This is the mail archive of the
mailing list for the GCC project.
Re: [tree-ssa] Miscompilation (SIGSEGV) with -ffast-math (!?)
- From: law at redhat dot com
- To: Steven Bosscher <s dot bosscher at student dot tudelft dot nl>
- Cc: Richard Guenther <rguenth at tat dot physik dot uni-tuebingen dot de>, "gcc at gcc dot gnu dot org" <gcc at gcc dot gnu dot org>
- Date: Tue, 24 Feb 2004 09:51:02 -0700
- Subject: Re: [tree-ssa] Miscompilation (SIGSEGV) with -ffast-math (!?)
- Reply-to: law at redhat dot com
In message <email@example.com>, Steven Bosscher
>On Tuesday 24 February 2004 05:00, firstname.lastname@example.org wrote:
>> In message <403A0162.email@example.com>, Richard Guenther
>> >My favorite testcase is miscompiled by g++-ssa (GCC) 3.5-tree-ssa
>> >20040220 (merged 20040211) causing it to segfault when using -O2
>> >-ffast-math as opposed to -O2 which is fine (cmdline params "-n 10").
>> Note I'm not getting the segfault. However, I was able to fix the
>> insane compile-time regression -- we were spending an absurd amount of
>> time walking page table lists in ggc_free.
>In that case you may want to revive a patch I had some time ago to
>have per-order page lists. See PR10944. Andrew Pinski reported that
>bug but I was never able to reproduce it and Andrew never tested the
This is different from the issues in PR10944.
We're not walking freelists -- we're walking the in-use list (which is a
per-allocation-order list). This was necessary when the page containing
the object to be freed has no other free objects as we need to move that
page to the head of the in-use list.