Date:	Mon, 8 Nov 1999 17:28:44 +0100 (CET)
From:	Ronald Wahl <rwahl AT gmx DOT net>
Sender:	Ronald DOT Wahl AT Informatik DOT TU-Chemnitz DOT DE
To:	pgcc AT delorie DOT com
Subject:	Re: pgcc-2.95.2: strange nbench results
In-Reply-To:	<19991103022858.I9319@cerebro.laendle>
Message-ID:	<Pine.LNX.4.10.9911081718130.17889-100000@goliath.csn.tu-chemnitz.de>
MIME-Version:	1.0
Reply-To:	pgcc AT delorie DOT com
X-Mailing-List:	pgcc AT delorie DOT com
X-Unsubscribes-To:	listserv AT delorie DOT com

On Wed, 3 Nov 1999, Marc Lehmann wrote:

> On Tue, Nov 02, 1999 at 02:49:05AM +0100, Ronald Wahl <rwahl AT gmx DOT net> wrote:
> 
> Iiiih, 'n gmx-er.

Was ist daran so schlimm? Ich benutz GMX ja auch nur als Forwarder...
 
> > moin,
> 
> Iiih, 'n Norddeutscher ;->

Nee, Sachse. ;-)

> > just run nbench-byte-2.1 compiled with pgcc-2.95.2 (-mk6 -O6 -ffast-math
> > -funroll-all-loops). Look especially at the neural net bench:
> 
> I think this is a large improvement.. IIRC, nbench always segfaulted on the
> neural net test, now it only gives a wrong result ;->

Not always... Last version was ok (at least with -mk6).

> Seriously, this is probably a bug in pgcc, although (having inspected the
> test some times) it's not clear to me wether the compiler isn't allowed to
> just optimize it away...

Hmm, I hope this doesn't affect other applications...

ron

-- 
/\/\  Ronald Wahl -- rwahl AT gmx DOT net     /\/\          C S N           /\/\
\/\/  -------------------------------  \/\/  ----------------------  \/\/
/\/\  http://www.tu-chemnitz.de/~row/  /\/\  network administration  /\/\
\/\/  PGP key available                \/\/    system  management    \/\/

- Raw text -