diff options
| author | Michael Niedermayer <michaelni@gmx.at> | 2007-08-20 22:41:47 +0000 |
|---|---|---|
| committer | Michael Niedermayer <michaelni@gmx.at> | 2007-08-20 22:41:47 +0000 |
| commit | be3b22f9cae9916405cde209fd93f8593e52ed4b (patch) | |
| tree | 0bf44e5716c9f460909a56feee968c375d324eaa /libavcodec/i386/snowdsp_mmx.c | |
| parent | 8510c519e7e2c666f84ba120245bd13790c4e22e (diff) | |
| download | ffmpeg-be3b22f9cae9916405cde209fd93f8593e52ed4b.tar.gz | |
remove idiotc double subtraction from the sse2 code (untested, no sse2 here)
Originally committed as revision 10158 to svn://svn.ffmpeg.org/ffmpeg/trunk
Diffstat (limited to 'libavcodec/i386/snowdsp_mmx.c')
| -rw-r--r-- | libavcodec/i386/snowdsp_mmx.c | 33 |
1 files changed, 15 insertions, 18 deletions
diff --git a/libavcodec/i386/snowdsp_mmx.c b/libavcodec/i386/snowdsp_mmx.c index 053e4bde69..96cb856e94 100644 --- a/libavcodec/i386/snowdsp_mmx.c +++ b/libavcodec/i386/snowdsp_mmx.c @@ -111,7 +111,8 @@ void ff_snow_horizontal_compose97i_sse2(DWTELEM *b, int width){ i = 0; asm volatile( - "pslld $1, %%xmm7 \n\t" /* xmm7 already holds a '4' from 2 lifts ago. */ + "pcmpeqd %%xmm7, %%xmm7 \n\t" + "psrad $29, %%xmm7 \n\t" ::); for(; i<w_l-7; i+=8){ asm volatile( @@ -121,22 +122,18 @@ void ff_snow_horizontal_compose97i_sse2(DWTELEM *b, int width){ "movdqu 20(%1), %%xmm4 \n\t" //FIXME try aligned reads and shifts "paddd %%xmm1, %%xmm0 \n\t" "paddd %%xmm5, %%xmm4 \n\t" - "movdqa %%xmm7, %%xmm1 \n\t" - "movdqa %%xmm7, %%xmm5 \n\t" - "psubd %%xmm0, %%xmm1 \n\t" - "psubd %%xmm4, %%xmm5 \n\t" - "movdqa (%0), %%xmm0 \n\t" - "movdqa 16(%0), %%xmm4 \n\t" - "pslld $2, %%xmm0 \n\t" - "pslld $2, %%xmm4 \n\t" - "psubd %%xmm0, %%xmm1 \n\t" - "psubd %%xmm4, %%xmm5 \n\t" - "psrad $4, %%xmm1 \n\t" - "psrad $4, %%xmm5 \n\t" - "movdqa (%0), %%xmm0 \n\t" - "movdqa 16(%0), %%xmm4 \n\t" - "psubd %%xmm1, %%xmm0 \n\t" - "psubd %%xmm5, %%xmm4 \n\t" + "paddd %%xmm7, %%xmm0 \n\t" + "paddd %%xmm7, %%xmm4 \n\t" + "movdqa (%0), %%xmm1 \n\t" + "movdqa 16(%0), %%xmm5 \n\t" + "psrad $2, %%xmm0 \n\t" + "psrad $2, %%xmm4 \n\t" + "paddd %%xmm1, %%xmm0 \n\t" + "paddd %%xmm5, %%xmm4 \n\t" + "psrad $2, %%xmm0 \n\t" + "psrad $2, %%xmm4 \n\t" + "paddd %%xmm1, %%xmm0 \n\t" + "paddd %%xmm5, %%xmm4 \n\t" "movdqa %%xmm0, (%0) \n\t" "movdqa %%xmm4, 16(%0) \n\t" :: "r"(&b[i]), "r"(&ref[i]) @@ -144,7 +141,7 @@ void ff_snow_horizontal_compose97i_sse2(DWTELEM *b, int width){ ); } snow_horizontal_compose_liftS_lead_out(i, b, b, ref, width, w_l); - b[0] = b_0 - (((-2 * ref[1] + W_BO) - 4 * b_0) >> W_BS); + b[0] = b_0 + ((2 * ref[1] + W_BO-1 + 4 * b_0) >> W_BS); } { // Lift 3 |