generic-library/ffmpeg
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

x86/sbrdsp: remove an unnecessary mova in sbr_autocorrelate

Browse Source 
Signed-off-by: James Almer <jamrial@gmail.com>
This commit is contained in:
James Almer 2015-08-03 19:25:43 -03:00
parent bee2e702cc
commit 9c0407e856
1 changed files with 17 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

34
libavcodec/x86/sbrdsp.asm
View File

@ -515,42 +515,42 @@ align 16
jl .loop jl .loop
movlhps m1, m1 movlhps m1, m1
mulps m4, m1, m2 mulps m2, m1
mulps m1, m1 mulps m1, m1
addps m4, m6 ; real_sum1 + x[38][0] * x[39][0], x[38][1] * x[39][1]; imag_sum1 + x[38][0] * x[39][1], x[38][1] * x[39][0]; addps m2, m6 ; real_sum1 + x[38][0] * x[39][0], x[38][1] * x[39][1]; imag_sum1 + x[38][0] * x[39][1], x[38][1] * x[39][0];
addps m1, m7 ; real_sum0 + x[38][0] * x[38][0], x[38][1] * x[38][1]; addps m1, m7 ; real_sum0 + x[38][0] * x[38][0], x[38][1] * x[38][1];
addps m6, [rsp ] ; real_sum1 + x[ 0][0] * x[ 1][0], x[ 0][1] * x[ 1][1]; imag_sum1 + x[ 0][0] * x[ 1][1], x[ 0][1] * x[ 1][0]; addps m6, [rsp ] ; real_sum1 + x[ 0][0] * x[ 1][0], x[ 0][1] * x[ 1][1]; imag_sum1 + x[ 0][0] * x[ 1][1], x[ 0][1] * x[ 1][0];
addps m7, [rsp+16] ; real_sum0 + x[ 0][0] * x[ 0][0], x[ 0][1] * x[ 0][1]; addps m7, [rsp+16] ; real_sum0 + x[ 0][0] * x[ 0][0], x[ 0][1] * x[ 0][1];
xorps m4, [ps_mask3] xorps m2, [ps_mask3]
xorps m5, [ps_mask3] xorps m5, [ps_mask3]
xorps m6, [ps_mask3] xorps m6, [ps_mask3]
%if cpuflag(sse3) %if cpuflag(sse3)
movshdup m2, m1 movshdup m0, m1
haddps m4, m5 haddps m2, m5
haddps m7, m6 haddps m7, m6
addss m1, m2 addss m1, m0
%else %else
movaps m3, m4 movaps m3, m2
movaps m2, m5 movaps m0, m5
movaps m0, m6 movaps m4, m6
shufps m3, m3, q0301 shufps m3, m3, q0301
shufps m2, m2, q0301
shufps m0, m0, q0301 shufps m0, m0, q0301
addps m4, m3 shufps m4, m4, q0301
addps m5, m2 addps m2, m3
addps m6, m0 addps m5, m0
addps m6, m4
movss m2, m7 movss m0, m7
movss m3, m1 movss m3, m1
shufps m7, m7, q0001 shufps m7, m7, q0001
shufps m1, m1, q0001 shufps m1, m1, q0001
addss m7, m2 addss m7, m0
addss m1, m3 addss m1, m3
shufps m4, m5, q2020 shufps m2, m5, q2020
shufps m7, m6, q2020 shufps m7, m6, q2020
%endif %endif
movaps [phiq ], m4 movaps [phiq ], m2
movhps [phiq+0x18], m7 movhps [phiq+0x18], m7
movss [phiq+0x28], m7 movss [phiq+0x28], m7
movss [phiq+0x10], m1 movss [phiq+0x10], m1