/* The copyright in this software is being made available under the BSD * License, included below. This software may be subject to other third party * and contributor rights, including patent rights, and no such rights are * granted under this license. * * Copyright (c) 2010-2023, ITU/ISO/IEC * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are met: * * * Redistributions of source code must retain the above copyright notice, * this list of conditions and the following disclaimer. * * Redistributions in binary form must reproduce the above copyright notice, * this list of conditions and the following disclaimer in the documentation * and/or other materials provided with the distribution. * * Neither the name of the ITU/ISO/IEC nor the names of its contributors may * be used to endorse or promote products derived from this software without * specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF * THE POSSIBILITY OF SUCH DAMAGE. */ /** \file EncSearch.cpp * \brief encoder intra search class */ #include "IntraSearch.h" #include "EncModeCtrl.h" #include "CommonLib/CommonDef.h" #include "CommonLib/Rom.h" #include "CommonLib/Picture.h" #include "CommonLib/UnitTools.h" #include "CommonLib/dtrace_next.h" #include "CommonLib/dtrace_buffer.h" #include #include //! \ingroup EncoderLib //! \{ #define PLTCtx(c) SubCtx( Ctx::Palette, c ) IntraSearch::IntraSearch() : m_pSplitCS(nullptr) , m_pFullCS(nullptr) , m_pBestCS(nullptr) , m_pcEncCfg(nullptr) , m_pcTrQuant(nullptr) , m_pcRdCost(nullptr) , m_pcReshape(nullptr) , m_CABACEstimator(nullptr) , m_ctxPool(nullptr) , m_isInitialized(false) { for( uint32_t ch = 0; ch < MAX_NUM_TBLOCKS; ch++ ) { m_pSharedPredTransformSkip[ch] = nullptr; } m_minErrorIndexMap = nullptr; for (unsigned i = 0; i < (MAXPLTSIZE + 1); i++) { m_indexError[i] = nullptr; } for (unsigned i = 0; i < NUM_TRELLIS_STATE; i++) { m_statePtRDOQ[i] = nullptr; } } void IntraSearch::destroy() { CHECK( !m_isInitialized, "Not initialized" ); if( m_pcEncCfg ) { const uint32_t numLayersToAllocateSplit = 1; const uint32_t numLayersToAllocateFull = 1; const int numSaveLayersToAllocate = 2; for (uint32_t layer = 0; layer < numSaveLayersToAllocate; layer++) { m_pSaveCS[layer]->destroy(); delete m_pSaveCS[layer]; } const uint32_t numWidths = gp_sizeIdxInfo->numWidths(); const uint32_t numHeights = gp_sizeIdxInfo->numHeights(); for( uint32_t width = 0; width < numWidths; width++ ) { for( uint32_t height = 0; height < numHeights; height++ ) { if( gp_sizeIdxInfo->isCuSize( gp_sizeIdxInfo->sizeFrom( width ) ) && gp_sizeIdxInfo->isCuSize( gp_sizeIdxInfo->sizeFrom( height ) ) ) { for (uint32_t layer = 0; layer < numLayersToAllocateSplit; layer++) { m_pSplitCS[width][height][layer]->destroy(); delete m_pSplitCS[width][height][layer]; } for (uint32_t layer = 0; layer < numLayersToAllocateFull; layer++) { m_pFullCS[width][height][layer]->destroy(); delete m_pFullCS[width][height][layer]; } delete[] m_pSplitCS[width][height]; delete[] m_pFullCS [width][height]; m_pBestCS[width][height]->destroy(); m_pTempCS[width][height]->destroy(); delete m_pTempCS[width][height]; delete m_pBestCS[width][height]; } } delete[] m_pSplitCS[width]; delete[] m_pFullCS [width]; delete[] m_pTempCS[width]; delete[] m_pBestCS[width]; } delete[] m_pSplitCS; delete[] m_pFullCS; delete[] m_pBestCS; delete[] m_pTempCS; delete[] m_pSaveCS; } m_pSplitCS = m_pFullCS = nullptr; m_pBestCS = m_pTempCS = nullptr; m_pSaveCS = nullptr; for( uint32_t ch = 0; ch < MAX_NUM_TBLOCKS; ch++ ) { delete[] m_pSharedPredTransformSkip[ch]; m_pSharedPredTransformSkip[ch] = nullptr; } m_tmpStorageCtu.destroy(); m_colorTransResiBuf.destroy(); m_isInitialized = false; if (m_indexError[0] != nullptr) { for (unsigned i = 0; i < (MAXPLTSIZE + 1); i++) { delete[] m_indexError[i]; m_indexError[i] = nullptr; } } if (m_minErrorIndexMap != nullptr) { delete[] m_minErrorIndexMap; m_minErrorIndexMap = nullptr; } if (m_statePtRDOQ[0] != nullptr) { for (unsigned i = 0; i < NUM_TRELLIS_STATE; i++) { delete[] m_statePtRDOQ[i]; m_statePtRDOQ[i] = nullptr; } } } IntraSearch::~IntraSearch() { if( m_isInitialized ) { destroy(); } } void IntraSearch::init(EncCfg *pcEncCfg, TrQuant *pcTrQuant, RdCost *pcRdCost, CABACWriter *CABACEstimator, CtxPool *ctxPool, const uint32_t maxCUWidth, const uint32_t maxCUHeight, const uint32_t maxTotalCUDepth, EncReshape *pcReshape, const unsigned bitDepthY) { CHECK(m_isInitialized, "Already initialized"); m_pcEncCfg = pcEncCfg; m_pcTrQuant = pcTrQuant; m_pcRdCost = pcRdCost; m_CABACEstimator = CABACEstimator; m_ctxPool = ctxPool; m_pcReshape = pcReshape; const ChromaFormat cform = pcEncCfg->getChromaFormatIdc(); IntraPrediction::init(cform, pcEncCfg->getBitDepth(ChannelType::LUMA)); m_tmpStorageCtu.create(UnitArea(cform, Area(0, 0, MAX_CU_SIZE, MAX_CU_SIZE))); m_colorTransResiBuf.create(UnitArea(cform, Area(0, 0, MAX_CU_SIZE, MAX_CU_SIZE))); for( uint32_t ch = 0; ch < MAX_NUM_TBLOCKS; ch++ ) { m_pSharedPredTransformSkip[ch] = new Pel[MAX_CU_SIZE * MAX_CU_SIZE]; } const uint32_t numWidths = gp_sizeIdxInfo->numWidths(); const uint32_t numHeights = gp_sizeIdxInfo->numHeights(); const uint32_t numLayersToAllocateSplit = 1; const uint32_t numLayersToAllocateFull = 1; m_pBestCS = new CodingStructure**[numWidths]; m_pTempCS = new CodingStructure**[numWidths]; m_pFullCS = new CodingStructure***[numWidths]; m_pSplitCS = new CodingStructure***[numWidths]; for( uint32_t width = 0; width < numWidths; width++ ) { m_pBestCS[width] = new CodingStructure*[numHeights]; m_pTempCS[width] = new CodingStructure*[numHeights]; m_pFullCS [width] = new CodingStructure**[numHeights]; m_pSplitCS[width] = new CodingStructure**[numHeights]; for( uint32_t height = 0; height < numHeights; height++ ) { if( gp_sizeIdxInfo->isCuSize( gp_sizeIdxInfo->sizeFrom( width ) ) && gp_sizeIdxInfo->isCuSize( gp_sizeIdxInfo->sizeFrom( height ) ) ) { m_pBestCS[width][height] = new CodingStructure(m_unitPool); m_pTempCS[width][height] = new CodingStructure(m_unitPool); m_pBestCS[width][height]->create(m_pcEncCfg->getChromaFormatIdc(), Area(0, 0, gp_sizeIdxInfo->sizeFrom(width), gp_sizeIdxInfo->sizeFrom(height)), false, (bool)pcEncCfg->getPLTMode()); m_pTempCS[width][height]->create(m_pcEncCfg->getChromaFormatIdc(), Area(0, 0, gp_sizeIdxInfo->sizeFrom(width), gp_sizeIdxInfo->sizeFrom(height)), false, (bool)pcEncCfg->getPLTMode()); m_pFullCS[width][height] = new CodingStructure *[numLayersToAllocateFull]; m_pSplitCS[width][height] = new CodingStructure *[numLayersToAllocateSplit]; for (uint32_t layer = 0; layer < numLayersToAllocateFull; layer++) { m_pFullCS[width][height][layer] = new CodingStructure(m_unitPool); m_pFullCS[width][height][layer]->create(m_pcEncCfg->getChromaFormatIdc(), Area(0, 0, gp_sizeIdxInfo->sizeFrom(width), gp_sizeIdxInfo->sizeFrom(height)), false, (bool)pcEncCfg->getPLTMode()); } for (uint32_t layer = 0; layer < numLayersToAllocateSplit; layer++) { m_pSplitCS[width][height][layer] = new CodingStructure(m_unitPool); m_pSplitCS[width][height][layer]->create(m_pcEncCfg->getChromaFormatIdc(), Area(0, 0, gp_sizeIdxInfo->sizeFrom(width), gp_sizeIdxInfo->sizeFrom(height)), false, (bool)pcEncCfg->getPLTMode()); } } else { m_pBestCS[width][height] = nullptr; m_pTempCS[width][height] = nullptr; m_pFullCS [width][height] = nullptr; m_pSplitCS[width][height] = nullptr; } } } const int numSaveLayersToAllocate = 2; m_pSaveCS = new CodingStructure *[numSaveLayersToAllocate]; for (uint32_t depth = 0; depth < numSaveLayersToAllocate; depth++) { m_pSaveCS[depth] = new CodingStructure(m_unitPool); m_pSaveCS[depth]->create(UnitArea(cform, Area(0, 0, maxCUWidth, maxCUHeight)), false, (bool)pcEncCfg->getPLTMode()); } m_isInitialized = true; if (pcEncCfg->getPLTMode()) { if (m_indexError[0] == nullptr) { for (unsigned i = 0; i < (MAXPLTSIZE + 1); i++) { m_indexError[i] = new double[MAX_CU_BLKSIZE_PLT*MAX_CU_BLKSIZE_PLT]; } } if (m_minErrorIndexMap == nullptr) { m_minErrorIndexMap = new uint8_t[MAX_CU_BLKSIZE_PLT*MAX_CU_BLKSIZE_PLT]; } if (m_statePtRDOQ[0] == nullptr) { for (unsigned i = 0; i < NUM_TRELLIS_STATE; i++) { m_statePtRDOQ[i] = new uint8_t[MAX_CU_BLKSIZE_PLT*MAX_CU_BLKSIZE_PLT]; } } } } ////////////////////////////////////////////////////////////////////////// // INTRA PREDICTION ////////////////////////////////////////////////////////////////////////// static constexpr double COST_UNKNOWN = -65536.0; double IntraSearch::findInterCUCost( CodingUnit &cu ) { if( cu.isConsIntra() && !cu.slice->isIntra() ) { //search corresponding inter CU cost for( int i = 0; i < m_numCuInSCIPU; i++ ) { if( cu.lumaPos() == m_cuAreaInSCIPU[i].pos() && cu.lumaSize() == m_cuAreaInSCIPU[i].size() ) { return m_cuCostInSCIPU[i]; } } } return COST_UNKNOWN; } #if GDR_ENABLED int IntraSearch::getNumTopRecons(PredictionUnit &pu, int lumaDirMode, bool isChroma) { const int w = isChroma ? pu.Cb().width : pu.Y().width; const int h = isChroma ? pu.Cb().height : pu.Y().height; int numOfTopRecons = w; const int refIdx = pu.multiRefIdx; const int predModeIntra = getModifiedWideAngle(w, h, lumaDirMode); const int isModeVer = predModeIntra >= DIA_IDX; const int intraPredAngleMode = (isModeVer) ? predModeIntra - VER_IDX : -(predModeIntra - HOR_IDX); const int absAngMode = abs(intraPredAngleMode); const int signAng = intraPredAngleMode < 0 ? -1 : 1; const int absAng = (lumaDirMode > DC_IDX && lumaDirMode < NUM_LUMA_MODE) ? angTable[absAngMode] : 0; const int invAngle = invAngTable[absAngMode]; const int intraPredAngle = signAng * absAng; const int sideSize = isModeVer ? h : w; const int maxScale = 2; const int angularScale = std::min(maxScale, floorLog2(sideSize) - (floorLog2(3 * invAngle - 2) - 8)); bool applyPDPC; // 1.0 derive PDPC applyPDPC = (refIdx == 0) ? true : false; if (lumaDirMode > DC_IDX && lumaDirMode < NUM_LUMA_MODE) { if (intraPredAngleMode < 0) { applyPDPC &= false; } else if (intraPredAngleMode > 0) { applyPDPC &= (angularScale >= 0); } } // 2.0 calculate number of recons switch (lumaDirMode) { case PLANAR_IDX: numOfTopRecons = applyPDPC ? (w + 1) : (w + 1); break; case DC_IDX: numOfTopRecons = applyPDPC ? (w) : (w); break; case HOR_IDX: numOfTopRecons = applyPDPC ? (w) : (w); break; case VER_IDX: numOfTopRecons = applyPDPC ? (w) : (w); break; default: // 2..66 // note: There should be a way to reduce the number of top recons, in case of non PDPC applyPDPC |= isChroma; if (predModeIntra >= DIA_IDX) { if (intraPredAngle < 0) { numOfTopRecons = (applyPDPC) ? (w + w) : (w + 1); } else { numOfTopRecons = (applyPDPC) ? (w + w) : (w + w); } } else { if (intraPredAngle < 0) { numOfTopRecons = (applyPDPC) ? (w + w) : (w); } else { numOfTopRecons = (applyPDPC) ? (w + w) : (w); } } break; } return numOfTopRecons; } bool IntraSearch::isValidIntraPredLuma(PredictionUnit &pu, int lumaDirMode) { bool isValid = true; if (pu.cs->picture->gdrParam.inGdrInterval) { int x = pu.Y().x; // count num of recons on the top int virX = pu.cs->picture->gdrParam.verBoundary; int numOfTopRecons = getNumTopRecons(pu, lumaDirMode, false); // check if recon is out of boundary if (x < virX && virX < (x + numOfTopRecons)) { isValid = false; } } return isValid; } bool IntraSearch::isValidIntraPredChroma(PredictionUnit &pu, int lumaDirMode, int chromaDirMode) { bool isValid = true; CodingStructure *cs = pu.cs; if (pu.cs->picture->gdrParam.inGdrInterval) { // note: chroma cordinate int cbX = pu.Cb().x; //int cbY = pu.Cb().y; int cbW = pu.Cb().width; int cbH = pu.Cb().height; int chromaScaleX = getComponentScaleX(COMPONENT_Cb, cs->area.chromaFormat); int chromaScaleY = getComponentScaleY(COMPONENT_Cb, cs->area.chromaFormat); int lumaX = cbX << chromaScaleX; // int lumaY = cbY << chromaScaleY; int lumaW = cbW << chromaScaleX; int lumaH = cbH << chromaScaleY; int numOfTopRecons = lumaW; int virX = pu.cs->picture->gdrParam.verBoundary; // count num of recons on the top switch (chromaDirMode) { case LM_CHROMA_IDX : numOfTopRecons = lumaW; break; case MDLM_L_IDX : numOfTopRecons = lumaW; break; // note: could reduce the actual #of case MDLM_T_IDX: numOfTopRecons = (lumaW + lumaH); break; case DM_CHROMA_IDX: numOfTopRecons = getNumTopRecons(pu, lumaDirMode, true) << chromaScaleX; break; default: numOfTopRecons = getNumTopRecons(pu, chromaDirMode, true) << chromaScaleX; break; } // check if recon is out of boundary if (lumaX < virX && virX < (lumaX + numOfTopRecons)) { isValid = false; } } return isValid; } #endif bool IntraSearch::estIntraPredLumaQT(CodingUnit &cu, Partitioner &partitioner, const double bestCostSoFar, bool mtsCheckRangeFlag, int mtsFirstCheckId, int mtsLastCheckId, bool moreProbMTSIdxFirst, CodingStructure* bestCS) { CodingStructure &cs = *cu.cs; const SPS &sps = *cs.sps; const uint32_t logWidth = floorLog2(partitioner.currArea().lwidth()); const uint32_t logHeight = floorLog2(partitioner.currArea().lheight()); // Lambda calculation at equivalent Qp of 4 is recommended because at that Qp, the quantization divisor is 1. const double sqrtLambdaForFirstPass = m_pcRdCost->getMotionLambda( ) * FRAC_BITS_SCALE; //===== loop over partitions ===== const TempCtx ctxStart(m_ctxPool, m_CABACEstimator->getCtx()); const TempCtx ctxStartMipFlag(m_ctxPool, SubCtx(Ctx::MipFlag, m_CABACEstimator->getCtx())); const TempCtx ctxStartIspMode(m_ctxPool, SubCtx(Ctx::ISPMode, m_CABACEstimator->getCtx())); const TempCtx ctxStartPlanarFlag(m_ctxPool, SubCtx(Ctx::IntraLumaPlanarFlag, m_CABACEstimator->getCtx())); const TempCtx ctxStartIntraMode(m_ctxPool, SubCtx(Ctx::IntraLumaMpmFlag, m_CABACEstimator->getCtx())); const TempCtx ctxStartMrlIdx(m_ctxPool, SubCtx(Ctx::MultiRefLineIdx, m_CABACEstimator->getCtx())); CHECK( !cu.firstPU, "CU has no PUs" ); // variables for saving fast intra modes scan results across multiple LFNST passes bool lfnstLoadFlag = sps.getUseLFNST() && cu.lfnstIdx != 0; bool lfnstSaveFlag = sps.getUseLFNST() && cu.lfnstIdx == 0; lfnstSaveFlag &= sps.getExplicitMtsIntraEnabled() ? cu.mtsFlag == 0 : true; const uint32_t lfnstIdx = cu.lfnstIdx; double costInterCU = findInterCUCost( cu ); const int width = partitioner.currArea().lwidth(); const int height = partitioner.currArea().lheight(); // Marking MTS usage for faster MTS // 0: MTS is either not applicable for current CU (cuWidth > MTS_INTRA_MAX_CU_SIZE or cuHeight > MTS_INTRA_MAX_CU_SIZE), not active in the config file or the fast decision algorithm is not used in this case // 1: MTS fast algorithm can be applied for the current CU, and the DCT2 is being checked // 2: MTS is being checked for current CU. Stored results of DCT2 can be utilized for speedup uint8_t mtsUsageFlag = 0; const int maxSizeEMT = MTS_INTRA_MAX_CU_SIZE; if (width <= maxSizeEMT && height <= maxSizeEMT && sps.getExplicitMtsIntraEnabled()) { mtsUsageFlag = ( sps.getUseLFNST() && cu.mtsFlag == 1 ) ? 2 : 1; } if( width * height < 64 && !m_pcEncCfg->getUseFastLFNST() ) { mtsUsageFlag = 0; } const bool colorTransformIsEnabled = sps.getUseColorTrans() && !CS::isDualITree(cs); const bool isFirstColorSpace = colorTransformIsEnabled && ((m_pcEncCfg->getRGBFormatFlag() && cu.colorTransform) || (!m_pcEncCfg->getRGBFormatFlag() && !cu.colorTransform)); const bool isSecondColorSpace = colorTransformIsEnabled && ((m_pcEncCfg->getRGBFormatFlag() && !cu.colorTransform) || (!m_pcEncCfg->getRGBFormatFlag() && cu.colorTransform)); double bestCurrentCost = bestCostSoFar; bool ispCanBeUsed = sps.getUseISP() && cu.mtsFlag == 0 && cu.lfnstIdx == 0 && CU::canUseISP(width, height, cu.cs->sps->getMaxTbSize()); bool saveDataForISP = ispCanBeUsed && (!colorTransformIsEnabled || isFirstColorSpace); bool testISP = ispCanBeUsed && (!colorTransformIsEnabled || !cu.colorTransform); if ( saveDataForISP ) { //reset the intra modes lists variables m_ispCandList[ISPType::HOR].clear(); m_ispCandList[ISPType::VER].clear(); } if( testISP ) { //reset the variables used for the tests m_regIntraRDListWithCosts.clear(); int numTotalPartsHor = (int)height >> floorLog2(CU::getISPSplitDim(width, height, TU_1D_HORZ_SPLIT)); int numTotalPartsVer = (int)width >> floorLog2(CU::getISPSplitDim(width, height, TU_1D_VERT_SPLIT)); m_ispTestedModes[0].init( numTotalPartsHor, numTotalPartsVer ); //the total number of subpartitions is modified to take into account the cases where LFNST cannot be combined with ISP due to size restrictions numTotalPartsHor = sps.getUseLFNST() && CU::canUseLfnstWithISP(cu.Y(), ISPType::HOR) ? numTotalPartsHor : 0; numTotalPartsVer = sps.getUseLFNST() && CU::canUseLfnstWithISP(cu.Y(), ISPType::VER) ? numTotalPartsVer : 0; for (int j = 1; j < NUM_LFNST_NUM_PER_SET; j++) { m_ispTestedModes[j].init(numTotalPartsHor, numTotalPartsVer); } } const bool testBDPCM = sps.getBDPCMEnabledFlag() && CU::bdpcmAllowed(cu, ComponentID(partitioner.chType)) && cu.mtsFlag == 0 && cu.lfnstIdx == 0; static_vector hadModeList; static_vector candCostList; static_vector candHadList; auto &pu = *cu.firstPU; #if GDR_ENABLED const bool isEncodeGdrClean = cs.sps->getGDREnabledFlag() && cs.pcv->isEncoder && cs.picture->gdrParam.inGdrInterval && cs.isClean(pu.Y().topRight(), ChannelType::LUMA); #endif bool validReturn = false; { candHadList.clear(); candCostList.clear(); hadModeList.clear(); CHECK(pu.cu != &cu, "PU is not contained in the CU"); //===== determine set of modes to be tested (using prediction signal only) ===== int numModesAvailable = NUM_LUMA_MODE; // total number of Intra modes const bool fastMip = sps.getUseMIP() && m_pcEncCfg->getUseFastMIP(); const bool mipAllowed = sps.getUseMIP() && isLuma(partitioner.chType) && ((cu.lfnstIdx == 0) || allowLfnstWithMip(cu.firstPU->lumaSize())); const bool testMip = mipAllowed && !(cu.lwidth() > (8 * cu.lheight()) || cu.lheight() > (8 * cu.lwidth())); const bool supportedMipBlkSize = pu.lwidth() <= MIP_MAX_WIDTH && pu.lheight() <= MIP_MAX_HEIGHT; static_vector rdModeList; int numModesForFullRD = g_intraModeNumFastUseMPM2D[logWidth - MIN_CU_LOG2][logHeight - MIN_CU_LOG2]; if (isSecondColorSpace) { rdModeList.clear(); if (m_numSavedRdModeFirstColorSpace[m_savedRdModeIdx] > 0) { for (int i = 0; i < m_numSavedRdModeFirstColorSpace[m_savedRdModeIdx]; i++) { rdModeList.push_back(m_savedRdModeFirstColorSpace[m_savedRdModeIdx][i]); } } else { return false; } } else { if (mtsUsageFlag != 2) { // this should always be true CHECK(!pu.Y().valid(), "PU is not valid"); bool isFirstLineOfCtu = (((pu.block(COMPONENT_Y).y) & ((pu.cs->sps)->getMaxCUWidth() - 1)) == 0); int numOfPassesExtendRef = ((!sps.getUseMRL() || isFirstLineOfCtu) ? 1 : MRL_NUM_REF_LINES); pu.multiRefIdx = 0; if (numModesForFullRD != numModesAvailable) { CHECK(numModesForFullRD >= numModesAvailable, "Too many modes for full RD search"); const CompArea &area = pu.Y(); PelBuf piOrg = cs.getOrgBuf(area); PelBuf piPred = cs.getPredBuf(area); DistParam distParamSad; DistParam distParamHad; if (cu.slice->getLmcsEnabledFlag() && m_pcReshape->getCTUFlag()) { CompArea tmpArea(COMPONENT_Y, area.chromaFormat, Position(0, 0), area.size()); PelBuf tmpOrg = m_tmpStorageCtu.getBuf(tmpArea); tmpOrg.copyFrom(piOrg); tmpOrg.rspSignal(m_pcReshape->getFwdLUT()); m_pcRdCost->setDistParam(distParamSad, tmpOrg, piPred, sps.getBitDepth(ChannelType::LUMA), COMPONENT_Y, false); // Use SAD cost m_pcRdCost->setDistParam(distParamHad, tmpOrg, piPred, sps.getBitDepth(ChannelType::LUMA), COMPONENT_Y, true); // Use HAD (SATD) cost } else { m_pcRdCost->setDistParam(distParamSad, piOrg, piPred, sps.getBitDepth(ChannelType::LUMA), COMPONENT_Y, false); // Use SAD cost m_pcRdCost->setDistParam(distParamHad, piOrg, piPred, sps.getBitDepth(ChannelType::LUMA), COMPONENT_Y, true); // Use HAD (SATD) cost } distParamSad.applyWeight = false; distParamHad.applyWeight = false; if (testMip && supportedMipBlkSize) { numModesForFullRD += fastMip ? std::max(numModesForFullRD, floorLog2(std::min(pu.lwidth(), pu.lheight())) - 1) : numModesForFullRD; } const int numHadCand = (testMip ? 2 : 1) * 3; //*** Derive (regular) candidates using Hadamard cu.mipFlag = false; //===== init pattern for luma prediction ===== initIntraPatternChType(cu, pu.Y(), true); bool satdChecked[NUM_INTRA_MODE]; std::fill_n(satdChecked, NUM_INTRA_MODE, false); if (!lfnstLoadFlag) { for (int modeIdx = 0; modeIdx < numModesAvailable; modeIdx++) { uint32_t mode = modeIdx; Distortion minSadHad = 0; // Skip checking extended Angular modes in the first round of SATD if (mode > DC_IDX && (mode & 1)) { continue; } satdChecked[mode] = true; pu.intraDir[ChannelType::LUMA] = modeIdx; initPredIntraParams(pu, pu.Y(), sps); predIntraAng(COMPONENT_Y, piPred, pu); // Use the min between SAD and HAD as the cost criterion // SAD is scaled by 2 to align with the scaling of HAD minSadHad += std::min(distParamSad.distFunc(distParamSad) * 2, distParamHad.distFunc(distParamHad)); // NB xFracModeBitsIntra will not affect the mode for chroma that may have already been pre-estimated. m_CABACEstimator->getCtx() = SubCtx( Ctx::MipFlag, ctxStartMipFlag ); m_CABACEstimator->getCtx() = SubCtx( Ctx::ISPMode, ctxStartIspMode ); m_CABACEstimator->getCtx() = SubCtx(Ctx::IntraLumaPlanarFlag, ctxStartPlanarFlag); m_CABACEstimator->getCtx() = SubCtx(Ctx::IntraLumaMpmFlag, ctxStartIntraMode); m_CABACEstimator->getCtx() = SubCtx( Ctx::MultiRefLineIdx, ctxStartMrlIdx ); uint64_t fracModeBits = xFracModeBitsIntra(pu, mode, ChannelType::LUMA); double cost = (double) minSadHad + (double) fracModeBits * sqrtLambdaForFirstPass; DTRACE(g_trace_ctx, D_INTRA_COST, "IntraHAD: %u, %llu, %f (%d)\n", minSadHad, fracModeBits, cost, mode); #if GDR_ENABLED if (!isEncodeGdrClean || isValidIntraPredLuma(pu, mode)) #endif { const ModeInfo mi(false, false, 0, ISPType::NONE, mode); updateCandList(mi, cost, rdModeList, candCostList, numModesForFullRD); updateCandList(mi, double(minSadHad), hadModeList, candHadList, numHadCand); } } if (!sps.getUseMIP() && lfnstSaveFlag) { // save found best modes m_savedNumRdModesLFNST = numModesForFullRD; m_savedRdModeListLFNST = rdModeList; m_savedModeCostLFNST = candCostList; // PBINTRA fast m_savedHadModeListLFNST = hadModeList; m_savedHadListLFNST = candHadList; lfnstSaveFlag = false; } } // NSSTFlag if (!sps.getUseMIP() && lfnstLoadFlag) { // restore saved modes numModesForFullRD = m_savedNumRdModesLFNST; rdModeList = m_savedRdModeListLFNST; candCostList = m_savedModeCostLFNST; // PBINTRA fast hadModeList = m_savedHadModeListLFNST; candHadList = m_savedHadListLFNST; } // !LFNSTFlag if (!(sps.getUseMIP() && lfnstLoadFlag)) { static_vector parentCandList = rdModeList; // Second round of SATD for extended Angular modes #if GDR_ENABLED int nn = numModesForFullRD; if (isEncodeGdrClean) { nn = std::min((int)numModesForFullRD, (int)parentCandList.size()); } for (int modeIdx = 0; modeIdx < nn; modeIdx++) #else for (int modeIdx = 0; modeIdx < numModesForFullRD; modeIdx++) #endif { unsigned parentMode = parentCandList[modeIdx].modeId; if (parentMode > (DC_IDX + 1) && parentMode < (NUM_LUMA_MODE - 1)) { for (int subModeIdx = -1; subModeIdx <= 1; subModeIdx += 2) { unsigned mode = parentMode + subModeIdx; if (!satdChecked[mode]) { pu.intraDir[ChannelType::LUMA] = mode; initPredIntraParams(pu, pu.Y(), sps); predIntraAng(COMPONENT_Y, piPred, pu); // Use the min between SAD and SATD as the cost criterion // SAD is scaled by 2 to align with the scaling of HAD Distortion minSadHad = std::min(distParamSad.distFunc(distParamSad) * 2, distParamHad.distFunc(distParamHad)); // NB xFracModeBitsIntra will not affect the mode for chroma that may have already been // pre-estimated. m_CABACEstimator->getCtx() = SubCtx(Ctx::MipFlag, ctxStartMipFlag); m_CABACEstimator->getCtx() = SubCtx(Ctx::ISPMode, ctxStartIspMode); m_CABACEstimator->getCtx() = SubCtx(Ctx::IntraLumaPlanarFlag, ctxStartPlanarFlag); m_CABACEstimator->getCtx() = SubCtx(Ctx::IntraLumaMpmFlag, ctxStartIntraMode); m_CABACEstimator->getCtx() = SubCtx(Ctx::MultiRefLineIdx, ctxStartMrlIdx); uint64_t fracModeBits = xFracModeBitsIntra(pu, mode, ChannelType::LUMA); double cost = (double) minSadHad + (double) fracModeBits * sqrtLambdaForFirstPass; #if GDR_ENABLED if (!isEncodeGdrClean || isValidIntraPredLuma(pu, mode)) #endif { const ModeInfo mi(false, false, 0, ISPType::NONE, mode); updateCandList(mi, cost, rdModeList, candCostList, numModesForFullRD); updateCandList(mi, double(minSadHad), hadModeList, candHadList, numHadCand); } satdChecked[mode] = true; } } } } if (saveDataForISP) { // we save the regular intra modes list m_ispCandList[ISPType::HOR] = rdModeList; } pu.multiRefIdx = 1; const int numMPMs = NUM_MOST_PROBABLE_MODES; unsigned multiRefMPM[numMPMs]; PU::getIntraMPMs(pu, multiRefMPM); for (int mRefNum = 1; mRefNum < numOfPassesExtendRef; mRefNum++) { int multiRefIdx = MULTI_REF_LINE_IDX[mRefNum]; pu.multiRefIdx = multiRefIdx; { initIntraPatternChType(cu, pu.Y(), true); } for (int x = 1; x < numMPMs; x++) { uint32_t mode = multiRefMPM[x]; { pu.intraDir[ChannelType::LUMA] = mode; initPredIntraParams(pu, pu.Y(), sps); predIntraAng(COMPONENT_Y, piPred, pu); // Use the min between SAD and SATD as the cost criterion // SAD is scaled by 2 to align with the scaling of HAD Distortion minSadHad = std::min(distParamSad.distFunc(distParamSad) * 2, distParamHad.distFunc(distParamHad)); // NB xFracModeBitsIntra will not affect the mode for chroma that may have already been pre-estimated. m_CABACEstimator->getCtx() = SubCtx(Ctx::MipFlag, ctxStartMipFlag); m_CABACEstimator->getCtx() = SubCtx(Ctx::ISPMode, ctxStartIspMode); m_CABACEstimator->getCtx() = SubCtx(Ctx::IntraLumaPlanarFlag, ctxStartPlanarFlag); m_CABACEstimator->getCtx() = SubCtx(Ctx::IntraLumaMpmFlag, ctxStartIntraMode); m_CABACEstimator->getCtx() = SubCtx(Ctx::MultiRefLineIdx, ctxStartMrlIdx); uint64_t fracModeBits = xFracModeBitsIntra(pu, mode, ChannelType::LUMA); double cost = (double) minSadHad + (double) fracModeBits * sqrtLambdaForFirstPass; #if GDR_ENABLED if (!isEncodeGdrClean || isValidIntraPredLuma(pu, mode)) #endif { const ModeInfo mi(false, false, multiRefIdx, ISPType::NONE, mode); updateCandList(mi, cost, rdModeList, candCostList, numModesForFullRD); updateCandList(mi, double(minSadHad), hadModeList, candHadList, numHadCand); } } } } #if GDR_ENABLED if (!isEncodeGdrClean) #endif { CHECKD(rdModeList.size() != numModesForFullRD, "Error: RD mode list size"); } if (lfnstSaveFlag && testMip && !allowLfnstWithMip(cu.firstPU->lumaSize())) // save a different set for the next run { // save found best modes m_savedRdModeListLFNST = rdModeList; m_savedModeCostLFNST = candCostList; // PBINTRA fast m_savedHadModeListLFNST = hadModeList; m_savedHadListLFNST = candHadList; m_savedNumRdModesLFNST = g_intraModeNumFastUseMPM2D[logWidth - MIN_CU_LOG2][logHeight - MIN_CU_LOG2]; m_savedRdModeListLFNST.resize(m_savedNumRdModesLFNST); m_savedModeCostLFNST.resize(m_savedNumRdModesLFNST); // PBINTRA fast m_savedHadModeListLFNST.resize(3); m_savedHadListLFNST.resize(3); lfnstSaveFlag = false; } //*** Derive MIP candidates using Hadamard if (testMip && !supportedMipBlkSize) { // avoid estimation for unsupported blk sizes const int transpOff = MatrixIntraPrediction::getNumModesMip(pu.Y()); const int numModesFull = (transpOff << 1); for (uint32_t modeFull = 0; modeFull < numModesFull; modeFull++) { const bool isTransposed = (modeFull >= transpOff ? true : false); const uint32_t mode = (isTransposed ? modeFull - transpOff : modeFull); numModesForFullRD++; rdModeList.push_back(ModeInfo(true, isTransposed, 0, ISPType::NONE, mode)); candCostList.push_back(0); } } else if (testMip) { cu.mipFlag = true; pu.multiRefIdx = 0; double mipHadCost[MAX_NUM_MIP_MODE] = { MAX_DOUBLE }; initIntraPatternChType(cu, pu.Y()); initIntraMip(pu, pu.Y()); const int transpOff = MatrixIntraPrediction::getNumModesMip(pu.Y()); const int numModesFull = (transpOff << 1); for (uint32_t modeFull = 0; modeFull < numModesFull; modeFull++) { const bool isTransposed = (modeFull >= transpOff ? true : false); const uint32_t mode = (isTransposed ? modeFull - transpOff : modeFull); pu.mipTransposedFlag = isTransposed; pu.intraDir[ChannelType::LUMA] = mode; predIntraMip(COMPONENT_Y, piPred, pu); // Use the min between SAD and HAD as the cost criterion // SAD is scaled by 2 to align with the scaling of HAD Distortion minSadHad = std::min(distParamSad.distFunc(distParamSad) * 2, distParamHad.distFunc(distParamHad)); m_CABACEstimator->getCtx() = SubCtx(Ctx::MipFlag, ctxStartMipFlag); uint64_t fracModeBits = xFracModeBitsIntra(pu, mode, ChannelType::LUMA); double cost = double(minSadHad) + double(fracModeBits) * sqrtLambdaForFirstPass; mipHadCost[modeFull] = cost; DTRACE(g_trace_ctx, D_INTRA_COST, "IntraMIP: %u, %llu, %f (%d)\n", minSadHad, fracModeBits, cost, modeFull); #if GDR_ENABLED if (!isEncodeGdrClean || isValidIntraPredLuma(pu, mode)) #endif { const ModeInfo mi(true, isTransposed, 0, ISPType::NONE, mode); updateCandList(mi, cost, rdModeList, candCostList, numModesForFullRD + 1); updateCandList(mi, 0.8 * double(minSadHad), hadModeList, candHadList, numHadCand); } } const double thresholdHadCost = 1.0 + 1.4 / sqrt((double) (pu.lwidth() * pu.lheight())); reduceHadCandList(rdModeList, candCostList, numModesForFullRD, thresholdHadCost, mipHadCost, pu, fastMip); } if (sps.getUseMIP() && lfnstSaveFlag) { // save found best modes m_savedNumRdModesLFNST = numModesForFullRD; m_savedRdModeListLFNST = rdModeList; m_savedModeCostLFNST = candCostList; // PBINTRA fast m_savedHadModeListLFNST = hadModeList; m_savedHadListLFNST = candHadList; lfnstSaveFlag = false; } } else // if( sps.getUseMIP() && lfnstLoadFlag) { // restore saved modes numModesForFullRD = m_savedNumRdModesLFNST; rdModeList = m_savedRdModeListLFNST; candCostList = m_savedModeCostLFNST; // PBINTRA fast hadModeList = m_savedHadModeListLFNST; candHadList = m_savedHadListLFNST; } if (m_pcEncCfg->getFastUDIUseMPMEnabled()) { const int numMPMs = NUM_MOST_PROBABLE_MODES; unsigned preds[numMPMs]; pu.multiRefIdx = 0; const int numCand = PU::getIntraMPMs(pu, preds); for (int j = 0; j < numCand; j++) { bool mostProbableModeIncluded = false; ModeInfo mostProbableMode(false, false, 0, ISPType::NONE, preds[j]); #if GDR_ENABLED int nn = numModesForFullRD; if (isEncodeGdrClean) { nn = std::min((int) numModesForFullRD, (int) rdModeList.size()); } for (int i = 0; i < nn; i++) #else for (int i = 0; i < numModesForFullRD; i++) #endif { mostProbableModeIncluded |= (mostProbableMode == rdModeList[i]); } #if GDR_ENABLED if (!isEncodeGdrClean && !mostProbableModeIncluded) #else if (!mostProbableModeIncluded) #endif { numModesForFullRD++; rdModeList.push_back(mostProbableMode); candCostList.push_back(0); } } if (saveDataForISP) { // we add the MPMs to the list that contains only regular intra modes for (int j = 0; j < numCand; j++) { bool mostProbableModeIncluded = false; ModeInfo mostProbableMode(false, false, 0, ISPType::NONE, preds[j]); for (const auto &x: m_ispCandList[ISPType::HOR]) { mostProbableModeIncluded |= mostProbableMode == x; } #if GDR_ENABLED if (!isEncodeGdrClean && !mostProbableModeIncluded) #else if (!mostProbableModeIncluded) #endif { m_ispCandList[ISPType::HOR].push_back(mostProbableMode); } } } } } else { THROW("Full search not supported for MIP"); } if (sps.getUseLFNST() && mtsUsageFlag == 1) { // Store the modes to be checked with RD m_savedNumRdModes[lfnstIdx] = numModesForFullRD; std::copy_n(rdModeList.begin(), numModesForFullRD, m_savedRdModeList[lfnstIdx]); } } else // mtsUsage = 2 (here we potentially reduce the number of modes that will be full-RD checked) { if ((m_pcEncCfg->getUseFastLFNST() || !cu.slice->isIntra()) && m_bestModeCostValid[lfnstIdx]) { numModesForFullRD = 0; double thresholdSkipMode = 1.0 + ((cu.lfnstIdx > 0) ? 0.1 : 1.0) * (1.4 / sqrt((double) (width * height))); // Skip checking the modes with much larger R-D cost than the best mode for (int i = 0; i < m_savedNumRdModes[lfnstIdx]; i++) { if (m_modeCostStore[lfnstIdx][i] <= thresholdSkipMode * m_bestModeCostStore[lfnstIdx]) { rdModeList.push_back(m_savedRdModeList[lfnstIdx][i]); numModesForFullRD++; } } } else // this is necessary because we skip the candidates list calculation, since it was already obtained for // the DCT-II. Now we load it { // Restore the modes to be checked with RD numModesForFullRD = m_savedNumRdModes[lfnstIdx]; rdModeList.resize(numModesForFullRD); std::copy_n(m_savedRdModeList[lfnstIdx], m_savedNumRdModes[lfnstIdx], rdModeList.begin()); candCostList.resize(numModesForFullRD); } } #if GDR_ENABLED if (!isEncodeGdrClean) #endif { CHECK(numModesForFullRD != rdModeList.size(), "Inconsistent state!"); } // after this point, don't use numModesForFullRD // PBINTRA fast if (m_pcEncCfg->getUsePbIntraFast() && !cs.slice->isIntra() && rdModeList.size() < numModesAvailable && !cs.slice->getDisableSATDForRD() && (mtsUsageFlag != 2 || lfnstIdx > 0)) { double pbintraRatio = (lfnstIdx > 0) ? 1.25 : PBINTRA_RATIO; int maxSize = -1; ModeInfo bestMipMode; int bestMipIdx = -1; for (int idx = 0; idx < rdModeList.size(); idx++) { if (rdModeList[idx].mipFlg) { bestMipMode = rdModeList[idx]; bestMipIdx = idx; break; } } const int numHadCand = 3; for (int k = numHadCand - 1; k >= 0; k--) { if (candHadList.size() < (k + 1) || candHadList[k] > cs.interHad * pbintraRatio) { maxSize = k; } } if (maxSize > 0) { rdModeList.resize(std::min(rdModeList.size(), maxSize)); if (bestMipIdx >= 0) { if (rdModeList.size() <= bestMipIdx) { rdModeList.push_back(bestMipMode); } } if (saveDataForISP) { m_ispCandList[ISPType::HOR].resize(std::min(m_ispCandList[ISPType::HOR].size(), maxSize)); } } if (maxSize == 0) { cs.dist = std::numeric_limits::max(); cs.interHad = 0; //===== reset context models ===== m_CABACEstimator->getCtx() = SubCtx(Ctx::MipFlag, ctxStartMipFlag); m_CABACEstimator->getCtx() = SubCtx(Ctx::ISPMode, ctxStartIspMode); m_CABACEstimator->getCtx() = SubCtx(Ctx::IntraLumaPlanarFlag, ctxStartPlanarFlag); m_CABACEstimator->getCtx() = SubCtx(Ctx::IntraLumaMpmFlag, ctxStartIntraMode); m_CABACEstimator->getCtx() = SubCtx(Ctx::MultiRefLineIdx, ctxStartMrlIdx); return false; } } } int numNonISPModes = (int) rdModeList.size(); if ( testISP ) { // we reserve positions for ISP in the common full RD list #if GDR_ENABLED if (!isEncodeGdrClean) #endif { const int maxNumRDModesISP = sps.getUseLFNST() ? 16 * NUM_LFNST_NUM_PER_SET : 16; m_curIspLfnstIdx = 0; for (int i = 0; i < maxNumRDModesISP; i++) { rdModeList.push_back(ModeInfo(false, false, 0, ISPType::RESERVED, 0)); } } } //===== check modes (using r-d costs) ===== ModeInfo bestPuMode; BdpcmMode bestBDPCMMode = BdpcmMode::NONE; double bestCostNonBDPCM = MAX_DOUBLE; CodingStructure *csTemp = m_pTempCS[gp_sizeIdxInfo->idxFrom( cu.lwidth() )][gp_sizeIdxInfo->idxFrom( cu.lheight() )]; CodingStructure *csBest = m_pBestCS[gp_sizeIdxInfo->idxFrom( cu.lwidth() )][gp_sizeIdxInfo->idxFrom( cu.lheight() )]; csTemp->slice = cs.slice; csBest->slice = cs.slice; csTemp->initStructData(); csBest->initStructData(); csTemp->picture = cs.picture; csBest->picture = cs.picture; // just to be sure numModesForFullRD = (int) rdModeList.size(); TUIntraSubPartitioner subTuPartitioner( partitioner ); if ( testISP ) { m_modeCtrl->setIspCost( MAX_DOUBLE ); m_modeCtrl->setMtsFirstPassNoIspCost( MAX_DOUBLE ); } int bestLfnstIdx = cu.lfnstIdx; for (int mode = isSecondColorSpace ? 0 : -2 * int(testBDPCM); mode < (int) rdModeList.size(); mode++) { // set CU/PU to luma prediction mode ModeInfo orgMode; if (sps.getUseColorTrans() && !m_pcEncCfg->getRGBFormatFlag() && isSecondColorSpace && mode) { continue; } if (mode < 0 || (isSecondColorSpace && m_savedBDPCMModeFirstColorSpace[m_savedRdModeIdx][mode] != BdpcmMode::NONE)) { cu.bdpcmMode = mode < 0 ? BdpcmMode(-mode) : m_savedBDPCMModeFirstColorSpace[m_savedRdModeIdx][mode]; orgMode = ModeInfo(false, false, 0, ISPType::NONE, cu.bdpcmMode == BdpcmMode::VER ? VER_IDX : HOR_IDX); } else { cu.bdpcmMode = BdpcmMode::NONE; orgMode = rdModeList[mode]; } if (cu.bdpcmMode == BdpcmMode::NONE && rdModeList[mode].ispMod == ISPType::RESERVED) { if (mode == numNonISPModes) // the list needs to be sorted only once { if (m_pcEncCfg->getUseFastISP()) { m_modeCtrl->setBestPredModeDCT2(bestPuMode.modeId, bestPuMode.mipFlg); } if (!xSortISPCandList(bestCurrentCost, csBest->cost, bestPuMode)) { break; } } xGetNextISPMode(rdModeList[mode], (mode > 0 ? &rdModeList[mode - 1] : nullptr), Size(width, height)); if (rdModeList[mode].ispMod == ISPType::RESERVED) { continue; } cu.lfnstIdx = m_curIspLfnstIdx; orgMode = rdModeList[mode]; } cu.mipFlag = orgMode.mipFlg; pu.mipTransposedFlag = orgMode.mipTrFlg; cu.ispMode = orgMode.ispMod; pu.multiRefIdx = orgMode.mRefId; pu.intraDir[ChannelType::LUMA] = orgMode.modeId; CHECK(cu.mipFlag && pu.multiRefIdx, "Error: combination of MIP and MRL not supported"); CHECK(pu.multiRefIdx && (pu.intraDir[ChannelType::LUMA] == PLANAR_IDX), "Error: combination of MRL and Planar mode not supported"); CHECK(cu.ispMode != ISPType::NONE && cu.mipFlag, "Error: combination of ISP and MIP not supported"); CHECK(cu.ispMode != ISPType::NONE && pu.multiRefIdx, "Error: combination of ISP and MRL not supported"); CHECK(cu.ispMode != ISPType::NONE && cu.colorTransform, "Error: combination of ISP and ACT not supported"); pu.intraDir[ChannelType::CHROMA] = cu.colorTransform ? DM_CHROMA_IDX : pu.intraDir[ChannelType::CHROMA]; // set context models m_CABACEstimator->getCtx() = ctxStart; // determine residual for partition cs.initSubStructure( *csTemp, partitioner.chType, cs.area, true ); bool tmpValidReturn = false; if (cu.ispMode != ISPType::NONE) { if ( m_pcEncCfg->getUseFastISP() ) { m_modeCtrl->setISPWasTested(true); } tmpValidReturn = xIntraCodingLumaISP(*csTemp, subTuPartitioner, bestCurrentCost); if (csTemp->tus.size() == 0) { // no TUs were coded csTemp->cost = MAX_DOUBLE; continue; } // we save the data for future tests m_ispTestedModes[m_curIspLfnstIdx].setModeResults( cu.ispMode, (int) orgMode.modeId, (int) csTemp->tus.size(), csTemp->cus[0]->firstTU->cbf[COMPONENT_Y] ? csTemp->cost : MAX_DOUBLE, csBest->cost); csTemp->cost = !tmpValidReturn ? MAX_DOUBLE : csTemp->cost; } else { if (cu.colorTransform) { tmpValidReturn = xRecurIntraCodingACTQT(*csTemp, partitioner, mtsCheckRangeFlag, mtsFirstCheckId, mtsLastCheckId, moreProbMTSIdxFirst); } else { tmpValidReturn = xRecurIntraCodingLumaQT(*csTemp, partitioner, mtsCheckRangeFlag, mtsFirstCheckId, mtsLastCheckId, moreProbMTSIdxFirst); } } if (cu.ispMode == ISPType::NONE && !cu.mtsFlag && !cu.lfnstIdx && cu.bdpcmMode == BdpcmMode::NONE && !pu.multiRefIdx && !cu.mipFlag && testISP) { m_regIntraRDListWithCosts.push_back( ModeInfoWithCost(cu.mipFlag, pu.mipTransposedFlag, pu.multiRefIdx, cu.ispMode, orgMode.modeId, csTemp->cost)); } if (cu.ispMode != ISPType::NONE && !csTemp->cus[0]->firstTU->cbf[COMPONENT_Y]) { csTemp->cost = MAX_DOUBLE; csTemp->costDbOffset = 0; tmpValidReturn = false; } validReturn |= tmpValidReturn; if (sps.getUseLFNST() && mtsUsageFlag == 1 && cu.ispMode == ISPType::NONE && mode >= 0) { m_modeCostStore[lfnstIdx][mode] = tmpValidReturn ? csTemp->cost : (MAX_DOUBLE / 2.0); //(MAX_DOUBLE / 2.0) ?? } DTRACE(g_trace_ctx, D_INTRA_COST, "IntraCost T [x=%d,y=%d,w=%d,h=%d] %f (%d,%d,%d,%d,%d,%d) \n", cu.blocks[0].x, cu.blocks[0].y, (int) width, (int) height, csTemp->cost, orgMode.modeId, orgMode.ispMod, pu.multiRefIdx, cu.mipFlag, cu.lfnstIdx, cu.mtsFlag); if( tmpValidReturn ) { if (isFirstColorSpace) { if (m_pcEncCfg->getRGBFormatFlag() || cu.ispMode == ISPType::NONE) { sortRdModeListFirstColorSpace( orgMode, csTemp->cost, cu.bdpcmMode, m_savedRdModeFirstColorSpace[m_savedRdModeIdx], m_savedRdCostFirstColorSpace[m_savedRdModeIdx], m_savedBDPCMModeFirstColorSpace[m_savedRdModeIdx], m_numSavedRdModeFirstColorSpace[m_savedRdModeIdx]); } } // check r-d cost if( csTemp->cost < csBest->cost ) { std::swap( csTemp, csBest ); bestPuMode = orgMode; bestBDPCMMode = cu.bdpcmMode; if (sps.getUseLFNST() && mtsUsageFlag == 1 && cu.ispMode == ISPType::NONE) { m_bestModeCostStore[ lfnstIdx ] = csBest->cost; //cs.cost; m_bestModeCostValid[ lfnstIdx ] = true; } if( csBest->cost < bestCurrentCost ) { bestCurrentCost = csBest->cost; } if (cu.ispMode != ISPType::NONE) { m_modeCtrl->setIspCost(csBest->cost); bestLfnstIdx = cu.lfnstIdx; } else if ( testISP ) { m_modeCtrl->setMtsFirstPassNoIspCost(csBest->cost); } } if (cu.ispMode == ISPType::NONE && cu.bdpcmMode == BdpcmMode::NONE && csBest->cost < bestCostNonBDPCM) { bestCostNonBDPCM = csBest->cost; } } csTemp->releaseIntermediateData(); if( m_pcEncCfg->getFastLocalDualTreeMode() ) { if( cu.isConsIntra() && !cu.slice->isIntra() && csBest->cost != MAX_DOUBLE && costInterCU != COST_UNKNOWN && mode >= 0 ) { if( m_pcEncCfg->getFastLocalDualTreeMode() == 2 ) { //Note: only try one intra mode, which is especially useful to reduce EncT for LDB case (around 4%) break; } else { if( csBest->cost > costInterCU * 1.5 ) { break; } } } } if (sps.getUseColorTrans() && !CS::isDualITree(cs)) { if ((m_pcEncCfg->getRGBFormatFlag() && !cu.colorTransform) && csBest->cost != MAX_DOUBLE && bestCS->cost != MAX_DOUBLE && mode >= 0) { if (csBest->cost > bestCS->cost) { break; } } } } // Mode loop cu.ispMode = bestPuMode.ispMod; cu.lfnstIdx = bestLfnstIdx; if( validReturn ) { if (cu.colorTransform) { cs.useSubStructure(*csBest, partitioner.chType, pu, true, true, KEEP_PRED_AND_RESI_SIGNALS, KEEP_PRED_AND_RESI_SIGNALS, true); } else { cs.useSubStructure(*csBest, partitioner.chType, pu.singleChan(ChannelType::LUMA), true, true, KEEP_PRED_AND_RESI_SIGNALS, KEEP_PRED_AND_RESI_SIGNALS, true); } } csBest->releaseIntermediateData(); if( validReturn ) { //=== update PU data ==== cu.mipFlag = bestPuMode.mipFlg; pu.mipTransposedFlag = bestPuMode.mipTrFlg; pu.multiRefIdx = bestPuMode.mRefId; pu.intraDir[ChannelType::LUMA] = bestPuMode.modeId; cu.bdpcmMode = bestBDPCMMode; if (cu.colorTransform) { CHECK(pu.intraDir[ChannelType::CHROMA] != DM_CHROMA_IDX, "chroma should use DM mode for adaptive color transform"); } } } //===== reset context models ===== m_CABACEstimator->getCtx() = ctxStart; return validReturn; } void IntraSearch::estIntraPredChromaQT( CodingUnit &cu, Partitioner &partitioner, const double maxCostAllowed ) { const ChromaFormat format = cu.chromaFormat; const uint32_t numberValidComponents = getNumberValidComponents(format); CodingStructure &cs = *cu.cs; const TempCtx ctxStart(m_ctxPool, m_CABACEstimator->getCtx()); cs.setDecomp( cs.area.Cb(), false ); double bestCostSoFar = maxCostAllowed; bool lumaUsesISP = !cu.isSepTree() && cu.ispMode != ISPType::NONE; PartSplit ispType = lumaUsesISP ? CU::getISPType( cu, COMPONENT_Y ) : TU_NO_ISP; CHECK(cu.ispMode != ISPType::NONE && bestCostSoFar < 0, "bestCostSoFar must be positive!"); auto &pu = *cu.firstPU; { uint32_t bestMode = 0; Distortion bestDist = 0; double bestCost = MAX_DOUBLE; BdpcmMode bestBDPCMMode = BdpcmMode::NONE; //----- init mode list ---- { int32_t minMode = 0; int32_t maxMode = NUM_CHROMA_MODE; //----- check chroma modes ----- uint32_t chromaCandModes[ NUM_CHROMA_MODE ]; PU::getIntraChromaCandModes( pu, chromaCandModes ); // create a temporary CS CodingStructure &saveCS = *m_pSaveCS[0]; saveCS.pcv = cs.pcv; saveCS.picture = cs.picture; saveCS.area.repositionTo( cs.area ); saveCS.clearTUs(); if (!cu.isSepTree() && cu.ispMode != ISPType::NONE) { saveCS.clearCUs(); saveCS.clearPUs(); } if( cu.isSepTree() ) { if( partitioner.canSplit( TU_MAX_TR_SPLIT, cs ) ) { partitioner.splitCurrArea( TU_MAX_TR_SPLIT, cs ); do { cs.addTU( CS::getArea( cs, partitioner.currArea(), partitioner.chType ), partitioner.chType ).depth = partitioner.currTrDepth; } while( partitioner.nextPart( cs ) ); partitioner.exitCurrSplit(); } else { cs.addTU(CS::getArea(cs, partitioner.currArea(), partitioner.chType), partitioner.chType); } } auto &orgTUs = m_orgTUs; orgTUs.clear(); if( lumaUsesISP ) { CodingUnit& auxCU = saveCS.addCU( cu, partitioner.chType ); auxCU.ispMode = cu.ispMode; saveCS.sps = cu.cs->sps; saveCS.addPU( *cu.firstPU, partitioner.chType ); } // create a store for the TUs for( const auto &ptu : cs.tus ) { // for split TUs in HEVC, add the TUs without Chroma parts for correct setting of Cbfs if (lumaUsesISP || pu.contains(*ptu, ChannelType::CHROMA)) { saveCS.addTU( *ptu, partitioner.chType ); orgTUs.push_back( ptu ); } } if( lumaUsesISP ) { saveCS.clearCUs(); } // SATD pre-selecting. int satdModeList[NUM_CHROMA_MODE]; int64_t satdSortedCost[NUM_CHROMA_MODE]; for (int i = 0; i < NUM_CHROMA_MODE; i++) { satdSortedCost[i] = 0; // for the mode not pre-select by SATD, do RDO by default, so set the initial value 0. satdModeList[i] = 0; } bool modeIsEnable[NUM_INTRA_MODE + 1]; // use intra mode idx to check whether enable for (int i = 0; i < NUM_INTRA_MODE + 1; i++) { modeIsEnable[i] = 1; } DistParam distParamSad; DistParam distParamSatd; pu.intraDir[ChannelType::CHROMA] = MDLM_L_IDX; // temporary assigned, just to indicate this is a MDLM mode. for luma down-sampling operation. initIntraPatternChType(cu, pu.Cb()); initIntraPatternChType(cu, pu.Cr()); xGetLumaRecPixels(pu, pu.Cb()); for (int idx = minMode; idx <= maxMode - 1; idx++) { int mode = chromaCandModes[idx]; satdModeList[idx] = mode; if (PU::isLMCMode(mode) && (!PU::isLMCModeEnabled(pu, mode) || cu.slice->getDisableLmChromaCheck())) { continue; } if ((mode == LM_CHROMA_IDX) || (mode == PLANAR_IDX) || (mode == DM_CHROMA_IDX)) // only pre-check regular modes and MDLM modes, not including DM ,Planar, and LM { continue; } pu.intraDir[ChannelType::CHROMA] = mode; // temporary assigned, for SATD checking. int64_t sad = 0; int64_t sadCb = 0; int64_t satdCb = 0; int64_t sadCr = 0; int64_t satdCr = 0; CodingStructure& cs = *(pu.cs); CompArea areaCb = pu.Cb(); PelBuf orgCb = cs.getOrgBuf(areaCb); PelBuf predCb = cs.getPredBuf(areaCb); m_pcRdCost->setDistParam(distParamSad, orgCb, predCb, pu.cs->sps->getBitDepth(ChannelType::CHROMA), COMPONENT_Cb, false); m_pcRdCost->setDistParam(distParamSatd, orgCb, predCb, pu.cs->sps->getBitDepth(ChannelType::CHROMA), COMPONENT_Cb, true); distParamSad.applyWeight = false; distParamSatd.applyWeight = false; if (PU::isLMCMode(mode)) { predIntraChromaLM(COMPONENT_Cb, predCb, pu, areaCb, mode); } else { initPredIntraParams(pu, pu.Cb(), *pu.cs->sps); predIntraAng(COMPONENT_Cb, predCb, pu); } sadCb = distParamSad.distFunc(distParamSad) * 2; satdCb = distParamSatd.distFunc(distParamSatd); sad += std::min(sadCb, satdCb); CompArea areaCr = pu.Cr(); PelBuf orgCr = cs.getOrgBuf(areaCr); PelBuf predCr = cs.getPredBuf(areaCr); m_pcRdCost->setDistParam(distParamSad, orgCr, predCr, pu.cs->sps->getBitDepth(ChannelType::CHROMA), COMPONENT_Cr, false); m_pcRdCost->setDistParam(distParamSatd, orgCr, predCr, pu.cs->sps->getBitDepth(ChannelType::CHROMA), COMPONENT_Cr, true); distParamSad.applyWeight = false; distParamSatd.applyWeight = false; if (PU::isLMCMode(mode)) { predIntraChromaLM(COMPONENT_Cr, predCr, pu, areaCr, mode); } else { initPredIntraParams(pu, pu.Cr(), *pu.cs->sps); predIntraAng(COMPONENT_Cr, predCr, pu); } sadCr = distParamSad.distFunc(distParamSad) * 2; satdCr = distParamSatd.distFunc(distParamSatd); sad += std::min(sadCr, satdCr); satdSortedCost[idx] = sad; } // sort the mode based on the cost from small to large. int tempIdx = 0; int64_t tempCost = 0; for (int i = minMode; i <= maxMode - 1; i++) { for (int j = i + 1; j <= maxMode - 1; j++) { if (satdSortedCost[j] < satdSortedCost[i]) { tempIdx = satdModeList[i]; satdModeList[i] = satdModeList[j]; satdModeList[j] = tempIdx; tempCost = satdSortedCost[i]; satdSortedCost[i] = satdSortedCost[j]; satdSortedCost[j] = tempCost; } } } int reducedModeNumber = 2; // reduce the number of chroma modes for (int i = 0; i < reducedModeNumber; i++) { modeIsEnable[satdModeList[maxMode - 1 - i]] = 0; // disable the last reducedModeNumber modes } // save the dist Distortion baseDist = cs.dist; const bool testBDPCM = CU::bdpcmAllowed(cu, COMPONENT_Cb) && cu.ispMode == ISPType::NONE && cu.mtsFlag == 0 && cu.lfnstIdx == 0; for (int32_t mode = minMode - (2 * int(testBDPCM)); mode < maxMode; mode++) { int chromaIntraMode; if (mode < 0) { cu.bdpcmModeChroma = BdpcmMode(-mode); chromaIntraMode = cu.bdpcmModeChroma == BdpcmMode::VER ? chromaCandModes[1] : chromaCandModes[2]; } else { chromaIntraMode = chromaCandModes[mode]; cu.bdpcmModeChroma = BdpcmMode::NONE; if( PU::isLMCMode( chromaIntraMode ) && ( !PU::isLMCModeEnabled( pu, chromaIntraMode ) || cu.slice->getDisableLmChromaCheck() ) ) { continue; } if (!modeIsEnable[chromaIntraMode] && PU::isLMCModeEnabled(pu, chromaIntraMode)) // when CCLM is disable, then MDLM is disable. not use satd checking { continue; } } cs.setDecomp( pu.Cb(), false ); cs.dist = baseDist; //----- restore context models ----- m_CABACEstimator->getCtx() = ctxStart; //----- chroma coding ----- pu.intraDir[ChannelType::CHROMA] = chromaIntraMode; xRecurIntraChromaCodingQT( cs, partitioner, bestCostSoFar, ispType ); if( lumaUsesISP && cs.dist == MAX_UINT ) { continue; } if (cs.sps->getTransformSkipEnabledFlag()) { m_CABACEstimator->getCtx() = ctxStart; } uint64_t fracBits = xGetIntraFracBitsQT( cs, partitioner, false, true, -1, ispType ); Distortion dist = cs.dist; double cost = m_pcRdCost->calcRdCost(fracBits, dist - baseDist); //----- compare ----- #if GDR_ENABLED bool allOk = (cost < bestCost); if (m_pcEncCfg->getGdrEnabled()) { allOk = allOk && bestCost && isValidIntraPredChroma(pu, (int) PU::getCoLocatedIntraLumaMode(pu), chromaIntraMode); } if (allOk) #else if (cost < bestCost) #endif { if (lumaUsesISP && cost < bestCostSoFar) { bestCostSoFar = cost; } for (uint32_t i = getFirstComponentOfChannel(ChannelType::CHROMA); i < numberValidComponents; i++) { const CompArea &area = pu.blocks[i]; saveCS.getRecoBuf ( area ).copyFrom( cs.getRecoBuf ( area ) ); #if KEEP_PRED_AND_RESI_SIGNALS saveCS.getPredBuf ( area ).copyFrom( cs.getPredBuf ( area ) ); saveCS.getResiBuf ( area ).copyFrom( cs.getResiBuf ( area ) ); #endif saveCS.getPredBuf ( area ).copyFrom( cs.getPredBuf (area ) ); cs.picture->getPredBuf( area ).copyFrom( cs.getPredBuf (area ) ); cs.picture->getRecoBuf( area ).copyFrom( cs.getRecoBuf( area ) ); for( uint32_t j = 0; j < saveCS.tus.size(); j++ ) { saveCS.tus[j]->copyComponentFrom( *orgTUs[j], area.compID ); } } bestCost = cost; bestDist = dist; bestMode = chromaIntraMode; bestBDPCMMode = cu.bdpcmModeChroma; } } for (uint32_t i = getFirstComponentOfChannel(ChannelType::CHROMA); i < numberValidComponents; i++) { const CompArea &area = pu.blocks[i]; cs.getRecoBuf ( area ).copyFrom( saveCS.getRecoBuf( area ) ); #if KEEP_PRED_AND_RESI_SIGNALS cs.getPredBuf ( area ).copyFrom( saveCS.getPredBuf( area ) ); cs.getResiBuf ( area ).copyFrom( saveCS.getResiBuf( area ) ); #endif cs.getPredBuf ( area ).copyFrom( saveCS.getPredBuf( area ) ); cs.picture->getPredBuf( area ).copyFrom( cs.getPredBuf ( area ) ); cs.picture->getRecoBuf( area ).copyFrom( cs. getRecoBuf( area ) ); for( uint32_t j = 0; j < saveCS.tus.size(); j++ ) { orgTUs[ j ]->copyComponentFrom( *saveCS.tus[ j ], area.compID ); } } } pu.intraDir[ChannelType::CHROMA] = bestMode; cs.dist = bestDist; cu.bdpcmModeChroma = bestBDPCMMode; } //----- restore context models ----- m_CABACEstimator->getCtx() = ctxStart; if( lumaUsesISP && bestCostSoFar >= maxCostAllowed ) { cu.ispMode = ISPType::NONE; } } void IntraSearch::saveCuAreaCostInSCIPU( Area area, double cost ) { if( m_numCuInSCIPU < NUM_INTER_CU_INFO_SAVE ) { m_cuAreaInSCIPU[m_numCuInSCIPU] = area; m_cuCostInSCIPU[m_numCuInSCIPU] = cost; m_numCuInSCIPU++; } } void IntraSearch::initCuAreaCostInSCIPU() { for( int i = 0; i < NUM_INTER_CU_INFO_SAVE; i++ ) { m_cuAreaInSCIPU[i] = Area(); m_cuCostInSCIPU[i] = 0; } m_numCuInSCIPU = 0; } void IntraSearch::PLTSearch(CodingStructure &cs, Partitioner& partitioner, ComponentID compBegin, uint32_t numComp) { CodingUnit &cu = *cs.getCU(partitioner.chType); TransformUnit &tu = *cs.getTU(partitioner.chType); uint32_t height = cu.block(compBegin).height; uint32_t width = cu.block(compBegin).width; if (m_pcEncCfg->getLmcs() && (cs.slice->getLmcsEnabledFlag() && m_pcReshape->getCTUFlag())) { cs.getPredBuf().copyFrom(cs.getOrgBuf()); cs.getPredBuf().Y().rspSignal(m_pcReshape->getFwdLUT()); } if( cu.isLocalSepTree() ) { cs.prevPLT.curPLTSize[compBegin] = cs.prevPLT.curPLTSize[COMPONENT_Y]; } cu.lastPLTSize[compBegin] = cs.prevPLT.curPLTSize[compBegin]; //derive palette derivePLTLossy(cs, partitioner, compBegin, numComp); reorderPLT(cs, partitioner, compBegin, numComp); bool idxExist[MAXPLTSIZE + 1] = { false }; preCalcPLTIndexRD(cs, partitioner, compBegin, numComp); // Pre-calculate distortions for each pixel double rdCost = MAX_DOUBLE; deriveIndexMap(cs, partitioner, compBegin, numComp, PLT_SCAN_HORTRAV, rdCost, idxExist); // Optimize palette index map (horizontal scan) if ((cu.curPLTSize[compBegin] + cu.useEscape[compBegin]) > 1) { deriveIndexMap(cs, partitioner, compBegin, numComp, PLT_SCAN_VERTRAV, rdCost, idxExist); // Optimize palette index map (vertical scan) } // Remove unused palette entries uint8_t newPLTSize = 0; int idxMapping[MAXPLTSIZE + 1]; memset(idxMapping, -1, sizeof(int) * (MAXPLTSIZE + 1)); for (int i = 0; i < cu.curPLTSize[compBegin]; i++) { if (idxExist[i]) { idxMapping[i] = newPLTSize; newPLTSize++; } } idxMapping[cu.curPLTSize[compBegin]] = cu.useEscape[compBegin]? newPLTSize: -1; if (newPLTSize != cu.curPLTSize[compBegin]) // there exist unused palette entries { // update palette table and reuseflag Pel curPLTtmp[MAX_NUM_COMPONENT][MAXPLTSIZE]; int reuseFlagIdx = 0, curPLTtmpIdx = 0, reuseEntrySize = 0; memset(cu.reuseflag[compBegin], false, sizeof(bool) * MAXPLTPREDSIZE); int compBeginTmp = compBegin; int numCompTmp = numComp; if( cu.isLocalSepTree() ) { memset(cu.reuseflag[COMPONENT_Y], false, sizeof(bool) * MAXPLTPREDSIZE); compBeginTmp = COMPONENT_Y; numCompTmp = getNumberValidComponents(cu.chromaFormat); } for (int curIdx = 0; curIdx < cu.curPLTSize[compBegin]; curIdx++) { if (idxExist[curIdx]) { for (int comp = compBeginTmp; comp < (compBeginTmp + numCompTmp); comp++) { curPLTtmp[comp][curPLTtmpIdx] = cu.curPLT[comp][curIdx]; } // Update reuse flags if (curIdx < cu.reusePLTSize[compBegin]) { bool match = false; for (; reuseFlagIdx < cs.prevPLT.curPLTSize[compBegin]; reuseFlagIdx++) { bool matchTmp = true; for (int comp = compBegin; comp < (compBegin + numComp); comp++) { matchTmp = matchTmp && (curPLTtmp[comp][curPLTtmpIdx] == cs.prevPLT.curPLT[comp][reuseFlagIdx]); } if (matchTmp) { match = true; break; } } if (match) { cu.reuseflag[compBegin][reuseFlagIdx] = true; if( cu.isLocalSepTree() ) { cu.reuseflag[COMPONENT_Y][reuseFlagIdx] = true; } reuseEntrySize++; } } curPLTtmpIdx++; } } cu.reusePLTSize[compBegin] = reuseEntrySize; // update palette table cu.curPLTSize[compBegin] = newPLTSize; if( cu.isLocalSepTree() ) { cu.curPLTSize[COMPONENT_Y] = newPLTSize; } for (int comp = compBeginTmp; comp < (compBeginTmp + numCompTmp); comp++) { memcpy( cu.curPLT[comp], curPLTtmp[comp], sizeof(Pel)*cu.curPLTSize[compBegin]); } } cu.useRotation[compBegin] = m_bestScanRotationMode; int indexMaxSize = cu.useEscape[compBegin] ? (cu.curPLTSize[compBegin] + 1) : cu.curPLTSize[compBegin]; if (indexMaxSize <= 1) { cu.useRotation[compBegin] = false; } //reconstruct pixel PelBuf curPLTIdx = tu.getcurPLTIdx(compBegin); for (uint32_t y = 0; y < height; y++) { for (uint32_t x = 0; x < width; x++) { curPLTIdx.at(x, y) = idxMapping[curPLTIdx.at(x, y)]; if (curPLTIdx.at(x, y) == cu.curPLTSize[compBegin]) { calcPixelPred(cs, partitioner, y, x, compBegin, numComp); } else { for (uint32_t compID = compBegin; compID < (compBegin + numComp); compID++) { CompArea area = cu.blocks[compID]; PelBuf recBuf = cs.getRecoBuf(area); uint32_t scaleX = getComponentScaleX((ComponentID)COMPONENT_Cb, cs.sps->getChromaFormatIdc()); uint32_t scaleY = getComponentScaleY((ComponentID)COMPONENT_Cb, cs.sps->getChromaFormatIdc()); if (compBegin != COMPONENT_Y || compID == COMPONENT_Y) { recBuf.at(x, y) = cu.curPLT[compID][curPLTIdx.at(x, y)]; } else if (compBegin == COMPONENT_Y && compID != COMPONENT_Y && y % (1 << scaleY) == 0 && x % (1 << scaleX) == 0) { recBuf.at(x >> scaleX, y >> scaleY) = cu.curPLT[compID][curPLTIdx.at(x, y)]; } } } } } cs.getPredBuf().fill(0); cs.getResiBuf().fill(0); cs.getOrgResiBuf().fill(0); cs.fracBits = MAX_UINT; cs.cost = MAX_DOUBLE; Distortion distortion = 0; for (uint32_t comp = compBegin; comp < (compBegin + numComp); comp++) { const ComponentID compID = ComponentID(comp); CPelBuf reco = cs.getRecoBuf(compID); CPelBuf org = cs.getOrgBuf(compID); #if WCG_EXT if (m_pcEncCfg->getLumaLevelToDeltaQPMapping().isEnabled() || ( m_pcEncCfg->getLmcs() && (cs.slice->getLmcsEnabledFlag() && m_pcReshape->getCTUFlag()))) { const CPelBuf orgLuma = cs.getOrgBuf(cs.area.blocks[COMPONENT_Y]); if (compID == COMPONENT_Y && !(m_pcEncCfg->getLumaLevelToDeltaQPMapping().isEnabled())) { const CompArea &areaY = cu.Y(); CompArea tmpArea1(COMPONENT_Y, areaY.chromaFormat, Position(0, 0), areaY.size()); PelBuf tmpRecLuma = m_tmpStorageCtu.getBuf(tmpArea1); tmpRecLuma.copyFrom(reco); tmpRecLuma.rspSignal(m_pcReshape->getInvLUT()); distortion += m_pcRdCost->getDistPart(org, tmpRecLuma, cs.sps->getBitDepth(toChannelType(compID)), compID, DFunc::SSE_WTD, &orgLuma); } else { distortion += m_pcRdCost->getDistPart(org, reco, cs.sps->getBitDepth(toChannelType(compID)), compID, DFunc::SSE_WTD, &orgLuma); } } else #endif { distortion += m_pcRdCost->getDistPart(org, reco, cs.sps->getBitDepth(toChannelType(compID)), compID, DFunc::SSE); } } cs.dist += distortion; const CompArea &area = cu.blocks[compBegin]; cs.setDecomp(area); cs.picture->getRecoBuf(area).copyFrom(cs.getRecoBuf(area)); } void IntraSearch::calcPixelPredRD(CodingStructure& cs, Partitioner& partitioner, Pel* orgBuf, Pel* paPixelValue, Pel* paRecoValue, ComponentID compBegin, uint32_t numComp) { CodingUnit &cu = *cs.getCU(partitioner.chType); TransformUnit &tu = *cs.getTU(partitioner.chType); int qp[3]; int qpRem[3]; int qpPer[3]; int quantiserScale[3]; int quantiserRightShift[3]; int rightShiftOffset[3]; int invquantiserRightShift[3]; int add[3]; for (uint32_t ch = compBegin; ch < (compBegin + numComp); ch++) { QpParam cQP(tu, ComponentID(ch)); qp[ch] = cQP.Qp(true); qpRem[ch] = qp[ch] % 6; qpPer[ch] = qp[ch] / 6; quantiserScale[ch] = g_quantScales[0][qpRem[ch]]; quantiserRightShift[ch] = QUANT_SHIFT + qpPer[ch]; rightShiftOffset[ch] = 1 << (quantiserRightShift[ch] - 1); invquantiserRightShift[ch] = IQUANT_SHIFT; add[ch] = 1 << (invquantiserRightShift[ch] - 1); } for (uint32_t ch = compBegin; ch < (compBegin + numComp); ch++) { const int channelBitDepth = cu.cs->sps->getBitDepth(toChannelType((ComponentID)ch)); paPixelValue[ch] = Pel(std::max(0, ((orgBuf[ch] * quantiserScale[ch] + rightShiftOffset[ch]) >> quantiserRightShift[ch]))); assert(paPixelValue[ch] < (1 << (channelBitDepth + 1))); paRecoValue[ch] = (((paPixelValue[ch] * g_invQuantScales[0][qpRem[ch]]) << qpPer[ch]) + add[ch]) >> invquantiserRightShift[ch]; paRecoValue[ch] = Pel(ClipBD(paRecoValue[ch], channelBitDepth));//to be checked } } void IntraSearch::preCalcPLTIndexRD(CodingStructure& cs, Partitioner& partitioner, ComponentID compBegin, uint32_t numComp) { CodingUnit &cu = *cs.getCU(partitioner.chType); uint32_t height = cu.block(compBegin).height; uint32_t width = cu.block(compBegin).width; bool lossless = (m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && cs.slice->isLossless()); CPelBuf orgBuf[3]; for (int comp = compBegin; comp < (compBegin + numComp); comp++) { CompArea area = cu.blocks[comp]; if (m_pcEncCfg->getLmcs() && (cs.slice->getLmcsEnabledFlag() && m_pcReshape->getCTUFlag())) { orgBuf[comp] = cs.getPredBuf(area); } else { orgBuf[comp] = cs.getOrgBuf(area); } } int rasPos; uint32_t scaleX = getComponentScaleX(COMPONENT_Cb, cs.sps->getChromaFormatIdc()); uint32_t scaleY = getComponentScaleY(COMPONENT_Cb, cs.sps->getChromaFormatIdc()); for (uint32_t y = 0; y < height; y++) { for (uint32_t x = 0; x < width; x++) { rasPos = y * width + x;; // chroma discard bool discardChroma = (compBegin == COMPONENT_Y) && (y&scaleY || x&scaleX); Pel curPel[3]; for (int comp = compBegin; comp < (compBegin + numComp); comp++) { uint32_t pX1 = (comp > 0 && compBegin == COMPONENT_Y) ? (x >> scaleX) : x; uint32_t pY1 = (comp > 0 && compBegin == COMPONENT_Y) ? (y >> scaleY) : y; curPel[comp] = orgBuf[comp].at(pX1, pY1); } uint8_t pltIdx = 0; double minError = MAX_DOUBLE; uint8_t bestIdx = 0; for (uint8_t z = 0; z < cu.curPLTSize[compBegin]; z++) { m_indexError[z][rasPos] = minError; } while (pltIdx < cu.curPLTSize[compBegin]) { uint64_t sqrtError = 0; if (lossless) { for (int comp = compBegin; comp < (discardChroma ? 1 : (compBegin + numComp)); comp++) { sqrtError += int64_t(abs(curPel[comp] - cu.curPLT[comp][pltIdx])); } if (sqrtError == 0) { m_indexError[pltIdx][rasPos] = (double) sqrtError; minError = (double) sqrtError; bestIdx = pltIdx; break; } } else { for (int comp = compBegin; comp < (discardChroma ? 1 : (compBegin + numComp)); comp++) { int64_t tmpErr = int64_t(curPel[comp] - cu.curPLT[comp][pltIdx]); if (isChroma((ComponentID) comp)) { sqrtError += uint64_t(tmpErr * tmpErr * ENC_CHROMA_WEIGHTING); } else { sqrtError += tmpErr * tmpErr; } } m_indexError[pltIdx][rasPos] = (double) sqrtError; if (sqrtError < minError) { minError = (double) sqrtError; bestIdx = pltIdx; } } pltIdx++; } Pel paPixelValue[3], paRecoValue[3]; if (!lossless) { calcPixelPredRD(cs, partitioner, curPel, paPixelValue, paRecoValue, compBegin, numComp); } uint64_t error = 0, rate = 0; for (int comp = compBegin; comp < (discardChroma ? 1 : (compBegin + numComp)); comp++) { if (lossless) { rate += getEpExGolombNumBins(curPel[comp], 5); } else { int64_t tmpErr = int64_t(curPel[comp] - paRecoValue[comp]); if (isChroma((ComponentID) comp)) { error += uint64_t(tmpErr * tmpErr * ENC_CHROMA_WEIGHTING); } else { error += tmpErr * tmpErr; } rate += getEpExGolombNumBins(paPixelValue[comp], 5); // encode quantized escape color } } double rdCost = (double)error + m_pcRdCost->getLambda()*(double)rate; m_indexError[cu.curPLTSize[compBegin]][rasPos] = rdCost; if (rdCost < minError) { minError = rdCost; bestIdx = (uint8_t)cu.curPLTSize[compBegin]; } m_minErrorIndexMap[rasPos] = bestIdx; // save the optimal index of the current pixel } } } void IntraSearch::deriveIndexMap(CodingStructure &cs, Partitioner &partitioner, ComponentID compBegin, uint32_t numComp, PLTScanMode pltScanMode, double &minCost, bool *idxExist) { CodingUnit &cu = *cs.getCU(partitioner.chType); TransformUnit &tu = *cs.getTU(partitioner.chType); uint32_t height = cu.block(compBegin).height; uint32_t width = cu.block(compBegin).width; int total = height*width; Pel *runIndex = tu.getPLTIndex(compBegin); bool *runType = tu.getRunTypes(toChannelType(compBegin)); m_scanOrder = g_scanOrder[SCAN_UNGROUPED][pltScanMode ? CoeffScanType::TRAV_VER : CoeffScanType::TRAV_HOR][gp_sizeIdxInfo->idxFrom(width)][gp_sizeIdxInfo->idxFrom(height)]; // Trellis initialization for (int i = 0; i < 2; i++) { memset(m_prevRunTypeRDOQ[i], 0, sizeof(Pel)*NUM_TRELLIS_STATE); memset(m_prevRunPosRDOQ[i], 0, sizeof(int)*NUM_TRELLIS_STATE); memset(m_stateCostRDOQ[i], 0, sizeof (double)*NUM_TRELLIS_STATE); } for (int state = 0; state < NUM_TRELLIS_STATE; state++) { m_statePtRDOQ[state][0] = 0; } // Context modeling const FracBitsAccess& fracBits = m_CABACEstimator->getCtx().getFracBitsAcess(); BinFracBits fracBitsPltCopyFlagIndex[RUN_IDX_THRE + 1]; for (int dist = 0; dist <= RUN_IDX_THRE; dist++) { const unsigned ctxId = DeriveCtx::CtxPltCopyFlag(PLT_RUN_INDEX, dist); fracBitsPltCopyFlagIndex[dist] = fracBits.getFracBitsArray(Ctx::IdxRunModel( ctxId ) ); } BinFracBits fracBitsPltCopyFlagAbove[RUN_IDX_THRE + 1]; for (int dist = 0; dist <= RUN_IDX_THRE; dist++) { const unsigned ctxId = DeriveCtx::CtxPltCopyFlag(PLT_RUN_COPY, dist); fracBitsPltCopyFlagAbove[dist] = fracBits.getFracBitsArray(Ctx::CopyRunModel( ctxId ) ); } const BinFracBits fracBitsPltRunType = fracBits.getFracBitsArray( Ctx::RunTypeFlag() ); // Trellis RDO per CG bool contTrellisRD = true; for (int subSetId = 0; ( subSetId <= (total - 1) >> LOG2_PALETTE_CG_SIZE ) && contTrellisRD; subSetId++) { int minSubPos = subSetId << LOG2_PALETTE_CG_SIZE; int maxSubPos = minSubPos + (1 << LOG2_PALETTE_CG_SIZE); maxSubPos = (maxSubPos > total) ? total : maxSubPos; // if last position is out of the current CU size contTrellisRD = deriveSubblockIndexMap(cs, partitioner, compBegin, pltScanMode, minSubPos, maxSubPos, fracBitsPltRunType, fracBitsPltCopyFlagIndex, fracBitsPltCopyFlagAbove, minCost, (bool) pltScanMode); } if (!contTrellisRD) { return; } // best state at the last scan position double sumRdCost = MAX_DOUBLE; uint8_t bestState = 0; for (uint8_t state = 0; state < NUM_TRELLIS_STATE; state++) { if (m_stateCostRDOQ[0][state] < sumRdCost) { sumRdCost = m_stateCostRDOQ[0][state]; bestState = state; } } bool checkRunTable[MAX_CU_BLKSIZE_PLT * MAX_CU_BLKSIZE_PLT]; uint8_t checkIndexTable[MAX_CU_BLKSIZE_PLT*MAX_CU_BLKSIZE_PLT]; uint8_t bestStateTable [MAX_CU_BLKSIZE_PLT*MAX_CU_BLKSIZE_PLT]; uint8_t nextState = bestState; // best trellis path for (int i = (width*height - 1); i >= 0; i--) { bestStateTable[i] = nextState; int rasterPos = m_scanOrder[i].idx; nextState = m_statePtRDOQ[nextState][rasterPos]; } // reconstruct index and runs based on the state pointers for (int i = 0; i < (width*height); i++) { int rasterPos = m_scanOrder[i].idx; int abovePos = (pltScanMode == PLT_SCAN_HORTRAV) ? m_scanOrder[i].idx - width : m_scanOrder[i].idx - 1; nextState = bestStateTable[i]; if ( nextState == 0 ) // same as the previous { checkRunTable[rasterPos] = checkRunTable[ m_scanOrder[i - 1].idx ]; if ( checkRunTable[rasterPos] == PLT_RUN_INDEX ) { checkIndexTable[rasterPos] = checkIndexTable[m_scanOrder[i - 1].idx]; } else { checkIndexTable[rasterPos] = checkIndexTable[ abovePos ]; } } else if (nextState == 1) // CopyAbove mode { checkRunTable[rasterPos] = PLT_RUN_COPY; checkIndexTable[rasterPos] = checkIndexTable[abovePos]; } else if (nextState == 2) // Index mode { checkRunTable[rasterPos] = PLT_RUN_INDEX; checkIndexTable[rasterPos] = m_minErrorIndexMap[rasterPos]; } } // Escape flag m_bestEscape = false; for (int pos = 0; pos < (width*height); pos++) { uint8_t index = checkIndexTable[pos]; if (index == cu.curPLTSize[compBegin]) { m_bestEscape = true; break; } } // Horizontal scan v.s vertical scan if (sumRdCost < minCost) { cu.useEscape[compBegin] = m_bestEscape; m_bestScanRotationMode = pltScanMode; memset(idxExist, false, sizeof(bool) * (MAXPLTSIZE + 1)); for (int pos = 0; pos < (width*height); pos++) { runIndex[pos] = checkIndexTable[pos]; runType[pos] = checkRunTable[pos]; idxExist[checkIndexTable[pos]] = true; } minCost = sumRdCost; } } bool IntraSearch::deriveSubblockIndexMap(CodingStructure &cs, Partitioner &partitioner, ComponentID compBegin, PLTScanMode pltScanMode, int minSubPos, int maxSubPos, const BinFracBits &fracBitsPltRunType, const BinFracBits *fracBitsPltIndexINDEX, const BinFracBits *fracBitsPltIndexCOPY, const double minCost, bool useRotate) { CodingUnit &cu = *cs.getCU(partitioner.chType); uint32_t height = cu.block(compBegin).height; uint32_t width = cu.block(compBegin).width; int indexMaxValue = cu.curPLTSize[compBegin]; int refId = 0; int currRasterPos, currScanPos, prevScanPos, aboveScanPos, roffset; int log2Width = (pltScanMode == PLT_SCAN_HORTRAV) ? floorLog2(width): floorLog2(height); int buffersize = (pltScanMode == PLT_SCAN_HORTRAV) ? 2*width: 2*height; for (int curPos = minSubPos; curPos < maxSubPos; curPos++) { currRasterPos = m_scanOrder[curPos].idx; prevScanPos = (curPos == 0) ? 0 : (curPos - 1) % buffersize; roffset = (curPos >> log2Width) << log2Width; aboveScanPos = roffset - (curPos - roffset + 1); aboveScanPos %= buffersize; currScanPos = curPos % buffersize; if ((pltScanMode == PLT_SCAN_HORTRAV && curPos < width) || (pltScanMode == PLT_SCAN_VERTRAV && curPos < height)) { aboveScanPos = -1; // first column/row: above row is not valid } // Trellis stats: // 1st state: same as previous scanned sample // 2nd state: Copy_Above mode // 3rd state: Index mode // Loop of current state for ( int curState = 0; curState < NUM_TRELLIS_STATE; curState++ ) { double minRdCost = MAX_DOUBLE; int minState = 0; // best prevState uint8_t bestRunIndex = 0; bool bestRunType = 0; bool bestPrevCodedType = 0; int bestPrevCodedPos = 0; if ( ( curState == 0 && curPos == 0 ) || ( curState == 1 && aboveScanPos < 0 ) ) // state not available { m_stateCostRDOQ[1 - refId][curState] = MAX_DOUBLE; continue; } bool runType = 0; uint8_t runIndex = 0; if ( curState == 1 ) // 2nd state: Copy_Above mode { runType = PLT_RUN_COPY; } else if ( curState == 2 ) // 3rd state: Index mode { runType = PLT_RUN_INDEX; runIndex = m_minErrorIndexMap[currRasterPos]; } // Loop of previous state for ( int stateID = 0; stateID < NUM_TRELLIS_STATE; stateID++ ) { if ( m_stateCostRDOQ[refId][stateID] == MAX_DOUBLE ) { continue; } if ( curState == 0 ) // 1st state: same as previous scanned sample { runType = m_runMapRDOQ[refId][stateID][prevScanPos]; runIndex = ( runType == PLT_RUN_INDEX ) ? m_indexMapRDOQ[refId][stateID][ prevScanPos ] : m_indexMapRDOQ[refId][stateID][ aboveScanPos ]; } else if ( curState == 1 ) // 2nd state: Copy_Above mode { runIndex = m_indexMapRDOQ[refId][stateID][aboveScanPos]; } bool prevRunType = m_runMapRDOQ[refId][stateID][prevScanPos]; uint8_t prevRunIndex = m_indexMapRDOQ[refId][stateID][prevScanPos]; uint8_t aboveRunIndex = (aboveScanPos >= 0) ? m_indexMapRDOQ[refId][stateID][aboveScanPos] : 0; int dist = curPos - m_prevRunPosRDOQ[refId][stateID] - 1; double rdCost = m_stateCostRDOQ[refId][stateID]; if (rdCost >= minRdCost) { continue; } // Calculate Rd cost bool prevCodedRunType = m_prevRunTypeRDOQ[refId][stateID]; int prevCodedPos = m_prevRunPosRDOQ [refId][stateID]; const BinFracBits* fracBitsPt = (m_prevRunTypeRDOQ[refId][stateID] == PLT_RUN_INDEX) ? fracBitsPltIndexINDEX : fracBitsPltIndexCOPY; rdCost += rateDistOptPLT(runType, runIndex, prevRunType, prevRunIndex, aboveRunIndex, prevCodedRunType, prevCodedPos, curPos, (pltScanMode == PLT_SCAN_HORTRAV) ? width : height, dist, indexMaxValue, fracBitsPt, fracBitsPltRunType); if (rdCost < minRdCost) // update minState ( minRdCost ) { minRdCost = rdCost; minState = stateID; bestRunType = runType; bestRunIndex = runIndex; bestPrevCodedType = prevCodedRunType; bestPrevCodedPos = prevCodedPos; } } // Update trellis info of current state m_stateCostRDOQ [1 - refId][curState] = minRdCost; m_prevRunTypeRDOQ[1 - refId][curState] = bestPrevCodedType; m_prevRunPosRDOQ [1 - refId][curState] = bestPrevCodedPos; m_statePtRDOQ[curState][currRasterPos] = minState; int buffer2update = std::min(buffersize, curPos); memcpy(m_indexMapRDOQ[1 - refId][curState], m_indexMapRDOQ[refId][minState], sizeof(uint8_t)*buffer2update); memcpy(m_runMapRDOQ[1 - refId][curState], m_runMapRDOQ[refId][minState], sizeof(bool)*buffer2update); m_indexMapRDOQ[1 - refId][curState][currScanPos] = bestRunIndex; m_runMapRDOQ [1 - refId][curState][currScanPos] = bestRunType; } if (useRotate) // early terminate: Rd cost >= min cost in horizontal scan { if ((m_stateCostRDOQ[1 - refId][0] >= minCost) && (m_stateCostRDOQ[1 - refId][1] >= minCost) && (m_stateCostRDOQ[1 - refId][2] >= minCost) ) { return 0; } } refId = 1 - refId; } return 1; } double IntraSearch::rateDistOptPLT(bool runType, uint8_t runIndex, bool prevRunType, uint8_t prevRunIndex, uint8_t aboveRunIndex, bool &prevCodedRunType, int &prevCodedPos, int scanPos, uint32_t width, int dist, int indexMaxValue, const BinFracBits *IndexfracBits, const BinFracBits &TypefracBits) { double rdCost = 0.0; bool identityFlag = !( (runType != prevRunType) || ( (runType == PLT_RUN_INDEX) && (runIndex != prevRunIndex) ) ); if ( ( !identityFlag && runType == PLT_RUN_INDEX ) || scanPos == 0 ) // encode index value { uint8_t refIndex = (prevRunType == PLT_RUN_INDEX) ? prevRunIndex : aboveRunIndex; refIndex = (scanPos == 0) ? ( indexMaxValue + 1) : refIndex; if ( runIndex == refIndex ) { rdCost = MAX_DOUBLE; return rdCost; } rdCost += m_pcRdCost->getLambda()*(getTruncBinBits((runIndex > refIndex) ? runIndex - 1 : runIndex, (scanPos == 0) ? (indexMaxValue + 1) : indexMaxValue) << SCALE_BITS); } rdCost += m_indexError[runIndex][m_scanOrder[scanPos].idx] * (1 << SCALE_BITS); if (scanPos > 0) { rdCost += m_pcRdCost->getLambda()*( identityFlag ? (IndexfracBits[(dist < RUN_IDX_THRE) ? dist : RUN_IDX_THRE].intBits[1]) : (IndexfracBits[(dist < RUN_IDX_THRE) ? dist : RUN_IDX_THRE].intBits[0] ) ); } if ( !identityFlag && scanPos >= width && prevRunType != PLT_RUN_COPY ) { rdCost += m_pcRdCost->getLambda()*TypefracBits.intBits[runType]; } if (!identityFlag || scanPos == 0) { prevCodedRunType = runType; prevCodedPos = scanPos; } return rdCost; } uint32_t IntraSearch::getEpExGolombNumBins(uint32_t symbol, uint32_t count) { uint32_t numBins = 0; while (symbol >= (uint32_t)(1 << count)) { numBins++; symbol -= 1 << count; count++; } numBins++; numBins += count; assert(numBins <= 32); return numBins; } uint32_t IntraSearch::getTruncBinBits(const uint32_t symbol, const uint32_t numSymbols) { CHECKD(symbol >= numSymbols, "symbol must be less than numSymbols"); const uint32_t thresh = floorLog2(numSymbols); const uint32_t val = 1 << thresh; const uint32_t b = numSymbols - val; return symbol < val - b ? thresh : thresh + 1; } void IntraSearch::calcPixelPred(CodingStructure& cs, Partitioner& partitioner, uint32_t yPos, uint32_t xPos, ComponentID compBegin, uint32_t numComp) { CodingUnit &cu = *cs.getCU(partitioner.chType); TransformUnit &tu = *cs.getTU(partitioner.chType); bool lossless = (m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && cs.slice->isLossless()); CPelBuf orgBuf[3]; for (int comp = compBegin; comp < (compBegin + numComp); comp++) { CompArea area = cu.blocks[comp]; if (m_pcEncCfg->getLmcs() && (cs.slice->getLmcsEnabledFlag() && m_pcReshape->getCTUFlag())) { orgBuf[comp] = cs.getPredBuf(area); } else { orgBuf[comp] = cs.getOrgBuf(area); } } int qp[3]; int qpRem[3]; int qpPer[3]; int quantiserScale[3]; int quantiserRightShift[3]; int rightShiftOffset[3]; int invquantiserRightShift[3]; int add[3]; if (!lossless) { for (uint32_t ch = compBegin; ch < (compBegin + numComp); ch++) { QpParam cQP(tu, ComponentID(ch)); qp[ch] = cQP.Qp(true); qpRem[ch] = qp[ch] % 6; qpPer[ch] = qp[ch] / 6; quantiserScale[ch] = g_quantScales[0][qpRem[ch]]; quantiserRightShift[ch] = QUANT_SHIFT + qpPer[ch]; rightShiftOffset[ch] = 1 << (quantiserRightShift[ch] - 1); invquantiserRightShift[ch] = IQUANT_SHIFT; add[ch] = 1 << (invquantiserRightShift[ch] - 1); } } uint32_t scaleX = getComponentScaleX(COMPONENT_Cb, cs.sps->getChromaFormatIdc()); uint32_t scaleY = getComponentScaleY(COMPONENT_Cb, cs.sps->getChromaFormatIdc()); for (uint32_t ch = compBegin; ch < (compBegin + numComp); ch++) { const int channelBitDepth = cu.cs->sps->getBitDepth(toChannelType((ComponentID)ch)); CompArea area = cu.blocks[ch]; PelBuf recBuf = cs.getRecoBuf(area); PLTescapeBuf escapeValue = tu.getescapeValue((ComponentID)ch); if (compBegin != COMPONENT_Y || ch == 0) { if (lossless) { escapeValue.at(xPos, yPos) = orgBuf[ch].at(xPos, yPos); recBuf.at(xPos, yPos) = orgBuf[ch].at(xPos, yPos); } else { escapeValue.at(xPos, yPos) = std::max( 0, ((orgBuf[ch].at(xPos, yPos) * quantiserScale[ch] + rightShiftOffset[ch]) >> quantiserRightShift[ch])); assert(escapeValue.at(xPos, yPos) < (TCoeff(1) << (channelBitDepth + 1))); TCoeff value = (((escapeValue.at(xPos, yPos) * g_invQuantScales[0][qpRem[ch]]) << qpPer[ch]) + add[ch]) >> invquantiserRightShift[ch]; recBuf.at(xPos, yPos) = Pel(ClipBD(value, channelBitDepth)); // to be checked } } else if (compBegin == COMPONENT_Y && ch > 0 && yPos % (1 << scaleY) == 0 && xPos % (1 << scaleX) == 0) { uint32_t yPosC = yPos >> scaleY; uint32_t xPosC = xPos >> scaleX; if (lossless) { escapeValue.at(xPosC, yPosC) = orgBuf[ch].at(xPosC, yPosC); recBuf.at(xPosC, yPosC) = orgBuf[ch].at(xPosC, yPosC); } else { escapeValue.at(xPosC, yPosC) = std::max( 0, ((orgBuf[ch].at(xPosC, yPosC) * quantiserScale[ch] + rightShiftOffset[ch]) >> quantiserRightShift[ch])); assert(escapeValue.at(xPosC, yPosC) < (TCoeff(1) << (channelBitDepth + 1))); TCoeff value = (((escapeValue.at(xPosC, yPosC) * g_invQuantScales[0][qpRem[ch]]) << qpPer[ch]) + add[ch]) >> invquantiserRightShift[ch]; recBuf.at(xPosC, yPosC) = Pel(ClipBD(value, channelBitDepth)); // to be checked } } } } void IntraSearch::derivePLTLossy(CodingStructure& cs, Partitioner& partitioner, ComponentID compBegin, uint32_t numComp) { CodingUnit &cu = *cs.getCU(partitioner.chType); const int channelBitDepth_L = cs.sps->getBitDepth(ChannelType::LUMA); const int channelBitDepth_C = cs.sps->getBitDepth(ChannelType::CHROMA); bool lossless = (m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && cs.slice->isLossless()); int pcmShiftRight_L = (channelBitDepth_L - PLT_ENCBITDEPTH); int pcmShiftRight_C = (channelBitDepth_C - PLT_ENCBITDEPTH); if (lossless) { pcmShiftRight_L = 0; pcmShiftRight_C = 0; } int maxPltSize = cu.isSepTree() ? MAXPLTSIZE_DUALTREE : MAXPLTSIZE; uint32_t height = cu.block(compBegin).height; uint32_t width = cu.block(compBegin).width; CPelBuf orgBuf[3]; for (int comp = compBegin; comp < (compBegin + numComp); comp++) { CompArea area = cu.blocks[comp]; if (m_pcEncCfg->getLmcs() && (cs.slice->getLmcsEnabledFlag() && m_pcReshape->getCTUFlag())) { orgBuf[comp] = cs.getPredBuf(area); } else { orgBuf[comp] = cs.getOrgBuf(area); } } TransformUnit &tu = *cs.getTU(partitioner.chType); QpParam cQP(tu, compBegin); int qp = cQP.Qp(true) - 6*(channelBitDepth_L - 8); qp = (qp < 0) ? 0 : ((qp > 56) ? 56 : qp); int errorLimit = g_paletteQuant[qp]; if (lossless) { errorLimit = 0; } uint32_t totalSize = height*width; SortingElement *pelList = new SortingElement[totalSize]; SortingElement element; SortingElement *pelListSort = new SortingElement[MAXPLTSIZE + 1]; uint32_t dictMaxSize = maxPltSize; uint32_t idx = 0; int last = -1; uint32_t scaleX = getComponentScaleX(COMPONENT_Cb, cs.sps->getChromaFormatIdc()); uint32_t scaleY = getComponentScaleY(COMPONENT_Cb, cs.sps->getChromaFormatIdc()); for (uint32_t y = 0; y < height; y++) { for (uint32_t x = 0; x < width; x++) { uint32_t org[3], pX, pY; for (int comp = compBegin; comp < (compBegin + numComp); comp++) { pX = (comp > 0 && compBegin == COMPONENT_Y) ? (x >> scaleX) : x; pY = (comp > 0 && compBegin == COMPONENT_Y) ? (y >> scaleY) : y; org[comp] = orgBuf[comp].at(pX, pY); } element.setAll(org, compBegin, numComp); ComponentID tmpCompBegin = compBegin; int tmpNumComp = numComp; if (cs.sps->getChromaFormatIdc() != ChromaFormat::_444 && numComp == 3 && (x != ((x >> scaleX) << scaleX) || (y != ((y >> scaleY) << scaleY)))) { tmpCompBegin = COMPONENT_Y; tmpNumComp = 1; } int besti = last, bestSAD = (last == -1) ? MAX_UINT : pelList[last].getSAD(element, cs.sps->getBitDepths(), tmpCompBegin, tmpNumComp, lossless); if (lossless) { if (bestSAD) { for (int i = idx - 1; i >= 0; i--) { uint32_t sad = pelList[i].getSAD(element, cs.sps->getBitDepths(), tmpCompBegin, tmpNumComp, lossless); if (sad == 0) { bestSAD = sad; besti = i; break; } } } } else { if (bestSAD) { for (int i = idx - 1; i >= 0; i--) { uint32_t sad = pelList[i].getSAD(element, cs.sps->getBitDepths(), tmpCompBegin, tmpNumComp, lossless); if (sad < bestSAD) { bestSAD = sad; besti = i; if (!sad) { break; } } } } } if (besti >= 0 && pelList[besti].almostEqualData(element, errorLimit, cs.sps->getBitDepths(), tmpCompBegin, tmpNumComp, lossless)) { pelList[besti].addElement(element, tmpCompBegin, tmpNumComp); last = besti; } else { pelList[idx].copyDataFrom(element, tmpCompBegin, tmpNumComp); for (int comp = tmpCompBegin; comp < (tmpCompBegin + tmpNumComp); comp++) { pelList[idx].setCnt(1, comp); } last = idx; idx++; } } } if (cs.sps->getChromaFormatIdc() != ChromaFormat::_444 && numComp == 3) { for( int i = 0; i < idx; i++ ) { pelList[i].setCnt( pelList[i].getCnt(COMPONENT_Y) + (pelList[i].getCnt(COMPONENT_Cb) >> 2), MAX_NUM_COMPONENT); } } else { if( compBegin == 0 ) { for( int i = 0; i < idx; i++ ) { pelList[i].setCnt(pelList[i].getCnt(COMPONENT_Y), COMPONENT_Cb); pelList[i].setCnt(pelList[i].getCnt(COMPONENT_Y), COMPONENT_Cr); pelList[i].setCnt(pelList[i].getCnt(COMPONENT_Y), MAX_NUM_COMPONENT); } } else { for( int i = 0; i < idx; i++ ) { pelList[i].setCnt(pelList[i].getCnt(COMPONENT_Cb), COMPONENT_Y); pelList[i].setCnt(pelList[i].getCnt(COMPONENT_Cb), MAX_NUM_COMPONENT); } } } for (int i = 0; i < dictMaxSize; i++) { pelListSort[i].setCnt(0, COMPONENT_Y); pelListSort[i].setCnt(0, COMPONENT_Cb); pelListSort[i].setCnt(0, COMPONENT_Cr); pelListSort[i].setCnt(0, MAX_NUM_COMPONENT); pelListSort[i].resetAll(compBegin, numComp); } //bubble sorting dictMaxSize = 1; for (int i = 0; i < idx; i++) { if( pelList[i].getCnt(MAX_NUM_COMPONENT) > pelListSort[dictMaxSize - 1].getCnt(MAX_NUM_COMPONENT) ) { int j; for (j = dictMaxSize; j > 0; j--) { if (pelList[i].getCnt(MAX_NUM_COMPONENT) > pelListSort[j - 1].getCnt(MAX_NUM_COMPONENT)) { pelListSort[j].copyAllFrom(pelListSort[j - 1], compBegin, numComp); dictMaxSize = std::min(dictMaxSize + 1, (uint32_t)maxPltSize); } else { break; } } pelListSort[j].copyAllFrom(pelList[i], compBegin, numComp); } } uint32_t paletteSize = 0; uint64_t numColorBits = 0; for (int comp = compBegin; comp < (compBegin + numComp); comp++) { numColorBits += (comp > 0) ? channelBitDepth_C : channelBitDepth_L; } const int plt_lambda_shift = (compBegin > 0) ? pcmShiftRight_C : pcmShiftRight_L; double bitCost = m_pcRdCost->getLambda() / (double) (1 << (2 * plt_lambda_shift)) * numColorBits; bool reuseflag[MAXPLTPREDSIZE] = { false }; int run; double reuseflagCost; for (int i = 0; i < maxPltSize; i++) { if( pelListSort[i].getCnt(MAX_NUM_COMPONENT) ) { ComponentID tmpCompBegin = compBegin; int tmpNumComp = numComp; if (cs.sps->getChromaFormatIdc() != ChromaFormat::_444 && numComp == 3 && pelListSort[i].getCnt(COMPONENT_Cb) == 0) { tmpCompBegin = COMPONENT_Y; tmpNumComp = 1; } for( int comp = tmpCompBegin; comp < (tmpCompBegin + tmpNumComp); comp++ ) { int half = pelListSort[i].getCnt(comp) >> 1; cu.curPLT[comp][paletteSize] = (pelListSort[i].getSumData(comp) + half) / pelListSort[i].getCnt(comp); } int best = -1; if( errorLimit ) { double pal[MAX_NUM_COMPONENT], err = 0.0, bestCost = 0.0; for( int comp = tmpCompBegin; comp < (tmpCompBegin + tmpNumComp); comp++ ) { pal[comp] = pelListSort[i].getSumData(comp) / (double)pelListSort[i].getCnt(comp); err = pal[comp] - cu.curPLT[comp][paletteSize]; if( isChroma((ComponentID) comp) ) { bestCost += (err * err * PLT_CHROMA_WEIGHTING) / (1 << (2 * pcmShiftRight_C)) * pelListSort[i].getCnt(comp); } else { bestCost += (err * err) / (1 << (2 * pcmShiftRight_L)) * pelListSort[i].getCnt(comp); } } bestCost += bitCost; for( int t = 0; t < cs.prevPLT.curPLTSize[compBegin]; t++ ) { double cost = 0.0; for( int comp = tmpCompBegin; comp < (tmpCompBegin + tmpNumComp); comp++ ) { err = pal[comp] - cs.prevPLT.curPLT[comp][t]; if( isChroma((ComponentID) comp) ) { cost += (err * err * PLT_CHROMA_WEIGHTING) / (1 << (2 * pcmShiftRight_C)) * pelListSort[i].getCnt(comp); } else { cost += (err * err) / (1 << (2 * pcmShiftRight_L)) * pelListSort[i].getCnt(comp); } } run = 0; for (int t2 = t; t2 >= 0; t2--) { if (!reuseflag[t2]) { run++; } else { break; } } reuseflagCost = m_pcRdCost->getLambda() / (double)(1 << (2 * plt_lambda_shift)) * getEpExGolombNumBins(run ? run + 1 : run, 0); cost += reuseflagCost; if( cost < bestCost ) { best = t; bestCost = cost; } } if( best != -1 ) { for( int comp = tmpCompBegin; comp < (tmpCompBegin + tmpNumComp); comp++ ) { cu.curPLT[comp][paletteSize] = cs.prevPLT.curPLT[comp][best]; } reuseflag[best] = true; } } bool duplicate = false; if( pelListSort[i].getCnt(MAX_NUM_COMPONENT) == 1 && best == -1 ) { duplicate = true; } else { for( int t = 0; t < paletteSize; t++ ) { bool duplicateTmp = true; for( int comp = tmpCompBegin; comp < (tmpCompBegin + tmpNumComp); comp++ ) { duplicateTmp = duplicateTmp && (cu.curPLT[comp][paletteSize] == cu.curPLT[comp][t]); } if( duplicateTmp ) { duplicate = true; break; } } } if( !duplicate ) { if (cs.sps->getChromaFormatIdc() != ChromaFormat::_444 && numComp == 3 && pelListSort[i].getCnt(COMPONENT_Cb) == 0) { if( best != -1 ) { cu.curPLT[COMPONENT_Cb][paletteSize] = cs.prevPLT.curPLT[COMPONENT_Cb][best]; cu.curPLT[COMPONENT_Cr][paletteSize] = cs.prevPLT.curPLT[COMPONENT_Cr][best]; } else { cu.curPLT[COMPONENT_Cb][paletteSize] = 1 << (channelBitDepth_C - 1); cu.curPLT[COMPONENT_Cr][paletteSize] = 1 << (channelBitDepth_C - 1); } } paletteSize++; } } else { break; } } cu.curPLTSize[compBegin] = paletteSize; if( cu.isLocalSepTree() ) { cu.curPLTSize[COMPONENT_Y] = paletteSize; } delete[] pelList; delete[] pelListSort; } // ------------------------------------------------------------------------------------------------------------------- // Intra search // ------------------------------------------------------------------------------------------------------------------- void IntraSearch::xEncIntraHeader(CodingStructure &cs, Partitioner &partitioner, const bool &hasLuma, const bool &hasChroma, const int subTuIdx) { CodingUnit &cu = *cs.getCU( partitioner.chType ); if (hasLuma) { bool isFirst = cu.ispMode != ISPType::NONE ? subTuIdx == 0 : partitioner.currArea().lumaPos() == cs.area.lumaPos(); // CU header if( isFirst ) { if ((!cs.slice->isIntra() || cs.slice->getSPS()->getIBCFlag() || cs.slice->getSPS()->getPLTMode()) && cu.Y().valid()) { m_CABACEstimator->cu_skip_flag( cu ); m_CABACEstimator->pred_mode ( cu ); } if (CU::isPLT(cu)) { return; } } PredictionUnit &pu = *cs.getPU(partitioner.currArea().lumaPos(), partitioner.chType); // luma prediction mode if (isFirst) { if ( !cu.Y().valid()) { m_CABACEstimator->pred_mode( cu ); } m_CABACEstimator->bdpcm_mode( cu, COMPONENT_Y ); m_CABACEstimator->intra_luma_pred_mode( pu ); } } if (hasChroma) { bool isFirst = partitioner.currArea().Cb().valid() && partitioner.currArea().chromaPos() == cs.area.chromaPos(); PredictionUnit &pu = *cs.getPU(partitioner.currArea().chromaPos(), ChannelType::CHROMA); if( isFirst ) { m_CABACEstimator->bdpcm_mode(cu, ComponentID(ChannelType::CHROMA)); m_CABACEstimator->intra_chroma_pred_mode( pu ); } } } void IntraSearch::xEncSubdivCbfQT(CodingStructure &cs, Partitioner &partitioner, const bool &hasLuma, const bool &hasChroma, const int subTuIdx, const PartSplit ispType) { const UnitArea &currArea = partitioner.currArea(); int subTuCounter = subTuIdx; TransformUnit &currTU = *cs.getTU(currArea.block(partitioner.chType), partitioner.chType, subTuCounter); CodingUnit &currCU = *currTU.cu; uint32_t currDepth = partitioner.currTrDepth; const bool subdiv = currTU.depth > currDepth; ComponentID compID = isLuma(partitioner.chType) ? COMPONENT_Y : COMPONENT_Cb; if (partitioner.canSplit(TU_MAX_TR_SPLIT, cs)) { CHECK(!subdiv, "TU split implied"); } else { CHECK(subdiv && currCU.ispMode == ISPType::NONE && isLuma(compID), "No TU subdivision is allowed with QTBT"); } if (hasChroma) { const bool chromaCbfISP = currArea.blocks[COMPONENT_Cb].valid() && currCU.ispMode != ISPType::NONE && !subdiv; if (currCU.ispMode == ISPType::NONE || chromaCbfISP) { const uint32_t numberValidComponents = getNumberValidComponents(currArea.chromaFormat); const uint32_t cbfDepth = (chromaCbfISP ? currDepth - 1 : currDepth); for (uint32_t ch = COMPONENT_Cb; ch < numberValidComponents; ch++) { const ComponentID compID = ComponentID(ch); if (currDepth == 0 || TU::getCbfAtDepth(currTU, compID, currDepth - 1) || chromaCbfISP) { const bool prevCbf = (compID == COMPONENT_Cr ? TU::getCbfAtDepth(currTU, COMPONENT_Cb, currDepth) : false); m_CABACEstimator->cbf_comp(TU::getCbfAtDepth(currTU, compID, currDepth), currArea.blocks[compID], cbfDepth, prevCbf, false, currCU.getBdpcmMode(compID)); } } } } if (subdiv) { if( partitioner.canSplit( TU_MAX_TR_SPLIT, cs ) ) { partitioner.splitCurrArea( TU_MAX_TR_SPLIT, cs ); } else if (currCU.ispMode != ISPType::NONE && isLuma(compID)) { partitioner.splitCurrArea( ispType, cs ); } else { THROW("Cannot perform an implicit split!"); } do { xEncSubdivCbfQT(cs, partitioner, hasLuma, hasChroma, subTuCounter, ispType); subTuCounter += subTuCounter != -1 ? 1 : 0; } while( partitioner.nextPart( cs ) ); partitioner.exitCurrSplit(); } else { //===== Cbfs ===== if (hasLuma) { bool previousCbf = false; bool lastCbfIsInferred = false; if( ispType != TU_NO_ISP ) { bool rootCbfSoFar = false; const uint32_t nTus = currCU.ispMode == ISPType::HOR ? currCU.lheight() >> floorLog2(currTU.lheight()) : currCU.lwidth() >> floorLog2(currTU.lwidth()); if( subTuCounter == nTus - 1 ) { TransformUnit* tuPointer = currCU.firstTU; for( int tuIdx = 0; tuIdx < nTus - 1; tuIdx++ ) { rootCbfSoFar |= TU::getCbfAtDepth( *tuPointer, COMPONENT_Y, currDepth ); tuPointer = tuPointer->next; } if( !rootCbfSoFar ) { lastCbfIsInferred = true; } } if( !lastCbfIsInferred ) { previousCbf = TU::getPrevTuCbfAtDepth( currTU, COMPONENT_Y, partitioner.currTrDepth ); } } if( !lastCbfIsInferred ) { m_CABACEstimator->cbf_comp(TU::getCbfAtDepth(currTU, COMPONENT_Y, currDepth), currTU.Y(), currTU.depth, previousCbf, currCU.ispMode != ISPType::NONE, currCU.getBdpcmMode(COMPONENT_Y)); } } } } void IntraSearch::xEncCoeffQT( CodingStructure &cs, Partitioner &partitioner, const ComponentID compID, const int subTuIdx, const PartSplit ispType, CUCtx* cuCtx ) { const UnitArea &currArea = partitioner.currArea(); int subTuCounter = subTuIdx; TransformUnit &currTU = *cs.getTU(currArea.block(partitioner.chType), partitioner.chType, subTuIdx); uint32_t currDepth = partitioner.currTrDepth; const bool subdiv = currTU.depth > currDepth; if (subdiv) { if (partitioner.canSplit(TU_MAX_TR_SPLIT, cs)) { partitioner.splitCurrArea(TU_MAX_TR_SPLIT, cs); } else if (currTU.cu->ispMode != ISPType::NONE) { partitioner.splitCurrArea( ispType, cs ); } else { THROW("Implicit TU split not available!"); } do { xEncCoeffQT( cs, partitioner, compID, subTuCounter, ispType, cuCtx ); subTuCounter += subTuCounter != -1 ? 1 : 0; } while( partitioner.nextPart( cs ) ); partitioner.exitCurrSplit(); } else { if (currArea.blocks[compID].valid()) { if (compID == COMPONENT_Cr) { const int cbfMask = (TU::getCbf(currTU, COMPONENT_Cb) ? CBF_MASK_CB : 0) + (TU::getCbf(currTU, COMPONENT_Cr) ? CBF_MASK_CR : 0); m_CABACEstimator->joint_cb_cr(currTU, cbfMask); } if (TU::getCbf(currTU, compID)) { if (isLuma(compID)) { m_CABACEstimator->residual_coding(currTU, compID, cuCtx); m_CABACEstimator->mts_idx(*currTU.cu, cuCtx); } else { m_CABACEstimator->residual_coding(currTU, compID); } } } } } uint64_t IntraSearch::xGetIntraFracBitsQT(CodingStructure &cs, Partitioner &partitioner, const bool &hasLuma, const bool &hasChroma, const int subTuIdx, const PartSplit ispType, CUCtx *cuCtx) { m_CABACEstimator->resetBits(); xEncIntraHeader(cs, partitioner, hasLuma, hasChroma, subTuIdx); xEncSubdivCbfQT(cs, partitioner, hasLuma, hasChroma, subTuIdx, ispType); if (hasLuma) { xEncCoeffQT( cs, partitioner, COMPONENT_Y, subTuIdx, ispType, cuCtx ); } if (hasChroma) { xEncCoeffQT( cs, partitioner, COMPONENT_Cb, subTuIdx, ispType ); xEncCoeffQT( cs, partitioner, COMPONENT_Cr, subTuIdx, ispType ); } CodingUnit& cu = *cs.getCU(partitioner.chType); if (cuCtx && hasLuma && cu.isSepTree() && (cu.ispMode == ISPType::NONE || (cu.lfnstIdx && subTuIdx == 0) || (!cu.lfnstIdx && subTuIdx == m_ispTestedModes[cu.lfnstIdx].numTotalParts[cu.ispMode] - 1))) { m_CABACEstimator->residual_lfnst_mode(cu, *cuCtx); } uint64_t fracBits = m_CABACEstimator->getEstFracBits(); return fracBits; } uint64_t IntraSearch::xGetIntraFracBitsQTSingleChromaComponent( CodingStructure &cs, Partitioner &partitioner, const ComponentID compID ) { m_CABACEstimator->resetBits(); if( compID == COMPONENT_Cb ) { //intra mode coding PredictionUnit &pu = *cs.getPU( partitioner.currArea().lumaPos(), partitioner.chType ); m_CABACEstimator->intra_chroma_pred_mode( pu ); //xEncIntraHeader(cs, partitioner, false, true); } CHECK( partitioner.currTrDepth != 1, "error in the depth!" ); const UnitArea &currArea = partitioner.currArea(); TransformUnit &currTU = *cs.getTU(currArea.block(partitioner.chType), partitioner.chType); //cbf coding const bool prevCbf = ( compID == COMPONENT_Cr ? TU::getCbfAtDepth( currTU, COMPONENT_Cb, partitioner.currTrDepth ) : false ); m_CABACEstimator->cbf_comp(TU::getCbfAtDepth(currTU, compID, partitioner.currTrDepth), currArea.blocks[compID], partitioner.currTrDepth - 1, prevCbf, false, currTU.cu->getBdpcmMode(compID)); //coeffs coding and cross comp coding if( TU::getCbf( currTU, compID ) ) { m_CABACEstimator->residual_coding( currTU, compID ); } uint64_t fracBits = m_CABACEstimator->getEstFracBits(); return fracBits; } uint64_t IntraSearch::xGetIntraFracBitsQTChroma(TransformUnit& currTU, const ComponentID &compID) { m_CABACEstimator->resetBits(); // Include Cbf and jointCbCr flags here as we make decisions across components if ( currTU.jointCbCr ) { const bool cbfMaskCb = TU::getCbf(currTU, COMPONENT_Cb); const bool cbfMaskCr = TU::getCbf(currTU, COMPONENT_Cr); const int cbfMask = (cbfMaskCb ? CBF_MASK_CB : 0) + (cbfMaskCr ? CBF_MASK_CR : 0); m_CABACEstimator->cbf_comp(cbfMaskCb, currTU.blocks[COMPONENT_Cb], currTU.depth, false, false, currTU.cu->getBdpcmMode(COMPONENT_Cb)); m_CABACEstimator->cbf_comp(cbfMaskCr, currTU.blocks[COMPONENT_Cr], currTU.depth, cbfMaskCb, false, currTU.cu->getBdpcmMode(COMPONENT_Cr)); if (cbfMask != 0) { m_CABACEstimator->joint_cb_cr( currTU, cbfMask ); } if (cbfMaskCb) { m_CABACEstimator->residual_coding( currTU, COMPONENT_Cb ); } if (cbfMaskCr) { m_CABACEstimator->residual_coding( currTU, COMPONENT_Cr ); } } else { if ( compID == COMPONENT_Cb ) { m_CABACEstimator->cbf_comp(TU::getCbf(currTU, compID), currTU.blocks[compID], currTU.depth, false, false, currTU.cu->getBdpcmMode(compID)); } else { const bool cbCbf = TU::getCbf( currTU, COMPONENT_Cb ); const bool crCbf = TU::getCbf( currTU, compID ); const int cbfMask = (cbCbf ? CBF_MASK_CB : 0) + (crCbf ? CBF_MASK_CR : 0); m_CABACEstimator->cbf_comp(crCbf, currTU.blocks[compID], currTU.depth, cbCbf, false, currTU.cu->getBdpcmMode(compID)); m_CABACEstimator->joint_cb_cr( currTU, cbfMask ); } } if( !currTU.jointCbCr && TU::getCbf( currTU, compID ) ) { m_CABACEstimator->residual_coding( currTU, compID ); } uint64_t fracBits = m_CABACEstimator->getEstFracBits(); return fracBits; } void IntraSearch::xIntraCodingTUBlock(TransformUnit &tu, const ComponentID &compID, Distortion &dist, const int &default0Save1Load2, uint32_t *numSig, TrModeList *trModes, const bool loadTr) { if (!tu.blocks[compID].valid()) { return; } CodingStructure &cs = *tu.cs; m_pcRdCost->setChromaFormat(cs.sps->getChromaFormatIdc()); const CompArea &area = tu.blocks[compID]; const SPS &sps = *cs.sps; const ChannelType chType = toChannelType(compID); const int bitDepth = sps.getBitDepth(chType); PelBuf piOrg = cs.getOrgBuf (area); PelBuf piPred = cs.getPredBuf (area); PelBuf piResi = cs.getResiBuf (area); PelBuf piReco = cs.getRecoBuf (area); const PredictionUnit &pu = *cs.getPU(area.pos(), chType); const uint32_t chFinalMode = PU::getFinalIntraMode(pu, chType); //===== init availability pattern ===== CHECK( tu.jointCbCr && compID == COMPONENT_Cr, "wrong combination of compID and jointCbCr" ); bool jointCbCr = tu.jointCbCr && compID == COMPONENT_Cb; if (compID == COMPONENT_Y) { PelBuf sharedPredTS( m_pSharedPredTransformSkip[compID], area ); if( default0Save1Load2 != 2 ) { bool predRegDiffFromTB = CU::isPredRegDiffFromTB(*tu.cu, compID); bool firstTBInPredReg = CU::isFirstTBInPredReg(*tu.cu, compID, area); CompArea areaPredReg(COMPONENT_Y, tu.chromaFormat, area); if (tu.cu->ispMode != ISPType::NONE && isLuma(compID)) { if (predRegDiffFromTB) { if (firstTBInPredReg) { CU::adjustPredArea(areaPredReg); initIntraPatternChTypeISP(*tu.cu, areaPredReg, piReco); } } else { initIntraPatternChTypeISP(*tu.cu, area, piReco); } } else { initIntraPatternChType(*tu.cu, area); } //===== get prediction signal ===== if (compID != COMPONENT_Y && tu.cu->bdpcmModeChroma == BdpcmMode::NONE && PU::isLMCMode(chFinalMode)) { xGetLumaRecPixels( pu, area ); predIntraChromaLM(compID, piPred, pu, area, chFinalMode); } else { if( PU::isMIP( pu, chType ) ) { initIntraMip( pu, area ); predIntraMip( compID, piPred, pu ); } else { if (predRegDiffFromTB) { if (firstTBInPredReg) { PelBuf piPredReg = cs.getPredBuf(areaPredReg); predIntraAng(compID, piPredReg, pu); } } else { predIntraAng(compID, piPred, pu); } } } // save prediction if( default0Save1Load2 == 1 ) { sharedPredTS.copyFrom( piPred ); } } else { // load prediction piPred.copyFrom( sharedPredTS ); } } DTRACE(g_trace_ctx, D_PRED, "@(%4d,%4d) [%2dx%2d] IMode=%d\n", tu.lx(), tu.ly(), tu.lwidth(), tu.lheight(), chFinalMode); //DTRACE_PEL_BUF( D_PRED, piPred, tu, tu.cu->predMode, COMPONENT_Y ); const Slice &slice = *cs.slice; bool flag = slice.getLmcsEnabledFlag() && (slice.isIntra() || (!slice.isIntra() && m_pcReshape->getCTUFlag())); if (isLuma(compID)) { //===== get residual signal ===== piResi.copyFrom( piOrg ); if (slice.getLmcsEnabledFlag() && m_pcReshape->getCTUFlag() && compID == COMPONENT_Y) { CompArea tmpArea(COMPONENT_Y, area.chromaFormat, Position(0, 0), area.size()); PelBuf tmpPred = m_tmpStorageCtu.getBuf(tmpArea); tmpPred.copyFrom(piPred); piResi.rspSignal(m_pcReshape->getFwdLUT()); piResi.subtract(tmpPred); } else { piResi.subtract( piPred ); } } //===== transform and quantization ===== //--- init rate estimation arrays for RDOQ --- //--- transform and quantization --- TCoeff absSum = 0; const QpParam cQP(tu, compID); #if RDOQ_CHROMA_LAMBDA m_pcTrQuant->selectLambda(compID); #endif flag =flag && (tu.blocks[compID].width*tu.blocks[compID].height > 4); if (flag && isChroma(compID) && slice.getPicHeader()->getLmcsChromaResidualScaleFlag() ) { int cResScaleInv = tu.getChromaAdj(); double cResScale = (double)(1 << CSCALE_FP_PREC) / (double)cResScaleInv; m_pcTrQuant->setLambda(m_pcTrQuant->getLambda() / (cResScale*cResScale)); } PelBuf crOrg; PelBuf crPred; PelBuf crResi; PelBuf crReco; if (isChroma(compID)) { const CompArea &crArea = tu.blocks[ COMPONENT_Cr ]; crOrg = cs.getOrgBuf ( crArea ); crPred = cs.getPredBuf ( crArea ); crResi = cs.getResiBuf ( crArea ); crReco = cs.getRecoBuf ( crArea ); } if ( jointCbCr ) { // Lambda is loosened for the joint mode with respect to single modes as the same residual is used for both chroma blocks const int absIct = abs( TU::getICTMode(tu) ); const double lfact = ( absIct == 1 || absIct == 3 ? 0.8 : 0.5 ); m_pcTrQuant->setLambda( lfact * m_pcTrQuant->getLambda() ); } if ( sps.getJointCbCrEnabledFlag() && isChroma(compID) && (tu.cu->cs->slice->getSliceQp() > 18) ) { m_pcTrQuant->setLambda( 1.3 * m_pcTrQuant->getLambda() ); } if( isLuma(compID) ) { if (trModes) { m_pcTrQuant->transformNxN(tu, compID, cQP, *trModes, m_pcEncCfg->getMTSIntraMaxCand()); tu.mtsIdx[compID] = trModes->at(0).first; } if (!(m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && slice.isLossless() && tu.mtsIdx[compID] == MtsType::DCT2_DCT2) || tu.cu->bdpcmMode != BdpcmMode::NONE) { m_pcTrQuant->transformNxN(tu, compID, cQP, absSum, m_CABACEstimator->getCtx(), loadTr); } DTRACE(g_trace_ctx, D_TU_ABS_SUM, "%d: comp=%d, abssum=%d\n", DTRACE_GET_COUNTER(g_trace_ctx, D_TU_ABS_SUM), compID, absSum); if (tu.cu->ispMode != ISPType::NONE && isLuma(compID) && CU::isISPLast(*tu.cu, area, area.compID) && CU::allLumaCBFsAreZero(*tu.cu)) { // ISP has to have at least one non-zero CBF dist = MAX_INT; return; } if (m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && slice.isLossless() && tu.mtsIdx[compID] == MtsType::DCT2_DCT2 && BdpcmMode::NONE == tu.cu->bdpcmMode) { absSum = 0; tu.getCoeffs(compID).fill(0); TU::setCbfAtDepth(tu, compID, tu.depth, 0); } //--- inverse transform --- if (absSum > 0) { m_pcTrQuant->invTransformNxN(tu, compID, piResi, cQP); } else { piResi.fill(0); } } else // chroma { ComponentID codeCompId = tu.jointCbCr != 0 ? ((tu.jointCbCr & CBF_MASK_CB) != 0 ? COMPONENT_Cb : COMPONENT_Cr) : compID; const QpParam qpCbCr(tu, codeCompId); if( tu.jointCbCr ) { ComponentID otherCompId = ( codeCompId==COMPONENT_Cr ? COMPONENT_Cb : COMPONENT_Cr ); tu.getCoeffs( otherCompId ).fill(0); // do we need that? TU::setCbfAtDepth (tu, otherCompId, tu.depth, false ); } PelBuf& codeResi = ( codeCompId == COMPONENT_Cr ? crResi : piResi ); absSum = 0; if (trModes) { m_pcTrQuant->transformNxN(tu, codeCompId, qpCbCr, *trModes, m_pcEncCfg->getMTSIntraMaxCand()); tu.mtsIdx[codeCompId] = trModes->at(0).first; if (tu.jointCbCr) { tu.mtsIdx[(codeCompId == COMPONENT_Cr) ? COMPONENT_Cb : COMPONENT_Cr] = MtsType::DCT2_DCT2; } } // encoder bugfix: Set loadTr to aovid redundant transform process if (!(m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && slice.isLossless() && tu.mtsIdx[compID] == MtsType::DCT2_DCT2) || tu.cu->bdpcmModeChroma != BdpcmMode::NONE) { m_pcTrQuant->transformNxN(tu, codeCompId, qpCbCr, absSum, m_CABACEstimator->getCtx(), loadTr); } if ((m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && slice.isLossless() && tu.mtsIdx[compID] == MtsType::DCT2_DCT2) && BdpcmMode::NONE == tu.cu->bdpcmModeChroma) { absSum = 0; tu.getCoeffs(compID).fill(0); TU::setCbfAtDepth(tu, compID, tu.depth, 0); } DTRACE(g_trace_ctx, D_TU_ABS_SUM, "%d: comp=%d, abssum=%d\n", DTRACE_GET_COUNTER(g_trace_ctx, D_TU_ABS_SUM), codeCompId, absSum); int codedCbfMask = 0; if (absSum > 0) { m_pcTrQuant->invTransformNxN(tu, codeCompId, codeResi, qpCbCr); codedCbfMask += codeCompId == COMPONENT_Cb ? CBF_MASK_CB : CBF_MASK_CR; } else { codeResi.fill(0); } if( tu.jointCbCr ) { if (tu.jointCbCr == 3 && codedCbfMask == CBF_MASK_CB) { codedCbfMask = CBF_MASK_CBCR; TU::setCbfAtDepth (tu, COMPONENT_Cr, tu.depth, true ); } if( tu.jointCbCr != codedCbfMask ) { dist = std::numeric_limits::max(); return; } m_pcTrQuant->invTransformICT( tu, piResi, crResi ); absSum = codedCbfMask; } } //===== reconstruction ===== if (flag && absSum > 0 && isChroma(compID) && slice.getPicHeader()->getLmcsChromaResidualScaleFlag()) { piResi.scaleSignal(tu.getChromaAdj(), 0, tu.cu->cs->slice->clpRng(compID)); if( jointCbCr ) { crResi.scaleSignal(tu.getChromaAdj(), 0, tu.cu->cs->slice->clpRng(COMPONENT_Cr)); } } if (slice.getLmcsEnabledFlag() && m_pcReshape->getCTUFlag() && compID == COMPONENT_Y) { CompArea tmpArea(COMPONENT_Y, area.chromaFormat, Position(0,0), area.size()); PelBuf tmpPred = m_tmpStorageCtu.getBuf(tmpArea); tmpPred.copyFrom(piPred); piReco.reconstruct(tmpPred, piResi, cs.slice->clpRng(compID)); } else { piReco.reconstruct(piPred, piResi, cs.slice->clpRng( compID )); if( jointCbCr ) { crReco.reconstruct(crPred, crResi, cs.slice->clpRng( COMPONENT_Cr )); } } //===== update distortion ===== #if WCG_EXT if (m_pcEncCfg->getLumaLevelToDeltaQPMapping().isEnabled() || (m_pcEncCfg->getLmcs() && slice.getLmcsEnabledFlag() && (m_pcReshape->getCTUFlag() || (isChroma(compID) && m_pcEncCfg->getReshapeIntraCMD())))) { const CPelBuf orgLuma = cs.getOrgBuf( cs.area.blocks[COMPONENT_Y] ); if (compID == COMPONENT_Y && !(m_pcEncCfg->getLumaLevelToDeltaQPMapping().isEnabled())) { CompArea tmpArea1(COMPONENT_Y, area.chromaFormat, Position(0, 0), area.size()); PelBuf tmpRecLuma = m_tmpStorageCtu.getBuf(tmpArea1); tmpRecLuma.copyFrom(piReco); tmpRecLuma.rspSignal(m_pcReshape->getInvLUT()); dist += m_pcRdCost->getDistPart(piOrg, tmpRecLuma, sps.getBitDepth(toChannelType(compID)), compID, DFunc::SSE_WTD, &orgLuma); } else { dist += m_pcRdCost->getDistPart(piOrg, piReco, bitDepth, compID, DFunc::SSE_WTD, &orgLuma); if( jointCbCr ) { dist += m_pcRdCost->getDistPart(crOrg, crReco, bitDepth, COMPONENT_Cr, DFunc::SSE_WTD, &orgLuma); } } } else #endif { dist += m_pcRdCost->getDistPart(piOrg, piReco, bitDepth, compID, DFunc::SSE); if( jointCbCr ) { dist += m_pcRdCost->getDistPart(crOrg, crReco, bitDepth, COMPONENT_Cr, DFunc::SSE); } } } void IntraSearch::xIntraCodingACTTUBlock(TransformUnit &tu, const ComponentID &compID, Distortion &dist, TrModeList *trModes, const bool loadTr) { if (!tu.blocks[compID].valid()) { THROW("tu does not exist"); } CodingStructure &cs = *tu.cs; const SPS &sps = *cs.sps; const Slice &slice = *cs.slice; const CompArea &area = tu.blocks[compID]; const CompArea &crArea = tu.blocks[COMPONENT_Cr]; PelBuf piOrgResi = cs.getOrgResiBuf(area); PelBuf piResi = cs.getResiBuf(area); PelBuf crOrgResi = cs.getOrgResiBuf(crArea); PelBuf crResi = cs.getResiBuf(crArea); TCoeff absSum = 0; CHECK(tu.jointCbCr && compID == COMPONENT_Cr, "wrong combination of compID and jointCbCr"); bool jointCbCr = tu.jointCbCr && compID == COMPONENT_Cb; m_pcRdCost->setChromaFormat(cs.sps->getChromaFormatIdc()); if (m_pcEncCfg->getCostMode() != COST_LOSSLESS_CODING || !slice.isLossless()) m_pcTrQuant->lambdaAdjustColorTrans(true); if (jointCbCr) { ComponentID compIdCode = (tu.jointCbCr >> 1 ? COMPONENT_Cb : COMPONENT_Cr); m_pcTrQuant->selectLambda(compIdCode); } else { m_pcTrQuant->selectLambda(compID); } bool flag = slice.getLmcsEnabledFlag() && (slice.isIntra() || (!slice.isIntra() && m_pcReshape->getCTUFlag())) && (tu.blocks[compID].width*tu.blocks[compID].height > 4); if (flag && isChroma(compID) && slice.getPicHeader()->getLmcsChromaResidualScaleFlag()) { int cResScaleInv = tu.getChromaAdj(); double cResScale = (double)(1 << CSCALE_FP_PREC) / (double)cResScaleInv; m_pcTrQuant->setLambda(m_pcTrQuant->getLambda() / (cResScale*cResScale)); } if (jointCbCr) { // Lambda is loosened for the joint mode with respect to single modes as the same residual is used for both chroma blocks const int absIct = abs(TU::getICTMode(tu)); const double lfact = (absIct == 1 || absIct == 3 ? 0.8 : 0.5); m_pcTrQuant->setLambda(lfact * m_pcTrQuant->getLambda()); } if (sps.getJointCbCrEnabledFlag() && isChroma(compID) && (slice.getSliceQp() > 18)) { m_pcTrQuant->setLambda(1.3 * m_pcTrQuant->getLambda()); } if (isLuma(compID)) { QpParam cQP(tu, compID); if (trModes) { m_pcTrQuant->transformNxN(tu, compID, cQP, *trModes, m_pcEncCfg->getMTSIntraMaxCand()); tu.mtsIdx[compID] = trModes->at(0).first; } if (!(m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && slice.isLossless() && tu.mtsIdx[compID] == MtsType::DCT2_DCT2) || tu.cu->bdpcmMode != BdpcmMode::NONE) { m_pcTrQuant->transformNxN(tu, compID, cQP, absSum, m_CABACEstimator->getCtx(), loadTr); } if ((m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && slice.isLossless() && tu.mtsIdx[compID] == MtsType::DCT2_DCT2) && tu.cu->bdpcmMode == BdpcmMode::NONE) { absSum = 0; tu.getCoeffs(compID).fill(0); TU::setCbfAtDepth(tu, compID, tu.depth, 0); } if (absSum > 0) { m_pcTrQuant->invTransformNxN(tu, compID, piResi, cQP); } else { piResi.fill(0); } } else { int codedCbfMask = 0; ComponentID codeCompId = (tu.jointCbCr ? (tu.jointCbCr >> 1 ? COMPONENT_Cb : COMPONENT_Cr) : compID); QpParam qpCbCr(tu, codeCompId); if (tu.jointCbCr) { ComponentID otherCompId = (codeCompId == COMPONENT_Cr ? COMPONENT_Cb : COMPONENT_Cr); tu.getCoeffs(otherCompId).fill(0); TU::setCbfAtDepth(tu, otherCompId, tu.depth, false); } PelBuf& codeResi = (codeCompId == COMPONENT_Cr ? crResi : piResi); absSum = 0; if (trModes) { m_pcTrQuant->transformNxN(tu, codeCompId, qpCbCr, *trModes, m_pcEncCfg->getMTSIntraMaxCand()); tu.mtsIdx[codeCompId] = trModes->at(0).first; if (tu.jointCbCr) { tu.mtsIdx[(codeCompId == COMPONENT_Cr) ? COMPONENT_Cb : COMPONENT_Cr] = MtsType::DCT2_DCT2; } } if (!(m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && slice.isLossless() && tu.mtsIdx[codeCompId] == MtsType::DCT2_DCT2) || tu.cu->bdpcmModeChroma != BdpcmMode::NONE) { m_pcTrQuant->transformNxN(tu, codeCompId, qpCbCr, absSum, m_CABACEstimator->getCtx(), loadTr); } if (absSum > 0) { m_pcTrQuant->invTransformNxN(tu, codeCompId, codeResi, qpCbCr); codedCbfMask += codeCompId == COMPONENT_Cb ? CBF_MASK_CB : CBF_MASK_CR; } else { codeResi.fill(0); } if (tu.jointCbCr) { if (tu.jointCbCr == 3 && codedCbfMask == CBF_MASK_CB) { codedCbfMask = CBF_MASK_CBCR; TU::setCbfAtDepth(tu, COMPONENT_Cr, tu.depth, true); } if (tu.jointCbCr != codedCbfMask) { dist = std::numeric_limits::max(); if (m_pcEncCfg->getCostMode() != COST_LOSSLESS_CODING || !slice.isLossless()) m_pcTrQuant->lambdaAdjustColorTrans(false); return; } m_pcTrQuant->invTransformICT(tu, piResi, crResi); absSum = codedCbfMask; } } if (m_pcEncCfg->getCostMode() != COST_LOSSLESS_CODING || !slice.isLossless()) { m_pcTrQuant->lambdaAdjustColorTrans(false); } dist += m_pcRdCost->getDistPart(piOrgResi, piResi, sps.getBitDepth(toChannelType(compID)), compID, DFunc::SSE); if (jointCbCr) { dist += m_pcRdCost->getDistPart(crOrgResi, crResi, sps.getBitDepth(toChannelType(COMPONENT_Cr)), COMPONENT_Cr, DFunc::SSE); } } bool IntraSearch::xIntraCodingLumaISP(CodingStructure& cs, Partitioner& partitioner, const double bestCostSoFar) { int subTuCounter = 0; const CodingUnit& cu = *cs.getCU(partitioner.currArea().lumaPos(), partitioner.chType); bool earlySkipISP = false; bool splitCbfLuma = false; const PartSplit ispType = CU::getISPType(cu, COMPONENT_Y); cs.cost = 0; partitioner.splitCurrArea(ispType, cs); CUCtx cuCtx; cuCtx.isDQPCoded = true; cuCtx.isChromaQpAdjCoded = true; do // subpartitions loop { uint32_t numSig = 0; Distortion singleDistTmpLuma = 0; uint64_t singleTmpFracBits = 0; double singleCostTmp = 0; TransformUnit& tu = cs.addTU(CS::getArea(cs, partitioner.currArea(), partitioner.chType), partitioner.chType); tu.depth = partitioner.currTrDepth; // Encode TU xIntraCodingTUBlock(tu, COMPONENT_Y, singleDistTmpLuma, 0, &numSig); if (singleDistTmpLuma == MAX_INT) // all zero CBF skip { earlySkipISP = true; partitioner.exitCurrSplit(); cs.cost = MAX_DOUBLE; return false; } if (m_pcRdCost->calcRdCost(cs.fracBits, cs.dist + singleDistTmpLuma) > bestCostSoFar) { // The accumulated cost + distortion is already larger than the best cost so far, so it is not necessary to // calculate the rate earlySkipISP = true; } else { singleTmpFracBits = xGetIntraFracBitsQT(cs, partitioner, true, false, subTuCounter, ispType, &cuCtx); } singleCostTmp = m_pcRdCost->calcRdCost(singleTmpFracBits, singleDistTmpLuma); cs.cost += singleCostTmp; cs.dist += singleDistTmpLuma; cs.fracBits += singleTmpFracBits; subTuCounter++; splitCbfLuma |= TU::getCbfAtDepth(*cs.getTU(partitioner.currArea().lumaPos(), partitioner.chType, subTuCounter - 1), COMPONENT_Y, partitioner.currTrDepth); int nSubPartitions = m_ispTestedModes[cu.lfnstIdx].numTotalParts[cu.ispMode]; if (subTuCounter < nSubPartitions) { // exit condition if the accumulated cost is already larger than the best cost so far (no impact in RD performance) if (cs.cost > bestCostSoFar) { earlySkipISP = true; break; } else if (subTuCounter < nSubPartitions) { // more restrictive exit condition double threshold = nSubPartitions == 2 ? 0.95 : subTuCounter == 1 ? 0.83 : 0.91; if (subTuCounter < nSubPartitions && cs.cost > bestCostSoFar * threshold) { earlySkipISP = true; break; } } } } while (partitioner.nextPart(cs)); // subpartitions loop partitioner.exitCurrSplit(); const UnitArea& currArea = partitioner.currArea(); const uint32_t currDepth = partitioner.currTrDepth; if (earlySkipISP) { cs.cost = MAX_DOUBLE; } else { cs.cost = m_pcRdCost->calcRdCost(cs.fracBits, cs.dist); // The cost check is necessary here again to avoid superfluous operations if the maximum number of coded subpartitions was reached and yet ISP did not win if (cs.cost < bestCostSoFar) { cs.setDecomp(cu.Y()); cs.picture->getRecoBuf(currArea.Y()).copyFrom(cs.getRecoBuf(currArea.Y())); for (auto& ptu : cs.tus) { if (currArea.Y().contains(ptu->Y())) { TU::setCbfAtDepth(*ptu, COMPONENT_Y, currDepth, splitCbfLuma ? 1 : 0); } } } else { earlySkipISP = true; } } return !earlySkipISP; } bool IntraSearch::xRecurIntraCodingLumaQT( CodingStructure &cs, Partitioner &partitioner, bool mtsCheckRangeFlag, int mtsFirstCheckId, int mtsLastCheckId, bool moreProbMTSIdxFirst ) { const UnitArea &currArea = partitioner.currArea(); const CodingUnit &cu = *cs.getCU( currArea.lumaPos(), partitioner.chType ); uint32_t currDepth = partitioner.currTrDepth; const SPS &sps = *cs.sps; bool checkFull = !partitioner.canSplit(TU_MAX_TR_SPLIT, cs); bool checkSplit = partitioner.canSplit(TU_MAX_TR_SPLIT, cs); const Slice &slice = *cs.slice; CHECK(cu.ispMode != ISPType::NONE, "Use the function xIntraCodingLumaISP for ISP cases."); uint32_t numSig = 0; double singleCost = MAX_DOUBLE; Distortion singleDistLuma = 0; uint64_t singleFracBits = 0; bool checkTransformSkip = sps.getTransformSkipEnabledFlag(); uint8_t nNumTransformCands = cu.mtsFlag ? 4 : 1; uint8_t numTransformIndexCands = nNumTransformCands; std::array bestModeIds; bestModeIds.fill(0); const TempCtx ctxStart(m_ctxPool, m_CABACEstimator->getCtx()); TempCtx ctxBest(m_ctxPool); CodingStructure *csSplit = nullptr; CodingStructure *csFull = nullptr; CUCtx cuCtx; cuCtx.isDQPCoded = true; cuCtx.isChromaQpAdjCoded = true; if (checkSplit) { csSplit = &cs; } else if (checkFull) { csFull = &cs; } bool validReturnFull = false; if (checkFull) { csFull->cost = 0.0; TransformUnit &tu = csFull->addTU( CS::getArea( *csFull, currArea, partitioner.chType ), partitioner.chType ); tu.depth = currDepth; const bool tsAllowed = TU::isTSAllowed( tu, COMPONENT_Y ); const bool mtsAllowed = CU::isMTSAllowed( cu, COMPONENT_Y ); TrModeList trModes; if( sps.getUseLFNST() ) { checkTransformSkip &= tsAllowed; checkTransformSkip &= !cu.mtsFlag; checkTransformSkip &= !cu.lfnstIdx; if( !cu.mtsFlag && checkTransformSkip ) { trModes.push_back(TrMode(MtsType::DCT2_DCT2, true)); // DCT2 trModes.push_back(TrMode(MtsType::SKIP, true)); // TS } } else { nNumTransformCands = 1 + ( tsAllowed ? 1 : 0 ) + ( mtsAllowed ? 4 : 0 ); // DCT + TS + 4 MTS = 6 tests if (m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && slice.isLossless()) { nNumTransformCands = 1; CHECK(!tsAllowed && cu.bdpcmMode == BdpcmMode::NONE, "transform skip should be enabled for LS"); if (cu.bdpcmMode != BdpcmMode::NONE) { trModes.push_back(TrMode(MtsType::DCT2_DCT2, true)); } else { trModes.push_back(TrMode(MtsType::SKIP, true)); } } else { trModes.push_back(TrMode(MtsType::DCT2_DCT2, true)); // DCT2 if (tsAllowed) { trModes.push_back(TrMode(MtsType::SKIP, true)); } if (mtsAllowed) { for (MtsType mtsIdx = MtsType::DST7_DST7; mtsIdx < MtsType::NUM; mtsIdx++) { trModes.push_back(TrMode(mtsIdx, true)); } } } } CHECK( !tu.Y().valid(), "Invalid TU" ); CodingStructure &saveCS = *m_pSaveCS[0]; TransformUnit *tmpTU = nullptr; Distortion singleDistTmpLuma = 0; uint64_t singleTmpFracBits = 0; double singleCostTmp = 0; int firstCheckId = (sps.getUseLFNST() && mtsCheckRangeFlag && cu.mtsFlag) ? mtsFirstCheckId : 0; //we add the MTS candidates to the loop. TransformSkip will still be the last one to be checked (when modeId == lastCheckId) as long as checkTransformSkip is true int lastCheckId = sps.getUseLFNST() ? ((mtsCheckRangeFlag && cu.mtsFlag) ? (mtsLastCheckId + (int) checkTransformSkip) : (numTransformIndexCands - (firstCheckId + 1) + (int) checkTransformSkip)) : trModes[nNumTransformCands - 1].first - MtsType::DCT2_DCT2; bool isNotOnlyOneMode = sps.getUseLFNST() ? lastCheckId != firstCheckId : nNumTransformCands != 1; if( isNotOnlyOneMode ) { saveCS.pcv = cs.pcv; saveCS.picture = cs.picture; saveCS.area.repositionTo(cs.area); saveCS.clearTUs(); tmpTU = &saveCS.addTU(currArea, partitioner.chType); } bool cbfBestMode = false; bool cbfBestModeValid = false; bool cbfDCT2 = true; for( int modeId = firstCheckId; modeId <= ( sps.getUseLFNST() ? lastCheckId : ( nNumTransformCands - 1 ) ); modeId++ ) { int transformIndex = modeId; if( sps.getUseLFNST() ) { if( ( transformIndex < lastCheckId ) || ( ( transformIndex == lastCheckId ) && !checkTransformSkip ) ) //we avoid this if the mode is transformSkip { // Skip checking other transform candidates if zero CBF is encountered and it is the best transform so far if( m_pcEncCfg->getUseFastLFNST() && transformIndex && !cbfBestMode && cbfBestModeValid ) { continue; } } } else { if (!(m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && slice.isLossless())) { if (!cbfDCT2 || (m_pcEncCfg->getUseTransformSkipFast() && MtsType::DCT2_DCT2 + bestModeIds[COMPONENT_Y] == MtsType::SKIP)) { break; } if (!trModes[modeId].second) { continue; } } tu.mtsIdx[COMPONENT_Y] = trModes[modeId].first; } if ((modeId != firstCheckId) && isNotOnlyOneMode) { m_CABACEstimator->getCtx() = ctxStart; } int default0Save1Load2 = 0; singleDistTmpLuma = 0; if( modeId == firstCheckId && ( sps.getUseLFNST() ? ( modeId != lastCheckId ) : ( nNumTransformCands > 1 ) ) ) { default0Save1Load2 = 1; } else if (modeId != firstCheckId) { if( sps.getUseLFNST() && !cbfBestModeValid ) { default0Save1Load2 = 1; } else { default0Save1Load2 = 2; } } if( sps.getUseLFNST() ) { if( cu.mtsFlag ) { if( moreProbMTSIdxFirst ) { const ChannelType chType = toChannelType( COMPONENT_Y ); const CompArea& area = tu.blocks[ COMPONENT_Y ]; const PredictionUnit& pu = *cs.getPU( area.pos(), chType ); uint32_t intraMode = pu.intraDir[chType]; if( transformIndex == 1 ) { tu.mtsIdx[COMPONENT_Y] = (intraMode < DIA_IDX) ? MtsType::DST7_DCT8 : MtsType::DCT8_DST7; } else if( transformIndex == 2 ) { tu.mtsIdx[COMPONENT_Y] = (intraMode < DIA_IDX) ? MtsType::DCT8_DST7 : MtsType::DST7_DCT8; } else { tu.mtsIdx[COMPONENT_Y] = MtsType::DST7_DST7 + transformIndex; } } else { tu.mtsIdx[COMPONENT_Y] = MtsType::DST7_DST7 + transformIndex; } } else { tu.mtsIdx[COMPONENT_Y] = MtsType::DCT2_DCT2 + transformIndex; } if( !cu.mtsFlag && checkTransformSkip ) { xIntraCodingTUBlock( tu, COMPONENT_Y, singleDistTmpLuma, default0Save1Load2, &numSig, modeId == 0 ? &trModes : nullptr, true ); if( modeId == 0 ) { for( int i = 0; i < 2; i++ ) { if( trModes[ i ].second ) { lastCheckId = trModes[i].first - MtsType::DCT2_DCT2; } } } } else { xIntraCodingTUBlock( tu, COMPONENT_Y, singleDistTmpLuma, default0Save1Load2, &numSig ); } } else { if( nNumTransformCands > 1 ) { xIntraCodingTUBlock( tu, COMPONENT_Y, singleDistTmpLuma, default0Save1Load2, &numSig, modeId == 0 ? &trModes : nullptr, true ); if( modeId == 0 ) { for( int i = 0; i < nNumTransformCands; i++ ) { if( trModes[ i ].second ) { lastCheckId = trModes[i].first - MtsType::DCT2_DCT2; } } } } else { xIntraCodingTUBlock( tu, COMPONENT_Y, singleDistTmpLuma, default0Save1Load2, &numSig ); } } cuCtx.mtsLastScanPos = false; cuCtx.violatesMtsCoeffConstraint = false; //----- determine rate and r-d cost ----- if ((sps.getUseLFNST() ? (modeId == lastCheckId && modeId != 0 && checkTransformSkip) : (trModes[modeId].first != MtsType::DCT2_DCT2)) && !TU::getCbfAtDepth(tu, COMPONENT_Y, currDepth)) { //In order not to code TS flag when cbf is zero, the case for TS with cbf being zero is forbidden. if (m_pcEncCfg->getCostMode() != COST_LOSSLESS_CODING || !slice.isLossless()) { singleCostTmp = MAX_DOUBLE; } else { singleTmpFracBits = xGetIntraFracBitsQT(*csFull, partitioner, true, false, -1, TU_NO_ISP, &cuCtx); singleCostTmp = m_pcRdCost->calcRdCost(singleTmpFracBits, singleDistTmpLuma); } } else { singleTmpFracBits = xGetIntraFracBitsQT(*csFull, partitioner, true, false, -1, TU_NO_ISP, &cuCtx); if (tu.mtsIdx[COMPONENT_Y] > MtsType::SKIP) { if (!cuCtx.mtsLastScanPos) { singleCostTmp = MAX_DOUBLE; } else { singleCostTmp = m_pcRdCost->calcRdCost(singleTmpFracBits, singleDistTmpLuma); } } else { singleCostTmp = m_pcRdCost->calcRdCost(singleTmpFracBits, singleDistTmpLuma); } } if (singleCostTmp < singleCost) { singleCost = singleCostTmp; singleDistLuma = singleDistTmpLuma; singleFracBits = singleTmpFracBits; if( sps.getUseLFNST() ) { bestModeIds[COMPONENT_Y] = modeId; cbfBestMode = TU::getCbfAtDepth( tu, COMPONENT_Y, currDepth ); cbfBestModeValid = true; validReturnFull = true; } else { bestModeIds[COMPONENT_Y] = trModes[modeId].first - MtsType::DCT2_DCT2; if (trModes[modeId].first == MtsType::DCT2_DCT2) { cbfDCT2 = TU::getCbfAtDepth( tu, COMPONENT_Y, currDepth ); } } if (bestModeIds[COMPONENT_Y] != lastCheckId) { saveCS.getPredBuf( tu.Y() ).copyFrom( csFull->getPredBuf( tu.Y() ) ); saveCS.getRecoBuf( tu.Y() ).copyFrom( csFull->getRecoBuf( tu.Y() ) ); if( KEEP_PRED_AND_RESI_SIGNALS ) { saveCS.getResiBuf ( tu.Y() ).copyFrom( csFull->getResiBuf ( tu.Y() ) ); saveCS.getOrgResiBuf( tu.Y() ).copyFrom( csFull->getOrgResiBuf( tu.Y() ) ); } tmpTU->copyComponentFrom( tu, COMPONENT_Y ); ctxBest = m_CABACEstimator->getCtx(); } } } if( sps.getUseLFNST() && !validReturnFull ) { csFull->cost = MAX_DOUBLE; if (checkSplit) { ctxBest = m_CABACEstimator->getCtx(); } } else { if (bestModeIds[COMPONENT_Y] != lastCheckId) { csFull->getPredBuf( tu.Y() ).copyFrom( saveCS.getPredBuf( tu.Y() ) ); csFull->getRecoBuf( tu.Y() ).copyFrom( saveCS.getRecoBuf( tu.Y() ) ); if( KEEP_PRED_AND_RESI_SIGNALS ) { csFull->getResiBuf ( tu.Y() ).copyFrom( saveCS.getResiBuf ( tu.Y() ) ); csFull->getOrgResiBuf( tu.Y() ).copyFrom( saveCS.getOrgResiBuf( tu.Y() ) ); } tu.copyComponentFrom( *tmpTU, COMPONENT_Y ); if (!checkSplit) { m_CABACEstimator->getCtx() = ctxBest; } } else if (checkSplit) { ctxBest = m_CABACEstimator->getCtx(); } csFull->cost += singleCost; csFull->dist += singleDistLuma; csFull->fracBits += singleFracBits; } } bool validReturnSplit = false; if (checkSplit) { //----- store full entropy coding status, load original entropy coding status ----- if (checkFull) { m_CABACEstimator->getCtx() = ctxStart; } //----- code splitted block ----- csSplit->cost = 0; bool splitCbfLuma = false; bool splitIsSelected = true; if( partitioner.canSplit( TU_MAX_TR_SPLIT, cs ) ) { partitioner.splitCurrArea( TU_MAX_TR_SPLIT, cs ); } do { bool tmpValidReturnSplit = xRecurIntraCodingLumaQT( *csSplit, partitioner, false, mtsCheckRangeFlag, mtsFirstCheckId, mtsLastCheckId ); if( sps.getUseLFNST() && !tmpValidReturnSplit ) { splitIsSelected = false; break; } csSplit->setDecomp(partitioner.currArea().Y()); splitCbfLuma |= TU::getCbfAtDepth(*csSplit->getTU(partitioner.currArea().lumaPos(), partitioner.chType, -1), COMPONENT_Y, partitioner.currTrDepth); } while( partitioner.nextPart( *csSplit ) ); partitioner.exitCurrSplit(); if( splitIsSelected ) { for( auto &ptu : csSplit->tus ) { if( currArea.Y().contains( ptu->Y() ) ) { TU::setCbfAtDepth(*ptu, COMPONENT_Y, currDepth, splitCbfLuma ? 1 : 0); } } //----- restore context states ----- m_CABACEstimator->getCtx() = ctxStart; cuCtx.violatesLfnstConstrained.fill(false); cuCtx.lfnstLastScanPos = false; cuCtx.violatesMtsCoeffConstraint = false; cuCtx.mtsLastScanPos = false; //----- determine rate and r-d cost ----- csSplit->fracBits = xGetIntraFracBitsQT( *csSplit, partitioner, true, false, -1, TU_NO_ISP, &cuCtx ); //--- update cost --- csSplit->cost = m_pcRdCost->calcRdCost(csSplit->fracBits, csSplit->dist); validReturnSplit = true; } } bool retVal = false; if( csFull || csSplit ) { if( !sps.getUseLFNST() || validReturnFull || validReturnSplit ) { // otherwise this would've happened in useSubStructure cs.picture->getRecoBuf(currArea.Y()).copyFrom(cs.getRecoBuf(currArea.Y())); cs.picture->getPredBuf(currArea.Y()).copyFrom(cs.getPredBuf(currArea.Y())); cs.cost = m_pcRdCost->calcRdCost(cs.fracBits, cs.dist); retVal = true; } } return retVal; } bool IntraSearch::xRecurIntraCodingACTQT(CodingStructure &cs, Partitioner &partitioner, bool mtsCheckRangeFlag, int mtsFirstCheckId, int mtsLastCheckId, bool moreProbMTSIdxFirst) { const UnitArea &currArea = partitioner.currArea(); uint32_t currDepth = partitioner.currTrDepth; const Slice &slice = *cs.slice; const SPS &sps = *cs.sps; bool checkFull = !partitioner.canSplit(TU_MAX_TR_SPLIT, cs); bool checkSplit = !checkFull; TempCtx ctxStart(m_ctxPool, m_CABACEstimator->getCtx()); TempCtx ctxBest(m_ctxPool); CodingStructure *csSplit = nullptr; CodingStructure *csFull = nullptr; if (checkSplit) { csSplit = &cs; } else if (checkFull) { csFull = &cs; } bool validReturnFull = false; if (checkFull) { TransformUnit &tu = csFull->addTU(CS::getArea(*csFull, currArea, partitioner.chType), partitioner.chType); tu.depth = currDepth; const CodingUnit &cu = *csFull->getCU(tu.Y().pos(), ChannelType::LUMA); const PredictionUnit &pu = *csFull->getPU(tu.Y().pos(), ChannelType::LUMA); CHECK(!tu.Y().valid() || !tu.Cb().valid() || !tu.Cr().valid(), "Invalid TU"); CHECK(tu.cu != &cu, "wrong CU fetch"); CHECK(cu.ispMode != ISPType::NONE, "adaptive color transform cannot be applied to ISP"); CHECK(pu.intraDir[ChannelType::CHROMA] != DM_CHROMA_IDX, "chroma should use DM mode for adaptive color transform"); // 1. intra prediction and forward color transform PelUnitBuf orgBuf = csFull->getOrgBuf(tu); PelUnitBuf predBuf = csFull->getPredBuf(tu); PelUnitBuf resiBuf = csFull->getResiBuf(tu); PelUnitBuf orgResiBuf = csFull->getOrgResiBuf(tu); bool doReshaping = (slice.getLmcsEnabledFlag() && slice.getPicHeader()->getLmcsChromaResidualScaleFlag() && (slice.isIntra() || m_pcReshape->getCTUFlag()) && (tu.blocks[COMPONENT_Cb].width * tu.blocks[COMPONENT_Cb].height > 4)); if (doReshaping) { const Area area = tu.Y().valid() ? tu.Y() : Area(recalcPosition(tu.chromaFormat, tu.chType, ChannelType::LUMA, tu.block(tu.chType).pos()), recalcSize(tu.chromaFormat, tu.chType, ChannelType::LUMA, tu.block(tu.chType).size())); const CompArea &areaY = CompArea(COMPONENT_Y, tu.chromaFormat, area); int adj = m_pcReshape->calculateChromaAdjVpduNei(tu, areaY); tu.setChromaAdj(adj); } for (int i = 0; i < getNumberValidComponents(tu.chromaFormat); i++) { ComponentID compID = (ComponentID)i; const CompArea &area = tu.blocks[compID]; const ChannelType chType = toChannelType(compID); PelBuf piOrg = orgBuf.bufs[compID]; PelBuf piPred = predBuf.bufs[compID]; PelBuf piResi = resiBuf.bufs[compID]; initIntraPatternChType(*tu.cu, area); if (PU::isMIP(pu, chType)) { initIntraMip(pu, area); predIntraMip(compID, piPred, pu); } else { predIntraAng(compID, piPred, pu); } piResi.copyFrom(piOrg); if (slice.getLmcsEnabledFlag() && m_pcReshape->getCTUFlag() && compID == COMPONENT_Y) { CompArea tmpArea(COMPONENT_Y, area.chromaFormat, Position(0, 0), area.size()); PelBuf tmpPred = m_tmpStorageCtu.getBuf(tmpArea); tmpPred.copyFrom(piPred); piResi.rspSignal(m_pcReshape->getFwdLUT()); piResi.subtract(tmpPred); } else if (doReshaping && (compID != COMPONENT_Y)) { piResi.subtract(piPred); int cResScaleInv = tu.getChromaAdj(); piResi.scaleSignal(cResScaleInv, 1, slice.clpRng(compID)); } else { piResi.subtract(piPred); } } resiBuf.colorSpaceConvert(orgResiBuf, true, cs.slice->clpRng(COMPONENT_Y)); // 2. luma residual optimization double singleCostLuma = MAX_DOUBLE; bool checkTransformSkip = sps.getTransformSkipEnabledFlag(); int bestLumaModeId = 0; uint8_t nNumTransformCands = cu.mtsFlag ? 4 : 1; uint8_t numTransformIndexCands = nNumTransformCands; const bool tsAllowed = TU::isTSAllowed(tu, COMPONENT_Y); const bool mtsAllowed = CU::isMTSAllowed(cu, COMPONENT_Y); const bool lossless = m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && slice.isLossless(); TrModeList trModes; if (sps.getUseLFNST()) { checkTransformSkip &= tsAllowed; checkTransformSkip &= !cu.mtsFlag; checkTransformSkip &= !cu.lfnstIdx; if (!cu.mtsFlag && checkTransformSkip) { trModes.push_back(TrMode(MtsType::DCT2_DCT2, true)); // DCT2 trModes.push_back(TrMode(MtsType::SKIP, true)); // TS } } else { if (lossless) { nNumTransformCands = 1; CHECK(!tsAllowed && cu.bdpcmMode == BdpcmMode::NONE, "transform skip should be enabled for LS"); if (cu.bdpcmMode != BdpcmMode::NONE) { trModes.push_back(TrMode(MtsType::DCT2_DCT2, true)); } else { trModes.push_back(TrMode(MtsType::SKIP, true)); } } else { nNumTransformCands = 1 + (tsAllowed ? 1 : 0) + (mtsAllowed ? 4 : 0); // DCT + TS + 4 MTS = 6 tests trModes.push_back(TrMode(MtsType::DCT2_DCT2, true)); // DCT2 if (tsAllowed) { trModes.push_back(TrMode(MtsType::SKIP, true)); } if (mtsAllowed) { for (int i = 2; i < 6; i++) { trModes.push_back(TrMode(MtsType(i), true)); } } } } CodingStructure &saveLumaCS = *m_pSaveCS[0]; TransformUnit *tmpTU = nullptr; Distortion singleDistTmpLuma = 0; uint64_t singleTmpFracBits = 0; double singleCostTmp = 0; int firstCheckId = (sps.getUseLFNST() && mtsCheckRangeFlag && cu.mtsFlag) ? mtsFirstCheckId : 0; int lastCheckId = sps.getUseLFNST() ? ((mtsCheckRangeFlag && cu.mtsFlag) ? (mtsLastCheckId + (int) checkTransformSkip) : (numTransformIndexCands - (firstCheckId + 1) + (int) checkTransformSkip)) : trModes[nNumTransformCands - 1].first - MtsType::DCT2_DCT2; bool isNotOnlyOneMode = sps.getUseLFNST() ? lastCheckId != firstCheckId : nNumTransformCands != 1; if (isNotOnlyOneMode) { saveLumaCS.pcv = csFull->pcv; saveLumaCS.picture = csFull->picture; saveLumaCS.area.repositionTo(csFull->area); saveLumaCS.clearTUs(); tmpTU = &saveLumaCS.addTU(currArea, partitioner.chType); } bool cbfBestMode = false; bool cbfBestModeValid = false; bool cbfDCT2 = true; if (m_pcEncCfg->getCostMode() != COST_LOSSLESS_CODING || !slice.isLossless()) { m_pcRdCost->lambdaAdjustColorTrans(true, COMPONENT_Y); } for (int modeIndex = firstCheckId; sps.getUseLFNST() || modeIndex < trModes.size(); modeIndex++) { const int modeId = sps.getUseLFNST() ? modeIndex : trModes[modeIndex].first - MtsType::DCT2_DCT2; if (modeId > lastCheckId) { break; } uint8_t transformIndex = modeId; csFull->getResiBuf(tu.Y()).copyFrom(csFull->getOrgResiBuf(tu.Y())); m_CABACEstimator->getCtx() = ctxStart; m_CABACEstimator->resetBits(); if (sps.getUseLFNST()) { if ((transformIndex < lastCheckId) || ((transformIndex == lastCheckId) && !checkTransformSkip)) //we avoid this if the mode is transformSkip { // Skip checking other transform candidates if zero CBF is encountered and it is the best transform so far if (m_pcEncCfg->getUseFastLFNST() && transformIndex && !cbfBestMode && cbfBestModeValid) { continue; } } } else { if (!lossless) { if (!cbfDCT2 || (m_pcEncCfg->getUseTransformSkipFast() && bestLumaModeId == 1)) { break; } if (!trModes[modeIndex].second) { continue; } } tu.mtsIdx[COMPONENT_Y] = MtsType::DCT2_DCT2 + modeId; } singleDistTmpLuma = 0; if (sps.getUseLFNST()) { if (cu.mtsFlag) { if (moreProbMTSIdxFirst) { const uint32_t intraMode = pu.intraDir[ChannelType::LUMA]; if (transformIndex == 1) { tu.mtsIdx[COMPONENT_Y] = (intraMode < DIA_IDX) ? MtsType::DST7_DCT8 : MtsType::DCT8_DST7; } else if (transformIndex == 2) { tu.mtsIdx[COMPONENT_Y] = (intraMode < DIA_IDX) ? MtsType::DCT8_DST7 : MtsType::DST7_DCT8; } else { tu.mtsIdx[COMPONENT_Y] = MtsType::DST7_DST7 + transformIndex; } } else { tu.mtsIdx[COMPONENT_Y] = MtsType::DST7_DST7 + transformIndex; } } else { tu.mtsIdx[COMPONENT_Y] = MtsType::DCT2_DCT2 + transformIndex; } if (!cu.mtsFlag && checkTransformSkip) { xIntraCodingACTTUBlock(tu, COMPONENT_Y, singleDistTmpLuma, modeId == 0 ? &trModes : nullptr, true); if (modeId == 0) { for (int i = 0; i < 2; i++) { if (trModes[i].second) { lastCheckId = trModes[i].first - MtsType::DCT2_DCT2; } } } } else { xIntraCodingACTTUBlock(tu, COMPONENT_Y, singleDistTmpLuma); } } else { if (nNumTransformCands > 1) { xIntraCodingACTTUBlock(tu, COMPONENT_Y, singleDistTmpLuma, modeId == 0 ? &trModes : nullptr, true); if (modeId == 0) { for (int i = 0; i < nNumTransformCands; i++) { if (trModes[i].second) { lastCheckId = trModes[i].first - MtsType::DCT2_DCT2; } } } } else { xIntraCodingACTTUBlock(tu, COMPONENT_Y, singleDistTmpLuma); } } CUCtx cuCtx; cuCtx.isDQPCoded = true; cuCtx.isChromaQpAdjCoded = true; //----- determine rate and r-d cost ----- if ((sps.getUseLFNST() ? (modeId == lastCheckId && modeId != 0 && checkTransformSkip) : (modeId != 0)) && !TU::getCbfAtDepth(tu, COMPONENT_Y, currDepth)) { //In order not to code TS flag when cbf is zero, the case for TS with cbf being zero is forbidden. if (m_pcEncCfg->getCostMode() != COST_LOSSLESS_CODING || !slice.isLossless()) singleCostTmp = MAX_DOUBLE; else { singleTmpFracBits = xGetIntraFracBitsQT(*csFull, partitioner, true, false, -1, TU_NO_ISP); singleCostTmp = m_pcRdCost->calcRdCost(singleTmpFracBits, singleDistTmpLuma, false); } } else { singleTmpFracBits = xGetIntraFracBitsQT(*csFull, partitioner, true, false, -1, TU_NO_ISP, &cuCtx); if (tu.mtsIdx[COMPONENT_Y] > MtsType::SKIP) { if (!cuCtx.mtsLastScanPos) { singleCostTmp = MAX_DOUBLE; } else { singleCostTmp = m_pcRdCost->calcRdCost(singleTmpFracBits, singleDistTmpLuma, false); } } else { singleCostTmp = m_pcRdCost->calcRdCost(singleTmpFracBits, singleDistTmpLuma, false); } } if (singleCostTmp < singleCostLuma) { singleCostLuma = singleCostTmp; validReturnFull = true; if (sps.getUseLFNST()) { bestLumaModeId = modeId; cbfBestMode = TU::getCbfAtDepth(tu, COMPONENT_Y, currDepth); cbfBestModeValid = true; } else { bestLumaModeId = modeId; if (modeId == 0) { cbfDCT2 = TU::getCbfAtDepth(tu, COMPONENT_Y, currDepth); } } if (bestLumaModeId != lastCheckId) { saveLumaCS.getResiBuf(tu.Y()).copyFrom(csFull->getResiBuf(tu.Y())); tmpTU->copyComponentFrom(tu, COMPONENT_Y); ctxBest = m_CABACEstimator->getCtx(); } } } if (m_pcEncCfg->getCostMode() != COST_LOSSLESS_CODING || !slice.isLossless()) { m_pcRdCost->lambdaAdjustColorTrans(false, COMPONENT_Y); } if (sps.getUseLFNST()) { if (!validReturnFull) { csFull->cost = MAX_DOUBLE; return false; } } else { CHECK(!validReturnFull, "no transform mode was tested for luma"); } csFull->setDecomp(currArea.Y(), true); csFull->setDecomp(currArea.Cb(), true); if (bestLumaModeId != lastCheckId) { csFull->getResiBuf(tu.Y()).copyFrom(saveLumaCS.getResiBuf(tu.Y())); tu.copyComponentFrom(*tmpTU, COMPONENT_Y); m_CABACEstimator->getCtx() = ctxBest; } // 3 chroma residual optimization CodingStructure &saveChromaCS = *m_pSaveCS[1]; saveChromaCS.pcv = csFull->pcv; saveChromaCS.picture = csFull->picture; saveChromaCS.area.repositionTo(csFull->area); saveChromaCS.initStructData(MAX_INT, true); tmpTU = &saveChromaCS.addTU(currArea, partitioner.chType); CompArea& cbArea = tu.blocks[COMPONENT_Cb]; CompArea& crArea = tu.blocks[COMPONENT_Cr]; tu.jointCbCr = 0; CompStorage orgResiCb[5], orgResiCr[5]; // 0:std, 1-3:jointCbCr (placeholder at this stage), 4:crossComp orgResiCb[0].create(cbArea); orgResiCr[0].create(crArea); orgResiCb[0].copyFrom(csFull->getOrgResiBuf(cbArea)); orgResiCr[0].copyFrom(csFull->getOrgResiBuf(crArea)); // 3.1 regular chroma residual coding csFull->getResiBuf(cbArea).copyFrom(orgResiCb[0]); csFull->getResiBuf(crArea).copyFrom(orgResiCr[0]); for (uint32_t c = COMPONENT_Cb; c < ::getNumberValidTBlocks(*csFull->pcv); c++) { const ComponentID compID = ComponentID(c); double singleBestCostChroma = MAX_DOUBLE; int bestModeId = -1; bool tsAllowed = TU::isTSAllowed(tu, compID) && (m_pcEncCfg->getUseChromaTS()) && !cu.lfnstIdx; uint8_t numTransformCands = 1 + (tsAllowed ? 1 : 0); // DCT + TS = 2 tests bool cbfDCT2 = true; trModes.clear(); if (lossless) { numTransformCands = 1; CHECK(!tsAllowed && cu.bdpcmModeChroma == BdpcmMode::NONE, "transform skip should be enabled for LS"); if (cu.bdpcmModeChroma != BdpcmMode::NONE) { trModes.push_back(TrMode(MtsType::DCT2_DCT2, true)); } else { trModes.push_back(TrMode(MtsType::SKIP, true)); } } else { trModes.push_back(TrMode(MtsType::DCT2_DCT2, true)); // DCT if (tsAllowed) { trModes.push_back(TrMode(MtsType::SKIP, true)); // TS } } if (m_pcEncCfg->getCostMode() != COST_LOSSLESS_CODING || !slice.isLossless()) { if (doReshaping) { int cResScaleInv = tu.getChromaAdj(); m_pcRdCost->lambdaAdjustColorTrans(true, compID, true, &cResScaleInv); } else { m_pcRdCost->lambdaAdjustColorTrans(true, compID); } } TempCtx ctxBegin(m_ctxPool); ctxBegin = m_CABACEstimator->getCtx(); for (int modeId = 0; modeId < numTransformCands; modeId++) { if (m_pcEncCfg->getCostMode() != COST_LOSSLESS_CODING || !slice.isLossless()) { if (modeId && !cbfDCT2) { continue; } if (!trModes[modeId].second) { continue; } } if (modeId > 0) { m_CABACEstimator->getCtx() = ctxBegin; } tu.mtsIdx[compID] = trModes[modeId].first; Distortion singleDistChroma = 0; if (numTransformCands > 1) { xIntraCodingACTTUBlock(tu, compID, singleDistChroma, modeId == 0 ? &trModes : nullptr, true); } else { xIntraCodingACTTUBlock(tu, compID, singleDistChroma); } if (tu.mtsIdx[compID] == MtsType::DCT2_DCT2) { cbfDCT2 = TU::getCbfAtDepth(tu, compID, currDepth); } uint64_t fracBitChroma = xGetIntraFracBitsQTChroma(tu, compID); double singleCostChroma = m_pcRdCost->calcRdCost(fracBitChroma, singleDistChroma, false); if (singleCostChroma < singleBestCostChroma) { singleBestCostChroma = singleCostChroma; bestModeId = modeId; if (bestModeId != (numTransformCands - 1)) { saveChromaCS.getResiBuf(tu.blocks[compID]).copyFrom(csFull->getResiBuf(tu.blocks[compID])); tmpTU->copyComponentFrom(tu, compID); ctxBest = m_CABACEstimator->getCtx(); } } } if (bestModeId != (numTransformCands - 1)) { csFull->getResiBuf(tu.blocks[compID]).copyFrom(saveChromaCS.getResiBuf(tu.blocks[compID])); tu.copyComponentFrom(*tmpTU, compID); m_CABACEstimator->getCtx() = ctxBest; } if (m_pcEncCfg->getCostMode() != COST_LOSSLESS_CODING || !slice.isLossless()) { m_pcRdCost->lambdaAdjustColorTrans(false, compID); } } Position tuPos = tu.Y(); tuPos.relativeTo(cu.Y()); const UnitArea relativeUnitArea(tu.chromaFormat, Area(tuPos, tu.Y().size())); PelUnitBuf invColorTransResidual = m_colorTransResiBuf.getBuf(relativeUnitArea); csFull->getResiBuf(tu).colorSpaceConvert(invColorTransResidual, false, cs.slice->clpRng(COMPONENT_Y)); Distortion totalDist = 0; for (uint32_t c = COMPONENT_Y; c < ::getNumberValidTBlocks(*csFull->pcv); c++) { const ComponentID compID = ComponentID(c); const CompArea& area = tu.blocks[compID]; PelBuf piOrg = csFull->getOrgBuf(area); PelBuf piReco = csFull->getRecoBuf(area); PelBuf piPred = csFull->getPredBuf(area); PelBuf piResi = invColorTransResidual.bufs[compID]; if (doReshaping && (compID != COMPONENT_Y)) { piResi.scaleSignal(tu.getChromaAdj(), 0, slice.clpRng(compID)); } piReco.reconstruct(piPred, piResi, cs.slice->clpRng(compID)); if (m_pcEncCfg->getLumaLevelToDeltaQPMapping().isEnabled() || (m_pcEncCfg->getLmcs() && slice.getLmcsEnabledFlag() && (m_pcReshape->getCTUFlag() || (isChroma(compID) && m_pcEncCfg->getReshapeIntraCMD())))) { const CPelBuf orgLuma = csFull->getOrgBuf(csFull->area.blocks[COMPONENT_Y]); if (compID == COMPONENT_Y && !(m_pcEncCfg->getLumaLevelToDeltaQPMapping().isEnabled())) { CompArea tmpArea1(COMPONENT_Y, area.chromaFormat, Position(0, 0), area.size()); PelBuf tmpRecLuma = m_tmpStorageCtu.getBuf(tmpArea1); tmpRecLuma.copyFrom(piReco); tmpRecLuma.rspSignal(m_pcReshape->getInvLUT()); totalDist += m_pcRdCost->getDistPart(piOrg, tmpRecLuma, sps.getBitDepth(toChannelType(compID)), compID, DFunc::SSE_WTD, &orgLuma); } else { totalDist += m_pcRdCost->getDistPart(piOrg, piReco, sps.getBitDepth(toChannelType(compID)), compID, DFunc::SSE_WTD, &orgLuma); } } else { totalDist += m_pcRdCost->getDistPart(piOrg, piReco, sps.getBitDepth(toChannelType(compID)), compID, DFunc::SSE); } } m_CABACEstimator->getCtx() = ctxStart; uint64_t totalBits = xGetIntraFracBitsQT(*csFull, partitioner, true, true, -1, TU_NO_ISP); double totalCost = m_pcRdCost->calcRdCost(totalBits, totalDist); saveChromaCS.getResiBuf(cbArea).copyFrom(csFull->getResiBuf(cbArea)); saveChromaCS.getResiBuf(crArea).copyFrom(csFull->getResiBuf(crArea)); saveChromaCS.getRecoBuf(tu).copyFrom(csFull->getRecoBuf(tu)); tmpTU->copyComponentFrom(tu, COMPONENT_Cb); tmpTU->copyComponentFrom(tu, COMPONENT_Cr); ctxBest = m_CABACEstimator->getCtx(); // 3.2 jointCbCr double bestCostJointCbCr = totalCost; Distortion bestDistJointCbCr = totalDist; uint64_t bestBitsJointCbCr = totalBits; int bestJointCbCr = tu.jointCbCr; assert(!bestJointCbCr); bool lastIsBest = false; CbfMaskList jointCbfMasksToTest; if (sps.getJointCbCrEnabledFlag() && (TU::getCbf(tu, COMPONENT_Cb) || TU::getCbf(tu, COMPONENT_Cr))) { m_pcTrQuant->selectICTCandidates(tu, orgResiCb, orgResiCr, jointCbfMasksToTest); } for (int cbfMask : jointCbfMasksToTest) { tu.jointCbCr = (uint8_t)cbfMask; ComponentID codeCompId = (cbfMask & CBF_MASK_CB) != 0 ? COMPONENT_Cb : COMPONENT_Cr; ComponentID otherCompId = codeCompId == COMPONENT_Cb ? COMPONENT_Cr : COMPONENT_Cb; bool tsAllowed = TU::isTSAllowed(tu, codeCompId) && (m_pcEncCfg->getUseChromaTS()) && !cu.lfnstIdx; uint8_t numTransformCands = 1 + (tsAllowed ? 1 : 0); // DCT + TS = 2 tests bool cbfDCT2 = true; trModes.clear(); trModes.push_back(TrMode(MtsType::DCT2_DCT2, true)); // DCT2 if (tsAllowed) { trModes.push_back(TrMode(MtsType::SKIP, true)); // TS } for (int modeId = 0; modeId < numTransformCands; modeId++) { if (modeId && !cbfDCT2) { continue; } if (!trModes[modeId].second) { continue; } Distortion distTmp = 0; tu.mtsIdx[codeCompId] = trModes[modeId].first; tu.mtsIdx[otherCompId] = MtsType::DCT2_DCT2; m_CABACEstimator->getCtx() = ctxStart; csFull->getResiBuf(cbArea).copyFrom(orgResiCb[cbfMask]); csFull->getResiBuf(crArea).copyFrom(orgResiCr[cbfMask]); if (nNumTransformCands > 1) { xIntraCodingACTTUBlock(tu, COMPONENT_Cb, distTmp, modeId == 0 ? &trModes : nullptr, true); } else { xIntraCodingACTTUBlock(tu, COMPONENT_Cb, distTmp); } double costTmp = std::numeric_limits::max(); uint64_t bitsTmp = 0; if (distTmp < std::numeric_limits::max()) { if (tu.mtsIdx[codeCompId] == MtsType::DCT2_DCT2) { cbfDCT2 = true; } csFull->getResiBuf(tu).colorSpaceConvert(invColorTransResidual, false, csFull->slice->clpRng(COMPONENT_Y)); distTmp = 0; for (uint32_t c = COMPONENT_Y; c < ::getNumberValidTBlocks(*csFull->pcv); c++) { const ComponentID compID = ComponentID(c); const CompArea & area = tu.blocks[compID]; PelBuf piOrg = csFull->getOrgBuf(area); PelBuf piReco = csFull->getRecoBuf(area); PelBuf piPred = csFull->getPredBuf(area); PelBuf piResi = invColorTransResidual.bufs[compID]; if (doReshaping && (compID != COMPONENT_Y)) { piResi.scaleSignal(tu.getChromaAdj(), 0, slice.clpRng(compID)); } piReco.reconstruct(piPred, piResi, cs.slice->clpRng(compID)); if (m_pcEncCfg->getLumaLevelToDeltaQPMapping().isEnabled() || (m_pcEncCfg->getLmcs() && slice.getLmcsEnabledFlag() && (m_pcReshape->getCTUFlag() || (isChroma(compID) && m_pcEncCfg->getReshapeIntraCMD())))) { const CPelBuf orgLuma = csFull->getOrgBuf(csFull->area.blocks[COMPONENT_Y]); if (compID == COMPONENT_Y && !(m_pcEncCfg->getLumaLevelToDeltaQPMapping().isEnabled())) { CompArea tmpArea1(COMPONENT_Y, area.chromaFormat, Position(0, 0), area.size()); PelBuf tmpRecLuma = m_tmpStorageCtu.getBuf(tmpArea1); tmpRecLuma.copyFrom(piReco); tmpRecLuma.rspSignal(m_pcReshape->getInvLUT()); distTmp += m_pcRdCost->getDistPart(piOrg, tmpRecLuma, sps.getBitDepth(toChannelType(compID)), compID, DFunc::SSE_WTD, &orgLuma); } else { distTmp += m_pcRdCost->getDistPart(piOrg, piReco, sps.getBitDepth(toChannelType(compID)), compID, DFunc::SSE_WTD, &orgLuma); } } else { distTmp += m_pcRdCost->getDistPart(piOrg, piReco, sps.getBitDepth(toChannelType(compID)), compID, DFunc::SSE); } } bitsTmp = xGetIntraFracBitsQT(*csFull, partitioner, true, true, -1, TU_NO_ISP); costTmp = m_pcRdCost->calcRdCost(bitsTmp, distTmp); } else if (tu.mtsIdx[codeCompId] == MtsType::DCT2_DCT2) { cbfDCT2 = false; } if (costTmp < bestCostJointCbCr) { bestCostJointCbCr = costTmp; bestDistJointCbCr = distTmp; bestBitsJointCbCr = bitsTmp; bestJointCbCr = tu.jointCbCr; lastIsBest = (cbfMask == jointCbfMasksToTest.back() && modeId == (numTransformCands - 1)); // store data if (!lastIsBest) { saveChromaCS.getResiBuf(cbArea).copyFrom(csFull->getResiBuf(cbArea)); saveChromaCS.getResiBuf(crArea).copyFrom(csFull->getResiBuf(crArea)); saveChromaCS.getRecoBuf(tu).copyFrom(csFull->getRecoBuf(tu)); tmpTU->copyComponentFrom(tu, COMPONENT_Cb); tmpTU->copyComponentFrom(tu, COMPONENT_Cr); ctxBest = m_CABACEstimator->getCtx(); } } } } if (!lastIsBest) { csFull->getResiBuf(cbArea).copyFrom(saveChromaCS.getResiBuf(cbArea)); csFull->getResiBuf(crArea).copyFrom(saveChromaCS.getResiBuf(crArea)); csFull->getRecoBuf(tu).copyFrom(saveChromaCS.getRecoBuf(tu)); tu.copyComponentFrom(*tmpTU, COMPONENT_Cb); tu.copyComponentFrom(*tmpTU, COMPONENT_Cr); m_CABACEstimator->getCtx() = ctxBest; } tu.jointCbCr = bestJointCbCr; csFull->picture->getRecoBuf(tu).copyFrom(csFull->getRecoBuf(tu)); csFull->dist += bestDistJointCbCr; csFull->fracBits += bestBitsJointCbCr; csFull->cost = m_pcRdCost->calcRdCost(csFull->fracBits, csFull->dist); } bool validReturnSplit = false; if (checkSplit) { if (partitioner.canSplit(TU_MAX_TR_SPLIT, *csSplit)) { partitioner.splitCurrArea(TU_MAX_TR_SPLIT, *csSplit); } bool splitIsSelected = true; do { bool tmpValidReturnSplit = xRecurIntraCodingACTQT(*csSplit, partitioner, mtsCheckRangeFlag, mtsFirstCheckId, mtsLastCheckId, moreProbMTSIdxFirst); if (sps.getUseLFNST()) { if (!tmpValidReturnSplit) { splitIsSelected = false; break; } } else { CHECK(!tmpValidReturnSplit, "invalid RD of sub-TU partitions for ACT"); } } while (partitioner.nextPart(*csSplit)); partitioner.exitCurrSplit(); if (splitIsSelected) { unsigned compCbf[3] = { 0, 0, 0 }; for (auto &currTU : csSplit->traverseTUs(currArea, partitioner.chType)) { for (unsigned ch = 0; ch < getNumberValidTBlocks(*csSplit->pcv); ch++) { compCbf[ch] |= (TU::getCbfAtDepth(currTU, ComponentID(ch), currDepth + 1) ? 1 : 0); } } for (auto &currTU : csSplit->traverseTUs(currArea, partitioner.chType)) { TU::setCbfAtDepth(currTU, COMPONENT_Y, currDepth, compCbf[COMPONENT_Y]); TU::setCbfAtDepth(currTU, COMPONENT_Cb, currDepth, compCbf[COMPONENT_Cb]); TU::setCbfAtDepth(currTU, COMPONENT_Cr, currDepth, compCbf[COMPONENT_Cr]); } m_CABACEstimator->getCtx() = ctxStart; csSplit->fracBits = xGetIntraFracBitsQT(*csSplit, partitioner, true, true, -1, TU_NO_ISP); csSplit->cost = m_pcRdCost->calcRdCost(csSplit->fracBits, csSplit->dist); validReturnSplit = true; } } bool retVal = false; if (csFull || csSplit) { if (sps.getUseLFNST()) { if (validReturnFull || validReturnSplit) { retVal = true; } } else { CHECK(!validReturnFull && !validReturnSplit, "illegal TU optimization"); retVal = true; } } return retVal; } ChromaCbfs IntraSearch::xRecurIntraChromaCodingQT( CodingStructure &cs, Partitioner& partitioner, const double bestCostSoFar, const PartSplit ispType ) { UnitArea currArea = partitioner.currArea(); const bool keepResi = cs.sps->getUseLMChroma() || KEEP_PRED_AND_RESI_SIGNALS; if (!currArea.Cb().valid()) { return ChromaCbfs(false); } const Slice &slice = *cs.slice; TransformUnit &currTU = *cs.getTU(currArea.chromaPos(), ChannelType::CHROMA); const PredictionUnit &pu = *cs.getPU(currArea.chromaPos(), ChannelType::CHROMA); bool lumaUsesISP = false; uint32_t currDepth = partitioner.currTrDepth; ChromaCbfs cbfs(false); if (currDepth == currTU.depth) { if (!currArea.Cb().valid() || !currArea.Cr().valid()) { return cbfs; } CodingStructure &saveCS = *m_pSaveCS[1]; saveCS.pcv = cs.pcv; saveCS.picture = cs.picture; saveCS.area.repositionTo( cs.area ); saveCS.initStructData( MAX_INT, true ); if (!currTU.cu->isSepTree() && currTU.cu->ispMode != ISPType::NONE) { saveCS.clearCUs(); CodingUnit& auxCU = saveCS.addCU( *currTU.cu, partitioner.chType ); auxCU.ispMode = currTU.cu->ispMode; saveCS.sps = currTU.cs->sps; saveCS.clearPUs(); saveCS.addPU( *currTU.cu->firstPU, partitioner.chType ); } TransformUnit &tmpTU = saveCS.addTU(currArea, partitioner.chType); cs.setDecomp(currArea.Cb(), true); // set in advance (required for Cb2/Cr2 in 4:2:2 video) const unsigned numTBlocks = ::getNumberValidTBlocks( *cs.pcv ); CompArea& cbArea = currTU.blocks[COMPONENT_Cb]; CompArea& crArea = currTU.blocks[COMPONENT_Cr]; double bestCostCb = MAX_DOUBLE; double bestCostCr = MAX_DOUBLE; Distortion bestDistCb = 0; Distortion bestDistCr = 0; int maxModesTested = 0; bool earlyExitISP = false; TempCtx ctxStartTU(m_ctxPool); TempCtx ctxStart(m_ctxPool); TempCtx ctxBest(m_ctxPool); ctxStartTU = m_CABACEstimator->getCtx(); currTU.jointCbCr = 0; // Do predictions here to avoid repeating the "default0Save1Load2" stuff int predMode = pu.cu->bdpcmModeChroma != BdpcmMode::NONE ? BDPCM_IDX : PU::getFinalIntraMode(pu, ChannelType::CHROMA); PelBuf piPredCb = cs.getPredBuf(cbArea); PelBuf piPredCr = cs.getPredBuf(crArea); initIntraPatternChType( *currTU.cu, cbArea); initIntraPatternChType( *currTU.cu, crArea); if( PU::isLMCMode( predMode ) ) { xGetLumaRecPixels( pu, cbArea ); predIntraChromaLM( COMPONENT_Cb, piPredCb, pu, cbArea, predMode ); predIntraChromaLM( COMPONENT_Cr, piPredCr, pu, crArea, predMode ); } else if (PU::isMIP(pu, ChannelType::CHROMA)) { initIntraMip(pu, cbArea); predIntraMip(COMPONENT_Cb, piPredCb, pu); initIntraMip(pu, crArea); predIntraMip(COMPONENT_Cr, piPredCr, pu); } else { predIntraAng( COMPONENT_Cb, piPredCb, pu); predIntraAng( COMPONENT_Cr, piPredCr, pu); } // determination of chroma residuals including reshaping and cross-component prediction //----- get chroma residuals ----- PelBuf resiCb = cs.getResiBuf(cbArea); PelBuf resiCr = cs.getResiBuf(crArea); resiCb.copyFrom( cs.getOrgBuf (cbArea) ); resiCr.copyFrom( cs.getOrgBuf (crArea) ); resiCb.subtract( piPredCb ); resiCr.subtract( piPredCr ); //----- get reshape parameter ---- bool doReshaping = ( cs.slice->getLmcsEnabledFlag() && cs.picHeader->getLmcsChromaResidualScaleFlag() && (cs.slice->isIntra() || m_pcReshape->getCTUFlag()) && (cbArea.width * cbArea.height > 4) ); if( doReshaping ) { const Area area = currTU.Y().valid() ? currTU.Y() : Area( recalcPosition(currTU.chromaFormat, currTU.chType, ChannelType::LUMA, currTU.block(currTU.chType).pos()), recalcSize(currTU.chromaFormat, currTU.chType, ChannelType::LUMA, currTU.block(currTU.chType).size())); const CompArea &areaY = CompArea(COMPONENT_Y, currTU.chromaFormat, area); int adj = m_pcReshape->calculateChromaAdjVpduNei(currTU, areaY); currTU.setChromaAdj(adj); } //----- get cross component prediction parameters ----- //===== store original residual signals ===== CompStorage orgResiCb[4], orgResiCr[4]; // 0:std, 1-3:jointCbCr (placeholder at this stage) orgResiCb[0].create( cbArea ); orgResiCr[0].create( crArea ); orgResiCb[0].copyFrom( resiCb ); orgResiCr[0].copyFrom( resiCr ); if( doReshaping ) { int cResScaleInv = currTU.getChromaAdj(); orgResiCb[0].scaleSignal( cResScaleInv, 1, currTU.cu->cs->slice->clpRng(COMPONENT_Cb) ); orgResiCr[0].scaleSignal( cResScaleInv, 1, currTU.cu->cs->slice->clpRng(COMPONENT_Cr) ); } for( uint32_t c = COMPONENT_Cb; c < numTBlocks; c++) { const ComponentID compID = ComponentID(c); const CompArea& area = currTU.blocks[compID]; double singleCost = MAX_DOUBLE; int bestModeId = 0; Distortion singleDistCTmp = 0; double singleCostTmp = 0; const bool tsAllowed = TU::isTSAllowed(currTU, compID) && m_pcEncCfg->getUseChromaTS() && !currTU.cu->lfnstIdx; uint8_t nNumTransformCands = 1 + (tsAllowed ? 1 : 0); // DCT + TS = 2 tests TrModeList trModes; if (m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && slice.isLossless()) { nNumTransformCands = 1; CHECK(!tsAllowed && currTU.cu->bdpcmModeChroma == BdpcmMode::NONE, "transform skip should be enabled for LS"); if (currTU.cu->bdpcmModeChroma != BdpcmMode::NONE) { trModes.push_back(TrMode(MtsType::DCT2_DCT2, true)); } else { trModes.push_back(TrMode(MtsType::SKIP, true)); } } else { trModes.push_back(TrMode(MtsType::DCT2_DCT2, true)); // DCT2 if (tsAllowed) { trModes.push_back(TrMode(MtsType::SKIP, true)); // TS } } CHECK(!currTU.Cb().valid(), "Invalid TU"); const int totalModesToTest = nNumTransformCands; bool cbfDCT2 = true; const bool isOneMode = false; maxModesTested = totalModesToTest > maxModesTested ? totalModesToTest : maxModesTested; int currModeId = 0; int default0Save1Load2 = 0; if (!isOneMode) { ctxStart = m_CABACEstimator->getCtx(); } for (int modeId = 0; modeId < nNumTransformCands; modeId++) { resiCb.copyFrom(orgResiCb[0]); resiCr.copyFrom(orgResiCr[0]); currTU.mtsIdx[compID] = currTU.cu->bdpcmModeChroma != BdpcmMode::NONE ? MtsType::SKIP : trModes[modeId].first; currModeId++; const bool isFirstMode = (currModeId == 1); const bool isLastMode = false; // Always store output to saveCS and tmpTU if (!(m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && slice.isLossless())) { // if DCT2's cbf==0, skip ts search if (!cbfDCT2 && trModes[modeId].first == MtsType::SKIP) { break; } if (!trModes[modeId].second) { continue; } } if (!isFirstMode) // if not first mode to be tested { m_CABACEstimator->getCtx() = ctxStart; } singleDistCTmp = 0; if (nNumTransformCands > 1) { xIntraCodingTUBlock(currTU, compID, singleDistCTmp, default0Save1Load2, nullptr, modeId == 0 ? &trModes : nullptr, true); } else { xIntraCodingTUBlock(currTU, compID, singleDistCTmp, default0Save1Load2); } if (((currTU.mtsIdx[compID] == MtsType::SKIP && currTU.cu->bdpcmModeChroma == BdpcmMode::NONE) && !TU::getCbf(currTU, compID))) // In order not to code TS flag when cbf is zero, the case for TS with // cbf being zero is forbidden. { if (m_pcEncCfg->getCostMode() != COST_LOSSLESS_CODING || !slice.isLossless()) { singleCostTmp = MAX_DOUBLE; } else { uint64_t fracBitsTmp = xGetIntraFracBitsQTChroma(currTU, compID); singleCostTmp = m_pcRdCost->calcRdCost(fracBitsTmp, singleDistCTmp); } } else if (lumaUsesISP && bestCostSoFar != MAX_DOUBLE && c == COMPONENT_Cb) { uint64_t fracBitsTmp = xGetIntraFracBitsQTSingleChromaComponent(cs, partitioner, ComponentID(c)); singleCostTmp = m_pcRdCost->calcRdCost(fracBitsTmp, singleDistCTmp); if (isOneMode || (!isOneMode && !isLastMode)) { m_CABACEstimator->getCtx() = ctxStart; } } else if (!isOneMode) { uint64_t fracBitsTmp = xGetIntraFracBitsQTChroma(currTU, compID); singleCostTmp = m_pcRdCost->calcRdCost(fracBitsTmp, singleDistCTmp); } if (singleCostTmp < singleCost) { singleCost = singleCostTmp; bestModeId = currModeId; if (c == COMPONENT_Cb) { bestCostCb = singleCostTmp; bestDistCb = singleDistCTmp; } else { bestCostCr = singleCostTmp; bestDistCr = singleDistCTmp; } if (currTU.mtsIdx[compID] == MtsType::DCT2_DCT2) { cbfDCT2 = TU::getCbfAtDepth(currTU, compID, currDepth); } if (!isLastMode) { #if KEEP_PRED_AND_RESI_SIGNALS saveCS.getPredBuf(area).copyFrom(cs.getPredBuf(area)); saveCS.getOrgResiBuf(area).copyFrom(cs.getOrgResiBuf(area)); #endif saveCS.getPredBuf(area).copyFrom(cs.getPredBuf(area)); if (keepResi) { saveCS.getResiBuf(area).copyFrom(cs.getResiBuf(area)); } saveCS.getRecoBuf(area).copyFrom(cs.getRecoBuf(area)); tmpTU.copyComponentFrom(currTU, compID); ctxBest = m_CABACEstimator->getCtx(); } } } if (lumaUsesISP && singleCost > bestCostSoFar && c == COMPONENT_Cb) { //Luma + Cb cost is already larger than the best cost, so we don't need to test Cr cs.dist = MAX_UINT; m_CABACEstimator->getCtx() = ctxStart; earlyExitISP = true; break; //return cbfs; } // Done with one component of separate coding of Cr and Cb, just switch to the best Cb contexts if Cr coding is still to be done if ((c == COMPONENT_Cb && bestModeId < totalModesToTest) || (c == COMPONENT_Cb && m_pcEncCfg->getCostMode() == COST_LOSSLESS_CODING && slice.isLossless())) { m_CABACEstimator->getCtx() = ctxBest; currTU.copyComponentFrom(tmpTU, COMPONENT_Cb); // Cbf of Cb is needed to estimate cost for Cr Cbf } } if ( !earlyExitISP ) { // Test using joint chroma residual coding double bestCostCbCr = bestCostCb + bestCostCr; Distortion bestDistCbCr = bestDistCb + bestDistCr; int bestJointCbCr = 0; CbfMaskList jointCbfMasksToTest; const bool cbfCb = TU::getCbf(tmpTU, COMPONENT_Cb); const bool cbfCr = TU::getCbf(tmpTU, COMPONENT_Cr); if (cs.sps->getJointCbCrEnabledFlag() && (cbfCb || cbfCr)) { m_pcTrQuant->selectICTCandidates(currTU, orgResiCb, orgResiCr, jointCbfMasksToTest); } const bool dctCb = cbfCb && tmpTU.mtsIdx[COMPONENT_Cb] == MtsType::DCT2_DCT2; const bool dctCr = cbfCr && tmpTU.mtsIdx[COMPONENT_Cr] == MtsType::DCT2_DCT2; const bool tsCb = cbfCb && tmpTU.mtsIdx[COMPONENT_Cb] == MtsType::SKIP; const bool tsCr = cbfCr && tmpTU.mtsIdx[COMPONENT_Cr] == MtsType::SKIP; const bool checkDctOnly = (dctCb && !cbfCr) || (dctCr && !cbfCb) || (dctCb && dctCr); const bool checkTsOnly = (tsCb && !cbfCr) || (tsCr && !cbfCb) || (tsCb && tsCr); if (jointCbfMasksToTest.size() && currTU.cu->bdpcmModeChroma != BdpcmMode::NONE) { CHECK(!checkTsOnly || checkDctOnly, "bdpcm only allows transform skip"); } for( int cbfMask : jointCbfMasksToTest ) { currTU.jointCbCr = (uint8_t) cbfMask; ComponentID codeCompId = (cbfMask & CBF_MASK_CB) != 0 ? COMPONENT_Cb : COMPONENT_Cr; ComponentID otherCompId = codeCompId == COMPONENT_Cb ? COMPONENT_Cr : COMPONENT_Cb; bool tsAllowed = TU::isTSAllowed(currTU, codeCompId) && (m_pcEncCfg->getUseChromaTS()) && !currTU.cu->lfnstIdx; uint8_t numTransformCands = 1 + (tsAllowed ? 1 : 0); // DCT + TS = 2 tests bool cbfDCT2 = true; TrModeList trModes; if (checkDctOnly || checkTsOnly) { numTransformCands = 1; } if (!checkTsOnly || currTU.cu->bdpcmModeChroma != BdpcmMode::NONE) { trModes.push_back(TrMode(MtsType::DCT2_DCT2, true)); // DCT2 } if (tsAllowed && !checkDctOnly) { trModes.push_back(TrMode(MtsType::SKIP, true)); // TS } for (int modeId = 0; modeId < numTransformCands; modeId++) { if (modeId && !cbfDCT2) { continue; } if (!trModes[modeId].second) { continue; } Distortion distTmp = 0; currTU.mtsIdx[codeCompId] = currTU.cu->bdpcmModeChroma != BdpcmMode::NONE ? MtsType::SKIP : trModes[modeId].first; currTU.mtsIdx[otherCompId] = MtsType::DCT2_DCT2; m_CABACEstimator->getCtx() = ctxStartTU; resiCb.copyFrom(orgResiCb[cbfMask]); resiCr.copyFrom(orgResiCr[cbfMask]); if (numTransformCands > 1) { xIntraCodingTUBlock(currTU, COMPONENT_Cb, distTmp, 0, nullptr, modeId == 0 ? &trModes : nullptr, true); } else { xIntraCodingTUBlock(currTU, COMPONENT_Cb, distTmp, 0); } double costTmp = std::numeric_limits::max(); if (distTmp < std::numeric_limits::max()) { uint64_t bits = xGetIntraFracBitsQTChroma(currTU, COMPONENT_Cb); costTmp = m_pcRdCost->calcRdCost(bits, distTmp); if (currTU.mtsIdx[codeCompId] == MtsType::DCT2_DCT2) { cbfDCT2 = true; } } else if (currTU.mtsIdx[codeCompId] == MtsType::DCT2_DCT2) { cbfDCT2 = false; } if (costTmp < bestCostCbCr) { bestCostCbCr = costTmp; bestDistCbCr = distTmp; bestJointCbCr = currTU.jointCbCr; // store data { #if KEEP_PRED_AND_RESI_SIGNALS saveCS.getOrgResiBuf(cbArea).copyFrom(cs.getOrgResiBuf(cbArea)); saveCS.getOrgResiBuf(crArea).copyFrom(cs.getOrgResiBuf(crArea)); #endif saveCS.getPredBuf(cbArea).copyFrom(cs.getPredBuf(cbArea)); saveCS.getPredBuf(crArea).copyFrom(cs.getPredBuf(crArea)); if (keepResi) { saveCS.getResiBuf(cbArea).copyFrom(cs.getResiBuf(cbArea)); saveCS.getResiBuf(crArea).copyFrom(cs.getResiBuf(crArea)); } saveCS.getRecoBuf(cbArea).copyFrom(cs.getRecoBuf(cbArea)); saveCS.getRecoBuf(crArea).copyFrom(cs.getRecoBuf(crArea)); tmpTU.copyComponentFrom(currTU, COMPONENT_Cb); tmpTU.copyComponentFrom(currTU, COMPONENT_Cr); ctxBest = m_CABACEstimator->getCtx(); } } } } // Retrieve the best CU data (unless it was the very last one tested) { #if KEEP_PRED_AND_RESI_SIGNALS cs.getPredBuf (cbArea).copyFrom(saveCS.getPredBuf (cbArea)); cs.getOrgResiBuf(cbArea).copyFrom(saveCS.getOrgResiBuf(cbArea)); cs.getPredBuf (crArea).copyFrom(saveCS.getPredBuf (crArea)); cs.getOrgResiBuf(crArea).copyFrom(saveCS.getOrgResiBuf(crArea)); #endif cs.getPredBuf (cbArea).copyFrom(saveCS.getPredBuf (cbArea)); cs.getPredBuf (crArea).copyFrom(saveCS.getPredBuf (crArea)); if( keepResi ) { cs.getResiBuf (cbArea).copyFrom(saveCS.getResiBuf (cbArea)); cs.getResiBuf (crArea).copyFrom(saveCS.getResiBuf (crArea)); } cs.getRecoBuf (cbArea).copyFrom(saveCS.getRecoBuf (cbArea)); cs.getRecoBuf (crArea).copyFrom(saveCS.getRecoBuf (crArea)); currTU.copyComponentFrom(tmpTU, COMPONENT_Cb); currTU.copyComponentFrom(tmpTU, COMPONENT_Cr); m_CABACEstimator->getCtx() = ctxBest; } // Copy results to the picture structures cs.picture->getRecoBuf(cbArea).copyFrom(cs.getRecoBuf(cbArea)); cs.picture->getRecoBuf(crArea).copyFrom(cs.getRecoBuf(crArea)); cs.picture->getPredBuf(cbArea).copyFrom(cs.getPredBuf(cbArea)); cs.picture->getPredBuf(crArea).copyFrom(cs.getPredBuf(crArea)); cbfs.cbf(COMPONENT_Cb) = TU::getCbf(currTU, COMPONENT_Cb); cbfs.cbf(COMPONENT_Cr) = TU::getCbf(currTU, COMPONENT_Cr); currTU.jointCbCr = ( (cbfs.cbf(COMPONENT_Cb) + cbfs.cbf(COMPONENT_Cr)) ? bestJointCbCr : 0 ); cs.dist += bestDistCbCr; } } else { unsigned numValidTBlocks = ::getNumberValidTBlocks( *cs.pcv ); ChromaCbfs SplitCbfs ( false ); if( partitioner.canSplit( TU_MAX_TR_SPLIT, cs ) ) { partitioner.splitCurrArea( TU_MAX_TR_SPLIT, cs ); } else if (currTU.cu->ispMode != ISPType::NONE) { partitioner.splitCurrArea( ispType, cs ); } else { THROW( "Implicit TU split not available" ); } do { ChromaCbfs subCbfs = xRecurIntraChromaCodingQT( cs, partitioner, bestCostSoFar, ispType ); for( uint32_t ch = COMPONENT_Cb; ch < numValidTBlocks; ch++ ) { const ComponentID compID = ComponentID( ch ); SplitCbfs.cbf( compID ) |= subCbfs.cbf( compID ); } } while( partitioner.nextPart( cs ) ); partitioner.exitCurrSplit(); if( lumaUsesISP && cs.dist == MAX_UINT ) { return cbfs; } cbfs.Cb |= SplitCbfs.Cb; cbfs.Cr |= SplitCbfs.Cr; if (!lumaUsesISP) { for (auto &ptu: cs.tus) { if (currArea.Cb().contains(ptu->Cb()) || (!ptu->Cb().valid() && currArea.Y().contains(ptu->Y()))) { TU::setCbfAtDepth(*ptu, COMPONENT_Cb, currDepth, SplitCbfs.Cb); TU::setCbfAtDepth(*ptu, COMPONENT_Cr, currDepth, SplitCbfs.Cr); } } } } return cbfs; } uint64_t IntraSearch::xFracModeBitsIntra(PredictionUnit &pu, const uint32_t &mode, const ChannelType &chType) { uint32_t orgMode = mode; if (!pu.ciipFlag) { std::swap(orgMode, pu.intraDir[chType]); } m_CABACEstimator->resetBits(); if( isLuma( chType ) ) { if (!pu.ciipFlag) { m_CABACEstimator->intra_luma_pred_mode(pu); } } else { m_CABACEstimator->intra_chroma_pred_mode( pu ); } if ( !pu.ciipFlag ) { std::swap(orgMode, pu.intraDir[chType]); } return m_CABACEstimator->getEstFracBits(); } void IntraSearch::sortRdModeListFirstColorSpace(ModeInfo mode, double cost, const BdpcmMode bdpcmMode, ModeInfo *rdModeList, double *rdCostList, BdpcmMode *bdpcmModeList, int &candNum) { if (candNum == 0) { rdModeList[0] = mode; rdCostList[0] = cost; bdpcmModeList[0] = bdpcmMode; candNum++; return; } int insertPos = -1; for (int pos = candNum - 1; pos >= 0; pos--) { if (cost < rdCostList[pos]) { insertPos = pos; } } if (insertPos >= 0) { for (int i = candNum - 1; i >= insertPos; i--) { rdModeList[i + 1] = rdModeList[i]; rdCostList[i + 1] = rdCostList[i]; bdpcmModeList[i + 1] = bdpcmModeList[i]; } rdModeList[insertPos] = mode; rdCostList[insertPos] = cost; bdpcmModeList[insertPos] = bdpcmMode; candNum++; } else { rdModeList[candNum] = mode; rdCostList[candNum] = cost; bdpcmModeList[candNum] = bdpcmMode; candNum++; } CHECK(candNum > FAST_UDI_MAX_RDMODE_NUM, "exceed intra mode candidate list capacity"); return; } void IntraSearch::invalidateBestRdModeFirstColorSpace() { int numSaveRdClass = 4 * NUM_LFNST_NUM_PER_SET * 2; int savedRdModeListSize = FAST_UDI_MAX_RDMODE_NUM; for (int i = 0; i < numSaveRdClass; i++) { m_numSavedRdModeFirstColorSpace[i] = 0; for (int j = 0; j < savedRdModeListSize; j++) { m_savedRdModeFirstColorSpace[i][j] = ModeInfo(false, false, 0, ISPType::NONE, 0); m_savedBDPCMModeFirstColorSpace[i][j] = BdpcmMode::NONE; m_savedRdCostFirstColorSpace[i][j] = MAX_DOUBLE; } } } template void IntraSearch::reduceHadCandList(static_vector& candModeList, static_vector& candCostList, int& numModesForFullRD, const double thresholdHadCost, const double* mipHadCost, const PredictionUnit &pu, const bool fastMip) { const int maxCandPerType = numModesForFullRD >> 1; static_vector tempRdModeList; static_vector tempCandCostList; const double minCost = candCostList[0]; bool keepOneMip = candModeList.size() > numModesForFullRD; int numConv = 0; int numMip = 0; for (int idx = 0; idx < candModeList.size() - (keepOneMip?0:1); idx++) { bool addMode = false; const ModeInfo& orgMode = candModeList[idx]; if (!orgMode.mipFlg) { addMode = (numConv < 3); numConv += addMode ? 1:0; } else { addMode = ( numMip < maxCandPerType || (candCostList[idx] < thresholdHadCost * minCost) || keepOneMip ); keepOneMip = false; numMip += addMode ? 1:0; } if( addMode ) { tempRdModeList.push_back(orgMode); tempCandCostList.push_back(candCostList[idx]); } } if ((pu.lwidth() > 8 && pu.lheight() > 8)) { // Sort MIP candidates by Hadamard cost const int transpOff = MatrixIntraPrediction::getNumModesMip(pu.Y()); static_vector sortedMipModes(0); static_vector sortedMipCost(0); for( uint8_t mode : { 0, 1, 2 } ) { uint8_t candMode = mode + uint8_t((mipHadCost[mode + transpOff] < mipHadCost[mode]) ? transpOff : 0); updateCandList(candMode, mipHadCost[candMode], sortedMipModes, sortedMipCost, 3); } // Append MIP mode to RD mode list const int modeListSize = int(tempRdModeList.size()); for (int idx = 0; idx < 3; idx++) { const bool isTransposed = (sortedMipModes[idx] >= transpOff ? true : false); const uint32_t mipIdx = (isTransposed ? sortedMipModes[idx] - transpOff : sortedMipModes[idx]); const ModeInfo mipMode(true, isTransposed, 0, ISPType::NONE, mipIdx); bool alreadyIncluded = false; for (int modeListIdx = 0; modeListIdx < modeListSize; modeListIdx++) { if (tempRdModeList[modeListIdx] == mipMode) { alreadyIncluded = true; break; } } if (!alreadyIncluded) { tempRdModeList.push_back(mipMode); tempCandCostList.push_back(0); if (fastMip) { break; } } } } candModeList = tempRdModeList; candCostList = tempCandCostList; numModesForFullRD = int(candModeList.size()); } // It decides which modes from the ISP lists can be full RD tested void IntraSearch::xGetNextISPMode(ModeInfo& modeInfo, const ModeInfo* lastMode, const Size cuSize) { if (m_curIspLfnstIdx >= NUM_LFNST_NUM_PER_SET) { // All LFNST indices have been checked return; } ISPType nextISPcandSplitType; auto &ispTestedModes = m_ispTestedModes[m_curIspLfnstIdx]; const bool horSplitIsTerminated = ispTestedModes.splitIsFinished[ISPType::HOR]; const bool verSplitIsTerminated = ispTestedModes.splitIsFinished[ISPType::VER]; if (!horSplitIsTerminated && !verSplitIsTerminated) { nextISPcandSplitType = !lastMode ? ISPType::HOR : lastMode->ispMod == ISPType::HOR ? ISPType::VER : ISPType::HOR; } else if (!horSplitIsTerminated && verSplitIsTerminated) { nextISPcandSplitType = ISPType::HOR; } else if (horSplitIsTerminated && !verSplitIsTerminated) { nextISPcandSplitType = ISPType::VER; } else { xFinishISPModes(); return; // no more modes will be tested } int maxNumSubPartitions = ispTestedModes.numTotalParts[nextISPcandSplitType]; // We try to break the split here for lfnst > 0 according to the first mode if (m_curIspLfnstIdx > 0 && ispTestedModes.numTestedModes[nextISPcandSplitType] == 1) { int firstModeThisSplit = ispTestedModes.getTestedIntraMode(nextISPcandSplitType, 0); int numSubPartsFirstModeThisSplit = ispTestedModes.getNumCompletedSubParts(nextISPcandSplitType, firstModeThisSplit); CHECK(numSubPartsFirstModeThisSplit < 0, "wrong number of subpartitions!"); bool stopThisSplit = false; bool stopThisSplitAllLfnsts = false; if (numSubPartsFirstModeThisSplit < maxNumSubPartitions) { stopThisSplit = true; if (m_pcEncCfg->getUseFastISP() && m_curIspLfnstIdx == 1 && numSubPartsFirstModeThisSplit < maxNumSubPartitions - 1) { stopThisSplitAllLfnsts = true; } } if (stopThisSplit) { ispTestedModes.splitIsFinished[nextISPcandSplitType] = true; if (m_curIspLfnstIdx == 1 && stopThisSplitAllLfnsts) { m_ispTestedModes[2].splitIsFinished[nextISPcandSplitType] = true; } return; } } // We try to break the split here for lfnst = 0 or all lfnst indices according to the first two modes if (m_curIspLfnstIdx == 0 && ispTestedModes.numTestedModes[nextISPcandSplitType] == 2) { // Split stop criteria after checking the performance of previously tested intra modes const int thresholdSplit1 = maxNumSubPartitions; bool stopThisSplit = false; bool stopThisSplitForAllLFNSTs = false; const int thresholdSplit1ForAllLFNSTs = maxNumSubPartitions - 1; std::array modes; int numSubPartsBestMode[2]; for (int i = 0; i < 2; i++) { modes[i] = ispTestedModes.getTestedIntraMode(nextISPcandSplitType, i); modes[i] = modes[i] == DC_IDX ? NOMODE_IDX : modes[i]; numSubPartsBestMode[i] = modes[i] != NOMODE_IDX ? ispTestedModes.getNumCompletedSubParts(nextISPcandSplitType, modes[i]) : -1; } // 1) The 2 most promising modes do not reach a certain number of sub-partitions if (numSubPartsBestMode[0] != -1 && numSubPartsBestMode[1] != -1) { if (numSubPartsBestMode[0] < thresholdSplit1 && numSubPartsBestMode[1] < thresholdSplit1) { stopThisSplit = true; if (m_curIspLfnstIdx == 0 && numSubPartsBestMode[0] < thresholdSplit1ForAllLFNSTs && numSubPartsBestMode[1] < thresholdSplit1ForAllLFNSTs) { stopThisSplitForAllLFNSTs = true; } } else { // we stop also if the cost is MAX_DOUBLE for all modes if (std::find_if(modes.begin(), modes.end(), [&](const int &x) { return ispTestedModes.getRDCost(nextISPcandSplitType, x) < MAX_DOUBLE; }) == modes.end()) { stopThisSplit = true; } } } if (!stopThisSplit) { int numSubPartsBestModeAltSplit[2]; // 2) One split type may be discarded by comparing the number of sub-partitions of the best angle modes of both splits const ISPType otherSplit = nextISPcandSplitType == ISPType::HOR ? ISPType::VER : ISPType::HOR; numSubPartsBestModeAltSplit[1] = modes[1] != NOMODE_IDX ? ispTestedModes.getNumCompletedSubParts(otherSplit, modes[1]) : -1; if (numSubPartsBestModeAltSplit[1] != -1 && numSubPartsBestMode[1] != -1 && ispTestedModes.bestSplitSoFar != nextISPcandSplitType) { if (numSubPartsBestModeAltSplit[1] > numSubPartsBestMode[1]) { stopThisSplit = true; } // both have the same number of subpartitions else if (numSubPartsBestModeAltSplit[1] == numSubPartsBestMode[1]) { // both have the maximum number of subpartitions, so it compares RD costs to decide if (numSubPartsBestModeAltSplit[1] == maxNumSubPartitions) { double rdCostBestMode2ThisSplit = ispTestedModes.getRDCost(nextISPcandSplitType, modes[1]); double rdCostBestMode2OtherSplit = ispTestedModes.getRDCost(otherSplit, modes[1]); double threshold = 1.3; if (rdCostBestMode2ThisSplit == MAX_DOUBLE || rdCostBestMode2OtherSplit < rdCostBestMode2ThisSplit * threshold) { stopThisSplit = true; } } else // none of them reached the maximum number of subpartitions with the best angle modes, so it compares the results with the the planar mode { numSubPartsBestModeAltSplit[0] = modes[0] != -1 ? ispTestedModes.getNumCompletedSubParts(otherSplit, modes[0]) : -1; if (numSubPartsBestModeAltSplit[0] != -1 && numSubPartsBestMode[0] != -1 && numSubPartsBestModeAltSplit[0] > numSubPartsBestMode[0]) { stopThisSplit = true; } } } } } if (stopThisSplit) { ispTestedModes.splitIsFinished[nextISPcandSplitType] = true; if (stopThisSplitForAllLFNSTs) { for (int lfnstIdx = 1; lfnstIdx < NUM_LFNST_NUM_PER_SET; lfnstIdx++) { m_ispTestedModes[lfnstIdx].splitIsFinished[nextISPcandSplitType] = true; } } return; } } // Now a new mode is retrieved from the list and it has to be decided whether it should be tested or not if (ispTestedModes.candIndexInList[nextISPcandSplitType] < m_ispCandList[nextISPcandSplitType].size()) { ModeInfo candidate = m_ispCandList[nextISPcandSplitType].at(ispTestedModes.candIndexInList[nextISPcandSplitType]); ispTestedModes.candIndexInList[nextISPcandSplitType]++; // extra modes are only tested if ISP has won so far if (ispTestedModes.candIndexInList[nextISPcandSplitType] > ispTestedModes.numOrigModesToTest) { if (ispTestedModes.bestSplitSoFar != candidate.ispMod || ispTestedModes.bestModeSoFar == PLANAR_IDX) { ispTestedModes.splitIsFinished[nextISPcandSplitType] = true; return; } } bool testCandidate = true; // we look for a reference mode that has already been tested within the window and decide to test the new one according to the reference mode costs if (maxNumSubPartitions > 2 && (m_curIspLfnstIdx > 0 || (candidate.modeId >= DC_IDX && ispTestedModes.numTestedModes[nextISPcandSplitType] >= 2))) { std::array similarModes; similarModes.fill(NOMODE_IDX); constexpr int ANG_WINDOW_SIZE = 5; const int windowSize = candidate.modeId > DC_IDX ? ANG_WINDOW_SIZE : 1; int refLfnstIdx = m_curIspLfnstIdx; xFindAlreadyTestedNearbyIntraModes((int) candidate.modeId, refLfnstIdx, similarModes, candidate.ispMod, windowSize); int numSubPartsRefMode = 0; if (refLfnstIdx != m_curIspLfnstIdx) { numSubPartsRefMode = m_ispTestedModes[refLfnstIdx].getNumCompletedSubParts(candidate.ispMod, candidate.modeId); CHECK(numSubPartsRefMode <= 0, "Wrong value of the number of subpartitions completed!"); } else { for (auto m: similarModes) { if (m != NOMODE_IDX) { numSubPartsRefMode = std::max(numSubPartsRefMode, ispTestedModes.getNumCompletedSubParts(candidate.ispMod, m)); } } } if (numSubPartsRefMode > 0) { const int numSamples = cuSize.width << floorLog2(cuSize.height); const int numSubPartsLimit = numSamples >= 256 ? maxNumSubPartitions - 1 : 2; // The mode was found. Now we check the condition testCandidate = numSubPartsRefMode > numSubPartsLimit; } } if (testCandidate) { modeInfo = candidate; } } else { //the end of the list was reached, so the split is invalidated ispTestedModes.splitIsFinished[nextISPcandSplitType] = true; } } void IntraSearch::xFindAlreadyTestedNearbyIntraModes(const int currentIntraMode, int &refLfnstIdx, std::array &similarModes, const ISPType ispOption, const int windowSize) { //first we check if the exact intra mode was already tested for another lfnstIdx value for (int idx = refLfnstIdx - 1; idx >= 0; idx--) { if (m_ispTestedModes[idx].modeHasBeenTested[currentIntraMode][ispOption]) { refLfnstIdx = idx; return; } } //The mode has not been checked for another lfnstIdx value, so now we look for a similar mode within a window using the same lfnstIdx for (int k = 1; k <= windowSize; k++) { const int leftMode = (currentIntraMode + NUM_INTRA_ANGULAR_MODES - ANGULAR_BASE - k) % NUM_INTRA_ANGULAR_MODES + ANGULAR_BASE; const int rightMode = currentIntraMode < ANGULAR_BASE ? PLANAR_IDX : (currentIntraMode - ANGULAR_BASE + k) % NUM_INTRA_ANGULAR_MODES + ANGULAR_BASE; auto found = [&](int m) { return m != currentIntraMode ? m_ispTestedModes[refLfnstIdx].modeHasBeenTested[m][ispOption] : false; }; const bool leftModeFound = found(leftMode); const bool rightModeFound = found(rightMode); if (leftModeFound || rightModeFound) { similarModes[0] = leftModeFound ? leftMode : NOMODE_IDX; similarModes[1] = rightModeFound ? rightMode : NOMODE_IDX; return; } } } //It prepares the list of potential intra modes candidates that will be tested using RD costs bool IntraSearch::xSortISPCandList(double bestCostSoFar, double bestNonISPCost, const ModeInfo &bestNonISPMode) { int bestISPModeInRelCU = NOMODE_IDX; m_modeCtrl->setStopNonDCT2Transforms(false); if (m_pcEncCfg->getUseFastISP()) { // check if the ISP tests can be skipped const double thSkipISP = 1.4; if (bestNonISPCost > bestCostSoFar * thSkipISP) { for (int j = 0; j < NUM_LFNST_NUM_PER_SET; j++) { m_ispTestedModes[j].splitIsFinished.fill(true); } return false; } if (!updateISPStatusFromRelCU(bestNonISPCost, bestNonISPMode, bestISPModeInRelCU)) { return false; } } for (auto &c: m_ispCandList[ISPType::HOR]) { // set the correct ISP split type value c.ispMod = ISPType::HOR; } auto origHadList = m_ispCandList[ISPType::HOR]; // save the original hadamard list of regular intra ModeInfo refMode = origHadList.front(); m_ispCandList[ISPType::HOR].clear(); m_ispCandList[ISPType::VER].clear(); // we sort the normal intra modes according to their full RD costs std::stable_sort(m_regIntraRDListWithCosts.begin(), m_regIntraRDListWithCosts.end(), ModeInfoWithCost::compare); // we get the best angle from the regular intra list const auto p = std::find_if(m_regIntraRDListWithCosts.begin(), m_regIntraRDListWithCosts.end(), [](const ModeInfoWithCost &mi) { return mi.modeId >= ANGULAR_BASE; }); const int bestNormalIntraAngle = p == m_regIntraRDListWithCosts.end() ? NOMODE_IDX : p->modeId; auto &destList = m_ispCandList[ISPType::HOR]; std::array modeIsInList; modeIsInList.fill(false); //List creation auto addMode = [&](const int m) -> bool { if (!modeIsInList[m]) { refMode.modeId = m; destList.push_back(refMode); modeIsInList[m] = true; return true; } return false; }; if (m_pcEncCfg->getUseFastISP() && bestISPModeInRelCU != NOMODE_IDX) // RelCU intra mode { addMode(bestISPModeInRelCU); } // Planar addMode(PLANAR_IDX); // Best angle in regular intra if (bestNormalIntraAngle != NOMODE_IDX) { addMode(bestNormalIntraAngle); } // Remaining regular intra modes that were full RD tested (except DC, which is added after the angles from regular intra) bool addDc = false; for (const auto &e: m_regIntraRDListWithCosts) { if (e.modeId == DC_IDX) { addDc = true; } else { addMode(e.modeId); } } // DC is added after the angles from regular intra if (addDc) { addMode(DC_IDX); } // We add extra candidates to the list that will only be tested if ISP is likely to win for (int j = 0; j < NUM_LFNST_NUM_PER_SET; j++) { m_ispTestedModes[j].numOrigModesToTest = (int) destList.size(); } const int addedModesFromHadList = 3; int newModesAdded = 0; for (const auto &e: origHadList) { if (addMode(e.modeId)) { if (++newModesAdded == addedModesFromHadList) { break; } } } if (m_pcEncCfg->getUseFastISP() && bestISPModeInRelCU != NOMODE_IDX) { destList.resize(1); } // Copy modes to other split-type list m_ispCandList[ISPType::VER] = m_ispCandList[ISPType::HOR]; for (auto &x: m_ispCandList[ISPType::VER]) { x.ispMod = ISPType::VER; } // Reset the tested modes information to 0 for (int j = 0; j < NUM_LFNST_NUM_PER_SET; j++) { for (const auto &x: m_ispCandList[ISPType::HOR]) { m_ispTestedModes[j].clearISPModeInfo(x.modeId); } } return true; } void IntraSearch::xSortISPCandListLFNST() { //It resorts the list of intra mode candidates for lfnstIdx > 0 by checking the RD costs for lfnstIdx = 0 ISPTestedModesInfo& ispTestedModesRef = m_ispTestedModes[0]; for (const auto ispMode: { ISPType::HOR, ISPType::VER }) { if (!m_ispTestedModes[m_curIspLfnstIdx].splitIsFinished[ispMode] && ispTestedModesRef.testedModes[ispMode].size() > 1) { auto &candList = m_ispCandList[ispMode]; int bestModeId = candList[1].modeId > DC_IDX ? candList[1].modeId : NOMODE_IDX; int bestSubParts = candList[1].modeId > DC_IDX ? ispTestedModesRef.getNumCompletedSubParts(ispMode, bestModeId) : -1; double bestCost = candList[1].modeId > DC_IDX ? ispTestedModesRef.getRDCost(ispMode, bestModeId) : MAX_DOUBLE; for (int i = 0; i < candList.size(); i++) { const int candSubParts = ispTestedModesRef.getNumCompletedSubParts(ispMode, candList[i].modeId); const double candCost = ispTestedModesRef.getRDCost(ispMode, candList[i].modeId); if (candSubParts > bestSubParts || candCost < bestCost) { bestModeId = candList[i].modeId; bestCost = candCost; bestSubParts = candSubParts; } } if (bestModeId != NOMODE_IDX && bestModeId != candList[0].modeId) { auto prevMode = candList[0]; candList[0].modeId = bestModeId; for (int i = 1; i < candList.size(); i++) { auto nextMode = candList[i]; candList[i] = prevMode; if (nextMode.modeId == bestModeId) { break; } prevMode = nextMode; } } } } } bool IntraSearch::updateISPStatusFromRelCU(double bestNonISPCostCurrCu, const ModeInfo &bestNonISPModeCurrCu, int &bestISPModeInRelCU) { //It compares the data of a related CU with the current CU to cancel or reduce the ISP tests bestISPModeInRelCU = NOMODE_IDX; if (m_modeCtrl->getRelatedCuIsValid()) { const IspPredModeVal ispPredModeVal = m_modeCtrl->getIspPredModeValRelCU(); const bool bestModeRelCuIsMip = ispPredModeVal.mipFlag; const int relatedCuIntraMode = ispPredModeVal.bestPredModeDCT2; double bestNonISPCostRelCU = m_modeCtrl->getBestDCT2NonISPCostRelCU(); double costRatio = bestNonISPCostCurrCu / bestNonISPCostRelCU; bool bestModeCurrCuIsMip = bestNonISPModeCurrCu.mipFlg; bool isSameTypeOfMode = bestModeRelCuIsMip == bestModeCurrCuIsMip; bool bothModesAreAngular = isSameTypeOfMode && !bestModeCurrCuIsMip && bestNonISPModeCurrCu.modeId > DC_IDX && relatedCuIntraMode > DC_IDX; bool modesAreComparable = isSameTypeOfMode && (bestNonISPModeCurrCu.modeId == relatedCuIntraMode || (bothModesAreAngular && abs(relatedCuIntraMode - (int) bestNonISPModeCurrCu.modeId) <= 5)); CHECK(!ispPredModeVal.valid, "Wrong ISP relCU status"); if (ispPredModeVal.notIsp) // ISP was not selected in the relCU { double bestNonDCT2Cost = m_modeCtrl->getBestNonDCT2Cost(); double ratioWithNonDCT2 = bestNonDCT2Cost / bestNonISPCostRelCU; const double margin = ratioWithNonDCT2 < 0.95 ? 0.2 : 0.1; if (costRatio > 1.0 - margin && costRatio < 1.0 + margin && modesAreComparable) { for (int lfnstVal = 0; lfnstVal < NUM_LFNST_NUM_PER_SET; lfnstVal++) { m_ispTestedModes[lfnstVal].splitIsFinished[ISPType::HOR] = true; m_ispTestedModes[lfnstVal].splitIsFinished[ISPType::VER] = true; } return false; } } else { const double margin = 0.05; if (costRatio > 1.0 - margin && costRatio < 1.0 + margin && modesAreComparable) { bestISPModeInRelCU = (int)m_modeCtrl->getBestISPIntraModeRelCU(); for (const auto splitIdx: { ISPType::HOR, ISPType::VER }) { for (int lfnstVal = 0; lfnstVal < NUM_LFNST_NUM_PER_SET; lfnstVal++) { if (lfnstVal == ispPredModeVal.ispLfnstIdx && splitIdx == (ispPredModeVal.verIsp == 0 ? ISPType::HOR : ISPType::VER)) { continue; } m_ispTestedModes[lfnstVal].splitIsFinished[splitIdx] = true; } } m_modeCtrl->setStopNonDCT2Transforms(ispPredModeVal.lowIspCost); } } } return true; } void IntraSearch::xFinishISPModes() { //Continue to the next lfnst index m_curIspLfnstIdx++; if (m_curIspLfnstIdx < NUM_LFNST_NUM_PER_SET) { //Check if LFNST is applicable if (m_curIspLfnstIdx == 1) { bool canTestLFNST = false; for (int lfnstIdx = 1; lfnstIdx < NUM_LFNST_NUM_PER_SET; lfnstIdx++) { canTestLFNST |= !m_ispTestedModes[lfnstIdx].splitIsFinished[ISPType::HOR] || !m_ispTestedModes[lfnstIdx].splitIsFinished[ISPType::VER]; } if (canTestLFNST) { //Construct the intra modes candidates list for the lfnst > 0 cases xSortISPCandListLFNST(); } } } }