Re: fmal and nexttowardl on alpha

[Jakub Jelinek <jakub at redhat dot com>]

On Thu, Apr 05, 2001 at 10:24:10AM +0200, Andreas Jaeger wrote:
> > On Thu, Apr 05, 2001 at 09:05:43AM +0200, Andreas Jaeger wrote:
> > > Stephen L Moshier <moshier@mediaone.net> writes:
> > > 
> > > > >It's not only Alpha - it's all platforms that don't have long double,
> > > > >e.g. also Sparc.
> > > > 
> > > > Sparc has 128-bit long double, supported by gcc.
> > > 
> > > In that case we should make that known for sparc32 (sparc64 does it
> > > right already).
> > > 
> > > Uli, Jakub, what do you think?  Shall I commit this?
> 
> > No. long double is the same type as double on sparc32, until we change this.
> 
> Thanks for the explanation.
> 
> > But the change needs to be well thought out in order to preserve binary
> > compatibility, similarly to PowerPC and Alpha.
> > gcc-2.96-RH and 3.0/3.1 support -mlong-double-64 and -mlong-double-128
> > command line switches (and -mlong-double-64 is the default) on sparc32, so
> > the steps are basically:
> > 1) glibc has to enforce gcc-2.96 and above on sparc32 in configure
> > 2) all functions which touch long double (including stuff like printf etc.)
> >    need to be versioned
> > 3) -mlong-double-128 has to be used, headers updated
> > 
> > and once this is done, gcc configure can check whether glibc supports
> > 128-bit long double and if yes, default to -mlong-double-128.
> > 
> > I assume all of Sparc32, Alpha and PowerPC could be changed at once...
> > 
> > So it is not that trivial.
> 
> I guess this is a task that can only be done when gcc 3.0 is out -
> let's wait for some time.

Well, IMHO it should be done before gcc 3.0 is released, so that e.g.
distributions shipping gcc 3.0 can already ship it defaulting to
-mlong-double-128 (and equivalents on alpha/powerpc).
I'll try to cook up preliminary patch soon.

	Jakub