This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: [tree-ssa] Mainline merge plan
- From: law at redhat dot com
- To: Jim Wilson <wilson at specifixinc dot com>
- Cc: ahs3 at fc dot hp dot com, Andrew Pinski <pinskia at physics dot uc dot edu>, "gcc at gcc dot gnu dot org" <gcc at gcc dot gnu dot org>
- Date: Tue, 02 Mar 2004 00:21:01 -0700
- Subject: Re: [tree-ssa] Mainline merge plan
- Reply-to: law at redhat dot com
In message <404422B7.6080404@specifixinc.com>, Jim Wilson writes:
>Al Stone wrote:
>> Probably true. Mainline has not bootstrapped on
>> ia64 for a couple of weeks; the day job has been
>
>I am not aware of any lingering bootstrap problem, and messages to the
>gcc-testresults mailing list seem to confirm that. There is a
>successful report on Feb 27.
>
>I did an update and bootstrap today, and it did fail. I tracked it down
>to a problem in global.c. The list of registers to allocate had two of
>them swapped. This is presumably a problem with allocno_compare, which
>does some FP arithmetic. I see Steve Ellcey checked in a patch which
>enabled inline FP divides by default. If I disable that (by changing
>TARGET_DEFAULT), then the bootstrap succeeds.
>
>So the inline FP divide code is somehow causing allocno_compare in
>global.c to be miscompiled. I haven't had a chance to look at why yet.
FWIW, one of the allocators has an unstable sort when presented with a
small number of qtys to allocate. That may (or may not) be related.
jeff