IntraSearch.cpp

              bestSAD = sad;
              besti   = i;
              break;
            }
          }
        }
      }
      else
      {
        if (bestSAD)
        {
          for (int i = idx - 1; i >= 0; i--)
          {
            uint32_t sad = pelList[i].getSAD(element, cs.sps->getBitDepths(), tmpCompBegin, tmpNumComp, lossless);
            if (sad < bestSAD)
            {
              bestSAD = sad;
              besti   = i;
              if (!sad)
              {
                break;
              }
            }
          }
        }
      }
      if (besti >= 0 && pelList[besti].almostEqualData(element, errorLimit, cs.sps->getBitDepths(), tmpCompBegin, tmpNumComp, lossless))
      {
        pelList[besti].addElement(element, tmpCompBegin, tmpNumComp);
        last = besti;
      }
      else
      {
        pelList[idx].copyDataFrom(element, tmpCompBegin, tmpNumComp);
        for (int comp = tmpCompBegin; comp < (tmpCompBegin + tmpNumComp); comp++)
        {
          pelList[idx].setCnt(1, comp);
        }
        last = idx;
        idx++;
      }
    }
  }

  if( cs.sps->getChromaFormatIdc() != CHROMA_444 && numComp == 3 )
  {
    for( int i = 0; i < idx; i++ )
    {
      pelList[i].setCnt( pelList[i].getCnt(COMPONENT_Y) + (pelList[i].getCnt(COMPONENT_Cb) >> 2), MAX_NUM_COMPONENT);
    }
  }
  else
  {
    if( compBegin == 0 )
    {
      for( int i = 0; i < idx; i++ )
      {
        pelList[i].setCnt(pelList[i].getCnt(COMPONENT_Y), COMPONENT_Cb);
        pelList[i].setCnt(pelList[i].getCnt(COMPONENT_Y), COMPONENT_Cr);
        pelList[i].setCnt(pelList[i].getCnt(COMPONENT_Y), MAX_NUM_COMPONENT);
      }
    }
    else
    {
      for( int i = 0; i < idx; i++ )
      {
        pelList[i].setCnt(pelList[i].getCnt(COMPONENT_Cb), COMPONENT_Y);
        pelList[i].setCnt(pelList[i].getCnt(COMPONENT_Cb), MAX_NUM_COMPONENT);
      }
    }
  }

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

  //bubble sorting
  dictMaxSize = 1;
  for (int i = 0; i < idx; i++)
  {
    if( pelList[i].getCnt(MAX_NUM_COMPONENT) > pelListSort[dictMaxSize - 1].getCnt(MAX_NUM_COMPONENT) )
    {
      int j;
      for (j = dictMaxSize; j > 0; j--)
      {
        if (pelList[i].getCnt(MAX_NUM_COMPONENT) > pelListSort[j - 1].getCnt(MAX_NUM_COMPONENT))
        {
          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;
  }
  const int plt_lambda_shift = (compBegin > 0) ? pcmShiftRight_C : pcmShiftRight_L;
  double    bitCost          = m_pcRdCost->getLambda() / (double) (1 << (2 * plt_lambda_shift)) * numColorBits;
  bool   reuseflag[MAXPLTPREDSIZE] = { false };
  int    run;
  double reuseflagCost;
  for (int i = 0; i < maxPltSize; i++)
  {
    if( pelListSort[i].getCnt(MAX_NUM_COMPONENT) )
    {
      ComponentID tmpCompBegin = compBegin;
      int tmpNumComp = numComp;
      if( cs.sps->getChromaFormatIdc() != CHROMA_444 && numComp == 3 && pelListSort[i].getCnt(COMPONENT_Cb) == 0 )
      {
        tmpCompBegin = COMPONENT_Y;
        tmpNumComp   = 1;
      }

      for( int comp = tmpCompBegin; comp < (tmpCompBegin + tmpNumComp); comp++ )
      {
        int half = pelListSort[i].getCnt(comp) >> 1;
        cu.curPLT[comp][paletteSize] = (pelListSort[i].getSumData(comp) + half) / pelListSort[i].getCnt(comp);
      }

      int best = -1;
      if( errorLimit )
      {
        double pal[MAX_NUM_COMPONENT], err = 0.0, bestCost = 0.0;
        for( int comp = tmpCompBegin; comp < (tmpCompBegin + tmpNumComp); comp++ )
        {
          pal[comp] = pelListSort[i].getSumData(comp) / (double)pelListSort[i].getCnt(comp);
          err = pal[comp] - cu.curPLT[comp][paletteSize];
          if( isChroma((ComponentID) comp) )
          {
            bestCost += (err * err * PLT_CHROMA_WEIGHTING) / (1 << (2 * pcmShiftRight_C)) * pelListSort[i].getCnt(comp);
          }
          else
          {
            bestCost += (err * err) / (1 << (2 * pcmShiftRight_L)) * pelListSort[i].getCnt(comp);
          }
        }
        bestCost += bitCost;

        for( int t = 0; t < cs.prevPLT.curPLTSize[compBegin]; t++ )
        {
          double cost = 0.0;
          for( int comp = tmpCompBegin; comp < (tmpCompBegin + tmpNumComp); comp++ )
          {
            err = pal[comp] - cs.prevPLT.curPLT[comp][t];
            if( isChroma((ComponentID) comp) )
            {
              cost += (err * err * PLT_CHROMA_WEIGHTING) / (1 << (2 * pcmShiftRight_C)) * pelListSort[i].getCnt(comp);
            }
            else
            {
              cost += (err * err) / (1 << (2 * pcmShiftRight_L)) * pelListSort[i].getCnt(comp);
            }
          }
          run = 0;
          for (int t2 = t; t2 >= 0; t2--)
          {
            if (!reuseflag[t2])
            {
              run++;
            }
            else
            {
              break;
            }
          }
          reuseflagCost = m_pcRdCost->getLambda() / (double)(1 << (2 * plt_lambda_shift)) * getEpExGolombNumBins(run ? run + 1 : run, 0);
          cost += reuseflagCost;

          if( cost < bestCost )
          {
            best = t;
            bestCost = cost;
          }
        }
        if( best != -1 )
        {
          for( int comp = tmpCompBegin; comp < (tmpCompBegin + tmpNumComp); comp++ )
          {
            cu.curPLT[comp][paletteSize] = cs.prevPLT.curPLT[comp][best];
          }
          reuseflag[best] = true;
        }
      }

      bool duplicate = false;
      if( pelListSort[i].getCnt(MAX_NUM_COMPONENT) == 1 && best == -1 )
      {
        duplicate = true;
      }
      else
      {
        for( int t = 0; t < paletteSize; t++ )
        {
          bool duplicateTmp = true;
          for( int comp = tmpCompBegin; comp < (tmpCompBegin + tmpNumComp); comp++ )
          {
            duplicateTmp = duplicateTmp && (cu.curPLT[comp][paletteSize] == cu.curPLT[comp][t]);
          }
          if( duplicateTmp )
          {
            duplicate = true;
            break;
          }
        }
      }
      if( !duplicate )
      {
        if( cs.sps->getChromaFormatIdc() != CHROMA_444 && numComp == 3 && pelListSort[i].getCnt(COMPONENT_Cb) == 0 )
        {
          if( best != -1 )
          {
            cu.curPLT[COMPONENT_Cb][paletteSize] = cs.prevPLT.curPLT[COMPONENT_Cb][best];
            cu.curPLT[COMPONENT_Cr][paletteSize] = cs.prevPLT.curPLT[COMPONENT_Cr][best];
          }
          else
          {
            cu.curPLT[COMPONENT_Cb][paletteSize] = 1 << (channelBitDepth_C - 1);
            cu.curPLT[COMPONENT_Cr][paletteSize] = 1 << (channelBitDepth_C - 1);
          }
        }
        paletteSize++;
      }
    }
    else
    {
      break;
    }
  }
  cu.curPLTSize[compBegin] = paletteSize;
#if !INTRA_RM_SMALL_BLOCK_SIZE_CONSTRAINTS
  if( cu.isLocalSepTree() )
  {
    cu.curPLTSize[COMPONENT_Y] = paletteSize;
  }
#endif
  delete[] pelList;
  delete[] pelListSort;
}
// -------------------------------------------------------------------------------------------------------------------
// 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() || cs.slice->getSPS()->getPLTMode())
          && cu.Y().valid())
      {
        m_CABACEstimator->cu_skip_flag( cu );
        m_CABACEstimator->pred_mode   ( cu );
      }
#if ENABLE_DIMD
      m_CABACEstimator->cu_dimd_flag(cu);
#endif
      if (CU::isPLT(cu))
      {
        return;
      }
    }

    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->bdpcm_mode( cu, COMPONENT_Y );
      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->bdpcm_mode( cu, ComponentID(CHANNEL_TYPE_CHROMA) );
      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;

  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)
  {
    const bool chromaCbfISP = currArea.blocks[COMPONENT_Cb].valid() && currCU.ispMode && !subdiv;
    if ( !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, CUCtx* cuCtx )
{
  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, cuCtx );
      subTuCounter += subTuCounter != -1 ? 1 : 0;
    } while( partitioner.nextPart( cs ) );

    partitioner.exitCurrSplit();
  }
  else
  {
    if (currArea.blocks[compID].valid())
    {
      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);
      }
      if (TU::getCbf(currTU, compID))
      {
        if (isLuma(compID))
        {
          m_CABACEstimator->residual_coding(currTU, compID, cuCtx);
          m_CABACEstimator->mts_idx(*currTU.cu, cuCtx);
        }
        else
        {
          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, CUCtx* cuCtx )
{
  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, cuCtx );
  }
  if( bChroma )
  {
    xEncCoeffQT( cs, partitioner, COMPONENT_Cb, subTuIdx, ispType );
    xEncCoeffQT( cs, partitioner, COMPONENT_Cr, subTuIdx, ispType );
  }

  CodingUnit& cu = *cs.getCU(partitioner.chType);
#if !INTRA_RM_SMALL_BLOCK_SIZE_CONSTRAINTS
  if ( cuCtx && bLuma && cu.isSepTree() && ( !cu.ispMode || ( cu.lfnstIdx && subTuIdx == 0 ) || ( !cu.lfnstIdx && subTuIdx == m_ispTestedModes[cu.lfnstIdx].numTotalParts[cu.ispMode - 1] - 1 ) ) )
#else
  if (cuCtx && bLuma && CS::isDualITree(cs) && (!cu.ispMode || (cu.lfnstIdx && subTuIdx == 0) || (!cu.lfnstIdx && subTuIdx == m_ispTestedModes[cu.lfnstIdx].numTotalParts[cu.ispMode - 1] - 1)))
#endif
  {
    m_CABACEstimator->residual_lfnst_mode(cu, *cuCtx);
  }

  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
  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 );
  //coeffs coding and cross comp coding
  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();
  // Include Cbf and jointCbCr flags here as we make decisions across components
  CodingStructure &cs = *currTU.cs;

  if ( currTU.jointCbCr )
  {
    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 ( compID == COMPONENT_Cb )
    {
      m_CABACEstimator->cbf_comp( cs, TU::getCbf( currTU, compID ), currTU.blocks[ compID ], currTU.depth, false );
    }
    else
    {
      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 );
    }
  }

  if( !currTU.jointCbCr && TU::getCbf( currTU, compID ) )
  {
    m_CABACEstimator->residual_coding( currTU, compID );
  }

  uint64_t fracBits = m_CABACEstimator->getEstFracBits();
  return fracBits;
}
#if JVET_W0103_INTRA_MTS
void IntraSearch::xSelectAMTForFullRD(TransformUnit &tu)
{
  if (!tu.blocks[COMPONENT_Y].valid())
  {
    return;
  }

  if (!tu.cu->mtsFlag)
  {
    return;
  }

  CodingStructure &cs = *tu.cs;
  m_pcRdCost->setChromaFormat(cs.sps->getChromaFormatIdc());

  const CompArea      &area = tu.blocks[COMPONENT_Y];

  const ChannelType    chType = toChannelType(COMPONENT_Y);


  PelBuf         piOrg = cs.getOrgBuf(area);
  PelBuf         piPred = cs.getPredBuf(area);
  PelBuf         piResi = cs.getResiBuf(area);


  const PredictionUnit &pu = *cs.getPU(area.pos(), chType);

  //===== init availability pattern =====

  PelBuf sharedPredTS(m_pSharedPredTransformSkip[COMPONENT_Y], area);
  initIntraPatternChType(*tu.cu, area);

  //===== get prediction signal =====
  if (PU::isMIP(pu, chType))
  {
    initIntraMip(pu, area);
    predIntraMip(COMPONENT_Y, piPred, pu);
  }
  else
  {
    predIntraAng(COMPONENT_Y, piPred, pu);
  }


  // save prediction
  sharedPredTS.copyFrom(piPred);

  const Slice           &slice = *cs.slice;
  //===== get residual signal =====
  piResi.copyFrom(piOrg);
  if (slice.getLmcsEnabledFlag() && m_pcReshape->getCTUFlag())
  {
    piResi.rspSignal(m_pcReshape->getFwdLUT());
    piResi.subtract(piPred);
  }
  else
  {
    piResi.subtract(piPred);
  }
  // do transform and calculate Coeff AbsSum for all MTS candidates
#if JVET_Y0142_ADAPT_INTRA_MTS
  int nCands = MTS_NCANDS[2];
  if (m_coeffAbsSumDCT2 >= 0 && m_coeffAbsSumDCT2 <= MTS_TH_COEFF[0])
  {
    nCands = MTS_NCANDS[0];
  }
  else if(m_coeffAbsSumDCT2 > MTS_TH_COEFF[0] && m_coeffAbsSumDCT2 <= MTS_TH_COEFF[1])
  {
    nCands = MTS_NCANDS[1];
  }
  std::vector<std::pair<int, uint64_t>> coeffAbsSum(nCands);
  for (int i = 0; i < nCands; i++)
#else
  std::vector<std::pair<int, uint64_t>> coeffAbsSum(4);
  for (int i = 0; i < 4; i++)
#endif
  {
    tu.mtsIdx[0] = i + MTS_DST7_DST7;
    uint64_t AbsSum = m_pcTrQuant->transformNxN(tu);
    coeffAbsSum[i] = { i, AbsSum };
  }
  std::stable_sort(coeffAbsSum.begin(), coeffAbsSum.end(), [](const std::pair<int, uint64_t> & l, const std::pair<int, uint64_t> & r) {return l.second < r.second; });
#if JVET_Y0142_ADAPT_INTRA_MTS
  for (int i = 0; i < nCands; i++)
#else
  for (int i = 0; i < 4; i++)
#endif
  {
    m_testAMTForFullRD[i] = coeffAbsSum[i].first;
  }
#if JVET_Y0142_ADAPT_INTRA_MTS
  m_numCandAMTForFullRD = nCands;
#else
  m_numCandAMTForFullRD = 4;
#endif
#if !JVET_Y0142_ADAPT_INTRA_MTS
  if (m_pcEncCfg->getUseFastLFNST())
  {
    double skipThreshold = 1.0 + 1.0 / sqrt((double)(area.width*area.height));
    skipThreshold = std::max(skipThreshold, 1.03);
    for (int i = 1; i < m_numCandAMTForFullRD; i++)
    {
      if (coeffAbsSum[i].second > skipThreshold * coeffAbsSum[0].second)
      {
        m_numCandAMTForFullRD = i;
        break;
      }
    }
  }
#endif
}
#endif
void IntraSearch::xIntraCodingTUBlock(TransformUnit &tu, const ComponentID &compID, 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;
#if JVET_V0094_BILATERAL_FILTER || JVET_X0071_CHROMA_BILATERAL_FILTER
  const PPS           &pps                  = *cs.pps;
#endif

  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         piReco                     = cs.getRecoBuf   (area);

#if JVET_AB0061_ITMP_BV_FOR_IBC
  PredictionUnit &pu = *cs.getPU(area.pos(), chType);
#else
  const PredictionUnit &pu                  = *cs.getPU(area.pos(), chType);
#endif
  const uint32_t           uiChFinalMode        = PU::getFinalIntraMode(pu, chType);

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

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

      //===== get prediction signal =====
      if(compID != COMPONENT_Y && !tu.cu->bdpcmModeChroma && PU::isLMCMode(uiChFinalMode))
      {
        xGetLumaRecPixels( pu, area );
        predIntraChromaLM( compID, piPred, pu, area, uiChFinalMode );
      }
      else
      {
#if JVET_V0130_INTRA_TMP
        if( PU::isTmp( pu, chType ) )
        {
          int foundCandiNum;
#if JVET_W0069_TMP_BOUNDARY
          RefTemplateType tempType = getRefTemplateType( *(tu.cu), tu.cu->blocks[COMPONENT_Y] );
          if( tempType != NO_TEMPLATE )
          {
            getTargetTemplate( tu.cu, pu.lwidth(), pu.lheight(), tempType );
            candidateSearchIntra( tu.cu, pu.lwidth(), pu.lheight(), tempType );
            generateTMPrediction( piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), foundCandiNum );
#if JVET_AB0061_ITMP_BV_FOR_IBC
            pu.interDir               = 1;             // use list 0 for IBC mode
            pu.refIdx[REF_PIC_LIST_0] = MAX_NUM_REF;   // last idx in the list
            pu.mv->set(m_tempLibFast.getX() << MV_FRACTIONAL_BITS_INTERNAL, m_tempLibFast.getY() << MV_FRACTIONAL_BITS_INTERNAL);
            pu.bv.set(m_tempLibFast.getX(), m_tempLibFast.getY());
#endif
          }
          else
          {
            foundCandiNum = 1;
            generateTmDcPrediction( piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), 1 << (tu.cu->cs->sps->getBitDepth( CHANNEL_TYPE_LUMA ) - 1) );
#if JVET_AB0061_ITMP_BV_FOR_IBC
            pu.interDir               = 1;             // use list 0 for IBC mode
            pu.refIdx[REF_PIC_LIST_0] = MAX_NUM_REF;   // last idx in the list
            pu.mv->set(0, 0);
            pu.bv.set(0, 0);
#endif
          }
#else
          getTargetTemplate( tu.cu, pu.lwidth(), pu.lheight() );
          candidateSearchIntra( tu.cu, pu.lwidth(), pu.lheight() );
          generateTMPrediction( piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), foundCandiNum );
#endif
          CHECK( foundCandiNum < 1, "" );
        }
        else if( PU::isMIP( pu, chType ) )
#else
        if( PU::isMIP( pu, chType ) )
#endif
        {
          initIntraMip( pu, area );
          predIntraMip( compID, piPred, pu );
        }
        else
        {
          if (predRegDiffFromTB)
          {
            if (firstTBInPredReg)
            {
              PelBuf piPredReg = cs.getPredBuf(areaPredReg);
              predIntraAng(compID, piPredReg, pu);
            }
          }
          else
          {
            predIntraAng(compID, piPred, pu);
          }
#if JVET_Z0050_DIMD_CHROMA_FUSION
          if (compID != COMPONENT_Y && pu.isChromaFusion)
          {
            geneChromaFusionPred(compID, piPred, pu);
          }
#endif
        }
      }

      // 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_W0103_INTRA_MTS
  if (!tu.cu->mtsFlag && isLuma(compID))
#else
  if (isLuma(compID))
#endif
  {
    //===== get residual signal =====
    piResi.copyFrom( piOrg  );
    if (slice.getLmcsEnabledFlag() && m_pcReshape->getCTUFlag() && compID == COMPONENT_Y)
    {
      piResi.rspSignal( m_pcReshape->getFwdLUT() );
      piResi.subtract( piPred );
    }
    else
    {
      piResi.subtract( piPred );
    }
  }

  //===== 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.getPicHeader()->getLmcsChromaResidualScaleFlag() )
  {
    int cResScaleInv = tu.getChromaAdj();
    double cResScale = (double)(1 << CSCALE_FP_PREC) / (double)cResScaleInv;
    m_pcTrQuant->setLambda(m_pcTrQuant->getLambda() / (cResScale*cResScale));
  }

  PelBuf          crOrg;
  PelBuf          crPred;
  PelBuf          crResi;
  PelBuf          crReco;

  if (isChroma(compID))
  {
    const CompArea &crArea = tu.blocks[ COMPONENT_Cr ];
    crOrg  = cs.getOrgBuf  ( crArea );
    crPred = cs.getPredBuf ( crArea );
    crResi = cs.getResiBuf ( crArea );
    crReco = cs.getRecoBuf ( crArea );
  }

  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) && (tu.cu->cs->slice->getSliceQp() > 18) )
  {
    m_pcTrQuant->setLambda( 1.3 * m_pcTrQuant->getLambda() );
  }

  if( isLuma(compID) )
  {
    if (trModes)
    {
#if JVET_Y0142_ADAPT_INTRA_MTS
      m_pcTrQuant->transformNxN(tu, compID, cQP, trModes, 8);
#else
      m_pcTrQuant->transformNxN(tu, compID, cQP, trModes, m_pcEncCfg->getMTSIntraMaxCand());
#endif
      tu.mtsIdx[compID] = trModes->at(0).first;
    }

    if (!(m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && slice.isLossless() && tu.mtsIdx[compID] == 0) || tu.cu->bdpcmMode != 0)
    {
      m_pcTrQuant->transformNxN(tu, compID, cQP, uiAbsSum, m_CABACEstimator->getCtx(), loadTr);
    }

    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 (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;
    }
#if JVET_Y0142_ADAPT_INTRA_MTS
    if (isLuma(compID) && tu.mtsIdx[compID] >= MTS_DST7_DST7)
    {
      bool signHiding = cs.slice->getSignDataHidingEnabledFlag();
      CoeffCodingContext  cctx(tu, compID, signHiding);
      const TCoeff*       coeff = tu.getCoeffs(compID).buf;
      int          scanPosLast = -1;
      uint64_t     coeffAbsSum = 0;

      for (int scanPos = 0; scanPos < cctx.maxNumCoeff(); scanPos++)
      {
        unsigned blkPos = cctx.blockPos(scanPos);
        if (coeff[blkPos])
        {
          scanPosLast = scanPos;
          coeffAbsSum += abs(coeff[blkPos]);
        }
      }
      int nCands = (coeffAbsSum > MTS_TH_COEFF[1]) ? MTS_NCANDS[2] : (coeffAbsSum > MTS_TH_COEFF[0]) ? MTS_NCANDS[1] : MTS_NCANDS[0];
      bool isInvalid = (scanPosLast <= 0) || ((tu.mtsIdx[COMPONENT_Y] - MTS_DST7_DST7) >= nCands);
      if (isInvalid)
      {
        m_validMTSReturn = false;
        ruiDist = MAX_INT;
        return;
      }
    }
#endif
    if ((m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && slice.isLossless() && tu.mtsIdx[compID] == 0)
        && 0 == tu.cu->bdpcmMode)
    {
      uiAbsSum = 0;
      tu.getCoeffs(compID).fill(0);
      TU::setCbfAtDepth(tu, compID, tu.depth, 0);
    }

    //--- inverse transform ---
    if (uiAbsSum > 0)
    {
      m_pcTrQuant->invTransformNxN(tu, compID, piResi, cQP);
    }
    else
    {
      piResi.fill(0);
    }
  }
  else // chroma
  {
    int         codedCbfMask  = 0;
    ComponentID codeCompId    = (tu.jointCbCr ? (tu.jointCbCr >> 1 ? COMPONENT_Cb : COMPONENT_Cr) : compID);
    const QpParam qpCbCr(tu, codeCompId);

    if( tu.jointCbCr )
    {
      ComponentID otherCompId = ( codeCompId==COMPONENT_Cr ? COMPONENT_Cb : COMPONENT_Cr );
      tu.getCoeffs( otherCompId ).fill(0); // do we need that?
      TU::setCbfAtDepth (tu, otherCompId, tu.depth, false );
    }
    PelBuf& codeResi = ( codeCompId == COMPONENT_Cr ? crResi : piResi );
    uiAbsSum = 0;

    if (trModes)
    {