Re: Altivec Linux support in GCC

[Stan Shebs <shebs at apple dot com>]

Kumar Gala wrote:

> On Tuesday, April 15, 2003, at 01:08 PM, Stan Shebs wrote:
>
> > Aldy Hernandez wrote:
> >
> > > > > > > > "Michel" == Michel LESPINASSE <walken at zoy dot org> writes:
> > >
> > >  > On Mon, Apr 14, 2003 at 12:49:48PM -0500, Kumar Gala wrote:
> > > >> 1. What are there two options to enable the support? (-maltivec 
> > > >> -mabi=altivec).  If I have a Linux PPC targeted GCC why do I need 
> > > to >> tell it to use the abi to get altivec code generation?  This 
> > > just seems >> confusing to the user (I dont have to specify the abi 
> > > for any other >> option).
> > >
> > > Still, only the minority of users do this.
> > >
> > > I agree with y'all that -maltivec without -mabi=altivec is useful,
> > > what I have argued for in the past (and again now), is having
> > > -maltivec imply -mabi=altivec as the default.  Then, if you want the
> > > functionality you describe, use -maltivec -mabi=no-altivec.
> > Sounds like a good idea.  I think when this all originally went in,
> > we were being cautious and incremental - -maltivec would always work,
> > but -mabi=altivec required cooperating libraries and such.  However,
> > I think we're past that initial phase now!
> >
> > Apple goes one further actually, forces AltiVec-friendly ABI always,
> > since system can trash you if you try to use AltiVec ops but not
> > the proper ABI.
>
>
> I guess this is my concern, especially with -mabi=no-altivec not being 
> very well specified to the user.  It is not clear at all as to what a 
> user gets with this option as far as AltiVec support goes.

Since the exact meaning of -mabi=altivec depends on the specific PowerPC
target, it's up to the Linuxites.  The basic ABI is described in the SVR4
doc, and the AltiVec ABI in the PIM, but Linux users would certainly
benefit from a distinct and succinct writeup specific to PPC Linux.

Stan