IntraSearch.cpp

      tu.setChromaAdj(adj);
    }

    CompStorage  orgResiCb[5], orgResiCr[5]; // 0:std, 1-3:jointCbCr (placeholder at this stage), 4:crossComp
    orgResiCb[0].create(cbArea);
    orgResiCr[0].create(crArea);
    orgResiCb[0].copyFrom(csFull->getOrgResiBuf(cbArea));
    orgResiCr[0].copyFrom(csFull->getOrgResiBuf(crArea));
    if (doReshaping)
    {
      int cResScaleInv = tu.getChromaAdj();
      orgResiCb[0].scaleSignal(cResScaleInv, 1, slice.clpRng(COMPONENT_Cb));
      orgResiCr[0].scaleSignal(cResScaleInv, 1, slice.clpRng(COMPONENT_Cr));
    }

    // 3.1 regular chroma residual coding
    csFull->getResiBuf(cbArea).copyFrom(orgResiCb[0]);
    csFull->getResiBuf(crArea).copyFrom(orgResiCr[0]);

    for (uint32_t c = COMPONENT_Cb; c < ::getNumberValidTBlocks(*csFull->pcv); c++)
    {
      const ComponentID compID = ComponentID(c);
      Distortion singleDistChroma = 0;
      xIntraCodingACTTUBlock(tu, compID, singleDistChroma);
      xGetIntraFracBitsQTChroma(tu, compID);
    }

    Position tuPos = tu.Y();
    tuPos.relativeTo(cu.Y());
    const UnitArea relativeUnitArea(tu.chromaFormat, Area(tuPos, tu.Y().size()));
    PelUnitBuf     invColorTransResidual = m_colorTransResiBuf.getBuf(relativeUnitArea);
    csFull->getResiBuf(tu).colorSpaceConvert(invColorTransResidual, false);

    Distortion totalDist = 0;
    for (uint32_t c = COMPONENT_Y; c < ::getNumberValidTBlocks(*csFull->pcv); c++)
    {
      const ComponentID compID = ComponentID(c);
      const CompArea&   area = tu.blocks[compID];
      PelBuf            piOrg = csFull->getOrgBuf(area);
      PelBuf            piReco = csFull->getRecoBuf(area);
      PelBuf            piPred = csFull->getPredBuf(area);
      PelBuf            piResi = invColorTransResidual.bufs[compID];

      piReco.reconstruct(piPred, piResi, cs.slice->clpRng(compID));

      if (m_pcEncCfg->getLumaLevelToDeltaQPMapping().isEnabled() || (m_pcEncCfg->getReshaper()
#if JVET_P1006_PICTURE_HEADER
        & slice.getPicHeader()->getLmcsEnabledFlag() && (m_pcReshape->getCTUFlag() || (isChroma(compID) && m_pcEncCfg->getReshapeIntraCMD()))))
#else
        & slice.getLmcsEnabledFlag() && (m_pcReshape->getCTUFlag() || (isChroma(compID) && m_pcEncCfg->getReshapeIntraCMD()))))
#endif
      {
        const CPelBuf orgLuma = csFull->getOrgBuf(csFull->area.blocks[COMPONENT_Y]);
        if (compID == COMPONENT_Y && !(m_pcEncCfg->getLumaLevelToDeltaQPMapping().isEnabled()))
        {
          CompArea      tmpArea1(COMPONENT_Y, area.chromaFormat, Position(0, 0), area.size());
          PelBuf tmpRecLuma = m_tmpStorageLCU.getBuf(tmpArea1);
          tmpRecLuma.copyFrom(piReco);
          tmpRecLuma.rspSignal(m_pcReshape->getInvLUT());
          totalDist += m_pcRdCost->getDistPart(piOrg, tmpRecLuma, sps.getBitDepth(toChannelType(compID)), compID, DF_SSE_WTD, &orgLuma);
        }
        else
        {
          totalDist += m_pcRdCost->getDistPart(piOrg, piReco, sps.getBitDepth(toChannelType(compID)), compID, DF_SSE_WTD, &orgLuma);
        }
      }
      else
      {
        totalDist += m_pcRdCost->getDistPart(piOrg, piReco, sps.getBitDepth(toChannelType(compID)), compID, DF_SSE);
      }
    }

    m_CABACEstimator->getCtx() = ctxStart;
    uint64_t totalBits = xGetIntraFracBitsQT(*csFull, partitioner, true, true, -1, TU_NO_ISP);
    double   totalCost = m_pcRdCost->calcRdCost(totalBits, totalDist);

    saveChromaCS.getResiBuf(cbArea).copyFrom(csFull->getResiBuf(cbArea));
    saveChromaCS.getResiBuf(crArea).copyFrom(csFull->getResiBuf(crArea));
    saveChromaCS.getRecoBuf(tu).copyFrom(csFull->getRecoBuf(tu));
    tmpTU->copyComponentFrom(tu, COMPONENT_Cb);
    tmpTU->copyComponentFrom(tu, COMPONENT_Cr);
    ctxBest = m_CABACEstimator->getCtx();

    // 3.2 jointCbCr
    double     bestCostJointCbCr = totalCost;
    Distortion bestDistJointCbCr = totalDist;
    uint64_t   bestBitsJointCbCr = totalBits;
    int        bestJointCbCr = tu.jointCbCr; assert(!bestJointCbCr);

    bool       lastIsBest = false;
    std::vector<int>  jointCbfMasksToTest;
    if (sps.getJointCbCrEnabledFlag() && (TU::getCbf(tu, COMPONENT_Cb) || TU::getCbf(tu, COMPONENT_Cr)))
    {
      jointCbfMasksToTest = m_pcTrQuant->selectICTCandidates(tu, orgResiCb, orgResiCr);
    }

    for (int cbfMask : jointCbfMasksToTest)
    {
      m_CABACEstimator->getCtx() = ctxStart;
      m_CABACEstimator->resetBits();

      Distortion distTmp = 0;
      tu.jointCbCr = (uint8_t)cbfMask;

      csFull->getResiBuf(cbArea).copyFrom(orgResiCb[cbfMask]);
      csFull->getResiBuf(crArea).copyFrom(orgResiCr[cbfMask]);
      xIntraCodingACTTUBlock(tu, COMPONENT_Cb, distTmp);

      double   costTmp = std::numeric_limits<double>::max();
      uint64_t bitsTmp = 0;
      if (distTmp < std::numeric_limits<Distortion>::max())
      {
        csFull->getResiBuf(tu).colorSpaceConvert(invColorTransResidual, false);
        distTmp = 0;
        for (uint32_t c = COMPONENT_Y; c < ::getNumberValidTBlocks(*csFull->pcv); c++)
        {
          const ComponentID compID = ComponentID(c);
          const CompArea&   area = tu.blocks[compID];
          PelBuf            piOrg = csFull->getOrgBuf(area);
          PelBuf            piReco = csFull->getRecoBuf(area);
          PelBuf            piPred = csFull->getPredBuf(area);
          PelBuf            piResi = invColorTransResidual.bufs[compID];

          piReco.reconstruct(piPred, piResi, cs.slice->clpRng(compID));
          if (m_pcEncCfg->getLumaLevelToDeltaQPMapping().isEnabled() || (m_pcEncCfg->getReshaper()
#if JVET_P1006_PICTURE_HEADER
            & slice.getPicHeader()->getLmcsEnabledFlag() && (m_pcReshape->getCTUFlag() || (isChroma(compID) && m_pcEncCfg->getReshapeIntraCMD()))))
#else
            & slice.getLmcsEnabledFlag() && (m_pcReshape->getCTUFlag() || (isChroma(compID) && m_pcEncCfg->getReshapeIntraCMD()))))
#endif
          {
            const CPelBuf orgLuma = csFull->getOrgBuf(csFull->area.blocks[COMPONENT_Y]);
            if (compID == COMPONENT_Y && !(m_pcEncCfg->getLumaLevelToDeltaQPMapping().isEnabled()))
            {
              CompArea      tmpArea1(COMPONENT_Y, area.chromaFormat, Position(0, 0), area.size());
              PelBuf tmpRecLuma = m_tmpStorageLCU.getBuf(tmpArea1);
              tmpRecLuma.copyFrom(piReco);
              tmpRecLuma.rspSignal(m_pcReshape->getInvLUT());
              distTmp += m_pcRdCost->getDistPart(piOrg, tmpRecLuma, sps.getBitDepth(toChannelType(compID)), compID, DF_SSE_WTD, &orgLuma);
            }
            else
            {
              distTmp += m_pcRdCost->getDistPart(piOrg, piReco, sps.getBitDepth(toChannelType(compID)), compID, DF_SSE_WTD, &orgLuma);
            }
          }
          else
          {
            distTmp += m_pcRdCost->getDistPart(piOrg, piReco, sps.getBitDepth(toChannelType(compID)), compID, DF_SSE);
          }
        }

        bitsTmp = xGetIntraFracBitsQT(*csFull, partitioner, true, true, -1, TU_NO_ISP);
        costTmp = m_pcRdCost->calcRdCost(bitsTmp, distTmp);
      }

      if (costTmp < bestCostJointCbCr)
      {
        bestCostJointCbCr = costTmp;
        bestDistJointCbCr = distTmp;
        bestBitsJointCbCr = bitsTmp;
        bestJointCbCr = tu.jointCbCr;
        lastIsBest = (cbfMask == jointCbfMasksToTest.back());

        // store data
        if (!lastIsBest)
        {
          saveChromaCS.getResiBuf(cbArea).copyFrom(csFull->getResiBuf(cbArea));
          saveChromaCS.getResiBuf(crArea).copyFrom(csFull->getResiBuf(crArea));
          saveChromaCS.getRecoBuf(tu).copyFrom(csFull->getRecoBuf(tu));
          tmpTU->copyComponentFrom(tu, COMPONENT_Cb);
          tmpTU->copyComponentFrom(tu, COMPONENT_Cr);

          ctxBest = m_CABACEstimator->getCtx();
        }
      }
    }

    if (!lastIsBest)
    {
      csFull->getResiBuf(cbArea).copyFrom(saveChromaCS.getResiBuf(cbArea));
      csFull->getResiBuf(crArea).copyFrom(saveChromaCS.getResiBuf(crArea));
      csFull->getRecoBuf(tu).copyFrom(saveChromaCS.getRecoBuf(tu));
      tu.copyComponentFrom(*tmpTU, COMPONENT_Cb);
      tu.copyComponentFrom(*tmpTU, COMPONENT_Cr);

      m_CABACEstimator->getCtx() = ctxBest;
    }
    tu.jointCbCr = bestJointCbCr;
    csFull->picture->getRecoBuf(tu).copyFrom(csFull->getRecoBuf(tu));

    csFull->dist += bestDistJointCbCr;
    csFull->fracBits += bestBitsJointCbCr;
    csFull->cost = m_pcRdCost->calcRdCost(csFull->fracBits, csFull->dist);
  }

  bool validReturnSplit = false;
  if (bCheckSplit)
  {
    if (partitioner.canSplit(TU_MAX_TR_SPLIT, *csSplit))
    {
      partitioner.splitCurrArea(TU_MAX_TR_SPLIT, *csSplit);
    }

    bool splitIsSelected = true;
    do
    {
      bool tmpValidReturnSplit = xRecurIntraCodingACTQT(*csSplit, partitioner, mtsCheckRangeFlag, mtsFirstCheckId, mtsLastCheckId, moreProbMTSIdxFirst);
      if (sps.getUseLFNST())
      {
        if (!tmpValidReturnSplit)
        {
          splitIsSelected = false;
          break;
        }
      }
      else
      {
        CHECK(!tmpValidReturnSplit, "invalid RD of sub-TU partitions for ACT");
      }
    } while (partitioner.nextPart(*csSplit));

    partitioner.exitCurrSplit();

    if (splitIsSelected)
    {
      unsigned compCbf[3] = { 0, 0, 0 };
      for (auto &currTU : csSplit->traverseTUs(currArea, partitioner.chType))
      {
        for (unsigned ch = 0; ch < getNumberValidTBlocks(*csSplit->pcv); ch++)
        {
          compCbf[ch] |= (TU::getCbfAtDepth(currTU, ComponentID(ch), currDepth + 1) ? 1 : 0);
        }
      }

      for (auto &currTU : csSplit->traverseTUs(currArea, partitioner.chType))
      {
        TU::setCbfAtDepth(currTU, COMPONENT_Y, currDepth, compCbf[COMPONENT_Y]);
        TU::setCbfAtDepth(currTU, COMPONENT_Cb, currDepth, compCbf[COMPONENT_Cb]);
        TU::setCbfAtDepth(currTU, COMPONENT_Cr, currDepth, compCbf[COMPONENT_Cr]);
      }

      m_CABACEstimator->getCtx() = ctxStart;
      csSplit->fracBits = xGetIntraFracBitsQT(*csSplit, partitioner, true, true, -1, TU_NO_ISP);
      csSplit->cost = m_pcRdCost->calcRdCost(csSplit->fracBits, csSplit->dist);

      validReturnSplit = true;
    }
  }

  bool retVal = false;
  if (csFull || csSplit)
  {
    if (sps.getUseLFNST())
    {
      if (validReturnFull || validReturnSplit)
      {
        retVal = true;
      }
    }
    else
    {
      CHECK(!validReturnFull && !validReturnSplit, "illegal TU optimization");
      retVal = true;
    }
  }
  return retVal;
}
#endif

ChromaCbfs IntraSearch::xRecurIntraChromaCodingQT( CodingStructure &cs, Partitioner& partitioner, const double bestCostSoFar, const PartSplit ispType )
{
  UnitArea currArea                   = partitioner.currArea();
  const bool keepResi                 = cs.sps->getUseLMChroma() || KEEP_PRED_AND_RESI_SIGNALS;
  if( !currArea.Cb().valid() ) return ChromaCbfs( false );


  TransformUnit &currTU               = *cs.getTU( currArea.chromaPos(), CHANNEL_TYPE_CHROMA );
  const PredictionUnit &pu            = *cs.getPU( currArea.chromaPos(), CHANNEL_TYPE_CHROMA );

  bool lumaUsesISP                    = false;
  uint32_t     currDepth                  = partitioner.currTrDepth;
  const PPS &pps                      = *cs.pps;
  ChromaCbfs cbfs                     ( false );

  if (currDepth == currTU.depth)
  {
    if (!currArea.Cb().valid() || !currArea.Cr().valid())
    {
      return cbfs;
    }


    CodingStructure &saveCS = *m_pSaveCS[1];
    saveCS.pcv      = cs.pcv;
    saveCS.picture  = cs.picture;
    saveCS.area.repositionTo( cs.area );
    saveCS.initStructData( MAX_INT, false, true );

    if( !currTU.cu->isSepTree() && currTU.cu->ispMode )
    {
      saveCS.clearCUs();
      CodingUnit& auxCU = saveCS.addCU( *currTU.cu, partitioner.chType );
      auxCU.ispMode = currTU.cu->ispMode;
      saveCS.sps = currTU.cs->sps;
      saveCS.clearPUs();
      saveCS.addPU( *currTU.cu->firstPU, partitioner.chType );
    }

    TransformUnit &tmpTU = saveCS.addTU(currArea, partitioner.chType);


    cs.setDecomp(currArea.Cb(), true); // set in advance (required for Cb2/Cr2 in 4:2:2 video)

    const unsigned      numTBlocks  = ::getNumberValidTBlocks( *cs.pcv );

    CompArea&  cbArea         = currTU.blocks[COMPONENT_Cb];
    CompArea&  crArea         = currTU.blocks[COMPONENT_Cr];
    double     bestCostCb     = MAX_DOUBLE;
    double     bestCostCr     = MAX_DOUBLE;
    Distortion bestDistCb     = 0;
    Distortion bestDistCr     = 0;
    int        maxModesTested = 0;
    bool       earlyExitISP   = false;

    TempCtx ctxStartTU( m_CtxCache );
    TempCtx ctxStart  ( m_CtxCache );
    TempCtx ctxBest   ( m_CtxCache );

    ctxStartTU       = m_CABACEstimator->getCtx();
    currTU.jointCbCr = 0;

    // Do predictions here to avoid repeating the "default0Save1Load2" stuff
#if JVET_P0059_CHROMA_BDPCM
    int  predMode   = pu.cu->bdpcmModeChroma ? BDPCM_IDX : PU::getFinalIntraMode(pu, CHANNEL_TYPE_CHROMA);
#else
    int  predMode   = PU::getFinalIntraMode( pu, CHANNEL_TYPE_CHROMA );
#endif

    PelBuf piPredCb = cs.getPredBuf(cbArea);
    PelBuf piPredCr = cs.getPredBuf(crArea);

    initIntraPatternChType( *currTU.cu, cbArea);
    initIntraPatternChType( *currTU.cu, crArea);

    if( PU::isLMCMode( predMode ) )
    {
      xGetLumaRecPixels( pu, cbArea );
      predIntraChromaLM( COMPONENT_Cb, piPredCb, pu, cbArea, predMode );
      predIntraChromaLM( COMPONENT_Cr, piPredCr, pu, crArea, predMode );
    }
    else
    {
      predIntraAng( COMPONENT_Cb, piPredCb, pu);
      predIntraAng( COMPONENT_Cr, piPredCr, pu);
    }

    // determination of chroma residuals including reshaping and cross-component prediction
    //----- get chroma residuals -----
    PelBuf resiCb  = cs.getResiBuf(cbArea);
    PelBuf resiCr  = cs.getResiBuf(crArea);
    resiCb.copyFrom( cs.getOrgBuf (cbArea) );
    resiCr.copyFrom( cs.getOrgBuf (crArea) );
    resiCb.subtract( piPredCb );
    resiCr.subtract( piPredCr );

    //----- get reshape parameter ----
#if JVET_P1006_PICTURE_HEADER
    bool doReshaping = ( cs.picHeader->getLmcsEnabledFlag() && cs.picHeader->getLmcsChromaResidualScaleFlag()
#else
    bool doReshaping = ( cs.slice->getLmcsEnabledFlag() && cs.slice->getLmcsChromaResidualScaleFlag()
#endif
                         && (cs.slice->isIntra() || m_pcReshape->getCTUFlag()) && (cbArea.width * cbArea.height > 4) );
    if( doReshaping )
    {
      const Area area = currTU.Y().valid() ? currTU.Y() : Area(recalcPosition(currTU.chromaFormat, currTU.chType, CHANNEL_TYPE_LUMA, currTU.blocks[currTU.chType].pos()), recalcSize(currTU.chromaFormat, currTU.chType, CHANNEL_TYPE_LUMA, currTU.blocks[currTU.chType].size()));
      const CompArea &areaY = CompArea(COMPONENT_Y, currTU.chromaFormat, area);
      int adj = m_pcReshape->calculateChromaAdjVpduNei(currTU, areaY);
      currTU.setChromaAdj(adj);
    }

    //----- get cross component prediction parameters -----
    bool checkCrossComponentPrediction = PU::isChromaIntraModeCrossCheckMode( pu ) && pps.getPpsRangeExtension().getCrossComponentPredictionEnabledFlag() && TU::getCbf( currTU, COMPONENT_Y );
    int  compAlpha[MAX_NUM_COMPONENT] = { 0, 0, 0 };
    if( checkCrossComponentPrediction )
    {
      compAlpha[COMPONENT_Cb] = xCalcCrossComponentPredictionAlpha( currTU, COMPONENT_Cb, m_pcEncCfg->getUseReconBasedCrossCPredictionEstimate() );
      compAlpha[COMPONENT_Cr] = xCalcCrossComponentPredictionAlpha( currTU, COMPONENT_Cr, m_pcEncCfg->getUseReconBasedCrossCPredictionEstimate() );
      if( compAlpha[COMPONENT_Cb] == 0 && compAlpha[COMPONENT_Cr] == 0 )
      {
        checkCrossComponentPrediction = false;
      }
    }

    //===== store original residual signals (std and crossCompPred) =====
    CompStorage  orgResiCb[5], orgResiCr[5]; // 0:std, 1-3:jointCbCr (placeholder at this stage), 4:crossComp
    for( int k = 0; k < (checkCrossComponentPrediction?5:1); k+=4 )
    {
      orgResiCb[k].create( cbArea );
      orgResiCr[k].create( crArea );
      if( k >= 4 ) {
        CrossComponentPrediction::crossComponentPrediction( currTU, COMPONENT_Cb, cs.getResiBuf(currTU.Y()), resiCb, orgResiCb[k], false);
        CrossComponentPrediction::crossComponentPrediction( currTU, COMPONENT_Cr, cs.getResiBuf(currTU.Y()), resiCr, orgResiCr[k], false);
      } else {
        orgResiCb[k].copyFrom( resiCb );
        orgResiCr[k].copyFrom( resiCr );
      }
      if( doReshaping )
      {
        int cResScaleInv = currTU.getChromaAdj();
        orgResiCb[k].scaleSignal( cResScaleInv, 1, currTU.cu->cs->slice->clpRng(COMPONENT_Cb) );
        orgResiCr[k].scaleSignal( cResScaleInv, 1, currTU.cu->cs->slice->clpRng(COMPONENT_Cr) );
      }
    }

    for( uint32_t c = COMPONENT_Cb; c < numTBlocks; c++)
    {
      const ComponentID compID  = ComponentID(c);
      const CompArea&   area    = currTU.blocks[compID];

      double     dSingleCost    = MAX_DOUBLE;
      int        bestModeId     = 0;
      Distortion singleDistCTmp = 0;
      double     singleCostTmp  = 0;
      const int  crossCPredictionModesToTest = checkCrossComponentPrediction ? 2 : 1;
#if JVET_P0058_CHROMA_TS
      const bool tsAllowed = TU::isTSAllowed(currTU, compID) && (m_pcEncCfg->getUseChromaTS());
      uint8_t nNumTransformCands = 1 + (tsAllowed ? 1 : 0); // DCT + TS = 2 tests
      std::vector<TrMode> trModes;
      trModes.push_back(TrMode(0, true)); // DCT2

      if (tsAllowed)
      {
          trModes.push_back(TrMode(1, true));//TS
      }
      CHECK(!currTU.Cb().valid(), "Invalid TU");
#endif

#if JVET_P0058_CHROMA_TS
      const int  totalModesToTest            = crossCPredictionModesToTest * nNumTransformCands;
      bool cbfDCT2 = true;
#else
      const int  totalModesToTest            = crossCPredictionModesToTest;
#endif
      const bool isOneMode                   = false;
      maxModesTested                         = totalModesToTest > maxModesTested ? totalModesToTest : maxModesTested;

      int currModeId = 0;
      int default0Save1Load2 = 0;


      if (!isOneMode)
      {
        ctxStart = m_CABACEstimator->getCtx();
      }

#if JVET_P0058_CHROMA_TS
      for (int modeId = 0; modeId < nNumTransformCands; modeId++)
#endif
      {
        for (int crossCPredictionModeId = 0; crossCPredictionModeId < crossCPredictionModesToTest; crossCPredictionModeId++)
        {
          resiCb.copyFrom( orgResiCb[4*crossCPredictionModeId] );
          resiCr.copyFrom( orgResiCr[4*crossCPredictionModeId] );

          currTU.compAlpha    [compID] = ( crossCPredictionModeId ? compAlpha[compID] : 0 );

#if JVET_P0058_CHROMA_TS
#if JVET_P0059_CHROMA_BDPCM
          currTU.mtsIdx[compID] = currTU.cu->bdpcmModeChroma ? MTS_SKIP : trModes[modeId].first;
#else
          currTU.mtsIdx[compID] = trModes[modeId].first;
#endif
#endif

          currModeId++;

          const bool isFirstMode = (currModeId == 1);
          const bool isLastMode  = false; // Always store output to saveCS and tmpTU

#if JVET_AHG14_LOSSLESS
          if( !( m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING ) )
          {
#endif
#if JVET_P0058_CHROMA_TS
           //if DCT2's cbf==0, skip ts search
          if (!cbfDCT2 && trModes[modeId].first == MTS_SKIP)
          {
              break;
          }
          if (!trModes[modeId].second)
          {
              continue;
          }
#endif
#if JVET_AHG14_LOSSLESS
          }
#endif

          if (!isFirstMode) // if not first mode to be tested
          {
            m_CABACEstimator->getCtx() = ctxStart;
          }

          singleDistCTmp = 0;

#if JVET_P0058_CHROMA_TS
          if (nNumTransformCands > 1)
          {
              xIntraCodingTUBlock(currTU, compID, crossCPredictionModeId != 0, singleDistCTmp, default0Save1Load2, nullptr, modeId == 0 ? &trModes : nullptr, true);
          }
          else
          {
              xIntraCodingTUBlock(currTU, compID, crossCPredictionModeId != 0, singleDistCTmp, default0Save1Load2);
          }
#else
          xIntraCodingTUBlock( currTU, compID, crossCPredictionModeId != 0, singleDistCTmp, default0Save1Load2 );
#endif

#if JVET_P0058_CHROMA_TS
#if JVET_P0059_CHROMA_BDPCM
          if (((crossCPredictionModeId == 1) && (currTU.compAlpha[compID] == 0)) || ((currTU.mtsIdx[compID] == MTS_SKIP && !currTU.cu->bdpcmModeChroma) && !TU::getCbf(currTU, compID))) //In order not to code TS flag when cbf is zero, the case for TS with cbf being zero is forbidden.
#else
          if (((crossCPredictionModeId == 1) && (currTU.compAlpha[compID] == 0)) || ((currTU.mtsIdx[compID] == MTS_SKIP) && !TU::getCbf(currTU, compID))) //In order not to code TS flag when cbf is zero, the case for TS with cbf being zero is forbidden.
#endif
#else
          if( ( ( crossCPredictionModeId == 1 ) && ( currTU.compAlpha[compID] == 0 ) ) ) //In order not to code TS flag when cbf is zero, the case for TS with cbf being zero is forbidden.
#endif
          {
            singleCostTmp = MAX_DOUBLE;
          }
          else if( lumaUsesISP && bestCostSoFar != MAX_DOUBLE && c == COMPONENT_Cb )
          {
            uint64_t fracBitsTmp = xGetIntraFracBitsQTSingleChromaComponent( cs, partitioner, ComponentID( c ) );
            singleCostTmp = m_pcRdCost->calcRdCost( fracBitsTmp, singleDistCTmp );
            if( isOneMode || ( !isOneMode && !isLastMode ) )
            {
              m_CABACEstimator->getCtx() = ctxStart;
            }
          }
          else if( !isOneMode )
          {
            uint64_t fracBitsTmp = xGetIntraFracBitsQTChroma( currTU, compID );
            singleCostTmp = m_pcRdCost->calcRdCost( fracBitsTmp, singleDistCTmp );
          }

          if( singleCostTmp < dSingleCost )
          {
            dSingleCost = singleCostTmp;
            bestModeId  = currModeId;

            if ( c == COMPONENT_Cb )
            {
              bestCostCb = singleCostTmp;
              bestDistCb = singleDistCTmp;
            }
            else
            {
              bestCostCr = singleCostTmp;
              bestDistCr = singleDistCTmp;
            }

#if JVET_P0058_CHROMA_TS
            if (currTU.mtsIdx[compID] == MTS_DCT2_DCT2)
            {
                cbfDCT2 = TU::getCbfAtDepth(currTU, compID, currDepth);
            }
#endif

            if( !isLastMode )
            {
#if KEEP_PRED_AND_RESI_SIGNALS
              saveCS.getPredBuf   (area).copyFrom(cs.getPredBuf   (area));
              saveCS.getOrgResiBuf(area).copyFrom(cs.getOrgResiBuf(area));
#endif
              saveCS.getPredBuf   (area).copyFrom(cs.getPredBuf   (area));
              if( keepResi )
              {
                saveCS.getResiBuf (area).copyFrom(cs.getResiBuf   (area));
              }
              saveCS.getRecoBuf   (area).copyFrom(cs.getRecoBuf   (area));

              tmpTU.copyComponentFrom(currTU, compID);

              ctxBest = m_CABACEstimator->getCtx();
            }
          }
        }
      }

      if( lumaUsesISP && dSingleCost > bestCostSoFar && c == COMPONENT_Cb )
      {
        //Luma + Cb cost is already larger than the best cost, so we don't need to test Cr
        cs.dist = MAX_UINT;
        m_CABACEstimator->getCtx() = ctxStart;
        earlyExitISP               = true;
        break;
        //return cbfs;
      }

      // Done with one component of separate coding of Cr and Cb, just switch to the best Cb contexts if Cr coding is still to be done
      if ( c == COMPONENT_Cb && bestModeId < totalModesToTest)
      {
        m_CABACEstimator->getCtx() = ctxBest;

        currTU.copyComponentFrom(tmpTU, COMPONENT_Cb); // Cbf of Cb is needed to estimate cost for Cr Cbf
      }
    }

    if ( !earlyExitISP )
    {
      // Test using joint chroma residual coding
      double     bestCostCbCr   = bestCostCb + bestCostCr;
      Distortion bestDistCbCr   = bestDistCb + bestDistCr;
      int        bestJointCbCr  = 0;
      bool       lastIsBest     = false;
      std::vector<int>  jointCbfMasksToTest;
      if ( cs.sps->getJointCbCrEnabledFlag() && (TU::getCbf(tmpTU, COMPONENT_Cb) || TU::getCbf(tmpTU, COMPONENT_Cr)))
      {
        jointCbfMasksToTest = m_pcTrQuant->selectICTCandidates(currTU, orgResiCb, orgResiCr);
      }
      for( int cbfMask : jointCbfMasksToTest )
      {
        Distortion distTmp = 0;

        currTU.jointCbCr               = (uint8_t)cbfMask;
        currTU.compAlpha[COMPONENT_Cb] = 0;
        currTU.compAlpha[COMPONENT_Cr] = 0;
#if JVET_P0058_CHROMA_TS
        // encoder bugfix: initialize mtsIdx for chroma under JointCbCrMode.
        currTU.mtsIdx[COMPONENT_Cb] = currTU.mtsIdx[COMPONENT_Cr]  = MTS_DCT2_DCT2;
#endif
        m_CABACEstimator->getCtx() = ctxStartTU;

        resiCb.copyFrom( orgResiCb[cbfMask] );
        resiCr.copyFrom( orgResiCr[cbfMask] );
        xIntraCodingTUBlock( currTU, COMPONENT_Cb, false, distTmp, 0 );

        double costTmp = std::numeric_limits<double>::max();
        if( distTmp < std::numeric_limits<Distortion>::max() )
        {
          uint64_t bits  = xGetIntraFracBitsQTChroma( currTU, COMPONENT_Cb );
          costTmp = m_pcRdCost->calcRdCost( bits, distTmp );
        }

        if( costTmp < bestCostCbCr )
        {
          bestCostCbCr  = costTmp;
          bestDistCbCr  = distTmp;
          bestJointCbCr = currTU.jointCbCr;

          // store data
          if( cbfMask != jointCbfMasksToTest.back() )
          {
#if KEEP_PRED_AND_RESI_SIGNALS
            saveCS.getOrgResiBuf(cbArea).copyFrom(cs.getOrgResiBuf(cbArea));
            saveCS.getOrgResiBuf(crArea).copyFrom(cs.getOrgResiBuf(crArea));
#endif
            saveCS.getPredBuf   (cbArea).copyFrom(cs.getPredBuf   (cbArea));
            saveCS.getPredBuf   (crArea).copyFrom(cs.getPredBuf   (crArea));
            if( keepResi )
            {
              saveCS.getResiBuf (cbArea).copyFrom(cs.getResiBuf   (cbArea));
              saveCS.getResiBuf (crArea).copyFrom(cs.getResiBuf   (crArea));
            }
            saveCS.getRecoBuf   (cbArea).copyFrom(cs.getRecoBuf   (cbArea));
            saveCS.getRecoBuf   (crArea).copyFrom(cs.getRecoBuf   (crArea));

            tmpTU.copyComponentFrom(currTU, COMPONENT_Cb);
            tmpTU.copyComponentFrom(currTU, COMPONENT_Cr);

            ctxBest = m_CABACEstimator->getCtx();
          }
          else
          {
            lastIsBest = true;
          }
        }
      }

      // Retrieve the best CU data (unless it was the very last one tested)
      if ( !( maxModesTested == 1 && jointCbfMasksToTest.empty() ) && !lastIsBest )
      {
#if KEEP_PRED_AND_RESI_SIGNALS
        cs.getPredBuf   (cbArea).copyFrom(saveCS.getPredBuf   (cbArea));
        cs.getOrgResiBuf(cbArea).copyFrom(saveCS.getOrgResiBuf(cbArea));
        cs.getPredBuf   (crArea).copyFrom(saveCS.getPredBuf   (crArea));
        cs.getOrgResiBuf(crArea).copyFrom(saveCS.getOrgResiBuf(crArea));
#endif
        cs.getPredBuf   (cbArea).copyFrom(saveCS.getPredBuf   (cbArea));
        cs.getPredBuf   (crArea).copyFrom(saveCS.getPredBuf   (crArea));

        if( keepResi )
        {
          cs.getResiBuf (cbArea).copyFrom(saveCS.getResiBuf   (cbArea));
          cs.getResiBuf (crArea).copyFrom(saveCS.getResiBuf   (crArea));
        }
        cs.getRecoBuf   (cbArea).copyFrom(saveCS.getRecoBuf   (cbArea));
        cs.getRecoBuf   (crArea).copyFrom(saveCS.getRecoBuf   (crArea));

        currTU.copyComponentFrom(tmpTU, COMPONENT_Cb);
        currTU.copyComponentFrom(tmpTU, COMPONENT_Cr);

        m_CABACEstimator->getCtx() = ctxBest;
      }

      // Copy results to the picture structures
      cs.picture->getRecoBuf(cbArea).copyFrom(cs.getRecoBuf(cbArea));
      cs.picture->getRecoBuf(crArea).copyFrom(cs.getRecoBuf(crArea));
      cs.picture->getPredBuf(cbArea).copyFrom(cs.getPredBuf(cbArea));
      cs.picture->getPredBuf(crArea).copyFrom(cs.getPredBuf(crArea));

      cbfs.cbf(COMPONENT_Cb) = TU::getCbf(currTU, COMPONENT_Cb);
      cbfs.cbf(COMPONENT_Cr) = TU::getCbf(currTU, COMPONENT_Cr);

      currTU.jointCbCr = ( (cbfs.cbf(COMPONENT_Cb) + cbfs.cbf(COMPONENT_Cr)) ? bestJointCbCr : 0 );
      cs.dist         += bestDistCbCr;
    }
  }
  else
  {
    unsigned    numValidTBlocks   = ::getNumberValidTBlocks( *cs.pcv );
    ChromaCbfs  SplitCbfs         ( false );

    if( partitioner.canSplit( TU_MAX_TR_SPLIT, cs ) )
    {
      partitioner.splitCurrArea( TU_MAX_TR_SPLIT, cs );
    }
    else if( currTU.cu->ispMode )
    {
      partitioner.splitCurrArea( ispType, cs );
    }
    else
      THROW( "Implicit TU split not available" );

    do
    {
      ChromaCbfs subCbfs = xRecurIntraChromaCodingQT( cs, partitioner, bestCostSoFar, ispType );

      for( uint32_t ch = COMPONENT_Cb; ch < numValidTBlocks; ch++ )
      {
        const ComponentID compID = ComponentID( ch );
        SplitCbfs.cbf( compID ) |= subCbfs.cbf( compID );
      }
    } while( partitioner.nextPart( cs ) );

    partitioner.exitCurrSplit();

    if( lumaUsesISP && cs.dist == MAX_UINT )
    {
      return cbfs;
    }
    {

      cbfs.Cb |= SplitCbfs.Cb;
      cbfs.Cr |= SplitCbfs.Cr;

      if( !lumaUsesISP )
      {
        for( auto &ptu : cs.tus )
        {
          if( currArea.Cb().contains( ptu->Cb() ) || ( !ptu->Cb().valid() && currArea.Y().contains( ptu->Y() ) ) )
          {
            TU::setCbfAtDepth( *ptu, COMPONENT_Cb, currDepth, SplitCbfs.Cb );
            TU::setCbfAtDepth( *ptu, COMPONENT_Cr, currDepth, SplitCbfs.Cr );
          }
        }
      }
    }
  }

  return cbfs;
}

uint64_t IntraSearch::xFracModeBitsIntra(PredictionUnit &pu, const uint32_t &uiMode, const ChannelType &chType)
{
  uint32_t orgMode = uiMode;

  if (!pu.mhIntraFlag)
  std::swap(orgMode, pu.intraDir[chType]);

  m_CABACEstimator->resetBits();

  if( isLuma( chType ) )
  {
    if (!pu.mhIntraFlag)
    {
      m_CABACEstimator->intra_luma_pred_mode(pu);
    }
  }
  else
  {
    m_CABACEstimator->intra_chroma_pred_mode( pu );
  }

  if ( !pu.mhIntraFlag )
  std::swap(orgMode, pu.intraDir[chType]);

  return m_CABACEstimator->getEstFracBits();
}

#if JVET_P0517_ADAPTIVE_COLOR_TRANSFORM 
void IntraSearch::sortRdModeListFirstColorSpace(ModeInfo mode, double cost, char bdpcmMode, ModeInfo* rdModeList, double* rdCostList, char* bdpcmModeList, int& candNum)
{
  if (candNum == 0)
  {
    rdModeList[0] = mode;
    rdCostList[0] = cost;
    bdpcmModeList[0] = bdpcmMode;
    candNum++;
    return;
  }

  int insertPos = -1;
  for (int pos = candNum - 1; pos >= 0; pos--)
  {
    if (cost < rdCostList[pos])
    {
      insertPos = pos;
    }
  }

  if (insertPos >= 0)
  {
    for (int i = candNum - 1; i >= insertPos; i--)
    {
      rdModeList[i + 1] = rdModeList[i];
      rdCostList[i + 1] = rdCostList[i];
      bdpcmModeList[i + 1] = bdpcmModeList[i];
    }
    rdModeList[insertPos] = mode;
    rdCostList[insertPos] = cost;
    bdpcmModeList[insertPos] = bdpcmMode;
    candNum++;
  }
  else
  {
    rdModeList[candNum] = mode;
    rdCostList[candNum] = cost;
    bdpcmModeList[candNum] = bdpcmMode;
    candNum++;
  }

  CHECK(candNum > FAST_UDI_MAX_RDMODE_NUM, "exceed intra mode candidate list capacity");

  return;
}

void IntraSearch::invalidateBestRdModeFirstColorSpace()
{
  int numSaveRdClass = 4 * NUM_LFNST_NUM_PER_SET * 2;
  int savedRdModeListSize = FAST_UDI_MAX_RDMODE_NUM;

  for (int i = 0; i < numSaveRdClass; i++)
  {
    m_numSavedRdModeFirstColorSpace[i] = 0;
    for (int j = 0; j < savedRdModeListSize; j++)
    {
#if JVET_P0803_COMBINED_MIP_CLEANUP
      m_savedRdModeFirstColorSpace[i][j] = ModeInfo(false, false, 0, NOT_INTRA_SUBPARTITIONS, 0);
#else
      m_savedRdModeFirstColorSpace[i][j] = ModeInfo(false, 0, NOT_INTRA_SUBPARTITIONS, 0);
#endif
      m_savedBDPCMModeFirstColorSpace[i][j] = 0;
      m_savedRdCostFirstColorSpace[i][j] = MAX_DOUBLE;
    }
  }
}
#endif

void IntraSearch::encPredIntraDPCM( const ComponentID &compID, PelBuf &pOrg, PelBuf &pDst, const uint32_t &uiDirMode )
{
  CHECK( pOrg.buf == 0, "Encoder DPCM called without original buffer" );

  const int srcStride = m_refBufferStride[compID];
  CPelBuf   pSrc = CPelBuf(getPredictorPtr(compID), srcStride, m_leftRefLength + 1);

  // Sample Adaptive intra-Prediction (SAP)
  if( uiDirMode == HOR_IDX )
  {
    // left column filled with reference samples, remaining columns filled with pOrg data
    for( int y = 0; y < pDst.height; y++ )
    {
      pDst.at(0, y) = pSrc.at(1 + y, 1);
    }
    CPelBuf orgRest  = pOrg.subBuf( 0, 0, pOrg.width - 1, pOrg.height );
    PelBuf  predRest = pDst.subBuf( 1, 0, pDst.width - 1, pDst.height );

    predRest.copyFrom( orgRest );
  }
  else // VER_IDX
  {
    // top row filled with reference samples, remaining rows filled with pOrg data
    for( int x = 0; x < pDst.width; x++ )
    {
      pDst.at( x, 0 ) = pSrc.at( 1 + x, 0 );
    }
    CPelBuf orgRest  = pOrg.subBuf( 0, 0, pOrg.width, pOrg.height - 1 );
    PelBuf  predRest = pDst.subBuf( 0, 1, pDst.width, pDst.height - 1 );

    predRest.copyFrom( orgRest );
  }
}

bool IntraSearch::useDPCMForFirstPassIntraEstimation( const PredictionUnit &pu, const uint32_t &uiDirMode )
{
  return CU::isRDPCMEnabled( *pu.cu ) && pu.cu->transQuantBypass && (uiDirMode == HOR_IDX || uiDirMode == VER_IDX);
}

template<typename T, size_t N>
void IntraSearch::reduceHadCandList(static_vector<T, N>& candModeList, static_vector<double, N>& candCostList, int& numModesForFullRD, const double thresholdHadCost, const double* mipHadCost, const PredictionUnit &pu, const bool fastMip)
{
  const int maxCandPerType = numModesForFullRD >> 1;
  static_vector<ModeInfo, FAST_UDI_MAX_RDMODE_NUM> tempRdModeList;
  static_vector<double, FAST_UDI_MAX_RDMODE_NUM> tempCandCostList;
  const double minCost = candCostList[0];
  bool keepOneMip = candModeList.size() > numModesForFullRD;

  int numConv = 0;
  int numMip = 0;
  for (int idx = 0; idx < candModeList.size() - (keepOneMip?0:1); idx++)
  {
    bool addMode = false;
    const ModeInfo& orgMode = candModeList[idx];

    if (!orgMode.mipFlg)
    {
      addMode = (numConv < 3);
      numConv += addMode ? 1:0;
    }
    else
    {
      addMode = ( numMip < maxCandPerType || (candCostList[idx] < thresholdHadCost * minCost) || keepOneMip );
      keepOneMip = false;
      numMip += addMode ? 1:0;
    }
    if( addMode )
    {
      tempRdModeList.push_back(orgMode);
      tempCandCostList.push_back(candCostList[idx]);
    }
  }

  if ((pu.lwidth() > 8 && pu.lheight() > 8))
  {
    // Sort MIP candidates by Hadamard cost
#if JVET_P0803_COMBINED_MIP_CLEANUP
    const int transpOff = getNumModesMip( pu.Y() );
#else
    const int transpOff = getNumModesMip(pu.Y()) / 2;
#endif
    static_vector<uint8_t, FAST_UDI_MAX_RDMODE_NUM> sortedMipModes(0);
    static_vector<double, FAST_UDI_MAX_RDMODE_NUM> sortedMipCost(0);
#if JVET_P0803_COMBINED_MIP_CLEANUP
    for( uint8_t mode : { 0, 1, 2 } )
#else
    for (uint8_t mode : { 3, 4, 5 })
#endif
    {
      uint8_t candMode = mode + uint8_t((mipHadCost[mode + transpOff] < mipHadCost[mode]) ? transpOff : 0);
      updateCandList(candMode, mipHadCost[candMode], sortedMipModes, sortedMipCost, 3);
    }

    // Append MIP mode to RD mode list
    const int modeListSize = int(tempRdModeList.size());
    for (int idx = 0; idx < 3; idx++)
    {
#if JVET_P0803_COMBINED_MIP_CLEANUP
      const bool     isTransposed = (sortedMipModes[idx] >= transpOff ? true : false);
      const uint32_t mipIdx       = (isTransposed ? sortedMipModes[idx] - transpOff : sortedMipModes[idx]);
      const ModeInfo mipMode( true, isTransposed, 0, NOT_INTRA_SUBPARTITIONS, mipIdx );
#else
      const ModeInfo mipMode(true, 0, NOT_INTRA_SUBPARTITIONS, sortedMipModes[idx]);
#endif
      bool alreadyIncluded = false;
      for (int modeListIdx = 0; modeListIdx < modeListSize; modeListIdx++)
      {
        if (tempRdModeList[modeListIdx] == mipMode)
        {
          alreadyIncluded = true;
          break;
        }
      }

      if (!alreadyIncluded)
      {
        tempRdModeList.push_back(mipMode);
        tempCandCostList.push_back(0);
        if( fastMip ) break;
      }
    }
  }

  candModeList = tempRdModeList;
  candCostList = tempCandCostList;
  numModesForFullRD = int(candModeList.size());
}

// It decides which modes from the ISP lists can be full RD tested
void IntraSearch::xGetNextISPMode(ModeInfo& modeInfo, const ModeInfo* lastMode, const Size cuSize)
{