[MPlayer-dev-eng] [PATCH] A new ASS renderer

Vladimir Mosgalin mosgalin at VM10124.spb.edu
Mon Sep 24 22:31:42 CEST 2012


Hi Reimar Döffinger!

 On 2012.09.23 at 21:32:07 +0200, Reimar Döffinger wrote next:

> 
> Have a sample at hand? The subtitles I have around are probably usually
> too simple.
> Also a "perf record"/"perf report" trace with MPlayer debug build and debug symbols
> for the driver binaries might help figuring out the cause.

Here are results of various tests:
http://178.130.36.83/perf-reports-all.tar.gz

Note that I used "scale" filter instead of -vf-clr option to get rid of
that crash that happens now and then on start with -vf-clr. It probably
should give more consistent comparison between -vf ass and vo_gl
rendering due to similar DR mode used; however, if it's a problem, I can
try to make tests with -vf-clr. Scale filter doesn't affect performance
much, it seems.

perf record/report output seems mostly meaningless, though. For example
ass-fullscreen-karaoke vs ass-window-karaoke tests - big difference in
CPU usage but same perf output. Or "isolate_freepages_block" which
appears only in glhack version (that's with your patch), but strangely
taking way more CPU usage in window.

Neither mplayer output nor perf seem to reflect how much mplayer was
lagging in gl-fullscreen-karaoke and glhack-fullscreen-karaoke modes,
with massive actual A-V difference due to sub rendering lagging so much.

Or, for example, gl-fullscreen-nokaraoke shows high framedrop but no
visible lag and gl-fullscreen-karaoke shows about the same framedrop but
with massive visible lags.
(-framedrop wasn't used in any of these tests)


I wonder if using system-wide perf record (starting before launching
mplayer, stopping soon after) would provide better results. But it will
include output from each and every app running on the system, so idk
if it'll be easy to analyze it.

-- 

Vladimir


More information about the MPlayer-dev-eng mailing list