diff --git a/source/Lib/CommonLib/BilateralFilter.cpp b/source/Lib/CommonLib/BilateralFilter.cpp
index 9b7137b869443d5ebe96691de5914951f6177fba..7d3b615d807a069a526274e99654f6ea84e5d8c5 100644
--- a/source/Lib/CommonLib/BilateralFilter.cpp
+++ b/source/Lib/CommonLib/BilateralFilter.cpp
@@ -53,6 +53,9 @@
BilateralFilter::BilateralFilter()
{
m_bilateralFilterDiamond5x5 = blockBilateralFilterDiamond5x5;
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ m_calcMAD = calcMAD;
+#endif
#if ENABLE_SIMD_BILATERAL_FILTER || JVET_X0071_CHROMA_BILATERAL_FILTER_ENABLE_SIMD
#ifdef TARGET_SIMD_X86
@@ -67,14 +70,53 @@ BilateralFilter::~BilateralFilter()
void BilateralFilter::create()
{
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ for (int qp = 0; qp < 26; qp++)
+ {
+ for (int i = 0; i < 16; i++)
+ {
+#if JVET_V0094_BILATERAL_FILTER
+ m_lut[qp][i] = ((int)m_wBIF[qp][i] * m_distFactor[0] + 4) >> 3;
+ m_lut[qp][16 + i] = ((int)m_wBIF[qp][i] * m_distFactor[1] + 4) >> 3;
+ m_lut[qp][32 + i] = ((int)m_wBIF[qp][i] * m_distFactor[2] + 4) >> 3;
+#endif
+#if JVET_X0071_CHROMA_BILATERAL_FILTER
+ m_lutChroma[qp][i] = ((int)m_wBIFChroma[qp][i] * m_distFactorChroma[0] + 4) >> 3;
+ m_lutChroma[qp][16 + i] = ((int)m_wBIFChroma[qp][i] * m_distFactorChroma[1] + 4) >> 3;
+ m_lutChroma[qp][32 + i] = ((int)m_wBIFChroma[qp][i] * m_distFactorChroma[2] + 4) >> 3;
+#endif
+ }
+ }
+#endif
}
void BilateralFilter::destroy()
{
}
#if JVET_V0094_BILATERAL_FILTER
-const char* BilateralFilter::getFilterLutParameters( const int size, const PredMode predMode, const int32_t qp, int& bfac )
+const char* BilateralFilter::getFilterLutParameters(int16_t* block, const int stride, const int width, const int height, const PredMode predMode, const int32_t qp, int& bfac)
{
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ int w = floorLog2(width);
+ int h = floorLog2(height);
+ int mad = m_calcMAD(block, stride, width, height, w + h);
+
+ w = std::min(w, 7);
+ h = std::min(h, 7);
+ mad = std::min(mad >> 4, 15);
+
+ bfac = m_tuSizeFactor[predMode == MODE_INTER][w * 8 + h];
+ if (bfac) // BIF is not applied if tuSizeFactor[(w, h)] = 0
+ {
+ bfac += m_tuMADFactor[predMode == MODE_INTER][mad];
+ }
+
+ int sqp = qp - 17;
+ sqp = std::min(sqp, 25);
+ sqp = std::max(sqp, 0);
+ return m_lut[sqp];
+#else
+ const int size = std::min(width, height);
if( size <= 4 )
{
bfac = 3;
@@ -117,99 +159,81 @@ const char* BilateralFilter::getFilterLutParameters( const int size, const PredM
}
return m_wBIF[sqp - 17];
+#endif
}
#endif
-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 )
+inline void bifApplyLut(int diff, int& res, int cutBitsNum, int bitsRound, int bitsRound2, int shift, const char* lutRowPtr, int lutShift)
+{
+ int sg0 = diff >> shift;
+ int v0 = (diff + sg0) ^ sg0;
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ int idx = (v0 + bitsRound) >> cutBitsNum;
+ idx = 15 + ((idx - 15) & ((idx - 15) >> shift));
+ int idx2 = (v0 + bitsRound2) >> cutBitsNum;
+ idx2 = 15 + ((idx2 - 15) & ((idx2 - 15) >> shift));
+ int w0 = (lutRowPtr[lutShift + idx] + lutRowPtr[lutShift + idx2] + 1) >> 1;
+#else
+ int idx = (v0 + 4) >> 3;
+ idx = 15 + ((idx - 15) & ((idx - 15) >> shift));
+ int w0 = lutRowPtr[idx] >> lutShift;
+#endif
+ res = (w0 + sg0) ^ sg0;
+}
+
+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 pad = 2;
+ int pad = NUMBER_PADDED_SAMPLES;
int padwidth = iWidthExtSIMD;
- int downbuffer[64];
- int downleftbuffer[65];
- int downrightbuffer[2][65];
- int shift, sg0, v0, idx, w0;
- shift = sizeof( int ) * 8 - 1;
- downbuffer[0] = 0;
+ int downbuffer[128];
+ int downleftbuffer[129];
+ int downrightbuffer[2][129];
+ int shift = sizeof( int ) * 8 - 1;
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ int lutShift1 = 0, lutShift2 = 16, lutShift3 = 32;
+ int bitsRound = 1 << (cutBitsNum - 2);
+ int bitsRound2 = bitsRound + (1 << (cutBitsNum - 1));
+#else
+ int lutShift1 = 0, lutShift2 = 1, lutShift3 = 1;
+ int bitsRound = 4;
+ int bitsRound2 = 4;
+#endif
for( int x = 0; x < uiWidth; x++ )
{
int pixel = block[(-1 + pad)*padwidth + x + pad];
int below = block[(-1 + pad + 1)*padwidth + x + pad];
- int diff = below - pixel;
- sg0 = diff >> shift;
- v0 = (diff + sg0) ^ sg0;
- v0 = (v0 + 4) >> 3;
- idx = 15 + ((v0 - 15)&((v0 - 15) >> shift));
- w0 = lutRowPtr[idx];
- int mod = (w0 + sg0) ^ sg0;
- downbuffer[x] = mod;
+ bifApplyLut(below - pixel, downbuffer[x], cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift1);
int belowright = block[(-1 + pad + 1)*padwidth + x + pad + 1];
- diff = belowright - pixel;
- sg0 = diff >> shift;
- v0 = (diff + sg0) ^ sg0;
- v0 = (v0 + 4) >> 3;
- idx = 15 + ((v0 - 15)&((v0 - 15) >> shift));
- w0 = lutRowPtr[idx] >> 1;
- mod = (w0 + sg0) ^ sg0;
- downrightbuffer[1][x + 1] = mod;
+ bifApplyLut(belowright - pixel, downrightbuffer[1][x + 1], cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift2);
int belowleft = block[(-1 + pad + 1)*padwidth + x + pad - 1];
- diff = belowleft - pixel;
- sg0 = diff >> shift;
- v0 = (diff + sg0) ^ sg0;
- v0 = (v0 + 4) >> 3;
- idx = 15 + ((v0 - 15)&((v0 - 15) >> shift));
- w0 = lutRowPtr[idx] >> 1;
- mod = (w0 + sg0) ^ sg0;
- downleftbuffer[x] = mod;
+ bifApplyLut(belowleft - pixel, downleftbuffer[x], cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift2);
}
int width = uiWidth;
for( int y = 0; y < uiHeight; y++ )
{
- int diff;
-
int16_t *rowStart = &block[(y + pad)*padwidth + pad];
int pixel = rowStart[-1];
- int right = rowStart[0];
- diff = right - pixel;
- sg0 = diff >> shift;
- v0 = (diff + sg0) ^ sg0;
- v0 = (v0 + 4) >> 3;
- idx = 15 + ((v0 - 15)&((v0 - 15) >> shift));
- w0 = lutRowPtr[idx];
- int mod = (w0 + sg0) ^ sg0;
- int rightmod = mod;
+ int right = rowStart[0], rightmod = 0;
+ bifApplyLut(right - pixel, rightmod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift1);
pixel = rowStart[-padwidth - 1];
int belowright = right;
- diff = belowright - pixel;
- sg0 = diff >> shift;
- v0 = (diff + sg0) ^ sg0;
- v0 = (v0 + 4) >> 3;
- idx = 15 + ((v0 - 15)&((v0 - 15) >> shift));
- w0 = lutRowPtr[idx] >> 1;
- mod = (w0 + sg0) ^ sg0;
- downrightbuffer[(y + 1) % 2][0] = mod;
+ bifApplyLut(belowright - pixel, downrightbuffer[(y + 1) % 2][0], cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift2);
pixel = rowStart[-padwidth + width];
int belowleft = rowStart[width - 1];
- diff = belowleft - pixel;
- sg0 = diff >> shift;
- v0 = (diff + sg0) ^ sg0;
- v0 = (v0 + 4) >> 3;
- idx = 15 + ((v0 - 15)&((v0 - 15) >> shift));
- w0 = lutRowPtr[idx] >> 1;
- mod = (w0 + sg0) ^ sg0;
- downleftbuffer[width] = mod;
+ bifApplyLut(belowleft - pixel, downleftbuffer[width], cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift2);
for( int x = 0; x < uiWidth; x++ )
{
pixel = rowStart[x];
- int modsum = 0;
+ int modsum = 0, mod = 0;
int abovemod = -downbuffer[x];
modsum += abovemod;
@@ -218,57 +242,28 @@ void BilateralFilter::blockBilateralFilterDiamond5x5( uint32_t uiWidth, uint32_t
modsum += leftmod;
right = rowStart[x + 1];
- diff = right - pixel;
- sg0 = diff >> shift;
- v0 = (diff + sg0) ^ sg0;
- v0 = (v0 + 4) >> 3;
- idx = 15 + ((v0 - 15)&((v0 - 15) >> shift));
- w0 = lutRowPtr[idx];
- mod = (w0 + sg0) ^ sg0;
-
+ bifApplyLut(right - pixel, mod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift1);
modsum += mod;
rightmod = mod;
int below = rowStart[x + padwidth];
- diff = below - pixel;
- sg0 = diff >> shift;
- v0 = (diff + sg0) ^ sg0;
- v0 = (v0 + 4) >> 3;
- idx = 15 + ((v0 - 15)&((v0 - 15) >> shift));
- w0 = lutRowPtr[idx];
- mod = (w0 + sg0) ^ sg0;
+ bifApplyLut(below - pixel, mod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift1);
modsum += mod;
downbuffer[x] = mod;
int aboverightmod = -downleftbuffer[x + 1];
- // modsum += ((int16_t)((uint16_t)((aboverightmod) >> 1)));
modsum += aboverightmod;
int aboveleftmod = -downrightbuffer[(y + 1) % 2][x];
- // modsum += ((int16_t)((uint16_t)((aboveleftmod) >> 1)));
modsum += aboveleftmod;
int belowleft = rowStart[x + padwidth - 1];
- diff = belowleft - pixel;
- sg0 = diff >> shift;
- v0 = (diff + sg0) ^ sg0;
- v0 = (v0 + 4) >> 3;
- idx = 15 + ((v0 - 15)&((v0 - 15) >> shift));
- w0 = lutRowPtr[idx] >> 1;
- mod = (w0 + sg0) ^ sg0;
- // modsum += ((int16_t)((uint16_t)((mod) >> 1)));
+ bifApplyLut(belowleft - pixel, mod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift2);
modsum += mod;
downleftbuffer[x] = mod;
int belowright = rowStart[x + padwidth + 1];
- diff = belowright - pixel;
- sg0 = diff >> shift;
- v0 = (diff + sg0) ^ sg0;
- v0 = (v0 + 4) >> 3;
- idx = 15 + ((v0 - 15)&((v0 - 15) >> shift));
- w0 = lutRowPtr[idx] >> 1;
- mod = (w0 + sg0) ^ sg0;
- //modsum += ((int16_t)((uint16_t)((mod) >> 1)));
+ bifApplyLut(belowright - pixel, mod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift2);
modsum += mod;
downrightbuffer[y % 2][x + 1] = mod;
@@ -277,56 +272,34 @@ void BilateralFilter::blockBilateralFilterDiamond5x5( uint32_t uiWidth, uint32_t
// speed when SIMD is turned off.
int above = rowStart[x - 2 * padwidth];
- diff = above - pixel;
- sg0 = diff >> shift;
- v0 = (diff + sg0) ^ sg0;
- v0 = (v0 + 4) >> 3;
- idx = 15 + ((v0 - 15)&((v0 - 15) >> shift));
- w0 = lutRowPtr[idx] >> 1;
- mod = (w0 + sg0) ^ sg0;
+ bifApplyLut(above - pixel, mod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift3);
modsum += mod;
below = rowStart[x + 2 * padwidth];
- diff = below - pixel;
- sg0 = diff >> shift;
- v0 = (diff + sg0) ^ sg0;
- v0 = (v0 + 4) >> 3;
- idx = 15 + ((v0 - 15)&((v0 - 15) >> shift));
- w0 = lutRowPtr[idx] >> 1;
- mod = (w0 + sg0) ^ sg0;
+ bifApplyLut(below - pixel, mod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift3);
modsum += mod;
int left = rowStart[x - 2];
- diff = left - pixel;
- sg0 = diff >> shift;
- v0 = (diff + sg0) ^ sg0;
- v0 = (v0 + 4) >> 3;
- idx = 15 + ((v0 - 15)&((v0 - 15) >> shift));
- w0 = lutRowPtr[idx] >> 1;
- mod = (w0 + sg0) ^ sg0;
+ bifApplyLut(left - pixel, mod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift3);
modsum += mod;
right = rowStart[x + 2];
- diff = right - pixel;
- sg0 = diff >> shift;
- v0 = (diff + sg0) ^ sg0;
- v0 = (v0 + 4) >> 3;
- idx = 15 + ((v0 - 15)&((v0 - 15) >> shift));
- w0 = lutRowPtr[idx] >> 1;
- mod = (w0 + sg0) ^ sg0;
+ bifApplyLut(right - pixel, mod, cutBitsNum, bitsRound, bitsRound2, shift, lutRowPtr, lutShift3);
modsum += mod;
- blkFilt[(y + pad)*(padwidth + 4) + x + pad] = (( int16_t ) (( uint16_t ) ((modsum*bfac + bifRoundAdd) >> bifRoundShift)));
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ blkFilt[(y + pad) * padwidth + x + pad] = ((int16_t)((uint16_t)((modsum * bfac + (bifRoundAdd << 3)) >> (bifRoundShift + 3))));
+#else
+ blkFilt[(y + pad) * padwidth + x + pad] = (( int16_t ) (( uint16_t ) ((modsum*bfac + bifRoundAdd) >> bifRoundShift)));
+#endif
}
}
// Copy back
- Pel *tempBlockPtr = ( short* ) blkFilt + (((padwidth + 4) << 1) + 2);
- int tempBlockStride = padwidth + 4;
+ Pel* tempBlockPtr = blkFilt + pad * padwidth + pad;
if( isRDO )
{
- Pel *srcBlockPtr = ( short* ) block + (((padwidth) << 1) + 2);
- int srcBlockStride = padwidth;
+ Pel *srcBlockPtr = block + pad * padwidth + pad;
for( uint32_t yy = 0; yy < uiHeight; yy++ )
{
for( uint32_t xx = 0; xx < uiWidth; xx++ )
@@ -334,8 +307,8 @@ void BilateralFilter::blockBilateralFilterDiamond5x5( uint32_t uiWidth, uint32_t
recPtr[xx] = ClipPel( srcBlockPtr[xx] + tempBlockPtr[xx], clpRng );
}
recPtr += recStride;
- tempBlockPtr += tempBlockStride;
- srcBlockPtr += srcBlockStride;
+ tempBlockPtr += padwidth;
+ srcBlockPtr += padwidth;
}
}
else if( noClip )
@@ -349,7 +322,7 @@ void BilateralFilter::blockBilateralFilterDiamond5x5( uint32_t uiWidth, uint32_t
recPtr[xx] = recPtr[xx] + tempBlockPtr[xx]; // clipping is done jointly for SAO/BIF/CCSAO
}
recPtr += recStride;
- tempBlockPtr += tempBlockStride;
+ tempBlockPtr += padwidth;
}
}
else
@@ -362,41 +335,44 @@ void BilateralFilter::blockBilateralFilterDiamond5x5( uint32_t uiWidth, uint32_t
recPtr[xx] = ClipPel<int>( recPtr[xx] + tempBlockPtr[xx], clpRng );
}
recPtr += recStride;
- tempBlockPtr += tempBlockStride;
+ tempBlockPtr += padwidth;
}
}
}
-#if JVET_V0094_BILATERAL_FILTER
-void BilateralFilter::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, std::vector<Pel>* pLUT)
-{
- uint32_t uiWidth = predBuf.width;
- uint32_t uiHeight = predBuf.height;
-
- int bfac = 1;
- const int bifRoundAdd = BIF_ROUND_ADD >> currTU.cs->pps->getBIFStrength();
- const int bifRoundShift = BIF_ROUND_SHIFT - currTU.cs->pps->getBIFStrength();
-
- const char* lutRowPtr = nullptr;
- if( isLuma( compID ) )
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+int BilateralFilter::calcMAD(int16_t* block, int stride, int width, int height, int whlog2)
+{
+ int average = 0;
+ for (int i = 0; i < height; i++)
{
- lutRowPtr = getFilterLutParameters( std::min( uiWidth, uiHeight ), currTU.cu->predMode, qp + currTU.cs->pps->getBIFQPOffset(), bfac );
+ for (int j = 0; j < width; j++)
+ {
+ average += block[j];
+ }
+ block += stride;
}
- else
+ block -= stride * height;
+ average = (average + (1 << (whlog2 - 1))) >> whlog2;
+ int mad = 0;
+ for (int i = 0; i < height; i++)
{
- int widthForStrength = currTU.blocks[compID].width;
- int heightForStrength = currTU.blocks[compID].height;
-
- if( currTU.blocks[COMPONENT_Y].valid() )
+ for (int j = 0; j < width; j++)
{
- widthForStrength = currTU.blocks[COMPONENT_Y].width;
- heightForStrength = currTU.blocks[COMPONENT_Y].height;
+ mad += std::abs(block[j] - average);
}
-
- lutRowPtr = getFilterLutParametersChroma( std::min( uiWidth, uiHeight ), currTU.cu->predMode, qp + currTU.cs->pps->getChromaBIFQPOffset(), bfac, widthForStrength, heightForStrength, currTU.blocks[COMPONENT_Y].valid() );
-
- CHECK( doReshape, "Reshape domain is not used for chroma" );
+ block += stride;
}
+ mad = (mad + (1 << (whlog2 - 1))) >> whlog2;
+ return mad;
+}
+#endif
+
+#if JVET_V0094_BILATERAL_FILTER
+void BilateralFilter::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, std::vector<Pel>* pLUT)
+{
+ uint32_t uiWidth = predBuf.width;
+ uint32_t uiHeight = predBuf.height;
const unsigned uiPredStride = predBuf.stride;
const unsigned uiStrideRes = resiBuf.stride;
@@ -416,11 +392,7 @@ void BilateralFilter::bilateralFilterRDOdiamond5x5(const ComponentID compID, Pel
const uint32_t uiWidthExt = uiWidth + (NUMBER_PADDED_SAMPLES << 1);
const uint32_t uiHeightExt = uiHeight + (NUMBER_PADDED_SAMPLES << 1);
- int iWidthExtSIMD = uiWidthExt | 0x04;
- if( uiWidth < 8 )
- {
- iWidthExtSIMD = 8 + (NUMBER_PADDED_SAMPLES << 1);
- }
+ int iWidthExtSIMD = uiWidthExt + 16;
memset(tempblock, 0, iWidthExtSIMD*uiHeightExt * sizeof(Pel));
Pel *tempBlockPtr = tempblock + NUMBER_PADDED_SAMPLES* iWidthExtSIMD + NUMBER_PADDED_SAMPLES;
@@ -589,7 +561,38 @@ void BilateralFilter::bilateralFilterRDOdiamond5x5(const ComponentID compID, Pel
std::copy( tempblock + iWidthExtSIMD, tempblock + iWidthExtSIMD + uiWidthExt, tempblock );
std::copy( tempblock + iWidthExtSIMD * ( uiHeightExt - 2 ), tempblock + iWidthExtSIMD * ( uiHeightExt - 2 ) + uiWidthExt, tempblock + iWidthExtSIMD * ( uiHeightExt - 1 ) );
- m_bilateralFilterDiamond5x5(uiWidth, uiHeight, tempblock, tempblockFiltered, clpRng, piReco, uiRecStride, iWidthExtSIMD, bfac, bifRoundAdd, bifRoundShift, true, lutRowPtr, false );
+ int bfac = 1;
+ const int bifRoundAdd = BIF_ROUND_ADD >> currTU.cs->pps->getBIFStrength();
+ const int bifRoundShift = BIF_ROUND_SHIFT - currTU.cs->pps->getBIFStrength();
+
+ const char* lutRowPtr = nullptr;
+ int cutBitsNum = 3;
+
+ if (isLuma(compID))
+ {
+ lutRowPtr = getFilterLutParameters(tempblock + iWidthExtSIMD * 2 + 2, iWidthExtSIMD, uiWidth, uiHeight, currTU.cu->predMode, qp + currTU.cs->pps->getBIFQPOffset(), bfac);
+ }
+ else
+ {
+#if JVET_X0071_CHROMA_BILATERAL_FILTER
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ cutBitsNum = 2;
+#endif
+ int widthForStrength = currTU.blocks[compID].width;
+ int heightForStrength = currTU.blocks[compID].height;
+
+ if (currTU.blocks[COMPONENT_Y].valid())
+ {
+ widthForStrength = currTU.blocks[COMPONENT_Y].width;
+ heightForStrength = currTU.blocks[COMPONENT_Y].height;
+ }
+
+ lutRowPtr = getFilterLutParametersChroma(tempblock + iWidthExtSIMD * 2 + 2, iWidthExtSIMD, uiWidth, uiHeight, currTU.cu->predMode, qp + currTU.cs->pps->getChromaBIFQPOffset(), bfac, widthForStrength, heightForStrength, currTU.blocks[COMPONENT_Y].valid());
+
+ CHECK(doReshape, "Reshape domain is not used for chroma");
+#endif
+ }
+ m_bilateralFilterDiamond5x5(uiWidth, uiHeight, tempblock, tempblockFiltered, clpRng, piReco, uiRecStride, iWidthExtSIMD, bfac, bifRoundAdd, bifRoundShift, true, lutRowPtr, false, cutBitsNum);
if( !useReco )
{
@@ -659,29 +662,6 @@ void BilateralFilter::bilateralFilterDiamond5x5( const ComponentID compID, const
int recStride = rec.get( compID ).stride;
Pel *recPtr = rec.get( compID ).bufAt(blkDst);
- int bfac = 1;
- const char *lutRowPtr = nullptr;
- if( isLuma( compID ) )
- {
- lutRowPtr = getFilterLutParameters( std::min( width, height ), currTU.cu->predMode, qp + currTU.cs->pps->getBIFQPOffset(), bfac );
- }
- else
- {
- int widthForStrength = currTU.blocks[compID].width;
- int heightForStrength = currTU.blocks[compID].height;
-
- if( currTU.blocks[COMPONENT_Y].valid() )
- {
- widthForStrength = currTU.blocks[COMPONENT_Y].width;
- heightForStrength = currTU.blocks[COMPONENT_Y].height;
- }
-
- lutRowPtr = getFilterLutParametersChroma(std::min( uiWidth, uiHeight ), currTU.cu->predMode, qp + currTU.cs->pps->getChromaBIFQPOffset(), bfac, widthForStrength, heightForStrength, currTU.blocks[COMPONENT_Y].valid() );
- }
-
- int bifRoundAdd = BIF_ROUND_ADD >> currTU.cs->pps->getBIFStrength();
- int bifRoundShift = BIF_ROUND_SHIFT - currTU.cs->pps->getBIFStrength();
-
bool topAltAvailable = !clipT;
bool leftAltAvailable = !clipL;
@@ -708,11 +688,7 @@ void BilateralFilter::bilateralFilterDiamond5x5( const ComponentID compID, const
uint32_t uiWidthExt = uiWidth + (NUMBER_PADDED_SAMPLES << 1);
uint32_t uiHeightExt = uiHeight + (NUMBER_PADDED_SAMPLES << 1);
- int iWidthExtSIMD = uiWidthExt | 0x04;
- if (uiWidth < 8)
- {
- iWidthExtSIMD = 8 + (NUMBER_PADDED_SAMPLES << 1);
- }
+ int iWidthExtSIMD = uiWidthExt + 16;
Pel *tempBlockPtr;
@@ -897,7 +873,36 @@ void BilateralFilter::bilateralFilterDiamond5x5( const ComponentID compID, const
}
}
- m_bilateralFilterDiamond5x5( uiWidth, uiHeight, tempblock, tempblockFiltered, clpRng, recPtr, recStride, iWidthExtSIMD, bfac, bifRoundAdd, bifRoundShift, false, lutRowPtr, false );
+ int bfac = 1;
+ const char* lutRowPtr = nullptr;
+ int cutBitsNum = 3;
+ if (isLuma(compID))
+ {
+ lutRowPtr = getFilterLutParameters(tempblock + iWidthExtSIMD * 2 + 2, iWidthExtSIMD, uiWidth, uiHeight, currTU.cu->predMode, qp + currTU.cs->pps->getBIFQPOffset(), bfac);
+ }
+ else
+ {
+#if JVET_X0071_CHROMA_BILATERAL_FILTER
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ cutBitsNum = 2;
+#endif
+ int widthForStrength = currTU.blocks[compID].width;
+ int heightForStrength = currTU.blocks[compID].height;
+
+ if (currTU.blocks[COMPONENT_Y].valid())
+ {
+ widthForStrength = currTU.blocks[COMPONENT_Y].width;
+ heightForStrength = currTU.blocks[COMPONENT_Y].height;
+ }
+
+ lutRowPtr = getFilterLutParametersChroma(tempblock + iWidthExtSIMD * 2 + 2, iWidthExtSIMD, uiWidth, uiHeight, currTU.cu->predMode, qp + currTU.cs->pps->getChromaBIFQPOffset(), bfac, widthForStrength, heightForStrength, currTU.blocks[COMPONENT_Y].valid());
+#endif
+ }
+
+ int bifRoundAdd = BIF_ROUND_ADD >> currTU.cs->pps->getBIFStrength();
+ int bifRoundShift = BIF_ROUND_SHIFT - currTU.cs->pps->getBIFStrength();
+
+ m_bilateralFilterDiamond5x5(uiWidth, uiHeight, tempblock, tempblockFiltered, clpRng, recPtr, recStride, iWidthExtSIMD, bfac, bifRoundAdd, bifRoundShift, false, lutRowPtr, false, cutBitsNum);
xStart = xEnd;
}
@@ -923,31 +928,6 @@ void BilateralFilter::bilateralFilterDiamond5x5( const ComponentID compID, const
int recStride = rec.get( compID ).stride;
Pel *recPtr = rec.get( compID ).bufAt(compArea);
- int bfac = 1;
-
- const char *lutRowPtr = nullptr;
-
- if( isLuma( compID ) )
- {
- lutRowPtr = getFilterLutParameters(std::min(uiWidth, uiHeight), currTU.cu->predMode, qp + currTU.cs->pps->getBIFQPOffset(), bfac);
- }
- else
- {
- int widthForStrength = currTU.blocks[compID].width;
- int heightForStrength = currTU.blocks[compID].height;
-
- if( currTU.blocks[COMPONENT_Y].valid() )
- {
- widthForStrength = currTU.blocks[COMPONENT_Y].width;
- heightForStrength = currTU.blocks[COMPONENT_Y].height;
- }
-
- lutRowPtr = getFilterLutParametersChroma( std::min( uiWidth, uiHeight ), currTU.cu->predMode, qp + currTU.cs->pps->getChromaBIFQPOffset(), bfac, widthForStrength, heightForStrength, currTU.blocks[COMPONENT_Y].valid() );
- }
-
- int bifRoundAdd = BIF_ROUND_ADD >> currTU.cs->pps->getBIFStrength();
- int bifRoundShift = BIF_ROUND_SHIFT - currTU.cs->pps->getBIFStrength();
-
const CompArea &myArea = currTU.blocks[compID];
topAltAvailable = myArea.y - NUMBER_PADDED_SAMPLES >= 0;
leftAltAvailable = myArea.x - NUMBER_PADDED_SAMPLES >= 0;
@@ -962,11 +942,7 @@ void BilateralFilter::bilateralFilterDiamond5x5( const ComponentID compID, const
uint32_t uiWidthExt = uiWidth + (NUMBER_PADDED_SAMPLES << 1);
uint32_t uiHeightExt = uiHeight + (NUMBER_PADDED_SAMPLES << 1);
- int iWidthExtSIMD = uiWidthExt | 0x04;
- if (uiWidth < 8)
- {
- iWidthExtSIMD = 8 + (NUMBER_PADDED_SAMPLES << 1);
- }
+ int iWidthExtSIMD = uiWidthExt + 16;
Pel *tempBlockPtr;
@@ -1163,7 +1139,37 @@ void BilateralFilter::bilateralFilterDiamond5x5( const ComponentID compID, const
}
}
- m_bilateralFilterDiamond5x5(uiWidth, uiHeight, tempblock, tempblockFiltered, clpRng, recPtr, recStride, iWidthExtSIMD, bfac, bifRoundAdd, bifRoundShift, false, lutRowPtr, noClip);
+ int bfac = 1;
+ const char* lutRowPtr = nullptr;
+ int cutBitsNum = 3;
+
+ if (isLuma(compID))
+ {
+ lutRowPtr = getFilterLutParameters(tempblock + iWidthExtSIMD * 2 + 2, iWidthExtSIMD, uiWidth, uiHeight, currTU.cu->predMode, qp + currTU.cs->pps->getBIFQPOffset(), bfac);
+ }
+ else
+ {
+#if JVET_X0071_CHROMA_BILATERAL_FILTER
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ cutBitsNum = 2;
+#endif
+ int widthForStrength = currTU.blocks[compID].width;
+ int heightForStrength = currTU.blocks[compID].height;
+
+ if (currTU.blocks[COMPONENT_Y].valid())
+ {
+ widthForStrength = currTU.blocks[COMPONENT_Y].width;
+ heightForStrength = currTU.blocks[COMPONENT_Y].height;
+ }
+
+ lutRowPtr = getFilterLutParametersChroma(tempblock + iWidthExtSIMD * 2 + 2, iWidthExtSIMD, uiWidth, uiHeight, currTU.cu->predMode, qp + currTU.cs->pps->getChromaBIFQPOffset(), bfac, widthForStrength, heightForStrength, currTU.blocks[COMPONENT_Y].valid());
+#endif
+ }
+
+ int bifRoundAdd = BIF_ROUND_ADD >> currTU.cs->pps->getBIFStrength();
+ int bifRoundShift = BIF_ROUND_SHIFT - currTU.cs->pps->getBIFStrength();
+
+ m_bilateralFilterDiamond5x5(uiWidth, uiHeight, tempblock, tempblockFiltered, clpRng, recPtr, recStride, iWidthExtSIMD, bfac, bifRoundAdd, bifRoundShift, false, lutRowPtr, noClip, cutBitsNum);
}
}
void BilateralFilter::clipNotBilaterallyFilteredBlocks(const ComponentID compID, const CPelUnitBuf& src, PelUnitBuf& rec, const ClpRng& clpRng, TransformUnit & currTU)
@@ -1395,8 +1401,6 @@ void BilateralFilter::bilateralFilterPicRDOperCTU( const ComponentID compID, Cod
for( auto &currCU : cs.traverseCUs( CS::getArea( cs, ctuArea, chType ), chType ) )
{
- bool isInter = ( currCU.predMode == MODE_INTER ) ? true : false;
-
bool valid = isLuma( compID ) ? true: currCU.blocks[compID].valid();
if( !valid )
{
@@ -1405,6 +1409,10 @@ void BilateralFilter::bilateralFilterPicRDOperCTU( const ComponentID compID, Cod
for (auto &currTU : CU::traverseTUs(currCU))
{
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ bool applyBIF = getApplyBIF(currTU, compID);
+#else
+ bool isInter = (currCU.predMode == MODE_INTER) ? true : false;
bool applyBIF = true;
if( isLuma( compID ) )
{
@@ -1430,6 +1438,7 @@ void BilateralFilter::bilateralFilterPicRDOperCTU( const ComponentID compID, Cod
applyBIF = ( tuCBF || isInter == false ) && ( currTU.cu->qp > 17 );
}
}
+#endif
// We should ideally also check the CTU-BIF-flag here. However, given that this function
// is only called by the encoder, and the encoder always has CTU-BIF-flag on, there is no
@@ -1579,8 +1588,28 @@ void BilateralFilter::bilateralFilterPicRDOperCTU( const ComponentID compID, Cod
#endif
#if JVET_X0071_CHROMA_BILATERAL_FILTER
-const char* BilateralFilter::getFilterLutParametersChroma( const int size, const PredMode predMode, const int32_t qp, int& bfac, int widthForStrength, int heightForStrength, bool isLumaValid)
+const char* BilateralFilter::getFilterLutParametersChroma(int16_t* block, const int stride, const int width, const int height, const PredMode predMode, const int32_t qp, int& bfac, int widthForStrength, int heightForStrength, bool isLumaValid)
{
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ int w = floorLog2(widthForStrength);
+ int h = floorLog2(heightForStrength);
+
+ int mad = m_calcMAD(block, stride, width, height, floorLog2(width) + floorLog2(height));
+
+ w = std::min(w, 7);
+ h = std::min(h, 7);
+ mad = std::min(mad >> 4, 15);
+ bfac = m_tuSizeFactorChroma[predMode == MODE_INTER][w * 8 + h];
+ if (bfac) // BIF is not applied if tuSizeFactor[(w, h)] = 0
+ {
+ bfac += m_tuMADFactorChroma[predMode == MODE_INTER][mad];
+ }
+
+ int sqp = qp - 17;
+ sqp = std::min(sqp, 25);
+ sqp = std::max(sqp, 0);
+ return m_lutChroma[sqp];
+#else
int conditionForStrength = std::min(widthForStrength, heightForStrength);
int T1 = 4;
int T2 = 16;
@@ -1632,6 +1661,39 @@ const char* BilateralFilter::getFilterLutParametersChroma( const int size, const
sqp = maxQP;
}
return m_wBIFChroma[sqp - 17];
+#endif
+}
+#endif
+
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+bool BilateralFilter::getApplyBIF(const TransformUnit& currTU, ComponentID compID)
+{
+ bool applyBIF = currTU.blocks[compID].valid() && (TU::getCbf(currTU, compID) || currTU.cu->predMode != MODE_INTER) && (currTU.cu->qp > 17);
+ if (applyBIF)
+ {
+ int w = currTU.blocks[compID].width, h = currTU.blocks[compID].height;
+ char(*factor)[64] = nullptr;
+#if JVET_V0094_BILATERAL_FILTER
+ if (isLuma(compID))
+ {
+ factor = m_tuSizeFactor;
+ }
+#endif
+#if JVET_X0071_CHROMA_BILATERAL_FILTER
+ if (!isLuma(compID))
+ {
+ if (currTU.blocks[COMPONENT_Y].valid())
+ {
+ w = currTU.blocks[COMPONENT_Y].width, h = currTU.blocks[COMPONENT_Y].height;
+ }
+ factor = m_tuSizeFactorChroma;
+ }
+#endif
+ w = floorLog2(w), h = floorLog2(h);
+ w = std::min(w, 7), h = std::min(h, 7);
+ applyBIF = factor && factor[currTU.cu->predMode == MODE_INTER][w * 8 + h];
+ }
+ return applyBIF;
}
#endif
diff --git a/source/Lib/CommonLib/BilateralFilter.h b/source/Lib/CommonLib/BilateralFilter.h
index 6fa4504564d336f61133b749211433a971dd2965..5bf80a14228949eda6bec0ed4dbb5f1699f921d0 100644
--- a/source/Lib/CommonLib/BilateralFilter.h
+++ b/source/Lib/CommonLib/BilateralFilter.h
@@ -55,27 +55,52 @@ public:
class BilateralFilter
{
private:
-#ifdef TARGET_SIMD_X86
- __m128i tempblockSIMD[2320];
- __m128i tempblockFilteredSIMD[2320];
-#else
- int64_t tempblockSIMD[2 * 2320];
- int64_t tempblockFilteredSIMD[2 * 2320];
-#endif
- Pel *tempblock = (Pel*) tempblockSIMD;
- Pel *tempblockFilteredTemp = (Pel*) (&tempblockFilteredSIMD[1]);
- // SIMD method writes to tempblockFiltered + 4 so that address
- // must be 128-aligned. Hence tempblockFilteredSIMD is a bit bigger than
- // it otherwise would need to be and we don't use the first 14 bytes.
- // We pad 8 bytes so required sizes is (128+8)*(128+8)*2+16 bytes = 37008 bytes
- // = 2313 128-bit words which has been rounded up to 2320 above.
- Pel *tempblockFiltered = &tempblockFilteredTemp[-2];
+ // 128x128 is the max TU size, 4 is the padding for the considered neighborhood, 16 is the AVX buffer size.
+ // (128 + 4 + 16) * (128 + 4) = 19536.
+ Pel tempblockSIMD[19536];
+ Pel tempblockFilteredSIMD[19536];
+
+ Pel *tempblock = (Pel*)tempblockSIMD;
+ Pel* tempblockFiltered = (Pel*)tempblockFilteredSIMD;
- 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);
- 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 );
+ 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);
+
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ int (*m_calcMAD)(int16_t* block, int stride, int width, int height, int whlog2);
+ static int calcMAD(int16_t* block, int stride, int width, int height, int whlog2);
+#endif
#if JVET_V0094_BILATERAL_FILTER
char m_wBIF[26][16] = {
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, }, /* 17 */
+ { 0, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, }, /* 18 */
+ { 0, 2, 2, 2, 2, 2, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, }, /* 19 */
+ { 0, 3, 4, 4, 4, 2, 2, 2, 1, 1, 1, 1, 0, 0, 0, -1, }, /* 20 */
+ { 0, 4, 5, 5, 5, 3, 2, 2, 1, 1, 1, 1, 0, 0, 0, -1, }, /* 21 */
+ { 0, 5, 6, 6, 6, 4, 3, 3, 1, 1, 1, 1, 0, 0, 0, -1, }, /* 22 */
+ { 0, 6, 7, 7, 7, 5, 4, 4, 2, 2, 1, 1, 1, 0, 0, -1, }, /* 23 */
+ { 0, 6, 8, 8, 8, 6, 5, 4, 3, 3, 2, 2, 1, 1, 0, -2, }, /* 24 */
+ { 0, 7, 8, 9, 9, 7, 7, 5, 4, 4, 2, 2, 2, 1, 1, -2, }, /* 25 */
+ { 0, 7, 9, 10, 10, 8, 8, 5, 5, 5, 3, 3, 2, 2, 1, -3, }, /* 26 */
+ { 0, 8, 10, 11, 11, 9, 9, 6, 6, 6, 3, 3, 3, 2, 1, -3, }, /* 27 */
+ { 0, 9, 11, 13, 13, 11, 11, 8, 8, 8, 5, 4, 4, 3, 2, -3, }, /* 28 */
+ { 0, 10, 12, 14, 14, 13, 13, 10, 9, 9, 6, 5, 5, 4, 3, -4, }, /* 29 */
+ { 0, 11, 13, 16, 16, 15, 15, 13, 11, 11, 8, 6, 6, 4, 4, -4, }, /* 30 */
+ { 0, 12, 14, 17, 17, 17, 17, 15, 12, 12, 9, 7, 7, 5, 5, -5, }, /* 31 */
+ { 0, 13, 15, 19, 19, 19, 19, 17, 14, 14, 11, 8, 8, 6, 6, -5, }, /* 32 */
+ { 0, 14, 17, 20, 21, 21, 21, 19, 17, 16, 13, 10, 10, 7, 7, -5, }, /* 33 */
+ { 0, 15, 19, 22, 23, 24, 23, 22, 20, 18, 15, 12, 11, 9, 7, -6, }, /* 34 */
+ { 0, 17, 20, 23, 26, 26, 26, 24, 23, 20, 18, 15, 13, 10, 8, -6, }, /* 35 */
+ { 0, 18, 22, 25, 28, 29, 28, 27, 26, 22, 20, 17, 14, 12, 8, -7, }, /* 36 */
+ { 0, 19, 24, 26, 30, 31, 30, 29, 29, 24, 22, 19, 16, 13, 9, -7, }, /* 37 */
+ { 0, 20, 26, 29, 32, 33, 32, 31, 31, 27, 24, 21, 17, 14, 11, -8, }, /* 38 */
+ { 0, 21, 28, 31, 34, 35, 35, 34, 34, 30, 26, 23, 19, 15, 13, -8, }, /* 39 */
+ { 0, 21, 30, 34, 36, 38, 37, 36, 36, 32, 29, 25, 20, 17, 15, -9, }, /* 40 */
+ { 0, 22, 32, 36, 38, 40, 40, 39, 39, 35, 31, 27, 22, 18, 17, -9, }, /* 41 */
+ { 0, 23, 34, 39, 40, 42, 42, 41, 41, 38, 33, 29, 23, 19, 19, -10, }, /* 42 */
+#else
{ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },
{ 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },
{ 0, 2, 2, 2, 1, 1, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, },
@@ -102,10 +127,49 @@ private:
{ 0, 15, 22, 31, 35, 39, 42, 42, 43, 41, 37, 25, 21, 17, 15, -3, },
{ 0, 16, 23, 32, 36, 40, 43, 43, 44, 42, 39, 26, 21, 17, 15, -3, },
{ 0, 17, 23, 33, 37, 41, 44, 44, 45, 44, 42, 27, 22, 17, 15, -3, },
+#endif
+ };
+ char m_lut[26][3 * 16];
+ char m_distFactor[3] = { 16, 12, 11, };
+ char m_tuSizeFactor[2][64] = {
+ { 10,10,10,9,8,6,5,4,10,10,10,9,8,6,5,4,10,10,10,9,8,6,5,4,9,9,9,7,6,5,5,4,8,8,8,6,5,5,5,4,6,6,6,5,5,5,5,4,5,5,5,5,5,5,5,4,4,4,4,4,4,4,4,4, },
+ { 10,10,10,8,6,6,4,0,10,10,10,8,6,6,4,0,10,10,10,8,6,6,4,0,8,8,8,6,5,4,4,0,6,6,6,5,4,4,4,0,6,6,6,4,4,0,0,0,4,4,4,4,4,0,0,0,0,0,0,0,0,0,0,0, },
+ };
+ char m_tuMADFactor[2][16] = {
+ { 0,0,0,1,2,3,4,5,6,6,6,6,7,7,7,7, },
+ { 0,0,0,0,0,0,1,2,2,4,4,4,4,5,5,5, },
};
#endif
#if JVET_X0071_CHROMA_BILATERAL_FILTER
char m_wBIFChroma[26][16] = {
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, }, /* 17 */
+ { 0, 1, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 0, }, /* 18 */
+ { 0, 2, 3, 3, 4, 3, 3, 3, 3, 3, 3, 2, 2, 2, 2, 0, }, /* 19 */
+ { 0, 3, 5, 5, 6, 5, 5, 5, 5, 5, 4, 2, 2, 2, 2, 1, }, /* 20 */
+ { 0, 4, 6, 6, 8, 6, 6, 6, 6, 6, 6, 3, 3, 3, 3, 1, }, /* 21 */
+ { 0, 5, 8, 8, 10, 8, 8, 8, 8, 8, 7, 4, 4, 4, 4, 1, }, /* 22 */
+ { 0, 5, 8, 8, 10, 9, 9, 9, 9, 9, 8, 5, 5, 5, 4, 1, }, /* 23 */
+ { 0, 5, 8, 8, 10, 9, 10, 10, 9, 9, 9, 6, 6, 6, 4, 2, }, /* 24 */
+ { 0, 5, 8, 8, 10, 10, 10, 10, 10, 10, 9, 8, 6, 6, 5, 2, }, /* 25 */
+ { 0, 5, 8, 8, 10, 10, 11, 11, 10, 10, 10, 9, 7, 7, 5, 3, }, /* 26 */
+ { 0, 5, 8, 8, 10, 11, 12, 12, 11, 11, 11, 10, 8, 8, 5, 3, }, /* 27 */
+ { 0, 5, 8, 9, 11, 12, 13, 13, 13, 13, 13, 13, 11, 11, 8, 5, }, /* 28 */
+ { 0, 5, 8, 9, 11, 13, 14, 15, 15, 15, 16, 15, 13, 13, 11, 6, }, /* 29 */
+ { 0, 6, 9, 10, 12, 13, 16, 16, 16, 16, 18, 18, 16, 16, 15, 8, }, /* 30 */
+ { 0, 6, 9, 10, 12, 14, 17, 18, 18, 18, 21, 20, 18, 18, 18, 9, }, /* 31 */
+ { 0, 6, 9, 11, 13, 15, 18, 19, 20, 20, 23, 23, 21, 21, 21, 11, }, /* 32 */
+ { 0, 6, 9, 12, 14, 16, 18, 20, 21, 21, 24, 24, 22, 24, 21, 13, }, /* 33 */
+ { 0, 6, 9, 13, 14, 17, 19, 21, 21, 21, 25, 25, 23, 27, 22, 14, }, /* 34 */
+ { 0, 6, 10, 13, 15, 17, 19, 21, 22, 22, 25, 25, 25, 29, 22, 16, }, /* 35 */
+ { 0, 6, 10, 14, 15, 18, 20, 22, 22, 22, 26, 26, 26, 32, 23, 17, }, /* 36 */
+ { 0, 6, 10, 15, 16, 19, 20, 23, 23, 23, 27, 27, 27, 35, 23, 19, }, /* 37 */
+ { 0, 6, 11, 15, 17, 20, 21, 24, 25, 25, 28, 29, 29, 36, 26, 21, }, /* 38 */
+ { 0, 6, 11, 15, 18, 21, 22, 25, 26, 26, 29, 30, 32, 37, 29, 24, }, /* 39 */
+ { 0, 7, 12, 16, 18, 21, 22, 27, 28, 28, 29, 32, 34, 37, 33, 26, }, /* 40 */
+ { 0, 7, 12, 16, 19, 22, 23, 28, 29, 29, 30, 33, 37, 38, 36, 29, }, /* 41 */
+ { 0, 7, 13, 16, 20, 23, 24, 29, 31, 31, 31, 35, 39, 39, 39, 31, }, /* 42 */
+#else
{ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },
{ 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },
{ 0, 2, 2, 2, 1, 1, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, },
@@ -132,6 +196,17 @@ private:
{ 0, 12, 17, 24, 27, 30, 33, 33, 34, 32, 29, 20, 16, 13, 12, -2, },
{ 0, 12, 18, 25, 28, 31, 34, 34, 34, 33, 30, 20, 16, 13, 12, -2, },
{ 0, 13, 18, 26, 29, 32, 34, 34, 35, 34, 33, 21, 17, 13, 12, -2, },
+#endif
+ };
+ char m_lutChroma[26][3 * 16];
+ char m_distFactorChroma[3] = { 16, 8, 6, };
+ char m_tuSizeFactorChroma[2][64] = {
+ { 9,9,9,9,8,8,8,0,9,9,9,9,8,8,8,0,9,9,8,8,8,8,8,0,9,9,8,8,8,8,8,0,8,8,8,8,8,8,8,0,8,8,8,8,8,8,8,0,8,8,8,8,8,8,8,0,0,0,0,0,0,0,0,0, },
+ { 13,13,13,13,9,9,9,0,13,13,13,13,9,9,9,0,13,13,13,13,9,9,9,0,13,13,13,9,9,9,9,0,9,9,9,9,9,9,9,0,9,9,9,9,9,0,0,0,9,9,9,9,9,0,0,0,0,0,0,0,0,0,0,0, },
+ };
+ char m_tuMADFactorChroma[2][16] = {
+ { 0,1,3,4,5,7,9,11,11,11,11,11,11,11,11,11, },
+ { 0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1, },
};
#endif
public:
@@ -142,26 +217,34 @@ public:
void destroy();
#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);
- void bilateralFilterDiamond5x5( const ComponentID compID, const CPelUnitBuf& src, PelUnitBuf& rec, int32_t qp, const ClpRng& clpRng, TransformUnit & currTU, bool noClip
+ void bilateralFilterPicRDOperCTU(const ComponentID compID, CodingStructure& cs, PelUnitBuf& src,BIFCabacEst* bifCABACEstimator);
+ void bilateralFilterDiamond5x5(const ComponentID compID, const CPelUnitBuf& src, PelUnitBuf& rec, int32_t qp, const ClpRng& clpRng, TransformUnit & currTU, bool noClip
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
, bool isCtuCrossedByVirtualBoundaries, int horVirBndryPos[], int verVirBndryPos[], int numHorVirBndry, int numVerVirBndry
, bool clipTop, bool clipBottom, bool clipLeft, bool clipRight
#endif
);
- const char* getFilterLutParameters( const int size, const PredMode predMode, const int qp, int& bfac );
- void clipNotBilaterallyFilteredBlocks( const ComponentID compID, const CPelUnitBuf& src, PelUnitBuf& rec, const ClpRng& clpRng, TransformUnit & currTU);
-#if JVET_X0071_CHROMA_BILATERAL_FILTER
- const char* getFilterLutParametersChroma( const int size, const PredMode predMode, const int qp, int& bfac, int widthForStrength, int heightForStrength, bool isLumaValid);
+ const char* getFilterLutParameters(int16_t* block, const int stride, const int width, const int height, const PredMode predMode, const int qp, int& bfac);
+ void clipNotBilaterallyFilteredBlocks(const ComponentID compID, const CPelUnitBuf& src, PelUnitBuf& rec, const ClpRng& clpRng, TransformUnit & currTU);
#endif
+
+#if JVET_X0071_CHROMA_BILATERAL_FILTER
+ const char* getFilterLutParametersChroma(int16_t* block, const int stride, const int width, const int height, const PredMode predMode, const int qp, int& bfac, int widthForStrength, int heightForStrength, bool isLumaValid);
#endif
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ bool getApplyBIF(const TransformUnit& currTU, ComponentID compID);
+#endif
#if ENABLE_SIMD_BILATERAL_FILTER || JVET_X0071_CHROMA_BILATERAL_FILTER_ENABLE_SIMD
#ifdef TARGET_SIMD_X86
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 );
+ 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);
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ template<X86_VEXT vext>
+ static int simdCalcMAD(int16_t* block, int stride, int width, int height, int whlog2);
+#endif
void initBilateralFilterX86();
template <X86_VEXT vext>
diff --git a/source/Lib/CommonLib/Picture.h b/source/Lib/CommonLib/Picture.h
index 4f0803624f908b98cd8cb395b16350385191f923..86c8084b573e09182d4b6a89e43f00e27c3ae2fa 100644
--- a/source/Lib/CommonLib/Picture.h
+++ b/source/Lib/CommonLib/Picture.h
@@ -364,7 +364,7 @@ public:
#if JVET_X0071_CHROMA_BILATERAL_FILTER
BifParams m_bifParams[MAX_NUM_COMPONENT];
#else
- BifParams m_bifParams;
+ BifParams m_bifParams[1];
#endif
#endif
std::vector<uint8_t> m_alfCtuEnableFlag[MAX_NUM_COMPONENT];
diff --git a/source/Lib/CommonLib/SampleAdaptiveOffset.cpp b/source/Lib/CommonLib/SampleAdaptiveOffset.cpp
index 97d8143d91f8034d7a21d7b1454d049118db233f..04f37c0117543433c742b93226b96c869510a502 100644
--- a/source/Lib/CommonLib/SampleAdaptiveOffset.cpp
+++ b/source/Lib/CommonLib/SampleAdaptiveOffset.cpp
@@ -174,6 +174,9 @@ void SampleAdaptiveOffset::create( int picWidth, int picHeight, ChromaFormat for
::memset(m_ccSaoControl[compIdx], 0, sizeof(uint8_t) * m_numCTUsInPic);
}
#endif
+#if JVET_V0094_BILATERAL_FILTER || JVET_X0071_CHROMA_BILATERAL_FILTER
+ m_bilateralFilter.create();
+#endif
}
void SampleAdaptiveOffset::destroy()
@@ -1156,8 +1159,17 @@ void SampleAdaptiveOffset::SAOProcess( CodingStructure& cs, SAOBlkParam* saoBlkP
{
for (auto& currTU : CU::traverseTUs(currCU))
{
- bool isInter = (currCU.predMode == MODE_INTER) ? true : false;
- if (bifParams.ctuOn[ctuRsAddr] && ((TU::getCbf(currTU, COMPONENT_Y) || isInter == false) && (currTU.cu->qp > 17)) && (128 > std::max(currTU.lumaSize().width, currTU.lumaSize().height)) && ((isInter == false) || (32 > std::min(currTU.lumaSize().width, currTU.lumaSize().height))))
+ bool applyBIF = bifParams.ctuOn[ctuRsAddr];
+ if (applyBIF)
+ {
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ applyBIF = m_bilateralFilter.getApplyBIF(currTU, COMPONENT_Y);
+#else
+ bool isInter = (currCU.predMode == MODE_INTER) ? true : false;
+ applyBIF = ((TU::getCbf(currTU, COMPONENT_Y) || isInter == false) && (currTU.cu->qp > 17)) && (128 > std::max(currTU.lumaSize().width, currTU.lumaSize().height)) && ((isInter == false) || (32 > std::min(currTU.lumaSize().width, currTU.lumaSize().height)));
+#endif
+ }
+ if (applyBIF)
{
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
bool clipTop = false, clipBottom = false, clipLeft = false, clipRight = false;
@@ -1181,8 +1193,6 @@ void SampleAdaptiveOffset::SAOProcess( CodingStructure& cs, SAOBlkParam* saoBlkP
#if JVET_X0071_CHROMA_BILATERAL_FILTER
if(cs.pps->getUseChromaBIF())
{
- bool tuValid = false;
- bool tuCBF = false;
bool isDualTree = CS::isDualITree(cs);
ChannelType chType = isDualTree ? CH_C : CH_L;
bool applyChromaBIF = false;
@@ -1202,7 +1212,6 @@ void SampleAdaptiveOffset::SAOProcess( CodingStructure& cs, SAOBlkParam* saoBlkP
}
for (auto &currTU : CU::traverseTUs(currCU))
{
- bool isInter = (currCU.predMode == MODE_INTER) ? true : false;
for(int compIdx = COMPONENT_Cb; compIdx < MAX_NUM_COMPONENT; compIdx++)
{
applyChromaBIF = false;
@@ -1210,6 +1219,12 @@ void SampleAdaptiveOffset::SAOProcess( CodingStructure& cs, SAOBlkParam* saoBlkP
BifParams& chromaBifParams = cs.picture->getBifParam( compID );
bool ctuEnableChromaBIF = chromaBifParams.ctuOn[ctuRsAddr];
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ applyChromaBIF = ctuEnableChromaBIF && m_bilateralFilter.getApplyBIF(currTU, compID);
+#else
+ bool tuValid = false;
+ bool tuCBF = false;
+ bool isInter = (currCU.predMode == MODE_INTER) ? true : false;
if(!isDualTree)
{
tuValid = currTU.blocks[compIdx].valid();
@@ -1225,6 +1240,7 @@ void SampleAdaptiveOffset::SAOProcess( CodingStructure& cs, SAOBlkParam* saoBlkP
tuCBF = TU::getCbf(currTU, compID);
applyChromaBIF = (ctuEnableChromaBIF && ((tuCBF || isInter == false) && (currTU.cu->qp > 17)));
}
+#endif
if(applyChromaBIF)
{
@@ -1301,20 +1317,29 @@ void SampleAdaptiveOffset::SAOProcess( CodingStructure& cs, SAOBlkParam* saoBlkP
for (auto &currTU : CU::traverseTUs(currCU))
{
- bool isInter = (currCU.predMode == MODE_INTER) ? true : false;
- if ( bifParams.ctuOn[ctuRsAddr] && ((TU::getCbf(currTU, COMPONENT_Y) || isInter == false) && (currTU.cu->qp > 17)) && (128 > std::max(currTU.lumaSize().width, currTU.lumaSize().height)) && ((isInter == false) || (32 > std::min(currTU.lumaSize().width, currTU.lumaSize().height))))
+ bool applyBIF = bifParams.ctuOn[ctuRsAddr];
+ if (applyBIF)
+ {
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ applyBIF = m_bilateralFilter.getApplyBIF(currTU, COMPONENT_Y);
+#else
+ bool isInter = (currCU.predMode == MODE_INTER) ? true : false;
+ applyBIF = ((TU::getCbf(currTU, COMPONENT_Y) || isInter == false) && (currTU.cu->qp > 17)) && (128 > std::max(currTU.lumaSize().width, currTU.lumaSize().height)) && ((isInter == false) || (32 > std::min(currTU.lumaSize().width, currTU.lumaSize().height)));
+#endif
+ }
+ if (applyBIF)
{
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
bool clipTop = false, clipBottom = false, clipLeft = false, clipRight = false;
int numHorVirBndry = 0, numVerVirBndry = 0;
int horVirBndryPos[] = { 0, 0, 0 };
int verVirBndryPos[] = { 0, 0, 0 };
- bool isTUCrossedByVirtualBoundaries = bilateralFilter.isCrossedByVirtualBoundaries(
+ bool isTUCrossedByVirtualBoundaries = m_bilateralFilter.isCrossedByVirtualBoundaries(
cs, currTU.Y().x, currTU.Y().y, currTU.lumaSize().width, currTU.lumaSize().height, clipTop, clipBottom,
clipLeft, clipRight, numHorVirBndry, numVerVirBndry, horVirBndryPos, verVirBndryPos);
#endif
- m_bilateralFilter.bilateralFilterDiamond5x5( COMPONENT_Y, m_tempBuf, rec, currTU.cu->qp, cs.slice->clpRng(COMPONENT_Y), currTU
+ m_bilateralFilter.bilateralFilterDiamond5x5( COMPONENT_Y, m_tempBuf, rec, currTU.cu->qp, cs.slice->clpRng(COMPONENT_Y), currTU, false
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
, isTUCrossedByVirtualBoundaries, horVirBndryPos, verVirBndryPos, numHorVirBndry, numVerVirBndry
, clipTop, clipBottom, clipLeft, clipRight
@@ -1347,9 +1372,7 @@ void SampleAdaptiveOffset::SAOProcess( CodingStructure& cs, SAOBlkParam* saoBlkP
}
}
if(cs.pps->getUseChromaBIF())
- {
- bool tuValid = false;
- bool tuCBF = false;
+ {
bool isDualTree = CS::isDualITree(cs);
ChannelType chType = isDualTree ? CH_C : CH_L;
bool applyChromaBIF = false;
@@ -1363,12 +1386,17 @@ void SampleAdaptiveOffset::SAOProcess( CodingStructure& cs, SAOBlkParam* saoBlkP
}
for (auto &currTU : CU::traverseTUs(currCU))
{
- bool isInter = (currCU.predMode == MODE_INTER) ? true : false;
for(int compIdx = COMPONENT_Cb; compIdx < MAX_NUM_COMPONENT; compIdx++)
{
ComponentID compID = ComponentID( compIdx );
BifParams& chromaBifParams = cs.picture->getBifParam( compID );
bool ctuEnableChromaBIF = chromaBifParams.ctuOn[ctuRsAddr];
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ applyChromaBIF = ctuEnableChromaBIF && m_bilateralFilter.getApplyBIF(currTU, compID);
+#else
+ bool tuValid = false;
+ bool tuCBF = false;
+ bool isInter = (currCU.predMode == MODE_INTER) ? true : false;
applyChromaBIF = false;
if(!isDualTree)
{
@@ -1385,6 +1413,7 @@ void SampleAdaptiveOffset::SAOProcess( CodingStructure& cs, SAOBlkParam* saoBlkP
tuCBF = TU::getCbf(currTU, compID);
applyChromaBIF = (ctuEnableChromaBIF && ((tuCBF || isInter == false) && (currTU.cu->qp > 17)));
}
+#endif
if(applyChromaBIF)
{
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
@@ -1402,7 +1431,7 @@ void SampleAdaptiveOffset::SAOProcess( CodingStructure& cs, SAOBlkParam* saoBlkP
clipLeft, clipRight, numHorVirBndry, numVerVirBndry, horVirBndryPos, verVirBndryPos);
#endif
- m_bilateralFilter.bilateralFilterDiamond5x5( compID, m_tempBuf, rec, currTU.cu->qp, cs.slice->clpRng(compID), currTU, false,
+ m_bilateralFilter.bilateralFilterDiamond5x5( compID, m_tempBuf, rec, currTU.cu->qp, cs.slice->clpRng(compID), currTU, false
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
, isTUCrossedByVirtualBoundaries, horVirBndryPos, verVirBndryPos, numHorVirBndry, numVerVirBndry, clipTop, clipBottom, clipLeft, clipRight
#endif
@@ -1517,6 +1546,9 @@ void SampleAdaptiveOffset::SAOProcess( CodingStructure& cs, SAOBlkParam* saoBlkP
BifParams& chromaBifParams = cs.picture->getBifParam( compID );
bool ctuEnableChromaBIF = chromaBifParams.ctuOn[ctuRsAddr];
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ applyChromaBIF = ctuEnableChromaBIF && m_bilateralFilter.getApplyBIF(currTU, compID);
+#else
applyChromaBIF = false;
if(!isDualTree)
{
@@ -1533,6 +1565,7 @@ void SampleAdaptiveOffset::SAOProcess( CodingStructure& cs, SAOBlkParam* saoBlkP
tuCBF = TU::getCbf(currTU, compID);
applyChromaBIF = (ctuEnableChromaBIF && ((tuCBF || isInter == false) && (currTU.cu->qp > 17)));
}
+#endif
if(applyChromaBIF)
{
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
@@ -1723,8 +1756,13 @@ void SampleAdaptiveOffset::jointClipSaoBifCcSao(CodingStructure& cs)
{
for( auto& currTU : CU::traverseTUs( currCU ) )
{
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ bool applyBIF = bifParams.ctuOn[ctuRsAddr] && m_bilateralFilter.getApplyBIF(currTU, compID);
+#else
bool isInter = ( currCU.predMode == MODE_INTER ) ? true : false;
- if( bifParams.ctuOn[ctuRsAddr] && ( ( TU::getCbf( currTU, compID ) || isInter == false ) && ( currTU.cu->qp > 17 ) ) && ( 128 > std::max( currTU.lumaSize().width, currTU.lumaSize().height ) ) && ( ( isInter == false ) || ( 32 > std::min( currTU.lumaSize().width, currTU.lumaSize().height ) ) ) )
+ bool applyBIF = bifParams.ctuOn[ctuRsAddr] && ((TU::getCbf(currTU, compID) || isInter == false) && (currTU.cu->qp > 17)) && (128 > std::max(currTU.lumaSize().width, currTU.lumaSize().height)) && ((isInter == false) || (32 > std::min(currTU.lumaSize().width, currTU.lumaSize().height)));
+#endif
+ if (applyBIF)
{
m_bilateralFilter.clipNotBilaterallyFilteredBlocks( compID, m_tempBuf, dstYuv, cs.slice->clpRng( compID ), currTU );
}
@@ -1734,8 +1772,6 @@ void SampleAdaptiveOffset::jointClipSaoBifCcSao(CodingStructure& cs)
#if JVET_X0071_CHROMA_BILATERAL_FILTER
if( cs.pps->getUseChromaBIF() && isChroma( compID ) )
{
- bool tuValid = false;
- bool tuCBF = false;
bool ctuEnableChromaBIF = false;
bool isDualTree = CS::isDualITree( cs );
ChannelType chType = isDualTree ? CH_C : CH_L;
@@ -1749,12 +1785,17 @@ void SampleAdaptiveOffset::jointClipSaoBifCcSao(CodingStructure& cs)
}
for( auto &currTU : CU::traverseTUs( currCU ) )
{
- bool isInter = ( currCU.predMode == MODE_INTER ) ? true : false;
//Cb or Cr
applyChromaBIF = false;
BifParams& chromaBifParams = cs.picture->getBifParam( compID );
ctuEnableChromaBIF = chromaBifParams.ctuOn[ctuRsAddr];
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ applyChromaBIF = ctuEnableChromaBIF && m_bilateralFilter.getApplyBIF(currTU, compID);
+#else
+ bool tuValid = false;
+ bool tuCBF = false;
+ bool isInter = (currCU.predMode == MODE_INTER) ? true : false;
if( !isDualTree )
{
tuValid = currTU.blocks[compIdx].valid();
@@ -1770,6 +1811,7 @@ void SampleAdaptiveOffset::jointClipSaoBifCcSao(CodingStructure& cs)
tuCBF = TU::getCbf( currTU, compID );
applyChromaBIF = ( ctuEnableChromaBIF && ( ( tuCBF || isInter == false ) && ( currTU.cu->qp > 17 ) ) );
}
+#endif
if( applyChromaBIF )
{
m_bilateralFilter.clipNotBilaterallyFilteredBlocks( compID, m_tempBuf, dstYuv, cs.slice->clpRng( compID ), currTU );
diff --git a/source/Lib/CommonLib/TypeDef.h b/source/Lib/CommonLib/TypeDef.h
index a03ea17bb02bb233c5b455567b50ac847c0666f8..b976de3bac19fc37ce44269b7a2093bce917d260 100644
--- a/source/Lib/CommonLib/TypeDef.h
+++ b/source/Lib/CommonLib/TypeDef.h
@@ -397,6 +397,7 @@
#define JVET_AE0139_ALF_IMPROVED_FIXFILTER 1 // JVET-AE0139: Improved ALF fixed filter
#define JVET_AE0151_CCSAO_HISTORY_OFFSETS_AND_EXT_EO 1 // JVET-AE0151: CCSAO with history offsets and extended edge classifiers
#define JVET_AF0197_LUMA_RESIDUAL_TAP_IN_CCALF 1 // JVET-AF0197: Luma Residual Tap in CCALF
+#define JVET_AF0112_BIF_DYNAMIC_SCALING 1 // JVET-AF0112: Dynamic TU scale factor for BIF with LUTs interpolation
// SIMD optimizations
#if IF_12TAP
diff --git a/source/Lib/CommonLib/x86/BilateralFilterX86.h b/source/Lib/CommonLib/x86/BilateralFilterX86.h
index 7f39a283fa644e9880eff236cd11733f4fc4e4c3..9e3262765f16f6344f0319d555a55582150dfcff 100644
--- a/source/Lib/CommonLib/x86/BilateralFilterX86.h
+++ b/source/Lib/CommonLib/x86/BilateralFilterX86.h
@@ -44,33 +44,317 @@
#endif
#if ENABLE_SIMD_BILATERAL_FILTER || JVET_X0071_CHROMA_BILATERAL_FILTER_ENABLE_SIMD
+
+#if USE_AVX2
+
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+inline void simdBifApplyLut(__m256i& val, __m256i& acc, int cutBitsNum, __m256i& bitsRound, __m256i& bitsRound2, __m256i& lut)
+#else
+inline void simdBifApplyLut(__m256i& val, __m256i& acc, __m256i& lut, int lutShift)
+#endif
+{
+ __m256i diffabs = _mm256_abs_epi16(val); /* absolute value */
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ __m256i diffabs2 = _mm256_add_epi16(diffabs, bitsRound2); /* + bitsRound2 */
+ diffabs2 = _mm256_srli_epi16(diffabs2, cutBitsNum); /* >> cutBitsNum */
+ diffabs = _mm256_add_epi16(diffabs, bitsRound); /* + bitsRound2 */
+ diffabs = _mm256_srli_epi16(diffabs, cutBitsNum); /* >> cutBitsNum */
+ diffabs = _mm256_packus_epi16(diffabs2, diffabs); /* convert to 8 bits */
+ diffabs = _mm256_permute4x64_epi64(diffabs, 0xD8); /* permute bytes */
+ diffabs = _mm256_min_epu8(diffabs, _mm256_set1_epi8(15)); /* min(x,15) */
+ diffabs = _mm256_shuffle_epi8(lut, diffabs); /* lut */
+ diffabs2 = _mm256_cvtepi8_epi16(*((__m128i*) & diffabs)); /* back to 16-bit */
+ diffabs = _mm256_cvtepi8_epi16(*((__m128i*) & diffabs + 1)); /* back to 16-bit */
+ diffabs = _mm256_srai_epi16(_mm256_add_epi16(_mm256_add_epi16(diffabs, diffabs2), _mm256_set1_epi16(1)), 1); /* (v1 + v2 + 1) >> 1 */
+#else
+ diffabs = _mm256_add_epi16(diffabs, _mm256_set1_epi16(4)); /* +4 */
+ diffabs = _mm256_srai_epi16(diffabs, 3); /* >> 3 */
+ diffabs = _mm256_min_epi16(diffabs, _mm256_set1_epi16(15)); /* min(x,15) */
+ diffabs = _mm256_packus_epi16(diffabs, _mm256_permute2f128_si256(diffabs, diffabs, 0x01)); /* convert to 8 */
+ 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
+ diffabs = _mm256_sign_epi16(diffabs, val); /* add original sign */
+ acc = _mm256_add_epi16(diffabs, acc); /* add to acc */
+}
+
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 )
+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)
+{
+ //if( uiWidth < 4 || ( uiWidth < 8 && isRDO ) )
+ if (uiWidth < 4)
+ {
+ return blockBilateralFilterDiamond5x5(uiWidth, uiHeight, block, blkFilt, clpRng, recPtr, recStride, iWidthExtSIMD, bfac, bifRoundAdd, bifRoundShift, isRDO, lutRowPtr, noClip, cutBitsNum);
+ }
+
+ int pad = 2;
+ int padwidth = iWidthExtSIMD;
+
+ __m256i center, left, right, up, down, lu, ld, ru, rd, acc, roundAdd, clipmin, clipmax, inputVals;
+ __m256i ll, rr, uu, dd;
+ __m128i lutTmp;
+
+ clipmin = _mm256_set1_epi16(clpRng.min);
+ clipmax = _mm256_set1_epi16(clpRng.max);
+
+ acc = _mm256_set1_epi32(0);
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ lutTmp = _mm_loadu_si128((__m128i*)(lutRowPtr));
+ __m256i lut1 = _mm256_set_m128i(lutTmp, lutTmp);
+ lutTmp = _mm_loadu_si128((__m128i*)(lutRowPtr + 16));
+ __m256i lut2 = _mm256_set_m128i(lutTmp, lutTmp);
+ lutTmp = _mm_loadu_si128((__m128i*)(lutRowPtr + 32));
+ __m256i lut3 = _mm256_set_m128i(lutTmp, lutTmp);
+ __m256i mmBfac = _mm256_set1_epi16(bfac);
+ 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)));
+#else
+ lutTmp = _mm_loadu_si128((__m128i*)(lutRowPtr));
+ __m256i lut = _mm256_set_m128i(lutTmp, lutTmp);
+ int lutShift1 = 0, lutShift2 = 1, lutShift3 = 1;
+ roundAdd = _mm256_set1_epi16(bifRoundAdd);
+#endif
+
+ for (int row = 0; row < uiHeight; row++)
+ {
+ for (int col = 0; col < uiWidth; col += 16)
+ {
+ acc = _mm256_set1_epi32(0);
+ int16_t* point = &block[(row + pad) * padwidth + pad + col];
+
+ center = _mm256_loadu_si256((__m256i*)(point));
+
+ //load neighbours
+ left = _mm256_loadu_si256((__m256i*)(point - 1));
+ right = _mm256_loadu_si256((__m256i*)(point + 1));
+ up = _mm256_loadu_si256((__m256i*)(point - padwidth));
+ down = _mm256_loadu_si256((__m256i*)(point + padwidth));
+
+ lu = _mm256_loadu_si256((__m256i*)(point - 1 - padwidth));
+ ld = _mm256_loadu_si256((__m256i*)(point - 1 + padwidth));
+ ru = _mm256_loadu_si256((__m256i*)(point + 1 - padwidth));
+ rd = _mm256_loadu_si256((__m256i*)(point + 1 + padwidth));
+
+ ll = _mm256_loadu_si256((__m256i*)(point - 2));
+ rr = _mm256_loadu_si256((__m256i*)(point + 2));
+ uu = _mm256_loadu_si256((__m256i*)(point - 2 * padwidth));
+ dd = _mm256_loadu_si256((__m256i*)(point + 2 * padwidth));
+
+ //calculate diffs
+ left = _mm256_sub_epi16(left, center);
+ right = _mm256_sub_epi16(right, center);
+ up = _mm256_sub_epi16(up, center);
+ down = _mm256_sub_epi16(down, center);
+
+ lu = _mm256_sub_epi16(lu, center);
+ ld = _mm256_sub_epi16(ld, center);
+ ru = _mm256_sub_epi16(ru, center);
+ rd = _mm256_sub_epi16(rd, center);
+
+ ll = _mm256_sub_epi16(ll, center);
+ rr = _mm256_sub_epi16(rr, center);
+ uu = _mm256_sub_epi16(uu, center);
+ dd = _mm256_sub_epi16(dd, center);
+
+ // apply LUT
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ simdBifApplyLut(left, acc, cutBitsNum, bitsRound, bitsRound2, lut1);
+ simdBifApplyLut(right, acc, cutBitsNum, bitsRound, bitsRound2, lut1);
+ simdBifApplyLut(up, acc, cutBitsNum, bitsRound, bitsRound2, lut1);
+ simdBifApplyLut(down, acc, cutBitsNum, bitsRound, bitsRound2, lut1);
+
+ simdBifApplyLut(lu, acc, cutBitsNum, bitsRound, bitsRound2, lut2);
+ simdBifApplyLut(ld, acc, cutBitsNum, bitsRound, bitsRound2, lut2);
+ simdBifApplyLut(ru, acc, cutBitsNum, bitsRound, bitsRound2, lut2);
+ simdBifApplyLut(rd, acc, cutBitsNum, bitsRound, bitsRound2, lut2);
+
+ simdBifApplyLut(ll, acc, cutBitsNum, bitsRound, bitsRound2, lut3);
+ simdBifApplyLut(rr, acc, cutBitsNum, bitsRound, bitsRound2, lut3);
+ simdBifApplyLut(uu, acc, cutBitsNum, bitsRound, bitsRound2, lut3);
+ simdBifApplyLut(dd, acc, cutBitsNum, bitsRound, bitsRound2, lut3);
+#else
+ simdBifApplyLut(left, acc, lut, lutShift1);
+ simdBifApplyLut(right, acc, lut, lutShift1);
+ simdBifApplyLut(up, acc, lut, lutShift1);
+ simdBifApplyLut(down, acc, lut, lutShift1);
+
+ simdBifApplyLut(lu, acc, lut, lutShift2);
+ simdBifApplyLut(ld, acc, lut, lutShift2);
+ simdBifApplyLut(ru, acc, lut, lutShift2);
+ simdBifApplyLut(rd, acc, lut, lutShift2);
+
+ simdBifApplyLut(ll, acc, lut, lutShift3);
+ simdBifApplyLut(rr, acc, lut, lutShift3);
+ simdBifApplyLut(uu, acc, lut, lutShift3);
+ simdBifApplyLut(dd, acc, lut, lutShift3);
+#endif
+
+ // TU scaling
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ acc = _mm256_mullo_epi16(acc, mmBfac);
+ acc = _mm256_adds_epi16(acc, roundAdd);
+ acc = _mm256_srai_epi16(acc, bifRoundShift + 3);
+#else
+ if (bfac == 2)
+ {
+ acc = _mm256_slli_epi16(acc, 1); // Shift left to get 2*
+ }
+ else if (bfac == 3)
+ {
+ acc = _mm256_add_epi16(acc, _mm256_slli_epi16(acc, 1)); // Multiply by two by shifting left and add original value to get 3*
+ }
+
+ // Add 16 and shift 5
+ acc = _mm256_add_epi16(acc, roundAdd);
+ acc = _mm256_srai_epi16(acc, bifRoundShift);
+#endif
+
+ // Instead we add our input values to the delta
+ if (isRDO)
+ {
+ acc = _mm256_add_epi16(acc, center);
+ }
+ else
+ {
+ int16_t* recpoint = &recPtr[row * recStride + col];
+ inputVals = _mm256_loadu_si256((__m256i*)(recpoint));
+ acc = _mm256_add_epi16(acc, inputVals);
+ }
+
+ // Clip
+#if JVET_W0066_CCSAO
+ if (isRDO || !noClip)
+#endif
+ {
+ acc = _mm256_max_epi16(acc, clipmin);
+ acc = _mm256_min_epi16(acc, clipmax);
+ }
+
+ _mm256_storeu_si256((__m256i*)(blkFilt + (row + pad) * padwidth + col + pad), acc);
+ }
+ }
+
+ // Copy back from tempbufFilter to recBuf
+ int onerow = uiWidth * sizeof(Pel);
+ // Copy back parameters
+ Pel* tempBlockPtr = blkFilt + pad * padwidth + pad;
+ for (uint32_t yy = 0; yy < uiHeight; yy++)
+ {
+ std::memcpy(recPtr, tempBlockPtr, onerow);
+ recPtr += recStride;
+ tempBlockPtr += padwidth;
+ }
+}
+
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+template<X86_VEXT vext>
+int BilateralFilter::simdCalcMAD(int16_t* block, int stride, int width, int height, int whlog2)
+{
+ if (width < 8)
+ {
+ return calcMAD(block, stride, width, height, whlog2);
+ }
+ int sum32[8];
+ __m256i acc32, val, average;
+ acc32 = _mm256_setzero_si256();
+ for (int i = 0; i < height; i++)
+ {
+ for (int j = 0; j < width; j += 8)
+ {
+ val = _mm256_cvtepi16_epi32(_mm_loadu_si128((__m128i*)(block + j)));
+ acc32 = _mm256_add_epi32(acc32, val);
+ }
+ block += stride;
+ }
+ block -= stride * height;
+ acc32 = _mm256_hadd_epi32(acc32, acc32);
+ acc32 = _mm256_hadd_epi32(acc32, acc32);
+ _mm256_storeu_si256((__m256i*)sum32, acc32);
+ average = _mm256_set1_epi32((sum32[0] + sum32[4] + (1 << (whlog2 - 1))) >> whlog2);
+ acc32 = _mm256_setzero_si256();
+ for (int i = 0; i < height; i++)
+ {
+ for (int j = 0; j < width; j += 8)
+ {
+ val = _mm256_cvtepi16_epi32(_mm_loadu_si128((__m128i*)(block + j)));
+ acc32 = _mm256_add_epi32(acc32, _mm256_abs_epi32(_mm256_sub_epi32(val, average)));
+ }
+ block += stride;
+ }
+ acc32 = _mm256_hadd_epi32(acc32, acc32);
+ acc32 = _mm256_hadd_epi32(acc32, acc32);
+ _mm256_storeu_si256((__m256i*)sum32, acc32);
+ return (sum32[0] + sum32[4] + (1 << (whlog2 - 1))) >> whlog2;
+}
+#endif
+
+#else // USE_AVX2
+
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+inline void simdBifApplyLut(__m128i& val, __m128i& acc, int cutBitsNum, __m128i& bitsRound, __m128i& bitsRound2, __m128i& lut)
+#else
+inline void simdBifApplyLut(__m128i& val, __m128i& acc, __m128i& lut, int lutShift)
+#endif
+{
+ __m128i diffabs = _mm_abs_epi16(val); /* absolute value */
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ __m128i diffabs2 = _mm_add_epi16(diffabs, bitsRound2); /* + bitsRound2 */
+ diffabs2 = _mm_srli_epi16(diffabs2, cutBitsNum); /* >> cutBitsNum */
+ diffabs = _mm_add_epi16(diffabs, bitsRound); /* + bitsRound2 */
+ diffabs = _mm_srli_epi16(diffabs, cutBitsNum); /* >> cutBitsNum */
+ diffabs = _mm_packus_epi16(diffabs2, diffabs); /* convert to 8 bits */
+ diffabs = _mm_min_epu8(diffabs, _mm_set1_epi8(15)); /* min(x,15) */
+ diffabs = _mm_shuffle_epi8(lut, diffabs); /* lut */
+ diffabs2 = _mm_cvtepi8_epi16(diffabs); /* back to 16-bit */
+ diffabs = _mm_cvtepi8_epi16(_mm_shuffle_epi32(diffabs, 0x4e)); /* back to 16-bit */
+ diffabs = _mm_srai_epi16(_mm_add_epi16(_mm_add_epi16(diffabs, diffabs2), _mm_set1_epi16(1)), 1); /* (v1 + v2 + 1) >> 1 */
+#else
+ diffabs = _mm_add_epi16(diffabs, _mm_set1_epi16(4)); /* +4 */
+ diffabs = _mm_srai_epi16(diffabs, 3); /* >> 3 */
+ diffabs = _mm_min_epi16(diffabs, _mm_set1_epi16(15)); /* min(x,15) */
+ diffabs = _mm_packus_epi16(diffabs, diffabs); /* convert to 8 */
+ 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
+ diffabs = _mm_sign_epi16(diffabs, val); /* add original sign */
+ acc = _mm_add_epi16(diffabs, acc); /* add to acc */
+}
+
+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)
{
//if( uiWidth < 4 || ( uiWidth < 8 && isRDO ) )
if( uiWidth < 4 )
{
- return blockBilateralFilterDiamond5x5( uiWidth, uiHeight, block, blkFilt, clpRng, recPtr, recStride, iWidthExtSIMD, bfac, bifRoundAdd, bifRoundShift, isRDO, lutRowPtr, noClip );
+ return blockBilateralFilterDiamond5x5(uiWidth, uiHeight, block, blkFilt, clpRng, recPtr, recStride, iWidthExtSIMD, bfac, bifRoundAdd, bifRoundShift, isRDO, lutRowPtr, noClip, cutBitsNum);
}
int pad = 2;
int padwidth = iWidthExtSIMD;
- __m128i center, left, right, up, down, lu, ld, ru, rd, diffabs, four, fifteen, lut, acc, temp, roundAdd, clipmin, clipmax, inputVals;
+ __m128i center, left, right, up, down, lu, ld, ru, rd, acc, roundAdd, clipmin, clipmax, inputVals;
__m128i ll, rr, uu, dd;
- four = _mm_set1_epi16(4);
- fifteen = _mm_set1_epi16(15);
- roundAdd = _mm_set1_epi16(bifRoundAdd);
clipmin = _mm_set1_epi16(clpRng.min);
clipmax = _mm_set1_epi16(clpRng.max);
- lut = _mm_loadu_si128((__m128i*)(lutRowPtr));
acc = _mm_set1_epi32(0);
-
- // Copy back parameters
- Pel *tempBlockPtr = (short*)blkFilt + (((padwidth+4) << 1) + 2);
- int tempBlockStride = padwidth+4;
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ __m128i lut1 = _mm_loadu_si128((__m128i*)(lutRowPtr));
+ __m128i lut2 = _mm_loadu_si128((__m128i*)(lutRowPtr + 16));
+ __m128i lut3 = _mm_loadu_si128((__m128i*)(lutRowPtr + 32));
+ __m128i mmBfac = _mm_set1_epi16(bfac);
+ 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)));
+#else
+ __m128i lut = _mm_loadu_si128((__m128i*)(lutRowPtr));
+ int lutShift1 = 0, lutShift2 = 1, lutShift3 = 1;
+ roundAdd = _mm_set1_epi16(bifRoundAdd);
+#endif
for (int col = 0; col < uiWidth; col += 8)
{
@@ -113,164 +397,59 @@ void BilateralFilter::simdFilterDiamond5x5( uint32_t uiWidth, uint32_t uiHeight,
uu = _mm_sub_epi16(uu, center);
dd = _mm_sub_epi16(dd, center);
- //LEFT!
- //calculate abs
- diffabs = _mm_abs_epi16(left); //abs
- diffabs = _mm_add_epi16(diffabs, four); //+4
- diffabs = _mm_srai_epi16(diffabs, 3); //>>3
- diffabs = _mm_min_epi16(diffabs, fifteen); //min(x,15)
- diffabs = _mm_packus_epi16(diffabs, diffabs); //convert to 8
- diffabs = _mm_shuffle_epi8(lut, diffabs);//lut
- diffabs = _mm_cvtepi8_epi16(diffabs);//back to 16-bit
- diffabs = _mm_sign_epi16(diffabs, left);//fix sign!
- acc = _mm_add_epi16(diffabs, acc); //add to acc
- //RIGHT!
- //calculate abs
- diffabs = _mm_abs_epi16(right); //abs
- diffabs = _mm_add_epi16(diffabs, four); //+4
- diffabs = _mm_srai_epi16(diffabs, 3); //>>3
- diffabs = _mm_min_epi16(diffabs, fifteen); //min(x,15)
- diffabs = _mm_packus_epi16(diffabs, diffabs); //convert to 8
- diffabs = _mm_shuffle_epi8(lut, diffabs);//lut
- diffabs = _mm_cvtepi8_epi16(diffabs);//back to 16-bit
- diffabs = _mm_sign_epi16(diffabs, right);//fix sign!
- acc = _mm_add_epi16(diffabs, acc); //add to acc
- //UP!
- //calculate abs
- diffabs = _mm_abs_epi16(up); //abs
- diffabs = _mm_add_epi16(diffabs, four); //+4
- diffabs = _mm_srai_epi16(diffabs, 3); //>>3
- diffabs = _mm_min_epi16(diffabs, fifteen); //min(x,15)
- diffabs = _mm_packus_epi16(diffabs, diffabs); //convert to 8
- diffabs = _mm_shuffle_epi8(lut, diffabs);//lut
- diffabs = _mm_cvtepi8_epi16(diffabs);//back to 16-bit
- diffabs = _mm_sign_epi16(diffabs, up);//fix sign!
- acc = _mm_add_epi16(diffabs, acc); //add to acc
-
- //DOWN!
- //calculate abs
- diffabs = _mm_abs_epi16(down); //abs
- diffabs = _mm_add_epi16(diffabs, four); //+4
- diffabs = _mm_srai_epi16(diffabs, 3); //>>3
- diffabs = _mm_min_epi16(diffabs, fifteen); //min(x,15)
- diffabs = _mm_packus_epi16(diffabs, diffabs); //convert to 8
- diffabs = _mm_shuffle_epi8(lut, diffabs);//lut
- diffabs = _mm_cvtepi8_epi16(diffabs);//back to 16-bit
- diffabs = _mm_sign_epi16(diffabs, down);//fix sign!
- acc = _mm_add_epi16(diffabs, acc); //add to acc
-
- //lu!
- //calculate abs
- diffabs = _mm_abs_epi16(lu); //abs
- diffabs = _mm_add_epi16(diffabs, four); //+4
- diffabs = _mm_srai_epi16(diffabs, 3); //>>3
- diffabs = _mm_min_epi16(diffabs, fifteen); //min(x,15)
- diffabs = _mm_packus_epi16(diffabs, diffabs); //convert to 8
- diffabs = _mm_shuffle_epi8(lut, diffabs);//lut
- diffabs = _mm_cvtepi8_epi16(diffabs);//back to 16-bit
- diffabs = _mm_srai_epi16(diffabs, 1);//diagonal shift!
- diffabs = _mm_sign_epi16(diffabs, lu);//fix sign!
- acc = _mm_add_epi16(diffabs, acc); //add to acc
- //ld!
- //calculate abs
- diffabs = _mm_abs_epi16(ld); //abs
- diffabs = _mm_add_epi16(diffabs, four); //+4
- diffabs = _mm_srai_epi16(diffabs, 3); //>>3
- diffabs = _mm_min_epi16(diffabs, fifteen); //min(x,15)
- diffabs = _mm_packus_epi16(diffabs, diffabs); //convert to 8
- diffabs = _mm_shuffle_epi8(lut, diffabs);//lut
- diffabs = _mm_cvtepi8_epi16(diffabs);//back to 16-bit
- diffabs = _mm_srai_epi16(diffabs, 1);//diagonal shift!
- diffabs = _mm_sign_epi16(diffabs, ld);//fix sign!
- acc = _mm_add_epi16(diffabs, acc); //add to acc
- //ru!
- //calculate abs
- diffabs = _mm_abs_epi16(ru); //abs
- diffabs = _mm_add_epi16(diffabs, four); //+4
- diffabs = _mm_srai_epi16(diffabs, 3); //>>3
- diffabs = _mm_min_epi16(diffabs, fifteen); //min(x,15)
- diffabs = _mm_packus_epi16(diffabs, diffabs); //convert to 8
- diffabs = _mm_shuffle_epi8(lut, diffabs);//lut
- diffabs = _mm_cvtepi8_epi16(diffabs);//back to 16-bit
- diffabs = _mm_srai_epi16(diffabs, 1);//diagonal shift!
- diffabs = _mm_sign_epi16(diffabs, ru);//fix sign!
- acc = _mm_add_epi16(diffabs, acc); //add to acc
- //rd!
- //calculate abs
- diffabs = _mm_abs_epi16(rd); //abs
- diffabs = _mm_add_epi16(diffabs, four); //+4
- diffabs = _mm_srai_epi16(diffabs, 3); //>>3
- diffabs = _mm_min_epi16(diffabs, fifteen); //min(x,15)
- diffabs = _mm_packus_epi16(diffabs, diffabs); //convert to 8
- diffabs = _mm_shuffle_epi8(lut, diffabs);//lut
- diffabs = _mm_cvtepi8_epi16(diffabs);//back to 16-bit
- diffabs = _mm_srai_epi16(diffabs, 1);//diagonal shift!
- diffabs = _mm_sign_epi16(diffabs, rd);//fix sign!
- acc = _mm_add_epi16(diffabs, acc); //add to acc
-
- //ll!
- //calculate abs
- diffabs = _mm_abs_epi16(ll); //abs
- diffabs = _mm_add_epi16(diffabs, four); //+4
- diffabs = _mm_srai_epi16(diffabs, 3); //>>3
- diffabs = _mm_min_epi16(diffabs, fifteen); //min(x,15)
- diffabs = _mm_packus_epi16(diffabs, diffabs); //convert to 8
- diffabs = _mm_shuffle_epi8(lut, diffabs);//lut
- diffabs = _mm_cvtepi8_epi16(diffabs);//back to 16-bit
- diffabs = _mm_srai_epi16(diffabs, 1);//diagonal shift!
- diffabs = _mm_sign_epi16(diffabs, ll);//fix sign!
- acc = _mm_add_epi16(diffabs, acc); //add to acc
- //rr!
- //calculate abs
- diffabs = _mm_abs_epi16(rr); //abs
- diffabs = _mm_add_epi16(diffabs, four); //+4
- diffabs = _mm_srai_epi16(diffabs, 3); //>>3
- diffabs = _mm_min_epi16(diffabs, fifteen); //min(x,15)
- diffabs = _mm_packus_epi16(diffabs, diffabs); //convert to 8
- diffabs = _mm_shuffle_epi8(lut, diffabs);//lut
- diffabs = _mm_cvtepi8_epi16(diffabs);//back to 16-bit
- diffabs = _mm_srai_epi16(diffabs, 1);//diagonal shift!
- diffabs = _mm_sign_epi16(diffabs, rr);//fix sign!
- acc = _mm_add_epi16(diffabs, acc); //add to acc
- //uu!
- //calculate abs
- diffabs = _mm_abs_epi16(uu); //abs
- diffabs = _mm_add_epi16(diffabs, four); //+4
- diffabs = _mm_srai_epi16(diffabs, 3); //>>3
- diffabs = _mm_min_epi16(diffabs, fifteen); //min(x,15)
- diffabs = _mm_packus_epi16(diffabs, diffabs); //convert to 8
- diffabs = _mm_shuffle_epi8(lut, diffabs);//lut
- diffabs = _mm_cvtepi8_epi16(diffabs);//back to 16-bit
- diffabs = _mm_srai_epi16(diffabs, 1);//diagonal shift!
- diffabs = _mm_sign_epi16(diffabs, uu);//fix sign!
- acc = _mm_add_epi16(diffabs, acc); //add to acc
- //dd!
- //calculate abs
- diffabs = _mm_abs_epi16(dd); //abs
- diffabs = _mm_add_epi16(diffabs, four); //+4
- diffabs = _mm_srai_epi16(diffabs, 3); //>>3
- diffabs = _mm_min_epi16(diffabs, fifteen); //min(x,15)
- diffabs = _mm_packus_epi16(diffabs, diffabs); //convert to 8
- diffabs = _mm_shuffle_epi8(lut, diffabs);//lut
- diffabs = _mm_cvtepi8_epi16(diffabs);//back to 16-bit
- diffabs = _mm_srai_epi16(diffabs, 1);//diagonal shift!
- diffabs = _mm_sign_epi16(diffabs, dd);//fix sign!
- acc = _mm_add_epi16(diffabs, acc); //add to acc
+ // apply LUT
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ simdBifApplyLut(left, acc, cutBitsNum, bitsRound, bitsRound2, lut1);
+ simdBifApplyLut(right, acc, cutBitsNum, bitsRound, bitsRound2, lut1);
+ simdBifApplyLut(up, acc, cutBitsNum, bitsRound, bitsRound2, lut1);
+ simdBifApplyLut(down, acc, cutBitsNum, bitsRound, bitsRound2, lut1);
+
+ simdBifApplyLut(lu, acc, cutBitsNum, bitsRound, bitsRound2, lut2);
+ simdBifApplyLut(ld, acc, cutBitsNum, bitsRound, bitsRound2, lut2);
+ simdBifApplyLut(ru, acc, cutBitsNum, bitsRound, bitsRound2, lut2);
+ simdBifApplyLut(rd, acc, cutBitsNum, bitsRound, bitsRound2, lut2);
+
+ simdBifApplyLut(ll, acc, cutBitsNum, bitsRound, bitsRound2, lut3);
+ simdBifApplyLut(rr, acc, cutBitsNum, bitsRound, bitsRound2, lut3);
+ simdBifApplyLut(uu, acc, cutBitsNum, bitsRound, bitsRound2, lut3);
+ simdBifApplyLut(dd, acc, cutBitsNum, bitsRound, bitsRound2, lut3);
+#else
+ simdBifApplyLut(left, acc, lut, lutShift1);
+ simdBifApplyLut(right, acc, lut, lutShift1);
+ simdBifApplyLut(up, acc, lut, lutShift1);
+ simdBifApplyLut(down, acc, lut, lutShift1);
+
+ simdBifApplyLut(lu, acc, lut, lutShift2);
+ simdBifApplyLut(ld, acc, lut, lutShift2);
+ simdBifApplyLut(ru, acc, lut, lutShift2);
+ simdBifApplyLut(rd, acc, lut, lutShift2);
+
+ simdBifApplyLut(ll, acc, lut, lutShift3);
+ simdBifApplyLut(rr, acc, lut, lutShift3);
+ simdBifApplyLut(uu, acc, lut, lutShift3);
+ simdBifApplyLut(dd, acc, lut, lutShift3);
+#endif
+ // TU scaling
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ acc = _mm_mullo_epi16(acc, mmBfac);
+ acc = _mm_adds_epi16(acc, roundAdd);
+ acc = _mm_srai_epi16(acc, bifRoundShift + 3);
+#else
if (bfac == 2)
{
acc = _mm_slli_epi16(acc, 1); // Shift left to get 2*
}
else if (bfac == 3)
{
- temp = _mm_slli_epi16(acc, 1); // Multiply by two by shifting left
- acc = _mm_add_epi16(acc, temp); // Add original value to get 3*
+ acc = _mm_add_epi16(acc, _mm_slli_epi16(acc, 1)); // Multiply by two by shifting left and add original value to get 3*
}
// Add 16 and shift 5
acc = _mm_add_epi16(acc, roundAdd);
acc = _mm_srai_epi16(acc, bifRoundShift);
-
+#endif
+
// Instead we add our input values to the delta
if(isRDO)
{
@@ -292,26 +471,76 @@ void BilateralFilter::simdFilterDiamond5x5( uint32_t uiWidth, uint32_t uiHeight,
acc = _mm_min_epi16( acc, clipmax );
}
- _mm_store_si128((__m128i*)(blkFilt + (row + pad) * (padwidth + 4) + col + pad), acc);
+ _mm_storeu_si128((__m128i*)(blkFilt + (row + pad) * padwidth + col + pad), acc);
}
}
// Copy back from tempbufFilter to recBuf
int onerow = uiWidth * sizeof(Pel);
- for(uint32_t yy = 0; yy < uiHeight; yy++)
+ // Copy back parameters
+ Pel* tempBlockPtr = (short*)blkFilt + pad * padwidth + pad;
+ for (uint32_t yy = 0; yy < uiHeight; yy++)
{
std::memcpy(recPtr, tempBlockPtr, onerow);
recPtr += recStride;
- tempBlockPtr += tempBlockStride;
+ tempBlockPtr += padwidth;
+ }
+}
+
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+template<X86_VEXT vext>
+int BilateralFilter::simdCalcMAD(int16_t* block, int stride, int width, int height, int whlog2)
+{
+ if (width < 4)
+ {
+ return calcMAD(block, stride, width, height, whlog2);
+ }
+ int sum32[4];
+ __m128i acc32, val, average;
+ acc32 = _mm_setzero_si128();
+ for (int i = 0; i < height; i++)
+ {
+ for (int j = 0; j < width; j += 4)
+ {
+ val = _mm_loadl_epi64((__m128i*)(block + j));
+ val = _mm_cvtepi16_epi32(val);
+ acc32 = _mm_add_epi32(acc32, val);
+ }
+ block += stride;
}
+ block -= stride * height;
+ acc32 = _mm_hadd_epi32(acc32, acc32);
+ _mm_storeu_si128((__m128i*)sum32, acc32);
+ average = _mm_set1_epi32((sum32[0] + sum32[1] + (1 << (whlog2 - 1))) >> whlog2);
+ acc32 = _mm_setzero_si128();
+ for (int i = 0; i < height; i++)
+ {
+ for (int j = 0; j < width; j += 4)
+ {
+ val = _mm_loadl_epi64((__m128i*)(block + j));
+ val = _mm_cvtepi16_epi32(val);
+ acc32 = _mm_add_epi32(acc32, _mm_abs_epi32(_mm_sub_epi32(val, average)));
+ }
+ block += stride;
+ }
+ acc32 = _mm_hadd_epi32(acc32, acc32);
+ _mm_storeu_si128((__m128i*)sum32, acc32);
+ return (sum32[0] + sum32[1] + (1 << (whlog2 - 1))) >> whlog2;
}
+#endif
+
+#endif // USE_AVX2
template <X86_VEXT vext>
void BilateralFilter::_initBilateralFilterX86()
{
- m_bilateralFilterDiamond5x5 = simdFilterDiamond5x5<vext>;
+ m_bilateralFilterDiamond5x5 = simdFilterDiamond5x5<vext>;
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ m_calcMAD = simdCalcMAD<vext>;
+#endif
}
template void BilateralFilter::_initBilateralFilterX86<SIMDX86>();
+
#endif
#endif // TARGET_SIMD_X86
diff --git a/source/Lib/EncoderLib/EncCu.cpp b/source/Lib/EncoderLib/EncCu.cpp
index 0640e4a2a3af3fb2b0a02b441633c052048b8fc8..ca0e53773812991c707ab37594bc204d420879a1 100644
--- a/source/Lib/EncoderLib/EncCu.cpp
+++ b/source/Lib/EncoderLib/EncCu.cpp
@@ -103,6 +103,11 @@ EncCu::EncCu() : m_GeoModeTest
void EncCu::create( EncCfg* encCfg )
{
+#if JVET_V0094_BILATERAL_FILTER || JVET_X0071_CHROMA_BILATERAL_FILTER
+ m_bilateralFilter = new BilateralFilter();;
+ m_bilateralFilter->create();
+#endif
+
unsigned uiMaxWidth = encCfg->getMaxCUWidth();
unsigned uiMaxHeight = encCfg->getMaxCUHeight();
ChromaFormat chromaFormat = encCfg->getChromaFormatIdc();
@@ -296,6 +301,7 @@ void EncCu::destroy()
unsigned numHeights = gp_sizeIdxInfo->numHeights();
#if JVET_V0094_BILATERAL_FILTER || JVET_X0071_CHROMA_BILATERAL_FILTER
+ m_bilateralFilter->destroy();
delete m_bilateralFilter;
#endif
@@ -16993,17 +16999,19 @@ void EncCu::xCalDebCost( CodingStructure &cs, Partitioner &partitioner, bool cal
for (auto &currTU : CU::traverseTUs(*cu))
{
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ bool applyBIF = m_bilateralFilter->getApplyBIF(currTU, COMPONENT_Y);
+#else
bool isInter = (cu->predMode == MODE_INTER) ? true : false;
- if ((TU::getCbf(currTU, COMPONENT_Y) || isInter == false) && (currTU.cu->qp > 17) && (128 > std::max(currTU.lumaSize().width, currTU.lumaSize().height)))
+ bool applyBIF = ((TU::getCbf(currTU, COMPONENT_Y) || isInter == false) && (currTU.cu->qp > 17)) && (128 > std::max(currTU.lumaSize().width, currTU.lumaSize().height)) && ((isInter == false) || (32 > std::min(currTU.lumaSize().width, currTU.lumaSize().height)));
+#endif
+ if (applyBIF)
{
- if ((isInter == false) || (32 > std::min(currTU.lumaSize().width, currTU.lumaSize().height)))
- {
- CompArea &compArea = currTU.block(COMPONENT_Y);
- PelBuf recBuf = picDbBuf.getBuf(compArea);
- PelBuf recIPredBuf = recBuf;
- std::vector<Pel> invLUT;
- m_bilateralFilter->bilateralFilterRDOdiamond5x5( COMPONENT_Y, recBuf, recBuf, recBuf, currTU.cu->qp, recIPredBuf, cs.slice->clpRng(COMPONENT_Y), currTU, true, false, &invLUT );
- }
+ CompArea& compArea = currTU.block(COMPONENT_Y);
+ PelBuf recBuf = picDbBuf.getBuf(compArea);
+ PelBuf recIPredBuf = recBuf;
+ std::vector<Pel> invLUT;
+ m_bilateralFilter->bilateralFilterRDOdiamond5x5(COMPONENT_Y, recBuf, recBuf, recBuf, currTU.cu->qp, recIPredBuf, cs.slice->clpRng(COMPONENT_Y), currTU, true, false, &invLUT);
}
}
}
@@ -17011,10 +17019,6 @@ void EncCu::xCalDebCost( CodingStructure &cs, Partitioner &partitioner, bool cal
#if JVET_X0071_CHROMA_BILATERAL_FILTER
if(cs.pps->getUseChromaBIF())
{
- bool tuValid = false;
- bool tuCBF = false;
- bool isDualTree = CS::isDualITree(cs);
- bool chromaValid = cu->Cb().valid() && cu->Cr().valid();
bool applyChromaBIF = false;
if (leftEdgeAvai && topEdgeAvai)
@@ -17030,10 +17034,17 @@ void EncCu::xCalDebCost( CodingStructure &cs, Partitioner &partitioner, bool cal
for (auto &currTU : CU::traverseTUs(*cu))
{
- bool isInter = (cu->predMode == MODE_INTER) ? true : false;
for(int compIdx = COMPONENT_Cb; compIdx < MAX_NUM_COMPONENT; compIdx++)
{
ComponentID compID = ComponentID( compIdx );
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ applyChromaBIF = m_bilateralFilter->getApplyBIF(currTU, compID);
+#else
+ bool tuValid = false;
+ bool tuCBF = false;
+ bool isInter = (cu->predMode == MODE_INTER) ? true : false;
+ bool isDualTree = CS::isDualITree(cs);
+ bool chromaValid = cu->Cb().valid() && cu->Cr().valid();
applyChromaBIF = false;
if(!isDualTree && chromaValid)
{
@@ -17051,6 +17062,7 @@ void EncCu::xCalDebCost( CodingStructure &cs, Partitioner &partitioner, bool cal
tuCBF = TU::getCbf(currTU, compID);
applyChromaBIF = ((tuCBF || isInter == false) && (currTU.cu->qp > 17));
}
+#endif
if (applyChromaBIF)
{
CompArea &compArea = currTU.block(compID);
@@ -17799,17 +17811,20 @@ void EncCu::xReuseCachedResult( CodingStructure *&tempCS, CodingStructure *&best
tmpRecLuma.rspSignal(m_pcReshape->getInvLUT());
}
}
- if(tempCS->pps->getUseBIF() && isLuma(compID) && cu.qp > 17)
+ if(tempCS->pps->getUseBIF() && isLuma(compID))
{
for (auto &currTU : CU::traverseTUs(cu))
{
- Position tuPosInCu = currTU.lumaPos() - cu.lumaPos();
- PelBuf tmpSubBuf = tmpRecLuma.subBuf(tuPosInCu, currTU.lumaSize());
-
- bool isInter = (cu.predMode == MODE_INTRA) ? false : true;
-
- if ((TU::getCbf(currTU, compID ) || isInter == false) && (currTU.cu->qp > 17) && (128 > std::max(currTU.lumaSize().width, currTU.lumaSize().height)) && ((isInter == false) || (32 > std::min(currTU.lumaSize().width, currTU.lumaSize().height))))
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ bool applyBIF = m_bilateralFilter->getApplyBIF(currTU, compID);
+#else
+ bool isInter = (cu.predMode == MODE_INTER) ? true : false;
+ bool applyBIF = ((TU::getCbf(currTU, COMPONENT_Y) || isInter == false) && (currTU.cu->qp > 17)) && (128 > std::max(currTU.lumaSize().width, currTU.lumaSize().height)) && ((isInter == false) || (32 > std::min(currTU.lumaSize().width, currTU.lumaSize().height)));
+#endif
+ if (applyBIF)
{
+ Position tuPosInCu = currTU.lumaPos() - cu.lumaPos();
+ PelBuf tmpSubBuf = tmpRecLuma.subBuf(tuPosInCu, currTU.lumaSize());
CompArea compArea = currTU.blocks[compID];
PelBuf recIPredBuf = tempCS->slice->getPic()->getRecoBuf(compArea);
// Do we need to use clipArea?
@@ -17837,18 +17852,19 @@ void EncCu::xReuseCachedResult( CodingStructure *&tempCS, CodingStructure *&best
tmpRecChroma.copyFrom(reco);
}
- if(tempCS->pps->getUseChromaBIF() && isChroma(compID) && (cu.qp > 17))
+ if(tempCS->pps->getUseChromaBIF() && isChroma(compID))
{
- bool tuValid = false;
- bool tuCBF = false;
- bool isDualTree = CS::isDualITree(*tempCS);
- bool chromaValid = cu.Cb().valid() && cu.Cr().valid();
bool applyChromaBIF = false;
for (auto &currTU : CU::traverseTUs(cu))
{
- Position tuPosInCu = currTU.chromaPos() - cu.chromaPos();
- PelBuf tmpSubBuf = tmpRecChroma.subBuf(tuPosInCu, currTU.chromaSize());
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ applyChromaBIF = m_bilateralFilter->getApplyBIF(currTU, compID);
+#else
+ bool tuValid = false;
+ bool tuCBF = false;
bool isInter = (cu.predMode == MODE_INTER) ? true : false;
+ bool isDualTree = CS::isDualITree(*tempCS);
+ bool chromaValid = cu.Cb().valid() && cu.Cr().valid();
applyChromaBIF = false;
if(!isDualTree && chromaValid)
{
@@ -17865,9 +17881,12 @@ void EncCu::xReuseCachedResult( CodingStructure *&tempCS, CodingStructure *&best
{
applyChromaBIF = ((TU::getCbf(currTU, compID) || isInter == false) && (currTU.cu->qp > 17));
}
+#endif
if(applyChromaBIF)
{
+ Position tuPosInCu = currTU.chromaPos() - cu.chromaPos();
+ PelBuf tmpSubBuf = tmpRecChroma.subBuf(tuPosInCu, currTU.chromaSize());
CompArea compArea = currTU.blocks[compID];
PelBuf recIPredBuf = tempCS->slice->getPic()->getRecoBuf(compArea);
m_bilateralFilter->bilateralFilterRDOdiamond5x5( compID, tmpSubBuf, tmpSubBuf, tmpSubBuf, currTU.cu->qp, recIPredBuf, tempCS->slice->clpRng(compID), currTU, true );
@@ -17917,7 +17936,7 @@ void EncCu::xReuseCachedResult( CodingStructure *&tempCS, CodingStructure *&best
tmpRecChroma.copyFrom(reco);
}
- if(tempCS->pps->getUseChromaBIF() && isChroma(compID) && (cu.qp > 17))
+ if(tempCS->pps->getUseChromaBIF() && isChroma(compID))
{
bool tuValid = false;
bool tuCBF = false;
@@ -17927,9 +17946,9 @@ void EncCu::xReuseCachedResult( CodingStructure *&tempCS, CodingStructure *&best
for (auto &currTU : CU::traverseTUs(cu))
{
- Position tuPosInCu = currTU.chromaPos() - cu.chromaPos();
- PelBuf tmpSubBuf = tmpRecChroma.subBuf(tuPosInCu, currTU.chromaSize());
-
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ applyChromaBIF = m_bilateralFilter->getApplyBIF(currTU, compID);
+#else
bool isInter = (cu.predMode == MODE_INTER) ? true : false;
applyChromaBIF = false;
if(!isDualTree && chromaValid)
@@ -17946,8 +17965,11 @@ void EncCu::xReuseCachedResult( CodingStructure *&tempCS, CodingStructure *&best
{
applyChromaBIF = ((TU::getCbf(currTU, compID) || isInter == false) && (currTU.cu->qp > 17));
}
+#endif
if(applyChromaBIF)
{
+ Position tuPosInCu = currTU.chromaPos() - cu.chromaPos();
+ PelBuf tmpSubBuf = tmpRecChroma.subBuf(tuPosInCu, currTU.chromaSize());
CompArea compArea = currTU.blocks[compID];
PelBuf recIPredBuf = tempCS->slice->getPic()->getRecoBuf(compArea);
m_bilateralFilter->bilateralFilterRDOdiamond5x5( compID, tmpSubBuf, tmpSubBuf, tmpSubBuf, currTU.cu->qp, recIPredBuf, tempCS->slice->clpRng(compID), currTU, true );
diff --git a/source/Lib/EncoderLib/EncCu.h b/source/Lib/EncoderLib/EncCu.h
index 5721eef609961649593b208c1f90c69d7ff97a2b..f1dd78ab07b5c5f5772cae100a312d793bbd78fa 100644
--- a/source/Lib/EncoderLib/EncCu.h
+++ b/source/Lib/EncoderLib/EncCu.h
@@ -433,7 +433,7 @@ public:
EncModeCtrl* getModeCtrl () { return m_modeCtrl; }
#if JVET_V0094_BILATERAL_FILTER || JVET_X0071_CHROMA_BILATERAL_FILTER
- BilateralFilter *m_bilateralFilter = new BilateralFilter();
+ BilateralFilter *m_bilateralFilter;
#endif
void setMergeBestSATDCost(double cost) { m_mergeBestSATDCost = cost; }
diff --git a/source/Lib/EncoderLib/EncSampleAdaptiveOffset.cpp b/source/Lib/EncoderLib/EncSampleAdaptiveOffset.cpp
index 74df2581a152cb798c517853c2f0700e21cc72c5..117a3a5a2ae3e13a13a448f2d4d4acec0cb363fe 100644
--- a/source/Lib/EncoderLib/EncSampleAdaptiveOffset.cpp
+++ b/source/Lib/EncoderLib/EncSampleAdaptiveOffset.cpp
@@ -1423,9 +1423,13 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
{
for (auto& currTU : CU::traverseTUs(currCU))
{
-
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ bool applyBIF = bifParams.ctuOn[ctuRsAddr] && m_bilateralFilter.getApplyBIF(currTU, COMPONENT_Y);
+#else
bool isInter = (currCU.predMode == MODE_INTER) ? true : false;
- if (bifParams.ctuOn[ctuRsAddr] && ((TU::getCbf(currTU, COMPONENT_Y) || isInter == false) && (currTU.cu->qp > 17)) && (128 > std::max(currTU.lumaSize().width, currTU.lumaSize().height)) && ((isInter == false) || (32 > std::min(currTU.lumaSize().width, currTU.lumaSize().height))))
+ bool applyBIF = bifParams.ctuOn[ctuRsAddr] && ((TU::getCbf(currTU, COMPONENT_Y) || isInter == false) && (currTU.cu->qp > 17)) && (128 > std::max(currTU.lumaSize().width, currTU.lumaSize().height)) && ((isInter == false) || (32 > std::min(currTU.lumaSize().width, currTU.lumaSize().height)));
+#endif
+ if (applyBIF)
{
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
bool clipTop = false, clipBottom = false, clipLeft = false, clipRight = false;
@@ -1449,8 +1453,6 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
#if JVET_X0071_CHROMA_BILATERAL_FILTER
if(cs.pps->getUseChromaBIF())
{
- bool tuValid = false;
- bool tuCBF = false;
bool isDualTree = CS::isDualITree(cs);
ChannelType chType = isDualTree ? CH_C : CH_L;
bool applyChromaBIF = false;
@@ -1469,13 +1471,18 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
}
for (auto &currTU : CU::traverseTUs(currCU))
{
- bool isInter = (currCU.predMode == MODE_INTER) ? true : false;
for(int compIdx = COMPONENT_Cb ; compIdx < MAX_NUM_COMPONENT; compIdx++)
{
ComponentID compID = ComponentID( compIdx );
BifParams& chromaBifParams = cs.picture->getBifParam( compID );
bool ctuEnableChromaBIF = chromaBifParams.ctuOn[ctuRsAddr];
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ applyChromaBIF = ctuEnableChromaBIF && m_bilateralFilter.getApplyBIF(currTU, compID);
+#else
+ bool tuValid = false;
+ bool tuCBF = false;
+ bool isInter = (currCU.predMode == MODE_INTER) ? true : false;
applyChromaBIF = false;
if(!isDualTree)
{
@@ -1492,6 +1499,7 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
tuCBF = TU::getCbf(currTU, compID);
applyChromaBIF = (ctuEnableChromaBIF && ((tuCBF || isInter == false) && (currTU.cu->qp > 17)));
}
+#endif
if(applyChromaBIF)
{
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
@@ -1534,7 +1542,7 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
// We don't need to clip if SAO was not performed on luma.
SAOBlkParam mySAOblkParam = cs.picture->getSAO()[ctuRsAddr];
SAOOffset& myCtbOffset = mySAOblkParam[0];
- BifParams& bifParams = cs.picture->getBifParam();
+ BifParams& bifParams = cs.picture->getBifParam(COMPONENT_Y);
bool clipLumaIfNoBilat = false;
if(myCtbOffset.modeIdc != SAO_MODE_OFF)
@@ -1561,9 +1569,14 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
{
for (auto &currTU : CU::traverseTUs(currCU))
{
-
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ bool applyBIF = bifParams.ctuOn[ctuRsAddr] && m_bilateralFilter.getApplyBIF(currTU, COMPONENT_Y);
+#else
bool isInter = (currCU.predMode == MODE_INTER) ? true : false;
- if ( bifParams.ctuOn[ctuRsAddr] && ((TU::getCbf(currTU, COMPONENT_Y) || isInter == false) && (currTU.cu->qp > 17)) && (128 > std::max(currTU.lumaSize().width, currTU.lumaSize().height)) && ((isInter == false) || (32 > std::min(currTU.lumaSize().width, currTU.lumaSize().height))))
+ bool applyBIF = bifParams.ctuOn[ctuRsAddr] && ((TU::getCbf(currTU, COMPONENT_Y) || isInter == false) && (currTU.cu->qp > 17)) && (128 > std::max(currTU.lumaSize().width, currTU.lumaSize().height)) && ((isInter == false) || (32 > std::min(currTU.lumaSize().width, currTU.lumaSize().height)));
+#endif
+
+ if (applyBIF)
{
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
bool clipTop = false, clipBottom = false, clipLeft = false, clipRight = false;
@@ -1574,7 +1587,7 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
cs, currTU.Y().x, currTU.Y().y, currTU.lumaSize().width, currTU.lumaSize().height, clipTop,
clipBottom, clipLeft, clipRight, numHorVirBndry, numVerVirBndry, horVirBndryPos, verVirBndryPos);
#endif
- bilateralFilter.bilateralFilterDiamond5x5( COMPONENT_Y, srcYuv, resYuv, currTU.cu->qp, cs.slice->clpRng(COMPONENT_Y), currTU
+ bilateralFilter.bilateralFilterDiamond5x5( COMPONENT_Y, srcYuv, resYuv, currTU.cu->qp, cs.slice->clpRng(COMPONENT_Y), currTU, false
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
, isTUCrossedByVirtualBoundaries, horVirBndryPos, verVirBndryPos, numHorVirBndry, numVerVirBndry
, clipTop, clipBottom, clipLeft, clipRight
@@ -1608,8 +1621,6 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
}
if(cs.pps->getUseChromaBIF())
{
- bool tuValid = false;
- bool tuCBF = false;
bool isDualTree = CS::isDualITree(cs);
ChannelType chType = isDualTree ? CH_C : CH_L;
bool applyChromaBIF = false;
@@ -1623,12 +1634,17 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
}
for (auto &currTU : CU::traverseTUs(currCU))
{
- bool isInter = (currCU.predMode == MODE_INTER) ? true : false;
for(int compIdx = COMPONENT_Cb; compIdx < MAX_NUM_COMPONENT; compIdx++)
{
ComponentID compID = ComponentID( compIdx );
BifParams& chromaBifParams = cs.picture->getBifParam( compID );
bool ctuEnableChromaBIF = chromaBifParams.ctuOn[ctuRsAddr];
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ applyChromaBIF = ctuEnableChromaBIF && m_bilateralFilter.getApplyBIF(currTU, compID);
+#else
+ bool tuValid = false;
+ bool tuCBF = false;
+ bool isInter = (currCU.predMode == MODE_INTER) ? true : false;
applyChromaBIF = false;
if(!isDualTree)
{
@@ -1645,6 +1661,7 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
tuCBF = TU::getCbf(currTU, compID);
applyChromaBIF = (ctuEnableChromaBIF && ((tuCBF || isInter == false) && (currTU.cu->qp > 17)));
}
+#endif
if(applyChromaBIF)
{
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
@@ -1662,7 +1679,7 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
clipLeft, clipRight, numHorVirBndry, numVerVirBndry, horVirBndryPos, verVirBndryPos);
#endif
- m_bilateralFilter.bilateralFilterDiamond5x5(compID, srcYuv, resYuv, currTU.cu->qp, cs.slice->clpRng(compID), currTU, false,
+ m_bilateralFilter.bilateralFilterDiamond5x5(compID, srcYuv, resYuv, currTU.cu->qp, cs.slice->clpRng(compID), currTU, false
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
, isTUCrossedByVirtualBoundaries, horVirBndryPos, verVirBndryPos, numHorVirBndry, numVerVirBndry, clipTop, clipBottom, clipLeft, clipRight
#endif
@@ -1771,6 +1788,9 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
ComponentID compID = ComponentID( compIdx );
BifParams& chromaBifParams = cs.picture->getBifParam( compID );
bool ctuEnableChromaBIF = chromaBifParams.ctuOn[ctuRsAddr];
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ applyChromaBIF = ctuEnableChromaBIF && m_bilateralFilter.getApplyBIF(currTU, compID);
+#else
applyChromaBIF = false;
if(!isDualTree)
{
@@ -1787,6 +1807,7 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
tuCBF = TU::getCbf(currTU, compID);
applyChromaBIF = (ctuEnableChromaBIF && ((tuCBF || isInter == false) && (currTU.cu->qp > 17)));
}
+#endif
if(applyChromaBIF)
{
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
@@ -1892,9 +1913,13 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
{
for (auto& currTU : CU::traverseTUs(currCU))
{
-
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ bool applyBIF = bifParams.ctuOn[ctuRsAddr] && m_bilateralFilter.getApplyBIF(currTU, COMPONENT_Y);
+#else
bool isInter = (currCU.predMode == MODE_INTER) ? true : false;
- if (bifParams.ctuOn[ctuRsAddr] && ((TU::getCbf(currTU, COMPONENT_Y) || isInter == false) && (currTU.cu->qp > 17)) && (128 > std::max(currTU.lumaSize().width, currTU.lumaSize().height)) && ((isInter == false) || (32 > std::min(currTU.lumaSize().width, currTU.lumaSize().height))))
+ bool applyBIF = bifParams.ctuOn[ctuRsAddr] && ((TU::getCbf(currTU, COMPONENT_Y) || isInter == false) && (currTU.cu->qp > 17)) && (128 > std::max(currTU.lumaSize().width, currTU.lumaSize().height)) && ((isInter == false) || (32 > std::min(currTU.lumaSize().width, currTU.lumaSize().height)));
+#endif
+ if (applyBIF)
{
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
bool clipTop = false, clipBottom = false, clipLeft = false, clipRight = false;
@@ -1919,8 +1944,6 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
#if JVET_X0071_CHROMA_BILATERAL_FILTER
if(cs.pps->getUseChromaBIF())
{
- bool tuValid = false;
- bool tuCBF = false;
bool isDualTree = CS::isDualITree(cs);
ChannelType chType = isDualTree ? CH_C : CH_L;
bool applyChromaBIF = false;
@@ -1939,13 +1962,18 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
for (auto &currTU : CU::traverseTUs(currCU))
{
- bool isInter = (currCU.predMode == MODE_INTER) ? true : false;
for(int compIdx = COMPONENT_Cb; compIdx < MAX_NUM_COMPONENT; compIdx++)
{
ComponentID compID = ComponentID( compIdx );
BifParams& chromaBifParams = cs.picture->getBifParam( compID );
bool ctuEnableChromaBIF = chromaBifParams.ctuOn[ctuRsAddr];
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ applyChromaBIF = ctuEnableChromaBIF && m_bilateralFilter.getApplyBIF(currTU, compID);
+#else
+ bool tuValid = false;
+ bool tuCBF = false;
+ bool isInter = (currCU.predMode == MODE_INTER) ? true : false;
applyChromaBIF = false;
if(!isDualTree)
{
@@ -1962,6 +1990,7 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
tuCBF = TU::getCbf(currTU, compID);
applyChromaBIF = (ctuEnableChromaBIF && ((tuCBF || isInter == false) && (currTU.cu->qp > 17)));
}
+#endif
if(applyChromaBIF)
{
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
@@ -2022,8 +2051,13 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
{
for (auto &currTU : CU::traverseTUs(currCU))
{
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ bool applyBIF = bifParams.ctuOn[ctuRsAddr] && m_bilateralFilter.getApplyBIF(currTU, COMPONENT_Y);
+#else
bool isInter = (currCU.predMode == MODE_INTER) ? true : false;
- if ( bifParams.ctuOn[ctuRsAddr] && ((TU::getCbf(currTU, COMPONENT_Y) || isInter == false) && (currTU.cu->qp > 17)) && (128 > std::max(currTU.lumaSize().width, currTU.lumaSize().height)) && ((isInter == false) || (32 > std::min(currTU.lumaSize().width, currTU.lumaSize().height))))
+ bool applyBIF = bifParams.ctuOn[ctuRsAddr] && ((TU::getCbf(currTU, COMPONENT_Y) || isInter == false) && (currTU.cu->qp > 17)) && (128 > std::max(currTU.lumaSize().width, currTU.lumaSize().height)) && ((isInter == false) || (32 > std::min(currTU.lumaSize().width, currTU.lumaSize().height)));
+#endif
+ if (applyBIF)
{
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
bool clipTop = false, clipBottom = false, clipLeft = false, clipRight = false;
@@ -2034,7 +2068,7 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
cs, currTU.Y().x, currTU.Y().y, currTU.lumaSize().width, currTU.lumaSize().height, clipTop,
clipBottom, clipLeft, clipRight, numHorVirBndry, numVerVirBndry, horVirBndryPos, verVirBndryPos);
#endif
- bilateralFilter.bilateralFilterDiamond5x5( COMPONENT_Y, srcYuv, resYuv, currTU.cu->qp, cs.slice->clpRng(COMPONENT_Y), currTU
+ bilateralFilter.bilateralFilterDiamond5x5( COMPONENT_Y, srcYuv, resYuv, currTU.cu->qp, cs.slice->clpRng(COMPONENT_Y), currTU, false
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
, isTUCrossedByVirtualBoundaries, horVirBndryPos, verVirBndryPos, numHorVirBndry, numVerVirBndry
, clipTop, clipBottom, clipLeft, clipRight
@@ -2068,8 +2102,6 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
if(cs.pps->getUseChromaBIF())
{
- bool tuValid = false;
- bool tuCBF = false;
bool isDualTree = CS::isDualITree(cs);
ChannelType chType = isDualTree ? CH_C : CH_L;
bool applyChromaBIF = false;
@@ -2083,13 +2115,18 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
}
for (auto &currTU : CU::traverseTUs(currCU))
{
- bool isInter = (currCU.predMode == MODE_INTER) ? true : false;
for(int compIdx = COMPONENT_Cb; compIdx < MAX_NUM_COMPONENT; compIdx++)
{
ComponentID compID = ComponentID( compIdx );
BifParams& chromaBifParams = cs.picture->getBifParam( compID );
bool ctuEnableChromaBIF = chromaBifParams.ctuOn[ctuRsAddr];
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ applyChromaBIF = ctuEnableChromaBIF && m_bilateralFilter.getApplyBIF(currTU, compID);
+#else
+ bool tuValid = false;
+ bool tuCBF = false;
+ bool isInter = (currCU.predMode == MODE_INTER) ? true : false;
applyChromaBIF = false;
if(!isDualTree)
{
@@ -2106,6 +2143,7 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
tuCBF = TU::getCbf(currTU, compID);
applyChromaBIF = (ctuEnableChromaBIF && ((tuCBF || isInter == false) && (currTU.cu->qp > 17)));
}
+#endif
if(applyChromaBIF)
{
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
@@ -2123,7 +2161,7 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
clipLeft, clipRight, numHorVirBndry, numVerVirBndry, horVirBndryPos, verVirBndryPos);
#endif
- bilateralFilter.bilateralFilterDiamond5x5(compID, srcYuv, resYuv, currTU.cu->qp, cs.slice->clpRng(compID), currTU, false,
+ bilateralFilter.bilateralFilterDiamond5x5(compID, srcYuv, resYuv, currTU.cu->qp, cs.slice->clpRng(compID), currTU, false
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
,isTUCrossedByVirtualBoundaries, horVirBndryPos, verVirBndryPos, numHorVirBndry, numVerVirBndry, clipTop, clipBottom, clipLeft, clipRight
#endif
@@ -2195,9 +2233,13 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
{
for (auto &currTU : CU::traverseTUs(currCU))
{
-
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ bool applyBIF = bifParams.ctuOn[ctuRsAddr] && m_bilateralFilter.getApplyBIF(currTU, COMPONENT_Y);
+#else
bool isInter = (currCU.predMode == MODE_INTER) ? true : false;
- if ( bifParams.ctuOn[ctuRsAddr] && ((TU::getCbf(currTU, COMPONENT_Y) || isInter == false) && (currTU.cu->qp > 17)) && (128 > std::max(currTU.lumaSize().width, currTU.lumaSize().height)) && ((isInter == false) || (32 > std::min(currTU.lumaSize().width, currTU.lumaSize().height))))
+ bool applyBIF = bifParams.ctuOn[ctuRsAddr] && ((TU::getCbf(currTU, COMPONENT_Y) || isInter == false) && (currTU.cu->qp > 17)) && (128 > std::max(currTU.lumaSize().width, currTU.lumaSize().height)) && ((isInter == false) || (32 > std::min(currTU.lumaSize().width, currTU.lumaSize().height)));
+#endif
+ if (applyBIF)
{
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
bool clipTop = false, clipBottom = false, clipLeft = false, clipRight = false;
@@ -2209,7 +2251,7 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
clipBottom, clipLeft, clipRight, numHorVirBndry, numVerVirBndry, horVirBndryPos, verVirBndryPos);
#endif
- bilateralFilter.bilateralFilterDiamond5x5( COMPONENT_Y, srcYuv, resYuv, currTU.cu->qp, cs.slice->clpRng(COMPONENT_Y), currTU
+ bilateralFilter.bilateralFilterDiamond5x5( COMPONENT_Y, srcYuv, resYuv, currTU.cu->qp, cs.slice->clpRng(COMPONENT_Y), currTU, false
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
, isTUCrossedByVirtualBoundaries, horVirBndryPos, verVirBndryPos, numHorVirBndry, numVerVirBndry
, clipTop, clipBottom, clipLeft, clipRight
@@ -2292,6 +2334,9 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
ComponentID compID = ComponentID( compIdx );
BifParams& chromaBifParams = cs.picture->getBifParam( compID );
bool ctuEnableChromaBIF = chromaBifParams.ctuOn[ctuRsAddr];
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ applyChromaBIF = ctuEnableChromaBIF && m_bilateralFilter.getApplyBIF(currTU, compID);
+#else
applyChromaBIF = false;
if(!isDualTree)
{
@@ -2308,6 +2353,7 @@ void EncSampleAdaptiveOffset::decideBlkParams(CodingStructure& cs, bool* sliceEn
tuCBF = TU::getCbf(currTU, compID);
applyChromaBIF = (ctuEnableChromaBIF && ((tuCBF || isInter == false) && (currTU.cu->qp > 17)));
}
+#endif
if(applyChromaBIF)
{
#if JVET_Z0105_LOOP_FILTER_VIRTUAL_BOUNDARY
diff --git a/source/Lib/EncoderLib/InterSearch.cpp b/source/Lib/EncoderLib/InterSearch.cpp
index aaeb87461e89e877287385cf628be316be757704..bfba7fdf823b5cc62d0523dafe14139403ed2431 100644
--- a/source/Lib/EncoderLib/InterSearch.cpp
+++ b/source/Lib/EncoderLib/InterSearch.cpp
@@ -13196,9 +13196,14 @@ void InterSearch::xEstimateInterResidualQT(CodingStructure &cs, Partitioner &par
}
#if JVET_V0094_BILATERAL_FILTER
- // getCbf() is going to be 1 since currAbsSum > 0 here, according to the if-statement a couple of lines up.
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ bool applyBIF = cs.pps->getUseBIF() && isLuma(compID) && m_bilateralFilter->getApplyBIF(tu, compID);
+#else
bool isInter = (cu.predMode == MODE_INTER) ? true : false;
- if( cs.pps->getUseBIF() && isLuma( compID ) && tu.cu->qp > 17 && 128 > std::max( tu.lumaSize().width, tu.lumaSize().height ) && ( !isInter || 32 > std::min( tu.lumaSize().width, tu.lumaSize().height ) ) )
+ // getCbf() is going to be 1 since currAbsSum > 0 here, according to the if-statement a couple of lines up.
+ bool applyBIF = cs.pps->getUseBIF() && isLuma(compID) && tu.cu->qp > 17 && 128 > std::max(tu.lumaSize().width, tu.lumaSize().height) && (!isInter || 32 > std::min(tu.lumaSize().width, tu.lumaSize().height));
+#endif
+ if(applyBIF)
{
CompArea tmpArea1( compID, tu.chromaFormat, Position(0, 0), Size(resiBuf.width, resiBuf.height));
PelBuf tmpRecLuma = m_tmpStorageLCU.getBuf(tmpArea1);
@@ -13217,7 +13222,12 @@ void InterSearch::xEstimateInterResidualQT(CodingStructure &cs, Partitioner &par
#if JVET_X0071_CHROMA_BILATERAL_FILTER
if(isChroma(compID))
{
- if (cs.pps->getUseChromaBIF() && isChroma(compID) && (tu.cu->qp > 17))
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ bool applyChromaBIF = cs.pps->getUseChromaBIF() && m_bilateralFilter->getApplyBIF(tu, compID);
+#else
+ bool applyChromaBIF = cs.pps->getUseChromaBIF() && isChroma(compID) && (tu.cu->qp > 17);
+#endif
+ if (applyChromaBIF)
{
//chroma and bilateral
CompArea tmpArea1(compID, tu.chromaFormat, Position(0, 0), Size(resiBuf.width, resiBuf.height));
@@ -14674,16 +14684,20 @@ void InterSearch::encodeResAndCalcRdInterCU(CodingStructure &cs, Partitioner &pa
tmpRecLuma.rspSignal(m_pcReshape->getInvLUT());
}
- if(cs.pps->getUseBIF() && isLuma(compID) && (cu.qp > 17))
+ if(cs.pps->getUseBIF() && isLuma(compID))
{
for (auto &currTU : CU::traverseTUs(cu))
- {
- Position tuPosInCu = currTU.lumaPos() - cu.lumaPos();
- PelBuf tmpSubBuf = tmpRecLuma.subBuf(tuPosInCu, currTU.lumaSize());
-
+ {
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ bool applyBIF = m_bilateralFilter->getApplyBIF(currTU, compID);
+#else
bool isInter = (cu.predMode == MODE_INTER) ? true : false;
- if( ( TU::getCbf( currTU, compID ) || !isInter ) && currTU.cu->qp > 17 && 128 > std::max( currTU.lumaSize().width, currTU.lumaSize().height ) && ( !isInter || 32 > std::min( currTU.lumaSize().width, currTU.lumaSize().height ) ) )
+ bool applyBIF = ((TU::getCbf(currTU, compID) || !isInter) && currTU.cu->qp > 17 && 128 > std::max(currTU.lumaSize().width, currTU.lumaSize().height) && (!isInter || 32 > std::min(currTU.lumaSize().width, currTU.lumaSize().height)));
+#endif
+ Position tuPosInCu = currTU.lumaPos() - cu.lumaPos();
+ if(applyBIF)
{
+ PelBuf tmpSubBuf = tmpRecLuma.subBuf(tuPosInCu, currTU.lumaSize());
CompArea compArea = currTU.blocks[compID];
PelBuf recIPredBuf = cs.slice->getPic()->getRecoBuf(compArea);
@@ -14716,14 +14730,19 @@ void InterSearch::encodeResAndCalcRdInterCU(CodingStructure &cs, Partitioner &pa
tmpRecChroma = m_tmpStorageLCU.getBuf(tmpArea2);
tmpRecChroma.copyFrom(reco);
- if(cs.pps->getUseChromaBIF() && isChroma(compID) && (cu.qp > 17))
+ if(cs.pps->getUseChromaBIF() && isChroma(compID))
{
for (auto &currTU : CU::traverseTUs(cu))
{
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ bool applyChromaBIF = m_bilateralFilter->getApplyBIF(currTU, compID);
+#else
+ bool isInter = (cu.predMode == MODE_INTER) ? true : false;
+ bool applyChromaBIF = (TU::getCbf(currTU, compID) || isInter == false) && (cu.qp > 17);
+#endif
Position tuPosInCu = currTU.chromaPos() - cu.chromaPos();
PelBuf tmpSubBuf = tmpRecChroma.subBuf(tuPosInCu, currTU.chromaSize());
- bool isInter = (cu.predMode == MODE_INTER) ? true : false;
- if ((TU::getCbf(currTU, compID) || isInter == false))
+ if (applyChromaBIF)
{
CompArea compArea = currTU.blocks[compID];
PelBuf recIPredBuf = cs.slice->getPic()->getRecoBuf(compArea);
@@ -14786,15 +14805,20 @@ void InterSearch::encodeResAndCalcRdInterCU(CodingStructure &cs, Partitioner &pa
CompArea tmpArea2( compID, area.chromaFormat, Position(0, 0), area.size());
tmpRecChroma = m_tmpStorageLCU.getBuf(tmpArea2);
tmpRecChroma.copyFrom(reco);
- if(cs.pps->getUseChromaBIF() && isChroma(compID) && (cu.qp > 17))
+ if(cs.pps->getUseChromaBIF() && isChroma(compID))
{
for (auto &currTU : CU::traverseTUs(cu))
{
- Position tuPosInCu = currTU.chromaPos() - cu.chromaPos();
- PelBuf tmpSubBuf = tmpRecChroma.subBuf(tuPosInCu, currTU.chromaSize());
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ bool applyChromaBIF = m_bilateralFilter->getApplyBIF(currTU, compID);
+#else
bool isInter = (cu.predMode == MODE_INTER) ? true : false;
- if ((TU::getCbf( currTU, compID ) || isInter == false))
+ bool applyChromaBIF = (TU::getCbf(currTU, compID) || isInter == false) && (cu.qp > 17);
+#endif
+ if (applyChromaBIF)
{
+ Position tuPosInCu = currTU.chromaPos() - cu.chromaPos();
+ PelBuf tmpSubBuf = tmpRecChroma.subBuf(tuPosInCu, currTU.chromaSize());
CompArea compArea = currTU.blocks[compID];
PelBuf recIPredBuf = cs.slice->getPic()->getRecoBuf(compArea);
m_bilateralFilter->bilateralFilterRDOdiamond5x5( compID, tmpSubBuf, tmpSubBuf, tmpSubBuf, currTU.cu->qp, recIPredBuf, cs.slice->clpRng(compID), currTU, true );
diff --git a/source/Lib/EncoderLib/IntraSearch.cpp b/source/Lib/EncoderLib/IntraSearch.cpp
index b48065f4d12deb16c8c37311744e11fa9e564d01..98a86a537a8622c1db850174096289f9c7f1d237 100644
--- a/source/Lib/EncoderLib/IntraSearch.cpp
+++ b/source/Lib/EncoderLib/IntraSearch.cpp
@@ -8032,7 +8032,12 @@ void IntraSearch::xIntraCodingTUBlock(TransformUnit &tu, const ComponentID &comp
tmpRecLuma.rspSignal(m_pcReshape->getInvLUT());
}
- if( pps.getUseBIF() /*&& (uiAbsSum > 0)*/ && tu.cu->qp > 17 && 128 > std::max( tu.lumaSize().width, tu.lumaSize().height ) )
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ bool applyBIF = pps.getUseBIF() && m_bilateralFilter->getApplyBIF(tu, compID);
+#else
+ bool applyBIF = pps.getUseBIF() /*&& (uiAbsSum > 0)*/ && tu.cu->qp > 17 && 128 > std::max(tu.lumaSize().width, tu.lumaSize().height);
+#endif
+ if(applyBIF)
{
CompArea compArea = tu.blocks[compID];
PelBuf recIPredBuf = cs.slice->getPic()->getRecoBuf(compArea);
@@ -8052,7 +8057,12 @@ void IntraSearch::xIntraCodingTUBlock(TransformUnit &tu, const ComponentID &comp
else
{
#if JVET_X0071_CHROMA_BILATERAL_FILTER
- if(pps.getUseChromaBIF() && isChroma(compID) && tu.cu->qp > 17)
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ bool applyChromaBIF = pps.getUseChromaBIF() && m_bilateralFilter->getApplyBIF(tu, compID);
+#else
+ bool applyChromaBIF = pps.getUseChromaBIF() && tu.cu->qp > 17;
+#endif
+ if(applyChromaBIF)
{
CompArea compArea = tu.blocks[compID];
PelBuf recIPredBuf = cs.slice->getPic()->getRecoBuf(compArea);
@@ -8084,7 +8094,12 @@ void IntraSearch::xIntraCodingTUBlock(TransformUnit &tu, const ComponentID &comp
{
if(isLuma(compID))
{
- if( pps.getUseBIF() /*&& (uiAbsSum > 0)*/ && tu.cu->qp > 17 && 128 > std::max( tu.lumaSize().width, tu.lumaSize().height ) )
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ bool applyBIF = pps.getUseBIF() && m_bilateralFilter->getApplyBIF(tu, compID);
+#else
+ bool applyBIF = pps.getUseBIF() /*&& (uiAbsSum > 0)*/ && tu.cu->qp > 17 && 128 > std::max(tu.lumaSize().width, tu.lumaSize().height);
+#endif
+ if(applyBIF)
{
CompArea compArea = tu.blocks[compID];
PelBuf recIPredBuf = cs.slice->getPic()->getRecoBuf(compArea);
@@ -8097,7 +8112,12 @@ void IntraSearch::xIntraCodingTUBlock(TransformUnit &tu, const ComponentID &comp
else
{
#if JVET_X0071_CHROMA_BILATERAL_FILTER
- if (pps.getUseChromaBIF() && isChroma(compID) && (tu.cu->qp > 17))
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ bool applyChromaBIF = pps.getUseChromaBIF() && m_bilateralFilter->getApplyBIF(tu, compID);
+#else
+ bool applyChromaBIF = pps.getUseChromaBIF() && isChroma(compID) && (tu.cu->qp > 17);
+#endif
+ if (applyChromaBIF)
{
CompArea compArea = tu.blocks[compID];
PelBuf recIPredBuf = cs.slice->getPic()->getRecoBuf(compArea);
@@ -8159,7 +8179,12 @@ void IntraSearch::xIntraCodingTUBlock(TransformUnit &tu, const ComponentID &comp
}
else
{
- if(pps.getUseChromaBIF() && isChroma(compID) && (tu.cu->qp > 17))
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ bool applyChromaBIF = pps.getUseChromaBIF() && m_bilateralFilter->getApplyBIF(tu, compID);
+#else
+ bool applyChromaBIF = pps.getUseChromaBIF() && tu.cu->qp > 17;
+#endif
+ if (applyChromaBIF)
{
CompArea compArea = tu.blocks[compID];
PelBuf recIPredBuf = cs.slice->getPic()->getRecoBuf(compArea);
@@ -8188,7 +8213,12 @@ void IntraSearch::xIntraCodingTUBlock(TransformUnit &tu, const ComponentID &comp
}
else
{
- if (pps.getUseChromaBIF() && isChroma(compID) && (tu.cu->qp > 17))
+#if JVET_AF0112_BIF_DYNAMIC_SCALING
+ bool applyChromaBIF = pps.getUseChromaBIF() && m_bilateralFilter->getApplyBIF(tu, compID);
+#else
+ bool applyChromaBIF = pps.getUseChromaBIF() && tu.cu->qp > 17;
+#endif
+ if (applyChromaBIF)
{
CompArea compArea = tu.blocks[compID];
PelBuf recIPredBuf = cs.slice->getPic()->getRecoBuf(compArea);