IntraSearch.cpp

        if (PU::isLMCMode(mode))
        {
          predIntraChromaLM(COMPONENT_Cb, predCb, pu, areaCb, mode);
        }
        else
        {
          predIntraAng(COMPONENT_Cb, predCb, pu, false);
        }

        sad += distParam.distFunc(distParam);

        CompArea areaCr = pu.Cr();
        PelBuf orgCr = cs.getOrgBuf(areaCr);
        PelBuf predCr = cs.getPredBuf(areaCr);

        m_pcRdCost->setDistParam(distParam, orgCr, predCr, pu.cs->sps->getBitDepth(CHANNEL_TYPE_CHROMA), COMPONENT_Cr, useHadamard);
        distParam.applyWeight = false;

        if (PU::isLMCMode(mode))
        {
          predIntraChromaLM(COMPONENT_Cr, predCr, pu, areaCr, mode);
        }
        else
        {
          predIntraAng(COMPONENT_Cr, predCr, pu, false);
        }
        sad += distParam.distFunc(distParam);
        satdSortedCost[idx] = sad;
      }
      // sort the mode based on the cost from small to large.
      int tempIdx = 0;
      int64_t tempCost = 0;
      for (int i = uiMinMode; i <= uiMaxMode - 1; i++)
      {
        for (int j = i + 1; j <= uiMaxMode - 1; j++)
        {
          if (satdSortedCost[j] < satdSortedCost[i])
          {
            tempIdx = satdModeList[i];
            satdModeList[i] = satdModeList[j];
            satdModeList[j] = tempIdx;

            tempCost = satdSortedCost[i];
            satdSortedCost[i] = satdSortedCost[j];
            satdSortedCost[j] = tempCost;

          }
        }
      }
      int reducedModeNumber = 2; // reduce the number of chroma modes
      for (int i = 0; i < reducedModeNumber; i++)
      {
        modeIsEnable[satdModeList[uiMaxMode - 1 - i]] = 0; // disable the last reducedModeNumber modes
      }

      // save the dist
      Distortion baseDist = cs.dist;

      for (uint32_t uiMode = uiMinMode; uiMode < uiMaxMode; uiMode++)
      {
        const int chromaIntraMode = chromaCandModes[uiMode];
        if( PU::isLMCMode( chromaIntraMode ) && ! PU::isLMCModeEnabled( pu, chromaIntraMode ) )
        {
          continue;
        }
        if (!modeIsEnable[chromaIntraMode] && PU::isLMCModeEnabled(pu, chromaIntraMode)) // when CCLM is disable, then MDLM is disable. not use satd checking
        {
          continue;
        }
        cs.setDecomp( pu.Cb(), false );
        cs.dist = baseDist;
        //----- restore context models -----
        m_CABACEstimator->getCtx() = ctxStart;

        //----- chroma coding -----
        pu.intraDir[1] = chromaIntraMode;

        xRecurIntraChromaCodingQT( cs, partitioner, bestCostSoFar, ispType );
        if( lumaUsesISP && cs.dist == MAX_UINT )
        {
          continue;
        }

        if (cs.pps->getUseTransformSkip())
        {
          m_CABACEstimator->getCtx() = ctxStart;
        }

        uint64_t fracBits   = xGetIntraFracBitsQT( cs, partitioner, false, true, -1, ispType );
        Distortion uiDist = cs.dist;
        double    dCost   = m_pcRdCost->calcRdCost( fracBits, uiDist - baseDist );

        //----- compare -----
        if( dCost < dBestCost )
        {
          if( lumaUsesISP && dCost < bestCostSoFar )
          {
            bestCostSoFar = dCost;
          }
          for( uint32_t i = getFirstComponentOfChannel( CHANNEL_TYPE_CHROMA ); i < numberValidComponents; i++ )
          {
            const CompArea &area = pu.blocks[i];

            saveCS.getRecoBuf     ( area ).copyFrom( cs.getRecoBuf   ( area ) );
#if KEEP_PRED_AND_RESI_SIGNALS
            saveCS.getPredBuf     ( area ).copyFrom( cs.getPredBuf   ( area ) );
            saveCS.getResiBuf     ( area ).copyFrom( cs.getResiBuf   ( area ) );
#endif
            saveCS.getPredBuf     ( area ).copyFrom( cs.getPredBuf   (area ) );
            cs.picture->getPredBuf( area ).copyFrom( cs.getPredBuf   (area ) );
            cs.picture->getRecoBuf( area ).copyFrom( cs.getRecoBuf( area ) );

            for( uint32_t j = 0; j < saveCS.tus.size(); j++ )
            {
              saveCS.tus[j]->copyComponentFrom( *orgTUs[j], area.compID );
            }
          }

          dBestCost  = dCost;
          uiBestDist = uiDist;
          uiBestMode = chromaIntraMode;
        }
      }

      for( uint32_t i = getFirstComponentOfChannel( CHANNEL_TYPE_CHROMA ); i < numberValidComponents; i++ )
      {
        const CompArea &area = pu.blocks[i];

        cs.getRecoBuf         ( area ).copyFrom( saveCS.getRecoBuf( area ) );
#if KEEP_PRED_AND_RESI_SIGNALS
        cs.getPredBuf         ( area ).copyFrom( saveCS.getPredBuf( area ) );
        cs.getResiBuf         ( area ).copyFrom( saveCS.getResiBuf( area ) );
#endif
        cs.getPredBuf         ( area ).copyFrom( saveCS.getPredBuf( area ) );
        cs.picture->getPredBuf( area ).copyFrom( cs.getPredBuf    ( area ) );

        cs.picture->getRecoBuf( area ).copyFrom( cs.    getRecoBuf( area ) );

        for( uint32_t j = 0; j < saveCS.tus.size(); j++ )
        {
          orgTUs[ j ]->copyComponentFrom( *saveCS.tus[ j ], area.compID );
        }
      }
    }

    pu.intraDir[1] = uiBestMode;
    cs.dist        = uiBestDist;
  }

  //----- restore context models -----
  m_CABACEstimator->getCtx() = ctxStart;
  if( lumaUsesISP && bestCostSoFar >= maxCostAllowed )
  {
    cu.ispMode = 0;
  }
}

void IntraSearch::IPCMSearch(CodingStructure &cs, Partitioner& partitioner)
{
  ComponentID compStr = (CS::isDualITree(cs) && !isLuma(partitioner.chType)) ? COMPONENT_Cb: COMPONENT_Y;
  ComponentID compEnd = (CS::isDualITree(cs) && isLuma(partitioner.chType)) ? COMPONENT_Y : COMPONENT_Cr;
  for( ComponentID compID = compStr; compID <= compEnd; compID = ComponentID(compID+1) )
  {

    xEncPCM(cs, partitioner, compID);
  }

  cs.getPredBuf().fill(0);
  cs.getResiBuf().fill(0);
  cs.getOrgResiBuf().fill(0);

  cs.dist     = 0;
  cs.fracBits = 0;
  cs.cost     = 0;

  cs.setDecomp(cs.area);
  cs.picture->getPredBuf(cs.area).copyFrom(cs.getPredBuf());
}

void IntraSearch::xEncPCM(CodingStructure &cs, Partitioner& partitioner, const ComponentID &compID)
{
  TransformUnit &tu = *cs.getTU( partitioner.chType );

  const int  channelBitDepth = cs.sps->getBitDepth(toChannelType(compID));
  const uint32_t uiPCMBitDepth = cs.sps->getPCMBitDepth(toChannelType(compID));

  const int pcmShiftRight = (channelBitDepth - int(uiPCMBitDepth));

  CompArea  area    = tu.blocks[compID];
  PelBuf    pcmBuf  = tu.getPcmbuf  (compID);
  PelBuf    recBuf  = cs.getRecoBuf ( area );
  CPelBuf   orgBuf  = cs.getOrgBuf  ( area );

  CHECK(pcmShiftRight < 0, "Negative shift");
  CompArea      tmpArea(COMPONENT_Y, area.chromaFormat, Position(0, 0), area.size());
  PelBuf tempOrgBuf = m_tmpStorageLCU.getBuf(tmpArea);
  tempOrgBuf.copyFrom(orgBuf);
  if (cs.slice->getReshapeInfo().getUseSliceReshaper() && m_pcReshape->getCTUFlag() && compID == COMPONENT_Y)
  {
    tempOrgBuf.rspSignal(m_pcReshape->getFwdLUT());
  }
  for (uint32_t uiY = 0; uiY < pcmBuf.height; uiY++)
  {
    for (uint32_t uiX = 0; uiX < pcmBuf.width; uiX++)
    {
      // Encode
      pcmBuf.at(uiX, uiY) = tempOrgBuf.at(uiX, uiY) >> pcmShiftRight;
      // Reconstruction
      recBuf.at(uiX, uiY) = pcmBuf.at(uiX, uiY) << pcmShiftRight;
    }
  }
}

// -------------------------------------------------------------------------------------------------------------------
// Intra search
// -------------------------------------------------------------------------------------------------------------------

void IntraSearch::xEncIntraHeader( CodingStructure &cs, Partitioner &partitioner, const bool &bLuma, const bool &bChroma, const int subTuIdx )
{
  CodingUnit &cu = *cs.getCU( partitioner.chType );

  if (bLuma)
  {
    bool isFirst = cu.ispMode ? subTuIdx == 0 : partitioner.currArea().lumaPos() == cs.area.lumaPos();

    // CU header
    if( isFirst )
    {
      if ((!cs.slice->isIntra() || cs.slice->getSPS()->getIBCFlag())
        && cu.Y().valid()
        )
      {
        if( cs.pps->getTransquantBypassEnabledFlag() )
        {
          m_CABACEstimator->cu_transquant_bypass_flag( cu );
        }
        m_CABACEstimator->cu_skip_flag( cu );
        m_CABACEstimator->pred_mode   ( cu );
      }
      if( CU::isIntra(cu) )
      {
        m_CABACEstimator->pcm_data( cu, partitioner );
        if( cu.ipcm )
        {
          return;
        }
      }
      m_CABACEstimator->extend_ref_line(cu);
      m_CABACEstimator->isp_mode      ( cu );
    }

    PredictionUnit &pu = *cs.getPU(partitioner.currArea().lumaPos(), partitioner.chType);

    // luma prediction mode
    if (isFirst)
    {
      if ( !cu.Y().valid())
        m_CABACEstimator->pred_mode( cu );
      m_CABACEstimator->intra_luma_pred_mode( pu );
    }
  }

  if (bChroma)
  {
    bool isFirst = partitioner.currArea().Cb().valid() && partitioner.currArea().chromaPos() == cs.area.chromaPos();

    PredictionUnit &pu = *cs.getPU( partitioner.currArea().chromaPos(), CHANNEL_TYPE_CHROMA );

    if( isFirst )
    {
      m_CABACEstimator->intra_chroma_pred_mode( pu );
    }
  }
}

void IntraSearch::xEncSubdivCbfQT( CodingStructure &cs, Partitioner &partitioner, const bool &bLuma, const bool &bChroma, const int subTuIdx, const PartSplit ispType )
{
  const UnitArea &currArea = partitioner.currArea();
          int subTuCounter = subTuIdx;
  TransformUnit &currTU = *cs.getTU( currArea.blocks[partitioner.chType], partitioner.chType, subTuCounter );
  CodingUnit    &currCU = *currTU.cu;
  uint32_t currDepth           = partitioner.currTrDepth;

  const bool subdiv        = currTU.depth > currDepth;
  ComponentID compID = partitioner.chType == CHANNEL_TYPE_LUMA ? COMPONENT_Y : COMPONENT_Cb;
  const bool chromaCbfISP = currArea.blocks[COMPONENT_Cb].valid() && currCU.ispMode && !subdiv;

  if( partitioner.canSplit( TU_MAX_TR_SPLIT, cs ) )
  {
    CHECK( !subdiv, "TU split implied" );
  }
  else
  {
    CHECK( subdiv && !currCU.ispMode && isLuma( compID ), "No TU subdivision is allowed with QTBT" );
  }

  if( bChroma && ( !currCU.ispMode || chromaCbfISP ) )
  {
    const uint32_t numberValidComponents = getNumberValidComponents(currArea.chromaFormat);
    const uint32_t cbfDepth = ( chromaCbfISP ? currDepth - 1 : currDepth );

    for (uint32_t ch = COMPONENT_Cb; ch < numberValidComponents; ch++)
    {
      const ComponentID compID = ComponentID(ch);

      if( currDepth == 0 || TU::getCbfAtDepth( currTU, compID, currDepth - 1 ) || chromaCbfISP )
      {
        const bool prevCbf = ( compID == COMPONENT_Cr ? TU::getCbfAtDepth( currTU, COMPONENT_Cb, currDepth ) : false );
        m_CABACEstimator->cbf_comp( cs, TU::getCbfAtDepth( currTU, compID, currDepth ), currArea.blocks[compID], cbfDepth, prevCbf );

      }
    }
  }

  if (subdiv)
  {

    if( partitioner.canSplit( TU_MAX_TR_SPLIT, cs ) )
    {
      partitioner.splitCurrArea( TU_MAX_TR_SPLIT, cs );
    }
    else if( currCU.ispMode && isLuma( compID ) )
    {
      partitioner.splitCurrArea( ispType, cs );
    }
    else
    THROW( "Cannot perform an implicit split!" );

    do
    {
      xEncSubdivCbfQT( cs, partitioner, bLuma, bChroma, subTuCounter, ispType );
      subTuCounter += subTuCounter != -1 ? 1 : 0;
    } while( partitioner.nextPart( cs ) );

    partitioner.exitCurrSplit();
  }
  else
  {
    //===== Cbfs =====
    if (bLuma)
    {
      bool previousCbf       = false;
      bool lastCbfIsInferred = false;
      if( ispType != TU_NO_ISP )
      {
        bool rootCbfSoFar = false;
        uint32_t nTus = currCU.ispMode == HOR_INTRA_SUBPARTITIONS ? currCU.lheight() >> g_aucLog2[currTU.lheight()] : currCU.lwidth() >> g_aucLog2[currTU.lwidth()];
        if( subTuCounter == nTus - 1 )
        {
          TransformUnit* tuPointer = currCU.firstTU;
          for( int tuIdx = 0; tuIdx < nTus - 1; tuIdx++ )
          {
            rootCbfSoFar |= TU::getCbfAtDepth( *tuPointer, COMPONENT_Y, currDepth );
            tuPointer = tuPointer->next;
          }
          if( !rootCbfSoFar )
          {
            lastCbfIsInferred = true;
          }
        }
        if( !lastCbfIsInferred )
        {
          previousCbf = TU::getPrevTuCbfAtDepth( currTU, COMPONENT_Y, partitioner.currTrDepth );
        }
      }
      if( !lastCbfIsInferred )
      {
        m_CABACEstimator->cbf_comp( cs, TU::getCbfAtDepth( currTU, COMPONENT_Y, currDepth ), currTU.Y(), currTU.depth, previousCbf, currCU.ispMode );
      }
    }
  }
}

void IntraSearch::xEncCoeffQT( CodingStructure &cs, Partitioner &partitioner, const ComponentID compID, const int subTuIdx, const PartSplit ispType )
{
  const UnitArea &currArea  = partitioner.currArea();

       int subTuCounter     = subTuIdx;
  TransformUnit &currTU     = *cs.getTU( currArea.blocks[partitioner.chType], partitioner.chType, subTuIdx );
  uint32_t      currDepth       = partitioner.currTrDepth;
  const bool subdiv         = currTU.depth > currDepth;

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

    do
    {
      xEncCoeffQT( cs, partitioner, compID, subTuCounter, ispType );
      subTuCounter += subTuCounter != -1 ? 1 : 0;
    } while( partitioner.nextPart( cs ) );

    partitioner.exitCurrSplit();
  }
  else

  if( currArea.blocks[compID].valid() )
  {
    if( TU::hasCrossCompPredInfo( currTU, compID ) )
    {
      m_CABACEstimator->cross_comp_pred( currTU, compID );
    }
    if( TU::getCbf( currTU, compID ) )
    {
      m_CABACEstimator->residual_coding( currTU, compID );
    }
  }
}

uint64_t IntraSearch::xGetIntraFracBitsQT( CodingStructure &cs, Partitioner &partitioner, const bool &bLuma, const bool &bChroma, const int subTuIdx, const PartSplit ispType )
{
  m_CABACEstimator->resetBits();

  xEncIntraHeader( cs, partitioner, bLuma, bChroma, subTuIdx );
  xEncSubdivCbfQT( cs, partitioner, bLuma, bChroma, subTuIdx, ispType );


  if( bLuma )
  {
    xEncCoeffQT( cs, partitioner, COMPONENT_Y, subTuIdx, ispType );
  }
  if( bChroma )
  {
    xEncCoeffQT( cs, partitioner, COMPONENT_Cb, subTuIdx, ispType );
    xEncCoeffQT( cs, partitioner, COMPONENT_Cr, subTuIdx, ispType );
  }

  uint64_t fracBits = m_CABACEstimator->getEstFracBits();
  return fracBits;
}

uint64_t IntraSearch::xGetIntraFracBitsQTSingleChromaComponent( CodingStructure &cs, Partitioner &partitioner, const ComponentID compID )
{
  m_CABACEstimator->resetBits();

  if( compID == COMPONENT_Cb )
  {
    //intra mode coding
    PredictionUnit &pu = *cs.getPU( partitioner.currArea().lumaPos(), partitioner.chType );
    m_CABACEstimator->intra_chroma_pred_mode( pu );
    //xEncIntraHeader(cs, partitioner, false, true);
  }
  CHECK( partitioner.currTrDepth != 1, "error in the depth!" );
  const UnitArea &currArea = partitioner.currArea();

  TransformUnit &currTU = *cs.getTU( currArea.blocks[partitioner.chType], partitioner.chType );

  //cbf coding
  m_CABACEstimator->cbf_comp( cs, TU::getCbfAtDepth( currTU, compID, partitioner.currTrDepth ), currArea.blocks[compID], partitioner.currTrDepth - 1 );
  //coeffs coding and cross comp coding
  if( TU::hasCrossCompPredInfo( currTU, compID ) )
  {
    m_CABACEstimator->cross_comp_pred( currTU, compID );
  }
  if( TU::getCbf( currTU, compID ) )
  {
    m_CABACEstimator->residual_coding( currTU, compID );
  }

  uint64_t fracBits = m_CABACEstimator->getEstFracBits();
  return fracBits;
}

uint64_t IntraSearch::xGetIntraFracBitsQTChroma(TransformUnit& currTU, const ComponentID &compID)
{
  m_CABACEstimator->resetBits();

  if( TU::hasCrossCompPredInfo( currTU, compID ) )
  {
    m_CABACEstimator->cross_comp_pred( currTU, compID );
  }
  if( TU::getCbf( currTU, compID ) )
  {
    m_CABACEstimator->residual_coding( currTU, compID );
  }

  uint64_t fracBits = m_CABACEstimator->getEstFracBits();
  return fracBits;
}

void IntraSearch::xIntraCodingTUBlock(TransformUnit &tu, const ComponentID &compID, const bool &checkCrossCPrediction, Distortion& ruiDist, const int &default0Save1Load2, uint32_t* numSig, std::vector<TrMode>* trModes, const bool loadTr)
{
  if (!tu.blocks[compID].valid())
  {
    return;
  }

  CodingStructure &cs                       = *tu.cs;

  const CompArea      &area                 = tu.blocks[compID];
  const SPS           &sps                  = *cs.sps;
  const PPS           &pps                  = *cs.pps;

  const ChannelType    chType               = toChannelType(compID);
  const int            bitDepth             = sps.getBitDepth(chType);

  PelBuf         piOrg                      = cs.getOrgBuf    (area);
  PelBuf         piPred                     = cs.getPredBuf   (area);
  PelBuf         piResi                     = cs.getResiBuf   (area);
  PelBuf         piOrgResi                  = cs.getOrgResiBuf(area);
  PelBuf         piReco                     = cs.getRecoBuf   (area);

  const PredictionUnit &pu                  = *cs.getPU(area.pos(), chType);
  const uint32_t           uiChFinalMode        = PU::getFinalIntraMode(pu, chType);

  const bool           bUseCrossCPrediction = pps.getPpsRangeExtension().getCrossComponentPredictionEnabledFlag() && isChroma( compID ) && PU::isChromaIntraModeCrossCheckMode( pu ) && checkCrossCPrediction;
  const bool           ccUseRecoResi        = m_pcEncCfg->getUseReconBasedCrossCPredictionEstimate();
  const bool           ispSplitIsAllowed    = CU::canUseISPSplit( *tu.cu, compID );


  //===== init availability pattern =====
  PelBuf sharedPredTS( m_pSharedPredTransformSkip[compID], area );
  if( default0Save1Load2 != 2 )
  {
    const bool bUseFilteredPredictions = IntraPrediction::useFilteredIntraRefSamples( compID, pu, true, tu );
    initIntraPatternChType( *tu.cu, area, bUseFilteredPredictions );

    //===== get prediction signal =====
    if( compID != COMPONENT_Y && PU::isLMCMode( uiChFinalMode ) )
    {
      {
        xGetLumaRecPixels( pu, area );
      }
      predIntraChromaLM( compID, piPred, pu, area, uiChFinalMode );
    }
    else
    {
      predIntraAng( compID, piPred, pu, bUseFilteredPredictions );
    }


    // save prediction
    if( default0Save1Load2 == 1 )
    {
      sharedPredTS.copyFrom( piPred );
    }
  }
  else
  {
    // load prediction
    piPred.copyFrom( sharedPredTS );
  }


  DTRACE( g_trace_ctx, D_PRED, "@(%4d,%4d) [%2dx%2d] IMode=%d\n", tu.lx(), tu.ly(), tu.lwidth(), tu.lheight(), uiChFinalMode );
  //DTRACE_PEL_BUF( D_PRED, piPred, tu, tu.cu->predMode, COMPONENT_Y );

  const Slice           &slice = *cs.slice;
  bool flag = slice.getReshapeInfo().getUseSliceReshaper() && (slice.isIntra() || (!slice.isIntra() && m_pcReshape->getCTUFlag()));
  if (flag && slice.getReshapeInfo().getSliceReshapeChromaAdj() && isChroma(compID))
  {
    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 );
    PelBuf piPredY;
    piPredY = cs.picture->getPredBuf(areaY);
    const Pel avgLuma = piPredY.computeAvg();
    int adj = m_pcReshape->calculateChromaAdj(avgLuma);
    tu.setChromaAdj(adj);
  }
  //===== get residual signal =====
  piResi.copyFrom( piOrg  );
  if (slice.getReshapeInfo().getUseSliceReshaper() && 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 );

  if (pps.getPpsRangeExtension().getCrossComponentPredictionEnabledFlag() && isLuma(compID))
  {
    piOrgResi.copyFrom (piResi);
  }

  if (bUseCrossCPrediction)
  {
    if (xCalcCrossComponentPredictionAlpha(tu, compID, ccUseRecoResi) == 0)
    {
      return;
    }
    CrossComponentPrediction::crossComponentPrediction(tu, compID, cs.getResiBuf(tu.Y()), piResi, piResi, false);
  }

  //===== transform and quantization =====
  //--- init rate estimation arrays for RDOQ ---
  //--- transform and quantization           ---
  TCoeff uiAbsSum = 0;

  const QpParam cQP(tu, compID);

#if RDOQ_CHROMA_LAMBDA
  m_pcTrQuant->selectLambda(compID);
#endif

  flag =flag && (tu.blocks[compID].width*tu.blocks[compID].height > 4);
  if (flag && isChroma(compID) && slice.getReshapeInfo().getSliceReshapeChromaAdj() )
  {
    int cResScaleInv = tu.getChromaAdj();
    double cResScale = round((double)(1 << CSCALE_FP_PREC) / (double)cResScaleInv);
    m_pcTrQuant->setLambda(m_pcTrQuant->getLambda() / (cResScale*cResScale));
    piResi.scaleSignal(cResScaleInv, 1, tu.cu->cs->slice->clpRng(compID));
  }

  double diagRatio = 0, horVerRatio = 0;

  if( trModes )
  {
    m_pcTrQuant->transformNxN( tu, compID, cQP, trModes, CU::isIntra( *tu.cu ) ? m_pcEncCfg->getIntraMTSMaxCand() : m_pcEncCfg->getInterMTSMaxCand(), ispSplitIsAllowed ? &diagRatio : nullptr, ispSplitIsAllowed ? &horVerRatio : nullptr );
    tu.mtsIdx = trModes->at(0).first;
  }
  m_pcTrQuant->transformNxN( tu, compID, cQP, uiAbsSum, m_CABACEstimator->getCtx(), loadTr, &diagRatio, &horVerRatio );
  if (!tu.cu->ispMode && isLuma(compID) && ispSplitIsAllowed &&
    tu.mtsIdx == 0
    )
  {
    m_intraModeDiagRatio        .push_back(diagRatio);
    m_intraModeHorVerRatio      .push_back(horVerRatio);
    m_intraModeTestedNormalIntra.push_back((int)uiChFinalMode);
  }


  DTRACE( g_trace_ctx, D_TU_ABS_SUM, "%d: comp=%d, abssum=%d\n", DTRACE_GET_COUNTER( g_trace_ctx, D_TU_ABS_SUM ), compID, uiAbsSum );


  //--- inverse transform ---
  if (uiAbsSum > 0)
  {
    m_pcTrQuant->invTransformNxN(tu, compID, piResi, cQP);
  }
  else
  {
    piResi.fill(0);
  }

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

  if (slice.getReshapeInfo().getUseSliceReshaper() && 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 ));

  //===== update distortion =====
#if WCG_EXT
  if (m_pcEncCfg->getLumaLevelToDeltaQPMapping().isEnabled() || (m_pcEncCfg->getReshaper()
    && slice.getReshapeInfo().getUseSliceReshaper() && (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);
  }
  else
#endif
  {
    ruiDist += m_pcRdCost->getDistPart( piOrg, piReco, bitDepth, compID, DF_SSE );
  }
}

void IntraSearch::xRecurIntraCodingLumaQT( CodingStructure &cs, Partitioner &partitioner, const double bestCostSoFar, const int subTuIdx, const PartSplit ispType, const bool ispIsCurrentWinnder )
{
        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 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;
  int        bestModeId[MAX_NUM_COMPONENT]      = { 0, 0, 0 };

  const 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;
  }

  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 = TU::isMTSAllowed( tu, COMPONENT_Y );
    uint8_t nNumTransformCands = 1 + ( tsAllowed ? 1 : 0 ) + ( mtsAllowed ? 4 : 0 ); // DCT + TS + 4 MTS = 6 tests
    std::vector<TrMode> trModes;
    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) );
      }
    }

    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      = 0;

    int       lastCheckId        = trModes[nNumTransformCands-1].first;
    bool isNotOnlyOneMode        = 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    cbfDCT2  = true;

    double bestDCT2cost = MAX_DOUBLE;
    double threshold = m_pcEncCfg->getUseFastISP() && !cu.ispMode && ispIsCurrentWinnder && nNumTransformCands > 1 ? 1 + 1.4 / sqrt( cu.lwidth() * cu.lheight() ) : 1;
    for( int modeId = firstCheckId; modeId < nNumTransformCands; modeId++ )
    {
      if( !cbfDCT2 || ( m_pcEncCfg->getUseTransformSkipFast() && bestModeId[COMPONENT_Y] == 1 ) )
      {
        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 && ispIsCurrentWinnder && trModes[modeId].first != 0 && ( trModes[modeId].first != 1 || !tsAllowed ) && bestDCT2cost > bestCostSoFar * threshold )
      {
        continue;
      }
      tu.mtsIdx = trModes[modeId].first;

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

      int default0Save1Load2 = 0;
      singleDistTmpLuma = 0;

      if( modeId == firstCheckId && nNumTransformCands > 1 )
      {
        default0Save1Load2 = 1;
      }
      else if (modeId != firstCheckId)
      {
        default0Save1Load2 = 2;
      }
      if( cu.ispMode )
      {
        default0Save1Load2 = 0;
      }
      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( ( 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;

        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( 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;
  }

  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
    {
      xRecurIntraCodingLumaQT( *csSplit, partitioner, bestCostSoFar, subTuCounter, ispType );
      subTuCounter += subTuCounter != -1 ? 1 : 0;

      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() >> g_aucLog2[cu.firstTU->lheight()] : cu.lwidth() >> g_aucLog2[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;