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

avutil/pixelutils: faster pixelutils_sad_16x16

Browse Source 
501 to 439 decicycles.

See 45c7f3997e.
This commit is contained in:
Clément Bœsch 2014-08-23 20:03:10 +02:00
parent f4dec0dba0
commit 554d819062
1 changed files with 12 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
libavutil/x86/pixelutils.asm
View File

@ -109,18 +109,24 @@ cglobal pixelutils_sad_16x16, 4,4,0, src1, stride1, src2, stride2
;-------------------------------------------------------------------------------
INIT_XMM sse2
cglobal pixelutils_sad_16x16, 4,4,5, src1, stride1, src2, stride2
pxor m4, m4
%rep 8
movu m0, [src1q]
movu m1, [src1q + stride1q]
movu m4, [src1q]
movu m2, [src2q]
movu m1, [src1q + stride1q]
movu m3, [src2q + stride2q]
psadbw m4, m2
psadbw m1, m3
paddw m4, m1
%rep 7
lea src1q, [src1q + 2*stride1q]
lea src2q, [src2q + 2*stride2q]
movu m0, [src1q]
movu m2, [src2q]
movu m1, [src1q + stride1q]
movu m3, [src2q + stride2q]
psadbw m0, m2
psadbw m1, m3
paddw m4, m0
paddw m4, m1
lea src1q, [src1q + 2*stride1q]
lea src2q, [src2q + 2*stride2q]
%endrep
movhlps m0, m4
paddw m4, m0