Date: Wed, 29 Oct 1997 11:04:31 +0200 (IST)
From: Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>
To: Bram Stolk <bram AT nuson DOT nl>
cc: djgpp AT delorie DOT com
Subject: Re: FYI: speed of Allegro/DJGPP
In-Reply-To: <34548295.731EF500@nuson.nl>
Message-ID: <Pine.SUN.3.91.971029110354.26592P-100000@is>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Precedence: bulk


On Mon, 27 Oct 1997, Bram Stolk wrote:

> With this option, a preprocessor, compiler, assembler, linkerloader are
> all started simultaneously, and communicate with IPC (pipes).
> 
> This is especially a very cool option to use if you have a
> multiprocessor machine, and enough ram.

I think it's high time we stop talking about ``cool options'' like my
kids do and start looking under the hood, don't you think?

I'd bet that with multiple CPUs and lots of RAM, the compilation runs
fast enough even without -pipe.  You might time your compilations some
day to see how much speedup does -pipe actually bring.

> I used to compile my sources with DJGPP.
> Now I crosscompile them from Linux, WITH the -pipe flag.
> Together with the much better filesystem linux has, my
> compilation times have almost vanished :-)

And I pointed my TMPDIR into a 5MB RAM drive, and my compilation times
disappeared also, even with the old good-for-nothing DOS filesystem
and *without* the ``cool'' -pipe option.

If you search the DJGPP mail archives, I think you will find that it
was shown here time and again that, given an optimal setup, lots of
RAM, and the same processor and disk, DJGPP on DOS doesn't fall behind
Lunix by more than 30%.  Anybody who looks at the numbers will realize
that ``cool'' OS features mean much less for performance than raw CPU
and I/O power and a good system setup.