From 01c0bb57696d552735dcc8e91eea1d75962ce172 Mon Sep 17 00:00:00 2001 From: Joao Carreira Date: Sat, 28 Feb 2015 03:42:17 +0000 Subject: [PATCH] Add support for SSE instructions (optional) --- CMakeLists.txt | 2 + config.h.in | 7 +- lib/PlaYUVerFramePixelFormats.cpp | 163 ++++++++++++++++++++++++++++-- 3 files changed, 160 insertions(+), 12 deletions(-) diff --git a/CMakeLists.txt b/CMakeLists.txt index 6056350..0c4d315 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -40,6 +40,7 @@ OPTION( PLAYUVER_LONGVERSIONNAME "Use long version name in deb" OFF ) OPTION( PLAYUVER_WERROR "Warnings as errors" OFF ) OPTION( PLAYUVER_THREADED_MODULES "Use threads for modules" OFF ) OPTION( USE_QT4 "Build with QT4" OFF ) +OPTION( USE_SSE "Build with SSE support" ON ) ##################################################################################### # Qt library @@ -99,6 +100,7 @@ IF( USE_FFMPEG ) ENDIF() SET_PACKAGE_PROPERTIES(FFmpeg PROPERTIES URL "http://ffmpeg.org/" DESCRIPTION "Libav library support in PlaYUVerStream" TYPE OPTIONAL) +ADD_FEATURE_INFO(SSE USE_SSE "SSE instructions support" ) ADD_FEATURE_INFO(Fervor USE_FERVOR "Fervor lib includes check for updates feature.") ADD_FEATURE_INFO(WErrors PLAYUVER_WERROR "Warnings as errors" ) diff --git a/config.h.in b/config.h.in index 62b33be..647c3d6 100644 --- a/config.h.in +++ b/config.h.in @@ -16,12 +16,12 @@ * with this program; if not, write to the Free Software Foundation, Inc., * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */ - + /** * \file config.h.in * \brief definition of the building options */ - + #ifndef __CONFIG_PLAYUVER_H__ #define __CONFIG_PLAYUVER_H__ @@ -35,6 +35,9 @@ #define DEPENDENCIES_STRING "@DEPENDENCIES_STRING@" +/* CPU features */ +#cmakedefine USE_SSE + /* QT definition */ #cmakedefine USE_QT4 diff --git a/lib/PlaYUVerFramePixelFormats.cpp b/lib/PlaYUVerFramePixelFormats.cpp index df05e68..c0aef07 100644 --- a/lib/PlaYUVerFramePixelFormats.cpp +++ b/lib/PlaYUVerFramePixelFormats.cpp @@ -25,6 +25,11 @@ #include "config.h" #include #include "LibMemory.h" +#ifdef USE_SSE +#include +#include +#include +#endif #ifdef USE_FFMPEG #include "LibAvContextHandle.h" #endif @@ -68,23 +73,24 @@ Void bufferFromFrameYUVProgressive( Pel*** in, Pel *out, UInt width, UInt height Void fillRGBbufferYUV( Pel*** in, UChar* out, UInt width, UInt height, UInt ratioChromaHor, UInt ratioChromaVer ) { Pel* pY = in[LUMA][0]; - Pel* pU = in[CHROMA_U][0]; - Pel* pV = in[CHROMA_V][0]; + Pel* pLineU = in[CHROMA_U][0]; + Pel* pLineV = in[CHROMA_V][0]; + UInt uiChromaStride = width >> ratioChromaHor; + Pel* pU; + Pel* pV; Int iY, iU, iV, iR, iG, iB; UInt* buff = ( UInt* )out; - pU -= width >> ratioChromaHor; - pV -= width >> ratioChromaHor; for( UInt y = 0; y < height >> ratioChromaVer; y++ ) { - pU += width >> ratioChromaHor; - pV += width >> ratioChromaHor; for( Int i = 0; i < 1 << ratioChromaVer; i++ ) { + pU = pLineU; + pV = pLineV; for( UInt x = 0; x < width >> ratioChromaHor; x++ ) { - iU = *pU++ - 128; - iV = *pV++ - 128; + iU = *pU++; + iV = *pV++; for( Int j = 0; j < 1 << ratioChromaHor; j++ ) { iY = *pY++; @@ -93,9 +99,9 @@ Void fillRGBbufferYUV( Pel*** in, UChar* out, UInt width, UInt height, UInt rati buff++; } } - pU -= width >> ratioChromaHor; - pV -= width >> ratioChromaHor; } + pLineU += uiChromaStride; + pLineV += uiChromaStride; } } @@ -113,7 +119,144 @@ Void bufferFromFrameYUV420p( Pel ***in, Pel* out, UInt width, UInt height ) } Void fillRGBbufferYUV420p( Pel*** in, UChar* out, UInt width, UInt height ) { +#ifndef USE_SSE fillRGBbufferYUV( in, out, width, height, 1, 1 ); +#else + uint8_t *yp = in[LUMA][0]; + uint8_t *up = in[CHROMA_U][0]; + uint8_t *vp = in[CHROMA_V][0]; + uint32_t suv = width >> 1; + uint32_t *rgb = ( uint32_t* )out; + + __m128i y0r0, y0r1, u0, v0; + __m128i y00r0, y01r0, y00r1, y01r1; + __m128i u00, u01, v00, v01; + __m128i rv00, rv01, gu00, gu01, gv00, gv01, bu00, bu01; + __m128i r00, r01, g00, g01, b00, b01; + __m128i rgb0123, rgb4567, rgb89ab, rgbcdef; + __m128i gbgb; + __m128i ysub, uvsub; + __m128i zero, facy, facrv, facgu, facgv, facbu; + __m128i *srcy128r0, *srcy128r1; + __m128i *dstrgb128r0, *dstrgb128r1; + __m64 *srcu64, *srcv64; + UInt x, y; + + ysub = _mm_set1_epi32( 0x00100010 ); + uvsub = _mm_set1_epi32( 0x00800080 ); + + facy = _mm_set1_epi32( 0x004a004a ); + facrv = _mm_set1_epi32( 0x00660066 ); + facgu = _mm_set1_epi32( 0x00190019 ); + facgv = _mm_set1_epi32( 0x00340034 ); + facbu = _mm_set1_epi32( 0x00810081 ); + + zero = _mm_set1_epi32( 0x00000000 ); + + for( y = 0; y < height; y += 2 ) + { + + srcy128r0 = ( __m128i * )( yp + width * y ); + srcy128r1 = ( __m128i * )( yp + width * y + width ); + srcu64 = ( __m64 * )( up + suv * ( y / 2 ) ); + srcv64 = ( __m64 * )( vp + suv * ( y / 2 ) ); + + dstrgb128r0 = ( __m128i * )( rgb + width * y ); + dstrgb128r1 = ( __m128i * )( rgb + width * y + width ); + + for( x = 0; x < width; x += 16 ) + { + + u0 = _mm_loadl_epi64( ( __m128i * )srcu64 ); + srcu64++; + v0 = _mm_loadl_epi64( ( __m128i * )srcv64 ); + srcv64++; + + y0r0 = _mm_load_si128( srcy128r0++ ); + y0r1 = _mm_load_si128( srcy128r1++ ); + + // constant y factors + y00r0 = _mm_mullo_epi16( _mm_sub_epi16( _mm_unpacklo_epi8( y0r0, zero ), ysub ), facy ); + y01r0 = _mm_mullo_epi16( _mm_sub_epi16( _mm_unpackhi_epi8( y0r0, zero ), ysub ), facy ); + y00r1 = _mm_mullo_epi16( _mm_sub_epi16( _mm_unpacklo_epi8( y0r1, zero ), ysub ), facy ); + y01r1 = _mm_mullo_epi16( _mm_sub_epi16( _mm_unpackhi_epi8( y0r1, zero ), ysub ), facy ); + + // expand u and v so they're aligned with y values + u0 = _mm_unpacklo_epi8( u0, zero ); + u00 = _mm_sub_epi16( _mm_unpacklo_epi16( u0, u0 ), uvsub ); + u01 = _mm_sub_epi16( _mm_unpackhi_epi16( u0, u0 ), uvsub ); + + v0 = _mm_unpacklo_epi8( v0, zero ); + v00 = _mm_sub_epi16( _mm_unpacklo_epi16( v0, v0 ), uvsub ); + v01 = _mm_sub_epi16( _mm_unpackhi_epi16( v0, v0 ), uvsub ); + + // common factors on both rows. + rv00 = _mm_mullo_epi16( facrv, v00 ); + rv01 = _mm_mullo_epi16( facrv, v01 ); + gu00 = _mm_mullo_epi16( facgu, u00 ); + gu01 = _mm_mullo_epi16( facgu, u01 ); + gv00 = _mm_mullo_epi16( facgv, v00 ); + gv01 = _mm_mullo_epi16( facgv, v01 ); + bu00 = _mm_mullo_epi16( facbu, u00 ); + bu01 = _mm_mullo_epi16( facbu, u01 ); + + // row 0 + r00 = _mm_srai_epi16( _mm_add_epi16( y00r0, rv00 ), 6 ); + r01 = _mm_srai_epi16( _mm_add_epi16( y01r0, rv01 ), 6 ); + g00 = _mm_srai_epi16( _mm_sub_epi16( _mm_sub_epi16( y00r0, gu00 ), gv00 ), 6 ); + g01 = _mm_srai_epi16( _mm_sub_epi16( _mm_sub_epi16( y01r0, gu01 ), gv01 ), 6 ); + b00 = _mm_srai_epi16( _mm_add_epi16( y00r0, bu00 ), 6 ); + b01 = _mm_srai_epi16( _mm_add_epi16( y01r0, bu01 ), 6 ); + + r00 = _mm_packus_epi16( r00, r01 );// rrrr.. saturated + g00 = _mm_packus_epi16( g00, g01 );// gggg.. saturated + b00 = _mm_packus_epi16( b00, b01 );// bbbb.. saturated + + r01 = _mm_unpacklo_epi8( r00, zero );// 0r0r.. + gbgb = _mm_unpacklo_epi8( b00, g00 );// gbgb.. + rgb0123 = _mm_unpacklo_epi16( gbgb, r01 );// 0rgb0rgb.. + rgb4567 = _mm_unpackhi_epi16( gbgb, r01 );// 0rgb0rgb.. + + r01 = _mm_unpackhi_epi8( r00, zero ); + gbgb = _mm_unpackhi_epi8( b00, g00 ); + rgb89ab = _mm_unpacklo_epi16( gbgb, r01 ); + rgbcdef = _mm_unpackhi_epi16( gbgb, r01 ); + + _mm_store_si128( dstrgb128r0++, rgb0123 ); + _mm_store_si128( dstrgb128r0++, rgb4567 ); + _mm_store_si128( dstrgb128r0++, rgb89ab ); + _mm_store_si128( dstrgb128r0++, rgbcdef ); + + // row 1 + r00 = _mm_srai_epi16( _mm_add_epi16( y00r1, rv00 ), 6 ); + r01 = _mm_srai_epi16( _mm_add_epi16( y01r1, rv01 ), 6 ); + g00 = _mm_srai_epi16( _mm_sub_epi16( _mm_sub_epi16( y00r1, gu00 ), gv00 ), 6 ); + g01 = _mm_srai_epi16( _mm_sub_epi16( _mm_sub_epi16( y01r1, gu01 ), gv01 ), 6 ); + b00 = _mm_srai_epi16( _mm_add_epi16( y00r1, bu00 ), 6 ); + b01 = _mm_srai_epi16( _mm_add_epi16( y01r1, bu01 ), 6 ); + + r00 = _mm_packus_epi16( r00, r01 );// rrrr.. saturated + g00 = _mm_packus_epi16( g00, g01 );// gggg.. saturated + b00 = _mm_packus_epi16( b00, b01 );// bbbb.. saturated + + r01 = _mm_unpacklo_epi8( r00, zero );// 0r0r.. + gbgb = _mm_unpacklo_epi8( b00, g00 );// gbgb.. + rgb0123 = _mm_unpacklo_epi16( gbgb, r01 );// 0rgb0rgb.. + rgb4567 = _mm_unpackhi_epi16( gbgb, r01 );// 0rgb0rgb.. + + r01 = _mm_unpackhi_epi8( r00, zero ); + gbgb = _mm_unpackhi_epi8( b00, g00 ); + rgb89ab = _mm_unpacklo_epi16( gbgb, r01 ); + rgbcdef = _mm_unpackhi_epi16( gbgb, r01 ); + + _mm_store_si128( dstrgb128r1++, rgb0123 ); + _mm_store_si128( dstrgb128r1++, rgb4567 ); + _mm_store_si128( dstrgb128r1++, rgb89ab ); + _mm_store_si128( dstrgb128r1++, rgbcdef ); + + } + } +#endif } PlaYUVerFramePelFormat yuv420p = {