---
libswscale/x86/rgb2rgb_template.c | 15 ++-------------
1 file changed, 2 insertions(+), 13 deletions(-)
diff --git a/libswscale/x86/rgb2rgb_template.c
b/libswscale/x86/rgb2rgb_template.c
index 5aeef8c..b7bb3ce 100644
--- a/libswscale/x86/rgb2rgb_template.c
+++ b/libswscale/x86/rgb2rgb_template.c
@@ -1881,12 +1881,12 @@ static void RENAME(interleaveBytes)(const uint8_t
*src1, const uint8_t *src2, ui
for (h=0; h < height; h++) {
int w;
-#if COMPILE_TEMPLATE_SSE2
__asm__(
"xor %%"REG_a", %%"REG_a" \n\t"
"1: \n\t"
PREFETCH" 64(%1, %%"REG_a") \n\t"
PREFETCH" 64(%2, %%"REG_a") \n\t"
+#if COMPILE_TEMPLATE_SSE2
"movdqa (%1, %%"REG_a"), %%xmm0 \n\t"
"movdqa (%1, %%"REG_a"), %%xmm1 \n\t"
"movdqa (%2, %%"REG_a"), %%xmm2 \n\t"
@@ -1894,18 +1894,7 @@ static void RENAME(interleaveBytes)(const uint8_t *src1,
const uint8_t *src2, ui
"punpckhbw %%xmm2, %%xmm1 \n\t"
"movntdq %%xmm0, (%0, %%"REG_a", 2) \n\t"
"movntdq %%xmm1, 16(%0, %%"REG_a", 2) \n\t"
- "add $16, %%"REG_a" \n\t"
- "cmp %3, %%"REG_a" \n\t"
- " jb 1b \n\t"
- ::"r"(dest), "r"(src1), "r"(src2), "r" ((x86_reg)width-15)
- : "memory", "%"REG_a""
- );
#else
- __asm__(
- "xor %%"REG_a", %%"REG_a" \n\t"
- "1: \n\t"
- PREFETCH" 64(%1, %%"REG_a") \n\t"
- PREFETCH" 64(%2, %%"REG_a") \n\t"
"movq (%1, %%"REG_a"), %%mm0 \n\t"
"movq 8(%1, %%"REG_a"), %%mm2 \n\t"
"movq %%mm0, %%mm1 \n\t"
@@ -1920,13 +1909,13 @@ static void RENAME(interleaveBytes)(const uint8_t
*src1, const uint8_t *src2, ui
MOVNTQ" %%mm1, 8(%0, %%"REG_a", 2) \n\t"
MOVNTQ" %%mm2, 16(%0, %%"REG_a", 2) \n\t"
MOVNTQ" %%mm3, 24(%0, %%"REG_a", 2) \n\t"
+#endif
"add $16, %%"REG_a" \n\t"
"cmp %3, %%"REG_a" \n\t"
" jb 1b \n\t"
::"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];
--
1.7.9.5
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel
