What fraction of work is iDCT?

[Jeffrey W. Baker]

On Mon, 25 Jun 2001, Christophe Massiot wrote:

> On Mon, Jun 25, 2001, Eugenio Jarosiewicz wrote:
>
> > Ok, so someone else who knows more about this than me already gave you
> > some #'s.  But of quite interest might be:
> >
> > Altivec IDCT sample plus much more...
> >
> > http://developer.apple.com/hardware/ve/examples.html
> >
> > http://developer.apple.com/hardware/ve/
> >
> > ( I wish I had done a more completesearch earlier )
> >
> > Jeff - have you taken a look this?  I'm very interested about getting
> > this into VLC for OSX and ppc in general.  I don't know much about
> > vector processing but the concept seems fairly straightforward.
>
> If I remember well, I think this is the code I borrowed for the
> currently-broken vdec_idct_altivec.c. There was apparently a deadlock
> in the code (?). One of my mistakes was that I forgot (or misplaced) the
> matrix transpose at the beginning, and I haven't had time (nor a machine)
> to debug it.

The code referenced above is different from the existing AltiVec IDCT in
the vlc source tree.  It is about 1/3rd as much code, and it definitely
works without deadlocks.  I have verified it against the C IDCT.

> Well, you can still ask me questions, but I don't have an OS X machine
> to test it - and BTW I'm going on vacation for 2 weeks on thursday, so
> be quick :-).

I'm not using OS X either.  Debian PPC on a PowerBook.

-jwb