diff options
| author | Michael Niedermayer <michaelni@gmx.at> | 2005-06-01 01:18:41 +0000 |
|---|---|---|
| committer | Michael Niedermayer <michaelni@gmx.at> | 2005-06-01 01:18:41 +0000 |
| commit | 1f3dbc09b14adcc7611e256b502e2ffeb6cf4625 (patch) | |
| tree | f8572e7fc56e6f534925b959486a49fa53fd3a8b /libavcodec/i386 | |
| parent | 1d62fc856067e7c2b1379ac1e2938b16497e9b25 (diff) | |
| download | ffmpeg-1f3dbc09b14adcc7611e256b502e2ffeb6cf4625.tar.gz | |
add rounding bias before the horizontal idct (765->730 dezicyles on duron)
Originally committed as revision 4330 to svn://svn.ffmpeg.org/ffmpeg/trunk
Diffstat (limited to 'libavcodec/i386')
| -rw-r--r-- | libavcodec/i386/idct_mmx.c | 5 |
1 files changed, 3 insertions, 2 deletions
diff --git a/libavcodec/i386/idct_mmx.c b/libavcodec/i386/idct_mmx.c index 6587455179..4ad7980d63 100644 --- a/libavcodec/i386/idct_mmx.c +++ b/libavcodec/i386/idct_mmx.c @@ -641,7 +641,6 @@ declare_idct (ff_mmx_idct, mmx_table, #define STORE_DIFF_4P( p, t, pw32, z, dst ) \ asm volatile(\ - "paddw "#pw32", "#p" \n\t"\ "psraw $6, "#p" \n\t"\ "movd (%0), "#t" \n\t"\ "punpcklbw "#z", "#t" \n\t"\ @@ -665,9 +664,12 @@ void ff_h264_idct_add_mmx2(uint8_t *dst, int16_t *block, int stride) /* mm1=s02+s13 mm2=s02-s13 mm4=d02+d13 mm0=d02-d13 */ IDCT4_1D( %%mm2, %%mm1, %%mm0, %%mm3, %%mm4, %%mm5 ) + "movq ff_pw_32, %%mm6 \n\t" /* in: 1,4,0,2 out: 1,2,3,0 */ TRANSPOSE4( %%mm1, %%mm4, %%mm0, %%mm2, %%mm3 ) + "paddw %%mm6, %%mm1 \n\t" + /* mm2=s02+s13 mm3=s02-s13 mm4=d02+d13 mm1=d02-d13 */ IDCT4_1D( %%mm3, %%mm2, %%mm1, %%mm0, %%mm4, %%mm5 ) @@ -675,7 +677,6 @@ void ff_h264_idct_add_mmx2(uint8_t *dst, int16_t *block, int stride) TRANSPOSE4( %%mm2, %%mm4, %%mm1, %%mm3, %%mm0 ) "pxor %%mm7, %%mm7 \n\t" - "movq ff_pw_32, %%mm6 \n\t" :: ); STORE_DIFF_4P( %%mm2, %%mm4, %%mm6, %%mm7, &dst[0*stride] ); |