Date: Wed, 12 Aug 1998 11:26:57 +0300 (IDT)
From: Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>
To: D DOT J DOT Eleveld AT anest DOT azg DOT nl
cc: djgpp AT delorie DOT com
Subject: Re: How safe is the -march=pentium flag?
In-Reply-To: <6qnh4u$db4$1@nnrp1.dejanews.com>
Message-ID: <Pine.SUN.3.91.980812112555.16917A-100000@is>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Precedence: bulk


On Mon, 10 Aug 1998 D DOT J DOT Eleveld AT anest DOT azg DOT nl wrote:

> > How huge is ``huge''?  Can you give some numbers?
> 
> Ok, I was playing with optimizing the inner loops of the AllegPeg Mpeg player
> library for Allegro.  I do my profiling stuff with some pentium-specific code
> that counts clock cycles.  With -march=pentium I get 27.7M clock cycles per
> frame with my Pentiun 133 with painfully slow (banked) video card.  Exactly
> the same flags and code (just changed the compiler options in Rhide, and
> rebuilt) gets me 32.6M clock cycles per frame.

The difference between 27.7 and 32.6 is merely 17%.  I would hardly
call that a ``huge speed increase''.  Numbers around 10%-20% are what
you would usually get by playing with different optimization
switches.  See section 14.2 of the DJGPP FAQ for more details.