This is the mail archive of the
mailing list for the GCC project.
Re: [PATCH] Fix PR42509, wrong-code from the RTL alias oracle
- From: Richard Guenther <rguenther at suse dot de>
- To: Ulrich Weigand <uweigand at de dot ibm dot com>
- Cc: gcc-patches at gcc dot gnu dot org, rearnsha at gcc dot gnu dot org, doko at ubuntu dot com
- Date: Tue, 27 Jul 2010 22:12:16 +0200 (CEST)
- Subject: Re: [PATCH] Fix PR42509, wrong-code from the RTL alias oracle
- References: <201007271902.o6RJ2VGQ013283@d12av02.megacenter.de.ibm.com>
On Tue, 27 Jul 2010, Ulrich Weigand wrote:
> Richard Guenther wrote:
> > PR middle-end/42509
> > * alias.c (nonoverlapping_memrefs_p): For spill-slot accesses
> > require a non-NULL MEM_OFFSET.
> What are your thoughts on backporting this to the 4.4 branch?
> It seems that this problem never existed in 4.3 and earlier because
> it was introduced by IRA and the spill_slot_decl logic; and the bug
> is fixed in 4.5 and later. However, it exists as latent bug in 4.4.
> Now, I was unable to construct a problem to deliberately trigger
> the issue (you need just the right interaction between reload,
> alias, and cross-jumping ...). However, it seems to have occured
> in an out-of-tree port:
> Since the fix looks to me as if it couldn't really have any adverse
> effects, and it fixes a potentially hard to track down latent wrong
> code generation bug, it might be good to backport ...
If you want to do it and do the testing a backport is fine with me.