[MPlayer-dev-eng] Fixing the GCC-3.3 and scale code mess

[Rich Felker]

On Tue, Jun 28, 2005 at 04:44:45PM +0200, Michael Niedermayer wrote:
> Hi
> 
> On Tuesday 28 June 2005 14:13, Guillaume POIRIER wrote:
> > Hello,
> > We're getting more and more messages about the code I broke
> > unwillingly to support GCC-4.0. I'd much better be able to reply "use
> > current CVS" than explaining what has gone wrong and why it may take
> > some time until it's fixed.
> >
> > Can I revert my patch so that GCC-3.3 (used by a large crowd of users)
> > works and GCC-4.0 (used only by bleeding edge users using FC4 or
> > ubuntu breezy) [unfortunately] doesn't?
> 
> yes
> 
> 
> >
> > That way, this leaves us some time to think about a good fix to
> > support all GCC version, if possible.
> >
> > What do you think?
> 
> gcc sucks, and we probably need some automated way to test if the code works 
> with several gcc versions, something with -vo md5 maybe, or maybe 
> postproc/swscale-example.c 

I have a proposal to fix all this mess. It seems like the problems is
coming from having to "fix" constraints like "m"((long)width) which
gcc4 does not like. Can we PLEASE just change all the functions to
take long arguments instead of int? It will make no difference on x86,
but it will allow us to remove all the casts on x86-64 so that no ugly
hacks are needed to load the values into the asm.

Opinions? IMO this is by far the simplest fix...

Rich