[MPlayer-dev-eng] [RFC] internal library copies

[Diego Biurrun]

On Sat, Mar 27, 2010 at 08:55:58PM +0100, Reimar Döffinger wrote:
> On Thu, Mar 25, 2010 at 07:09:56PM +0100, Diego Biurrun wrote:
> > On Wed, Mar 17, 2010 at 07:50:08PM +0100, Reimar Döffinger wrote:
> > > On Wed, Mar 17, 2010 at 01:03:21PM +0100, Diego Biurrun wrote:
> > > > * liba52
> > > > 
> > > > The internal copy was disabled some time ago, FFmpeg is both faster and
> > > > more featureful.  We support building against the external version.  Our
> > > > copy is patched, but upstream is dead and the patches have been rejected
> > > > IIRC.
> > > > verdict: drop
> > > 
> > > Hm? IIRC FFmpeg was a lot slower originally and even now is still a bit slower.
> > > And upstream is a lot slower, so it seems to me that removing it would remove
> > > the fastest decoder available.
> > > Not that I mind much.
> > 
> > Hmm, you disabled liba52.  Can somebody provide numbers?  I think FFmpeg
> > was faster, even for my K6-III...
> 
> These are the numbers I got back then:
> As of 2010-01-04 mplayer/ffmpeg-mt:
> $ time mplayer -nocache -novideo -vc null -vo null \
> -ao pcm:fast:nowaveheader:file=/dev/null -benchmark /path/to/file.mkv
> 
> 2ch/224kbps/-aid 0
> liba52: real    0m10.027s user   0m9.433s sys   0m0.424s
> ffac3:  real    0m11.079s user  0m10.289s sys   0m0.532s
> 6ch/448kbps->downmix 2ch/-aid 1 -channels 2
> liba52: real    0m17.246s user  0m16.269s sys   0m0.564s
> ffac3:  real    0m19.884s user  0m18.873s sys   0m0.536s
> 6ch/448kbps->6ch/-aid 1 -channels 6
> liba52: real    0m26.187s user  0m24.410s sys   0m0.816s
> ffac3:  real    0m27.093s user  0m25.790s sys   0m0.808s

So these are your current numbers?  What sort of system did you test on?
Where are the samples?

FFmpeg was faster in two out of three cases for you, all cases for Loren.
I'll look into running some benchmarks myself.

> > > > * libmpeg2
> > > > 
> > > > Nowadays FFmpeg is faster and should be better maintained.  Unfortunately
> > > > we cannot compile against the external version.  This should not be hard
> > > > to fix.  If it gets fixed, we could remove libmpeg2 IMO.
> > > > Is anybody interested in fixing external libmpeg2 compilation?
> > > 
> > > Not particularly, but I wonder when FFmpeg got faster?
> > > IIRC it was always slower... Of course that might depend on the CPU,
> > > libmpeg2 only supports up to SSE I think, so with CPUs that support
> > > more than that libavcodec probably has a chance.
> > 
> > Well, you switched to FFmpeg as the default decoder for MPEG-2 (IIRC)..
> > I don't think there is a point in keeping local copies of libraries that
> > are not even used by default...
> 
> That was due to stability reasons, not speed.
> However I don't really mind removing it if nobody has any reasons to use it I forgot
> (does dvdnav work with ffmpeg12 nowadays?).
> 
> There's one missing:
> mp3lib.
> IIRC everything relevant has been merged into libmpg123, development on that is still active,
> so all that's needed is someone either writing a ad_mpg123 or changing ad_mp3lib to work with
> the external one and we can remove it IMO.

Agreed.

Diego