This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Fix for semi-latent bug in vectorizer
- From: Diego Novillo <dnovillo at redhat dot com>
- To: Keith Besaw <kbesaw at us dot ibm dot com>
- Cc: law at redhat dot com, gcc-patches at gcc dot gnu dot org
- Date: Tue, 19 Apr 2005 13:22:40 -0400
- Subject: Re: Fix for semi-latent bug in vectorizer
- Organization: Red Hat Canada
- References: <1113581885.5447.144.camel@localhost.localdomain> <OF8C477F7F.3FB1C301-ON86256FE8.00292090-86256FE8.002BA0EA@us.ibm.com>
On Tue, Apr 19, 2005 at 02:56:32AM -0500, Keith Besaw wrote:
> is incorrect. If someone found an inconsistency then I'd like to see it
> because the other calls to copy_virtual_operands in the vectorizer should
> also be changed.
>
No. This was the only place where it was needed, because it's
where the vectorized statement is emitted. The other copies made
in the vectorizer are not getting new alias information.
> I see that Zdenek found a problem with the vectorizer when he changed
> verify_ssa, but I haven't seen the problem yet in my testing so I'm not
> sure what the problem is. In any case, I don't think that this is the
> right fix.
>
I am working on changes to the aliaser that should allow us to do
what the vectorizer and ivopts are trying to do. The core of the
problem is that the aliaser does not support using arbitrary
symbols as memory tags. I tried fixing this by convincing ivopts
and the vectorizer to use a type tag, but it's starting to look
like a really bad idea. Moreover, Zdenek's TARGET_MEM_REF patch
will also need these changes.
I am travelling now, so I can't really take a look at the
problem, sorry. I also have some queued changes in the
vectorizer to prevent it from doing so much work in the SSA
updating phase. I will take care of this problem early next week.
Diego.