Add support for 16x16 (single face) tile reduce and pack untilize

tenstorrent · Feb 29, 2024 · 60345bc · 60345bc
1 parent de157f0
commit 60345bc
Show file tree

Hide file tree

Showing 2 changed files with 12 additions and 6 deletions.
diff --git a/llk_lib/llk_math_reduce.h b/llk_lib/llk_math_reduce.h
@@ -172,7 +172,7 @@ inline void _llk_math_reduce_(const uint dst_index, bool narrow_tile = false, co
                     TTI_GAPOOL(p_setrwc::CLR_NONE, p_gpool::DIM_16X16, ADDR_MOD_0, p_gpool::INDEX_DIS, 0);
                 }
             }
-            if (!narrow_tile) {
+            if ((!narrow_tile) && (num_faces>1)) {
                 TTI_SETRWC(p_setrwc::CLR_NONE, p_setrwc::CR_D, 8, 0, 0, p_setrwc::SET_D);
                 TTI_SETRWC(p_setrwc::CLR_AB, p_setrwc::CR_D, 8, 0, 0, p_setrwc::SET_D);
 

diff --git a/llk_lib/llk_pack_untilize.h b/llk_lib/llk_pack_untilize.h
@@ -40,12 +40,18 @@ inline void _llk_pack_untilize_mop_config_(const std::uint32_t face_r_dim = FACE
 
     constexpr uint MOP_OUTER_LOOP = block_ct_dim;
 
-    // Inc ch0_y+=1 (addr_mod_0 will increment by 15)
-    ckernel::ckernel_template tmp(MOP_OUTER_LOOP, MOP_INNER_LOOP, TT_OP_INCADCXY(p_setadc::PAC, 0, 0, 1, 0));
-    tmp.set_start_op(TT_OP_PACR(ADDR_MOD_0, ZERO_OUTPUT_FLAG, PACK_SEL(PACKCNT), 0, MEGAROW, 0, 0));
-    tmp.set_end_ops(TT_OP_PACR(ADDR_MOD_1, ZERO_OUTPUT_FLAG, PACK_SEL(PACKCNT), 0, MEGAROW, 0, 0),
+    if (num_faces>1) {
+        // Inc ch0_y+=1 (addr_mod_0 will increment by 15)
+        ckernel::ckernel_template tmp(MOP_OUTER_LOOP, MOP_INNER_LOOP, TT_OP_INCADCXY(p_setadc::PAC, 0, 0, 1, 0));
+        tmp.set_start_op(TT_OP_PACR(ADDR_MOD_0, ZERO_OUTPUT_FLAG, PACK_SEL(PACKCNT), 0, MEGAROW, 0, 0));
+        tmp.set_end_ops(TT_OP_PACR(ADDR_MOD_1, ZERO_OUTPUT_FLAG, PACK_SEL(PACKCNT), 0, MEGAROW, 0, 0),
                     TT_OP_INCADCZW(p_setadc::PAC, 0, 0, 1, 0)); // w cnt points to the next tile
-    tmp.program(instrn_buffer);
+        tmp.program(instrn_buffer);
+    } else {
+        ckernel::ckernel_template tmp(MOP_OUTER_LOOP, MOP_INNER_LOOP, TT_OP_PACR(ADDR_MOD_1, ZERO_OUTPUT_FLAG, PACK_SEL(PACKCNT), 0, MEGAROW, 0, 0));
+        tmp.set_end_op(TT_OP_INCADCZW(p_setadc::PAC, 0, 0, 1, 0)); // w cnt points to the next tile
+        tmp.program(instrn_buffer);
+    }
 }
 
 template <std::uint32_t block_ct_dim>