From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 14701 invoked by alias); 19 Aug 2010 15:53:53 -0000 Received: (qmail 14662 invoked by uid 22791); 19 Aug 2010 15:53:52 -0000 X-SWARE-Spam-Status: No, hits=-1.8 required=5.0 tests=AWL,BAYES_00,DKIM_SIGNED,DKIM_VALID,DKIM_VALID_AU,FREEMAIL_FROM,RCVD_IN_DNSWL_NONE X-Spam-Check-By: sourceware.org Received: from mail-px0-f175.google.com (HELO mail-px0-f175.google.com) (209.85.212.175) by sourceware.org (qpsmtpd/0.43rc1) with ESMTP; Thu, 19 Aug 2010 15:53:42 +0000 Received: by pxi11 with SMTP id 11so840486pxi.20 for ; Thu, 19 Aug 2010 08:53:41 -0700 (PDT) MIME-Version: 1.0 Received: by 10.114.77.11 with SMTP id z11mr26812waa.112.1282233219330; Thu, 19 Aug 2010 08:53:39 -0700 (PDT) Received: by 10.220.164.142 with HTTP; Thu, 19 Aug 2010 08:53:39 -0700 (PDT) In-Reply-To: <4C6D5164.4060803@ispras.ru> References: <4C6B9A2F.4090805@ispras.ru> <4C6D39D7.3090709@ispras.ru> <4C6D3C2F.4010602@ispras.ru> <4C6D5164.4060803@ispras.ru> Date: Thu, 19 Aug 2010 16:03:00 -0000 Message-ID: Subject: Re: [PATCH] Fix PR 44691 From: "H.J. Lu" To: Andrey Belevantsev Cc: GCC Patches , "Vladimir N. Makarov" Content-Type: text/plain; charset=ISO-8859-1 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: 2010-08/txt/msg01461.txt.bz2 2010/8/19 Andrey Belevantsev : > On 19.08.2010 18:14, Andrey Belevantsev wrote: >> >> On 19.08.2010 18:12, H.J. Lu wrote: >>> >>> 2010/8/19 Andrey Belevantsev: >>>> >>>> On 19.08.2010 17:28, H.J. Lu wrote: >>>>> >>>>> 2010/8/18 Andrey Belevantsev: >>>>>> >>>>>> Hello, >>>>>> >>>>>> As explained in the audit trail, the problem was that in the selecti= ve >>>>>> scheduler I assumed that SUBREG_REG will always be a REG, which seems >>>>>> to >>>>>> be >>>>>> not the case. This is not quite in line with what documentation says, >>>>>> if >>>>>> I >>>>>> read it correctly, but it seems to be used in a number of backends, = so >>>>>> the >>>>>> below patch just gives up substitution also when SUBREG_REG is not a >>>>>> register. Bootstrapped and tested on ia64, and verified that the test >>>>>> is >>>>>> fixed on x86_64. >>>>>> >>>>>> I think that this qualifies as obvious, so unless Vlad or other peop= le >>>>>> have >>>>>> any comments, I'll commit it tomorrow. >>>>>> >>>>>> Yours, Andrey >>>>>> >>>>>> 2010-08-18 Andrey Belevantsev >>>>>> >>>>>> PR rtl-optimization/44691 >>>>>> >>>>>> * sel-sched.c (count_occurrences_1): Also punt when SUBREG_REG >>>>>> is not a register. >>>>>> >>>>> >>>>> Shouldn't we add the testcase? >>>> >>>> The test is fortran.dg/pr42294.f which is actually mentioned in the bug >>>> report. Sorry for not saying this explicitly in the mail. >>>> >>> >>> Normally this bug isn't trigged. You need to pass -O2 >>> -fselective-scheduling2 >>> to see it. You should copy gfortran.dg/pr42294.f and add -O2 >>> -fselective-scheduling2. >> >> Ah, ok, I forgot about the explicit options. I will do that. > > Looking closely, pr42294.f happens to be another sel-sched bug, so it > already has "-O2 -fselective-scheduling2 -fsel-sched-pipelining > -funroll-all-loops" as dg-options. =A0So I guess this test should be enou= gh, > what do you think? Why didn't it fail before? Does this bug fail only with "-O2 -fselective-scheduling2"? --=20 H.J.