Re: mplayer / divx
On Wed, Aug 29, 2001 at 08:47:53AM +0200, Wichert Akkerman wrote:
> Previously Aaron Lehmann wrote:
> > I'd *like* to optimize the package at least for a Pentium, because I
> > imagine that MMX may be useful and I really doubt that the package
> > would have any utility whatsoever on anything < i586.
> There is i386, i386 with mmx, i386 with 3dnow and a few other
> variations. Lots of options to choose from.
So with the ones you listed, we're talking about i386 vs. Pentium
MMX and up vs. K6/Athlon. As I said, my impulse is to go with the Pentium
optimization since it's a very safe bet. Is there a problem with doing
this in a distribution that's supposed to be for i386? I share the view
with many that we should try to preserve a least common denominator,
but trying to watch DivX videos on a 386 is just crazy.
The *right* thing to do about the whole IA32 optimization issue is
to compile *all* the cores in and let the program chose the
appropriate one based on the results of the CPUID instruction. This is
what the package zsnes (that I maintain) does. But this has to come
from upstream. I really ought to request this feature from them and
hope they implement it before I try to package mplayer. As for the
opendivx stuff, I tried very naively removing all references into the
OpenDivX code (there were very few). This resulted in an X crash,
indicating that the DivX code may be used more than I had thought.
Whatever the problem is, it's going to take some human inspection. It
could just be that it had been trying to, say, register the OpenDivX
codec and my removal of a line of code right in the middle of that
messed everything up. I'll have to look at it, or alternatively, it
would be great if upstream cooperated with me and made it a
configure-time option (thus making it easier to remove from the
tarball). I'll have to ask them.