From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 8832 invoked by alias); 7 May 2010 20:31:42 -0000 Received: (qmail 8741 invoked by uid 22791); 7 May 2010 20:31:40 -0000 X-SWARE-Spam-Status: No, hits=-1.4 required=5.0 tests=AWL,BAYES_00,DKIM_SIGNED,DKIM_VALID,DKIM_VALID_AU,FREEMAIL_FROM,RCVD_IN_DNSWL_NONE,SARE_MSGID_LONG45 X-Spam-Check-By: sourceware.org Received: from mail-vw0-f47.google.com (HELO mail-vw0-f47.google.com) (209.85.212.47) by sourceware.org (qpsmtpd/0.43rc1) with ESMTP; Fri, 07 May 2010 20:31:35 +0000 Received: by vws13 with SMTP id 13so1242803vws.20 for ; Fri, 07 May 2010 13:31:33 -0700 (PDT) MIME-Version: 1.0 Received: by 10.220.126.222 with SMTP id d30mr409227vcs.78.1273264293359; Fri, 07 May 2010 13:31:33 -0700 (PDT) Received: by 10.220.100.77 with HTTP; Fri, 7 May 2010 13:31:33 -0700 (PDT) In-Reply-To: <6dc9ffc80910171021t73b5633am64086a14d5ccf6b9@mail.gmail.com> References: <6dc9ffc80910171021t73b5633am64086a14d5ccf6b9@mail.gmail.com> Date: Fri, 07 May 2010 20:31:00 -0000 Message-ID: Subject: Re: [PR debug/41535] reset debug insns affected by scheduling From: "H.J. Lu" To: Alexandre Oliva Cc: gcc-patches@gcc.gnu.org, jakub@redhat.com Content-Type: text/plain; charset=windows-1252 Content-Transfer-Encoding: quoted-printable X-IsSubscribed: yes Mailing-List: contact gcc-patches-help@gcc.gnu.org; run by ezmlm Precedence: bulk List-Id: List-Archive: List-Post: List-Help: Sender: gcc-patches-owner@gcc.gnu.org X-SW-Source: 2010-05/txt/msg00545.txt.bz2 On Sat, Oct 17, 2009 at 10:21 AM, H.J. Lu wrote: > On Fri, Oct 16, 2009 at 12:25 AM, Alexandre Oliva wro= te: >> On Oct 15, 2009, Alexandre Oliva wrote: >> >>> This patch enables nondebug insns to =93depend=94 on debug insns, but t= hese >>> dependencies are handled in a special way: they don't stop the nondebug >>> insn from being scheduled. =A0Rather, they enable the nondebug insn, on= ce >>> scheduled, to quickly reset debug insns that remain as unresolved >>> dependencies. >> >>> To avoid using up more memory and complicating scheduler logic, I didn't >>> add more dependency lists. >> >> Turns out when I was about to finish composing the e-mail above, this >> idea occurred to me: >> >>> perhaps even discounting them in the list length counter. >> >> and it was pretty easy to implement and, surprise, it worked beautifully >> and far more easily and more efficient than the earlier patch! >> >> Also, Jakub kindly tested the patch on other arches and found >> -fmodulo-sched regressions: assertions failed that I'd put in there >> precisely to prevent =93debug deps=94. =A0I'm yet to determine whether t= heir >> presence might cause -fcompare-debug differences; it didn't for the >> testcases I tested, but they hardly exercised broad coverage. =A0I'll >> start testing some optimized bootstrap-debugs and see if I run into any >> such problems. =A0Fixing them will be for another patch, though. >> >> Here's the simplified patch I've regstrapped, so far, on >> x86_64-linux-gnu. =A0Ok to install? >> > > This may have caused: > > http://gcc.gnu.org/bugzilla/show_bug.cgi?id=3D41739 > This also caused: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=3D44028 --=20 H.J.