This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Final intermodule patch
Op za 12-07-2003, om 03:19 schreef Geoff Keating:
> Richard Henderson <rth@redhat.com> writes:
>
> > Of course, the final show stopper, once you make it beyond the
> > crt files, is that the actual intermodule bits hork badly.
> >
> > In file included from <command line>:72,
> > from ../../../src-gcc/gcc/alias.c:2932:
> > ../../../src-gcc/gcc/rtl.def:796: warning: comma@end of enumerator list
> > ../../../src-gcc/gcc/rtl.def:796: error: parse error before "const"
> > ../../../src-gcc/gcc/rtl.def:947: warning: comma@end of enumerator list
> > ../../../src-gcc/gcc/rtl.def:947: error: parse error before ',' token
> > [thousands of more error lines removed]
> >
> > This happens on both i686 and alpha linux. Does it really
> > work on ppc-darwin?
>
> It worked for me once, *at the default -g -O2*. It's known to chew up
> all RAM on the system and die@-O3, because loop can't handle the
> large functions generated, and then if you do -fno-loop-optimize I
> believe it dies in a different way having to do deleting tablejump
> tables.
It still fails bootstrap on i686-pc-linux-gnu, cvs minutes ago.
> I find that using the intermodule stuff is an excellent way to find
> optimizer and codegen problems, and encourage everyone to try it!
<grin> I suppose it exposes worst cases, yes :-)
You apparently added support for bootstrapping with intermodule
optimizations (--enable_intermodule), but you didn't indicate what it
does and if it had any effect on performance. Do you have some numbers
to make us enthousiastic about this patch?
Gr.
Steven