diff options
| author | Michael Niedermayer <michael@niedermayer.cc> | 2015-12-15 02:50:20 +0100 |
|---|---|---|
| committer | Carl Eugen Hoyos <cehoyos@ag.or.at> | 2015-12-15 11:21:17 +0100 |
| commit | ef23c40d435bb9bf44842d57168cd3f32465d903 (patch) | |
| tree | 43534881a32ec7601394f24245b5b766d5a28ef8 /libswscale/x86/rgb2rgb_template.c | |
| parent | 442e7c910a593319e2024c76d699d916a650f2f7 (diff) | |
| download | ffmpeg-ef23c40d435bb9bf44842d57168cd3f32465d903.tar.gz | |
swscale/x86/rgb2rgb_template: Fallback to mmx in interleaveBytes() if the alignment is insufficient for SSE*
This also as a sideeffect fixes the non aligned case
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
(cherry picked from commit a066ff89bcbae6033c2ffda9271cad84f6c1b807)
Diffstat (limited to 'libswscale/x86/rgb2rgb_template.c')
| -rw-r--r-- | libswscale/x86/rgb2rgb_template.c | 12 |
1 files changed, 6 insertions, 6 deletions
diff --git a/libswscale/x86/rgb2rgb_template.c b/libswscale/x86/rgb2rgb_template.c index dc8ca252c0..fca0ad9f29 100644 --- a/libswscale/x86/rgb2rgb_template.c +++ b/libswscale/x86/rgb2rgb_template.c @@ -1874,10 +1874,9 @@ static void RENAME(interleaveBytes)(const uint8_t *src1, const uint8_t *src2, ui for (h=0; h < height; h++) { int w; - if (width >= 16 + if (width >= 16) { #if COMPILE_TEMPLATE_SSE2 - && !((((intptr_t)src1) | ((intptr_t)src2) | ((intptr_t)dest))&15) - ) + if (!((((intptr_t)src1) | ((intptr_t)src2) | ((intptr_t)dest))&15)) { __asm__( "xor %%"REG_a", %%"REG_a" \n\t" "1: \n\t" @@ -1896,8 +1895,8 @@ static void RENAME(interleaveBytes)(const uint8_t *src1, const uint8_t *src2, ui ::"r"(dest), "r"(src1), "r"(src2), "r" ((x86_reg)width-15) : "memory", "%"REG_a"" ); -#else - ) + } else +#endif __asm__( "xor %%"REG_a", %%"REG_a" \n\t" "1: \n\t" @@ -1923,7 +1922,8 @@ static void RENAME(interleaveBytes)(const uint8_t *src1, const uint8_t *src2, ui ::"r"(dest), "r"(src1), "r"(src2), "r" ((x86_reg)width-15) : "memory", "%"REG_a ); -#endif + + } for (w= (width&(~15)); w < width; w++) { dest[2*w+0] = src1[w]; dest[2*w+1] = src2[w]; |