EncCu.cpp

      for (int y = lumaArea.y; y < lumaArea.y + lumaArea.height; y += MIN_PU_SIZE)
      {
        for (int x = lumaArea.x; x < lumaArea.x + lumaArea.width; x += MIN_PU_SIZE)
        {
          const MotionInfo &curMi = pu.cs->picture->cs->getMotionInfo(Position{ x, y });

          subPu.UnitArea::operator=(UnitArea(pu.chromaFormat, Area(x, y, MIN_PU_SIZE, MIN_PU_SIZE)));
          Position offsetRef = subPu.blocks[compID].pos().offset((curMi.bv.getHor() >> shiftHor), (curMi.bv.getVer() >> shiftVer));
          Position refEndPos(offsetRef.x + subPu.blocks[compID].size().width - 1, offsetRef.y + subPu.blocks[compID].size().height - 1 );

          if (!subPu.cs->isDecomp(refEndPos, toChannelType(compID)) || !subPu.cs->isDecomp(offsetRef, toChannelType(compID))) // ref block is not yet available for this chroma sub-block
          {
            success = false;
            break;
          }
        }
        if (!success)
          break;
      }
      ////////////////////////////////////////////////////////////////////////////

      if (success)
      {
        //pu.mergeType = MRG_TYPE_CPR;
        m_pcInterSearch->motionCompensation(pu, REF_PIC_LIST_0, false, true); // luma=0, chroma=1
        m_pcInterSearch->encodeResAndCalcRdInterCU(*tempCS, partitioner, false, false, true);

        xEncodeDontSplit(*tempCS, partitioner);

        xCheckDQP(*tempCS, partitioner);

        DTRACE_MODE_COST(*tempCS, m_pcRdCost->getLambda());

        xCheckBestMode(tempCS, bestCS, partitioner, encTestMode);
      }
      else
      {
        tempCS->dist = 0;
        tempCS->fracBits = 0;
        tempCS->cost = MAX_DOUBLE;
      }
    }
  }
  // check cpr mode in encoder RD
  //////////////////////////////////////////////////////////////////////////////////////////////

void EncCu::xCheckRDCostInter( CodingStructure *&tempCS, CodingStructure *&bestCS, Partitioner &partitioner, const EncTestMode& encTestMode )
{
  tempCS->initStructData( encTestMode.qp, encTestMode.lossless );

  
  m_pcInterSearch->setAffineModeSelected(false);

  if( tempCS->slice->getCheckLDC() )
  {
    m_bestGbiCost[0] = m_bestGbiCost[1] = std::numeric_limits<double>::max();
    m_bestGbiIdx[0] = m_bestGbiIdx[1] = -1;
  }

  m_pcInterSearch->resetBufferedUniMotions();
  int gbiLoopNum = (tempCS->slice->isInterB() ? GBI_NUM : 1);
  gbiLoopNum = (tempCS->sps->getSpsNext().getUseGBi() ? gbiLoopNum : 1);

  if( tempCS->area.lwidth() * tempCS->area.lheight() < GBI_SIZE_CONSTRAINT )
  {
    gbiLoopNum = 1;
  }

  double curBestCost = bestCS->cost;
  double equGBiCost = MAX_DOUBLE;

  for( int gbiLoopIdx = 0; gbiLoopIdx < gbiLoopNum; gbiLoopIdx++ )
  {
    if( m_pcEncCfg->getUseGBiFast() )
    {
      auto blkCache = dynamic_cast< CacheBlkInfoCtrl* >(m_modeCtrl);

      if( blkCache )
      {
        bool isBestInter = blkCache->getInter(bestCS->area);
        uint8_t bestGBiIdx = blkCache->getGbiIdx(bestCS->area);

        if( isBestInter && g_GbiSearchOrder[gbiLoopIdx] != GBI_DEFAULT && g_GbiSearchOrder[gbiLoopIdx] != bestGBiIdx )
        {
          continue;
        }
      }
    }
    if( !tempCS->slice->getCheckLDC() )
    {
      if( gbiLoopIdx != 0 && gbiLoopIdx != 3 && gbiLoopIdx != 4 )
      {
        continue;
      }
    }

  CodingUnit &cu      = tempCS->addCU( tempCS->area, partitioner.chType );

  partitioner.setCUData( cu );
  cu.slice            = tempCS->slice;
#if HEVC_TILES_WPP
  cu.tileIdx          = tempCS->picture->tileMap->getTileIdxMap( tempCS->area.lumaPos() );
#endif
  cu.skip             = false;
  cu.mmvdSkip = false;
//cu.affine
  cu.predMode         = MODE_INTER;
  cu.transQuantBypass = encTestMode.lossless;
  cu.chromaQpAdj      = cu.transQuantBypass ? 0 : m_cuChromaQpOffsetIdxPlus1;
  cu.qp               = encTestMode.qp;
  CU::addPUs( cu );

  cu.GBiIdx = g_GbiSearchOrder[gbiLoopIdx];
  uint8_t gbiIdx = cu.GBiIdx;
  bool  testGbi = (gbiIdx != GBI_DEFAULT);

  m_pcInterSearch->predInterSearch( cu, partitioner );

  const unsigned wIdx = gp_sizeIdxInfo->idxFrom( tempCS->area.lwidth () );

  gbiIdx = CU::getValidGbiIdx(cu);
  if( testGbi && gbiIdx == GBI_DEFAULT ) // Enabled GBi but the search results is uni.
  {
    tempCS->initStructData(encTestMode.qp, encTestMode.lossless);
    continue;
  }
  CHECK(!(testGbi || (!testGbi && gbiIdx == GBI_DEFAULT)), " !( bTestGbi || (!bTestGbi && gbiIdx == GBI_DEFAULT ) )");

  bool isEqualUni = false;
  if( m_pcEncCfg->getUseGBiFast() )
  {
    if( cu.firstPU->interDir != 3 && testGbi == 0 )
    {
      isEqualUni = true;
    }
  }

  xEncodeInterResidual( tempCS, bestCS, partitioner, encTestMode, 0
    , m_pImvTempCS ? m_pImvTempCS[wIdx] : NULL
    , 1
    , 0
    , &equGBiCost
  );

  if( g_GbiSearchOrder[gbiLoopIdx] == GBI_DEFAULT )
    m_pcInterSearch->setAffineModeSelected((bestCS->cus.front()->affine && !(bestCS->cus.front()->firstPU->mergeFlag)));

  tempCS->initStructData(encTestMode.qp, encTestMode.lossless);

  double skipTH = MAX_DOUBLE;
  skipTH = (m_pcEncCfg->getUseGBiFast() ? 1.05 : MAX_DOUBLE);
  if( equGBiCost > curBestCost * skipTH )
  {
    break;
  }

  if( m_pcEncCfg->getUseGBiFast() )
  {
    if( isEqualUni == true && m_pcEncCfg->getIntraPeriod() == -1 )
    {
      break;
    }
  }
  if( g_GbiSearchOrder[gbiLoopIdx] == GBI_DEFAULT && xIsGBiSkip(cu) && m_pcEncCfg->getUseGBiFast() )
  {
    break;
  }
 }  // for( UChar gbiLoopIdx = 0; gbiLoopIdx < gbiLoopNum; gbiLoopIdx++ )
}





bool EncCu::xCheckRDCostInterIMV( CodingStructure *&tempCS, CodingStructure *&bestCS, Partitioner &partitioner, const EncTestMode& encTestMode )
{
  int iIMV = int( ( encTestMode.opts & ETO_IMV ) >> ETO_IMV_SHIFT );
  m_pcInterSearch->setAffineModeSelected(false);
  // Only int-Pel, 4-Pel and fast 4-Pel allowed
  CHECK( iIMV != 1 && iIMV != 2 && iIMV != 3, "Unsupported IMV Mode" );
  // Fast 4-Pel Mode

  EncTestMode encTestModeBase = encTestMode;                                        // copy for clearing non-IMV options
  encTestModeBase.opts        = EncTestModeOpts( encTestModeBase.opts & ETO_IMV );  // clear non-IMV options (is that intended?)

  tempCS->initStructData( encTestMode.qp, encTestMode.lossless );

  CodingStructure* pcCUInfo2Reuse = nullptr;

  m_pcInterSearch->resetBufferedUniMotions();
  int gbiLoopNum = (tempCS->slice->isInterB() ? GBI_NUM : 1);
  gbiLoopNum = (pcCUInfo2Reuse != NULL ? 1 : gbiLoopNum);
  gbiLoopNum = (tempCS->slice->getSPS()->getSpsNext().getUseGBi() ? gbiLoopNum : 1);

  if( tempCS->area.lwidth() * tempCS->area.lheight() < GBI_SIZE_CONSTRAINT )
  {
    gbiLoopNum = 1;
  }

  double curBestCost = bestCS->cost;
  double equGBiCost = MAX_DOUBLE;

  for( int gbiLoopIdx = 0; gbiLoopIdx < gbiLoopNum; gbiLoopIdx++ )
  {
    if( m_pcEncCfg->getUseGBiFast() )
    {
      auto blkCache = dynamic_cast< CacheBlkInfoCtrl* >(m_modeCtrl);

      if( blkCache )
      {
        bool isBestInter = blkCache->getInter(bestCS->area);
        uint8_t bestGBiIdx = blkCache->getGbiIdx(bestCS->area);

        if( isBestInter && g_GbiSearchOrder[gbiLoopIdx] != GBI_DEFAULT && g_GbiSearchOrder[gbiLoopIdx] != bestGBiIdx )
        {
          continue;
        }
      }
    }

    if( !tempCS->slice->getCheckLDC() )
    {
      if( gbiLoopIdx != 0 && gbiLoopIdx != 3 && gbiLoopIdx != 4 )
      {
        continue;
      }
    }

    if( m_pcEncCfg->getUseGBiFast() && tempCS->slice->getCheckLDC() && g_GbiSearchOrder[gbiLoopIdx] != GBI_DEFAULT
      && (m_bestGbiIdx[0] >= 0 && g_GbiSearchOrder[gbiLoopIdx] != m_bestGbiIdx[0])
      && (m_bestGbiIdx[1] >= 0 && g_GbiSearchOrder[gbiLoopIdx] != m_bestGbiIdx[1]))
    {
      continue;
    }

  CodingUnit &cu = ( pcCUInfo2Reuse != nullptr ) ? *tempCS->getCU( partitioner.chType ) : tempCS->addCU( tempCS->area, partitioner.chType );

  if( pcCUInfo2Reuse == nullptr )
  {
    partitioner.setCUData( cu );
    cu.slice            = tempCS->slice;
#if HEVC_TILES_WPP
    cu.tileIdx          = tempCS->picture->tileMap->getTileIdxMap( tempCS->area.lumaPos() );
#endif
    cu.skip             = false;
    cu.mmvdSkip = false;
  //cu.affine
    cu.predMode         = MODE_INTER;
    cu.transQuantBypass = encTestMode.lossless;
    cu.chromaQpAdj      = cu.transQuantBypass ? 0 : m_cuChromaQpOffsetIdxPlus1;
    cu.qp               = encTestMode.qp;

    CU::addPUs( cu );
  }
  else
  {
    CHECK( cu.skip,                                "Mismatch" );
    CHECK( cu.qtDepth  != partitioner.currQtDepth, "Mismatch" );
    CHECK( cu.btDepth  != partitioner.currBtDepth, "Mismatch" );
    CHECK( cu.mtDepth  != partitioner.currMtDepth, "Mismatch" );
    CHECK( cu.depth    != partitioner.currDepth,   "Mismatch" );
  }

  cu.imv      = iIMV > 1 ? 2 : 1;
  cu.emtFlag  = false;

  bool testGbi;
  uint8_t gbiIdx;
  
  if( pcCUInfo2Reuse != nullptr )
  {
    // reuse the motion info from pcCUInfo2Reuse
    CU::resetMVDandMV2Int( cu, m_pcInterSearch );

    CHECK(cu.GBiIdx < 0 || cu.GBiIdx >= GBI_NUM, "cu.GBiIdx < 0 || cu.GBiIdx >= GBI_NUM");
    gbiIdx = CU::getValidGbiIdx(cu);
    testGbi = (gbiIdx != GBI_DEFAULT);

    if( !CU::hasSubCUNonZeroMVd( cu ) )
    {
      if (m_modeCtrl->useModeResult(encTestModeBase, tempCS, partitioner))
      {
        std::swap(tempCS, bestCS);
        // store temp best CI for next CU coding
        m_CurrCtx->best = m_CABACEstimator->getCtx();
      }
      return false;
    }
    else
    {
      m_pcInterSearch->motionCompensation( cu );
    }
  }
  else
  {
    cu.GBiIdx = g_GbiSearchOrder[gbiLoopIdx];
    gbiIdx = cu.GBiIdx;
    testGbi = (gbiIdx != GBI_DEFAULT);

    m_pcInterSearch->predInterSearch( cu, partitioner );

    gbiIdx = CU::getValidGbiIdx(cu);
  }

  if( testGbi && gbiIdx == GBI_DEFAULT ) // Enabled GBi but the search results is uni.
  {
    tempCS->initStructData(encTestMode.qp, encTestMode.lossless);
    continue;
  }
  CHECK(!(testGbi || (!testGbi && gbiIdx == GBI_DEFAULT)), " !( bTestGbi || (!bTestGbi && gbiIdx == GBI_DEFAULT ) )");

  bool isEqualUni = false;
  if( m_pcEncCfg->getUseGBiFast() )
  {
    if( cu.firstPU->interDir != 3 && testGbi == 0 )
    {
      isEqualUni = true;
    }
  }

  if( !CU::hasSubCUNonZeroMVd( cu ) )
  {
    if (m_modeCtrl->useModeResult(encTestModeBase, tempCS, partitioner))
    {
      std::swap(tempCS, bestCS);
      // store temp best CI for next CU coding
      m_CurrCtx->best = m_CABACEstimator->getCtx();
    }
    return false;
  }

  xEncodeInterResidual( tempCS, bestCS, partitioner, encTestModeBase, 0
    , NULL
    , true
    , 0
    , &equGBiCost
  );

  tempCS->initStructData(encTestMode.qp, encTestMode.lossless);

  double skipTH = MAX_DOUBLE;
  skipTH = (m_pcEncCfg->getUseGBiFast() ? 1.05 : MAX_DOUBLE);
  if( equGBiCost > curBestCost * skipTH )
  {
    break;
  }

  if( m_pcEncCfg->getUseGBiFast() )
  {
    if( isEqualUni == true && m_pcEncCfg->getIntraPeriod() == -1 )
    {
      break;
    }
  }
  if( g_GbiSearchOrder[gbiLoopIdx] == GBI_DEFAULT && xIsGBiSkip(cu) && m_pcEncCfg->getUseGBiFast() )
  {
    break;
  }
 } // for( UChar gbiLoopIdx = 0; gbiLoopIdx < gbiLoopNum; gbiLoopIdx++ )

  return true;
}

void EncCu::xEncodeInterResidual( CodingStructure *&tempCS, CodingStructure *&bestCS, Partitioner &partitioner, const EncTestMode& encTestMode, int residualPass
  , CodingStructure* imvCS
  , int emtMode
  , bool* bestHasNonResi
  , double* equGBiCost
  )
{
  if( residualPass == 1 && encTestMode.lossless )
  {
    return;
  }

  CodingUnit*            cu        = tempCS->getCU( partitioner.chType );
  double   bestCostInternal        = MAX_DOUBLE;
  double           bestCost        = bestCS->cost;
  const SPS&            sps        = *tempCS->sps;
  const int      maxSizeEMT        = EMT_INTER_MAX_CU_WITH_QTBT;
  bool              swapped        = false; // avoid unwanted data copy
  bool             reloadCU        = false;
  const bool considerEmtSecondPass = emtMode && sps.getSpsNext().getUseInterEMT() && partitioner.currArea().lwidth() <= maxSizeEMT && partitioner.currArea().lheight() <= maxSizeEMT;

  int minEMTMode = 0;
  int maxEMTMode = (considerEmtSecondPass?1:0);

  // Not allow very big |MVd| to avoid CABAC crash caused by too large MVd. Normally no impact on coding performance.
  const int maxMvd = 1 << 15;
  const PredictionUnit& pu = *cu->firstPU;
  if (!cu->affine)
  {
    if ((pu.refIdx[0] >= 0 && (pu.mvd[0].getAbsHor() >= maxMvd || pu.mvd[0].getAbsVer() >= maxMvd))
      || (pu.refIdx[1] >= 0 && (pu.mvd[1].getAbsHor() >= maxMvd || pu.mvd[1].getAbsVer() >= maxMvd)))
    {
      return;
    }
  }
  else
  {
    for (int refList = 0; refList < NUM_REF_PIC_LIST_01; refList++)
    {
      if (pu.refIdx[refList] >= 0)
      {
        for (int ctrlP = 1 + (cu->affineType == AFFINEMODEL_6PARAM); ctrlP >= 0; ctrlP--)
        {
          if (pu.mvdAffi[refList][ctrlP].getAbsHor() >= maxMvd || pu.mvdAffi[refList][ctrlP].getAbsVer() >= maxMvd)
          {
            return;
          }
        }
      }
    }
  }

  if( emtMode == 2 )
  {
    minEMTMode = maxEMTMode = (cu->emtFlag?1:0);
  }

  for( int curEmtMode = minEMTMode; curEmtMode <= maxEMTMode; curEmtMode++ )
  {
    if( reloadCU )
    {
      if( bestCost == bestCS->cost ) //The first EMT pass didn't become the bestCS, so we clear the TUs generated
      {
        tempCS->clearTUs();
      }
      else if( false == swapped )
      {
        tempCS->initStructData( encTestMode.qp, encTestMode.lossless );
        tempCS->copyStructure( *bestCS, partitioner.chType );
        tempCS->getPredBuf().copyFrom( bestCS->getPredBuf() );
        bestCost = bestCS->cost;
        cu       = tempCS->getCU( partitioner.chType );
        swapped = true;
      }
      else
      {
        tempCS->clearTUs();
        bestCost = bestCS->cost;
        cu       = tempCS->getCU( partitioner.chType );
      }

      //we need to restart the distortion for the new tempCS, the bit count and the cost
      tempCS->dist     = 0;
      tempCS->fracBits = 0;
      tempCS->cost     = MAX_DOUBLE;
    }

    reloadCU    = true; // enable cu reloading
    cu->skip    = false;
    cu->emtFlag = curEmtMode;

    const bool skipResidual = residualPass == 1;
    m_pcInterSearch->encodeResAndCalcRdInterCU( *tempCS, partitioner, skipResidual );

    xEncodeDontSplit( *tempCS, partitioner );

    xCheckDQP( *tempCS, partitioner );

    if( ETM_INTER_ME == encTestMode.type )
    {
      if( equGBiCost != NULL )
      {
        if( tempCS->cost < (*equGBiCost) && cu->GBiIdx == GBI_DEFAULT )
        {
          (*equGBiCost) = tempCS->cost;
        }
      }
      else
      {
        CHECK(equGBiCost == NULL, "equGBiCost == NULL");
      }
      if( tempCS->slice->getCheckLDC() && !cu->imv && cu->GBiIdx != GBI_DEFAULT && tempCS->cost < m_bestGbiCost[1] )
      {
        if( tempCS->cost < m_bestGbiCost[0] )
        {
          m_bestGbiCost[1] = m_bestGbiCost[0];
          m_bestGbiCost[0] = tempCS->cost;
          m_bestGbiIdx[1] = m_bestGbiIdx[0];
          m_bestGbiIdx[0] = cu->GBiIdx;
        }
        else
        {
          m_bestGbiCost[1] = tempCS->cost;
          m_bestGbiIdx[1] = cu->GBiIdx;
        }
      }
    }

    double emtFirstPassCost = tempCS->cost;
    if( imvCS && (tempCS->cost < imvCS->cost) )
    {
      if( imvCS->cost != MAX_DOUBLE )
      {
        imvCS->initStructData( encTestMode.qp, encTestMode.lossless );
      }
      imvCS->copyStructure( *tempCS, partitioner.chType );
    }
    if( NULL != bestHasNonResi && (bestCostInternal > tempCS->cost) )
    {
      bestCostInternal = tempCS->cost;
      if (!(tempCS->getPU(partitioner.chType)->mhIntraFlag))
      *bestHasNonResi  = !cu->rootCbf;
    }

    if (cu->rootCbf == false)
    {
      if (tempCS->getPU(partitioner.chType)->mhIntraFlag)
      {
        tempCS->cost = MAX_DOUBLE;
        return;
      }
    }

#if WCG_EXT
    DTRACE_MODE_COST( *tempCS, m_pcRdCost->getLambda( true ) );
#else
    DTRACE_MODE_COST( *tempCS, m_pcRdCost->getLambda() );
#endif
    xCheckBestMode( tempCS, bestCS, partitioner, encTestMode );

    //now we check whether the second pass should be skipped or not
    if( !curEmtMode && maxEMTMode )
    {
      const double thresholdToSkipEmtSecondPass = 1.1; // Skip checking EMT transforms
      const bool bCond1 = !cu->firstTU->cbf[COMPONENT_Y];

      const bool bCond3 = emtFirstPassCost > ( bestCost * thresholdToSkipEmtSecondPass );

      if( m_pcEncCfg->getFastInterEMT() && (bCond1 || bCond3 ) ) 
      {
        maxEMTMode = 0; // do not test EMT
      }
    }
  } //end emt loop
}


void EncCu::xEncodeDontSplit( CodingStructure &cs, Partitioner &partitioner )
{
  m_CABACEstimator->resetBits();

  {
    if( partitioner.canSplit( CU_QUAD_SPLIT, cs ) )
    {
      m_CABACEstimator->split_cu_flag( false, cs, partitioner );
    }
    if( partitioner.canSplit( CU_MT_SPLIT, cs ) )
    {
      m_CABACEstimator->split_cu_mode_mt( CU_DONT_SPLIT, cs, partitioner );
    }
  }

  cs.fracBits += m_CABACEstimator->getEstFracBits(); // split bits
  cs.cost      = m_pcRdCost->calcRdCost( cs.fracBits, cs.dist );

}

#if REUSE_CU_RESULTS
void EncCu::xReuseCachedResult( CodingStructure *&tempCS, CodingStructure *&bestCS, Partitioner &partitioner )
{
  const SPS &sps = *tempCS->sps;

  BestEncInfoCache* bestEncCache = dynamic_cast<BestEncInfoCache*>( m_modeCtrl );
  CHECK( !bestEncCache, "If this mode is chosen, mode controller has to implement the mode caching capabilities" );
  EncTestMode cachedMode;

  if( bestEncCache->setCsFrom( *tempCS, cachedMode, partitioner ) )
  {
    CodingUnit& cu = *tempCS->cus.front();
    partitioner.setCUData( cu );

    if( CU::isIntra( cu ) )
    {
      xReconIntraQT( cu );
    }
    else
    {
      xDeriveCUMV( cu );
      xReconInter( cu );
    }

    Distortion finalDistortion = 0;
    const int  numValidComponents = getNumberValidComponents( tempCS->area.chromaFormat );

    for( int comp = 0; comp < numValidComponents; comp++ )
    {
      const ComponentID compID = ComponentID( comp );

      if( CS::isDualITree( *tempCS ) && toChannelType( compID ) != partitioner.chType )
      {
        continue;
      }

      CPelBuf reco = tempCS->getRecoBuf( compID );
      CPelBuf org  = tempCS->getOrgBuf ( compID );

#if WCG_EXT
      if( m_pcEncCfg->getLumaLevelToDeltaQPMapping().isEnabled() )
      {
        const CPelBuf orgLuma = tempCS->getOrgBuf(tempCS->area.blocks[COMPONENT_Y]);
        finalDistortion += m_pcRdCost->getDistPart( org, reco, sps.getBitDepth( toChannelType( compID ) ), compID, DF_SSE_WTD, &orgLuma );
      }
      else
#endif
      finalDistortion += m_pcRdCost->getDistPart( org, reco, sps.getBitDepth( toChannelType( compID ) ), compID, DF_SSE );
    }

    m_CABACEstimator->getCtx() = m_CurrCtx->start;
    m_CABACEstimator->resetBits();

    CUCtx cuCtx;
    cuCtx.isDQPCoded = true;
    cuCtx.isChromaQpAdjCoded = true;
    m_CABACEstimator->coding_unit( cu, partitioner, cuCtx );

    tempCS->dist     = finalDistortion;
    tempCS->fracBits = m_CABACEstimator->getEstFracBits();
    tempCS->cost     = m_pcRdCost->calcRdCost( tempCS->fracBits, tempCS->dist );

    xEncodeDontSplit( *tempCS,         partitioner );
    xCheckDQP       ( *tempCS,         partitioner );
    xCheckBestMode  (  tempCS, bestCS, partitioner, cachedMode );
  }
  else
  {
    THROW( "Should never happen!" );
  }
}

#endif


//! \}