InterPrediction.cpp

    padStr = m_filteredBlockTmp[2 + refList][COMPONENT_Y] + 2 * stridePredMC + 1;
    ::memcpy(padStr - stridePredMC, padStr, sizeof(Pel)*(widthG));
    ::memcpy(padStr + height*stridePredMC, padStr + (height - 1)*stridePredMC, sizeof(Pel)*(widthG));
  }

  const ClpRng& clpRng = pu.cu->cs->slice->clpRng(COMPONENT_Y);
  const int   bitDepth = clipBitDepths.recon[toChannelType(COMPONENT_Y)];
  const int   shiftNum = IF_INTERNAL_PREC + 1 - bitDepth;
  const int   offset = (1 << (shiftNum - 1)) + 2 * IF_INTERNAL_OFFS;
  const int   limit = ((int)1 << (4 + IF_INTERNAL_PREC - bitDepth - 5));

  int*     dotProductTemp1 = m_dotProduct1;
  int*     dotProductTemp2 = m_dotProduct2;
  int*     dotProductTemp3 = m_dotProduct3;
  int*     dotProductTemp5 = m_dotProduct5;
  int*     dotProductTemp6 = m_dotProduct6;

  xCalcBIOPar(srcY0Temp, srcY1Temp, gradX0, gradX1, gradY0, gradY1, dotProductTemp1, dotProductTemp2, dotProductTemp3, dotProductTemp5, dotProductTemp6, src0Stride, src1Stride, widthG, widthG, heightG);

  int xUnit = (width >> 2);
  int yUnit = (height >> 2);

  Pel *dstY0 = dstY;
  gradX0 = m_gradX0; gradX1 = m_gradX1;
  gradY0 = m_gradY0; gradY1 = m_gradY1;

  for (int yu = 0; yu < yUnit; yu++)
  {
    for (int xu = 0; xu < xUnit; xu++)
    {
      if (m_bioPredSubBlkDist[yu*xUnit + xu] < m_bioSubBlkDistThres)
      {
        srcY0Temp = srcY0 + (stridePredMC + 1) + ((yu*src0Stride + xu) << 2);
        srcY1Temp = srcY1 + (stridePredMC + 1) + ((yu*src1Stride + xu) << 2);
        dstY0 = dstY + ((yu*dstStride + xu) << 2);
        PelBuf dstPelBuf(dstY0, dstStride, Size(4, 4));
        dstPelBuf.addAvg(CPelBuf(srcY0Temp, src0Stride, Size(4, 4)), CPelBuf(srcY1Temp, src1Stride, Size(4, 4)), clpRng);
        continue;
      }

      int     sGxdI = 0, sGydI = 0, sGxGy = 0, sGx2 = 0, sGy2 = 0;
      int     tmpx = 0, tmpy = 0;

      dotProductTemp1 = m_dotProduct1 + offsetPos + ((yu*widthG + xu) << 2);
      dotProductTemp2 = m_dotProduct2 + offsetPos + ((yu*widthG + xu) << 2);
      dotProductTemp3 = m_dotProduct3 + offsetPos + ((yu*widthG + xu) << 2);
      dotProductTemp5 = m_dotProduct5 + offsetPos + ((yu*widthG + xu) << 2);
      dotProductTemp6 = m_dotProduct6 + offsetPos + ((yu*widthG + xu) << 2);

      xCalcBlkGradient(xu << 2, yu << 2, dotProductTemp1, dotProductTemp2, dotProductTemp3, dotProductTemp5, dotProductTemp6, sGx2, sGy2, sGxGy, sGxdI, sGydI, widthG, heightG, (1 << 2));

      if (sGx2 > 0)
      {
        tmpx = rightShiftMSB(sGxdI << 3, sGx2);
        tmpx = Clip3(-limit, limit, tmpx);
      }
      if (sGy2 > 0)
      {
        int     mainsGxGy = sGxGy >> 12;
        int     secsGxGy = sGxGy & ((1 << 12) - 1);
        int     tmpData = tmpx * mainsGxGy;
        tmpData = ((tmpData << 12) + tmpx*secsGxGy) >> 1;
        tmpy = rightShiftMSB(((sGydI << 3) - tmpData), sGy2);
        tmpy = Clip3(-limit, limit, tmpy);
      }

      srcY0Temp = srcY0 + (stridePredMC + 1) + ((yu*src0Stride + xu) << 2);
      srcY1Temp = srcY1 + (stridePredMC + 1) + ((yu*src0Stride + xu) << 2);
      gradX0 = m_gradX0 + offsetPos + ((yu*widthG + xu) << 2);
      gradX1 = m_gradX1 + offsetPos + ((yu*widthG + xu) << 2);
      gradY0 = m_gradY0 + offsetPos + ((yu*widthG + xu) << 2);
      gradY1 = m_gradY1 + offsetPos + ((yu*widthG + xu) << 2);

      dstY0 = dstY + ((yu*dstStride + xu) << 2);
      xAddBIOAvg4(srcY0Temp, src0Stride, srcY1Temp, src1Stride, dstY0, dstStride, gradX0, gradX1, gradY0, gradY1, widthG, (1 << 2), (1 << 2), (int)tmpx, (int)tmpy, shiftNum, offset, clpRng);
    }  // xu
  }  // yu
}

void InterPrediction::bioSampleExtendBilinearFilter(Pel const* src, int srcStride, Pel *dst, int dstStride, int width, int height, int dim, int fracX, int fracY, bool isLast, const ChromaFormat fmt, const ClpRng& clpRng)
{
  Pel const* pSrc = NULL;
  Pel*       pDst = NULL;

  int vFilterSize = NTAPS_BILINEAR;
  int widthTmp = 0;
  int heightTmp = 0;

  for (int cand = 0; cand < 4; cand++)  // top, left, bottom and right
  {

    if (cand == 0)  // top
    {
      pSrc = src;
      pDst = dst;
      widthTmp = width;
      heightTmp = dim;
    }
    else if (cand == 1)  // left
    {
      pSrc = src + dim*srcStride;
      pDst = dst + dim*dstStride;
      widthTmp = dim;
      heightTmp = height - 2 * dim;
    }
    else if (cand == 2)  // bottom
    {
      pSrc = src + (height - dim)*srcStride;
      pDst = dst + (height - dim)*dstStride;
      widthTmp = width;
      heightTmp = dim;
    }
    else if (cand == 3)  // right
    {
      pSrc = src + dim*srcStride + width - dim;
      pDst = dst + dim*dstStride + width - dim;
      widthTmp = dim;
      heightTmp = height - 2 * dim;
    }

    if (fracY == 0)
    {
      m_if.filterHor(COMPONENT_Y, pSrc, srcStride, pDst, dstStride, widthTmp, heightTmp, fracX, isLast, fmt, clpRng, 1);
    }
    else if (fracX == 0)
    {
      m_if.filterVer(COMPONENT_Y, pSrc, srcStride, pDst, dstStride, widthTmp, heightTmp, fracY, true, isLast, fmt, clpRng, 1);
    }
    else
    {
      PelBuf tmpBuf = PelBuf(m_filteredBlockTmp[0][COMPONENT_Y], Size(width, height));
      tmpBuf.stride = width;

      m_if.filterHor(COMPONENT_Y, pSrc - ((vFilterSize >> 1) - 1) * srcStride, srcStride, tmpBuf.buf, tmpBuf.stride, widthTmp, heightTmp + vFilterSize - 1, fracX, false, fmt, clpRng, 1);
      JVET_J0090_SET_CACHE_ENABLE( false );
      m_if.filterVer(COMPONENT_Y, tmpBuf.buf + ((vFilterSize >> 1) - 1) * tmpBuf.stride, tmpBuf.stride, pDst, dstStride, widthTmp, heightTmp, fracY, false, isLast, fmt, clpRng, 1);
      JVET_J0090_SET_CACHE_ENABLE( true );
    }
  }
}

bool InterPrediction::xCalcBiPredSubBlkDist(const PredictionUnit &pu, const Pel* pYuvSrc0, const int src0Stride, const Pel* pYuvSrc1, const int src1Stride, const BitDepths &clipBitDepths)
{
  const int     width = pu.lwidth();
  const int     height = pu.lheight();
  const int     clipbd = clipBitDepths.recon[toChannelType(COMPONENT_Y)];
  const uint32_t distortionShift = DISTORTION_PRECISION_ADJUSTMENT(clipbd);
  const int     shift = std::max<int>(2, (IF_INTERNAL_PREC - clipbd));
  const int     xUnit = (width >> 2);
  const int     yUnit = (height >> 2);

  m_bioDistThres = (shift <= 5) ? (((32 << (clipbd - 8))*width*height) >> (5 - shift)) : (((32 << (clipbd - 8))*width*height) << (shift - 5));
  m_bioSubBlkDistThres = (shift <= 5) ? (((64 << (clipbd - 8)) << 4) >> (5 - shift)) : (((64 << (clipbd - 8)) << 4) << (shift - 5));

  m_bioDistThres >>= distortionShift;
  m_bioSubBlkDistThres >>= distortionShift;

  DistParam cDistParam;
  Distortion dist = 0;
  for (int yu = 0, blkIdx = 0; yu < yUnit; yu++)
  {
    for (int xu = 0; xu < xUnit; xu++, blkIdx++)
    {
      const Pel* pPred0 = pYuvSrc0 + ((yu*src0Stride + xu) << 2);
      const Pel* pPred1 = pYuvSrc1 + ((yu*src1Stride + xu) << 2);

      m_pcRdCost->setDistParam(cDistParam, pPred0, pPred1, src0Stride, src1Stride, clipbd, COMPONENT_Y, (1 << 2), (1 << 2), 0, 1, false, true);
      m_bioPredSubBlkDist[blkIdx] = cDistParam.distFunc(cDistParam);
      dist += m_bioPredSubBlkDist[blkIdx];
    }
  }

  return (dist >= m_bioDistThres);
}

void InterPrediction::xAddBIOAvg4(const Pel* src0, int src0Stride, const Pel* src1, int src1Stride, Pel *dst, int dstStride, const Pel *gradX0, const Pel *gradX1, const Pel *gradY0, const Pel*gradY1, int gradStride, int width, int height, int tmpx, int tmpy, int shift, int offset, const ClpRng& clpRng)
{
#if ENABLE_SIMD_OPT_BIO
  g_pelBufOP.addBIOAvg4(src0, src0Stride, src1, src1Stride, dst, dstStride, gradX0, gradX1, gradY0, gradY1, gradStride, width, height, tmpx, tmpy, shift, offset, clpRng);
#else
  int b = 0;

  for (int y = 0; y < height; y++)
  {
    for (int x = 0; x < width; x += 4)
    {
      b = tmpx * (gradX0[x] - gradX1[x]) + tmpy * (gradY0[x] - gradY1[x]);
      b = ((b + 1) >> 1);
      dst[x] = ClipPel((int16_t)rightShift((src0[x] + src1[x] + b + offset), shift), clpRng);

      b = tmpx * (gradX0[x + 1] - gradX1[x + 1]) + tmpy * (gradY0[x + 1] - gradY1[x + 1]);
      b = ((b + 1) >> 1);
      dst[x + 1] = ClipPel((int16_t)rightShift((src0[x + 1] + src1[x + 1] + b + offset), shift), clpRng);

      b = tmpx * (gradX0[x + 2] - gradX1[x + 2]) + tmpy * (gradY0[x + 2] - gradY1[x + 2]);
      b = ((b + 1) >> 1);
      dst[x + 2] = ClipPel((int16_t)rightShift((src0[x + 2] + src1[x + 2] + b + offset), shift), clpRng);

      b = tmpx * (gradX0[x + 3] - gradX1[x + 3]) + tmpy * (gradY0[x + 3] - gradY1[x + 3]);
      b = ((b + 1) >> 1);
      dst[x + 3] = ClipPel((int16_t)rightShift((src0[x + 3] + src1[x + 3] + b + offset), shift), clpRng);
    }
    dst += dstStride;       src0 += src0Stride;     src1 += src1Stride;
    gradX0 += gradStride; gradX1 += gradStride; gradY0 += gradStride; gradY1 += gradStride;
  }
#endif
}

void InterPrediction::xBioGradFilter(Pel* pSrc, int srcStride, int width, int height, int gradStride, Pel* gradX, Pel* gradY)
{
#if ENABLE_SIMD_OPT_BIO
  g_pelBufOP.bioGradFilter(pSrc, srcStride, width, height, gradStride, gradX, gradY);
#else
  Pel* srcTmp = pSrc + srcStride + 1;
  Pel* gradXTmp = gradX + gradStride + 1;
  Pel* gradYTmp = gradY + gradStride + 1;

  for (int y = 0; y < (height - 2 * BIO_EXTEND_SIZE); y++)
  {
    for (int x = 0; x < (width - 2 * BIO_EXTEND_SIZE); x++)
    {
      gradYTmp[x] = (srcTmp[x + srcStride] - srcTmp[x - srcStride]) >> 4;
      gradXTmp[x] = (srcTmp[x + 1] - srcTmp[x - 1]) >> 4;
    }
    gradXTmp += gradStride;
    gradYTmp += gradStride;
    srcTmp += srcStride;
  }

  gradXTmp = gradX + gradStride + 1;
  gradYTmp = gradY + gradStride + 1;
  for (int y = 0; y < (height - 2 * BIO_EXTEND_SIZE); y++)
  {
    gradXTmp[-1] = gradXTmp[0];
    gradXTmp[width - 2 * BIO_EXTEND_SIZE] = gradXTmp[width - 2 * BIO_EXTEND_SIZE - 1];
    gradXTmp += gradStride;

    gradYTmp[-1] = gradYTmp[0];
    gradYTmp[width - 2 * BIO_EXTEND_SIZE] = gradYTmp[width - 2 * BIO_EXTEND_SIZE - 1];
    gradYTmp += gradStride;
  }

  gradXTmp = gradX + gradStride;
  gradYTmp = gradY + gradStride;
  ::memcpy(gradXTmp - gradStride, gradXTmp, sizeof(Pel)*(width));
  ::memcpy(gradXTmp + (height - 2 * BIO_EXTEND_SIZE)*gradStride, gradXTmp + (height - 2 * BIO_EXTEND_SIZE - 1)*gradStride, sizeof(Pel)*(width));
  ::memcpy(gradYTmp - gradStride, gradYTmp, sizeof(Pel)*(width));
  ::memcpy(gradYTmp + (height - 2 * BIO_EXTEND_SIZE)*gradStride, gradYTmp + (height - 2 * BIO_EXTEND_SIZE - 1)*gradStride, sizeof(Pel)*(width));
#endif
}

void InterPrediction::xCalcBIOPar(const Pel* srcY0Temp, const Pel* srcY1Temp, const Pel* gradX0, const Pel* gradX1, const Pel* gradY0, const Pel* gradY1, int* dotProductTemp1, int* dotProductTemp2, int* dotProductTemp3, int* dotProductTemp5, int* dotProductTemp6, const int src0Stride, const int src1Stride, const int gradStride, const int widthG, const int heightG)
{
#if ENABLE_SIMD_OPT_BIO 
  g_pelBufOP.calcBIOPar(srcY0Temp, srcY1Temp, gradX0, gradX1, gradY0, gradY1, dotProductTemp1, dotProductTemp2, dotProductTemp3, dotProductTemp5, dotProductTemp6, src0Stride, src1Stride, gradStride, widthG, heightG);
#else
  for (int y = 0; y < heightG; y++)
  {
    for (int x = 0; x < widthG; x++)
    {
      int temp = (srcY0Temp[x] >> 6) - (srcY1Temp[x] >> 6);
      int tempX = (gradX0[x] + gradX1[x]) >> 3;
      int tempY = (gradY0[x] + gradY1[x]) >> 3;
      dotProductTemp1[x] = tempX * tempX;
      dotProductTemp2[x] = tempX * tempY;
      dotProductTemp3[x] = -tempX * temp;
      dotProductTemp5[x] = tempY * tempY;
      dotProductTemp6[x] = -tempY * temp;
    }
    srcY0Temp += src0Stride;
    srcY1Temp += src1Stride;
    gradX0 += gradStride;
    gradX1 += gradStride;
    gradY0 += gradStride;
    gradY1 += gradStride;
    dotProductTemp1 += widthG;
    dotProductTemp2 += widthG;
    dotProductTemp3 += widthG;
    dotProductTemp5 += widthG;
    dotProductTemp6 += widthG;
  }
#endif
}

void InterPrediction::xCalcBlkGradient(int sx, int sy, int    *arraysGx2, int     *arraysGxGy, int     *arraysGxdI, int     *arraysGy2, int     *arraysGydI, int     &sGx2, int     &sGy2, int     &sGxGy, int     &sGxdI, int     &sGydI, int width, int height, int unitSize)
{
#if ENABLE_SIMD_OPT_BIO
  g_pelBufOP.calcBlkGradient(sx, sy, arraysGx2, arraysGxGy, arraysGxdI, arraysGy2, arraysGydI, sGx2, sGy2, sGxGy, sGxdI, sGydI, width, height, unitSize);
#else
  int     *Gx2 = arraysGx2;
  int     *Gy2 = arraysGy2;
  int     *GxGy = arraysGxGy;
  int     *GxdI = arraysGxdI;
  int     *GydI = arraysGydI;

  // set to the above row due to JVET_K0485_BIO_EXTEND_SIZE
  Gx2 -= (BIO_EXTEND_SIZE*width);
  Gy2 -= (BIO_EXTEND_SIZE*width);
  GxGy -= (BIO_EXTEND_SIZE*width);
  GxdI -= (BIO_EXTEND_SIZE*width);
  GydI -= (BIO_EXTEND_SIZE*width);

  for (int y = -BIO_EXTEND_SIZE; y < unitSize + BIO_EXTEND_SIZE; y++)
  {
    for (int x = -BIO_EXTEND_SIZE; x < unitSize + BIO_EXTEND_SIZE; x++)
    {
      sGx2 += Gx2[x];
      sGy2 += Gy2[x];
      sGxGy += GxGy[x];
      sGxdI += GxdI[x];
      sGydI += GydI[x];
    }
    Gx2 += width;
    Gy2 += width;
    GxGy += width;
    GxdI += width;
    GydI += width;
  }
#endif
}
#endif

#if JVET_L0256_BIO
void InterPrediction::xWeightedAverage(const PredictionUnit& pu, const CPelUnitBuf& pcYuvSrc0, const CPelUnitBuf& pcYuvSrc1, PelUnitBuf& pcYuvDst, const BitDepths& clipBitDepths, const ClpRngs& clpRngs, const bool& bioApplied )
#else
void InterPrediction::xWeightedAverage( const PredictionUnit& pu, const CPelUnitBuf& pcYuvSrc0, const CPelUnitBuf& pcYuvSrc1, PelUnitBuf& pcYuvDst, const BitDepths& clipBitDepths, const ClpRngs& clpRngs )
#endif
{
  const int iRefIdx0 = pu.refIdx[0];
  const int iRefIdx1 = pu.refIdx[1];

  if( iRefIdx0 >= 0 && iRefIdx1 >= 0 )
  {
#if JVET_L0646_GBI
    if( pu.cu->GBiIdx != GBI_DEFAULT )
    {
#if JVET_L0256_BIO
      CHECK(bioApplied, "GBi is disallowed with BIO");
#endif
      pcYuvDst.addWeightedAvg(pcYuvSrc0, pcYuvSrc1, clpRngs, pu.cu->GBiIdx);
      return;
    }
#endif
#if JVET_L0256_BIO
    if (bioApplied)
    {
      const int  src0Stride = pu.lwidth() + 2 * BIO_EXTEND_SIZE + 2;
      const int  src1Stride = pu.lwidth() + 2 * BIO_EXTEND_SIZE + 2;
      const Pel* pSrcY0 = m_filteredBlockTmp[2][COMPONENT_Y] + 2 * src0Stride + 2;
      const Pel* pSrcY1 = m_filteredBlockTmp[3][COMPONENT_Y] + 2 * src1Stride + 2;

      bool bioEnabled = xCalcBiPredSubBlkDist(pu, pSrcY0, src0Stride, pSrcY1, src1Stride, clipBitDepths);
      if (bioEnabled)
      {
        applyBiOptFlow(pu, pcYuvSrc0, pcYuvSrc1, iRefIdx0, iRefIdx1, pcYuvDst, clipBitDepths);
      }
      else
      {
        pcYuvDst.bufs[0].addAvg(CPelBuf(pSrcY0, src0Stride, pu.lumaSize()), CPelBuf(pSrcY1, src1Stride, pu.lumaSize()), clpRngs.comp[0]);
      }
    }
    pcYuvDst.addAvg(pcYuvSrc0, pcYuvSrc1, clpRngs, bioApplied);
#else
    pcYuvDst.addAvg( pcYuvSrc0, pcYuvSrc1, clpRngs );
#endif
  }
  else if( iRefIdx0 >= 0 && iRefIdx1 < 0 )
  {
#if JVET_L0124_L0208_TRIANGLE
    if( pu.cu->triangle )
    {
      pcYuvDst.copyFrom( pcYuvSrc0 );
    }
    else
#endif 
    pcYuvDst.copyClip( pcYuvSrc0, clpRngs );
  }
  else if( iRefIdx0 < 0 && iRefIdx1 >= 0 )
  {
#if JVET_L0124_L0208_TRIANGLE
    if( pu.cu->triangle )
    {
      pcYuvDst.copyFrom( pcYuvSrc1 );
    }
    else
#endif
    pcYuvDst.copyClip( pcYuvSrc1, clpRngs );
  }
}

void InterPrediction::motionCompensation( PredictionUnit &pu, PelUnitBuf &predBuf, const RefPicList &eRefPicList 
#if JVET_L0293_CPR
  , const bool luma, const bool chroma
#endif
)
{
#if JVET_L0293_CPR
  // dual tree handling for CPR as the only ref
  if (!luma || !chroma)
  {
    if (!luma && chroma)
    {
      xChromaMC(pu, predBuf);
      return;
    }
    else // (luma && !chroma)
    {
      xPredInterUni(pu, eRefPicList, predBuf, false
#if JVET_L0256_BIO
        , false
#endif
        , luma, chroma);
      return;
    }
  }
  // else, go with regular MC below
#endif
        CodingStructure &cs = *pu.cs;
  const PPS &pps            = *cs.pps;
  const SliceType sliceType =  cs.slice->getSliceType();

  if( eRefPicList != REF_PIC_LIST_X )
  {
    if( ( ( sliceType == P_SLICE && pps.getUseWP() ) || ( sliceType == B_SLICE && pps.getWPBiPred() ) ) )
    {
      xPredInterUni         ( pu,          eRefPicList, predBuf, true 
#if JVET_L0256_BIO
        , false
#endif
#if JVET_L0293_CPR     
        , true, true
#endif
      );
      xWeightedPredictionUni( pu, predBuf, eRefPicList, predBuf, -1, m_maxCompIDToPred );
    }
    else
    {
      xPredInterUni( pu, eRefPicList, predBuf, false 
#if JVET_L0256_BIO
        , false
#endif
#if JVET_L0293_CPR     
        , true, true
#endif
      );
    }
  }
  else
  {
#if JVET_L0293_CPR
    if (pu.mergeType != MRG_TYPE_DEFAULT_N && pu.mergeType != MRG_TYPE_CPR)
#else
    if( pu.mergeType != MRG_TYPE_DEFAULT_N )
#endif
    {
      xSubPuMC( pu, predBuf, eRefPicList );
    }
    else if( xCheckIdenticalMotion( pu ) )
    {
      xPredInterUni( pu, REF_PIC_LIST_0, predBuf, false 
#if JVET_L0256_BIO
        , false
#endif
#if JVET_L0293_CPR     
        , true, true
#endif
      );
    }
    else
    {
      xPredInterBi( pu, predBuf );
    }
  }
  return;
}

void InterPrediction::motionCompensation( CodingUnit &cu, const RefPicList &eRefPicList 
#if JVET_L0293_CPR
  , const bool luma, const bool chroma
#endif
)
{
  for( auto &pu : CU::traversePUs( cu ) )
  {
    PelUnitBuf predBuf = cu.cs->getPredBuf( pu );
    motionCompensation( pu, predBuf, eRefPicList 
#if JVET_L0293_CPR
      , luma, chroma
#endif
    );
  }
}

void InterPrediction::motionCompensation( PredictionUnit &pu, const RefPicList &eRefPicList /*= REF_PIC_LIST_X*/ 
#if JVET_L0293_CPR
  , const bool luma, const bool chroma
#endif
)
{
  PelUnitBuf predBuf = pu.cs->getPredBuf( pu );
  motionCompensation( pu, predBuf, eRefPicList 
#if JVET_L0293_CPR
    , luma, chroma
#endif
  );
}

#if JVET_L0256_BIO
int InterPrediction::rightShiftMSB(int numer, int denom)
{
  int     d;
  int msbIdx = 0;
  for (msbIdx = 0; msbIdx<32; msbIdx++)
  {
    if (denom < ((int)1 << msbIdx))
    {
      break;
    }
  }

  int shiftIdx = msbIdx - 1;
  d = (numer >> shiftIdx);

  return d;
}
#endif

#if JVET_L0124_L0208_TRIANGLE
void InterPrediction::motionCompensation4Triangle( CodingUnit &cu, MergeCtx &triangleMrgCtx, const bool splitDir, const uint8_t candIdx0, const uint8_t candIdx1 )
{
  for( auto &pu : CU::traversePUs( cu ) )
  {
    const UnitArea localUnitArea( cu.cs->area.chromaFormat, Area( 0, 0, pu.lwidth(), pu.lheight() ) );
    PelUnitBuf tmpTriangleBuf = m_triangleBuf.getBuf( localUnitArea );
    PelUnitBuf predBuf        = cu.cs->getPredBuf( pu );
     
    triangleMrgCtx.setMergeInfo( pu, candIdx0 );
    PU::spanMotionInfo( pu );
    motionCompensation( pu, tmpTriangleBuf );
   
    triangleMrgCtx.setMergeInfo( pu, candIdx1 );
    PU::spanMotionInfo( pu );
    motionCompensation( pu, predBuf );

    weightedTriangleBlk( pu, PU::getTriangleWeights(pu, triangleMrgCtx, candIdx0, candIdx1), splitDir, MAX_NUM_CHANNEL_TYPE, predBuf, tmpTriangleBuf, predBuf );
  }
}

void InterPrediction::weightedTriangleBlk( PredictionUnit &pu, bool weights, const bool splitDir, int32_t channel, PelUnitBuf& predDst, PelUnitBuf& predSrc0, PelUnitBuf& predSrc1 )
{
  if( channel == CHANNEL_TYPE_LUMA )
  {
    xWeightedTriangleBlk( pu, pu.lumaSize().width, pu.lumaSize().height, COMPONENT_Y, splitDir, weights, predDst, predSrc0, predSrc1 );
  }
  else if( channel == CHANNEL_TYPE_CHROMA )
  {
    xWeightedTriangleBlk( pu, pu.chromaSize().width, pu.chromaSize().height, COMPONENT_Cb, splitDir, weights, predDst, predSrc0, predSrc1 );
    xWeightedTriangleBlk( pu, pu.chromaSize().width, pu.chromaSize().height, COMPONENT_Cr, splitDir, weights, predDst, predSrc0, predSrc1 );
  }
  else
  {
    xWeightedTriangleBlk( pu, pu.lumaSize().width,   pu.lumaSize().height,   COMPONENT_Y,  splitDir, weights, predDst, predSrc0, predSrc1 );
    xWeightedTriangleBlk( pu, pu.chromaSize().width, pu.chromaSize().height, COMPONENT_Cb, splitDir, weights, predDst, predSrc0, predSrc1 );
    xWeightedTriangleBlk( pu, pu.chromaSize().width, pu.chromaSize().height, COMPONENT_Cr, splitDir, weights, predDst, predSrc0, predSrc1 );
  }
}

void InterPrediction::xWeightedTriangleBlk( const PredictionUnit &pu, const uint32_t width, const uint32_t height, const ComponentID compIdx, const bool splitDir, const bool weights, PelUnitBuf& predDst, PelUnitBuf& predSrc0, PelUnitBuf& predSrc1 )
{
  Pel*    dst        = predDst .get(compIdx).buf;
  Pel*    src0       = predSrc0.get(compIdx).buf;
  Pel*    src1       = predSrc1.get(compIdx).buf;
  int32_t strideDst  = predDst .get(compIdx).stride  - width;
  int32_t strideSrc0 = predSrc0.get(compIdx).stride  - width;
  int32_t strideSrc1 = predSrc1.get(compIdx).stride  - width;

  const char    log2WeightBase    = 3;
  const ClpRng  clipRng           = pu.cu->slice->clpRngs().comp[compIdx];
  const int32_t clipbd            = clipRng.bd;
  const int32_t shiftDefault      = std::max<int>(2, (IF_INTERNAL_PREC - clipbd));
  const int32_t offsetDefault     = (1<<(shiftDefault-1)) + IF_INTERNAL_OFFS;
  const int32_t shiftWeighted     = std::max<int>(2, (IF_INTERNAL_PREC - clipbd)) + log2WeightBase;
  const int32_t offsetWeighted    = (1 << (shiftWeighted - 1)) + (IF_INTERNAL_OFFS << log2WeightBase);
                                  
  const int32_t ratioWH           = (width > height) ? (width / height) : 1;
  const int32_t ratioHW           = (width > height) ? 1 : (height / width);
  const Pel*    pelWeighted       = (compIdx == COMPONENT_Y) ? g_trianglePelWeightedLuma[splitDir][weights] : g_trianglePelWeightedChroma[predDst.chromaFormat == CHROMA_444 ? 0 : 1][splitDir][weights];
  const int32_t weightedLength    = (compIdx == COMPONENT_Y) ? g_triangleWeightLengthLuma[weights] : g_triangleWeightLengthChroma[predDst.chromaFormat == CHROMA_444 ? 0 : 1][weights];
        int32_t weightedStartPos  = ( splitDir == 0 ) ? ( 0 - (weightedLength >> 1) * ratioWH ) : ( width - ((weightedLength + 1) >> 1) * ratioWH );
        int32_t weightedEndPos    = weightedStartPos + weightedLength * ratioWH - 1;
        int32_t weightedPosoffset =( splitDir == 0 ) ? ratioWH : -ratioWH;
  
  const Pel*    tmpPelWeighted;
        int32_t x, y, tmpX, tmpY, tmpWeightedStart, tmpWeightedEnd;
  
  for( y = 0; y < height; y+= ratioHW )
  {
    for( tmpY = ratioHW; tmpY > 0; tmpY-- )
    {
      for( x = 0; x < weightedStartPos; x++ )
      {
        *dst++ = ClipPel( rightShift( (splitDir == 0 ? *src1 : *src0) + offsetDefault, shiftDefault), clipRng );
        src0++;
        src1++;
      }

      tmpWeightedStart = std::max((int32_t)0, weightedStartPos);
      tmpWeightedEnd   = std::min(weightedEndPos, (int32_t)(width - 1));
      tmpPelWeighted   = pelWeighted;
      if( weightedStartPos < 0 )
      {
        tmpPelWeighted += abs(weightedStartPos) / ratioWH;
      }
      for( x = tmpWeightedStart; x <= tmpWeightedEnd; x+= ratioWH )
      {
        for( tmpX = ratioWH; tmpX > 0; tmpX-- )
        {
          *dst++ = ClipPel( rightShift( ((*tmpPelWeighted)*(*src0++) + ((8 - (*tmpPelWeighted)) * (*src1++)) + offsetWeighted), shiftWeighted ), clipRng );
        }
        tmpPelWeighted++;
      }

      for( x = weightedEndPos + 1; x < width; x++ )
      {
        *dst++ = ClipPel( rightShift( (splitDir == 0 ? *src0 : *src1) + offsetDefault, shiftDefault ), clipRng );
        src0++;
        src1++;
      }

      dst  += strideDst;
      src0 += strideSrc0;
      src1 += strideSrc1;
    }
    weightedStartPos += weightedPosoffset;
    weightedEndPos   += weightedPosoffset;
  }
}
#endif

#if JVET_J0090_MEMORY_BANDWITH_MEASURE
void InterPrediction::cacheAssign( CacheModel *cache )
{
  m_cacheModel = cache;
  m_if.cacheAssign( cache );
  m_if.initInterpolationFilter( !cache->isCacheEnable() );
}
#endif

//! \}