ref: 316a3ba3a7f4770198c66830536474801e1fd5ed
parent: 73647654a1619809aa11b6261fda94405c3d0bf1
parent: 26d3346518c74e252a3c742f56e88d656ddbf94b
author: huili2 <[email protected]>
date: Tue Jan 15 11:51:12 EST 2019
Merge pull request #3080 from xiaotiansf/GeneralFix Update and fix temporal direct prediction and fix multi-reference pic…
--- a/codec/decoder/core/inc/mv_pred.h
+++ b/codec/decoder/core/inc/mv_pred.h
@@ -111,7 +111,8 @@
/*!
* \brief get the motion predictor for B-slice temporal direct mode 16x16
*/
-int32_t PredBDirectTemporal (PWelsDecoderContext pCtx, int16_t iMvp[LIST_A][2], int8_t ref[LIST_A]);
+int32_t PredBDirectTemporal (PWelsDecoderContext pCtx, int16_t iMvp[LIST_A][2], int8_t ref[LIST_A],
+ SubMbType& subMbType);
/*!
* \brief get the motion params for B-slice spatial direct mode
@@ -142,6 +143,39 @@
*/
void PredInter8x16Mv (int16_t iMotionVector[LIST_A][30][MV_A], int8_t iRefIndex[LIST_A][30],
int32_t listIdx, int32_t iPartIdx, int8_t iRef, int16_t iMVP[2]);
+
+/*!
+* \brief Fill the spatial direct motion vectors for 8x8 direct MB
+* \param
+* \param output motion vector cache and motion vector deviation cache
+*/
+void FillSpatialDirect8x8Mv (PDqLayer pCurLayer, const int16_t& iIdx8, const int8_t& iPartCount, const int8_t& iPartW,
+ const SubMbType& subMbType, const bool& bIsLongRef, int16_t pMvDirect[LIST_A][2], int8_t iRef[LIST_A],
+ int16_t pMotionVector[LIST_A][30][MV_A], int16_t pMvdCache[LIST_A][30][MV_A]);
+
+/*!
+* \brief Fill the temporal direct motion vectors for 8x8 direct MB
+* \param
+* \param output motion vector cache and motion vector deviation cache
+*/
+void FillTemporalDirect8x8Mv (PDqLayer pCurLayer, const int16_t& iIdx8, const int8_t& iPartCount, const int8_t& iPartW,
+ const SubMbType& subMbType, int8_t iRef[LIST_A], int16_t (*mvColoc)[2],
+ int16_t pMotionVector[LIST_A][30][MV_A], int16_t pMvdCache[LIST_A][30][MV_A]);
+
+/*!
+* \brief returns ref_index in List_0 from the colocated ref_index in LIST_0.
+* \param
+* returns ref_index in List_0 of ref picture LIST_0
+*/
+int8_t MapColToList0 (PWelsDecoderContext& pCtx, const int8_t& colocRefIndexL0,
+ const int32_t& ref0Count); //ISO/IEC 14496-10:2009(E) (8-193)
+
+/*!
+* \brief update ref_index cache for current MB, for 8x8
+* \param
+* \param
+*/
+void Update8x8RefIdx (PDqLayer& pCurDqLayer, const int16_t& iPartIdx, const int32_t& listIdx, const int8_t& iRef);
} // namespace WelsDec
--- a/codec/decoder/core/inc/parse_mb_syn_cabac.h
+++ b/codec/decoder/core/inc/parse_mb_syn_cabac.h
@@ -82,6 +82,8 @@
const int8_t iListIdx);
void UpdateP8x8DirectCabac (PDqLayer pCurDqLayer, int32_t iPartIdx);
void UpdateP16x16DirectCabac (PDqLayer pCurDqLayer);
+void UpdateP8x8RefCacheIdxCabac (int8_t pRefIndex[LIST_A][30], const int16_t& iPartIdx, const int32_t& listIdx,
+ const int8_t& iRef);
}
//#pragma pack()
#endif
--- a/codec/decoder/core/inc/slice.h
+++ b/codec/decoder/core/inc/slice.h
@@ -122,7 +122,7 @@
PPps pPps;
int32_t iSpsId;
int32_t iPpsId;
- bool bIdrFlag;
+ bool bIdrFlag;
/*********************got from other layer for efficency if possible*********************/
SRefPicListReorderSyn pRefPicListReordering; // Reference picture list reordering syntaxs
--- a/codec/decoder/core/src/decode_slice.cpp
+++ b/codec/decoder/core/src/decode_slice.cpp
@@ -54,12 +54,6 @@
extern void FreePicture (PPicture pPic, CMemoryAlign* pMa);
-static inline int32_t iAbs (int32_t x) {
- static const int32_t INT_BITS = (sizeof (int) * CHAR_BIT) - 1;
- int32_t y = x >> INT_BITS;
- return (x ^ y) - y;
-}
-
extern PPicture AllocPicture (PWelsDecoderContext pCtx, const int32_t kiPicWidth, const int32_t kiPicHeight);
int32_t WelsTargetSliceConstruction (PWelsDecoderContext pCtx) {
@@ -1417,11 +1411,10 @@
pCtx->bMbRefConcealed = pCtx->bRPLRError || pCtx->bMbRefConcealed || ! (ppRefPicL0[0] && ppRefPicL0[0]->bIsComplete)
|| ! (ppRefPicL1[0] && ppRefPicL1[0]->bIsComplete);
-
+ SubMbType subMbType;
if (pSliceHeader->iDirectSpatialMvPredFlag) {
//predict direct spatial mv
- SubMbType subMbType;
int32_t ret = PredMvBDirectSpatial (pCtx, pMv, ref, subMbType);
if (ret != ERR_NONE) {
return ret;
@@ -1428,8 +1421,7 @@
}
} else {
//temporal direct mode
- ComputeColocated (pCtx);
- int32_t ret = PredBDirectTemporal (pCtx, pMv, ref);
+ int32_t ret = PredBDirectTemporal (pCtx, pMv, ref, subMbType);
if (ret != ERR_NONE) {
return ret;
}
@@ -2380,10 +2372,10 @@
pCtx->bMbRefConcealed = pCtx->bRPLRError || pCtx->bMbRefConcealed || ! (ppRefPicL0[0] && ppRefPicL0[0]->bIsComplete)
|| ! (ppRefPicL1[0] && ppRefPicL1[0]->bIsComplete);
//predict iMv
+ SubMbType subMbType;
if (pSliceHeader->iDirectSpatialMvPredFlag) {
//predict direct spatial mv
- SubMbType subMbType;
int32_t ret = PredMvBDirectSpatial (pCtx, iMv, ref, subMbType);
if (ret != ERR_NONE) {
return ret;
@@ -2390,8 +2382,7 @@
}
} else {
//temporal direct mode
- ComputeColocated (pCtx);
- int32_t ret = PredBDirectTemporal (pCtx, iMv, ref);
+ int32_t ret = PredBDirectTemporal (pCtx, iMv, ref, subMbType);
if (ret != ERR_NONE) {
return ret;
}
@@ -2832,17 +2823,21 @@
PSlice pCurSlice = &pCurLayer->sLayerInfo.sSliceInLayer;
PSliceHeader pSliceHeader = &pCurSlice->sSliceHeaderExt.sSliceHeader;
if (!pSliceHeader->iDirectSpatialMvPredFlag) {
- uint32_t uiShortRefCount = pCtx->sRefPic.uiShortRefCount[LIST_0];
- for (int32_t listIdx = LIST_0; listIdx < LIST_A; ++listIdx) {
- for (uint32_t i = 0; i < uiShortRefCount; ++i) {
- int32_t iTRb = WELS_CLIP3 (-128, 127, pSliceHeader->iPicOrderCntLsb - pCtx->sRefPic.pRefList[listIdx][i]->iFramePoc);
- int32_t iTRp = WELS_CLIP3 (-128, 127,
- pCtx->sRefPic.pRefList[LIST_1][i]->iFramePoc - pCtx->sRefPic.pRefList[LIST_0][i]->iFramePoc);
- if (iTRp != 0) {
- int32_t prescale = (16384 + iAbs (iTRp / 2)) / iTRp;
- pCurSlice->iMvScale[listIdx][i] = WELS_CLIP3 (-1024, 1023, (iTRb * prescale + 32) >> 6);
- } else {
- pCurSlice->iMvScale[listIdx][i] = 0x03FFF;
+ uint32_t uiRefCount = pSliceHeader->uiRefCount[LIST_0];
+ if (pCtx->sRefPic.pRefList[LIST_1][0] != NULL) {
+ for (uint32_t i = 0; i < uiRefCount; ++i) {
+ if (pCtx->sRefPic.pRefList[LIST_0][i] != NULL) {
+ const int32_t poc0 = pCtx->sRefPic.pRefList[LIST_0][i]->iFramePoc;
+ const int32_t poc1 = pCtx->sRefPic.pRefList[LIST_1][0]->iFramePoc;
+ const int32_t poc = pSliceHeader->iPicOrderCntLsb;
+ const int32_t td = WELS_CLIP3 (poc1 - poc0, -128, 127);
+ if (td == 0) {
+ pCurSlice->iMvScale[LIST_0][i] = 1 << 8;
+ } else {
+ int32_t tb = WELS_CLIP3 (poc - poc0, -128, 127);
+ int32_t tx = (16384 + (abs (td) >> 1)) / td;
+ pCurSlice->iMvScale[LIST_0][i] = WELS_CLIP3 ((tb * tx + 32) >> 6, -1024, 1023);
+ }
}
}
}
--- a/codec/decoder/core/src/decoder_core.cpp
+++ b/codec/decoder/core/src/decoder_core.cpp
@@ -2550,6 +2550,9 @@
}
}
}
+ //calculate Colocated mv scaler factor for temporal direct prediction
+ if (pSh->eSliceType == B_SLICE && !pSh->iDirectSpatialMvPredFlag)
+ ComputeColocated (pCtx);
iRet = WelsDecodeSlice (pCtx, bFreshSliceAvailable, pNalCur);
@@ -2648,8 +2651,10 @@
memcpy (pCtx->pDec->pRefIndex[LIST_1], pCtx->pCurDqLayer->pRefIndex[LIST_1],
pCtx->sMb.iMbWidth * pCtx->sMb.iMbHeight * sizeof (int8_t) * MB_BLOCK4x4_NUM);
for (int32_t listIdx = LIST_0; listIdx < LIST_A; ++listIdx) {
- for (uint32_t i = 0; i < pCtx->sRefPic.uiRefCount[listIdx]; ++i) {
+ uint32_t i = 0;
+ while (pCtx->sRefPic.pRefList[listIdx][i]) {
pCtx->pDec->pRefPic[listIdx][i] = pCtx->sRefPic.pRefList[listIdx][i];
+ ++i;
}
}
iRet = WelsMarkAsRef (pCtx);
--- a/codec/decoder/core/src/manage_dec_ref.cpp
+++ b/codec/decoder/core/src/manage_dec_ref.cpp
@@ -346,8 +346,8 @@
for (int32_t listIdx = 0; listIdx < ListCount; ++listIdx) {
PPicture pPic = NULL;
PPicture* ppRefList = pCtx->sRefPic.pRefList[listIdx];
- int32_t iMaxRefIdx = pCtx->pSps->iNumRefFrames;
- int32_t iRefCount = pCtx->sRefPic.uiRefCount[listIdx];
+ int32_t iMaxRefIdx = pCtx->iPicQueueNumber;
+ int32_t iRefCount = pSliceHeader->uiRefCount[listIdx];
int32_t iPredFrameNum = pSliceHeader->iFrameNum;
int32_t iMaxPicNum = 1 << pSliceHeader->pSps->uiLog2MaxFrameNum;
int32_t iAbsDiffPicNum = -1;
--- a/codec/decoder/core/src/mv_pred.cpp
+++ b/codec/decoder/core/src/mv_pred.cpp
@@ -584,75 +584,7 @@
pSubPartCount[i] = 4;
pPartW[i] = 1;
}
-
- int8_t iPartCount = pSubPartCount[i];
- int16_t iPartIdx, iBlockW = pPartW[i];
-
- for (int32_t j = 0; j < iPartCount; j++) {
- iPartIdx = iIdx8 + j * iBlockW;
- uint8_t iScan4Idx = g_kuiScan4[iPartIdx];
- uint8_t iColocIdx = g_kuiScan4[iPartIdx];
- //uint8_t iCacheIdx = g_kuiCache30ScanIdx[iPartIdx];
-
- int16_t pMV[4] = { 0 };
- if (IS_SUB_8x8 (subMbType)) {
- * (uint32_t*)pMV = * (uint32_t*)iMvp[LIST_0];
- ST32 ((pMV + 2), LD32 (pMV));
- ST64 (pCurLayer->pMv[LIST_0][iMbXy][iScan4Idx], LD64 (pMV));
- ST64 (pCurLayer->pMv[LIST_0][iMbXy][iScan4Idx + 4], LD64 (pMV));
- ST64 (pCurLayer->pMvd[LIST_0][iMbXy][iScan4Idx], 0);
- ST64 (pCurLayer->pMvd[LIST_0][iMbXy][iScan4Idx + 4], 0);
- * (uint32_t*)pMV = * (uint32_t*)iMvp[LIST_1];
- ST32 ((pMV + 2), LD32 (pMV));
- ST64 (pCurLayer->pMv[LIST_1][iMbXy][iScan4Idx], LD64 (pMV));
- ST64 (pCurLayer->pMv[LIST_1][iMbXy][iScan4Idx + 4], LD64 (pMV));
- ST64 (pCurLayer->pMvd[LIST_1][iMbXy][iScan4Idx], 0);
- ST64 (pCurLayer->pMvd[LIST_1][iMbXy][iScan4Idx + 4], 0);
- } else { //SUB_4x4
- * (uint32_t*)pMV = * (uint32_t*)iMvp[LIST_0];
- ST32 (pCurLayer->pMv[LIST_0][iMbXy][iScan4Idx], LD32 (pMV));
- ST32 (pCurLayer->pMvd[LIST_0][iMbXy][iScan4Idx], 0);
- * (uint32_t*)pMV = * (uint32_t*)iMvp[LIST_1];
- ST32 (pCurLayer->pMv[LIST_1][iMbXy][iScan4Idx], LD32 (pMV));
- ST32 (pCurLayer->pMvd[LIST_1][iMbXy][iScan4Idx], 0);
- }
- if ((* (int32_t*)iMvp[LIST_0] | * (int32_t*)iMvp[LIST_1])) {
- uint32_t uiColZeroFlag = (0 == pCurLayer->iColocIntra[iColocIdx]) && !bIsLongRef &&
- (pCurLayer->iColocRefIndex[LIST_0][iColocIdx] == 0 || (pCurLayer->iColocRefIndex[LIST_0][iColocIdx] < 0
- && pCurLayer->iColocRefIndex[LIST_1][iColocIdx] == 0));
- const int16_t (*mvColoc)[2] = 0 == pCurLayer->iColocRefIndex[LIST_0][iColocIdx] ? pCurLayer->iColocMv[LIST_0] :
- pCurLayer->iColocMv[LIST_1];
- const int16_t* mv = mvColoc[iColocIdx];
- if (IS_SUB_8x8 (subMbType)) {
- if (uiColZeroFlag && ((unsigned) (mv[0] + 1) <= 2 && (unsigned) (mv[1] + 1) <= 2)) {
- if (ref[LIST_0] == 0) {
- ST64 (pCurLayer->pMv[LIST_0][iMbXy][iScan4Idx], 0);
- ST64 (pCurLayer->pMv[LIST_0][iMbXy][iScan4Idx + 4], 0);
- ST64 (pCurLayer->pMvd[LIST_0][iMbXy][iScan4Idx], 0);
- ST64 (pCurLayer->pMvd[LIST_0][iMbXy][iScan4Idx + 4], 0);
- }
-
- if (ref[LIST_1] == 0) {
- ST64 (pCurLayer->pMv[LIST_1][iMbXy][iScan4Idx], 0);
- ST64 (pCurLayer->pMv[LIST_1][iMbXy][iScan4Idx + 4], 0);
- ST64 (pCurLayer->pMvd[LIST_1][iMbXy][iScan4Idx], 0);
- ST64 (pCurLayer->pMvd[LIST_1][iMbXy][iScan4Idx + 4], 0);
- }
- }
- } else {
- if (uiColZeroFlag && ((unsigned) (mv[0] + 1) <= 2 && (unsigned) (mv[1] + 1) <= 2)) {
- if (ref[LIST_0] == 0) {
- ST32 (pCurLayer->pMv[LIST_0][iMbXy][iScan4Idx], 0);
- ST32 (pCurLayer->pMvd[LIST_0][iMbXy][iScan4Idx], 0);
- }
- if (ref[LIST_1] == 0) {
- ST32 (pCurLayer->pMv[LIST_1][iMbXy][iScan4Idx], 0);
- ST32 (pCurLayer->pMvd[LIST_1][iMbXy][iScan4Idx], 0);
- }
- }
- }
- }
- }
+ FillSpatialDirect8x8Mv (pCurLayer, iIdx8, pSubPartCount[i], pPartW[i], subMbType, bIsLongRef, iMvp, ref, NULL, NULL);
}
}
}
@@ -659,21 +591,29 @@
return ret;
}
-int32_t PredBDirectTemporal (PWelsDecoderContext pCtx, int16_t iMvp[LIST_A][2], int8_t ref[LIST_A]) {
+int32_t PredBDirectTemporal (PWelsDecoderContext pCtx, int16_t iMvp[LIST_A][2], int8_t ref[LIST_A],
+ SubMbType& subMbType) {
int32_t ret = ERR_NONE;
PDqLayer pCurLayer = pCtx->pCurDqLayer;
int32_t iMbXy = pCurLayer->iMbXyIndex;
bool bSkipOrDirect = (IS_SKIP (pCurLayer->pMbType[iMbXy]) | IS_DIRECT (pCurLayer->pMbType[iMbXy])) > 0;
+
MbType mbType;
- SubMbType subMbType;
ret = GetColocatedMb (pCtx, mbType, subMbType);
if (ret != ERR_NONE) {
return ret;
}
+
+ pCurLayer->pMbType[iMbXy] = mbType;
+
PSlice pSlice = &pCurLayer->sLayerInfo.sSliceInLayer;
+ PSliceHeader pSliceHeader = &pSlice->sSliceHeaderExt.sSliceHeader;
+ int16_t pMvd[4] = { 0 };
+ const int32_t ref0Count = WELS_MIN (pSliceHeader->uiRefCount[LIST_0], pCtx->sRefPic.uiRefCount[LIST_0]);
if (IS_INTER_16x16 (mbType)) {
ref[LIST_0] = 0;
ref[LIST_1] = 0;
+ UpdateP16x16DirectCabac (pCurLayer);
UpdateP16x16RefIdx (pCurLayer, LIST_1, ref[LIST_1]);
ST64 (iMvp, 0);
if (pCurLayer->iColocIntra[0]) {
@@ -681,10 +621,14 @@
UpdateP16x16MotionOnly (pCurLayer, LIST_1, iMvp[LIST_1]);
UpdateP16x16RefIdx (pCurLayer, LIST_0, ref[LIST_0]);
} else {
- ref[LIST_0] = pCurLayer->iColocRefIndex[LIST_0][0] >= 0 ? pCurLayer->iColocRefIndex[LIST_0][0] :
- pCurLayer->iColocRefIndex[LIST_1][0];
- const int16_t (*mvColoc)[2] = 0 == ref[LIST_0] ? pCurLayer->iColocMv[LIST_0] : pCurLayer->iColocMv[LIST_1];
- const int16_t* mv = mvColoc[0];
+ ref[LIST_0] = 0;
+ int16_t* mv = pCurLayer->iColocMv[LIST_0][0];
+ int8_t colocRefIndexL0 = pCurLayer->iColocRefIndex[LIST_0][0];
+ if (colocRefIndexL0 >= 0) {
+ ref[LIST_0] = MapColToList0 (pCtx, colocRefIndexL0, ref0Count);
+ } else {
+ mv = pCurLayer->iColocMv[LIST_1][0];
+ }
UpdateP16x16RefIdx (pCurLayer, LIST_0, ref[LIST_0]);
iMvp[LIST_0][0] = (pSlice->iMvScale[LIST_0][ref[LIST_0]] * mv[0] + 128) >> 8;
@@ -694,26 +638,35 @@
iMvp[LIST_1][1] = iMvp[LIST_0][1] - mv[1];
UpdateP16x16MotionOnly (pCurLayer, LIST_1, iMvp[LIST_1]);
}
+ UpdateP16x16MvdCabac (pCurLayer, pMvd, LIST_0);
+ UpdateP16x16MvdCabac (pCurLayer, pMvd, LIST_1);
} else {
if (bSkipOrDirect) {
int8_t pSubPartCount[4], pPartW[4];
+ int8_t pRefIndex[LIST_A][30];
for (int32_t i = 0; i < 4; i++) {
int16_t iIdx8 = i << 2;
+ const uint8_t iScan4Idx = g_kuiScan4[iIdx8];
pCurLayer->pSubMbType[iMbXy][i] = subMbType;
+ int16_t (*mvColoc)[2] = pCurLayer->iColocMv[LIST_0];
+
ref[LIST_1] = 0;
- if (pCurLayer->iColocIntra[g_kuiScan4[iIdx8]]) {
+ UpdateP8x8RefIdxCabac (pCurLayer, pRefIndex, iIdx8, ref[LIST_1], LIST_1);
+ if (pCurLayer->iColocIntra[iScan4Idx]) {
ref[LIST_0] = 0;
+ UpdateP8x8RefIdxCabac (pCurLayer, pRefIndex, iIdx8, ref[LIST_0], LIST_0);
+ ST64 (iMvp, 0);
} else {
- if (pCurLayer->iColocRefIndex[LIST_0][iIdx8] >= 0) {
- ref[LIST_0] = pCurLayer->iColocRefIndex[LIST_0][iIdx8];
+ ref[LIST_0] = 0;
+ int8_t colocRefIndexL0 = pCurLayer->iColocRefIndex[LIST_0][iScan4Idx];
+ if (colocRefIndexL0 >= 0) {
+ ref[LIST_0] = MapColToList0 (pCtx, colocRefIndexL0, ref0Count);
} else {
- ref[LIST_0] = pCurLayer->iColocRefIndex[LIST_1][iIdx8];
+ mvColoc = pCurLayer->iColocMv[LIST_1];
}
+ UpdateP8x8RefIdxCabac (pCurLayer, pRefIndex, iIdx8, ref[LIST_0], LIST_0);
}
- int8_t pRefIndex[LIST_A][30];
- UpdateP8x8RefIdxCabac (pCurLayer, pRefIndex, iIdx8, ref[LIST_0], LIST_0);
- UpdateP8x8RefIdxCabac (pCurLayer, pRefIndex, iIdx8, ref[LIST_1], LIST_1);
UpdateP8x8DirectCabac (pCurLayer, iIdx8);
pSubPartCount[i] = g_ksInterBSubMbTypeInfo[0].iPartCount;
@@ -723,46 +676,7 @@
pSubPartCount[i] = 4;
pPartW[i] = 1;
}
-
- int8_t iPartCount = pSubPartCount[i];
- int16_t iPartIdx, iBlockW = pPartW[i];
- for (int32_t j = 0; j < iPartCount; j++) {
- iPartIdx = iIdx8 + j * iBlockW;
- uint8_t iScan4Idx = g_kuiScan4[iPartIdx];
- uint8_t iColocIdx = g_kuiScan4[iPartIdx];
-
- int16_t (*mvColoc)[2] = pCurLayer->iColocMv[LIST_0];
- int16_t* mv = mvColoc[iColocIdx];
-
- int16_t pMV[4] = { 0 };
- if (IS_SUB_8x8 (subMbType)) {
- iMvp[LIST_0][0] = (pSlice->iMvScale[LIST_0][ref[LIST_0]] * mv[0] + 128) >> 8;
- iMvp[LIST_0][1] = (pSlice->iMvScale[LIST_0][ref[LIST_0]] * mv[1] + 128) >> 8;
- ST32 (pMV, LD32 (iMvp[LIST_0]));
- ST32 ((pMV + 2), LD32 (iMvp[LIST_0]));
- ST64 (pCurLayer->pMv[LIST_0][iMbXy][iScan4Idx], LD64 (pMV));
- ST64 (pCurLayer->pMv[LIST_0][iMbXy][iScan4Idx + 4], LD64 (pMV));
- ST64 (pCurLayer->pMvd[LIST_0][iMbXy][iScan4Idx], 0);
- ST64 (pCurLayer->pMvd[LIST_0][iMbXy][iScan4Idx + 4], 0);
- iMvp[LIST_1][0] -= iMvp[LIST_0][0] - mv[0];
- iMvp[LIST_1][1] -= iMvp[LIST_0][0] - mv[1];
- ST32 (pMV, LD32 (iMvp[LIST_1]));
- ST32 ((pMV + 2), LD32 (iMvp[LIST_1]));
- ST64 (pCurLayer->pMv[LIST_1][iMbXy][iScan4Idx], LD64 (pMV));
- ST64 (pCurLayer->pMv[LIST_1][iMbXy][iScan4Idx + 4], LD64 (pMV));
- ST64 (pCurLayer->pMvd[LIST_1][iMbXy][iScan4Idx], 0);
- ST64 (pCurLayer->pMvd[LIST_1][iMbXy][iScan4Idx + 4], 0);
- } else { //SUB_4x4
- iMvp[LIST_0][0] = (pSlice->iMvScale[LIST_0][ref[LIST_0]] * mv[0] + 128) >> 8;
- iMvp[LIST_0][1] = (pSlice->iMvScale[LIST_0][ref[LIST_0]] * mv[1] + 128) >> 8;
- ST32 (pCurLayer->pMv[LIST_0][iMbXy][iScan4Idx], LD32 (iMvp[LIST_0]));
- ST32 (pCurLayer->pMvd[LIST_0][iMbXy][iScan4Idx], 0);
- iMvp[LIST_1][0] -= iMvp[LIST_0][0] - mv[0];
- iMvp[LIST_1][1] -= iMvp[LIST_0][0] - mv[1];
- ST32 (pCurLayer->pMv[LIST_1][iMbXy][iScan4Idx], LD32 (iMvp[LIST_1]));
- ST32 (pCurLayer->pMvd[LIST_1][iMbXy][iScan4Idx], 0);
- }
- }
+ FillTemporalDirect8x8Mv (pCurLayer, iIdx8, pSubPartCount[i], pPartW[i], subMbType, ref, mvColoc, NULL, NULL);
}
}
}
@@ -981,4 +895,231 @@
}
}
+void FillSpatialDirect8x8Mv (PDqLayer pCurLayer, const int16_t& iIdx8, const int8_t& iPartCount, const int8_t& iPartW,
+ const SubMbType& subMbType, const bool& bIsLongRef, int16_t pMvDirect[LIST_A][2], int8_t iRef[LIST_A],
+ int16_t pMotionVector[LIST_A][30][MV_A], int16_t pMvdCache[LIST_A][30][MV_A]) {
+ int32_t iMbXy = pCurLayer->iMbXyIndex;
+ for (int32_t j = 0; j < iPartCount; j++) {
+ int8_t iPartIdx = iIdx8 + j * iPartW;
+ uint8_t iScan4Idx = g_kuiScan4[iPartIdx];
+ uint8_t iColocIdx = g_kuiScan4[iPartIdx];
+ uint8_t iCacheIdx = g_kuiCache30ScanIdx[iPartIdx];
+
+ int16_t pMV[4] = { 0 };
+ if (IS_SUB_8x8 (subMbType)) {
+ * (uint32_t*)pMV = * (uint32_t*)pMvDirect[LIST_0];
+ ST32 ((pMV + 2), LD32 (pMV));
+ ST64 (pCurLayer->pMv[LIST_0][iMbXy][iScan4Idx], LD64 (pMV));
+ ST64 (pCurLayer->pMv[LIST_0][iMbXy][iScan4Idx + 4], LD64 (pMV));
+ ST64 (pCurLayer->pMvd[LIST_0][iMbXy][iScan4Idx], 0);
+ ST64 (pCurLayer->pMvd[LIST_0][iMbXy][iScan4Idx + 4], 0);
+ if (pMotionVector != NULL) {
+ ST64 (pMotionVector[LIST_0][iCacheIdx], LD64 (pMV));
+ ST64 (pMotionVector[LIST_0][iCacheIdx + 6], LD64 (pMV));
+ }
+ if (pMvdCache != NULL) {
+ ST64 (pMvdCache[LIST_0][iCacheIdx], 0);
+ ST64 (pMvdCache[LIST_0][iCacheIdx + 6], 0);
+ }
+ * (uint32_t*)pMV = * (uint32_t*)pMvDirect[LIST_1];
+ ST32 ((pMV + 2), LD32 (pMV));
+ ST64 (pCurLayer->pMv[LIST_1][iMbXy][iScan4Idx], LD64 (pMV));
+ ST64 (pCurLayer->pMv[LIST_1][iMbXy][iScan4Idx + 4], LD64 (pMV));
+ ST64 (pCurLayer->pMvd[LIST_1][iMbXy][iScan4Idx], 0);
+ ST64 (pCurLayer->pMvd[LIST_1][iMbXy][iScan4Idx + 4], 0);
+ if (pMotionVector != NULL) {
+ ST64 (pMotionVector[LIST_1][iCacheIdx], LD64 (pMV));
+ ST64 (pMotionVector[LIST_1][iCacheIdx + 6], LD64 (pMV));
+ }
+ if (pMvdCache != NULL) {
+ ST64 (pMvdCache[LIST_1][iCacheIdx], 0);
+ ST64 (pMvdCache[LIST_1][iCacheIdx + 6], 0);
+ }
+ } else { //SUB_4x4
+ * (uint32_t*)pMV = * (uint32_t*)pMvDirect[LIST_0];
+ ST32 (pCurLayer->pMv[LIST_0][iMbXy][iScan4Idx], LD32 (pMV));
+ ST32 (pCurLayer->pMvd[LIST_0][iMbXy][iScan4Idx], 0);
+ if (pMotionVector != NULL) {
+ ST32 (pMotionVector[LIST_0][iCacheIdx], LD32 (pMV));
+ }
+ if (pMvdCache != NULL) {
+ ST32 (pMvdCache[LIST_0][iCacheIdx], 0);
+ }
+ * (uint32_t*)pMV = * (uint32_t*)pMvDirect[LIST_1];
+ ST32 (pCurLayer->pMv[LIST_1][iMbXy][iScan4Idx], LD32 (pMV));
+ ST32 (pCurLayer->pMvd[LIST_1][iMbXy][iScan4Idx], 0);
+ if (pMotionVector != NULL) {
+ ST32 (pMotionVector[LIST_1][iCacheIdx], LD32 (pMV));
+ }
+ if (pMvdCache != NULL) {
+ ST32 (pMvdCache[LIST_1][iCacheIdx], 0);
+ }
+ }
+ if ((* (int32_t*)pMvDirect[LIST_0] | * (int32_t*)pMvDirect[LIST_1])) {
+ uint32_t uiColZeroFlag = (0 == pCurLayer->iColocIntra[iColocIdx]) && !bIsLongRef &&
+ (pCurLayer->iColocRefIndex[LIST_0][iColocIdx] == 0 || (pCurLayer->iColocRefIndex[LIST_0][iColocIdx] < 0
+ && pCurLayer->iColocRefIndex[LIST_1][iColocIdx] == 0));
+ const int16_t (*mvColoc)[2] = 0 == pCurLayer->iColocRefIndex[LIST_0][iColocIdx] ? pCurLayer->iColocMv[LIST_0] :
+ pCurLayer->iColocMv[LIST_1];
+ const int16_t* mv = mvColoc[iColocIdx];
+ if (IS_SUB_8x8 (subMbType)) {
+ if (uiColZeroFlag && ((unsigned) (mv[0] + 1) <= 2 && (unsigned) (mv[1] + 1) <= 2)) {
+ if (iRef[LIST_0] == 0) {
+ ST64 (pCurLayer->pMv[LIST_0][iMbXy][iScan4Idx], 0);
+ ST64 (pCurLayer->pMv[LIST_0][iMbXy][iScan4Idx + 4], 0);
+ ST64 (pCurLayer->pMvd[LIST_0][iMbXy][iScan4Idx], 0);
+ ST64 (pCurLayer->pMvd[LIST_0][iMbXy][iScan4Idx + 4], 0);
+ if (pMotionVector != NULL) {
+ ST64 (pMotionVector[LIST_0][iCacheIdx], 0);
+ ST64 (pMotionVector[LIST_0][iCacheIdx + 6], 0);
+ }
+ if (pMvdCache != NULL) {
+ ST64 (pMvdCache[LIST_0][iCacheIdx], 0);
+ ST64 (pMvdCache[LIST_0][iCacheIdx + 6], 0);
+ }
+ }
+
+ if (iRef[LIST_1] == 0) {
+ ST64 (pCurLayer->pMv[LIST_1][iMbXy][iScan4Idx], 0);
+ ST64 (pCurLayer->pMv[LIST_1][iMbXy][iScan4Idx + 4], 0);
+ ST64 (pCurLayer->pMvd[LIST_1][iMbXy][iScan4Idx], 0);
+ ST64 (pCurLayer->pMvd[LIST_1][iMbXy][iScan4Idx + 4], 0);
+ if (pMotionVector != NULL) {
+ ST64 (pMotionVector[LIST_1][iCacheIdx], 0);
+ ST64 (pMotionVector[LIST_1][iCacheIdx + 6], 0);
+ }
+ if (pMvdCache != NULL) {
+ ST64 (pMvdCache[LIST_1][iCacheIdx], 0);
+ ST64 (pMvdCache[LIST_1][iCacheIdx + 6], 0);
+ }
+ }
+ }
+ } else {
+ if (uiColZeroFlag && ((unsigned) (mv[0] + 1) <= 2 && (unsigned) (mv[1] + 1) <= 2)) {
+ if (iRef[LIST_0] == 0) {
+ ST32 (pCurLayer->pMv[LIST_0][iMbXy][iScan4Idx], 0);
+ ST32 (pCurLayer->pMvd[LIST_0][iMbXy][iScan4Idx], 0);
+ if (pMotionVector != NULL) {
+ ST32 (pMotionVector[LIST_0][iCacheIdx], 0);
+ }
+ if (pMvdCache != NULL) {
+ ST32 (pMvdCache[LIST_0][iCacheIdx], 0);
+ }
+ }
+ if (iRef[LIST_1] == 0) {
+ ST32 (pCurLayer->pMv[LIST_1][iMbXy][iScan4Idx], 0);
+ ST32 (pCurLayer->pMvd[LIST_1][iMbXy][iScan4Idx], 0);
+ if (pMotionVector != NULL) {
+ ST32 (pMotionVector[LIST_1][iCacheIdx], 0);
+ }
+ if (pMvdCache != NULL) {
+ ST32 (pMvdCache[LIST_1][iCacheIdx], 0);
+ }
+ }
+ }
+ }
+ }
+ }
+}
+
+void FillTemporalDirect8x8Mv (PDqLayer pCurLayer, const int16_t& iIdx8, const int8_t& iPartCount, const int8_t& iPartW,
+ const SubMbType& subMbType, int8_t iRef[LIST_A], int16_t (*mvColoc)[2], int16_t pMotionVector[LIST_A][30][MV_A],
+ int16_t pMvdCache[LIST_A][30][MV_A]) {
+ PSlice pSlice = &pCurLayer->sLayerInfo.sSliceInLayer;
+ int32_t iMbXy = pCurLayer->iMbXyIndex;
+ int16_t pMvDirect[LIST_A][2] = { { 0, 0 }, { 0, 0 } };
+ for (int32_t j = 0; j < iPartCount; j++) {
+ int8_t iPartIdx = iIdx8 + j * iPartW;
+ uint8_t iScan4Idx = g_kuiScan4[iPartIdx];
+ uint8_t iColocIdx = g_kuiScan4[iPartIdx];
+ uint8_t iCacheIdx = g_kuiCache30ScanIdx[iPartIdx];
+
+ int16_t* mv = mvColoc[iColocIdx];
+
+ int16_t pMV[4] = { 0 };
+ if (IS_SUB_8x8 (subMbType)) {
+ if (!pCurLayer->iColocIntra[iColocIdx]) {
+ pMvDirect[LIST_0][0] = (pSlice->iMvScale[LIST_0][iRef[LIST_0]] * mv[0] + 128) >> 8;
+ pMvDirect[LIST_0][1] = (pSlice->iMvScale[LIST_0][iRef[LIST_0]] * mv[1] + 128) >> 8;
+ }
+ ST32 (pMV, LD32 (pMvDirect[LIST_0]));
+ ST32 ((pMV + 2), LD32 (pMvDirect[LIST_0]));
+ ST64 (pCurLayer->pMv[LIST_0][iMbXy][iScan4Idx], LD64 (pMV));
+ ST64 (pCurLayer->pMv[LIST_0][iMbXy][iScan4Idx + 4], LD64 (pMV));
+ ST64 (pCurLayer->pMvd[LIST_0][iMbXy][iScan4Idx], 0);
+ ST64 (pCurLayer->pMvd[LIST_0][iMbXy][iScan4Idx + 4], 0);
+ if (pMotionVector != NULL) {
+ ST64 (pMotionVector[LIST_0][iCacheIdx], LD64 (pMV));
+ ST64 (pMotionVector[LIST_0][iCacheIdx + 6], LD64 (pMV));
+ }
+ if (pMvdCache != NULL) {
+ ST64 (pMvdCache[LIST_0][iCacheIdx], 0);
+ ST64 (pMvdCache[LIST_0][iCacheIdx + 6], 0);
+ }
+ if (!pCurLayer->iColocIntra[g_kuiScan4[iIdx8]]) {
+ pMvDirect[LIST_1][0] = pMvDirect[LIST_0][0] - mv[0];
+ pMvDirect[LIST_1][1] = pMvDirect[LIST_0][1] - mv[1];
+ }
+ ST32 (pMV, LD32 (pMvDirect[LIST_1]));
+ ST32 ((pMV + 2), LD32 (pMvDirect[LIST_1]));
+ ST64 (pCurLayer->pMv[LIST_1][iMbXy][iScan4Idx], LD64 (pMV));
+ ST64 (pCurLayer->pMv[LIST_1][iMbXy][iScan4Idx + 4], LD64 (pMV));
+ ST64 (pCurLayer->pMvd[LIST_1][iMbXy][iScan4Idx], 0);
+ ST64 (pCurLayer->pMvd[LIST_1][iMbXy][iScan4Idx + 4], 0);
+ if (pMotionVector != NULL) {
+ ST64 (pMotionVector[LIST_1][iCacheIdx], LD64 (pMV));
+ ST64 (pMotionVector[LIST_1][iCacheIdx + 6], LD64 (pMV));
+ }
+ if (pMvdCache != NULL) {
+ ST64 (pMvdCache[LIST_1][iCacheIdx], 0);
+ ST64 (pMvdCache[LIST_1][iCacheIdx + 6], 0);
+ }
+ } else { //SUB_4x4
+ if (!pCurLayer->iColocIntra[iColocIdx]) {
+ pMvDirect[LIST_0][0] = (pSlice->iMvScale[LIST_0][iRef[LIST_0]] * mv[0] + 128) >> 8;
+ pMvDirect[LIST_0][1] = (pSlice->iMvScale[LIST_0][iRef[LIST_0]] * mv[1] + 128) >> 8;
+ }
+ ST32 (pCurLayer->pMv[LIST_0][iMbXy][iScan4Idx], LD32 (pMvDirect[LIST_0]));
+ ST32 (pCurLayer->pMvd[LIST_0][iMbXy][iScan4Idx], 0);
+ if (pMotionVector != NULL) {
+ ST32 (pMotionVector[LIST_0][iCacheIdx], LD32 (pMvDirect[LIST_0]));
+ }
+ if (pMvdCache != NULL) {
+ ST32 (pMvdCache[LIST_0][iCacheIdx], 0);
+ }
+ if (!pCurLayer->iColocIntra[iColocIdx]) {
+ pMvDirect[LIST_1][0] = pMvDirect[LIST_0][0] - mv[0];
+ pMvDirect[LIST_1][1] = pMvDirect[LIST_0][1] - mv[1];
+ }
+ ST32 (pCurLayer->pMv[LIST_1][iMbXy][iScan4Idx], LD32 (pMvDirect[LIST_1]));
+ ST32 (pCurLayer->pMvd[LIST_1][iMbXy][iScan4Idx], 0);
+ if (pMotionVector != NULL) {
+ ST32 (pMotionVector[LIST_1][iCacheIdx], LD32 (pMvDirect[LIST_1]));
+ }
+ if (pMvdCache != NULL) {
+ ST32 (pMvdCache[LIST_1][iCacheIdx], 0);
+ }
+ }
+ }
+}
+int8_t MapColToList0 (PWelsDecoderContext& pCtx, const int8_t& colocRefIndexL0,
+ const int32_t& ref0Count) { //ISO/IEC 14496-10:2009(E) (8-193)
+ PPicture pic1 = pCtx->sRefPic.pRefList[LIST_1][0];
+ if (pic1 && pic1->pRefPic[LIST_0][colocRefIndexL0]) {
+ const int32_t iFramePoc = pic1->pRefPic[LIST_0][colocRefIndexL0]->iFramePoc;
+ for (int32_t i = 0; i < ref0Count; i++) {
+ if (pCtx->sRefPic.pRefList[LIST_0][i]->iFramePoc == iFramePoc) {
+ return i;
+ }
+ }
+ }
+ return 0;
+}
+void Update8x8RefIdx (PDqLayer& pCurDqLayer, const int16_t& iPartIdx, const int32_t& listIdx, const int8_t& iRef) {
+ int32_t iMbXy = pCurDqLayer->iMbXyIndex;
+ const uint8_t iScan4Idx = g_kuiScan4[iPartIdx];
+ pCurDqLayer->pRefIndex[listIdx][iMbXy][iScan4Idx] = pCurDqLayer->pRefIndex[listIdx][iMbXy][iScan4Idx + 1] =
+ pCurDqLayer->pRefIndex[listIdx][iMbXy][iScan4Idx + 4] = pCurDqLayer->pRefIndex[listIdx][iMbXy][iScan4Idx + 5] = iRef;
+
+}
} // namespace WelsDec
--- a/codec/decoder/core/src/parse_mb_syn_cabac.cpp
+++ b/codec/decoder/core/src/parse_mb_syn_cabac.cpp
@@ -739,9 +739,9 @@
if (IS_DIRECT (mbType)) {
int16_t pMvDirect[LIST_A][2] = { { 0, 0 }, { 0, 0 } };
+ SubMbType subMbType;
if (pSliceHeader->iDirectSpatialMvPredFlag) {
//predict direct spatial mv
- SubMbType subMbType;
int32_t ret = PredMvBDirectSpatial (pCtx, pMvDirect, iRef, subMbType);
if (ret != ERR_NONE) {
return ret;
@@ -748,8 +748,7 @@
}
} else {
//temporal direct 16x16 mode
- ComputeColocated (pCtx);
- int32_t ret = PredBDirectTemporal (pCtx, pMvDirect, iRef);
+ int32_t ret = PredBDirectTemporal (pCtx, pMvDirect, iRef, subMbType);
if (ret != ERR_NONE) {
return ret;
}
@@ -883,6 +882,8 @@
uint32_t uiSubMbType;
//sub_mb_type, partition
int16_t pMvDirect[LIST_A][2] = { {0, 0}, {0, 0} };
+ bool bIsLongRef = pCtx->sRefPic.pRefList[LIST_1][0]->bIsLongRef;
+ const int32_t ref0Count = WELS_MIN (pSliceHeader->uiRefCount[LIST_0], pCtx->sRefPic.uiRefCount[LIST_0]);
bool has_direct_called = false;
SubMbType directSubMbType = 0;
for (int32_t i = 0; i < 4; i++) {
@@ -908,8 +909,7 @@
} else {
//temporal direct mode
- ComputeColocated (pCtx);
- int32_t ret = PredBDirectTemporal (pCtx, pMvDirect, iRef);
+ int32_t ret = PredBDirectTemporal (pCtx, pMvDirect, iRef, directSubMbType);
if (ret != ERR_NONE) {
return ret;
}
@@ -926,152 +926,32 @@
}
}
for (int32_t i = 0; i < 4; i++) { //Direct 8x8 Ref and mv
-
int16_t iIdx8 = i << 2;
if (IS_DIRECT (pCurDqLayer->pSubMbType[iMbXy][i])) {
-
- int8_t iPartCount = pSubPartCount[i];
- int16_t iPartIdx, iBlockW = pPartW[i];
- uint8_t iScan4Idx, iCacheIdx, iColocIdx;
- iCacheIdx = g_kuiCache30ScanIdx[iIdx8];
-
- if (!pSliceHeader->iDirectSpatialMvPredFlag) {
+ if (pSliceHeader->iDirectSpatialMvPredFlag) {
+ FillSpatialDirect8x8Mv (pCurDqLayer, iIdx8, pSubPartCount[i], pPartW[i], directSubMbType, bIsLongRef, pMvDirect, iRef,
+ pMotionVector, pMvdCache);
+ } else {
+ int16_t (*mvColoc)[2] = pCurDqLayer->iColocMv[LIST_0];
iRef[LIST_1] = 0;
- if (pCurDqLayer->iColocIntra[g_kuiScan4[iIdx8]]) {
+ iRef[LIST_0] = 0;
+ const uint8_t uiColoc4Idx = g_kuiScan4[iIdx8];
+ if (!pCurDqLayer->iColocIntra[uiColoc4Idx]) {
iRef[LIST_0] = 0;
- } else {
- if (pCurDqLayer->iColocRefIndex[LIST_0][iIdx8] >= 0) {
- iRef[LIST_0] = pCurDqLayer->iColocRefIndex[LIST_0][iIdx8];
+ int8_t colocRefIndexL0 = pCurDqLayer->iColocRefIndex[LIST_0][uiColoc4Idx];
+ if (colocRefIndexL0 >= 0) {
+ iRef[LIST_0] = MapColToList0 (pCtx, colocRefIndexL0, ref0Count);
} else {
- iRef[LIST_0] = pCurDqLayer->iColocRefIndex[LIST_1][iIdx8];
+ mvColoc = pCurDqLayer->iColocMv[LIST_1];
}
}
+ Update8x8RefIdx (pCurDqLayer, iIdx8, LIST_0, iRef[LIST_0]);
+ Update8x8RefIdx (pCurDqLayer, iIdx8, LIST_1, iRef[LIST_1]);
+ UpdateP8x8RefCacheIdxCabac (pRefIndex, iIdx8, LIST_0, iRef[LIST_0]);
+ UpdateP8x8RefCacheIdxCabac (pRefIndex, iIdx8, LIST_1, iRef[LIST_1]);
+ FillTemporalDirect8x8Mv (pCurDqLayer, iIdx8, pSubPartCount[i], pPartW[i], directSubMbType, iRef, mvColoc, pMotionVector,
+ pMvdCache);
}
- for (int32_t j = 0; j < iPartCount; j++) {
- iPartIdx = iIdx8 + j * iBlockW;
- iColocIdx = g_kuiScan4[iPartIdx];
- iScan4Idx = g_kuiScan4[iPartIdx];
- iCacheIdx = g_kuiCache30ScanIdx[iPartIdx];
-
- if (pSliceHeader->iDirectSpatialMvPredFlag) {
- int16_t pMV[4] = { 0 };
- if (IS_SUB_8x8 (pCurDqLayer->pSubMbType[iMbXy][i])) {
- * (uint32_t*)pMV = * (uint32_t*)pMvDirect[LIST_0];
- ST32 ((pMV + 2), LD32 (pMV));
- ST64 (pCurDqLayer->pMv[LIST_0][iMbXy][iScan4Idx], LD64 (pMV));
- ST64 (pCurDqLayer->pMv[LIST_0][iMbXy][iScan4Idx + 4], LD64 (pMV));
- ST64 (pCurDqLayer->pMvd[LIST_0][iMbXy][iScan4Idx], 0);
- ST64 (pCurDqLayer->pMvd[LIST_0][iMbXy][iScan4Idx + 4], 0);
- ST64 (pMotionVector[LIST_0][iCacheIdx], LD64 (pMV));
- ST64 (pMotionVector[LIST_0][iCacheIdx + 6], LD64 (pMV));
- ST64 (pMvdCache[LIST_0][iCacheIdx], 0);
- ST64 (pMvdCache[LIST_0][iCacheIdx + 6], 0);
- * (uint32_t*)pMV = * (uint32_t*)pMvDirect[LIST_1];
- ST32 ((pMV + 2), LD32 (pMV));
- ST64 (pCurDqLayer->pMv[LIST_1][iMbXy][iScan4Idx], LD64 (pMV));
- ST64 (pCurDqLayer->pMv[LIST_1][iMbXy][iScan4Idx + 4], LD64 (pMV));
- ST64 (pCurDqLayer->pMvd[LIST_1][iMbXy][iScan4Idx], 0);
- ST64 (pCurDqLayer->pMvd[LIST_1][iMbXy][iScan4Idx + 4], 0);
- ST64 (pMotionVector[LIST_1][iCacheIdx], LD64 (pMV));
- ST64 (pMotionVector[LIST_1][iCacheIdx + 6], LD64 (pMV));
- ST64 (pMvdCache[LIST_1][iCacheIdx], 0);
- ST64 (pMvdCache[LIST_1][iCacheIdx + 6], 0);
- } else { //SUB_4x4
- * (uint32_t*)pMV = * (uint32_t*)pMvDirect[LIST_0];
- ST32 (pCurDqLayer->pMv[LIST_0][iMbXy][iScan4Idx], LD32 (pMV));
- ST32 (pCurDqLayer->pMvd[LIST_0][iMbXy][iScan4Idx], 0);
- ST32 (pMotionVector[LIST_0][iCacheIdx], LD32 (pMV));
- ST32 (pMvdCache[LIST_0][iCacheIdx], 0);
- * (uint32_t*)pMV = * (uint32_t*)pMvDirect[LIST_1];
- ST32 (pCurDqLayer->pMv[LIST_1][iMbXy][iScan4Idx], LD32 (pMV));
- ST32 (pCurDqLayer->pMvd[LIST_1][iMbXy][iScan4Idx], 0);
- ST32 (pMotionVector[LIST_1][iCacheIdx], LD32 (pMV));
- ST32 (pMvdCache[LIST_1][iCacheIdx], 0);
- }
-
- if ((* (int32_t*)pMvDirect[LIST_0] | * (int32_t*)pMvDirect[LIST_1])) {
- bool bIsLongRef = pCtx->sRefPic.pRefList[LIST_1][0]->bIsLongRef;
- uint32_t uiColZeroFlag = (0 == pCurDqLayer->iColocIntra[iColocIdx]) && !bIsLongRef &&
- (pCurDqLayer->iColocRefIndex[LIST_0][iColocIdx] == 0 || (pCurDqLayer->iColocRefIndex[LIST_0][iColocIdx] < 0
- && pCurDqLayer->iColocRefIndex[LIST_1][iColocIdx] == 0));
- const int16_t (*mvColoc)[2] = pCurDqLayer->iColocRefIndex[LIST_0][iColocIdx] == 0 ? pCurDqLayer->iColocMv[LIST_0] :
- pCurDqLayer->iColocMv[LIST_1];
- const int16_t* mv = mvColoc[iColocIdx];
- if (IS_SUB_8x8 (pCurDqLayer->pSubMbType[iMbXy][i])) {
- if (uiColZeroFlag && ((unsigned) (mv[0] + 1) <= 2 && (unsigned) (mv[1] + 1) <= 2)) {
- if (iRef[LIST_0] == 0) {
- ST64 (pCurDqLayer->pMv[LIST_0][iMbXy][iScan4Idx], 0);
- ST64 (pCurDqLayer->pMv[LIST_0][iMbXy][iScan4Idx + 4], 0);
- ST64 (pCurDqLayer->pMvd[LIST_0][iMbXy][iScan4Idx], 0);
- ST64 (pCurDqLayer->pMvd[LIST_0][iMbXy][iScan4Idx + 4], 0);
- ST64 (pMotionVector[LIST_0][iCacheIdx], 0);
- ST64 (pMotionVector[LIST_0][iCacheIdx + 6], 0);
- ST64 (pMvdCache[LIST_0][iCacheIdx], 0);
- ST64 (pMvdCache[LIST_0][iCacheIdx + 6], 0);
- }
-
- if (iRef[LIST_1] == 0) {
- ST64 (pCurDqLayer->pMv[LIST_1][iMbXy][iScan4Idx], 0);
- ST64 (pCurDqLayer->pMv[LIST_1][iMbXy][iScan4Idx + 4], 0);
- ST64 (pCurDqLayer->pMvd[LIST_1][iMbXy][iScan4Idx], 0);
- ST64 (pCurDqLayer->pMvd[LIST_1][iMbXy][iScan4Idx + 4], 0);
- ST64 (pMotionVector[LIST_1][iCacheIdx], 0);
- ST64 (pMotionVector[LIST_1][iCacheIdx + 6], 0);
- ST64 (pMvdCache[LIST_1][iCacheIdx], 0);
- ST64 (pMvdCache[LIST_1][iCacheIdx + 6], 0);
- }
- }
- } else {
- if (uiColZeroFlag && ((unsigned) (mv[0] + 1) <= 2 && (unsigned) (mv[1] + 1) <= 2)) {
- if (iRef[LIST_0] == 0) {
- ST32 (pCurDqLayer->pMv[LIST_0][iMbXy][iScan4Idx], 0);
- ST32 (pCurDqLayer->pMvd[LIST_0][iMbXy][iScan4Idx], 0);
- ST32 (pMotionVector[LIST_0][iCacheIdx], 0);
- ST32 (pMvdCache[LIST_0][iCacheIdx], 0);
- }
- if (iRef[LIST_1] == 0) {
- ST32 (pCurDqLayer->pMv[LIST_1][iMbXy][iScan4Idx], 0);
- ST32 (pCurDqLayer->pMvd[LIST_1][iMbXy][iScan4Idx], 0);
- ST32 (pMotionVector[LIST_1][iCacheIdx], 0);
- ST32 (pMvdCache[LIST_1][iCacheIdx], 0);
- }
- }
- }
- }
- } else {
- int16_t (*mvColoc)[2] = pCurDqLayer->iColocMv[LIST_0];
- int16_t* mv = mvColoc[iColocIdx];
- int16_t pMV[4] = { 0 };
- int16_t iMvp[LIST_A][2];
- if (IS_SUB_8x8 (pCurDqLayer->pSubMbType[iMbXy][i])) {
- iMvp[LIST_0][0] = (pSlice->iMvScale[LIST_0][iRef[LIST_0]] * mv[0] + 128) >> 8;
- iMvp[LIST_0][1] = (pSlice->iMvScale[LIST_0][iRef[LIST_0]] * mv[1] + 128) >> 8;
- ST32 (pMV, LD32 (iMvp[LIST_0]));
- ST32 ((pMV + 2), LD32 (iMvp[LIST_0]));
- ST64 (pCurDqLayer->pMv[LIST_0][iMbXy][iScan4Idx], LD64 (pMV));
- ST64 (pCurDqLayer->pMv[LIST_0][iMbXy][iScan4Idx + 4], LD64 (pMV));
- ST64 (pCurDqLayer->pMvd[LIST_0][iMbXy][iScan4Idx], 0);
- ST64 (pCurDqLayer->pMvd[LIST_0][iMbXy][iScan4Idx + 4], 0);
- iMvp[LIST_1][0] -= iMvp[LIST_0][0] - mv[0];
- iMvp[LIST_1][1] -= iMvp[LIST_0][0] - mv[1];
- ST32 (pMV, LD32 (iMvp[LIST_1]));
- ST32 ((pMV + 2), LD32 (iMvp[LIST_1]));
- ST64 (pCurDqLayer->pMv[LIST_1][iMbXy][iScan4Idx], LD64 (pMV));
- ST64 (pCurDqLayer->pMv[LIST_1][iMbXy][iScan4Idx + 4], LD64 (pMV));
- ST64 (pCurDqLayer->pMvd[LIST_1][iMbXy][iScan4Idx], 0);
- ST64 (pCurDqLayer->pMvd[LIST_1][iMbXy][iScan4Idx + 4], 0);
- } else { //SUB_4x4
- iMvp[LIST_0][0] = (pSlice->iMvScale[LIST_0][iRef[LIST_0]] * mv[0] + 128) >> 8;
- iMvp[LIST_0][1] = (pSlice->iMvScale[LIST_0][iRef[LIST_0]] * mv[1] + 128) >> 8;
- ST32 (pCurDqLayer->pMv[LIST_0][iMbXy][iScan4Idx], LD32 (iMvp[LIST_0]));
- ST32 (pCurDqLayer->pMvd[LIST_0][iMbXy][iScan4Idx], 0);
- iMvp[LIST_1][0] -= iMvp[LIST_0][0] - mv[0];
- iMvp[LIST_1][1] -= iMvp[LIST_0][0] - mv[1];
- ST32 (pCurDqLayer->pMv[LIST_1][iMbXy][iScan4Idx], LD32 (iMvp[LIST_1]));
- ST32 (pCurDqLayer->pMvd[LIST_1][iMbXy][iScan4Idx], 0);
- }
- }
- }
}
}
//ref no-direct
@@ -1083,18 +963,8 @@
int8_t iref = REF_NOT_IN_LIST;
if (IS_DIRECT (subMbType)) {
if (pSliceHeader->iDirectSpatialMvPredFlag) {
- iref = iRef[listIdx];
- } else {
- iref = 0;
- if (listIdx == LIST_0) {
- if (!pCurDqLayer->iColocIntra[g_kuiScan4[iIdx8]]) {
- if (pCurDqLayer->iColocRefIndex[LIST_0][iIdx8] >= 0) {
- iref = pCurDqLayer->iColocRefIndex[LIST_0][iIdx8];
- } else {
- iref = pCurDqLayer->iColocRefIndex[LIST_1][iIdx8];
- }
- }
- }
+ Update8x8RefIdx (pCurDqLayer, iIdx8, listIdx, iRef[listIdx]);
+ ref_idx_list[listIdx][i] = iRef[listIdx];
}
UpdateP8x8DirectCabac (pCurDqLayer, iIdx8);
} else {
@@ -1114,29 +984,30 @@
pCtx->bMbRefConcealed = pCtx->bRPLRError || pCtx->bMbRefConcealed || ! (pCtx->sRefPic.pRefList[listIdx][iref]
&& pCtx->sRefPic.pRefList[listIdx][iref]->bIsComplete);
}
+ Update8x8RefIdx (pCurDqLayer, iIdx8, listIdx, iref);
+ ref_idx_list[listIdx][i] = iref;
}
- UpdateP8x8RefIdxCabac (pCurDqLayer, pRefIndex, iIdx8, iref, listIdx);
- ref_idx_list[listIdx][i] = iref;
}
}
//mv
for (int32_t listIdx = LIST_0; listIdx < LIST_A; ++listIdx) {
for (int32_t i = 0; i < 4; i++) {
- int8_t iPartCount = pSubPartCount[i];
- int16_t iPartIdx, iBlockW = pPartW[i];
- uint8_t iScan4Idx, iCacheIdx;
+ int16_t iIdx8 = i << 2;
- iCacheIdx = g_kuiCache30ScanIdx[i << 2];
+ uint32_t subMbType = pCurDqLayer->pSubMbType[iMbXy][i];
+ if (IS_DIRECT (subMbType) && !pSliceHeader->iDirectSpatialMvPredFlag)
+ continue;
int8_t iref = ref_idx_list[listIdx][i];
- pRefIndex[listIdx][iCacheIdx] = pRefIndex[listIdx][iCacheIdx + 1]
- = pRefIndex[listIdx][iCacheIdx + 6] = pRefIndex[listIdx][iCacheIdx + 7] = iref;
+ UpdateP8x8RefCacheIdxCabac (pRefIndex, iIdx8, listIdx, iref);
- uint32_t subMbType = pCurDqLayer->pSubMbType[iMbXy][i];
- if (IS_DIRECT (subMbType)) {
+ if (IS_DIRECT (subMbType))
continue;
- }
+
bool is_dir = IS_DIR (subMbType, 0, listIdx) > 0;
+ int8_t iPartCount = pSubPartCount[i];
+ int16_t iBlockW = pPartW[i];
+ uint8_t iScan4Idx, iCacheIdx;
for (int32_t j = 0; j < iPartCount; j++) {
iPartIdx = (i << 2) + j * iBlockW;
iScan4Idx = g_kuiScan4[iPartIdx];
@@ -1668,5 +1539,11 @@
WELS_READ_VERIFY (InitReadBits (pBsAux, 1));
WELS_READ_VERIFY (InitCabacDecEngineFromBS (pCabacDecEngine, pBsAux));
return ERR_NONE;
+}
+void UpdateP8x8RefCacheIdxCabac (int8_t pRefIndex[LIST_A][30], const int16_t& iPartIdx,
+ const int32_t& listIdx, const int8_t& iRef) {
+ const uint8_t uiCacheIdx = g_kuiCache30ScanIdx[iPartIdx];
+ pRefIndex[listIdx][uiCacheIdx] = pRefIndex[listIdx][uiCacheIdx + 1] = pRefIndex[listIdx][uiCacheIdx + 6] =
+ pRefIndex[listIdx][uiCacheIdx + 7] = iRef;
}
}
--- a/codec/decoder/core/src/parse_mb_syn_cavlc.cpp
+++ b/codec/decoder/core/src/parse_mb_syn_cavlc.cpp
@@ -1349,9 +1349,9 @@
if (IS_DIRECT (mbType)) {
int16_t pMvDirect[LIST_A][2] = { { 0, 0 }, { 0, 0 } };
+ SubMbType subMbType;
if (pSliceHeader->iDirectSpatialMvPredFlag) {
//predict direct spatial mv
- SubMbType subMbType;
int32_t ret = PredMvBDirectSpatial (pCtx, pMvDirect, iRef, subMbType);
if (ret != ERR_NONE) {
return ret;
@@ -1358,8 +1358,7 @@
}
} else {
//temporal direct 16x16 mode
- ComputeColocated (pCtx);
- int32_t ret = PredBDirectTemporal (pCtx, pMvDirect, iRef);
+ int32_t ret = PredBDirectTemporal (pCtx, pMvDirect, iRef, subMbType);
if (ret != ERR_NONE) {
return ret;
}
@@ -1369,7 +1368,7 @@
for (int32_t listIdx = LIST_0; listIdx < LIST_A; ++listIdx) {
if (IS_DIR (mbType, 0, listIdx)) {
WELS_READ_VERIFY (BsGetOneBit (pBs, &uiCode)); //motion_prediction_flag_l0/l1[ mbPartIdx ]
- iMotionPredFlag[listIdx][0] = uiCode > 0;
+ iMotionPredFlag[listIdx][0] = uiCode;
}
}
}
@@ -1417,7 +1416,7 @@
for (int32_t i = 0; i < 2; ++i) {
if (IS_DIR (mbType, i, listIdx)) {
WELS_READ_VERIFY (BsGetOneBit (pBs, &uiCode)); //motion_prediction_flag_l0/l1[ mbPartIdx ]
- iMotionPredFlag[listIdx][i] = uiCode > 0;
+ iMotionPredFlag[listIdx][i] = uiCode;
}
}
}
@@ -1474,7 +1473,7 @@
for (int32_t i = 0; i < 2; ++i) {
if (IS_DIR (mbType, i, listIdx)) {
WELS_READ_VERIFY (BsGetOneBit (pBs, &uiCode)); //motion_prediction_flag_l0/l1[ mbPartIdx ]
- iMotionPredFlag[listIdx][i] = uiCode > 0;
+ iMotionPredFlag[listIdx][i] = uiCode;
}
}
}
@@ -1529,6 +1528,8 @@
uint32_t uiSubMbType;
//sub_mb_type, partition
int16_t pMvDirect[LIST_A][2] = { { 0, 0 }, { 0, 0 } };
+ bool bIsLongRef = pCtx->sRefPic.pRefList[LIST_1][0]->bIsLongRef;
+ const int32_t ref0Count = WELS_MIN (pSliceHeader->uiRefCount[LIST_0], pCtx->sRefPic.uiRefCount[LIST_0]);
bool has_direct_called = false;
SubMbType directSubMbType = 0;
@@ -1556,8 +1557,7 @@
} else {
//temporal direct mode
- ComputeColocated (pCtx);
- int32_t ret = PredBDirectTemporal (pCtx, pMvDirect, iRef);
+ int32_t ret = PredBDirectTemporal (pCtx, pMvDirect, iRef, directSubMbType);
if (ret != ERR_NONE) {
return ret;
}
@@ -1579,7 +1579,7 @@
bool is_dir = IS_DIR (pCurDqLayer->pSubMbType[iMbXy][i], 0, listIdx) > 0;
if (is_dir) {
WELS_READ_VERIFY (BsGetOneBit (pBs, &uiCode)); //motion_prediction_flag_l0[ mbPartIdx ]
- iMotionPredFlag[listIdx][i] = uiCode > 0;
+ iMotionPredFlag[listIdx][i] = uiCode;
}
}
}
@@ -1587,136 +1587,28 @@
for (int32_t i = 0; i < 4; i++) { //Direct 8x8 Ref and mv
int16_t iIdx8 = i << 2;
if (IS_DIRECT (pCurDqLayer->pSubMbType[iMbXy][i])) {
- int8_t iPartCount = pSubPartCount[i];
- int16_t iPartIdx, iBlockW = pPartW[i];
- uint8_t iScan4Idx, iCacheIdx, iColocIdx;
- iCacheIdx = g_kuiCache30ScanIdx[iIdx8];
-
- if (!pSliceHeader->iDirectSpatialMvPredFlag) {
+ if (pSliceHeader->iDirectSpatialMvPredFlag) {
+ FillSpatialDirect8x8Mv (pCurDqLayer, iIdx8, pSubPartCount[i], pPartW[i], directSubMbType, bIsLongRef, pMvDirect, iRef,
+ iMvArray, NULL);
+ } else {
+ int16_t (*mvColoc)[2] = pCurDqLayer->iColocMv[LIST_0];
iRef[LIST_1] = 0;
- if (pCurDqLayer->iColocIntra[g_kuiScan4[iIdx8]]) {
+ iRef[LIST_0] = 0;
+ const uint8_t uiColoc4Idx = g_kuiScan4[iIdx8];
+ if (!pCurDqLayer->iColocIntra[uiColoc4Idx]) {
iRef[LIST_0] = 0;
- } else {
- if (pCurDqLayer->iColocRefIndex[LIST_0][iIdx8] >= 0) {
- iRef[LIST_0] = pCurDqLayer->iColocRefIndex[LIST_0][iIdx8];
+ int8_t colocRefIndexL0 = pCurDqLayer->iColocRefIndex[LIST_0][uiColoc4Idx];
+ if (colocRefIndexL0 >= 0) {
+ iRef[LIST_0] = MapColToList0 (pCtx, colocRefIndexL0, ref0Count);
} else {
- iRef[LIST_0] = pCurDqLayer->iColocRefIndex[LIST_1][iIdx8];
+ mvColoc = pCurDqLayer->iColocMv[LIST_1];
}
}
+ Update8x8RefIdx (pCurDqLayer, iIdx8, LIST_0, iRef[LIST_0]);
+ Update8x8RefIdx (pCurDqLayer, iIdx8, LIST_1, iRef[LIST_1]);
+ FillTemporalDirect8x8Mv (pCurDqLayer, iIdx8, pSubPartCount[i], pPartW[i], directSubMbType, iRef, mvColoc, iMvArray,
+ NULL);
}
- for (int32_t j = 0; j < iPartCount; j++) {
- iPartIdx = iIdx8 + j * iBlockW;
- iColocIdx = g_kuiScan4[iPartIdx];
- iScan4Idx = g_kuiScan4[iPartIdx];
- iCacheIdx = g_kuiCache30ScanIdx[iPartIdx];
-
- if (pSliceHeader->iDirectSpatialMvPredFlag) {
- int16_t pMV[4] = { 0 };
- if (IS_SUB_8x8 (pCurDqLayer->pSubMbType[iMbXy][i])) {
- * (uint32_t*)pMV = * (uint32_t*)pMvDirect[LIST_0];
- ST32 ((pMV + 2), LD32 (pMV));
- ST64 (pCurDqLayer->pMv[LIST_0][iMbXy][iScan4Idx], LD64 (pMV));
- ST64 (pCurDqLayer->pMv[LIST_0][iMbXy][iScan4Idx + 4], LD64 (pMV));
- ST64 (pCurDqLayer->pMvd[LIST_0][iMbXy][iScan4Idx], 0);
- ST64 (pCurDqLayer->pMvd[LIST_0][iMbXy][iScan4Idx + 4], 0);
- ST64 (iMvArray[LIST_0][iCacheIdx], LD64 (pMV));
- ST64 (iMvArray[LIST_0][iCacheIdx + 6], LD64 (pMV));
- * (uint32_t*)pMV = * (uint32_t*)pMvDirect[LIST_1];
- ST32 ((pMV + 2), LD32 (pMV));
- ST64 (pCurDqLayer->pMv[LIST_1][iMbXy][iScan4Idx], LD64 (pMV));
- ST64 (pCurDqLayer->pMv[LIST_1][iMbXy][iScan4Idx + 4], LD64 (pMV));
- ST64 (pCurDqLayer->pMvd[LIST_1][iMbXy][iScan4Idx], 0);
- ST64 (pCurDqLayer->pMvd[LIST_1][iMbXy][iScan4Idx + 4], 0);
- ST64 (iMvArray[LIST_1][iCacheIdx], LD64 (pMV));
- ST64 (iMvArray[LIST_1][iCacheIdx + 6], LD64 (pMV));
- } else { //SUB_4x4
- * (uint32_t*)pMV = * (uint32_t*)pMvDirect[LIST_0];
- ST32 (pCurDqLayer->pMv[LIST_0][iMbXy][iScan4Idx], LD32 (pMV));
- ST32 (pCurDqLayer->pMvd[LIST_0][iMbXy][iScan4Idx], 0);
- ST32 (iMvArray[LIST_0][iCacheIdx], LD32 (pMV));
- * (uint32_t*)pMV = * (uint32_t*)pMvDirect[LIST_1];
- ST32 (pCurDqLayer->pMv[LIST_1][iMbXy][iScan4Idx], LD32 (pMV));
- ST32 (pCurDqLayer->pMvd[LIST_1][iMbXy][iScan4Idx], 0);
- ST32 (iMvArray[LIST_1][iCacheIdx], LD32 (pMV));
- }
-
- if ((* (int32_t*)pMvDirect[LIST_0] | * (int32_t*)pMvDirect[LIST_1])) {
- bool bIsLongRef = pCtx->sRefPic.pRefList[LIST_1][0]->bIsLongRef;
- uint32_t uiColZeroFlag = (0 == pCurDqLayer->iColocIntra[iColocIdx]) && !bIsLongRef &&
- (pCurDqLayer->iColocRefIndex[LIST_0][iColocIdx] == 0 || (pCurDqLayer->iColocRefIndex[LIST_0][iColocIdx] < 0
- && pCurDqLayer->iColocRefIndex[LIST_1][iColocIdx] == 0));
- const int16_t (*mvColoc)[2] = pCurDqLayer->iColocRefIndex[LIST_0][iColocIdx] == 0 ? pCurDqLayer->iColocMv[LIST_0] :
- pCurDqLayer->iColocMv[LIST_1];
- const int16_t* mv = mvColoc[iColocIdx];
- if (IS_SUB_8x8 (pCurDqLayer->pSubMbType[iMbXy][i])) {
- if (uiColZeroFlag && ((unsigned) (mv[0] + 1) <= 2 && (unsigned) (mv[1] + 1) <= 2)) {
- if (iRef[LIST_0] == 0) {
- ST64 (pCurDqLayer->pMv[LIST_0][iMbXy][iScan4Idx], 0);
- ST64 (pCurDqLayer->pMv[LIST_0][iMbXy][iScan4Idx + 4], 0);
- ST64 (pCurDqLayer->pMvd[LIST_0][iMbXy][iScan4Idx], 0);
- ST64 (pCurDqLayer->pMvd[LIST_0][iMbXy][iScan4Idx + 4], 0);
- ST64 (iMvArray[LIST_0][iCacheIdx], 0);
- ST64 (iMvArray[LIST_0][iCacheIdx + 6], 0);
- }
-
- if (iRef[LIST_1] == 0) {
- ST64 (pCurDqLayer->pMv[LIST_1][iMbXy][iScan4Idx], 0);
- ST64 (pCurDqLayer->pMv[LIST_1][iMbXy][iScan4Idx + 4], 0);
- ST64 (pCurDqLayer->pMvd[LIST_1][iMbXy][iScan4Idx], 0);
- ST64 (pCurDqLayer->pMvd[LIST_1][iMbXy][iScan4Idx + 4], 0);
- ST64 (iMvArray[LIST_1][iCacheIdx], 0);
- ST64 (iMvArray[LIST_1][iCacheIdx + 6], 0);
- }
- }
- } else {
- if (uiColZeroFlag && ((unsigned) (mv[0] + 1) <= 2 && (unsigned) (mv[1] + 1) <= 2)) {
- if (iRef[LIST_0] == 0) {
- ST32 (pCurDqLayer->pMv[LIST_0][iMbXy][iScan4Idx], 0);
- ST32 (pCurDqLayer->pMvd[LIST_0][iMbXy][iScan4Idx], 0);
- ST32 (iMvArray[LIST_0][iCacheIdx], 0);
- }
- if (iRef[LIST_1] == 0) {
- ST32 (pCurDqLayer->pMv[LIST_1][iMbXy][iScan4Idx], 0);
- ST32 (pCurDqLayer->pMvd[LIST_1][iMbXy][iScan4Idx], 0);
- ST32 (iMvArray[LIST_1][iCacheIdx], 0);
- }
- }
- }
- }
- } else {
- int16_t (*mvColoc)[2] = pCurDqLayer->iColocMv[LIST_0];
- int16_t* mv = mvColoc[iColocIdx];
- int16_t pMV[4] = { 0 };
- int16_t iMvp[LIST_A][2];
- if (IS_SUB_8x8 (pCurDqLayer->pSubMbType[iMbXy][i])) {
- iMvp[LIST_0][0] = (pSlice->iMvScale[LIST_0][iRef[LIST_0]] * mv[0] + 128) >> 8;
- iMvp[LIST_0][1] = (pSlice->iMvScale[LIST_0][iRef[LIST_0]] * mv[1] + 128) >> 8;
- ST32 (pMV, LD32 (iMvp[LIST_0]));
- ST32 ((pMV + 2), LD32 (iMvp[LIST_0]));
- ST64 (pCurDqLayer->pMv[LIST_0][iMbXy][iScan4Idx], LD64 (pMV));
- ST64 (pCurDqLayer->pMv[LIST_0][iMbXy][iScan4Idx + 4], LD64 (pMV));
- ST64 (pCurDqLayer->pMvd[LIST_0][iMbXy][iScan4Idx], 0);
- ST64 (pCurDqLayer->pMvd[LIST_0][iMbXy][iScan4Idx + 4], 0);
- iMvp[LIST_1][0] -= iMvp[LIST_0][0] - mv[0];
- iMvp[LIST_1][1] -= iMvp[LIST_0][0] - mv[1];
- ST32 (pMV, LD32 (iMvp[LIST_1]));
- ST32 ((pMV + 2), LD32 (iMvp[LIST_1]));
- ST64 (pCurDqLayer->pMv[LIST_1][iMbXy][iScan4Idx], LD64 (pMV));
- ST64 (pCurDqLayer->pMv[LIST_1][iMbXy][iScan4Idx + 4], LD64 (pMV));
- ST64 (pCurDqLayer->pMvd[LIST_1][iMbXy][iScan4Idx], 0);
- ST64 (pCurDqLayer->pMvd[LIST_1][iMbXy][iScan4Idx + 4], 0);
- } else { //SUB_4x4
- iMvp[LIST_0][0] = (pSlice->iMvScale[LIST_0][iRef[LIST_0]] * mv[0] + 128) >> 8;
- iMvp[LIST_0][1] = (pSlice->iMvScale[LIST_0][iRef[LIST_0]] * mv[1] + 128) >> 8;
- ST32 (pCurDqLayer->pMv[LIST_0][iMbXy][iScan4Idx], LD32 (iMvp[LIST_0]));
- ST32 (pCurDqLayer->pMvd[LIST_0][iMbXy][iScan4Idx], 0);
- iMvp[LIST_1][0] -= iMvp[LIST_0][0] - mv[0];
- iMvp[LIST_1][1] -= iMvp[LIST_0][0] - mv[1];
- ST32 (pCurDqLayer->pMv[LIST_1][iMbXy][iScan4Idx], LD32 (iMvp[LIST_1]));
- ST32 (pCurDqLayer->pMvd[LIST_1][iMbXy][iScan4Idx], 0);
- }
- }
- }
}
}
//ref no-direct
@@ -1723,23 +1615,12 @@
for (int32_t listIdx = LIST_0; listIdx < LIST_A; ++listIdx) {
for (int32_t i = 0; i < 4; i++) {
int16_t iIdx8 = i << 2;
- uint8_t uiScan4Idx = g_kuiScan4[iIdx8];
int32_t subMbType = pCurDqLayer->pSubMbType[iMbXy][i];
int8_t iref = REF_NOT_IN_LIST;
if (IS_DIRECT (subMbType)) {
if (pSliceHeader->iDirectSpatialMvPredFlag) {
- iref = iRef[listIdx];
- } else {
- iref = 0;
- if (listIdx == LIST_0) {
- if (!pCurDqLayer->iColocIntra[g_kuiScan4[iIdx8]]) {
- if (pCurDqLayer->iColocRefIndex[LIST_0][iIdx8] >= 0) {
- iref = pCurDqLayer->iColocRefIndex[LIST_0][iIdx8];
- } else {
- iref = pCurDqLayer->iColocRefIndex[LIST_1][iIdx8];
- }
- }
- }
+ Update8x8RefIdx (pCurDqLayer, iIdx8, listIdx, iRef[listIdx]);
+ ref_idx_list[listIdx][i] = iRef[listIdx];
}
} else {
if (IS_DIR (subMbType, 0, listIdx)) {
@@ -1746,7 +1627,7 @@
if (iMotionPredFlag[listIdx][i] == 0) {
WELS_READ_VERIFY (BsGetTe0 (pBs, iRefCount[listIdx], &uiCode)); //ref_idx_l0[ mbPartIdx ]
iref = uiCode;
- if ((iref < 0) || (iref >= iRefCount[0]) || (ppRefPic[listIdx][iref] == NULL)) { //error ref_idx
+ if ((iref < 0) || (iref >= iRefCount[listIdx]) || (ppRefPic[listIdx][iref] == NULL)) { //error ref_idx
pCtx->bMbRefConcealed = true;
if (pCtx->pParam->eEcActiveIdc != ERROR_CON_DISABLE) {
iref = 0;
@@ -1762,10 +1643,9 @@
return GENERATE_ERROR_NO (ERR_LEVEL_MB_DATA, ERR_INFO_UNSUPPORTED_ILP);
}
}
+ Update8x8RefIdx (pCurDqLayer, iIdx8, listIdx, iref);
+ ref_idx_list[listIdx][i] = iref;
}
- pCurDqLayer->pRefIndex[listIdx][iMbXy][uiScan4Idx] = pCurDqLayer->pRefIndex[listIdx][iMbXy][uiScan4Idx + 1] =
- pCurDqLayer->pRefIndex[listIdx][iMbXy][uiScan4Idx + 4] = pCurDqLayer->pRefIndex[listIdx][iMbXy][uiScan4Idx + 5] = iref;
- ref_idx_list[listIdx][i] = iref;
}
}
//mv
--- a/codec/decoder/plus/inc/welsDecoderExt.h
+++ b/codec/decoder/plus/inc/welsDecoderExt.h
@@ -112,7 +112,7 @@
typedef struct tagPictInfo {
SBufferInfo sBufferInfo;
int32_t iPOC;
- int32_t iFrameNum;
+ int32_t iPicBuffIdx;
bool bLastGOP;
unsigned char* pData[3];
} SPictInfo, *PPictInfo;
--- a/codec/decoder/plus/src/welsDecoderExt.cpp
+++ b/codec/decoder/plus/src/welsDecoderExt.cpp
@@ -690,6 +690,7 @@
ppDst[1] = m_sPictInfoList[m_iPictInfoIndex].pData[1];
ppDst[2] = m_sPictInfoList[m_iPictInfoIndex].pData[2];
m_sPictInfoList[m_iPictInfoIndex].iPOC = sIMinInt32;
+ m_pDecContext->pPicBuff->ppPic[m_sPictInfoList[m_iPictInfoIndex].iPicBuffIdx]->bAvailableFlag = true;
m_sPictInfoList[m_iPictInfoIndex].bLastGOP = false;
m_iMinPOC = sIMinInt32;
--m_iNumOfPicts;
@@ -743,6 +744,7 @@
}
DECODING_STATE CWelsDecoder::ReorderPicturesInDisplay (unsigned char** ppDst, SBufferInfo* pDstInfo) {
+ DECODING_STATE iRet = dsErrorFree;
if (pDstInfo->iBufferStatus == 1 && m_pDecContext->pSps->uiProfileIdc != 66) {
if (m_pDecContext->pSliceHeader->iPicOrderCntLsb == 0) {
if (m_iNumOfPicts > 0) {
@@ -761,7 +763,8 @@
m_sPictInfoList[i].pData[1] = ppDst[1];
m_sPictInfoList[i].pData[2] = ppDst[2];
m_sPictInfoList[i].iPOC = m_pDecContext->pSliceHeader->iPicOrderCntLsb;
- m_sPictInfoList[i].iFrameNum = m_pDecContext->pSliceHeader->iFrameNum;
+ m_sPictInfoList[i].iPicBuffIdx = m_pDecContext->pPicBuff->iCurrentIdx;
+ m_pDecContext->pPicBuff->ppPic[m_sPictInfoList[i].iPicBuffIdx]->bAvailableFlag = false;
m_sPictInfoList[i].bLastGOP = false;
pDstInfo->iBufferStatus = 0;
++m_iNumOfPicts;
@@ -794,6 +797,7 @@
ppDst[1] = m_sPictInfoList[m_iPictInfoIndex].pData[1];
ppDst[2] = m_sPictInfoList[m_iPictInfoIndex].pData[2];
m_sPictInfoList[m_iPictInfoIndex].iPOC = sIMinInt32;
+ m_pDecContext->pPicBuff->ppPic[m_sPictInfoList[m_iPictInfoIndex].iPicBuffIdx]->bAvailableFlag = true;
m_sPictInfoList[m_iPictInfoIndex].bLastGOP = false;
m_iMinPOC = sIMinInt32;
--m_iNumOfPicts;
@@ -801,7 +805,7 @@
if (m_iLastGOPRemainPicts == 0) {
m_LastWrittenPOC = sIMinInt32;
}
- return dsErrorFree;
+ return iRet;
}
if (m_iNumOfPicts > 0) {
m_iMinPOC = sIMinInt32;
@@ -830,15 +834,16 @@
ppDst[1] = m_sPictInfoList[m_iPictInfoIndex].pData[1];
ppDst[2] = m_sPictInfoList[m_iPictInfoIndex].pData[2];
m_sPictInfoList[m_iPictInfoIndex].iPOC = sIMinInt32;
+ m_pDecContext->pPicBuff->ppPic[m_sPictInfoList[m_iPictInfoIndex].iPicBuffIdx]->bAvailableFlag = true;
m_sPictInfoList[m_iPictInfoIndex].bLastGOP = false;
m_iMinPOC = sIMinInt32;
--m_iNumOfPicts;
- return dsErrorFree;
+ return iRet;
}
}
}
- return dsErrorFree;
+ return iRet;
}
DECODING_STATE CWelsDecoder::DecodeParser (const unsigned char* kpSrc,
binary files /dev/null b/res/VID_1280x544_cabac_temporal_direct.264 differ
binary files /dev/null b/res/VID_1280x544_cavlc_temporal_direct.264 differ
binary files /dev/null b/res/VID_1280x720_cabac_temporal_direct.264 differ
binary files /dev/null b/res/VID_1280x720_cavlc_temporal_direct.264 differ
binary files /dev/null b/res/VID_1920x1080_cabac_temporal_direct.264 differ
binary files /dev/null b/res/VID_1920x1080_cavlc_temporal_direct.264 differ
--- a/test/api/decoder_test.cpp
+++ b/test/api/decoder_test.cpp
@@ -134,6 +134,12 @@
{"res/Cisco_Men_whisper_640x320_CABAC_Bframe_9.264", "88b8864a69cee7656202bc54d2ffa8b7b6f1f6c5"},
{"res/Cisco_Men_whisper_640x320_CAVLC_Bframe_9.264", "270a500d2f91c9e2c8ffabc03f62e0dc0b3a24ed"},
{"res/Cisco_Adobe_PDF_sample_a_1024x768_CAVLC_Bframe_9.264", "d3b2b986178ce3eafa806cd984543d0da830f408"},
+ {"res/VID_1280x544_cabac_temporal_direct.264", "46ec1d702ffef31492f933afdd0b1ea155aac922"},
+ {"res/VID_1280x720_cabac_temporal_direct.264", "449e70d68b2b216d7d1cf3439d80298df3c3d50c"},
+ {"res/VID_1920x1080_cabac_temporal_direct.264", "7da13d15f660c953094e238220fdba29f29465e5"},
+ {"res/VID_1280x544_cavlc_temporal_direct.264", "290b06a9be6a991269d2d2dd42d85bd4ba02783f"},
+ {"res/VID_1280x720_cavlc_temporal_direct.264", "a339f84be1a3268638cc389f611ac4648c8c26c8"},
+ {"res/VID_1920x1080_cavlc_temporal_direct.264", "fbc5a3253180f1e24d9a2d3c4a2ae0d43630c8f9"},
};
INSTANTIATE_TEST_CASE_P (DecodeFile, DecoderOutputTest,