From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 30994 invoked by alias); 1 Mar 2012 09:15:16 -0000 Received: (qmail 30976 invoked by uid 22791); 1 Mar 2012 09:15:15 -0000 X-SWARE-Spam-Status: No, hits=-2.8 required=5.0 tests=ALL_TRUSTED,AWL,BAYES_00 X-Spam-Check-By: sourceware.org Received: from localhost (HELO gcc.gnu.org) (127.0.0.1) by sourceware.org (qpsmtpd/0.43rc1) with ESMTP; Thu, 01 Mar 2012 09:15:03 +0000 From: "jakub at gcc dot gnu.org" To: gcc-bugs@gcc.gnu.org Subject: [Bug tree-optimization/52445] [4.6/4.7 Regression] conditional store replacement causes segfault in generated code Date: Thu, 01 Mar 2012 09:15:00 -0000 X-Bugzilla-Reason: CC X-Bugzilla-Type: changed X-Bugzilla-Watch-Reason: None X-Bugzilla-Product: gcc X-Bugzilla-Component: tree-optimization X-Bugzilla-Keywords: wrong-code X-Bugzilla-Severity: normal X-Bugzilla-Who: jakub at gcc dot gnu.org X-Bugzilla-Status: NEW X-Bugzilla-Priority: P3 X-Bugzilla-Assigned-To: unassigned at gcc dot gnu.org X-Bugzilla-Target-Milestone: 4.6.3 X-Bugzilla-Changed-Fields: Message-ID: In-Reply-To: References: X-Bugzilla-URL: http://gcc.gnu.org/bugzilla/ Auto-Submitted: auto-generated Content-Type: text/plain; charset="UTF-8" MIME-Version: 1.0 Mailing-List: contact gcc-bugs-help@gcc.gnu.org; run by ezmlm Precedence: bulk List-Id: List-Archive: List-Post: List-Help: Sender: gcc-bugs-owner@gcc.gnu.org X-SW-Source: 2012-03/txt/msg00026.txt.bz2 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52445 --- Comment #5 from Jakub Jelinek 2012-03-01 09:14:42 UTC --- The quick fix that would IMHO brings us back to pre-161655 decisions would be just to store also the offset and size into the hash table entries and use them as a requirement for the lookups. I think that would essentially map to what we did before. A better, but more complicated, change, would be to only keep using ssa_name and store as lookup criteria as we do right now, add a linked list of offset/sizes and consider non-trapping stores if the [offset, offset+size) interval is subset of the non-trapping bytes. This would be able to optimize even the cases where say there are is a larger store (or several smaller stores) that cover the area. We might need to prune the chains in nt_fini_block though. Richard, do you think for 4.7.0/4.6.4 just implementing the simpler approach would be fine?