summaryrefslogtreecommitdiffstats
path: root/libswscale/x86
diff options
context:
space:
mode:
authorMichael Niedermayer <michael@niedermayer.cc>2015-12-15 02:50:20 +0100
committerMichael Niedermayer <michael@niedermayer.cc>2015-12-15 03:06:33 +0100
commita066ff89bcbae6033c2ffda9271cad84f6c1b807 (patch)
tree4cac642a323ddd84847c5a833e4a830acd5a2047 /libswscale/x86
parent80bfce35ccd11458e97f68f417fc094c5347070c (diff)
downloadffmpeg-streaming-a066ff89bcbae6033c2ffda9271cad84f6c1b807.zip
ffmpeg-streaming-a066ff89bcbae6033c2ffda9271cad84f6c1b807.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>
Diffstat (limited to 'libswscale/x86')
-rw-r--r--libswscale/x86/rgb2rgb_template.c14
1 files changed, 6 insertions, 8 deletions
diff --git a/libswscale/x86/rgb2rgb_template.c b/libswscale/x86/rgb2rgb_template.c
index 6524461..a10f268 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];
@@ -1946,9 +1946,7 @@ static void RENAME(interleaveBytes)(const uint8_t *src1, const uint8_t *src2, ui
src2 += src2Stride;
}
__asm__(
-#if !COMPILE_TEMPLATE_SSE2
EMMS" \n\t"
-#endif
SFENCE" \n\t"
::: "memory"
);
OpenPOWER on IntegriCloud