MrBesen/ffmpeg
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

New implementation of rgb32tobgr32

Browse Source 
The previous implementation segfaulted with MMX enabled when fed an image
smaller than the size of the units the MMX code processed. The new code:
- is faster for MMX, MMX2 and plain C
- processes small images correctly
- is LGPL

Originally committed as revision 23009 to svn://svn.mplayerhq.hu/mplayer/trunk/libswscale
This commit is contained in:
Ivo van Poorten 2007-04-16 21:41:03 +00:00
parent 4f99f93268
commit b38d487466
1 changed files with 57 additions and 40 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

97
libswscale/rgb2rgb_template.c
View File

@ -1364,49 +1364,66 @@ static inline void RENAME(rgb16to32)(const uint8_t *src, uint8_t *dst, long src_
static inline void RENAME(rgb32tobgr32)(const uint8_t *src, uint8_t *dst, long src_size) static inline void RENAME(rgb32tobgr32)(const uint8_t *src, uint8_t *dst, long src_size)
{ {
uint8_t *d = dst, *s = (uint8_t *) src;
const uint8_t *end = s + src_size;
#ifdef HAVE_MMX #ifdef HAVE_MMX
/* TODO: unroll this loop */ __asm __volatile(
asm volatile ( " "PREFETCH" (%1) \n"
"xor %%"REG_a", %%"REG_a" \n\t" " movq %3, %%mm7 \n"
ASMALIGN(4) " pxor %4, %%mm7 \n"
"1: \n\t" " movq %%mm7, %%mm6 \n"
PREFETCH" 32(%0, %%"REG_a") \n\t" " pxor %5, %%mm7 \n"
"movq (%0, %%"REG_a"), %%mm0 \n\t" " jmp 2f \n"
"movq %%mm0, %%mm1 \n\t" ASMALIGN(4)
"movq %%mm0, %%mm2 \n\t" "1: \n"
"pslld $16, %%mm0 \n\t" " "PREFETCH" 32(%1) \n"
"psrld $16, %%mm1 \n\t" " movq (%1), %%mm0 \n"
"pand "MANGLE(mask32r)", %%mm0 \n\t" " movq 8(%1), %%mm1 \n"
"pand "MANGLE(mask32g)", %%mm2 \n\t" # ifdef HAVE_MMX2
"pand "MANGLE(mask32b)", %%mm1 \n\t" " pshufw $177, %%mm0, %%mm3 \n"
"por %%mm0, %%mm2 \n\t" " pshufw $177, %%mm1, %%mm5 \n"
"por %%mm1, %%mm2 \n\t" " pand %%mm7, %%mm0 \n"
MOVNTQ" %%mm2, (%1, %%"REG_a") \n\t" " pand %%mm6, %%mm3 \n"
"add $8, %%"REG_a" \n\t" " pand %%mm7, %%mm1 \n"
"cmp %2, %%"REG_a" \n\t" " pand %%mm6, %%mm5 \n"
" jb 1b \n\t" " por %%mm3, %%mm0 \n"
:: "r" (src), "r"(dst), "r" (src_size-7) " por %%mm5, %%mm1 \n"
: "%"REG_a # else
); " movq %%mm0, %%mm2 \n"
" movq %%mm1, %%mm4 \n"
__asm __volatile(SFENCE:::"memory"); " pand %%mm7, %%mm0 \n"
__asm __volatile(EMMS:::"memory"); " pand %%mm6, %%mm2 \n"
#else " pand %%mm7, %%mm1 \n"
unsigned i; " pand %%mm6, %%mm4 \n"
unsigned num_pixels = src_size >> 2; " movq %%mm2, %%mm3 \n"
for(i=0; i<num_pixels; i++) " movq %%mm4, %%mm5 \n"
{ " pslld $16, %%mm2 \n"
#ifdef WORDS_BIGENDIAN " psrld $16, %%mm3 \n"
dst[4*i + 1] = src[4*i + 3]; " pslld $16, %%mm4 \n"
dst[4*i + 2] = src[4*i + 2]; " psrld $16, %%mm5 \n"
dst[4*i + 3] = src[4*i + 1]; " por %%mm2, %%mm0 \n"
#else " por %%mm4, %%mm1 \n"
dst[4*i + 0] = src[4*i + 2]; " por %%mm3, %%mm0 \n"
dst[4*i + 1] = src[4*i + 1]; " por %%mm5, %%mm1 \n"
dst[4*i + 2] = src[4*i + 0]; # endif
" "MOVNTQ" %%mm0, (%0) \n"
" "MOVNTQ" %%mm1, 8(%0) \n"
" add $16, %0 \n"
" add $16, %1 \n"
"2: \n"
" cmp %1, %2 \n"
" ja 1b \n"
" "SFENCE" \n"
" "EMMS" \n"
: "+r"(d), "+r"(s)
: "r" (end-15), "m" (mask32b), "m" (mask32r), "m" (mmx_one)
: "memory");
#endif #endif
for (; s<end; s+=4, d+=4) {
int v = *(uint32_t *)s, g = v & 0xff00;
v &= 0xff00ff;
*(uint32_t *)d = (v>>16) + g + (v<<16);
} }
#endif
} }
static inline void RENAME(rgb24tobgr24)(const uint8_t *src, uint8_t *dst, long src_size) static inline void RENAME(rgb24tobgr24)(const uint8_t *src, uint8_t *dst, long src_size)