From 1c1b257d552e984e2953b31d4b0b2d203c664a8a Mon Sep 17 00:00:00 2001
From: Vadim Seregin <vseregin@qti.qualcomm.com>
Date: Thu, 21 Oct 2021 08:49:46 -0700
Subject: [PATCH] Lossless code cleanup, mainly TMP related
---
source/Lib/CommonLib/InterPrediction.cpp | 4 +
source/Lib/CommonLib/IntraPrediction.cpp | 546 +++++++++++++++++-
source/Lib/CommonLib/IntraPrediction.h | 87 ++-
source/Lib/CommonLib/TrQuant.cpp | 543 -----------------
source/Lib/CommonLib/TrQuant.h | 83 +--
source/Lib/CommonLib/TypeDef.h | 16 +-
source/Lib/CommonLib/x86/InitX86.cpp | 25 +-
source/Lib/CommonLib/x86/IntraX86.h | 321 ++++++++++
source/Lib/CommonLib/x86/TrQuantX86.h | 270 +--------
source/Lib/CommonLib/x86/avx/Intra_avx.cpp | 1 +
source/Lib/CommonLib/x86/avx2/Intra_avx2.cpp | 1 +
.../Lib/CommonLib/x86/sse41/Intra_sse41.cpp | 1 +
.../Lib/CommonLib/x86/sse42/Intra_sse42.cpp | 1 +
source/Lib/DecoderLib/DecCu.cpp | 25 +-
source/Lib/EncoderLib/IntraSearch.cpp | 42 +-
15 files changed, 1030 insertions(+), 936 deletions(-)
create mode 100644 source/Lib/CommonLib/x86/IntraX86.h
create mode 100644 source/Lib/CommonLib/x86/avx/Intra_avx.cpp
create mode 100644 source/Lib/CommonLib/x86/avx2/Intra_avx2.cpp
create mode 100644 source/Lib/CommonLib/x86/sse41/Intra_sse41.cpp
create mode 100644 source/Lib/CommonLib/x86/sse42/Intra_sse42.cpp
diff --git a/source/Lib/CommonLib/InterPrediction.cpp b/source/Lib/CommonLib/InterPrediction.cpp
index 7fe099653..e4873b7a5 100644
--- a/source/Lib/CommonLib/InterPrediction.cpp
+++ b/source/Lib/CommonLib/InterPrediction.cpp
@@ -48,6 +48,10 @@
#include "Reshape.h"
#endif
+#if ENABLE_SIMD_TMP
+#include "CommonDefX86.h"
+#endif
+
//! \ingroup CommonLib
//! \{
diff --git a/source/Lib/CommonLib/IntraPrediction.cpp b/source/Lib/CommonLib/IntraPrediction.cpp
index 5e5a66392..f16ddaa80 100644
--- a/source/Lib/CommonLib/IntraPrediction.cpp
+++ b/source/Lib/CommonLib/IntraPrediction.cpp
@@ -107,6 +107,9 @@ IntraPrediction::IntraPrediction()
#if MMLM
m_encPreRDRun = false;
#endif
+#if JVET_V0130_INTRA_TMP
+ m_pppTarPatch = NULL;
+#endif
}
IntraPrediction::~IntraPrediction()
@@ -140,6 +143,31 @@ void IntraPrediction::destroy()
buffer.destroy();
}
m_tempBuffer.clear();
+
+#if JVET_V0130_INTRA_TMP
+ if( m_pppTarPatch != NULL )
+ {
+ for( unsigned int uiDepth = 0; uiDepth < USE_MORE_BLOCKSIZE_DEPTH_MAX; uiDepth++ )
+ {
+ unsigned int blkSize = g_uiDepth2Width[uiDepth];
+
+ unsigned int patchSize = blkSize + TMP_TEMPLATE_SIZE;
+ for( unsigned int uiRow = 0; uiRow < patchSize; uiRow++ )
+ {
+ if( m_pppTarPatch[uiDepth][uiRow] != NULL )
+ {
+ delete[]m_pppTarPatch[uiDepth][uiRow]; m_pppTarPatch[uiDepth][uiRow] = NULL;
+ }
+ }
+ if( m_pppTarPatch[uiDepth] != NULL )
+ {
+ delete[]m_pppTarPatch[uiDepth]; m_pppTarPatch[uiDepth] = NULL;
+ }
+ }
+ delete[] m_pppTarPatch;
+ m_pppTarPatch = NULL;
+ }
+#endif
}
void IntraPrediction::init(ChromaFormat chromaFormatIDC, const unsigned bitDepthY)
@@ -210,6 +238,34 @@ void IntraPrediction::init(ChromaFormat chromaFormatIDC, const unsigned bitDepth
{
buffer.create( chromaFormatIDC, Area( 0, 0, MAX_CU_SIZE, MAX_CU_SIZE ) );
}
+
+#if JVET_V0130_INTRA_TMP
+ unsigned int blkSize;
+
+ if( m_pppTarPatch == NULL )
+ {
+ m_pppTarPatch = new Pel * *[USE_MORE_BLOCKSIZE_DEPTH_MAX];
+ for( unsigned int uiDepth = 0; uiDepth < USE_MORE_BLOCKSIZE_DEPTH_MAX; uiDepth++ )
+ {
+ blkSize = g_uiDepth2Width[uiDepth];
+
+ unsigned int patchSize = blkSize + TMP_TEMPLATE_SIZE;
+ m_pppTarPatch[uiDepth] = new Pel *[patchSize];
+ for( unsigned int uiRow = 0; uiRow < patchSize; uiRow++ )
+ {
+ m_pppTarPatch[uiDepth][uiRow] = new Pel[patchSize];
+ }
+ }
+ }
+
+ m_calcTemplateDiff = calcTemplateDiff;
+#endif
+
+#if ENABLE_SIMD_TMP
+#ifdef TARGET_SIMD_X86
+ initIntraX86();
+#endif
+#endif
}
#if JVET_W0123_TIMD_FUSION
@@ -1984,7 +2040,7 @@ void IntraPrediction::initIntraPatternChTypeISP(const CodingUnit& cu, const Comp
#if JVET_V0130_INTRA_TMP
#if JVET_W0069_TMP_BOUNDARY
-RefTemplateType IntraPrediction::GetRefTemplateType(CodingUnit& cu, CompArea& area)
+RefTemplateType IntraPrediction::getRefTemplateType(CodingUnit& cu, CompArea& area)
#else
bool IntraPrediction::isRefTemplateAvailable(CodingUnit& cu, CompArea& area)
#endif
@@ -5556,4 +5612,492 @@ int &a, int &b, int &iShift)
}
#endif
+#if JVET_V0130_INTRA_TMP
+void insertNode( int diff, int& iXOffset, int& iYOffset, int& pDiff, int& pX, int& pY, short& pId, unsigned int& setId )
+{
+ pDiff = diff;
+ pX = iXOffset;
+ pY = iYOffset;
+ pId = setId;
+}
+
+void clipMvIntraConstraint( CodingUnit* pcCU, int regionId, int& iHorMin, int& iHorMax, int& iVerMin, int& iVerMax, unsigned int uiTemplateSize, unsigned int uiBlkWidth, unsigned int uiBlkHeight, int iCurrY, int iCurrX, int offsetLCUY, int offsetLCUX )
+{
+ int searchRangeWidth = TMP_SEARCH_RANGE_MULT_FACTOR * uiBlkWidth;
+ int searchRangeHeight = TMP_SEARCH_RANGE_MULT_FACTOR * uiBlkHeight;
+ int iMvShift = 0;
+ int iTemplateSize = uiTemplateSize;
+ int iBlkWidth = uiBlkWidth;
+ int iBlkHeight = uiBlkHeight;
+ if( regionId == 0 ) //above outside LCU
+ {
+ iHorMax = std::min( (iCurrX + searchRangeWidth) << iMvShift, ( int ) ((pcCU->cs->pps->getPicWidthInLumaSamples() - iBlkWidth) << iMvShift) );
+ iHorMin = std::max( (iTemplateSize) << iMvShift, (iCurrX - searchRangeWidth) << iMvShift );
+
+ iVerMax = (iCurrY - iBlkHeight - offsetLCUY) << iMvShift;
+ iVerMin = std::max( ((iTemplateSize) << iMvShift), ((iCurrY - searchRangeHeight) << iMvShift) );
+
+ iHorMin = iHorMin - iCurrX;
+ iHorMax = iHorMax - iCurrX;
+ iVerMax = iVerMax - iCurrY;
+ iVerMin = iVerMin - iCurrY;
+ }
+ else if( regionId == 1 ) //left outside LCU
+ {
+ iHorMax = (iCurrX - offsetLCUX - iBlkWidth) << iMvShift;
+ iHorMin = std::max( (iTemplateSize) << iMvShift, (iCurrX - searchRangeWidth) << iMvShift );
+
+ iVerMin = std::max( (iTemplateSize) << iMvShift, (iCurrY - iBlkHeight - offsetLCUY) << iMvShift );
+ iVerMax = (iCurrY) << iMvShift;
+
+ iHorMin = iHorMin - iCurrX;
+ iHorMax = iHorMax - iCurrX;
+ iVerMax = iVerMax - iCurrY;
+ iVerMin = iVerMin - iCurrY;
+ }
+ else if( regionId == 2 ) //left outside LCU (can reach the bottom row of LCU)
+ {
+ iHorMin = std::max( (iTemplateSize) << iMvShift, (iCurrX - searchRangeWidth) << iMvShift );
+ iHorMax = (iCurrX - offsetLCUX - iBlkWidth) << iMvShift;
+ iVerMin = (iCurrY + 1) << iMvShift;
+ iVerMax = std::min( pcCU->cs->pps->getPicHeightInLumaSamples() - iBlkHeight, (iCurrY - offsetLCUY + pcCU->cs->sps->getCTUSize() - iBlkHeight) << iMvShift );
+
+ iHorMin = iHorMin - iCurrX;
+ iHorMax = iHorMax - iCurrX;
+ iVerMax = iVerMax - iCurrY;
+ iVerMin = iVerMin - iCurrY;
+ }
+}
+
+TempLibFast::TempLibFast()
+{
+}
+
+TempLibFast::~TempLibFast()
+{
+}
+
+void TempLibFast::initTemplateDiff( unsigned int uiPatchWidth, unsigned int uiPatchHeight, unsigned int uiBlkWidth, unsigned int uiBlkHeight, int bitDepth )
+{
+ int maxValue = ((1 << bitDepth) >> (INIT_THRESHOULD_SHIFTBITS)) * (uiPatchHeight * uiPatchWidth - uiBlkHeight * uiBlkWidth);
+ m_diffMax = maxValue;
+ {
+ m_pDiff = maxValue;
+ }
+}
+
+#if JVET_W0069_TMP_BOUNDARY
+void IntraPrediction::getTargetTemplate( CodingUnit* pcCU, unsigned int uiBlkWidth, unsigned int uiBlkHeight, RefTemplateType tempType )
+#else
+void IntraPrediction::getTargetTemplate( CodingUnit* pcCU, unsigned int uiBlkWidth, unsigned int uiBlkHeight )
+#endif
+{
+ const ComponentID compID = COMPONENT_Y;
+ unsigned int uiPatchWidth = uiBlkWidth + TMP_TEMPLATE_SIZE;
+ unsigned int uiPatchHeight = uiBlkHeight + TMP_TEMPLATE_SIZE;
+ unsigned int uiTarDepth = floorLog2( std::max( uiBlkHeight, uiBlkWidth ) ) - 2;
+ Pel** tarPatch = m_pppTarPatch[uiTarDepth];
+ CompArea area = pcCU->blocks[compID];
+ Pel* pCurrStart = pcCU->cs->picture->getRecoBuf( area ).buf;
+ unsigned int uiPicStride = pcCU->cs->picture->getRecoBuf( compID ).stride;
+ unsigned int uiY, uiX;
+
+ //fill template
+ //up-left & up
+ Pel* tarTemp;
+#if JVET_W0069_TMP_BOUNDARY
+ if( tempType == L_SHAPE_TEMPLATE )
+ {
+#endif
+ Pel* pCurrTemp = pCurrStart - TMP_TEMPLATE_SIZE * uiPicStride - TMP_TEMPLATE_SIZE;
+ for( uiY = 0; uiY < TMP_TEMPLATE_SIZE; uiY++ )
+ {
+ tarTemp = tarPatch[uiY];
+ for( uiX = 0; uiX < uiPatchWidth; uiX++ )
+ {
+ tarTemp[uiX] = pCurrTemp[uiX];
+ }
+ pCurrTemp += uiPicStride;
+ }
+ //left
+ for( uiY = TMP_TEMPLATE_SIZE; uiY < uiPatchHeight; uiY++ )
+ {
+ tarTemp = tarPatch[uiY];
+ for( uiX = 0; uiX < TMP_TEMPLATE_SIZE; uiX++ )
+ {
+ tarTemp[uiX] = pCurrTemp[uiX];
+ }
+ pCurrTemp += uiPicStride;
+ }
+#if JVET_W0069_TMP_BOUNDARY
+ }
+ else if( tempType == ABOVE_TEMPLATE )
+ {
+ Pel* pCurrTemp = pCurrStart - TMP_TEMPLATE_SIZE * uiPicStride;
+ for( uiY = 0; uiY < TMP_TEMPLATE_SIZE; uiY++ )
+ {
+ tarTemp = tarPatch[uiY];
+ for( uiX = 0; uiX < uiBlkWidth; uiX++ )
+ {
+ tarTemp[uiX] = pCurrTemp[uiX];
+ }
+ pCurrTemp += uiPicStride;
+ }
+ }
+ else if( tempType == LEFT_TEMPLATE )
+ {
+ Pel* pCurrTemp = pCurrStart - TMP_TEMPLATE_SIZE;
+ for( uiY = TMP_TEMPLATE_SIZE; uiY < uiPatchHeight; uiY++ )
+ {
+ tarTemp = tarPatch[uiY];
+ for( uiX = 0; uiX < TMP_TEMPLATE_SIZE; uiX++ )
+ {
+ tarTemp[uiX] = pCurrTemp[uiX];
+ }
+ pCurrTemp += uiPicStride;
+ }
+ }
+#endif
+}
+
+#if JVET_W0069_TMP_BOUNDARY
+void IntraPrediction::candidateSearchIntra( CodingUnit* pcCU, unsigned int uiBlkWidth, unsigned int uiBlkHeight, RefTemplateType tempType )
+#else
+void IntraPrediction::candidateSearchIntra( CodingUnit* pcCU, unsigned int uiBlkWidth, unsigned int uiBlkHeight )
+#endif
+{
+ const ComponentID compID = COMPONENT_Y;
+ const int channelBitDepth = pcCU->cs->sps->getBitDepth( toChannelType( compID ) );
+ unsigned int uiPatchWidth = uiBlkWidth + TMP_TEMPLATE_SIZE;
+ unsigned int uiPatchHeight = uiBlkHeight + TMP_TEMPLATE_SIZE;
+ unsigned int uiTarDepth = floorLog2( std::max( uiBlkWidth, uiBlkHeight ) ) - 2;
+ Pel** tarPatch = getTargetPatch( uiTarDepth );
+ //Initialize the library for saving the best candidates
+ m_tempLibFast.initTemplateDiff( uiPatchWidth, uiPatchHeight, uiBlkWidth, uiBlkHeight, channelBitDepth );
+ short setId = 0; //record the reference picture.
+#if JVET_W0069_TMP_BOUNDARY
+ searchCandidateFromOnePicIntra( pcCU, tarPatch, uiPatchWidth, uiPatchHeight, setId, tempType );
+#else
+ searchCandidateFromOnePicIntra( pcCU, tarPatch, uiPatchWidth, uiPatchHeight, setId );
+#endif
+ //count collected candidate number
+ int pDiff = m_tempLibFast.getDiff();
+ int maxDiff = m_tempLibFast.getDiffMax();
+
+
+ if( pDiff < maxDiff )
+ {
+ m_uiVaildCandiNum = 1;
+ }
+ else
+ {
+ m_uiVaildCandiNum = 0;
+ }
+}
+
+#if JVET_W0069_TMP_BOUNDARY
+void IntraPrediction::searchCandidateFromOnePicIntra( CodingUnit* pcCU, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, unsigned int setId, RefTemplateType tempType )
+#else
+void IntraPrediction::searchCandidateFromOnePicIntra( CodingUnit* pcCU, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, unsigned int setId )
+#endif
+{
+ const ComponentID compID = COMPONENT_Y;
+ unsigned int uiBlkWidth = uiPatchWidth - TMP_TEMPLATE_SIZE;
+ unsigned int uiBlkHeight = uiPatchHeight - TMP_TEMPLATE_SIZE;
+
+ int pX = m_tempLibFast.getX();
+ int pY = m_tempLibFast.getY();
+ int pDiff = m_tempLibFast.getDiff();
+ short pId = m_tempLibFast.getId();
+ CompArea area = pcCU->blocks[compID];
+ int refStride = pcCU->cs->picture->getRecoBuf( compID ).stride;
+
+ Pel* ref = pcCU->cs->picture->getRecoBuf( area ).buf;
+
+ setRefPicUsed( ref ); //facilitate the access of each candidate point
+ setStride( refStride );
+
+ Mv cTmpMvPred;
+ cTmpMvPred.setZero();
+
+ unsigned int uiCUPelY = area.pos().y;
+ unsigned int uiCUPelX = area.pos().x;
+ int blkX = 0;
+ int blkY = 0;
+ int iCurrY = uiCUPelY + blkY;
+ int iCurrX = uiCUPelX + blkX;
+
+ Position ctuRsAddr = CU::getCtuXYAddr( *pcCU );
+ int offsetLCUY = iCurrY - ctuRsAddr.y;
+ int offsetLCUX = iCurrX - ctuRsAddr.x;
+
+ int iYOffset, iXOffset;
+ int diff;
+ Pel* refCurr;
+
+ const int regionNum = 3;
+ int mvYMins[regionNum];
+ int mvYMaxs[regionNum];
+ int mvXMins[regionNum];
+ int mvXMaxs[regionNum];
+ int regionId = 0;
+
+ //1. check the near pixels within LCU
+ //above pixels in LCU
+ int iTemplateSize = TMP_TEMPLATE_SIZE;
+ int iBlkWidth = uiBlkWidth;
+ int iBlkHeight = uiBlkHeight;
+ regionId = 0;
+ int iMvShift = 0;
+
+ int iVerMin = std::max( ((iTemplateSize) << iMvShift), (iCurrY - offsetLCUY - iBlkHeight + 1) << iMvShift );
+ int iVerMax = (iCurrY - iBlkHeight) << iMvShift;
+ int iHorMin = std::max( (iTemplateSize) << iMvShift, (iCurrX - offsetLCUX - iBlkWidth + 1) << iMvShift );
+ int iHorMax = (iCurrX - iBlkWidth);
+
+ mvXMins[regionId] = iHorMin - iCurrX;
+ mvXMaxs[regionId] = iHorMax - iCurrX;
+ mvYMins[regionId] = iVerMin - iCurrY;
+ mvYMaxs[regionId] = iVerMax - iCurrY;
+
+ //check within CTU pixels
+ for( regionId = 0; regionId < 1; regionId++ )
+ {
+ int mvYMin = mvYMins[regionId];
+ int mvYMax = mvYMaxs[regionId];
+ int mvXMin = mvXMins[regionId];
+ int mvXMax = mvXMaxs[regionId];
+ if( mvYMax < mvYMin || mvXMax < mvXMin )
+ {
+ continue;
+ }
+
+ for( iYOffset = mvYMax; iYOffset >= mvYMin; iYOffset-- )
+ {
+ for( iXOffset = mvXMax; iXOffset >= mvXMin; iXOffset-- )
+ {
+ refCurr = ref + iYOffset * refStride + iXOffset;
+#if JVET_W0069_TMP_BOUNDARY
+ diff = m_calcTemplateDiff( refCurr, refStride, tarPatch, uiPatchWidth, uiPatchHeight, pDiff, tempType );
+#else
+ diff = m_calcTemplateDiff( refCurr, refStride, tarPatch, uiPatchWidth, uiPatchHeight, pDiff );
+#endif
+ if( diff < (pDiff) )
+ {
+ insertNode( diff, iXOffset, iYOffset, pDiff, pX, pY, pId, setId );
+ }
+ if( pDiff == 0 )
+ {
+ regionId++;
+ }
+ }
+ }
+ }
+
+ //2. check the pixels outside CTU
+ for( regionId = 0; regionId < regionNum; regionId++ )
+ {
+ // this function fills in the range the template matching for pixels outside the current CTU
+ clipMvIntraConstraint( pcCU, regionId, mvXMins[regionId], mvXMaxs[regionId], mvYMins[regionId], mvYMaxs[regionId], TMP_TEMPLATE_SIZE, uiBlkWidth, uiBlkHeight, iCurrY, iCurrX, offsetLCUY, offsetLCUX );
+ }
+
+ for( regionId = 0; regionId < regionNum; regionId++ )
+ {
+ int mvYMin = mvYMins[regionId];
+ int mvYMax = mvYMaxs[regionId];
+ int mvXMin = mvXMins[regionId];
+ int mvXMax = mvXMaxs[regionId];
+ if( mvYMax < mvYMin || mvXMax < mvXMin )
+ {
+ continue;
+ }
+ for( iYOffset = mvYMax; iYOffset >= mvYMin; iYOffset-- )
+ {
+ for( iXOffset = mvXMax; iXOffset >= mvXMin; iXOffset-- )
+ {
+ refCurr = ref + iYOffset * refStride + iXOffset;
+#if JVET_W0069_TMP_BOUNDARY
+ diff = m_calcTemplateDiff( refCurr, refStride, tarPatch, uiPatchWidth, uiPatchHeight, pDiff, tempType );
+#else
+ diff = m_calcTemplateDiff( refCurr, refStride, tarPatch, uiPatchWidth, uiPatchHeight, pDiff );
+#endif
+
+ if( diff < (pDiff) )
+ {
+ insertNode( diff, iXOffset, iYOffset, pDiff, pX, pY, pId, setId );
+ }
+
+ if( pDiff == 0 )
+ {
+ regionId = regionNum;
+ }
+ }
+ }
+ }
+
+ m_tempLibFast.m_pX = pX;
+ m_tempLibFast.m_pY = pY;
+ m_tempLibFast.m_pDiff = pDiff;
+ m_tempLibFast.m_pId = pId;
+}
+bool IntraPrediction::generateTMPrediction( Pel* piPred, unsigned int uiStride, unsigned int uiBlkWidth, unsigned int uiBlkHeight, int& foundCandiNum )
+{
+ bool bSucceedFlag = true;
+ unsigned int uiPatchWidth = uiBlkWidth + TMP_TEMPLATE_SIZE;
+ unsigned int uiPatchHeight = uiBlkHeight + TMP_TEMPLATE_SIZE;
+
+ foundCandiNum = m_uiVaildCandiNum;
+ if( foundCandiNum < 1 )
+ {
+ return false;
+ }
+
+ int pX = m_tempLibFast.getX();
+ int pY = m_tempLibFast.getY();
+ Pel* ref;
+ int picStride = getStride();
+ int iOffsetY, iOffsetX;
+ Pel* refTarget;
+ unsigned int uiHeight = uiPatchHeight - TMP_TEMPLATE_SIZE;
+ unsigned int uiWidth = uiPatchWidth - TMP_TEMPLATE_SIZE;
+
+ //the data center: we use the prediction block as the center now.
+ //collect the candidates
+ ref = getRefPicUsed();
+ {
+ iOffsetY = pY;
+ iOffsetX = pX;
+ refTarget = ref + iOffsetY * picStride + iOffsetX;
+ for( unsigned int uiY = 0; uiY < uiHeight; uiY++ )
+ {
+ for( unsigned int uiX = 0; uiX < uiWidth; uiX++ )
+ {
+ piPred[uiX] = refTarget[uiX];
+ }
+ refTarget += picStride;
+ piPred += uiStride;
+ }
+ }
+ return bSucceedFlag;
+}
+
+#if JVET_W0069_TMP_BOUNDARY
+bool IntraPrediction::generateTmDcPrediction( Pel* piPred, unsigned int uiStride, unsigned int uiBlkWidth, unsigned int uiBlkHeight, int DC_Val )
+{
+ bool bSucceedFlag = true;
+ {
+ for( unsigned int uiY = 0; uiY < uiBlkHeight; uiY++ )
+ {
+ for( unsigned int uiX = 0; uiX < uiBlkWidth; uiX++ )
+ {
+ piPred[uiX] = DC_Val;
+ }
+ piPred += uiStride;
+ }
+ }
+ return bSucceedFlag;
+}
+#endif
+
+#if JVET_W0069_TMP_BOUNDARY
+int IntraPrediction::calcTemplateDiff( Pel* ref, unsigned int uiStride, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, int iMax, RefTemplateType tempType )
+#else
+int IntraPrediction::calcTemplateDiff( Pel* ref, unsigned int uiStride, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, int iMax )
+#endif
+{
+ int diffSum = 0;
+#if JVET_W0069_TMP_BOUNDARY
+ Pel* refPatchRow;
+ if( tempType == L_SHAPE_TEMPLATE )
+ {
+ refPatchRow = ref - TMP_TEMPLATE_SIZE * uiStride - TMP_TEMPLATE_SIZE;
+ }
+ else if( tempType == LEFT_TEMPLATE )
+ {
+ refPatchRow = ref - TMP_TEMPLATE_SIZE;
+ }
+ else if( tempType == ABOVE_TEMPLATE )
+ {
+ refPatchRow = ref - TMP_TEMPLATE_SIZE * uiStride;
+ }
+#else
+ Pel* refPatchRow = ref - TMP_TEMPLATE_SIZE * uiStride - TMP_TEMPLATE_SIZE;
+#endif
+ Pel* tarPatchRow;
+
+#if JVET_W0069_TMP_BOUNDARY
+ if( tempType == L_SHAPE_TEMPLATE )
+ {
+#endif
+ // horizontal difference
+ for( int iY = 0; iY < TMP_TEMPLATE_SIZE; iY++ )
+ {
+ tarPatchRow = tarPatch[iY];
+ for( int iX = 0; iX < uiPatchWidth; iX++ )
+ {
+ diffSum += abs( refPatchRow[iX] - tarPatchRow[iX] );
+ }
+ if( diffSum > iMax ) //for speeding up
+ {
+ return diffSum;
+ }
+ refPatchRow += uiStride;
+ }
+
+ // vertical difference
+ for( int iY = TMP_TEMPLATE_SIZE; iY < uiPatchHeight; iY++ )
+ {
+ tarPatchRow = tarPatch[iY];
+ for( int iX = 0; iX < TMP_TEMPLATE_SIZE; iX++ )
+ {
+ diffSum += abs( refPatchRow[iX] - tarPatchRow[iX] );
+ }
+ if( diffSum > iMax ) //for speeding up
+ {
+ return diffSum;
+ }
+ refPatchRow += uiStride;
+ }
+#if JVET_W0069_TMP_BOUNDARY
+ }
+ else if( tempType == ABOVE_TEMPLATE )
+ {
+ // top template difference
+ for( int iY = 0; iY < TMP_TEMPLATE_SIZE; iY++ )
+ {
+ tarPatchRow = tarPatch[iY];
+ for( int iX = 0; iX < uiPatchWidth - TMP_TEMPLATE_SIZE; iX++ )
+ {
+ diffSum += abs( refPatchRow[iX] - tarPatchRow[iX] );
+ }
+ if( diffSum > iMax ) //for speeding up
+ {
+ return diffSum;
+ }
+ refPatchRow += uiStride;
+ }
+ }
+ else if( tempType == LEFT_TEMPLATE )
+ {
+ // left template difference
+ for( int iY = TMP_TEMPLATE_SIZE; iY < uiPatchHeight; iY++ )
+ {
+ tarPatchRow = tarPatch[iY];
+ for( int iX = 0; iX < TMP_TEMPLATE_SIZE; iX++ )
+ {
+ diffSum += abs( refPatchRow[iX] - tarPatchRow[iX] );
+ }
+ if( diffSum > iMax ) //for speeding up
+ {
+ return diffSum;
+ }
+ refPatchRow += uiStride;
+ }
+ }
+#endif
+
+ return diffSum;
+}
+#endif
+
//! \}
diff --git a/source/Lib/CommonLib/IntraPrediction.h b/source/Lib/CommonLib/IntraPrediction.h
index d89f5d2e7..b97ad8975 100644
--- a/source/Lib/CommonLib/IntraPrediction.h
+++ b/source/Lib/CommonLib/IntraPrediction.h
@@ -66,6 +66,43 @@ enum PredBuf
static const uint32_t MAX_INTRA_FILTER_DEPTHS=8;
+#if JVET_V0130_INTRA_TMP
+extern unsigned int g_uiDepth2Width[5];
+extern unsigned int g_uiDepth2MaxCandiNum[5];
+
+class TempLibFast
+{
+public:
+ int m_pX; //offset X
+ int m_pY; //offset Y
+ int m_pXInteger; //offset X for integer pixel search
+ int m_pYInteger; //offset Y for integer pixel search
+ int m_pDiffInteger;
+ int getXInteger() { return m_pXInteger; }
+ int getYInteger() { return m_pYInteger; }
+ int getDiffInteger() { return m_pDiffInteger; }
+ short m_pIdInteger; //frame id
+ short getIdInteger() { return m_pIdInteger; }
+ int m_pDiff; //mse
+ short m_pId; //frame id
+
+ TempLibFast();
+ ~TempLibFast();
+ //void init();
+ int getX() { return m_pX; }
+ int getY() { return m_pY; }
+ int getDiff() { return m_pDiff; }
+ short getId() { return m_pId; }
+ /*void initDiff(unsigned int uiPatchSize, int bitDepth);
+ void initDiff(unsigned int uiPatchSize, int bitDepth, int iCandiNumber);*/
+ void initTemplateDiff( unsigned int uiPatchWidth, unsigned int uiPatchHeight, unsigned int uiBlkWidth, unsigned int uiBlkHeight, int bitDepth );
+ int m_diffMax;
+ int getDiffMax() { return m_diffMax; }
+};
+
+typedef short TrainDataType;
+#endif
+
class IntraPrediction
{
#if MMLM
@@ -140,6 +177,16 @@ protected:
bool m_bestScanRotationMode;
std::vector<PelStorage> m_tempBuffer;
+#if JVET_V0130_INTRA_TMP
+ int m_uiPartLibSize;
+ TempLibFast m_tempLibFast;
+ Pel* m_refPicUsed;
+ Picture* m_refPicBuf;
+ unsigned int m_uiPicStride;
+ unsigned int m_uiVaildCandiNum;
+ Pel*** m_pppTarPatch;
+#endif
+
// prediction
void xPredIntraPlanar ( const CPelBuf &pSrc, PelBuf &pDst );
void xPredIntraDc ( const CPelBuf &pSrc, PelBuf &pDst, const ChannelType channelType, const bool enableBoundaryFilter = true );
@@ -204,7 +251,7 @@ public:
virtual ~IntraPrediction();
#if JVET_W0069_TMP_BOUNDARY
- RefTemplateType GetRefTemplateType(CodingUnit& cu, CompArea& area);
+ RefTemplateType getRefTemplateType(CodingUnit& cu, CompArea& area);
#endif
void init (ChromaFormat chromaFormatIDC, const unsigned bitDepthY);
@@ -259,6 +306,44 @@ public:
Pel* getPredictorPtr2 (const ComponentID compID, uint32_t idx) { return m_yuvExt2[compID][idx]; }
void switchBuffer (const PredictionUnit &pu, ComponentID compID, PelBuf srcBuff, Pel *dst);
#endif
+
+#if JVET_V0130_INTRA_TMP
+#if JVET_W0069_TMP_BOUNDARY
+ int( *m_calcTemplateDiff )(Pel* ref, unsigned int uiStride, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, int iMax, RefTemplateType TempType);
+ static int calcTemplateDiff( Pel* ref, unsigned int uiStride, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, int iMax, RefTemplateType TempType );
+#else
+ int( *m_calcTemplateDiff )(Pel* ref, unsigned int uiStride, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, int iMax);
+ static int calcTemplateDiff( Pel* ref, unsigned int uiStride, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, int iMax );
+#endif
+ Pel** getTargetPatch( unsigned int uiDepth ) { return m_pppTarPatch[uiDepth]; }
+ Pel* getRefPicUsed() { return m_refPicUsed; }
+ void setRefPicUsed( Pel* ref ) { m_refPicUsed = ref; }
+ unsigned int getStride() { return m_uiPicStride; }
+ void setStride( unsigned int uiPicStride ) { m_uiPicStride = uiPicStride; }
+
+#if JVET_W0069_TMP_BOUNDARY
+ void searchCandidateFromOnePicIntra( CodingUnit* pcCU, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, unsigned int setId, RefTemplateType tempType );
+ void candidateSearchIntra( CodingUnit* pcCU, unsigned int uiBlkWidth, unsigned int uiBlkHeight, RefTemplateType tempType );
+#else
+ void searchCandidateFromOnePicIntra( CodingUnit* pcCU, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, unsigned int setId );
+ void candidateSearchIntra( CodingUnit* pcCU, unsigned int uiBlkWidth, unsigned int uiBlkHeight );
+#endif
+ bool generateTMPrediction( Pel* piPred, unsigned int uiStride, unsigned int uiBlkWidth, unsigned int uiBlkHeight, int& foundCandiNum );
+#if JVET_W0069_TMP_BOUNDARY
+ bool generateTmDcPrediction( Pel* piPred, unsigned int uiStride, unsigned int uiBlkWidth, unsigned int uiBlkHeight, int DC_Val );
+ void getTargetTemplate( CodingUnit* pcCU, unsigned int uiBlkWidth, unsigned int uiBlkHeight, RefTemplateType tempType );
+#else
+ void getTargetTemplate( CodingUnit* pcCU, unsigned int uiBlkWidth, unsigned int uiBlkHeight );
+#endif
+#endif
+
+#if ENABLE_SIMD_TMP
+#ifdef TARGET_SIMD_X86
+ void initIntraX86();
+ template <X86_VEXT vext>
+ void _initIntraX86();
+#endif
+#endif
};
#if ENABLE_DIMD
int buildHistogram(const Pel *pReco, int iStride, uint32_t uiHeight, uint32_t uiWidth, int* piHistogram, int direction, int bw, int bh);
diff --git a/source/Lib/CommonLib/TrQuant.cpp b/source/Lib/CommonLib/TrQuant.cpp
index ce6634820..349675100 100644
--- a/source/Lib/CommonLib/TrQuant.cpp
+++ b/source/Lib/CommonLib/TrQuant.cpp
@@ -55,10 +55,6 @@
#include "CommonLib/CodingStatistics.h"
#endif
-#if ENABLE_SIMD_TMP
-#include "CommonDefX86.h"
-#endif
-
#if JVET_V0130_INTRA_TMP
unsigned int g_uiDepth2Width[5] = { 4, 8, 16, 32, 64 };
#endif
@@ -195,9 +191,6 @@ TrQuant::TrQuant() : m_quant( nullptr )
m_fwdICT[-2] = fwdTransformCbCr<-2>;
m_fwdICT[ 3] = fwdTransformCbCr< 3>;
m_fwdICT[-3] = fwdTransformCbCr<-3>;
-#if JVET_V0130_INTRA_TMP
- m_pppTarPatch = NULL;
-#endif
}
}
@@ -208,31 +201,6 @@ TrQuant::~TrQuant()
delete m_quant;
m_quant = nullptr;
}
-
-#if JVET_V0130_INTRA_TMP
- if (m_pppTarPatch != NULL)
- {
- for (unsigned int uiDepth = 0; uiDepth < USE_MORE_BLOCKSIZE_DEPTH_MAX; uiDepth++)
- {
- unsigned int blkSize = g_uiDepth2Width[uiDepth];
-
- unsigned int patchSize = blkSize + TMP_TEMPLATE_SIZE;
- for (unsigned int uiRow = 0; uiRow < patchSize; uiRow++)
- {
- if (m_pppTarPatch[uiDepth][uiRow] != NULL)
- {
- delete[]m_pppTarPatch[uiDepth][uiRow]; m_pppTarPatch[uiDepth][uiRow] = NULL;
- }
- }
- if (m_pppTarPatch[uiDepth] != NULL)
- {
- delete[]m_pppTarPatch[uiDepth]; m_pppTarPatch[uiDepth] = NULL;
- }
- }
- delete[] m_pppTarPatch;
- m_pppTarPatch = NULL;
- }
-#endif
}
#if ENABLE_SPLIT_PARALLELISM
@@ -270,27 +238,6 @@ void TrQuant::init( const Quant* otherQuant,
m_quant->init( uiMaxTrSize, bUseRDOQ, bUseRDOQTS, useSelectiveRDOQ );
}
-
-#if JVET_V0130_INTRA_TMP
- unsigned int blkSize;
-
- if (m_pppTarPatch == NULL)
- {
- m_pppTarPatch = new Pel * *[USE_MORE_BLOCKSIZE_DEPTH_MAX];
- for (unsigned int uiDepth = 0; uiDepth < USE_MORE_BLOCKSIZE_DEPTH_MAX; uiDepth++)
- {
- blkSize = g_uiDepth2Width[uiDepth];
-
- unsigned int patchSize = blkSize + TMP_TEMPLATE_SIZE;
- m_pppTarPatch[uiDepth] = new Pel * [patchSize];
- for (unsigned int uiRow = 0; uiRow < patchSize; uiRow++)
- {
- m_pppTarPatch[uiDepth][uiRow] = new Pel[patchSize];
- }
- }
-}
-#endif
-
#if TU_256
fastFwdTrans =
{ {
@@ -348,9 +295,6 @@ void TrQuant::init( const Quant* otherQuant,
#if ENABLE_SIMD_SIGN_PREDICTION
m_computeSAD = xComputeSAD;
#endif
-#if JVET_V0130_INTRA_TMP
- m_calcTemplateDiff = calcTemplateDiff;
-#endif
#if ENABLE_SIMD_SIGN_PREDICTION || TRANSFORM_SIMD_OPT
#ifdef TARGET_SIMD_X86
@@ -468,493 +412,6 @@ void TrQuant::invLfnstNxN( int* src, int* dst, const uint32_t mode, const uint32
}
}
-#if JVET_V0130_INTRA_TMP
-void insertNode(int diff, int& iXOffset, int& iYOffset, int& pDiff, int& pX, int& pY, short& pId, unsigned int& setId)
-{
- pDiff = diff;
- pX = iXOffset;
- pY = iYOffset;
- pId = setId;
-}
-
-void clipMvIntraConstraint(CodingUnit* pcCU, int regionId, int& iHorMin, int& iHorMax, int& iVerMin, int& iVerMax, unsigned int uiTemplateSize, unsigned int uiBlkWidth, unsigned int uiBlkHeight, int iCurrY, int iCurrX, int offsetLCUY, int offsetLCUX)
-{
- int searchRangeWidth = TMP_SEARCH_RANGE_MULT_FACTOR * uiBlkWidth;
- int searchRangeHeight = TMP_SEARCH_RANGE_MULT_FACTOR * uiBlkHeight;
- int iMvShift = 0;
- int iTemplateSize = uiTemplateSize;
- int iBlkWidth = uiBlkWidth;
- int iBlkHeight = uiBlkHeight;
- if (regionId == 0) //above outside LCU
- {
- iHorMax = std::min((iCurrX + searchRangeWidth) << iMvShift, (int)((pcCU->cs->pps->getPicWidthInLumaSamples() - iBlkWidth) << iMvShift) );
- iHorMin = std::max((iTemplateSize) << iMvShift, (iCurrX - searchRangeWidth) << iMvShift);
-
- iVerMax = (iCurrY - iBlkHeight - offsetLCUY) << iMvShift;
- iVerMin = std::max(((iTemplateSize) << iMvShift), ((iCurrY - searchRangeHeight) << iMvShift));
-
- iHorMin = iHorMin - iCurrX;
- iHorMax = iHorMax - iCurrX;
- iVerMax = iVerMax - iCurrY;
- iVerMin = iVerMin - iCurrY;
- }
- else if (regionId == 1) //left outside LCU
- {
- iHorMax = (iCurrX - offsetLCUX - iBlkWidth) << iMvShift;
- iHorMin = std::max((iTemplateSize) << iMvShift, (iCurrX - searchRangeWidth) << iMvShift);
-
- iVerMin = std::max((iTemplateSize) << iMvShift, (iCurrY - iBlkHeight - offsetLCUY) << iMvShift);
- iVerMax = (iCurrY) << iMvShift;
-
- iHorMin = iHorMin - iCurrX;
- iHorMax = iHorMax - iCurrX;
- iVerMax = iVerMax - iCurrY;
- iVerMin = iVerMin - iCurrY;
- }
- else if (regionId == 2) //left outside LCU (can reach the bottom row of LCU)
- {
- iHorMin = std::max((iTemplateSize) << iMvShift, (iCurrX - searchRangeWidth) << iMvShift);
- iHorMax = (iCurrX - offsetLCUX - iBlkWidth) << iMvShift;
- iVerMin = (iCurrY + 1) << iMvShift;
- iVerMax = std::min(pcCU->cs->pps->getPicHeightInLumaSamples() - iBlkHeight, (iCurrY - offsetLCUY + pcCU->cs->sps->getCTUSize() - iBlkHeight) << iMvShift);
-
- iHorMin = iHorMin - iCurrX;
- iHorMax = iHorMax - iCurrX;
- iVerMax = iVerMax - iCurrY;
- iVerMin = iVerMin - iCurrY;
- }
-}
-
-TempLibFast::TempLibFast()
-{
-}
-
-TempLibFast::~TempLibFast()
-{
-}
-
-void TempLibFast::initTemplateDiff(unsigned int uiPatchWidth, unsigned int uiPatchHeight, unsigned int uiBlkWidth, unsigned int uiBlkHeight, int bitDepth)
-{
- int maxValue = ((1 << bitDepth) >> (INIT_THRESHOULD_SHIFTBITS)) * (uiPatchHeight * uiPatchWidth - uiBlkHeight * uiBlkWidth);
- m_diffMax = maxValue;
- {
- m_pDiff = maxValue;
- }
-}
-
-#if JVET_W0069_TMP_BOUNDARY
-void TrQuant::getTargetTemplate(CodingUnit* pcCU, unsigned int uiBlkWidth, unsigned int uiBlkHeight, RefTemplateType TempType)
-#else
-void TrQuant::getTargetTemplate(CodingUnit* pcCU, unsigned int uiBlkWidth, unsigned int uiBlkHeight)
-#endif
-{
- const ComponentID compID = COMPONENT_Y;
- unsigned int uiPatchWidth = uiBlkWidth + TMP_TEMPLATE_SIZE;
- unsigned int uiPatchHeight = uiBlkHeight + TMP_TEMPLATE_SIZE;
- unsigned int uiTarDepth = floorLog2(std::max(uiBlkHeight, uiBlkWidth)) - 2;
- Pel** tarPatch = m_pppTarPatch[uiTarDepth];
- CompArea area = pcCU->blocks[compID];
- Pel* pCurrStart = pcCU->cs->picture->getRecoBuf(area).buf;
- unsigned int uiPicStride = pcCU->cs->picture->getRecoBuf(compID).stride;
- unsigned int uiY, uiX;
-
- //fill template
- //up-left & up
- Pel* tarTemp;
-#if JVET_W0069_TMP_BOUNDARY
- if (TempType == L_SHAPE_TEMPLATE)
- {
-#endif
- Pel* pCurrTemp = pCurrStart - TMP_TEMPLATE_SIZE * uiPicStride - TMP_TEMPLATE_SIZE;
- for (uiY = 0; uiY < TMP_TEMPLATE_SIZE; uiY++)
- {
- tarTemp = tarPatch[uiY];
- for (uiX = 0; uiX < uiPatchWidth; uiX++)
- {
- tarTemp[uiX] = pCurrTemp[uiX];
- }
- pCurrTemp += uiPicStride;
- }
- //left
- for (uiY = TMP_TEMPLATE_SIZE; uiY < uiPatchHeight; uiY++)
- {
- tarTemp = tarPatch[uiY];
- for (uiX = 0; uiX < TMP_TEMPLATE_SIZE; uiX++)
- {
- tarTemp[uiX] = pCurrTemp[uiX];
- }
- pCurrTemp += uiPicStride;
- }
-#if JVET_W0069_TMP_BOUNDARY
- }
- else if (TempType == ABOVE_TEMPLATE)
- {
- Pel* pCurrTemp = pCurrStart - TMP_TEMPLATE_SIZE * uiPicStride;
- for (uiY = 0; uiY < TMP_TEMPLATE_SIZE; uiY++)
- {
- tarTemp = tarPatch[uiY];
- for (uiX = 0; uiX < uiBlkWidth; uiX++)
- {
- tarTemp[uiX] = pCurrTemp[uiX];
- }
- pCurrTemp += uiPicStride;
- }
- }
- else if (TempType == LEFT_TEMPLATE)
- {
- Pel* pCurrTemp = pCurrStart - TMP_TEMPLATE_SIZE;
- for (uiY = TMP_TEMPLATE_SIZE; uiY < uiPatchHeight; uiY++)
- {
- tarTemp = tarPatch[uiY];
- for (uiX = 0; uiX < TMP_TEMPLATE_SIZE; uiX++)
- {
- tarTemp[uiX] = pCurrTemp[uiX];
- }
- pCurrTemp += uiPicStride;
- }
- }
-#endif
-}
-
-#if JVET_W0069_TMP_BOUNDARY
-void TrQuant::candidateSearchIntra(CodingUnit* pcCU, unsigned int uiBlkWidth, unsigned int uiBlkHeight, RefTemplateType TempType)
-#else
-void TrQuant::candidateSearchIntra(CodingUnit* pcCU, unsigned int uiBlkWidth, unsigned int uiBlkHeight)
-#endif
-{
- const ComponentID compID = COMPONENT_Y;
- const int channelBitDepth = pcCU->cs->sps->getBitDepth(toChannelType(compID));
- unsigned int uiPatchWidth = uiBlkWidth + TMP_TEMPLATE_SIZE;
- unsigned int uiPatchHeight = uiBlkHeight + TMP_TEMPLATE_SIZE;
- unsigned int uiTarDepth = floorLog2(std::max(uiBlkWidth, uiBlkHeight)) - 2;
- Pel** tarPatch = getTargetPatch(uiTarDepth);
- //Initialize the library for saving the best candidates
- m_tempLibFast.initTemplateDiff(uiPatchWidth, uiPatchHeight, uiBlkWidth, uiBlkHeight, channelBitDepth);
- short setId = 0; //record the reference picture.
-#if JVET_W0069_TMP_BOUNDARY
- searchCandidateFromOnePicIntra(pcCU, tarPatch, uiPatchWidth, uiPatchHeight, setId, TempType);
-#else
- searchCandidateFromOnePicIntra(pcCU, tarPatch, uiPatchWidth, uiPatchHeight, setId);
-#endif
- //count collected candidate number
- int pDiff = m_tempLibFast.getDiff();
- int maxDiff = m_tempLibFast.getDiffMax();
-
-
- if( pDiff < maxDiff )
- {
- m_uiVaildCandiNum = 1;
- }
- else
- {
- m_uiVaildCandiNum = 0;
- }
-}
-
-#if JVET_W0069_TMP_BOUNDARY
-void TrQuant::searchCandidateFromOnePicIntra(CodingUnit* pcCU, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, unsigned int setId, RefTemplateType TempType)
-#else
-void TrQuant::searchCandidateFromOnePicIntra(CodingUnit* pcCU, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, unsigned int setId)
-#endif
-{
- const ComponentID compID = COMPONENT_Y;
- unsigned int uiBlkWidth = uiPatchWidth - TMP_TEMPLATE_SIZE;
- unsigned int uiBlkHeight = uiPatchHeight - TMP_TEMPLATE_SIZE;
-
- int pX = m_tempLibFast.getX();
- int pY = m_tempLibFast.getY();
- int pDiff = m_tempLibFast.getDiff();
- short pId = m_tempLibFast.getId();
- CompArea area = pcCU->blocks[compID];
- int refStride = pcCU->cs->picture->getRecoBuf(compID).stride;
-
- Pel* ref = pcCU->cs->picture->getRecoBuf(area).buf;
-
- setRefPicUsed(ref); //facilitate the access of each candidate point
- setStride(refStride);
-
- Mv cTmpMvPred;
- cTmpMvPred.setZero();
-
- unsigned int uiCUPelY = area.pos().y;
- unsigned int uiCUPelX = area.pos().x;
- int blkX = 0;
- int blkY = 0;
- int iCurrY = uiCUPelY + blkY;
- int iCurrX = uiCUPelX + blkX;
-
- Position ctuRsAddr = CU::getCtuXYAddr(*pcCU);
- int offsetLCUY = iCurrY - ctuRsAddr.y;
- int offsetLCUX = iCurrX - ctuRsAddr.x;
-
- int iYOffset, iXOffset;
- int diff;
- Pel* refCurr;
-
- const int regionNum = 3;
- int mvYMins[regionNum];
- int mvYMaxs[regionNum];
- int mvXMins[regionNum];
- int mvXMaxs[regionNum];
- int regionId = 0;
-
- //1. check the near pixels within LCU
- //above pixels in LCU
- int iTemplateSize = TMP_TEMPLATE_SIZE;
- int iBlkWidth = uiBlkWidth;
- int iBlkHeight = uiBlkHeight;
- regionId = 0;
- int iMvShift = 0;
-
- int iVerMin = std::max(((iTemplateSize) << iMvShift), (iCurrY - offsetLCUY - iBlkHeight + 1) << iMvShift);
- int iVerMax = (iCurrY - iBlkHeight) << iMvShift;
- int iHorMin = std::max((iTemplateSize) << iMvShift, (iCurrX - offsetLCUX - iBlkWidth + 1) << iMvShift);
- int iHorMax = (iCurrX - iBlkWidth);
-
- mvXMins[regionId] = iHorMin - iCurrX;
- mvXMaxs[regionId] = iHorMax - iCurrX;
- mvYMins[regionId] = iVerMin - iCurrY;
- mvYMaxs[regionId] = iVerMax - iCurrY;
-
- //check within CTU pixels
- for (regionId = 0; regionId < 1; regionId++)
- {
- int mvYMin = mvYMins[regionId];
- int mvYMax = mvYMaxs[regionId];
- int mvXMin = mvXMins[regionId];
- int mvXMax = mvXMaxs[regionId];
- if (mvYMax < mvYMin || mvXMax < mvXMin)
- {
- continue;
- }
-
- for (iYOffset = mvYMax; iYOffset >= mvYMin; iYOffset--)
- {
- for (iXOffset = mvXMax; iXOffset >= mvXMin; iXOffset--)
- {
- refCurr = ref + iYOffset * refStride + iXOffset;
-#if JVET_W0069_TMP_BOUNDARY
- diff = m_calcTemplateDiff(refCurr, refStride, tarPatch, uiPatchWidth, uiPatchHeight, pDiff, TempType);
-#else
- diff = m_calcTemplateDiff(refCurr, refStride, tarPatch, uiPatchWidth, uiPatchHeight, pDiff);
-#endif
- if (diff < (pDiff))
- {
- insertNode(diff, iXOffset, iYOffset, pDiff, pX, pY, pId, setId);
- }
- if (pDiff == 0)
- {
- regionId++;
- }
- }
- }
- }
-
- //2. check the pixels outside CTU
- for (regionId = 0; regionId < regionNum; regionId++)
- {// this function fills in the range the template matching for pixels outside the current CTU
- clipMvIntraConstraint(pcCU, regionId, mvXMins[regionId], mvXMaxs[regionId], mvYMins[regionId], mvYMaxs[regionId], TMP_TEMPLATE_SIZE, uiBlkWidth, uiBlkHeight, iCurrY, iCurrX, offsetLCUY, offsetLCUX);
- }
-
- for (regionId = 0; regionId < regionNum; regionId++)
- {
- int mvYMin = mvYMins[regionId];
- int mvYMax = mvYMaxs[regionId];
- int mvXMin = mvXMins[regionId];
- int mvXMax = mvXMaxs[regionId];
- if ( mvYMax < mvYMin || mvXMax < mvXMin )
- {
- continue;
- }
- for (iYOffset = mvYMax; iYOffset >= mvYMin; iYOffset--)
- {
- for (iXOffset = mvXMax; iXOffset >= mvXMin; iXOffset--)
- {
- refCurr = ref + iYOffset * refStride + iXOffset;
-#if JVET_W0069_TMP_BOUNDARY
- diff = m_calcTemplateDiff(refCurr, refStride, tarPatch, uiPatchWidth, uiPatchHeight, pDiff, TempType);
-#else
- diff = m_calcTemplateDiff(refCurr, refStride, tarPatch, uiPatchWidth, uiPatchHeight, pDiff);
-#endif
-
- if (diff < (pDiff))
- {
- insertNode(diff, iXOffset, iYOffset, pDiff, pX, pY, pId, setId);
- }
-
- if (pDiff == 0)
- {
- regionId = regionNum;
- }
- }
- }
- }
-
- m_tempLibFast.m_pX = pX;
- m_tempLibFast.m_pY = pY;
- m_tempLibFast.m_pDiff = pDiff;
- m_tempLibFast.m_pId = pId;
-}
-bool TrQuant::generateTMPrediction(Pel* piPred, unsigned int uiStride, unsigned int uiBlkWidth, unsigned int uiBlkHeight, int& foundCandiNum)
-{
- bool bSucceedFlag = true;
- unsigned int uiPatchWidth = uiBlkWidth + TMP_TEMPLATE_SIZE;
- unsigned int uiPatchHeight = uiBlkHeight + TMP_TEMPLATE_SIZE;
-
- foundCandiNum = m_uiVaildCandiNum;
- if (foundCandiNum < 1)
- {
- return false;
- }
-
- int pX = m_tempLibFast.getX();
- int pY = m_tempLibFast.getY();
- Pel* ref;
- int picStride = getStride();
- int iOffsetY, iOffsetX;
- Pel* refTarget;
- unsigned int uiHeight = uiPatchHeight - TMP_TEMPLATE_SIZE;
- unsigned int uiWidth = uiPatchWidth - TMP_TEMPLATE_SIZE;
-
- //the data center: we use the prediction block as the center now.
- //collect the candidates
- ref = getRefPicUsed();
- {
- iOffsetY = pY;
- iOffsetX = pX;
- refTarget = ref + iOffsetY * picStride + iOffsetX;
- for (unsigned int uiY = 0; uiY < uiHeight; uiY++)
- {
- for (unsigned int uiX = 0; uiX < uiWidth; uiX++)
- {
- piPred[uiX] = refTarget[uiX];
- }
- refTarget += picStride;
- piPred += uiStride;
- }
- }
- return bSucceedFlag;
-}
-
-#if JVET_W0069_TMP_BOUNDARY
-bool TrQuant::generateTM_DC_Prediction(Pel* piPred, unsigned int uiStride, unsigned int uiBlkWidth, unsigned int uiBlkHeight, int DC_Val)
-{
- bool bSucceedFlag = true;
- {
- for (unsigned int uiY = 0; uiY < uiBlkHeight; uiY++)
- {
- for (unsigned int uiX = 0; uiX < uiBlkWidth; uiX++)
- {
- piPred[uiX] = DC_Val;
- }
- piPred += uiStride;
- }
- }
- return bSucceedFlag;
-}
-#endif
-
-#if JVET_W0069_TMP_BOUNDARY
-int TrQuant::calcTemplateDiff(Pel* ref, unsigned int uiStride, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, int iMax, RefTemplateType TempType)
-#else
-int TrQuant::calcTemplateDiff(Pel* ref, unsigned int uiStride, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, int iMax)
-#endif
-{
- int iDiffSum = 0;
-#if JVET_W0069_TMP_BOUNDARY
- Pel* refPatchRow;
- if( TempType == L_SHAPE_TEMPLATE )
- {
- refPatchRow = ref - TMP_TEMPLATE_SIZE * uiStride - TMP_TEMPLATE_SIZE;
- }
- else if( TempType == LEFT_TEMPLATE )
- {
- refPatchRow = ref - TMP_TEMPLATE_SIZE;
- }
- else if( TempType == ABOVE_TEMPLATE )
- {
- refPatchRow = ref - TMP_TEMPLATE_SIZE * uiStride;
- }
-#else
- Pel* refPatchRow = ref - TMP_TEMPLATE_SIZE * uiStride - TMP_TEMPLATE_SIZE;
-#endif
- Pel* tarPatchRow;
-
-#if JVET_W0069_TMP_BOUNDARY
- if( TempType == L_SHAPE_TEMPLATE )
- {
-#endif
- // horizontal difference
- for( int iY = 0; iY < TMP_TEMPLATE_SIZE; iY++ )
- {
- tarPatchRow = tarPatch[iY];
- for( int iX = 0; iX < uiPatchWidth; iX++ )
- {
- iDiffSum += abs( refPatchRow[iX] - tarPatchRow[iX] );
- }
- if( iDiffSum > iMax ) //for speeding up
- {
- return iDiffSum;
- }
- refPatchRow += uiStride;
- }
-
- // vertical difference
- for( int iY = TMP_TEMPLATE_SIZE; iY < uiPatchHeight; iY++ )
- {
- tarPatchRow = tarPatch[iY];
- for( int iX = 0; iX < TMP_TEMPLATE_SIZE; iX++ )
- {
- iDiffSum += abs( refPatchRow[iX] - tarPatchRow[iX] );
- }
- if( iDiffSum > iMax ) //for speeding up
- {
- return iDiffSum;
- }
- refPatchRow += uiStride;
- }
-#if JVET_W0069_TMP_BOUNDARY
- }
- else if (TempType == ABOVE_TEMPLATE)
- {
- // top template difference
- for (int iY = 0; iY < TMP_TEMPLATE_SIZE; iY++)
- {
- tarPatchRow = tarPatch[iY];
- for (int iX = 0; iX < uiPatchWidth - TMP_TEMPLATE_SIZE; iX++)
- {
- iDiffSum += abs(refPatchRow[iX] - tarPatchRow[iX]);
- }
- if (iDiffSum > iMax) //for speeding up
- {
- return iDiffSum;
- }
- refPatchRow += uiStride;
- }
- }
- else if (TempType == LEFT_TEMPLATE)
- {
- // left template difference
- for (int iY = TMP_TEMPLATE_SIZE; iY < uiPatchHeight; iY++)
- {
- tarPatchRow = tarPatch[iY];
- for (int iX = 0; iX < TMP_TEMPLATE_SIZE; iX++)
- {
- iDiffSum += abs(refPatchRow[iX] - tarPatchRow[iX]);
- }
- if (iDiffSum > iMax) //for speeding up
- {
- return iDiffSum;
- }
- refPatchRow += uiStride;
- }
- }
-#endif
-
- return iDiffSum;
-}
-#endif
-
uint32_t TrQuant::getLFNSTIntraMode( int wideAngPredMode )
{
uint32_t intraMode;
diff --git a/source/Lib/CommonLib/TrQuant.h b/source/Lib/CommonLib/TrQuant.h
index ec9b42c35..3bc60d777 100644
--- a/source/Lib/CommonLib/TrQuant.h
+++ b/source/Lib/CommonLib/TrQuant.h
@@ -56,47 +56,6 @@ typedef void FwdTrans(const TCoeff*, TCoeff*, int, int, int, int);
typedef void InvTrans(const TCoeff*, TCoeff*, int, int, int, int, const TCoeff, const TCoeff);
-
-#if JVET_V0130_INTRA_TMP
-extern unsigned int g_uiDepth2Width[5];
-extern unsigned int g_uiDepth2MaxCandiNum[5];
-
-class TempLibFast
-{
-public:
- int m_pX; //offset X
- int m_pY; //offset Y
- int m_pXInteger; //offset X for integer pixel search
- int m_pYInteger; //offset Y for integer pixel search
- int m_pDiffInteger;
- int getXInteger() { return m_pXInteger; }
- int getYInteger() { return m_pYInteger; }
- int getDiffInteger() { return m_pDiffInteger; }
- short m_pIdInteger; //frame id
- short getIdInteger() { return m_pIdInteger; }
- int m_pDiff; //mse
- short m_pId; //frame id
-
-
- TempLibFast();
- ~TempLibFast();
- //void init();
- int getX() { return m_pX; }
- int getY() { return m_pY; }
- int getDiff() { return m_pDiff; }
- short getId() { return m_pId; }
- /*void initDiff(unsigned int uiPatchSize, int bitDepth);
- void initDiff(unsigned int uiPatchSize, int bitDepth, int iCandiNumber);*/
- void initTemplateDiff(unsigned int uiPatchWidth, unsigned int uiPatchHeight, unsigned int uiBlkWidth, unsigned int uiBlkHeight, int bitDepth);
- int m_diffMax;
- int getDiffMax() { return m_diffMax; }
-};
-
-
-typedef short TrainDataType;
-#endif
-
-
// ====================================================================================================================
// Class definition
// ====================================================================================================================
@@ -128,37 +87,6 @@ public:
#else
void fwdLfnstNxN( int* src, int* dst, const uint32_t mode, const uint32_t index, const uint32_t size, int zeroOutSize );
void invLfnstNxN( int* src, int* dst, const uint32_t mode, const uint32_t index, const uint32_t size, int zeroOutSize );
-#endif
-#if JVET_V0130_INTRA_TMP
-#if JVET_W0069_TMP_BOUNDARY
- int (*m_calcTemplateDiff)(Pel* ref, unsigned int uiStride, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, int iMax, RefTemplateType TempType);
- static int calcTemplateDiff(Pel* ref, unsigned int uiStride, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, int iMax, RefTemplateType TempType);
-#else
- int ( *m_calcTemplateDiff )(Pel* ref, unsigned int uiStride, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, int iMax);
- static int calcTemplateDiff(Pel* ref, unsigned int uiStride, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, int iMax);
-#endif
- Pel** getTargetPatch(unsigned int uiDepth) { return m_pppTarPatch[uiDepth]; }
- Pel* getRefPicUsed() { return m_refPicUsed; }
- void setRefPicUsed(Pel* ref) { m_refPicUsed = ref; }
- unsigned int getStride() { return m_uiPicStride; }
- void setStride(unsigned int uiPicStride) { m_uiPicStride = uiPicStride; }
-
-#if JVET_W0069_TMP_BOUNDARY
- void searchCandidateFromOnePicIntra(CodingUnit* pcCU, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, unsigned int setId, RefTemplateType TempType);
- void candidateSearchIntra(CodingUnit* pcCU, unsigned int uiBlkWidth, unsigned int uiBlkHeight, RefTemplateType TempType);
-#else
- void searchCandidateFromOnePicIntra(CodingUnit* pcCU, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, unsigned int setId);
- void candidateSearchIntra(CodingUnit* pcCU, unsigned int uiBlkWidth, unsigned int uiBlkHeight);
-#endif
- bool generateTMPrediction(Pel* piPred, unsigned int uiStride, unsigned int uiBlkWidth, unsigned int uiBlkHeight, int& foundCandiNum);
-#if JVET_W0069_TMP_BOUNDARY
- bool generateTM_DC_Prediction(Pel* piPred, unsigned int uiStride, unsigned int uiBlkWidth, unsigned int uiBlkHeight, int DC_Val);
-#endif
-#if JVET_W0069_TMP_BOUNDARY
- void getTargetTemplate(CodingUnit* pcCU, unsigned int uiBlkWidth, unsigned int uiBlkHeight, RefTemplateType TempType);
-#else
- void getTargetTemplate(CodingUnit* pcCU, unsigned int uiBlkWidth, unsigned int uiBlkHeight);
-#endif
#endif
uint32_t getLFNSTIntraMode( int wideAngPredMode );
@@ -217,15 +145,6 @@ public:
protected:
TCoeff m_tempCoeff[MAX_TB_SIZEY * MAX_TB_SIZEY];
-#if JVET_V0130_INTRA_TMP
- int m_uiPartLibSize;
- TempLibFast m_tempLibFast;
- Pel* m_refPicUsed;
- Picture* m_refPicBuf;
- unsigned int m_uiPicStride;
- unsigned int m_uiVaildCandiNum;
- Pel*** m_pppTarPatch;
-#endif
#if SIGN_PREDICTION
Pel m_tempSignPredResid[SIGN_PRED_MAX_BS * SIGN_PRED_MAX_BS * 2]{0};
Pel m_signPredTemplate[SIGN_PRED_FREQ_RANGE*SIGN_PRED_FREQ_RANGE*SIGN_PRED_MAX_BS*2];
@@ -305,7 +224,7 @@ private:
static void fastInverseTransform_SIMD( const TCoeff *coeff, TCoeff *block, int shift, int line, int iSkipLine, int iSkipLine2, const TCoeff outputMinimum, const TCoeff outputMaximum );
#endif
-#if ENABLE_SIMD_SIGN_PREDICTION || TRANSFORM_SIMD_OPT || ENABLE_SIMD_TMP
+#if ENABLE_SIMD_SIGN_PREDICTION || TRANSFORM_SIMD_OPT
#ifdef TARGET_SIMD_X86
void initTrQuantX86();
template <X86_VEXT vext>
diff --git a/source/Lib/CommonLib/TypeDef.h b/source/Lib/CommonLib/TypeDef.h
index 351e19286..38d9110f4 100644
--- a/source/Lib/CommonLib/TypeDef.h
+++ b/source/Lib/CommonLib/TypeDef.h
@@ -65,6 +65,8 @@
#define REMOVE_PCM 1 // Remove PCM related code for memory reduction and speedup
+#define JVET_X0049_BDMVR_SW_OPT 1 // JVET-X0049: software optimization for BDMVR (lossless)
+
// SIMD optimizations
#define MCIF_SIMD_NEW 1 // SIMD for interpolation
#define DIST_SSE_ENABLE 1 // Enable SIMD for SSE
@@ -109,15 +111,19 @@
#define INTRA_6TAP 1 // 6TapCubic + 6 TapGaussian + left side 4 tap weak filtering for intra.
#define SECONDARY_MPM 1 // Primary MPM and Secondary MPM: Add neighbouring modes into MPMs from positions AR, BL, AL, derived modes
#define ENABLE_DIMD 1 // Decoder side intra mode derivation
-#define JVET_V0087_DIMD_NO_ISP ENABLE_DIMD // JVET-V0087: disallow combination of DIMD and ISP
+#if ENABLE_DIMD
+#define JVET_V0087_DIMD_NO_ISP 1 // JVET-V0087: disallow combination of DIMD and ISP
+#define JVET_X0124_TMP_SIGNAL 1 // JVET-X0124: cleanup on signalling of intra template matching
+#endif
#define JVET_V0130_INTRA_TMP 1 // JVET-V0130: template matching prediction
#define JVET_W0069_TMP_BOUNDARY 1 // JVET-W0069: boundary handling for TMP
#define JVET_W0123_TIMD_FUSION 1 // JVET-W0123: Template based intra mode derivation and fusion
-#define JVET_X0148_TIMD_PDPC JVET_W0123_TIMD_FUSION // JVET-X0148: PDPC handling for TIMD
+#if JVET_W0123_TIMD_FUSION
+#define JVET_X0148_TIMD_PDPC 1 // JVET-X0148: PDPC handling for TIMD
+#endif
#if ENABLE_DIMD || JVET_W0123_TIMD_FUSION
#define JVET_X0149_TIMD_DIMD_LUT 1 // JVET-X0149: LUT-based derivation of DIMD and TIMD
#endif
-#define JVET_X0124_TMP_SIGNAL ENABLE_DIMD // JVET-X0124: cleanup on signalling of intra template matching
// Inter
#define CIIP_PDPC 1 // apply pdpc to megre prediction as a new CIIP mode (CIIP_PDPC) additional to CIIP mode
@@ -131,13 +137,11 @@
#define IF_12TAP 1 // 12-tap IF
#define ENABLE_OBMC 1 // Enable Overlapped Block Motion Compensation
-#define JVET_X0049_BDMVR_SW_OPT 1 // JVET-X0049: software optimization for BDMVR (lossless)
#if JVET_X0049_BDMVR_SW_OPT
#define JVET_X0049_ADAPT_DMVR 1 // JVET-X0049: Adaptive DMVR
#endif
#define JVET_X0056_DMVD_EARLY_TERMINATION 1 // JVET-X0056: Early termination for DMVR and TM
-
-#define JVET_X0083_BM_AMVP_MERGE_MODE 1
+#define JVET_X0083_BM_AMVP_MERGE_MODE 1 // JVET-X0083: AMVP-merge mode
// Inter template matching tools
#define ENABLE_INTER_TEMPLATE_MATCHING 1 // It controls whether template matching is enabled for inter prediction
diff --git a/source/Lib/CommonLib/x86/InitX86.cpp b/source/Lib/CommonLib/x86/InitX86.cpp
index e12809508..7766a4ce2 100644
--- a/source/Lib/CommonLib/x86/InitX86.cpp
+++ b/source/Lib/CommonLib/x86/InitX86.cpp
@@ -193,7 +193,7 @@ void IbcHashMap::initIbcHashMapX86()
}
#endif
-#if ENABLE_SIMD_SIGN_PREDICTION || TRANSFORM_SIMD_OPT || ENABLE_SIMD_TMP
+#if ENABLE_SIMD_SIGN_PREDICTION || TRANSFORM_SIMD_OPT
void TrQuant::initTrQuantX86()
{
auto vext = read_x86_extension_flags();
@@ -216,6 +216,29 @@ void TrQuant::initTrQuantX86()
}
#endif
+#if ENABLE_SIMD_TMP
+void IntraPrediction::initIntraX86()
+{
+ auto vext = read_x86_extension_flags();
+ switch( vext )
+ {
+ case AVX512:
+ case AVX2:
+ _initIntraX86<AVX2>();
+ break;
+ case AVX:
+ _initIntraX86<AVX>();
+ break;
+ case SSE42:
+ case SSE41:
+ _initIntraX86<SSE41>();
+ break;
+ default:
+ break;
+ }
+}
+#endif
+
#if ENABLE_SIMD_BILATERAL_FILTER || JVET_X0071_CHROMA_BILATERAL_FILTER_ENABLE_SIMD
void BilateralFilter::initBilateralFilterX86()
{
diff --git a/source/Lib/CommonLib/x86/IntraX86.h b/source/Lib/CommonLib/x86/IntraX86.h
new file mode 100644
index 000000000..e4f5617d4
--- /dev/null
+++ b/source/Lib/CommonLib/x86/IntraX86.h
@@ -0,0 +1,321 @@
+/* The copyright in this software is being made available under the BSD
+ * License, included below. This software may be subject to other third party
+ * and contributor rights, including patent rights, and no such rights are
+ * granted under this license.
+ *
+ * Copyright (c) 2010-2021, ITU/ISO/IEC
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * * Redistributions of source code must retain the above copyright notice,
+ * this list of conditions and the following disclaimer.
+ * * Redistributions in binary form must reproduce the above copyright notice,
+ * this list of conditions and the following disclaimer in the documentation
+ * and/or other materials provided with the distribution.
+ * * Neither the name of the ITU/ISO/IEC nor the names of its contributors may
+ * be used to endorse or promote products derived from this software without
+ * specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT 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.
+ */
+
+/**
+ * \file
+ * \brief Implementation of IbcHashMap class
+ */
+
+#include "CommonDefX86.h"
+#include "../IntraPrediction.h"
+
+#ifdef TARGET_SIMD_X86
+
+#include <nmmintrin.h>
+
+#if ENABLE_SIMD_TMP
+template< X86_VEXT vext >
+#if JVET_W0069_TMP_BOUNDARY
+int calcTemplateDiffSIMD(Pel* ref, unsigned int uiStride, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, int iMax, RefTemplateType tempType)
+#else
+int calcTemplateDiffSIMD( Pel* ref, unsigned int uiStride, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, int iMax )
+#endif
+{
+ int diffSum = 0;
+ int iY;
+#if JVET_W0069_TMP_BOUNDARY
+ Pel* refPatchRow;
+ if( tempType == L_SHAPE_TEMPLATE )
+ {
+ refPatchRow = ref - TMP_TEMPLATE_SIZE * uiStride - TMP_TEMPLATE_SIZE;
+ }
+ else if( tempType == LEFT_TEMPLATE )
+ {
+ refPatchRow = ref - TMP_TEMPLATE_SIZE;
+ }
+ else if( tempType == ABOVE_TEMPLATE )
+ {
+ refPatchRow = ref - TMP_TEMPLATE_SIZE * uiStride;
+ }
+#else
+ Pel* refPatchRow = ref - TMP_TEMPLATE_SIZE * uiStride - TMP_TEMPLATE_SIZE;
+#endif
+ Pel* tarPatchRow;
+ uint32_t uiSum;
+
+ // horizontal difference
+#if JVET_W0069_TMP_BOUNDARY
+ if( tempType == L_SHAPE_TEMPLATE )
+ {
+#endif
+ for( iY = 0; iY < TMP_TEMPLATE_SIZE; iY++ )
+ {
+ tarPatchRow = tarPatch[iY];
+ const short* pSrc1 = ( const short* ) tarPatchRow;
+ const short* pSrc2 = ( const short* ) refPatchRow;
+
+ // SIMD difference
+ //int iRows = uiPatchHeight;
+ int iCols = uiPatchWidth;
+ if( (iCols & 7) == 0 )
+ {
+ // Do with step of 8
+ __m128i vzero = _mm_setzero_si128();
+ __m128i vsum32 = vzero;
+ //for (int iY = 0; iY < iRows; iY += iSubStep)
+ {
+ __m128i vsum16 = vzero;
+ for( int iX = 0; iX < iCols; iX += 8 )
+ {
+ __m128i vsrc1 = _mm_loadu_si128( (const __m128i*)(&pSrc1[iX]) );
+ __m128i vsrc2 = _mm_lddqu_si128( (const __m128i*)(&pSrc2[iX]) );
+ vsum16 = _mm_add_epi16( vsum16, _mm_abs_epi16( _mm_sub_epi16( vsrc1, vsrc2 ) ) );
+ }
+ __m128i vsumtemp = _mm_add_epi32( _mm_unpacklo_epi16( vsum16, vzero ), _mm_unpackhi_epi16( vsum16, vzero ) );
+ vsum32 = _mm_add_epi32( vsum32, vsumtemp );
+ //pSrc1 += iStrideSrc1;
+ //pSrc2 += iStrideSrc2;
+ }
+ vsum32 = _mm_add_epi32( vsum32, _mm_shuffle_epi32( vsum32, 0x4e ) ); // 01001110
+ vsum32 = _mm_add_epi32( vsum32, _mm_shuffle_epi32( vsum32, 0xb1 ) ); // 10110001
+ uiSum = _mm_cvtsi128_si32( vsum32 );
+ }
+ else
+ {
+ // Do with step of 4
+ __m128i vzero = _mm_setzero_si128();
+ __m128i vsum32 = vzero;
+ //for (int iY = 0; iY < iRows; iY += iSubStep)
+ {
+ __m128i vsum16 = vzero;
+ for( int iX = 0; iX < iCols; iX += 4 )
+ {
+ __m128i vsrc1 = _mm_loadl_epi64( (const __m128i*) & pSrc1[iX] );
+ __m128i vsrc2 = _mm_loadl_epi64( (const __m128i*) & pSrc2[iX] );
+ vsum16 = _mm_add_epi16( vsum16, _mm_abs_epi16( _mm_sub_epi16( vsrc1, vsrc2 ) ) );
+ }
+ __m128i vsumtemp = _mm_add_epi32( _mm_unpacklo_epi16( vsum16, vzero ), _mm_unpackhi_epi16( vsum16, vzero ) );
+ vsum32 = _mm_add_epi32( vsum32, vsumtemp );
+ //pSrc1 += iStrideSrc1;
+ //pSrc2 += iStrideSrc2;
+ }
+ vsum32 = _mm_add_epi32( vsum32, _mm_shuffle_epi32( vsum32, 0x4e ) ); // 01001110
+ vsum32 = _mm_add_epi32( vsum32, _mm_shuffle_epi32( vsum32, 0xb1 ) ); // 10110001
+ uiSum = _mm_cvtsi128_si32( vsum32 );
+ }
+ diffSum += uiSum;
+
+ if( diffSum > iMax ) //for speeding up
+ {
+ return diffSum;
+ }
+ // update location
+ refPatchRow += uiStride;
+ }
+
+ // vertical difference
+ int iCols = TMP_TEMPLATE_SIZE;
+
+ for( iY = TMP_TEMPLATE_SIZE; iY < uiPatchHeight; iY++ )
+ {
+ tarPatchRow = tarPatch[iY];
+ const short* pSrc1 = ( const short* ) tarPatchRow;
+ const short* pSrc2 = ( const short* ) refPatchRow;
+
+ // SIMD difference
+
+ // Do with step of 4
+ __m128i vzero = _mm_setzero_si128();
+ __m128i vsum32 = vzero;
+ //for (int iY = 0; iY < iRows; iY += iSubStep)
+ {
+ __m128i vsum16 = vzero;
+ for( int iX = 0; iX < iCols; iX += 4 )
+ {
+ __m128i vsrc1 = _mm_loadl_epi64( (const __m128i*) & pSrc1[iX] );
+ __m128i vsrc2 = _mm_loadl_epi64( (const __m128i*) & pSrc2[iX] );
+ vsum16 = _mm_add_epi16( vsum16, _mm_abs_epi16( _mm_sub_epi16( vsrc1, vsrc2 ) ) );
+ }
+ __m128i vsumtemp = _mm_add_epi32( _mm_unpacklo_epi16( vsum16, vzero ), _mm_unpackhi_epi16( vsum16, vzero ) );
+ vsum32 = _mm_add_epi32( vsum32, vsumtemp );
+ //pSrc1 += iStrideSrc1;
+ //pSrc2 += iStrideSrc2;
+ }
+ vsum32 = _mm_add_epi32( vsum32, _mm_shuffle_epi32( vsum32, 0x4e ) ); // 01001110
+ vsum32 = _mm_add_epi32( vsum32, _mm_shuffle_epi32( vsum32, 0xb1 ) ); // 10110001
+ uiSum = _mm_cvtsi128_si32( vsum32 );
+
+ diffSum += uiSum;
+
+ if( diffSum > iMax ) //for speeding up
+ {
+ return diffSum;
+ }
+ // update location
+ refPatchRow += uiStride;
+ }
+#if JVET_W0069_TMP_BOUNDARY
+ }
+ else if (tempType == ABOVE_TEMPLATE)
+ {
+ // horizontal difference
+ for (iY = 0; iY < TMP_TEMPLATE_SIZE; iY++)
+ {
+ tarPatchRow = tarPatch[iY];
+ const short* pSrc1 = (const short*)tarPatchRow;
+ const short* pSrc2 = (const short*)refPatchRow;
+
+ // SIMD difference
+ //int iRows = uiPatchHeight;
+ int iCols = uiPatchWidth - TMP_TEMPLATE_SIZE;
+ if ((iCols & 7) == 0)
+ {
+ // Do with step of 8
+ __m128i vzero = _mm_setzero_si128();
+ __m128i vsum32 = vzero;
+ //for (int iY = 0; iY < iRows; iY += iSubStep)
+ {
+ __m128i vsum16 = vzero;
+ for (int iX = 0; iX < iCols; iX += 8)
+ {
+ __m128i vsrc1 = _mm_loadu_si128((const __m128i*)(&pSrc1[iX]));
+ __m128i vsrc2 = _mm_lddqu_si128((const __m128i*)(&pSrc2[iX]));
+ vsum16 = _mm_add_epi16(vsum16, _mm_abs_epi16(_mm_sub_epi16(vsrc1, vsrc2)));
+ }
+ __m128i vsumtemp = _mm_add_epi32(_mm_unpacklo_epi16(vsum16, vzero), _mm_unpackhi_epi16(vsum16, vzero));
+ vsum32 = _mm_add_epi32(vsum32, vsumtemp);
+ //pSrc1 += iStrideSrc1;
+ //pSrc2 += iStrideSrc2;
+ }
+ vsum32 = _mm_add_epi32(vsum32, _mm_shuffle_epi32(vsum32, 0x4e)); // 01001110
+ vsum32 = _mm_add_epi32(vsum32, _mm_shuffle_epi32(vsum32, 0xb1)); // 10110001
+ uiSum = _mm_cvtsi128_si32(vsum32);
+ }
+ else
+ {
+ // Do with step of 4
+ __m128i vzero = _mm_setzero_si128();
+ __m128i vsum32 = vzero;
+ //for (int iY = 0; iY < iRows; iY += iSubStep)
+ {
+ __m128i vsum16 = vzero;
+ for (int iX = 0; iX < iCols; iX += 4)
+ {
+ __m128i vsrc1 = _mm_loadl_epi64((const __m128i*) & pSrc1[iX]);
+ __m128i vsrc2 = _mm_loadl_epi64((const __m128i*) & pSrc2[iX]);
+ vsum16 = _mm_add_epi16(vsum16, _mm_abs_epi16(_mm_sub_epi16(vsrc1, vsrc2)));
+ }
+ __m128i vsumtemp = _mm_add_epi32(_mm_unpacklo_epi16(vsum16, vzero), _mm_unpackhi_epi16(vsum16, vzero));
+ vsum32 = _mm_add_epi32(vsum32, vsumtemp);
+ //pSrc1 += iStrideSrc1;
+ //pSrc2 += iStrideSrc2;
+ }
+ vsum32 = _mm_add_epi32(vsum32, _mm_shuffle_epi32(vsum32, 0x4e)); // 01001110
+ vsum32 = _mm_add_epi32(vsum32, _mm_shuffle_epi32(vsum32, 0xb1)); // 10110001
+ uiSum = _mm_cvtsi128_si32(vsum32);
+ }
+ diffSum += uiSum;
+
+ if (diffSum > iMax) //for speeding up
+ {
+ return diffSum;
+ }
+ // update location
+ refPatchRow += uiStride;
+ }
+
+
+ }
+ else if (tempType == LEFT_TEMPLATE)
+ {
+
+ // vertical difference
+ int iCols = TMP_TEMPLATE_SIZE;
+
+ for (iY = TMP_TEMPLATE_SIZE; iY < uiPatchHeight; iY++)
+ {
+ tarPatchRow = tarPatch[iY];
+ const short* pSrc1 = (const short*)tarPatchRow;
+ const short* pSrc2 = (const short*)refPatchRow;
+
+ // SIMD difference
+
+ // Do with step of 4
+ __m128i vzero = _mm_setzero_si128();
+ __m128i vsum32 = vzero;
+ //for (int iY = 0; iY < iRows; iY += iSubStep)
+ {
+ __m128i vsum16 = vzero;
+ for (int iX = 0; iX < iCols; iX += 4)
+ {
+ __m128i vsrc1 = _mm_loadl_epi64((const __m128i*) & pSrc1[iX]);
+ __m128i vsrc2 = _mm_loadl_epi64((const __m128i*) & pSrc2[iX]);
+ vsum16 = _mm_add_epi16(vsum16, _mm_abs_epi16(_mm_sub_epi16(vsrc1, vsrc2)));
+ }
+ __m128i vsumtemp = _mm_add_epi32(_mm_unpacklo_epi16(vsum16, vzero), _mm_unpackhi_epi16(vsum16, vzero));
+ vsum32 = _mm_add_epi32(vsum32, vsumtemp);
+ //pSrc1 += iStrideSrc1;
+ //pSrc2 += iStrideSrc2;
+ }
+ vsum32 = _mm_add_epi32(vsum32, _mm_shuffle_epi32(vsum32, 0x4e)); // 01001110
+ vsum32 = _mm_add_epi32(vsum32, _mm_shuffle_epi32(vsum32, 0xb1)); // 10110001
+ uiSum = _mm_cvtsi128_si32(vsum32);
+
+ diffSum += uiSum;
+
+ if (diffSum > iMax) //for speeding up
+ {
+ return diffSum;
+ }
+ // update location
+ refPatchRow += uiStride;
+ }
+ }
+#endif
+
+ return diffSum;
+}
+
+template <X86_VEXT vext>
+void IntraPrediction::_initIntraX86()
+{
+#if ENABLE_SIMD_TMP
+ m_calcTemplateDiff = calcTemplateDiffSIMD<vext>;
+#endif
+}
+
+template void IntraPrediction::_initIntraX86<SIMDX86>();
+#endif
+
+#endif //#ifdef TARGET_SIMD_X86
+//! \}
diff --git a/source/Lib/CommonLib/x86/TrQuantX86.h b/source/Lib/CommonLib/x86/TrQuantX86.h
index ac45c865e..99c806a31 100644
--- a/source/Lib/CommonLib/x86/TrQuantX86.h
+++ b/source/Lib/CommonLib/x86/TrQuantX86.h
@@ -410,271 +410,7 @@ uint32_t computeSAD_SIMD( const Pel* ref, const Pel* cur, const int size )
}
#endif
-#if ENABLE_SIMD_TMP
-template< X86_VEXT vext >
-#if JVET_W0069_TMP_BOUNDARY
-int calcTemplateDiffSIMD(Pel* ref, unsigned int uiStride, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, int iMax, RefTemplateType TempType)
-#else
-int calcTemplateDiffSIMD( Pel* ref, unsigned int uiStride, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, int iMax )
-#endif
-{
- int iDiffSum = 0;
- int iY;
-#if JVET_W0069_TMP_BOUNDARY
- Pel* refPatchRow;
- if( TempType == L_SHAPE_TEMPLATE )
- {
- refPatchRow = ref - TMP_TEMPLATE_SIZE * uiStride - TMP_TEMPLATE_SIZE;
- }
- else if( TempType == LEFT_TEMPLATE )
- {
- refPatchRow = ref - TMP_TEMPLATE_SIZE;
- }
- else if( TempType == ABOVE_TEMPLATE )
- {
- refPatchRow = ref - TMP_TEMPLATE_SIZE * uiStride;
- }
-#else
- Pel* refPatchRow = ref - TMP_TEMPLATE_SIZE * uiStride - TMP_TEMPLATE_SIZE;
-#endif
- Pel* tarPatchRow;
- uint32_t uiSum;
-
- // horizontal difference
-#if JVET_W0069_TMP_BOUNDARY
- if (TempType == L_SHAPE_TEMPLATE)
- {
-#endif
- for( iY = 0; iY < TMP_TEMPLATE_SIZE; iY++ )
- {
- tarPatchRow = tarPatch[iY];
- const short* pSrc1 = ( const short* ) tarPatchRow;
- const short* pSrc2 = ( const short* ) refPatchRow;
-
- // SIMD difference
- //int iRows = uiPatchHeight;
- int iCols = uiPatchWidth;
- if( (iCols & 7) == 0 )
- {
- // Do with step of 8
- __m128i vzero = _mm_setzero_si128();
- __m128i vsum32 = vzero;
- //for (int iY = 0; iY < iRows; iY += iSubStep)
- {
- __m128i vsum16 = vzero;
- for( int iX = 0; iX < iCols; iX += 8 )
- {
- __m128i vsrc1 = _mm_loadu_si128( (const __m128i*)(&pSrc1[iX]) );
- __m128i vsrc2 = _mm_lddqu_si128( (const __m128i*)(&pSrc2[iX]) );
- vsum16 = _mm_add_epi16( vsum16, _mm_abs_epi16( _mm_sub_epi16( vsrc1, vsrc2 ) ) );
- }
- __m128i vsumtemp = _mm_add_epi32( _mm_unpacklo_epi16( vsum16, vzero ), _mm_unpackhi_epi16( vsum16, vzero ) );
- vsum32 = _mm_add_epi32( vsum32, vsumtemp );
- //pSrc1 += iStrideSrc1;
- //pSrc2 += iStrideSrc2;
- }
- vsum32 = _mm_add_epi32( vsum32, _mm_shuffle_epi32( vsum32, 0x4e ) ); // 01001110
- vsum32 = _mm_add_epi32( vsum32, _mm_shuffle_epi32( vsum32, 0xb1 ) ); // 10110001
- uiSum = _mm_cvtsi128_si32( vsum32 );
- }
- else
- {
- // Do with step of 4
- __m128i vzero = _mm_setzero_si128();
- __m128i vsum32 = vzero;
- //for (int iY = 0; iY < iRows; iY += iSubStep)
- {
- __m128i vsum16 = vzero;
- for( int iX = 0; iX < iCols; iX += 4 )
- {
- __m128i vsrc1 = _mm_loadl_epi64( (const __m128i*) & pSrc1[iX] );
- __m128i vsrc2 = _mm_loadl_epi64( (const __m128i*) & pSrc2[iX] );
- vsum16 = _mm_add_epi16( vsum16, _mm_abs_epi16( _mm_sub_epi16( vsrc1, vsrc2 ) ) );
- }
- __m128i vsumtemp = _mm_add_epi32( _mm_unpacklo_epi16( vsum16, vzero ), _mm_unpackhi_epi16( vsum16, vzero ) );
- vsum32 = _mm_add_epi32( vsum32, vsumtemp );
- //pSrc1 += iStrideSrc1;
- //pSrc2 += iStrideSrc2;
- }
- vsum32 = _mm_add_epi32( vsum32, _mm_shuffle_epi32( vsum32, 0x4e ) ); // 01001110
- vsum32 = _mm_add_epi32( vsum32, _mm_shuffle_epi32( vsum32, 0xb1 ) ); // 10110001
- uiSum = _mm_cvtsi128_si32( vsum32 );
- }
- iDiffSum += uiSum;
-
- if( iDiffSum > iMax ) //for speeding up
- {
- return iDiffSum;
- }
- // update location
- refPatchRow += uiStride;
- }
-
- // vertical difference
- int iCols = TMP_TEMPLATE_SIZE;
-
- for( iY = TMP_TEMPLATE_SIZE; iY < uiPatchHeight; iY++ )
- {
- tarPatchRow = tarPatch[iY];
- const short* pSrc1 = ( const short* ) tarPatchRow;
- const short* pSrc2 = ( const short* ) refPatchRow;
-
- // SIMD difference
-
- // Do with step of 4
- __m128i vzero = _mm_setzero_si128();
- __m128i vsum32 = vzero;
- //for (int iY = 0; iY < iRows; iY += iSubStep)
- {
- __m128i vsum16 = vzero;
- for( int iX = 0; iX < iCols; iX += 4 )
- {
- __m128i vsrc1 = _mm_loadl_epi64( (const __m128i*) & pSrc1[iX] );
- __m128i vsrc2 = _mm_loadl_epi64( (const __m128i*) & pSrc2[iX] );
- vsum16 = _mm_add_epi16( vsum16, _mm_abs_epi16( _mm_sub_epi16( vsrc1, vsrc2 ) ) );
- }
- __m128i vsumtemp = _mm_add_epi32( _mm_unpacklo_epi16( vsum16, vzero ), _mm_unpackhi_epi16( vsum16, vzero ) );
- vsum32 = _mm_add_epi32( vsum32, vsumtemp );
- //pSrc1 += iStrideSrc1;
- //pSrc2 += iStrideSrc2;
- }
- vsum32 = _mm_add_epi32( vsum32, _mm_shuffle_epi32( vsum32, 0x4e ) ); // 01001110
- vsum32 = _mm_add_epi32( vsum32, _mm_shuffle_epi32( vsum32, 0xb1 ) ); // 10110001
- uiSum = _mm_cvtsi128_si32( vsum32 );
-
- iDiffSum += uiSum;
-
- if( iDiffSum > iMax ) //for speeding up
- {
- return iDiffSum;
- }
- // update location
- refPatchRow += uiStride;
- }
-#if JVET_W0069_TMP_BOUNDARY
- }
- else if (TempType == ABOVE_TEMPLATE)
- {
- // horizontal difference
- for (iY = 0; iY < TMP_TEMPLATE_SIZE; iY++)
- {
- tarPatchRow = tarPatch[iY];
- const short* pSrc1 = (const short*)tarPatchRow;
- const short* pSrc2 = (const short*)refPatchRow;
-
- // SIMD difference
- //int iRows = uiPatchHeight;
- int iCols = uiPatchWidth - TMP_TEMPLATE_SIZE;
- if ((iCols & 7) == 0)
- {
- // Do with step of 8
- __m128i vzero = _mm_setzero_si128();
- __m128i vsum32 = vzero;
- //for (int iY = 0; iY < iRows; iY += iSubStep)
- {
- __m128i vsum16 = vzero;
- for (int iX = 0; iX < iCols; iX += 8)
- {
- __m128i vsrc1 = _mm_loadu_si128((const __m128i*)(&pSrc1[iX]));
- __m128i vsrc2 = _mm_lddqu_si128((const __m128i*)(&pSrc2[iX]));
- vsum16 = _mm_add_epi16(vsum16, _mm_abs_epi16(_mm_sub_epi16(vsrc1, vsrc2)));
- }
- __m128i vsumtemp = _mm_add_epi32(_mm_unpacklo_epi16(vsum16, vzero), _mm_unpackhi_epi16(vsum16, vzero));
- vsum32 = _mm_add_epi32(vsum32, vsumtemp);
- //pSrc1 += iStrideSrc1;
- //pSrc2 += iStrideSrc2;
- }
- vsum32 = _mm_add_epi32(vsum32, _mm_shuffle_epi32(vsum32, 0x4e)); // 01001110
- vsum32 = _mm_add_epi32(vsum32, _mm_shuffle_epi32(vsum32, 0xb1)); // 10110001
- uiSum = _mm_cvtsi128_si32(vsum32);
- }
- else
- {
- // Do with step of 4
- __m128i vzero = _mm_setzero_si128();
- __m128i vsum32 = vzero;
- //for (int iY = 0; iY < iRows; iY += iSubStep)
- {
- __m128i vsum16 = vzero;
- for (int iX = 0; iX < iCols; iX += 4)
- {
- __m128i vsrc1 = _mm_loadl_epi64((const __m128i*) & pSrc1[iX]);
- __m128i vsrc2 = _mm_loadl_epi64((const __m128i*) & pSrc2[iX]);
- vsum16 = _mm_add_epi16(vsum16, _mm_abs_epi16(_mm_sub_epi16(vsrc1, vsrc2)));
- }
- __m128i vsumtemp = _mm_add_epi32(_mm_unpacklo_epi16(vsum16, vzero), _mm_unpackhi_epi16(vsum16, vzero));
- vsum32 = _mm_add_epi32(vsum32, vsumtemp);
- //pSrc1 += iStrideSrc1;
- //pSrc2 += iStrideSrc2;
- }
- vsum32 = _mm_add_epi32(vsum32, _mm_shuffle_epi32(vsum32, 0x4e)); // 01001110
- vsum32 = _mm_add_epi32(vsum32, _mm_shuffle_epi32(vsum32, 0xb1)); // 10110001
- uiSum = _mm_cvtsi128_si32(vsum32);
- }
- iDiffSum += uiSum;
-
- if (iDiffSum > iMax) //for speeding up
- {
- return iDiffSum;
- }
- // update location
- refPatchRow += uiStride;
- }
-
-
- }
- else if (TempType == LEFT_TEMPLATE)
- {
-
- // vertical difference
- int iCols = TMP_TEMPLATE_SIZE;
-
- for (iY = TMP_TEMPLATE_SIZE; iY < uiPatchHeight; iY++)
- {
- tarPatchRow = tarPatch[iY];
- const short* pSrc1 = (const short*)tarPatchRow;
- const short* pSrc2 = (const short*)refPatchRow;
-
- // SIMD difference
-
- // Do with step of 4
- __m128i vzero = _mm_setzero_si128();
- __m128i vsum32 = vzero;
- //for (int iY = 0; iY < iRows; iY += iSubStep)
- {
- __m128i vsum16 = vzero;
- for (int iX = 0; iX < iCols; iX += 4)
- {
- __m128i vsrc1 = _mm_loadl_epi64((const __m128i*) & pSrc1[iX]);
- __m128i vsrc2 = _mm_loadl_epi64((const __m128i*) & pSrc2[iX]);
- vsum16 = _mm_add_epi16(vsum16, _mm_abs_epi16(_mm_sub_epi16(vsrc1, vsrc2)));
- }
- __m128i vsumtemp = _mm_add_epi32(_mm_unpacklo_epi16(vsum16, vzero), _mm_unpackhi_epi16(vsum16, vzero));
- vsum32 = _mm_add_epi32(vsum32, vsumtemp);
- //pSrc1 += iStrideSrc1;
- //pSrc2 += iStrideSrc2;
- }
- vsum32 = _mm_add_epi32(vsum32, _mm_shuffle_epi32(vsum32, 0x4e)); // 01001110
- vsum32 = _mm_add_epi32(vsum32, _mm_shuffle_epi32(vsum32, 0xb1)); // 10110001
- uiSum = _mm_cvtsi128_si32(vsum32);
-
- iDiffSum += uiSum;
-
- if (iDiffSum > iMax) //for speeding up
- {
- return iDiffSum;
- }
- // update location
- refPatchRow += uiStride;
- }
- }
-#endif
-
- return iDiffSum;
-}
-#endif
-
-#if ENABLE_SIMD_SIGN_PREDICTION || TRANSFORM_SIMD_OPT || ENABLE_SIMD_TMP
+#if ENABLE_SIMD_SIGN_PREDICTION || TRANSFORM_SIMD_OPT
template <X86_VEXT vext>
void TrQuant::_initTrQuantX86()
{
@@ -897,10 +633,6 @@ void TrQuant::_initTrQuantX86()
fastInvTrans[2][5] = fastInverseTransform_SIMD<DST7, 64>;
#endif
#endif
-
-#if ENABLE_SIMD_TMP
- m_calcTemplateDiff = calcTemplateDiffSIMD<vext>;
-#endif
}
template void TrQuant::_initTrQuantX86<SIMDX86>();
diff --git a/source/Lib/CommonLib/x86/avx/Intra_avx.cpp b/source/Lib/CommonLib/x86/avx/Intra_avx.cpp
new file mode 100644
index 000000000..59a692f72
--- /dev/null
+++ b/source/Lib/CommonLib/x86/avx/Intra_avx.cpp
@@ -0,0 +1 @@
+#include "../IntraX86.h"
diff --git a/source/Lib/CommonLib/x86/avx2/Intra_avx2.cpp b/source/Lib/CommonLib/x86/avx2/Intra_avx2.cpp
new file mode 100644
index 000000000..59a692f72
--- /dev/null
+++ b/source/Lib/CommonLib/x86/avx2/Intra_avx2.cpp
@@ -0,0 +1 @@
+#include "../IntraX86.h"
diff --git a/source/Lib/CommonLib/x86/sse41/Intra_sse41.cpp b/source/Lib/CommonLib/x86/sse41/Intra_sse41.cpp
new file mode 100644
index 000000000..59a692f72
--- /dev/null
+++ b/source/Lib/CommonLib/x86/sse41/Intra_sse41.cpp
@@ -0,0 +1 @@
+#include "../IntraX86.h"
diff --git a/source/Lib/CommonLib/x86/sse42/Intra_sse42.cpp b/source/Lib/CommonLib/x86/sse42/Intra_sse42.cpp
new file mode 100644
index 000000000..59a692f72
--- /dev/null
+++ b/source/Lib/CommonLib/x86/sse42/Intra_sse42.cpp
@@ -0,0 +1 @@
+#include "../IntraX86.h"
diff --git a/source/Lib/DecoderLib/DecCu.cpp b/source/Lib/DecoderLib/DecCu.cpp
index 0e196a07f..6d3fb02d4 100644
--- a/source/Lib/DecoderLib/DecCu.cpp
+++ b/source/Lib/DecoderLib/DecCu.cpp
@@ -385,22 +385,23 @@ void DecCu::xIntraRecBlk( TransformUnit& tu, const ComponentID compID )
{
int foundCandiNum;
#if JVET_W0069_TMP_BOUNDARY
- RefTemplateType TempType = m_pcIntraPred->GetRefTemplateType(*(tu.cu), tu.cu->blocks[COMPONENT_Y]);
- if (TempType != NO_TEMPLATE)
+ RefTemplateType tempType = m_pcIntraPred->getRefTemplateType(*(tu.cu), tu.cu->blocks[COMPONENT_Y]);
+
+ if( tempType != NO_TEMPLATE )
{
- m_pcTrQuant->getTargetTemplate(tu.cu, pu.lwidth(), pu.lheight(), TempType);
- m_pcTrQuant->candidateSearchIntra(tu.cu, pu.lwidth(), pu.lheight(), TempType);
- m_pcTrQuant->generateTMPrediction(piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), foundCandiNum);
+ m_pcIntraPred->getTargetTemplate(tu.cu, pu.lwidth(), pu.lheight(), tempType);
+ m_pcIntraPred->candidateSearchIntra(tu.cu, pu.lwidth(), pu.lheight(), tempType);
+ m_pcIntraPred->generateTMPrediction(piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), foundCandiNum);
}
else
{
foundCandiNum = 1;
- m_pcTrQuant->generateTM_DC_Prediction(piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), 1 << (tu.cu->cs->sps->getBitDepth(CHANNEL_TYPE_LUMA) - 1));
+ m_pcIntraPred->generateTmDcPrediction(piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), 1 << (tu.cu->cs->sps->getBitDepth(CHANNEL_TYPE_LUMA) - 1));
}
#else
- m_pcTrQuant->getTargetTemplate(tu.cu, pu.lwidth(), pu.lheight());
- m_pcTrQuant->candidateSearchIntra(tu.cu, pu.lwidth(), pu.lheight());
- m_pcTrQuant->generateTMPrediction(piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), foundCandiNum);
+ m_pcIntraPred->getTargetTemplate(tu.cu, pu.lwidth(), pu.lheight());
+ m_pcIntraPred->candidateSearchIntra(tu.cu, pu.lwidth(), pu.lheight());
+ m_pcIntraPred->generateTMPrediction(piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), foundCandiNum);
#endif
assert(foundCandiNum >= 1);
}
@@ -605,9 +606,9 @@ void DecCu::xIntraRecACTBlk(TransformUnit& tu)
{
int foundCandiNum;
const unsigned int uiStride = cs.picture->getRecoBuf(COMPONENT_Y).stride;
- m_pcTrQuant->getTargetTemplate(tu.cu, pu.lwidth(), pu.lheight());
- m_pcTrQuant->candidateSearchIntra(tu.cu, pu.lwidth(), pu.lheight());
- m_pcTrQuant->generateTMPrediction(piPred.buf, uiStride, pu.lwidth(), pu.lheight(), foundCandiNum);
+ m_pcIntraPred->getTargetTemplate(tu.cu, pu.lwidth(), pu.lheight());
+ m_pcIntraPred->candidateSearchIntra(tu.cu, pu.lwidth(), pu.lheight());
+ m_pcIntraPred->generateTMPrediction(piPred.buf, uiStride, pu.lwidth(), pu.lheight(), foundCandiNum);
}
else if (PU::isMIP(pu, chType))
#else
diff --git a/source/Lib/EncoderLib/IntraSearch.cpp b/source/Lib/EncoderLib/IntraSearch.cpp
index e97365623..c4edea054 100644
--- a/source/Lib/EncoderLib/IntraSearch.cpp
+++ b/source/Lib/EncoderLib/IntraSearch.cpp
@@ -892,27 +892,27 @@ bool IntraSearch::estIntraPredLumaQT(CodingUnit &cu, Partitioner &partitioner, c
CodingUnit cu_cpy = cu;
#if JVET_W0069_TMP_BOUNDARY
- RefTemplateType TemplateType = GetRefTemplateType( cu_cpy, cu_cpy.blocks[COMPONENT_Y] );
- if( TemplateType != NO_TEMPLATE )
+ RefTemplateType templateType = getRefTemplateType( cu_cpy, cu_cpy.blocks[COMPONENT_Y] );
+ if( templateType != NO_TEMPLATE )
#else
if( isRefTemplateAvailable( cu_cpy, cu_cpy.blocks[COMPONENT_Y] ) )
#endif
{
#if JVET_W0069_TMP_BOUNDARY
- m_pcTrQuant->getTargetTemplate( &cu_cpy, pu.lwidth(), pu.lheight(), TemplateType );
- m_pcTrQuant->candidateSearchIntra( &cu_cpy, pu.lwidth(), pu.lheight(), TemplateType );
- bsuccessfull = m_pcTrQuant->generateTMPrediction( piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), foundCandiNum );
+ getTargetTemplate( &cu_cpy, pu.lwidth(), pu.lheight(), templateType );
+ candidateSearchIntra( &cu_cpy, pu.lwidth(), pu.lheight(), templateType );
+ bsuccessfull = generateTMPrediction( piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), foundCandiNum );
#else
- m_pcTrQuant->getTargetTemplate( &cu_cpy, pu.lwidth(), pu.lheight() );
- m_pcTrQuant->candidateSearchIntra( &cu_cpy, pu.lwidth(), pu.lheight() );
- bsuccessfull = m_pcTrQuant->generateTMPrediction( piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), foundCandiNum );
+ getTargetTemplate( &cu_cpy, pu.lwidth(), pu.lheight() );
+ candidateSearchIntra( &cu_cpy, pu.lwidth(), pu.lheight() );
+ bsuccessfull = generateTMPrediction( piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), foundCandiNum );
#endif
}
#if JVET_W0069_TMP_BOUNDARY
else
{
foundCandiNum = 1;
- bsuccessfull = m_pcTrQuant->generateTM_DC_Prediction( piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), 1 << (cu_cpy.cs->sps->getBitDepth( CHANNEL_TYPE_LUMA ) - 1) );
+ bsuccessfull = generateTmDcPrediction( piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), 1 << (cu_cpy.cs->sps->getBitDepth( CHANNEL_TYPE_LUMA ) - 1) );
}
#endif
if( bsuccessfull && foundCandiNum >= 1 )
@@ -3679,22 +3679,22 @@ void IntraSearch::xIntraCodingTUBlock(TransformUnit &tu, const ComponentID &comp
{
int foundCandiNum;
#if JVET_W0069_TMP_BOUNDARY
- RefTemplateType TempType = GetRefTemplateType( *(tu.cu), tu.cu->blocks[COMPONENT_Y] );
- if( TempType != NO_TEMPLATE )
+ RefTemplateType tempType = getRefTemplateType( *(tu.cu), tu.cu->blocks[COMPONENT_Y] );
+ if( tempType != NO_TEMPLATE )
{
- m_pcTrQuant->getTargetTemplate( tu.cu, pu.lwidth(), pu.lheight(), TempType );
- m_pcTrQuant->candidateSearchIntra( tu.cu, pu.lwidth(), pu.lheight(), TempType );
- m_pcTrQuant->generateTMPrediction( piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), foundCandiNum );
+ getTargetTemplate( tu.cu, pu.lwidth(), pu.lheight(), tempType );
+ candidateSearchIntra( tu.cu, pu.lwidth(), pu.lheight(), tempType );
+ generateTMPrediction( piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), foundCandiNum );
}
else
{
foundCandiNum = 1;
- m_pcTrQuant->generateTM_DC_Prediction( piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), 1 << (tu.cu->cs->sps->getBitDepth( CHANNEL_TYPE_LUMA ) - 1) );
+ generateTmDcPrediction( piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), 1 << (tu.cu->cs->sps->getBitDepth( CHANNEL_TYPE_LUMA ) - 1) );
}
#else
- m_pcTrQuant->getTargetTemplate( tu.cu, pu.lwidth(), pu.lheight() );
- m_pcTrQuant->candidateSearchIntra( tu.cu, pu.lwidth(), pu.lheight() );
- m_pcTrQuant->generateTMPrediction( piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), foundCandiNum );
+ getTargetTemplate( tu.cu, pu.lwidth(), pu.lheight() );
+ candidateSearchIntra( tu.cu, pu.lwidth(), pu.lheight() );
+ generateTMPrediction( piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), foundCandiNum );
#endif
CHECK( foundCandiNum < 1, "" );
}
@@ -5094,9 +5094,9 @@ bool IntraSearch::xRecurIntraCodingACTQT(CodingStructure &cs, Partitioner &parti
if( PU::isTmp( pu, chType ) )
{
int foundCandiNum;
- m_pcTrQuant->getTargetTemplate( pu.cu, pu.lwidth(), pu.lheight() );
- m_pcTrQuant->candidateSearchIntra( pu.cu, pu.lwidth(), pu.lheight() );
- m_pcTrQuant->generateTMPrediction( piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), foundCandiNum );
+ getTargetTemplate( pu.cu, pu.lwidth(), pu.lheight() );
+ candidateSearchIntra( pu.cu, pu.lwidth(), pu.lheight() );
+ generateTMPrediction( piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), foundCandiNum );
CHECK( foundCandiNum < 1, "" );
}
--
GitLab