IntraSearch.cpp

            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( KEEP_PRED_AND_RESI_SIGNALS || JVET_AC0162_ALF_RESIDUAL_SAMPLES_INPUT )
          {
            saveCS.getResiBuf   ( tu.Y() ).copyFrom( csFull->getResiBuf   ( tu.Y() ) );
          }
          if( KEEP_PRED_AND_RESI_SIGNALS )
          {
            saveCS.getOrgResiBuf( tu.Y() ).copyFrom( csFull->getOrgResiBuf( tu.Y() ) );
          }

          tmpTU->copyComponentFrom( tu, COMPONENT_Y );

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

#if JVET_AH0103_LOW_DELAY_LFNST_NSPT
    if( spsIntraLfnstEnabled && !validReturnFull )
#else
    if( sps.getUseLFNST() && !validReturnFull )
#endif
    {
      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( KEEP_PRED_AND_RESI_SIGNALS || JVET_AC0162_ALF_RESIDUAL_SAMPLES_INPUT )
        {
          csFull->getResiBuf   ( tu.Y() ).copyFrom( saveCS.getResiBuf   ( tu.Y() ) );
        }
        if( KEEP_PRED_AND_RESI_SIGNALS )
        {
          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 );
#if SIGN_PREDICTION
#if JVET_Z0118_GDR
      cs.updateReconMotIPM(partitioner.currArea());
#else
      cs.picture->getRecoBuf(  partitioner.currArea()  ).copyFrom( cs.getRecoBuf( partitioner.currArea() ) );
#endif
#endif
      subTuCounter += subTuCounter != -1 ? 1 : 0;
#if JVET_AH0103_LOW_DELAY_LFNST_NSPT
      if( spsIntraLfnstEnabled && !tmpValidReturnSplit )
#else
      if( sps.getUseLFNST() && !tmpValidReturnSplit )
#endif
      {
        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;

      cuCtx.violatesLfnstConstrained[CHANNEL_TYPE_LUMA] = false;
      cuCtx.violatesLfnstConstrained[CHANNEL_TYPE_CHROMA] = false;
      cuCtx.lfnstLastScanPos = false;
      cuCtx.violatesMtsCoeffConstraint = false;
      cuCtx.mtsLastScanPos = false;
#if JVET_Y0142_ADAPT_INTRA_MTS
      cuCtx.mtsCoeffAbsSum = 0;
#endif

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

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

      validReturnSplit = true;
    }
  }

  bool retVal = false;
  if( csFull || csSplit )
  {
#if JVET_AH0103_LOW_DELAY_LFNST_NSPT
    if( !spsIntraLfnstEnabled || validReturnFull || validReturnSplit )
#else
    if( !sps.getUseLFNST() || validReturnFull || validReturnSplit )
#endif
    {
      // otherwise this would've happened in useSubStructure
#if JVET_Z0118_GDR
      cs.updateReconMotIPM(currArea.Y());
#else
      cs.picture->getRecoBuf(currArea.Y()).copyFrom(cs.getRecoBuf(currArea.Y()));
#endif
      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;
}

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 JVET_AH0103_LOW_DELAY_LFNST_NSPT
  bool spsIntraLfnstEnabled = ( ( slice.getSliceType() == I_SLICE && sps.getUseIntraLFNSTISlice() ) ||
                                ( slice.getSliceType() != I_SLICE && sps.getUseIntraLFNSTPBSlice() ) );
#endif

  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);
#if JVET_S0234_ACT_CRS_FIX
    bool doReshaping = (slice.getLmcsEnabledFlag() && slice.getPicHeader()->getLmcsChromaResidualScaleFlag() && (slice.isIntra() || m_pcReshape->getCTUFlag()) && (tu.blocks[COMPONENT_Cb].width * tu.blocks[COMPONENT_Cb].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);
    }
#endif

    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 JVET_V0130_INTRA_TMP && !JVET_W0069_TMP_BOUNDARY
      if( PU::isTmp( pu, chType ) )
      {
        int foundCandiNum;
        getTargetTemplate( pu.cu, pu.lwidth(), pu.lheight() );
        candidateSearchIntra( pu.cu, pu.lwidth(), pu.lheight() );
        generateTMPrediction( piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), foundCandiNum );
        CHECK( foundCandiNum < 1, "" );

      }
      else if( PU::isMIP( pu, chType ) )
#else
      if (PU::isMIP(pu, chType))
#endif
      {
        initIntraMip(pu, area);
#if JVET_AB0067_MIP_DIMD_LFNST
#if JVET_AH0076_OBIC
        predIntraMip(compID, piPred, pu, true);
#else
        predIntraMip(compID, piPred, pu, pu.cu->lfnstIdx > 0 ? true : false);
#endif
#else
        predIntraMip(compID, piPred, pu);
#endif
      }
      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);
        piResi.rspSignal( piPred, m_pcReshape->getFwdLUT() );
        piResi.subtract(tmpPred);
      }
#if JVET_S0234_ACT_CRS_FIX
      else if (doReshaping && (compID != COMPONENT_Y))
      {
        piResi.subtract(piPred);
        int cResScaleInv = tu.getChromaAdj();
        piResi.scaleSignal(cResScaleInv, 1, slice.clpRng(compID));
      }
#endif
      else
      {
        piResi.subtract(piPred);
      }
    }

    resiBuf.colorSpaceConvert(orgResiBuf, true, cs.slice->clpRng(COMPONENT_Y));

    // 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);
    const bool mtsAllowed = CU::isMTSAllowed(cu, COMPONENT_Y);
    std::vector<TrMode> trModes;

#if JVET_AH0103_LOW_DELAY_LFNST_NSPT
    if( spsIntraLfnstEnabled )
#else
    if (sps.getUseLFNST())
#endif
    {
      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
    {
      if (m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && slice.isLossless())
      {
        nNumTransformCands = 1;
        CHECK(!tsAllowed && !cu.bdpcmMode, "transform skip should be enabled for LS");
        if (cu.bdpcmMode)
        {
          trModes.push_back(TrMode(0, true));
        }
        else
        {
          trModes.push_back(TrMode(1, true));
        }
      }
      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;
#if JVET_AH0103_LOW_DELAY_LFNST_NSPT
    int             firstCheckId = ( spsIntraLfnstEnabled && mtsCheckRangeFlag && cu.mtsFlag ) ? mtsFirstCheckId : 0;
    int             lastCheckId = spsIntraLfnstEnabled ? ( ( mtsCheckRangeFlag && cu.mtsFlag ) ? ( mtsLastCheckId + ( int ) checkTransformSkip ) : ( numTransformIndexCands - ( firstCheckId + 1 ) + ( int ) checkTransformSkip ) ) : trModes[ nNumTransformCands - 1 ].first;
    bool            isNotOnlyOneMode = spsIntraLfnstEnabled ? lastCheckId != firstCheckId : nNumTransformCands != 1;
#else
    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;
#endif

    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;
    if (m_pcEncCfg->getCostMode() != COST_LOSSLESS_CODING || !slice.isLossless())
    m_pcRdCost->lambdaAdjustColorTrans(true, COMPONENT_Y);
    for (int modeId = firstCheckId; modeId <= ((m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && slice.isLossless()) ? (nNumTransformCands - 1) : lastCheckId); modeId++)
    {
      uint8_t transformIndex = modeId;
      csFull->getResiBuf(tu.Y()).copyFrom(csFull->getOrgResiBuf(tu.Y()));

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

#if JVET_AH0103_LOW_DELAY_LFNST_NSPT
      if( spsIntraLfnstEnabled )
#else
      if (sps.getUseLFNST())
#endif
      {
        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 (!(m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && slice.isLossless()))
        {
          if (!cbfDCT2 || (m_pcEncCfg->getUseTransformSkipFast() && bestLumaModeId == 1))
          {
            break;
          }
          if (!trModes[modeId].second)
          {
            continue;
          }
        }
        tu.mtsIdx[COMPONENT_Y] = trModes[modeId].first;
      }

      singleDistTmpLuma = 0;
#if JVET_AH0103_LOW_DELAY_LFNST_NSPT
      if( spsIntraLfnstEnabled )
#else
      if (sps.getUseLFNST())
#endif
      {
        if (cu.mtsFlag)
        {
          if (moreProbMTSIdxFirst)
          {
            uint32_t uiIntraMode = pu.intraDir[CHANNEL_TYPE_LUMA];

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

        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);
        }
      }

      CUCtx cuCtx;
      cuCtx.isDQPCoded = true;
      cuCtx.isChromaQpAdjCoded = true;
      //----- determine rate and r-d cost -----
#if JVET_AH0103_LOW_DELAY_LFNST_NSPT
      if( ( spsIntraLfnstEnabled ? ( modeId == lastCheckId && modeId != 0 && checkTransformSkip ) : ( trModes[ modeId ].first != 0 ) ) && !TU::getCbfAtDepth( tu, COMPONENT_Y, currDepth ) )
#else
      if ((sps.getUseLFNST() ? (modeId == lastCheckId && modeId != 0 && checkTransformSkip) : (trModes[modeId].first != 0)) && !TU::getCbfAtDepth(tu, COMPONENT_Y, currDepth))
#endif
      {
        //In order not to code TS flag when cbf is zero, the case for TS with cbf being zero is forbidden.
        if (m_pcEncCfg->getCostMode() != COST_LOSSLESS_CODING || !slice.isLossless())
        singleCostTmp = MAX_DOUBLE;
        else
        {
          singleTmpFracBits = xGetIntraFracBitsQT(*csFull, partitioner, true, false, -1, TU_NO_ISP);
          singleCostTmp = m_pcRdCost->calcRdCost(singleTmpFracBits, singleDistTmpLuma, false);
        }
      }
      else
      {
        singleTmpFracBits = xGetIntraFracBitsQT(*csFull, partitioner, true, false, -1, TU_NO_ISP, &cuCtx);

        if (tu.mtsIdx[COMPONENT_Y] > MTS_SKIP)
        {
#if JVET_Y0142_ADAPT_INTRA_MTS
          int nCands = (cuCtx.mtsCoeffAbsSum > MTS_TH_COEFF[1]) ? MTS_NCANDS[2] : (cuCtx.mtsCoeffAbsSum > MTS_TH_COEFF[0]) ? MTS_NCANDS[1] : MTS_NCANDS[0];
          bool isInvalid = !cuCtx.mtsLastScanPos || ((tu.mtsIdx[COMPONENT_Y] - MTS_DST7_DST7) >= nCands);
          if (isInvalid)
#else
          if (!cuCtx.mtsLastScanPos)
#endif
          {
            singleCostTmp = MAX_DOUBLE;
          }
          else
          {
            singleCostTmp = m_pcRdCost->calcRdCost(singleTmpFracBits, singleDistTmpLuma, false);
          }
        }
        else
        {
          singleCostTmp = m_pcRdCost->calcRdCost(singleTmpFracBits, singleDistTmpLuma, false);
        }
      }

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

#if JVET_AH0103_LOW_DELAY_LFNST_NSPT
        if( spsIntraLfnstEnabled )
#else
        if (sps.getUseLFNST())
#endif
        {
          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();
        }
      }
    }
    if (m_pcEncCfg->getCostMode() != COST_LOSSLESS_CODING || !slice.isLossless())
    m_pcRdCost->lambdaAdjustColorTrans(false, COMPONENT_Y);

#if JVET_AH0103_LOW_DELAY_LFNST_NSPT
    if( spsIntraLfnstEnabled )
#else
    if (sps.getUseLFNST())
#endif
    {
      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, true);
    tmpTU = &saveChromaCS.addTU(currArea, partitioner.chType);

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

    tu.jointCbCr = 0;

#if !JVET_S0234_ACT_CRS_FIX
    bool doReshaping = (slice.getLmcsEnabledFlag() && slice.getPicHeader()->getLmcsChromaResidualScaleFlag() && (slice.isIntra() || m_pcReshape->getCTUFlag()) && (cbArea.width * cbArea.height > 4));
    if (doReshaping)
    {
#if LMCS_CHROMA_CALC_CU
      const Area      area = tu.cu->Y().valid() ? tu.cu->Y() : Area(recalcPosition(tu.chromaFormat, tu.chType, CHANNEL_TYPE_LUMA, tu.cu->blocks[tu.chType].pos()), recalcSize(tu.chromaFormat, tu.chType, CHANNEL_TYPE_LUMA, tu.cu->blocks[tu.chType].size()));
#else
      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()));
#endif
      const CompArea &areaY = CompArea(COMPONENT_Y, tu.chromaFormat, area);
      int             adj = m_pcReshape->calculateChromaAdjVpduNei(tu, areaY);
      tu.setChromaAdj(adj);
    }
#endif

    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 !JVET_S0234_ACT_CRS_FIX
    if (doReshaping)
    {
      int cResScaleInv = tu.getChromaAdj();
      orgResiCb[0].scaleSignal(cResScaleInv, 1, slice.clpRng(COMPONENT_Cb));
      orgResiCr[0].scaleSignal(cResScaleInv, 1, slice.clpRng(COMPONENT_Cr));
    }
#endif

    // 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);
      double  dSingleBestCostChroma = MAX_DOUBLE;
      int     bestModeId = -1;
      bool    tsAllowed = TU::isTSAllowed(tu, compID) && (m_pcEncCfg->getUseChromaTS()) && !cu.lfnstIdx;
      uint8_t numTransformCands = 1 + (tsAllowed ? 1 : 0);  // DCT + TS = 2 tests
      bool        cbfDCT2 = true;

      trModes.clear();
      if (m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && slice.isLossless())
      {
        numTransformCands = 1;
        CHECK(!tsAllowed && !cu.bdpcmModeChroma, "transform skip should be enabled for LS");
        if (cu.bdpcmModeChroma)
        {
          trModes.push_back(TrMode(0, true));
        }
        else
        {
          trModes.push_back(TrMode(1, true));
        }
      }
      else
      {
        trModes.push_back(TrMode(0, true));                    // DCT
        if (tsAllowed)
        {
          trModes.push_back(TrMode(1, true));                  // TS
        }
      }
      if (m_pcEncCfg->getCostMode() != COST_LOSSLESS_CODING || !slice.isLossless())
#if JVET_S0234_ACT_CRS_FIX
      {
        if (doReshaping)
        {
          int cResScaleInv = tu.getChromaAdj();
          m_pcRdCost->lambdaAdjustColorTrans(true, compID, true, &cResScaleInv);
        }
        else
        {
          m_pcRdCost->lambdaAdjustColorTrans(true, compID);
        }
      }
#else
      {
        m_pcRdCost->lambdaAdjustColorTrans(true, compID);
      }
#endif

      TempCtx ctxBegin(m_ctxCache);
      ctxBegin = m_CABACEstimator->getCtx();

      for (int modeId = 0; modeId < numTransformCands; modeId++)
      {
        if (m_pcEncCfg->getCostMode() != COST_LOSSLESS_CODING || !slice.isLossless())
        {
          if (modeId && !cbfDCT2)
          {
            continue;
          }
          if (!trModes[modeId].second)
          {
            continue;
          }
        }

        if (modeId > 0)
        {
          m_CABACEstimator->getCtx() = ctxBegin;
        }

        tu.mtsIdx[compID] = trModes[modeId].first;
        Distortion singleDistChroma = 0;
        if (numTransformCands > 1)
        {
          xIntraCodingACTTUBlock(tu, compID, singleDistChroma, modeId == 0 ? &trModes : nullptr, true);
        }
        else
        {
          xIntraCodingACTTUBlock(tu, compID, singleDistChroma);
        }
        if (!tu.mtsIdx[compID])
        {
          cbfDCT2 = TU::getCbfAtDepth(tu, compID, currDepth);
        }
        uint64_t fracBitChroma     = xGetIntraFracBitsQTChroma(tu, compID);
        double   dSingleCostChroma = m_pcRdCost->calcRdCost(fracBitChroma, singleDistChroma, false);
        if (dSingleCostChroma < dSingleBestCostChroma)
        {
          dSingleBestCostChroma = dSingleCostChroma;
          bestModeId            = modeId;
          if (bestModeId != (numTransformCands - 1))
          {
            saveChromaCS.getResiBuf(tu.blocks[compID]).copyFrom(csFull->getResiBuf(tu.blocks[compID]));
            tmpTU->copyComponentFrom(tu, compID);
            ctxBest = m_CABACEstimator->getCtx();
          }
        }
      }

      if (bestModeId != (numTransformCands - 1))
      {
        csFull->getResiBuf(tu.blocks[compID]).copyFrom(saveChromaCS.getResiBuf(tu.blocks[compID]));
        tu.copyComponentFrom(*tmpTU, compID);
        m_CABACEstimator->getCtx() = ctxBest;
      }
      if (m_pcEncCfg->getCostMode() != COST_LOSSLESS_CODING || !slice.isLossless())
      {
        m_pcRdCost->lambdaAdjustColorTrans(false, 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, cs.slice->clpRng(COMPONENT_Y));

    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];

#if JVET_S0234_ACT_CRS_FIX
      if (doReshaping && (compID != COMPONENT_Y))
      {
        piResi.scaleSignal(tu.getChromaAdj(), 0, slice.clpRng(compID));
      }
#endif
      piReco.reconstruct(piPred, piResi, cs.slice->clpRng(compID));

      if (m_pcEncCfg->getLumaLevelToDeltaQPMapping().isEnabled() || (m_pcEncCfg->getLmcs()
        && 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.rspSignal( piReco, 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)
    {
      tu.jointCbCr = (uint8_t)cbfMask;

      ComponentID codeCompId = ((cbfMask >> 1) ? COMPONENT_Cb : COMPONENT_Cr);
      ComponentID otherCompId = ((codeCompId == COMPONENT_Cb) ? COMPONENT_Cr : COMPONENT_Cb);
      bool        tsAllowed = TU::isTSAllowed(tu, codeCompId) && (m_pcEncCfg->getUseChromaTS()) && !cu.lfnstIdx;
      uint8_t     numTransformCands = 1 + (tsAllowed ? 1 : 0); // DCT + TS = 2 tests
      bool        cbfDCT2 = true;

      trModes.clear();
      trModes.push_back(TrMode(0, true)); // DCT2
      if (tsAllowed)
      {
        trModes.push_back(TrMode(1, true));//TS
      }

      for (int modeId = 0; modeId < numTransformCands; modeId++)
      {
        if (modeId && !cbfDCT2)
        {
          continue;
        }
        if (!trModes[modeId].second)
        {
          continue;
        }
        Distortion distTmp = 0;
        tu.mtsIdx[codeCompId] = trModes[modeId].first;
        tu.mtsIdx[otherCompId] = MTS_DCT2_DCT2;
        m_CABACEstimator->getCtx() = ctxStart;
        csFull->getResiBuf(cbArea).copyFrom(orgResiCb[cbfMask]);
        csFull->getResiBuf(crArea).copyFrom(orgResiCr[cbfMask]);
        if (nNumTransformCands > 1)
        {
          xIntraCodingACTTUBlock(tu, COMPONENT_Cb, distTmp, modeId == 0 ? &trModes : nullptr, true);
        }
        else
        {
          xIntraCodingACTTUBlock(tu, COMPONENT_Cb, distTmp);
        }

        double   costTmp = std::numeric_limits<double>::max();
        uint64_t bitsTmp = 0;
        if (distTmp < std::numeric_limits<Distortion>::max())
        {
          if (!tu.mtsIdx[codeCompId])
          {
            cbfDCT2 = true;
          }
          csFull->getResiBuf(tu).colorSpaceConvert(invColorTransResidual, false, csFull->slice->clpRng(COMPONENT_Y));
          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];

#if JVET_S0234_ACT_CRS_FIX
            if (doReshaping && (compID != COMPONENT_Y))
            {
              piResi.scaleSignal(tu.getChromaAdj(), 0, slice.clpRng(compID));
            }
#endif
            piReco.reconstruct(piPred, piResi, cs.slice->clpRng(compID));
            if (m_pcEncCfg->getLumaLevelToDeltaQPMapping().isEnabled()
                || (m_pcEncCfg->getLmcs() && 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.rspSignal(piReco, 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);
        }
        else if (!tu.mtsIdx[codeCompId])
        {
          cbfDCT2 = false;
        }

        if (costTmp < bestCostJointCbCr)
        {
          bestCostJointCbCr = costTmp;
          bestDistJointCbCr = distTmp;
          bestBitsJointCbCr = bitsTmp;
          bestJointCbCr     = tu.jointCbCr;
          lastIsBest        = (cbfMask == jointCbfMasksToTest.back() && modeId == (numTransformCands - 1));

          // 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);