virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2024-12-23 12:43:46 +02:00
Code Issues Releases Activity

x86/aacpsdsp: optimize add_squares loop

Browse Source 
Signed-off-by: James Almer <jamrial@gmail.com>
This commit is contained in:
James Almer 2016-06-14 12:41:23 -03:00
parent d5ded429f2
commit ede4ec1f8f
1 changed files with 9 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
libavcodec/x86/aacpsdsp.asm
View File

@ -33,18 +33,22 @@ SECTION .text
;************************************************************************* ;*************************************************************************
%macro PS_ADD_SQUARES 1 %macro PS_ADD_SQUARES 1
cglobal ps_add_squares, 3, 3, %1, dst, src, n cglobal ps_add_squares, 3, 3, %1, dst, src, n
shl nd, 3
add srcq, nq
neg nq
align 16
.loop: .loop:
movaps m0, [srcq] movaps m0, [srcq+nq]
movaps m1, [srcq+mmsize] movaps m1, [srcq+nq+mmsize]
mulps m0, m0 mulps m0, m0
mulps m1, m1 mulps m1, m1
HADDPS m0, m1, m2 HADDPS m0, m1, m2
addps m0, [dstq] addps m0, [dstq]
movaps [dstq], m0 movaps [dstq], m0
add dstq, mmsize add dstq, mmsize
add srcq, mmsize*2 add nq, mmsize*2
sub nd, mmsize/4 jl .loop
jg .loop
REP_RET REP_RET
%endmacro %endmacro