Message-Id: <3.0.2.32.19990510212615.012c72e8@pop3.digibel.be>
X-Sender: pascutto AT pop3 DOT digibel DOT be
X-Mailer: QUALCOMM Windows Eudora Light Version 3.0.2 (32)
Date: Mon, 10 May 1999 21:26:15 +0200
To: pgcc AT delorie DOT com
From: Gian-Carlo Pascutto <GianCarlo DOT Pascutto AT advalvas DOT be>
Subject: Re: Optimization question
In-Reply-To: <Pine.LNX.4.04.9905100834040.13078-100000@sal.physics.ucsb.
 edu>
References: <Pine DOT SOL DOT 3 DOT 96 DOT 990510125208 DOT 12696D-100000 AT ursa DOT cus DOT cam DOT ac DOT uk>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Reply-To: pgcc AT delorie DOT com

At 09:09 10/05/99 -0700, you wrote:
>On Mon, 10 May 1999, Dr H. T. Leung wrote:
>
>>If you had read the mailing list archive, it is terribly unfair to 
>>people on the list to cc you replies when you are not on the list. 
>>If you want to ask a question, subscribe, read on for a while, then 
>>post (then maybe unsubscribe). 
>
>That's EXTREMELY strange.

What's so strange about it ?

>Optimizing comiplers like egcs/pgcc do "strength reduction"

I always thought strength reduction was generally considered a 
bad idea on register starved machines like the x86 architecture.
Then again, I might be mistaken...

>According to my timing tests pgcc hasn't gained me anything over egcs so
>far, but I keep hoping.

You might try reducing the optimization level.  I've had code nearly
doubling speed after changing -O6 into -O and setting the cpu switch
wrong on purpose. At this kind of optimization levels things don't 
always work out as they were ment to be. But then again, my code was
using long longs, which is a long way from floating point math.
Trying won't hurt though...

Also, as egcs is incorporating more and more of pgcc's optimizations
that are known to work well, the performance gap is closing fast.

Greetings,
Gian-Carlo