diff --git a/source/App/EncoderApp/EncApp.cpp b/source/App/EncoderApp/EncApp.cpp
index b7f45d32aae6c42cb38ab92c06ae4207a3b5534c..34cd83ca8e9b0858bd7ec57dd6b35c6ef4f9a5f4 100644
--- a/source/App/EncoderApp/EncApp.cpp
+++ b/source/App/EncoderApp/EncApp.cpp
@@ -736,6 +736,9 @@ void EncApp::xInitLibCfg()
#endif
m_cEncLib.setPROF ( m_PROF );
m_cEncLib.setBIO (m_BIO);
+#if ARMC_TM
+ m_cEncLib.setAML ( m_AML );
+#endif
m_cEncLib.setUseLMChroma ( m_LMChroma );
m_cEncLib.setHorCollocatedChromaFlag ( m_horCollocatedChromaFlag );
m_cEncLib.setVerCollocatedChromaFlag ( m_verCollocatedChromaFlag );
diff --git a/source/App/EncoderApp/EncAppCfg.cpp b/source/App/EncoderApp/EncAppCfg.cpp
index 9d312e8c8656ce6e11391a3800eaa870fd622000..f684965eb91ce3b440f30ff157fb102e75212873 100644
--- a/source/App/EncoderApp/EncAppCfg.cpp
+++ b/source/App/EncoderApp/EncAppCfg.cpp
@@ -978,6 +978,9 @@ bool EncAppCfg::parseCfg( int argc, char* argv[] )
#endif
("PROF", m_PROF, false, "Enable Prediction refinement with optical flow for affine mode (0:off, 1:on) [default: off]")
("BIO", m_BIO, false, "Enable bi-directional optical flow")
+#if ARMC_TM
+ ("AML", m_AML, true, "Enable adaptive merge list")
+#endif
("IMV", m_ImvMode, 1, "Adaptive MV precision Mode (IMV)\n"
"\t0: disabled\n"
"\t1: enabled (1/2-Pel, Full-Pel and 4-PEL)\n")
@@ -4266,6 +4269,9 @@ void EncAppCfg::xPrintParameter()
msg( VERBOSE, "DualITree:%d ", m_dualTree );
msg( VERBOSE, "IMV:%d ", m_ImvMode );
msg( VERBOSE, "BIO:%d ", m_BIO );
+#if ARMC_TM
+ msg( VERBOSE, "AML:%d ", m_AML );
+#endif
msg( VERBOSE, "LMChroma:%d ", m_LMChroma );
msg( VERBOSE, "HorCollocatedChroma:%d ", m_horCollocatedChromaFlag );
msg( VERBOSE, "VerCollocatedChroma:%d ", m_verCollocatedChromaFlag );
diff --git a/source/App/EncoderApp/EncAppCfg.h b/source/App/EncoderApp/EncAppCfg.h
index 965161ce8943a1f2b4d519332fb100ed409cd8c6..3a6a3d1b4ddef2838bf9177945a8060638beb078 100644
--- a/source/App/EncoderApp/EncAppCfg.h
+++ b/source/App/EncoderApp/EncAppCfg.h
@@ -358,6 +358,9 @@ protected:
#endif
bool m_PROF;
bool m_BIO;
+#if ARMC_TM
+ bool m_AML;
+#endif
int m_LMChroma;
bool m_horCollocatedChromaFlag;
bool m_verCollocatedChromaFlag;
diff --git a/source/Lib/CommonLib/Buffer.cpp b/source/Lib/CommonLib/Buffer.cpp
index 1c62b65a642917ef775f102a4e988400a4724a89..fe9c674101b97287aa7fc7f6976615de495cb5a9 100644
--- a/source/Lib/CommonLib/Buffer.cpp
+++ b/source/Lib/CommonLib/Buffer.cpp
@@ -913,11 +913,13 @@ void AreaBuf<Pel>::copyClip( const AreaBuf<const Pel> &src, const ClpRng& clpRng
const unsigned srcStride = src.stride;
const unsigned destStride = stride;
+#if !ARMC_TM
if( width == 1 )
{
THROW( "Blocks of width = 1 not supported" );
}
else
+#endif
{
#define RECO_OP( ADDR ) dest[ADDR] = ClipPel( srcp[ADDR], clpRng )
#define RECO_INC \
@@ -1008,6 +1010,18 @@ void AreaBuf<Pel>::linearTransform( const int scale, const int shift, const int
const Pel* src = buf;
Pel* dst = buf;
+#if ARMC_TM
+#if ENABLE_SIMD_OPT_BUFFER && defined(TARGET_SIMD_X86)
+ if ((width & 7) == 0)
+ {
+ g_pelBufOP.linTf8(src, stride, dst, stride, width, height, scale, shift, offset, clpRng, bClip);
+ }
+ else if ((width & 3) == 0)
+ {
+ g_pelBufOP.linTf4(src, stride, dst, stride, width, height, scale, shift, offset, clpRng, bClip);
+ }
+#endif
+#else
if( width == 1 )
{
THROW( "Blocks of width = 1 not supported" );
@@ -1021,6 +1035,7 @@ void AreaBuf<Pel>::linearTransform( const int scale, const int shift, const int
{
g_pelBufOP.linTf4( src, stride, dst, stride, width, height, scale, shift, offset, clpRng, bClip );
}
+#endif
#endif
else
{
diff --git a/source/Lib/CommonLib/CommonDef.h b/source/Lib/CommonLib/CommonDef.h
index 510b0f039f154372866fc88ca2fac83780dc0699..197fc3ceb92883fd6d11a41490da8b12ae8251c9 100644
--- a/source/Lib/CommonLib/CommonDef.h
+++ b/source/Lib/CommonLib/CommonDef.h
@@ -578,6 +578,12 @@ static const double AMAXBT_TH64 = 30.0;
static const double AMAXBT_TH128 = 60.0;
#endif
+#if ARMC_TM
+static const int AML_MERGE_TEMPLATE_SIZE = 1;
+static const int ADAPTIVE_SUB_GROUP_SIZE = 5;
+static const int ADAPTIVE_AFFINE_SUB_GROUP_SIZE = 3;
+#endif
+
// need to know for static memory allocation
static const int MAX_DELTA_QP = 7; ///< maximum supported delta QP value
static const int MAX_TESTED_QPs = ( 1 + 1 + ( MAX_DELTA_QP << 1 ) ); ///< dqp=0 +- max_delta_qp + lossless mode
diff --git a/source/Lib/CommonLib/InterPrediction.cpp b/source/Lib/CommonLib/InterPrediction.cpp
index be9096990085991782b9590b417a6bf1297cf9d7..1c49d1ec186526497d5167d732f70f3d9dd23704 100644
--- a/source/Lib/CommonLib/InterPrediction.cpp
+++ b/source/Lib/CommonLib/InterPrediction.cpp
@@ -44,7 +44,7 @@
#include <memory.h>
#include <algorithm>
-#if INTER_LIC || (TM_AMVP || TM_MRG)
+#if INTER_LIC || (TM_AMVP || TM_MRG) || ARMC_TM
#include "Reshape.h"
#endif
@@ -181,6 +181,18 @@ InterPrediction::InterPrediction()
CHECK(mvSearchIdx_bilMrg != (2 * BDMVR_INTME_RANGE + 1) * (2 * BDMVR_INTME_RANGE + 1),
"this is wrong, mvSearchIdx_bilMrg != (2 * BDMVR_INTME_RANGE + 1) * (2 * BDMVR_INTME_RANGE + 1)");
#endif
+#if ARMC_TM
+ for (uint32_t ch = 0; ch < MAX_NUM_COMPONENT; ch++)
+ {
+ for (uint32_t tmplt = 0; tmplt < 2; tmplt++)
+ {
+ m_acYuvCurAMLTemplate[tmplt][ch] = nullptr;
+ m_acYuvRefAboveTemplate[tmplt][ch] = nullptr;
+ m_acYuvRefLeftTemplate[tmplt][ch] = nullptr;
+ m_acYuvRefAMLTemplate[tmplt][ch] = nullptr;
+ }
+ }
+#endif
}
InterPrediction::~InterPrediction()
@@ -276,16 +288,32 @@ void InterPrediction::destroy()
#if MULTI_HYP_PRED
m_additionalHypothesisStorage.destroy();
#endif
+#if ARMC_TM
+ for (uint32_t ch = 0; ch < MAX_NUM_COMPONENT; ch++)
+ {
+ for (uint32_t tmplt = 0; tmplt < 2; tmplt++)
+ {
+ xFree(m_acYuvCurAMLTemplate[tmplt][ch]);
+ xFree(m_acYuvRefAboveTemplate[tmplt][ch]);
+ xFree(m_acYuvRefLeftTemplate[tmplt][ch]);
+ xFree(m_acYuvRefAMLTemplate[tmplt][ch]);
+ m_acYuvCurAMLTemplate[tmplt][ch] = nullptr;
+ m_acYuvRefAboveTemplate[tmplt][ch] = nullptr;
+ m_acYuvRefLeftTemplate[tmplt][ch] = nullptr;
+ m_acYuvRefAMLTemplate[tmplt][ch] = nullptr;
+ }
+}
+#endif
}
-#if INTER_LIC || (TM_AMVP || TM_MRG)
+#if INTER_LIC || (TM_AMVP || TM_MRG) || ARMC_TM
void InterPrediction::init( RdCost* pcRdCost, ChromaFormat chromaFormatIDC, const int ctuSize, Reshape* reshape )
#else
void InterPrediction::init( RdCost* pcRdCost, ChromaFormat chromaFormatIDC, const int ctuSize )
#endif
{
m_pcRdCost = pcRdCost;
-#if INTER_LIC || (TM_AMVP || TM_MRG)
+#if INTER_LIC || (TM_AMVP || TM_MRG) || ARMC_TM
m_pcReshape = reshape;
#endif
@@ -411,6 +439,18 @@ void InterPrediction::init( RdCost* pcRdCost, ChromaFormat chromaFormatIDC, cons
m_pcLICRecAboveTemplate = (Pel*)xMalloc(Pel, MAX_CU_SIZE);
}
#endif
+#if ARMC_TM
+ for (uint32_t ch = 0; ch < MAX_NUM_COMPONENT; ch++)
+ {
+ for (uint32_t tmplt = 0; tmplt < 2; tmplt++)
+ {
+ m_acYuvCurAMLTemplate[tmplt][ch] = (Pel*)xMalloc(Pel, MAX_CU_SIZE * MAX_CU_SIZE);
+ m_acYuvRefAboveTemplate[tmplt][ch] = (Pel*)xMalloc(Pel, MAX_CU_SIZE * MAX_CU_SIZE);
+ m_acYuvRefLeftTemplate[tmplt][ch] = (Pel*)xMalloc(Pel, MAX_CU_SIZE * MAX_CU_SIZE);
+ m_acYuvRefAMLTemplate[tmplt][ch] = (Pel*)xMalloc(Pel, MAX_CU_SIZE * MAX_CU_SIZE);
+ }
+ }
+#endif
if (m_storedMv == nullptr)
{
@@ -1462,8 +1502,22 @@ void InterPrediction::xPredInterBlk ( const ComponentID& compID, const Predictio
, bool bilinearMC
, Pel *srcPadBuf
, int32_t srcPadStride
+#if ARMC_TM
+ , bool AML
+#if INTER_LIC
+ , bool doLic
+ , Mv mvCurr
+#endif
+#endif
)
{
+#if ARMC_TM
+ int nFilterIdx = AML ? 0 : 0;
+ if (bilinearMC)
+ {
+ nFilterIdx = 1;
+ }
+#endif
JVET_J0090_SET_REF_PICTURE( refPic, compID );
const ChromaFormat chFmt = pu.chromaFormat;
const bool rndRes = !bi;
@@ -1571,11 +1625,19 @@ void InterPrediction::xPredInterBlk ( const ComponentID& compID, const Predictio
if( yFrac == 0 )
{
+#if ARMC_TM
+ m_if.filterHor( compID, (Pel*)refBuf.buf, refBuf.stride, dstBuf.buf, dstBuf.stride, backupWidth, backupHeight, xFrac, rndRes, chFmt, clpRng, nFilterIdx, bilinearMC, useAltHpelIf );
+#else
m_if.filterHor( compID, ( Pel* ) refBuf.buf, refBuf.stride, dstBuf.buf, dstBuf.stride, backupWidth, backupHeight, xFrac, rndRes, chFmt, clpRng, bilinearMC, bilinearMC, useAltHpelIf);
+#endif
}
else if( xFrac == 0 )
{
+#if ARMC_TM
+ m_if.filterVer( compID, (Pel*)refBuf.buf, refBuf.stride, dstBuf.buf, dstBuf.stride, backupWidth, backupHeight, yFrac, true, rndRes, chFmt, clpRng, nFilterIdx, bilinearMC, useAltHpelIf );
+#else
m_if.filterVer( compID, ( Pel* ) refBuf.buf, refBuf.stride, dstBuf.buf, dstBuf.stride, backupWidth, backupHeight, yFrac, true, rndRes, chFmt, clpRng, bilinearMC, bilinearMC, useAltHpelIf);
+#endif
}
else
{
@@ -1607,13 +1669,23 @@ void InterPrediction::xPredInterBlk ( const ComponentID& compID, const Predictio
#else
int vFilterSize = isLuma( compID ) ? NTAPS_LUMA : NTAPS_CHROMA;
#endif
+#if ARMC_TM
+ if (isLuma(compID) && nFilterIdx == 1)
+#else
if (bilinearMC)
+#endif
{
vFilterSize = NTAPS_BILINEAR;
}
+#if ARMC_TM
+ m_if.filterHor(compID, (Pel*)refBuf.buf - ((vFilterSize >> 1) - 1) * refBuf.stride, refBuf.stride, tmpBuf.buf, tmpBuf.stride, backupWidth, backupHeight + vFilterSize - 1, xFrac, false, chFmt, clpRng, nFilterIdx, bilinearMC, useAltHpelIf);
+ JVET_J0090_SET_CACHE_ENABLE(false);
+ m_if.filterVer(compID, (Pel*)tmpBuf.buf + ((vFilterSize >> 1) - 1) * tmpBuf.stride, tmpBuf.stride, dstBuf.buf, dstBuf.stride, backupWidth, backupHeight, yFrac, false, rndRes, chFmt, clpRng, nFilterIdx, bilinearMC, useAltHpelIf);
+#else
m_if.filterHor( compID, ( Pel* ) refBuf.buf - ( ( vFilterSize >> 1 ) - 1 ) * refBuf.stride, refBuf.stride, tmpBuf.buf, tmpBuf.stride, backupWidth, backupHeight + vFilterSize - 1, xFrac, false, chFmt, clpRng, bilinearMC, bilinearMC, useAltHpelIf);
JVET_J0090_SET_CACHE_ENABLE( false );
m_if.filterVer( compID, ( Pel* ) tmpBuf.buf + ( ( vFilterSize >> 1 ) - 1 ) * tmpBuf.stride, tmpBuf.stride, dstBuf.buf, dstBuf.stride, backupWidth, backupHeight, yFrac, false, rndRes, chFmt, clpRng, bilinearMC, bilinearMC, useAltHpelIf);
+#endif
#if SIMD_4x4_12
}
#endif
@@ -1677,12 +1749,25 @@ void InterPrediction::xPredInterBlk ( const ComponentID& compID, const Predictio
m_predictionBeforeLIC.bufs[compID].copyFrom( dstBuf );
}
+#if ARMC_TM
+ if (pu.cu->LICFlag && (!pu.ciipFlag || doLic))
+#else
if( pu.cu->LICFlag && !pu.ciipFlag )
+#endif
{
CHECK( pu.cu->geoFlag, "Geometric mode is not used with LIC" );
CHECK( CU::isIBC( *pu.cu ), "IBC mode is not used with LIC" );
CHECK( pu.interDir == 3, "Bi-prediction is not used with LIC" );
+#if !ARMC_TM
CHECK( pu.ciipFlag, "CIIP mode is not used with LIC" );
+#endif
+#if ARMC_TM
+ if (AML)
+ {
+ xLocalIlluComp(pu, compID, *refPic, mvCurr, bi, dstBuf);
+ }
+ else
+#endif
xLocalIlluComp( pu, compID, *refPic, _mv, bi, dstBuf );
}
#endif
@@ -2838,6 +2923,204 @@ void InterPrediction::xWeightedAverage(
}
}
+#if ARMC_TM
+#if !INTER_LIC
+template <bool TrueA_FalseL>
+void InterPrediction::xGetPredBlkTpl(const CodingUnit& cu, const ComponentID compID, const CPelBuf& refBuf, const Mv& mv, const int posW, const int posH, const int tplSize, Pel* predBlkTpl)
+{
+ const int lumaShift = 2 + MV_FRACTIONAL_BITS_DIFF;
+ const int horShift = (lumaShift + ::getComponentScaleX(compID, cu.chromaFormat));
+ const int verShift = (lumaShift + ::getComponentScaleY(compID, cu.chromaFormat));
+
+ const int xInt = mv.getHor() >> horShift;
+ const int yInt = mv.getVer() >> verShift;
+ const int xFrac = mv.getHor() & ((1 << horShift) - 1);
+ const int yFrac = mv.getVer() & ((1 << verShift) - 1);
+
+ const Pel* ref;
+ Pel* dst;
+ int refStride, dstStride, bw, bh;
+ if (TrueA_FalseL)
+ {
+ ref = refBuf.bufAt(cu.blocks[compID].pos().offset(xInt + posW, yInt + posH - 1));
+ dst = predBlkTpl + posW;
+ refStride = refBuf.stride;
+ dstStride = tplSize;
+ bw = tplSize;
+ bh = 1;
+ }
+ else
+ {
+ ref = refBuf.bufAt(cu.blocks[compID].pos().offset(xInt + posW - 1, yInt + posH));
+ dst = predBlkTpl + posH;
+ refStride = refBuf.stride;
+ dstStride = 1;
+ bw = 1;
+ bh = tplSize;
+ }
+
+ const int nFilterIdx = 0;
+ const bool useAltHpelIf = false;
+
+ if (yFrac == 0)
+ {
+ m_if.filterHor(compID, (Pel*)ref, refStride, dst, dstStride, bw, bh, xFrac, true, cu.chromaFormat, cu.slice->clpRng(compID), nFilterIdx, false, useAltHpelIf);
+ }
+ else if (xFrac == 0)
+ {
+ m_if.filterVer(compID, (Pel*)ref, refStride, dst, dstStride, bw, bh, yFrac, true, true, cu.chromaFormat, cu.slice->clpRng(compID), nFilterIdx, false, useAltHpelIf);
+ }
+ else
+ {
+#if IF_12TAP
+ const int vFilterSize = isLuma(compID) ? NTAPS_LUMA(0) : NTAPS_CHROMA;
+#else
+ const int vFilterSize = isLuma(compID) ? NTAPS_LUMA : NTAPS_CHROMA;
+#endif
+ PelBuf tmpBuf = PelBuf(m_filteredBlockTmp[0][compID], Size(bw, bh + vFilterSize - 1));
+
+ m_if.filterHor(compID, (Pel*)ref - ((vFilterSize >> 1) - 1)*refStride, refStride, tmpBuf.buf, tmpBuf.stride, bw, bh + vFilterSize - 1, xFrac, false, cu.chromaFormat, cu.slice->clpRng(compID), nFilterIdx, false, useAltHpelIf);
+ JVET_J0090_SET_CACHE_ENABLE(false);
+ m_if.filterVer(compID, tmpBuf.buf + ((vFilterSize >> 1) - 1)*tmpBuf.stride, tmpBuf.stride, dst, dstStride, bw, bh, yFrac, false, true, cu.chromaFormat, cu.slice->clpRng(compID), nFilterIdx, false, useAltHpelIf);
+ JVET_J0090_SET_CACHE_ENABLE(true);
+ }
+}
+#endif
+void InterPrediction::xWeightedAverageY(const PredictionUnit& pu, const CPelUnitBuf& pcYuvSrc0, const CPelUnitBuf& pcYuvSrc1, PelUnitBuf& pcYuvDst, const BitDepths& clipBitDepths, const ClpRngs& clpRngs)
+{
+ const int iRefIdx0 = pu.refIdx[0];
+ const int iRefIdx1 = pu.refIdx[1];
+
+ if (iRefIdx0 >= 0 && iRefIdx1 >= 0)
+ {
+ if (pu.cu->BcwIdx != BCW_DEFAULT)
+ {
+ pcYuvDst.addWeightedAvg(pcYuvSrc0, pcYuvSrc1, clpRngs, pu.cu->BcwIdx, false, true);
+ }
+ else
+ {
+ pcYuvDst.addAvg(pcYuvSrc0, pcYuvSrc1, clpRngs, false, true);
+ }
+ }
+ else if (iRefIdx0 >= 0 && iRefIdx1 < 0)
+ {
+ pcYuvDst.copyClip(pcYuvSrc0, clpRngs, true);
+ }
+ else if (iRefIdx0 < 0 && iRefIdx1 >= 0)
+ {
+ pcYuvDst.copyClip(pcYuvSrc1, clpRngs, true);
+ }
+}
+void InterPrediction::xPredAffineTpl(const PredictionUnit &pu, const RefPicList &eRefPicList, int* numTemplate, Pel* refLeftTemplate, Pel* refAboveTemplate)
+{
+ int iRefIdx = pu.refIdx[eRefPicList];
+ CHECK(iRefIdx < 0, "iRefIdx incorrect.");
+ const Picture* refPic = pu.cu->slice->getRefPic(eRefPicList, iRefIdx)->unscaledPic;
+ Mv mvLT = pu.mvAffi[eRefPicList][0];
+ Mv mvRT = pu.mvAffi[eRefPicList][1];
+ Mv mvLB = pu.mvAffi[eRefPicList][2];
+ // get affine sub-block width and height
+ const int width = pu.Y().width;
+ const int height = pu.Y().height;
+ int blockWidth = AFFINE_MIN_BLOCK_SIZE;
+ int blockHeight = AFFINE_MIN_BLOCK_SIZE;
+
+ CHECK(blockWidth > width, "Sub Block width > Block width");
+ CHECK(blockHeight > height, "Sub Block height > Block height");
+
+ const int cxWidth = width;
+ const int cxHeight = height;
+ const int iBit = MAX_CU_DEPTH;
+ int iDMvHorX, iDMvHorY, iDMvVerX, iDMvVerY;
+ iDMvHorX = (mvRT - mvLT).getHor() << (iBit - floorLog2(width));
+ iDMvHorY = (mvRT - mvLT).getVer() << (iBit - floorLog2(width));
+ if (pu.cu->affineType == AFFINEMODEL_6PARAM)
+ {
+ iDMvVerX = (mvLB - mvLT).getHor() << (iBit - floorLog2(height));
+ iDMvVerY = (mvLB - mvLT).getVer() << (iBit - floorLog2(height));
+ }
+ else
+ {
+ iDMvVerX = -iDMvHorY;
+ iDMvVerY = iDMvHorX;
+ }
+ int iMvScaleHor = mvLT.getHor() << iBit;
+ int iMvScaleVer = mvLT.getVer() << iBit;
+
+ const int shift = iBit - 4 + MV_FRACTIONAL_BITS_INTERNAL;
+#if !AFFINE_RM_CONSTRAINTS_AND_OPT
+ const bool subblkMVSpreadOverLimit = isSubblockVectorSpreadOverLimit(iDMvHorX, iDMvHorY, iDMvVerX, iDMvVerY, pu.interDir);
+#endif
+#if AFFINE_RM_CONSTRAINTS_AND_OPT
+ if (iDMvHorX == 0 && iDMvHorY == 0)
+ blockWidth = width;
+ else
+ {
+ int maxDmv = std::max(abs(iDMvHorX), abs(iDMvHorY)) * blockWidth;
+ int TH = 1 << (iBit - 1); // Half pel
+ while (maxDmv < TH && blockWidth < width)
+ {
+ blockWidth <<= 1;
+ maxDmv <<= 1;
+ }
+ }
+ if (iDMvVerX == 0 && iDMvVerY == 0)
+ blockHeight = height;
+ else
+ {
+ int maxDmv = std::max(abs(iDMvVerX), abs(iDMvVerY)) * blockHeight;
+ int TH = 1 << (iBit - 1); // Half pel
+ while (maxDmv < TH && blockHeight < height)
+ {
+ blockHeight <<= 1;
+ maxDmv <<= 1;
+ }
+ }
+#endif
+ int iMvScaleTmpHor0 = iMvScaleHor + ((iDMvHorX * blockWidth + iDMvVerX * blockHeight) >> 1);
+ int iMvScaleTmpVer0 = iMvScaleVer + ((iDMvHorY * blockWidth + iDMvVerY * blockHeight) >> 1);
+ // get prediction block by block
+ for (int h = 0; h < cxHeight; h += blockHeight)
+ {
+ for (int w = 0; w < cxWidth; w += blockWidth)
+ {
+ if (w == 0 || h == 0)
+ {
+ int iMvScaleTmpHor, iMvScaleTmpVer;
+
+#if !AFFINE_RM_CONSTRAINTS_AND_OPT
+ if (!subblkMVSpreadOverLimit)
+#endif
+ {
+ iMvScaleTmpHor = iMvScaleTmpHor0 + iDMvHorX * w + iDMvVerX * h;
+ iMvScaleTmpVer = iMvScaleTmpVer0 + iDMvHorY * w + iDMvVerY * h;
+ }
+#if !AFFINE_RM_CONSTRAINTS_AND_OPT
+ else
+ {
+ iMvScaleTmpHor = iMvScaleHor + iDMvHorX * (cxWidth >> 1) + iDMvVerX * (cxHeight >> 1);
+ iMvScaleTmpVer = iMvScaleVer + iDMvHorY * (cxWidth >> 1) + iDMvVerY * (cxHeight >> 1);
+ }
+#endif
+ roundAffineMv(iMvScaleTmpHor, iMvScaleTmpVer, shift);
+ Mv tmpMv(iMvScaleTmpHor, iMvScaleTmpVer);
+ tmpMv.clipToStorageBitDepth();
+ iMvScaleTmpHor = tmpMv.getHor();
+ iMvScaleTmpVer = tmpMv.getVer();
+
+ // clip and scale
+ if (refPic->isRefScaled(pu.cs->pps) == false)
+ {
+ clipMv(tmpMv, pu.lumaPos(), pu.lumaSize(), *pu.cs->sps, *pu.cs->pps);
+ iMvScaleTmpHor = tmpMv.getHor();
+ iMvScaleTmpVer = tmpMv.getVer();
+ }
+ xGetSublkAMLTemplate(*pu.cu, COMPONENT_Y, *refPic, Mv(iMvScaleTmpHor, iMvScaleTmpVer), blockWidth, blockHeight, w, h, numTemplate, refLeftTemplate, refAboveTemplate);
+ }
+ }
+ }
+}
+#endif
void InterPrediction::motionCompensation( PredictionUnit &pu, PelUnitBuf &predBuf, const RefPicList &eRefPicList
, const bool luma, const bool chroma
@@ -4192,6 +4475,734 @@ void InterPrediction::cacheAssign( CacheModel *cache )
}
#endif
+#if ARMC_TM
+void InterPrediction::adjustInterMergeCandidates(PredictionUnit &pu, MergeCtx& mrgCtx, int mrgCandIdx)
+{
+ uint32_t RdCandList[MRG_MAX_NUM_CANDS][MRG_MAX_NUM_CANDS];
+ Distortion candCostList[MRG_MAX_NUM_CANDS][MRG_MAX_NUM_CANDS];
+
+ for (uint32_t i = 0; i < MRG_MAX_NUM_CANDS; i++)
+ {
+ for (uint32_t j = 0; j < MRG_MAX_NUM_CANDS; j++)
+ {
+ RdCandList[i][j] = j;
+ candCostList[i][j] = MAX_UINT;
+ }
+ }
+
+ Distortion uiCost;
+
+ DistParam cDistParam;
+ cDistParam.applyWeight = false;
+
+ /*const SPS &sps = *pu.cs->sps;
+ Position puPos = pu.lumaPos();*/
+ int nWidth = pu.lumaSize().width;
+ int nHeight = pu.lumaSize().height;
+
+ if (!xAMLGetCurBlkTemplate(pu, nWidth, nHeight))
+ {
+ return;
+ }
+
+ for (uint32_t uiMergeCand = ((mrgCandIdx < 0) ? 0 : (mrgCandIdx / ADAPTIVE_SUB_GROUP_SIZE)*ADAPTIVE_SUB_GROUP_SIZE); uiMergeCand < (((mrgCandIdx < 0) || ((mrgCandIdx / ADAPTIVE_SUB_GROUP_SIZE + 1)*ADAPTIVE_SUB_GROUP_SIZE > mrgCtx.numValidMergeCand)) ? mrgCtx.numValidMergeCand : ((mrgCandIdx / ADAPTIVE_SUB_GROUP_SIZE + 1)*ADAPTIVE_SUB_GROUP_SIZE)); ++uiMergeCand)
+ {
+ bool firstGroup = (uiMergeCand / ADAPTIVE_SUB_GROUP_SIZE) == 0 ? true : false;
+ bool lastGroup = ((uiMergeCand / ADAPTIVE_SUB_GROUP_SIZE + 1)*ADAPTIVE_SUB_GROUP_SIZE >= mrgCtx.numValidMergeCand) ? true : false;
+ if (lastGroup && !firstGroup)
+ {
+ break;
+ }
+ uiCost = 0;
+
+ mrgCtx.setMergeInfo(pu, uiMergeCand);
+ PU::spanMotionInfo(pu, mrgCtx);
+
+ PelUnitBuf pcBufPredRefTop = (PelUnitBuf(pu.chromaFormat, PelBuf(m_acYuvRefAMLTemplate[0][0], nWidth, AML_MERGE_TEMPLATE_SIZE)));
+ PelUnitBuf pcBufPredCurTop = (PelUnitBuf(pu.chromaFormat, PelBuf(m_acYuvCurAMLTemplate[0][0], nWidth, AML_MERGE_TEMPLATE_SIZE)));
+ PelUnitBuf pcBufPredRefLeft = (PelUnitBuf(pu.chromaFormat, PelBuf(m_acYuvRefAMLTemplate[1][0], AML_MERGE_TEMPLATE_SIZE, nHeight)));
+ PelUnitBuf pcBufPredCurLeft = (PelUnitBuf(pu.chromaFormat, PelBuf(m_acYuvCurAMLTemplate[1][0], AML_MERGE_TEMPLATE_SIZE, nHeight)));
+
+ getBlkAMLRefTemplate(pu, pcBufPredRefTop, pcBufPredRefLeft);
+
+ if (m_bAMLTemplateAvailabe[0])
+ {
+ m_pcRdCost->setDistParam(cDistParam, pcBufPredCurTop.Y(), pcBufPredRefTop.Y(), pu.cs->sps->getBitDepth(CHANNEL_TYPE_LUMA), COMPONENT_Y, false);
+
+ uiCost += cDistParam.distFunc(cDistParam);
+ }
+
+ if (m_bAMLTemplateAvailabe[1])
+ {
+ m_pcRdCost->setDistParam(cDistParam, pcBufPredCurLeft.Y(), pcBufPredRefLeft.Y(), pu.cs->sps->getBitDepth(CHANNEL_TYPE_LUMA), COMPONENT_Y, false);
+
+ uiCost += cDistParam.distFunc(cDistParam);
+ }
+ updateCandList(uiMergeCand, uiCost, ADAPTIVE_SUB_GROUP_SIZE, RdCandList[uiMergeCand / ADAPTIVE_SUB_GROUP_SIZE], candCostList[uiMergeCand / ADAPTIVE_SUB_GROUP_SIZE]);
+ }
+ pu.mergeIdx = mrgCandIdx; //restore the merge index
+ updateCandInfo(mrgCtx, RdCandList
+ , mrgCandIdx
+ );
+
+}
+bool InterPrediction::xAMLGetCurBlkTemplate(PredictionUnit& pu, int nCurBlkWidth, int nCurBlkHeight)
+{
+ m_bAMLTemplateAvailabe[0] = xAMLIsTopTempAvailable(pu);
+ m_bAMLTemplateAvailabe[1] = xAMLIsLeftTempAvailable(pu);
+
+ if (!m_bAMLTemplateAvailabe[0] && !m_bAMLTemplateAvailabe[1])
+ {
+ return false;
+ }
+
+ /* const int lumaShift = 2 + MV_FRACTIONAL_BITS_DIFF;
+ const int horShift = (lumaShift + ::getComponentScaleX(COMPONENT_Y, pu.chromaFormat));
+ const int verShift = (lumaShift + ::getComponentScaleY(COMPONENT_Y, pu.chromaFormat));*/
+ const Picture& currPic = *pu.cs->picture;
+ const CPelBuf recBuf = currPic.getRecoBuf(pu.cs->picture->blocks[COMPONENT_Y]);
+ std::vector<Pel>& invLUT = m_pcReshape->getInvLUT();
+
+ if (m_bAMLTemplateAvailabe[0])
+ {
+ const Pel* rec = recBuf.bufAt(pu.blocks[COMPONENT_Y].pos().offset(0, -AML_MERGE_TEMPLATE_SIZE));
+ PelBuf pcYBuf = PelBuf(m_acYuvCurAMLTemplate[0][0], nCurBlkWidth, AML_MERGE_TEMPLATE_SIZE);
+ Pel* pcY = pcYBuf.bufAt(0, 0);
+ for (int k = 0; k < nCurBlkWidth; k++)
+ {
+ for (int l = 0; l < AML_MERGE_TEMPLATE_SIZE; l++)
+ {
+ int recVal = rec[k + l * recBuf.stride];
+
+ if (m_pcReshape->getSliceReshaperInfo().getUseSliceReshaper() && m_pcReshape->getCTUFlag())
+ {
+ recVal = invLUT[recVal];
+ }
+
+ pcY[k + l * nCurBlkWidth] = recVal;
+ }
+ }
+ }
+
+ if (m_bAMLTemplateAvailabe[1])
+ {
+ PelBuf pcYBuf = PelBuf(m_acYuvCurAMLTemplate[1][0], AML_MERGE_TEMPLATE_SIZE, nCurBlkHeight);
+ Pel* pcY = pcYBuf.bufAt(0, 0);
+ const Pel* rec = recBuf.bufAt(pu.blocks[COMPONENT_Y].pos().offset(-AML_MERGE_TEMPLATE_SIZE, 0));
+ for (int k = 0; k < nCurBlkHeight; k++)
+ {
+ for (int l = 0; l < AML_MERGE_TEMPLATE_SIZE; l++)
+ {
+ int recVal = rec[recBuf.stride * k + l];
+
+ if (m_pcReshape->getSliceReshaperInfo().getUseSliceReshaper() && m_pcReshape->getCTUFlag())
+ {
+ recVal = invLUT[recVal];
+ }
+
+ pcY[AML_MERGE_TEMPLATE_SIZE * k + l] = recVal;
+ }
+ }
+ }
+
+ return true;
+}
+
+bool InterPrediction::xAMLIsTopTempAvailable(PredictionUnit& pu)
+{
+ const CodingStructure &cs = *pu.cs;
+ Position posRT = pu.Y().topRight();
+ const PredictionUnit *puAbove = cs.getPURestricted(posRT.offset(0, -1), pu, pu.chType);
+
+ return (puAbove && pu.cu != puAbove->cu);
+}
+
+bool InterPrediction::xAMLIsLeftTempAvailable(PredictionUnit& pu)
+{
+ const CodingStructure &cs = *pu.cs;
+ Position posLB = pu.Y().bottomLeft();
+ const PredictionUnit *puLeft = cs.getPURestricted(posLB.offset(-1, 0), pu, pu.chType);
+
+ return (puLeft && pu.cu != puLeft->cu);
+}
+void InterPrediction::updateCandList(uint32_t uiCand, Distortion uiCost, uint32_t uiMrgCandNum, uint32_t* RdCandList, Distortion* CandCostList)
+{
+ uint32_t i;
+ uint32_t shift = 0;
+
+ while (shift < uiMrgCandNum && uiCost < CandCostList[uiMrgCandNum - 1 - shift])
+ {
+ shift++;
+ }
+
+ if (shift != 0)
+ {
+ for (i = 1; i < shift; i++)
+ {
+ RdCandList[uiMrgCandNum - i] = RdCandList[uiMrgCandNum - 1 - i];
+ CandCostList[uiMrgCandNum - i] = CandCostList[uiMrgCandNum - 1 - i];
+ }
+ RdCandList[uiMrgCandNum - shift] = uiCand;
+ CandCostList[uiMrgCandNum - shift] = uiCost;
+ }
+}
+void InterPrediction::updateCandInfo(MergeCtx& mrgCtx, uint32_t(*RdCandList)[MRG_MAX_NUM_CANDS], int mrgCandIdx)
+{
+ MergeCtx mrgCtxTmp;
+ for (uint32_t ui = 0; ui < MRG_MAX_NUM_CANDS; ++ui)
+ {
+ mrgCtxTmp.BcwIdx[ui] = BCW_DEFAULT;
+ mrgCtxTmp.interDirNeighbours[ui] = 0;
+ mrgCtxTmp.mvFieldNeighbours[(ui << 1)].refIdx = NOT_VALID;
+ mrgCtxTmp.mvFieldNeighbours[(ui << 1) + 1].refIdx = NOT_VALID;
+ mrgCtxTmp.useAltHpelIf[ui] = false;
+#if INTER_LIC
+ mrgCtxTmp.LICFlags[ui] = false;
+#endif
+#if MULTI_HYP_PRED
+ mrgCtxTmp.addHypNeighbours[ui].clear();
+#endif
+ }
+ for (uint32_t uiMergeCand = ((mrgCandIdx < 0) ? 0 : (mrgCandIdx / ADAPTIVE_SUB_GROUP_SIZE)*ADAPTIVE_SUB_GROUP_SIZE); uiMergeCand < (((mrgCandIdx < 0) || ((mrgCandIdx / ADAPTIVE_SUB_GROUP_SIZE + 1)*ADAPTIVE_SUB_GROUP_SIZE > mrgCtx.numValidMergeCand)) ? mrgCtx.numValidMergeCand : ((mrgCandIdx / ADAPTIVE_SUB_GROUP_SIZE + 1)*ADAPTIVE_SUB_GROUP_SIZE)); ++uiMergeCand)
+ {
+ bool firstGroup = (uiMergeCand / ADAPTIVE_SUB_GROUP_SIZE) == 0 ? true : false;
+ bool lastGroup = ((uiMergeCand / ADAPTIVE_SUB_GROUP_SIZE + 1)*ADAPTIVE_SUB_GROUP_SIZE >= mrgCtx.numValidMergeCand) ? true : false;
+ if (lastGroup && !firstGroup)
+ {
+ break;
+ }
+ mrgCtxTmp.BcwIdx[uiMergeCand] = mrgCtx.BcwIdx[uiMergeCand];
+ mrgCtxTmp.interDirNeighbours[uiMergeCand] = mrgCtx.interDirNeighbours[uiMergeCand];
+ mrgCtxTmp.mvFieldNeighbours[(uiMergeCand << 1)] = mrgCtx.mvFieldNeighbours[(uiMergeCand << 1)];
+ mrgCtxTmp.mvFieldNeighbours[(uiMergeCand << 1) + 1] = mrgCtx.mvFieldNeighbours[(uiMergeCand << 1) + 1];
+ mrgCtxTmp.useAltHpelIf[uiMergeCand] = mrgCtx.useAltHpelIf[uiMergeCand];
+#if INTER_LIC
+ mrgCtxTmp.LICFlags[uiMergeCand] = mrgCtx.LICFlags[uiMergeCand];
+#endif
+#if MULTI_HYP_PRED
+ mrgCtxTmp.addHypNeighbours[uiMergeCand] = mrgCtx.addHypNeighbours[uiMergeCand];
+#endif
+ }
+ //update
+ for (uint32_t uiMergeCand = ((mrgCandIdx < 0) ? 0 : (mrgCandIdx / ADAPTIVE_SUB_GROUP_SIZE)*ADAPTIVE_SUB_GROUP_SIZE); uiMergeCand < (((mrgCandIdx < 0) || ((mrgCandIdx / ADAPTIVE_SUB_GROUP_SIZE + 1)*ADAPTIVE_SUB_GROUP_SIZE > mrgCtx.numValidMergeCand)) ? mrgCtx.numValidMergeCand : ((mrgCandIdx / ADAPTIVE_SUB_GROUP_SIZE + 1)*ADAPTIVE_SUB_GROUP_SIZE)); ++uiMergeCand)
+ {
+ bool firstGroup = (uiMergeCand / ADAPTIVE_SUB_GROUP_SIZE) == 0 ? true : false;
+ bool lastGroup = ((uiMergeCand / ADAPTIVE_SUB_GROUP_SIZE + 1)*ADAPTIVE_SUB_GROUP_SIZE >= mrgCtx.numValidMergeCand) ? true : false;
+ if (lastGroup && !firstGroup)
+ {
+ break;
+ }
+
+ mrgCtx.BcwIdx[uiMergeCand] = mrgCtxTmp.BcwIdx[RdCandList[uiMergeCand / ADAPTIVE_SUB_GROUP_SIZE][uiMergeCand%ADAPTIVE_SUB_GROUP_SIZE]];
+ mrgCtx.interDirNeighbours[uiMergeCand] = mrgCtxTmp.interDirNeighbours[RdCandList[uiMergeCand / ADAPTIVE_SUB_GROUP_SIZE][uiMergeCand%ADAPTIVE_SUB_GROUP_SIZE]];
+ mrgCtx.mvFieldNeighbours[(uiMergeCand << 1)] = mrgCtxTmp.mvFieldNeighbours[(RdCandList[uiMergeCand / ADAPTIVE_SUB_GROUP_SIZE][uiMergeCand%ADAPTIVE_SUB_GROUP_SIZE] << 1)];
+ mrgCtx.mvFieldNeighbours[(uiMergeCand << 1) + 1] = mrgCtxTmp.mvFieldNeighbours[(RdCandList[uiMergeCand / ADAPTIVE_SUB_GROUP_SIZE][uiMergeCand%ADAPTIVE_SUB_GROUP_SIZE] << 1) + 1];
+ mrgCtx.useAltHpelIf[uiMergeCand] = mrgCtxTmp.useAltHpelIf[RdCandList[uiMergeCand / ADAPTIVE_SUB_GROUP_SIZE][uiMergeCand%ADAPTIVE_SUB_GROUP_SIZE]];
+#if INTER_LIC
+ mrgCtx.LICFlags[uiMergeCand] = mrgCtxTmp.LICFlags[RdCandList[uiMergeCand / ADAPTIVE_SUB_GROUP_SIZE][uiMergeCand%ADAPTIVE_SUB_GROUP_SIZE]];
+#endif
+#if MULTI_HYP_PRED
+ mrgCtx.addHypNeighbours[uiMergeCand] = mrgCtxTmp.addHypNeighbours[RdCandList[uiMergeCand / ADAPTIVE_SUB_GROUP_SIZE][uiMergeCand%ADAPTIVE_SUB_GROUP_SIZE]];
+#endif
+ }
+}
+void InterPrediction::getBlkAMLRefTemplate(PredictionUnit &pu, PelUnitBuf &pcBufPredRefTop, PelUnitBuf &pcBufPredRefLeft)
+{
+ Mv mvCurr;
+ const int lumaShift = 2 + MV_FRACTIONAL_BITS_DIFF;
+ const int horShift = (lumaShift + ::getComponentScaleX(COMPONENT_Y, pu.chromaFormat));
+ const int verShift = (lumaShift + ::getComponentScaleY(COMPONENT_Y, pu.chromaFormat));
+
+ if (xCheckIdenticalMotion(pu))
+ {
+ mvCurr = pu.mv[0];
+ /*const int horIntMv = (mvCurr.getHor() + ((1 << horShift) >> 1)) >> horShift;
+ const int verIntMv = (mvCurr.getVer() + ((1 << verShift) >> 1)) >> verShift;
+ Mv subPelMv(horIntMv << horShift, verIntMv << verShift);*/
+ Mv subPelMv = mvCurr;
+ clipMv(mvCurr, pu.lumaPos(), pu.lumaSize(), *pu.cs->sps, *pu.cs->pps);
+ CHECK(pu.refIdx[0] < 0, "invalid ref idx");
+
+ if (m_bAMLTemplateAvailabe[0])
+ {
+ Mv mvTop(0, -(AML_MERGE_TEMPLATE_SIZE << verShift));
+ mvTop += subPelMv;
+
+ clipMv(mvTop, pu.lumaPos(), pu.lumaSize(), *pu.cs->sps, *pu.cs->pps);
+
+#if INTER_LIC
+ xPredInterBlk(COMPONENT_Y, pu, pu.cu->slice->getRefPic(REF_PIC_LIST_0, pu.refIdx[0]), mvTop, pcBufPredRefTop,
+ false, pu.cu->slice->clpRng(COMPONENT_Y), false, false, SCALE_1X, 0, 0, false, NULL, 0, true, true,
+ mvCurr);
+#else
+ xPredInterBlk(COMPONENT_Y, pu, pu.cu->slice->getRefPic(REF_PIC_LIST_0, pu.refIdx[0]), mvTop, pcBufPredRefTop,
+ false, pu.cu->slice->clpRng(COMPONENT_Y), false, false, SCALE_1X, 0, 0, false, NULL, 0, true);
+#endif
+ }
+ if (m_bAMLTemplateAvailabe[1])
+ {
+ Mv mvLeft(-(AML_MERGE_TEMPLATE_SIZE << horShift), 0);
+ mvLeft += subPelMv;
+
+ clipMv(mvLeft, pu.lumaPos(), pu.lumaSize(), *pu.cs->sps, *pu.cs->pps);
+
+#if INTER_LIC
+ xPredInterBlk(COMPONENT_Y, pu, pu.cu->slice->getRefPic(REF_PIC_LIST_0, pu.refIdx[0]), mvLeft, pcBufPredRefLeft,
+ false, pu.cu->slice->clpRng(COMPONENT_Y), false, false, SCALE_1X, 0, 0, false, NULL, 0, true, true,
+ mvCurr);
+#else
+ xPredInterBlk(COMPONENT_Y, pu, pu.cu->slice->getRefPic(REF_PIC_LIST_0, pu.refIdx[0]), mvLeft, pcBufPredRefLeft,
+ false, pu.cu->slice->clpRng(COMPONENT_Y), false, false, SCALE_1X, 0, 0, false, NULL, 0, true);
+#endif
+ }
+ }
+ else
+ {
+ for (uint32_t refList = 0; refList < NUM_REF_PIC_LIST_01; refList++)
+ {
+ if (pu.refIdx[refList] < 0)
+ {
+ continue;
+ }
+ RefPicList eRefPicList = (refList ? REF_PIC_LIST_1 : REF_PIC_LIST_0);
+ CHECK(pu.refIdx[refList] >= pu.cu->slice->getNumRefIdx(eRefPicList), "Invalid reference index");
+
+ m_iRefListIdx = refList;
+ mvCurr = pu.mv[refList];
+ /*const int horIntMv = (mvCurr.getHor() + ((1 << horShift) >> 1)) >> horShift;
+ const int verIntMv = (mvCurr.getVer() + ((1 << verShift) >> 1)) >> verShift;
+ Mv subPelMv(horIntMv << horShift, verIntMv << verShift);*/
+ Mv subPelMv = mvCurr;
+ clipMv(mvCurr, pu.lumaPos(), pu.lumaSize(), *pu.cs->sps, *pu.cs->pps);
+
+ if (m_bAMLTemplateAvailabe[0])
+ {
+ Mv mvTop(0, -(AML_MERGE_TEMPLATE_SIZE << verShift));
+ mvTop += subPelMv;
+
+ clipMv(mvTop, pu.lumaPos(), pu.lumaSize(), *pu.cs->sps, *pu.cs->pps);
+
+ PelUnitBuf pcMbBuf =
+ PelUnitBuf(pu.chromaFormat, PelBuf(m_acYuvRefAboveTemplate[refList][0], pcBufPredRefTop.Y()));
+
+ if (pu.refIdx[0] >= 0 && pu.refIdx[1] >= 0)
+ {
+#if INTER_LIC
+ xPredInterBlk(COMPONENT_Y, pu, pu.cu->slice->getRefPic(eRefPicList, pu.refIdx[refList]), mvTop, pcMbBuf, true,
+ pu.cu->slice->clpRng(COMPONENT_Y), false, false, SCALE_1X, 0, 0, false, NULL, 0, true, true,
+ mvCurr);
+#else
+ xPredInterBlk(COMPONENT_Y, pu, pu.cu->slice->getRefPic(eRefPicList, pu.refIdx[refList]), mvTop, pcMbBuf, true,
+ pu.cu->slice->clpRng(COMPONENT_Y), false, false, SCALE_1X, 0, 0, false, NULL, 0, true);
+#endif
+ }
+ else
+ {
+#if INTER_LIC
+ xPredInterBlk(COMPONENT_Y, pu, pu.cu->slice->getRefPic(eRefPicList, pu.refIdx[refList]), mvTop, pcMbBuf,
+ false, pu.cu->slice->clpRng(COMPONENT_Y), false, false, SCALE_1X, 0, 0, false, NULL, 0, true,
+ true, mvCurr);
+#else
+ xPredInterBlk(COMPONENT_Y, pu, pu.cu->slice->getRefPic(eRefPicList, pu.refIdx[refList]), mvTop, pcMbBuf,
+ false, pu.cu->slice->clpRng(COMPONENT_Y), false, false, SCALE_1X, 0, 0, false, NULL, 0, true);
+#endif
+ }
+ }
+ if (m_bAMLTemplateAvailabe[1])
+ {
+ Mv mvLeft(-(AML_MERGE_TEMPLATE_SIZE << horShift), 0);
+ mvLeft += subPelMv;
+
+ clipMv(mvLeft, pu.lumaPos(), pu.lumaSize(), *pu.cs->sps, *pu.cs->pps);
+
+ PelUnitBuf pcMbBuf =
+ PelUnitBuf(pu.chromaFormat, PelBuf(m_acYuvRefLeftTemplate[refList][0], pcBufPredRefLeft.Y()));
+
+ if (pu.refIdx[0] >= 0 && pu.refIdx[1] >= 0)
+ {
+#if INTER_LIC
+ xPredInterBlk(COMPONENT_Y, pu, pu.cu->slice->getRefPic(eRefPicList, pu.refIdx[refList]), mvLeft, pcMbBuf,
+ true, pu.cu->slice->clpRng(COMPONENT_Y), false, false, SCALE_1X, 0, 0, false, NULL, 0, true,
+ true, mvCurr);
+#else
+ xPredInterBlk(COMPONENT_Y, pu, pu.cu->slice->getRefPic(eRefPicList, pu.refIdx[refList]), mvLeft, pcMbBuf,
+ true, pu.cu->slice->clpRng(COMPONENT_Y), false, false, SCALE_1X, 0, 0, false, NULL, 0, true);
+#endif
+ }
+ else
+ {
+#if INTER_LIC
+ xPredInterBlk(COMPONENT_Y, pu, pu.cu->slice->getRefPic(eRefPicList, pu.refIdx[refList]), mvLeft, pcMbBuf,
+ false, pu.cu->slice->clpRng(COMPONENT_Y), false, false, SCALE_1X, 0, 0, false, NULL, 0, true,
+ true, mvCurr);
+#else
+ xPredInterBlk(COMPONENT_Y, pu, pu.cu->slice->getRefPic(eRefPicList, pu.refIdx[refList]), mvLeft, pcMbBuf,
+ false, pu.cu->slice->clpRng(COMPONENT_Y), false, false, SCALE_1X, 0, 0, false, NULL, 0, true);
+#endif
+ }
+ }
+ }
+ if (m_bAMLTemplateAvailabe[0])
+ {
+ CPelUnitBuf srcPred0 = CPelUnitBuf(pu.chromaFormat, PelBuf(m_acYuvRefAboveTemplate[0][0], pcBufPredRefTop.Y()));
+ CPelUnitBuf srcPred1 = CPelUnitBuf(pu.chromaFormat, PelBuf(m_acYuvRefAboveTemplate[1][0], pcBufPredRefTop.Y()));
+ xWeightedAverageY(pu, srcPred0, srcPred1, pcBufPredRefTop, pu.cu->slice->getSPS()->getBitDepths(),
+ pu.cu->slice->clpRngs());
+ }
+ if (m_bAMLTemplateAvailabe[1])
+ {
+ CPelUnitBuf srcPred0 = CPelUnitBuf(pu.chromaFormat, PelBuf(m_acYuvRefLeftTemplate[0][0], pcBufPredRefLeft.Y()));
+ CPelUnitBuf srcPred1 = CPelUnitBuf(pu.chromaFormat, PelBuf(m_acYuvRefLeftTemplate[1][0], pcBufPredRefLeft.Y()));
+ xWeightedAverageY(pu, srcPred0, srcPred1, pcBufPredRefLeft, pu.cu->slice->getSPS()->getBitDepths(),
+ pu.cu->slice->clpRngs());
+ }
+ }
+}
+
+void InterPrediction::adjustAffineMergeCandidates(PredictionUnit &pu, AffineMergeCtx& affMrgCtx, int mrgCandIdx)
+{
+ uint32_t RdCandList[AFFINE_MRG_MAX_NUM_CANDS][AFFINE_MRG_MAX_NUM_CANDS];
+ Distortion candCostList[AFFINE_MRG_MAX_NUM_CANDS][AFFINE_MRG_MAX_NUM_CANDS];
+
+ for (uint32_t i = 0; i < AFFINE_MRG_MAX_NUM_CANDS; i++)
+ {
+ for (uint32_t j = 0; j < AFFINE_MRG_MAX_NUM_CANDS; j++)
+ {
+ RdCandList[i][j] = j;
+ candCostList[i][j] = MAX_UINT;
+ }
+ }
+
+ Distortion uiCost;
+
+ DistParam cDistParam;
+ cDistParam.applyWeight = false;
+
+ int nWidth = pu.lumaSize().width;
+ int nHeight = pu.lumaSize().height;
+
+ if (!xAMLGetCurBlkTemplate(pu, nWidth, nHeight))
+ {
+ return;
+ }
+
+ for (uint32_t uiMergeCand = ((mrgCandIdx < 0) ? 0 : (mrgCandIdx / ADAPTIVE_AFFINE_SUB_GROUP_SIZE)*ADAPTIVE_AFFINE_SUB_GROUP_SIZE); uiMergeCand < (((mrgCandIdx < 0) || ((mrgCandIdx / ADAPTIVE_AFFINE_SUB_GROUP_SIZE + 1)*ADAPTIVE_AFFINE_SUB_GROUP_SIZE > affMrgCtx.maxNumMergeCand)) ? affMrgCtx.maxNumMergeCand : ((mrgCandIdx / ADAPTIVE_AFFINE_SUB_GROUP_SIZE + 1)*ADAPTIVE_AFFINE_SUB_GROUP_SIZE)); ++uiMergeCand)
+ {
+ bool firstGroup = (uiMergeCand / ADAPTIVE_AFFINE_SUB_GROUP_SIZE) == 0 ? true : false;
+ bool lastGroup = ((uiMergeCand / ADAPTIVE_AFFINE_SUB_GROUP_SIZE + 1)*ADAPTIVE_AFFINE_SUB_GROUP_SIZE >= affMrgCtx.maxNumMergeCand) ? true : false;
+ if (lastGroup && !firstGroup)
+ {
+ break;
+ }
+ uiCost = 0;
+
+ // set merge information
+ pu.interDir = affMrgCtx.interDirNeighbours[uiMergeCand];
+ pu.mergeFlag = true;
+ pu.regularMergeFlag = false;
+ pu.mergeIdx = uiMergeCand;
+ pu.cu->affine = true;
+ pu.cu->affineType = affMrgCtx.affineType[uiMergeCand];
+#if AFFINE_MMVD
+ pu.afMmvdFlag = false;
+#endif
+ pu.cu->BcwIdx = affMrgCtx.BcwIdx[uiMergeCand];
+#if INTER_LIC
+ pu.cu->LICFlag = affMrgCtx.LICFlags[uiMergeCand];
+#endif
+
+ pu.mergeType = affMrgCtx.mergeType[uiMergeCand];
+ if (pu.mergeType == MRG_TYPE_SUBPU_ATMVP)
+ {
+ pu.refIdx[0] = affMrgCtx.mvFieldNeighbours[(uiMergeCand << 1) + 0][0].refIdx;
+ pu.refIdx[1] = affMrgCtx.mvFieldNeighbours[(uiMergeCand << 1) + 1][0].refIdx;
+ PU::spanMotionInfo(pu, *affMrgCtx.mrgCtx);
+ }
+ else
+ {
+ for (uint32_t refList = 0; refList < NUM_REF_PIC_LIST_01; refList++)
+ {
+ for (int i = 0; i < 3; i++)
+ {
+ pu.mvAffi[refList][i] = affMrgCtx.mvFieldNeighbours[(uiMergeCand << 1) + refList][i].mv;
+ }
+ pu.refIdx[refList] = affMrgCtx.mvFieldNeighbours[(uiMergeCand << 1) + refList][0].refIdx;
+ }
+
+ PelUnitBuf pcBufPredRefTop = (PelUnitBuf(pu.chromaFormat, PelBuf(m_acYuvRefAMLTemplate[0][0], nWidth, AML_MERGE_TEMPLATE_SIZE)));
+ PelUnitBuf pcBufPredCurTop = (PelUnitBuf(pu.chromaFormat, PelBuf(m_acYuvCurAMLTemplate[0][0], nWidth, AML_MERGE_TEMPLATE_SIZE)));
+ PelUnitBuf pcBufPredRefLeft = (PelUnitBuf(pu.chromaFormat, PelBuf(m_acYuvRefAMLTemplate[1][0], AML_MERGE_TEMPLATE_SIZE, nHeight)));
+ PelUnitBuf pcBufPredCurLeft = (PelUnitBuf(pu.chromaFormat, PelBuf(m_acYuvCurAMLTemplate[1][0], AML_MERGE_TEMPLATE_SIZE, nHeight)));
+
+ getAffAMLRefTemplate(pu, pcBufPredRefTop, pcBufPredRefLeft);
+
+ if (m_bAMLTemplateAvailabe[0])
+ {
+ m_pcRdCost->setDistParam(cDistParam, pcBufPredCurTop.Y(), pcBufPredRefTop.Y(), pu.cs->sps->getBitDepth(CHANNEL_TYPE_LUMA), COMPONENT_Y, false);
+
+ uiCost += cDistParam.distFunc(cDistParam);
+ }
+
+ if (m_bAMLTemplateAvailabe[1])
+ {
+ m_pcRdCost->setDistParam(cDistParam, pcBufPredCurLeft.Y(), pcBufPredRefLeft.Y(), pu.cs->sps->getBitDepth(CHANNEL_TYPE_LUMA), COMPONENT_Y, false);
+
+ uiCost += cDistParam.distFunc(cDistParam);
+ }
+ }
+ updateCandList(uiMergeCand, uiCost, ADAPTIVE_AFFINE_SUB_GROUP_SIZE, RdCandList[uiMergeCand / ADAPTIVE_AFFINE_SUB_GROUP_SIZE], candCostList[uiMergeCand / ADAPTIVE_AFFINE_SUB_GROUP_SIZE]);
+ }
+ pu.mergeIdx = mrgCandIdx; //restore the merge index
+ updateAffineCandInfo(pu, affMrgCtx, RdCandList
+ , mrgCandIdx
+ );
+
+}
+void InterPrediction::updateAffineCandInfo(PredictionUnit &pu, AffineMergeCtx& affMrgCtx, uint32_t(*RdCandList)[AFFINE_MRG_MAX_NUM_CANDS], int mrgCandIdx)
+{
+ AffineMergeCtx affMrgCtxTmp;
+ const uint32_t maxNumAffineMergeCand = pu.cs->slice->getPicHeader()->getMaxNumAffineMergeCand();
+ for (int i = 0; i < maxNumAffineMergeCand; i++)
+ {
+ for (int mvNum = 0; mvNum < 3; mvNum++)
+ {
+ affMrgCtxTmp.mvFieldNeighbours[(i << 1) + 0][mvNum].setMvField(Mv(), -1);
+ affMrgCtxTmp.mvFieldNeighbours[(i << 1) + 1][mvNum].setMvField(Mv(), -1);
+ }
+ affMrgCtxTmp.interDirNeighbours[i] = 0;
+ affMrgCtxTmp.affineType[i] = AFFINEMODEL_4PARAM;
+ affMrgCtxTmp.mergeType[i] = MRG_TYPE_DEFAULT_N;
+ affMrgCtxTmp.BcwIdx[i] = BCW_DEFAULT;
+#if INTER_LIC
+ affMrgCtxTmp.LICFlags[i] = false;
+#endif
+ }
+ for (uint32_t uiMergeCand = ((mrgCandIdx < 0) ? 0 : (mrgCandIdx / ADAPTIVE_AFFINE_SUB_GROUP_SIZE)*ADAPTIVE_AFFINE_SUB_GROUP_SIZE); uiMergeCand < (((mrgCandIdx < 0) || ((mrgCandIdx / ADAPTIVE_AFFINE_SUB_GROUP_SIZE + 1)*ADAPTIVE_AFFINE_SUB_GROUP_SIZE > affMrgCtx.maxNumMergeCand)) ? affMrgCtx.maxNumMergeCand : ((mrgCandIdx / ADAPTIVE_AFFINE_SUB_GROUP_SIZE + 1)*ADAPTIVE_AFFINE_SUB_GROUP_SIZE)); ++uiMergeCand)
+ {
+ bool firstGroup = (uiMergeCand / ADAPTIVE_AFFINE_SUB_GROUP_SIZE) == 0 ? true : false;
+ bool lastGroup = ((uiMergeCand / ADAPTIVE_AFFINE_SUB_GROUP_SIZE + 1)*ADAPTIVE_AFFINE_SUB_GROUP_SIZE >= affMrgCtx.maxNumMergeCand) ? true : false;
+ if (lastGroup && !firstGroup)
+ {
+ break;
+ }
+ for (int mvNum = 0; mvNum < 3; mvNum++)
+ {
+ affMrgCtxTmp.mvFieldNeighbours[(uiMergeCand << 1) + 0][mvNum] = affMrgCtx.mvFieldNeighbours[(uiMergeCand << 1) + 0][mvNum];
+ affMrgCtxTmp.mvFieldNeighbours[(uiMergeCand << 1) + 1][mvNum] = affMrgCtx.mvFieldNeighbours[(uiMergeCand << 1) + 1][mvNum];
+ }
+ affMrgCtxTmp.interDirNeighbours[uiMergeCand] = affMrgCtx.interDirNeighbours[uiMergeCand];
+ affMrgCtxTmp.affineType[uiMergeCand] = affMrgCtx.affineType[uiMergeCand];
+ affMrgCtxTmp.mergeType[uiMergeCand] = affMrgCtx.mergeType[uiMergeCand];
+ affMrgCtxTmp.BcwIdx[uiMergeCand] = affMrgCtx.BcwIdx[uiMergeCand];
+#if INTER_LIC
+ affMrgCtxTmp.LICFlags[uiMergeCand] = affMrgCtx.LICFlags[uiMergeCand];
+#endif
+ }
+ //update
+ for (uint32_t uiMergeCand = ((mrgCandIdx < 0) ? 0 : (mrgCandIdx / ADAPTIVE_AFFINE_SUB_GROUP_SIZE)*ADAPTIVE_AFFINE_SUB_GROUP_SIZE); uiMergeCand < (((mrgCandIdx < 0) || ((mrgCandIdx / ADAPTIVE_AFFINE_SUB_GROUP_SIZE + 1)*ADAPTIVE_AFFINE_SUB_GROUP_SIZE > affMrgCtx.maxNumMergeCand)) ? affMrgCtx.maxNumMergeCand : ((mrgCandIdx / ADAPTIVE_AFFINE_SUB_GROUP_SIZE + 1)*ADAPTIVE_AFFINE_SUB_GROUP_SIZE)); ++uiMergeCand)
+ {
+ bool firstGroup = (uiMergeCand / ADAPTIVE_AFFINE_SUB_GROUP_SIZE) == 0 ? true : false;
+ bool lastGroup = ((uiMergeCand / ADAPTIVE_AFFINE_SUB_GROUP_SIZE + 1)*ADAPTIVE_AFFINE_SUB_GROUP_SIZE >= affMrgCtx.maxNumMergeCand) ? true : false;
+ if (lastGroup && !firstGroup)
+ {
+ break;
+ }
+ for (int mvNum = 0; mvNum < 3; mvNum++)
+ {
+ affMrgCtx.mvFieldNeighbours[(uiMergeCand << 1) + 0][mvNum] = affMrgCtxTmp.mvFieldNeighbours[(RdCandList[uiMergeCand / ADAPTIVE_AFFINE_SUB_GROUP_SIZE][uiMergeCand%ADAPTIVE_AFFINE_SUB_GROUP_SIZE] << 1) + 0][mvNum];
+ affMrgCtx.mvFieldNeighbours[(uiMergeCand << 1) + 1][mvNum] = affMrgCtxTmp.mvFieldNeighbours[(RdCandList[uiMergeCand / ADAPTIVE_AFFINE_SUB_GROUP_SIZE][uiMergeCand%ADAPTIVE_AFFINE_SUB_GROUP_SIZE] << 1) + 1][mvNum];
+ }
+ affMrgCtx.interDirNeighbours[uiMergeCand] = affMrgCtxTmp.interDirNeighbours[RdCandList[uiMergeCand / ADAPTIVE_AFFINE_SUB_GROUP_SIZE][uiMergeCand%ADAPTIVE_AFFINE_SUB_GROUP_SIZE]];
+ affMrgCtx.affineType[uiMergeCand] = affMrgCtxTmp.affineType[RdCandList[uiMergeCand / ADAPTIVE_AFFINE_SUB_GROUP_SIZE][uiMergeCand%ADAPTIVE_AFFINE_SUB_GROUP_SIZE]];
+ affMrgCtx.mergeType[uiMergeCand] = affMrgCtxTmp.mergeType[RdCandList[uiMergeCand / ADAPTIVE_AFFINE_SUB_GROUP_SIZE][uiMergeCand%ADAPTIVE_AFFINE_SUB_GROUP_SIZE]];
+ affMrgCtx.BcwIdx[uiMergeCand] = affMrgCtxTmp.BcwIdx[RdCandList[uiMergeCand / ADAPTIVE_AFFINE_SUB_GROUP_SIZE][uiMergeCand%ADAPTIVE_AFFINE_SUB_GROUP_SIZE]];
+#if INTER_LIC
+ affMrgCtx.LICFlags[uiMergeCand] = affMrgCtxTmp.LICFlags[RdCandList[uiMergeCand / ADAPTIVE_AFFINE_SUB_GROUP_SIZE][uiMergeCand%ADAPTIVE_AFFINE_SUB_GROUP_SIZE]];
+#endif
+ }
+}
+void InterPrediction::xGetSublkAMLTemplate(const CodingUnit& cu,
+ const ComponentID compID,
+ const Picture& refPic,
+ const Mv& mv,
+ const int sublkWidth,
+ const int sublkHeight,
+ const int posW,
+ const int posH,
+ int* numTemplate,
+ Pel* refLeftTemplate,
+ Pel* refAboveTemplate)
+{
+ const int bitDepth = cu.cs->sps->getBitDepth(toChannelType(compID));
+ const int precShift = std::max(0, bitDepth - 12);
+
+ const CodingUnit* const cuAbove = cu.cs->getCU(cu.blocks[compID].pos().offset(0, -1), toChannelType(compID));
+ const CodingUnit* const cuLeft = cu.cs->getCU(cu.blocks[compID].pos().offset(-1, 0), toChannelType(compID));
+ const CPelBuf refBuf = cuAbove || cuLeft ? refPic.getRecoBuf(refPic.blocks[compID]) : CPelBuf();
+
+ // above
+ if (cuAbove && posH == 0)
+ {
+ xGetPredBlkTpl<true>(cu, compID, refBuf, mv, posW, posH, sublkWidth, refAboveTemplate);
+
+ for (int k = posW; k < posW + sublkWidth; k++)
+ {
+ int refVal = refAboveTemplate[k];
+ refVal >>= precShift;
+ refAboveTemplate[k] = refVal;
+ numTemplate[0]++;
+ }
+ }
+
+ // left
+ if (cuLeft && posW == 0)
+ {
+ xGetPredBlkTpl<false>(cu, compID, refBuf, mv, posW, posH, sublkHeight, refLeftTemplate);
+
+ for (int k = posH; k < posH + sublkHeight; k++)
+ {
+ int refVal = refLeftTemplate[k];
+ refVal >>= precShift;
+ refLeftTemplate[k] = refVal;
+ numTemplate[1]++;
+ }
+ }
+}
+void InterPrediction::getAffAMLRefTemplate(PredictionUnit &pu, PelUnitBuf &pcBufPredRefTop, PelUnitBuf &pcBufPredRefLeft)
+{
+#if INTER_LIC
+ int LICshift[2] = { 0 };
+ int scale[2] = { 0 };
+ int offset[2] = { 0 };
+#endif
+ const int bitDepth = pu.cs->sps->getBitDepth(CHANNEL_TYPE_LUMA);
+ if (xCheckIdenticalMotion(pu))
+ {
+ Pel * refLeftTemplate = m_acYuvRefAMLTemplate[1][0];
+ Pel * refAboveTemplate = m_acYuvRefAMLTemplate[0][0];
+ int numTemplate[2] = { 0, 0 }; // 0:Above, 1:Left
+ const RefPicList eRefPicList = REF_PIC_LIST_0;
+ xPredAffineTpl(pu, eRefPicList, numTemplate, refLeftTemplate, refAboveTemplate);
+#if INTER_LIC
+ if (pu.cu->LICFlag)
+ {
+ Pel *recLeftTemplate = m_acYuvCurAMLTemplate[1][0];
+ Pel *recAboveTemplate = m_acYuvCurAMLTemplate[0][0];
+ xGetLICParamGeneral(*pu.cu, COMPONENT_Y, numTemplate, refLeftTemplate, refAboveTemplate, recLeftTemplate,
+ recAboveTemplate, LICshift[0], scale[0], offset[0]);
+ if (m_bAMLTemplateAvailabe[0])
+ {
+ PelBuf & dstBuf = pcBufPredRefTop.bufs[0];
+ const ClpRng &clpRng = pu.cu->cs->slice->clpRng(COMPONENT_Y);
+ dstBuf.linearTransform(scale[0], LICshift[0], offset[0], true, clpRng);
+ }
+ if (m_bAMLTemplateAvailabe[1])
+ {
+ PelBuf & dstBuf = pcBufPredRefLeft.bufs[0];
+ const ClpRng &clpRng = pu.cu->cs->slice->clpRng(COMPONENT_Y);
+ dstBuf.linearTransform(scale[0], LICshift[0], offset[0], true, clpRng);
+ }
+ }
+#endif
+ }
+ else
+ {
+ for (uint32_t refList = 0; refList < NUM_REF_PIC_LIST_01; refList++)
+ {
+ if (pu.refIdx[refList] < 0)
+ {
+ continue;
+ }
+ RefPicList eRefPicList = (refList ? REF_PIC_LIST_1 : REF_PIC_LIST_0);
+ CHECK(pu.refIdx[refList] >= pu.cu->slice->getNumRefIdx(eRefPicList), "Invalid reference index");
+ Pel *refLeftTemplate = m_acYuvRefLeftTemplate[refList][0];
+ Pel *refAboveTemplate = m_acYuvRefAboveTemplate[refList][0];
+ int numTemplate[2] = { 0, 0 }; // 0:Above, 1:Left
+ xPredAffineTpl(pu, eRefPicList, numTemplate, refLeftTemplate, refAboveTemplate);
+#if INTER_LIC
+ if (pu.cu->LICFlag)
+ {
+ Pel *recLeftTemplate = m_acYuvCurAMLTemplate[1][0];
+ Pel *recAboveTemplate = m_acYuvCurAMLTemplate[0][0];
+ xGetLICParamGeneral(*pu.cu, COMPONENT_Y, numTemplate, refLeftTemplate, refAboveTemplate, recLeftTemplate,
+ recAboveTemplate, LICshift[refList], scale[refList], offset[refList]);
+ }
+#endif
+ }
+ if (m_bAMLTemplateAvailabe[0])
+ {
+ PelUnitBuf srcPred[2];
+ srcPred[0] = PelUnitBuf(pu.chromaFormat, PelBuf(m_acYuvRefAboveTemplate[0][0], pcBufPredRefTop.Y()));
+ srcPred[1] = PelUnitBuf(pu.chromaFormat, PelBuf(m_acYuvRefAboveTemplate[1][0], pcBufPredRefTop.Y()));
+#if INTER_LIC
+ if (pu.cu->LICFlag)
+ {
+ for (int i = 0; i < 2; i++)
+ {
+ if (pu.refIdx[i] >= 0)
+ {
+ PelBuf & dstBuf = srcPred[i].bufs[0];
+ const ClpRng &clpRng = pu.cu->cs->slice->clpRng(COMPONENT_Y);
+ dstBuf.linearTransform(scale[i], LICshift[i], offset[i], true, clpRng);
+ }
+ }
+ }
+#endif
+ const int iRefIdx0 = pu.refIdx[0];
+ const int iRefIdx1 = pu.refIdx[1];
+ if (iRefIdx0 >= 0 && iRefIdx1 >= 0)
+ {
+ for (int i = 0; i < 2; i++)
+ {
+ PelBuf & dstBuf = srcPred[i].bufs[0];
+ const int biShift = IF_INTERNAL_PREC - bitDepth;
+ const Pel biOffset = -IF_INTERNAL_OFFS;
+ ClpRng clpRngDummy;
+ dstBuf.linearTransform(1, -biShift, biOffset, false, clpRngDummy);
+ }
+ }
+ xWeightedAverageY(pu, srcPred[0], srcPred[1], pcBufPredRefTop, pu.cu->slice->getSPS()->getBitDepths(),
+ pu.cu->slice->clpRngs());
+ }
+ if (m_bAMLTemplateAvailabe[1])
+ {
+ PelUnitBuf srcPred[2];
+ srcPred[0] = PelUnitBuf(pu.chromaFormat, PelBuf(m_acYuvRefLeftTemplate[0][0], pcBufPredRefLeft.Y()));
+ srcPred[1] = PelUnitBuf(pu.chromaFormat, PelBuf(m_acYuvRefLeftTemplate[1][0], pcBufPredRefLeft.Y()));
+#if INTER_LIC
+ if (pu.cu->LICFlag)
+ {
+ for (int i = 0; i < 2; i++)
+ {
+ if (pu.refIdx[i] >= 0)
+ {
+ PelBuf & dstBuf = srcPred[i].bufs[0];
+ const ClpRng &clpRng = pu.cu->cs->slice->clpRng(COMPONENT_Y);
+ dstBuf.linearTransform(scale[i], LICshift[i], offset[i], true, clpRng);
+ }
+ }
+ }
+#endif
+ const int iRefIdx0 = pu.refIdx[0];
+ const int iRefIdx1 = pu.refIdx[1];
+ if (iRefIdx0 >= 0 && iRefIdx1 >= 0)
+ {
+ for (int i = 0; i < 2; i++)
+ {
+ PelBuf & dstBuf = srcPred[i].bufs[0];
+ const int biShift = IF_INTERNAL_PREC - bitDepth;
+ const Pel biOffset = -IF_INTERNAL_OFFS;
+ ClpRng clpRngDummy;
+ dstBuf.linearTransform(1, -biShift, biOffset, false, clpRngDummy);
+ }
+ }
+ xWeightedAverageY(pu, srcPred[0], srcPred[1], pcBufPredRefLeft, pu.cu->slice->getSPS()->getBitDepths(),
+ pu.cu->slice->clpRngs());
+ }
+ }
+}
+#endif
+
void InterPrediction::xFillIBCBuffer(CodingUnit &cu)
{
for (auto &currPU : CU::traverseTUs(cu))
diff --git a/source/Lib/CommonLib/InterPrediction.h b/source/Lib/CommonLib/InterPrediction.h
index 0333d844882cb197cc72f69d681d7e2d92bcc57b..e3a13e149d5df0ef4c2ac9e18857f12d96191376 100644
--- a/source/Lib/CommonLib/InterPrediction.h
+++ b/source/Lib/CommonLib/InterPrediction.h
@@ -51,7 +51,7 @@
#include "ContextModelling.h"
// forward declaration
class Mv;
-#if INTER_LIC || (TM_AMVP || TM_MRG)
+#if INTER_LIC || (TM_AMVP || TM_MRG) || ARMC_TM
class Reshape;
#endif
@@ -70,7 +70,7 @@ public:
PelUnitBuf m_predictionBeforeLIC;
bool m_storeBeforeLIC;
#endif
-#if INTER_LIC || (TM_AMVP || TM_MRG) // note: already refactor
+#if INTER_LIC || (TM_AMVP || TM_MRG) || ARMC_TM // note: already refactor
Reshape* m_pcReshape;
#endif
@@ -216,6 +216,13 @@ protected:
, bool bilinearMC = false
, Pel *srcPadBuf = NULL
, int32_t srcPadStride = 0
+#if ARMC_TM
+ , bool AML = false
+#if INTER_LIC
+ , bool doLic = false
+ , Mv mvCurr = Mv(0, 0)
+#endif
+#endif
);
void xAddBIOAvg4 (const Pel* src0, int src0Stride, const Pel* src1, int src1Stride, Pel *dst, int dstStride, const Pel *gradX0, const Pel *gradX1, const Pel *gradY0, const Pel*gradY1, int gradStride, int width, int height, int tmpx, int tmpy, int shift, int offset, const ClpRng& clpRng);
@@ -227,6 +234,14 @@ protected:
#else
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 );
#endif
+#if ARMC_TM
+#if !INTER_LIC
+ template <bool TrueA_FalseL>
+ void xGetPredBlkTpl(const CodingUnit& cu, const ComponentID compID, const CPelBuf& refBuf, const Mv& mv, const int posW, const int posH, const int tplSize, Pel* predBlkTpl);
+#endif
+ void xWeightedAverageY(const PredictionUnit& pu, const CPelUnitBuf& pcYuvSrc0, const CPelUnitBuf& pcYuvSrc1, PelUnitBuf& pcYuvDst, const BitDepths& clipBitDepths, const ClpRngs& clpRngs);
+ void xPredAffineTpl(const PredictionUnit &pu, const RefPicList &eRefPicList, int* numTemplate, Pel* refLeftTemplate, Pel* refAboveTemplate);
+#endif
#if AFFINE_ENC_OPT
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, const bool calGradient = false);
#else
@@ -248,6 +263,13 @@ protected:
MotionInfo m_SubPuMiBuf[(MAX_CU_SIZE * MAX_CU_SIZE) >> (MIN_CU_LOG2 << 1)];
+#if ARMC_TM
+ Pel* m_acYuvCurAMLTemplate[2][MAX_NUM_COMPONENT]; //0: top, 1: left
+ bool m_bAMLTemplateAvailabe[2];
+ Pel* m_acYuvRefAboveTemplate[2][MAX_NUM_COMPONENT]; //0: list0, 1: list1
+ Pel* m_acYuvRefLeftTemplate[2][MAX_NUM_COMPONENT]; //0: list0, 1: list1
+ Pel* m_acYuvRefAMLTemplate[2][MAX_NUM_COMPONENT]; //0: top, 1: left
+#endif
#if JVET_J0090_MEMORY_BANDWITH_MEASURE
CacheModel *m_cacheModel;
#endif
@@ -260,7 +282,7 @@ public:
InterPrediction();
virtual ~InterPrediction();
-#if INTER_LIC || (TM_AMVP || TM_MRG)
+#if INTER_LIC || (TM_AMVP || TM_MRG) || ARMC_TM
void init (RdCost* pcRdCost, ChromaFormat chromaFormatIDC, const int ctuSize, Reshape* reshape);
#else
void init (RdCost* pcRdCost, ChromaFormat chromaFormatIDC, const int ctuSize);
@@ -299,6 +321,19 @@ public:
#if !AFFINE_RM_CONSTRAINTS_AND_OPT
static bool isSubblockVectorSpreadOverLimit( int a, int b, int c, int d, int predType );
#endif
+#if ARMC_TM
+ void adjustInterMergeCandidates(PredictionUnit &pu, MergeCtx& mrgCtx, int mrgCandIdx = -1);
+ bool xAMLGetCurBlkTemplate(PredictionUnit& pu, int nCurBlkWidth, int nCurBlkHeight);
+ bool xAMLIsTopTempAvailable(PredictionUnit& pu);
+ bool xAMLIsLeftTempAvailable(PredictionUnit& pu);
+ void updateCandList(uint32_t uiCand, Distortion uiCost, uint32_t uiMrgCandNum, uint32_t* RdCandList, Distortion* CandCostList);
+ void updateCandInfo(MergeCtx& mrgCtx, uint32_t(*RdCandList)[MRG_MAX_NUM_CANDS], int mrgCandIdx = -1);
+ void getBlkAMLRefTemplate(PredictionUnit &pu, PelUnitBuf &pcBufPredRefTop, PelUnitBuf &pcBufPredRefLeft);
+ void adjustAffineMergeCandidates(PredictionUnit &pu, AffineMergeCtx& affMrgCtx, int mrgCandIdx = -1);
+ void updateAffineCandInfo(PredictionUnit &pu, AffineMergeCtx& affMrgCtx, uint32_t(*RdCandList)[AFFINE_MRG_MAX_NUM_CANDS], int mrgCandIdx = -1);
+ void xGetSublkAMLTemplate(const CodingUnit& cu, const ComponentID compID, const Picture& refPic, const Mv& mv, const int sublkWidth, const int sublkHeight, const int posW, const int posH, int* numTemplate, Pel* refLeftTemplate, Pel* refAboveTemplate);
+ void getAffAMLRefTemplate(PredictionUnit &pu, PelUnitBuf &pcBufPredRefTop, PelUnitBuf &pcBufPredRefLeft);
+#endif
#if INTER_LIC
void xGetLICParamGeneral (const CodingUnit& cu, const ComponentID compID, int* numTemplate, Pel* refLeftTemplate, Pel* refAboveTemplate, Pel* recLeftTemplate, Pel* recAboveTemplate, int& shift, int& scale, int& offset);
void xGetSublkTemplate (const CodingUnit& cu, const ComponentID compID, const Picture& refPic, const Mv& mv, const int sublkWidth, const int sublkHeight, const int posW, const int posH, int* numTemplate, Pel* refLeftTemplate, Pel* refAboveTemplate, Pel* recLeftTemplate, Pel* recAboveTemplate);
diff --git a/source/Lib/CommonLib/InterpolationFilter.cpp b/source/Lib/CommonLib/InterpolationFilter.cpp
index b2a4f35265b1c6c9062483ffff23bac04ebe269f..ace9eb99c6b902049556c8fe1078e8e8e4faf6b8 100644
--- a/source/Lib/CommonLib/InterpolationFilter.cpp
+++ b/source/Lib/CommonLib/InterpolationFilter.cpp
@@ -1482,7 +1482,7 @@ void InterpolationFilter::filterHor(const ComponentID compID, Pel const *src, in
CHECK(frac < 0 || frac >= LUMA_INTERPOLATION_FILTER_SUB_SAMPLE_POSITIONS, "Invalid fraction");
if (nFilterIdx == 1)
{
-#if TM_AMVP || TM_MRG
+#if TM_AMVP || TM_MRG || ARMC_TM
filterHor<NTAPS_BILINEAR>( clpRng, src, srcStride, dst, dstStride, width, height, isLast, (biMCForDMVR ? m_bilinearFilterPrec4 : m_bilinearFilter)[frac], biMCForDMVR );
#else
filterHor<NTAPS_BILINEAR>(clpRng, src, srcStride, dst, dstStride, width, height, isLast, m_bilinearFilterPrec4[frac], biMCForDMVR);
@@ -1528,7 +1528,7 @@ void InterpolationFilter::filterHor(const ComponentID compID, Pel const *src, in
CHECK( frac < 0 || frac >= LUMA_INTERPOLATION_FILTER_SUB_SAMPLE_POSITIONS, "Invalid fraction" );
if( nFilterIdx == 1 )
{
-#if TM_AMVP || TM_MRG
+#if TM_AMVP || TM_MRG || ARMC_TM
filterHor<NTAPS_BILINEAR>( clpRng, src, srcStride, dst, dstStride, width, height, isLast, (biMCForDMVR ? m_bilinearFilterPrec4 : m_bilinearFilter)[frac], biMCForDMVR );
#else
filterHor<NTAPS_BILINEAR>( clpRng, src, srcStride, dst, dstStride, width, height, isLast, m_bilinearFilterPrec4[frac], biMCForDMVR );
@@ -1618,7 +1618,7 @@ void InterpolationFilter::filterVer(const ComponentID compID, Pel const *src, in
CHECK(frac < 0 || frac >= LUMA_INTERPOLATION_FILTER_SUB_SAMPLE_POSITIONS, "Invalid fraction");
if (nFilterIdx == 1)
{
-#if TM_AMVP || TM_MRG
+#if TM_AMVP || TM_MRG || ARMC_TM
filterVer<NTAPS_BILINEAR>( clpRng, src, srcStride, dst, dstStride, width, height, isFirst, isLast, (biMCForDMVR ? m_bilinearFilterPrec4 : m_bilinearFilter)[frac], biMCForDMVR );
#else
filterVer<NTAPS_BILINEAR>(clpRng, src, srcStride, dst, dstStride, width, height, isFirst, isLast, m_bilinearFilterPrec4[frac], biMCForDMVR);
@@ -1662,7 +1662,7 @@ void InterpolationFilter::filterVer(const ComponentID compID, Pel const *src, in
CHECK( frac < 0 || frac >= LUMA_INTERPOLATION_FILTER_SUB_SAMPLE_POSITIONS, "Invalid fraction" );
if( nFilterIdx == 1 )
{
-#if TM_AMVP || TM_MRG
+#if TM_AMVP || TM_MRG || ARMC_TM
filterVer<NTAPS_BILINEAR>( clpRng, src, srcStride, dst, dstStride, width, height, isFirst, isLast, (biMCForDMVR ? m_bilinearFilterPrec4 : m_bilinearFilter)[frac], biMCForDMVR );
#else
filterVer<NTAPS_BILINEAR>( clpRng, src, srcStride, dst, dstStride, width, height, isFirst, isLast, m_bilinearFilterPrec4[frac], biMCForDMVR );
diff --git a/source/Lib/CommonLib/Slice.cpp b/source/Lib/CommonLib/Slice.cpp
index d82e0bd2969d87b2dc6110911f75e9d22e0df70e..19d9e489e2db1ec06a5b29884a00f9f763eb885d 100644
--- a/source/Lib/CommonLib/Slice.cpp
+++ b/source/Lib/CommonLib/Slice.cpp
@@ -3142,6 +3142,9 @@ SPS::SPS()
#endif
, m_MRL ( false )
, m_MIP ( false )
+#if ARMC_TM
+, m_AML ( false )
+#endif
, m_GDREnabledFlag ( true )
, m_SubLayerCbpParametersPresentFlag ( true )
, m_rprEnabledFlag ( false )
diff --git a/source/Lib/CommonLib/Slice.h b/source/Lib/CommonLib/Slice.h
index 70d7b8bc6b0106e57954db4266f2769999574c9c..2ad0c76fcd6f912ee8357b6884a11d916a82bf26 100644
--- a/source/Lib/CommonLib/Slice.h
+++ b/source/Lib/CommonLib/Slice.h
@@ -1691,6 +1691,9 @@ private:
#endif
bool m_MRL;
bool m_MIP;
+#if ARMC_TM
+ bool m_AML;
+#endif
ChromaQpMappingTable m_chromaQpMappingTable;
bool m_GDREnabledFlag;
bool m_SubLayerCbpParametersPresentFlag;
@@ -2117,6 +2120,10 @@ void setCCALFEnabledFlag( bool b )
void setUseMIP ( bool b ) { m_MIP = b; }
bool getUseMIP () const { return m_MIP; }
+#if ARMC_TM
+ void setUseAML ( bool b ) { m_AML = b; }
+ bool getUseAML () const { return m_AML; }
+#endif
bool getUseWP () const { return m_useWeightPred; }
bool getUseWPBiPred () const { return m_useWeightedBiPred; }
void setUseWP ( bool b ) { m_useWeightPred = b; }
diff --git a/source/Lib/CommonLib/TypeDef.h b/source/Lib/CommonLib/TypeDef.h
index 83d7c38c4c00b07468f3815a5f00b0e51a6bad89..1bfdbdec7b08905d272d241d85a0715091164e6c 100644
--- a/source/Lib/CommonLib/TypeDef.h
+++ b/source/Lib/CommonLib/TypeDef.h
@@ -50,7 +50,7 @@
#include <assert.h>
#include <cassert>
-
+#define ARMC_TM 1 //Adaptive reordering of merge candidates with template matching
diff --git a/source/Lib/DecoderLib/DecCu.cpp b/source/Lib/DecoderLib/DecCu.cpp
index f3c5ec18b4cb47fa5a83083a94c4056233f287b7..0e07891a9c9be8131af12200c91c17a9112a0502 100644
--- a/source/Lib/DecoderLib/DecCu.cpp
+++ b/source/Lib/DecoderLib/DecCu.cpp
@@ -1298,7 +1298,12 @@ void DecCu::xDeriveCUMV( CodingUnit &cu )
affineMergeCtx.mrgCtx = &mrgCtx;
}
#if AFFINE_MMVD
+#if ARMC_TM
+ int affMrgIdx = pu.cs->sps->getUseAML() && (((pu.mergeIdx / ADAPTIVE_AFFINE_SUB_GROUP_SIZE + 1)*ADAPTIVE_AFFINE_SUB_GROUP_SIZE < pu.cs->sps->getMaxNumAffineMergeCand()) || (pu.mergeIdx / ADAPTIVE_AFFINE_SUB_GROUP_SIZE) == 0) ? pu.mergeIdx / ADAPTIVE_AFFINE_SUB_GROUP_SIZE * ADAPTIVE_AFFINE_SUB_GROUP_SIZE + ADAPTIVE_AFFINE_SUB_GROUP_SIZE - 1 : pu.mergeIdx;
+ PU::getAffineMergeCand(pu, affineMergeCtx, (pu.afMmvdFlag ? pu.afMmvdBaseIdx : affMrgIdx), pu.afMmvdFlag);
+#else
PU::getAffineMergeCand(pu, affineMergeCtx, (pu.afMmvdFlag ? pu.afMmvdBaseIdx : pu.mergeIdx), pu.afMmvdFlag);
+#endif
if (pu.afMmvdFlag)
{
@@ -1306,8 +1311,26 @@ void DecCu::xDeriveCUMV( CodingUnit &cu )
CHECK(pu.mergeIdx >= pu.cu->slice->getPicHeader()->getMaxNumAffineMergeCand(), "Affine MMVD mode doesn't have a valid base candidate!");
PU::getAfMmvdMvf(pu, affineMergeCtx, affineMergeCtx.mvFieldNeighbours + (pu.mergeIdx << 1), pu.mergeIdx, pu.afMmvdStep, pu.afMmvdDir);
}
+#if ARMC_TM
+ else
+ {
+ if (pu.cs->sps->getUseAML())
+ {
+ m_pcInterPred->adjustAffineMergeCandidates(pu, affineMergeCtx, pu.mergeIdx);
+ }
+ }
+#endif
+#else
+#if ARMC_TM
+ int affMrgIdx = pu.cs->sps->getUseAML() && (((pu.mergeIdx / ADAPTIVE_AFFINE_SUB_GROUP_SIZE + 1)*ADAPTIVE_AFFINE_SUB_GROUP_SIZE < pu.cs->sps->getMaxNumAffineMergeCand()) || (pu.mergeIdx / ADAPTIVE_AFFINE_SUB_GROUP_SIZE) == 0) ? pu.mergeIdx / ADAPTIVE_AFFINE_SUB_GROUP_SIZE * ADAPTIVE_AFFINE_SUB_GROUP_SIZE + ADAPTIVE_AFFINE_SUB_GROUP_SIZE - 1 : pu.mergeIdx;
+ PU::getAffineMergeCand(pu, affineMergeCtx, affMrgIdx);
+ if (pu.cs->sps->getUseAML())
+ {
+ m_pcInterPred->adjustAffineMergeCandidates(pu, affineMergeCtx, pu.mergeIdx);
+ }
#else
PU::getAffineMergeCand( pu, affineMergeCtx, pu.mergeIdx );
+#endif
#endif
pu.interDir = affineMergeCtx.interDirNeighbours[pu.mergeIdx];
pu.cu->affineType = affineMergeCtx.affineType[pu.mergeIdx];
@@ -1345,7 +1368,19 @@ void DecCu::xDeriveCUMV( CodingUnit &cu )
if (CU::isIBC(*pu.cu))
PU::getIBCMergeCandidates(pu, mrgCtx, pu.mergeIdx);
else
+#if ARMC_TM
+ if (pu.cs->sps->getUseAML())
+ {
+ PU::getInterMergeCandidates(pu, mrgCtx, 0, pu.cs->sps->getUseAML() && (((pu.mergeIdx / ADAPTIVE_SUB_GROUP_SIZE + 1)*ADAPTIVE_SUB_GROUP_SIZE < pu.cs->sps->getMaxNumMergeCand()) || (pu.mergeIdx / ADAPTIVE_SUB_GROUP_SIZE) == 0) ? pu.mergeIdx / ADAPTIVE_SUB_GROUP_SIZE * ADAPTIVE_SUB_GROUP_SIZE + ADAPTIVE_SUB_GROUP_SIZE - 1 : pu.mergeIdx);
+ m_pcInterPred->adjustInterMergeCandidates(pu, mrgCtx, pu.mergeIdx);
+ }
+ else
+ {
+ PU::getInterMergeCandidates(pu, mrgCtx, 0, pu.mergeIdx);
+ }
+#else
PU::getInterMergeCandidates(pu, mrgCtx, 0, pu.mergeIdx);
+#endif
mrgCtx.setMergeInfo( pu, pu.mergeIdx );
#if TM_MRG && !MULTI_PASS_DMVR
if (pu.tmMergeFlag)
diff --git a/source/Lib/DecoderLib/DecLib.cpp b/source/Lib/DecoderLib/DecLib.cpp
index f2808eb733dc8302fd443ecce9314ed72466aee3..ec27df33991409c01a52574c12d9c264bff541b0 100644
--- a/source/Lib/DecoderLib/DecLib.cpp
+++ b/source/Lib/DecoderLib/DecLib.cpp
@@ -1643,7 +1643,7 @@ void DecLib::xActivateParameterSets( const InputNALUnit nalu )
log2SaoOffsetScaleLuma, log2SaoOffsetScaleChroma );
m_cLoopFilter.create(maxDepth);
m_cIntraPred.init( sps->getChromaFormatIdc(), sps->getBitDepth( CHANNEL_TYPE_LUMA ) );
-#if INTER_LIC || (TM_AMVP || TM_MRG)
+#if INTER_LIC || (TM_AMVP || TM_MRG) || ARMC_TM
m_cInterPred.init( &m_cRdCost, sps->getChromaFormatIdc(), sps->getMaxCUHeight(), &m_cReshaper);
#else
m_cInterPred.init( &m_cRdCost, sps->getChromaFormatIdc(), sps->getMaxCUHeight() );
diff --git a/source/Lib/DecoderLib/VLCReader.cpp b/source/Lib/DecoderLib/VLCReader.cpp
index 96d62b5f8372b48ce07d3d99461577d8834ff1fd..2eb37b75586f8c745dc16d4355bd0a49b0e2d765 100644
--- a/source/Lib/DecoderLib/VLCReader.cpp
+++ b/source/Lib/DecoderLib/VLCReader.cpp
@@ -2153,6 +2153,9 @@ void HLSyntaxReader::parseSPS(SPS* pcSPS)
READ_FLAG( uiCode, "sps_amvr_enabled_flag" ); pcSPS->setAMVREnabledFlag ( uiCode != 0 );
+#if ARMC_TM
+ READ_FLAG( uiCode, "sps_aml_enabled_flag"); pcSPS->setUseAML ( uiCode != 0 );
+#endif
READ_FLAG( uiCode, "sps_bdof_enabled_flag" ); pcSPS->setBDOFEnabledFlag ( uiCode != 0 );
if (pcSPS->getBDOFEnabledFlag())
{
diff --git a/source/Lib/EncoderLib/EncCfg.h b/source/Lib/EncoderLib/EncCfg.h
index 14ecf41a012ef787653f69c1d7c6ab738dd6b4ae..84929d8a066564622400a086a33adf84a04c27a1 100644
--- a/source/Lib/EncoderLib/EncCfg.h
+++ b/source/Lib/EncoderLib/EncCfg.h
@@ -373,6 +373,9 @@ protected:
#endif
bool m_PROF;
bool m_BIO;
+#if ARMC_TM
+ bool m_AML;
+#endif
bool m_SMVD;
bool m_compositeRefEnabled; //composite reference
@@ -1204,6 +1207,10 @@ public:
bool getPROF () const { return m_PROF; }
void setBIO(bool b) { m_BIO = b; }
bool getBIO() const { return m_BIO; }
+#if ARMC_TM
+ void setAML(bool b) { m_AML = b; }
+ bool getAML() const { return m_AML; }
+#endif
#if ENABLE_OBMC
void setUseOBMC ( bool n ) { m_OBMC = n; }
bool getUseOBMC () const { return m_OBMC; }
diff --git a/source/Lib/EncoderLib/EncCu.cpp b/source/Lib/EncoderLib/EncCu.cpp
index d6dc2b317dd81897ab461b923c9eca00e19e1048..3160705f07a621c6dee2874ef15743a451c32468 100644
--- a/source/Lib/EncoderLib/EncCu.cpp
+++ b/source/Lib/EncoderLib/EncCu.cpp
@@ -2708,6 +2708,9 @@ void EncCu::xCheckRDCostMerge2Nx2N( CodingStructure *&tempCS, CodingStructure *&
tempCS->initStructData( encTestMode.qp );
MergeCtx mergeCtx;
+#if ARMC_TM
+ MergeCtx mergeCtxtmp;
+#endif
const SPS &sps = *tempCS->sps;
#if MERGE_ENC_OPT
@@ -2715,6 +2718,9 @@ void EncCu::xCheckRDCostMerge2Nx2N( CodingStructure *&tempCS, CodingStructure *&
&& !(bestCS->area.lumaSize().width < 8 || bestCS->area.lumaSize().height < 8);
AffineMergeCtx affineMergeCtx;
+#if ARMC_TM
+ AffineMergeCtx affineMergeCtxTmp;
+#endif
MergeCtx mrgCtx;
#if TM_MRG
MergeCtx tmMrgCtx;
@@ -2764,12 +2770,25 @@ void EncCu::xCheckRDCostMerge2Nx2N( CodingStructure *&tempCS, CodingStructure *&
, 0
);
PU::getInterMMVDMergeCandidates(pu, mergeCtx);
+#if ARMC_TM
+ mergeCtxtmp = mergeCtx;
+ if (sps.getUseAML())
+ {
+ m_pcInterSearch->adjustInterMergeCandidates(pu, mergeCtx);
+ }
+#endif
#if TM_MRG && MERGE_ENC_OPT
if (cu.cs->sps->getUseDMVDMode())
{
cu.firstPU = &pu;
pu.tmMergeFlag = true;
PU::getInterMergeCandidates(pu, tmMrgCtx, 0);
+#if ARMC_TM
+ if (sps.getUseAML())
+ {
+ m_pcInterSearch->adjustInterMergeCandidates(pu, tmMrgCtx);
+ }
+#endif
for( uint32_t uiMergeCand = 0; uiMergeCand < tmMrgCtx.numValidMergeCand; uiMergeCand++ )
{
@@ -2817,6 +2836,13 @@ void EncCu::xCheckRDCostMerge2Nx2N( CodingStructure *&tempCS, CodingStructure *&
pu.regularMergeFlag = false;
cu.affine = true;
PU::getAffineMergeCand(pu, affineMergeCtx);
+#if ARMC_TM
+ affineMergeCtxTmp = affineMergeCtx;
+ if (sps.getUseAML())
+ {
+ m_pcInterSearch->adjustAffineMergeCandidates(pu, affineMergeCtx);
+ }
+#endif
cu.affine = false;
}
#endif
@@ -3319,7 +3345,11 @@ void EncCu::xCheckRDCostMerge2Nx2N( CodingStructure *&tempCS, CodingStructure *&
if ( pu.cs->sps->getUseMMVD() )
{
#if MERGE_ENC_OPT
+#if ARMC_TM
+ xCheckSATDCostMmvdMerge(tempCS, cu, pu, mergeCtxtmp, acMergeTempBuffer, singleMergeTempBuffer, uiNumMrgSATDCand, RdModeList, candCostList, distParam, ctxStart);
+#else
xCheckSATDCostMmvdMerge(tempCS, cu, pu, mergeCtx, acMergeTempBuffer, singleMergeTempBuffer, uiNumMrgSATDCand, RdModeList, candCostList, distParam, ctxStart);
+#endif
#else
cu.mmvdSkip = true;
pu.regularMergeFlag = true;
@@ -3330,7 +3360,11 @@ void EncCu::xCheckRDCostMerge2Nx2N( CodingStructure *&tempCS, CodingStructure *&
int refineStep = (mmvdMergeCand - (baseIdx * MMVD_MAX_REFINE_NUM)) / 4;
if (refineStep >= m_pcEncCfg->getMmvdDisNum())
continue;
+#if ARMC_TM
+ mergeCtxtmp.setMmvdMergeCandiInfo(pu, mmvdMergeCand);
+#else
mergeCtx.setMmvdMergeCandiInfo(pu, mmvdMergeCand);
+#endif
PU::spanMotionInfo(pu, mergeCtx);
pu.mvRefine = true;
@@ -3393,7 +3427,11 @@ void EncCu::xCheckRDCostMerge2Nx2N( CodingStructure *&tempCS, CodingStructure *&
#if AFFINE_MMVD
if (affineMmvdAvail)
{
+#if ARMC_TM
+ xCheckSATDCostAffineMmvdMerge(tempCS, cu, pu, affineMergeCtxTmp, mrgCtx, acMergeTempBuffer, singleMergeTempBuffer, uiNumMrgSATDCand, RdModeList, candCostList, distParam, ctxStart);
+#else
xCheckSATDCostAffineMmvdMerge(tempCS, cu, pu, affineMergeCtx, mrgCtx, acMergeTempBuffer, singleMergeTempBuffer, uiNumMrgSATDCand, RdModeList, candCostList, distParam, ctxStart);
+#endif
}
#endif
#endif
@@ -3544,7 +3582,11 @@ void EncCu::xCheckRDCostMerge2Nx2N( CodingStructure *&tempCS, CodingStructure *&
{
cu.mmvdSkip = true;
pu.regularMergeFlag = true;
+#if ARMC_TM
+ mergeCtxtmp.setMmvdMergeCandiInfo(pu, uiMergeCand);
+#else
mergeCtx.setMmvdMergeCandiInfo(pu, uiMergeCand);
+#endif
}
#if MERGE_ENC_OPT
#if AFFINE_MMVD
@@ -3567,6 +3609,22 @@ void EncCu::xCheckRDCostMerge2Nx2N( CodingStructure *&tempCS, CodingStructure *&
pu.afMmvdDir = (uint8_t)dirIdx;
pu.afMmvdStep = (uint8_t)stepIdx;
pu.mergeIdx = (uint8_t)(baseIdx + afMmvdBaseIdxToMergeIdxOffset);
+#if ARMC_TM
+ pu.mergeType = affineMergeCtxTmp.mergeType[pu.mergeIdx];
+#if INTER_LIC
+ pu.cu->LICFlag = affineMergeCtxTmp.LICFlags[pu.mergeIdx];
+#endif
+ pu.interDir = affineMergeCtxTmp.interDirNeighbours[pu.mergeIdx];
+ pu.cu->affineType = affineMergeCtxTmp.affineType[pu.mergeIdx];
+ pu.cu->BcwIdx = affineMergeCtxTmp.BcwIdx[pu.mergeIdx];
+ pu.mmvdMergeFlag = false;
+ pu.ciipFlag = false;
+
+ CHECK(pu.mergeIdx >= affineMergeCtxTmp.numValidMergeCand, "Invalid merge index for AffineMMVD");
+
+ MvField mvfMmvd[2][3];
+ PU::getAfMmvdMvf(pu, affineMergeCtxTmp, mvfMmvd, pu.mergeIdx, pu.afMmvdStep, pu.afMmvdDir);
+#else
pu.mergeType = affineMergeCtx.mergeType [pu.mergeIdx];
#if INTER_LIC
pu.cu->LICFlag = affineMergeCtx.LICFlags [pu.mergeIdx];
@@ -3581,6 +3639,7 @@ void EncCu::xCheckRDCostMerge2Nx2N( CodingStructure *&tempCS, CodingStructure *&
MvField mvfMmvd[2][3];
PU::getAfMmvdMvf(pu, affineMergeCtx, mvfMmvd, pu.mergeIdx, pu.afMmvdStep, pu.afMmvdDir);
+#endif
for (int i = 0; i < 2; i++)
{
@@ -5098,6 +5157,12 @@ void EncCu::xCheckRDCostAffineMerge2Nx2N( CodingStructure *&tempCS, CodingStruct
pu.cs = tempCS;
pu.regularMergeFlag = false;
PU::getAffineMergeCand( pu, affineMergeCtx );
+#if ARMC_TM
+ if (sps.getUseAML())
+ {
+ m_pcInterSearch->adjustAffineMergeCandidates(pu, affineMergeCtx);
+ }
+#endif
if ( affineMergeCtx.numValidMergeCand <= 0 )
{
@@ -5736,6 +5801,12 @@ void EncCu::xCheckRDCostTMMerge2Nx2N(CodingStructure *&tempCS, CodingStructure *
pu.tmMergeFlag = true;
PU::getInterMergeCandidates(pu, mergeCtx, 0);
+#if ARMC_TM
+ if (sps.getUseAML())
+ {
+ m_pcInterSearch->adjustInterMergeCandidates(pu, mergeCtx);
+ }
+#endif
for( uint32_t uiMergeCand = 0; uiMergeCand < mergeCtx.numValidMergeCand; uiMergeCand++ )
{
diff --git a/source/Lib/EncoderLib/EncLib.cpp b/source/Lib/EncoderLib/EncLib.cpp
index 16965b1ac1592b814f890798740006b84e007d93..65361885e723be5c9a19b1eb9302057b72bd409c 100644
--- a/source/Lib/EncoderLib/EncLib.cpp
+++ b/source/Lib/EncoderLib/EncLib.cpp
@@ -1370,6 +1370,9 @@ void EncLib::xInitSPS( SPS& sps )
sps.setSbTMVPEnabledFlag(m_sbTmvpEnableFlag);
sps.setAMVREnabledFlag ( m_ImvMode != IMV_OFF );
sps.setBDOFEnabledFlag ( m_BIO );
+#if ARMC_TM
+ sps.setUseAML ( m_AML );
+#endif
sps.setMaxNumMergeCand(getMaxNumMergeCand());
sps.setMaxNumAffineMergeCand(getMaxNumAffineMergeCand());
sps.setMaxNumIBCMergeCand(getMaxNumIBCMergeCand());
diff --git a/source/Lib/EncoderLib/InterSearch.cpp b/source/Lib/EncoderLib/InterSearch.cpp
index 2a7145c80f8ea7bacd578ea73f6005b05dc961c7..061b1b82bb58637175d2d97b65e8adcbcd0ada28 100644
--- a/source/Lib/EncoderLib/InterSearch.cpp
+++ b/source/Lib/EncoderLib/InterSearch.cpp
@@ -275,7 +275,7 @@ void InterSearch::init( EncCfg* pcEncCfg,
}
const ChromaFormat cform = pcEncCfg->getChromaFormatIdc();
-#if INTER_LIC || (TM_AMVP || TM_MRG)
+#if INTER_LIC || (TM_AMVP || TM_MRG) || ARMC_TM
InterPrediction::init( pcRdCost, cform, maxCUHeight, m_pcReshape );
#else
InterPrediction::init( pcRdCost, cform, maxCUHeight );
diff --git a/source/Lib/EncoderLib/VLCWriter.cpp b/source/Lib/EncoderLib/VLCWriter.cpp
index 19aeb75d35f7e95bb82c2507245036e6afbf7d34..aad9ff6f41b04b17a339d32f6fa5cff89f5d970d 100644
--- a/source/Lib/EncoderLib/VLCWriter.cpp
+++ b/source/Lib/EncoderLib/VLCWriter.cpp
@@ -1307,6 +1307,9 @@ void HLSWriter::codeSPS( const SPS* pcSPS )
WRITE_FLAG( pcSPS->getAMVREnabledFlag() ? 1 : 0, "sps_amvr_enabled_flag" );
+#if ARMC_TM
+ WRITE_FLAG( pcSPS->getUseAML() ? 1 : 0, "sps_aml_enabled_flag" );
+#endif
WRITE_FLAG( pcSPS->getBDOFEnabledFlag() ? 1 : 0, "sps_bdof_enabled_flag" );
if (pcSPS->getBDOFEnabledFlag())
{