diff options
| author | Michael Niedermayer <michaelni@gmx.at> | 2015-03-16 14:28:11 +0100 |
|---|---|---|
| committer | Michael Niedermayer <michaelni@gmx.at> | 2015-03-16 14:51:51 +0100 |
| commit | d79f7bf0d63a81ee66026ee92a6946a7303d04bd (patch) | |
| tree | a0ac63dd8c622c15a19637f4a3b692159e19e0b5 /libavcodec/x86/cavsdsp.c | |
| parent | 12cf61c3b26151806c87ba8774503673e407ac33 (diff) | |
| download | ffmpeg-d79f7bf0d63a81ee66026ee92a6946a7303d04bd.tar.gz | |
avcodec/x86/cavsdsp: remove incorrect LOCAL_ALIGN tmp
This is faster and simpler as well
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Diffstat (limited to 'libavcodec/x86/cavsdsp.c')
| -rw-r--r-- | libavcodec/x86/cavsdsp.c | 24 |
1 files changed, 11 insertions, 13 deletions
diff --git a/libavcodec/x86/cavsdsp.c b/libavcodec/x86/cavsdsp.c index f142db6eba..2ac3bb583a 100644 --- a/libavcodec/x86/cavsdsp.c +++ b/libavcodec/x86/cavsdsp.c @@ -142,8 +142,6 @@ static void cavs_idct8_add_mmx(uint8_t *dst, int16_t *block, int stride) LOCAL_ALIGNED(16, int16_t, b2, [64]); for(i=0; i<2; i++){ - LOCAL_ALIGNED(8, uint64_t, tmp); - cavs_idct8_1d(block+4*i, ff_pw_4.a); __asm__ volatile( @@ -155,19 +153,19 @@ static void cavs_idct8_add_mmx(uint8_t *dst, int16_t *block, int stride) "psraw $3, %%mm2 \n\t" "psraw $3, %%mm1 \n\t" "psraw $3, %%mm0 \n\t" - "movq %%mm7, %0 \n\t" + "movq %%mm7, (%0) \n\t" TRANSPOSE4( %%mm0, %%mm2, %%mm4, %%mm6, %%mm7 ) - "movq %%mm0, 8(%1) \n\t" - "movq %%mm6, 24(%1) \n\t" - "movq %%mm7, 40(%1) \n\t" - "movq %%mm4, 56(%1) \n\t" - "movq %0, %%mm7 \n\t" + "movq %%mm0, 8(%0) \n\t" + "movq %%mm6, 24(%0) \n\t" + "movq %%mm7, 40(%0) \n\t" + "movq %%mm4, 56(%0) \n\t" + "movq (%0), %%mm7 \n\t" TRANSPOSE4( %%mm7, %%mm5, %%mm3, %%mm1, %%mm0 ) - "movq %%mm7, (%1) \n\t" - "movq %%mm1, 16(%1) \n\t" - "movq %%mm0, 32(%1) \n\t" - "movq %%mm3, 48(%1) \n\t" - : "=m"(tmp) + "movq %%mm7, (%0) \n\t" + "movq %%mm1, 16(%0) \n\t" + "movq %%mm0, 32(%0) \n\t" + "movq %%mm3, 48(%0) \n\t" + : : "r"(b2+32*i) : "memory" ); |