From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 1772 invoked by alias); 22 Oct 2008 10:17:24 -0000 Received: (qmail 1760 invoked by uid 22791); 22 Oct 2008 10:17:23 -0000 X-Spam-Check-By: sourceware.org Received: from fk-out-0910.google.com (HELO fk-out-0910.google.com) (209.85.128.189) by sourceware.org (qpsmtpd/0.31) with ESMTP; Wed, 22 Oct 2008 10:16:38 +0000 Received: by fk-out-0910.google.com with SMTP id 19so2962506fkr.8 for ; Wed, 22 Oct 2008 03:16:35 -0700 (PDT) Received: by 10.187.172.9 with SMTP id z9mr1444057fao.33.1224670595824; Wed, 22 Oct 2008 03:16:35 -0700 (PDT) Received: by 10.187.174.3 with HTTP; Wed, 22 Oct 2008 03:16:35 -0700 (PDT) Message-ID: <84fc9c000810220316t20257102t9ce3b53e95d3400b@mail.gmail.com> Date: Wed, 22 Oct 2008 10:42:00 -0000 From: "Richard Guenther" To: "Ralph Loader" Subject: Re: [PATCH] Fix for PR 37809 and 37807 Cc: gcc-patches@gcc.gnu.org In-Reply-To: <20081022181234.2acf7548@i.geek.nz> MIME-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline References: <20081022181234.2acf7548@i.geek.nz> 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: 2008-10/txt/msg00922.txt.bz2 On Wed, Oct 22, 2008 at 7:12 AM, Ralph Loader wrote: > Hi, > > This patch has a fix for prs 37807 (exponential time with MMX builtins) > and 37809 (wrong code with MMX builtins). > > Could an expert check that I'm not inadvertantly disabling important > optimisations for vector types? I believe the code in question was > only ever intended to be used for scalar types, but it is possible that > I've missed something. The rtlanal.c parts are ok. I wonder if we ever hit the _or_truncate case with vector modes in force_to_mode though - can you check if using gen_lowpart instead of gen_lowpart_or_truncate works? Btw, you missa ChangeLog entry for the testcase for testsuite/ChangeLog. Thanks, Richard. > 2008-10-19 Ralph Loader > > PR middle-end/37807, middle-end/37809 > * combine.c (force_to_mode): Do not process vector types. > > * rtlanal.c (nonzero_bits1): Do not process vector types. > (num_sign_bit_copies1): Likewise. > > Cheers, > Ralph. >