[MPlayer-dev-eng] liba52 0.7.4

[Diego Biurrun]

On Mon, Jan 16, 2006 at 01:19:38AM +0100, Dominik 'Rathann' Mierzejewski wrote:
> On Saturday, 14 January 2006 at 19:23, Diego Biurrun wrote:
> > On Sat, Jan 07, 2006 at 01:29:41PM +0100, Emanuele Giaquinta wrote:
> > > 
> > > Apart from the changes already integrated upstream, I've not ported the ones to
> > > the non accelerated imdct in imdct.c (it has been totally rewritten).  I've
> > > tested it succesfully with the accelerated resample/downmix/imdct on
> > > ppc/altivec and x86/mmx+sse.  To test it create a new liba52 directory and do
> > > the following:
> > 
> > I'd appreciate if somebody could look into this.  We have a bug in
> > Bugzilla about failing AC3 samples that a newer version of liba52 should
> > hopefully be able to play.  Furthermore an upgrade promises much
> > improved performance.
> > 
> > Benchmarks comparing old and new version would be good.
> 
> My CPU is AMD Athlon Thunderbird 1GHz, MMX2, 3DNowEx, no SSE.
> MPlayer compiled with -O3 -ffast-math -march=athlon (which is the default).
> 
> With 5:32 long 448kbit/s AC3 5.1 sample, the new liba52 is slower:

Hmm, this is mysterious, since the liba52 homepage promises much
improved performance..  Did the 3dnow optimizations get ported from our
version to the updated one?

Diego