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

[strites]

I am actually benchmarking those patches. I found a combination that 
gave speed improvements, but messed with git and lost it. I am trying to 
redo it.

In the meanwhile, I did some test with "av_always_inline", and I am 
obtaining a slight speed improvement.

What I did:
Create 2 files, mpg1 and mpg2. I used the method described by Robert 
Swain. (input file is a 1024x576 video - an extract of a japanese drama)

I used two copies of the tree, one without patches (named "vanilla") and 
one with the 6 patches submitted and the seventh provided as attachment 
(may need adjustment - I made this file with git diff).


MPEG 1 FILE

vanilla:
time for a in `seq 1 10` ; do ./ffmpeg -y -benchmark -i 
'/home/strites/RAM/testmpg1.mpg' -f rawvideo /dev/null 2>/dev/null; done
bench: utime=27.640s
bench: utime=27.630s
bench: utime=27.594s
bench: utime=27.750s
bench: utime=27.848s
bench: utime=27.673s
bench: utime=27.776s
bench: utime=27.707s
bench: utime=27.775s
bench: utime=27.843s

real    4m40.284s
user    4m37.340s
sys     0m2.390s

patched:
time for a in `seq 1 10` ; do ./ffmpeg -y -benchmark -i 
'/home/strites/RAM/testmpg1.mpg' -f rawvideo /dev/null 2>/dev/null; done
bench: utime=27.680s
bench: utime=28.066s
bench: utime=27.707s
bench: utime=27.883s
bench: utime=27.881s
bench: utime=28.311s
bench: utime=27.983s
bench: utime=27.682s
bench: utime=27.733s
bench: utime=27.809s

real    4m41.804s
user    4m38.834s
sys     0m2.380s


MPEG2 FILE

vanilla:
time for a in `seq 1 10` ; do ./ffmpeg -y -benchmark -i 
'/home/strites/RAM/testmpg2.mpg' -f rawvideo /dev/null 2>/dev/null; done
bench: utime=29.538s
bench: utime=29.213s
bench: utime=29.187s
bench: utime=29.316s
bench: utime=29.897s
bench: utime=29.150s
bench: utime=29.393s
bench: utime=29.320s
bench: utime=29.500s
bench: utime=29.382s

real    4m56.940s
user    4m54.002s
sys     0m2.280s


patched:
time for a in `seq 1 10` ; do ./ffmpeg -y -benchmark -i 
'/home/strites/RAM/testmpg2.mpg' -f rawvideo /dev/null 2>/dev/null; done
bench: utime=29.083s
bench: utime=29.453s
bench: utime=29.384s
bench: utime=29.313s
bench: utime=30.144s
bench: utime=29.133s
bench: utime=29.232s
bench: utime=29.576s
bench: utime=29.138s
bench: utime=29.264s

real    4m56.835s
user    4m53.816s
sys     0m2.412s

Seems there is no speed gain at the moment. I am further investigating.
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: 0007-inline-modification.patch
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20080416/9ca2e83c/attachment.asc>