github/mpv
1
Fork 0
mirror of https://github.com/mpv-player/mpv synced 2025-01-20 21:07:29 +01:00
Code Releases Activity

slightly faster rgb32tobgr32; avoid one add and one cmp

Browse Source 
git-svn-id: svn://svn.mplayerhq.hu/mplayer/trunk@23012 b3059339-0415-0410-9bf9-f77b7e298cf2
This commit is contained in:
ivo 2007-04-17 20:38:17 +00:00
parent 1400ea6fbf
commit f854708977
1 changed files with 17 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

35
libswscale/rgb2rgb_template.c
View File

@ -1364,21 +1364,22 @@ 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)
{
uint8_t *d = dst, *s = (uint8_t *) src;
const uint8_t *end = s + src_size;
long idx = 15 - src_size;
uint8_t *s = (uint8_t *) src-idx, *d = dst-idx;
#ifdef HAVE_MMX
__asm __volatile(
" "PREFETCH" (%1) \n"
" test %0, %0 \n"
" jns 2f \n"
" "PREFETCH" (%1, %0) \n"
" movq %3, %%mm7 \n"
" pxor %4, %%mm7 \n"
" movq %%mm7, %%mm6 \n"
" pxor %5, %%mm7 \n"
" jmp 2f \n"
ASMALIGN(4)
"1: \n"
" "PREFETCH" 32(%1) \n"
" movq (%1), %%mm0 \n"
" movq 8(%1), %%mm1 \n"
" "PREFETCH" 32(%1, %0) \n"
" movq (%1, %0), %%mm0 \n"
" movq 8(%1, %0), %%mm1 \n"
# ifdef HAVE_MMX2
" pshufw $177, %%mm0, %%mm3 \n"
" pshufw $177, %%mm1, %%mm5 \n"
@ -1406,23 +1407,21 @@ static inline void RENAME(rgb32tobgr32)(const uint8_t *src, uint8_t *dst, long s
" por %%mm3, %%mm0 \n"
" por %%mm5, %%mm1 \n"
# endif
" "MOVNTQ" %%mm0, (%0) \n"
" "MOVNTQ" %%mm1, 8(%0) \n"
" "MOVNTQ" %%mm0, (%2, %0) \n"
" "MOVNTQ" %%mm1, 8(%2, %0) \n"
" add $16, %0 \n"
" add $16, %1 \n"
"2: \n"
" cmp %1, %2 \n"
" ja 1b \n"
" js 1b \n"
" "SFENCE" \n"
" "EMMS" \n"
: "+r"(d), "+r"(s)
: "r" (end-15), "m" (mask32b), "m" (mask32r), "m" (mmx_one)
"2: \n"
: "+&r"(idx)
: "r" (s), "r" (d), "m" (mask32b), "m" (mask32r), "m" (mmx_one)
: "memory");
#endif
for (; s<end; s+=4, d+=4) {
int v = *(uint32_t *)s, g = v & 0xff00;
for (; idx<15; idx+=4) {
register int v = *(uint32_t *)&s[idx], g = v & 0xff00;
v &= 0xff00ff;
*(uint32_t *)d = (v>>16) + g + (v<<16);
*(uint32_t *)&d[idx] = (v>>16) + g + (v<<16);
}
}