From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 12374 invoked by alias); 15 Jan 2004 18:40:44 -0000 Mailing-List: contact gcc-bugs-help@gcc.gnu.org; run by ezmlm Precedence: bulk List-Archive: List-Post: List-Help: Sender: gcc-bugs-owner@gcc.gnu.org Received: (qmail 12353 invoked by alias); 15 Jan 2004 18:40:43 -0000 Date: Thu, 15 Jan 2004 18:40:00 -0000 Message-ID: <20040115184043.12352.qmail@sources.redhat.com> From: "dje at watson dot ibm dot com" To: gcc-bugs@gcc.gnu.org In-Reply-To: <20040114231753.13689.dje@gcc.gnu.org> References: <20040114231753.13689.dje@gcc.gnu.org> Reply-To: gcc-bugzilla@gcc.gnu.org Subject: [Bug pch/13689] GGC PCH breaks aliasing X-Bugzilla-Reason: CC X-SW-Source: 2004-01/txt/msg01777.txt.bz2 List-Id: ------- Additional Comments From dje at watson dot ibm dot com 2004-01-15 18:40 ------- Subject: Re: GGC PCH breaks aliasing The example in the original report ICEs (with Honza's patch) on AIX due to an internal varray consistency check. With the splay tree implementation, we presumably gets zeroes for the data that was not stored in PCH. If the zeroes represent to alias sets that alias with everything, then the code will be correct but potentially missing optimizations. If we re-use those alias sets numbers, then we may perform incorrect optimizations assuming that types or expressions do not alias. David -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13689