[MPlayer-dev-eng] [PATCH] remove mp3lib

[Reimar Döffinger]

On Mon, Dec 05, 2011 at 10:50:25AM +0100, Thomas Orgis wrote:
> Hi... long time no see.
> 
> I started working on the issue of mp3lib vs. libmpg123 and so far had
> some interesting insights. Yes, there are still some improvements to
> merge in from mp3lib to mpg123 code! That does not already explain the
> distinct drop in performance when using libmpg123 from MPlayer instad
> of stand-alone mpg123, but it's a step towards parity.
> 
> I do have both a K6-3+ @550 MHz and a Duron @800 MHz set up here as
> test systems. On the latter, I can use perf to get some counters. Have
> a read about my ongoing efforts on
> 
> 	svn://svn.orgis.org/mpg123/notes/beating_mp3lib_in_mplayer.tex
> 
> (or
> http://scm.orgis.org/view/mpg123/notes/beating_mp3lib_in_mplayer.tex?revision=2988&content-type=text%2Fplain)
> 
> I can easily reproduce significant performance differences on those
> systems, more than you on your Athlon, even.
> 
> I do have a blunt question regarding your diagnostics, though.
> 
> Am Tue, 27 Sep 2011 04:39:12 +0300
> schrieb Ivan Kalvachev <ikalvachev at gmail.com>: 
> 
> > athlon# perf record -e cycles:u -e instructions:u -e l1-dcache-loads:u
> > -e l1-dcache-load-misses:u -e branch-misses ./mplayer -loop 10
> > -benchmark -ao pcm:fast:file=/dev/null -quiet -nocache stream.dump -ac
> > mp3
> > athlon# perf report
> > 
> > # Events: 2K cycles
> > 
> > # Events: 2K instructions
> 
> > # Events: 2K L1-dcache-loads
> 
> > # Events: 1K L1-dcache-load-misses
> 
> > # Events: 2K branch-misses
> 
> I can run the perf report command fine, but how do you extract those
> reports about the different events/counters exactly? Just running 'perf
> report' does always give me the same picture about overall overhead.
> Sorry if this is RTFM material, but I managed to not stumble over this
> info:-/

perf list
and then use those with -e, e.g. perf record -e L1-dcache-loads
(I didn't actually test it though).