From fd970b23feebe4e96fd42fd75a7a7e8ef739d3d9 Mon Sep 17 00:00:00 2001
From: Xiaoyu Xiu <xiaoyuxiu@kwai.com>
Date: Tue, 23 Jul 2019 21:41:52 +0200
Subject: [PATCH] use m_pcEncCfg->getUseAffineAmvrEncOpt() to control the
effect of m_pcEncCfg->getIntraPeriod() != ( uint32_t ) -1, i.e., low delay
setting.
---
source/Lib/CommonLib/Rom.cpp | 8 +
source/Lib/CommonLib/Rom.h | 6 +
source/Lib/CommonLib/TypeDef.h | 2 +
source/Lib/EncoderLib/EncCu.cpp | 17 ++
source/Lib/EncoderLib/EncModeCtrl.cpp | 12 ++
source/Lib/EncoderLib/EncModeCtrl.h | 9 +
source/Lib/EncoderLib/EncSlice.cpp | 3 +
source/Lib/EncoderLib/InterSearch.cpp | 285 ++++++++++++++++++++++++++
source/Lib/EncoderLib/InterSearch.h | 100 +++++++++
9 files changed, 442 insertions(+)
diff --git a/source/Lib/CommonLib/Rom.cpp b/source/Lib/CommonLib/Rom.cpp
index 7cc906bc1..e32c236ea 100644
--- a/source/Lib/CommonLib/Rom.cpp
+++ b/source/Lib/CommonLib/Rom.cpp
@@ -430,6 +430,10 @@ void initROM()
}
}
#endif
+
+#if JVET_O0592_ENC_ME_IMP
+ ::memset(g_isReusedUniMVsFilled, 0, sizeof(g_isReusedUniMVsFilled));
+#endif
}
void destroyROM()
@@ -753,5 +757,9 @@ uint8_t g_triangleMvStorage[TRIANGLE_DIR_NUM][MAX_CU_DEPTH - MIN_CU_LOG2 + 1][MA
#if JVET_O0280_SIMD_TRIANGLE_WEIGHTING
int16_t *g_triangleWeights[2][TRIANGLE_DIR_NUM][MAX_CU_DEPTH - MIN_CU_LOG2 + 2][MAX_CU_DEPTH - MIN_CU_LOG2 + 2];
#endif
+#if JVET_O0592_ENC_ME_IMP
+Mv g_reusedUniMVs[32][32][8][8][2][33];
+bool g_isReusedUniMVsFilled[32][32][8][8];
+#endif
//! \}
diff --git a/source/Lib/CommonLib/Rom.h b/source/Lib/CommonLib/Rom.h
index ac26cd734..877b947d7 100644
--- a/source/Lib/CommonLib/Rom.h
+++ b/source/Lib/CommonLib/Rom.h
@@ -226,5 +226,11 @@ extern int16_t *g_triangleWeights[2][TRIANGLE_DIR_NUM][MAX_CU_DEPTH - MIN_CU_LOG
extern bool g_mctsDecCheckEnabled;
+#if JVET_O0592_ENC_ME_IMP
+class Mv;
+extern Mv g_reusedUniMVs[32][32][8][8][2][33];
+extern bool g_isReusedUniMVsFilled[32][32][8][8];
+#endif
+
#endif //__TCOMROM__
diff --git a/source/Lib/CommonLib/TypeDef.h b/source/Lib/CommonLib/TypeDef.h
index 384a0c6cf..20b82d923 100644
--- a/source/Lib/CommonLib/TypeDef.h
+++ b/source/Lib/CommonLib/TypeDef.h
@@ -114,6 +114,8 @@
#define JVET_O0126_BPWA_INDEX_CODING_FIX 1 // JVET-O0126 align BPWA index coding with specification
+#define JVET_O0592_ENC_ME_IMP 1 // JVET-O0592 encoder ME improvement
+
#define FIX_DB_MAX_TRANSFORM_SIZE 1
#define MRG_SHARELIST_SHARSIZE 32
diff --git a/source/Lib/EncoderLib/EncCu.cpp b/source/Lib/EncoderLib/EncCu.cpp
index 2ae45ae5a..167cc7935 100644
--- a/source/Lib/EncoderLib/EncCu.cpp
+++ b/source/Lib/EncoderLib/EncCu.cpp
@@ -266,6 +266,9 @@ void EncCu::init( EncLib* pcEncLib, const SPS& sps PARL_PARAM( const int tId ) )
m_modeCtrl->init( m_pcEncCfg, m_pcRateCtrl, m_pcRdCost );
m_pcInterSearch->setModeCtrl( m_modeCtrl );
+#if JVET_O0592_ENC_ME_IMP
+ m_modeCtrl->setInterSearch(m_pcInterSearch);
+#endif
m_pcIntraSearch->setModeCtrl( m_modeCtrl );
if ( ( m_pcEncCfg->getIBCHashSearch() && m_pcEncCfg->getIBCMode() ) || m_pcEncCfg->getAllowDisFracMMVD() )
@@ -1188,6 +1191,14 @@ void EncCu::xCheckModeSplit(CodingStructure *&tempCS, CodingStructure *&bestCS,
AffineMVInfo tmpMVInfo;
bool isAffMVInfoSaved;
m_pcInterSearch->savePrevAffMVInfo(0, tmpMVInfo, isAffMVInfoSaved);
+#if JVET_O0592_ENC_ME_IMP
+ BlkUniMvInfo tmpUniMvInfo;
+ bool isUniMvInfoSaved = false;
+ if (!tempCS->slice->isIntra())
+ {
+ m_pcInterSearch->savePrevUniMvInfo(tempCS->area.Y(), tmpUniMvInfo, isUniMvInfoSaved);
+ }
+#endif
do
{
@@ -1319,6 +1330,12 @@ void EncCu::xCheckModeSplit(CodingStructure *&tempCS, CodingStructure *&bestCS,
if (isAffMVInfoSaved)
m_pcInterSearch->addAffMVInfo(tmpMVInfo);
+#if JVET_O0592_ENC_ME_IMP
+ if (!tempCS->slice->isIntra() && isUniMvInfoSaved)
+ {
+ m_pcInterSearch->addUniMvInfo(tmpUniMvInfo);
+ }
+#endif
tempCS->motionLut = oldMotionLut;
diff --git a/source/Lib/EncoderLib/EncModeCtrl.cpp b/source/Lib/EncoderLib/EncModeCtrl.cpp
index 1c1622bcd..e91d16737 100644
--- a/source/Lib/EncoderLib/EncModeCtrl.cpp
+++ b/source/Lib/EncoderLib/EncModeCtrl.cpp
@@ -1795,6 +1795,18 @@ bool EncModeCtrlMTnoRQT::tryMode( const EncTestMode& encTestmode, const CodingSt
{
CHECK( encTestmode.type != ETM_POST_DONT_SPLIT, "Unknown mode" );
+#if JVET_O0592_ENC_ME_IMP
+ if ((cuECtx.get<double>(BEST_NO_IMV_COST) == (MAX_DOUBLE * .5) || cuECtx.get<bool>(IS_REUSING_CU)) && !slice.isIntra())
+ {
+ unsigned idx1, idx2, idx3, idx4;
+ getAreaIdx(partitioner.currArea().Y(), *slice.getPPS()->pcv, idx1, idx2, idx3, idx4);
+ if (g_isReusedUniMVsFilled[idx1][idx2][idx3][idx4])
+ {
+ m_pcInterSearch->insertUniMvCands(partitioner.currArea().Y(), g_reusedUniMVs[idx1][idx2][idx3][idx4]);
+ }
+ }
+#endif
+
if( !bestCS || ( bestCS && isModeSplit( bestMode ) ) )
{
return false;
diff --git a/source/Lib/EncoderLib/EncModeCtrl.h b/source/Lib/EncoderLib/EncModeCtrl.h
index 9fe9e3189..d29d095c4 100644
--- a/source/Lib/EncoderLib/EncModeCtrl.h
+++ b/source/Lib/EncoderLib/EncModeCtrl.h
@@ -43,6 +43,9 @@
#include "CommonLib/CommonDef.h"
#include "CommonLib/CodingStructure.h"
+#if JVET_O0592_ENC_ME_IMP
+#include "InterSearch.h"
+#endif
#include <typeinfo>
#include <vector>
@@ -258,6 +261,9 @@ protected:
#if ENABLE_SPLIT_PARALLELISM
int m_runNextInParallel;
#endif
+#if JVET_O0592_ENC_ME_IMP
+ InterSearch* m_pcInterSearch;
+#endif
public:
@@ -315,6 +321,9 @@ public:
void setBestCostWithoutSplitFlags ( double cost ) { m_ComprCUCtxList.back().bestCostWithoutSplitFlags = cost; }
double getMtsFirstPassNoIspCost () const { return m_ComprCUCtxList.back().bestCostMtsFirstPassNoIsp; }
void setMtsFirstPassNoIspCost ( double cost ) { m_ComprCUCtxList.back().bestCostMtsFirstPassNoIsp = cost; }
+#if JVET_O0592_ENC_ME_IMP
+ void setInterSearch (InterSearch* pcInterSearch) { m_pcInterSearch = pcInterSearch; }
+#endif
protected:
void xExtractFeatures ( const EncTestMode encTestmode, CodingStructure& cs );
diff --git a/source/Lib/EncoderLib/EncSlice.cpp b/source/Lib/EncoderLib/EncSlice.cpp
index 9fd85ab2f..1e75780bb 100644
--- a/source/Lib/EncoderLib/EncSlice.cpp
+++ b/source/Lib/EncoderLib/EncSlice.cpp
@@ -1415,6 +1415,9 @@ void EncSlice::compressSlice( Picture* pcPic, const bool bCompressEntireSlice, c
writeBlockStatisticsHeader(sps);
#endif
m_pcInterSearch->resetAffineMVList();
+#if JVET_O0592_ENC_ME_IMP
+ m_pcInterSearch->resetUniMvList();
+#endif
encodeCtus( pcPic, bCompressEntireSlice, bFastDeltaQP, startCtuTsAddr, boundingCtuTsAddr, m_pcLib );
diff --git a/source/Lib/EncoderLib/InterSearch.cpp b/source/Lib/EncoderLib/InterSearch.cpp
index 9827166bd..6d82bae56 100644
--- a/source/Lib/EncoderLib/InterSearch.cpp
+++ b/source/Lib/EncoderLib/InterSearch.cpp
@@ -112,6 +112,11 @@ InterSearch::InterSearch()
m_affMVList = nullptr;
m_affMVListSize = 0;
m_affMVListIdx = 0;
+#if JVET_O0592_ENC_ME_IMP
+ m_uniMvList = nullptr;
+ m_uniMvListSize = 0;
+ m_uniMvListIdx = 0;
+#endif
m_histBestSbt = MAX_UCHAR;
m_histBestMtsIdx = MAX_UCHAR;
}
@@ -156,6 +161,15 @@ void InterSearch::destroy()
}
m_affMVListIdx = 0;
m_affMVListSize = 0;
+#if JVET_O0592_ENC_ME_IMP
+ if (m_uniMvList)
+ {
+ delete[] m_uniMvList;
+ m_uniMvList = nullptr;
+ }
+ m_uniMvListIdx = 0;
+ m_uniMvListSize = 0;
+#endif
m_isInitialized = false;
}
@@ -255,6 +269,15 @@ void InterSearch::init( EncCfg* pcEncCfg,
m_affMVList = new AffineMVInfo[m_affMVListMaxSize];
m_affMVListIdx = 0;
m_affMVListSize = 0;
+#if JVET_O0592_ENC_ME_IMP
+ m_uniMvListMaxSize = 15;
+ if (!m_uniMvList)
+ {
+ m_uniMvList = new BlkUniMvInfo[m_uniMvListMaxSize];
+ }
+ m_uniMvListIdx = 0;
+ m_uniMvListSize = 0;
+#endif
m_isInitialized = true;
}
@@ -2399,6 +2422,18 @@ void InterSearch::predInterSearch(CodingUnit& cu, Partitioner& partitioner)
}
}
+#if JVET_O0592_ENC_ME_IMP
+ ::memcpy(cMvHevcTemp, cMvTemp, sizeof(cMvTemp));
+ if (cu.imv == 0 && (!cu.slice->getSPS()->getUseGBi() || gbiIdx == GBI_DEFAULT))
+ {
+ insertUniMvCands(pu.Y(), cMvTemp);
+
+ unsigned idx1, idx2, idx3, idx4;
+ getAreaIdx(cu.Y(), *cu.slice->getPPS()->pcv, idx1, idx2, idx3, idx4);
+ ::memcpy(&(g_reusedUniMVs[idx1][idx2][idx3][idx4][0][0]), cMvTemp, 2 * 33 * sizeof(Mv));
+ g_isReusedUniMVsFilled[idx1][idx2][idx3][idx4] = true;
+ }
+#else
if (cu.Y().width > 8 && cu.Y().height > 8 && cu.slice->getSPS()->getUseAffine()
&& checkAffine
&& (gbiIdx == GBI_DEFAULT || m_affineModeSelected || !m_pcEncCfg->getUseGBiFast())
@@ -2406,6 +2441,7 @@ void InterSearch::predInterSearch(CodingUnit& cu, Partitioner& partitioner)
{
::memcpy( cMvHevcTemp, cMvTemp, sizeof( cMvTemp ) );
}
+#endif
// Bi-predictive Motion estimation
if( ( cs.slice->isInterB() ) && ( PU::isBipredRestriction( pu ) == false )
&& (cu.slice->getCheckLDC() || gbiIdx == GBI_DEFAULT || !m_affineModeSelected || !m_pcEncCfg->getUseGBiFast())
@@ -2648,11 +2684,48 @@ void InterSearch::predInterSearch(CodingUnit& cu, Partitioner& partitioner)
costStart += m_pcRdCost->getCost(bits);
std::vector<Mv> symmvdCands;
+#if JVET_O0592_ENC_ME_IMP
+ auto smmvdCandsGen = [&](Mv mvCand, bool mvPrecAdj)
+ {
+ if (mvPrecAdj && pu.cu->imv)
+ {
+ mvCand.roundIbcPrecInternal2Amvr(pu.cu->imv);
+ }
+
+ bool toAddMvCand = true;
+ for (std::vector<Mv>::iterator pos = symmvdCands.begin(); pos != symmvdCands.end(); pos++)
+ {
+ if (*pos == mvCand)
+ {
+ toAddMvCand = false;
+ break;
+ }
+ }
+
+ if (toAddMvCand)
+ {
+ symmvdCands.push_back(mvCand);
+ }
+ };
+
+ smmvdCandsGen(cMvHevcTemp[curRefList][refIdxCur], false);
+ smmvdCandsGen(cMvTemp[curRefList][refIdxCur], false);
+ if (iRefIdxBi[curRefList] == refIdxCur)
+ {
+ smmvdCandsGen(cMvBi[curRefList], false);
+ }
+ for (int i = 0; i < m_uniMvListSize; i++)
+ {
+ BlkUniMvInfo* curMvInfo = m_uniMvList + ((m_uniMvListIdx - 1 - i + m_uniMvListMaxSize) % (m_uniMvListMaxSize));
+ smmvdCandsGen(curMvInfo->uniMvs[curRefList][refIdxCur], true);
+ }
+#else
symmvdCands.push_back(cMvTemp[curRefList][refIdxCur]);
if (iRefIdxBi[curRefList] == refIdxCur && cMvBi[curRefList] != cMvTemp[curRefList][refIdxCur])
{
symmvdCands.push_back(cMvBi[curRefList]);
}
+#endif
for (auto mvStart : symmvdCands)
{
@@ -2662,7 +2735,13 @@ void InterSearch::predInterSearch(CodingUnit& cu, Partitioner& partitioner)
checked |= (mvStart == aacAMVPInfo[curRefList][refIdxCur].mvCand[i]);
}
if (checked)
+#if JVET_O0592_ENC_ME_IMP
+ {
+ continue;
+ }
+#else
break;
+#endif
Distortion bestCost = costStart;
symmvdCheckBestMvp(pu, origBuf, mvStart, (RefPicList)curRefList, aacAMVPInfo, gbiIdx, cMvPredSym, mvpIdxSym, costStart);
@@ -3288,13 +3367,63 @@ void InterSearch::xMotionEstimation(PredictionUnit& pu, PelUnitBuf& origBuf, Ref
// Do integer search
if( ( m_motionEstimationSearchMethod == MESEARCH_FULL ) || bBi || bQTBTMV )
{
+#if JVET_O0592_ENC_ME_IMP
+ cStruct.subShiftMode = m_pcEncCfg->getFastInterSearchMode() == FASTINTERSEARCH_MODE1 || m_pcEncCfg->getFastInterSearchMode() == FASTINTERSEARCH_MODE3 ? 2 : 0;
+ m_pcRdCost->setDistParam(m_cDistParam, *cStruct.pcPatternKey, cStruct.piRefY, cStruct.iRefStride, m_lumaClpRng.bd, COMPONENT_Y, cStruct.subShiftMode);
+
+ Mv bestInitMv = (bBi ? rcMv : rcMvPred);
+ Mv cTmpMv = bestInitMv;
+ clipMv(cTmpMv, pu.cu->lumaPos(), pu.cu->lumaSize(), *pu.cs->sps);
+ cTmpMv.changePrecision(MV_PRECISION_INTERNAL, MV_PRECISION_INT);
+ m_cDistParam.cur.buf = cStruct.piRefY + (cTmpMv.ver * cStruct.iRefStride) + cTmpMv.hor;
+ Distortion uiBestSad = m_cDistParam.distFunc(m_cDistParam);
+ uiBestSad += m_pcRdCost->getCostOfVectorWithPredictor(cTmpMv.hor, cTmpMv.ver, cStruct.imvShift);
+
+ for (int i = 0; i < m_uniMvListSize; i++)
+ {
+ BlkUniMvInfo* curMvInfo = m_uniMvList + ((m_uniMvListIdx - 1 - i + m_uniMvListMaxSize) % (m_uniMvListMaxSize));
+
+ int j = 0;
+ for (; j < i; j++)
+ {
+ BlkUniMvInfo *prevMvInfo = m_uniMvList + ((m_uniMvListIdx - 1 - j + m_uniMvListMaxSize) % (m_uniMvListMaxSize));
+ if (curMvInfo->uniMvs[eRefPicList][iRefIdxPred] == prevMvInfo->uniMvs[eRefPicList][iRefIdxPred])
+ {
+ break;
+ }
+ }
+ if (j < i)
+ continue;
+
+ cTmpMv = curMvInfo->uniMvs[eRefPicList][iRefIdxPred];
+ clipMv(cTmpMv, pu.cu->lumaPos(), pu.cu->lumaSize(), *pu.cs->sps);
+ cTmpMv.changePrecision(MV_PRECISION_INTERNAL, MV_PRECISION_INT);
+ m_cDistParam.cur.buf = cStruct.piRefY + (cTmpMv.ver * cStruct.iRefStride) + cTmpMv.hor;
+
+ Distortion uiSad = m_cDistParam.distFunc(m_cDistParam);
+ uiSad += m_pcRdCost->getCostOfVectorWithPredictor(cTmpMv.hor, cTmpMv.ver, cStruct.imvShift);
+ if (uiSad < uiBestSad)
+ {
+ uiBestSad = uiSad;
+ bestInitMv = curMvInfo->uniMvs[eRefPicList][iRefIdxPred];
+ m_cDistParam.maximumDistortionForEarlyExit = uiSad;
+ }
+ }
+#endif
+
if( !bQTBTMV )
{
+#if JVET_O0592_ENC_ME_IMP
+ xSetSearchRange(pu, bestInitMv, iSrchRng, cStruct.searchRange
+#else
xSetSearchRange(pu, (bBi ? rcMv : rcMvPred), iSrchRng, cStruct.searchRange
+#endif
, cStruct
);
}
+#if !JVET_O0592_ENC_ME_IMP
cStruct.subShiftMode = m_pcEncCfg->getFastInterSearchMode() == FASTINTERSEARCH_MODE1 || m_pcEncCfg->getFastInterSearchMode() == FASTINTERSEARCH_MODE3 ? 2 : 0;
+#endif
xPatternSearch( cStruct, rcMv, ruiCost);
}
else if( bQTBTMV2 )
@@ -3303,7 +3432,11 @@ void InterSearch::xMotionEstimation(PredictionUnit& pu, PelUnitBuf& origBuf, Ref
cStruct.subShiftMode = ( !m_pcEncCfg->getRestrictMESampling() && m_pcEncCfg->getMotionEstimationSearchMethod() == MESEARCH_SELECTIVE ) ? 1 :
( m_pcEncCfg->getFastInterSearchMode() == FASTINTERSEARCH_MODE1 || m_pcEncCfg->getFastInterSearchMode() == FASTINTERSEARCH_MODE3 ) ? 2 : 0;
+#if JVET_O0592_ENC_ME_IMP
+ xTZSearch(pu, eRefPicList, iRefIdxPred, cStruct, rcMv, ruiCost, NULL, false, true);
+#else
xTZSearch( pu, cStruct, rcMv, ruiCost, NULL, false, true );
+#endif
}
else
{
@@ -3311,7 +3444,11 @@ void InterSearch::xMotionEstimation(PredictionUnit& pu, PelUnitBuf& origBuf, Ref
( m_pcEncCfg->getFastInterSearchMode() == FASTINTERSEARCH_MODE1 || m_pcEncCfg->getFastInterSearchMode() == FASTINTERSEARCH_MODE3 ) ? 2 : 0;
rcMv = rcMvPred;
const Mv *pIntegerMv2Nx2NPred = 0;
+#if JVET_O0592_ENC_ME_IMP
+ xPatternSearchFast(pu, eRefPicList, iRefIdxPred, cStruct, rcMv, ruiCost, pIntegerMv2Nx2NPred);
+#else
xPatternSearchFast( pu, cStruct, rcMv, ruiCost, pIntegerMv2Nx2NPred );
+#endif
if( blkCache )
{
blkCache->setMv( pu.cs->area, eRefPicList, iRefIdxPred, rcMv );
@@ -3471,6 +3608,10 @@ void InterSearch::xPatternSearch( IntTZSearchStruct& cStruct,
void InterSearch::xPatternSearchFast( const PredictionUnit& pu,
+#if JVET_O0592_ENC_ME_IMP
+ RefPicList eRefPicList,
+ int iRefIdxPred,
+#endif
IntTZSearchStruct& cStruct,
Mv& rcMv,
Distortion& ruiSAD,
@@ -3479,15 +3620,27 @@ void InterSearch::xPatternSearchFast( const PredictionUnit& pu,
switch ( m_motionEstimationSearchMethod )
{
case MESEARCH_DIAMOND:
+#if JVET_O0592_ENC_ME_IMP
+ xTZSearch ( pu, eRefPicList, iRefIdxPred, cStruct, rcMv, ruiSAD, pIntegerMv2Nx2NPred, false );
+#else
xTZSearch ( pu, cStruct, rcMv, ruiSAD, pIntegerMv2Nx2NPred, false );
+#endif
break;
case MESEARCH_SELECTIVE:
+#if JVET_O0592_ENC_ME_IMP
+ xTZSearchSelective( pu, eRefPicList, iRefIdxPred, cStruct, rcMv, ruiSAD, pIntegerMv2Nx2NPred );
+#else
xTZSearchSelective( pu, cStruct, rcMv, ruiSAD, pIntegerMv2Nx2NPred );
+#endif
break;
case MESEARCH_DIAMOND_ENHANCED:
+#if JVET_O0592_ENC_ME_IMP
+ xTZSearch ( pu, eRefPicList, iRefIdxPred, cStruct, rcMv, ruiSAD, pIntegerMv2Nx2NPred, true );
+#else
xTZSearch ( pu, cStruct, rcMv, ruiSAD, pIntegerMv2Nx2NPred, true );
+#endif
break;
case MESEARCH_FULL: // shouldn't get here.
@@ -3498,6 +3651,10 @@ void InterSearch::xPatternSearchFast( const PredictionUnit& pu,
void InterSearch::xTZSearch( const PredictionUnit& pu,
+#if JVET_O0592_ENC_ME_IMP
+ RefPicList eRefPicList,
+ int iRefIdxPred,
+#endif
IntTZSearchStruct& cStruct,
Mv& rcMv,
Distortion& ruiSAD,
@@ -3588,6 +3745,41 @@ void InterSearch::xTZSearch( const PredictionUnit& pu,
xTZSearchHelp( cStruct, integerMv2Nx2NPred.getHor(), integerMv2Nx2NPred.getVer(), 0, 0);
}
}
+
+#if JVET_O0592_ENC_ME_IMP
+ for (int i = 0; i < m_uniMvListSize; i++)
+ {
+ BlkUniMvInfo* curMvInfo = m_uniMvList + ((m_uniMvListIdx - 1 - i + m_uniMvListMaxSize) % (m_uniMvListMaxSize));
+
+ int j = 0;
+ for (; j < i; j++)
+ {
+ BlkUniMvInfo *prevMvInfo = m_uniMvList + ((m_uniMvListIdx - 1 - j + m_uniMvListMaxSize) % (m_uniMvListMaxSize));
+ if (curMvInfo->uniMvs[eRefPicList][iRefIdxPred] == prevMvInfo->uniMvs[eRefPicList][iRefIdxPred])
+ {
+ break;
+ }
+ }
+ if (j < i)
+ continue;
+
+ Mv cTmpMv = curMvInfo->uniMvs[eRefPicList][iRefIdxPred];
+ clipMv(cTmpMv, pu.cu->lumaPos(), pu.cu->lumaSize(), *pu.cs->sps);
+ cTmpMv.changePrecision(MV_PRECISION_INTERNAL, MV_PRECISION_INT);
+ m_cDistParam.cur.buf = cStruct.piRefY + (cTmpMv.ver * cStruct.iRefStride) + cTmpMv.hor;
+
+ Distortion uiSad = m_cDistParam.distFunc(m_cDistParam);
+ uiSad += m_pcRdCost->getCostOfVectorWithPredictor(cTmpMv.hor, cTmpMv.ver, cStruct.imvShift);
+ if (uiSad < cStruct.uiBestSad)
+ {
+ cStruct.uiBestSad = uiSad;
+ cStruct.iBestX = cTmpMv.hor;
+ cStruct.iBestY = cTmpMv.ver;
+ m_cDistParam.maximumDistortionForEarlyExit = uiSad;
+ }
+ }
+#endif
+
{
// set search range
Mv currBestMv(cStruct.iBestX, cStruct.iBestY );
@@ -3804,6 +3996,10 @@ void InterSearch::xTZSearch( const PredictionUnit& pu,
void InterSearch::xTZSearchSelective( const PredictionUnit& pu,
+#if JVET_O0592_ENC_ME_IMP
+ RefPicList eRefPicList,
+ int iRefIdxPred,
+#endif
IntTZSearchStruct& cStruct,
Mv &rcMv,
Distortion &ruiSAD,
@@ -3859,6 +4055,41 @@ void InterSearch::xTZSearchSelective( const PredictionUnit& pu,
xTZSearchHelp( cStruct, integerMv2Nx2NPred.getHor(), integerMv2Nx2NPred.getVer(), 0, 0);
}
+
+#if JVET_O0592_ENC_ME_IMP
+ for (int i = 0; i < m_uniMvListSize; i++)
+ {
+ BlkUniMvInfo* curMvInfo = m_uniMvList + ((m_uniMvListIdx - 1 - i + m_uniMvListMaxSize) % (m_uniMvListMaxSize));
+
+ int j = 0;
+ for (; j < i; j++)
+ {
+ BlkUniMvInfo *prevMvInfo = m_uniMvList + ((m_uniMvListIdx - 1 - j + m_uniMvListMaxSize) % (m_uniMvListMaxSize));
+ if (curMvInfo->uniMvs[eRefPicList][iRefIdxPred] == prevMvInfo->uniMvs[eRefPicList][iRefIdxPred])
+ {
+ break;
+ }
+ }
+ if (j < i)
+ continue;
+
+ Mv cTmpMv = curMvInfo->uniMvs[eRefPicList][iRefIdxPred];
+ clipMv(cTmpMv, pu.cu->lumaPos(), pu.cu->lumaSize(), *pu.cs->sps);
+ cTmpMv.changePrecision(MV_PRECISION_INTERNAL, MV_PRECISION_INT);
+ m_cDistParam.cur.buf = cStruct.piRefY + (cTmpMv.ver * cStruct.iRefStride) + cTmpMv.hor;
+
+ Distortion uiSad = m_cDistParam.distFunc(m_cDistParam);
+ uiSad += m_pcRdCost->getCostOfVectorWithPredictor(cTmpMv.hor, cTmpMv.ver, cStruct.imvShift);
+ if (uiSad < cStruct.uiBestSad)
+ {
+ cStruct.uiBestSad = uiSad;
+ cStruct.iBestX = cTmpMv.hor;
+ cStruct.iBestY = cTmpMv.ver;
+ m_cDistParam.maximumDistortionForEarlyExit = uiSad;
+ }
+ }
+#endif
+
{
// set search range
Mv currBestMv(cStruct.iBestX, cStruct.iBestY );
@@ -5445,6 +5676,7 @@ void InterSearch::xAffineMotionEstimation( PredictionUnit& pu,
const uint32_t mvShift = mvShiftTable[pu.cu->imv];
if (uiCostBest <= AFFINE_ME_LIST_MVP_TH*m_hevcCost)
{
+#if !JVET_O0592_ENC_ME_IMP
//search 8 nearest neighbors; integer distance
int testPos[8][2] = { { -1, 0 },{ 0, -1 },{ 0, 1 },{ 1, 0 },{ -1, -1 },{ -1, 1 },{ 1, 1 },{ 1, -1 } };
const int maxSearchRound = 3;
@@ -5491,6 +5723,7 @@ void InterSearch::xAffineMotionEstimation( PredictionUnit& pu,
}
}
}
+#endif
Mv mvPredTmp[3] = { acMvPred[0], acMvPred[1], acMvPred[2] };
Mv mvME[3];
@@ -5535,6 +5768,57 @@ void InterSearch::xAffineMotionEstimation( PredictionUnit& pu,
checkCPMVRdCost(acMvTemp);
}
+#if JVET_O0592_ENC_ME_IMP
+ // 8 nearest neighbor search
+ int testPos[8][2] = { { -1, 0 },{ 0, -1 },{ 0, 1 },{ 1, 0 },{ -1, -1 },{ -1, 1 },{ 1, 1 },{ 1, -1 } };
+ const int maxSearchRound = (pu.cu->imv) ? 3 : ((m_pcEncCfg->getUseAffineAmvrEncOpt() && m_pcEncCfg->getIntraPeriod() == (uint32_t)-1) ? 2 : 3);
+
+ for (int rnd = 0; rnd < maxSearchRound; rnd++)
+ {
+ bool modelChange = false;
+ //search the model parameters with finear granularity;
+ for (int j = 0; j < mvNum; j++)
+ {
+ bool loopChange = false;
+ for (int iter = 0; iter < 2; iter++)
+ {
+ if (iter == 1 && !loopChange)
+ {
+ break;
+ }
+ Mv centerMv[3];
+ memcpy(centerMv, acMv, sizeof(Mv) * 3);
+ memcpy(acMvTemp, acMv, sizeof(Mv) * 3);
+
+ for (int i = ((iter == 0) ? 0 : 4); i < ((iter == 0) ? 4 : 8); i++)
+ {
+ acMvTemp[j].set(centerMv[j].getHor() + (testPos[i][0] << mvShift), centerMv[j].getVer() + (testPos[i][1] << mvShift));
+ clipMv(acMvTemp[j], pu.cu->lumaPos(), pu.cu->lumaSize(), *pu.cs->sps);
+ xPredAffineBlk(COMPONENT_Y, pu, refPic, acMvTemp, predBuf, false, pu.cu->slice->clpRng(COMPONENT_Y));
+
+ Distortion costTemp = m_pcRdCost->getDistPart(predBuf.Y(), pBuf->Y(), pu.cs->sps->getBitDepth(CHANNEL_TYPE_LUMA), COMPONENT_Y, distFunc);
+ uint32_t bitsTemp = ruiBits;
+ bitsTemp += xCalcAffineMVBits(pu, acMvTemp, acMvPred);
+ costTemp = (Distortion)(floor(fWeight * (double)costTemp) + (double)m_pcRdCost->getCost(bitsTemp));
+
+ if (costTemp < uiCostBest)
+ {
+ uiCostBest = costTemp;
+ uiBitsBest = bitsTemp;
+ ::memcpy(acMv, acMvTemp, sizeof(Mv) * 3);
+ modelChange = true;
+ loopChange = true;
+ }
+ }
+ }
+ }
+
+ if (!modelChange)
+ {
+ break;
+ }
+ }
+#else
{
const int threshold = 4 << (MV_PRECISION_INTERNAL - MV_PRECISION_SIXTEENTH);
dMv = acMv[1] - acMv[0];
@@ -5551,6 +5835,7 @@ void InterSearch::xAffineMotionEstimation( PredictionUnit& pu,
}
}
}
+#endif
}
acMvPred[0] = aamvpi.mvCandLT[mvpIdx];
acMvPred[1] = aamvpi.mvCandRT[mvpIdx];
diff --git a/source/Lib/EncoderLib/InterSearch.h b/source/Lib/EncoderLib/InterSearch.h
index fa1541039..105d51808 100644
--- a/source/Lib/EncoderLib/InterSearch.h
+++ b/source/Lib/EncoderLib/InterSearch.h
@@ -78,6 +78,14 @@ struct AffineMVInfo
int x, y, w, h;
};
+#if JVET_O0592_ENC_ME_IMP
+struct BlkUniMvInfo
+{
+ Mv uniMvs[2][33];
+ int x, y, w, h;
+};
+#endif
+
typedef struct
{
Mv acMvAffine4Para[2][3];
@@ -116,6 +124,12 @@ private:
int m_affMVListIdx;
int m_affMVListSize;
int m_affMVListMaxSize;
+#if JVET_O0592_ENC_ME_IMP
+ BlkUniMvInfo* m_uniMvList;
+ int m_uniMvListIdx;
+ int m_uniMvListSize;
+ int m_uniMvListMaxSize;
+#endif
Distortion m_hevcCost;
EncAffineMotion m_affineMotion;
PatentBvCand m_defaultCachedBvs;
@@ -232,6 +246,80 @@ public:
m_affMVListSize = std::min(m_affMVListSize + 1, m_affMVListMaxSize);
}
}
+#if JVET_O0592_ENC_ME_IMP
+ void resetUniMvList() { m_uniMvListIdx = 0; m_uniMvListSize = 0; }
+ void insertUniMvCands(CompArea blkArea, Mv cMvTemp[2][33])
+ {
+ BlkUniMvInfo* curMvInfo = m_uniMvList + m_uniMvListIdx;
+ int j = 0;
+ for (; j < m_uniMvListSize; j++)
+ {
+ BlkUniMvInfo* prevMvInfo = m_uniMvList + ((m_uniMvListIdx - 1 - j + m_uniMvListMaxSize) % (m_uniMvListMaxSize));
+ if ((blkArea.x == prevMvInfo->x) && (blkArea.y == prevMvInfo->y) && (blkArea.width == prevMvInfo->w) && (blkArea.height == prevMvInfo->h))
+ {
+ break;
+ }
+ }
+
+ if (j < m_uniMvListSize)
+ {
+ curMvInfo = m_uniMvList + ((m_uniMvListIdx - 1 - j + m_uniMvListMaxSize) % (m_uniMvListMaxSize));
+ }
+
+ ::memcpy(curMvInfo->uniMvs, cMvTemp, 2 * 33 * sizeof(Mv));
+ if (j == m_uniMvListSize) // new element
+ {
+ curMvInfo->x = blkArea.x;
+ curMvInfo->y = blkArea.y;
+ curMvInfo->w = blkArea.width;
+ curMvInfo->h = blkArea.height;
+ m_uniMvListSize = std::min(m_uniMvListSize + 1, m_uniMvListMaxSize);
+ m_uniMvListIdx = (m_uniMvListIdx + 1) % (m_uniMvListMaxSize);
+ }
+ }
+ void savePrevUniMvInfo(CompArea blkArea, BlkUniMvInfo &tmpUniMvInfo, bool& isUniMvInfoSaved)
+ {
+ int j = 0;
+ BlkUniMvInfo* curUniMvInfo = nullptr;
+ for (; j < m_uniMvListSize; j++)
+ {
+ curUniMvInfo = m_uniMvList + ((m_uniMvListIdx - 1 - j + m_uniMvListMaxSize) % (m_uniMvListMaxSize));
+ if ((blkArea.x == curUniMvInfo->x) && (blkArea.y == curUniMvInfo->y) && (blkArea.width == curUniMvInfo->w) && (blkArea.height == curUniMvInfo->h))
+ {
+ break;
+ }
+ }
+
+ if (j < m_uniMvListSize)
+ {
+ isUniMvInfoSaved = true;
+ tmpUniMvInfo = *curUniMvInfo;
+ }
+ }
+ void addUniMvInfo(BlkUniMvInfo &tmpUniMVInfo)
+ {
+ int j = 0;
+ BlkUniMvInfo* prevUniMvInfo = nullptr;
+ for (; j < m_uniMvListSize; j++)
+ {
+ prevUniMvInfo = m_uniMvList + ((m_uniMvListIdx - 1 - j + m_uniMvListMaxSize) % (m_uniMvListMaxSize));
+ if ((tmpUniMVInfo.x == prevUniMvInfo->x) && (tmpUniMVInfo.y == prevUniMvInfo->y) && (tmpUniMVInfo.w == prevUniMvInfo->w) && (tmpUniMVInfo.h == prevUniMvInfo->h))
+ {
+ break;
+ }
+ }
+ if (j < m_uniMvListSize)
+ {
+ *prevUniMvInfo = tmpUniMVInfo;
+ }
+ else
+ {
+ m_uniMvList[m_uniMvListIdx] = tmpUniMVInfo;
+ m_uniMvListIdx = (m_uniMvListIdx + 1) % m_uniMvListMaxSize;
+ m_uniMvListSize = std::min(m_uniMvListSize + 1, m_uniMvListMaxSize);
+ }
+ }
+#endif
void resetSavedAffineMotion();
void storeAffineMotion( Mv acAffineMv[2][3], int16_t affineRefIdx[2], EAffineModel affineType, int gbiIdx );
protected:
@@ -364,6 +452,10 @@ protected:
);
void xTZSearch ( const PredictionUnit& pu,
+#if JVET_O0592_ENC_ME_IMP
+ RefPicList eRefPicList,
+ int iRefIdxPred,
+#endif
IntTZSearchStruct& cStruct,
Mv& rcMv,
Distortion& ruiSAD,
@@ -373,6 +465,10 @@ protected:
);
void xTZSearchSelective ( const PredictionUnit& pu,
+#if JVET_O0592_ENC_ME_IMP
+ RefPicList eRefPicList,
+ int iRefIdxPred,
+#endif
IntTZSearchStruct& cStruct,
Mv& rcMv,
Distortion& ruiSAD,
@@ -387,6 +483,10 @@ protected:
);
void xPatternSearchFast ( const PredictionUnit& pu,
+#if JVET_O0592_ENC_ME_IMP
+ RefPicList eRefPicList,
+ int iRefIdxPred,
+#endif
IntTZSearchStruct& cStruct,
Mv& rcMv,
Distortion& ruiSAD,
--
GitLab