Skip to content

F
ffmpeg
Commit 0922c6b0 authored by James Almer's avatar James Almer
Browse files
Options

x86/lpc: use fused negative multiply-add instructions where useful 

Signed-off-by: 's avatarJames Almer <jamrial@gmail.com>
parent a1c6f4b6
Hide whitespace changes
Inline Side-by-side
Showing with 15 additions and 2 deletions
libavcodec/x86/lpc.asm
View file @ 0922c6b0
......@@ -79,11 +79,12 @@ cglobal lpc_apply_welch_window, 3, 5, 8, data, len, out, off1, off2
.loop_o:
movapd m1, m6
mulpd m2, m0, m0
subpd m1, m2
%if cpuflag(avx2)
fnmaddpd m1, m0, m0, m1
vpermpd m2, m1, q0123
%else
mulpd m2, m0, m0
subpd m1, m2
shufpd m2, m1, m1, 01b
%endif
......@@ -116,8 +117,12 @@ cglobal lpc_apply_welch_window, 3, 5, 8, data, len, out, off1, off2
.loop_o_scalar:
movapd xm1, xm6
%if cpuflag(avx2)
fnmaddpd xm1, xm0, xm0, xm1
%else
mulpd xm2, xm0, xm0
subpd xm1, xm2
%endif
cvtdq2pd xm3, [dataq + off1q]
cvtdq2pd xm4, [dataq + off2q]
......@@ -174,8 +179,12 @@ cglobal lpc_apply_welch_window, 3, 5, 8, data, len, out, off1, off2
.loop_e:
movapd m1, m6
%if cpuflag(avx2)
fnmaddpd m1, m0, m0, m1
%else
mulpd m2, m0, m0
subpd m1, m2
%endif
%if cpuflag(avx2)
vpermpd m2, m1, q0123
%else
......@@ -210,8 +219,12 @@ cglobal lpc_apply_welch_window, 3, 5, 8, data, len, out, off1, off2
.loop_e_scalar:
movapd xm1, xm6
%if cpuflag(avx2)
fnmaddpd xm1, xm0, xm0, xm1
%else
mulpd xm2, xm0, xm0
subpd xm1, xm2
%endif
cvtdq2pd xm3, [dataq + off1q]
cvtdq2pd xm4, [dataq + off2q]
......
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