From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 7827 invoked by alias); 23 Jan 2004 22:55:42 -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 7799 invoked by alias); 23 Jan 2004 22:55:42 -0000 Date: Fri, 23 Jan 2004 22:55:00 -0000 Message-ID: <20040123225542.7798.qmail@sources.redhat.com> From: "wilson at specifixinc dot com" To: gcc-bugs@gcc.gnu.org In-Reply-To: <20040117230657.13722.schwab@suse.de> References: <20040117230657.13722.schwab@suse.de> Reply-To: gcc-bugzilla@gcc.gnu.org Subject: [Bug target/13722] [3.4/3.5 regression] [ia64] ICE in push_secondary_reload X-Bugzilla-Reason: CC X-SW-Source: 2004-01/txt/msg03016.txt.bz2 List-Id: ------- Additional Comments From wilson at specifixinc dot com 2004-01-23 22:55 ------- Subject: Re: PR 13722 candidate fix On Thu, 2004-01-22 at 23:43, Zack Weinberg wrote: > Under what circumstances should REG_INC notes be tagged onto the insns > emitted here? It's currently done only for POST_MODIFYs, but this > patch will generate POST_INC and POST_DEC as well. I believe the answer is that we need it for all auto-inc addresses. The REG_INC note indicates that the instruction has another side-effect besides the sets and clobbers. Any routines that care about instruction side-effects aren't going to notice the auto-inc address unless the REG_INC note is there. The fact that we only do this after reload reduces the chances of a problem, but the rtlanal.c reg_set_p function needs it, and it is likely that there is already some code somewhere after reload that directly or indirectly calls reg_set_p. Thus the REG_INC notes should be added to be safe. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13722