IntraSearch.cpp

#endif
          }
        }
        else
        {
#if JVET_P0058_CHROMA_TS
          tu.mtsIdx[COMPONENT_Y] = transformIndex;
#else
          tu.mtsIdx = transformIndex;
#endif
        }

        if( !cu.mtsFlag && checkTransformSkip )
        {
          xIntraCodingTUBlock( tu, COMPONENT_Y, false, singleDistTmpLuma, default0Save1Load2, &numSig, modeId == 0 ? &trModes : nullptr, true );
          if( modeId == 0 )
          {
            for( int i = 0; i < 2; i++ )
            {
              if( trModes[ i ].second )
              {
                lastCheckId = trModes[ i ].first;
              }
            }
          }
        }
        else
        {
          xIntraCodingTUBlock( tu, COMPONENT_Y, false, singleDistTmpLuma, default0Save1Load2, &numSig );
        }
      }
      else
      {
        if( nNumTransformCands > 1 )
        {
          xIntraCodingTUBlock( tu, COMPONENT_Y, false, singleDistTmpLuma, default0Save1Load2, &numSig, modeId == 0 ? &trModes : nullptr, true );
          if( modeId == 0 )
          {
            for( int i = 0; i < nNumTransformCands; i++ )
            {
              if( trModes[ i ].second )
              {
                lastCheckId = trModes[ i ].first;
              }
            }
          }
        }
        else
        {
          xIntraCodingTUBlock( tu, COMPONENT_Y, false, singleDistTmpLuma, default0Save1Load2, &numSig );
        }
      }

      //----- determine rate and r-d cost -----
      if( ( sps.getUseLFNST() ? ( modeId == lastCheckId && modeId != 0 && checkTransformSkip ) : ( trModes[ modeId ].first != 0 ) ) && !TU::getCbfAtDepth( tu, COMPONENT_Y, currDepth ) )
      {
        //In order not to code TS flag when cbf is zero, the case for TS with cbf being zero is forbidden.
        singleCostTmp = MAX_DOUBLE;
      }
      else
      {
        if( cu.ispMode && m_pcRdCost->calcRdCost( csFull->fracBits, csFull->dist + singleDistTmpLuma ) > bestCostSoFar )
        {
          earlySkipISP = true;
        }
        else
        {
          singleTmpFracBits = xGetIntraFracBitsQT( *csFull, partitioner, true, false, subTuCounter, ispType );
        }
        singleCostTmp     = m_pcRdCost->calcRdCost( singleTmpFracBits, singleDistTmpLuma );
      }

      if ( !cu.ispMode && nNumTransformCands > 1 && modeId == firstCheckId )
      {
        bestDCT2cost = singleCostTmp;
      }

      if (singleCostTmp < dSingleCost)
      {
        dSingleCost       = singleCostTmp;
        uiSingleDistLuma  = singleDistTmpLuma;
        singleFracBits    = singleTmpFracBits;

        if( sps.getUseLFNST() )
        {
          bestModeId[ COMPONENT_Y ] = modeId;
          cbfBestMode = TU::getCbfAtDepth( tu, COMPONENT_Y, currDepth );
          cbfBestModeValid = true;
          validReturnFull = true;
        }
        else
        {
          bestModeId[ COMPONENT_Y ] = trModes[ modeId ].first;
          if( trModes[ modeId ].first == 0 )
          {
            cbfDCT2 = TU::getCbfAtDepth( tu, COMPONENT_Y, currDepth );
          }
        }

        if( bestModeId[COMPONENT_Y] != lastCheckId )
        {
          saveCS.getPredBuf( tu.Y() ).copyFrom( csFull->getPredBuf( tu.Y() ) );
          saveCS.getRecoBuf( tu.Y() ).copyFrom( csFull->getRecoBuf( tu.Y() ) );

          if( keepResi )
          {
            saveCS.getResiBuf   ( tu.Y() ).copyFrom( csFull->getResiBuf   ( tu.Y() ) );
            saveCS.getOrgResiBuf( tu.Y() ).copyFrom( csFull->getOrgResiBuf( tu.Y() ) );
          }

          tmpTU->copyComponentFrom( tu, COMPONENT_Y );

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

    if( sps.getUseLFNST() && !validReturnFull )
    {
      csFull->cost = MAX_DOUBLE;

      if( bCheckSplit )
      {
        ctxBest = m_CABACEstimator->getCtx();
      }
    }
    else
    {
      if( bestModeId[COMPONENT_Y] != lastCheckId )
      {
        csFull->getPredBuf( tu.Y() ).copyFrom( saveCS.getPredBuf( tu.Y() ) );
        csFull->getRecoBuf( tu.Y() ).copyFrom( saveCS.getRecoBuf( tu.Y() ) );

        if( keepResi )
        {
          csFull->getResiBuf   ( tu.Y() ).copyFrom( saveCS.getResiBuf   ( tu.Y() ) );
          csFull->getOrgResiBuf( tu.Y() ).copyFrom( saveCS.getOrgResiBuf( tu.Y() ) );
        }

        tu.copyComponentFrom( *tmpTU, COMPONENT_Y );

        if( !bCheckSplit )
        {
          m_CABACEstimator->getCtx() = ctxBest;
        }
      }
      else if( bCheckSplit )
      {
        ctxBest = m_CABACEstimator->getCtx();
      }

      csFull->cost     += dSingleCost;
      csFull->dist     += uiSingleDistLuma;
      csFull->fracBits += singleFracBits;
    }
  }

  bool validReturnSplit = false;
  if( bCheckSplit )
  {
    //----- store full entropy coding status, load original entropy coding status -----
    if( bCheckFull )
    {
      m_CABACEstimator->getCtx() = ctxStart;
    }
    //----- code splitted block -----
    csSplit->cost = 0;

    bool uiSplitCbfLuma  = false;
    bool splitIsSelected = true;
    if( partitioner.canSplit( TU_MAX_TR_SPLIT, cs ) )
    {
      partitioner.splitCurrArea( TU_MAX_TR_SPLIT, cs );
    }

    if( cu.ispMode )
    {
      partitioner.splitCurrArea( ispType, *csSplit );
    }
    do
    {
      bool tmpValidReturnSplit = xRecurIntraCodingLumaQT( *csSplit, partitioner, bestCostSoFar, subTuCounter, ispType, false, mtsCheckRangeFlag, mtsFirstCheckId, mtsLastCheckId );
      subTuCounter += subTuCounter != -1 ? 1 : 0;
      if( sps.getUseLFNST() && !tmpValidReturnSplit )
      {
        splitIsSelected = false;
        break;
      }

      if( !cu.ispMode )
      {
        csSplit->setDecomp( partitioner.currArea().Y() );
      }
      else if( CU::isISPFirst( cu, partitioner.currArea().Y(), COMPONENT_Y ) )
      {
        csSplit->setDecomp( cu.Y() );
      }

      uiSplitCbfLuma |= TU::getCbfAtDepth( *csSplit->getTU( partitioner.currArea().lumaPos(), partitioner.chType, subTuCounter - 1 ), COMPONENT_Y, partitioner.currTrDepth );
      if( cu.ispMode )
      {
        //exit condition if the accumulated cost is already larger than the best cost so far (no impact in RD performance)
        if( csSplit->cost > bestCostSoFar )
        {
          earlySkipISP    = true;
          splitIsSelected = false;
          break;
        }
        else
        {
          //more restrictive exit condition
          bool tuIsDividedInRows = CU::divideTuInRows( cu );
          int nSubPartitions = tuIsDividedInRows ? cu.lheight() >> floorLog2(cu.firstTU->lheight()) : cu.lwidth() >> floorLog2(cu.firstTU->lwidth());
          double threshold = nSubPartitions == 2 ? 0.95 : subTuCounter == 1 ? 0.83 : 0.91;
          if( subTuCounter < nSubPartitions && csSplit->cost > bestCostSoFar*threshold )
          {
            earlySkipISP    = true;
            splitIsSelected = false;
            break;
          }
        }
      }



    } while( partitioner.nextPart( *csSplit ) );

    partitioner.exitCurrSplit();

    if( splitIsSelected )
    {
      for( auto &ptu : csSplit->tus )
      {
        if( currArea.Y().contains( ptu->Y() ) )
        {
          TU::setCbfAtDepth( *ptu, COMPONENT_Y, currDepth, uiSplitCbfLuma ? 1 : 0 );
        }
      }

      //----- restore context states -----
      m_CABACEstimator->getCtx() = ctxStart;

      //----- determine rate and r-d cost -----
      csSplit->fracBits = xGetIntraFracBitsQT( *csSplit, partitioner, true, false, cu.ispMode ? 0 : -1, ispType );

      //--- update cost ---
      csSplit->cost     = m_pcRdCost->calcRdCost(csSplit->fracBits, csSplit->dist);

      validReturnSplit = true;
    }
  }

  bool retVal = false;
  if( csFull || csSplit )
  {
    if( !sps.getUseLFNST() || validReturnFull || validReturnSplit )
    {
      {
        // otherwise this would've happened in useSubStructure
        cs.picture->getRecoBuf( currArea.Y() ).copyFrom( cs.getRecoBuf( currArea.Y() ) );
        cs.picture->getPredBuf( currArea.Y() ).copyFrom( cs.getPredBuf( currArea.Y() ) );
      }

      if( cu.ispMode && earlySkipISP )
      {
        cs.cost = MAX_DOUBLE;
      }
      else
      {
        cs.cost = m_pcRdCost->calcRdCost( cs.fracBits, cs.dist );
        retVal = true;
      }
    }
  }
  return retVal;
}

#if JVET_P0517_ADAPTIVE_COLOR_TRANSFORM
bool IntraSearch::xRecurIntraCodingACTQT(CodingStructure &cs, Partitioner &partitioner, bool mtsCheckRangeFlag, int mtsFirstCheckId, int mtsLastCheckId, bool moreProbMTSIdxFirst)
{
  const UnitArea &currArea = partitioner.currArea();
  uint32_t       currDepth = partitioner.currTrDepth;
  const Slice    &slice = *cs.slice;
  const SPS      &sps = *cs.sps;

  bool bCheckFull = !partitioner.canSplit(TU_MAX_TR_SPLIT, cs);
  bool bCheckSplit = !bCheckFull;

  TempCtx ctxStart(m_CtxCache, m_CABACEstimator->getCtx());
  TempCtx ctxBest(m_CtxCache);

  CodingStructure *csSplit = nullptr;
  CodingStructure *csFull = nullptr;
  if (bCheckSplit)
  {
    csSplit = &cs;
  }
  else if (bCheckFull)
  {
    csFull = &cs;
  }

  bool validReturnFull = false;

  if (bCheckFull)
  {
    TransformUnit        &tu = csFull->addTU(CS::getArea(*csFull, currArea, partitioner.chType), partitioner.chType);
    tu.depth = currDepth;
    const CodingUnit     &cu = *csFull->getCU(tu.Y().pos(), CHANNEL_TYPE_LUMA);
    const PredictionUnit &pu = *csFull->getPU(tu.Y().pos(), CHANNEL_TYPE_LUMA);
    CHECK(!tu.Y().valid() || !tu.Cb().valid() || !tu.Cr().valid(), "Invalid TU");
    CHECK(tu.cu != &cu, "wrong CU fetch");
    CHECK(cu.ispMode, "adaptive color transform cannot be applied to ISP");
    CHECK(pu.intraDir[CHANNEL_TYPE_CHROMA] != DM_CHROMA_IDX, "chroma should use DM mode for adaptive color transform");

    // 1. intra prediction and forward color transform

    PelUnitBuf orgBuf = csFull->getOrgBuf(tu);
    PelUnitBuf predBuf = csFull->getPredBuf(tu);
    PelUnitBuf resiBuf = csFull->getResiBuf(tu);
    PelUnitBuf orgResiBuf = csFull->getOrgResiBuf(tu);

    for (int i = 0; i < getNumberValidComponents(tu.chromaFormat); i++)
    {
      ComponentID          compID = (ComponentID)i;
      const CompArea       &area = tu.blocks[compID];
      const ChannelType    chType = toChannelType(compID);

      PelBuf         piOrg = orgBuf.bufs[compID];
      PelBuf         piPred = predBuf.bufs[compID];
      PelBuf         piResi = resiBuf.bufs[compID];

      initIntraPatternChType(*tu.cu, area);
      if (PU::isMIP(pu, chType))
      {
#if JVET_P0803_COMBINED_MIP_CLEANUP
        initIntraMip(pu, area);
#endif
        predIntraMip(compID, piPred, pu);
      }
      else
      {
        predIntraAng(compID, piPred, pu);
      }

      piResi.copyFrom(piOrg);
      if (slice.getLmcsEnabledFlag() && m_pcReshape->getCTUFlag() && compID == COMPONENT_Y)
      {
        CompArea tmpArea(COMPONENT_Y, area.chromaFormat, Position(0, 0), area.size());
        PelBuf   tmpPred = m_tmpStorageLCU.getBuf(tmpArea);
        tmpPred.copyFrom(piPred);
        piResi.rspSignal(m_pcReshape->getFwdLUT());
        piResi.subtract(tmpPred);
      }
      else
        piResi.subtract(piPred);
    }

    resiBuf.colorSpaceConvert(orgResiBuf, true);

    // 2. luma residual optimization 
    double     dSingleCostLuma = MAX_DOUBLE;
    bool       checkTransformSkip = sps.getTransformSkipEnabledFlag();
    int        bestLumaModeId = 0;
    uint8_t    nNumTransformCands = cu.mtsFlag ? 4 : 1;
    uint8_t    numTransformIndexCands = nNumTransformCands;

    const bool tsAllowed = TU::isTSAllowed(tu, COMPONENT_Y);
#if JVET_P1026_MTS_SIGNALLING
    const bool mtsAllowed = CU::isMTSAllowed(cu, COMPONENT_Y);
#else
    const bool mtsAllowed = TU::isMTSAllowed(tu, COMPONENT_Y);
#endif
    std::vector<TrMode> trModes;

    if (sps.getUseLFNST())
    {
      checkTransformSkip &= tsAllowed;
      checkTransformSkip &= !cu.mtsFlag;
      checkTransformSkip &= !cu.lfnstIdx;

      if (!cu.mtsFlag && checkTransformSkip)
      {
        trModes.push_back(TrMode(0, true)); //DCT2
        trModes.push_back(TrMode(1, true)); //TS
      }
    }
    else
    {
      nNumTransformCands = 1 + (tsAllowed ? 1 : 0) + (mtsAllowed ? 4 : 0); // DCT + TS + 4 MTS = 6 tests

      trModes.push_back(TrMode(0, true)); //DCT2
      if (tsAllowed)
      {
        trModes.push_back(TrMode(1, true));
      }
      if (mtsAllowed)
      {
        for (int i = 2; i < 6; i++)
        {
          trModes.push_back(TrMode(i, true));
        }
      }
    }

    CodingStructure &saveLumaCS = *m_pSaveCS[0];
    TransformUnit   *tmpTU = nullptr;
    Distortion      singleDistTmpLuma = 0;
    uint64_t        singleTmpFracBits = 0;
    double          singleCostTmp = 0;
    int             firstCheckId = (sps.getUseLFNST() && mtsCheckRangeFlag && cu.mtsFlag) ? mtsFirstCheckId : 0;
    int             lastCheckId = sps.getUseLFNST() ? ((mtsCheckRangeFlag && cu.mtsFlag) ? (mtsLastCheckId + (int)checkTransformSkip) : (numTransformIndexCands - (firstCheckId + 1) + (int)checkTransformSkip)) : trModes[nNumTransformCands - 1].first;
    bool            isNotOnlyOneMode = sps.getUseLFNST() ? lastCheckId != firstCheckId : nNumTransformCands != 1;

    if (isNotOnlyOneMode)
    {
      saveLumaCS.pcv = csFull->pcv;
      saveLumaCS.picture = csFull->picture;
      saveLumaCS.area.repositionTo(csFull->area);
      saveLumaCS.clearTUs();
      tmpTU = &saveLumaCS.addTU(currArea, partitioner.chType);
    }

    bool    cbfBestMode = false;
    bool    cbfBestModeValid = false;
    bool    cbfDCT2 = true;

    m_pcRdCost->lambdaAdjustColorTrans(true, COMPONENT_Y);

    for (int modeId = firstCheckId; modeId <= lastCheckId; modeId++)
    {
      uint8_t transformIndex = modeId;
      csFull->getResiBuf(tu.Y()).copyFrom(csFull->getOrgResiBuf(tu.Y()));

      m_CABACEstimator->getCtx() = ctxStart;
      m_CABACEstimator->resetBits();

      if (sps.getUseLFNST())
      {
        if ((transformIndex < lastCheckId) || ((transformIndex == lastCheckId) && !checkTransformSkip)) //we avoid this if the mode is transformSkip
        {
          // Skip checking other transform candidates if zero CBF is encountered and it is the best transform so far
          if (m_pcEncCfg->getUseFastLFNST() && transformIndex && !cbfBestMode && cbfBestModeValid)
          {
            continue;
          }
        }
      }
      else
      {
        if (!cbfDCT2 || (m_pcEncCfg->getUseTransformSkipFast() && bestLumaModeId == 1))
        {
          break;
        }
        if (!trModes[modeId].second)
        {
          continue;
        }
#if JVET_P0058_CHROMA_TS
        tu.mtsIdx[COMPONENT_Y] = trModes[modeId].first;
#else
        tu.mtsIdx = trModes[modeId].first;
#endif
      }

      singleDistTmpLuma = 0;
      if (sps.getUseLFNST())
      {
        if (cu.mtsFlag)
        {
          if (moreProbMTSIdxFirst)
          {
            uint32_t uiIntraMode = pu.intraDir[CHANNEL_TYPE_LUMA];

            if (transformIndex == 1)
            {
#if JVET_P0058_CHROMA_TS
              tu.mtsIdx[COMPONENT_Y] = (uiIntraMode < 34) ? MTS_DST7_DCT8 : MTS_DCT8_DST7;
#else
              tu.mtsIdx = (uiIntraMode < 34) ? MTS_DST7_DCT8 : MTS_DCT8_DST7;
#endif
            }
            else if (transformIndex == 2)
            {
#if JVET_P0058_CHROMA_TS
              tu.mtsIdx[COMPONENT_Y] = (uiIntraMode < 34) ? MTS_DCT8_DST7 : MTS_DST7_DCT8;
#else
              tu.mtsIdx = (uiIntraMode < 34) ? MTS_DCT8_DST7 : MTS_DST7_DCT8;
#endif
            }
            else
            {
#if JVET_P0058_CHROMA_TS
              tu.mtsIdx[COMPONENT_Y] = MTS_DST7_DST7 + transformIndex;
#else
              tu.mtsIdx = MTS_DST7_DST7 + transformIndex;
#endif
            }
          }
          else
          {
#if JVET_P0058_CHROMA_TS
            tu.mtsIdx[COMPONENT_Y] = MTS_DST7_DST7 + transformIndex;
#else
            tu.mtsIdx = MTS_DST7_DST7 + transformIndex;
#endif
          }
        }
        else
        {
#if JVET_P0058_CHROMA_TS
          tu.mtsIdx[COMPONENT_Y] = transformIndex;
#else
          tu.mtsIdx = transformIndex;
#endif
        }

        if (!cu.mtsFlag && checkTransformSkip)
        {
          xIntraCodingACTTUBlock(tu, COMPONENT_Y, singleDistTmpLuma, modeId == 0 ? &trModes : nullptr, true);
          if (modeId == 0)
          {
            for (int i = 0; i < 2; i++)
            {
              if (trModes[i].second)
              {
                lastCheckId = trModes[i].first;
              }
            }
          }
        }
        else
        {
          xIntraCodingACTTUBlock(tu, COMPONENT_Y, singleDistTmpLuma);
        }
      }
      else
      {
        if (nNumTransformCands > 1)
        {
          xIntraCodingACTTUBlock(tu, COMPONENT_Y, singleDistTmpLuma, modeId == 0 ? &trModes : nullptr, true);
          if (modeId == 0)
          {
            for (int i = 0; i < nNumTransformCands; i++)
            {
              if (trModes[i].second)
              {
                lastCheckId = trModes[i].first;
              }
            }
          }
        }
        else
        {
          xIntraCodingACTTUBlock(tu, COMPONENT_Y, singleDistTmpLuma);
        }
      }

      //----- determine rate and r-d cost -----
      if ((sps.getUseLFNST() ? (modeId == lastCheckId && modeId != 0 && checkTransformSkip) : (trModes[modeId].first != 0)) && !TU::getCbfAtDepth(tu, COMPONENT_Y, currDepth))
      {
        //In order not to code TS flag when cbf is zero, the case for TS with cbf being zero is forbidden.
        singleCostTmp = MAX_DOUBLE;
      }
      else
      {
        singleTmpFracBits = xGetIntraFracBitsQT(*csFull, partitioner, true, false, -1, TU_NO_ISP);
        singleCostTmp = m_pcRdCost->calcRdCost(singleTmpFracBits, singleDistTmpLuma);
      }

      if (singleCostTmp < dSingleCostLuma)
      {
        dSingleCostLuma = singleCostTmp;
        validReturnFull = true;

        if (sps.getUseLFNST())
        {
          bestLumaModeId = modeId;
          cbfBestMode = TU::getCbfAtDepth(tu, COMPONENT_Y, currDepth);
          cbfBestModeValid = true;
        }
        else
        {
          bestLumaModeId = trModes[modeId].first;
          if (trModes[modeId].first == 0)
          {
            cbfDCT2 = TU::getCbfAtDepth(tu, COMPONENT_Y, currDepth);
          }
        }

        if (bestLumaModeId != lastCheckId)
        {
          saveLumaCS.getResiBuf(tu.Y()).copyFrom(csFull->getResiBuf(tu.Y()));
          tmpTU->copyComponentFrom(tu, COMPONENT_Y);
          ctxBest = m_CABACEstimator->getCtx();
        }
      }
    }

    m_pcRdCost->lambdaAdjustColorTrans(false, COMPONENT_Y);

    if (sps.getUseLFNST())
    {
      if (!validReturnFull)
      {
        csFull->cost = MAX_DOUBLE;
        return false;
      }
    }
    else
    {
      CHECK(!validReturnFull, "no transform mode was tested for luma");
    }

    csFull->setDecomp(currArea.Y(), true);
    csFull->setDecomp(currArea.Cb(), true);

    if (bestLumaModeId != lastCheckId)
    {
      csFull->getResiBuf(tu.Y()).copyFrom(saveLumaCS.getResiBuf(tu.Y()));
      tu.copyComponentFrom(*tmpTU, COMPONENT_Y);
      m_CABACEstimator->getCtx() = ctxBest;
    }

    // 3 chroma residual optimization
    CodingStructure &saveChromaCS = *m_pSaveCS[1];
    saveChromaCS.pcv = csFull->pcv;
    saveChromaCS.picture = csFull->picture;
    saveChromaCS.area.repositionTo(csFull->area);
    saveChromaCS.initStructData(MAX_INT, false, true);
    tmpTU = &saveChromaCS.addTU(currArea, partitioner.chType);

    CompArea&  cbArea = tu.blocks[COMPONENT_Cb];
    CompArea&  crArea = tu.blocks[COMPONENT_Cr];

    ctxStart = m_CABACEstimator->getCtx();
    m_CABACEstimator->resetBits();
    tu.jointCbCr = 0;

    bool doReshaping = (slice.getLmcsEnabledFlag() && slice.getLmcsChromaResidualScaleFlag() && (slice.isIntra() || m_pcReshape->getCTUFlag()) && (cbArea.width * cbArea.height > 4));
    if (doReshaping)
    {
      const Area      area = tu.Y().valid() ? tu.Y() : Area(recalcPosition(tu.chromaFormat, tu.chType, CHANNEL_TYPE_LUMA, tu.blocks[tu.chType].pos()), recalcSize(tu.chromaFormat, tu.chType, CHANNEL_TYPE_LUMA, tu.blocks[tu.chType].size()));
      const CompArea &areaY = CompArea(COMPONENT_Y, tu.chromaFormat, area);
      int             adj = m_pcReshape->calculateChromaAdjVpduNei(tu, areaY);
      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()
        & slice.getLmcsEnabledFlag() && (m_pcReshape->getCTUFlag() || (isChroma(compID) && m_pcEncCfg->getReshapeIntraCMD()))))
      {
        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()
            & slice.getLmcsEnabledFlag() && (m_pcReshape->getCTUFlag() || (isChroma(compID) && m_pcEncCfg->getReshapeIntraCMD()))))
          {
            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
    int  predMode   = PU::getFinalIntraMode( pu, CHANNEL_TYPE_CHROMA );

    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 ----
    bool doReshaping = ( cs.slice->getLmcsEnabledFlag() && cs.slice->getLmcsChromaResidualScaleFlag()