Message-ID: <B0000054188@stargate.astr.lu.lv>
From: "Andris Pavenis" <pavenis AT lanet DOT lv>
To: "Salvador Eduardo Tropea (SET)" <salvador AT inti DOT gov DOT ar>, djgpp AT delorie DOT com
Date: Tue, 8 Dec 1998 10:30:47 +0200
MIME-Version: 1.0
Content-type: text/plain; charset=US-ASCII
Content-transfer-encoding: 7BIT
Subject: RE: [Announce] Benchmarks result tested with gcc, egcs and pgcc 
In-reply-to: <m0zn3T4-000S5eC@inti.gov.ar>
References: <199812042040 DOT UAA18311 AT remus DOT clara DOT net>
X-mailer: Pegasus Mail for Win32 (v3.01d)
Reply-To: djgpp AT delorie DOT com

On 7 Dec 98, at 13:33, Salvador Eduardo Tropea (SET) wrote:

> "Arthur" <arfa AT clara DOT net> wrote:
> 
> > >   I put some results of compiling the BYTE benchmarks with gcc,
> > > egcs and pgcc
> > > using differentswitchs. I included the results of an EXE compiled
> > > with Watcom
> > > 10.0a.
> > >
> > 
> > Very interesting. One minor gripe: your tables are wider than the page so I
> > had to print them all in landscape to see the figures! Suppose it can't be
> > helped...
> 
> Reduce the font, both (Netscape and IE) have this option. I used it at home 
> (800x600) and don't needed it at work (1024x768).
>  
> > >From what I can gather from my own experience and the experience of a few
> > other people I know, PGCC works very well on non-intel processors, while GCC
> > 2.8.0 doesn't seem to like non-intels that much. I've found that PGCC really
> > slows down my code with -O6, while gcc 2.8.0 with -march=pentium etc. gives
> > me a rather large speed boost (Large being one or two percent :-).
> > 
> > BTW, according to the Byte benchmark that I had a few moths ago, my P200MMX
> > is a 234.3MHz Pentium on average (I think it was roughly the same for both
> > integer and floats) - that particular benchmark isn't that accurate. No,
> > I've not overclocked my processor.
> 
> That's very close! the benchmark says your computer is XXX MHz compared with 
> normal Pentium, no MMX Pentium. There are 10% of difference due to a much 
> better cache strategy used in MMX CPUs so you should spect 220 MHz from the 
> Benchmark ;-) 
> 
> In some days I'll upload a new table using geometric average, that's a little 
> bit better for comparisson in this case (can be much better in other cases).
> 

Tried also my build of pgcc-1.1 (I built it in September but didn't really used
it). There were no significant gain (less than 5%) but it benchmarks
were crashing near the end. With egcs-1.1.1 all went Ok.

One more thing to test is to compare egcs-1.1.1 built with HAIFA scheduler
enabled (all currently available binaries for egcs-1.1X) and disabled.
Perhaps it is necessary to test it before released normal release
of egcs-1.1.1 fpr DJGPP

Andris