Sender:	law AT sgi DOT com
Message-ID:	<38AD6A13.AA6FA791@sgi.com>
Date:	Fri, 18 Feb 2000 07:49:39 -0800
From:	Linda Walsh <law AT sgi DOT com>
X-Mailer:	Mozilla 4.7 [en] (X11; I; Linux 2.2.14 i686)
X-Accept-Language:	en
MIME-Version:	1.0
To:	Marc Lehmann <marc AT gimp DOT org>
CC:	pgcc AT delorie DOT com
Subject:	Re: Big slowdown on -mpentium/-mpentiumpro, short_int add
References:	<38AC20C0 DOT 55E33FA4 AT sgi DOT com> <20000218130553 DOT C1399 AT cerebro DOT laendle>
Reply-To:	pgcc AT delorie DOT com

Marc Lehmann wrote:
> 
> On Thu, Feb 17, 2000 at 08:24:32AM -0800, Linda Walsh <law AT sgi DOT com> wrote:
> > Showed about a 30-40% slowdown when using the -mpentium or -mpentiumpro
> 
> on which cpu, and slowdown compared to what?
---
	on a Pentium-III.

Flags fast on short_int:

CCOPT=-O2 -fomit-frame-pointer -m486 -malign-loops=2 -malign-jumps=2 -malign-functions=2 -DCPU=686  

vs.  Adding the -mpentiumpro or -mpentium to the above set.

> > Note the test suite is copyright, but I'm excerpting this section for
> > educational (to demonatrate a problem) purposes.
> 
> Das that make it more legal ;) Anyway, thanks a lot for the snippet!
---
	Under copyright law, excerpting bits for educational purposes
is permitted! :-)  Education is a protected area.  Another is for
purposes of a review of an article or something (unless UTICA passes, then
all bets are off).  I can see a post-UTICA world where books are not
sold to people, but "licensed" -- and the terms of the license would
prohibit any review critical of the work.  Oi vey.

> I do not know the reason for the slowdown (yet), but "short" is about
> the slowest datatype for pentium-type systems, and it could very well be
> that suboptimal scheduling even of a single insns can make such a large
> difference.
---
	I also have some P-II's here, I can try it on that later when I
get the time.  I hadn't had time to dump assembler and peek yet either,
My work has only been sucking up most of my brain lately...:-)
-l

- Raw text -