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

Simplify code in HorFilterInput16bit in MC

Browse Source 
This avoids a gcc optimizer bug (which seems to be present in some
gcc 4.6 and 4.7 versions) at the -O3 level.
This commit is contained in:
Martin Storsjö 2015-01-26 22:57:20 +02:00
parent 279e14b34e
commit 23b20fb14c
2 changed files with 9 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
codec/decoder/core/src/mc.cpp
View File

@ -134,9 +134,9 @@ static inline void McCopyWidthEq16_c (const uint8_t* pSrc, int32_t iSrcStride, u
//--------------------Luma sample MC------------------//
static inline int32_t HorFilterInput16bit_c (const int16_t* pSrc) {
int32_t iPix05 = pSrc[-2] + pSrc[3];
int32_t iPix14 = pSrc[-1] + pSrc[2];
int32_t iPix23 = pSrc[ 0] + pSrc[1];
int32_t iPix05 = pSrc[0] + pSrc[5];
int32_t iPix14 = pSrc[1] + pSrc[4];
int32_t iPix23 = pSrc[2] + pSrc[3];
return (iPix05 - (iPix14 * 5) + (iPix23 * 20));
}
@ -213,7 +213,7 @@ static inline void McHorVer22_c (const uint8_t* pSrc, int32_t iSrcStride, uint8_
iTmp[j] = FilterInput8bitWithStride_c (pSrc - 2 + j, iSrcStride);
}
for (k = 0; k < iWidth; k++) {
pDst[k] = WelsClip1 ((HorFilterInput16bit_c (&iTmp[2 + k]) + 512) >> 10);
pDst[k] = WelsClip1 ((HorFilterInput16bit_c (&iTmp[k]) + 512) >> 10);
}
pSrc += iSrcStride;
pDst += iDstStride;

10
codec/encoder/core/src/mc.cpp
View File

@ -126,9 +126,9 @@ static inline int32_t HorFilter_c (const uint8_t* pSrc) {
}
static inline int32_t HorFilterInput16bit1_c (const int16_t* pSrc) {
int32_t iPix05 = pSrc[-2] + pSrc[3];
int32_t iPix14 = pSrc[-1] + pSrc[2];
int32_t iPix23 = pSrc[ 0] + pSrc[1];
int32_t iPix05 = pSrc[0] + pSrc[5];
int32_t iPix14 = pSrc[1] + pSrc[4];
int32_t iPix23 = pSrc[2] + pSrc[3];
return (iPix05 - ((iPix14 << 2) + iPix14) + (iPix23 << 4) + (iPix23 << 2));
}
@ -203,7 +203,7 @@ static inline void McHorVer22WidthEq16_c (const uint8_t* pSrc, int32_t iSrcStrid
pTmp[j] = VerFilter_c (pSrc - 2 + j, iSrcStride);
}
for (k = 0; k < 16; k++) {
pDst[k] = WelsClip1 ((HorFilterInput16bit1_c (&pTmp[2 + k]) + 512) >> 10);
pDst[k] = WelsClip1 ((HorFilterInput16bit1_c (&pTmp[k]) + 512) >> 10);
}
pSrc += iSrcStride;
pDst += iDstStride;
@ -342,7 +342,7 @@ static inline void McHorVer22_c (const uint8_t* pSrc, int32_t iSrcStride, uint8_
pTmp[j] = VerFilter_c (pSrc - 2 + j, iSrcStride);
}
for (k = 0; k < iWidth; k++) {
pDst[k] = WelsClip1 ((HorFilterInput16bit1_c (&pTmp[2 + k]) + 512) >> 10);
pDst[k] = WelsClip1 ((HorFilterInput16bit1_c (&pTmp[k]) + 512) >> 10);
}
pSrc += iSrcStride;
pDst += iDstStride;