[vorbis-dev] alloca() problem with CodeWarrior

[Firelight Technologies]

> >
> > Are you saying it takes only 22cycles per sample (at 44khz) to
decode a
> > vorbis track? (or .1% if you mean a 1ghz g4).  Some pcm software
mixers
> > are only this fast (well it's a bit slow for a mixer but not
unusual).
> > I find this hard to believe when the current reference decoder takes
> > about 20-30 times more cpu than that on my 1.2 ghz machine.
> 
> I mean it takes .17% time on my ppc7410/500MHz.  This is more than 19
> cycles per output sample, which is perfectly reasonable.  Anything a
> lot slower is just badly written (or not written for good
performance).

<p>This sounds amazingly fast .. you are actually cycle counting this and
not using 3rd party cpu usage meters?  Just looking at something like
mdct.c it seems the sheer number of multiplies makes it impossible to
execute in 19 cycles without vectorizing it massively and doing the
decode in a different way (im not sure how wide the powerpc is but I
assume it is 128bit max).. this is your own private source branch right?
as it sounds impossible for the reference implementation to get anywhere
near that - without vectorization adpcm probably wouldn't even be able
to decode in 19 cycles per output sample.

<p>--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to 'vorbis-dev-request at xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is needed.
Unsubscribe messages sent to the list will be ignored/filtered.