This is the mail archive of the
mailing list for the GCC project.
Re: 3.4 IMA uses incorrect alias sets
- From: Andrew Pinski <pinskia at physics dot uc dot edu>
- To: Ulrich Weigand <weigand at i1 dot informatik dot uni-erlangen dot de>
- Cc: gcc at gcc dot gnu dot org
- Date: Tue, 27 Jul 2004 15:19:04 -0700
- Subject: Re: 3.4 IMA uses incorrect alias sets
- References: <200407272212.AAA14442@faui1m.informatik.uni-erlangen.de>
On Jul 27, 2004, at 3:12 PM, Ulrich Weigand wrote:
we're seeing a miscompilation using GCC 3.4's inter-module
optimization. The bug is caused by scheduling violating a
dependency. This in turn is because a struct type defined
in both translation units gets assigned *different* alias
set numbers, so that (after inter-module inlining) the use
of a local variable gets a different alias set than the
initialization of the variable.
Initial debugging shows that there are simply two different
instances of the type, each of which then gets assigned its
own alias set. What is supposed to prevent this?
The 3.5 IMA doesn't show this bug in my testcase. However,
it appears to be rather completely rewritten ...
Any suggestions where to look further?
But I would say that this is not going to be fixed.
I filed the bug to remind me to test again and then I saw that
it was fixed when I helped fix IMI (and also when the tree-ssa
was merged in really) so I closed it as fixed.