Newer
Older

Karsten Suehring
committed
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
#if JVET_AA0126_GLM
for (int32_t uiMode = 0; uiMode < NUM_LMC_MODE; uiMode++)
{
int chromaIntraMode = LM_CHROMA_IDX + uiMode;
if ( PU::isLMCModeEnabled( pu, chromaIntraMode ) && PU::hasGlmFlag( pu, chromaIntraMode ) )
{
if ( satdGlmIdcBest[chromaIntraMode - LM_CHROMA_IDX].isActive() )
{
pu.intraDir[1] = chromaIntraMode;
pu.glmIdc = satdGlmIdcBest[chromaIntraMode - LM_CHROMA_IDX];
// RD search replicated from above
cs.setDecomp( pu.Cb(), false );
cs.dist = baseDist;
//----- restore context models -----
m_CABACEstimator->getCtx() = ctxStart;
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 uiDist = cs.dist;
double dCost = m_pcRdCost->calcRdCost( fracBits, uiDist - baseDist );
//----- compare -----
if( dCost < dBestCost )
{
if( lumaUsesISP && dCost < bestCostSoFar )
{
bestCostSoFar = dCost;
}
for( uint32_t i = getFirstComponentOfChannel( CHANNEL_TYPE_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 );
}
}
dBestCost = dCost;
uiBestDist = uiDist;
uiBestMode = chromaIntraMode;
bestBDPCMMode = cu.bdpcmModeChroma;
bestGlmIdc = pu.glmIdc;
}
if ( chromaIntraMode == LM_CHROMA_IDX && !bestGlmIdc.isActive() )
{
break;
}
}
}
}
pu.glmIdc.setAllZero();
#endif
#if JVET_Z0050_DIMD_CHROMA_FUSION
// RDO for chroma fusion mode
for (int32_t uiMode = 0; uiMode < 1; uiMode++)
{
int chromaIntraMode = bestNonLmMode;
#if ENABLE_DIMD
if (!pu.cs->slice->isIntra() && cu.slice->getSPS()->getUseDimd())
chromaIntraMode = DIMD_CHROMA_IDX;
#endif
if (PU::hasChromaFusionFlag(pu, chromaIntraMode))
pu.isChromaFusion = true;
cs.setDecomp(pu.Cb(), false);
cs.dist = baseDist;
//----- restore context models -----
m_CABACEstimator->getCtx() = ctxStart;
//----- chroma coding -----
pu.intraDir[1] = chromaIntraMode;
xRecurIntraChromaCodingQT(cs, partitioner, bestCostSoFar, ispType);
if (lumaUsesISP && cs.dist == MAX_UINT)
if (cs.sps->getTransformSkipEnabledFlag())
m_CABACEstimator->getCtx() = ctxStart;
}
uint64_t fracBits = xGetIntraFracBitsQT(cs, partitioner, false, true, -1, ispType);
Distortion uiDist = cs.dist;
double dCost = m_pcRdCost->calcRdCost(fracBits, uiDist - baseDist);
if (dCost < dBestCost)
{
if (lumaUsesISP && dCost < bestCostSoFar)
for (uint32_t i = getFirstComponentOfChannel(CHANNEL_TYPE_CHROMA); i < numberValidComponents; i++)
{
const CompArea &area = pu.blocks[i];
saveCS.getRecoBuf(area).copyFrom(cs.getRecoBuf(area));
saveCS.getPredBuf(area).copyFrom(cs.getPredBuf(area));
saveCS.getResiBuf(area).copyFrom(cs.getResiBuf(area));
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);
}
}
dBestCost = dCost;
uiBestDist = uiDist;
uiBestMode = chromaIntraMode;
bestBDPCMMode = cu.bdpcmModeChroma;
isChromaFusion = pu.isChromaFusion;
}
}
pu.isChromaFusion = false;
#endif
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
#if JVET_Z0050_CCLM_SLOPE
#if MMLM
for (int32_t uiMode = 0; uiMode < 2; uiMode++)
{
int chromaIntraMode = uiMode ? MMLM_CHROMA_IDX : LM_CHROMA_IDX;
#else
for (int32_t uiMode = 0; uiMode < 1; uiMode++)
{
int chromaIntraMode = LM_CHROMA_IDX;
#endif
if ( PU::isLMCModeEnabled( pu, chromaIntraMode ) && PU::hasCclmDeltaFlag( pu, chromaIntraMode ) )
{
if ( satdCclmOffsetsBest[chromaIntraMode - LM_CHROMA_IDX].isActive() )
{
pu.intraDir[1] = chromaIntraMode;
pu.cclmOffsets = satdCclmOffsetsBest[chromaIntraMode - LM_CHROMA_IDX];
#if JVET_Z0050_DIMD_CHROMA_FUSION
pu.isChromaFusion = false;
#endif
// RD search replicated from above
cs.setDecomp( pu.Cb(), false );
cs.dist = baseDist;
//----- restore context models -----
m_CABACEstimator->getCtx() = ctxStart;
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 uiDist = cs.dist;
double dCost = m_pcRdCost->calcRdCost( fracBits, uiDist - baseDist );
//----- compare -----
if( dCost < dBestCost )
{
if( lumaUsesISP && dCost < bestCostSoFar )
{
bestCostSoFar = dCost;
}
for( uint32_t i = getFirstComponentOfChannel( CHANNEL_TYPE_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 );
}
}
dBestCost = dCost;
uiBestDist = uiDist;
uiBestMode = chromaIntraMode;
bestBDPCMMode = cu.bdpcmModeChroma;
bestCclmOffsets = pu.cclmOffsets;
#if JVET_Z0050_DIMD_CHROMA_FUSION
isChromaFusion = pu.isChromaFusion;
#endif
#if JVET_AA0126_GLM
bestGlmIdc = pu.glmIdc;
#endif
}
}
}
}
pu.cclmOffsets.setAllZero();
#endif
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
#if JVET_AB0143_CCCM_TS
int chromaIntraModeInCCCM = LM_CHROMA_IDX;
isCCCMEnabled = isCccmFullEnabled;
pu.cccmFlag = 1;
#if MMLM
for (int32_t uiMode = 0; uiMode < 6; uiMode++)
#else
for (int32_t uiMode = 0; uiMode < 3; uiMode++)
#endif
{
if (uiMode == 1)
{
chromaIntraModeInCCCM = MDLM_L_IDX;
isCCCMEnabled = isCccmLeftEnabled;
pu.cccmFlag = 2;
}
else if (uiMode == 2)
{
chromaIntraModeInCCCM = MDLM_T_IDX;
isCCCMEnabled = isCccmTopEnabled;
pu.cccmFlag = 3;
}
#if MMLM
else if (uiMode == 3)
{
chromaIntraModeInCCCM = MMLM_CHROMA_IDX;
isCCCMEnabled = isMultiCccmFullEnabled;
pu.cccmFlag = 1;
}
else if (uiMode == 4)
{
chromaIntraModeInCCCM = MMLM_L_IDX;
isCCCMEnabled = isMultiCccmLeftEnabled;
pu.cccmFlag = 2;
}
else if (uiMode == 5)
{
chromaIntraModeInCCCM = MMLM_T_IDX;
isCCCMEnabled = isMultiCccmTopEnabled;
pu.cccmFlag = 3;
}
#endif
if (!isCccmModeEnabledInRdo[chromaIntraModeInCCCM])
{
continue;
}
if (isCCCMEnabled)
{
#else
#if MMLM
for (int32_t uiMode = 0; uiMode < 2; uiMode++)
{
int chromaIntraMode = uiMode ? MMLM_CHROMA_IDX : LM_CHROMA_IDX;
#else
for (int32_t uiMode = 0; uiMode < 1; uiMode++)
{
int chromaIntraMode = LM_CHROMA_IDX;
#endif
if ( PU::cccmSingleModeAvail(pu, chromaIntraMode) || PU::cccmMultiModeAvail(pu, chromaIntraMode) )
{
pu.cccmFlag = 1;
// Original RD check code replicated from above
cs.setDecomp( pu.Cb(), false );
cs.dist = baseDist;
//----- restore context models -----
m_CABACEstimator->getCtx() = ctxStart;
//----- chroma coding -----
#if JVET_AB0143_CCCM_TS
pu.intraDir[1] = chromaIntraModeInCCCM;
xRecurIntraChromaCodingQT(cs, partitioner, bestCostSoFar, ispType, cccmStorage[uiMode]);
#else
pu.intraDir[1] = chromaIntraMode;
xRecurIntraChromaCodingQT( cs, partitioner, bestCostSoFar, ispType );
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
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 uiDist = cs.dist;
double dCost = m_pcRdCost->calcRdCost( fracBits, uiDist - baseDist );
//----- compare -----
if( dCost < dBestCost )
{
if( lumaUsesISP && dCost < bestCostSoFar )
{
bestCostSoFar = dCost;
}
for( uint32_t i = getFirstComponentOfChannel( CHANNEL_TYPE_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 );
}
}
dBestCost = dCost;
uiBestDist = uiDist;
#if JVET_AB0143_CCCM_TS
uiBestMode = chromaIntraModeInCCCM;
#else
bestBDPCMMode = cu.bdpcmModeChroma;
#if JVET_Z0050_DIMD_CHROMA_FUSION
isChromaFusion = pu.isChromaFusion;
#endif
#if JVET_Z0050_CCLM_SLOPE
bestCclmOffsets = pu.cclmOffsets;
#endif
cccmModeBest = pu.cccmFlag;
#if JVET_AA0126_GLM
bestGlmIdc = pu.glmIdc;
#endif
}
}
}
pu.cccmFlag = 0;
#endif

Karsten Suehring
committed
for( uint32_t i = getFirstComponentOfChannel( CHANNEL_TYPE_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 ) );
#if JVET_Z0118_GDR
cs.updateReconMotIPM(area);
#else

Karsten Suehring
committed
cs.picture->getRecoBuf( area ).copyFrom( cs. getRecoBuf( area ) );

Karsten Suehring
committed
for( uint32_t j = 0; j < saveCS.tus.size(); j++ )
{
orgTUs[ j ]->copyComponentFrom( *saveCS.tus[ j ], area.compID );
}
}
}
pu.intraDir[1] = uiBestMode;
cs.dist = uiBestDist;
cu.bdpcmModeChroma = bestBDPCMMode;
#if JVET_Z0050_CCLM_SLOPE
pu.cclmOffsets = bestCclmOffsets;
#if JVET_AA0057_CCCM
pu.cccmFlag = cccmModeBest;
#endif
#if JVET_Z0050_DIMD_CHROMA_FUSION
pu.isChromaFusion = isChromaFusion;
#endif
#if JVET_AA0126_GLM
pu.glmIdc = bestGlmIdc;

Karsten Suehring
committed
}
//----- restore context models -----
m_CABACEstimator->getCtx() = ctxStart;
if( lumaUsesISP && bestCostSoFar >= maxCostAllowed )
{
cu.ispMode = 0;
}

Karsten Suehring
committed
}
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
#if JVET_Z0050_CCLM_SLOPE
void IntraSearch::xFindBestCclmDeltaSlopeSATD(PredictionUnit &pu, ComponentID compID, int cclmModel, int &deltaBest, int64_t &sadBest )
{
CclmModel cclmModelStored;
CodingStructure& cs = *(pu.cs);
CompArea area = compID == COMPONENT_Cb ? pu.Cb() : pu.Cr();
PelBuf orgBuf = cs.getOrgBuf(area);
PelBuf predBuf = cs.getPredBuf(area);
int maxOffset = 4;
int mode = pu.intraDir[1];
bool createNewModel = true;
DistParam distParamSad;
DistParam distParamSatd;
m_pcRdCost->setDistParam(distParamSad, orgBuf, predBuf, pu.cs->sps->getBitDepth(CHANNEL_TYPE_CHROMA), compID, false);
m_pcRdCost->setDistParam(distParamSatd, orgBuf, predBuf, pu.cs->sps->getBitDepth(CHANNEL_TYPE_CHROMA), compID, true);
distParamSad.applyWeight = false;
distParamSatd.applyWeight = false;
sadBest = -1;
// Search positive offsets
for ( int offset = 0; offset <= maxOffset; offset++)
{
pu.cclmOffsets.setOffset(compID, cclmModel, offset);
predIntraChromaLM( compID, predBuf, pu, area, mode, createNewModel, &cclmModelStored );
createNewModel = false; // Need to calculate the base model just once
int64_t sad = distParamSad.distFunc(distParamSad) * 2;
int64_t satd = distParamSatd.distFunc(distParamSatd);
int64_t sadThis = std::min(sad, satd);
if ( sadBest == -1 || sadThis < sadBest )
{
sadBest = sadThis;
deltaBest = offset;
}
else
{
break;
}
}
// Search negative offsets only if positives didn't help
if ( deltaBest == 0 )
{
for ( int offset = -1; offset >= -maxOffset; offset--)
{
pu.cclmOffsets.setOffset(compID, cclmModel, offset);
predIntraChromaLM( compID, predBuf, pu, area, mode, createNewModel, &cclmModelStored );
int64_t sad = distParamSad.distFunc(distParamSad) * 2;
int64_t satd = distParamSatd.distFunc(distParamSatd);
int64_t sadThis = std::min(sad, satd);
if ( sadThis < sadBest )
{
sadBest = sadThis;
deltaBest = offset;
}
else
{
break;
}
}
}
}
#endif
#if JVET_AA0126_GLM
void IntraSearch::xFindBestGlmIdcSATD(PredictionUnit &pu, ComponentID compID, int &idcBest, int64_t &sadBest )
{
CodingStructure& cs = *(pu.cs);
CompArea area = compID == COMPONENT_Cb ? pu.Cb() : pu.Cr();
PelBuf orgBuf = cs.getOrgBuf(area);
PelBuf predBuf = cs.getPredBuf(area);
#if JVET_AB0092_GLM_WITH_LUMA
int maxIdc = NUM_GLM_PATTERN * NUM_GLM_WEIGHT;
#else
int mode = pu.intraDir[1];
DistParam distParamSad;
DistParam distParamSatd;
m_pcRdCost->setDistParam(distParamSad, orgBuf, predBuf, pu.cs->sps->getBitDepth(CHANNEL_TYPE_CHROMA), compID, false);
m_pcRdCost->setDistParam(distParamSatd, orgBuf, predBuf, pu.cs->sps->getBitDepth(CHANNEL_TYPE_CHROMA), compID, true);
distParamSad.applyWeight = false;
distParamSatd.applyWeight = false;
sadBest = -1;
#if JVET_AB0092_GLM_WITH_LUMA
CompArea areacr = pu.Cr();
PelBuf orgBufcr = cs.getOrgBuf(areacr);
PelBuf predBufcr = cs.getPredBuf(areacr);
DistParam distParamSadcr;
DistParam distParamSatdcr;
m_pcRdCost->setDistParam(distParamSadcr, orgBufcr, predBufcr, pu.cs->sps->getBitDepth(CHANNEL_TYPE_CHROMA), COMPONENT_Cr, false);
m_pcRdCost->setDistParam(distParamSatdcr, orgBufcr, predBufcr, pu.cs->sps->getBitDepth(CHANNEL_TYPE_CHROMA), COMPONENT_Cr, true);
distParamSadcr.applyWeight = false;
distParamSatdcr.applyWeight = false;
#endif
// Search positive idcs
for ( int idc = 0; idc <= maxIdc; idc++ )
{
pu.glmIdc.setIdc(compID, 0, idc);
pu.glmIdc.setIdc(compID, 1, idc);
predIntraChromaLM( compID, predBuf, pu, area, mode );
int64_t sad = distParamSad.distFunc(distParamSad) * 2;
int64_t satd = distParamSatd.distFunc(distParamSatd);
int64_t sadThis = std::min(sad, satd);
#if JVET_AB0092_GLM_WITH_LUMA
pu.glmIdc.setIdc(COMPONENT_Cr, 0, idc);
pu.glmIdc.setIdc(COMPONENT_Cr, 1, idc);
predIntraChromaLM(COMPONENT_Cr, predBufcr, pu, areacr, mode);
int64_t sadcr = distParamSadcr.distFunc(distParamSadcr) * 2;
int64_t satdcr = distParamSatdcr.distFunc(distParamSatdcr);
int64_t sadThiscr = std::min(sadcr, satdcr);
sadThis += sadThiscr;
#endif
if ( sadBest == -1 || sadThis < sadBest )
{
sadBest = sadThis;
idcBest = idc;
}
}
}
#endif
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)
Yung-Hsuan Chao (Jessie)
committed
{
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());
}
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);
Yin Zhao
committed
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;
memset(cu.reuseflag[COMPONENT_Y], false, sizeof(bool) * MAXPLTPREDSIZE);
compBeginTmp = COMPONENT_Y;
numCompTmp = (cu.chromaFormat != CHROMA_400) ? 3 : 1;
}
for (int curIdx = 0; curIdx < cu.curPLTSize[compBegin]; curIdx++)
{
if (idxExist[curIdx])
{
for (int comp = compBeginTmp; comp < (compBeginTmp + numCompTmp); comp++)
3663
3664
3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
3677
3678
3679
3680
3681
3682
3683
3684
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;
cu.reuseflag[COMPONENT_Y][reuseFlagIdx] = true;
reuseEntrySize++;
}
}
curPLTtmpIdx++;
}
}
cu.reusePLTSize[compBegin] = reuseEntrySize;
// update palette table
cu.curPLTSize[compBegin] = newPLTSize;
cu.curPLTSize[COMPONENT_Y] = newPLTSize;
for (int comp = compBeginTmp; comp < (compBeginTmp + numCompTmp); comp++)
memcpy( cu.curPLT[comp], curPLTtmp[comp], sizeof(Pel)*cu.curPLTSize[compBegin]);
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);
Yung-Hsuan Chao (Jessie)
committed
#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_tmpStorageLCU.getBuf(tmpArea1);
distortion += m_pcRdCost->getDistPart(org, tmpRecLuma, cs.sps->getBitDepth(toChannelType(compID)), compID, DF_SSE_WTD, &orgLuma);
}
else
{
distortion += m_pcRdCost->getDistPart(org, reco, cs.sps->getBitDepth(toChannelType(compID)), compID, DF_SSE_WTD, &orgLuma);
}
}
else
Yung-Hsuan Chao (Jessie)
committed
#endif
distortion += m_pcRdCost->getDistPart(org, reco, cs.sps->getBitDepth(toChannelType(compID)), compID, DF_SSE);
Yung-Hsuan Chao (Jessie)
committed
cs.dist += distortion;
const CompArea &area = cu.blocks[compBegin];
cs.setDecomp(area);
#if JVET_Z0118_GDR
cs.updateReconMotIPM(area);
#else
cs.picture->getRecoBuf(area).copyFrom(cs.getRecoBuf(area));
Yung-Hsuan Chao (Jessie)
committed
}
3796
3797
3798
3799
3800
3801
3802
3803
3804
3805
3806
3807
3808
3809
3810
3811
3812
3813
3814
3815
3816
3817
3818
3819
3820
3821
3822
3823
3824
3825
3826
3827
3828
3829
3830
3831
3832
3833
3834
3835
3836
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<int>(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<int>(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()))
3844
3845
3846
3847
3848
3849
3850
3851
3852
3853
3854
3855
3856
3857
3858
3859
3860
3861
3862
3863
3864
3865
3866
3867
3868
3869
3870
3871
3872
3873
{
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 += m_escapeNumBins[curPel[comp]];
}
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 += m_escapeNumBins[paPixelValue[comp]]; // 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& dMinCost, bool* idxExist)
3958
3959
3960
3961
3962
3963
3964
3965
3966
3967
3968
3969
3970
3971
3972
3973
3974
3975
3976
3977
3978
3979
3980
3981
3982
3983
3984
3985
3986
3987
3988
3989
3990
3991
3992
3993
3994
3995
3996
3997
3998
3999
4000
{
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(compBegin);
m_scanOrder = g_scanOrder[SCAN_UNGROUPED][pltScanMode ? SCAN_TRAV_VER : SCAN_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);