Skip to content

Commit

Permalink
Vers 2.029
Browse files Browse the repository at this point in the history 
Updated with Spout 2.007.001 files
  • Loading branch information
@leadedge
leadedge committed Jul 25, 2023
1 parent 28ab85b commit e0d47fc
Show file tree
Hide file tree
Showing 24 changed files with 752 additions and 293 deletions.
20 changes: 10 additions & 10 deletions SpoutDX/source/SpoutCommon.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -34,8 +34,12 @@
LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
03.07.23 - Remove _MSC_VER condition from SPOUT_DLLEXP define
(#PR93 Fix MinGW error (beta branch)
*/

#pragma once

#ifndef __SpoutCommon__
Expand All @@ -46,17 +50,13 @@
// SPOUT_BUILD_DLL in the preprocessor defines.
// Properties > C++ > Preprocessor > Preprocessor Definitions
//
#if defined(_MSC_VER)
#if defined(SPOUT_BUILD_DLL)
#define SPOUT_DLLEXP __declspec(dllexport)
#elif defined(SPOUT_IMPORT_DLL)
#define SPOUT_DLLEXP __declspec(dllimport)
#else
#define SPOUT_DLLEXP
#endif
#else // _MSC_VER
#if defined(SPOUT_BUILD_DLL)
#define SPOUT_DLLEXP __declspec(dllexport)
#elif defined(SPOUT_IMPORT_DLL)
#define SPOUT_DLLEXP __declspec(dllimport)
#else
#define SPOUT_DLLEXP
#endif // _MSC_VERR
#endif

// Common utility functions namespace
#include "SpoutUtils.h"
Expand Down
285 changes: 266 additions & 19 deletions SpoutDX/source/SpoutCopy.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -61,6 +61,14 @@
11.12.22 - test for null args in conversion functions
22.12.22 - Compiler compatibility check
Change all {} initializations to "={}"
02-04-23 - Corrected source pointer increment in rgba2rgba when not inverted
Version 2.007.11
17.04.23 - Add rgba_to_rgb_sse and rgba_to_bgr_sse
18.04.23 - Rename to rgba_to_rgb_sse3 and rgba_to_bgr_sse3
Add bSwapRG to rgba_to_rgb_sse3
Remove rgba_to_bgr_sse3
Add experimental rgb_to_bgra_sse3
Version 2.007.012
*/

Expand Down Expand Up @@ -298,8 +306,8 @@ void spoutCopy::rgba2rgba(const void* rgba_source, void* rgba_dest,
// first
// Casting first avoids warning C26451: Arithmetic overflow with VS2022 code review
// https://docs.microsoft.com/en-us/visualstudio/code-quality/c26451
unsigned long YxW = (unsigned long)(y * width);
const unsigned long YxSP4 = (unsigned long)sourcePitch / 4;
unsigned long YxW = (unsigned long)(y * width);
const unsigned long YxSP4 = (unsigned long)(y * sourcePitch / 4);
const unsigned long InvYxSP4 = (unsigned long)((height - 1 - y) * sourcePitch / 4);

// Increment to current line
Expand Down Expand Up @@ -524,7 +532,8 @@ void spoutCopy::bgra2rgba(const void *bgra_source, void *rgba_dest, unsigned int

//---------------------------------------------------------
// Function: rgba2rgb
// Copy RGBA to RGB allowing for source line pitch
// Copy RGBA to RGB or BGR allowing for source line pitch using the fastest method
//
void spoutCopy::rgba2rgb(const void* rgba_source, void* rgb_dest,
unsigned int width, unsigned int height,
unsigned int rgba_pitch, bool bInvert, bool bMirror, bool bSwapRB) const
Expand All @@ -535,6 +544,33 @@ void spoutCopy::rgba2rgb(const void* rgba_source, void* rgb_dest,
if (!rgb || !rgba)
return;

//
// SSE3 copy
// No mirror option, image size 16 bit byte aligned, SSE3 intrinsics support
//
// Timing tests show more than twice as fast
// (Intel(R) Core(TM) i7-3770K CPU @ 3.50GHz)
//
// SSE
// 1280x720 1.6 msec
// 1920x1080 3.7 msec
// 3840x2160 14.0 msec
// Byte copy
// 1280x720 4.0 msec
// 1920x1080 9.3 msec
// 3840x2160 35.9 msec
//
unsigned int pitch = rgba_pitch;
if(pitch == 0) pitch = width*4;
if (!bMirror && width >= 320 && (width % 16) == 0 && m_bSSE3) {
rgba_to_rgb_sse3(rgba_source, rgb_dest, width, height, pitch, bInvert, bSwapRB);
return;
}

//
// Byte pointer copy
//

// RGB dest does not have padding
uint64_t rgbsize = (uint64_t)width * (uint64_t)height * 3;
uint64_t rgbpitch = (uint64_t)width * 3;
Expand Down Expand Up @@ -580,6 +616,7 @@ void spoutCopy::rgba2rgb(const void* rgba_source, void* rgb_dest,

} // end rgba2rgb


//---------------------------------------------------------
// Function: rgb2rgba
//
Expand Down Expand Up @@ -814,35 +851,37 @@ void spoutCopy::rgb2bgra(const void *rgb_source, void *bgra_dest,
} // end rgb2bgra


// Experimental
//
// Testing shows gains of 1 fps
// TODO : exact frame timing
// =====================================================================================
//
// Experimental
// https://exchangetuts.com/fast-vectorized-conversion-from-rgb-to-bgra-1640844423877396
//
// in and out must be 16-byte aligned
// See also SIMD library
// https://github.com/ermig1979/Simd
//

//---------------------------------------------------------
// Function: rgb_to_bgrx_sse
// Experimental pending testing
void spoutCopy::rgb_to_bgrx_sse(unsigned int w, const void* inpix, void* outpix) const
// Single line function
void spoutCopy::rgb_to_bgrx_sse(unsigned int npixels, const void* rgb_source, void* bgrx_dest) const
{
const __m128i* in_vec = static_cast<const __m128i*>(inpix);
__m128i* out_vec = static_cast<__m128i*>(outpix);
const __m128i* in_vec = static_cast<const __m128i*>(rgb_source);
__m128i* out_vec = static_cast<__m128i*>(bgrx_dest);

if (!in_vec || !out_vec)
return;

w /= 16;
npixels /= 16;

while (w-- > 0) {
while (npixels-- > 0) {

// 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
// in_vec[0] Ra Ga Ba Rb Gb Bb Rc Gc Bc Rd Gd Bd Re Ge Be Rf
// in_vec[1] Gf Bf Rg Gg Bg Rh Gh Bh Ri Gi Bi Rj Gj Bj Rk Gk
// in_vec[2] Bk Rl Gl Bl Rm Gm Bm Rn Gn Bn Ro Go Bo Rp Gp Bp
//

/* 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
* in_vec[0] Ra Ga Ba Rb Gb Bb Rc Gc Bc Rd Gd Bd Re Ge Be Rf
* in_vec[1] Gf Bf Rg Gg Bg Rh Gh Bh Ri Gi Bi Rj Gj Bj Rk Gk
* in_vec[2] Bk Rl Gl Bl Rm Gm Bm Rn Gn Bn Ro Go Bo Rp Gp Bp
*/
__m128i in1={};
__m128i in2={};
__m128i in3={};
Expand Down Expand Up @@ -894,6 +933,214 @@ void spoutCopy::rgb_to_bgrx_sse(unsigned int w, const void* inpix, void* outpix)

} // end rgb_to_bgrx_sse


//---------------------------------------------------------
// Function: rgb_to_bgra_sse
// Experimental pending testing
// Full image height
void spoutCopy::rgb_to_bgra_sse3 (
void* rgb_source,
void* rgba_dest,
unsigned int width,
unsigned int height) const
{
if ((width % 16) != 0)
return;

auto rgb = static_cast<const unsigned char*>(rgb_source); // rgb/bgr
auto rgba = static_cast<unsigned char*>(rgba_dest); // rgba/bgra
if (!rgb || !rgba)
return;

for (unsigned int y = 0; y < height-2; y++) {
rgb_to_bgrx_sse(width*3, rgb, rgba);
rgb += width*3;
rgba += width*4;
}

} // end rgb_to_bgra_sse3

//
// =====================================================================================


//---------------------------------------------------------
// Function: rgba_to_rgb_sse3
//
void spoutCopy::rgba_to_rgb_sse3(const void* rgba_source, void* rgb_dest,
unsigned int width, unsigned int height, unsigned int rgba_pitch,
bool bInvert, bool bSwapRB) const
{
const __m128i* in_vec = static_cast<const __m128i*>(rgba_source); // rgba
__m128i* out_vec = static_cast<__m128i*>(rgb_dest); // rgb

if (!out_vec || !in_vec)
return;

// RGB dest does not have padding
unsigned int rgbsize = width*height*3;
unsigned int rgbpitch = width*3;

// Dest and source must be the same dimensions otherwise
unsigned int rgba_padding = rgba_pitch-width*4; // byte line padding

// Flip image option, move to the beginning of the last rgb line
if (bInvert) {
out_vec += rgbsize/16; // end of rgb buffer
out_vec -= rgbpitch/16; // beginning of the last rgb line
}

unsigned int w = width/16;
for (unsigned int y = 0; y < height; y++) {

while (w-- > 0) {

__m128i in0={};
__m128i in1={};
__m128i in2={};
__m128i in3={};

__m128i out0={};
__m128i out1={};

in0 = in_vec[0]; // First 128 bits RGBA
in1 = in_vec[1]; // Second 128 bits RGBA
in2 = in_vec[2]; // Third 128 bits RGBA
in3 = in_vec[3]; // Fourth 128 bits RGBA

if (!bSwapRB) {

// RGBA > RGB

//
// RGBA in 4x16 = 64 bytes (4 pixels at 4 bytes each = 16 bytes)
// 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
// in_vec[0] Ra Ga Ba Xa Rb Gb Bb Xb Rc Gc Bc Xc Rd Gd Bd Xd
// in_vec[1] Re Ge Be Xe Rf Gf Bf Xf Rg Gg Bg Xg Rh Gh Bh Xh
// in_vec[2] Ri Gi Bi Xi Rj Gj Bj Xj Rk Gk Bk Xk Rl Gl Bl Xl
// in_vec[3] Rm Gm Bm Xm Rn Gn Bn Xn Ro Go Bo Xo Rp Gp Bp Xp

//
// RGB out 3x16 = 48 bytes (4 pixels at 3 bytes each = 12 bytes)
//
// out_vec[0] Rf Be Ge Re Bd Gd Rd Bc Gc Rc Bb Gb Rb Ba Ga Ra
// 4 2 1 0 14 13 12 10 9 8 6 5 4 2 1 0
// in_vec[1] | in_vec[0] |
// out_vec[1] Gk Rk Bj Gj Rj Bi Gi Ri Bh Gh Rh Bg Gg Rg Bf Gf
// 9 8 6 5 4 2 1 0 14 13 12 10 9 8 6 5
// in_vec[2] | in_vec[1]
// out_vec[2] Bp Gp Rp Bo Go Ro Bn Gn Rn Bm Gm Rm Bl Gl Rl Bk
// 14 13 12 10 9 8 6 5 4 2 1 0 14 13 12 10
// in_vec[3] | in_vec[2]

// First 16 RGB bytes
// out_vec[0] Rf Be Ge Re Bd Gd Rd Bc Gc Rc Bb Gb Rb Ba Ga Ra
// 4 2 1 0 14 13 12 10 9 8 6 5 4 2 1 0
// in_vec[1] | in_vec[0] |
out0 = _mm_shuffle_epi8(in0,
_mm_set_epi8('\xff', '\xff', '\xff', '\xff', 14, 13, 12, 10, 9, 8, 6, 5, 4, 2, 1, 0));
out1 = _mm_shuffle_epi8(in1,
_mm_set_epi8(4, 2, 1, 0, '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff'));
out_vec[0] = _mm_or_si128(out0, out1);

// Second 16 RGB bytes
// out_vec[1] Gk Rk Bj Gj Rj Bi Gi Ri Bh Gh Rh Bg Gg Rg Bf Gf
// 9 8 6 5 4 2 1 0 14 13 12 10 9 8 6 5
// in_vec[2] | in_vec[1]
out0 = _mm_shuffle_epi8(in1,
_mm_set_epi8('\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', 14, 13, 12, 10, 9, 8, 6, 5));
out1 = _mm_shuffle_epi8(in2,
_mm_set_epi8(9, 8, 6, 5, 4, 2, 1, 0, '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff'));
out_vec[1] = _mm_or_si128(out0, out1);

// Third 16 RGB bytes
// out_vec[2] Bp Gp Rp Bo Go Ro Bn Gn Rn Bm Gm Rm Bl Gl Rl Bk
// 14 13 12 10 9 8 6 5 4 2 1 0 14 13 12 10
// in_vec[3] | in_vec[2]
out0 = _mm_shuffle_epi8(in2,
_mm_set_epi8('\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', 14, 13, 12, 10));
out1 = _mm_shuffle_epi8(in3,
_mm_set_epi8(14, 13, 12, 10, 9, 8, 6, 5, 4, 2, 1, 0, '\xff', '\xff', '\xff', '\xff'));
out_vec[2] = _mm_or_si128(out0, out1);
} // end RGBA >RGB
else {

// RGBA > BGR

//
// RGBA in 4x16 = 64 bytes (4 pixels at 4 bytes each = 16 bytes)
// 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
// in_vec[0] Ra Ga Ba Xa Rb Gb Bb Xb Rc Gc Bc Xc Rd Gd Bd Xd
// in_vec[1] Re Ge Be Xe Rf Gf Bf Xf Rg Gg Bg Xg Rh Gh Bh Xh
// in_vec[2] Ri Gi Bi Xi Rj Gj Bj Xj Rk Gk Bk Xk Rl Gl Bl Xl
// in_vec[3] Rm Gm Bm Xm Rn Gn Bn Xn Ro Go Bo Xo Rp Gp Bp Xp

//
// BGR out 3x16 = 48 bytes (4 pixels at 3 bytes each = 12 bytes)
//
// out_vec[0] Bf Re Ge Be Rd Gd Bd Rc Gc Bc Rb Gb Bb Ra Ga Ba
// 6 0 1 2 12 13 14 8 9 10 4 5 6 0 1 2
// in_vec[1] | in_vec[0] |
// out_vec[1] Gk Bk Rj Gj Bj Ri Gi Bi Rh Gh Bh Rg Gg Bg Rf Gf
// 9 10 4 5 6 0 1 2 12 13 14 8 9 10 4 5
// in_vec[2] | in_vec[1]
// out_vec[2] Rp Gp Bp Ro Go Bo Rn Gn Bn Rm Gm Bm Rl Gl Bl Rk
// 12 13 14 8 9 10 4 5 6 0 1 2 12 13 14 8
// in_vec[3] | in_vec[2]

// First 16 BGR bytes
// out_vec[0] Rf Be Ge Re Bd Gd Rd Bc Gc Rc Bb Gb Rb Ba Ga Ra
// 6 0 1 2 12 13 14 8 9 10 4 5 6 0 1 2
// in_vec[1] | in_vec[0] |
out0 = _mm_shuffle_epi8(in0,
_mm_set_epi8('\xff', '\xff', '\xff', '\xff', 12, 13, 14, 8, 9, 10, 4, 5, 6, 0, 1, 2));
out1 = _mm_shuffle_epi8(in1,
_mm_set_epi8(6, 0, 1, 2, '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff'));
out_vec[0] = _mm_or_si128(out0, out1);

// Second 16 BGR bytes
// out_vec[1] Gk Rk Bj Gj Rj Bi Gi Ri Bh Gh Rh Bg Gg Rg Bf Gf
// 9 10 4 5 6 0 1 2 12 13 14 8 9 10 4 5
// in_vec[2] | in_vec[1]
out0 = _mm_shuffle_epi8(in1,
_mm_set_epi8('\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', 12, 13, 14, 8, 9, 10, 4, 5));
out1 = _mm_shuffle_epi8(in2,
_mm_set_epi8(9, 10, 4, 5, 6, 0, 1, 2, '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff'));
out_vec[1] = _mm_or_si128(out0, out1);

// Third 16 BGR bytes
// out_vec[2] Bp Gp Rp Bo Go Ro Bn Gn Rn Bm Gm Rm Bl Gl Rl Bk
// 12 13 14 8 9 10 4 5 6 0 1 2 12 13 14 8
// in_vec[3] | in_vec[2]
out0 = _mm_shuffle_epi8(in2,
_mm_set_epi8('\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', '\xff', 12, 13, 14, 8));
out1 = _mm_shuffle_epi8(in3,
_mm_set_epi8(12, 13, 14, 8, 9, 10, 4, 5, 6, 0, 1, 2, '\xff', '\xff', '\xff', '\xff'));
out_vec[2] = _mm_or_si128(out0, out1);

} // end RGBA > BGR

in_vec += 4; // RGBA 4x16
out_vec += 3; // RGB 3x16

} // done the line

// Reset width
w = width/16;

// Allow for padding at the end of the rgba source line
in_vec += rgba_padding/16;

// move up a line for invert
if (bInvert) {
out_vec -= w*3*2;
}

}

} // end rgba_to_rgb_sse


//---------------------------------------------------------
// Function: bgr2bgra
//
Expand Down Expand Up @@ -1040,7 +1287,7 @@ void spoutCopy::rgba2rgbResample(const void* source, void* dest,

if (bMirror) {
if (bInvert)
pixel = (destHeight - i - 1)*destWidth * 3 + (destWidth - j - 1) * 3; // flip vertically
pixel = (destHeight - i - 1)*destWidth * 3 + (destWidth - j - 1) * 3; // flip horizontally
else
pixel = i * destWidth * 3 + (destWidth - j - 1) * 3;
}
Expand Down
Loading

0 comments on commit e0d47fc

Please sign in to comment.