On Sat, 2005-08-27 at 16:20 +0200, Zdenek Dvorak wrote:
> > >we lose the VUSE, which could in worst case lead to misscompilations.
> > >In this particular case it only confuses ivopts that expect
> > >name_mem_tag
> > >or type_mem_tag to be set for every dereferenced pointer, and leads to
> > >ICE.
> > Then how does copy prop works? This is exactly what copy prop does
> > and I don't see we losing VUSE there.
> copy propagation merges the alias information (see
> tree-ssa-copy.c:merge_alias_info). Which of course is an
> alternative solution for the problem; I may prepare a patch to
> use it if preferable.
I'm not sure i understand:
Why is it preferrable to not do the propagation at all here when the
other places in the compiler simply merge the alias info?
Why not just make this case consistent and merge the alias info?
> > This looks more like a bug in
> > IV-OPTs not recording the aliasing info right when it transforms the
> > loop.
> The loss of VUSEs occurs before ivopts, so they have nothing to do with