[Encoder] save memory usage and improve error return logic under VLCOVERFLOW
Showing the changes to codec/encoder/core/src/encoder_ext.cpp (1 of 7 files). Show all
Unified · Split
| @@ -702,7 +702,12 @@ static inline int32_t AcquireLayersNals (sWelsEncCtx** ppCtx, SWelsSvcCodingPara · 701 unmodified lines | ||
| 702 | 702 | if (iDIndex == 0) |
| 703 | 703 | iCountNumNals += MAX_SLICES_NUM; |
| 704 | 704 | // MAX_SLICES_NUM < MAX_LAYER_NUM_OF_FRAME ensured at svc_enc_slice_segment.h |
| 705 | − assert (iCountNumNals - iOrgNumNals <= MAX_NAL_UNITS_IN_LAYER); | |
| 705 | + if (iCountNumNals - iOrgNumNals > MAX_NAL_UNITS_IN_LAYER) { | |
| 706 | + WelsLog (& (*ppCtx)->sLogCtx, WELS_LOG_ERROR, | |
| 707 | + "AcquireLayersNals(), num_of_slice(%d) > existing slice(%d) at (iDid= %d), max=%d", | |
| 708 | + iCountNumNals, iOrgNumNals, iDIndex, MAX_NAL_UNITS_IN_LAYER); | |
| 709 | + return 1; | |
| 710 | + } | |
| 706 | 711 | } else { /*if ( SM_SINGLE_SLICE != pDLayer->sSliceCfg.uiSliceMode )*/ |
| 707 | 712 | const int32_t kiNumOfSlice = GetInitialSliceNum ((pDLayer->iVideoWidth + 0x0f) >> 4, |
| 708 | 713 | (pDLayer->iVideoHeight + 0x0f) >> 4, |
| @@ -1754,6 +1759,11 @@ int32_t RequestMemorySvc (sWelsEncCtx** ppCtx, SExistingParasetList* pExistingPa · 1045 unmodified lines | ||
| 1754 | 1759 | const int32_t kiPpsSize = (*ppCtx)->GetNeededPpsNum() * PPS_BUFFER_SIZE; |
| 1755 | 1760 | iNonVclLayersBsSizeCount = SSEI_BUFFER_SIZE + kiSpsSize + kiPpsSize; |
| 1756 | 1761 | |
| 1762 | + bool bDynamicSlice = false; | |
| 1763 | + uint32_t uiMaxSliceNumEstimation = 0; | |
| 1764 | + int32_t iSliceBufferSize = 0; | |
| 1765 | + int32_t iMaxSliceBufferSize = 0; | |
| 1766 | + int32_t iTotalLength = 0; | |
| 1757 | 1767 | int32_t iLayerBsSize = 0; |
| 1758 | 1768 | iIndex = 0; |
| 1759 | 1769 | while (iIndex < pParam->iSpatialLayerNum) { |
| @@ -1765,11 +1775,29 @@ int32_t RequestMemorySvc (sWelsEncCtx** ppCtx, SExistingParasetList* pExistingPa · 5 unmodified lines | ||
| 1765 | 1775 | MAX_MACROBLOCK_SIZE_IN_BYTE_x2; |
| 1766 | 1776 | iLayerBsSize = WELS_ALIGN (iLayerBsSize, 4); // 4 bytes alinged |
| 1767 | 1777 | iVclLayersBsSizeCount += iLayerBsSize; |
| 1778 | + | |
| 1779 | + SSliceConfig* pMso = & (fDlp->sSliceCfg); | |
| 1780 | + if (pMso->uiSliceMode == SM_DYN_SLICE) { | |
| 1781 | + bDynamicSlice = true; | |
| 1782 | + uiMaxSliceNumEstimation = WELS_MIN (AVERSLICENUM_CONSTRAINT, | |
| 1783 | + (iLayerBsSize / pMso->sSliceArgument.uiSliceSizeConstraint) + 1); | |
| 1784 | + (*ppCtx)->iMaxSliceCount = WELS_MAX ((*ppCtx)->iMaxSliceCount, uiMaxSliceNumEstimation); | |
| 1785 | + iSliceBufferSize = (WELS_MAX(pMso->sSliceArgument.uiSliceSizeConstraint, iLayerBsSize/uiMaxSliceNumEstimation)<<1) + MAX_MACROBLOCK_SIZE_IN_BYTE_x2; | |
| 1786 | + } else { | |
| 1787 | + (*ppCtx)->iMaxSliceCount = WELS_MAX ((*ppCtx)->iMaxSliceCount, pMso->sSliceArgument.uiSliceNum); | |
| 1788 | + iSliceBufferSize = ((iLayerBsSize / pMso->sSliceArgument.uiSliceNum)<<1) + MAX_MACROBLOCK_SIZE_IN_BYTE_x2; | |
| 1789 | + } | |
| 1790 | + iMaxSliceBufferSize = WELS_MAX(iMaxSliceBufferSize, iSliceBufferSize); | |
| 1791 | + | |
| 1768 | 1792 | ++ iIndex; |
| 1769 | 1793 | } |
| 1770 | 1794 | iTargetSpatialBsSize = iLayerBsSize; |
| 1771 | 1795 | iCountBsLen = iNonVclLayersBsSizeCount + iVclLayersBsSizeCount; |
| 1772 | 1796 | |
| 1797 | + iMaxSliceBufferSize = WELS_MIN (iMaxSliceBufferSize, iTargetSpatialBsSize); | |
| 1798 | + iTotalLength = (pParam->iMultipleThreadIdc == 1) ? iCountBsLen : (iCountBsLen + ((*ppCtx)->iMaxSliceCount - 1) * | |
| 1799 | + iMaxSliceBufferSize); | |
| 1800 | + | |
| 1773 | 1801 | pParam->iNumRefFrame = WELS_CLIP3 (pParam->iNumRefFrame, MIN_REF_PIC_COUNT, |
| 1774 | 1802 | (pParam->iUsageType == CAMERA_VIDEO_REAL_TIME ? MAX_REFERENCE_PICTURE_COUNT_NUM_CAMERA : |
| 1775 | 1803 | MAX_REFERENCE_PICTURE_COUNT_NUM_SCREEN)); |
| @@ -1787,20 +1815,14 @@ int32_t RequestMemorySvc (sWelsEncCtx** ppCtx, SExistingParasetList* pExistingPa · 11 unmodified lines | ||
| 1787 | 1815 | (*ppCtx)->pOut->iCountNals = iCountNals; |
| 1788 | 1816 | (*ppCtx)->pOut->iNalIndex = 0; |
| 1789 | 1817 | |
| 1790 | − if (pParam->iMultipleThreadIdc > 1) { | |
| 1791 | − const int32_t iTotalLength = iCountBsLen + (iTargetSpatialBsSize * ((*ppCtx)->iMaxSliceCount - 1)); | |
| 1792 | − (*ppCtx)->pFrameBs = (uint8_t*)pMa->WelsMalloc (iTotalLength, "pFrameBs"); | |
| 1793 | − WELS_VERIFY_RETURN_PROC_IF (1, (NULL == (*ppCtx)->pFrameBs), FreeMemorySvc (ppCtx)) | |
| 1794 | − (*ppCtx)->iFrameBsSize = iTotalLength; | |
| 1795 | − } else { | |
| 1796 | − (*ppCtx)->pFrameBs = (uint8_t*)pMa->WelsMalloc (iCountBsLen, "pFrameBs"); | |
| 1797 | − WELS_VERIFY_RETURN_PROC_IF (1, (NULL == (*ppCtx)->pFrameBs), FreeMemorySvc (ppCtx)) | |
| 1798 | − (*ppCtx)->iFrameBsSize = iCountBsLen; | |
| 1799 | − } | |
| 1818 | + | |
| 1819 | + (*ppCtx)->pFrameBs = (uint8_t*)pMa->WelsMalloc (iTotalLength, "pFrameBs"); | |
| 1820 | + WELS_VERIFY_RETURN_PROC_IF (1, (NULL == (*ppCtx)->pFrameBs), FreeMemorySvc (ppCtx)) | |
| 1821 | + (*ppCtx)->iFrameBsSize = iTotalLength; | |
| 1800 | 1822 | (*ppCtx)->iPosBsBuffer = 0; |
| 1801 | 1823 | |
| 1802 | 1824 | // for pSlice bs buffers |
| 1803 | − if (pParam->iMultipleThreadIdc > 1 && RequestMtResource (ppCtx, pParam, iCountBsLen, iTargetSpatialBsSize)) { | |
| 1825 | + if (pParam->iMultipleThreadIdc > 1 && RequestMtResource (ppCtx, pParam, iCountBsLen, iMaxSliceBufferSize, bDynamicSlice)) { | |
| 1804 | 1826 | WelsLog (& (*ppCtx)->sLogCtx, WELS_LOG_WARNING, "RequestMemorySvc(), RequestMtResource failed!"); |
| 1805 | 1827 | FreeMemorySvc (ppCtx); |
| 1806 | 1828 | return 1; |
| @@ -4075,7 +4097,7 @@ int32_t WelsEncoderEncodeExt (sWelsEncCtx* pCtx, SFrameBSInfo* pFbi, const SSour · 2268 unmodified lines | ||
| 4075 | 4097 | // pick up succeeding slice for threading |
| 4076 | 4098 | // thread_id equal to iEventId per implementation here |
| 4077 | 4099 | pCtx->pSliceThreading->pThreadPEncCtx[iEventId].iSliceIndex = iIndexOfSliceToBeCoded; |
| 4078 | − SetOneSliceBsBufferUnderMultithread(pCtx, iEventId, iIndexOfSliceToBeCoded); | |
| 4100 | + SetOneSliceBsBufferUnderMultithread (pCtx, iEventId, iIndexOfSliceToBeCoded); | |
| 4079 | 4101 | WelsEventSignal (&pCtx->pSliceThreading->pReadySliceCodingEvent[iEventId]); |
| 4080 | 4102 | WelsEventSignal (&pCtx->pSliceThreading->pThreadMasterEvent[iEventId]); |
| 4081 | 4103 | ++ iIndexOfSliceToBeCoded; |