diff options
| author | Michael Niedermayer <michaelni@gmx.at> | 2005-06-01 11:56:58 +0000 |
|---|---|---|
| committer | Michael Niedermayer <michaelni@gmx.at> | 2005-06-01 11:56:58 +0000 |
| commit | 6510f43cf3e06cb3dae4a501fb80fe59d085b4b5 (patch) | |
| tree | 5cf5fa06c12e63d8e54b1d7be7d9d95f51b9de32 /libavcodec | |
| parent | 987ae784e63d7b68ade73d5a7a5426c0a09eab9f (diff) | |
| download | ffmpeg-6510f43cf3e06cb3dae4a501fb80fe59d085b4b5.tar.gz | |
merge a few asm blocks so gcc cant unoptimize it (658->631 dezicycles on duron)
Originally committed as revision 4334 to svn://svn.ffmpeg.org/ffmpeg/trunk
Diffstat (limited to 'libavcodec')
| -rw-r--r-- | libavcodec/i386/idct_mmx.c | 20 |
1 files changed, 13 insertions, 7 deletions
diff --git a/libavcodec/i386/idct_mmx.c b/libavcodec/i386/idct_mmx.c index c2bd6c88a1..7be3cf592b 100644 --- a/libavcodec/i386/idct_mmx.c +++ b/libavcodec/i386/idct_mmx.c @@ -637,14 +637,13 @@ declare_idct (ff_mmx_idct, mmx_table, SBUTTERFLY( a, c, d, dq ) /* a=aeim d=bfjn */\ SBUTTERFLY( t, b, c, dq ) /* t=cgko c=dhlp */ -#define STORE_DIFF_4P( p, t, z, dst ) \ - asm volatile(\ +#define STORE_DIFF_4P( p, t, z ) \ "psraw $6, "#p" \n\t"\ "movd (%0), "#t" \n\t"\ "punpcklbw "#z", "#t" \n\t"\ "paddsw "#t", "#p" \n\t"\ "packuswb "#z", "#p" \n\t"\ - "movd "#p", (%0) \n\t" :: "r"(dst) ) + "movd "#p", (%0) \n\t" static const uint64_t ff_pw_32 attribute_used __attribute__ ((aligned(8))) = 0x0020002000200020ULL; @@ -674,8 +673,15 @@ void ff_h264_idct_add_mmx2(uint8_t *dst, int16_t *block, int stride) "pxor %%mm7, %%mm7 \n\t" :: "m"(ff_pw_32)); - STORE_DIFF_4P( %%mm0, %%mm1, %%mm7, &dst[0*stride] ); - STORE_DIFF_4P( %%mm2, %%mm1, %%mm7, &dst[1*stride] ); - STORE_DIFF_4P( %%mm3, %%mm1, %%mm7, &dst[2*stride] ); - STORE_DIFF_4P( %%mm4, %%mm1, %%mm7, &dst[3*stride] ); + asm volatile( + STORE_DIFF_4P( %%mm0, %%mm1, %%mm7) + "addl %1, %0 \n\t" + STORE_DIFF_4P( %%mm2, %%mm1, %%mm7) + "addl %1, %0 \n\t" + STORE_DIFF_4P( %%mm3, %%mm1, %%mm7) + "addl %1, %0 \n\t" + STORE_DIFF_4P( %%mm4, %%mm1, %%mm7) + : "+r"(dst) + : "r" (stride) + ); } |