har0ke/FFmpeg
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Do not misuse HAVE_ defines. Introduce COMPILE_TEMPLATE_ defines and use them

Browse Source 
instead.

Originally committed as revision 29446 to svn://svn.mplayerhq.hu/mplayer/trunk/libswscale
This commit is contained in:
Ramiro Polla 2009-07-27 06:47:41 +00:00
parent 18c61752fd
commit 94daf2e94c
2 changed files with 79 additions and 85 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

70
libswscale/swscale.c
View File

@ -1324,14 +1324,10 @@ static inline void monoblack2Y(uint8_t *dst, const uint8_t *src, long width, uin
#endif
#endif //ARCH_X86
#undef HAVE_MMX
#undef HAVE_MMX2
#undef HAVE_AMD3DNOW
#undef HAVE_ALTIVEC
#define HAVE_MMX 0
#define HAVE_MMX2 0
#define HAVE_AMD3DNOW 0
#define HAVE_ALTIVEC 0
#define COMPILE_TEMPLATE_MMX 0
#define COMPILE_TEMPLATE_MMX2 0
#define COMPILE_TEMPLATE_AMD3DNOW 0
#define COMPILE_TEMPLATE_ALTIVEC 0
#ifdef COMPILE_C
#define RENAME(a) a ## _C
@ -1340,8 +1336,8 @@ static inline void monoblack2Y(uint8_t *dst, const uint8_t *src, long width, uin
#ifdef COMPILE_ALTIVEC
#undef RENAME
#undef HAVE_ALTIVEC
#define HAVE_ALTIVEC 1
#undef COMPILE_TEMPLATE_ALTIVEC
#define COMPILE_TEMPLATE_ALTIVEC 1
#define RENAME(a) a ## _altivec
#include "swscale_template.c"
#endif
@ -1351,12 +1347,12 @@ static inline void monoblack2Y(uint8_t *dst, const uint8_t *src, long width, uin
//MMX versions
#ifdef COMPILE_MMX
#undef RENAME
#undef HAVE_MMX
#undef HAVE_MMX2
#undef HAVE_AMD3DNOW
#define HAVE_MMX 1
#define HAVE_MMX2 0
#define HAVE_AMD3DNOW 0
#undef COMPILE_TEMPLATE_MMX
#undef COMPILE_TEMPLATE_MMX2
#undef COMPILE_TEMPLATE_AMD3DNOW
#define COMPILE_TEMPLATE_MMX 1
#define COMPILE_TEMPLATE_MMX2 0
#define COMPILE_TEMPLATE_AMD3DNOW 0
#define RENAME(a) a ## _MMX
#include "swscale_template.c"
#endif
@ -1364,12 +1360,12 @@ static inline void monoblack2Y(uint8_t *dst, const uint8_t *src, long width, uin
//MMX2 versions
#ifdef COMPILE_MMX2
#undef RENAME
#undef HAVE_MMX
#undef HAVE_MMX2
#undef HAVE_AMD3DNOW
#define HAVE_MMX 1
#define HAVE_MMX2 1
#define HAVE_AMD3DNOW 0
#undef COMPILE_TEMPLATE_MMX
#undef COMPILE_TEMPLATE_MMX2
#undef COMPILE_TEMPLATE_AMD3DNOW
#define COMPILE_TEMPLATE_MMX 1
#define COMPILE_TEMPLATE_MMX2 1
#define COMPILE_TEMPLATE_AMD3DNOW 0
#define RENAME(a) a ## _MMX2
#include "swscale_template.c"
#endif
@ -1377,20 +1373,18 @@ static inline void monoblack2Y(uint8_t *dst, const uint8_t *src, long width, uin
//3DNOW versions
#ifdef COMPILE_3DNOW
#undef RENAME
#undef HAVE_MMX
#undef HAVE_MMX2
#undef HAVE_AMD3DNOW
#define HAVE_MMX 1
#define HAVE_MMX2 0
#define HAVE_AMD3DNOW 1
#undef COMPILE_TEMPLATE_MMX
#undef COMPILE_TEMPLATE_MMX2
#undef COMPILE_TEMPLATE_AMD3DNOW
#define COMPILE_TEMPLATE_MMX 1
#define COMPILE_TEMPLATE_MMX2 0
#define COMPILE_TEMPLATE_AMD3DNOW 1
#define RENAME(a) a ## _3DNow
#include "swscale_template.c"
#endif
#endif //ARCH_X86
// minor note: the HAVE_xyz are messed up after this line so don't use them
static double getSplineCoeff(double a, double b, double c, double d, double dist)
{
// printf("%f %f %f %f %f\n", a,b,c,d,dist);
@ -2003,16 +1997,16 @@ static SwsFunc getSwsFunc(SwsContext *c)
return swScale_C;
#endif /* ARCH_X86 && CONFIG_GPL */
#else //CONFIG_RUNTIME_CPUDETECT
#if HAVE_MMX2
#if COMPILE_TEMPLATE_MMX2
sws_init_swScale_MMX2(c);
return swScale_MMX2;
#elif HAVE_AMD3DNOW
#elif COMPILE_TEMPLATE_AMD3DNOW
sws_init_swScale_3DNow(c);
return swScale_3DNow;
#elif HAVE_MMX
#elif COMPILE_TEMPLATE_MMX
sws_init_swScale_MMX(c);
return swScale_MMX;
#elif HAVE_ALTIVEC
#elif COMPILE_TEMPLATE_ALTIVEC
sws_init_swScale_altivec(c);
return swScale_altivec;
#else
@ -2565,13 +2559,13 @@ SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat, int d
#if !CONFIG_RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC|SWS_CPU_CAPS_BFIN);
#if HAVE_MMX2
#if COMPILE_TEMPLATE_MMX2
flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
#elif HAVE_AMD3DNOW
#elif COMPILE_TEMPLATE_AMD3DNOW
flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
#elif HAVE_MMX
#elif COMPILE_TEMPLATE_MMX
flags |= SWS_CPU_CAPS_MMX;
#elif HAVE_ALTIVEC
#elif COMPILE_TEMPLATE_ALTIVEC
flags |= SWS_CPU_CAPS_ALTIVEC;
#elif ARCH_BFIN
flags |= SWS_CPU_CAPS_BFIN;

94
libswscale/swscale_template.c
View File

@ -27,10 +27,10 @@
#undef PREFETCH
#undef PREFETCHW
#if HAVE_AMD3DNOW
#if COMPILE_TEMPLATE_AMD3DNOW
#define PREFETCH "prefetch"
#define PREFETCHW "prefetchw"
#elif HAVE_MMX2
#elif COMPILE_TEMPLATE_MMX2
#define PREFETCH "prefetchnta"
#define PREFETCHW "prefetcht0"
#else
@ -38,20 +38,20 @@
#define PREFETCHW " # nop"
#endif
#if HAVE_MMX2
#if COMPILE_TEMPLATE_MMX2
#define PAVGB(a,b) "pavgb " #a ", " #b " \n\t"
#elif HAVE_AMD3DNOW
#elif COMPILE_TEMPLATE_AMD3DNOW
#define PAVGB(a,b) "pavgusb " #a ", " #b " \n\t"
#endif
#if HAVE_MMX2
#if COMPILE_TEMPLATE_MMX2
#define REAL_MOVNTQ(a,b) "movntq " #a ", " #b " \n\t"
#else
#define REAL_MOVNTQ(a,b) "movq " #a ", " #b " \n\t"
#endif
#define MOVNTQ(a,b) REAL_MOVNTQ(a,b)
#if HAVE_ALTIVEC
#if COMPILE_TEMPLATE_ALTIVEC
#include "ppc/swscale_altivec_template.c"
#endif
@ -875,7 +875,7 @@
"cmp "#dstw", "#index" \n\t"\
" jb 1b \n\t"
#if HAVE_MMX2
#if COMPILE_TEMPLATE_MMX2
#undef WRITEBGR24
#define WRITEBGR24(dst, dstw, index) WRITEBGR24MMX2(dst, dstw, index)
#else
@ -905,7 +905,7 @@ static inline void RENAME(yuv2yuvX)(SwsContext *c, const int16_t *lumFilter, con
const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize, const int16_t **alpSrc,
uint8_t *dest, uint8_t *uDest, uint8_t *vDest, uint8_t *aDest, long dstW, long chrDstW)
{
#if HAVE_MMX
#if COMPILE_TEMPLATE_MMX
if(!(c->flags & SWS_BITEXACT)){
if (c->flags & SWS_ACCURATE_RND){
if (uDest){
@ -931,15 +931,15 @@ static inline void RENAME(yuv2yuvX)(SwsContext *c, const int16_t *lumFilter, con
return;
}
#endif
#if HAVE_ALTIVEC
#if COMPILE_TEMPLATE_ALTIVEC
yuv2yuvX_altivec_real(lumFilter, lumSrc, lumFilterSize,
chrFilter, chrSrc, chrFilterSize,
dest, uDest, vDest, dstW, chrDstW);
#else //HAVE_ALTIVEC
#else //COMPILE_TEMPLATE_ALTIVEC
yuv2yuvXinC(lumFilter, lumSrc, lumFilterSize,
chrFilter, chrSrc, chrFilterSize,
alpSrc, dest, uDest, vDest, aDest, dstW, chrDstW);
#endif //!HAVE_ALTIVEC
#endif //!COMPILE_TEMPLATE_ALTIVEC
}
static inline void RENAME(yuv2nv12X)(SwsContext *c, const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
@ -955,7 +955,7 @@ static inline void RENAME(yuv2yuv1)(SwsContext *c, const int16_t *lumSrc, const
uint8_t *dest, uint8_t *uDest, uint8_t *vDest, uint8_t *aDest, long dstW, long chrDstW)
{
int i;
#if HAVE_MMX
#if COMPILE_TEMPLATE_MMX
if(!(c->flags & SWS_BITEXACT)){
long p= 4;
uint8_t *src[4]= {alpSrc + dstW, lumSrc + dstW, chrSrc + chrDstW, chrSrc + VOFW + chrDstW};
@ -1032,7 +1032,7 @@ static inline void RENAME(yuv2packedX)(SwsContext *c, const int16_t *lumFilter,
const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
const int16_t **alpSrc, uint8_t *dest, long dstW, long dstY)
{
#if HAVE_MMX
#if COMPILE_TEMPLATE_MMX
x86_reg dummy=0;
if(!(c->flags & SWS_BITEXACT)){
if (c->flags & SWS_ACCURATE_RND){
@ -1193,8 +1193,8 @@ static inline void RENAME(yuv2packedX)(SwsContext *c, const int16_t *lumFilter,
}
}
}
#endif /* HAVE_MMX */
#if HAVE_ALTIVEC
#endif /* COMPILE_TEMPLATE_MMX */
#if COMPILE_TEMPLATE_ALTIVEC
/* The following list of supported dstFormat values should
match what's found in the body of ff_yuv2packedX_altivec() */
if (!(c->flags & SWS_BITEXACT) && !c->alpPixBuf &&
@ -1221,7 +1221,7 @@ static inline void RENAME(yuv2packed2)(SwsContext *c, const uint16_t *buf0, cons
int uvalpha1=4095-uvalpha;
int i;
#if HAVE_MMX
#if COMPILE_TEMPLATE_MMX
if(!(c->flags & SWS_BITEXACT)){
switch(c->dstFormat)
{
@ -1357,7 +1357,7 @@ static inline void RENAME(yuv2packed2)(SwsContext *c, const uint16_t *buf0, cons
default: break;
}
}
#endif //HAVE_MMX
#endif //COMPILE_TEMPLATE_MMX
YSCALE_YUV_2_ANYRGB_C(YSCALE_YUV_2_RGB2_C, YSCALE_YUV_2_PACKED2_C(void,0), YSCALE_YUV_2_GRAY16_2_C, YSCALE_YUV_2_MONO2_C)
}
@ -1379,7 +1379,7 @@ static inline void RENAME(yuv2packed1)(SwsContext *c, const uint16_t *buf0, cons
return;
}
#if HAVE_MMX
#if COMPILE_TEMPLATE_MMX
if(!(flags & SWS_BITEXACT)){
if (uvalpha < 2048) // note this is not correct (shifts chrominance by 0.5 pixels) but it is a bit faster
{
@ -1600,7 +1600,7 @@ static inline void RENAME(yuv2packed1)(SwsContext *c, const uint16_t *buf0, cons
}
}
}
#endif /* HAVE_MMX */
#endif /* COMPILE_TEMPLATE_MMX */
if (uvalpha < 2048)
{
YSCALE_YUV_2_ANYRGB_C(YSCALE_YUV_2_RGB1_C, YSCALE_YUV_2_PACKED1_C(void,0), YSCALE_YUV_2_GRAY16_1_C, YSCALE_YUV_2_MONO2_C)
@ -1613,7 +1613,7 @@ static inline void RENAME(yuv2packed1)(SwsContext *c, const uint16_t *buf0, cons
static inline void RENAME(yuy2ToY)(uint8_t *dst, const uint8_t *src, long width, uint32_t *unused)
{
#if HAVE_MMX
#if COMPILE_TEMPLATE_MMX
__asm__ volatile(
"movq "MANGLE(bm01010101)", %%mm2 \n\t"
"mov %0, %%"REG_a" \n\t"
@ -1638,7 +1638,7 @@ static inline void RENAME(yuy2ToY)(uint8_t *dst, const uint8_t *src, long width,
static inline void RENAME(yuy2ToUV)(uint8_t *dstU, uint8_t *dstV, const uint8_t *src1, const uint8_t *src2, long width, uint32_t *unused)
{
#if HAVE_MMX
#if COMPILE_TEMPLATE_MMX
__asm__ volatile(
"movq "MANGLE(bm01010101)", %%mm4 \n\t"
"mov %0, %%"REG_a" \n\t"
@ -1673,7 +1673,7 @@ static inline void RENAME(yuy2ToUV)(uint8_t *dstU, uint8_t *dstV, const uint8_t
static inline void RENAME(LEToUV)(uint8_t *dstU, uint8_t *dstV, const uint8_t *src1, const uint8_t *src2, long width, uint32_t *unused)
{
#if HAVE_MMX
#if COMPILE_TEMPLATE_MMX
__asm__ volatile(
"mov %0, %%"REG_a" \n\t"
"1: \n\t"
@ -1708,7 +1708,7 @@ static inline void RENAME(LEToUV)(uint8_t *dstU, uint8_t *dstV, const uint8_t *s
* yuy2ToY/UV)(dst, src+1, ...) would have 100% unaligned accesses. */
static inline void RENAME(uyvyToY)(uint8_t *dst, const uint8_t *src, long width, uint32_t *unused)
{
#if HAVE_MMX
#if COMPILE_TEMPLATE_MMX
__asm__ volatile(
"mov %0, %%"REG_a" \n\t"
"1: \n\t"
@ -1732,7 +1732,7 @@ static inline void RENAME(uyvyToY)(uint8_t *dst, const uint8_t *src, long width,
static inline void RENAME(uyvyToUV)(uint8_t *dstU, uint8_t *dstV, const uint8_t *src1, const uint8_t *src2, long width, uint32_t *unused)
{
#if HAVE_MMX
#if COMPILE_TEMPLATE_MMX
__asm__ volatile(
"movq "MANGLE(bm01010101)", %%mm4 \n\t"
"mov %0, %%"REG_a" \n\t"
@ -1767,7 +1767,7 @@ static inline void RENAME(uyvyToUV)(uint8_t *dstU, uint8_t *dstV, const uint8_t
static inline void RENAME(BEToUV)(uint8_t *dstU, uint8_t *dstV, const uint8_t *src1, const uint8_t *src2, long width, uint32_t *unused)
{
#if HAVE_MMX
#if COMPILE_TEMPLATE_MMX
__asm__ volatile(
"movq "MANGLE(bm01010101)", %%mm4 \n\t"
"mov %0, %%"REG_a" \n\t"
@ -1799,7 +1799,7 @@ static inline void RENAME(BEToUV)(uint8_t *dstU, uint8_t *dstV, const uint8_t *s
#endif
}
#if HAVE_MMX
#if COMPILE_TEMPLATE_MMX
static inline void RENAME(bgr24ToY_mmx)(uint8_t *dst, const uint8_t *src, long width, int srcFormat)
{
@ -1914,7 +1914,7 @@ static inline void RENAME(bgr24ToUV_mmx)(uint8_t *dstU, uint8_t *dstV, const uin
static inline void RENAME(bgr24ToY)(uint8_t *dst, const uint8_t *src, long width, uint32_t *unused)
{
#if HAVE_MMX
#if COMPILE_TEMPLATE_MMX
RENAME(bgr24ToY_mmx)(dst, src, width, PIX_FMT_BGR24);
#else
int i;
@ -1926,12 +1926,12 @@ static inline void RENAME(bgr24ToY)(uint8_t *dst, const uint8_t *src, long width
dst[i]= ((RY*r + GY*g + BY*b + (33<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
}
#endif /* HAVE_MMX */
#endif /* COMPILE_TEMPLATE_MMX */
}
static inline void RENAME(bgr24ToUV)(uint8_t *dstU, uint8_t *dstV, const uint8_t *src1, const uint8_t *src2, long width, uint32_t *unused)
{
#if HAVE_MMX
#if COMPILE_TEMPLATE_MMX
RENAME(bgr24ToUV_mmx)(dstU, dstV, src1, width, PIX_FMT_BGR24);
#else
int i;
@ -1944,7 +1944,7 @@ static inline void RENAME(bgr24ToUV)(uint8_t *dstU, uint8_t *dstV, const uint8_t
dstU[i]= (RU*r + GU*g + BU*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT;
dstV[i]= (RV*r + GV*g + BV*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT;
}
#endif /* HAVE_MMX */
#endif /* COMPILE_TEMPLATE_MMX */
assert(src1 == src2);
}
@ -1965,7 +1965,7 @@ static inline void RENAME(bgr24ToUV_half)(uint8_t *dstU, uint8_t *dstV, const ui
static inline void RENAME(rgb24ToY)(uint8_t *dst, const uint8_t *src, long width, uint32_t *unused)
{
#if HAVE_MMX
#if COMPILE_TEMPLATE_MMX
RENAME(bgr24ToY_mmx)(dst, src, width, PIX_FMT_RGB24);
#else
int i;
@ -1982,7 +1982,7 @@ static inline void RENAME(rgb24ToY)(uint8_t *dst, const uint8_t *src, long width
static inline void RENAME(rgb24ToUV)(uint8_t *dstU, uint8_t *dstV, const uint8_t *src1, const uint8_t *src2, long width, uint32_t *unused)
{
#if HAVE_MMX
#if COMPILE_TEMPLATE_MMX
assert(src1==src2);
RENAME(bgr24ToUV_mmx)(dstU, dstV, src1, width, PIX_FMT_RGB24);
#else
@ -2020,7 +2020,7 @@ static inline void RENAME(rgb24ToUV_half)(uint8_t *dstU, uint8_t *dstV, const ui
static inline void RENAME(hScale)(int16_t *dst, int dstW, const uint8_t *src, int srcW, int xInc,
const int16_t *filter, const int16_t *filterPos, long filterSize)
{
#if HAVE_MMX
#if COMPILE_TEMPLATE_MMX
assert(filterSize % 4 == 0 && filterSize>0);
if (filterSize==4) // Always true for upscaling, sometimes for down, too.
{
@ -2176,7 +2176,7 @@ static inline void RENAME(hScale)(int16_t *dst, int dstW, const uint8_t *src, in
);
}
#else
#if HAVE_ALTIVEC
#if COMPILE_TEMPLATE_ALTIVEC
hScale_altivec_real(dst, dstW, src, srcW, xInc, filter, filterPos, filterSize);
#else
int i;
@ -2195,8 +2195,8 @@ static inline void RENAME(hScale)(int16_t *dst, int dstW, const uint8_t *src, in
dst[i] = FFMIN(val>>7, (1<<15)-1); // the cubic equation does overflow ...
//dst[i] = val>>7;
}
#endif /* HAVE_ALTIVEC */
#endif /* HAVE_MMX */
#endif /* COMPILE_ALTIVEC */
#endif /* COMPILE_MMX */
}
#define FAST_BILINEAR_X86 \
@ -2251,7 +2251,7 @@ static inline void RENAME(hyscale)(SwsContext *c, uint16_t *dst, long dstWidth,
src= formatConvBuffer;
}
#if HAVE_MMX
#if COMPILE_TEMPLATE_MMX
// Use the new MMX scaler if the MMX2 one can't be used (it is faster than the x86 ASM one).
if (!(flags&SWS_FAST_BILINEAR) || (!canMMX2BeUsed))
#else
@ -2263,7 +2263,7 @@ static inline void RENAME(hyscale)(SwsContext *c, uint16_t *dst, long dstWidth,
else // fast bilinear upscale / crap downscale
{
#if ARCH_X86 && CONFIG_GPL
#if HAVE_MMX2
#if COMPILE_TEMPLATE_MMX2
int i;
#if defined(PIC)
DECLARE_ALIGNED(8, uint64_t, ebxsave);
@ -2331,7 +2331,7 @@ FUNNY_Y_CODE
}
else
{
#endif /* HAVE_MMX2 */
#endif /* COMPILE_TEMPLATE_MMX2 */
x86_reg xInc_shr16 = xInc >> 16;
uint16_t xInc_mask = xInc & 0xffff;
//NO MMX just normal asm ...
@ -2364,7 +2364,7 @@ FUNNY_Y_CODE
:: "r" (src), "m" (dst), "m" (dstWidth), "m" (xInc_shr16), "m" (xInc_mask)
: "%"REG_a, "%"REG_d, "%ecx", "%"REG_D, "%esi"
);
#if HAVE_MMX2
#if COMPILE_TEMPLATE_MMX2
} //if MMX2 can't be used
#endif
#else
@ -2436,7 +2436,7 @@ inline static void RENAME(hcscale)(SwsContext *c, uint16_t *dst, long dstWidth,
src2= formatConvBuffer+VOFW;
}
#if HAVE_MMX
#if COMPILE_TEMPLATE_MMX
// Use the new MMX scaler if the MMX2 one can't be used (it is faster than the x86 ASM one).
if (!(flags&SWS_FAST_BILINEAR) || (!canMMX2BeUsed))
#else
@ -2449,7 +2449,7 @@ inline static void RENAME(hcscale)(SwsContext *c, uint16_t *dst, long dstWidth,
else // fast bilinear upscale / crap downscale
{
#if ARCH_X86 && CONFIG_GPL
#if HAVE_MMX2
#if COMPILE_TEMPLATE_MMX2
int i;
#if defined(PIC)
DECLARE_ALIGNED(8, uint64_t, ebxsave);
@ -2530,7 +2530,7 @@ FUNNY_UV_CODE
}
else
{
#endif /* HAVE_MMX2 */
#endif /* COMPILE_TEMPLATE_MMX2 */
x86_reg xInc_shr16 = (x86_reg) (xInc >> 16);
uint16_t xInc_mask = xInc & 0xffff;
__asm__ volatile(
@ -2566,7 +2566,7 @@ FUNNY_UV_CODE
"r" (src2)
: "%"REG_a, "%"REG_d, "%ecx", "%"REG_D, "%esi"
);
#if HAVE_MMX2
#if COMPILE_TEMPLATE_MMX2
} //if MMX2 can't be used
#endif
#else
@ -2807,7 +2807,7 @@ static int RENAME(swScale)(SwsContext *c, uint8_t* src[], int srcStride[], int s
break; //we can't output a dstY line so let's try with the next slice
}
#if HAVE_MMX
#if COMPILE_TEMPLATE_MMX
c->blueDither= ff_dither8[dstY&1];
if (c->dstFormat == PIX_FMT_RGB555 || c->dstFormat == PIX_FMT_BGR555)
c->greenDither= ff_dither8[dstY&1];
@ -2820,7 +2820,7 @@ static int RENAME(swScale)(SwsContext *c, uint8_t* src[], int srcStride[], int s
const int16_t **lumSrcPtr= (const int16_t **) lumPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize;
const int16_t **chrSrcPtr= (const int16_t **) chrPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;
const int16_t **alpSrcPtr= (CONFIG_SWSCALE_ALPHA && alpPixBuf) ? (const int16_t **) alpPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize : NULL;
#if HAVE_MMX
#if COMPILE_TEMPLATE_MMX
int i;
if (flags & SWS_ACCURATE_RND){
int s= APCK_SIZE / 8;
@ -2993,7 +2993,7 @@ static int RENAME(swScale)(SwsContext *c, uint8_t* src[], int srcStride[], int s
if ((dstFormat == PIX_FMT_YUVA420P) && !alpPixBuf)
fillPlane(dst[3], dstStride[3], dstW, dstY-lastDstY, lastDstY, 255);
#if HAVE_MMX
#if COMPILE_TEMPLATE_MMX
if (flags & SWS_CPU_CAPS_MMX2 ) __asm__ volatile("sfence":::"memory");
/* On K6 femms is faster than emms. On K7 femms is directly mapped to emms. */
if (flags & SWS_CPU_CAPS_3DNOW) __asm__ volatile("femms" :::"memory");