[MEncoder-users] [BUG] non-deterministic PSNR calculation with -lavcopts threads=2

[Corey Hickey]

I'm currently in the final stage of preparing an lavc mpeg-4 mencoder
command for Doom9: testing threads=2 and seeing how much quality is
lost. I have a full encode going right now, but meanwhile I've been
testing a short clip.

Quality is visibly a bit lower on the clip I tested when two threads are
used, but, oddly enough, the PSNR shown is not only about 4-5% higher,
but varies somewhat each time I run mencoder. The output file is the
same each time, however, so the encoding itself should be correct.

I haven't used TOOLS/psnr-video.sh yet to see what the PSNR really is,
but the variation didn't look right.

You can test it with this file:
http://fatooh.org/files/doom9tmp/matrix-clip.vob

...and this command:
 mencoder matrix-clip.vob -nosound -ovc lavc \
-lavcopts vcodec=mpeg4:threads=2:psnr -ofps 24000/1001

Thanks,
Corey