IntraSearch.cpp

          tu.mtsIdx[(codeCompId == COMPONENT_Cr) ? COMPONENT_Cb : COMPONENT_Cr] = MTS_DCT2_DCT2;
        }
#else
        m_pcTrQuant->transformNxN(tu, compID, qpCbCr, trModes, m_pcEncCfg->getMTSIntraMaxCand());
        tu.mtsIdx[compID] = trModes->at(0).first;
#endif
    }
    // encoder bugfix: Set loadTr to aovid redundant transform process
#if JVET_AHG14_LOSSLESS
    if (!(m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && tu.mtsIdx[compID] == 0) || tu.cu->bdpcmModeChroma != 0)
    {
        m_pcTrQuant->transformNxN(tu, codeCompId, qpCbCr, uiAbsSum, m_CABACEstimator->getCtx(), loadTr);
    }
#else
    m_pcTrQuant->transformNxN(tu, codeCompId, qpCbCr, uiAbsSum, m_CABACEstimator->getCtx(), loadTr);
#endif

#if JVET_AHG14_LOSSLESS
    if ((m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && tu.mtsIdx[compID] == 0) && 0 == tu.cu->bdpcmModeChroma)
    {
        uiAbsSum = 0;
        tu.getCoeffs(compID).fill(0);
        TU::setCbfAtDepth(tu, compID, tu.depth, 0);
    }
#endif

    DTRACE( g_trace_ctx, D_TU_ABS_SUM, "%d: comp=%d, abssum=%d\n", DTRACE_GET_COUNTER( g_trace_ctx, D_TU_ABS_SUM ), codeCompId, uiAbsSum );
    if( uiAbsSum > 0 )
    {
      m_pcTrQuant->invTransformNxN(tu, codeCompId, codeResi, qpCbCr);
      codedCbfMask += ( codeCompId == COMPONENT_Cb ? 2 : 1 );
    }
    else
    {
      codeResi.fill(0);
    }

    if( tu.jointCbCr )
    {
      if( tu.jointCbCr == 3 && codedCbfMask == 2 )
      {
        codedCbfMask = 3;
        TU::setCbfAtDepth (tu, COMPONENT_Cr, tu.depth, true );
      }
      if( tu.jointCbCr != codedCbfMask )
      {
        ruiDist = std::numeric_limits<Distortion>::max();
        return;
      }
      m_pcTrQuant->invTransformICT( tu, piResi, crResi );
      uiAbsSum = codedCbfMask;
    }
  }

  //===== reconstruction =====
  if ( flag && uiAbsSum > 0 && isChroma(compID) && slice.getPicHeader()->getLmcsChromaResidualScaleFlag() )
  {
    piResi.scaleSignal(tu.getChromaAdj(), 0, tu.cu->cs->slice->clpRng(compID));
    if( jointCbCr )
    {
      crResi.scaleSignal(tu.getChromaAdj(), 0, tu.cu->cs->slice->clpRng(COMPONENT_Cr));
    }
  }
  if (bUseCrossCPrediction)
  {
    CrossComponentPrediction::crossComponentPrediction(tu, compID, cs.getResiBuf(tu.Y()), piResi, piResi, true);
    if( jointCbCr )
    {
      CrossComponentPrediction::crossComponentPrediction(tu, COMPONENT_Cr, cs.getResiBuf(tu.Y()), crResi, crResi, true);
    }
  }

  if (slice.getPicHeader()->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);
    piReco.reconstruct(tmpPred, piResi, cs.slice->clpRng(compID));
  }
  else
  {
    piReco.reconstruct(piPred, piResi, cs.slice->clpRng( compID ));
    if( jointCbCr )
    {
      crReco.reconstruct(crPred, crResi, cs.slice->clpRng( COMPONENT_Cr ));
    }
  }


  //===== update distortion =====
#if WCG_EXT
  if (m_pcEncCfg->getLumaLevelToDeltaQPMapping().isEnabled() || (m_pcEncCfg->getLmcs()
    && slice.getPicHeader()->getLmcsEnabledFlag() && (m_pcReshape->getCTUFlag() || (isChroma(compID) && m_pcEncCfg->getReshapeIntraCMD()))))
  {
    const CPelBuf orgLuma = cs.getOrgBuf( cs.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());
      ruiDist += m_pcRdCost->getDistPart(piOrg, tmpRecLuma, sps.getBitDepth(toChannelType(compID)), compID, DF_SSE_WTD, &orgLuma);
    }
    else
    {
      ruiDist += m_pcRdCost->getDistPart(piOrg, piReco, bitDepth, compID, DF_SSE_WTD, &orgLuma);
      if( jointCbCr )
      {
        ruiDist += m_pcRdCost->getDistPart(crOrg, crReco, bitDepth, COMPONENT_Cr, DF_SSE_WTD, &orgLuma);
      }
    }
  }
  else
#endif
  {
    ruiDist += m_pcRdCost->getDistPart( piOrg, piReco, bitDepth, compID, DF_SSE );
    if( jointCbCr )
    {
      ruiDist += m_pcRdCost->getDistPart( crOrg, crReco, bitDepth, COMPONENT_Cr, DF_SSE );
    }
  }
}

void IntraSearch::xIntraCodingACTTUBlock(TransformUnit &tu, const ComponentID &compID, Distortion& ruiDist, std::vector<TrMode>* trModes, const bool loadTr)
{
  if (!tu.blocks[compID].valid())
  {
    CHECK(1, "tu does not exist");
  }

  CodingStructure     &cs = *tu.cs;
  const SPS           &sps = *cs.sps;
  const Slice         &slice = *cs.slice;
  const CompArea      &area = tu.blocks[compID];
  const CompArea &crArea = tu.blocks[COMPONENT_Cr];

  PelBuf              piOrgResi = cs.getOrgResiBuf(area);
  PelBuf              piResi = cs.getResiBuf(area);
  PelBuf              crOrgResi = cs.getOrgResiBuf(crArea);
  PelBuf              crResi = cs.getResiBuf(crArea);
  TCoeff              uiAbsSum = 0;

  CHECK(tu.jointCbCr && compID == COMPONENT_Cr, "wrong combination of compID and jointCbCr");
  bool jointCbCr = tu.jointCbCr && compID == COMPONENT_Cb;

  m_pcRdCost->setChromaFormat(cs.sps->getChromaFormatIdc());

#if JVET_Q0820_ACT
  if (m_pcEncCfg->getCostMode() != COST_LOSSLESS_CODING)
#endif 
  m_pcTrQuant->lambdaAdjustColorTrans(true);

  if (jointCbCr)
  {
    ComponentID compIdCode = (tu.jointCbCr >> 1 ? COMPONENT_Cb : COMPONENT_Cr);
    m_pcTrQuant->selectLambda(compIdCode);
  }
  else
  {
    m_pcTrQuant->selectLambda(compID);
  }

  bool flag = slice.getPicHeader()->getLmcsEnabledFlag() && (slice.isIntra() || (!slice.isIntra() && m_pcReshape->getCTUFlag())) && (tu.blocks[compID].width*tu.blocks[compID].height > 4);
  if (flag && isChroma(compID) && slice.getPicHeader()->getLmcsChromaResidualScaleFlag())
  {
    int    cResScaleInv = tu.getChromaAdj();
    double cResScale = (double)(1 << CSCALE_FP_PREC) / (double)cResScaleInv;
    m_pcTrQuant->setLambda(m_pcTrQuant->getLambda() / (cResScale*cResScale));
  }

  if (jointCbCr)
  {
    // Lambda is loosened for the joint mode with respect to single modes as the same residual is used for both chroma blocks
    const int    absIct = abs(TU::getICTMode(tu));
    const double lfact = (absIct == 1 || absIct == 3 ? 0.8 : 0.5);
    m_pcTrQuant->setLambda(lfact * m_pcTrQuant->getLambda());
  }
  if (sps.getJointCbCrEnabledFlag() && isChroma(compID) && (slice.getSliceQp() > 18))
  {
    m_pcTrQuant->setLambda(1.3 * m_pcTrQuant->getLambda());
  }

  if (isLuma(compID))
  {
    QpParam cQP(tu, compID);
#if !JVET_Q0820_ACT 
    for (int qpIdx = 0; qpIdx < 2; qpIdx++)
    {
      cQP.Qps[qpIdx] = cQP.Qps[qpIdx] + (compID == COMPONENT_Cr ? DELTA_QP_FOR_Co : DELTA_QP_FOR_Y_Cg);
      cQP.pers[qpIdx] = cQP.Qps[qpIdx] / 6;
      cQP.rems[qpIdx] = cQP.Qps[qpIdx] % 6;
    }
#endif

    if (trModes)
    {
      m_pcTrQuant->transformNxN(tu, compID, cQP, trModes, m_pcEncCfg->getMTSIntraMaxCand());
      tu.mtsIdx[compID] = trModes->at(0).first;
    }
#if JVET_Q0820_ACT 
    if (!(m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && tu.mtsIdx[compID] == 0) || tu.cu->bdpcmMode != 0)
#endif 
    m_pcTrQuant->transformNxN(tu, compID, cQP, uiAbsSum, m_CABACEstimator->getCtx(), loadTr);
#if JVET_Q0820_ACT
    if ((m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && tu.mtsIdx[compID] == 0) && tu.cu->bdpcmMode == 0)
    {
      uiAbsSum = 0;
      tu.getCoeffs(compID).fill(0);
      TU::setCbfAtDepth(tu, compID, tu.depth, 0);
    }
#endif 

    if (uiAbsSum > 0)
    {
      m_pcTrQuant->invTransformNxN(tu, compID, piResi, cQP);
    }
    else
    {
      piResi.fill(0);
    }
  }
  else
  {
    int         codedCbfMask = 0;
    ComponentID codeCompId = (tu.jointCbCr ? (tu.jointCbCr >> 1 ? COMPONENT_Cb : COMPONENT_Cr) : compID);
    QpParam qpCbCr(tu, codeCompId);
#if !JVET_Q0820_ACT 
    for (int qpIdx = 0; qpIdx < 2; qpIdx++)
    {
      qpCbCr.Qps[qpIdx] = qpCbCr.Qps[qpIdx] + (codeCompId == COMPONENT_Cr ? DELTA_QP_FOR_Co : DELTA_QP_FOR_Y_Cg);
      qpCbCr.pers[qpIdx] = qpCbCr.Qps[qpIdx] / 6;
      qpCbCr.rems[qpIdx] = qpCbCr.Qps[qpIdx] % 6;
    }
#endif

    if (tu.jointCbCr)
    {
      ComponentID otherCompId = (codeCompId == COMPONENT_Cr ? COMPONENT_Cb : COMPONENT_Cr);
      tu.getCoeffs(otherCompId).fill(0);
      TU::setCbfAtDepth(tu, otherCompId, tu.depth, false);
    }

    PelBuf& codeResi = (codeCompId == COMPONENT_Cr ? crResi : piResi);
    uiAbsSum = 0;
#if JVET_Q0820_ACT
    if (trModes)
    {
      m_pcTrQuant->transformNxN(tu, codeCompId, qpCbCr, trModes, m_pcEncCfg->getMTSIntraMaxCand());
      tu.mtsIdx[codeCompId] = trModes->at(0).first;
      if (tu.jointCbCr)
      {
        tu.mtsIdx[(codeCompId == COMPONENT_Cr) ? COMPONENT_Cb : COMPONENT_Cr] = MTS_DCT2_DCT2;
      }
    }
    if (!(m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && tu.mtsIdx[codeCompId] == 0) || tu.cu->bdpcmModeChroma != 0)
    {
      m_pcTrQuant->transformNxN(tu, codeCompId, qpCbCr, uiAbsSum, m_CABACEstimator->getCtx(), loadTr);
    }
#else
    m_pcTrQuant->transformNxN(tu, codeCompId, qpCbCr, uiAbsSum, m_CABACEstimator->getCtx());
#endif
    if (uiAbsSum > 0)
    {
      m_pcTrQuant->invTransformNxN(tu, codeCompId, codeResi, qpCbCr);
      codedCbfMask += (codeCompId == COMPONENT_Cb ? 2 : 1);
    }
    else
    {
      codeResi.fill(0);
    }

    if (tu.jointCbCr)
    {
      if (tu.jointCbCr == 3 && codedCbfMask == 2)
      {
        codedCbfMask = 3;
        TU::setCbfAtDepth(tu, COMPONENT_Cr, tu.depth, true);
      }
      if (tu.jointCbCr != codedCbfMask)
      {
        ruiDist = std::numeric_limits<Distortion>::max();
#if JVET_Q0820_ACT
        if (m_pcEncCfg->getCostMode() != COST_LOSSLESS_CODING)
#endif 
        m_pcTrQuant->lambdaAdjustColorTrans(false);
        return;
      }
      m_pcTrQuant->invTransformICT(tu, piResi, crResi);
      uiAbsSum = codedCbfMask;
    }
  }

  if (flag && uiAbsSum > 0 && isChroma(compID) && slice.getPicHeader()->getLmcsChromaResidualScaleFlag())
  {
    piResi.scaleSignal(tu.getChromaAdj(), 0, slice.clpRng(compID));
    if (jointCbCr)
    {
      crResi.scaleSignal(tu.getChromaAdj(), 0, slice.clpRng(COMPONENT_Cr));
    }
  }

#if JVET_Q0820_ACT
  if (m_pcEncCfg->getCostMode() != COST_LOSSLESS_CODING)
#endif 
  m_pcTrQuant->lambdaAdjustColorTrans(false);

  ruiDist += m_pcRdCost->getDistPart(piOrgResi, piResi, sps.getBitDepth(toChannelType(compID)), compID, DF_SSE);
  if (jointCbCr)
  {
    ruiDist += m_pcRdCost->getDistPart(crOrgResi, crResi, sps.getBitDepth(toChannelType(COMPONENT_Cr)), COMPONENT_Cr, DF_SSE);
  }
}

bool IntraSearch::xIntraCodingLumaISP(CodingStructure& cs, Partitioner& partitioner, const double bestCostSoFar)
{
  int               subTuCounter = 0;
  const CodingUnit& cu = *cs.getCU(partitioner.currArea().lumaPos(), partitioner.chType);
  bool              earlySkipISP = false;
  bool              splitCbfLuma = false;
  const PartSplit   ispType = CU::getISPType(cu, COMPONENT_Y);

  cs.cost = 0;

  partitioner.splitCurrArea(ispType, cs);

  CUCtx cuCtx;
  cuCtx.isDQPCoded = true;
  cuCtx.isChromaQpAdjCoded = true;

  do   // subpartitions loop
  {
    uint32_t   numSig = 0;
    Distortion singleDistTmpLuma = 0;
    uint64_t   singleTmpFracBits = 0;
    double     singleCostTmp = 0;

    TransformUnit& tu = cs.addTU(CS::getArea(cs, partitioner.currArea(), partitioner.chType), partitioner.chType);
    tu.depth = partitioner.currTrDepth;

    // Encode TU
    xIntraCodingTUBlock(tu, COMPONENT_Y, false, singleDistTmpLuma, 0, &numSig);

    if (singleDistTmpLuma == MAX_INT)   // all zero CBF skip
    {
      earlySkipISP = true;
      partitioner.exitCurrSplit();
      cs.cost = MAX_DOUBLE;
      return false;
    }

    {
      if (m_pcRdCost->calcRdCost(cs.fracBits, cs.dist + singleDistTmpLuma) > bestCostSoFar)
      {
        // The accumulated cost + distortion is already larger than the best cost so far, so it is not necessary to calculate the rate
        earlySkipISP = true;
      }
      else
      {
        singleTmpFracBits = xGetIntraFracBitsQT(cs, partitioner, true, false, subTuCounter, ispType, &cuCtx);
      }
      singleCostTmp = m_pcRdCost->calcRdCost(singleTmpFracBits, singleDistTmpLuma);
    }

    cs.cost += singleCostTmp;
    cs.dist += singleDistTmpLuma;
    cs.fracBits += singleTmpFracBits;

    subTuCounter++;

    splitCbfLuma |= TU::getCbfAtDepth(*cs.getTU(partitioner.currArea().lumaPos(), partitioner.chType, subTuCounter - 1), COMPONENT_Y, partitioner.currTrDepth);
    int nSubPartitions = m_ispTestedModes[cu.lfnstIdx].numTotalParts[cu.ispMode - 1];
    if (subTuCounter < nSubPartitions)
    {
      // exit condition if the accumulated cost is already larger than the best cost so far (no impact in RD performance)
      if (cs.cost > bestCostSoFar)
      {
        earlySkipISP = true;
        break;
      }
      else if (subTuCounter < nSubPartitions)
      {
        // more restrictive exit condition
        double threshold = nSubPartitions == 2 ? 0.95 : subTuCounter == 1 ? 0.83 : 0.91;
        if (subTuCounter < nSubPartitions && cs.cost > bestCostSoFar * threshold)
        {
          earlySkipISP = true;
          break;
        }
      }
    }
  } while (partitioner.nextPart(cs));   // subpartitions loop

  partitioner.exitCurrSplit();
  const UnitArea& currArea = partitioner.currArea();
  const uint32_t  currDepth = partitioner.currTrDepth;

  if (earlySkipISP)
  {
    cs.cost = MAX_DOUBLE;
  }
  else
  {
    cs.cost = m_pcRdCost->calcRdCost(cs.fracBits, cs.dist);
    // The cost check is necessary here again to avoid superfluous operations if the maximum number of coded subpartitions was reached and yet ISP did not win
    if (cs.cost < bestCostSoFar)
    {
      cs.setDecomp(cu.Y());
      cs.picture->getRecoBuf(currArea.Y()).copyFrom(cs.getRecoBuf(currArea.Y()));

      for (auto& ptu : cs.tus)
      {
        if (currArea.Y().contains(ptu->Y()))
        {
          TU::setCbfAtDepth(*ptu, COMPONENT_Y, currDepth, splitCbfLuma ? 1 : 0);
        }
      }
    }
    else
    {
      earlySkipISP = true;
    }
  }
  return !earlySkipISP;
}


bool IntraSearch::xRecurIntraCodingLumaQT( CodingStructure &cs, Partitioner &partitioner, const double bestCostSoFar, const int subTuIdx, const PartSplit ispType, const bool ispIsCurrentWinner, bool mtsCheckRangeFlag, int mtsFirstCheckId, int mtsLastCheckId, bool moreProbMTSIdxFirst )
{
        int   subTuCounter = subTuIdx;
  const UnitArea &currArea = partitioner.currArea();
  const CodingUnit     &cu = *cs.getCU( currArea.lumaPos(), partitioner.chType );
        bool  earlySkipISP = false;
  uint32_t currDepth       = partitioner.currTrDepth;
  const SPS &sps           = *cs.sps;
  const PPS &pps           = *cs.pps;
  const bool keepResi      = pps.getPpsRangeExtension().getCrossComponentPredictionEnabledFlag() || KEEP_PRED_AND_RESI_SIGNALS;
  bool bCheckFull          = true;
  bool bCheckSplit         = false;
  bCheckFull               = !partitioner.canSplit( TU_MAX_TR_SPLIT, cs );
  bCheckSplit              = partitioner.canSplit( TU_MAX_TR_SPLIT, cs );

  if( cu.ispMode )
  {
    bCheckSplit = partitioner.canSplit( ispType, cs );
    bCheckFull = !bCheckSplit;
  }
  uint32_t    numSig           = 0;

  double     dSingleCost                        = MAX_DOUBLE;
  Distortion uiSingleDistLuma                   = 0;
  uint64_t   singleFracBits                     = 0;
  bool       checkTransformSkip                 = sps.getTransformSkipEnabledFlag();
  int        bestModeId[ MAX_NUM_COMPONENT ]    = { 0, 0, 0 };
  uint8_t    nNumTransformCands                 = cu.mtsFlag ? 4 : 1;
  uint8_t    numTransformIndexCands             = nNumTransformCands;

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

  CodingStructure *csSplit = nullptr;
  CodingStructure *csFull  = nullptr;

  CUCtx cuCtx;
  cuCtx.isDQPCoded = true;
  cuCtx.isChromaQpAdjCoded = true;

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

  bool validReturnFull = false;

  if( bCheckFull )
  {
    csFull->cost = 0.0;

    TransformUnit &tu = csFull->addTU( CS::getArea( *csFull, currArea, partitioner.chType ), partitioner.chType );
    tu.depth = currDepth;

    const bool tsAllowed  = TU::isTSAllowed( tu, COMPONENT_Y );
    const bool mtsAllowed = CU::isMTSAllowed( cu, COMPONENT_Y );
    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

#if JVET_Q0820_ACT 
      if (m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING)
      {
        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
      {
#endif
      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 ) );
        }
      }
#if JVET_Q0820_ACT 
      }
#endif 
    }

    CHECK( !tu.Y().valid(), "Invalid TU" );

    CodingStructure &saveCS = *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;

    //we add the MTS candidates to the loop. TransformSkip will still be the last one to be checked (when modeId == lastCheckId) as long as checkTransformSkip is true
    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 )
    {
      saveCS.pcv     = cs.pcv;
      saveCS.picture = cs.picture;
      saveCS.area.repositionTo(cs.area);
      saveCS.clearTUs();
      tmpTU = &saveCS.addTU(currArea, partitioner.chType);
    }

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

    double bestDCT2cost = MAX_DOUBLE;
    double threshold = m_pcEncCfg->getUseFastISP() && !cu.ispMode && ispIsCurrentWinner && nNumTransformCands > 1 ? 1 + 1.4 / sqrt( cu.lwidth() * cu.lheight() ) : 1;
    for( int modeId = firstCheckId; modeId <= ( sps.getUseLFNST() ? lastCheckId : ( nNumTransformCands - 1 ) ); modeId++ )
    {
      uint8_t transformIndex = modeId;

      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 JVET_AHG14_LOSSLESS
        if( !( m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING ) )
        {
#endif
        if( !cbfDCT2 || ( m_pcEncCfg->getUseTransformSkipFast() && bestModeId[ COMPONENT_Y ] == MTS_SKIP))
        {
          break;
        }
        if( !trModes[ modeId ].second )
        {
          continue;
        }
        //we compare the DCT-II cost against the best ISP cost so far (except for TS)
        if (m_pcEncCfg->getUseFastISP() && !cu.ispMode && ispIsCurrentWinner && trModes[modeId].first != MTS_DCT2_DCT2 && (trModes[modeId].first != MTS_SKIP || !tsAllowed) && bestDCT2cost > bestCostSoFar * threshold)
        {
          continue;
        }
#if JVET_AHG14_LOSSLESS
        }
#endif
        tu.mtsIdx[COMPONENT_Y] = trModes[modeId].first;
      }


      if ((modeId != firstCheckId) && isNotOnlyOneMode)
      {
        m_CABACEstimator->getCtx() = ctxStart;
      }

      int default0Save1Load2 = 0;
      singleDistTmpLuma = 0;

      if( modeId == firstCheckId && ( sps.getUseLFNST() ? ( modeId != lastCheckId ) : ( nNumTransformCands > 1 ) ) )
      {
        default0Save1Load2 = 1;
      }
      else if (modeId != firstCheckId)
      {
        if( sps.getUseLFNST() && !cbfBestModeValid )
        {
          default0Save1Load2 = 1;
        }
        else
        {
          default0Save1Load2 = 2;
        }
      }
      if( cu.ispMode )
      {
        default0Save1Load2 = 0;
      }
      if( sps.getUseLFNST() )
      {
        if( cu.mtsFlag )
        {
          if( moreProbMTSIdxFirst )
          {
            const ChannelType     chType      = toChannelType( COMPONENT_Y );
            const CompArea&       area        = tu.blocks[ COMPONENT_Y ];
            const PredictionUnit& pu          = *cs.getPU( area.pos(), chType );
            uint32_t              uiIntraMode = pu.intraDir[ chType ];

            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 )
        {
          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 );
        }
      }

#if JVET_Q0516_MTS_SIGNALLING_DC_ONLY_COND 
      cuCtx.mtsLastScanPos = false;
#endif
      //----- 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.
#if JVET_Q0820_ACT 
        if (m_pcEncCfg->getCostMode() != COST_LOSSLESS_CODING)
#endif 
        singleCostTmp = MAX_DOUBLE;
#if JVET_Q0820_ACT 
        else
        {
          singleTmpFracBits = xGetIntraFracBitsQT(*csFull, partitioner, true, false, subTuCounter, ispType, &cuCtx);
          singleCostTmp = m_pcRdCost->calcRdCost(singleTmpFracBits, singleDistTmpLuma);
        }
#endif 
      }
      else
      {
        if( cu.ispMode && m_pcRdCost->calcRdCost( csFull->fracBits, csFull->dist + singleDistTmpLuma ) > bestCostSoFar )
        {
          earlySkipISP = true;
        }
        else
        {
          singleTmpFracBits = xGetIntraFracBitsQT( *csFull, partitioner, true, false, subTuCounter, ispType, &cuCtx );
        }
#if JVET_Q0516_MTS_SIGNALLING_DC_ONLY_COND  
        if (tu.mtsIdx[COMPONENT_Y] > MTS_SKIP)
        {
          if (!cuCtx.mtsLastScanPos)
          {
            singleCostTmp = MAX_DOUBLE;
          }
          else
          {
            singleCostTmp = m_pcRdCost->calcRdCost(singleTmpFracBits, singleDistTmpLuma);
          }
        }
        else
#endif
        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;

      cuCtx.violatesLfnstConstrained[CHANNEL_TYPE_LUMA] = false;
      cuCtx.violatesLfnstConstrained[CHANNEL_TYPE_CHROMA] = false;
      cuCtx.lfnstLastScanPos = false;
      cuCtx.violatesMtsCoeffConstraint = false;
#if JVET_Q0516_MTS_SIGNALLING_DC_ONLY_COND 
      cuCtx.mtsLastScanPos = false;
#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( !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;
}

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);