second commit for threaded decoding support: Add function WelsDecodeAndConstructSlice (and add associted functions) which combines decoding->reconstructing->deblocking in single MB loop so that it allows multile frames to decode MBs in parallel.

xiaotiansf committed 7 years ago

Commit fd42f041 parent 2f5ecec22 files · +924 −190Browse files

Showing the changes to codec/decoder/core/src/decode_slice.cpp (1 of 22 files). Show all

Unified · Split

codec/decoder/core/src/decode_slice.cpp+202−35

@@ -93,7 +93,7 @@ int32_t WelsTargetSliceConstruction (PWelsDecoderContext pCtx) { · 92 unmodified lines
93	93
94	94	int32_t iTotalNumMb = pCurSlice->iTotalMbInCurSlice;
95	95	int32_t iCountNumMb = 0;
96		− PDeblockingFilterMbFunc pDeblockMb;
	96	+ PDeblockingFilterMbFunc pDeblockMb = WelsDeblockingMb;
97	97
98	98	if (!pCtx->sSpsPpsCtx.bAvcBasedFlag && iCurLayerWidth != pCtx->iCurSeqIntervalMaxPicWidth) {
99	99	return ERR_INFO_WIDTH_MISMATCH;
@@ -163,8 +163,6 @@ int32_t WelsTargetSliceConstruction (PWelsDecoderContext pCtx) { · 63 unmodified lines
163	163	if (pCtx->pParam->bParseOnly) //for parse only, deblocking should not go on
164	164	return ERR_NONE;
165	165
166		− pDeblockMb = WelsDeblockingMb;
167		−
168	166	if (1 == pSliceHeader->uiDisableDeblockingFilterIdc
169	167	\|\| pCtx->pCurDqLayer->sLayerInfo.sSliceInLayer.iTotalMbInCurSlice <= 0) {
170	168	return ERR_NONE;//NO_SUPPORTED_FILTER_IDX
@@ -238,8 +236,10 @@ int32_t WelsMbInterConstruction (PWelsDecoderContext pCtx, PDqLayer pCurDqLayer) · 67 unmodified lines
238	236	}
239	237	WelsMbInterSampleConstruction (pCtx, pCurDqLayer, pDstY, pDstCb, pDstCr, iLumaStride, iChromaStride);
240	238
241		− pCtx->sBlockFunc.pWelsSetNonZeroCountFunc (
242		− pCurDqLayer->pNzc[pCurDqLayer->iMbXyIndex]); // set all none-zero nzc to 1; dbk can be opti!
	239	+ if (pCtx->pThreadCtx == NULL) {
	240	+ pCtx->sBlockFunc.pWelsSetNonZeroCountFunc (
	241	+ pCurDqLayer->pNzc[pCurDqLayer->iMbXyIndex]); // set all none-zero nzc to 1; dbk can be opti!
	242	+ }
243	243	return ERR_NONE;
244	244	}
245	245
@@ -723,7 +723,7 @@ int32_t WelsDecodeMbCabacISliceBaseMode0 (PWelsDecoderContext pCtx, uint32_t& ui · 477 unmodified lines
723	723	pCurDqLayer->pLumaQp[iMbXy] = pSlice->iLastMbQp;
724	724	for (i = 0; i < 2; i++) {
725	725	pCurDqLayer->pChromaQp[iMbXy][i] = g_kuiChromaQpTable[WELS_CLIP3 ((pCurDqLayer->pLumaQp[iMbXy] +
726		− pSliceHeader->pPps->iChromaQpIndexOffset[i]), 0, 51)];
	726	+ pSliceHeader->pPps->iChromaQpIndexOffset[i]), 0, 51)];
727	727	}
728	728	}
729	729
@@ -738,7 +738,7 @@ int32_t WelsDecodeMbCabacISliceBaseMode0 (PWelsDecoderContext pCtx, uint32_t& ui · 8 unmodified lines
738	738	pSlice->iLastMbQp = pCurDqLayer->pLumaQp[iMbXy];
739	739	for (i = 0; i < 2; i++) {
740	740	pCurDqLayer->pChromaQp[iMbXy][i] = g_kuiChromaQpTable[WELS_CLIP3 ((pSlice->iLastMbQp +
741		− pSliceHeader->pPps->iChromaQpIndexOffset[i]), 0, 51)];
	741	+ pSliceHeader->pPps->iChromaQpIndexOffset[i]), 0, 51)];
742	742	}
743	743	if (MB_TYPE_INTRA16x16 == pCurDqLayer->pDec->pMbType[iMbXy]) {
744	744	//step1: Luma DC
@@ -973,7 +973,7 @@ int32_t WelsDecodeMbCabacPSliceBaseMode0 (PWelsDecoderContext pCtx, PWelsNeighAv · 228 unmodified lines
973	973	pSlice->iLastMbQp = pCurDqLayer->pLumaQp[iMbXy];
974	974	for (i = 0; i < 2; i++) {
975	975	pCurDqLayer->pChromaQp[iMbXy][i] = g_kuiChromaQpTable[WELS_CLIP3 (pSlice->iLastMbQp +
976		− pSliceHeader->pPps->iChromaQpIndexOffset[i], 0, 51)];
	976	+ pSliceHeader->pPps->iChromaQpIndexOffset[i], 0, 51)];
977	977	}
978	978
979	979	if (MB_TYPE_INTRA16x16 == pCurDqLayer->pDec->pMbType[iMbXy]) {
@@ -1080,7 +1080,7 @@ int32_t WelsDecodeMbCabacPSliceBaseMode0 (PWelsDecoderContext pCtx, PWelsNeighAv · 100 unmodified lines
1080	1080	pCurDqLayer->pLumaQp[iMbXy] = pSlice->iLastMbQp;
1081	1081	for (i = 0; i < 2; i++) {
1082	1082	pCurDqLayer->pChromaQp[iMbXy][i] = g_kuiChromaQpTable[WELS_CLIP3 (pCurDqLayer->pLumaQp[iMbXy] +
1083		− pSliceHeader->pPps->iChromaQpIndexOffset[i], 0, 51)];
	1083	+ pSliceHeader->pPps->iChromaQpIndexOffset[i], 0, 51)];
1084	1084	}
1085	1085	}
1086	1086
@@ -1214,7 +1214,7 @@ int32_t WelsDecodeMbCabacBSliceBaseMode0 (PWelsDecoderContext pCtx, PWelsNeighAv · 127 unmodified lines
1214	1214	pSlice->iLastMbQp = pCurDqLayer->pLumaQp[iMbXy];
1215	1215	for (i = 0; i < 2; i++) {
1216	1216	pCurDqLayer->pChromaQp[iMbXy][i] = g_kuiChromaQpTable[WELS_CLIP3 (pSlice->iLastMbQp +
1217		− pSliceHeader->pPps->iChromaQpIndexOffset[i], 0, 51)];
	1217	+ pSliceHeader->pPps->iChromaQpIndexOffset[i], 0, 51)];
1218	1218	}
1219	1219
1220	1220	if (MB_TYPE_INTRA16x16 == pCurDqLayer->pDec->pMbType[iMbXy]) {
@@ -1321,7 +1321,7 @@ int32_t WelsDecodeMbCabacBSliceBaseMode0 (PWelsDecoderContext pCtx, PWelsNeighAv · 100 unmodified lines
1321	1321	pCurDqLayer->pLumaQp[iMbXy] = pSlice->iLastMbQp;
1322	1322	for (i = 0; i < 2; i++) {
1323	1323	pCurDqLayer->pChromaQp[iMbXy][i] = g_kuiChromaQpTable[WELS_CLIP3 (pCurDqLayer->pLumaQp[iMbXy] +
1324		− pSliceHeader->pPps->iChromaQpIndexOffset[i], 0, 51)];
	1324	+ pSliceHeader->pPps->iChromaQpIndexOffset[i], 0, 51)];
1325	1325	}
1326	1326	}
1327	1327
@@ -1365,7 +1365,9 @@ int32_t WelsDecodeMbCabacPSlice (PWelsDecoderContext pCtx, PNalUnit pNalCur, uin · 37 unmodified lines
1365	1365
1366	1366	pCurDqLayer->pInterPredictionDoneFlag[iMbXy] = 0;
1367	1367	memset (pCurDqLayer->pDec->pRefIndex[0][iMbXy], 0, sizeof (int8_t) * 16);
1368		− pCtx->bMbRefConcealed = pCtx->bRPLRError \|\| pCtx->bMbRefConcealed \|\| ! (ppRefPic[0] && ppRefPic[0]->bIsComplete);
	1368	+ bool bIsPending = pCtx->pThreadCtx != NULL;
	1369	+ pCtx->bMbRefConcealed = pCtx->bRPLRError \|\| pCtx->bMbRefConcealed \|\| ! (ppRefPic[0] && (ppRefPic[0]->bIsComplete
	1370	+ \|\| bIsPending));
1369	1371	//predict mv
1370	1372	PredPSkipMvFromNeighbor (pCurDqLayer, pMv);
1371	1373	for (i = 0; i < 16; i++) {
@@ -1381,7 +1383,7 @@ int32_t WelsDecodeMbCabacPSlice (PWelsDecoderContext pCtx, PNalUnit pNalCur, uin · 9 unmodified lines
1381	1383	pCurDqLayer->pLumaQp[iMbXy] = pSlice->iLastMbQp; //??????????????? dqaunt of previous mb
1382	1384	for (i = 0; i < 2; i++) {
1383	1385	pCurDqLayer->pChromaQp[iMbXy][i] = g_kuiChromaQpTable[WELS_CLIP3 (pCurDqLayer->pLumaQp[iMbXy] +
1384		− pSliceHeader->pPps->iChromaQpIndexOffset[i], 0, 51)];
	1386	+ pSliceHeader->pPps->iChromaQpIndexOffset[i], 0, 51)];
1385	1387	}
1386	1388
1387	1389	//for neighboring CABAC usage
@@ -1419,6 +1421,8 @@ int32_t WelsDecodeMbCabacBSlice (PWelsDecoderContext pCtx, PNalUnit pNalCur, uin · 31 unmodified lines
1419	1421
1420	1422	memset (pCurDqLayer->pDirect[iMbXy], 0, sizeof (int8_t) * 16);
1421	1423
	1424	+ bool bIsPending = pCtx->pThreadCtx != NULL;
	1425	+
1422	1426	if (uiCode) {
1423	1427	int16_t pMv[LIST_A][2] = { {0, 0}, { 0, 0 } };
1424	1428	int8_t ref[LIST_A] = { 0 };
@@ -1433,8 +1437,8 @@ int32_t WelsDecodeMbCabacBSlice (PWelsDecoderContext pCtx, PNalUnit pNalCur, uin · 8 unmodified lines
1433	1437	pCurDqLayer->pInterPredictionDoneFlag[iMbXy] = 0;
1434	1438	memset (pCurDqLayer->pDec->pRefIndex[LIST_0][iMbXy], 0, sizeof (int8_t) * 16);
1435	1439	memset (pCurDqLayer->pDec->pRefIndex[LIST_1][iMbXy], 0, sizeof (int8_t) * 16);
1436		− pCtx->bMbRefConcealed = pCtx->bRPLRError \|\| pCtx->bMbRefConcealed \|\| ! (ppRefPicL0[0] && ppRefPicL0[0]->bIsComplete)
1437		− \|\| ! (ppRefPicL1[0] && ppRefPicL1[0]->bIsComplete);
	1440	+ pCtx->bMbRefConcealed = pCtx->bRPLRError \|\| pCtx->bMbRefConcealed \|\| ! (ppRefPicL0[0] && (ppRefPicL0[0]->bIsComplete
	1441	+ \|\| bIsPending)) \|\| ! (ppRefPicL1[0] && (ppRefPicL1[0]->bIsComplete \|\| bIsPending));
1438	1442
1439	1443	if (pCtx->bMbRefConcealed) {
1440	1444	SLogContext* pLogCtx = & (pCtx->sLogCtx);
@@ -1463,7 +1467,7 @@ int32_t WelsDecodeMbCabacBSlice (PWelsDecoderContext pCtx, PNalUnit pNalCur, uin · 22 unmodified lines
1463	1467	pCurDqLayer->pLumaQp[iMbXy] = pSlice->iLastMbQp; //??????????????? dqaunt of previous mb
1464	1468	for (i = 0; i < 2; i++) {
1465	1469	pCurDqLayer->pChromaQp[iMbXy][i] = g_kuiChromaQpTable[WELS_CLIP3 (pCurDqLayer->pLumaQp[iMbXy] +
1466		− pSliceHeader->pPps->iChromaQpIndexOffset[i], 0, 51)];
	1470	+ pSliceHeader->pPps->iChromaQpIndexOffset[i], 0, 51)];
1467	1471	}
1468	1472
1469	1473	//for neighboring CABAC usage
@@ -1613,6 +1617,161 @@ int32_t WelsDecodeSlice (PWelsDecoderContext pCtx, bool bFirstSliceInLayer, PNal · 143 unmodified lines
1613	1617	return ERR_NONE;
1614	1618	}
1615	1619
	1620	+int32_t WelsDecodeAndConstructSlice (PWelsDecoderContext pCtx) {
	1621	+ PNalUnit pNalCur = pCtx->pNalCur;
	1622	+ PDqLayer pCurDqLayer = pCtx->pCurDqLayer;
	1623	+ PFmo pFmo = pCtx->pFmo;
	1624	+ int32_t iRet;
	1625	+ int32_t iNextMbXyIndex, iSliceIdc;
	1626	+
	1627	+ PSlice pSlice = &pCurDqLayer->sLayerInfo.sSliceInLayer;
	1628	+ PSliceHeaderExt pSliceHeaderExt = &pSlice->sSliceHeaderExt;
	1629	+ PSliceHeader pSliceHeader = &pSliceHeaderExt->sSliceHeader;
	1630	+ int32_t iMbX, iMbY;
	1631	+ const int32_t kiCountNumMb = pSliceHeader->pSps->uiTotalMbCount; //need to be correct when fmo or multi slice
	1632	+ int32_t iTotalMbTargetLayer = kiCountNumMb;
	1633	+ uint32_t uiEosFlag = 0;
	1634	+ PWelsDecMbFunc pDecMbFunc;
	1635	+
	1636	+ pSlice->iTotalMbInCurSlice = 0; //initialize at the starting of slice decoding.
	1637	+
	1638	+ if (pCtx->pPps->bEntropyCodingModeFlag) {
	1639	+ if (pSlice->sSliceHeaderExt.bAdaptiveMotionPredFlag \|\|
	1640	+ pSlice->sSliceHeaderExt.bAdaptiveBaseModeFlag \|\|
	1641	+ pSlice->sSliceHeaderExt.bAdaptiveResidualPredFlag) {
	1642	+ WelsLog (& (pCtx->sLogCtx), WELS_LOG_ERROR,
	1643	+ "WelsDecodeSlice()::::ILP flag exist, not supported with CABAC enabled!");
	1644	+ pCtx->iErrorCode \|= dsBitstreamError;
	1645	+ return dsBitstreamError;
	1646	+ }
	1647	+ if (P_SLICE == pSliceHeader->eSliceType)
	1648	+ pDecMbFunc = WelsDecodeMbCabacPSlice;
	1649	+ else if (B_SLICE == pSliceHeader->eSliceType)
	1650	+ pDecMbFunc = WelsDecodeMbCabacBSlice;
	1651	+ else //I_SLICE. B_SLICE is being supported
	1652	+ pDecMbFunc = WelsDecodeMbCabacISlice;
	1653	+ } else {
	1654	+ if (P_SLICE == pSliceHeader->eSliceType) {
	1655	+ pDecMbFunc = WelsDecodeMbCavlcPSlice;
	1656	+ } else if (B_SLICE == pSliceHeader->eSliceType) {
	1657	+ pDecMbFunc = WelsDecodeMbCavlcBSlice;
	1658	+ } else { //I_SLICE
	1659	+ pDecMbFunc = WelsDecodeMbCavlcISlice;
	1660	+ }
	1661	+ }
	1662	+
	1663	+ if (pSliceHeader->pPps->bConstainedIntraPredFlag) {
	1664	+ pCtx->pFillInfoCacheIntraNxNFunc = WelsFillCacheConstrain1IntraNxN;
	1665	+ pCtx->pMapNxNNeighToSampleFunc = WelsMapNxNNeighToSampleConstrain1;
	1666	+ pCtx->pMap16x16NeighToSampleFunc = WelsMap16x16NeighToSampleConstrain1;
	1667	+ } else {
	1668	+ pCtx->pFillInfoCacheIntraNxNFunc = WelsFillCacheConstrain0IntraNxN;
	1669	+ pCtx->pMapNxNNeighToSampleFunc = WelsMapNxNNeighToSampleNormal;
	1670	+ pCtx->pMap16x16NeighToSampleFunc = WelsMap16x16NeighToSampleNormal;
	1671	+ }
	1672	+
	1673	+ pCtx->eSliceType = pSliceHeader->eSliceType;
	1674	+ if (pCurDqLayer->sLayerInfo.pPps->bEntropyCodingModeFlag == 1) {
	1675	+ int32_t iQp = pSlice->sSliceHeaderExt.sSliceHeader.iSliceQp;
	1676	+ int32_t iCabacInitIdc = pSlice->sSliceHeaderExt.sSliceHeader.iCabacInitIdc;
	1677	+ WelsCabacContextInit (pCtx, pSlice->eSliceType, iCabacInitIdc, iQp);
	1678	+ //InitCabacCtx (pCtx->pCabacCtx, pSlice->eSliceType, iCabacInitIdc, iQp);
	1679	+ pSlice->iLastDeltaQp = 0;
	1680	+ WELS_READ_VERIFY (InitCabacDecEngineFromBS (pCtx->pCabacDecEngine, pCtx->pCurDqLayer->pBitStringAux));
	1681	+ }
	1682	+ //try to calculate the dequant_coeff
	1683	+ WelsCalcDeqCoeffScalingList (pCtx);
	1684	+
	1685	+ iNextMbXyIndex = pSliceHeader->iFirstMbInSlice;
	1686	+ iMbX = iNextMbXyIndex % pCurDqLayer->iMbWidth;
	1687	+ iMbY = iNextMbXyIndex / pCurDqLayer->iMbWidth; // error is introduced by multiple slices case, 11/23/2009
	1688	+ pSlice->iMbSkipRun = -1;
	1689	+ iSliceIdc = (pSliceHeader->iFirstMbInSlice << 7) + pCurDqLayer->uiLayerDqId;
	1690	+
	1691	+ pCurDqLayer->iMbX = iMbX;
	1692	+ pCurDqLayer->iMbY = iMbY;
	1693	+ pCurDqLayer->iMbXyIndex = iNextMbXyIndex;
	1694	+
	1695	+ PDeblockingFilterMbFunc pDeblockMb = WelsDeblockingMb;
	1696	+
	1697	+ SDeblockingFilter pFilter;
	1698	+ int32_t iFilterIdc = 1;
	1699	+ if (pCtx->pThreadCtx && pSliceHeader->uiDisableDeblockingFilterIdc != 1) {
	1700	+ WelsDeblockingInitFilter (pCtx, pFilter, iFilterIdc);
	1701	+ }
	1702	+
	1703	+ do {
	1704	+ if ((-1 == iNextMbXyIndex) \|\| (iNextMbXyIndex >= kiCountNumMb)) { // slice group boundary or end of a frame
	1705	+ break;
	1706	+ }
	1707	+
	1708	+ pCurDqLayer->pSliceIdc[iNextMbXyIndex] = iSliceIdc;
	1709	+ pCtx->bMbRefConcealed = false;
	1710	+ iRet = pDecMbFunc (pCtx, pNalCur, uiEosFlag);
	1711	+ pCurDqLayer->pMbRefConcealedFlag[iNextMbXyIndex] = pCtx->bMbRefConcealed;
	1712	+ if (iRet != ERR_NONE) {
	1713	+ return iRet;
	1714	+ }
	1715	+ if (WelsTargetMbConstruction (pCtx)) {
	1716	+ WelsLog (& (pCtx->sLogCtx), WELS_LOG_WARNING,
	1717	+ "WelsTargetSliceConstruction():::MB(%d, %d) construction error. pCurSlice_type:%d",
	1718	+ pCurDqLayer->iMbX, pCurDqLayer->iMbY, pSlice->eSliceType);
	1719	+
	1720	+ return ERR_INFO_MB_RECON_FAIL;
	1721	+ }
	1722	+ WelsDeblockingFilterMB (pCurDqLayer, pFilter, iFilterIdc, pDeblockMb);
	1723	+ if (pCtx->uiNalRefIdc > 0) {
	1724	+ if (pCurDqLayer->iMbX == 0 \|\| pCurDqLayer->iMbX == pCurDqLayer->iMbWidth - 1 \|\| pCurDqLayer->iMbY == 0
	1725	+ \|\| pCurDqLayer->iMbY == pCurDqLayer->iMbHeight - 1) {
	1726	+ PadMBLuma_c (pCurDqLayer->pDec->pData[0], pCurDqLayer->pDec->iLinesize[0], pCurDqLayer->pDec->iWidthInPixel,
	1727	+ pCurDqLayer->pDec->iHeightInPixel, pCurDqLayer->iMbX, pCurDqLayer->iMbY, pCurDqLayer->iMbWidth, pCurDqLayer->iMbHeight);
	1728	+ PadMBChroma_c (pCurDqLayer->pDec->pData[1], pCurDqLayer->pDec->iLinesize[1], pCurDqLayer->pDec->iWidthInPixel / 2,
	1729	+ pCurDqLayer->pDec->iHeightInPixel / 2, pCurDqLayer->iMbX, pCurDqLayer->iMbY, pCurDqLayer->iMbWidth,
	1730	+ pCurDqLayer->iMbHeight);
	1731	+ PadMBChroma_c (pCurDqLayer->pDec->pData[2], pCurDqLayer->pDec->iLinesize[2], pCurDqLayer->pDec->iWidthInPixel / 2,
	1732	+ pCurDqLayer->pDec->iHeightInPixel / 2, pCurDqLayer->iMbX, pCurDqLayer->iMbY, pCurDqLayer->iMbWidth,
	1733	+ pCurDqLayer->iMbHeight);
	1734	+ }
	1735	+ }
	1736	+ if (!pCurDqLayer->pMbCorrectlyDecodedFlag[iNextMbXyIndex]) { //already con-ed, overwrite
	1737	+ pCurDqLayer->pMbCorrectlyDecodedFlag[iNextMbXyIndex] = true;
	1738	+ pCtx->pDec->iMbEcedPropNum += (pCurDqLayer->pMbRefConcealedFlag[iNextMbXyIndex] ? 1 : 0);
	1739	+ ++pCtx->iTotalNumMbRec;
	1740	+ }
	1741	+
	1742	+ if (pCtx->iTotalNumMbRec > iTotalMbTargetLayer) {
	1743	+ WelsLog (& (pCtx->sLogCtx), WELS_LOG_WARNING,
	1744	+ "WelsTargetSliceConstruction():::pCtx->iTotalNumMbRec:%d, iTotalMbTargetLayer:%d",
	1745	+ pCtx->iTotalNumMbRec, iTotalMbTargetLayer);
	1746	+
	1747	+ return ERR_INFO_MB_NUM_EXCEED_FAIL;
	1748	+ }
	1749	+
	1750	+ ++pSlice->iTotalMbInCurSlice;
	1751	+ if (uiEosFlag) { //end of slice
	1752	+ SET_EVENT (&pCtx->pDec->pReadyEvent[pCurDqLayer->iMbY]);
	1753	+ break;
	1754	+ }
	1755	+ if (pSliceHeader->pPps->uiNumSliceGroups > 1) {
	1756	+ iNextMbXyIndex = FmoNextMb (pFmo, iNextMbXyIndex);
	1757	+ } else {
	1758	+ ++iNextMbXyIndex;
	1759	+ }
	1760	+ int32_t iLastMby = iMbY;
	1761	+ int32_t iLastMbx = iMbX;
	1762	+ iMbX = iNextMbXyIndex % pCurDqLayer->iMbWidth;
	1763	+ iMbY = iNextMbXyIndex / pCurDqLayer->iMbWidth;
	1764	+ pCurDqLayer->iMbX = iMbX;
	1765	+ pCurDqLayer->iMbY = iMbY;
	1766	+ pCurDqLayer->iMbXyIndex = iNextMbXyIndex;
	1767	+ if ((iMbY > iLastMby) && (iLastMbx == pCurDqLayer->iMbWidth - 1)) {
	1768	+ SET_EVENT (&pCtx->pDec->pReadyEvent[iLastMby]);
	1769	+ }
	1770	+ } while (1);
	1771	+ SET_EVENT (&pCtx->pDec->pReadyEvent[pCurDqLayer->iMbY]);
	1772	+ return ERR_NONE;
	1773	+}
	1774	+
1616	1775	int32_t WelsActualDecodeMbCavlcISlice (PWelsDecoderContext pCtx) {
1617	1776	SVlcTable* pVlcTable = pCtx->pVlcTable;
1618	1777	PDqLayer pCurDqLayer = pCtx->pCurDqLayer;
@@ -1762,7 +1921,7 @@ int32_t WelsActualDecodeMbCavlcISlice (PWelsDecoderContext pCtx) { · 143 unmodified lines
1762	1921	pCurDqLayer->pLumaQp[iMbXy] = pSlice->iLastMbQp;
1763	1922	for (i = 0; i < 2; i++) {
1764	1923	pCurDqLayer->pChromaQp[iMbXy][i] = g_kuiChromaQpTable[WELS_CLIP3 (pCurDqLayer->pLumaQp[iMbXy] +
1765		− pSliceHeader->pPps->iChromaQpIndexOffset[i], 0, 51)];
	1924	+ pSliceHeader->pPps->iChromaQpIndexOffset[i], 0, 51)];
1766	1925	}
1767	1926
1768	1927	}
@@ -1782,8 +1941,8 @@ int32_t WelsActualDecodeMbCavlcISlice (PWelsDecoderContext pCtx) { · 13 unmodified lines
1782	1941	pSlice->iLastMbQp = pCurDqLayer->pLumaQp[iMbXy];
1783	1942	for (i = 0; i < 2; i++) {
1784	1943	pCurDqLayer->pChromaQp[iMbXy][i] = g_kuiChromaQpTable[WELS_CLIP3 (pSlice->iLastMbQp +
1785		− pSliceHeader->pPps->iChromaQpIndexOffset[i], 0,
1786		− 51)];
	1944	+ pSliceHeader->pPps->iChromaQpIndexOffset[i], 0,
	1945	+ 51)];
1787	1946	}
1788	1947
1789	1948
@@ -1876,7 +2035,8 @@ int32_t WelsActualDecodeMbCavlcISlice (PWelsDecoderContext pCtx) { · 86 unmodified lines
1876	2035	int32_t iIndex = 16 + (i << 2);
1877	2036	for (iId4x4 = 0; iId4x4 < 4; iId4x4++) {
1878	2037	if ((iRet = WelsResidualBlockCavlc (pVlcTable, pNonZeroCount, pBs, iIndex, iScanIdxEnd - WELS_MAX (iScanIdxStart,
1879		− 1) + 1, g_kuiZigzagScan + WELS_MAX (iScanIdxStart, 1), iMbResProperty, pCurDqLayer->pScaledTCoeff[iMbXy] + (iIndex << 4),
	2038	+ 1) + 1, g_kuiZigzagScan + WELS_MAX (iScanIdxStart, 1), iMbResProperty,
	2039	+ pCurDqLayer->pScaledTCoeff[iMbXy] + (iIndex << 4),
1880	2040	pCurDqLayer->pChromaQp[iMbXy][i], pCtx)) != ERR_NONE) {
1881	2041	return iRet;//abnormal
1882	2042	}
@@ -2126,11 +2286,12 @@ int32_t WelsActualDecodeMbCavlcPSlice (PWelsDecoderContext pCtx) { · 243 unmodified lines
2126	2286	ST32A4 (&pNzc[12], 0);
2127	2287	ST32A4 (&pNzc[16], 0);
2128	2288	ST32A4 (&pNzc[20], 0);
2129		− if (pCurDqLayer->pCbp[iMbXy] == 0 && !IS_INTRA16x16 (pCurDqLayer->pDec->pMbType[iMbXy]) && !IS_I_BL (pCurDqLayer->pDec->pMbType[iMbXy])) {
	2289	+ if (pCurDqLayer->pCbp[iMbXy] == 0 && !IS_INTRA16x16 (pCurDqLayer->pDec->pMbType[iMbXy])
	2290	+ && !IS_I_BL (pCurDqLayer->pDec->pMbType[iMbXy])) {
2130	2291	pCurDqLayer->pLumaQp[iMbXy] = pSlice->iLastMbQp;
2131	2292	for (i = 0; i < 2; i++) {
2132	2293	pCurDqLayer->pChromaQp[iMbXy][i] = g_kuiChromaQpTable[WELS_CLIP3 (pCurDqLayer->pLumaQp[iMbXy] +
2133		− pSliceHeader->pPps->iChromaQpIndexOffset[i], 0, 51)];
	2294	+ pSliceHeader->pPps->iChromaQpIndexOffset[i], 0, 51)];
2134	2295	}
2135	2296	}
2136	2297
@@ -2148,8 +2309,8 @@ int32_t WelsActualDecodeMbCavlcPSlice (PWelsDecoderContext pCtx) { · 11 unmodified lines
2148	2309	pSlice->iLastMbQp = pCurDqLayer->pLumaQp[iMbXy];
2149	2310	for (i = 0; i < 2; i++) {
2150	2311	pCurDqLayer->pChromaQp[iMbXy][i] = g_kuiChromaQpTable[WELS_CLIP3 (pSlice->iLastMbQp +
2151		− pSliceHeader->pPps->iChromaQpIndexOffset[i], 0,
2152		− 51)];
	2312	+ pSliceHeader->pPps->iChromaQpIndexOffset[i], 0,
	2313	+ 51)];
2153	2314	}
2154	2315
2155	2316	BsStartCavlc (pBs);
@@ -2251,7 +2412,8 @@ int32_t WelsActualDecodeMbCavlcPSlice (PWelsDecoderContext pCtx) { · 95 unmodified lines
2251	2412	int32_t iIndex = 16 + (i << 2);
2252	2413	for (iId4x4 = 0; iId4x4 < 4; iId4x4++) {
2253	2414	if ((iRet = WelsResidualBlockCavlc (pVlcTable, pNonZeroCount, pBs, iIndex, iScanIdxEnd - WELS_MAX (iScanIdxStart,
2254		− 1) + 1, g_kuiZigzagScan + WELS_MAX (iScanIdxStart, 1), iMbResProperty, pCurDqLayer->pScaledTCoeff[iMbXy] + (iIndex << 4),
	2415	+ 1) + 1, g_kuiZigzagScan + WELS_MAX (iScanIdxStart, 1), iMbResProperty,
	2416	+ pCurDqLayer->pScaledTCoeff[iMbXy] + (iIndex << 4),
2255	2417	pCurDqLayer->pChromaQp[iMbXy][i], pCtx)) != ERR_NONE) {
2256	2418	return iRet;//abnormal
2257	2419	}
@@ -2305,7 +2467,9 @@ int32_t WelsDecodeMbCavlcPSlice (PWelsDecoderContext pCtx, PNalUnit pNalCur, uin · 47 unmodified lines
2305	2467
2306	2468	pCurDqLayer->pInterPredictionDoneFlag[iMbXy] = 0;
2307	2469	memset (pCurDqLayer->pDec->pRefIndex[0][iMbXy], 0, sizeof (int8_t) * 16);
2308		− pCtx->bMbRefConcealed = pCtx->bRPLRError \|\| pCtx->bMbRefConcealed \|\| ! (ppRefPic[0] && ppRefPic[0]->bIsComplete);
	2470	+ bool bIsPending = pCtx->pThreadCtx != NULL;
	2471	+ pCtx->bMbRefConcealed = pCtx->bRPLRError \|\| pCtx->bMbRefConcealed \|\| ! (ppRefPic[0] && (ppRefPic[0]->bIsComplete
	2472	+ \|\| bIsPending));
2309	2473	//predict iMv
2310	2474	PredPSkipMvFromNeighbor (pCurDqLayer, iMv);
2311	2475	for (i = 0; i < 16; i++) {
@@ -2322,7 +2486,7 @@ int32_t WelsDecodeMbCavlcPSlice (PWelsDecoderContext pCtx, PNalUnit pNalCur, uin · 10 unmodified lines
2322	2486	pCurDqLayer->pLumaQp[iMbXy] = pSlice->iLastMbQp;
2323	2487	for (i = 0; i < 2; i++) {
2324	2488	pCurDqLayer->pChromaQp[iMbXy][i] = g_kuiChromaQpTable[WELS_CLIP3 (pCurDqLayer->pLumaQp[iMbXy] +
2325		− pSliceHeader->pPps->iChromaQpIndexOffset[i], 0, 51)];
	2489	+ pSliceHeader->pPps->iChromaQpIndexOffset[i], 0, 51)];
2326	2490	}
2327	2491	}
2328	2492
@@ -2400,8 +2564,9 @@ int32_t WelsDecodeMbCavlcBSlice (PWelsDecoderContext pCtx, PNalUnit pNalCur, uin · 71 unmodified lines
2400	2564	pCurDqLayer->pInterPredictionDoneFlag[iMbXy] = 0;
2401	2565	memset (pCurDqLayer->pDec->pRefIndex[LIST_0][iMbXy], 0, sizeof (int8_t) * 16);
2402	2566	memset (pCurDqLayer->pDec->pRefIndex[LIST_1][iMbXy], 0, sizeof (int8_t) * 16);
2403		− pCtx->bMbRefConcealed = pCtx->bRPLRError \|\| pCtx->bMbRefConcealed \|\| ! (ppRefPicL0[0] && ppRefPicL0[0]->bIsComplete)
2404		− \|\| ! (ppRefPicL1[0] && ppRefPicL1[0]->bIsComplete);
	2567	+ bool bIsPending = pCtx->pThreadCtx != NULL;
	2568	+ pCtx->bMbRefConcealed = pCtx->bRPLRError \|\| pCtx->bMbRefConcealed \|\| ! (ppRefPicL0[0] && (ppRefPicL0[0]->bIsComplete
	2569	+ \|\| bIsPending)) \|\| ! (ppRefPicL1[0] && (ppRefPicL1[0]->bIsComplete \|\| bIsPending));
2405	2570
2406	2571	/*if (pCtx->bMbRefConcealed) {
2407	2572	SLogContext* pLogCtx = & (pCtx->sLogCtx);
@@ -2435,7 +2600,7 @@ int32_t WelsDecodeMbCavlcBSlice (PWelsDecoderContext pCtx, PNalUnit pNalCur, uin · 27 unmodified lines
2435	2600	pCurDqLayer->pLumaQp[iMbXy] = pSlice->iLastMbQp;
2436	2601	for (i = 0; i < 2; i++) {
2437	2602	pCurDqLayer->pChromaQp[iMbXy][i] = g_kuiChromaQpTable[WELS_CLIP3 (pCurDqLayer->pLumaQp[iMbXy] +
2438		− pSliceHeader->pPps->iChromaQpIndexOffset[i], 0, 51)];
	2603	+ pSliceHeader->pPps->iChromaQpIndexOffset[i], 0, 51)];
2439	2604	}
2440	2605	}
2441	2606
@@ -2665,11 +2830,12 @@ int32_t WelsActualDecodeMbCavlcBSlice (PWelsDecoderContext pCtx) { · 223 unmodified lines
2665	2830	ST32A4 (&pNzc[12], 0);
2666	2831	ST32A4 (&pNzc[16], 0);
2667	2832	ST32A4 (&pNzc[20], 0);
2668		− if (pCurDqLayer->pCbp[iMbXy] == 0 && !IS_INTRA16x16 (pCurDqLayer->pDec->pMbType[iMbXy]) && !IS_I_BL (pCurDqLayer->pDec->pMbType[iMbXy])) {
	2833	+ if (pCurDqLayer->pCbp[iMbXy] == 0 && !IS_INTRA16x16 (pCurDqLayer->pDec->pMbType[iMbXy])
	2834	+ && !IS_I_BL (pCurDqLayer->pDec->pMbType[iMbXy])) {
2669	2835	pCurDqLayer->pLumaQp[iMbXy] = pSlice->iLastMbQp;
2670	2836	for (i = 0; i < 2; i++) {
2671	2837	pCurDqLayer->pChromaQp[iMbXy][i] = g_kuiChromaQpTable[WELS_CLIP3 (pCurDqLayer->pLumaQp[iMbXy] +
2672		− pSliceHeader->pPps->iChromaQpIndexOffset[i], 0, 51)];
	2838	+ pSliceHeader->pPps->iChromaQpIndexOffset[i], 0, 51)];
2673	2839	}
2674	2840	}
2675	2841
@@ -2687,8 +2853,8 @@ int32_t WelsActualDecodeMbCavlcBSlice (PWelsDecoderContext pCtx) { · 11 unmodified lines
2687	2853	pSlice->iLastMbQp = pCurDqLayer->pLumaQp[iMbXy];
2688	2854	for (i = 0; i < 2; i++) {
2689	2855	pCurDqLayer->pChromaQp[iMbXy][i] = g_kuiChromaQpTable[WELS_CLIP3 (pSlice->iLastMbQp +
2690		− pSliceHeader->pPps->iChromaQpIndexOffset[i], 0,
2691		− 51)];
	2856	+ pSliceHeader->pPps->iChromaQpIndexOffset[i], 0,
	2857	+ 51)];
2692	2858	}
2693	2859
2694	2860	BsStartCavlc (pBs);
@@ -2790,7 +2956,8 @@ int32_t WelsActualDecodeMbCavlcBSlice (PWelsDecoderContext pCtx) { · 95 unmodified lines
2790	2956	int32_t iIndex = 16 + (i << 2);
2791	2957	for (iId4x4 = 0; iId4x4 < 4; iId4x4++) {
2792	2958	if ((iRet = WelsResidualBlockCavlc (pVlcTable, pNonZeroCount, pBs, iIndex, iScanIdxEnd - WELS_MAX (iScanIdxStart,
2793		− 1) + 1, g_kuiZigzagScan + WELS_MAX (iScanIdxStart, 1), iMbResProperty, pCurDqLayer->pScaledTCoeff[iMbXy] + (iIndex << 4),
	2959	+ 1) + 1, g_kuiZigzagScan + WELS_MAX (iScanIdxStart, 1), iMbResProperty,
	2960	+ pCurDqLayer->pScaledTCoeff[iMbXy] + (iIndex << 4),
2794	2961	pCurDqLayer->pChromaQp[iMbXy][i], pCtx)) != ERR_NONE) {
2795	2962	return iRet;//abnormal
2796	2963	}

second commit for threaded decoding support: Add function WelsDecodeAndConstructSlice (and add associted functions) which combines decoding->reconstructing->deblocking in single MB loop so that it allows multile frames to decode MBs in parallel.

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code