[FFmpeg-devel] [PATCH] adpcm: Collapse nodes with similar state only after checking ssd

[Martin Storsjö]

Hi,

In the attached patch, I shuffled around some checks in the adpcm trellis 
code, moving the costly check for collapsing nodes with similar state to 
after checking if the ssd is worse than the one to be replaced.

With this patch in place, the run time for -trellis 8 on one test sample 
drops from 158 to 79 seconds, on my machine.

Runtime vs PSNR graphs are available at 
http://albin.abo.fi/~mstorsjo/adpcm-graphs2/. For the music1 and music2 
samples, this gives quite clear improvments in performance. For the 
trailer sample, I notice that all three algorithms seem to converge to the 
same output quality at the end.

// Martin
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-adpcm-Collapse-nodes-with-similar-state-only-after-c.patch
Type: text/x-diff
Size: 3611 bytes
Desc: 
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20101115/cb6fbf27/attachment.patch>