diff --git a/source/Lib/CommonLib/InterPrediction.cpp b/source/Lib/CommonLib/InterPrediction.cpp
index 22dcc819e57e47ba9e5c6389a9eeaffa32463e78..b0baa9c197e4caa6a9811c97d9005060c9f1a27d 100644
--- a/source/Lib/CommonLib/InterPrediction.cpp
+++ b/source/Lib/CommonLib/InterPrediction.cpp
@@ -915,7 +915,11 @@ void InterPrediction::xPredAffineBlk( const ComponentID& compID, const Predictio
#if JVET_P0653_BDOF_PROF_PARA_DEV
const int mvShift = 8;
+#if JVET_P0491_BDOFPROF_MVD_RANGE
+ const int dmvLimit = ( 1 << 5 ) - 1;
+#else
const int dmvLimit = (1 << 5);
+#endif
#else
#if JVET_P0057_BDOF_PROF_HARMONIZATION
const int mvShift = shift + MV_FRACTIONAL_BITS_INTERNAL + 2 - std::max<int>(5, clpRng.bd - 7);
@@ -935,8 +939,13 @@ void InterPrediction::xPredAffineBlk( const ComponentID& compID, const Predictio
#else
roundAffineMv(dMvScaleHor[idx], dMvScaleVer[idx], shift);
#endif
+#if JVET_P0491_BDOFPROF_MVD_RANGE
+ dMvScaleHor[idx] = Clip3( -dmvLimit, dmvLimit, dMvScaleHor[idx] );
+ dMvScaleVer[idx] = Clip3( -dmvLimit, dmvLimit, dMvScaleVer[idx] );
+#else
dMvScaleHor[idx] = Clip3(-dmvLimit, dmvLimit - 1, dMvScaleHor[idx]);
dMvScaleVer[idx] = Clip3(-dmvLimit, dmvLimit - 1, dMvScaleVer[idx]);
+#endif
}
}
else
@@ -1176,7 +1185,11 @@ void InterPrediction::applyBiOptFlow(const PredictionUnit &pu, const CPelUnitBuf
const int shiftNum = IF_INTERNAL_PREC + 1 - bitDepth;
const int offset = (1 << (shiftNum - 1)) + 2 * IF_INTERNAL_OFFS;
#if JVET_P0653_BDOF_PROF_PARA_DEV
+#if JVET_P0491_BDOFPROF_MVD_RANGE
+ const int limit = ( 1 << 5 ) - 1;
+#else
const int limit = (1 << 5);
+#endif
#else
const int limit = (1<<(std::max<int>(5, bitDepth - 7)));
#endif
@@ -1205,7 +1218,7 @@ void InterPrediction::applyBiOptFlow(const PredictionUnit &pu, const CPelUnitBuf
g_pelBufOP.calcBIOSums(SrcY0Tmp, SrcY1Tmp, pGradX0Tmp, pGradX1Tmp, pGradY0Tmp, pGradY1Tmp, xu, yu, src0Stride, src1Stride, widthG, bitDepth, &sumAbsGX, &sumAbsGY, &sumDIX, &sumDIY, &sumSignGY_GX);
tmpx = (sumAbsGX == 0 ? 0 : rightShiftMSB(sumDIX << 3, sumAbsGX));
-#if JVET_P0057_BDOF_PROF_HARMONIZATION
+#if JVET_P0057_BDOF_PROF_HARMONIZATION && !JVET_P0491_BDOFPROF_MVD_RANGE
tmpx = Clip3(-limit, limit - 1, tmpx);
#else
tmpx = Clip3(-limit, limit, tmpx);
@@ -1216,7 +1229,7 @@ void InterPrediction::applyBiOptFlow(const PredictionUnit &pu, const CPelUnitBuf
int tmpData = tmpx * mainsGxGy;
tmpData = ((tmpData << 12) + tmpx*secsGxGy) >> 1;
tmpy = (sumAbsGY == 0 ? 0 : rightShiftMSB(((sumDIY << 3) - tmpData), sumAbsGY));
-#if JVET_P0057_BDOF_PROF_HARMONIZATION
+#if JVET_P0057_BDOF_PROF_HARMONIZATION && !JVET_P0491_BDOFPROF_MVD_RANGE
tmpy = Clip3(-limit, limit - 1, tmpy);
#else
tmpy = Clip3(-limit, limit, tmpy);
@@ -1364,7 +1377,11 @@ void InterPrediction::xApplyBiPROF(const PredictionUnit &pu, const CPelBuf& pcYu
const int bit = MAX_CU_DEPTH;
#if JVET_P0653_BDOF_PROF_PARA_DEV
const int mvShift = 8;
+#if JVET_P0491_BDOFPROF_MVD_RANGE
+ const int dmvLimit = ( 1 << 5 ) - 1;
+#else
const int dmvLimit = (1 << 5);
+#endif
#else
const int shift = bit - 4 + MV_FRACTIONAL_BITS_INTERNAL;
#if JVET_P0057_BDOF_PROF_HARMONIZATION
@@ -1439,8 +1456,13 @@ void InterPrediction::xApplyBiPROF(const PredictionUnit &pu, const CPelBuf& pcYu
#else
roundAffineMv(dMvScaleHor[idx], dMvScaleVer[idx], shift);
#endif
+#if JVET_P0491_BDOFPROF_MVD_RANGE
+ dMvScaleHor[idx] = Clip3( -dmvLimit, dmvLimit, dMvScaleHor[idx] );
+ dMvScaleVer[idx] = Clip3( -dmvLimit, dmvLimit, dMvScaleVer[idx] );
+#else
dMvScaleHor[idx] = Clip3(-dmvLimit, dmvLimit - 1, dMvScaleHor[idx]);
dMvScaleVer[idx] = Clip3(-dmvLimit, dmvLimit - 1, dMvScaleVer[idx]);
+#endif
}
}
else
diff --git a/source/Lib/CommonLib/TypeDef.h b/source/Lib/CommonLib/TypeDef.h
index bd2c51b8c510fc3e88bb4020770bccc033676103..841a7f0f43b1a53f68c3f889ef6c9a8702c9e4e7 100644
--- a/source/Lib/CommonLib/TypeDef.h
+++ b/source/Lib/CommonLib/TypeDef.h
@@ -50,6 +50,8 @@
#include <assert.h>
#include <cassert>
+#define JVET_P0491_BDOFPROF_MVD_RANGE 1 // JVET-P0491: clip the MVD in BDOF/PROF to [-31 31]
+
#define JVET_P0460_PLT_TS_MIN_QP 1 // JVET-P0460: Use TS min QP for Palette Escape mode
#define JVET_P1001_DEBLOCKING_CHROMAQP_FIX 1 //JVET-P1001/P1002: Align Chroma QP used in deblocking with the one used in Transform/invTransform
diff --git a/source/Lib/CommonLib/x86/BufferX86.h b/source/Lib/CommonLib/x86/BufferX86.h
index e6459f4a9033da5dc374c869ae01afad16c6339e..e9df4e45549e363603a883103993bb8a5981a023 100644
--- a/source/Lib/CommonLib/x86/BufferX86.h
+++ b/source/Lib/CommonLib/x86/BufferX86.h
@@ -529,7 +529,11 @@ void roundIntVector_SIMD(int* v, int size, unsigned int nShift, const int dmvLim
if (vext >= AVX512 && size >= 16)
{
__m512i dMvMin = _mm256_set1_epi32(-dmvLimit);
+#if JVET_P0491_BDOFPROF_MVD_RANGE
+ __m512i dMvMax = _mm256_set1_epi32( dmvLimit );
+#else
__m512i dMvMax = _mm256_set1_epi32(dmvLimit - 1 );
+#endif
__m512i nOffset = _mm512_set1_epi32((1 << (nShift - 1)));
__m512i vones = _mm512_set1_epi32(1);
__m512i vzero = _mm512_setzero_si512();
@@ -549,7 +553,11 @@ void roundIntVector_SIMD(int* v, int size, unsigned int nShift, const int dmvLim
if (vext >= AVX2 && size >= 8)
{
__m256i dMvMin = _mm256_set1_epi32(-dmvLimit);
+#if JVET_P0491_BDOFPROF_MVD_RANGE
+ __m256i dMvMax = _mm256_set1_epi32( dmvLimit );
+#else
__m256i dMvMax = _mm256_set1_epi32(dmvLimit - 1);
+#endif
__m256i nOffset = _mm256_set1_epi32(1 << (nShift - 1));
__m256i vzero = _mm256_setzero_si256();
for (int i = 0; i < size; i += 8, v += 8)
@@ -565,7 +573,11 @@ void roundIntVector_SIMD(int* v, int size, unsigned int nShift, const int dmvLim
#endif
{
__m128i dMvMin = _mm_set1_epi32(-dmvLimit);
+#if JVET_P0491_BDOFPROF_MVD_RANGE
+ __m128i dMvMax = _mm_set1_epi32( dmvLimit );
+#else
__m128i dMvMax = _mm_set1_epi32(dmvLimit - 1);
+#endif
__m128i nOffset = _mm_set1_epi32((1 << (nShift - 1)));
__m128i vzero = _mm_setzero_si128();
for (int i = 0; i < size; i += 4, v += 4)