From c0f0d7b6993e9be41d18e289538e85a985f016e2 Mon Sep 17 00:00:00 2001
From: Ruoyang Yu <ruoyyu@qti.qualcomm.com>
Date: Fri, 15 Nov 2024 17:37:12 +0000
Subject: [PATCH] JVET-AJ0237: 12-bit internal bit depth modifications for ECM
---
source/App/EncoderApp/EncAppCfg.cpp | 7 +
source/Lib/CommonLib/AdaptiveLoopFilter.cpp | 81 +++++++
source/Lib/CommonLib/AdaptiveLoopFilter.h | 8 +
source/Lib/CommonLib/AlfParameters.h | 15 +-
source/Lib/CommonLib/BilateralFilter.cpp | 107 ++++++++-
source/Lib/CommonLib/BilateralFilter.h | 17 ++
source/Lib/CommonLib/CommonDef.h | 11 +
source/Lib/CommonLib/InterPrediction.cpp | 107 +++++++++
source/Lib/CommonLib/InterPrediction.h | 12 +
source/Lib/CommonLib/InterpolationFilter.cpp | 28 +++
source/Lib/CommonLib/InterpolationFilter.h | 4 +
source/Lib/CommonLib/IntraPrediction.cpp | 222 +++++++++++++++++-
source/Lib/CommonLib/IntraPrediction.h | 28 +++
source/Lib/CommonLib/Picture.cpp | 3 +
source/Lib/CommonLib/Rom.cpp | 12 +
source/Lib/CommonLib/Rom.h | 12 +
source/Lib/CommonLib/SampleAdaptiveOffset.cpp | 173 ++++++++++++++
source/Lib/CommonLib/Slice.h | 14 ++
source/Lib/CommonLib/TrQuant.cpp | 60 +++++
source/Lib/CommonLib/TypeDef.h | 1 +
.../Lib/CommonLib/x86/AdaptiveLoopFilterX86.h | 152 +++++++++++-
source/Lib/CommonLib/x86/BilateralFilterX86.h | 107 +++++++++
source/Lib/CommonLib/x86/BufferX86.h | 4 +
.../CommonLib/x86/InterpolationFilterX86.h | 160 +++++++++++++
source/Lib/CommonLib/x86/IntraX86.h | 56 +++++
source/Lib/CommonLib/x86/RdCostX86.h | 40 ++++
source/Lib/DecoderLib/DecLib.cpp | 23 ++
source/Lib/DecoderLib/VLCReader.cpp | 15 ++
.../Lib/EncoderLib/EncAdaptiveLoopFilter.cpp | 12 +
source/Lib/EncoderLib/EncCu.cpp | 3 +
source/Lib/EncoderLib/EncGOP.cpp | 27 +++
source/Lib/EncoderLib/EncGOP.h | 8 +
source/Lib/EncoderLib/EncLib.cpp | 9 +
.../EncoderLib/EncSampleAdaptiveOffset.cpp | 128 ++++++++++
.../Lib/EncoderLib/EncSampleAdaptiveOffset.h | 9 +
source/Lib/EncoderLib/EncTemporalFilter.cpp | 67 ++++++
source/Lib/EncoderLib/EncTemporalFilter.h | 16 ++
source/Lib/EncoderLib/InterSearch.cpp | 8 +
source/Lib/EncoderLib/IntraSearch.cpp | 24 ++
source/Lib/EncoderLib/IntraSearch.h | 4 +
source/Lib/EncoderLib/VLCWriter.cpp | 11 +
41 files changed, 1797 insertions(+), 8 deletions(-)
diff --git a/source/App/EncoderApp/EncAppCfg.cpp b/source/App/EncoderApp/EncAppCfg.cpp
index 75c962989..64ff0e13e 100644
--- a/source/App/EncoderApp/EncAppCfg.cpp
+++ b/source/App/EncoderApp/EncAppCfg.cpp
@@ -3588,6 +3588,13 @@ bool EncAppCfg::parseCfg( int argc, char* argv[] )
}
#endif
+#if JVET_AJ0237_INTERNAL_12BIT
+ if ((m_internalBitDepth[CHANNEL_TYPE_LUMA] > 10) && m_CCSAO && (m_iQP >= 37) && (m_sourceWidth * m_sourceHeight > 1920 * 1080))
+ {
+ m_CCSAO = false;
+ }
+#endif
+
// check validity of input parameters
if( xCheckParameter() )
{
diff --git a/source/Lib/CommonLib/AdaptiveLoopFilter.cpp b/source/Lib/CommonLib/AdaptiveLoopFilter.cpp
index ceaa4d3d1..875310240 100644
--- a/source/Lib/CommonLib/AdaptiveLoopFilter.cpp
+++ b/source/Lib/CommonLib/AdaptiveLoopFilter.cpp
@@ -2544,7 +2544,11 @@ void AdaptiveLoopFilter::alfAddCorrect( AlfClassifier** classifier, const PelUn
int dstStride2 = dstStride * clsSizeY;
int srcStride2 = srcStride * clsSizeY;
+#if JVET_AJ0237_INTERNAL_12BIT
+ const Pel currBase = 1 << (clpRng.bd - 1);
+#else
const Pel currBase = 512;
+#endif
const int adjustOffCorr = (1 << (shiftCorr + shiftPrecis - 1));
const int offsetN = adjustOffCorr;
const int offsetP = offsetN - 1;
@@ -3182,7 +3186,11 @@ double AdaptiveLoopFilter::getScaleCorrDouble( const int s )
#if JVET_AG0157_ALF_CHROMA_FIXED_FILTER
void AdaptiveLoopFilter::deriveFixFilterResultsBlkChroma(AlfClassifier ***classifier, Pel ***fixedFilterResults, const CPelBuf &src, const CPelBuf &srcBeforeDb, const Area &blkDst, const Area &blk, const int bits, CodingStructure& cs, const ClpRng &clpRng, const Pel clippingValues[4], int qp, int fixedFilterSetIdx, int mappingDir[NUM_DIR_FIX][NUM_DIR_FIX], uint32_t **laplacian[NUM_DIRECTIONS])
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ m_deriveVariance(src, blkDst, blk, laplacian, bits);
+#else
m_deriveVariance(src, blkDst, blk, laplacian);
+#endif
m_deriveClassificationLaplacian(src, blkDst, blk, laplacian, ALF_CLASSIFIER_FL_CHROMA);
m_calcClass0(classifier[ALF_NUM_CLASSIFIER + 1], blkDst, blk, ALF_CLASSIFIER_FL_CHROMA + 10, 1, NUM_DIR_FIX, NUM_ACT_FIX, bits, 2, mappingDir, laplacian);
@@ -5054,8 +5062,12 @@ void AdaptiveLoopFilter::deriveClassificationAndFixFilterResultsBlk( AlfClassifi
#endif
{
#if JVET_AE0139_ALF_IMPROVED_FIXFILTER
+#if JVET_AJ0237_INTERNAL_12BIT
+ m_deriveVariance(srcLuma, blkDst, blk, laplacian, bits);
+#else
m_deriveVariance(srcLuma, blkDst, blk, laplacian);
#endif
+#endif
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
#if JVET_AG0157_ALF_CHROMA_FIXED_FILTER
m_deriveClassificationLaplacian(srcLuma, blkDst, blk, laplacian, ALF_CLASSIFIER_FL);
@@ -5622,7 +5634,11 @@ void AdaptiveLoopFilter::filterBlk(AlfClassifier **classifier, const PelUnitBuf
adjustShift -= shiftPrecis; // add more precision
}
const int shift = adjustShift;
+#if JVET_AJ0237_INTERNAL_12BIT
+ const Pel currBase = 1 << (clpRng.bd - 1);
+#else
const Pel currBase = 512; // 10-bits
+#endif
#else
#if JVET_AG0158_ALF_LUMA_COEFF_PRECISION
const int shift = coeffBits - 1;
@@ -7055,8 +7071,16 @@ void AdaptiveLoopFilter::paddingFixedFilterResultsCtu(Pel*** fixedFilterResultsP
}
#if JVET_AE0139_ALF_IMPROVED_FIXFILTER
+#if JVET_AJ0237_INTERNAL_12BIT
+void AdaptiveLoopFilter::deriveVariance(const CPelBuf& srcLuma, const Area& blkDst, const Area& blk, uint32_t*** laplacian, int bits)
+#else
void AdaptiveLoopFilter::deriveVariance(const CPelBuf &srcLuma, const Area &blkDst, const Area &blk, uint32_t ***laplacian)
+#endif
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ int64_t tempData[4][(m_CLASSIFICATION_BLK_SIZE + 10) >> 1][((m_CLASSIFICATION_BLK_SIZE + 16) >> 1) + 8] = { { { 0 } } };
+#endif
+
int fl = DIST_CLASS;
int stride = srcLuma.stride;
int stride2 = 2 * stride;
@@ -7072,12 +7096,43 @@ void AdaptiveLoopFilter::deriveVariance(const CPelBuf &srcLuma, const Area &blkD
for (int j = 0; j < blk.width + fl2; j += 2)
{
int jOffset = j >> 1;
+
+#if JVET_AJ0237_INTERNAL_12BIT
+ tempData[0][iOffset][jOffset] = src[j] + src[j + 1] + src1[j] + src1[j + 1];
+ tempData[1][iOffset][jOffset] = src[j] * src[j] + src[j + 1] * src[j + 1] + src1[j] * src1[j] + src1[j + 1] * src1[j + 1];
+#else
laplacian[0][iOffset][jOffset] = src[j] + src[j + 1] + src1[j] + src1[j + 1];
laplacian[1][iOffset][jOffset] = src[j] * src[j] + src[j + 1] * src[j + 1] + src1[j] * src1[j] + src1[j + 1] * src1[j + 1];
+#endif
int iOffsetM4 = iOffset - 4;
int jOffsetM4 = jOffset - 4;
+#if JVET_AJ0237_INTERNAL_12BIT
+ if (jOffsetM4 == 0)
+ {
+ tempData[2][iOffset][jOffsetM4] = tempData[0][iOffset][jOffset - 4] + tempData[0][iOffset][jOffset - 3] + tempData[0][iOffset][jOffset - 2] + tempData[0][iOffset][jOffset - 1] + tempData[0][iOffset][jOffset];
+ tempData[3][iOffset][jOffsetM4] = tempData[1][iOffset][jOffset - 4] + tempData[1][iOffset][jOffset - 3] + tempData[1][iOffset][jOffset - 2] + tempData[1][iOffset][jOffset - 1] + tempData[1][iOffset][jOffset];
+ }
+ else if (jOffsetM4 > 0)
+ {
+ tempData[2][iOffset][jOffsetM4] = tempData[2][iOffset][jOffset - 5] - tempData[0][iOffset][jOffset - 5] + tempData[0][iOffset][jOffset];
+ tempData[3][iOffset][jOffsetM4] = tempData[3][iOffset][jOffset - 5] - tempData[1][iOffset][jOffset - 5] + tempData[1][iOffset][jOffset];
+ }
+
+ if ((iOffsetM4 >= 0) && (jOffsetM4 >= 0))
+ {
+ if (iOffsetM4 == 0)
+ {
+ tempData[0][iOffsetM4][jOffsetM4] = tempData[2][iOffsetM4][jOffsetM4] + tempData[2][iOffset - 3][jOffsetM4] + tempData[2][iOffset - 2][jOffsetM4] + tempData[2][iOffset - 1][jOffsetM4] + tempData[2][iOffset][jOffsetM4];
+ tempData[1][iOffsetM4][jOffsetM4] = tempData[3][iOffsetM4][jOffsetM4] + tempData[3][iOffset - 3][jOffsetM4] + tempData[3][iOffset - 2][jOffsetM4] + tempData[3][iOffset - 1][jOffsetM4] + tempData[3][iOffset][jOffsetM4];
+ }
+ else
+ {
+ tempData[0][iOffsetM4][jOffsetM4] = tempData[0][iOffsetM4 - 1][jOffsetM4] - tempData[2][iOffsetM4 - 1][jOffsetM4] + tempData[2][iOffset][jOffsetM4];
+ tempData[1][iOffsetM4][jOffsetM4] = tempData[1][iOffsetM4 - 1][jOffsetM4] - tempData[3][iOffsetM4 - 1][jOffsetM4] + tempData[3][iOffset][jOffsetM4];
+ }
+#else
if (jOffsetM4 == 0)
{
laplacian[2][iOffset][jOffsetM4] = laplacian[0][iOffset][jOffset - 4] + laplacian[0][iOffset][jOffset - 3] + laplacian[0][iOffset][jOffset - 2] + laplacian[0][iOffset][jOffset - 1] + laplacian[0][iOffset][jOffset];
@@ -7101,8 +7156,14 @@ void AdaptiveLoopFilter::deriveVariance(const CPelBuf &srcLuma, const Area &blkD
laplacian[0][iOffsetM4][jOffsetM4] = laplacian[0][iOffsetM4 - 1][jOffsetM4] - laplacian[2][iOffsetM4 - 1][jOffsetM4] + laplacian[2][iOffset][jOffsetM4];
laplacian[1][iOffsetM4][jOffsetM4] = laplacian[1][iOffsetM4 - 1][jOffsetM4] - laplacian[3][iOffsetM4 - 1][jOffsetM4] + laplacian[3][iOffset][jOffsetM4];
}
+#endif
+#if JVET_AJ0237_INTERNAL_12BIT
+ int bdShift = 2 * std::max(0, bits - 10);
+ laplacian[VARIANCE][iOffsetM4][jOffsetM4] = (uint32_t)((13 * ((numSample * tempData[1][iOffsetM4][jOffsetM4] - tempData[0][iOffsetM4][jOffsetM4] * tempData[0][iOffsetM4][jOffsetM4] + offset) >> 3)) >> (14 + bdShift));
+#else
laplacian[VARIANCE][iOffsetM4][jOffsetM4] = (13 * ((numSample * laplacian[1][iOffsetM4][jOffsetM4] - laplacian[0][iOffsetM4][jOffsetM4] * laplacian[0][iOffsetM4][jOffsetM4] + offset) >> 3)) >> 14;
+#endif
}
}
src += stride2;
@@ -7184,7 +7245,11 @@ void AdaptiveLoopFilter::deriveFixedFilterResultsPerBlkChroma(AlfClassifier ***c
if (useSimd)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ m_deriveVariance(src, blk, blk, laplacian, bits);
+#else
m_deriveVariance(src, blk, blk, laplacian);
+#endif
m_deriveClassificationLaplacian(src, blk, blk, laplacian, ALF_CLASSIFIER_FL_CHROMA);
m_calcClass0(classifier[ALF_NUM_CLASSIFIER + 1], blk, blk, ALF_CLASSIFIER_FL_CHROMA + 10, 1, NUM_DIR_FIX, NUM_ACT_FIX, bits, 2, mappingDir, laplacian);
alfFixedFilterBlk(classifier[ALF_NUM_CLASSIFIER + 1], src, blk, blk, srcBeforeDb, fixedFilterResults, m_picWidth, fixedFilterSetIdx, targetFixedFilterSetInd, 0, clpRng, clippingValues, false
@@ -7195,7 +7260,11 @@ void AdaptiveLoopFilter::deriveFixedFilterResultsPerBlkChroma(AlfClassifier ***c
}
else
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ deriveVariance(src, blk, blk, laplacian, bits);
+#else
deriveVariance(src, blk, blk, laplacian);
+#endif
deriveClassificationLaplacian(src, blk, blk, laplacian, ALF_CLASSIFIER_FL_CHROMA);
calcClass0Var(classifier[ALF_NUM_CLASSIFIER + 1], blk, blk, ALF_CLASSIFIER_FL_CHROMA + 10, 1, NUM_DIR_FIX, NUM_ACT_FIX, bits, 2, mappingDir, laplacian);
alfFixedFilterBlkNonSimd(classifier[ALF_NUM_CLASSIFIER + 1], src, blk, blk, srcBeforeDb, fixedFilterResults, m_picWidth, fixedFilterSetIdx, targetFixedFilterSetInd, 0, clpRng, clippingValues, false
@@ -7704,8 +7773,12 @@ void AdaptiveLoopFilter::deriveFixedFilterResultsPerBlk( AlfClassifier **classif
if(useSimd)
{
#if JVET_AE0139_ALF_IMPROVED_FIXFILTER
+#if JVET_AJ0237_INTERNAL_12BIT
+ m_deriveVariance(srcLuma, blkCur, blkCur, laplacian, bits);
+#else
m_deriveVariance(srcLuma, blkCur, blkCur, laplacian);
#endif
+#endif
#if JVET_AG0157_ALF_CHROMA_FIXED_FILTER
m_deriveClassificationLaplacian(srcLuma, blkCur, blkCur, laplacian, ALF_CLASSIFIER_FL);
#else
@@ -7715,8 +7788,12 @@ void AdaptiveLoopFilter::deriveFixedFilterResultsPerBlk( AlfClassifier **classif
else
{
#if JVET_AE0139_ALF_IMPROVED_FIXFILTER
+#if JVET_AJ0237_INTERNAL_12BIT
+ deriveVariance(srcLuma, blkCur, blkCur, laplacian, bits);
+#else
deriveVariance(srcLuma, blkCur, blkCur, laplacian);
#endif
+#endif
#if JVET_AG0157_ALF_CHROMA_FIXED_FILTER
deriveClassificationLaplacian(srcLuma, blkCur, blkCur, laplacian, ALF_CLASSIFIER_FL);
#else
@@ -8126,7 +8203,11 @@ void AdaptiveLoopFilter::gaussFiltering(CodingStructure &cs, Pel ***gaussPic, co
int padSize = ALF_PADDING_SIZE_GAUSS_RESULTS;
int shift = 10;
const int numCoeff = 12;
+#if JVET_AJ0237_INTERNAL_12BIT
+ int diffTH = 32 << std::max(0, cs.sps->getBitDepth(CHANNEL_TYPE_LUMA) - 10);
+#else
int diffTH = 32;
+#endif
#if JVET_AJ0188_CODING_INFO_CLASSIFICATION
const bool isIntraSlice = cs.slice->isIntra();
const bool isSpsAdjust = cs.sps->getAlfLumaFixedFilterAdjust();
diff --git a/source/Lib/CommonLib/AdaptiveLoopFilter.h b/source/Lib/CommonLib/AdaptiveLoopFilter.h
index d2dacb490..6a31efba7 100644
--- a/source/Lib/CommonLib/AdaptiveLoopFilter.h
+++ b/source/Lib/CommonLib/AdaptiveLoopFilter.h
@@ -180,14 +180,22 @@ public:
void deriveFixedFilterResultsBlk( AlfClassifier*** classifier, const CPelBuf& srcLuma, const CPelBuf& srcLumaBeforeDb, const Area& blkDst, const Area& blk, const int bits, CodingStructure &cs, const ClpRng &clpRng, const Pel clippingValues[4], int qp, int winIdx, int fixedFilterSetIdx );
void deriveFixedFilterResults( AlfClassifier*** classifier, const CPelBuf& srcLuma, const CPelBuf& srcLumaBeforeDb, const Area& blkDst, const Area& blk, CodingStructure &cs, int winIdx, int fixedFilterSetIdx );
static void calcClass0Var( AlfClassifier **classifier, const Area &blkDst, const Area &cu, int dirWindSize, int classDir, int noDir, int noAct, int bitDepth, int subBlkSize, int mappingDir[NUM_DIR_FIX][NUM_DIR_FIX], uint32_t **laplacian[NUM_DIRECTIONS] );
+#if JVET_AJ0237_INTERNAL_12BIT
+ static void deriveVariance(const CPelBuf& srcLuma, const Area& blkDst, const Area& blk, uint32_t ***laplacian, int bd);
+#else
static void deriveVariance( const CPelBuf &srcLuma, const Area &blkDst, const Area &blk, uint32_t ***laplacian );
+#endif
void deriveFixedFilterResultsCtuBoundary( AlfClassifier ***classifier, Pel ***fixedFilterResults, const CPelBuf &srcLuma, const CPelBuf &srcLumaBeforeDb, const Area &blkDst, const int bits, CodingStructure& cs, const ClpRng &clpRng, const Pel clippingValues[4], int qp, int fixedFilterSetIdx, int mappingDir[NUM_DIR_FIX][NUM_DIR_FIX], uint32_t **laplacian[NUM_DIRECTIONS], uint8_t* ctuEnableFlagLuma, uint8_t* ctuEnableOnlineLuma, int ctuIdx, int classifierIdx
#if JVET_AJ0188_CODING_INFO_CLASSIFICATION
, const CPelBuf& srcCodingInfo, const CPelBuf& srcResi
#endif
);
void deriveFixedFilterResultsPerBlk( AlfClassifier ***classifier, Pel ***fixedFilterResults, const CPelBuf &srcLuma, const CPelBuf &srcLumaBeforeDb, const Area &blkCur, const int bits, CodingStructure& cs, const ClpRng &clpRng, const Pel clippingValues[4], int qp, int fixedFilterSetIdx, int mappingDir[NUM_DIR_FIX][NUM_DIR_FIX], uint32_t **laplacian[NUM_DIRECTIONS], const int classifierIdx );
+#if JVET_AJ0237_INTERNAL_12BIT
+ void(*m_deriveVariance)(const CPelBuf& srcLuma, const Area& blkDst, const Area& blk, uint32_t ***variance, int bd);
+#else
void(*m_deriveVariance)(const CPelBuf &srcLuma, const Area &blkDst, const Area &blk, uint32_t ***variance);
+#endif
#if JVET_AG0157_ALF_CHROMA_FIXED_FILTER
void deriveFixedFilterResultsCtuBoundaryChroma(AlfClassifier ***classifier, Pel ***fixedFilterResults, const CPelBuf &src, const CPelBuf &srcBeforeDb, const Area &blkDst, const int bits, CodingStructure& cs, const ClpRng &clpRng, const Pel clippingValues[4], int qp, int fixedFilterSetIdx, int mappingDir[NUM_DIR_FIX][NUM_DIR_FIX], uint32_t **laplacian[NUM_DIRECTIONS], uint8_t* ctuEnableFlag, int ctuIdx);
void deriveFixedFilterResultsPerBlkChroma(AlfClassifier ***classifier, Pel ***fixedFilterResults, const CPelBuf &src, const CPelBuf &srcBeforeDb, const Area &blk, const int bits, CodingStructure& cs, const ClpRng &clpRng, const Pel clippingValues[4], int qp, int fixedFilterSetIdx, int mappingDir[NUM_DIR_FIX][NUM_DIR_FIX], uint32_t **laplacian[NUM_DIRECTIONS]);
diff --git a/source/Lib/CommonLib/AlfParameters.h b/source/Lib/CommonLib/AlfParameters.h
index 0a507c46b..1712e1ba5 100644
--- a/source/Lib/CommonLib/AlfParameters.h
+++ b/source/Lib/CommonLib/AlfParameters.h
@@ -584,6 +584,9 @@ struct ScaleAlf
bool usePrev;
int apsIdx;
+#if JVET_AJ0237_INTERNAL_12BIT
+ int bitDepth;
+#endif
void reset()
{
@@ -608,14 +611,20 @@ struct ScaleAlf
void setMinMax( const Pel lumaMin = 0, const Pel lumaMax = 1024, const bool bCheckClassifier = true )
{
const int c = classifierIdx;
+#if !JVET_AJ0237_INTERNAL_12BIT
const int bitDepth = 10;
+#endif
idxClassMin = (!bCheckClassifier || c == 1) ? ((lumaMin * ALF_NUM_CLASSES_CLASSIFIER[c]) >> bitDepth) : 0 ;
idxClassMax = (!bCheckClassifier || c == 1) ? ((lumaMax * ALF_NUM_CLASSES_CLASSIFIER[c]) >> bitDepth) : (ALF_NUM_CLASSES_CLASSIFIER[c] - 1) ;
initMinMaxDone = true;
}
- void init( const int f, const int a, const int c )
+#if JVET_AJ0237_INTERNAL_12BIT
+ void init(const int f, const int a, const int c, const int bDepth)
+#else
+ void init( const int f, const int a, const int c )
+#endif
{
filterSetIndex = f;
alt_num = a;
@@ -625,7 +634,9 @@ struct ScaleAlf
idxClassMin = 0 ;
idxClassMax = ALF_NUM_CLASSES_CLASSIFIER[c] - 1 ;
-
+#if JVET_AJ0237_INTERNAL_12BIT
+ bitDepth = bDepth;
+#endif
initDone = true;
}
diff --git a/source/Lib/CommonLib/BilateralFilter.cpp b/source/Lib/CommonLib/BilateralFilter.cpp
index 7d3b615d8..0f6c0f4e3 100644
--- a/source/Lib/CommonLib/BilateralFilter.cpp
+++ b/source/Lib/CommonLib/BilateralFilter.cpp
@@ -62,6 +62,9 @@ BilateralFilter::BilateralFilter()
initBilateralFilterX86();
#endif
#endif
+#if JVET_AJ0237_INTERNAL_12BIT
+ internalBitDepth = 10;
+#endif
}
BilateralFilter::~BilateralFilter()
@@ -101,6 +104,12 @@ const char* BilateralFilter::getFilterLutParameters(int16_t* block, const int st
int h = floorLog2(height);
int mad = m_calcMAD(block, stride, width, height, w + h);
+#if JVET_AJ0237_INTERNAL_12BIT
+ int bdShift = std::max(0, internalBitDepth - 10);
+ int offset = (bdShift == 0) ? 0 : (1 << (bdShift - 1));
+ mad = (mad + offset) >> bdShift;
+#endif
+
w = std::min(w, 7);
h = std::min(h, 7);
mad = std::min(mad >> 4, 15);
@@ -163,7 +172,11 @@ const char* BilateralFilter::getFilterLutParameters(int16_t* block, const int st
}
#endif
+#if JVET_AJ0237_INTERNAL_12BIT
+inline void bifApplyLut(int diff, int& res, int cutBitsNum, int bitsRound, int bitsRound2, int shift, const char* lutRowPtr, int lutShift, int bdShift)
+#else
inline void bifApplyLut(int diff, int& res, int cutBitsNum, int bitsRound, int bitsRound2, int shift, const char* lutRowPtr, int lutShift)
+#endif
{
int sg0 = diff >> shift;
int v0 = (diff + sg0) ^ sg0;
@@ -177,14 +190,26 @@ inline void bifApplyLut(int diff, int& res, int cutBitsNum, int bitsRound, int b
int idx = (v0 + 4) >> 3;
idx = 15 + ((idx - 15) & ((idx - 15) >> shift));
int w0 = lutRowPtr[idx] >> lutShift;
+#endif
+#if JVET_AJ0237_INTERNAL_12BIT
+ w0 = (w0 << bdShift);
#endif
res = (w0 + sg0) ^ sg0;
}
+#if JVET_AJ0237_INTERNAL_12BIT
+void BilateralFilter::blockBilateralFilterDiamond5x5(uint32_t uiWidth, uint32_t uiHeight, int16_t block[], int16_t blkFilt[], const ClpRng& clpRng, Pel* recPtr, int recStride, int iWidthExtSIMD, int bfac, int bifRoundAdd, int bifRoundShift, bool isRDO, const char* lutRowPtr, bool noClip, int cutBitsNum, int bdShift)
+#else
void BilateralFilter::blockBilateralFilterDiamond5x5( uint32_t uiWidth, uint32_t uiHeight, int16_t block[], int16_t blkFilt[], const ClpRng& clpRng, Pel* recPtr, int recStride, int iWidthExtSIMD, int bfac, int bifRoundAdd, int bifRoundShift, bool isRDO, const char* lutRowPtr, bool noClip, int cutBitsNum)
+#endif
{
int pad = NUMBER_PADDED_SAMPLES;
+#if JVET_AJ0237_INTERNAL_12BIT
+ cutBitsNum += bdShift;
+#endif
+
+
int padwidth = iWidthExtSIMD;
int downbuffer[128];
int downleftbuffer[129];
@@ -204,13 +229,25 @@ void BilateralFilter::blockBilateralFilterDiamond5x5( uint32_t uiWidth, uint32_t
{
int pixel = block[(-1 + pad)*padwidth + x + pad];
int below = block[(-1 + pad + 1)*padwidth + x + pad];
+#if JVET_AJ0237_INTERNAL_12BIT
+ bifApplyLut(below - pixel, downbuffer[x], cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift1, bdShift);
+#else
bifApplyLut(below - pixel, downbuffer[x], cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift1);
+#endif
int belowright = block[(-1 + pad + 1)*padwidth + x + pad + 1];
+#if JVET_AJ0237_INTERNAL_12BIT
+ bifApplyLut(belowright - pixel, downrightbuffer[1][x + 1], cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift2, bdShift);
+#else
bifApplyLut(belowright - pixel, downrightbuffer[1][x + 1], cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift2);
+#endif
int belowleft = block[(-1 + pad + 1)*padwidth + x + pad - 1];
+#if JVET_AJ0237_INTERNAL_12BIT
+ bifApplyLut(belowleft - pixel, downleftbuffer[x], cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift2, bdShift);
+#else
bifApplyLut(belowleft - pixel, downleftbuffer[x], cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift2);
+#endif
}
int width = uiWidth;
for( int y = 0; y < uiHeight; y++ )
@@ -220,15 +257,27 @@ void BilateralFilter::blockBilateralFilterDiamond5x5( uint32_t uiWidth, uint32_t
int pixel = rowStart[-1];
int right = rowStart[0], rightmod = 0;
+#if JVET_AJ0237_INTERNAL_12BIT
+ bifApplyLut(right - pixel, rightmod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift1, bdShift);
+#else
bifApplyLut(right - pixel, rightmod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift1);
+#endif
pixel = rowStart[-padwidth - 1];
int belowright = right;
+#if JVET_AJ0237_INTERNAL_12BIT
+ bifApplyLut(belowright - pixel, downrightbuffer[(y + 1) % 2][0], cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift2, bdShift);
+#else
bifApplyLut(belowright - pixel, downrightbuffer[(y + 1) % 2][0], cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift2);
+#endif
pixel = rowStart[-padwidth + width];
int belowleft = rowStart[width - 1];
+#if JVET_AJ0237_INTERNAL_12BIT
+ bifApplyLut(belowleft - pixel, downleftbuffer[width], cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift2, bdShift);
+#else
bifApplyLut(belowleft - pixel, downleftbuffer[width], cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift2);
+#endif
for( int x = 0; x < uiWidth; x++ )
{
@@ -242,12 +291,20 @@ void BilateralFilter::blockBilateralFilterDiamond5x5( uint32_t uiWidth, uint32_t
modsum += leftmod;
right = rowStart[x + 1];
+#if JVET_AJ0237_INTERNAL_12BIT
+ bifApplyLut(right - pixel, mod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift1, bdShift);
+#else
bifApplyLut(right - pixel, mod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift1);
+#endif
modsum += mod;
rightmod = mod;
int below = rowStart[x + padwidth];
+#if JVET_AJ0237_INTERNAL_12BIT
+ bifApplyLut(below - pixel, mod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift1, bdShift);
+#else
bifApplyLut(below - pixel, mod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift1);
+#endif
modsum += mod;
downbuffer[x] = mod;
@@ -258,12 +315,20 @@ void BilateralFilter::blockBilateralFilterDiamond5x5( uint32_t uiWidth, uint32_t
modsum += aboveleftmod;
int belowleft = rowStart[x + padwidth - 1];
+#if JVET_AJ0237_INTERNAL_12BIT
+ bifApplyLut(belowleft - pixel, mod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift2, bdShift);
+#else
bifApplyLut(belowleft - pixel, mod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift2);
+#endif
modsum += mod;
downleftbuffer[x] = mod;
int belowright = rowStart[x + padwidth + 1];
+#if JVET_AJ0237_INTERNAL_12BIT
+ bifApplyLut(belowright - pixel, mod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift2, bdShift);
+#else
bifApplyLut(belowright - pixel, mod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift2);
+#endif
modsum += mod;
downrightbuffer[y % 2][x + 1] = mod;
@@ -272,23 +337,43 @@ void BilateralFilter::blockBilateralFilterDiamond5x5( uint32_t uiWidth, uint32_t
// speed when SIMD is turned off.
int above = rowStart[x - 2 * padwidth];
+#if JVET_AJ0237_INTERNAL_12BIT
+ bifApplyLut(above - pixel, mod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift3, bdShift);
+#else
bifApplyLut(above - pixel, mod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift3);
+#endif
modsum += mod;
below = rowStart[x + 2 * padwidth];
+#if JVET_AJ0237_INTERNAL_12BIT
+ bifApplyLut(below - pixel, mod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift3, bdShift);
+#else
bifApplyLut(below - pixel, mod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift3);
+#endif
modsum += mod;
int left = rowStart[x - 2];
+#if JVET_AJ0237_INTERNAL_12BIT
+ bifApplyLut(left - pixel, mod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift3, bdShift);
+#else
bifApplyLut(left - pixel, mod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift3);
+#endif
modsum += mod;
right = rowStart[x + 2];
+#if JVET_AJ0237_INTERNAL_12BIT
+ bifApplyLut(right - pixel, mod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift3, bdShift);
+#else
bifApplyLut(right - pixel, mod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift3);
+#endif
modsum += mod;
#if JVET_AF0112_BIF_DYNAMIC_SCALING
+#if JVET_AJ0237_INTERNAL_12BIT
+ blkFilt[(y + pad) * padwidth + x + pad] = ((int16_t)((modsum * bfac + (bifRoundAdd << 3)) >> (bifRoundShift + 3)));
+#else
blkFilt[(y + pad) * padwidth + x + pad] = ((int16_t)((uint16_t)((modsum * bfac + (bifRoundAdd << 3)) >> (bifRoundShift + 3))));
+#endif
#else
blkFilt[(y + pad) * padwidth + x + pad] = (( int16_t ) (( uint16_t ) ((modsum*bfac + bifRoundAdd) >> bifRoundShift)));
#endif
@@ -592,7 +677,12 @@ void BilateralFilter::bilateralFilterRDOdiamond5x5(const ComponentID compID, Pel
CHECK(doReshape, "Reshape domain is not used for chroma");
#endif
}
+#if JVET_AJ0237_INTERNAL_12BIT
+ int bdShift = std::max(0, internalBitDepth - 10);
+ m_bilateralFilterDiamond5x5(uiWidth, uiHeight, tempblock, tempblockFiltered, clpRng, piReco, uiRecStride, iWidthExtSIMD, bfac, bifRoundAdd, bifRoundShift, true, lutRowPtr, false, cutBitsNum, bdShift);
+#else
m_bilateralFilterDiamond5x5(uiWidth, uiHeight, tempblock, tempblockFiltered, clpRng, piReco, uiRecStride, iWidthExtSIMD, bfac, bifRoundAdd, bifRoundShift, true, lutRowPtr, false, cutBitsNum);
+#endif
if( !useReco )
{
@@ -619,6 +709,9 @@ void BilateralFilter::bilateralFilterDiamond5x5( const ComponentID compID, const
#endif
)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ int bdShift = std::max(0, internalBitDepth - 10);
+#endif
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
const int scaleX = getChannelTypeScaleX( toChannelType( compID ), currTU.cu->cs->pcv->chrFormat );
const int scaleY = getChannelTypeScaleY( toChannelType( compID ), currTU.cu->cs->pcv->chrFormat );
@@ -902,7 +995,11 @@ void BilateralFilter::bilateralFilterDiamond5x5( const ComponentID compID, const
int bifRoundAdd = BIF_ROUND_ADD >> currTU.cs->pps->getBIFStrength();
int bifRoundShift = BIF_ROUND_SHIFT - currTU.cs->pps->getBIFStrength();
+#if JVET_AJ0237_INTERNAL_12BIT
+ m_bilateralFilterDiamond5x5(uiWidth, uiHeight, tempblock, tempblockFiltered, clpRng, recPtr, recStride, iWidthExtSIMD, bfac, bifRoundAdd, bifRoundShift, false, lutRowPtr, false, cutBitsNum, bdShift);
+#else
m_bilateralFilterDiamond5x5(uiWidth, uiHeight, tempblock, tempblockFiltered, clpRng, recPtr, recStride, iWidthExtSIMD, bfac, bifRoundAdd, bifRoundShift, false, lutRowPtr, false, cutBitsNum);
+#endif
xStart = xEnd;
}
@@ -1168,8 +1265,11 @@ void BilateralFilter::bilateralFilterDiamond5x5( const ComponentID compID, const
int bifRoundAdd = BIF_ROUND_ADD >> currTU.cs->pps->getBIFStrength();
int bifRoundShift = BIF_ROUND_SHIFT - currTU.cs->pps->getBIFStrength();
-
+#if JVET_AJ0237_INTERNAL_12BIT
+ m_bilateralFilterDiamond5x5(uiWidth, uiHeight, tempblock, tempblockFiltered, clpRng, recPtr, recStride, iWidthExtSIMD, bfac, bifRoundAdd, bifRoundShift, false, lutRowPtr, noClip, cutBitsNum, bdShift);
+#else
m_bilateralFilterDiamond5x5(uiWidth, uiHeight, tempblock, tempblockFiltered, clpRng, recPtr, recStride, iWidthExtSIMD, bfac, bifRoundAdd, bifRoundShift, false, lutRowPtr, noClip, cutBitsNum);
+#endif
}
}
void BilateralFilter::clipNotBilaterallyFilteredBlocks(const ComponentID compID, const CPelUnitBuf& src, PelUnitBuf& rec, const ClpRng& clpRng, TransformUnit & currTU)
@@ -1595,6 +1695,11 @@ const char* BilateralFilter::getFilterLutParametersChroma(int16_t* block, const
int h = floorLog2(heightForStrength);
int mad = m_calcMAD(block, stride, width, height, floorLog2(width) + floorLog2(height));
+#if JVET_AJ0237_INTERNAL_12BIT
+ int bdShift = std::max(0, internalBitDepth - 10);
+ int offset = (bdShift == 0) ? 0 : (1 << (bdShift - 1));
+ mad = (mad + offset) >> bdShift;
+#endif
w = std::min(w, 7);
h = std::min(h, 7);
diff --git a/source/Lib/CommonLib/BilateralFilter.h b/source/Lib/CommonLib/BilateralFilter.h
index 27cece33b..3cba3eefc 100644
--- a/source/Lib/CommonLib/BilateralFilter.h
+++ b/source/Lib/CommonLib/BilateralFilter.h
@@ -67,8 +67,13 @@ private:
Pel *tempblock = (Pel*)tempblockSIMD;
Pel* tempblockFiltered = (Pel*)tempblockFilteredSIMD;
+#if JVET_AJ0237_INTERNAL_12BIT
+ void (*m_bilateralFilterDiamond5x5)(uint32_t uiWidth, uint32_t uiHeight, int16_t block[], int16_t blkFilt[], const ClpRng& clpRng, Pel* recPtr, int recStride, int iWidthExtSIMD, int bfac, int bifRoundAdd, int bifRoundShift, bool isRDO, const char* lutRowPtr, bool noClip, int cutBitsNum, int bdShift);
+ static void blockBilateralFilterDiamond5x5(uint32_t uiWidth, uint32_t uiHeight, int16_t block[], int16_t blkFilt[], const ClpRng& clpRng, Pel* recPtr, int recStride, int iWidthExtSIMD, int bfac, int bifRoundAdd, int bifRoundShift, bool isRDO, const char* lutRowPtr, bool noClip, int cutBitsNum, int bdShift);
+#else
void (*m_bilateralFilterDiamond5x5)(uint32_t uiWidth, uint32_t uiHeight, int16_t block[], int16_t blkFilt[], const ClpRng& clpRng, Pel* recPtr, int recStride, int iWidthExtSIMD, int bfac, int bifRoundAdd, int bifRoundShift, bool isRDO, const char* lutRowPtr, bool noClip, int cutBitsNum);
static void blockBilateralFilterDiamond5x5(uint32_t uiWidth, uint32_t uiHeight, int16_t block[], int16_t blkFilt[], const ClpRng& clpRng, Pel* recPtr, int recStride, int iWidthExtSIMD, int bfac, int bifRoundAdd, int bifRoundShift, bool isRDO, const char* lutRowPtr, bool noClip, int cutBitsNum);
+#endif
#if JVET_AF0112_BIF_DYNAMIC_SCALING
int (*m_calcMAD)(int16_t* block, int stride, int width, int height, int whlog2);
@@ -213,12 +218,19 @@ private:
{ 0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1, },
};
#endif
+
+#if JVET_AJ0237_INTERNAL_12BIT
+ int internalBitDepth;
+#endif
public:
BilateralFilter();
~BilateralFilter();
void create();
void destroy();
+#if JVET_AJ0237_INTERNAL_12BIT
+ void setInternalBitDepth(int bdDepth) { internalBitDepth = bdDepth; }
+#endif
#if JVET_V0094_BILATERAL_FILTER
void bilateralFilterRDOdiamond5x5(const ComponentID compID, PelBuf& resiBuf, const CPelBuf& predBuf, PelBuf& recoBuf, int32_t qp, const CPelBuf& recIPredBuf, const ClpRng& clpRng, TransformUnit & currTU, bool useReco, bool doReshape = false, std::vector<Pel>* pLUT = nullptr);
void bilateralFilterPicRDOperCTU(const ComponentID compID, CodingStructure& cs, PelUnitBuf& src,BIFCabacEst* bifCABACEstimator);
@@ -243,8 +255,13 @@ public:
#if ENABLE_SIMD_BILATERAL_FILTER || JVET_X0071_CHROMA_BILATERAL_FILTER_ENABLE_SIMD
#ifdef TARGET_SIMD_X86
+#if JVET_AJ0237_INTERNAL_12BIT
+ template<X86_VEXT vext>
+ static void simdFilterDiamond5x5(uint32_t uiWidth, uint32_t uiHeight, int16_t block[], int16_t blkFilt[], const ClpRng& clpRng, Pel* recPtr, int recStride, int iWidthExtSIMD, int bfac, int bifRoundAdd, int bifRoundShift, bool isRDO, const char* lutRowPtr, bool noClip, int cutBitsNum, int bdShift);
+#else
template<X86_VEXT vext>
static void simdFilterDiamond5x5(uint32_t uiWidth, uint32_t uiHeight, int16_t block[], int16_t blkFilt[], const ClpRng& clpRng, Pel* recPtr, int recStride, int iWidthExtSIMD, int bfac, int bifRoundAdd, int bifRoundShift, bool isRDO, const char* lutRowPtr, bool noClip, int cutBitsNum);
+#endif
#if JVET_AF0112_BIF_DYNAMIC_SCALING
template<X86_VEXT vext>
static int simdCalcMAD(int16_t* block, int stride, int width, int height, int whlog2);
diff --git a/source/Lib/CommonLib/CommonDef.h b/source/Lib/CommonLib/CommonDef.h
index 29b7eb3dd..9bebc2f78 100644
--- a/source/Lib/CommonLib/CommonDef.h
+++ b/source/Lib/CommonLib/CommonDef.h
@@ -1419,6 +1419,10 @@ static const int CCCM_MAX_REF_SAMPLES = 4 * ( 2 * CCCM_WINDOW_SIZE * ( 2 * M
#else
static const int CCCM_MAX_REF_SAMPLES = ( 2 * CCCM_WINDOW_SIZE * ( 2 * MAX_CU_SIZE + CCCM_WINDOW_SIZE ) );
#endif
+#if JVET_AJ0237_INTERNAL_12BIT
+static const int CCCM_MATRIX_BITS_HBD = 32;
+static const int CCCM_DECIM_BITS_HBD = 22;
+#endif
#if JVET_AB0174_CCCM_DIV_FREE
static const int CCCM_MATRIX_BITS = 22;
static const int CCCM_DECIM_BITS = 16;
@@ -1426,7 +1430,9 @@ static const int CCCM_DECIM_BITS = 16;
static const int CCCM_MATRIX_BITS = 28;
static const int CCCM_DECIM_BITS = 22;
#endif
+#if !JVET_AJ0237_INTERNAL_12BIT
static const int CCCM_DECIM_ROUND = ( 1 << (CCCM_DECIM_BITS - 1 ) );
+#endif
#if JVET_AB0143_CCCM_TS
#if MMLM
#if JVET_AC0054_GLCCCM
@@ -1531,6 +1537,11 @@ static const int MAX_DELTA_QP = 7; ///<
static const int MAX_TESTED_QPs = ( 1 + 1 + ( MAX_DELTA_QP << 1 ) ); ///< dqp=0 +- max_delta_qp + lossless mode
static const int COM16_C806_TRANS_PREC = 0;
+
+#if JVET_AJ0237_INTERNAL_12BIT
+#define DECIM_BITS(x) ( (x) > 10 ? CCCM_DECIM_BITS_HBD : CCCM_DECIM_BITS )
+#endif
+
#if IF_12TAP
#define NTAPS_LUMA(x) ( (x) == 0 ? 12 : 8 ) // 12-tap filter for index 0. 8-tap fitler for other indices.
#else
diff --git a/source/Lib/CommonLib/InterPrediction.cpp b/source/Lib/CommonLib/InterPrediction.cpp
index 70b847afd..806653b25 100644
--- a/source/Lib/CommonLib/InterPrediction.cpp
+++ b/source/Lib/CommonLib/InterPrediction.cpp
@@ -347,6 +347,9 @@ InterPrediction::InterPrediction()
#if JVET_AG0276_NLIC
m_skipDoLic = false;
#endif
+#if JVET_AJ0237_INTERNAL_12BIT
+ m_dmvrCostLambda = 1;
+#endif
}
InterPrediction::~InterPrediction()
@@ -570,10 +573,18 @@ void InterPrediction::destroy()
#if INTER_LIC || (TM_AMVP || TM_MRG || JVET_Z0084_IBC_TM) || JVET_W0090_ARMC_TM || JVET_Z0056_GPM_SPLIT_MODE_REORDERING || JVET_Z0061_TM_OBMC
#if JVET_Z0153_IBC_EXT_REF
#if JVET_AJ0172_IBC_ITMP_ALIGN_REF_AREA
+#if JVET_AJ0237_INTERNAL_12BIT
+void InterPrediction::init( RdCost* pcRdCost, ChromaFormat chromaFormatIDC, const int ctuSize, Reshape* reshape, const int picWidth, const int picHeight, const int bitDepth )
+#else
void InterPrediction::init( RdCost* pcRdCost, ChromaFormat chromaFormatIDC, const int ctuSize, Reshape* reshape, const int picWidth, const int picHeight )
+#endif
+#else
+#if JVET_AJ0237_INTERNAL_12BIT
+void InterPrediction::init( RdCost* pcRdCost, ChromaFormat chromaFormatIDC, const int ctuSize, Reshape* reshape, const int picWidth, const int bitDepth )
#else
void InterPrediction::init( RdCost* pcRdCost, ChromaFormat chromaFormatIDC, const int ctuSize, Reshape* reshape, const int picWidth )
#endif
+#endif
#else
void InterPrediction::init( RdCost* pcRdCost, ChromaFormat chromaFormatIDC, const int ctuSize, Reshape* reshape )
#endif
@@ -593,6 +604,10 @@ void InterPrediction::init( RdCost* pcRdCost, ChromaFormat chromaFormatIDC, cons
}
m_currChromaFormat = chromaFormatIDC;
+#if JVET_AJ0237_INTERNAL_12BIT
+ m_dmvrCostLambda = 1 << std::max(0, std::min(14, bitDepth) - 10); // 14 is the maximum possible DMVR internal precision value, 10 is the baseline
+#endif
+
if( m_acYuvPred[REF_PIC_LIST_0][COMPONENT_Y] == nullptr ) // check if first is null (in which case, nothing initialised yet)
{
#if JVET_AF0057
@@ -15486,7 +15501,11 @@ Distortion InterPrediction::deriveBcwBlending( PredictionUnit& pu, bool bUniDir[
blendModel.params[1] = (int)((bcwModel.params[1] + offsetA) >> shiftA);
blendModel.params[2] = (int)((bcwModel.params[2] + offsetA) >> shiftA);
+#if JVET_AJ0237_INTERNAL_12BIT
+ blendModel.shift = bcwModel.decimBits - shiftA - bcwBlendingLog2WeightBase;
+#else
blendModel.shift = CCCM_DECIM_BITS - shiftA - bcwBlendingLog2WeightBase;
+#endif
blendModel.offset = blendModel.shift ? (1 << (blendModel.shift - 1)) : 0;
if (blendModel.shift < 0)
{
@@ -27643,7 +27662,11 @@ bool InterPrediction::processBDMVRPU2Dir(PredictionUnit& pu, bool subPURefine[2]
#else
Distortion initCost = xBDMVRGetMatchingError(pu, mvInitial_PU, bUseMR, false);
#endif
+#if JVET_AJ0237_INTERNAL_12BIT
+ if (initCost < lumaArea * m_dmvrCostLambda)
+#else
if (initCost < lumaArea)
+#endif
{
subPURefine[0] = false;
subPURefine[1] = false;
@@ -27661,7 +27684,11 @@ bool InterPrediction::processBDMVRPU2Dir(PredictionUnit& pu, bool subPURefine[2]
#else
minCost = xBDMVRMvOneTemplateHPelSquareSearch<1>(mvFinal, initCost, pu, mvInitial_PU, 2, MV_FRACTIONAL_BITS_INTERNAL - 1, bUseMR, false);
#endif
+#if JVET_AJ0237_INTERNAL_12BIT
+ subPURefine[0] = minCost >= lumaArea * m_dmvrCostLambda;
+#else
subPURefine[0] = minCost >= lumaArea;
+#endif
finalMvDir[0] = mvFinal[0];
#if JVET_AA0093_REFINED_MOTION_FOR_ARMC
}
@@ -27678,7 +27705,11 @@ bool InterPrediction::processBDMVRPU2Dir(PredictionUnit& pu, bool subPURefine[2]
#else
minCost = xBDMVRMvOneTemplateHPelSquareSearch<2>(mvFinal, initCost, pu, mvInitial_PU, 2, MV_FRACTIONAL_BITS_INTERNAL - 1, bUseMR, false);
#endif
+#if JVET_AJ0237_INTERNAL_12BIT
+ subPURefine[1] = minCost >= lumaArea * m_dmvrCostLambda;
+#else
subPURefine[1] = minCost >= lumaArea;
+#endif
finalMvDir[1] = mvFinal[1];
#if JVET_AA0093_REFINED_MOTION_FOR_ARMC
}
@@ -27732,7 +27763,11 @@ void InterPrediction::processBDMVRSubPU(PredictionUnit& pu, bool subPURefine)
Mv mvFinal[2] = { pu.mv[0], pu.mv[1] };
Mv mvOffset;
+#if JVET_AJ0237_INTERNAL_12BIT
+ const Distortion earlyTerminateTh = dx * dy * m_dmvrCostLambda;
+#else
const Distortion earlyTerminateTh = dx * dy;
+#endif
const int adaptiveSearchRangeHor = (dx >> 1) < BDMVR_INTME_RANGE ? (dx >> 1) : BDMVR_INTME_RANGE;
const int adaptiveSearchRangeVer = (dy >> 1) < BDMVR_INTME_RANGE ? (dy >> 1) : BDMVR_INTME_RANGE;
const bool adaptRange = (adaptiveSearchRangeHor != BDMVR_INTME_RANGE || adaptiveSearchRangeVer != BDMVR_INTME_RANGE);
@@ -28154,7 +28189,11 @@ void InterPrediction::bmAdaptiveAffineIntSearch(const PredictionUnit &pu, Mv(&mv
}
else
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ bmCostShift = bitDepth > 8 ? 2 : 0;
+#else
bmCostShift = bitDepth > 8 ? bitDepth - 8 : 0;
+#endif
}
#else
bmCostShift = 0;
@@ -28585,7 +28624,11 @@ void InterPrediction::bmAffineIntSearch(const PredictionUnit &pu, Mv(&mvOffset)[
}
else
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ bmCostShift = bitDepth > 8 ? 2 : 0;
+#else
bmCostShift = bitDepth > 8 ? bitDepth - 8 : 0;
+#endif
}
#else
bmCostShift = 0;
@@ -28802,7 +28845,11 @@ void InterPrediction::xInitBilateralMatching(const int width, const int height,
}
else
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ m_bmCostShift = bitDepth > 8 ? 2 : 0;
+#else
m_bmCostShift = bitDepth > 8 ? bitDepth - 8 : 0;
+#endif
}
#else
m_bmCostShift = 0;
@@ -29395,7 +29442,11 @@ bool InterPrediction::processBDMVR4Affine(PredictionUnit& pu
minCost = xGetBilateralMatchingErrorAffine(pu, pu.mvAffi, true);
}
const int lumaArea = pu.lumaSize().area();
+#if JVET_AJ0237_INTERNAL_12BIT
+ const bool isTooSmallDist = minCost < lumaArea * m_dmvrCostLambda;
+#else
const bool isTooSmallDist = minCost < lumaArea;
+#endif
if (!isTooSmallDist)
{
minCost = xBDMVRMv6ParameterSearchAffine(minCost, pu);
@@ -30555,7 +30606,11 @@ bool InterPrediction::processBDMVR4AdaptiveAffine(PredictionUnit& pu, Mv(&mvAffi
#else
bmAdaptiveAffineIntSearch(pu, mvFinalPUL0, minCostL0, mvFinalPUL1, minCostL1);
#endif
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int lumaArea = pu.lumaSize().area() * m_dmvrCostLambda;
+#else
const int lumaArea = pu.lumaSize().area();
+#endif
// sub-pel search for L0
if (minCostL0 > lumaArea)
@@ -30680,7 +30735,11 @@ bool InterPrediction::processBDMVR4AdaptiveAffine(PredictionUnit& pu, Mv(&mvAffi
{
minCostL0 = xGetBilateralMatchingErrorAffine(pu, pu.mvAffi, true);
}
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int lumaArea = pu.lumaSize().area() * m_dmvrCostLambda;
+#else
const int lumaArea = pu.lumaSize().area();
+#endif
const bool isTooSmallDist = minCostL0 < lumaArea;
if (!isTooSmallDist)
{
@@ -30761,7 +30820,11 @@ bool InterPrediction::processBDMVR4AdaptiveAffine(PredictionUnit& pu, Mv(&mvAffi
{
minCostL1 = xGetBilateralMatchingErrorAffine(pu, pu.mvAffi, true);
}
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int lumaArea = pu.lumaSize().area() * m_dmvrCostLambda;
+#else
const int lumaArea = pu.lumaSize().area();
+#endif
const bool isTooSmallDist = minCostL1 < lumaArea;
if (!isTooSmallDist)
{
@@ -30848,7 +30911,11 @@ bool InterPrediction::processBDMVR4AdaptiveAffine(PredictionUnit& pu, Mv(&mvAffi
{
minCost = xGetBilateralMatchingErrorAffine(pu, pu.mvAffi, true);
}
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int lumaArea = pu.lumaSize().area() * m_dmvrCostLambda;
+#else
const int lumaArea = pu.lumaSize().area();
+#endif
const bool isTooSmallDist = minCost < lumaArea;
if (!isTooSmallDist)
{
@@ -30934,7 +31001,11 @@ bool InterPrediction::processBDMVR(PredictionUnit& pu)
#else
minCost = xBDMVRGetMatchingError(pu, mvInitial_PU, bUseMR, false);
#endif
+#if JVET_AJ0237_INTERNAL_12BIT
+ if (minCost >= lumaArea * m_dmvrCostLambda)
+#else
if (minCost >= lumaArea)
+#endif
{
#if JVET_AI0185_ADAPTIVE_COST_IN_MERGE_MODE
minCost = xBDMVRMvOneTemplateHPelSquareSearch<1>(mvFinal_PU, minCost, pu, mvInitial_PU, 2, MV_FRACTIONAL_BITS_INTERNAL - 1, bUseMR, useHadmard);
@@ -30950,7 +31021,11 @@ bool InterPrediction::processBDMVR(PredictionUnit& pu)
#else
minCost = xBDMVRGetMatchingError(pu, mvInitial_PU, bUseMR, false);
#endif
+#if JVET_AJ0237_INTERNAL_12BIT
+ if (minCost >= lumaArea * m_dmvrCostLambda)
+#else
if (minCost >= lumaArea)
+#endif
{
#if JVET_AI0185_ADAPTIVE_COST_IN_MERGE_MODE
minCost = xBDMVRMvOneTemplateHPelSquareSearch<2>(mvFinal_PU, minCost, pu, mvInitial_PU, 2, MV_FRACTIONAL_BITS_INTERNAL - 1, bUseMR, useHadmard);
@@ -31106,7 +31181,11 @@ bool InterPrediction::processBDMVR(PredictionUnit& pu)
minCost = xBDMVRMvSquareSearch( mvFinal_PU, minCost, pu, mvInitial_PU, 2, MV_FRACTIONAL_BITS_INTERNAL - 1, bUseMR, false );
#endif
+#if JVET_AJ0237_INTERNAL_12BIT
+ subPURefine = minCost >= (lumaArea * m_dmvrCostLambda);
+#else
subPURefine = minCost >= lumaArea;
+#endif
#if JVET_AG0067_DMVR_EXTENSIONS
pu.mv[REF_PIC_LIST_0] = (mvFinal_PU[0] - puOrgMv[0]).scaleMv(scale0) + puOrgMv[0];
pu.mv[REF_PIC_LIST_1] = (mvFinal_PU[1] - puOrgMv[1]).scaleMv(scale1) + puOrgMv[1];
@@ -31188,7 +31267,11 @@ bool InterPrediction::processBDMVR(PredictionUnit& pu)
Mv mvFinal[2] = { pu.mv[0], pu.mv[1] };
Mv mvOffset;
+#if JVET_AJ0237_INTERNAL_12BIT
+ const Distortion earlyTerminateTh = dx * dy * m_dmvrCostLambda;
+#else
const Distortion earlyTerminateTh = dx * dy;
+#endif
const int adaptiveSearchRangeHor = (dx >> 1) < BDMVR_INTME_RANGE ? (dx >> 1) : BDMVR_INTME_RANGE;
const int adaptiveSearchRangeVer = (dy >> 1) < BDMVR_INTME_RANGE ? (dy >> 1) : BDMVR_INTME_RANGE;
const bool adaptRange = (adaptiveSearchRangeHor != BDMVR_INTME_RANGE || adaptiveSearchRangeVer != BDMVR_INTME_RANGE);
@@ -32077,7 +32160,11 @@ Distortion InterPrediction::xBDMVRMvIntPelFullSearch(Mv&mvOffset, Distortion cur
}
else
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ int32_t precisionAdj = cDistParam.bitDepth > 8 ? 2 : 0;
+#else
int32_t precisionAdj = cDistParam.bitDepth > 8 ? cDistParam.bitDepth - 8 : 0;
+#endif
curBestCost = cDistParam.distFunc(cDistParam) >> precisionAdj;
}
#else
@@ -32124,7 +32211,11 @@ Distortion InterPrediction::xBDMVRMvIntPelFullSearch(Mv&mvOffset, Distortion cur
}
else
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ int32_t precisionAdj = cDistParam.bitDepth > 8 ? 2 : 0;
+#else
int32_t precisionAdj = cDistParam.bitDepth > 8 ? cDistParam.bitDepth - 8 : 0;
+#endif
m_sadEnlargeArrayBilMrg[searchOffsetIdx] = cDistParam.distFunc(cDistParam) >> precisionAdj;
}
#else
@@ -33093,7 +33184,11 @@ Distortion InterPrediction::xBDMVRGetMatchingError(const PredictionUnit& pu, con
}
else
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ int32_t precisionAdj = cDistParam.bitDepth > 8 ? 2 : 0;
+#else
int32_t precisionAdj = cDistParam.bitDepth > 8 ? cDistParam.bitDepth - 8 : 0;
+#endif
return cDistParam.distFunc(cDistParam) >> precisionAdj;
}
#else
@@ -33259,7 +33354,11 @@ Distortion InterPrediction::xBDMVRGetMatchingError(const PredictionUnit& pu, con
}
else
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ int32_t precisionAdj = cDistParam.bitDepth > 8 ? 2 : 0;
+#else
int32_t precisionAdj = cDistParam.bitDepth > 8 ? cDistParam.bitDepth - 8 : 0;
+#endif
return cDistParam.distFunc( cDistParam ) >> precisionAdj;
}
#else
@@ -41495,6 +41594,10 @@ std::vector<Mv> InterPrediction::deriveMVDFromMVSDIdxAffineSI(PredictionUnit& pu
{
const ComponentID ch = ComponentID(chan);
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int maxValue = (1 << slice.getSPS()->getBitDepth(toChannelType(ch))) - 1;
+#endif
+
Pel *piTxtBuff =
pPadBuffYUV->getBuf(blkUnitAreaBuff)
.bufs[ch]
@@ -41512,7 +41615,11 @@ std::vector<Mv> InterPrediction::deriveMVDFromMVSDIdxAffineSI(PredictionUnit& pu
piTxtBuff[idx] += CompDiff[chan];
piTxtBuff[idx] = (piTxtBuff[idx] < 0) ? 0 : piTxtBuff[idx];
+#if JVET_AJ0237_INTERNAL_12BIT
+ piTxtBuff[idx] = (piTxtBuff[idx] > maxValue) ? maxValue : piTxtBuff[idx];
+#else
piTxtBuff[idx] = (piTxtBuff[idx] > 1023) ? 1023 : piTxtBuff[idx];
+#endif
}
piTxtBuff += iStrideBuff;
piTmpBuff += iStrideTmp;
diff --git a/source/Lib/CommonLib/InterPrediction.h b/source/Lib/CommonLib/InterPrediction.h
index 9e7ba7f3e..f1d9664ef 100644
--- a/source/Lib/CommonLib/InterPrediction.h
+++ b/source/Lib/CommonLib/InterPrediction.h
@@ -310,6 +310,10 @@ protected:
PelStorage m_obmcPelStorage;
#endif
+#if JVET_AJ0237_INTERNAL_12BIT
+ uint8_t m_dmvrCostLambda;
+#endif
+
ChromaFormat m_currChromaFormat;
ComponentID m_maxCompIDToPred; ///< tells the predictor to only process the components up to (inklusive) this one - useful to skip chroma components during RD-search
@@ -687,10 +691,18 @@ public:
#if INTER_LIC || (TM_AMVP || TM_MRG || JVET_Z0084_IBC_TM) || JVET_W0090_ARMC_TM || JVET_Z0056_GPM_SPLIT_MODE_REORDERING || JVET_Z0061_TM_OBMC
#if JVET_Z0153_IBC_EXT_REF
#if JVET_AJ0172_IBC_ITMP_ALIGN_REF_AREA
+#if JVET_AJ0237_INTERNAL_12BIT
+ void init (RdCost* pcRdCost, ChromaFormat chromaFormatIDC, const int ctuSize, Reshape* reshape, const int picWidth, const int picHeight, const int bitDepth);
+#else
void init (RdCost* pcRdCost, ChromaFormat chromaFormatIDC, const int ctuSize, Reshape* reshape, const int picWidth, const int picHeight);
+#endif
+#else
+#if JVET_AJ0237_INTERNAL_12BIT
+ void init (RdCost* pcRdCost, ChromaFormat chromaFormatIDC, const int ctuSize, Reshape* reshape, const int picWidth, const int bitDepth);
#else
void init (RdCost* pcRdCost, ChromaFormat chromaFormatIDC, const int ctuSize, Reshape* reshape, const int picWidth);
#endif
+#endif
#else
void init (RdCost* pcRdCost, ChromaFormat chromaFormatIDC, const int ctuSize, Reshape* reshape);
#endif
diff --git a/source/Lib/CommonLib/InterpolationFilter.cpp b/source/Lib/CommonLib/InterpolationFilter.cpp
index 09be2f148..914cceb7d 100644
--- a/source/Lib/CommonLib/InterpolationFilter.cpp
+++ b/source/Lib/CommonLib/InterpolationFilter.cpp
@@ -1596,9 +1596,17 @@ void InterpolationFilter::filterCopy( const ClpRng& clpRng, const Pel *src, int
if (biMCForDMVR)
{
int shift10BitOut, offset;
+#if JVET_AJ0237_INTERNAL_12BIT
+ if ((clpRng.bd - IF_INTERNAL_PREC_BILINEAR(clpRng.bd)) > 0)
+#else
if ((clpRng.bd - IF_INTERNAL_PREC_BILINEAR) > 0)
+#endif
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ shift10BitOut = (clpRng.bd - IF_INTERNAL_PREC_BILINEAR(clpRng.bd));
+#else
shift10BitOut = (clpRng.bd - IF_INTERNAL_PREC_BILINEAR);
+#endif
offset = (1 << (shift10BitOut - 1));
for (row = 0; row < height; row++)
{
@@ -1612,7 +1620,11 @@ void InterpolationFilter::filterCopy( const ClpRng& clpRng, const Pel *src, int
}
else
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ shift10BitOut = (IF_INTERNAL_PREC_BILINEAR(clpRng.bd) - clpRng.bd);
+#else
shift10BitOut = (IF_INTERNAL_PREC_BILINEAR - clpRng.bd);
+#endif
for (row = 0; row < height; row++)
{
for (col = 0; col < width; col++)
@@ -1649,9 +1661,17 @@ void InterpolationFilter::filterCopy( const ClpRng& clpRng, const Pel *src, int
if (biMCForDMVR)
{
int shift10BitOut, offset;
+#if JVET_AJ0237_INTERNAL_12BIT
+ if ((clpRng.bd - IF_INTERNAL_PREC_BILINEAR(clpRng.bd)) > 0)
+#else
if ((clpRng.bd - IF_INTERNAL_PREC_BILINEAR) > 0)
+#endif
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ shift10BitOut = (clpRng.bd - IF_INTERNAL_PREC_BILINEAR(clpRng.bd));
+#else
shift10BitOut = (clpRng.bd - IF_INTERNAL_PREC_BILINEAR);
+#endif
offset = (1 << (shift10BitOut - 1));
for (row = 0; row < height; row++)
{
@@ -1665,7 +1685,11 @@ void InterpolationFilter::filterCopy( const ClpRng& clpRng, const Pel *src, int
}
else
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ shift10BitOut = (IF_INTERNAL_PREC_BILINEAR(clpRng.bd) - clpRng.bd);
+#else
shift10BitOut = (IF_INTERNAL_PREC_BILINEAR - clpRng.bd);
+#endif
for (row = 0; row < height; row++)
{
for (col = 0; col < width; col++)
@@ -1865,7 +1889,11 @@ void InterpolationFilter::filter(const ClpRng& clpRng, Pel const *src, int srcSt
{
if( isFirst )
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ shift = IF_FILTER_PREC_BILINEAR - (IF_INTERNAL_PREC_BILINEAR(clpRng.bd) - clpRng.bd);
+#else
shift = IF_FILTER_PREC_BILINEAR - (IF_INTERNAL_PREC_BILINEAR - clpRng.bd);
+#endif
offset = 1 << (shift - 1);
}
else
diff --git a/source/Lib/CommonLib/InterpolationFilter.h b/source/Lib/CommonLib/InterpolationFilter.h
index 827258848..4b494a388 100644
--- a/source/Lib/CommonLib/InterpolationFilter.h
+++ b/source/Lib/CommonLib/InterpolationFilter.h
@@ -52,7 +52,11 @@
#define IF_FILTER_PREC 6 ///< Log2 of sum of filter taps
#endif
#define IF_INTERNAL_OFFS (1<<(IF_INTERNAL_PREC-1)) ///< Offset used internally
+#if JVET_AJ0237_INTERNAL_12BIT
+#define IF_INTERNAL_PREC_BILINEAR(bd) std::min(IF_INTERNAL_PREC, int(bd))
+#else
#define IF_INTERNAL_PREC_BILINEAR 10 ///< Number of bits for internal precision
+#endif
#define IF_FILTER_PREC_BILINEAR 4 ///< Bilinear filter coeff precision so that intermediate value will not exceed 16 bit for SIMD - bit exact
#if JVET_R0351_HIGH_BIT_DEPTH_SUPPORT
#define IF_INTERNAL_FRAC_BITS(bd) std::max(2, IF_INTERNAL_PREC - int(bd))
diff --git a/source/Lib/CommonLib/IntraPrediction.cpp b/source/Lib/CommonLib/IntraPrediction.cpp
index c18de13d3..c6631009b 100644
--- a/source/Lib/CommonLib/IntraPrediction.cpp
+++ b/source/Lib/CommonLib/IntraPrediction.cpp
@@ -16521,7 +16521,11 @@ void IntraPrediction::reorderPLT(CodingStructure& cs, Partitioner& partitioner,
}
#if MMLM && LMS_LINEAR_MODEL
+#if JVET_AJ0237_INTERNAL_12BIT
+int IntraPrediction::xCalcLMParametersGeneralized(int64_t x, int64_t y, int64_t xx, int64_t xy, int count, int bitDepth, int& a, int& b, int& iShift)
+#else
int IntraPrediction::xCalcLMParametersGeneralized(int x, int y, int xx, int xy, int count, int bitDepth, int &a, int &b, int &iShift)
+#endif
{
uint32_t uiInternalBitDepth = bitDepth;
@@ -16537,23 +16541,42 @@ int IntraPrediction::xCalcLMParametersGeneralized(int x, int y, int xx, int xy,
int iCountShift = g_aucLog2[count];
+#if JVET_AJ0237_INTERNAL_12BIT
+ int iTempShift = uiInternalBitDepth + iCountShift - ((uiInternalBitDepth > 10) ? 31 : 15);
+#else
int iTempShift = uiInternalBitDepth + iCountShift - 15;
+#endif
if (iTempShift > 0)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ x = (x + ((int64_t)1 << (iTempShift - 1))) >> iTempShift;
+ y = (y + ((int64_t)1 << (iTempShift - 1))) >> iTempShift;
+ xx = (xx + ((int64_t)1 << (iTempShift - 1))) >> iTempShift;
+ xy = (xy + ((int64_t)1 << (iTempShift - 1))) >> iTempShift;
+#else
x = (x + (1 << (iTempShift - 1))) >> iTempShift;
y = (y + (1 << (iTempShift - 1))) >> iTempShift;
xx = (xx + (1 << (iTempShift - 1))) >> iTempShift;
xy = (xy + (1 << (iTempShift - 1))) >> iTempShift;
+#endif
iCountShift -= iTempShift;
}
/////// xCalcLMParameters
+#if JVET_AJ0237_INTERNAL_12BIT
+ int64_t avgX = x >> iCountShift;
+ int64_t avgY = y >> iCountShift;
+
+ int64_t RErrX = x & ((1 << iCountShift) - 1);
+ int64_t RErrY = y & ((1 << iCountShift) - 1);
+#else
int avgX = x >> iCountShift;
int avgY = y >> iCountShift;
int RErrX = x & ((1 << iCountShift) - 1);
int RErrY = y & ((1 << iCountShift) - 1);
+#endif
int iB = 7;
iShift = 13 - iB;
@@ -16566,19 +16589,33 @@ int IntraPrediction::xCalcLMParametersGeneralized(int x, int y, int xx, int xy,
}
else
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ int64_t a1 = xy - (avgX * avgY << iCountShift) - avgX * RErrY - avgY * RErrX;
+ int64_t a2 = xx - (avgX * avgX << iCountShift) - 2 * avgX * RErrX;
+#else
int a1 = xy - (avgX * avgY << iCountShift) - avgX * RErrY - avgY * RErrX;
int a2 = xx - (avgX * avgX << iCountShift) - 2 * avgX * RErrX;
+#endif
const int iShiftA1 = uiInternalBitDepth - 2;
const int iShiftA2 = 5;
const int iAccuracyShift = uiInternalBitDepth + 4;
int iScaleShiftA2 = 0;
int iScaleShiftA1 = 0;
+
+#if JVET_AJ0237_INTERNAL_12BIT
+ int64_t a1s = a1;
+ int64_t a2s = a2;
+
+ iScaleShiftA1 = a1 == 0 ? 0 : floorLog2Uint64(abs(a1)) - iShiftA1;
+ iScaleShiftA2 = a2 == 0 ? 0 : floorLog2Uint64(abs(a2)) - iShiftA2;
+#else
int a1s = a1;
int a2s = a2;
iScaleShiftA1 = a1 == 0 ? 0 : floorLog2(abs(a1)) - iShiftA1;
iScaleShiftA2 = a2 == 0 ? 0 : floorLog2(abs(a2)) - iShiftA2;
+#endif
if (iScaleShiftA1 < 0)
{
@@ -16599,7 +16636,11 @@ int IntraPrediction::xCalcLMParametersGeneralized(int x, int y, int xx, int xy,
if (a2s >= 32)
{
uint32_t a2t = m_auShiftLM[a2s - 32];
+#if JVET_AJ0237_INTERNAL_12BIT
+ a = int(a1s * a2t);
+#else
a = a1s * a2t;
+#endif
}
else
{
@@ -16625,8 +16666,11 @@ int IntraPrediction::xCalcLMParametersGeneralized(int x, int y, int xx, int xy,
iShift = (iShift + iB) - n;
a = a >> n;
-
+#if JVET_AJ0237_INTERNAL_12BIT
+ b = int(avgY - ((a * avgX) >> iShift));
+#else
b = avgY - ((a * avgX) >> iShift);
+#endif
}
return 0;
}
@@ -16699,7 +16743,11 @@ int IntraPrediction::xLMSampleClassifiedTraining(int count, int mean, int meanC,
}
}
+#if JVET_AJ0237_INTERNAL_12BIT
+ int64_t x[2], y[2], xy[2], xx[2];
+#else
int x[2], y[2], xy[2], xx[2];
+#endif
for (int group = 0; group < 2; group++)
{
x[group] = y[group] = xy[group] = xx[group] = 0;
@@ -16851,7 +16899,11 @@ void IntraPrediction::xGetLMParametersLMS(const PredictionUnit &pu, const Compon
srcColor0 = temp.bufAt(0, 0);
curChroma0 = getPredictorPtr(compID);
+#if JVET_AJ0237_INTERNAL_12BIT
+ int64_t x = 0, y = 0, xx = 0, xy = 0;
+#else
int x = 0, y = 0, xx = 0, xy = 0;
+#endif
int iCountShift = 0;
unsigned uiInternalBitDepth = sps.getBitDepth(CHANNEL_TYPE_CHROMA);
@@ -17091,25 +17143,42 @@ void IntraPrediction::xGetLMParametersLMS(const PredictionUnit &pu, const Compon
return;
}
}
-
+#if JVET_AJ0237_INTERNAL_12BIT
+ int iTempShift = uiInternalBitDepth + iCountShift - ((uiInternalBitDepth > 10) ? 31 : 15);
+#else
int iTempShift = uiInternalBitDepth + iCountShift - 15;
+#endif
if (iTempShift > 0)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ x = (x + ((int64_t)1 << (iTempShift - 1))) >> iTempShift;
+ y = (y + ((int64_t)1 << (iTempShift - 1))) >> iTempShift;
+ xx = (xx + ((int64_t)1 << (iTempShift - 1))) >> iTempShift;
+ xy = (xy + ((int64_t)1 << (iTempShift - 1))) >> iTempShift;
+#else
x = (x + (1 << (iTempShift - 1))) >> iTempShift;
y = (y + (1 << (iTempShift - 1))) >> iTempShift;
xx = (xx + (1 << (iTempShift - 1))) >> iTempShift;
xy = (xy + (1 << (iTempShift - 1))) >> iTempShift;
+#endif
iCountShift -= iTempShift;
}
/////// xCalcLMParameters
+#if JVET_AJ0237_INTERNAL_12BIT
+ int64_t avgX = x >> iCountShift;
+ int64_t avgY = y >> iCountShift;
+ int64_t RErrX = x & ((1 << iCountShift) - 1);
+ int64_t RErrY = y & ((1 << iCountShift) - 1);
+#else
int avgX = x >> iCountShift;
int avgY = y >> iCountShift;
int RErrX = x & ((1 << iCountShift) - 1);
int RErrY = y & ((1 << iCountShift) - 1);
+#endif
int iB = 7;
int a = 0;
@@ -17122,20 +17191,32 @@ void IntraPrediction::xGetLMParametersLMS(const PredictionUnit &pu, const Compon
}
else
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ int64_t a1 = xy - (avgX * avgY << iCountShift) - avgX * RErrY - avgY * RErrX;
+ int64_t a2 = xx - (avgX * avgX << iCountShift) - 2 * avgX * RErrX;
+#else
int a1 = xy - (avgX * avgY << iCountShift) - avgX * RErrY - avgY * RErrX;
int a2 = xx - (avgX * avgX << iCountShift) - 2 * avgX * RErrX;
+#endif
const int iShiftA1 = uiInternalBitDepth - 2;
const int iShiftA2 = 5;
const int iAccuracyShift = uiInternalBitDepth + 4;
int iScaleShiftA2 = 0;
int iScaleShiftA1 = 0;
+#if JVET_AJ0237_INTERNAL_12BIT
+ int64_t a1s = a1;
+ int64_t a2s = a2;
+
+ iScaleShiftA1 = a1 == 0 ? 0 : floorLog2Uint64(abs(a1)) - iShiftA1;
+ iScaleShiftA2 = a2 == 0 ? 0 : floorLog2Uint64(abs(a2)) - iShiftA2;
+#else
int a1s = a1;
int a2s = a2;
iScaleShiftA1 = a1 == 0 ? 0 : floorLog2(abs(a1)) - iShiftA1;
iScaleShiftA2 = a2 == 0 ? 0 : floorLog2(abs(a2)) - iShiftA2;
-
+#endif
if (iScaleShiftA1 < 0)
{
iScaleShiftA1 = 0;
@@ -17155,7 +17236,11 @@ void IntraPrediction::xGetLMParametersLMS(const PredictionUnit &pu, const Compon
if (a2s >= 32)
{
uint32_t a2t = m_auShiftLM[a2s - 32];
+#if JVET_AJ0237_INTERNAL_12BIT
+ a = int(a1s * a2t);
+#else
a = a1s * a2t;
+#endif
}
else
{
@@ -17181,7 +17266,11 @@ void IntraPrediction::xGetLMParametersLMS(const PredictionUnit &pu, const Compon
iShift = (iShift + iB) - n;
a = a >> n;
+#if JVET_AJ0237_INTERNAL_12BIT
+ b = int(avgY - ((a * avgX) >> iShift));
+#else
b = avgY - ((a * avgX) >> iShift);
+#endif
cclmModel.setFirstModel( a, b, iShift );
}
@@ -23308,7 +23397,11 @@ void IntraPrediction::xCclmApplyModel(const PredictionUnit &pu, const ComponentI
samples[0] = refLumaBlk.at(x, y); // C
samples[1] = cccmModel.bias();
+#if JVET_AJ0237_INTERNAL_12BIT
+ piPred.at(x, y) = ClipPel<Pel>(Pel((cccmModel.params[0] * samples[0] + cccmModel.params[1] * samples[1] + cccmModel.decimRound) >> cccmModel.decimBits), clpRng);
+#else
piPred.at(x, y) = ClipPel<Pel>(Pel((cccmModel.params[0] * samples[0] + cccmModel.params[1] * samples[1] + CCCM_DECIM_ROUND) >> CCCM_DECIM_BITS), clpRng);
+#endif
}
}
}
@@ -24292,7 +24385,11 @@ void IntraPrediction::combineCcpAndInter(PredictionUnit& pu, PelBuf& inPredCb, P
#define DIV_INTR_BITS (DIV_PREC_BITS - DIV_SLOT_BITS)
#define DIV_INTR_ROUND (1 << DIV_INTR_BITS >> 1)
+#if JVET_AJ0237_INTERNAL_12BIT
+int64_t xDivide(int64_t num, int64_t denom, int decimBits) // Note: assumes positive denominator
+#else
int64_t xDivide(int64_t num, int64_t denom) // Note: assumes positive denominator
+#endif
{
static const int pow2W[8] = { 214, 153, 113, 86, 67, 53, 43, 35 }; // DIV_PREC_BITS_POW2
static const int pow2O[8] = { 4822, 5952, 6624, 6792, 6408, 5424, 3792, 1466 }; // DIV_PREC_BITS
@@ -24306,11 +24403,19 @@ int64_t xDivide(int64_t num, int64_t denom) // Note: assumes positive denominato
int scale = ((pow2W[diffFull] * ((normDiff2 * normDiff2) >> DIV_PREC_BITS)) >> DIV_PREC_BITS_POW2) - (normDiff2 >> 1) + pow2B[diffFull];
+#if JVET_AJ0237_INTERNAL_12BIT
+ return ((num << (decimBits - DIV_PREC_BITS)) * scale + round) >> shift;
+#else
return ( (num << (CCCM_DECIM_BITS - DIV_PREC_BITS)) * scale + round) >> shift;
+#endif
}
#if JVET_AC0053_GAUSSIAN_SOLVER
+#if JVET_AJ0237_INTERNAL_12BIT
+void xGetDivScaleRoundShift(int64_t denom, int decimBits, int& scale, int& round, int& shift) // Note: assumes positive denominator
+#else
void xGetDivScaleRoundShift(int64_t denom, int &scale, int &round, int &shift) // Note: assumes positive denominator
+#endif
{
static const int pow2W[8] = { 214, 153, 113, 86, 67, 53, 43, 35 }; // DIV_PREC_BITS_POW2
static const int pow2O[8] = { 4822, 5952, 6624, 6792, 6408, 5424, 3792, 1466 }; // DIV_PREC_BITS
@@ -24323,7 +24428,11 @@ void xGetDivScaleRoundShift(int64_t denom, int &scale, int &round, int &shift) /
int normDiff2 = normDiff - pow2O[diffFull];
scale = ((pow2W[diffFull] * ((normDiff2 * normDiff2) >> DIV_PREC_BITS)) >> DIV_PREC_BITS_POW2) - (normDiff2 >> 1) + pow2B[diffFull];
+#if JVET_AJ0237_INTERNAL_12BIT
+ scale <<= decimBits - DIV_PREC_BITS;
+#else
scale <<= CCCM_DECIM_BITS - DIV_PREC_BITS;
+#endif
}
#endif
@@ -24333,8 +24442,22 @@ void xGetDivScaleRoundShift(int64_t denom, int &scale, int &round, int &shift) /
#undef DIV_INTR_BITS
#undef DIV_INTR_ROUND
+#if JVET_AJ0237_INTERNAL_12BIT
+int xCccmDivideLowPrec(int64_t num, int64_t denom, int decimBits)
+#else
int xCccmDivideLowPrec(int64_t num, int64_t denom)
+#endif
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ if (num < 0)
+ {
+ return -int(xDivide(-num, denom, decimBits) >> decimBits);
+ }
+ else
+ {
+ return int(xDivide(num, denom, decimBits) >> decimBits);
+ }
+#else
if ( num < 0 )
{
return -int(xDivide(-num, denom) >> CCCM_DECIM_BITS);
@@ -24343,13 +24466,21 @@ int xCccmDivideLowPrec(int64_t num, int64_t denom)
{
return int(xDivide(num, denom) >> CCCM_DECIM_BITS);
}
+#endif
}
+#if JVET_AJ0237_INTERNAL_12BIT
+int64_t xCccmDivide(int64_t num, int64_t denom, int decimBits) // Note: assumes positive denominator
+{
+ return xDivide(num, denom, decimBits);
+}
+#else
int64_t xCccmDivide(int64_t num, int64_t denom) // Note: assumes positive denominator
{
return xDivide(num, denom);
}
#endif
+#endif
#if JVET_AD0120_LBCCP || JVET_AG0154_DECODER_DERIVED_CCP_FUSION
#if JVET_AA0057_CCCM || JVET_AG0154_DECODER_DERIVED_CCP_FUSION
@@ -26328,7 +26459,16 @@ int IntraPrediction::xBvgCccmCalcBlkAver(const PredictionUnit& pu) const
}
}
#if JVET_AB0174_CCCM_DIV_FREE
+#if JVET_AJ0237_INTERNAL_12BIT
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int bd = pu.cu->slice->getSPS()->getBitDepth(CHANNEL_TYPE_LUMA);
+ return numSamples == 0 ? (1 << (bd - 1)) : xCccmDivideLowPrec(sumSamples, numSamples, DECIM_BITS(bd));
+#else
+ return numSamples == 0 ? ((1 << pu.cu->slice->getSPS()->getBitDepth(CHANNEL_TYPE_LUMA)) - 1) : xCccmDivideLowPrec(sumSamples, numSamples);
+#endif
+#else
return numSamples == 0 ? 512 : xCccmDivideLowPrec(sumSamples, numSamples);
+#endif
#else
return numSamples == 0 ? 512 : ( sumSamples + numSamples/2) / numSamples;
#endif
@@ -26856,7 +26996,16 @@ int IntraPrediction::xCccmCalcRefAver(const PredictionUnit& pu
#endif
#if JVET_AB0174_CCCM_DIV_FREE
+#if JVET_AJ0237_INTERNAL_12BIT
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int bd = pu.cu->slice->getSPS()->getBitDepth(CHANNEL_TYPE_LUMA);
+ return numSamples == 0 ? (1 << (bd - 1)) : xCccmDivideLowPrec(sumSamples, numSamples, DECIM_BITS(bd));
+#else
+ return numSamples == 0 ? ((1 << pu.cu->slice->getSPS()->getBitDepth(CHANNEL_TYPE_LUMA)) - 1) : xCccmDivideLowPrec(sumSamples, numSamples);
+#endif
+#else
return numSamples == 0 ? 512 : xCccmDivideLowPrec(sumSamples, numSamples);
+#endif
#else
return numSamples == 0 ? 512 : ( sumSamples + numSamples/2) / numSamples;
#endif
@@ -27787,7 +27936,11 @@ int IntraPrediction::xCflmCalcRefAver(const PredictionUnit& pu, const CompArea&
}
#if JVET_AD0184_REMOVAL_OF_DIVISION_OPERATIONS
+#if JVET_AJ0237_INTERNAL_12BIT
+ return numSamples == 0 ? (1 << (pu.cu->slice->getSPS()->getBitDepth(CHANNEL_TYPE_LUMA) - 1)) : PU::getMeanValue(sumSamples + (numSamples >> 1), numSamples);
+#else
return numSamples == 0 ? 512 : PU::getMeanValue( sumSamples + (numSamples >> 1), numSamples);
+#endif
#else
return numSamples == 0 ? 512 : (sumSamples + numSamples / 2) / numSamples;
#endif
@@ -27797,7 +27950,11 @@ int IntraPrediction::xCflmCalcRefAver(const PredictionUnit& pu, const CompArea&
#if JVET_AA0057_CCCM || JVET_AB0092_GLM_WITH_LUMA || JVET_AC0119_LM_CHROMA_FUSION || JVET_AG0058_EIP || JVET_AG0154_DECODER_DERIVED_CCP_FUSION
#if JVET_AC0053_GAUSSIAN_SOLVER
+#if JVET_AJ0237_INTERNAL_12BIT
+void CccmCovariance::gaussBacksubstitution( TCccmCoeff* x, int numEq, int col, int round, int bits)
+#else
void CccmCovariance::gaussBacksubstitution( TCccmCoeff* x, int numEq, int col )
+#endif
{
x[numEq-1] = C[numEq-1][col];
@@ -27807,7 +27964,11 @@ void CccmCovariance::gaussBacksubstitution( TCccmCoeff* x, int numEq, int col )
for( int j = i+1; j < numEq; j++ )
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ x[i] -= FIXED_MULT(C[i][j], x[j], round, bits);
+#else
x[i] -= FIXED_MULT(C[i][j], x[j]);
+#endif
}
}
}
@@ -27825,6 +27986,11 @@ void CccmCovariance::gaussElimination( TCccmCoeff A[CCCM_NUM_PARAMS_MAX][CCCM_NU
#else
int reg = 2 << (bd - 8);
#endif
+
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int decimBits = DECIM_BITS(bd);
+ const int decimRound = (1 << (decimBits - 1));
+#endif
// Create an [M][M+2] matrix system (could have been done already when calculating auto/cross-correlations)
for( int i = 0; i < numEq; i++ )
@@ -27847,7 +28013,11 @@ void CccmCovariance::gaussElimination( TCccmCoeff A[CCCM_NUM_PARAMS_MAX][CCCM_NU
#if JVET_AB0174_CCCM_DIV_FREE
int scale, round, shift;
+#if JVET_AJ0237_INTERNAL_12BIT
+ xGetDivScaleRoundShift(diag, decimBits, scale, round, shift);
+#else
xGetDivScaleRoundShift(diag, scale, round, shift);
+#endif
#endif
for( int j = i+1; j < numEq+numFilters; j++ )
@@ -27867,7 +28037,11 @@ void CccmCovariance::gaussElimination( TCccmCoeff A[CCCM_NUM_PARAMS_MAX][CCCM_NU
// On row j all elements with k < i+1 are now zero (not zeroing those here as backsubstitution does not need them)
for( int k = i + 1; k < numEq+numFilters; k++ )
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[k] -= FIXED_MULT(scale, src[k], decimRound, decimBits);
+#else
dst[k] -= FIXED_MULT(scale, src[k]);
+#endif
}
}
}
@@ -27875,12 +28049,21 @@ void CccmCovariance::gaussElimination( TCccmCoeff A[CCCM_NUM_PARAMS_MAX][CCCM_NU
// Solve with backsubstitution
if ( numFilters == 2 )
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ gaussBacksubstitution(x0, numEq, colChr0, decimRound, decimBits);
+ gaussBacksubstitution(x1, numEq, colChr1, decimRound, decimBits);
+#else
gaussBacksubstitution(x0, numEq, colChr0);
gaussBacksubstitution(x1, numEq, colChr1);
+#endif
}
else
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ gaussBacksubstitution(x0, numEq, colChr0, decimRound, decimBits);
+#else
gaussBacksubstitution(x0, numEq, colChr0);
+#endif
}
}
@@ -28055,7 +28238,11 @@ void CccmCovariance::solve1( const Pel A[CCCM_NUM_PARAMS_MAX][CCCM_REF_SAMPLES_M
#endif
// Scale the matrix and vector to selected dynamic range
+#if JVET_AJ0237_INTERNAL_12BIT
+ int matrixShift = ((model.bd > 10) ? CCCM_MATRIX_BITS_HBD : 28) - 2 * model.bd - ceilLog2(sampleNum);
+#else
int matrixShift = 28 - 2 * model.bd - ceilLog2( sampleNum );
+#endif
if( matrixShift > 0 )
{
@@ -28104,8 +28291,12 @@ void CccmCovariance::solve1( const Pel A[CCCM_NUM_PARAMS_MAX][CCCM_REF_SAMPLES_M
#if JVET_AB0174_CCCM_DIV_FREE
// Add the chroma offset to bias term (after shifting up by CCCM_DECIM_BITS and down by cccmModelCb.bd - 1)
+#if JVET_AJ0237_INTERNAL_12BIT
+ model.params[numParams - 1] += chromaOffset << (model.decimBits - (model.bd - 1));
+#else
model.params[numParams - 1] += chromaOffset << (CCCM_DECIM_BITS - (model.bd - 1));
#endif
+#endif
}
#if JVET_AB0174_CCCM_DIV_FREE
@@ -28174,7 +28365,11 @@ void CccmCovariance::solve2( const Pel A[CCCM_NUM_PARAMS_MAX][CCCM_REF_SAMPLES_M
// Scale the matrix and vector to selected dynamic range
CHECK( modelCb.bd != modelCr.bd, "Bitdepth of Cb and Cr is different" );
#if JVET_AE0059_INTER_CCCM
+#if JVET_AJ0237_INTERNAL_12BIT
+ int matrixShift = ((modelCb.bd > 10) ? CCCM_MATRIX_BITS_HBD : (interCccmMode ? 28 : CCCM_MATRIX_BITS)) - 2 * modelCb.bd - ceilLog2(sampleNum);
+#else
int matrixShift = (interCccmMode ? 28 : CCCM_MATRIX_BITS) - 2 * modelCb.bd - ceilLog2( sampleNum );
+#endif
#else
int matrixShift = CCCM_MATRIX_BITS - 2 * modelCb.bd - ceilLog2( sampleNum );
#endif
@@ -28242,9 +28437,14 @@ void CccmCovariance::solve2( const Pel A[CCCM_NUM_PARAMS_MAX][CCCM_REF_SAMPLES_M
#if JVET_AB0174_CCCM_DIV_FREE
// Add the chroma offset to bias term (after shifting up by CCCM_DECIM_BITS and down by cccmModelCb.bd - 1)
+#if JVET_AJ0237_INTERNAL_12BIT
+ modelCb.params[numParams - 1] += chromaOffsetCb << (modelCb.decimBits - (modelCb.bd - 1));
+ modelCr.params[numParams - 1] += chromaOffsetCr << (modelCr.decimBits - (modelCr.bd - 1));
+#else
modelCb.params[numParams - 1] += chromaOffsetCb << (CCCM_DECIM_BITS - (modelCb.bd - 1));
modelCr.params[numParams - 1] += chromaOffsetCr << (CCCM_DECIM_BITS - (modelCr.bd - 1));
#endif
+#endif
}
#endif
@@ -30445,8 +30645,16 @@ void CccmCovariance::solveEip(const TCccmCoeff* A, const TCccmCoeff* Y, const in
{
regularizationParam = (sampleNum <= REGULARIZED_EIP_L2_SAMPLE_THRESHOLD) ? REGULARIZED_EIP_L2_SMALL * numParams : REGULARIZED_EIP_L2_LARGE * numParams;
}
+#if JVET_AJ0237_INTERNAL_12BIT
+ regularizationParam <<= 2 * std::max(0, model.bd - 10);
+#endif
+#else
+#if JVET_AJ0237_INTERNAL_12BIT
+ int regShift = 2 * std::max(0, model.bd - 10);
+ const int regularizationParam = ((sampleNum <= REGULARIZED_EIP_L2_SAMPLE_THRESHOLD) ? REGULARIZED_EIP_L2_SMALL * numParams : REGULARIZED_EIP_L2_LARGE * numParams) << regShift;
#else
const int regularizationParam = (sampleNum <= REGULARIZED_EIP_L2_SAMPLE_THRESHOLD) ? REGULARIZED_EIP_L2_SMALL * numParams : REGULARIZED_EIP_L2_LARGE * numParams;
+#endif
#endif
for (int coli0 = 0; coli0 < numParams - 1; coli0++) // The last term (bias) is not regularized.
{
@@ -30467,7 +30675,11 @@ void CccmCovariance::solveEip(const TCccmCoeff* A, const TCccmCoeff* Y, const in
}
#endif
// Scale the matrix and vector to selected dynamic range
+#if JVET_AJ0237_INTERNAL_12BIT
+ int matrixShift = ((model.bd > 10) ? CCCM_MATRIX_BITS_HBD : 28) - 2 * model.bd - ceilLog2(sampleNum);
+#else
int matrixShift = 28 - 2 * model.bd - ceilLog2(sampleNum);
+#endif
if (matrixShift > 0)
{
@@ -30515,8 +30727,12 @@ void CccmCovariance::solveEip(const TCccmCoeff* A, const TCccmCoeff* Y, const in
#if JVET_AB0174_CCCM_DIV_FREE
// Add the chroma offset to bias term (after shifting up by CCCM_DECIM_BITS and down by cccmModelCb.bd - 1)
+#if JVET_AJ0237_INTERNAL_12BIT
+ model.params[numParams - 1] += lumaOffset << (model.decimBits - (model.bd - 1));
+#else
model.params[numParams - 1] += lumaOffset << (CCCM_DECIM_BITS - (model.bd - 1));
#endif
+#endif
}
void IntraPrediction::initEipParams(const PredictionUnit& pu, const ComponentID compId)
diff --git a/source/Lib/CommonLib/IntraPrediction.h b/source/Lib/CommonLib/IntraPrediction.h
index 9601371d8..66ae8a1e5 100644
--- a/source/Lib/CommonLib/IntraPrediction.h
+++ b/source/Lib/CommonLib/IntraPrediction.h
@@ -145,7 +145,11 @@ typedef short TrainDataType;
#if JVET_AA0057_CCCM || JVET_AB0092_GLM_WITH_LUMA || JVET_AC0119_LM_CHROMA_FUSION || JVET_AG0058_EIP || JVET_AG0154_DECODER_DERIVED_CCP_FUSION
typedef int64_t TCccmCoeff;
+#if JVET_AJ0237_INTERNAL_12BIT
+#define FIXED_MULT(x, y, round, bits) TCccmCoeff((int64_t(x)*(y) + round) >> bits )
+#else
#define FIXED_MULT(x, y) TCccmCoeff((int64_t(x)*(y) + CCCM_DECIM_ROUND) >> CCCM_DECIM_BITS )
+#endif
#if !JVET_AB0174_CCCM_DIV_FREE
#define FIXED_DIV(x, y) TCccmCoeff((int64_t(x) << CCCM_DECIM_BITS ) / (y) )
#endif
@@ -157,6 +161,10 @@ struct CccmModel
bd = bitdepth;
midVal = ( 1 << ( bitdepth - 1 ) );
params.resize( num );
+#if JVET_AJ0237_INTERNAL_12BIT
+ decimBits = DECIM_BITS(bd);
+ decimRound = (1 << (decimBits - 1));
+#endif
}
~CccmModel() {}
@@ -164,6 +172,10 @@ struct CccmModel
std::vector<TCccmCoeff> params;
int bd;
int midVal;
+#if JVET_AJ0237_INTERNAL_12BIT
+ int decimRound;
+ int decimBits;
+#endif
const int getNumParams() const
{
@@ -176,7 +188,11 @@ struct CccmModel
std::fill( params.begin(), params.end(), 0 );
+#if JVET_AJ0237_INTERNAL_12BIT
+ params[numParams - 1] = (TCccmCoeff)1 << decimBits; // Default bias to 1
+#else
params[numParams - 1] = 1 << CCCM_DECIM_BITS; // Default bias to 1
+#endif
}
Pel convolve(Pel* vector)
@@ -188,7 +204,11 @@ struct CccmModel
sum += params[i] * vector[i];
}
+#if JVET_AJ0237_INTERNAL_12BIT
+ return Pel( (sum + decimRound) >> decimBits);
+#else
return Pel( (sum + CCCM_DECIM_ROUND ) >> CCCM_DECIM_BITS );
+#endif
}
Pel nonlinear(const Pel val) { return (val * val + midVal) >> bd; }
@@ -238,7 +258,11 @@ private:
TCccmCoeff C[CCCM_NUM_PARAMS_MAX][CCCM_NUM_PARAMS_MAX + 2];
#if JVET_AC0053_GAUSSIAN_SOLVER
+#if JVET_AJ0237_INTERNAL_12BIT
+ void gaussBacksubstitution ( TCccmCoeff* x, int numEq, int col, int round, int bits);
+#else
void gaussBacksubstitution ( TCccmCoeff* x, int numEq, int col );
+#endif
#if JVET_AE0059_INTER_CCCM
void gaussElimination ( TCccmCoeff A[CCCM_NUM_PARAMS_MAX][CCCM_NUM_PARAMS_MAX], TCccmCoeff* y0, TCccmCoeff* x0, TCccmCoeff* y1, TCccmCoeff* x1, int numEq, int numFilters, int bd, const bool interCccmMode = false);
#else
@@ -659,7 +683,11 @@ public:
int b;
int shift;
};
+#if JVET_AJ0237_INTERNAL_12BIT
+ int xCalcLMParametersGeneralized(int64_t x, int64_t y, int64_t xx, int64_t xy, int count, int bitDepth, int& a, int& b, int& iShift);
+#else
int xCalcLMParametersGeneralized(int x, int y, int xx, int xy, int count, int bitDepth, int &a, int &b, int &iShift);
+#endif
int xLMSampleClassifiedTraining (int count, int mean, int meanC, int lumaSamples[], int chrmSamples[], int bitDepth, MMLMParameters parameters[]);
#if JVET_AG0136_INTRA_TMP_LIC
std::array<int, 7>& getMemLicParams(const int licIdc, const int idx) { return m_memLicParams[licIdc][idx]; }
diff --git a/source/Lib/CommonLib/Picture.cpp b/source/Lib/CommonLib/Picture.cpp
index c3e95010c..e1e609a75 100644
--- a/source/Lib/CommonLib/Picture.cpp
+++ b/source/Lib/CommonLib/Picture.cpp
@@ -1810,6 +1810,9 @@ void Picture::calcLumaClpParams()
clipDeltaShift = ADAPTIVE_CLIP_SHIFT_DELTA_VALUE_0;
cs->slice->setAdaptiveClipQuant(false);
}
+#if JVET_AJ0237_INTERNAL_12BIT
+ clipDeltaShift += std::max(0, cs->sps->getBitDepth(toChannelType(COMPONENT_Y)) - 10);
+#endif
int pelMaxOF = 0;
int pelMinOF = (1 << cs->sps->getBitDepth(toChannelType(COMPONENT_Y))) - 1;
const int orgPelMin = pelMin;
diff --git a/source/Lib/CommonLib/Rom.cpp b/source/Lib/CommonLib/Rom.cpp
index 3dfaab825..b4aba4478 100644
--- a/source/Lib/CommonLib/Rom.cpp
+++ b/source/Lib/CommonLib/Rom.cpp
@@ -625,14 +625,26 @@ MsgLevel g_verbosity = VERBOSE;
#if JVET_Y0141_SIGN_PRED_IMPROVE
#if JVET_W0119_LFNST_EXTENSION || EXTENDED_LFNST
#if JVET_AJ0175_NSPT_FOR_NONREG_MODES
+#if JVET_AJ0237_INTERNAL_12BIT
+int16_t* g_resiBorderTemplateLFNST[NUM_NSPT_BLOCK_TYPES][6][6][210];
+#else
int8_t * g_resiBorderTemplateLFNST[NUM_NSPT_BLOCK_TYPES][6][6][210];
+#endif
+#else
+#if JVET_AJ0237_INTERNAL_12BIT
+int16_t* g_resiBorderTemplateLFNST[6][6][210];
#else
int8_t * g_resiBorderTemplateLFNST[6][6][210];
#endif
+#endif
#else
int8_t * g_resiBorderTemplateLFNST[6][6][16];
#endif
+#if JVET_AJ0237_INTERNAL_12BIT
+int16_t* g_resiBorderTemplate[6][6][NUM_TRANS_TYPE * NUM_TRANS_TYPE];
+#else
int8_t * g_resiBorderTemplate[6][6][NUM_TRANS_TYPE*NUM_TRANS_TYPE];
+#endif
#else
const int8_t * g_resiBorderTemplate[6][6][NUM_TRANS_TYPE*NUM_TRANS_TYPE];
#endif
diff --git a/source/Lib/CommonLib/Rom.h b/source/Lib/CommonLib/Rom.h
index eb045838d..6f832c7a0 100644
--- a/source/Lib/CommonLib/Rom.h
+++ b/source/Lib/CommonLib/Rom.h
@@ -82,14 +82,26 @@ void destroyMipFilters();
#if JVET_Y0141_SIGN_PRED_IMPROVE
#if JVET_W0119_LFNST_EXTENSION || EXTENDED_LFNST
#if JVET_AJ0175_NSPT_FOR_NONREG_MODES
+#if JVET_AJ0237_INTERNAL_12BIT
+extern int16_t* g_resiBorderTemplateLFNST[NUM_NSPT_BLOCK_TYPES][6][6][210];
+#else
extern int8_t * g_resiBorderTemplateLFNST[NUM_NSPT_BLOCK_TYPES][6][6][210];
+#endif
+#else
+#if JVET_AJ0237_INTERNAL_12BIT
+extern int16_t* g_resiBorderTemplateLFNST[6][6][210];
#else
extern int8_t * g_resiBorderTemplateLFNST[6][6][210];
#endif
+#endif
#else
extern int8_t * g_resiBorderTemplateLFNST[6][6][16];
#endif
+#if JVET_AJ0237_INTERNAL_12BIT
+extern int16_t* g_resiBorderTemplate[6][6][NUM_TRANS_TYPE * NUM_TRANS_TYPE];
+#else
extern int8_t * g_resiBorderTemplate[6][6][NUM_TRANS_TYPE*NUM_TRANS_TYPE];
+#endif
#else
extern const int8_t * g_resiBorderTemplate[6][6][NUM_TRANS_TYPE*NUM_TRANS_TYPE];
#endif
diff --git a/source/Lib/CommonLib/SampleAdaptiveOffset.cpp b/source/Lib/CommonLib/SampleAdaptiveOffset.cpp
index 04f37c011..f2a946054 100644
--- a/source/Lib/CommonLib/SampleAdaptiveOffset.cpp
+++ b/source/Lib/CommonLib/SampleAdaptiveOffset.cpp
@@ -2158,7 +2158,12 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClipEdge(const ComponentID compID,
const int edgePosXB = g_ccSaoEdgePosX[edgeDir][1], edgePosYB = g_ccSaoEdgePosY[edgeDir][1];
const int bandCmp = g_ccSaoBandTab [bandIdc][0];
const int bandNum = g_ccSaoBandTab [bandIdc][1];
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int bdShift = std::max(0, bitDepth - 10);
+ const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr] << bdShift;
+#else
const int edgeThrVal = g_ccSaoEdgeThr [edgeIdc][edgeThr];
+#endif
const int edgeNum = g_ccSaoEdgeNum [edgeIdc][0];
const int edgeNumUni = g_ccSaoEdgeNum [edgeIdc][1];
const int srcStrideE = edgeCmp == COMPONENT_Y ? srcStrideY : edgeCmp == COMPONENT_Cb ? srcStrideU : srcStrideV;
@@ -2216,7 +2221,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClipEdge(const ComponentID compID,
const int bandIdx = (*col[bandCmp] * bandNum) >> bitDepth;
const int classIdx = bandIdx * edgeNum + edgeIdx;
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
#else
const Pel *colY = srcY + x;
const Pel *colA = srcY + x + srcStrideY * candPosYYA + candPosYXA;
@@ -2295,7 +2304,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClipEdge(const ComponentID compID,
const int bandIdx = (*col[bandCmp] * bandNum) >> bitDepth;
const int classIdx = bandIdx * edgeNum + edgeIdx;
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
#else
const Pel *colY = srcY + x;
const Pel *colA = srcY + x + srcStrideY * candPosYYA + candPosYXA;
@@ -2369,7 +2382,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClipEdge(const ComponentID compID,
const int bandIdx = (*col[bandCmp] * bandNum) >> bitDepth;
const int classIdx = bandIdx * edgeNum + edgeIdx;
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
#else
const Pel *colY = srcY + x;
const Pel *colA = srcY + x + srcStrideY * candPosYYA + candPosYXA;
@@ -2436,7 +2453,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClipEdge(const ComponentID compID,
const int bandIdx = (*col[bandCmp] * bandNum) >> bitDepth;
const int classIdx = bandIdx * edgeNum + edgeIdx;
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
#else
const Pel *colY = srcY + x;
const Pel *colA = srcY + x + srcStrideY * candPosYYA + candPosYXA;
@@ -2504,7 +2525,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClipEdge(const ComponentID compID,
const int bandIdx = (*col[bandCmp] * bandNum) >> bitDepth;
const int classIdx = bandIdx * edgeNum + edgeIdx;
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
#else
const Pel *colY = srcY + x;
const Pel *colA = srcY + x + srcStrideY * candPosYYA + candPosYXA;
@@ -2573,7 +2598,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClipEdge(const ComponentID compID,
const int bandIdx = (*col[bandCmp] * bandNum) >> bitDepth;
const int classIdx = bandIdx * edgeNum + edgeIdx;
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
#else
const Pel *colY = srcY + x;
const Pel *colA = srcY + x + srcStrideY * candPosYYA + candPosYXA;
@@ -2640,7 +2669,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClipEdge(const ComponentID compID,
const int bandIdx = (*col[bandCmp] * bandNum) >> bitDepth;
const int classIdx = bandIdx * edgeNum + edgeIdx;
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
#else
const Pel *colY = srcY + x;
const Pel *colA = srcY + x + srcStrideY * candPosYYA + candPosYXA;
@@ -2708,7 +2741,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClipEdge(const ComponentID compID,
const int bandIdx = (*col[bandCmp] * bandNum) >> bitDepth;
const int classIdx = bandIdx * edgeNum + edgeIdx;
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
#else
const Pel *colY = srcY + x;
const Pel *colA = srcY + x + srcStrideY * candPosYYA + candPosYXA;
@@ -2788,7 +2825,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClipEdge(const ComponentID compID,
const int bandIdx = (*col[bandCmp] * bandNum) >> bitDepth;
const int classIdx = bandIdx * edgeNum + edgeIdx;
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
#else
const Pel *colY = srcY + (x << chromaScaleX);
const Pel *colA = srcY + (x << chromaScaleX) + srcStrideY * candPosYYA + candPosYXA;
@@ -2868,7 +2909,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClipEdge(const ComponentID compID,
const int bandIdx = (*col[bandCmp] * bandNum) >> bitDepth;
const int classIdx = bandIdx * edgeNum + edgeIdx;
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
#else
const Pel *colY = srcY + (x << chromaScaleX);
const Pel *colA = srcY + (x << chromaScaleX) + srcStrideY * candPosYYA + candPosYXA;
@@ -2943,7 +2988,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClipEdge(const ComponentID compID,
const int bandIdx = (*col[bandCmp] * bandNum) >> bitDepth;
const int classIdx = bandIdx * edgeNum + edgeIdx;
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
#else
const Pel *colY = srcY + (x << chromaScaleX);
const Pel *colA = srcY + (x << chromaScaleX) + srcStrideY * candPosYYA + candPosYXA;
@@ -3011,7 +3060,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClipEdge(const ComponentID compID,
const int bandIdx = (*col[bandCmp] * bandNum) >> bitDepth;
const int classIdx = bandIdx * edgeNum + edgeIdx;
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
#else
const Pel *colY = srcY + (x << chromaScaleX);
const Pel *colA = srcY + (x << chromaScaleX) + srcStrideY * candPosYYA + candPosYXA;
@@ -3080,7 +3133,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClipEdge(const ComponentID compID,
const int bandIdx = (*col[bandCmp] * bandNum) >> bitDepth;
const int classIdx = bandIdx * edgeNum + edgeIdx;
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
#else
const Pel *colY = srcY + (x << chromaScaleX);
const Pel *colA = srcY + (x << chromaScaleX) + srcStrideY * candPosYYA + candPosYXA;
@@ -3150,7 +3207,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClipEdge(const ComponentID compID,
const int bandIdx = (*col[bandCmp] * bandNum) >> bitDepth;
const int classIdx = bandIdx * edgeNum + edgeIdx;
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
#else
const Pel *colY = srcY + (x << chromaScaleX);
const Pel *colA = srcY + (x << chromaScaleX) + srcStrideY * candPosYYA + candPosYXA;
@@ -3218,7 +3279,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClipEdge(const ComponentID compID,
const int bandIdx = (*col[bandCmp] * bandNum) >> bitDepth;
const int classIdx = bandIdx * edgeNum + edgeIdx;
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
#else
const Pel *colY = srcY + (x << chromaScaleX);
const Pel *colA = srcY + (x << chromaScaleX) + srcStrideY * candPosYYA + candPosYXA;
@@ -3287,7 +3352,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClipEdge(const ComponentID compID,
const int bandIdx = (*col[bandCmp] * bandNum) >> bitDepth;
const int classIdx = bandIdx * edgeNum + edgeIdx;
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
#else
const Pel *colY = srcY + (x << chromaScaleX);
const Pel *colA = srcY + (x << chromaScaleX) + srcStrideY * candPosYYA + candPosYXA;
@@ -3538,7 +3607,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
srcY += srcStrideY;
srcU += srcStrideU * chromaScaleYM1;
@@ -3565,7 +3638,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
srcY += srcStrideY;
srcU += srcStrideU * ((y & 0x1) | chromaScaleYM1);
@@ -3604,7 +3681,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
srcY += srcStrideY;
@@ -3638,7 +3719,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
srcY += srcStrideY;
@@ -3666,7 +3751,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
srcY += srcStrideY;
@@ -3700,7 +3789,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
srcY += srcStrideY;
srcU += srcStrideU * ((y & 0x1) | chromaScaleYM1);
@@ -3727,7 +3820,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
srcY += srcStrideY;
srcU += srcStrideU * ((y & 0x1) | chromaScaleYM1);
@@ -3760,7 +3857,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
srcY += srcStrideY;
srcU += srcStrideU * ((y & 0x1) | chromaScaleYM1);
@@ -3793,7 +3894,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
srcY += srcStrideY;
srcU += srcStrideU * ((y & 0x1) | chromaScaleYM1);
@@ -3822,7 +3927,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
break;
}
@@ -3850,7 +3959,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
srcY += srcStrideY;
srcU += srcStrideU * ((y & 0x1) | chromaScaleYM1);
@@ -3883,7 +3996,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
srcY += srcStrideY;
srcU += srcStrideU * ((y & 0x1) | chromaScaleYM1);
@@ -3911,7 +4028,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
break;
}
@@ -3950,7 +4071,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
srcY += srcStrideY << chromaScaleY;
srcU += srcStrideU;
@@ -3977,7 +4102,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
srcY += srcStrideY << chromaScaleY;
srcU += srcStrideU;
@@ -4016,7 +4145,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
srcY += srcStrideY << chromaScaleY;
@@ -4050,7 +4183,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
srcY += srcStrideY << chromaScaleY;
@@ -4078,7 +4215,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
srcY += srcStrideY << chromaScaleY;
@@ -4112,7 +4253,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
srcY += srcStrideY << chromaScaleY;
srcU += srcStrideU;
@@ -4139,7 +4284,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
srcY += srcStrideY << chromaScaleY;
srcU += srcStrideU;
@@ -4172,7 +4321,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
srcY += srcStrideY << chromaScaleY;
srcU += srcStrideU;
@@ -4205,7 +4358,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
srcY += srcStrideY << chromaScaleY;
srcU += srcStrideU;
@@ -4234,7 +4391,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
break;
}
@@ -4262,7 +4423,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
srcY += srcStrideY << chromaScaleY;
srcU += srcStrideU;
@@ -4295,7 +4460,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
srcY += srcStrideY << chromaScaleY;
srcU += srcStrideU;
@@ -4322,7 +4491,11 @@ void SampleAdaptiveOffset::offsetBlockCcSaoNoClip(const ComponentID compID, cons
const int classIdx = bandIdx;
// dst[x] = ClipPel<int>(dst[x] + offset[classIdx], clpRng);
+#if JVET_AJ0237_INTERNAL_12BIT
+ dst[x] = dst[x] + (offset[classIdx] << m_offsetStepLog2[compID]);
+#else
dst[x] = dst[x] + offset[classIdx];
+#endif
}
}
break;
diff --git a/source/Lib/CommonLib/Slice.h b/source/Lib/CommonLib/Slice.h
index 29864a45a..ec4cababa 100644
--- a/source/Lib/CommonLib/Slice.h
+++ b/source/Lib/CommonLib/Slice.h
@@ -3531,8 +3531,15 @@ public:
void setColRefIdx( uint32_t refIdx) { m_colRefIdx = refIdx; }
uint32_t getColRefIdx() { return m_colRefIdx; }
#if JVET_AA0093_DIVERSITY_CRITERION_FOR_ARMC
+#if JVET_AJ0237_INTERNAL_12BIT
+ void setCostForARMC(uint32_t cost, int bitDepth) { m_costForARMC = (cost << (std::max(0, bitDepth - 10))); }
+#else
void setCostForARMC(uint32_t cost) { m_costForARMC = cost; }
+#endif
uint32_t getCostForARMC() { return m_costForARMC; }
+#if JVET_AJ0237_INTERNAL_12BIT
+ uint32_t getCostForARMC(int bitDepth) { return m_costForARMC >> (std::max(0, bitDepth - 10)); } // for header parsing/writing purpose
+#endif
#endif
#if JVET_AC0185_ENHANCED_TEMPORAL_MOTION_DERIVATION
void setPicColFromL0Flag2nd(bool val) { m_picColFromL0Flag2nd = val; }
@@ -4084,6 +4091,9 @@ public:
#endif
void checkColRefIdx(uint32_t curSliceSegmentIdx, const Picture* pic);
#if JVET_AA0093_DIVERSITY_CRITERION_FOR_ARMC
+#if JVET_AJ0237_INTERNAL_12BIT
+ uint32_t getCostForARMC(int bitDepth) const { return m_costForARMC >> (std::max(0, bitDepth - 10)); } // for header parsing/writing purpose
+#endif
uint32_t getCostForARMC() const { return m_costForARMC; }
#endif
#if JVET_Y0134_TMVP_NAMVP_CAND_REORDERING
@@ -4284,7 +4294,11 @@ public:
void setExtAmvpLevel(int b) { m_extAmvpLevel = b; }
#endif
#if JVET_AA0093_DIVERSITY_CRITERION_FOR_ARMC
+#if JVET_AJ0237_INTERNAL_12BIT
+ void setCostForARMC(uint32_t cost, int bitDepth) { m_costForARMC = (cost << (std::max(0, bitDepth - 10))); }
+#else
void setCostForARMC(uint32_t cost) { m_costForARMC = cost; }
+#endif
#endif
void setBiDirPred( bool b, int refIdx0, int refIdx1 ) { m_biDirPred = b; m_symRefIdx[0] = refIdx0; m_symRefIdx[1] = refIdx1; }
bool getBiDirPred() const { return m_biDirPred; }
diff --git a/source/Lib/CommonLib/TrQuant.cpp b/source/Lib/CommonLib/TrQuant.cpp
index ac707949d..c731c3976 100644
--- a/source/Lib/CommonLib/TrQuant.cpp
+++ b/source/Lib/CommonLib/TrQuant.cpp
@@ -2529,7 +2529,11 @@ void TrQuant::predCoeffSigns(TransformUnit &tu, const ComponentID compID, const
ComponentID residCompID = compID;
bool bJccrWithCr = bIsJCCR && !(tu.jointCbCr >> 1);
#if JVET_AI0096_SIGN_PRED_BIT_DEPTH_FIX
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int signPredShift = SIGN_PRED_RESIDUAL_BITS;
+#else
const int signPredShift = 10 + SIGN_PRED_RESIDUAL_BITS - tu.cs->sps->getBitDepth(toChannelType(COMPONENT_Y));
+#endif
const int signPredOffset = 1 << (signPredShift - 1);
#endif
if(bJccrWithCr)
@@ -2627,14 +2631,23 @@ void TrQuant::predCoeffSigns(TransformUnit &tu, const ComponentID compID, const
int spArea = tu.cs->sps->getSignPredArea();
int signPredWidth = std::min((int)width, spArea);
int signPredHeight = std::min((int)height, spArea);
+#if JVET_AJ0237_INTERNAL_12BIT
+ int16_t* pTemplate = (int16_t*)xMalloc(int16_t, stride * h * w);
+ AreaBuf<int16_t> templateBuf(pTemplate, stride, length, h * w);
+#else
int8_t *pTemplate = (int8_t *) xMalloc(int8_t, stride * h * w);
AreaBuf<int8_t> templateBuf(pTemplate, stride, length, h * w);
+#endif
#else
int8_t *pTemplate = (int8_t *) xMalloc(int8_t, stride * SIGN_PRED_FREQ_RANGE * SIGN_PRED_FREQ_RANGE);
AreaBuf<int8_t> templateBuf(pTemplate, stride, length, SIGN_PRED_FREQ_RANGE * SIGN_PRED_FREQ_RANGE);
#endif
Position prev(0,0);
+#if JVET_AJ0237_INTERNAL_12BIT
+ int16_t* templ = templateBuf.buf;
+#else
int8_t *templ = templateBuf.buf;
+#endif
#if JVET_Y0141_SIGN_PRED_IMPROVE
for (int j = 0; j < signPredHeight*signPredWidth; ++j)
{
@@ -2660,8 +2673,12 @@ void TrQuant::predCoeffSigns(TransformUnit &tu, const ComponentID compID, const
for (uint32_t i = 0; i < height; i++)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ templ[i] = (int16_t)(*pelResi);
+#else
CHECK(*pelResi < -128 || *pelResi > 127, "value exceeds 8-bit range");
templ[i] = (int8_t)(*pelResi);
+#endif
pelResi -= resi.stride;
}
@@ -2669,8 +2686,12 @@ void TrQuant::predCoeffSigns(TransformUnit &tu, const ComponentID compID, const
for (uint32_t i = 0; i < width; i++)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ templ[i + height] = (int16_t)pelResi[i];
+#else
CHECK(pelResi[i] < -128 || pelResi[i] > 127, "value exceeds 8-bit range");
templ[i + height] = (int8_t) pelResi[i];
+#endif
}
#if !JVET_Y0141_SIGN_PRED_IMPROVE
templ += templateBuf.stride;
@@ -2701,9 +2722,15 @@ void TrQuant::predCoeffSigns(TransformUnit &tu, const ComponentID compID, const
PelBuf resi(memTmpResid, width, height);
int signPredHeight = 4;
int signPredWidth = 4;
+#if JVET_AJ0237_INTERNAL_12BIT
+ int16_t* pTemplate = (int16_t*)xMalloc(int16_t, stride * signPredHeight * signPredWidth);
+ AreaBuf<int16_t> templateBuf(pTemplate, stride, length, signPredHeight* signPredWidth);
+ int16_t* templ = templateBuf.buf;
+#else
int8_t *pTemplate = (int8_t *) xMalloc(int8_t, stride * signPredHeight * signPredWidth);
AreaBuf<int8_t> templateBuf(pTemplate, stride, length, signPredHeight * signPredWidth);
int8_t *templ = templateBuf.buf;
+#endif
for (int j = 0; j < signPredHeight*signPredWidth; ++j)
{
coeff.fill(0);
@@ -2721,8 +2748,12 @@ void TrQuant::predCoeffSigns(TransformUnit &tu, const ComponentID compID, const
for (uint32_t i = 0; i < height; i++)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ templ[i] = (int16_t)(*pelResi);
+#else
CHECK(*pelResi < -128 || *pelResi > 127, "value exceeds 8-bit range");
templ[i] = (int8_t)(*pelResi);
+#endif
pelResi -= resi.stride;
}
@@ -2730,8 +2761,12 @@ void TrQuant::predCoeffSigns(TransformUnit &tu, const ComponentID compID, const
for (uint32_t i = 0; i < width; i++)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ templ[i + height] = (int16_t)pelResi[i];
+#else
CHECK(pelResi[i] < -128 || pelResi[i] > 127, "value exceeds 8-bit range");
templ[i + height] = (int8_t) pelResi[i];
+#endif
}
templ += templateBuf.stride;
}
@@ -2838,23 +2873,44 @@ void TrQuant::predCoeffSigns(TransformUnit &tu, const ComponentID compID, const
const uint32_t w = std::min(uiWidth, (uint32_t)SIGN_PRED_FREQ_RANGE);
const uint32_t h = std::min(uiHeight, (uint32_t)SIGN_PRED_FREQ_RANGE);
+#if JVET_AJ0237_INTERNAL_12BIT
+ AreaBuf<const int16_t> templateNormalizedBuf =
+ (lfnstEnabled ? AreaBuf<const int16_t>()
+ : AreaBuf<const int16_t>(g_resiBorderTemplate[log2Width - 2][log2Height - 2][actualTrIdx], stride,
+ length, w * h));
+#else
AreaBuf<const int8_t> templateNormalizedBuf =
(lfnstEnabled ? AreaBuf<const int8_t>()
: AreaBuf<const int8_t>(g_resiBorderTemplate[log2Width - 2][log2Height - 2][actualTrIdx], stride,
length, w * h));
+#endif
#if JVET_AJ0175_NSPT_FOR_NONREG_MODES
bool allowNSPT = CU::isNSPTAllowed( tu, compID, uiWidth, uiHeight, spsIntraLfnstEnabled && CU::isIntra( *( tu.cu ) ) );
int nsptBucketIdx = allowNSPT ? PU::getNSPTBucket(tu) : 0;
+#if JVET_AJ0237_INTERNAL_12BIT
+ AreaBuf<const int16_t> templateLfnstNormalizedBuf =
+ (lfnstEnabled ? AreaBuf<const int16_t>(g_resiBorderTemplateLFNST[nsptBucketIdx][log2Width - 2][log2Height - 2][actualLfnstIdx],
+ stride, length, signPredWidth * signPredHeight)
+ : AreaBuf<const int16_t>());
+#else
AreaBuf<const int8_t> templateLfnstNormalizedBuf =
(lfnstEnabled ? AreaBuf<const int8_t>(g_resiBorderTemplateLFNST[nsptBucketIdx][log2Width - 2][log2Height - 2][actualLfnstIdx],
stride, length, signPredWidth * signPredHeight)
: AreaBuf<const int8_t>());
+#endif
+#else
+#if JVET_AJ0237_INTERNAL_12BIT
+ AreaBuf<const int16_t> templateLfnstNormalizedBuf =
+ (lfnstEnabled ? AreaBuf<const int16_t>(g_resiBorderTemplateLFNST[log2Width - 2][log2Height - 2][actualLfnstIdx],
+ stride, length, signPredWidth * signPredHeight)
+ : AreaBuf<const int16_t>());
#else
AreaBuf<const int8_t> templateLfnstNormalizedBuf =
(lfnstEnabled ? AreaBuf<const int8_t>(g_resiBorderTemplateLFNST[log2Width - 2][log2Height - 2][actualLfnstIdx],
stride, length, signPredWidth * signPredHeight)
: AreaBuf<const int8_t>());
#endif
+#endif
#else
AreaBuf<const int8_t> templateNormalizedBuf(g_resiBorderTemplate[log2Width - 2][log2Height - 2][actualTrIdx], stride,
length, SIGN_PRED_FREQ_RANGE * SIGN_PRED_FREQ_RANGE);
@@ -2915,7 +2971,11 @@ void TrQuant::predCoeffSigns(TransformUnit &tu, const ComponentID compID, const
CHECK(coeffVal == 0, "coefficient value should be nonzero");
#endif
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int16_t* templateBasisVec;
+#else
const int8_t *templateBasisVec;
+#endif
#if JVET_Y0141_SIGN_PRED_IMPROVE
if (lfnstEnabled)
diff --git a/source/Lib/CommonLib/TypeDef.h b/source/Lib/CommonLib/TypeDef.h
index a2ab6604d..d5e901ea3 100644
--- a/source/Lib/CommonLib/TypeDef.h
+++ b/source/Lib/CommonLib/TypeDef.h
@@ -525,6 +525,7 @@
#define JVET_Z0150_MEMORY_USAGE_PRINT 1 // JVET-Z0150: Print memory usage
#define JVET_Z0118_GDR 1 // JVET-Z0118: GDR
#define JVET_AD0169_SMALL_SCALE_DOWNSAMPLING 1 // JVET-AD0169: Downsampling filters in range 1.1 to 1.35 based on Kaiser(7) windowed sinc
+#define JVET_AJ0237_INTERNAL_12BIT 1 // JVET-AJ0237: Modifications for better operation at 12-bit internal bitdepth
#if JVET_Z0118_GDR
#define GDR_LEAK_TEST 0
diff --git a/source/Lib/CommonLib/x86/AdaptiveLoopFilterX86.h b/source/Lib/CommonLib/x86/AdaptiveLoopFilterX86.h
index bd42c0c21..b6f449d96 100644
--- a/source/Lib/CommonLib/x86/AdaptiveLoopFilterX86.h
+++ b/source/Lib/CommonLib/x86/AdaptiveLoopFilterX86.h
@@ -3761,7 +3761,11 @@ static void simdFilter13x13BlkExtDbResiDirect(
adjustShift -= shiftPrecis; // add more precision
}
const int shift = adjustShift;
+#if JVET_AJ0237_INTERNAL_12BIT
+ const Pel currBase = 1 << (clpRng.bd - 1);
+#else
const Pel currBase = 512;
+#endif
int round = 1 << (shift - 1);
#if !( USE_AVX2 && JVET_AJ0188_CODING_INFO_CLASSIFICATION )
@@ -4976,7 +4980,11 @@ static void simdFilter13x13BlkExtDbResi(
adjustShift -= shiftPrecis; // add more precision
}
const int shift = adjustShift;
+#if JVET_AJ0237_INTERNAL_12BIT
+ const Pel currBase = 1 << (clpRng.bd - 1);
+#else
const Pel currBase = 512;
+#endif
int round = 1 << (shift - 1);
#if !( USE_AVX2 && JVET_AJ0188_CODING_INFO_CLASSIFICATION )
@@ -6115,7 +6123,11 @@ static void simdGaussFiltering(CodingStructure &cs, Pel ***gaussPic, const CPelB
int clipIdx = gaussClipIdxTable[filterSetIdx][i];
gaussClipTable[i] = clippingValues[clipIdx];
}
+#if JVET_AJ0237_INTERNAL_12BIT
+ int16_t diffTH = 32 << std::max(0, cs.sps->getBitDepth(CHANNEL_TYPE_LUMA) - 10);
+#else
int16_t diffTH = 32;
+#endif
#if JVET_AJ0188_CODING_INFO_CLASSIFICATION
const bool isIntraSlice = cs.slice->isIntra();
@@ -8372,9 +8384,19 @@ static void simdFixFilter9x9Db9Blk(AlfClassifier **classifier, const CPelBuf &sr
#endif
}
+#if JVET_AJ0237_INTERNAL_12BIT
+template<X86_VEXT vext>
+static void simdDeriveVariance(const CPelBuf& srcLuma, const Area& blkDst, const Area& blk, uint32_t ***variance, int bits)
+#else
template<X86_VEXT vext>
static void simdDeriveVariance(const CPelBuf &srcLuma, const Area &blkDst, const Area &blk, uint32_t ***variance)
+#endif
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ // temporary buffer, could be optimized
+ int64_t tempData[2][(256 + 10) >> 1][((256 + 16) >> 1) + 8] = { { { 0 } } };
+ int bdShift = 2 * std::max(0, bits - 10);
+#endif
const size_t imgStride = srcLuma.stride;
const Pel * srcExt = srcLuma.buf;
int fl = DIST_CLASS;
@@ -8386,16 +8408,19 @@ static void simdDeriveVariance(const CPelBuf &srcLuma, const Area &blkDst, const
int numSample2 = 128 * 128;
int offset = numSample2 >> 1;
+#if !JVET_AJ0237_INTERNAL_12BIT
int num[8]{ numSample, numSample, numSample, numSample, numSample, numSample, numSample, numSample };
int mul[8]{ 13, 13, 13, 13, 13, 13, 13, 13 };
int off[8]{ offset, offset, offset, offset, offset, offset, offset, offset };
+#endif
#if USE_AVX2
if (vext >= AVX2 && (blk.width % 32) == 0)
{
+#if !JVET_AJ0237_INTERNAL_12BIT
__m256i n = _mm256_loadu_si256((__m256i *) num);
__m256i m13 = _mm256_loadu_si256((__m256i *) mul);
__m256i o = _mm256_loadu_si256((__m256i *) off);
-
+#endif
const int posX = blk.pos().x;
const int posY = blk.pos().y;
@@ -8525,6 +8550,41 @@ static void simdDeriveVariance(const CPelBuf &srcLuma, const Area &blkDst, const
if (i == 8)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ for (int kk = 0; kk < 4; kk++)
+ {
+ __m256i x8Low = _mm256_cvtepi32_epi64(_mm_loadu_si128((__m128i*) & variance[2][iOffset - 4][jOffset + kk * 4]));
+ __m256i y8Low = _mm256_cvtepi32_epi64(_mm_loadu_si128((__m128i*) & variance[3][iOffset - 4][jOffset + kk * 4]));
+ __m256i x6Low = _mm256_cvtepi32_epi64(_mm_loadu_si128((__m128i*) & variance[2][iOffset - 3][jOffset + kk * 4]));
+ __m256i y6Low = _mm256_cvtepi32_epi64(_mm_loadu_si128((__m128i*) & variance[3][iOffset - 3][jOffset + kk * 4]));
+ __m256i x4Low = _mm256_cvtepi32_epi64(_mm_loadu_si128((__m128i*) & variance[2][iOffset - 2][jOffset + kk * 4]));
+ __m256i y4Low = _mm256_cvtepi32_epi64(_mm_loadu_si128((__m128i*) & variance[3][iOffset - 2][jOffset + kk * 4]));
+ __m256i x2Low = _mm256_cvtepi32_epi64(_mm_loadu_si128((__m128i*) & variance[2][iOffset - 1][jOffset + kk * 4]));
+ __m256i y2Low = _mm256_cvtepi32_epi64(_mm_loadu_si128((__m128i*) & variance[3][iOffset - 1][jOffset + kk * 4]));
+ __m256i sumLow = _mm256_cvtepi32_epi64(_mm_loadu_si128((__m128i*) & variance[2][iOffset ][jOffset + kk * 4]));
+ __m256i sum2Low = _mm256_cvtepi32_epi64(_mm_loadu_si128((__m128i*) & variance[3][iOffset ][jOffset + kk * 4]));
+
+ x8Low = _mm256_add_epi64(sumLow, x8Low);
+ y8Low = _mm256_add_epi64(sum2Low, y8Low);
+
+ x4Low = _mm256_add_epi64(x6Low, x4Low);
+ y4Low = _mm256_add_epi64(y6Low, y4Low);
+
+ x2Low = _mm256_add_epi64(x8Low, x2Low);
+ y2Low = _mm256_add_epi64(y8Low, y2Low);
+
+ sumLow = _mm256_add_epi64(x4Low, x2Low);
+ sum2Low = _mm256_add_epi64(y4Low, y2Low);
+
+ _mm256_storeu_si256((__m256i*) &tempData[0][iOffset - 4][jOffset + kk * 4], sumLow);
+ _mm256_storeu_si256((__m256i*) &tempData[1][iOffset - 4][jOffset + kk * 4], sum2Low);
+
+ variance[VARIANCE][iOffset - 4][jOffset + kk * 4] = (uint32_t)((13 * ((numSample * tempData[1][iOffset - 4][jOffset + kk * 4] - tempData[0][iOffset - 4][jOffset + kk * 4] * tempData[0][iOffset - 4][jOffset + kk * 4] + offset) >> 3)) >> (14 + bdShift));
+ variance[VARIANCE][iOffset - 4][jOffset + kk * 4 + 1] = (uint32_t)((13 * ((numSample * tempData[1][iOffset - 4][jOffset + kk * 4 + 1] - tempData[0][iOffset - 4][jOffset + kk * 4 + 1] * tempData[0][iOffset - 4][jOffset + kk * 4 + 1] + offset) >> 3)) >> (14 + bdShift));
+ variance[VARIANCE][iOffset - 4][jOffset + kk * 4 + 2] = (uint32_t)((13 * ((numSample * tempData[1][iOffset - 4][jOffset + kk * 4 + 2] - tempData[0][iOffset - 4][jOffset + kk * 4 + 2] * tempData[0][iOffset - 4][jOffset + kk * 4 + 2] + offset) >> 3)) >> (14 + bdShift));
+ variance[VARIANCE][iOffset - 4][jOffset + kk * 4 + 3] = (uint32_t)((13 * ((numSample * tempData[1][iOffset - 4][jOffset + kk * 4 + 3] - tempData[0][iOffset - 4][jOffset + kk * 4 + 3] * tempData[0][iOffset - 4][jOffset + kk * 4 + 3] + offset) >> 3)) >> (14 + bdShift));
+ }
+#else
x8 = _mm256_loadu_si256((__m256i *)&variance[2][iOffset - 4][jOffset]);
y8 = _mm256_loadu_si256((__m256i *)&variance[3][iOffset - 4][jOffset]);
x6 = _mm256_loadu_si256((__m256i *)&variance[2][iOffset - 3][jOffset]);
@@ -8583,9 +8643,37 @@ static void simdDeriveVariance(const CPelBuf &srcLuma, const Area &blkDst, const
summ2 = _mm256_srli_epi32(summ2, 14);
_mm256_storeu_si256((__m256i *) &variance[VARIANCE][iOffset - 4][jOffset], sum2);
_mm256_storeu_si256((__m256i *) &variance[VARIANCE][iOffset - 4][jOffset + 8], summ2);
+#endif
}
else if (i > 8)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ for (int kk = 0; kk < 4; kk++)
+ {
+ __m256i x8Low = _mm256_cvtepi32_epi64(_mm_loadu_si128((__m128i*) &variance[2][iOffset - 5][jOffset + kk * 4]));
+ __m256i y8Low = _mm256_cvtepi32_epi64(_mm_loadu_si128((__m128i*) &variance[3][iOffset - 5][jOffset + kk * 4]));
+
+ __m256i x6Low = _mm256_loadu_si256((__m256i*) &tempData[0][iOffset - 5][jOffset + kk * 4]);
+ __m256i y6Low = _mm256_loadu_si256((__m256i*) &tempData[1][iOffset - 5][jOffset + kk * 4]);
+
+ __m256i sumLow = _mm256_cvtepi32_epi64(_mm_loadu_si128((__m128i*) &variance[2][iOffset][jOffset + kk * 4]));
+ __m256i sum2Low = _mm256_cvtepi32_epi64(_mm_loadu_si128((__m128i*) &variance[3][iOffset][jOffset + kk * 4]));
+
+ x6Low = _mm256_sub_epi64(x6Low, x8Low);
+ y6Low = _mm256_sub_epi64(y6Low, y8Low);
+
+ sumLow = _mm256_add_epi64(x6Low, sumLow);
+ sum2Low = _mm256_add_epi64(y6Low, sum2Low);
+
+ _mm256_storeu_si256((__m256i*)& tempData[0][iOffset - 4][jOffset + kk * 4], sumLow);
+ _mm256_storeu_si256((__m256i*)& tempData[1][iOffset - 4][jOffset + kk * 4], sum2Low);
+
+ variance[VARIANCE][iOffset - 4][jOffset + kk * 4] = (uint32_t)((13 * ((numSample * tempData[1][iOffset - 4][jOffset + kk * 4] - tempData[0][iOffset - 4][jOffset + kk * 4] * tempData[0][iOffset - 4][jOffset + kk * 4] + offset) >> 3)) >> (14 + bdShift));
+ variance[VARIANCE][iOffset - 4][jOffset + kk * 4 + 1] = (uint32_t)((13 * ((numSample * tempData[1][iOffset - 4][jOffset + kk * 4 + 1] - tempData[0][iOffset - 4][jOffset + kk * 4 + 1] * tempData[0][iOffset - 4][jOffset + kk * 4 + 1] + offset) >> 3)) >> (14 + bdShift));
+ variance[VARIANCE][iOffset - 4][jOffset + kk * 4 + 2] = (uint32_t)((13 * ((numSample * tempData[1][iOffset - 4][jOffset + kk * 4 + 2] - tempData[0][iOffset - 4][jOffset + kk * 4 + 2] * tempData[0][iOffset - 4][jOffset + kk * 4 + 2] + offset) >> 3)) >> (14 + bdShift));
+ variance[VARIANCE][iOffset - 4][jOffset + kk * 4 + 3] = (uint32_t)((13 * ((numSample * tempData[1][iOffset - 4][jOffset + kk * 4 + 3] - tempData[0][iOffset - 4][jOffset + kk * 4 + 3] * tempData[0][iOffset - 4][jOffset + kk * 4 + 3] + offset) >> 3)) >> (14 + bdShift));
+ }
+#else
x8 = _mm256_loadu_si256((__m256i *)&variance[2][iOffset - 5][jOffset]);
xx8 = _mm256_loadu_si256((__m256i *)&variance[2][iOffset - 5][jOffset + 8]);
y8 = _mm256_loadu_si256((__m256i *)&variance[3][iOffset - 5][jOffset]);
@@ -8625,6 +8713,7 @@ static void simdDeriveVariance(const CPelBuf &srcLuma, const Area &blkDst, const
summ2 = _mm256_srli_epi32(summ2, 14);
_mm256_storeu_si256((__m256i *) &variance[VARIANCE][iOffset - 4][jOffset], sum2);
_mm256_storeu_si256((__m256i *) &variance[VARIANCE][iOffset - 4][jOffset + 8], summ2);
+#endif
}
}
@@ -8633,10 +8722,11 @@ static void simdDeriveVariance(const CPelBuf &srcLuma, const Area &blkDst, const
else
{
#endif
+#if !JVET_AJ0237_INTERNAL_12BIT
__m128i n = _mm_loadu_si128((__m128i *) num);
__m128i m13 = _mm_loadu_si128((__m128i *) mul);
__m128i o = _mm_loadu_si128((__m128i *) off);
-
+#endif
const int posX = blk.pos().x;
const int posY = blk.pos().y;
@@ -8706,6 +8796,39 @@ static void simdDeriveVariance(const CPelBuf &srcLuma, const Area &blkDst, const
if (i == 8)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ for (int kk = 0; kk < 2; kk++)
+ {
+ __m128i x8Low = _mm_cvtepi32_epi64(_mm_loadu_si128((__m128i*) & variance[2][iOffset - 4][jOffset + kk * 2]));
+ __m128i y8Low = _mm_cvtepi32_epi64(_mm_loadu_si128((__m128i*) & variance[3][iOffset - 4][jOffset + kk * 2]));
+ __m128i x6Low = _mm_cvtepi32_epi64(_mm_loadu_si128((__m128i*) & variance[2][iOffset - 3][jOffset + kk * 2]));
+ __m128i y6Low = _mm_cvtepi32_epi64(_mm_loadu_si128((__m128i*) & variance[3][iOffset - 3][jOffset + kk * 2]));
+ __m128i x4Low = _mm_cvtepi32_epi64(_mm_loadu_si128((__m128i*) & variance[2][iOffset - 2][jOffset + kk * 2]));
+ __m128i y4Low = _mm_cvtepi32_epi64(_mm_loadu_si128((__m128i*) & variance[3][iOffset - 2][jOffset + kk * 2]));
+ __m128i x2Low = _mm_cvtepi32_epi64(_mm_loadu_si128((__m128i*) & variance[2][iOffset - 1][jOffset + kk * 2]));
+ __m128i y2Low = _mm_cvtepi32_epi64(_mm_loadu_si128((__m128i*) & variance[3][iOffset - 1][jOffset + kk * 2]));
+ __m128i sumLow = _mm_cvtepi32_epi64(_mm_loadu_si128((__m128i*) & variance[2][iOffset ][jOffset + kk * 2]));
+ __m128i sum2Low = _mm_cvtepi32_epi64(_mm_loadu_si128((__m128i*) & variance[3][iOffset ][jOffset + kk * 2]));
+
+ x8Low = _mm_add_epi64(sumLow, x8Low);
+ y8Low = _mm_add_epi64(sum2Low, y8Low);
+
+ x4Low = _mm_add_epi64(x6Low, x4Low);
+ y4Low = _mm_add_epi64(y6Low, y4Low);
+
+ x2Low = _mm_add_epi64(x8Low, x2Low);
+ y2Low = _mm_add_epi64(y8Low, y2Low);
+
+ sumLow = _mm_add_epi64(x4Low, x2Low);
+ sum2Low = _mm_add_epi64(y4Low, y2Low);
+
+ _mm_storeu_si128((__m128i*) &tempData[0][iOffset - 4][jOffset + kk * 2], sumLow);
+ _mm_storeu_si128((__m128i*) &tempData[1][iOffset - 4][jOffset + kk * 2], sum2Low);
+
+ variance[VARIANCE][iOffset - 4][jOffset + kk * 2] = (uint32_t)((13 * ((numSample * tempData[1][iOffset - 4][jOffset + kk * 2] - tempData[0][iOffset - 4][jOffset + kk * 2] * tempData[0][iOffset - 4][jOffset + kk * 2] + offset) >> 3)) >> (14 + bdShift));
+ variance[VARIANCE][iOffset - 4][jOffset + kk * 2 + 1] = (uint32_t)((13 * ((numSample * tempData[1][iOffset - 4][jOffset + kk * 2 + 1] - tempData[0][iOffset - 4][jOffset + kk * 2 + 1] * tempData[0][iOffset - 4][jOffset + kk * 2 + 1] + offset) >> 3)) >> (14 + bdShift));
+ }
+#else
x8 = _mm_loadu_si128((__m128i *)&variance[2][iOffset - 4][jOffset]);
y8 = _mm_loadu_si128((__m128i *)&variance[3][iOffset - 4][jOffset]);
x6 = _mm_loadu_si128((__m128i *)&variance[2][iOffset - 3][jOffset]);
@@ -8737,9 +8860,33 @@ static void simdDeriveVariance(const CPelBuf &srcLuma, const Area &blkDst, const
sum2 = _mm_mullo_epi32(sum2, m13);
sum2 = _mm_srli_epi32(sum2, 14);
_mm_storeu_si128((__m128i *) &variance[VARIANCE][iOffset - 4][jOffset], sum2);
+#endif
}
else if (i > 8)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ for (int kk = 0; kk < 2; kk++)
+ {
+ __m128i x8Low = _mm_cvtepi32_epi64(_mm_loadu_si128((__m128i*) & variance[2][iOffset - 5][jOffset + kk * 2]));
+ __m128i y8Low = _mm_cvtepi32_epi64(_mm_loadu_si128((__m128i*) & variance[3][iOffset - 5][jOffset + kk * 2]));
+ __m128i x6Low = _mm_loadu_si128((__m128i*) &tempData[0][iOffset - 5][jOffset + kk * 2]);
+ __m128i y6Low = _mm_loadu_si128((__m128i*) &tempData[1][iOffset - 5][jOffset + kk * 2]);
+ __m128i sumLow = _mm_cvtepi32_epi64(_mm_loadu_si128((__m128i*) & variance[2][iOffset][jOffset + kk * 2]));
+ __m128i sum2Low = _mm_cvtepi32_epi64(_mm_loadu_si128((__m128i*) & variance[3][iOffset][jOffset + kk * 2]));
+
+ x6Low = _mm_sub_epi32(x6Low, x8Low);
+ y6Low = _mm_sub_epi32(y6Low, y8Low);
+
+ sumLow = _mm_add_epi32(x6Low, sumLow);
+ sum2Low = _mm_add_epi32(y6Low, sum2Low);
+
+ _mm_storeu_si128((__m128i*) & tempData[0][iOffset - 4][jOffset + kk * 2], sumLow);
+ _mm_storeu_si128((__m128i*) & tempData[1][iOffset - 4][jOffset + kk * 2], sum2Low);
+
+ variance[VARIANCE][iOffset - 4][jOffset + kk * 2] = (uint32_t)((13 * ((numSample * tempData[1][iOffset - 4][jOffset + kk * 2] - tempData[0][iOffset - 4][jOffset + kk * 2] * tempData[0][iOffset - 4][jOffset + kk * 2] + offset) >> 3)) >> (14 + bdShift));
+ variance[VARIANCE][iOffset - 4][jOffset + kk * 2 + 1] = (uint32_t)((13 * ((numSample * tempData[1][iOffset - 4][jOffset + kk * 2 + 1] - tempData[0][iOffset - 4][jOffset + kk * 2 + 1] * tempData[0][iOffset - 4][jOffset + kk * 2 + 1] + offset) >> 3)) >> (14 + bdShift));
+ }
+#else
x8 = _mm_loadu_si128((__m128i *)&variance[2][iOffset - 5][jOffset]);
y8 = _mm_loadu_si128((__m128i *)&variance[3][iOffset - 5][jOffset]);
x6 = _mm_loadu_si128((__m128i *)&variance[0][iOffset - 5][jOffset]);
@@ -8761,6 +8908,7 @@ static void simdDeriveVariance(const CPelBuf &srcLuma, const Area &blkDst, const
sum2 = _mm_mullo_epi32(sum2, m13);
sum2 = _mm_srli_epi32(sum2, 14);
_mm_storeu_si128((__m128i *) &variance[VARIANCE][iOffset - 4][jOffset], sum2);
+#endif
}
}
diff --git a/source/Lib/CommonLib/x86/BilateralFilterX86.h b/source/Lib/CommonLib/x86/BilateralFilterX86.h
index 9e3262765..2c60e66cc 100644
--- a/source/Lib/CommonLib/x86/BilateralFilterX86.h
+++ b/source/Lib/CommonLib/x86/BilateralFilterX86.h
@@ -48,7 +48,11 @@
#if USE_AVX2
#if JVET_AF0112_BIF_DYNAMIC_SCALING
+#if JVET_AJ0237_INTERNAL_12BIT
+inline void simdBifApplyLut(__m256i& val, __m256i& acc, int cutBitsNum, __m256i& bitsRound, __m256i& bitsRound2, __m256i& lut, int bdShift)
+#else
inline void simdBifApplyLut(__m256i& val, __m256i& acc, int cutBitsNum, __m256i& bitsRound, __m256i& bitsRound2, __m256i& lut)
+#endif
#else
inline void simdBifApplyLut(__m256i& val, __m256i& acc, __m256i& lut, int lutShift)
#endif
@@ -74,23 +78,40 @@ inline void simdBifApplyLut(__m256i& val, __m256i& acc, __m256i& lut, int lutShi
diffabs = _mm256_shuffle_epi8(lut, diffabs); /* lut */
diffabs = _mm256_cvtepi8_epi16(_mm256_castsi256_si128(diffabs)); /* back to 16-bit */
diffabs = _mm256_srai_epi16(diffabs, lutShift); /* diagonal shift! */
+#endif
+#if JVET_AJ0237_INTERNAL_12BIT
+ diffabs = _mm256_slli_epi16(diffabs, bdShift);
#endif
diffabs = _mm256_sign_epi16(diffabs, val); /* add original sign */
acc = _mm256_add_epi16(diffabs, acc); /* add to acc */
}
+#if JVET_AJ0237_INTERNAL_12BIT
+template<X86_VEXT vext>
+void BilateralFilter::simdFilterDiamond5x5(uint32_t uiWidth, uint32_t uiHeight, int16_t block[], int16_t blkFilt[], const ClpRng& clpRng, Pel* recPtr, int recStride, int iWidthExtSIMD, int bfac, int bifRoundAdd, int bifRoundShift, bool isRDO, const char* lutRowPtr, bool noClip, int cutBitsNum, int bdShift)
+#else
template<X86_VEXT vext>
void BilateralFilter::simdFilterDiamond5x5(uint32_t uiWidth, uint32_t uiHeight, int16_t block[], int16_t blkFilt[], const ClpRng& clpRng, Pel* recPtr, int recStride, int iWidthExtSIMD, int bfac, int bifRoundAdd, int bifRoundShift, bool isRDO, const char* lutRowPtr, bool noClip, int cutBitsNum)
+#endif
{
//if( uiWidth < 4 || ( uiWidth < 8 && isRDO ) )
if (uiWidth < 4)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ return blockBilateralFilterDiamond5x5(uiWidth, uiHeight, block, blkFilt, clpRng, recPtr, recStride, iWidthExtSIMD, bfac, bifRoundAdd, bifRoundShift, isRDO, lutRowPtr, noClip, cutBitsNum, bdShift);
+#else
return blockBilateralFilterDiamond5x5(uiWidth, uiHeight, block, blkFilt, clpRng, recPtr, recStride, iWidthExtSIMD, bfac, bifRoundAdd, bifRoundShift, isRDO, lutRowPtr, noClip, cutBitsNum);
+#endif
}
int pad = 2;
int padwidth = iWidthExtSIMD;
+#if JVET_AJ0237_INTERNAL_12BIT
+ cutBitsNum += bdShift;
+#endif
+
+
__m256i center, left, right, up, down, lu, ld, ru, rd, acc, roundAdd, clipmin, clipmax, inputVals;
__m256i ll, rr, uu, dd;
__m128i lutTmp;
@@ -106,7 +127,11 @@ void BilateralFilter::simdFilterDiamond5x5(uint32_t uiWidth, uint32_t uiHeight,
__m256i lut2 = _mm256_set_m128i(lutTmp, lutTmp);
lutTmp = _mm_loadu_si128((__m128i*)(lutRowPtr + 32));
__m256i lut3 = _mm256_set_m128i(lutTmp, lutTmp);
+#if JVET_AJ0237_INTERNAL_12BIT
+ __m256i mmBfac = _mm256_unpacklo_epi16(_mm256_set1_epi16(bfac), _mm256_set1_epi16(1));
+#else
__m256i mmBfac = _mm256_set1_epi16(bfac);
+#endif
roundAdd = _mm256_set1_epi16(bifRoundAdd << 3);
__m256i bitsRound = _mm256_set1_epi16(1 << (cutBitsNum - 2));
__m256i bitsRound2 = _mm256_set1_epi16((1 << (cutBitsNum - 2)) + (1 << (cutBitsNum - 1)));
@@ -160,6 +185,22 @@ void BilateralFilter::simdFilterDiamond5x5(uint32_t uiWidth, uint32_t uiHeight,
// apply LUT
#if JVET_AF0112_BIF_DYNAMIC_SCALING
+#if JVET_AJ0237_INTERNAL_12BIT
+ simdBifApplyLut(left, acc, cutBitsNum, bitsRound, bitsRound2, lut1, bdShift);
+ simdBifApplyLut(right, acc, cutBitsNum, bitsRound, bitsRound2, lut1, bdShift);
+ simdBifApplyLut(up, acc, cutBitsNum, bitsRound, bitsRound2, lut1, bdShift);
+ simdBifApplyLut(down, acc, cutBitsNum, bitsRound, bitsRound2, lut1, bdShift);
+
+ simdBifApplyLut(lu, acc, cutBitsNum, bitsRound, bitsRound2, lut2, bdShift);
+ simdBifApplyLut(ld, acc, cutBitsNum, bitsRound, bitsRound2, lut2, bdShift);
+ simdBifApplyLut(ru, acc, cutBitsNum, bitsRound, bitsRound2, lut2, bdShift);
+ simdBifApplyLut(rd, acc, cutBitsNum, bitsRound, bitsRound2, lut2, bdShift);
+
+ simdBifApplyLut(ll, acc, cutBitsNum, bitsRound, bitsRound2, lut3, bdShift);
+ simdBifApplyLut(rr, acc, cutBitsNum, bitsRound, bitsRound2, lut3, bdShift);
+ simdBifApplyLut(uu, acc, cutBitsNum, bitsRound, bitsRound2, lut3, bdShift);
+ simdBifApplyLut(dd, acc, cutBitsNum, bitsRound, bitsRound2, lut3, bdShift);
+#else
simdBifApplyLut(left, acc, cutBitsNum, bitsRound, bitsRound2, lut1);
simdBifApplyLut(right, acc, cutBitsNum, bitsRound, bitsRound2, lut1);
simdBifApplyLut(up, acc, cutBitsNum, bitsRound, bitsRound2, lut1);
@@ -174,6 +215,7 @@ void BilateralFilter::simdFilterDiamond5x5(uint32_t uiWidth, uint32_t uiHeight,
simdBifApplyLut(rr, acc, cutBitsNum, bitsRound, bitsRound2, lut3);
simdBifApplyLut(uu, acc, cutBitsNum, bitsRound, bitsRound2, lut3);
simdBifApplyLut(dd, acc, cutBitsNum, bitsRound, bitsRound2, lut3);
+#endif
#else
simdBifApplyLut(left, acc, lut, lutShift1);
simdBifApplyLut(right, acc, lut, lutShift1);
@@ -193,9 +235,21 @@ void BilateralFilter::simdFilterDiamond5x5(uint32_t uiWidth, uint32_t uiHeight,
// TU scaling
#if JVET_AF0112_BIF_DYNAMIC_SCALING
+#if JVET_AJ0237_INTERNAL_12BIT
+ __m256i accLow = _mm256_unpacklo_epi16(acc, roundAdd);
+ __m256i accHigh = _mm256_unpackhi_epi16(acc, roundAdd);
+ __m256i accLowPack = _mm256_madd_epi16(accLow, mmBfac);
+ __m256i accHighPack = _mm256_madd_epi16(accHigh, mmBfac);
+
+ accLow = _mm256_srai_epi32(accLowPack, bifRoundShift + 3);
+ accHigh = _mm256_srai_epi32(accHighPack, bifRoundShift + 3);
+
+ acc = _mm256_packs_epi32(accLow, accHigh);
+#else
acc = _mm256_mullo_epi16(acc, mmBfac);
acc = _mm256_adds_epi16(acc, roundAdd);
acc = _mm256_srai_epi16(acc, bifRoundShift + 3);
+#endif
#else
if (bfac == 2)
{
@@ -293,7 +347,11 @@ int BilateralFilter::simdCalcMAD(int16_t* block, int stride, int width, int heig
#else // USE_AVX2
#if JVET_AF0112_BIF_DYNAMIC_SCALING
+#if JVET_AJ0237_INTERNAL_12BIT
+inline void simdBifApplyLut(__m128i& val, __m128i& acc, int cutBitsNum, __m128i& bitsRound, __m128i& bitsRound2, __m128i& lut, int bdShift)
+#else
inline void simdBifApplyLut(__m128i& val, __m128i& acc, int cutBitsNum, __m128i& bitsRound, __m128i& bitsRound2, __m128i& lut)
+#endif
#else
inline void simdBifApplyLut(__m128i& val, __m128i& acc, __m128i& lut, int lutShift)
#endif
@@ -318,23 +376,39 @@ inline void simdBifApplyLut(__m128i& val, __m128i& acc, __m128i& lut, int lutShi
diffabs = _mm_shuffle_epi8(lut, diffabs); /* lut */
diffabs = _mm_cvtepi8_epi16(diffabs); /* back to 16-bit */
diffabs = _mm_srai_epi16(diffabs, lutShift); /* diagonal shift! */
+#endif
+#if JVET_AJ0237_INTERNAL_12BIT
+ diffabs = _mm_slli_epi16(diffabs, bdShift);
#endif
diffabs = _mm_sign_epi16(diffabs, val); /* add original sign */
acc = _mm_add_epi16(diffabs, acc); /* add to acc */
}
+#if JVET_AJ0237_INTERNAL_12BIT
+template<X86_VEXT vext>
+void BilateralFilter::simdFilterDiamond5x5(uint32_t uiWidth, uint32_t uiHeight, int16_t block[], int16_t blkFilt[], const ClpRng& clpRng, Pel* recPtr, int recStride, int iWidthExtSIMD, int bfac, int bifRoundAdd, int bifRoundShift, bool isRDO, const char* lutRowPtr, bool noClip, int cutBitsNum, int bdShift)
+#else
template<X86_VEXT vext>
void BilateralFilter::simdFilterDiamond5x5(uint32_t uiWidth, uint32_t uiHeight, int16_t block[], int16_t blkFilt[], const ClpRng& clpRng, Pel* recPtr, int recStride, int iWidthExtSIMD, int bfac, int bifRoundAdd, int bifRoundShift, bool isRDO, const char* lutRowPtr, bool noClip, int cutBitsNum)
+#endif
{
//if( uiWidth < 4 || ( uiWidth < 8 && isRDO ) )
if( uiWidth < 4 )
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ return blockBilateralFilterDiamond5x5(uiWidth, uiHeight, block, blkFilt, clpRng, recPtr, recStride, iWidthExtSIMD, bfac, bifRoundAdd, bifRoundShift, isRDO, lutRowPtr, noClip, cutBitsNum, bdShift);
+#else
return blockBilateralFilterDiamond5x5(uiWidth, uiHeight, block, blkFilt, clpRng, recPtr, recStride, iWidthExtSIMD, bfac, bifRoundAdd, bifRoundShift, isRDO, lutRowPtr, noClip, cutBitsNum);
+#endif
}
int pad = 2;
int padwidth = iWidthExtSIMD;
+#if JVET_AJ0237_INTERNAL_12BIT
+ cutBitsNum += bdShift;
+#endif
+
__m128i center, left, right, up, down, lu, ld, ru, rd, acc, roundAdd, clipmin, clipmax, inputVals;
__m128i ll, rr, uu, dd;
@@ -346,7 +420,11 @@ void BilateralFilter::simdFilterDiamond5x5(uint32_t uiWidth, uint32_t uiHeight,
__m128i lut1 = _mm_loadu_si128((__m128i*)(lutRowPtr));
__m128i lut2 = _mm_loadu_si128((__m128i*)(lutRowPtr + 16));
__m128i lut3 = _mm_loadu_si128((__m128i*)(lutRowPtr + 32));
+#if JVET_AJ0237_INTERNAL_12BIT
+ __m128i mmBfac = _mm_unpacklo_epi16(_mm_set1_epi16(bfac), _mm_set1_epi16(1));
+#else
__m128i mmBfac = _mm_set1_epi16(bfac);
+#endif
roundAdd = _mm_set1_epi16(bifRoundAdd << 3);
__m128i bitsRound = _mm_set1_epi16(1 << (cutBitsNum - 2));
__m128i bitsRound2 = _mm_set1_epi16((1 << (cutBitsNum - 2)) + (1 << (cutBitsNum - 1)));
@@ -399,6 +477,22 @@ void BilateralFilter::simdFilterDiamond5x5(uint32_t uiWidth, uint32_t uiHeight,
// apply LUT
#if JVET_AF0112_BIF_DYNAMIC_SCALING
+#if JVET_AJ0237_INTERNAL_12BIT
+ simdBifApplyLut(left, acc, cutBitsNum, bitsRound, bitsRound2, lut1, bdShift);
+ simdBifApplyLut(right, acc, cutBitsNum, bitsRound, bitsRound2, lut1, bdShift);
+ simdBifApplyLut(up, acc, cutBitsNum, bitsRound, bitsRound2, lut1, bdShift);
+ simdBifApplyLut(down, acc, cutBitsNum, bitsRound, bitsRound2, lut1, bdShift);
+
+ simdBifApplyLut(lu, acc, cutBitsNum, bitsRound, bitsRound2, lut2, bdShift);
+ simdBifApplyLut(ld, acc, cutBitsNum, bitsRound, bitsRound2, lut2, bdShift);
+ simdBifApplyLut(ru, acc, cutBitsNum, bitsRound, bitsRound2, lut2, bdShift);
+ simdBifApplyLut(rd, acc, cutBitsNum, bitsRound, bitsRound2, lut2, bdShift);
+
+ simdBifApplyLut(ll, acc, cutBitsNum, bitsRound, bitsRound2, lut3, bdShift);
+ simdBifApplyLut(rr, acc, cutBitsNum, bitsRound, bitsRound2, lut3, bdShift);
+ simdBifApplyLut(uu, acc, cutBitsNum, bitsRound, bitsRound2, lut3, bdShift);
+ simdBifApplyLut(dd, acc, cutBitsNum, bitsRound, bitsRound2, lut3, bdShift);
+#else
simdBifApplyLut(left, acc, cutBitsNum, bitsRound, bitsRound2, lut1);
simdBifApplyLut(right, acc, cutBitsNum, bitsRound, bitsRound2, lut1);
simdBifApplyLut(up, acc, cutBitsNum, bitsRound, bitsRound2, lut1);
@@ -413,6 +507,7 @@ void BilateralFilter::simdFilterDiamond5x5(uint32_t uiWidth, uint32_t uiHeight,
simdBifApplyLut(rr, acc, cutBitsNum, bitsRound, bitsRound2, lut3);
simdBifApplyLut(uu, acc, cutBitsNum, bitsRound, bitsRound2, lut3);
simdBifApplyLut(dd, acc, cutBitsNum, bitsRound, bitsRound2, lut3);
+#endif
#else
simdBifApplyLut(left, acc, lut, lutShift1);
simdBifApplyLut(right, acc, lut, lutShift1);
@@ -432,9 +527,21 @@ void BilateralFilter::simdFilterDiamond5x5(uint32_t uiWidth, uint32_t uiHeight,
// TU scaling
#if JVET_AF0112_BIF_DYNAMIC_SCALING
+#if JVET_AJ0237_INTERNAL_12BIT
+ __m128i accLow = _mm_unpacklo_epi16(acc, roundAdd);
+ __m128i accHigh = _mm_unpackhi_epi16(acc, roundAdd);
+ __m128i accLowPack = _mm_madd_epi16(accLow, mmBfac);
+ __m128i accHighPack = _mm_madd_epi16(accHigh, mmBfac);
+
+ accLow = _mm_srai_epi32(accLowPack, bifRoundShift + 3);
+ accHigh = _mm_srai_epi32(accHighPack, bifRoundShift + 3);
+
+ acc = _mm_packs_epi32(accLow, accHigh);
+#else
acc = _mm_mullo_epi16(acc, mmBfac);
acc = _mm_adds_epi16(acc, roundAdd);
acc = _mm_srai_epi16(acc, bifRoundShift + 3);
+#endif
#else
if (bfac == 2)
{
diff --git a/source/Lib/CommonLib/x86/BufferX86.h b/source/Lib/CommonLib/x86/BufferX86.h
index 5730bd2ec..b5028b225 100644
--- a/source/Lib/CommonLib/x86/BufferX86.h
+++ b/source/Lib/CommonLib/x86/BufferX86.h
@@ -3907,7 +3907,11 @@ void getAbsoluteDifferencePerSample_SSE(Pel* dst, int dstStride, const Pel* src0
template <X86_VEXT vext, uint8_t maskType>
int64_t getMaskedSampleSum_SSE(Pel* src, int srcStride, int width, int height, int bitDepth, short* weightMask, int maskStepX, int maskStride, int maskStride2)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ if ((width & 7) != 0 || bitDepth > 12)
+#else
if ((width & 7) != 0 || bitDepth > 10)
+#endif
{
return getMaskedSampleSumCore<maskType>(src, srcStride, width, height, bitDepth, weightMask, maskStepX, maskStride, maskStride2);
}
diff --git a/source/Lib/CommonLib/x86/InterpolationFilterX86.h b/source/Lib/CommonLib/x86/InterpolationFilterX86.h
index 0d92aca83..039b840f0 100644
--- a/source/Lib/CommonLib/x86/InterpolationFilterX86.h
+++ b/source/Lib/CommonLib/x86/InterpolationFilterX86.h
@@ -79,6 +79,11 @@ static void fullPelCopySSE( const ClpRng& clpRng, const void*_src, int srcStride
#endif
int headroom_offset = 1 << ( headroom - 1 );
int offset = IF_INTERNAL_OFFS;
+
+#if JVET_AJ0237_INTERNAL_12BIT
+ int dmvrHeadRoom = IF_INTERNAL_PREC_BILINEAR(clpRng.bd) - clpRng.bd; // in the current setup, dmvr headroom should either be 0 or negative
+#endif
+
__m128i voffset = _mm_set1_epi16( offset );
__m128i voffset_headroom = _mm_set1_epi16( headroom_offset );
@@ -106,7 +111,29 @@ static void fullPelCopySSE( const ClpRng& clpRng, const void*_src, int srcStride
#if MCIF_SIMD_NEW
if (biMCForDMVR)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ if ((isFirst == isLast) || (isFirst && dmvrHeadRoom == 0))
+ {
+ vsum = vsrc;
+ }
+ else if (isFirst)
+ {
+ if (dmvrHeadRoom > 0)
+ {
+ vsum = _mm_slli_epi16(vsrc, dmvrHeadRoom);
+ }
+ else
+ {
+ vsum = _mm_srai_epi16(vsrc, -dmvrHeadRoom);
+ }
+ }
+ else
+ {
+ CHECK(1, "Impossible to have isFirst being false and isLast being true, when biMCForDMVR is true");
+ }
+#else
vsum = _mm_min_epi16(vibdimax, _mm_max_epi16(vibdimin, vsrc));
+#endif
}
else if (isFirst == isLast)
{
@@ -152,6 +179,10 @@ static void fullPelCopyVerSSE(const ClpRng& clpRng, const void*_src, int srcStri
int headroom = IF_INTERNAL_PREC - clpRng.bd;
int headroom_offset = 1 << (headroom - 1);
int offset = IF_INTERNAL_OFFS;
+#if JVET_AJ0237_INTERNAL_12BIT
+ int dmvrHeadRoom = IF_INTERNAL_PREC_BILINEAR(clpRng.bd) - clpRng.bd; // in the current setup, dmvr headroom should either be 0 or negative
+#endif
+
__m128i voffset = _mm_set1_epi16(offset);
__m128i voffset_headroom = _mm_set1_epi16(headroom_offset);
@@ -176,7 +207,29 @@ static void fullPelCopyVerSSE(const ClpRng& clpRng, const void*_src, int srcStri
}
if (biMCForDMVR)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ if ((isFirst == isLast) || (isFirst && dmvrHeadRoom == 0))
+ {
+ vsum = vsrc;
+ }
+ else if (isFirst)
+ {
+ if (dmvrHeadRoom > 0)
+ {
+ vsum = _mm_slli_epi16(vsrc, dmvrHeadRoom);
+ }
+ else
+ {
+ vsum = _mm_srai_epi16(vsrc, -dmvrHeadRoom);
+ }
+ }
+ else
+ {
+ CHECK(1, "Impossible to have isFirst being false and isLast being true, when biMCForDMVR is true");
+ }
+#else
vsum = _mm_min_epi16(vibdimax, _mm_max_epi16(vibdimin, vsrc));
+#endif
}
else if (isFirst == isLast)
{
@@ -218,6 +271,10 @@ static void fullPelCopySSE_M4(const ClpRng& clpRng, const void*_src, ptrdiff_t s
int headroom = IF_INTERNAL_PREC - clpRng.bd;
int headroom_offset = 1 << (headroom - 1);
int offset = IF_INTERNAL_OFFS;
+#if JVET_AJ0237_INTERNAL_12BIT
+ int dmvrHeadRoom = IF_INTERNAL_PREC_BILINEAR(clpRng.bd) - clpRng.bd; // in the current setup, dmvr headroom should either be 0 or negative
+#endif
+
__m128i voffset = _mm_set1_epi16(offset);
__m128i voffset_headroom = _mm_set1_epi16(headroom_offset);
__m128i vibdimin = _mm_set1_epi16(clpRng.min);
@@ -243,7 +300,29 @@ static void fullPelCopySSE_M4(const ClpRng& clpRng, const void*_src, ptrdiff_t s
}
if (biMCForDMVR)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ if ((isFirst == isLast) || (isFirst && dmvrHeadRoom == 0))
+ {
+ vsum = vsrc;
+ }
+ else if (isFirst)
+ {
+ if (dmvrHeadRoom > 0)
+ {
+ vsum = _mm_slli_epi16(vsrc, dmvrHeadRoom);
+ }
+ else
+ {
+ vsum = _mm_srai_epi16(vsrc, -dmvrHeadRoom);
+ }
+ }
+ else
+ {
+ CHECK(1, "Impossible to have isFirst being false and isLast being true, when biMCForDMVR is true");
+ }
+#else
vsum = _mm_min_epi16(vibdimax, _mm_max_epi16(vibdimin, vsrc));
+#endif
}
else if (isFirst == isLast)
{
@@ -280,6 +359,10 @@ static void fullPelCopyVerSSE_M4(const ClpRng& clpRng, const void*_src, ptrdiff_
int headroom = IF_INTERNAL_PREC - clpRng.bd;
int headroom_offset = 1 << (headroom - 1);
int offset = IF_INTERNAL_OFFS;
+#if JVET_AJ0237_INTERNAL_12BIT
+ int dmvrHeadRoom = IF_INTERNAL_PREC_BILINEAR(clpRng.bd) - clpRng.bd; // in the current setup, dmvr headroom should either be 0 or negative
+#endif
+
__m128i voffset = _mm_set1_epi16(offset);
__m128i voffset_headroom = _mm_set1_epi16(headroom_offset);
__m128i vibdimin = _mm_set1_epi16(clpRng.min);
@@ -301,7 +384,29 @@ static void fullPelCopyVerSSE_M4(const ClpRng& clpRng, const void*_src, ptrdiff_
}
if (biMCForDMVR)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ if ((isFirst == isLast) || (isFirst && dmvrHeadRoom == 0))
+ {
+ vsum = vsrc;
+ }
+ else if (isFirst)
+ {
+ if (dmvrHeadRoom > 0)
+ {
+ vsum = _mm_slli_epi16(vsrc, dmvrHeadRoom);
+ }
+ else
+ {
+ vsum = _mm_srai_epi16(vsrc, -dmvrHeadRoom);
+ }
+ }
+ else
+ {
+ CHECK(1, "Impossible to have isFirst being false and isLast being true, when biMCForDMVR is true");
+ }
+#else
vsum = _mm_min_epi16(vibdimax, _mm_max_epi16(vibdimin, vsrc));
+#endif
}
else if (isFirst == isLast)
{
@@ -355,6 +460,9 @@ static void fullPelCopyAVX2( const ClpRng& clpRng, const void*_src, int srcStrid
__m256i vibdimax = _mm256_set1_epi16( clpRng.max );
__m256i vsrc, vsum;
+#if JVET_AJ0237_INTERNAL_12BIT
+ int dmvrHeadRoom = IF_INTERNAL_PREC_BILINEAR(clpRng.bd) - clpRng.bd; // in the current setup, dmvr headroom should either be 0 or negative
+#endif
for( int row = 0; row < height; row++ )
{
@@ -376,7 +484,29 @@ static void fullPelCopyAVX2( const ClpRng& clpRng, const void*_src, int srcStrid
#if MCIF_SIMD_NEW
if (biMCForDMVR)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ if ((isFirst == isLast) || (isFirst && dmvrHeadRoom == 0))
+ {
+ vsum = vsrc;
+ }
+ else if (isFirst)
+ {
+ if (dmvrHeadRoom > 0)
+ {
+ vsum = _mm256_slli_epi16(vsrc, dmvrHeadRoom);
+ }
+ else
+ {
+ vsum = _mm256_srai_epi16(vsrc, -dmvrHeadRoom);
+ }
+ }
+ else
+ {
+ CHECK(1, "Impossible to have isFirst being false and isLast being true, when biMCForDMVR is true");
+ }
+#else
vsum = _mm256_min_epi16(vibdimax, _mm256_max_epi16(vibdimin, vsrc));
+#endif
}
else if (isFirst == isLast)
{
@@ -426,6 +556,10 @@ static void fullPelCopyVerAVX2(const ClpRng& clpRng, const void*_src, int srcStr
int offset = 1 << (headroom - 1);
int internal_offset = IF_INTERNAL_OFFS;
+#if JVET_AJ0237_INTERNAL_12BIT
+ int dmvrHeadRoom = IF_INTERNAL_PREC_BILINEAR(clpRng.bd) - clpRng.bd; // in the current setup, dmvr headroom should either be 0 or negative
+#endif
+
__m256i vinternal_offset = _mm256_set1_epi16(internal_offset);
__m256i vheadroom_offset = _mm256_set1_epi16(offset);
@@ -452,7 +586,29 @@ static void fullPelCopyVerAVX2(const ClpRng& clpRng, const void*_src, int srcStr
}
if (biMCForDMVR)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ if ((isFirst == isLast) || (isFirst && dmvrHeadRoom == 0))
+ {
+ vsum = vsrc;
+ }
+ else if (isFirst)
+ {
+ if (dmvrHeadRoom > 0)
+ {
+ vsum = _mm256_slli_epi16(vsrc, dmvrHeadRoom);
+ }
+ else
+ {
+ vsum = _mm256_srai_epi16(vsrc, -dmvrHeadRoom);
+ }
+ }
+ else
+ {
+ CHECK(1, "Impossible to have isFirst being false and isLast being true, when biMCForDMVR is true");
+ }
+#else
vsum = _mm256_min_epi16(vibdimax, _mm256_max_epi16(vibdimin, vsrc));
+#endif
}
else if (isFirst == isLast)
{
@@ -3262,7 +3418,11 @@ static void simdFilter( const ClpRng& clpRng, Pel const *src, int srcStride, Pel
{
if( isFirst )
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ shift = IF_FILTER_PREC_BILINEAR - (IF_INTERNAL_PREC_BILINEAR(clpRng.bd) - clpRng.bd);
+#else
shift = IF_FILTER_PREC_BILINEAR - (IF_INTERNAL_PREC_BILINEAR - clpRng.bd);
+#endif
offset = 1 << (shift - 1);
}
else
diff --git a/source/Lib/CommonLib/x86/IntraX86.h b/source/Lib/CommonLib/x86/IntraX86.h
index 8d03a3abc..d505792cd 100644
--- a/source/Lib/CommonLib/x86/IntraX86.h
+++ b/source/Lib/CommonLib/x86/IntraX86.h
@@ -2723,7 +2723,11 @@ bool xPredIntraOpt_SIMD(PelBuf &pDst, const PredictionUnit &pu, const uint32_t m
const int addShift = 1 << 13;
const __m128i offset = _mm_set1_epi32( addShift );
+#if JVET_AJ0237_INTERNAL_12BIT
+ const __m128i max = _mm_set1_epi32((1 << clpRng.bd) - 1);
+#else
const __m128i max = _mm_set1_epi32( 1023 );
+#endif
const __m128i zeros = _mm_setzero_si128();
__m128i vmat[ 4 ], vcoef[ 4 ], vsrc;
@@ -2862,6 +2866,27 @@ int64_t calcAeipGroupSumSIMD(const Pel* src1, const Pel* src2, const int numSamp
__m256i vzero = _mm256_setzero_si256();
__m256i vsum32 = vzero;
const int samplesBySIMD = (numSamples >> 4) << 4;
+#if JVET_AJ0237_INTERNAL_12BIT
+ int64_t sum = 0;
+ const int simdSampleBatchCnt = (samplesBySIMD >> 4) >> 2;
+ for (int batchIdx = 0; batchIdx < simdSampleBatchCnt; batchIdx++)
+ {
+ vsum32 = vzero;
+ for (; i < ((batchIdx + 1) * 64); i += 16)
+ {
+ __m256i vsrc1 = _mm256_lddqu_si256((__m256i*)(&src1[i]));
+ __m256i vsrc2 = _mm256_lddqu_si256((__m256i*)(&src2[i]));
+ __m256i vsumtemp = _mm256_madd_epi16(vsrc1, vsrc2);
+ vsum32 = _mm256_add_epi32(vsum32, vsumtemp);
+ }
+ vsum32 = _mm256_hadd_epi32(vsum32, vsum32);
+ vsum32 = _mm256_hadd_epi32(vsum32, vsum32);
+ sum += (_mm_cvtsi128_si32(_mm256_castsi256_si128(vsum32)) + _mm_cvtsi128_si32(_mm256_castsi256_si128(_mm256_permute2x128_si256(vsum32, vsum32, 0x11))));
+ }
+ if (i < samplesBySIMD)
+ {
+ vsum32 = vzero;
+#endif
for (; i < samplesBySIMD; i += 16)
{
__m256i vsrc1 = _mm256_lddqu_si256((__m256i*)(&src1[i]));
@@ -2871,11 +2896,37 @@ int64_t calcAeipGroupSumSIMD(const Pel* src1, const Pel* src2, const int numSamp
}
vsum32 = _mm256_hadd_epi32(vsum32, vsum32);
vsum32 = _mm256_hadd_epi32(vsum32, vsum32);
+#if JVET_AJ0237_INTERNAL_12BIT
+ sum += (_mm_cvtsi128_si32(_mm256_castsi256_si128(vsum32)) + _mm_cvtsi128_si32(_mm256_castsi256_si128(_mm256_permute2x128_si256(vsum32, vsum32, 0x11))));
+ }
+#else
int64_t sum = (_mm_cvtsi128_si32(_mm256_castsi256_si128(vsum32)) + _mm_cvtsi128_si32(_mm256_castsi256_si128(_mm256_permute2x128_si256(vsum32, vsum32, 0x11))));
+#endif
#else
__m128i vzero = _mm_setzero_si128();
__m128i vsum32 = vzero;
const int samplesBySIMD = (numSamples >> 4) << 4;
+#if JVET_AJ0237_INTERNAL_12BIT
+ int64_t sum = 0;
+ const int simdSampleBatchCnt = (samplesBySIMD >> 4) >> 2;
+ for (int batchIdx = 0; batchIdx < simdSampleBatchCnt; batchIdx++)
+ {
+ vsum32 = vzero;
+ for (; i < ((batchIdx + 1) * 64); i += 8)
+ {
+ __m128i vsrc1 = _mm_loadu_si128((__m128i*)(&src1[i]));
+ __m128i vsrc2 = _mm_loadu_si128((__m128i*)(&src2[i]));
+ __m128i vsumtemp = _mm_madd_epi16(vsrc1, vsrc2);
+ vsum32 = _mm_add_epi32(vsum32, vsumtemp);
+ }
+ vsum32 = _mm_add_epi32(vsum32, _mm_shuffle_epi32(vsum32, 0x4e));
+ vsum32 = _mm_add_epi32(vsum32, _mm_shuffle_epi32(vsum32, 0xb1));
+ sum += _mm_cvtsi128_si32(vsum32);
+ }
+ if (i < samplesBySIMD)
+ {
+ vsum32 = vzero;
+#endif
for (; i < samplesBySIMD; i += 8)
{
__m128i vsrc1 = _mm_loadu_si128((__m128i*)(&src1[i]));
@@ -2885,7 +2936,12 @@ int64_t calcAeipGroupSumSIMD(const Pel* src1, const Pel* src2, const int numSamp
}
vsum32 = _mm_add_epi32(vsum32, _mm_shuffle_epi32(vsum32, 0x4e)); // 01001110
vsum32 = _mm_add_epi32(vsum32, _mm_shuffle_epi32(vsum32, 0xb1)); // 10110001
+#if JVET_AJ0237_INTERNAL_12BIT
+ sum += _mm_cvtsi128_si32(vsum32);
+ }
+#else
int64_t sum = _mm_cvtsi128_si32(vsum32);
+#endif
#endif
for (; i < numSamples; i++)
{
diff --git a/source/Lib/CommonLib/x86/RdCostX86.h b/source/Lib/CommonLib/x86/RdCostX86.h
index 5a877a07e..f0c53c1a2 100644
--- a/source/Lib/CommonLib/x86/RdCostX86.h
+++ b/source/Lib/CommonLib/x86/RdCostX86.h
@@ -50,7 +50,11 @@ template<X86_VEXT vext >
Distortion RdCost::xGetSSE_SIMD( const DistParam &rcDtParam )
{
#if DIST_SSE_ENABLE
+#if JVET_AJ0237_INTERNAL_12BIT
+ if (rcDtParam.org.width < 4 || rcDtParam.bitDepth > 12 || rcDtParam.applyWeight)
+#else
if (rcDtParam.org.width < 4 || rcDtParam.bitDepth > 10 || rcDtParam.applyWeight)
+#endif
#else
if( rcDtParam.bitDepth > 10 )
#endif
@@ -179,7 +183,11 @@ Distortion RdCost::xGetSSE_SIMD( const DistParam &rcDtParam )
template<int iWidth, X86_VEXT vext >
Distortion RdCost::xGetSSE_NxN_SIMD( const DistParam &rcDtParam )
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ if (rcDtParam.bitDepth > 12 || rcDtParam.applyWeight)
+#else
if( rcDtParam.bitDepth > 10 || rcDtParam.applyWeight )
+#endif
return RdCost::xGetSSE( rcDtParam );
const Torg* pSrc1 = (const Torg*)rcDtParam.org.buf;
@@ -354,7 +362,11 @@ Distortion RdCost::xGetSSE_NxN_SIMD( const DistParam &rcDtParam )
template<X86_VEXT vext>
Distortion RdCost::xGetSSE_NxN_SIMD(const DistParam &rcDtParam)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ if (rcDtParam.bitDepth > 12 || rcDtParam.applyWeight)
+#else
if (rcDtParam.bitDepth > 10 || rcDtParam.applyWeight)
+#endif
return RdCost::xGetSSE(rcDtParam);
const Torg *pSrc1 = (const Torg *) rcDtParam.org.buf;
@@ -500,7 +512,11 @@ Distortion RdCost::xGetSSE_NxN_SIMD(const DistParam &rcDtParam)
template< X86_VEXT vext >
Distortion RdCost::xGetSAD_SIMD( const DistParam &rcDtParam )
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ if (rcDtParam.org.width < 4 || rcDtParam.bitDepth > 12 || rcDtParam.applyWeight)
+#else
if( rcDtParam.org.width < 4 || rcDtParam.bitDepth > 10 || rcDtParam.applyWeight )
+#endif
return RdCost::xGetSAD( rcDtParam );
const short* pSrc1 = (const short*)rcDtParam.org.buf;
@@ -593,7 +609,11 @@ Distortion RdCost::xGetSAD_SIMD( const DistParam &rcDtParam )
template< X86_VEXT vext >
Distortion RdCost::xGetSAD_IBD_SIMD(const DistParam &rcDtParam)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ if (rcDtParam.org.width < 4 || rcDtParam.bitDepth > 12 || rcDtParam.applyWeight)
+#else
if (rcDtParam.org.width < 4 || rcDtParam.bitDepth > 10 || rcDtParam.applyWeight)
+#endif
return RdCost::xGetSAD(rcDtParam);
const short* src0 = (const short*)rcDtParam.org.buf;
@@ -631,7 +651,11 @@ Distortion RdCost::xGetSAD_IBD_SIMD(const DistParam &rcDtParam)
template< int iWidth, X86_VEXT vext >
Distortion RdCost::xGetSAD_NxN_SIMD( const DistParam &rcDtParam )
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ if (rcDtParam.bitDepth > 12 || rcDtParam.applyWeight)
+#else
if( rcDtParam.bitDepth > 10 || rcDtParam.applyWeight )
+#endif
return RdCost::xGetSAD( rcDtParam );
// assert( rcDtParam.iCols == iWidth);
@@ -2493,7 +2517,11 @@ static uint32_t xCalcHAD8x16_AVX2( const Pel* piOrg, const Pel* piCur, const int
template< X86_VEXT vext >
Distortion RdCost::xGetSADwMask_SIMD( const DistParam &rcDtParam )
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ if (rcDtParam.org.width < 4 || rcDtParam.bitDepth > 12 || rcDtParam.applyWeight)
+#else
if (rcDtParam.org.width < 4 || rcDtParam.bitDepth > 10 || rcDtParam.applyWeight)
+#endif
return RdCost::xGetSADwMask( rcDtParam );
const short* src1 = (const short*)rcDtParam.org.buf;
@@ -2583,7 +2611,11 @@ Distortion RdCost::xGetSADwMask_SIMD( const DistParam &rcDtParam )
template<X86_VEXT vext>
Distortion RdCost::xGetHADs_SIMD( const DistParam &rcDtParam )
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ if (rcDtParam.bitDepth > 12 || rcDtParam.applyWeight)
+#else
if( rcDtParam.bitDepth > 10 || rcDtParam.applyWeight )
+#endif
{
return RdCost::xGetHADs( rcDtParam );
}
@@ -2842,7 +2874,11 @@ Distortion RdCost::xGetMRSAD_SIMD(const DistParam &rcDtParam)
{
int width = rcDtParam.org.width;
+#if JVET_AJ0237_INTERNAL_12BIT
+ if (width < 4 || rcDtParam.bitDepth > 12 || rcDtParam.applyWeight)
+#else
if (width < 4 || rcDtParam.bitDepth > 10 || rcDtParam.applyWeight)
+#endif
{
return RdCost::xGetMRSAD(rcDtParam);
}
@@ -3056,7 +3092,11 @@ Distortion RdCost::xGetTMErrorFull_SIMD(const DistParam& rcDtParam)
{
if ( rcDtParam.org.width < 4
|| ( trueAfalseL && (rcDtParam.org.width & 15) ) // (Above template) multiple of 16
+#if JVET_AJ0237_INTERNAL_12BIT
+ || rcDtParam.bitDepth > 12
+#else
|| rcDtParam.bitDepth > 10
+#endif
|| rcDtParam.applyWeight
)
{
diff --git a/source/Lib/DecoderLib/DecLib.cpp b/source/Lib/DecoderLib/DecLib.cpp
index 955a7d0b2..8d36328a2 100644
--- a/source/Lib/DecoderLib/DecLib.cpp
+++ b/source/Lib/DecoderLib/DecLib.cpp
@@ -1119,9 +1119,17 @@ void DecLib::finishPicture(int& poc, PicList*& rpcListPic, MsgLevel msgl )
#if JVET_AA0096_MC_BOUNDARY_PADDING
m_cFrameMcPadPrediction.init(&m_cRdCost, pcSlice->getSPS()->getChromaFormatIdc(), pcSlice->getSPS()->getMaxCUHeight(),
#if JVET_AJ0172_IBC_ITMP_ALIGN_REF_AREA
+#if JVET_AJ0237_INTERNAL_12BIT
+ NULL, m_pcPic->getPicWidthInLumaSamples(),m_pcPic->getPicHeightInLumaSamples(), pcSlice->getSPS()->getBitDepth(CHANNEL_TYPE_LUMA));
+#else
NULL, m_pcPic->getPicWidthInLumaSamples(),m_pcPic->getPicHeightInLumaSamples());
+#endif
+#else
+#if JVET_AJ0237_INTERNAL_12BIT
+ NULL, m_pcPic->getPicWidthInLumaSamples(), pcSlice->getSPS()->getBitDepth(CHANNEL_TYPE_LUMA));
#else
NULL, m_pcPic->getPicWidthInLumaSamples());
+#endif
#endif
m_cFrameMcPadPrediction.mcFramePad(m_pcPic, *(m_pcPic->slices[0]));
#endif
@@ -1969,6 +1977,10 @@ void DecLib::xActivateParameterSets( const InputNALUnit nalu )
sps->getMaxCUWidth(), sps->getMaxCUHeight(),
maxDepth,
log2SaoOffsetScaleLuma, log2SaoOffsetScaleChroma );
+#if JVET_AJ0237_INTERNAL_12BIT
+ m_cSAO.m_bilateralFilter.setInternalBitDepth(sps->getBitDepth(CHANNEL_TYPE_LUMA));
+ m_cBilateralFilter.setInternalBitDepth(sps->getBitDepth(CHANNEL_TYPE_LUMA));
+#endif
#if JVET_W0066_CCSAO
pSlice->m_ccSaoControl[COMPONENT_Y ] = m_cSAO.getCcSaoControlIdc(COMPONENT_Y);
pSlice->m_ccSaoControl[COMPONENT_Cb] = m_cSAO.getCcSaoControlIdc(COMPONENT_Cb);
@@ -1979,10 +1991,18 @@ void DecLib::xActivateParameterSets( const InputNALUnit nalu )
#if INTER_LIC || (TM_AMVP || TM_MRG || JVET_Z0084_IBC_TM) || JVET_W0090_ARMC_TM || JVET_Z0056_GPM_SPLIT_MODE_REORDERING
#if JVET_Z0153_IBC_EXT_REF
#if JVET_AJ0172_IBC_ITMP_ALIGN_REF_AREA
+#if JVET_AJ0237_INTERNAL_12BIT
+ m_cInterPred.init(&m_cRdCost, sps->getChromaFormatIdc(), sps->getMaxCUHeight(), &m_cReshaper, sps->getMaxPicWidthInLumaSamples(),sps->getMaxPicHeightInLumaSamples(), sps->getBitDepth(CHANNEL_TYPE_LUMA));
+#else
m_cInterPred.init(&m_cRdCost, sps->getChromaFormatIdc(), sps->getMaxCUHeight(), &m_cReshaper, sps->getMaxPicWidthInLumaSamples(),sps->getMaxPicHeightInLumaSamples());
+#endif
+#else
+#if JVET_AJ0237_INTERNAL_12BIT
+ m_cInterPred.init(&m_cRdCost, sps->getChromaFormatIdc(), sps->getMaxCUHeight(), &m_cReshaper, sps->getMaxPicWidthInLumaSamples(), sps->getBitDepth(CHANNEL_TYPE_LUMA));
#else
m_cInterPred.init(&m_cRdCost, sps->getChromaFormatIdc(), sps->getMaxCUHeight(), &m_cReshaper, sps->getMaxPicWidthInLumaSamples());
#endif
+#endif
#else
m_cInterPred.init( &m_cRdCost, sps->getChromaFormatIdc(), sps->getMaxCUHeight(), &m_cReshaper);
#endif
@@ -3086,6 +3106,9 @@ bool DecLib::xDecodeSlice(InputNALUnit &nalu, int &iSkipFrame, int iPOCLastDispl
{
clipDeltaShift = ADAPTIVE_CLIP_SHIFT_DELTA_VALUE_0;
}
+#if JVET_AJ0237_INTERNAL_12BIT
+ clipDeltaShift += std::max(0, pcSlice->getSPS()->getBitDepth(toChannelType(COMPONENT_Y)) - 10);
+#endif
if (pcSlice->getSliceType() != I_SLICE)
{
int deltaMax = pcSlice->getLumaPelMax();
diff --git a/source/Lib/DecoderLib/VLCReader.cpp b/source/Lib/DecoderLib/VLCReader.cpp
index f7a37c708..103394181 100644
--- a/source/Lib/DecoderLib/VLCReader.cpp
+++ b/source/Lib/DecoderLib/VLCReader.cpp
@@ -4711,7 +4711,11 @@ void HLSyntaxReader::parseScaleAlf( Slice* pcSlice, SPS* sps, ParameterSetManage
{
ScaleAlf& curScaleAlfParam = pcSlice->getAlfScale( i , j );
+#if JVET_AJ0237_INTERNAL_12BIT
+ curScaleAlfParam.init(apsIdx, j, alfParam.lumaClassifierIdx[j], sps->getBitDepth(CHANNEL_TYPE_LUMA));
+#else
curScaleAlfParam.init( apsIdx, j, alfParam.lumaClassifierIdx[j] );
+#endif
curScaleAlfParam.apsIdx = apsIdx;
if ( !bReadUseAlfScale )
@@ -5807,7 +5811,11 @@ void HLSyntaxReader::parseSliceHeader (Slice* pcSlice, PicHeader* picHeader, Par
if (index != -1)
{
lambdaCanBePredicted = true;
+#if JVET_AJ0237_INTERNAL_12BIT
+ pcSlice->setCostForARMC(sps->getLambdaVal((int)index), sps->getBitDepth(CHANNEL_TYPE_LUMA));
+#else
pcSlice->setCostForARMC(sps->getLambdaVal((int) index));
+#endif
}
if (!lambdaCanBePredicted)
{
@@ -5816,7 +5824,11 @@ void HLSyntaxReader::parseSliceHeader (Slice* pcSlice, PicHeader* picHeader, Par
#else
READ_CODE(9, uiCode, "Lambda");
#endif
+#if JVET_AJ0237_INTERNAL_12BIT
+ pcSlice->setCostForARMC((uint32_t)uiCode, sps->getBitDepth(CHANNEL_TYPE_LUMA));
+#else
pcSlice->setCostForARMC((uint32_t)uiCode);
+#endif
}
}
#endif
@@ -5861,6 +5873,9 @@ void HLSyntaxReader::parseSliceHeader (Slice* pcSlice, PicHeader* picHeader, Par
{
clipDeltaShift = ADAPTIVE_CLIP_SHIFT_DELTA_VALUE_0;
}
+#if JVET_AJ0237_INTERNAL_12BIT
+ clipDeltaShift += std::max(0, sps->getBitDepth(toChannelType(COMPONENT_Y)) - 10);
+#endif
if (pcSlice->getSliceType() == I_SLICE)
{
READ_SVLC(iCode, "clip_luma_pel_max");
diff --git a/source/Lib/EncoderLib/EncAdaptiveLoopFilter.cpp b/source/Lib/EncoderLib/EncAdaptiveLoopFilter.cpp
index 38e77f597..79bad478e 100644
--- a/source/Lib/EncoderLib/EncAdaptiveLoopFilter.cpp
+++ b/source/Lib/EncoderLib/EncAdaptiveLoopFilter.cpp
@@ -7051,7 +7051,11 @@ void EncAdaptiveLoopFilter::alfCorrection( CodingStructure& cs, const PelUnitBuf
{
curScaleAlfEncParam.reset();
+#if JVET_AJ0237_INTERNAL_12BIT
+ curScaleAlfParam.init(filterSetIndex, alt_num, classifierIdx, cs.slice->getSPS()->getBitDepth(CHANNEL_TYPE_LUMA));
+#else
curScaleAlfParam.init( filterSetIndex, alt_num, classifierIdx );
+#endif
curScaleAlfParam.setMinMax( cs.slice->getLumaPelMin(), cs.slice->getLumaPelMax() );
const int apsIdx = cs.slice->getTileGroupApsIdLuma()[filterSetIndex - NUM_FIXED_FILTER_SETS];
@@ -7061,7 +7065,11 @@ void EncAdaptiveLoopFilter::alfCorrection( CodingStructure& cs, const PelUnitBuf
CHECK( curScaleAlfParam.classifierIdx != classifierIdx || curScaleAlfParam.filterSetIndex != filterSetIndex || curScaleAlfParam.alt_num != alt_num, "alfCorrection() failed.");
char coeffBits = m_coeffBitsApsLuma[filterSetIndex - NUM_FIXED_FILTER_SETS][alt_num];
+#if JVET_AJ0237_INTERNAL_12BIT
+ const Pel currBase = 1 << (curScaleAlfParam.bitDepth - 1);
+#else
const Pel currBase = 512; // 10-bits
+#endif
if ( !bModeAnalysis )
{
@@ -9463,7 +9471,11 @@ void EncAdaptiveLoopFilter::countLumaSwingGreaterThanThreshold(const Pel* luma,
void EncAdaptiveLoopFilter::countChromaSampleValueNearMidPoint(const Pel* chroma, int chromaStride, int height, int width, int log2BlockWidth, int log2BlockHeight, uint64_t* chromaSampleCountNearMidPoint, int chromaSampleCountNearMidPointStride)
{
const int midPoint = (1 << m_inputBitDepth[CH_C]) >> 1;
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int threshold = 16 << std::max(0, m_inputBitDepth[CH_C] - 10);
+#else
const int threshold = 16;
+#endif
for (int y = 0; y < height; y += (1 << log2BlockHeight))
{
diff --git a/source/Lib/EncoderLib/EncCu.cpp b/source/Lib/EncoderLib/EncCu.cpp
index f06e35c17..bcf1afb77 100644
--- a/source/Lib/EncoderLib/EncCu.cpp
+++ b/source/Lib/EncoderLib/EncCu.cpp
@@ -122,6 +122,9 @@ void EncCu::create( EncCfg* encCfg )
#if JVET_V0094_BILATERAL_FILTER || JVET_X0071_CHROMA_BILATERAL_FILTER
m_bilateralFilter = new BilateralFilter();;
m_bilateralFilter->create();
+#if JVET_AJ0237_INTERNAL_12BIT
+ m_bilateralFilter->setInternalBitDepth(encCfg->getBitDepth(CHANNEL_TYPE_LUMA));
+#endif
#endif
unsigned uiMaxWidth = encCfg->getMaxCUWidth();
diff --git a/source/Lib/EncoderLib/EncGOP.cpp b/source/Lib/EncoderLib/EncGOP.cpp
index 472f0680d..fb066e072 100644
--- a/source/Lib/EncoderLib/EncGOP.cpp
+++ b/source/Lib/EncoderLib/EncGOP.cpp
@@ -2707,11 +2707,19 @@ void EncGOP::compressGOP(int iPOCLast, int iNumPicRcvd, PicList &rcListPic, std:
if (index != -1)
{
const SPS* sps = pcSlice->getSPS();
+#if JVET_AJ0237_INTERNAL_12BIT
+ pcSlice->setCostForARMC(sps->getLambdaVal(index), sps->getBitDepth(CHANNEL_TYPE_LUMA));
+#else
pcSlice->setCostForARMC(sps->getLambdaVal(index));
+#endif
}
else
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ pcSlice->setCostForARMC((uint32_t)LAMBDA_DEC_SIDE[min(max(pcSlice->getSliceQp(), 0), MAX_QP)], pcSlice->getSPS()->getBitDepth(CHANNEL_TYPE_LUMA));
+#else
pcSlice->setCostForARMC((uint32_t) LAMBDA_DEC_SIDE[min(max(pcSlice->getSliceQp(), 0), MAX_QP)]);
+#endif
}
if (pcSlice->getCheckLDC())
@@ -2738,12 +2746,20 @@ void EncGOP::compressGOP(int iPOCLast, int iNumPicRcvd, PicList &rcListPic, std:
}
if (mindist != 1 )
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ pcSlice->setCostForARMC((uint32_t)LAMBDA_DEC_SIDE[min(max(pcSlice->getSliceQp() - 4, 0), MAX_QP)], pcSlice->getSPS()->getBitDepth(CHANNEL_TYPE_LUMA));
+#else
pcSlice->setCostForARMC((uint32_t) LAMBDA_DEC_SIDE[min(max(pcSlice->getSliceQp() - 4, 0), MAX_QP)]);
+#endif
}
}
else
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ pcSlice->setCostForARMC((uint32_t)LAMBDA_DEC_SIDE[min(max(pcSlice->getSliceQp() - 4, 0), MAX_QP)], pcSlice->getSPS()->getBitDepth(CHANNEL_TYPE_LUMA));
+#else
pcSlice->setCostForARMC((uint32_t) LAMBDA_DEC_SIDE[min(max(pcSlice->getSliceQp() - 4, 0), MAX_QP)]);
+#endif
}
}
#endif
@@ -3803,6 +3819,9 @@ void EncGOP::compressGOP(int iPOCLast, int iNumPicRcvd, PicList &rcListPic, std:
if ( saoSize.width != picWidth || saoSize.height != picHeight )
{
m_pcSAO->create(picWidth, picHeight, chromaFormatIDC, maxCUWidth, maxCUHeight, maxTotalCUDepth, log2SaoOffsetScaleLuma, log2SaoOffsetScaleChroma);
+#if JVET_AJ0237_INTERNAL_12BIT
+ m_pcSAO->m_bilateralFilter.setInternalBitDepth(pcSlice->getSPS()->getBitDepth(CHANNEL_TYPE_LUMA));
+#endif
m_pcSAO->setReshaper(m_pcReshaper);
}
@@ -4852,9 +4871,17 @@ void EncGOP::compressGOP(int iPOCLast, int iNumPicRcvd, PicList &rcListPic, std:
#if JVET_AA0096_MC_BOUNDARY_PADDING
m_pcFrameMcPadPrediction->init(m_pcEncLib->getRdCost(), pcSlice->getSPS()->getChromaFormatIdc(),
#if JVET_AJ0172_IBC_ITMP_ALIGN_REF_AREA
+#if JVET_AJ0237_INTERNAL_12BIT
+ pcSlice->getSPS()->getMaxCUHeight(), NULL, pcPic->getPicWidthInLumaSamples(),pcPic->getPicHeightInLumaSamples(), pcSlice->getSPS()->getBitDepth(CHANNEL_TYPE_LUMA));
+#else
pcSlice->getSPS()->getMaxCUHeight(), NULL, pcPic->getPicWidthInLumaSamples(),pcPic->getPicHeightInLumaSamples());
+#endif
+#else
+#if JVET_AJ0237_INTERNAL_12BIT
+ pcSlice->getSPS()->getMaxCUHeight(), NULL, pcPic->getPicWidthInLumaSamples(), pcSlice->getSPS()->getBitDepth(CHANNEL_TYPE_LUMA));
#else
pcSlice->getSPS()->getMaxCUHeight(), NULL, pcPic->getPicWidthInLumaSamples());
+#endif
#endif
m_pcFrameMcPadPrediction->mcFramePad(pcPic, *(pcPic->slices[0]));
m_pcFrameMcPadPrediction->destroy();
diff --git a/source/Lib/EncoderLib/EncGOP.h b/source/Lib/EncoderLib/EncGOP.h
index 907cf60ff..7fad71fa8 100644
--- a/source/Lib/EncoderLib/EncGOP.h
+++ b/source/Lib/EncoderLib/EncGOP.h
@@ -140,8 +140,10 @@ private:
PicList* m_pcListPic;
HLSWriter* m_HLSWriter;
+#if !JVET_AJ0237_INTERNAL_12BIT
#if JVET_V0094_BILATERAL_FILTER || JVET_X0071_CHROMA_BILATERAL_FILTER
BilateralFilter m_cBilateralFilter;
+#endif
#endif
LoopFilter* m_pcLoopFilter;
@@ -214,6 +216,12 @@ private:
#endif
public:
+#if JVET_AJ0237_INTERNAL_12BIT
+#if JVET_V0094_BILATERAL_FILTER || JVET_X0071_CHROMA_BILATERAL_FILTER
+ BilateralFilter m_cBilateralFilter;
+#endif
+#endif
+
EncGOP();
virtual ~EncGOP();
diff --git a/source/Lib/EncoderLib/EncLib.cpp b/source/Lib/EncoderLib/EncLib.cpp
index 964180341..613ba5a02 100644
--- a/source/Lib/EncoderLib/EncLib.cpp
+++ b/source/Lib/EncoderLib/EncLib.cpp
@@ -101,6 +101,9 @@ void EncLib::create( const int layerId )
m_iPOCLast = m_compositeRefEnabled ? -2 : -1;
// create processing unit classes
m_cGOPEncoder. create( );
+#if JVET_AJ0237_INTERNAL_12BIT
+ m_cGOPEncoder.m_cBilateralFilter.setInternalBitDepth(m_bitDepth[COMPONENT_Y]);
+#endif
#if ENABLE_SPLIT_PARALLELISM
#if ENABLE_SPLIT_PARALLELISM
m_numCuEncStacks = m_numSplitThreads == 1 ? 1 : NUM_RESERVERD_SPLIT_JOBS;
@@ -131,6 +134,9 @@ void EncLib::create( const int layerId )
m_cCuEncoder. create( this );
#if JVET_V0094_BILATERAL_FILTER || JVET_X0071_CHROMA_BILATERAL_FILTER
m_bilateralFilter. create();
+#if JVET_AJ0237_INTERNAL_12BIT
+ m_bilateralFilter.setInternalBitDepth(m_bitDepth[COMPONENT_Y]);
+#endif
#endif
#endif
#if JVET_J0090_MEMORY_BANDWITH_MEASURE
@@ -219,6 +225,9 @@ void EncLib::create( const int layerId )
m_cEncSAO.create(m_iSourceWidth, m_iSourceHeight, m_chromaFormatIDC, m_maxCUWidth, m_maxCUHeight, floorLog2(m_maxCUWidth) - m_log2MinCUSize, (uint32_t)std::max(0, m_bitDepth[CHANNEL_TYPE_LUMA] - MAX_SAO_TRUNCATED_BITDEPTH), (uint32_t)std::max(0, m_bitDepth[CHANNEL_TYPE_CHROMA] - MAX_SAO_TRUNCATED_BITDEPTH));
#endif
m_cEncSAO.createEncData(m_saoCtuBoundary, numCtuInFrame);
+#if JVET_AJ0237_INTERNAL_12BIT
+ m_cEncSAO.m_bilateralFilter.setInternalBitDepth(m_bitDepth[COMPONENT_Y]);
+#endif
}
}
diff --git a/source/Lib/EncoderLib/EncSampleAdaptiveOffset.cpp b/source/Lib/EncoderLib/EncSampleAdaptiveOffset.cpp
index e850b03da..04ff642e8 100644
--- a/source/Lib/EncoderLib/EncSampleAdaptiveOffset.cpp
+++ b/source/Lib/EncoderLib/EncSampleAdaptiveOffset.cpp
@@ -413,6 +413,9 @@ void EncSampleAdaptiveOffset::SAOProcess( CodingStructure& cs, bool* sliceEnable
if(!cs.sps->getSAOEnabledFlag() && (cs.pps->getUseBIF() || cs.pps->getUseChromaBIF()))
{
bilateralFilter.create();
+#if JVET_AJ0237_INTERNAL_12BIT
+ bilateralFilter.setInternalBitDepth(cs.sps->getBitDepth(CHANNEL_TYPE_LUMA));
+#endif
if( cs.pps->getUseBIF() )
{
bilateralFilter.bilateralFilterPicRDOperCTU( COMPONENT_Y, cs, src, bifCABACEstimator ); // Filters from src to res
@@ -468,6 +471,9 @@ void EncSampleAdaptiveOffset::SAOProcess( CodingStructure& cs, bool* sliceEnable
if( cs.pps->getUseBIF() || cs.pps->getUseChromaBIF() )
{
bilateralFilter.create();
+#if JVET_AJ0237_INTERNAL_12BIT
+ bilateralFilter.setInternalBitDepth(cs.sps->getBitDepth(CHANNEL_TYPE_LUMA));
+#endif
if( cs.pps->getUseBIF() )
{
bilateralFilter.bilateralFilterPicRDOperCTU( COMPONENT_Y, cs, src, bifCABACEstimator ); // Filters from src to res'
@@ -774,9 +780,18 @@ int64_t EncSampleAdaptiveOffset::getDistortion(const int channelBitDepth, int ty
return dist;
}
+#if JVET_AJ0237_INTERNAL_12BIT
+inline int64_t EncSampleAdaptiveOffset::estSaoDist(int64_t count, int64_t offset, int64_t diffSum, int shift, int bdShift)
+#else
inline int64_t EncSampleAdaptiveOffset::estSaoDist(int64_t count, int64_t offset, int64_t diffSum, int shift)
+#endif
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ int64_t tmpOffset = offset << bdShift;
+ return ((count * tmpOffset * tmpOffset - diffSum * tmpOffset * 2) >> shift);
+#else
return (( count*offset*offset-diffSum*offset*2 ) >> shift);
+#endif
}
@@ -1154,6 +1169,9 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
#if JVET_V0094_BILATERAL_FILTER || JVET_X0071_CHROMA_BILATERAL_FILTER
BilateralFilter bilateralFilter;
bilateralFilter.create();
+#if JVET_AJ0237_INTERNAL_12BIT
+ bilateralFilter.setInternalBitDepth(cs.sps->getBitDepth(CHANNEL_TYPE_LUMA));
+#endif
#endif
const TempCtx ctxPicStart ( m_ctxCache, SAOCtx( m_CABACEstimator->getCtx() ) );
@@ -3011,6 +3029,19 @@ void EncSampleAdaptiveOffset::CCSAOProcess(CodingStructure& cs, const double* la
}
#endif
+#if JVET_AJ0237_INTERNAL_12BIT
+ if (!cs.slice->isIntra() && !cs.slice->getCheckLDC() && (cs.slice->getSPS()->getBitDepth(CHANNEL_TYPE_LUMA) > 10) && (cs.slice->getSliceQp() > 45) && (m_picWidth * m_picHeight <= 1920 * 1080))
+ {
+ for (int compIdx = COMPONENT_Y; compIdx < MAX_NUM_COMPONENT; compIdx++)
+ {
+ ComponentID compID = (ComponentID)compIdx;
+ m_ccSaoComParam.reset(compID);
+ memset(m_ccSaoControl[compID], 0, sizeof(uint8_t) * m_numCTUsInPic);
+ }
+ return;
+ }
+#endif
+
PelUnitBuf orgYuv = cs.getOrgBuf();
PelUnitBuf dstYuv = cs.getRecoBuf();
PelUnitBuf srcYuv = m_ccSaoBuf.getBuf( cs.area );
@@ -3413,6 +3444,10 @@ void EncSampleAdaptiveOffset::setupInitCcSaoParam(CodingStructure& cs, const Com
initCcSaoParam.reset();
memset(initCcSaoControl, 0, sizeof(uint8_t) * m_numCTUsInPic);
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int shift = 2 * DISTORTION_PRECISION_ADJUSTMENT(cs.sps->getBitDepth(CHANNEL_TYPE_LUMA));
+#endif
+
if (setNum == 1)
{
std::fill_n(initCcSaoControl, m_numCTUsInPic, 1);
@@ -3432,7 +3467,11 @@ void EncSampleAdaptiveOffset::setupInitCcSaoParam(CodingStructure& cs, const Com
#if JVET_AE0151_CCSAO_HISTORY_OFFSETS_AND_EXT_EO
getCcSaoDistortion(compID, setIdx
, bestCcSaoParam.setType[setIdx] == CCSAO_SET_TYPE_BAND ? blkStats : blkStatsEdge
+#if JVET_AJ0237_INTERNAL_12BIT
+ , bestCcSaoParam.offset, trainingDistortion, shift);
+#else
, bestCcSaoParam.offset, trainingDistortion);
+#endif
#else
if (bestCcSaoParam.setType[setIdx] == 0) /* band */
{
@@ -3941,6 +3980,10 @@ void EncSampleAdaptiveOffset::getCcSaoBlkStatsEdgeNew(const ComponentID compID,
const int srcStrideTab[MAX_NUM_COMPONENT] = { srcStrideY, srcStrideU, srcStrideU };
#endif
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int bdShift = std::max(0, bitDepth - 10);
+#endif
+
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
int x, y, startX, startY, endX, endY;
int firstLineStartX, firstLineEndX;
@@ -3997,7 +4040,11 @@ void EncSampleAdaptiveOffset::getCcSaoBlkStatsEdgeNew(const ComponentID compID,
const int edgeNumUni = g_ccSaoEdgeNum[edgeIdc][1];
for (int edgeThr = 0; edgeThr < MAX_CCSAO_EDGE_THR; edgeThr++)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr] << bdShift;
+#else
const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr];
+#endif
const int edgeIdxA = getCcSaoEdgeIdx(*colE, *colA, edgeThrVal, edgeIdc);
const int edgeIdxB = getCcSaoEdgeIdx(*colE, *colB, edgeThrVal, edgeIdc);
const int edgeIdx = edgeIdxA * edgeNumUni + edgeIdxB;
@@ -4067,7 +4114,11 @@ void EncSampleAdaptiveOffset::getCcSaoBlkStatsEdgeNew(const ComponentID compID,
const int edgeNumUni = g_ccSaoEdgeNum[edgeIdc][1];
for (int edgeThr = 0; edgeThr < MAX_CCSAO_EDGE_THR; edgeThr++)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr] << bdShift;
+#else
const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr];
+#endif
const int edgeIdxA = getCcSaoEdgeIdx(*colE, *colA, edgeThrVal, edgeIdc);
const int edgeIdxB = getCcSaoEdgeIdx(*colE, *colB, edgeThrVal, edgeIdc);
const int edgeIdx = edgeIdxA * edgeNumUni + edgeIdxB;
@@ -4131,7 +4182,11 @@ void EncSampleAdaptiveOffset::getCcSaoBlkStatsEdgeNew(const ComponentID compID,
const int edgeNumUni = g_ccSaoEdgeNum[edgeIdc][1];
for (int edgeThr = 0; edgeThr < MAX_CCSAO_EDGE_THR; edgeThr++)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr] << bdShift;
+#else
const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr];
+#endif
const int edgeIdxA = getCcSaoEdgeIdx(*colE, *colA, edgeThrVal, edgeIdc);
const int edgeIdxB = getCcSaoEdgeIdx(*colE, *colB, edgeThrVal, edgeIdc);
const int edgeIdx = edgeIdxA * edgeNumUni + edgeIdxB;
@@ -4188,7 +4243,11 @@ void EncSampleAdaptiveOffset::getCcSaoBlkStatsEdgeNew(const ComponentID compID,
const int edgeNumUni = g_ccSaoEdgeNum[edgeIdc][1];
for (int edgeThr = 0; edgeThr < MAX_CCSAO_EDGE_THR; edgeThr++)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr] << bdShift;
+#else
const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr];
+#endif
const int edgeIdxA = getCcSaoEdgeIdx(*colE, *colA, edgeThrVal, edgeIdc);
const int edgeIdxB = getCcSaoEdgeIdx(*colE, *colB, edgeThrVal, edgeIdc);
const int edgeIdx = edgeIdxA * edgeNumUni + edgeIdxB;
@@ -4252,7 +4311,11 @@ void EncSampleAdaptiveOffset::getCcSaoBlkStatsEdgeNew(const ComponentID compID,
const int edgeNumUni = g_ccSaoEdgeNum[edgeIdc][1];
for (int edgeThr = 0; edgeThr < MAX_CCSAO_EDGE_THR; edgeThr++)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr] << bdShift;
+#else
const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr];
+#endif
const int edgeIdxA = getCcSaoEdgeIdx(*colE, *colA, edgeThrVal, edgeIdc);
const int edgeIdxB = getCcSaoEdgeIdx(*colE, *colB, edgeThrVal, edgeIdc);
const int edgeIdx = edgeIdxA * edgeNumUni + edgeIdxB;
@@ -4309,7 +4372,11 @@ void EncSampleAdaptiveOffset::getCcSaoBlkStatsEdgeNew(const ComponentID compID,
const int edgeNumUni = g_ccSaoEdgeNum[edgeIdc][1];
for (int edgeThr = 0; edgeThr < MAX_CCSAO_EDGE_THR; edgeThr++)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr] << bdShift;
+#else
const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr];
+#endif
const int edgeIdxA = getCcSaoEdgeIdx(*colE, *colA, edgeThrVal, edgeIdc);
const int edgeIdxB = getCcSaoEdgeIdx(*colE, *colB, edgeThrVal, edgeIdc);
const int edgeIdx = edgeIdxA * edgeNumUni + edgeIdxB;
@@ -4779,7 +4846,11 @@ void EncSampleAdaptiveOffset::getCcSaoBlkStatsEdgeNew(const ComponentID compID,
const int edgeNumUni = g_ccSaoEdgeNum[edgeIdc][1];
for (int edgeThr = 0; edgeThr < MAX_CCSAO_EDGE_THR; edgeThr++)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr] << bdShift;
+#else
const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr];
+#endif
const int edgeIdxA = getCcSaoEdgeIdx(*colE, *colA, edgeThrVal, edgeIdc);
const int edgeIdxB = getCcSaoEdgeIdx(*colE, *colB, edgeThrVal, edgeIdc);
const int edgeIdx = edgeIdxA * edgeNumUni + edgeIdxB;
@@ -4849,7 +4920,11 @@ void EncSampleAdaptiveOffset::getCcSaoBlkStatsEdgeNew(const ComponentID compID,
const int edgeNumUni = g_ccSaoEdgeNum[edgeIdc][1];
for (int edgeThr = 0; edgeThr < MAX_CCSAO_EDGE_THR; edgeThr++)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr] << bdShift;
+#else
const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr];
+#endif
const int edgeIdxA = getCcSaoEdgeIdx(*colE, *colA, edgeThrVal, edgeIdc);
const int edgeIdxB = getCcSaoEdgeIdx(*colE, *colB, edgeThrVal, edgeIdc);
const int edgeIdx = edgeIdxA * edgeNumUni + edgeIdxB;
@@ -4913,7 +4988,11 @@ void EncSampleAdaptiveOffset::getCcSaoBlkStatsEdgeNew(const ComponentID compID,
const int edgeNumUni = g_ccSaoEdgeNum[edgeIdc][1];
for (int edgeThr = 0; edgeThr < MAX_CCSAO_EDGE_THR; edgeThr++)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr] << bdShift;
+#else
const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr];
+#endif
const int edgeIdxA = getCcSaoEdgeIdx(*colE, *colA, edgeThrVal, edgeIdc);
const int edgeIdxB = getCcSaoEdgeIdx(*colE, *colB, edgeThrVal, edgeIdc);
const int edgeIdx = edgeIdxA * edgeNumUni + edgeIdxB;
@@ -4970,7 +5049,11 @@ void EncSampleAdaptiveOffset::getCcSaoBlkStatsEdgeNew(const ComponentID compID,
const int edgeNumUni = g_ccSaoEdgeNum[edgeIdc][1];
for (int edgeThr = 0; edgeThr < MAX_CCSAO_EDGE_THR; edgeThr++)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr] << bdShift;
+#else
const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr];
+#endif
const int edgeIdxA = getCcSaoEdgeIdx(*colE, *colA, edgeThrVal, edgeIdc);
const int edgeIdxB = getCcSaoEdgeIdx(*colE, *colB, edgeThrVal, edgeIdc);
const int edgeIdx = edgeIdxA * edgeNumUni + edgeIdxB;
@@ -5034,7 +5117,11 @@ void EncSampleAdaptiveOffset::getCcSaoBlkStatsEdgeNew(const ComponentID compID,
const int edgeNumUni = g_ccSaoEdgeNum[edgeIdc][1];
for (int edgeThr = 0; edgeThr < MAX_CCSAO_EDGE_THR; edgeThr++)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr] << bdShift;
+#else
const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr];
+#endif
const int edgeIdxA = getCcSaoEdgeIdx(*colE, *colA, edgeThrVal, edgeIdc);
const int edgeIdxB = getCcSaoEdgeIdx(*colE, *colB, edgeThrVal, edgeIdc);
const int edgeIdx = edgeIdxA * edgeNumUni + edgeIdxB;
@@ -5091,7 +5178,11 @@ void EncSampleAdaptiveOffset::getCcSaoBlkStatsEdgeNew(const ComponentID compID,
const int edgeNumUni = g_ccSaoEdgeNum[edgeIdc][1];
for (int edgeThr = 0; edgeThr < MAX_CCSAO_EDGE_THR; edgeThr++)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr] << bdShift;
+#else
const int edgeThrVal = g_ccSaoEdgeThr[edgeIdc][edgeThr];
+#endif
const int edgeIdxA = getCcSaoEdgeIdx(*colE, *colA, edgeThrVal, edgeIdc);
const int edgeIdxB = getCcSaoEdgeIdx(*colE, *colB, edgeThrVal, edgeIdc);
const int edgeIdx = edgeIdxA * edgeNumUni + edgeIdxB;
@@ -6726,14 +6817,23 @@ void EncSampleAdaptiveOffset::deriveCcSaoOffsets(const ComponentID compID, const
{
int quantOffsets[MAX_CCSAO_CLASS_NUM] = { 0 };
+#if JVET_AJ0237_INTERNAL_12BIT
+ int shift = 2 * DISTORTION_PRECISION_ADJUSTMENT(bitDepth);
+#endif
+
for(int k = 0; k < MAX_CCSAO_CLASS_NUM; k++)
{
if(frameStats[setIdx].count[k] == 0)
continue;
quantOffsets[k] =
+#if JVET_AJ0237_INTERNAL_12BIT
+ (int) xRoundIbdi(bitDepth, (double)(frameStats[setIdx].diff [k] << DISTORTION_PRECISION_ADJUSTMENT(bitDepth))
+ / (double)((int64_t)frameStats[setIdx].count[k] << m_offsetStepLog2[compID]));
+#else
(int) xRoundIbdi(bitDepth, (double)(frameStats[setIdx].diff [k] << DISTORTION_PRECISION_ADJUSTMENT(bitDepth))
/ (double)(frameStats[setIdx].count[k]));
+#endif
quantOffsets[k] = Clip3(-MAX_CCSAO_OFFSET_THR, MAX_CCSAO_OFFSET_THR, quantOffsets[k]);
}
@@ -6744,7 +6844,11 @@ void EncSampleAdaptiveOffset::deriveCcSaoOffsets(const ComponentID compID, const
cost[k] = m_lambda[compID];
if (quantOffsets[k] != 0)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ quantOffsets[k] = estCcSaoIterOffset(m_lambda[compID], quantOffsets[k], frameStats[setIdx].count[k], frameStats[setIdx].diff[k], shift, m_offsetStepLog2[compID], dist[k], cost[k], MAX_CCSAO_OFFSET_THR);
+#else
quantOffsets[k] = estCcSaoIterOffset(m_lambda[compID], quantOffsets[k], frameStats[setIdx].count[k], frameStats[setIdx].diff[k], 0, 0, dist[k], cost[k], MAX_CCSAO_OFFSET_THR);
+#endif
}
}
@@ -6757,7 +6861,11 @@ void EncSampleAdaptiveOffset::deriveCcSaoOffsets(const ComponentID compID, const
void EncSampleAdaptiveOffset::getCcSaoDistortion(const ComponentID compID, const int setIdx, CcSaoStatData* blkStats[MAX_CCSAO_SET_NUM]
, short offset[MAX_CCSAO_SET_NUM][MAX_CCSAO_CLASS_NUM]
+#if JVET_AJ0237_INTERNAL_12BIT
+ , int64_t* trainingDistortion[MAX_CCSAO_SET_NUM], const int shift)
+#else
, int64_t* trainingDistortion[MAX_CCSAO_SET_NUM])
+#endif
{
::memset(trainingDistortion[setIdx], 0, sizeof(int64_t) * m_numCTUsInPic);
@@ -6766,7 +6874,11 @@ void EncSampleAdaptiveOffset::getCcSaoDistortion(const ComponentID compID, const
for (int k = 0; k < MAX_CCSAO_CLASS_NUM; k++)
{
trainingDistortion[setIdx][ctbIdx]
+#if JVET_AJ0237_INTERNAL_12BIT
+ += estSaoDist(blkStats[setIdx][ctbIdx].count[k], offset[setIdx][k], blkStats[setIdx][ctbIdx].diff[k], shift, m_offsetStepLog2[toChannelType(compID)]);
+#else
+= estSaoDist(blkStats[setIdx][ctbIdx].count[k], offset[setIdx][k], blkStats[setIdx][ctbIdx].diff[k], 0);
+#endif
}
}
}
@@ -7064,6 +7176,10 @@ void EncSampleAdaptiveOffset::deriveCcSaoRDO(CodingStructure& cs, const Componen
const TempCtx ctxStartCcSaoControlFlag ( m_ctxCache, SubCtx( Ctx::CcSaoControlIdc, m_CABACEstimator->getCtx() ) );
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int shift = 2 * DISTORTION_PRECISION_ADJUSTMENT(cs.sps->getBitDepth(toChannelType(compID)));
+#endif
+
int trainingIter = 0;
bool keepTraining = true;
bool improved = false;
@@ -7081,7 +7197,11 @@ void EncSampleAdaptiveOffset::deriveCcSaoRDO(CodingStructure& cs, const Componen
{
getCcSaoDistortion(compID, setIdx
, tempCcSaoParam.setType[setIdx] == CCSAO_SET_TYPE_BAND ? blkStats : blkStatsEdge
+#if JVET_AJ0237_INTERNAL_12BIT
+ , tempCcSaoParam.offset, trainingDistortion, shift);
+#else
, tempCcSaoParam.offset, trainingDistortion);
+#endif
}
else
{
@@ -7099,12 +7219,20 @@ void EncSampleAdaptiveOffset::deriveCcSaoRDO(CodingStructure& cs, const Componen
#endif
{
deriveCcSaoOffsets(compID, cs.sps->getBitDepth(toChannelType(compID)), setIdx, frameStats, tempCcSaoParam.offset);
+#if JVET_AJ0237_INTERNAL_12BIT
+ getCcSaoDistortion(compID, setIdx, blkStats, tempCcSaoParam.offset, trainingDistortion, shift);
+#else
getCcSaoDistortion(compID, setIdx, blkStats, tempCcSaoParam.offset, trainingDistortion);
+#endif
}
else
{
deriveCcSaoOffsets(compID, cs.sps->getBitDepth(toChannelType(compID)), setIdx, frameStatsEdge, tempCcSaoParam.offset);
+#if JVET_AJ0237_INTERNAL_12BIT
+ getCcSaoDistortion(compID, setIdx, blkStatsEdge, tempCcSaoParam.offset, trainingDistortion, shift);
+#else
getCcSaoDistortion(compID, setIdx, blkStatsEdge, tempCcSaoParam.offset, trainingDistortion);
+#endif
}
#else
deriveCcSaoOffsets(compID, cs.sps->getBitDepth(toChannelType(compID)), setIdx, frameStats,
diff --git a/source/Lib/EncoderLib/EncSampleAdaptiveOffset.h b/source/Lib/EncoderLib/EncSampleAdaptiveOffset.h
index 7cfd6e040..1ef5732cb 100644
--- a/source/Lib/EncoderLib/EncSampleAdaptiveOffset.h
+++ b/source/Lib/EncoderLib/EncSampleAdaptiveOffset.h
@@ -216,7 +216,11 @@ private: //methods
void deriveModeMergeRDO(const BitDepths &bitDepths, int ctuRsAddr, SAOBlkParam* mergeList[NUM_SAO_MERGE_TYPES], bool* sliceEnabled, std::vector<SAOStatData**>& blkStats, SAOBlkParam& modeParam, double& modeNormCost );
int64_t getDistortion(const int channelBitDepth, int typeIdc, int typeAuxInfo, int* offsetVal, SAOStatData& statData);
void deriveOffsets(ComponentID compIdx, const int channelBitDepth, int typeIdc, SAOStatData& statData, int* quantOffsets, int& typeAuxInfo);
+#if JVET_AJ0237_INTERNAL_12BIT
+ inline int64_t estSaoDist(int64_t count, int64_t offset, int64_t diffSum, int shift, int bdShift = 0);
+#else
inline int64_t estSaoDist(int64_t count, int64_t offset, int64_t diffSum, int shift);
+#endif
inline int estIterOffset(int typeIdx, double lambda, int offsetInput, int64_t count, int64_t diffSum, int shift, int bitIncrease, int64_t& bestDist, double& bestCost, int offsetTh );
void addPreDBFStatistics(std::vector<SAOStatData**>& blkStats);
#if JVET_W0066_CCSAO
@@ -335,8 +339,13 @@ private: //methods
, CcSaoStatData frameStats[MAX_CCSAO_SET_NUM]
, short offset[MAX_CCSAO_SET_NUM][MAX_CCSAO_CLASS_NUM]);
inline int estCcSaoIterOffset(const double lambda, const int offsetInput, const int64_t count, const int64_t diffSum, const int shift, const int bitIncrease, int64_t& bestDist, double& bestCost, const int offsetTh);
+#if JVET_AJ0237_INTERNAL_12BIT
+ void getCcSaoDistortion(const ComponentID compID, const int setIdx, CcSaoStatData* blkStats[MAX_CCSAO_SET_NUM]
+ , short offset[MAX_CCSAO_SET_NUM][MAX_CCSAO_CLASS_NUM], int64_t* trainingDistortion[MAX_CCSAO_SET_NUM], const int shift);
+#else
void getCcSaoDistortion(const ComponentID compID, const int setIdx, CcSaoStatData* blkStats[MAX_CCSAO_SET_NUM]
, short offset[MAX_CCSAO_SET_NUM][MAX_CCSAO_CLASS_NUM], int64_t* trainingDistortion[MAX_CCSAO_SET_NUM]);
+#endif
#if JVET_Y0106_CCSAO_EDGE_CLASSIFIER && !JVET_AE0151_CCSAO_HISTORY_OFFSETS_AND_EXT_EO
void getCcSaoDistortionEdge(const ComponentID compID, const int setIdx,
CcSaoStatData *blkStatsEdge[MAX_CCSAO_SET_NUM],
diff --git a/source/Lib/EncoderLib/EncTemporalFilter.cpp b/source/Lib/EncoderLib/EncTemporalFilter.cpp
index 5cbf713dd..17e202fec 100644
--- a/source/Lib/EncoderLib/EncTemporalFilter.cpp
+++ b/source/Lib/EncoderLib/EncTemporalFilter.cpp
@@ -349,6 +349,16 @@ void EncTemporalFilter::subsampleLuma(const PelStorage &input, PelStorage &outpu
output.extendBorderPel(m_padding, m_padding);
}
+#if JVET_AJ0237_INTERNAL_12BIT
+int64_t EncTemporalFilter::motionErrorLuma(const PelStorage& orig,
+ const PelStorage& buffer,
+ const int x,
+ const int y,
+ int dx,
+ int dy,
+ const int bs,
+ const int64_t besterror) const
+#else
int EncTemporalFilter::motionErrorLuma(const PelStorage &orig,
const PelStorage &buffer,
const int x,
@@ -357,13 +367,18 @@ int EncTemporalFilter::motionErrorLuma(const PelStorage &orig,
int dy,
const int bs,
const int besterror = 8 * 8 * 1024 * 1024) const
+#endif
{
const Pel* origOrigin = orig.Y().buf;
const int origStride = orig.Y().stride;
const Pel* buffOrigin = buffer.Y().buf;
const int buffStride = buffer.Y().stride;
+#if JVET_AJ0237_INTERNAL_12BIT
+ int64_t error = 0;
+#else
int error = 0;
+#endif
if (((dx | dy) & 0xF) == 0)
{
dx /= m_motionVectorFactor;
@@ -454,6 +469,12 @@ void EncTemporalFilter::motionEstimationLuma(Array2D<MotionVector> &mvs, const P
const int origWidth = orig.Y().width;
const int origHeight = orig.Y().height;
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int bitShift = 2 * (16 - m_internalBitDepth[CHANNEL_TYPE_LUMA]);
+ const int denorm = 204800 / (1 << bitShift);
+ const double offset = 20480 / (1 << bitShift);
+#endif
+
#if JVET_V0056
for (int blockY = 0; blockY + blockSize <= origHeight; blockY += stepSize)
{
@@ -489,7 +510,11 @@ void EncTemporalFilter::motionEstimationLuma(Array2D<MotionVector> &mvs, const P
if ((testx >= 0) && (testx < origWidth / (2 * blockSize)) && (testy >= 0) && (testy < origHeight / (2 * blockSize)))
{
MotionVector old = previous->get(testx, testy);
+#if JVET_AJ0237_INTERNAL_12BIT
+ int64_t error = motionErrorLuma(orig, buffer, blockX, blockY, old.x * factor, old.y * factor, blockSize, best.error);
+#else
int error = motionErrorLuma(orig, buffer, blockX, blockY, old.x * factor, old.y * factor, blockSize, best.error);
+#endif
if (error < best.error)
{
best.set(old.x * factor, old.y * factor, error);
@@ -498,7 +523,11 @@ void EncTemporalFilter::motionEstimationLuma(Array2D<MotionVector> &mvs, const P
}
}
#if JVET_V0056
+#if JVET_AJ0237_INTERNAL_12BIT
+ int64_t error = motionErrorLuma(orig, buffer, blockX, blockY, 0, 0, blockSize, best.error);
+#else
int error = motionErrorLuma(orig, buffer, blockX, blockY, 0, 0, blockSize, best.error);
+#endif
if (error < best.error)
{
best.set(0, 0, error);
@@ -510,7 +539,11 @@ void EncTemporalFilter::motionEstimationLuma(Array2D<MotionVector> &mvs, const P
{
for (int x2 = prevBest.x / m_motionVectorFactor - range; x2 <= prevBest.x / m_motionVectorFactor + range; x2++)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ int64_t error = motionErrorLuma(orig, buffer, blockX, blockY, x2 * m_motionVectorFactor, y2 * m_motionVectorFactor, blockSize, best.error);
+#else
int error = motionErrorLuma(orig, buffer, blockX, blockY, x2 * m_motionVectorFactor, y2 * m_motionVectorFactor, blockSize, best.error);
+#endif
if (error < best.error)
{
best.set(x2 * m_motionVectorFactor, y2 * m_motionVectorFactor, error);
@@ -525,7 +558,11 @@ void EncTemporalFilter::motionEstimationLuma(Array2D<MotionVector> &mvs, const P
{
for (int x2 = prevBest.x - doubleRange; x2 <= prevBest.x + doubleRange; x2 += 4)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ int64_t error = motionErrorLuma(orig, buffer, blockX, blockY, x2, y2, blockSize, best.error);
+#else
int error = motionErrorLuma(orig, buffer, blockX, blockY, x2, y2, blockSize, best.error);
+#endif
if (error < best.error)
{
best.set(x2, y2, error);
@@ -539,7 +576,11 @@ void EncTemporalFilter::motionEstimationLuma(Array2D<MotionVector> &mvs, const P
{
for (int x2 = prevBest.x - doubleRange; x2 <= prevBest.x + doubleRange; x2++)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ int64_t error = motionErrorLuma(orig, buffer, blockX, blockY, x2, y2, blockSize, best.error);
+#else
int error = motionErrorLuma(orig, buffer, blockX, blockY, x2, y2, blockSize, best.error);
+#endif
if (error < best.error)
{
best.set(x2, y2, error);
@@ -552,7 +593,11 @@ void EncTemporalFilter::motionEstimationLuma(Array2D<MotionVector> &mvs, const P
if (blockY > 0)
{
MotionVector aboveMV = mvs.get(blockX / stepSize, (blockY - stepSize) / stepSize);
+#if JVET_AJ0237_INTERNAL_12BIT
+ int64_t error = motionErrorLuma(orig, buffer, blockX, blockY, aboveMV.x, aboveMV.y, blockSize, best.error);
+#else
int error = motionErrorLuma(orig, buffer, blockX, blockY, aboveMV.x, aboveMV.y, blockSize, best.error);
+#endif
if (error < best.error)
{
best.set(aboveMV.x, aboveMV.y, error);
@@ -561,7 +606,11 @@ void EncTemporalFilter::motionEstimationLuma(Array2D<MotionVector> &mvs, const P
if (blockX > 0)
{
MotionVector leftMV = mvs.get((blockX - stepSize) / stepSize, blockY / stepSize);
+#if JVET_AJ0237_INTERNAL_12BIT
+ int64_t error = motionErrorLuma(orig, buffer, blockX, blockY, leftMV.x, leftMV.y, blockSize, best.error);
+#else
int error = motionErrorLuma(orig, buffer, blockX, blockY, leftMV.x, leftMV.y, blockSize, best.error);
+#endif
if (error < best.error)
{
best.set(leftMV.x, leftMV.y, error);
@@ -589,7 +638,11 @@ void EncTemporalFilter::motionEstimationLuma(Array2D<MotionVector> &mvs, const P
variance = variance + (pix - avg) * (pix - avg);
}
}
+#if JVET_AJ0237_INTERNAL_12BIT
+ best.error = (int)(20 * ((best.error + offset) / (variance + offset)) + (best.error / (blockSize * blockSize)) / denorm);
+#else
best.error = (int)(20 * ((best.error + 5.0) / (variance + 5.0)) + (best.error / (blockSize * blockSize)) / 50);
+#endif
#endif
mvs.get(blockX / stepSize, blockY / stepSize) = best;
}
@@ -731,6 +784,12 @@ void EncTemporalFilter::bilateralFilter(const PelStorage &orgPic,
const double weightScaling = overallStrength * (isChroma(compID) ? m_chromaFactor : 0.4);
const Pel maxSampleValue = (1 << m_internalBitDepth[toChannelType(compID)]) - 1;
const double bitDepthDiffWeighting = 1024.0 / (maxSampleValue + 1);
+
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int bitShift = 2 * (16 - m_internalBitDepth[toChannelType(compID)]);
+ const double offset = 20480 / (1 << bitShift);
+#endif
+
#if JVET_V0056
const int lumaBlockSize = 8;
const int csx = getComponentScaleX(compID, m_chromaFormatIDC);
@@ -783,7 +842,11 @@ void EncTemporalFilter::bilateralFilter(const PelStorage &orgPic,
const int cntV = blockSizeX * blockSizeY;
const int cntD = 2 * cntV - blockSizeX - blockSizeY;
srcFrameInfo[i].mvs.get( x / blockSizeX, y / blockSizeY ).noise =
+#if JVET_AJ0237_INTERNAL_12BIT
+ ( int ) round( (15.0 * cntD / cntV * variance + offset) / (diffsum + offset) );
+#else
( int ) round( (15.0 * cntD / cntV * variance + 5.0) / (diffsum + 5.0) );
+#endif
}
}
double minError = 9999999;
@@ -795,7 +858,11 @@ void EncTemporalFilter::bilateralFilter(const PelStorage &orgPic,
for (int i = 0; i < numRefs; i++)
{
#if JVET_V0056
+#if JVET_AJ0237_INTERNAL_12BIT
+ const int64_t error = srcFrameInfo[i].mvs.get(x / blockSizeX, y / blockSizeY).error;
+#else
const int error = srcFrameInfo[i].mvs.get(x / blockSizeX, y / blockSizeY).error;
+#endif
const int noise = srcFrameInfo[i].mvs.get(x / blockSizeX, y / blockSizeY).noise;
#endif
const Pel* pCorrectedPelPtr = srcFrameInfo[i].picBuffer.bufs[c].buf + (y * srcFrameInfo[i].picBuffer.bufs[c].stride + x);
diff --git a/source/Lib/EncoderLib/EncTemporalFilter.h b/source/Lib/EncoderLib/EncTemporalFilter.h
index 38b4f3279..c5cb475d3 100644
--- a/source/Lib/EncoderLib/EncTemporalFilter.h
+++ b/source/Lib/EncoderLib/EncTemporalFilter.h
@@ -50,14 +50,26 @@
struct MotionVector
{
int x, y;
+#if JVET_AJ0237_INTERNAL_12BIT
+ int64_t error;
+#else
int error;
+#endif
#if JVET_V0056
int noise;
+#if JVET_AJ0237_INTERNAL_12BIT
+ MotionVector() : x(0), y(0), error(INT_LEAST64_MAX), noise(0) {}
+#else
MotionVector() : x(0), y(0), error(INT_LEAST32_MAX), noise(0) {}
+#endif
#else
MotionVector() : x(0), y(0), error(INT_LEAST32_MAX) {}
#endif
+#if JVET_AJ0237_INTERNAL_12BIT
+ void set(int vectorX, int vectorY, int64_t errorValue) { x = vectorX; y = vectorY; error = errorValue; }
+#else
void set(int vectorX, int vectorY, int errorValue) { x = vectorX; y = vectorY; error = errorValue; }
+#endif
};
template <class T>
@@ -169,7 +181,11 @@ private:
// Private functions
void subsampleLuma(const PelStorage &input, PelStorage &output, const int factor = 2) const;
+#if JVET_AJ0237_INTERNAL_12BIT
+ int64_t motionErrorLuma(const PelStorage& orig, const PelStorage& buffer, const int x, const int y, int dx, int dy, const int bs, const int64_t besterror) const;
+#else
int motionErrorLuma(const PelStorage &orig, const PelStorage &buffer, const int x, const int y, int dx, int dy, const int bs, const int besterror) const;
+#endif
void motionEstimationLuma(Array2D<MotionVector> &mvs, const PelStorage &orig, const PelStorage &buffer, const int bs,
const Array2D<MotionVector> *previous=0, const int factor = 1, const bool doubleRes = false) const;
void motionEstimation(Array2D<MotionVector> &mvs, const PelStorage &orgPic, const PelStorage &buffer, const PelStorage &origSubsampled2, const PelStorage &origSubsampled4) const;
diff --git a/source/Lib/EncoderLib/InterSearch.cpp b/source/Lib/EncoderLib/InterSearch.cpp
index f4824391e..ff0f3e365 100644
--- a/source/Lib/EncoderLib/InterSearch.cpp
+++ b/source/Lib/EncoderLib/InterSearch.cpp
@@ -498,10 +498,18 @@ void InterSearch::init( EncCfg* pcEncCfg,
#if INTER_LIC || (TM_AMVP || TM_MRG || JVET_Z0084_IBC_TM) || JVET_W0090_ARMC_TM || JVET_Z0056_GPM_SPLIT_MODE_REORDERING
#if JVET_Z0153_IBC_EXT_REF
#if JVET_AJ0172_IBC_ITMP_ALIGN_REF_AREA
+#if JVET_AJ0237_INTERNAL_12BIT
+ InterPrediction::init( pcRdCost, cform, maxCUHeight, m_pcReshape, curPicWidthY, curPicHeightY, pcEncCfg->getBitDepth(CHANNEL_TYPE_LUMA));
+#else
InterPrediction::init( pcRdCost, cform, maxCUHeight, m_pcReshape, curPicWidthY, curPicHeightY );
+#endif
+#else
+#if JVET_AJ0237_INTERNAL_12BIT
+ InterPrediction::init( pcRdCost, cform, maxCUHeight, m_pcReshape, curPicWidthY, pcEncCfg->getBitDepth(CHANNEL_TYPE_LUMA));
#else
InterPrediction::init( pcRdCost, cform, maxCUHeight, m_pcReshape, curPicWidthY );
#endif
+#endif
#else
InterPrediction::init( pcRdCost, cform, maxCUHeight, m_pcReshape );
#endif
diff --git a/source/Lib/EncoderLib/IntraSearch.cpp b/source/Lib/EncoderLib/IntraSearch.cpp
index f8f9b5bd8..8453b1015 100644
--- a/source/Lib/EncoderLib/IntraSearch.cpp
+++ b/source/Lib/EncoderLib/IntraSearch.cpp
@@ -114,8 +114,10 @@ IntraSearch::IntraSearch()
m_dimdPredBuf = nullptr;
m_obicPredBuf = nullptr;
#endif
+#if !JVET_AJ0237_INTERNAL_12BIT
m_truncBinBits = nullptr;
m_escapeNumBins = nullptr;
+#endif
m_minErrorIndexMap = nullptr;
for (unsigned i = 0; i < (MAXPLTSIZE + 1); i++)
{
@@ -324,6 +326,7 @@ void IntraSearch::destroy()
m_obicPredBuf = nullptr;
#endif
m_isInitialized = false;
+#if !JVET_AJ0237_INTERNAL_12BIT
if (m_truncBinBits != nullptr)
{
for (unsigned i = 0; i < m_symbolSize; i++)
@@ -339,6 +342,7 @@ void IntraSearch::destroy()
delete[] m_escapeNumBins;
m_escapeNumBins = nullptr;
}
+#endif
if (m_indexError[0] != nullptr)
{
for (unsigned i = 0; i < (MAXPLTSIZE + 1); i++)
@@ -597,6 +601,7 @@ void IntraSearch::init( EncCfg* pcEncCfg,
m_isInitialized = true;
if (pcEncCfg->getPLTMode())
{
+#if !JVET_AJ0237_INTERNAL_12BIT
m_symbolSize = (1 << bitDepthY); // pixel values are within [0, SymbolSize-1] with size SymbolSize
if (m_truncBinBits == nullptr)
{
@@ -611,6 +616,7 @@ void IntraSearch::init( EncCfg* pcEncCfg,
m_escapeNumBins = new uint16_t[m_symbolSize];
}
initTBCTable(bitDepthY);
+#endif
if (m_indexError[0] == nullptr)
{
for (unsigned i = 0; i < (MAXPLTSIZE + 1); i++)
@@ -9119,7 +9125,11 @@ void IntraSearch::preCalcPLTIndexRD(CodingStructure& cs, Partitioner& partitione
{
if (lossless)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ rate += getEpExGolombNumBins(curPel[comp], 5);
+#else
rate += m_escapeNumBins[curPel[comp]];
+#endif
}
else
{
@@ -9132,7 +9142,11 @@ void IntraSearch::preCalcPLTIndexRD(CodingStructure& cs, Partitioner& partitione
{
error += tmpErr * tmpErr;
}
+#if JVET_AJ0237_INTERNAL_12BIT
+ rate += getEpExGolombNumBins(paPixelValue[comp], 5);
+#else
rate += m_escapeNumBins[paPixelValue[comp]]; // encode quantized escape color
+#endif
}
}
double rdCost = (double)error + m_pcRdCost->getLambda()*(double)rate;
@@ -9439,7 +9453,11 @@ double IntraSearch::rateDistOptPLT(
rdCost = MAX_DOUBLE;
return rdCost;
}
+#if JVET_AJ0237_INTERNAL_12BIT
+ rdCost += m_pcRdCost->getLambda() * (getTruncBinBits((runIndex > refIndex) ? runIndex - 1 : runIndex, (scanPos == 0) ? (indexMaxValue + 1) : indexMaxValue) << SCALE_BITS);
+#else
rdCost += m_pcRdCost->getLambda()*(m_truncBinBits[(runIndex > refIndex) ? runIndex - 1 : runIndex][(scanPos == 0) ? (indexMaxValue + 1) : indexMaxValue] << SCALE_BITS);
+#endif
}
rdCost += m_indexError[runIndex][m_scanOrder[scanPos].idx] * (1 << SCALE_BITS);
if (scanPos > 0)
@@ -9508,6 +9526,7 @@ uint32_t IntraSearch::getTruncBinBits(uint32_t symbol, uint32_t maxSymbol)
return idxCodeBit;
}
+#if !JVET_AJ0237_INTERNAL_12BIT
void IntraSearch::initTBCTable(int bitDepth)
{
for (uint32_t i = 0; i < m_symbolSize; i++)
@@ -9527,6 +9546,7 @@ void IntraSearch::initTBCTable(int bitDepth)
m_escapeNumBins[i] = getEpExGolombNumBins(i, 5);
}
}
+#endif
void IntraSearch::calcPixelPred(CodingStructure& cs, Partitioner& partitioner, uint32_t yPos, uint32_t xPos, ComponentID compBegin, uint32_t numComp)
{
@@ -9680,7 +9700,11 @@ void IntraSearch::derivePLTLossy(CodingStructure& cs, Partitioner& partitioner,
TransformUnit &tu = *cs.getTU(partitioner.chType);
QpParam cQP(tu, compBegin);
+#if JVET_AJ0237_INTERNAL_12BIT
+ int qp = cQP.Qp(true) - 6 * (channelBitDepth_L - 8);
+#else
int qp = cQP.Qp(true) - 12;
+#endif
qp = (qp < 0) ? 0 : ((qp > 56) ? 56 : qp);
int errorLimit = g_paletteQuant[qp];
if (lossless)
diff --git a/source/Lib/EncoderLib/IntraSearch.h b/source/Lib/EncoderLib/IntraSearch.h
index 081574c43..d08dc5706 100644
--- a/source/Lib/EncoderLib/IntraSearch.h
+++ b/source/Lib/EncoderLib/IntraSearch.h
@@ -742,9 +742,11 @@ protected:
CtxCache* m_ctxCache;
bool m_isInitialized;
+#if !JVET_AJ0237_INTERNAL_12BIT
uint32_t m_symbolSize;
uint16_t** m_truncBinBits;
uint16_t* m_escapeNumBins;
+#endif
bool m_bestEscape;
double* m_indexError[MAXPLTSIZE + 1];
uint8_t* m_minErrorIndexMap; // store the best index in terms of distortion for each pixel
@@ -932,7 +934,9 @@ protected:
void deriveIndexMap (CodingStructure& cs, Partitioner& partitioner, ComponentID compBegin, uint32_t numComp, PLTScanMode pltScanMode, double& dCost, bool* idxExist);
bool 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);
double rateDistOptPLT (bool RunType, uint8_t RunIndex, bool prevRunType, uint8_t prevRunIndex, uint8_t aboveRunIndex, bool& prevCodedRunType, int& prevCodedRunPos, int scanPos, uint32_t width, int dist, int indexMaxValue, const BinFracBits* IndexfracBits, const BinFracBits& TypefracBits);
+#if !JVET_AJ0237_INTERNAL_12BIT
void initTBCTable (int bitDepth);
+#endif
uint32_t getTruncBinBits (uint32_t symbol, uint32_t maxSymbol);
uint32_t getEpExGolombNumBins (uint32_t symbol, uint32_t count);
void xGetNextISPMode ( ModeInfo& modeInfo, const ModeInfo* lastMode, const Size cuSize );
diff --git a/source/Lib/EncoderLib/VLCWriter.cpp b/source/Lib/EncoderLib/VLCWriter.cpp
index b92443151..af74cd1fb 100644
--- a/source/Lib/EncoderLib/VLCWriter.cpp
+++ b/source/Lib/EncoderLib/VLCWriter.cpp
@@ -3604,7 +3604,11 @@ void HLSWriter::codeSliceHeader ( Slice* pcSlice )
bool lambdaCanBePredicted = false;
if (index !=-1)
{
+#if JVET_AJ0237_INTERNAL_12BIT
+ if (pcSlice->getSPS()->getLambdaVal(index) == pcSlice->getCostForARMC(pcSlice->getSPS()->getBitDepth(CHANNEL_TYPE_LUMA)))
+#else
if(pcSlice->getSPS()->getLambdaVal(index) == pcSlice->getCostForARMC())
+#endif
{
lambdaCanBePredicted = true;
}
@@ -3612,7 +3616,11 @@ void HLSWriter::codeSliceHeader ( Slice* pcSlice )
if (!lambdaCanBePredicted)
{
#if JVET_AB0082
+#if JVET_AJ0237_INTERNAL_12BIT
+ WRITE_CODE(pcSlice->getCostForARMC(pcSlice->getSPS()->getBitDepth(CHANNEL_TYPE_LUMA)), 10, "Lambda");
+#else
WRITE_CODE(pcSlice->getCostForARMC(), 10, "Lambda");
+#endif
#else
WRITE_CODE(pcSlice->getCostForARMC(), 9, "Lambda");
#endif
@@ -3648,6 +3656,9 @@ void HLSWriter::codeSliceHeader ( Slice* pcSlice )
{
clipDeltaShift = ADAPTIVE_CLIP_SHIFT_DELTA_VALUE_0;
}
+#if JVET_AJ0237_INTERNAL_12BIT
+ clipDeltaShift += std::max(0, pcSlice->getSPS()->getBitDepth(toChannelType(COMPONENT_Y)) - 10);
+#endif
if (pcSlice->getSliceType() == I_SLICE)
{
#if JVET_AI0096_ADAPTIVE_CLIPPING_BIT_DEPTH_FIX
--
GitLab