Message-ID: <19990604152712.39337@atrey.karlin.mff.cuni.cz>
Date: Fri, 4 Jun 1999 15:27:12 +0200
From: Jan Hubicka <hubicka AT atrey DOT karlin DOT mff DOT cuni DOT cz>
To: pgcc AT delorie DOT com
Subject: Re: Pgcc 1.1.3 - bad performance on P6
References: <m10prMP-00021bC AT chkw386 DOT ch DOT pwr DOT wroc DOT pl>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
X-Mailer: Mutt 0.84
In-Reply-To: <m10prMP-00021bC@chkw386.ch.pwr.wroc.pl>; from Krzysztof Strasburger on Fri, Jun 04, 1999 at 12:54:13PM +0000
Reply-To: pgcc AT delorie DOT com
X-Mailing-List: pgcc AT delorie DOT com
X-Unsubscribes-To: listserv AT delorie DOT com
Precedence: bulk

Hi
I've downloaded the loop and did some tests with new egcs snapshots. Results are not so bad:
2.7.2 -O2 -m486
0m4.764s
egcs -O2 -mpentium
0m3.472s
egcs -O3 -mpentium gausil.c -lm  -ffast-math -funroll-all-loops -mno-ieee-fp
-fomit-frame-pointer
0m2.848s
after patching the fpmul parameters in i386.md:
0m2.762s
2.7.2 with same parameters:
0m4.661s
egcs 1.1.2 with same parameters:
0m3.364s

At least compared to 2.7.2 in -O3 -ffast-math etc... it is almost twice as
fast.... According to other letter here it is getting closer to VC,
that was more than twice as fast compared to 1.1.2.
At least there is IMO visible quite good progress.

The tests was done on my P5/160Mhz...
Note that -mno-ieee-fp is _VERY_ dangerous on anything older than two weeks
old egcs snapshot. (about 30% of compares are reversed in resulting code)

Honza