Skip to content

F
ffmpeg
Commit b3446862 authored by James Almer's avatar James Almer
Browse files
Options

x86/vorbisdsp: optimize ff_vorbis_inverse_coupling_sse 

About 7% faster.
parent c0607d88
Hide whitespace changes
Inline Side-by-side
Showing with 11 additions and 8 deletions
libavcodec/x86/vorbisdsp.asm
View file @ b3446862
...@@ -57,13 +57,17 @@ cglobal vorbis_inverse_coupling, 3, 3, 6, mag, ang, block_size ...@@ -57,13 +57,17 @@ cglobal vorbis_inverse_coupling, 3, 3, 6, mag, ang, block_size
%endif %endif
INIT_XMM sse 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] mova m5, [pdw_80000000]
xor cntrq, cntrq shl block_sized, 2
add magq, block_sizeq
add angq, block_sizeq
neg block_sizeq
align 16 align 16
.loop: .loop:
mova m0, [magq+cntrq*4] mova m0, [magq+block_sizeq]
mova m1, [angq+cntrq*4] mova m1, [angq+block_sizeq]
xorps m2, m2 xorps m2, m2
xorps m3, m3 xorps m3, m3
cmpleps m2, m0 ; m <= 0.0 cmpleps m2, m0 ; m <= 0.0
...@@ -75,9 +79,8 @@ align 16 ...@@ -75,9 +79,8 @@ align 16
andnps m4, m1 andnps m4, m1
addps m3, m0 ; a = m + ((a < 0) & (a ^ sign(m))) addps m3, m0 ; a = m + ((a < 0) & (a ^ sign(m)))
subps m0, m4 ; m = m + ((a > 0) & (a ^ sign(m))) subps m0, m4 ; m = m + ((a > 0) & (a ^ sign(m)))
mova [angq+cntrq*4], m3 mova [angq+block_sizeq], m3
mova [magq+cntrq*4], m0 mova [magq+block_sizeq], m0
add cntrq, 4 add block_sizeq, mmsize
cmp cntrq, block_sizeq
jl .loop jl .loop
RET RET
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment