Date: Tue, 2 Mar 1999 13:03:52 -0600 (CST)
From: Dustin Marquess <jailbird AT alcatraz DOT fdf DOT net>
To: "Eugene M. Indenbom" <bom AT classic DOT iki DOT rssi DOT ru>
cc: "'pgcc AT delorie DOT com'" <pgcc AT delorie DOT com>
Subject: Re: gcc-2.7.0 creates faster code than pgcc-1.1.1
In-Reply-To: <01BE64C5.8B5FD140@valya.iki.rssi.ru>
Message-ID: <Pine.LNX.4.04.9903021302300.1724-100000@alcatraz.fdf.net>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Reply-To: pgcc AT delorie DOT com

On Tue, 2 Mar 1999, Eugene M. Indenbom wrote:

> I make gzip using pgcc with the following oprions:
> 
> make CC="/home/ddict/glibc/pgcc-1.1.1/gcc/xgcc -B/home/ddict/glibc/pgcc-1.1.1/gcc/" \
> CFLAGS="-Os -fomit-frame-pointer -mpentiumpro -march=pentiumpro -fno-exceptions 
> -malign-loops=2 -malign-jumps=2 -malign-functions=2"
> 
> This combination appears to produce the fastest code I can get on my Pentium Pro 180MHz (192MB RAM):
> 
> $ time ./gzip -cd egcs-1.1.1.tar.gz > /dev/null
> 
> real    0m7.296s
> user    0m7.170s
> sys     0m0.130s
> 
> But gzip that I get from very old Slackware distribution (1995) works
> much faster. It was compiled with gcc-2.7.0 -O2 and works more than 5% faster!
> 
> $ time /bin/gzip -cd egcs-1.1.1.tar.gz > /dev/null
> 
> real    0m6.942s
> user    0m6.730s
> sys     0m0.210s
> 
> Stock egcs-1.1.1 looses competition too:
> make CFLAGS="-Os -fomit-frame-pointer -mpentiumpro -march=pentiumpro -fno-exceptions 
> -malign-loops=2 -malign-jumps=2 -malign-functions=2"
> 
> $ time ./gzip -cd egcs-1.1.1.tar.gz > /dev/null
> 
> real    0m7.571s
> user    0m7.350s
> sys     0m0.230s
> 
> I send you this fancy executable as attachment and really want to know
> how can this be, why we loose 5% of speed in the 2.9.x gcc code using
> all optimizations invented during past 4 years?
> 
> Sincerely Yours, Eugene.
>  
> P.S. I am not subscribed to this mailing list 

	-Os optimizes for smallest binary size, not speed.. Try:

CFLAGS="-O6 -fomit-frame-pointer -mpentiumpro -march=pentiumpro
-fno-exceptions -pipe -s -malign-loops=2 -malign-jumps=2 -malign-functions=2"
LDFLAGS="-s"

	Also, is this old Slackware still a.out, or is it one of the
ELF versions?  ELF can be 1-5% slower than a.out..

						Thanks,
						-Dustin