RepoMirrors
/
ffmpeg
mirror of https://git.ffmpeg.org/ffmpeg.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

remove idiotc double subtraction from the sse2 code (untested, no sse2 here) 

Originally committed as revision 10158 to svn://svn.ffmpeg.org/ffmpeg/trunk
This commit is contained in:
Michael Niedermayer 2007-08-20 22:41:47 +00:00
parent 8510c519e7
commit be3b22f9ca
1 changed files with 15 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

33
libavcodec/i386/snowdsp_mmx.c
View File

@ -111,7 +111,8 @@ void ff_snow_horizontal_compose97i_sse2(DWTELEM *b, int width){
i = 0; i = 0;
asm volatile( 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){ for(; i<w_l-7; i+=8){
asm volatile( 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 "movdqu 20(%1), %%xmm4 \n\t" //FIXME try aligned reads and shifts
"paddd %%xmm1, %%xmm0 \n\t" "paddd %%xmm1, %%xmm0 \n\t"
"paddd %%xmm5, %%xmm4 \n\t" "paddd %%xmm5, %%xmm4 \n\t"
"movdqa %%xmm7, %%xmm1 \n\t" "paddd %%xmm7, %%xmm0 \n\t"
"movdqa %%xmm7, %%xmm5 \n\t" "paddd %%xmm7, %%xmm4 \n\t"
"psubd %%xmm0, %%xmm1 \n\t" "movdqa (%0), %%xmm1 \n\t"
"psubd %%xmm4, %%xmm5 \n\t" "movdqa 16(%0), %%xmm5 \n\t"
"movdqa (%0), %%xmm0 \n\t" "psrad $2, %%xmm0 \n\t"
"movdqa 16(%0), %%xmm4 \n\t" "psrad $2, %%xmm4 \n\t"
"pslld $2, %%xmm0 \n\t" "paddd %%xmm1, %%xmm0 \n\t"
"pslld $2, %%xmm4 \n\t" "paddd %%xmm5, %%xmm4 \n\t"
"psubd %%xmm0, %%xmm1 \n\t" "psrad $2, %%xmm0 \n\t"
"psubd %%xmm4, %%xmm5 \n\t" "psrad $2, %%xmm4 \n\t"
"psrad $4, %%xmm1 \n\t" "paddd %%xmm1, %%xmm0 \n\t"
"psrad $4, %%xmm5 \n\t" "paddd %%xmm5, %%xmm4 \n\t"
"movdqa (%0), %%xmm0 \n\t"
"movdqa 16(%0), %%xmm4 \n\t"
"psubd %%xmm1, %%xmm0 \n\t"
"psubd %%xmm5, %%xmm4 \n\t"
"movdqa %%xmm0, (%0) \n\t" "movdqa %%xmm0, (%0) \n\t"
"movdqa %%xmm4, 16(%0) \n\t" "movdqa %%xmm4, 16(%0) \n\t"
:: "r"(&b[i]), "r"(&ref[i]) :: "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); 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 { // Lift 3