhbw       m5, m0, m1
+    pmaddubsw       m4, m11
+    pmaddubsw       m5, m11
+    paddw           m6, m4
+    paddw           m7, m5
+%ifidn %1,pp
+    pmulhrsw        m6, m12
+    pmulhrsw        m7, m12
+    packuswb        m6, m7
+    movu            [r2 + r3 * 2], m6
+%else
+    psubw           m6, m12
+    psubw           m7, m12
+    vperm2i128      m0, m6, m7, 0x20
+    vperm2i128      m6, m6, m7, 0x31
+    movu            [r2 + r3 * 2], m0
+    movu            [r2 + r3 * 2 + mmsize], m6
+%endif
+
+    movu            m0, [r0 + r1 * 2]               ; m0 = row 6
+    punpcklbw       m6, m1, m0
+    punpckhbw       m7, m1, m0
+    pmaddubsw       m6, m11
+    pmaddubsw       m7, m11
+    paddw           m2, m6
+    paddw           m3, m7
+%ifidn %1,pp
+    pmulhrsw        m2, m12
+    pmulhrsw        m3, m12
+    packuswb        m2, m3
+    movu            [r2 + r5], m2
+%else
+    psubw           m2, m12
+    psubw           m3, m12
+    vperm2i128      m0, m2, m3, 0x20
+    vperm2i128      m2, m2, m3, 0x31
+    movu            [r2 + r5], m0
+    movu            [r2 + r5 + mmsize], m2
+%endif
+    lea             r2, [r2 + r3 * 4]
+    dec             r6d
+    jnz             .loopW
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_32xN pp, 64
+    FILTER_VER_CHROMA_AVX2_32xN pp, 48
+    FILTER_VER_CHROMA_AVX2_32xN pp, 32
+    FILTER_VER_CHROMA_AVX2_32xN pp, 24
+    FILTER_VER_CHROMA_AVX2_32xN pp, 16
+    FILTER_VER_CHROMA_AVX2_32xN pp, 8
+    FILTER_VER_CHROMA_AVX2_32xN ps, 64
+    FILTER_VER_CHROMA_AVX2_32xN ps, 48
+    FILTER_VER_CHROMA_AVX2_32xN ps, 32
+    FILTER_VER_CHROMA_AVX2_32xN ps, 24
+    FILTER_VER_CHROMA_AVX2_32xN ps, 16
+    FILTER_VER_CHROMA_AVX2_32xN ps, 8
+
+%macro FILTER_VER_CHROMA_AVX2_48x64 1
+%if ARCH_X86_64 == 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_48x64, 4, 8, 13
+    mov             r4d, r4m
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    mova            m10, [r5]
+    mova            m11, [r5 + mmsize]
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,pp
+    mova            m12, [pw_512]
+%else
+    add             r3d, r3d
+    vbroadcasti128  m12, [pw_2000]
+%endif
+    lea             r5, [r3 * 3]
+    lea             r7, [r1 * 4]
+    mov             r6d, 16
+.loopH:
+    movu            m0, [r0]                        ; m0 = row 0
+    movu            m1, [r0 + r1]                   ; m1 = row 1
+    punpcklbw       m2, m0, m1
+    punpckhbw       m3, m0, m1
+    pmaddubsw       m2, m10
+    pmaddubsw       m3, m10
+    movu            m0, [r0 + r1 * 2]               ; m0 = row 2
+    punpcklbw       m4, m1, m0
+    punpckhbw       m5, m1, m0
+    pmaddubsw       m4, m10
+    pmaddubsw       m5, m10
+    movu            m1, [r0 + r4]                   ; m1 = row 3
+    punpcklbw       m6, m0, m1
+    punpckhbw       m7, m0, m1
+    pmaddubsw       m8, m6, m11
+    pmaddubsw       m9, m7, m11
+    pmaddubsw       m6, m10
+    pmaddubsw       m7, m10
+    paddw           m2, m8
+    paddw           m3, m9
+%ifidn %1,pp
+    pmulhrsw        m2, m12
+    pmulhrsw        m3, m12
+    packuswb        m2, m3
+    movu            [r2], m2
+%else
+    psubw           m2, m12
+    psubw           m3, m12
+    vperm2i128      m0, m2, m3, 0x20
+    vperm2i128      m2, m2, m3, 0x31
+    movu            [r2], m0
+    movu            [r2 + mmsize], m2
+%endif
+    lea             r0, [r0 + r1 * 4]
+    movu            m0, [r0]                        ; m0 = row 4
+    punpcklbw       m2, m1, m0
+    punpckhbw       m3, m1, m0
+    pmaddubsw       m8, m2, m11
+    pmaddubsw       m9, m3, m11
+    pmaddubsw       m2, m10
+    pmaddubsw       m3, m10
+    paddw           m4, m8
+    paddw           m5, m9
+%ifidn %1,pp
+    pmulhrsw        m4, m12
+    pmulhrsw        m5, m12
+    packuswb        m4, m5
+    movu            [r2 + r3], m4
+%else
+    psubw           m4, m12
+    psubw           m5, m12
+    vperm2i128      m1, m4, m5, 0x20
+    vperm2i128      m4, m4, m5, 0x31
+    movu            [r2 + r3], m1
+    movu            [r2 + r3 + mmsize], m4
+%endif
+
+    movu            m1, [r0 + r1]                   ; m1 = row 5
+    punpcklbw       m4, m0, m1
+    punpckhbw       m5, m0, m1
+    pmaddubsw       m4, m11
+    pmaddubsw       m5, m11
+    paddw           m6, m4
+    paddw           m7, m5
+%ifidn %1,pp
+    pmulhrsw        m6, m12
+    pmulhrsw        m7, m12
+    packuswb        m6, m7
+    movu            [r2 + r3 * 2], m6
+%else
+    psubw           m6, m12
+    psubw           m7, m12
+    vperm2i128      m0, m6, m7, 0x20
+    vperm2i128      m6, m6, m7, 0x31
+    movu            [r2 + r3 * 2], m0
+    movu            [r2 + r3 * 2 + mmsize], m6
+%endif
+
+    movu            m0, [r0 + r1 * 2]               ; m0 = row 6
+    punpcklbw       m6, m1, m0
+    punpckhbw       m7, m1, m0
+    pmaddubsw       m6, m11
+    pmaddubsw       m7, m11
+    paddw           m2, m6
+    paddw           m3, m7
+%ifidn %1,pp
+    pmulhrsw        m2, m12
+    pmulhrsw        m3, m12
+    packuswb        m2, m3
+    movu            [r2 + r5], m2
+    add             r2, 32
+%else
+    psubw           m2, m12
+    psubw           m3, m12
+    vperm2i128      m0, m2, m3, 0x20
+    vperm2i128      m2, m2, m3, 0x31
+    movu            [r2 + r5], m0
+    movu            [r2 + r5 + mmsize], m2
+    add             r2, 64
+%endif
+    sub             r0, r7
+
+    movu            xm0, [r0 + 32]                  ; m0 = row 0
+    movu            xm1, [r0 + r1 + 32]             ; m1 = row 1
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddubsw       m0, m10
+    movu            xm2, [r0 + r1 * 2 + 32]         ; m2 = row 2
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m1, m10
+    movu            xm3, [r0 + r4 + 32]             ; m3 = row 3
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m4, m2, m11
+    paddw           m0, m4
+    pmaddubsw       m2, m10
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0 + 32]                  ; m4 = row 4
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddubsw       m5, m3, m11
+    paddw           m1, m5
+    pmaddubsw       m3, m10
+    movu            xm5, [r0 + r1 + 32]             ; m5 = row 5
+    punpckhbw       xm6, xm4, xm5
+    punpcklbw       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddubsw       m4, m11
+    paddw           m2, m4
+    movu            xm6, [r0 + r1 * 2 + 32]         ; m6 = row 6
+    punpckhbw       xm7, xm5, xm6
+    punpcklbw       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddubsw       m5, m11
+    paddw           m3, m5
+%ifidn %1,pp
+    pmulhrsw        m0, m12                         ; m0 = word: row 0
+    pmulhrsw        m1, m12                         ; m1 = word: row 1
+    pmulhrsw        m2, m12                         ; m2 = word: row 2
+    pmulhrsw        m3, m12                         ; m3 = word: row 3
+    packuswb        m0, m1
+    packuswb        m2, m3
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m2, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r5], xm3
+    lea             r2, [r2 + r3 * 4 - 32]
+%else
+    psubw           m0, m12                         ; m0 = word: row 0
+    psubw           m1, m12                         ; m1 = word: row 1
+    psubw           m2, m12                         ; m2 = word: row 2
+    psubw           m3, m12                         ; m3 = word: row 3
+    movu            [r2], m0
+    movu            [r2 + r3], m1
+    movu            [r2 + r3 * 2], m2
+    movu            [r2 + r5], m3
+    lea             r2, [r2 + r3 * 4 - 64]
+%endif
+    dec             r6d
+    jnz             .loopH
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_48x64 pp
+    FILTER_VER_CHROMA_AVX2_48x64 ps
+
+%macro FILTER_VER_CHROMA_AVX2_64xN 2
+%if ARCH_X86_64 == 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_64x%2, 4, 8, 13
+    mov             r4d, r4m
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    mova            m10, [r5]
+    mova            m11, [r5 + mmsize]
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,pp
+    mova            m12, [pw_512]
+%else
+    add             r3d, r3d
+    vbroadcasti128  m12, [pw_2000]
+%endif
+    lea             r5, [r3 * 3]
+    lea             r7, [r1 * 4]
+    mov             r6d, %2 / 4
+.loopH:
+%assign x 0
+%rep 2
+    movu            m0, [r0 + x]                    ; m0 = row 0
+    movu            m1, [r0 + r1 + x]               ; m1 = row 1
+    punpcklbw       m2, m0, m1
+    punpckhbw       m3, m0, m1
+    pmaddubsw       m2, m10
+    pmaddubsw       m3, m10
+    movu            m0, [r0 + r1 * 2 + x]           ; m0 = row 2
+    punpcklbw       m4, m1, m0
+    punpckhbw       m5, m1, m0
+    pmaddubsw       m4, m10
+    pmaddubsw       m5, m10
+    movu            m1, [r0 + r4 + x]               ; m1 = row 3
+    punpcklbw       m6, m0, m1
+    punpckhbw       m7, m0, m1
+    pmaddubsw       m8, m6, m11
+    pmaddubsw       m9, m7, m11
+    pmaddubsw       m6, m10
+    pmaddubsw       m7, m10
+    paddw           m2, m8
+    paddw           m3, m9
+%ifidn %1,pp
+    pmulhrsw        m2, m12
+    pmulhrsw        m3, m12
+    packuswb        m2, m3
+    movu            [r2], m2
+%else
+    psubw           m2, m12
+    psubw           m3, m12
+    vperm2i128      m0, m2, m3, 0x20
+    vperm2i128      m2, m2, m3, 0x31
+    movu            [r2], m0
+    movu            [r2 + mmsize], m2
+%endif
+    lea             r0, [r0 + r1 * 4]
+    movu            m0, [r0 + x]                    ; m0 = row 4
+    punpcklbw       m2, m1, m0
+    punpckhbw       m3, m1, m0
+    pmaddubsw       m8, m2, m11
+    pmaddubsw       m9, m3, m11
+    pmaddubsw       m2, m10
+    pmaddubsw       m3, m10
+    paddw           m4, m8
+    paddw           m5, m9
+%ifidn %1,pp
+    pmulhrsw        m4, m12
+    pmulhrsw        m5, m12
+    packuswb        m4, m5
+    movu            [r2 + r3], m4
+%else
+    psubw           m4, m12
+    psubw           m5, m12
+    vperm2i128      m1, m4, m5, 0x20
+    vperm2i128      m4, m4, m5, 0x31
+    movu            [r2 + r3], m1
+    movu            [r2 + r3 + mmsize], m4
+%endif
+
+    movu            m1, [r0 + r1 + x]               ; m1 = row 5
+    punpcklbw       m4, m0, m1
+    punpckhbw       m5, m0, m1
+    pmaddubsw       m4, m11
+    pmaddubsw       m5, m11
+    paddw           m6, m4
+    paddw           m7, m5
+%ifidn %1,pp
+    pmulhrsw        m6, m12
+    pmulhrsw        m7, m12
+    packuswb        m6, m7
+    movu            [r2 + r3 * 2], m6
+%else
+    psubw           m6, m12
+    psubw           m7, m12
+    vperm2i128      m0, m6, m7, 0x20
+    vperm2i128      m6, m6, m7, 0x31
+    movu            [r2 + r3 * 2], m0
+    movu            [r2 + r3 * 2 + mmsize], m6
+%endif
+
+    movu            m0, [r0 + r1 * 2 + x]           ; m0 = row 6
+    punpcklbw       m6, m1, m0
+    punpckhbw       m7, m1, m0
+    pmaddubsw       m6, m11
+    pmaddubsw       m7, m11
+    paddw           m2, m6
+    paddw           m3, m7
+%ifidn %1,pp
+    pmulhrsw        m2, m12
+    pmulhrsw        m3, m12
+    packuswb        m2, m3
+    movu            [r2 + r5], m2
+    add             r2, 32
+%else
+    psubw           m2, m12
+    psubw           m3, m12
+    vperm2i128      m0, m2, m3, 0x20
+    vperm2i128      m2, m2, m3, 0x31
+    movu            [r2 + r5], m0
+    movu            [r2 + r5 + mmsize], m2
+    add             r2, 64
+%endif
+    sub             r0, r7
+%assign x x+32
+%endrep
+%ifidn %1,pp
+    lea             r2, [r2 + r3 * 4 - 64]
+%else
+    lea             r2, [r2 + r3 * 4 - 128]
+%endif
+    add             r0, r7
+    dec             r6d
+    jnz             .loopH
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_AVX2_64xN pp, 64
+    FILTER_VER_CHROMA_AVX2_64xN pp, 48
+    FILTER_VER_CHROMA_AVX2_64xN pp, 32
+    FILTER_VER_CHROMA_AVX2_64xN pp, 16
+    FILTER_VER_CHROMA_AVX2_64xN ps, 64
+    FILTER_VER_CHROMA_AVX2_64xN ps, 48
+    FILTER_VER_CHROMA_AVX2_64xN ps, 32
+    FILTER_VER_CHROMA_AVX2_64xN ps, 16
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_vert_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W16n_H2 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_pp_%1x%2, 4, 7, 8
+
+    mov         r4d,       r4m
+    sub         r0,        r1
+
+%ifdef PIC
+    lea         r5,        [tab_ChromaCoeff]
+    movd        m0,        [r5 + r4 * 4]
+%else
+    movd        m0,        [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb      m1,        m0,       [tab_Vm]
+    pshufb      m0,        [tab_Vm + 16]
+
+    mov         r4d,       %2/2
+
+.loop:
+
+    mov         r6d,       %1/16
+
+.loopW:
+
+    movu        m2,        [r0]
+    movu        m3,        [r0 + r1]
+
+    punpcklbw   m4,        m2,        m3
+    punpckhbw   m2,        m3
+
+    pmaddubsw   m4,        m1
+    pmaddubsw   m2,        m1
+
+    lea         r5,        [r0 + 2 * r1]
+    movu        m5,        [r5]
+    movu        m6,        [r5 + r1]
+
+    punpckhbw   m7,        m5,        m6
+    pmaddubsw   m7,        m0
+    paddw       m2,        m7
+
+    punpcklbw   m7,        m5,        m6
+    pmaddubsw   m7,        m0
+    paddw       m4,        m7
+
+    mova        m7,        [pw_512]
+
+    pmulhrsw    m4,        m7
+    pmulhrsw    m2,        m7
+
+    packuswb    m4,        m2
+
+    movu        [r2],      m4
+
+    punpcklbw   m4,        m3,        m5
+    punpckhbw   m3,        m5
+
+    pmaddubsw   m4,        m1
+    pmaddubsw   m3,        m1
+
+    movu        m5,        [r5 + 2 * r1]
+
+    punpcklbw   m2,        m6,        m5
+    punpckhbw   m6,        m5
+
+    pmaddubsw   m2,        m0
+    pmaddubsw   m6,        m0
+
+    paddw       m4,        m2
+    paddw       m3,        m6
+
+    pmulhrsw    m4,        m7
+    pmulhrsw    m3,        m7
+
+    packuswb    m4,        m3
+
+    movu        [r2 + r3],      m4
+
+    add         r0,        16
+    add         r2,        16
+    dec         r6d
+    jnz         .loopW
+
+    lea         r0,        [r0 + r1 * 2 - %1]
+    lea         r2,        [r2 + r3 * 2 - %1]
+
+    dec         r4d
+    jnz        .loop
+    RET
+%endmacro
+
+    FILTER_V4_W16n_H2 64, 64
+    FILTER_V4_W16n_H2 64, 32
+    FILTER_V4_W16n_H2 64, 48
+    FILTER_V4_W16n_H2 48, 64
+    FILTER_V4_W16n_H2 64, 16
+
+%macro PROCESS_CHROMA_SP_W4_4R 0
+    movq       m0, [r0]
+    movq       m1, [r0 + r1]
+    punpcklwd  m0, m1                          ;m0=[0 1]
+    pmaddwd    m0, [r6 + 0 *16]                ;m0=[0+1]         Row1
+
+    lea        r0, [r0 + 2 * r1]
+    movq       m4, [r0]
+    punpcklwd  m1, m4                          ;m1=[1 2]
+    pmaddwd    m1, [r6 + 0 *16]                ;m1=[1+2]         Row2
+
+    movq       m5, [r0 + r1]
+    punpcklwd  m4, m5                          ;m4=[2 3]
+    pmaddwd    m2, m4, [r6 + 0 *16]            ;m2=[2+3]         Row3
+    pmaddwd    m4, [r6 + 1 * 16]
+    paddd      m0, m4                          ;m0=[0+1+2+3]     Row1 done
+
+    lea        r0, [r0 + 2 * r1]
+    movq       m4, [r0]
+    punpcklwd  m5, m4                          ;m5=[3 4]
+    pmaddwd    m3, m5, [r6 + 0 *16]            ;m3=[3+4]         Row4
+    pmaddwd    m5, [r6 + 1 * 16]
+    paddd      m1, m5                          ;m1 = [1+2+3+4]   Row2
+
+    movq       m5, [r0 + r1]
+    punpcklwd  m4, m5                          ;m4=[4 5]
+    pmaddwd    m4, [r6 + 1 * 16]
+    paddd      m2, m4                          ;m2=[2+3+4+5]     Row3
+
+    movq       m4, [r0 + 2 * r1]
+    punpcklwd  m5, m4                          ;m5=[5 6]
+    pmaddwd    m5, [r6 + 1 * 16]
+    paddd      m3, m5                          ;m3=[3+4+5+6]     Row4
+%endmacro
+
+;--------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_sp_%1x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_SP 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_sp_%1x%2, 5, 7, 7 ,0-gprsize
+
+    add       r1d, r1d
+    sub       r0, r1
+    shl       r4d, 5
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r6, [r5 + r4]
+%else
+    lea       r6, [tab_ChromaCoeffV + r4]
+%endif
+
+    mova      m6, [v4_pd_526336]
+
+    mov       dword [rsp], %2/4
+
+.loopH:
+    mov       r4d, (%1/4)
+.loopW:
+    PROCESS_CHROMA_SP_W4_4R
+
+    paddd     m0, m6
+    paddd     m1, m6
+    paddd     m2, m6
+    paddd     m3, m6
+
+    psrad     m0, 12
+    psrad     m1, 12
+    psrad     m2, 12
+    psrad     m3, 12
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+
+    packuswb  m0, m2
+
+    movd      [r2], m0
+    pextrd    [r2 + r3], m0, 1
+    lea       r5, [r2 + 2 * r3]
+    pextrd    [r5], m0, 2
+    pextrd    [r5 + r3], m0, 3
+
+    lea       r5, [4 * r1 - 2 * 4]
+    sub       r0, r5
+    add       r2, 4
+
+    dec       r4d
+    jnz       .loopW
+
+    lea       r0, [r0 + 4 * r1 - 2 * %1]
+    lea       r2, [r2 + 4 * r3 - %1]
+
+    dec       dword [rsp]
+    jnz       .loopH
+
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_SP 4, 4
+    FILTER_VER_CHROMA_SP 4, 8
+    FILTER_VER_CHROMA_SP 16, 16
+    FILTER_VER_CHROMA_SP 16, 8
+    FILTER_VER_CHROMA_SP 16, 12
+    FILTER_VER_CHROMA_SP 12, 16
+    FILTER_VER_CHROMA_SP 16, 4
+    FILTER_VER_CHROMA_SP 4, 16
+    FILTER_VER_CHROMA_SP 32, 32
+    FILTER_VER_CHROMA_SP 32, 16
+    FILTER_VER_CHROMA_SP 16, 32
+    FILTER_VER_CHROMA_SP 32, 24
+    FILTER_VER_CHROMA_SP 24, 32
+    FILTER_VER_CHROMA_SP 32, 8
+
+    FILTER_VER_CHROMA_SP 16, 24
+    FILTER_VER_CHROMA_SP 16, 64
+    FILTER_VER_CHROMA_SP 12, 32
+    FILTER_VER_CHROMA_SP 4, 32
+    FILTER_VER_CHROMA_SP 32, 64
+    FILTER_VER_CHROMA_SP 32, 48
+    FILTER_VER_CHROMA_SP 24, 64
+
+    FILTER_VER_CHROMA_SP 64, 64
+    FILTER_VER_CHROMA_SP 64, 32
+    FILTER_VER_CHROMA_SP 64, 48
+    FILTER_VER_CHROMA_SP 48, 64
+    FILTER_VER_CHROMA_SP 64, 16
+
+
+%macro PROCESS_CHROMA_SP_W2_4R 1
+    movd       m0, [r0]
+    movd       m1, [r0 + r1]
+    punpcklwd  m0, m1                          ;m0=[0 1]
+
+    lea        r0, [r0 + 2 * r1]
+    movd       m2, [r0]
+    punpcklwd  m1, m2                          ;m1=[1 2]
+    punpcklqdq m0, m1                          ;m0=[0 1 1 2]
+    pmaddwd    m0, [%1 + 0 *16]                ;m0=[0+1 1+2] Row 1-2
+
+    movd       m1, [r0 + r1]
+    punpcklwd  m2, m1                          ;m2=[2 3]
+
+    lea        r0, [r0 + 2 * r1]
+    movd       m3, [r0]
+    punpcklwd  m1, m3                          ;m2=[3 4]
+    punpcklqdq m2, m1                          ;m2=[2 3 3 4]
+
+    pmaddwd    m4, m2, [%1 + 1 * 16]           ;m4=[2+3 3+4] Row 1-2
+    pmaddwd    m2, [%1 + 0 * 16]               ;m2=[2+3 3+4] Row 3-4
+    paddd      m0, m4                          ;m0=[0+1+2+3 1+2+3+4] Row 1-2
+
+    movd       m1, [r0 + r1]
+    punpcklwd  m3, m1                          ;m3=[4 5]
+
+    movd       m4, [r0 + 2 * r1]
+    punpcklwd  m1, m4                          ;m1=[5 6]
+    punpcklqdq m3, m1                          ;m2=[4 5 5 6]
+    pmaddwd    m3, [%1 + 1 * 16]               ;m3=[4+5 5+6] Row 3-4
+    paddd      m2, m3                          ;m2=[2+3+4+5 3+4+5+6] Row 3-4
+%endmacro
+
+;-------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vertical_sp_%1x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_SP_W2_4R 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_sp_%1x%2, 5, 6, 6
+
+    add       r1d, r1d
+    sub       r0, r1
+    shl       r4d, 5
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r5, [r5 + r4]
+%else
+    lea       r5, [tab_ChromaCoeffV + r4]
+%endif
+
+    mova      m5, [v4_pd_526336]
+
+    mov       r4d, (%2/4)
+
+.loopH:
+    PROCESS_CHROMA_SP_W2_4R r5
+
+    paddd     m0, m5
+    paddd     m2, m5
+
+    psrad     m0, 12
+    psrad     m2, 12
+
+    packssdw  m0, m2
+    packuswb  m0, m0
+
+    pextrw    [r2], m0, 0
+    pextrw    [r2 + r3], m0, 1
+    lea       r2, [r2 + 2 * r3]
+    pextrw    [r2], m0, 2
+    pextrw    [r2 + r3], m0, 3
+
+    lea       r2, [r2 + 2 * r3]
+
+    dec       r4d
+    jnz       .loopH
+
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_SP_W2_4R 2, 4
+    FILTER_VER_CHROMA_SP_W2_4R 2, 8
+
+    FILTER_VER_CHROMA_SP_W2_4R 2, 16
+
+;--------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_sp_4x2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+INIT_XMM sse4
+cglobal interp_4tap_vert_sp_4x2, 5, 6, 5
+
+    add        r1d, r1d
+    sub        r0, r1
+    shl        r4d, 5
+
+%ifdef PIC
+    lea        r5, [tab_ChromaCoeffV]
+    lea        r5, [r5 + r4]
+%else
+    lea        r5, [tab_ChromaCoeffV + r4]
+%endif
+
+    mova       m4, [v4_pd_526336]
+
+    movq       m0, [r0]
+    movq       m1, [r0 + r1]
+    punpcklwd  m0, m1                          ;m0=[0 1]
+    pmaddwd    m0, [r5 + 0 *16]                ;m0=[0+1]  Row1
+
+    lea        r0, [r0 + 2 * r1]
+    movq       m2, [r0]
+    punpcklwd  m1, m2                          ;m1=[1 2]
+    pmaddwd    m1, [r5 + 0 *16]                ;m1=[1+2]  Row2
+
+    movq       m3, [r0 + r1]
+    punpcklwd  m2, m3                          ;m4=[2 3]
+    pmaddwd    m2, [r5 + 1 * 16]
+    paddd      m0, m2                          ;m0=[0+1+2+3]  Row1 done
+    paddd      m0, m4
+    psrad      m0, 12
+
+    movq       m2, [r0 + 2 * r1]
+    punpcklwd  m3, m2                          ;m5=[3 4]
+    pmaddwd    m3, [r5 + 1 * 16]
+    paddd      m1, m3                          ;m1 = [1+2+3+4]  Row2 done
+    paddd      m1, m4
+    psrad      m1, 12
+
+    packssdw   m0, m1
+    packuswb   m0, m0
+
+    movd       [r2], m0
+    pextrd     [r2 + r3], m0, 1
+
+    RET
+
+;-------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vertical_sp_6x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_SP_W6_H4 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_sp_6x%2, 5, 7, 7
+
+    add       r1d, r1d
+    sub       r0, r1
+    shl       r4d, 5
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r6, [r5 + r4]
+%else
+    lea       r6, [tab_ChromaCoeffV + r4]
+%endif
+
+    mova      m6, [v4_pd_526336]
+
+    mov       r4d, %2/4
+
+.loopH:
+    PROCESS_CHROMA_SP_W4_4R
+
+    paddd     m0, m6
+    paddd     m1, m6
+    paddd     m2, m6
+    paddd     m3, m6
+
+    psrad     m0, 12
+    psrad     m1, 12
+    psrad     m2, 12
+    psrad     m3, 12
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+
+    packuswb  m0, m2
+
+    movd      [r2], m0
+    pextrd    [r2 + r3], m0, 1
+    lea       r5, [r2 + 2 * r3]
+    pextrd    [r5], m0, 2
+    pextrd    [r5 + r3], m0, 3
+
+    lea       r5, [4 * r1 - 2 * 4]
+    sub       r0, r5
+    add       r2, 4
+
+    PROCESS_CHROMA_SP_W2_4R r6
+
+    paddd     m0, m6
+    paddd     m2, m6
+
+    psrad     m0, 12
+    psrad     m2, 12
+
+    packssdw  m0, m2
+    packuswb  m0, m0
+
+    pextrw    [r2], m0, 0
+    pextrw    [r2 + r3], m0, 1
+    lea       r2, [r2 + 2 * r3]
+    pextrw    [r2], m0, 2
+    pextrw    [r2 + r3], m0, 3
+
+    sub       r0, 2 * 4
+    lea       r2, [r2 + 2 * r3 - 4]
+
+    dec       r4d
+    jnz       .loopH
+
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_SP_W6_H4 6, 8
+
+    FILTER_VER_CHROMA_SP_W6_H4 6, 16
+
+%macro PROCESS_CHROMA_SP_W8_2R 0
+    movu       m1, [r0]
+    movu       m3, [r0 + r1]
+    punpcklwd  m0, m1, m3
+    pmaddwd    m0, [r5 + 0 * 16]                ;m0 = [0l+1l]  Row1l
+    punpckhwd  m1, m3
+    pmaddwd    m1, [r5 + 0 * 16]                ;m1 = [0h+1h]  Row1h
+
+    movu       m4, [r0 + 2 * r1]
+    punpcklwd  m2, m3, m4
+    pmaddwd    m2, [r5 + 0 * 16]                ;m2 = [1l+2l]  Row2l
+    punpckhwd  m3, m4
+    pmaddwd    m3, [r5 + 0 * 16]                ;m3 = [1h+2h]  Row2h
+
+    lea        r0, [r0 + 2 * r1]
+    movu       m5, [r0 + r1]
+    punpcklwd  m6, m4, m5
+    pmaddwd    m6, [r5 + 1 * 16]                ;m6 = [2l+3l]  Row1l
+    paddd      m0, m6                           ;m0 = [0l+1l+2l+3l]  Row1l sum
+    punpckhwd  m4, m5
+    pmaddwd    m4, [r5 + 1 * 16]                ;m6 = [2h+3h]  Row1h
+    paddd      m1, m4                           ;m1 = [0h+1h+2h+3h]  Row1h sum
+
+    movu       m4, [r0 + 2 * r1]
+    punpcklwd  m6, m5, m4
+    pmaddwd    m6, [r5 + 1 * 16]                ;m6 = [3l+4l]  Row2l
+    paddd      m2, m6                           ;m2 = [1l+2l+3l+4l]  Row2l sum
+    punpckhwd  m5, m4
+    pmaddwd    m5, [r5 + 1 * 16]                ;m1 = [3h+4h]  Row2h
+    paddd      m3, m5                           ;m3 = [1h+2h+3h+4h]  Row2h sum
+%endmacro
+
+;--------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_sp_8x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_SP_W8_H2 2
+INIT_XMM sse2
+cglobal interp_4tap_vert_sp_%1x%2, 5, 6, 8
+
+    add       r1d, r1d
+    sub       r0, r1
+    shl       r4d, 5
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r5, [r5 + r4]
+%else
+    lea       r5, [tab_ChromaCoeffV + r4]
+%endif
+
+    mova      m7, [v4_pd_526336]
+
+    mov       r4d, %2/2
+.loopH:
+    PROCESS_CHROMA_SP_W8_2R
+
+    paddd     m0, m7
+    paddd     m1, m7
+    paddd     m2, m7
+    paddd     m3, m7
+
+    psrad     m0, 12
+    psrad     m1, 12
+    psrad     m2, 12
+    psrad     m3, 12
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+
+    packuswb  m0, m2
+
+    movlps    [r2], m0
+    movhps    [r2 + r3], m0
+
+    lea       r2, [r2 + 2 * r3]
+
+    dec r4d
+    jnz .loopH
+
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_SP_W8_H2 8, 2
+    FILTER_VER_CHROMA_SP_W8_H2 8, 4
+    FILTER_VER_CHROMA_SP_W8_H2 8, 6
+    FILTER_VER_CHROMA_SP_W8_H2 8, 8
+    FILTER_VER_CHROMA_SP_W8_H2 8, 16
+    FILTER_VER_CHROMA_SP_W8_H2 8, 32
+
+    FILTER_VER_CHROMA_SP_W8_H2 8, 12
+    FILTER_VER_CHROMA_SP_W8_H2 8, 64
+
+
+;---------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ps_%1x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;---------------------------------------------------------------------------------------------------------------
+%macro FILTER_V_PS_W16n 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_ps_%1x%2, 4, 7, 8
+
+    mov        r4d, r4m
+    sub        r0, r1
+    add        r3d, r3d
+
+%ifdef PIC
+    lea        r5, [tab_ChromaCoeff]
+    movd       m0, [r5 + r4 * 4]
+%else
+    movd       m0, [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb     m1, m0, [tab_Vm]
+    pshufb     m0, [tab_Vm + 16]
+    mov        r4d, %2/2
+
+.loop:
+
+    mov         r6d,       %1/16
+
+.loopW:
+
+    movu       m2, [r0]
+    movu       m3, [r0 + r1]
+
+    punpcklbw  m4, m2, m3
+    punpckhbw  m2, m3
+
+    pmaddubsw  m4, m1
+    pmaddubsw  m2, m1
+
+    lea        r5, [r0 + 2 * r1]
+    movu       m5, [r5]
+    movu       m7, [r5 + r1]
+
+    punpcklbw  m6, m5, m7
+    pmaddubsw  m6, m0
+    paddw      m4, m6
+
+    punpckhbw  m6, m5, m7
+    pmaddubsw  m6, m0
+    paddw      m2, m6
+
+    mova       m6, [pw_2000]
+
+    psubw      m4, m6
+    psubw      m2, m6
+
+    movu       [r2], m4
+    movu       [r2 + 16], m2
+
+    punpcklbw  m4, m3, m5
+    punpckhbw  m3, m5
+
+    pmaddubsw  m4, m1
+    pmaddubsw  m3, m1
+
+    movu       m5, [r5 + 2 * r1]
+
+    punpcklbw  m2, m7, m5
+    punpckhbw  m7, m5
+
+    pmaddubsw  m2, m0
+    pmaddubsw  m7, m0
+
+    paddw      m4, m2
+    paddw      m3, m7
+
+    psubw      m4, m6
+    psubw      m3, m6
+
+    movu       [r2 + r3], m4
+    movu       [r2 + r3 + 16], m3
+
+    add         r0,        16
+    add         r2,        32
+    dec         r6d
+    jnz         .loopW
+
+    lea         r0,        [r0 + r1 * 2 - %1]
+    lea         r2,        [r2 + r3 * 2 - %1 * 2]
+
+    dec        r4d
+    jnz        .loop
+    RET
+%endmacro
+
+    FILTER_V_PS_W16n 64, 64
+    FILTER_V_PS_W16n 64, 32
+    FILTER_V_PS_W16n 64, 48
+    FILTER_V_PS_W16n 48, 64
+    FILTER_V_PS_W16n 64, 16
+
+
+;------------------------------------------------------------------------------------------------------------
+;void interp_4tap_vert_ps_2x4(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;------------------------------------------------------------------------------------------------------------
+INIT_XMM sse4
+cglobal interp_4tap_vert_ps_2x4, 4, 6, 7
+
+    mov         r4d, r4m
+    sub         r0, r1
+    add         r3d, r3d
+
+%ifdef PIC
+    lea         r5, [tab_ChromaCoeff]
+    movd        m0, [r5 + r4 * 4]
+%else
+    movd        m0, [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb      m0, [tab_Cm]
+
+    lea         r5, [3 * r1]
+
+    movd        m2, [r0]
+    movd        m3, [r0 + r1]
+    movd        m4, [r0 + 2 * r1]
+    movd        m5, [r0 + r5]
+
+    punpcklbw   m2, m3
+    punpcklbw   m6, m4, m5
+    punpcklbw   m2, m6
+
+    pmaddubsw   m2, m0
+
+    lea         r0, [r0 + 4 * r1]
+    movd        m6, [r0]
+
+    punpcklbw   m3, m4
+    punpcklbw   m1, m5, m6
+    punpcklbw   m3, m1
+
+    pmaddubsw   m3, m0
+    phaddw      m2, m3
+
+    mova        m1, [pw_2000]
+
+    psubw       m2, m1
+
+    movd        [r2], m2
+    pextrd      [r2 + r3], m2, 2
+
+    movd        m2, [r0 + r1]
+
+    punpcklbw   m4, m5
+    punpcklbw   m3, m6, m2
+    punpcklbw   m4, m3
+
+    pmaddubsw   m4, m0
+
+    movd        m3, [r0 + 2 * r1]
+
+    punpcklbw   m5, m6
+    punpcklbw   m2, m3
+    punpcklbw   m5, m2
+
+    pmaddubsw   m5, m0
+    phaddw      m4, m5
+    psubw       m4, m1
+
+    lea         r2, [r2 + 2 * r3]
+    movd        [r2], m4
+    pextrd      [r2 + r3], m4, 2
+
+    RET
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ps_2x8(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+%macro FILTER_V_PS_W2 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_ps_2x%2, 4, 6, 8
+
+    mov        r4d, r4m
+    sub        r0, r1
+    add        r3d, r3d
+
+%ifdef PIC
+    lea        r5, [tab_ChromaCoeff]
+    movd       m0, [r5 + r4 * 4]
+%else
+    movd       m0, [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    pshufb     m0, [tab_Cm]
+
+    mova       m1, [pw_2000]
+    lea        r5, [3 * r1]
+    mov        r4d, %2/4
+.loop:
+    movd       m2, [r0]
+    movd       m3, [r0 + r1]
+    movd       m4, [r0 + 2 * r1]
+    movd       m5, [r0 + r5]
+
+    punpcklbw  m2, m3
+    punpcklbw  m6, m4, m5
+    punpcklbw  m2, m6
+
+    pmaddubsw  m2, m0
+
+    lea        r0, [r0 + 4 * r1]
+    movd       m6, [r0]
+
+    punpcklbw  m3, m4
+    punpcklbw  m7, m5, m6
+    punpcklbw  m3, m7
+
+    pmaddubsw  m3, m0
+
+    phaddw     m2, m3
+    psubw      m2, m1
+
+
+    movd       [r2], m2
+    pshufd     m2, m2, 2
+    movd       [r2 + r3], m2
+
+    movd       m2, [r0 + r1]
+
+    punpcklbw  m4, m5
+    punpcklbw  m3, m6, m2
+    punpcklbw  m4, m3
+
+    pmaddubsw  m4, m0
+
+    movd       m3, [r0 + 2 * r1]
+
+    punpcklbw  m5, m6
+    punpcklbw  m2, m3
+    punpcklbw  m5, m2
+
+    pmaddubsw  m5, m0
+
+    phaddw     m4, m5
+
+    psubw      m4, m1
+
+    lea        r2, [r2 + 2 * r3]
+    movd       [r2], m4
+    pshufd     m4 , m4 ,2
+    movd       [r2 + r3], m4
+
+    lea        r2, [r2 + 2 * r3]
+
+    dec        r4d
+    jnz        .loop
+
+    RET
+%endmacro
+
+    FILTER_V_PS_W2 2, 8
+
+    FILTER_V_PS_W2 2, 16
+
+;-----------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ss_%1x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_SS 2
+INIT_XMM sse2
+cglobal interp_4tap_vert_ss_%1x%2, 5, 7, 6 ,0-gprsize
+
+    add       r1d, r1d
+    add       r3d, r3d
+    sub       r0, r1
+    shl       r4d, 5
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r6, [r5 + r4]
+%else
+    lea       r6, [tab_ChromaCoeffV + r4]
+%endif
+
+    mov       dword [rsp], %2/4
+
+.loopH:
+    mov       r4d, (%1/4)
+.loopW:
+    PROCESS_CHROMA_SP_W4_4R
+
+    psrad     m0, 6
+    psrad     m1, 6
+    psrad     m2, 6
+    psrad     m3, 6
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+
+    movlps    [r2], m0
+    movhps    [r2 + r3], m0
+    lea       r5, [r2 + 2 * r3]
+    movlps    [r5], m2
+    movhps    [r5 + r3], m2
+
+    lea       r5, [4 * r1 - 2 * 4]
+    sub       r0, r5
+    add       r2, 2 * 4
+
+    dec       r4d
+    jnz       .loopW
+
+    lea       r0, [r0 + 4 * r1 - 2 * %1]
+    lea       r2, [r2 + 4 * r3 - 2 * %1]
+
+    dec       dword [rsp]
+    jnz       .loopH
+
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_SS 4, 4
+    FILTER_VER_CHROMA_SS 4, 8
+    FILTER_VER_CHROMA_SS 16, 16
+    FILTER_VER_CHROMA_SS 16, 8
+    FILTER_VER_CHROMA_SS 16, 12
+    FILTER_VER_CHROMA_SS 12, 16
+    FILTER_VER_CHROMA_SS 16, 4
+    FILTER_VER_CHROMA_SS 4, 16
+    FILTER_VER_CHROMA_SS 32, 32
+    FILTER_VER_CHROMA_SS 32, 16
+    FILTER_VER_CHROMA_SS 16, 32
+    FILTER_VER_CHROMA_SS 32, 24
+    FILTER_VER_CHROMA_SS 24, 32
+    FILTER_VER_CHROMA_SS 32, 8
+
+    FILTER_VER_CHROMA_SS 16, 24
+    FILTER_VER_CHROMA_SS 12, 32
+    FILTER_VER_CHROMA_SS 4, 32
+    FILTER_VER_CHROMA_SS 32, 64
+    FILTER_VER_CHROMA_SS 16, 64
+    FILTER_VER_CHROMA_SS 32, 48
+    FILTER_VER_CHROMA_SS 24, 64
+
+    FILTER_VER_CHROMA_SS 64, 64
+    FILTER_VER_CHROMA_SS 64, 32
+    FILTER_VER_CHROMA_SS 64, 48
+    FILTER_VER_CHROMA_SS 48, 64
+    FILTER_VER_CHROMA_SS 64, 16
+
+%macro FILTER_VER_CHROMA_S_AVX2_4x4 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_4x4, 4, 6, 7
+    mov             r4d, r4m
+    add             r1d, r1d
+    shl             r4d, 6
+    sub             r0, r1
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+%ifidn %1,sp
+    mova            m6, [v4_pd_526336]
+%else
+    add             r3d, r3d
+%endif
+
+    movq            xm0, [r0]
+    movq            xm1, [r0 + r1]
+    punpcklwd       xm0, xm1
+    movq            xm2, [r0 + r1 * 2]
+    punpcklwd       xm1, xm2
+    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
+    pmaddwd         m0, [r5]
+    movq            xm3, [r0 + r4]
+    punpcklwd       xm2, xm3
+    lea             r0, [r0 + 4 * r1]
+    movq            xm4, [r0]
+    punpcklwd       xm3, xm4
+    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
+    pmaddwd         m5, m2, [r5 + 1 * mmsize]
+    pmaddwd         m2, [r5]
+    paddd           m0, m5
+    movq            xm3, [r0 + r1]
+    punpcklwd       xm4, xm3
+    movq            xm1, [r0 + r1 * 2]
+    punpcklwd       xm3, xm1
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
+    pmaddwd         m4, [r5 + 1 * mmsize]
+    paddd           m2, m4
+
+%ifidn %1,sp
+    paddd           m0, m6
+    paddd           m2, m6
+    psrad           m0, 12
+    psrad           m2, 12
+%else
+    psrad           m0, 6
+    psrad           m2, 6
+%endif
+    packssdw        m0, m2
+    vextracti128    xm2, m0, 1
+    lea             r4, [r3 * 3]
+
+%ifidn %1,sp
+    packuswb        xm0, xm2
+    movd            [r2], xm0
+    pextrd          [r2 + r3], xm0, 2
+    pextrd          [r2 + r3 * 2], xm0, 1
+    pextrd          [r2 + r4], xm0, 3
+%else
+    movq            [r2], xm0
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r4], xm2
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_4x4 sp
+    FILTER_VER_CHROMA_S_AVX2_4x4 ss
+
+%macro FILTER_VER_CHROMA_S_AVX2_4x8 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_4x8, 4, 6, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+    sub             r0, r1
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+%ifidn %1,sp
+    mova            m7, [v4_pd_526336]
+%else
+    add             r3d, r3d
+%endif
+
+    movq            xm0, [r0]
+    movq            xm1, [r0 + r1]
+    punpcklwd       xm0, xm1
+    movq            xm2, [r0 + r1 * 2]
+    punpcklwd       xm1, xm2
+    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
+    pmaddwd         m0, [r5]
+    movq            xm3, [r0 + r4]
+    punpcklwd       xm2, xm3
+    lea             r0, [r0 + 4 * r1]
+    movq            xm4, [r0]
+    punpcklwd       xm3, xm4
+    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
+    pmaddwd         m5, m2, [r5 + 1 * mmsize]
+    pmaddwd         m2, [r5]
+    paddd           m0, m5
+    movq            xm3, [r0 + r1]
+    punpcklwd       xm4, xm3
+    movq            xm1, [r0 + r1 * 2]
+    punpcklwd       xm3, xm1
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
+    pmaddwd         m5, m4, [r5 + 1 * mmsize]
+    paddd           m2, m5
+    pmaddwd         m4, [r5]
+    movq            xm3, [r0 + r4]
+    punpcklwd       xm1, xm3
+    lea             r0, [r0 + 4 * r1]
+    movq            xm6, [r0]
+    punpcklwd       xm3, xm6
+    vinserti128     m1, m1, xm3, 1                  ; m1 = [8 7 7 6]
+    pmaddwd         m5, m1, [r5 + 1 * mmsize]
+    paddd           m4, m5
+    pmaddwd         m1, [r5]
+    movq            xm3, [r0 + r1]
+    punpcklwd       xm6, xm3
+    movq            xm5, [r0 + 2 * r1]
+    punpcklwd       xm3, xm5
+    vinserti128     m6, m6, xm3, 1                  ; m6 = [A 9 9 8]
+    pmaddwd         m6, [r5 + 1 * mmsize]
+    paddd           m1, m6
+    lea             r4, [r3 * 3]
+
+%ifidn %1,sp
+    paddd           m0, m7
+    paddd           m2, m7
+    paddd           m4, m7
+    paddd           m1, m7
+    psrad           m0, 12
+    psrad           m2, 12
+    psrad           m4, 12
+    psrad           m1, 12
+%else
+    psrad           m0, 6
+    psrad           m2, 6
+    psrad           m4, 6
+    psrad           m1, 6
+%endif
+    packssdw        m0, m2
+    packssdw        m4, m1
+%ifidn %1,sp
+    packuswb        m0, m4
+    vextracti128    xm2, m0, 1
+    movd            [r2], xm0
+    movd            [r2 + r3], xm2
+    pextrd          [r2 + r3 * 2], xm0, 1
+    pextrd          [r2 + r4], xm2, 1
+    lea             r2, [r2 + r3 * 4]
+    pextrd          [r2], xm0, 2
+    pextrd          [r2 + r3], xm2, 2
+    pextrd          [r2 + r3 * 2], xm0, 3
+    pextrd          [r2 + r4], xm2, 3
+%else
+    vextracti128    xm2, m0, 1
+    vextracti128    xm1, m4, 1
+    movq            [r2], xm0
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r4], xm2
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm4
+    movq            [r2 + r3], xm1
+    movhps          [r2 + r3 * 2], xm4
+    movhps          [r2 + r4], xm1
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_4x8 sp
+    FILTER_VER_CHROMA_S_AVX2_4x8 ss
+
+%macro PROCESS_CHROMA_AVX2_W4_16R 1
+    movq            xm0, [r0]
+    movq            xm1, [r0 + r1]
+    punpcklwd       xm0, xm1
+    movq            xm2, [r0 + r1 * 2]
+    punpcklwd       xm1, xm2
+    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
+    pmaddwd         m0, [r5]
+    movq            xm3, [r0 + r4]
+    punpcklwd       xm2, xm3
+    lea             r0, [r0 + 4 * r1]
+    movq            xm4, [r0]
+    punpcklwd       xm3, xm4
+    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
+    pmaddwd         m5, m2, [r5 + 1 * mmsize]
+    pmaddwd         m2, [r5]
+    paddd           m0, m5
+    movq            xm3, [r0 + r1]
+    punpcklwd       xm4, xm3
+    movq            xm1, [r0 + r1 * 2]
+    punpcklwd       xm3, xm1
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
+    pmaddwd         m5, m4, [r5 + 1 * mmsize]
+    paddd           m2, m5
+    pmaddwd         m4, [r5]
+    movq            xm3, [r0 + r4]
+    punpcklwd       xm1, xm3
+    lea             r0, [r0 + 4 * r1]
+    movq            xm6, [r0]
+    punpcklwd       xm3, xm6
+    vinserti128     m1, m1, xm3, 1                  ; m1 = [8 7 7 6]
+    pmaddwd         m5, m1, [r5 + 1 * mmsize]
+    paddd           m4, m5
+    pmaddwd         m1, [r5]
+    movq            xm3, [r0 + r1]
+    punpcklwd       xm6, xm3
+    movq            xm5, [r0 + 2 * r1]
+    punpcklwd       xm3, xm5
+    vinserti128     m6, m6, xm3, 1                  ; m6 = [10 9 9 8]
+    pmaddwd         m3, m6, [r5 + 1 * mmsize]
+    paddd           m1, m3
+    pmaddwd         m6, [r5]
+
+%ifidn %1,sp
+    paddd           m0, m7
+    paddd           m2, m7
+    paddd           m4, m7
+    paddd           m1, m7
+    psrad           m4, 12
+    psrad           m1, 12
+    psrad           m0, 12
+    psrad           m2, 12
+%else
+    psrad           m0, 6
+    psrad           m2, 6
+    psrad           m4, 6
+    psrad           m1, 6
+%endif
+    packssdw        m0, m2
+    packssdw        m4, m1
+%ifidn %1,sp
+    packuswb        m0, m4
+    vextracti128    xm4, m0, 1
+    movd            [r2], xm0
+    movd            [r2 + r3], xm4
+    pextrd          [r2 + r3 * 2], xm0, 1
+    pextrd          [r2 + r6], xm4, 1
+    lea             r2, [r2 + r3 * 4]
+    pextrd          [r2], xm0, 2
+    pextrd          [r2 + r3], xm4, 2
+    pextrd          [r2 + r3 * 2], xm0, 3
+    pextrd          [r2 + r6], xm4, 3
+%else
+    vextracti128    xm2, m0, 1
+    vextracti128    xm1, m4, 1
+    movq            [r2], xm0
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r6], xm2
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm4
+    movq            [r2 + r3], xm1
+    movhps          [r2 + r3 * 2], xm4
+    movhps          [r2 + r6], xm1
+%endif
+
+    movq            xm2, [r0 + r4]
+    punpcklwd       xm5, xm2
+    lea             r0, [r0 + 4 * r1]
+    movq            xm0, [r0]
+    punpcklwd       xm2, xm0
+    vinserti128     m5, m5, xm2, 1                  ; m5 = [12 11 11 10]
+    pmaddwd         m2, m5, [r5 + 1 * mmsize]
+    paddd           m6, m2
+    pmaddwd         m5, [r5]
+    movq            xm2, [r0 + r1]
+    punpcklwd       xm0, xm2
+    movq            xm3, [r0 + 2 * r1]
+    punpcklwd       xm2, xm3
+    vinserti128     m0, m0, xm2, 1                  ; m0 = [14 13 13 12]
+    pmaddwd         m2, m0, [r5 + 1 * mmsize]
+    paddd           m5, m2
+    pmaddwd         m0, [r5]
+    movq            xm4, [r0 + r4]
+    punpcklwd       xm3, xm4
+    lea             r0, [r0 + 4 * r1]
+    movq            xm1, [r0]
+    punpcklwd       xm4, xm1
+    vinserti128     m3, m3, xm4, 1                  ; m3 = [16 15 15 14]
+    pmaddwd         m4, m3, [r5 + 1 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m3, [r5]
+    movq            xm4, [r0 + r1]
+    punpcklwd       xm1, xm4
+    movq            xm2, [r0 + 2 * r1]
+    punpcklwd       xm4, xm2
+    vinserti128     m1, m1, xm4, 1                  ; m1 = [18 17 17 16]
+    pmaddwd         m1, [r5 + 1 * mmsize]
+    paddd           m3, m1
+
+%ifidn %1,sp
+    paddd           m6, m7
+    paddd           m5, m7
+    paddd           m0, m7
+    paddd           m3, m7
+    psrad           m6, 12
+    psrad           m5, 12
+    psrad           m0, 12
+    psrad           m3, 12
+%else
+    psrad           m6, 6
+    psrad           m5, 6
+    psrad           m0, 6
+    psrad           m3, 6
+%endif
+    packssdw        m6, m5
+    packssdw        m0, m3
+    lea             r2, [r2 + r3 * 4]
+
+%ifidn %1,sp
+    packuswb        m6, m0
+    vextracti128    xm0, m6, 1
+    movd            [r2], xm6
+    movd            [r2 + r3], xm0
+    pextrd          [r2 + r3 * 2], xm6, 1
+    pextrd          [r2 + r6], xm0, 1
+    lea             r2, [r2 + r3 * 4]
+    pextrd          [r2], xm6, 2
+    pextrd          [r2 + r3], xm0, 2
+    pextrd          [r2 + r3 * 2], xm6, 3
+    pextrd          [r2 + r6], xm0, 3
+%else
+    vextracti128    xm5, m6, 1
+    vextracti128    xm3, m0, 1
+    movq            [r2], xm6
+    movq            [r2 + r3], xm5
+    movhps          [r2 + r3 * 2], xm6
+    movhps          [r2 + r6], xm5
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm0
+    movq            [r2 + r3], xm3
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r6], xm3
+%endif
+%endmacro
+
+%macro FILTER_VER_CHROMA_S_AVX2_4x16 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_4x16, 4, 7, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+    sub             r0, r1
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+%ifidn %1,sp
+    mova            m7, [v4_pd_526336]
+%else
+    add             r3d, r3d
+%endif
+    lea             r6, [r3 * 3]
+    PROCESS_CHROMA_AVX2_W4_16R %1
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_4x16 sp
+    FILTER_VER_CHROMA_S_AVX2_4x16 ss
+
+%macro FILTER_VER_CHROMA_S_AVX2_4x32 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_4x32, 4, 7, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+    sub             r0, r1
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+%ifidn %1,sp
+    mova            m7, [v4_pd_526336]
+%else
+    add             r3d, r3d
+%endif
+    lea             r6, [r3 * 3]
+%rep 2
+    PROCESS_CHROMA_AVX2_W4_16R %1
+    lea             r2, [r2 + r3 * 4]
+%endrep
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_4x32 sp
+    FILTER_VER_CHROMA_S_AVX2_4x32 ss
+
+%macro FILTER_VER_CHROMA_S_AVX2_4x2 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_4x2, 4, 6, 6
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+    sub             r0, r1
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+%ifidn %1,sp
+    mova            m5, [v4_pd_526336]
+%else
+    add             r3d, r3d
+%endif
+    movq            xm0, [r0]
+    movq            xm1, [r0 + r1]
+    punpcklwd       xm0, xm1
+    movq            xm2, [r0 + r1 * 2]
+    punpcklwd       xm1, xm2
+    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
+    pmaddwd         m0, [r5]
+    movq            xm3, [r0 + r4]
+    punpcklwd       xm2, xm3
+    movq            xm4, [r0 + 4 * r1]
+    punpcklwd       xm3, xm4
+    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m0, m2
+%ifidn %1,sp
+    paddd           m0, m5
+    psrad           m0, 12
+%else
+    psrad           m0, 6
+%endif
+    vextracti128    xm1, m0, 1
+    packssdw        xm0, xm1
+%ifidn %1,sp
+    packuswb        xm0, xm0
+    movd            [r2], xm0
+    pextrd          [r2 + r3], xm0, 1
+%else
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_4x2 sp
+    FILTER_VER_CHROMA_S_AVX2_4x2 ss
+
+%macro FILTER_VER_CHROMA_S_AVX2_2x4 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_2x4, 4, 6, 6
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+    sub             r0, r1
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+%ifidn %1,sp
+    mova            m5, [v4_pd_526336]
+%else
+    add             r3d, r3d
+%endif
+    movd            xm0, [r0]
+    movd            xm1, [r0 + r1]
+    punpcklwd       xm0, xm1
+    movd            xm2, [r0 + r1 * 2]
+    punpcklwd       xm1, xm2
+    punpcklqdq      xm0, xm1                        ; m0 = [2 1 1 0]
+    movd            xm3, [r0 + r4]
+    punpcklwd       xm2, xm3
+    lea             r0, [r0 + 4 * r1]
+    movd            xm4, [r0]
+    punpcklwd       xm3, xm4
+    punpcklqdq      xm2, xm3                        ; m2 = [4 3 3 2]
+    vinserti128     m0, m0, xm2, 1                  ; m0 = [4 3 3 2 2 1 1 0]
+    movd            xm1, [r0 + r1]
+    punpcklwd       xm4, xm1
+    movd            xm3, [r0 + r1 * 2]
+    punpcklwd       xm1, xm3
+    punpcklqdq      xm4, xm1                        ; m4 = [6 5 5 4]
+    vinserti128     m2, m2, xm4, 1                  ; m2 = [6 5 5 4 4 3 3 2]
+    pmaddwd         m0, [r5]
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m0, m2
+%ifidn %1,sp
+    paddd           m0, m5
+    psrad           m0, 12
+%else
+    psrad           m0, 6
+%endif
+    vextracti128    xm1, m0, 1
+    packssdw        xm0, xm1
+    lea             r4, [r3 * 3]
+%ifidn %1,sp
+    packuswb        xm0, xm0
+    pextrw          [r2], xm0, 0
+    pextrw          [r2 + r3], xm0, 1
+    pextrw          [r2 + 2 * r3], xm0, 2
+    pextrw          [r2 + r4], xm0, 3
+%else
+    movd            [r2], xm0
+    pextrd          [r2 + r3], xm0, 1
+    pextrd          [r2 + 2 * r3], xm0, 2
+    pextrd          [r2 + r4], xm0, 3
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_2x4 sp
+    FILTER_VER_CHROMA_S_AVX2_2x4 ss
+
+%macro FILTER_VER_CHROMA_S_AVX2_8x8 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_8x8, 4, 6, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,sp
+    mova            m7, [v4_pd_526336]
+%else
+    add             r3d, r3d
+%endif
+
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    pmaddwd         m2, [r5]
+    paddd           m0, m4
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    pmaddwd         m3, [r5]
+    paddd           m1, m5
+%ifidn %1,sp
+    paddd           m0, m7
+    paddd           m1, m7
+    psrad           m0, 12
+    psrad           m1, 12
+%else
+    psrad           m0, 6
+    psrad           m1, 6
+%endif
+    packssdw        m0, m1
+
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m2, m6
+    pmaddwd         m4, [r5]
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm1, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm1, 1
+    pmaddwd         m1, m5, [r5 + 1 * mmsize]
+    pmaddwd         m5, [r5]
+    paddd           m3, m1
+%ifidn %1,sp
+    paddd           m2, m7
+    paddd           m3, m7
+    psrad           m2, 12
+    psrad           m3, 12
+%else
+    psrad           m2, 6
+    psrad           m3, 6
+%endif
+    packssdw        m2, m3
+
+    movu            xm1, [r0 + r4]                  ; m1 = row 7
+    punpckhwd       xm3, xm6, xm1
+    punpcklwd       xm6, xm1
+    vinserti128     m6, m6, xm3, 1
+    pmaddwd         m3, m6, [r5 + 1 * mmsize]
+    pmaddwd         m6, [r5]
+    paddd           m4, m3
+
+    lea             r4, [r3 * 3]
+%ifidn %1,sp
+    packuswb        m0, m2
+    mova            m3, [v4_interp8_hps_shuf]
+    vpermd          m0, m3, m0
+    vextracti128    xm2, m0, 1
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r4], xm2
+%else
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    movu            [r2], xm0
+    vextracti128    xm0, m0, 1
+    vextracti128    xm3, m2, 1
+    movu            [r2 + r3], xm0
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r4], xm3
+%endif
+    lea             r2, [r2 + r3 * 4]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm0, [r0]                       ; m0 = row 8
+    punpckhwd       xm2, xm1, xm0
+    punpcklwd       xm1, xm0
+    vinserti128     m1, m1, xm2, 1
+    pmaddwd         m2, m1, [r5 + 1 * mmsize]
+    pmaddwd         m1, [r5]
+    paddd           m5, m2
+%ifidn %1,sp
+    paddd           m4, m7
+    paddd           m5, m7
+    psrad           m4, 12
+    psrad           m5, 12
+%else
+    psrad           m4, 6
+    psrad           m5, 6
+%endif
+    packssdw        m4, m5
+
+    movu            xm2, [r0 + r1]                  ; m2 = row 9
+    punpckhwd       xm5, xm0, xm2
+    punpcklwd       xm0, xm2
+    vinserti128     m0, m0, xm5, 1
+    pmaddwd         m0, [r5 + 1 * mmsize]
+    paddd           m6, m0
+    movu            xm5, [r0 + r1 * 2]              ; m5 = row 10
+    punpckhwd       xm0, xm2, xm5
+    punpcklwd       xm2, xm5
+    vinserti128     m2, m2, xm0, 1
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m1, m2
+
+%ifidn %1,sp
+    paddd           m6, m7
+    paddd           m1, m7
+    psrad           m6, 12
+    psrad           m1, 12
+%else
+    psrad           m6, 6
+    psrad           m1, 6
+%endif
+    packssdw        m6, m1
+%ifidn %1,sp
+    packuswb        m4, m6
+    vpermd          m4, m3, m4
+    vextracti128    xm6, m4, 1
+    movq            [r2], xm4
+    movhps          [r2 + r3], xm4
+    movq            [r2 + r3 * 2], xm6
+    movhps          [r2 + r4], xm6
+%else
+    vpermq          m4, m4, 11011000b
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm5, m4, 1
+    vextracti128    xm1, m6, 1
+    movu            [r2], xm4
+    movu            [r2 + r3], xm5
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r4], xm1
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_8x8 sp
+    FILTER_VER_CHROMA_S_AVX2_8x8 ss
+
+%macro PROCESS_CHROMA_S_AVX2_W8_16R 1
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m2, [r5]
+    lea             r7, [r0 + r1 * 4]
+    movu            xm4, [r7]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    paddd           m1, m5
+    pmaddwd         m3, [r5]
+    movu            xm5, [r7 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m2, m6
+    pmaddwd         m4, [r5]
+    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm7, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddwd         m7, m5, [r5 + 1 * mmsize]
+    paddd           m3, m7
+    pmaddwd         m5, [r5]
+%ifidn %1,sp
+    paddd           m0, m9
+    paddd           m1, m9
+    paddd           m2, m9
+    paddd           m3, m9
+    psrad           m0, 12
+    psrad           m1, 12
+    psrad           m2, 12
+    psrad           m3, 12
+%else
+    psrad           m0, 6
+    psrad           m1, 6
+    psrad           m2, 6
+    psrad           m3, 6
+%endif
+    packssdw        m0, m1
+    packssdw        m2, m3
+%ifidn %1,sp
+    packuswb        m0, m2
+    mova            m3, [v4_interp8_hps_shuf]
+    vpermd          m0, m3, m0
+    vextracti128    xm2, m0, 1
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r6], xm2
+%else
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m2, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
+%endif
+
+    movu            xm7, [r7 + r4]                  ; m7 = row 7
+    punpckhwd       xm8, xm6, xm7
+    punpcklwd       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddwd         m8, m6, [r5 + 1 * mmsize]
+    paddd           m4, m8
+    pmaddwd         m6, [r5]
+    lea             r7, [r7 + r1 * 4]
+    movu            xm8, [r7]                       ; m8 = row 8
+    punpckhwd       xm0, xm7, xm8
+    punpcklwd       xm7, xm8
+    vinserti128     m7, m7, xm0, 1
+    pmaddwd         m0, m7, [r5 + 1 * mmsize]
+    paddd           m5, m0
+    pmaddwd         m7, [r5]
+    movu            xm0, [r7 + r1]                  ; m0 = row 9
+    punpckhwd       xm1, xm8, xm0
+    punpcklwd       xm8, xm0
+    vinserti128     m8, m8, xm1, 1
+    pmaddwd         m1, m8, [r5 + 1 * mmsize]
+    paddd           m6, m1
+    pmaddwd         m8, [r5]
+    movu            xm1, [r7 + r1 * 2]              ; m1 = row 10
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m2, m0, [r5 + 1 * mmsize]
+    paddd           m7, m2
+    pmaddwd         m0, [r5]
+%ifidn %1,sp
+    paddd           m4, m9
+    paddd           m5, m9
+    psrad           m4, 12
+    psrad           m5, 12
+    paddd           m6, m9
+    paddd           m7, m9
+    psrad           m6, 12
+    psrad           m7, 12
+%else
+    psrad           m4, 6
+    psrad           m5, 6
+    psrad           m6, 6
+    psrad           m7, 6
+%endif
+    packssdw        m4, m5
+    packssdw        m6, m7
+    lea             r8, [r2 + r3 * 4]
+%ifidn %1,sp
+    packuswb        m4, m6
+    vpermd          m4, m3, m4
+    vextracti128    xm6, m4, 1
+    movq            [r8], xm4
+    movhps          [r8 + r3], xm4
+    movq            [r8 + r3 * 2], xm6
+    movhps          [r8 + r6], xm6
+%else
+    vpermq          m4, m4, 11011000b
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm5, m4, 1
+    vextracti128    xm7, m6, 1
+    movu            [r8], xm4
+    movu            [r8 + r3], xm5
+    movu            [r8 + r3 * 2], xm6
+    movu            [r8 + r6], xm7
+%endif
+
+    movu            xm2, [r7 + r4]                  ; m2 = row 11
+    punpckhwd       xm4, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm4, 1
+    pmaddwd         m4, m1, [r5 + 1 * mmsize]
+    paddd           m8, m4
+    pmaddwd         m1, [r5]
+    lea             r7, [r7 + r1 * 4]
+    movu            xm4, [r7]                       ; m4 = row 12
+    punpckhwd       xm5, xm2, xm4
+    punpcklwd       xm2, xm4
+    vinserti128     m2, m2, xm5, 1
+    pmaddwd         m5, m2, [r5 + 1 * mmsize]
+    paddd           m0, m5
+    pmaddwd         m2, [r5]
+    movu            xm5, [r7 + r1]                  ; m5 = row 13
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m1, m6
+    pmaddwd         m4, [r5]
+    movu            xm6, [r7 + r1 * 2]              ; m6 = row 14
+    punpckhwd       xm7, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddwd         m7, m5, [r5 + 1 * mmsize]
+    paddd           m2, m7
+    pmaddwd         m5, [r5]
+%ifidn %1,sp
+    paddd           m8, m9
+    paddd           m0, m9
+    paddd           m1, m9
+    paddd           m2, m9
+    psrad           m8, 12
+    psrad           m0, 12
+    psrad           m1, 12
+    psrad           m2, 12
+%else
+    psrad           m8, 6
+    psrad           m0, 6
+    psrad           m1, 6
+    psrad           m2, 6
+%endif
+    packssdw        m8, m0
+    packssdw        m1, m2
+    lea             r8, [r8 + r3 * 4]
+%ifidn %1,sp
+    packuswb        m8, m1
+    vpermd          m8, m3, m8
+    vextracti128    xm1, m8, 1
+    movq            [r8], xm8
+    movhps          [r8 + r3], xm8
+    movq            [r8 + r3 * 2], xm1
+    movhps          [r8 + r6], xm1
+%else
+    vpermq          m8, m8, 11011000b
+    vpermq          m1, m1, 11011000b
+    vextracti128    xm0, m8, 1
+    vextracti128    xm2, m1, 1
+    movu            [r8], xm8
+    movu            [r8 + r3], xm0
+    movu            [r8 + r3 * 2], xm1
+    movu            [r8 + r6], xm2
+%endif
+    lea             r8, [r8 + r3 * 4]
+
+    movu            xm7, [r7 + r4]                  ; m7 = row 15
+    punpckhwd       xm2, xm6, xm7
+    punpcklwd       xm6, xm7
+    vinserti128     m6, m6, xm2, 1
+    pmaddwd         m2, m6, [r5 + 1 * mmsize]
+    paddd           m4, m2
+    pmaddwd         m6, [r5]
+    lea             r7, [r7 + r1 * 4]
+    movu            xm2, [r7]                       ; m2 = row 16
+    punpckhwd       xm1, xm7, xm2
+    punpcklwd       xm7, xm2
+    vinserti128     m7, m7, xm1, 1
+    pmaddwd         m1, m7, [r5 + 1 * mmsize]
+    paddd           m5, m1
+    pmaddwd         m7, [r5]
+    movu            xm1, [r7 + r1]                  ; m1 = row 17
+    punpckhwd       xm0, xm2, xm1
+    punpcklwd       xm2, xm1
+    vinserti128     m2, m2, xm0, 1
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m6, m2
+    movu            xm0, [r7 + r1 * 2]              ; m0 = row 18
+    punpckhwd       xm2, xm1, xm0
+    punpcklwd       xm1, xm0
+    vinserti128     m1, m1, xm2, 1
+    pmaddwd         m1, [r5 + 1 * mmsize]
+    paddd           m7, m1
+
+%ifidn %1,sp
+    paddd           m4, m9
+    paddd           m5, m9
+    paddd           m6, m9
+    paddd           m7, m9
+    psrad           m4, 12
+    psrad           m5, 12
+    psrad           m6, 12
+    psrad           m7, 12
+%else
+    psrad           m4, 6
+    psrad           m5, 6
+    psrad           m6, 6
+    psrad           m7, 6
+%endif
+    packssdw        m4, m5
+    packssdw        m6, m7
+%ifidn %1,sp
+    packuswb        m4, m6
+    vpermd          m4, m3, m4
+    vextracti128    xm6, m4, 1
+    movq            [r8], xm4
+    movhps          [r8 + r3], xm4
+    movq            [r8 + r3 * 2], xm6
+    movhps          [r8 + r6], xm6
+%else
+    vpermq          m4, m4, 11011000b
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm5, m4, 1
+    vextracti128    xm7, m6, 1
+    movu            [r8], xm4
+    movu            [r8 + r3], xm5
+    movu            [r8 + r3 * 2], xm6
+    movu            [r8 + r6], xm7
+%endif
+%endmacro
+
+%macro FILTER_VER_CHROMA_S_AVX2_Nx16 2
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_vert_%1_%2x16, 4, 10, 10
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,sp
+    mova            m9, [v4_pd_526336]
+%else
+    add             r3d, r3d
+%endif
+    lea             r6, [r3 * 3]
+    mov             r9d, %2 / 8
+.loopW:
+    PROCESS_CHROMA_S_AVX2_W8_16R %1
+%ifidn %1,sp
+    add             r2, 8
+%else
+    add             r2, 16
+%endif
+    add             r0, 16
+    dec             r9d
+    jnz             .loopW
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_Nx16 sp, 16
+    FILTER_VER_CHROMA_S_AVX2_Nx16 sp, 32
+    FILTER_VER_CHROMA_S_AVX2_Nx16 sp, 64
+    FILTER_VER_CHROMA_S_AVX2_Nx16 ss, 16
+    FILTER_VER_CHROMA_S_AVX2_Nx16 ss, 32
+    FILTER_VER_CHROMA_S_AVX2_Nx16 ss, 64
+
+%macro FILTER_VER_CHROMA_S_AVX2_NxN 3
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_vert_%3_%1x%2, 4, 11, 10
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %3,sp
+    mova            m9, [v4_pd_526336]
+%else
+    add             r3d, r3d
+%endif
+    lea             r6, [r3 * 3]
+    mov             r9d, %2 / 16
+.loopH:
+    mov             r10d, %1 / 8
+.loopW:
+    PROCESS_CHROMA_S_AVX2_W8_16R %3
+%ifidn %3,sp
+    add             r2, 8
+%else
+    add             r2, 16
+%endif
+    add             r0, 16
+    dec             r10d
+    jnz             .loopW
+    lea             r0, [r7 - 2 * %1 + 16]
+%ifidn %3,sp
+    lea             r2, [r8 + r3 * 4 - %1 + 8]
+%else
+    lea             r2, [r8 + r3 * 4 - 2 * %1 + 16]
+%endif
+    dec             r9d
+    jnz             .loopH
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_NxN 16, 32, sp
+    FILTER_VER_CHROMA_S_AVX2_NxN 24, 32, sp
+    FILTER_VER_CHROMA_S_AVX2_NxN 32, 32, sp
+    FILTER_VER_CHROMA_S_AVX2_NxN 16, 32, ss
+    FILTER_VER_CHROMA_S_AVX2_NxN 24, 32, ss
+    FILTER_VER_CHROMA_S_AVX2_NxN 32, 32, ss
+    FILTER_VER_CHROMA_S_AVX2_NxN 16, 64, sp
+    FILTER_VER_CHROMA_S_AVX2_NxN 24, 64, sp
+    FILTER_VER_CHROMA_S_AVX2_NxN 32, 64, sp
+    FILTER_VER_CHROMA_S_AVX2_NxN 32, 48, sp
+    FILTER_VER_CHROMA_S_AVX2_NxN 32, 48, ss
+    FILTER_VER_CHROMA_S_AVX2_NxN 16, 64, ss
+    FILTER_VER_CHROMA_S_AVX2_NxN 24, 64, ss
+    FILTER_VER_CHROMA_S_AVX2_NxN 32, 64, ss
+    FILTER_VER_CHROMA_S_AVX2_NxN 64, 64, sp
+    FILTER_VER_CHROMA_S_AVX2_NxN 64, 32, sp
+    FILTER_VER_CHROMA_S_AVX2_NxN 64, 48, sp
+    FILTER_VER_CHROMA_S_AVX2_NxN 48, 64, sp
+    FILTER_VER_CHROMA_S_AVX2_NxN 64, 64, ss
+    FILTER_VER_CHROMA_S_AVX2_NxN 64, 32, ss
+    FILTER_VER_CHROMA_S_AVX2_NxN 64, 48, ss
+    FILTER_VER_CHROMA_S_AVX2_NxN 48, 64, ss
+
+%macro PROCESS_CHROMA_S_AVX2_W8_4R 1
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m2, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    paddd           m1, m5
+    pmaddwd         m3, [r5]
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m4, [r5 + 1 * mmsize]
+    paddd           m2, m4
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm4, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm4, 1
+    pmaddwd         m5, [r5 + 1 * mmsize]
+    paddd           m3, m5
+%ifidn %1,sp
+    paddd           m0, m7
+    paddd           m1, m7
+    paddd           m2, m7
+    paddd           m3, m7
+    psrad           m0, 12
+    psrad           m1, 12
+    psrad           m2, 12
+    psrad           m3, 12
+%else
+    psrad           m0, 6
+    psrad           m1, 6
+    psrad           m2, 6
+    psrad           m3, 6
+%endif
+    packssdw        m0, m1
+    packssdw        m2, m3
+%ifidn %1,sp
+    packuswb        m0, m2
+    mova            m3, [v4_interp8_hps_shuf]
+    vpermd          m0, m3, m0
+    vextracti128    xm2, m0, 1
+%else
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m2, 1
+%endif
+%endmacro
+
+%macro FILTER_VER_CHROMA_S_AVX2_8x4 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_8x4, 4, 6, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,sp
+    mova            m7, [v4_pd_526336]
+%else
+    add             r3d, r3d
+%endif
+
+    PROCESS_CHROMA_S_AVX2_W8_4R %1
+    lea             r4, [r3 * 3]
+%ifidn %1,sp
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r4], xm2
+%else
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r4], xm3
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_8x4 sp
+    FILTER_VER_CHROMA_S_AVX2_8x4 ss
+
+%macro FILTER_VER_CHROMA_S_AVX2_12x16 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_vert_%1_12x16, 4, 9, 10
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,sp
+    mova            m9, [v4_pd_526336]
+%else
+    add             r3d, r3d
+%endif
+    lea             r6, [r3 * 3]
+    PROCESS_CHROMA_S_AVX2_W8_16R %1
+%ifidn %1,sp
+    add             r2, 8
+%else
+    add             r2, 16
+%endif
+    add             r0, 16
+    mova            m7, m9
+    PROCESS_CHROMA_AVX2_W4_16R %1
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_12x16 sp
+    FILTER_VER_CHROMA_S_AVX2_12x16 ss
+
+%macro FILTER_VER_CHROMA_S_AVX2_12x32 1
+%if ARCH_X86_64 == 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_12x32, 4, 9, 10
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1, sp
+    mova            m9, [v4_pd_526336]
+%else
+    add             r3d, r3d
+%endif
+    lea             r6, [r3 * 3]
+%rep 2
+    PROCESS_CHROMA_S_AVX2_W8_16R %1
+%ifidn %1, sp
+    add             r2, 8
+%else
+    add             r2, 16
+%endif
+    add             r0, 16
+    mova            m7, m9
+    PROCESS_CHROMA_AVX2_W4_16R %1
+    sub             r0, 16
+%ifidn %1, sp
+    lea             r2, [r2 + r3 * 4 - 8]
+%else
+    lea             r2, [r2 + r3 * 4 - 16]
+%endif
+%endrep
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_12x32 sp
+    FILTER_VER_CHROMA_S_AVX2_12x32 ss
+
+%macro FILTER_VER_CHROMA_S_AVX2_16x12 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_vert_%1_16x12, 4, 9, 9
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,sp
+    mova            m8, [v4_pd_526336]
+%else
+    add             r3d, r3d
+%endif
+    lea             r6, [r3 * 3]
+%rep 2
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m2, [r5]
+    lea             r7, [r0 + r1 * 4]
+    movu            xm4, [r7]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    paddd           m1, m5
+    pmaddwd         m3, [r5]
+%ifidn %1,sp
+    paddd           m0, m8
+    paddd           m1, m8
+    psrad           m0, 12
+    psrad           m1, 12
+%else
+    psrad           m0, 6
+    psrad           m1, 6
+%endif
+    packssdw        m0, m1
+
+    movu            xm5, [r7 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m2, m6
+    pmaddwd         m4, [r5]
+    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm1, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm1, 1
+    pmaddwd         m1, m5, [r5 + 1 * mmsize]
+    pmaddwd         m5, [r5]
+    paddd           m3, m1
+%ifidn %1,sp
+    paddd           m2, m8
+    paddd           m3, m8
+    psrad           m2, 12
+    psrad           m3, 12
+%else
+    psrad           m2, 6
+    psrad           m3, 6
+%endif
+    packssdw        m2, m3
+%ifidn %1,sp
+    packuswb        m0, m2
+    mova            m3, [v4_interp8_hps_shuf]
+    vpermd          m0, m3, m0
+    vextracti128    xm2, m0, 1
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r6], xm2
+%else
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    movu            [r2], xm0
+    vextracti128    xm0, m0, 1
+    vextracti128    xm3, m2, 1
+    movu            [r2 + r3], xm0
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
+%endif
+    lea             r8, [r2 + r3 * 4]
+
+    movu            xm1, [r7 + r4]                  ; m1 = row 7
+    punpckhwd       xm0, xm6, xm1
+    punpcklwd       xm6, xm1
+    vinserti128     m6, m6, xm0, 1
+    pmaddwd         m0, m6, [r5 + 1 * mmsize]
+    pmaddwd         m6, [r5]
+    paddd           m4, m0
+    lea             r7, [r7 + r1 * 4]
+    movu            xm0, [r7]                       ; m0 = row 8
+    punpckhwd       xm2, xm1, xm0
+    punpcklwd       xm1, xm0
+    vinserti128     m1, m1, xm2, 1
+    pmaddwd         m2, m1, [r5 + 1 * mmsize]
+    pmaddwd         m1, [r5]
+    paddd           m5, m2
+%ifidn %1,sp
+    paddd           m4, m8
+    paddd           m5, m8
+    psrad           m4, 12
+    psrad           m5, 12
+%else
+    psrad           m4, 6
+    psrad           m5, 6
+%endif
+    packssdw        m4, m5
+
+    movu            xm2, [r7 + r1]                  ; m2 = row 9
+    punpckhwd       xm5, xm0, xm2
+    punpcklwd       xm0, xm2
+    vinserti128     m0, m0, xm5, 1
+    pmaddwd         m5, m0, [r5 + 1 * mmsize]
+    paddd           m6, m5
+    pmaddwd         m0, [r5]
+    movu            xm5, [r7 + r1 * 2]              ; m5 = row 10
+    punpckhwd       xm7, xm2, xm5
+    punpcklwd       xm2, xm5
+    vinserti128     m2, m2, xm7, 1
+    pmaddwd         m7, m2, [r5 + 1 * mmsize]
+    paddd           m1, m7
+    pmaddwd         m2, [r5]
+
+%ifidn %1,sp
+    paddd           m6, m8
+    paddd           m1, m8
+    psrad           m6, 12
+    psrad           m1, 12
+%else
+    psrad           m6, 6
+    psrad           m1, 6
+%endif
+    packssdw        m6, m1
+%ifidn %1,sp
+    packuswb        m4, m6
+    vpermd          m4, m3, m4
+    vextracti128    xm6, m4, 1
+    movq            [r8], xm4
+    movhps          [r8 + r3], xm4
+    movq            [r8 + r3 * 2], xm6
+    movhps          [r8 + r6], xm6
+%else
+    vpermq          m4, m4, 11011000b
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm7, m4, 1
+    vextracti128    xm1, m6, 1
+    movu            [r8], xm4
+    movu            [r8 + r3], xm7
+    movu            [r8 + r3 * 2], xm6
+    movu            [r8 + r6], xm1
+%endif
+    lea             r8, [r8 + r3 * 4]
+
+    movu            xm7, [r7 + r4]                  ; m7 = row 11
+    punpckhwd       xm1, xm5, xm7
+    punpcklwd       xm5, xm7
+    vinserti128     m5, m5, xm1, 1
+    pmaddwd         m1, m5, [r5 + 1 * mmsize]
+    paddd           m0, m1
+    pmaddwd         m5, [r5]
+    lea             r7, [r7 + r1 * 4]
+    movu            xm1, [r7]                       ; m1 = row 12
+    punpckhwd       xm4, xm7, xm1
+    punpcklwd       xm7, xm1
+    vinserti128     m7, m7, xm4, 1
+    pmaddwd         m4, m7, [r5 + 1 * mmsize]
+    paddd           m2, m4
+    pmaddwd         m7, [r5]
+%ifidn %1,sp
+    paddd           m0, m8
+    paddd           m2, m8
+    psrad           m0, 12
+    psrad           m2, 12
+%else
+    psrad           m0, 6
+    psrad           m2, 6
+%endif
+    packssdw        m0, m2
+
+    movu            xm4, [r7 + r1]                  ; m4 = row 13
+    punpckhwd       xm2, xm1, xm4
+    punpcklwd       xm1, xm4
+    vinserti128     m1, m1, xm2, 1
+    pmaddwd         m1, [r5 + 1 * mmsize]
+    paddd           m5, m1
+    movu            xm2, [r7 + r1 * 2]              ; m2 = row 14
+    punpckhwd       xm6, xm4, xm2
+    punpcklwd       xm4, xm2
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m4, [r5 + 1 * mmsize]
+    paddd           m7, m4
+%ifidn %1,sp
+    paddd           m5, m8
+    paddd           m7, m8
+    psrad           m5, 12
+    psrad           m7, 12
+%else
+    psrad           m5, 6
+    psrad           m7, 6
+%endif
+    packssdw        m5, m7
+%ifidn %1,sp
+    packuswb        m0, m5
+    vpermd          m0, m3, m0
+    vextracti128    xm5, m0, 1
+    movq            [r8], xm0
+    movhps          [r8 + r3], xm0
+    movq            [r8 + r3 * 2], xm5
+    movhps          [r8 + r6], xm5
+    add             r2, 8
+%else
+    vpermq          m0, m0, 11011000b
+    vpermq          m5, m5, 11011000b
+    vextracti128    xm7, m0, 1
+    vextracti128    xm6, m5, 1
+    movu            [r8], xm0
+    movu            [r8 + r3], xm7
+    movu            [r8 + r3 * 2], xm5
+    movu            [r8 + r6], xm6
+    add             r2, 16
+%endif
+    add             r0, 16
+%endrep
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_16x12 sp
+    FILTER_VER_CHROMA_S_AVX2_16x12 ss
+
+%macro FILTER_VER_CHROMA_S_AVX2_8x12 1
+%if ARCH_X86_64 == 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_8x12, 4, 7, 9
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,sp
+    mova            m8, [v4_pd_526336]
+%else
+    add             r3d, r3d
+%endif
+    lea             r6, [r3 * 3]
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m2, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    paddd           m1, m5
+    pmaddwd         m3, [r5]
+%ifidn %1,sp
+    paddd           m0, m8
+    paddd           m1, m8
+    psrad           m0, 12
+    psrad           m1, 12
+%else
+    psrad           m0, 6
+    psrad           m1, 6
+%endif
+    packssdw        m0, m1
+
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m2, m6
+    pmaddwd         m4, [r5]
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm1, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm1, 1
+    pmaddwd         m1, m5, [r5 + 1 * mmsize]
+    pmaddwd         m5, [r5]
+    paddd           m3, m1
+%ifidn %1,sp
+    paddd           m2, m8
+    paddd           m3, m8
+    psrad           m2, 12
+    psrad           m3, 12
+%else
+    psrad           m2, 6
+    psrad           m3, 6
+%endif
+    packssdw        m2, m3
+%ifidn %1,sp
+    packuswb        m0, m2
+    mova            m3, [v4_interp8_hps_shuf]
+    vpermd          m0, m3, m0
+    vextracti128    xm2, m0, 1
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r6], xm2
+%else
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    movu            [r2], xm0
+    vextracti128    xm0, m0, 1
+    vextracti128    xm3, m2, 1
+    movu            [r2 + r3], xm0
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
+%endif
+    lea             r2, [r2 + r3 * 4]
+
+    movu            xm1, [r0 + r4]                  ; m1 = row 7
+    punpckhwd       xm0, xm6, xm1
+    punpcklwd       xm6, xm1
+    vinserti128     m6, m6, xm0, 1
+    pmaddwd         m0, m6, [r5 + 1 * mmsize]
+    pmaddwd         m6, [r5]
+    paddd           m4, m0
+    lea             r0, [r0 + r1 * 4]
+    movu            xm0, [r0]                       ; m0 = row 8
+    punpckhwd       xm2, xm1, xm0
+    punpcklwd       xm1, xm0
+    vinserti128     m1, m1, xm2, 1
+    pmaddwd         m2, m1, [r5 + 1 * mmsize]
+    pmaddwd         m1, [r5]
+    paddd           m5, m2
+%ifidn %1,sp
+    paddd           m4, m8
+    paddd           m5, m8
+    psrad           m4, 12
+    psrad           m5, 12
+%else
+    psrad           m4, 6
+    psrad           m5, 6
+%endif
+    packssdw        m4, m5
+
+    movu            xm2, [r0 + r1]                  ; m2 = row 9
+    punpckhwd       xm5, xm0, xm2
+    punpcklwd       xm0, xm2
+    vinserti128     m0, m0, xm5, 1
+    pmaddwd         m5, m0, [r5 + 1 * mmsize]
+    paddd           m6, m5
+    pmaddwd         m0, [r5]
+    movu            xm5, [r0 + r1 * 2]              ; m5 = row 10
+    punpckhwd       xm7, xm2, xm5
+    punpcklwd       xm2, xm5
+    vinserti128     m2, m2, xm7, 1
+    pmaddwd         m7, m2, [r5 + 1 * mmsize]
+    paddd           m1, m7
+    pmaddwd         m2, [r5]
+
+%ifidn %1,sp
+    paddd           m6, m8
+    paddd           m1, m8
+    psrad           m6, 12
+    psrad           m1, 12
+%else
+    psrad           m6, 6
+    psrad           m1, 6
+%endif
+    packssdw        m6, m1
+%ifidn %1,sp
+    packuswb        m4, m6
+    vpermd          m4, m3, m4
+    vextracti128    xm6, m4, 1
+    movq            [r2], xm4
+    movhps          [r2 + r3], xm4
+    movq            [r2 + r3 * 2], xm6
+    movhps          [r2 + r6], xm6
+%else
+    vpermq          m4, m4, 11011000b
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm7, m4, 1
+    vextracti128    xm1, m6, 1
+    movu            [r2], xm4
+    movu            [r2 + r3], xm7
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r6], xm1
+%endif
+    lea             r2, [r2 + r3 * 4]
+
+    movu            xm7, [r0 + r4]                  ; m7 = row 11
+    punpckhwd       xm1, xm5, xm7
+    punpcklwd       xm5, xm7
+    vinserti128     m5, m5, xm1, 1
+    pmaddwd         m1, m5, [r5 + 1 * mmsize]
+    paddd           m0, m1
+    pmaddwd         m5, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm1, [r0]                       ; m1 = row 12
+    punpckhwd       xm4, xm7, xm1
+    punpcklwd       xm7, xm1
+    vinserti128     m7, m7, xm4, 1
+    pmaddwd         m4, m7, [r5 + 1 * mmsize]
+    paddd           m2, m4
+    pmaddwd         m7, [r5]
+%ifidn %1,sp
+    paddd           m0, m8
+    paddd           m2, m8
+    psrad           m0, 12
+    psrad           m2, 12
+%else
+    psrad           m0, 6
+    psrad           m2, 6
+%endif
+    packssdw        m0, m2
+
+    movu            xm4, [r0 + r1]                  ; m4 = row 13
+    punpckhwd       xm2, xm1, xm4
+    punpcklwd       xm1, xm4
+    vinserti128     m1, m1, xm2, 1
+    pmaddwd         m1, [r5 + 1 * mmsize]
+    paddd           m5, m1
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 14
+    punpckhwd       xm6, xm4, xm2
+    punpcklwd       xm4, xm2
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m4, [r5 + 1 * mmsize]
+    paddd           m7, m4
+%ifidn %1,sp
+    paddd           m5, m8
+    paddd           m7, m8
+    psrad           m5, 12
+    psrad           m7, 12
+%else
+    psrad           m5, 6
+    psrad           m7, 6
+%endif
+    packssdw        m5, m7
+%ifidn %1,sp
+    packuswb        m0, m5
+    vpermd          m0, m3, m0
+    vextracti128    xm5, m0, 1
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm5
+    movhps          [r2 + r6], xm5
+%else
+    vpermq          m0, m0, 11011000b
+    vpermq          m5, m5, 11011000b
+    vextracti128    xm7, m0, 1
+    vextracti128    xm6, m5, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm7
+    movu            [r2 + r3 * 2], xm5
+    movu            [r2 + r6], xm6
+%endif
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_8x12 sp
+    FILTER_VER_CHROMA_S_AVX2_8x12 ss
+
+%macro FILTER_VER_CHROMA_S_AVX2_16x4 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_16x4, 4, 7, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,sp
+    mova            m7, [v4_pd_526336]
+%else
+    add             r3d, r3d
+%endif
+%rep 2
+    PROCESS_CHROMA_S_AVX2_W8_4R %1
+    lea             r6, [r3 * 3]
+%ifidn %1,sp
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r6], xm2
+    add             r2, 8
+%else
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
+    add             r2, 16
+%endif
+    lea             r6, [4 * r1 - 16]
+    sub             r0, r6
+%endrep
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_16x4 sp
+    FILTER_VER_CHROMA_S_AVX2_16x4 ss
+
+%macro PROCESS_CHROMA_S_AVX2_W8_8R 1
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m2, [r5]
+    lea             r7, [r0 + r1 * 4]
+    movu            xm4, [r7]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    paddd           m1, m5
+    pmaddwd         m3, [r5]
+%ifidn %1,sp
+    paddd           m0, m7
+    paddd           m1, m7
+    psrad           m0, 12
+    psrad           m1, 12
+%else
+    psrad           m0, 6
+    psrad           m1, 6
+%endif
+    packssdw        m0, m1
+
+    movu            xm5, [r7 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m2, m6
+    pmaddwd         m4, [r5]
+    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm1, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm1, 1
+    pmaddwd         m1, m5, [r5 + 1 * mmsize]
+    pmaddwd         m5, [r5]
+    paddd           m3, m1
+%ifidn %1,sp
+    paddd           m2, m7
+    paddd           m3, m7
+    psrad           m2, 12
+    psrad           m3, 12
+%else
+    psrad           m2, 6
+    psrad           m3, 6
+%endif
+    packssdw        m2, m3
+%ifidn %1,sp
+    packuswb        m0, m2
+    mova            m3, [v4_interp8_hps_shuf]
+    vpermd          m0, m3, m0
+    vextracti128    xm2, m0, 1
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r6], xm2
+%else
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    movu            [r2], xm0
+    vextracti128    xm0, m0, 1
+    vextracti128    xm3, m2, 1
+    movu            [r2 + r3], xm0
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
+%endif
+    lea             r8, [r2 + r3 * 4]
+
+    movu            xm1, [r7 + r4]                  ; m1 = row 7
+    punpckhwd       xm0, xm6, xm1
+    punpcklwd       xm6, xm1
+    vinserti128     m6, m6, xm0, 1
+    pmaddwd         m0, m6, [r5 + 1 * mmsize]
+    pmaddwd         m6, [r5]
+    paddd           m4, m0
+    lea             r7, [r7 + r1 * 4]
+    movu            xm0, [r7]                       ; m0 = row 8
+    punpckhwd       xm2, xm1, xm0
+    punpcklwd       xm1, xm0
+    vinserti128     m1, m1, xm2, 1
+    pmaddwd         m2, m1, [r5 + 1 * mmsize]
+    pmaddwd         m1, [r5]
+    paddd           m5, m2
+%ifidn %1,sp
+    paddd           m4, m7
+    paddd           m5, m7
+    psrad           m4, 12
+    psrad           m5, 12
+%else
+    psrad           m4, 6
+    psrad           m5, 6
+%endif
+    packssdw        m4, m5
+
+    movu            xm2, [r7 + r1]                  ; m2 = row 9
+    punpckhwd       xm5, xm0, xm2
+    punpcklwd       xm0, xm2
+    vinserti128     m0, m0, xm5, 1
+    pmaddwd         m0, [r5 + 1 * mmsize]
+    paddd           m6, m0
+    movu            xm5, [r7 + r1 * 2]              ; m5 = row 10
+    punpckhwd       xm0, xm2, xm5
+    punpcklwd       xm2, xm5
+    vinserti128     m2, m2, xm0, 1
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m1, m2
+
+%ifidn %1,sp
+    paddd           m6, m7
+    paddd           m1, m7
+    psrad           m6, 12
+    psrad           m1, 12
+%else
+    psrad           m6, 6
+    psrad           m1, 6
+%endif
+    packssdw        m6, m1
+%ifidn %1,sp
+    packuswb        m4, m6
+    vpermd          m4, m3, m4
+    vextracti128    xm6, m4, 1
+    movq            [r8], xm4
+    movhps          [r8 + r3], xm4
+    movq            [r8 + r3 * 2], xm6
+    movhps          [r8 + r6], xm6
+%else
+    vpermq          m4, m4, 11011000b
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm7, m4, 1
+    vextracti128    xm1, m6, 1
+    movu            [r8], xm4
+    movu            [r8 + r3], xm7
+    movu            [r8 + r3 * 2], xm6
+    movu            [r8 + r6], xm1
+%endif
+%endmacro
+
+%macro FILTER_VER_CHROMA_S_AVX2_Nx8 2
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_vert_%1_%2x8, 4, 9, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,sp
+    mova            m7, [v4_pd_526336]
+%else
+    add             r3d, r3d
+%endif
+    lea             r6, [r3 * 3]
+%rep %2 / 8
+    PROCESS_CHROMA_S_AVX2_W8_8R %1
+%ifidn %1,sp
+    add             r2, 8
+%else
+    add             r2, 16
+%endif
+    add             r0, 16
+%endrep
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_Nx8 sp, 32
+    FILTER_VER_CHROMA_S_AVX2_Nx8 sp, 16
+    FILTER_VER_CHROMA_S_AVX2_Nx8 ss, 32
+    FILTER_VER_CHROMA_S_AVX2_Nx8 ss, 16
+
+%macro FILTER_VER_CHROMA_S_AVX2_8x2 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_8x2, 4, 6, 6
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,sp
+    mova            m5, [v4_pd_526336]
+%else
+    add             r3d, r3d
+%endif
+
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m0, m2
+    movu            xm4, [r0 + r1 * 4]              ; m4 = row 4
+    punpckhwd       xm2, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm2, 1
+    pmaddwd         m3, [r5 + 1 * mmsize]
+    paddd           m1, m3
+%ifidn %1,sp
+    paddd           m0, m5
+    paddd           m1, m5
+    psrad           m0, 12
+    psrad           m1, 12
+%else
+    psrad           m0, 6
+    psrad           m1, 6
+%endif
+    packssdw        m0, m1
+%ifidn %1,sp
+    vextracti128    xm1, m0, 1
+    packuswb        xm0, xm1
+    pshufd          xm0, xm0, 11011000b
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+%else
+    vpermq          m0, m0, 11011000b
+    vextracti128    xm1, m0, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_8x2 sp
+    FILTER_VER_CHROMA_S_AVX2_8x2 ss
+
+%macro FILTER_VER_CHROMA_S_AVX2_8x6 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_8x6, 4, 6, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,sp
+    mova            m7, [v4_pd_526336]
+%else
+    add             r3d, r3d
+%endif
+
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    pmaddwd         m2, [r5]
+    paddd           m0, m4
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    pmaddwd         m3, [r5]
+    paddd           m1, m5
+%ifidn %1,sp
+    paddd           m0, m7
+    paddd           m1, m7
+    psrad           m0, 12
+    psrad           m1, 12
+%else
+    psrad           m0, 6
+    psrad           m1, 6
+%endif
+    packssdw        m0, m1
+
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m2, m6
+    pmaddwd         m4, [r5]
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm1, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm1, 1
+    pmaddwd         m1, m5, [r5 + 1 * mmsize]
+    pmaddwd         m5, [r5]
+    paddd           m3, m1
+%ifidn %1,sp
+    paddd           m2, m7
+    paddd           m3, m7
+    psrad           m2, 12
+    psrad           m3, 12
+%else
+    psrad           m2, 6
+    psrad           m3, 6
+%endif
+    packssdw        m2, m3
+
+    movu            xm1, [r0 + r4]                  ; m1 = row 7
+    punpckhwd       xm3, xm6, xm1
+    punpcklwd       xm6, xm1
+    vinserti128     m6, m6, xm3, 1
+    pmaddwd         m6, [r5 + 1 * mmsize]
+    paddd           m4, m6
+    movu            xm6, [r0 + r1 * 4]              ; m6 = row 8
+    punpckhwd       xm3, xm1, xm6
+    punpcklwd       xm1, xm6
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5 + 1 * mmsize]
+    paddd           m5, m1
+%ifidn %1,sp
+    paddd           m4, m7
+    paddd           m5, m7
+    psrad           m4, 12
+    psrad           m5, 12
+%else
+    psrad           m4, 6
+    psrad           m5, 6
+%endif
+    packssdw        m4, m5
+    lea             r4, [r3 * 3]
+%ifidn %1,sp
+    packuswb        m0, m2
+    mova            m3, [v4_interp8_hps_shuf]
+    vpermd          m0, m3, m0
+    vextracti128    xm2, m0, 1
+    vextracti128    xm5, m4, 1
+    packuswb        xm4, xm5
+    pshufd          xm4, xm4, 11011000b
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r4], xm2
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm4
+    movhps          [r2 + r3], xm4
+%else
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    vpermq          m4, m4, 11011000b
+    movu            [r2], xm0
+    vextracti128    xm0, m0, 1
+    vextracti128    xm3, m2, 1
+    vextracti128    xm5, m4, 1
+    movu            [r2 + r3], xm0
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r4], xm3
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm4
+    movu            [r2 + r3], xm5
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_8x6 sp
+    FILTER_VER_CHROMA_S_AVX2_8x6 ss
+
+%macro FILTER_VER_CHROMA_S_AVX2_8xN 2
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_vert_%1_8x%2, 4, 7, 9
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,sp
+    mova            m8, [v4_pd_526336]
+%else
+    add             r3d, r3d
+%endif
+    lea             r6, [r3 * 3]
+%rep %2 / 16
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m2, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    paddd           m1, m5
+    pmaddwd         m3, [r5]
+%ifidn %1,sp
+    paddd           m0, m8
+    paddd           m1, m8
+    psrad           m0, 12
+    psrad           m1, 12
+%else
+    psrad           m0, 6
+    psrad           m1, 6
+%endif
+    packssdw        m0, m1
+
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m2, m6
+    pmaddwd         m4, [r5]
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm1, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm1, 1
+    pmaddwd         m1, m5, [r5 + 1 * mmsize]
+    pmaddwd         m5, [r5]
+    paddd           m3, m1
+%ifidn %1,sp
+    paddd           m2, m8
+    paddd           m3, m8
+    psrad           m2, 12
+    psrad           m3, 12
+%else
+    psrad           m2, 6
+    psrad           m3, 6
+%endif
+    packssdw        m2, m3
+%ifidn %1,sp
+    packuswb        m0, m2
+    mova            m3, [v4_interp8_hps_shuf]
+    vpermd          m0, m3, m0
+    vextracti128    xm2, m0, 1
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r6], xm2
+%else
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    movu            [r2], xm0
+    vextracti128    xm0, m0, 1
+    vextracti128    xm3, m2, 1
+    movu            [r2 + r3], xm0
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm3
+%endif
+    lea             r2, [r2 + r3 * 4]
+
+    movu            xm1, [r0 + r4]                  ; m1 = row 7
+    punpckhwd       xm0, xm6, xm1
+    punpcklwd       xm6, xm1
+    vinserti128     m6, m6, xm0, 1
+    pmaddwd         m0, m6, [r5 + 1 * mmsize]
+    pmaddwd         m6, [r5]
+    paddd           m4, m0
+    lea             r0, [r0 + r1 * 4]
+    movu            xm0, [r0]                       ; m0 = row 8
+    punpckhwd       xm2, xm1, xm0
+    punpcklwd       xm1, xm0
+    vinserti128     m1, m1, xm2, 1
+    pmaddwd         m2, m1, [r5 + 1 * mmsize]
+    pmaddwd         m1, [r5]
+    paddd           m5, m2
+%ifidn %1,sp
+    paddd           m4, m8
+    paddd           m5, m8
+    psrad           m4, 12
+    psrad           m5, 12
+%else
+    psrad           m4, 6
+    psrad           m5, 6
+%endif
+    packssdw        m4, m5
+
+    movu            xm2, [r0 + r1]                  ; m2 = row 9
+    punpckhwd       xm5, xm0, xm2
+    punpcklwd       xm0, xm2
+    vinserti128     m0, m0, xm5, 1
+    pmaddwd         m5, m0, [r5 + 1 * mmsize]
+    paddd           m6, m5
+    pmaddwd         m0, [r5]
+    movu            xm5, [r0 + r1 * 2]              ; m5 = row 10
+    punpckhwd       xm7, xm2, xm5
+    punpcklwd       xm2, xm5
+    vinserti128     m2, m2, xm7, 1
+    pmaddwd         m7, m2, [r5 + 1 * mmsize]
+    paddd           m1, m7
+    pmaddwd         m2, [r5]
+
+%ifidn %1,sp
+    paddd           m6, m8
+    paddd           m1, m8
+    psrad           m6, 12
+    psrad           m1, 12
+%else
+    psrad           m6, 6
+    psrad           m1, 6
+%endif
+    packssdw        m6, m1
+%ifidn %1,sp
+    packuswb        m4, m6
+    vpermd          m4, m3, m4
+    vextracti128    xm6, m4, 1
+    movq            [r2], xm4
+    movhps          [r2 + r3], xm4
+    movq            [r2 + r3 * 2], xm6
+    movhps          [r2 + r6], xm6
+%else
+    vpermq          m4, m4, 11011000b
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm7, m4, 1
+    vextracti128    xm1, m6, 1
+    movu            [r2], xm4
+    movu            [r2 + r3], xm7
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r6], xm1
+%endif
+    lea             r2, [r2 + r3 * 4]
+
+    movu            xm7, [r0 + r4]                  ; m7 = row 11
+    punpckhwd       xm1, xm5, xm7
+    punpcklwd       xm5, xm7
+    vinserti128     m5, m5, xm1, 1
+    pmaddwd         m1, m5, [r5 + 1 * mmsize]
+    paddd           m0, m1
+    pmaddwd         m5, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm1, [r0]                       ; m1 = row 12
+    punpckhwd       xm4, xm7, xm1
+    punpcklwd       xm7, xm1
+    vinserti128     m7, m7, xm4, 1
+    pmaddwd         m4, m7, [r5 + 1 * mmsize]
+    paddd           m2, m4
+    pmaddwd         m7, [r5]
+%ifidn %1,sp
+    paddd           m0, m8
+    paddd           m2, m8
+    psrad           m0, 12
+    psrad           m2, 12
+%else
+    psrad           m0, 6
+    psrad           m2, 6
+%endif
+    packssdw        m0, m2
+
+    movu            xm4, [r0 + r1]                  ; m4 = row 13
+    punpckhwd       xm2, xm1, xm4
+    punpcklwd       xm1, xm4
+    vinserti128     m1, m1, xm2, 1
+    pmaddwd         m2, m1, [r5 + 1 * mmsize]
+    paddd           m5, m2
+    pmaddwd         m1, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 14
+    punpckhwd       xm6, xm4, xm2
+    punpcklwd       xm4, xm2
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m7, m6
+    pmaddwd         m4, [r5]
+%ifidn %1,sp
+    paddd           m5, m8
+    paddd           m7, m8
+    psrad           m5, 12
+    psrad           m7, 12
+%else
+    psrad           m5, 6
+    psrad           m7, 6
+%endif
+    packssdw        m5, m7
+%ifidn %1,sp
+    packuswb        m0, m5
+    vpermd          m0, m3, m0
+    vextracti128    xm5, m0, 1
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm5
+    movhps          [r2 + r6], xm5
+%else
+    vpermq          m0, m0, 11011000b
+    vpermq          m5, m5, 11011000b
+    vextracti128    xm7, m0, 1
+    vextracti128    xm6, m5, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm7
+    movu            [r2 + r3 * 2], xm5
+    movu            [r2 + r6], xm6
+%endif
+    lea             r2, [r2 + r3 * 4]
+
+    movu            xm6, [r0 + r4]                  ; m6 = row 15
+    punpckhwd       xm5, xm2, xm6
+    punpcklwd       xm2, xm6
+    vinserti128     m2, m2, xm5, 1
+    pmaddwd         m5, m2, [r5 + 1 * mmsize]
+    paddd           m1, m5
+    pmaddwd         m2, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm0, [r0]                       ; m0 = row 16
+    punpckhwd       xm5, xm6, xm0
+    punpcklwd       xm6, xm0
+    vinserti128     m6, m6, xm5, 1
+    pmaddwd         m5, m6, [r5 + 1 * mmsize]
+    paddd           m4, m5
+    pmaddwd         m6, [r5]
+%ifidn %1,sp
+    paddd           m1, m8
+    paddd           m4, m8
+    psrad           m1, 12
+    psrad           m4, 12
+%else
+    psrad           m1, 6
+    psrad           m4, 6
+%endif
+    packssdw        m1, m4
+
+    movu            xm5, [r0 + r1]                  ; m5 = row 17
+    punpckhwd       xm4, xm0, xm5
+    punpcklwd       xm0, xm5
+    vinserti128     m0, m0, xm4, 1
+    pmaddwd         m0, [r5 + 1 * mmsize]
+    paddd           m2, m0
+    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
+    punpckhwd       xm0, xm5, xm4
+    punpcklwd       xm5, xm4
+    vinserti128     m5, m5, xm0, 1
+    pmaddwd         m5, [r5 + 1 * mmsize]
+    paddd           m6, m5
+%ifidn %1,sp
+    paddd           m2, m8
+    paddd           m6, m8
+    psrad           m2, 12
+    psrad           m6, 12
+%else
+    psrad           m2, 6
+    psrad           m6, 6
+%endif
+    packssdw        m2, m6
+%ifidn %1,sp
+    packuswb        m1, m2
+    vpermd          m1, m3, m1
+    vextracti128    xm2, m1, 1
+    movq            [r2], xm1
+    movhps          [r2 + r3], xm1
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r6], xm2
+%else
+    vpermq          m1, m1, 11011000b
+    vpermq          m2, m2, 11011000b
+    vextracti128    xm6, m1, 1
+    vextracti128    xm4, m2, 1
+    movu            [r2], xm1
+    movu            [r2 + r3], xm6
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm4
+%endif
+    lea             r2, [r2 + r3 * 4]
+%endrep
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_8xN sp, 16
+    FILTER_VER_CHROMA_S_AVX2_8xN sp, 32
+    FILTER_VER_CHROMA_S_AVX2_8xN sp, 64
+    FILTER_VER_CHROMA_S_AVX2_8xN ss, 16
+    FILTER_VER_CHROMA_S_AVX2_8xN ss, 32
+    FILTER_VER_CHROMA_S_AVX2_8xN ss, 64
+
+%macro FILTER_VER_CHROMA_S_AVX2_Nx24 2
+%if ARCH_X86_64 == 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_%2x24, 4, 10, 10
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,sp
+    mova            m9, [v4_pd_526336]
+%else
+    add             r3d, r3d
+%endif
+    lea             r6, [r3 * 3]
+    mov             r9d, %2 / 8
+.loopW:
+    PROCESS_CHROMA_S_AVX2_W8_16R %1
+%ifidn %1,sp
+    add             r2, 8
+%else
+    add             r2, 16
+%endif
+    add             r0, 16
+    dec             r9d
+    jnz             .loopW
+%ifidn %1,sp
+    lea             r2, [r8 + r3 * 4 - %2 + 8]
+%else
+    lea             r2, [r8 + r3 * 4 - 2 * %2 + 16]
+%endif
+    lea             r0, [r7 - 2 * %2 + 16]
+    mova            m7, m9
+    mov             r9d, %2 / 8
+.loop:
+    PROCESS_CHROMA_S_AVX2_W8_8R %1
+%ifidn %1,sp
+    add             r2, 8
+%else
+    add             r2, 16
+%endif
+    add             r0, 16
+    dec             r9d
+    jnz             .loop
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_Nx24 sp, 32
+    FILTER_VER_CHROMA_S_AVX2_Nx24 sp, 16
+    FILTER_VER_CHROMA_S_AVX2_Nx24 ss, 32
+    FILTER_VER_CHROMA_S_AVX2_Nx24 ss, 16
+
+%macro FILTER_VER_CHROMA_S_AVX2_2x8 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_2x8, 4, 6, 7
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+    sub             r0, r1
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+%ifidn %1,sp
+    mova            m6, [v4_pd_526336]
+%else
+    add             r3d, r3d
+%endif
+    movd            xm0, [r0]
+    movd            xm1, [r0 + r1]
+    punpcklwd       xm0, xm1
+    movd            xm2, [r0 + r1 * 2]
+    punpcklwd       xm1, xm2
+    punpcklqdq      xm0, xm1                        ; m0 = [2 1 1 0]
+    movd            xm3, [r0 + r4]
+    punpcklwd       xm2, xm3
+    lea             r0, [r0 + 4 * r1]
+    movd            xm4, [r0]
+    punpcklwd       xm3, xm4
+    punpcklqdq      xm2, xm3                        ; m2 = [4 3 3 2]
+    vinserti128     m0, m0, xm2, 1                  ; m0 = [4 3 3 2 2 1 1 0]
+    movd            xm1, [r0 + r1]
+    punpcklwd       xm4, xm1
+    movd            xm3, [r0 + r1 * 2]
+    punpcklwd       xm1, xm3
+    punpcklqdq      xm4, xm1                        ; m4 = [6 5 5 4]
+    vinserti128     m2, m2, xm4, 1                  ; m2 = [6 5 5 4 4 3 3 2]
+    pmaddwd         m0, [r5]
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m0, m2
+    movd            xm1, [r0 + r4]
+    punpcklwd       xm3, xm1
+    lea             r0, [r0 + 4 * r1]
+    movd            xm2, [r0]
+    punpcklwd       xm1, xm2
+    punpcklqdq      xm3, xm1                        ; m3 = [8 7 7 6]
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [8 7 7 6 6 5 5 4]
+    movd            xm1, [r0 + r1]
+    punpcklwd       xm2, xm1
+    movd            xm5, [r0 + r1 * 2]
+    punpcklwd       xm1, xm5
+    punpcklqdq      xm2, xm1                        ; m2 = [10 9 9 8]
+    vinserti128     m3, m3, xm2, 1                  ; m3 = [10 9 9 8 8 7 7 6]
+    pmaddwd         m4, [r5]
+    pmaddwd         m3, [r5 + 1 * mmsize]
+    paddd           m4, m3
+%ifidn %1,sp
+    paddd           m0, m6
+    paddd           m4, m6
+    psrad           m0, 12
+    psrad           m4, 12
+%else
+    psrad           m0, 6
+    psrad           m4, 6
+%endif
+    packssdw        m0, m4
+    vextracti128    xm4, m0, 1
+    lea             r4, [r3 * 3]
+%ifidn %1,sp
+    packuswb        xm0, xm4
+    pextrw          [r2], xm0, 0
+    pextrw          [r2 + r3], xm0, 1
+    pextrw          [r2 + 2 * r3], xm0, 4
+    pextrw          [r2 + r4], xm0, 5
+    lea             r2, [r2 + r3 * 4]
+    pextrw          [r2], xm0, 2
+    pextrw          [r2 + r3], xm0, 3
+    pextrw          [r2 + 2 * r3], xm0, 6
+    pextrw          [r2 + r4], xm0, 7
+%else
+    movd            [r2], xm0
+    pextrd          [r2 + r3], xm0, 1
+    movd            [r2 + 2 * r3], xm4
+    pextrd          [r2 + r4], xm4, 1
+    lea             r2, [r2 + r3 * 4]
+    pextrd          [r2], xm0, 2
+    pextrd          [r2 + r3], xm0, 3
+    pextrd          [r2 + 2 * r3], xm4, 2
+    pextrd          [r2 + r4], xm4, 3
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_2x8 sp
+    FILTER_VER_CHROMA_S_AVX2_2x8 ss
+
+%macro FILTER_VER_CHROMA_S_AVX2_2x16 1
+%if ARCH_X86_64 == 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_2x16, 4, 6, 9
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+    sub             r0, r1
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+%ifidn %1,sp
+    mova            m6, [v4_pd_526336]
+%else
+    add             r3d, r3d
+%endif
+    movd            xm0, [r0]
+    movd            xm1, [r0 + r1]
+    punpcklwd       xm0, xm1
+    movd            xm2, [r0 + r1 * 2]
+    punpcklwd       xm1, xm2
+    punpcklqdq      xm0, xm1                        ; m0 = [2 1 1 0]
+    movd            xm3, [r0 + r4]
+    punpcklwd       xm2, xm3
+    lea             r0, [r0 + 4 * r1]
+    movd            xm4, [r0]
+    punpcklwd       xm3, xm4
+    punpcklqdq      xm2, xm3                        ; m2 = [4 3 3 2]
+    vinserti128     m0, m0, xm2, 1                  ; m0 = [4 3 3 2 2 1 1 0]
+    movd            xm1, [r0 + r1]
+    punpcklwd       xm4, xm1
+    movd            xm3, [r0 + r1 * 2]
+    punpcklwd       xm1, xm3
+    punpcklqdq      xm4, xm1                        ; m4 = [6 5 5 4]
+    vinserti128     m2, m2, xm4, 1                  ; m2 = [6 5 5 4 4 3 3 2]
+    pmaddwd         m0, [r5]
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m0, m2
+    movd            xm1, [r0 + r4]
+    punpcklwd       xm3, xm1
+    lea             r0, [r0 + 4 * r1]
+    movd            xm2, [r0]
+    punpcklwd       xm1, xm2
+    punpcklqdq      xm3, xm1                        ; m3 = [8 7 7 6]
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [8 7 7 6 6 5 5 4]
+    movd            xm1, [r0 + r1]
+    punpcklwd       xm2, xm1
+    movd            xm5, [r0 + r1 * 2]
+    punpcklwd       xm1, xm5
+    punpcklqdq      xm2, xm1                        ; m2 = [10 9 9 8]
+    vinserti128     m3, m3, xm2, 1                  ; m3 = [10 9 9 8 8 7 7 6]
+    pmaddwd         m4, [r5]
+    pmaddwd         m3, [r5 + 1 * mmsize]
+    paddd           m4, m3
+    movd            xm1, [r0 + r4]
+    punpcklwd       xm5, xm1
+    lea             r0, [r0 + 4 * r1]
+    movd            xm3, [r0]
+    punpcklwd       xm1, xm3
+    punpcklqdq      xm5, xm1                        ; m5 = [12 11 11 10]
+    vinserti128     m2, m2, xm5, 1                  ; m2 = [12 11 11 10 10 9 9 8]
+    movd            xm1, [r0 + r1]
+    punpcklwd       xm3, xm1
+    movd            xm7, [r0 + r1 * 2]
+    punpcklwd       xm1, xm7
+    punpcklqdq      xm3, xm1                        ; m3 = [14 13 13 12]
+    vinserti128     m5, m5, xm3, 1                  ; m5 = [14 13 13 12 12 11 11 10]
+    pmaddwd         m2, [r5]
+    pmaddwd         m5, [r5 + 1 * mmsize]
+    paddd           m2, m5
+    movd            xm5, [r0 + r4]
+    punpcklwd       xm7, xm5
+    lea             r0, [r0 + 4 * r1]
+    movd            xm1, [r0]
+    punpcklwd       xm5, xm1
+    punpcklqdq      xm7, xm5                        ; m7 = [16 15 15 14]
+    vinserti128     m3, m3, xm7, 1                  ; m3 = [16 15 15 14 14 13 13 12]
+    movd            xm5, [r0 + r1]
+    punpcklwd       xm1, xm5
+    movd            xm8, [r0 + r1 * 2]
+    punpcklwd       xm5, xm8
+    punpcklqdq      xm1, xm5                        ; m1 = [18 17 17 16]
+    vinserti128     m7, m7, xm1, 1                  ; m7 = [18 17 17 16 16 15 15 14]
+    pmaddwd         m3, [r5]
+    pmaddwd         m7, [r5 + 1 * mmsize]
+    paddd           m3, m7
+%ifidn %1,sp
+    paddd           m0, m6
+    paddd           m4, m6
+    paddd           m2, m6
+    paddd           m3, m6
+    psrad           m0, 12
+    psrad           m4, 12
+    psrad           m2, 12
+    psrad           m3, 12
+%else
+    psrad           m0, 6
+    psrad           m4, 6
+    psrad           m2, 6
+    psrad           m3, 6
+%endif
+    packssdw        m0, m4
+    packssdw        m2, m3
+    lea             r4, [r3 * 3]
+%ifidn %1,sp
+    packuswb        m0, m2
+    vextracti128    xm2, m0, 1
+    pextrw          [r2], xm0, 0
+    pextrw          [r2 + r3], xm0, 1
+    pextrw          [r2 + 2 * r3], xm2, 0
+    pextrw          [r2 + r4], xm2, 1
+    lea             r2, [r2 + r3 * 4]
+    pextrw          [r2], xm0, 2
+    pextrw          [r2 + r3], xm0, 3
+    pextrw          [r2 + 2 * r3], xm2, 2
+    pextrw          [r2 + r4], xm2, 3
+    lea             r2, [r2 + r3 * 4]
+    pextrw          [r2], xm0, 4
+    pextrw          [r2 + r3], xm0, 5
+    pextrw          [r2 + 2 * r3], xm2, 4
+    pextrw          [r2 + r4], xm2, 5
+    lea             r2, [r2 + r3 * 4]
+    pextrw          [r2], xm0, 6
+    pextrw          [r2 + r3], xm0, 7
+    pextrw          [r2 + 2 * r3], xm2, 6
+    pextrw          [r2 + r4], xm2, 7
+%else
+    vextracti128    xm4, m0, 1
+    vextracti128    xm3, m2, 1
+    movd            [r2], xm0
+    pextrd          [r2 + r3], xm0, 1
+    movd            [r2 + 2 * r3], xm4
+    pextrd          [r2 + r4], xm4, 1
+    lea             r2, [r2 + r3 * 4]
+    pextrd          [r2], xm0, 2
+    pextrd          [r2 + r3], xm0, 3
+    pextrd          [r2 + 2 * r3], xm4, 2
+    pextrd          [r2 + r4], xm4, 3
+    lea             r2, [r2 + r3 * 4]
+    movd            [r2], xm2
+    pextrd          [r2 + r3], xm2, 1
+    movd            [r2 + 2 * r3], xm3
+    pextrd          [r2 + r4], xm3, 1
+    lea             r2, [r2 + r3 * 4]
+    pextrd          [r2], xm2, 2
+    pextrd          [r2 + r3], xm2, 3
+    pextrd          [r2 + 2 * r3], xm3, 2
+    pextrd          [r2 + r4], xm3, 3
+%endif
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_2x16 sp
+    FILTER_VER_CHROMA_S_AVX2_2x16 ss
+
+%macro FILTER_VER_CHROMA_S_AVX2_6x8 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_6x8, 4, 6, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,sp
+    mova            m7, [v4_pd_526336]
+%else
+    add             r3d, r3d
+%endif
+
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    pmaddwd         m2, [r5]
+    paddd           m0, m4
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    pmaddwd         m3, [r5]
+    paddd           m1, m5
+%ifidn %1,sp
+    paddd           m0, m7
+    paddd           m1, m7
+    psrad           m0, 12
+    psrad           m1, 12
+%else
+    psrad           m0, 6
+    psrad           m1, 6
+%endif
+    packssdw        m0, m1
+
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m2, m6
+    pmaddwd         m4, [r5]
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm1, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm1, 1
+    pmaddwd         m1, m5, [r5 + 1 * mmsize]
+    pmaddwd         m5, [r5]
+    paddd           m3, m1
+%ifidn %1,sp
+    paddd           m2, m7
+    paddd           m3, m7
+    psrad           m2, 12
+    psrad           m3, 12
+%else
+    psrad           m2, 6
+    psrad           m3, 6
+%endif
+    packssdw        m2, m3
+
+    movu            xm1, [r0 + r4]                  ; m1 = row 7
+    punpckhwd       xm3, xm6, xm1
+    punpcklwd       xm6, xm1
+    vinserti128     m6, m6, xm3, 1
+    pmaddwd         m3, m6, [r5 + 1 * mmsize]
+    pmaddwd         m6, [r5]
+    paddd           m4, m3
+
+    lea             r4, [r3 * 3]
+%ifidn %1,sp
+    packuswb        m0, m2
+    vextracti128    xm2, m0, 1
+    movd            [r2], xm0
+    pextrw          [r2 + 4], xm2, 0
+    pextrd          [r2 + r3], xm0, 1
+    pextrw          [r2 + r3 + 4], xm2, 2
+    pextrd          [r2 + r3 * 2], xm0, 2
+    pextrw          [r2 + r3 * 2 + 4], xm2, 4
+    pextrd          [r2 + r4], xm0, 3
+    pextrw          [r2 + r4 + 4], xm2, 6
+%else
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r4], xm2
+    vextracti128    xm0, m0, 1
+    vextracti128    xm3, m2, 1
+    movd            [r2 + 8], xm0
+    pextrd          [r2 + r3 + 8], xm0, 2
+    movd            [r2 + r3 * 2 + 8], xm3
+    pextrd          [r2 + r4 + 8], xm3, 2
+%endif
+    lea             r2, [r2 + r3 * 4]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm0, [r0]                       ; m0 = row 8
+    punpckhwd       xm2, xm1, xm0
+    punpcklwd       xm1, xm0
+    vinserti128     m1, m1, xm2, 1
+    pmaddwd         m2, m1, [r5 + 1 * mmsize]
+    pmaddwd         m1, [r5]
+    paddd           m5, m2
+%ifidn %1,sp
+    paddd           m4, m7
+    paddd           m5, m7
+    psrad           m4, 12
+    psrad           m5, 12
+%else
+    psrad           m4, 6
+    psrad           m5, 6
+%endif
+    packssdw        m4, m5
+
+    movu            xm2, [r0 + r1]                  ; m2 = row 9
+    punpckhwd       xm5, xm0, xm2
+    punpcklwd       xm0, xm2
+    vinserti128     m0, m0, xm5, 1
+    pmaddwd         m0, [r5 + 1 * mmsize]
+    paddd           m6, m0
+    movu            xm5, [r0 + r1 * 2]              ; m5 = row 10
+    punpckhwd       xm0, xm2, xm5
+    punpcklwd       xm2, xm5
+    vinserti128     m2, m2, xm0, 1
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m1, m2
+
+%ifidn %1,sp
+    paddd           m6, m7
+    paddd           m1, m7
+    psrad           m6, 12
+    psrad           m1, 12
+%else
+    psrad           m6, 6
+    psrad           m1, 6
+%endif
+    packssdw        m6, m1
+%ifidn %1,sp
+    packuswb        m4, m6
+    vextracti128    xm6, m4, 1
+    movd            [r2], xm4
+    pextrw          [r2 + 4], xm6, 0
+    pextrd          [r2 + r3], xm4, 1
+    pextrw          [r2 + r3 + 4], xm6, 2
+    pextrd          [r2 + r3 * 2], xm4, 2
+    pextrw          [r2 + r3 * 2 + 4], xm6, 4
+    pextrd          [r2 + r4], xm4, 3
+    pextrw          [r2 + r4 + 4], xm6, 6
+%else
+    movq            [r2], xm4
+    movhps          [r2 + r3], xm4
+    movq            [r2 + r3 * 2], xm6
+    movhps          [r2 + r4], xm6
+    vextracti128    xm5, m4, 1
+    vextracti128    xm1, m6, 1
+    movd            [r2 + 8], xm5
+    pextrd          [r2 + r3 + 8], xm5, 2
+    movd            [r2 + r3 * 2 + 8], xm1
+    pextrd          [r2 + r4 + 8], xm1, 2
+%endif
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_6x8 sp
+    FILTER_VER_CHROMA_S_AVX2_6x8 ss
+
+%macro FILTER_VER_CHROMA_S_AVX2_6x16 1
+%if ARCH_X86_64 == 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_6x16, 4, 7, 9
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,sp
+    mova            m8, [v4_pd_526336]
+%else
+    add             r3d, r3d
+%endif
+    lea             r6, [r3 * 3]
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m2, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    paddd           m1, m5
+    pmaddwd         m3, [r5]
+%ifidn %1,sp
+    paddd           m0, m8
+    paddd           m1, m8
+    psrad           m0, 12
+    psrad           m1, 12
+%else
+    psrad           m0, 6
+    psrad           m1, 6
+%endif
+    packssdw        m0, m1
+
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m2, m6
+    pmaddwd         m4, [r5]
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm1, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm1, 1
+    pmaddwd         m1, m5, [r5 + 1 * mmsize]
+    pmaddwd         m5, [r5]
+    paddd           m3, m1
+%ifidn %1,sp
+    paddd           m2, m8
+    paddd           m3, m8
+    psrad           m2, 12
+    psrad           m3, 12
+%else
+    psrad           m2, 6
+    psrad           m3, 6
+%endif
+    packssdw        m2, m3
+%ifidn %1,sp
+    packuswb        m0, m2
+    vextracti128    xm2, m0, 1
+    movd            [r2], xm0
+    pextrw          [r2 + 4], xm2, 0
+    pextrd          [r2 + r3], xm0, 1
+    pextrw          [r2 + r3 + 4], xm2, 2
+    pextrd          [r2 + r3 * 2], xm0, 2
+    pextrw          [r2 + r3 * 2 + 4], xm2, 4
+    pextrd          [r2 + r6], xm0, 3
+    pextrw          [r2 + r6 + 4], xm2, 6
+%else
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r6], xm2
+    vextracti128    xm0, m0, 1
+    vextracti128    xm3, m2, 1
+    movd            [r2 + 8], xm0
+    pextrd          [r2 + r3 + 8], xm0, 2
+    movd            [r2 + r3 * 2 + 8], xm3
+    pextrd          [r2 + r6 + 8], xm3, 2
+%endif
+    lea             r2, [r2 + r3 * 4]
+    movu            xm1, [r0 + r4]                  ; m1 = row 7
+    punpckhwd       xm0, xm6, xm1
+    punpcklwd       xm6, xm1
+    vinserti128     m6, m6, xm0, 1
+    pmaddwd         m0, m6, [r5 + 1 * mmsize]
+    pmaddwd         m6, [r5]
+    paddd           m4, m0
+    lea             r0, [r0 + r1 * 4]
+    movu            xm0, [r0]                       ; m0 = row 8
+    punpckhwd       xm2, xm1, xm0
+    punpcklwd       xm1, xm0
+    vinserti128     m1, m1, xm2, 1
+    pmaddwd         m2, m1, [r5 + 1 * mmsize]
+    pmaddwd         m1, [r5]
+    paddd           m5, m2
+%ifidn %1,sp
+    paddd           m4, m8
+    paddd           m5, m8
+    psrad           m4, 12
+    psrad           m5, 12
+%else
+    psrad           m4, 6
+    psrad           m5, 6
+%endif
+    packssdw        m4, m5
+
+    movu            xm2, [r0 + r1]                  ; m2 = row 9
+    punpckhwd       xm5, xm0, xm2
+    punpcklwd       xm0, xm2
+    vinserti128     m0, m0, xm5, 1
+    pmaddwd         m5, m0, [r5 + 1 * mmsize]
+    paddd           m6, m5
+    pmaddwd         m0, [r5]
+    movu            xm5, [r0 + r1 * 2]              ; m5 = row 10
+    punpckhwd       xm7, xm2, xm5
+    punpcklwd       xm2, xm5
+    vinserti128     m2, m2, xm7, 1
+    pmaddwd         m7, m2, [r5 + 1 * mmsize]
+    paddd           m1, m7
+    pmaddwd         m2, [r5]
+
+%ifidn %1,sp
+    paddd           m6, m8
+    paddd           m1, m8
+    psrad           m6, 12
+    psrad           m1, 12
+%else
+    psrad           m6, 6
+    psrad           m1, 6
+%endif
+    packssdw        m6, m1
+%ifidn %1,sp
+    packuswb        m4, m6
+    vextracti128    xm6, m4, 1
+    movd            [r2], xm4
+    pextrw          [r2 + 4], xm6, 0
+    pextrd          [r2 + r3], xm4, 1
+    pextrw          [r2 + r3 + 4], xm6, 2
+    pextrd          [r2 + r3 * 2], xm4, 2
+    pextrw          [r2 + r3 * 2 + 4], xm6, 4
+    pextrd          [r2 + r6], xm4, 3
+    pextrw          [r2 + r6 + 4], xm6, 6
+%else
+    movq            [r2], xm4
+    movhps          [r2 + r3], xm4
+    movq            [r2 + r3 * 2], xm6
+    movhps          [r2 + r6], xm6
+    vextracti128    xm4, m4, 1
+    vextracti128    xm1, m6, 1
+    movd            [r2 + 8], xm4
+    pextrd          [r2 + r3 + 8], xm4, 2
+    movd            [r2 + r3 * 2 + 8], xm1
+    pextrd          [r2 + r6 + 8], xm1, 2
+%endif
+    lea             r2, [r2 + r3 * 4]
+    movu            xm7, [r0 + r4]                  ; m7 = row 11
+    punpckhwd       xm1, xm5, xm7
+    punpcklwd       xm5, xm7
+    vinserti128     m5, m5, xm1, 1
+    pmaddwd         m1, m5, [r5 + 1 * mmsize]
+    paddd           m0, m1
+    pmaddwd         m5, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm1, [r0]                       ; m1 = row 12
+    punpckhwd       xm4, xm7, xm1
+    punpcklwd       xm7, xm1
+    vinserti128     m7, m7, xm4, 1
+    pmaddwd         m4, m7, [r5 + 1 * mmsize]
+    paddd           m2, m4
+    pmaddwd         m7, [r5]
+%ifidn %1,sp
+    paddd           m0, m8
+    paddd           m2, m8
+    psrad           m0, 12
+    psrad           m2, 12
+%else
+    psrad           m0, 6
+    psrad           m2, 6
+%endif
+    packssdw        m0, m2
+
+    movu            xm4, [r0 + r1]                  ; m4 = row 13
+    punpckhwd       xm2, xm1, xm4
+    punpcklwd       xm1, xm4
+    vinserti128     m1, m1, xm2, 1
+    pmaddwd         m2, m1, [r5 + 1 * mmsize]
+    paddd           m5, m2
+    pmaddwd         m1, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 14
+    punpckhwd       xm6, xm4, xm2
+    punpcklwd       xm4, xm2
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m7, m6
+    pmaddwd         m4, [r5]
+%ifidn %1,sp
+    paddd           m5, m8
+    paddd           m7, m8
+    psrad           m5, 12
+    psrad           m7, 12
+%else
+    psrad           m5, 6
+    psrad           m7, 6
+%endif
+    packssdw        m5, m7
+%ifidn %1,sp
+    packuswb        m0, m5
+    vextracti128    xm5, m0, 1
+    movd            [r2], xm0
+    pextrw          [r2 + 4], xm5, 0
+    pextrd          [r2 + r3], xm0, 1
+    pextrw          [r2 + r3 + 4], xm5, 2
+    pextrd          [r2 + r3 * 2], xm0, 2
+    pextrw          [r2 + r3 * 2 + 4], xm5, 4
+    pextrd          [r2 + r6], xm0, 3
+    pextrw          [r2 + r6 + 4], xm5, 6
+%else
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm5
+    movhps          [r2 + r6], xm5
+    vextracti128    xm0, m0, 1
+    vextracti128    xm7, m5, 1
+    movd            [r2 + 8], xm0
+    pextrd          [r2 + r3 + 8], xm0, 2
+    movd            [r2 + r3 * 2 + 8], xm7
+    pextrd          [r2 + r6 + 8], xm7, 2
+%endif
+    lea             r2, [r2 + r3 * 4]
+
+    movu            xm6, [r0 + r4]                  ; m6 = row 15
+    punpckhwd       xm5, xm2, xm6
+    punpcklwd       xm2, xm6
+    vinserti128     m2, m2, xm5, 1
+    pmaddwd         m5, m2, [r5 + 1 * mmsize]
+    paddd           m1, m5
+    pmaddwd         m2, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm0, [r0]                       ; m0 = row 16
+    punpckhwd       xm5, xm6, xm0
+    punpcklwd       xm6, xm0
+    vinserti128     m6, m6, xm5, 1
+    pmaddwd         m5, m6, [r5 + 1 * mmsize]
+    paddd           m4, m5
+    pmaddwd         m6, [r5]
+%ifidn %1,sp
+    paddd           m1, m8
+    paddd           m4, m8
+    psrad           m1, 12
+    psrad           m4, 12
+%else
+    psrad           m1, 6
+    psrad           m4, 6
+%endif
+    packssdw        m1, m4
+
+    movu            xm5, [r0 + r1]                  ; m5 = row 17
+    punpckhwd       xm4, xm0, xm5
+    punpcklwd       xm0, xm5
+    vinserti128     m0, m0, xm4, 1
+    pmaddwd         m0, [r5 + 1 * mmsize]
+    paddd           m2, m0
+    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
+    punpckhwd       xm0, xm5, xm4
+    punpcklwd       xm5, xm4
+    vinserti128     m5, m5, xm0, 1
+    pmaddwd         m5, [r5 + 1 * mmsize]
+    paddd           m6, m5
+%ifidn %1,sp
+    paddd           m2, m8
+    paddd           m6, m8
+    psrad           m2, 12
+    psrad           m6, 12
+%else
+    psrad           m2, 6
+    psrad           m6, 6
+%endif
+    packssdw        m2, m6
+%ifidn %1,sp
+    packuswb        m1, m2
+    vextracti128    xm2, m1, 1
+    movd            [r2], xm1
+    pextrw          [r2 + 4], xm2, 0
+    pextrd          [r2 + r3], xm1, 1
+    pextrw          [r2 + r3 + 4], xm2, 2
+    pextrd          [r2 + r3 * 2], xm1, 2
+    pextrw          [r2 + r3 * 2 + 4], xm2, 4
+    pextrd          [r2 + r6], xm1, 3
+    pextrw          [r2 + r6 + 4], xm2, 6
+%else
+    movq            [r2], xm1
+    movhps          [r2 + r3], xm1
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r6], xm2
+    vextracti128    xm4, m1, 1
+    vextracti128    xm6, m2, 1
+    movd            [r2 + 8], xm4
+    pextrd          [r2 + r3 + 8], xm4, 2
+    movd            [r2 + r3 * 2 + 8], xm6
+    pextrd          [r2 + r6 + 8], xm6, 2
+%endif
+    RET
+%endif
+%endmacro
+
+    FILTER_VER_CHROMA_S_AVX2_6x16 sp
+    FILTER_VER_CHROMA_S_AVX2_6x16 ss
+
+;---------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vertical_ss_%1x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;---------------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_SS_W2_4R 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_ss_%1x%2, 5, 6, 5
+
+    add       r1d, r1d
+    add       r3d, r3d
+    sub       r0, r1
+    shl       r4d, 5
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r5, [r5 + r4]
+%else
+    lea       r5, [tab_ChromaCoeffV + r4]
+%endif
+
+    mov       r4d, (%2/4)
+
+.loopH:
+    PROCESS_CHROMA_SP_W2_4R r5
+
+    psrad     m0, 6
+    psrad     m2, 6
+
+    packssdw  m0, m2
+
+    movd      [r2], m0
+    pextrd    [r2 + r3], m0, 1
+    lea       r2, [r2 + 2 * r3]
+    pextrd    [r2], m0, 2
+    pextrd    [r2 + r3], m0, 3
+
+    lea       r2, [r2 + 2 * r3]
+
+    dec       r4d
+    jnz       .loopH
+
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_SS_W2_4R 2, 4
+    FILTER_VER_CHROMA_SS_W2_4R 2, 8
+
+    FILTER_VER_CHROMA_SS_W2_4R 2, 16
+
+;---------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ss_4x2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;---------------------------------------------------------------------------------------------------------------
+INIT_XMM sse2
+cglobal interp_4tap_vert_ss_4x2, 5, 6, 4
+
+    add        r1d, r1d
+    add        r3d, r3d
+    sub        r0, r1
+    shl        r4d, 5
+
+%ifdef PIC
+    lea        r5, [tab_ChromaCoeffV]
+    lea        r5, [r5 + r4]
+%else
+    lea        r5, [tab_ChromaCoeffV + r4]
+%endif
+
+    movq       m0, [r0]
+    movq       m1, [r0 + r1]
+    punpcklwd  m0, m1                          ;m0=[0 1]
+    pmaddwd    m0, [r5 + 0 *16]                ;m0=[0+1]  Row1
+
+    lea        r0, [r0 + 2 * r1]
+    movq       m2, [r0]
+    punpcklwd  m1, m2                          ;m1=[1 2]
+    pmaddwd    m1, [r5 + 0 *16]                ;m1=[1+2]  Row2
+
+    movq       m3, [r0 + r1]
+    punpcklwd  m2, m3                          ;m4=[2 3]
+    pmaddwd    m2, [r5 + 1 * 16]
+    paddd      m0, m2                          ;m0=[0+1+2+3]  Row1 done
+    psrad      m0, 6
+
+    movq       m2, [r0 + 2 * r1]
+    punpcklwd  m3, m2                          ;m5=[3 4]
+    pmaddwd    m3, [r5 + 1 * 16]
+    paddd      m1, m3                          ;m1=[1+2+3+4]  Row2 done
+    psrad      m1, 6
+
+    packssdw   m0, m1
+
+    movlps     [r2], m0
+    movhps     [r2 + r3], m0
+
+    RET
+
+;-------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vertical_ss_6x8(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_SS_W6_H4 2
+INIT_XMM sse4
+cglobal interp_4tap_vert_ss_6x%2, 5, 7, 6
+
+    add       r1d, r1d
+    add       r3d, r3d
+    sub       r0, r1
+    shl       r4d, 5
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r6, [r5 + r4]
+%else
+    lea       r6, [tab_ChromaCoeffV + r4]
+%endif
+
+    mov       r4d, %2/4
+
+.loopH:
+    PROCESS_CHROMA_SP_W4_4R
+
+    psrad     m0, 6
+    psrad     m1, 6
+    psrad     m2, 6
+    psrad     m3, 6
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+
+    movlps    [r2], m0
+    movhps    [r2 + r3], m0
+    lea       r5, [r2 + 2 * r3]
+    movlps    [r5], m2
+    movhps    [r5 + r3], m2
+
+    lea       r5, [4 * r1 - 2 * 4]
+    sub       r0, r5
+    add       r2, 2 * 4
+
+    PROCESS_CHROMA_SP_W2_4R r6
+
+    psrad     m0, 6
+    psrad     m2, 6
+
+    packssdw  m0, m2
+
+    movd      [r2], m0
+    pextrd    [r2 + r3], m0, 1
+    lea       r2, [r2 + 2 * r3]
+    pextrd    [r2], m0, 2
+    pextrd    [r2 + r3], m0, 3
+
+    sub       r0, 2 * 4
+    lea       r2, [r2 + 2 * r3 - 2 * 4]
+
+    dec       r4d
+    jnz       .loopH
+
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_SS_W6_H4 6, 8
+
+    FILTER_VER_CHROMA_SS_W6_H4 6, 16
+
+
+;----------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_ss_8x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;----------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_SS_W8_H2 2
+INIT_XMM sse2
+cglobal interp_4tap_vert_ss_%1x%2, 5, 6, 7
+
+    add       r1d, r1d
+    add       r3d, r3d
+    sub       r0, r1
+    shl       r4d, 5
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r5, [r5 + r4]
+%else
+    lea       r5, [tab_ChromaCoeffV + r4]
+%endif
+
+    mov       r4d, %2/2
+.loopH:
+    PROCESS_CHROMA_SP_W8_2R
+
+    psrad     m0, 6
+    psrad     m1, 6
+    psrad     m2, 6
+    psrad     m3, 6
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+
+    movu      [r2], m0
+    movu      [r2 + r3], m2
+
+    lea       r2, [r2 + 2 * r3]
+
+    dec       r4d
+    jnz       .loopH
+
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_SS_W8_H2 8, 2
+    FILTER_VER_CHROMA_SS_W8_H2 8, 4
+    FILTER_VER_CHROMA_SS_W8_H2 8, 6
+    FILTER_VER_CHROMA_SS_W8_H2 8, 8
+    FILTER_VER_CHROMA_SS_W8_H2 8, 16
+    FILTER_VER_CHROMA_SS_W8_H2 8, 32
+
+    FILTER_VER_CHROMA_SS_W8_H2 8, 12
+    FILTER_VER_CHROMA_SS_W8_H2 8, 64
+

From 1a6e8f9c054ee34f4cd5e98db7fa9055a710fd66 Mon Sep 17 00:00:00 2001
From: Mythreyi P <mythreyi@multicorewareinc.com>
Date: Thu, 15 Feb 2018 00:56:16 -0800
Subject: [PATCH 48/51] x86: Split ipfilter16 kernals into two different source
 files.

Port horizonal kernals from ipfilter16.asm to a new source file, h-ipfilter16.asm to improve build time.
---
 source/common/CMakeLists.txt       |    2 +-
 source/common/x86/h-ipfilter16.asm | 5000 +++++++++++++++++++++++++++
 source/common/x86/ipfilter16.asm   | 5104 +---------------------------
 3 files changed, 5090 insertions(+), 5016 deletions(-)
 create mode 100644 source/common/x86/h-ipfilter16.asm

diff --git a/source/common/CMakeLists.txt b/source/common/CMakeLists.txt
index 8261fac0c3..079b64982c 100644
--- a/source/common/CMakeLists.txt
+++ b/source/common/CMakeLists.txt
@@ -61,7 +61,7 @@ if(ENABLE_ASSEMBLY AND X86)
                mc-a2.asm pixel-util8.asm blockcopy8.asm
                pixeladd8.asm dct8.asm seaintegral.asm)
     if(HIGH_BIT_DEPTH)
-        set(A_SRCS ${A_SRCS} sad16-a.asm intrapred16.asm ipfilter16.asm loopfilter.asm)
+        set(A_SRCS ${A_SRCS} sad16-a.asm intrapred16.asm h-ipfilter16.asm ipfilter16.asm loopfilter.asm)
     else()
         set(A_SRCS ${A_SRCS} sad-a.asm intrapred8.asm intrapred8_allangs.asm v4-ipfilter8.asm h-ipfilter8.asm ipfilter8.asm loopfilter.asm)
     endif()
diff --git a/source/common/x86/h-ipfilter16.asm b/source/common/x86/h-ipfilter16.asm
new file mode 100644
index 0000000000..92c329b036
--- /dev/null
+++ b/source/common/x86/h-ipfilter16.asm
@@ -0,0 +1,5000 @@
+;*****************************************************************************
+;* Copyright (C) 2013-2017 MulticoreWare, Inc
+;*
+;* Authors: Nabajit Deka <nabajit@multicorewareinc.com>
+;*          Murugan Vairavel <murugan@multicorewareinc.com>
+;*          Min Chen <chenm003@163.com>
+;*
+;* This program is free software; you can redistribute it and/or modify
+;* it under the terms of the GNU General Public License as published by
+;* the Free Software Foundation; either version 2 of the License, or
+;* (at your option) any later version.
+;*
+;* This program is distributed in the hope that it will be useful,
+;* but WITHOUT ANY WARRANTY; without even the implied warranty of
+;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+;* GNU General Public License for more details.
+;*
+;* You should have received a copy of the GNU General Public License
+;* along with this program; if not, write to the Free Software
+;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
+;*
+;* This program is also available under a commercial proprietary license.
+;* For more information, contact us at license @ x265.com.
+;*****************************************************************************/
+%include "x86inc.asm"
+%include "x86util.asm"
+
+
+%define INTERP_OFFSET_PP        pd_32
+%define INTERP_SHIFT_PP         6
+
+%if BIT_DEPTH == 10
+    %define INTERP_SHIFT_PS         2
+    %define INTERP_OFFSET_PS        pd_n32768
+    %define INTERP_SHIFT_SP         10
+    %define INTERP_OFFSET_SP        h_pd_524800
+%elif BIT_DEPTH == 12
+    %define INTERP_SHIFT_PS         4
+    %define INTERP_OFFSET_PS        pd_n131072
+    %define INTERP_SHIFT_SP         8
+    %define INTERP_OFFSET_SP        pd_524416
+%else
+    %error Unsupport bit depth!
+%endif
+
+SECTION_RODATA 32
+
+tab_c_32:         times 8 dd 32
+h_pd_524800:        times 8 dd 524800
+
+tab_Tm16:         db 0, 1, 2, 3, 4,  5,  6, 7, 2, 3, 4,  5, 6, 7, 8, 9
+
+
+tab_ChromaCoeff:  dw  0, 64,  0,  0
+                  dw -2, 58, 10, -2
+                  dw -4, 54, 16, -2
+                  dw -6, 46, 28, -4
+                  dw -4, 36, 36, -4
+                  dw -4, 28, 46, -6
+                  dw -2, 16, 54, -4
+                  dw -2, 10, 58, -2
+                                    
+tab_LumaCoeff:    dw   0, 0,  0,  64,  0,   0,  0,  0
+                  dw  -1, 4, -10, 58,  17, -5,  1,  0
+                  dw  -1, 4, -11, 40,  40, -11, 4, -1
+                  dw   0, 1, -5,  17,  58, -10, 4, -1
+                  
+ALIGN 32
+h_tab_LumaCoeffV:   times 4 dw 0, 0
+                  times 4 dw 0, 64
+                  times 4 dw 0, 0
+                  times 4 dw 0, 0
+
+                  times 4 dw -1, 4
+                  times 4 dw -10, 58
+                  times 4 dw 17, -5
+                  times 4 dw 1, 0
+
+                  times 4 dw -1, 4
+                  times 4 dw -11, 40
+                  times 4 dw 40, -11
+                  times 4 dw 4, -1
+
+                  times 4 dw 0, 1
+                  times 4 dw -5, 17
+                  times 4 dw 58, -10
+                  times 4 dw 4, -1
+                  
+const interp8_hps_shuf,     dd 0, 4, 1, 5, 2, 6, 3, 7                 
+
+const interp8_hpp_shuf,     db 0, 1, 2, 3, 4, 5, 6, 7, 2, 3, 4, 5, 6, 7, 8, 9
+                            db 4, 5, 6, 7, 8, 9, 10, 11, 6, 7, 8, 9, 10, 11, 12, 13 
+
+const interp8_hpp_shuf_new, db 0, 1, 2, 3, 2, 3, 4, 5, 4, 5, 6, 7, 6, 7, 8, 9
+                            db 4, 5, 6, 7, 6, 7, 8, 9, 8, 9, 10, 11, 10, 11, 12, 13                         
+                            
+SECTION .text
+cextern pd_8
+cextern pd_32
+cextern pw_pixel_max
+cextern pd_524416
+cextern pd_n32768
+cextern pd_n131072
+cextern pw_2000
+cextern idct8_shuf2
+
+%macro FILTER_LUMA_HOR_4_sse2 1
+    movu        m4,     [r0 + %1]       ; m4 = src[0-7]
+    movu        m5,     [r0 + %1 + 2]   ; m5 = src[1-8]
+    pmaddwd     m4,     m0
+    pmaddwd     m5,     m0
+    pshufd      m2,     m4,     q2301
+    paddd       m4,     m2
+    pshufd      m2,     m5,     q2301
+    paddd       m5,     m2
+    pshufd      m4,     m4,     q3120
+    pshufd      m5,     m5,     q3120
+    punpcklqdq  m4,     m5
+
+    movu        m5,     [r0 + %1 + 4]   ; m5 = src[2-9]
+    movu        m3,     [r0 + %1 + 6]   ; m3 = src[3-10]
+    pmaddwd     m5,     m0
+    pmaddwd     m3,     m0
+    pshufd      m2,     m5,     q2301
+    paddd       m5,     m2
+    pshufd      m2,     m3,     q2301
+    paddd       m3,     m2
+    pshufd      m5,     m5,     q3120
+    pshufd      m3,     m3,     q3120
+    punpcklqdq  m5,     m3
+
+    pshufd      m2,     m4,     q2301
+    paddd       m4,     m2
+    pshufd      m2,     m5,     q2301
+    paddd       m5,     m2
+    pshufd      m4,     m4,     q3120
+    pshufd      m5,     m5,     q3120
+    punpcklqdq  m4,     m5
+    paddd       m4,     m1
+%endmacro
+
+%macro FILTER_LUMA_HOR_8_sse2 1
+    movu        m4,     [r0 + %1]       ; m4 = src[0-7]
+    movu        m5,     [r0 + %1 + 2]   ; m5 = src[1-8]
+    pmaddwd     m4,     m0
+    pmaddwd     m5,     m0
+    pshufd      m2,     m4,     q2301
+    paddd       m4,     m2
+    pshufd      m2,     m5,     q2301
+    paddd       m5,     m2
+    pshufd      m4,     m4,     q3120
+    pshufd      m5,     m5,     q3120
+    punpcklqdq  m4,     m5
+
+    movu        m5,     [r0 + %1 + 4]   ; m5 = src[2-9]
+    movu        m3,     [r0 + %1 + 6]   ; m3 = src[3-10]
+    pmaddwd     m5,     m0
+    pmaddwd     m3,     m0
+    pshufd      m2,     m5,     q2301
+    paddd       m5,     m2
+    pshufd      m2,     m3,     q2301
+    paddd       m3,     m2
+    pshufd      m5,     m5,     q3120
+    pshufd      m3,     m3,     q3120
+    punpcklqdq  m5,     m3
+
+    pshufd      m2,     m4,     q2301
+    paddd       m4,     m2
+    pshufd      m2,     m5,     q2301
+    paddd       m5,     m2
+    pshufd      m4,     m4,     q3120
+    pshufd      m5,     m5,     q3120
+    punpcklqdq  m4,     m5
+    paddd       m4,     m1
+
+    movu        m5,     [r0 + %1 + 8]   ; m5 = src[4-11]
+    movu        m6,     [r0 + %1 + 10]  ; m6 = src[5-12]
+    pmaddwd     m5,     m0
+    pmaddwd     m6,     m0
+    pshufd      m2,     m5,     q2301
+    paddd       m5,     m2
+    pshufd      m2,     m6,     q2301
+    paddd       m6,     m2
+    pshufd      m5,     m5,     q3120
+    pshufd      m6,     m6,     q3120
+    punpcklqdq  m5,     m6
+
+    movu        m6,     [r0 + %1 + 12]  ; m6 = src[6-13]
+    movu        m3,     [r0 + %1 + 14]  ; m3 = src[7-14]
+    pmaddwd     m6,     m0
+    pmaddwd     m3,     m0
+    pshufd      m2,     m6,     q2301
+    paddd       m6,     m2
+    pshufd      m2,     m3,     q2301
+    paddd       m3,     m2
+    pshufd      m6,     m6,     q3120
+    pshufd      m3,     m3,     q3120
+    punpcklqdq  m6,     m3
+
+    pshufd      m2,     m5,     q2301
+    paddd       m5,     m2
+    pshufd      m2,     m6,     q2301
+    paddd       m6,     m2
+    pshufd      m5,     m5,     q3120
+    pshufd      m6,     m6,     q3120
+    punpcklqdq  m5,     m6
+    paddd       m5,     m1
+%endmacro
+
+;------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_p%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;------------------------------------------------------------------------------------------------------------
+%macro FILTER_HOR_LUMA_sse2 3
+INIT_XMM sse2
+cglobal interp_8tap_horiz_%3_%1x%2, 4, 7, 8
+    mov         r4d,    r4m
+    sub         r0,     6
+    shl         r4d,    4
+    add         r1d,    r1d
+    add         r3d,    r3d
+
+%ifdef PIC
+    lea         r6,     [tab_LumaCoeff]
+    mova        m0,     [r6 + r4]
+%else
+    mova        m0,     [tab_LumaCoeff + r4]
+%endif
+
+%ifidn %3, pp
+    mova        m1,     [pd_32]
+    pxor        m7,     m7
+%else
+    mova        m1,     [INTERP_OFFSET_PS]
+%endif
+
+    mov         r4d,    %2
+%ifidn %3, ps
+    cmp         r5m,    byte 0
+    je          .loopH
+    lea         r6,     [r1 + 2 * r1]
+    sub         r0,     r6
+    add         r4d,    7
+%endif
+
+.loopH:
+%assign x 0
+%rep %1/8
+    FILTER_LUMA_HOR_8_sse2 x
+
+%ifidn %3, pp
+    psrad       m4,     6
+    psrad       m5,     6
+    packssdw    m4,     m5
+    CLIPW       m4,     m7,     [pw_pixel_max]
+%else
+  %if BIT_DEPTH == 10
+    psrad       m4,     2
+    psrad       m5,     2
+  %elif BIT_DEPTH == 12
+    psrad       m4,     4
+    psrad       m5,     4
+  %endif
+    packssdw    m4,     m5
+%endif
+
+    movu        [r2 + x], m4
+%assign x x+16
+%endrep
+
+%rep (%1 % 8)/4
+    FILTER_LUMA_HOR_4_sse2 x
+
+%ifidn %3, pp
+    psrad       m4,     6
+    packssdw    m4,     m4
+    CLIPW       m4,     m7,     [pw_pixel_max]
+%else
+  %if BIT_DEPTH == 10
+    psrad       m4,     2
+  %elif BIT_DEPTH == 12
+    psrad       m4,     4
+  %endif
+    packssdw    m4,     m4
+%endif
+
+    movh        [r2 + x], m4
+%endrep
+
+    add         r0,     r1
+    add         r2,     r3
+
+    dec         r4d
+    jnz         .loopH
+    RET
+
+%endmacro
+
+;------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
+;------------------------------------------------------------------------------------------------------------
+    FILTER_HOR_LUMA_sse2 4, 4, pp
+    FILTER_HOR_LUMA_sse2 4, 8, pp
+    FILTER_HOR_LUMA_sse2 4, 16, pp
+    FILTER_HOR_LUMA_sse2 8, 4, pp
+    FILTER_HOR_LUMA_sse2 8, 8, pp
+    FILTER_HOR_LUMA_sse2 8, 16, pp
+    FILTER_HOR_LUMA_sse2 8, 32, pp
+    FILTER_HOR_LUMA_sse2 12, 16, pp
+    FILTER_HOR_LUMA_sse2 16, 4, pp
+    FILTER_HOR_LUMA_sse2 16, 8, pp
+    FILTER_HOR_LUMA_sse2 16, 12, pp
+    FILTER_HOR_LUMA_sse2 16, 16, pp
+    FILTER_HOR_LUMA_sse2 16, 32, pp
+    FILTER_HOR_LUMA_sse2 16, 64, pp
+    FILTER_HOR_LUMA_sse2 24, 32, pp
+    FILTER_HOR_LUMA_sse2 32, 8, pp
+    FILTER_HOR_LUMA_sse2 32, 16, pp
+    FILTER_HOR_LUMA_sse2 32, 24, pp
+    FILTER_HOR_LUMA_sse2 32, 32, pp
+    FILTER_HOR_LUMA_sse2 32, 64, pp
+    FILTER_HOR_LUMA_sse2 48, 64, pp
+    FILTER_HOR_LUMA_sse2 64, 16, pp
+    FILTER_HOR_LUMA_sse2 64, 32, pp
+    FILTER_HOR_LUMA_sse2 64, 48, pp
+    FILTER_HOR_LUMA_sse2 64, 64, pp
+
+;---------------------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_ps_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;---------------------------------------------------------------------------------------------------------------------------
+    FILTER_HOR_LUMA_sse2 4, 4, ps
+    FILTER_HOR_LUMA_sse2 4, 8, ps
+    FILTER_HOR_LUMA_sse2 4, 16, ps
+    FILTER_HOR_LUMA_sse2 8, 4, ps
+    FILTER_HOR_LUMA_sse2 8, 8, ps
+    FILTER_HOR_LUMA_sse2 8, 16, ps
+    FILTER_HOR_LUMA_sse2 8, 32, ps
+    FILTER_HOR_LUMA_sse2 12, 16, ps
+    FILTER_HOR_LUMA_sse2 16, 4, ps
+    FILTER_HOR_LUMA_sse2 16, 8, ps
+    FILTER_HOR_LUMA_sse2 16, 12, ps
+    FILTER_HOR_LUMA_sse2 16, 16, ps
+    FILTER_HOR_LUMA_sse2 16, 32, ps
+    FILTER_HOR_LUMA_sse2 16, 64, ps
+    FILTER_HOR_LUMA_sse2 24, 32, ps
+    FILTER_HOR_LUMA_sse2 32, 8, ps
+    FILTER_HOR_LUMA_sse2 32, 16, ps
+    FILTER_HOR_LUMA_sse2 32, 24, ps
+    FILTER_HOR_LUMA_sse2 32, 32, ps
+    FILTER_HOR_LUMA_sse2 32, 64, ps
+    FILTER_HOR_LUMA_sse2 48, 64, ps
+    FILTER_HOR_LUMA_sse2 64, 16, ps
+    FILTER_HOR_LUMA_sse2 64, 32, ps
+    FILTER_HOR_LUMA_sse2 64, 48, ps
+    FILTER_HOR_LUMA_sse2 64, 64, ps
+    
+%macro FILTERH_W2_4_sse3 2
+    movh        m3,     [r0 + %1]
+    movhps      m3,     [r0 + %1 + 2]
+    pmaddwd     m3,     m0
+    movh        m4,     [r0 + r1 + %1]
+    movhps      m4,     [r0 + r1 + %1 + 2]
+    pmaddwd     m4,     m0
+    pshufd      m2,     m3,     q2301
+    paddd       m3,     m2
+    pshufd      m2,     m4,     q2301
+    paddd       m4,     m2
+    pshufd      m3,     m3,     q3120
+    pshufd      m4,     m4,     q3120
+    punpcklqdq  m3,     m4
+    paddd       m3,     m1
+    movh        m5,     [r0 + 2 * r1 + %1]
+    movhps      m5,     [r0 + 2 * r1 + %1 + 2]
+    pmaddwd     m5,     m0
+    movh        m4,     [r0 + r4 + %1]
+    movhps      m4,     [r0 + r4 + %1 + 2]
+    pmaddwd     m4,     m0
+    pshufd      m2,     m5,     q2301
+    paddd       m5,     m2
+    pshufd      m2,     m4,     q2301
+    paddd       m4,     m2
+    pshufd      m5,     m5,     q3120
+    pshufd      m4,     m4,     q3120
+    punpcklqdq  m5,     m4
+    paddd       m5,     m1
+%ifidn %2, pp
+    psrad       m3,     6
+    psrad       m5,     6
+    packssdw    m3,     m5
+    CLIPW       m3,     m7,     m6
+%else
+    psrad       m3,     INTERP_SHIFT_PS
+    psrad       m5,     INTERP_SHIFT_PS
+    packssdw    m3,     m5
+%endif
+    movd        [r2 + %1], m3
+    psrldq      m3,     4
+    movd        [r2 + r3 + %1], m3
+    psrldq      m3,     4
+    movd        [r2 + r3 * 2 + %1], m3
+    psrldq      m3,     4
+    movd        [r2 + r5 + %1], m3
+%endmacro
+
+%macro FILTERH_W2_3_sse3 1
+    movh        m3,     [r0 + %1]
+    movhps      m3,     [r0 + %1 + 2]
+    pmaddwd     m3,     m0
+    movh        m4,     [r0 + r1 + %1]
+    movhps      m4,     [r0 + r1 + %1 + 2]
+    pmaddwd     m4,     m0
+    pshufd      m2,     m3,     q2301
+    paddd       m3,     m2
+    pshufd      m2,     m4,     q2301
+    paddd       m4,     m2
+    pshufd      m3,     m3,     q3120
+    pshufd      m4,     m4,     q3120
+    punpcklqdq  m3,     m4
+    paddd       m3,     m1
+
+    movh        m5,     [r0 + 2 * r1 + %1]
+    movhps      m5,     [r0 + 2 * r1 + %1 + 2]
+    pmaddwd     m5,     m0
+
+    pshufd      m2,     m5,     q2301
+    paddd       m5,     m2
+    pshufd      m5,     m5,     q3120
+    paddd       m5,     m1
+
+    psrad       m3,     INTERP_SHIFT_PS
+    psrad       m5,     INTERP_SHIFT_PS
+    packssdw    m3,     m5
+
+    movd        [r2 + %1], m3
+    psrldq      m3,     4
+    movd        [r2 + r3 + %1], m3
+    psrldq      m3,     4
+    movd        [r2 + r3 * 2 + %1], m3
+%endmacro
+
+%macro FILTERH_W4_2_sse3 2
+    movh        m3,     [r0 + %1]
+    movhps      m3,     [r0 + %1 + 2]
+    pmaddwd     m3,     m0
+    movh        m4,     [r0 + %1 + 4]
+    movhps      m4,     [r0 + %1 + 6]
+    pmaddwd     m4,     m0
+    pshufd      m2,     m3,     q2301
+    paddd       m3,     m2
+    pshufd      m2,     m4,     q2301
+    paddd       m4,     m2
+    pshufd      m3,     m3,     q3120
+    pshufd      m4,     m4,     q3120
+    punpcklqdq  m3,     m4
+    paddd       m3,     m1
+
+    movh        m5,     [r0 + r1 + %1]
+    movhps      m5,     [r0 + r1 + %1 + 2]
+    pmaddwd     m5,     m0
+    movh        m4,     [r0 + r1 + %1 + 4]
+    movhps      m4,     [r0 + r1 + %1 + 6]
+    pmaddwd     m4,     m0
+    pshufd      m2,     m5,     q2301
+    paddd       m5,     m2
+    pshufd      m2,     m4,     q2301
+    paddd       m4,     m2
+    pshufd      m5,     m5,     q3120
+    pshufd      m4,     m4,     q3120
+    punpcklqdq  m5,     m4
+    paddd       m5,     m1
+%ifidn %2, pp
+    psrad       m3,     6
+    psrad       m5,     6
+    packssdw    m3,     m5
+    CLIPW       m3,     m7,     m6
+%else
+    psrad       m3,     INTERP_SHIFT_PS
+    psrad       m5,     INTERP_SHIFT_PS
+    packssdw    m3,     m5
+%endif
+    movh        [r2 + %1], m3
+    movhps      [r2 + r3 + %1], m3
+%endmacro
+
+%macro FILTERH_W4_1_sse3 1
+    movh        m3,     [r0 + 2 * r1 + %1]
+    movhps      m3,     [r0 + 2 * r1 + %1 + 2]
+    pmaddwd     m3,     m0
+    movh        m4,     [r0 + 2 * r1 + %1 + 4]
+    movhps      m4,     [r0 + 2 * r1 + %1 + 6]
+    pmaddwd     m4,     m0
+    pshufd      m2,     m3,     q2301
+    paddd       m3,     m2
+    pshufd      m2,     m4,     q2301
+    paddd       m4,     m2
+    pshufd      m3,     m3,     q3120
+    pshufd      m4,     m4,     q3120
+    punpcklqdq  m3,     m4
+    paddd       m3,     m1
+
+    psrad       m3,     INTERP_SHIFT_PS
+    packssdw    m3,     m3
+    movh        [r2 + r3 * 2 + %1], m3
+%endmacro
+
+%macro FILTERH_W8_1_sse3 2
+    movh        m3,     [r0 + %1]
+    movhps      m3,     [r0 + %1 + 2]
+    pmaddwd     m3,     m0
+    movh        m4,     [r0 + %1 + 4]
+    movhps      m4,     [r0 + %1 + 6]
+    pmaddwd     m4,     m0
+    pshufd      m2,     m3,     q2301
+    paddd       m3,     m2
+    pshufd      m2,     m4,     q2301
+    paddd       m4,     m2
+    pshufd      m3,     m3,     q3120
+    pshufd      m4,     m4,     q3120
+    punpcklqdq  m3,     m4
+    paddd       m3,     m1
+
+    movh        m5,     [r0 + %1 + 8]
+    movhps      m5,     [r0 + %1 + 10]
+    pmaddwd     m5,     m0
+    movh        m4,     [r0 + %1 + 12]
+    movhps      m4,     [r0 + %1 + 14]
+    pmaddwd     m4,     m0
+    pshufd      m2,     m5,     q2301
+    paddd       m5,     m2
+    pshufd      m2,     m4,     q2301
+    paddd       m4,     m2
+    pshufd      m5,     m5,     q3120
+    pshufd      m4,     m4,     q3120
+    punpcklqdq  m5,     m4
+    paddd       m5,     m1
+%ifidn %2, pp
+    psrad       m3,     6
+    psrad       m5,     6
+    packssdw    m3,     m5
+    CLIPW       m3,     m7,     m6
+%else
+    psrad       m3,     INTERP_SHIFT_PS
+    psrad       m5,     INTERP_SHIFT_PS
+    packssdw    m3,     m5
+%endif
+    movdqu      [r2 + %1], m3
+%endmacro
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_horiz_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_HOR_CHROMA_sse3 3
+INIT_XMM sse3
+cglobal interp_4tap_horiz_%3_%1x%2, 4, 7, 8
+    add         r3,     r3
+    add         r1,     r1
+    sub         r0,     2
+    mov         r4d,    r4m
+    add         r4d,    r4d
+
+%ifdef PIC
+    lea         r6,     [tab_ChromaCoeff]
+    movddup     m0,     [r6 + r4 * 4]
+%else
+    movddup     m0,     [tab_ChromaCoeff + r4 * 4]
+%endif
+
+%ifidn %3, ps
+    mova        m1,     [INTERP_OFFSET_PS]
+    cmp         r5m,    byte 0
+%if %1 <= 6
+    lea         r4,     [r1 * 3]
+    lea         r5,     [r3 * 3]
+%endif
+    je          .skip
+    sub         r0,     r1
+%if %1 <= 6
+%assign y 1
+%else
+%assign y 3
+%endif
+%assign z 0
+%rep y
+%assign x 0
+%rep %1/8
+    FILTERH_W8_1_sse3 x, %3
+%assign x x+16
+%endrep
+%if %1 == 4 || (%1 == 6 && z == 0) || (%1 == 12 && z == 0)
+    FILTERH_W4_2_sse3 x, %3
+    FILTERH_W4_1_sse3 x
+%assign x x+8
+%endif
+%if %1 == 2 || (%1 == 6 && z == 0)
+    FILTERH_W2_3_sse3 x
+%endif
+%if %1 <= 6
+    lea         r0,     [r0 + r4]
+    lea         r2,     [r2 + r5]
+%else
+    lea         r0,     [r0 + r1]
+    lea         r2,     [r2 + r3]
+%endif
+%assign z z+1
+%endrep
+.skip:
+%elifidn %3, pp
+    pxor        m7,     m7
+    mova        m6,     [pw_pixel_max]
+    mova        m1,     [tab_c_32]
+%if %1 == 2 || %1 == 6
+    lea         r4,     [r1 * 3]
+    lea         r5,     [r3 * 3]
+%endif
+%endif
+
+%if %1 == 2
+%assign y %2/4
+%elif %1 <= 6
+%assign y %2/2
+%else
+%assign y %2
+%endif
+%assign z 0
+%rep y
+%assign x 0
+%rep %1/8
+    FILTERH_W8_1_sse3 x, %3
+%assign x x+16
+%endrep
+%if %1 == 4 || %1 == 6 || (%1 == 12 && (z % 2) == 0)
+    FILTERH_W4_2_sse3 x, %3
+%assign x x+8
+%endif
+%if %1 == 2 || (%1 == 6 && (z % 2) == 0)
+    FILTERH_W2_4_sse3 x, %3
+%endif
+%assign z z+1
+%if z < y
+%if %1 == 2
+    lea         r0,     [r0 + 4 * r1]
+    lea         r2,     [r2 + 4 * r3]
+%elif %1 <= 6
+    lea         r0,     [r0 + 2 * r1]
+    lea         r2,     [r2 + 2 * r3]
+%else
+    lea         r0,     [r0 + r1]
+    lea         r2,     [r2 + r3]
+%endif
+%endif ;z < y
+%endrep
+
+    RET
+%endmacro
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_horiz_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+
+FILTER_HOR_CHROMA_sse3 2, 4, pp
+FILTER_HOR_CHROMA_sse3 2, 8, pp
+FILTER_HOR_CHROMA_sse3 2, 16, pp
+FILTER_HOR_CHROMA_sse3 4, 2, pp
+FILTER_HOR_CHROMA_sse3 4, 4, pp
+FILTER_HOR_CHROMA_sse3 4, 8, pp
+FILTER_HOR_CHROMA_sse3 4, 16, pp
+FILTER_HOR_CHROMA_sse3 4, 32, pp
+FILTER_HOR_CHROMA_sse3 6, 8, pp
+FILTER_HOR_CHROMA_sse3 6, 16, pp
+FILTER_HOR_CHROMA_sse3 8, 2, pp
+FILTER_HOR_CHROMA_sse3 8, 4, pp
+FILTER_HOR_CHROMA_sse3 8, 6, pp
+FILTER_HOR_CHROMA_sse3 8, 8, pp
+FILTER_HOR_CHROMA_sse3 8, 12, pp
+FILTER_HOR_CHROMA_sse3 8, 16, pp
+FILTER_HOR_CHROMA_sse3 8, 32, pp
+FILTER_HOR_CHROMA_sse3 8, 64, pp
+FILTER_HOR_CHROMA_sse3 12, 16, pp
+FILTER_HOR_CHROMA_sse3 12, 32, pp
+FILTER_HOR_CHROMA_sse3 16, 4, pp
+FILTER_HOR_CHROMA_sse3 16, 8, pp
+FILTER_HOR_CHROMA_sse3 16, 12, pp
+FILTER_HOR_CHROMA_sse3 16, 16, pp
+FILTER_HOR_CHROMA_sse3 16, 24, pp
+FILTER_HOR_CHROMA_sse3 16, 32, pp
+FILTER_HOR_CHROMA_sse3 16, 64, pp
+FILTER_HOR_CHROMA_sse3 24, 32, pp
+FILTER_HOR_CHROMA_sse3 24, 64, pp
+FILTER_HOR_CHROMA_sse3 32, 8, pp
+FILTER_HOR_CHROMA_sse3 32, 16, pp
+FILTER_HOR_CHROMA_sse3 32, 24, pp
+FILTER_HOR_CHROMA_sse3 32, 32, pp
+FILTER_HOR_CHROMA_sse3 32, 48, pp
+FILTER_HOR_CHROMA_sse3 32, 64, pp
+FILTER_HOR_CHROMA_sse3 48, 64, pp
+FILTER_HOR_CHROMA_sse3 64, 16, pp
+FILTER_HOR_CHROMA_sse3 64, 32, pp
+FILTER_HOR_CHROMA_sse3 64, 48, pp
+FILTER_HOR_CHROMA_sse3 64, 64, pp
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_horiz_ps_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+
+FILTER_HOR_CHROMA_sse3 2, 4, ps
+FILTER_HOR_CHROMA_sse3 2, 8, ps
+FILTER_HOR_CHROMA_sse3 2, 16, ps
+FILTER_HOR_CHROMA_sse3 4, 2, ps
+FILTER_HOR_CHROMA_sse3 4, 4, ps
+FILTER_HOR_CHROMA_sse3 4, 8, ps
+FILTER_HOR_CHROMA_sse3 4, 16, ps
+FILTER_HOR_CHROMA_sse3 4, 32, ps
+FILTER_HOR_CHROMA_sse3 6, 8, ps
+FILTER_HOR_CHROMA_sse3 6, 16, ps
+FILTER_HOR_CHROMA_sse3 8, 2, ps
+FILTER_HOR_CHROMA_sse3 8, 4, ps
+FILTER_HOR_CHROMA_sse3 8, 6, ps
+FILTER_HOR_CHROMA_sse3 8, 8, ps
+FILTER_HOR_CHROMA_sse3 8, 12, ps
+FILTER_HOR_CHROMA_sse3 8, 16, ps
+FILTER_HOR_CHROMA_sse3 8, 32, ps
+FILTER_HOR_CHROMA_sse3 8, 64, ps
+FILTER_HOR_CHROMA_sse3 12, 16, ps
+FILTER_HOR_CHROMA_sse3 12, 32, ps
+FILTER_HOR_CHROMA_sse3 16, 4, ps
+FILTER_HOR_CHROMA_sse3 16, 8, ps
+FILTER_HOR_CHROMA_sse3 16, 12, ps
+FILTER_HOR_CHROMA_sse3 16, 16, ps
+FILTER_HOR_CHROMA_sse3 16, 24, ps
+FILTER_HOR_CHROMA_sse3 16, 32, ps
+FILTER_HOR_CHROMA_sse3 16, 64, ps
+FILTER_HOR_CHROMA_sse3 24, 32, ps
+FILTER_HOR_CHROMA_sse3 24, 64, ps
+FILTER_HOR_CHROMA_sse3 32, 8, ps
+FILTER_HOR_CHROMA_sse3 32, 16, ps
+FILTER_HOR_CHROMA_sse3 32, 24, ps
+FILTER_HOR_CHROMA_sse3 32, 32, ps
+FILTER_HOR_CHROMA_sse3 32, 48, ps
+FILTER_HOR_CHROMA_sse3 32, 64, ps
+FILTER_HOR_CHROMA_sse3 48, 64, ps
+FILTER_HOR_CHROMA_sse3 64, 16, ps
+FILTER_HOR_CHROMA_sse3 64, 32, ps
+FILTER_HOR_CHROMA_sse3 64, 48, ps
+FILTER_HOR_CHROMA_sse3 64, 64, ps
+
+%macro FILTER_P2S_2_4_sse2 1
+    movd        m0,     [r0 + %1]
+    movd        m2,     [r0 + r1 * 2 + %1]
+    movhps      m0,     [r0 + r1 + %1]
+    movhps      m2,     [r0 + r4 + %1]
+    psllw       m0,     (14 - BIT_DEPTH)
+    psllw       m2,     (14 - BIT_DEPTH)
+    psubw       m0,     m1
+    psubw       m2,     m1
+
+    movd        [r2 + r3 * 0 + %1], m0
+    movd        [r2 + r3 * 2 + %1], m2
+    movhlps     m0,     m0
+    movhlps     m2,     m2
+    movd        [r2 + r3 * 1 + %1], m0
+    movd        [r2 + r5 + %1], m2
+%endmacro
+
+%macro FILTER_P2S_4_4_sse2 1
+    movh        m0,     [r0 + %1]
+    movhps      m0,     [r0 + r1 + %1]
+    psllw       m0,     (14 - BIT_DEPTH)
+    psubw       m0,     m1
+    movh        [r2 + r3 * 0 + %1], m0
+    movhps      [r2 + r3 * 1 + %1], m0
+
+    movh        m2,     [r0 + r1 * 2 + %1]
+    movhps      m2,     [r0 + r4 + %1]
+    psllw       m2,     (14 - BIT_DEPTH)
+    psubw       m2,     m1
+    movh        [r2 + r3 * 2 + %1], m2
+    movhps      [r2 + r5 + %1], m2
+%endmacro
+
+%macro FILTER_P2S_4_2_sse2 0
+    movh        m0,     [r0]
+    movhps      m0,     [r0 + r1 * 2]
+    psllw       m0,     (14 - BIT_DEPTH)
+    psubw       m0,     [pw_2000]
+    movh        [r2 + r3 * 0], m0
+    movhps      [r2 + r3 * 2], m0
+%endmacro
+
+%macro FILTER_P2S_8_4_sse2 1
+    movu        m0,     [r0 + %1]
+    movu        m2,     [r0 + r1 + %1]
+    psllw       m0,     (14 - BIT_DEPTH)
+    psllw       m2,     (14 - BIT_DEPTH)
+    psubw       m0,     m1
+    psubw       m2,     m1
+    movu        [r2 + r3 * 0 + %1], m0
+    movu        [r2 + r3 * 1 + %1], m2
+
+    movu        m3,     [r0 + r1 * 2 + %1]
+    movu        m4,     [r0 + r4 + %1]
+    psllw       m3,     (14 - BIT_DEPTH)
+    psllw       m4,     (14 - BIT_DEPTH)
+    psubw       m3,     m1
+    psubw       m4,     m1
+    movu        [r2 + r3 * 2 + %1], m3
+    movu        [r2 + r5 + %1], m4
+%endmacro
+
+%macro FILTER_P2S_8_2_sse2 1
+    movu        m0,     [r0 + %1]
+    movu        m2,     [r0 + r1 + %1]
+    psllw       m0,     (14 - BIT_DEPTH)
+    psllw       m2,     (14 - BIT_DEPTH)
+    psubw       m0,     m1
+    psubw       m2,     m1
+    movu        [r2 + r3 * 0 + %1], m0
+    movu        [r2 + r3 * 1 + %1], m2
+%endmacro
+
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride)
+;-----------------------------------------------------------------------------
+%macro FILTER_PIX_TO_SHORT_sse2 2
+INIT_XMM sse2
+cglobal filterPixelToShort_%1x%2, 4, 6, 3
+%if %2 == 2
+%if %1 == 4
+    FILTER_P2S_4_2_sse2
+%elif %1 == 8
+    add        r1d, r1d
+    add        r3d, r3d
+    mova       m1, [pw_2000]
+    FILTER_P2S_8_2_sse2 0
+%endif
+%else
+    add        r1d, r1d
+    add        r3d, r3d
+    mova       m1, [pw_2000]
+    lea        r4, [r1 * 3]
+    lea        r5, [r3 * 3]
+%assign y 1
+%rep %2/4
+%assign x 0
+%rep %1/8
+    FILTER_P2S_8_4_sse2 x
+%if %2 == 6
+    lea         r0,     [r0 + 4 * r1]
+    lea         r2,     [r2 + 4 * r3]
+    FILTER_P2S_8_2_sse2 x
+%endif
+%assign x x+16
+%endrep
+%rep (%1 % 8)/4
+    FILTER_P2S_4_4_sse2 x
+%assign x x+8
+%endrep
+%rep (%1 % 4)/2
+    FILTER_P2S_2_4_sse2 x
+%endrep
+%if y < %2/4
+    lea         r0,     [r0 + 4 * r1]
+    lea         r2,     [r2 + 4 * r3]
+%assign y y+1
+%endif
+%endrep
+%endif
+RET
+%endmacro
+
+    FILTER_PIX_TO_SHORT_sse2 2, 4
+    FILTER_PIX_TO_SHORT_sse2 2, 8
+    FILTER_PIX_TO_SHORT_sse2 2, 16
+    FILTER_PIX_TO_SHORT_sse2 4, 2
+    FILTER_PIX_TO_SHORT_sse2 4, 4
+    FILTER_PIX_TO_SHORT_sse2 4, 8
+    FILTER_PIX_TO_SHORT_sse2 4, 16
+    FILTER_PIX_TO_SHORT_sse2 4, 32
+    FILTER_PIX_TO_SHORT_sse2 6, 8
+    FILTER_PIX_TO_SHORT_sse2 6, 16
+    FILTER_PIX_TO_SHORT_sse2 8, 2
+    FILTER_PIX_TO_SHORT_sse2 8, 4
+    FILTER_PIX_TO_SHORT_sse2 8, 6
+    FILTER_PIX_TO_SHORT_sse2 8, 8
+    FILTER_PIX_TO_SHORT_sse2 8, 12
+    FILTER_PIX_TO_SHORT_sse2 8, 16
+    FILTER_PIX_TO_SHORT_sse2 8, 32
+    FILTER_PIX_TO_SHORT_sse2 8, 64
+    FILTER_PIX_TO_SHORT_sse2 12, 16
+    FILTER_PIX_TO_SHORT_sse2 12, 32
+    FILTER_PIX_TO_SHORT_sse2 16, 4
+    FILTER_PIX_TO_SHORT_sse2 16, 8
+    FILTER_PIX_TO_SHORT_sse2 16, 12
+    FILTER_PIX_TO_SHORT_sse2 16, 16
+    FILTER_PIX_TO_SHORT_sse2 16, 24
+    FILTER_PIX_TO_SHORT_sse2 16, 32
+    FILTER_PIX_TO_SHORT_sse2 16, 64
+    FILTER_PIX_TO_SHORT_sse2 24, 32
+    FILTER_PIX_TO_SHORT_sse2 24, 64
+    FILTER_PIX_TO_SHORT_sse2 32, 8
+    FILTER_PIX_TO_SHORT_sse2 32, 16
+    FILTER_PIX_TO_SHORT_sse2 32, 24
+    FILTER_PIX_TO_SHORT_sse2 32, 32
+    FILTER_PIX_TO_SHORT_sse2 32, 48
+    FILTER_PIX_TO_SHORT_sse2 32, 64
+    FILTER_PIX_TO_SHORT_sse2 48, 64
+    FILTER_PIX_TO_SHORT_sse2 64, 16
+    FILTER_PIX_TO_SHORT_sse2 64, 32
+    FILTER_PIX_TO_SHORT_sse2 64, 48
+    FILTER_PIX_TO_SHORT_sse2 64, 64
+
+;------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_pp_4x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;------------------------------------------------------------------------------------------------------------
+%macro FILTER_HOR_LUMA_W4 3
+INIT_XMM sse4
+cglobal interp_8tap_horiz_%3_%1x%2, 4, 7, 8
+    mov         r4d, r4m
+    sub         r0, 6
+    shl         r4d, 4
+    add         r1, r1
+    add         r3, r3
+
+%ifdef PIC
+    lea         r6, [tab_LumaCoeff]
+    mova        m0, [r6 + r4]
+%else
+    mova        m0, [tab_LumaCoeff + r4]
+%endif
+
+%ifidn %3, pp
+    mova        m1, [pd_32]
+    pxor        m6, m6
+    mova        m7, [pw_pixel_max]
+%else
+    mova        m1, [INTERP_OFFSET_PS]
+%endif
+
+    mov         r4d, %2
+%ifidn %3, ps
+    cmp         r5m, byte 0
+    je          .loopH
+    lea         r6, [r1 + 2 * r1]
+    sub         r0, r6
+    add         r4d, 7
+%endif
+
+.loopH:
+    movu        m2, [r0]                     ; m2 = src[0-7]
+    movu        m3, [r0 + 16]                ; m3 = src[8-15]
+
+    pmaddwd     m4, m2, m0
+    palignr     m5, m3, m2, 2                ; m5 = src[1-8]
+    pmaddwd     m5, m0
+    phaddd      m4, m5
+
+    palignr     m5, m3, m2, 4                ; m5 = src[2-9]
+    pmaddwd     m5, m0
+    palignr     m3, m2, 6                    ; m3 = src[3-10]
+    pmaddwd     m3, m0
+    phaddd      m5, m3
+
+    phaddd      m4, m5
+    paddd       m4, m1
+%ifidn %3, pp
+    psrad       m4, 6
+    packusdw    m4, m4
+    CLIPW       m4, m6, m7
+%else
+    psrad       m4, INTERP_SHIFT_PS
+    packssdw    m4, m4
+%endif
+
+    movh        [r2], m4
+
+    add         r0, r1
+    add         r2, r3
+
+    dec         r4d
+    jnz         .loopH
+    RET
+%endmacro
+
+;------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_pp_4x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
+;------------------------------------------------------------------------------------------------------------
+FILTER_HOR_LUMA_W4 4, 4, pp
+FILTER_HOR_LUMA_W4 4, 8, pp
+FILTER_HOR_LUMA_W4 4, 16, pp
+
+;---------------------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_ps_4x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;---------------------------------------------------------------------------------------------------------------------------
+FILTER_HOR_LUMA_W4 4, 4, ps
+FILTER_HOR_LUMA_W4 4, 8, ps
+FILTER_HOR_LUMA_W4 4, 16, ps
+
+;------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;------------------------------------------------------------------------------------------------------------
+%macro FILTER_HOR_LUMA_W8 3
+INIT_XMM sse4
+cglobal interp_8tap_horiz_%3_%1x%2, 4, 7, 8
+
+    add         r1, r1
+    add         r3, r3
+    mov         r4d, r4m
+    sub         r0, 6
+    shl         r4d, 4
+
+%ifdef PIC
+    lea         r6, [tab_LumaCoeff]
+    mova        m0, [r6 + r4]
+%else
+    mova        m0, [tab_LumaCoeff + r4]
+%endif
+
+%ifidn %3, pp
+    mova        m1, [pd_32]
+    pxor        m7, m7
+%else
+    mova        m1, [INTERP_OFFSET_PS]
+%endif
+
+    mov         r4d, %2
+%ifidn %3, ps
+    cmp         r5m, byte 0
+    je         .loopH
+    lea         r6, [r1 + 2 * r1]
+    sub         r0, r6
+    add         r4d, 7
+%endif
+
+.loopH:
+    movu        m2, [r0]                     ; m2 = src[0-7]
+    movu        m3, [r0 + 16]                ; m3 = src[8-15]
+
+    pmaddwd     m4, m2, m0
+    palignr     m5, m3, m2, 2                ; m5 = src[1-8]
+    pmaddwd     m5, m0
+    phaddd      m4, m5
+
+    palignr     m5, m3, m2, 4                ; m5 = src[2-9]
+    pmaddwd     m5, m0
+    palignr     m6, m3, m2, 6                ; m6 = src[3-10]
+    pmaddwd     m6, m0
+    phaddd      m5, m6
+    phaddd      m4, m5
+    paddd       m4, m1
+
+    palignr     m5, m3, m2, 8                ; m5 = src[4-11]
+    pmaddwd     m5, m0
+    palignr     m6, m3, m2, 10               ; m6 = src[5-12]
+    pmaddwd     m6, m0
+    phaddd      m5, m6
+
+    palignr     m6, m3, m2, 12               ; m6 = src[6-13]
+    pmaddwd     m6, m0
+    palignr     m3, m2, 14                   ; m3 = src[7-14]
+    pmaddwd     m3, m0
+    phaddd      m6, m3
+    phaddd      m5, m6
+    paddd       m5, m1
+%ifidn %3, pp
+    psrad       m4, 6
+    psrad       m5, 6
+    packusdw    m4, m5
+    CLIPW       m4, m7, [pw_pixel_max]
+%else
+    psrad       m4, INTERP_SHIFT_PS
+    psrad       m5, INTERP_SHIFT_PS
+    packssdw    m4, m5
+%endif
+
+    movu        [r2], m4
+
+    add         r0, r1
+    add         r2, r3
+
+    dec         r4d
+    jnz        .loopH
+    RET
+%endmacro
+
+;------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_pp_8x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
+;------------------------------------------------------------------------------------------------------------
+FILTER_HOR_LUMA_W8 8, 4, pp
+FILTER_HOR_LUMA_W8 8, 8, pp
+FILTER_HOR_LUMA_W8 8, 16, pp
+FILTER_HOR_LUMA_W8 8, 32, pp
+
+;---------------------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_ps_8x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;---------------------------------------------------------------------------------------------------------------------------
+FILTER_HOR_LUMA_W8 8, 4, ps
+FILTER_HOR_LUMA_W8 8, 8, ps
+FILTER_HOR_LUMA_W8 8, 16, ps
+FILTER_HOR_LUMA_W8 8, 32, ps
+
+;--------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+%macro FILTER_HOR_LUMA_W12 3
+INIT_XMM sse4
+cglobal interp_8tap_horiz_%3_%1x%2, 4, 7, 8
+
+    add         r1, r1
+    add         r3, r3
+    mov         r4d, r4m
+    sub         r0, 6
+    shl         r4d, 4
+
+%ifdef PIC
+    lea         r6, [tab_LumaCoeff]
+    mova        m0, [r6 + r4]
+%else
+    mova        m0, [tab_LumaCoeff + r4]
+%endif
+%ifidn %3, pp
+    mova        m1, [INTERP_OFFSET_PP]
+%else
+    mova        m1, [INTERP_OFFSET_PS]
+%endif
+
+    mov         r4d, %2
+%ifidn %3, ps
+    cmp         r5m, byte 0
+    je          .loopH
+    lea         r6, [r1 + 2 * r1]
+    sub         r0, r6
+    add         r4d, 7
+%endif
+
+.loopH:
+    movu        m2, [r0]                     ; m2 = src[0-7]
+    movu        m3, [r0 + 16]                ; m3 = src[8-15]
+
+    pmaddwd     m4, m2, m0
+    palignr     m5, m3, m2, 2                ; m5 = src[1-8]
+    pmaddwd     m5, m0
+    phaddd      m4, m5
+
+    palignr     m5, m3, m2, 4                ; m5 = src[2-9]
+    pmaddwd     m5, m0
+    palignr     m6, m3, m2, 6                ; m6 = src[3-10]
+    pmaddwd     m6, m0
+    phaddd      m5, m6
+    phaddd      m4, m5
+    paddd       m4, m1
+
+    palignr     m5, m3, m2, 8                ; m5 = src[4-11]
+    pmaddwd     m5, m0
+    palignr     m6, m3, m2, 10               ; m6 = src[5-12]
+    pmaddwd     m6, m0
+    phaddd      m5, m6
+
+    palignr     m6, m3, m2, 12               ; m6 = src[6-13]
+    pmaddwd     m6, m0
+    palignr     m7, m3, m2, 14               ; m2 = src[7-14]
+    pmaddwd     m7, m0
+    phaddd      m6, m7
+    phaddd      m5, m6
+    paddd       m5, m1
+%ifidn %3, pp
+    psrad       m4, INTERP_SHIFT_PP
+    psrad       m5, INTERP_SHIFT_PP
+    packusdw    m4, m5
+    pxor        m5, m5
+    CLIPW       m4, m5, [pw_pixel_max]
+%else
+    psrad       m4, INTERP_SHIFT_PS
+    psrad       m5, INTERP_SHIFT_PS
+    packssdw    m4, m5
+%endif
+
+    movu        [r2], m4
+
+    movu        m2, [r0 + 32]                ; m2 = src[16-23]
+
+    pmaddwd     m4, m3, m0                   ; m3 = src[8-15]
+    palignr     m5, m2, m3, 2                ; m5 = src[9-16]
+    pmaddwd     m5, m0
+    phaddd      m4, m5
+
+    palignr     m5, m2, m3, 4                ; m5 = src[10-17]
+    pmaddwd     m5, m0
+    palignr     m2, m3, 6                    ; m2 = src[11-18]
+    pmaddwd     m2, m0
+    phaddd      m5, m2
+    phaddd      m4, m5
+    paddd       m4, m1
+%ifidn %3, pp
+    psrad       m4, INTERP_SHIFT_PP
+    packusdw    m4, m4
+    pxor        m5, m5
+    CLIPW       m4, m5, [pw_pixel_max]
+%else
+    psrad       m4, INTERP_SHIFT_PS
+    packssdw    m4, m4
+%endif
+
+    movh        [r2 + 16], m4
+
+    add         r0, r1
+    add         r2, r3
+
+    dec         r4d
+    jnz         .loopH
+    RET
+%endmacro
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_pp_12x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+FILTER_HOR_LUMA_W12 12, 16, pp
+
+;----------------------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_ps_12x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;----------------------------------------------------------------------------------------------------------------------------
+FILTER_HOR_LUMA_W12 12, 16, ps
+
+;--------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+%macro FILTER_HOR_LUMA_W16 3
+INIT_XMM sse4
+cglobal interp_8tap_horiz_%3_%1x%2, 4, 7, 8
+
+    add         r1, r1
+    add         r3, r3
+    mov         r4d, r4m
+    sub         r0, 6
+    shl         r4d, 4
+
+%ifdef PIC
+    lea         r6, [tab_LumaCoeff]
+    mova        m0, [r6 + r4]
+%else
+    mova        m0, [tab_LumaCoeff + r4]
+%endif
+
+%ifidn %3, pp
+    mova        m1, [pd_32]
+%else
+    mova        m1, [INTERP_OFFSET_PS]
+%endif
+
+    mov         r4d, %2
+%ifidn %3, ps
+    cmp         r5m, byte 0
+    je          .loopH
+    lea         r6, [r1 + 2 * r1]
+    sub         r0, r6
+    add         r4d, 7
+%endif
+
+.loopH:
+%assign x 0
+%rep %1 / 16
+    movu        m2, [r0 + x]                 ; m2 = src[0-7]
+    movu        m3, [r0 + 16 + x]            ; m3 = src[8-15]
+
+    pmaddwd     m4, m2, m0
+    palignr     m5, m3, m2, 2                ; m5 = src[1-8]
+    pmaddwd     m5, m0
+    phaddd      m4, m5
+
+    palignr     m5, m3, m2, 4                ; m5 = src[2-9]
+    pmaddwd     m5, m0
+    palignr     m6, m3, m2, 6                ; m6 = src[3-10]
+    pmaddwd     m6, m0
+    phaddd      m5, m6
+    phaddd      m4, m5
+    paddd       m4, m1
+
+    palignr     m5, m3, m2, 8                ; m5 = src[4-11]
+    pmaddwd     m5, m0
+    palignr     m6, m3, m2, 10               ; m6 = src[5-12]
+    pmaddwd     m6, m0
+    phaddd      m5, m6
+
+    palignr     m6, m3, m2, 12               ; m6 = src[6-13]
+    pmaddwd     m6, m0
+    palignr     m7, m3, m2, 14               ; m2 = src[7-14]
+    pmaddwd     m7, m0
+    phaddd      m6, m7
+    phaddd      m5, m6
+    paddd       m5, m1
+%ifidn %3, pp
+    psrad       m4, INTERP_SHIFT_PP
+    psrad       m5, INTERP_SHIFT_PP
+    packusdw    m4, m5
+    pxor        m5, m5
+    CLIPW       m4, m5, [pw_pixel_max]
+%else
+    psrad       m4, INTERP_SHIFT_PS
+    psrad       m5, INTERP_SHIFT_PS
+    packssdw    m4, m5
+%endif
+    movu        [r2 + x], m4
+
+    movu        m2, [r0 + 32 + x]            ; m2 = src[16-23]
+
+    pmaddwd     m4, m3, m0                   ; m3 = src[8-15]
+    palignr     m5, m2, m3, 2                ; m5 = src[9-16]
+    pmaddwd     m5, m0
+    phaddd      m4, m5
+
+    palignr     m5, m2, m3, 4                ; m5 = src[10-17]
+    pmaddwd     m5, m0
+    palignr     m6, m2, m3, 6                ; m6 = src[11-18]
+    pmaddwd     m6, m0
+    phaddd      m5, m6
+    phaddd      m4, m5
+    paddd       m4, m1
+
+    palignr     m5, m2, m3, 8                ; m5 = src[12-19]
+    pmaddwd     m5, m0
+    palignr     m6, m2, m3, 10               ; m6 = src[13-20]
+    pmaddwd     m6, m0
+    phaddd      m5, m6
+
+    palignr     m6, m2, m3, 12               ; m6 = src[14-21]
+    pmaddwd     m6, m0
+    palignr     m2, m3, 14                   ; m3 = src[15-22]
+    pmaddwd     m2, m0
+    phaddd      m6, m2
+    phaddd      m5, m6
+    paddd       m5, m1
+%ifidn %3, pp
+    psrad       m4, INTERP_SHIFT_PP
+    psrad       m5, INTERP_SHIFT_PP
+    packusdw    m4, m5
+    pxor        m5, m5
+    CLIPW       m4, m5, [pw_pixel_max]
+%else
+    psrad       m4, INTERP_SHIFT_PS
+    psrad       m5, INTERP_SHIFT_PS
+    packssdw    m4, m5
+%endif
+    movu        [r2 + 16 + x], m4
+
+%assign x x+32
+%endrep
+
+    add         r0, r1
+    add         r2, r3
+
+    dec         r4d
+    jnz         .loopH
+    RET
+%endmacro
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_pp_16x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+FILTER_HOR_LUMA_W16 16, 4, pp
+FILTER_HOR_LUMA_W16 16, 8, pp
+FILTER_HOR_LUMA_W16 16, 12, pp
+FILTER_HOR_LUMA_W16 16, 16, pp
+FILTER_HOR_LUMA_W16 16, 32, pp
+FILTER_HOR_LUMA_W16 16, 64, pp
+
+;----------------------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_ps_16x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;----------------------------------------------------------------------------------------------------------------------------
+FILTER_HOR_LUMA_W16 16, 4, ps
+FILTER_HOR_LUMA_W16 16, 8, ps
+FILTER_HOR_LUMA_W16 16, 12, ps
+FILTER_HOR_LUMA_W16 16, 16, ps
+FILTER_HOR_LUMA_W16 16, 32, ps
+FILTER_HOR_LUMA_W16 16, 64, ps
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_pp_32x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+FILTER_HOR_LUMA_W16 32, 8, pp
+FILTER_HOR_LUMA_W16 32, 16, pp
+FILTER_HOR_LUMA_W16 32, 24, pp
+FILTER_HOR_LUMA_W16 32, 32, pp
+FILTER_HOR_LUMA_W16 32, 64, pp
+
+;----------------------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_ps_32x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;----------------------------------------------------------------------------------------------------------------------------
+FILTER_HOR_LUMA_W16 32, 8, ps
+FILTER_HOR_LUMA_W16 32, 16, ps
+FILTER_HOR_LUMA_W16 32, 24, ps
+FILTER_HOR_LUMA_W16 32, 32, ps
+FILTER_HOR_LUMA_W16 32, 64, ps
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_pp_48x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+FILTER_HOR_LUMA_W16 48, 64, pp
+
+;----------------------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_ps_48x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;----------------------------------------------------------------------------------------------------------------------------
+FILTER_HOR_LUMA_W16 48, 64, ps
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_pp_64x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+FILTER_HOR_LUMA_W16 64, 16, pp
+FILTER_HOR_LUMA_W16 64, 32, pp
+FILTER_HOR_LUMA_W16 64, 48, pp
+FILTER_HOR_LUMA_W16 64, 64, pp
+
+;----------------------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_ps_64x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;----------------------------------------------------------------------------------------------------------------------------
+FILTER_HOR_LUMA_W16 64, 16, ps
+FILTER_HOR_LUMA_W16 64, 32, ps
+FILTER_HOR_LUMA_W16 64, 48, ps
+FILTER_HOR_LUMA_W16 64, 64, ps
+
+;--------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+%macro FILTER_HOR_LUMA_W24 3
+INIT_XMM sse4
+cglobal interp_8tap_horiz_%3_%1x%2, 4, 7, 8
+
+    add         r1, r1
+    add         r3, r3
+    mov         r4d, r4m
+    sub         r0, 6
+    shl         r4d, 4
+
+%ifdef PIC
+    lea         r6, [tab_LumaCoeff]
+    mova        m0, [r6 + r4]
+%else
+    mova        m0, [tab_LumaCoeff + r4]
+%endif
+%ifidn %3, pp
+    mova        m1, [pd_32]
+%else
+    mova        m1, [INTERP_OFFSET_PS]
+%endif
+
+    mov         r4d, %2
+%ifidn %3, ps
+    cmp         r5m, byte 0
+    je          .loopH
+    lea         r6, [r1 + 2 * r1]
+    sub         r0, r6
+    add         r4d, 7
+%endif
+
+.loopH:
+    movu        m2, [r0]                     ; m2 = src[0-7]
+    movu        m3, [r0 + 16]                ; m3 = src[8-15]
+
+    pmaddwd     m4, m2, m0
+    palignr     m5, m3, m2, 2                ; m5 = src[1-8]
+    pmaddwd     m5, m0
+    phaddd      m4, m5
+
+    palignr     m5, m3, m2, 4                ; m5 = src[2-9]
+    pmaddwd     m5, m0
+    palignr     m6, m3, m2, 6                ; m6 = src[3-10]
+    pmaddwd     m6, m0
+    phaddd      m5, m6
+    phaddd      m4, m5
+    paddd       m4, m1
+
+    palignr     m5, m3, m2, 8                ; m5 = src[4-11]
+    pmaddwd     m5, m0
+    palignr     m6, m3, m2, 10               ; m6 = src[5-12]
+    pmaddwd     m6, m0
+    phaddd      m5, m6
+
+    palignr     m6, m3, m2, 12               ; m6 = src[6-13]
+    pmaddwd     m6, m0
+    palignr     m7, m3, m2, 14               ; m7 = src[7-14]
+    pmaddwd     m7, m0
+    phaddd      m6, m7
+    phaddd      m5, m6
+    paddd       m5, m1
+%ifidn %3, pp
+    psrad       m4, INTERP_SHIFT_PP
+    psrad       m5, INTERP_SHIFT_PP
+    packusdw    m4, m5
+    pxor        m5, m5
+    CLIPW       m4, m5, [pw_pixel_max]
+%else
+    psrad       m4, INTERP_SHIFT_PS
+    psrad       m5, INTERP_SHIFT_PS
+    packssdw    m4, m5
+%endif
+    movu        [r2], m4
+
+    movu        m2, [r0 + 32]                ; m2 = src[16-23]
+
+    pmaddwd     m4, m3, m0                   ; m3 = src[8-15]
+    palignr     m5, m2, m3, 2                ; m5 = src[1-8]
+    pmaddwd     m5, m0
+    phaddd      m4, m5
+
+    palignr     m5, m2, m3, 4                ; m5 = src[2-9]
+    pmaddwd     m5, m0
+    palignr     m6, m2, m3, 6                ; m6 = src[3-10]
+    pmaddwd     m6, m0
+    phaddd      m5, m6
+    phaddd      m4, m5
+    paddd       m4, m1
+
+    palignr     m5, m2, m3, 8                ; m5 = src[4-11]
+    pmaddwd     m5, m0
+    palignr     m6, m2, m3, 10               ; m6 = src[5-12]
+    pmaddwd     m6, m0
+    phaddd      m5, m6
+
+    palignr     m6, m2, m3, 12               ; m6 = src[6-13]
+    pmaddwd     m6, m0
+    palignr     m7, m2, m3, 14               ; m7 = src[7-14]
+    pmaddwd     m7, m0
+    phaddd      m6, m7
+    phaddd      m5, m6
+    paddd       m5, m1
+%ifidn %3, pp
+    psrad       m4, INTERP_SHIFT_PP
+    psrad       m5, INTERP_SHIFT_PP
+    packusdw    m4, m5
+    pxor        m5, m5
+    CLIPW       m4, m5, [pw_pixel_max]
+%else
+    psrad       m4, INTERP_SHIFT_PS
+    psrad       m5, INTERP_SHIFT_PS
+    packssdw    m4, m5
+%endif
+    movu        [r2 + 16], m4
+
+    movu        m3, [r0 + 48]                ; m3 = src[24-31]
+
+    pmaddwd     m4, m2, m0                   ; m2 = src[16-23]
+    palignr     m5, m3, m2, 2                ; m5 = src[1-8]
+    pmaddwd     m5, m0
+    phaddd      m4, m5
+
+    palignr     m5, m3, m2, 4                ; m5 = src[2-9]
+    pmaddwd     m5, m0
+    palignr     m6, m3, m2, 6                ; m6 = src[3-10]
+    pmaddwd     m6, m0
+    phaddd      m5, m6
+    phaddd      m4, m5
+    paddd       m4, m1
+
+    palignr     m5, m3, m2, 8                ; m5 = src[4-11]
+    pmaddwd     m5, m0
+    palignr     m6, m3, m2, 10               ; m6 = src[5-12]
+    pmaddwd     m6, m0
+    phaddd      m5, m6
+
+    palignr     m6, m3, m2, 12               ; m6 = src[6-13]
+    pmaddwd     m6, m0
+    palignr     m7, m3, m2, 14               ; m7 = src[7-14]
+    pmaddwd     m7, m0
+    phaddd      m6, m7
+    phaddd      m5, m6
+    paddd       m5, m1
+%ifidn %3, pp
+    psrad       m4, INTERP_SHIFT_PP
+    psrad       m5, INTERP_SHIFT_PP
+    packusdw    m4, m5
+    pxor        m5, m5
+    CLIPW       m4, m5, [pw_pixel_max]
+%else
+    psrad       m4, INTERP_SHIFT_PS
+    psrad       m5, INTERP_SHIFT_PS
+    packssdw    m4, m5
+%endif
+    movu        [r2 + 32], m4
+
+    add         r0, r1
+    add         r2, r3
+
+    dec         r4d
+    jnz         .loopH
+    RET
+%endmacro
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_pp_24x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+FILTER_HOR_LUMA_W24 24, 32, pp
+
+;----------------------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_ps_24x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;----------------------------------------------------------------------------------------------------------------------------
+FILTER_HOR_LUMA_W24 24, 32, ps
+
+%macro FILTER_W2_2 1
+    movu        m3,         [r0]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + r1]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    packusdw    m3,         m3
+    CLIPW       m3,         m7,    m6
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    packssdw    m3,         m3
+%endif
+    movd        [r2],       m3
+    pextrd      [r2 + r3],  m3, 1
+%endmacro
+
+%macro FILTER_W4_2 1
+    movu        m3,         [r0]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 4]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + r1]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + r1 + 4]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m7,    m6
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2],       m3
+    movhps      [r2 + r3],  m3
+%endmacro
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_pp(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+%macro FILTER_HOR_LUMA_W4_avx2 1
+INIT_YMM avx2
+cglobal interp_8tap_horiz_pp_4x%1, 4,7,7
+    add              r1d, r1d
+    add              r3d, r3d
+    sub              r0, 6
+    mov              r4d, r4m
+    shl              r4d, 4
+%ifdef PIC
+    lea              r5, [tab_LumaCoeff]
+    vpbroadcastq     m0, [r5 + r4]
+    vpbroadcastq     m1, [r5 + r4 + 8]
+%else
+    vpbroadcastq     m0, [tab_LumaCoeff + r4]
+    vpbroadcastq     m1, [tab_LumaCoeff + r4 + 8]
+%endif
+    lea              r6, [pw_pixel_max]
+    mova             m3, [interp8_hpp_shuf]
+    mova             m6, [pd_32]
+    pxor             m2, m2
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    mov              r4d, %1/2
+
+.loop:
+    vbroadcasti128   m4, [r0]
+    vbroadcasti128   m5, [r0 + 8]
+    pshufb           m4, m3
+    pshufb           m5, m3
+
+    pmaddwd          m4, m0
+    pmaddwd          m5, m1
+    paddd            m4, m5
+
+    phaddd           m4, m4
+    vpermq           m4, m4, q3120
+    paddd            m4, m6
+    psrad            m4, INTERP_SHIFT_PP
+
+    packusdw         m4, m4
+    vpermq           m4, m4, q2020
+    CLIPW            m4, m2, [r6]
+    movq             [r2], xm4
+
+    vbroadcasti128   m4, [r0 + r1]
+    vbroadcasti128   m5, [r0 + r1 + 8]
+    pshufb           m4, m3
+    pshufb           m5, m3
+
+    pmaddwd          m4, m0
+    pmaddwd          m5, m1
+    paddd            m4, m5
+
+    phaddd           m4, m4
+    vpermq           m4, m4, q3120
+    paddd            m4, m6
+    psrad            m4, INTERP_SHIFT_PP
+
+    packusdw         m4, m4
+    vpermq           m4, m4, q2020
+    CLIPW            m4, m2, [r6]
+    movq             [r2 + r3], xm4
+
+    lea              r2, [r2 + 2 * r3]
+    lea              r0, [r0 + 2 * r1]
+    dec              r4d
+    jnz              .loop
+    RET
+%endmacro
+FILTER_HOR_LUMA_W4_avx2 4
+FILTER_HOR_LUMA_W4_avx2 8
+FILTER_HOR_LUMA_W4_avx2 16
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_pp(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+%macro FILTER_HOR_LUMA_W8 1
+INIT_YMM avx2
+cglobal interp_8tap_horiz_pp_8x%1, 4,6,8
+    add              r1d, r1d
+    add              r3d, r3d
+    sub              r0, 6
+    mov              r4d, r4m
+    shl              r4d, 4
+%ifdef PIC
+    lea              r5, [tab_LumaCoeff]
+    vpbroadcastq     m0, [r5 + r4]
+    vpbroadcastq     m1, [r5 + r4 + 8]
+%else
+    vpbroadcastq     m0, [tab_LumaCoeff + r4]
+    vpbroadcastq     m1, [h_ab_LumaCoeff + r4 + 8]
+%endif
+    mova             m3, [interp8_hpp_shuf]
+    mova             m7, [pd_32]
+    pxor             m2, m2
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    mov              r4d, %1/2
+
+.loop:
+    vbroadcasti128   m4, [r0]
+    vbroadcasti128   m5, [r0 + 8]
+    pshufb           m4, m3
+    pshufb           m5, m3
+
+    pmaddwd          m4, m0
+    pmaddwd          m5, m1
+    paddd            m4, m5
+
+    vbroadcasti128   m5, [r0 + 8]
+    vbroadcasti128   m6, [r0 + 16]
+    pshufb           m5, m3
+    pshufb           m6, m3
+
+    pmaddwd          m5, m0
+    pmaddwd          m6, m1
+    paddd            m5, m6
+
+    phaddd           m4, m5
+    vpermq           m4, m4, q3120
+    paddd            m4, m7
+    psrad            m4, INTERP_SHIFT_PP
+
+    packusdw         m4, m4
+    vpermq           m4, m4, q2020
+    CLIPW            m4, m2, [pw_pixel_max]
+    movu             [r2], xm4
+
+    vbroadcasti128   m4, [r0 + r1]
+    vbroadcasti128   m5, [r0 + r1 + 8]
+    pshufb           m4, m3
+    pshufb           m5, m3
+
+    pmaddwd          m4, m0
+    pmaddwd          m5, m1
+    paddd            m4, m5
+
+    vbroadcasti128   m5, [r0 + r1 + 8]
+    vbroadcasti128   m6, [r0 + r1 + 16]
+    pshufb           m5, m3
+    pshufb           m6, m3
+
+    pmaddwd          m5, m0
+    pmaddwd          m6, m1
+    paddd            m5, m6
+
+    phaddd           m4, m5
+    vpermq           m4, m4, q3120
+    paddd            m4, m7
+    psrad            m4, INTERP_SHIFT_PP
+
+    packusdw         m4, m4
+    vpermq           m4, m4, q2020
+    CLIPW            m4, m2, [pw_pixel_max]
+    movu             [r2 + r3], xm4
+
+    lea              r2, [r2 + 2 * r3]
+    lea              r0, [r0 + 2 * r1]
+    dec              r4d
+    jnz              .loop
+    RET
+%endmacro
+FILTER_HOR_LUMA_W8 4
+FILTER_HOR_LUMA_W8 8
+FILTER_HOR_LUMA_W8 16
+FILTER_HOR_LUMA_W8 32
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_pp(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+%macro FILTER_HOR_LUMA_W16 1
+INIT_YMM avx2
+cglobal interp_8tap_horiz_pp_16x%1, 4,6,8
+    add              r1d, r1d
+    add              r3d, r3d
+    sub              r0, 6
+    mov              r4d, r4m
+    shl              r4d, 4
+%ifdef PIC
+    lea              r5, [tab_LumaCoeff]
+    vpbroadcastq     m0, [r5 + r4]
+    vpbroadcastq     m1, [r5 + r4 + 8]
+%else
+    vpbroadcastq     m0, [tab_LumaCoeff + r4]
+    vpbroadcastq     m1, [tab_LumaCoeff + r4 + 8]
+%endif
+    mova             m3, [interp8_hpp_shuf]
+    mova             m7, [pd_32]
+    pxor             m2, m2
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    mov              r4d, %1
+
+.loop:
+    vbroadcasti128   m4, [r0]
+    vbroadcasti128   m5, [r0 + 8]
+    pshufb           m4, m3
+    pshufb           m5, m3
+
+    pmaddwd          m4, m0
+    pmaddwd          m5, m1
+    paddd            m4, m5
+
+    vbroadcasti128   m5, [r0 + 8]
+    vbroadcasti128   m6, [r0 + 16]
+    pshufb           m5, m3
+    pshufb           m6, m3
+
+    pmaddwd          m5, m0
+    pmaddwd          m6, m1
+    paddd            m5, m6
+
+    phaddd           m4, m5
+    vpermq           m4, m4, q3120
+    paddd            m4, m7
+    psrad            m4, INTERP_SHIFT_PP
+
+    packusdw         m4, m4
+    vpermq           m4, m4, q2020
+    CLIPW            m4, m2, [pw_pixel_max]
+    movu             [r2], xm4
+
+    vbroadcasti128   m4, [r0 + 16]
+    vbroadcasti128   m5, [r0 + 24]
+    pshufb           m4, m3
+    pshufb           m5, m3
+
+    pmaddwd          m4, m0
+    pmaddwd          m5, m1
+    paddd            m4, m5
+
+    vbroadcasti128   m5, [r0 + 24]
+    vbroadcasti128   m6, [r0 + 32]
+    pshufb           m5, m3
+    pshufb           m6, m3
+
+    pmaddwd          m5, m0
+    pmaddwd          m6, m1
+    paddd            m5, m6
+
+    phaddd           m4, m5
+    vpermq           m4, m4, q3120
+    paddd            m4, m7
+    psrad            m4, INTERP_SHIFT_PP
+
+    packusdw         m4, m4
+    vpermq           m4, m4, q2020
+    CLIPW            m4, m2, [pw_pixel_max]
+    movu             [r2 + 16], xm4
+
+    add              r2, r3
+    add              r0, r1
+    dec              r4d
+    jnz              .loop
+    RET
+%endmacro
+FILTER_HOR_LUMA_W16 4
+FILTER_HOR_LUMA_W16 8
+FILTER_HOR_LUMA_W16 12
+FILTER_HOR_LUMA_W16 16
+FILTER_HOR_LUMA_W16 32
+FILTER_HOR_LUMA_W16 64
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_pp(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+%macro FILTER_HOR_LUMA_W32 2
+INIT_YMM avx2
+cglobal interp_8tap_horiz_pp_%1x%2, 4,6,8
+    add              r1d, r1d
+    add              r3d, r3d
+    sub              r0, 6
+    mov              r4d, r4m
+    shl              r4d, 4
+%ifdef PIC
+    lea              r5, [tab_LumaCoeff]
+    vpbroadcastq     m0, [r5 + r4]
+    vpbroadcastq     m1, [r5 + r4 + 8]
+%else
+    vpbroadcastq     m0, [tab_LumaCoeff + r4]
+    vpbroadcastq     m1, [tab_LumaCoeff + r4 + 8]
+%endif
+    mova             m3, [interp8_hpp_shuf]
+    mova             m7, [pd_32]
+    pxor             m2, m2
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    mov              r4d, %2
+
+.loop:
+%assign x 0
+%rep %1/16
+    vbroadcasti128   m4, [r0 + x]
+    vbroadcasti128   m5, [r0 + 8 + x]
+    pshufb           m4, m3
+    pshufb           m5, m3
+
+    pmaddwd          m4, m0
+    pmaddwd          m5, m1
+    paddd            m4, m5
+
+    vbroadcasti128   m5, [r0 + 8 + x]
+    vbroadcasti128   m6, [r0 + 16 + x]
+    pshufb           m5, m3
+    pshufb           m6, m3
+
+    pmaddwd          m5, m0
+    pmaddwd          m6, m1
+    paddd            m5, m6
+
+    phaddd           m4, m5
+    vpermq           m4, m4, q3120
+    paddd            m4, m7
+    psrad            m4, INTERP_SHIFT_PP
+
+    packusdw         m4, m4
+    vpermq           m4, m4, q2020
+    CLIPW            m4, m2, [pw_pixel_max]
+    movu             [r2 + x], xm4
+
+    vbroadcasti128   m4, [r0 + 16 + x]
+    vbroadcasti128   m5, [r0 + 24 + x]
+    pshufb           m4, m3
+    pshufb           m5, m3
+
+    pmaddwd          m4, m0
+    pmaddwd          m5, m1
+    paddd            m4, m5
+
+    vbroadcasti128   m5, [r0 + 24 + x]
+    vbroadcasti128   m6, [r0 + 32 + x]
+    pshufb           m5, m3
+    pshufb           m6, m3
+
+    pmaddwd          m5, m0
+    pmaddwd          m6, m1
+    paddd            m5, m6
+
+    phaddd           m4, m5
+    vpermq           m4, m4, q3120
+    paddd            m4, m7
+    psrad            m4, INTERP_SHIFT_PP
+
+    packusdw         m4, m4
+    vpermq           m4, m4, q2020
+    CLIPW            m4, m2, [pw_pixel_max]
+    movu             [r2 + 16 + x], xm4
+
+%assign x x+32
+%endrep
+
+    add              r2, r3
+    add              r0, r1
+    dec              r4d
+    jnz              .loop
+    RET
+%endmacro
+FILTER_HOR_LUMA_W32 32, 8
+FILTER_HOR_LUMA_W32 32, 16
+FILTER_HOR_LUMA_W32 32, 24
+FILTER_HOR_LUMA_W32 32, 32
+FILTER_HOR_LUMA_W32 32, 64
+FILTER_HOR_LUMA_W32 64, 16
+FILTER_HOR_LUMA_W32 64, 32
+FILTER_HOR_LUMA_W32 64, 48
+FILTER_HOR_LUMA_W32 64, 64
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_pp(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+cglobal interp_8tap_horiz_pp_12x16, 4,6,8
+    add              r1d, r1d
+    add              r3d, r3d
+    sub              r0, 6
+    mov              r4d, r4m
+    shl              r4d, 4
+%ifdef PIC
+    lea              r5, [tab_LumaCoeff]
+    vpbroadcastq     m0, [r5 + r4]
+    vpbroadcastq     m1, [r5 + r4 + 8]
+%else
+    vpbroadcastq     m0, [tab_LumaCoeff + r4]
+    vpbroadcastq     m1, [tab_LumaCoeff + r4 + 8]
+%endif
+    mova             m3, [interp8_hpp_shuf]
+    mova             m7, [pd_32]
+    pxor             m2, m2
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    mov              r4d, 16
+
+.loop:
+    vbroadcasti128   m4, [r0]
+    vbroadcasti128   m5, [r0 + 8]
+    pshufb           m4, m3
+    pshufb           m5, m3
+
+    pmaddwd          m4, m0
+    pmaddwd          m5, m1
+    paddd            m4, m5
+
+    vbroadcasti128   m5, [r0 + 8]
+    vbroadcasti128   m6, [r0 + 16]
+    pshufb           m5, m3
+    pshufb           m6, m3
+
+    pmaddwd          m5, m0
+    pmaddwd          m6, m1
+    paddd            m5, m6
+
+    phaddd           m4, m5
+    vpermq           m4, m4, q3120
+    paddd            m4, m7
+    psrad            m4, INTERP_SHIFT_PP
+
+    packusdw         m4, m4
+    vpermq           m4, m4, q2020
+    CLIPW            m4, m2, [pw_pixel_max]
+    movu             [r2], xm4
+
+    vbroadcasti128   m4, [r0 + 16]
+    vbroadcasti128   m5, [r0 + 24]
+    pshufb           m4, m3
+    pshufb           m5, m3
+
+    pmaddwd          m4, m0
+    pmaddwd          m5, m1
+    paddd            m4, m5
+
+    vbroadcasti128   m5, [r0 + 24]
+    vbroadcasti128   m6, [r0 + 32]
+    pshufb           m5, m3
+    pshufb           m6, m3
+
+    pmaddwd          m5, m0
+    pmaddwd          m6, m1
+    paddd            m5, m6
+
+    phaddd           m4, m5
+    vpermq           m4, m4, q3120
+    paddd            m4, m7
+    psrad            m4, INTERP_SHIFT_PP
+
+    packusdw         m4, m4
+    vpermq           m4, m4, q2020
+    CLIPW            m4, m2, [pw_pixel_max]
+    movq             [r2 + 16], xm4
+
+    add              r2, r3
+    add              r0, r1
+    dec              r4d
+    jnz              .loop
+    RET
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_pp(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+cglobal interp_8tap_horiz_pp_24x32, 4,6,8
+    add              r1d, r1d
+    add              r3d, r3d
+    sub              r0, 6
+    mov              r4d, r4m
+    shl              r4d, 4
+%ifdef PIC
+    lea              r5, [tab_LumaCoeff]
+    vpbroadcastq     m0, [r5 + r4]
+    vpbroadcastq     m1, [r5 + r4 + 8]
+%else
+    vpbroadcastq     m0, [tab_LumaCoeff + r4]
+    vpbroadcastq     m1, [tab_LumaCoeff + r4 + 8]
+%endif
+    mova             m3, [interp8_hpp_shuf]
+    mova             m7, [pd_32]
+    pxor             m2, m2
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    mov              r4d, 32
+
+.loop:
+    vbroadcasti128   m4, [r0]
+    vbroadcasti128   m5, [r0 + 8]
+    pshufb           m4, m3
+    pshufb           m5, m3
+
+    pmaddwd          m4, m0
+    pmaddwd          m5, m1
+    paddd            m4, m5
+
+    vbroadcasti128   m5, [r0 + 8]
+    vbroadcasti128   m6, [r0 + 16]
+    pshufb           m5, m3
+    pshufb           m6, m3
+
+    pmaddwd          m5, m0
+    pmaddwd          m6, m1
+    paddd            m5, m6
+
+    phaddd           m4, m5
+    vpermq           m4, m4, q3120
+    paddd            m4, m7
+    psrad            m4, INTERP_SHIFT_PP
+
+    packusdw         m4, m4
+    vpermq           m4, m4, q2020
+    CLIPW            m4, m2, [pw_pixel_max]
+    movu             [r2], xm4
+
+    vbroadcasti128   m4, [r0 + 16]
+    vbroadcasti128   m5, [r0 + 24]
+    pshufb           m4, m3
+    pshufb           m5, m3
+
+    pmaddwd          m4, m0
+    pmaddwd          m5, m1
+    paddd            m4, m5
+
+    vbroadcasti128   m5, [r0 + 24]
+    vbroadcasti128   m6, [r0 + 32]
+    pshufb           m5, m3
+    pshufb           m6, m3
+
+    pmaddwd          m5, m0
+    pmaddwd          m6, m1
+    paddd            m5, m6
+
+    phaddd           m4, m5
+    vpermq           m4, m4, q3120
+    paddd            m4, m7
+    psrad            m4, INTERP_SHIFT_PP
+
+    packusdw         m4, m4
+    vpermq           m4, m4, q2020
+    CLIPW            m4, m2, [pw_pixel_max]
+    movu             [r2 + 16], xm4
+
+    vbroadcasti128   m4, [r0 + 32]
+    vbroadcasti128   m5, [r0 + 40]
+    pshufb           m4, m3
+    pshufb           m5, m3
+
+    pmaddwd          m4, m0
+    pmaddwd          m5, m1
+    paddd            m4, m5
+
+    vbroadcasti128   m5, [r0 + 40]
+    vbroadcasti128   m6, [r0 + 48]
+    pshufb           m5, m3
+    pshufb           m6, m3
+
+    pmaddwd          m5, m0
+    pmaddwd          m6, m1
+    paddd            m5, m6
+
+    phaddd           m4, m5
+    vpermq           m4, m4, q3120
+    paddd            m4, m7
+    psrad            m4, INTERP_SHIFT_PP
+
+    packusdw         m4, m4
+    vpermq           m4, m4, q2020
+    CLIPW            m4, m2, [pw_pixel_max]
+    movu             [r2 + 32], xm4
+
+    add              r2, r3
+    add              r0, r1
+    dec              r4d
+    jnz              .loop
+    RET
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_horiz_pp(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+cglobal interp_8tap_horiz_pp_48x64, 4,6,8
+    add              r1d, r1d
+    add              r3d, r3d
+    sub              r0, 6
+    mov              r4d, r4m
+    shl              r4d, 4
+%ifdef PIC
+    lea              r5, [tab_LumaCoeff]
+    vpbroadcastq     m0, [r5 + r4]
+    vpbroadcastq     m1, [r5 + r4 + 8]
+%else
+    vpbroadcastq     m0, [tab_LumaCoeff + r4]
+    vpbroadcastq     m1, [tab_LumaCoeff + r4 + 8]
+%endif
+    mova             m3, [interp8_hpp_shuf]
+    mova             m7, [pd_32]
+    pxor             m2, m2
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    mov              r4d, 64
+
+.loop:
+%assign x 0
+%rep 2
+    vbroadcasti128   m4, [r0 + x]
+    vbroadcasti128   m5, [r0 + 8 + x]
+    pshufb           m4, m3
+    pshufb           m5, m3
+
+    pmaddwd          m4, m0
+    pmaddwd          m5, m1
+    paddd            m4, m5
+
+    vbroadcasti128   m5, [r0 + 8 + x]
+    vbroadcasti128   m6, [r0 + 16 + x]
+    pshufb           m5, m3
+    pshufb           m6, m3
+
+    pmaddwd          m5, m0
+    pmaddwd          m6, m1
+    paddd            m5, m6
+
+    phaddd           m4, m5
+    vpermq           m4, m4, q3120
+    paddd            m4, m7
+    psrad            m4, INTERP_SHIFT_PP
+
+    packusdw         m4, m4
+    vpermq           m4, m4, q2020
+    CLIPW            m4, m2, [pw_pixel_max]
+    movu             [r2 + x], xm4
+
+    vbroadcasti128   m4, [r0 + 16 + x]
+    vbroadcasti128   m5, [r0 + 24 + x]
+    pshufb           m4, m3
+    pshufb           m5, m3
+
+    pmaddwd          m4, m0
+    pmaddwd          m5, m1
+    paddd            m4, m5
+
+    vbroadcasti128   m5, [r0 + 24 + x]
+    vbroadcasti128   m6, [r0 + 32 + x]
+    pshufb           m5, m3
+    pshufb           m6, m3
+
+    pmaddwd          m5, m0
+    pmaddwd          m6, m1
+    paddd            m5, m6
+
+    phaddd           m4, m5
+    vpermq           m4, m4, q3120
+    paddd            m4, m7
+    psrad            m4, INTERP_SHIFT_PP
+
+    packusdw         m4, m4
+    vpermq           m4, m4, q2020
+    CLIPW            m4, m2, [pw_pixel_max]
+    movu             [r2 + 16 + x], xm4
+
+    vbroadcasti128   m4, [r0 + 32 + x]
+    vbroadcasti128   m5, [r0 + 40 + x]
+    pshufb           m4, m3
+    pshufb           m5, m3
+
+    pmaddwd          m4, m0
+    pmaddwd          m5, m1
+    paddd            m4, m5
+
+    vbroadcasti128   m5, [r0 + 40 + x]
+    vbroadcasti128   m6, [r0 + 48 + x]
+    pshufb           m5, m3
+    pshufb           m6, m3
+
+    pmaddwd          m5, m0
+    pmaddwd          m6, m1
+    paddd            m5, m6
+
+    phaddd           m4, m5
+    vpermq           m4, m4, q3120
+    paddd            m4, m7
+    psrad            m4, INTERP_SHIFT_PP
+
+    packusdw         m4, m4
+    vpermq           m4, m4, q2020
+    CLIPW            m4, m2, [pw_pixel_max]
+    movu             [r2 + 32 + x], xm4
+
+%assign x x+48
+%endrep
+
+    add              r2, r3
+    add              r0, r1
+    dec              r4d
+    jnz              .loop
+    RET
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_horiz_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_CHROMA_H 6
+INIT_XMM sse4
+cglobal interp_4tap_horiz_%3_%1x%2, 4, %4, %5
+
+    add         r3,       r3
+    add         r1,       r1
+    sub         r0,       2
+    mov         r4d,      r4m
+    add         r4d,      r4d
+
+%ifdef PIC
+    lea         r%6,      [tab_ChromaCoeff]
+    movh        m0,       [r%6 + r4 * 4]
+%else
+    movh        m0,       [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    punpcklqdq  m0,       m0
+    mova        m2,       [tab_Tm16]
+
+%ifidn %3, ps
+    mova        m1,       [INTERP_OFFSET_PS]
+    cmp         r5m, byte 0
+    je          .skip
+    sub         r0,       r1
+    movu        m3,       [r0]
+    pshufb      m3,       m3, m2
+    pmaddwd     m3,       m0
+
+  %if %1 == 4
+    movu        m4,       [r0 + 4]
+    pshufb      m4,       m4, m2
+    pmaddwd     m4,       m0
+    phaddd      m3,       m4
+  %else
+    phaddd      m3,       m3
+  %endif
+
+    paddd       m3,       m1
+    psrad       m3,       INTERP_SHIFT_PS
+    packssdw    m3,       m3
+
+  %if %1 == 2
+    movd        [r2],     m3
+  %else
+    movh        [r2],     m3
+  %endif
+
+    add         r0,       r1
+    add         r2,       r3
+    FILTER_W%1_2 %3
+    lea         r0,       [r0 + 2 * r1]
+    lea         r2,       [r2 + 2 * r3]
+
+.skip:
+
+%else     ;%ifidn %3, ps
+    pxor        m7,       m7
+    mova        m6,       [pw_pixel_max]
+    mova        m1,       [tab_c_32]
+%endif    ;%ifidn %3, ps
+
+    FILTER_W%1_2 %3
+
+%rep (%2/2) - 1
+    lea         r0,       [r0 + 2 * r1]
+    lea         r2,       [r2 + 2 * r3]
+    FILTER_W%1_2 %3
+%endrep
+    RET
+%endmacro
+
+FILTER_CHROMA_H 2, 4, pp, 6, 8, 5
+FILTER_CHROMA_H 2, 8, pp, 6, 8, 5
+FILTER_CHROMA_H 4, 2, pp, 6, 8, 5
+FILTER_CHROMA_H 4, 4, pp, 6, 8, 5
+FILTER_CHROMA_H 4, 8, pp, 6, 8, 5
+FILTER_CHROMA_H 4, 16, pp, 6, 8, 5
+
+FILTER_CHROMA_H 2, 4, ps, 7, 5, 6
+FILTER_CHROMA_H 2, 8, ps, 7, 5, 6
+FILTER_CHROMA_H 4, 2, ps, 7, 6, 6
+FILTER_CHROMA_H 4, 4, ps, 7, 6, 6
+FILTER_CHROMA_H 4, 8, ps, 7, 6, 6
+FILTER_CHROMA_H 4, 16, ps, 7, 6, 6
+
+FILTER_CHROMA_H 2, 16, pp, 6, 8, 5
+FILTER_CHROMA_H 4, 32, pp, 6, 8, 5
+FILTER_CHROMA_H 2, 16, ps, 7, 5, 6
+FILTER_CHROMA_H 4, 32, ps, 7, 6, 6
+
+
+%macro FILTER_W6_1 1
+    movu        m3,         [r0]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 4]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m4,         [r0 + 8]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m4,         m4
+    paddd       m4,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m4,         INTERP_SHIFT_PP
+    packusdw    m3,         m4
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m4,         INTERP_SHIFT_PS
+    packssdw    m3,         m4
+%endif
+    movh        [r2],       m3
+    pextrd      [r2 + 8],   m3, 2
+%endmacro
+
+cglobal chroma_filter_pp_6x1_internal
+    FILTER_W6_1 pp
+    ret
+
+cglobal chroma_filter_ps_6x1_internal
+    FILTER_W6_1 ps
+    ret
+
+%macro FILTER_W8_1 1
+    movu        m3,         [r0]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 4]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + 8]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + 12]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2],       m3
+    movhps      [r2 + 8],   m3
+%endmacro
+
+cglobal chroma_filter_pp_8x1_internal
+    FILTER_W8_1 pp
+    ret
+
+cglobal chroma_filter_ps_8x1_internal
+    FILTER_W8_1 ps
+    ret
+
+%macro FILTER_W12_1 1
+    movu        m3,         [r0]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 4]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + 8]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + 12]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2],       m3
+    movhps      [r2 + 8],   m3
+
+    movu        m3,         [r0 + 16]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 20]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    packusdw    m3,         m3
+    CLIPW       m3,         m6, m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    packssdw    m3,         m3
+%endif
+    movh        [r2 + 16],  m3
+%endmacro
+
+cglobal chroma_filter_pp_12x1_internal
+    FILTER_W12_1 pp
+    ret
+
+cglobal chroma_filter_ps_12x1_internal
+    FILTER_W12_1 ps
+    ret
+
+%macro FILTER_W16_1 1
+    movu        m3,         [r0]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 4]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + 8]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + 12]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2],       m3
+    movhps      [r2 + 8],   m3
+
+    movu        m3,         [r0 + 16]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 20]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + 24]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + 28]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2 + 16],  m3
+    movhps      [r2 + 24],  m3
+%endmacro
+
+cglobal chroma_filter_pp_16x1_internal
+    FILTER_W16_1 pp
+    ret
+
+cglobal chroma_filter_ps_16x1_internal
+    FILTER_W16_1 ps
+    ret
+
+%macro FILTER_W24_1 1
+    movu        m3,         [r0]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 4]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + 8]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + 12]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2],       m3
+    movhps      [r2 + 8],   m3
+
+    movu        m3,         [r0 + 16]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 20]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + 24]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + 28]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2 + 16],  m3
+    movhps      [r2 + 24],  m3
+
+    movu        m3,         [r0 + 32]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 36]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + 40]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + 44]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2 + 32],  m3
+    movhps      [r2 + 40],  m3
+%endmacro
+
+cglobal chroma_filter_pp_24x1_internal
+    FILTER_W24_1 pp
+    ret
+
+cglobal chroma_filter_ps_24x1_internal
+    FILTER_W24_1 ps
+    ret
+
+%macro FILTER_W32_1 1
+    movu        m3,         [r0]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 4]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + 8]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + 12]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2],       m3
+    movhps      [r2 + 8],   m3
+
+    movu        m3,         [r0 + 16]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 20]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + 24]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + 28]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2 + 16],  m3
+    movhps      [r2 + 24],  m3
+
+    movu        m3,         [r0 + 32]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 36]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + 40]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + 44]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2 + 32],  m3
+    movhps      [r2 + 40],  m3
+
+    movu        m3,         [r0 + 48]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 52]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + 56]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + 60]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2 + 48],  m3
+    movhps      [r2 + 56],  m3
+%endmacro
+
+cglobal chroma_filter_pp_32x1_internal
+    FILTER_W32_1 pp
+    ret
+
+cglobal chroma_filter_ps_32x1_internal
+    FILTER_W32_1 ps
+    ret
+
+%macro FILTER_W8o_1 2
+    movu        m3,         [r0 + %2]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + %2 + 4]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + %2 + 8]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + %2 + 12]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2 + %2],       m3
+    movhps      [r2 + %2 + 8],   m3
+%endmacro
+
+%macro FILTER_W48_1 1
+    FILTER_W8o_1 %1, 0
+    FILTER_W8o_1 %1, 16
+    FILTER_W8o_1 %1, 32
+    FILTER_W8o_1 %1, 48
+    FILTER_W8o_1 %1, 64
+    FILTER_W8o_1 %1, 80
+%endmacro
+
+cglobal chroma_filter_pp_48x1_internal
+    FILTER_W48_1 pp
+    ret
+
+cglobal chroma_filter_ps_48x1_internal
+    FILTER_W48_1 ps
+    ret
+
+%macro FILTER_W64_1 1
+    FILTER_W8o_1 %1, 0
+    FILTER_W8o_1 %1, 16
+    FILTER_W8o_1 %1, 32
+    FILTER_W8o_1 %1, 48
+    FILTER_W8o_1 %1, 64
+    FILTER_W8o_1 %1, 80
+    FILTER_W8o_1 %1, 96
+    FILTER_W8o_1 %1, 112
+%endmacro
+
+cglobal chroma_filter_pp_64x1_internal
+    FILTER_W64_1 pp
+    ret
+
+cglobal chroma_filter_ps_64x1_internal
+    FILTER_W64_1 ps
+    ret
+    
+;-----------------------------------------------------------------------------
+; void interp_4tap_horiz_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+
+INIT_XMM sse4
+%macro IPFILTER_CHROMA 6
+cglobal interp_4tap_horiz_%3_%1x%2, 4, %5, %6
+
+    add         r3,        r3
+    add         r1,        r1
+    sub         r0,         2
+    mov         r4d,        r4m
+    add         r4d,        r4d
+
+%ifdef PIC
+    lea         r%4,       [tab_ChromaCoeff]
+    movh        m0,       [r%4 + r4 * 4]
+%else
+    movh        m0,       [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    punpcklqdq  m0,       m0
+    mova        m2,       [tab_Tm16]
+
+%ifidn %3, ps
+    mova        m1,       [INTERP_OFFSET_PS]
+    cmp         r5m, byte 0
+    je          .skip
+    sub         r0, r1
+    call chroma_filter_%3_%1x1_internal
+    add         r0, r1
+    add         r2, r3
+    call chroma_filter_%3_%1x1_internal
+    add         r0, r1
+    add         r2, r3
+    call chroma_filter_%3_%1x1_internal
+    add         r0, r1
+    add         r2, r3
+.skip:
+%else
+    mova        m1,         [tab_c_32]
+    pxor        m6,         m6
+    mova        m7,         [pw_pixel_max]
+%endif
+
+    call chroma_filter_%3_%1x1_internal
+%rep %2 - 1
+    add         r0,       r1
+    add         r2,       r3
+    call chroma_filter_%3_%1x1_internal
+%endrep
+RET
+%endmacro
+IPFILTER_CHROMA 6, 8, pp, 5, 6, 8
+IPFILTER_CHROMA 8, 2, pp, 5, 6, 8
+IPFILTER_CHROMA 8, 4, pp, 5, 6, 8
+IPFILTER_CHROMA 8, 6, pp, 5, 6, 8
+IPFILTER_CHROMA 8, 8, pp, 5, 6, 8
+IPFILTER_CHROMA 8, 16, pp, 5, 6, 8
+IPFILTER_CHROMA 8, 32, pp, 5, 6, 8
+IPFILTER_CHROMA 12, 16, pp, 5, 6, 8
+IPFILTER_CHROMA 16, 4, pp, 5, 6, 8
+IPFILTER_CHROMA 16, 8, pp, 5, 6, 8
+IPFILTER_CHROMA 16, 12, pp, 5, 6, 8
+IPFILTER_CHROMA 16, 16, pp, 5, 6, 8
+IPFILTER_CHROMA 16, 32, pp, 5, 6, 8
+IPFILTER_CHROMA 24, 32, pp, 5, 6, 8
+IPFILTER_CHROMA 32, 8, pp, 5, 6, 8
+IPFILTER_CHROMA 32, 16, pp, 5, 6, 8
+IPFILTER_CHROMA 32, 24, pp, 5, 6, 8
+IPFILTER_CHROMA 32, 32, pp, 5, 6, 8
+
+IPFILTER_CHROMA 6, 8, ps, 6, 7, 6
+IPFILTER_CHROMA 8, 2, ps, 6, 7, 6
+IPFILTER_CHROMA 8, 4, ps, 6, 7, 6
+IPFILTER_CHROMA 8, 6, ps, 6, 7, 6
+IPFILTER_CHROMA 8, 8, ps, 6, 7, 6
+IPFILTER_CHROMA 8, 16, ps, 6, 7, 6
+IPFILTER_CHROMA 8, 32, ps, 6, 7, 6
+IPFILTER_CHROMA 12, 16, ps, 6, 7, 6
+IPFILTER_CHROMA 16, 4, ps, 6, 7, 6
+IPFILTER_CHROMA 16, 8, ps, 6, 7, 6
+IPFILTER_CHROMA 16, 12, ps, 6, 7, 6
+IPFILTER_CHROMA 16, 16, ps, 6, 7, 6
+IPFILTER_CHROMA 16, 32, ps, 6, 7, 6
+IPFILTER_CHROMA 24, 32, ps, 6, 7, 6
+IPFILTER_CHROMA 32, 8, ps, 6, 7, 6
+IPFILTER_CHROMA 32, 16, ps, 6, 7, 6
+IPFILTER_CHROMA 32, 24, ps, 6, 7, 6
+IPFILTER_CHROMA 32, 32, ps, 6, 7, 6
+
+IPFILTER_CHROMA 6, 16, pp, 5, 6, 8
+IPFILTER_CHROMA 8, 12, pp, 5, 6, 8
+IPFILTER_CHROMA 8, 64, pp, 5, 6, 8
+IPFILTER_CHROMA 12, 32, pp, 5, 6, 8
+IPFILTER_CHROMA 16, 24, pp, 5, 6, 8
+IPFILTER_CHROMA 16, 64, pp, 5, 6, 8
+IPFILTER_CHROMA 24, 64, pp, 5, 6, 8
+IPFILTER_CHROMA 32, 48, pp, 5, 6, 8
+IPFILTER_CHROMA 32, 64, pp, 5, 6, 8
+IPFILTER_CHROMA 6, 16, ps, 6, 7, 6
+IPFILTER_CHROMA 8, 12, ps, 6, 7, 6
+IPFILTER_CHROMA 8, 64, ps, 6, 7, 6
+IPFILTER_CHROMA 12, 32, ps, 6, 7, 6
+IPFILTER_CHROMA 16, 24, ps, 6, 7, 6
+IPFILTER_CHROMA 16, 64, ps, 6, 7, 6
+IPFILTER_CHROMA 24, 64, ps, 6, 7, 6
+IPFILTER_CHROMA 32, 48, ps, 6, 7, 6
+IPFILTER_CHROMA 32, 64, ps, 6, 7, 6
+
+IPFILTER_CHROMA 48, 64, pp, 5, 6, 8
+IPFILTER_CHROMA 64, 48, pp, 5, 6, 8
+IPFILTER_CHROMA 64, 64, pp, 5, 6, 8
+IPFILTER_CHROMA 64, 32, pp, 5, 6, 8
+IPFILTER_CHROMA 64, 16, pp, 5, 6, 8
+IPFILTER_CHROMA 48, 64, ps, 6, 7, 6
+IPFILTER_CHROMA 64, 48, ps, 6, 7, 6
+IPFILTER_CHROMA 64, 64, ps, 6, 7, 6
+IPFILTER_CHROMA 64, 32, ps, 6, 7, 6
+IPFILTER_CHROMA 64, 16, ps, 6, 7, 6
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+%macro IPFILTER_CHROMA_avx2_6xN 1
+cglobal interp_4tap_horiz_pp_6x%1, 5,6,8
+    add             r1d, r1d
+    add             r3d, r3d
+    sub             r0, 2
+    mov             r4d, r4m
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova            m1, [interp8_hpp_shuf]
+    vpbroadcastd    m2, [pd_32]
+    pxor            m5, m5
+    mova            m6, [idct8_shuf2]
+    mova            m7, [pw_pixel_max]
+
+    mov             r4d, %1/2
+.loop:
+    vbroadcasti128  m3, [r0]
+    vbroadcasti128  m4, [r0 + 8]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, INTERP_SHIFT_PP           ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3, q2020
+    pshufb          xm3, xm6                      ; m3 = WORD[7 6 5 4 3 2 1 0]
+    CLIPW           xm3, xm5, xm7
+    movq            [r2], xm3
+    pextrd          [r2 + 8], xm3, 2
+
+    vbroadcasti128  m3, [r0 + r1]
+    vbroadcasti128  m4, [r0 + r1 + 8]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, INTERP_SHIFT_PP           ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3, q2020
+    pshufb          xm3, xm6                      ; m3 = WORD[7 6 5 4 3 2 1 0]
+    CLIPW           xm3, xm5, xm7
+    movq            [r2 + r3], xm3
+    pextrd          [r2 + r3 + 8], xm3, 2
+
+    lea             r0, [r0 + r1 * 2]
+    lea             r2, [r2 + r3 * 2]
+    dec             r4d
+    jnz             .loop
+    RET
+%endmacro
+IPFILTER_CHROMA_avx2_6xN 8
+IPFILTER_CHROMA_avx2_6xN 16
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_8x2, 5,6,8
+    add             r1d, r1d
+    add             r3d, r3d
+    sub             r0, 2
+    mov             r4d, r4m
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova            m1, [interp8_hpp_shuf]
+    vpbroadcastd    m2, [pd_32]
+    pxor            m5, m5
+    mova            m6, [idct8_shuf2]
+    mova            m7, [pw_pixel_max]
+
+    vbroadcasti128  m3, [r0]
+    vbroadcasti128  m4, [r0 + 8]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, INTERP_SHIFT_PP          ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3,q2020
+    pshufb          xm3, xm6                     ; m3 = WORD[7 6 5 4 3 2 1 0]
+    CLIPW           xm3, xm5, xm7
+    movu            [r2], xm3
+
+    vbroadcasti128  m3, [r0 + r1]
+    vbroadcasti128  m4, [r0 + r1 + 8]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, INTERP_SHIFT_PP           ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3,q2020
+    pshufb          xm3, xm6                      ; m3 = WORD[7 6 5 4 3 2 1 0]
+    CLIPW           xm3, xm5, xm7
+    movu            [r2 + r3], xm3
+    RET
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_8x4, 5,6,8
+    add             r1d, r1d
+    add             r3d, r3d
+    sub             r0, 2
+    mov             r4d, r4m
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova            m1, [interp8_hpp_shuf]
+    vpbroadcastd    m2, [pd_32]
+    pxor            m5, m5
+    mova            m6, [idct8_shuf2]
+    mova            m7, [pw_pixel_max]
+
+%rep 2
+    vbroadcasti128  m3, [r0]
+    vbroadcasti128  m4, [r0 + 8]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3,q2020
+    pshufb          xm3, xm6                    ; m3 = WORD[7 6 5 4 3 2 1 0]
+    CLIPW           xm3, xm5, xm7
+    movu            [r2], xm3
+
+    vbroadcasti128  m3, [r0 + r1]
+    vbroadcasti128  m4, [r0 + r1 + 8]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3,q2020
+    pshufb          xm3, xm6                    ; m3 = WORD[7 6 5 4 3 2 1 0]
+    CLIPW           xm3, xm5, xm7
+    movu            [r2 + r3], xm3
+
+    lea             r0, [r0 + r1 * 2]
+    lea             r2, [r2 + r3 * 2]
+%endrep
+    RET
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+%macro IPFILTER_CHROMA_avx2_8xN 1
+cglobal interp_4tap_horiz_pp_8x%1, 5,6,8
+    add             r1d, r1d
+    add             r3d, r3d
+    sub             r0, 2
+    mov             r4d, r4m
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova            m1, [interp8_hpp_shuf]
+    vpbroadcastd    m2, [pd_32]
+    pxor            m5, m5
+    mova            m6, [idct8_shuf2]
+    mova            m7, [pw_pixel_max]
+
+    mov             r4d, %1/2
+.loop:
+    vbroadcasti128  m3, [r0]
+    vbroadcasti128  m4, [r0 + 8]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6                         ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3, q2020
+    pshufb          xm3, xm6                      ; m3 = WORD[7 6 5 4 3 2 1 0]
+    CLIPW           xm3, xm5, xm7
+    movu            [r2], xm3
+
+    vbroadcasti128  m3, [r0 + r1]
+    vbroadcasti128  m4, [r0 + r1 + 8]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6                         ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3, q2020
+    pshufb          xm3, xm6                      ; m3 = WORD[7 6 5 4 3 2 1 0]
+    CLIPW           xm3, xm5, xm7
+    movu            [r2 + r3], xm3
+
+    lea             r0, [r0 + r1 * 2]
+    lea             r2, [r2 + r3 * 2]
+    dec             r4d
+    jnz             .loop
+    RET
+%endmacro
+IPFILTER_CHROMA_avx2_8xN 6
+IPFILTER_CHROMA_avx2_8xN 8
+IPFILTER_CHROMA_avx2_8xN 12
+IPFILTER_CHROMA_avx2_8xN 16
+IPFILTER_CHROMA_avx2_8xN 32
+IPFILTER_CHROMA_avx2_8xN 64
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+%macro IPFILTER_CHROMA_avx2_16xN 1
+%if ARCH_X86_64
+cglobal interp_4tap_horiz_pp_16x%1, 5,6,9
+    add             r1d, r1d
+    add             r3d, r3d
+    sub             r0, 2
+    mov             r4d, r4m
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova            m1, [interp8_hpp_shuf]
+    vpbroadcastd    m2, [pd_32]
+    pxor            m5, m5
+    mova            m6, [idct8_shuf2]
+    mova            m7, [pw_pixel_max]
+
+    mov             r4d, %1
+.loop:
+    vbroadcasti128  m3, [r0]
+    vbroadcasti128  m4, [r0 + 8]
+
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3, q2020
+    pshufb          xm3, xm6                     ; m3 = WORD[7 6 5 4 3 2 1 0]
+
+    vbroadcasti128  m4, [r0 + 16]
+    vbroadcasti128  m8, [r0 + 24]
+
+    pshufb          m4, m1
+    pshufb          m8, m1
+
+    pmaddwd         m4, m0
+    pmaddwd         m8, m0
+    phaddd          m4, m8
+    paddd           m4, m2
+    psrad           m4, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m4, m4
+    vpermq          m4, m4, q2020
+    pshufb          xm4, xm6                    ; m3 = WORD[7 6 5 4 3 2 1 0]
+    vinserti128     m3, m3, xm4, 1
+    CLIPW           m3, m5, m7
+    movu            [r2], m3
+
+    add             r0, r1
+    add             r2, r3
+    dec             r4d
+    jnz             .loop
+    RET
+%endif
+%endmacro
+IPFILTER_CHROMA_avx2_16xN 4
+IPFILTER_CHROMA_avx2_16xN 8
+IPFILTER_CHROMA_avx2_16xN 12
+IPFILTER_CHROMA_avx2_16xN 16
+IPFILTER_CHROMA_avx2_16xN 24
+IPFILTER_CHROMA_avx2_16xN 32
+IPFILTER_CHROMA_avx2_16xN 64
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+%macro IPFILTER_CHROMA_avx2_32xN 1
+%if ARCH_X86_64
+cglobal interp_4tap_horiz_pp_32x%1, 5,6,9
+    add             r1d, r1d
+    add             r3d, r3d
+    sub             r0, 2
+    mov             r4d, r4m
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova            m1, [interp8_hpp_shuf]
+    vpbroadcastd    m2, [pd_32]
+    pxor            m5, m5
+    mova            m6, [idct8_shuf2]
+    mova            m7, [pw_pixel_max]
+
+    mov             r6d, %1
+.loop:
+%assign x 0
+%rep 2
+    vbroadcasti128  m3, [r0 + x]
+    vbroadcasti128  m4, [r0 + 8 + x]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3, q2020
+    pshufb          xm3, xm6                     ; m3 = WORD[7 6 5 4 3 2 1 0]
+
+    vbroadcasti128  m4, [r0 + 16 + x]
+    vbroadcasti128  m8, [r0 + 24 + x]
+    pshufb          m4, m1
+    pshufb          m8, m1
+
+    pmaddwd         m4, m0
+    pmaddwd         m8, m0
+    phaddd          m4, m8
+    paddd           m4, m2
+    psrad           m4, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m4, m4
+    vpermq          m4, m4, q2020
+    pshufb          xm4, xm6                    ; m3 = WORD[7 6 5 4 3 2 1 0]
+    vinserti128     m3, m3, xm4, 1
+    CLIPW           m3, m5, m7
+    movu            [r2 + x], m3
+    %assign x x+32
+    %endrep
+
+    add             r0, r1
+    add             r2, r3
+    dec             r6d
+    jnz             .loop
+    RET
+%endif
+%endmacro
+IPFILTER_CHROMA_avx2_32xN 8
+IPFILTER_CHROMA_avx2_32xN 16
+IPFILTER_CHROMA_avx2_32xN 24
+IPFILTER_CHROMA_avx2_32xN 32
+IPFILTER_CHROMA_avx2_32xN 48
+IPFILTER_CHROMA_avx2_32xN 64
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+%macro IPFILTER_CHROMA_avx2_12xN 1
+%if ARCH_X86_64
+cglobal interp_4tap_horiz_pp_12x%1, 5,6,8
+    add             r1d, r1d
+    add             r3d, r3d
+    sub             r0, 2
+    mov             r4d, r4m
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova            m1, [interp8_hpp_shuf]
+    vpbroadcastd    m2, [pd_32]
+    pxor            m5, m5
+    mova            m6, [idct8_shuf2]
+    mova            m7, [pw_pixel_max]
+
+    mov             r4d, %1
+.loop:
+    vbroadcasti128  m3, [r0]
+    vbroadcasti128  m4, [r0 + 8]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3, q2020
+    pshufb          xm3, xm6                     ; m3 = WORD[7 6 5 4 3 2 1 0]
+    CLIPW           xm3, xm5, xm7
+    movu            [r2], xm3
+
+    vbroadcasti128  m3, [r0 + 16]
+    vbroadcasti128  m4, [r0 + 24]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3, q2020
+    pshufb          xm3, xm6                    ; m3 = WORD[7 6 5 4 3 2 1 0]
+    CLIPW           xm3, xm5, xm7
+    movq            [r2 + 16], xm3
+
+    add             r0, r1
+    add             r2, r3
+    dec             r4d
+    jnz             .loop
+    RET
+%endif
+%endmacro
+IPFILTER_CHROMA_avx2_12xN 16
+IPFILTER_CHROMA_avx2_12xN 32
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+%macro IPFILTER_CHROMA_avx2_24xN 1
+%if ARCH_X86_64
+cglobal interp_4tap_horiz_pp_24x%1, 5,6,9
+    add             r1d, r1d
+    add             r3d, r3d
+    sub             r0, 2
+    mov             r4d, r4m
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova            m1, [interp8_hpp_shuf]
+    vpbroadcastd    m2, [pd_32]
+    pxor            m5, m5
+    mova            m6, [idct8_shuf2]
+    mova            m7, [pw_pixel_max]
+
+    mov             r4d, %1
+.loop:
+    vbroadcasti128  m3, [r0]
+    vbroadcasti128  m4, [r0 + 8]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6
+
+    vbroadcasti128  m4, [r0 + 16]
+    vbroadcasti128  m8, [r0 + 24]
+    pshufb          m4, m1
+    pshufb          m8, m1
+
+    pmaddwd         m4, m0
+    pmaddwd         m8, m0
+    phaddd          m4, m8
+    paddd           m4, m2
+    psrad           m4, 6
+
+    packusdw        m3, m4
+    vpermq          m3, m3, q3120
+    pshufb          m3, m6
+    CLIPW           m3, m5, m7
+    movu            [r2], m3
+
+    vbroadcasti128  m3, [r0 + 32]
+    vbroadcasti128  m4, [r0 + 40]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6
+
+    packusdw        m3, m3
+    vpermq          m3, m3, q2020
+    pshufb          xm3, xm6
+    CLIPW           xm3, xm5, xm7
+    movu            [r2 + 32], xm3
+
+    add             r0, r1
+    add             r2, r3
+    dec             r4d
+    jnz             .loop
+    RET
+%endif
+%endmacro
+IPFILTER_CHROMA_avx2_24xN 32
+IPFILTER_CHROMA_avx2_24xN 64
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+%macro IPFILTER_CHROMA_avx2_64xN 1
+%if ARCH_X86_64
+cglobal interp_4tap_horiz_pp_64x%1, 5,6,9
+    add             r1d, r1d
+    add             r3d, r3d
+    sub             r0, 2
+    mov             r4d, r4m
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova            m1, [interp8_hpp_shuf]
+    vpbroadcastd    m2, [pd_32]
+    pxor            m5, m5
+    mova            m6, [idct8_shuf2]
+    mova            m7, [pw_pixel_max]
+
+    mov             r6d, %1
+.loop:
+%assign x 0
+%rep 4
+    vbroadcasti128  m3, [r0 + x]
+    vbroadcasti128  m4, [r0 + 8 + x]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6
+
+    vbroadcasti128  m4, [r0 + 16 + x]
+    vbroadcasti128  m8, [r0 + 24 + x]
+    pshufb          m4, m1
+    pshufb          m8, m1
+
+    pmaddwd         m4, m0
+    pmaddwd         m8, m0
+    phaddd          m4, m8
+    paddd           m4, m2
+    psrad           m4, 6
+
+    packusdw        m3, m4
+    vpermq          m3, m3, q3120
+    pshufb          m3, m6
+    CLIPW           m3, m5, m7
+    movu            [r2 + x], m3
+    %assign x x+32
+    %endrep
+
+    add             r0, r1
+    add             r2, r3
+    dec             r6d
+    jnz             .loop
+    RET
+%endif
+%endmacro
+IPFILTER_CHROMA_avx2_64xN 16
+IPFILTER_CHROMA_avx2_64xN 32
+IPFILTER_CHROMA_avx2_64xN 48
+IPFILTER_CHROMA_avx2_64xN 64
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+%if ARCH_X86_64
+cglobal interp_4tap_horiz_pp_48x64, 5,6,9
+    add             r1d, r1d
+    add             r3d, r3d
+    sub             r0, 2
+    mov             r4d, r4m
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova            m1, [interp8_hpp_shuf]
+    vpbroadcastd    m2, [pd_32]
+    pxor            m5, m5
+    mova            m6, [idct8_shuf2]
+    mova            m7, [pw_pixel_max]
+
+    mov             r4d, 64
+.loop:
+%assign x 0
+%rep 3
+    vbroadcasti128  m3, [r0 + x]
+    vbroadcasti128  m4, [r0 + 8 + x]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6
+
+    vbroadcasti128  m4, [r0 + 16 + x]
+    vbroadcasti128  m8, [r0 + 24 + x]
+    pshufb          m4, m1
+    pshufb          m8, m1
+
+    pmaddwd         m4, m0
+    pmaddwd         m8, m0
+    phaddd          m4, m8
+    paddd           m4, m2
+    psrad           m4, 6
+
+    packusdw        m3, m4
+    vpermq          m3, m3, q3120
+    pshufb          m3, m6
+    CLIPW           m3, m5, m7
+    movu            [r2 + x], m3
+%assign x x+32
+%endrep
+
+    add             r0, r1
+    add             r2, r3
+    dec             r4d
+    jnz             .loop
+    RET
+%endif
+
+;-----------------------------------------------------------------------------------------------------------------------------
+;void interp_horiz_ps_c(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;-----------------------------------------------------------------------------------------------------------------------------
+
+%macro IPFILTER_LUMA_PS_4xN_AVX2 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_horiz_ps_4x%1, 6,8,7
+    mov                         r5d,               r5m
+    mov                         r4d,               r4m
+    add                         r1d,               r1d
+    add                         r3d,               r3d
+
+%ifdef PIC
+    lea                         r6,                [tab_LumaCoeff]
+    lea                         r4,                [r4 * 8]
+    vbroadcasti128              m0,                [r6 + r4 * 2]
+%else
+    lea                         r4,                [r4 * 8]
+    vbroadcasti128              m0,                [tab_LumaCoeff + r4 * 2]
+%endif
+
+    vbroadcasti128              m2,                [INTERP_OFFSET_PS]
+
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - pw_2000
+
+    sub                         r0,                6
+    test                        r5d,               r5d
+    mov                         r7d,               %1                                    ; loop count variable - height
+    jz                         .preloop
+    lea                         r6,                [r1 * 3]                              ; r6 = (N / 2 - 1) * srcStride
+    sub                         r0,                r6                                    ; r0(src) - 3 * srcStride
+    add                         r7d,               6                                     ;7 - 1(since last row not in loop)                            ; need extra 7 rows, just set a specially flag here, blkheight += N - 1  (7 - 3 = 4 ; since the last three rows not in loop)
+
+.preloop:
+    lea                         r6,                [r3 * 3]
+.loop:
+    ; Row 0
+    movu                        xm3,                [r0]                                 ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    movu                        xm4,                [r0 + 2]                             ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    vinserti128                 m3,                 m3,                xm4,       1
+    movu                        xm4,                [r0 + 4]
+    movu                        xm5,                [r0 + 6]
+    vinserti128                 m4,                 m4,                xm5,       1
+    pmaddwd                     m3,                m0
+    pmaddwd                     m4,                m0
+    phaddd                      m3,                m4                                    ; DWORD [R1D R1C R0D R0C R1B R1A R0B R0A]
+
+    ; Row 1
+    movu                        xm4,                [r0 + r1]                            ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    movu                        xm5,                [r0 + r1 + 2]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    vinserti128                 m4,                 m4,                xm5,       1
+    movu                        xm5,                [r0 + r1 + 4]
+    movu                        xm6,                [r0 + r1 + 6]
+    vinserti128                 m5,                 m5,                xm6,       1
+    pmaddwd                     m4,                m0
+    pmaddwd                     m5,                m0
+    phaddd                      m4,                m5                                     ; DWORD [R3D R3C R2D R2C R3B R3A R2B R2A]
+    phaddd                      m3,                m4                                     ; all rows and col completed.
+
+    mova                        m5,                [interp8_hps_shuf]
+    vpermd                      m3,                m5,                  m3
+    paddd                       m3,                m2
+    vextracti128                xm4,               m3,                  1
+    psrad                       xm3,               INTERP_SHIFT_PS
+    psrad                       xm4,               INTERP_SHIFT_PS
+    packssdw                    xm3,               xm3
+    packssdw                    xm4,               xm4
+
+    movq                        [r2],              xm3                                   ;row 0
+    movq                        [r2 + r3],         xm4                                   ;row 1
+    lea                         r0,                [r0 + r1 * 2]                         ; first loop src ->5th row(i.e 4)
+    lea                         r2,                [r2 + r3 * 2]                         ; first loop dst ->5th row(i.e 4)
+
+    sub                         r7d,               2
+    jg                          .loop
+    test                        r5d,               r5d
+    jz                          .end
+
+    ; Row 10
+    movu                        xm3,                [r0]
+    movu                        xm4,                [r0 + 2]
+    vinserti128                 m3,                 m3,                 xm4,      1
+    movu                        xm4,                [r0 + 4]
+    movu                        xm5,                [r0 + 6]
+    vinserti128                 m4,                 m4,                 xm5,      1
+    pmaddwd                     m3,                m0
+    pmaddwd                     m4,                m0
+    phaddd                      m3,                m4
+
+    ; Row11
+    phaddd                      m3,                m4                                    ; all rows and col completed.
+
+    mova                        m5,                [interp8_hps_shuf]
+    vpermd                      m3,                m5,                  m3
+    paddd                       m3,                m2
+    vextracti128                xm4,               m3,                  1
+    psrad                       xm3,               INTERP_SHIFT_PS
+    psrad                       xm4,               INTERP_SHIFT_PS
+    packssdw                    xm3,               xm3
+    packssdw                    xm4,               xm4
+
+    movq                        [r2],              xm3                                   ;row 0
+.end:
+    RET
+%endif
+%endmacro
+
+    IPFILTER_LUMA_PS_4xN_AVX2 4
+    IPFILTER_LUMA_PS_4xN_AVX2 8
+    IPFILTER_LUMA_PS_4xN_AVX2 16
+
+%macro IPFILTER_LUMA_PS_8xN_AVX2 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_horiz_ps_8x%1, 4, 6, 8
+    add                 r1d, r1d
+    add                 r3d, r3d
+    mov                 r4d, r4m
+    mov                 r5d, r5m
+    shl                 r4d, 4
+%ifdef PIC
+    lea                 r6, [tab_LumaCoeff]
+    vpbroadcastq        m0, [r6 + r4]
+    vpbroadcastq        m1, [r6 + r4 + 8]
+%else
+    vpbroadcastq        m0, [tab_LumaCoeff + r4]
+    vpbroadcastq        m1, [tab_LumaCoeff + r4 + 8]
+%endif
+    mova                m3, [interp8_hpp_shuf]
+    vbroadcasti128      m2, [INTERP_OFFSET_PS]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    sub                 r0, 6
+    test                r5d, r5d
+    mov                 r4d, %1
+    jz                  .loop0
+    lea                 r6, [r1*3]
+    sub                 r0, r6
+    add                 r4d, 7
+
+.loop0:
+    vbroadcasti128      m4, [r0]
+    vbroadcasti128      m5, [r0 + 8]
+    pshufb              m4, m3
+    pshufb              m7, m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m7, m1
+    paddd               m4, m7
+
+    vbroadcasti128      m6, [r0 + 16]
+    pshufb              m5, m3
+    pshufb              m6, m3
+    pmaddwd             m5, m0
+    pmaddwd             m6, m1
+    paddd               m5, m6
+
+    phaddd              m4, m5
+    vpermq              m4, m4, q3120
+    paddd               m4, m2
+    vextracti128        xm5,m4, 1
+    psrad               xm4, INTERP_SHIFT_PS
+    psrad               xm5, INTERP_SHIFT_PS
+    packssdw            xm4, xm5
+
+    movu                [r2], xm4
+    add                 r2, r3
+    add                 r0, r1
+    dec                 r4d
+    jnz                 .loop0
+    RET
+%endif
+%endmacro
+
+    IPFILTER_LUMA_PS_8xN_AVX2 4
+    IPFILTER_LUMA_PS_8xN_AVX2 8
+    IPFILTER_LUMA_PS_8xN_AVX2 16
+    IPFILTER_LUMA_PS_8xN_AVX2 32
+
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_horiz_ps_24x32, 4, 6, 8
+    add                 r1d, r1d
+    add                 r3d, r3d
+    mov                 r4d, r4m
+    mov                 r5d, r5m
+    shl                 r4d, 4
+%ifdef PIC
+    lea                 r6, [tab_LumaCoeff]
+    vpbroadcastq        m0, [r6 + r4]
+    vpbroadcastq        m1, [r6 + r4 + 8]
+%else
+    vpbroadcastq        m0, [tab_LumaCoeff + r4]
+    vpbroadcastq        m1, [tab_LumaCoeff + r4 + 8]
+%endif
+    mova                m3, [interp8_hpp_shuf]
+    vbroadcasti128      m2, [INTERP_OFFSET_PS]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    sub                 r0, 6
+    test                r5d, r5d
+    mov                 r4d, 32
+    jz                  .loop0
+    lea                 r6, [r1*3]
+    sub                 r0, r6
+    add                 r4d, 7
+
+.loop0:
+%assign x 0
+%rep 24/8
+    vbroadcasti128      m4, [r0 + x]
+    vbroadcasti128      m5, [r0 + 8 + x]
+    pshufb              m4, m3
+    pshufb              m7, m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m7, m1
+    paddd               m4, m7
+
+    vbroadcasti128      m6, [r0 + 16 + x]
+    pshufb              m5, m3
+    pshufb              m6, m3
+    pmaddwd             m5, m0
+    pmaddwd             m6, m1
+    paddd               m5, m6
+
+    phaddd              m4, m5
+    vpermq              m4, m4, q3120
+    paddd               m4, m2
+    vextracti128        xm5,m4, 1
+    psrad               xm4, INTERP_SHIFT_PS
+    psrad               xm5, INTERP_SHIFT_PS
+    packssdw            xm4, xm5
+
+    movu                [r2 + x], xm4
+    %assign x x+16
+    %endrep
+
+    add                 r2, r3
+    add                 r0, r1
+    dec                 r4d
+    jnz                 .loop0
+    RET
+%endif
+%macro IPFILTER_LUMA_PS_32_64_AVX2 2
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_horiz_ps_%1x%2, 4, 6, 8
+
+    add                 r1d, r1d
+    add                 r3d, r3d
+    mov                 r4d, r4m
+    mov                 r5d, r5m
+    shl                 r4d, 6
+%ifdef PIC
+    lea                 r6, [h_tab_LumaCoeffV]
+    movu                m0, [r6 + r4]
+    movu                m1, [r6 + r4 + mmsize]
+%else
+    movu                m0, [h_tab_LumaCoeffV + r4]
+    movu                m1, [h_tab_LumaCoeffV + r4 + mmsize]
+%endif
+    mova                m3, [interp8_hpp_shuf_new]
+    vbroadcasti128      m2, [INTERP_OFFSET_PS]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    sub                 r0, 6
+    test                r5d, r5d
+    mov                 r4d, %2
+    jz                 .loop0
+    lea                 r6, [r1*3]
+    sub                 r0, r6
+    add                 r4d, 7
+
+.loop0:
+%assign x 0
+%rep %1/16
+    vbroadcasti128      m4, [r0 + x]
+    vbroadcasti128      m5, [r0 + 4 * SIZEOF_PIXEL + x]
+    pshufb              m4, m3
+    pshufb              m5, m3
+
+    pmaddwd             m4, m0
+    pmaddwd             m7, m5, m1
+    paddd               m4, m7
+    vextracti128        xm7, m4, 1
+    paddd               xm4, xm7
+    paddd               xm4, xm2
+    psrad               xm4, INTERP_SHIFT_PS
+
+    vbroadcasti128      m6, [r0 + 16 + x]
+    pshufb              m6, m3
+
+    pmaddwd             m5, m0
+    pmaddwd             m7, m6, m1
+    paddd               m5, m7
+    vextracti128        xm7, m5, 1
+    paddd               xm5, xm7
+    paddd               xm5, xm2
+    psrad               xm5, INTERP_SHIFT_PS
+
+    packssdw            xm4, xm5
+    movu                [r2 + x], xm4
+
+    vbroadcasti128      m5, [r0 + 24 + x]
+    pshufb              m5, m3
+
+    pmaddwd             m6, m0
+    pmaddwd             m7, m5, m1
+    paddd               m6, m7
+    vextracti128        xm7, m6, 1
+    paddd               xm6, xm7
+    paddd               xm6, xm2
+    psrad               xm6, INTERP_SHIFT_PS
+
+    vbroadcasti128      m7, [r0 + 32 + x]
+    pshufb              m7, m3
+
+    pmaddwd             m5, m0
+    pmaddwd             m7, m1
+    paddd               m5, m7
+    vextracti128        xm7, m5, 1
+    paddd               xm5, xm7
+    paddd               xm5, xm2
+    psrad               xm5, INTERP_SHIFT_PS
+
+    packssdw            xm6, xm5
+    movu                [r2 + 16 + x], xm6
+
+%assign x x+32
+%endrep
+
+    add                 r2, r3
+    add                 r0, r1
+    dec                 r4d
+    jnz                .loop0
+    RET
+%endif
+%endmacro
+
+    IPFILTER_LUMA_PS_32_64_AVX2 32, 8
+    IPFILTER_LUMA_PS_32_64_AVX2 32, 16
+    IPFILTER_LUMA_PS_32_64_AVX2 32, 24
+    IPFILTER_LUMA_PS_32_64_AVX2 32, 32
+    IPFILTER_LUMA_PS_32_64_AVX2 32, 64
+
+    IPFILTER_LUMA_PS_32_64_AVX2 64, 16
+    IPFILTER_LUMA_PS_32_64_AVX2 64, 32
+    IPFILTER_LUMA_PS_32_64_AVX2 64, 48
+    IPFILTER_LUMA_PS_32_64_AVX2 64, 64
+
+    IPFILTER_LUMA_PS_32_64_AVX2 48, 64
+
+%macro IPFILTER_LUMA_PS_16xN_AVX2 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_horiz_ps_16x%1, 4, 6, 8
+
+    add                 r1d, r1d
+    add                 r3d, r3d
+    mov                 r4d, r4m
+    mov                 r5d, r5m
+    shl                 r4d, 4
+%ifdef PIC
+    lea                 r6, [tab_LumaCoeff]
+    vpbroadcastq        m0, [r6 + r4]
+    vpbroadcastq        m1, [r6 + r4 + 8]
+%else
+    vpbroadcastq        m0, [tab_LumaCoeff + r4]
+    vpbroadcastq        m1, [tab_LumaCoeff + r4 + 8]
+%endif
+    mova                m3, [interp8_hpp_shuf]
+    vbroadcasti128      m2, [INTERP_OFFSET_PS]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    sub                 r0, 6
+    test                r5d, r5d
+    mov                 r4d, %1
+    jz                  .loop0
+    lea                 r6, [r1*3]
+    sub                 r0, r6
+    add                 r4d, 7
+
+.loop0:
+    vbroadcasti128      m4, [r0]
+    vbroadcasti128      m5, [r0 + 8]
+    pshufb              m4, m3
+    pshufb              m7, m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m7, m1
+    paddd               m4, m7
+
+    vbroadcasti128      m6, [r0 + 16]
+    pshufb              m5, m3
+    pshufb              m7, m6, m3
+    pmaddwd             m5, m0
+    pmaddwd             m7, m1
+    paddd               m5, m7
+
+    phaddd              m4, m5
+    vpermq              m4, m4, q3120
+    paddd               m4, m2
+    vextracti128        xm5, m4, 1
+    psrad               xm4, INTERP_SHIFT_PS
+    psrad               xm5, INTERP_SHIFT_PS
+    packssdw            xm4, xm5
+    movu                [r2], xm4
+
+    vbroadcasti128      m5, [r0 + 24]
+    pshufb              m6, m3
+    pshufb              m7, m5, m3
+    pmaddwd             m6, m0
+    pmaddwd             m7, m1
+    paddd               m6, m7
+
+    vbroadcasti128      m7, [r0 + 32]
+    pshufb              m5, m3
+    pshufb              m7, m3
+    pmaddwd             m5, m0
+    pmaddwd             m7, m1
+    paddd               m5, m7
+
+    phaddd              m6, m5
+    vpermq              m6, m6, q3120
+    paddd               m6, m2
+    vextracti128        xm5,m6, 1
+    psrad               xm6, INTERP_SHIFT_PS
+    psrad               xm5, INTERP_SHIFT_PS
+    packssdw            xm6, xm5
+    movu                [r2 + 16], xm6
+
+    add                 r2, r3
+    add                 r0, r1
+    dec                 r4d
+    jnz                 .loop0
+    RET
+%endif
+%endmacro
+
+    IPFILTER_LUMA_PS_16xN_AVX2 4
+    IPFILTER_LUMA_PS_16xN_AVX2 8
+    IPFILTER_LUMA_PS_16xN_AVX2 12
+    IPFILTER_LUMA_PS_16xN_AVX2 16
+    IPFILTER_LUMA_PS_16xN_AVX2 32
+    IPFILTER_LUMA_PS_16xN_AVX2 64
+
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_horiz_ps_12x16, 4, 6, 8
+    add                 r1d, r1d
+    add                 r3d, r3d
+    mov                 r4d, r4m
+    mov                 r5d, r5m
+    shl                 r4d, 4
+%ifdef PIC
+    lea                 r6, [tab_LumaCoeff]
+    vpbroadcastq        m0, [r6 + r4]
+    vpbroadcastq        m1, [r6 + r4 + 8]
+%else
+    vpbroadcastq        m0, [tab_LumaCoeff + r4]
+    vpbroadcastq        m1, [tab_LumaCoeff + r4 + 8]
+%endif
+    mova                m3, [interp8_hpp_shuf]
+    vbroadcasti128      m2, [INTERP_OFFSET_PS]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    sub                 r0, 6
+    test                r5d, r5d
+    mov                 r4d, 16
+    jz                  .loop0
+    lea                 r6, [r1*3]
+    sub                 r0, r6
+    add                 r4d, 7
+
+.loop0:
+    vbroadcasti128      m4, [r0]
+    vbroadcasti128      m5, [r0 + 8]
+    pshufb              m4, m3
+    pshufb              m7, m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m7, m1
+    paddd               m4, m7
+
+    vbroadcasti128      m6, [r0 + 16]
+    pshufb              m5, m3
+    pshufb              m7, m6, m3
+    pmaddwd             m5, m0
+    pmaddwd             m7, m1
+    paddd               m5, m7
+
+    phaddd              m4, m5
+    vpermq              m4, m4, q3120
+    paddd               m4, m2
+    vextracti128        xm5,m4, 1
+    psrad               xm4, INTERP_SHIFT_PS
+    psrad               xm5, INTERP_SHIFT_PS
+    packssdw            xm4, xm5
+    movu                [r2], xm4
+
+    vbroadcasti128      m5, [r0 + 24]
+    pshufb              m6, m3
+    pshufb              m5, m3
+    pmaddwd             m6, m0
+    pmaddwd             m5, m1
+    paddd               m6, m5
+
+    phaddd              m6, m6
+    vpermq              m6, m6, q3120
+    paddd               xm6, xm2
+    psrad               xm6, INTERP_SHIFT_PS
+    packssdw            xm6, xm6
+    movq                [r2 + 16], xm6
+
+    add                 r2, r3
+    add                 r0, r1
+    dec                 r4d
+    jnz                 .loop0
+    RET
+%endif
+%macro IPFILTER_CHROMA_PS_8xN_AVX2 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_horiz_ps_8x%1, 4, 7, 6
+    add                 r1d, r1d
+    add                 r3d, r3d
+    mov                 r4d, r4m
+    mov                 r5d, r5m
+
+%ifdef PIC
+    lea                 r6, [tab_ChromaCoeff]
+    vpbroadcastq        m0, [r6 + r4 * 8]
+%else
+    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova                m3, [interp8_hpp_shuf]
+    vbroadcasti128      m2, [INTERP_OFFSET_PS]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    sub                 r0, 2
+    test                r5d, r5d
+    mov                 r4d, %1
+    jz                  .loop0
+    sub                 r0, r1
+    add                 r4d, 3
+
+.loop0:
+    vbroadcasti128      m4, [r0]
+    vbroadcasti128      m5, [r0 + 8]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2], xm4
+    add                 r2, r3
+    add                 r0, r1
+    dec                 r4d
+    jnz                 .loop0
+    RET
+%endif
+%endmacro
+
+    IPFILTER_CHROMA_PS_8xN_AVX2 4
+    IPFILTER_CHROMA_PS_8xN_AVX2 8
+    IPFILTER_CHROMA_PS_8xN_AVX2 16
+    IPFILTER_CHROMA_PS_8xN_AVX2 32
+    IPFILTER_CHROMA_PS_8xN_AVX2 6
+    IPFILTER_CHROMA_PS_8xN_AVX2 2
+    IPFILTER_CHROMA_PS_8xN_AVX2 12
+    IPFILTER_CHROMA_PS_8xN_AVX2 64
+
+%macro IPFILTER_CHROMA_PS_16xN_AVX2 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_horiz_ps_16x%1, 4, 7, 6
+    add                 r1d, r1d
+    add                 r3d, r3d
+    mov                 r4d, r4m
+    mov                 r5d, r5m
+
+%ifdef PIC
+    lea                 r6, [tab_ChromaCoeff]
+    vpbroadcastq        m0, [r6 + r4 * 8]
+%else
+    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova                m3, [interp8_hpp_shuf]
+    vbroadcasti128      m2, [INTERP_OFFSET_PS]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    sub                 r0, 2
+    test                r5d, r5d
+    mov                 r4d, %1
+    jz                  .loop0
+    sub                 r0, r1
+    add                 r4d, 3
+
+.loop0:
+    vbroadcasti128      m4, [r0]
+    vbroadcasti128      m5, [r0 + 8]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2], xm4
+
+    vbroadcasti128      m4, [r0 + 16]
+    vbroadcasti128      m5, [r0 + 24]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 16], xm4
+
+    add                 r2, r3
+    add                 r0, r1
+    dec                 r4d
+    jnz                 .loop0
+    RET
+%endif
+%endmacro
+
+IPFILTER_CHROMA_PS_16xN_AVX2 16
+IPFILTER_CHROMA_PS_16xN_AVX2 8
+IPFILTER_CHROMA_PS_16xN_AVX2 32
+IPFILTER_CHROMA_PS_16xN_AVX2 12
+IPFILTER_CHROMA_PS_16xN_AVX2 4
+IPFILTER_CHROMA_PS_16xN_AVX2 64
+IPFILTER_CHROMA_PS_16xN_AVX2 24
+
+%macro IPFILTER_CHROMA_PS_24xN_AVX2 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_horiz_ps_24x%1, 4, 7, 6
+    add                 r1d, r1d
+    add                 r3d, r3d
+    mov                 r4d, r4m
+    mov                 r5d, r5m
+
+%ifdef PIC
+    lea                 r6, [tab_ChromaCoeff]
+    vpbroadcastq        m0, [r6 + r4 * 8]
+%else
+    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova                m3, [interp8_hpp_shuf]
+    vbroadcasti128      m2, [INTERP_OFFSET_PS]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    sub                 r0, 2
+    test                r5d, r5d
+    mov                 r4d, %1
+    jz                  .loop0
+    sub                 r0, r1
+    add                 r4d, 3
+
+.loop0:
+    vbroadcasti128      m4, [r0]
+    vbroadcasti128      m5, [r0 + 8]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2], xm4
+
+    vbroadcasti128      m4, [r0 + 16]
+    vbroadcasti128      m5, [r0 + 24]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 16], xm4
+
+    vbroadcasti128      m4, [r0 + 32]
+    vbroadcasti128      m5, [r0 + 40]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 32], xm4
+
+    add                 r2, r3
+    add                 r0, r1
+    dec                 r4d
+    jnz                 .loop0
+    RET
+%endif
+%endmacro
+
+IPFILTER_CHROMA_PS_24xN_AVX2 32
+IPFILTER_CHROMA_PS_24xN_AVX2 64
+
+%macro IPFILTER_CHROMA_PS_12xN_AVX2 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_horiz_ps_12x%1, 4, 7, 6
+    add                 r1d, r1d
+    add                 r3d, r3d
+    mov                 r4d, r4m
+    mov                 r5d, r5m
+
+%ifdef PIC
+    lea                 r6, [tab_ChromaCoeff]
+    vpbroadcastq        m0, [r6 + r4 * 8]
+%else
+    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova                m3, [interp8_hpp_shuf]
+    vbroadcasti128      m2, [INTERP_OFFSET_PS]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    sub                 r0, 2
+    test                r5d, r5d
+    mov                 r4d, %1
+    jz                  .loop0
+    sub                 r0, r1
+    add                 r4d, 3
+
+.loop0:
+    vbroadcasti128      m4, [r0]
+    vbroadcasti128      m5, [r0 + 8]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2], xm4
+
+    vbroadcasti128      m4, [r0 + 16]
+    pshufb              m4, m3
+    pmaddwd             m4, m0
+    phaddd              m4, m4
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movq                [r2 + 16], xm4
+
+    add                 r2, r3
+    add                 r0, r1
+    dec                 r4d
+    jnz                 .loop0
+    RET
+%endif
+%endmacro
+
+IPFILTER_CHROMA_PS_12xN_AVX2 16
+IPFILTER_CHROMA_PS_12xN_AVX2 32
+
+%macro IPFILTER_CHROMA_PS_32xN_AVX2 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_horiz_ps_32x%1, 4, 7, 6
+    add                 r1d, r1d
+    add                 r3d, r3d
+    mov                 r4d, r4m
+    mov                 r5d, r5m
+
+%ifdef PIC
+    lea                 r6, [tab_ChromaCoeff]
+    vpbroadcastq        m0, [r6 + r4 * 8]
+%else
+    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova                m3, [interp8_hpp_shuf]
+    vbroadcasti128      m2, [INTERP_OFFSET_PS]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    sub                 r0, 2
+    test                r5d, r5d
+    mov                 r4d, %1
+    jz                  .loop0
+    sub                 r0, r1
+    add                 r4d, 3
+
+.loop0:
+    vbroadcasti128      m4, [r0]
+    vbroadcasti128      m5, [r0 + 8]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2], xm4
+
+    vbroadcasti128      m4, [r0 + 16]
+    vbroadcasti128      m5, [r0 + 24]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 16], xm4
+
+    vbroadcasti128      m4, [r0 + 32]
+    vbroadcasti128      m5, [r0 + 40]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 32], xm4
+
+    vbroadcasti128      m4, [r0 + 48]
+    vbroadcasti128      m5, [r0 + 56]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 48], xm4
+
+    add                 r2, r3
+    add                 r0, r1
+    dec                 r4d
+    jnz                 .loop0
+    RET
+%endif
+%endmacro
+
+IPFILTER_CHROMA_PS_32xN_AVX2 32
+IPFILTER_CHROMA_PS_32xN_AVX2 16
+IPFILTER_CHROMA_PS_32xN_AVX2 24
+IPFILTER_CHROMA_PS_32xN_AVX2 8
+IPFILTER_CHROMA_PS_32xN_AVX2 64
+IPFILTER_CHROMA_PS_32xN_AVX2 48
+
+
+%macro IPFILTER_CHROMA_PS_64xN_AVX2 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_horiz_ps_64x%1, 4, 7, 6
+    add                 r1d, r1d
+    add                 r3d, r3d
+    mov                 r4d, r4m
+    mov                 r5d, r5m
+
+%ifdef PIC
+    lea                 r6, [tab_ChromaCoeff]
+    vpbroadcastq        m0, [r6 + r4 * 8]
+%else
+    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova                m3, [interp8_hpp_shuf]
+    vbroadcasti128      m2, [INTERP_OFFSET_PS]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    sub                 r0, 2
+    test                r5d, r5d
+    mov                 r4d, %1
+    jz                  .loop0
+    sub                 r0, r1
+    add                 r4d, 3
+
+.loop0:
+    vbroadcasti128      m4, [r0]
+    vbroadcasti128      m5, [r0 + 8]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2], xm4
+
+    vbroadcasti128      m4, [r0 + 16]
+    vbroadcasti128      m5, [r0 + 24]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 16], xm4
+
+    vbroadcasti128      m4, [r0 + 32]
+    vbroadcasti128      m5, [r0 + 40]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 32], xm4
+
+    vbroadcasti128      m4, [r0 + 48]
+    vbroadcasti128      m5, [r0 + 56]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 48], xm4
+
+    vbroadcasti128      m4, [r0 + 64]
+    vbroadcasti128      m5, [r0 + 72]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 64], xm4
+
+    vbroadcasti128      m4, [r0 + 80]
+    vbroadcasti128      m5, [r0 + 88]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 80], xm4
+
+    vbroadcasti128      m4, [r0 + 96]
+    vbroadcasti128      m5, [r0 + 104]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 96], xm4
+
+    vbroadcasti128      m4, [r0 + 112]
+    vbroadcasti128      m5, [r0 + 120]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 112], xm4
+
+    add                 r2, r3
+    add                 r0, r1
+    dec                 r4d
+    jnz                 .loop0
+    RET
+%endif
+%endmacro
+
+IPFILTER_CHROMA_PS_64xN_AVX2 64
+IPFILTER_CHROMA_PS_64xN_AVX2 48
+IPFILTER_CHROMA_PS_64xN_AVX2 32
+IPFILTER_CHROMA_PS_64xN_AVX2 16
+
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_horiz_ps_48x64, 4, 7, 6
+    add                 r1d, r1d
+    add                 r3d, r3d
+    mov                 r4d, r4m
+    mov                 r5d, r5m
+
+%ifdef PIC
+    lea                 r6, [tab_ChromaCoeff]
+    vpbroadcastq        m0, [r6 + r4 * 8]
+%else
+    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova                m3, [interp8_hpp_shuf]
+    vbroadcasti128      m2, [INTERP_OFFSET_PS]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    sub                 r0, 2
+    test                r5d, r5d
+    mov                 r4d, 64
+    jz                  .loop0
+    sub                 r0, r1
+    add                 r4d, 3
+
+.loop0:
+    vbroadcasti128      m4, [r0]
+    vbroadcasti128      m5, [r0 + 8]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2], xm4
+
+    vbroadcasti128      m4, [r0 + 16]
+    vbroadcasti128      m5, [r0 + 24]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 16], xm4
+
+    vbroadcasti128      m4, [r0 + 32]
+    vbroadcasti128      m5, [r0 + 40]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 32], xm4
+
+    vbroadcasti128      m4, [r0 + 48]
+    vbroadcasti128      m5, [r0 + 56]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 48], xm4
+
+    vbroadcasti128      m4, [r0 + 64]
+    vbroadcasti128      m5, [r0 + 72]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 64], xm4
+
+    vbroadcasti128      m4, [r0 + 80]
+    vbroadcasti128      m5, [r0 + 88]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 80], xm4
+
+    add                 r2, r3
+    add                 r0, r1
+    dec                 r4d
+    jnz                 .loop0
+    RET
+%endif
+
+%macro IPFILTER_CHROMA_PS_6xN_AVX2 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_horiz_ps_6x%1, 4, 7, 6
+    add                 r1d, r1d
+    add                 r3d, r3d
+    mov                 r4d, r4m
+    mov                 r5d, r5m
+
+%ifdef PIC
+    lea                 r6, [tab_ChromaCoeff]
+    vpbroadcastq        m0, [r6 + r4 * 8]
+%else
+    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova                m3, [interp8_hpp_shuf]
+    vbroadcasti128      m2, [INTERP_OFFSET_PS]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    sub                 r0, 2
+    test                r5d, r5d
+    mov                 r4d, %1
+    jz                  .loop0
+    sub                 r0, r1
+    add                 r4d, 3
+
+.loop0:
+    vbroadcasti128      m4, [r0]
+    vbroadcasti128      m5, [r0 + 8]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movq                [r2], xm4
+    pextrd              [r2 + 8], xm4, 2
+    add                 r2, r3
+    add                 r0, r1
+    dec                 r4d
+    jnz                 .loop0
+    RET
+%endif
+%endmacro
+
+    IPFILTER_CHROMA_PS_6xN_AVX2 8
+    IPFILTER_CHROMA_PS_6xN_AVX2 16
diff --git a/source/common/x86/ipfilter16.asm b/source/common/x86/ipfilter16.asm
index d44cfd9edb..c5b775121e 100644
--- a/source/common/x86/ipfilter16.asm
+++ b/source/common/x86/ipfilter16.asm
@@ -47,22 +47,10 @@
 
 SECTION_RODATA 32
 
-tab_c_32:         times 8 dd 32
 tab_c_524800:     times 4 dd 524800
 tab_c_n8192:      times 8 dw -8192
 pd_524800:        times 8 dd 524800
 
-tab_Tm16:         db 0, 1, 2, 3, 4,  5,  6, 7, 2, 3, 4,  5, 6, 7, 8, 9
-
-tab_ChromaCoeff:  dw  0, 64,  0,  0
-                  dw -2, 58, 10, -2
-                  dw -4, 54, 16, -2
-                  dw -6, 46, 28, -4
-                  dw -4, 36, 36, -4
-                  dw -4, 28, 46, -6
-                  dw -2, 16, 54, -4
-                  dw -2, 10, 58, -2
-
 const tab_ChromaCoeffV,  times 8 dw 0, 64
                          times 8 dw 0, 0
 
@@ -111,11 +99,6 @@ tab_ChromaCoeffVer: times 8 dw 0, 64
                     times 8 dw -2, 10
                     times 8 dw 58, -2
 
-tab_LumaCoeff:    dw   0, 0,  0,  64,  0,   0,  0,  0
-                  dw  -1, 4, -10, 58,  17, -5,  1,  0
-                  dw  -1, 4, -11, 40,  40, -11, 4, -1
-                  dw   0, 1, -5,  17,  58, -10, 4, -1
-
 ALIGN 32
 tab_LumaCoeffV:   times 4 dw 0, 0
                   times 4 dw 0, 64
@@ -157,14 +140,6 @@ tab_LumaCoeffVer: times 8 dw 0, 0
                   times 8 dw 58, -10
                   times 8 dw 4, -1
 
-const interp8_hps_shuf,     dd 0, 4, 1, 5, 2, 6, 3, 7
-
-const interp8_hpp_shuf,     db 0, 1, 2, 3, 4, 5, 6, 7, 2, 3, 4, 5, 6, 7, 8, 9
-                            db 4, 5, 6, 7, 8, 9, 10, 11, 6, 7, 8, 9, 10, 11, 12, 13
-
-const interp8_hpp_shuf_new, db 0, 1, 2, 3, 2, 3, 4, 5, 4, 5, 6, 7, 6, 7, 8, 9
-                            db 4, 5, 6, 7, 6, 7, 8, 9, 8, 9, 10, 11, 10, 11, 12, 13
-
 SECTION .text
 cextern pd_8
 cextern pd_32
@@ -175,255 +150,6 @@ cextern pd_n131072
 cextern pw_2000
 cextern idct8_shuf2
 
-%macro FILTER_LUMA_HOR_4_sse2 1
-    movu        m4,     [r0 + %1]       ; m4 = src[0-7]
-    movu        m5,     [r0 + %1 + 2]   ; m5 = src[1-8]
-    pmaddwd     m4,     m0
-    pmaddwd     m5,     m0
-    pshufd      m2,     m4,     q2301
-    paddd       m4,     m2
-    pshufd      m2,     m5,     q2301
-    paddd       m5,     m2
-    pshufd      m4,     m4,     q3120
-    pshufd      m5,     m5,     q3120
-    punpcklqdq  m4,     m5
-
-    movu        m5,     [r0 + %1 + 4]   ; m5 = src[2-9]
-    movu        m3,     [r0 + %1 + 6]   ; m3 = src[3-10]
-    pmaddwd     m5,     m0
-    pmaddwd     m3,     m0
-    pshufd      m2,     m5,     q2301
-    paddd       m5,     m2
-    pshufd      m2,     m3,     q2301
-    paddd       m3,     m2
-    pshufd      m5,     m5,     q3120
-    pshufd      m3,     m3,     q3120
-    punpcklqdq  m5,     m3
-
-    pshufd      m2,     m4,     q2301
-    paddd       m4,     m2
-    pshufd      m2,     m5,     q2301
-    paddd       m5,     m2
-    pshufd      m4,     m4,     q3120
-    pshufd      m5,     m5,     q3120
-    punpcklqdq  m4,     m5
-    paddd       m4,     m1
-%endmacro
-
-%macro FILTER_LUMA_HOR_8_sse2 1
-    movu        m4,     [r0 + %1]       ; m4 = src[0-7]
-    movu        m5,     [r0 + %1 + 2]   ; m5 = src[1-8]
-    pmaddwd     m4,     m0
-    pmaddwd     m5,     m0
-    pshufd      m2,     m4,     q2301
-    paddd       m4,     m2
-    pshufd      m2,     m5,     q2301
-    paddd       m5,     m2
-    pshufd      m4,     m4,     q3120
-    pshufd      m5,     m5,     q3120
-    punpcklqdq  m4,     m5
-
-    movu        m5,     [r0 + %1 + 4]   ; m5 = src[2-9]
-    movu        m3,     [r0 + %1 + 6]   ; m3 = src[3-10]
-    pmaddwd     m5,     m0
-    pmaddwd     m3,     m0
-    pshufd      m2,     m5,     q2301
-    paddd       m5,     m2
-    pshufd      m2,     m3,     q2301
-    paddd       m3,     m2
-    pshufd      m5,     m5,     q3120
-    pshufd      m3,     m3,     q3120
-    punpcklqdq  m5,     m3
-
-    pshufd      m2,     m4,     q2301
-    paddd       m4,     m2
-    pshufd      m2,     m5,     q2301
-    paddd       m5,     m2
-    pshufd      m4,     m4,     q3120
-    pshufd      m5,     m5,     q3120
-    punpcklqdq  m4,     m5
-    paddd       m4,     m1
-
-    movu        m5,     [r0 + %1 + 8]   ; m5 = src[4-11]
-    movu        m6,     [r0 + %1 + 10]  ; m6 = src[5-12]
-    pmaddwd     m5,     m0
-    pmaddwd     m6,     m0
-    pshufd      m2,     m5,     q2301
-    paddd       m5,     m2
-    pshufd      m2,     m6,     q2301
-    paddd       m6,     m2
-    pshufd      m5,     m5,     q3120
-    pshufd      m6,     m6,     q3120
-    punpcklqdq  m5,     m6
-
-    movu        m6,     [r0 + %1 + 12]  ; m6 = src[6-13]
-    movu        m3,     [r0 + %1 + 14]  ; m3 = src[7-14]
-    pmaddwd     m6,     m0
-    pmaddwd     m3,     m0
-    pshufd      m2,     m6,     q2301
-    paddd       m6,     m2
-    pshufd      m2,     m3,     q2301
-    paddd       m3,     m2
-    pshufd      m6,     m6,     q3120
-    pshufd      m3,     m3,     q3120
-    punpcklqdq  m6,     m3
-
-    pshufd      m2,     m5,     q2301
-    paddd       m5,     m2
-    pshufd      m2,     m6,     q2301
-    paddd       m6,     m2
-    pshufd      m5,     m5,     q3120
-    pshufd      m6,     m6,     q3120
-    punpcklqdq  m5,     m6
-    paddd       m5,     m1
-%endmacro
-
-;------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_p%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;------------------------------------------------------------------------------------------------------------
-%macro FILTER_HOR_LUMA_sse2 3
-INIT_XMM sse2
-cglobal interp_8tap_horiz_%3_%1x%2, 4, 7, 8
-    mov         r4d,    r4m
-    sub         r0,     6
-    shl         r4d,    4
-    add         r1d,    r1d
-    add         r3d,    r3d
-
-%ifdef PIC
-    lea         r6,     [tab_LumaCoeff]
-    mova        m0,     [r6 + r4]
-%else
-    mova        m0,     [tab_LumaCoeff + r4]
-%endif
-
-%ifidn %3, pp
-    mova        m1,     [pd_32]
-    pxor        m7,     m7
-%else
-    mova        m1,     [INTERP_OFFSET_PS]
-%endif
-
-    mov         r4d,    %2
-%ifidn %3, ps
-    cmp         r5m,    byte 0
-    je          .loopH
-    lea         r6,     [r1 + 2 * r1]
-    sub         r0,     r6
-    add         r4d,    7
-%endif
-
-.loopH:
-%assign x 0
-%rep %1/8
-    FILTER_LUMA_HOR_8_sse2 x
-
-%ifidn %3, pp
-    psrad       m4,     6
-    psrad       m5,     6
-    packssdw    m4,     m5
-    CLIPW       m4,     m7,     [pw_pixel_max]
-%else
-  %if BIT_DEPTH == 10
-    psrad       m4,     2
-    psrad       m5,     2
-  %elif BIT_DEPTH == 12
-    psrad       m4,     4
-    psrad       m5,     4
-  %endif
-    packssdw    m4,     m5
-%endif
-
-    movu        [r2 + x], m4
-%assign x x+16
-%endrep
-
-%rep (%1 % 8)/4
-    FILTER_LUMA_HOR_4_sse2 x
-
-%ifidn %3, pp
-    psrad       m4,     6
-    packssdw    m4,     m4
-    CLIPW       m4,     m7,     [pw_pixel_max]
-%else
-  %if BIT_DEPTH == 10
-    psrad       m4,     2
-  %elif BIT_DEPTH == 12
-    psrad       m4,     4
-  %endif
-    packssdw    m4,     m4
-%endif
-
-    movh        [r2 + x], m4
-%endrep
-
-    add         r0,     r1
-    add         r2,     r3
-
-    dec         r4d
-    jnz         .loopH
-    RET
-
-%endmacro
-
-;------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
-;------------------------------------------------------------------------------------------------------------
-    FILTER_HOR_LUMA_sse2 4, 4, pp
-    FILTER_HOR_LUMA_sse2 4, 8, pp
-    FILTER_HOR_LUMA_sse2 4, 16, pp
-    FILTER_HOR_LUMA_sse2 8, 4, pp
-    FILTER_HOR_LUMA_sse2 8, 8, pp
-    FILTER_HOR_LUMA_sse2 8, 16, pp
-    FILTER_HOR_LUMA_sse2 8, 32, pp
-    FILTER_HOR_LUMA_sse2 12, 16, pp
-    FILTER_HOR_LUMA_sse2 16, 4, pp
-    FILTER_HOR_LUMA_sse2 16, 8, pp
-    FILTER_HOR_LUMA_sse2 16, 12, pp
-    FILTER_HOR_LUMA_sse2 16, 16, pp
-    FILTER_HOR_LUMA_sse2 16, 32, pp
-    FILTER_HOR_LUMA_sse2 16, 64, pp
-    FILTER_HOR_LUMA_sse2 24, 32, pp
-    FILTER_HOR_LUMA_sse2 32, 8, pp
-    FILTER_HOR_LUMA_sse2 32, 16, pp
-    FILTER_HOR_LUMA_sse2 32, 24, pp
-    FILTER_HOR_LUMA_sse2 32, 32, pp
-    FILTER_HOR_LUMA_sse2 32, 64, pp
-    FILTER_HOR_LUMA_sse2 48, 64, pp
-    FILTER_HOR_LUMA_sse2 64, 16, pp
-    FILTER_HOR_LUMA_sse2 64, 32, pp
-    FILTER_HOR_LUMA_sse2 64, 48, pp
-    FILTER_HOR_LUMA_sse2 64, 64, pp
-
-;---------------------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_ps_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;---------------------------------------------------------------------------------------------------------------------------
-    FILTER_HOR_LUMA_sse2 4, 4, ps
-    FILTER_HOR_LUMA_sse2 4, 8, ps
-    FILTER_HOR_LUMA_sse2 4, 16, ps
-    FILTER_HOR_LUMA_sse2 8, 4, ps
-    FILTER_HOR_LUMA_sse2 8, 8, ps
-    FILTER_HOR_LUMA_sse2 8, 16, ps
-    FILTER_HOR_LUMA_sse2 8, 32, ps
-    FILTER_HOR_LUMA_sse2 12, 16, ps
-    FILTER_HOR_LUMA_sse2 16, 4, ps
-    FILTER_HOR_LUMA_sse2 16, 8, ps
-    FILTER_HOR_LUMA_sse2 16, 12, ps
-    FILTER_HOR_LUMA_sse2 16, 16, ps
-    FILTER_HOR_LUMA_sse2 16, 32, ps
-    FILTER_HOR_LUMA_sse2 16, 64, ps
-    FILTER_HOR_LUMA_sse2 24, 32, ps
-    FILTER_HOR_LUMA_sse2 32, 8, ps
-    FILTER_HOR_LUMA_sse2 32, 16, ps
-    FILTER_HOR_LUMA_sse2 32, 24, ps
-    FILTER_HOR_LUMA_sse2 32, 32, ps
-    FILTER_HOR_LUMA_sse2 32, 64, ps
-    FILTER_HOR_LUMA_sse2 48, 64, ps
-    FILTER_HOR_LUMA_sse2 64, 16, ps
-    FILTER_HOR_LUMA_sse2 64, 32, ps
-    FILTER_HOR_LUMA_sse2 64, 48, ps
-    FILTER_HOR_LUMA_sse2 64, 64, ps
-
 %macro PROCESS_LUMA_VER_W4_4R_sse2 0
     movq       m0, [r0]
     movq       m1, [r0 + r1]
@@ -556,3518 +282,107 @@ cglobal interp_8tap_vert_%1_%2x%3, 5, 7, 8
 %assign x x+8
 %endrep
 
-    lea       r0, [r0 + 4 * r1 - 2 * %2]
-    lea       r2, [r2 + 4 * r3]
-
-    dec         r4d
-    jnz       .loopH
-
-    RET
-%endmacro
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_pp_%2x%3(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_sse2 pp, 4, 4
-    FILTER_VER_LUMA_sse2 pp, 8, 8
-    FILTER_VER_LUMA_sse2 pp, 8, 4
-    FILTER_VER_LUMA_sse2 pp, 4, 8
-    FILTER_VER_LUMA_sse2 pp, 16, 16
-    FILTER_VER_LUMA_sse2 pp, 16, 8
-    FILTER_VER_LUMA_sse2 pp, 8, 16
-    FILTER_VER_LUMA_sse2 pp, 16, 12
-    FILTER_VER_LUMA_sse2 pp, 12, 16
-    FILTER_VER_LUMA_sse2 pp, 16, 4
-    FILTER_VER_LUMA_sse2 pp, 4, 16
-    FILTER_VER_LUMA_sse2 pp, 32, 32
-    FILTER_VER_LUMA_sse2 pp, 32, 16
-    FILTER_VER_LUMA_sse2 pp, 16, 32
-    FILTER_VER_LUMA_sse2 pp, 32, 24
-    FILTER_VER_LUMA_sse2 pp, 24, 32
-    FILTER_VER_LUMA_sse2 pp, 32, 8
-    FILTER_VER_LUMA_sse2 pp, 8, 32
-    FILTER_VER_LUMA_sse2 pp, 64, 64
-    FILTER_VER_LUMA_sse2 pp, 64, 32
-    FILTER_VER_LUMA_sse2 pp, 32, 64
-    FILTER_VER_LUMA_sse2 pp, 64, 48
-    FILTER_VER_LUMA_sse2 pp, 48, 64
-    FILTER_VER_LUMA_sse2 pp, 64, 16
-    FILTER_VER_LUMA_sse2 pp, 16, 64
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_ps_%2x%3(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_sse2 ps, 4, 4
-    FILTER_VER_LUMA_sse2 ps, 8, 8
-    FILTER_VER_LUMA_sse2 ps, 8, 4
-    FILTER_VER_LUMA_sse2 ps, 4, 8
-    FILTER_VER_LUMA_sse2 ps, 16, 16
-    FILTER_VER_LUMA_sse2 ps, 16, 8
-    FILTER_VER_LUMA_sse2 ps, 8, 16
-    FILTER_VER_LUMA_sse2 ps, 16, 12
-    FILTER_VER_LUMA_sse2 ps, 12, 16
-    FILTER_VER_LUMA_sse2 ps, 16, 4
-    FILTER_VER_LUMA_sse2 ps, 4, 16
-    FILTER_VER_LUMA_sse2 ps, 32, 32
-    FILTER_VER_LUMA_sse2 ps, 32, 16
-    FILTER_VER_LUMA_sse2 ps, 16, 32
-    FILTER_VER_LUMA_sse2 ps, 32, 24
-    FILTER_VER_LUMA_sse2 ps, 24, 32
-    FILTER_VER_LUMA_sse2 ps, 32, 8
-    FILTER_VER_LUMA_sse2 ps, 8, 32
-    FILTER_VER_LUMA_sse2 ps, 64, 64
-    FILTER_VER_LUMA_sse2 ps, 64, 32
-    FILTER_VER_LUMA_sse2 ps, 32, 64
-    FILTER_VER_LUMA_sse2 ps, 64, 48
-    FILTER_VER_LUMA_sse2 ps, 48, 64
-    FILTER_VER_LUMA_sse2 ps, 64, 16
-    FILTER_VER_LUMA_sse2 ps, 16, 64
-
-%macro FILTERH_W2_4_sse3 2
-    movh        m3,     [r0 + %1]
-    movhps      m3,     [r0 + %1 + 2]
-    pmaddwd     m3,     m0
-    movh        m4,     [r0 + r1 + %1]
-    movhps      m4,     [r0 + r1 + %1 + 2]
-    pmaddwd     m4,     m0
-    pshufd      m2,     m3,     q2301
-    paddd       m3,     m2
-    pshufd      m2,     m4,     q2301
-    paddd       m4,     m2
-    pshufd      m3,     m3,     q3120
-    pshufd      m4,     m4,     q3120
-    punpcklqdq  m3,     m4
-    paddd       m3,     m1
-    movh        m5,     [r0 + 2 * r1 + %1]
-    movhps      m5,     [r0 + 2 * r1 + %1 + 2]
-    pmaddwd     m5,     m0
-    movh        m4,     [r0 + r4 + %1]
-    movhps      m4,     [r0 + r4 + %1 + 2]
-    pmaddwd     m4,     m0
-    pshufd      m2,     m5,     q2301
-    paddd       m5,     m2
-    pshufd      m2,     m4,     q2301
-    paddd       m4,     m2
-    pshufd      m5,     m5,     q3120
-    pshufd      m4,     m4,     q3120
-    punpcklqdq  m5,     m4
-    paddd       m5,     m1
-%ifidn %2, pp
-    psrad       m3,     6
-    psrad       m5,     6
-    packssdw    m3,     m5
-    CLIPW       m3,     m7,     m6
-%else
-    psrad       m3,     INTERP_SHIFT_PS
-    psrad       m5,     INTERP_SHIFT_PS
-    packssdw    m3,     m5
-%endif
-    movd        [r2 + %1], m3
-    psrldq      m3,     4
-    movd        [r2 + r3 + %1], m3
-    psrldq      m3,     4
-    movd        [r2 + r3 * 2 + %1], m3
-    psrldq      m3,     4
-    movd        [r2 + r5 + %1], m3
-%endmacro
-
-%macro FILTERH_W2_3_sse3 1
-    movh        m3,     [r0 + %1]
-    movhps      m3,     [r0 + %1 + 2]
-    pmaddwd     m3,     m0
-    movh        m4,     [r0 + r1 + %1]
-    movhps      m4,     [r0 + r1 + %1 + 2]
-    pmaddwd     m4,     m0
-    pshufd      m2,     m3,     q2301
-    paddd       m3,     m2
-    pshufd      m2,     m4,     q2301
-    paddd       m4,     m2
-    pshufd      m3,     m3,     q3120
-    pshufd      m4,     m4,     q3120
-    punpcklqdq  m3,     m4
-    paddd       m3,     m1
-
-    movh        m5,     [r0 + 2 * r1 + %1]
-    movhps      m5,     [r0 + 2 * r1 + %1 + 2]
-    pmaddwd     m5,     m0
-
-    pshufd      m2,     m5,     q2301
-    paddd       m5,     m2
-    pshufd      m5,     m5,     q3120
-    paddd       m5,     m1
-
-    psrad       m3,     INTERP_SHIFT_PS
-    psrad       m5,     INTERP_SHIFT_PS
-    packssdw    m3,     m5
-
-    movd        [r2 + %1], m3
-    psrldq      m3,     4
-    movd        [r2 + r3 + %1], m3
-    psrldq      m3,     4
-    movd        [r2 + r3 * 2 + %1], m3
-%endmacro
-
-%macro FILTERH_W4_2_sse3 2
-    movh        m3,     [r0 + %1]
-    movhps      m3,     [r0 + %1 + 2]
-    pmaddwd     m3,     m0
-    movh        m4,     [r0 + %1 + 4]
-    movhps      m4,     [r0 + %1 + 6]
-    pmaddwd     m4,     m0
-    pshufd      m2,     m3,     q2301
-    paddd       m3,     m2
-    pshufd      m2,     m4,     q2301
-    paddd       m4,     m2
-    pshufd      m3,     m3,     q3120
-    pshufd      m4,     m4,     q3120
-    punpcklqdq  m3,     m4
-    paddd       m3,     m1
-
-    movh        m5,     [r0 + r1 + %1]
-    movhps      m5,     [r0 + r1 + %1 + 2]
-    pmaddwd     m5,     m0
-    movh        m4,     [r0 + r1 + %1 + 4]
-    movhps      m4,     [r0 + r1 + %1 + 6]
-    pmaddwd     m4,     m0
-    pshufd      m2,     m5,     q2301
-    paddd       m5,     m2
-    pshufd      m2,     m4,     q2301
-    paddd       m4,     m2
-    pshufd      m5,     m5,     q3120
-    pshufd      m4,     m4,     q3120
-    punpcklqdq  m5,     m4
-    paddd       m5,     m1
-%ifidn %2, pp
-    psrad       m3,     6
-    psrad       m5,     6
-    packssdw    m3,     m5
-    CLIPW       m3,     m7,     m6
-%else
-    psrad       m3,     INTERP_SHIFT_PS
-    psrad       m5,     INTERP_SHIFT_PS
-    packssdw    m3,     m5
-%endif
-    movh        [r2 + %1], m3
-    movhps      [r2 + r3 + %1], m3
-%endmacro
-
-%macro FILTERH_W4_1_sse3 1
-    movh        m3,     [r0 + 2 * r1 + %1]
-    movhps      m3,     [r0 + 2 * r1 + %1 + 2]
-    pmaddwd     m3,     m0
-    movh        m4,     [r0 + 2 * r1 + %1 + 4]
-    movhps      m4,     [r0 + 2 * r1 + %1 + 6]
-    pmaddwd     m4,     m0
-    pshufd      m2,     m3,     q2301
-    paddd       m3,     m2
-    pshufd      m2,     m4,     q2301
-    paddd       m4,     m2
-    pshufd      m3,     m3,     q3120
-    pshufd      m4,     m4,     q3120
-    punpcklqdq  m3,     m4
-    paddd       m3,     m1
-
-    psrad       m3,     INTERP_SHIFT_PS
-    packssdw    m3,     m3
-    movh        [r2 + r3 * 2 + %1], m3
-%endmacro
-
-%macro FILTERH_W8_1_sse3 2
-    movh        m3,     [r0 + %1]
-    movhps      m3,     [r0 + %1 + 2]
-    pmaddwd     m3,     m0
-    movh        m4,     [r0 + %1 + 4]
-    movhps      m4,     [r0 + %1 + 6]
-    pmaddwd     m4,     m0
-    pshufd      m2,     m3,     q2301
-    paddd       m3,     m2
-    pshufd      m2,     m4,     q2301
-    paddd       m4,     m2
-    pshufd      m3,     m3,     q3120
-    pshufd      m4,     m4,     q3120
-    punpcklqdq  m3,     m4
-    paddd       m3,     m1
-
-    movh        m5,     [r0 + %1 + 8]
-    movhps      m5,     [r0 + %1 + 10]
-    pmaddwd     m5,     m0
-    movh        m4,     [r0 + %1 + 12]
-    movhps      m4,     [r0 + %1 + 14]
-    pmaddwd     m4,     m0
-    pshufd      m2,     m5,     q2301
-    paddd       m5,     m2
-    pshufd      m2,     m4,     q2301
-    paddd       m4,     m2
-    pshufd      m5,     m5,     q3120
-    pshufd      m4,     m4,     q3120
-    punpcklqdq  m5,     m4
-    paddd       m5,     m1
-%ifidn %2, pp
-    psrad       m3,     6
-    psrad       m5,     6
-    packssdw    m3,     m5
-    CLIPW       m3,     m7,     m6
-%else
-    psrad       m3,     INTERP_SHIFT_PS
-    psrad       m5,     INTERP_SHIFT_PS
-    packssdw    m3,     m5
-%endif
-    movdqu      [r2 + %1], m3
-%endmacro
-
-;-----------------------------------------------------------------------------
-; void interp_4tap_horiz_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro FILTER_HOR_CHROMA_sse3 3
-INIT_XMM sse3
-cglobal interp_4tap_horiz_%3_%1x%2, 4, 7, 8
-    add         r3,     r3
-    add         r1,     r1
-    sub         r0,     2
-    mov         r4d,    r4m
-    add         r4d,    r4d
-
-%ifdef PIC
-    lea         r6,     [tab_ChromaCoeff]
-    movddup     m0,     [r6 + r4 * 4]
-%else
-    movddup     m0,     [tab_ChromaCoeff + r4 * 4]
-%endif
-
-%ifidn %3, ps
-    mova        m1,     [INTERP_OFFSET_PS]
-    cmp         r5m,    byte 0
-%if %1 <= 6
-    lea         r4,     [r1 * 3]
-    lea         r5,     [r3 * 3]
-%endif
-    je          .skip
-    sub         r0,     r1
-%if %1 <= 6
-%assign y 1
-%else
-%assign y 3
-%endif
-%assign z 0
-%rep y
-%assign x 0
-%rep %1/8
-    FILTERH_W8_1_sse3 x, %3
-%assign x x+16
-%endrep
-%if %1 == 4 || (%1 == 6 && z == 0) || (%1 == 12 && z == 0)
-    FILTERH_W4_2_sse3 x, %3
-    FILTERH_W4_1_sse3 x
-%assign x x+8
-%endif
-%if %1 == 2 || (%1 == 6 && z == 0)
-    FILTERH_W2_3_sse3 x
-%endif
-%if %1 <= 6
-    lea         r0,     [r0 + r4]
-    lea         r2,     [r2 + r5]
-%else
-    lea         r0,     [r0 + r1]
-    lea         r2,     [r2 + r3]
-%endif
-%assign z z+1
-%endrep
-.skip:
-%elifidn %3, pp
-    pxor        m7,     m7
-    mova        m6,     [pw_pixel_max]
-    mova        m1,     [tab_c_32]
-%if %1 == 2 || %1 == 6
-    lea         r4,     [r1 * 3]
-    lea         r5,     [r3 * 3]
-%endif
-%endif
-
-%if %1 == 2
-%assign y %2/4
-%elif %1 <= 6
-%assign y %2/2
-%else
-%assign y %2
-%endif
-%assign z 0
-%rep y
-%assign x 0
-%rep %1/8
-    FILTERH_W8_1_sse3 x, %3
-%assign x x+16
-%endrep
-%if %1 == 4 || %1 == 6 || (%1 == 12 && (z % 2) == 0)
-    FILTERH_W4_2_sse3 x, %3
-%assign x x+8
-%endif
-%if %1 == 2 || (%1 == 6 && (z % 2) == 0)
-    FILTERH_W2_4_sse3 x, %3
-%endif
-%assign z z+1
-%if z < y
-%if %1 == 2
-    lea         r0,     [r0 + 4 * r1]
-    lea         r2,     [r2 + 4 * r3]
-%elif %1 <= 6
-    lea         r0,     [r0 + 2 * r1]
-    lea         r2,     [r2 + 2 * r3]
-%else
-    lea         r0,     [r0 + r1]
-    lea         r2,     [r2 + r3]
-%endif
-%endif ;z < y
-%endrep
-
-    RET
-%endmacro
-
-;-----------------------------------------------------------------------------
-; void interp_4tap_horiz_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-
-FILTER_HOR_CHROMA_sse3 2, 4, pp
-FILTER_HOR_CHROMA_sse3 2, 8, pp
-FILTER_HOR_CHROMA_sse3 2, 16, pp
-FILTER_HOR_CHROMA_sse3 4, 2, pp
-FILTER_HOR_CHROMA_sse3 4, 4, pp
-FILTER_HOR_CHROMA_sse3 4, 8, pp
-FILTER_HOR_CHROMA_sse3 4, 16, pp
-FILTER_HOR_CHROMA_sse3 4, 32, pp
-FILTER_HOR_CHROMA_sse3 6, 8, pp
-FILTER_HOR_CHROMA_sse3 6, 16, pp
-FILTER_HOR_CHROMA_sse3 8, 2, pp
-FILTER_HOR_CHROMA_sse3 8, 4, pp
-FILTER_HOR_CHROMA_sse3 8, 6, pp
-FILTER_HOR_CHROMA_sse3 8, 8, pp
-FILTER_HOR_CHROMA_sse3 8, 12, pp
-FILTER_HOR_CHROMA_sse3 8, 16, pp
-FILTER_HOR_CHROMA_sse3 8, 32, pp
-FILTER_HOR_CHROMA_sse3 8, 64, pp
-FILTER_HOR_CHROMA_sse3 12, 16, pp
-FILTER_HOR_CHROMA_sse3 12, 32, pp
-FILTER_HOR_CHROMA_sse3 16, 4, pp
-FILTER_HOR_CHROMA_sse3 16, 8, pp
-FILTER_HOR_CHROMA_sse3 16, 12, pp
-FILTER_HOR_CHROMA_sse3 16, 16, pp
-FILTER_HOR_CHROMA_sse3 16, 24, pp
-FILTER_HOR_CHROMA_sse3 16, 32, pp
-FILTER_HOR_CHROMA_sse3 16, 64, pp
-FILTER_HOR_CHROMA_sse3 24, 32, pp
-FILTER_HOR_CHROMA_sse3 24, 64, pp
-FILTER_HOR_CHROMA_sse3 32, 8, pp
-FILTER_HOR_CHROMA_sse3 32, 16, pp
-FILTER_HOR_CHROMA_sse3 32, 24, pp
-FILTER_HOR_CHROMA_sse3 32, 32, pp
-FILTER_HOR_CHROMA_sse3 32, 48, pp
-FILTER_HOR_CHROMA_sse3 32, 64, pp
-FILTER_HOR_CHROMA_sse3 48, 64, pp
-FILTER_HOR_CHROMA_sse3 64, 16, pp
-FILTER_HOR_CHROMA_sse3 64, 32, pp
-FILTER_HOR_CHROMA_sse3 64, 48, pp
-FILTER_HOR_CHROMA_sse3 64, 64, pp
-
-;-----------------------------------------------------------------------------
-; void interp_4tap_horiz_ps_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-
-FILTER_HOR_CHROMA_sse3 2, 4, ps
-FILTER_HOR_CHROMA_sse3 2, 8, ps
-FILTER_HOR_CHROMA_sse3 2, 16, ps
-FILTER_HOR_CHROMA_sse3 4, 2, ps
-FILTER_HOR_CHROMA_sse3 4, 4, ps
-FILTER_HOR_CHROMA_sse3 4, 8, ps
-FILTER_HOR_CHROMA_sse3 4, 16, ps
-FILTER_HOR_CHROMA_sse3 4, 32, ps
-FILTER_HOR_CHROMA_sse3 6, 8, ps
-FILTER_HOR_CHROMA_sse3 6, 16, ps
-FILTER_HOR_CHROMA_sse3 8, 2, ps
-FILTER_HOR_CHROMA_sse3 8, 4, ps
-FILTER_HOR_CHROMA_sse3 8, 6, ps
-FILTER_HOR_CHROMA_sse3 8, 8, ps
-FILTER_HOR_CHROMA_sse3 8, 12, ps
-FILTER_HOR_CHROMA_sse3 8, 16, ps
-FILTER_HOR_CHROMA_sse3 8, 32, ps
-FILTER_HOR_CHROMA_sse3 8, 64, ps
-FILTER_HOR_CHROMA_sse3 12, 16, ps
-FILTER_HOR_CHROMA_sse3 12, 32, ps
-FILTER_HOR_CHROMA_sse3 16, 4, ps
-FILTER_HOR_CHROMA_sse3 16, 8, ps
-FILTER_HOR_CHROMA_sse3 16, 12, ps
-FILTER_HOR_CHROMA_sse3 16, 16, ps
-FILTER_HOR_CHROMA_sse3 16, 24, ps
-FILTER_HOR_CHROMA_sse3 16, 32, ps
-FILTER_HOR_CHROMA_sse3 16, 64, ps
-FILTER_HOR_CHROMA_sse3 24, 32, ps
-FILTER_HOR_CHROMA_sse3 24, 64, ps
-FILTER_HOR_CHROMA_sse3 32, 8, ps
-FILTER_HOR_CHROMA_sse3 32, 16, ps
-FILTER_HOR_CHROMA_sse3 32, 24, ps
-FILTER_HOR_CHROMA_sse3 32, 32, ps
-FILTER_HOR_CHROMA_sse3 32, 48, ps
-FILTER_HOR_CHROMA_sse3 32, 64, ps
-FILTER_HOR_CHROMA_sse3 48, 64, ps
-FILTER_HOR_CHROMA_sse3 64, 16, ps
-FILTER_HOR_CHROMA_sse3 64, 32, ps
-FILTER_HOR_CHROMA_sse3 64, 48, ps
-FILTER_HOR_CHROMA_sse3 64, 64, ps
-
-%macro FILTER_P2S_2_4_sse2 1
-    movd        m0,     [r0 + %1]
-    movd        m2,     [r0 + r1 * 2 + %1]
-    movhps      m0,     [r0 + r1 + %1]
-    movhps      m2,     [r0 + r4 + %1]
-    psllw       m0,     (14 - BIT_DEPTH)
-    psllw       m2,     (14 - BIT_DEPTH)
-    psubw       m0,     m1
-    psubw       m2,     m1
-
-    movd        [r2 + r3 * 0 + %1], m0
-    movd        [r2 + r3 * 2 + %1], m2
-    movhlps     m0,     m0
-    movhlps     m2,     m2
-    movd        [r2 + r3 * 1 + %1], m0
-    movd        [r2 + r5 + %1], m2
-%endmacro
-
-%macro FILTER_P2S_4_4_sse2 1
-    movh        m0,     [r0 + %1]
-    movhps      m0,     [r0 + r1 + %1]
-    psllw       m0,     (14 - BIT_DEPTH)
-    psubw       m0,     m1
-    movh        [r2 + r3 * 0 + %1], m0
-    movhps      [r2 + r3 * 1 + %1], m0
-
-    movh        m2,     [r0 + r1 * 2 + %1]
-    movhps      m2,     [r0 + r4 + %1]
-    psllw       m2,     (14 - BIT_DEPTH)
-    psubw       m2,     m1
-    movh        [r2 + r3 * 2 + %1], m2
-    movhps      [r2 + r5 + %1], m2
-%endmacro
-
-%macro FILTER_P2S_4_2_sse2 0
-    movh        m0,     [r0]
-    movhps      m0,     [r0 + r1 * 2]
-    psllw       m0,     (14 - BIT_DEPTH)
-    psubw       m0,     [pw_2000]
-    movh        [r2 + r3 * 0], m0
-    movhps      [r2 + r3 * 2], m0
-%endmacro
-
-%macro FILTER_P2S_8_4_sse2 1
-    movu        m0,     [r0 + %1]
-    movu        m2,     [r0 + r1 + %1]
-    psllw       m0,     (14 - BIT_DEPTH)
-    psllw       m2,     (14 - BIT_DEPTH)
-    psubw       m0,     m1
-    psubw       m2,     m1
-    movu        [r2 + r3 * 0 + %1], m0
-    movu        [r2 + r3 * 1 + %1], m2
-
-    movu        m3,     [r0 + r1 * 2 + %1]
-    movu        m4,     [r0 + r4 + %1]
-    psllw       m3,     (14 - BIT_DEPTH)
-    psllw       m4,     (14 - BIT_DEPTH)
-    psubw       m3,     m1
-    psubw       m4,     m1
-    movu        [r2 + r3 * 2 + %1], m3
-    movu        [r2 + r5 + %1], m4
-%endmacro
-
-%macro FILTER_P2S_8_2_sse2 1
-    movu        m0,     [r0 + %1]
-    movu        m2,     [r0 + r1 + %1]
-    psllw       m0,     (14 - BIT_DEPTH)
-    psllw       m2,     (14 - BIT_DEPTH)
-    psubw       m0,     m1
-    psubw       m2,     m1
-    movu        [r2 + r3 * 0 + %1], m0
-    movu        [r2 + r3 * 1 + %1], m2
-%endmacro
-
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride)
-;-----------------------------------------------------------------------------
-%macro FILTER_PIX_TO_SHORT_sse2 2
-INIT_XMM sse2
-cglobal filterPixelToShort_%1x%2, 4, 6, 3
-%if %2 == 2
-%if %1 == 4
-    FILTER_P2S_4_2_sse2
-%elif %1 == 8
-    add        r1d, r1d
-    add        r3d, r3d
-    mova       m1, [pw_2000]
-    FILTER_P2S_8_2_sse2 0
-%endif
-%else
-    add        r1d, r1d
-    add        r3d, r3d
-    mova       m1, [pw_2000]
-    lea        r4, [r1 * 3]
-    lea        r5, [r3 * 3]
-%assign y 1
-%rep %2/4
-%assign x 0
-%rep %1/8
-    FILTER_P2S_8_4_sse2 x
-%if %2 == 6
-    lea         r0,     [r0 + 4 * r1]
-    lea         r2,     [r2 + 4 * r3]
-    FILTER_P2S_8_2_sse2 x
-%endif
-%assign x x+16
-%endrep
-%rep (%1 % 8)/4
-    FILTER_P2S_4_4_sse2 x
-%assign x x+8
-%endrep
-%rep (%1 % 4)/2
-    FILTER_P2S_2_4_sse2 x
-%endrep
-%if y < %2/4
-    lea         r0,     [r0 + 4 * r1]
-    lea         r2,     [r2 + 4 * r3]
-%assign y y+1
-%endif
-%endrep
-%endif
-RET
-%endmacro
-
-    FILTER_PIX_TO_SHORT_sse2 2, 4
-    FILTER_PIX_TO_SHORT_sse2 2, 8
-    FILTER_PIX_TO_SHORT_sse2 2, 16
-    FILTER_PIX_TO_SHORT_sse2 4, 2
-    FILTER_PIX_TO_SHORT_sse2 4, 4
-    FILTER_PIX_TO_SHORT_sse2 4, 8
-    FILTER_PIX_TO_SHORT_sse2 4, 16
-    FILTER_PIX_TO_SHORT_sse2 4, 32
-    FILTER_PIX_TO_SHORT_sse2 6, 8
-    FILTER_PIX_TO_SHORT_sse2 6, 16
-    FILTER_PIX_TO_SHORT_sse2 8, 2
-    FILTER_PIX_TO_SHORT_sse2 8, 4
-    FILTER_PIX_TO_SHORT_sse2 8, 6
-    FILTER_PIX_TO_SHORT_sse2 8, 8
-    FILTER_PIX_TO_SHORT_sse2 8, 12
-    FILTER_PIX_TO_SHORT_sse2 8, 16
-    FILTER_PIX_TO_SHORT_sse2 8, 32
-    FILTER_PIX_TO_SHORT_sse2 8, 64
-    FILTER_PIX_TO_SHORT_sse2 12, 16
-    FILTER_PIX_TO_SHORT_sse2 12, 32
-    FILTER_PIX_TO_SHORT_sse2 16, 4
-    FILTER_PIX_TO_SHORT_sse2 16, 8
-    FILTER_PIX_TO_SHORT_sse2 16, 12
-    FILTER_PIX_TO_SHORT_sse2 16, 16
-    FILTER_PIX_TO_SHORT_sse2 16, 24
-    FILTER_PIX_TO_SHORT_sse2 16, 32
-    FILTER_PIX_TO_SHORT_sse2 16, 64
-    FILTER_PIX_TO_SHORT_sse2 24, 32
-    FILTER_PIX_TO_SHORT_sse2 24, 64
-    FILTER_PIX_TO_SHORT_sse2 32, 8
-    FILTER_PIX_TO_SHORT_sse2 32, 16
-    FILTER_PIX_TO_SHORT_sse2 32, 24
-    FILTER_PIX_TO_SHORT_sse2 32, 32
-    FILTER_PIX_TO_SHORT_sse2 32, 48
-    FILTER_PIX_TO_SHORT_sse2 32, 64
-    FILTER_PIX_TO_SHORT_sse2 48, 64
-    FILTER_PIX_TO_SHORT_sse2 64, 16
-    FILTER_PIX_TO_SHORT_sse2 64, 32
-    FILTER_PIX_TO_SHORT_sse2 64, 48
-    FILTER_PIX_TO_SHORT_sse2 64, 64
-
-;------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_pp_4x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;------------------------------------------------------------------------------------------------------------
-%macro FILTER_HOR_LUMA_W4 3
-INIT_XMM sse4
-cglobal interp_8tap_horiz_%3_%1x%2, 4, 7, 8
-    mov         r4d, r4m
-    sub         r0, 6
-    shl         r4d, 4
-    add         r1, r1
-    add         r3, r3
-
-%ifdef PIC
-    lea         r6, [tab_LumaCoeff]
-    mova        m0, [r6 + r4]
-%else
-    mova        m0, [tab_LumaCoeff + r4]
-%endif
-
-%ifidn %3, pp
-    mova        m1, [pd_32]
-    pxor        m6, m6
-    mova        m7, [pw_pixel_max]
-%else
-    mova        m1, [INTERP_OFFSET_PS]
-%endif
-
-    mov         r4d, %2
-%ifidn %3, ps
-    cmp         r5m, byte 0
-    je          .loopH
-    lea         r6, [r1 + 2 * r1]
-    sub         r0, r6
-    add         r4d, 7
-%endif
-
-.loopH:
-    movu        m2, [r0]                     ; m2 = src[0-7]
-    movu        m3, [r0 + 16]                ; m3 = src[8-15]
-
-    pmaddwd     m4, m2, m0
-    palignr     m5, m3, m2, 2                ; m5 = src[1-8]
-    pmaddwd     m5, m0
-    phaddd      m4, m5
-
-    palignr     m5, m3, m2, 4                ; m5 = src[2-9]
-    pmaddwd     m5, m0
-    palignr     m3, m2, 6                    ; m3 = src[3-10]
-    pmaddwd     m3, m0
-    phaddd      m5, m3
-
-    phaddd      m4, m5
-    paddd       m4, m1
-%ifidn %3, pp
-    psrad       m4, 6
-    packusdw    m4, m4
-    CLIPW       m4, m6, m7
-%else
-    psrad       m4, INTERP_SHIFT_PS
-    packssdw    m4, m4
-%endif
-
-    movh        [r2], m4
-
-    add         r0, r1
-    add         r2, r3
-
-    dec         r4d
-    jnz         .loopH
-    RET
-%endmacro
-
-;------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_pp_4x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
-;------------------------------------------------------------------------------------------------------------
-FILTER_HOR_LUMA_W4 4, 4, pp
-FILTER_HOR_LUMA_W4 4, 8, pp
-FILTER_HOR_LUMA_W4 4, 16, pp
-
-;---------------------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_ps_4x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;---------------------------------------------------------------------------------------------------------------------------
-FILTER_HOR_LUMA_W4 4, 4, ps
-FILTER_HOR_LUMA_W4 4, 8, ps
-FILTER_HOR_LUMA_W4 4, 16, ps
-
-;------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;------------------------------------------------------------------------------------------------------------
-%macro FILTER_HOR_LUMA_W8 3
-INIT_XMM sse4
-cglobal interp_8tap_horiz_%3_%1x%2, 4, 7, 8
-
-    add         r1, r1
-    add         r3, r3
-    mov         r4d, r4m
-    sub         r0, 6
-    shl         r4d, 4
-
-%ifdef PIC
-    lea         r6, [tab_LumaCoeff]
-    mova        m0, [r6 + r4]
-%else
-    mova        m0, [tab_LumaCoeff + r4]
-%endif
-
-%ifidn %3, pp
-    mova        m1, [pd_32]
-    pxor        m7, m7
-%else
-    mova        m1, [INTERP_OFFSET_PS]
-%endif
-
-    mov         r4d, %2
-%ifidn %3, ps
-    cmp         r5m, byte 0
-    je         .loopH
-    lea         r6, [r1 + 2 * r1]
-    sub         r0, r6
-    add         r4d, 7
-%endif
-
-.loopH:
-    movu        m2, [r0]                     ; m2 = src[0-7]
-    movu        m3, [r0 + 16]                ; m3 = src[8-15]
-
-    pmaddwd     m4, m2, m0
-    palignr     m5, m3, m2, 2                ; m5 = src[1-8]
-    pmaddwd     m5, m0
-    phaddd      m4, m5
-
-    palignr     m5, m3, m2, 4                ; m5 = src[2-9]
-    pmaddwd     m5, m0
-    palignr     m6, m3, m2, 6                ; m6 = src[3-10]
-    pmaddwd     m6, m0
-    phaddd      m5, m6
-    phaddd      m4, m5
-    paddd       m4, m1
-
-    palignr     m5, m3, m2, 8                ; m5 = src[4-11]
-    pmaddwd     m5, m0
-    palignr     m6, m3, m2, 10               ; m6 = src[5-12]
-    pmaddwd     m6, m0
-    phaddd      m5, m6
-
-    palignr     m6, m3, m2, 12               ; m6 = src[6-13]
-    pmaddwd     m6, m0
-    palignr     m3, m2, 14                   ; m3 = src[7-14]
-    pmaddwd     m3, m0
-    phaddd      m6, m3
-    phaddd      m5, m6
-    paddd       m5, m1
-%ifidn %3, pp
-    psrad       m4, 6
-    psrad       m5, 6
-    packusdw    m4, m5
-    CLIPW       m4, m7, [pw_pixel_max]
-%else
-    psrad       m4, INTERP_SHIFT_PS
-    psrad       m5, INTERP_SHIFT_PS
-    packssdw    m4, m5
-%endif
-
-    movu        [r2], m4
-
-    add         r0, r1
-    add         r2, r3
-
-    dec         r4d
-    jnz        .loopH
-    RET
-%endmacro
-
-;------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_pp_8x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
-;------------------------------------------------------------------------------------------------------------
-FILTER_HOR_LUMA_W8 8, 4, pp
-FILTER_HOR_LUMA_W8 8, 8, pp
-FILTER_HOR_LUMA_W8 8, 16, pp
-FILTER_HOR_LUMA_W8 8, 32, pp
-
-;---------------------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_ps_8x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;---------------------------------------------------------------------------------------------------------------------------
-FILTER_HOR_LUMA_W8 8, 4, ps
-FILTER_HOR_LUMA_W8 8, 8, ps
-FILTER_HOR_LUMA_W8 8, 16, ps
-FILTER_HOR_LUMA_W8 8, 32, ps
-
-;--------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-%macro FILTER_HOR_LUMA_W12 3
-INIT_XMM sse4
-cglobal interp_8tap_horiz_%3_%1x%2, 4, 7, 8
-
-    add         r1, r1
-    add         r3, r3
-    mov         r4d, r4m
-    sub         r0, 6
-    shl         r4d, 4
-
-%ifdef PIC
-    lea         r6, [tab_LumaCoeff]
-    mova        m0, [r6 + r4]
-%else
-    mova        m0, [tab_LumaCoeff + r4]
-%endif
-%ifidn %3, pp
-    mova        m1, [INTERP_OFFSET_PP]
-%else
-    mova        m1, [INTERP_OFFSET_PS]
-%endif
-
-    mov         r4d, %2
-%ifidn %3, ps
-    cmp         r5m, byte 0
-    je          .loopH
-    lea         r6, [r1 + 2 * r1]
-    sub         r0, r6
-    add         r4d, 7
-%endif
-
-.loopH:
-    movu        m2, [r0]                     ; m2 = src[0-7]
-    movu        m3, [r0 + 16]                ; m3 = src[8-15]
-
-    pmaddwd     m4, m2, m0
-    palignr     m5, m3, m2, 2                ; m5 = src[1-8]
-    pmaddwd     m5, m0
-    phaddd      m4, m5
-
-    palignr     m5, m3, m2, 4                ; m5 = src[2-9]
-    pmaddwd     m5, m0
-    palignr     m6, m3, m2, 6                ; m6 = src[3-10]
-    pmaddwd     m6, m0
-    phaddd      m5, m6
-    phaddd      m4, m5
-    paddd       m4, m1
-
-    palignr     m5, m3, m2, 8                ; m5 = src[4-11]
-    pmaddwd     m5, m0
-    palignr     m6, m3, m2, 10               ; m6 = src[5-12]
-    pmaddwd     m6, m0
-    phaddd      m5, m6
-
-    palignr     m6, m3, m2, 12               ; m6 = src[6-13]
-    pmaddwd     m6, m0
-    palignr     m7, m3, m2, 14               ; m2 = src[7-14]
-    pmaddwd     m7, m0
-    phaddd      m6, m7
-    phaddd      m5, m6
-    paddd       m5, m1
-%ifidn %3, pp
-    psrad       m4, INTERP_SHIFT_PP
-    psrad       m5, INTERP_SHIFT_PP
-    packusdw    m4, m5
-    pxor        m5, m5
-    CLIPW       m4, m5, [pw_pixel_max]
-%else
-    psrad       m4, INTERP_SHIFT_PS
-    psrad       m5, INTERP_SHIFT_PS
-    packssdw    m4, m5
-%endif
-
-    movu        [r2], m4
-
-    movu        m2, [r0 + 32]                ; m2 = src[16-23]
-
-    pmaddwd     m4, m3, m0                   ; m3 = src[8-15]
-    palignr     m5, m2, m3, 2                ; m5 = src[9-16]
-    pmaddwd     m5, m0
-    phaddd      m4, m5
-
-    palignr     m5, m2, m3, 4                ; m5 = src[10-17]
-    pmaddwd     m5, m0
-    palignr     m2, m3, 6                    ; m2 = src[11-18]
-    pmaddwd     m2, m0
-    phaddd      m5, m2
-    phaddd      m4, m5
-    paddd       m4, m1
-%ifidn %3, pp
-    psrad       m4, INTERP_SHIFT_PP
-    packusdw    m4, m4
-    pxor        m5, m5
-    CLIPW       m4, m5, [pw_pixel_max]
-%else
-    psrad       m4, INTERP_SHIFT_PS
-    packssdw    m4, m4
-%endif
-
-    movh        [r2 + 16], m4
-
-    add         r0, r1
-    add         r2, r3
-
-    dec         r4d
-    jnz         .loopH
-    RET
-%endmacro
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_pp_12x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-FILTER_HOR_LUMA_W12 12, 16, pp
-
-;----------------------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_ps_12x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;----------------------------------------------------------------------------------------------------------------------------
-FILTER_HOR_LUMA_W12 12, 16, ps
-
-;--------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-%macro FILTER_HOR_LUMA_W16 3
-INIT_XMM sse4
-cglobal interp_8tap_horiz_%3_%1x%2, 4, 7, 8
-
-    add         r1, r1
-    add         r3, r3
-    mov         r4d, r4m
-    sub         r0, 6
-    shl         r4d, 4
-
-%ifdef PIC
-    lea         r6, [tab_LumaCoeff]
-    mova        m0, [r6 + r4]
-%else
-    mova        m0, [tab_LumaCoeff + r4]
-%endif
-
-%ifidn %3, pp
-    mova        m1, [pd_32]
-%else
-    mova        m1, [INTERP_OFFSET_PS]
-%endif
-
-    mov         r4d, %2
-%ifidn %3, ps
-    cmp         r5m, byte 0
-    je          .loopH
-    lea         r6, [r1 + 2 * r1]
-    sub         r0, r6
-    add         r4d, 7
-%endif
-
-.loopH:
-%assign x 0
-%rep %1 / 16
-    movu        m2, [r0 + x]                 ; m2 = src[0-7]
-    movu        m3, [r0 + 16 + x]            ; m3 = src[8-15]
-
-    pmaddwd     m4, m2, m0
-    palignr     m5, m3, m2, 2                ; m5 = src[1-8]
-    pmaddwd     m5, m0
-    phaddd      m4, m5
-
-    palignr     m5, m3, m2, 4                ; m5 = src[2-9]
-    pmaddwd     m5, m0
-    palignr     m6, m3, m2, 6                ; m6 = src[3-10]
-    pmaddwd     m6, m0
-    phaddd      m5, m6
-    phaddd      m4, m5
-    paddd       m4, m1
-
-    palignr     m5, m3, m2, 8                ; m5 = src[4-11]
-    pmaddwd     m5, m0
-    palignr     m6, m3, m2, 10               ; m6 = src[5-12]
-    pmaddwd     m6, m0
-    phaddd      m5, m6
-
-    palignr     m6, m3, m2, 12               ; m6 = src[6-13]
-    pmaddwd     m6, m0
-    palignr     m7, m3, m2, 14               ; m2 = src[7-14]
-    pmaddwd     m7, m0
-    phaddd      m6, m7
-    phaddd      m5, m6
-    paddd       m5, m1
-%ifidn %3, pp
-    psrad       m4, INTERP_SHIFT_PP
-    psrad       m5, INTERP_SHIFT_PP
-    packusdw    m4, m5
-    pxor        m5, m5
-    CLIPW       m4, m5, [pw_pixel_max]
-%else
-    psrad       m4, INTERP_SHIFT_PS
-    psrad       m5, INTERP_SHIFT_PS
-    packssdw    m4, m5
-%endif
-    movu        [r2 + x], m4
-
-    movu        m2, [r0 + 32 + x]            ; m2 = src[16-23]
-
-    pmaddwd     m4, m3, m0                   ; m3 = src[8-15]
-    palignr     m5, m2, m3, 2                ; m5 = src[9-16]
-    pmaddwd     m5, m0
-    phaddd      m4, m5
-
-    palignr     m5, m2, m3, 4                ; m5 = src[10-17]
-    pmaddwd     m5, m0
-    palignr     m6, m2, m3, 6                ; m6 = src[11-18]
-    pmaddwd     m6, m0
-    phaddd      m5, m6
-    phaddd      m4, m5
-    paddd       m4, m1
-
-    palignr     m5, m2, m3, 8                ; m5 = src[12-19]
-    pmaddwd     m5, m0
-    palignr     m6, m2, m3, 10               ; m6 = src[13-20]
-    pmaddwd     m6, m0
-    phaddd      m5, m6
-
-    palignr     m6, m2, m3, 12               ; m6 = src[14-21]
-    pmaddwd     m6, m0
-    palignr     m2, m3, 14                   ; m3 = src[15-22]
-    pmaddwd     m2, m0
-    phaddd      m6, m2
-    phaddd      m5, m6
-    paddd       m5, m1
-%ifidn %3, pp
-    psrad       m4, INTERP_SHIFT_PP
-    psrad       m5, INTERP_SHIFT_PP
-    packusdw    m4, m5
-    pxor        m5, m5
-    CLIPW       m4, m5, [pw_pixel_max]
-%else
-    psrad       m4, INTERP_SHIFT_PS
-    psrad       m5, INTERP_SHIFT_PS
-    packssdw    m4, m5
-%endif
-    movu        [r2 + 16 + x], m4
-
-%assign x x+32
-%endrep
-
-    add         r0, r1
-    add         r2, r3
-
-    dec         r4d
-    jnz         .loopH
-    RET
-%endmacro
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_pp_16x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-FILTER_HOR_LUMA_W16 16, 4, pp
-FILTER_HOR_LUMA_W16 16, 8, pp
-FILTER_HOR_LUMA_W16 16, 12, pp
-FILTER_HOR_LUMA_W16 16, 16, pp
-FILTER_HOR_LUMA_W16 16, 32, pp
-FILTER_HOR_LUMA_W16 16, 64, pp
-
-;----------------------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_ps_16x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;----------------------------------------------------------------------------------------------------------------------------
-FILTER_HOR_LUMA_W16 16, 4, ps
-FILTER_HOR_LUMA_W16 16, 8, ps
-FILTER_HOR_LUMA_W16 16, 12, ps
-FILTER_HOR_LUMA_W16 16, 16, ps
-FILTER_HOR_LUMA_W16 16, 32, ps
-FILTER_HOR_LUMA_W16 16, 64, ps
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_pp_32x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-FILTER_HOR_LUMA_W16 32, 8, pp
-FILTER_HOR_LUMA_W16 32, 16, pp
-FILTER_HOR_LUMA_W16 32, 24, pp
-FILTER_HOR_LUMA_W16 32, 32, pp
-FILTER_HOR_LUMA_W16 32, 64, pp
-
-;----------------------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_ps_32x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;----------------------------------------------------------------------------------------------------------------------------
-FILTER_HOR_LUMA_W16 32, 8, ps
-FILTER_HOR_LUMA_W16 32, 16, ps
-FILTER_HOR_LUMA_W16 32, 24, ps
-FILTER_HOR_LUMA_W16 32, 32, ps
-FILTER_HOR_LUMA_W16 32, 64, ps
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_pp_48x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-FILTER_HOR_LUMA_W16 48, 64, pp
-
-;----------------------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_ps_48x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;----------------------------------------------------------------------------------------------------------------------------
-FILTER_HOR_LUMA_W16 48, 64, ps
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_pp_64x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-FILTER_HOR_LUMA_W16 64, 16, pp
-FILTER_HOR_LUMA_W16 64, 32, pp
-FILTER_HOR_LUMA_W16 64, 48, pp
-FILTER_HOR_LUMA_W16 64, 64, pp
-
-;----------------------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_ps_64x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;----------------------------------------------------------------------------------------------------------------------------
-FILTER_HOR_LUMA_W16 64, 16, ps
-FILTER_HOR_LUMA_W16 64, 32, ps
-FILTER_HOR_LUMA_W16 64, 48, ps
-FILTER_HOR_LUMA_W16 64, 64, ps
-
-;--------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-%macro FILTER_HOR_LUMA_W24 3
-INIT_XMM sse4
-cglobal interp_8tap_horiz_%3_%1x%2, 4, 7, 8
-
-    add         r1, r1
-    add         r3, r3
-    mov         r4d, r4m
-    sub         r0, 6
-    shl         r4d, 4
-
-%ifdef PIC
-    lea         r6, [tab_LumaCoeff]
-    mova        m0, [r6 + r4]
-%else
-    mova        m0, [tab_LumaCoeff + r4]
-%endif
-%ifidn %3, pp
-    mova        m1, [pd_32]
-%else
-    mova        m1, [INTERP_OFFSET_PS]
-%endif
-
-    mov         r4d, %2
-%ifidn %3, ps
-    cmp         r5m, byte 0
-    je          .loopH
-    lea         r6, [r1 + 2 * r1]
-    sub         r0, r6
-    add         r4d, 7
-%endif
-
-.loopH:
-    movu        m2, [r0]                     ; m2 = src[0-7]
-    movu        m3, [r0 + 16]                ; m3 = src[8-15]
-
-    pmaddwd     m4, m2, m0
-    palignr     m5, m3, m2, 2                ; m5 = src[1-8]
-    pmaddwd     m5, m0
-    phaddd      m4, m5
-
-    palignr     m5, m3, m2, 4                ; m5 = src[2-9]
-    pmaddwd     m5, m0
-    palignr     m6, m3, m2, 6                ; m6 = src[3-10]
-    pmaddwd     m6, m0
-    phaddd      m5, m6
-    phaddd      m4, m5
-    paddd       m4, m1
-
-    palignr     m5, m3, m2, 8                ; m5 = src[4-11]
-    pmaddwd     m5, m0
-    palignr     m6, m3, m2, 10               ; m6 = src[5-12]
-    pmaddwd     m6, m0
-    phaddd      m5, m6
-
-    palignr     m6, m3, m2, 12               ; m6 = src[6-13]
-    pmaddwd     m6, m0
-    palignr     m7, m3, m2, 14               ; m7 = src[7-14]
-    pmaddwd     m7, m0
-    phaddd      m6, m7
-    phaddd      m5, m6
-    paddd       m5, m1
-%ifidn %3, pp
-    psrad       m4, INTERP_SHIFT_PP
-    psrad       m5, INTERP_SHIFT_PP
-    packusdw    m4, m5
-    pxor        m5, m5
-    CLIPW       m4, m5, [pw_pixel_max]
-%else
-    psrad       m4, INTERP_SHIFT_PS
-    psrad       m5, INTERP_SHIFT_PS
-    packssdw    m4, m5
-%endif
-    movu        [r2], m4
-
-    movu        m2, [r0 + 32]                ; m2 = src[16-23]
-
-    pmaddwd     m4, m3, m0                   ; m3 = src[8-15]
-    palignr     m5, m2, m3, 2                ; m5 = src[1-8]
-    pmaddwd     m5, m0
-    phaddd      m4, m5
-
-    palignr     m5, m2, m3, 4                ; m5 = src[2-9]
-    pmaddwd     m5, m0
-    palignr     m6, m2, m3, 6                ; m6 = src[3-10]
-    pmaddwd     m6, m0
-    phaddd      m5, m6
-    phaddd      m4, m5
-    paddd       m4, m1
-
-    palignr     m5, m2, m3, 8                ; m5 = src[4-11]
-    pmaddwd     m5, m0
-    palignr     m6, m2, m3, 10               ; m6 = src[5-12]
-    pmaddwd     m6, m0
-    phaddd      m5, m6
-
-    palignr     m6, m2, m3, 12               ; m6 = src[6-13]
-    pmaddwd     m6, m0
-    palignr     m7, m2, m3, 14               ; m7 = src[7-14]
-    pmaddwd     m7, m0
-    phaddd      m6, m7
-    phaddd      m5, m6
-    paddd       m5, m1
-%ifidn %3, pp
-    psrad       m4, INTERP_SHIFT_PP
-    psrad       m5, INTERP_SHIFT_PP
-    packusdw    m4, m5
-    pxor        m5, m5
-    CLIPW       m4, m5, [pw_pixel_max]
-%else
-    psrad       m4, INTERP_SHIFT_PS
-    psrad       m5, INTERP_SHIFT_PS
-    packssdw    m4, m5
-%endif
-    movu        [r2 + 16], m4
-
-    movu        m3, [r0 + 48]                ; m3 = src[24-31]
-
-    pmaddwd     m4, m2, m0                   ; m2 = src[16-23]
-    palignr     m5, m3, m2, 2                ; m5 = src[1-8]
-    pmaddwd     m5, m0
-    phaddd      m4, m5
-
-    palignr     m5, m3, m2, 4                ; m5 = src[2-9]
-    pmaddwd     m5, m0
-    palignr     m6, m3, m2, 6                ; m6 = src[3-10]
-    pmaddwd     m6, m0
-    phaddd      m5, m6
-    phaddd      m4, m5
-    paddd       m4, m1
-
-    palignr     m5, m3, m2, 8                ; m5 = src[4-11]
-    pmaddwd     m5, m0
-    palignr     m6, m3, m2, 10               ; m6 = src[5-12]
-    pmaddwd     m6, m0
-    phaddd      m5, m6
-
-    palignr     m6, m3, m2, 12               ; m6 = src[6-13]
-    pmaddwd     m6, m0
-    palignr     m7, m3, m2, 14               ; m7 = src[7-14]
-    pmaddwd     m7, m0
-    phaddd      m6, m7
-    phaddd      m5, m6
-    paddd       m5, m1
-%ifidn %3, pp
-    psrad       m4, INTERP_SHIFT_PP
-    psrad       m5, INTERP_SHIFT_PP
-    packusdw    m4, m5
-    pxor        m5, m5
-    CLIPW       m4, m5, [pw_pixel_max]
-%else
-    psrad       m4, INTERP_SHIFT_PS
-    psrad       m5, INTERP_SHIFT_PS
-    packssdw    m4, m5
-%endif
-    movu        [r2 + 32], m4
-
-    add         r0, r1
-    add         r2, r3
-
-    dec         r4d
-    jnz         .loopH
-    RET
-%endmacro
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_pp_24x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-FILTER_HOR_LUMA_W24 24, 32, pp
-
-;----------------------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_ps_24x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;----------------------------------------------------------------------------------------------------------------------------
-FILTER_HOR_LUMA_W24 24, 32, ps
-
-%macro FILTER_W2_2 1
-    movu        m3,         [r0]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + r1]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    packusdw    m3,         m3
-    CLIPW       m3,         m7,    m6
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    packssdw    m3,         m3
-%endif
-    movd        [r2],       m3
-    pextrd      [r2 + r3],  m3, 1
-%endmacro
-
-%macro FILTER_W4_2 1
-    movu        m3,         [r0]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 4]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + r1]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + r1 + 4]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m7,    m6
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2],       m3
-    movhps      [r2 + r3],  m3
-%endmacro
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_pp(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-%macro FILTER_HOR_LUMA_W4_avx2 1
-INIT_YMM avx2
-cglobal interp_8tap_horiz_pp_4x%1, 4,7,7
-    add              r1d, r1d
-    add              r3d, r3d
-    sub              r0, 6
-    mov              r4d, r4m
-    shl              r4d, 4
-%ifdef PIC
-    lea              r5, [tab_LumaCoeff]
-    vpbroadcastq     m0, [r5 + r4]
-    vpbroadcastq     m1, [r5 + r4 + 8]
-%else
-    vpbroadcastq     m0, [tab_LumaCoeff + r4]
-    vpbroadcastq     m1, [tab_LumaCoeff + r4 + 8]
-%endif
-    lea              r6, [pw_pixel_max]
-    mova             m3, [interp8_hpp_shuf]
-    mova             m6, [pd_32]
-    pxor             m2, m2
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    mov              r4d, %1/2
-
-.loop:
-    vbroadcasti128   m4, [r0]
-    vbroadcasti128   m5, [r0 + 8]
-    pshufb           m4, m3
-    pshufb           m5, m3
-
-    pmaddwd          m4, m0
-    pmaddwd          m5, m1
-    paddd            m4, m5
-
-    phaddd           m4, m4
-    vpermq           m4, m4, q3120
-    paddd            m4, m6
-    psrad            m4, INTERP_SHIFT_PP
-
-    packusdw         m4, m4
-    vpermq           m4, m4, q2020
-    CLIPW            m4, m2, [r6]
-    movq             [r2], xm4
-
-    vbroadcasti128   m4, [r0 + r1]
-    vbroadcasti128   m5, [r0 + r1 + 8]
-    pshufb           m4, m3
-    pshufb           m5, m3
-
-    pmaddwd          m4, m0
-    pmaddwd          m5, m1
-    paddd            m4, m5
-
-    phaddd           m4, m4
-    vpermq           m4, m4, q3120
-    paddd            m4, m6
-    psrad            m4, INTERP_SHIFT_PP
-
-    packusdw         m4, m4
-    vpermq           m4, m4, q2020
-    CLIPW            m4, m2, [r6]
-    movq             [r2 + r3], xm4
-
-    lea              r2, [r2 + 2 * r3]
-    lea              r0, [r0 + 2 * r1]
-    dec              r4d
-    jnz              .loop
-    RET
-%endmacro
-FILTER_HOR_LUMA_W4_avx2 4
-FILTER_HOR_LUMA_W4_avx2 8
-FILTER_HOR_LUMA_W4_avx2 16
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_pp(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-%macro FILTER_HOR_LUMA_W8 1
-INIT_YMM avx2
-cglobal interp_8tap_horiz_pp_8x%1, 4,6,8
-    add              r1d, r1d
-    add              r3d, r3d
-    sub              r0, 6
-    mov              r4d, r4m
-    shl              r4d, 4
-%ifdef PIC
-    lea              r5, [tab_LumaCoeff]
-    vpbroadcastq     m0, [r5 + r4]
-    vpbroadcastq     m1, [r5 + r4 + 8]
-%else
-    vpbroadcastq     m0, [tab_LumaCoeff + r4]
-    vpbroadcastq     m1, [tab_LumaCoeff + r4 + 8]
-%endif
-    mova             m3, [interp8_hpp_shuf]
-    mova             m7, [pd_32]
-    pxor             m2, m2
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    mov              r4d, %1/2
-
-.loop:
-    vbroadcasti128   m4, [r0]
-    vbroadcasti128   m5, [r0 + 8]
-    pshufb           m4, m3
-    pshufb           m5, m3
-
-    pmaddwd          m4, m0
-    pmaddwd          m5, m1
-    paddd            m4, m5
-
-    vbroadcasti128   m5, [r0 + 8]
-    vbroadcasti128   m6, [r0 + 16]
-    pshufb           m5, m3
-    pshufb           m6, m3
-
-    pmaddwd          m5, m0
-    pmaddwd          m6, m1
-    paddd            m5, m6
-
-    phaddd           m4, m5
-    vpermq           m4, m4, q3120
-    paddd            m4, m7
-    psrad            m4, INTERP_SHIFT_PP
-
-    packusdw         m4, m4
-    vpermq           m4, m4, q2020
-    CLIPW            m4, m2, [pw_pixel_max]
-    movu             [r2], xm4
-
-    vbroadcasti128   m4, [r0 + r1]
-    vbroadcasti128   m5, [r0 + r1 + 8]
-    pshufb           m4, m3
-    pshufb           m5, m3
-
-    pmaddwd          m4, m0
-    pmaddwd          m5, m1
-    paddd            m4, m5
-
-    vbroadcasti128   m5, [r0 + r1 + 8]
-    vbroadcasti128   m6, [r0 + r1 + 16]
-    pshufb           m5, m3
-    pshufb           m6, m3
-
-    pmaddwd          m5, m0
-    pmaddwd          m6, m1
-    paddd            m5, m6
-
-    phaddd           m4, m5
-    vpermq           m4, m4, q3120
-    paddd            m4, m7
-    psrad            m4, INTERP_SHIFT_PP
-
-    packusdw         m4, m4
-    vpermq           m4, m4, q2020
-    CLIPW            m4, m2, [pw_pixel_max]
-    movu             [r2 + r3], xm4
-
-    lea              r2, [r2 + 2 * r3]
-    lea              r0, [r0 + 2 * r1]
-    dec              r4d
-    jnz              .loop
-    RET
-%endmacro
-FILTER_HOR_LUMA_W8 4
-FILTER_HOR_LUMA_W8 8
-FILTER_HOR_LUMA_W8 16
-FILTER_HOR_LUMA_W8 32
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_pp(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-%macro FILTER_HOR_LUMA_W16 1
-INIT_YMM avx2
-cglobal interp_8tap_horiz_pp_16x%1, 4,6,8
-    add              r1d, r1d
-    add              r3d, r3d
-    sub              r0, 6
-    mov              r4d, r4m
-    shl              r4d, 4
-%ifdef PIC
-    lea              r5, [tab_LumaCoeff]
-    vpbroadcastq     m0, [r5 + r4]
-    vpbroadcastq     m1, [r5 + r4 + 8]
-%else
-    vpbroadcastq     m0, [tab_LumaCoeff + r4]
-    vpbroadcastq     m1, [tab_LumaCoeff + r4 + 8]
-%endif
-    mova             m3, [interp8_hpp_shuf]
-    mova             m7, [pd_32]
-    pxor             m2, m2
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    mov              r4d, %1
-
-.loop:
-    vbroadcasti128   m4, [r0]
-    vbroadcasti128   m5, [r0 + 8]
-    pshufb           m4, m3
-    pshufb           m5, m3
-
-    pmaddwd          m4, m0
-    pmaddwd          m5, m1
-    paddd            m4, m5
-
-    vbroadcasti128   m5, [r0 + 8]
-    vbroadcasti128   m6, [r0 + 16]
-    pshufb           m5, m3
-    pshufb           m6, m3
-
-    pmaddwd          m5, m0
-    pmaddwd          m6, m1
-    paddd            m5, m6
-
-    phaddd           m4, m5
-    vpermq           m4, m4, q3120
-    paddd            m4, m7
-    psrad            m4, INTERP_SHIFT_PP
-
-    packusdw         m4, m4
-    vpermq           m4, m4, q2020
-    CLIPW            m4, m2, [pw_pixel_max]
-    movu             [r2], xm4
-
-    vbroadcasti128   m4, [r0 + 16]
-    vbroadcasti128   m5, [r0 + 24]
-    pshufb           m4, m3
-    pshufb           m5, m3
-
-    pmaddwd          m4, m0
-    pmaddwd          m5, m1
-    paddd            m4, m5
-
-    vbroadcasti128   m5, [r0 + 24]
-    vbroadcasti128   m6, [r0 + 32]
-    pshufb           m5, m3
-    pshufb           m6, m3
-
-    pmaddwd          m5, m0
-    pmaddwd          m6, m1
-    paddd            m5, m6
-
-    phaddd           m4, m5
-    vpermq           m4, m4, q3120
-    paddd            m4, m7
-    psrad            m4, INTERP_SHIFT_PP
-
-    packusdw         m4, m4
-    vpermq           m4, m4, q2020
-    CLIPW            m4, m2, [pw_pixel_max]
-    movu             [r2 + 16], xm4
-
-    add              r2, r3
-    add              r0, r1
-    dec              r4d
-    jnz              .loop
-    RET
-%endmacro
-FILTER_HOR_LUMA_W16 4
-FILTER_HOR_LUMA_W16 8
-FILTER_HOR_LUMA_W16 12
-FILTER_HOR_LUMA_W16 16
-FILTER_HOR_LUMA_W16 32
-FILTER_HOR_LUMA_W16 64
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_pp(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-%macro FILTER_HOR_LUMA_W32 2
-INIT_YMM avx2
-cglobal interp_8tap_horiz_pp_%1x%2, 4,6,8
-    add              r1d, r1d
-    add              r3d, r3d
-    sub              r0, 6
-    mov              r4d, r4m
-    shl              r4d, 4
-%ifdef PIC
-    lea              r5, [tab_LumaCoeff]
-    vpbroadcastq     m0, [r5 + r4]
-    vpbroadcastq     m1, [r5 + r4 + 8]
-%else
-    vpbroadcastq     m0, [tab_LumaCoeff + r4]
-    vpbroadcastq     m1, [tab_LumaCoeff + r4 + 8]
-%endif
-    mova             m3, [interp8_hpp_shuf]
-    mova             m7, [pd_32]
-    pxor             m2, m2
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    mov              r4d, %2
-
-.loop:
-%assign x 0
-%rep %1/16
-    vbroadcasti128   m4, [r0 + x]
-    vbroadcasti128   m5, [r0 + 8 + x]
-    pshufb           m4, m3
-    pshufb           m5, m3
-
-    pmaddwd          m4, m0
-    pmaddwd          m5, m1
-    paddd            m4, m5
-
-    vbroadcasti128   m5, [r0 + 8 + x]
-    vbroadcasti128   m6, [r0 + 16 + x]
-    pshufb           m5, m3
-    pshufb           m6, m3
-
-    pmaddwd          m5, m0
-    pmaddwd          m6, m1
-    paddd            m5, m6
-
-    phaddd           m4, m5
-    vpermq           m4, m4, q3120
-    paddd            m4, m7
-    psrad            m4, INTERP_SHIFT_PP
-
-    packusdw         m4, m4
-    vpermq           m4, m4, q2020
-    CLIPW            m4, m2, [pw_pixel_max]
-    movu             [r2 + x], xm4
-
-    vbroadcasti128   m4, [r0 + 16 + x]
-    vbroadcasti128   m5, [r0 + 24 + x]
-    pshufb           m4, m3
-    pshufb           m5, m3
-
-    pmaddwd          m4, m0
-    pmaddwd          m5, m1
-    paddd            m4, m5
-
-    vbroadcasti128   m5, [r0 + 24 + x]
-    vbroadcasti128   m6, [r0 + 32 + x]
-    pshufb           m5, m3
-    pshufb           m6, m3
-
-    pmaddwd          m5, m0
-    pmaddwd          m6, m1
-    paddd            m5, m6
-
-    phaddd           m4, m5
-    vpermq           m4, m4, q3120
-    paddd            m4, m7
-    psrad            m4, INTERP_SHIFT_PP
-
-    packusdw         m4, m4
-    vpermq           m4, m4, q2020
-    CLIPW            m4, m2, [pw_pixel_max]
-    movu             [r2 + 16 + x], xm4
-
-%assign x x+32
-%endrep
-
-    add              r2, r3
-    add              r0, r1
-    dec              r4d
-    jnz              .loop
-    RET
-%endmacro
-FILTER_HOR_LUMA_W32 32, 8
-FILTER_HOR_LUMA_W32 32, 16
-FILTER_HOR_LUMA_W32 32, 24
-FILTER_HOR_LUMA_W32 32, 32
-FILTER_HOR_LUMA_W32 32, 64
-FILTER_HOR_LUMA_W32 64, 16
-FILTER_HOR_LUMA_W32 64, 32
-FILTER_HOR_LUMA_W32 64, 48
-FILTER_HOR_LUMA_W32 64, 64
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_pp(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-cglobal interp_8tap_horiz_pp_12x16, 4,6,8
-    add              r1d, r1d
-    add              r3d, r3d
-    sub              r0, 6
-    mov              r4d, r4m
-    shl              r4d, 4
-%ifdef PIC
-    lea              r5, [tab_LumaCoeff]
-    vpbroadcastq     m0, [r5 + r4]
-    vpbroadcastq     m1, [r5 + r4 + 8]
-%else
-    vpbroadcastq     m0, [tab_LumaCoeff + r4]
-    vpbroadcastq     m1, [tab_LumaCoeff + r4 + 8]
-%endif
-    mova             m3, [interp8_hpp_shuf]
-    mova             m7, [pd_32]
-    pxor             m2, m2
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    mov              r4d, 16
-
-.loop:
-    vbroadcasti128   m4, [r0]
-    vbroadcasti128   m5, [r0 + 8]
-    pshufb           m4, m3
-    pshufb           m5, m3
-
-    pmaddwd          m4, m0
-    pmaddwd          m5, m1
-    paddd            m4, m5
-
-    vbroadcasti128   m5, [r0 + 8]
-    vbroadcasti128   m6, [r0 + 16]
-    pshufb           m5, m3
-    pshufb           m6, m3
-
-    pmaddwd          m5, m0
-    pmaddwd          m6, m1
-    paddd            m5, m6
-
-    phaddd           m4, m5
-    vpermq           m4, m4, q3120
-    paddd            m4, m7
-    psrad            m4, INTERP_SHIFT_PP
-
-    packusdw         m4, m4
-    vpermq           m4, m4, q2020
-    CLIPW            m4, m2, [pw_pixel_max]
-    movu             [r2], xm4
-
-    vbroadcasti128   m4, [r0 + 16]
-    vbroadcasti128   m5, [r0 + 24]
-    pshufb           m4, m3
-    pshufb           m5, m3
-
-    pmaddwd          m4, m0
-    pmaddwd          m5, m1
-    paddd            m4, m5
-
-    vbroadcasti128   m5, [r0 + 24]
-    vbroadcasti128   m6, [r0 + 32]
-    pshufb           m5, m3
-    pshufb           m6, m3
-
-    pmaddwd          m5, m0
-    pmaddwd          m6, m1
-    paddd            m5, m6
-
-    phaddd           m4, m5
-    vpermq           m4, m4, q3120
-    paddd            m4, m7
-    psrad            m4, INTERP_SHIFT_PP
-
-    packusdw         m4, m4
-    vpermq           m4, m4, q2020
-    CLIPW            m4, m2, [pw_pixel_max]
-    movq             [r2 + 16], xm4
-
-    add              r2, r3
-    add              r0, r1
-    dec              r4d
-    jnz              .loop
-    RET
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_pp(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-cglobal interp_8tap_horiz_pp_24x32, 4,6,8
-    add              r1d, r1d
-    add              r3d, r3d
-    sub              r0, 6
-    mov              r4d, r4m
-    shl              r4d, 4
-%ifdef PIC
-    lea              r5, [tab_LumaCoeff]
-    vpbroadcastq     m0, [r5 + r4]
-    vpbroadcastq     m1, [r5 + r4 + 8]
-%else
-    vpbroadcastq     m0, [tab_LumaCoeff + r4]
-    vpbroadcastq     m1, [tab_LumaCoeff + r4 + 8]
-%endif
-    mova             m3, [interp8_hpp_shuf]
-    mova             m7, [pd_32]
-    pxor             m2, m2
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    mov              r4d, 32
-
-.loop:
-    vbroadcasti128   m4, [r0]
-    vbroadcasti128   m5, [r0 + 8]
-    pshufb           m4, m3
-    pshufb           m5, m3
-
-    pmaddwd          m4, m0
-    pmaddwd          m5, m1
-    paddd            m4, m5
-
-    vbroadcasti128   m5, [r0 + 8]
-    vbroadcasti128   m6, [r0 + 16]
-    pshufb           m5, m3
-    pshufb           m6, m3
-
-    pmaddwd          m5, m0
-    pmaddwd          m6, m1
-    paddd            m5, m6
-
-    phaddd           m4, m5
-    vpermq           m4, m4, q3120
-    paddd            m4, m7
-    psrad            m4, INTERP_SHIFT_PP
-
-    packusdw         m4, m4
-    vpermq           m4, m4, q2020
-    CLIPW            m4, m2, [pw_pixel_max]
-    movu             [r2], xm4
-
-    vbroadcasti128   m4, [r0 + 16]
-    vbroadcasti128   m5, [r0 + 24]
-    pshufb           m4, m3
-    pshufb           m5, m3
-
-    pmaddwd          m4, m0
-    pmaddwd          m5, m1
-    paddd            m4, m5
-
-    vbroadcasti128   m5, [r0 + 24]
-    vbroadcasti128   m6, [r0 + 32]
-    pshufb           m5, m3
-    pshufb           m6, m3
-
-    pmaddwd          m5, m0
-    pmaddwd          m6, m1
-    paddd            m5, m6
-
-    phaddd           m4, m5
-    vpermq           m4, m4, q3120
-    paddd            m4, m7
-    psrad            m4, INTERP_SHIFT_PP
-
-    packusdw         m4, m4
-    vpermq           m4, m4, q2020
-    CLIPW            m4, m2, [pw_pixel_max]
-    movu             [r2 + 16], xm4
-
-    vbroadcasti128   m4, [r0 + 32]
-    vbroadcasti128   m5, [r0 + 40]
-    pshufb           m4, m3
-    pshufb           m5, m3
-
-    pmaddwd          m4, m0
-    pmaddwd          m5, m1
-    paddd            m4, m5
-
-    vbroadcasti128   m5, [r0 + 40]
-    vbroadcasti128   m6, [r0 + 48]
-    pshufb           m5, m3
-    pshufb           m6, m3
-
-    pmaddwd          m5, m0
-    pmaddwd          m6, m1
-    paddd            m5, m6
-
-    phaddd           m4, m5
-    vpermq           m4, m4, q3120
-    paddd            m4, m7
-    psrad            m4, INTERP_SHIFT_PP
-
-    packusdw         m4, m4
-    vpermq           m4, m4, q2020
-    CLIPW            m4, m2, [pw_pixel_max]
-    movu             [r2 + 32], xm4
-
-    add              r2, r3
-    add              r0, r1
-    dec              r4d
-    jnz              .loop
-    RET
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_horiz_pp(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-cglobal interp_8tap_horiz_pp_48x64, 4,6,8
-    add              r1d, r1d
-    add              r3d, r3d
-    sub              r0, 6
-    mov              r4d, r4m
-    shl              r4d, 4
-%ifdef PIC
-    lea              r5, [tab_LumaCoeff]
-    vpbroadcastq     m0, [r5 + r4]
-    vpbroadcastq     m1, [r5 + r4 + 8]
-%else
-    vpbroadcastq     m0, [tab_LumaCoeff + r4]
-    vpbroadcastq     m1, [tab_LumaCoeff + r4 + 8]
-%endif
-    mova             m3, [interp8_hpp_shuf]
-    mova             m7, [pd_32]
-    pxor             m2, m2
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    mov              r4d, 64
-
-.loop:
-%assign x 0
-%rep 2
-    vbroadcasti128   m4, [r0 + x]
-    vbroadcasti128   m5, [r0 + 8 + x]
-    pshufb           m4, m3
-    pshufb           m5, m3
-
-    pmaddwd          m4, m0
-    pmaddwd          m5, m1
-    paddd            m4, m5
-
-    vbroadcasti128   m5, [r0 + 8 + x]
-    vbroadcasti128   m6, [r0 + 16 + x]
-    pshufb           m5, m3
-    pshufb           m6, m3
-
-    pmaddwd          m5, m0
-    pmaddwd          m6, m1
-    paddd            m5, m6
-
-    phaddd           m4, m5
-    vpermq           m4, m4, q3120
-    paddd            m4, m7
-    psrad            m4, INTERP_SHIFT_PP
-
-    packusdw         m4, m4
-    vpermq           m4, m4, q2020
-    CLIPW            m4, m2, [pw_pixel_max]
-    movu             [r2 + x], xm4
-
-    vbroadcasti128   m4, [r0 + 16 + x]
-    vbroadcasti128   m5, [r0 + 24 + x]
-    pshufb           m4, m3
-    pshufb           m5, m3
-
-    pmaddwd          m4, m0
-    pmaddwd          m5, m1
-    paddd            m4, m5
-
-    vbroadcasti128   m5, [r0 + 24 + x]
-    vbroadcasti128   m6, [r0 + 32 + x]
-    pshufb           m5, m3
-    pshufb           m6, m3
-
-    pmaddwd          m5, m0
-    pmaddwd          m6, m1
-    paddd            m5, m6
-
-    phaddd           m4, m5
-    vpermq           m4, m4, q3120
-    paddd            m4, m7
-    psrad            m4, INTERP_SHIFT_PP
-
-    packusdw         m4, m4
-    vpermq           m4, m4, q2020
-    CLIPW            m4, m2, [pw_pixel_max]
-    movu             [r2 + 16 + x], xm4
-
-    vbroadcasti128   m4, [r0 + 32 + x]
-    vbroadcasti128   m5, [r0 + 40 + x]
-    pshufb           m4, m3
-    pshufb           m5, m3
-
-    pmaddwd          m4, m0
-    pmaddwd          m5, m1
-    paddd            m4, m5
-
-    vbroadcasti128   m5, [r0 + 40 + x]
-    vbroadcasti128   m6, [r0 + 48 + x]
-    pshufb           m5, m3
-    pshufb           m6, m3
-
-    pmaddwd          m5, m0
-    pmaddwd          m6, m1
-    paddd            m5, m6
-
-    phaddd           m4, m5
-    vpermq           m4, m4, q3120
-    paddd            m4, m7
-    psrad            m4, INTERP_SHIFT_PP
-
-    packusdw         m4, m4
-    vpermq           m4, m4, q2020
-    CLIPW            m4, m2, [pw_pixel_max]
-    movu             [r2 + 32 + x], xm4
-
-%assign x x+48
-%endrep
-
-    add              r2, r3
-    add              r0, r1
-    dec              r4d
-    jnz              .loop
-    RET
-
-;-----------------------------------------------------------------------------
-; void interp_4tap_horiz_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro FILTER_CHROMA_H 6
-INIT_XMM sse4
-cglobal interp_4tap_horiz_%3_%1x%2, 4, %4, %5
-
-    add         r3,       r3
-    add         r1,       r1
-    sub         r0,       2
-    mov         r4d,      r4m
-    add         r4d,      r4d
-
-%ifdef PIC
-    lea         r%6,      [tab_ChromaCoeff]
-    movh        m0,       [r%6 + r4 * 4]
-%else
-    movh        m0,       [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    punpcklqdq  m0,       m0
-    mova        m2,       [tab_Tm16]
-
-%ifidn %3, ps
-    mova        m1,       [INTERP_OFFSET_PS]
-    cmp         r5m, byte 0
-    je          .skip
-    sub         r0,       r1
-    movu        m3,       [r0]
-    pshufb      m3,       m3, m2
-    pmaddwd     m3,       m0
-
-  %if %1 == 4
-    movu        m4,       [r0 + 4]
-    pshufb      m4,       m4, m2
-    pmaddwd     m4,       m0
-    phaddd      m3,       m4
-  %else
-    phaddd      m3,       m3
-  %endif
-
-    paddd       m3,       m1
-    psrad       m3,       INTERP_SHIFT_PS
-    packssdw    m3,       m3
-
-  %if %1 == 2
-    movd        [r2],     m3
-  %else
-    movh        [r2],     m3
-  %endif
-
-    add         r0,       r1
-    add         r2,       r3
-    FILTER_W%1_2 %3
-    lea         r0,       [r0 + 2 * r1]
-    lea         r2,       [r2 + 2 * r3]
-
-.skip:
-
-%else     ;%ifidn %3, ps
-    pxor        m7,       m7
-    mova        m6,       [pw_pixel_max]
-    mova        m1,       [tab_c_32]
-%endif    ;%ifidn %3, ps
-
-    FILTER_W%1_2 %3
-
-%rep (%2/2) - 1
-    lea         r0,       [r0 + 2 * r1]
-    lea         r2,       [r2 + 2 * r3]
-    FILTER_W%1_2 %3
-%endrep
-    RET
-%endmacro
-
-FILTER_CHROMA_H 2, 4, pp, 6, 8, 5
-FILTER_CHROMA_H 2, 8, pp, 6, 8, 5
-FILTER_CHROMA_H 4, 2, pp, 6, 8, 5
-FILTER_CHROMA_H 4, 4, pp, 6, 8, 5
-FILTER_CHROMA_H 4, 8, pp, 6, 8, 5
-FILTER_CHROMA_H 4, 16, pp, 6, 8, 5
-
-FILTER_CHROMA_H 2, 4, ps, 7, 5, 6
-FILTER_CHROMA_H 2, 8, ps, 7, 5, 6
-FILTER_CHROMA_H 4, 2, ps, 7, 6, 6
-FILTER_CHROMA_H 4, 4, ps, 7, 6, 6
-FILTER_CHROMA_H 4, 8, ps, 7, 6, 6
-FILTER_CHROMA_H 4, 16, ps, 7, 6, 6
-
-FILTER_CHROMA_H 2, 16, pp, 6, 8, 5
-FILTER_CHROMA_H 4, 32, pp, 6, 8, 5
-FILTER_CHROMA_H 2, 16, ps, 7, 5, 6
-FILTER_CHROMA_H 4, 32, ps, 7, 6, 6
-
-
-%macro FILTER_W6_1 1
-    movu        m3,         [r0]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 4]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m4,         [r0 + 8]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m4,         m4
-    paddd       m4,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m4,         INTERP_SHIFT_PP
-    packusdw    m3,         m4
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m4,         INTERP_SHIFT_PS
-    packssdw    m3,         m4
-%endif
-    movh        [r2],       m3
-    pextrd      [r2 + 8],   m3, 2
-%endmacro
-
-cglobal chroma_filter_pp_6x1_internal
-    FILTER_W6_1 pp
-    ret
-
-cglobal chroma_filter_ps_6x1_internal
-    FILTER_W6_1 ps
-    ret
-
-%macro FILTER_W8_1 1
-    movu        m3,         [r0]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 4]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + 8]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + 12]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2],       m3
-    movhps      [r2 + 8],   m3
-%endmacro
-
-cglobal chroma_filter_pp_8x1_internal
-    FILTER_W8_1 pp
-    ret
-
-cglobal chroma_filter_ps_8x1_internal
-    FILTER_W8_1 ps
-    ret
-
-%macro FILTER_W12_1 1
-    movu        m3,         [r0]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 4]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + 8]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + 12]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2],       m3
-    movhps      [r2 + 8],   m3
-
-    movu        m3,         [r0 + 16]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 20]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    packusdw    m3,         m3
-    CLIPW       m3,         m6, m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    packssdw    m3,         m3
-%endif
-    movh        [r2 + 16],  m3
-%endmacro
-
-cglobal chroma_filter_pp_12x1_internal
-    FILTER_W12_1 pp
-    ret
-
-cglobal chroma_filter_ps_12x1_internal
-    FILTER_W12_1 ps
-    ret
-
-%macro FILTER_W16_1 1
-    movu        m3,         [r0]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 4]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + 8]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + 12]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2],       m3
-    movhps      [r2 + 8],   m3
-
-    movu        m3,         [r0 + 16]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 20]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + 24]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + 28]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2 + 16],  m3
-    movhps      [r2 + 24],  m3
-%endmacro
-
-cglobal chroma_filter_pp_16x1_internal
-    FILTER_W16_1 pp
-    ret
-
-cglobal chroma_filter_ps_16x1_internal
-    FILTER_W16_1 ps
-    ret
-
-%macro FILTER_W24_1 1
-    movu        m3,         [r0]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 4]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + 8]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + 12]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2],       m3
-    movhps      [r2 + 8],   m3
-
-    movu        m3,         [r0 + 16]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 20]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + 24]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + 28]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2 + 16],  m3
-    movhps      [r2 + 24],  m3
-
-    movu        m3,         [r0 + 32]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 36]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + 40]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + 44]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2 + 32],  m3
-    movhps      [r2 + 40],  m3
-%endmacro
-
-cglobal chroma_filter_pp_24x1_internal
-    FILTER_W24_1 pp
-    ret
-
-cglobal chroma_filter_ps_24x1_internal
-    FILTER_W24_1 ps
-    ret
-
-%macro FILTER_W32_1 1
-    movu        m3,         [r0]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 4]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + 8]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + 12]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2],       m3
-    movhps      [r2 + 8],   m3
-
-    movu        m3,         [r0 + 16]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 20]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + 24]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + 28]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2 + 16],  m3
-    movhps      [r2 + 24],  m3
-
-    movu        m3,         [r0 + 32]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 36]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + 40]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + 44]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2 + 32],  m3
-    movhps      [r2 + 40],  m3
-
-    movu        m3,         [r0 + 48]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 52]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + 56]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + 60]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2 + 48],  m3
-    movhps      [r2 + 56],  m3
-%endmacro
-
-cglobal chroma_filter_pp_32x1_internal
-    FILTER_W32_1 pp
-    ret
-
-cglobal chroma_filter_ps_32x1_internal
-    FILTER_W32_1 ps
-    ret
-
-%macro FILTER_W8o_1 2
-    movu        m3,         [r0 + %2]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + %2 + 4]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + %2 + 8]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + %2 + 12]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2 + %2],       m3
-    movhps      [r2 + %2 + 8],   m3
-%endmacro
-
-%macro FILTER_W48_1 1
-    FILTER_W8o_1 %1, 0
-    FILTER_W8o_1 %1, 16
-    FILTER_W8o_1 %1, 32
-    FILTER_W8o_1 %1, 48
-    FILTER_W8o_1 %1, 64
-    FILTER_W8o_1 %1, 80
-%endmacro
-
-cglobal chroma_filter_pp_48x1_internal
-    FILTER_W48_1 pp
-    ret
-
-cglobal chroma_filter_ps_48x1_internal
-    FILTER_W48_1 ps
-    ret
-
-%macro FILTER_W64_1 1
-    FILTER_W8o_1 %1, 0
-    FILTER_W8o_1 %1, 16
-    FILTER_W8o_1 %1, 32
-    FILTER_W8o_1 %1, 48
-    FILTER_W8o_1 %1, 64
-    FILTER_W8o_1 %1, 80
-    FILTER_W8o_1 %1, 96
-    FILTER_W8o_1 %1, 112
-%endmacro
-
-cglobal chroma_filter_pp_64x1_internal
-    FILTER_W64_1 pp
-    ret
-
-cglobal chroma_filter_ps_64x1_internal
-    FILTER_W64_1 ps
-    ret
-
-
-;-----------------------------------------------------------------------------
-; void interp_4tap_horiz_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-
-INIT_XMM sse4
-%macro IPFILTER_CHROMA 6
-cglobal interp_4tap_horiz_%3_%1x%2, 4, %5, %6
-
-    add         r3,        r3
-    add         r1,        r1
-    sub         r0,         2
-    mov         r4d,        r4m
-    add         r4d,        r4d
-
-%ifdef PIC
-    lea         r%4,       [tab_ChromaCoeff]
-    movh        m0,       [r%4 + r4 * 4]
-%else
-    movh        m0,       [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    punpcklqdq  m0,       m0
-    mova        m2,       [tab_Tm16]
-
-%ifidn %3, ps
-    mova        m1,       [INTERP_OFFSET_PS]
-    cmp         r5m, byte 0
-    je          .skip
-    sub         r0, r1
-    call chroma_filter_%3_%1x1_internal
-    add         r0, r1
-    add         r2, r3
-    call chroma_filter_%3_%1x1_internal
-    add         r0, r1
-    add         r2, r3
-    call chroma_filter_%3_%1x1_internal
-    add         r0, r1
-    add         r2, r3
-.skip:
-%else
-    mova        m1,         [tab_c_32]
-    pxor        m6,         m6
-    mova        m7,         [pw_pixel_max]
-%endif
-
-    call chroma_filter_%3_%1x1_internal
-%rep %2 - 1
-    add         r0,       r1
-    add         r2,       r3
-    call chroma_filter_%3_%1x1_internal
-%endrep
-RET
-%endmacro
-IPFILTER_CHROMA 6, 8, pp, 5, 6, 8
-IPFILTER_CHROMA 8, 2, pp, 5, 6, 8
-IPFILTER_CHROMA 8, 4, pp, 5, 6, 8
-IPFILTER_CHROMA 8, 6, pp, 5, 6, 8
-IPFILTER_CHROMA 8, 8, pp, 5, 6, 8
-IPFILTER_CHROMA 8, 16, pp, 5, 6, 8
-IPFILTER_CHROMA 8, 32, pp, 5, 6, 8
-IPFILTER_CHROMA 12, 16, pp, 5, 6, 8
-IPFILTER_CHROMA 16, 4, pp, 5, 6, 8
-IPFILTER_CHROMA 16, 8, pp, 5, 6, 8
-IPFILTER_CHROMA 16, 12, pp, 5, 6, 8
-IPFILTER_CHROMA 16, 16, pp, 5, 6, 8
-IPFILTER_CHROMA 16, 32, pp, 5, 6, 8
-IPFILTER_CHROMA 24, 32, pp, 5, 6, 8
-IPFILTER_CHROMA 32, 8, pp, 5, 6, 8
-IPFILTER_CHROMA 32, 16, pp, 5, 6, 8
-IPFILTER_CHROMA 32, 24, pp, 5, 6, 8
-IPFILTER_CHROMA 32, 32, pp, 5, 6, 8
-
-IPFILTER_CHROMA 6, 8, ps, 6, 7, 6
-IPFILTER_CHROMA 8, 2, ps, 6, 7, 6
-IPFILTER_CHROMA 8, 4, ps, 6, 7, 6
-IPFILTER_CHROMA 8, 6, ps, 6, 7, 6
-IPFILTER_CHROMA 8, 8, ps, 6, 7, 6
-IPFILTER_CHROMA 8, 16, ps, 6, 7, 6
-IPFILTER_CHROMA 8, 32, ps, 6, 7, 6
-IPFILTER_CHROMA 12, 16, ps, 6, 7, 6
-IPFILTER_CHROMA 16, 4, ps, 6, 7, 6
-IPFILTER_CHROMA 16, 8, ps, 6, 7, 6
-IPFILTER_CHROMA 16, 12, ps, 6, 7, 6
-IPFILTER_CHROMA 16, 16, ps, 6, 7, 6
-IPFILTER_CHROMA 16, 32, ps, 6, 7, 6
-IPFILTER_CHROMA 24, 32, ps, 6, 7, 6
-IPFILTER_CHROMA 32, 8, ps, 6, 7, 6
-IPFILTER_CHROMA 32, 16, ps, 6, 7, 6
-IPFILTER_CHROMA 32, 24, ps, 6, 7, 6
-IPFILTER_CHROMA 32, 32, ps, 6, 7, 6
-
-IPFILTER_CHROMA 6, 16, pp, 5, 6, 8
-IPFILTER_CHROMA 8, 12, pp, 5, 6, 8
-IPFILTER_CHROMA 8, 64, pp, 5, 6, 8
-IPFILTER_CHROMA 12, 32, pp, 5, 6, 8
-IPFILTER_CHROMA 16, 24, pp, 5, 6, 8
-IPFILTER_CHROMA 16, 64, pp, 5, 6, 8
-IPFILTER_CHROMA 24, 64, pp, 5, 6, 8
-IPFILTER_CHROMA 32, 48, pp, 5, 6, 8
-IPFILTER_CHROMA 32, 64, pp, 5, 6, 8
-IPFILTER_CHROMA 6, 16, ps, 6, 7, 6
-IPFILTER_CHROMA 8, 12, ps, 6, 7, 6
-IPFILTER_CHROMA 8, 64, ps, 6, 7, 6
-IPFILTER_CHROMA 12, 32, ps, 6, 7, 6
-IPFILTER_CHROMA 16, 24, ps, 6, 7, 6
-IPFILTER_CHROMA 16, 64, ps, 6, 7, 6
-IPFILTER_CHROMA 24, 64, ps, 6, 7, 6
-IPFILTER_CHROMA 32, 48, ps, 6, 7, 6
-IPFILTER_CHROMA 32, 64, ps, 6, 7, 6
-
-IPFILTER_CHROMA 48, 64, pp, 5, 6, 8
-IPFILTER_CHROMA 64, 48, pp, 5, 6, 8
-IPFILTER_CHROMA 64, 64, pp, 5, 6, 8
-IPFILTER_CHROMA 64, 32, pp, 5, 6, 8
-IPFILTER_CHROMA 64, 16, pp, 5, 6, 8
-IPFILTER_CHROMA 48, 64, ps, 6, 7, 6
-IPFILTER_CHROMA 64, 48, ps, 6, 7, 6
-IPFILTER_CHROMA 64, 64, ps, 6, 7, 6
-IPFILTER_CHROMA 64, 32, ps, 6, 7, 6
-IPFILTER_CHROMA 64, 16, ps, 6, 7, 6
-
-
-%macro PROCESS_CHROMA_SP_W4_4R 0
-    movq       m0, [r0]
-    movq       m1, [r0 + r1]
-    punpcklwd  m0, m1                          ;m0=[0 1]
-    pmaddwd    m0, [r6 + 0 *32]                ;m0=[0+1]         Row1
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m4, [r0]
-    punpcklwd  m1, m4                          ;m1=[1 2]
-    pmaddwd    m1, [r6 + 0 *32]                ;m1=[1+2]         Row2
-
-    movq       m5, [r0 + r1]
-    punpcklwd  m4, m5                          ;m4=[2 3]
-    pmaddwd    m2, m4, [r6 + 0 *32]            ;m2=[2+3]         Row3
-    pmaddwd    m4, [r6 + 1 * 32]
-    paddd      m0, m4                          ;m0=[0+1+2+3]     Row1 done
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m4, [r0]
-    punpcklwd  m5, m4                          ;m5=[3 4]
-    pmaddwd    m3, m5, [r6 + 0 *32]            ;m3=[3+4]         Row4
-    pmaddwd    m5, [r6 + 1 * 32]
-    paddd      m1, m5                          ;m1 = [1+2+3+4]   Row2
-
-    movq       m5, [r0 + r1]
-    punpcklwd  m4, m5                          ;m4=[4 5]
-    pmaddwd    m4, [r6 + 1 * 32]
-    paddd      m2, m4                          ;m2=[2+3+4+5]     Row3
-
-    movq       m4, [r0 + 2 * r1]
-    punpcklwd  m5, m4                          ;m5=[5 6]
-    pmaddwd    m5, [r6 + 1 * 32]
-    paddd      m3, m5                          ;m3=[3+4+5+6]     Row4
-%endmacro
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-%macro IPFILTER_CHROMA_avx2_6xN 1
-cglobal interp_4tap_horiz_pp_6x%1, 5,6,8
-    add             r1d, r1d
-    add             r3d, r3d
-    sub             r0, 2
-    mov             r4d, r4m
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeff]
-    vpbroadcastq    m0, [r5 + r4 * 8]
-%else
-    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova            m1, [interp8_hpp_shuf]
-    vpbroadcastd    m2, [pd_32]
-    pxor            m5, m5
-    mova            m6, [idct8_shuf2]
-    mova            m7, [pw_pixel_max]
-
-    mov             r4d, %1/2
-.loop:
-    vbroadcasti128  m3, [r0]
-    vbroadcasti128  m4, [r0 + 8]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, INTERP_SHIFT_PP           ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m3, m3
-    vpermq          m3, m3, q2020
-    pshufb          xm3, xm6                      ; m3 = WORD[7 6 5 4 3 2 1 0]
-    CLIPW           xm3, xm5, xm7
-    movq            [r2], xm3
-    pextrd          [r2 + 8], xm3, 2
-
-    vbroadcasti128  m3, [r0 + r1]
-    vbroadcasti128  m4, [r0 + r1 + 8]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, INTERP_SHIFT_PP           ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m3, m3
-    vpermq          m3, m3, q2020
-    pshufb          xm3, xm6                      ; m3 = WORD[7 6 5 4 3 2 1 0]
-    CLIPW           xm3, xm5, xm7
-    movq            [r2 + r3], xm3
-    pextrd          [r2 + r3 + 8], xm3, 2
-
-    lea             r0, [r0 + r1 * 2]
-    lea             r2, [r2 + r3 * 2]
-    dec             r4d
-    jnz             .loop
-    RET
-%endmacro
-IPFILTER_CHROMA_avx2_6xN 8
-IPFILTER_CHROMA_avx2_6xN 16
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_8x2, 5,6,8
-    add             r1d, r1d
-    add             r3d, r3d
-    sub             r0, 2
-    mov             r4d, r4m
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeff]
-    vpbroadcastq    m0, [r5 + r4 * 8]
-%else
-    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova            m1, [interp8_hpp_shuf]
-    vpbroadcastd    m2, [pd_32]
-    pxor            m5, m5
-    mova            m6, [idct8_shuf2]
-    mova            m7, [pw_pixel_max]
-
-    vbroadcasti128  m3, [r0]
-    vbroadcasti128  m4, [r0 + 8]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, INTERP_SHIFT_PP          ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m3, m3
-    vpermq          m3, m3,q2020
-    pshufb          xm3, xm6                     ; m3 = WORD[7 6 5 4 3 2 1 0]
-    CLIPW           xm3, xm5, xm7
-    movu            [r2], xm3
-
-    vbroadcasti128  m3, [r0 + r1]
-    vbroadcasti128  m4, [r0 + r1 + 8]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, INTERP_SHIFT_PP           ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m3, m3
-    vpermq          m3, m3,q2020
-    pshufb          xm3, xm6                      ; m3 = WORD[7 6 5 4 3 2 1 0]
-    CLIPW           xm3, xm5, xm7
-    movu            [r2 + r3], xm3
-    RET
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_8x4, 5,6,8
-    add             r1d, r1d
-    add             r3d, r3d
-    sub             r0, 2
-    mov             r4d, r4m
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeff]
-    vpbroadcastq    m0, [r5 + r4 * 8]
-%else
-    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova            m1, [interp8_hpp_shuf]
-    vpbroadcastd    m2, [pd_32]
-    pxor            m5, m5
-    mova            m6, [idct8_shuf2]
-    mova            m7, [pw_pixel_max]
-
-%rep 2
-    vbroadcasti128  m3, [r0]
-    vbroadcasti128  m4, [r0 + 8]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m3, m3
-    vpermq          m3, m3,q2020
-    pshufb          xm3, xm6                    ; m3 = WORD[7 6 5 4 3 2 1 0]
-    CLIPW           xm3, xm5, xm7
-    movu            [r2], xm3
-
-    vbroadcasti128  m3, [r0 + r1]
-    vbroadcasti128  m4, [r0 + r1 + 8]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m3, m3
-    vpermq          m3, m3,q2020
-    pshufb          xm3, xm6                    ; m3 = WORD[7 6 5 4 3 2 1 0]
-    CLIPW           xm3, xm5, xm7
-    movu            [r2 + r3], xm3
-
-    lea             r0, [r0 + r1 * 2]
-    lea             r2, [r2 + r3 * 2]
-%endrep
-    RET
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-%macro IPFILTER_CHROMA_avx2_8xN 1
-cglobal interp_4tap_horiz_pp_8x%1, 5,6,8
-    add             r1d, r1d
-    add             r3d, r3d
-    sub             r0, 2
-    mov             r4d, r4m
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeff]
-    vpbroadcastq    m0, [r5 + r4 * 8]
-%else
-    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova            m1, [interp8_hpp_shuf]
-    vpbroadcastd    m2, [pd_32]
-    pxor            m5, m5
-    mova            m6, [idct8_shuf2]
-    mova            m7, [pw_pixel_max]
-
-    mov             r4d, %1/2
-.loop:
-    vbroadcasti128  m3, [r0]
-    vbroadcasti128  m4, [r0 + 8]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, 6                         ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m3, m3
-    vpermq          m3, m3, q2020
-    pshufb          xm3, xm6                      ; m3 = WORD[7 6 5 4 3 2 1 0]
-    CLIPW           xm3, xm5, xm7
-    movu            [r2], xm3
-
-    vbroadcasti128  m3, [r0 + r1]
-    vbroadcasti128  m4, [r0 + r1 + 8]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, 6                         ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m3, m3
-    vpermq          m3, m3, q2020
-    pshufb          xm3, xm6                      ; m3 = WORD[7 6 5 4 3 2 1 0]
-    CLIPW           xm3, xm5, xm7
-    movu            [r2 + r3], xm3
-
-    lea             r0, [r0 + r1 * 2]
-    lea             r2, [r2 + r3 * 2]
-    dec             r4d
-    jnz             .loop
-    RET
-%endmacro
-IPFILTER_CHROMA_avx2_8xN 6
-IPFILTER_CHROMA_avx2_8xN 8
-IPFILTER_CHROMA_avx2_8xN 12
-IPFILTER_CHROMA_avx2_8xN 16
-IPFILTER_CHROMA_avx2_8xN 32
-IPFILTER_CHROMA_avx2_8xN 64
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-%macro IPFILTER_CHROMA_avx2_16xN 1
-%if ARCH_X86_64
-cglobal interp_4tap_horiz_pp_16x%1, 5,6,9
-    add             r1d, r1d
-    add             r3d, r3d
-    sub             r0, 2
-    mov             r4d, r4m
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeff]
-    vpbroadcastq    m0, [r5 + r4 * 8]
-%else
-    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova            m1, [interp8_hpp_shuf]
-    vpbroadcastd    m2, [pd_32]
-    pxor            m5, m5
-    mova            m6, [idct8_shuf2]
-    mova            m7, [pw_pixel_max]
-
-    mov             r4d, %1
-.loop:
-    vbroadcasti128  m3, [r0]
-    vbroadcasti128  m4, [r0 + 8]
-
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m3, m3
-    vpermq          m3, m3, q2020
-    pshufb          xm3, xm6                     ; m3 = WORD[7 6 5 4 3 2 1 0]
-
-    vbroadcasti128  m4, [r0 + 16]
-    vbroadcasti128  m8, [r0 + 24]
-
-    pshufb          m4, m1
-    pshufb          m8, m1
-
-    pmaddwd         m4, m0
-    pmaddwd         m8, m0
-    phaddd          m4, m8
-    paddd           m4, m2
-    psrad           m4, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m4, m4
-    vpermq          m4, m4, q2020
-    pshufb          xm4, xm6                    ; m3 = WORD[7 6 5 4 3 2 1 0]
-    vinserti128     m3, m3, xm4, 1
-    CLIPW           m3, m5, m7
-    movu            [r2], m3
-
-    add             r0, r1
-    add             r2, r3
-    dec             r4d
-    jnz             .loop
-    RET
-%endif
-%endmacro
-IPFILTER_CHROMA_avx2_16xN 4
-IPFILTER_CHROMA_avx2_16xN 8
-IPFILTER_CHROMA_avx2_16xN 12
-IPFILTER_CHROMA_avx2_16xN 16
-IPFILTER_CHROMA_avx2_16xN 24
-IPFILTER_CHROMA_avx2_16xN 32
-IPFILTER_CHROMA_avx2_16xN 64
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-%macro IPFILTER_CHROMA_avx2_32xN 1
-%if ARCH_X86_64
-cglobal interp_4tap_horiz_pp_32x%1, 5,6,9
-    add             r1d, r1d
-    add             r3d, r3d
-    sub             r0, 2
-    mov             r4d, r4m
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeff]
-    vpbroadcastq    m0, [r5 + r4 * 8]
-%else
-    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova            m1, [interp8_hpp_shuf]
-    vpbroadcastd    m2, [pd_32]
-    pxor            m5, m5
-    mova            m6, [idct8_shuf2]
-    mova            m7, [pw_pixel_max]
-
-    mov             r6d, %1
-.loop:
-%assign x 0
-%rep 2
-    vbroadcasti128  m3, [r0 + x]
-    vbroadcasti128  m4, [r0 + 8 + x]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m3, m3
-    vpermq          m3, m3, q2020
-    pshufb          xm3, xm6                     ; m3 = WORD[7 6 5 4 3 2 1 0]
-
-    vbroadcasti128  m4, [r0 + 16 + x]
-    vbroadcasti128  m8, [r0 + 24 + x]
-    pshufb          m4, m1
-    pshufb          m8, m1
-
-    pmaddwd         m4, m0
-    pmaddwd         m8, m0
-    phaddd          m4, m8
-    paddd           m4, m2
-    psrad           m4, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m4, m4
-    vpermq          m4, m4, q2020
-    pshufb          xm4, xm6                    ; m3 = WORD[7 6 5 4 3 2 1 0]
-    vinserti128     m3, m3, xm4, 1
-    CLIPW           m3, m5, m7
-    movu            [r2 + x], m3
-    %assign x x+32
-    %endrep
-
-    add             r0, r1
-    add             r2, r3
-    dec             r6d
-    jnz             .loop
-    RET
-%endif
-%endmacro
-IPFILTER_CHROMA_avx2_32xN 8
-IPFILTER_CHROMA_avx2_32xN 16
-IPFILTER_CHROMA_avx2_32xN 24
-IPFILTER_CHROMA_avx2_32xN 32
-IPFILTER_CHROMA_avx2_32xN 48
-IPFILTER_CHROMA_avx2_32xN 64
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-%macro IPFILTER_CHROMA_avx2_12xN 1
-%if ARCH_X86_64
-cglobal interp_4tap_horiz_pp_12x%1, 5,6,8
-    add             r1d, r1d
-    add             r3d, r3d
-    sub             r0, 2
-    mov             r4d, r4m
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeff]
-    vpbroadcastq    m0, [r5 + r4 * 8]
-%else
-    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova            m1, [interp8_hpp_shuf]
-    vpbroadcastd    m2, [pd_32]
-    pxor            m5, m5
-    mova            m6, [idct8_shuf2]
-    mova            m7, [pw_pixel_max]
-
-    mov             r4d, %1
-.loop:
-    vbroadcasti128  m3, [r0]
-    vbroadcasti128  m4, [r0 + 8]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m3, m3
-    vpermq          m3, m3, q2020
-    pshufb          xm3, xm6                     ; m3 = WORD[7 6 5 4 3 2 1 0]
-    CLIPW           xm3, xm5, xm7
-    movu            [r2], xm3
-
-    vbroadcasti128  m3, [r0 + 16]
-    vbroadcasti128  m4, [r0 + 24]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m3, m3
-    vpermq          m3, m3, q2020
-    pshufb          xm3, xm6                    ; m3 = WORD[7 6 5 4 3 2 1 0]
-    CLIPW           xm3, xm5, xm7
-    movq            [r2 + 16], xm3
-
-    add             r0, r1
-    add             r2, r3
-    dec             r4d
-    jnz             .loop
-    RET
-%endif
-%endmacro
-IPFILTER_CHROMA_avx2_12xN 16
-IPFILTER_CHROMA_avx2_12xN 32
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-%macro IPFILTER_CHROMA_avx2_24xN 1
-%if ARCH_X86_64
-cglobal interp_4tap_horiz_pp_24x%1, 5,6,9
-    add             r1d, r1d
-    add             r3d, r3d
-    sub             r0, 2
-    mov             r4d, r4m
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeff]
-    vpbroadcastq    m0, [r5 + r4 * 8]
-%else
-    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova            m1, [interp8_hpp_shuf]
-    vpbroadcastd    m2, [pd_32]
-    pxor            m5, m5
-    mova            m6, [idct8_shuf2]
-    mova            m7, [pw_pixel_max]
-
-    mov             r4d, %1
-.loop:
-    vbroadcasti128  m3, [r0]
-    vbroadcasti128  m4, [r0 + 8]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, 6
-
-    vbroadcasti128  m4, [r0 + 16]
-    vbroadcasti128  m8, [r0 + 24]
-    pshufb          m4, m1
-    pshufb          m8, m1
-
-    pmaddwd         m4, m0
-    pmaddwd         m8, m0
-    phaddd          m4, m8
-    paddd           m4, m2
-    psrad           m4, 6
-
-    packusdw        m3, m4
-    vpermq          m3, m3, q3120
-    pshufb          m3, m6
-    CLIPW           m3, m5, m7
-    movu            [r2], m3
-
-    vbroadcasti128  m3, [r0 + 32]
-    vbroadcasti128  m4, [r0 + 40]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, 6
-
-    packusdw        m3, m3
-    vpermq          m3, m3, q2020
-    pshufb          xm3, xm6
-    CLIPW           xm3, xm5, xm7
-    movu            [r2 + 32], xm3
-
-    add             r0, r1
-    add             r2, r3
-    dec             r4d
-    jnz             .loop
-    RET
-%endif
-%endmacro
-IPFILTER_CHROMA_avx2_24xN 32
-IPFILTER_CHROMA_avx2_24xN 64
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-%macro IPFILTER_CHROMA_avx2_64xN 1
-%if ARCH_X86_64
-cglobal interp_4tap_horiz_pp_64x%1, 5,6,9
-    add             r1d, r1d
-    add             r3d, r3d
-    sub             r0, 2
-    mov             r4d, r4m
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeff]
-    vpbroadcastq    m0, [r5 + r4 * 8]
-%else
-    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova            m1, [interp8_hpp_shuf]
-    vpbroadcastd    m2, [pd_32]
-    pxor            m5, m5
-    mova            m6, [idct8_shuf2]
-    mova            m7, [pw_pixel_max]
-
-    mov             r6d, %1
-.loop:
-%assign x 0
-%rep 4
-    vbroadcasti128  m3, [r0 + x]
-    vbroadcasti128  m4, [r0 + 8 + x]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, 6
-
-    vbroadcasti128  m4, [r0 + 16 + x]
-    vbroadcasti128  m8, [r0 + 24 + x]
-    pshufb          m4, m1
-    pshufb          m8, m1
+    lea       r0, [r0 + 4 * r1 - 2 * %2]
+    lea       r2, [r2 + 4 * r3]
 
-    pmaddwd         m4, m0
-    pmaddwd         m8, m0
-    phaddd          m4, m8
-    paddd           m4, m2
-    psrad           m4, 6
+    dec         r4d
+    jnz       .loopH
 
-    packusdw        m3, m4
-    vpermq          m3, m3, q3120
-    pshufb          m3, m6
-    CLIPW           m3, m5, m7
-    movu            [r2 + x], m3
-    %assign x x+32
-    %endrep
-
-    add             r0, r1
-    add             r2, r3
-    dec             r6d
-    jnz             .loop
     RET
-%endif
 %endmacro
-IPFILTER_CHROMA_avx2_64xN 16
-IPFILTER_CHROMA_avx2_64xN 32
-IPFILTER_CHROMA_avx2_64xN 48
-IPFILTER_CHROMA_avx2_64xN 64
 
 ;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+; void interp_8tap_vert_pp_%2x%3(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
 ;-------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-%if ARCH_X86_64
-cglobal interp_4tap_horiz_pp_48x64, 5,6,9
-    add             r1d, r1d
-    add             r3d, r3d
-    sub             r0, 2
-    mov             r4d, r4m
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeff]
-    vpbroadcastq    m0, [r5 + r4 * 8]
-%else
-    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova            m1, [interp8_hpp_shuf]
-    vpbroadcastd    m2, [pd_32]
-    pxor            m5, m5
-    mova            m6, [idct8_shuf2]
-    mova            m7, [pw_pixel_max]
+    FILTER_VER_LUMA_sse2 pp, 4, 4
+    FILTER_VER_LUMA_sse2 pp, 8, 8
+    FILTER_VER_LUMA_sse2 pp, 8, 4
+    FILTER_VER_LUMA_sse2 pp, 4, 8
+    FILTER_VER_LUMA_sse2 pp, 16, 16
+    FILTER_VER_LUMA_sse2 pp, 16, 8
+    FILTER_VER_LUMA_sse2 pp, 8, 16
+    FILTER_VER_LUMA_sse2 pp, 16, 12
+    FILTER_VER_LUMA_sse2 pp, 12, 16
+    FILTER_VER_LUMA_sse2 pp, 16, 4
+    FILTER_VER_LUMA_sse2 pp, 4, 16
+    FILTER_VER_LUMA_sse2 pp, 32, 32
+    FILTER_VER_LUMA_sse2 pp, 32, 16
+    FILTER_VER_LUMA_sse2 pp, 16, 32
+    FILTER_VER_LUMA_sse2 pp, 32, 24
+    FILTER_VER_LUMA_sse2 pp, 24, 32
+    FILTER_VER_LUMA_sse2 pp, 32, 8
+    FILTER_VER_LUMA_sse2 pp, 8, 32
+    FILTER_VER_LUMA_sse2 pp, 64, 64
+    FILTER_VER_LUMA_sse2 pp, 64, 32
+    FILTER_VER_LUMA_sse2 pp, 32, 64
+    FILTER_VER_LUMA_sse2 pp, 64, 48
+    FILTER_VER_LUMA_sse2 pp, 48, 64
+    FILTER_VER_LUMA_sse2 pp, 64, 16
+    FILTER_VER_LUMA_sse2 pp, 16, 64
 
-    mov             r4d, 64
-.loop:
-%assign x 0
-%rep 3
-    vbroadcasti128  m3, [r0 + x]
-    vbroadcasti128  m4, [r0 + 8 + x]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, 6
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_ps_%2x%3(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_sse2 ps, 4, 4
+    FILTER_VER_LUMA_sse2 ps, 8, 8
+    FILTER_VER_LUMA_sse2 ps, 8, 4
+    FILTER_VER_LUMA_sse2 ps, 4, 8
+    FILTER_VER_LUMA_sse2 ps, 16, 16
+    FILTER_VER_LUMA_sse2 ps, 16, 8
+    FILTER_VER_LUMA_sse2 ps, 8, 16
+    FILTER_VER_LUMA_sse2 ps, 16, 12
+    FILTER_VER_LUMA_sse2 ps, 12, 16
+    FILTER_VER_LUMA_sse2 ps, 16, 4
+    FILTER_VER_LUMA_sse2 ps, 4, 16
+    FILTER_VER_LUMA_sse2 ps, 32, 32
+    FILTER_VER_LUMA_sse2 ps, 32, 16
+    FILTER_VER_LUMA_sse2 ps, 16, 32
+    FILTER_VER_LUMA_sse2 ps, 32, 24
+    FILTER_VER_LUMA_sse2 ps, 24, 32
+    FILTER_VER_LUMA_sse2 ps, 32, 8
+    FILTER_VER_LUMA_sse2 ps, 8, 32
+    FILTER_VER_LUMA_sse2 ps, 64, 64
+    FILTER_VER_LUMA_sse2 ps, 64, 32
+    FILTER_VER_LUMA_sse2 ps, 32, 64
+    FILTER_VER_LUMA_sse2 ps, 64, 48
+    FILTER_VER_LUMA_sse2 ps, 48, 64
+    FILTER_VER_LUMA_sse2 ps, 64, 16
+    FILTER_VER_LUMA_sse2 ps, 16, 64
 
-    vbroadcasti128  m4, [r0 + 16 + x]
-    vbroadcasti128  m8, [r0 + 24 + x]
-    pshufb          m4, m1
-    pshufb          m8, m1
+%macro PROCESS_CHROMA_SP_W4_4R 0
+    movq       m0, [r0]
+    movq       m1, [r0 + r1]
+    punpcklwd  m0, m1                          ;m0=[0 1]
+    pmaddwd    m0, [r6 + 0 *32]                ;m0=[0+1]         Row1
 
-    pmaddwd         m4, m0
-    pmaddwd         m8, m0
-    phaddd          m4, m8
-    paddd           m4, m2
-    psrad           m4, 6
+    lea        r0, [r0 + 2 * r1]
+    movq       m4, [r0]
+    punpcklwd  m1, m4                          ;m1=[1 2]
+    pmaddwd    m1, [r6 + 0 *32]                ;m1=[1+2]         Row2
 
-    packusdw        m3, m4
-    vpermq          m3, m3, q3120
-    pshufb          m3, m6
-    CLIPW           m3, m5, m7
-    movu            [r2 + x], m3
-%assign x x+32
-%endrep
+    movq       m5, [r0 + r1]
+    punpcklwd  m4, m5                          ;m4=[2 3]
+    pmaddwd    m2, m4, [r6 + 0 *32]            ;m2=[2+3]         Row3
+    pmaddwd    m4, [r6 + 1 * 32]
+    paddd      m0, m4                          ;m0=[0+1+2+3]     Row1 done
 
-    add             r0, r1
-    add             r2, r3
-    dec             r4d
-    jnz             .loop
-    RET
-%endif
+    lea        r0, [r0 + 2 * r1]
+    movq       m4, [r0]
+    punpcklwd  m5, m4                          ;m5=[3 4]
+    pmaddwd    m3, m5, [r6 + 0 *32]            ;m3=[3+4]         Row4
+    pmaddwd    m5, [r6 + 1 * 32]
+    paddd      m1, m5                          ;m1 = [1+2+3+4]   Row2
+
+    movq       m5, [r0 + r1]
+    punpcklwd  m4, m5                          ;m4=[4 5]
+    pmaddwd    m4, [r6 + 1 * 32]
+    paddd      m2, m4                          ;m2=[2+3+4+5]     Row3
+
+    movq       m4, [r0 + 2 * r1]
+    punpcklwd  m5, m4                          ;m5=[5 6]
+    pmaddwd    m5, [r6 + 1 * 32]
+    paddd      m3, m5                          ;m3=[3+4+5+6]     Row4
+%endmacro
 
 ;-----------------------------------------------------------------------------------------------------------------
 ; void interp_4tap_vert_%3_%1x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
@@ -10275,1247 +6590,6 @@ cglobal filterPixelToShort_48x64, 3, 7, 4
     RET
 
 
-;-----------------------------------------------------------------------------------------------------------------------------
-;void interp_horiz_ps_c(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;-----------------------------------------------------------------------------------------------------------------------------
-
-%macro IPFILTER_LUMA_PS_4xN_AVX2 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_horiz_ps_4x%1, 6,8,7
-    mov                         r5d,               r5m
-    mov                         r4d,               r4m
-    add                         r1d,               r1d
-    add                         r3d,               r3d
-
-%ifdef PIC
-    lea                         r6,                [tab_LumaCoeff]
-    lea                         r4,                [r4 * 8]
-    vbroadcasti128              m0,                [r6 + r4 * 2]
-%else
-    lea                         r4,                [r4 * 8]
-    vbroadcasti128              m0,                [tab_LumaCoeff + r4 * 2]
-%endif
-
-    vbroadcasti128              m2,                [INTERP_OFFSET_PS]
-
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - pw_2000
-
-    sub                         r0,                6
-    test                        r5d,               r5d
-    mov                         r7d,               %1                                    ; loop count variable - height
-    jz                         .preloop
-    lea                         r6,                [r1 * 3]                              ; r6 = (N / 2 - 1) * srcStride
-    sub                         r0,                r6                                    ; r0(src) - 3 * srcStride
-    add                         r7d,               6                                     ;7 - 1(since last row not in loop)                            ; need extra 7 rows, just set a specially flag here, blkheight += N - 1  (7 - 3 = 4 ; since the last three rows not in loop)
-
-.preloop:
-    lea                         r6,                [r3 * 3]
-.loop:
-    ; Row 0
-    movu                        xm3,                [r0]                                 ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    movu                        xm4,                [r0 + 2]                             ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    vinserti128                 m3,                 m3,                xm4,       1
-    movu                        xm4,                [r0 + 4]
-    movu                        xm5,                [r0 + 6]
-    vinserti128                 m4,                 m4,                xm5,       1
-    pmaddwd                     m3,                m0
-    pmaddwd                     m4,                m0
-    phaddd                      m3,                m4                                    ; DWORD [R1D R1C R0D R0C R1B R1A R0B R0A]
-
-    ; Row 1
-    movu                        xm4,                [r0 + r1]                            ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    movu                        xm5,                [r0 + r1 + 2]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    vinserti128                 m4,                 m4,                xm5,       1
-    movu                        xm5,                [r0 + r1 + 4]
-    movu                        xm6,                [r0 + r1 + 6]
-    vinserti128                 m5,                 m5,                xm6,       1
-    pmaddwd                     m4,                m0
-    pmaddwd                     m5,                m0
-    phaddd                      m4,                m5                                     ; DWORD [R3D R3C R2D R2C R3B R3A R2B R2A]
-    phaddd                      m3,                m4                                     ; all rows and col completed.
-
-    mova                        m5,                [interp8_hps_shuf]
-    vpermd                      m3,                m5,                  m3
-    paddd                       m3,                m2
-    vextracti128                xm4,               m3,                  1
-    psrad                       xm3,               INTERP_SHIFT_PS
-    psrad                       xm4,               INTERP_SHIFT_PS
-    packssdw                    xm3,               xm3
-    packssdw                    xm4,               xm4
-
-    movq                        [r2],              xm3                                   ;row 0
-    movq                        [r2 + r3],         xm4                                   ;row 1
-    lea                         r0,                [r0 + r1 * 2]                         ; first loop src ->5th row(i.e 4)
-    lea                         r2,                [r2 + r3 * 2]                         ; first loop dst ->5th row(i.e 4)
-
-    sub                         r7d,               2
-    jg                          .loop
-    test                        r5d,               r5d
-    jz                          .end
-
-    ; Row 10
-    movu                        xm3,                [r0]
-    movu                        xm4,                [r0 + 2]
-    vinserti128                 m3,                 m3,                 xm4,      1
-    movu                        xm4,                [r0 + 4]
-    movu                        xm5,                [r0 + 6]
-    vinserti128                 m4,                 m4,                 xm5,      1
-    pmaddwd                     m3,                m0
-    pmaddwd                     m4,                m0
-    phaddd                      m3,                m4
-
-    ; Row11
-    phaddd                      m3,                m4                                    ; all rows and col completed.
-
-    mova                        m5,                [interp8_hps_shuf]
-    vpermd                      m3,                m5,                  m3
-    paddd                       m3,                m2
-    vextracti128                xm4,               m3,                  1
-    psrad                       xm3,               INTERP_SHIFT_PS
-    psrad                       xm4,               INTERP_SHIFT_PS
-    packssdw                    xm3,               xm3
-    packssdw                    xm4,               xm4
-
-    movq                        [r2],              xm3                                   ;row 0
-.end:
-    RET
-%endif
-%endmacro
-
-    IPFILTER_LUMA_PS_4xN_AVX2 4
-    IPFILTER_LUMA_PS_4xN_AVX2 8
-    IPFILTER_LUMA_PS_4xN_AVX2 16
-
-%macro IPFILTER_LUMA_PS_8xN_AVX2 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_horiz_ps_8x%1, 4, 6, 8
-    add                 r1d, r1d
-    add                 r3d, r3d
-    mov                 r4d, r4m
-    mov                 r5d, r5m
-    shl                 r4d, 4
-%ifdef PIC
-    lea                 r6, [tab_LumaCoeff]
-    vpbroadcastq        m0, [r6 + r4]
-    vpbroadcastq        m1, [r6 + r4 + 8]
-%else
-    vpbroadcastq        m0, [tab_LumaCoeff + r4]
-    vpbroadcastq        m1, [tab_LumaCoeff + r4 + 8]
-%endif
-    mova                m3, [interp8_hpp_shuf]
-    vbroadcasti128      m2, [INTERP_OFFSET_PS]
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    sub                 r0, 6
-    test                r5d, r5d
-    mov                 r4d, %1
-    jz                  .loop0
-    lea                 r6, [r1*3]
-    sub                 r0, r6
-    add                 r4d, 7
-
-.loop0:
-    vbroadcasti128      m4, [r0]
-    vbroadcasti128      m5, [r0 + 8]
-    pshufb              m4, m3
-    pshufb              m7, m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m7, m1
-    paddd               m4, m7
-
-    vbroadcasti128      m6, [r0 + 16]
-    pshufb              m5, m3
-    pshufb              m6, m3
-    pmaddwd             m5, m0
-    pmaddwd             m6, m1
-    paddd               m5, m6
-
-    phaddd              m4, m5
-    vpermq              m4, m4, q3120
-    paddd               m4, m2
-    vextracti128        xm5,m4, 1
-    psrad               xm4, INTERP_SHIFT_PS
-    psrad               xm5, INTERP_SHIFT_PS
-    packssdw            xm4, xm5
-
-    movu                [r2], xm4
-    add                 r2, r3
-    add                 r0, r1
-    dec                 r4d
-    jnz                 .loop0
-    RET
-%endif
-%endmacro
-
-    IPFILTER_LUMA_PS_8xN_AVX2 4
-    IPFILTER_LUMA_PS_8xN_AVX2 8
-    IPFILTER_LUMA_PS_8xN_AVX2 16
-    IPFILTER_LUMA_PS_8xN_AVX2 32
-
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_horiz_ps_24x32, 4, 6, 8
-    add                 r1d, r1d
-    add                 r3d, r3d
-    mov                 r4d, r4m
-    mov                 r5d, r5m
-    shl                 r4d, 4
-%ifdef PIC
-    lea                 r6, [tab_LumaCoeff]
-    vpbroadcastq        m0, [r6 + r4]
-    vpbroadcastq        m1, [r6 + r4 + 8]
-%else
-    vpbroadcastq        m0, [tab_LumaCoeff + r4]
-    vpbroadcastq        m1, [tab_LumaCoeff + r4 + 8]
-%endif
-    mova                m3, [interp8_hpp_shuf]
-    vbroadcasti128      m2, [INTERP_OFFSET_PS]
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    sub                 r0, 6
-    test                r5d, r5d
-    mov                 r4d, 32
-    jz                  .loop0
-    lea                 r6, [r1*3]
-    sub                 r0, r6
-    add                 r4d, 7
-
-.loop0:
-%assign x 0
-%rep 24/8
-    vbroadcasti128      m4, [r0 + x]
-    vbroadcasti128      m5, [r0 + 8 + x]
-    pshufb              m4, m3
-    pshufb              m7, m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m7, m1
-    paddd               m4, m7
-
-    vbroadcasti128      m6, [r0 + 16 + x]
-    pshufb              m5, m3
-    pshufb              m6, m3
-    pmaddwd             m5, m0
-    pmaddwd             m6, m1
-    paddd               m5, m6
-
-    phaddd              m4, m5
-    vpermq              m4, m4, q3120
-    paddd               m4, m2
-    vextracti128        xm5,m4, 1
-    psrad               xm4, INTERP_SHIFT_PS
-    psrad               xm5, INTERP_SHIFT_PS
-    packssdw            xm4, xm5
-
-    movu                [r2 + x], xm4
-    %assign x x+16
-    %endrep
-
-    add                 r2, r3
-    add                 r0, r1
-    dec                 r4d
-    jnz                 .loop0
-    RET
-%endif
-
-
-%macro IPFILTER_LUMA_PS_32_64_AVX2 2
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_horiz_ps_%1x%2, 4, 6, 8
-
-    add                 r1d, r1d
-    add                 r3d, r3d
-    mov                 r4d, r4m
-    mov                 r5d, r5m
-    shl                 r4d, 6
-%ifdef PIC
-    lea                 r6, [tab_LumaCoeffV]
-    movu                m0, [r6 + r4]
-    movu                m1, [r6 + r4 + mmsize]
-%else
-    movu                m0, [tab_LumaCoeffV + r4]
-    movu                m1, [tab_LumaCoeffV + r4 + mmsize]
-%endif
-    mova                m3, [interp8_hpp_shuf_new]
-    vbroadcasti128      m2, [INTERP_OFFSET_PS]
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    sub                 r0, 6
-    test                r5d, r5d
-    mov                 r4d, %2
-    jz                 .loop0
-    lea                 r6, [r1*3]
-    sub                 r0, r6
-    add                 r4d, 7
-
-.loop0:
-%assign x 0
-%rep %1/16
-    vbroadcasti128      m4, [r0 + x]
-    vbroadcasti128      m5, [r0 + 4 * SIZEOF_PIXEL + x]
-    pshufb              m4, m3
-    pshufb              m5, m3
-
-    pmaddwd             m4, m0
-    pmaddwd             m7, m5, m1
-    paddd               m4, m7
-    vextracti128        xm7, m4, 1
-    paddd               xm4, xm7
-    paddd               xm4, xm2
-    psrad               xm4, INTERP_SHIFT_PS
-
-    vbroadcasti128      m6, [r0 + 16 + x]
-    pshufb              m6, m3
-
-    pmaddwd             m5, m0
-    pmaddwd             m7, m6, m1
-    paddd               m5, m7
-    vextracti128        xm7, m5, 1
-    paddd               xm5, xm7
-    paddd               xm5, xm2
-    psrad               xm5, INTERP_SHIFT_PS
-
-    packssdw            xm4, xm5
-    movu                [r2 + x], xm4
-
-    vbroadcasti128      m5, [r0 + 24 + x]
-    pshufb              m5, m3
-
-    pmaddwd             m6, m0
-    pmaddwd             m7, m5, m1
-    paddd               m6, m7
-    vextracti128        xm7, m6, 1
-    paddd               xm6, xm7
-    paddd               xm6, xm2
-    psrad               xm6, INTERP_SHIFT_PS
-
-    vbroadcasti128      m7, [r0 + 32 + x]
-    pshufb              m7, m3
-
-    pmaddwd             m5, m0
-    pmaddwd             m7, m1
-    paddd               m5, m7
-    vextracti128        xm7, m5, 1
-    paddd               xm5, xm7
-    paddd               xm5, xm2
-    psrad               xm5, INTERP_SHIFT_PS
-
-    packssdw            xm6, xm5
-    movu                [r2 + 16 + x], xm6
-
-%assign x x+32
-%endrep
-
-    add                 r2, r3
-    add                 r0, r1
-    dec                 r4d
-    jnz                .loop0
-    RET
-%endif
-%endmacro
-
-    IPFILTER_LUMA_PS_32_64_AVX2 32, 8
-    IPFILTER_LUMA_PS_32_64_AVX2 32, 16
-    IPFILTER_LUMA_PS_32_64_AVX2 32, 24
-    IPFILTER_LUMA_PS_32_64_AVX2 32, 32
-    IPFILTER_LUMA_PS_32_64_AVX2 32, 64
-
-    IPFILTER_LUMA_PS_32_64_AVX2 64, 16
-    IPFILTER_LUMA_PS_32_64_AVX2 64, 32
-    IPFILTER_LUMA_PS_32_64_AVX2 64, 48
-    IPFILTER_LUMA_PS_32_64_AVX2 64, 64
-
-    IPFILTER_LUMA_PS_32_64_AVX2 48, 64
-
-%macro IPFILTER_LUMA_PS_16xN_AVX2 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_horiz_ps_16x%1, 4, 6, 8
-
-    add                 r1d, r1d
-    add                 r3d, r3d
-    mov                 r4d, r4m
-    mov                 r5d, r5m
-    shl                 r4d, 4
-%ifdef PIC
-    lea                 r6, [tab_LumaCoeff]
-    vpbroadcastq        m0, [r6 + r4]
-    vpbroadcastq        m1, [r6 + r4 + 8]
-%else
-    vpbroadcastq        m0, [tab_LumaCoeff + r4]
-    vpbroadcastq        m1, [tab_LumaCoeff + r4 + 8]
-%endif
-    mova                m3, [interp8_hpp_shuf]
-    vbroadcasti128      m2, [INTERP_OFFSET_PS]
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    sub                 r0, 6
-    test                r5d, r5d
-    mov                 r4d, %1
-    jz                  .loop0
-    lea                 r6, [r1*3]
-    sub                 r0, r6
-    add                 r4d, 7
-
-.loop0:
-    vbroadcasti128      m4, [r0]
-    vbroadcasti128      m5, [r0 + 8]
-    pshufb              m4, m3
-    pshufb              m7, m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m7, m1
-    paddd               m4, m7
-
-    vbroadcasti128      m6, [r0 + 16]
-    pshufb              m5, m3
-    pshufb              m7, m6, m3
-    pmaddwd             m5, m0
-    pmaddwd             m7, m1
-    paddd               m5, m7
-
-    phaddd              m4, m5
-    vpermq              m4, m4, q3120
-    paddd               m4, m2
-    vextracti128        xm5, m4, 1
-    psrad               xm4, INTERP_SHIFT_PS
-    psrad               xm5, INTERP_SHIFT_PS
-    packssdw            xm4, xm5
-    movu                [r2], xm4
-
-    vbroadcasti128      m5, [r0 + 24]
-    pshufb              m6, m3
-    pshufb              m7, m5, m3
-    pmaddwd             m6, m0
-    pmaddwd             m7, m1
-    paddd               m6, m7
-
-    vbroadcasti128      m7, [r0 + 32]
-    pshufb              m5, m3
-    pshufb              m7, m3
-    pmaddwd             m5, m0
-    pmaddwd             m7, m1
-    paddd               m5, m7
-
-    phaddd              m6, m5
-    vpermq              m6, m6, q3120
-    paddd               m6, m2
-    vextracti128        xm5,m6, 1
-    psrad               xm6, INTERP_SHIFT_PS
-    psrad               xm5, INTERP_SHIFT_PS
-    packssdw            xm6, xm5
-    movu                [r2 + 16], xm6
-
-    add                 r2, r3
-    add                 r0, r1
-    dec                 r4d
-    jnz                 .loop0
-    RET
-%endif
-%endmacro
-
-    IPFILTER_LUMA_PS_16xN_AVX2 4
-    IPFILTER_LUMA_PS_16xN_AVX2 8
-    IPFILTER_LUMA_PS_16xN_AVX2 12
-    IPFILTER_LUMA_PS_16xN_AVX2 16
-    IPFILTER_LUMA_PS_16xN_AVX2 32
-    IPFILTER_LUMA_PS_16xN_AVX2 64
-
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_horiz_ps_12x16, 4, 6, 8
-    add                 r1d, r1d
-    add                 r3d, r3d
-    mov                 r4d, r4m
-    mov                 r5d, r5m
-    shl                 r4d, 4
-%ifdef PIC
-    lea                 r6, [tab_LumaCoeff]
-    vpbroadcastq        m0, [r6 + r4]
-    vpbroadcastq        m1, [r6 + r4 + 8]
-%else
-    vpbroadcastq        m0, [tab_LumaCoeff + r4]
-    vpbroadcastq        m1, [tab_LumaCoeff + r4 + 8]
-%endif
-    mova                m3, [interp8_hpp_shuf]
-    vbroadcasti128      m2, [INTERP_OFFSET_PS]
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    sub                 r0, 6
-    test                r5d, r5d
-    mov                 r4d, 16
-    jz                  .loop0
-    lea                 r6, [r1*3]
-    sub                 r0, r6
-    add                 r4d, 7
-
-.loop0:
-    vbroadcasti128      m4, [r0]
-    vbroadcasti128      m5, [r0 + 8]
-    pshufb              m4, m3
-    pshufb              m7, m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m7, m1
-    paddd               m4, m7
-
-    vbroadcasti128      m6, [r0 + 16]
-    pshufb              m5, m3
-    pshufb              m7, m6, m3
-    pmaddwd             m5, m0
-    pmaddwd             m7, m1
-    paddd               m5, m7
-
-    phaddd              m4, m5
-    vpermq              m4, m4, q3120
-    paddd               m4, m2
-    vextracti128        xm5,m4, 1
-    psrad               xm4, INTERP_SHIFT_PS
-    psrad               xm5, INTERP_SHIFT_PS
-    packssdw            xm4, xm5
-    movu                [r2], xm4
-
-    vbroadcasti128      m5, [r0 + 24]
-    pshufb              m6, m3
-    pshufb              m5, m3
-    pmaddwd             m6, m0
-    pmaddwd             m5, m1
-    paddd               m6, m5
-
-    phaddd              m6, m6
-    vpermq              m6, m6, q3120
-    paddd               xm6, xm2
-    psrad               xm6, INTERP_SHIFT_PS
-    packssdw            xm6, xm6
-    movq                [r2 + 16], xm6
-
-    add                 r2, r3
-    add                 r0, r1
-    dec                 r4d
-    jnz                 .loop0
-    RET
-%endif
-
-%macro IPFILTER_CHROMA_PS_8xN_AVX2 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_horiz_ps_8x%1, 4, 7, 6
-    add                 r1d, r1d
-    add                 r3d, r3d
-    mov                 r4d, r4m
-    mov                 r5d, r5m
-
-%ifdef PIC
-    lea                 r6, [tab_ChromaCoeff]
-    vpbroadcastq        m0, [r6 + r4 * 8]
-%else
-    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova                m3, [interp8_hpp_shuf]
-    vbroadcasti128      m2, [INTERP_OFFSET_PS]
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    sub                 r0, 2
-    test                r5d, r5d
-    mov                 r4d, %1
-    jz                  .loop0
-    sub                 r0, r1
-    add                 r4d, 3
-
-.loop0:
-    vbroadcasti128      m4, [r0]
-    vbroadcasti128      m5, [r0 + 8]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2], xm4
-    add                 r2, r3
-    add                 r0, r1
-    dec                 r4d
-    jnz                 .loop0
-    RET
-%endif
-%endmacro
-
-    IPFILTER_CHROMA_PS_8xN_AVX2 4
-    IPFILTER_CHROMA_PS_8xN_AVX2 8
-    IPFILTER_CHROMA_PS_8xN_AVX2 16
-    IPFILTER_CHROMA_PS_8xN_AVX2 32
-    IPFILTER_CHROMA_PS_8xN_AVX2 6
-    IPFILTER_CHROMA_PS_8xN_AVX2 2
-    IPFILTER_CHROMA_PS_8xN_AVX2 12
-    IPFILTER_CHROMA_PS_8xN_AVX2 64
-
-%macro IPFILTER_CHROMA_PS_16xN_AVX2 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_horiz_ps_16x%1, 4, 7, 6
-    add                 r1d, r1d
-    add                 r3d, r3d
-    mov                 r4d, r4m
-    mov                 r5d, r5m
-
-%ifdef PIC
-    lea                 r6, [tab_ChromaCoeff]
-    vpbroadcastq        m0, [r6 + r4 * 8]
-%else
-    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova                m3, [interp8_hpp_shuf]
-    vbroadcasti128      m2, [INTERP_OFFSET_PS]
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    sub                 r0, 2
-    test                r5d, r5d
-    mov                 r4d, %1
-    jz                  .loop0
-    sub                 r0, r1
-    add                 r4d, 3
-
-.loop0:
-    vbroadcasti128      m4, [r0]
-    vbroadcasti128      m5, [r0 + 8]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2], xm4
-
-    vbroadcasti128      m4, [r0 + 16]
-    vbroadcasti128      m5, [r0 + 24]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 16], xm4
-
-    add                 r2, r3
-    add                 r0, r1
-    dec                 r4d
-    jnz                 .loop0
-    RET
-%endif
-%endmacro
-
-IPFILTER_CHROMA_PS_16xN_AVX2 16
-IPFILTER_CHROMA_PS_16xN_AVX2 8
-IPFILTER_CHROMA_PS_16xN_AVX2 32
-IPFILTER_CHROMA_PS_16xN_AVX2 12
-IPFILTER_CHROMA_PS_16xN_AVX2 4
-IPFILTER_CHROMA_PS_16xN_AVX2 64
-IPFILTER_CHROMA_PS_16xN_AVX2 24
-
-%macro IPFILTER_CHROMA_PS_24xN_AVX2 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_horiz_ps_24x%1, 4, 7, 6
-    add                 r1d, r1d
-    add                 r3d, r3d
-    mov                 r4d, r4m
-    mov                 r5d, r5m
-
-%ifdef PIC
-    lea                 r6, [tab_ChromaCoeff]
-    vpbroadcastq        m0, [r6 + r4 * 8]
-%else
-    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova                m3, [interp8_hpp_shuf]
-    vbroadcasti128      m2, [INTERP_OFFSET_PS]
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    sub                 r0, 2
-    test                r5d, r5d
-    mov                 r4d, %1
-    jz                  .loop0
-    sub                 r0, r1
-    add                 r4d, 3
-
-.loop0:
-    vbroadcasti128      m4, [r0]
-    vbroadcasti128      m5, [r0 + 8]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2], xm4
-
-    vbroadcasti128      m4, [r0 + 16]
-    vbroadcasti128      m5, [r0 + 24]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 16], xm4
-
-    vbroadcasti128      m4, [r0 + 32]
-    vbroadcasti128      m5, [r0 + 40]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 32], xm4
-
-    add                 r2, r3
-    add                 r0, r1
-    dec                 r4d
-    jnz                 .loop0
-    RET
-%endif
-%endmacro
-
-IPFILTER_CHROMA_PS_24xN_AVX2 32
-IPFILTER_CHROMA_PS_24xN_AVX2 64
-
-%macro IPFILTER_CHROMA_PS_12xN_AVX2 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_horiz_ps_12x%1, 4, 7, 6
-    add                 r1d, r1d
-    add                 r3d, r3d
-    mov                 r4d, r4m
-    mov                 r5d, r5m
-
-%ifdef PIC
-    lea                 r6, [tab_ChromaCoeff]
-    vpbroadcastq        m0, [r6 + r4 * 8]
-%else
-    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova                m3, [interp8_hpp_shuf]
-    vbroadcasti128      m2, [INTERP_OFFSET_PS]
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    sub                 r0, 2
-    test                r5d, r5d
-    mov                 r4d, %1
-    jz                  .loop0
-    sub                 r0, r1
-    add                 r4d, 3
-
-.loop0:
-    vbroadcasti128      m4, [r0]
-    vbroadcasti128      m5, [r0 + 8]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2], xm4
-
-    vbroadcasti128      m4, [r0 + 16]
-    pshufb              m4, m3
-    pmaddwd             m4, m0
-    phaddd              m4, m4
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movq                [r2 + 16], xm4
-
-    add                 r2, r3
-    add                 r0, r1
-    dec                 r4d
-    jnz                 .loop0
-    RET
-%endif
-%endmacro
-
-IPFILTER_CHROMA_PS_12xN_AVX2 16
-IPFILTER_CHROMA_PS_12xN_AVX2 32
-
-%macro IPFILTER_CHROMA_PS_32xN_AVX2 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_horiz_ps_32x%1, 4, 7, 6
-    add                 r1d, r1d
-    add                 r3d, r3d
-    mov                 r4d, r4m
-    mov                 r5d, r5m
-
-%ifdef PIC
-    lea                 r6, [tab_ChromaCoeff]
-    vpbroadcastq        m0, [r6 + r4 * 8]
-%else
-    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova                m3, [interp8_hpp_shuf]
-    vbroadcasti128      m2, [INTERP_OFFSET_PS]
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    sub                 r0, 2
-    test                r5d, r5d
-    mov                 r4d, %1
-    jz                  .loop0
-    sub                 r0, r1
-    add                 r4d, 3
-
-.loop0:
-    vbroadcasti128      m4, [r0]
-    vbroadcasti128      m5, [r0 + 8]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2], xm4
-
-    vbroadcasti128      m4, [r0 + 16]
-    vbroadcasti128      m5, [r0 + 24]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 16], xm4
-
-    vbroadcasti128      m4, [r0 + 32]
-    vbroadcasti128      m5, [r0 + 40]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 32], xm4
-
-    vbroadcasti128      m4, [r0 + 48]
-    vbroadcasti128      m5, [r0 + 56]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 48], xm4
-
-    add                 r2, r3
-    add                 r0, r1
-    dec                 r4d
-    jnz                 .loop0
-    RET
-%endif
-%endmacro
-
-IPFILTER_CHROMA_PS_32xN_AVX2 32
-IPFILTER_CHROMA_PS_32xN_AVX2 16
-IPFILTER_CHROMA_PS_32xN_AVX2 24
-IPFILTER_CHROMA_PS_32xN_AVX2 8
-IPFILTER_CHROMA_PS_32xN_AVX2 64
-IPFILTER_CHROMA_PS_32xN_AVX2 48
-
-
-%macro IPFILTER_CHROMA_PS_64xN_AVX2 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_horiz_ps_64x%1, 4, 7, 6
-    add                 r1d, r1d
-    add                 r3d, r3d
-    mov                 r4d, r4m
-    mov                 r5d, r5m
-
-%ifdef PIC
-    lea                 r6, [tab_ChromaCoeff]
-    vpbroadcastq        m0, [r6 + r4 * 8]
-%else
-    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova                m3, [interp8_hpp_shuf]
-    vbroadcasti128      m2, [INTERP_OFFSET_PS]
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    sub                 r0, 2
-    test                r5d, r5d
-    mov                 r4d, %1
-    jz                  .loop0
-    sub                 r0, r1
-    add                 r4d, 3
-
-.loop0:
-    vbroadcasti128      m4, [r0]
-    vbroadcasti128      m5, [r0 + 8]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2], xm4
-
-    vbroadcasti128      m4, [r0 + 16]
-    vbroadcasti128      m5, [r0 + 24]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 16], xm4
-
-    vbroadcasti128      m4, [r0 + 32]
-    vbroadcasti128      m5, [r0 + 40]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 32], xm4
-
-    vbroadcasti128      m4, [r0 + 48]
-    vbroadcasti128      m5, [r0 + 56]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 48], xm4
-
-    vbroadcasti128      m4, [r0 + 64]
-    vbroadcasti128      m5, [r0 + 72]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 64], xm4
-
-    vbroadcasti128      m4, [r0 + 80]
-    vbroadcasti128      m5, [r0 + 88]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 80], xm4
-
-    vbroadcasti128      m4, [r0 + 96]
-    vbroadcasti128      m5, [r0 + 104]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 96], xm4
-
-    vbroadcasti128      m4, [r0 + 112]
-    vbroadcasti128      m5, [r0 + 120]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 112], xm4
-
-    add                 r2, r3
-    add                 r0, r1
-    dec                 r4d
-    jnz                 .loop0
-    RET
-%endif
-%endmacro
-
-IPFILTER_CHROMA_PS_64xN_AVX2 64
-IPFILTER_CHROMA_PS_64xN_AVX2 48
-IPFILTER_CHROMA_PS_64xN_AVX2 32
-IPFILTER_CHROMA_PS_64xN_AVX2 16
-
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_horiz_ps_48x64, 4, 7, 6
-    add                 r1d, r1d
-    add                 r3d, r3d
-    mov                 r4d, r4m
-    mov                 r5d, r5m
-
-%ifdef PIC
-    lea                 r6, [tab_ChromaCoeff]
-    vpbroadcastq        m0, [r6 + r4 * 8]
-%else
-    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova                m3, [interp8_hpp_shuf]
-    vbroadcasti128      m2, [INTERP_OFFSET_PS]
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    sub                 r0, 2
-    test                r5d, r5d
-    mov                 r4d, 64
-    jz                  .loop0
-    sub                 r0, r1
-    add                 r4d, 3
-
-.loop0:
-    vbroadcasti128      m4, [r0]
-    vbroadcasti128      m5, [r0 + 8]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2], xm4
-
-    vbroadcasti128      m4, [r0 + 16]
-    vbroadcasti128      m5, [r0 + 24]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 16], xm4
-
-    vbroadcasti128      m4, [r0 + 32]
-    vbroadcasti128      m5, [r0 + 40]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 32], xm4
-
-    vbroadcasti128      m4, [r0 + 48]
-    vbroadcasti128      m5, [r0 + 56]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 48], xm4
-
-    vbroadcasti128      m4, [r0 + 64]
-    vbroadcasti128      m5, [r0 + 72]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 64], xm4
-
-    vbroadcasti128      m4, [r0 + 80]
-    vbroadcasti128      m5, [r0 + 88]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 80], xm4
-
-    add                 r2, r3
-    add                 r0, r1
-    dec                 r4d
-    jnz                 .loop0
-    RET
-%endif
-
-%macro IPFILTER_CHROMA_PS_6xN_AVX2 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_horiz_ps_6x%1, 4, 7, 6
-    add                 r1d, r1d
-    add                 r3d, r3d
-    mov                 r4d, r4m
-    mov                 r5d, r5m
-
-%ifdef PIC
-    lea                 r6, [tab_ChromaCoeff]
-    vpbroadcastq        m0, [r6 + r4 * 8]
-%else
-    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova                m3, [interp8_hpp_shuf]
-    vbroadcasti128      m2, [INTERP_OFFSET_PS]
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    sub                 r0, 2
-    test                r5d, r5d
-    mov                 r4d, %1
-    jz                  .loop0
-    sub                 r0, r1
-    add                 r4d, 3
-
-.loop0:
-    vbroadcasti128      m4, [r0]
-    vbroadcasti128      m5, [r0 + 8]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movq                [r2], xm4
-    pextrd              [r2 + 8], xm4, 2
-    add                 r2, r3
-    add                 r0, r1
-    dec                 r4d
-    jnz                 .loop0
-    RET
-%endif
-%endmacro
-
-    IPFILTER_CHROMA_PS_6xN_AVX2 8
-    IPFILTER_CHROMA_PS_6xN_AVX2 16
-
 %macro FILTER_VER_CHROMA_AVX2_8xN 2
 INIT_YMM avx2
 %if ARCH_X86_64 == 1

From f0bb1a50a4342bd5164b6a761ae42e9cd8cfa2ca Mon Sep 17 00:00:00 2001
From: Mythreyi P <mythreyi@multicorewareinc.com>
Date: Tue, 13 Feb 2018 00:28:25 -0800
Subject: [PATCH 49/51] x86: Split ipfilter16 kernals part2

Port horizonal 4tap kernals from h-ipfilter16.asm to a new source file, h4-ipfilter16.asm to improve build time.
---
 source/common/CMakeLists.txt        |    2 +-
 source/common/x86/h-ipfilter16.asm  | 3111 +++------------------------
 source/common/x86/h4-ipfilter16.asm | 2632 ++++++++++++++++++++++
 3 files changed, 2957 insertions(+), 2788 deletions(-)
 create mode 100644 source/common/x86/h4-ipfilter16.asm

diff --git a/source/common/CMakeLists.txt b/source/common/CMakeLists.txt
index 079b64982c..0eb2dfb8f2 100644
--- a/source/common/CMakeLists.txt
+++ b/source/common/CMakeLists.txt
@@ -61,7 +61,7 @@ if(ENABLE_ASSEMBLY AND X86)
                mc-a2.asm pixel-util8.asm blockcopy8.asm
                pixeladd8.asm dct8.asm seaintegral.asm)
     if(HIGH_BIT_DEPTH)
-        set(A_SRCS ${A_SRCS} sad16-a.asm intrapred16.asm h-ipfilter16.asm ipfilter16.asm loopfilter.asm)
+        set(A_SRCS ${A_SRCS} sad16-a.asm intrapred16.asm h4-ipfilter16.asm h-ipfilter16.asm ipfilter16.asm loopfilter.asm)
     else()
         set(A_SRCS ${A_SRCS} sad-a.asm intrapred8.asm intrapred8_allangs.asm v4-ipfilter8.asm h-ipfilter8.asm ipfilter8.asm loopfilter.asm)
     endif()
diff --git a/source/common/x86/h-ipfilter16.asm b/source/common/x86/h-ipfilter16.asm
index 92c329b036..8f593e1b67 100644
--- a/source/common/x86/h-ipfilter16.asm
+++ b/source/common/x86/h-ipfilter16.asm
@@ -45,20 +45,7 @@
 
 SECTION_RODATA 32
 
-tab_c_32:         times 8 dd 32
 h_pd_524800:        times 8 dd 524800
-
-tab_Tm16:         db 0, 1, 2, 3, 4,  5,  6, 7, 2, 3, 4,  5, 6, 7, 8, 9
-
-
-tab_ChromaCoeff:  dw  0, 64,  0,  0
-                  dw -2, 58, 10, -2
-                  dw -4, 54, 16, -2
-                  dw -6, 46, 28, -4
-                  dw -4, 36, 36, -4
-                  dw -4, 28, 46, -6
-                  dw -2, 16, 54, -4
-                  dw -2, 10, 58, -2
                                     
 tab_LumaCoeff:    dw   0, 0,  0,  64,  0,   0,  0,  0
                   dw  -1, 4, -10, 58,  17, -5,  1,  0
@@ -352,198 +339,6 @@ cglobal interp_8tap_horiz_%3_%1x%2, 4, 7, 8
     FILTER_HOR_LUMA_sse2 64, 32, ps
     FILTER_HOR_LUMA_sse2 64, 48, ps
     FILTER_HOR_LUMA_sse2 64, 64, ps
-    
-%macro FILTERH_W2_4_sse3 2
-    movh        m3,     [r0 + %1]
-    movhps      m3,     [r0 + %1 + 2]
-    pmaddwd     m3,     m0
-    movh        m4,     [r0 + r1 + %1]
-    movhps      m4,     [r0 + r1 + %1 + 2]
-    pmaddwd     m4,     m0
-    pshufd      m2,     m3,     q2301
-    paddd       m3,     m2
-    pshufd      m2,     m4,     q2301
-    paddd       m4,     m2
-    pshufd      m3,     m3,     q3120
-    pshufd      m4,     m4,     q3120
-    punpcklqdq  m3,     m4
-    paddd       m3,     m1
-    movh        m5,     [r0 + 2 * r1 + %1]
-    movhps      m5,     [r0 + 2 * r1 + %1 + 2]
-    pmaddwd     m5,     m0
-    movh        m4,     [r0 + r4 + %1]
-    movhps      m4,     [r0 + r4 + %1 + 2]
-    pmaddwd     m4,     m0
-    pshufd      m2,     m5,     q2301
-    paddd       m5,     m2
-    pshufd      m2,     m4,     q2301
-    paddd       m4,     m2
-    pshufd      m5,     m5,     q3120
-    pshufd      m4,     m4,     q3120
-    punpcklqdq  m5,     m4
-    paddd       m5,     m1
-%ifidn %2, pp
-    psrad       m3,     6
-    psrad       m5,     6
-    packssdw    m3,     m5
-    CLIPW       m3,     m7,     m6
-%else
-    psrad       m3,     INTERP_SHIFT_PS
-    psrad       m5,     INTERP_SHIFT_PS
-    packssdw    m3,     m5
-%endif
-    movd        [r2 + %1], m3
-    psrldq      m3,     4
-    movd        [r2 + r3 + %1], m3
-    psrldq      m3,     4
-    movd        [r2 + r3 * 2 + %1], m3
-    psrldq      m3,     4
-    movd        [r2 + r5 + %1], m3
-%endmacro
-
-%macro FILTERH_W2_3_sse3 1
-    movh        m3,     [r0 + %1]
-    movhps      m3,     [r0 + %1 + 2]
-    pmaddwd     m3,     m0
-    movh        m4,     [r0 + r1 + %1]
-    movhps      m4,     [r0 + r1 + %1 + 2]
-    pmaddwd     m4,     m0
-    pshufd      m2,     m3,     q2301
-    paddd       m3,     m2
-    pshufd      m2,     m4,     q2301
-    paddd       m4,     m2
-    pshufd      m3,     m3,     q3120
-    pshufd      m4,     m4,     q3120
-    punpcklqdq  m3,     m4
-    paddd       m3,     m1
-
-    movh        m5,     [r0 + 2 * r1 + %1]
-    movhps      m5,     [r0 + 2 * r1 + %1 + 2]
-    pmaddwd     m5,     m0
-
-    pshufd      m2,     m5,     q2301
-    paddd       m5,     m2
-    pshufd      m5,     m5,     q3120
-    paddd       m5,     m1
-
-    psrad       m3,     INTERP_SHIFT_PS
-    psrad       m5,     INTERP_SHIFT_PS
-    packssdw    m3,     m5
-
-    movd        [r2 + %1], m3
-    psrldq      m3,     4
-    movd        [r2 + r3 + %1], m3
-    psrldq      m3,     4
-    movd        [r2 + r3 * 2 + %1], m3
-%endmacro
-
-%macro FILTERH_W4_2_sse3 2
-    movh        m3,     [r0 + %1]
-    movhps      m3,     [r0 + %1 + 2]
-    pmaddwd     m3,     m0
-    movh        m4,     [r0 + %1 + 4]
-    movhps      m4,     [r0 + %1 + 6]
-    pmaddwd     m4,     m0
-    pshufd      m2,     m3,     q2301
-    paddd       m3,     m2
-    pshufd      m2,     m4,     q2301
-    paddd       m4,     m2
-    pshufd      m3,     m3,     q3120
-    pshufd      m4,     m4,     q3120
-    punpcklqdq  m3,     m4
-    paddd       m3,     m1
-
-    movh        m5,     [r0 + r1 + %1]
-    movhps      m5,     [r0 + r1 + %1 + 2]
-    pmaddwd     m5,     m0
-    movh        m4,     [r0 + r1 + %1 + 4]
-    movhps      m4,     [r0 + r1 + %1 + 6]
-    pmaddwd     m4,     m0
-    pshufd      m2,     m5,     q2301
-    paddd       m5,     m2
-    pshufd      m2,     m4,     q2301
-    paddd       m4,     m2
-    pshufd      m5,     m5,     q3120
-    pshufd      m4,     m4,     q3120
-    punpcklqdq  m5,     m4
-    paddd       m5,     m1
-%ifidn %2, pp
-    psrad       m3,     6
-    psrad       m5,     6
-    packssdw    m3,     m5
-    CLIPW       m3,     m7,     m6
-%else
-    psrad       m3,     INTERP_SHIFT_PS
-    psrad       m5,     INTERP_SHIFT_PS
-    packssdw    m3,     m5
-%endif
-    movh        [r2 + %1], m3
-    movhps      [r2 + r3 + %1], m3
-%endmacro
-
-%macro FILTERH_W4_1_sse3 1
-    movh        m3,     [r0 + 2 * r1 + %1]
-    movhps      m3,     [r0 + 2 * r1 + %1 + 2]
-    pmaddwd     m3,     m0
-    movh        m4,     [r0 + 2 * r1 + %1 + 4]
-    movhps      m4,     [r0 + 2 * r1 + %1 + 6]
-    pmaddwd     m4,     m0
-    pshufd      m2,     m3,     q2301
-    paddd       m3,     m2
-    pshufd      m2,     m4,     q2301
-    paddd       m4,     m2
-    pshufd      m3,     m3,     q3120
-    pshufd      m4,     m4,     q3120
-    punpcklqdq  m3,     m4
-    paddd       m3,     m1
-
-    psrad       m3,     INTERP_SHIFT_PS
-    packssdw    m3,     m3
-    movh        [r2 + r3 * 2 + %1], m3
-%endmacro
-
-%macro FILTERH_W8_1_sse3 2
-    movh        m3,     [r0 + %1]
-    movhps      m3,     [r0 + %1 + 2]
-    pmaddwd     m3,     m0
-    movh        m4,     [r0 + %1 + 4]
-    movhps      m4,     [r0 + %1 + 6]
-    pmaddwd     m4,     m0
-    pshufd      m2,     m3,     q2301
-    paddd       m3,     m2
-    pshufd      m2,     m4,     q2301
-    paddd       m4,     m2
-    pshufd      m3,     m3,     q3120
-    pshufd      m4,     m4,     q3120
-    punpcklqdq  m3,     m4
-    paddd       m3,     m1
-
-    movh        m5,     [r0 + %1 + 8]
-    movhps      m5,     [r0 + %1 + 10]
-    pmaddwd     m5,     m0
-    movh        m4,     [r0 + %1 + 12]
-    movhps      m4,     [r0 + %1 + 14]
-    pmaddwd     m4,     m0
-    pshufd      m2,     m5,     q2301
-    paddd       m5,     m2
-    pshufd      m2,     m4,     q2301
-    paddd       m4,     m2
-    pshufd      m5,     m5,     q3120
-    pshufd      m4,     m4,     q3120
-    punpcklqdq  m5,     m4
-    paddd       m5,     m1
-%ifidn %2, pp
-    psrad       m3,     6
-    psrad       m5,     6
-    packssdw    m3,     m5
-    CLIPW       m3,     m7,     m6
-%else
-    psrad       m3,     INTERP_SHIFT_PS
-    psrad       m5,     INTERP_SHIFT_PS
-    packssdw    m3,     m5
-%endif
-    movdqu      [r2 + %1], m3
-%endmacro
 
 ;-----------------------------------------------------------------------------
 ; void interp_4tap_horiz_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
@@ -652,96 +447,6 @@ cglobal interp_4tap_horiz_%3_%1x%2, 4, 7, 8
     RET
 %endmacro
 
-;-----------------------------------------------------------------------------
-; void interp_4tap_horiz_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-
-FILTER_HOR_CHROMA_sse3 2, 4, pp
-FILTER_HOR_CHROMA_sse3 2, 8, pp
-FILTER_HOR_CHROMA_sse3 2, 16, pp
-FILTER_HOR_CHROMA_sse3 4, 2, pp
-FILTER_HOR_CHROMA_sse3 4, 4, pp
-FILTER_HOR_CHROMA_sse3 4, 8, pp
-FILTER_HOR_CHROMA_sse3 4, 16, pp
-FILTER_HOR_CHROMA_sse3 4, 32, pp
-FILTER_HOR_CHROMA_sse3 6, 8, pp
-FILTER_HOR_CHROMA_sse3 6, 16, pp
-FILTER_HOR_CHROMA_sse3 8, 2, pp
-FILTER_HOR_CHROMA_sse3 8, 4, pp
-FILTER_HOR_CHROMA_sse3 8, 6, pp
-FILTER_HOR_CHROMA_sse3 8, 8, pp
-FILTER_HOR_CHROMA_sse3 8, 12, pp
-FILTER_HOR_CHROMA_sse3 8, 16, pp
-FILTER_HOR_CHROMA_sse3 8, 32, pp
-FILTER_HOR_CHROMA_sse3 8, 64, pp
-FILTER_HOR_CHROMA_sse3 12, 16, pp
-FILTER_HOR_CHROMA_sse3 12, 32, pp
-FILTER_HOR_CHROMA_sse3 16, 4, pp
-FILTER_HOR_CHROMA_sse3 16, 8, pp
-FILTER_HOR_CHROMA_sse3 16, 12, pp
-FILTER_HOR_CHROMA_sse3 16, 16, pp
-FILTER_HOR_CHROMA_sse3 16, 24, pp
-FILTER_HOR_CHROMA_sse3 16, 32, pp
-FILTER_HOR_CHROMA_sse3 16, 64, pp
-FILTER_HOR_CHROMA_sse3 24, 32, pp
-FILTER_HOR_CHROMA_sse3 24, 64, pp
-FILTER_HOR_CHROMA_sse3 32, 8, pp
-FILTER_HOR_CHROMA_sse3 32, 16, pp
-FILTER_HOR_CHROMA_sse3 32, 24, pp
-FILTER_HOR_CHROMA_sse3 32, 32, pp
-FILTER_HOR_CHROMA_sse3 32, 48, pp
-FILTER_HOR_CHROMA_sse3 32, 64, pp
-FILTER_HOR_CHROMA_sse3 48, 64, pp
-FILTER_HOR_CHROMA_sse3 64, 16, pp
-FILTER_HOR_CHROMA_sse3 64, 32, pp
-FILTER_HOR_CHROMA_sse3 64, 48, pp
-FILTER_HOR_CHROMA_sse3 64, 64, pp
-
-;-----------------------------------------------------------------------------
-; void interp_4tap_horiz_ps_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-
-FILTER_HOR_CHROMA_sse3 2, 4, ps
-FILTER_HOR_CHROMA_sse3 2, 8, ps
-FILTER_HOR_CHROMA_sse3 2, 16, ps
-FILTER_HOR_CHROMA_sse3 4, 2, ps
-FILTER_HOR_CHROMA_sse3 4, 4, ps
-FILTER_HOR_CHROMA_sse3 4, 8, ps
-FILTER_HOR_CHROMA_sse3 4, 16, ps
-FILTER_HOR_CHROMA_sse3 4, 32, ps
-FILTER_HOR_CHROMA_sse3 6, 8, ps
-FILTER_HOR_CHROMA_sse3 6, 16, ps
-FILTER_HOR_CHROMA_sse3 8, 2, ps
-FILTER_HOR_CHROMA_sse3 8, 4, ps
-FILTER_HOR_CHROMA_sse3 8, 6, ps
-FILTER_HOR_CHROMA_sse3 8, 8, ps
-FILTER_HOR_CHROMA_sse3 8, 12, ps
-FILTER_HOR_CHROMA_sse3 8, 16, ps
-FILTER_HOR_CHROMA_sse3 8, 32, ps
-FILTER_HOR_CHROMA_sse3 8, 64, ps
-FILTER_HOR_CHROMA_sse3 12, 16, ps
-FILTER_HOR_CHROMA_sse3 12, 32, ps
-FILTER_HOR_CHROMA_sse3 16, 4, ps
-FILTER_HOR_CHROMA_sse3 16, 8, ps
-FILTER_HOR_CHROMA_sse3 16, 12, ps
-FILTER_HOR_CHROMA_sse3 16, 16, ps
-FILTER_HOR_CHROMA_sse3 16, 24, ps
-FILTER_HOR_CHROMA_sse3 16, 32, ps
-FILTER_HOR_CHROMA_sse3 16, 64, ps
-FILTER_HOR_CHROMA_sse3 24, 32, ps
-FILTER_HOR_CHROMA_sse3 24, 64, ps
-FILTER_HOR_CHROMA_sse3 32, 8, ps
-FILTER_HOR_CHROMA_sse3 32, 16, ps
-FILTER_HOR_CHROMA_sse3 32, 24, ps
-FILTER_HOR_CHROMA_sse3 32, 32, ps
-FILTER_HOR_CHROMA_sse3 32, 48, ps
-FILTER_HOR_CHROMA_sse3 32, 64, ps
-FILTER_HOR_CHROMA_sse3 48, 64, ps
-FILTER_HOR_CHROMA_sse3 64, 16, ps
-FILTER_HOR_CHROMA_sse3 64, 32, ps
-FILTER_HOR_CHROMA_sse3 64, 48, ps
-FILTER_HOR_CHROMA_sse3 64, 64, ps
-
 %macro FILTER_P2S_2_4_sse2 1
     movd        m0,     [r0 + %1]
     movd        m2,     [r0 + r1 * 2 + %1]
@@ -1590,59 +1295,6 @@ FILTER_HOR_LUMA_W24 24, 32, pp
 ;----------------------------------------------------------------------------------------------------------------------------
 FILTER_HOR_LUMA_W24 24, 32, ps
 
-%macro FILTER_W2_2 1
-    movu        m3,         [r0]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + r1]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    packusdw    m3,         m3
-    CLIPW       m3,         m7,    m6
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    packssdw    m3,         m3
-%endif
-    movd        [r2],       m3
-    pextrd      [r2 + r3],  m3, 1
-%endmacro
-
-%macro FILTER_W4_2 1
-    movu        m3,         [r0]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 4]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + r1]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + r1 + 4]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m7,    m6
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2],       m3
-    movhps      [r2 + r3],  m3
-%endmacro
-
 ;-------------------------------------------------------------------------------------------------------------
 ; void interp_8tap_horiz_pp(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx
 ;-------------------------------------------------------------------------------------------------------------
@@ -2352,2019 +2004,137 @@ cglobal interp_8tap_horiz_pp_48x64, 4,6,8
     jnz              .loop
     RET
 
-;-----------------------------------------------------------------------------
-; void interp_4tap_horiz_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro FILTER_CHROMA_H 6
-INIT_XMM sse4
-cglobal interp_4tap_horiz_%3_%1x%2, 4, %4, %5
+;-----------------------------------------------------------------------------------------------------------------------------
+;void interp_horiz_ps_c(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt)
+;-----------------------------------------------------------------------------------------------------------------------------
 
-    add         r3,       r3
-    add         r1,       r1
-    sub         r0,       2
-    mov         r4d,      r4m
-    add         r4d,      r4d
+%macro IPFILTER_LUMA_PS_4xN_AVX2 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_horiz_ps_4x%1, 6,8,7
+    mov                         r5d,               r5m
+    mov                         r4d,               r4m
+    add                         r1d,               r1d
+    add                         r3d,               r3d
 
 %ifdef PIC
-    lea         r%6,      [tab_ChromaCoeff]
-    movh        m0,       [r%6 + r4 * 4]
+    lea                         r6,                [tab_LumaCoeff]
+    lea                         r4,                [r4 * 8]
+    vbroadcasti128              m0,                [r6 + r4 * 2]
 %else
-    movh        m0,       [tab_ChromaCoeff + r4 * 4]
+    lea                         r4,                [r4 * 8]
+    vbroadcasti128              m0,                [tab_LumaCoeff + r4 * 2]
 %endif
 
-    punpcklqdq  m0,       m0
-    mova        m2,       [tab_Tm16]
+    vbroadcasti128              m2,                [INTERP_OFFSET_PS]
 
-%ifidn %3, ps
-    mova        m1,       [INTERP_OFFSET_PS]
-    cmp         r5m, byte 0
-    je          .skip
-    sub         r0,       r1
-    movu        m3,       [r0]
-    pshufb      m3,       m3, m2
-    pmaddwd     m3,       m0
-
-  %if %1 == 4
-    movu        m4,       [r0 + 4]
-    pshufb      m4,       m4, m2
-    pmaddwd     m4,       m0
-    phaddd      m3,       m4
-  %else
-    phaddd      m3,       m3
-  %endif
+    ; register map
+    ; m0 - interpolate coeff
+    ; m1 - shuffle order table
+    ; m2 - pw_2000
 
-    paddd       m3,       m1
-    psrad       m3,       INTERP_SHIFT_PS
-    packssdw    m3,       m3
+    sub                         r0,                6
+    test                        r5d,               r5d
+    mov                         r7d,               %1                                    ; loop count variable - height
+    jz                         .preloop
+    lea                         r6,                [r1 * 3]                              ; r6 = (N / 2 - 1) * srcStride
+    sub                         r0,                r6                                    ; r0(src) - 3 * srcStride
+    add                         r7d,               6                                     ;7 - 1(since last row not in loop)                            ; need extra 7 rows, just set a specially flag here, blkheight += N - 1  (7 - 3 = 4 ; since the last three rows not in loop)
 
-  %if %1 == 2
-    movd        [r2],     m3
-  %else
-    movh        [r2],     m3
-  %endif
+.preloop:
+    lea                         r6,                [r3 * 3]
+.loop:
+    ; Row 0
+    movu                        xm3,                [r0]                                 ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    movu                        xm4,                [r0 + 2]                             ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    vinserti128                 m3,                 m3,                xm4,       1
+    movu                        xm4,                [r0 + 4]
+    movu                        xm5,                [r0 + 6]
+    vinserti128                 m4,                 m4,                xm5,       1
+    pmaddwd                     m3,                m0
+    pmaddwd                     m4,                m0
+    phaddd                      m3,                m4                                    ; DWORD [R1D R1C R0D R0C R1B R1A R0B R0A]
 
-    add         r0,       r1
-    add         r2,       r3
-    FILTER_W%1_2 %3
-    lea         r0,       [r0 + 2 * r1]
-    lea         r2,       [r2 + 2 * r3]
+    ; Row 1
+    movu                        xm4,                [r0 + r1]                            ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    movu                        xm5,                [r0 + r1 + 2]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+    vinserti128                 m4,                 m4,                xm5,       1
+    movu                        xm5,                [r0 + r1 + 4]
+    movu                        xm6,                [r0 + r1 + 6]
+    vinserti128                 m5,                 m5,                xm6,       1
+    pmaddwd                     m4,                m0
+    pmaddwd                     m5,                m0
+    phaddd                      m4,                m5                                     ; DWORD [R3D R3C R2D R2C R3B R3A R2B R2A]
+    phaddd                      m3,                m4                                     ; all rows and col completed.
 
-.skip:
+    mova                        m5,                [interp8_hps_shuf]
+    vpermd                      m3,                m5,                  m3
+    paddd                       m3,                m2
+    vextracti128                xm4,               m3,                  1
+    psrad                       xm3,               INTERP_SHIFT_PS
+    psrad                       xm4,               INTERP_SHIFT_PS
+    packssdw                    xm3,               xm3
+    packssdw                    xm4,               xm4
 
-%else     ;%ifidn %3, ps
-    pxor        m7,       m7
-    mova        m6,       [pw_pixel_max]
-    mova        m1,       [tab_c_32]
-%endif    ;%ifidn %3, ps
+    movq                        [r2],              xm3                                   ;row 0
+    movq                        [r2 + r3],         xm4                                   ;row 1
+    lea                         r0,                [r0 + r1 * 2]                         ; first loop src ->5th row(i.e 4)
+    lea                         r2,                [r2 + r3 * 2]                         ; first loop dst ->5th row(i.e 4)
 
-    FILTER_W%1_2 %3
+    sub                         r7d,               2
+    jg                          .loop
+    test                        r5d,               r5d
+    jz                          .end
 
-%rep (%2/2) - 1
-    lea         r0,       [r0 + 2 * r1]
-    lea         r2,       [r2 + 2 * r3]
-    FILTER_W%1_2 %3
-%endrep
-    RET
-%endmacro
+    ; Row 10
+    movu                        xm3,                [r0]
+    movu                        xm4,                [r0 + 2]
+    vinserti128                 m3,                 m3,                 xm4,      1
+    movu                        xm4,                [r0 + 4]
+    movu                        xm5,                [r0 + 6]
+    vinserti128                 m4,                 m4,                 xm5,      1
+    pmaddwd                     m3,                m0
+    pmaddwd                     m4,                m0
+    phaddd                      m3,                m4
 
-FILTER_CHROMA_H 2, 4, pp, 6, 8, 5
-FILTER_CHROMA_H 2, 8, pp, 6, 8, 5
-FILTER_CHROMA_H 4, 2, pp, 6, 8, 5
-FILTER_CHROMA_H 4, 4, pp, 6, 8, 5
-FILTER_CHROMA_H 4, 8, pp, 6, 8, 5
-FILTER_CHROMA_H 4, 16, pp, 6, 8, 5
-
-FILTER_CHROMA_H 2, 4, ps, 7, 5, 6
-FILTER_CHROMA_H 2, 8, ps, 7, 5, 6
-FILTER_CHROMA_H 4, 2, ps, 7, 6, 6
-FILTER_CHROMA_H 4, 4, ps, 7, 6, 6
-FILTER_CHROMA_H 4, 8, ps, 7, 6, 6
-FILTER_CHROMA_H 4, 16, ps, 7, 6, 6
-
-FILTER_CHROMA_H 2, 16, pp, 6, 8, 5
-FILTER_CHROMA_H 4, 32, pp, 6, 8, 5
-FILTER_CHROMA_H 2, 16, ps, 7, 5, 6
-FILTER_CHROMA_H 4, 32, ps, 7, 6, 6
-
-
-%macro FILTER_W6_1 1
-    movu        m3,         [r0]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 4]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m4,         [r0 + 8]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m4,         m4
-    paddd       m4,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m4,         INTERP_SHIFT_PP
-    packusdw    m3,         m4
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m4,         INTERP_SHIFT_PS
-    packssdw    m3,         m4
-%endif
-    movh        [r2],       m3
-    pextrd      [r2 + 8],   m3, 2
-%endmacro
+    ; Row11
+    phaddd                      m3,                m4                                    ; all rows and col completed.
 
-cglobal chroma_filter_pp_6x1_internal
-    FILTER_W6_1 pp
-    ret
-
-cglobal chroma_filter_ps_6x1_internal
-    FILTER_W6_1 ps
-    ret
-
-%macro FILTER_W8_1 1
-    movu        m3,         [r0]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 4]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + 8]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + 12]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2],       m3
-    movhps      [r2 + 8],   m3
-%endmacro
+    mova                        m5,                [interp8_hps_shuf]
+    vpermd                      m3,                m5,                  m3
+    paddd                       m3,                m2
+    vextracti128                xm4,               m3,                  1
+    psrad                       xm3,               INTERP_SHIFT_PS
+    psrad                       xm4,               INTERP_SHIFT_PS
+    packssdw                    xm3,               xm3
+    packssdw                    xm4,               xm4
 
-cglobal chroma_filter_pp_8x1_internal
-    FILTER_W8_1 pp
-    ret
-
-cglobal chroma_filter_ps_8x1_internal
-    FILTER_W8_1 ps
-    ret
-
-%macro FILTER_W12_1 1
-    movu        m3,         [r0]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 4]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + 8]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + 12]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2],       m3
-    movhps      [r2 + 8],   m3
-
-    movu        m3,         [r0 + 16]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 20]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    packusdw    m3,         m3
-    CLIPW       m3,         m6, m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    packssdw    m3,         m3
+    movq                        [r2],              xm3                                   ;row 0
+.end:
+    RET
 %endif
-    movh        [r2 + 16],  m3
 %endmacro
 
-cglobal chroma_filter_pp_12x1_internal
-    FILTER_W12_1 pp
-    ret
-
-cglobal chroma_filter_ps_12x1_internal
-    FILTER_W12_1 ps
-    ret
-
-%macro FILTER_W16_1 1
-    movu        m3,         [r0]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 4]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + 8]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + 12]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2],       m3
-    movhps      [r2 + 8],   m3
-
-    movu        m3,         [r0 + 16]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 20]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + 24]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + 28]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2 + 16],  m3
-    movhps      [r2 + 24],  m3
-%endmacro
+    IPFILTER_LUMA_PS_4xN_AVX2 4
+    IPFILTER_LUMA_PS_4xN_AVX2 8
+    IPFILTER_LUMA_PS_4xN_AVX2 16
 
-cglobal chroma_filter_pp_16x1_internal
-    FILTER_W16_1 pp
-    ret
-
-cglobal chroma_filter_ps_16x1_internal
-    FILTER_W16_1 ps
-    ret
-
-%macro FILTER_W24_1 1
-    movu        m3,         [r0]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 4]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + 8]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + 12]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2],       m3
-    movhps      [r2 + 8],   m3
-
-    movu        m3,         [r0 + 16]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 20]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + 24]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + 28]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2 + 16],  m3
-    movhps      [r2 + 24],  m3
-
-    movu        m3,         [r0 + 32]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 36]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + 40]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + 44]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m6,    m7
+%macro IPFILTER_LUMA_PS_8xN_AVX2 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_horiz_ps_8x%1, 4, 6, 8
+    add                 r1d, r1d
+    add                 r3d, r3d
+    mov                 r4d, r4m
+    mov                 r5d, r5m
+    shl                 r4d, 4
+%ifdef PIC
+    lea                 r6, [tab_LumaCoeff]
+    vpbroadcastq        m0, [r6 + r4]
+    vpbroadcastq        m1, [r6 + r4 + 8]
 %else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2 + 32],  m3
-    movhps      [r2 + 40],  m3
-%endmacro
-
-cglobal chroma_filter_pp_24x1_internal
-    FILTER_W24_1 pp
-    ret
-
-cglobal chroma_filter_ps_24x1_internal
-    FILTER_W24_1 ps
-    ret
-
-%macro FILTER_W32_1 1
-    movu        m3,         [r0]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 4]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + 8]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + 12]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2],       m3
-    movhps      [r2 + 8],   m3
-
-    movu        m3,         [r0 + 16]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 20]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + 24]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + 28]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2 + 16],  m3
-    movhps      [r2 + 24],  m3
-
-    movu        m3,         [r0 + 32]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 36]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + 40]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + 44]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2 + 32],  m3
-    movhps      [r2 + 40],  m3
-
-    movu        m3,         [r0 + 48]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + 52]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + 56]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + 60]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2 + 48],  m3
-    movhps      [r2 + 56],  m3
-%endmacro
-
-cglobal chroma_filter_pp_32x1_internal
-    FILTER_W32_1 pp
-    ret
-
-cglobal chroma_filter_ps_32x1_internal
-    FILTER_W32_1 ps
-    ret
-
-%macro FILTER_W8o_1 2
-    movu        m3,         [r0 + %2]
-    pshufb      m3,         m3, m2
-    pmaddwd     m3,         m0
-    movu        m4,         [r0 + %2 + 4]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m3,         m4
-    paddd       m3,         m1
-
-    movu        m5,         [r0 + %2 + 8]
-    pshufb      m5,         m5, m2
-    pmaddwd     m5,         m0
-    movu        m4,         [r0 + %2 + 12]
-    pshufb      m4,         m4, m2
-    pmaddwd     m4,         m0
-    phaddd      m5,         m4
-    paddd       m5,         m1
-%ifidn %1, pp
-    psrad       m3,         INTERP_SHIFT_PP
-    psrad       m5,         INTERP_SHIFT_PP
-    packusdw    m3,         m5
-    CLIPW       m3,         m6,    m7
-%else
-    psrad       m3,         INTERP_SHIFT_PS
-    psrad       m5,         INTERP_SHIFT_PS
-    packssdw    m3,         m5
-%endif
-    movh        [r2 + %2],       m3
-    movhps      [r2 + %2 + 8],   m3
-%endmacro
-
-%macro FILTER_W48_1 1
-    FILTER_W8o_1 %1, 0
-    FILTER_W8o_1 %1, 16
-    FILTER_W8o_1 %1, 32
-    FILTER_W8o_1 %1, 48
-    FILTER_W8o_1 %1, 64
-    FILTER_W8o_1 %1, 80
-%endmacro
-
-cglobal chroma_filter_pp_48x1_internal
-    FILTER_W48_1 pp
-    ret
-
-cglobal chroma_filter_ps_48x1_internal
-    FILTER_W48_1 ps
-    ret
-
-%macro FILTER_W64_1 1
-    FILTER_W8o_1 %1, 0
-    FILTER_W8o_1 %1, 16
-    FILTER_W8o_1 %1, 32
-    FILTER_W8o_1 %1, 48
-    FILTER_W8o_1 %1, 64
-    FILTER_W8o_1 %1, 80
-    FILTER_W8o_1 %1, 96
-    FILTER_W8o_1 %1, 112
-%endmacro
-
-cglobal chroma_filter_pp_64x1_internal
-    FILTER_W64_1 pp
-    ret
-
-cglobal chroma_filter_ps_64x1_internal
-    FILTER_W64_1 ps
-    ret
-    
-;-----------------------------------------------------------------------------
-; void interp_4tap_horiz_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-
-INIT_XMM sse4
-%macro IPFILTER_CHROMA 6
-cglobal interp_4tap_horiz_%3_%1x%2, 4, %5, %6
-
-    add         r3,        r3
-    add         r1,        r1
-    sub         r0,         2
-    mov         r4d,        r4m
-    add         r4d,        r4d
-
-%ifdef PIC
-    lea         r%4,       [tab_ChromaCoeff]
-    movh        m0,       [r%4 + r4 * 4]
-%else
-    movh        m0,       [tab_ChromaCoeff + r4 * 4]
-%endif
-
-    punpcklqdq  m0,       m0
-    mova        m2,       [tab_Tm16]
-
-%ifidn %3, ps
-    mova        m1,       [INTERP_OFFSET_PS]
-    cmp         r5m, byte 0
-    je          .skip
-    sub         r0, r1
-    call chroma_filter_%3_%1x1_internal
-    add         r0, r1
-    add         r2, r3
-    call chroma_filter_%3_%1x1_internal
-    add         r0, r1
-    add         r2, r3
-    call chroma_filter_%3_%1x1_internal
-    add         r0, r1
-    add         r2, r3
-.skip:
-%else
-    mova        m1,         [tab_c_32]
-    pxor        m6,         m6
-    mova        m7,         [pw_pixel_max]
-%endif
-
-    call chroma_filter_%3_%1x1_internal
-%rep %2 - 1
-    add         r0,       r1
-    add         r2,       r3
-    call chroma_filter_%3_%1x1_internal
-%endrep
-RET
-%endmacro
-IPFILTER_CHROMA 6, 8, pp, 5, 6, 8
-IPFILTER_CHROMA 8, 2, pp, 5, 6, 8
-IPFILTER_CHROMA 8, 4, pp, 5, 6, 8
-IPFILTER_CHROMA 8, 6, pp, 5, 6, 8
-IPFILTER_CHROMA 8, 8, pp, 5, 6, 8
-IPFILTER_CHROMA 8, 16, pp, 5, 6, 8
-IPFILTER_CHROMA 8, 32, pp, 5, 6, 8
-IPFILTER_CHROMA 12, 16, pp, 5, 6, 8
-IPFILTER_CHROMA 16, 4, pp, 5, 6, 8
-IPFILTER_CHROMA 16, 8, pp, 5, 6, 8
-IPFILTER_CHROMA 16, 12, pp, 5, 6, 8
-IPFILTER_CHROMA 16, 16, pp, 5, 6, 8
-IPFILTER_CHROMA 16, 32, pp, 5, 6, 8
-IPFILTER_CHROMA 24, 32, pp, 5, 6, 8
-IPFILTER_CHROMA 32, 8, pp, 5, 6, 8
-IPFILTER_CHROMA 32, 16, pp, 5, 6, 8
-IPFILTER_CHROMA 32, 24, pp, 5, 6, 8
-IPFILTER_CHROMA 32, 32, pp, 5, 6, 8
-
-IPFILTER_CHROMA 6, 8, ps, 6, 7, 6
-IPFILTER_CHROMA 8, 2, ps, 6, 7, 6
-IPFILTER_CHROMA 8, 4, ps, 6, 7, 6
-IPFILTER_CHROMA 8, 6, ps, 6, 7, 6
-IPFILTER_CHROMA 8, 8, ps, 6, 7, 6
-IPFILTER_CHROMA 8, 16, ps, 6, 7, 6
-IPFILTER_CHROMA 8, 32, ps, 6, 7, 6
-IPFILTER_CHROMA 12, 16, ps, 6, 7, 6
-IPFILTER_CHROMA 16, 4, ps, 6, 7, 6
-IPFILTER_CHROMA 16, 8, ps, 6, 7, 6
-IPFILTER_CHROMA 16, 12, ps, 6, 7, 6
-IPFILTER_CHROMA 16, 16, ps, 6, 7, 6
-IPFILTER_CHROMA 16, 32, ps, 6, 7, 6
-IPFILTER_CHROMA 24, 32, ps, 6, 7, 6
-IPFILTER_CHROMA 32, 8, ps, 6, 7, 6
-IPFILTER_CHROMA 32, 16, ps, 6, 7, 6
-IPFILTER_CHROMA 32, 24, ps, 6, 7, 6
-IPFILTER_CHROMA 32, 32, ps, 6, 7, 6
-
-IPFILTER_CHROMA 6, 16, pp, 5, 6, 8
-IPFILTER_CHROMA 8, 12, pp, 5, 6, 8
-IPFILTER_CHROMA 8, 64, pp, 5, 6, 8
-IPFILTER_CHROMA 12, 32, pp, 5, 6, 8
-IPFILTER_CHROMA 16, 24, pp, 5, 6, 8
-IPFILTER_CHROMA 16, 64, pp, 5, 6, 8
-IPFILTER_CHROMA 24, 64, pp, 5, 6, 8
-IPFILTER_CHROMA 32, 48, pp, 5, 6, 8
-IPFILTER_CHROMA 32, 64, pp, 5, 6, 8
-IPFILTER_CHROMA 6, 16, ps, 6, 7, 6
-IPFILTER_CHROMA 8, 12, ps, 6, 7, 6
-IPFILTER_CHROMA 8, 64, ps, 6, 7, 6
-IPFILTER_CHROMA 12, 32, ps, 6, 7, 6
-IPFILTER_CHROMA 16, 24, ps, 6, 7, 6
-IPFILTER_CHROMA 16, 64, ps, 6, 7, 6
-IPFILTER_CHROMA 24, 64, ps, 6, 7, 6
-IPFILTER_CHROMA 32, 48, ps, 6, 7, 6
-IPFILTER_CHROMA 32, 64, ps, 6, 7, 6
-
-IPFILTER_CHROMA 48, 64, pp, 5, 6, 8
-IPFILTER_CHROMA 64, 48, pp, 5, 6, 8
-IPFILTER_CHROMA 64, 64, pp, 5, 6, 8
-IPFILTER_CHROMA 64, 32, pp, 5, 6, 8
-IPFILTER_CHROMA 64, 16, pp, 5, 6, 8
-IPFILTER_CHROMA 48, 64, ps, 6, 7, 6
-IPFILTER_CHROMA 64, 48, ps, 6, 7, 6
-IPFILTER_CHROMA 64, 64, ps, 6, 7, 6
-IPFILTER_CHROMA 64, 32, ps, 6, 7, 6
-IPFILTER_CHROMA 64, 16, ps, 6, 7, 6
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-%macro IPFILTER_CHROMA_avx2_6xN 1
-cglobal interp_4tap_horiz_pp_6x%1, 5,6,8
-    add             r1d, r1d
-    add             r3d, r3d
-    sub             r0, 2
-    mov             r4d, r4m
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeff]
-    vpbroadcastq    m0, [r5 + r4 * 8]
-%else
-    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova            m1, [interp8_hpp_shuf]
-    vpbroadcastd    m2, [pd_32]
-    pxor            m5, m5
-    mova            m6, [idct8_shuf2]
-    mova            m7, [pw_pixel_max]
-
-    mov             r4d, %1/2
-.loop:
-    vbroadcasti128  m3, [r0]
-    vbroadcasti128  m4, [r0 + 8]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, INTERP_SHIFT_PP           ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m3, m3
-    vpermq          m3, m3, q2020
-    pshufb          xm3, xm6                      ; m3 = WORD[7 6 5 4 3 2 1 0]
-    CLIPW           xm3, xm5, xm7
-    movq            [r2], xm3
-    pextrd          [r2 + 8], xm3, 2
-
-    vbroadcasti128  m3, [r0 + r1]
-    vbroadcasti128  m4, [r0 + r1 + 8]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, INTERP_SHIFT_PP           ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m3, m3
-    vpermq          m3, m3, q2020
-    pshufb          xm3, xm6                      ; m3 = WORD[7 6 5 4 3 2 1 0]
-    CLIPW           xm3, xm5, xm7
-    movq            [r2 + r3], xm3
-    pextrd          [r2 + r3 + 8], xm3, 2
-
-    lea             r0, [r0 + r1 * 2]
-    lea             r2, [r2 + r3 * 2]
-    dec             r4d
-    jnz             .loop
-    RET
-%endmacro
-IPFILTER_CHROMA_avx2_6xN 8
-IPFILTER_CHROMA_avx2_6xN 16
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_8x2, 5,6,8
-    add             r1d, r1d
-    add             r3d, r3d
-    sub             r0, 2
-    mov             r4d, r4m
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeff]
-    vpbroadcastq    m0, [r5 + r4 * 8]
-%else
-    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova            m1, [interp8_hpp_shuf]
-    vpbroadcastd    m2, [pd_32]
-    pxor            m5, m5
-    mova            m6, [idct8_shuf2]
-    mova            m7, [pw_pixel_max]
-
-    vbroadcasti128  m3, [r0]
-    vbroadcasti128  m4, [r0 + 8]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, INTERP_SHIFT_PP          ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m3, m3
-    vpermq          m3, m3,q2020
-    pshufb          xm3, xm6                     ; m3 = WORD[7 6 5 4 3 2 1 0]
-    CLIPW           xm3, xm5, xm7
-    movu            [r2], xm3
-
-    vbroadcasti128  m3, [r0 + r1]
-    vbroadcasti128  m4, [r0 + r1 + 8]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, INTERP_SHIFT_PP           ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m3, m3
-    vpermq          m3, m3,q2020
-    pshufb          xm3, xm6                      ; m3 = WORD[7 6 5 4 3 2 1 0]
-    CLIPW           xm3, xm5, xm7
-    movu            [r2 + r3], xm3
-    RET
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-cglobal interp_4tap_horiz_pp_8x4, 5,6,8
-    add             r1d, r1d
-    add             r3d, r3d
-    sub             r0, 2
-    mov             r4d, r4m
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeff]
-    vpbroadcastq    m0, [r5 + r4 * 8]
-%else
-    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova            m1, [interp8_hpp_shuf]
-    vpbroadcastd    m2, [pd_32]
-    pxor            m5, m5
-    mova            m6, [idct8_shuf2]
-    mova            m7, [pw_pixel_max]
-
-%rep 2
-    vbroadcasti128  m3, [r0]
-    vbroadcasti128  m4, [r0 + 8]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m3, m3
-    vpermq          m3, m3,q2020
-    pshufb          xm3, xm6                    ; m3 = WORD[7 6 5 4 3 2 1 0]
-    CLIPW           xm3, xm5, xm7
-    movu            [r2], xm3
-
-    vbroadcasti128  m3, [r0 + r1]
-    vbroadcasti128  m4, [r0 + r1 + 8]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m3, m3
-    vpermq          m3, m3,q2020
-    pshufb          xm3, xm6                    ; m3 = WORD[7 6 5 4 3 2 1 0]
-    CLIPW           xm3, xm5, xm7
-    movu            [r2 + r3], xm3
-
-    lea             r0, [r0 + r1 * 2]
-    lea             r2, [r2 + r3 * 2]
-%endrep
-    RET
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-%macro IPFILTER_CHROMA_avx2_8xN 1
-cglobal interp_4tap_horiz_pp_8x%1, 5,6,8
-    add             r1d, r1d
-    add             r3d, r3d
-    sub             r0, 2
-    mov             r4d, r4m
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeff]
-    vpbroadcastq    m0, [r5 + r4 * 8]
-%else
-    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova            m1, [interp8_hpp_shuf]
-    vpbroadcastd    m2, [pd_32]
-    pxor            m5, m5
-    mova            m6, [idct8_shuf2]
-    mova            m7, [pw_pixel_max]
-
-    mov             r4d, %1/2
-.loop:
-    vbroadcasti128  m3, [r0]
-    vbroadcasti128  m4, [r0 + 8]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, 6                         ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m3, m3
-    vpermq          m3, m3, q2020
-    pshufb          xm3, xm6                      ; m3 = WORD[7 6 5 4 3 2 1 0]
-    CLIPW           xm3, xm5, xm7
-    movu            [r2], xm3
-
-    vbroadcasti128  m3, [r0 + r1]
-    vbroadcasti128  m4, [r0 + r1 + 8]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, 6                         ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m3, m3
-    vpermq          m3, m3, q2020
-    pshufb          xm3, xm6                      ; m3 = WORD[7 6 5 4 3 2 1 0]
-    CLIPW           xm3, xm5, xm7
-    movu            [r2 + r3], xm3
-
-    lea             r0, [r0 + r1 * 2]
-    lea             r2, [r2 + r3 * 2]
-    dec             r4d
-    jnz             .loop
-    RET
-%endmacro
-IPFILTER_CHROMA_avx2_8xN 6
-IPFILTER_CHROMA_avx2_8xN 8
-IPFILTER_CHROMA_avx2_8xN 12
-IPFILTER_CHROMA_avx2_8xN 16
-IPFILTER_CHROMA_avx2_8xN 32
-IPFILTER_CHROMA_avx2_8xN 64
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-%macro IPFILTER_CHROMA_avx2_16xN 1
-%if ARCH_X86_64
-cglobal interp_4tap_horiz_pp_16x%1, 5,6,9
-    add             r1d, r1d
-    add             r3d, r3d
-    sub             r0, 2
-    mov             r4d, r4m
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeff]
-    vpbroadcastq    m0, [r5 + r4 * 8]
-%else
-    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova            m1, [interp8_hpp_shuf]
-    vpbroadcastd    m2, [pd_32]
-    pxor            m5, m5
-    mova            m6, [idct8_shuf2]
-    mova            m7, [pw_pixel_max]
-
-    mov             r4d, %1
-.loop:
-    vbroadcasti128  m3, [r0]
-    vbroadcasti128  m4, [r0 + 8]
-
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m3, m3
-    vpermq          m3, m3, q2020
-    pshufb          xm3, xm6                     ; m3 = WORD[7 6 5 4 3 2 1 0]
-
-    vbroadcasti128  m4, [r0 + 16]
-    vbroadcasti128  m8, [r0 + 24]
-
-    pshufb          m4, m1
-    pshufb          m8, m1
-
-    pmaddwd         m4, m0
-    pmaddwd         m8, m0
-    phaddd          m4, m8
-    paddd           m4, m2
-    psrad           m4, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m4, m4
-    vpermq          m4, m4, q2020
-    pshufb          xm4, xm6                    ; m3 = WORD[7 6 5 4 3 2 1 0]
-    vinserti128     m3, m3, xm4, 1
-    CLIPW           m3, m5, m7
-    movu            [r2], m3
-
-    add             r0, r1
-    add             r2, r3
-    dec             r4d
-    jnz             .loop
-    RET
-%endif
-%endmacro
-IPFILTER_CHROMA_avx2_16xN 4
-IPFILTER_CHROMA_avx2_16xN 8
-IPFILTER_CHROMA_avx2_16xN 12
-IPFILTER_CHROMA_avx2_16xN 16
-IPFILTER_CHROMA_avx2_16xN 24
-IPFILTER_CHROMA_avx2_16xN 32
-IPFILTER_CHROMA_avx2_16xN 64
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-%macro IPFILTER_CHROMA_avx2_32xN 1
-%if ARCH_X86_64
-cglobal interp_4tap_horiz_pp_32x%1, 5,6,9
-    add             r1d, r1d
-    add             r3d, r3d
-    sub             r0, 2
-    mov             r4d, r4m
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeff]
-    vpbroadcastq    m0, [r5 + r4 * 8]
-%else
-    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova            m1, [interp8_hpp_shuf]
-    vpbroadcastd    m2, [pd_32]
-    pxor            m5, m5
-    mova            m6, [idct8_shuf2]
-    mova            m7, [pw_pixel_max]
-
-    mov             r6d, %1
-.loop:
-%assign x 0
-%rep 2
-    vbroadcasti128  m3, [r0 + x]
-    vbroadcasti128  m4, [r0 + 8 + x]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m3, m3
-    vpermq          m3, m3, q2020
-    pshufb          xm3, xm6                     ; m3 = WORD[7 6 5 4 3 2 1 0]
-
-    vbroadcasti128  m4, [r0 + 16 + x]
-    vbroadcasti128  m8, [r0 + 24 + x]
-    pshufb          m4, m1
-    pshufb          m8, m1
-
-    pmaddwd         m4, m0
-    pmaddwd         m8, m0
-    phaddd          m4, m8
-    paddd           m4, m2
-    psrad           m4, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m4, m4
-    vpermq          m4, m4, q2020
-    pshufb          xm4, xm6                    ; m3 = WORD[7 6 5 4 3 2 1 0]
-    vinserti128     m3, m3, xm4, 1
-    CLIPW           m3, m5, m7
-    movu            [r2 + x], m3
-    %assign x x+32
-    %endrep
-
-    add             r0, r1
-    add             r2, r3
-    dec             r6d
-    jnz             .loop
-    RET
-%endif
-%endmacro
-IPFILTER_CHROMA_avx2_32xN 8
-IPFILTER_CHROMA_avx2_32xN 16
-IPFILTER_CHROMA_avx2_32xN 24
-IPFILTER_CHROMA_avx2_32xN 32
-IPFILTER_CHROMA_avx2_32xN 48
-IPFILTER_CHROMA_avx2_32xN 64
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-%macro IPFILTER_CHROMA_avx2_12xN 1
-%if ARCH_X86_64
-cglobal interp_4tap_horiz_pp_12x%1, 5,6,8
-    add             r1d, r1d
-    add             r3d, r3d
-    sub             r0, 2
-    mov             r4d, r4m
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeff]
-    vpbroadcastq    m0, [r5 + r4 * 8]
-%else
-    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova            m1, [interp8_hpp_shuf]
-    vpbroadcastd    m2, [pd_32]
-    pxor            m5, m5
-    mova            m6, [idct8_shuf2]
-    mova            m7, [pw_pixel_max]
-
-    mov             r4d, %1
-.loop:
-    vbroadcasti128  m3, [r0]
-    vbroadcasti128  m4, [r0 + 8]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m3, m3
-    vpermq          m3, m3, q2020
-    pshufb          xm3, xm6                     ; m3 = WORD[7 6 5 4 3 2 1 0]
-    CLIPW           xm3, xm5, xm7
-    movu            [r2], xm3
-
-    vbroadcasti128  m3, [r0 + 16]
-    vbroadcasti128  m4, [r0 + 24]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
-
-    packusdw        m3, m3
-    vpermq          m3, m3, q2020
-    pshufb          xm3, xm6                    ; m3 = WORD[7 6 5 4 3 2 1 0]
-    CLIPW           xm3, xm5, xm7
-    movq            [r2 + 16], xm3
-
-    add             r0, r1
-    add             r2, r3
-    dec             r4d
-    jnz             .loop
-    RET
-%endif
-%endmacro
-IPFILTER_CHROMA_avx2_12xN 16
-IPFILTER_CHROMA_avx2_12xN 32
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-%macro IPFILTER_CHROMA_avx2_24xN 1
-%if ARCH_X86_64
-cglobal interp_4tap_horiz_pp_24x%1, 5,6,9
-    add             r1d, r1d
-    add             r3d, r3d
-    sub             r0, 2
-    mov             r4d, r4m
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeff]
-    vpbroadcastq    m0, [r5 + r4 * 8]
-%else
-    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova            m1, [interp8_hpp_shuf]
-    vpbroadcastd    m2, [pd_32]
-    pxor            m5, m5
-    mova            m6, [idct8_shuf2]
-    mova            m7, [pw_pixel_max]
-
-    mov             r4d, %1
-.loop:
-    vbroadcasti128  m3, [r0]
-    vbroadcasti128  m4, [r0 + 8]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, 6
-
-    vbroadcasti128  m4, [r0 + 16]
-    vbroadcasti128  m8, [r0 + 24]
-    pshufb          m4, m1
-    pshufb          m8, m1
-
-    pmaddwd         m4, m0
-    pmaddwd         m8, m0
-    phaddd          m4, m8
-    paddd           m4, m2
-    psrad           m4, 6
-
-    packusdw        m3, m4
-    vpermq          m3, m3, q3120
-    pshufb          m3, m6
-    CLIPW           m3, m5, m7
-    movu            [r2], m3
-
-    vbroadcasti128  m3, [r0 + 32]
-    vbroadcasti128  m4, [r0 + 40]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, 6
-
-    packusdw        m3, m3
-    vpermq          m3, m3, q2020
-    pshufb          xm3, xm6
-    CLIPW           xm3, xm5, xm7
-    movu            [r2 + 32], xm3
-
-    add             r0, r1
-    add             r2, r3
-    dec             r4d
-    jnz             .loop
-    RET
-%endif
-%endmacro
-IPFILTER_CHROMA_avx2_24xN 32
-IPFILTER_CHROMA_avx2_24xN 64
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-%macro IPFILTER_CHROMA_avx2_64xN 1
-%if ARCH_X86_64
-cglobal interp_4tap_horiz_pp_64x%1, 5,6,9
-    add             r1d, r1d
-    add             r3d, r3d
-    sub             r0, 2
-    mov             r4d, r4m
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeff]
-    vpbroadcastq    m0, [r5 + r4 * 8]
-%else
-    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova            m1, [interp8_hpp_shuf]
-    vpbroadcastd    m2, [pd_32]
-    pxor            m5, m5
-    mova            m6, [idct8_shuf2]
-    mova            m7, [pw_pixel_max]
-
-    mov             r6d, %1
-.loop:
-%assign x 0
-%rep 4
-    vbroadcasti128  m3, [r0 + x]
-    vbroadcasti128  m4, [r0 + 8 + x]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, 6
-
-    vbroadcasti128  m4, [r0 + 16 + x]
-    vbroadcasti128  m8, [r0 + 24 + x]
-    pshufb          m4, m1
-    pshufb          m8, m1
-
-    pmaddwd         m4, m0
-    pmaddwd         m8, m0
-    phaddd          m4, m8
-    paddd           m4, m2
-    psrad           m4, 6
-
-    packusdw        m3, m4
-    vpermq          m3, m3, q3120
-    pshufb          m3, m6
-    CLIPW           m3, m5, m7
-    movu            [r2 + x], m3
-    %assign x x+32
-    %endrep
-
-    add             r0, r1
-    add             r2, r3
-    dec             r6d
-    jnz             .loop
-    RET
-%endif
-%endmacro
-IPFILTER_CHROMA_avx2_64xN 16
-IPFILTER_CHROMA_avx2_64xN 32
-IPFILTER_CHROMA_avx2_64xN 48
-IPFILTER_CHROMA_avx2_64xN 64
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
-;-------------------------------------------------------------------------------------------------------------
-INIT_YMM avx2
-%if ARCH_X86_64
-cglobal interp_4tap_horiz_pp_48x64, 5,6,9
-    add             r1d, r1d
-    add             r3d, r3d
-    sub             r0, 2
-    mov             r4d, r4m
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeff]
-    vpbroadcastq    m0, [r5 + r4 * 8]
-%else
-    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova            m1, [interp8_hpp_shuf]
-    vpbroadcastd    m2, [pd_32]
-    pxor            m5, m5
-    mova            m6, [idct8_shuf2]
-    mova            m7, [pw_pixel_max]
-
-    mov             r4d, 64
-.loop:
-%assign x 0
-%rep 3
-    vbroadcasti128  m3, [r0 + x]
-    vbroadcasti128  m4, [r0 + 8 + x]
-    pshufb          m3, m1
-    pshufb          m4, m1
-
-    pmaddwd         m3, m0
-    pmaddwd         m4, m0
-    phaddd          m3, m4
-    paddd           m3, m2
-    psrad           m3, 6
-
-    vbroadcasti128  m4, [r0 + 16 + x]
-    vbroadcasti128  m8, [r0 + 24 + x]
-    pshufb          m4, m1
-    pshufb          m8, m1
-
-    pmaddwd         m4, m0
-    pmaddwd         m8, m0
-    phaddd          m4, m8
-    paddd           m4, m2
-    psrad           m4, 6
-
-    packusdw        m3, m4
-    vpermq          m3, m3, q3120
-    pshufb          m3, m6
-    CLIPW           m3, m5, m7
-    movu            [r2 + x], m3
-%assign x x+32
-%endrep
-
-    add             r0, r1
-    add             r2, r3
-    dec             r4d
-    jnz             .loop
-    RET
-%endif
-
-;-----------------------------------------------------------------------------------------------------------------------------
-;void interp_horiz_ps_c(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt)
-;-----------------------------------------------------------------------------------------------------------------------------
-
-%macro IPFILTER_LUMA_PS_4xN_AVX2 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_horiz_ps_4x%1, 6,8,7
-    mov                         r5d,               r5m
-    mov                         r4d,               r4m
-    add                         r1d,               r1d
-    add                         r3d,               r3d
-
-%ifdef PIC
-    lea                         r6,                [tab_LumaCoeff]
-    lea                         r4,                [r4 * 8]
-    vbroadcasti128              m0,                [r6 + r4 * 2]
-%else
-    lea                         r4,                [r4 * 8]
-    vbroadcasti128              m0,                [tab_LumaCoeff + r4 * 2]
-%endif
-
-    vbroadcasti128              m2,                [INTERP_OFFSET_PS]
-
-    ; register map
-    ; m0 - interpolate coeff
-    ; m1 - shuffle order table
-    ; m2 - pw_2000
-
-    sub                         r0,                6
-    test                        r5d,               r5d
-    mov                         r7d,               %1                                    ; loop count variable - height
-    jz                         .preloop
-    lea                         r6,                [r1 * 3]                              ; r6 = (N / 2 - 1) * srcStride
-    sub                         r0,                r6                                    ; r0(src) - 3 * srcStride
-    add                         r7d,               6                                     ;7 - 1(since last row not in loop)                            ; need extra 7 rows, just set a specially flag here, blkheight += N - 1  (7 - 3 = 4 ; since the last three rows not in loop)
-
-.preloop:
-    lea                         r6,                [r3 * 3]
-.loop:
-    ; Row 0
-    movu                        xm3,                [r0]                                 ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    movu                        xm4,                [r0 + 2]                             ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    vinserti128                 m3,                 m3,                xm4,       1
-    movu                        xm4,                [r0 + 4]
-    movu                        xm5,                [r0 + 6]
-    vinserti128                 m4,                 m4,                xm5,       1
-    pmaddwd                     m3,                m0
-    pmaddwd                     m4,                m0
-    phaddd                      m3,                m4                                    ; DWORD [R1D R1C R0D R0C R1B R1A R0B R0A]
-
-    ; Row 1
-    movu                        xm4,                [r0 + r1]                            ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    movu                        xm5,                [r0 + r1 + 2]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
-    vinserti128                 m4,                 m4,                xm5,       1
-    movu                        xm5,                [r0 + r1 + 4]
-    movu                        xm6,                [r0 + r1 + 6]
-    vinserti128                 m5,                 m5,                xm6,       1
-    pmaddwd                     m4,                m0
-    pmaddwd                     m5,                m0
-    phaddd                      m4,                m5                                     ; DWORD [R3D R3C R2D R2C R3B R3A R2B R2A]
-    phaddd                      m3,                m4                                     ; all rows and col completed.
-
-    mova                        m5,                [interp8_hps_shuf]
-    vpermd                      m3,                m5,                  m3
-    paddd                       m3,                m2
-    vextracti128                xm4,               m3,                  1
-    psrad                       xm3,               INTERP_SHIFT_PS
-    psrad                       xm4,               INTERP_SHIFT_PS
-    packssdw                    xm3,               xm3
-    packssdw                    xm4,               xm4
-
-    movq                        [r2],              xm3                                   ;row 0
-    movq                        [r2 + r3],         xm4                                   ;row 1
-    lea                         r0,                [r0 + r1 * 2]                         ; first loop src ->5th row(i.e 4)
-    lea                         r2,                [r2 + r3 * 2]                         ; first loop dst ->5th row(i.e 4)
-
-    sub                         r7d,               2
-    jg                          .loop
-    test                        r5d,               r5d
-    jz                          .end
-
-    ; Row 10
-    movu                        xm3,                [r0]
-    movu                        xm4,                [r0 + 2]
-    vinserti128                 m3,                 m3,                 xm4,      1
-    movu                        xm4,                [r0 + 4]
-    movu                        xm5,                [r0 + 6]
-    vinserti128                 m4,                 m4,                 xm5,      1
-    pmaddwd                     m3,                m0
-    pmaddwd                     m4,                m0
-    phaddd                      m3,                m4
-
-    ; Row11
-    phaddd                      m3,                m4                                    ; all rows and col completed.
-
-    mova                        m5,                [interp8_hps_shuf]
-    vpermd                      m3,                m5,                  m3
-    paddd                       m3,                m2
-    vextracti128                xm4,               m3,                  1
-    psrad                       xm3,               INTERP_SHIFT_PS
-    psrad                       xm4,               INTERP_SHIFT_PS
-    packssdw                    xm3,               xm3
-    packssdw                    xm4,               xm4
-
-    movq                        [r2],              xm3                                   ;row 0
-.end:
-    RET
-%endif
-%endmacro
-
-    IPFILTER_LUMA_PS_4xN_AVX2 4
-    IPFILTER_LUMA_PS_4xN_AVX2 8
-    IPFILTER_LUMA_PS_4xN_AVX2 16
-
-%macro IPFILTER_LUMA_PS_8xN_AVX2 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_horiz_ps_8x%1, 4, 6, 8
-    add                 r1d, r1d
-    add                 r3d, r3d
-    mov                 r4d, r4m
-    mov                 r5d, r5m
-    shl                 r4d, 4
-%ifdef PIC
-    lea                 r6, [tab_LumaCoeff]
-    vpbroadcastq        m0, [r6 + r4]
-    vpbroadcastq        m1, [r6 + r4 + 8]
-%else
-    vpbroadcastq        m0, [tab_LumaCoeff + r4]
-    vpbroadcastq        m1, [tab_LumaCoeff + r4 + 8]
-%endif
-    mova                m3, [interp8_hpp_shuf]
-    vbroadcasti128      m2, [INTERP_OFFSET_PS]
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    sub                 r0, 6
-    test                r5d, r5d
-    mov                 r4d, %1
-    jz                  .loop0
-    lea                 r6, [r1*3]
-    sub                 r0, r6
-    add                 r4d, 7
-
-.loop0:
-    vbroadcasti128      m4, [r0]
-    vbroadcasti128      m5, [r0 + 8]
-    pshufb              m4, m3
-    pshufb              m7, m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m7, m1
-    paddd               m4, m7
-
-    vbroadcasti128      m6, [r0 + 16]
-    pshufb              m5, m3
-    pshufb              m6, m3
-    pmaddwd             m5, m0
-    pmaddwd             m6, m1
-    paddd               m5, m6
-
-    phaddd              m4, m5
-    vpermq              m4, m4, q3120
-    paddd               m4, m2
-    vextracti128        xm5,m4, 1
-    psrad               xm4, INTERP_SHIFT_PS
-    psrad               xm5, INTERP_SHIFT_PS
-    packssdw            xm4, xm5
-
-    movu                [r2], xm4
-    add                 r2, r3
-    add                 r0, r1
-    dec                 r4d
-    jnz                 .loop0
-    RET
-%endif
-%endmacro
-
-    IPFILTER_LUMA_PS_8xN_AVX2 4
-    IPFILTER_LUMA_PS_8xN_AVX2 8
-    IPFILTER_LUMA_PS_8xN_AVX2 16
-    IPFILTER_LUMA_PS_8xN_AVX2 32
-
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_horiz_ps_24x32, 4, 6, 8
-    add                 r1d, r1d
-    add                 r3d, r3d
-    mov                 r4d, r4m
-    mov                 r5d, r5m
-    shl                 r4d, 4
-%ifdef PIC
-    lea                 r6, [tab_LumaCoeff]
-    vpbroadcastq        m0, [r6 + r4]
-    vpbroadcastq        m1, [r6 + r4 + 8]
-%else
-    vpbroadcastq        m0, [tab_LumaCoeff + r4]
-    vpbroadcastq        m1, [tab_LumaCoeff + r4 + 8]
-%endif
-    mova                m3, [interp8_hpp_shuf]
-    vbroadcasti128      m2, [INTERP_OFFSET_PS]
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    sub                 r0, 6
-    test                r5d, r5d
-    mov                 r4d, 32
-    jz                  .loop0
-    lea                 r6, [r1*3]
-    sub                 r0, r6
-    add                 r4d, 7
-
-.loop0:
-%assign x 0
-%rep 24/8
-    vbroadcasti128      m4, [r0 + x]
-    vbroadcasti128      m5, [r0 + 8 + x]
-    pshufb              m4, m3
-    pshufb              m7, m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m7, m1
-    paddd               m4, m7
-
-    vbroadcasti128      m6, [r0 + 16 + x]
-    pshufb              m5, m3
-    pshufb              m6, m3
-    pmaddwd             m5, m0
-    pmaddwd             m6, m1
-    paddd               m5, m6
-
-    phaddd              m4, m5
-    vpermq              m4, m4, q3120
-    paddd               m4, m2
-    vextracti128        xm5,m4, 1
-    psrad               xm4, INTERP_SHIFT_PS
-    psrad               xm5, INTERP_SHIFT_PS
-    packssdw            xm4, xm5
-
-    movu                [r2 + x], xm4
-    %assign x x+16
-    %endrep
-
-    add                 r2, r3
-    add                 r0, r1
-    dec                 r4d
-    jnz                 .loop0
-    RET
-%endif
-%macro IPFILTER_LUMA_PS_32_64_AVX2 2
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_horiz_ps_%1x%2, 4, 6, 8
-
-    add                 r1d, r1d
-    add                 r3d, r3d
-    mov                 r4d, r4m
-    mov                 r5d, r5m
-    shl                 r4d, 6
-%ifdef PIC
-    lea                 r6, [h_tab_LumaCoeffV]
-    movu                m0, [r6 + r4]
-    movu                m1, [r6 + r4 + mmsize]
-%else
-    movu                m0, [h_tab_LumaCoeffV + r4]
-    movu                m1, [h_tab_LumaCoeffV + r4 + mmsize]
-%endif
-    mova                m3, [interp8_hpp_shuf_new]
-    vbroadcasti128      m2, [INTERP_OFFSET_PS]
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    sub                 r0, 6
-    test                r5d, r5d
-    mov                 r4d, %2
-    jz                 .loop0
-    lea                 r6, [r1*3]
-    sub                 r0, r6
-    add                 r4d, 7
-
-.loop0:
-%assign x 0
-%rep %1/16
-    vbroadcasti128      m4, [r0 + x]
-    vbroadcasti128      m5, [r0 + 4 * SIZEOF_PIXEL + x]
-    pshufb              m4, m3
-    pshufb              m5, m3
-
-    pmaddwd             m4, m0
-    pmaddwd             m7, m5, m1
-    paddd               m4, m7
-    vextracti128        xm7, m4, 1
-    paddd               xm4, xm7
-    paddd               xm4, xm2
-    psrad               xm4, INTERP_SHIFT_PS
-
-    vbroadcasti128      m6, [r0 + 16 + x]
-    pshufb              m6, m3
-
-    pmaddwd             m5, m0
-    pmaddwd             m7, m6, m1
-    paddd               m5, m7
-    vextracti128        xm7, m5, 1
-    paddd               xm5, xm7
-    paddd               xm5, xm2
-    psrad               xm5, INTERP_SHIFT_PS
-
-    packssdw            xm4, xm5
-    movu                [r2 + x], xm4
-
-    vbroadcasti128      m5, [r0 + 24 + x]
-    pshufb              m5, m3
-
-    pmaddwd             m6, m0
-    pmaddwd             m7, m5, m1
-    paddd               m6, m7
-    vextracti128        xm7, m6, 1
-    paddd               xm6, xm7
-    paddd               xm6, xm2
-    psrad               xm6, INTERP_SHIFT_PS
-
-    vbroadcasti128      m7, [r0 + 32 + x]
-    pshufb              m7, m3
-
-    pmaddwd             m5, m0
-    pmaddwd             m7, m1
-    paddd               m5, m7
-    vextracti128        xm7, m5, 1
-    paddd               xm5, xm7
-    paddd               xm5, xm2
-    psrad               xm5, INTERP_SHIFT_PS
-
-    packssdw            xm6, xm5
-    movu                [r2 + 16 + x], xm6
-
-%assign x x+32
-%endrep
-
-    add                 r2, r3
-    add                 r0, r1
-    dec                 r4d
-    jnz                .loop0
-    RET
-%endif
-%endmacro
-
-    IPFILTER_LUMA_PS_32_64_AVX2 32, 8
-    IPFILTER_LUMA_PS_32_64_AVX2 32, 16
-    IPFILTER_LUMA_PS_32_64_AVX2 32, 24
-    IPFILTER_LUMA_PS_32_64_AVX2 32, 32
-    IPFILTER_LUMA_PS_32_64_AVX2 32, 64
-
-    IPFILTER_LUMA_PS_32_64_AVX2 64, 16
-    IPFILTER_LUMA_PS_32_64_AVX2 64, 32
-    IPFILTER_LUMA_PS_32_64_AVX2 64, 48
-    IPFILTER_LUMA_PS_32_64_AVX2 64, 64
-
-    IPFILTER_LUMA_PS_32_64_AVX2 48, 64
-
-%macro IPFILTER_LUMA_PS_16xN_AVX2 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_horiz_ps_16x%1, 4, 6, 8
-
-    add                 r1d, r1d
-    add                 r3d, r3d
-    mov                 r4d, r4m
-    mov                 r5d, r5m
-    shl                 r4d, 4
-%ifdef PIC
-    lea                 r6, [tab_LumaCoeff]
-    vpbroadcastq        m0, [r6 + r4]
-    vpbroadcastq        m1, [r6 + r4 + 8]
-%else
-    vpbroadcastq        m0, [tab_LumaCoeff + r4]
-    vpbroadcastq        m1, [tab_LumaCoeff + r4 + 8]
-%endif
-    mova                m3, [interp8_hpp_shuf]
-    vbroadcasti128      m2, [INTERP_OFFSET_PS]
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    sub                 r0, 6
-    test                r5d, r5d
-    mov                 r4d, %1
-    jz                  .loop0
-    lea                 r6, [r1*3]
-    sub                 r0, r6
-    add                 r4d, 7
-
-.loop0:
-    vbroadcasti128      m4, [r0]
-    vbroadcasti128      m5, [r0 + 8]
-    pshufb              m4, m3
-    pshufb              m7, m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m7, m1
-    paddd               m4, m7
-
-    vbroadcasti128      m6, [r0 + 16]
-    pshufb              m5, m3
-    pshufb              m7, m6, m3
-    pmaddwd             m5, m0
-    pmaddwd             m7, m1
-    paddd               m5, m7
-
-    phaddd              m4, m5
-    vpermq              m4, m4, q3120
-    paddd               m4, m2
-    vextracti128        xm5, m4, 1
-    psrad               xm4, INTERP_SHIFT_PS
-    psrad               xm5, INTERP_SHIFT_PS
-    packssdw            xm4, xm5
-    movu                [r2], xm4
-
-    vbroadcasti128      m5, [r0 + 24]
-    pshufb              m6, m3
-    pshufb              m7, m5, m3
-    pmaddwd             m6, m0
-    pmaddwd             m7, m1
-    paddd               m6, m7
-
-    vbroadcasti128      m7, [r0 + 32]
-    pshufb              m5, m3
-    pshufb              m7, m3
-    pmaddwd             m5, m0
-    pmaddwd             m7, m1
-    paddd               m5, m7
-
-    phaddd              m6, m5
-    vpermq              m6, m6, q3120
-    paddd               m6, m2
-    vextracti128        xm5,m6, 1
-    psrad               xm6, INTERP_SHIFT_PS
-    psrad               xm5, INTERP_SHIFT_PS
-    packssdw            xm6, xm5
-    movu                [r2 + 16], xm6
-
-    add                 r2, r3
-    add                 r0, r1
-    dec                 r4d
-    jnz                 .loop0
-    RET
-%endif
-%endmacro
-
-    IPFILTER_LUMA_PS_16xN_AVX2 4
-    IPFILTER_LUMA_PS_16xN_AVX2 8
-    IPFILTER_LUMA_PS_16xN_AVX2 12
-    IPFILTER_LUMA_PS_16xN_AVX2 16
-    IPFILTER_LUMA_PS_16xN_AVX2 32
-    IPFILTER_LUMA_PS_16xN_AVX2 64
-
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_horiz_ps_12x16, 4, 6, 8
-    add                 r1d, r1d
-    add                 r3d, r3d
-    mov                 r4d, r4m
-    mov                 r5d, r5m
-    shl                 r4d, 4
-%ifdef PIC
-    lea                 r6, [tab_LumaCoeff]
-    vpbroadcastq        m0, [r6 + r4]
-    vpbroadcastq        m1, [r6 + r4 + 8]
-%else
-    vpbroadcastq        m0, [tab_LumaCoeff + r4]
-    vpbroadcastq        m1, [tab_LumaCoeff + r4 + 8]
-%endif
-    mova                m3, [interp8_hpp_shuf]
-    vbroadcasti128      m2, [INTERP_OFFSET_PS]
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    sub                 r0, 6
-    test                r5d, r5d
-    mov                 r4d, 16
-    jz                  .loop0
-    lea                 r6, [r1*3]
-    sub                 r0, r6
-    add                 r4d, 7
-
-.loop0:
-    vbroadcasti128      m4, [r0]
-    vbroadcasti128      m5, [r0 + 8]
-    pshufb              m4, m3
-    pshufb              m7, m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m7, m1
-    paddd               m4, m7
-
-    vbroadcasti128      m6, [r0 + 16]
-    pshufb              m5, m3
-    pshufb              m7, m6, m3
-    pmaddwd             m5, m0
-    pmaddwd             m7, m1
-    paddd               m5, m7
-
-    phaddd              m4, m5
-    vpermq              m4, m4, q3120
-    paddd               m4, m2
-    vextracti128        xm5,m4, 1
-    psrad               xm4, INTERP_SHIFT_PS
-    psrad               xm5, INTERP_SHIFT_PS
-    packssdw            xm4, xm5
-    movu                [r2], xm4
-
-    vbroadcasti128      m5, [r0 + 24]
-    pshufb              m6, m3
-    pshufb              m5, m3
-    pmaddwd             m6, m0
-    pmaddwd             m5, m1
-    paddd               m6, m5
-
-    phaddd              m6, m6
-    vpermq              m6, m6, q3120
-    paddd               xm6, xm2
-    psrad               xm6, INTERP_SHIFT_PS
-    packssdw            xm6, xm6
-    movq                [r2 + 16], xm6
-
-    add                 r2, r3
-    add                 r0, r1
-    dec                 r4d
-    jnz                 .loop0
-    RET
-%endif
-%macro IPFILTER_CHROMA_PS_8xN_AVX2 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_horiz_ps_8x%1, 4, 7, 6
-    add                 r1d, r1d
-    add                 r3d, r3d
-    mov                 r4d, r4m
-    mov                 r5d, r5m
-
-%ifdef PIC
-    lea                 r6, [tab_ChromaCoeff]
-    vpbroadcastq        m0, [r6 + r4 * 8]
-%else
-    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova                m3, [interp8_hpp_shuf]
-    vbroadcasti128      m2, [INTERP_OFFSET_PS]
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    sub                 r0, 2
-    test                r5d, r5d
-    mov                 r4d, %1
-    jz                  .loop0
-    sub                 r0, r1
-    add                 r4d, 3
-
-.loop0:
-    vbroadcasti128      m4, [r0]
-    vbroadcasti128      m5, [r0 + 8]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2], xm4
-    add                 r2, r3
-    add                 r0, r1
-    dec                 r4d
-    jnz                 .loop0
-    RET
-%endif
-%endmacro
-
-    IPFILTER_CHROMA_PS_8xN_AVX2 4
-    IPFILTER_CHROMA_PS_8xN_AVX2 8
-    IPFILTER_CHROMA_PS_8xN_AVX2 16
-    IPFILTER_CHROMA_PS_8xN_AVX2 32
-    IPFILTER_CHROMA_PS_8xN_AVX2 6
-    IPFILTER_CHROMA_PS_8xN_AVX2 2
-    IPFILTER_CHROMA_PS_8xN_AVX2 12
-    IPFILTER_CHROMA_PS_8xN_AVX2 64
-
-%macro IPFILTER_CHROMA_PS_16xN_AVX2 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_horiz_ps_16x%1, 4, 7, 6
-    add                 r1d, r1d
-    add                 r3d, r3d
-    mov                 r4d, r4m
-    mov                 r5d, r5m
-
-%ifdef PIC
-    lea                 r6, [tab_ChromaCoeff]
-    vpbroadcastq        m0, [r6 + r4 * 8]
-%else
-    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
+    vpbroadcastq        m0, [tab_LumaCoeff + r4]
+    vpbroadcastq        m1, [tab_LumaCoeff + r4 + 8]
 %endif
     mova                m3, [interp8_hpp_shuf]
     vbroadcasti128      m2, [INTERP_OFFSET_PS]
@@ -4372,194 +2142,39 @@ cglobal interp_4tap_horiz_ps_16x%1, 4, 7, 6
     ; register map
     ; m0 , m1 interpolate coeff
 
-    sub                 r0, 2
+    sub                 r0, 6
     test                r5d, r5d
     mov                 r4d, %1
     jz                  .loop0
-    sub                 r0, r1
-    add                 r4d, 3
+    lea                 r6, [r1*3]
+    sub                 r0, r6
+    add                 r4d, 7
 
 .loop0:
     vbroadcasti128      m4, [r0]
     vbroadcasti128      m5, [r0 + 8]
     pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2], xm4
-
-    vbroadcasti128      m4, [r0 + 16]
-    vbroadcasti128      m5, [r0 + 24]
-    pshufb              m4, m3
-    pshufb              m5, m3
+    pshufb              m7, m5, m3
     pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 16], xm4
-
-    add                 r2, r3
-    add                 r0, r1
-    dec                 r4d
-    jnz                 .loop0
-    RET
-%endif
-%endmacro
-
-IPFILTER_CHROMA_PS_16xN_AVX2 16
-IPFILTER_CHROMA_PS_16xN_AVX2 8
-IPFILTER_CHROMA_PS_16xN_AVX2 32
-IPFILTER_CHROMA_PS_16xN_AVX2 12
-IPFILTER_CHROMA_PS_16xN_AVX2 4
-IPFILTER_CHROMA_PS_16xN_AVX2 64
-IPFILTER_CHROMA_PS_16xN_AVX2 24
-
-%macro IPFILTER_CHROMA_PS_24xN_AVX2 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_horiz_ps_24x%1, 4, 7, 6
-    add                 r1d, r1d
-    add                 r3d, r3d
-    mov                 r4d, r4m
-    mov                 r5d, r5m
-
-%ifdef PIC
-    lea                 r6, [tab_ChromaCoeff]
-    vpbroadcastq        m0, [r6 + r4 * 8]
-%else
-    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova                m3, [interp8_hpp_shuf]
-    vbroadcasti128      m2, [INTERP_OFFSET_PS]
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    sub                 r0, 2
-    test                r5d, r5d
-    mov                 r4d, %1
-    jz                  .loop0
-    sub                 r0, r1
-    add                 r4d, 3
+    pmaddwd             m7, m1
+    paddd               m4, m7
 
-.loop0:
-    vbroadcasti128      m4, [r0]
-    vbroadcasti128      m5, [r0 + 8]
-    pshufb              m4, m3
+    vbroadcasti128      m6, [r0 + 16]
     pshufb              m5, m3
-    pmaddwd             m4, m0
+    pshufb              m6, m3
     pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2], xm4
+    pmaddwd             m6, m1
+    paddd               m5, m6
 
-    vbroadcasti128      m4, [r0 + 16]
-    vbroadcasti128      m5, [r0 + 24]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
     phaddd              m4, m5
-    paddd               m4, m2
     vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 16], xm4
-
-    vbroadcasti128      m4, [r0 + 32]
-    vbroadcasti128      m5, [r0 + 40]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
     paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
+    vextracti128        xm5,m4, 1
+    psrad               xm4, INTERP_SHIFT_PS
+    psrad               xm5, INTERP_SHIFT_PS
     packssdw            xm4, xm5
-    movu                [r2 + 32], xm4
-
-    add                 r2, r3
-    add                 r0, r1
-    dec                 r4d
-    jnz                 .loop0
-    RET
-%endif
-%endmacro
-
-IPFILTER_CHROMA_PS_24xN_AVX2 32
-IPFILTER_CHROMA_PS_24xN_AVX2 64
-
-%macro IPFILTER_CHROMA_PS_12xN_AVX2 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_horiz_ps_12x%1, 4, 7, 6
-    add                 r1d, r1d
-    add                 r3d, r3d
-    mov                 r4d, r4m
-    mov                 r5d, r5m
-
-%ifdef PIC
-    lea                 r6, [tab_ChromaCoeff]
-    vpbroadcastq        m0, [r6 + r4 * 8]
-%else
-    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
-%endif
-    mova                m3, [interp8_hpp_shuf]
-    vbroadcasti128      m2, [INTERP_OFFSET_PS]
-
-    ; register map
-    ; m0 , m1 interpolate coeff
-
-    sub                 r0, 2
-    test                r5d, r5d
-    mov                 r4d, %1
-    jz                  .loop0
-    sub                 r0, r1
-    add                 r4d, 3
 
-.loop0:
-    vbroadcasti128      m4, [r0]
-    vbroadcasti128      m5, [r0 + 8]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
     movu                [r2], xm4
-
-    vbroadcasti128      m4, [r0 + 16]
-    pshufb              m4, m3
-    pmaddwd             m4, m0
-    phaddd              m4, m4
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movq                [r2 + 16], xm4
-
     add                 r2, r3
     add                 r0, r1
     dec                 r4d
@@ -4568,23 +2183,26 @@ cglobal interp_4tap_horiz_ps_12x%1, 4, 7, 6
 %endif
 %endmacro
 
-IPFILTER_CHROMA_PS_12xN_AVX2 16
-IPFILTER_CHROMA_PS_12xN_AVX2 32
+    IPFILTER_LUMA_PS_8xN_AVX2 4
+    IPFILTER_LUMA_PS_8xN_AVX2 8
+    IPFILTER_LUMA_PS_8xN_AVX2 16
+    IPFILTER_LUMA_PS_8xN_AVX2 32
 
-%macro IPFILTER_CHROMA_PS_32xN_AVX2 1
 INIT_YMM avx2
 %if ARCH_X86_64 == 1
-cglobal interp_4tap_horiz_ps_32x%1, 4, 7, 6
+cglobal interp_8tap_horiz_ps_24x32, 4, 6, 8
     add                 r1d, r1d
     add                 r3d, r3d
     mov                 r4d, r4m
     mov                 r5d, r5m
-
+    shl                 r4d, 4
 %ifdef PIC
-    lea                 r6, [tab_ChromaCoeff]
-    vpbroadcastq        m0, [r6 + r4 * 8]
+    lea                 r6, [tab_LumaCoeff]
+    vpbroadcastq        m0, [r6 + r4]
+    vpbroadcastq        m1, [r6 + r4 + 8]
 %else
-    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
+    vpbroadcastq        m0, [tab_LumaCoeff + r4]
+    vpbroadcastq        m1, [tab_LumaCoeff + r4 + 8]
 %endif
     mova                m3, [interp8_hpp_shuf]
     vbroadcasti128      m2, [INTERP_OFFSET_PS]
@@ -4592,69 +2210,43 @@ cglobal interp_4tap_horiz_ps_32x%1, 4, 7, 6
     ; register map
     ; m0 , m1 interpolate coeff
 
-    sub                 r0, 2
+    sub                 r0, 6
     test                r5d, r5d
-    mov                 r4d, %1
+    mov                 r4d, 32
     jz                  .loop0
-    sub                 r0, r1
-    add                 r4d, 3
+    lea                 r6, [r1*3]
+    sub                 r0, r6
+    add                 r4d, 7
 
 .loop0:
-    vbroadcasti128      m4, [r0]
-    vbroadcasti128      m5, [r0 + 8]
+%assign x 0
+%rep 24/8
+    vbroadcasti128      m4, [r0 + x]
+    vbroadcasti128      m5, [r0 + 8 + x]
     pshufb              m4, m3
-    pshufb              m5, m3
+    pshufb              m7, m5, m3
     pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2], xm4
+    pmaddwd             m7, m1
+    paddd               m4, m7
 
-    vbroadcasti128      m4, [r0 + 16]
-    vbroadcasti128      m5, [r0 + 24]
-    pshufb              m4, m3
+    vbroadcasti128      m6, [r0 + 16 + x]
     pshufb              m5, m3
-    pmaddwd             m4, m0
+    pshufb              m6, m3
     pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 16], xm4
+    pmaddwd             m6, m1
+    paddd               m5, m6
 
-    vbroadcasti128      m4, [r0 + 32]
-    vbroadcasti128      m5, [r0 + 40]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
     phaddd              m4, m5
-    paddd               m4, m2
     vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 32], xm4
-
-    vbroadcasti128      m4, [r0 + 48]
-    vbroadcasti128      m5, [r0 + 56]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
     paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
+    vextracti128        xm5,m4, 1
+    psrad               xm4, INTERP_SHIFT_PS
+    psrad               xm5, INTERP_SHIFT_PS
     packssdw            xm4, xm5
-    movu                [r2 + 48], xm4
+
+    movu                [r2 + x], xm4
+    %assign x x+16
+    %endrep
 
     add                 r2, r3
     add                 r0, r1
@@ -4662,183 +2254,134 @@ cglobal interp_4tap_horiz_ps_32x%1, 4, 7, 6
     jnz                 .loop0
     RET
 %endif
-%endmacro
-
-IPFILTER_CHROMA_PS_32xN_AVX2 32
-IPFILTER_CHROMA_PS_32xN_AVX2 16
-IPFILTER_CHROMA_PS_32xN_AVX2 24
-IPFILTER_CHROMA_PS_32xN_AVX2 8
-IPFILTER_CHROMA_PS_32xN_AVX2 64
-IPFILTER_CHROMA_PS_32xN_AVX2 48
-
-
-%macro IPFILTER_CHROMA_PS_64xN_AVX2 1
+%macro IPFILTER_LUMA_PS_32_64_AVX2 2
 INIT_YMM avx2
 %if ARCH_X86_64 == 1
-cglobal interp_4tap_horiz_ps_64x%1, 4, 7, 6
+cglobal interp_8tap_horiz_ps_%1x%2, 4, 6, 8
+
     add                 r1d, r1d
     add                 r3d, r3d
     mov                 r4d, r4m
     mov                 r5d, r5m
-
+    shl                 r4d, 6
 %ifdef PIC
-    lea                 r6, [tab_ChromaCoeff]
-    vpbroadcastq        m0, [r6 + r4 * 8]
+    lea                 r6, [h_tab_LumaCoeffV]
+    movu                m0, [r6 + r4]
+    movu                m1, [r6 + r4 + mmsize]
 %else
-    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
+    movu                m0, [h_tab_LumaCoeffV + r4]
+    movu                m1, [h_tab_LumaCoeffV + r4 + mmsize]
 %endif
-    mova                m3, [interp8_hpp_shuf]
+    mova                m3, [interp8_hpp_shuf_new]
     vbroadcasti128      m2, [INTERP_OFFSET_PS]
 
     ; register map
     ; m0 , m1 interpolate coeff
 
-    sub                 r0, 2
+    sub                 r0, 6
     test                r5d, r5d
-    mov                 r4d, %1
-    jz                  .loop0
-    sub                 r0, r1
-    add                 r4d, 3
-
-.loop0:
-    vbroadcasti128      m4, [r0]
-    vbroadcasti128      m5, [r0 + 8]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2], xm4
-
-    vbroadcasti128      m4, [r0 + 16]
-    vbroadcasti128      m5, [r0 + 24]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 16], xm4
+    mov                 r4d, %2
+    jz                 .loop0
+    lea                 r6, [r1*3]
+    sub                 r0, r6
+    add                 r4d, 7
 
-    vbroadcasti128      m4, [r0 + 32]
-    vbroadcasti128      m5, [r0 + 40]
+.loop0:
+%assign x 0
+%rep %1/16
+    vbroadcasti128      m4, [r0 + x]
+    vbroadcasti128      m5, [r0 + 4 * SIZEOF_PIXEL + x]
     pshufb              m4, m3
     pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 32], xm4
 
-    vbroadcasti128      m4, [r0 + 48]
-    vbroadcasti128      m5, [r0 + 56]
-    pshufb              m4, m3
-    pshufb              m5, m3
     pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 48], xm4
+    pmaddwd             m7, m5, m1
+    paddd               m4, m7
+    vextracti128        xm7, m4, 1
+    paddd               xm4, xm7
+    paddd               xm4, xm2
+    psrad               xm4, INTERP_SHIFT_PS
 
-    vbroadcasti128      m4, [r0 + 64]
-    vbroadcasti128      m5, [r0 + 72]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 64], xm4
+    vbroadcasti128      m6, [r0 + 16 + x]
+    pshufb              m6, m3
 
-    vbroadcasti128      m4, [r0 + 80]
-    vbroadcasti128      m5, [r0 + 88]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
     pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 80], xm4
+    pmaddwd             m7, m6, m1
+    paddd               m5, m7
+    vextracti128        xm7, m5, 1
+    paddd               xm5, xm7
+    paddd               xm5, xm2
+    psrad               xm5, INTERP_SHIFT_PS
 
-    vbroadcasti128      m4, [r0 + 96]
-    vbroadcasti128      m5, [r0 + 104]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
     packssdw            xm4, xm5
-    movu                [r2 + 96], xm4
+    movu                [r2 + x], xm4
 
-    vbroadcasti128      m4, [r0 + 112]
-    vbroadcasti128      m5, [r0 + 120]
-    pshufb              m4, m3
+    vbroadcasti128      m5, [r0 + 24 + x]
     pshufb              m5, m3
-    pmaddwd             m4, m0
+
+    pmaddwd             m6, m0
+    pmaddwd             m7, m5, m1
+    paddd               m6, m7
+    vextracti128        xm7, m6, 1
+    paddd               xm6, xm7
+    paddd               xm6, xm2
+    psrad               xm6, INTERP_SHIFT_PS
+
+    vbroadcasti128      m7, [r0 + 32 + x]
+    pshufb              m7, m3
+
     pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 112], xm4
+    pmaddwd             m7, m1
+    paddd               m5, m7
+    vextracti128        xm7, m5, 1
+    paddd               xm5, xm7
+    paddd               xm5, xm2
+    psrad               xm5, INTERP_SHIFT_PS
+
+    packssdw            xm6, xm5
+    movu                [r2 + 16 + x], xm6
+
+%assign x x+32
+%endrep
 
     add                 r2, r3
     add                 r0, r1
     dec                 r4d
-    jnz                 .loop0
+    jnz                .loop0
     RET
 %endif
 %endmacro
 
-IPFILTER_CHROMA_PS_64xN_AVX2 64
-IPFILTER_CHROMA_PS_64xN_AVX2 48
-IPFILTER_CHROMA_PS_64xN_AVX2 32
-IPFILTER_CHROMA_PS_64xN_AVX2 16
+    IPFILTER_LUMA_PS_32_64_AVX2 32, 8
+    IPFILTER_LUMA_PS_32_64_AVX2 32, 16
+    IPFILTER_LUMA_PS_32_64_AVX2 32, 24
+    IPFILTER_LUMA_PS_32_64_AVX2 32, 32
+    IPFILTER_LUMA_PS_32_64_AVX2 32, 64
+
+    IPFILTER_LUMA_PS_32_64_AVX2 64, 16
+    IPFILTER_LUMA_PS_32_64_AVX2 64, 32
+    IPFILTER_LUMA_PS_32_64_AVX2 64, 48
+    IPFILTER_LUMA_PS_32_64_AVX2 64, 64
+
+    IPFILTER_LUMA_PS_32_64_AVX2 48, 64
 
+%macro IPFILTER_LUMA_PS_16xN_AVX2 1
 INIT_YMM avx2
 %if ARCH_X86_64 == 1
-cglobal interp_4tap_horiz_ps_48x64, 4, 7, 6
+cglobal interp_8tap_horiz_ps_16x%1, 4, 6, 8
+
     add                 r1d, r1d
     add                 r3d, r3d
     mov                 r4d, r4m
     mov                 r5d, r5m
-
+    shl                 r4d, 4
 %ifdef PIC
-    lea                 r6, [tab_ChromaCoeff]
-    vpbroadcastq        m0, [r6 + r4 * 8]
+    lea                 r6, [tab_LumaCoeff]
+    vpbroadcastq        m0, [r6 + r4]
+    vpbroadcastq        m1, [r6 + r4 + 8]
 %else
-    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
+    vpbroadcastq        m0, [tab_LumaCoeff + r4]
+    vpbroadcastq        m1, [tab_LumaCoeff + r4 + 8]
 %endif
     mova                m3, [interp8_hpp_shuf]
     vbroadcasti128      m2, [INTERP_OFFSET_PS]
@@ -4846,97 +2389,61 @@ cglobal interp_4tap_horiz_ps_48x64, 4, 7, 6
     ; register map
     ; m0 , m1 interpolate coeff
 
-    sub                 r0, 2
+    sub                 r0, 6
     test                r5d, r5d
-    mov                 r4d, 64
+    mov                 r4d, %1
     jz                  .loop0
-    sub                 r0, r1
-    add                 r4d, 3
+    lea                 r6, [r1*3]
+    sub                 r0, r6
+    add                 r4d, 7
 
 .loop0:
     vbroadcasti128      m4, [r0]
     vbroadcasti128      m5, [r0 + 8]
     pshufb              m4, m3
-    pshufb              m5, m3
+    pshufb              m7, m5, m3
     pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2], xm4
+    pmaddwd             m7, m1
+    paddd               m4, m7
 
-    vbroadcasti128      m4, [r0 + 16]
-    vbroadcasti128      m5, [r0 + 24]
-    pshufb              m4, m3
+    vbroadcasti128      m6, [r0 + 16]
     pshufb              m5, m3
-    pmaddwd             m4, m0
+    pshufb              m7, m6, m3
     pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 16], xm4
+    pmaddwd             m7, m1
+    paddd               m5, m7
 
-    vbroadcasti128      m4, [r0 + 32]
-    vbroadcasti128      m5, [r0 + 40]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
     phaddd              m4, m5
-    paddd               m4, m2
     vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 32], xm4
-
-    vbroadcasti128      m4, [r0 + 48]
-    vbroadcasti128      m5, [r0 + 56]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
     paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
     vextracti128        xm5, m4, 1
+    psrad               xm4, INTERP_SHIFT_PS
+    psrad               xm5, INTERP_SHIFT_PS
     packssdw            xm4, xm5
-    movu                [r2 + 48], xm4
+    movu                [r2], xm4
 
-    vbroadcasti128      m4, [r0 + 64]
-    vbroadcasti128      m5, [r0 + 72]
-    pshufb              m4, m3
-    pshufb              m5, m3
-    pmaddwd             m4, m0
-    pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 64], xm4
+    vbroadcasti128      m5, [r0 + 24]
+    pshufb              m6, m3
+    pshufb              m7, m5, m3
+    pmaddwd             m6, m0
+    pmaddwd             m7, m1
+    paddd               m6, m7
 
-    vbroadcasti128      m4, [r0 + 80]
-    vbroadcasti128      m5, [r0 + 88]
-    pshufb              m4, m3
+    vbroadcasti128      m7, [r0 + 32]
     pshufb              m5, m3
-    pmaddwd             m4, m0
+    pshufb              m7, m3
     pmaddwd             m5, m0
-    phaddd              m4, m5
-    paddd               m4, m2
-    vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
-    packssdw            xm4, xm5
-    movu                [r2 + 80], xm4
+    pmaddwd             m7, m1
+    paddd               m5, m7
+
+    phaddd              m6, m5
+    vpermq              m6, m6, q3120
+    paddd               m6, m2
+    vextracti128        xm5,m6, 1
+    psrad               xm6, INTERP_SHIFT_PS
+    psrad               xm5, INTERP_SHIFT_PS
+    packssdw            xm6, xm5
+    movu                [r2 + 16], xm6
 
     add                 r2, r3
     add                 r0, r1
@@ -4944,21 +2451,30 @@ cglobal interp_4tap_horiz_ps_48x64, 4, 7, 6
     jnz                 .loop0
     RET
 %endif
+%endmacro
+
+    IPFILTER_LUMA_PS_16xN_AVX2 4
+    IPFILTER_LUMA_PS_16xN_AVX2 8
+    IPFILTER_LUMA_PS_16xN_AVX2 12
+    IPFILTER_LUMA_PS_16xN_AVX2 16
+    IPFILTER_LUMA_PS_16xN_AVX2 32
+    IPFILTER_LUMA_PS_16xN_AVX2 64
 
-%macro IPFILTER_CHROMA_PS_6xN_AVX2 1
 INIT_YMM avx2
 %if ARCH_X86_64 == 1
-cglobal interp_4tap_horiz_ps_6x%1, 4, 7, 6
+cglobal interp_8tap_horiz_ps_12x16, 4, 6, 8
     add                 r1d, r1d
     add                 r3d, r3d
     mov                 r4d, r4m
     mov                 r5d, r5m
-
+    shl                 r4d, 4
 %ifdef PIC
-    lea                 r6, [tab_ChromaCoeff]
-    vpbroadcastq        m0, [r6 + r4 * 8]
+    lea                 r6, [tab_LumaCoeff]
+    vpbroadcastq        m0, [r6 + r4]
+    vpbroadcastq        m1, [r6 + r4 + 8]
 %else
-    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
+    vpbroadcastq        m0, [tab_LumaCoeff + r4]
+    vpbroadcastq        m1, [tab_LumaCoeff + r4 + 8]
 %endif
     mova                m3, [interp8_hpp_shuf]
     vbroadcasti128      m2, [INTERP_OFFSET_PS]
@@ -4966,35 +2482,56 @@ cglobal interp_4tap_horiz_ps_6x%1, 4, 7, 6
     ; register map
     ; m0 , m1 interpolate coeff
 
-    sub                 r0, 2
+    sub                 r0, 6
     test                r5d, r5d
-    mov                 r4d, %1
+    mov                 r4d, 16
     jz                  .loop0
-    sub                 r0, r1
-    add                 r4d, 3
+    lea                 r6, [r1*3]
+    sub                 r0, r6
+    add                 r4d, 7
 
 .loop0:
     vbroadcasti128      m4, [r0]
     vbroadcasti128      m5, [r0 + 8]
     pshufb              m4, m3
-    pshufb              m5, m3
+    pshufb              m7, m5, m3
     pmaddwd             m4, m0
+    pmaddwd             m7, m1
+    paddd               m4, m7
+
+    vbroadcasti128      m6, [r0 + 16]
+    pshufb              m5, m3
+    pshufb              m7, m6, m3
     pmaddwd             m5, m0
+    pmaddwd             m7, m1
+    paddd               m5, m7
+
     phaddd              m4, m5
-    paddd               m4, m2
     vpermq              m4, m4, q3120
-    psrad               m4, INTERP_SHIFT_PS
-    vextracti128        xm5, m4, 1
+    paddd               m4, m2
+    vextracti128        xm5,m4, 1
+    psrad               xm4, INTERP_SHIFT_PS
+    psrad               xm5, INTERP_SHIFT_PS
     packssdw            xm4, xm5
-    movq                [r2], xm4
-    pextrd              [r2 + 8], xm4, 2
+    movu                [r2], xm4
+
+    vbroadcasti128      m5, [r0 + 24]
+    pshufb              m6, m3
+    pshufb              m5, m3
+    pmaddwd             m6, m0
+    pmaddwd             m5, m1
+    paddd               m6, m5
+
+    phaddd              m6, m6
+    vpermq              m6, m6, q3120
+    paddd               xm6, xm2
+    psrad               xm6, INTERP_SHIFT_PS
+    packssdw            xm6, xm6
+    movq                [r2 + 16], xm6
+
     add                 r2, r3
     add                 r0, r1
     dec                 r4d
     jnz                 .loop0
     RET
 %endif
-%endmacro
-
-    IPFILTER_CHROMA_PS_6xN_AVX2 8
-    IPFILTER_CHROMA_PS_6xN_AVX2 16
diff --git a/source/common/x86/h4-ipfilter16.asm b/source/common/x86/h4-ipfilter16.asm
new file mode 100644
index 0000000000..727ea95e79
--- /dev/null
+++ b/source/common/x86/h4-ipfilter16.asm
@@ -0,0 +1,2632 @@
+;*****************************************************************************
+;* Copyright (C) 2013-2017 MulticoreWare, Inc
+;*
+;* Authors: Nabajit Deka <nabajit@multicorewareinc.com>
+;*          Murugan Vairavel <murugan@multicorewareinc.com>
+;*          Min Chen <chenm003@163.com>
+;*
+;* This program is free software; you can redistribute it and/or modify
+;* it under the terms of the GNU General Public License as published by
+;* the Free Software Foundation; either version 2 of the License, or
+;* (at your option) any later version.
+;*
+;* This program is distributed in the hope that it will be useful,
+;* but WITHOUT ANY WARRANTY; without even the implied warranty of
+;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+;* GNU General Public License for more details.
+;*
+;* You should have received a copy of the GNU General Public License
+;* along with this program; if not, write to the Free Software
+;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
+;*
+;* This program is also available under a commercial proprietary license.
+;* For more information, contact us at license @ x265.com.
+;*****************************************************************************/
+
+%include "x86inc.asm"
+%include "x86util.asm"
+
+
+%define INTERP_OFFSET_PP        pd_32
+%define INTERP_SHIFT_PP         6
+
+%if BIT_DEPTH == 10
+    %define INTERP_SHIFT_PS         2
+    %define INTERP_OFFSET_PS        pd_n32768
+    %define INTERP_SHIFT_SP         10
+    %define INTERP_OFFSET_SP        h4_pd_524800
+%elif BIT_DEPTH == 12
+    %define INTERP_SHIFT_PS         4
+    %define INTERP_OFFSET_PS        pd_n131072
+    %define INTERP_SHIFT_SP         8
+    %define INTERP_OFFSET_SP        pd_524416
+%else
+    %error Unsupport bit depth!
+%endif
+
+
+SECTION_RODATA 32
+
+tab_c_32:         times 8 dd 32
+h4_pd_524800:        times 8 dd 524800
+
+tab_Tm16:         db 0, 1, 2, 3, 4,  5,  6, 7, 2, 3, 4,  5, 6, 7, 8, 9
+
+tab_ChromaCoeff:  dw  0, 64,  0,  0
+                  dw -2, 58, 10, -2
+                  dw -4, 54, 16, -2
+                  dw -6, 46, 28, -4
+                  dw -4, 36, 36, -4
+                  dw -4, 28, 46, -6
+                  dw -2, 16, 54, -4
+                  dw -2, 10, 58, -2
+
+const h4_interp8_hpp_shuf,     db 0, 1, 2, 3, 4, 5, 6, 7, 2, 3, 4, 5, 6, 7, 8, 9
+                            db 4, 5, 6, 7, 8, 9, 10, 11, 6, 7, 8, 9, 10, 11, 12, 13
+
+SECTION .text
+cextern pd_8
+cextern pd_32
+cextern pw_pixel_max
+cextern pd_524416
+cextern pd_n32768
+cextern pd_n131072
+cextern pw_2000
+cextern idct8_shuf2
+
+%macro FILTERH_W2_4_sse3 2
+    movh        m3,     [r0 + %1]
+    movhps      m3,     [r0 + %1 + 2]
+    pmaddwd     m3,     m0
+    movh        m4,     [r0 + r1 + %1]
+    movhps      m4,     [r0 + r1 + %1 + 2]
+    pmaddwd     m4,     m0
+    pshufd      m2,     m3,     q2301
+    paddd       m3,     m2
+    pshufd      m2,     m4,     q2301
+    paddd       m4,     m2
+    pshufd      m3,     m3,     q3120
+    pshufd      m4,     m4,     q3120
+    punpcklqdq  m3,     m4
+    paddd       m3,     m1
+    movh        m5,     [r0 + 2 * r1 + %1]
+    movhps      m5,     [r0 + 2 * r1 + %1 + 2]
+    pmaddwd     m5,     m0
+    movh        m4,     [r0 + r4 + %1]
+    movhps      m4,     [r0 + r4 + %1 + 2]
+    pmaddwd     m4,     m0
+    pshufd      m2,     m5,     q2301
+    paddd       m5,     m2
+    pshufd      m2,     m4,     q2301
+    paddd       m4,     m2
+    pshufd      m5,     m5,     q3120
+    pshufd      m4,     m4,     q3120
+    punpcklqdq  m5,     m4
+    paddd       m5,     m1
+%ifidn %2, pp
+    psrad       m3,     6
+    psrad       m5,     6
+    packssdw    m3,     m5
+    CLIPW       m3,     m7,     m6
+%else
+    psrad       m3,     INTERP_SHIFT_PS
+    psrad       m5,     INTERP_SHIFT_PS
+    packssdw    m3,     m5
+%endif
+    movd        [r2 + %1], m3
+    psrldq      m3,     4
+    movd        [r2 + r3 + %1], m3
+    psrldq      m3,     4
+    movd        [r2 + r3 * 2 + %1], m3
+    psrldq      m3,     4
+    movd        [r2 + r5 + %1], m3
+%endmacro
+
+%macro FILTERH_W2_3_sse3 1
+    movh        m3,     [r0 + %1]
+    movhps      m3,     [r0 + %1 + 2]
+    pmaddwd     m3,     m0
+    movh        m4,     [r0 + r1 + %1]
+    movhps      m4,     [r0 + r1 + %1 + 2]
+    pmaddwd     m4,     m0
+    pshufd      m2,     m3,     q2301
+    paddd       m3,     m2
+    pshufd      m2,     m4,     q2301
+    paddd       m4,     m2
+    pshufd      m3,     m3,     q3120
+    pshufd      m4,     m4,     q3120
+    punpcklqdq  m3,     m4
+    paddd       m3,     m1
+
+    movh        m5,     [r0 + 2 * r1 + %1]
+    movhps      m5,     [r0 + 2 * r1 + %1 + 2]
+    pmaddwd     m5,     m0
+
+    pshufd      m2,     m5,     q2301
+    paddd       m5,     m2
+    pshufd      m5,     m5,     q3120
+    paddd       m5,     m1
+
+    psrad       m3,     INTERP_SHIFT_PS
+    psrad       m5,     INTERP_SHIFT_PS
+    packssdw    m3,     m5
+
+    movd        [r2 + %1], m3
+    psrldq      m3,     4
+    movd        [r2 + r3 + %1], m3
+    psrldq      m3,     4
+    movd        [r2 + r3 * 2 + %1], m3
+%endmacro
+
+%macro FILTERH_W4_2_sse3 2
+    movh        m3,     [r0 + %1]
+    movhps      m3,     [r0 + %1 + 2]
+    pmaddwd     m3,     m0
+    movh        m4,     [r0 + %1 + 4]
+    movhps      m4,     [r0 + %1 + 6]
+    pmaddwd     m4,     m0
+    pshufd      m2,     m3,     q2301
+    paddd       m3,     m2
+    pshufd      m2,     m4,     q2301
+    paddd       m4,     m2
+    pshufd      m3,     m3,     q3120
+    pshufd      m4,     m4,     q3120
+    punpcklqdq  m3,     m4
+    paddd       m3,     m1
+
+    movh        m5,     [r0 + r1 + %1]
+    movhps      m5,     [r0 + r1 + %1 + 2]
+    pmaddwd     m5,     m0
+    movh        m4,     [r0 + r1 + %1 + 4]
+    movhps      m4,     [r0 + r1 + %1 + 6]
+    pmaddwd     m4,     m0
+    pshufd      m2,     m5,     q2301
+    paddd       m5,     m2
+    pshufd      m2,     m4,     q2301
+    paddd       m4,     m2
+    pshufd      m5,     m5,     q3120
+    pshufd      m4,     m4,     q3120
+    punpcklqdq  m5,     m4
+    paddd       m5,     m1
+%ifidn %2, pp
+    psrad       m3,     6
+    psrad       m5,     6
+    packssdw    m3,     m5
+    CLIPW       m3,     m7,     m6
+%else
+    psrad       m3,     INTERP_SHIFT_PS
+    psrad       m5,     INTERP_SHIFT_PS
+    packssdw    m3,     m5
+%endif
+    movh        [r2 + %1], m3
+    movhps      [r2 + r3 + %1], m3
+%endmacro
+
+%macro FILTERH_W4_1_sse3 1
+    movh        m3,     [r0 + 2 * r1 + %1]
+    movhps      m3,     [r0 + 2 * r1 + %1 + 2]
+    pmaddwd     m3,     m0
+    movh        m4,     [r0 + 2 * r1 + %1 + 4]
+    movhps      m4,     [r0 + 2 * r1 + %1 + 6]
+    pmaddwd     m4,     m0
+    pshufd      m2,     m3,     q2301
+    paddd       m3,     m2
+    pshufd      m2,     m4,     q2301
+    paddd       m4,     m2
+    pshufd      m3,     m3,     q3120
+    pshufd      m4,     m4,     q3120
+    punpcklqdq  m3,     m4
+    paddd       m3,     m1
+
+    psrad       m3,     INTERP_SHIFT_PS
+    packssdw    m3,     m3
+    movh        [r2 + r3 * 2 + %1], m3
+%endmacro
+
+%macro FILTERH_W8_1_sse3 2
+    movh        m3,     [r0 + %1]
+    movhps      m3,     [r0 + %1 + 2]
+    pmaddwd     m3,     m0
+    movh        m4,     [r0 + %1 + 4]
+    movhps      m4,     [r0 + %1 + 6]
+    pmaddwd     m4,     m0
+    pshufd      m2,     m3,     q2301
+    paddd       m3,     m2
+    pshufd      m2,     m4,     q2301
+    paddd       m4,     m2
+    pshufd      m3,     m3,     q3120
+    pshufd      m4,     m4,     q3120
+    punpcklqdq  m3,     m4
+    paddd       m3,     m1
+
+    movh        m5,     [r0 + %1 + 8]
+    movhps      m5,     [r0 + %1 + 10]
+    pmaddwd     m5,     m0
+    movh        m4,     [r0 + %1 + 12]
+    movhps      m4,     [r0 + %1 + 14]
+    pmaddwd     m4,     m0
+    pshufd      m2,     m5,     q2301
+    paddd       m5,     m2
+    pshufd      m2,     m4,     q2301
+    paddd       m4,     m2
+    pshufd      m5,     m5,     q3120
+    pshufd      m4,     m4,     q3120
+    punpcklqdq  m5,     m4
+    paddd       m5,     m1
+%ifidn %2, pp
+    psrad       m3,     6
+    psrad       m5,     6
+    packssdw    m3,     m5
+    CLIPW       m3,     m7,     m6
+%else
+    psrad       m3,     INTERP_SHIFT_PS
+    psrad       m5,     INTERP_SHIFT_PS
+    packssdw    m3,     m5
+%endif
+    movdqu      [r2 + %1], m3
+%endmacro
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_horiz_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_HOR_CHROMA_sse3 3
+INIT_XMM sse3
+cglobal interp_4tap_horiz_%3_%1x%2, 4, 7, 8
+    add         r3,     r3
+    add         r1,     r1
+    sub         r0,     2
+    mov         r4d,    r4m
+    add         r4d,    r4d
+
+%ifdef PIC
+    lea         r6,     [tab_ChromaCoeff]
+    movddup     m0,     [r6 + r4 * 4]
+%else
+    movddup     m0,     [tab_ChromaCoeff + r4 * 4]
+%endif
+
+%ifidn %3, ps
+    mova        m1,     [INTERP_OFFSET_PS]
+    cmp         r5m,    byte 0
+%if %1 <= 6
+    lea         r4,     [r1 * 3]
+    lea         r5,     [r3 * 3]
+%endif
+    je          .skip
+    sub         r0,     r1
+%if %1 <= 6
+%assign y 1
+%else
+%assign y 3
+%endif
+%assign z 0
+%rep y
+%assign x 0
+%rep %1/8
+    FILTERH_W8_1_sse3 x, %3
+%assign x x+16
+%endrep
+%if %1 == 4 || (%1 == 6 && z == 0) || (%1 == 12 && z == 0)
+    FILTERH_W4_2_sse3 x, %3
+    FILTERH_W4_1_sse3 x
+%assign x x+8
+%endif
+%if %1 == 2 || (%1 == 6 && z == 0)
+    FILTERH_W2_3_sse3 x
+%endif
+%if %1 <= 6
+    lea         r0,     [r0 + r4]
+    lea         r2,     [r2 + r5]
+%else
+    lea         r0,     [r0 + r1]
+    lea         r2,     [r2 + r3]
+%endif
+%assign z z+1
+%endrep
+.skip:
+%elifidn %3, pp
+    pxor        m7,     m7
+    mova        m6,     [pw_pixel_max]
+    mova        m1,     [tab_c_32]
+%if %1 == 2 || %1 == 6
+    lea         r4,     [r1 * 3]
+    lea         r5,     [r3 * 3]
+%endif
+%endif
+
+%if %1 == 2
+%assign y %2/4
+%elif %1 <= 6
+%assign y %2/2
+%else
+%assign y %2
+%endif
+%assign z 0
+%rep y
+%assign x 0
+%rep %1/8
+    FILTERH_W8_1_sse3 x, %3
+%assign x x+16
+%endrep
+%if %1 == 4 || %1 == 6 || (%1 == 12 && (z % 2) == 0)
+    FILTERH_W4_2_sse3 x, %3
+%assign x x+8
+%endif
+%if %1 == 2 || (%1 == 6 && (z % 2) == 0)
+    FILTERH_W2_4_sse3 x, %3
+%endif
+%assign z z+1
+%if z < y
+%if %1 == 2
+    lea         r0,     [r0 + 4 * r1]
+    lea         r2,     [r2 + 4 * r3]
+%elif %1 <= 6
+    lea         r0,     [r0 + 2 * r1]
+    lea         r2,     [r2 + 2 * r3]
+%else
+    lea         r0,     [r0 + r1]
+    lea         r2,     [r2 + r3]
+%endif
+%endif ;z < y
+%endrep
+
+    RET
+%endmacro
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_horiz_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+
+FILTER_HOR_CHROMA_sse3 2, 4, pp
+FILTER_HOR_CHROMA_sse3 2, 8, pp
+FILTER_HOR_CHROMA_sse3 2, 16, pp
+FILTER_HOR_CHROMA_sse3 4, 2, pp
+FILTER_HOR_CHROMA_sse3 4, 4, pp
+FILTER_HOR_CHROMA_sse3 4, 8, pp
+FILTER_HOR_CHROMA_sse3 4, 16, pp
+FILTER_HOR_CHROMA_sse3 4, 32, pp
+FILTER_HOR_CHROMA_sse3 6, 8, pp
+FILTER_HOR_CHROMA_sse3 6, 16, pp
+FILTER_HOR_CHROMA_sse3 8, 2, pp
+FILTER_HOR_CHROMA_sse3 8, 4, pp
+FILTER_HOR_CHROMA_sse3 8, 6, pp
+FILTER_HOR_CHROMA_sse3 8, 8, pp
+FILTER_HOR_CHROMA_sse3 8, 12, pp
+FILTER_HOR_CHROMA_sse3 8, 16, pp
+FILTER_HOR_CHROMA_sse3 8, 32, pp
+FILTER_HOR_CHROMA_sse3 8, 64, pp
+FILTER_HOR_CHROMA_sse3 12, 16, pp
+FILTER_HOR_CHROMA_sse3 12, 32, pp
+FILTER_HOR_CHROMA_sse3 16, 4, pp
+FILTER_HOR_CHROMA_sse3 16, 8, pp
+FILTER_HOR_CHROMA_sse3 16, 12, pp
+FILTER_HOR_CHROMA_sse3 16, 16, pp
+FILTER_HOR_CHROMA_sse3 16, 24, pp
+FILTER_HOR_CHROMA_sse3 16, 32, pp
+FILTER_HOR_CHROMA_sse3 16, 64, pp
+FILTER_HOR_CHROMA_sse3 24, 32, pp
+FILTER_HOR_CHROMA_sse3 24, 64, pp
+FILTER_HOR_CHROMA_sse3 32, 8, pp
+FILTER_HOR_CHROMA_sse3 32, 16, pp
+FILTER_HOR_CHROMA_sse3 32, 24, pp
+FILTER_HOR_CHROMA_sse3 32, 32, pp
+FILTER_HOR_CHROMA_sse3 32, 48, pp
+FILTER_HOR_CHROMA_sse3 32, 64, pp
+FILTER_HOR_CHROMA_sse3 48, 64, pp
+FILTER_HOR_CHROMA_sse3 64, 16, pp
+FILTER_HOR_CHROMA_sse3 64, 32, pp
+FILTER_HOR_CHROMA_sse3 64, 48, pp
+FILTER_HOR_CHROMA_sse3 64, 64, pp
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_horiz_ps_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+
+FILTER_HOR_CHROMA_sse3 2, 4, ps
+FILTER_HOR_CHROMA_sse3 2, 8, ps
+FILTER_HOR_CHROMA_sse3 2, 16, ps
+FILTER_HOR_CHROMA_sse3 4, 2, ps
+FILTER_HOR_CHROMA_sse3 4, 4, ps
+FILTER_HOR_CHROMA_sse3 4, 8, ps
+FILTER_HOR_CHROMA_sse3 4, 16, ps
+FILTER_HOR_CHROMA_sse3 4, 32, ps
+FILTER_HOR_CHROMA_sse3 6, 8, ps
+FILTER_HOR_CHROMA_sse3 6, 16, ps
+FILTER_HOR_CHROMA_sse3 8, 2, ps
+FILTER_HOR_CHROMA_sse3 8, 4, ps
+FILTER_HOR_CHROMA_sse3 8, 6, ps
+FILTER_HOR_CHROMA_sse3 8, 8, ps
+FILTER_HOR_CHROMA_sse3 8, 12, ps
+FILTER_HOR_CHROMA_sse3 8, 16, ps
+FILTER_HOR_CHROMA_sse3 8, 32, ps
+FILTER_HOR_CHROMA_sse3 8, 64, ps
+FILTER_HOR_CHROMA_sse3 12, 16, ps
+FILTER_HOR_CHROMA_sse3 12, 32, ps
+FILTER_HOR_CHROMA_sse3 16, 4, ps
+FILTER_HOR_CHROMA_sse3 16, 8, ps
+FILTER_HOR_CHROMA_sse3 16, 12, ps
+FILTER_HOR_CHROMA_sse3 16, 16, ps
+FILTER_HOR_CHROMA_sse3 16, 24, ps
+FILTER_HOR_CHROMA_sse3 16, 32, ps
+FILTER_HOR_CHROMA_sse3 16, 64, ps
+FILTER_HOR_CHROMA_sse3 24, 32, ps
+FILTER_HOR_CHROMA_sse3 24, 64, ps
+FILTER_HOR_CHROMA_sse3 32, 8, ps
+FILTER_HOR_CHROMA_sse3 32, 16, ps
+FILTER_HOR_CHROMA_sse3 32, 24, ps
+FILTER_HOR_CHROMA_sse3 32, 32, ps
+FILTER_HOR_CHROMA_sse3 32, 48, ps
+FILTER_HOR_CHROMA_sse3 32, 64, ps
+FILTER_HOR_CHROMA_sse3 48, 64, ps
+FILTER_HOR_CHROMA_sse3 64, 16, ps
+FILTER_HOR_CHROMA_sse3 64, 32, ps
+FILTER_HOR_CHROMA_sse3 64, 48, ps
+FILTER_HOR_CHROMA_sse3 64, 64, ps
+
+%macro FILTER_W2_2 1
+    movu        m3,         [r0]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + r1]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    packusdw    m3,         m3
+    CLIPW       m3,         m7,    m6
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    packssdw    m3,         m3
+%endif
+    movd        [r2],       m3
+    pextrd      [r2 + r3],  m3, 1
+%endmacro
+
+%macro FILTER_W4_2 1
+    movu        m3,         [r0]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 4]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + r1]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + r1 + 4]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m7,    m6
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2],       m3
+    movhps      [r2 + r3],  m3
+%endmacro
+
+;-----------------------------------------------------------------------------
+; void interp_4tap_horiz_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_CHROMA_H 6
+INIT_XMM sse4
+cglobal interp_4tap_horiz_%3_%1x%2, 4, %4, %5
+
+    add         r3,       r3
+    add         r1,       r1
+    sub         r0,       2
+    mov         r4d,      r4m
+    add         r4d,      r4d
+
+%ifdef PIC
+    lea         r%6,      [tab_ChromaCoeff]
+    movh        m0,       [r%6 + r4 * 4]
+%else
+    movh        m0,       [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    punpcklqdq  m0,       m0
+    mova        m2,       [tab_Tm16]
+
+%ifidn %3, ps
+    mova        m1,       [INTERP_OFFSET_PS]
+    cmp         r5m, byte 0
+    je          .skip
+    sub         r0,       r1
+    movu        m3,       [r0]
+    pshufb      m3,       m3, m2
+    pmaddwd     m3,       m0
+
+  %if %1 == 4
+    movu        m4,       [r0 + 4]
+    pshufb      m4,       m4, m2
+    pmaddwd     m4,       m0
+    phaddd      m3,       m4
+  %else
+    phaddd      m3,       m3
+  %endif
+
+    paddd       m3,       m1
+    psrad       m3,       INTERP_SHIFT_PS
+    packssdw    m3,       m3
+
+  %if %1 == 2
+    movd        [r2],     m3
+  %else
+    movh        [r2],     m3
+  %endif
+
+    add         r0,       r1
+    add         r2,       r3
+    FILTER_W%1_2 %3
+    lea         r0,       [r0 + 2 * r1]
+    lea         r2,       [r2 + 2 * r3]
+
+.skip:
+
+%else     ;%ifidn %3, ps
+    pxor        m7,       m7
+    mova        m6,       [pw_pixel_max]
+    mova        m1,       [tab_c_32]
+%endif    ;%ifidn %3, ps
+
+    FILTER_W%1_2 %3
+
+%rep (%2/2) - 1
+    lea         r0,       [r0 + 2 * r1]
+    lea         r2,       [r2 + 2 * r3]
+    FILTER_W%1_2 %3
+%endrep
+    RET
+%endmacro
+
+FILTER_CHROMA_H 2, 4, pp, 6, 8, 5
+FILTER_CHROMA_H 2, 8, pp, 6, 8, 5
+FILTER_CHROMA_H 4, 2, pp, 6, 8, 5
+FILTER_CHROMA_H 4, 4, pp, 6, 8, 5
+FILTER_CHROMA_H 4, 8, pp, 6, 8, 5
+FILTER_CHROMA_H 4, 16, pp, 6, 8, 5
+
+FILTER_CHROMA_H 2, 4, ps, 7, 5, 6
+FILTER_CHROMA_H 2, 8, ps, 7, 5, 6
+FILTER_CHROMA_H 4, 2, ps, 7, 6, 6
+FILTER_CHROMA_H 4, 4, ps, 7, 6, 6
+FILTER_CHROMA_H 4, 8, ps, 7, 6, 6
+FILTER_CHROMA_H 4, 16, ps, 7, 6, 6
+
+FILTER_CHROMA_H 2, 16, pp, 6, 8, 5
+FILTER_CHROMA_H 4, 32, pp, 6, 8, 5
+FILTER_CHROMA_H 2, 16, ps, 7, 5, 6
+FILTER_CHROMA_H 4, 32, ps, 7, 6, 6
+
+
+%macro FILTER_W6_1 1
+    movu        m3,         [r0]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 4]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m4,         [r0 + 8]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m4,         m4
+    paddd       m4,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m4,         INTERP_SHIFT_PP
+    packusdw    m3,         m4
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m4,         INTERP_SHIFT_PS
+    packssdw    m3,         m4
+%endif
+    movh        [r2],       m3
+    pextrd      [r2 + 8],   m3, 2
+%endmacro
+
+cglobal chroma_filter_pp_6x1_internal
+    FILTER_W6_1 pp
+    ret
+
+cglobal chroma_filter_ps_6x1_internal
+    FILTER_W6_1 ps
+    ret
+
+%macro FILTER_W8_1 1
+    movu        m3,         [r0]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 4]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + 8]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + 12]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2],       m3
+    movhps      [r2 + 8],   m3
+%endmacro
+
+cglobal chroma_filter_pp_8x1_internal
+    FILTER_W8_1 pp
+    ret
+
+cglobal chroma_filter_ps_8x1_internal
+    FILTER_W8_1 ps
+    ret
+
+%macro FILTER_W12_1 1
+    movu        m3,         [r0]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 4]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + 8]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + 12]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2],       m3
+    movhps      [r2 + 8],   m3
+
+    movu        m3,         [r0 + 16]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 20]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    packusdw    m3,         m3
+    CLIPW       m3,         m6, m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    packssdw    m3,         m3
+%endif
+    movh        [r2 + 16],  m3
+%endmacro
+
+cglobal chroma_filter_pp_12x1_internal
+    FILTER_W12_1 pp
+    ret
+
+cglobal chroma_filter_ps_12x1_internal
+    FILTER_W12_1 ps
+    ret
+
+%macro FILTER_W16_1 1
+    movu        m3,         [r0]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 4]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + 8]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + 12]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2],       m3
+    movhps      [r2 + 8],   m3
+
+    movu        m3,         [r0 + 16]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 20]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + 24]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + 28]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2 + 16],  m3
+    movhps      [r2 + 24],  m3
+%endmacro
+
+cglobal chroma_filter_pp_16x1_internal
+    FILTER_W16_1 pp
+    ret
+
+cglobal chroma_filter_ps_16x1_internal
+    FILTER_W16_1 ps
+    ret
+
+%macro FILTER_W24_1 1
+    movu        m3,         [r0]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 4]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + 8]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + 12]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2],       m3
+    movhps      [r2 + 8],   m3
+
+    movu        m3,         [r0 + 16]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 20]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + 24]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + 28]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2 + 16],  m3
+    movhps      [r2 + 24],  m3
+
+    movu        m3,         [r0 + 32]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 36]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + 40]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + 44]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2 + 32],  m3
+    movhps      [r2 + 40],  m3
+%endmacro
+
+cglobal chroma_filter_pp_24x1_internal
+    FILTER_W24_1 pp
+    ret
+
+cglobal chroma_filter_ps_24x1_internal
+    FILTER_W24_1 ps
+    ret
+
+%macro FILTER_W32_1 1
+    movu        m3,         [r0]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 4]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + 8]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + 12]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2],       m3
+    movhps      [r2 + 8],   m3
+
+    movu        m3,         [r0 + 16]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 20]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + 24]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + 28]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2 + 16],  m3
+    movhps      [r2 + 24],  m3
+
+    movu        m3,         [r0 + 32]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 36]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + 40]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + 44]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2 + 32],  m3
+    movhps      [r2 + 40],  m3
+
+    movu        m3,         [r0 + 48]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + 52]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + 56]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + 60]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2 + 48],  m3
+    movhps      [r2 + 56],  m3
+%endmacro
+
+cglobal chroma_filter_pp_32x1_internal
+    FILTER_W32_1 pp
+    ret
+
+cglobal chroma_filter_ps_32x1_internal
+    FILTER_W32_1 ps
+    ret
+
+%macro FILTER_W8o_1 2
+    movu        m3,         [r0 + %2]
+    pshufb      m3,         m3, m2
+    pmaddwd     m3,         m0
+    movu        m4,         [r0 + %2 + 4]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m3,         m4
+    paddd       m3,         m1
+
+    movu        m5,         [r0 + %2 + 8]
+    pshufb      m5,         m5, m2
+    pmaddwd     m5,         m0
+    movu        m4,         [r0 + %2 + 12]
+    pshufb      m4,         m4, m2
+    pmaddwd     m4,         m0
+    phaddd      m5,         m4
+    paddd       m5,         m1
+%ifidn %1, pp
+    psrad       m3,         INTERP_SHIFT_PP
+    psrad       m5,         INTERP_SHIFT_PP
+    packusdw    m3,         m5
+    CLIPW       m3,         m6,    m7
+%else
+    psrad       m3,         INTERP_SHIFT_PS
+    psrad       m5,         INTERP_SHIFT_PS
+    packssdw    m3,         m5
+%endif
+    movh        [r2 + %2],       m3
+    movhps      [r2 + %2 + 8],   m3
+%endmacro
+
+%macro FILTER_W48_1 1
+    FILTER_W8o_1 %1, 0
+    FILTER_W8o_1 %1, 16
+    FILTER_W8o_1 %1, 32
+    FILTER_W8o_1 %1, 48
+    FILTER_W8o_1 %1, 64
+    FILTER_W8o_1 %1, 80
+%endmacro
+
+cglobal chroma_filter_pp_48x1_internal
+    FILTER_W48_1 pp
+    ret
+
+cglobal chroma_filter_ps_48x1_internal
+    FILTER_W48_1 ps
+    ret
+
+%macro FILTER_W64_1 1
+    FILTER_W8o_1 %1, 0
+    FILTER_W8o_1 %1, 16
+    FILTER_W8o_1 %1, 32
+    FILTER_W8o_1 %1, 48
+    FILTER_W8o_1 %1, 64
+    FILTER_W8o_1 %1, 80
+    FILTER_W8o_1 %1, 96
+    FILTER_W8o_1 %1, 112
+%endmacro
+
+cglobal chroma_filter_pp_64x1_internal
+    FILTER_W64_1 pp
+    ret
+
+cglobal chroma_filter_ps_64x1_internal
+    FILTER_W64_1 ps
+    ret
+;-----------------------------------------------------------------------------
+; void interp_4tap_horiz_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+
+INIT_XMM sse4
+%macro IPFILTER_CHROMA 6
+cglobal interp_4tap_horiz_%3_%1x%2, 4, %5, %6
+
+    add         r3,        r3
+    add         r1,        r1
+    sub         r0,         2
+    mov         r4d,        r4m
+    add         r4d,        r4d
+
+%ifdef PIC
+    lea         r%4,       [tab_ChromaCoeff]
+    movh        m0,       [r%4 + r4 * 4]
+%else
+    movh        m0,       [tab_ChromaCoeff + r4 * 4]
+%endif
+
+    punpcklqdq  m0,       m0
+    mova        m2,       [tab_Tm16]
+
+%ifidn %3, ps
+    mova        m1,       [INTERP_OFFSET_PS]
+    cmp         r5m, byte 0
+    je          .skip
+    sub         r0, r1
+    call chroma_filter_%3_%1x1_internal
+    add         r0, r1
+    add         r2, r3
+    call chroma_filter_%3_%1x1_internal
+    add         r0, r1
+    add         r2, r3
+    call chroma_filter_%3_%1x1_internal
+    add         r0, r1
+    add         r2, r3
+.skip:
+%else
+    mova        m1,         [tab_c_32]
+    pxor        m6,         m6
+    mova        m7,         [pw_pixel_max]
+%endif
+
+    call chroma_filter_%3_%1x1_internal
+%rep %2 - 1
+    add         r0,       r1
+    add         r2,       r3
+    call chroma_filter_%3_%1x1_internal
+%endrep
+RET
+%endmacro
+IPFILTER_CHROMA 6, 8, pp, 5, 6, 8
+IPFILTER_CHROMA 8, 2, pp, 5, 6, 8
+IPFILTER_CHROMA 8, 4, pp, 5, 6, 8
+IPFILTER_CHROMA 8, 6, pp, 5, 6, 8
+IPFILTER_CHROMA 8, 8, pp, 5, 6, 8
+IPFILTER_CHROMA 8, 16, pp, 5, 6, 8
+IPFILTER_CHROMA 8, 32, pp, 5, 6, 8
+IPFILTER_CHROMA 12, 16, pp, 5, 6, 8
+IPFILTER_CHROMA 16, 4, pp, 5, 6, 8
+IPFILTER_CHROMA 16, 8, pp, 5, 6, 8
+IPFILTER_CHROMA 16, 12, pp, 5, 6, 8
+IPFILTER_CHROMA 16, 16, pp, 5, 6, 8
+IPFILTER_CHROMA 16, 32, pp, 5, 6, 8
+IPFILTER_CHROMA 24, 32, pp, 5, 6, 8
+IPFILTER_CHROMA 32, 8, pp, 5, 6, 8
+IPFILTER_CHROMA 32, 16, pp, 5, 6, 8
+IPFILTER_CHROMA 32, 24, pp, 5, 6, 8
+IPFILTER_CHROMA 32, 32, pp, 5, 6, 8
+
+IPFILTER_CHROMA 6, 8, ps, 6, 7, 6
+IPFILTER_CHROMA 8, 2, ps, 6, 7, 6
+IPFILTER_CHROMA 8, 4, ps, 6, 7, 6
+IPFILTER_CHROMA 8, 6, ps, 6, 7, 6
+IPFILTER_CHROMA 8, 8, ps, 6, 7, 6
+IPFILTER_CHROMA 8, 16, ps, 6, 7, 6
+IPFILTER_CHROMA 8, 32, ps, 6, 7, 6
+IPFILTER_CHROMA 12, 16, ps, 6, 7, 6
+IPFILTER_CHROMA 16, 4, ps, 6, 7, 6
+IPFILTER_CHROMA 16, 8, ps, 6, 7, 6
+IPFILTER_CHROMA 16, 12, ps, 6, 7, 6
+IPFILTER_CHROMA 16, 16, ps, 6, 7, 6
+IPFILTER_CHROMA 16, 32, ps, 6, 7, 6
+IPFILTER_CHROMA 24, 32, ps, 6, 7, 6
+IPFILTER_CHROMA 32, 8, ps, 6, 7, 6
+IPFILTER_CHROMA 32, 16, ps, 6, 7, 6
+IPFILTER_CHROMA 32, 24, ps, 6, 7, 6
+IPFILTER_CHROMA 32, 32, ps, 6, 7, 6
+
+IPFILTER_CHROMA 6, 16, pp, 5, 6, 8
+IPFILTER_CHROMA 8, 12, pp, 5, 6, 8
+IPFILTER_CHROMA 8, 64, pp, 5, 6, 8
+IPFILTER_CHROMA 12, 32, pp, 5, 6, 8
+IPFILTER_CHROMA 16, 24, pp, 5, 6, 8
+IPFILTER_CHROMA 16, 64, pp, 5, 6, 8
+IPFILTER_CHROMA 24, 64, pp, 5, 6, 8
+IPFILTER_CHROMA 32, 48, pp, 5, 6, 8
+IPFILTER_CHROMA 32, 64, pp, 5, 6, 8
+IPFILTER_CHROMA 6, 16, ps, 6, 7, 6
+IPFILTER_CHROMA 8, 12, ps, 6, 7, 6
+IPFILTER_CHROMA 8, 64, ps, 6, 7, 6
+IPFILTER_CHROMA 12, 32, ps, 6, 7, 6
+IPFILTER_CHROMA 16, 24, ps, 6, 7, 6
+IPFILTER_CHROMA 16, 64, ps, 6, 7, 6
+IPFILTER_CHROMA 24, 64, ps, 6, 7, 6
+IPFILTER_CHROMA 32, 48, ps, 6, 7, 6
+IPFILTER_CHROMA 32, 64, ps, 6, 7, 6
+
+IPFILTER_CHROMA 48, 64, pp, 5, 6, 8
+IPFILTER_CHROMA 64, 48, pp, 5, 6, 8
+IPFILTER_CHROMA 64, 64, pp, 5, 6, 8
+IPFILTER_CHROMA 64, 32, pp, 5, 6, 8
+IPFILTER_CHROMA 64, 16, pp, 5, 6, 8
+IPFILTER_CHROMA 48, 64, ps, 6, 7, 6
+IPFILTER_CHROMA 64, 48, ps, 6, 7, 6
+IPFILTER_CHROMA 64, 64, ps, 6, 7, 6
+IPFILTER_CHROMA 64, 32, ps, 6, 7, 6
+IPFILTER_CHROMA 64, 16, ps, 6, 7, 6
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+%macro IPFILTER_CHROMA_avx2_6xN 1
+cglobal interp_4tap_horiz_pp_6x%1, 5,6,8
+    add             r1d, r1d
+    add             r3d, r3d
+    sub             r0, 2
+    mov             r4d, r4m
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova            m1, [h4_interp8_hpp_shuf]
+    vpbroadcastd    m2, [pd_32]
+    pxor            m5, m5
+    mova            m6, [idct8_shuf2]
+    mova            m7, [pw_pixel_max]
+
+    mov             r4d, %1/2
+.loop:
+    vbroadcasti128  m3, [r0]
+    vbroadcasti128  m4, [r0 + 8]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, INTERP_SHIFT_PP           ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3, q2020
+    pshufb          xm3, xm6                      ; m3 = WORD[7 6 5 4 3 2 1 0]
+    CLIPW           xm3, xm5, xm7
+    movq            [r2], xm3
+    pextrd          [r2 + 8], xm3, 2
+
+    vbroadcasti128  m3, [r0 + r1]
+    vbroadcasti128  m4, [r0 + r1 + 8]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, INTERP_SHIFT_PP           ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3, q2020
+    pshufb          xm3, xm6                      ; m3 = WORD[7 6 5 4 3 2 1 0]
+    CLIPW           xm3, xm5, xm7
+    movq            [r2 + r3], xm3
+    pextrd          [r2 + r3 + 8], xm3, 2
+
+    lea             r0, [r0 + r1 * 2]
+    lea             r2, [r2 + r3 * 2]
+    dec             r4d
+    jnz             .loop
+    RET
+%endmacro
+IPFILTER_CHROMA_avx2_6xN 8
+IPFILTER_CHROMA_avx2_6xN 16
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_8x2, 5,6,8
+    add             r1d, r1d
+    add             r3d, r3d
+    sub             r0, 2
+    mov             r4d, r4m
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova            m1, [h4_interp8_hpp_shuf]
+    vpbroadcastd    m2, [pd_32]
+    pxor            m5, m5
+    mova            m6, [idct8_shuf2]
+    mova            m7, [pw_pixel_max]
+
+    vbroadcasti128  m3, [r0]
+    vbroadcasti128  m4, [r0 + 8]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, INTERP_SHIFT_PP          ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3,q2020
+    pshufb          xm3, xm6                     ; m3 = WORD[7 6 5 4 3 2 1 0]
+    CLIPW           xm3, xm5, xm7
+    movu            [r2], xm3
+
+    vbroadcasti128  m3, [r0 + r1]
+    vbroadcasti128  m4, [r0 + r1 + 8]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, INTERP_SHIFT_PP           ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3,q2020
+    pshufb          xm3, xm6                      ; m3 = WORD[7 6 5 4 3 2 1 0]
+    CLIPW           xm3, xm5, xm7
+    movu            [r2 + r3], xm3
+    RET
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_8x4, 5,6,8
+    add             r1d, r1d
+    add             r3d, r3d
+    sub             r0, 2
+    mov             r4d, r4m
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova            m1, [h4_interp8_hpp_shuf]
+    vpbroadcastd    m2, [pd_32]
+    pxor            m5, m5
+    mova            m6, [idct8_shuf2]
+    mova            m7, [pw_pixel_max]
+
+%rep 2
+    vbroadcasti128  m3, [r0]
+    vbroadcasti128  m4, [r0 + 8]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3,q2020
+    pshufb          xm3, xm6                    ; m3 = WORD[7 6 5 4 3 2 1 0]
+    CLIPW           xm3, xm5, xm7
+    movu            [r2], xm3
+
+    vbroadcasti128  m3, [r0 + r1]
+    vbroadcasti128  m4, [r0 + r1 + 8]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3,q2020
+    pshufb          xm3, xm6                    ; m3 = WORD[7 6 5 4 3 2 1 0]
+    CLIPW           xm3, xm5, xm7
+    movu            [r2 + r3], xm3
+
+    lea             r0, [r0 + r1 * 2]
+    lea             r2, [r2 + r3 * 2]
+%endrep
+    RET
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+%macro IPFILTER_CHROMA_avx2_8xN 1
+cglobal interp_4tap_horiz_pp_8x%1, 5,6,8
+    add             r1d, r1d
+    add             r3d, r3d
+    sub             r0, 2
+    mov             r4d, r4m
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova            m1, [h4_interp8_hpp_shuf]
+    vpbroadcastd    m2, [pd_32]
+    pxor            m5, m5
+    mova            m6, [idct8_shuf2]
+    mova            m7, [pw_pixel_max]
+
+    mov             r4d, %1/2
+.loop:
+    vbroadcasti128  m3, [r0]
+    vbroadcasti128  m4, [r0 + 8]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6                         ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3, q2020
+    pshufb          xm3, xm6                      ; m3 = WORD[7 6 5 4 3 2 1 0]
+    CLIPW           xm3, xm5, xm7
+    movu            [r2], xm3
+
+    vbroadcasti128  m3, [r0 + r1]
+    vbroadcasti128  m4, [r0 + r1 + 8]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6                         ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3, q2020
+    pshufb          xm3, xm6                      ; m3 = WORD[7 6 5 4 3 2 1 0]
+    CLIPW           xm3, xm5, xm7
+    movu            [r2 + r3], xm3
+
+    lea             r0, [r0 + r1 * 2]
+    lea             r2, [r2 + r3 * 2]
+    dec             r4d
+    jnz             .loop
+    RET
+%endmacro
+IPFILTER_CHROMA_avx2_8xN 6
+IPFILTER_CHROMA_avx2_8xN 8
+IPFILTER_CHROMA_avx2_8xN 12
+IPFILTER_CHROMA_avx2_8xN 16
+IPFILTER_CHROMA_avx2_8xN 32
+IPFILTER_CHROMA_avx2_8xN 64
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+%macro IPFILTER_CHROMA_avx2_16xN 1
+%if ARCH_X86_64
+cglobal interp_4tap_horiz_pp_16x%1, 5,6,9
+    add             r1d, r1d
+    add             r3d, r3d
+    sub             r0, 2
+    mov             r4d, r4m
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova            m1, [h4_interp8_hpp_shuf]
+    vpbroadcastd    m2, [pd_32]
+    pxor            m5, m5
+    mova            m6, [idct8_shuf2]
+    mova            m7, [pw_pixel_max]
+
+    mov             r4d, %1
+.loop:
+    vbroadcasti128  m3, [r0]
+    vbroadcasti128  m4, [r0 + 8]
+
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3, q2020
+    pshufb          xm3, xm6                     ; m3 = WORD[7 6 5 4 3 2 1 0]
+
+    vbroadcasti128  m4, [r0 + 16]
+    vbroadcasti128  m8, [r0 + 24]
+
+    pshufb          m4, m1
+    pshufb          m8, m1
+
+    pmaddwd         m4, m0
+    pmaddwd         m8, m0
+    phaddd          m4, m8
+    paddd           m4, m2
+    psrad           m4, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m4, m4
+    vpermq          m4, m4, q2020
+    pshufb          xm4, xm6                    ; m3 = WORD[7 6 5 4 3 2 1 0]
+    vinserti128     m3, m3, xm4, 1
+    CLIPW           m3, m5, m7
+    movu            [r2], m3
+
+    add             r0, r1
+    add             r2, r3
+    dec             r4d
+    jnz             .loop
+    RET
+%endif
+%endmacro
+IPFILTER_CHROMA_avx2_16xN 4
+IPFILTER_CHROMA_avx2_16xN 8
+IPFILTER_CHROMA_avx2_16xN 12
+IPFILTER_CHROMA_avx2_16xN 16
+IPFILTER_CHROMA_avx2_16xN 24
+IPFILTER_CHROMA_avx2_16xN 32
+IPFILTER_CHROMA_avx2_16xN 64
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+%macro IPFILTER_CHROMA_avx2_32xN 1
+%if ARCH_X86_64
+cglobal interp_4tap_horiz_pp_32x%1, 5,6,9
+    add             r1d, r1d
+    add             r3d, r3d
+    sub             r0, 2
+    mov             r4d, r4m
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova            m1, [h4_interp8_hpp_shuf]
+    vpbroadcastd    m2, [pd_32]
+    pxor            m5, m5
+    mova            m6, [idct8_shuf2]
+    mova            m7, [pw_pixel_max]
+
+    mov             r6d, %1
+.loop:
+%assign x 0
+%rep 2
+    vbroadcasti128  m3, [r0 + x]
+    vbroadcasti128  m4, [r0 + 8 + x]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3, q2020
+    pshufb          xm3, xm6                     ; m3 = WORD[7 6 5 4 3 2 1 0]
+
+    vbroadcasti128  m4, [r0 + 16 + x]
+    vbroadcasti128  m8, [r0 + 24 + x]
+    pshufb          m4, m1
+    pshufb          m8, m1
+
+    pmaddwd         m4, m0
+    pmaddwd         m8, m0
+    phaddd          m4, m8
+    paddd           m4, m2
+    psrad           m4, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m4, m4
+    vpermq          m4, m4, q2020
+    pshufb          xm4, xm6                    ; m3 = WORD[7 6 5 4 3 2 1 0]
+    vinserti128     m3, m3, xm4, 1
+    CLIPW           m3, m5, m7
+    movu            [r2 + x], m3
+    %assign x x+32
+    %endrep
+
+    add             r0, r1
+    add             r2, r3
+    dec             r6d
+    jnz             .loop
+    RET
+%endif
+%endmacro
+IPFILTER_CHROMA_avx2_32xN 8
+IPFILTER_CHROMA_avx2_32xN 16
+IPFILTER_CHROMA_avx2_32xN 24
+IPFILTER_CHROMA_avx2_32xN 32
+IPFILTER_CHROMA_avx2_32xN 48
+IPFILTER_CHROMA_avx2_32xN 64
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+%macro IPFILTER_CHROMA_avx2_12xN 1
+%if ARCH_X86_64
+cglobal interp_4tap_horiz_pp_12x%1, 5,6,8
+    add             r1d, r1d
+    add             r3d, r3d
+    sub             r0, 2
+    mov             r4d, r4m
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova            m1, [h4_interp8_hpp_shuf]
+    vpbroadcastd    m2, [pd_32]
+    pxor            m5, m5
+    mova            m6, [idct8_shuf2]
+    mova            m7, [pw_pixel_max]
+
+    mov             r4d, %1
+.loop:
+    vbroadcasti128  m3, [r0]
+    vbroadcasti128  m4, [r0 + 8]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3, q2020
+    pshufb          xm3, xm6                     ; m3 = WORD[7 6 5 4 3 2 1 0]
+    CLIPW           xm3, xm5, xm7
+    movu            [r2], xm3
+
+    vbroadcasti128  m3, [r0 + 16]
+    vbroadcasti128  m4, [r0 + 24]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3, q2020
+    pshufb          xm3, xm6                    ; m3 = WORD[7 6 5 4 3 2 1 0]
+    CLIPW           xm3, xm5, xm7
+    movq            [r2 + 16], xm3
+
+    add             r0, r1
+    add             r2, r3
+    dec             r4d
+    jnz             .loop
+    RET
+%endif
+%endmacro
+IPFILTER_CHROMA_avx2_12xN 16
+IPFILTER_CHROMA_avx2_12xN 32
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+%macro IPFILTER_CHROMA_avx2_24xN 1
+%if ARCH_X86_64
+cglobal interp_4tap_horiz_pp_24x%1, 5,6,9
+    add             r1d, r1d
+    add             r3d, r3d
+    sub             r0, 2
+    mov             r4d, r4m
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova            m1, [h4_interp8_hpp_shuf]
+    vpbroadcastd    m2, [pd_32]
+    pxor            m5, m5
+    mova            m6, [idct8_shuf2]
+    mova            m7, [pw_pixel_max]
+
+    mov             r4d, %1
+.loop:
+    vbroadcasti128  m3, [r0]
+    vbroadcasti128  m4, [r0 + 8]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6
+
+    vbroadcasti128  m4, [r0 + 16]
+    vbroadcasti128  m8, [r0 + 24]
+    pshufb          m4, m1
+    pshufb          m8, m1
+
+    pmaddwd         m4, m0
+    pmaddwd         m8, m0
+    phaddd          m4, m8
+    paddd           m4, m2
+    psrad           m4, 6
+
+    packusdw        m3, m4
+    vpermq          m3, m3, q3120
+    pshufb          m3, m6
+    CLIPW           m3, m5, m7
+    movu            [r2], m3
+
+    vbroadcasti128  m3, [r0 + 32]
+    vbroadcasti128  m4, [r0 + 40]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6
+
+    packusdw        m3, m3
+    vpermq          m3, m3, q2020
+    pshufb          xm3, xm6
+    CLIPW           xm3, xm5, xm7
+    movu            [r2 + 32], xm3
+
+    add             r0, r1
+    add             r2, r3
+    dec             r4d
+    jnz             .loop
+    RET
+%endif
+%endmacro
+IPFILTER_CHROMA_avx2_24xN 32
+IPFILTER_CHROMA_avx2_24xN 64
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+%macro IPFILTER_CHROMA_avx2_64xN 1
+%if ARCH_X86_64
+cglobal interp_4tap_horiz_pp_64x%1, 5,6,9
+    add             r1d, r1d
+    add             r3d, r3d
+    sub             r0, 2
+    mov             r4d, r4m
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova            m1, [h4_interp8_hpp_shuf]
+    vpbroadcastd    m2, [pd_32]
+    pxor            m5, m5
+    mova            m6, [idct8_shuf2]
+    mova            m7, [pw_pixel_max]
+
+    mov             r6d, %1
+.loop:
+%assign x 0
+%rep 4
+    vbroadcasti128  m3, [r0 + x]
+    vbroadcasti128  m4, [r0 + 8 + x]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6
+
+    vbroadcasti128  m4, [r0 + 16 + x]
+    vbroadcasti128  m8, [r0 + 24 + x]
+    pshufb          m4, m1
+    pshufb          m8, m1
+
+    pmaddwd         m4, m0
+    pmaddwd         m8, m0
+    phaddd          m4, m8
+    paddd           m4, m2
+    psrad           m4, 6
+
+    packusdw        m3, m4
+    vpermq          m3, m3, q3120
+    pshufb          m3, m6
+    CLIPW           m3, m5, m7
+    movu            [r2 + x], m3
+    %assign x x+32
+    %endrep
+
+    add             r0, r1
+    add             r2, r3
+    dec             r6d
+    jnz             .loop
+    RET
+%endif
+%endmacro
+IPFILTER_CHROMA_avx2_64xN 16
+IPFILTER_CHROMA_avx2_64xN 32
+IPFILTER_CHROMA_avx2_64xN 48
+IPFILTER_CHROMA_avx2_64xN 64
+
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+%if ARCH_X86_64
+cglobal interp_4tap_horiz_pp_48x64, 5,6,9
+    add             r1d, r1d
+    add             r3d, r3d
+    sub             r0, 2
+    mov             r4d, r4m
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova            m1, [h4_interp8_hpp_shuf]
+    vpbroadcastd    m2, [pd_32]
+    pxor            m5, m5
+    mova            m6, [idct8_shuf2]
+    mova            m7, [pw_pixel_max]
+
+    mov             r4d, 64
+.loop:
+%assign x 0
+%rep 3
+    vbroadcasti128  m3, [r0 + x]
+    vbroadcasti128  m4, [r0 + 8 + x]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6
+
+    vbroadcasti128  m4, [r0 + 16 + x]
+    vbroadcasti128  m8, [r0 + 24 + x]
+    pshufb          m4, m1
+    pshufb          m8, m1
+
+    pmaddwd         m4, m0
+    pmaddwd         m8, m0
+    phaddd          m4, m8
+    paddd           m4, m2
+    psrad           m4, 6
+
+    packusdw        m3, m4
+    vpermq          m3, m3, q3120
+    pshufb          m3, m6
+    CLIPW           m3, m5, m7
+    movu            [r2 + x], m3
+%assign x x+32
+%endrep
+
+    add             r0, r1
+    add             r2, r3
+    dec             r4d
+    jnz             .loop
+    RET
+%endif
+
+%macro IPFILTER_CHROMA_PS_8xN_AVX2 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_horiz_ps_8x%1, 4, 7, 6
+    add                 r1d, r1d
+    add                 r3d, r3d
+    mov                 r4d, r4m
+    mov                 r5d, r5m
+
+%ifdef PIC
+    lea                 r6, [tab_ChromaCoeff]
+    vpbroadcastq        m0, [r6 + r4 * 8]
+%else
+    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova                m3, [h4_interp8_hpp_shuf]
+    vbroadcasti128      m2, [INTERP_OFFSET_PS]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    sub                 r0, 2
+    test                r5d, r5d
+    mov                 r4d, %1
+    jz                  .loop0
+    sub                 r0, r1
+    add                 r4d, 3
+
+.loop0:
+    vbroadcasti128      m4, [r0]
+    vbroadcasti128      m5, [r0 + 8]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2], xm4
+    add                 r2, r3
+    add                 r0, r1
+    dec                 r4d
+    jnz                 .loop0
+    RET
+%endif
+%endmacro
+
+    IPFILTER_CHROMA_PS_8xN_AVX2 4
+    IPFILTER_CHROMA_PS_8xN_AVX2 8
+    IPFILTER_CHROMA_PS_8xN_AVX2 16
+    IPFILTER_CHROMA_PS_8xN_AVX2 32
+    IPFILTER_CHROMA_PS_8xN_AVX2 6
+    IPFILTER_CHROMA_PS_8xN_AVX2 2
+    IPFILTER_CHROMA_PS_8xN_AVX2 12
+    IPFILTER_CHROMA_PS_8xN_AVX2 64
+
+%macro IPFILTER_CHROMA_PS_16xN_AVX2 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_horiz_ps_16x%1, 4, 7, 6
+    add                 r1d, r1d
+    add                 r3d, r3d
+    mov                 r4d, r4m
+    mov                 r5d, r5m
+
+%ifdef PIC
+    lea                 r6, [tab_ChromaCoeff]
+    vpbroadcastq        m0, [r6 + r4 * 8]
+%else
+    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova                m3, [h4_interp8_hpp_shuf]
+    vbroadcasti128      m2, [INTERP_OFFSET_PS]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    sub                 r0, 2
+    test                r5d, r5d
+    mov                 r4d, %1
+    jz                  .loop0
+    sub                 r0, r1
+    add                 r4d, 3
+
+.loop0:
+    vbroadcasti128      m4, [r0]
+    vbroadcasti128      m5, [r0 + 8]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2], xm4
+
+    vbroadcasti128      m4, [r0 + 16]
+    vbroadcasti128      m5, [r0 + 24]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 16], xm4
+
+    add                 r2, r3
+    add                 r0, r1
+    dec                 r4d
+    jnz                 .loop0
+    RET
+%endif
+%endmacro
+
+IPFILTER_CHROMA_PS_16xN_AVX2 16
+IPFILTER_CHROMA_PS_16xN_AVX2 8
+IPFILTER_CHROMA_PS_16xN_AVX2 32
+IPFILTER_CHROMA_PS_16xN_AVX2 12
+IPFILTER_CHROMA_PS_16xN_AVX2 4
+IPFILTER_CHROMA_PS_16xN_AVX2 64
+IPFILTER_CHROMA_PS_16xN_AVX2 24
+
+%macro IPFILTER_CHROMA_PS_24xN_AVX2 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_horiz_ps_24x%1, 4, 7, 6
+    add                 r1d, r1d
+    add                 r3d, r3d
+    mov                 r4d, r4m
+    mov                 r5d, r5m
+
+%ifdef PIC
+    lea                 r6, [tab_ChromaCoeff]
+    vpbroadcastq        m0, [r6 + r4 * 8]
+%else
+    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova                m3, [h4_interp8_hpp_shuf]
+    vbroadcasti128      m2, [INTERP_OFFSET_PS]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    sub                 r0, 2
+    test                r5d, r5d
+    mov                 r4d, %1
+    jz                  .loop0
+    sub                 r0, r1
+    add                 r4d, 3
+
+.loop0:
+    vbroadcasti128      m4, [r0]
+    vbroadcasti128      m5, [r0 + 8]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2], xm4
+
+    vbroadcasti128      m4, [r0 + 16]
+    vbroadcasti128      m5, [r0 + 24]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 16], xm4
+
+    vbroadcasti128      m4, [r0 + 32]
+    vbroadcasti128      m5, [r0 + 40]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 32], xm4
+
+    add                 r2, r3
+    add                 r0, r1
+    dec                 r4d
+    jnz                 .loop0
+    RET
+%endif
+%endmacro
+
+IPFILTER_CHROMA_PS_24xN_AVX2 32
+IPFILTER_CHROMA_PS_24xN_AVX2 64
+
+%macro IPFILTER_CHROMA_PS_12xN_AVX2 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_horiz_ps_12x%1, 4, 7, 6
+    add                 r1d, r1d
+    add                 r3d, r3d
+    mov                 r4d, r4m
+    mov                 r5d, r5m
+
+%ifdef PIC
+    lea                 r6, [tab_ChromaCoeff]
+    vpbroadcastq        m0, [r6 + r4 * 8]
+%else
+    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova                m3, [h4_interp8_hpp_shuf]
+    vbroadcasti128      m2, [INTERP_OFFSET_PS]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    sub                 r0, 2
+    test                r5d, r5d
+    mov                 r4d, %1
+    jz                  .loop0
+    sub                 r0, r1
+    add                 r4d, 3
+
+.loop0:
+    vbroadcasti128      m4, [r0]
+    vbroadcasti128      m5, [r0 + 8]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2], xm4
+
+    vbroadcasti128      m4, [r0 + 16]
+    pshufb              m4, m3
+    pmaddwd             m4, m0
+    phaddd              m4, m4
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movq                [r2 + 16], xm4
+
+    add                 r2, r3
+    add                 r0, r1
+    dec                 r4d
+    jnz                 .loop0
+    RET
+%endif
+%endmacro
+
+IPFILTER_CHROMA_PS_12xN_AVX2 16
+IPFILTER_CHROMA_PS_12xN_AVX2 32
+
+%macro IPFILTER_CHROMA_PS_32xN_AVX2 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_horiz_ps_32x%1, 4, 7, 6
+    add                 r1d, r1d
+    add                 r3d, r3d
+    mov                 r4d, r4m
+    mov                 r5d, r5m
+
+%ifdef PIC
+    lea                 r6, [tab_ChromaCoeff]
+    vpbroadcastq        m0, [r6 + r4 * 8]
+%else
+    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova                m3, [h4_interp8_hpp_shuf]
+    vbroadcasti128      m2, [INTERP_OFFSET_PS]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    sub                 r0, 2
+    test                r5d, r5d
+    mov                 r4d, %1
+    jz                  .loop0
+    sub                 r0, r1
+    add                 r4d, 3
+
+.loop0:
+    vbroadcasti128      m4, [r0]
+    vbroadcasti128      m5, [r0 + 8]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2], xm4
+
+    vbroadcasti128      m4, [r0 + 16]
+    vbroadcasti128      m5, [r0 + 24]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 16], xm4
+
+    vbroadcasti128      m4, [r0 + 32]
+    vbroadcasti128      m5, [r0 + 40]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 32], xm4
+
+    vbroadcasti128      m4, [r0 + 48]
+    vbroadcasti128      m5, [r0 + 56]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 48], xm4
+
+    add                 r2, r3
+    add                 r0, r1
+    dec                 r4d
+    jnz                 .loop0
+    RET
+%endif
+%endmacro
+
+IPFILTER_CHROMA_PS_32xN_AVX2 32
+IPFILTER_CHROMA_PS_32xN_AVX2 16
+IPFILTER_CHROMA_PS_32xN_AVX2 24
+IPFILTER_CHROMA_PS_32xN_AVX2 8
+IPFILTER_CHROMA_PS_32xN_AVX2 64
+IPFILTER_CHROMA_PS_32xN_AVX2 48
+
+
+%macro IPFILTER_CHROMA_PS_64xN_AVX2 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_horiz_ps_64x%1, 4, 7, 6
+    add                 r1d, r1d
+    add                 r3d, r3d
+    mov                 r4d, r4m
+    mov                 r5d, r5m
+
+%ifdef PIC
+    lea                 r6, [tab_ChromaCoeff]
+    vpbroadcastq        m0, [r6 + r4 * 8]
+%else
+    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova                m3, [h4_interp8_hpp_shuf]
+    vbroadcasti128      m2, [INTERP_OFFSET_PS]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    sub                 r0, 2
+    test                r5d, r5d
+    mov                 r4d, %1
+    jz                  .loop0
+    sub                 r0, r1
+    add                 r4d, 3
+
+.loop0:
+    vbroadcasti128      m4, [r0]
+    vbroadcasti128      m5, [r0 + 8]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2], xm4
+
+    vbroadcasti128      m4, [r0 + 16]
+    vbroadcasti128      m5, [r0 + 24]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 16], xm4
+
+    vbroadcasti128      m4, [r0 + 32]
+    vbroadcasti128      m5, [r0 + 40]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 32], xm4
+
+    vbroadcasti128      m4, [r0 + 48]
+    vbroadcasti128      m5, [r0 + 56]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 48], xm4
+
+    vbroadcasti128      m4, [r0 + 64]
+    vbroadcasti128      m5, [r0 + 72]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 64], xm4
+
+    vbroadcasti128      m4, [r0 + 80]
+    vbroadcasti128      m5, [r0 + 88]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 80], xm4
+
+    vbroadcasti128      m4, [r0 + 96]
+    vbroadcasti128      m5, [r0 + 104]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 96], xm4
+
+    vbroadcasti128      m4, [r0 + 112]
+    vbroadcasti128      m5, [r0 + 120]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 112], xm4
+
+    add                 r2, r3
+    add                 r0, r1
+    dec                 r4d
+    jnz                 .loop0
+    RET
+%endif
+%endmacro
+
+IPFILTER_CHROMA_PS_64xN_AVX2 64
+IPFILTER_CHROMA_PS_64xN_AVX2 48
+IPFILTER_CHROMA_PS_64xN_AVX2 32
+IPFILTER_CHROMA_PS_64xN_AVX2 16
+
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_horiz_ps_48x64, 4, 7, 6
+    add                 r1d, r1d
+    add                 r3d, r3d
+    mov                 r4d, r4m
+    mov                 r5d, r5m
+
+%ifdef PIC
+    lea                 r6, [tab_ChromaCoeff]
+    vpbroadcastq        m0, [r6 + r4 * 8]
+%else
+    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova                m3, [h4_interp8_hpp_shuf]
+    vbroadcasti128      m2, [INTERP_OFFSET_PS]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    sub                 r0, 2
+    test                r5d, r5d
+    mov                 r4d, 64
+    jz                  .loop0
+    sub                 r0, r1
+    add                 r4d, 3
+
+.loop0:
+    vbroadcasti128      m4, [r0]
+    vbroadcasti128      m5, [r0 + 8]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2], xm4
+
+    vbroadcasti128      m4, [r0 + 16]
+    vbroadcasti128      m5, [r0 + 24]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 16], xm4
+
+    vbroadcasti128      m4, [r0 + 32]
+    vbroadcasti128      m5, [r0 + 40]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 32], xm4
+
+    vbroadcasti128      m4, [r0 + 48]
+    vbroadcasti128      m5, [r0 + 56]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 48], xm4
+
+    vbroadcasti128      m4, [r0 + 64]
+    vbroadcasti128      m5, [r0 + 72]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 64], xm4
+
+    vbroadcasti128      m4, [r0 + 80]
+    vbroadcasti128      m5, [r0 + 88]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movu                [r2 + 80], xm4
+
+    add                 r2, r3
+    add                 r0, r1
+    dec                 r4d
+    jnz                 .loop0
+    RET
+%endif
+
+%macro IPFILTER_CHROMA_PS_6xN_AVX2 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_horiz_ps_6x%1, 4, 7, 6
+    add                 r1d, r1d
+    add                 r3d, r3d
+    mov                 r4d, r4m
+    mov                 r5d, r5m
+
+%ifdef PIC
+    lea                 r6, [tab_ChromaCoeff]
+    vpbroadcastq        m0, [r6 + r4 * 8]
+%else
+    vpbroadcastq        m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova                m3, [h4_interp8_hpp_shuf]
+    vbroadcasti128      m2, [INTERP_OFFSET_PS]
+
+    ; register map
+    ; m0 , m1 interpolate coeff
+
+    sub                 r0, 2
+    test                r5d, r5d
+    mov                 r4d, %1
+    jz                  .loop0
+    sub                 r0, r1
+    add                 r4d, 3
+
+.loop0:
+    vbroadcasti128      m4, [r0]
+    vbroadcasti128      m5, [r0 + 8]
+    pshufb              m4, m3
+    pshufb              m5, m3
+    pmaddwd             m4, m0
+    pmaddwd             m5, m0
+    phaddd              m4, m5
+    paddd               m4, m2
+    vpermq              m4, m4, q3120
+    psrad               m4, INTERP_SHIFT_PS
+    vextracti128        xm5, m4, 1
+    packssdw            xm4, xm5
+    movq                [r2], xm4
+    pextrd              [r2 + 8], xm4, 2
+    add                 r2, r3
+    add                 r0, r1
+    dec                 r4d
+    jnz                 .loop0
+    RET
+%endif
+%endmacro
+
+    IPFILTER_CHROMA_PS_6xN_AVX2 8
+    IPFILTER_CHROMA_PS_6xN_AVX2 16

From 36301974e97b0b7a978f2d8e0ff9a2b863f9b623 Mon Sep 17 00:00:00 2001
From: Mythreyi P <mythreyi@multicorewareinc.com>
Date: Thu, 15 Feb 2018 02:21:26 -0800
Subject: [PATCH 50/51] x86: Split ipfilter16 kernals part3

Port vertical 4tap kernals from ipfilter16.asm to new source file, v4-ipfilter16.asm to reduce compile time of x265.
---
 source/common/CMakeLists.txt        |    2 +-
 source/common/x86/ipfilter16.asm    | 7604 ++++++++-------------------
 source/common/x86/v4-ipfilter16.asm | 3529 +++++++++++++
 3 files changed, 5598 insertions(+), 5537 deletions(-)
 create mode 100644 source/common/x86/v4-ipfilter16.asm

diff --git a/source/common/CMakeLists.txt b/source/common/CMakeLists.txt
index 0eb2dfb8f2..c70bb108c9 100644
--- a/source/common/CMakeLists.txt
+++ b/source/common/CMakeLists.txt
@@ -61,7 +61,7 @@ if(ENABLE_ASSEMBLY AND X86)
                mc-a2.asm pixel-util8.asm blockcopy8.asm
                pixeladd8.asm dct8.asm seaintegral.asm)
     if(HIGH_BIT_DEPTH)
-        set(A_SRCS ${A_SRCS} sad16-a.asm intrapred16.asm h4-ipfilter16.asm h-ipfilter16.asm ipfilter16.asm loopfilter.asm)
+        set(A_SRCS ${A_SRCS} sad16-a.asm intrapred16.asm v4-ipfilter16.asm h4-ipfilter16.asm h-ipfilter16.asm ipfilter16.asm loopfilter.asm)
     else()
         set(A_SRCS ${A_SRCS} sad-a.asm intrapred8.asm intrapred8_allangs.asm v4-ipfilter8.asm h-ipfilter8.asm ipfilter8.asm loopfilter.asm)
     endif()
diff --git a/source/common/x86/ipfilter16.asm b/source/common/x86/ipfilter16.asm
index c5b775121e..9582ccabeb 100644
--- a/source/common/x86/ipfilter16.asm
+++ b/source/common/x86/ipfilter16.asm
@@ -51,54 +51,6 @@ tab_c_524800:     times 4 dd 524800
 tab_c_n8192:      times 8 dw -8192
 pd_524800:        times 8 dd 524800
 
-const tab_ChromaCoeffV,  times 8 dw 0, 64
-                         times 8 dw 0, 0
-
-                         times 8 dw -2, 58
-                         times 8 dw 10, -2
-
-                         times 8 dw -4, 54
-                         times 8 dw 16, -2
-
-                         times 8 dw -6, 46
-                         times 8 dw 28, -4
-
-                         times 8 dw -4, 36
-                         times 8 dw 36, -4
-
-                         times 8 dw -4, 28
-                         times 8 dw 46, -6
-
-                         times 8 dw -2, 16
-                         times 8 dw 54, -4
-
-                         times 8 dw -2, 10
-                         times 8 dw 58, -2
-
-tab_ChromaCoeffVer: times 8 dw 0, 64
-                    times 8 dw 0, 0
-
-                    times 8 dw -2, 58
-                    times 8 dw 10, -2
-
-                    times 8 dw -4, 54
-                    times 8 dw 16, -2
-
-                    times 8 dw -6, 46
-                    times 8 dw 28, -4
-
-                    times 8 dw -4, 36
-                    times 8 dw 36, -4
-
-                    times 8 dw -4, 28
-                    times 8 dw 46, -6
-
-                    times 8 dw -2, 16
-                    times 8 dw 54, -4
-
-                    times 8 dw -2, 10
-                    times 8 dw 58, -2
-
 ALIGN 32
 tab_LumaCoeffV:   times 4 dw 0, 0
                   times 4 dw 0, 64
@@ -349,121 +301,115 @@ cglobal interp_8tap_vert_%1_%2x%3, 5, 7, 8
     FILTER_VER_LUMA_sse2 ps, 64, 16
     FILTER_VER_LUMA_sse2 ps, 16, 64
 
-%macro PROCESS_CHROMA_SP_W4_4R 0
+
+%macro PROCESS_LUMA_VER_W4_4R 0
     movq       m0, [r0]
     movq       m1, [r0 + r1]
     punpcklwd  m0, m1                          ;m0=[0 1]
-    pmaddwd    m0, [r6 + 0 *32]                ;m0=[0+1]         Row1
+    pmaddwd    m0, [r6 + 0 *16]                ;m0=[0+1]  Row1
 
     lea        r0, [r0 + 2 * r1]
     movq       m4, [r0]
     punpcklwd  m1, m4                          ;m1=[1 2]
-    pmaddwd    m1, [r6 + 0 *32]                ;m1=[1+2]         Row2
+    pmaddwd    m1, [r6 + 0 *16]                ;m1=[1+2]  Row2
 
     movq       m5, [r0 + r1]
     punpcklwd  m4, m5                          ;m4=[2 3]
-    pmaddwd    m2, m4, [r6 + 0 *32]            ;m2=[2+3]         Row3
-    pmaddwd    m4, [r6 + 1 * 32]
-    paddd      m0, m4                          ;m0=[0+1+2+3]     Row1 done
+    pmaddwd    m2, m4, [r6 + 0 *16]            ;m2=[2+3]  Row3
+    pmaddwd    m4, [r6 + 1 * 16]
+    paddd      m0, m4                          ;m0=[0+1+2+3]  Row1
 
     lea        r0, [r0 + 2 * r1]
     movq       m4, [r0]
     punpcklwd  m5, m4                          ;m5=[3 4]
-    pmaddwd    m3, m5, [r6 + 0 *32]            ;m3=[3+4]         Row4
-    pmaddwd    m5, [r6 + 1 * 32]
-    paddd      m1, m5                          ;m1 = [1+2+3+4]   Row2
+    pmaddwd    m3, m5, [r6 + 0 *16]            ;m3=[3+4]  Row4
+    pmaddwd    m5, [r6 + 1 * 16]
+    paddd      m1, m5                          ;m1 = [1+2+3+4]  Row2
 
     movq       m5, [r0 + r1]
     punpcklwd  m4, m5                          ;m4=[4 5]
-    pmaddwd    m4, [r6 + 1 * 32]
-    paddd      m2, m4                          ;m2=[2+3+4+5]     Row3
+    pmaddwd    m6, m4, [r6 + 1 * 16]
+    paddd      m2, m6                          ;m2=[2+3+4+5]  Row3
+    pmaddwd    m4, [r6 + 2 * 16]
+    paddd      m0, m4                          ;m0=[0+1+2+3+4+5]  Row1
 
-    movq       m4, [r0 + 2 * r1]
+    lea        r0, [r0 + 2 * r1]
+    movq       m4, [r0]
     punpcklwd  m5, m4                          ;m5=[5 6]
-    pmaddwd    m5, [r6 + 1 * 32]
-    paddd      m3, m5                          ;m3=[3+4+5+6]     Row4
+    pmaddwd    m6, m5, [r6 + 1 * 16]
+    paddd      m3, m6                          ;m3=[3+4+5+6]  Row4
+    pmaddwd    m5, [r6 + 2 * 16]
+    paddd      m1, m5                          ;m1=[1+2+3+4+5+6]  Row2
+
+    movq       m5, [r0 + r1]
+    punpcklwd  m4, m5                          ;m4=[6 7]
+    pmaddwd    m6, m4, [r6 + 2 * 16]
+    paddd      m2, m6                          ;m2=[2+3+4+5+6+7]  Row3
+    pmaddwd    m4, [r6 + 3 * 16]
+    paddd      m0, m4                          ;m0=[0+1+2+3+4+5+6+7]  Row1 end
+
+    lea        r0, [r0 + 2 * r1]
+    movq       m4, [r0]
+    punpcklwd  m5, m4                          ;m5=[7 8]
+    pmaddwd    m6, m5, [r6 + 2 * 16]
+    paddd      m3, m6                          ;m3=[3+4+5+6+7+8]  Row4
+    pmaddwd    m5, [r6 + 3 * 16]
+    paddd      m1, m5                          ;m1=[1+2+3+4+5+6+7+8]  Row2 end
+
+    movq       m5, [r0 + r1]
+    punpcklwd  m4, m5                          ;m4=[8 9]
+    pmaddwd    m4, [r6 + 3 * 16]
+    paddd      m2, m4                          ;m2=[2+3+4+5+6+7+8+9]  Row3 end
+
+    movq       m4, [r0 + 2 * r1]
+    punpcklwd  m5, m4                          ;m5=[9 10]
+    pmaddwd    m5, [r6 + 3 * 16]
+    paddd      m3, m5                          ;m3=[3+4+5+6+7+8+9+10]  Row4 end
 %endmacro
 
-;-----------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_%3_%1x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_SS 4
-INIT_XMM sse2
-cglobal interp_4tap_vert_%3_%1x%2, 5, 7, %4 ,0-gprsize
+;--------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_LUMA_PP 2
+INIT_XMM sse4
+cglobal interp_8tap_vert_pp_%1x%2, 5, 7, 8 ,0-gprsize
 
     add       r1d, r1d
     add       r3d, r3d
-    sub       r0, r1
+    lea       r5, [r1 + 2 * r1]
+    sub       r0, r5
     shl       r4d, 6
 
 %ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
+    lea       r5, [tab_LumaCoeffV]
     lea       r6, [r5 + r4]
 %else
-    lea       r6, [tab_ChromaCoeffV + r4]
+    lea       r6, [tab_LumaCoeffV + r4]
 %endif
 
-    mov       dword [rsp], %2/4
-
-%ifnidn %3, ss
-    %ifnidn %3, ps
-        mova      m7, [pw_pixel_max]
-        %ifidn %3, pp
-            mova      m6, [INTERP_OFFSET_PP]
-        %else
-            mova      m6, [INTERP_OFFSET_SP]
-        %endif
-    %else
-        mova      m6, [INTERP_OFFSET_PS]
-    %endif
-%endif
+    mova      m7, [INTERP_OFFSET_PP]
 
+    mov       dword [rsp], %2/4
 .loopH:
     mov       r4d, (%1/4)
 .loopW:
-    PROCESS_CHROMA_SP_W4_4R
+    PROCESS_LUMA_VER_W4_4R
 
-%ifidn %3, ss
-    psrad     m0, 6
-    psrad     m1, 6
-    psrad     m2, 6
-    psrad     m3, 6
+    paddd     m0, m7
+    paddd     m1, m7
+    paddd     m2, m7
+    paddd     m3, m7
 
-    packssdw  m0, m1
-    packssdw  m2, m3
-%elifidn %3, ps
-    paddd     m0, m6
-    paddd     m1, m6
-    paddd     m2, m6
-    paddd     m3, m6
-    psrad     m0, INTERP_SHIFT_PS
-    psrad     m1, INTERP_SHIFT_PS
-    psrad     m2, INTERP_SHIFT_PS
-    psrad     m3, INTERP_SHIFT_PS
+    psrad     m0, INTERP_SHIFT_PP
+    psrad     m1, INTERP_SHIFT_PP
+    psrad     m2, INTERP_SHIFT_PP
+    psrad     m3, INTERP_SHIFT_PP
 
     packssdw  m0, m1
     packssdw  m2, m3
-%else
-    paddd     m0, m6
-    paddd     m1, m6
-    paddd     m2, m6
-    paddd     m3, m6
-    %ifidn %3, pp
-        psrad     m0, INTERP_SHIFT_PP
-        psrad     m1, INTERP_SHIFT_PP
-        psrad     m2, INTERP_SHIFT_PP
-        psrad     m3, INTERP_SHIFT_PP
-    %else
-        psrad     m0, INTERP_SHIFT_SP
-        psrad     m1, INTERP_SHIFT_SP
-        psrad     m2, INTERP_SHIFT_SP
-        psrad     m3, INTERP_SHIFT_SP
-    %endif
-    packssdw  m0, m1
-    packssdw  m2, m3
-    pxor      m5, m5
-    CLIPW2    m0, m2, m5, m7
-%endif
+
+    pxor      m1, m1
+    CLIPW2    m0, m2, m1, [pw_pixel_max]
 
     movh      [r2], m0
     movhps    [r2 + r3], m0
@@ -471,7 +417,7 @@ cglobal interp_4tap_vert_%3_%1x%2, 5, 7, %4 ,0-gprsize
     movh      [r5], m2
     movhps    [r5 + r3], m2
 
-    lea       r5, [4 * r1 - 2 * 4]
+    lea       r5, [8 * r1 - 2 * 4]
     sub       r0, r5
     add       r2, 2 * 4
 
@@ -483,776 +429,287 @@ cglobal interp_4tap_vert_%3_%1x%2, 5, 7, %4 ,0-gprsize
 
     dec       dword [rsp]
     jnz       .loopH
-
     RET
 %endmacro
 
-    FILTER_VER_CHROMA_SS 4, 4, ss, 6
-    FILTER_VER_CHROMA_SS 4, 8, ss, 6
-    FILTER_VER_CHROMA_SS 16, 16, ss, 6
-    FILTER_VER_CHROMA_SS 16, 8, ss, 6
-    FILTER_VER_CHROMA_SS 16, 12, ss, 6
-    FILTER_VER_CHROMA_SS 12, 16, ss, 6
-    FILTER_VER_CHROMA_SS 16, 4, ss, 6
-    FILTER_VER_CHROMA_SS 4, 16, ss, 6
-    FILTER_VER_CHROMA_SS 32, 32, ss, 6
-    FILTER_VER_CHROMA_SS 32, 16, ss, 6
-    FILTER_VER_CHROMA_SS 16, 32, ss, 6
-    FILTER_VER_CHROMA_SS 32, 24, ss, 6
-    FILTER_VER_CHROMA_SS 24, 32, ss, 6
-    FILTER_VER_CHROMA_SS 32, 8, ss, 6
-
-    FILTER_VER_CHROMA_SS 4, 4, ps, 7
-    FILTER_VER_CHROMA_SS 4, 8, ps, 7
-    FILTER_VER_CHROMA_SS 16, 16, ps, 7
-    FILTER_VER_CHROMA_SS 16, 8, ps, 7
-    FILTER_VER_CHROMA_SS 16, 12, ps, 7
-    FILTER_VER_CHROMA_SS 12, 16, ps, 7
-    FILTER_VER_CHROMA_SS 16, 4, ps, 7
-    FILTER_VER_CHROMA_SS 4, 16, ps, 7
-    FILTER_VER_CHROMA_SS 32, 32, ps, 7
-    FILTER_VER_CHROMA_SS 32, 16, ps, 7
-    FILTER_VER_CHROMA_SS 16, 32, ps, 7
-    FILTER_VER_CHROMA_SS 32, 24, ps, 7
-    FILTER_VER_CHROMA_SS 24, 32, ps, 7
-    FILTER_VER_CHROMA_SS 32, 8, ps, 7
-
-    FILTER_VER_CHROMA_SS 4, 4, sp, 8
-    FILTER_VER_CHROMA_SS 4, 8, sp, 8
-    FILTER_VER_CHROMA_SS 16, 16, sp, 8
-    FILTER_VER_CHROMA_SS 16, 8, sp, 8
-    FILTER_VER_CHROMA_SS 16, 12, sp, 8
-    FILTER_VER_CHROMA_SS 12, 16, sp, 8
-    FILTER_VER_CHROMA_SS 16, 4, sp, 8
-    FILTER_VER_CHROMA_SS 4, 16, sp, 8
-    FILTER_VER_CHROMA_SS 32, 32, sp, 8
-    FILTER_VER_CHROMA_SS 32, 16, sp, 8
-    FILTER_VER_CHROMA_SS 16, 32, sp, 8
-    FILTER_VER_CHROMA_SS 32, 24, sp, 8
-    FILTER_VER_CHROMA_SS 24, 32, sp, 8
-    FILTER_VER_CHROMA_SS 32, 8, sp, 8
-
-    FILTER_VER_CHROMA_SS 4, 4, pp, 8
-    FILTER_VER_CHROMA_SS 4, 8, pp, 8
-    FILTER_VER_CHROMA_SS 16, 16, pp, 8
-    FILTER_VER_CHROMA_SS 16, 8, pp, 8
-    FILTER_VER_CHROMA_SS 16, 12, pp, 8
-    FILTER_VER_CHROMA_SS 12, 16, pp, 8
-    FILTER_VER_CHROMA_SS 16, 4, pp, 8
-    FILTER_VER_CHROMA_SS 4, 16, pp, 8
-    FILTER_VER_CHROMA_SS 32, 32, pp, 8
-    FILTER_VER_CHROMA_SS 32, 16, pp, 8
-    FILTER_VER_CHROMA_SS 16, 32, pp, 8
-    FILTER_VER_CHROMA_SS 32, 24, pp, 8
-    FILTER_VER_CHROMA_SS 24, 32, pp, 8
-    FILTER_VER_CHROMA_SS 32, 8, pp, 8
-
-
-    FILTER_VER_CHROMA_SS 16, 24, ss, 6
-    FILTER_VER_CHROMA_SS 12, 32, ss, 6
-    FILTER_VER_CHROMA_SS 4, 32, ss, 6
-    FILTER_VER_CHROMA_SS 32, 64, ss, 6
-    FILTER_VER_CHROMA_SS 16, 64, ss, 6
-    FILTER_VER_CHROMA_SS 32, 48, ss, 6
-    FILTER_VER_CHROMA_SS 24, 64, ss, 6
-
-    FILTER_VER_CHROMA_SS 16, 24, ps, 7
-    FILTER_VER_CHROMA_SS 12, 32, ps, 7
-    FILTER_VER_CHROMA_SS 4, 32, ps, 7
-    FILTER_VER_CHROMA_SS 32, 64, ps, 7
-    FILTER_VER_CHROMA_SS 16, 64, ps, 7
-    FILTER_VER_CHROMA_SS 32, 48, ps, 7
-    FILTER_VER_CHROMA_SS 24, 64, ps, 7
-
-    FILTER_VER_CHROMA_SS 16, 24, sp, 8
-    FILTER_VER_CHROMA_SS 12, 32, sp, 8
-    FILTER_VER_CHROMA_SS 4, 32, sp, 8
-    FILTER_VER_CHROMA_SS 32, 64, sp, 8
-    FILTER_VER_CHROMA_SS 16, 64, sp, 8
-    FILTER_VER_CHROMA_SS 32, 48, sp, 8
-    FILTER_VER_CHROMA_SS 24, 64, sp, 8
-
-    FILTER_VER_CHROMA_SS 16, 24, pp, 8
-    FILTER_VER_CHROMA_SS 12, 32, pp, 8
-    FILTER_VER_CHROMA_SS 4, 32, pp, 8
-    FILTER_VER_CHROMA_SS 32, 64, pp, 8
-    FILTER_VER_CHROMA_SS 16, 64, pp, 8
-    FILTER_VER_CHROMA_SS 32, 48, pp, 8
-    FILTER_VER_CHROMA_SS 24, 64, pp, 8
-
-
-    FILTER_VER_CHROMA_SS 48, 64, ss, 6
-    FILTER_VER_CHROMA_SS 64, 48, ss, 6
-    FILTER_VER_CHROMA_SS 64, 64, ss, 6
-    FILTER_VER_CHROMA_SS 64, 32, ss, 6
-    FILTER_VER_CHROMA_SS 64, 16, ss, 6
-
-    FILTER_VER_CHROMA_SS 48, 64, ps, 7
-    FILTER_VER_CHROMA_SS 64, 48, ps, 7
-    FILTER_VER_CHROMA_SS 64, 64, ps, 7
-    FILTER_VER_CHROMA_SS 64, 32, ps, 7
-    FILTER_VER_CHROMA_SS 64, 16, ps, 7
-
-    FILTER_VER_CHROMA_SS 48, 64, sp, 8
-    FILTER_VER_CHROMA_SS 64, 48, sp, 8
-    FILTER_VER_CHROMA_SS 64, 64, sp, 8
-    FILTER_VER_CHROMA_SS 64, 32, sp, 8
-    FILTER_VER_CHROMA_SS 64, 16, sp, 8
-
-    FILTER_VER_CHROMA_SS 48, 64, pp, 8
-    FILTER_VER_CHROMA_SS 64, 48, pp, 8
-    FILTER_VER_CHROMA_SS 64, 64, pp, 8
-    FILTER_VER_CHROMA_SS 64, 32, pp, 8
-    FILTER_VER_CHROMA_SS 64, 16, pp, 8
-
-
-%macro PROCESS_CHROMA_SP_W2_4R 1
-    movd       m0, [r0]
-    movd       m1, [r0 + r1]
-    punpcklwd  m0, m1                          ;m0=[0 1]
-
-    lea        r0, [r0 + 2 * r1]
-    movd       m2, [r0]
-    punpcklwd  m1, m2                          ;m1=[1 2]
-    punpcklqdq m0, m1                          ;m0=[0 1 1 2]
-    pmaddwd    m0, [%1 + 0 *32]                ;m0=[0+1 1+2] Row 1-2
-
-    movd       m1, [r0 + r1]
-    punpcklwd  m2, m1                          ;m2=[2 3]
-
-    lea        r0, [r0 + 2 * r1]
-    movd       m3, [r0]
-    punpcklwd  m1, m3                          ;m2=[3 4]
-    punpcklqdq m2, m1                          ;m2=[2 3 3 4]
-
-    pmaddwd    m4, m2, [%1 + 1 * 32]           ;m4=[2+3 3+4] Row 1-2
-    pmaddwd    m2, [%1 + 0 * 32]               ;m2=[2+3 3+4] Row 3-4
-    paddd      m0, m4                          ;m0=[0+1+2+3 1+2+3+4] Row 1-2
-
-    movd       m1, [r0 + r1]
-    punpcklwd  m3, m1                          ;m3=[4 5]
-
-    movd       m4, [r0 + 2 * r1]
-    punpcklwd  m1, m4                          ;m1=[5 6]
-    punpcklqdq m3, m1                          ;m2=[4 5 5 6]
-    pmaddwd    m3, [%1 + 1 * 32]               ;m3=[4+5 5+6] Row 3-4
-    paddd      m2, m3                          ;m2=[2+3+4+5 3+4+5+6] Row 3-4
-%endmacro
-
-;---------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vertical_%2_2x%1(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;---------------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_W2 3
-INIT_XMM sse4
-cglobal interp_4tap_vert_%2_2x%1, 5, 6, %3
+;-------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_PP 4, 4
+    FILTER_VER_LUMA_PP 8, 8
+    FILTER_VER_LUMA_PP 8, 4
+    FILTER_VER_LUMA_PP 4, 8
+    FILTER_VER_LUMA_PP 16, 16
+    FILTER_VER_LUMA_PP 16, 8
+    FILTER_VER_LUMA_PP 8, 16
+    FILTER_VER_LUMA_PP 16, 12
+    FILTER_VER_LUMA_PP 12, 16
+    FILTER_VER_LUMA_PP 16, 4
+    FILTER_VER_LUMA_PP 4, 16
+    FILTER_VER_LUMA_PP 32, 32
+    FILTER_VER_LUMA_PP 32, 16
+    FILTER_VER_LUMA_PP 16, 32
+    FILTER_VER_LUMA_PP 32, 24
+    FILTER_VER_LUMA_PP 24, 32
+    FILTER_VER_LUMA_PP 32, 8
+    FILTER_VER_LUMA_PP 8, 32
+    FILTER_VER_LUMA_PP 64, 64
+    FILTER_VER_LUMA_PP 64, 32
+    FILTER_VER_LUMA_PP 32, 64
+    FILTER_VER_LUMA_PP 64, 48
+    FILTER_VER_LUMA_PP 48, 64
+    FILTER_VER_LUMA_PP 64, 16
+    FILTER_VER_LUMA_PP 16, 64
 
-    add       r1d, r1d
-    add       r3d, r3d
-    sub       r0, r1
-    shl       r4d, 6
+%macro FILTER_VER_LUMA_AVX2_4x4 1
+INIT_YMM avx2
+cglobal interp_8tap_vert_%1_4x4, 4, 6, 7
+    mov             r4d, r4m
+    add             r1d, r1d
+    add             r3d, r3d
+    shl             r4d, 7
 
 %ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
-    lea       r5, [r5 + r4]
+    lea             r5, [tab_LumaCoeffVer]
+    add             r5, r4
 %else
-    lea       r5, [tab_ChromaCoeffV + r4]
+    lea             r5, [tab_LumaCoeffVer + r4]
 %endif
 
-    mov       r4d, (%1/4)
-%ifnidn %2, ss
-    %ifnidn %2, ps
-        pxor      m7, m7
-        mova      m6, [pw_pixel_max]
-        %ifidn %2, pp
-            mova      m5, [INTERP_OFFSET_PP]
-        %else
-            mova      m5, [INTERP_OFFSET_SP]
-        %endif
-    %else
-        mova      m5, [INTERP_OFFSET_PS]
-    %endif
-%endif
+    lea             r4, [r1 * 3]
+    sub             r0, r4
 
-.loopH:
-    PROCESS_CHROMA_SP_W2_4R r5
-%ifidn %2, ss
-    psrad     m0, 6
-    psrad     m2, 6
-    packssdw  m0, m2
-%elifidn %2, ps
-    paddd     m0, m5
-    paddd     m2, m5
-    psrad     m0, INTERP_SHIFT_PS
-    psrad     m2, INTERP_SHIFT_PS
-    packssdw  m0, m2
+%ifidn %1,pp
+    vbroadcasti128  m6, [pd_32]
+%elifidn %1, sp
+    vbroadcasti128  m6, [INTERP_OFFSET_SP]
 %else
-    paddd     m0, m5
-    paddd     m2, m5
-    %ifidn %2, pp
-        psrad     m0, INTERP_SHIFT_PP
-        psrad     m2, INTERP_SHIFT_PP
-    %else
-        psrad     m0, INTERP_SHIFT_SP
-        psrad     m2, INTERP_SHIFT_SP
-    %endif
-    packusdw  m0, m2
-    CLIPW     m0, m7,    m6
+    vbroadcasti128  m6, [INTERP_OFFSET_PS]
 %endif
 
-    movd      [r2], m0
-    pextrd    [r2 + r3], m0, 1
-    lea       r2, [r2 + 2 * r3]
-    pextrd    [r2], m0, 2
-    pextrd    [r2 + r3], m0, 3
-
-    lea       r2, [r2 + 2 * r3]
-
-    dec       r4d
-    jnz       .loopH
-    RET
-%endmacro
-
-FILTER_VER_CHROMA_W2 4, ss, 5
-FILTER_VER_CHROMA_W2 8, ss, 5
-
-FILTER_VER_CHROMA_W2 4, pp, 8
-FILTER_VER_CHROMA_W2 8, pp, 8
-
-FILTER_VER_CHROMA_W2 4, ps, 6
-FILTER_VER_CHROMA_W2 8, ps, 6
-
-FILTER_VER_CHROMA_W2 4, sp, 8
-FILTER_VER_CHROMA_W2 8, sp, 8
-
-FILTER_VER_CHROMA_W2 16, ss, 5
-FILTER_VER_CHROMA_W2 16, pp, 8
-FILTER_VER_CHROMA_W2 16, ps, 6
-FILTER_VER_CHROMA_W2 16, sp, 8
-
-
-;---------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_%1_4x2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;---------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_W4 3
-INIT_XMM sse4
-cglobal interp_4tap_vert_%2_4x%1, 5, 6, %3
-    add        r1d, r1d
-    add        r3d, r3d
-    sub        r0, r1
-    shl        r4d, 6
+    movq            xm0, [r0]
+    movq            xm1, [r0 + r1]
+    punpcklwd       xm0, xm1
+    movq            xm2, [r0 + r1 * 2]
+    punpcklwd       xm1, xm2
+    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
+    pmaddwd         m0, [r5]
+    movq            xm3, [r0 + r4]
+    punpcklwd       xm2, xm3
+    lea             r0, [r0 + 4 * r1]
+    movq            xm4, [r0]
+    punpcklwd       xm3, xm4
+    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
+    pmaddwd         m5, m2, [r5 + 1 * mmsize]
+    pmaddwd         m2, [r5]
+    paddd           m0, m5
+    movq            xm3, [r0 + r1]
+    punpcklwd       xm4, xm3
+    movq            xm1, [r0 + r1 * 2]
+    punpcklwd       xm3, xm1
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
+    pmaddwd         m5, m4, [r5 + 2 * mmsize]
+    pmaddwd         m4, [r5 + 1 * mmsize]
+    paddd           m0, m5
+    paddd           m2, m4
+    movq            xm3, [r0 + r4]
+    punpcklwd       xm1, xm3
+    lea             r0, [r0 + 4 * r1]
+    movq            xm4, [r0]
+    punpcklwd       xm3, xm4
+    vinserti128     m1, m1, xm3, 1                  ; m1 = [8 7 7 6]
+    pmaddwd         m5, m1, [r5 + 3 * mmsize]
+    pmaddwd         m1, [r5 + 2 * mmsize]
+    paddd           m0, m5
+    paddd           m2, m1
+    movq            xm3, [r0 + r1]
+    punpcklwd       xm4, xm3
+    movq            xm1, [r0 + 2 * r1]
+    punpcklwd       xm3, xm1
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [A 9 9 8]
+    pmaddwd         m4, [r5 + 3 * mmsize]
+    paddd           m2, m4
 
-%ifdef PIC
-    lea        r5, [tab_ChromaCoeffV]
-    lea        r5, [r5 + r4]
+%ifidn %1,ss
+    psrad           m0, 6
+    psrad           m2, 6
 %else
-    lea        r5, [tab_ChromaCoeffV + r4]
-%endif
-
-%ifnidn %2, 2
-    mov        r4d, %1/2
+    paddd           m0, m6
+    paddd           m2, m6
+%ifidn %1,pp
+    psrad           m0, INTERP_SHIFT_PP
+    psrad           m2, INTERP_SHIFT_PP
+%elifidn %1, sp
+    psrad           m0, INTERP_SHIFT_SP
+    psrad           m2, INTERP_SHIFT_SP
+%else
+    psrad           m0, INTERP_SHIFT_PS
+    psrad           m2, INTERP_SHIFT_PS
 %endif
-
-%ifnidn %2, ss
-    %ifnidn %2, ps
-        pxor      m6, m6
-        mova      m5, [pw_pixel_max]
-        %ifidn %2, pp
-            mova      m4, [INTERP_OFFSET_PP]
-        %else
-            mova      m4, [INTERP_OFFSET_SP]
-        %endif
-    %else
-        mova      m4, [INTERP_OFFSET_PS]
-    %endif
 %endif
 
-%ifnidn %2, 2
-.loop:
+    packssdw        m0, m2
+    pxor            m1, m1
+%ifidn %1,pp
+    CLIPW           m0, m1, [pw_pixel_max]
+%elifidn %1, sp
+    CLIPW           m0, m1, [pw_pixel_max]
 %endif
 
-    movh       m0, [r0]
-    movh       m1, [r0 + r1]
-    punpcklwd  m0, m1                          ;m0=[0 1]
-    pmaddwd    m0, [r5 + 0 *32]                ;m0=[0+1]  Row1
-
-    lea        r0, [r0 + 2 * r1]
-    movh       m2, [r0]
-    punpcklwd  m1, m2                          ;m1=[1 2]
-    pmaddwd    m1, [r5 + 0 *32]                ;m1=[1+2]  Row2
-
-    movh       m3, [r0 + r1]
-    punpcklwd  m2, m3                          ;m4=[2 3]
-    pmaddwd    m2, [r5 + 1 * 32]
-    paddd      m0, m2                          ;m0=[0+1+2+3]  Row1 done
-
-    movh       m2, [r0 + 2 * r1]
-    punpcklwd  m3, m2                          ;m5=[3 4]
-    pmaddwd    m3, [r5 + 1 * 32]
-    paddd      m1, m3                          ;m1=[1+2+3+4]  Row2 done
-
-%ifidn %2, ss
-    psrad     m0, 6
-    psrad     m1, 6
-    packssdw  m0, m1
-%elifidn %2, ps
-    paddd     m0, m4
-    paddd     m1, m4
-    psrad     m0, INTERP_SHIFT_PS
-    psrad     m1, INTERP_SHIFT_PS
-    packssdw  m0, m1
-%else
-    paddd     m0, m4
-    paddd     m1, m4
-    %ifidn %2, pp
-        psrad     m0, INTERP_SHIFT_PP
-        psrad     m1, INTERP_SHIFT_PP
-    %else
-        psrad     m0, INTERP_SHIFT_SP
-        psrad     m1, INTERP_SHIFT_SP
-    %endif
-    packusdw  m0, m1
-    CLIPW     m0, m6,    m5
-%endif
-
-    movh       [r2], m0
-    movhps     [r2 + r3], m0
-
-%ifnidn %2, 2
-    lea        r2, [r2 + r3 * 2]
-    dec        r4d
-    jnz        .loop
-%endif
-    RET
-%endmacro
-
-FILTER_VER_CHROMA_W4 2, ss, 4
-FILTER_VER_CHROMA_W4 2, pp, 7
-FILTER_VER_CHROMA_W4 2, ps, 5
-FILTER_VER_CHROMA_W4 2, sp, 7
-
-FILTER_VER_CHROMA_W4 4, ss, 4
-FILTER_VER_CHROMA_W4 4, pp, 7
-FILTER_VER_CHROMA_W4 4, ps, 5
-FILTER_VER_CHROMA_W4 4, sp, 7
-
-;-------------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vertical_%1_6x8(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_W6 3
-INIT_XMM sse4
-cglobal interp_4tap_vert_%2_6x%1, 5, 7, %3
-    add       r1d, r1d
-    add       r3d, r3d
-    sub       r0, r1
-    shl       r4d, 6
-
-%ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
-    lea       r6, [r5 + r4]
-%else
-    lea       r6, [tab_ChromaCoeffV + r4]
-%endif
-
-    mov       r4d, %1/4
-
-%ifnidn %2, ss
-    %ifnidn %2, ps
-        mova      m7, [pw_pixel_max]
-        %ifidn %2, pp
-            mova      m6, [INTERP_OFFSET_PP]
-        %else
-            mova      m6, [INTERP_OFFSET_SP]
-        %endif
-    %else
-        mova      m6, [INTERP_OFFSET_PS]
-    %endif
-%endif
-
-.loopH:
-    PROCESS_CHROMA_SP_W4_4R
-
-%ifidn %2, ss
-    psrad     m0, 6
-    psrad     m1, 6
-    psrad     m2, 6
-    psrad     m3, 6
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-%elifidn %2, ps
-    paddd     m0, m6
-    paddd     m1, m6
-    paddd     m2, m6
-    paddd     m3, m6
-    psrad     m0, INTERP_SHIFT_PS
-    psrad     m1, INTERP_SHIFT_PS
-    psrad     m2, INTERP_SHIFT_PS
-    psrad     m3, INTERP_SHIFT_PS
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-%else
-    paddd     m0, m6
-    paddd     m1, m6
-    paddd     m2, m6
-    paddd     m3, m6
-    %ifidn %2, pp
-        psrad     m0, INTERP_SHIFT_PP
-        psrad     m1, INTERP_SHIFT_PP
-        psrad     m2, INTERP_SHIFT_PP
-        psrad     m3, INTERP_SHIFT_PP
-    %else
-        psrad     m0, INTERP_SHIFT_SP
-        psrad     m1, INTERP_SHIFT_SP
-        psrad     m2, INTERP_SHIFT_SP
-        psrad     m3, INTERP_SHIFT_SP
-    %endif
-    packssdw  m0, m1
-    packssdw  m2, m3
-    pxor      m5, m5
-    CLIPW2    m0, m2, m5, m7
-%endif
-
-    movh      [r2], m0
-    movhps    [r2 + r3], m0
-    lea       r5, [r2 + 2 * r3]
-    movh      [r5], m2
-    movhps    [r5 + r3], m2
-
-    lea       r5, [4 * r1 - 2 * 4]
-    sub       r0, r5
-    add       r2, 2 * 4
-
-    PROCESS_CHROMA_SP_W2_4R r6
-
-%ifidn %2, ss
-    psrad     m0, 6
-    psrad     m2, 6
-    packssdw  m0, m2
-%elifidn %2, ps
-    paddd     m0, m6
-    paddd     m2, m6
-    psrad     m0, INTERP_SHIFT_PS
-    psrad     m2, INTERP_SHIFT_PS
-    packssdw  m0, m2
-%else
-    paddd     m0, m6
-    paddd     m2, m6
-    %ifidn %2, pp
-        psrad     m0, INTERP_SHIFT_PP
-        psrad     m2, INTERP_SHIFT_PP
-    %else
-        psrad     m0, INTERP_SHIFT_SP
-        psrad     m2, INTERP_SHIFT_SP
-    %endif
-    packusdw  m0, m2
-    CLIPW     m0, m5,    m7
-%endif
-
-    movd      [r2], m0
-    pextrd    [r2 + r3], m0, 1
-    lea       r2, [r2 + 2 * r3]
-    pextrd    [r2], m0, 2
-    pextrd    [r2 + r3], m0, 3
-
-    sub       r0, 2 * 4
-    lea       r2, [r2 + 2 * r3 - 2 * 4]
-
-    dec       r4d
-    jnz       .loopH
-    RET
-%endmacro
-
-FILTER_VER_CHROMA_W6 8, ss, 6
-FILTER_VER_CHROMA_W6 8, ps, 7
-FILTER_VER_CHROMA_W6 8, sp, 8
-FILTER_VER_CHROMA_W6 8, pp, 8
-
-FILTER_VER_CHROMA_W6 16, ss, 6
-FILTER_VER_CHROMA_W6 16, ps, 7
-FILTER_VER_CHROMA_W6 16, sp, 8
-FILTER_VER_CHROMA_W6 16, pp, 8
-
-%macro PROCESS_CHROMA_SP_W8_2R 0
-    movu       m1, [r0]
-    movu       m3, [r0 + r1]
-    punpcklwd  m0, m1, m3
-    pmaddwd    m0, [r5 + 0 * 32]                ;m0 = [0l+1l]  Row1l
-    punpckhwd  m1, m3
-    pmaddwd    m1, [r5 + 0 * 32]                ;m1 = [0h+1h]  Row1h
-
-    movu       m4, [r0 + 2 * r1]
-    punpcklwd  m2, m3, m4
-    pmaddwd    m2, [r5 + 0 * 32]                ;m2 = [1l+2l]  Row2l
-    punpckhwd  m3, m4
-    pmaddwd    m3, [r5 + 0 * 32]                ;m3 = [1h+2h]  Row2h
-
-    lea        r0, [r0 + 2 * r1]
-    movu       m5, [r0 + r1]
-    punpcklwd  m6, m4, m5
-    pmaddwd    m6, [r5 + 1 * 32]                ;m6 = [2l+3l]  Row1l
-    paddd      m0, m6                           ;m0 = [0l+1l+2l+3l]  Row1l sum
-    punpckhwd  m4, m5
-    pmaddwd    m4, [r5 + 1 * 32]                ;m6 = [2h+3h]  Row1h
-    paddd      m1, m4                           ;m1 = [0h+1h+2h+3h]  Row1h sum
-
-    movu       m4, [r0 + 2 * r1]
-    punpcklwd  m6, m5, m4
-    pmaddwd    m6, [r5 + 1 * 32]                ;m6 = [3l+4l]  Row2l
-    paddd      m2, m6                           ;m2 = [1l+2l+3l+4l]  Row2l sum
-    punpckhwd  m5, m4
-    pmaddwd    m5, [r5 + 1 * 32]                ;m1 = [3h+4h]  Row2h
-    paddd      m3, m5                           ;m3 = [1h+2h+3h+4h]  Row2h sum
-%endmacro
-
-;----------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert_%3_%1x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;----------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_W8 4
-INIT_XMM sse2
-cglobal interp_4tap_vert_%3_%1x%2, 5, 6, %4
-
-    add       r1d, r1d
-    add       r3d, r3d
-    sub       r0, r1
-    shl       r4d, 6
-
-%ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
-    lea       r5, [r5 + r4]
-%else
-    lea       r5, [tab_ChromaCoeffV + r4]
-%endif
-
-    mov       r4d, %2/2
-
-%ifidn %3, pp
-    mova      m7, [INTERP_OFFSET_PP]
-%elifidn %3, sp
-    mova      m7, [INTERP_OFFSET_SP]
-%elifidn %3, ps
-    mova      m7, [INTERP_OFFSET_PS]
-%endif
-
-.loopH:
-    PROCESS_CHROMA_SP_W8_2R
-
-%ifidn %3, ss
-    psrad     m0, 6
-    psrad     m1, 6
-    psrad     m2, 6
-    psrad     m3, 6
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-%elifidn %3, ps
-    paddd     m0, m7
-    paddd     m1, m7
-    paddd     m2, m7
-    paddd     m3, m7
-    psrad     m0, INTERP_SHIFT_PS
-    psrad     m1, INTERP_SHIFT_PS
-    psrad     m2, INTERP_SHIFT_PS
-    psrad     m3, INTERP_SHIFT_PS
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-%else
-    paddd     m0, m7
-    paddd     m1, m7
-    paddd     m2, m7
-    paddd     m3, m7
-    %ifidn %3, pp
-        psrad     m0, INTERP_SHIFT_PP
-        psrad     m1, INTERP_SHIFT_PP
-        psrad     m2, INTERP_SHIFT_PP
-        psrad     m3, INTERP_SHIFT_PP
-    %else
-        psrad     m0, INTERP_SHIFT_SP
-        psrad     m1, INTERP_SHIFT_SP
-        psrad     m2, INTERP_SHIFT_SP
-        psrad     m3, INTERP_SHIFT_SP
-    %endif
-    packssdw  m0, m1
-    packssdw  m2, m3
-    pxor      m5, m5
-    mova      m6, [pw_pixel_max]
-    CLIPW2    m0, m2, m5, m6
-%endif
-
-    movu      [r2], m0
-    movu      [r2 + r3], m2
-
-    lea       r2, [r2 + 2 * r3]
-
-    dec       r4d
-    jnz       .loopH
+    vextracti128    xm2, m0, 1
+    lea             r4, [r3 * 3]
+    movq            [r2], xm0
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r4], xm2
     RET
 %endmacro
 
-FILTER_VER_CHROMA_W8 8, 2, ss, 7
-FILTER_VER_CHROMA_W8 8, 4, ss, 7
-FILTER_VER_CHROMA_W8 8, 6, ss, 7
-FILTER_VER_CHROMA_W8 8, 8, ss, 7
-FILTER_VER_CHROMA_W8 8, 16, ss, 7
-FILTER_VER_CHROMA_W8 8, 32, ss, 7
-
-FILTER_VER_CHROMA_W8 8, 2, sp, 8
-FILTER_VER_CHROMA_W8 8, 4, sp, 8
-FILTER_VER_CHROMA_W8 8, 6, sp, 8
-FILTER_VER_CHROMA_W8 8, 8, sp, 8
-FILTER_VER_CHROMA_W8 8, 16, sp, 8
-FILTER_VER_CHROMA_W8 8, 32, sp, 8
-
-FILTER_VER_CHROMA_W8 8, 2, ps, 8
-FILTER_VER_CHROMA_W8 8, 4, ps, 8
-FILTER_VER_CHROMA_W8 8, 6, ps, 8
-FILTER_VER_CHROMA_W8 8, 8, ps, 8
-FILTER_VER_CHROMA_W8 8, 16, ps, 8
-FILTER_VER_CHROMA_W8 8, 32, ps, 8
-
-FILTER_VER_CHROMA_W8 8, 2, pp, 8
-FILTER_VER_CHROMA_W8 8, 4, pp, 8
-FILTER_VER_CHROMA_W8 8, 6, pp, 8
-FILTER_VER_CHROMA_W8 8, 8, pp, 8
-FILTER_VER_CHROMA_W8 8, 16, pp, 8
-FILTER_VER_CHROMA_W8 8, 32, pp, 8
-
-FILTER_VER_CHROMA_W8 8, 12, ss, 7
-FILTER_VER_CHROMA_W8 8, 64, ss, 7
-FILTER_VER_CHROMA_W8 8, 12, sp, 8
-FILTER_VER_CHROMA_W8 8, 64, sp, 8
-FILTER_VER_CHROMA_W8 8, 12, ps, 8
-FILTER_VER_CHROMA_W8 8, 64, ps, 8
-FILTER_VER_CHROMA_W8 8, 12, pp, 8
-FILTER_VER_CHROMA_W8 8, 64, pp, 8
-
-%macro PROCESS_CHROMA_VERT_W16_2R 0
-    movu       m1, [r0]
-    movu       m3, [r0 + r1]
-    punpcklwd  m0, m1, m3
-    pmaddwd    m0, [r5 + 0 * 32]
-    punpckhwd  m1, m3
-    pmaddwd    m1, [r5 + 0 * 32]
-
-    movu       m4, [r0 + 2 * r1]
-    punpcklwd  m2, m3, m4
-    pmaddwd    m2, [r5 + 0 * 32]
-    punpckhwd  m3, m4
-    pmaddwd    m3, [r5 + 0 * 32]
-
-    lea        r0, [r0 + 2 * r1]
-    movu       m5, [r0 + r1]
-    punpcklwd  m6, m4, m5
-    pmaddwd    m6, [r5 + 1 * 32]
-    paddd      m0, m6
-    punpckhwd  m4, m5
-    pmaddwd    m4, [r5 + 1 * 32]
-    paddd      m1, m4
-
-    movu       m4, [r0 + 2 * r1]
-    punpcklwd  m6, m5, m4
-    pmaddwd    m6, [r5 + 1 * 32]
-    paddd      m2, m6
-    punpckhwd  m5, m4
-    pmaddwd    m5, [r5 + 1 * 32]
-    paddd      m3, m5
-%endmacro
+FILTER_VER_LUMA_AVX2_4x4 pp
+FILTER_VER_LUMA_AVX2_4x4 ps
+FILTER_VER_LUMA_AVX2_4x4 sp
+FILTER_VER_LUMA_AVX2_4x4 ss
 
-;-----------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_AVX2_6xN 2
+%macro FILTER_VER_LUMA_AVX2_8x8 1
 INIT_YMM avx2
-%if ARCH_X86_64
-cglobal interp_4tap_vert_%2_6x%1, 4, 7, 10
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_8x8, 4, 6, 12
     mov             r4d, r4m
     add             r1d, r1d
     add             r3d, r3d
-    shl             r4d, 6
+    shl             r4d, 7
 
 %ifdef PIC
-    lea             r5, [tab_ChromaCoeffV]
+    lea             r5, [tab_LumaCoeffVer]
     add             r5, r4
 %else
-    lea             r5, [tab_ChromaCoeffV + r4]
+    lea             r5, [tab_LumaCoeffVer + r4]
 %endif
 
-    sub             r0, r1
-    mov             r6d, %1/4
+    lea             r4, [r1 * 3]
+    sub             r0, r4
 
-%ifidn %2,pp
-    vbroadcasti128  m8, [INTERP_OFFSET_PP]
-%elifidn %2, sp
-    vbroadcasti128  m8, [INTERP_OFFSET_SP]
+%ifidn %1,pp
+    vbroadcasti128  m11, [pd_32]
+%elifidn %1, sp
+    vbroadcasti128  m11, [INTERP_OFFSET_SP]
 %else
-    vbroadcasti128  m8, [INTERP_OFFSET_PS]
+    vbroadcasti128  m11, [INTERP_OFFSET_PS]
 %endif
 
-.loopH:
-    movu            xm0, [r0]
-    movu            xm1, [r0 + r1]
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
     punpckhwd       xm2, xm0, xm1
     punpcklwd       xm0, xm1
     vinserti128     m0, m0, xm2, 1
     pmaddwd         m0, [r5]
-
-    movu            xm2, [r0 + r1 * 2]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
     punpckhwd       xm3, xm1, xm2
     punpcklwd       xm1, xm2
     vinserti128     m1, m1, xm3, 1
     pmaddwd         m1, [r5]
-
-    lea             r4, [r1 * 3]
-    movu            xm3, [r0 + r4]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
     punpckhwd       xm4, xm2, xm3
     punpcklwd       xm2, xm3
     vinserti128     m2, m2, xm4, 1
     pmaddwd         m4, m2, [r5 + 1 * mmsize]
     pmaddwd         m2, [r5]
     paddd           m0, m4
-
     lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]
+    movu            xm4, [r0]                       ; m4 = row 4
     punpckhwd       xm5, xm3, xm4
     punpcklwd       xm3, xm4
     vinserti128     m3, m3, xm5, 1
     pmaddwd         m5, m3, [r5 + 1 * mmsize]
     pmaddwd         m3, [r5]
     paddd           m1, m5
-
-    movu            xm5, [r0 + r1]
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
     punpckhwd       xm6, xm4, xm5
     punpcklwd       xm4, xm5
     vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 2 * mmsize]
+    paddd           m0, m6
     pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    pmaddwd         m4, [r5]
     paddd           m2, m6
-
-    movu            xm6, [r0 + r1 * 2]
+    pmaddwd         m4, [r5]
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
     punpckhwd       xm7, xm5, xm6
     punpcklwd       xm5, xm6
     vinserti128     m5, m5, xm7, 1
+    pmaddwd         m7, m5, [r5 + 2 * mmsize]
+    paddd           m1, m7
     pmaddwd         m7, m5, [r5 + 1 * mmsize]
     pmaddwd         m5, [r5]
     paddd           m3, m7
+    movu            xm7, [r0 + r4]                  ; m7 = row 7
+    punpckhwd       xm8, xm6, xm7
+    punpcklwd       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddwd         m8, m6, [r5 + 3 * mmsize]
+    paddd           m0, m8
+    pmaddwd         m8, m6, [r5 + 2 * mmsize]
+    paddd           m2, m8
+    pmaddwd         m8, m6, [r5 + 1 * mmsize]
+    pmaddwd         m6, [r5]
+    paddd           m4, m8
+    lea             r0, [r0 + r1 * 4]
+    movu            xm8, [r0]                       ; m8 = row 8
+    punpckhwd       xm9, xm7, xm8
+    punpcklwd       xm7, xm8
+    vinserti128     m7, m7, xm9, 1
+    pmaddwd         m9, m7, [r5 + 3 * mmsize]
+    paddd           m1, m9
+    pmaddwd         m9, m7, [r5 + 2 * mmsize]
+    paddd           m3, m9
+    pmaddwd         m9, m7, [r5 + 1 * mmsize]
+    pmaddwd         m7, [r5]
+    paddd           m5, m9
+    movu            xm9, [r0 + r1]                  ; m9 = row 9
+    punpckhwd       xm10, xm8, xm9
+    punpcklwd       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddwd         m10, m8, [r5 + 3 * mmsize]
+    paddd           m2, m10
+    pmaddwd         m10, m8, [r5 + 2 * mmsize]
+    pmaddwd         m8, [r5 + 1 * mmsize]
+    paddd           m4, m10
+    paddd           m6, m8
+    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
+    punpckhwd       xm8, xm9, xm10
+    punpcklwd       xm9, xm10
+    vinserti128     m9, m9, xm8, 1
+    pmaddwd         m8, m9, [r5 + 3 * mmsize]
+    paddd           m3, m8
+    pmaddwd         m8, m9, [r5 + 2 * mmsize]
+    pmaddwd         m9, [r5 + 1 * mmsize]
+    paddd           m5, m8
+    paddd           m7, m9
+    movu            xm8, [r0 + r4]                  ; m8 = row 11
+    punpckhwd       xm9, xm10, xm8
+    punpcklwd       xm10, xm8
+    vinserti128     m10, m10, xm9, 1
+    pmaddwd         m9, m10, [r5 + 3 * mmsize]
+    pmaddwd         m10, [r5 + 2 * mmsize]
+    paddd           m4, m9
+    paddd           m6, m10
+
     lea             r4, [r3 * 3]
-%ifidn %2,ss
+%ifidn %1,ss
     psrad           m0, 6
     psrad           m1, 6
     psrad           m2, 6
     psrad           m3, 6
 %else
-    paddd           m0, m8
-    paddd           m1, m8
-    paddd           m2, m8
-    paddd           m3, m8
-%ifidn %2,pp
+    paddd           m0, m11
+    paddd           m1, m11
+    paddd           m2, m11
+    paddd           m3, m11
+%ifidn %1,pp
     psrad           m0, INTERP_SHIFT_PP
     psrad           m1, INTERP_SHIFT_PP
     psrad           m2, INTERP_SHIFT_PP
     psrad           m3, INTERP_SHIFT_PP
-%elifidn %2, sp
+%elifidn %1, sp
     psrad           m0, INTERP_SHIFT_SP
     psrad           m1, INTERP_SHIFT_SP
     psrad           m2, INTERP_SHIFT_SP
@@ -1267,1108 +724,633 @@ cglobal interp_4tap_vert_%2_6x%1, 4, 7, 10
 
     packssdw        m0, m1
     packssdw        m2, m3
-    vpermq          m0, m0, q3120
-    vpermq          m2, m2, q3120
-    pxor            m5, m5
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    pxor            m10, m10
     mova            m9, [pw_pixel_max]
-%ifidn %2,pp
-    CLIPW           m0, m5, m9
-    CLIPW           m2, m5, m9
-%elifidn %2, sp
-    CLIPW           m0, m5, m9
-    CLIPW           m2, m5, m9
+%ifidn %1,pp
+    CLIPW           m0, m10, m9
+    CLIPW           m2, m10, m9
+%elifidn %1, sp
+    CLIPW           m0, m10, m9
+    CLIPW           m2, m10, m9
 %endif
 
     vextracti128    xm1, m0, 1
     vextracti128    xm3, m2, 1
-    movq            [r2], xm0
-    pextrd          [r2 + 8], xm0, 2
-    movq            [r2 + r3], xm1
-    pextrd          [r2 + r3 + 8], xm1, 2
-    movq            [r2 + r3 * 2], xm2
-    pextrd          [r2 + r3 * 2 + 8], xm2, 2
-    movq            [r2 + r4], xm3
-    pextrd          [r2 + r4 + 8], xm3, 2
-
-    lea             r2, [r2 + r3 * 4]
-    dec r6d
-    jnz .loopH
-    RET
-%endif
-%endmacro
-FILTER_VER_CHROMA_AVX2_6xN 8, pp
-FILTER_VER_CHROMA_AVX2_6xN 8, ps
-FILTER_VER_CHROMA_AVX2_6xN 8, ss
-FILTER_VER_CHROMA_AVX2_6xN 8, sp
-FILTER_VER_CHROMA_AVX2_6xN 16, pp
-FILTER_VER_CHROMA_AVX2_6xN 16, ps
-FILTER_VER_CHROMA_AVX2_6xN 16, ss
-FILTER_VER_CHROMA_AVX2_6xN 16, sp
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r4], xm3
 
-;-----------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_W16_16xN_avx2 3
-INIT_YMM avx2
-cglobal interp_4tap_vert_%2_16x%1, 5, 6, %3
-    add       r1d, r1d
-    add       r3d, r3d
-    sub       r0, r1
-    shl       r4d, 6
+    lea             r0, [r0 + r1 * 4]
+    movu            xm2, [r0]                       ; m2 = row 12
+    punpckhwd       xm3, xm8, xm2
+    punpcklwd       xm8, xm2
+    vinserti128     m8, m8, xm3, 1
+    pmaddwd         m3, m8, [r5 + 3 * mmsize]
+    pmaddwd         m8, [r5 + 2 * mmsize]
+    paddd           m5, m3
+    paddd           m7, m8
+    movu            xm3, [r0 + r1]                  ; m3 = row 13
+    punpckhwd       xm0, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm0, 1
+    pmaddwd         m2, [r5 + 3 * mmsize]
+    paddd           m6, m2
+    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
+    punpckhwd       xm1, xm3, xm0
+    punpcklwd       xm3, xm0
+    vinserti128     m3, m3, xm1, 1
+    pmaddwd         m3, [r5 + 3 * mmsize]
+    paddd           m7, m3
 
-%ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
-    lea       r5, [r5 + r4]
+%ifidn %1,ss
+    psrad           m4, 6
+    psrad           m5, 6
+    psrad           m6, 6
+    psrad           m7, 6
+%else
+    paddd           m4, m11
+    paddd           m5, m11
+    paddd           m6, m11
+    paddd           m7, m11
+%ifidn %1,pp
+    psrad           m4, INTERP_SHIFT_PP
+    psrad           m5, INTERP_SHIFT_PP
+    psrad           m6, INTERP_SHIFT_PP
+    psrad           m7, INTERP_SHIFT_PP
+%elifidn %1, sp
+    psrad           m4, INTERP_SHIFT_SP
+    psrad           m5, INTERP_SHIFT_SP
+    psrad           m6, INTERP_SHIFT_SP
+    psrad           m7, INTERP_SHIFT_SP
 %else
-    lea       r5, [tab_ChromaCoeffV + r4]
+    psrad           m4, INTERP_SHIFT_PS
+    psrad           m5, INTERP_SHIFT_PS
+    psrad           m6, INTERP_SHIFT_PS
+    psrad           m7, INTERP_SHIFT_PS
+%endif
 %endif
 
-    mov       r4d, %1/2
-
-%ifidn %2, pp
-    vbroadcasti128  m7, [INTERP_OFFSET_PP]
-%elifidn %2, sp
-    vbroadcasti128  m7, [INTERP_OFFSET_SP]
-%elifidn %2, ps
-    vbroadcasti128  m7, [INTERP_OFFSET_PS]
+    packssdw        m4, m5
+    packssdw        m6, m7
+    vpermq          m4, m4, 11011000b
+    vpermq          m6, m6, 11011000b
+%ifidn %1,pp
+    CLIPW           m4, m10, m9
+    CLIPW           m6, m10, m9
+%elifidn %1, sp
+    CLIPW           m4, m10, m9
+    CLIPW           m6, m10, m9
+%endif
+    vextracti128    xm5, m4, 1
+    vextracti128    xm7, m6, 1
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm4
+    movu            [r2 + r3], xm5
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r4], xm7
+    RET
 %endif
+%endmacro
 
-.loopH:
-    PROCESS_CHROMA_VERT_W16_2R
-%ifidn %2, ss
-    psrad     m0, 6
-    psrad     m1, 6
-    psrad     m2, 6
-    psrad     m3, 6
+FILTER_VER_LUMA_AVX2_8x8 pp
+FILTER_VER_LUMA_AVX2_8x8 ps
+FILTER_VER_LUMA_AVX2_8x8 sp
+FILTER_VER_LUMA_AVX2_8x8 ss
 
-    packssdw  m0, m1
-    packssdw  m2, m3
-%elifidn %2, ps
-    paddd     m0, m7
-    paddd     m1, m7
-    paddd     m2, m7
-    paddd     m3, m7
-    psrad     m0, INTERP_SHIFT_PS
-    psrad     m1, INTERP_SHIFT_PS
-    psrad     m2, INTERP_SHIFT_PS
-    psrad     m3, INTERP_SHIFT_PS
+%macro PROCESS_LUMA_AVX2_W8_16R 1
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m2, [r5]
+    lea             r7, [r0 + r1 * 4]
+    movu            xm4, [r7]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    paddd           m1, m5
+    pmaddwd         m3, [r5]
+    movu            xm5, [r7 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 2 * mmsize]
+    paddd           m0, m6
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m2, m6
+    pmaddwd         m4, [r5]
+    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm7, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddwd         m7, m5, [r5 + 2 * mmsize]
+    paddd           m1, m7
+    pmaddwd         m7, m5, [r5 + 1 * mmsize]
+    paddd           m3, m7
+    pmaddwd         m5, [r5]
+    movu            xm7, [r7 + r4]                  ; m7 = row 7
+    punpckhwd       xm8, xm6, xm7
+    punpcklwd       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddwd         m8, m6, [r5 + 3 * mmsize]
+    paddd           m0, m8
+    pmaddwd         m8, m6, [r5 + 2 * mmsize]
+    paddd           m2, m8
+    pmaddwd         m8, m6, [r5 + 1 * mmsize]
+    paddd           m4, m8
+    pmaddwd         m6, [r5]
+    lea             r7, [r7 + r1 * 4]
+    movu            xm8, [r7]                       ; m8 = row 8
+    punpckhwd       xm9, xm7, xm8
+    punpcklwd       xm7, xm8
+    vinserti128     m7, m7, xm9, 1
+    pmaddwd         m9, m7, [r5 + 3 * mmsize]
+    paddd           m1, m9
+    pmaddwd         m9, m7, [r5 + 2 * mmsize]
+    paddd           m3, m9
+    pmaddwd         m9, m7, [r5 + 1 * mmsize]
+    paddd           m5, m9
+    pmaddwd         m7, [r5]
+    movu            xm9, [r7 + r1]                  ; m9 = row 9
+    punpckhwd       xm10, xm8, xm9
+    punpcklwd       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddwd         m10, m8, [r5 + 3 * mmsize]
+    paddd           m2, m10
+    pmaddwd         m10, m8, [r5 + 2 * mmsize]
+    paddd           m4, m10
+    pmaddwd         m10, m8, [r5 + 1 * mmsize]
+    paddd           m6, m10
+    pmaddwd         m8, [r5]
+    movu            xm10, [r7 + r1 * 2]             ; m10 = row 10
+    punpckhwd       xm11, xm9, xm10
+    punpcklwd       xm9, xm10
+    vinserti128     m9, m9, xm11, 1
+    pmaddwd         m11, m9, [r5 + 3 * mmsize]
+    paddd           m3, m11
+    pmaddwd         m11, m9, [r5 + 2 * mmsize]
+    paddd           m5, m11
+    pmaddwd         m11, m9, [r5 + 1 * mmsize]
+    paddd           m7, m11
+    pmaddwd         m9, [r5]
+    movu            xm11, [r7 + r4]                 ; m11 = row 11
+    punpckhwd       xm12, xm10, xm11
+    punpcklwd       xm10, xm11
+    vinserti128     m10, m10, xm12, 1
+    pmaddwd         m12, m10, [r5 + 3 * mmsize]
+    paddd           m4, m12
+    pmaddwd         m12, m10, [r5 + 2 * mmsize]
+    paddd           m6, m12
+    pmaddwd         m12, m10, [r5 + 1 * mmsize]
+    paddd           m8, m12
+    pmaddwd         m10, [r5]
+    lea             r7, [r7 + r1 * 4]
+    movu            xm12, [r7]                      ; m12 = row 12
+    punpckhwd       xm13, xm11, xm12
+    punpcklwd       xm11, xm12
+    vinserti128     m11, m11, xm13, 1
+    pmaddwd         m13, m11, [r5 + 3 * mmsize]
+    paddd           m5, m13
+    pmaddwd         m13, m11, [r5 + 2 * mmsize]
+    paddd           m7, m13
+    pmaddwd         m13, m11, [r5 + 1 * mmsize]
+    paddd           m9, m13
+    pmaddwd         m11, [r5]
 
-    packssdw  m0, m1
-    packssdw  m2, m3
+%ifidn %1,ss
+    psrad           m0, 6
+    psrad           m1, 6
+    psrad           m2, 6
+    psrad           m3, 6
+    psrad           m4, 6
+    psrad           m5, 6
 %else
-    paddd     m0, m7
-    paddd     m1, m7
-    paddd     m2, m7
-    paddd     m3, m7
- %ifidn %2, pp
-    psrad     m0, INTERP_SHIFT_PP
-    psrad     m1, INTERP_SHIFT_PP
-    psrad     m2, INTERP_SHIFT_PP
-    psrad     m3, INTERP_SHIFT_PP
+    paddd           m0, m14
+    paddd           m1, m14
+    paddd           m2, m14
+    paddd           m3, m14
+    paddd           m4, m14
+    paddd           m5, m14
+%ifidn %1,pp
+    psrad           m0, INTERP_SHIFT_PP
+    psrad           m1, INTERP_SHIFT_PP
+    psrad           m2, INTERP_SHIFT_PP
+    psrad           m3, INTERP_SHIFT_PP
+    psrad           m4, INTERP_SHIFT_PP
+    psrad           m5, INTERP_SHIFT_PP
+%elifidn %1, sp
+    psrad           m0, INTERP_SHIFT_SP
+    psrad           m1, INTERP_SHIFT_SP
+    psrad           m2, INTERP_SHIFT_SP
+    psrad           m3, INTERP_SHIFT_SP
+    psrad           m4, INTERP_SHIFT_SP
+    psrad           m5, INTERP_SHIFT_SP
 %else
-    psrad     m0, INTERP_SHIFT_SP
-    psrad     m1, INTERP_SHIFT_SP
-    psrad     m2, INTERP_SHIFT_SP
-    psrad     m3, INTERP_SHIFT_SP
+    psrad           m0, INTERP_SHIFT_PS
+    psrad           m1, INTERP_SHIFT_PS
+    psrad           m2, INTERP_SHIFT_PS
+    psrad           m3, INTERP_SHIFT_PS
+    psrad           m4, INTERP_SHIFT_PS
+    psrad           m5, INTERP_SHIFT_PS
 %endif
-    packssdw  m0, m1
-    packssdw  m2, m3
-    pxor      m5, m5
-    CLIPW2    m0, m2, m5, [pw_pixel_max]
 %endif
 
-    movu      [r2], m0
-    movu      [r2 + r3], m2
-    lea       r2, [r2 + 2 * r3]
-    dec       r4d
-    jnz       .loopH
-    RET
-%endmacro
-    FILTER_VER_CHROMA_W16_16xN_avx2 4, pp, 8
-    FILTER_VER_CHROMA_W16_16xN_avx2 8, pp, 8
-    FILTER_VER_CHROMA_W16_16xN_avx2 12, pp, 8
-    FILTER_VER_CHROMA_W16_16xN_avx2 24, pp, 8
-    FILTER_VER_CHROMA_W16_16xN_avx2 16, pp, 8
-    FILTER_VER_CHROMA_W16_16xN_avx2 32, pp, 8
-    FILTER_VER_CHROMA_W16_16xN_avx2 64, pp, 8
-
-    FILTER_VER_CHROMA_W16_16xN_avx2 4, ps, 8
-    FILTER_VER_CHROMA_W16_16xN_avx2 8, ps, 8
-    FILTER_VER_CHROMA_W16_16xN_avx2 12, ps, 8
-    FILTER_VER_CHROMA_W16_16xN_avx2 24, ps, 8
-    FILTER_VER_CHROMA_W16_16xN_avx2 16, ps, 8
-    FILTER_VER_CHROMA_W16_16xN_avx2 32, ps, 8
-    FILTER_VER_CHROMA_W16_16xN_avx2 64, ps, 8
-
-    FILTER_VER_CHROMA_W16_16xN_avx2 4, ss, 7
-    FILTER_VER_CHROMA_W16_16xN_avx2 8, ss, 7
-    FILTER_VER_CHROMA_W16_16xN_avx2 12, ss, 7
-    FILTER_VER_CHROMA_W16_16xN_avx2 24, ss, 7
-    FILTER_VER_CHROMA_W16_16xN_avx2 16, ss, 7
-    FILTER_VER_CHROMA_W16_16xN_avx2 32, ss, 7
-    FILTER_VER_CHROMA_W16_16xN_avx2 64, ss, 7
-
-    FILTER_VER_CHROMA_W16_16xN_avx2 4, sp, 8
-    FILTER_VER_CHROMA_W16_16xN_avx2 8, sp, 8
-    FILTER_VER_CHROMA_W16_16xN_avx2 12, sp, 8
-    FILTER_VER_CHROMA_W16_16xN_avx2 24, sp, 8
-    FILTER_VER_CHROMA_W16_16xN_avx2 16, sp, 8
-    FILTER_VER_CHROMA_W16_16xN_avx2 32, sp, 8
-    FILTER_VER_CHROMA_W16_16xN_avx2 64, sp, 8
-
-%macro PROCESS_CHROMA_VERT_W32_2R 0
-    movu       m1, [r0]
-    movu       m3, [r0 + r1]
-    punpcklwd  m0, m1, m3
-    pmaddwd    m0, [r5 + 0 * mmsize]
-    punpckhwd  m1, m3
-    pmaddwd    m1, [r5 + 0 * mmsize]
-
-    movu       m9, [r0 + mmsize]
-    movu       m11, [r0 + r1 + mmsize]
-    punpcklwd  m8, m9, m11
-    pmaddwd    m8, [r5 + 0 * mmsize]
-    punpckhwd  m9, m11
-    pmaddwd    m9, [r5 + 0 * mmsize]
-
-    movu       m4, [r0 + 2 * r1]
-    punpcklwd  m2, m3, m4
-    pmaddwd    m2, [r5 + 0 * mmsize]
-    punpckhwd  m3, m4
-    pmaddwd    m3, [r5 + 0 * mmsize]
-
-    movu       m12, [r0 + 2 * r1 + mmsize]
-    punpcklwd  m10, m11, m12
-    pmaddwd    m10, [r5 + 0 * mmsize]
-    punpckhwd  m11, m12
-    pmaddwd    m11, [r5 + 0 * mmsize]
-
-    lea        r6, [r0 + 2 * r1]
-    movu       m5, [r6 + r1]
-    punpcklwd  m6, m4, m5
-    pmaddwd    m6, [r5 + 1 * mmsize]
-    paddd      m0, m6
-    punpckhwd  m4, m5
-    pmaddwd    m4, [r5 + 1 * mmsize]
-    paddd      m1, m4
-
-    movu       m13, [r6 + r1 + mmsize]
-    punpcklwd  m14, m12, m13
-    pmaddwd    m14, [r5 + 1 * mmsize]
-    paddd      m8, m14
-    punpckhwd  m12, m13
-    pmaddwd    m12, [r5 + 1 * mmsize]
-    paddd      m9, m12
-
-    movu       m4, [r6 + 2 * r1]
-    punpcklwd  m6, m5, m4
-    pmaddwd    m6, [r5 + 1 * mmsize]
-    paddd      m2, m6
-    punpckhwd  m5, m4
-    pmaddwd    m5, [r5 + 1 * mmsize]
-    paddd      m3, m5
-
-    movu       m12, [r6 + 2 * r1 + mmsize]
-    punpcklwd  m14, m13, m12
-    pmaddwd    m14, [r5 + 1 * mmsize]
-    paddd      m10, m14
-    punpckhwd  m13, m12
-    pmaddwd    m13, [r5 + 1 * mmsize]
-    paddd      m11, m13
-%endmacro
-
-;-----------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_W16_32xN_avx2 3
-INIT_YMM avx2
-%if ARCH_X86_64
-cglobal interp_4tap_vert_%2_32x%1, 5, 7, %3
-    add       r1d, r1d
-    add       r3d, r3d
-    sub       r0, r1
-    shl       r4d, 6
-
-%ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
-    lea       r5, [r5 + r4]
-%else
-    lea       r5, [tab_ChromaCoeffV + r4]
-%endif
-    mov       r4d, %1/2
-
-%ifidn %2, pp
-    vbroadcasti128  m7, [INTERP_OFFSET_PP]
-%elifidn %2, sp
-    vbroadcasti128  m7, [INTERP_OFFSET_SP]
-%elifidn %2, ps
-    vbroadcasti128  m7, [INTERP_OFFSET_PS]
+    packssdw        m0, m1
+    packssdw        m2, m3
+    packssdw        m4, m5
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    vpermq          m4, m4, 11011000b
+    pxor            m5, m5
+    mova            m3, [pw_pixel_max]
+%ifidn %1,pp
+    CLIPW           m0, m5, m3
+    CLIPW           m2, m5, m3
+    CLIPW           m4, m5, m3
+%elifidn %1, sp
+    CLIPW           m0, m5, m3
+    CLIPW           m2, m5, m3
+    CLIPW           m4, m5, m3
 %endif
 
-.loopH:
-    PROCESS_CHROMA_VERT_W32_2R
-%ifidn %2, ss
-    psrad     m0, 6
-    psrad     m1, 6
-    psrad     m2, 6
-    psrad     m3, 6
-
-    psrad     m8, 6
-    psrad     m9, 6
-    psrad     m10, 6
-    psrad     m11, 6
+    vextracti128    xm1, m0, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    vextracti128    xm1, m2, 1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm1
+    lea             r8, [r2 + r3 * 4]
+    vextracti128    xm1, m4, 1
+    movu            [r8], xm4
+    movu            [r8 + r3], xm1
 
-    packssdw  m0, m1
-    packssdw  m2, m3
-    packssdw  m8, m9
-    packssdw  m10, m11
-%elifidn %2, ps
-    paddd     m0, m7
-    paddd     m1, m7
-    paddd     m2, m7
-    paddd     m3, m7
-    psrad     m0, INTERP_SHIFT_PS
-    psrad     m1, INTERP_SHIFT_PS
-    psrad     m2, INTERP_SHIFT_PS
-    psrad     m3, INTERP_SHIFT_PS
-    paddd     m8, m7
-    paddd     m9, m7
-    paddd     m10, m7
-    paddd     m11, m7
-    psrad     m8, INTERP_SHIFT_PS
-    psrad     m9, INTERP_SHIFT_PS
-    psrad     m10, INTERP_SHIFT_PS
-    psrad     m11, INTERP_SHIFT_PS
+    movu            xm13, [r7 + r1]                 ; m13 = row 13
+    punpckhwd       xm0, xm12, xm13
+    punpcklwd       xm12, xm13
+    vinserti128     m12, m12, xm0, 1
+    pmaddwd         m0, m12, [r5 + 3 * mmsize]
+    paddd           m6, m0
+    pmaddwd         m0, m12, [r5 + 2 * mmsize]
+    paddd           m8, m0
+    pmaddwd         m0, m12, [r5 + 1 * mmsize]
+    paddd           m10, m0
+    pmaddwd         m12, [r5]
+    movu            xm0, [r7 + r1 * 2]              ; m0 = row 14
+    punpckhwd       xm1, xm13, xm0
+    punpcklwd       xm13, xm0
+    vinserti128     m13, m13, xm1, 1
+    pmaddwd         m1, m13, [r5 + 3 * mmsize]
+    paddd           m7, m1
+    pmaddwd         m1, m13, [r5 + 2 * mmsize]
+    paddd           m9, m1
+    pmaddwd         m1, m13, [r5 + 1 * mmsize]
+    paddd           m11, m1
+    pmaddwd         m13, [r5]
 
-    packssdw  m0, m1
-    packssdw  m2, m3
-    packssdw  m8, m9
-    packssdw  m10, m11
+%ifidn %1,ss
+    psrad           m6, 6
+    psrad           m7, 6
 %else
-    paddd     m0, m7
-    paddd     m1, m7
-    paddd     m2, m7
-    paddd     m3, m7
-    paddd     m8, m7
-    paddd     m9, m7
-    paddd     m10, m7
-    paddd     m11, m7
- %ifidn %2, pp
-    psrad     m0, INTERP_SHIFT_PP
-    psrad     m1, INTERP_SHIFT_PP
-    psrad     m2, INTERP_SHIFT_PP
-    psrad     m3, INTERP_SHIFT_PP
-    psrad     m8, INTERP_SHIFT_PP
-    psrad     m9, INTERP_SHIFT_PP
-    psrad     m10, INTERP_SHIFT_PP
-    psrad     m11, INTERP_SHIFT_PP
+    paddd           m6, m14
+    paddd           m7, m14
+%ifidn %1,pp
+    psrad           m6, INTERP_SHIFT_PP
+    psrad           m7, INTERP_SHIFT_PP
+%elifidn %1, sp
+    psrad           m6, INTERP_SHIFT_SP
+    psrad           m7, INTERP_SHIFT_SP
 %else
-    psrad     m0, INTERP_SHIFT_SP
-    psrad     m1, INTERP_SHIFT_SP
-    psrad     m2, INTERP_SHIFT_SP
-    psrad     m3, INTERP_SHIFT_SP
-    psrad     m8, INTERP_SHIFT_SP
-    psrad     m9, INTERP_SHIFT_SP
-    psrad     m10, INTERP_SHIFT_SP
-    psrad     m11, INTERP_SHIFT_SP
+    psrad           m6, INTERP_SHIFT_PS
+    psrad           m7, INTERP_SHIFT_PS
 %endif
-    packssdw  m0, m1
-    packssdw  m2, m3
-    packssdw  m8, m9
-    packssdw  m10, m11
-    pxor      m5, m5
-    CLIPW2    m0, m2, m5, [pw_pixel_max]
-    CLIPW2    m8, m10, m5, [pw_pixel_max]
 %endif
 
-    movu      [r2], m0
-    movu      [r2 + r3], m2
-    movu      [r2 + mmsize], m8
-    movu      [r2 + r3 + mmsize], m10
-    lea       r2, [r2 + 2 * r3]
-    lea       r0, [r0 + 2 * r1]
-    dec       r4d
-    jnz       .loopH
-    RET
+    packssdw        m6, m7
+    vpermq          m6, m6, 11011000b
+%ifidn %1,pp
+    CLIPW           m6, m5, m3
+%elifidn %1, sp
+    CLIPW           m6, m5, m3
 %endif
-%endmacro
-    FILTER_VER_CHROMA_W16_32xN_avx2 8, pp, 15
-    FILTER_VER_CHROMA_W16_32xN_avx2 16, pp, 15
-    FILTER_VER_CHROMA_W16_32xN_avx2 24, pp, 15
-    FILTER_VER_CHROMA_W16_32xN_avx2 32, pp, 15
-    FILTER_VER_CHROMA_W16_32xN_avx2 48, pp, 15
-    FILTER_VER_CHROMA_W16_32xN_avx2 64, pp, 15
-
-    FILTER_VER_CHROMA_W16_32xN_avx2 8, ps, 15
-    FILTER_VER_CHROMA_W16_32xN_avx2 16, ps, 15
-    FILTER_VER_CHROMA_W16_32xN_avx2 24, ps, 15
-    FILTER_VER_CHROMA_W16_32xN_avx2 32, ps, 15
-    FILTER_VER_CHROMA_W16_32xN_avx2 48, ps, 15
-    FILTER_VER_CHROMA_W16_32xN_avx2 64, ps, 15
-
-    FILTER_VER_CHROMA_W16_32xN_avx2 8, ss, 15
-    FILTER_VER_CHROMA_W16_32xN_avx2 16, ss, 15
-    FILTER_VER_CHROMA_W16_32xN_avx2 24, ss, 15
-    FILTER_VER_CHROMA_W16_32xN_avx2 32, ss, 15
-    FILTER_VER_CHROMA_W16_32xN_avx2 48, ss, 15
-    FILTER_VER_CHROMA_W16_32xN_avx2 64, ss, 15
-
-    FILTER_VER_CHROMA_W16_32xN_avx2 8, sp, 15
-    FILTER_VER_CHROMA_W16_32xN_avx2 16, sp, 15
-    FILTER_VER_CHROMA_W16_32xN_avx2 24, sp, 15
-    FILTER_VER_CHROMA_W16_32xN_avx2 32, sp, 15
-    FILTER_VER_CHROMA_W16_32xN_avx2 48, sp, 15
-    FILTER_VER_CHROMA_W16_32xN_avx2 64, sp, 15
+    vextracti128    xm7, m6, 1
+    movu            [r8 + r3 * 2], xm6
+    movu            [r8 + r6], xm7
 
-;-----------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_W16_64xN_avx2 3
-INIT_YMM avx2
-cglobal interp_4tap_vert_%2_64x%1, 5, 7, %3
-    add       r1d, r1d
-    add       r3d, r3d
-    sub       r0, r1
-    shl       r4d, 6
+    movu            xm1, [r7 + r4]                  ; m1 = row 15
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m2, m0, [r5 + 3 * mmsize]
+    paddd           m8, m2
+    pmaddwd         m2, m0, [r5 + 2 * mmsize]
+    paddd           m10, m2
+    pmaddwd         m2, m0, [r5 + 1 * mmsize]
+    paddd           m12, m2
+    pmaddwd         m0, [r5]
+    lea             r7, [r7 + r1 * 4]
+    movu            xm2, [r7]                       ; m2 = row 16
+    punpckhwd       xm6, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm6, 1
+    pmaddwd         m6, m1, [r5 + 3 * mmsize]
+    paddd           m9, m6
+    pmaddwd         m6, m1, [r5 + 2 * mmsize]
+    paddd           m11, m6
+    pmaddwd         m6, m1, [r5 + 1 * mmsize]
+    paddd           m13, m6
+    pmaddwd         m1, [r5]
+    movu            xm6, [r7 + r1]                  ; m6 = row 17
+    punpckhwd       xm4, xm2, xm6
+    punpcklwd       xm2, xm6
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 3 * mmsize]
+    paddd           m10, m4
+    pmaddwd         m4, m2, [r5 + 2 * mmsize]
+    paddd           m12, m4
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m0, m2
+    movu            xm4, [r7 + r1 * 2]              ; m4 = row 18
+    punpckhwd       xm2, xm6, xm4
+    punpcklwd       xm6, xm4
+    vinserti128     m6, m6, xm2, 1
+    pmaddwd         m2, m6, [r5 + 3 * mmsize]
+    paddd           m11, m2
+    pmaddwd         m2, m6, [r5 + 2 * mmsize]
+    paddd           m13, m2
+    pmaddwd         m6, [r5 + 1 * mmsize]
+    paddd           m1, m6
+    movu            xm2, [r7 + r4]                  ; m2 = row 19
+    punpckhwd       xm6, xm4, xm2
+    punpcklwd       xm4, xm2
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 3 * mmsize]
+    paddd           m12, m6
+    pmaddwd         m4, [r5 + 2 * mmsize]
+    paddd           m0, m4
+    lea             r7, [r7 + r1 * 4]
+    movu            xm6, [r7]                       ; m6 = row 20
+    punpckhwd       xm7, xm2, xm6
+    punpcklwd       xm2, xm6
+    vinserti128     m2, m2, xm7, 1
+    pmaddwd         m7, m2, [r5 + 3 * mmsize]
+    paddd           m13, m7
+    pmaddwd         m2, [r5 + 2 * mmsize]
+    paddd           m1, m2
+    movu            xm7, [r7 + r1]                  ; m7 = row 21
+    punpckhwd       xm2, xm6, xm7
+    punpcklwd       xm6, xm7
+    vinserti128     m6, m6, xm2, 1
+    pmaddwd         m6, [r5 + 3 * mmsize]
+    paddd           m0, m6
+    movu            xm2, [r7 + r1 * 2]              ; m2 = row 22
+    punpckhwd       xm6, xm7, xm2
+    punpcklwd       xm7, xm2
+    vinserti128     m7, m7, xm6, 1
+    pmaddwd         m7, [r5 + 3 * mmsize]
+    paddd           m1, m7
 
-%ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
-    lea       r5, [r5 + r4]
+%ifidn %1,ss
+    psrad           m8, 6
+    psrad           m9, 6
+    psrad           m10, 6
+    psrad           m11, 6
+    psrad           m12, 6
+    psrad           m13, 6
+    psrad           m0, 6
+    psrad           m1, 6
+%else
+    paddd           m8, m14
+    paddd           m9, m14
+    paddd           m10, m14
+    paddd           m11, m14
+    paddd           m12, m14
+    paddd           m13, m14
+    paddd           m0, m14
+    paddd           m1, m14
+%ifidn %1,pp
+    psrad           m8, INTERP_SHIFT_PP
+    psrad           m9, INTERP_SHIFT_PP
+    psrad           m10, INTERP_SHIFT_PP
+    psrad           m11, INTERP_SHIFT_PP
+    psrad           m12, INTERP_SHIFT_PP
+    psrad           m13, INTERP_SHIFT_PP
+    psrad           m0, INTERP_SHIFT_PP
+    psrad           m1, INTERP_SHIFT_PP
+%elifidn %1, sp
+    psrad           m8, INTERP_SHIFT_SP
+    psrad           m9, INTERP_SHIFT_SP
+    psrad           m10, INTERP_SHIFT_SP
+    psrad           m11, INTERP_SHIFT_SP
+    psrad           m12, INTERP_SHIFT_SP
+    psrad           m13, INTERP_SHIFT_SP
+    psrad           m0, INTERP_SHIFT_SP
+    psrad           m1, INTERP_SHIFT_SP
 %else
-    lea       r5, [tab_ChromaCoeffV + r4]
+    psrad           m8, INTERP_SHIFT_PS
+    psrad           m9, INTERP_SHIFT_PS
+    psrad           m10, INTERP_SHIFT_PS
+    psrad           m11, INTERP_SHIFT_PS
+    psrad           m12, INTERP_SHIFT_PS
+    psrad           m13, INTERP_SHIFT_PS
+    psrad           m0, INTERP_SHIFT_PS
+    psrad           m1, INTERP_SHIFT_PS
 %endif
-    mov       r4d, %1/2
-
-%ifidn %2, pp
-    vbroadcasti128  m7, [INTERP_OFFSET_PP]
-%elifidn %2, sp
-    vbroadcasti128  m7, [INTERP_OFFSET_SP]
-%elifidn %2, ps
-    vbroadcasti128  m7, [INTERP_OFFSET_PS]
 %endif
 
-.loopH:
-%assign x 0
-%rep 4
-    movu       m1, [r0 + x]
-    movu       m3, [r0 + r1 + x]
-    movu       m5, [r5 + 0 * mmsize]
-    punpcklwd  m0, m1, m3
-    pmaddwd    m0, m5
-    punpckhwd  m1, m3
-    pmaddwd    m1, m5
-
-    movu       m4, [r0 + 2 * r1 + x]
-    punpcklwd  m2, m3, m4
-    pmaddwd    m2, m5
-    punpckhwd  m3, m4
-    pmaddwd    m3, m5
-
-    lea        r6, [r0 + 2 * r1]
-    movu       m5, [r6 + r1 + x]
-    punpcklwd  m6, m4, m5
-    pmaddwd    m6, [r5 + 1 * mmsize]
-    paddd      m0, m6
-    punpckhwd  m4, m5
-    pmaddwd    m4, [r5 + 1 * mmsize]
-    paddd      m1, m4
-
-    movu       m4, [r6 + 2 * r1 + x]
-    punpcklwd  m6, m5, m4
-    pmaddwd    m6, [r5 + 1 * mmsize]
-    paddd      m2, m6
-    punpckhwd  m5, m4
-    pmaddwd    m5, [r5 + 1 * mmsize]
-    paddd      m3, m5
-
-%ifidn %2, ss
-    psrad     m0, 6
-    psrad     m1, 6
-    psrad     m2, 6
-    psrad     m3, 6
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-%elifidn %2, ps
-    paddd     m0, m7
-    paddd     m1, m7
-    paddd     m2, m7
-    paddd     m3, m7
-    psrad     m0, INTERP_SHIFT_PS
-    psrad     m1, INTERP_SHIFT_PS
-    psrad     m2, INTERP_SHIFT_PS
-    psrad     m3, INTERP_SHIFT_PS
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-%else
-    paddd     m0, m7
-    paddd     m1, m7
-    paddd     m2, m7
-    paddd     m3, m7
-%ifidn %2, pp
-    psrad     m0, INTERP_SHIFT_PP
-    psrad     m1, INTERP_SHIFT_PP
-    psrad     m2, INTERP_SHIFT_PP
-    psrad     m3, INTERP_SHIFT_PP
-%else
-    psrad     m0, INTERP_SHIFT_SP
-    psrad     m1, INTERP_SHIFT_SP
-    psrad     m2, INTERP_SHIFT_SP
-    psrad     m3, INTERP_SHIFT_SP
-%endif
-    packssdw  m0, m1
-    packssdw  m2, m3
-    pxor      m5, m5
-    CLIPW2    m0, m2, m5, [pw_pixel_max]
+    packssdw        m8, m9
+    packssdw        m10, m11
+    packssdw        m12, m13
+    packssdw        m0, m1
+    vpermq          m8, m8, 11011000b
+    vpermq          m10, m10, 11011000b
+    vpermq          m12, m12, 11011000b
+    vpermq          m0, m0, 11011000b
+%ifidn %1,pp
+    CLIPW           m8, m5, m3
+    CLIPW           m10, m5, m3
+    CLIPW           m12, m5, m3
+    CLIPW           m0, m5, m3
+%elifidn %1, sp
+    CLIPW           m8, m5, m3
+    CLIPW           m10, m5, m3
+    CLIPW           m12, m5, m3
+    CLIPW           m0, m5, m3
 %endif
-
-    movu      [r2 + x], m0
-    movu      [r2 + r3 + x], m2
-%assign x x+mmsize
-%endrep
-
-    lea       r2, [r2 + 2 * r3]
-    lea       r0, [r0 + 2 * r1]
-    dec       r4d
-    jnz       .loopH
-    RET
+    vextracti128    xm9, m8, 1
+    vextracti128    xm11, m10, 1
+    vextracti128    xm13, m12, 1
+    vextracti128    xm1, m0, 1
+    lea             r8, [r8 + r3 * 4]
+    movu            [r8], xm8
+    movu            [r8 + r3], xm9
+    movu            [r8 + r3 * 2], xm10
+    movu            [r8 + r6], xm11
+    lea             r8, [r8 + r3 * 4]
+    movu            [r8], xm12
+    movu            [r8 + r3], xm13
+    movu            [r8 + r3 * 2], xm0
+    movu            [r8 + r6], xm1
 %endmacro
-    FILTER_VER_CHROMA_W16_64xN_avx2 16, ss, 7
-    FILTER_VER_CHROMA_W16_64xN_avx2 32, ss, 7
-    FILTER_VER_CHROMA_W16_64xN_avx2 48, ss, 7
-    FILTER_VER_CHROMA_W16_64xN_avx2 64, ss, 7
-    FILTER_VER_CHROMA_W16_64xN_avx2 16, sp, 8
-    FILTER_VER_CHROMA_W16_64xN_avx2 32, sp, 8
-    FILTER_VER_CHROMA_W16_64xN_avx2 48, sp, 8
-    FILTER_VER_CHROMA_W16_64xN_avx2 64, sp, 8
-    FILTER_VER_CHROMA_W16_64xN_avx2 16, ps, 8
-    FILTER_VER_CHROMA_W16_64xN_avx2 32, ps, 8
-    FILTER_VER_CHROMA_W16_64xN_avx2 48, ps, 8
-    FILTER_VER_CHROMA_W16_64xN_avx2 64, ps, 8
-    FILTER_VER_CHROMA_W16_64xN_avx2 16, pp, 8
-    FILTER_VER_CHROMA_W16_64xN_avx2 32, pp, 8
-    FILTER_VER_CHROMA_W16_64xN_avx2 48, pp, 8
-    FILTER_VER_CHROMA_W16_64xN_avx2 64, pp, 8
 
-;-----------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_W16_12xN_avx2 3
+%macro FILTER_VER_LUMA_AVX2_Nx16 2
 INIT_YMM avx2
-cglobal interp_4tap_vert_%2_12x%1, 5, 8, %3
-    add       r1d, r1d
-    add       r3d, r3d
-    sub       r0, r1
-    shl       r4d, 6
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_%2x16, 4, 10, 15
+    mov             r4d, r4m
+    shl             r4d, 7
+    add             r1d, r1d
+    add             r3d, r3d
 
 %ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
-    lea       r5, [r5 + r4]
+    lea             r5, [tab_LumaCoeffVer]
+    add             r5, r4
 %else
-    lea       r5, [tab_ChromaCoeffV + r4]
-%endif
-    mov       r4d, %1/2
-
-%ifidn %2, pp
-    vbroadcasti128  m7, [INTERP_OFFSET_PP]
-%elifidn %2, sp
-    vbroadcasti128  m7, [INTERP_OFFSET_SP]
-%elifidn %2, ps
-    vbroadcasti128  m7, [INTERP_OFFSET_PS]
+    lea             r5, [tab_LumaCoeffVer + r4]
 %endif
 
-.loopH:
-    PROCESS_CHROMA_VERT_W16_2R
-%ifidn %2, ss
-    psrad     m0, 6
-    psrad     m1, 6
-    psrad     m2, 6
-    psrad     m3, 6
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-%elifidn %2, ps
-    paddd     m0, m7
-    paddd     m1, m7
-    paddd     m2, m7
-    paddd     m3, m7
-    psrad     m0, INTERP_SHIFT_PS
-    psrad     m1, INTERP_SHIFT_PS
-    psrad     m2, INTERP_SHIFT_PS
-    psrad     m3, INTERP_SHIFT_PS
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-%else
-    paddd     m0, m7
-    paddd     m1, m7
-    paddd     m2, m7
-    paddd     m3, m7
- %ifidn %2, pp
-    psrad     m0, INTERP_SHIFT_PP
-    psrad     m1, INTERP_SHIFT_PP
-    psrad     m2, INTERP_SHIFT_PP
-    psrad     m3, INTERP_SHIFT_PP
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+%ifidn %1,pp
+    vbroadcasti128  m14, [pd_32]
+%elifidn %1, sp
+    vbroadcasti128  m14, [INTERP_OFFSET_SP]
 %else
-    psrad     m0, INTERP_SHIFT_SP
-    psrad     m1, INTERP_SHIFT_SP
-    psrad     m2, INTERP_SHIFT_SP
-    psrad     m3, INTERP_SHIFT_SP
-%endif
-    packssdw  m0, m1
-    packssdw  m2, m3
-    pxor      m5, m5
-    CLIPW2    m0, m2, m5, [pw_pixel_max]
+    vbroadcasti128  m14, [INTERP_OFFSET_PS]
 %endif
-
-    movu      [r2], xm0
-    movu      [r2 + r3], xm2
-    vextracti128 xm0, m0, 1
-    vextracti128 xm2, m2, 1
-    movq      [r2 + 16], xm0
-    movq      [r2 + r3 + 16], xm2
-    lea       r2, [r2 + 2 * r3]
-    dec       r4d
-    jnz       .loopH
+    lea             r6, [r3 * 3]
+    mov             r9d, %2 / 8
+.loopW:
+    PROCESS_LUMA_AVX2_W8_16R %1
+    add             r2, 16
+    add             r0, 16
+    dec             r9d
+    jnz             .loopW
     RET
-%endmacro
-    FILTER_VER_CHROMA_W16_12xN_avx2 16, ss, 7
-    FILTER_VER_CHROMA_W16_12xN_avx2 16, sp, 8
-    FILTER_VER_CHROMA_W16_12xN_avx2 16, ps, 8
-    FILTER_VER_CHROMA_W16_12xN_avx2 16, pp, 8
-    FILTER_VER_CHROMA_W16_12xN_avx2 32, ss, 7
-    FILTER_VER_CHROMA_W16_12xN_avx2 32, sp, 8
-    FILTER_VER_CHROMA_W16_12xN_avx2 32, ps, 8
-    FILTER_VER_CHROMA_W16_12xN_avx2 32, pp, 8
-
-%macro PROCESS_CHROMA_VERT_W24_2R 0
-    movu       m1, [r0]
-    movu       m3, [r0 + r1]
-    punpcklwd  m0, m1, m3
-    pmaddwd    m0, [r5 + 0 * mmsize]
-    punpckhwd  m1, m3
-    pmaddwd    m1, [r5 + 0 * mmsize]
-
-    movu       xm9, [r0 + mmsize]
-    movu       xm11, [r0 + r1 + mmsize]
-    punpcklwd  xm8, xm9, xm11
-    pmaddwd    xm8, [r5 + 0 * mmsize]
-    punpckhwd  xm9, xm11
-    pmaddwd    xm9, [r5 + 0 * mmsize]
-
-    movu       m4, [r0 + 2 * r1]
-    punpcklwd  m2, m3, m4
-    pmaddwd    m2, [r5 + 0 * mmsize]
-    punpckhwd  m3, m4
-    pmaddwd    m3, [r5 + 0 * mmsize]
-
-    movu       xm12, [r0 + 2 * r1 + mmsize]
-    punpcklwd  xm10, xm11, xm12
-    pmaddwd    xm10, [r5 + 0 * mmsize]
-    punpckhwd  xm11, xm12
-    pmaddwd    xm11, [r5 + 0 * mmsize]
-
-    lea        r6, [r0 + 2 * r1]
-    movu       m5, [r6 + r1]
-    punpcklwd  m6, m4, m5
-    pmaddwd    m6, [r5 + 1 * mmsize]
-    paddd      m0, m6
-    punpckhwd  m4, m5
-    pmaddwd    m4, [r5 + 1 * mmsize]
-    paddd      m1, m4
-
-    movu       xm13, [r6 + r1 + mmsize]
-    punpcklwd  xm14, xm12, xm13
-    pmaddwd    xm14, [r5 + 1 * mmsize]
-    paddd      xm8, xm14
-    punpckhwd  xm12, xm13
-    pmaddwd    xm12, [r5 + 1 * mmsize]
-    paddd      xm9, xm12
-
-    movu       m4, [r6 + 2 * r1]
-    punpcklwd  m6, m5, m4
-    pmaddwd    m6, [r5 + 1 * mmsize]
-    paddd      m2, m6
-    punpckhwd  m5, m4
-    pmaddwd    m5, [r5 + 1 * mmsize]
-    paddd      m3, m5
-
-    movu       xm12, [r6 + 2 * r1 + mmsize]
-    punpcklwd  xm14, xm13, xm12
-    pmaddwd    xm14, [r5 + 1 * mmsize]
-    paddd      xm10, xm14
-    punpckhwd  xm13, xm12
-    pmaddwd    xm13, [r5 + 1 * mmsize]
-    paddd      xm11, xm13
+%endif
 %endmacro
 
-;-----------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_W16_24xN_avx2 3
+FILTER_VER_LUMA_AVX2_Nx16 pp, 16
+FILTER_VER_LUMA_AVX2_Nx16 pp, 32
+FILTER_VER_LUMA_AVX2_Nx16 pp, 64
+FILTER_VER_LUMA_AVX2_Nx16 ps, 16
+FILTER_VER_LUMA_AVX2_Nx16 ps, 32
+FILTER_VER_LUMA_AVX2_Nx16 ps, 64
+FILTER_VER_LUMA_AVX2_Nx16 sp, 16
+FILTER_VER_LUMA_AVX2_Nx16 sp, 32
+FILTER_VER_LUMA_AVX2_Nx16 sp, 64
+FILTER_VER_LUMA_AVX2_Nx16 ss, 16
+FILTER_VER_LUMA_AVX2_Nx16 ss, 32
+FILTER_VER_LUMA_AVX2_Nx16 ss, 64
+
+%macro FILTER_VER_LUMA_AVX2_NxN 3
 INIT_YMM avx2
-%if ARCH_X86_64
-cglobal interp_4tap_vert_%2_24x%1, 5, 7, %3
-    add       r1d, r1d
-    add       r3d, r3d
-    sub       r0, r1
-    shl       r4d, 6
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%3_%1x%2, 4, 12, 15
+    mov             r4d, r4m
+    shl             r4d, 7
+    add             r1d, r1d
+    add             r3d, r3d
 
 %ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
-    lea       r5, [r5 + r4]
+    lea             r5, [tab_LumaCoeffVer]
+    add             r5, r4
 %else
-    lea       r5, [tab_ChromaCoeffV + r4]
-%endif
-    mov       r4d, %1/2
-
-%ifidn %2, pp
-    vbroadcasti128  m7, [INTERP_OFFSET_PP]
-%elifidn %2, sp
-    vbroadcasti128  m7, [INTERP_OFFSET_SP]
-%elifidn %2, ps
-    vbroadcasti128  m7, [INTERP_OFFSET_PS]
+    lea             r5, [tab_LumaCoeffVer + r4]
 %endif
 
-.loopH:
-    PROCESS_CHROMA_VERT_W24_2R
-%ifidn %2, ss
-    psrad     m0, 6
-    psrad     m1, 6
-    psrad     m2, 6
-    psrad     m3, 6
-
-    psrad     m8, 6
-    psrad     m9, 6
-    psrad     m10, 6
-    psrad     m11, 6
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-    packssdw  m8, m9
-    packssdw  m10, m11
-%elifidn %2, ps
-    paddd     m0, m7
-    paddd     m1, m7
-    paddd     m2, m7
-    paddd     m3, m7
-    psrad     m0, INTERP_SHIFT_PS
-    psrad     m1, INTERP_SHIFT_PS
-    psrad     m2, INTERP_SHIFT_PS
-    psrad     m3, INTERP_SHIFT_PS
-    paddd     m8, m7
-    paddd     m9, m7
-    paddd     m10, m7
-    paddd     m11, m7
-    psrad     m8, INTERP_SHIFT_PS
-    psrad     m9, INTERP_SHIFT_PS
-    psrad     m10, INTERP_SHIFT_PS
-    psrad     m11, INTERP_SHIFT_PS
+    lea             r4, [r1 * 3]
+    sub             r0, r4
 
-    packssdw  m0, m1
-    packssdw  m2, m3
-    packssdw  m8, m9
-    packssdw  m10, m11
-%else
-    paddd     m0, m7
-    paddd     m1, m7
-    paddd     m2, m7
-    paddd     m3, m7
-    paddd     m8, m7
-    paddd     m9, m7
-    paddd     m10, m7
-    paddd     m11, m7
- %ifidn %2, pp
-    psrad     m0, INTERP_SHIFT_PP
-    psrad     m1, INTERP_SHIFT_PP
-    psrad     m2, INTERP_SHIFT_PP
-    psrad     m3, INTERP_SHIFT_PP
-    psrad     m8, INTERP_SHIFT_PP
-    psrad     m9, INTERP_SHIFT_PP
-    psrad     m10, INTERP_SHIFT_PP
-    psrad     m11, INTERP_SHIFT_PP
+%ifidn %3,pp
+    vbroadcasti128  m14, [pd_32]
+%elifidn %3, sp
+    vbroadcasti128  m14, [INTERP_OFFSET_SP]
 %else
-    psrad     m0, INTERP_SHIFT_SP
-    psrad     m1, INTERP_SHIFT_SP
-    psrad     m2, INTERP_SHIFT_SP
-    psrad     m3, INTERP_SHIFT_SP
-    psrad     m8, INTERP_SHIFT_SP
-    psrad     m9, INTERP_SHIFT_SP
-    psrad     m10, INTERP_SHIFT_SP
-    psrad     m11, INTERP_SHIFT_SP
-%endif
-    packssdw  m0, m1
-    packssdw  m2, m3
-    packssdw  m8, m9
-    packssdw  m10, m11
-    pxor      m5, m5
-    CLIPW2    m0, m2, m5, [pw_pixel_max]
-    CLIPW2    m8, m10, m5, [pw_pixel_max]
+    vbroadcasti128  m14, [INTERP_OFFSET_PS]
 %endif
 
-    movu      [r2], m0
-    movu      [r2 + r3], m2
-    movu      [r2 + mmsize], xm8
-    movu      [r2 + r3 + mmsize], xm10
-    lea       r2, [r2 + 2 * r3]
-    lea       r0, [r0 + 2 * r1]
-    dec       r4d
-    jnz       .loopH
+    lea             r6, [r3 * 3]
+    lea             r11, [r1 * 4]
+    mov             r9d, %2 / 16
+.loopH:
+    mov             r10d, %1 / 8
+.loopW:
+    PROCESS_LUMA_AVX2_W8_16R %3
+    add             r2, 16
+    add             r0, 16
+    dec             r10d
+    jnz             .loopW
+    sub             r7, r11
+    lea             r0, [r7 - 2 * %1 + 16]
+    lea             r2, [r8 + r3 * 4 - 2 * %1 + 16]
+    dec             r9d
+    jnz             .loopH
     RET
 %endif
 %endmacro
-    FILTER_VER_CHROMA_W16_24xN_avx2 32, ss, 15
-    FILTER_VER_CHROMA_W16_24xN_avx2 32, sp, 15
-    FILTER_VER_CHROMA_W16_24xN_avx2 32, ps, 15
-    FILTER_VER_CHROMA_W16_24xN_avx2 32, pp, 15
-    FILTER_VER_CHROMA_W16_24xN_avx2 64, ss, 15
-    FILTER_VER_CHROMA_W16_24xN_avx2 64, sp, 15
-    FILTER_VER_CHROMA_W16_24xN_avx2 64, ps, 15
-    FILTER_VER_CHROMA_W16_24xN_avx2 64, pp, 15
-
-;-----------------------------------------------------------------------------------------------------------------
-; void interp_4tap_vert(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_CHROMA_W16_48x64_avx2 2
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_48x64, 5, 7, %2
-    add       r1d, r1d
-    add       r3d, r3d
-    sub       r0, r1
-    shl       r4d, 6
 
-%ifdef PIC
-    lea       r5, [tab_ChromaCoeffV]
-    lea       r5, [r5 + r4]
-%else
-    lea       r5, [tab_ChromaCoeffV + r4]
-%endif
-    mov       r4d, 32
+FILTER_VER_LUMA_AVX2_NxN 16, 32, pp
+FILTER_VER_LUMA_AVX2_NxN 16, 64, pp
+FILTER_VER_LUMA_AVX2_NxN 24, 32, pp
+FILTER_VER_LUMA_AVX2_NxN 32, 32, pp
+FILTER_VER_LUMA_AVX2_NxN 32, 64, pp
+FILTER_VER_LUMA_AVX2_NxN 48, 64, pp
+FILTER_VER_LUMA_AVX2_NxN 64, 32, pp
+FILTER_VER_LUMA_AVX2_NxN 64, 48, pp
+FILTER_VER_LUMA_AVX2_NxN 64, 64, pp
+FILTER_VER_LUMA_AVX2_NxN 16, 32, ps
+FILTER_VER_LUMA_AVX2_NxN 16, 64, ps
+FILTER_VER_LUMA_AVX2_NxN 24, 32, ps
+FILTER_VER_LUMA_AVX2_NxN 32, 32, ps
+FILTER_VER_LUMA_AVX2_NxN 32, 64, ps
+FILTER_VER_LUMA_AVX2_NxN 48, 64, ps
+FILTER_VER_LUMA_AVX2_NxN 64, 32, ps
+FILTER_VER_LUMA_AVX2_NxN 64, 48, ps
+FILTER_VER_LUMA_AVX2_NxN 64, 64, ps
+FILTER_VER_LUMA_AVX2_NxN 16, 32, sp
+FILTER_VER_LUMA_AVX2_NxN 16, 64, sp
+FILTER_VER_LUMA_AVX2_NxN 24, 32, sp
+FILTER_VER_LUMA_AVX2_NxN 32, 32, sp
+FILTER_VER_LUMA_AVX2_NxN 32, 64, sp
+FILTER_VER_LUMA_AVX2_NxN 48, 64, sp
+FILTER_VER_LUMA_AVX2_NxN 64, 32, sp
+FILTER_VER_LUMA_AVX2_NxN 64, 48, sp
+FILTER_VER_LUMA_AVX2_NxN 64, 64, sp
+FILTER_VER_LUMA_AVX2_NxN 16, 32, ss
+FILTER_VER_LUMA_AVX2_NxN 16, 64, ss
+FILTER_VER_LUMA_AVX2_NxN 24, 32, ss
+FILTER_VER_LUMA_AVX2_NxN 32, 32, ss
+FILTER_VER_LUMA_AVX2_NxN 32, 64, ss
+FILTER_VER_LUMA_AVX2_NxN 48, 64, ss
+FILTER_VER_LUMA_AVX2_NxN 64, 32, ss
+FILTER_VER_LUMA_AVX2_NxN 64, 48, ss
+FILTER_VER_LUMA_AVX2_NxN 64, 64, ss
 
-%ifidn %1, pp
-    vbroadcasti128  m7, [INTERP_OFFSET_PP]
-%elifidn %1, sp
-    vbroadcasti128  m7, [INTERP_OFFSET_SP]
-%elifidn %1, ps
-    vbroadcasti128  m7, [INTERP_OFFSET_PS]
-%endif
-
-.loopH:
-%assign x 0
-%rep 3
-    movu       m1, [r0 + x]
-    movu       m3, [r0 + r1 + x]
-    movu       m5, [r5 + 0 * mmsize]
-    punpcklwd  m0, m1, m3
-    pmaddwd    m0, m5
-    punpckhwd  m1, m3
-    pmaddwd    m1, m5
-
-    movu       m4, [r0 + 2 * r1 + x]
-    punpcklwd  m2, m3, m4
-    pmaddwd    m2, m5
-    punpckhwd  m3, m4
-    pmaddwd    m3, m5
-
-    lea        r6, [r0 + 2 * r1]
-    movu       m5, [r6 + r1 + x]
-    punpcklwd  m6, m4, m5
-    pmaddwd    m6, [r5 + 1 * mmsize]
-    paddd      m0, m6
-    punpckhwd  m4, m5
-    pmaddwd    m4, [r5 + 1 * mmsize]
-    paddd      m1, m4
-
-    movu       m4, [r6 + 2 * r1 + x]
-    punpcklwd  m6, m5, m4
-    pmaddwd    m6, [r5 + 1 * mmsize]
-    paddd      m2, m6
-    punpckhwd  m5, m4
-    pmaddwd    m5, [r5 + 1 * mmsize]
-    paddd      m3, m5
-
-%ifidn %1, ss
-    psrad     m0, 6
-    psrad     m1, 6
-    psrad     m2, 6
-    psrad     m3, 6
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-%elifidn %1, ps
-    paddd     m0, m7
-    paddd     m1, m7
-    paddd     m2, m7
-    paddd     m3, m7
-    psrad     m0, INTERP_SHIFT_PS
-    psrad     m1, INTERP_SHIFT_PS
-    psrad     m2, INTERP_SHIFT_PS
-    psrad     m3, INTERP_SHIFT_PS
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-%else
-    paddd     m0, m7
-    paddd     m1, m7
-    paddd     m2, m7
-    paddd     m3, m7
-%ifidn %1, pp
-    psrad     m0, INTERP_SHIFT_PP
-    psrad     m1, INTERP_SHIFT_PP
-    psrad     m2, INTERP_SHIFT_PP
-    psrad     m3, INTERP_SHIFT_PP
-%else
-    psrad     m0, INTERP_SHIFT_SP
-    psrad     m1, INTERP_SHIFT_SP
-    psrad     m2, INTERP_SHIFT_SP
-    psrad     m3, INTERP_SHIFT_SP
-%endif
-    packssdw  m0, m1
-    packssdw  m2, m3
-    pxor      m5, m5
-    CLIPW2    m0, m2, m5, [pw_pixel_max]
-%endif
-
-    movu      [r2 + x], m0
-    movu      [r2 + r3 + x], m2
-%assign x x+mmsize
-%endrep
-
-    lea       r2, [r2 + 2 * r3]
-    lea       r0, [r0 + 2 * r1]
-    dec       r4d
-    jnz       .loopH
-    RET
-%endmacro
-
-    FILTER_VER_CHROMA_W16_48x64_avx2 pp, 8
-    FILTER_VER_CHROMA_W16_48x64_avx2 ps, 8
-    FILTER_VER_CHROMA_W16_48x64_avx2 ss, 7
-    FILTER_VER_CHROMA_W16_48x64_avx2 sp, 8
-
-INIT_XMM sse2
-cglobal chroma_p2s, 3, 7, 3
-    ; load width and height
-    mov         r3d, r3m
-    mov         r4d, r4m
-    add         r1, r1
-
-    ; load constant
-    mova        m2, [tab_c_n8192]
-
-.loopH:
-
-    xor         r5d, r5d
-.loopW:
-    lea         r6, [r0 + r5 * 2]
-
-    movu        m0, [r6]
-    psllw       m0, (14 - BIT_DEPTH)
-    paddw       m0, m2
-
-    movu        m1, [r6 + r1]
-    psllw       m1, (14 - BIT_DEPTH)
-    paddw       m1, m2
-
-    add         r5d, 8
-    cmp         r5d, r3d
-    lea         r6, [r2 + r5 * 2]
-    jg          .width4
-    movu        [r6 + FENC_STRIDE / 2 * 0 - 16], m0
-    movu        [r6 + FENC_STRIDE / 2 * 2 - 16], m1
-    je          .nextH
-    jmp         .loopW
-
-.width4:
-    test        r3d, 4
-    jz          .width2
-    test        r3d, 2
-    movh        [r6 + FENC_STRIDE / 2 * 0 - 16], m0
-    movh        [r6 + FENC_STRIDE / 2 * 2 - 16], m1
-    lea         r6, [r6 + 8]
-    pshufd      m0, m0, 2
-    pshufd      m1, m1, 2
-    jz          .nextH
-
-.width2:
-    movd        [r6 + FENC_STRIDE / 2 * 0 - 16], m0
-    movd        [r6 + FENC_STRIDE / 2 * 2 - 16], m1
-
-.nextH:
-    lea         r0, [r0 + r1 * 2]
-    add         r2, FENC_STRIDE / 2 * 4
-
-    sub         r4d, 2
-    jnz         .loopH
-    RET
-
-%macro PROCESS_LUMA_VER_W4_4R 0
-    movq       m0, [r0]
-    movq       m1, [r0 + r1]
-    punpcklwd  m0, m1                          ;m0=[0 1]
-    pmaddwd    m0, [r6 + 0 *16]                ;m0=[0+1]  Row1
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m4, [r0]
-    punpcklwd  m1, m4                          ;m1=[1 2]
-    pmaddwd    m1, [r6 + 0 *16]                ;m1=[1+2]  Row2
-
-    movq       m5, [r0 + r1]
-    punpcklwd  m4, m5                          ;m4=[2 3]
-    pmaddwd    m2, m4, [r6 + 0 *16]            ;m2=[2+3]  Row3
-    pmaddwd    m4, [r6 + 1 * 16]
-    paddd      m0, m4                          ;m0=[0+1+2+3]  Row1
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m4, [r0]
-    punpcklwd  m5, m4                          ;m5=[3 4]
-    pmaddwd    m3, m5, [r6 + 0 *16]            ;m3=[3+4]  Row4
-    pmaddwd    m5, [r6 + 1 * 16]
-    paddd      m1, m5                          ;m1 = [1+2+3+4]  Row2
-
-    movq       m5, [r0 + r1]
-    punpcklwd  m4, m5                          ;m4=[4 5]
-    pmaddwd    m6, m4, [r6 + 1 * 16]
-    paddd      m2, m6                          ;m2=[2+3+4+5]  Row3
-    pmaddwd    m4, [r6 + 2 * 16]
-    paddd      m0, m4                          ;m0=[0+1+2+3+4+5]  Row1
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m4, [r0]
-    punpcklwd  m5, m4                          ;m5=[5 6]
-    pmaddwd    m6, m5, [r6 + 1 * 16]
-    paddd      m3, m6                          ;m3=[3+4+5+6]  Row4
-    pmaddwd    m5, [r6 + 2 * 16]
-    paddd      m1, m5                          ;m1=[1+2+3+4+5+6]  Row2
-
-    movq       m5, [r0 + r1]
-    punpcklwd  m4, m5                          ;m4=[6 7]
-    pmaddwd    m6, m4, [r6 + 2 * 16]
-    paddd      m2, m6                          ;m2=[2+3+4+5+6+7]  Row3
-    pmaddwd    m4, [r6 + 3 * 16]
-    paddd      m0, m4                          ;m0=[0+1+2+3+4+5+6+7]  Row1 end
-
-    lea        r0, [r0 + 2 * r1]
-    movq       m4, [r0]
-    punpcklwd  m5, m4                          ;m5=[7 8]
-    pmaddwd    m6, m5, [r6 + 2 * 16]
-    paddd      m3, m6                          ;m3=[3+4+5+6+7+8]  Row4
-    pmaddwd    m5, [r6 + 3 * 16]
-    paddd      m1, m5                          ;m1=[1+2+3+4+5+6+7+8]  Row2 end
-
-    movq       m5, [r0 + r1]
-    punpcklwd  m4, m5                          ;m4=[8 9]
-    pmaddwd    m4, [r6 + 3 * 16]
-    paddd      m2, m4                          ;m2=[2+3+4+5+6+7+8+9]  Row3 end
-
-    movq       m4, [r0 + 2 * r1]
-    punpcklwd  m5, m4                          ;m5=[9 10]
-    pmaddwd    m5, [r6 + 3 * 16]
-    paddd      m3, m5                          ;m3=[3+4+5+6+7+8+9+10]  Row4 end
-%endmacro
-
-;--------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_LUMA_PP 2
-INIT_XMM sse4
-cglobal interp_8tap_vert_pp_%1x%2, 5, 7, 8 ,0-gprsize
-
-    add       r1d, r1d
-    add       r3d, r3d
-    lea       r5, [r1 + 2 * r1]
-    sub       r0, r5
-    shl       r4d, 6
-
-%ifdef PIC
-    lea       r5, [tab_LumaCoeffV]
-    lea       r6, [r5 + r4]
-%else
-    lea       r6, [tab_LumaCoeffV + r4]
-%endif
-
-    mova      m7, [INTERP_OFFSET_PP]
-
-    mov       dword [rsp], %2/4
-.loopH:
-    mov       r4d, (%1/4)
-.loopW:
-    PROCESS_LUMA_VER_W4_4R
-
-    paddd     m0, m7
-    paddd     m1, m7
-    paddd     m2, m7
-    paddd     m3, m7
-
-    psrad     m0, INTERP_SHIFT_PP
-    psrad     m1, INTERP_SHIFT_PP
-    psrad     m2, INTERP_SHIFT_PP
-    psrad     m3, INTERP_SHIFT_PP
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-
-    pxor      m1, m1
-    CLIPW2    m0, m2, m1, [pw_pixel_max]
-
-    movh      [r2], m0
-    movhps    [r2 + r3], m0
-    lea       r5, [r2 + 2 * r3]
-    movh      [r5], m2
-    movhps    [r5 + r3], m2
-
-    lea       r5, [8 * r1 - 2 * 4]
-    sub       r0, r5
-    add       r2, 2 * 4
-
-    dec       r4d
-    jnz       .loopW
-
-    lea       r0, [r0 + 4 * r1 - 2 * %1]
-    lea       r2, [r2 + 4 * r3 - 2 * %1]
-
-    dec       dword [rsp]
-    jnz       .loopH
-    RET
-%endmacro
-
-;-------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_PP 4, 4
-    FILTER_VER_LUMA_PP 8, 8
-    FILTER_VER_LUMA_PP 8, 4
-    FILTER_VER_LUMA_PP 4, 8
-    FILTER_VER_LUMA_PP 16, 16
-    FILTER_VER_LUMA_PP 16, 8
-    FILTER_VER_LUMA_PP 8, 16
-    FILTER_VER_LUMA_PP 16, 12
-    FILTER_VER_LUMA_PP 12, 16
-    FILTER_VER_LUMA_PP 16, 4
-    FILTER_VER_LUMA_PP 4, 16
-    FILTER_VER_LUMA_PP 32, 32
-    FILTER_VER_LUMA_PP 32, 16
-    FILTER_VER_LUMA_PP 16, 32
-    FILTER_VER_LUMA_PP 32, 24
-    FILTER_VER_LUMA_PP 24, 32
-    FILTER_VER_LUMA_PP 32, 8
-    FILTER_VER_LUMA_PP 8, 32
-    FILTER_VER_LUMA_PP 64, 64
-    FILTER_VER_LUMA_PP 64, 32
-    FILTER_VER_LUMA_PP 32, 64
-    FILTER_VER_LUMA_PP 64, 48
-    FILTER_VER_LUMA_PP 48, 64
-    FILTER_VER_LUMA_PP 64, 16
-    FILTER_VER_LUMA_PP 16, 64
-
-%macro FILTER_VER_LUMA_AVX2_4x4 1
+%macro FILTER_VER_LUMA_AVX2_8xN 2
 INIT_YMM avx2
-cglobal interp_8tap_vert_%1_4x4, 4, 6, 7
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_8x%2, 4, 9, 15
     mov             r4d, r4m
+    shl             r4d, 7
     add             r1d, r1d
     add             r3d, r3d
-    shl             r4d, 7
 
 %ifdef PIC
     lea             r5, [tab_LumaCoeffVer]
@@ -2379,151 +1361,43 @@ cglobal interp_8tap_vert_%1_4x4, 4, 6, 7
 
     lea             r4, [r1 * 3]
     sub             r0, r4
-
 %ifidn %1,pp
-    vbroadcasti128  m6, [pd_32]
+    vbroadcasti128  m14, [pd_32]
 %elifidn %1, sp
-    vbroadcasti128  m6, [INTERP_OFFSET_SP]
+    vbroadcasti128  m14, [INTERP_OFFSET_SP]
 %else
-    vbroadcasti128  m6, [INTERP_OFFSET_PS]
+    vbroadcasti128  m14, [INTERP_OFFSET_PS]
 %endif
-
-    movq            xm0, [r0]
-    movq            xm1, [r0 + r1]
+    lea             r6, [r3 * 3]
+    lea             r7, [r1 * 4]
+    mov             r8d, %2 / 16
+.loopH:
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
     punpcklwd       xm0, xm1
-    movq            xm2, [r0 + r1 * 2]
-    punpcklwd       xm1, xm2
-    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
+    vinserti128     m0, m0, xm2, 1
     pmaddwd         m0, [r5]
-    movq            xm3, [r0 + r4]
-    punpcklwd       xm2, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm4, [r0]
-    punpcklwd       xm3, xm4
-    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
-    pmaddwd         m5, m2, [r5 + 1 * mmsize]
-    pmaddwd         m2, [r5]
-    paddd           m0, m5
-    movq            xm3, [r0 + r1]
-    punpcklwd       xm4, xm3
-    movq            xm1, [r0 + r1 * 2]
-    punpcklwd       xm3, xm1
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
-    pmaddwd         m5, m4, [r5 + 2 * mmsize]
-    pmaddwd         m4, [r5 + 1 * mmsize]
-    paddd           m0, m5
-    paddd           m2, m4
-    movq            xm3, [r0 + r4]
-    punpcklwd       xm1, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm4, [r0]
-    punpcklwd       xm3, xm4
-    vinserti128     m1, m1, xm3, 1                  ; m1 = [8 7 7 6]
-    pmaddwd         m5, m1, [r5 + 3 * mmsize]
-    pmaddwd         m1, [r5 + 2 * mmsize]
-    paddd           m0, m5
-    paddd           m2, m1
-    movq            xm3, [r0 + r1]
-    punpcklwd       xm4, xm3
-    movq            xm1, [r0 + 2 * r1]
-    punpcklwd       xm3, xm1
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [A 9 9 8]
-    pmaddwd         m4, [r5 + 3 * mmsize]
-    paddd           m2, m4
-
-%ifidn %1,ss
-    psrad           m0, 6
-    psrad           m2, 6
-%else
-    paddd           m0, m6
-    paddd           m2, m6
-%ifidn %1,pp
-    psrad           m0, INTERP_SHIFT_PP
-    psrad           m2, INTERP_SHIFT_PP
-%elifidn %1, sp
-    psrad           m0, INTERP_SHIFT_SP
-    psrad           m2, INTERP_SHIFT_SP
-%else
-    psrad           m0, INTERP_SHIFT_PS
-    psrad           m2, INTERP_SHIFT_PS
-%endif
-%endif
-
-    packssdw        m0, m2
-    pxor            m1, m1
-%ifidn %1,pp
-    CLIPW           m0, m1, [pw_pixel_max]
-%elifidn %1, sp
-    CLIPW           m0, m1, [pw_pixel_max]
-%endif
-
-    vextracti128    xm2, m0, 1
-    lea             r4, [r3 * 3]
-    movq            [r2], xm0
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r4], xm2
-    RET
-%endmacro
-
-FILTER_VER_LUMA_AVX2_4x4 pp
-FILTER_VER_LUMA_AVX2_4x4 ps
-FILTER_VER_LUMA_AVX2_4x4 sp
-FILTER_VER_LUMA_AVX2_4x4 ss
-
-%macro FILTER_VER_LUMA_AVX2_8x8 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_8x8, 4, 6, 12
-    mov             r4d, r4m
-    add             r1d, r1d
-    add             r3d, r3d
-    shl             r4d, 7
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-
-%ifidn %1,pp
-    vbroadcasti128  m11, [pd_32]
-%elifidn %1, sp
-    vbroadcasti128  m11, [INTERP_OFFSET_SP]
-%else
-    vbroadcasti128  m11, [INTERP_OFFSET_PS]
-%endif
-
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
     punpcklwd       xm2, xm3
     vinserti128     m2, m2, xm4, 1
     pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    pmaddwd         m2, [r5]
     paddd           m0, m4
+    pmaddwd         m2, [r5]
     lea             r0, [r0 + r1 * 4]
     movu            xm4, [r0]                       ; m4 = row 4
     punpckhwd       xm5, xm3, xm4
     punpcklwd       xm3, xm4
     vinserti128     m3, m3, xm5, 1
     pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    pmaddwd         m3, [r5]
     paddd           m1, m5
+    pmaddwd         m3, [r5]
     movu            xm5, [r0 + r1]                  ; m5 = row 5
     punpckhwd       xm6, xm4, xm5
     punpcklwd       xm4, xm5
@@ -2540,8 +1414,8 @@ cglobal interp_8tap_vert_%1_8x8, 4, 6, 12
     pmaddwd         m7, m5, [r5 + 2 * mmsize]
     paddd           m1, m7
     pmaddwd         m7, m5, [r5 + 1 * mmsize]
-    pmaddwd         m5, [r5]
     paddd           m3, m7
+    pmaddwd         m5, [r5]
     movu            xm7, [r0 + r4]                  ; m7 = row 7
     punpckhwd       xm8, xm6, xm7
     punpcklwd       xm6, xm7
@@ -2551,8 +1425,8 @@ cglobal interp_8tap_vert_%1_8x8, 4, 6, 12
     pmaddwd         m8, m6, [r5 + 2 * mmsize]
     paddd           m2, m8
     pmaddwd         m8, m6, [r5 + 1 * mmsize]
-    pmaddwd         m6, [r5]
     paddd           m4, m8
+    pmaddwd         m6, [r5]
     lea             r0, [r0 + r1 * 4]
     movu            xm8, [r0]                       ; m8 = row 8
     punpckhwd       xm9, xm7, xm8
@@ -2563,8 +1437,8 @@ cglobal interp_8tap_vert_%1_8x8, 4, 6, 12
     pmaddwd         m9, m7, [r5 + 2 * mmsize]
     paddd           m3, m9
     pmaddwd         m9, m7, [r5 + 1 * mmsize]
-    pmaddwd         m7, [r5]
     paddd           m5, m9
+    pmaddwd         m7, [r5]
     movu            xm9, [r0 + r1]                  ; m9 = row 9
     punpckhwd       xm10, xm8, xm9
     punpcklwd       xm8, xm9
@@ -2572,156 +1446,336 @@ cglobal interp_8tap_vert_%1_8x8, 4, 6, 12
     pmaddwd         m10, m8, [r5 + 3 * mmsize]
     paddd           m2, m10
     pmaddwd         m10, m8, [r5 + 2 * mmsize]
-    pmaddwd         m8, [r5 + 1 * mmsize]
     paddd           m4, m10
-    paddd           m6, m8
+    pmaddwd         m10, m8, [r5 + 1 * mmsize]
+    paddd           m6, m10
+    pmaddwd         m8, [r5]
     movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
-    punpckhwd       xm8, xm9, xm10
+    punpckhwd       xm11, xm9, xm10
     punpcklwd       xm9, xm10
-    vinserti128     m9, m9, xm8, 1
-    pmaddwd         m8, m9, [r5 + 3 * mmsize]
-    paddd           m3, m8
-    pmaddwd         m8, m9, [r5 + 2 * mmsize]
-    pmaddwd         m9, [r5 + 1 * mmsize]
-    paddd           m5, m8
-    paddd           m7, m9
-    movu            xm8, [r0 + r4]                  ; m8 = row 11
-    punpckhwd       xm9, xm10, xm8
-    punpcklwd       xm10, xm8
-    vinserti128     m10, m10, xm9, 1
-    pmaddwd         m9, m10, [r5 + 3 * mmsize]
-    pmaddwd         m10, [r5 + 2 * mmsize]
-    paddd           m4, m9
-    paddd           m6, m10
+    vinserti128     m9, m9, xm11, 1
+    pmaddwd         m11, m9, [r5 + 3 * mmsize]
+    paddd           m3, m11
+    pmaddwd         m11, m9, [r5 + 2 * mmsize]
+    paddd           m5, m11
+    pmaddwd         m11, m9, [r5 + 1 * mmsize]
+    paddd           m7, m11
+    pmaddwd         m9, [r5]
+    movu            xm11, [r0 + r4]                 ; m11 = row 11
+    punpckhwd       xm12, xm10, xm11
+    punpcklwd       xm10, xm11
+    vinserti128     m10, m10, xm12, 1
+    pmaddwd         m12, m10, [r5 + 3 * mmsize]
+    paddd           m4, m12
+    pmaddwd         m12, m10, [r5 + 2 * mmsize]
+    paddd           m6, m12
+    pmaddwd         m12, m10, [r5 + 1 * mmsize]
+    paddd           m8, m12
+    pmaddwd         m10, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm12, [r0]                      ; m12 = row 12
+    punpckhwd       xm13, xm11, xm12
+    punpcklwd       xm11, xm12
+    vinserti128     m11, m11, xm13, 1
+    pmaddwd         m13, m11, [r5 + 3 * mmsize]
+    paddd           m5, m13
+    pmaddwd         m13, m11, [r5 + 2 * mmsize]
+    paddd           m7, m13
+    pmaddwd         m13, m11, [r5 + 1 * mmsize]
+    paddd           m9, m13
+    pmaddwd         m11, [r5]
 
-    lea             r4, [r3 * 3]
 %ifidn %1,ss
     psrad           m0, 6
     psrad           m1, 6
     psrad           m2, 6
     psrad           m3, 6
+    psrad           m4, 6
+    psrad           m5, 6
 %else
-    paddd           m0, m11
-    paddd           m1, m11
-    paddd           m2, m11
-    paddd           m3, m11
+    paddd           m0, m14
+    paddd           m1, m14
+    paddd           m2, m14
+    paddd           m3, m14
+    paddd           m4, m14
+    paddd           m5, m14
 %ifidn %1,pp
     psrad           m0, INTERP_SHIFT_PP
     psrad           m1, INTERP_SHIFT_PP
     psrad           m2, INTERP_SHIFT_PP
     psrad           m3, INTERP_SHIFT_PP
+    psrad           m4, INTERP_SHIFT_PP
+    psrad           m5, INTERP_SHIFT_PP
 %elifidn %1, sp
     psrad           m0, INTERP_SHIFT_SP
     psrad           m1, INTERP_SHIFT_SP
     psrad           m2, INTERP_SHIFT_SP
     psrad           m3, INTERP_SHIFT_SP
+    psrad           m4, INTERP_SHIFT_SP
+    psrad           m5, INTERP_SHIFT_SP
 %else
     psrad           m0, INTERP_SHIFT_PS
     psrad           m1, INTERP_SHIFT_PS
     psrad           m2, INTERP_SHIFT_PS
     psrad           m3, INTERP_SHIFT_PS
+    psrad           m4, INTERP_SHIFT_PS
+    psrad           m5, INTERP_SHIFT_PS
 %endif
 %endif
 
     packssdw        m0, m1
     packssdw        m2, m3
+    packssdw        m4, m5
     vpermq          m0, m0, 11011000b
     vpermq          m2, m2, 11011000b
-    pxor            m10, m10
-    mova            m9, [pw_pixel_max]
+    vpermq          m4, m4, 11011000b
+    pxor            m5, m5
+    mova            m3, [pw_pixel_max]
 %ifidn %1,pp
-    CLIPW           m0, m10, m9
-    CLIPW           m2, m10, m9
+    CLIPW           m0, m5, m3
+    CLIPW           m2, m5, m3
+    CLIPW           m4, m5, m3
 %elifidn %1, sp
-    CLIPW           m0, m10, m9
-    CLIPW           m2, m10, m9
+    CLIPW           m0, m5, m3
+    CLIPW           m2, m5, m3
+    CLIPW           m4, m5, m3
 %endif
 
     vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
     movu            [r2], xm0
     movu            [r2 + r3], xm1
+    vextracti128    xm1, m2, 1
     movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r4], xm3
+    movu            [r2 + r6], xm1
+    lea             r2, [r2 + r3 * 4]
+    vextracti128    xm1, m4, 1
+    movu            [r2], xm4
+    movu            [r2 + r3], xm1
 
-    lea             r0, [r0 + r1 * 4]
-    movu            xm2, [r0]                       ; m2 = row 12
-    punpckhwd       xm3, xm8, xm2
-    punpcklwd       xm8, xm2
-    vinserti128     m8, m8, xm3, 1
-    pmaddwd         m3, m8, [r5 + 3 * mmsize]
-    pmaddwd         m8, [r5 + 2 * mmsize]
-    paddd           m5, m3
-    paddd           m7, m8
-    movu            xm3, [r0 + r1]                  ; m3 = row 13
-    punpckhwd       xm0, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm0, 1
-    pmaddwd         m2, [r5 + 3 * mmsize]
-    paddd           m6, m2
+    movu            xm13, [r0 + r1]                 ; m13 = row 13
+    punpckhwd       xm0, xm12, xm13
+    punpcklwd       xm12, xm13
+    vinserti128     m12, m12, xm0, 1
+    pmaddwd         m0, m12, [r5 + 3 * mmsize]
+    paddd           m6, m0
+    pmaddwd         m0, m12, [r5 + 2 * mmsize]
+    paddd           m8, m0
+    pmaddwd         m0, m12, [r5 + 1 * mmsize]
+    paddd           m10, m0
+    pmaddwd         m12, [r5]
     movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
-    punpckhwd       xm1, xm3, xm0
-    punpcklwd       xm3, xm0
-    vinserti128     m3, m3, xm1, 1
-    pmaddwd         m3, [r5 + 3 * mmsize]
-    paddd           m7, m3
-
-%ifidn %1,ss
-    psrad           m4, 6
-    psrad           m5, 6
-    psrad           m6, 6
-    psrad           m7, 6
-%else
-    paddd           m4, m11
-    paddd           m5, m11
-    paddd           m6, m11
-    paddd           m7, m11
+    punpckhwd       xm1, xm13, xm0
+    punpcklwd       xm13, xm0
+    vinserti128     m13, m13, xm1, 1
+    pmaddwd         m1, m13, [r5 + 3 * mmsize]
+    paddd           m7, m1
+    pmaddwd         m1, m13, [r5 + 2 * mmsize]
+    paddd           m9, m1
+    pmaddwd         m1, m13, [r5 + 1 * mmsize]
+    paddd           m11, m1
+    pmaddwd         m13, [r5]
+
+%ifidn %1,ss
+    psrad           m6, 6
+    psrad           m7, 6
+%else
+    paddd           m6, m14
+    paddd           m7, m14
 %ifidn %1,pp
-    psrad           m4, INTERP_SHIFT_PP
-    psrad           m5, INTERP_SHIFT_PP
     psrad           m6, INTERP_SHIFT_PP
     psrad           m7, INTERP_SHIFT_PP
 %elifidn %1, sp
-    psrad           m4, INTERP_SHIFT_SP
-    psrad           m5, INTERP_SHIFT_SP
     psrad           m6, INTERP_SHIFT_SP
     psrad           m7, INTERP_SHIFT_SP
 %else
-    psrad           m4, INTERP_SHIFT_PS
-    psrad           m5, INTERP_SHIFT_PS
     psrad           m6, INTERP_SHIFT_PS
     psrad           m7, INTERP_SHIFT_PS
 %endif
 %endif
 
-    packssdw        m4, m5
     packssdw        m6, m7
-    vpermq          m4, m4, 11011000b
     vpermq          m6, m6, 11011000b
 %ifidn %1,pp
-    CLIPW           m4, m10, m9
-    CLIPW           m6, m10, m9
+    CLIPW           m6, m5, m3
 %elifidn %1, sp
-    CLIPW           m4, m10, m9
-    CLIPW           m6, m10, m9
+    CLIPW           m6, m5, m3
 %endif
-    vextracti128    xm5, m4, 1
     vextracti128    xm7, m6, 1
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm4
-    movu            [r2 + r3], xm5
     movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r4], xm7
+    movu            [r2 + r6], xm7
+
+    movu            xm1, [r0 + r4]                  ; m1 = row 15
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m2, m0, [r5 + 3 * mmsize]
+    paddd           m8, m2
+    pmaddwd         m2, m0, [r5 + 2 * mmsize]
+    paddd           m10, m2
+    pmaddwd         m2, m0, [r5 + 1 * mmsize]
+    paddd           m12, m2
+    pmaddwd         m0, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm2, [r0]                       ; m2 = row 16
+    punpckhwd       xm6, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm6, 1
+    pmaddwd         m6, m1, [r5 + 3 * mmsize]
+    paddd           m9, m6
+    pmaddwd         m6, m1, [r5 + 2 * mmsize]
+    paddd           m11, m6
+    pmaddwd         m6, m1, [r5 + 1 * mmsize]
+    paddd           m13, m6
+    pmaddwd         m1, [r5]
+    movu            xm6, [r0 + r1]                  ; m6 = row 17
+    punpckhwd       xm4, xm2, xm6
+    punpcklwd       xm2, xm6
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 3 * mmsize]
+    paddd           m10, m4
+    pmaddwd         m4, m2, [r5 + 2 * mmsize]
+    paddd           m12, m4
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m0, m2
+    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
+    punpckhwd       xm2, xm6, xm4
+    punpcklwd       xm6, xm4
+    vinserti128     m6, m6, xm2, 1
+    pmaddwd         m2, m6, [r5 + 3 * mmsize]
+    paddd           m11, m2
+    pmaddwd         m2, m6, [r5 + 2 * mmsize]
+    paddd           m13, m2
+    pmaddwd         m6, [r5 + 1 * mmsize]
+    paddd           m1, m6
+    movu            xm2, [r0 + r4]                  ; m2 = row 19
+    punpckhwd       xm6, xm4, xm2
+    punpcklwd       xm4, xm2
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 3 * mmsize]
+    paddd           m12, m6
+    pmaddwd         m4, [r5 + 2 * mmsize]
+    paddd           m0, m4
+    lea             r0, [r0 + r1 * 4]
+    movu            xm6, [r0]                       ; m6 = row 20
+    punpckhwd       xm7, xm2, xm6
+    punpcklwd       xm2, xm6
+    vinserti128     m2, m2, xm7, 1
+    pmaddwd         m7, m2, [r5 + 3 * mmsize]
+    paddd           m13, m7
+    pmaddwd         m2, [r5 + 2 * mmsize]
+    paddd           m1, m2
+    movu            xm7, [r0 + r1]                  ; m7 = row 21
+    punpckhwd       xm2, xm6, xm7
+    punpcklwd       xm6, xm7
+    vinserti128     m6, m6, xm2, 1
+    pmaddwd         m6, [r5 + 3 * mmsize]
+    paddd           m0, m6
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 22
+    punpckhwd       xm6, xm7, xm2
+    punpcklwd       xm7, xm2
+    vinserti128     m7, m7, xm6, 1
+    pmaddwd         m7, [r5 + 3 * mmsize]
+    paddd           m1, m7
+
+%ifidn %1,ss
+    psrad           m8, 6
+    psrad           m9, 6
+    psrad           m10, 6
+    psrad           m11, 6
+    psrad           m12, 6
+    psrad           m13, 6
+    psrad           m0, 6
+    psrad           m1, 6
+%else
+    paddd           m8, m14
+    paddd           m9, m14
+    paddd           m10, m14
+    paddd           m11, m14
+    paddd           m12, m14
+    paddd           m13, m14
+    paddd           m0, m14
+    paddd           m1, m14
+%ifidn %1,pp
+    psrad           m8, INTERP_SHIFT_PP
+    psrad           m9, INTERP_SHIFT_PP
+    psrad           m10, INTERP_SHIFT_PP
+    psrad           m11, INTERP_SHIFT_PP
+    psrad           m12, INTERP_SHIFT_PP
+    psrad           m13, INTERP_SHIFT_PP
+    psrad           m0, INTERP_SHIFT_PP
+    psrad           m1, INTERP_SHIFT_PP
+%elifidn %1, sp
+    psrad           m8, INTERP_SHIFT_SP
+    psrad           m9, INTERP_SHIFT_SP
+    psrad           m10, INTERP_SHIFT_SP
+    psrad           m11, INTERP_SHIFT_SP
+    psrad           m12, INTERP_SHIFT_SP
+    psrad           m13, INTERP_SHIFT_SP
+    psrad           m0, INTERP_SHIFT_SP
+    psrad           m1, INTERP_SHIFT_SP
+%else
+    psrad           m8, INTERP_SHIFT_PS
+    psrad           m9, INTERP_SHIFT_PS
+    psrad           m10, INTERP_SHIFT_PS
+    psrad           m11, INTERP_SHIFT_PS
+    psrad           m12, INTERP_SHIFT_PS
+    psrad           m13, INTERP_SHIFT_PS
+    psrad           m0, INTERP_SHIFT_PS
+    psrad           m1, INTERP_SHIFT_PS
+%endif
+%endif
+
+    packssdw        m8, m9
+    packssdw        m10, m11
+    packssdw        m12, m13
+    packssdw        m0, m1
+    vpermq          m8, m8, 11011000b
+    vpermq          m10, m10, 11011000b
+    vpermq          m12, m12, 11011000b
+    vpermq          m0, m0, 11011000b
+%ifidn %1,pp
+    CLIPW           m8, m5, m3
+    CLIPW           m10, m5, m3
+    CLIPW           m12, m5, m3
+    CLIPW           m0, m5, m3
+%elifidn %1, sp
+    CLIPW           m8, m5, m3
+    CLIPW           m10, m5, m3
+    CLIPW           m12, m5, m3
+    CLIPW           m0, m5, m3
+%endif
+    vextracti128    xm9, m8, 1
+    vextracti128    xm11, m10, 1
+    vextracti128    xm13, m12, 1
+    vextracti128    xm1, m0, 1
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm8
+    movu            [r2 + r3], xm9
+    movu            [r2 + r3 * 2], xm10
+    movu            [r2 + r6], xm11
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm12
+    movu            [r2 + r3], xm13
+    movu            [r2 + r3 * 2], xm0
+    movu            [r2 + r6], xm1
+    lea             r2, [r2 + r3 * 4]
+    sub             r0, r7
+    dec             r8d
+    jnz             .loopH
     RET
 %endif
 %endmacro
 
-FILTER_VER_LUMA_AVX2_8x8 pp
-FILTER_VER_LUMA_AVX2_8x8 ps
-FILTER_VER_LUMA_AVX2_8x8 sp
-FILTER_VER_LUMA_AVX2_8x8 ss
+FILTER_VER_LUMA_AVX2_8xN pp, 16
+FILTER_VER_LUMA_AVX2_8xN pp, 32
+FILTER_VER_LUMA_AVX2_8xN ps, 16
+FILTER_VER_LUMA_AVX2_8xN ps, 32
+FILTER_VER_LUMA_AVX2_8xN sp, 16
+FILTER_VER_LUMA_AVX2_8xN sp, 32
+FILTER_VER_LUMA_AVX2_8xN ss, 16
+FILTER_VER_LUMA_AVX2_8xN ss, 32
 
-%macro PROCESS_LUMA_AVX2_W8_16R 1
+%macro PROCESS_LUMA_AVX2_W8_8R 1
     movu            xm0, [r0]                       ; m0 = row 0
     movu            xm1, [r0 + r1]                  ; m1 = row 1
     punpckhwd       xm2, xm0, xm1
@@ -2797,43 +1851,35 @@ FILTER_VER_LUMA_AVX2_8x8 ss
     paddd           m2, m10
     pmaddwd         m10, m8, [r5 + 2 * mmsize]
     paddd           m4, m10
-    pmaddwd         m10, m8, [r5 + 1 * mmsize]
-    paddd           m6, m10
-    pmaddwd         m8, [r5]
+    pmaddwd         m8, [r5 + 1 * mmsize]
+    paddd           m6, m8
     movu            xm10, [r7 + r1 * 2]             ; m10 = row 10
-    punpckhwd       xm11, xm9, xm10
+    punpckhwd       xm8, xm9, xm10
     punpcklwd       xm9, xm10
-    vinserti128     m9, m9, xm11, 1
-    pmaddwd         m11, m9, [r5 + 3 * mmsize]
-    paddd           m3, m11
-    pmaddwd         m11, m9, [r5 + 2 * mmsize]
-    paddd           m5, m11
-    pmaddwd         m11, m9, [r5 + 1 * mmsize]
-    paddd           m7, m11
-    pmaddwd         m9, [r5]
-    movu            xm11, [r7 + r4]                 ; m11 = row 11
-    punpckhwd       xm12, xm10, xm11
-    punpcklwd       xm10, xm11
-    vinserti128     m10, m10, xm12, 1
-    pmaddwd         m12, m10, [r5 + 3 * mmsize]
-    paddd           m4, m12
-    pmaddwd         m12, m10, [r5 + 2 * mmsize]
-    paddd           m6, m12
-    pmaddwd         m12, m10, [r5 + 1 * mmsize]
-    paddd           m8, m12
-    pmaddwd         m10, [r5]
+    vinserti128     m9, m9, xm8, 1
+    pmaddwd         m8, m9, [r5 + 3 * mmsize]
+    paddd           m3, m8
+    pmaddwd         m8, m9, [r5 + 2 * mmsize]
+    paddd           m5, m8
+    pmaddwd         m9, [r5 + 1 * mmsize]
+    paddd           m7, m9
+    movu            xm8, [r7 + r4]                  ; m8 = row 11
+    punpckhwd       xm9, xm10, xm8
+    punpcklwd       xm10, xm8
+    vinserti128     m10, m10, xm9, 1
+    pmaddwd         m9, m10, [r5 + 3 * mmsize]
+    paddd           m4, m9
+    pmaddwd         m10, [r5 + 2 * mmsize]
+    paddd           m6, m10
     lea             r7, [r7 + r1 * 4]
-    movu            xm12, [r7]                      ; m12 = row 12
-    punpckhwd       xm13, xm11, xm12
-    punpcklwd       xm11, xm12
-    vinserti128     m11, m11, xm13, 1
-    pmaddwd         m13, m11, [r5 + 3 * mmsize]
-    paddd           m5, m13
-    pmaddwd         m13, m11, [r5 + 2 * mmsize]
-    paddd           m7, m13
-    pmaddwd         m13, m11, [r5 + 1 * mmsize]
-    paddd           m9, m13
-    pmaddwd         m11, [r5]
+    movu            xm9, [r7]                       ; m9 = row 12
+    punpckhwd       xm10, xm8, xm9
+    punpcklwd       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddwd         m10, m8, [r5 + 3 * mmsize]
+    paddd           m5, m10
+    pmaddwd         m8, [r5 + 2 * mmsize]
+    paddd           m7, m8
 
 %ifidn %1,ss
     psrad           m0, 6
@@ -2843,12 +1889,12 @@ FILTER_VER_LUMA_AVX2_8x8 ss
     psrad           m4, 6
     psrad           m5, 6
 %else
-    paddd           m0, m14
-    paddd           m1, m14
-    paddd           m2, m14
-    paddd           m3, m14
-    paddd           m4, m14
-    paddd           m5, m14
+    paddd           m0, m11
+    paddd           m1, m11
+    paddd           m2, m11
+    paddd           m3, m11
+    paddd           m4, m11
+    paddd           m5, m11
 %ifidn %1,pp
     psrad           m0, INTERP_SHIFT_PP
     psrad           m1, INTERP_SHIFT_PP
@@ -2879,58 +1925,47 @@ FILTER_VER_LUMA_AVX2_8x8 ss
     vpermq          m0, m0, 11011000b
     vpermq          m2, m2, 11011000b
     vpermq          m4, m4, 11011000b
-    pxor            m5, m5
-    mova            m3, [pw_pixel_max]
+    pxor            m8, m8
 %ifidn %1,pp
-    CLIPW           m0, m5, m3
-    CLIPW           m2, m5, m3
-    CLIPW           m4, m5, m3
+    CLIPW           m0, m8, m12
+    CLIPW           m2, m8, m12
+    CLIPW           m4, m8, m12
 %elifidn %1, sp
-    CLIPW           m0, m5, m3
-    CLIPW           m2, m5, m3
-    CLIPW           m4, m5, m3
+    CLIPW           m0, m8, m12
+    CLIPW           m2, m8, m12
+    CLIPW           m4, m8, m12
 %endif
 
     vextracti128    xm1, m0, 1
+    vextracti128    xm3, m2, 1
+    vextracti128    xm5, m4, 1
     movu            [r2], xm0
     movu            [r2 + r3], xm1
-    vextracti128    xm1, m2, 1
     movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm1
+    movu            [r2 + r6], xm3
     lea             r8, [r2 + r3 * 4]
-    vextracti128    xm1, m4, 1
     movu            [r8], xm4
-    movu            [r8 + r3], xm1
+    movu            [r8 + r3], xm5
 
-    movu            xm13, [r7 + r1]                 ; m13 = row 13
-    punpckhwd       xm0, xm12, xm13
-    punpcklwd       xm12, xm13
-    vinserti128     m12, m12, xm0, 1
-    pmaddwd         m0, m12, [r5 + 3 * mmsize]
-    paddd           m6, m0
-    pmaddwd         m0, m12, [r5 + 2 * mmsize]
-    paddd           m8, m0
-    pmaddwd         m0, m12, [r5 + 1 * mmsize]
-    paddd           m10, m0
-    pmaddwd         m12, [r5]
+    movu            xm10, [r7 + r1]                 ; m10 = row 13
+    punpckhwd       xm0, xm9, xm10
+    punpcklwd       xm9, xm10
+    vinserti128     m9, m9, xm0, 1
+    pmaddwd         m9, [r5 + 3 * mmsize]
+    paddd           m6, m9
     movu            xm0, [r7 + r1 * 2]              ; m0 = row 14
-    punpckhwd       xm1, xm13, xm0
-    punpcklwd       xm13, xm0
-    vinserti128     m13, m13, xm1, 1
-    pmaddwd         m1, m13, [r5 + 3 * mmsize]
-    paddd           m7, m1
-    pmaddwd         m1, m13, [r5 + 2 * mmsize]
-    paddd           m9, m1
-    pmaddwd         m1, m13, [r5 + 1 * mmsize]
-    paddd           m11, m1
-    pmaddwd         m13, [r5]
+    punpckhwd       xm1, xm10, xm0
+    punpcklwd       xm10, xm0
+    vinserti128     m10, m10, xm1, 1
+    pmaddwd         m10, [r5 + 3 * mmsize]
+    paddd           m7, m10
 
 %ifidn %1,ss
     psrad           m6, 6
     psrad           m7, 6
 %else
-    paddd           m6, m14
-    paddd           m7, m14
+    paddd           m6, m11
+    paddd           m7, m11
 %ifidn %1,pp
     psrad           m6, INTERP_SHIFT_PP
     psrad           m7, INTERP_SHIFT_PP
@@ -2946,174 +1981,19 @@ FILTER_VER_LUMA_AVX2_8x8 ss
     packssdw        m6, m7
     vpermq          m6, m6, 11011000b
 %ifidn %1,pp
-    CLIPW           m6, m5, m3
+    CLIPW           m6, m8, m12
 %elifidn %1, sp
-    CLIPW           m6, m5, m3
+    CLIPW           m6, m8, m12
 %endif
     vextracti128    xm7, m6, 1
     movu            [r8 + r3 * 2], xm6
     movu            [r8 + r6], xm7
-
-    movu            xm1, [r7 + r4]                  ; m1 = row 15
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m2, m0, [r5 + 3 * mmsize]
-    paddd           m8, m2
-    pmaddwd         m2, m0, [r5 + 2 * mmsize]
-    paddd           m10, m2
-    pmaddwd         m2, m0, [r5 + 1 * mmsize]
-    paddd           m12, m2
-    pmaddwd         m0, [r5]
-    lea             r7, [r7 + r1 * 4]
-    movu            xm2, [r7]                       ; m2 = row 16
-    punpckhwd       xm6, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm6, 1
-    pmaddwd         m6, m1, [r5 + 3 * mmsize]
-    paddd           m9, m6
-    pmaddwd         m6, m1, [r5 + 2 * mmsize]
-    paddd           m11, m6
-    pmaddwd         m6, m1, [r5 + 1 * mmsize]
-    paddd           m13, m6
-    pmaddwd         m1, [r5]
-    movu            xm6, [r7 + r1]                  ; m6 = row 17
-    punpckhwd       xm4, xm2, xm6
-    punpcklwd       xm2, xm6
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 3 * mmsize]
-    paddd           m10, m4
-    pmaddwd         m4, m2, [r5 + 2 * mmsize]
-    paddd           m12, m4
-    pmaddwd         m2, [r5 + 1 * mmsize]
-    paddd           m0, m2
-    movu            xm4, [r7 + r1 * 2]              ; m4 = row 18
-    punpckhwd       xm2, xm6, xm4
-    punpcklwd       xm6, xm4
-    vinserti128     m6, m6, xm2, 1
-    pmaddwd         m2, m6, [r5 + 3 * mmsize]
-    paddd           m11, m2
-    pmaddwd         m2, m6, [r5 + 2 * mmsize]
-    paddd           m13, m2
-    pmaddwd         m6, [r5 + 1 * mmsize]
-    paddd           m1, m6
-    movu            xm2, [r7 + r4]                  ; m2 = row 19
-    punpckhwd       xm6, xm4, xm2
-    punpcklwd       xm4, xm2
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 3 * mmsize]
-    paddd           m12, m6
-    pmaddwd         m4, [r5 + 2 * mmsize]
-    paddd           m0, m4
-    lea             r7, [r7 + r1 * 4]
-    movu            xm6, [r7]                       ; m6 = row 20
-    punpckhwd       xm7, xm2, xm6
-    punpcklwd       xm2, xm6
-    vinserti128     m2, m2, xm7, 1
-    pmaddwd         m7, m2, [r5 + 3 * mmsize]
-    paddd           m13, m7
-    pmaddwd         m2, [r5 + 2 * mmsize]
-    paddd           m1, m2
-    movu            xm7, [r7 + r1]                  ; m7 = row 21
-    punpckhwd       xm2, xm6, xm7
-    punpcklwd       xm6, xm7
-    vinserti128     m6, m6, xm2, 1
-    pmaddwd         m6, [r5 + 3 * mmsize]
-    paddd           m0, m6
-    movu            xm2, [r7 + r1 * 2]              ; m2 = row 22
-    punpckhwd       xm6, xm7, xm2
-    punpcklwd       xm7, xm2
-    vinserti128     m7, m7, xm6, 1
-    pmaddwd         m7, [r5 + 3 * mmsize]
-    paddd           m1, m7
-
-%ifidn %1,ss
-    psrad           m8, 6
-    psrad           m9, 6
-    psrad           m10, 6
-    psrad           m11, 6
-    psrad           m12, 6
-    psrad           m13, 6
-    psrad           m0, 6
-    psrad           m1, 6
-%else
-    paddd           m8, m14
-    paddd           m9, m14
-    paddd           m10, m14
-    paddd           m11, m14
-    paddd           m12, m14
-    paddd           m13, m14
-    paddd           m0, m14
-    paddd           m1, m14
-%ifidn %1,pp
-    psrad           m8, INTERP_SHIFT_PP
-    psrad           m9, INTERP_SHIFT_PP
-    psrad           m10, INTERP_SHIFT_PP
-    psrad           m11, INTERP_SHIFT_PP
-    psrad           m12, INTERP_SHIFT_PP
-    psrad           m13, INTERP_SHIFT_PP
-    psrad           m0, INTERP_SHIFT_PP
-    psrad           m1, INTERP_SHIFT_PP
-%elifidn %1, sp
-    psrad           m8, INTERP_SHIFT_SP
-    psrad           m9, INTERP_SHIFT_SP
-    psrad           m10, INTERP_SHIFT_SP
-    psrad           m11, INTERP_SHIFT_SP
-    psrad           m12, INTERP_SHIFT_SP
-    psrad           m13, INTERP_SHIFT_SP
-    psrad           m0, INTERP_SHIFT_SP
-    psrad           m1, INTERP_SHIFT_SP
-%else
-    psrad           m8, INTERP_SHIFT_PS
-    psrad           m9, INTERP_SHIFT_PS
-    psrad           m10, INTERP_SHIFT_PS
-    psrad           m11, INTERP_SHIFT_PS
-    psrad           m12, INTERP_SHIFT_PS
-    psrad           m13, INTERP_SHIFT_PS
-    psrad           m0, INTERP_SHIFT_PS
-    psrad           m1, INTERP_SHIFT_PS
-%endif
-%endif
-
-    packssdw        m8, m9
-    packssdw        m10, m11
-    packssdw        m12, m13
-    packssdw        m0, m1
-    vpermq          m8, m8, 11011000b
-    vpermq          m10, m10, 11011000b
-    vpermq          m12, m12, 11011000b
-    vpermq          m0, m0, 11011000b
-%ifidn %1,pp
-    CLIPW           m8, m5, m3
-    CLIPW           m10, m5, m3
-    CLIPW           m12, m5, m3
-    CLIPW           m0, m5, m3
-%elifidn %1, sp
-    CLIPW           m8, m5, m3
-    CLIPW           m10, m5, m3
-    CLIPW           m12, m5, m3
-    CLIPW           m0, m5, m3
-%endif
-    vextracti128    xm9, m8, 1
-    vextracti128    xm11, m10, 1
-    vextracti128    xm13, m12, 1
-    vextracti128    xm1, m0, 1
-    lea             r8, [r8 + r3 * 4]
-    movu            [r8], xm8
-    movu            [r8 + r3], xm9
-    movu            [r8 + r3 * 2], xm10
-    movu            [r8 + r6], xm11
-    lea             r8, [r8 + r3 * 4]
-    movu            [r8], xm12
-    movu            [r8 + r3], xm13
-    movu            [r8 + r3 * 2], xm0
-    movu            [r8 + r6], xm1
 %endmacro
 
-%macro FILTER_VER_LUMA_AVX2_Nx16 2
+%macro FILTER_VER_LUMA_AVX2_Nx8 2
 INIT_YMM avx2
 %if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_%2x16, 4, 10, 15
+cglobal interp_8tap_vert_%1_%2x8, 4, 10, 13
     mov             r4d, r4m
     shl             r4d, 7
     add             r1d, r1d
@@ -3129,16 +2009,17 @@ cglobal interp_8tap_vert_%1_%2x16, 4, 10, 15
     lea             r4, [r1 * 3]
     sub             r0, r4
 %ifidn %1,pp
-    vbroadcasti128  m14, [pd_32]
+    vbroadcasti128  m11, [pd_32]
 %elifidn %1, sp
-    vbroadcasti128  m14, [INTERP_OFFSET_SP]
+    vbroadcasti128  m11, [INTERP_OFFSET_SP]
 %else
-    vbroadcasti128  m14, [INTERP_OFFSET_PS]
+    vbroadcasti128  m11, [INTERP_OFFSET_PS]
 %endif
+    mova            m12, [pw_pixel_max]
     lea             r6, [r3 * 3]
     mov             r9d, %2 / 8
 .loopW:
-    PROCESS_LUMA_AVX2_W8_16R %1
+    PROCESS_LUMA_AVX2_W8_8R %1
     add             r2, 16
     add             r0, 16
     dec             r9d
@@ -3147,23 +2028,19 @@ cglobal interp_8tap_vert_%1_%2x16, 4, 10, 15
 %endif
 %endmacro
 
-FILTER_VER_LUMA_AVX2_Nx16 pp, 16
-FILTER_VER_LUMA_AVX2_Nx16 pp, 32
-FILTER_VER_LUMA_AVX2_Nx16 pp, 64
-FILTER_VER_LUMA_AVX2_Nx16 ps, 16
-FILTER_VER_LUMA_AVX2_Nx16 ps, 32
-FILTER_VER_LUMA_AVX2_Nx16 ps, 64
-FILTER_VER_LUMA_AVX2_Nx16 sp, 16
-FILTER_VER_LUMA_AVX2_Nx16 sp, 32
-FILTER_VER_LUMA_AVX2_Nx16 sp, 64
-FILTER_VER_LUMA_AVX2_Nx16 ss, 16
-FILTER_VER_LUMA_AVX2_Nx16 ss, 32
-FILTER_VER_LUMA_AVX2_Nx16 ss, 64
+FILTER_VER_LUMA_AVX2_Nx8 pp, 32
+FILTER_VER_LUMA_AVX2_Nx8 pp, 16
+FILTER_VER_LUMA_AVX2_Nx8 ps, 32
+FILTER_VER_LUMA_AVX2_Nx8 ps, 16
+FILTER_VER_LUMA_AVX2_Nx8 sp, 32
+FILTER_VER_LUMA_AVX2_Nx8 sp, 16
+FILTER_VER_LUMA_AVX2_Nx8 ss, 32
+FILTER_VER_LUMA_AVX2_Nx8 ss, 16
 
-%macro FILTER_VER_LUMA_AVX2_NxN 3
+%macro FILTER_VER_LUMA_AVX2_32x24 1
 INIT_YMM avx2
 %if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%3_%1x%2, 4, 12, 15
+cglobal interp_8tap_vert_%1_32x24, 4, 10, 15
     mov             r4d, r4m
     shl             r4d, 7
     add             r1d, r1d
@@ -3178,101 +2055,44 @@ cglobal interp_8tap_vert_%3_%1x%2, 4, 12, 15
 
     lea             r4, [r1 * 3]
     sub             r0, r4
-
-%ifidn %3,pp
+%ifidn %1,pp
     vbroadcasti128  m14, [pd_32]
-%elifidn %3, sp
+%elifidn %1, sp
     vbroadcasti128  m14, [INTERP_OFFSET_SP]
 %else
     vbroadcasti128  m14, [INTERP_OFFSET_PS]
 %endif
-
     lea             r6, [r3 * 3]
-    lea             r11, [r1 * 4]
-    mov             r9d, %2 / 16
-.loopH:
-    mov             r10d, %1 / 8
+    mov             r9d, 4
 .loopW:
-    PROCESS_LUMA_AVX2_W8_16R %3
+    PROCESS_LUMA_AVX2_W8_16R %1
     add             r2, 16
     add             r0, 16
-    dec             r10d
+    dec             r9d
     jnz             .loopW
-    sub             r7, r11
-    lea             r0, [r7 - 2 * %1 + 16]
-    lea             r2, [r8 + r3 * 4 - 2 * %1 + 16]
+    lea             r9, [r1 * 4]
+    sub             r7, r9
+    lea             r0, [r7 - 48]
+    lea             r2, [r8 + r3 * 4 - 48]
+    mova            m11, m14
+    mova            m12, m3
+    mov             r9d, 4
+.loop:
+    PROCESS_LUMA_AVX2_W8_8R %1
+    add             r2, 16
+    add             r0, 16
     dec             r9d
-    jnz             .loopH
+    jnz             .loop
     RET
 %endif
 %endmacro
 
-FILTER_VER_LUMA_AVX2_NxN 16, 32, pp
-FILTER_VER_LUMA_AVX2_NxN 16, 64, pp
-FILTER_VER_LUMA_AVX2_NxN 24, 32, pp
-FILTER_VER_LUMA_AVX2_NxN 32, 32, pp
-FILTER_VER_LUMA_AVX2_NxN 32, 64, pp
-FILTER_VER_LUMA_AVX2_NxN 48, 64, pp
-FILTER_VER_LUMA_AVX2_NxN 64, 32, pp
-FILTER_VER_LUMA_AVX2_NxN 64, 48, pp
-FILTER_VER_LUMA_AVX2_NxN 64, 64, pp
-FILTER_VER_LUMA_AVX2_NxN 16, 32, ps
-FILTER_VER_LUMA_AVX2_NxN 16, 64, ps
-FILTER_VER_LUMA_AVX2_NxN 24, 32, ps
-FILTER_VER_LUMA_AVX2_NxN 32, 32, ps
-FILTER_VER_LUMA_AVX2_NxN 32, 64, ps
-FILTER_VER_LUMA_AVX2_NxN 48, 64, ps
-FILTER_VER_LUMA_AVX2_NxN 64, 32, ps
-FILTER_VER_LUMA_AVX2_NxN 64, 48, ps
-FILTER_VER_LUMA_AVX2_NxN 64, 64, ps
-FILTER_VER_LUMA_AVX2_NxN 16, 32, sp
-FILTER_VER_LUMA_AVX2_NxN 16, 64, sp
-FILTER_VER_LUMA_AVX2_NxN 24, 32, sp
-FILTER_VER_LUMA_AVX2_NxN 32, 32, sp
-FILTER_VER_LUMA_AVX2_NxN 32, 64, sp
-FILTER_VER_LUMA_AVX2_NxN 48, 64, sp
-FILTER_VER_LUMA_AVX2_NxN 64, 32, sp
-FILTER_VER_LUMA_AVX2_NxN 64, 48, sp
-FILTER_VER_LUMA_AVX2_NxN 64, 64, sp
-FILTER_VER_LUMA_AVX2_NxN 16, 32, ss
-FILTER_VER_LUMA_AVX2_NxN 16, 64, ss
-FILTER_VER_LUMA_AVX2_NxN 24, 32, ss
-FILTER_VER_LUMA_AVX2_NxN 32, 32, ss
-FILTER_VER_LUMA_AVX2_NxN 32, 64, ss
-FILTER_VER_LUMA_AVX2_NxN 48, 64, ss
-FILTER_VER_LUMA_AVX2_NxN 64, 32, ss
-FILTER_VER_LUMA_AVX2_NxN 64, 48, ss
-FILTER_VER_LUMA_AVX2_NxN 64, 64, ss
-
-%macro FILTER_VER_LUMA_AVX2_8xN 2
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_8x%2, 4, 9, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-    add             r1d, r1d
-    add             r3d, r3d
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer + r4]
-%endif
+FILTER_VER_LUMA_AVX2_32x24 pp
+FILTER_VER_LUMA_AVX2_32x24 ps
+FILTER_VER_LUMA_AVX2_32x24 sp
+FILTER_VER_LUMA_AVX2_32x24 ss
 
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-%ifidn %1,pp
-    vbroadcasti128  m14, [pd_32]
-%elifidn %1, sp
-    vbroadcasti128  m14, [INTERP_OFFSET_SP]
-%else
-    vbroadcasti128  m14, [INTERP_OFFSET_PS]
-%endif
-    lea             r6, [r3 * 3]
-    lea             r7, [r1 * 4]
-    mov             r8d, %2 / 16
-.loopH:
+%macro PROCESS_LUMA_AVX2_W8_4R 1
     movu            xm0, [r0]                       ; m0 = row 0
     movu            xm1, [r0 + r1]                  ; m1 = row 1
     punpckhwd       xm2, xm0, xm1
@@ -3305,409 +2125,236 @@ cglobal interp_8tap_vert_%1_8x%2, 4, 9, 15
     vinserti128     m4, m4, xm6, 1
     pmaddwd         m6, m4, [r5 + 2 * mmsize]
     paddd           m0, m6
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m2, m6
-    pmaddwd         m4, [r5]
+    pmaddwd         m4, [r5 + 1 * mmsize]
+    paddd           m2, m4
     movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm7, xm5, xm6
+    punpckhwd       xm4, xm5, xm6
     punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddwd         m7, m5, [r5 + 2 * mmsize]
-    paddd           m1, m7
-    pmaddwd         m7, m5, [r5 + 1 * mmsize]
-    paddd           m3, m7
-    pmaddwd         m5, [r5]
-    movu            xm7, [r0 + r4]                  ; m7 = row 7
-    punpckhwd       xm8, xm6, xm7
-    punpcklwd       xm6, xm7
-    vinserti128     m6, m6, xm8, 1
-    pmaddwd         m8, m6, [r5 + 3 * mmsize]
-    paddd           m0, m8
-    pmaddwd         m8, m6, [r5 + 2 * mmsize]
-    paddd           m2, m8
-    pmaddwd         m8, m6, [r5 + 1 * mmsize]
-    paddd           m4, m8
-    pmaddwd         m6, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm8, [r0]                       ; m8 = row 8
-    punpckhwd       xm9, xm7, xm8
-    punpcklwd       xm7, xm8
-    vinserti128     m7, m7, xm9, 1
-    pmaddwd         m9, m7, [r5 + 3 * mmsize]
-    paddd           m1, m9
-    pmaddwd         m9, m7, [r5 + 2 * mmsize]
-    paddd           m3, m9
-    pmaddwd         m9, m7, [r5 + 1 * mmsize]
-    paddd           m5, m9
-    pmaddwd         m7, [r5]
-    movu            xm9, [r0 + r1]                  ; m9 = row 9
-    punpckhwd       xm10, xm8, xm9
-    punpcklwd       xm8, xm9
-    vinserti128     m8, m8, xm10, 1
-    pmaddwd         m10, m8, [r5 + 3 * mmsize]
-    paddd           m2, m10
-    pmaddwd         m10, m8, [r5 + 2 * mmsize]
-    paddd           m4, m10
-    pmaddwd         m10, m8, [r5 + 1 * mmsize]
-    paddd           m6, m10
-    pmaddwd         m8, [r5]
-    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
-    punpckhwd       xm11, xm9, xm10
-    punpcklwd       xm9, xm10
-    vinserti128     m9, m9, xm11, 1
-    pmaddwd         m11, m9, [r5 + 3 * mmsize]
-    paddd           m3, m11
-    pmaddwd         m11, m9, [r5 + 2 * mmsize]
-    paddd           m5, m11
-    pmaddwd         m11, m9, [r5 + 1 * mmsize]
-    paddd           m7, m11
-    pmaddwd         m9, [r5]
-    movu            xm11, [r0 + r4]                 ; m11 = row 11
-    punpckhwd       xm12, xm10, xm11
-    punpcklwd       xm10, xm11
-    vinserti128     m10, m10, xm12, 1
-    pmaddwd         m12, m10, [r5 + 3 * mmsize]
-    paddd           m4, m12
-    pmaddwd         m12, m10, [r5 + 2 * mmsize]
-    paddd           m6, m12
-    pmaddwd         m12, m10, [r5 + 1 * mmsize]
-    paddd           m8, m12
-    pmaddwd         m10, [r5]
+    vinserti128     m5, m5, xm4, 1
+    pmaddwd         m4, m5, [r5 + 2 * mmsize]
+    paddd           m1, m4
+    pmaddwd         m5, [r5 + 1 * mmsize]
+    paddd           m3, m5
+    movu            xm4, [r0 + r4]                  ; m4 = row 7
+    punpckhwd       xm5, xm6, xm4
+    punpcklwd       xm6, xm4
+    vinserti128     m6, m6, xm5, 1
+    pmaddwd         m5, m6, [r5 + 3 * mmsize]
+    paddd           m0, m5
+    pmaddwd         m6, [r5 + 2 * mmsize]
+    paddd           m2, m6
     lea             r0, [r0 + r1 * 4]
-    movu            xm12, [r0]                      ; m12 = row 12
-    punpckhwd       xm13, xm11, xm12
-    punpcklwd       xm11, xm12
-    vinserti128     m11, m11, xm13, 1
-    pmaddwd         m13, m11, [r5 + 3 * mmsize]
-    paddd           m5, m13
-    pmaddwd         m13, m11, [r5 + 2 * mmsize]
-    paddd           m7, m13
-    pmaddwd         m13, m11, [r5 + 1 * mmsize]
-    paddd           m9, m13
-    pmaddwd         m11, [r5]
+    movu            xm5, [r0]                       ; m5 = row 8
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 3 * mmsize]
+    paddd           m1, m6
+    pmaddwd         m4, [r5 + 2 * mmsize]
+    paddd           m3, m4
+    movu            xm6, [r0 + r1]                  ; m6 = row 9
+    punpckhwd       xm4, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm4, 1
+    pmaddwd         m5, [r5 + 3 * mmsize]
+    paddd           m2, m5
+    movu            xm4, [r0 + r1 * 2]              ; m4 = row 10
+    punpckhwd       xm5, xm6, xm4
+    punpcklwd       xm6, xm4
+    vinserti128     m6, m6, xm5, 1
+    pmaddwd         m6, [r5 + 3 * mmsize]
+    paddd           m3, m6
 
 %ifidn %1,ss
     psrad           m0, 6
     psrad           m1, 6
     psrad           m2, 6
     psrad           m3, 6
-    psrad           m4, 6
-    psrad           m5, 6
 %else
-    paddd           m0, m14
-    paddd           m1, m14
-    paddd           m2, m14
-    paddd           m3, m14
-    paddd           m4, m14
-    paddd           m5, m14
+    paddd           m0, m7
+    paddd           m1, m7
+    paddd           m2, m7
+    paddd           m3, m7
 %ifidn %1,pp
     psrad           m0, INTERP_SHIFT_PP
     psrad           m1, INTERP_SHIFT_PP
     psrad           m2, INTERP_SHIFT_PP
     psrad           m3, INTERP_SHIFT_PP
-    psrad           m4, INTERP_SHIFT_PP
-    psrad           m5, INTERP_SHIFT_PP
 %elifidn %1, sp
     psrad           m0, INTERP_SHIFT_SP
     psrad           m1, INTERP_SHIFT_SP
     psrad           m2, INTERP_SHIFT_SP
     psrad           m3, INTERP_SHIFT_SP
-    psrad           m4, INTERP_SHIFT_SP
-    psrad           m5, INTERP_SHIFT_SP
 %else
     psrad           m0, INTERP_SHIFT_PS
     psrad           m1, INTERP_SHIFT_PS
     psrad           m2, INTERP_SHIFT_PS
     psrad           m3, INTERP_SHIFT_PS
-    psrad           m4, INTERP_SHIFT_PS
-    psrad           m5, INTERP_SHIFT_PS
 %endif
 %endif
 
     packssdw        m0, m1
     packssdw        m2, m3
-    packssdw        m4, m5
     vpermq          m0, m0, 11011000b
     vpermq          m2, m2, 11011000b
-    vpermq          m4, m4, 11011000b
-    pxor            m5, m5
-    mova            m3, [pw_pixel_max]
+    pxor            m4, m4
 %ifidn %1,pp
-    CLIPW           m0, m5, m3
-    CLIPW           m2, m5, m3
-    CLIPW           m4, m5, m3
+    CLIPW           m0, m4, [pw_pixel_max]
+    CLIPW           m2, m4, [pw_pixel_max]
 %elifidn %1, sp
-    CLIPW           m0, m5, m3
-    CLIPW           m2, m5, m3
-    CLIPW           m4, m5, m3
+    CLIPW           m0, m4, [pw_pixel_max]
+    CLIPW           m2, m4, [pw_pixel_max]
 %endif
 
     vextracti128    xm1, m0, 1
+    vextracti128    xm3, m2, 1
+%endmacro
+
+%macro FILTER_VER_LUMA_AVX2_16x4 1
+INIT_YMM avx2
+cglobal interp_8tap_vert_%1_16x4, 4, 7, 8, 0-gprsize
+    mov             r4d, r4m
+    shl             r4d, 7
+    add             r1d, r1d
+    add             r3d, r3d
+
+%ifdef PIC
+    lea             r5, [tab_LumaCoeffVer]
+    add             r5, r4
+%else
+    lea             r5, [tab_LumaCoeffVer + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+%ifidn %1,pp
+    vbroadcasti128  m7, [pd_32]
+%elifidn %1, sp
+    vbroadcasti128  m7, [INTERP_OFFSET_SP]
+%else
+    vbroadcasti128  m7, [INTERP_OFFSET_PS]
+%endif
+    mov             dword [rsp], 2
+.loopW:
+    PROCESS_LUMA_AVX2_W8_4R %1
     movu            [r2], xm0
     movu            [r2 + r3], xm1
-    vextracti128    xm1, m2, 1
     movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm1
-    lea             r2, [r2 + r3 * 4]
-    vextracti128    xm1, m4, 1
-    movu            [r2], xm4
-    movu            [r2 + r3], xm1
+    lea             r6, [r3 * 3]
+    movu            [r2 + r6], xm3
+    add             r2, 16
+    lea             r6, [8 * r1 - 16]
+    sub             r0, r6
+    dec             dword [rsp]
+    jnz             .loopW
+    RET
+%endmacro
 
-    movu            xm13, [r0 + r1]                 ; m13 = row 13
-    punpckhwd       xm0, xm12, xm13
-    punpcklwd       xm12, xm13
-    vinserti128     m12, m12, xm0, 1
-    pmaddwd         m0, m12, [r5 + 3 * mmsize]
-    paddd           m6, m0
-    pmaddwd         m0, m12, [r5 + 2 * mmsize]
-    paddd           m8, m0
-    pmaddwd         m0, m12, [r5 + 1 * mmsize]
-    paddd           m10, m0
-    pmaddwd         m12, [r5]
-    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
-    punpckhwd       xm1, xm13, xm0
-    punpcklwd       xm13, xm0
-    vinserti128     m13, m13, xm1, 1
-    pmaddwd         m1, m13, [r5 + 3 * mmsize]
-    paddd           m7, m1
-    pmaddwd         m1, m13, [r5 + 2 * mmsize]
-    paddd           m9, m1
-    pmaddwd         m1, m13, [r5 + 1 * mmsize]
-    paddd           m11, m1
-    pmaddwd         m13, [r5]
+FILTER_VER_LUMA_AVX2_16x4 pp
+FILTER_VER_LUMA_AVX2_16x4 ps
+FILTER_VER_LUMA_AVX2_16x4 sp
+FILTER_VER_LUMA_AVX2_16x4 ss
 
-%ifidn %1,ss
-    psrad           m6, 6
-    psrad           m7, 6
+%macro FILTER_VER_LUMA_AVX2_8x4 1
+INIT_YMM avx2
+cglobal interp_8tap_vert_%1_8x4, 4, 6, 8
+    mov             r4d, r4m
+    shl             r4d, 7
+    add             r1d, r1d
+    add             r3d, r3d
+
+%ifdef PIC
+    lea             r5, [tab_LumaCoeffVer]
+    add             r5, r4
 %else
-    paddd           m6, m14
-    paddd           m7, m14
+    lea             r5, [tab_LumaCoeffVer + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r4
 %ifidn %1,pp
-    psrad           m6, INTERP_SHIFT_PP
-    psrad           m7, INTERP_SHIFT_PP
+    vbroadcasti128  m7, [pd_32]
 %elifidn %1, sp
-    psrad           m6, INTERP_SHIFT_SP
-    psrad           m7, INTERP_SHIFT_SP
+    vbroadcasti128  m7, [INTERP_OFFSET_SP]
 %else
-    psrad           m6, INTERP_SHIFT_PS
-    psrad           m7, INTERP_SHIFT_PS
+    vbroadcasti128  m7, [INTERP_OFFSET_PS]
 %endif
+
+    PROCESS_LUMA_AVX2_W8_4R %1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    lea             r4, [r3 * 3]
+    movu            [r2 + r4], xm3
+    RET
+%endmacro
+
+FILTER_VER_LUMA_AVX2_8x4 pp
+FILTER_VER_LUMA_AVX2_8x4 ps
+FILTER_VER_LUMA_AVX2_8x4 sp
+FILTER_VER_LUMA_AVX2_8x4 ss
+
+%macro FILTER_VER_LUMA_AVX2_16x12 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_16x12, 4, 10, 15
+    mov             r4d, r4m
+    shl             r4d, 7
+    add             r1d, r1d
+    add             r3d, r3d
+
+%ifdef PIC
+    lea             r5, [tab_LumaCoeffVer]
+    add             r5, r4
+%else
+    lea             r5, [tab_LumaCoeffVer + r4]
 %endif
 
-    packssdw        m6, m7
-    vpermq          m6, m6, 11011000b
+    lea             r4, [r1 * 3]
+    sub             r0, r4
 %ifidn %1,pp
-    CLIPW           m6, m5, m3
+    vbroadcasti128  m14, [pd_32]
 %elifidn %1, sp
-    CLIPW           m6, m5, m3
+    vbroadcasti128  m14, [INTERP_OFFSET_SP]
+%else
+    vbroadcasti128  m14, [INTERP_OFFSET_PS]
 %endif
-    vextracti128    xm7, m6, 1
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r6], xm7
-
-    movu            xm1, [r0 + r4]                  ; m1 = row 15
+    mova            m13, [pw_pixel_max]
+    pxor            m12, m12
+    lea             r6, [r3 * 3]
+    mov             r9d, 2
+.loopW:
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
     punpckhwd       xm2, xm0, xm1
     punpcklwd       xm0, xm1
     vinserti128     m0, m0, xm2, 1
-    pmaddwd         m2, m0, [r5 + 3 * mmsize]
-    paddd           m8, m2
-    pmaddwd         m2, m0, [r5 + 2 * mmsize]
-    paddd           m10, m2
-    pmaddwd         m2, m0, [r5 + 1 * mmsize]
-    paddd           m12, m2
     pmaddwd         m0, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm2, [r0]                       ; m2 = row 16
-    punpckhwd       xm6, xm1, xm2
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
     punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm6, 1
-    pmaddwd         m6, m1, [r5 + 3 * mmsize]
-    paddd           m9, m6
-    pmaddwd         m6, m1, [r5 + 2 * mmsize]
-    paddd           m11, m6
-    pmaddwd         m6, m1, [r5 + 1 * mmsize]
-    paddd           m13, m6
+    vinserti128     m1, m1, xm3, 1
     pmaddwd         m1, [r5]
-    movu            xm6, [r0 + r1]                  ; m6 = row 17
-    punpckhwd       xm4, xm2, xm6
-    punpcklwd       xm2, xm6
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
     vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 3 * mmsize]
-    paddd           m10, m4
-    pmaddwd         m4, m2, [r5 + 2 * mmsize]
-    paddd           m12, m4
-    pmaddwd         m2, [r5 + 1 * mmsize]
-    paddd           m0, m2
-    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
-    punpckhwd       xm2, xm6, xm4
-    punpcklwd       xm6, xm4
-    vinserti128     m6, m6, xm2, 1
-    pmaddwd         m2, m6, [r5 + 3 * mmsize]
-    paddd           m11, m2
-    pmaddwd         m2, m6, [r5 + 2 * mmsize]
-    paddd           m13, m2
-    pmaddwd         m6, [r5 + 1 * mmsize]
-    paddd           m1, m6
-    movu            xm2, [r0 + r4]                  ; m2 = row 19
-    punpckhwd       xm6, xm4, xm2
-    punpcklwd       xm4, xm2
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 3 * mmsize]
-    paddd           m12, m6
-    pmaddwd         m4, [r5 + 2 * mmsize]
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
     paddd           m0, m4
-    lea             r0, [r0 + r1 * 4]
-    movu            xm6, [r0]                       ; m6 = row 20
-    punpckhwd       xm7, xm2, xm6
-    punpcklwd       xm2, xm6
-    vinserti128     m2, m2, xm7, 1
-    pmaddwd         m7, m2, [r5 + 3 * mmsize]
-    paddd           m13, m7
-    pmaddwd         m2, [r5 + 2 * mmsize]
-    paddd           m1, m2
-    movu            xm7, [r0 + r1]                  ; m7 = row 21
-    punpckhwd       xm2, xm6, xm7
-    punpcklwd       xm6, xm7
-    vinserti128     m6, m6, xm2, 1
-    pmaddwd         m6, [r5 + 3 * mmsize]
-    paddd           m0, m6
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 22
-    punpckhwd       xm6, xm7, xm2
-    punpcklwd       xm7, xm2
-    vinserti128     m7, m7, xm6, 1
-    pmaddwd         m7, [r5 + 3 * mmsize]
-    paddd           m1, m7
-
-%ifidn %1,ss
-    psrad           m8, 6
-    psrad           m9, 6
-    psrad           m10, 6
-    psrad           m11, 6
-    psrad           m12, 6
-    psrad           m13, 6
-    psrad           m0, 6
-    psrad           m1, 6
-%else
-    paddd           m8, m14
-    paddd           m9, m14
-    paddd           m10, m14
-    paddd           m11, m14
-    paddd           m12, m14
-    paddd           m13, m14
-    paddd           m0, m14
-    paddd           m1, m14
-%ifidn %1,pp
-    psrad           m8, INTERP_SHIFT_PP
-    psrad           m9, INTERP_SHIFT_PP
-    psrad           m10, INTERP_SHIFT_PP
-    psrad           m11, INTERP_SHIFT_PP
-    psrad           m12, INTERP_SHIFT_PP
-    psrad           m13, INTERP_SHIFT_PP
-    psrad           m0, INTERP_SHIFT_PP
-    psrad           m1, INTERP_SHIFT_PP
-%elifidn %1, sp
-    psrad           m8, INTERP_SHIFT_SP
-    psrad           m9, INTERP_SHIFT_SP
-    psrad           m10, INTERP_SHIFT_SP
-    psrad           m11, INTERP_SHIFT_SP
-    psrad           m12, INTERP_SHIFT_SP
-    psrad           m13, INTERP_SHIFT_SP
-    psrad           m0, INTERP_SHIFT_SP
-    psrad           m1, INTERP_SHIFT_SP
-%else
-    psrad           m8, INTERP_SHIFT_PS
-    psrad           m9, INTERP_SHIFT_PS
-    psrad           m10, INTERP_SHIFT_PS
-    psrad           m11, INTERP_SHIFT_PS
-    psrad           m12, INTERP_SHIFT_PS
-    psrad           m13, INTERP_SHIFT_PS
-    psrad           m0, INTERP_SHIFT_PS
-    psrad           m1, INTERP_SHIFT_PS
-%endif
-%endif
-
-    packssdw        m8, m9
-    packssdw        m10, m11
-    packssdw        m12, m13
-    packssdw        m0, m1
-    vpermq          m8, m8, 11011000b
-    vpermq          m10, m10, 11011000b
-    vpermq          m12, m12, 11011000b
-    vpermq          m0, m0, 11011000b
-%ifidn %1,pp
-    CLIPW           m8, m5, m3
-    CLIPW           m10, m5, m3
-    CLIPW           m12, m5, m3
-    CLIPW           m0, m5, m3
-%elifidn %1, sp
-    CLIPW           m8, m5, m3
-    CLIPW           m10, m5, m3
-    CLIPW           m12, m5, m3
-    CLIPW           m0, m5, m3
-%endif
-    vextracti128    xm9, m8, 1
-    vextracti128    xm11, m10, 1
-    vextracti128    xm13, m12, 1
-    vextracti128    xm1, m0, 1
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm8
-    movu            [r2 + r3], xm9
-    movu            [r2 + r3 * 2], xm10
-    movu            [r2 + r6], xm11
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm12
-    movu            [r2 + r3], xm13
-    movu            [r2 + r3 * 2], xm0
-    movu            [r2 + r6], xm1
-    lea             r2, [r2 + r3 * 4]
-    sub             r0, r7
-    dec             r8d
-    jnz             .loopH
-    RET
-%endif
-%endmacro
-
-FILTER_VER_LUMA_AVX2_8xN pp, 16
-FILTER_VER_LUMA_AVX2_8xN pp, 32
-FILTER_VER_LUMA_AVX2_8xN ps, 16
-FILTER_VER_LUMA_AVX2_8xN ps, 32
-FILTER_VER_LUMA_AVX2_8xN sp, 16
-FILTER_VER_LUMA_AVX2_8xN sp, 32
-FILTER_VER_LUMA_AVX2_8xN ss, 16
-FILTER_VER_LUMA_AVX2_8xN ss, 32
-
-%macro PROCESS_LUMA_AVX2_W8_8R 1
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m2, [r5]
-    lea             r7, [r0 + r1 * 4]
-    movu            xm4, [r7]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    paddd           m1, m5
-    pmaddwd         m3, [r5]
-    movu            xm5, [r7 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 2 * mmsize]
+    pmaddwd         m2, [r5]
+    lea             r7, [r0 + r1 * 4]
+    movu            xm4, [r7]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    paddd           m1, m5
+    pmaddwd         m3, [r5]
+    movu            xm5, [r7 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 2 * mmsize]
     paddd           m0, m6
     pmaddwd         m6, m4, [r5 + 1 * mmsize]
     paddd           m2, m6
@@ -3752,175 +2399,235 @@ FILTER_VER_LUMA_AVX2_8xN ss, 32
     paddd           m2, m10
     pmaddwd         m10, m8, [r5 + 2 * mmsize]
     paddd           m4, m10
-    pmaddwd         m8, [r5 + 1 * mmsize]
-    paddd           m6, m8
+    pmaddwd         m10, m8, [r5 + 1 * mmsize]
+    paddd           m6, m10
+    pmaddwd         m8, [r5]
     movu            xm10, [r7 + r1 * 2]             ; m10 = row 10
-    punpckhwd       xm8, xm9, xm10
+    punpckhwd       xm11, xm9, xm10
     punpcklwd       xm9, xm10
-    vinserti128     m9, m9, xm8, 1
-    pmaddwd         m8, m9, [r5 + 3 * mmsize]
-    paddd           m3, m8
-    pmaddwd         m8, m9, [r5 + 2 * mmsize]
-    paddd           m5, m8
-    pmaddwd         m9, [r5 + 1 * mmsize]
-    paddd           m7, m9
-    movu            xm8, [r7 + r4]                  ; m8 = row 11
-    punpckhwd       xm9, xm10, xm8
-    punpcklwd       xm10, xm8
-    vinserti128     m10, m10, xm9, 1
-    pmaddwd         m9, m10, [r5 + 3 * mmsize]
-    paddd           m4, m9
-    pmaddwd         m10, [r5 + 2 * mmsize]
-    paddd           m6, m10
-    lea             r7, [r7 + r1 * 4]
-    movu            xm9, [r7]                       ; m9 = row 12
-    punpckhwd       xm10, xm8, xm9
-    punpcklwd       xm8, xm9
-    vinserti128     m8, m8, xm10, 1
-    pmaddwd         m10, m8, [r5 + 3 * mmsize]
-    paddd           m5, m10
-    pmaddwd         m8, [r5 + 2 * mmsize]
-    paddd           m7, m8
+    vinserti128     m9, m9, xm11, 1
+    pmaddwd         m11, m9, [r5 + 3 * mmsize]
+    paddd           m3, m11
+    pmaddwd         m11, m9, [r5 + 2 * mmsize]
+    paddd           m5, m11
+    pmaddwd         m11, m9, [r5 + 1 * mmsize]
+    paddd           m7, m11
+    pmaddwd         m9, [r5]
 
 %ifidn %1,ss
     psrad           m0, 6
     psrad           m1, 6
     psrad           m2, 6
     psrad           m3, 6
-    psrad           m4, 6
-    psrad           m5, 6
 %else
-    paddd           m0, m11
-    paddd           m1, m11
-    paddd           m2, m11
-    paddd           m3, m11
-    paddd           m4, m11
-    paddd           m5, m11
+    paddd           m0, m14
+    paddd           m1, m14
+    paddd           m2, m14
+    paddd           m3, m14
 %ifidn %1,pp
     psrad           m0, INTERP_SHIFT_PP
     psrad           m1, INTERP_SHIFT_PP
     psrad           m2, INTERP_SHIFT_PP
     psrad           m3, INTERP_SHIFT_PP
-    psrad           m4, INTERP_SHIFT_PP
-    psrad           m5, INTERP_SHIFT_PP
 %elifidn %1, sp
     psrad           m0, INTERP_SHIFT_SP
     psrad           m1, INTERP_SHIFT_SP
     psrad           m2, INTERP_SHIFT_SP
     psrad           m3, INTERP_SHIFT_SP
-    psrad           m4, INTERP_SHIFT_SP
-    psrad           m5, INTERP_SHIFT_SP
 %else
     psrad           m0, INTERP_SHIFT_PS
     psrad           m1, INTERP_SHIFT_PS
     psrad           m2, INTERP_SHIFT_PS
     psrad           m3, INTERP_SHIFT_PS
-    psrad           m4, INTERP_SHIFT_PS
-    psrad           m5, INTERP_SHIFT_PS
 %endif
 %endif
 
     packssdw        m0, m1
     packssdw        m2, m3
-    packssdw        m4, m5
     vpermq          m0, m0, 11011000b
     vpermq          m2, m2, 11011000b
-    vpermq          m4, m4, 11011000b
-    pxor            m8, m8
 %ifidn %1,pp
-    CLIPW           m0, m8, m12
-    CLIPW           m2, m8, m12
-    CLIPW           m4, m8, m12
+    CLIPW           m0, m12, m13
+    CLIPW           m2, m12, m13
 %elifidn %1, sp
-    CLIPW           m0, m8, m12
-    CLIPW           m2, m8, m12
-    CLIPW           m4, m8, m12
+    CLIPW           m0, m12, m13
+    CLIPW           m2, m12, m13
 %endif
 
     vextracti128    xm1, m0, 1
     vextracti128    xm3, m2, 1
-    vextracti128    xm5, m4, 1
     movu            [r2], xm0
     movu            [r2 + r3], xm1
     movu            [r2 + r3 * 2], xm2
     movu            [r2 + r6], xm3
-    lea             r8, [r2 + r3 * 4]
-    movu            [r8], xm4
-    movu            [r8 + r3], xm5
 
-    movu            xm10, [r7 + r1]                 ; m10 = row 13
-    punpckhwd       xm0, xm9, xm10
-    punpcklwd       xm9, xm10
-    vinserti128     m9, m9, xm0, 1
-    pmaddwd         m9, [r5 + 3 * mmsize]
-    paddd           m6, m9
+    movu            xm11, [r7 + r4]                 ; m11 = row 11
+    punpckhwd       xm0, xm10, xm11
+    punpcklwd       xm10, xm11
+    vinserti128     m10, m10, xm0, 1
+    pmaddwd         m0, m10, [r5 + 3 * mmsize]
+    paddd           m4, m0
+    pmaddwd         m0, m10, [r5 + 2 * mmsize]
+    paddd           m6, m0
+    pmaddwd         m0, m10, [r5 + 1 * mmsize]
+    paddd           m8, m0
+    pmaddwd         m10, [r5]
+    lea             r7, [r7 + r1 * 4]
+    movu            xm0, [r7]                      ; m0 = row 12
+    punpckhwd       xm1, xm11, xm0
+    punpcklwd       xm11, xm0
+    vinserti128     m11, m11, xm1, 1
+    pmaddwd         m1, m11, [r5 + 3 * mmsize]
+    paddd           m5, m1
+    pmaddwd         m1, m11, [r5 + 2 * mmsize]
+    paddd           m7, m1
+    pmaddwd         m1, m11, [r5 + 1 * mmsize]
+    paddd           m9, m1
+    pmaddwd         m11, [r5]
+    movu            xm2, [r7 + r1]                 ; m2 = row 13
+    punpckhwd       xm1, xm0, xm2
+    punpcklwd       xm0, xm2
+    vinserti128     m0, m0, xm1, 1
+    pmaddwd         m1, m0, [r5 + 3 * mmsize]
+    paddd           m6, m1
+    pmaddwd         m1, m0, [r5 + 2 * mmsize]
+    paddd           m8, m1
+    pmaddwd         m0, [r5 + 1 * mmsize]
+    paddd           m10, m0
     movu            xm0, [r7 + r1 * 2]              ; m0 = row 14
-    punpckhwd       xm1, xm10, xm0
-    punpcklwd       xm10, xm0
-    vinserti128     m10, m10, xm1, 1
-    pmaddwd         m10, [r5 + 3 * mmsize]
-    paddd           m7, m10
+    punpckhwd       xm1, xm2, xm0
+    punpcklwd       xm2, xm0
+    vinserti128     m2, m2, xm1, 1
+    pmaddwd         m1, m2, [r5 + 3 * mmsize]
+    paddd           m7, m1
+    pmaddwd         m1, m2, [r5 + 2 * mmsize]
+    paddd           m9, m1
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m11, m2
 
 %ifidn %1,ss
+    psrad           m4, 6
+    psrad           m5, 6
     psrad           m6, 6
     psrad           m7, 6
 %else
-    paddd           m6, m11
-    paddd           m7, m11
+    paddd           m4, m14
+    paddd           m5, m14
+    paddd           m6, m14
+    paddd           m7, m14
 %ifidn %1,pp
+    psrad           m4, INTERP_SHIFT_PP
+    psrad           m5, INTERP_SHIFT_PP
     psrad           m6, INTERP_SHIFT_PP
     psrad           m7, INTERP_SHIFT_PP
 %elifidn %1, sp
+    psrad           m4, INTERP_SHIFT_SP
+    psrad           m5, INTERP_SHIFT_SP
     psrad           m6, INTERP_SHIFT_SP
     psrad           m7, INTERP_SHIFT_SP
 %else
+    psrad           m4, INTERP_SHIFT_PS
+    psrad           m5, INTERP_SHIFT_PS
     psrad           m6, INTERP_SHIFT_PS
     psrad           m7, INTERP_SHIFT_PS
 %endif
 %endif
 
+    packssdw        m4, m5
     packssdw        m6, m7
+    vpermq          m4, m4, 11011000b
     vpermq          m6, m6, 11011000b
 %ifidn %1,pp
-    CLIPW           m6, m8, m12
+    CLIPW           m4, m12, m13
+    CLIPW           m6, m12, m13
 %elifidn %1, sp
-    CLIPW           m6, m8, m12
+    CLIPW           m4, m12, m13
+    CLIPW           m6, m12, m13
 %endif
+    lea             r8, [r2 + r3 * 4]
+    vextracti128    xm1, m4, 1
     vextracti128    xm7, m6, 1
+    movu            [r8], xm4
+    movu            [r8 + r3], xm1
     movu            [r8 + r3 * 2], xm6
     movu            [r8 + r6], xm7
-%endmacro
 
-%macro FILTER_VER_LUMA_AVX2_Nx8 2
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_%2x8, 4, 10, 13
-    mov             r4d, r4m
-    shl             r4d, 7
-    add             r1d, r1d
-    add             r3d, r3d
+    movu            xm1, [r7 + r4]                  ; m1 = row 15
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m2, m0, [r5 + 3 * mmsize]
+    paddd           m8, m2
+    pmaddwd         m0, [r5 + 2 * mmsize]
+    paddd           m10, m0
+    lea             r7, [r7 + r1 * 4]
+    movu            xm2, [r7]                       ; m2 = row 16
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m3, m1, [r5 + 3 * mmsize]
+    paddd           m9, m3
+    pmaddwd         m1, [r5 + 2 * mmsize]
+    paddd           m11, m1
+    movu            xm3, [r7 + r1]                  ; m3 = row 17
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m2, [r5 + 3 * mmsize]
+    paddd           m10, m2
+    movu            xm4, [r7 + r1 * 2]              ; m4 = row 18
+    punpckhwd       xm2, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm2, 1
+    pmaddwd         m3, [r5 + 3 * mmsize]
+    paddd           m11, m3
 
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer]
-    add             r5, r4
+%ifidn %1,ss
+    psrad           m8, 6
+    psrad           m9, 6
+    psrad           m10, 6
+    psrad           m11, 6
 %else
-    lea             r5, [tab_LumaCoeffVer + r4]
+    paddd           m8, m14
+    paddd           m9, m14
+    paddd           m10, m14
+    paddd           m11, m14
+%ifidn %1,pp
+    psrad           m8, INTERP_SHIFT_PP
+    psrad           m9, INTERP_SHIFT_PP
+    psrad           m10, INTERP_SHIFT_PP
+    psrad           m11, INTERP_SHIFT_PP
+%elifidn %1, sp
+    psrad           m8, INTERP_SHIFT_SP
+    psrad           m9, INTERP_SHIFT_SP
+    psrad           m10, INTERP_SHIFT_SP
+    psrad           m11, INTERP_SHIFT_SP
+%else
+    psrad           m8, INTERP_SHIFT_PS
+    psrad           m9, INTERP_SHIFT_PS
+    psrad           m10, INTERP_SHIFT_PS
+    psrad           m11, INTERP_SHIFT_PS
+%endif
 %endif
 
-    lea             r4, [r1 * 3]
-    sub             r0, r4
+    packssdw        m8, m9
+    packssdw        m10, m11
+    vpermq          m8, m8, 11011000b
+    vpermq          m10, m10, 11011000b
 %ifidn %1,pp
-    vbroadcasti128  m11, [pd_32]
+    CLIPW           m8, m12, m13
+    CLIPW           m10, m12, m13
 %elifidn %1, sp
-    vbroadcasti128  m11, [INTERP_OFFSET_SP]
-%else
-    vbroadcasti128  m11, [INTERP_OFFSET_PS]
+    CLIPW           m8, m12, m13
+    CLIPW           m10, m12, m13
 %endif
-    mova            m12, [pw_pixel_max]
-    lea             r6, [r3 * 3]
-    mov             r9d, %2 / 8
-.loopW:
-    PROCESS_LUMA_AVX2_W8_8R %1
+    vextracti128    xm9, m8, 1
+    vextracti128    xm11, m10, 1
+    lea             r8, [r8 + r3 * 4]
+    movu            [r8], xm8
+    movu            [r8 + r3], xm9
+    movu            [r8 + r3 * 2], xm10
+    movu            [r8 + r6], xm11
     add             r2, 16
     add             r0, 16
     dec             r9d
@@ -3929,19 +2636,14 @@ cglobal interp_8tap_vert_%1_%2x8, 4, 10, 13
 %endif
 %endmacro
 
-FILTER_VER_LUMA_AVX2_Nx8 pp, 32
-FILTER_VER_LUMA_AVX2_Nx8 pp, 16
-FILTER_VER_LUMA_AVX2_Nx8 ps, 32
-FILTER_VER_LUMA_AVX2_Nx8 ps, 16
-FILTER_VER_LUMA_AVX2_Nx8 sp, 32
-FILTER_VER_LUMA_AVX2_Nx8 sp, 16
-FILTER_VER_LUMA_AVX2_Nx8 ss, 32
-FILTER_VER_LUMA_AVX2_Nx8 ss, 16
+FILTER_VER_LUMA_AVX2_16x12 pp
+FILTER_VER_LUMA_AVX2_16x12 ps
+FILTER_VER_LUMA_AVX2_16x12 sp
+FILTER_VER_LUMA_AVX2_16x12 ss
 
-%macro FILTER_VER_LUMA_AVX2_32x24 1
+%macro FILTER_VER_LUMA_AVX2_4x8 1
 INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_32x24, 4, 10, 15
+cglobal interp_8tap_vert_%1_4x8, 4, 7, 8
     mov             r4d, r4m
     shl             r4d, 7
     add             r1d, r1d
@@ -3956,595 +2658,413 @@ cglobal interp_8tap_vert_%1_32x24, 4, 10, 15
 
     lea             r4, [r1 * 3]
     sub             r0, r4
+
 %ifidn %1,pp
-    vbroadcasti128  m14, [pd_32]
+    vbroadcasti128  m7, [pd_32]
 %elifidn %1, sp
-    vbroadcasti128  m14, [INTERP_OFFSET_SP]
+    vbroadcasti128  m7, [INTERP_OFFSET_SP]
 %else
-    vbroadcasti128  m14, [INTERP_OFFSET_PS]
+    vbroadcasti128  m7, [INTERP_OFFSET_PS]
 %endif
     lea             r6, [r3 * 3]
-    mov             r9d, 4
-.loopW:
-    PROCESS_LUMA_AVX2_W8_16R %1
-    add             r2, 16
-    add             r0, 16
-    dec             r9d
-    jnz             .loopW
-    lea             r9, [r1 * 4]
-    sub             r7, r9
-    lea             r0, [r7 - 48]
-    lea             r2, [r8 + r3 * 4 - 48]
-    mova            m11, m14
-    mova            m12, m3
-    mov             r9d, 4
-.loop:
-    PROCESS_LUMA_AVX2_W8_8R %1
-    add             r2, 16
-    add             r0, 16
-    dec             r9d
-    jnz             .loop
-    RET
-%endif
-%endmacro
-
-FILTER_VER_LUMA_AVX2_32x24 pp
-FILTER_VER_LUMA_AVX2_32x24 ps
-FILTER_VER_LUMA_AVX2_32x24 sp
-FILTER_VER_LUMA_AVX2_32x24 ss
 
-%macro PROCESS_LUMA_AVX2_W8_4R 1
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
+    movq            xm0, [r0]
+    movq            xm1, [r0 + r1]
     punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
+    movq            xm2, [r0 + r1 * 2]
     punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
+    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
+    pmaddwd         m0, [r5]
+    movq            xm3, [r0 + r4]
     punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
+    lea             r0, [r0 + 4 * r1]
+    movq            xm4, [r0]
     punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    paddd           m1, m5
-    pmaddwd         m3, [r5]
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 2 * mmsize]
-    paddd           m0, m6
-    pmaddwd         m4, [r5 + 1 * mmsize]
-    paddd           m2, m4
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm4, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm4, 1
-    pmaddwd         m4, m5, [r5 + 2 * mmsize]
-    paddd           m1, m4
-    pmaddwd         m5, [r5 + 1 * mmsize]
-    paddd           m3, m5
-    movu            xm4, [r0 + r4]                  ; m4 = row 7
-    punpckhwd       xm5, xm6, xm4
-    punpcklwd       xm6, xm4
-    vinserti128     m6, m6, xm5, 1
-    pmaddwd         m5, m6, [r5 + 3 * mmsize]
+    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
+    pmaddwd         m5, m2, [r5 + 1 * mmsize]
+    pmaddwd         m2, [r5]
     paddd           m0, m5
-    pmaddwd         m6, [r5 + 2 * mmsize]
-    paddd           m2, m6
-    lea             r0, [r0 + r1 * 4]
-    movu            xm5, [r0]                       ; m5 = row 8
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 3 * mmsize]
-    paddd           m1, m6
-    pmaddwd         m4, [r5 + 2 * mmsize]
-    paddd           m3, m4
-    movu            xm6, [r0 + r1]                  ; m6 = row 9
-    punpckhwd       xm4, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm4, 1
-    pmaddwd         m5, [r5 + 3 * mmsize]
+    movq            xm3, [r0 + r1]
+    punpcklwd       xm4, xm3
+    movq            xm1, [r0 + r1 * 2]
+    punpcklwd       xm3, xm1
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
+    pmaddwd         m5, m4, [r5 + 2 * mmsize]
+    paddd           m0, m5
+    pmaddwd         m5, m4, [r5 + 1 * mmsize]
     paddd           m2, m5
-    movu            xm4, [r0 + r1 * 2]              ; m4 = row 10
-    punpckhwd       xm5, xm6, xm4
-    punpcklwd       xm6, xm4
-    vinserti128     m6, m6, xm5, 1
-    pmaddwd         m6, [r5 + 3 * mmsize]
-    paddd           m3, m6
+    pmaddwd         m4, [r5]
+    movq            xm3, [r0 + r4]
+    punpcklwd       xm1, xm3
+    lea             r0, [r0 + 4 * r1]
+    movq            xm6, [r0]
+    punpcklwd       xm3, xm6
+    vinserti128     m1, m1, xm3, 1                  ; m1 = [8 7 7 6]
+    pmaddwd         m5, m1, [r5 + 3 * mmsize]
+    paddd           m0, m5
+    pmaddwd         m5, m1, [r5 + 2 * mmsize]
+    paddd           m2, m5
+    pmaddwd         m5, m1, [r5 + 1 * mmsize]
+    paddd           m4, m5
+    pmaddwd         m1, [r5]
+    movq            xm3, [r0 + r1]
+    punpcklwd       xm6, xm3
+    movq            xm5, [r0 + 2 * r1]
+    punpcklwd       xm3, xm5
+    vinserti128     m6, m6, xm3, 1                  ; m6 = [A 9 9 8]
+    pmaddwd         m3, m6, [r5 + 3 * mmsize]
+    paddd           m2, m3
+    pmaddwd         m3, m6, [r5 + 2 * mmsize]
+    paddd           m4, m3
+    pmaddwd         m6, [r5 + 1 * mmsize]
+    paddd           m1, m6
 
 %ifidn %1,ss
     psrad           m0, 6
-    psrad           m1, 6
     psrad           m2, 6
-    psrad           m3, 6
 %else
     paddd           m0, m7
-    paddd           m1, m7
     paddd           m2, m7
-    paddd           m3, m7
 %ifidn %1,pp
     psrad           m0, INTERP_SHIFT_PP
-    psrad           m1, INTERP_SHIFT_PP
     psrad           m2, INTERP_SHIFT_PP
-    psrad           m3, INTERP_SHIFT_PP
 %elifidn %1, sp
     psrad           m0, INTERP_SHIFT_SP
-    psrad           m1, INTERP_SHIFT_SP
     psrad           m2, INTERP_SHIFT_SP
-    psrad           m3, INTERP_SHIFT_SP
 %else
     psrad           m0, INTERP_SHIFT_PS
-    psrad           m1, INTERP_SHIFT_PS
     psrad           m2, INTERP_SHIFT_PS
-    psrad           m3, INTERP_SHIFT_PS
 %endif
 %endif
 
-    packssdw        m0, m1
-    packssdw        m2, m3
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
-    pxor            m4, m4
+    packssdw        m0, m2
+    pxor            m6, m6
+    mova            m3, [pw_pixel_max]
 %ifidn %1,pp
-    CLIPW           m0, m4, [pw_pixel_max]
-    CLIPW           m2, m4, [pw_pixel_max]
+    CLIPW           m0, m6, m3
 %elifidn %1, sp
-    CLIPW           m0, m4, [pw_pixel_max]
-    CLIPW           m2, m4, [pw_pixel_max]
+    CLIPW           m0, m6, m3
 %endif
 
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
-%endmacro
+    vextracti128    xm2, m0, 1
+    movq            [r2], xm0
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r6], xm2
 
-%macro FILTER_VER_LUMA_AVX2_16x4 1
-INIT_YMM avx2
-cglobal interp_8tap_vert_%1_16x4, 4, 7, 8, 0-gprsize
-    mov             r4d, r4m
-    shl             r4d, 7
-    add             r1d, r1d
-    add             r3d, r3d
+    movq            xm2, [r0 + r4]
+    punpcklwd       xm5, xm2
+    lea             r0, [r0 + 4 * r1]
+    movq            xm0, [r0]
+    punpcklwd       xm2, xm0
+    vinserti128     m5, m5, xm2, 1                  ; m5 = [C B B A]
+    pmaddwd         m2, m5, [r5 + 3 * mmsize]
+    paddd           m4, m2
+    pmaddwd         m5, [r5 + 2 * mmsize]
+    paddd           m1, m5
+    movq            xm2, [r0 + r1]
+    punpcklwd       xm0, xm2
+    movq            xm5, [r0 + 2 * r1]
+    punpcklwd       xm2, xm5
+    vinserti128     m0, m0, xm2, 1                  ; m0 = [E D D C]
+    pmaddwd         m0, [r5 + 3 * mmsize]
+    paddd           m1, m0
 
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer]
-    add             r5, r4
+%ifidn %1,ss
+    psrad           m4, 6
+    psrad           m1, 6
 %else
-    lea             r5, [tab_LumaCoeffVer + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
+    paddd           m4, m7
+    paddd           m1, m7
 %ifidn %1,pp
-    vbroadcasti128  m7, [pd_32]
+    psrad           m4, INTERP_SHIFT_PP
+    psrad           m1, INTERP_SHIFT_PP
 %elifidn %1, sp
-    vbroadcasti128  m7, [INTERP_OFFSET_SP]
+    psrad           m4, INTERP_SHIFT_SP
+    psrad           m1, INTERP_SHIFT_SP
 %else
-    vbroadcasti128  m7, [INTERP_OFFSET_PS]
+    psrad           m4, INTERP_SHIFT_PS
+    psrad           m1, INTERP_SHIFT_PS
 %endif
-    mov             dword [rsp], 2
-.loopW:
-    PROCESS_LUMA_AVX2_W8_4R %1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    lea             r6, [r3 * 3]
-    movu            [r2 + r6], xm3
-    add             r2, 16
-    lea             r6, [8 * r1 - 16]
-    sub             r0, r6
-    dec             dword [rsp]
-    jnz             .loopW
-    RET
-%endmacro
-
-FILTER_VER_LUMA_AVX2_16x4 pp
-FILTER_VER_LUMA_AVX2_16x4 ps
-FILTER_VER_LUMA_AVX2_16x4 sp
-FILTER_VER_LUMA_AVX2_16x4 ss
-
-%macro FILTER_VER_LUMA_AVX2_8x4 1
-INIT_YMM avx2
-cglobal interp_8tap_vert_%1_8x4, 4, 6, 8
-    mov             r4d, r4m
-    shl             r4d, 7
-    add             r1d, r1d
-    add             r3d, r3d
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer + r4]
 %endif
 
-    lea             r4, [r1 * 3]
-    sub             r0, r4
+    packssdw        m4, m1
 %ifidn %1,pp
-    vbroadcasti128  m7, [pd_32]
+    CLIPW           m4, m6, m3
 %elifidn %1, sp
-    vbroadcasti128  m7, [INTERP_OFFSET_SP]
-%else
-    vbroadcasti128  m7, [INTERP_OFFSET_PS]
+    CLIPW           m4, m6, m3
 %endif
 
-    PROCESS_LUMA_AVX2_W8_4R %1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    lea             r4, [r3 * 3]
-    movu            [r2 + r4], xm3
+    vextracti128    xm1, m4, 1
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm4
+    movq            [r2 + r3], xm1
+    movhps          [r2 + r3 * 2], xm4
+    movhps          [r2 + r6], xm1
     RET
 %endmacro
 
-FILTER_VER_LUMA_AVX2_8x4 pp
-FILTER_VER_LUMA_AVX2_8x4 ps
-FILTER_VER_LUMA_AVX2_8x4 sp
-FILTER_VER_LUMA_AVX2_8x4 ss
+FILTER_VER_LUMA_AVX2_4x8 pp
+FILTER_VER_LUMA_AVX2_4x8 ps
+FILTER_VER_LUMA_AVX2_4x8 sp
+FILTER_VER_LUMA_AVX2_4x8 ss
 
-%macro FILTER_VER_LUMA_AVX2_16x12 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_16x12, 4, 10, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-    add             r1d, r1d
-    add             r3d, r3d
+%macro PROCESS_LUMA_AVX2_W4_16R 1
+    movq            xm0, [r0]
+    movq            xm1, [r0 + r1]
+    punpcklwd       xm0, xm1
+    movq            xm2, [r0 + r1 * 2]
+    punpcklwd       xm1, xm2
+    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
+    pmaddwd         m0, [r5]
+    movq            xm3, [r0 + r4]
+    punpcklwd       xm2, xm3
+    lea             r0, [r0 + 4 * r1]
+    movq            xm4, [r0]
+    punpcklwd       xm3, xm4
+    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
+    pmaddwd         m5, m2, [r5 + 1 * mmsize]
+    pmaddwd         m2, [r5]
+    paddd           m0, m5
+    movq            xm3, [r0 + r1]
+    punpcklwd       xm4, xm3
+    movq            xm1, [r0 + r1 * 2]
+    punpcklwd       xm3, xm1
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
+    pmaddwd         m5, m4, [r5 + 2 * mmsize]
+    paddd           m0, m5
+    pmaddwd         m5, m4, [r5 + 1 * mmsize]
+    paddd           m2, m5
+    pmaddwd         m4, [r5]
+    movq            xm3, [r0 + r4]
+    punpcklwd       xm1, xm3
+    lea             r0, [r0 + 4 * r1]
+    movq            xm6, [r0]
+    punpcklwd       xm3, xm6
+    vinserti128     m1, m1, xm3, 1                  ; m1 = [8 7 7 6]
+    pmaddwd         m5, m1, [r5 + 3 * mmsize]
+    paddd           m0, m5
+    pmaddwd         m5, m1, [r5 + 2 * mmsize]
+    paddd           m2, m5
+    pmaddwd         m5, m1, [r5 + 1 * mmsize]
+    paddd           m4, m5
+    pmaddwd         m1, [r5]
+    movq            xm3, [r0 + r1]
+    punpcklwd       xm6, xm3
+    movq            xm5, [r0 + 2 * r1]
+    punpcklwd       xm3, xm5
+    vinserti128     m6, m6, xm3, 1                  ; m6 = [10 9 9 8]
+    pmaddwd         m3, m6, [r5 + 3 * mmsize]
+    paddd           m2, m3
+    pmaddwd         m3, m6, [r5 + 2 * mmsize]
+    paddd           m4, m3
+    pmaddwd         m3, m6, [r5 + 1 * mmsize]
+    paddd           m1, m3
+    pmaddwd         m6, [r5]
 
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer]
-    add             r5, r4
+%ifidn %1,ss
+    psrad           m0, 6
+    psrad           m2, 6
 %else
-    lea             r5, [tab_LumaCoeffVer + r4]
+    paddd           m0, m7
+    paddd           m2, m7
+%ifidn %1,pp
+    psrad           m0, INTERP_SHIFT_PP
+    psrad           m2, INTERP_SHIFT_PP
+%elifidn %1, sp
+    psrad           m0, INTERP_SHIFT_SP
+    psrad           m2, INTERP_SHIFT_SP
+%else
+    psrad           m0, INTERP_SHIFT_PS
+    psrad           m2, INTERP_SHIFT_PS
+%endif
 %endif
 
-    lea             r4, [r1 * 3]
-    sub             r0, r4
+    packssdw        m0, m2
+    pxor            m3, m3
 %ifidn %1,pp
-    vbroadcasti128  m14, [pd_32]
+    CLIPW           m0, m3, [pw_pixel_max]
 %elifidn %1, sp
-    vbroadcasti128  m14, [INTERP_OFFSET_SP]
-%else
-    vbroadcasti128  m14, [INTERP_OFFSET_PS]
+    CLIPW           m0, m3, [pw_pixel_max]
 %endif
-    mova            m13, [pw_pixel_max]
-    pxor            m12, m12
-    lea             r6, [r3 * 3]
-    mov             r9d, 2
-.loopW:
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m2, [r5]
-    lea             r7, [r0 + r1 * 4]
-    movu            xm4, [r7]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    paddd           m1, m5
-    pmaddwd         m3, [r5]
-    movu            xm5, [r7 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 2 * mmsize]
-    paddd           m0, m6
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m2, m6
-    pmaddwd         m4, [r5]
-    movu            xm6, [r7 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm7, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddwd         m7, m5, [r5 + 2 * mmsize]
-    paddd           m1, m7
-    pmaddwd         m7, m5, [r5 + 1 * mmsize]
-    paddd           m3, m7
+
+    vextracti128    xm2, m0, 1
+    movq            [r2], xm0
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r6], xm2
+
+    movq            xm2, [r0 + r4]
+    punpcklwd       xm5, xm2
+    lea             r0, [r0 + 4 * r1]
+    movq            xm0, [r0]
+    punpcklwd       xm2, xm0
+    vinserti128     m5, m5, xm2, 1                  ; m5 = [12 11 11 10]
+    pmaddwd         m2, m5, [r5 + 3 * mmsize]
+    paddd           m4, m2
+    pmaddwd         m2, m5, [r5 + 2 * mmsize]
+    paddd           m1, m2
+    pmaddwd         m2, m5, [r5 + 1 * mmsize]
+    paddd           m6, m2
     pmaddwd         m5, [r5]
-    movu            xm7, [r7 + r4]                  ; m7 = row 7
-    punpckhwd       xm8, xm6, xm7
-    punpcklwd       xm6, xm7
-    vinserti128     m6, m6, xm8, 1
-    pmaddwd         m8, m6, [r5 + 3 * mmsize]
-    paddd           m0, m8
-    pmaddwd         m8, m6, [r5 + 2 * mmsize]
-    paddd           m2, m8
-    pmaddwd         m8, m6, [r5 + 1 * mmsize]
-    paddd           m4, m8
-    pmaddwd         m6, [r5]
-    lea             r7, [r7 + r1 * 4]
-    movu            xm8, [r7]                       ; m8 = row 8
-    punpckhwd       xm9, xm7, xm8
-    punpcklwd       xm7, xm8
-    vinserti128     m7, m7, xm9, 1
-    pmaddwd         m9, m7, [r5 + 3 * mmsize]
-    paddd           m1, m9
-    pmaddwd         m9, m7, [r5 + 2 * mmsize]
-    paddd           m3, m9
-    pmaddwd         m9, m7, [r5 + 1 * mmsize]
-    paddd           m5, m9
-    pmaddwd         m7, [r5]
-    movu            xm9, [r7 + r1]                  ; m9 = row 9
-    punpckhwd       xm10, xm8, xm9
-    punpcklwd       xm8, xm9
-    vinserti128     m8, m8, xm10, 1
-    pmaddwd         m10, m8, [r5 + 3 * mmsize]
-    paddd           m2, m10
-    pmaddwd         m10, m8, [r5 + 2 * mmsize]
-    paddd           m4, m10
-    pmaddwd         m10, m8, [r5 + 1 * mmsize]
-    paddd           m6, m10
-    pmaddwd         m8, [r5]
-    movu            xm10, [r7 + r1 * 2]             ; m10 = row 10
-    punpckhwd       xm11, xm9, xm10
-    punpcklwd       xm9, xm10
-    vinserti128     m9, m9, xm11, 1
-    pmaddwd         m11, m9, [r5 + 3 * mmsize]
-    paddd           m3, m11
-    pmaddwd         m11, m9, [r5 + 2 * mmsize]
-    paddd           m5, m11
-    pmaddwd         m11, m9, [r5 + 1 * mmsize]
-    paddd           m7, m11
-    pmaddwd         m9, [r5]
+    movq            xm2, [r0 + r1]
+    punpcklwd       xm0, xm2
+    movq            xm3, [r0 + 2 * r1]
+    punpcklwd       xm2, xm3
+    vinserti128     m0, m0, xm2, 1                  ; m0 = [14 13 13 12]
+    pmaddwd         m2, m0, [r5 + 3 * mmsize]
+    paddd           m1, m2
+    pmaddwd         m2, m0, [r5 + 2 * mmsize]
+    paddd           m6, m2
+    pmaddwd         m2, m0, [r5 + 1 * mmsize]
+    paddd           m5, m2
+    pmaddwd         m0, [r5]
 
 %ifidn %1,ss
-    psrad           m0, 6
+    psrad           m4, 6
     psrad           m1, 6
-    psrad           m2, 6
-    psrad           m3, 6
 %else
-    paddd           m0, m14
-    paddd           m1, m14
-    paddd           m2, m14
-    paddd           m3, m14
+    paddd           m4, m7
+    paddd           m1, m7
 %ifidn %1,pp
-    psrad           m0, INTERP_SHIFT_PP
+    psrad           m4, INTERP_SHIFT_PP
     psrad           m1, INTERP_SHIFT_PP
-    psrad           m2, INTERP_SHIFT_PP
-    psrad           m3, INTERP_SHIFT_PP
 %elifidn %1, sp
-    psrad           m0, INTERP_SHIFT_SP
+    psrad           m4, INTERP_SHIFT_SP
     psrad           m1, INTERP_SHIFT_SP
-    psrad           m2, INTERP_SHIFT_SP
-    psrad           m3, INTERP_SHIFT_SP
 %else
-    psrad           m0, INTERP_SHIFT_PS
+    psrad           m4, INTERP_SHIFT_PS
     psrad           m1, INTERP_SHIFT_PS
-    psrad           m2, INTERP_SHIFT_PS
-    psrad           m3, INTERP_SHIFT_PS
 %endif
 %endif
 
-    packssdw        m0, m1
-    packssdw        m2, m3
-    vpermq          m0, m0, 11011000b
-    vpermq          m2, m2, 11011000b
+    packssdw        m4, m1
+    pxor            m2, m2
 %ifidn %1,pp
-    CLIPW           m0, m12, m13
-    CLIPW           m2, m12, m13
+    CLIPW           m4, m2, [pw_pixel_max]
 %elifidn %1, sp
-    CLIPW           m0, m12, m13
-    CLIPW           m2, m12, m13
+    CLIPW           m4, m2, [pw_pixel_max]
 %endif
 
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm3
+    vextracti128    xm1, m4, 1
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm4
+    movq            [r2 + r3], xm1
+    movhps          [r2 + r3 * 2], xm4
+    movhps          [r2 + r6], xm1
 
-    movu            xm11, [r7 + r4]                 ; m11 = row 11
-    punpckhwd       xm0, xm10, xm11
-    punpcklwd       xm10, xm11
-    vinserti128     m10, m10, xm0, 1
-    pmaddwd         m0, m10, [r5 + 3 * mmsize]
-    paddd           m4, m0
-    pmaddwd         m0, m10, [r5 + 2 * mmsize]
-    paddd           m6, m0
-    pmaddwd         m0, m10, [r5 + 1 * mmsize]
-    paddd           m8, m0
-    pmaddwd         m10, [r5]
-    lea             r7, [r7 + r1 * 4]
-    movu            xm0, [r7]                      ; m0 = row 12
-    punpckhwd       xm1, xm11, xm0
-    punpcklwd       xm11, xm0
-    vinserti128     m11, m11, xm1, 1
-    pmaddwd         m1, m11, [r5 + 3 * mmsize]
-    paddd           m5, m1
-    pmaddwd         m1, m11, [r5 + 2 * mmsize]
-    paddd           m7, m1
-    pmaddwd         m1, m11, [r5 + 1 * mmsize]
-    paddd           m9, m1
-    pmaddwd         m11, [r5]
-    movu            xm2, [r7 + r1]                 ; m2 = row 13
-    punpckhwd       xm1, xm0, xm2
-    punpcklwd       xm0, xm2
-    vinserti128     m0, m0, xm1, 1
-    pmaddwd         m1, m0, [r5 + 3 * mmsize]
-    paddd           m6, m1
-    pmaddwd         m1, m0, [r5 + 2 * mmsize]
-    paddd           m8, m1
-    pmaddwd         m0, [r5 + 1 * mmsize]
-    paddd           m10, m0
-    movu            xm0, [r7 + r1 * 2]              ; m0 = row 14
-    punpckhwd       xm1, xm2, xm0
-    punpcklwd       xm2, xm0
-    vinserti128     m2, m2, xm1, 1
-    pmaddwd         m1, m2, [r5 + 3 * mmsize]
-    paddd           m7, m1
-    pmaddwd         m1, m2, [r5 + 2 * mmsize]
-    paddd           m9, m1
-    pmaddwd         m2, [r5 + 1 * mmsize]
-    paddd           m11, m2
+    movq            xm4, [r0 + r4]
+    punpcklwd       xm3, xm4
+    lea             r0, [r0 + 4 * r1]
+    movq            xm1, [r0]
+    punpcklwd       xm4, xm1
+    vinserti128     m3, m3, xm4, 1                  ; m3 = [16 15 15 14]
+    pmaddwd         m4, m3, [r5 + 3 * mmsize]
+    paddd           m6, m4
+    pmaddwd         m4, m3, [r5 + 2 * mmsize]
+    paddd           m5, m4
+    pmaddwd         m4, m3, [r5 + 1 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m3, [r5]
+    movq            xm4, [r0 + r1]
+    punpcklwd       xm1, xm4
+    movq            xm2, [r0 + 2 * r1]
+    punpcklwd       xm4, xm2
+    vinserti128     m1, m1, xm4, 1                  ; m1 = [18 17 17 16]
+    pmaddwd         m4, m1, [r5 + 3 * mmsize]
+    paddd           m5, m4
+    pmaddwd         m4, m1, [r5 + 2 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m1, [r5 + 1 * mmsize]
+    paddd           m3, m1
 
 %ifidn %1,ss
-    psrad           m4, 6
-    psrad           m5, 6
     psrad           m6, 6
-    psrad           m7, 6
+    psrad           m5, 6
 %else
-    paddd           m4, m14
-    paddd           m5, m14
-    paddd           m6, m14
-    paddd           m7, m14
+    paddd           m6, m7
+    paddd           m5, m7
 %ifidn %1,pp
-    psrad           m4, INTERP_SHIFT_PP
-    psrad           m5, INTERP_SHIFT_PP
     psrad           m6, INTERP_SHIFT_PP
-    psrad           m7, INTERP_SHIFT_PP
+    psrad           m5, INTERP_SHIFT_PP
 %elifidn %1, sp
-    psrad           m4, INTERP_SHIFT_SP
-    psrad           m5, INTERP_SHIFT_SP
     psrad           m6, INTERP_SHIFT_SP
-    psrad           m7, INTERP_SHIFT_SP
+    psrad           m5, INTERP_SHIFT_SP
 %else
-    psrad           m4, INTERP_SHIFT_PS
-    psrad           m5, INTERP_SHIFT_PS
     psrad           m6, INTERP_SHIFT_PS
-    psrad           m7, INTERP_SHIFT_PS
+    psrad           m5, INTERP_SHIFT_PS
 %endif
 %endif
 
-    packssdw        m4, m5
-    packssdw        m6, m7
-    vpermq          m4, m4, 11011000b
-    vpermq          m6, m6, 11011000b
+    packssdw        m6, m5
+    pxor            m1, m1
 %ifidn %1,pp
-    CLIPW           m4, m12, m13
-    CLIPW           m6, m12, m13
+    CLIPW           m6, m1, [pw_pixel_max]
 %elifidn %1, sp
-    CLIPW           m4, m12, m13
-    CLIPW           m6, m12, m13
+    CLIPW           m6, m1, [pw_pixel_max]
 %endif
-    lea             r8, [r2 + r3 * 4]
-    vextracti128    xm1, m4, 1
-    vextracti128    xm7, m6, 1
-    movu            [r8], xm4
-    movu            [r8 + r3], xm1
-    movu            [r8 + r3 * 2], xm6
-    movu            [r8 + r6], xm7
 
-    movu            xm1, [r7 + r4]                  ; m1 = row 15
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m2, m0, [r5 + 3 * mmsize]
-    paddd           m8, m2
-    pmaddwd         m0, [r5 + 2 * mmsize]
-    paddd           m10, m0
-    lea             r7, [r7 + r1 * 4]
-    movu            xm2, [r7]                       ; m2 = row 16
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m3, m1, [r5 + 3 * mmsize]
-    paddd           m9, m3
-    pmaddwd         m1, [r5 + 2 * mmsize]
-    paddd           m11, m1
-    movu            xm3, [r7 + r1]                  ; m3 = row 17
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m2, [r5 + 3 * mmsize]
-    paddd           m10, m2
-    movu            xm4, [r7 + r1 * 2]              ; m4 = row 18
-    punpckhwd       xm2, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm2, 1
-    pmaddwd         m3, [r5 + 3 * mmsize]
-    paddd           m11, m3
+    vextracti128    xm5, m6, 1
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm6
+    movq            [r2 + r3], xm5
+    movhps          [r2 + r3 * 2], xm6
+    movhps          [r2 + r6], xm5
+
+    movq            xm4, [r0 + r4]
+    punpcklwd       xm2, xm4
+    lea             r0, [r0 + 4 * r1]
+    movq            xm6, [r0]
+    punpcklwd       xm4, xm6
+    vinserti128     m2, m2, xm4, 1                  ; m2 = [20 19 19 18]
+    pmaddwd         m4, m2, [r5 + 3 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m2, [r5 + 2 * mmsize]
+    paddd           m3, m2
+    movq            xm4, [r0 + r1]
+    punpcklwd       xm6, xm4
+    movq            xm2, [r0 + 2 * r1]
+    punpcklwd       xm4, xm2
+    vinserti128     m6, m6, xm4, 1                  ; m6 = [22 21 21 20]
+    pmaddwd         m6, [r5 + 3 * mmsize]
+    paddd           m3, m6
 
 %ifidn %1,ss
-    psrad           m8, 6
-    psrad           m9, 6
-    psrad           m10, 6
-    psrad           m11, 6
+    psrad           m0, 6
+    psrad           m3, 6
 %else
-    paddd           m8, m14
-    paddd           m9, m14
-    paddd           m10, m14
-    paddd           m11, m14
+    paddd           m0, m7
+    paddd           m3, m7
 %ifidn %1,pp
-    psrad           m8, INTERP_SHIFT_PP
-    psrad           m9, INTERP_SHIFT_PP
-    psrad           m10, INTERP_SHIFT_PP
-    psrad           m11, INTERP_SHIFT_PP
+    psrad           m0, INTERP_SHIFT_PP
+    psrad           m3, INTERP_SHIFT_PP
 %elifidn %1, sp
-    psrad           m8, INTERP_SHIFT_SP
-    psrad           m9, INTERP_SHIFT_SP
-    psrad           m10, INTERP_SHIFT_SP
-    psrad           m11, INTERP_SHIFT_SP
+    psrad           m0, INTERP_SHIFT_SP
+    psrad           m3, INTERP_SHIFT_SP
 %else
-    psrad           m8, INTERP_SHIFT_PS
-    psrad           m9, INTERP_SHIFT_PS
-    psrad           m10, INTERP_SHIFT_PS
-    psrad           m11, INTERP_SHIFT_PS
+    psrad           m0, INTERP_SHIFT_PS
+    psrad           m3, INTERP_SHIFT_PS
 %endif
 %endif
 
-    packssdw        m8, m9
-    packssdw        m10, m11
-    vpermq          m8, m8, 11011000b
-    vpermq          m10, m10, 11011000b
+    packssdw        m0, m3
 %ifidn %1,pp
-    CLIPW           m8, m12, m13
-    CLIPW           m10, m12, m13
+    CLIPW           m0, m1, [pw_pixel_max]
 %elifidn %1, sp
-    CLIPW           m8, m12, m13
-    CLIPW           m10, m12, m13
-%endif
-    vextracti128    xm9, m8, 1
-    vextracti128    xm11, m10, 1
-    lea             r8, [r8 + r3 * 4]
-    movu            [r8], xm8
-    movu            [r8 + r3], xm9
-    movu            [r8 + r3 * 2], xm10
-    movu            [r8 + r6], xm11
-    add             r2, 16
-    add             r0, 16
-    dec             r9d
-    jnz             .loopW
-    RET
+    CLIPW           m0, m1, [pw_pixel_max]
 %endif
-%endmacro
 
-FILTER_VER_LUMA_AVX2_16x12 pp
-FILTER_VER_LUMA_AVX2_16x12 ps
-FILTER_VER_LUMA_AVX2_16x12 sp
-FILTER_VER_LUMA_AVX2_16x12 ss
+    vextracti128    xm3, m0, 1
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm0
+    movq            [r2 + r3], xm3
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r6], xm3
+%endmacro
 
-%macro FILTER_VER_LUMA_AVX2_4x8 1
+%macro FILTER_VER_LUMA_AVX2_4x16 1
 INIT_YMM avx2
-cglobal interp_8tap_vert_%1_4x8, 4, 7, 8
+cglobal interp_8tap_vert_%1_4x16, 4, 7, 8
     mov             r4d, r4m
     shl             r4d, 7
     add             r1d, r1d
@@ -4559,7 +3079,6 @@ cglobal interp_8tap_vert_%1_4x8, 4, 7, 8
 
     lea             r4, [r1 * 3]
     sub             r0, r4
-
 %ifidn %1,pp
     vbroadcasti128  m7, [pd_32]
 %elifidn %1, sp
@@ -4568,698 +3087,278 @@ cglobal interp_8tap_vert_%1_4x8, 4, 7, 8
     vbroadcasti128  m7, [INTERP_OFFSET_PS]
 %endif
     lea             r6, [r3 * 3]
+    PROCESS_LUMA_AVX2_W4_16R %1
+    RET
+%endmacro
 
-    movq            xm0, [r0]
-    movq            xm1, [r0 + r1]
-    punpcklwd       xm0, xm1
-    movq            xm2, [r0 + r1 * 2]
-    punpcklwd       xm1, xm2
-    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
-    pmaddwd         m0, [r5]
-    movq            xm3, [r0 + r4]
-    punpcklwd       xm2, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm4, [r0]
-    punpcklwd       xm3, xm4
-    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
-    pmaddwd         m5, m2, [r5 + 1 * mmsize]
-    pmaddwd         m2, [r5]
-    paddd           m0, m5
-    movq            xm3, [r0 + r1]
-    punpcklwd       xm4, xm3
-    movq            xm1, [r0 + r1 * 2]
-    punpcklwd       xm3, xm1
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
-    pmaddwd         m5, m4, [r5 + 2 * mmsize]
-    paddd           m0, m5
-    pmaddwd         m5, m4, [r5 + 1 * mmsize]
-    paddd           m2, m5
-    pmaddwd         m4, [r5]
-    movq            xm3, [r0 + r4]
-    punpcklwd       xm1, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm6, [r0]
-    punpcklwd       xm3, xm6
-    vinserti128     m1, m1, xm3, 1                  ; m1 = [8 7 7 6]
-    pmaddwd         m5, m1, [r5 + 3 * mmsize]
-    paddd           m0, m5
-    pmaddwd         m5, m1, [r5 + 2 * mmsize]
-    paddd           m2, m5
-    pmaddwd         m5, m1, [r5 + 1 * mmsize]
-    paddd           m4, m5
-    pmaddwd         m1, [r5]
-    movq            xm3, [r0 + r1]
-    punpcklwd       xm6, xm3
-    movq            xm5, [r0 + 2 * r1]
-    punpcklwd       xm3, xm5
-    vinserti128     m6, m6, xm3, 1                  ; m6 = [A 9 9 8]
-    pmaddwd         m3, m6, [r5 + 3 * mmsize]
-    paddd           m2, m3
-    pmaddwd         m3, m6, [r5 + 2 * mmsize]
-    paddd           m4, m3
-    pmaddwd         m6, [r5 + 1 * mmsize]
-    paddd           m1, m6
+FILTER_VER_LUMA_AVX2_4x16 pp
+FILTER_VER_LUMA_AVX2_4x16 ps
+FILTER_VER_LUMA_AVX2_4x16 sp
+FILTER_VER_LUMA_AVX2_4x16 ss
 
-%ifidn %1,ss
-    psrad           m0, 6
-    psrad           m2, 6
+%macro FILTER_VER_LUMA_AVX2_12x16 1
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_%1_12x16, 4, 9, 15
+    mov             r4d, r4m
+    shl             r4d, 7
+    add             r1d, r1d
+    add             r3d, r3d
+
+%ifdef PIC
+    lea             r5, [tab_LumaCoeffVer]
+    add             r5, r4
 %else
-    paddd           m0, m7
-    paddd           m2, m7
+    lea             r5, [tab_LumaCoeffVer + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r4
 %ifidn %1,pp
-    psrad           m0, INTERP_SHIFT_PP
-    psrad           m2, INTERP_SHIFT_PP
+    vbroadcasti128  m14, [pd_32]
 %elifidn %1, sp
-    psrad           m0, INTERP_SHIFT_SP
-    psrad           m2, INTERP_SHIFT_SP
+    vbroadcasti128  m14, [INTERP_OFFSET_SP]
 %else
-    psrad           m0, INTERP_SHIFT_PS
-    psrad           m2, INTERP_SHIFT_PS
+    vbroadcasti128  m14, [INTERP_OFFSET_PS]
 %endif
+    lea             r6, [r3 * 3]
+    PROCESS_LUMA_AVX2_W8_16R %1
+    add             r2, 16
+    add             r0, 16
+    mova            m7, m14
+    PROCESS_LUMA_AVX2_W4_16R %1
+    RET
 %endif
+%endmacro
 
-    packssdw        m0, m2
-    pxor            m6, m6
-    mova            m3, [pw_pixel_max]
-%ifidn %1,pp
-    CLIPW           m0, m6, m3
-%elifidn %1, sp
-    CLIPW           m0, m6, m3
-%endif
+FILTER_VER_LUMA_AVX2_12x16 pp
+FILTER_VER_LUMA_AVX2_12x16 ps
+FILTER_VER_LUMA_AVX2_12x16 sp
+FILTER_VER_LUMA_AVX2_12x16 ss
 
-    vextracti128    xm2, m0, 1
-    movq            [r2], xm0
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r6], xm2
+;---------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_ps_%1x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;---------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_LUMA_PS 2
+INIT_XMM sse4
+cglobal interp_8tap_vert_ps_%1x%2, 5, 7, 8 ,0-gprsize
 
-    movq            xm2, [r0 + r4]
-    punpcklwd       xm5, xm2
-    lea             r0, [r0 + 4 * r1]
-    movq            xm0, [r0]
-    punpcklwd       xm2, xm0
-    vinserti128     m5, m5, xm2, 1                  ; m5 = [C B B A]
-    pmaddwd         m2, m5, [r5 + 3 * mmsize]
-    paddd           m4, m2
-    pmaddwd         m5, [r5 + 2 * mmsize]
-    paddd           m1, m5
-    movq            xm2, [r0 + r1]
-    punpcklwd       xm0, xm2
-    movq            xm5, [r0 + 2 * r1]
-    punpcklwd       xm2, xm5
-    vinserti128     m0, m0, xm2, 1                  ; m0 = [E D D C]
-    pmaddwd         m0, [r5 + 3 * mmsize]
-    paddd           m1, m0
+    add       r1d, r1d
+    add       r3d, r3d
+    lea       r5, [r1 + 2 * r1]
+    sub       r0, r5
+    shl       r4d, 6
 
-%ifidn %1,ss
-    psrad           m4, 6
-    psrad           m1, 6
-%else
-    paddd           m4, m7
-    paddd           m1, m7
-%ifidn %1,pp
-    psrad           m4, INTERP_SHIFT_PP
-    psrad           m1, INTERP_SHIFT_PP
-%elifidn %1, sp
-    psrad           m4, INTERP_SHIFT_SP
-    psrad           m1, INTERP_SHIFT_SP
+%ifdef PIC
+    lea       r5, [tab_LumaCoeffV]
+    lea       r6, [r5 + r4]
 %else
-    psrad           m4, INTERP_SHIFT_PS
-    psrad           m1, INTERP_SHIFT_PS
-%endif
+    lea       r6, [tab_LumaCoeffV + r4]
 %endif
 
-    packssdw        m4, m1
-%ifidn %1,pp
-    CLIPW           m4, m6, m3
-%elifidn %1, sp
-    CLIPW           m4, m6, m3
-%endif
+    mova      m7, [INTERP_OFFSET_PS]
 
-    vextracti128    xm1, m4, 1
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm4
-    movq            [r2 + r3], xm1
-    movhps          [r2 + r3 * 2], xm4
-    movhps          [r2 + r6], xm1
+    mov       dword [rsp], %2/4
+.loopH:
+    mov       r4d, (%1/4)
+.loopW:
+    PROCESS_LUMA_VER_W4_4R
+
+    paddd     m0, m7
+    paddd     m1, m7
+    paddd     m2, m7
+    paddd     m3, m7
+
+    psrad     m0, INTERP_SHIFT_PS
+    psrad     m1, INTERP_SHIFT_PS
+    psrad     m2, INTERP_SHIFT_PS
+    psrad     m3, INTERP_SHIFT_PS
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+
+    movh      [r2], m0
+    movhps    [r2 + r3], m0
+    lea       r5, [r2 + 2 * r3]
+    movh      [r5], m2
+    movhps    [r5 + r3], m2
+
+    lea       r5, [8 * r1 - 2 * 4]
+    sub       r0, r5
+    add       r2, 2 * 4
+
+    dec       r4d
+    jnz       .loopW
+
+    lea       r0, [r0 + 4 * r1 - 2 * %1]
+    lea       r2, [r2 + 4 * r3 - 2 * %1]
+
+    dec       dword [rsp]
+    jnz       .loopH
     RET
 %endmacro
 
-FILTER_VER_LUMA_AVX2_4x8 pp
-FILTER_VER_LUMA_AVX2_4x8 ps
-FILTER_VER_LUMA_AVX2_4x8 sp
-FILTER_VER_LUMA_AVX2_4x8 ss
+;---------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_ps_%1x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;---------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_PS 4, 4
+    FILTER_VER_LUMA_PS 8, 8
+    FILTER_VER_LUMA_PS 8, 4
+    FILTER_VER_LUMA_PS 4, 8
+    FILTER_VER_LUMA_PS 16, 16
+    FILTER_VER_LUMA_PS 16, 8
+    FILTER_VER_LUMA_PS 8, 16
+    FILTER_VER_LUMA_PS 16, 12
+    FILTER_VER_LUMA_PS 12, 16
+    FILTER_VER_LUMA_PS 16, 4
+    FILTER_VER_LUMA_PS 4, 16
+    FILTER_VER_LUMA_PS 32, 32
+    FILTER_VER_LUMA_PS 32, 16
+    FILTER_VER_LUMA_PS 16, 32
+    FILTER_VER_LUMA_PS 32, 24
+    FILTER_VER_LUMA_PS 24, 32
+    FILTER_VER_LUMA_PS 32, 8
+    FILTER_VER_LUMA_PS 8, 32
+    FILTER_VER_LUMA_PS 64, 64
+    FILTER_VER_LUMA_PS 64, 32
+    FILTER_VER_LUMA_PS 32, 64
+    FILTER_VER_LUMA_PS 64, 48
+    FILTER_VER_LUMA_PS 48, 64
+    FILTER_VER_LUMA_PS 64, 16
+    FILTER_VER_LUMA_PS 16, 64
 
-%macro PROCESS_LUMA_AVX2_W4_16R 1
-    movq            xm0, [r0]
-    movq            xm1, [r0 + r1]
-    punpcklwd       xm0, xm1
-    movq            xm2, [r0 + r1 * 2]
-    punpcklwd       xm1, xm2
-    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
-    pmaddwd         m0, [r5]
-    movq            xm3, [r0 + r4]
-    punpcklwd       xm2, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm4, [r0]
-    punpcklwd       xm3, xm4
-    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
-    pmaddwd         m5, m2, [r5 + 1 * mmsize]
-    pmaddwd         m2, [r5]
-    paddd           m0, m5
-    movq            xm3, [r0 + r1]
-    punpcklwd       xm4, xm3
-    movq            xm1, [r0 + r1 * 2]
-    punpcklwd       xm3, xm1
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
-    pmaddwd         m5, m4, [r5 + 2 * mmsize]
-    paddd           m0, m5
-    pmaddwd         m5, m4, [r5 + 1 * mmsize]
-    paddd           m2, m5
-    pmaddwd         m4, [r5]
-    movq            xm3, [r0 + r4]
-    punpcklwd       xm1, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm6, [r0]
-    punpcklwd       xm3, xm6
-    vinserti128     m1, m1, xm3, 1                  ; m1 = [8 7 7 6]
-    pmaddwd         m5, m1, [r5 + 3 * mmsize]
-    paddd           m0, m5
-    pmaddwd         m5, m1, [r5 + 2 * mmsize]
-    paddd           m2, m5
-    pmaddwd         m5, m1, [r5 + 1 * mmsize]
-    paddd           m4, m5
-    pmaddwd         m1, [r5]
-    movq            xm3, [r0 + r1]
-    punpcklwd       xm6, xm3
-    movq            xm5, [r0 + 2 * r1]
-    punpcklwd       xm3, xm5
-    vinserti128     m6, m6, xm3, 1                  ; m6 = [10 9 9 8]
-    pmaddwd         m3, m6, [r5 + 3 * mmsize]
-    paddd           m2, m3
-    pmaddwd         m3, m6, [r5 + 2 * mmsize]
-    paddd           m4, m3
-    pmaddwd         m3, m6, [r5 + 1 * mmsize]
-    paddd           m1, m3
-    pmaddwd         m6, [r5]
+;--------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_sp_%1x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_LUMA_SP 2
+INIT_XMM sse4
+cglobal interp_8tap_vert_sp_%1x%2, 5, 7, 8 ,0-gprsize
 
-%ifidn %1,ss
-    psrad           m0, 6
-    psrad           m2, 6
-%else
-    paddd           m0, m7
-    paddd           m2, m7
-%ifidn %1,pp
-    psrad           m0, INTERP_SHIFT_PP
-    psrad           m2, INTERP_SHIFT_PP
-%elifidn %1, sp
-    psrad           m0, INTERP_SHIFT_SP
-    psrad           m2, INTERP_SHIFT_SP
+    add       r1d, r1d
+    add       r3d, r3d
+    lea       r5, [r1 + 2 * r1]
+    sub       r0, r5
+    shl       r4d, 6
+
+%ifdef PIC
+    lea       r5, [tab_LumaCoeffV]
+    lea       r6, [r5 + r4]
 %else
-    psrad           m0, INTERP_SHIFT_PS
-    psrad           m2, INTERP_SHIFT_PS
-%endif
+    lea       r6, [tab_LumaCoeffV + r4]
 %endif
 
-    packssdw        m0, m2
-    pxor            m3, m3
-%ifidn %1,pp
-    CLIPW           m0, m3, [pw_pixel_max]
-%elifidn %1, sp
-    CLIPW           m0, m3, [pw_pixel_max]
-%endif
+    mova      m7, [INTERP_OFFSET_SP]
 
-    vextracti128    xm2, m0, 1
-    movq            [r2], xm0
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r6], xm2
+    mov       dword [rsp], %2/4
+.loopH:
+    mov       r4d, (%1/4)
+.loopW:
+    PROCESS_LUMA_VER_W4_4R
 
-    movq            xm2, [r0 + r4]
-    punpcklwd       xm5, xm2
-    lea             r0, [r0 + 4 * r1]
-    movq            xm0, [r0]
-    punpcklwd       xm2, xm0
-    vinserti128     m5, m5, xm2, 1                  ; m5 = [12 11 11 10]
-    pmaddwd         m2, m5, [r5 + 3 * mmsize]
-    paddd           m4, m2
-    pmaddwd         m2, m5, [r5 + 2 * mmsize]
-    paddd           m1, m2
-    pmaddwd         m2, m5, [r5 + 1 * mmsize]
-    paddd           m6, m2
-    pmaddwd         m5, [r5]
-    movq            xm2, [r0 + r1]
-    punpcklwd       xm0, xm2
-    movq            xm3, [r0 + 2 * r1]
-    punpcklwd       xm2, xm3
-    vinserti128     m0, m0, xm2, 1                  ; m0 = [14 13 13 12]
-    pmaddwd         m2, m0, [r5 + 3 * mmsize]
-    paddd           m1, m2
-    pmaddwd         m2, m0, [r5 + 2 * mmsize]
-    paddd           m6, m2
-    pmaddwd         m2, m0, [r5 + 1 * mmsize]
-    paddd           m5, m2
-    pmaddwd         m0, [r5]
-
-%ifidn %1,ss
-    psrad           m4, 6
-    psrad           m1, 6
-%else
-    paddd           m4, m7
-    paddd           m1, m7
-%ifidn %1,pp
-    psrad           m4, INTERP_SHIFT_PP
-    psrad           m1, INTERP_SHIFT_PP
-%elifidn %1, sp
-    psrad           m4, INTERP_SHIFT_SP
-    psrad           m1, INTERP_SHIFT_SP
-%else
-    psrad           m4, INTERP_SHIFT_PS
-    psrad           m1, INTERP_SHIFT_PS
-%endif
-%endif
+    paddd     m0, m7
+    paddd     m1, m7
+    paddd     m2, m7
+    paddd     m3, m7
 
-    packssdw        m4, m1
-    pxor            m2, m2
-%ifidn %1,pp
-    CLIPW           m4, m2, [pw_pixel_max]
-%elifidn %1, sp
-    CLIPW           m4, m2, [pw_pixel_max]
-%endif
+    psrad     m0, INTERP_SHIFT_SP
+    psrad     m1, INTERP_SHIFT_SP
+    psrad     m2, INTERP_SHIFT_SP
+    psrad     m3, INTERP_SHIFT_SP
 
-    vextracti128    xm1, m4, 1
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm4
-    movq            [r2 + r3], xm1
-    movhps          [r2 + r3 * 2], xm4
-    movhps          [r2 + r6], xm1
+    packssdw  m0, m1
+    packssdw  m2, m3
 
-    movq            xm4, [r0 + r4]
-    punpcklwd       xm3, xm4
-    lea             r0, [r0 + 4 * r1]
-    movq            xm1, [r0]
-    punpcklwd       xm4, xm1
-    vinserti128     m3, m3, xm4, 1                  ; m3 = [16 15 15 14]
-    pmaddwd         m4, m3, [r5 + 3 * mmsize]
-    paddd           m6, m4
-    pmaddwd         m4, m3, [r5 + 2 * mmsize]
-    paddd           m5, m4
-    pmaddwd         m4, m3, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m3, [r5]
-    movq            xm4, [r0 + r1]
-    punpcklwd       xm1, xm4
-    movq            xm2, [r0 + 2 * r1]
-    punpcklwd       xm4, xm2
-    vinserti128     m1, m1, xm4, 1                  ; m1 = [18 17 17 16]
-    pmaddwd         m4, m1, [r5 + 3 * mmsize]
-    paddd           m5, m4
-    pmaddwd         m4, m1, [r5 + 2 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m1, [r5 + 1 * mmsize]
-    paddd           m3, m1
+    pxor      m1, m1
+    CLIPW2    m0, m2, m1, [pw_pixel_max]
 
-%ifidn %1,ss
-    psrad           m6, 6
-    psrad           m5, 6
-%else
-    paddd           m6, m7
-    paddd           m5, m7
-%ifidn %1,pp
-    psrad           m6, INTERP_SHIFT_PP
-    psrad           m5, INTERP_SHIFT_PP
-%elifidn %1, sp
-    psrad           m6, INTERP_SHIFT_SP
-    psrad           m5, INTERP_SHIFT_SP
-%else
-    psrad           m6, INTERP_SHIFT_PS
-    psrad           m5, INTERP_SHIFT_PS
-%endif
-%endif
+    movh      [r2], m0
+    movhps    [r2 + r3], m0
+    lea       r5, [r2 + 2 * r3]
+    movh      [r5], m2
+    movhps    [r5 + r3], m2
 
-    packssdw        m6, m5
-    pxor            m1, m1
-%ifidn %1,pp
-    CLIPW           m6, m1, [pw_pixel_max]
-%elifidn %1, sp
-    CLIPW           m6, m1, [pw_pixel_max]
-%endif
+    lea       r5, [8 * r1 - 2 * 4]
+    sub       r0, r5
+    add       r2, 2 * 4
 
-    vextracti128    xm5, m6, 1
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm6
-    movq            [r2 + r3], xm5
-    movhps          [r2 + r3 * 2], xm6
-    movhps          [r2 + r6], xm5
+    dec       r4d
+    jnz       .loopW
 
-    movq            xm4, [r0 + r4]
-    punpcklwd       xm2, xm4
-    lea             r0, [r0 + 4 * r1]
-    movq            xm6, [r0]
-    punpcklwd       xm4, xm6
-    vinserti128     m2, m2, xm4, 1                  ; m2 = [20 19 19 18]
-    pmaddwd         m4, m2, [r5 + 3 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m2, [r5 + 2 * mmsize]
-    paddd           m3, m2
-    movq            xm4, [r0 + r1]
-    punpcklwd       xm6, xm4
-    movq            xm2, [r0 + 2 * r1]
-    punpcklwd       xm4, xm2
-    vinserti128     m6, m6, xm4, 1                  ; m6 = [22 21 21 20]
-    pmaddwd         m6, [r5 + 3 * mmsize]
-    paddd           m3, m6
+    lea       r0, [r0 + 4 * r1 - 2 * %1]
+    lea       r2, [r2 + 4 * r3 - 2 * %1]
 
-%ifidn %1,ss
-    psrad           m0, 6
-    psrad           m3, 6
-%else
-    paddd           m0, m7
-    paddd           m3, m7
-%ifidn %1,pp
-    psrad           m0, INTERP_SHIFT_PP
-    psrad           m3, INTERP_SHIFT_PP
-%elifidn %1, sp
-    psrad           m0, INTERP_SHIFT_SP
-    psrad           m3, INTERP_SHIFT_SP
-%else
-    psrad           m0, INTERP_SHIFT_PS
-    psrad           m3, INTERP_SHIFT_PS
-%endif
-%endif
+    dec       dword [rsp]
+    jnz       .loopH
+    RET
+%endmacro
 
-    packssdw        m0, m3
-%ifidn %1,pp
-    CLIPW           m0, m1, [pw_pixel_max]
-%elifidn %1, sp
-    CLIPW           m0, m1, [pw_pixel_max]
-%endif
+;--------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_sp_%1x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;--------------------------------------------------------------------------------------------------------------
+    FILTER_VER_LUMA_SP 4, 4
+    FILTER_VER_LUMA_SP 8, 8
+    FILTER_VER_LUMA_SP 8, 4
+    FILTER_VER_LUMA_SP 4, 8
+    FILTER_VER_LUMA_SP 16, 16
+    FILTER_VER_LUMA_SP 16, 8
+    FILTER_VER_LUMA_SP 8, 16
+    FILTER_VER_LUMA_SP 16, 12
+    FILTER_VER_LUMA_SP 12, 16
+    FILTER_VER_LUMA_SP 16, 4
+    FILTER_VER_LUMA_SP 4, 16
+    FILTER_VER_LUMA_SP 32, 32
+    FILTER_VER_LUMA_SP 32, 16
+    FILTER_VER_LUMA_SP 16, 32
+    FILTER_VER_LUMA_SP 32, 24
+    FILTER_VER_LUMA_SP 24, 32
+    FILTER_VER_LUMA_SP 32, 8
+    FILTER_VER_LUMA_SP 8, 32
+    FILTER_VER_LUMA_SP 64, 64
+    FILTER_VER_LUMA_SP 64, 32
+    FILTER_VER_LUMA_SP 32, 64
+    FILTER_VER_LUMA_SP 64, 48
+    FILTER_VER_LUMA_SP 48, 64
+    FILTER_VER_LUMA_SP 64, 16
+    FILTER_VER_LUMA_SP 16, 64
 
-    vextracti128    xm3, m0, 1
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm0
-    movq            [r2 + r3], xm3
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r6], xm3
-%endmacro
+;-----------------------------------------------------------------------------------------------------------------
+; void interp_8tap_vert_ss_%1x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_LUMA_SS 2
+INIT_XMM sse2
+cglobal interp_8tap_vert_ss_%1x%2, 5, 7, 7 ,0-gprsize
 
-%macro FILTER_VER_LUMA_AVX2_4x16 1
-INIT_YMM avx2
-cglobal interp_8tap_vert_%1_4x16, 4, 7, 8
-    mov             r4d, r4m
-    shl             r4d, 7
-    add             r1d, r1d
-    add             r3d, r3d
+    add        r1d, r1d
+    add        r3d, r3d
+    lea        r5, [3 * r1]
+    sub        r0, r5
+    shl        r4d, 6
 
 %ifdef PIC
-    lea             r5, [tab_LumaCoeffVer]
-    add             r5, r4
+    lea        r5, [tab_LumaCoeffV]
+    lea        r6, [r5 + r4]
 %else
-    lea             r5, [tab_LumaCoeffVer + r4]
+    lea        r6, [tab_LumaCoeffV + r4]
 %endif
 
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-%ifidn %1,pp
-    vbroadcasti128  m7, [pd_32]
-%elifidn %1, sp
-    vbroadcasti128  m7, [INTERP_OFFSET_SP]
-%else
-    vbroadcasti128  m7, [INTERP_OFFSET_PS]
-%endif
-    lea             r6, [r3 * 3]
-    PROCESS_LUMA_AVX2_W4_16R %1
-    RET
-%endmacro
+    mov        dword [rsp], %2/4
+.loopH:
+    mov        r4d, (%1/4)
+.loopW:
+    PROCESS_LUMA_VER_W4_4R
 
-FILTER_VER_LUMA_AVX2_4x16 pp
-FILTER_VER_LUMA_AVX2_4x16 ps
-FILTER_VER_LUMA_AVX2_4x16 sp
-FILTER_VER_LUMA_AVX2_4x16 ss
+    psrad      m0, 6
+    psrad      m1, 6
+    packssdw   m0, m1
+    movlps     [r2], m0
+    movhps     [r2 + r3], m0
 
-%macro FILTER_VER_LUMA_AVX2_12x16 1
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_8tap_vert_%1_12x16, 4, 9, 15
-    mov             r4d, r4m
-    shl             r4d, 7
-    add             r1d, r1d
-    add             r3d, r3d
-
-%ifdef PIC
-    lea             r5, [tab_LumaCoeffVer]
-    add             r5, r4
-%else
-    lea             r5, [tab_LumaCoeffVer + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r4
-%ifidn %1,pp
-    vbroadcasti128  m14, [pd_32]
-%elifidn %1, sp
-    vbroadcasti128  m14, [INTERP_OFFSET_SP]
-%else
-    vbroadcasti128  m14, [INTERP_OFFSET_PS]
-%endif
-    lea             r6, [r3 * 3]
-    PROCESS_LUMA_AVX2_W8_16R %1
-    add             r2, 16
-    add             r0, 16
-    mova            m7, m14
-    PROCESS_LUMA_AVX2_W4_16R %1
-    RET
-%endif
-%endmacro
-
-FILTER_VER_LUMA_AVX2_12x16 pp
-FILTER_VER_LUMA_AVX2_12x16 ps
-FILTER_VER_LUMA_AVX2_12x16 sp
-FILTER_VER_LUMA_AVX2_12x16 ss
-
-;---------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_ps_%1x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;---------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_LUMA_PS 2
-INIT_XMM sse4
-cglobal interp_8tap_vert_ps_%1x%2, 5, 7, 8 ,0-gprsize
-
-    add       r1d, r1d
-    add       r3d, r3d
-    lea       r5, [r1 + 2 * r1]
-    sub       r0, r5
-    shl       r4d, 6
-
-%ifdef PIC
-    lea       r5, [tab_LumaCoeffV]
-    lea       r6, [r5 + r4]
-%else
-    lea       r6, [tab_LumaCoeffV + r4]
-%endif
-
-    mova      m7, [INTERP_OFFSET_PS]
-
-    mov       dword [rsp], %2/4
-.loopH:
-    mov       r4d, (%1/4)
-.loopW:
-    PROCESS_LUMA_VER_W4_4R
-
-    paddd     m0, m7
-    paddd     m1, m7
-    paddd     m2, m7
-    paddd     m3, m7
-
-    psrad     m0, INTERP_SHIFT_PS
-    psrad     m1, INTERP_SHIFT_PS
-    psrad     m2, INTERP_SHIFT_PS
-    psrad     m3, INTERP_SHIFT_PS
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-
-    movh      [r2], m0
-    movhps    [r2 + r3], m0
-    lea       r5, [r2 + 2 * r3]
-    movh      [r5], m2
-    movhps    [r5 + r3], m2
-
-    lea       r5, [8 * r1 - 2 * 4]
-    sub       r0, r5
-    add       r2, 2 * 4
-
-    dec       r4d
-    jnz       .loopW
-
-    lea       r0, [r0 + 4 * r1 - 2 * %1]
-    lea       r2, [r2 + 4 * r3 - 2 * %1]
-
-    dec       dword [rsp]
-    jnz       .loopH
-    RET
-%endmacro
-
-;---------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_ps_%1x%2(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;---------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_PS 4, 4
-    FILTER_VER_LUMA_PS 8, 8
-    FILTER_VER_LUMA_PS 8, 4
-    FILTER_VER_LUMA_PS 4, 8
-    FILTER_VER_LUMA_PS 16, 16
-    FILTER_VER_LUMA_PS 16, 8
-    FILTER_VER_LUMA_PS 8, 16
-    FILTER_VER_LUMA_PS 16, 12
-    FILTER_VER_LUMA_PS 12, 16
-    FILTER_VER_LUMA_PS 16, 4
-    FILTER_VER_LUMA_PS 4, 16
-    FILTER_VER_LUMA_PS 32, 32
-    FILTER_VER_LUMA_PS 32, 16
-    FILTER_VER_LUMA_PS 16, 32
-    FILTER_VER_LUMA_PS 32, 24
-    FILTER_VER_LUMA_PS 24, 32
-    FILTER_VER_LUMA_PS 32, 8
-    FILTER_VER_LUMA_PS 8, 32
-    FILTER_VER_LUMA_PS 64, 64
-    FILTER_VER_LUMA_PS 64, 32
-    FILTER_VER_LUMA_PS 32, 64
-    FILTER_VER_LUMA_PS 64, 48
-    FILTER_VER_LUMA_PS 48, 64
-    FILTER_VER_LUMA_PS 64, 16
-    FILTER_VER_LUMA_PS 16, 64
-
-;--------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_sp_%1x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_LUMA_SP 2
-INIT_XMM sse4
-cglobal interp_8tap_vert_sp_%1x%2, 5, 7, 8 ,0-gprsize
-
-    add       r1d, r1d
-    add       r3d, r3d
-    lea       r5, [r1 + 2 * r1]
-    sub       r0, r5
-    shl       r4d, 6
-
-%ifdef PIC
-    lea       r5, [tab_LumaCoeffV]
-    lea       r6, [r5 + r4]
-%else
-    lea       r6, [tab_LumaCoeffV + r4]
-%endif
-
-    mova      m7, [INTERP_OFFSET_SP]
-
-    mov       dword [rsp], %2/4
-.loopH:
-    mov       r4d, (%1/4)
-.loopW:
-    PROCESS_LUMA_VER_W4_4R
-
-    paddd     m0, m7
-    paddd     m1, m7
-    paddd     m2, m7
-    paddd     m3, m7
-
-    psrad     m0, INTERP_SHIFT_SP
-    psrad     m1, INTERP_SHIFT_SP
-    psrad     m2, INTERP_SHIFT_SP
-    psrad     m3, INTERP_SHIFT_SP
-
-    packssdw  m0, m1
-    packssdw  m2, m3
-
-    pxor      m1, m1
-    CLIPW2    m0, m2, m1, [pw_pixel_max]
-
-    movh      [r2], m0
-    movhps    [r2 + r3], m0
-    lea       r5, [r2 + 2 * r3]
-    movh      [r5], m2
-    movhps    [r5 + r3], m2
-
-    lea       r5, [8 * r1 - 2 * 4]
-    sub       r0, r5
-    add       r2, 2 * 4
-
-    dec       r4d
-    jnz       .loopW
-
-    lea       r0, [r0 + 4 * r1 - 2 * %1]
-    lea       r2, [r2 + 4 * r3 - 2 * %1]
-
-    dec       dword [rsp]
-    jnz       .loopH
-    RET
-%endmacro
-
-;--------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_sp_%1x%2(int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;--------------------------------------------------------------------------------------------------------------
-    FILTER_VER_LUMA_SP 4, 4
-    FILTER_VER_LUMA_SP 8, 8
-    FILTER_VER_LUMA_SP 8, 4
-    FILTER_VER_LUMA_SP 4, 8
-    FILTER_VER_LUMA_SP 16, 16
-    FILTER_VER_LUMA_SP 16, 8
-    FILTER_VER_LUMA_SP 8, 16
-    FILTER_VER_LUMA_SP 16, 12
-    FILTER_VER_LUMA_SP 12, 16
-    FILTER_VER_LUMA_SP 16, 4
-    FILTER_VER_LUMA_SP 4, 16
-    FILTER_VER_LUMA_SP 32, 32
-    FILTER_VER_LUMA_SP 32, 16
-    FILTER_VER_LUMA_SP 16, 32
-    FILTER_VER_LUMA_SP 32, 24
-    FILTER_VER_LUMA_SP 24, 32
-    FILTER_VER_LUMA_SP 32, 8
-    FILTER_VER_LUMA_SP 8, 32
-    FILTER_VER_LUMA_SP 64, 64
-    FILTER_VER_LUMA_SP 64, 32
-    FILTER_VER_LUMA_SP 32, 64
-    FILTER_VER_LUMA_SP 64, 48
-    FILTER_VER_LUMA_SP 48, 64
-    FILTER_VER_LUMA_SP 64, 16
-    FILTER_VER_LUMA_SP 16, 64
-
-;-----------------------------------------------------------------------------------------------------------------
-; void interp_8tap_vert_ss_%1x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------------------------------------------
-%macro FILTER_VER_LUMA_SS 2
-INIT_XMM sse2
-cglobal interp_8tap_vert_ss_%1x%2, 5, 7, 7 ,0-gprsize
-
-    add        r1d, r1d
-    add        r3d, r3d
-    lea        r5, [3 * r1]
-    sub        r0, r5
-    shl        r4d, 6
-
-%ifdef PIC
-    lea        r5, [tab_LumaCoeffV]
-    lea        r6, [r5 + r4]
-%else
-    lea        r6, [tab_LumaCoeffV + r4]
-%endif
-
-    mov        dword [rsp], %2/4
-.loopH:
-    mov        r4d, (%1/4)
-.loopW:
-    PROCESS_LUMA_VER_W4_4R
-
-    psrad      m0, 6
-    psrad      m1, 6
-    packssdw   m0, m1
-    movlps     [r2], m0
-    movhps     [r2 + r3], m0
-
-    psrad      m2, 6
-    psrad      m3, 6
-    packssdw   m2, m3
-    movlps     [r2 + 2 * r3], m2
-    lea        r5, [3 * r3]
-    movhps     [r2 + r5], m2
+    psrad      m2, 6
+    psrad      m3, 6
+    packssdw   m2, m3
+    movlps     [r2 + 2 * r3], m2
+    lea        r5, [3 * r3]
+    movhps     [r2 + r5], m2
 
     lea        r5, [8 * r1 - 2 * 4]
     sub        r0, r5
@@ -6512,1570 +4611,3 @@ cglobal filterPixelToShort_48x64, 3, 7, 5
     jnz        .loop
     RET
 
-;-----------------------------------------------------------------------------
-; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride)
-;-----------------------------------------------------------------------------
-INIT_YMM avx2
-cglobal filterPixelToShort_48x64, 3, 7, 4
-    add        r1d, r1d
-    mov        r3d, r3m
-    add        r3d, r3d
-    lea        r4, [r3 * 3]
-    lea        r5, [r1 * 3]
-
-    ; load height
-    mov        r6d, 16
-
-    ; load constant
-    mova       m3, [pw_2000]
-
-.loop:
-    movu       m0, [r0]
-    movu       m1, [r0 + 32]
-    movu       m2, [r0 + 64]
-    psllw      m0, (14 - BIT_DEPTH)
-    psllw      m1, (14 - BIT_DEPTH)
-    psllw      m2, (14 - BIT_DEPTH)
-    psubw      m0, m3
-    psubw      m1, m3
-    psubw      m2, m3
-    movu       [r2 + r3 * 0], m0
-    movu       [r2 + r3 * 0 + 32], m1
-    movu       [r2 + r3 * 0 + 64], m2
-
-    movu       m0, [r0 + r1]
-    movu       m1, [r0 + r1 + 32]
-    movu       m2, [r0 + r1 + 64]
-    psllw      m0, (14 - BIT_DEPTH)
-    psllw      m1, (14 - BIT_DEPTH)
-    psllw      m2, (14 - BIT_DEPTH)
-    psubw      m0, m3
-    psubw      m1, m3
-    psubw      m2, m3
-    movu       [r2 + r3 * 1], m0
-    movu       [r2 + r3 * 1 + 32], m1
-    movu       [r2 + r3 * 1 + 64], m2
-
-    movu       m0, [r0 + r1 * 2]
-    movu       m1, [r0 + r1 * 2 + 32]
-    movu       m2, [r0 + r1 * 2 + 64]
-    psllw      m0, (14 - BIT_DEPTH)
-    psllw      m1, (14 - BIT_DEPTH)
-    psllw      m2, (14 - BIT_DEPTH)
-    psubw      m0, m3
-    psubw      m1, m3
-    psubw      m2, m3
-    movu       [r2 + r3 * 2], m0
-    movu       [r2 + r3 * 2 + 32], m1
-    movu       [r2 + r3 * 2 + 64], m2
-
-    movu       m0, [r0 + r5]
-    movu       m1, [r0 + r5 + 32]
-    movu       m2, [r0 + r5 + 64]
-    psllw      m0, (14 - BIT_DEPTH)
-    psllw      m1, (14 - BIT_DEPTH)
-    psllw      m2, (14 - BIT_DEPTH)
-    psubw      m0, m3
-    psubw      m1, m3
-    psubw      m2, m3
-    movu       [r2 + r4], m0
-    movu       [r2 + r4 + 32], m1
-    movu       [r2 + r4 + 64], m2
-
-    lea        r0, [r0 + r1 * 4]
-    lea        r2, [r2 + r3 * 4]
-
-    dec        r6d
-    jnz        .loop
-    RET
-
-
-%macro FILTER_VER_CHROMA_AVX2_8xN 2
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_vert_%1_8x%2, 4, 9, 15
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-    add             r3d, r3d
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,pp
-    vbroadcasti128  m14, [pd_32]
-%elifidn %1, sp
-    vbroadcasti128  m14, [INTERP_OFFSET_SP]
-%else
-    vbroadcasti128  m14, [INTERP_OFFSET_PS]
-%endif
-    lea             r6, [r3 * 3]
-    lea             r7, [r1 * 4]
-    mov             r8d, %2 / 16
-.loopH:
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m2, [r5]
-
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    paddd           m1, m5
-    pmaddwd         m3, [r5]
-
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m2, m6
-    pmaddwd         m4, [r5]
-
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm7, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddwd         m7, m5, [r5 + 1 * mmsize]
-    paddd           m3, m7
-    pmaddwd         m5, [r5]
-
-    movu            xm7, [r0 + r4]                  ; m7 = row 7
-    punpckhwd       xm8, xm6, xm7
-    punpcklwd       xm6, xm7
-    vinserti128     m6, m6, xm8, 1
-    pmaddwd         m8, m6, [r5 + 1 * mmsize]
-    paddd           m4, m8
-    pmaddwd         m6, [r5]
-
-    lea             r0, [r0 + r1 * 4]
-    movu            xm8, [r0]                       ; m8 = row 8
-    punpckhwd       xm9, xm7, xm8
-    punpcklwd       xm7, xm8
-    vinserti128     m7, m7, xm9, 1
-    pmaddwd         m9, m7, [r5 + 1 * mmsize]
-    paddd           m5, m9
-    pmaddwd         m7, [r5]
-
-
-    movu            xm9, [r0 + r1]                  ; m9 = row 9
-    punpckhwd       xm10, xm8, xm9
-    punpcklwd       xm8, xm9
-    vinserti128     m8, m8, xm10, 1
-    pmaddwd         m10, m8, [r5 + 1 * mmsize]
-    paddd           m6, m10
-    pmaddwd         m8, [r5]
-
-
-    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
-    punpckhwd       xm11, xm9, xm10
-    punpcklwd       xm9, xm10
-    vinserti128     m9, m9, xm11, 1
-    pmaddwd         m11, m9, [r5 + 1 * mmsize]
-    paddd           m7, m11
-    pmaddwd         m9, [r5]
-
-    movu            xm11, [r0 + r4]                 ; m11 = row 11
-    punpckhwd       xm12, xm10, xm11
-    punpcklwd       xm10, xm11
-    vinserti128     m10, m10, xm12, 1
-    pmaddwd         m12, m10, [r5 + 1 * mmsize]
-    paddd           m8, m12
-    pmaddwd         m10, [r5]
-
-    lea             r0, [r0 + r1 * 4]
-    movu            xm12, [r0]                      ; m12 = row 12
-    punpckhwd       xm13, xm11, xm12
-    punpcklwd       xm11, xm12
-    vinserti128     m11, m11, xm13, 1
-    pmaddwd         m13, m11, [r5 + 1 * mmsize]
-    paddd           m9, m13
-    pmaddwd         m11, [r5]
-
-%ifidn %1,ss
-    psrad           m0, 6
-    psrad           m1, 6
-    psrad           m2, 6
-    psrad           m3, 6
-    psrad           m4, 6
-    psrad           m5, 6
-%else
-    paddd           m0, m14
-    paddd           m1, m14
-    paddd           m2, m14
-    paddd           m3, m14
-    paddd           m4, m14
-    paddd           m5, m14
-%ifidn %1,pp
-    psrad           m0, 6
-    psrad           m1, 6
-    psrad           m2, 6
-    psrad           m3, 6
-    psrad           m4, 6
-    psrad           m5, 6
-%elifidn %1, sp
-    psrad           m0, INTERP_SHIFT_SP
-    psrad           m1, INTERP_SHIFT_SP
-    psrad           m2, INTERP_SHIFT_SP
-    psrad           m3, INTERP_SHIFT_SP
-    psrad           m4, INTERP_SHIFT_SP
-    psrad           m5, INTERP_SHIFT_SP
-%else
-    psrad           m0, INTERP_SHIFT_PS
-    psrad           m1, INTERP_SHIFT_PS
-    psrad           m2, INTERP_SHIFT_PS
-    psrad           m3, INTERP_SHIFT_PS
-    psrad           m4, INTERP_SHIFT_PS
-    psrad           m5, INTERP_SHIFT_PS
-%endif
-%endif
-
-    packssdw        m0, m1
-    packssdw        m2, m3
-    packssdw        m4, m5
-    vpermq          m0, m0, q3120
-    vpermq          m2, m2, q3120
-    vpermq          m4, m4, q3120
-    pxor            m5, m5
-    mova            m3, [pw_pixel_max]
-%ifidn %1,pp
-    CLIPW           m0, m5, m3
-    CLIPW           m2, m5, m3
-    CLIPW           m4, m5, m3
-%elifidn %1, sp
-    CLIPW           m0, m5, m3
-    CLIPW           m2, m5, m3
-    CLIPW           m4, m5, m3
-%endif
-
-    vextracti128    xm1, m0, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    vextracti128    xm1, m2, 1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm1
-    lea             r2, [r2 + r3 * 4]
-    vextracti128    xm1, m4, 1
-    movu            [r2], xm4
-    movu            [r2 + r3], xm1
-
-    movu            xm13, [r0 + r1]                 ; m13 = row 13
-    punpckhwd       xm0, xm12, xm13
-    punpcklwd       xm12, xm13
-    vinserti128     m12, m12, xm0, 1
-    pmaddwd         m0, m12, [r5 + 1 * mmsize]
-    paddd           m10, m0
-    pmaddwd         m12, [r5]
-
-    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
-    punpckhwd       xm1, xm13, xm0
-    punpcklwd       xm13, xm0
-    vinserti128     m13, m13, xm1, 1
-    pmaddwd         m1, m13, [r5 + 1 * mmsize]
-    paddd           m11, m1
-    pmaddwd         m13, [r5]
-
-%ifidn %1,ss
-    psrad           m6, 6
-    psrad           m7, 6
-%else
-    paddd           m6, m14
-    paddd           m7, m14
-%ifidn %1,pp
-    psrad           m6, 6
-    psrad           m7, 6
-%elifidn %1, sp
-    psrad           m6, INTERP_SHIFT_SP
-    psrad           m7, INTERP_SHIFT_SP
-%else
-    psrad           m6, INTERP_SHIFT_PS
-    psrad           m7, INTERP_SHIFT_PS
-%endif
-%endif
-
-    packssdw        m6, m7
-    vpermq          m6, m6, q3120
-%ifidn %1,pp
-    CLIPW           m6, m5, m3
-%elifidn %1, sp
-    CLIPW           m6, m5, m3
-%endif
-    vextracti128    xm7, m6, 1
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r6], xm7
-
-    movu            xm1, [r0 + r4]                  ; m1 = row 15
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m2, m0, [r5 + 1 * mmsize]
-    paddd           m12, m2
-    pmaddwd         m0, [r5]
-
-    lea             r0, [r0 + r1 * 4]
-    movu            xm2, [r0]                       ; m2 = row 16
-    punpckhwd       xm6, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm6, 1
-    pmaddwd         m6, m1, [r5 + 1 * mmsize]
-    paddd           m13, m6
-    pmaddwd         m1, [r5]
-
-    movu            xm6, [r0 + r1]                  ; m6 = row 17
-    punpckhwd       xm4, xm2, xm6
-    punpcklwd       xm2, xm6
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m2, [r5 + 1 * mmsize]
-    paddd           m0, m2
-
-    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
-    punpckhwd       xm2, xm6, xm4
-    punpcklwd       xm6, xm4
-    vinserti128     m6, m6, xm2, 1
-    pmaddwd         m6, [r5 + 1 * mmsize]
-    paddd           m1, m6
-
-%ifidn %1,ss
-    psrad           m8, 6
-    psrad           m9, 6
-    psrad           m10, 6
-    psrad           m11, 6
-    psrad           m12, 6
-    psrad           m13, 6
-    psrad           m0, 6
-    psrad           m1, 6
-%else
-    paddd           m8, m14
-    paddd           m9, m14
-    paddd           m10, m14
-    paddd           m11, m14
-    paddd           m12, m14
-    paddd           m13, m14
-    paddd           m0, m14
-    paddd           m1, m14
-%ifidn %1,pp
-    psrad           m8, 6
-    psrad           m9, 6
-    psrad           m10, 6
-    psrad           m11, 6
-    psrad           m12, 6
-    psrad           m13, 6
-    psrad           m0, 6
-    psrad           m1, 6
-%elifidn %1, sp
-    psrad           m8, INTERP_SHIFT_SP
-    psrad           m9, INTERP_SHIFT_SP
-    psrad           m10, INTERP_SHIFT_SP
-    psrad           m11, INTERP_SHIFT_SP
-    psrad           m12, INTERP_SHIFT_SP
-    psrad           m13, INTERP_SHIFT_SP
-    psrad           m0, INTERP_SHIFT_SP
-    psrad           m1, INTERP_SHIFT_SP
-%else
-    psrad           m8, INTERP_SHIFT_PS
-    psrad           m9, INTERP_SHIFT_PS
-    psrad           m10, INTERP_SHIFT_PS
-    psrad           m11, INTERP_SHIFT_PS
-    psrad           m12, INTERP_SHIFT_PS
-    psrad           m13, INTERP_SHIFT_PS
-    psrad           m0, INTERP_SHIFT_PS
-    psrad           m1, INTERP_SHIFT_PS
-%endif
-%endif
-
-    packssdw        m8, m9
-    packssdw        m10, m11
-    packssdw        m12, m13
-    packssdw        m0, m1
-    vpermq          m8, m8, q3120
-    vpermq          m10, m10, q3120
-    vpermq          m12, m12, q3120
-    vpermq          m0, m0, q3120
-%ifidn %1,pp
-    CLIPW           m8, m5, m3
-    CLIPW           m10, m5, m3
-    CLIPW           m12, m5, m3
-    CLIPW           m0, m5, m3
-%elifidn %1, sp
-    CLIPW           m8, m5, m3
-    CLIPW           m10, m5, m3
-    CLIPW           m12, m5, m3
-    CLIPW           m0, m5, m3
-%endif
-    vextracti128    xm9, m8, 1
-    vextracti128    xm11, m10, 1
-    vextracti128    xm13, m12, 1
-    vextracti128    xm1, m0, 1
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm8
-    movu            [r2 + r3], xm9
-    movu            [r2 + r3 * 2], xm10
-    movu            [r2 + r6], xm11
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm12
-    movu            [r2 + r3], xm13
-    movu            [r2 + r3 * 2], xm0
-    movu            [r2 + r6], xm1
-    lea             r2, [r2 + r3 * 4]
-    dec             r8d
-    jnz             .loopH
-    RET
-%endif
-%endmacro
-
-FILTER_VER_CHROMA_AVX2_8xN pp, 16
-FILTER_VER_CHROMA_AVX2_8xN ps, 16
-FILTER_VER_CHROMA_AVX2_8xN ss, 16
-FILTER_VER_CHROMA_AVX2_8xN sp, 16
-FILTER_VER_CHROMA_AVX2_8xN pp, 32
-FILTER_VER_CHROMA_AVX2_8xN ps, 32
-FILTER_VER_CHROMA_AVX2_8xN sp, 32
-FILTER_VER_CHROMA_AVX2_8xN ss, 32
-FILTER_VER_CHROMA_AVX2_8xN pp, 64
-FILTER_VER_CHROMA_AVX2_8xN ps, 64
-FILTER_VER_CHROMA_AVX2_8xN sp, 64
-FILTER_VER_CHROMA_AVX2_8xN ss, 64
-
-%macro PROCESS_CHROMA_AVX2_8x2 3
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m2, m2, [r5 + 1 * mmsize]
-    paddd           m0, m2
-
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m3, m3, [r5 + 1 * mmsize]
-    paddd           m1, m3
-
-%ifnidn %1,ss
-    paddd           m0, m7
-    paddd           m1, m7
-%endif
-    psrad           m0, %3
-    psrad           m1, %3
-
-    packssdw        m0, m1
-    vpermq          m0, m0, q3120
-    pxor            m4, m4
-
-%if %2
-    CLIPW           m0, m4, [pw_pixel_max]
-%endif
-    vextracti128    xm1, m0, 1
-%endmacro
-
-
-%macro FILTER_VER_CHROMA_AVX2_8x2 3
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x2, 4, 6, 8
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-    add             r3d, r3d
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,pp
-    vbroadcasti128  m7, [pd_32]
-%elifidn %1, sp
-    vbroadcasti128  m7, [INTERP_OFFSET_SP]
-%else
-    vbroadcasti128  m7, [INTERP_OFFSET_PS]
-%endif
-
-    PROCESS_CHROMA_AVX2_8x2 %1, %2, %3
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    RET
-%endmacro
-
-FILTER_VER_CHROMA_AVX2_8x2 pp, 1, 6
-FILTER_VER_CHROMA_AVX2_8x2 ps, 0, INTERP_SHIFT_PS
-FILTER_VER_CHROMA_AVX2_8x2 sp, 1, INTERP_SHIFT_SP
-FILTER_VER_CHROMA_AVX2_8x2 ss, 0, 6
-
-%macro FILTER_VER_CHROMA_AVX2_4x2 3
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_4x2, 4, 6, 7
-    mov             r4d, r4m
-    add             r1d, r1d
-    add             r3d, r3d
-    shl             r4d, 6
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-
-%ifidn %1,pp
-    vbroadcasti128  m6, [pd_32]
-%elifidn %1, sp
-    vbroadcasti128  m6, [INTERP_OFFSET_SP]
-%else
-    vbroadcasti128  m6, [INTERP_OFFSET_PS]
-%endif
-
-    movq            xm0, [r0]                       ; row 0
-    movq            xm1, [r0 + r1]                  ; row 1
-    punpcklwd       xm0, xm1
-
-    movq            xm2, [r0 + r1 * 2]              ; row 2
-    punpcklwd       xm1, xm2
-    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
-    pmaddwd         m0, [r5]
-
-    movq            xm3, [r0 + r4]                  ; row 3
-    punpcklwd       xm2, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm4, [r0]                       ; row 4
-    punpcklwd       xm3, xm4
-    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
-    pmaddwd         m5, m2, [r5 + 1 * mmsize]
-    paddd           m0, m5
-
-%ifnidn %1, ss
-    paddd           m0, m6
-%endif
-    psrad           m0, %3
-    packssdw        m0, m0
-    pxor            m1, m1
-
-%if %2
-    CLIPW           m0, m1, [pw_pixel_max]
-%endif
-
-    vextracti128    xm2, m0, 1
-    lea             r4, [r3 * 3]
-    movq            [r2], xm0
-    movq            [r2 + r3], xm2
-    RET
-%endmacro
-
-FILTER_VER_CHROMA_AVX2_4x2 pp, 1, 6
-FILTER_VER_CHROMA_AVX2_4x2 ps, 0, INTERP_SHIFT_PS
-FILTER_VER_CHROMA_AVX2_4x2 sp, 1, INTERP_SHIFT_SP
-FILTER_VER_CHROMA_AVX2_4x2 ss, 0, 6
-
-%macro FILTER_VER_CHROMA_AVX2_4x4 3
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_4x4, 4, 6, 7
-    mov             r4d, r4m
-    add             r1d, r1d
-    add             r3d, r3d
-    shl             r4d, 6
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-
-%ifidn %1,pp
-   vbroadcasti128  m6, [pd_32]
-%elifidn %1, sp
-   vbroadcasti128  m6, [INTERP_OFFSET_SP]
-%else
-    vbroadcasti128  m6, [INTERP_OFFSET_PS]
-%endif
-    movq            xm0, [r0]                       ; row 0
-    movq            xm1, [r0 + r1]                  ; row 1
-    punpcklwd       xm0, xm1
-
-    movq            xm2, [r0 + r1 * 2]              ; row 2
-    punpcklwd       xm1, xm2
-    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
-    pmaddwd         m0, [r5]
-
-    movq            xm3, [r0 + r4]                  ; row 3
-    punpcklwd       xm2, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm4, [r0]                       ; row 4
-    punpcklwd       xm3, xm4
-    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
-    pmaddwd         m5, m2, [r5 + 1 * mmsize]
-    pmaddwd         m2, [r5]
-    paddd           m0, m5
-
-    movq            xm3, [r0 + r1]                  ; row 5
-    punpcklwd       xm4, xm3
-    movq            xm1, [r0 + r1 * 2]              ; row 6
-    punpcklwd       xm3, xm1
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
-    pmaddwd         m4, [r5 + 1 * mmsize]
-    paddd           m2, m4
-
-%ifnidn %1,ss
-    paddd           m0, m6
-    paddd           m2, m6
-%endif
-    psrad           m0, %3
-    psrad           m2, %3
-
-    packssdw        m0, m2
-    pxor            m1, m1
-%if %2
-    CLIPW           m0, m1, [pw_pixel_max]
-%endif
-
-    vextracti128    xm2, m0, 1
-    lea             r4, [r3 * 3]
-    movq            [r2], xm0
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r4], xm2
-    RET
-%endmacro
-
-FILTER_VER_CHROMA_AVX2_4x4 pp, 1, 6
-FILTER_VER_CHROMA_AVX2_4x4 ps, 0, INTERP_SHIFT_PS
-FILTER_VER_CHROMA_AVX2_4x4 sp, 1, INTERP_SHIFT_SP
-FILTER_VER_CHROMA_AVX2_4x4 ss, 0, 6
-
-
-%macro FILTER_VER_CHROMA_AVX2_4x8 3
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_4x8, 4, 7, 8
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-    add             r3d, r3d
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-
-%ifidn %1,pp
-    vbroadcasti128  m7, [pd_32]
-%elifidn %1, sp
-    vbroadcasti128  m7, [INTERP_OFFSET_SP]
-%else
-    vbroadcasti128  m7, [INTERP_OFFSET_PS]
-%endif
-    lea             r6, [r3 * 3]
-
-    movq            xm0, [r0]                       ; row 0
-    movq            xm1, [r0 + r1]                  ; row 1
-    punpcklwd       xm0, xm1
-    movq            xm2, [r0 + r1 * 2]              ; row 2
-    punpcklwd       xm1, xm2
-    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
-    pmaddwd         m0, [r5]
-
-    movq            xm3, [r0 + r4]                  ; row 3
-    punpcklwd       xm2, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm4, [r0]                       ; row 4
-    punpcklwd       xm3, xm4
-    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
-    pmaddwd         m5, m2, [r5 + 1 * mmsize]
-    pmaddwd         m2, [r5]
-    paddd           m0, m5
-
-    movq            xm3, [r0 + r1]                  ; row 5
-    punpcklwd       xm4, xm3
-    movq            xm1, [r0 + r1 * 2]              ; row 6
-    punpcklwd       xm3, xm1
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
-    pmaddwd         m5, m4, [r5 + 1 * mmsize]
-    paddd           m2, m5
-    pmaddwd         m4, [r5]
-
-    movq            xm3, [r0 + r4]                  ; row 7
-    punpcklwd       xm1, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm6, [r0]                       ; row 8
-    punpcklwd       xm3, xm6
-    vinserti128     m1, m1, xm3, 1                  ; m1 = [8 7 7 6]
-    pmaddwd         m5, m1, [r5 + 1 * mmsize]
-    paddd           m4, m5
-    pmaddwd         m1, [r5]
-
-    movq            xm3, [r0 + r1]                  ; row 9
-    punpcklwd       xm6, xm3
-    movq            xm5, [r0 + 2 * r1]              ; row 10
-    punpcklwd       xm3, xm5
-    vinserti128     m6, m6, xm3, 1                  ; m6 = [A 9 9 8]
-    pmaddwd         m6, [r5 + 1 * mmsize]
-    paddd           m1, m6
-%ifnidn %1,ss
-    paddd           m0, m7
-    paddd           m2, m7
-%endif
-    psrad           m0, %3
-    psrad           m2, %3
-    packssdw        m0, m2
-    pxor            m6, m6
-    mova            m3, [pw_pixel_max]
-%if %2
-    CLIPW           m0, m6, m3
-%endif
-    vextracti128    xm2, m0, 1
-    movq            [r2], xm0
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r6], xm2
-%ifnidn %1,ss
-    paddd           m4, m7
-    paddd           m1, m7
-%endif
-    psrad           m4, %3
-    psrad           m1, %3
-    packssdw        m4, m1
-%if %2
-    CLIPW           m4, m6, m3
-%endif
-    vextracti128    xm1, m4, 1
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm4
-    movq            [r2 + r3], xm1
-    movhps          [r2 + r3 * 2], xm4
-    movhps          [r2 + r6], xm1
-    RET
-%endmacro
-
-FILTER_VER_CHROMA_AVX2_4x8 pp, 1, 6
-FILTER_VER_CHROMA_AVX2_4x8 ps, 0, INTERP_SHIFT_PS
-FILTER_VER_CHROMA_AVX2_4x8 sp, 1, INTERP_SHIFT_SP
-FILTER_VER_CHROMA_AVX2_4x8 ss, 0 , 6
-
-%macro PROCESS_LUMA_AVX2_W4_16R_4TAP 3
-    movq            xm0, [r0]                       ; row 0
-    movq            xm1, [r0 + r1]                  ; row 1
-    punpcklwd       xm0, xm1
-    movq            xm2, [r0 + r1 * 2]              ; row 2
-    punpcklwd       xm1, xm2
-    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
-    pmaddwd         m0, [r5]
-    movq            xm3, [r0 + r4]                  ; row 3
-    punpcklwd       xm2, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm4, [r0]                       ; row 4
-    punpcklwd       xm3, xm4
-    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
-    pmaddwd         m5, m2, [r5 + 1 * mmsize]
-    pmaddwd         m2, [r5]
-    paddd           m0, m5
-    movq            xm3, [r0 + r1]                  ; row 5
-    punpcklwd       xm4, xm3
-    movq            xm1, [r0 + r1 * 2]              ; row 6
-    punpcklwd       xm3, xm1
-    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
-    pmaddwd         m5, m4, [r5 + 1 * mmsize]
-    paddd           m2, m5
-    pmaddwd         m4, [r5]
-    movq            xm3, [r0 + r4]                  ; row 7
-    punpcklwd       xm1, xm3
-    lea             r0, [r0 + 4 * r1]
-    movq            xm6, [r0]                       ; row 8
-    punpcklwd       xm3, xm6
-    vinserti128     m1, m1, xm3, 1                  ; m1 = [8 7 7 6]
-    pmaddwd         m5, m1, [r5 + 1 * mmsize]
-    paddd           m4, m5
-    pmaddwd         m1, [r5]
-    movq            xm3, [r0 + r1]                  ; row 9
-    punpcklwd       xm6, xm3
-    movq            xm5, [r0 + 2 * r1]              ; row 10
-    punpcklwd       xm3, xm5
-    vinserti128     m6, m6, xm3, 1                  ; m6 = [10 9 9 8]
-    pmaddwd         m3, m6, [r5 + 1 * mmsize]
-    paddd           m1, m3
-    pmaddwd         m6, [r5]
-%ifnidn %1,ss
-    paddd           m0, m7
-    paddd           m2, m7
-%endif
-    psrad           m0, %3
-    psrad           m2, %3
-    packssdw        m0, m2
-    pxor            m3, m3
-%if %2
-    CLIPW           m0, m3, [pw_pixel_max]
-%endif
-    vextracti128    xm2, m0, 1
-    movq            [r2], xm0
-    movq            [r2 + r3], xm2
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r6], xm2
-    movq            xm2, [r0 + r4]                  ;row 11
-    punpcklwd       xm5, xm2
-    lea             r0, [r0 + 4 * r1]
-    movq            xm0, [r0]                       ; row 12
-    punpcklwd       xm2, xm0
-    vinserti128     m5, m5, xm2, 1                  ; m5 = [12 11 11 10]
-    pmaddwd         m2, m5, [r5 + 1 * mmsize]
-    paddd           m6, m2
-    pmaddwd         m5, [r5]
-    movq            xm2, [r0 + r1]                  ; row 13
-    punpcklwd       xm0, xm2
-    movq            xm3, [r0 + 2 * r1]              ; row 14
-    punpcklwd       xm2, xm3
-    vinserti128     m0, m0, xm2, 1                  ; m0 = [14 13 13 12]
-    pmaddwd         m2, m0, [r5 + 1 * mmsize]
-    paddd           m5, m2
-    pmaddwd         m0, [r5]
-%ifnidn %1,ss
-    paddd           m4, m7
-    paddd           m1, m7
-%endif
-    psrad           m4, %3
-    psrad           m1, %3
-    packssdw        m4, m1
-    pxor            m2, m2
-%if %2
-    CLIPW           m4, m2, [pw_pixel_max]
-%endif
-
-    vextracti128    xm1, m4, 1
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm4
-    movq            [r2 + r3], xm1
-    movhps          [r2 + r3 * 2], xm4
-    movhps          [r2 + r6], xm1
-    movq            xm4, [r0 + r4]                  ; row 15
-    punpcklwd       xm3, xm4
-    lea             r0, [r0 + 4 * r1]
-    movq            xm1, [r0]                       ; row 16
-    punpcklwd       xm4, xm1
-    vinserti128     m3, m3, xm4, 1                  ; m3 = [16 15 15 14]
-    pmaddwd         m4, m3, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m3, [r5]
-    movq            xm4, [r0 + r1]                  ; row 17
-    punpcklwd       xm1, xm4
-    movq            xm2, [r0 + 2 * r1]              ; row 18
-    punpcklwd       xm4, xm2
-    vinserti128     m1, m1, xm4, 1                  ; m1 = [18 17 17 16]
-    pmaddwd         m1, [r5 + 1 * mmsize]
-    paddd           m3, m1
-
-%ifnidn %1,ss
-    paddd           m6, m7
-    paddd           m5, m7
-%endif
-    psrad           m6, %3
-    psrad           m5, %3
-    packssdw        m6, m5
-    pxor            m1, m1
-%if %2
-    CLIPW           m6, m1, [pw_pixel_max]
-%endif
-    vextracti128    xm5, m6, 1
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm6
-    movq            [r2 + r3], xm5
-    movhps          [r2 + r3 * 2], xm6
-    movhps          [r2 + r6], xm5
-%ifnidn %1,ss
-    paddd           m0, m7
-    paddd           m3, m7
-%endif
-    psrad           m0, %3
-    psrad           m3, %3
-    packssdw        m0, m3
-%if %2
-    CLIPW           m0, m1, [pw_pixel_max]
-%endif
-    vextracti128    xm3, m0, 1
-    lea             r2, [r2 + r3 * 4]
-    movq            [r2], xm0
-    movq            [r2 + r3], xm3
-    movhps          [r2 + r3 * 2], xm0
-    movhps          [r2 + r6], xm3
-%endmacro
-
-
-%macro FILTER_VER_CHROMA_AVX2_4xN 4
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_4x%2, 4, 8, 8
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-    add             r3d, r3d
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-    mov             r7d, %2 / 16
-%ifidn %1,pp
-    vbroadcasti128  m7, [pd_32]
-%elifidn %1, sp
-    vbroadcasti128  m7, [INTERP_OFFSET_SP]
-%else
-    vbroadcasti128  m7, [INTERP_OFFSET_PS]
-%endif
-    lea             r6, [r3 * 3]
-.loopH:
-    PROCESS_LUMA_AVX2_W4_16R_4TAP %1, %3, %4
-    lea             r2, [r2 + r3 * 4]
-    dec             r7d
-    jnz             .loopH
-    RET
-%endmacro
-
-FILTER_VER_CHROMA_AVX2_4xN pp, 16, 1, 6
-FILTER_VER_CHROMA_AVX2_4xN ps, 16, 0, INTERP_SHIFT_PS
-FILTER_VER_CHROMA_AVX2_4xN sp, 16, 1, INTERP_SHIFT_SP
-FILTER_VER_CHROMA_AVX2_4xN ss, 16, 0, 6
-FILTER_VER_CHROMA_AVX2_4xN pp, 32, 1, 6
-FILTER_VER_CHROMA_AVX2_4xN ps, 32, 0, INTERP_SHIFT_PS
-FILTER_VER_CHROMA_AVX2_4xN sp, 32, 1, INTERP_SHIFT_SP
-FILTER_VER_CHROMA_AVX2_4xN ss, 32, 0, 6
-
-%macro FILTER_VER_CHROMA_AVX2_8x8 3
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_vert_%1_8x8, 4, 6, 12
-    mov             r4d, r4m
-    add             r1d, r1d
-    add             r3d, r3d
-    shl             r4d, 6
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-
-%ifidn %1,pp
-    vbroadcasti128  m11, [pd_32]
-%elifidn %1, sp
-    vbroadcasti128  m11, [INTERP_OFFSET_SP]
-%else
-    vbroadcasti128  m11, [INTERP_OFFSET_PS]
-%endif
-
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    pmaddwd         m2, [r5]
-    paddd           m0, m4                          ; res row0 done(0,1,2,3)
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    pmaddwd         m3, [r5]
-    paddd           m1, m5                          ;res row1 done(1, 2, 3, 4)
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    pmaddwd         m4, [r5]
-    paddd           m2, m6                          ;res row2 done(2,3,4,5)
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm7, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddwd         m7, m5, [r5 + 1 * mmsize]
-    pmaddwd         m5, [r5]
-    paddd           m3, m7                          ;res row3 done(3,4,5,6)
-    movu            xm7, [r0 + r4]                  ; m7 = row 7
-    punpckhwd       xm8, xm6, xm7
-    punpcklwd       xm6, xm7
-    vinserti128     m6, m6, xm8, 1
-    pmaddwd         m8, m6, [r5 + 1 * mmsize]
-    pmaddwd         m6, [r5]
-    paddd           m4, m8                          ;res row4 done(4,5,6,7)
-    lea             r0, [r0 + r1 * 4]
-    movu            xm8, [r0]                       ; m8 = row 8
-    punpckhwd       xm9, xm7, xm8
-    punpcklwd       xm7, xm8
-    vinserti128     m7, m7, xm9, 1
-    pmaddwd         m9, m7, [r5 + 1 * mmsize]
-    pmaddwd         m7, [r5]
-    paddd           m5, m9                          ;res row5 done(5,6,7,8)
-    movu            xm9, [r0 + r1]                  ; m9 = row 9
-    punpckhwd       xm10, xm8, xm9
-    punpcklwd       xm8, xm9
-    vinserti128     m8, m8, xm10, 1
-    pmaddwd         m8, [r5 + 1 * mmsize]
-    paddd           m6, m8                          ;res row6 done(6,7,8,9)
-    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
-    punpckhwd       xm8, xm9, xm10
-    punpcklwd       xm9, xm10
-    vinserti128     m9, m9, xm8, 1
-    pmaddwd         m9, [r5 + 1 * mmsize]
-    paddd           m7, m9                          ;res row7 done 7,8,9,10
-    lea             r4, [r3 * 3]
-%ifnidn %1,ss
-    paddd           m0, m11
-    paddd           m1, m11
-    paddd           m2, m11
-    paddd           m3, m11
-%endif
-    psrad           m0, %3
-    psrad           m1, %3
-    psrad           m2, %3
-    psrad           m3, %3
-    packssdw        m0, m1
-    packssdw        m2, m3
-    vpermq          m0, m0, q3120
-    vpermq          m2, m2, q3120
-    pxor            m1, m1
-    mova            m3, [pw_pixel_max]
-%if %2
-    CLIPW           m0, m1, m3
-    CLIPW           m2, m1, m3
-%endif
-    vextracti128    xm9, m0, 1
-    vextracti128    xm8, m2, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm9
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r4], xm8
-%ifnidn %1,ss
-    paddd           m4, m11
-    paddd           m5, m11
-    paddd           m6, m11
-    paddd           m7, m11
-%endif
-    psrad           m4, %3
-    psrad           m5, %3
-    psrad           m6, %3
-    psrad           m7, %3
-    packssdw        m4, m5
-    packssdw        m6, m7
-    vpermq          m4, m4, q3120
-    vpermq          m6, m6, q3120
-%if %2
-    CLIPW           m4, m1, m3
-    CLIPW           m6, m1, m3
-%endif
-    vextracti128    xm5, m4, 1
-    vextracti128    xm7, m6, 1
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm4
-    movu            [r2 + r3], xm5
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r4], xm7
-    RET
-%endif
-%endmacro
-
-FILTER_VER_CHROMA_AVX2_8x8 pp, 1, 6
-FILTER_VER_CHROMA_AVX2_8x8 ps, 0, INTERP_SHIFT_PS
-FILTER_VER_CHROMA_AVX2_8x8 sp, 1, INTERP_SHIFT_SP
-FILTER_VER_CHROMA_AVX2_8x8 ss, 0, 6
-
-%macro FILTER_VER_CHROMA_AVX2_8x6 3
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_vert_%1_8x6, 4, 6, 12
-    mov             r4d, r4m
-    add             r1d, r1d
-    add             r3d, r3d
-    shl             r4d, 6
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-
-%ifidn %1,pp
-    vbroadcasti128  m11, [pd_32]
-%elifidn %1, sp
-    vbroadcasti128  m11, [INTERP_OFFSET_SP]
-%else
-    vbroadcasti128  m11, [INTERP_OFFSET_PS]
-%endif
-
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    pmaddwd         m2, [r5]
-    paddd           m0, m4                          ; r0 done(0,1,2,3)
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    pmaddwd         m3, [r5]
-    paddd           m1, m5                          ;r1 done(1, 2, 3, 4)
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    pmaddwd         m4, [r5]
-    paddd           m2, m6                          ;r2 done(2,3,4,5)
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm7, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddwd         m7, m5, [r5 + 1 * mmsize]
-    pmaddwd         m5, [r5]
-    paddd           m3, m7                          ;r3 done(3,4,5,6)
-    movu            xm7, [r0 + r4]                  ; m7 = row 7
-    punpckhwd       xm8, xm6, xm7
-    punpcklwd       xm6, xm7
-    vinserti128     m6, m6, xm8, 1
-    pmaddwd         m8, m6, [r5 + 1 * mmsize]
-    paddd           m4, m8                          ;r4 done(4,5,6,7)
-    lea             r0, [r0 + r1 * 4]
-    movu            xm8, [r0]                       ; m8 = row 8
-    punpckhwd       xm9, xm7, xm8
-    punpcklwd       xm7, xm8
-    vinserti128     m7, m7, xm9, 1
-    pmaddwd         m7, m7, [r5 + 1 * mmsize]
-    paddd           m5, m7                          ;r5 done(5,6,7,8)
-    lea             r4, [r3 * 3]
-%ifnidn %1,ss
-    paddd           m0, m11
-    paddd           m1, m11
-    paddd           m2, m11
-    paddd           m3, m11
-%endif
-    psrad           m0, %3
-    psrad           m1, %3
-    psrad           m2, %3
-    psrad           m3, %3
-    packssdw        m0, m1
-    packssdw        m2, m3
-    vpermq          m0, m0, q3120
-    vpermq          m2, m2, q3120
-    pxor            m10, m10
-    mova            m9, [pw_pixel_max]
-%if %2
-    CLIPW           m0, m10, m9
-    CLIPW           m2, m10, m9
-%endif
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r4], xm3
-%ifnidn %1,ss
-    paddd           m4, m11
-    paddd           m5, m11
-%endif
-    psrad           m4, %3
-    psrad           m5, %3
-    packssdw        m4, m5
-    vpermq          m4, m4, 11011000b
-%if %2
-    CLIPW           m4, m10, m9
-%endif
-    vextracti128    xm5, m4, 1
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm4
-    movu            [r2 + r3], xm5
-    RET
-%endif
-%endmacro
-
-FILTER_VER_CHROMA_AVX2_8x6 pp, 1, 6
-FILTER_VER_CHROMA_AVX2_8x6 ps, 0, INTERP_SHIFT_PS
-FILTER_VER_CHROMA_AVX2_8x6 sp, 1, INTERP_SHIFT_SP
-FILTER_VER_CHROMA_AVX2_8x6 ss, 0, 6
-
-%macro PROCESS_CHROMA_AVX2 3
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    paddd           m1, m5
-    pmaddwd         m3, [r5]
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m4, [r5 + 1 * mmsize]
-    paddd           m2, m4
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm4, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm4, 1
-    pmaddwd         m5, [r5 + 1 * mmsize]
-    paddd           m3, m5
-%ifnidn %1,ss
-    paddd           m0, m7
-    paddd           m1, m7
-    paddd           m2, m7
-    paddd           m3, m7
-%endif
-    psrad           m0, %3
-    psrad           m1, %3
-    psrad           m2, %3
-    psrad           m3, %3
-    packssdw        m0, m1
-    packssdw        m2, m3
-    vpermq          m0, m0, q3120
-    vpermq          m2, m2, q3120
-    pxor            m4, m4
-%if %2
-    CLIPW           m0, m4, [pw_pixel_max]
-    CLIPW           m2, m4, [pw_pixel_max]
-%endif
-    vextracti128    xm1, m0, 1
-    vextracti128    xm3, m2, 1
-%endmacro
-
-
-%macro FILTER_VER_CHROMA_AVX2_8x4 3
-INIT_YMM avx2
-cglobal interp_4tap_vert_%1_8x4, 4, 6, 8
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-    add             r3d, r3d
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer + r4]
-%endif
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,pp
-    vbroadcasti128  m7, [pd_32]
-%elifidn %1, sp
-    vbroadcasti128  m7, [INTERP_OFFSET_SP]
-%else
-    vbroadcasti128  m7, [INTERP_OFFSET_PS]
-%endif
-    PROCESS_CHROMA_AVX2 %1, %2, %3
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    movu            [r2 + r3 * 2], xm2
-    lea             r4, [r3 * 3]
-    movu            [r2 + r4], xm3
-    RET
-%endmacro
-
-FILTER_VER_CHROMA_AVX2_8x4 pp, 1, 6
-FILTER_VER_CHROMA_AVX2_8x4 ps, 0, INTERP_SHIFT_PS
-FILTER_VER_CHROMA_AVX2_8x4 sp, 1, INTERP_SHIFT_SP
-FILTER_VER_CHROMA_AVX2_8x4 ss, 0, 6
-
-%macro FILTER_VER_CHROMA_AVX2_8x12 3
-INIT_YMM avx2
-%if ARCH_X86_64 == 1
-cglobal interp_4tap_vert_%1_8x12, 4, 7, 15
-    mov             r4d, r4m
-    shl             r4d, 6
-    add             r1d, r1d
-    add             r3d, r3d
-
-%ifdef PIC
-    lea             r5, [tab_ChromaCoeffVer]
-    add             r5, r4
-%else
-    lea             r5, [tab_ChromaCoeffVer + r4]
-%endif
-
-    lea             r4, [r1 * 3]
-    sub             r0, r1
-%ifidn %1,pp
-    vbroadcasti128  m14, [pd_32]
-%elifidn %1, sp
-    vbroadcasti128  m14, [INTERP_OFFSET_SP]
-%else
-    vbroadcasti128  m14, [INTERP_OFFSET_PS]
-%endif
-    lea             r6, [r3 * 3]
-    movu            xm0, [r0]                       ; m0 = row 0
-    movu            xm1, [r0 + r1]                  ; m1 = row 1
-    punpckhwd       xm2, xm0, xm1
-    punpcklwd       xm0, xm1
-    vinserti128     m0, m0, xm2, 1
-    pmaddwd         m0, [r5]
-    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
-    punpckhwd       xm3, xm1, xm2
-    punpcklwd       xm1, xm2
-    vinserti128     m1, m1, xm3, 1
-    pmaddwd         m1, [r5]
-    movu            xm3, [r0 + r4]                  ; m3 = row 3
-    punpckhwd       xm4, xm2, xm3
-    punpcklwd       xm2, xm3
-    vinserti128     m2, m2, xm4, 1
-    pmaddwd         m4, m2, [r5 + 1 * mmsize]
-    paddd           m0, m4
-    pmaddwd         m2, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm4, [r0]                       ; m4 = row 4
-    punpckhwd       xm5, xm3, xm4
-    punpcklwd       xm3, xm4
-    vinserti128     m3, m3, xm5, 1
-    pmaddwd         m5, m3, [r5 + 1 * mmsize]
-    paddd           m1, m5
-    pmaddwd         m3, [r5]
-    movu            xm5, [r0 + r1]                  ; m5 = row 5
-    punpckhwd       xm6, xm4, xm5
-    punpcklwd       xm4, xm5
-    vinserti128     m4, m4, xm6, 1
-    pmaddwd         m6, m4, [r5 + 1 * mmsize]
-    paddd           m2, m6
-    pmaddwd         m4, [r5]
-    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
-    punpckhwd       xm7, xm5, xm6
-    punpcklwd       xm5, xm6
-    vinserti128     m5, m5, xm7, 1
-    pmaddwd         m7, m5, [r5 + 1 * mmsize]
-    paddd           m3, m7
-    pmaddwd         m5, [r5]
-    movu            xm7, [r0 + r4]                  ; m7 = row 7
-    punpckhwd       xm8, xm6, xm7
-    punpcklwd       xm6, xm7
-    vinserti128     m6, m6, xm8, 1
-    pmaddwd         m8, m6, [r5 + 1 * mmsize]
-    paddd           m4, m8
-    pmaddwd         m6, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm8, [r0]                       ; m8 = row 8
-    punpckhwd       xm9, xm7, xm8
-    punpcklwd       xm7, xm8
-    vinserti128     m7, m7, xm9, 1
-    pmaddwd         m9, m7, [r5 + 1 * mmsize]
-    paddd           m5, m9
-    pmaddwd         m7, [r5]
-    movu            xm9, [r0 + r1]                  ; m9 = row 9
-    punpckhwd       xm10, xm8, xm9
-    punpcklwd       xm8, xm9
-    vinserti128     m8, m8, xm10, 1
-    pmaddwd         m10, m8, [r5 + 1 * mmsize]
-    paddd           m6, m10
-    pmaddwd         m8, [r5]
-    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
-    punpckhwd       xm11, xm9, xm10
-    punpcklwd       xm9, xm10
-    vinserti128     m9, m9, xm11, 1
-    pmaddwd         m11, m9, [r5 + 1 * mmsize]
-    paddd           m7, m11
-    pmaddwd         m9, [r5]
-    movu            xm11, [r0 + r4]                 ; m11 = row 11
-    punpckhwd       xm12, xm10, xm11
-    punpcklwd       xm10, xm11
-    vinserti128     m10, m10, xm12, 1
-    pmaddwd         m12, m10, [r5 + 1 * mmsize]
-    paddd           m8, m12
-    pmaddwd         m10, [r5]
-    lea             r0, [r0 + r1 * 4]
-    movu            xm12, [r0]                      ; m12 = row 12
-    punpckhwd       xm13, xm11, xm12
-    punpcklwd       xm11, xm12
-    vinserti128     m11, m11, xm13, 1
-    pmaddwd         m13, m11, [r5 + 1 * mmsize]
-    paddd           m9, m13
-    pmaddwd         m11, [r5]
-%ifnidn %1,ss
-    paddd           m0, m14
-    paddd           m1, m14
-    paddd           m2, m14
-    paddd           m3, m14
-    paddd           m4, m14
-    paddd           m5, m14
-%endif
-    psrad           m0, %3
-    psrad           m1, %3
-    psrad           m2, %3
-    psrad           m3, %3
-    psrad           m4, %3
-    psrad           m5, %3
-    packssdw        m0, m1
-    packssdw        m2, m3
-    packssdw        m4, m5
-    vpermq          m0, m0, q3120
-    vpermq          m2, m2, q3120
-    vpermq          m4, m4, q3120
-    pxor            m5, m5
-    mova            m3, [pw_pixel_max]
-%if %2
-    CLIPW           m0, m5, m3
-    CLIPW           m2, m5, m3
-    CLIPW           m4, m5, m3
-%endif
-    vextracti128    xm1, m0, 1
-    movu            [r2], xm0
-    movu            [r2 + r3], xm1
-    vextracti128    xm1, m2, 1
-    movu            [r2 + r3 * 2], xm2
-    movu            [r2 + r6], xm1
-    lea             r2, [r2 + r3 * 4]
-    vextracti128    xm1, m4, 1
-    movu            [r2], xm4
-    movu            [r2 + r3], xm1
-    movu            xm13, [r0 + r1]                 ; m13 = row 13
-    punpckhwd       xm0, xm12, xm13
-    punpcklwd       xm12, xm13
-    vinserti128     m12, m12, xm0, 1
-    pmaddwd         m12, m12, [r5 + 1 * mmsize]
-    paddd           m10, m12
-    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
-    punpckhwd       xm1, xm13, xm0
-    punpcklwd       xm13, xm0
-    vinserti128     m13, m13, xm1, 1
-    pmaddwd         m13, m13, [r5 + 1 * mmsize]
-    paddd           m11, m13
-%ifnidn %1,ss
-    paddd           m6, m14
-    paddd           m7, m14
-    paddd           m8, m14
-    paddd           m9, m14
-    paddd           m10, m14
-    paddd           m11, m14
-%endif
-    psrad           m6, %3
-    psrad           m7, %3
-    psrad           m8, %3
-    psrad           m9, %3
-    psrad           m10, %3
-    psrad           m11, %3
-    packssdw        m6, m7
-    packssdw        m8, m9
-    packssdw        m10, m11
-    vpermq          m6, m6, q3120
-    vpermq          m8, m8, q3120
-    vpermq          m10, m10, q3120
-%if %2
-    CLIPW           m6, m5, m3
-    CLIPW           m8, m5, m3
-    CLIPW           m10, m5, m3
-%endif
-    vextracti128    xm7, m6, 1
-    vextracti128    xm9, m8, 1
-    vextracti128    xm11, m10, 1
-    movu            [r2 + r3 * 2], xm6
-    movu            [r2 + r6], xm7
-    lea             r2, [r2 + r3 * 4]
-    movu            [r2], xm8
-    movu            [r2 + r3], xm9
-    movu            [r2 + r3 * 2], xm10
-    movu            [r2 + r6], xm11
-    RET
-%endif
-%endmacro
-
-FILTER_VER_CHROMA_AVX2_8x12 pp, 1, 6
-FILTER_VER_CHROMA_AVX2_8x12 ps, 0, INTERP_SHIFT_PS
-FILTER_VER_CHROMA_AVX2_8x12 sp, 1, INTERP_SHIFT_SP
-FILTER_VER_CHROMA_AVX2_8x12 ss, 0, 6
diff --git a/source/common/x86/v4-ipfilter16.asm b/source/common/x86/v4-ipfilter16.asm
new file mode 100644
index 0000000000..8c51872bc3
--- /dev/null
+++ b/source/common/x86/v4-ipfilter16.asm
@@ -0,0 +1,3529 @@
+;*****************************************************************************
+;* Copyright (C) 2013-2017 MulticoreWare, Inc
+;*
+;* Authors: Nabajit Deka <nabajit@multicorewareinc.com>
+;*          Murugan Vairavel <murugan@multicorewareinc.com>
+;*          Min Chen <chenm003@163.com>
+;*
+;* This program is free software; you can redistribute it and/or modify
+;* it under the terms of the GNU General Public License as published by
+;* the Free Software Foundation; either version 2 of the License, or
+;* (at your option) any later version.
+;*
+;* This program is distributed in the hope that it will be useful,
+;* but WITHOUT ANY WARRANTY; without even the implied warranty of
+;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+;* GNU General Public License for more details.
+;*
+;* You should have received a copy of the GNU General Public License
+;* along with this program; if not, write to the Free Software
+;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
+;*
+;* This program is also available under a commercial proprietary license.
+;* For more information, contact us at license @ x265.com.
+;*****************************************************************************/
+
+%include "x86inc.asm"
+%include "x86util.asm"
+
+
+%define INTERP_OFFSET_PP        pd_32
+%define INTERP_SHIFT_PP         6
+
+%if BIT_DEPTH == 10
+    %define INTERP_SHIFT_PS         2
+    %define INTERP_OFFSET_PS        pd_n32768
+    %define INTERP_SHIFT_SP         10
+    %define INTERP_OFFSET_SP        v4_pd_524800
+%elif BIT_DEPTH == 12
+    %define INTERP_SHIFT_PS         4
+    %define INTERP_OFFSET_PS        pd_n131072
+    %define INTERP_SHIFT_SP         8
+    %define INTERP_OFFSET_SP        pd_524416
+%else
+    %error Unsupport bit depth!
+%endif
+
+
+SECTION_RODATA 32
+
+v4_pd_524800:        times 8 dd 524800
+tab_c_n8192:      times 8 dw -8192
+
+const tab_ChromaCoeffV,  times 8 dw 0, 64
+                         times 8 dw 0, 0
+
+                         times 8 dw -2, 58
+                         times 8 dw 10, -2
+
+                         times 8 dw -4, 54
+                         times 8 dw 16, -2
+
+                         times 8 dw -6, 46
+                         times 8 dw 28, -4
+
+                         times 8 dw -4, 36
+                         times 8 dw 36, -4
+
+                         times 8 dw -4, 28
+                         times 8 dw 46, -6
+
+                         times 8 dw -2, 16
+                         times 8 dw 54, -4
+
+                         times 8 dw -2, 10
+                         times 8 dw 58, -2
+                        
+tab_ChromaCoeffVer: times 8 dw 0, 64
+                    times 8 dw 0, 0
+
+                    times 8 dw -2, 58
+                    times 8 dw 10, -2
+
+                    times 8 dw -4, 54
+                    times 8 dw 16, -2
+
+                    times 8 dw -6, 46
+                    times 8 dw 28, -4
+
+                    times 8 dw -4, 36
+                    times 8 dw 36, -4
+
+                    times 8 dw -4, 28
+                    times 8 dw 46, -6
+
+                    times 8 dw -2, 16
+                    times 8 dw 54, -4
+
+                    times 8 dw -2, 10
+                    times 8 dw 58, -2
+                        
+SECTION .text
+cextern pd_8
+cextern pd_32
+cextern pw_pixel_max
+cextern pd_524416
+cextern pd_n32768
+cextern pd_n131072
+cextern pw_2000
+cextern idct8_shuf2
+
+%macro PROCESS_CHROMA_SP_W4_4R 0
+    movq       m0, [r0]
+    movq       m1, [r0 + r1]
+    punpcklwd  m0, m1                          ;m0=[0 1]
+    pmaddwd    m0, [r6 + 0 *32]                ;m0=[0+1]         Row1
+
+    lea        r0, [r0 + 2 * r1]
+    movq       m4, [r0]
+    punpcklwd  m1, m4                          ;m1=[1 2]
+    pmaddwd    m1, [r6 + 0 *32]                ;m1=[1+2]         Row2
+
+    movq       m5, [r0 + r1]
+    punpcklwd  m4, m5                          ;m4=[2 3]
+    pmaddwd    m2, m4, [r6 + 0 *32]            ;m2=[2+3]         Row3
+    pmaddwd    m4, [r6 + 1 * 32]
+    paddd      m0, m4                          ;m0=[0+1+2+3]     Row1 done
+
+    lea        r0, [r0 + 2 * r1]
+    movq       m4, [r0]
+    punpcklwd  m5, m4                          ;m5=[3 4]
+    pmaddwd    m3, m5, [r6 + 0 *32]            ;m3=[3+4]         Row4
+    pmaddwd    m5, [r6 + 1 * 32]
+    paddd      m1, m5                          ;m1 = [1+2+3+4]   Row2
+
+    movq       m5, [r0 + r1]
+    punpcklwd  m4, m5                          ;m4=[4 5]
+    pmaddwd    m4, [r6 + 1 * 32]
+    paddd      m2, m4                          ;m2=[2+3+4+5]     Row3
+
+    movq       m4, [r0 + 2 * r1]
+    punpcklwd  m5, m4                          ;m5=[5 6]
+    pmaddwd    m5, [r6 + 1 * 32]
+    paddd      m3, m5                          ;m3=[3+4+5+6]     Row4
+%endmacro
+
+;-----------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_%3_%1x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_SS 4
+INIT_XMM sse2
+cglobal interp_4tap_vert_%3_%1x%2, 5, 7, %4 ,0-gprsize
+
+    add       r1d, r1d
+    add       r3d, r3d
+    sub       r0, r1
+    shl       r4d, 6
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r6, [r5 + r4]
+%else
+    lea       r6, [tab_ChromaCoeffV + r4]
+%endif
+
+    mov       dword [rsp], %2/4
+
+%ifnidn %3, ss
+    %ifnidn %3, ps
+        mova      m7, [pw_pixel_max]
+        %ifidn %3, pp
+            mova      m6, [INTERP_OFFSET_PP]
+        %else
+            mova      m6, [INTERP_OFFSET_SP]
+        %endif
+    %else
+        mova      m6, [INTERP_OFFSET_PS]
+    %endif
+%endif
+
+.loopH:
+    mov       r4d, (%1/4)
+.loopW:
+    PROCESS_CHROMA_SP_W4_4R
+
+%ifidn %3, ss
+    psrad     m0, 6
+    psrad     m1, 6
+    psrad     m2, 6
+    psrad     m3, 6
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+%elifidn %3, ps
+    paddd     m0, m6
+    paddd     m1, m6
+    paddd     m2, m6
+    paddd     m3, m6
+    psrad     m0, INTERP_SHIFT_PS
+    psrad     m1, INTERP_SHIFT_PS
+    psrad     m2, INTERP_SHIFT_PS
+    psrad     m3, INTERP_SHIFT_PS
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+%else
+    paddd     m0, m6
+    paddd     m1, m6
+    paddd     m2, m6
+    paddd     m3, m6
+    %ifidn %3, pp
+        psrad     m0, INTERP_SHIFT_PP
+        psrad     m1, INTERP_SHIFT_PP
+        psrad     m2, INTERP_SHIFT_PP
+        psrad     m3, INTERP_SHIFT_PP
+    %else
+        psrad     m0, INTERP_SHIFT_SP
+        psrad     m1, INTERP_SHIFT_SP
+        psrad     m2, INTERP_SHIFT_SP
+        psrad     m3, INTERP_SHIFT_SP
+    %endif
+    packssdw  m0, m1
+    packssdw  m2, m3
+    pxor      m5, m5
+    CLIPW2    m0, m2, m5, m7
+%endif
+
+    movh      [r2], m0
+    movhps    [r2 + r3], m0
+    lea       r5, [r2 + 2 * r3]
+    movh      [r5], m2
+    movhps    [r5 + r3], m2
+
+    lea       r5, [4 * r1 - 2 * 4]
+    sub       r0, r5
+    add       r2, 2 * 4
+
+    dec       r4d
+    jnz       .loopW
+
+    lea       r0, [r0 + 4 * r1 - 2 * %1]
+    lea       r2, [r2 + 4 * r3 - 2 * %1]
+
+    dec       dword [rsp]
+    jnz       .loopH
+
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_SS 4, 4, ss, 6
+    FILTER_VER_CHROMA_SS 4, 8, ss, 6
+    FILTER_VER_CHROMA_SS 16, 16, ss, 6
+    FILTER_VER_CHROMA_SS 16, 8, ss, 6
+    FILTER_VER_CHROMA_SS 16, 12, ss, 6
+    FILTER_VER_CHROMA_SS 12, 16, ss, 6
+    FILTER_VER_CHROMA_SS 16, 4, ss, 6
+    FILTER_VER_CHROMA_SS 4, 16, ss, 6
+    FILTER_VER_CHROMA_SS 32, 32, ss, 6
+    FILTER_VER_CHROMA_SS 32, 16, ss, 6
+    FILTER_VER_CHROMA_SS 16, 32, ss, 6
+    FILTER_VER_CHROMA_SS 32, 24, ss, 6
+    FILTER_VER_CHROMA_SS 24, 32, ss, 6
+    FILTER_VER_CHROMA_SS 32, 8, ss, 6
+
+    FILTER_VER_CHROMA_SS 4, 4, ps, 7
+    FILTER_VER_CHROMA_SS 4, 8, ps, 7
+    FILTER_VER_CHROMA_SS 16, 16, ps, 7
+    FILTER_VER_CHROMA_SS 16, 8, ps, 7
+    FILTER_VER_CHROMA_SS 16, 12, ps, 7
+    FILTER_VER_CHROMA_SS 12, 16, ps, 7
+    FILTER_VER_CHROMA_SS 16, 4, ps, 7
+    FILTER_VER_CHROMA_SS 4, 16, ps, 7
+    FILTER_VER_CHROMA_SS 32, 32, ps, 7
+    FILTER_VER_CHROMA_SS 32, 16, ps, 7
+    FILTER_VER_CHROMA_SS 16, 32, ps, 7
+    FILTER_VER_CHROMA_SS 32, 24, ps, 7
+    FILTER_VER_CHROMA_SS 24, 32, ps, 7
+    FILTER_VER_CHROMA_SS 32, 8, ps, 7
+
+    FILTER_VER_CHROMA_SS 4, 4, sp, 8
+    FILTER_VER_CHROMA_SS 4, 8, sp, 8
+    FILTER_VER_CHROMA_SS 16, 16, sp, 8
+    FILTER_VER_CHROMA_SS 16, 8, sp, 8
+    FILTER_VER_CHROMA_SS 16, 12, sp, 8
+    FILTER_VER_CHROMA_SS 12, 16, sp, 8
+    FILTER_VER_CHROMA_SS 16, 4, sp, 8
+    FILTER_VER_CHROMA_SS 4, 16, sp, 8
+    FILTER_VER_CHROMA_SS 32, 32, sp, 8
+    FILTER_VER_CHROMA_SS 32, 16, sp, 8
+    FILTER_VER_CHROMA_SS 16, 32, sp, 8
+    FILTER_VER_CHROMA_SS 32, 24, sp, 8
+    FILTER_VER_CHROMA_SS 24, 32, sp, 8
+    FILTER_VER_CHROMA_SS 32, 8, sp, 8
+
+    FILTER_VER_CHROMA_SS 4, 4, pp, 8
+    FILTER_VER_CHROMA_SS 4, 8, pp, 8
+    FILTER_VER_CHROMA_SS 16, 16, pp, 8
+    FILTER_VER_CHROMA_SS 16, 8, pp, 8
+    FILTER_VER_CHROMA_SS 16, 12, pp, 8
+    FILTER_VER_CHROMA_SS 12, 16, pp, 8
+    FILTER_VER_CHROMA_SS 16, 4, pp, 8
+    FILTER_VER_CHROMA_SS 4, 16, pp, 8
+    FILTER_VER_CHROMA_SS 32, 32, pp, 8
+    FILTER_VER_CHROMA_SS 32, 16, pp, 8
+    FILTER_VER_CHROMA_SS 16, 32, pp, 8
+    FILTER_VER_CHROMA_SS 32, 24, pp, 8
+    FILTER_VER_CHROMA_SS 24, 32, pp, 8
+    FILTER_VER_CHROMA_SS 32, 8, pp, 8
+
+
+    FILTER_VER_CHROMA_SS 16, 24, ss, 6
+    FILTER_VER_CHROMA_SS 12, 32, ss, 6
+    FILTER_VER_CHROMA_SS 4, 32, ss, 6
+    FILTER_VER_CHROMA_SS 32, 64, ss, 6
+    FILTER_VER_CHROMA_SS 16, 64, ss, 6
+    FILTER_VER_CHROMA_SS 32, 48, ss, 6
+    FILTER_VER_CHROMA_SS 24, 64, ss, 6
+
+    FILTER_VER_CHROMA_SS 16, 24, ps, 7
+    FILTER_VER_CHROMA_SS 12, 32, ps, 7
+    FILTER_VER_CHROMA_SS 4, 32, ps, 7
+    FILTER_VER_CHROMA_SS 32, 64, ps, 7
+    FILTER_VER_CHROMA_SS 16, 64, ps, 7
+    FILTER_VER_CHROMA_SS 32, 48, ps, 7
+    FILTER_VER_CHROMA_SS 24, 64, ps, 7
+
+    FILTER_VER_CHROMA_SS 16, 24, sp, 8
+    FILTER_VER_CHROMA_SS 12, 32, sp, 8
+    FILTER_VER_CHROMA_SS 4, 32, sp, 8
+    FILTER_VER_CHROMA_SS 32, 64, sp, 8
+    FILTER_VER_CHROMA_SS 16, 64, sp, 8
+    FILTER_VER_CHROMA_SS 32, 48, sp, 8
+    FILTER_VER_CHROMA_SS 24, 64, sp, 8
+
+    FILTER_VER_CHROMA_SS 16, 24, pp, 8
+    FILTER_VER_CHROMA_SS 12, 32, pp, 8
+    FILTER_VER_CHROMA_SS 4, 32, pp, 8
+    FILTER_VER_CHROMA_SS 32, 64, pp, 8
+    FILTER_VER_CHROMA_SS 16, 64, pp, 8
+    FILTER_VER_CHROMA_SS 32, 48, pp, 8
+    FILTER_VER_CHROMA_SS 24, 64, pp, 8
+
+
+    FILTER_VER_CHROMA_SS 48, 64, ss, 6
+    FILTER_VER_CHROMA_SS 64, 48, ss, 6
+    FILTER_VER_CHROMA_SS 64, 64, ss, 6
+    FILTER_VER_CHROMA_SS 64, 32, ss, 6
+    FILTER_VER_CHROMA_SS 64, 16, ss, 6
+
+    FILTER_VER_CHROMA_SS 48, 64, ps, 7
+    FILTER_VER_CHROMA_SS 64, 48, ps, 7
+    FILTER_VER_CHROMA_SS 64, 64, ps, 7
+    FILTER_VER_CHROMA_SS 64, 32, ps, 7
+    FILTER_VER_CHROMA_SS 64, 16, ps, 7
+
+    FILTER_VER_CHROMA_SS 48, 64, sp, 8
+    FILTER_VER_CHROMA_SS 64, 48, sp, 8
+    FILTER_VER_CHROMA_SS 64, 64, sp, 8
+    FILTER_VER_CHROMA_SS 64, 32, sp, 8
+    FILTER_VER_CHROMA_SS 64, 16, sp, 8
+
+    FILTER_VER_CHROMA_SS 48, 64, pp, 8
+    FILTER_VER_CHROMA_SS 64, 48, pp, 8
+    FILTER_VER_CHROMA_SS 64, 64, pp, 8
+    FILTER_VER_CHROMA_SS 64, 32, pp, 8
+    FILTER_VER_CHROMA_SS 64, 16, pp, 8
+
+
+%macro PROCESS_CHROMA_SP_W2_4R 1
+    movd       m0, [r0]
+    movd       m1, [r0 + r1]
+    punpcklwd  m0, m1                          ;m0=[0 1]
+
+    lea        r0, [r0 + 2 * r1]
+    movd       m2, [r0]
+    punpcklwd  m1, m2                          ;m1=[1 2]
+    punpcklqdq m0, m1                          ;m0=[0 1 1 2]
+    pmaddwd    m0, [%1 + 0 *32]                ;m0=[0+1 1+2] Row 1-2
+
+    movd       m1, [r0 + r1]
+    punpcklwd  m2, m1                          ;m2=[2 3]
+
+    lea        r0, [r0 + 2 * r1]
+    movd       m3, [r0]
+    punpcklwd  m1, m3                          ;m2=[3 4]
+    punpcklqdq m2, m1                          ;m2=[2 3 3 4]
+
+    pmaddwd    m4, m2, [%1 + 1 * 32]           ;m4=[2+3 3+4] Row 1-2
+    pmaddwd    m2, [%1 + 0 * 32]               ;m2=[2+3 3+4] Row 3-4
+    paddd      m0, m4                          ;m0=[0+1+2+3 1+2+3+4] Row 1-2
+
+    movd       m1, [r0 + r1]
+    punpcklwd  m3, m1                          ;m3=[4 5]
+
+    movd       m4, [r0 + 2 * r1]
+    punpcklwd  m1, m4                          ;m1=[5 6]
+    punpcklqdq m3, m1                          ;m2=[4 5 5 6]
+    pmaddwd    m3, [%1 + 1 * 32]               ;m3=[4+5 5+6] Row 3-4
+    paddd      m2, m3                          ;m2=[2+3+4+5 3+4+5+6] Row 3-4
+%endmacro
+;---------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vertical_%2_2x%1(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;---------------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_W2 3
+INIT_XMM sse4
+cglobal interp_4tap_vert_%2_2x%1, 5, 6, %3
+
+    add       r1d, r1d
+    add       r3d, r3d
+    sub       r0, r1
+    shl       r4d, 6
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r5, [r5 + r4]
+%else
+    lea       r5, [tab_ChromaCoeffV + r4]
+%endif
+
+    mov       r4d, (%1/4)
+%ifnidn %2, ss
+    %ifnidn %2, ps
+        pxor      m7, m7
+        mova      m6, [pw_pixel_max]
+        %ifidn %2, pp
+            mova      m5, [INTERP_OFFSET_PP]
+        %else
+            mova      m5, [INTERP_OFFSET_SP]
+        %endif
+    %else
+        mova      m5, [INTERP_OFFSET_PS]
+    %endif
+%endif
+
+.loopH:
+    PROCESS_CHROMA_SP_W2_4R r5
+%ifidn %2, ss
+    psrad     m0, 6
+    psrad     m2, 6
+    packssdw  m0, m2
+%elifidn %2, ps
+    paddd     m0, m5
+    paddd     m2, m5
+    psrad     m0, INTERP_SHIFT_PS
+    psrad     m2, INTERP_SHIFT_PS
+    packssdw  m0, m2
+%else
+    paddd     m0, m5
+    paddd     m2, m5
+    %ifidn %2, pp
+        psrad     m0, INTERP_SHIFT_PP
+        psrad     m2, INTERP_SHIFT_PP
+    %else
+        psrad     m0, INTERP_SHIFT_SP
+        psrad     m2, INTERP_SHIFT_SP
+    %endif
+    packusdw  m0, m2
+    CLIPW     m0, m7,    m6
+%endif
+
+    movd      [r2], m0
+    pextrd    [r2 + r3], m0, 1
+    lea       r2, [r2 + 2 * r3]
+    pextrd    [r2], m0, 2
+    pextrd    [r2 + r3], m0, 3
+
+    lea       r2, [r2 + 2 * r3]
+
+    dec       r4d
+    jnz       .loopH
+    RET
+%endmacro
+
+FILTER_VER_CHROMA_W2 4, ss, 5
+FILTER_VER_CHROMA_W2 8, ss, 5
+
+FILTER_VER_CHROMA_W2 4, pp, 8
+FILTER_VER_CHROMA_W2 8, pp, 8
+
+FILTER_VER_CHROMA_W2 4, ps, 6
+FILTER_VER_CHROMA_W2 8, ps, 6
+
+FILTER_VER_CHROMA_W2 4, sp, 8
+FILTER_VER_CHROMA_W2 8, sp, 8
+
+FILTER_VER_CHROMA_W2 16, ss, 5
+FILTER_VER_CHROMA_W2 16, pp, 8
+FILTER_VER_CHROMA_W2 16, ps, 6
+FILTER_VER_CHROMA_W2 16, sp, 8
+
+
+;---------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_%1_4x2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;---------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_W4 3
+INIT_XMM sse4
+cglobal interp_4tap_vert_%2_4x%1, 5, 6, %3
+    add        r1d, r1d
+    add        r3d, r3d
+    sub        r0, r1
+    shl        r4d, 6
+
+%ifdef PIC
+    lea        r5, [tab_ChromaCoeffV]
+    lea        r5, [r5 + r4]
+%else
+    lea        r5, [tab_ChromaCoeffV + r4]
+%endif
+
+%ifnidn %2, 2
+    mov        r4d, %1/2
+%endif
+
+%ifnidn %2, ss
+    %ifnidn %2, ps
+        pxor      m6, m6
+        mova      m5, [pw_pixel_max]
+        %ifidn %2, pp
+            mova      m4, [INTERP_OFFSET_PP]
+        %else
+            mova      m4, [INTERP_OFFSET_SP]
+        %endif
+    %else
+        mova      m4, [INTERP_OFFSET_PS]
+    %endif
+%endif
+
+%ifnidn %2, 2
+.loop:
+%endif
+
+    movh       m0, [r0]
+    movh       m1, [r0 + r1]
+    punpcklwd  m0, m1                          ;m0=[0 1]
+    pmaddwd    m0, [r5 + 0 *32]                ;m0=[0+1]  Row1
+
+    lea        r0, [r0 + 2 * r1]
+    movh       m2, [r0]
+    punpcklwd  m1, m2                          ;m1=[1 2]
+    pmaddwd    m1, [r5 + 0 *32]                ;m1=[1+2]  Row2
+
+    movh       m3, [r0 + r1]
+    punpcklwd  m2, m3                          ;m4=[2 3]
+    pmaddwd    m2, [r5 + 1 * 32]
+    paddd      m0, m2                          ;m0=[0+1+2+3]  Row1 done
+
+    movh       m2, [r0 + 2 * r1]
+    punpcklwd  m3, m2                          ;m5=[3 4]
+    pmaddwd    m3, [r5 + 1 * 32]
+    paddd      m1, m3                          ;m1=[1+2+3+4]  Row2 done
+
+%ifidn %2, ss
+    psrad     m0, 6
+    psrad     m1, 6
+    packssdw  m0, m1
+%elifidn %2, ps
+    paddd     m0, m4
+    paddd     m1, m4
+    psrad     m0, INTERP_SHIFT_PS
+    psrad     m1, INTERP_SHIFT_PS
+    packssdw  m0, m1
+%else
+    paddd     m0, m4
+    paddd     m1, m4
+    %ifidn %2, pp
+        psrad     m0, INTERP_SHIFT_PP
+        psrad     m1, INTERP_SHIFT_PP
+    %else
+        psrad     m0, INTERP_SHIFT_SP
+        psrad     m1, INTERP_SHIFT_SP
+    %endif
+    packusdw  m0, m1
+    CLIPW     m0, m6,    m5
+%endif
+
+    movh       [r2], m0
+    movhps     [r2 + r3], m0
+
+%ifnidn %2, 2
+    lea        r2, [r2 + r3 * 2]
+    dec        r4d
+    jnz        .loop
+%endif
+    RET
+%endmacro
+
+FILTER_VER_CHROMA_W4 2, ss, 4
+FILTER_VER_CHROMA_W4 2, pp, 7
+FILTER_VER_CHROMA_W4 2, ps, 5
+FILTER_VER_CHROMA_W4 2, sp, 7
+
+FILTER_VER_CHROMA_W4 4, ss, 4
+FILTER_VER_CHROMA_W4 4, pp, 7
+FILTER_VER_CHROMA_W4 4, ps, 5
+FILTER_VER_CHROMA_W4 4, sp, 7
+
+;-------------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vertical_%1_6x8(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;-------------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_W6 3
+INIT_XMM sse4
+cglobal interp_4tap_vert_%2_6x%1, 5, 7, %3
+    add       r1d, r1d
+    add       r3d, r3d
+    sub       r0, r1
+    shl       r4d, 6
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r6, [r5 + r4]
+%else
+    lea       r6, [tab_ChromaCoeffV + r4]
+%endif
+
+    mov       r4d, %1/4
+
+%ifnidn %2, ss
+    %ifnidn %2, ps
+        mova      m7, [pw_pixel_max]
+        %ifidn %2, pp
+            mova      m6, [INTERP_OFFSET_PP]
+        %else
+            mova      m6, [INTERP_OFFSET_SP]
+        %endif
+    %else
+        mova      m6, [INTERP_OFFSET_PS]
+    %endif
+%endif
+
+.loopH:
+    PROCESS_CHROMA_SP_W4_4R
+
+%ifidn %2, ss
+    psrad     m0, 6
+    psrad     m1, 6
+    psrad     m2, 6
+    psrad     m3, 6
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+%elifidn %2, ps
+    paddd     m0, m6
+    paddd     m1, m6
+    paddd     m2, m6
+    paddd     m3, m6
+    psrad     m0, INTERP_SHIFT_PS
+    psrad     m1, INTERP_SHIFT_PS
+    psrad     m2, INTERP_SHIFT_PS
+    psrad     m3, INTERP_SHIFT_PS
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+%else
+    paddd     m0, m6
+    paddd     m1, m6
+    paddd     m2, m6
+    paddd     m3, m6
+    %ifidn %2, pp
+        psrad     m0, INTERP_SHIFT_PP
+        psrad     m1, INTERP_SHIFT_PP
+        psrad     m2, INTERP_SHIFT_PP
+        psrad     m3, INTERP_SHIFT_PP
+    %else
+        psrad     m0, INTERP_SHIFT_SP
+        psrad     m1, INTERP_SHIFT_SP
+        psrad     m2, INTERP_SHIFT_SP
+        psrad     m3, INTERP_SHIFT_SP
+    %endif
+    packssdw  m0, m1
+    packssdw  m2, m3
+    pxor      m5, m5
+    CLIPW2    m0, m2, m5, m7
+%endif
+
+    movh      [r2], m0
+    movhps    [r2 + r3], m0
+    lea       r5, [r2 + 2 * r3]
+    movh      [r5], m2
+    movhps    [r5 + r3], m2
+
+    lea       r5, [4 * r1 - 2 * 4]
+    sub       r0, r5
+    add       r2, 2 * 4
+
+    PROCESS_CHROMA_SP_W2_4R r6
+
+%ifidn %2, ss
+    psrad     m0, 6
+    psrad     m2, 6
+    packssdw  m0, m2
+%elifidn %2, ps
+    paddd     m0, m6
+    paddd     m2, m6
+    psrad     m0, INTERP_SHIFT_PS
+    psrad     m2, INTERP_SHIFT_PS
+    packssdw  m0, m2
+%else
+    paddd     m0, m6
+    paddd     m2, m6
+    %ifidn %2, pp
+        psrad     m0, INTERP_SHIFT_PP
+        psrad     m2, INTERP_SHIFT_PP
+    %else
+        psrad     m0, INTERP_SHIFT_SP
+        psrad     m2, INTERP_SHIFT_SP
+    %endif
+    packusdw  m0, m2
+    CLIPW     m0, m5,    m7
+%endif
+
+    movd      [r2], m0
+    pextrd    [r2 + r3], m0, 1
+    lea       r2, [r2 + 2 * r3]
+    pextrd    [r2], m0, 2
+    pextrd    [r2 + r3], m0, 3
+
+    sub       r0, 2 * 4
+    lea       r2, [r2 + 2 * r3 - 2 * 4]
+
+    dec       r4d
+    jnz       .loopH
+    RET
+%endmacro
+
+FILTER_VER_CHROMA_W6 8, ss, 6
+FILTER_VER_CHROMA_W6 8, ps, 7
+FILTER_VER_CHROMA_W6 8, sp, 8
+FILTER_VER_CHROMA_W6 8, pp, 8
+
+FILTER_VER_CHROMA_W6 16, ss, 6
+FILTER_VER_CHROMA_W6 16, ps, 7
+FILTER_VER_CHROMA_W6 16, sp, 8
+FILTER_VER_CHROMA_W6 16, pp, 8
+
+%macro PROCESS_CHROMA_SP_W8_2R 0
+    movu       m1, [r0]
+    movu       m3, [r0 + r1]
+    punpcklwd  m0, m1, m3
+    pmaddwd    m0, [r5 + 0 * 32]                ;m0 = [0l+1l]  Row1l
+    punpckhwd  m1, m3
+    pmaddwd    m1, [r5 + 0 * 32]                ;m1 = [0h+1h]  Row1h
+
+    movu       m4, [r0 + 2 * r1]
+    punpcklwd  m2, m3, m4
+    pmaddwd    m2, [r5 + 0 * 32]                ;m2 = [1l+2l]  Row2l
+    punpckhwd  m3, m4
+    pmaddwd    m3, [r5 + 0 * 32]                ;m3 = [1h+2h]  Row2h
+
+    lea        r0, [r0 + 2 * r1]
+    movu       m5, [r0 + r1]
+    punpcklwd  m6, m4, m5
+    pmaddwd    m6, [r5 + 1 * 32]                ;m6 = [2l+3l]  Row1l
+    paddd      m0, m6                           ;m0 = [0l+1l+2l+3l]  Row1l sum
+    punpckhwd  m4, m5
+    pmaddwd    m4, [r5 + 1 * 32]                ;m6 = [2h+3h]  Row1h
+    paddd      m1, m4                           ;m1 = [0h+1h+2h+3h]  Row1h sum
+
+    movu       m4, [r0 + 2 * r1]
+    punpcklwd  m6, m5, m4
+    pmaddwd    m6, [r5 + 1 * 32]                ;m6 = [3l+4l]  Row2l
+    paddd      m2, m6                           ;m2 = [1l+2l+3l+4l]  Row2l sum
+    punpckhwd  m5, m4
+    pmaddwd    m5, [r5 + 1 * 32]                ;m1 = [3h+4h]  Row2h
+    paddd      m3, m5                           ;m3 = [1h+2h+3h+4h]  Row2h sum
+%endmacro
+
+;----------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert_%3_%1x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;----------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_W8 4
+INIT_XMM sse2
+cglobal interp_4tap_vert_%3_%1x%2, 5, 6, %4
+
+    add       r1d, r1d
+    add       r3d, r3d
+    sub       r0, r1
+    shl       r4d, 6
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r5, [r5 + r4]
+%else
+    lea       r5, [tab_ChromaCoeffV + r4]
+%endif
+
+    mov       r4d, %2/2
+
+%ifidn %3, pp
+    mova      m7, [INTERP_OFFSET_PP]
+%elifidn %3, sp
+    mova      m7, [INTERP_OFFSET_SP]
+%elifidn %3, ps
+    mova      m7, [INTERP_OFFSET_PS]
+%endif
+
+.loopH:
+    PROCESS_CHROMA_SP_W8_2R
+
+%ifidn %3, ss
+    psrad     m0, 6
+    psrad     m1, 6
+    psrad     m2, 6
+    psrad     m3, 6
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+%elifidn %3, ps
+    paddd     m0, m7
+    paddd     m1, m7
+    paddd     m2, m7
+    paddd     m3, m7
+    psrad     m0, INTERP_SHIFT_PS
+    psrad     m1, INTERP_SHIFT_PS
+    psrad     m2, INTERP_SHIFT_PS
+    psrad     m3, INTERP_SHIFT_PS
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+%else
+    paddd     m0, m7
+    paddd     m1, m7
+    paddd     m2, m7
+    paddd     m3, m7
+    %ifidn %3, pp
+        psrad     m0, INTERP_SHIFT_PP
+        psrad     m1, INTERP_SHIFT_PP
+        psrad     m2, INTERP_SHIFT_PP
+        psrad     m3, INTERP_SHIFT_PP
+    %else
+        psrad     m0, INTERP_SHIFT_SP
+        psrad     m1, INTERP_SHIFT_SP
+        psrad     m2, INTERP_SHIFT_SP
+        psrad     m3, INTERP_SHIFT_SP
+    %endif
+    packssdw  m0, m1
+    packssdw  m2, m3
+    pxor      m5, m5
+    mova      m6, [pw_pixel_max]
+    CLIPW2    m0, m2, m5, m6
+%endif
+
+    movu      [r2], m0
+    movu      [r2 + r3], m2
+
+    lea       r2, [r2 + 2 * r3]
+
+    dec       r4d
+    jnz       .loopH
+    RET
+%endmacro
+
+FILTER_VER_CHROMA_W8 8, 2, ss, 7
+FILTER_VER_CHROMA_W8 8, 4, ss, 7
+FILTER_VER_CHROMA_W8 8, 6, ss, 7
+FILTER_VER_CHROMA_W8 8, 8, ss, 7
+FILTER_VER_CHROMA_W8 8, 16, ss, 7
+FILTER_VER_CHROMA_W8 8, 32, ss, 7
+
+FILTER_VER_CHROMA_W8 8, 2, sp, 8
+FILTER_VER_CHROMA_W8 8, 4, sp, 8
+FILTER_VER_CHROMA_W8 8, 6, sp, 8
+FILTER_VER_CHROMA_W8 8, 8, sp, 8
+FILTER_VER_CHROMA_W8 8, 16, sp, 8
+FILTER_VER_CHROMA_W8 8, 32, sp, 8
+
+FILTER_VER_CHROMA_W8 8, 2, ps, 8
+FILTER_VER_CHROMA_W8 8, 4, ps, 8
+FILTER_VER_CHROMA_W8 8, 6, ps, 8
+FILTER_VER_CHROMA_W8 8, 8, ps, 8
+FILTER_VER_CHROMA_W8 8, 16, ps, 8
+FILTER_VER_CHROMA_W8 8, 32, ps, 8
+
+FILTER_VER_CHROMA_W8 8, 2, pp, 8
+FILTER_VER_CHROMA_W8 8, 4, pp, 8
+FILTER_VER_CHROMA_W8 8, 6, pp, 8
+FILTER_VER_CHROMA_W8 8, 8, pp, 8
+FILTER_VER_CHROMA_W8 8, 16, pp, 8
+FILTER_VER_CHROMA_W8 8, 32, pp, 8
+
+FILTER_VER_CHROMA_W8 8, 12, ss, 7
+FILTER_VER_CHROMA_W8 8, 64, ss, 7
+FILTER_VER_CHROMA_W8 8, 12, sp, 8
+FILTER_VER_CHROMA_W8 8, 64, sp, 8
+FILTER_VER_CHROMA_W8 8, 12, ps, 8
+FILTER_VER_CHROMA_W8 8, 64, ps, 8
+FILTER_VER_CHROMA_W8 8, 12, pp, 8
+FILTER_VER_CHROMA_W8 8, 64, pp, 8
+
+%macro PROCESS_CHROMA_VERT_W16_2R 0
+    movu       m1, [r0]
+    movu       m3, [r0 + r1]
+    punpcklwd  m0, m1, m3
+    pmaddwd    m0, [r5 + 0 * 32]
+    punpckhwd  m1, m3
+    pmaddwd    m1, [r5 + 0 * 32]
+
+    movu       m4, [r0 + 2 * r1]
+    punpcklwd  m2, m3, m4
+    pmaddwd    m2, [r5 + 0 * 32]
+    punpckhwd  m3, m4
+    pmaddwd    m3, [r5 + 0 * 32]
+
+    lea        r0, [r0 + 2 * r1]
+    movu       m5, [r0 + r1]
+    punpcklwd  m6, m4, m5
+    pmaddwd    m6, [r5 + 1 * 32]
+    paddd      m0, m6
+    punpckhwd  m4, m5
+    pmaddwd    m4, [r5 + 1 * 32]
+    paddd      m1, m4
+
+    movu       m4, [r0 + 2 * r1]
+    punpcklwd  m6, m5, m4
+    pmaddwd    m6, [r5 + 1 * 32]
+    paddd      m2, m6
+    punpckhwd  m5, m4
+    pmaddwd    m5, [r5 + 1 * 32]
+    paddd      m3, m5
+%endmacro
+
+;-----------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_AVX2_6xN 2
+INIT_YMM avx2
+%if ARCH_X86_64
+cglobal interp_4tap_vert_%2_6x%1, 4, 7, 10
+    mov             r4d, r4m
+    add             r1d, r1d
+    add             r3d, r3d
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffV + r4]
+%endif
+
+    sub             r0, r1
+    mov             r6d, %1/4
+
+%ifidn %2,pp
+    vbroadcasti128  m8, [INTERP_OFFSET_PP]
+%elifidn %2, sp
+    vbroadcasti128  m8, [INTERP_OFFSET_SP]
+%else
+    vbroadcasti128  m8, [INTERP_OFFSET_PS]
+%endif
+
+.loopH:
+    movu            xm0, [r0]
+    movu            xm1, [r0 + r1]
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+
+    movu            xm2, [r0 + r1 * 2]
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+
+    lea             r4, [r1 * 3]
+    movu            xm3, [r0 + r4]
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    pmaddwd         m2, [r5]
+    paddd           m0, m4
+
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    pmaddwd         m3, [r5]
+    paddd           m1, m5
+
+    movu            xm5, [r0 + r1]
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    pmaddwd         m4, [r5]
+    paddd           m2, m6
+
+    movu            xm6, [r0 + r1 * 2]
+    punpckhwd       xm7, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddwd         m7, m5, [r5 + 1 * mmsize]
+    pmaddwd         m5, [r5]
+    paddd           m3, m7
+    lea             r4, [r3 * 3]
+%ifidn %2,ss
+    psrad           m0, 6
+    psrad           m1, 6
+    psrad           m2, 6
+    psrad           m3, 6
+%else
+    paddd           m0, m8
+    paddd           m1, m8
+    paddd           m2, m8
+    paddd           m3, m8
+%ifidn %2,pp
+    psrad           m0, INTERP_SHIFT_PP
+    psrad           m1, INTERP_SHIFT_PP
+    psrad           m2, INTERP_SHIFT_PP
+    psrad           m3, INTERP_SHIFT_PP
+%elifidn %2, sp
+    psrad           m0, INTERP_SHIFT_SP
+    psrad           m1, INTERP_SHIFT_SP
+    psrad           m2, INTERP_SHIFT_SP
+    psrad           m3, INTERP_SHIFT_SP
+%else
+    psrad           m0, INTERP_SHIFT_PS
+    psrad           m1, INTERP_SHIFT_PS
+    psrad           m2, INTERP_SHIFT_PS
+    psrad           m3, INTERP_SHIFT_PS
+%endif
+%endif
+
+    packssdw        m0, m1
+    packssdw        m2, m3
+    vpermq          m0, m0, q3120
+    vpermq          m2, m2, q3120
+    pxor            m5, m5
+    mova            m9, [pw_pixel_max]
+%ifidn %2,pp
+    CLIPW           m0, m5, m9
+    CLIPW           m2, m5, m9
+%elifidn %2, sp
+    CLIPW           m0, m5, m9
+    CLIPW           m2, m5, m9
+%endif
+
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m2, 1
+    movq            [r2], xm0
+    pextrd          [r2 + 8], xm0, 2
+    movq            [r2 + r3], xm1
+    pextrd          [r2 + r3 + 8], xm1, 2
+    movq            [r2 + r3 * 2], xm2
+    pextrd          [r2 + r3 * 2 + 8], xm2, 2
+    movq            [r2 + r4], xm3
+    pextrd          [r2 + r4 + 8], xm3, 2
+
+    lea             r2, [r2 + r3 * 4]
+    dec r6d
+    jnz .loopH
+    RET
+%endif
+%endmacro
+FILTER_VER_CHROMA_AVX2_6xN 8, pp
+FILTER_VER_CHROMA_AVX2_6xN 8, ps
+FILTER_VER_CHROMA_AVX2_6xN 8, ss
+FILTER_VER_CHROMA_AVX2_6xN 8, sp
+FILTER_VER_CHROMA_AVX2_6xN 16, pp
+FILTER_VER_CHROMA_AVX2_6xN 16, ps
+FILTER_VER_CHROMA_AVX2_6xN 16, ss
+FILTER_VER_CHROMA_AVX2_6xN 16, sp
+
+;-----------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_W16_16xN_avx2 3
+INIT_YMM avx2
+cglobal interp_4tap_vert_%2_16x%1, 5, 6, %3
+    add       r1d, r1d
+    add       r3d, r3d
+    sub       r0, r1
+    shl       r4d, 6
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r5, [r5 + r4]
+%else
+    lea       r5, [tab_ChromaCoeffV + r4]
+%endif
+
+    mov       r4d, %1/2
+
+%ifidn %2, pp
+    vbroadcasti128  m7, [INTERP_OFFSET_PP]
+%elifidn %2, sp
+    vbroadcasti128  m7, [INTERP_OFFSET_SP]
+%elifidn %2, ps
+    vbroadcasti128  m7, [INTERP_OFFSET_PS]
+%endif
+
+.loopH:
+    PROCESS_CHROMA_VERT_W16_2R
+%ifidn %2, ss
+    psrad     m0, 6
+    psrad     m1, 6
+    psrad     m2, 6
+    psrad     m3, 6
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+%elifidn %2, ps
+    paddd     m0, m7
+    paddd     m1, m7
+    paddd     m2, m7
+    paddd     m3, m7
+    psrad     m0, INTERP_SHIFT_PS
+    psrad     m1, INTERP_SHIFT_PS
+    psrad     m2, INTERP_SHIFT_PS
+    psrad     m3, INTERP_SHIFT_PS
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+%else
+    paddd     m0, m7
+    paddd     m1, m7
+    paddd     m2, m7
+    paddd     m3, m7
+ %ifidn %2, pp
+    psrad     m0, INTERP_SHIFT_PP
+    psrad     m1, INTERP_SHIFT_PP
+    psrad     m2, INTERP_SHIFT_PP
+    psrad     m3, INTERP_SHIFT_PP
+%else
+    psrad     m0, INTERP_SHIFT_SP
+    psrad     m1, INTERP_SHIFT_SP
+    psrad     m2, INTERP_SHIFT_SP
+    psrad     m3, INTERP_SHIFT_SP
+%endif
+    packssdw  m0, m1
+    packssdw  m2, m3
+    pxor      m5, m5
+    CLIPW2    m0, m2, m5, [pw_pixel_max]
+%endif
+
+    movu      [r2], m0
+    movu      [r2 + r3], m2
+    lea       r2, [r2 + 2 * r3]
+    dec       r4d
+    jnz       .loopH
+    RET
+%endmacro
+    FILTER_VER_CHROMA_W16_16xN_avx2 4, pp, 8
+    FILTER_VER_CHROMA_W16_16xN_avx2 8, pp, 8
+    FILTER_VER_CHROMA_W16_16xN_avx2 12, pp, 8
+    FILTER_VER_CHROMA_W16_16xN_avx2 24, pp, 8
+    FILTER_VER_CHROMA_W16_16xN_avx2 16, pp, 8
+    FILTER_VER_CHROMA_W16_16xN_avx2 32, pp, 8
+    FILTER_VER_CHROMA_W16_16xN_avx2 64, pp, 8
+
+    FILTER_VER_CHROMA_W16_16xN_avx2 4, ps, 8
+    FILTER_VER_CHROMA_W16_16xN_avx2 8, ps, 8
+    FILTER_VER_CHROMA_W16_16xN_avx2 12, ps, 8
+    FILTER_VER_CHROMA_W16_16xN_avx2 24, ps, 8
+    FILTER_VER_CHROMA_W16_16xN_avx2 16, ps, 8
+    FILTER_VER_CHROMA_W16_16xN_avx2 32, ps, 8
+    FILTER_VER_CHROMA_W16_16xN_avx2 64, ps, 8
+
+    FILTER_VER_CHROMA_W16_16xN_avx2 4, ss, 7
+    FILTER_VER_CHROMA_W16_16xN_avx2 8, ss, 7
+    FILTER_VER_CHROMA_W16_16xN_avx2 12, ss, 7
+    FILTER_VER_CHROMA_W16_16xN_avx2 24, ss, 7
+    FILTER_VER_CHROMA_W16_16xN_avx2 16, ss, 7
+    FILTER_VER_CHROMA_W16_16xN_avx2 32, ss, 7
+    FILTER_VER_CHROMA_W16_16xN_avx2 64, ss, 7
+
+    FILTER_VER_CHROMA_W16_16xN_avx2 4, sp, 8
+    FILTER_VER_CHROMA_W16_16xN_avx2 8, sp, 8
+    FILTER_VER_CHROMA_W16_16xN_avx2 12, sp, 8
+    FILTER_VER_CHROMA_W16_16xN_avx2 24, sp, 8
+    FILTER_VER_CHROMA_W16_16xN_avx2 16, sp, 8
+    FILTER_VER_CHROMA_W16_16xN_avx2 32, sp, 8
+    FILTER_VER_CHROMA_W16_16xN_avx2 64, sp, 8
+
+%macro PROCESS_CHROMA_VERT_W32_2R 0
+    movu       m1, [r0]
+    movu       m3, [r0 + r1]
+    punpcklwd  m0, m1, m3
+    pmaddwd    m0, [r5 + 0 * mmsize]
+    punpckhwd  m1, m3
+    pmaddwd    m1, [r5 + 0 * mmsize]
+
+    movu       m9, [r0 + mmsize]
+    movu       m11, [r0 + r1 + mmsize]
+    punpcklwd  m8, m9, m11
+    pmaddwd    m8, [r5 + 0 * mmsize]
+    punpckhwd  m9, m11
+    pmaddwd    m9, [r5 + 0 * mmsize]
+
+    movu       m4, [r0 + 2 * r1]
+    punpcklwd  m2, m3, m4
+    pmaddwd    m2, [r5 + 0 * mmsize]
+    punpckhwd  m3, m4
+    pmaddwd    m3, [r5 + 0 * mmsize]
+
+    movu       m12, [r0 + 2 * r1 + mmsize]
+    punpcklwd  m10, m11, m12
+    pmaddwd    m10, [r5 + 0 * mmsize]
+    punpckhwd  m11, m12
+    pmaddwd    m11, [r5 + 0 * mmsize]
+
+    lea        r6, [r0 + 2 * r1]
+    movu       m5, [r6 + r1]
+    punpcklwd  m6, m4, m5
+    pmaddwd    m6, [r5 + 1 * mmsize]
+    paddd      m0, m6
+    punpckhwd  m4, m5
+    pmaddwd    m4, [r5 + 1 * mmsize]
+    paddd      m1, m4
+
+    movu       m13, [r6 + r1 + mmsize]
+    punpcklwd  m14, m12, m13
+    pmaddwd    m14, [r5 + 1 * mmsize]
+    paddd      m8, m14
+    punpckhwd  m12, m13
+    pmaddwd    m12, [r5 + 1 * mmsize]
+    paddd      m9, m12
+
+    movu       m4, [r6 + 2 * r1]
+    punpcklwd  m6, m5, m4
+    pmaddwd    m6, [r5 + 1 * mmsize]
+    paddd      m2, m6
+    punpckhwd  m5, m4
+    pmaddwd    m5, [r5 + 1 * mmsize]
+    paddd      m3, m5
+
+    movu       m12, [r6 + 2 * r1 + mmsize]
+    punpcklwd  m14, m13, m12
+    pmaddwd    m14, [r5 + 1 * mmsize]
+    paddd      m10, m14
+    punpckhwd  m13, m12
+    pmaddwd    m13, [r5 + 1 * mmsize]
+    paddd      m11, m13
+%endmacro
+
+;-----------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_W16_32xN_avx2 3
+INIT_YMM avx2
+%if ARCH_X86_64
+cglobal interp_4tap_vert_%2_32x%1, 5, 7, %3
+    add       r1d, r1d
+    add       r3d, r3d
+    sub       r0, r1
+    shl       r4d, 6
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r5, [r5 + r4]
+%else
+    lea       r5, [tab_ChromaCoeffV + r4]
+%endif
+    mov       r4d, %1/2
+
+%ifidn %2, pp
+    vbroadcasti128  m7, [INTERP_OFFSET_PP]
+%elifidn %2, sp
+    vbroadcasti128  m7, [INTERP_OFFSET_SP]
+%elifidn %2, ps
+    vbroadcasti128  m7, [INTERP_OFFSET_PS]
+%endif
+
+.loopH:
+    PROCESS_CHROMA_VERT_W32_2R
+%ifidn %2, ss
+    psrad     m0, 6
+    psrad     m1, 6
+    psrad     m2, 6
+    psrad     m3, 6
+
+    psrad     m8, 6
+    psrad     m9, 6
+    psrad     m10, 6
+    psrad     m11, 6
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+    packssdw  m8, m9
+    packssdw  m10, m11
+%elifidn %2, ps
+    paddd     m0, m7
+    paddd     m1, m7
+    paddd     m2, m7
+    paddd     m3, m7
+    psrad     m0, INTERP_SHIFT_PS
+    psrad     m1, INTERP_SHIFT_PS
+    psrad     m2, INTERP_SHIFT_PS
+    psrad     m3, INTERP_SHIFT_PS
+    paddd     m8, m7
+    paddd     m9, m7
+    paddd     m10, m7
+    paddd     m11, m7
+    psrad     m8, INTERP_SHIFT_PS
+    psrad     m9, INTERP_SHIFT_PS
+    psrad     m10, INTERP_SHIFT_PS
+    psrad     m11, INTERP_SHIFT_PS
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+    packssdw  m8, m9
+    packssdw  m10, m11
+%else
+    paddd     m0, m7
+    paddd     m1, m7
+    paddd     m2, m7
+    paddd     m3, m7
+    paddd     m8, m7
+    paddd     m9, m7
+    paddd     m10, m7
+    paddd     m11, m7
+ %ifidn %2, pp
+    psrad     m0, INTERP_SHIFT_PP
+    psrad     m1, INTERP_SHIFT_PP
+    psrad     m2, INTERP_SHIFT_PP
+    psrad     m3, INTERP_SHIFT_PP
+    psrad     m8, INTERP_SHIFT_PP
+    psrad     m9, INTERP_SHIFT_PP
+    psrad     m10, INTERP_SHIFT_PP
+    psrad     m11, INTERP_SHIFT_PP
+%else
+    psrad     m0, INTERP_SHIFT_SP
+    psrad     m1, INTERP_SHIFT_SP
+    psrad     m2, INTERP_SHIFT_SP
+    psrad     m3, INTERP_SHIFT_SP
+    psrad     m8, INTERP_SHIFT_SP
+    psrad     m9, INTERP_SHIFT_SP
+    psrad     m10, INTERP_SHIFT_SP
+    psrad     m11, INTERP_SHIFT_SP
+%endif
+    packssdw  m0, m1
+    packssdw  m2, m3
+    packssdw  m8, m9
+    packssdw  m10, m11
+    pxor      m5, m5
+    CLIPW2    m0, m2, m5, [pw_pixel_max]
+    CLIPW2    m8, m10, m5, [pw_pixel_max]
+%endif
+
+    movu      [r2], m0
+    movu      [r2 + r3], m2
+    movu      [r2 + mmsize], m8
+    movu      [r2 + r3 + mmsize], m10
+    lea       r2, [r2 + 2 * r3]
+    lea       r0, [r0 + 2 * r1]
+    dec       r4d
+    jnz       .loopH
+    RET
+%endif
+%endmacro
+    FILTER_VER_CHROMA_W16_32xN_avx2 8, pp, 15
+    FILTER_VER_CHROMA_W16_32xN_avx2 16, pp, 15
+    FILTER_VER_CHROMA_W16_32xN_avx2 24, pp, 15
+    FILTER_VER_CHROMA_W16_32xN_avx2 32, pp, 15
+    FILTER_VER_CHROMA_W16_32xN_avx2 48, pp, 15
+    FILTER_VER_CHROMA_W16_32xN_avx2 64, pp, 15
+
+    FILTER_VER_CHROMA_W16_32xN_avx2 8, ps, 15
+    FILTER_VER_CHROMA_W16_32xN_avx2 16, ps, 15
+    FILTER_VER_CHROMA_W16_32xN_avx2 24, ps, 15
+    FILTER_VER_CHROMA_W16_32xN_avx2 32, ps, 15
+    FILTER_VER_CHROMA_W16_32xN_avx2 48, ps, 15
+    FILTER_VER_CHROMA_W16_32xN_avx2 64, ps, 15
+
+    FILTER_VER_CHROMA_W16_32xN_avx2 8, ss, 15
+    FILTER_VER_CHROMA_W16_32xN_avx2 16, ss, 15
+    FILTER_VER_CHROMA_W16_32xN_avx2 24, ss, 15
+    FILTER_VER_CHROMA_W16_32xN_avx2 32, ss, 15
+    FILTER_VER_CHROMA_W16_32xN_avx2 48, ss, 15
+    FILTER_VER_CHROMA_W16_32xN_avx2 64, ss, 15
+
+    FILTER_VER_CHROMA_W16_32xN_avx2 8, sp, 15
+    FILTER_VER_CHROMA_W16_32xN_avx2 16, sp, 15
+    FILTER_VER_CHROMA_W16_32xN_avx2 24, sp, 15
+    FILTER_VER_CHROMA_W16_32xN_avx2 32, sp, 15
+    FILTER_VER_CHROMA_W16_32xN_avx2 48, sp, 15
+    FILTER_VER_CHROMA_W16_32xN_avx2 64, sp, 15
+
+;-----------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_W16_64xN_avx2 3
+INIT_YMM avx2
+cglobal interp_4tap_vert_%2_64x%1, 5, 7, %3
+    add       r1d, r1d
+    add       r3d, r3d
+    sub       r0, r1
+    shl       r4d, 6
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r5, [r5 + r4]
+%else
+    lea       r5, [tab_ChromaCoeffV + r4]
+%endif
+    mov       r4d, %1/2
+
+%ifidn %2, pp
+    vbroadcasti128  m7, [INTERP_OFFSET_PP]
+%elifidn %2, sp
+    vbroadcasti128  m7, [INTERP_OFFSET_SP]
+%elifidn %2, ps
+    vbroadcasti128  m7, [INTERP_OFFSET_PS]
+%endif
+
+.loopH:
+%assign x 0
+%rep 4
+    movu       m1, [r0 + x]
+    movu       m3, [r0 + r1 + x]
+    movu       m5, [r5 + 0 * mmsize]
+    punpcklwd  m0, m1, m3
+    pmaddwd    m0, m5
+    punpckhwd  m1, m3
+    pmaddwd    m1, m5
+
+    movu       m4, [r0 + 2 * r1 + x]
+    punpcklwd  m2, m3, m4
+    pmaddwd    m2, m5
+    punpckhwd  m3, m4
+    pmaddwd    m3, m5
+
+    lea        r6, [r0 + 2 * r1]
+    movu       m5, [r6 + r1 + x]
+    punpcklwd  m6, m4, m5
+    pmaddwd    m6, [r5 + 1 * mmsize]
+    paddd      m0, m6
+    punpckhwd  m4, m5
+    pmaddwd    m4, [r5 + 1 * mmsize]
+    paddd      m1, m4
+
+    movu       m4, [r6 + 2 * r1 + x]
+    punpcklwd  m6, m5, m4
+    pmaddwd    m6, [r5 + 1 * mmsize]
+    paddd      m2, m6
+    punpckhwd  m5, m4
+    pmaddwd    m5, [r5 + 1 * mmsize]
+    paddd      m3, m5
+
+%ifidn %2, ss
+    psrad     m0, 6
+    psrad     m1, 6
+    psrad     m2, 6
+    psrad     m3, 6
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+%elifidn %2, ps
+    paddd     m0, m7
+    paddd     m1, m7
+    paddd     m2, m7
+    paddd     m3, m7
+    psrad     m0, INTERP_SHIFT_PS
+    psrad     m1, INTERP_SHIFT_PS
+    psrad     m2, INTERP_SHIFT_PS
+    psrad     m3, INTERP_SHIFT_PS
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+%else
+    paddd     m0, m7
+    paddd     m1, m7
+    paddd     m2, m7
+    paddd     m3, m7
+%ifidn %2, pp
+    psrad     m0, INTERP_SHIFT_PP
+    psrad     m1, INTERP_SHIFT_PP
+    psrad     m2, INTERP_SHIFT_PP
+    psrad     m3, INTERP_SHIFT_PP
+%else
+    psrad     m0, INTERP_SHIFT_SP
+    psrad     m1, INTERP_SHIFT_SP
+    psrad     m2, INTERP_SHIFT_SP
+    psrad     m3, INTERP_SHIFT_SP
+%endif
+    packssdw  m0, m1
+    packssdw  m2, m3
+    pxor      m5, m5
+    CLIPW2    m0, m2, m5, [pw_pixel_max]
+%endif
+
+    movu      [r2 + x], m0
+    movu      [r2 + r3 + x], m2
+%assign x x+mmsize
+%endrep
+
+    lea       r2, [r2 + 2 * r3]
+    lea       r0, [r0 + 2 * r1]
+    dec       r4d
+    jnz       .loopH
+    RET
+%endmacro
+    FILTER_VER_CHROMA_W16_64xN_avx2 16, ss, 7
+    FILTER_VER_CHROMA_W16_64xN_avx2 32, ss, 7
+    FILTER_VER_CHROMA_W16_64xN_avx2 48, ss, 7
+    FILTER_VER_CHROMA_W16_64xN_avx2 64, ss, 7
+    FILTER_VER_CHROMA_W16_64xN_avx2 16, sp, 8
+    FILTER_VER_CHROMA_W16_64xN_avx2 32, sp, 8
+    FILTER_VER_CHROMA_W16_64xN_avx2 48, sp, 8
+    FILTER_VER_CHROMA_W16_64xN_avx2 64, sp, 8
+    FILTER_VER_CHROMA_W16_64xN_avx2 16, ps, 8
+    FILTER_VER_CHROMA_W16_64xN_avx2 32, ps, 8
+    FILTER_VER_CHROMA_W16_64xN_avx2 48, ps, 8
+    FILTER_VER_CHROMA_W16_64xN_avx2 64, ps, 8
+    FILTER_VER_CHROMA_W16_64xN_avx2 16, pp, 8
+    FILTER_VER_CHROMA_W16_64xN_avx2 32, pp, 8
+    FILTER_VER_CHROMA_W16_64xN_avx2 48, pp, 8
+    FILTER_VER_CHROMA_W16_64xN_avx2 64, pp, 8
+
+;-----------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_W16_12xN_avx2 3
+INIT_YMM avx2
+cglobal interp_4tap_vert_%2_12x%1, 5, 8, %3
+    add       r1d, r1d
+    add       r3d, r3d
+    sub       r0, r1
+    shl       r4d, 6
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r5, [r5 + r4]
+%else
+    lea       r5, [tab_ChromaCoeffV + r4]
+%endif
+    mov       r4d, %1/2
+
+%ifidn %2, pp
+    vbroadcasti128  m7, [INTERP_OFFSET_PP]
+%elifidn %2, sp
+    vbroadcasti128  m7, [INTERP_OFFSET_SP]
+%elifidn %2, ps
+    vbroadcasti128  m7, [INTERP_OFFSET_PS]
+%endif
+
+.loopH:
+    PROCESS_CHROMA_VERT_W16_2R
+%ifidn %2, ss
+    psrad     m0, 6
+    psrad     m1, 6
+    psrad     m2, 6
+    psrad     m3, 6
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+%elifidn %2, ps
+    paddd     m0, m7
+    paddd     m1, m7
+    paddd     m2, m7
+    paddd     m3, m7
+    psrad     m0, INTERP_SHIFT_PS
+    psrad     m1, INTERP_SHIFT_PS
+    psrad     m2, INTERP_SHIFT_PS
+    psrad     m3, INTERP_SHIFT_PS
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+%else
+    paddd     m0, m7
+    paddd     m1, m7
+    paddd     m2, m7
+    paddd     m3, m7
+ %ifidn %2, pp
+    psrad     m0, INTERP_SHIFT_PP
+    psrad     m1, INTERP_SHIFT_PP
+    psrad     m2, INTERP_SHIFT_PP
+    psrad     m3, INTERP_SHIFT_PP
+%else
+    psrad     m0, INTERP_SHIFT_SP
+    psrad     m1, INTERP_SHIFT_SP
+    psrad     m2, INTERP_SHIFT_SP
+    psrad     m3, INTERP_SHIFT_SP
+%endif
+    packssdw  m0, m1
+    packssdw  m2, m3
+    pxor      m5, m5
+    CLIPW2    m0, m2, m5, [pw_pixel_max]
+%endif
+
+    movu      [r2], xm0
+    movu      [r2 + r3], xm2
+    vextracti128 xm0, m0, 1
+    vextracti128 xm2, m2, 1
+    movq      [r2 + 16], xm0
+    movq      [r2 + r3 + 16], xm2
+    lea       r2, [r2 + 2 * r3]
+    dec       r4d
+    jnz       .loopH
+    RET
+%endmacro
+    FILTER_VER_CHROMA_W16_12xN_avx2 16, ss, 7
+    FILTER_VER_CHROMA_W16_12xN_avx2 16, sp, 8
+    FILTER_VER_CHROMA_W16_12xN_avx2 16, ps, 8
+    FILTER_VER_CHROMA_W16_12xN_avx2 16, pp, 8
+    FILTER_VER_CHROMA_W16_12xN_avx2 32, ss, 7
+    FILTER_VER_CHROMA_W16_12xN_avx2 32, sp, 8
+    FILTER_VER_CHROMA_W16_12xN_avx2 32, ps, 8
+    FILTER_VER_CHROMA_W16_12xN_avx2 32, pp, 8
+
+%macro PROCESS_CHROMA_VERT_W24_2R 0
+    movu       m1, [r0]
+    movu       m3, [r0 + r1]
+    punpcklwd  m0, m1, m3
+    pmaddwd    m0, [r5 + 0 * mmsize]
+    punpckhwd  m1, m3
+    pmaddwd    m1, [r5 + 0 * mmsize]
+
+    movu       xm9, [r0 + mmsize]
+    movu       xm11, [r0 + r1 + mmsize]
+    punpcklwd  xm8, xm9, xm11
+    pmaddwd    xm8, [r5 + 0 * mmsize]
+    punpckhwd  xm9, xm11
+    pmaddwd    xm9, [r5 + 0 * mmsize]
+
+    movu       m4, [r0 + 2 * r1]
+    punpcklwd  m2, m3, m4
+    pmaddwd    m2, [r5 + 0 * mmsize]
+    punpckhwd  m3, m4
+    pmaddwd    m3, [r5 + 0 * mmsize]
+
+    movu       xm12, [r0 + 2 * r1 + mmsize]
+    punpcklwd  xm10, xm11, xm12
+    pmaddwd    xm10, [r5 + 0 * mmsize]
+    punpckhwd  xm11, xm12
+    pmaddwd    xm11, [r5 + 0 * mmsize]
+
+    lea        r6, [r0 + 2 * r1]
+    movu       m5, [r6 + r1]
+    punpcklwd  m6, m4, m5
+    pmaddwd    m6, [r5 + 1 * mmsize]
+    paddd      m0, m6
+    punpckhwd  m4, m5
+    pmaddwd    m4, [r5 + 1 * mmsize]
+    paddd      m1, m4
+
+    movu       xm13, [r6 + r1 + mmsize]
+    punpcklwd  xm14, xm12, xm13
+    pmaddwd    xm14, [r5 + 1 * mmsize]
+    paddd      xm8, xm14
+    punpckhwd  xm12, xm13
+    pmaddwd    xm12, [r5 + 1 * mmsize]
+    paddd      xm9, xm12
+
+    movu       m4, [r6 + 2 * r1]
+    punpcklwd  m6, m5, m4
+    pmaddwd    m6, [r5 + 1 * mmsize]
+    paddd      m2, m6
+    punpckhwd  m5, m4
+    pmaddwd    m5, [r5 + 1 * mmsize]
+    paddd      m3, m5
+
+    movu       xm12, [r6 + 2 * r1 + mmsize]
+    punpcklwd  xm14, xm13, xm12
+    pmaddwd    xm14, [r5 + 1 * mmsize]
+    paddd      xm10, xm14
+    punpckhwd  xm13, xm12
+    pmaddwd    xm13, [r5 + 1 * mmsize]
+    paddd      xm11, xm13
+%endmacro
+
+;-----------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_W16_24xN_avx2 3
+INIT_YMM avx2
+%if ARCH_X86_64
+cglobal interp_4tap_vert_%2_24x%1, 5, 7, %3
+    add       r1d, r1d
+    add       r3d, r3d
+    sub       r0, r1
+    shl       r4d, 6
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r5, [r5 + r4]
+%else
+    lea       r5, [tab_ChromaCoeffV + r4]
+%endif
+    mov       r4d, %1/2
+
+%ifidn %2, pp
+    vbroadcasti128  m7, [INTERP_OFFSET_PP]
+%elifidn %2, sp
+    vbroadcasti128  m7, [INTERP_OFFSET_SP]
+%elifidn %2, ps
+    vbroadcasti128  m7, [INTERP_OFFSET_PS]
+%endif
+
+.loopH:
+    PROCESS_CHROMA_VERT_W24_2R
+%ifidn %2, ss
+    psrad     m0, 6
+    psrad     m1, 6
+    psrad     m2, 6
+    psrad     m3, 6
+
+    psrad     m8, 6
+    psrad     m9, 6
+    psrad     m10, 6
+    psrad     m11, 6
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+    packssdw  m8, m9
+    packssdw  m10, m11
+%elifidn %2, ps
+    paddd     m0, m7
+    paddd     m1, m7
+    paddd     m2, m7
+    paddd     m3, m7
+    psrad     m0, INTERP_SHIFT_PS
+    psrad     m1, INTERP_SHIFT_PS
+    psrad     m2, INTERP_SHIFT_PS
+    psrad     m3, INTERP_SHIFT_PS
+    paddd     m8, m7
+    paddd     m9, m7
+    paddd     m10, m7
+    paddd     m11, m7
+    psrad     m8, INTERP_SHIFT_PS
+    psrad     m9, INTERP_SHIFT_PS
+    psrad     m10, INTERP_SHIFT_PS
+    psrad     m11, INTERP_SHIFT_PS
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+    packssdw  m8, m9
+    packssdw  m10, m11
+%else
+    paddd     m0, m7
+    paddd     m1, m7
+    paddd     m2, m7
+    paddd     m3, m7
+    paddd     m8, m7
+    paddd     m9, m7
+    paddd     m10, m7
+    paddd     m11, m7
+ %ifidn %2, pp
+    psrad     m0, INTERP_SHIFT_PP
+    psrad     m1, INTERP_SHIFT_PP
+    psrad     m2, INTERP_SHIFT_PP
+    psrad     m3, INTERP_SHIFT_PP
+    psrad     m8, INTERP_SHIFT_PP
+    psrad     m9, INTERP_SHIFT_PP
+    psrad     m10, INTERP_SHIFT_PP
+    psrad     m11, INTERP_SHIFT_PP
+%else
+    psrad     m0, INTERP_SHIFT_SP
+    psrad     m1, INTERP_SHIFT_SP
+    psrad     m2, INTERP_SHIFT_SP
+    psrad     m3, INTERP_SHIFT_SP
+    psrad     m8, INTERP_SHIFT_SP
+    psrad     m9, INTERP_SHIFT_SP
+    psrad     m10, INTERP_SHIFT_SP
+    psrad     m11, INTERP_SHIFT_SP
+%endif
+    packssdw  m0, m1
+    packssdw  m2, m3
+    packssdw  m8, m9
+    packssdw  m10, m11
+    pxor      m5, m5
+    CLIPW2    m0, m2, m5, [pw_pixel_max]
+    CLIPW2    m8, m10, m5, [pw_pixel_max]
+%endif
+
+    movu      [r2], m0
+    movu      [r2 + r3], m2
+    movu      [r2 + mmsize], xm8
+    movu      [r2 + r3 + mmsize], xm10
+    lea       r2, [r2 + 2 * r3]
+    lea       r0, [r0 + 2 * r1]
+    dec       r4d
+    jnz       .loopH
+    RET
+%endif
+%endmacro
+    FILTER_VER_CHROMA_W16_24xN_avx2 32, ss, 15
+    FILTER_VER_CHROMA_W16_24xN_avx2 32, sp, 15
+    FILTER_VER_CHROMA_W16_24xN_avx2 32, ps, 15
+    FILTER_VER_CHROMA_W16_24xN_avx2 32, pp, 15
+    FILTER_VER_CHROMA_W16_24xN_avx2 64, ss, 15
+    FILTER_VER_CHROMA_W16_24xN_avx2 64, sp, 15
+    FILTER_VER_CHROMA_W16_24xN_avx2 64, ps, 15
+    FILTER_VER_CHROMA_W16_24xN_avx2 64, pp, 15
+
+   
+;-----------------------------------------------------------------------------------------------------------------
+; void interp_4tap_vert(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------------------------------------------
+%macro FILTER_VER_CHROMA_W16_48x64_avx2 2
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_48x64, 5, 7, %2
+    add       r1d, r1d
+    add       r3d, r3d
+    sub       r0, r1
+    shl       r4d, 6
+
+%ifdef PIC
+    lea       r5, [tab_ChromaCoeffV]
+    lea       r5, [r5 + r4]
+%else
+    lea       r5, [tab_ChromaCoeffV + r4]
+%endif
+    mov       r4d, 32
+
+%ifidn %1, pp
+    vbroadcasti128  m7, [INTERP_OFFSET_PP]
+%elifidn %1, sp
+    vbroadcasti128  m7, [INTERP_OFFSET_SP]
+%elifidn %1, ps
+    vbroadcasti128  m7, [INTERP_OFFSET_PS]
+%endif
+
+.loopH:
+%assign x 0
+%rep 3
+    movu       m1, [r0 + x]
+    movu       m3, [r0 + r1 + x]
+    movu       m5, [r5 + 0 * mmsize]
+    punpcklwd  m0, m1, m3
+    pmaddwd    m0, m5
+    punpckhwd  m1, m3
+    pmaddwd    m1, m5
+
+    movu       m4, [r0 + 2 * r1 + x]
+    punpcklwd  m2, m3, m4
+    pmaddwd    m2, m5
+    punpckhwd  m3, m4
+    pmaddwd    m3, m5
+
+    lea        r6, [r0 + 2 * r1]
+    movu       m5, [r6 + r1 + x]
+    punpcklwd  m6, m4, m5
+    pmaddwd    m6, [r5 + 1 * mmsize]
+    paddd      m0, m6
+    punpckhwd  m4, m5
+    pmaddwd    m4, [r5 + 1 * mmsize]
+    paddd      m1, m4
+
+    movu       m4, [r6 + 2 * r1 + x]
+    punpcklwd  m6, m5, m4
+    pmaddwd    m6, [r5 + 1 * mmsize]
+    paddd      m2, m6
+    punpckhwd  m5, m4
+    pmaddwd    m5, [r5 + 1 * mmsize]
+    paddd      m3, m5
+
+%ifidn %1, ss
+    psrad     m0, 6
+    psrad     m1, 6
+    psrad     m2, 6
+    psrad     m3, 6
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+%elifidn %1, ps
+    paddd     m0, m7
+    paddd     m1, m7
+    paddd     m2, m7
+    paddd     m3, m7
+    psrad     m0, INTERP_SHIFT_PS
+    psrad     m1, INTERP_SHIFT_PS
+    psrad     m2, INTERP_SHIFT_PS
+    psrad     m3, INTERP_SHIFT_PS
+
+    packssdw  m0, m1
+    packssdw  m2, m3
+%else
+    paddd     m0, m7
+    paddd     m1, m7
+    paddd     m2, m7
+    paddd     m3, m7
+%ifidn %1, pp
+    psrad     m0, INTERP_SHIFT_PP
+    psrad     m1, INTERP_SHIFT_PP
+    psrad     m2, INTERP_SHIFT_PP
+    psrad     m3, INTERP_SHIFT_PP
+%else
+    psrad     m0, INTERP_SHIFT_SP
+    psrad     m1, INTERP_SHIFT_SP
+    psrad     m2, INTERP_SHIFT_SP
+    psrad     m3, INTERP_SHIFT_SP
+%endif
+    packssdw  m0, m1
+    packssdw  m2, m3
+    pxor      m5, m5
+    CLIPW2    m0, m2, m5, [pw_pixel_max]
+%endif
+
+    movu      [r2 + x], m0
+    movu      [r2 + r3 + x], m2
+%assign x x+mmsize
+%endrep
+
+    lea       r2, [r2 + 2 * r3]
+    lea       r0, [r0 + 2 * r1]
+    dec       r4d
+    jnz       .loopH
+    RET
+%endmacro
+
+    FILTER_VER_CHROMA_W16_48x64_avx2 pp, 8
+    FILTER_VER_CHROMA_W16_48x64_avx2 ps, 8
+    FILTER_VER_CHROMA_W16_48x64_avx2 ss, 7
+    FILTER_VER_CHROMA_W16_48x64_avx2 sp, 8
+
+INIT_XMM sse2
+cglobal chroma_p2s, 3, 7, 3
+    ; load width and height
+    mov         r3d, r3m
+    mov         r4d, r4m
+    add         r1, r1
+
+    ; load constant
+    mova        m2, [tab_c_n8192]
+
+.loopH:
+
+    xor         r5d, r5d
+.loopW:
+    lea         r6, [r0 + r5 * 2]
+
+    movu        m0, [r6]
+    psllw       m0, (14 - BIT_DEPTH)
+    paddw       m0, m2
+
+    movu        m1, [r6 + r1]
+    psllw       m1, (14 - BIT_DEPTH)
+    paddw       m1, m2
+
+    add         r5d, 8
+    cmp         r5d, r3d
+    lea         r6, [r2 + r5 * 2]
+    jg          .width4
+    movu        [r6 + FENC_STRIDE / 2 * 0 - 16], m0
+    movu        [r6 + FENC_STRIDE / 2 * 2 - 16], m1
+    je          .nextH
+    jmp         .loopW
+
+.width4:
+    test        r3d, 4
+    jz          .width2
+    test        r3d, 2
+    movh        [r6 + FENC_STRIDE / 2 * 0 - 16], m0
+    movh        [r6 + FENC_STRIDE / 2 * 2 - 16], m1
+    lea         r6, [r6 + 8]
+    pshufd      m0, m0, 2
+    pshufd      m1, m1, 2
+    jz          .nextH
+
+.width2:
+    movd        [r6 + FENC_STRIDE / 2 * 0 - 16], m0
+    movd        [r6 + FENC_STRIDE / 2 * 2 - 16], m1
+
+.nextH:
+    lea         r0, [r0 + r1 * 2]
+    add         r2, FENC_STRIDE / 2 * 4
+
+    sub         r4d, 2
+    jnz         .loopH
+    RET
+
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride)
+;-----------------------------------------------------------------------------
+INIT_YMM avx2
+cglobal filterPixelToShort_48x64, 3, 7, 4
+    add        r1d, r1d
+    mov        r3d, r3m
+    add        r3d, r3d
+    lea        r4, [r3 * 3]
+    lea        r5, [r1 * 3]
+
+    ; load height
+    mov        r6d, 16
+
+    ; load constant
+    mova       m3, [pw_2000]
+
+.loop:
+    movu       m0, [r0]
+    movu       m1, [r0 + 32]
+    movu       m2, [r0 + 64]
+    psllw      m0, (14 - BIT_DEPTH)
+    psllw      m1, (14 - BIT_DEPTH)
+    psllw      m2, (14 - BIT_DEPTH)
+    psubw      m0, m3
+    psubw      m1, m3
+    psubw      m2, m3
+    movu       [r2 + r3 * 0], m0
+    movu       [r2 + r3 * 0 + 32], m1
+    movu       [r2 + r3 * 0 + 64], m2
+
+    movu       m0, [r0 + r1]
+    movu       m1, [r0 + r1 + 32]
+    movu       m2, [r0 + r1 + 64]
+    psllw      m0, (14 - BIT_DEPTH)
+    psllw      m1, (14 - BIT_DEPTH)
+    psllw      m2, (14 - BIT_DEPTH)
+    psubw      m0, m3
+    psubw      m1, m3
+    psubw      m2, m3
+    movu       [r2 + r3 * 1], m0
+    movu       [r2 + r3 * 1 + 32], m1
+    movu       [r2 + r3 * 1 + 64], m2
+
+    movu       m0, [r0 + r1 * 2]
+    movu       m1, [r0 + r1 * 2 + 32]
+    movu       m2, [r0 + r1 * 2 + 64]
+    psllw      m0, (14 - BIT_DEPTH)
+    psllw      m1, (14 - BIT_DEPTH)
+    psllw      m2, (14 - BIT_DEPTH)
+    psubw      m0, m3
+    psubw      m1, m3
+    psubw      m2, m3
+    movu       [r2 + r3 * 2], m0
+    movu       [r2 + r3 * 2 + 32], m1
+    movu       [r2 + r3 * 2 + 64], m2
+
+    movu       m0, [r0 + r5]
+    movu       m1, [r0 + r5 + 32]
+    movu       m2, [r0 + r5 + 64]
+    psllw      m0, (14 - BIT_DEPTH)
+    psllw      m1, (14 - BIT_DEPTH)
+    psllw      m2, (14 - BIT_DEPTH)
+    psubw      m0, m3
+    psubw      m1, m3
+    psubw      m2, m3
+    movu       [r2 + r4], m0
+    movu       [r2 + r4 + 32], m1
+    movu       [r2 + r4 + 64], m2
+
+    lea        r0, [r0 + r1 * 4]
+    lea        r2, [r2 + r3 * 4]
+
+    dec        r6d
+    jnz        .loop
+    RET
+
+   %macro FILTER_VER_CHROMA_AVX2_8xN 2
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_vert_%1_8x%2, 4, 9, 15
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+    add             r3d, r3d
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,pp
+    vbroadcasti128  m14, [pd_32]
+%elifidn %1, sp
+    vbroadcasti128  m14, [INTERP_OFFSET_SP]
+%else
+    vbroadcasti128  m14, [INTERP_OFFSET_PS]
+%endif
+    lea             r6, [r3 * 3]
+    lea             r7, [r1 * 4]
+    mov             r8d, %2 / 16
+.loopH:
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m2, [r5]
+
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    paddd           m1, m5
+    pmaddwd         m3, [r5]
+
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m2, m6
+    pmaddwd         m4, [r5]
+
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm7, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddwd         m7, m5, [r5 + 1 * mmsize]
+    paddd           m3, m7
+    pmaddwd         m5, [r5]
+
+    movu            xm7, [r0 + r4]                  ; m7 = row 7
+    punpckhwd       xm8, xm6, xm7
+    punpcklwd       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddwd         m8, m6, [r5 + 1 * mmsize]
+    paddd           m4, m8
+    pmaddwd         m6, [r5]
+
+    lea             r0, [r0 + r1 * 4]
+    movu            xm8, [r0]                       ; m8 = row 8
+    punpckhwd       xm9, xm7, xm8
+    punpcklwd       xm7, xm8
+    vinserti128     m7, m7, xm9, 1
+    pmaddwd         m9, m7, [r5 + 1 * mmsize]
+    paddd           m5, m9
+    pmaddwd         m7, [r5]
+
+
+    movu            xm9, [r0 + r1]                  ; m9 = row 9
+    punpckhwd       xm10, xm8, xm9
+    punpcklwd       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddwd         m10, m8, [r5 + 1 * mmsize]
+    paddd           m6, m10
+    pmaddwd         m8, [r5]
+
+
+    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
+    punpckhwd       xm11, xm9, xm10
+    punpcklwd       xm9, xm10
+    vinserti128     m9, m9, xm11, 1
+    pmaddwd         m11, m9, [r5 + 1 * mmsize]
+    paddd           m7, m11
+    pmaddwd         m9, [r5]
+
+    movu            xm11, [r0 + r4]                 ; m11 = row 11
+    punpckhwd       xm12, xm10, xm11
+    punpcklwd       xm10, xm11
+    vinserti128     m10, m10, xm12, 1
+    pmaddwd         m12, m10, [r5 + 1 * mmsize]
+    paddd           m8, m12
+    pmaddwd         m10, [r5]
+
+    lea             r0, [r0 + r1 * 4]
+    movu            xm12, [r0]                      ; m12 = row 12
+    punpckhwd       xm13, xm11, xm12
+    punpcklwd       xm11, xm12
+    vinserti128     m11, m11, xm13, 1
+    pmaddwd         m13, m11, [r5 + 1 * mmsize]
+    paddd           m9, m13
+    pmaddwd         m11, [r5]
+
+%ifidn %1,ss
+    psrad           m0, 6
+    psrad           m1, 6
+    psrad           m2, 6
+    psrad           m3, 6
+    psrad           m4, 6
+    psrad           m5, 6
+%else
+    paddd           m0, m14
+    paddd           m1, m14
+    paddd           m2, m14
+    paddd           m3, m14
+    paddd           m4, m14
+    paddd           m5, m14
+%ifidn %1,pp
+    psrad           m0, 6
+    psrad           m1, 6
+    psrad           m2, 6
+    psrad           m3, 6
+    psrad           m4, 6
+    psrad           m5, 6
+%elifidn %1, sp
+    psrad           m0, INTERP_SHIFT_SP
+    psrad           m1, INTERP_SHIFT_SP
+    psrad           m2, INTERP_SHIFT_SP
+    psrad           m3, INTERP_SHIFT_SP
+    psrad           m4, INTERP_SHIFT_SP
+    psrad           m5, INTERP_SHIFT_SP
+%else
+    psrad           m0, INTERP_SHIFT_PS
+    psrad           m1, INTERP_SHIFT_PS
+    psrad           m2, INTERP_SHIFT_PS
+    psrad           m3, INTERP_SHIFT_PS
+    psrad           m4, INTERP_SHIFT_PS
+    psrad           m5, INTERP_SHIFT_PS
+%endif
+%endif
+
+    packssdw        m0, m1
+    packssdw        m2, m3
+    packssdw        m4, m5
+    vpermq          m0, m0, q3120
+    vpermq          m2, m2, q3120
+    vpermq          m4, m4, q3120
+    pxor            m5, m5
+    mova            m3, [pw_pixel_max]
+%ifidn %1,pp
+    CLIPW           m0, m5, m3
+    CLIPW           m2, m5, m3
+    CLIPW           m4, m5, m3
+%elifidn %1, sp
+    CLIPW           m0, m5, m3
+    CLIPW           m2, m5, m3
+    CLIPW           m4, m5, m3
+%endif
+
+    vextracti128    xm1, m0, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    vextracti128    xm1, m2, 1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm1
+    lea             r2, [r2 + r3 * 4]
+    vextracti128    xm1, m4, 1
+    movu            [r2], xm4
+    movu            [r2 + r3], xm1
+
+    movu            xm13, [r0 + r1]                 ; m13 = row 13
+    punpckhwd       xm0, xm12, xm13
+    punpcklwd       xm12, xm13
+    vinserti128     m12, m12, xm0, 1
+    pmaddwd         m0, m12, [r5 + 1 * mmsize]
+    paddd           m10, m0
+    pmaddwd         m12, [r5]
+
+    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
+    punpckhwd       xm1, xm13, xm0
+    punpcklwd       xm13, xm0
+    vinserti128     m13, m13, xm1, 1
+    pmaddwd         m1, m13, [r5 + 1 * mmsize]
+    paddd           m11, m1
+    pmaddwd         m13, [r5]
+
+%ifidn %1,ss
+    psrad           m6, 6
+    psrad           m7, 6
+%else
+    paddd           m6, m14
+    paddd           m7, m14
+%ifidn %1,pp
+    psrad           m6, 6
+    psrad           m7, 6
+%elifidn %1, sp
+    psrad           m6, INTERP_SHIFT_SP
+    psrad           m7, INTERP_SHIFT_SP
+%else
+    psrad           m6, INTERP_SHIFT_PS
+    psrad           m7, INTERP_SHIFT_PS
+%endif
+%endif
+
+    packssdw        m6, m7
+    vpermq          m6, m6, q3120
+%ifidn %1,pp
+    CLIPW           m6, m5, m3
+%elifidn %1, sp
+    CLIPW           m6, m5, m3
+%endif
+    vextracti128    xm7, m6, 1
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r6], xm7
+
+    movu            xm1, [r0 + r4]                  ; m1 = row 15
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m2, m0, [r5 + 1 * mmsize]
+    paddd           m12, m2
+    pmaddwd         m0, [r5]
+
+    lea             r0, [r0 + r1 * 4]
+    movu            xm2, [r0]                       ; m2 = row 16
+    punpckhwd       xm6, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm6, 1
+    pmaddwd         m6, m1, [r5 + 1 * mmsize]
+    paddd           m13, m6
+    pmaddwd         m1, [r5]
+
+    movu            xm6, [r0 + r1]                  ; m6 = row 17
+    punpckhwd       xm4, xm2, xm6
+    punpcklwd       xm2, xm6
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m0, m2
+
+    movu            xm4, [r0 + r1 * 2]              ; m4 = row 18
+    punpckhwd       xm2, xm6, xm4
+    punpcklwd       xm6, xm4
+    vinserti128     m6, m6, xm2, 1
+    pmaddwd         m6, [r5 + 1 * mmsize]
+    paddd           m1, m6
+
+%ifidn %1,ss
+    psrad           m8, 6
+    psrad           m9, 6
+    psrad           m10, 6
+    psrad           m11, 6
+    psrad           m12, 6
+    psrad           m13, 6
+    psrad           m0, 6
+    psrad           m1, 6
+%else
+    paddd           m8, m14
+    paddd           m9, m14
+    paddd           m10, m14
+    paddd           m11, m14
+    paddd           m12, m14
+    paddd           m13, m14
+    paddd           m0, m14
+    paddd           m1, m14
+%ifidn %1,pp
+    psrad           m8, 6
+    psrad           m9, 6
+    psrad           m10, 6
+    psrad           m11, 6
+    psrad           m12, 6
+    psrad           m13, 6
+    psrad           m0, 6
+    psrad           m1, 6
+%elifidn %1, sp
+    psrad           m8, INTERP_SHIFT_SP
+    psrad           m9, INTERP_SHIFT_SP
+    psrad           m10, INTERP_SHIFT_SP
+    psrad           m11, INTERP_SHIFT_SP
+    psrad           m12, INTERP_SHIFT_SP
+    psrad           m13, INTERP_SHIFT_SP
+    psrad           m0, INTERP_SHIFT_SP
+    psrad           m1, INTERP_SHIFT_SP
+%else
+    psrad           m8, INTERP_SHIFT_PS
+    psrad           m9, INTERP_SHIFT_PS
+    psrad           m10, INTERP_SHIFT_PS
+    psrad           m11, INTERP_SHIFT_PS
+    psrad           m12, INTERP_SHIFT_PS
+    psrad           m13, INTERP_SHIFT_PS
+    psrad           m0, INTERP_SHIFT_PS
+    psrad           m1, INTERP_SHIFT_PS
+%endif
+%endif
+
+    packssdw        m8, m9
+    packssdw        m10, m11
+    packssdw        m12, m13
+    packssdw        m0, m1
+    vpermq          m8, m8, q3120
+    vpermq          m10, m10, q3120
+    vpermq          m12, m12, q3120
+    vpermq          m0, m0, q3120
+%ifidn %1,pp
+    CLIPW           m8, m5, m3
+    CLIPW           m10, m5, m3
+    CLIPW           m12, m5, m3
+    CLIPW           m0, m5, m3
+%elifidn %1, sp
+    CLIPW           m8, m5, m3
+    CLIPW           m10, m5, m3
+    CLIPW           m12, m5, m3
+    CLIPW           m0, m5, m3
+%endif
+    vextracti128    xm9, m8, 1
+    vextracti128    xm11, m10, 1
+    vextracti128    xm13, m12, 1
+    vextracti128    xm1, m0, 1
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm8
+    movu            [r2 + r3], xm9
+    movu            [r2 + r3 * 2], xm10
+    movu            [r2 + r6], xm11
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm12
+    movu            [r2 + r3], xm13
+    movu            [r2 + r3 * 2], xm0
+    movu            [r2 + r6], xm1
+    lea             r2, [r2 + r3 * 4]
+    dec             r8d
+    jnz             .loopH
+    RET
+%endif
+%endmacro
+
+FILTER_VER_CHROMA_AVX2_8xN pp, 16
+FILTER_VER_CHROMA_AVX2_8xN ps, 16
+FILTER_VER_CHROMA_AVX2_8xN ss, 16
+FILTER_VER_CHROMA_AVX2_8xN sp, 16
+FILTER_VER_CHROMA_AVX2_8xN pp, 32
+FILTER_VER_CHROMA_AVX2_8xN ps, 32
+FILTER_VER_CHROMA_AVX2_8xN sp, 32
+FILTER_VER_CHROMA_AVX2_8xN ss, 32
+FILTER_VER_CHROMA_AVX2_8xN pp, 64
+FILTER_VER_CHROMA_AVX2_8xN ps, 64
+FILTER_VER_CHROMA_AVX2_8xN sp, 64
+FILTER_VER_CHROMA_AVX2_8xN ss, 64
+
+%macro PROCESS_CHROMA_AVX2_8x2 3
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m2, m2, [r5 + 1 * mmsize]
+    paddd           m0, m2
+
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m3, m3, [r5 + 1 * mmsize]
+    paddd           m1, m3
+
+%ifnidn %1,ss
+    paddd           m0, m7
+    paddd           m1, m7
+%endif
+    psrad           m0, %3
+    psrad           m1, %3
+
+    packssdw        m0, m1
+    vpermq          m0, m0, q3120
+    pxor            m4, m4
+
+%if %2
+    CLIPW           m0, m4, [pw_pixel_max]
+%endif
+    vextracti128    xm1, m0, 1
+%endmacro
+
+
+%macro FILTER_VER_CHROMA_AVX2_8x2 3
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_8x2, 4, 6, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+    add             r3d, r3d
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,pp
+    vbroadcasti128  m7, [pd_32]
+%elifidn %1, sp
+    vbroadcasti128  m7, [INTERP_OFFSET_SP]
+%else
+    vbroadcasti128  m7, [INTERP_OFFSET_PS]
+%endif
+
+    PROCESS_CHROMA_AVX2_8x2 %1, %2, %3
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    RET
+%endmacro
+
+FILTER_VER_CHROMA_AVX2_8x2 pp, 1, 6
+FILTER_VER_CHROMA_AVX2_8x2 ps, 0, INTERP_SHIFT_PS
+FILTER_VER_CHROMA_AVX2_8x2 sp, 1, INTERP_SHIFT_SP
+FILTER_VER_CHROMA_AVX2_8x2 ss, 0, 6
+
+%macro FILTER_VER_CHROMA_AVX2_4x2 3
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_4x2, 4, 6, 7
+    mov             r4d, r4m
+    add             r1d, r1d
+    add             r3d, r3d
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+
+%ifidn %1,pp
+    vbroadcasti128  m6, [pd_32]
+%elifidn %1, sp
+    vbroadcasti128  m6, [INTERP_OFFSET_SP]
+%else
+    vbroadcasti128  m6, [INTERP_OFFSET_PS]
+%endif
+
+    movq            xm0, [r0]                       ; row 0
+    movq            xm1, [r0 + r1]                  ; row 1
+    punpcklwd       xm0, xm1
+
+    movq            xm2, [r0 + r1 * 2]              ; row 2
+    punpcklwd       xm1, xm2
+    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
+    pmaddwd         m0, [r5]
+
+    movq            xm3, [r0 + r4]                  ; row 3
+    punpcklwd       xm2, xm3
+    lea             r0, [r0 + 4 * r1]
+    movq            xm4, [r0]                       ; row 4
+    punpcklwd       xm3, xm4
+    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
+    pmaddwd         m5, m2, [r5 + 1 * mmsize]
+    paddd           m0, m5
+
+%ifnidn %1, ss
+    paddd           m0, m6
+%endif
+    psrad           m0, %3
+    packssdw        m0, m0
+    pxor            m1, m1
+
+%if %2
+    CLIPW           m0, m1, [pw_pixel_max]
+%endif
+
+    vextracti128    xm2, m0, 1
+    lea             r4, [r3 * 3]
+    movq            [r2], xm0
+    movq            [r2 + r3], xm2
+    RET
+%endmacro
+
+FILTER_VER_CHROMA_AVX2_4x2 pp, 1, 6
+FILTER_VER_CHROMA_AVX2_4x2 ps, 0, INTERP_SHIFT_PS
+FILTER_VER_CHROMA_AVX2_4x2 sp, 1, INTERP_SHIFT_SP
+FILTER_VER_CHROMA_AVX2_4x2 ss, 0, 6
+
+%macro FILTER_VER_CHROMA_AVX2_4x4 3
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_4x4, 4, 6, 7
+    mov             r4d, r4m
+    add             r1d, r1d
+    add             r3d, r3d
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+
+%ifidn %1,pp
+   vbroadcasti128  m6, [pd_32]
+%elifidn %1, sp
+   vbroadcasti128  m6, [INTERP_OFFSET_SP]
+%else
+    vbroadcasti128  m6, [INTERP_OFFSET_PS]
+%endif
+    movq            xm0, [r0]                       ; row 0
+    movq            xm1, [r0 + r1]                  ; row 1
+    punpcklwd       xm0, xm1
+
+    movq            xm2, [r0 + r1 * 2]              ; row 2
+    punpcklwd       xm1, xm2
+    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
+    pmaddwd         m0, [r5]
+
+    movq            xm3, [r0 + r4]                  ; row 3
+    punpcklwd       xm2, xm3
+    lea             r0, [r0 + 4 * r1]
+    movq            xm4, [r0]                       ; row 4
+    punpcklwd       xm3, xm4
+    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
+    pmaddwd         m5, m2, [r5 + 1 * mmsize]
+    pmaddwd         m2, [r5]
+    paddd           m0, m5
+
+    movq            xm3, [r0 + r1]                  ; row 5
+    punpcklwd       xm4, xm3
+    movq            xm1, [r0 + r1 * 2]              ; row 6
+    punpcklwd       xm3, xm1
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
+    pmaddwd         m4, [r5 + 1 * mmsize]
+    paddd           m2, m4
+
+%ifnidn %1,ss
+    paddd           m0, m6
+    paddd           m2, m6
+%endif
+    psrad           m0, %3
+    psrad           m2, %3
+
+    packssdw        m0, m2
+    pxor            m1, m1
+%if %2
+    CLIPW           m0, m1, [pw_pixel_max]
+%endif
+
+    vextracti128    xm2, m0, 1
+    lea             r4, [r3 * 3]
+    movq            [r2], xm0
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r4], xm2
+    RET
+%endmacro
+
+FILTER_VER_CHROMA_AVX2_4x4 pp, 1, 6
+FILTER_VER_CHROMA_AVX2_4x4 ps, 0, INTERP_SHIFT_PS
+FILTER_VER_CHROMA_AVX2_4x4 sp, 1, INTERP_SHIFT_SP
+FILTER_VER_CHROMA_AVX2_4x4 ss, 0, 6
+
+
+%macro FILTER_VER_CHROMA_AVX2_4x8 3
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_4x8, 4, 7, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+    add             r3d, r3d
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+
+%ifidn %1,pp
+    vbroadcasti128  m7, [pd_32]
+%elifidn %1, sp
+    vbroadcasti128  m7, [INTERP_OFFSET_SP]
+%else
+    vbroadcasti128  m7, [INTERP_OFFSET_PS]
+%endif
+    lea             r6, [r3 * 3]
+
+    movq            xm0, [r0]                       ; row 0
+    movq            xm1, [r0 + r1]                  ; row 1
+    punpcklwd       xm0, xm1
+    movq            xm2, [r0 + r1 * 2]              ; row 2
+    punpcklwd       xm1, xm2
+    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
+    pmaddwd         m0, [r5]
+
+    movq            xm3, [r0 + r4]                  ; row 3
+    punpcklwd       xm2, xm3
+    lea             r0, [r0 + 4 * r1]
+    movq            xm4, [r0]                       ; row 4
+    punpcklwd       xm3, xm4
+    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
+    pmaddwd         m5, m2, [r5 + 1 * mmsize]
+    pmaddwd         m2, [r5]
+    paddd           m0, m5
+
+    movq            xm3, [r0 + r1]                  ; row 5
+    punpcklwd       xm4, xm3
+    movq            xm1, [r0 + r1 * 2]              ; row 6
+    punpcklwd       xm3, xm1
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
+    pmaddwd         m5, m4, [r5 + 1 * mmsize]
+    paddd           m2, m5
+    pmaddwd         m4, [r5]
+
+    movq            xm3, [r0 + r4]                  ; row 7
+    punpcklwd       xm1, xm3
+    lea             r0, [r0 + 4 * r1]
+    movq            xm6, [r0]                       ; row 8
+    punpcklwd       xm3, xm6
+    vinserti128     m1, m1, xm3, 1                  ; m1 = [8 7 7 6]
+    pmaddwd         m5, m1, [r5 + 1 * mmsize]
+    paddd           m4, m5
+    pmaddwd         m1, [r5]
+
+    movq            xm3, [r0 + r1]                  ; row 9
+    punpcklwd       xm6, xm3
+    movq            xm5, [r0 + 2 * r1]              ; row 10
+    punpcklwd       xm3, xm5
+    vinserti128     m6, m6, xm3, 1                  ; m6 = [A 9 9 8]
+    pmaddwd         m6, [r5 + 1 * mmsize]
+    paddd           m1, m6
+%ifnidn %1,ss
+    paddd           m0, m7
+    paddd           m2, m7
+%endif
+    psrad           m0, %3
+    psrad           m2, %3
+    packssdw        m0, m2
+    pxor            m6, m6
+    mova            m3, [pw_pixel_max]
+%if %2
+    CLIPW           m0, m6, m3
+%endif
+    vextracti128    xm2, m0, 1
+    movq            [r2], xm0
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r6], xm2
+%ifnidn %1,ss
+    paddd           m4, m7
+    paddd           m1, m7
+%endif
+    psrad           m4, %3
+    psrad           m1, %3
+    packssdw        m4, m1
+%if %2
+    CLIPW           m4, m6, m3
+%endif
+    vextracti128    xm1, m4, 1
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm4
+    movq            [r2 + r3], xm1
+    movhps          [r2 + r3 * 2], xm4
+    movhps          [r2 + r6], xm1
+    RET
+%endmacro
+
+FILTER_VER_CHROMA_AVX2_4x8 pp, 1, 6
+FILTER_VER_CHROMA_AVX2_4x8 ps, 0, INTERP_SHIFT_PS
+FILTER_VER_CHROMA_AVX2_4x8 sp, 1, INTERP_SHIFT_SP
+FILTER_VER_CHROMA_AVX2_4x8 ss, 0 , 6
+
+%macro PROCESS_LUMA_AVX2_W4_16R_4TAP 3
+    movq            xm0, [r0]                       ; row 0
+    movq            xm1, [r0 + r1]                  ; row 1
+    punpcklwd       xm0, xm1
+    movq            xm2, [r0 + r1 * 2]              ; row 2
+    punpcklwd       xm1, xm2
+    vinserti128     m0, m0, xm1, 1                  ; m0 = [2 1 1 0]
+    pmaddwd         m0, [r5]
+    movq            xm3, [r0 + r4]                  ; row 3
+    punpcklwd       xm2, xm3
+    lea             r0, [r0 + 4 * r1]
+    movq            xm4, [r0]                       ; row 4
+    punpcklwd       xm3, xm4
+    vinserti128     m2, m2, xm3, 1                  ; m2 = [4 3 3 2]
+    pmaddwd         m5, m2, [r5 + 1 * mmsize]
+    pmaddwd         m2, [r5]
+    paddd           m0, m5
+    movq            xm3, [r0 + r1]                  ; row 5
+    punpcklwd       xm4, xm3
+    movq            xm1, [r0 + r1 * 2]              ; row 6
+    punpcklwd       xm3, xm1
+    vinserti128     m4, m4, xm3, 1                  ; m4 = [6 5 5 4]
+    pmaddwd         m5, m4, [r5 + 1 * mmsize]
+    paddd           m2, m5
+    pmaddwd         m4, [r5]
+    movq            xm3, [r0 + r4]                  ; row 7
+    punpcklwd       xm1, xm3
+    lea             r0, [r0 + 4 * r1]
+    movq            xm6, [r0]                       ; row 8
+    punpcklwd       xm3, xm6
+    vinserti128     m1, m1, xm3, 1                  ; m1 = [8 7 7 6]
+    pmaddwd         m5, m1, [r5 + 1 * mmsize]
+    paddd           m4, m5
+    pmaddwd         m1, [r5]
+    movq            xm3, [r0 + r1]                  ; row 9
+    punpcklwd       xm6, xm3
+    movq            xm5, [r0 + 2 * r1]              ; row 10
+    punpcklwd       xm3, xm5
+    vinserti128     m6, m6, xm3, 1                  ; m6 = [10 9 9 8]
+    pmaddwd         m3, m6, [r5 + 1 * mmsize]
+    paddd           m1, m3
+    pmaddwd         m6, [r5]
+%ifnidn %1,ss
+    paddd           m0, m7
+    paddd           m2, m7
+%endif
+    psrad           m0, %3
+    psrad           m2, %3
+    packssdw        m0, m2
+    pxor            m3, m3
+%if %2
+    CLIPW           m0, m3, [pw_pixel_max]
+%endif
+    vextracti128    xm2, m0, 1
+    movq            [r2], xm0
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r6], xm2
+    movq            xm2, [r0 + r4]                  ;row 11
+    punpcklwd       xm5, xm2
+    lea             r0, [r0 + 4 * r1]
+    movq            xm0, [r0]                       ; row 12
+    punpcklwd       xm2, xm0
+    vinserti128     m5, m5, xm2, 1                  ; m5 = [12 11 11 10]
+    pmaddwd         m2, m5, [r5 + 1 * mmsize]
+    paddd           m6, m2
+    pmaddwd         m5, [r5]
+    movq            xm2, [r0 + r1]                  ; row 13
+    punpcklwd       xm0, xm2
+    movq            xm3, [r0 + 2 * r1]              ; row 14
+    punpcklwd       xm2, xm3
+    vinserti128     m0, m0, xm2, 1                  ; m0 = [14 13 13 12]
+    pmaddwd         m2, m0, [r5 + 1 * mmsize]
+    paddd           m5, m2
+    pmaddwd         m0, [r5]
+%ifnidn %1,ss
+    paddd           m4, m7
+    paddd           m1, m7
+%endif
+    psrad           m4, %3
+    psrad           m1, %3
+    packssdw        m4, m1
+    pxor            m2, m2
+%if %2
+    CLIPW           m4, m2, [pw_pixel_max]
+%endif
+
+    vextracti128    xm1, m4, 1
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm4
+    movq            [r2 + r3], xm1
+    movhps          [r2 + r3 * 2], xm4
+    movhps          [r2 + r6], xm1
+    movq            xm4, [r0 + r4]                  ; row 15
+    punpcklwd       xm3, xm4
+    lea             r0, [r0 + 4 * r1]
+    movq            xm1, [r0]                       ; row 16
+    punpcklwd       xm4, xm1
+    vinserti128     m3, m3, xm4, 1                  ; m3 = [16 15 15 14]
+    pmaddwd         m4, m3, [r5 + 1 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m3, [r5]
+    movq            xm4, [r0 + r1]                  ; row 17
+    punpcklwd       xm1, xm4
+    movq            xm2, [r0 + 2 * r1]              ; row 18
+    punpcklwd       xm4, xm2
+    vinserti128     m1, m1, xm4, 1                  ; m1 = [18 17 17 16]
+    pmaddwd         m1, [r5 + 1 * mmsize]
+    paddd           m3, m1
+
+%ifnidn %1,ss
+    paddd           m6, m7
+    paddd           m5, m7
+%endif
+    psrad           m6, %3
+    psrad           m5, %3
+    packssdw        m6, m5
+    pxor            m1, m1
+%if %2
+    CLIPW           m6, m1, [pw_pixel_max]
+%endif
+    vextracti128    xm5, m6, 1
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm6
+    movq            [r2 + r3], xm5
+    movhps          [r2 + r3 * 2], xm6
+    movhps          [r2 + r6], xm5
+%ifnidn %1,ss
+    paddd           m0, m7
+    paddd           m3, m7
+%endif
+    psrad           m0, %3
+    psrad           m3, %3
+    packssdw        m0, m3
+%if %2
+    CLIPW           m0, m1, [pw_pixel_max]
+%endif
+    vextracti128    xm3, m0, 1
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm0
+    movq            [r2 + r3], xm3
+    movhps          [r2 + r3 * 2], xm0
+    movhps          [r2 + r6], xm3
+%endmacro
+
+%macro FILTER_VER_CHROMA_AVX2_4xN 4
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_4x%2, 4, 8, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+    add             r3d, r3d
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+    mov             r7d, %2 / 16
+%ifidn %1,pp
+    vbroadcasti128  m7, [pd_32]
+%elifidn %1, sp
+    vbroadcasti128  m7, [INTERP_OFFSET_SP]
+%else
+    vbroadcasti128  m7, [INTERP_OFFSET_PS]
+%endif
+    lea             r6, [r3 * 3]
+.loopH:
+    PROCESS_LUMA_AVX2_W4_16R_4TAP %1, %3, %4
+    lea             r2, [r2 + r3 * 4]
+    dec             r7d
+    jnz             .loopH
+    RET
+%endmacro
+
+FILTER_VER_CHROMA_AVX2_4xN pp, 16, 1, 6
+FILTER_VER_CHROMA_AVX2_4xN ps, 16, 0, INTERP_SHIFT_PS
+FILTER_VER_CHROMA_AVX2_4xN sp, 16, 1, INTERP_SHIFT_SP
+FILTER_VER_CHROMA_AVX2_4xN ss, 16, 0, 6
+FILTER_VER_CHROMA_AVX2_4xN pp, 32, 1, 6
+FILTER_VER_CHROMA_AVX2_4xN ps, 32, 0, INTERP_SHIFT_PS
+FILTER_VER_CHROMA_AVX2_4xN sp, 32, 1, INTERP_SHIFT_SP
+FILTER_VER_CHROMA_AVX2_4xN ss, 32, 0, 6
+
+%macro FILTER_VER_CHROMA_AVX2_8x8 3
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_vert_%1_8x8, 4, 6, 12
+    mov             r4d, r4m
+    add             r1d, r1d
+    add             r3d, r3d
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+
+%ifidn %1,pp
+    vbroadcasti128  m11, [pd_32]
+%elifidn %1, sp
+    vbroadcasti128  m11, [INTERP_OFFSET_SP]
+%else
+    vbroadcasti128  m11, [INTERP_OFFSET_PS]
+%endif
+
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    pmaddwd         m2, [r5]
+    paddd           m0, m4                          ; res row0 done(0,1,2,3)
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    pmaddwd         m3, [r5]
+    paddd           m1, m5                          ;res row1 done(1, 2, 3, 4)
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    pmaddwd         m4, [r5]
+    paddd           m2, m6                          ;res row2 done(2,3,4,5)
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm7, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddwd         m7, m5, [r5 + 1 * mmsize]
+    pmaddwd         m5, [r5]
+    paddd           m3, m7                          ;res row3 done(3,4,5,6)
+    movu            xm7, [r0 + r4]                  ; m7 = row 7
+    punpckhwd       xm8, xm6, xm7
+    punpcklwd       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddwd         m8, m6, [r5 + 1 * mmsize]
+    pmaddwd         m6, [r5]
+    paddd           m4, m8                          ;res row4 done(4,5,6,7)
+    lea             r0, [r0 + r1 * 4]
+    movu            xm8, [r0]                       ; m8 = row 8
+    punpckhwd       xm9, xm7, xm8
+    punpcklwd       xm7, xm8
+    vinserti128     m7, m7, xm9, 1
+    pmaddwd         m9, m7, [r5 + 1 * mmsize]
+    pmaddwd         m7, [r5]
+    paddd           m5, m9                          ;res row5 done(5,6,7,8)
+    movu            xm9, [r0 + r1]                  ; m9 = row 9
+    punpckhwd       xm10, xm8, xm9
+    punpcklwd       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddwd         m8, [r5 + 1 * mmsize]
+    paddd           m6, m8                          ;res row6 done(6,7,8,9)
+    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
+    punpckhwd       xm8, xm9, xm10
+    punpcklwd       xm9, xm10
+    vinserti128     m9, m9, xm8, 1
+    pmaddwd         m9, [r5 + 1 * mmsize]
+    paddd           m7, m9                          ;res row7 done 7,8,9,10
+    lea             r4, [r3 * 3]
+%ifnidn %1,ss
+    paddd           m0, m11
+    paddd           m1, m11
+    paddd           m2, m11
+    paddd           m3, m11
+%endif
+    psrad           m0, %3
+    psrad           m1, %3
+    psrad           m2, %3
+    psrad           m3, %3
+    packssdw        m0, m1
+    packssdw        m2, m3
+    vpermq          m0, m0, q3120
+    vpermq          m2, m2, q3120
+    pxor            m1, m1
+    mova            m3, [pw_pixel_max]
+%if %2
+    CLIPW           m0, m1, m3
+    CLIPW           m2, m1, m3
+%endif
+    vextracti128    xm9, m0, 1
+    vextracti128    xm8, m2, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm9
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r4], xm8
+%ifnidn %1,ss
+    paddd           m4, m11
+    paddd           m5, m11
+    paddd           m6, m11
+    paddd           m7, m11
+%endif
+    psrad           m4, %3
+    psrad           m5, %3
+    psrad           m6, %3
+    psrad           m7, %3
+    packssdw        m4, m5
+    packssdw        m6, m7
+    vpermq          m4, m4, q3120
+    vpermq          m6, m6, q3120
+%if %2
+    CLIPW           m4, m1, m3
+    CLIPW           m6, m1, m3
+%endif
+    vextracti128    xm5, m4, 1
+    vextracti128    xm7, m6, 1
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm4
+    movu            [r2 + r3], xm5
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r4], xm7
+    RET
+%endif
+%endmacro
+
+FILTER_VER_CHROMA_AVX2_8x8 pp, 1, 6
+FILTER_VER_CHROMA_AVX2_8x8 ps, 0, INTERP_SHIFT_PS
+FILTER_VER_CHROMA_AVX2_8x8 sp, 1, INTERP_SHIFT_SP
+FILTER_VER_CHROMA_AVX2_8x8 ss, 0, 6
+
+%macro FILTER_VER_CHROMA_AVX2_8x6 3
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_vert_%1_8x6, 4, 6, 12
+    mov             r4d, r4m
+    add             r1d, r1d
+    add             r3d, r3d
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+
+%ifidn %1,pp
+    vbroadcasti128  m11, [pd_32]
+%elifidn %1, sp
+    vbroadcasti128  m11, [INTERP_OFFSET_SP]
+%else
+    vbroadcasti128  m11, [INTERP_OFFSET_PS]
+%endif
+
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    pmaddwd         m2, [r5]
+    paddd           m0, m4                          ; r0 done(0,1,2,3)
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    pmaddwd         m3, [r5]
+    paddd           m1, m5                          ;r1 done(1, 2, 3, 4)
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    pmaddwd         m4, [r5]
+    paddd           m2, m6                          ;r2 done(2,3,4,5)
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm7, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddwd         m7, m5, [r5 + 1 * mmsize]
+    pmaddwd         m5, [r5]
+    paddd           m3, m7                          ;r3 done(3,4,5,6)
+    movu            xm7, [r0 + r4]                  ; m7 = row 7
+    punpckhwd       xm8, xm6, xm7
+    punpcklwd       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddwd         m8, m6, [r5 + 1 * mmsize]
+    paddd           m4, m8                          ;r4 done(4,5,6,7)
+    lea             r0, [r0 + r1 * 4]
+    movu            xm8, [r0]                       ; m8 = row 8
+    punpckhwd       xm9, xm7, xm8
+    punpcklwd       xm7, xm8
+    vinserti128     m7, m7, xm9, 1
+    pmaddwd         m7, m7, [r5 + 1 * mmsize]
+    paddd           m5, m7                          ;r5 done(5,6,7,8)
+    lea             r4, [r3 * 3]
+%ifnidn %1,ss
+    paddd           m0, m11
+    paddd           m1, m11
+    paddd           m2, m11
+    paddd           m3, m11
+%endif
+    psrad           m0, %3
+    psrad           m1, %3
+    psrad           m2, %3
+    psrad           m3, %3
+    packssdw        m0, m1
+    packssdw        m2, m3
+    vpermq          m0, m0, q3120
+    vpermq          m2, m2, q3120
+    pxor            m10, m10
+    mova            m9, [pw_pixel_max]
+%if %2
+    CLIPW           m0, m10, m9
+    CLIPW           m2, m10, m9
+%endif
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m2, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r4], xm3
+%ifnidn %1,ss
+    paddd           m4, m11
+    paddd           m5, m11
+%endif
+    psrad           m4, %3
+    psrad           m5, %3
+    packssdw        m4, m5
+    vpermq          m4, m4, 11011000b
+%if %2
+    CLIPW           m4, m10, m9
+%endif
+    vextracti128    xm5, m4, 1
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm4
+    movu            [r2 + r3], xm5
+    RET
+%endif
+%endmacro
+
+FILTER_VER_CHROMA_AVX2_8x6 pp, 1, 6
+FILTER_VER_CHROMA_AVX2_8x6 ps, 0, INTERP_SHIFT_PS
+FILTER_VER_CHROMA_AVX2_8x6 sp, 1, INTERP_SHIFT_SP
+FILTER_VER_CHROMA_AVX2_8x6 ss, 0, 6
+
+%macro PROCESS_CHROMA_AVX2 3
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m2, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    paddd           m1, m5
+    pmaddwd         m3, [r5]
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m4, [r5 + 1 * mmsize]
+    paddd           m2, m4
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm4, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm4, 1
+    pmaddwd         m5, [r5 + 1 * mmsize]
+    paddd           m3, m5
+%ifnidn %1,ss
+    paddd           m0, m7
+    paddd           m1, m7
+    paddd           m2, m7
+    paddd           m3, m7
+%endif
+    psrad           m0, %3
+    psrad           m1, %3
+    psrad           m2, %3
+    psrad           m3, %3
+    packssdw        m0, m1
+    packssdw        m2, m3
+    vpermq          m0, m0, q3120
+    vpermq          m2, m2, q3120
+    pxor            m4, m4
+%if %2
+    CLIPW           m0, m4, [pw_pixel_max]
+    CLIPW           m2, m4, [pw_pixel_max]
+%endif
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m2, 1
+%endmacro
+
+
+%macro FILTER_VER_CHROMA_AVX2_8x4 3
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_8x4, 4, 6, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+    add             r3d, r3d
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer + r4]
+%endif
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,pp
+    vbroadcasti128  m7, [pd_32]
+%elifidn %1, sp
+    vbroadcasti128  m7, [INTERP_OFFSET_SP]
+%else
+    vbroadcasti128  m7, [INTERP_OFFSET_PS]
+%endif
+    PROCESS_CHROMA_AVX2 %1, %2, %3
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    movu            [r2 + r3 * 2], xm2
+    lea             r4, [r3 * 3]
+    movu            [r2 + r4], xm3
+    RET
+%endmacro
+
+FILTER_VER_CHROMA_AVX2_8x4 pp, 1, 6
+FILTER_VER_CHROMA_AVX2_8x4 ps, 0, INTERP_SHIFT_PS
+FILTER_VER_CHROMA_AVX2_8x4 sp, 1, INTERP_SHIFT_SP
+FILTER_VER_CHROMA_AVX2_8x4 ss, 0, 6
+
+%macro FILTER_VER_CHROMA_AVX2_8x12 3
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_4tap_vert_%1_8x12, 4, 7, 15
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+    add             r3d, r3d
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,pp
+    vbroadcasti128  m14, [pd_32]
+%elifidn %1, sp
+    vbroadcasti128  m14, [INTERP_OFFSET_SP]
+%else
+    vbroadcasti128  m14, [INTERP_OFFSET_PS]
+%endif
+    lea             r6, [r3 * 3]
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    paddd           m0, m4
+    pmaddwd         m2, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    paddd           m1, m5
+    pmaddwd         m3, [r5]
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m2, m6
+    pmaddwd         m4, [r5]
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm7, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddwd         m7, m5, [r5 + 1 * mmsize]
+    paddd           m3, m7
+    pmaddwd         m5, [r5]
+    movu            xm7, [r0 + r4]                  ; m7 = row 7
+    punpckhwd       xm8, xm6, xm7
+    punpcklwd       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddwd         m8, m6, [r5 + 1 * mmsize]
+    paddd           m4, m8
+    pmaddwd         m6, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm8, [r0]                       ; m8 = row 8
+    punpckhwd       xm9, xm7, xm8
+    punpcklwd       xm7, xm8
+    vinserti128     m7, m7, xm9, 1
+    pmaddwd         m9, m7, [r5 + 1 * mmsize]
+    paddd           m5, m9
+    pmaddwd         m7, [r5]
+    movu            xm9, [r0 + r1]                  ; m9 = row 9
+    punpckhwd       xm10, xm8, xm9
+    punpcklwd       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddwd         m10, m8, [r5 + 1 * mmsize]
+    paddd           m6, m10
+    pmaddwd         m8, [r5]
+    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
+    punpckhwd       xm11, xm9, xm10
+    punpcklwd       xm9, xm10
+    vinserti128     m9, m9, xm11, 1
+    pmaddwd         m11, m9, [r5 + 1 * mmsize]
+    paddd           m7, m11
+    pmaddwd         m9, [r5]
+    movu            xm11, [r0 + r4]                 ; m11 = row 11
+    punpckhwd       xm12, xm10, xm11
+    punpcklwd       xm10, xm11
+    vinserti128     m10, m10, xm12, 1
+    pmaddwd         m12, m10, [r5 + 1 * mmsize]
+    paddd           m8, m12
+    pmaddwd         m10, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm12, [r0]                      ; m12 = row 12
+    punpckhwd       xm13, xm11, xm12
+    punpcklwd       xm11, xm12
+    vinserti128     m11, m11, xm13, 1
+    pmaddwd         m13, m11, [r5 + 1 * mmsize]
+    paddd           m9, m13
+    pmaddwd         m11, [r5]
+%ifnidn %1,ss
+    paddd           m0, m14
+    paddd           m1, m14
+    paddd           m2, m14
+    paddd           m3, m14
+    paddd           m4, m14
+    paddd           m5, m14
+%endif
+    psrad           m0, %3
+    psrad           m1, %3
+    psrad           m2, %3
+    psrad           m3, %3
+    psrad           m4, %3
+    psrad           m5, %3
+    packssdw        m0, m1
+    packssdw        m2, m3
+    packssdw        m4, m5
+    vpermq          m0, m0, q3120
+    vpermq          m2, m2, q3120
+    vpermq          m4, m4, q3120
+    pxor            m5, m5
+    mova            m3, [pw_pixel_max]
+%if %2
+    CLIPW           m0, m5, m3
+    CLIPW           m2, m5, m3
+    CLIPW           m4, m5, m3
+%endif
+    vextracti128    xm1, m0, 1
+    movu            [r2], xm0
+    movu            [r2 + r3], xm1
+    vextracti128    xm1, m2, 1
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r6], xm1
+    lea             r2, [r2 + r3 * 4]
+    vextracti128    xm1, m4, 1
+    movu            [r2], xm4
+    movu            [r2 + r3], xm1
+    movu            xm13, [r0 + r1]                 ; m13 = row 13
+    punpckhwd       xm0, xm12, xm13
+    punpcklwd       xm12, xm13
+    vinserti128     m12, m12, xm0, 1
+    pmaddwd         m12, m12, [r5 + 1 * mmsize]
+    paddd           m10, m12
+    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
+    punpckhwd       xm1, xm13, xm0
+    punpcklwd       xm13, xm0
+    vinserti128     m13, m13, xm1, 1
+    pmaddwd         m13, m13, [r5 + 1 * mmsize]
+    paddd           m11, m13
+%ifnidn %1,ss
+    paddd           m6, m14
+    paddd           m7, m14
+    paddd           m8, m14
+    paddd           m9, m14
+    paddd           m10, m14
+    paddd           m11, m14
+%endif
+    psrad           m6, %3
+    psrad           m7, %3
+    psrad           m8, %3
+    psrad           m9, %3
+    psrad           m10, %3
+    psrad           m11, %3
+    packssdw        m6, m7
+    packssdw        m8, m9
+    packssdw        m10, m11
+    vpermq          m6, m6, q3120
+    vpermq          m8, m8, q3120
+    vpermq          m10, m10, q3120
+%if %2
+    CLIPW           m6, m5, m3
+    CLIPW           m8, m5, m3
+    CLIPW           m10, m5, m3
+%endif
+    vextracti128    xm7, m6, 1
+    vextracti128    xm9, m8, 1
+    vextracti128    xm11, m10, 1
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r6], xm7
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm8
+    movu            [r2 + r3], xm9
+    movu            [r2 + r3 * 2], xm10
+    movu            [r2 + r6], xm11
+    RET
+%endif
+%endmacro
+
+FILTER_VER_CHROMA_AVX2_8x12 pp, 1, 6
+FILTER_VER_CHROMA_AVX2_8x12 ps, 0, INTERP_SHIFT_PS
+FILTER_VER_CHROMA_AVX2_8x12 sp, 1, INTERP_SHIFT_SP
+FILTER_VER_CHROMA_AVX2_8x12 ss, 0, 6
\ No newline at end of file

From 09265145041bee53ade04a47f23b1a7c232007fb Mon Sep 17 00:00:00 2001
From: Ashok Kumar Mishra <ashok@multicorewareinc.com>
Date: Wed, 21 Feb 2018 14:25:56 +0530
Subject: [PATCH 51/51] release: Release notes for v2.7

---
 doc/reST/releasenotes.rst | 26 ++++++++++++++++++++++++++
 1 file changed, 26 insertions(+)

diff --git a/doc/reST/releasenotes.rst b/doc/reST/releasenotes.rst
index 89eca798ef..19de52c38b 100644
--- a/doc/reST/releasenotes.rst
+++ b/doc/reST/releasenotes.rst
@@ -2,6 +2,32 @@
 Release Notes
 *************
 
+Version 2.7
+===========
+
+Release date - 21st Feb, 2018.
+
+New features
+------------
+1. :option:`--gop-lookahead` can be used to extend the gop boundary(set by `--keyint`). The GOP will be extended, if a scene-cut frame is found within this many number of frames. 
+2. Support for RADL pictures added in x265.
+   :option:`--radl` can be used to decide number of RADL pictures preceding the IDR picture.
+
+Encoder enhancements
+--------------------
+1. Moved from YASM to NASM assembler. Supports NASM assembler version 2.13 and greater.
+2. Enable analysis save and load in a single run. Introduces two new cli options `--analysis-save <filename>` and `--analysis-load <filename>`.
+3. Comply to HDR10+ LLC specification.
+4. Reduced x265 build time by more than 50% by re-factoring ipfilter.asm.  
+
+Bug fixes
+---------
+1. Fixed inconsistent output issue in deblock filter and --const-vbv.
+2. Fixed Mac OS build warnings.
+3. Fixed inconsistency in pass-2 when weightp and cutree are enabled.
+4. Fixed deadlock issue due to dropping of BREF frames, while forcing slice types through qp file.
+
+
 Version 2.6
 ===========