X-Authentication-Warning: delorie.com: mail set sender to djgpp-bounces using -f
X-Recipient: djgpp AT delorie DOT com
Date: Tue, 10 Feb 2009 11:31:19 +0200
From: Eli Zaretskii <eliz AT gnu DOT org>
Subject: Re: djgpp on 386
In-reply-to: <89823734-86ca-4c73-b8d8-9e08f564223c@m40g2000yqh.googlegroups.com>
X-012-Sender: halo1 AT inter DOT net DOT il
To: djgpp AT delorie DOT com
Message-id: <u4oz2skns.fsf@gnu.org>
References: <598659 DOT 78008 DOT qm AT web31913 DOT mail DOT mud DOT yahoo DOT com> <89823734-86ca-4c73-b8d8-9e08f564223c AT m40g2000yqh DOT googlegroups DOT com>
Reply-To: djgpp AT delorie DOT com
Errors-To: nobody AT delorie DOT com
X-Mailing-List: djgpp AT delorie DOT com
X-Unsubscribes-To: listserv AT delorie DOT com
Precedence: bulk

> From: Rugxulo <rugxulo AT gmail DOT com>
> Date: Mon, 9 Feb 2009 21:15:10 -0800 (PST)
> 
> GCC 3.4.4 (compiled by me specifically for -march=pentium") runs at
> least four times slower on my P166 (at least compiling TDE 5.1v).

Like I said: 2.7.2 is the leanest, meanest GCC in the West.

> I don't know why, Haifa scheduler?

Who knows?  I guess the main reason is that being lean and mean is no
longer a worthy goal in the eyes of GCC developers as much as it was
before.

> I wonder if 386/486 optimizations regressed since then but haven't
> tested. Some people also claim .EXE size has gone up ever since,
> which is actually fairly true.

I challenge anyone who cares to show a test case where a real-life C
program (not one of those "benchmarks" prepared solely to show off
optimizer features) runs more than 10% faster under GCC 3.x or 4.x
than when compiled by GCC 2.7.2.

Me, I will trade 10% speedup for the ability to debug optimized code
any time.  To say nothing of 4-fold compilation speedup.

Thanks for making a new build of GCC 2.7.2.3 available.  Too bad it
didn't support DWARF-2 debug info, but that disadvantage pales in
comparison with its advantages.