Re: [3.1.1] x86-64 fixes

[Jan Hubicka <jh at suse dot cz>]

> 
> 
> --On Wednesday, May 22, 2002 08:04:16 PM +0200 Andreas Jaeger <aj@suse.de> 
> wrote:
> 
> >Mark Mitchell <mark@codesourcery.com> writes:
> >
> >>[...]
> >>Again, is this a regression?
> >
> >The changes are fixes for x86-64 that we haven't noticed earlier.
> >Since x86-64 is a new architecture, we cannot say it's a regression.
> 
> That is what I thought.
> 
> In that case, I do not want to apply changes that could affect plain
> x86.  Please conditionalize the patches with TARGET_64BIT, or some
> other appropriate macro, so that we can be sure they don't affect
> x86.

They were tested by building many packages using it I believe.  All
changes have nature of not modifying the behaviour of 32bit compiler.
> 
> I know that you've tested these with ordinary x86s, but there's no
> reason to take the chance.
Hmm, that will be tricky at least for unix.h that will need to use
inline thunk generation in some case and offline in the other.

Number of patches are already condtionalized by TARGET_64BIT.  Is that
enought?  Can you take a look at these?
This includes the thunk code of unix.h partially as well - the function
is verbatim copy of the macro just with TARGET_64BIT block changed.

Honza
> 
> If you conditionalize them in this way, and also deal with the other
> nits I raised, the patches will be fine for the branch.
> 
> Thanks!
> 
> --
> Mark Mitchell                   mark@codesourcery.com
> CodeSourcery, LLC               http://www.codesourcery.com