[MPlayer-dev-eng] Make ffmpeg MPEG1/2 decoder the default?

[Diego Biurrun]

On Mon, Feb 18, 2008 at 12:47:52PM -0500, Rich Felker wrote:
> On Mon, Feb 18, 2008 at 06:17:42PM +0100, Diego Biurrun wrote:
> > On Mon, Feb 18, 2008 at 06:12:29PM +0100, Reimar Döffinger wrote:
> > > On Mon, Feb 18, 2008 at 10:18:38AM +0100, Diego Biurrun wrote:
> > > > On Sat, Feb 16, 2008 at 03:38:36PM +0100, Reimar Döffinger wrote:
> > > > > libmpeg2 may be still slightly faster (though probably not on the newer CPUs),
> > > > > but it is quite crash-happy and otherwise buggy (see also 34582.mxf in
> > > > > incoming).
> > > > > Thus I am for making FFmpeg decoder the default, any objections?
> > > > 
> > > > I'm not too convinced of this since libmpeg2 was still faster the last
> > > > time I tried and now we have a pending patch with further speedup.  I
> > > > propose running benchmarks on many different CPUs instead.  Let's find
> > > > out which one really is the fastest.  Maybe we can motivate somebody to
> > > > try making FFmpeg speedier...
> > > 
> > > My benchmarks say it completely depends on the CPU, though the speed on
> > > the old Athlon would be more relevant in practice (though I doubt that many
> > > are still in use).
> > 
> > Why aren't you using -vo null like we have always done for these
> > benchmarks?
> 
> Because it hides all sorts of performance data which is relevant. I
> care about the performance difference between decoders when memcpy to
> video memory and execution of vo code is contending for the cache, not
> in some idealized-but-irrelevant environment with no vo.

OK, so let's do this properly: We need a standard command line that
everybody uses for testing, a set of sample files that exercise
different parts of the decoders and some testing methodology.

I propose the following command lines:

-vc ffmpeg12 -benchmark -nosound -quiet -vo null
-vc ffmpeg12 -benchmark -nosound -quiet -vo x11
-vc mpeg12 -benchmark -nosound -quiet -vo null
-vc mpeg12 -benchmark -nosound -quiet -vo x11

There should be multiple runs of each file, I'm not sure whether to use
the fastest result or the average of the runs.

What would be good samples for testing?  Some MPEG-1 file, some DVD
content, some HD content, maybe something else?

Diego