diff --git a/source/Lib/CommonLib/Buffer.h b/source/Lib/CommonLib/Buffer.h
index e714d2869888ebeb6cf2a80c2cb0288acdac23a8..216f8b57598cf0b37c6d87620ec93d4ba107fe0d 100644
--- a/source/Lib/CommonLib/Buffer.h
+++ b/source/Lib/CommonLib/Buffer.h
@@ -686,9 +686,17 @@ struct UnitBuf
const AreaBuf<T>& Cr() const { return bufs[2]; }
void fill ( const T &val );
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ void copyFrom ( const UnitBuf<const T> &other, const bool lumaOnly = false, const bool chromaOnly = false );
+#else
void copyFrom ( const UnitBuf<const T> &other );
+#endif
void reconstruct ( const UnitBuf<const T> &pred, const UnitBuf<const T> &resi, const ClpRngs& clpRngs );
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ void copyClip ( const UnitBuf<const T> &src, const ClpRngs& clpRngs, const bool lumaOnly = false, const bool chromaOnly = false );
+#else
void copyClip ( const UnitBuf<const T> &src, const ClpRngs& clpRngs );
+#endif
void subtract ( const UnitBuf<const T> &other );
void addWeightedAvg ( const UnitBuf<const T> &other1, const UnitBuf<const T> &other2, const ClpRngs& clpRngs, const uint8_t gbiIdx = GBI_DEFAULT, const bool chromaOnly = false, const bool lumaOnly = false);
void addAvg ( const UnitBuf<const T> &other1, const UnitBuf<const T> &other2, const ClpRngs& clpRngs, const bool chromaOnly = false, const bool lumaOnly = false);
@@ -718,11 +726,22 @@ void UnitBuf<T>::fill( const T &val )
}
template<typename T>
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+void UnitBuf<T>::copyFrom(const UnitBuf<const T> &other, const bool lumaOnly, const bool chromaOnly )
+#else
void UnitBuf<T>::copyFrom( const UnitBuf<const T> &other )
+#endif
{
CHECK( chromaFormat != other.chromaFormat, "Incompatible formats" );
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ CHECK( lumaOnly && chromaOnly, "Not allowed to have both lumaOnly and chromaOnly selected" );
+ const size_t compStart = chromaOnly ? 1 : 0;
+ const size_t compEnd = lumaOnly ? 1 : (unsigned) bufs.size();
+ for( size_t i = compStart; i < compEnd; i++ )
+#else
for( unsigned i = 0; i < bufs.size(); i++ )
+#endif
{
bufs[i].copyFrom( other.bufs[i] );
}
@@ -742,11 +761,22 @@ void UnitBuf<T>::subtract( const UnitBuf<const T> &other )
}
template<typename T>
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+void UnitBuf<T>::copyClip(const UnitBuf<const T> &src, const ClpRngs &clpRngs, const bool lumaOnly, const bool chromaOnly )
+#else
void UnitBuf<T>::copyClip(const UnitBuf<const T> &src, const ClpRngs& clpRngs)
+#endif
{
CHECK( chromaFormat != src.chromaFormat, "Incompatible formats" );
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ CHECK( lumaOnly && chromaOnly, "Not allowed to have both lumaOnly and chromaOnly selected" );
+ const size_t compStart = chromaOnly ? 1 : 0;
+ const size_t compEnd = lumaOnly ? 1 : bufs.size();
+ for( size_t i = compStart; i < compEnd; i++ )
+#else
for( unsigned i = 0; i < bufs.size(); i++ )
+#endif
{
bufs[i].copyClip( src.bufs[i], clpRngs.comp[i] );
}
diff --git a/source/Lib/CommonLib/InterPrediction.cpp b/source/Lib/CommonLib/InterPrediction.cpp
index ef51d6960c16156e393fe28174a96d9b31f482c6..2cb8fc51a357c061b40cfbab162e186fa969bfc5 100644
--- a/source/Lib/CommonLib/InterPrediction.cpp
+++ b/source/Lib/CommonLib/InterPrediction.cpp
@@ -264,7 +264,11 @@ bool InterPrediction::xCheckIdenticalMotion( const PredictionUnit &pu )
return false;
}
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+void InterPrediction::xSubPuMC( PredictionUnit& pu, PelUnitBuf& predBuf, const RefPicList &eRefPicList /*= REF_PIC_LIST_X*/, const bool luma /*= true*/, const bool chroma /*= true*/)
+#else
void InterPrediction::xSubPuMC( PredictionUnit& pu, PelUnitBuf& predBuf, const RefPicList &eRefPicList /*= REF_PIC_LIST_X*/ )
+#endif
{
// compute the location of the current PU
@@ -331,7 +335,11 @@ void InterPrediction::xSubPuMC( PredictionUnit& pu, PelUnitBuf& predBuf, const R
PelUnitBuf subPredBuf = predBuf.subBuf(UnitAreaRelative(pu, subPu));
subPu.mmvdEncOptMode = 0;
subPu.mvRefine = false;
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ motionCompensation(subPu, subPredBuf, eRefPicList, luma, chroma);
+#else
motionCompensation(subPu, subPredBuf, eRefPicList);
+#endif
secDim = later - secStep;
}
}
@@ -433,7 +441,12 @@ void InterPrediction::xPredInterUni(const PredictionUnit& pu, const RefPicList&
{
CHECK( bioApplied, "BIO is not allowed with affine" );
m_iRefListIdx = eRefPicList;
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ bool genChromaMv = (!luma && chroma && compID == COMPONENT_Cb);
+ xPredAffineBlk( compID, pu, pu.cu->slice->getRefPic( eRefPicList, iRefIdx )->unscaledPic, mv, pcYuvPred, bi, pu.cu->slice->clpRng( compID ), genChromaMv, pu.cu->slice->getScalingRatio( eRefPicList, iRefIdx ));
+#else
xPredAffineBlk( compID, pu, pu.cu->slice->getRefPic( eRefPicList, iRefIdx )->unscaledPic, mv, pcYuvPred, bi, pu.cu->slice->clpRng( compID ), pu.cu->slice->getScalingRatio( eRefPicList, iRefIdx ));
+#endif
}
else
{
@@ -452,7 +465,11 @@ void InterPrediction::xPredInterUni(const PredictionUnit& pu, const RefPicList&
}
}
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+void InterPrediction::xPredInterBi(PredictionUnit &pu, PelUnitBuf &pcYuvPred, const bool luma, const bool chroma, PelUnitBuf *yuvPredTmp /*= NULL*/)
+#else
void InterPrediction::xPredInterBi(PredictionUnit& pu, PelUnitBuf &pcYuvPred, PelUnitBuf* yuvPredTmp /*= NULL*/)
+#endif
{
const PPS &pps = *pu.cs->pps;
const Slice &slice = *pu.cs->slice;
@@ -537,12 +554,20 @@ void InterPrediction::xPredInterBi(PredictionUnit& pu, PelUnitBuf &pcYuvPred, Pe
if (dmvrApplied)
{
if (yuvPredTmp)
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ xPredInterUni(pu, eRefPicList, pcMbBuf, true, false, luma, chroma);
+#else
xPredInterUni(pu, eRefPicList, pcMbBuf, true, false, true, true);
+#endif
continue;
}
xPredInterUni ( pu, eRefPicList, pcMbBuf, true
, bioApplied
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ , luma, chroma
+#else
, true, true
+#endif
);
}
else
@@ -551,14 +576,22 @@ void InterPrediction::xPredInterBi(PredictionUnit& pu, PelUnitBuf &pcYuvPred, Pe
{
xPredInterUni ( pu, eRefPicList, pcMbBuf, true
, bioApplied
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ , luma, chroma
+#else
, true, true
+#endif
);
}
else
{
xPredInterUni( pu, eRefPicList, pcMbBuf, pu.cu->triangle
, bioApplied
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ , luma, chroma
+#else
, true, true
+#endif
);
}
}
@@ -569,15 +602,27 @@ void InterPrediction::xPredInterBi(PredictionUnit& pu, PelUnitBuf &pcYuvPred, Pe
CPelUnitBuf srcPred1 = ( pu.chromaFormat == CHROMA_400 ?
CPelUnitBuf(pu.chromaFormat, PelBuf(m_acYuvPred[1][0], pcYuvPred.Y())) :
CPelUnitBuf(pu.chromaFormat, PelBuf(m_acYuvPred[1][0], pcYuvPred.Y()), PelBuf(m_acYuvPred[1][1], pcYuvPred.Cb()), PelBuf(m_acYuvPred[1][2], pcYuvPred.Cr())) );
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ bool lumaOnly = luma && !chroma;
+ bool chromaOnly = !luma && chroma;
+#endif
if( !pu.cu->triangle && (!dmvrApplied) && (!bioApplied) && pps.getWPBiPred() && slice.getSliceType() == B_SLICE && pu.cu->GBiIdx==GBI_DEFAULT)
{
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ xWeightedPredictionBi( pu, srcPred0, srcPred1, pcYuvPred, m_maxCompIDToPred, lumaOnly, chromaOnly );
+#else
xWeightedPredictionBi( pu, srcPred0, srcPred1, pcYuvPred, m_maxCompIDToPred );
+#endif
if (yuvPredTmp)
yuvPredTmp->copyFrom(pcYuvPred);
}
else if( !pu.cu->triangle && pps.getUseWP() && slice.getSliceType() == P_SLICE )
{
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ xWeightedPredictionUni( pu, srcPred0, REF_PIC_LIST_0, pcYuvPred, -1, m_maxCompIDToPred, lumaOnly, chromaOnly );
+#else
xWeightedPredictionUni( pu, srcPred0, REF_PIC_LIST_0, pcYuvPred, -1, m_maxCompIDToPred );
+#endif
if (yuvPredTmp)
yuvPredTmp->copyFrom(pcYuvPred);
}
@@ -593,7 +638,11 @@ void InterPrediction::xPredInterBi(PredictionUnit& pu, PelUnitBuf &pcYuvPred, Pe
}
else
{
- xWeightedAverage( pu, srcPred0, srcPred1, pcYuvPred, slice.getSPS()->getBitDepths(), slice.clpRngs(), bioApplied, yuvPredTmp);
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ xWeightedAverage( pu, srcPred0, srcPred1, pcYuvPred, slice.getSPS()->getBitDepths(), slice.clpRngs(), bioApplied, lumaOnly, chromaOnly, yuvPredTmp );
+#else
+ xWeightedAverage( pu, srcPred0, srcPred1, pcYuvPred, slice.getSPS()->getBitDepths(), slice.clpRngs(), bioApplied, yuvPredTmp );
+#endif
}
}
}
@@ -790,7 +839,11 @@ bool InterPrediction::isSubblockVectorSpreadOverLimit( int a, int b, int c, int
return false;
}
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+void InterPrediction::xPredAffineBlk(const ComponentID &compID, const PredictionUnit &pu, const Picture *refPic, const Mv *_mv, PelUnitBuf &dstPic, const bool &bi, const ClpRng &clpRng, bool genChromaMv, const std::pair<int, int> scalingRatio)
+#else
void InterPrediction::xPredAffineBlk( const ComponentID& compID, const PredictionUnit& pu, const Picture* refPic, const Mv* _mv, PelUnitBuf& dstPic, const bool& bi, const ClpRng& clpRng, const std::pair<int, int> scalingRatio )
+#endif
{
JVET_J0090_SET_REF_PICTURE( refPic, compID );
@@ -946,6 +999,68 @@ void InterPrediction::xPredAffineBlk( const ComponentID& compID, const Predictio
#endif
}
}
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ int iScaleXLuma = ::getComponentScaleX(COMPONENT_Y, chFmt);
+ int iScaleYLuma = ::getComponentScaleY(COMPONENT_Y, chFmt);
+
+ if (genChromaMv && pu.chromaFormat != CHROMA_444)
+ {
+ CHECK(compID == COMPONENT_Y, "Chroma only subblock MV calculation should not apply to Luma");
+ int lumaBlockWidth = AFFINE_MIN_BLOCK_SIZE;
+ int lumaBlockHeight = AFFINE_MIN_BLOCK_SIZE;
+
+ CHECK(lumaBlockWidth > (width >> iScaleXLuma), "Sub Block width > Block width");
+ CHECK(lumaBlockHeight > (height >> iScaleYLuma), "Sub Block height > Block height");
+
+ const int cxWidthLuma = width >> iScaleXLuma;
+ const int cxHeightLuma = height >> iScaleYLuma;
+ const int iHalfBWLuma = lumaBlockWidth >> 1;
+ const int iHalfBHLuma = lumaBlockHeight >> 1;
+
+ int iDMvHorXLuma, iDMvHorYLuma, iDMvVerXLuma, iDMvVerYLuma;
+ iDMvHorXLuma = (mvRT - mvLT).getHor() << (iBit - floorLog2(cxWidthLuma));
+ iDMvHorYLuma = (mvRT - mvLT).getVer() << (iBit - floorLog2(cxWidthLuma));
+ if (pu.cu->affineType == AFFINEMODEL_6PARAM)
+ {
+ iDMvVerXLuma = (mvLB - mvLT).getHor() << (iBit - floorLog2(cxHeightLuma));
+ iDMvVerYLuma = (mvLB - mvLT).getVer() << (iBit - floorLog2(cxHeightLuma));
+ }
+ else
+ {
+ iDMvVerXLuma = -iDMvHorYLuma;
+ iDMvVerYLuma = iDMvHorXLuma;
+ }
+
+ const bool subblkMVSpreadOverLimitLuma = isSubblockVectorSpreadOverLimit(iDMvHorXLuma, iDMvHorYLuma, iDMvVerXLuma, iDMvVerYLuma, pu.interDir);
+
+ // get prediction block by block
+ for (int h = 0; h < cxHeightLuma; h += lumaBlockHeight)
+ {
+ for (int w = 0; w < cxWidthLuma; w += lumaBlockWidth)
+ {
+ int iMvScaleTmpHor, iMvScaleTmpVer;
+ if (!subblkMVSpreadOverLimitLuma)
+ {
+ iMvScaleTmpHor = iMvScaleHor + iDMvHorXLuma * (iHalfBWLuma + w) + iDMvVerXLuma * (iHalfBHLuma + h);
+ iMvScaleTmpVer = iMvScaleVer + iDMvHorYLuma * (iHalfBWLuma + w) + iDMvVerYLuma * (iHalfBHLuma + h);
+ }
+ else
+ {
+ iMvScaleTmpHor = iMvScaleHor + iDMvHorXLuma * (cxWidthLuma >> 1) + iDMvVerXLuma * (cxHeightLuma >> 1);
+ iMvScaleTmpVer = iMvScaleVer + iDMvHorYLuma * (cxWidthLuma >> 1) + iDMvVerYLuma * (cxHeightLuma >> 1);
+ }
+
+ roundAffineMv(iMvScaleTmpHor, iMvScaleTmpVer, shift);
+ Mv tmpMv(iMvScaleTmpHor, iMvScaleTmpVer);
+ tmpMv.clipToStorageBitDepth();
+ iMvScaleTmpHor = tmpMv.getHor();
+ iMvScaleTmpVer = tmpMv.getVer();
+
+ m_storedMv[h / AFFINE_MIN_BLOCK_SIZE * MVBUFFER_SIZE + w / AFFINE_MIN_BLOCK_SIZE].set(iMvScaleTmpHor, iMvScaleTmpVer);
+ }
+ }
+ }
+#endif
// get prediction block by block
for ( int h = 0; h < cxHeight; h += blockHeight )
{
@@ -1247,8 +1362,16 @@ void InterPrediction::xCalcBlkGradient(int sx, int sy, int *arraysGx2, int
g_pelBufOP.calcBlkGradient(sx, sy, arraysGx2, arraysGxGy, arraysGxdI, arraysGy2, arraysGydI, sGx2, sGy2, sGxGy, sGxdI, sGydI, width, height, unitSize);
}
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+void InterPrediction::xWeightedAverage(const PredictionUnit& pu, const CPelUnitBuf& pcYuvSrc0, const CPelUnitBuf& pcYuvSrc1, PelUnitBuf& pcYuvDst, const BitDepths& clipBitDepths, const ClpRngs& clpRngs, const bool& bioApplied, bool lumaOnly, bool chromaOnly, PelUnitBuf* yuvDstTmp /*= NULL*/)
+#else
void InterPrediction::xWeightedAverage(const PredictionUnit& pu, const CPelUnitBuf& pcYuvSrc0, const CPelUnitBuf& pcYuvSrc1, PelUnitBuf& pcYuvDst, const BitDepths& clipBitDepths, const ClpRngs& clpRngs, const bool& bioApplied, PelUnitBuf* yuvDstTmp /*= NULL*/)
+#endif
{
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ CHECK( (chromaOnly && lumaOnly), "should not happen" );
+#endif
+
const int iRefIdx0 = pu.refIdx[0];
const int iRefIdx1 = pu.refIdx[1];
@@ -1264,9 +1387,17 @@ void InterPrediction::xWeightedAverage(const PredictionUnit& pu, const CPelUnitB
if( pu.cu->GBiIdx != GBI_DEFAULT && (yuvDstTmp || !pu.mhIntraFlag) )
{
CHECK(bioApplied, "GBi is disallowed with BIO");
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ pcYuvDst.addWeightedAvg(pcYuvSrc0, pcYuvSrc1, clpRngs, pu.cu->GBiIdx, chromaOnly, lumaOnly);
+#else
pcYuvDst.addWeightedAvg(pcYuvSrc0, pcYuvSrc1, clpRngs, pu.cu->GBiIdx);
+#endif
if (yuvDstTmp)
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ yuvDstTmp->addAvg(pcYuvSrc0, pcYuvSrc1, clpRngs, chromaOnly, lumaOnly);
+#else
yuvDstTmp->addAvg(pcYuvSrc0, pcYuvSrc1, clpRngs, false);
+#endif
return;
}
if (bioApplied)
@@ -1299,13 +1430,30 @@ void InterPrediction::xWeightedAverage(const PredictionUnit& pu, const CPelUnitB
getWpScaling(pu.cu->slice, iRefIdx0, iRefIdx1, pwp0, pwp1);
if (!bioApplied)
{
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ if (!chromaOnly)
+#endif
addWeightBiComponent(pcYuvSrc0, pcYuvSrc1, pu.cu->slice->clpRngs(), pwp0, pwp1, pcYuvDst, true, COMPONENT_Y);
}
- addWeightBiComponent(pcYuvSrc0, pcYuvSrc1, pu.cu->slice->clpRngs(), pwp0, pwp1, pcYuvDst, true, COMPONENT_Cb);
- addWeightBiComponent(pcYuvSrc0, pcYuvSrc1, pu.cu->slice->clpRngs(), pwp0, pwp1, pcYuvDst, true, COMPONENT_Cr);
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ if (!lumaOnly)
+ {
+#endif
+ addWeightBiComponent(pcYuvSrc0, pcYuvSrc1, pu.cu->slice->clpRngs(), pwp0, pwp1, pcYuvDst, true, COMPONENT_Cb);
+ addWeightBiComponent(pcYuvSrc0, pcYuvSrc1, pu.cu->slice->clpRngs(), pwp0, pwp1, pcYuvDst, true, COMPONENT_Cr);
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ }
+#endif
}
else
{
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ if (!bioApplied && (lumaOnly || chromaOnly))
+ {
+ pcYuvDst.addAvg(pcYuvSrc0, pcYuvSrc1, clpRngs, chromaOnly, lumaOnly);
+ }
+ else
+#endif
pcYuvDst.addAvg(pcYuvSrc0, pcYuvSrc1, clpRngs, bioApplied);
}
if (yuvDstTmp)
@@ -1316,7 +1464,11 @@ void InterPrediction::xWeightedAverage(const PredictionUnit& pu, const CPelUnitB
yuvDstTmp->bufs[2].copyFrom(pcYuvDst.bufs[2]);
}
else
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ yuvDstTmp->copyFrom(pcYuvDst, lumaOnly, chromaOnly);
+#else
yuvDstTmp->copyFrom(pcYuvDst);
+#endif
}
}
else if( iRefIdx0 >= 0 && iRefIdx1 < 0 )
@@ -1326,9 +1478,17 @@ void InterPrediction::xWeightedAverage(const PredictionUnit& pu, const CPelUnitB
pcYuvDst.copyFrom( pcYuvSrc0 );
}
else
- pcYuvDst.copyClip( pcYuvSrc0, clpRngs );
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ pcYuvDst.copyClip( pcYuvSrc0, clpRngs, lumaOnly, chromaOnly );
+#else
+ pcYuvDst.copyClip( pcYuvSrc0, clpRngs );
+#endif
if (yuvDstTmp)
- yuvDstTmp->copyFrom(pcYuvDst);
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ yuvDstTmp->copyFrom( pcYuvDst, lumaOnly, chromaOnly );
+#else
+ yuvDstTmp->copyFrom( pcYuvDst );
+#endif
}
else if( iRefIdx0 < 0 && iRefIdx1 >= 0 )
{
@@ -1337,9 +1497,17 @@ void InterPrediction::xWeightedAverage(const PredictionUnit& pu, const CPelUnitB
pcYuvDst.copyFrom( pcYuvSrc1 );
}
else
- pcYuvDst.copyClip( pcYuvSrc1, clpRngs );
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ pcYuvDst.copyClip( pcYuvSrc1, clpRngs, lumaOnly, chromaOnly );
+#else
+ pcYuvDst.copyClip( pcYuvSrc1, clpRngs );
+#endif
if (yuvDstTmp)
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ yuvDstTmp->copyFrom(pcYuvDst, lumaOnly, chromaOnly);
+#else
yuvDstTmp->copyFrom(pcYuvDst);
+#endif
}
}
@@ -1514,15 +1682,27 @@ void InterPrediction::motionCompensation( PredictionUnit &pu, PelUnitBuf &predBu
{
xPredInterUni ( pu, eRefPicList, predBuf, true
, false
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ , luma, chroma
+#else
, true, true
+#endif
+ );
+ xWeightedPredictionUni( pu, predBuf, eRefPicList, predBuf, -1, m_maxCompIDToPred
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ , (luma && !chroma), (!luma && chroma)
+#endif
);
- xWeightedPredictionUni( pu, predBuf, eRefPicList, predBuf, -1, m_maxCompIDToPred );
}
else
{
xPredInterUni( pu, eRefPicList, predBuf, false
, false
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ , luma, chroma
+#else
, true, true
+#endif
);
}
}
@@ -1593,20 +1773,36 @@ void InterPrediction::motionCompensation( PredictionUnit &pu, PelUnitBuf &predBu
if (pu.mergeType != MRG_TYPE_DEFAULT_N && pu.mergeType != MRG_TYPE_IBC)
{
CHECK(predBufWOBIO != NULL, "the case should not happen!");
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ xSubPuMC( pu, predBuf, eRefPicList, luma, chroma );
+#else
xSubPuMC( pu, predBuf, eRefPicList );
+#endif
}
else if( xCheckIdenticalMotion( pu ) )
{
xPredInterUni( pu, REF_PIC_LIST_0, predBuf, false
, false
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ , luma, chroma
+#else
, true, true
+#endif
);
if (predBufWOBIO)
+#if 0//JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ predBufWOBIO->copyFrom(predBuf, (luma && !chroma), (chroma && !luma));
+#else
predBufWOBIO->copyFrom(predBuf);
+#endif
}
else
{
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ xPredInterBi(pu, predBuf, luma, chroma, predBufWOBIO);
+#else
xPredInterBi(pu, predBuf, predBufWOBIO);
+#endif
}
}
return;
diff --git a/source/Lib/CommonLib/InterPrediction.h b/source/Lib/CommonLib/InterPrediction.h
index b28b49dd3c93969837a8ef870ff9575f4ba593e8..0a03a6567f449f0b99e898e982175a6cf9b64f6f 100644
--- a/source/Lib/CommonLib/InterPrediction.h
+++ b/source/Lib/CommonLib/InterPrediction.h
@@ -124,7 +124,11 @@ protected:
, const bool& bioApplied
, const bool luma, const bool chroma
);
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ void xPredInterBi ( PredictionUnit& pu, PelUnitBuf &pcYuvPred, const bool luma = true, const bool chroma = true, PelUnitBuf* yuvPredTmp = NULL );
+#else
void xPredInterBi ( PredictionUnit& pu, PelUnitBuf &pcYuvPred, PelUnitBuf* yuvPredTmp = NULL );
+#endif
void xPredInterBlk ( const ComponentID& compID, const PredictionUnit& pu, const Picture* refPic, const Mv& _mv, PelUnitBuf& dstPic, const bool& bi, const ClpRng& clpRng
, const bool& bioApplied
, bool isIBC
@@ -140,15 +144,26 @@ protected:
void xBioGradFilter (Pel* pSrc, int srcStride, int width, int height, int gradStride, Pel* gradX, Pel* gradY, int bitDepth);
void xCalcBIOPar (const Pel* srcY0Temp, const Pel* srcY1Temp, const Pel* gradX0, const Pel* gradX1, const Pel* gradY0, const Pel* gradY1, int* dotProductTemp1, int* dotProductTemp2, int* dotProductTemp3, int* dotProductTemp5, int* dotProductTemp6, const int src0Stride, const int src1Stride, const int gradStride, const int widthG, const int heightG, int bitDepth);
void xCalcBlkGradient (int sx, int sy, int *arraysGx2, int *arraysGxGy, int *arraysGxdI, int *arraysGy2, int *arraysGydI, int &sGx2, int &sGy2, int &sGxGy, int &sGxdI, int &sGydI, int width, int height, int unitSize);
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ void xWeightedAverage ( const PredictionUnit& pu, const CPelUnitBuf& pcYuvSrc0, const CPelUnitBuf& pcYuvSrc1, PelUnitBuf& pcYuvDst, const BitDepths& clipBitDepths, const ClpRngs& clpRngs, const bool& bioApplied, const bool lumaOnly = false, const bool chromaOnly = false, PelUnitBuf* yuvDstTmp = NULL );
+#else
void xWeightedAverage ( const PredictionUnit& pu, const CPelUnitBuf& pcYuvSrc0, const CPelUnitBuf& pcYuvSrc1, PelUnitBuf& pcYuvDst, const BitDepths& clipBitDepths, const ClpRngs& clpRngs, const bool& bioApplied, PelUnitBuf* yuvDstTmp = NULL );
+#endif
void xApplyBiPROF (const PredictionUnit& pu, const CPelBuf& pcYuvSrc0, const CPelBuf& pcYuvSrc1, PelBuf& pcYuvDst, const ClpRng& clpRng);
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ void xPredAffineBlk ( const ComponentID& compID, const PredictionUnit& pu, const Picture* refPic, const Mv* _mv, PelUnitBuf& dstPic, const bool& bi, const ClpRng& clpRng, const bool genChromaMv = false, const std::pair<int, int> scalingRatio = SCALE_1X );
+#else
void xPredAffineBlk ( const ComponentID& compID, const PredictionUnit& pu, const Picture* refPic, const Mv* _mv, PelUnitBuf& dstPic, const bool& bi, const ClpRng& clpRng, const std::pair<int, int> scalingRatio = SCALE_1X );
-
+#endif
void xWeightedTriangleBlk ( const PredictionUnit &pu, const uint32_t width, const uint32_t height, const ComponentID compIdx, const bool splitDir, PelUnitBuf& predDst, PelUnitBuf& predSrc0, PelUnitBuf& predSrc1 );
static bool xCheckIdenticalMotion( const PredictionUnit& pu );
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ void xSubPuMC(PredictionUnit& pu, PelUnitBuf& predBuf, const RefPicList &eRefPicList = REF_PIC_LIST_X, const bool luma = true, const bool chroma = true);
+#else
void xSubPuMC(PredictionUnit& pu, PelUnitBuf& predBuf, const RefPicList &eRefPicList = REF_PIC_LIST_X);
+#endif
void xSubPuBio(PredictionUnit& pu, PelUnitBuf& predBuf, const RefPicList &eRefPicList = REF_PIC_LIST_X, PelUnitBuf* yuvDstTmp = NULL);
void destroy();
diff --git a/source/Lib/CommonLib/TypeDef.h b/source/Lib/CommonLib/TypeDef.h
index fbb7edbfbfc8055154a63dc3c684cf7e9afc3d78..435427f2b2b7067f3a8952d4381a7613a8b19369 100644
--- a/source/Lib/CommonLib/TypeDef.h
+++ b/source/Lib/CommonLib/TypeDef.h
@@ -50,6 +50,8 @@
#include <assert.h>
#include <cassert>
+#define JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP 1 // JVET-P0445: encoder speed up for sub-block based merge candidate search
+
#define JVET_P0057_BDOF_PROF_HARMONIZATION 1 // JVET-P0057: harmonization of BDOF and PROF on motion refinement precis
#define JVET_P0400_REMOVE_SHARED_MERGE_LIST 1 // JVET-P0400: removeal of shared merge list
diff --git a/source/Lib/CommonLib/WeightPrediction.cpp b/source/Lib/CommonLib/WeightPrediction.cpp
index 007fb08fc5a487d02b8c18df6826acc73719dfe3..d39c65b4d283324ad6f19a6fe3d3f28b77bad47e 100644
--- a/source/Lib/CommonLib/WeightPrediction.cpp
+++ b/source/Lib/CommonLib/WeightPrediction.cpp
@@ -161,13 +161,25 @@ void WeightPrediction::addWeightBi(const CPelUnitBuf &pcYuvSrc0,
const WPScalingParam *const wp1,
PelUnitBuf &rpcYuvDst,
const bool bRoundLuma /*= true*/,
- const ComponentID maxNumComp)
+ const ComponentID maxNumComp
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ , bool lumaOnly
+ , bool chromaOnly
+#endif
+)
{
const bool enableRounding[MAX_NUM_COMPONENT] = { bRoundLuma, true, true };
const uint32_t numValidComponent = (const uint32_t)pcYuvSrc0.bufs.size();
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ CHECK( lumaOnly && chromaOnly, "Not allowed to have both lumaOnly and chromaOnly selected" );
+ int firstComponent = chromaOnly ? 1 : 0;
+ int lastComponent = lumaOnly ? 0 : maxNumComp;
+ for (int componentIndex = firstComponent; componentIndex < numValidComponent && componentIndex <= lastComponent; componentIndex++)
+#else
for (int componentIndex = 0; componentIndex < numValidComponent && componentIndex <= maxNumComp; componentIndex++)
+#endif
{
const ComponentID compID = ComponentID(componentIndex);
@@ -275,11 +287,24 @@ void WeightPrediction::addWeightUni(const CPelUnitBuf &pcYuvSrc0,
const ClpRngs &clpRngs,
const WPScalingParam *const wp0,
PelUnitBuf &rpcYuvDst,
- const ComponentID maxNumComp)
+ const ComponentID maxNumComp
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ , bool lumaOnly
+ , bool chromaOnly
+#endif
+)
{
const uint32_t numValidComponent = (const uint32_t)pcYuvSrc0.bufs.size();
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ CHECK( lumaOnly && chromaOnly, "Not allowed to have both lumaOnly and chromaOnly selected" );
+ int firstComponent = chromaOnly ? 1 : 0;
+ int lastComponent = lumaOnly ? 0 : maxNumComp;
+ for (int componentIndex = firstComponent; componentIndex < numValidComponent && componentIndex <= lastComponent;
+ componentIndex++)
+#else
for (int componentIndex = 0; componentIndex < numValidComponent && componentIndex <= maxNumComp; componentIndex++)
+#endif
{
const ComponentID compID = ComponentID(componentIndex);
@@ -371,7 +396,12 @@ void WeightPrediction::xWeightedPredictionUni(const PredictionUnit &pu,
const RefPicList &eRefPicList,
PelUnitBuf &pcYuvPred,
const int iRefIdx_input/* = -1*/,
- const ComponentID maxNumComp)
+ const ComponentID maxNumComp
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ , bool lumaOnly
+ , bool chromaOnly
+#endif
+)
{
WPScalingParam *pwp, *pwpTmp;
@@ -391,14 +421,23 @@ void WeightPrediction::xWeightedPredictionUni(const PredictionUnit &pu,
{
getWpScaling(pu.cs->slice, -1, iRefIdx, pwpTmp, pwp, maxNumComp);
}
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ addWeightUni(pcYuvSrc, pu.cu->slice->clpRngs(), pwp, pcYuvPred, maxNumComp, lumaOnly, chromaOnly);
+#else
addWeightUni(pcYuvSrc, pu.cu->slice->clpRngs(), pwp, pcYuvPred, maxNumComp);
+#endif
}
void WeightPrediction::xWeightedPredictionBi(const PredictionUnit &pu,
const CPelUnitBuf &pcYuvSrc0,
const CPelUnitBuf &pcYuvSrc1,
PelUnitBuf &rpcYuvDst,
- const ComponentID maxNumComp)
+ const ComponentID maxNumComp
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ , bool lumaOnly
+ , bool chromaOnly
+#endif
+)
{
const int iRefIdx0 = pu.refIdx[0];
const int iRefIdx1 = pu.refIdx[1];
@@ -413,15 +452,27 @@ void WeightPrediction::xWeightedPredictionBi(const PredictionUnit &pu,
if (iRefIdx0 >= 0 && iRefIdx1 >= 0)
{
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ addWeightBi(pcYuvSrc0, pcYuvSrc1, pu.cu->slice->clpRngs(), pwp0, pwp1, rpcYuvDst, true, maxNumComp, lumaOnly, chromaOnly);
+#else
addWeightBi(pcYuvSrc0, pcYuvSrc1, pu.cu->slice->clpRngs(), pwp0, pwp1, rpcYuvDst, true, maxNumComp);
+#endif
}
else if (iRefIdx0 >= 0 && iRefIdx1 < 0)
{
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ addWeightUni(pcYuvSrc0, pu.cu->slice->clpRngs(), pwp0, rpcYuvDst, maxNumComp, lumaOnly, chromaOnly);
+#else
addWeightUni(pcYuvSrc0, pu.cu->slice->clpRngs(), pwp0, rpcYuvDst, maxNumComp);
+#endif
}
else if (iRefIdx0 < 0 && iRefIdx1 >= 0)
{
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ addWeightUni(pcYuvSrc1, pu.cu->slice->clpRngs(), pwp1, rpcYuvDst, maxNumComp, lumaOnly, chromaOnly);
+#else
addWeightUni(pcYuvSrc1, pu.cu->slice->clpRngs(), pwp1, rpcYuvDst, maxNumComp);
+#endif
}
else
{
diff --git a/source/Lib/CommonLib/WeightPrediction.h b/source/Lib/CommonLib/WeightPrediction.h
index a038dd2571dbbe69de1b33f475b6ed05e5bfb31c..5df9ad1cc5a100641aeaf9267607a473f1d39681 100644
--- a/source/Lib/CommonLib/WeightPrediction.h
+++ b/source/Lib/CommonLib/WeightPrediction.h
@@ -69,7 +69,12 @@ public:
const WPScalingParam *const wp1,
PelUnitBuf &rpcYuvDst,
const bool bRoundLuma = true,
- const ComponentID maxNumComp = MAX_NUM_COMPONENT );
+ const ComponentID maxNumComp = MAX_NUM_COMPONENT
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ , bool lumaOnly = false
+ , bool chromaOnly = false
+#endif
+ );
void addWeightBiComponent( const CPelUnitBuf &pcYuvSrc0,
const CPelUnitBuf &pcYuvSrc1,
@@ -84,20 +89,35 @@ public:
const ClpRngs &clpRngs,
const WPScalingParam *const wp0,
PelUnitBuf &rpcYuvDst,
- const ComponentID maxNumComp = MAX_NUM_COMPONENT);
+ const ComponentID maxNumComp = MAX_NUM_COMPONENT
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ , bool lumaOnly = false
+ , bool chromaOnly = false
+#endif
+ );
void xWeightedPredictionUni( const PredictionUnit &pu,
const CPelUnitBuf &pcYuvSrc,
const RefPicList &eRefPicList,
PelUnitBuf &pcYuvPred,
const int iRefIdx=-1,
- const ComponentID maxNumComp = MAX_NUM_COMPONENT);
+ const ComponentID maxNumComp = MAX_NUM_COMPONENT
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ , bool lumaOnly = false
+ , bool chromaOnly = false
+#endif
+ );
void xWeightedPredictionBi( const PredictionUnit &pu,
const CPelUnitBuf &pcYuvSrc0,
const CPelUnitBuf &pcYuvSrc1,
PelUnitBuf &pcYuvDst,
- const ComponentID maxNumComp = MAX_NUM_COMPONENT );
+ const ComponentID maxNumComp = MAX_NUM_COMPONENT
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ , bool lumaOnly = false
+ , bool chromaOnly = false
+#endif
+ );
};
#endif
diff --git a/source/Lib/EncoderLib/EncCu.cpp b/source/Lib/EncoderLib/EncCu.cpp
index 10d54df714e00d0fe6efb503aaa74413b5297f2c..5485e57ba51f56b8006fdda253116306f0498aaf 100644
--- a/source/Lib/EncoderLib/EncCu.cpp
+++ b/source/Lib/EncoderLib/EncCu.cpp
@@ -3149,7 +3149,11 @@ void EncCu::xCheckRDCostAffineMerge2Nx2N( CodingStructure *&tempCS, CodingStruct
distParam.cur = acMergeBuffer[uiMergeCand].Y();
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ m_pcInterSearch->motionCompensation( pu, acMergeBuffer[uiMergeCand], REF_PIC_LIST_X, true, false );
+#else
m_pcInterSearch->motionCompensation( pu, acMergeBuffer[uiMergeCand] );
+#endif
Distortion uiSad = distParam.distFunc( distParam );
uint32_t uiBitsCand = uiMergeCand + 1;
@@ -3248,7 +3252,12 @@ void EncCu::xCheckRDCostAffineMerge2Nx2N( CodingStructure *&tempCS, CodingStruct
}
if ( mrgTempBufSet )
{
+#if JVET_P0445_SUBBLOCK_MERGE_ENC_SPEEDUP
+ tempCS->getPredBuf().copyFrom(acMergeBuffer[uiMergeCand], true, false); // Copy Luma Only
+ m_pcInterSearch->motionCompensation(pu, REF_PIC_LIST_X, false, true);
+#else
tempCS->getPredBuf().copyFrom( acMergeBuffer[uiMergeCand] );
+#endif
}
else
{