IntraSearch.cpp

  }
  assert(run >= 1);
  double costPerPixel = (double)m_CABACEstimator->getEstFracBits() / (double)run;
  return costPerPixel;
}
void IntraSearch::preCalcPLTIndex(CodingStructure& cs, Partitioner& partitioner, ComponentID compBegin, uint32_t numComp)
{
  CodingUnit &cu = *cs.getCU(partitioner.chType);
  TransformUnit &tu = *cs.getTU(partitioner.chType);
  const int  channelBitDepth_L = cs.sps->getBitDepth(CHANNEL_TYPE_LUMA);
  const int  channelBitDepth_C = cs.sps->getBitDepth(CHANNEL_TYPE_CHROMA);
  const int  pcmShiftRight_L = (channelBitDepth_L - PLT_ENCBITDEPTH);
  const int  pcmShiftRight_C = (channelBitDepth_C - PLT_ENCBITDEPTH);

  uint32_t height = cu.block(compBegin).height;
  uint32_t width = cu.block(compBegin).width;

  CPelBuf   orgBuf[3];
  for (int comp = compBegin; comp < (compBegin + numComp); comp++)
  {
    CompArea  area = cu.blocks[comp];
    if (m_pcEncCfg->getReshaper() && (cs.slice->getLmcsEnabledFlag() && m_pcReshape->getCTUFlag()))
    {
      orgBuf[comp] = cs.getPredBuf(area);
    }
    else
    {
      orgBuf[comp] = cs.getOrgBuf(area);
    }
  }

  PelBuf   curPLTIdx = tu.getcurPLTIdx(compBegin);
  int      errorLimit = numComp * g_paletteQuant[cu.qp];
  uint32_t bestIdx = 0;
  uint32_t scaleX = getComponentScaleX(COMPONENT_Cb, cs.sps->getChromaFormatIdc());
  uint32_t scaleY = getComponentScaleY(COMPONENT_Cb, cs.sps->getChromaFormatIdc());
  for (uint32_t y = 0; y < height; y++)
  {
    for (uint32_t x = 0; x < width; x++)
    {
      uint32_t pltIdx = 0;
      uint32_t minError = MAX_UINT;
      while (pltIdx < cu.curPLTSize[compBegin])
      {
        uint32_t absError = 0, pX, pY;
        for (int comp = compBegin; comp < (compBegin + numComp); comp++)
        {
          pX = (comp > 0 && compBegin == COMPONENT_Y) ? (x >> scaleX) : x;
          pY = (comp > 0 && compBegin == COMPONENT_Y) ? (y >> scaleY) : y;
          int shift = (comp > 0) ? pcmShiftRight_C : pcmShiftRight_L;
          absError += abs(cu.curPLT[comp][pltIdx] - orgBuf[comp].at(pX, pY)) >> shift;
        }

        if (absError < minError)
        {
          bestIdx = pltIdx;
          minError = absError;
          if (minError == 0)
          {
            break;
          }
        }
        pltIdx++;
      }
      curPLTIdx.at(x, y) = bestIdx;
      if (minError > errorLimit)
      {
        curPLTIdx.at(x, y) = cu.curPLTSize[compBegin];
        cu.useEscape[compBegin] = true;
        calcPixelPred(cs, partitioner, y, x, compBegin, numComp);
      }
    }
  }
}
void IntraSearch::calcPixelPred(CodingStructure& cs, Partitioner& partitioner, uint32_t yPos, uint32_t xPos, ComponentID compBegin, uint32_t numComp)
{
  CodingUnit    &cu = *cs.getCU(partitioner.chType);
  TransformUnit &tu = *cs.getTU(partitioner.chType);

  CPelBuf   orgBuf[3];
  for (int comp = compBegin; comp < (compBegin + numComp); comp++)
  {
    CompArea  area = cu.blocks[comp];
    if (m_pcEncCfg->getReshaper() && (cs.slice->getLmcsEnabledFlag() && m_pcReshape->getCTUFlag()))
    {
      orgBuf[comp] = cs.getPredBuf(area);
    }
    else
    {
      orgBuf[comp] = cs.getOrgBuf(area);
    }
  }

  int qp[3];
  int qpRem[3];
  int qpPer[3];
  int quantiserScale[3];
  int quantiserRightShift[3];
  int rightShiftOffset[3];
  int InvquantiserRightShift[3];
  int add[3];
  for (uint32_t ch = compBegin; ch < (compBegin + numComp); ch++)
  {
    QpParam cQP(tu, ComponentID(ch));
#if JVET_O0919_TS_MIN_QP
    qp[ch] = cQP.Qp(false);
#else
    qp[ch] = cQP.Qp;
#endif
    qpRem[ch] = qp[ch] % 6;
    qpPer[ch] = qp[ch] / 6;
    quantiserScale[ch] = g_quantScales[0][qpRem[ch]];
    quantiserRightShift[ch] = QUANT_SHIFT + qpPer[ch];
    rightShiftOffset[ch] = 1 << (quantiserRightShift[ch] - 1);
    InvquantiserRightShift[ch] = IQUANT_SHIFT;
    add[ch] = 1 << (InvquantiserRightShift[ch] - 1);
  }

  uint32_t scaleX = getComponentScaleX(COMPONENT_Cb, cs.sps->getChromaFormatIdc());
  uint32_t scaleY = getComponentScaleY(COMPONENT_Cb, cs.sps->getChromaFormatIdc());
  for (uint32_t ch = compBegin; ch < (compBegin + numComp); ch++)
  {
    const int channelBitDepth = cu.cs->sps->getBitDepth(toChannelType((ComponentID)ch));
    CompArea  area = cu.blocks[ch];
    PelBuf    recBuf = cs.getRecoBuf(area);
    PLTescapeBuf escapeValue = tu.getescapeValue((ComponentID)ch);
    if (compBegin != COMPONENT_Y || ch == 0)
    {
      escapeValue.at(xPos, yPos) = TCoeff(std::max<int>(0, ((orgBuf[ch].at(xPos, yPos) * quantiserScale[ch] + rightShiftOffset[ch]) >> quantiserRightShift[ch])));
      assert(escapeValue.at(xPos, yPos) < (1 << (channelBitDepth + 1)));
      recBuf.at(xPos, yPos) = (((escapeValue.at(xPos, yPos)*g_invQuantScales[0][qpRem[ch]]) << qpPer[ch]) + add[ch]) >> InvquantiserRightShift[ch];
      recBuf.at(xPos, yPos) = Pel(ClipBD<int>(recBuf.at(xPos, yPos), channelBitDepth));//to be checked
    }
    else if (compBegin == COMPONENT_Y && ch > 0 && yPos % (1 << scaleY) == 0 && xPos % (1 << scaleX) == 0)
    {
      uint32_t yPosC = yPos >> scaleY;
      uint32_t xPosC = xPos >> scaleX;
      escapeValue.at(xPosC, yPosC) = TCoeff(std::max<int>(0, ((orgBuf[ch].at(xPosC, yPosC) * quantiserScale[ch] + rightShiftOffset[ch]) >> quantiserRightShift[ch])));
      assert(escapeValue.at(xPosC, yPosC) < (1 << (channelBitDepth + 1)));
      recBuf.at(xPosC, yPosC) = (((escapeValue.at(xPosC, yPosC)*g_invQuantScales[0][qpRem[ch]]) << qpPer[ch]) + add[ch]) >> InvquantiserRightShift[ch];
      recBuf.at(xPosC, yPosC) = Pel(ClipBD<int>(recBuf.at(xPosC, yPosC), channelBitDepth));//to be checked
    }
  }
}
void IntraSearch::derivePLTLossy(CodingStructure& cs, Partitioner& partitioner, ComponentID compBegin, uint32_t numComp)
{
  CodingUnit &cu = *cs.getCU(partitioner.chType);
  const int channelBitDepth_L = cs.sps->getBitDepth(CHANNEL_TYPE_LUMA);
  const int channelBitDepth_C = cs.sps->getBitDepth(CHANNEL_TYPE_CHROMA);
  const int pcmShiftRight_L = (channelBitDepth_L - PLT_ENCBITDEPTH);
  const int pcmShiftRight_C = (channelBitDepth_C - PLT_ENCBITDEPTH);

  uint32_t height = cu.block(compBegin).height;
  uint32_t width = cu.block(compBegin).width;

  CPelBuf   orgBuf[3];
  for (int comp = compBegin; comp < (compBegin + numComp); comp++)
  {
    CompArea  area = cu.blocks[comp];
    if (m_pcEncCfg->getReshaper() && (cs.slice->getLmcsEnabledFlag() && m_pcReshape->getCTUFlag()))
    {
      orgBuf[comp] = cs.getPredBuf(area);
    }
    else
    {
      orgBuf[comp] = cs.getOrgBuf(area);
    }
  }

  int errorLimit = g_paletteQuant[cu.qp];
  uint32_t totalSize = height*width;
  SortingElement *pelList = new SortingElement[totalSize];
  SortingElement  element;
  SortingElement *pelListSort = new SortingElement[MAXPLTSIZE + 1];
  uint32_t dictMaxSize = MAXPLTSIZE;
  uint32_t idx = 0;
  int last = -1;

  uint32_t scaleX = getComponentScaleX(COMPONENT_Cb, cs.sps->getChromaFormatIdc());
  uint32_t scaleY = getComponentScaleY(COMPONENT_Cb, cs.sps->getChromaFormatIdc());
  for (uint32_t y = 0; y < height; y++)
  {
    for (uint32_t x = 0; x < width; x++)
    {
      uint32_t org[3], pX, pY;
      for (int comp = compBegin; comp < (compBegin + numComp); comp++)
      {
        pX = (comp > 0 && compBegin == COMPONENT_Y) ? (x >> scaleX) : x;
        pY = (comp > 0 && compBegin == COMPONENT_Y) ? (y >> scaleY) : y;
        org[comp] = orgBuf[comp].at(pX, pY);
      }
      element.setAll(org, compBegin, numComp);
      int besti = last, bestSAD = (last == -1) ? MAX_UINT : pelList[last].getSAD(element, cs.sps->getBitDepths(), compBegin, numComp);
      if (bestSAD)
      {
        for (int i = idx - 1; i >= 0; i--)
        {
          uint32_t sad = pelList[i].getSAD(element, cs.sps->getBitDepths(), compBegin, numComp);
          if (sad < bestSAD)
          {
            bestSAD = sad;
            besti = i;
            if (!sad) break;
          }
        }
      }
      if (besti >= 0 && pelList[besti].almostEqualData(element, errorLimit, cs.sps->getBitDepths(), compBegin, numComp))
      {
        pelList[besti].addElement(element, compBegin, numComp);
        last = besti;
      }
      else
      {
        pelList[idx].copyDataFrom(element, compBegin, numComp);
        pelList[idx].setCnt(1);
        last = idx;
        idx++;
      }
    }
  }

  for (int i = 0; i < dictMaxSize; i++)
  {
    pelListSort[i].setCnt(0);
    pelListSort[i].resetAll(compBegin, numComp);
  }

  //bubble sorting
  dictMaxSize = 1;
  for (int i = 0; i < idx; i++)
  {
    if (pelList[i].getCnt() > pelListSort[dictMaxSize - 1].getCnt())
    {
      int j;
      for (j = dictMaxSize; j > 0; j--)
      {
        if (pelList[i].getCnt() > pelListSort[j - 1].getCnt() )
        {
          pelListSort[j].copyAllFrom(pelListSort[j - 1], compBegin, numComp);
          dictMaxSize = std::min(dictMaxSize + 1, (uint32_t)MAXPLTSIZE);
        }
        else
        {
          break;
        }
      }
      pelListSort[j].copyAllFrom(pelList[i], compBegin, numComp);
    }
  }

  uint32_t paletteSize = 0;
  uint64_t numColorBits = 0;
  for (int comp = compBegin; comp < (compBegin + numComp); comp++)
  {
    numColorBits += (comp > 0) ? channelBitDepth_C : channelBitDepth_L;
  }

  double bitCost = m_pcRdCost->getLambda()*numColorBits;
  for (int i = 0; i < MAXPLTSIZE; i++)
  {
    if (pelListSort[i].getCnt())
    {
      int half = pelListSort[i].getCnt() >> 1;
      for (int comp = compBegin; comp < (compBegin + numComp); comp++)
      {
        cu.curPLT[comp][paletteSize] = (pelListSort[i].getSumData(comp) + half) / pelListSort[i].getCnt();
      }

      int best = -1;
      if (errorLimit)
      {
        double pal[MAX_NUM_COMPONENT], err = 0.0, bestCost = 0.0;
        for (int comp = compBegin; comp < (compBegin + numComp); comp++)
        {
          const int shift = (comp > 0) ? pcmShiftRight_C : pcmShiftRight_L;
          pal[comp] = pelListSort[i].getSumData(comp) / (double)pelListSort[i].getCnt();
          err = pal[comp] - cu.curPLT[comp][paletteSize];
          bestCost += (err*err) / (1 << (2 * shift));
        }
        bestCost = bestCost * pelListSort[i].getCnt() + bitCost;

        for (int t = 0; t < cs.prevPLT.curPLTSize[compBegin]; t++)
        {
          double cost = 0.0;
          for (int comp = compBegin; comp < (compBegin + numComp); comp++)
          {
            const int shift = (comp > 0) ? pcmShiftRight_C : pcmShiftRight_L;
            err = pal[comp] - cs.prevPLT.curPLT[comp][t];
            cost += (err*err) / (1 << (2 * shift));
          }
          cost *= pelListSort[i].getCnt();
          if (cost < bestCost)
          {
            best = t;
            bestCost = cost;
          }
        }
        if (best != -1)
        {
          for (int comp = compBegin; comp < (compBegin + numComp); comp++)
          {
            cu.curPLT[comp][paletteSize] = cs.prevPLT.curPLT[comp][best];
          }
        }
      }

      bool duplicate = false;
      if (pelListSort[i].getCnt() == 1 && best == -1)
      {
        duplicate = true;
      }
      else
      {
        for (int t = 0; t<paletteSize; t++)
        {
          bool duplicateTmp = true;
          for (int comp = compBegin; comp < (compBegin + numComp); comp++)
          {
            duplicateTmp = duplicateTmp && (cu.curPLT[comp][paletteSize] == cu.curPLT[comp][t]);
          }
          if (duplicateTmp)
          {
            duplicate = true;
            break;
          }
        }
      }
      if (!duplicate) paletteSize++;
    }
    else
    {
      break;
    }
  }
  cu.curPLTSize[compBegin] = paletteSize;

  delete[] pelList;
  delete[] pelListSort;
}
#endif
// -------------------------------------------------------------------------------------------------------------------
// 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 JVET_O0119_BASE_PALETTE_444
      if ((!cs.slice->isIntra() || cs.slice->getSPS()->getIBCFlag() || cs.slice->getSPS()->getPLTMode())
      && cu.Y().valid()
      )
#else
      if ((!cs.slice->isIntra() || cs.slice->getSPS()->getIBCFlag())
        && cu.Y().valid()
        )
#endif
      {
        if( cs.pps->getTransquantBypassEnabledFlag() )
        {
          m_CABACEstimator->cu_transquant_bypass_flag( cu );
        }
        m_CABACEstimator->cu_skip_flag( cu );
        m_CABACEstimator->pred_mode   ( cu );
      }
#if JVET_O0119_BASE_PALETTE_444
      if (CU::isPLT(cu))
      {
        return;
      }
#endif
      m_CABACEstimator->bdpcm_mode  ( cu, ComponentID(partitioner.chType) );
#if !JVET_O0525_REMOVE_PCM
      if( CU::isIntra(cu) )
      {
        m_CABACEstimator->pcm_data( cu, partitioner );
        if( cu.ipcm )
        {
          return;
        }
      }
#endif
    }

    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() >> floorLog2(currTU.lheight()) : currCU.lwidth() >> floorLog2(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 JVET_O0105_ICT
    if( compID == COMPONENT_Cr )
    {
      const int cbfMask = ( TU::getCbf( currTU, COMPONENT_Cb ) ? 2 : 0 ) + ( TU::getCbf( currTU, COMPONENT_Cr ) ? 1 : 0 );
      m_CABACEstimator->joint_cb_cr( currTU, cbfMask );
    }
#endif
    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
#if JVET_O0105_ICT
  const bool prevCbf = ( compID == COMPONENT_Cr ? TU::getCbfAtDepth( currTU, COMPONENT_Cb, partitioner.currTrDepth ) : false );
  m_CABACEstimator->cbf_comp( cs, TU::getCbfAtDepth( currTU, compID, partitioner.currTrDepth ), currArea.blocks[compID], partitioner.currTrDepth - 1, prevCbf );
#else
  m_CABACEstimator->cbf_comp( cs, TU::getCbfAtDepth( currTU, compID, partitioner.currTrDepth ), currArea.blocks[compID], partitioner.currTrDepth - 1 );
#endif
  //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 );
  }

  // Include Cbf and jointCbCr flags here as we make decisions across components
  CodingStructure &cs = *currTU.cs;

  if ( currTU.jointCbCr )
  {
#if JVET_O0105_ICT
    const int cbfMask = ( TU::getCbf( currTU, COMPONENT_Cb ) ? 2 : 0 ) + ( TU::getCbf( currTU, COMPONENT_Cr ) ? 1 : 0 );
    m_CABACEstimator->cbf_comp( cs, cbfMask>>1, currTU.blocks[ COMPONENT_Cb ], currTU.depth, false );
    m_CABACEstimator->cbf_comp( cs, cbfMask &1, currTU.blocks[ COMPONENT_Cr ], currTU.depth, cbfMask>>1 );
    if( cbfMask )
      m_CABACEstimator->joint_cb_cr( currTU, cbfMask );
    if( cbfMask >> 1 )
      m_CABACEstimator->residual_coding( currTU, COMPONENT_Cb );
    if( cbfMask & 1 )
      m_CABACEstimator->residual_coding( currTU, COMPONENT_Cr );
#else
    if ( TU::getCbf( currTU, COMPONENT_Cb ) )
    {
      m_CABACEstimator->cbf_comp( cs, true, currTU.blocks[ COMPONENT_Cb ], currTU.depth, false );
      m_CABACEstimator->cbf_comp( cs, true, currTU.blocks[ COMPONENT_Cr ], currTU.depth, true );
      m_CABACEstimator->joint_cb_cr( currTU );
    }
    else
    {
      m_CABACEstimator->cbf_comp( cs, false, currTU.blocks[ COMPONENT_Cb ], currTU.depth, false );
      m_CABACEstimator->cbf_comp( cs, false, currTU.blocks[ COMPONENT_Cr ], currTU.depth, false );
    }
#endif
  }
  else
  {
    if ( compID == COMPONENT_Cb )
      m_CABACEstimator->cbf_comp( cs, TU::getCbf( currTU, compID ), currTU.blocks[ compID ], currTU.depth, false );
    else
#if JVET_O0105_ICT
    {
      const bool cbCbf    = TU::getCbf( currTU, COMPONENT_Cb );
      const bool crCbf    = TU::getCbf( currTU, compID );
      const int  cbfMask  = ( cbCbf ? 2 : 0 ) + ( crCbf ? 1 : 0 );
      m_CABACEstimator->cbf_comp( cs, crCbf, currTU.blocks[ compID ], currTU.depth, cbCbf );
      m_CABACEstimator->joint_cb_cr( currTU, cbfMask );
    }
#else
      m_CABACEstimator->cbf_comp( cs, TU::getCbf( currTU, compID ), currTU.blocks[ compID ], currTU.depth, TU::getCbf( currTU, COMPONENT_Cb ) );
#endif
  }

#if JVET_O0105_ICT
  if( !currTU.jointCbCr && TU::getCbf( currTU, compID ) )
#else
  if( TU::getCbf( currTU, compID ) )
#endif
  {
    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;
  m_pcRdCost->setChromaFormat(cs.sps->getChromaFormatIdc());

  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();
#if !JVET_O0502_ISP_CLEANUP
  const bool           ispSplitIsAllowed    = sps.getUseISP() && CU::canUseISP( *tu.cu, compID );
#endif


  //===== init availability pattern =====
#if JVET_O0105_ICT
  CHECK( tu.jointCbCr && compID == COMPONENT_Cr, "wrong combination of compID and jointCbCr" );
#endif
  bool jointCbCr = tu.jointCbCr && compID == COMPONENT_Cb;

  if ( compID == COMPONENT_Y )
  {
  PelBuf sharedPredTS( m_pSharedPredTransformSkip[compID], area );
  if( default0Save1Load2 != 2 )
  {
#if JVET_O0502_ISP_CLEANUP
#if JVET_O0106_ISP_4xN_PREDREG_FOR_1xN_2xN
    bool predRegDiffFromTB = CU::isPredRegDiffFromTB(*tu.cu, compID);
    bool firstTBInPredReg = CU::isFirstTBInPredReg(*tu.cu, compID, area);
    CompArea areaPredReg(COMPONENT_Y, tu.chromaFormat, area);
#endif
    if (tu.cu->ispMode && isLuma(compID))
    {
#if JVET_O0106_ISP_4xN_PREDREG_FOR_1xN_2xN
      if (predRegDiffFromTB)
      {
        if (firstTBInPredReg)
        {
          CU::adjustPredArea(areaPredReg);
          initIntraPatternChTypeISP(*tu.cu, areaPredReg, piReco);
        }
      }
      else
#endif
        initIntraPatternChTypeISP(*tu.cu, area, piReco);
    }
    else
    {
      initIntraPatternChType(*tu.cu, area);
    }
#else
#if JVET_O0106_ISP_4xN_PREDREG_FOR_1xN_2xN
    bool predRegDiffFromTB = CU::isPredRegDiffFromTB(*tu.cu, compID);
    bool firstTBInPredReg = CU::isFirstTBInPredReg(*tu.cu, compID, area);
    CompArea areaPredReg(COMPONENT_Y, tu.chromaFormat, area);
    if (predRegDiffFromTB)
    {
      if (firstTBInPredReg)
      {
        CU::adjustPredArea(areaPredReg);
        initIntraPatternChType(*tu.cu, areaPredReg);
      }
    }
    else
#endif
      initIntraPatternChType(*tu.cu, area);
#endif

    //===== get prediction signal =====
    if( compID != COMPONENT_Y && PU::isLMCMode( uiChFinalMode ) )
    {
      {
        xGetLumaRecPixels( pu, area );
      }
      predIntraChromaLM( compID, piPred, pu, area, uiChFinalMode );
    }
    else
    {
      if( PU::isMIP( pu, chType ) )
      {
        predIntraMip( compID, piPred, pu );
      }
      else
      {
#if JVET_O0106_ISP_4xN_PREDREG_FOR_1xN_2xN
        if (predRegDiffFromTB)
        {
          if (firstTBInPredReg)
          {
            PelBuf piPredReg = cs.getPredBuf(areaPredReg);
            predIntraAng(compID, piPredReg, pu);
          }
        }
        else
#endif
          predIntraAng(compID, piPred, pu);
      }
    }


    // 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.getLmcsEnabledFlag() && (slice.isIntra() || (!slice.isIntra() && m_pcReshape->getCTUFlag()));
#if JVET_O0105_ICT
  if (isLuma(compID))
  {
#else
  if (flag && slice.getLmcsChromaResidualScaleFlag() && 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 );
#if JVET_O1109_UNFIY_CRS
    int adj = m_pcReshape->calculateChromaAdjVpduNei(tu, areaY);
#else
    PelBuf piPredY;
    piPredY = cs.picture->getPredBuf(areaY);
    const Pel avgLuma = piPredY.computeAvg();
    int adj = m_pcReshape->calculateChromaAdj(avgLuma);
#endif
    tu.setChromaAdj(adj);
  }
#endif
  //===== get residual signal =====
  piResi.copyFrom( piOrg  );
  if (slice.getLmcsEnabledFlag() && m_pcReshape->getCTUFlag() && compID == COMPONENT_Y)
  {
    CompArea      tmpArea(COMPONENT_Y, area.chromaFormat, Position(0, 0), area.size());
    PelBuf tmpPred = m_tmpStorageLCU.getBuf(tmpArea);
    tmpPred.copyFrom(piPred);
    piResi.rspSignal(m_pcReshape->getFwdLUT());
    piResi.subtract(tmpPred);
  }
  else
  piResi.subtract( piPred );

  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);
  }
#if JVET_O0105_ICT
  }
#endif

  //===== 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.getLmcsChromaResidualScaleFlag() )
  {
    int cResScaleInv = tu.getChromaAdj();
#if JVET_O0429_CRS_LAMBDA_FIX
    double cResScale = (double)(1 << CSCALE_FP_PREC) / (double)cResScaleInv;
#else
    double cResScale = round((double)(1 << CSCALE_FP_PREC) / (double)cResScaleInv);
#endif
    m_pcTrQuant->setLambda(m_pcTrQuant->getLambda() / (cResScale*cResScale));
#if !JVET_O0105_ICT
    if ( !jointCbCr ) // Joint CbCr signal is to be scaled in the case of joint chroma
    piResi.scaleSignal(cResScaleInv, 1, tu.cu->cs->slice->clpRng(compID));
#endif
  }

  const CompArea &crArea = tu.blocks     [ COMPONENT_Cr ];
  PelBuf          crOrg  = cs.getOrgBuf  ( crArea );
  PelBuf          crPred = cs.getPredBuf ( crArea );
  PelBuf          crResi = cs.getResiBuf ( crArea );
  PelBuf          crReco = cs.getRecoBuf ( crArea );

  if ( jointCbCr )
  {
#if !JVET_O0105_ICT
    // Get Cr prediction and residual
    crResi.copyFrom( crOrg  );
    crResi.subtract( crPred );

    // Create joint residual and store it for Cb component: jointResi = (cbResi - crResi)/2
    piResi.subtractAndHalve( crResi );

    // Scale the joint signal
    if ( flag && slice.getLmcsChromaResidualScaleFlag() )
      piResi.scaleSignal(tu.getChromaAdj(), 1, tu.cu->cs->slice->clpRng(compID));
#endif
    // Lambda is loosened for the joint mode with respect to single modes as the same residual is used for both chroma blocks
#if JVET_O0105_ICT
    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() );
#else
    m_pcTrQuant->setLambda( 0.60 * m_pcTrQuant->getLambda() );
#endif
  }
#if JVET_O0105_ICT
#if JVET_O0376_SPS_JOINTCBCR_FLAG
  if ( sps.getJointCbCrEnabledFlag() && isChroma(compID) && (tu.cu->cs->slice->getSliceQp() > 18) )
  {
    m_pcTrQuant->setLambda( 1.3 * m_pcTrQuant->getLambda() );
  }
#else
  if( isChroma(compID) && tu.cu->cs->slice->getSliceQp() > 18 )
  {
    m_pcTrQuant->setLambda( 1.3 * m_pcTrQuant->getLambda() );
  }
#endif
#else
#if JVET_O0376_SPS_JOINTCBCR_FLAG
  else if ( sps.getJointCbCrEnabledFlag() && isChroma(compID) && (tu.cu->cs->slice->getSliceQp() > 18) )
#else
  else if ( isChroma(compID) && tu.cu->cs->slice->getSliceQp() > 18 )
#endif
    m_pcTrQuant->setLambda( 1.10 * m_pcTrQuant->getLambda() );
#endif

#if JVET_O0105_ICT
  if( isLuma(compID) )
  {
#endif
#if JVET_O0502_ISP_CLEANUP
    if (trModes)
    {
      m_pcTrQuant->transformNxN(tu, compID, cQP, trModes, CU::isIntra(*tu.cu) ? m_pcEncCfg->getIntraMTSMaxCand() : m_pcEncCfg->getInterMTSMaxCand());
      tu.mtsIdx = trModes->at(0).first;
    }
    m_pcTrQuant->transformNxN(tu, compID, cQP, uiAbsSum, m_CABACEstimator->getCtx(), loadTr);
#else
  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 == MTS_DCT2_DCT2 && ispSplitIsAllowed )
  {
    m_intraModeDiagRatio        .push_back(diagRatio);
    m_intraModeHorVerRatio      .push_back(horVerRatio);
    m_intraModeTestedNormalIntra.push_back((int)uiChFinalMode);
  }
#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 ), compID, uiAbsSum );

#if JVET_O0502_ISP_CLEANUP
  if (tu.cu->ispMode && isLuma(compID) && CU::isISPLast(*tu.cu, area, area.compID) && CU::allLumaCBFsAreZero(*tu.cu))
  {
    // ISP has to have at least one non-zero CBF
    ruiDist = MAX_INT;
    return;
  }
#endif


  //--- inverse transform ---
  if (uiAbsSum > 0)