IntraSearch.cpp

            m_bestModeCostValid[ lfnstIdx ] = true;
          }
          if( csBest->cost < bestCurrentCost )
          {
            bestCurrentCost = csBest->cost;
          }
          if( !cu.ispMode && !cu.mtsFlag )
          {
            m_modeCtrl->setMtsFirstPassNoIspCost( csBest->cost );
          }
        }
        if( !cu.ispMode && !cu.bdpcmMode && csBest->cost < bestCostNonBDPCM )
        {
          bestCostNonBDPCM = csBest->cost;
        }
      }

      csTemp->releaseIntermediateData();
      if( m_pcEncCfg->getFastLocalDualTreeMode() )
      {
        if( cu.isConsIntra() && !cu.slice->isIntra() && csBest->cost != MAX_DOUBLE && costInterCU != COST_UNKNOWN && mode >= 0 )
        {
          if( m_pcEncCfg->getFastLocalDualTreeMode() == 2 )
          {
            //Note: only try one intra mode, which is especially useful to reduce EncT for LDB case (around 4%)
            break;
          }
          else
          {
            if( csBest->cost > costInterCU * 1.5 )
            {
              break;
            }
          }
        }
      }
    } // Mode loop
    cu.ispMode = uiBestPUMode.ispMod;

    if( validReturn )
    {
      cs.useSubStructure( *csBest, partitioner.chType, pu.singleChan( CHANNEL_TYPE_LUMA ), true, true, keepResi, keepResi );
    }
    csBest->releaseIntermediateData();
    if( validReturn )
    {
      //=== update PU data ====
      cu.mipFlag = uiBestPUMode.mipFlg;
      pu.multiRefIdx = uiBestPUMode.mRefId;
      pu.intraDir[ CHANNEL_TYPE_LUMA ] = uiBestPUMode.modeId;
      cu.bdpcmMode = bestBDPCMMode;
    }
  }

  //===== reset context models =====
  m_CABACEstimator->getCtx() = ctxStart;

  return validReturn;
}

void IntraSearch::estIntraPredChromaQT( CodingUnit &cu, Partitioner &partitioner, const double maxCostAllowed )
{
  const ChromaFormat format   = cu.chromaFormat;
  const uint32_t    numberValidComponents = getNumberValidComponents(format);
  CodingStructure &cs = *cu.cs;
  const TempCtx ctxStart  ( m_CtxCache, m_CABACEstimator->getCtx() );

  cs.setDecomp( cs.area.Cb(), false );

  double    bestCostSoFar = maxCostAllowed;
  bool      lumaUsesISP   = !cu.isSepTree() && cu.ispMode;
  PartSplit ispType       = lumaUsesISP ? CU::getISPType( cu, COMPONENT_Y ) : TU_NO_ISP;
  CHECK( cu.ispMode && bestCostSoFar < 0, "bestCostSoFar must be positive!" );

  auto &pu = *cu.firstPU;

  {
    uint32_t       uiBestMode = 0;
    Distortion uiBestDist = 0;
    double     dBestCost = MAX_DOUBLE;

    //----- init mode list ----
    {
      uint32_t  uiMinMode = 0;
      uint32_t  uiMaxMode = NUM_CHROMA_MODE;

      //----- check chroma modes -----
      uint32_t chromaCandModes[ NUM_CHROMA_MODE ];
      PU::getIntraChromaCandModes( pu, chromaCandModes );

      // create a temporary CS
      CodingStructure &saveCS = *m_pSaveCS[0];
      saveCS.pcv      = cs.pcv;
      saveCS.picture  = cs.picture;
      saveCS.area.repositionTo( cs.area );
      saveCS.clearTUs();

      if( !cu.isSepTree() && cu.ispMode )
      {
        saveCS.clearCUs();
        saveCS.clearPUs();
      }

      if( cu.isSepTree() )
      {
        if( partitioner.canSplit( TU_MAX_TR_SPLIT, cs ) )
        {
          partitioner.splitCurrArea( TU_MAX_TR_SPLIT, cs );

          do
          {
            cs.addTU( CS::getArea( cs, partitioner.currArea(), partitioner.chType ), partitioner.chType ).depth = partitioner.currTrDepth;
          } while( partitioner.nextPart( cs ) );

          partitioner.exitCurrSplit();
        }
        else
        cs.addTU( CS::getArea( cs, partitioner.currArea(), partitioner.chType ), partitioner.chType );
      }

      std::vector<TransformUnit*> orgTUs;

      if( lumaUsesISP )
      {
        CodingUnit& auxCU = saveCS.addCU( cu, partitioner.chType );
        auxCU.ispMode = cu.ispMode;
        saveCS.sps = cu.cs->sps;
        saveCS.addPU( *cu.firstPU, partitioner.chType );
      }


      // create a store for the TUs
      for( const auto &ptu : cs.tus )
      {
        // for split TUs in HEVC, add the TUs without Chroma parts for correct setting of Cbfs
        if( lumaUsesISP || pu.contains( *ptu, CHANNEL_TYPE_CHROMA ) )
        {
          saveCS.addTU( *ptu, partitioner.chType );
          orgTUs.push_back( ptu );
        }
      }
      if( lumaUsesISP )
      {
        saveCS.clearCUs();
      }
      // SATD pre-selecting.
      int satdModeList[NUM_CHROMA_MODE];
      int64_t satdSortedCost[NUM_CHROMA_MODE];
      for (int i = 0; i < NUM_CHROMA_MODE; i++)
      {
        satdSortedCost[i] = 0; // for the mode not pre-select by SATD, do RDO by default, so set the initial value 0.
        satdModeList[i] = 0;
      }
      bool modeIsEnable[NUM_INTRA_MODE + 1]; // use intra mode idx to check whether enable
      for (int i = 0; i < NUM_INTRA_MODE + 1; i++)
      {
        modeIsEnable[i] = 1;
      }
#if JVET_P0058_CHROMA_TS_ENCODER_INTRA_SAD_MOD
      DistParam distParamSad;
      DistParam distParamSatd;
#else
      DistParam distParam;
#endif
#if JVET_P0058_CHROMA_TS_ENCODER_INTRA_SAD_MOD
#else
      const bool useHadamard = !cu.transQuantBypass;
#endif
      pu.intraDir[1] = MDLM_L_IDX; // temporary assigned, just to indicate this is a MDLM mode. for luma down-sampling operation.

      initIntraPatternChType(cu, pu.Cb());
      initIntraPatternChType(cu, pu.Cr());
      xGetLumaRecPixels(pu, pu.Cb());

      for (int idx = uiMinMode; idx <= uiMaxMode - 1; idx++)
      {
        int mode = chromaCandModes[idx];
        satdModeList[idx] = mode;
        if (PU::isLMCMode(mode) && !PU::isLMCModeEnabled(pu, mode))
        {
          continue;
        }
        if ((mode == LM_CHROMA_IDX) || (mode == PLANAR_IDX) || (mode == DM_CHROMA_IDX)) // only pre-check regular modes and MDLM modes, not including DM ,Planar, and LM
        {
          continue;
        }
        pu.intraDir[1] = mode; // temporary assigned, for SATD checking.

#if JVET_P0058_CHROMA_TS_ENCODER_INTRA_SAD_MOD
        int64_t sad = 0;
        int64_t sadCb = 0;
        int64_t satdCb = 0;
        int64_t sadCr = 0;
        int64_t satdCr = 0;
#else
        int64_t sad = 0;
#endif
        CodingStructure& cs = *(pu.cs);

        CompArea areaCb = pu.Cb();
        PelBuf orgCb = cs.getOrgBuf(areaCb);
        PelBuf predCb = cs.getPredBuf(areaCb);
#if JVET_P0058_CHROMA_TS_ENCODER_INTRA_SAD_MOD
        m_pcRdCost->setDistParam(distParamSad, orgCb, predCb, pu.cs->sps->getBitDepth(CHANNEL_TYPE_CHROMA), COMPONENT_Cb, false);
        m_pcRdCost->setDistParam(distParamSatd, orgCb, predCb, pu.cs->sps->getBitDepth(CHANNEL_TYPE_CHROMA), COMPONENT_Cb, true);
#else
        m_pcRdCost->setDistParam(distParam, orgCb, predCb, pu.cs->sps->getBitDepth(CHANNEL_TYPE_CHROMA), COMPONENT_Cb, useHadamard);
#endif
#if JVET_P0058_CHROMA_TS_ENCODER_INTRA_SAD_MOD
        distParamSad.applyWeight = false;
        distParamSatd.applyWeight = false;
#else
        distParam.applyWeight = false;
#endif
        if (PU::isLMCMode(mode))
        {
          predIntraChromaLM(COMPONENT_Cb, predCb, pu, areaCb, mode);
        }
        else
        {
          initPredIntraParams(pu, pu.Cb(), *pu.cs->sps);
          predIntraAng(COMPONENT_Cb, predCb, pu);
        }
#if JVET_P0058_CHROMA_TS_ENCODER_INTRA_SAD_MOD
        sadCb = distParamSad.distFunc(distParamSad) * 2;
        satdCb = distParamSatd.distFunc(distParamSatd);
        sad += std::min(sadCb, satdCb);
#else
        sad += distParam.distFunc(distParam);
#endif
        CompArea areaCr = pu.Cr();
        PelBuf orgCr = cs.getOrgBuf(areaCr);
        PelBuf predCr = cs.getPredBuf(areaCr);
#if JVET_P0058_CHROMA_TS_ENCODER_INTRA_SAD_MOD
        m_pcRdCost->setDistParam(distParamSad, orgCr, predCr, pu.cs->sps->getBitDepth(CHANNEL_TYPE_CHROMA), COMPONENT_Cr, false);
        m_pcRdCost->setDistParam(distParamSatd, orgCr, predCr, pu.cs->sps->getBitDepth(CHANNEL_TYPE_CHROMA), COMPONENT_Cr, true);
#else
        m_pcRdCost->setDistParam(distParam, orgCr, predCr, pu.cs->sps->getBitDepth(CHANNEL_TYPE_CHROMA), COMPONENT_Cr, useHadamard);
#endif
#if JVET_P0058_CHROMA_TS_ENCODER_INTRA_SAD_MOD
        distParamSad.applyWeight = false;
        distParamSatd.applyWeight = false;
#else
        distParam.applyWeight = false;
#endif
        if (PU::isLMCMode(mode))
        {
          predIntraChromaLM(COMPONENT_Cr, predCr, pu, areaCr, mode);
        }
        else
        {
          initPredIntraParams(pu, pu.Cr(), *pu.cs->sps);
          predIntraAng(COMPONENT_Cr, predCr, pu);
        }
#if JVET_P0058_CHROMA_TS_ENCODER_INTRA_SAD_MOD
        sadCr = distParamSad.distFunc(distParamSad) * 2;
        satdCr = distParamSatd.distFunc(distParamSatd);
        sad += std::min(sadCr, satdCr);
#else
        sad += distParam.distFunc(distParam);
#endif
        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.sps->getTransformSkipEnabledFlag())
        {
          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::saveCuAreaCostInSCIPU( Area area, double cost )
{
  if( m_numCuInSCIPU < NUM_INTER_CU_INFO_SAVE )
  {
    m_cuAreaInSCIPU[m_numCuInSCIPU] = area;
    m_cuCostInSCIPU[m_numCuInSCIPU] = cost;
    m_numCuInSCIPU++;
  }
}

void IntraSearch::initCuAreaCostInSCIPU()
{
  for( int i = 0; i < NUM_INTER_CU_INFO_SAVE; i++ )
  {
    m_cuAreaInSCIPU[i] = Area();
    m_cuCostInSCIPU[i] = 0;
  }
  m_numCuInSCIPU = 0;
}
void IntraSearch::PLTSearch(CodingStructure &cs, Partitioner& partitioner, ComponentID compBegin, uint32_t numComp)
{
  CodingUnit    &cu = *cs.getCU(partitioner.chType);
  TransformUnit &tu = *cs.getTU(partitioner.chType);
  uint32_t height = cu.block(compBegin).height;
  uint32_t width = cu.block(compBegin).width;
  m_orgCtxRD = PLTCtx(m_CABACEstimator->getCtx());

  if (m_pcEncCfg->getReshaper() && (cs.slice->getLmcsEnabledFlag() && m_pcReshape->getCTUFlag()))
  {
    cs.getPredBuf().copyFrom(cs.getOrgBuf());
    cs.getPredBuf().Y().rspSignal(m_pcReshape->getFwdLUT());
  }

  Pel  *runLength = tu.getRunLens (compBegin);
  bool *runType   = tu.getRunTypes(compBegin);
  cu.lastPLTSize[compBegin] = cs.prevPLT.curPLTSize[compBegin];
  //derive palette
  derivePLTLossy(cs, partitioner, compBegin, numComp);
  reorderPLT(cs, partitioner, compBegin, numComp);

  //calculate palette index
  preCalcPLTIndex(cs, partitioner, compBegin, numComp);
  //derive run
  uint64_t bits = MAX_UINT64;
  deriveRunAndCalcBits(cs, partitioner, compBegin, numComp, PLT_SCAN_HORTRAV, bits);
  if ((cu.curPLTSize[compBegin] + cu.useEscape[compBegin]) > 1)
  {
    deriveRunAndCalcBits(cs, partitioner, compBegin, numComp, PLT_SCAN_VERTRAV, bits);
  }
  cu.useRotation[compBegin] = m_bestScanRotationMode;
  memcpy(runType, m_runTypeRD, sizeof(bool)*width*height);
  memcpy(runLength, m_runLengthRD, sizeof(Pel)*width*height);
  //reconstruct pixel
  PelBuf    curPLTIdx = tu.getcurPLTIdx(compBegin);
  for (uint32_t y = 0; y < height; y++)
  {
    for (uint32_t x = 0; x < width; x++)
    {
      if (curPLTIdx.at(x, y) == cu.curPLTSize[compBegin])
      {

      }
      else
      {
        for (uint32_t compID = compBegin; compID < (compBegin + numComp); compID++)
        {
          CompArea area = cu.blocks[compID];
          PelBuf   recBuf = cs.getRecoBuf(area);
          uint32_t scaleX = getComponentScaleX((ComponentID)COMPONENT_Cb, cs.sps->getChromaFormatIdc());
          uint32_t scaleY = getComponentScaleY((ComponentID)COMPONENT_Cb, cs.sps->getChromaFormatIdc());
          if (compBegin != COMPONENT_Y || compID == COMPONENT_Y)
          {
            recBuf.at(x, y) = cu.curPLT[compID][curPLTIdx.at(x, y)];
          }
          else if (compBegin == COMPONENT_Y && compID != COMPONENT_Y && y % (1 << scaleY) == 0 && x % (1 << scaleX) == 0)
          {
            recBuf.at(x >> scaleX, y >> scaleY) = cu.curPLT[compID][curPLTIdx.at(x, y)];
          }
        }
      }
    }
  }

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

  cs.fracBits = MAX_UINT;
  cs.cost = MAX_DOUBLE;
  Distortion distortion = 0;
  for (uint32_t comp = compBegin; comp < (compBegin + numComp); comp++)
  {
    const ComponentID compID = ComponentID(comp);
    CPelBuf reco = cs.getRecoBuf(compID);
    CPelBuf org = cs.getOrgBuf(compID);
#if WCG_EXT
    if (m_pcEncCfg->getLumaLevelToDeltaQPMapping().isEnabled() || (
      m_pcEncCfg->getReshaper() && (cs.slice->getLmcsEnabledFlag() && m_pcReshape->getCTUFlag())))
    {
      const CPelBuf orgLuma = cs.getOrgBuf(cs.area.blocks[COMPONENT_Y]);

      if (compID == COMPONENT_Y && !(m_pcEncCfg->getLumaLevelToDeltaQPMapping().isEnabled()))
      {
        const CompArea &areaY = cu.Y();
        CompArea tmpArea1(COMPONENT_Y, areaY.chromaFormat, Position(0, 0), areaY.size());
        PelBuf   tmpRecLuma = m_tmpStorageLCU.getBuf(tmpArea1);
        tmpRecLuma.copyFrom(reco);
        tmpRecLuma.rspSignal(m_pcReshape->getInvLUT());
        distortion += m_pcRdCost->getDistPart(org, tmpRecLuma, cs.sps->getBitDepth(toChannelType(compID)), compID, DF_SSE_WTD, &orgLuma);
      }
      else
      {
        distortion += m_pcRdCost->getDistPart(org, reco, cs.sps->getBitDepth(toChannelType(compID)), compID, DF_SSE_WTD, &orgLuma);
      }
    }
    else
#endif
      distortion += m_pcRdCost->getDistPart(org, reco, cs.sps->getBitDepth(toChannelType(compID)), compID, DF_SSE);
  }

  cs.dist += distortion;
  const CompArea &area = cu.blocks[compBegin];
  cs.setDecomp(area);
  cs.picture->getRecoBuf(area).copyFrom(cs.getRecoBuf(area));
}
void IntraSearch::deriveRunAndCalcBits(CodingStructure& cs, Partitioner& partitioner, ComponentID compBegin, uint32_t numComp, PLTScanMode pltScanMode, uint64_t& minBits)
{
  CodingUnit    &cu = *cs.getCU(partitioner.chType);
  TransformUnit &tu = *cs.getTU(partitioner.chType);
  uint32_t height = cu.block(compBegin).height;
  uint32_t width = cu.block(compBegin).width;
  Pel  *runLength = tu.getRunLens (compBegin);
  bool *runType   = tu.getRunTypes(compBegin);

  cu.useRotation[compBegin] = (pltScanMode == PLT_SCAN_VERTRAV);
  m_scanOrder = g_scanOrder[SCAN_UNGROUPED][(cu.useRotation[compBegin]) ? SCAN_TRAV_VER : SCAN_TRAV_HOR][gp_sizeIdxInfo->idxFrom(width)][gp_sizeIdxInfo->idxFrom(height)];
  deriveRun(cs, partitioner, compBegin);

  m_CABACEstimator->getCtx() = PLTCtx(m_orgCtxRD);
  m_CABACEstimator->resetBits();
  CUCtx cuCtx;
  cuCtx.isDQPCoded = true;
  cuCtx.isChromaQpAdjCoded = true;
  m_CABACEstimator->cu_palette_info(cu, compBegin, numComp, cuCtx);
  uint64_t bitsTemp = m_CABACEstimator->getEstFracBits();
  if (minBits > bitsTemp)
  {
    m_bestScanRotationMode = pltScanMode;
    memcpy(m_runTypeRD, runType, sizeof(bool)*width*height);
    memcpy(m_runLengthRD, runLength, sizeof(Pel)*width*height);
    minBits = bitsTemp;
  }
}
void IntraSearch::deriveRun(CodingStructure& cs, Partitioner& partitioner, ComponentID compBegin)
{
  CodingUnit    &cu = *cs.getCU(partitioner.chType);
  TransformUnit &tu = *cs.getTU(partitioner.chType);
  uint32_t height = cu.block(compBegin).height;
  uint32_t width = cu.block(compBegin).width;
  uint32_t total = height * width, idx = 0;
  uint32_t startPos = 0;
  uint64_t indexBits = 0, runBitsIndex = 0, runBitsCopy = 0;
  m_storeCtxRun = PLTCtx(m_orgCtxRD);

  PLTtypeBuf  runType = tu.getrunType(compBegin);
  PelBuf      runLength = tu.getrunLength(compBegin);
  while (idx < total)
  {
    startPos = idx;
    double aveBitsPerPix[NUM_PLT_RUN];
    uint32_t indexRun = 0;
    bool runValid = calIndexRun(cs, partitioner, startPos, total, indexRun, compBegin);
    m_CABACEstimator->getCtx() = PLTCtx(m_storeCtxRun);
    aveBitsPerPix[PLT_RUN_INDEX] = runValid ? getRunBits(cu, indexRun, startPos, PLT_RUN_INDEX, &indexBits, &runBitsIndex, compBegin) : MAX_DOUBLE;
    m_storeCtxRunIndex = PLTCtx(m_CABACEstimator->getCtx());

    uint32_t copyRun = 0;
    bool copyValid = calCopyRun(cs, partitioner, startPos, total, copyRun, compBegin);
    m_CABACEstimator->getCtx() = PLTCtx(m_storeCtxRun);
    aveBitsPerPix[PLT_RUN_COPY] = copyValid ? getRunBits(cu, copyRun, startPos, PLT_RUN_COPY, &indexBits, &runBitsCopy, compBegin) : MAX_DOUBLE;
    m_storeCtxRunCopy = PLTCtx(m_CABACEstimator->getCtx());

    if (copyValid == 0 && runValid == 0)
    {
      assert(0);
    }
    else
    {
      if (aveBitsPerPix[PLT_RUN_COPY] <= aveBitsPerPix[PLT_RUN_INDEX])
      {
        for (int runidx = 0; runidx <copyRun; runidx++)
        {
          uint32_t posy = m_scanOrder[idx + runidx].y;
          uint32_t posx = m_scanOrder[idx + runidx].x;
          runType.at(posx, posy) = PLT_RUN_COPY;
          runLength.at(posx, posy) = copyRun;
        }
        idx += copyRun;
        m_storeCtxRun = PLTCtx(m_storeCtxRunCopy);

      }
      else
      {
        for (int runidx = 0; runidx <indexRun; runidx++)
        {
          uint32_t posy = m_scanOrder[idx + runidx].y;
          uint32_t posx = m_scanOrder[idx + runidx].x;
          runType.at(posx, posy) = PLT_RUN_INDEX;
          runLength.at(posx, posy) = indexRun;
        }
        idx += indexRun;
        m_storeCtxRun = PLTCtx(m_storeCtxRunIndex);

      }
    }
  }
  assert(idx == total);
}
double IntraSearch::getRunBits(const CodingUnit&  cu, uint32_t run, uint32_t strPos, PLTRunMode paletteRunMode, uint64_t* indexBits, uint64_t* runBits, ComponentID compBegin)
{
  TransformUnit&   tu = *cu.firstTU;
  uint32_t height = cu.block(compBegin).height;
  uint32_t width  = cu.block(compBegin).width;
  uint32_t endPos = height*width;
  PLTtypeBuf runType   = tu.getrunType(compBegin);
  PelBuf     curPLTIdx = tu.getcurPLTIdx(compBegin);
  uint32_t   indexMaxSize = (cu.useEscape[compBegin]) ? (cu.curPLTSize[compBegin] + 1) : cu.curPLTSize[compBegin];

  m_CABACEstimator->resetBits();
  ///////////////// encode Run Type
  m_CABACEstimator->encodeRunType(cu, runType, strPos, m_scanOrder, compBegin);
  uint64_t runTypeBits = m_CABACEstimator->getEstFracBits();
  uint32_t curLevel = 0;
  switch (paletteRunMode)
  {
  case PLT_RUN_INDEX:
    curLevel = m_CABACEstimator->writePLTIndex(cu, strPos, curPLTIdx, runType, indexMaxSize, compBegin);
    *indexBits = m_CABACEstimator->getEstFracBits() - runTypeBits;
    m_CABACEstimator->cu_run_val(run - 1, PLT_RUN_INDEX, curLevel, endPos - strPos - 1);
    *runBits = m_CABACEstimator->getEstFracBits() - runTypeBits - (*indexBits);
    break;
  case PLT_RUN_COPY:
    m_CABACEstimator->cu_run_val(run - 1, PLT_RUN_COPY, curLevel, endPos - strPos - 1);
    *runBits = m_CABACEstimator->getEstFracBits() - runTypeBits;
    break;
  default:
    assert(0);
  }
  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_P0460_PLT_TS_MIN_QP
    qp[ch] = cQP.Qp(true);
#else
    qp[ch] = cQP.Qp(false);
#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;
}
// -------------------------------------------------------------------------------------------------------------------
// 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()
      )
      {
        if( cs.pps->getTransquantBypassEnabledFlag() )
        {
          m_CABACEstimator->cu_transquant_bypass_flag( cu );