[MPlayer-users] Problem with Windows Media audio

[Vladimir Mosgalin]

Hi Rich Felker!

 On 2005.06.16 at 11:55:46 -0400, Rich Felker wrote next:

> > > > >   Thanks a lot, I tried -afm dmo, works perfectly! I don't think I'd run
> > > > >   across that one yet. :)
> > > > 
> > > > Well, next times you experience problems with wmv1/2 codecs, try
> > > > -vfm dmo ;) It helps even more often than with wma. You'll lose nice
> > > > postprocessing, though, but that's much better than watching as your
> > > > video is falling apart (and what the hell are those j-frames, does
> > > > anyone know?).
> > > > 
> > > > But afm = dmo has no drawbacks, that's why I suggested to append it to
> > > > config file and forget about this problem.
> > > 
> > > No drawbacks?? Won't it cause windows dll crap to get loaded for many
> > > formats that are playable with free decoders, resulting in worse
> > > performance if nothing else..?
> > 
> > Worse performance for audio decoding? No, not really. Besides, (AFAIR it
> > I heard it on this list), dmo codecs are heavily optimized and they
> > quite fast. I checked it now, ffmpeg audio decoder uses 0.3-0.4% cpu on
> > wma2 files I tested. acm decoder uses 0.4-0.5% cpu. dmo one - 0.5-0.6%.
> 
> You're right that they're optimized. In fact they depend on particular
> cpu extensions (mmx2? sse?) and will crash with invalid opcode on
> older cpus.

I thought that windows is supposed to work even on P1, no? Microsoft
surely uses a lot of bad tricks, but I don't remember MS forcing people
to upgrade your CPU. They are concerned about slow systems too.

> Anyway your cpu %'s already show that they're slower.

They can't be trusted with audio files, for example after any seeking
numbers are always at zero.

Anyway! Instead of that small 22khz file, I found another one -
title.wma ("No Hay Problema" by Pink Martini) from standart windows
distribution (%windir%/system32/oobe/images/title.wma). 64kbit 44khz
wma2 file which is 6 minutes long. It shows these numbers:

With -afm ffmpeg:
1.84user 0.02system 0:01.88elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+3491minor)pagefaults 0swaps
With -afm dmo:
1.29user 0.03system 0:01.38elapsed 96%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+3554minor)pagefaults 0swaps
With -afm acm:
2.99user 0.03system 0:03.05elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+3676minor)pagefaults 0swaps

The numbers don't change in multiple tests (less than 0.01s difference).
CPU load is alywas 96% with dmo, don't know why.

The conclusion? Don't use ffmpeg for decoding wma, period. And don't
even think about using acm/dshow codecs for wma/wmv, dmo ones are
better.

-- 

Vladimir