IntraPrediction.cpp

    ptrSrc = srcBuf - (1 + multiRefIdx) * srcStride - (1 + multiRefIdx);
    ptrDst = refBufUnfiltered;
    if (neighborFlags[totalLeftUnits])
    {
      ptrDst[0] = ptrSrc[0];
      ptrDst[predStride] = ptrSrc[0];
      for (int i = 1; i <= multiRefIdx; i++)
      {
        ptrDst[i] = ptrSrc[i];
        ptrDst[i + predStride] = ptrSrc[i * srcStride];
      }
    }

    // Fill left & below-left samples if available (downwards)
    ptrSrc += (1 + multiRefIdx) * srcStride;
    ptrDst += (1 + multiRefIdx) + predStride;
    for (int unitIdx = totalLeftUnits - 1; unitIdx > 0; unitIdx--)
    {
      if (neighborFlags[unitIdx])
      {
        for (int i = 0; i < unitHeight; i++)
        {
          ptrDst[i] = ptrSrc[i * srcStride];
        }
      }
      ptrSrc += unitHeight * srcStride;
      ptrDst += unitHeight;
    }
    // Fill last below-left sample(s)
    if (neighborFlags[0])
    {
      int lastSample = (predHSize % unitHeight == 0) ? unitHeight : predHSize % unitHeight;
      for (int i = 0; i < lastSample; i++)
      {
        ptrDst[i] = ptrSrc[i * srcStride];
      }
    }

    // Fill above & above-right samples if available (left-to-right)
    ptrSrc = srcBuf - srcStride * (1 + multiRefIdx);
    ptrDst = refBufUnfiltered + 1 + multiRefIdx;
    for (int unitIdx = totalLeftUnits + 1; unitIdx < totalUnits - 1; unitIdx++)
    {
      if (neighborFlags[unitIdx])
      {
        for (int j = 0; j < unitWidth; j++)
        {
          ptrDst[j] = ptrSrc[j];
        }
      }
      ptrSrc += unitWidth;
      ptrDst += unitWidth;
    }
    // Fill last above-right sample(s)
    if (neighborFlags[totalUnits - 1])
    {
      int lastSample = (predSize % unitWidth == 0) ? unitWidth : predSize % unitWidth;
      for (int j = 0; j < lastSample; j++)
      {
        ptrDst[j] = ptrSrc[j];
      }
    }

    // pad from first available down to the last below-left
    ptrDst = refBufUnfiltered;
    int lastAvailUnit = 0;
    if (!neighborFlags[0])
    {
      int firstAvailUnit = 1;
      while (firstAvailUnit < totalUnits && !neighborFlags[firstAvailUnit])
      {
        firstAvailUnit++;
      }

      // first available sample
      int firstAvailRow = -1;
      int firstAvailCol = 0;
      if (firstAvailUnit < totalLeftUnits)
      {
        firstAvailRow = (totalLeftUnits - firstAvailUnit) * unitHeight + multiRefIdx;
      }
      else if (firstAvailUnit == totalLeftUnits)
      {
        firstAvailRow = multiRefIdx;
      }
      else
      {
        firstAvailCol = (firstAvailUnit - totalLeftUnits - 1) * unitWidth + 1 + multiRefIdx;
      }
      const Pel firstAvailSample = ptrDst[firstAvailRow < 0 ? firstAvailCol : firstAvailRow + predStride];

      // last sample below-left (n.a.)
      int lastRow = predHSize + multiRefIdx;

      // fill left column
      for (int i = lastRow; i > firstAvailRow; i--)
      {
        ptrDst[i + predStride] = firstAvailSample;
      }
      // fill top row
      if (firstAvailCol > 0)
      {
        for (int j = 0; j < firstAvailCol; j++)
        {
          ptrDst[j] = firstAvailSample;
        }
      }
      lastAvailUnit = firstAvailUnit;
    }

    // pad all other reference samples.
    int currUnit = lastAvailUnit + 1;
    while (currUnit < totalUnits)
    {
      if (!neighborFlags[currUnit]) // samples not available
      {
        // last available sample
        int lastAvailRow = -1;
        int lastAvailCol = 0;
        if (lastAvailUnit < totalLeftUnits)
        {
          lastAvailRow = (totalLeftUnits - lastAvailUnit - 1) * unitHeight + multiRefIdx + 1;
        }
        else if (lastAvailUnit == totalLeftUnits)
        {
          lastAvailCol = multiRefIdx;
        }
        else
        {
          lastAvailCol = (lastAvailUnit - totalLeftUnits) * unitWidth + multiRefIdx;
        }
        const Pel lastAvailSample = ptrDst[lastAvailRow < 0 ? lastAvailCol : lastAvailRow + predStride];

        // fill current unit with last available sample
        if (currUnit < totalLeftUnits)
        {
          for (int i = lastAvailRow - 1; i >= lastAvailRow - unitHeight; i--)
          {
            ptrDst[i + predStride] = lastAvailSample;
          }
        }
        else if (currUnit == totalLeftUnits)
        {
          for (int i = 0; i < multiRefIdx + 1; i++)
          {
            ptrDst[i + predStride] = lastAvailSample;
          }
          for (int j = 0; j < multiRefIdx + 1; j++)
          {
            ptrDst[j] = lastAvailSample;
          }
        }
        else
        {
          int numSamplesInUnit = (currUnit == totalUnits - 1) ? ((predSize % unitWidth == 0) ? unitWidth : predSize % unitWidth) : unitWidth;
          for (int j = lastAvailCol + 1; j <= lastAvailCol + numSamplesInUnit; j++)
          {
            ptrDst[j] = lastAvailSample;
          }
        }
      }
      lastAvailUnit = currUnit;
      currUnit++;
    }
  }
}

void IntraPrediction::xFilterReferenceSamples(const Pel *refBufUnfiltered, Pel *refBufFiltered, const CompArea &area,
                                              const SPS &sps, int multiRefIdx
)
{
  if (area.compID != COMPONENT_Y)
  {
    multiRefIdx = 0;
  }
  const int predSize = m_topRefLength + multiRefIdx;
  const int predHSize = m_leftRefLength + multiRefIdx;
  const size_t predStride = m_refBufferStride[area.compID];

  const Pel topLeft =
    (refBufUnfiltered[0] + refBufUnfiltered[1] + refBufUnfiltered[predStride] + refBufUnfiltered[predStride + 1] + 2)
    >> 2;

  refBufFiltered[0] = topLeft;

  for (int i = 1; i < predSize; i++)
  {
    refBufFiltered[i] = (refBufUnfiltered[i - 1] + 2 * refBufUnfiltered[i] + refBufUnfiltered[i + 1] + 2) >> 2;
  }
  refBufFiltered[predSize] = refBufUnfiltered[predSize];

  refBufFiltered += predStride;
  refBufUnfiltered += predStride;

  refBufFiltered[0] = topLeft;

  for (int i = 1; i < predHSize; i++)
  {
    refBufFiltered[i] = (refBufUnfiltered[i - 1] + 2 * refBufUnfiltered[i] + refBufUnfiltered[i + 1] + 2) >> 2;
  }
  refBufFiltered[predHSize] = refBufUnfiltered[predHSize];
}

bool isAboveLeftAvailable(const CodingUnit &cu, const ChannelType &chType, const Position &posLT)
{
  const CodingStructure& cs = *cu.cs;
  const Position refPos = posLT.offset(-1, -1);

  if (!cs.isDecomp(refPos, chType))
  {
    return false;
  }

  return (cs.getCURestricted(refPos, cu, chType) != NULL);
}

int isAboveAvailable(const CodingUnit &cu, const ChannelType &chType, const Position &posLT, const uint32_t uiNumUnitsInPU, const uint32_t unitWidth, bool *bValidFlags)
{
  const CodingStructure& cs = *cu.cs;

  bool *    validFlags = bValidFlags;
  int       numIntra   = 0;
  const int maxDx      = uiNumUnitsInPU * unitWidth;

  for (int dx = 0; dx < maxDx; dx += unitWidth)
  {
    const Position refPos = posLT.offset(dx, -1);

    if (!cs.isDecomp(refPos, chType))
    {
      break;
    }

    const bool valid = (cs.getCURestricted(refPos, cu, chType) != NULL);
    numIntra += valid ? 1 : 0;
    *validFlags = valid;

    validFlags++;
  }

  return numIntra;
}

int isLeftAvailable(const CodingUnit &cu, const ChannelType &chType, const Position &posLT, const uint32_t uiNumUnitsInPU, const uint32_t unitHeight, bool *bValidFlags)
{
  const CodingStructure& cs = *cu.cs;

  bool *    validFlags = bValidFlags;
  int       numIntra   = 0;
  const int maxDy      = uiNumUnitsInPU * unitHeight;

  for (int dy = 0; dy < maxDy; dy += unitHeight)
  {
    const Position refPos = posLT.offset(-1, dy);

    if (!cs.isDecomp(refPos, chType))
    {
      break;
    }

    const bool valid = (cs.getCURestricted(refPos, cu, chType) != NULL);
    numIntra += valid ? 1 : 0;
    *validFlags = valid;

    validFlags--;
  }

  return numIntra;
}

int isAboveRightAvailable(const CodingUnit &cu, const ChannelType &chType, const Position &posRT, const uint32_t uiNumUnitsInPU, const uint32_t unitWidth, bool *bValidFlags )
{
  const CodingStructure& cs = *cu.cs;

  bool *    validFlags = bValidFlags;
  int       numIntra   = 0;
  const int maxDx      = uiNumUnitsInPU * unitWidth;

  for (int dx = 0; dx < maxDx; dx += unitWidth)
  {
    const Position refPos = posRT.offset(unitWidth + dx, -1);

    if (!cs.isDecomp(refPos, chType))
    {
      break;
    }

    const bool valid = (cs.getCURestricted(refPos, cu, chType) != NULL);
    numIntra += valid ? 1 : 0;
    *validFlags = valid;

    validFlags++;
  }

  return numIntra;
}

int isBelowLeftAvailable(const CodingUnit &cu, const ChannelType &chType, const Position &posLB, const uint32_t uiNumUnitsInPU, const uint32_t unitHeight, bool *bValidFlags )
{
  const CodingStructure& cs = *cu.cs;

  bool *    validFlags = bValidFlags;
  int       numIntra   = 0;
  const int maxDy      = uiNumUnitsInPU * unitHeight;

  for (int dy = 0; dy < maxDy; dy += unitHeight)
  {
    const Position refPos = posLB.offset(-1, unitHeight + dy);

    if (!cs.isDecomp(refPos, chType))
    {
      break;
    }

    const bool valid = (cs.getCURestricted(refPos, cu, chType) != NULL);
    numIntra += valid ? 1 : 0;
    *validFlags = valid;

    validFlags--;
  }

  return numIntra;
}

// LumaRecPixels
void IntraPrediction::xGetLumaRecPixels(const PredictionUnit &pu, CompArea chromaArea)
{
  int iDstStride = 0;
  Pel* pDst0 = 0;
  int curChromaMode = pu.intraDir[1];
  if ((curChromaMode == MDLM_L_IDX) || (curChromaMode == MDLM_T_IDX))
  {
    iDstStride = 2 * MAX_CU_SIZE + 1;
    pDst0 = m_pMdlmTemp + iDstStride + 1;
  }
  else
  {
    iDstStride = MAX_CU_SIZE + 1;
    pDst0 = m_piTemp + iDstStride + 1; //MMLM_SAMPLE_NEIGHBOR_LINES;
  }
  //assert 420 chroma subsampling
  CompArea lumaArea = CompArea( COMPONENT_Y, pu.chromaFormat, chromaArea.lumaPos(), recalcSize( pu.chromaFormat, CHANNEL_TYPE_CHROMA, CHANNEL_TYPE_LUMA, chromaArea.size() ) );//needed for correct pos/size (4x4 Tus)

  CHECK(lumaArea.width == chromaArea.width && CHROMA_444 != pu.chromaFormat, "");
  CHECK(lumaArea.height == chromaArea.height && CHROMA_444 != pu.chromaFormat && CHROMA_422 != pu.chromaFormat, "");

  const SizeType uiCWidth = chromaArea.width;
  const SizeType uiCHeight = chromaArea.height;

  const CPelBuf Src = pu.cs->picture->getRecoBuf( lumaArea );
  Pel const* pRecSrc0   = Src.bufAt( 0, 0 );
  int iRecStride        = Src.stride;
  int logSubWidthC  = getChannelTypeScaleX(CHANNEL_TYPE_CHROMA, pu.chromaFormat);
  int logSubHeightC = getChannelTypeScaleY(CHANNEL_TYPE_CHROMA, pu.chromaFormat);

  int iRecStride2       = iRecStride << logSubHeightC;

  const CodingUnit& lumaCU = isChroma( pu.chType ) ? *pu.cs->picture->cs->getCU( lumaArea.pos(), CH_L ) : *pu.cu;
  const CodingUnit&     cu = *pu.cu;

  const CompArea& area = isChroma( pu.chType ) ? chromaArea : lumaArea;

  const uint32_t uiTuWidth  = area.width;
  const uint32_t uiTuHeight = area.height;

  int iBaseUnitSize = ( 1 << MIN_CU_LOG2 );

  const int  iUnitWidth       = iBaseUnitSize >> getComponentScaleX( area.compID, area.chromaFormat );
  const int  iUnitHeight = iBaseUnitSize >> getComponentScaleY(area.compID, area.chromaFormat);

  const int  iTUWidthInUnits = uiTuWidth / iUnitWidth;
  const int  iTUHeightInUnits = uiTuHeight / iUnitHeight;
  const int  iAboveUnits      = iTUWidthInUnits;
  const int  iLeftUnits       = iTUHeightInUnits;
  const int  chromaUnitWidth = iBaseUnitSize >> getComponentScaleX(COMPONENT_Cb, area.chromaFormat);
  const int  chromaUnitHeight = iBaseUnitSize >> getComponentScaleY(COMPONENT_Cb, area.chromaFormat);
  const int  topTemplateSampNum = 2 * uiCWidth; // for MDLM, the number of template samples is 2W or 2H.
  const int  leftTemplateSampNum = 2 * uiCHeight;
  assert(m_topRefLength >= topTemplateSampNum);
  assert(m_leftRefLength >= leftTemplateSampNum);
  const int  totalAboveUnits = (topTemplateSampNum + (chromaUnitWidth - 1)) / chromaUnitWidth;
  const int  totalLeftUnits = (leftTemplateSampNum + (chromaUnitHeight - 1)) / chromaUnitHeight;
  const int  totalUnits = totalLeftUnits + totalAboveUnits + 1;
  const int  aboveRightUnits = totalAboveUnits - iAboveUnits;
  const int  leftBelowUnits = totalLeftUnits - iLeftUnits;

  int avaiAboveRightUnits = 0;
  int avaiLeftBelowUnits = 0;
  bool  bNeighborFlags[4 * MAX_NUM_PART_IDXS_IN_CTU_WIDTH + 1];
  memset(bNeighborFlags, 0, totalUnits);
  bool aboveIsAvailable, leftIsAvailable;

  int availlableUnit = isLeftAvailable( isChroma( pu.chType ) ? cu : lumaCU, toChannelType( area.compID ), area.pos(), iLeftUnits, iUnitHeight,
  ( bNeighborFlags + iLeftUnits + leftBelowUnits - 1 ) );

  leftIsAvailable = availlableUnit == iTUHeightInUnits;

  availlableUnit = isAboveAvailable( isChroma( pu.chType ) ? cu : lumaCU, toChannelType( area.compID ), area.pos(), iAboveUnits, iUnitWidth,
  ( bNeighborFlags + iLeftUnits + leftBelowUnits + 1 ) );

  aboveIsAvailable = availlableUnit == iTUWidthInUnits;

  if (leftIsAvailable)   // if left is not available, then the below left is not available
  {
    avaiLeftBelowUnits = isBelowLeftAvailable(isChroma(pu.chType) ? cu : lumaCU, toChannelType(area.compID), area.bottomLeftComp(area.compID), leftBelowUnits, iUnitHeight, (bNeighborFlags + leftBelowUnits - 1));
  }

  if (aboveIsAvailable)   // if above is not available, then  the above right is not available.
  {
    avaiAboveRightUnits = isAboveRightAvailable(isChroma(pu.chType) ? cu : lumaCU, toChannelType(area.compID), area.topRightComp(area.compID), aboveRightUnits, iUnitWidth, (bNeighborFlags + iLeftUnits + leftBelowUnits + iAboveUnits + 1));
  }

  Pel*       pDst  = nullptr;
  Pel const* piSrc = nullptr;

  bool isFirstRowOfCtu = (lumaArea.y & ((pu.cs->sps)->getCTUSize() - 1)) == 0;

  if (aboveIsAvailable)
  {
    pDst  = pDst0    - iDstStride;
    int addedAboveRight = 0;
    if ((curChromaMode == MDLM_L_IDX) || (curChromaMode == MDLM_T_IDX))
    {
      addedAboveRight = avaiAboveRightUnits*chromaUnitWidth;
    }
    for (int i = 0; i < uiCWidth + addedAboveRight; i++)
    {
      const bool leftPadding = i == 0 && !leftIsAvailable;
      if (pu.chromaFormat == CHROMA_444)
      {
        piSrc = pRecSrc0 - iRecStride;
        pDst[i] = piSrc[i];
      }
      else if (isFirstRowOfCtu)
      {
        piSrc   = pRecSrc0 - iRecStride;
        pDst[i] = (piSrc[2 * i] * 2 + piSrc[2 * i - (leftPadding ? 0 : 1)] + piSrc[2 * i + 1] + 2) >> 2;
      }
      else if (pu.chromaFormat == CHROMA_422)
      {
        piSrc = pRecSrc0 - iRecStride2;

        int s = 2;
        s += piSrc[2 * i] * 2;
        s += piSrc[2 * i - (leftPadding ? 0 : 1)];
        s += piSrc[2 * i + 1];
        pDst[i] = s >> 2;
      }
      else if (pu.cs->sps->getCclmCollocatedChromaFlag())
      {
        piSrc = pRecSrc0 - iRecStride2;

        int s = 4;
        s += piSrc[2 * i - iRecStride];
        s += piSrc[2 * i] * 4;
        s += piSrc[2 * i - (leftPadding ? 0 : 1)];
        s += piSrc[2 * i + 1];
        s += piSrc[2 * i + iRecStride];
        pDst[i] = s >> 3;
      }
      else
      {
        piSrc = pRecSrc0 - iRecStride2;
        int s = 4;
        s += piSrc[2 * i] * 2;
        s += piSrc[2 * i + 1];
        s += piSrc[2 * i - (leftPadding ? 0 : 1)];
        s += piSrc[2 * i + iRecStride] * 2;
        s += piSrc[2 * i + 1 + iRecStride];
        s += piSrc[2 * i + iRecStride - (leftPadding ? 0 : 1)];
        pDst[i] = s >> 3;
      }
    }
  }

  if (leftIsAvailable)
  {
    pDst  = pDst0    - 1;
    piSrc = pRecSrc0 - 1 - logSubWidthC;

    int addedLeftBelow = 0;
    if ((curChromaMode == MDLM_L_IDX) || (curChromaMode == MDLM_T_IDX))
    {
      addedLeftBelow = avaiLeftBelowUnits*chromaUnitHeight;
    }

    for (int j = 0; j < uiCHeight + addedLeftBelow; j++)
    {
      if (pu.chromaFormat == CHROMA_444)
      {
        pDst[0] = piSrc[0];
      }
      else if (pu.chromaFormat == CHROMA_422)
      {
        int s = 2;
        s += piSrc[0] * 2;
        s += piSrc[-1];
        s += piSrc[1];
        pDst[0] = s >> 2;
      }
      else if (pu.cs->sps->getCclmCollocatedChromaFlag())
      {
        const bool abovePadding = j == 0 && !aboveIsAvailable;

        int s = 4;
        s += piSrc[-(abovePadding ? 0 : iRecStride)];
        s += piSrc[0] * 4;
        s += piSrc[-1];
        s += piSrc[1];
        s += piSrc[iRecStride];
        pDst[0] = s >> 3;
      }
      else
      {
        int s = 4;
        s += piSrc[0] * 2;
        s += piSrc[1];
        s += piSrc[-1];
        s += piSrc[iRecStride] * 2;
        s += piSrc[iRecStride + 1];
        s += piSrc[iRecStride - 1];
        pDst[0] = s >> 3;
      }

      piSrc += iRecStride2;
      pDst  += iDstStride;
    }
  }

  // inner part from reconstructed picture buffer
  for( int j = 0; j < uiCHeight; j++ )
  {
    for( int i = 0; i < uiCWidth; i++ )
    {
      if (pu.chromaFormat == CHROMA_444)
      {
        pDst0[i] = pRecSrc0[i];
      }
      else if (pu.chromaFormat == CHROMA_422)
      {
        const bool leftPadding  = i == 0 && !leftIsAvailable;

        int s = 2;
        s += pRecSrc0[2 * i] * 2;
        s += pRecSrc0[2 * i - (leftPadding ? 0 : 1)];
        s += pRecSrc0[2 * i + 1];
        pDst0[i] = s >> 2;
      }
      else if (pu.cs->sps->getCclmCollocatedChromaFlag())
      {
        const bool leftPadding  = i == 0 && !leftIsAvailable;
        const bool abovePadding = j == 0 && !aboveIsAvailable;

        int s = 4;
        s += pRecSrc0[2 * i - (abovePadding ? 0 : iRecStride)];
        s += pRecSrc0[2 * i] * 4;
        s += pRecSrc0[2 * i - (leftPadding ? 0 : 1)];
        s += pRecSrc0[2 * i + 1];
        s += pRecSrc0[2 * i + iRecStride];
        pDst0[i] = s >> 3;
      }
      else
      {
        CHECK(pu.chromaFormat != CHROMA_420, "Chroma format must be 4:2:0 for vertical filtering");
        const bool leftPadding = i == 0 && !leftIsAvailable;

        int s = 4;
        s += pRecSrc0[2 * i] * 2;
        s += pRecSrc0[2 * i + 1];
        s += pRecSrc0[2 * i - (leftPadding ? 0 : 1)];
        s += pRecSrc0[2 * i + iRecStride] * 2;
        s += pRecSrc0[2 * i + 1 + iRecStride];
        s += pRecSrc0[2 * i + iRecStride - (leftPadding ? 0 : 1)];
        pDst0[i] = s >> 3;
      }
    }

    pDst0    += iDstStride;
    pRecSrc0 += iRecStride2;
  }
}
void IntraPrediction::xGetLMParameters(const PredictionUnit &pu, const ComponentID compID,
                                              const CompArea &chromaArea,
                                              int &a, int &b, int &iShift)
{
  CHECK(compID == COMPONENT_Y, "");

  const SizeType cWidth  = chromaArea.width;
  const SizeType cHeight = chromaArea.height;

  const Position posLT = chromaArea;

  CodingStructure & cs = *(pu.cs);
  const CodingUnit &cu = *(pu.cu);

  const SPS &        sps           = *cs.sps;
  const uint32_t     tuWidth     = chromaArea.width;
  const uint32_t     tuHeight    = chromaArea.height;
  const ChromaFormat nChromaFormat = sps.getChromaFormatIdc();

  const int baseUnitSize = 1 << MIN_CU_LOG2;
  const int unitWidth    = baseUnitSize >> getComponentScaleX(chromaArea.compID, nChromaFormat);
  const int unitHeight   = baseUnitSize >> getComponentScaleY(chromaArea.compID, nChromaFormat);

  const int tuWidthInUnits  = tuWidth / unitWidth;
  const int tuHeightInUnits = tuHeight / unitHeight;
  const int aboveUnits      = tuWidthInUnits;
  const int leftUnits       = tuHeightInUnits;
  int topTemplateSampNum = 2 * cWidth; // for MDLM, the template sample number is 2W or 2H;
  int leftTemplateSampNum = 2 * cHeight;
  assert(m_topRefLength >= topTemplateSampNum);
  assert(m_leftRefLength >= leftTemplateSampNum);
  int totalAboveUnits = (topTemplateSampNum + (unitWidth - 1)) / unitWidth;
  int totalLeftUnits = (leftTemplateSampNum + (unitHeight - 1)) / unitHeight;
  int totalUnits = totalLeftUnits + totalAboveUnits + 1;
  int aboveRightUnits = totalAboveUnits - aboveUnits;
  int leftBelowUnits = totalLeftUnits - leftUnits;
  int avaiAboveRightUnits = 0;
  int avaiLeftBelowUnits = 0;
  int avaiAboveUnits = 0;
  int avaiLeftUnits = 0;

  int curChromaMode = pu.intraDir[1];
  bool neighborFlags[4 * MAX_NUM_PART_IDXS_IN_CTU_WIDTH + 1];
  memset(neighborFlags, 0, totalUnits);

  bool aboveAvailable, leftAvailable;

  int availableUnit =
    isAboveAvailable(cu, CHANNEL_TYPE_CHROMA, posLT, aboveUnits, unitWidth,
    (neighborFlags + leftUnits + leftBelowUnits + 1));
  aboveAvailable = availableUnit == tuWidthInUnits;

  availableUnit =
    isLeftAvailable(cu, CHANNEL_TYPE_CHROMA, posLT, leftUnits, unitHeight,
    (neighborFlags + leftUnits + leftBelowUnits - 1));
  leftAvailable = availableUnit == tuHeightInUnits;
  if (leftAvailable) // if left is not available, then the below left is not available
  {
    avaiLeftUnits = tuHeightInUnits;
    avaiLeftBelowUnits = isBelowLeftAvailable(cu, CHANNEL_TYPE_CHROMA, chromaArea.bottomLeftComp(chromaArea.compID), leftBelowUnits, unitHeight, (neighborFlags + leftBelowUnits - 1));
  }
  if (aboveAvailable) // if above is not available, then  the above right is not available.
  {
    avaiAboveUnits = tuWidthInUnits;
    avaiAboveRightUnits = isAboveRightAvailable(cu, CHANNEL_TYPE_CHROMA, chromaArea.topRightComp(chromaArea.compID), aboveRightUnits, unitWidth, (neighborFlags + leftUnits + leftBelowUnits + aboveUnits + 1));
  }
  Pel *srcColor0, *curChroma0;
  int srcStride;

  PelBuf temp;
  if ((curChromaMode == MDLM_L_IDX) || (curChromaMode == MDLM_T_IDX))
  {
    srcStride = 2 * MAX_CU_SIZE + 1;
    temp = PelBuf(m_pMdlmTemp + srcStride + 1, srcStride, Size(chromaArea));
  }
  else
  {
    srcStride = MAX_CU_SIZE + 1;
    temp        = PelBuf(m_piTemp + srcStride + 1, srcStride, Size(chromaArea));
  }
  srcColor0 = temp.bufAt(0, 0);
  curChroma0 = getPredictorPtr(compID);


  unsigned internalBitDepth = sps.getBitDepth(CHANNEL_TYPE_CHROMA);

  int minLuma[2] = {  MAX_INT, 0 };
  int maxLuma[2] = { -MAX_INT, 0 };

  Pel *src = srcColor0 - srcStride;
  int actualTopTemplateSampNum = 0;
  int actualLeftTemplateSampNum = 0;
  if (curChromaMode == MDLM_T_IDX)
  {
    leftAvailable = 0;
    avaiAboveRightUnits = avaiAboveRightUnits > (cHeight/unitWidth) ?  cHeight/unitWidth : avaiAboveRightUnits;
    actualTopTemplateSampNum = unitWidth*(avaiAboveUnits + avaiAboveRightUnits);
  }
  else if (curChromaMode == MDLM_L_IDX)
  {
    aboveAvailable = 0;
    avaiLeftBelowUnits = avaiLeftBelowUnits > (cWidth/unitHeight) ? cWidth/unitHeight : avaiLeftBelowUnits;
    actualLeftTemplateSampNum = unitHeight*(avaiLeftUnits + avaiLeftBelowUnits);
  }
  else if (curChromaMode == LM_CHROMA_IDX)
  {
    actualTopTemplateSampNum = cWidth;
    actualLeftTemplateSampNum = cHeight;
  }
  int startPos[2]; //0:Above, 1: Left
  int pickStep[2];

  int aboveIs4 = leftAvailable  ? 0 : 1;
  int leftIs4 =  aboveAvailable ? 0 : 1;

  startPos[0] = actualTopTemplateSampNum >> (2 + aboveIs4);
  pickStep[0] = std::max(1, actualTopTemplateSampNum >> (1 + aboveIs4));

  startPos[1] = actualLeftTemplateSampNum >> (2 + leftIs4);
  pickStep[1] = std::max(1, actualLeftTemplateSampNum >> (1 + leftIs4));

  Pel selectLumaPix[4] = { 0, 0, 0, 0 };
  Pel selectChromaPix[4] = { 0, 0, 0, 0 };

  int cntT, cntL;
  cntT = cntL = 0;
  int cnt = 0;
  if (aboveAvailable)
  {
    cntT = std::min(actualTopTemplateSampNum, (1 + aboveIs4) << 1);
    src = srcColor0 - srcStride;
    const Pel *cur = curChroma0 + 1;
    for (int pos = startPos[0]; cnt < cntT; pos += pickStep[0], cnt++)
    {
      selectLumaPix[cnt] = src[pos];
      selectChromaPix[cnt] = cur[pos];
    }
  }

  if (leftAvailable)
  {
    cntL = std::min(actualLeftTemplateSampNum, ( 1 + leftIs4 ) << 1 );
    src = srcColor0 - 1;
    const Pel *cur = curChroma0 + m_refBufferStride[compID] + 1;
    for (int pos = startPos[1], cnt = 0; cnt < cntL; pos += pickStep[1], cnt++)
    {
      selectLumaPix[cnt + cntT] = src[pos * srcStride];
      selectChromaPix[cnt + cntT] = cur[pos];
    }
  }
  cnt = cntL + cntT;

  if (cnt == 2)
  {
    selectLumaPix[3] = selectLumaPix[0]; selectChromaPix[3] = selectChromaPix[0];
    selectLumaPix[2] = selectLumaPix[1]; selectChromaPix[2] = selectChromaPix[1];
    selectLumaPix[0] = selectLumaPix[1]; selectChromaPix[0] = selectChromaPix[1];
    selectLumaPix[1] = selectLumaPix[3]; selectChromaPix[1] = selectChromaPix[3];
  }

  int minGrpIdx[2] = { 0, 2 };
  int maxGrpIdx[2] = { 1, 3 };
  int *tmpMinGrp = minGrpIdx;
  int *tmpMaxGrp = maxGrpIdx;
  if (selectLumaPix[tmpMinGrp[0]] > selectLumaPix[tmpMinGrp[1]]) std::swap(tmpMinGrp[0], tmpMinGrp[1]);
  if (selectLumaPix[tmpMaxGrp[0]] > selectLumaPix[tmpMaxGrp[1]]) std::swap(tmpMaxGrp[0], tmpMaxGrp[1]);
  if (selectLumaPix[tmpMinGrp[0]] > selectLumaPix[tmpMaxGrp[1]]) std::swap(tmpMinGrp, tmpMaxGrp);
  if (selectLumaPix[tmpMinGrp[1]] > selectLumaPix[tmpMaxGrp[0]]) std::swap(tmpMinGrp[1], tmpMaxGrp[0]);

  minLuma[0] = (selectLumaPix[tmpMinGrp[0]] + selectLumaPix[tmpMinGrp[1]] + 1 )>>1;
  minLuma[1] = (selectChromaPix[tmpMinGrp[0]] + selectChromaPix[tmpMinGrp[1]] + 1) >> 1;
  maxLuma[0] = (selectLumaPix[tmpMaxGrp[0]] + selectLumaPix[tmpMaxGrp[1]] + 1 )>>1;
  maxLuma[1] = (selectChromaPix[tmpMaxGrp[0]] + selectChromaPix[tmpMaxGrp[1]] + 1) >> 1;

  if (leftAvailable || aboveAvailable)
  {
    int diff = maxLuma[0] - minLuma[0];
    if (diff > 0)
    {
      int diffC = maxLuma[1] - minLuma[1];
      int x = floorLog2( diff );
      static const uint8_t DivSigTable[1 << 4] = {
        // 4bit significands - 8 ( MSB is omitted )
        0,  7,  6,  5,  5,  4,  4,  3,  3,  2,  2,  1,  1,  1,  1,  0
      };
      int normDiff = (diff << 4 >> x) & 15;
      int v = DivSigTable[normDiff] | 8;
      x += normDiff != 0;

      int y = floorLog2( abs( diffC ) ) + 1;
      int add = 1 << y >> 1;
      a = (diffC * v + add) >> y;
      iShift = 3 + x - y;
      if ( iShift < 1 )
      {
        iShift = 1;
        a = ( (a == 0)? 0: (a < 0)? -15 : 15 );   // a=Sign(a)*15
      }
      b = minLuma[1] - ((a * minLuma[0]) >> iShift);
    }
    else
    {
      a = 0;
      b = minLuma[1];
      iShift = 0;
    }
  }
  else
  {
    a = 0;

    b = 1 << (internalBitDepth - 1);

    iShift = 0;
  }
}

void IntraPrediction::initIntraMip( const PredictionUnit &pu, const CompArea &area )
{
  CHECK( area.width > MIP_MAX_WIDTH || area.height > MIP_MAX_HEIGHT, "Error: block size not supported for MIP" );

  // prepare input (boundary) data for prediction
  CHECK( m_ipaParam.refFilterFlag, "ERROR: unfiltered refs expected for MIP" );
  Pel       *ptrSrc     = getPredictorPtr(area.compID);
  const int  srcStride  = m_refBufferStride[area.compID];
  const int  srcHStride = 2;

  m_matrixIntraPred.prepareInputForPred(CPelBuf(ptrSrc, srcStride, srcHStride), area,
                                        pu.cu->slice->getSPS()->getBitDepth(toChannelType(area.compID)), area.compID);
}

void IntraPrediction::predIntraMip( const ComponentID compId, PelBuf &piPred, const PredictionUnit &pu )
{
  CHECK( piPred.width > MIP_MAX_WIDTH || piPred.height > MIP_MAX_HEIGHT, "Error: block size not supported for MIP" );
  CHECK( piPred.width != (1 << floorLog2(piPred.width)) || piPred.height != (1 << floorLog2(piPred.height)), "Error: expecting blocks of size 2^M x 2^N" );

  // generate mode-specific prediction
  uint32_t modeIdx       = MAX_NUM_MIP_MODE;
  bool     transposeFlag = false;
  if (compId == COMPONENT_Y)
  {
    modeIdx       = pu.intraDir[CHANNEL_TYPE_LUMA];
    transposeFlag = pu.mipTransposedFlag;
  }
  else
  {
    const PredictionUnit &coLocatedLumaPU = PU::getCoLocatedLumaPU(pu);

    CHECK(pu.intraDir[CHANNEL_TYPE_CHROMA] != DM_CHROMA_IDX, "Error: MIP is only supported for chroma with DM_CHROMA.");
    CHECK(!coLocatedLumaPU.cu->mipFlag, "Error: Co-located luma CU should use MIP.");

    modeIdx       = coLocatedLumaPU.intraDir[CHANNEL_TYPE_LUMA];
    transposeFlag = coLocatedLumaPU.mipTransposedFlag;
  }
  const int bitDepth = pu.cu->slice->getSPS()->getBitDepth(toChannelType(compId));

  CHECK(modeIdx >= getNumModesMip(piPred), "Error: Wrong MIP mode index");

  static_vector<int, MIP_MAX_WIDTH* MIP_MAX_HEIGHT> predMip( piPred.width * piPred.height );
  m_matrixIntraPred.predBlock(predMip.data(), modeIdx, transposeFlag, bitDepth, compId);

  for( int y = 0; y < piPred.height; y++ )
  {
    for( int x = 0; x < piPred.width; x++ )
    {
      piPred.at( x, y ) = Pel(predMip[y * piPred.width + x]);
    }
  }
}
void IntraPrediction::reorderPLT(CodingStructure& cs, Partitioner& partitioner, ComponentID compBegin, uint32_t numComp)
{
  CodingUnit &cu = *cs.getCU(partitioner.chType);

  uint8_t        reusePLTSizetmp = 0;
  uint8_t        pltSizetmp = 0;
  Pel            curPLTtmp[MAX_NUM_COMPONENT][MAXPLTSIZE];
  bool           curPLTpred[MAXPLTPREDSIZE];

  for (int idx = 0; idx < MAXPLTPREDSIZE; idx++)
  {
    curPLTpred[idx] = false;
    cu.reuseflag[compBegin][idx] = false;
  }
  for (int idx = 0; idx < MAXPLTSIZE; idx++)
  {
    curPLTpred[idx] = false;
  }

  for (int predidx = 0; predidx < cs.prevPLT.curPLTSize[compBegin]; predidx++)
  {
    bool match = false;
    int curidx = 0;

    for (curidx = 0; curidx < cu.curPLTSize[compBegin]; curidx++)
    {
      if( curPLTpred[curidx] )
        continue;
      bool matchTmp = true;
      for (int comp = compBegin; comp < (compBegin + numComp); comp++)
      {
        matchTmp = matchTmp && (cu.curPLT[comp][curidx] == cs.prevPLT.curPLT[comp][predidx]);
      }
      if (matchTmp)
      {
        match = true;
        break;
      }
    }

    if (match)
    {
      cu.reuseflag[compBegin][predidx] = true;
      curPLTpred[curidx] = true;
      if( cu.isLocalSepTree() )
      {
        cu.reuseflag[COMPONENT_Y][predidx] = true;
        for( int comp = COMPONENT_Y; comp < MAX_NUM_COMPONENT; comp++ )
        {
          curPLTtmp[comp][reusePLTSizetmp] = cs.prevPLT.curPLT[comp][predidx];
        }
      }
      else
      {
      for (int comp = compBegin; comp < (compBegin + numComp); comp++)
      {
        curPLTtmp[comp][reusePLTSizetmp] = cs.prevPLT.curPLT[comp][predidx];
      }
      }
      reusePLTSizetmp++;
      pltSizetmp++;
    }
  }
  cu.reusePLTSize[compBegin] = reusePLTSizetmp;
  for (int curidx = 0; curidx < cu.curPLTSize[compBegin]; curidx++)
  {
    if (!curPLTpred[curidx])
    {
      if( cu.isLocalSepTree() )
      {
        for( int comp = compBegin; comp < (compBegin + numComp); comp++ )
        {
          curPLTtmp[comp][pltSizetmp] = cu.curPLT[comp][curidx];
        }
        if( isLuma(partitioner.chType) )
        {
          curPLTtmp[COMPONENT_Cb][pltSizetmp] = 1 << (cs.sps->getBitDepth(CHANNEL_TYPE_CHROMA) - 1);
          curPLTtmp[COMPONENT_Cr][pltSizetmp] = 1 << (cs.sps->getBitDepth(CHANNEL_TYPE_CHROMA) - 1);
        }
        else
        {
          curPLTtmp[COMPONENT_Y][pltSizetmp] = 1 << (cs.sps->getBitDepth(CHANNEL_TYPE_LUMA) - 1);
        }
      }
      else
      {
      for (int comp = compBegin; comp < (compBegin + numComp); comp++)
      {
        curPLTtmp[comp][pltSizetmp] = cu.curPLT[comp][curidx];
      }
      }
      pltSizetmp++;
    }
  }
  assert(pltSizetmp == cu.curPLTSize[compBegin]);
  for (int curidx = 0; curidx < cu.curPLTSize[compBegin]; curidx++)
  {
    if( cu.isLocalSepTree() )
    {
      for( int comp = COMPONENT_Y; comp < MAX_NUM_COMPONENT; comp++ )
      {
        cu.curPLT[comp][curidx] = curPLTtmp[comp][curidx];
      }
    }
    else
    {
    for (int comp = compBegin; comp < (compBegin + numComp); comp++)
    {
      cu.curPLT[comp][curidx] = curPLTtmp[comp][curidx];
    }
    }
  }
}
//! \}