[FFmpeg-cvslog] swscale/x86/rgb2rgb_template: Fallback to mmx in interleaveBytes() if the alignment is insufficient for SSE*
Michael Niedermayer
git at videolan.org
Tue Dec 15 11:23:05 CET 2015
ffmpeg | branch: release/2.6 | Michael Niedermayer <michael at niedermayer.cc> | Tue Dec 15 02:50:20 2015 +0100| [26e98f9ae54e0c519d8d3ea233b7ecea9c4b66ea] | committer: Carl Eugen Hoyos
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 at niedermayer.cc>
(cherry picked from commit a066ff89bcbae6033c2ffda9271cad84f6c1b807)
> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=26e98f9ae54e0c519d8d3ea233b7ecea9c4b66ea
---
libswscale/x86/rgb2rgb_template.c | 12 ++++++------
1 file changed, 6 insertions(+), 6 deletions(-)
diff --git a/libswscale/x86/rgb2rgb_template.c b/libswscale/x86/rgb2rgb_template.c
index 03153fd..6f218dd 100644
--- a/libswscale/x86/rgb2rgb_template.c
+++ b/libswscale/x86/rgb2rgb_template.c
@@ -1887,10 +1887,9 @@ 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"
@@ -1909,8 +1908,8 @@ 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", XMM_CLOBBERS("xmm0", "xmm1", "xmm2",) "%"REG_a
);
-#else
- )
+ } else
+#endif
__asm__(
"xor %%"REG_a", %%"REG_a" \n\t"
"1: \n\t"
@@ -1936,7 +1935,8 @@ 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];
More information about the ffmpeg-cvslog
mailing list