summaryrefslogtreecommitdiff
path: root/libavcodec/i386/idct_mmx.c
diff options
context:
space:
mode:
authorMichael Niedermayer <michaelni@gmx.at>2005-06-01 08:43:40 +0000
committerMichael Niedermayer <michaelni@gmx.at>2005-06-01 08:43:40 +0000
commite4b36d443428f86d6f9db8af1897553818e76fa0 (patch)
treec3376e868077580614ed6ecc328b3a84182ef956 /libavcodec/i386/idct_mmx.c
parent85bbfcd4eead2d26df88de890962cbda97fac88b (diff)
avoid one transpose (730->680 dezicycles on duron)
Originally committed as revision 4332 to svn://svn.ffmpeg.org/ffmpeg/trunk
Diffstat (limited to 'libavcodec/i386/idct_mmx.c')
-rw-r--r--libavcodec/i386/idct_mmx.c11
1 files changed, 4 insertions, 7 deletions
diff --git a/libavcodec/i386/idct_mmx.c b/libavcodec/i386/idct_mmx.c
index 6ce73ae2c3..7e50f4c86b 100644
--- a/libavcodec/i386/idct_mmx.c
+++ b/libavcodec/i386/idct_mmx.c
@@ -673,14 +673,11 @@ void ff_h264_idct_add_mmx2(uint8_t *dst, int16_t *block, int stride)
/* mm2=s02+s13 mm3=s02-s13 mm4=d02+d13 mm1=d02-d13 */
IDCT4_1D( %%mm3, %%mm2, %%mm1, %%mm0, %%mm4, %%mm5 )
- /* in: 2,4,1,3 out: 2,3,0,1 */
- TRANSPOSE4( %%mm2, %%mm4, %%mm1, %%mm3, %%mm0 )
-
"pxor %%mm7, %%mm7 \n\t"
:: "m"(ff_pw_32));
- STORE_DIFF_4P( %%mm2, %%mm4, %%mm7, &dst[0*stride] );
- STORE_DIFF_4P( %%mm3, %%mm4, %%mm7, &dst[1*stride] );
- STORE_DIFF_4P( %%mm0, %%mm4, %%mm7, &dst[2*stride] );
- STORE_DIFF_4P( %%mm1, %%mm4, %%mm7, &dst[3*stride] );
+ STORE_DIFF_4P( %%mm2, %%mm0, %%mm7, &dst[0*stride] );
+ STORE_DIFF_4P( %%mm4, %%mm0, %%mm7, &dst[1*stride] );
+ STORE_DIFF_4P( %%mm1, %%mm0, %%mm7, &dst[2*stride] );
+ STORE_DIFF_4P( %%mm3, %%mm0, %%mm7, &dst[3*stride] );
}