EncCu.cpp

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

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


  const unsigned int iteration = encTestMode.lossless ? 1 : 2;

  // 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

            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;
            }
        }
      }
    }
  }

}

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);

    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;

    CU::addPUs(cu);

    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);
          }

          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;
            }
          }
#endif
        }

      } // bValid
      else
      {
        tempCS->dist = 0;
        tempCS->fracBits = 0;
        tempCS->cost = MAX_DOUBLE;
      }
    }
 // 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);

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

        xCheckBestMode(tempCS, bestCS, partitioner, encTestMode);
      }
      else
      {
        tempCS->dist = 0;
        tempCS->fracBits = 0;
        tempCS->cost = MAX_DOUBLE;
      }
    }
  }
  // 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->getSpsNext().getUseGBi() ? gbiLoopNum : 1);

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

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

  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;
    }
  }

#if JVET_M0464_UNI_MTS
  xEncodeInterResidual( tempCS, bestCS, partitioner, encTestMode, 0
                        , m_pImvTempCS ? m_pImvTempCS[wIdx] : NULL
                        , 0
                        , &equGBiCost
#else
  xEncodeInterResidual( tempCS, bestCS, partitioner, encTestMode, 0
    , m_pImvTempCS ? m_pImvTempCS[wIdx] : NULL
    , 1
    , 0
    , &equGBiCost
#endif
  );

  if( g_GbiSearchOrder[gbiLoopIdx] == GBI_DEFAULT )
    m_pcInterSearch->setAffineModeSelected((bestCS->cus.front()->affine && !(bestCS->cus.front()->firstPU->mergeFlag)));

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

  double skipTH = MAX_DOUBLE;
  skipTH = (m_pcEncCfg->getUseGBiFast() ? 1.05 : MAX_DOUBLE);
  if( equGBiCost > curBestCost * skipTH )
  {
    break;
  }

  if( m_pcEncCfg->getUseGBiFast() )
  {
    if( isEqualUni == true && m_pcEncCfg->getIntraPeriod() == -1 )
    {
      break;
    }
  }
  if( g_GbiSearchOrder[gbiLoopIdx] == GBI_DEFAULT && xIsGBiSkip(cu) && m_pcEncCfg->getUseGBiFast() )
  {
    break;
  }
 }  // for( UChar gbiLoopIdx = 0; gbiLoopIdx < gbiLoopNum; gbiLoopIdx++ )
}





bool EncCu::xCheckRDCostInterIMV( CodingStructure *&tempCS, CodingStructure *&bestCS, Partitioner &partitioner, const EncTestMode& encTestMode )
{
  int iIMV = int( ( encTestMode.opts & ETO_IMV ) >> ETO_IMV_SHIFT );
  m_pcInterSearch->setAffineModeSelected(false);
  // Only int-Pel, 4-Pel and fast 4-Pel allowed
  CHECK( iIMV != 1 && iIMV != 2 && iIMV != 3, "Unsupported IMV Mode" );
  // Fast 4-Pel Mode

  EncTestMode encTestModeBase = encTestMode;                                        // copy for clearing non-IMV options
  encTestModeBase.opts        = EncTestModeOpts( encTestModeBase.opts & ETO_IMV );  // clear non-IMV options (is that intended?)

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

  CodingStructure* pcCUInfo2Reuse = nullptr;

  m_pcInterSearch->resetBufferedUniMotions();
  int gbiLoopNum = (tempCS->slice->isInterB() ? GBI_NUM : 1);
  gbiLoopNum = (pcCUInfo2Reuse != NULL ? 1 : gbiLoopNum);
  gbiLoopNum = (tempCS->slice->getSPS()->getSpsNext().getUseGBi() ? gbiLoopNum : 1);

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

#if JVET_M0246_AFFINE_AMVR
  bool validMode = false;
#endif
  double curBestCost = bestCS->cost;
  double equGBiCost = MAX_DOUBLE;

  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;
      }
    }

    if( m_pcEncCfg->getUseGBiFast() && tempCS->slice->getCheckLDC() && g_GbiSearchOrder[gbiLoopIdx] != GBI_DEFAULT
      && (m_bestGbiIdx[0] >= 0 && g_GbiSearchOrder[gbiLoopIdx] != m_bestGbiIdx[0])
      && (m_bestGbiIdx[1] >= 0 && g_GbiSearchOrder[gbiLoopIdx] != m_bestGbiIdx[1]))
    {
      continue;
    }

  CodingUnit &cu = ( pcCUInfo2Reuse != nullptr ) ? *tempCS->getCU( partitioner.chType ) : tempCS->addCU( tempCS->area, partitioner.chType );

  if( pcCUInfo2Reuse == nullptr )
  {
    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 );
  }
  else
  {
    CHECK( cu.skip,                                "Mismatch" );
    CHECK( cu.qtDepth  != partitioner.currQtDepth, "Mismatch" );
    CHECK( cu.btDepth  != partitioner.currBtDepth, "Mismatch" );
    CHECK( cu.mtDepth  != partitioner.currMtDepth, "Mismatch" );
    CHECK( cu.depth    != partitioner.currDepth,   "Mismatch" );
  }

  cu.imv      = iIMV > 1 ? 2 : 1;
#if !JVET_M0464_UNI_MTS
  cu.emtFlag  = false;
#endif

  bool testGbi;
  uint8_t gbiIdx;
#if JVET_M0246_AFFINE_AMVR
  bool affineAmvrEanbledFlag = cu.slice->getSPS()->getAffineAmvrEnabledFlag();
#endif
  
  if( pcCUInfo2Reuse != nullptr )
  {
    // reuse the motion info from pcCUInfo2Reuse
    CU::resetMVDandMV2Int( cu, m_pcInterSearch );

    CHECK(cu.GBiIdx < 0 || cu.GBiIdx >= GBI_NUM, "cu.GBiIdx < 0 || cu.GBiIdx >= GBI_NUM");
    gbiIdx = CU::getValidGbiIdx(cu);
    testGbi = (gbiIdx != GBI_DEFAULT);

#if JVET_M0246_AFFINE_AMVR
    if ( !CU::hasSubCUNonZeroMVd( cu ) && !CU::hasSubCUNonZeroAffineMVd( cu ) )
#else
    if( !CU::hasSubCUNonZeroMVd( cu ) )
#endif
    {
      if (m_modeCtrl->useModeResult(encTestModeBase, tempCS, partitioner))
      {
        std::swap(tempCS, bestCS);
        // store temp best CI for next CU coding
        m_CurrCtx->best = m_CABACEstimator->getCtx();
      }
#if JVET_M0246_AFFINE_AMVR
      if ( affineAmvrEanbledFlag )
      {
        tempCS->initStructData( encTestMode.qp, encTestMode.lossless );
        continue;
      }
      else
      {
        return false;
      }
#else
      return false;
#endif
    }
    else
    {
      m_pcInterSearch->motionCompensation( cu );
    }
  }
  else
  {
    cu.GBiIdx = g_GbiSearchOrder[gbiLoopIdx];
    gbiIdx = cu.GBiIdx;
    testGbi = (gbiIdx != GBI_DEFAULT);

#if JVET_M0246_AFFINE_AMVR
    cu.firstPU->interDir = 10;
#endif

    m_pcInterSearch->predInterSearch( cu, partitioner );

#if JVET_M0246_AFFINE_AMVR
    if ( cu.firstPU->interDir <= 3 )
    {
      gbiIdx = CU::getValidGbiIdx(cu);
    }
    else
    {
      return false;
    }
#else
    gbiIdx = CU::getValidGbiIdx(cu);
#endif
  }

  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;
    }
  }

#if JVET_M0246_AFFINE_AMVR
  if ( !CU::hasSubCUNonZeroMVd( cu ) && !CU::hasSubCUNonZeroAffineMVd( cu ) )
#else
  if( !CU::hasSubCUNonZeroMVd( cu ) )
#endif
  {
    if (m_modeCtrl->useModeResult(encTestModeBase, tempCS, partitioner))
    {
      std::swap(tempCS, bestCS);
      // store temp best CI for next CU coding
      m_CurrCtx->best = m_CABACEstimator->getCtx();
    }
#if JVET_M0246_AFFINE_AMVR
    if ( affineAmvrEanbledFlag )
    {
      tempCS->initStructData( encTestMode.qp, encTestMode.lossless );
      continue;
    }
    else
    {
      return false;
    }
#else
    return false;
#endif
  }

#if JVET_M0464_UNI_MTS
  xEncodeInterResidual( tempCS, bestCS, partitioner, encTestModeBase, 0
                        , NULL
                        , 0
                        , &equGBiCost
#else
  xEncodeInterResidual( tempCS, bestCS, partitioner, encTestModeBase, 0
    , NULL
    , true
    , 0
    , &equGBiCost
#endif
  );

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

  double skipTH = MAX_DOUBLE;
  skipTH = (m_pcEncCfg->getUseGBiFast() ? 1.05 : MAX_DOUBLE);
  if( equGBiCost > curBestCost * skipTH )
  {
    break;
  }

  if( m_pcEncCfg->getUseGBiFast() )
  {
    if( isEqualUni == true && m_pcEncCfg->getIntraPeriod() == -1 )
    {
      break;
    }
  }
  if( g_GbiSearchOrder[gbiLoopIdx] == GBI_DEFAULT && xIsGBiSkip(cu) && m_pcEncCfg->getUseGBiFast() )
  {
    break;
  }
#if JVET_M0246_AFFINE_AMVR
  validMode = true;
#endif
 } // for( UChar gbiLoopIdx = 0; gbiLoopIdx < gbiLoopNum; gbiLoopIdx++ )

#if JVET_M0246_AFFINE_AMVR
  return tempCS->slice->getSPS()->getAffineAmvrEnabledFlag() ? validMode : true;
#else
  return true;
#endif
}

#if JVET_M0464_UNI_MTS
void EncCu::xEncodeInterResidual(   CodingStructure *&tempCS
                                  , CodingStructure *&bestCS
                                  , Partitioner &partitioner
                                  , const EncTestMode& encTestMode
                                  , int residualPass
                                  , CodingStructure* imvCS
                                  , bool* bestHasNonResi
                                  , double* equGBiCost
#else
void EncCu::xEncodeInterResidual( CodingStructure *&tempCS, CodingStructure *&bestCS, Partitioner &partitioner, const EncTestMode& encTestMode, int residualPass
  , CodingStructure* imvCS
  , int emtMode
  , bool* bestHasNonResi
  , double* equGBiCost
#endif
  )
{
  if( residualPass == 1 && encTestMode.lossless )
  {
    return;
  }

  CodingUnit*            cu        = tempCS->getCU( partitioner.chType );
  double   bestCostInternal        = MAX_DOUBLE;
  double           bestCost        = bestCS->cost;
#if !JVET_M0464_UNI_MTS
  const SPS&            sps        = *tempCS->sps;
  const int      maxSizeEMT        = EMT_INTER_MAX_CU_WITH_QTBT;
#endif
  bool              swapped        = false; // avoid unwanted data copy
  bool             reloadCU        = false;
#if !JVET_M0464_UNI_MTS
  const bool considerEmtSecondPass = emtMode && sps.getSpsNext().getUseInterEMT() && partitioner.currArea().lwidth() <= maxSizeEMT && partitioner.currArea().lheight() <= maxSizeEMT;

  int minEMTMode = 0;
  int maxEMTMode = (considerEmtSecondPass?1:0);
#endif

  // Not allow very big |MVd| to avoid CABAC crash caused by too large MVd. Normally no impact on coding performance.
  const int maxMvd = 1 << 15;
  const PredictionUnit& pu = *cu->firstPU;
  if (!cu->affine)
  {
    if ((pu.refIdx[0] >= 0 && (pu.mvd[0].getAbsHor() >= maxMvd || pu.mvd[0].getAbsVer() >= maxMvd))
      || (pu.refIdx[1] >= 0 && (pu.mvd[1].getAbsHor() >= maxMvd || pu.mvd[1].getAbsVer() >= maxMvd)))
    {
      return;
    }
  }
  else
  {
    for (int refList = 0; refList < NUM_REF_PIC_LIST_01; refList++)
    {
      if (pu.refIdx[refList] >= 0)
      {
        for (int ctrlP = 1 + (cu->affineType == AFFINEMODEL_6PARAM); ctrlP >= 0; ctrlP--)
        {
          if (pu.mvdAffi[refList][ctrlP].getAbsHor() >= maxMvd || pu.mvdAffi[refList][ctrlP].getAbsVer() >= maxMvd)
          {
            return;
          }
        }
      }
    }
  }

#if !JVET_M0464_UNI_MTS
  if( emtMode == 2 )
  {
    minEMTMode = maxEMTMode = (cu->emtFlag?1:0);
  }

  for( int curEmtMode = minEMTMode; curEmtMode <= maxEMTMode; curEmtMode++ )
#endif
  {
    if( reloadCU )
    {
      if( bestCost == bestCS->cost ) //The first EMT pass didn't become the bestCS, so we clear the TUs generated
      {
        tempCS->clearTUs();
      }
      else if( false == swapped )
      {
        tempCS->initStructData( encTestMode.qp, encTestMode.lossless );
        tempCS->copyStructure( *bestCS, partitioner.chType );
        tempCS->getPredBuf().copyFrom( bestCS->getPredBuf() );
        bestCost = bestCS->cost;
        cu       = tempCS->getCU( partitioner.chType );
        swapped = true;
      }
      else
      {
        tempCS->clearTUs();
        bestCost = bestCS->cost;
        cu       = tempCS->getCU( partitioner.chType );
      }

      //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;
    }

    reloadCU    = true; // enable cu reloading
    cu->skip    = false;
#if !JVET_M0464_UNI_MTS
    cu->emtFlag = curEmtMode;
#endif

    const bool skipResidual = residualPass == 1;
    m_pcInterSearch->encodeResAndCalcRdInterCU( *tempCS, partitioner, skipResidual );

    xEncodeDontSplit( *tempCS, partitioner );

    xCheckDQP( *tempCS, partitioner );

    if( ETM_INTER_ME == encTestMode.type )
    {
      if( equGBiCost != NULL )
      {
        if( tempCS->cost < (*equGBiCost) && cu->GBiIdx == GBI_DEFAULT )
        {
          (*equGBiCost) = tempCS->cost;
        }
      }
      else
      {
        CHECK(equGBiCost == NULL, "equGBiCost == NULL");
      }
      if( tempCS->slice->getCheckLDC() && !cu->imv && cu->GBiIdx != GBI_DEFAULT && tempCS->cost < m_bestGbiCost[1] )
      {
        if( tempCS->cost < m_bestGbiCost[0] )
        {
          m_bestGbiCost[1] = m_bestGbiCost[0];
          m_bestGbiCost[0] = tempCS->cost;
          m_bestGbiIdx[1] = m_bestGbiIdx[0];
          m_bestGbiIdx[0] = cu->GBiIdx;
        }
        else
        {
          m_bestGbiCost[1] = tempCS->cost;
          m_bestGbiIdx[1] = cu->GBiIdx;
        }
      }
    }

#if !JVET_M0464_UNI_MTS
    double emtFirstPassCost = tempCS->cost;
#endif
    if( imvCS && (tempCS->cost < imvCS->cost) )
    {
      if( imvCS->cost != MAX_DOUBLE )
      {
        imvCS->initStructData( encTestMode.qp, encTestMode.lossless );
      }
      imvCS->copyStructure( *tempCS, partitioner.chType );
    }
    if( NULL != bestHasNonResi && (bestCostInternal > tempCS->cost) )
    {