X-Authentication-Warning: delorie.com: mailnull set sender to pgcc-bounces using -f Date: Wed, 16 Jan 2002 18:01:55 +0100 From: Jan Hubicka To: pgcc AT delorie DOT com Subject: Re: compiler-news? Message-ID: <20020116170155.GG14914@atrey.karlin.mff.cuni.cz> References: Mime-Version: 1.0 Content-Type: text/plain; charset=iso-8859-2 Content-Disposition: inline In-Reply-To: User-Agent: Mutt/1.3.24i X-MIME-Autoconverted: from 8bit to quoted-printable by atrey.karlin.mff.cuni.cz id SAA20857 Content-Transfer-Encoding: 8bit X-MIME-Autoconverted: from quoted-printable to 8bit by delorie.com id g0GH1xL31546 Reply-To: pgcc AT delorie DOT com Errors-To: nobody AT delorie DOT com X-Mailing-List: pgcc AT delorie DOT com X-Unsubscribes-To: listserv AT delorie DOT com Precedence: bulk > On Tue, 15 Jan 2002, Henrik Berglund SdU wrote: > > > On Tue, 15 Jan 2002, Marc wrote: > > > > > I should at least make the patch for 2.95.4, yes, but gcc-3.x produces slower > > > code (backend), but has a different internal structure so I expect much > > > better optimizations "soon". > > Yes i have installed gcc 3.0.3 on redhat and get nice nbench results > > on my duron 1.2GHz > > > > How 3.0.3 is compared to 2.95? The very first versions of gcc-3 were > slower in athlon than 2.95. Only in the FP code. 3.0.3 fixes number of FP issues, but the compariosns are still quite balanced. You may take a look at www.suse.de/~aj/SPEC for SPEC results - the 3.0.x is at average better than 2.95 in both SPECint and SPECfp. There are however important problems with scheduling on some numeric code, mainly Atlas. They are partly tracked down in 3.0.2 and 3.0.3. 3.0.0 additionally had problem with stack missalignment caused due to glibc2 misscompiled by buggy 2.95 stack alignment code distributed by most of major distros. Honza > > > ___ _ .''`. > | |_ _. _| _ |_) _ ._ ._ _. : :' : > | | |(_|(_|(/_|_| | (/_| || |(_| `. `'` > Linux User #50500 `- > Prof.Adjunto - Instituto de Física ----Debian- > Universidade Federal Fluminense Alpha/K6/K7