Date:	Fri, 29 Sep 2000 11:02:26 -0300 (BRT)
From:	Thadeu Penna <tjpp AT if DOT uff DOT br>
To:	pgcc AT delorie DOT com
Subject:	Re: pgcc didn't perform for me
In-Reply-To:	<00092908391500.02822@struppi>
Message-ID:	<Pine.LNX.4.21.0009291056410.1232-100000@complex.if.uff.br>
MIME-Version:	1.0
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

On Fri, 29 Sep 2000, Michael Hanke wrote:

> Thadeu Penna wrote on Don, 28 Sep 2000:
> > We have used pgcc-2.95.3 on Athlons with great success (35% or more
> > compared to gcc). We work on simulations on Statistical Physics  and other
> > number-crunching Condensed Matter stuff (lot of matrix operations, etc.)
> > Actually pgcc-2.95.3 is as good as Agcc (Athlon specific patch) that
> > appeared recently on Freshmeat.
> Hi!
> 
> This sounds rather interesting for me since I am working with number crunching
> applications on an Athlon, too. I made some tests with egcs-1.1.2 and
> pgcc-2.95.3. It turned out that (besides other switches) -march=486 on egcs
> gave best performance. So my questions are: Which switches do you use for this
> increase in performance for (double precision) floating point arithmetic? Where
> can I find Agcc? I have some short (1000 lines) but timecritical routines.
> 

AGCC can be downloaded from athlonlinux.org. I use the same switches as
described there (-O6 -mpentiumpro -march=pentiumpro are responsible for
more than 90% of the performance increase, all the other switches are
almost cosmetic changes).
 
-- 
Thadeu Penna          Linux user #50500      -O) 
Instituto de Fisica -  UFF                   /\\ 
Niteroi -  RJ -                  Brazil     _\/V  

- Raw text -