diff --git a/source/Lib/CommonLib/IntraPrediction.cpp b/source/Lib/CommonLib/IntraPrediction.cpp
index 0c0781095685db5f1f4236e5a35ab56be99df3ba..1812e6e72b9b4de7934b9af20e43e9e91bcea1e7 100644
--- a/source/Lib/CommonLib/IntraPrediction.cpp
+++ b/source/Lib/CommonLib/IntraPrediction.cpp
@@ -1773,7 +1773,11 @@ void IntraPrediction::initIntraPatternChTypeISP(const CodingUnit& cu, const Comp
}
#if JVET_V0130_INTRA_TMP
+#if JVET_W0069_TMP_BOUNDARY
+RefTemplateType IntraPrediction::GetRefTemplateType(CodingUnit& cu, CompArea& area)
+#else
bool IntraPrediction::isRefTemplateAvailable(CodingUnit& cu, CompArea& area)
+#endif
{
const ChannelType chType = toChannelType(area.compID);
const CodingStructure& cs = *cu.cs;
@@ -1799,7 +1803,11 @@ bool IntraPrediction::isRefTemplateAvailable(CodingUnit& cu, CompArea& area)
if( numAboveUnits <= 0 || numLeftUnits <= 0 || numAboveRightUnits <= 0 || numLeftBelowUnits <= 0 )
{
+#if JVET_W0069_TMP_BOUNDARY
+ return No_Template;
+#else
return false;
+#endif
}
// ----- Step 1: analyze neighborhood -----
@@ -1814,7 +1822,19 @@ bool IntraPrediction::isRefTemplateAvailable(CodingUnit& cu, CompArea& area)
//bool retVal = 1;
+#if JVET_W0069_TMP_BOUNDARY
+ if (isAboveLeftAvailable(cu, chType, posLT) && isAboveAvailable(cu, chType, posLT, numAboveUnits, unitWidth, (neighborFlags + totalLeftUnits + 1)) && isLeftAvailable(cu, chType, posLT, numLeftUnits, unitHeight, (neighborFlags + totalLeftUnits - 1)))
+ return L_Shape_Template;
+ else if (isAboveLeftAvailable(cu, chType, posLT))
+ return Left_Template;
+ else if (isAboveAvailable(cu, chType, posLT, numAboveUnits, unitWidth, (neighborFlags + totalLeftUnits + 1)))
+ return Up_Template;
+ else
+ return No_Template;
+ CHECK(1, "un defined template type");
+#else
return isAboveLeftAvailable(cu, chType, posLT) && isAboveAvailable(cu, chType, posLT, numAboveUnits, unitWidth, (neighborFlags + totalLeftUnits + 1)) && isLeftAvailable(cu, chType, posLT, numLeftUnits, unitHeight, (neighborFlags + totalLeftUnits - 1));
+#endif
//return retVal;
}
diff --git a/source/Lib/CommonLib/IntraPrediction.h b/source/Lib/CommonLib/IntraPrediction.h
index 9cf085f4b8859084f5d38f22a552f6e4dca7d033..b132622a284f45e968aa649af8ee7486ec988b9a 100644
--- a/source/Lib/CommonLib/IntraPrediction.h
+++ b/source/Lib/CommonLib/IntraPrediction.h
@@ -158,7 +158,7 @@ protected:
void xPredIntraBDPCM ( const CPelBuf &pSrc, PelBuf &pDst, const uint32_t dirMode, const ClpRng& clpRng );
Pel xGetPredValDc ( const CPelBuf &pSrc, const Size &dstSize );
-#if JVET_V0130_INTRA_TMP
+#if JVET_V0130_INTRA_TMP && !JVET_W0069_TMP_BOUNDARY
bool isRefTemplateAvailable(CodingUnit& cu, CompArea& area);
#endif
@@ -203,6 +203,10 @@ public:
IntraPrediction();
virtual ~IntraPrediction();
+#if JVET_W0069_TMP_BOUNDARY
+ RefTemplateType GetRefTemplateType(CodingUnit& cu, CompArea& area);
+#endif
+
void init (ChromaFormat chromaFormatIDC, const unsigned bitDepthY);
#if ENABLE_DIMD
static void deriveDimdMode (const CPelBuf &recoBuf, const CompArea &area, CodingUnit &cu);
diff --git a/source/Lib/CommonLib/TrQuant.cpp b/source/Lib/CommonLib/TrQuant.cpp
index 40d9dea86175ead1382cc38b0383f967be0025a9..548f8cc34819fedee563a6957f02a97ac21e7059 100644
--- a/source/Lib/CommonLib/TrQuant.cpp
+++ b/source/Lib/CommonLib/TrQuant.cpp
@@ -542,7 +542,11 @@ void TempLibFast::initTemplateDiff(unsigned int uiPatchWidth, unsigned int uiPat
}
}
+#if JVET_W0069_TMP_BOUNDARY
+void TrQuant::getTargetTemplate(CodingUnit* pcCU, unsigned int uiBlkWidth, unsigned int uiBlkHeight, RefTemplateType TempType)
+#else
void TrQuant::getTargetTemplate(CodingUnit* pcCU, unsigned int uiBlkWidth, unsigned int uiBlkHeight)
+#endif
{
const ComponentID compID = COMPONENT_Y;
unsigned int uiPatchWidth = uiBlkWidth + TMP_TEMPLATE_SIZE;
@@ -557,6 +561,10 @@ void TrQuant::getTargetTemplate(CodingUnit* pcCU, unsigned int uiBlkWidth, unsig
//fill template
//up-left & up
Pel* tarTemp;
+#if JVET_W0069_TMP_BOUNDARY
+ if (TempType == L_Shape_Template)
+ {
+#endif
Pel* pCurrTemp = pCurrStart - TMP_TEMPLATE_SIZE * uiPicStride - TMP_TEMPLATE_SIZE;
for (uiY = 0; uiY < TMP_TEMPLATE_SIZE; uiY++)
{
@@ -577,9 +585,42 @@ void TrQuant::getTargetTemplate(CodingUnit* pcCU, unsigned int uiBlkWidth, unsig
}
pCurrTemp += uiPicStride;
}
+#if JVET_W0069_TMP_BOUNDARY
+ }
+ else if (TempType == Up_Template)
+ {
+ Pel* pCurrTemp = pCurrStart - TMP_TEMPLATE_SIZE * uiPicStride;
+ for (uiY = 0; uiY < TMP_TEMPLATE_SIZE; uiY++)
+ {
+ tarTemp = tarPatch[uiY];
+ for (uiX = 0; uiX < uiBlkWidth; uiX++)
+ {
+ tarTemp[uiX] = pCurrTemp[uiX];
+ }
+ pCurrTemp += uiPicStride;
+ }
+ }
+ else if (TempType == Left_Template)
+ {
+ Pel* pCurrTemp = pCurrStart - TMP_TEMPLATE_SIZE;
+ for (uiY = TMP_TEMPLATE_SIZE; uiY < uiPatchHeight; uiY++)
+ {
+ tarTemp = tarPatch[uiY];
+ for (uiX = 0; uiX < TMP_TEMPLATE_SIZE; uiX++)
+ {
+ tarTemp[uiX] = pCurrTemp[uiX];
+ }
+ pCurrTemp += uiPicStride;
+ }
+ }
+#endif
}
+#if JVET_W0069_TMP_BOUNDARY
+void TrQuant::candidateSearchIntra(CodingUnit* pcCU, unsigned int uiBlkWidth, unsigned int uiBlkHeight, RefTemplateType TempType)
+#else
void TrQuant::candidateSearchIntra(CodingUnit* pcCU, unsigned int uiBlkWidth, unsigned int uiBlkHeight)
+#endif
{
const ComponentID compID = COMPONENT_Y;
const int channelBitDepth = pcCU->cs->sps->getBitDepth(toChannelType(compID));
@@ -590,7 +631,11 @@ void TrQuant::candidateSearchIntra(CodingUnit* pcCU, unsigned int uiBlkWidth, un
//Initialize the library for saving the best candidates
m_tempLibFast.initTemplateDiff(uiPatchWidth, uiPatchHeight, uiBlkWidth, uiBlkHeight, channelBitDepth);
short setId = 0; //record the reference picture.
+#if JVET_W0069_TMP_BOUNDARY
+ searchCandidateFromOnePicIntra(pcCU, tarPatch, uiPatchWidth, uiPatchHeight, setId, TempType);
+#else
searchCandidateFromOnePicIntra(pcCU, tarPatch, uiPatchWidth, uiPatchHeight, setId);
+#endif
//count collected candidate number
int pDiff = m_tempLibFast.getDiff();
int maxDiff = m_tempLibFast.getDiffMax();
@@ -606,7 +651,11 @@ void TrQuant::candidateSearchIntra(CodingUnit* pcCU, unsigned int uiBlkWidth, un
}
}
+#if JVET_W0069_TMP_BOUNDARY
+void TrQuant::searchCandidateFromOnePicIntra(CodingUnit* pcCU, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, unsigned int setId, RefTemplateType TempType)
+#else
void TrQuant::searchCandidateFromOnePicIntra(CodingUnit* pcCU, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, unsigned int setId)
+#endif
{
const ComponentID compID = COMPONENT_Y;
unsigned int uiBlkWidth = uiPatchWidth - TMP_TEMPLATE_SIZE;
@@ -685,7 +734,11 @@ void TrQuant::searchCandidateFromOnePicIntra(CodingUnit* pcCU, Pel** tarPatch,
for (iXOffset = mvXMax; iXOffset >= mvXMin; iXOffset--)
{
refCurr = ref + iYOffset * refStride + iXOffset;
+#if JVET_W0069_TMP_BOUNDARY
+ diff = m_calcTemplateDiff(refCurr, refStride, tarPatch, uiPatchWidth, uiPatchHeight, pDiff, TempType);
+#else
diff = m_calcTemplateDiff(refCurr, refStride, tarPatch, uiPatchWidth, uiPatchHeight, pDiff);
+#endif
if (diff < (pDiff))
{
insertNode(diff, iXOffset, iYOffset, pDiff, pX, pY, pId, setId);
@@ -719,7 +772,11 @@ void TrQuant::searchCandidateFromOnePicIntra(CodingUnit* pcCU, Pel** tarPatch,
for (iXOffset = mvXMax; iXOffset >= mvXMin; iXOffset--)
{
refCurr = ref + iYOffset * refStride + iXOffset;
+#if JVET_W0069_TMP_BOUNDARY
+ diff = m_calcTemplateDiff(refCurr, refStride, tarPatch, uiPatchWidth, uiPatchHeight, pDiff, TempType);
+#else
diff = m_calcTemplateDiff(refCurr, refStride, tarPatch, uiPatchWidth, uiPatchHeight, pDiff);
+#endif
if (diff < (pDiff))
{
@@ -780,12 +837,48 @@ bool TrQuant::generateTMPrediction(Pel* piPred, unsigned int uiStride, unsigned
return bSucceedFlag;
}
+#if JVET_W0069_TMP_BOUNDARY
+bool TrQuant::generateTM_DC_Prediction(Pel* piPred, unsigned int uiStride, unsigned int uiBlkWidth, unsigned int uiBlkHeight, int DC_Val)
+{
+ bool bSucceedFlag = true;
+ {
+ for (unsigned int uiY = 0; uiY < uiBlkHeight; uiY++)
+ {
+ for (unsigned int uiX = 0; uiX < uiBlkWidth; uiX++)
+ {
+ piPred[uiX] = DC_Val;
+ }
+ piPred += uiStride;
+ }
+ }
+ return bSucceedFlag;
+}
+#endif
+
+#if JVET_W0069_TMP_BOUNDARY
+int TrQuant::calcTemplateDiff(Pel* ref, unsigned int uiStride, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, int iMax, RefTemplateType TempType)
+#else
int TrQuant::calcTemplateDiff(Pel* ref, unsigned int uiStride, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, int iMax)
+#endif
{
int iDiffSum = 0;
+#if JVET_W0069_TMP_BOUNDARY
+ Pel* refPatchRow;
+ if (TempType == L_Shape_Template)
+ refPatchRow = ref - TMP_TEMPLATE_SIZE * uiStride - TMP_TEMPLATE_SIZE;
+ else if (TempType == Left_Template)
+ refPatchRow = ref - TMP_TEMPLATE_SIZE;
+ else if (TempType == Up_Template)
+ refPatchRow = ref - TMP_TEMPLATE_SIZE * uiStride;
+#else
Pel* refPatchRow = ref - TMP_TEMPLATE_SIZE * uiStride - TMP_TEMPLATE_SIZE;
+#endif
Pel* tarPatchRow;
+#if JVET_W0069_TMP_BOUNDARY
+ if (TempType == L_Shape_Template)
+ {
+#endif
// horizontal difference
for( int iY = 0; iY < TMP_TEMPLATE_SIZE; iY++ )
{
@@ -815,6 +908,43 @@ int TrQuant::calcTemplateDiff(Pel* ref, unsigned int uiStride, Pel** tarPatch, u
}
refPatchRow += uiStride;
}
+#if JVET_W0069_TMP_BOUNDARY
+ }
+ else if (TempType == Up_Template)
+ {
+ // top template difference
+ for (int iY = 0; iY < TMP_TEMPLATE_SIZE; iY++)
+ {
+ tarPatchRow = tarPatch[iY];
+ for (int iX = 0; iX < uiPatchWidth - TMP_TEMPLATE_SIZE; iX++)
+ {
+ iDiffSum += abs(refPatchRow[iX] - tarPatchRow[iX]);
+ }
+ if (iDiffSum > iMax) //for speeding up
+ {
+ return iDiffSum;
+ }
+ refPatchRow += uiStride;
+ }
+ }
+ else if (TempType == Left_Template)
+ {
+ // left template difference
+ for (int iY = TMP_TEMPLATE_SIZE; iY < uiPatchHeight; iY++)
+ {
+ tarPatchRow = tarPatch[iY];
+ for (int iX = 0; iX < TMP_TEMPLATE_SIZE; iX++)
+ {
+ iDiffSum += abs(refPatchRow[iX] - tarPatchRow[iX]);
+ }
+ if (iDiffSum > iMax) //for speeding up
+ {
+ return iDiffSum;
+ }
+ refPatchRow += uiStride;
+ }
+ }
+#endif
return iDiffSum;
}
diff --git a/source/Lib/CommonLib/TrQuant.h b/source/Lib/CommonLib/TrQuant.h
index 4145f64047924ac5623726cdac87aa9c71b32a10..ec9b42c35f72db6814722650e3c3749e222ec066 100644
--- a/source/Lib/CommonLib/TrQuant.h
+++ b/source/Lib/CommonLib/TrQuant.h
@@ -130,18 +130,35 @@ public:
void invLfnstNxN( int* src, int* dst, const uint32_t mode, const uint32_t index, const uint32_t size, int zeroOutSize );
#endif
#if JVET_V0130_INTRA_TMP
+#if JVET_W0069_TMP_BOUNDARY
+ int (*m_calcTemplateDiff)(Pel* ref, unsigned int uiStride, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, int iMax, RefTemplateType TempType);
+ static int calcTemplateDiff(Pel* ref, unsigned int uiStride, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, int iMax, RefTemplateType TempType);
+#else
int ( *m_calcTemplateDiff )(Pel* ref, unsigned int uiStride, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, int iMax);
static int calcTemplateDiff(Pel* ref, unsigned int uiStride, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, int iMax);
+#endif
Pel** getTargetPatch(unsigned int uiDepth) { return m_pppTarPatch[uiDepth]; }
Pel* getRefPicUsed() { return m_refPicUsed; }
void setRefPicUsed(Pel* ref) { m_refPicUsed = ref; }
unsigned int getStride() { return m_uiPicStride; }
void setStride(unsigned int uiPicStride) { m_uiPicStride = uiPicStride; }
+#if JVET_W0069_TMP_BOUNDARY
+ void searchCandidateFromOnePicIntra(CodingUnit* pcCU, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, unsigned int setId, RefTemplateType TempType);
+ void candidateSearchIntra(CodingUnit* pcCU, unsigned int uiBlkWidth, unsigned int uiBlkHeight, RefTemplateType TempType);
+#else
void searchCandidateFromOnePicIntra(CodingUnit* pcCU, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, unsigned int setId);
void candidateSearchIntra(CodingUnit* pcCU, unsigned int uiBlkWidth, unsigned int uiBlkHeight);
+#endif
bool generateTMPrediction(Pel* piPred, unsigned int uiStride, unsigned int uiBlkWidth, unsigned int uiBlkHeight, int& foundCandiNum);
+#if JVET_W0069_TMP_BOUNDARY
+ bool generateTM_DC_Prediction(Pel* piPred, unsigned int uiStride, unsigned int uiBlkWidth, unsigned int uiBlkHeight, int DC_Val);
+#endif
+#if JVET_W0069_TMP_BOUNDARY
+ void getTargetTemplate(CodingUnit* pcCU, unsigned int uiBlkWidth, unsigned int uiBlkHeight, RefTemplateType TempType);
+#else
void getTargetTemplate(CodingUnit* pcCU, unsigned int uiBlkWidth, unsigned int uiBlkHeight);
+#endif
#endif
uint32_t getLFNSTIntraMode( int wideAngPredMode );
diff --git a/source/Lib/CommonLib/TypeDef.h b/source/Lib/CommonLib/TypeDef.h
index 9762eb814ab022abc8a4301133a8cde2e110baa6..8f49a782e0cc37885e23a1267b5f4e7236210fbf 100644
--- a/source/Lib/CommonLib/TypeDef.h
+++ b/source/Lib/CommonLib/TypeDef.h
@@ -111,6 +111,7 @@
#define ENABLE_DIMD 1 // Decoder side intra mode derivation
#define JVET_V0087_DIMD_NO_ISP ENABLE_DIMD // disallow combination of DIMD and ISP
#define JVET_V0130_INTRA_TMP 1 // JVET-V0130: template matching prediction
+#define JVET_W0069_TMP_BOUNDARY 1
#define JVET_W0123_TIMD_FUSION 1 // Template based intra mode derivation and fusion
@@ -656,6 +657,15 @@ enum ChannelType
CHANNEL_TYPE_CHROMA = 1,
MAX_NUM_CHANNEL_TYPE = 2
};
+#if JVET_W0069_TMP_BOUNDARY
+enum RefTemplateType
+{
+ L_Shape_Template = 1,
+ Left_Template = 2,
+ Up_Template = 3,
+ No_Template = 4,
+};
+#endif
#if !INTRA_RM_SMALL_BLOCK_SIZE_CONSTRAINTS
enum TreeType
{
diff --git a/source/Lib/CommonLib/x86/TrQuantX86.h b/source/Lib/CommonLib/x86/TrQuantX86.h
index 7404f8fdc78506b73bbcaca8c743dda9447e090e..c560f9fbb46da8ee56874d6a3705e54df53f7b35 100644
--- a/source/Lib/CommonLib/x86/TrQuantX86.h
+++ b/source/Lib/CommonLib/x86/TrQuantX86.h
@@ -412,15 +412,33 @@ uint32_t computeSAD_SIMD( const Pel* ref, const Pel* cur, const int size )
#if ENABLE_SIMD_TMP
template< X86_VEXT vext >
+#if JVET_W0069_TMP_BOUNDARY
+int calcTemplateDiffSIMD(Pel* ref, unsigned int uiStride, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, int iMax, RefTemplateType TempType)
+#else
int calcTemplateDiffSIMD( Pel* ref, unsigned int uiStride, Pel** tarPatch, unsigned int uiPatchWidth, unsigned int uiPatchHeight, int iMax )
+#endif
{
int iDiffSum = 0;
int iY;
+#if JVET_W0069_TMP_BOUNDARY
+ Pel* refPatchRow;
+ if (TempType == L_Shape_Template)
+ refPatchRow = ref - TMP_TEMPLATE_SIZE * uiStride - TMP_TEMPLATE_SIZE;
+ else if (TempType == Left_Template)
+ refPatchRow = ref - TMP_TEMPLATE_SIZE;
+ else if (TempType == Up_Template)
+ refPatchRow = ref - TMP_TEMPLATE_SIZE * uiStride;
+#else
Pel* refPatchRow = ref - TMP_TEMPLATE_SIZE * uiStride - TMP_TEMPLATE_SIZE;
+#endif
Pel* tarPatchRow;
uint32_t uiSum;
// horizontal difference
+#if JVET_W0069_TMP_BOUNDARY
+ if (TempType == L_Shape_Template)
+ {
+#endif
for( iY = 0; iY < TMP_TEMPLATE_SIZE; iY++ )
{
tarPatchRow = tarPatch[iY];
@@ -527,6 +545,124 @@ int calcTemplateDiffSIMD( Pel* ref, unsigned int uiStride, Pel** tarPatch, unsig
// update location
refPatchRow += uiStride;
}
+#if JVET_W0069_TMP_BOUNDARY
+ }
+ else if (TempType == Up_Template)
+ {
+ // horizontal difference
+ for (iY = 0; iY < TMP_TEMPLATE_SIZE; iY++)
+ {
+ tarPatchRow = tarPatch[iY];
+ const short* pSrc1 = (const short*)tarPatchRow;
+ const short* pSrc2 = (const short*)refPatchRow;
+
+ // SIMD difference
+ //int iRows = uiPatchHeight;
+ int iCols = uiPatchWidth - TMP_TEMPLATE_SIZE;
+ if ((iCols & 7) == 0)
+ {
+ // Do with step of 8
+ __m128i vzero = _mm_setzero_si128();
+ __m128i vsum32 = vzero;
+ //for (int iY = 0; iY < iRows; iY += iSubStep)
+ {
+ __m128i vsum16 = vzero;
+ for (int iX = 0; iX < iCols; iX += 8)
+ {
+ __m128i vsrc1 = _mm_loadu_si128((const __m128i*)(&pSrc1[iX]));
+ __m128i vsrc2 = _mm_lddqu_si128((const __m128i*)(&pSrc2[iX]));
+ vsum16 = _mm_add_epi16(vsum16, _mm_abs_epi16(_mm_sub_epi16(vsrc1, vsrc2)));
+ }
+ __m128i vsumtemp = _mm_add_epi32(_mm_unpacklo_epi16(vsum16, vzero), _mm_unpackhi_epi16(vsum16, vzero));
+ vsum32 = _mm_add_epi32(vsum32, vsumtemp);
+ //pSrc1 += iStrideSrc1;
+ //pSrc2 += iStrideSrc2;
+ }
+ vsum32 = _mm_add_epi32(vsum32, _mm_shuffle_epi32(vsum32, 0x4e)); // 01001110
+ vsum32 = _mm_add_epi32(vsum32, _mm_shuffle_epi32(vsum32, 0xb1)); // 10110001
+ uiSum = _mm_cvtsi128_si32(vsum32);
+ }
+ else
+ {
+ // Do with step of 4
+ __m128i vzero = _mm_setzero_si128();
+ __m128i vsum32 = vzero;
+ //for (int iY = 0; iY < iRows; iY += iSubStep)
+ {
+ __m128i vsum16 = vzero;
+ for (int iX = 0; iX < iCols; iX += 4)
+ {
+ __m128i vsrc1 = _mm_loadl_epi64((const __m128i*) & pSrc1[iX]);
+ __m128i vsrc2 = _mm_loadl_epi64((const __m128i*) & pSrc2[iX]);
+ vsum16 = _mm_add_epi16(vsum16, _mm_abs_epi16(_mm_sub_epi16(vsrc1, vsrc2)));
+ }
+ __m128i vsumtemp = _mm_add_epi32(_mm_unpacklo_epi16(vsum16, vzero), _mm_unpackhi_epi16(vsum16, vzero));
+ vsum32 = _mm_add_epi32(vsum32, vsumtemp);
+ //pSrc1 += iStrideSrc1;
+ //pSrc2 += iStrideSrc2;
+ }
+ vsum32 = _mm_add_epi32(vsum32, _mm_shuffle_epi32(vsum32, 0x4e)); // 01001110
+ vsum32 = _mm_add_epi32(vsum32, _mm_shuffle_epi32(vsum32, 0xb1)); // 10110001
+ uiSum = _mm_cvtsi128_si32(vsum32);
+ }
+ iDiffSum += uiSum;
+
+ if (iDiffSum > iMax) //for speeding up
+ {
+ return iDiffSum;
+ }
+ // update location
+ refPatchRow += uiStride;
+ }
+
+
+ }
+ else if (TempType == Left_Template)
+ {
+
+ // vertical difference
+ int iCols = TMP_TEMPLATE_SIZE;
+
+ for (iY = TMP_TEMPLATE_SIZE; iY < uiPatchHeight; iY++)
+ {
+ tarPatchRow = tarPatch[iY];
+ const short* pSrc1 = (const short*)tarPatchRow;
+ const short* pSrc2 = (const short*)refPatchRow;
+
+ // SIMD difference
+
+ // Do with step of 4
+ __m128i vzero = _mm_setzero_si128();
+ __m128i vsum32 = vzero;
+ //for (int iY = 0; iY < iRows; iY += iSubStep)
+ {
+ __m128i vsum16 = vzero;
+ for (int iX = 0; iX < iCols; iX += 4)
+ {
+ __m128i vsrc1 = _mm_loadl_epi64((const __m128i*) & pSrc1[iX]);
+ __m128i vsrc2 = _mm_loadl_epi64((const __m128i*) & pSrc2[iX]);
+ vsum16 = _mm_add_epi16(vsum16, _mm_abs_epi16(_mm_sub_epi16(vsrc1, vsrc2)));
+ }
+ __m128i vsumtemp = _mm_add_epi32(_mm_unpacklo_epi16(vsum16, vzero), _mm_unpackhi_epi16(vsum16, vzero));
+ vsum32 = _mm_add_epi32(vsum32, vsumtemp);
+ //pSrc1 += iStrideSrc1;
+ //pSrc2 += iStrideSrc2;
+ }
+ vsum32 = _mm_add_epi32(vsum32, _mm_shuffle_epi32(vsum32, 0x4e)); // 01001110
+ vsum32 = _mm_add_epi32(vsum32, _mm_shuffle_epi32(vsum32, 0xb1)); // 10110001
+ uiSum = _mm_cvtsi128_si32(vsum32);
+
+ iDiffSum += uiSum;
+
+ if (iDiffSum > iMax) //for speeding up
+ {
+ return iDiffSum;
+ }
+ // update location
+ refPatchRow += uiStride;
+ }
+ }
+#endif
return iDiffSum;
}
diff --git a/source/Lib/DecoderLib/DecCu.cpp b/source/Lib/DecoderLib/DecCu.cpp
index 59f5ca00eae3380566c8f06d1968edd7244ea708..27aa5ae24fb9cdd8b59ab5c8f01dd590b2266ef3 100644
--- a/source/Lib/DecoderLib/DecCu.cpp
+++ b/source/Lib/DecoderLib/DecCu.cpp
@@ -384,9 +384,24 @@ void DecCu::xIntraRecBlk( TransformUnit& tu, const ComponentID compID )
if (PU::isTmp(pu, chType))
{
int foundCandiNum;
+#if JVET_W0069_TMP_BOUNDARY
+ RefTemplateType TempType = m_pcIntraPred->GetRefTemplateType(*(tu.cu), tu.cu->blocks[COMPONENT_Y]);
+ if (TempType != No_Template)
+ {
+ m_pcTrQuant->getTargetTemplate(tu.cu, pu.lwidth(), pu.lheight(), TempType);
+ m_pcTrQuant->candidateSearchIntra(tu.cu, pu.lwidth(), pu.lheight(), TempType);
+ m_pcTrQuant->generateTMPrediction(piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), foundCandiNum);
+ }
+ else
+ {
+ foundCandiNum = 1;
+ m_pcTrQuant->generateTM_DC_Prediction(piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), 1 << (tu.cu->cs->sps->getBitDepth(CHANNEL_TYPE_LUMA) - 1));
+ }
+#else
m_pcTrQuant->getTargetTemplate(tu.cu, pu.lwidth(), pu.lheight());
m_pcTrQuant->candidateSearchIntra(tu.cu, pu.lwidth(), pu.lheight());
m_pcTrQuant->generateTMPrediction(piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), foundCandiNum);
+#endif
assert(foundCandiNum >= 1);
}
else if (PU::isMIP(pu, chType))
@@ -585,7 +600,7 @@ void DecCu::xIntraRecACTBlk(TransformUnit& tu)
PelBuf piPred = cs.getPredBuf(area);
m_pcIntraPred->initIntraPatternChType(*tu.cu, area);
-#if JVET_V0130_INTRA_TMP
+#if JVET_V0130_INTRA_TMP && ! JVET_W0069_TMP_BOUNDARY
if (PU::isTmp(pu, chType))
{
int foundCandiNum;
diff --git a/source/Lib/EncoderLib/IntraSearch.cpp b/source/Lib/EncoderLib/IntraSearch.cpp
index 821eca75970bf218bee449080e33d97af1dbdc84..ac3df4bf2bcbfc44388c7ae22ecab14969c5c18b 100644
--- a/source/Lib/EncoderLib/IntraSearch.cpp
+++ b/source/Lib/EncoderLib/IntraSearch.cpp
@@ -891,12 +891,30 @@ bool IntraSearch::estIntraPredLumaQT(CodingUnit &cu, Partitioner &partitioner, c
bool bsuccessfull = 0;
CodingUnit cu_cpy = cu;
+#if JVET_W0069_TMP_BOUNDARY
+ RefTemplateType TemplateType = GetRefTemplateType(cu_cpy, cu_cpy.blocks[COMPONENT_Y]);
+ if (TemplateType != No_Template)
+#else
if( isRefTemplateAvailable( cu_cpy, cu_cpy.blocks[COMPONENT_Y] ) )
+#endif
{
+#if JVET_W0069_TMP_BOUNDARY
+ m_pcTrQuant->getTargetTemplate(&cu_cpy, pu.lwidth(), pu.lheight(), TemplateType);
+ m_pcTrQuant->candidateSearchIntra(&cu_cpy, pu.lwidth(), pu.lheight(), TemplateType);
+ bsuccessfull = m_pcTrQuant->generateTMPrediction(piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), foundCandiNum);
+#else
m_pcTrQuant->getTargetTemplate( &cu_cpy, pu.lwidth(), pu.lheight() );
m_pcTrQuant->candidateSearchIntra( &cu_cpy, pu.lwidth(), pu.lheight() );
bsuccessfull = m_pcTrQuant->generateTMPrediction( piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), foundCandiNum );
+#endif
}
+#if JVET_W0069_TMP_BOUNDARY
+ else
+ {
+ foundCandiNum = 1;
+ bsuccessfull = m_pcTrQuant->generateTM_DC_Prediction(piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), 1 << (cu_cpy.cs->sps->getBitDepth(CHANNEL_TYPE_LUMA) - 1));
+ }
+#endif
if( bsuccessfull && foundCandiNum >= 1 )
{
@@ -3660,9 +3678,24 @@ void IntraSearch::xIntraCodingTUBlock(TransformUnit &tu, const ComponentID &comp
if( PU::isTmp( pu, chType ) )
{
int foundCandiNum;
+#if JVET_W0069_TMP_BOUNDARY
+ RefTemplateType TempType = GetRefTemplateType(*(tu.cu), tu.cu->blocks[COMPONENT_Y]);
+ if (TempType != No_Template)
+ {
+ m_pcTrQuant->getTargetTemplate(tu.cu, pu.lwidth(), pu.lheight(), TempType);
+ m_pcTrQuant->candidateSearchIntra(tu.cu, pu.lwidth(), pu.lheight(), TempType);
+ m_pcTrQuant->generateTMPrediction(piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), foundCandiNum);
+ }
+ else
+ {
+ foundCandiNum = 1;
+ m_pcTrQuant->generateTM_DC_Prediction(piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), 1 << (tu.cu->cs->sps->getBitDepth(CHANNEL_TYPE_LUMA) - 1));
+ }
+#else
m_pcTrQuant->getTargetTemplate( tu.cu, pu.lwidth(), pu.lheight() );
m_pcTrQuant->candidateSearchIntra( tu.cu, pu.lwidth(), pu.lheight() );
m_pcTrQuant->generateTMPrediction( piPred.buf, piPred.stride, pu.lwidth(), pu.lheight(), foundCandiNum );
+#endif
CHECK( foundCandiNum < 1, "" );
}
else if( PU::isMIP( pu, chType ) )
@@ -4922,7 +4955,7 @@ bool IntraSearch::xRecurIntraCodingACTQT(CodingStructure &cs, Partitioner &parti
PelBuf piResi = resiBuf.bufs[compID];
initIntraPatternChType(*tu.cu, area);
-#if JVET_V0130_INTRA_TMP
+#if JVET_V0130_INTRA_TMP && !JVET_W0069_TMP_BOUNDARY
if( PU::isTmp( pu, chType ) )
{
int foundCandiNum;