RepoMirrors/ffmpeg
1
0
Fork 0
mirror of https://git.ffmpeg.org/ffmpeg.git synced 2025-02-18 21:07:01 +00:00
Code Issues Projects Releases Wiki Activity

avutil/pixelutils: faster pixelutils_sad_[au]_16x16

Browse Source 
~560 → ~500 decicycles

This is following the comments from Michael in
https://ffmpeg.org/pipermail/ffmpeg-devel/2014-August/160599.html

Using 2 registers for accumulator didn't help. On the other hand,
some re-ordering between the movs and psadbw allowed going ~538 to ~500.
This commit is contained in:
Clément Bœsch 2014-08-14 22:30:55 +02:00
parent c82a288f87
commit 45c7f3997e
1 changed files with 9 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
libavutil/x86/pixelutils.asm
View File

@ -134,16 +134,20 @@ cglobal pixelutils_sad_16x16, 4,4,5, src1, stride1, src2, stride2
%macro SAD_XMM_16x16 1 %macro SAD_XMM_16x16 1
INIT_XMM sse2 INIT_XMM sse2
cglobal pixelutils_sad_%1_16x16, 4,4,3, src1, stride1, src2, stride2 cglobal pixelutils_sad_%1_16x16, 4,4,3, src1, stride1, src2, stride2
pxor m2, m2 mov%1 m2, [src2q]
%rep 8 psadbw m2, [src1q]
mov%1 m0, [src2q]
mov%1 m1, [src2q + stride2q] mov%1 m1, [src2q + stride2q]
psadbw m1, [src1q + stride1q]
paddw m2, m1
%rep 7
lea src1q, [src1q + 2*stride1q]
lea src2q, [src2q + 2*stride2q]
mov%1 m0, [src2q]
psadbw m0, [src1q] psadbw m0, [src1q]
mov%1 m1, [src2q + stride2q]
psadbw m1, [src1q + stride1q] psadbw m1, [src1q + stride1q]
paddw m2, m0 paddw m2, m0
paddw m2, m1 paddw m2, m1
lea src1q, [src1q + 2*stride1q]
lea src2q, [src2q + 2*stride2q]
%endrep %endrep
movhlps m0, m2 movhlps m0, m2
paddw m2, m0 paddw m2, m0