RepoMirrors/ffmpeg
1
0
Fork 0
mirror of https://git.ffmpeg.org/ffmpeg.git synced 2025-04-25 12:48:03 +00:00
Code Issues Projects Releases Wiki Activity

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

Browse Source 
Signed-off-by: James Almer <jamrial@gmail.com>
This commit is contained in:
James Almer 2022-09-22 17:10:37 -03:00
parent a1c6f4b653
commit 0922c6b01b
1 changed files with 15 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
libavcodec/x86/lpc.asm
View File

@ -79,11 +79,12 @@ cglobal lpc_apply_welch_window, 3, 5, 8, data, len, out, off1, off2
.loop_o: .loop_o:
movapd m1, m6 movapd m1, m6
mulpd m2, m0, m0
subpd m1, m2
%if cpuflag(avx2) %if cpuflag(avx2)
fnmaddpd m1, m0, m0, m1
vpermpd m2, m1, q0123 vpermpd m2, m1, q0123
%else %else
mulpd m2, m0, m0
subpd m1, m2
shufpd m2, m1, m1, 01b shufpd m2, m1, m1, 01b
%endif %endif
@ -116,8 +117,12 @@ cglobal lpc_apply_welch_window, 3, 5, 8, data, len, out, off1, off2
.loop_o_scalar: .loop_o_scalar:
movapd xm1, xm6 movapd xm1, xm6
%if cpuflag(avx2)
fnmaddpd xm1, xm0, xm0, xm1
%else
mulpd xm2, xm0, xm0 mulpd xm2, xm0, xm0
subpd xm1, xm2 subpd xm1, xm2
%endif
cvtdq2pd xm3, [dataq + off1q] cvtdq2pd xm3, [dataq + off1q]
cvtdq2pd xm4, [dataq + off2q] cvtdq2pd xm4, [dataq + off2q]
@ -174,8 +179,12 @@ cglobal lpc_apply_welch_window, 3, 5, 8, data, len, out, off1, off2
.loop_e: .loop_e:
movapd m1, m6 movapd m1, m6
%if cpuflag(avx2)
fnmaddpd m1, m0, m0, m1
%else
mulpd m2, m0, m0 mulpd m2, m0, m0
subpd m1, m2 subpd m1, m2
%endif
%if cpuflag(avx2) %if cpuflag(avx2)
vpermpd m2, m1, q0123 vpermpd m2, m1, q0123
%else %else
@ -210,8 +219,12 @@ cglobal lpc_apply_welch_window, 3, 5, 8, data, len, out, off1, off2
.loop_e_scalar: .loop_e_scalar:
movapd xm1, xm6 movapd xm1, xm6
%if cpuflag(avx2)
fnmaddpd xm1, xm0, xm0, xm1
%else
mulpd xm2, xm0, xm0 mulpd xm2, xm0, xm0
subpd xm1, xm2 subpd xm1, xm2
%endif
cvtdq2pd xm3, [dataq + off1q] cvtdq2pd xm3, [dataq + off1q]
cvtdq2pd xm4, [dataq + off2q] cvtdq2pd xm4, [dataq + off2q]