[FFmpeg-devel] [PATCH] Patch cleanup for MPEG 1 & 2 optimizations

Luca Barbato lu_zero
Thu Apr 17 01:55:29 CEST 2008


Michael Niedermayer wrote:
> Jeffs benchmarks indicated that there still was a speed difference left
> but it could be as well random variation. If there is still a difference
> left i would like to know what is causing this difference. Is it just
> the overhead for the extra if() and call or is there more?

I tried some benchmarks myself on a different file (on a phenom and on a 
cell)

this is the phenom

time for a in `seq 1 10` ; do ./ffmpeg_inlined -y -benchmark -i 
'/tmp/test1.mpg' -f rawvideo /dev/null 2>/dev/null; done
bench: utime=6.992s
bench: utime=6.928s
bench: utime=6.928s
bench: utime=6.936s
bench: utime=6.980s
bench: utime=6.952s
bench: utime=6.964s
bench: utime=6.932s
bench: utime=6.944s
bench: utime=7.112s

real	1m10.543s
user	1m9.756s
sys	0m0.792s

time for a in `seq 1 10` ; do ./ffmpeg_hardcoded -y -benchmark -i 
'/tmp/test1.mpg' -f rawvideo /dev/null 2>/dev/null; done
bench: utime=7.064s
bench: utime=7.012s
bench: utime=6.976s
bench: utime=7.032s
bench: utime=6.904s
bench: utime=7.012s
bench: utime=6.940s
bench: utime=6.956s
bench: utime=7.004s
bench: utime=6.944s

real	1m10.664s
user	1m9.932s
sys	0m0.736s

this is the cell

time for a in `seq 1 10` ; do ./ffmpeg_inlined -y -benchmark -i 
'/tmp/test1.mpg' -f rawvideo /dev/null 2>/dev/null; done
bench: utime=27.354s
bench: utime=27.089s
bench: utime=27.353s
bench: utime=27.052s
bench: utime=26.898s
bench: utime=27.568s
bench: utime=26.829s
bench: utime=27.426s
bench: utime=27.159s
bench: utime=27.219s

real	4m38.073s
user	4m32.165s
sys	0m5.878s

time for a in `seq 1 10` ; do ./ffmpeg_hardcoded -y -benchmark -i 
'/tmp/test1.mpg' -f rawvideo /dev/null 2>/dev/null; done
bench: utime=27.111s
bench: utime=26.804s
bench: utime=27.060s
bench: utime=26.760s
bench: utime=26.900s
bench: utime=27.011s
bench: utime=26.754s
bench: utime=27.041s
bench: utime=27.023s
bench: utime=27.079s

real	4m35.595s
user	4m29.766s
sys	0m5.826s

As you can see different arches, apparently different results.

lu

-- 

Luca Barbato
Gentoo Council Member
Gentoo/linux Gentoo/PPC
http://dev.gentoo.org/~lu_zero





More information about the ffmpeg-devel mailing list