This is the mail archive of the
mailing list for the GCC project.
Re: [PATCH] Fix PR61221
- From: Richard Biener <rguenther at suse dot de>
- To: Jeff Law <law at redhat dot com>,gcc-patches at gcc dot gnu dot org
- Date: Mon, 19 May 2014 19:00:00 +0200
- Subject: Re: [PATCH] Fix PR61221
- Authentication-results: sourceware.org; auth=none
- References: <alpine dot LSU dot 2 dot 11 dot 1405191451130 dot 13304 at zhemvz dot fhfr dot qr> <537A3810 dot 7010708 at redhat dot com>
On May 19, 2014 6:57:52 PM CEST, Jeff Law <firstname.lastname@example.org> wrote:
>On 05/19/14 06:54, Richard Biener wrote:
>> In this PR we run into the issue that releasing SSA names from
>> FRE/PRE elimination corrupts the VN lattice and thus the VN lookup
>> we perform for removing redudnant stores ICEs. The patch works
>> around the particular case by making unreachable code detection
>> in SCCVN more optimistic by ignoring backedges during reachability
>> computation and by not doing any elimination on unreachable blocks.
>> I still have to think about a "proper" fix to the underlying issue
>> (propagate_value_into_stmt updating stmts on our back).
>> Bootstrap and regtest running on x86_64-unknown-linux-gnu.
>> 2014-05-19 Richard Biener <email@example.com>
>> PR tree-optimization/61221
>> * tree-ssa-pre.c (eliminate_dom_walker::before_dom_children):
>> Do nothing for unreachable blocks.
>> * tree-ssa-sccvn.c (cond_dom_walker::before_dom_children):
>> Improve unreachability detection.
>> * gcc.dg/torture/pr61221.c: New testcase.
>Is this another case where having two lists rather than just one in the
>SSA_NAME manager would help? It's something I really need to sit down
>and just do.
No, it's unfortunately not that easy.