EncCu.cpp

        pu.mergeFlag = true;

#if JVET_M0883_TRIANGLE_SIGNALING
        PU::spanTriangleMotionInfo(pu, triangleMrgCtx, splitDir, candIdx0, candIdx1 );
#else
        PU::spanTriangleMotionInfo(pu, triangleMrgCtx, mergeCand, splitDir, candIdx0, candIdx1 );
#endif

        if( tempBufSet )
        {
          tempCS->getPredBuf().copyFrom( triangleWeightedBuffer[mergeCand] );
        }
        else
        {
          triangleBuffer[candIdx0] = m_acMergeBuffer[candIdx0].getBuf( localUnitArea );
          triangleBuffer[candIdx1] = m_acMergeBuffer[candIdx1].getBuf( localUnitArea );
          PelUnitBuf predBuf         = tempCS->getPredBuf();
#if JVET_M0328_KEEP_ONE_WEIGHT_GROUP
          m_pcInterSearch->weightedTriangleBlk( pu, splitDir, MAX_NUM_CHANNEL_TYPE, predBuf, triangleBuffer[candIdx0], triangleBuffer[candIdx1] );
#else
          m_pcInterSearch->weightedTriangleBlk( pu, PU::getTriangleWeights(pu, triangleMrgCtx, candIdx0, candIdx1), splitDir, MAX_NUM_CHANNEL_TYPE, predBuf, triangleBuffer[candIdx0], triangleBuffer[candIdx1] );
#endif
        }

#if JVET_M0464_UNI_MTS
        xEncodeInterResidual( tempCS, bestCS, partitioner, encTestMode, noResidualPass, NULL, ( noResidualPass == 0 ? &trianglecandHasNoResidual[mergeCand] : NULL ) );
#else
        xEncodeInterResidual( tempCS, bestCS, partitioner, encTestMode, noResidualPass, NULL, true, ( (noResidualPass == 0 ) ? &trianglecandHasNoResidual[mergeCand] : NULL ) );
#endif

        if (m_pcEncCfg->getUseFastDecisionForMerge() && !bestIsSkip)
        {
          bestIsSkip = bestCS->getCU(partitioner.chType)->rootCbf == 0;
        }
        tempCS->initStructData(encTestMode.qp, encTestMode.lossless);
      }// end loop mrgHADIdx
    }
  }
#if JVET_M0428_ENC_DB_OPT
  if ( m_bestModeUpdated && bestCS->cost != MAX_DOUBLE )
  {
    xCalDebCost( *bestCS, partitioner );
  }
#endif
}

void EncCu::xCheckRDCostAffineMerge2Nx2N( CodingStructure *&tempCS, CodingStructure *&bestCS, Partitioner &partitioner, const EncTestMode& encTestMode )
{
  if( m_modeCtrl->getFastDeltaQp() )
  {
    return;
  }

  if ( bestCS->area.lumaSize().width < 8 || bestCS->area.lumaSize().height < 8 )
  {
    return;
  }
#if JVET_M0428_ENC_DB_OPT
  m_bestModeUpdated = tempCS->useDbCost = bestCS->useDbCost = false;
#endif
  const Slice &slice = *tempCS->slice;

  CHECK( slice.getSliceType() == I_SLICE, "Affine Merge modes not available for I-slices" );

  tempCS->initStructData( encTestMode.qp, encTestMode.lossless );

  AffineMergeCtx affineMergeCtx;
  const SPS &sps = *tempCS->sps;

  MergeCtx mrgCtx;
  if ( sps.getSBTMVPEnabledFlag() )
  {
    Size bufSize = g_miScaling.scale( tempCS->area.lumaSize() );
    mrgCtx.subPuMvpMiBuf = MotionBuf( m_SubPuMiBuf, bufSize );
    affineMergeCtx.mrgCtx = &mrgCtx;
  }

  {
    // first get merge candidates
    CodingUnit cu( tempCS->area );
    cu.cs = tempCS;
    cu.predMode = MODE_INTER;
    cu.slice = tempCS->slice;
#if HEVC_TILES_WPP
    cu.tileIdx = tempCS->picture->tileMap->getTileIdxMap( tempCS->area.lumaPos() );
#endif
    cu.mmvdSkip = false;

    PredictionUnit pu( tempCS->area );
    pu.cu = &cu;
    pu.cs = tempCS;

    PU::getAffineMergeCand( pu, affineMergeCtx );

    if ( affineMergeCtx.numValidMergeCand <= 0 )
    {
      return;
    }
  }

  bool candHasNoResidual[AFFINE_MRG_MAX_NUM_CANDS];
  for ( uint32_t ui = 0; ui < affineMergeCtx.numValidMergeCand; ui++ )
  {
    candHasNoResidual[ui] = false;
  }

  bool                                        bestIsSkip = false;
  uint32_t                                    uiNumMrgSATDCand = affineMergeCtx.numValidMergeCand;
  PelUnitBuf                                  acMergeBuffer[AFFINE_MRG_MAX_NUM_CANDS];
  static_vector<uint32_t, AFFINE_MRG_MAX_NUM_CANDS>  RdModeList;
  bool                                        mrgTempBufSet = false;

  for ( uint32_t i = 0; i < AFFINE_MRG_MAX_NUM_CANDS; i++ )
  {
    RdModeList.push_back( i );
  }

  if ( m_pcEncCfg->getUseFastMerge() )
  {
    uiNumMrgSATDCand = std::min( NUM_AFF_MRG_SATD_CAND, affineMergeCtx.numValidMergeCand );
    bestIsSkip = false;

    if ( auto blkCache = dynamic_cast<CacheBlkInfoCtrl*>(m_modeCtrl) )
    {
      bestIsSkip = blkCache->isSkip( tempCS->area );
    }

    static_vector<double, AFFINE_MRG_MAX_NUM_CANDS> candCostList;

    // 1. Pass: get SATD-cost for selected candidates and reduce their count
    if ( !bestIsSkip )
    {
      RdModeList.clear();
      mrgTempBufSet = true;
      const double sqrtLambdaForFirstPass = m_pcRdCost->getMotionLambda( encTestMode.lossless );

      CodingUnit &cu = tempCS->addCU( tempCS->area, partitioner.chType );

      partitioner.setCUData( cu );
      cu.slice = tempCS->slice;
#if HEVC_TILES_WPP
      cu.tileIdx = tempCS->picture->tileMap->getTileIdxMap( tempCS->area.lumaPos() );
#endif
      cu.skip = false;
      cu.affine = true;
      cu.predMode = MODE_INTER;
      cu.transQuantBypass = encTestMode.lossless;
      cu.chromaQpAdj = cu.transQuantBypass ? 0 : m_cuChromaQpOffsetIdxPlus1;
      cu.qp = encTestMode.qp;

      PredictionUnit &pu = tempCS->addPU( cu, partitioner.chType );

      DistParam distParam;
      const bool bUseHadamard = !encTestMode.lossless;
      m_pcRdCost->setDistParam( distParam, tempCS->getOrgBuf().Y(), m_acMergeBuffer[0].Y(), sps.getBitDepth( CHANNEL_TYPE_LUMA ), COMPONENT_Y, bUseHadamard );

      const UnitArea localUnitArea( tempCS->area.chromaFormat, Area( 0, 0, tempCS->area.Y().width, tempCS->area.Y().height ) );

      for ( uint32_t uiMergeCand = 0; uiMergeCand < affineMergeCtx.numValidMergeCand; uiMergeCand++ )
      {
        acMergeBuffer[uiMergeCand] = m_acMergeBuffer[uiMergeCand].getBuf( localUnitArea );

        // set merge information
        pu.interDir = affineMergeCtx.interDirNeighbours[uiMergeCand];
        pu.mergeFlag = true;
        pu.mergeIdx = uiMergeCand;
        cu.affineType = affineMergeCtx.affineType[uiMergeCand];
        cu.GBiIdx = affineMergeCtx.GBiIdx[uiMergeCand];

        pu.mergeType = affineMergeCtx.mergeType[uiMergeCand];
        if ( pu.mergeType == MRG_TYPE_SUBPU_ATMVP )
        {
          pu.refIdx[0] = affineMergeCtx.mvFieldNeighbours[(uiMergeCand << 1) + 0][0].refIdx;
          pu.refIdx[1] = affineMergeCtx.mvFieldNeighbours[(uiMergeCand << 1) + 1][0].refIdx;
          PU::spanMotionInfo( pu, mrgCtx );
        }
        else
        {
          PU::setAllAffineMvField( pu, affineMergeCtx.mvFieldNeighbours[(uiMergeCand << 1) + 0], REF_PIC_LIST_0 );
          PU::setAllAffineMvField( pu, affineMergeCtx.mvFieldNeighbours[(uiMergeCand << 1) + 1], REF_PIC_LIST_1 );

          PU::spanMotionInfo( pu );
        }

        distParam.cur = acMergeBuffer[uiMergeCand].Y();

        m_pcInterSearch->motionCompensation( pu, acMergeBuffer[uiMergeCand] );

        Distortion uiSad = distParam.distFunc( distParam );
        uint32_t   uiBitsCand = uiMergeCand + 1;
        if ( uiMergeCand == tempCS->slice->getMaxNumAffineMergeCand() - 1 )
        {
          uiBitsCand--;
        }
        double cost = (double)uiSad + (double)uiBitsCand * sqrtLambdaForFirstPass;
        static_vector<int, AFFINE_MRG_MAX_NUM_CANDS> emptyList;
        updateCandList( uiMergeCand, cost, RdModeList, candCostList
          , emptyList, -1
          , uiNumMrgSATDCand );

        CHECK( std::min( uiMergeCand + 1, uiNumMrgSATDCand ) != RdModeList.size(), "" );
      }

      // Try to limit number of candidates using SATD-costs
      for ( uint32_t i = 1; i < uiNumMrgSATDCand; i++ )
      {
        if ( candCostList[i] > MRG_FAST_RATIO * candCostList[0] )
        {
          uiNumMrgSATDCand = i;
          break;
        }
      }

      tempCS->initStructData( encTestMode.qp, encTestMode.lossless );
    }
    else
    {
      uiNumMrgSATDCand = affineMergeCtx.numValidMergeCand;
    }
  }

#if !JVET_M0253_HASH_ME
  const uint32_t iteration = encTestMode.lossless ? 1 : 2;

  // 2. Pass: check candidates using full RD test
  for ( uint32_t uiNoResidualPass = 0; uiNoResidualPass < iteration; uiNoResidualPass++ )
#else
  uint32_t iteration;
  uint32_t iterationBegin = m_modeCtrl->getIsHashPerfectMatch() ? 1 : 0;
  if (encTestMode.lossless)
  {
    iteration = 1;
    iterationBegin = 0;
  }
  else
  {
    iteration = 2;
  }
  for (uint32_t uiNoResidualPass = iterationBegin; uiNoResidualPass < iteration; ++uiNoResidualPass)
#endif
  {
    for ( uint32_t uiMrgHADIdx = 0; uiMrgHADIdx < uiNumMrgSATDCand; uiMrgHADIdx++ )
    {
      uint32_t uiMergeCand = RdModeList[uiMrgHADIdx];

      if ( ((uiNoResidualPass != 0) && candHasNoResidual[uiMergeCand])
        || ((uiNoResidualPass == 0) && bestIsSkip) )
      {
        continue;
      }

      // first get merge candidates
      CodingUnit &cu = tempCS->addCU( tempCS->area, partitioner.chType );

      partitioner.setCUData( cu );
      cu.slice = tempCS->slice;
#if HEVC_TILES_WPP
      cu.tileIdx = tempCS->picture->tileMap->getTileIdxMap( tempCS->area.lumaPos() );
#endif
      cu.skip = false;
      cu.affine = true;
      cu.predMode = MODE_INTER;
      cu.transQuantBypass = encTestMode.lossless;
      cu.chromaQpAdj = cu.transQuantBypass ? 0 : m_cuChromaQpOffsetIdxPlus1;
      cu.qp = encTestMode.qp;
      PredictionUnit &pu = tempCS->addPU( cu, partitioner.chType );

      // set merge information
      pu.mergeFlag = true;
      pu.mergeIdx = uiMergeCand;
      pu.interDir = affineMergeCtx.interDirNeighbours[uiMergeCand];
      cu.affineType = affineMergeCtx.affineType[uiMergeCand];
      cu.GBiIdx = affineMergeCtx.GBiIdx[uiMergeCand];

      pu.mergeType = affineMergeCtx.mergeType[uiMergeCand];
      if ( pu.mergeType == MRG_TYPE_SUBPU_ATMVP )
      {
        pu.refIdx[0] = affineMergeCtx.mvFieldNeighbours[(uiMergeCand << 1) + 0][0].refIdx;
        pu.refIdx[1] = affineMergeCtx.mvFieldNeighbours[(uiMergeCand << 1) + 1][0].refIdx;
        PU::spanMotionInfo( pu, mrgCtx );
      }
      else
      {
        PU::setAllAffineMvField( pu, affineMergeCtx.mvFieldNeighbours[(uiMergeCand << 1) + 0], REF_PIC_LIST_0 );
        PU::setAllAffineMvField( pu, affineMergeCtx.mvFieldNeighbours[(uiMergeCand << 1) + 1], REF_PIC_LIST_1 );

        PU::spanMotionInfo( pu );
      }

      if ( mrgTempBufSet )
      {
        tempCS->getPredBuf().copyFrom( acMergeBuffer[uiMergeCand] );
      }
      else
      {
        m_pcInterSearch->motionCompensation( pu );
      }

#if JVET_M0464_UNI_MTS
      xEncodeInterResidual( tempCS, bestCS, partitioner, encTestMode, uiNoResidualPass, NULL, ( uiNoResidualPass == 0 ? &candHasNoResidual[uiMergeCand] : NULL ) );
#else
      xEncodeInterResidual( tempCS, bestCS, partitioner, encTestMode, uiNoResidualPass, NULL, true, ((uiNoResidualPass == 0) ? &candHasNoResidual[uiMergeCand] : NULL) );
#endif

      if ( m_pcEncCfg->getUseFastDecisionForMerge() && !bestIsSkip )
      {
        bestIsSkip = bestCS->getCU( partitioner.chType )->rootCbf == 0;
      }
      tempCS->initStructData( encTestMode.qp, encTestMode.lossless );
    }// end loop uiMrgHADIdx

    if ( uiNoResidualPass == 0 && m_pcEncCfg->getUseEarlySkipDetection() )
    {
      const CodingUnit     &bestCU = *bestCS->getCU( partitioner.chType );
      const PredictionUnit &bestPU = *bestCS->getPU( partitioner.chType );

      if ( bestCU.rootCbf == 0 )
      {
        if ( bestPU.mergeFlag )
        {
          m_modeCtrl->setEarlySkipDetected();
        }
        else if ( m_pcEncCfg->getMotionEstimationSearchMethod() != MESEARCH_SELECTIVE )
        {
          int absolute_MV = 0;

          for ( uint32_t uiRefListIdx = 0; uiRefListIdx < 2; uiRefListIdx++ )
          {
            if ( slice.getNumRefIdx( RefPicList( uiRefListIdx ) ) > 0 )
            {
              absolute_MV += bestPU.mvd[uiRefListIdx].getAbsHor() + bestPU.mvd[uiRefListIdx].getAbsVer();
            }
          }

          if ( absolute_MV == 0 )
          {
            m_modeCtrl->setEarlySkipDetected();
          }
        }
      }
    }
  }
#if JVET_M0428_ENC_DB_OPT
  if ( m_bestModeUpdated && bestCS->cost != MAX_DOUBLE )
  {
    xCalDebCost( *bestCS, partitioner );
  }
#endif
}
//////////////////////////////////////////////////////////////////////////////////////////////
// ibc merge/skip mode check
void EncCu::xCheckRDCostIBCModeMerge2Nx2N(CodingStructure *&tempCS, CodingStructure *&bestCS, Partitioner &partitioner, const EncTestMode& encTestMode)
{
  assert(tempCS->chType != CHANNEL_TYPE_CHROMA); // chroma IBC is derived

  if (tempCS->area.lwidth() > IBC_MAX_CAND_SIZE || tempCS->area.lheight() > IBC_MAX_CAND_SIZE) // currently only check 32x32 and below block for ibc merge/skip
  {
    return;
  }
  const SPS &sps = *tempCS->sps;

  tempCS->initStructData(encTestMode.qp, encTestMode.lossless);
  MergeCtx mergeCtx;


  if (sps.getSBTMVPEnabledFlag())
  {
    Size bufSize = g_miScaling.scale(tempCS->area.lumaSize());
    mergeCtx.subPuMvpMiBuf = MotionBuf(m_SubPuMiBuf, bufSize);
  }

  {
    // first get merge candidates
    CodingUnit cu(tempCS->area);
    cu.cs = tempCS;
#if JVET_M0483_IBC
    cu.predMode = MODE_IBC;
#else
    cu.predMode = MODE_INTER;
    cu.ibc = true;
#endif
    cu.slice = tempCS->slice;
#if HEVC_TILES_WPP
    cu.tileIdx = tempCS->picture->tileMap->getTileIdxMap(tempCS->area.lumaPos());
#endif
    PredictionUnit pu(tempCS->area);
    pu.cu = &cu;
    pu.cs = tempCS;
    cu.mmvdSkip = false;
    pu.mmvdMergeFlag = false;
    cu.triangle = false;
#if JVET_M0170_MRG_SHARELIST
    pu.shareParentPos = tempCS->sharedBndPos;
    pu.shareParentSize = tempCS->sharedBndSize;
#endif
#if JVET_M0483_IBC
    PU::getIBCMergeCandidates(pu, mergeCtx);
#else
    PU::getInterMergeCandidates(pu, mergeCtx
      , 0
    );
#endif
  }

  int candHasNoResidual[MRG_MAX_NUM_CANDS];
  for (unsigned int ui = 0; ui < mergeCtx.numValidMergeCand; ui++)
  {
    candHasNoResidual[ui] = 0;
  }

  bool                                        bestIsSkip = false;
  unsigned                                    numMrgSATDCand = mergeCtx.numValidMergeCand;
  static_vector<unsigned, MRG_MAX_NUM_CANDS>  RdModeList(MRG_MAX_NUM_CANDS);
  for (unsigned i = 0; i < MRG_MAX_NUM_CANDS; i++)
  {
    RdModeList[i] = i;
  }

  //{
    static_vector<double, MRG_MAX_NUM_CANDS>  candCostList(MRG_MAX_NUM_CANDS, MAX_DOUBLE);
    // 1. Pass: get SATD-cost for selected candidates and reduce their count
    {
      const double sqrtLambdaForFirstPass = m_pcRdCost->getMotionLambda(encTestMode.lossless);

      CodingUnit &cu = tempCS->addCU(CS::getArea(*tempCS, tempCS->area, (const ChannelType)partitioner.chType), (const ChannelType)partitioner.chType);

      partitioner.setCUData(cu);
      cu.slice = tempCS->slice;
#if HEVC_TILES_WPP
      cu.tileIdx = tempCS->picture->tileMap->getTileIdxMap(tempCS->area.lumaPos());
#endif
      cu.skip = false;
#if JVET_M0483_IBC
      cu.predMode = MODE_IBC;
#else
      cu.predMode = MODE_INTER;
      cu.ibc = true;
#endif
      cu.transQuantBypass = encTestMode.lossless;
      cu.chromaQpAdj = cu.transQuantBypass ? 0 : m_cuChromaQpOffsetIdxPlus1;
      cu.qp = encTestMode.qp;
      cu.mmvdSkip = false;
      cu.triangle = false;
      DistParam distParam;
      const bool bUseHadamard = !encTestMode.lossless;
      PredictionUnit &pu = tempCS->addPU(cu, partitioner.chType); //tempCS->addPU(cu);
      pu.mmvdMergeFlag = false;
      Picture* refPic = pu.cu->slice->getPic();
      const CPelBuf refBuf = refPic->getRecoBuf(pu.blocks[COMPONENT_Y]);
      const Pel*        piRefSrch = refBuf.buf;
#if JVET_M0427_INLOOP_RESHAPER
      if (tempCS->slice->getReshapeInfo().getUseSliceReshaper() && m_pcReshape->getCTUFlag())
      {
        const CompArea &area = cu.blocks[COMPONENT_Y];
        CompArea    tmpArea(COMPONENT_Y, area.chromaFormat, Position(0, 0), area.size());
        PelBuf tmpLuma = m_tmpStorageLCU->getBuf(tmpArea);
        tmpLuma.copyFrom(tempCS->getOrgBuf().Y());
        tmpLuma.rspSignal(m_pcReshape->getFwdLUT());
        m_pcRdCost->setDistParam(distParam, tmpLuma, refBuf, sps.getBitDepth(CHANNEL_TYPE_LUMA), COMPONENT_Y, bUseHadamard);
      }
      else
#endif
      m_pcRdCost->setDistParam(distParam, tempCS->getOrgBuf().Y(), refBuf, sps.getBitDepth(CHANNEL_TYPE_LUMA), COMPONENT_Y, bUseHadamard);
      int refStride = refBuf.stride;
      const UnitArea localUnitArea(tempCS->area.chromaFormat, Area(0, 0, tempCS->area.Y().width, tempCS->area.Y().height));
      int numValidBv = mergeCtx.numValidMergeCand;
      for (unsigned int mergeCand = 0; mergeCand < mergeCtx.numValidMergeCand; mergeCand++)
      {
#if JVET_M0483_IBC==0
        if (mergeCtx.interDirNeighbours[mergeCand] != 1)
        {
          numValidBv--;
          continue;
        }
        if (tempCS->slice->getRefPic(REF_PIC_LIST_0, mergeCtx.mvFieldNeighbours[mergeCand << 1].refIdx)->getPOC() != tempCS->slice->getPOC())
        {
          numValidBv--;
          continue;
        }
#endif
        mergeCtx.setMergeInfo(pu, mergeCand); // set bv info in merge mode
        const int cuPelX = pu.Y().x;
        const int cuPelY = pu.Y().y;
        int roiWidth = pu.lwidth();
        int roiHeight = pu.lheight();
        const int picWidth = pu.cs->slice->getSPS()->getPicWidthInLumaSamples();
        const int picHeight = pu.cs->slice->getSPS()->getPicHeightInLumaSamples();
        const unsigned int  lcuWidth = pu.cs->slice->getSPS()->getMaxCUWidth();
        int xPred = pu.bv.getHor();
        int yPred = pu.bv.getVer();

        if (!PU::isBlockVectorValid(pu, cuPelX, cuPelY, roiWidth, roiHeight, picWidth, picHeight, 0, 0, xPred, yPred, lcuWidth)) // not valid bv derived
        {
          numValidBv--;
          continue;
        }
        PU::spanMotionInfo(pu, mergeCtx);

        distParam.cur.buf = piRefSrch + refStride * yPred + xPred;

        Distortion sad = distParam.distFunc(distParam);
        unsigned int bitsCand = mergeCand + 1;
        if (mergeCand == tempCS->slice->getMaxNumMergeCand() - 1)
        {
          bitsCand--;
        }
        double cost = (double)sad + (double)bitsCand * sqrtLambdaForFirstPass;
        static_vector<int, MRG_MAX_NUM_CANDS> * nullList = nullptr;

        updateCandList(mergeCand, cost, RdModeList, candCostList
          , *nullList, -1
         , numMrgSATDCand);
      }

      // Try to limit number of candidates using SATD-costs
      if (numValidBv)
      {
        numMrgSATDCand = numValidBv;
        for (unsigned int i = 1; i < numValidBv; i++)
        {
          if (candCostList[i] > MRG_FAST_RATIO*candCostList[0])
          {
            numMrgSATDCand = i;
            break;
          }
        }
      }
      else
      {
        tempCS->dist = 0;
        tempCS->fracBits = 0;
        tempCS->cost = MAX_DOUBLE;
#if JVET_M0428_ENC_DB_OPT
        tempCS->costDbOffset = 0;
#endif
        tempCS->initStructData(encTestMode.qp, encTestMode.lossless);
        return;
      }

      tempCS->initStructData(encTestMode.qp, encTestMode.lossless);
    }
  //}


  const unsigned int iteration = encTestMode.lossless ? 1 : 2;
#if JVET_M0428_ENC_DB_OPT
  m_bestModeUpdated = tempCS->useDbCost = bestCS->useDbCost = false;
#endif
  // 2. Pass: check candidates using full RD test
  for (unsigned int numResidualPass = 0; numResidualPass < iteration; numResidualPass++)
  {
    for (unsigned int mrgHADIdx = 0; mrgHADIdx < numMrgSATDCand; mrgHADIdx++)
    {
      unsigned int mergeCand = RdModeList[mrgHADIdx];
#if JVET_M0483_IBC==0
      if (mergeCtx.interDirNeighbours[mergeCand] != 1)
      {
        continue;
      }
      if (tempCS->slice->getRefPic(REF_PIC_LIST_0, mergeCtx.mvFieldNeighbours[mergeCand << 1].refIdx)->getPOC() != tempCS->slice->getPOC())
      {
        continue;
      }
#endif
      if (!(numResidualPass == 1 && candHasNoResidual[mergeCand] == 1))
      {
        if (!(bestIsSkip && (numResidualPass == 0)))
        {
#if JVET_M0464_UNI_MTS
          {
#else
          unsigned char considerEmtSecondPass = 0;
          bool skipSecondEmtPass = true;
          bool hasResidual[2] = { false, false };
          double emtCost[2] = { MAX_DOUBLE, MAX_DOUBLE };

          // CU-level optimization
          for (unsigned char emtCuFlag = 0; emtCuFlag <= considerEmtSecondPass; emtCuFlag++)
          {
            if (m_pcEncCfg->getFastInterEMT() && emtCuFlag && skipSecondEmtPass)
            {
              continue;
            }
#endif

            // first get merge candidates
            CodingUnit &cu = tempCS->addCU(CS::getArea(*tempCS, tempCS->area, (const ChannelType)partitioner.chType), (const ChannelType)partitioner.chType);

            partitioner.setCUData(cu);
            cu.slice = tempCS->slice;
#if HEVC_TILES_WPP
            cu.tileIdx = tempCS->picture->tileMap->getTileIdxMap(tempCS->area.lumaPos());
#endif
            cu.skip = false;
#if JVET_M0483_IBC
            cu.predMode = MODE_IBC;
#else
            cu.predMode = MODE_INTER;
            cu.ibc = true;
#endif
            cu.transQuantBypass = encTestMode.lossless;
            cu.chromaQpAdj = cu.transQuantBypass ? 0 : m_cuChromaQpOffsetIdxPlus1;
            cu.qp = encTestMode.qp;
#if !JVET_M0464_UNI_MTS
            cu.emtFlag = false;
#endif
#if JVET_M0140_SBT
            cu.sbtInfo = 0;
#endif

            PredictionUnit &pu = tempCS->addPU(cu, partitioner.chType);// tempCS->addPU(cu);
            pu.intraDir[0] = DC_IDX; // set intra pred for ibc block
            pu.intraDir[1] = PLANAR_IDX; // set intra pred for ibc block
            cu.mmvdSkip = false;
            pu.mmvdMergeFlag = false;
            cu.triangle = false;
            mergeCtx.setMergeInfo(pu, mergeCand);
            PU::spanMotionInfo(pu, mergeCtx);

            assert(mergeCtx.mrgTypeNeighbours[mergeCand] == MRG_TYPE_IBC); //  should be IBC candidate at this round
            const bool chroma = !(CS::isDualITree(*tempCS));

            //  MC
            m_pcInterSearch->motionCompensation(pu,REF_PIC_LIST_0, true, chroma);
            m_CABACEstimator->getCtx() = m_CurrCtx->start;

            m_pcInterSearch->encodeResAndCalcRdInterCU(*tempCS, partitioner, (numResidualPass != 0), true, chroma);
            xEncodeDontSplit(*tempCS, partitioner);

            if (tempCS->pps->getUseDQP() && (partitioner.currDepth) <= tempCS->pps->getMaxCuDQPDepth())
            {
              xCheckDQP(*tempCS, partitioner);
            }

#if !JVET_M0464_UNI_MTS
            hasResidual[emtCuFlag] = cu.rootCbf;
            emtCost[emtCuFlag] = tempCS->cost;
#endif

            DTRACE_MODE_COST(*tempCS, m_pcRdCost->getLambda());
            xCheckBestMode(tempCS, bestCS, partitioner, encTestMode);

            tempCS->initStructData(encTestMode.qp, encTestMode.lossless);
          }
#if !JVET_M0464_UNI_MTS
          if (numResidualPass == 0 && (emtCost[0] <= emtCost[1] ? !hasResidual[0] : !hasResidual[1]))

            {
              // If no residual when allowing for one, then set mark to not try case where residual is forced to 0
              candHasNoResidual[mergeCand] = 1;
            }
#endif

            if (m_pcEncCfg->getUseFastDecisionForMerge() && !bestIsSkip)
            {
              if (bestCS->getCU(partitioner.chType) == NULL)
                bestIsSkip = 0;
              else
              bestIsSkip = bestCS->getCU(partitioner.chType)->rootCbf == 0;
            }
        }
      }
    }
  }
#if JVET_M0428_ENC_DB_OPT
  if ( m_bestModeUpdated && bestCS->cost != MAX_DOUBLE )
  {
    xCalDebCost( *bestCS, partitioner );
  }
#endif
}

void EncCu::xCheckRDCostIBCMode(CodingStructure *&tempCS, CodingStructure *&bestCS, Partitioner &partitioner, const EncTestMode& encTestMode)
{
  if (tempCS->area.lwidth() > IBC_MAX_CAND_SIZE || tempCS->area.lheight() > IBC_MAX_CAND_SIZE) // currently only check 32x32 and below block for ibc merge/skip
  {
    return;
  }

    tempCS->initStructData(encTestMode.qp, encTestMode.lossless);

#if JVET_M0428_ENC_DB_OPT
    m_bestModeUpdated = tempCS->useDbCost = bestCS->useDbCost = false;
#endif

    CodingUnit &cu = tempCS->addCU(CS::getArea(*tempCS, tempCS->area, partitioner.chType), partitioner.chType);

    partitioner.setCUData(cu);
    cu.slice = tempCS->slice;
#if HEVC_TILES_WPP
    cu.tileIdx = tempCS->picture->tileMap->getTileIdxMap(tempCS->area.lumaPos());
#endif
    cu.skip = false;
#if JVET_M0483_IBC
    cu.predMode = MODE_IBC;
#else
    cu.predMode = MODE_INTER;
#endif
    cu.transQuantBypass = encTestMode.lossless;
    cu.chromaQpAdj = cu.transQuantBypass ? 0 : m_cuChromaQpOffsetIdxPlus1;
    cu.qp = encTestMode.qp;
#if JVET_M0483_IBC==0
    cu.ibc = true;
#endif
    cu.imv = 0;
#if JVET_M0140_SBT
    cu.sbtInfo = 0;
#endif

    CU::addPUs(cu);

#if JVET_M0428_ENC_DB_OPT
    m_bestModeUpdated = tempCS->useDbCost = bestCS->useDbCost = false;
#endif

    PredictionUnit& pu = *cu.firstPU;
    cu.mmvdSkip = false;
    pu.mmvdMergeFlag = false;

    pu.intraDir[0] = DC_IDX; // set intra pred for ibc block
    pu.intraDir[1] = PLANAR_IDX; // set intra pred for ibc block

    pu.interDir = 1; // use list 0 for IBC mode
#if JVET_M0483_IBC
    pu.refIdx[REF_PIC_LIST_0] = MAX_NUM_REF; // last idx in the list
#else
    pu.refIdx[REF_PIC_LIST_0] = pu.cs->slice->getNumRefIdx(REF_PIC_LIST_0) - 1; // last idx in the list
#endif

    if (partitioner.chType == CHANNEL_TYPE_LUMA)
    {
      bool bValid = m_pcInterSearch->predIBCSearch(cu, partitioner, m_ctuIbcSearchRangeX, m_ctuIbcSearchRangeY, m_ibcHashMap);

      if (bValid)
      {
        PU::spanMotionInfo(pu);
        const bool chroma = !(CS::isDualITree(*tempCS));
        //  MC
        m_pcInterSearch->motionCompensation(pu, REF_PIC_LIST_0, true, chroma);

#if JVET_M0464_UNI_MTS
        {
#else
        double    bestCost = bestCS->cost;
        unsigned char    considerEmtSecondPass = 0;
        bool      skipSecondEmtPass = true;
        double    emtFirstPassCost = MAX_DOUBLE;

        // CU-level optimization

        for (unsigned char emtCuFlag = 0; emtCuFlag <= considerEmtSecondPass; emtCuFlag++)
        {
          if (m_pcEncCfg->getFastInterEMT() && emtCuFlag && skipSecondEmtPass)
          {
            continue;
          }

          tempCS->getCU(tempCS->chType)->emtFlag = emtCuFlag;
#endif

          m_pcInterSearch->encodeResAndCalcRdInterCU(*tempCS, partitioner, false, true, chroma);

#if !JVET_M0464_UNI_MTS
          if (m_pcEncCfg->getFastInterEMT())
          {
            emtFirstPassCost = (!emtCuFlag) ? tempCS->cost : emtFirstPassCost;
          }
#endif
          xEncodeDontSplit(*tempCS, partitioner);

          if (tempCS->pps->getUseDQP() && (partitioner.currDepth) <= tempCS->pps->getMaxCuDQPDepth())
          {
            xCheckDQP(*tempCS, partitioner);
          }

#if JVET_M0428_ENC_DB_OPT
          tempCS->useDbCost = m_pcEncCfg->getUseEncDbOpt();
          if ( m_bestModeUpdated )
          {
            xCalDebCost( *tempCS, partitioner );
          }
#endif

          DTRACE_MODE_COST(*tempCS, m_pcRdCost->getLambda());
          xCheckBestMode(tempCS, bestCS, partitioner, encTestMode);

#if !JVET_M0464_UNI_MTS
          //now we check whether the second pass should be skipped or not
          if (!emtCuFlag && considerEmtSecondPass)
          {
            static const double thresholdToSkipEmtSecondPass = 1.1; // Skip checking EMT transforms
            if (m_pcEncCfg->getFastInterEMT() && (!cu.firstTU->cbf[COMPONENT_Y] || emtFirstPassCost > bestCost * thresholdToSkipEmtSecondPass))
            {
              skipSecondEmtPass = true;
            }
            else //EMT will be checked
            {
              if (bestCost == bestCS->cost) //The first EMT pass didn't become the bestCS, so we clear the TUs generated
              {
                tempCS->clearTUs();
              }
              else
              {
                tempCS->initStructData(bestCS->currQP[bestCS->chType], bestCS->isLossless);

                tempCS->copyStructure(*bestCS, partitioner.chType);
                tempCS->getPredBuf().copyFrom(bestCS->getPredBuf());
              }

              //we need to restart the distortion for the new tempCS, the bit count and the cost
              tempCS->dist = 0;
              tempCS->fracBits = 0;
              tempCS->cost = MAX_DOUBLE;
#if JVET_M0428_ENC_DB_OPT
              tempCS->costDbOffset = 0;
#endif
            }
          }
#endif
        }

      } // bValid
      else
      {
        tempCS->dist = 0;
        tempCS->fracBits = 0;
        tempCS->cost = MAX_DOUBLE;
#if JVET_M0428_ENC_DB_OPT
        tempCS->costDbOffset = 0;
#endif
      }
    }
 // chroma CU ibc comp
    else
    {
      bool success = true;
      // chroma tree, reuse luma bv at minimal block level
      // enabled search only when each chroma sub-block has a BV from its luma sub-block
      assert(tempCS->getIbcLumaCoverage(pu.Cb()) == IBC_LUMA_COVERAGE_FULL);
      // check if each BV for the chroma sub-block is valid
      //static const UInt unitArea = MIN_PU_SIZE * MIN_PU_SIZE;
      const CompArea lumaArea = CompArea(COMPONENT_Y, pu.chromaFormat, pu.Cb().lumaPos(), recalcSize(pu.chromaFormat, CHANNEL_TYPE_CHROMA, CHANNEL_TYPE_LUMA, pu.Cb().size()));
      PredictionUnit subPu;
      subPu.cs = pu.cs;
      subPu.cu = pu.cu;
      const ComponentID compID = COMPONENT_Cb; // use Cb to represent both Cb and CR, as their structures are the same
      int shiftHor = ::getComponentScaleX(compID, pu.chromaFormat);
      int shiftVer = ::getComponentScaleY(compID, pu.chromaFormat);
      //const ChromaFormat  chFmt = pu.chromaFormat;

      for (int y = lumaArea.y; y < lumaArea.y + lumaArea.height; y += MIN_PU_SIZE)
      {
        for (int x = lumaArea.x; x < lumaArea.x + lumaArea.width; x += MIN_PU_SIZE)
        {
          const MotionInfo &curMi = pu.cs->picture->cs->getMotionInfo(Position{ x, y });

          subPu.UnitArea::operator=(UnitArea(pu.chromaFormat, Area(x, y, MIN_PU_SIZE, MIN_PU_SIZE)));
          Position offsetRef = subPu.blocks[compID].pos().offset((curMi.bv.getHor() >> shiftHor), (curMi.bv.getVer() >> shiftVer));
          Position refEndPos(offsetRef.x + subPu.blocks[compID].size().width - 1, offsetRef.y + subPu.blocks[compID].size().height - 1 );

          if (!subPu.cs->isDecomp(refEndPos, toChannelType(compID)) || !subPu.cs->isDecomp(offsetRef, toChannelType(compID))) // ref block is not yet available for this chroma sub-block
          {
            success = false;
            break;
          }
        }
        if (!success)
          break;
      }
      ////////////////////////////////////////////////////////////////////////////

      if (success)
      {
        //pu.mergeType = MRG_TYPE_IBC;
        m_pcInterSearch->motionCompensation(pu, REF_PIC_LIST_0, false, true); // luma=0, chroma=1
        m_pcInterSearch->encodeResAndCalcRdInterCU(*tempCS, partitioner, false, false, true);

        xEncodeDontSplit(*tempCS, partitioner);

        xCheckDQP(*tempCS, partitioner);
#if JVET_M0428_ENC_DB_OPT
        tempCS->useDbCost = m_pcEncCfg->getUseEncDbOpt();
        if ( m_bestModeUpdated )
        {
          xCalDebCost( *tempCS, partitioner );
        }
#endif

        DTRACE_MODE_COST(*tempCS, m_pcRdCost->getLambda());

        xCheckBestMode(tempCS, bestCS, partitioner, encTestMode);
      }
      else
      {
        tempCS->dist = 0;
        tempCS->fracBits = 0;
        tempCS->cost = MAX_DOUBLE;
#if JVET_M0428_ENC_DB_OPT
        tempCS->costDbOffset = 0;
#endif
      }
    }
  }
  // check ibc mode in encoder RD
  //////////////////////////////////////////////////////////////////////////////////////////////

void EncCu::xCheckRDCostInter( CodingStructure *&tempCS, CodingStructure *&bestCS, Partitioner &partitioner, const EncTestMode& encTestMode )
{
  tempCS->initStructData( encTestMode.qp, encTestMode.lossless );


  m_pcInterSearch->setAffineModeSelected(false);

  if( tempCS->slice->getCheckLDC() )
  {
    m_bestGbiCost[0] = m_bestGbiCost[1] = std::numeric_limits<double>::max();
    m_bestGbiIdx[0] = m_bestGbiIdx[1] = -1;
  }

  m_pcInterSearch->resetBufferedUniMotions();
  int gbiLoopNum = (tempCS->slice->isInterB() ? GBI_NUM : 1);
  gbiLoopNum = (tempCS->sps->getUseGBi() ? gbiLoopNum : 1);

  if( tempCS->area.lwidth() * tempCS->area.lheight() < GBI_SIZE_CONSTRAINT )
  {
    gbiLoopNum = 1;
  }

  double curBestCost = bestCS->cost;
  double equGBiCost = MAX_DOUBLE;

#if JVET_M0428_ENC_DB_OPT
  m_bestModeUpdated = tempCS->useDbCost = bestCS->useDbCost = false;
#endif

  for( int gbiLoopIdx = 0; gbiLoopIdx < gbiLoopNum; gbiLoopIdx++ )
  {
    if( m_pcEncCfg->getUseGBiFast() )
    {
      auto blkCache = dynamic_cast< CacheBlkInfoCtrl* >(m_modeCtrl);

      if( blkCache )
      {
        bool isBestInter = blkCache->getInter(bestCS->area);
        uint8_t bestGBiIdx = blkCache->getGbiIdx(bestCS->area);

        if( isBestInter && g_GbiSearchOrder[gbiLoopIdx] != GBI_DEFAULT && g_GbiSearchOrder[gbiLoopIdx] != bestGBiIdx )
        {
          continue;
        }
      }
    }
    if( !tempCS->slice->getCheckLDC() )
    {
      if( gbiLoopIdx != 0 && gbiLoopIdx != 3 && gbiLoopIdx != 4 )
      {
        continue;
      }
    }

  CodingUnit &cu      = tempCS->addCU( tempCS->area, partitioner.chType );

  partitioner.setCUData( cu );
  cu.slice            = tempCS->slice;
#if HEVC_TILES_WPP
  cu.tileIdx          = tempCS->picture->tileMap->getTileIdxMap( tempCS->area.lumaPos() );
#endif
  cu.skip             = false;
  cu.mmvdSkip = false;
//cu.affine
  cu.predMode         = MODE_INTER;
  cu.transQuantBypass = encTestMode.lossless;
  cu.chromaQpAdj      = cu.transQuantBypass ? 0 : m_cuChromaQpOffsetIdxPlus1;
  cu.qp               = encTestMode.qp;
  CU::addPUs( cu );

  cu.GBiIdx = g_GbiSearchOrder[gbiLoopIdx];
  uint8_t gbiIdx = cu.GBiIdx;
  bool  testGbi = (gbiIdx != GBI_DEFAULT);

  m_pcInterSearch->predInterSearch( cu, partitioner );

  const unsigned wIdx = gp_sizeIdxInfo->idxFrom( tempCS->area.lwidth () );

  gbiIdx = CU::getValidGbiIdx(cu);
  if( testGbi && gbiIdx == GBI_DEFAULT ) // Enabled GBi but the search results is uni.
  {
    tempCS->initStructData(encTestMode.qp, encTestMode.lossless);
    continue;
  }
  CHECK(!(testGbi || (!testGbi && gbiIdx == GBI_DEFAULT)), " !( bTestGbi || (!bTestGbi && gbiIdx == GBI_DEFAULT ) )");

  bool isEqualUni = false;
  if( m_pcEncCfg->getUseGBiFast() )
  {
    if( cu.firstPU->interDir != 3 && testGbi == 0 )
    {
      isEqualUni = true;
    }
  }