xenia-project/FFmpeg
1
0
Fork 0
mirror of https://github.com/xenia-project/FFmpeg.git synced 2024-11-27 21:40:34 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

x86/vorbisdsp: optimize ff_vorbis_inverse_coupling_sse

Browse Source 
About 7% faster.
This commit is contained in:
James Almer 2017-06-15 23:20:05 -03:00
parent c0607d88ee
commit b3446862bf
1 changed files with 11 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
libavcodec/x86/vorbisdsp.asm
View File

@ -57,13 +57,17 @@ cglobal vorbis_inverse_coupling, 3, 3, 6, mag, ang, block_size
%endif
INIT_XMM sse
cglobal vorbis_inverse_coupling, 3, 4, 6, mag, ang, block_size, cntr
cglobal vorbis_inverse_coupling, 3, 3, 6, mag, ang, block_size
mova m5, [pdw_80000000]
xor cntrq, cntrq
shl block_sized, 2
add magq, block_sizeq
add angq, block_sizeq
neg block_sizeq
align 16
.loop:
mova m0, [magq+cntrq*4]
mova m1, [angq+cntrq*4]
mova m0, [magq+block_sizeq]
mova m1, [angq+block_sizeq]
xorps m2, m2
xorps m3, m3
cmpleps m2, m0 ; m <= 0.0
@ -75,9 +79,8 @@ align 16
andnps m4, m1
addps m3, m0 ; a = m + ((a < 0) & (a ^ sign(m)))
subps m0, m4 ; m = m + ((a > 0) & (a ^ sign(m)))
mova [angq+cntrq*4], m3
mova [magq+cntrq*4], m0
add cntrq, 4
cmp cntrq, block_sizeq
mova [angq+block_sizeq], m3
mova [magq+block_sizeq], m0
add block_sizeq, mmsize
jl .loop
RET