UnitTools.cpp

  // get Inter Dir
  interDirNeighbours = puFirstNeighbour->getMotionInfo().interDir;

  pu.cu->affineType = puFirstNeighbour->cu->affineType;

  // derive Mv from neighbor affine block
  Mv cMv[3];
  if ( interDirNeighbours != 2 )
  {
    xInheritedAffineMv( pu, puFirstNeighbour, REF_PIC_LIST_0, cMv );
    for ( int mvNum = 0; mvNum < 3; mvNum++ )
    {
      mvFieldNeighbours[0][mvNum].setMvField( cMv[mvNum], puFirstNeighbour->refIdx[0] );
    }
  }

  if ( pu.cs->slice->isInterB() )
  {
    if ( interDirNeighbours != 1 )
    {
      xInheritedAffineMv( pu, puFirstNeighbour, REF_PIC_LIST_1, cMv );
      for ( int mvNum = 0; mvNum < 3; mvNum++ )
      {
        mvFieldNeighbours[1][mvNum].setMvField( cMv[mvNum], puFirstNeighbour->refIdx[1] );
      }
    }
  }
}

void PU::setAllAffineMvField( PredictionUnit &pu, MvField *mvField, RefPicList eRefList )
{
  // Set Mv
  Mv mv[3];
  for ( int i = 0; i < 3; i++ )
  {
    mv[i] = mvField[i].mv;
  }
  setAllAffineMv( pu, mv[0], mv[1], mv[2], eRefList );

  // Set RefIdx
  CHECK( mvField[0].refIdx != mvField[1].refIdx || mvField[0].refIdx != mvField[2].refIdx, "Affine mv corners don't have the same refIdx." );
  pu.refIdx[eRefList] = mvField[0].refIdx;
}

void PU::setAllAffineMv( PredictionUnit& pu, Mv affLT, Mv affRT, Mv affLB, RefPicList eRefList 
#if REMOVE_MV_ADAPT_PREC
  , bool setHighPrec
#endif
)
{
  int width  = pu.Y().width;
  int shift = MAX_CU_DEPTH;
#if REMOVE_MV_ADAPT_PREC
  if (setHighPrec)
  {
    affLT.hor = affLT.hor << VCEG_AZ07_MV_ADD_PRECISION_BIT_FOR_STORE;
    affLT.ver = affLT.ver << VCEG_AZ07_MV_ADD_PRECISION_BIT_FOR_STORE;
    affRT.hor = affRT.hor << VCEG_AZ07_MV_ADD_PRECISION_BIT_FOR_STORE;
    affRT.ver = affRT.ver << VCEG_AZ07_MV_ADD_PRECISION_BIT_FOR_STORE;
    affLB.hor = affLB.hor << VCEG_AZ07_MV_ADD_PRECISION_BIT_FOR_STORE;
    affLB.ver = affLB.ver << VCEG_AZ07_MV_ADD_PRECISION_BIT_FOR_STORE;
  }
#else
  affLT.setHighPrec();
  affRT.setHighPrec();
  affLB.setHighPrec();
#endif
  int deltaMvHorX, deltaMvHorY, deltaMvVerX, deltaMvVerY;
  deltaMvHorX = (affRT - affLT).getHor() << (shift - g_aucLog2[width]);
  deltaMvHorY = (affRT - affLT).getVer() << (shift - g_aucLog2[width]);
  int height = pu.Y().height;
  if ( pu.cu->affineType == AFFINEMODEL_6PARAM )
  {
    deltaMvVerX = (affLB - affLT).getHor() << (shift - g_aucLog2[height]);
    deltaMvVerY = (affLB - affLT).getVer() << (shift - g_aucLog2[height]);
  }
  else
  {
    deltaMvVerX = -deltaMvHorY;
    deltaMvVerY = deltaMvHorX;
  }

  int mvScaleHor = affLT.getHor() << shift;
  int mvScaleVer = affLT.getVer() << shift;

  int blockWidth = AFFINE_MIN_BLOCK_SIZE;
  int blockHeight = AFFINE_MIN_BLOCK_SIZE;
  const int halfBW = blockWidth >> 1;
  const int halfBH = blockHeight >> 1;

  MotionBuf mb = pu.getMotionBuf();
  int mvScaleTmpHor, mvScaleTmpVer;
  for ( int h = 0; h < pu.Y().height; h += blockHeight )
  {
    for ( int w = 0; w < pu.Y().width; w += blockWidth )
    {
      mvScaleTmpHor = mvScaleHor + deltaMvHorX * (halfBW + w) + deltaMvVerX * (halfBH + h);
      mvScaleTmpVer = mvScaleVer + deltaMvHorY * (halfBW + w) + deltaMvVerY * (halfBH + h);
      roundAffineMv( mvScaleTmpHor, mvScaleTmpVer, shift );

      for ( int y = (h >> MIN_CU_LOG2); y < ((h + blockHeight) >> MIN_CU_LOG2); y++ )
      {
        for ( int x = (w >> MIN_CU_LOG2); x < ((w + blockHeight) >> MIN_CU_LOG2); x++ )
        {
#if REMOVE_MV_ADAPT_PREC
          mb.at(x, y).mv[eRefList].hor = mvScaleTmpHor;
          mb.at(x, y).mv[eRefList].ver = mvScaleTmpVer;
#else
          mb.at(x, y).mv[eRefList] = Mv(mvScaleTmpHor, mvScaleTmpVer, true);
#endif
        }
      }
    }
  }

  // Set AffineMvField for affine motion compensation LT, RT, LB and RB
  mb.at(            0,             0 ).mv[eRefList] = affLT;
  mb.at( mb.width - 1,             0 ).mv[eRefList] = affRT;

  if ( pu.cu->affineType == AFFINEMODEL_6PARAM )
  {
    mb.at( 0, mb.height - 1 ).mv[eRefList] = affLB;
  }
}

static bool deriveScaledMotionTemporal( const Slice&      slice,
                                        const Position&   colPos,
                                        const Picture*    pColPic,
                                        const RefPicList  eCurrRefPicList,
                                        Mv&         cColMv,
                                        const RefPicList  eFetchRefPicList)
{
  const MotionInfo &mi = pColPic->cs->getMotionInfo(colPos);
  const Slice *pColSlice = nullptr;

  for (const auto &pSlice : pColPic->slices)
  {
    if (pSlice->getIndependentSliceIdx() == mi.sliceIdx)
    {
      pColSlice = pSlice;
      break;
    }
  }

  CHECK(pColSlice == nullptr, "Couldn't find the colocated slice");

  int iColPOC, iColRefPOC, iCurrPOC, iCurrRefPOC, iScale;
  bool bAllowMirrorMV = true;
  RefPicList eColRefPicList = slice.getCheckLDC() ? eCurrRefPicList : RefPicList(1 - eFetchRefPicList);
  if (pColPic == slice.getRefPic(RefPicList(slice.isInterB() ? 1 - slice.getColFromL0Flag() : 0), slice.getColRefIdx()))
  {
    eColRefPicList = eCurrRefPicList;   //67 -> disable, 64 -> enable
    bAllowMirrorMV = false;
  }

  // Although it might make sense to keep the unavailable motion field per direction still be unavailable, I made the MV prediction the same way as in TMVP
  // So there is an interaction between MV0 and MV1 of the corresponding blocks identified by TV.

  // Grab motion and do necessary scaling.{{
  iCurrPOC = slice.getPOC();

  int iColRefIdx = mi.refIdx[eColRefPicList];

  if (iColRefIdx < 0 && (slice.getCheckLDC() || bAllowMirrorMV))
  {
    eColRefPicList = RefPicList(1 - eColRefPicList);
    iColRefIdx = mi.refIdx[eColRefPicList];

    if (iColRefIdx < 0)
    {
      return false;
    }
  }

  if (iColRefIdx >= 0 && slice.getNumRefIdx(eCurrRefPicList) > 0)
  {
    iColPOC = pColSlice->getPOC();
    iColRefPOC = pColSlice->getRefPOC(eColRefPicList, iColRefIdx);
    ///////////////////////////////////////////////////////////////
    // Set the target reference index to 0, may be changed later //
    ///////////////////////////////////////////////////////////////
    iCurrRefPOC = slice.getRefPic(eCurrRefPicList, 0)->getPOC();
    // Scale the vector.
    cColMv = mi.mv[eColRefPicList];
    //pcMvFieldSP[2*iPartition + eCurrRefPicList].getMv();
    // Assume always short-term for now
    iScale = xGetDistScaleFactor(iCurrPOC, iCurrRefPOC, iColPOC, iColRefPOC);

    if (iScale != 4096)
    {
#if !REMOVE_MV_ADAPT_PREC
      if (slice.getSPS()->getSpsNext().getUseHighPrecMv())
      {
        cColMv.setHighPrec();
      }
#endif

      cColMv = cColMv.scaleMv(iScale);
    }

    return true;
  }
  return false;
}

void clipColBlkMv(int& mvX, int& mvY, const PredictionUnit& pu)
{
  Position puPos = pu.lumaPos();
  Size     puSize = pu.lumaSize();

  int ctuSize = pu.cs->sps->getSpsNext().getCTUSize();
  int ctuX = puPos.x / ctuSize*ctuSize;
  int ctuY = puPos.y / ctuSize*ctuSize;

  int horMax = std::min((int)pu.cs->sps->getPicWidthInLumaSamples(), ctuX + ctuSize + 4) - puSize.width;
  int horMin = std::max((int)0, ctuX);
  int verMax = std::min((int)pu.cs->sps->getPicHeightInLumaSamples(), ctuY + ctuSize) - puSize.height;
  int verMin = std::min((int)0, ctuY);

  horMax = horMax - puPos.x;
  horMin = horMin - puPos.x;
  verMax = verMax - puPos.y;
  verMin = verMin - puPos.y;

  mvX = std::min(horMax, std::max(horMin, mvX));
  mvY = std::min(verMax, std::max(verMin, mvY));
}

bool PU::getInterMergeSubPuMvpCand(const PredictionUnit &pu, MergeCtx& mrgCtx, bool& LICFlag, const int count
)
{
  const Slice   &slice = *pu.cs->slice;
  const unsigned scale = 4 * std::max<int>(1, 4 * AMVP_DECIMATION_FACTOR / 4);
  const unsigned mask = ~(scale - 1);

  const Picture *pColPic = slice.getRefPic(RefPicList(slice.isInterB() ? 1 - slice.getColFromL0Flag() : 0), slice.getColRefIdx());
  Mv cTMv;
  RefPicList fetchRefPicList = RefPicList(slice.isInterB() ? 1 - slice.getColFromL0Flag() : 0);

  bool terminate = false;
  for (unsigned currRefListId = 0; currRefListId < (slice.getSliceType() == B_SLICE ? 2 : 1) && !terminate; currRefListId++)
  {
    for (int uiN = 0; uiN < count && !terminate; uiN++)
    {
      RefPicList currRefPicList = RefPicList(slice.getCheckLDC() ? (slice.getColFromL0Flag() ? currRefListId : 1 - currRefListId) : currRefListId);

      if ((mrgCtx.interDirNeighbours[uiN] & (1 << currRefPicList)) && slice.getRefPic(currRefPicList, mrgCtx.mvFieldNeighbours[uiN * 2 + currRefPicList].refIdx) == pColPic)
      {
        cTMv = mrgCtx.mvFieldNeighbours[uiN * 2 + currRefPicList].mv;
        terminate = true;
        fetchRefPicList = currRefPicList;
        break;
      }
    }
  }

  ///////////////////////////////////////////////////////////////////////
  ////////          GET Initial Temporal Vector                  ////////
  ///////////////////////////////////////////////////////////////////////
  int mvPrec = 2;
#if !REMOVE_MV_ADAPT_PREC
  if (pu.cs->sps->getSpsNext().getUseHighPrecMv())
  {
    cTMv.setHighPrec();
#endif
    mvPrec += VCEG_AZ07_MV_ADD_PRECISION_BIT_FOR_STORE;
#if !REMOVE_MV_ADAPT_PREC
  }
#endif
  int mvRndOffs = (1 << mvPrec) >> 1;

  Mv cTempVector = cTMv;
  bool  tempLICFlag = false;

  // compute the location of the current PU
  Position puPos = pu.lumaPos();
  Size puSize = pu.lumaSize();
  int numPartLine = std::max(puSize.width >> slice.getSubPuMvpSubblkLog2Size(), 1u);
  int numPartCol = std::max(puSize.height >> slice.getSubPuMvpSubblkLog2Size(), 1u);
  int puHeight = numPartCol == 1 ? puSize.height : 1 << slice.getSubPuMvpSubblkLog2Size();
  int puWidth = numPartLine == 1 ? puSize.width : 1 << slice.getSubPuMvpSubblkLog2Size();

  Mv cColMv;
  // use coldir.
  bool     bBSlice = slice.isInterB();

  Position centerPos;

  bool found = false;
  cTempVector = cTMv;
  int tempX = ((cTempVector.getHor() + mvRndOffs) >> mvPrec);
  int tempY = ((cTempVector.getVer() + mvRndOffs) >> mvPrec);
  clipColBlkMv(tempX, tempY, pu);

  if (puSize.width == puWidth && puSize.height == puHeight)
  {
    centerPos.x = puPos.x + (puSize.width >> 1) + tempX;
    centerPos.y = puPos.y + (puSize.height >> 1) + tempY;
  }
  else
  {
    centerPos.x = puPos.x + ((puSize.width / puWidth) >> 1)   * puWidth + (puWidth >> 1) + tempX;
    centerPos.y = puPos.y + ((puSize.height / puHeight) >> 1) * puHeight + (puHeight >> 1) + tempY;
  }

  centerPos.x = Clip3(0, (int)pColPic->lwidth() - 1, centerPos.x);
  centerPos.y = Clip3(0, (int)pColPic->lheight() - 1, centerPos.y);

  centerPos = Position{ PosType(centerPos.x & mask), PosType(centerPos.y & mask) };

  // derivation of center motion parameters from the collocated CU
  const MotionInfo &mi = pColPic->cs->getMotionInfo(centerPos);

  if (mi.isInter)
  {
    for (unsigned currRefListId = 0; currRefListId < (bBSlice ? 2 : 1); currRefListId++)
    {
      RefPicList  currRefPicList = RefPicList(currRefListId);

      if (deriveScaledMotionTemporal(slice, centerPos, pColPic, currRefPicList, cColMv, fetchRefPicList))
      {
        // set as default, for further motion vector field spanning
        mrgCtx.mvFieldNeighbours[(count << 1) + currRefListId].setMvField(cColMv, 0);
        mrgCtx.interDirNeighbours[count] |= (1 << currRefListId);
        LICFlag = tempLICFlag;
        found = true;
      }
      else
      {
        mrgCtx.mvFieldNeighbours[(count << 1) + currRefListId].setMvField(Mv(), NOT_VALID);
        mrgCtx.interDirNeighbours[count] &= ~(1 << currRefListId);
      }
    }
  }

  if (!found)
  {
    return false;
  }
  
  int xOff = puWidth / 2;
  int yOff = puHeight / 2;

  // compute the location of the current PU
  xOff += tempX;
  yOff += tempY;

  int iPicWidth = pColPic->lwidth() - 1;
  int iPicHeight = pColPic->lheight() - 1;

  MotionBuf& mb = mrgCtx.subPuMvpMiBuf;

  const bool isBiPred = isBipredRestriction(pu);

  for (int y = puPos.y; y < puPos.y + puSize.height; y += puHeight)
  {
    for (int x = puPos.x; x < puPos.x + puSize.width; x += puWidth)
    {
      Position colPos{ x + xOff, y + yOff };

      colPos.x = Clip3(0, iPicWidth, colPos.x);
      colPos.y = Clip3(0, iPicHeight, colPos.y);

      colPos = Position{ PosType(colPos.x & mask), PosType(colPos.y & mask) };

      const MotionInfo &colMi = pColPic->cs->getMotionInfo(colPos);

      MotionInfo mi;

      mi.isInter = true;
      mi.sliceIdx = slice.getIndependentSliceIdx();

      if (colMi.isInter)
      {
        for (unsigned currRefListId = 0; currRefListId < (bBSlice ? 2 : 1); currRefListId++)
        {
          RefPicList currRefPicList = RefPicList(currRefListId);
          if (deriveScaledMotionTemporal(slice, colPos, pColPic, currRefPicList, cColMv, fetchRefPicList))
          {
            mi.refIdx[currRefListId] = 0;
            mi.mv[currRefListId] = cColMv;
          }
        }
        }
      else
      {
        // intra coded, in this case, no motion vector is available for list 0 or list 1, so use default
        mi.mv[0] = mrgCtx.mvFieldNeighbours[(count << 1) + 0].mv;
        mi.mv[1] = mrgCtx.mvFieldNeighbours[(count << 1) + 1].mv;
        mi.refIdx[0] = mrgCtx.mvFieldNeighbours[(count << 1) + 0].refIdx;
        mi.refIdx[1] = mrgCtx.mvFieldNeighbours[(count << 1) + 1].refIdx;
      }

      mi.interDir = (mi.refIdx[0] != -1 ? 1 : 0) + (mi.refIdx[1] != -1 ? 2 : 0);

      if (isBiPred && mi.interDir == 3)
      {
        mi.interDir = 1;
        mi.mv[1] = Mv();
        mi.refIdx[1] = NOT_VALID;
      }

      mb.subBuf(g_miScaling.scale(Position{ x, y } -pu.lumaPos()), g_miScaling.scale(Size(puWidth, puHeight))).fill(mi);
      }
    }

  return true;
  }

void PU::spanMotionInfo( PredictionUnit &pu, const MergeCtx &mrgCtx )
{
  MotionBuf mb = pu.getMotionBuf();

  if( !pu.mergeFlag || pu.mergeType == MRG_TYPE_DEFAULT_N )
  {
    MotionInfo mi;

    mi.isInter  = CU::isInter( *pu.cu );
    mi.sliceIdx = pu.cu->slice->getIndependentSliceIdx();

    if( mi.isInter )
    {
      mi.interDir = pu.interDir;

      for( int i = 0; i < NUM_REF_PIC_LIST_01; i++ )
      {
        mi.mv[i]     = pu.mv[i];
        mi.refIdx[i] = pu.refIdx[i];
      }
    }

    if( pu.cu->affine )
    {
      for( int y = 0; y < mb.height; y++ )
      {
        for( int x = 0; x < mb.width; x++ )
        {
          MotionInfo &dest = mb.at( x, y );
          dest.isInter  = mi.isInter;
          dest.interDir = mi.interDir;
          dest.sliceIdx = mi.sliceIdx;
          for( int i = 0; i < NUM_REF_PIC_LIST_01; i++ )
          {
            if( mi.refIdx[i] == -1 )
            {
              dest.mv[i] = Mv();
            }
            dest.refIdx[i] = mi.refIdx[i];
          }
        }
      }
    }
    else
    {
      mb.fill( mi );
    }
  }
  else if (pu.mergeType == MRG_TYPE_SUBPU_ATMVP)
  {
    CHECK(mrgCtx.subPuMvpMiBuf.area() == 0 || !mrgCtx.subPuMvpMiBuf.buf, "Buffer not initialized");
    mb.copyFrom(mrgCtx.subPuMvpMiBuf);
  }
  else
  {

    if( isBipredRestriction( pu ) )
    {
      for( int y = 0; y < mb.height; y++ )
      {
        for( int x = 0; x < mb.width; x++ )
        {
          MotionInfo &mi = mb.at( x, y );
          if( mi.interDir == 3 )
          {
            mi.interDir  = 1;
            mi.mv    [1] = Mv();
            mi.refIdx[1] = NOT_VALID;
          }
        }
      }
    }
  }
}

void PU::applyImv( PredictionUnit& pu, MergeCtx &mrgCtx, InterPrediction *interPred )
{
  if( !pu.mergeFlag )
  {
    unsigned imvShift = pu.cu->imv << 1;
    if( pu.interDir != 2 /* PRED_L1 */ )
    {
      if (pu.cu->imv)
      {
#if !REMOVE_MV_ADAPT_PREC
        CHECK(pu.mvd[0].highPrec, "Motion vector difference should never be high precision");
#endif
        pu.mvd[0] = Mv( pu.mvd[0].hor << imvShift, pu.mvd[0].ver << imvShift );
      }
      unsigned mvp_idx = pu.mvpIdx[0];
      AMVPInfo amvpInfo;
      PU::fillMvpCand(pu, REF_PIC_LIST_0, pu.refIdx[0], amvpInfo);
      pu.mvpNum[0] = amvpInfo.numCand;
      pu.mvpIdx[0] = mvp_idx;
      pu.mv    [0] = amvpInfo.mvCand[mvp_idx] + pu.mvd[0];
#if REMOVE_MV_ADAPT_PREC
      pu.mv[0].hor = pu.mv[0].hor << VCEG_AZ07_MV_ADD_PRECISION_BIT_FOR_STORE;
      pu.mv[0].ver = pu.mv[0].ver << VCEG_AZ07_MV_ADD_PRECISION_BIT_FOR_STORE;
#endif
    }

    if (pu.interDir != 1 /* PRED_L0 */)
    {
      if( !( pu.cu->cs->slice->getMvdL1ZeroFlag() && pu.interDir == 3 ) && pu.cu->imv )/* PRED_BI */
      {
#if !REMOVE_MV_ADAPT_PREC
        CHECK(pu.mvd[1].highPrec, "Motion vector difference should never be high precision");
#endif
        pu.mvd[1] = Mv( pu.mvd[1].hor << imvShift, pu.mvd[1].ver << imvShift );
      }
      unsigned mvp_idx = pu.mvpIdx[1];
      AMVPInfo amvpInfo;
      PU::fillMvpCand(pu, REF_PIC_LIST_1, pu.refIdx[1], amvpInfo);
      pu.mvpNum[1] = amvpInfo.numCand;
      pu.mvpIdx[1] = mvp_idx;
      pu.mv    [1] = amvpInfo.mvCand[mvp_idx] + pu.mvd[1];
#if REMOVE_MV_ADAPT_PREC
      pu.mv[1].hor = pu.mv[1].hor << VCEG_AZ07_MV_ADD_PRECISION_BIT_FOR_STORE;
      pu.mv[1].ver = pu.mv[1].ver << VCEG_AZ07_MV_ADD_PRECISION_BIT_FOR_STORE;
#endif
    }
  }
  else
  {
    // this function is never called for merge
    THROW("unexpected");
    PU::getInterMergeCandidates ( pu, mrgCtx );
    PU::restrictBiPredMergeCands( pu, mrgCtx );

    mrgCtx.setMergeInfo( pu, pu.mergeIdx );
  }

  PU::spanMotionInfo( pu, mrgCtx );
}

bool PU::isBiPredFromDifferentDir( const PredictionUnit& pu )
{
  if ( pu.refIdx[0] >= 0 && pu.refIdx[1] >= 0 )
  {
    const int iPOC0 = pu.cu->slice->getRefPOC( REF_PIC_LIST_0, pu.refIdx[0] );
    const int iPOC1 = pu.cu->slice->getRefPOC( REF_PIC_LIST_1, pu.refIdx[1] );
    const int iPOC  = pu.cu->slice->getPOC();
    if ( (iPOC - iPOC0)*(iPOC - iPOC1) < 0 )
    {
      return true;
    }
  }

  return false;
}

void PU::restrictBiPredMergeCands( const PredictionUnit &pu, MergeCtx& mergeCtx )
{
  if( PU::isBipredRestriction( pu ) )
  {
    for( uint32_t mergeCand = 0; mergeCand < mergeCtx.numValidMergeCand; ++mergeCand )
    {
      if( mergeCtx.interDirNeighbours[ mergeCand ] == 3 )
      {
        mergeCtx.interDirNeighbours[ mergeCand ] = 1;
        mergeCtx.mvFieldNeighbours[( mergeCand << 1 ) + 1].setMvField( Mv( 0, 0 ), -1 );
      }
    }
  }
}

void CU::resetMVDandMV2Int( CodingUnit& cu, InterPrediction *interPred )
{
  for( auto &pu : CU::traversePUs( cu ) )
  {
    MergeCtx mrgCtx;

    if( !pu.mergeFlag )
    {
      unsigned imvShift = cu.imv << 1;
      if( pu.interDir != 2 /* PRED_L1 */ )
      {
        Mv mv        = pu.mv[0];
        Mv mvPred;
        AMVPInfo amvpInfo;
        PU::fillMvpCand(pu, REF_PIC_LIST_0, pu.refIdx[0], amvpInfo);
        pu.mvpNum[0] = amvpInfo.numCand;

        mvPred       = amvpInfo.mvCand[pu.mvpIdx[0]];
        roundMV      ( mv, imvShift );
        pu.mv[0]     = mv;
        Mv mvDiff    = mv - mvPred;
        pu.mvd[0]    = mvDiff;
      }
      if( pu.interDir != 1 /* PRED_L0 */ )
      {
        Mv mv        = pu.mv[1];
        Mv mvPred;
        AMVPInfo amvpInfo;
        PU::fillMvpCand(pu, REF_PIC_LIST_1, pu.refIdx[1], amvpInfo);
        pu.mvpNum[1] = amvpInfo.numCand;

        mvPred       = amvpInfo.mvCand[pu.mvpIdx[1]];
        roundMV      ( mv, imvShift );
        Mv mvDiff    = mv - mvPred;

        if( pu.cu->cs->slice->getMvdL1ZeroFlag() && pu.interDir == 3 /* PRED_BI */ )
        {
          pu.mvd[1] = Mv();
          mv = mvPred;
        }
        else
        {
          pu.mvd[1] = mvDiff;
        }
        pu.mv[1] = mv;
      }

    }
    else
    {
        PU::getInterMergeCandidates ( pu, mrgCtx );
        PU::restrictBiPredMergeCands( pu, mrgCtx );

        mrgCtx.setMergeInfo( pu, pu.mergeIdx );
    }

    PU::spanMotionInfo( pu, mrgCtx );
  }
}

bool CU::hasSubCUNonZeroMVd( const CodingUnit& cu )
{
  bool bNonZeroMvd = false;

  for( const auto &pu : CU::traversePUs( cu ) )
  {
    if( ( !pu.mergeFlag ) && ( !cu.skip ) )
    {
      if( pu.interDir != 2 /* PRED_L1 */ )
      {
        bNonZeroMvd |= pu.mvd[REF_PIC_LIST_0].getHor() != 0;
        bNonZeroMvd |= pu.mvd[REF_PIC_LIST_0].getVer() != 0;
      }
      if( pu.interDir != 1 /* PRED_L0 */ )
      {
        if( !pu.cu->cs->slice->getMvdL1ZeroFlag() || pu.interDir != 3 /* PRED_BI */ )
        {
          bNonZeroMvd |= pu.mvd[REF_PIC_LIST_1].getHor() != 0;
          bNonZeroMvd |= pu.mvd[REF_PIC_LIST_1].getVer() != 0;
        }
      }
    }
  }

  return bNonZeroMvd;
}

int CU::getMaxNeighboriMVCandNum( const CodingStructure& cs, const Position& pos )
{
  const int  numDefault     = 0;
  int        maxImvNumCand  = 0;

  // Get BCBP of left PU
#if HEVC_TILES_WPP
  const CodingUnit *cuLeft  = cs.getCURestricted( pos.offset( -1, 0 ), cs.slice->getIndependentSliceIdx(), cs.picture->tileMap->getTileIdxMap( pos ), CH_L );
#else
  const CodingUnit *cuLeft  = cs.getCURestricted( pos.offset( -1, 0 ), cs.slice->getIndependentSliceIdx(), CH_L );
#endif
  maxImvNumCand = ( cuLeft ) ? cuLeft->imvNumCand : numDefault;

  // Get BCBP of above PU
#if HEVC_TILES_WPP
  const CodingUnit *cuAbove = cs.getCURestricted( pos.offset( 0, -1 ), cs.slice->getIndependentSliceIdx(), cs.picture->tileMap->getTileIdxMap( pos ), CH_L );
#else
  const CodingUnit *cuAbove = cs.getCURestricted( pos.offset( 0, -1 ), cs.slice->getIndependentSliceIdx(), CH_L );
#endif
  maxImvNumCand = std::max( maxImvNumCand, ( cuAbove ) ? cuAbove->imvNumCand : numDefault );

  return maxImvNumCand;
}





// TU tools

#if HEVC_USE_4x4_DSTVII
bool TU::useDST(const TransformUnit &tu, const ComponentID &compID)
{
  return isLuma(compID) && tu.cu->predMode == MODE_INTRA;
}

#endif

bool TU::isNonTransformedResidualRotated(const TransformUnit &tu, const ComponentID &compID)
{
  return tu.cs->sps->getSpsRangeExtension().getTransformSkipRotationEnabledFlag() && tu.blocks[compID].width == 4 && tu.cu->predMode == MODE_INTRA;
}

bool TU::getCbf( const TransformUnit &tu, const ComponentID &compID )
{
#if ENABLE_BMS
  return getCbfAtDepth( tu, compID, tu.depth );
#else
  return tu.cbf[compID];
#endif
}

#if ENABLE_BMS
bool TU::getCbfAtDepth(const TransformUnit &tu, const ComponentID &compID, const unsigned &depth)
{
  return ((tu.cbf[compID] >> depth) & 1) == 1;
}

void TU::setCbfAtDepth(TransformUnit &tu, const ComponentID &compID, const unsigned &depth, const bool &cbf)
{
  // first clear the CBF at the depth
  tu.cbf[compID] &= ~(1  << depth);
  // then set the CBF
  tu.cbf[compID] |= ((cbf ? 1 : 0) << depth);
}
#else
void TU::setCbf( TransformUnit &tu, const ComponentID &compID, const bool &cbf )
{
  tu.cbf[compID] = cbf;
}
#endif

bool TU::hasTransformSkipFlag(const CodingStructure& cs, const CompArea& area)
{
  uint32_t transformSkipLog2MaxSize = cs.pps->getPpsRangeExtension().getLog2MaxTransformSkipBlockSize();

  if( cs.pcv->rectCUs )
  {
    return ( area.width * area.height <= (1 << ( transformSkipLog2MaxSize << 1 )) );
  }
  return ( area.width <= (1 << transformSkipLog2MaxSize) );
}

uint32_t TU::getGolombRiceStatisticsIndex(const TransformUnit &tu, const ComponentID &compID)
{
  const bool transformSkip    = tu.transformSkip[compID];
  const bool transquantBypass = tu.cu->transQuantBypass;

  //--------

  const uint32_t channelTypeOffset = isChroma(compID) ? 2 : 0;
  const uint32_t nonTransformedOffset = (transformSkip || transquantBypass) ? 1 : 0;

  //--------

  const uint32_t selectedIndex = channelTypeOffset + nonTransformedOffset;
  CHECK( selectedIndex >= RExt__GOLOMB_RICE_ADAPTATION_STATISTICS_SETS, "Invalid golomb rice adaptation statistics set" );

  return selectedIndex;
}

#if HEVC_USE_MDCS
uint32_t TU::getCoefScanIdx(const TransformUnit &tu, const ComponentID &compID)
{
  //------------------------------------------------

  //this mechanism is available for intra only

  if( !CU::isIntra( *tu.cu ) )
  {
    return SCAN_DIAG;
  }

  //------------------------------------------------

  //check that MDCS can be used for this TU


  const CompArea &area      = tu.blocks[compID];
  const SPS &sps            = *tu.cs->sps;
  const ChromaFormat format = sps.getChromaFormatIdc();


  const uint32_t maximumWidth  = MDCS_MAXIMUM_WIDTH  >> getComponentScaleX(compID, format);
  const uint32_t maximumHeight = MDCS_MAXIMUM_HEIGHT >> getComponentScaleY(compID, format);

  if ((area.width > maximumWidth) || (area.height > maximumHeight))
  {
    return SCAN_DIAG;
  }

  //------------------------------------------------

  //otherwise, select the appropriate mode

  const PredictionUnit &pu = *tu.cs->getPU( area.pos(), toChannelType( compID ) );

  uint32_t uiDirMode = PU::getFinalIntraMode(pu, toChannelType(compID));

  //------------------

       if (abs((int) uiDirMode - VER_IDX) <= MDCS_ANGLE_LIMIT)
  {
    return SCAN_HOR;
  }
  else if (abs((int) uiDirMode - HOR_IDX) <= MDCS_ANGLE_LIMIT)
  {
    return SCAN_VER;
  }
  else
  {
    return SCAN_DIAG;
  }
}

#endif
bool TU::hasCrossCompPredInfo( const TransformUnit &tu, const ComponentID &compID )
{
  return ( isChroma(compID) && tu.cs->pps->getPpsRangeExtension().getCrossComponentPredictionEnabledFlag() && TU::getCbf( tu, COMPONENT_Y ) &&
         ( CU::isInter(*tu.cu) || PU::isChromaIntraModeCrossCheckMode( *tu.cs->getPU( tu.blocks[compID].pos(), toChannelType( compID ) ) ) ) );
}

uint32_t TU::getNumNonZeroCoeffsNonTS( const TransformUnit& tu, const bool bLuma, const bool bChroma )
{
  uint32_t count = 0;
  for( uint32_t i = 0; i < ::getNumberValidTBlocks( *tu.cs->pcv ); i++ )
  {
    if( tu.blocks[i].valid() && !tu.transformSkip[i] && TU::getCbf( tu, ComponentID( i ) ) )
    {
      if( isLuma  ( tu.blocks[i].compID ) && !bLuma   ) continue;
      if( isChroma( tu.blocks[i].compID ) && !bChroma ) continue;

      uint32_t area = tu.blocks[i].area();
      const TCoeff* coeff = tu.getCoeffs( ComponentID( i ) ).buf;
      for( uint32_t j = 0; j < area; j++ )
      {
        count += coeff[j] != 0;
      }
    }
  }
  return count;
}

bool TU::needsSqrt2Scale( const Size& size )
{
  return (((g_aucLog2[size.width] + g_aucLog2[size.height]) & 1) == 1);
}

#if HM_QTBT_AS_IN_JEM_QUANT

bool TU::needsBlockSizeTrafoScale( const Size& size )
{
  return needsSqrt2Scale( size ) || isNonLog2BlockSize( size );
}
#else
bool TU::needsQP3Offset(const TransformUnit &tu, const ComponentID &compID)
{
  if( tu.cs->pcv->rectCUs && !tu.transformSkip[compID] )
  {
    return ( ( ( g_aucLog2[tu.blocks[compID].width] + g_aucLog2[tu.blocks[compID].height] ) & 1 ) == 1 );
  }
  return false;
}
#endif





// other tools

uint32_t getCtuAddr( const Position& pos, const PreCalcValues& pcv )
{
  return ( pos.x >> pcv.maxCUWidthLog2 ) + ( pos.y >> pcv.maxCUHeightLog2 ) * pcv.widthInCtus;
}