generic-library/openh264
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

replace conditional judgements with function pointer for acceleration

Browse Source
This commit is contained in:
Sijia Chen 2015-03-02 17:11:11 +08:00
parent 2da0ce6451
commit 303bf77e42
3 changed files with 7 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
codec/encoder/core/inc/md.h
View File

@ -122,6 +122,7 @@ uint8_t* pHalfPixHV;
uint8_t* pQuarPixBest;
uint8_t* pQuarPixTmp;
PCopyFunc pfCopyBlockByMode;
} SMeRefinePointer;
void FillNeighborCacheIntra (SMbCache* pMbCache, SMB* pCurMb, int32_t iMbWidth/*, bool constrained_intra_pred_flag*/);

13
codec/encoder/core/src/md.cpp
View File

@ -765,17 +765,8 @@ void MeRefineFracPixel (sWelsEncCtx* pEncCtx, uint8_t* pMemPredInterMb, SWelsME*
pBestPredInter = pRef;
iInterBlk4Stride = kiStrideRef;
}
if (MB_WIDTH_LUMA == iWidth && MB_HEIGHT_LUMA == iHeight) { //P16x16
pFunc->pfCopy16x16NotAligned (pMemPredInterMb, MB_WIDTH_LUMA, pBestPredInter,
iInterBlk4Stride); // dst can be align with 16 bytes, but not sure at pSrc, 12/29/2011
} else if (MB_WIDTH_LUMA == iWidth && MB_HEIGHT_CHROMA == iHeight) { //P16x8
pFunc->pfCopy16x8NotAligned (pMemPredInterMb, MB_WIDTH_LUMA, pBestPredInter,
iInterBlk4Stride); // dst can be align with 16 bytes, but not sure at pSrc, 12/29/2011
} else if (MB_WIDTH_CHROMA == iWidth && MB_HEIGHT_LUMA == iHeight) { //P8x16
pFunc->pfCopy8x16Aligned (pMemPredInterMb, MB_WIDTH_LUMA, pBestPredInter, iInterBlk4Stride);
} else { //P8x8
pFunc->pfCopy8x8Aligned (pMemPredInterMb, MB_WIDTH_LUMA, pBestPredInter, iInterBlk4Stride);
}
pMeRefine->pfCopyBlockByMode(pMemPredInterMb, MB_WIDTH_LUMA, pBestPredInter,
iInterBlk4Stride);
}
void InitBlkStrideWithRef (int32_t* pBlkStride, const int32_t kiStrideRef) {

5
codec/encoder/core/src/svc_base_layer_md.cpp
View File

@ -1448,6 +1448,7 @@ void WelsMdInterMbRefinement (sWelsEncCtx* pEncCtx, SWelsMD* pWelsMd, SMB* pCurM
case MB_TYPE_16x16:
//luma
InitMeRefinePointer (&sMeRefine, pMbCache, 0);
sMeRefine.pfCopyBlockByMode = pEncCtx->pFuncList->pfCopy16x16NotAligned; // dst can be align with 16 bytes, but not sure at pSrc, 12/29/2011
MeRefineFracPixel (pEncCtx, pDstLuma, &pWelsMd->sMe.sMe16x16, &sMeRefine, 16, 16);
UpdateP16x16MotionInfo (pMbCache, pCurMb, pWelsMd->uiRef, &pWelsMd->sMe.sMe16x16.sMv);
@ -1474,6 +1475,7 @@ void WelsMdInterMbRefinement (sWelsEncCtx* pEncCtx, SWelsMD* pWelsMd, SMB* pCurM
case MB_TYPE_16x8:
iPixStride = 0;
sMeRefine.pfCopyBlockByMode = pEncCtx->pFuncList->pfCopy16x8NotAligned; // dst can be align with 16 bytes, but not sure at pSrc, 12/29/2011
for (i = 0; i < 2; i++) {
//luma
iIdx = i << 3;
@ -1503,6 +1505,7 @@ void WelsMdInterMbRefinement (sWelsEncCtx* pEncCtx, SWelsMD* pWelsMd, SMB* pCurM
case MB_TYPE_8x16:
iPixStride = 0;
sMeRefine.pfCopyBlockByMode = pEncCtx->pFuncList->pfCopy8x16Aligned;
for (i = 0; i < 2; i++) {
//luma
iIdx = i << 2;
@ -1528,8 +1531,8 @@ void WelsMdInterMbRefinement (sWelsEncCtx* pEncCtx, SWelsMD* pWelsMd, SMB* pCurM
pEncCtx->pFuncList->sMcFuncs.pMcChromaFunc (pTmpRefCr, iLineSizeRefUV, pTmpDstCr, 8, pMv->iMvX, pMv->iMvY, 4, 8); //Cr
}
break;
case MB_TYPE_8x8:
sMeRefine.pfCopyBlockByMode = pEncCtx->pFuncList->pfCopy8x8Aligned;
for (i = 0; i < 4; i++) {
int32_t iBlk8Idx = i << 2; //0, 4, 8, 12
int32_t iBlk4X, iBlk4Y;