This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: biarch gcc
On Wed, Aug 13, 2003 at 12:48:37AM +0200, Michael Matz wrote:
> On Tue, 12 Aug 2003, Peter Bergner wrote:
> > > The Glibc headers for sparc, s390 and x86_64 were written with that in
> > > mind and do not have any problems that I'm aware of. What problems
> > > exactly do you have?
> >
> > I believe Alan was talking about kernel headers here and not GLIBC headers.
Yes, I was mostly referring to kernel headers. glibc headers for
powerpc-linux and powerpc64-linux did show some differences too, but it
looks like you'd be OK if you installed ppc64 glibc after ppc32 glibc.
> Well, having and maintaining such stub headers is a small cost compared to
> the advantage of having a real biarch compiler, which should be done, like
> it's done already for many other 32/64 platforms.
Well, it might be a small cost for distributions, but I'd say building a
cross toolchain is now quite a bit more complicated. Besides the header
issue, you also need to make little shell scripts to invoke the tools,
so that xyz package's configure sees that you have both a
powerpc-linux-gcc and a powerpc64-linux-gcc. Not hard to do, but you've
got to be aware that such a step is necessary.
> In fact I'm not
> entirely sure, about what Alan was complaining, but at least the
> difference between 32 and 64 bit in header files is no reason to not build
> a biarch compiler. About the suggestion of only creating a biarch
> compiler, when --enable-biarch is given I'm also not entirely sure. What
> is the difference to not fulfilling the prerequisites? Better
> error-messages from configure? That would be good, but I felt Alan was
> also suggesting this for other reasons I'm not getting?
I wasn't so much complaining as giving some background for why I think
there ought to be a way of building a non-biarch compiler. I also
wrote the email to give some pointers to people trying to build cross
toolchains.
--
Alan Modra
IBM OzLabs - Linux Technology Centre