[FFmpeg-devel] [PATCH] Further optimization for vp3.c:unpack_vlcs()

Mike Melanson mike
Sun Sep 20 08:38:31 CEST 2009


Reimar had a good idea with this patch:

http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/2008-June/047993.html

I got rid of a bunch of other structure dereferences in the inner loop. 
For one sample I was using (and profiling the calls to the unpack_vlcs() 
for the DC coefficients), these are the before and after numbers:

original:
3019769 dezicycles in unpack_vlcs (DC coeffs), 2048 runs, 0 skips

faster:
2894529 dezicycles in unpack_vlcs (DC coeffs), 2048 runs, 0 skips

Further, when using OProfile, I see unpack_vlcs()'s overall percentage 
drop from 44.1% -> 41.9% when decoding the first 15 seconds of Big Buck 
Bunny/1080p.

-- 
     -Mike Melanson
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: faster-unpack-vlcs.patch
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20090919/c28ad7f6/attachment.asc>



More information about the ffmpeg-devel mailing list