>From [log in to unmask] Wed Sep 29 00:25:52 1999
Return-Path: <[log in to unmask]>
X-Useless: [log in to unmask]
Received: (qmail 9266 invoked by uid 1013); 29 Sep 1999 00:25:52 -0000
X-Useless: [log in to unmask]
Received: (qmail 9261 invoked from network); 29 Sep 1999 00:25:49 -0000
Received: from mailout1.mailbase.ac.uk (128.240.226.11)
  by nemo.netservice.pl with SMTP; 29 Sep 1999 00:25:49 -0000
Received: from naga.mailbase.ac.uk (naga.mailbase.ac.uk [128.240.226.3])
	by mailout1.mailbase.ac.uk (8.9.1a/8.9.1) with ESMTP id BAA18813;
	Wed, 29 Sep 1999 01:23:53 +0100 (BST)
Received: (from daemon@localhost) 
        by naga.mailbase.ac.uk (8.8.x/Mailbase) id BAA10592;
        Wed, 29 Sep 1999 01:18:16 +0100 (BST)
Received: from augean.eleceng.adelaide.edu.au (augean.eleceng.adelaide.edu.au [129.127.28.4]) 
        by naga.mailbase.ac.uk (8.8.x/Mailbase) with ESMTP id BAA10579;
        Wed, 29 Sep 1999 01:18:10 +0100 (BST)
Received: from franklin.eleceng.adelaide.edu.au (franklin [129.127.28.130])
	by augean.eleceng.adelaide.edu.au (8.8.5/8.8.5/ElecEng-1.0) with ESMTP id JAA23377;
	Wed, 29 Sep 1999 09:48:05 +0930 (CST)
Message-Id: <[log in to unmask]>
Received: from FRANKLIN/SpoolDir by franklin.eleceng.adelaide.edu.au (Mercury 1.44);
    29 Sep 99 09:48:02 GMT +0930
Received: from SpoolDir by FRANKLIN (Mercury 1.44); 29 Sep 99 09:47:53 GMT +0930
Organization: The University of Adelaide
Date: Wed, 29 Sep 1999 09:47:48 +0930
MIME-Version: 1.0
Content-type: text/plain; charset=US-ASCII
Content-transfer-encoding: 7BIT
Priority: normal
In-reply-to: <[log in to unmask]>
X-mailer: Pegasus Mail for Win32 (v3.12a)
Subject: Re: Linux F90 comparisons at Polyhedron
From: "David Vowles" <[log in to unmask]>
To: [log in to unmask]
Cc: [log in to unmask]
X-List: [log in to unmask]
X-Unsub: To leave, send text 'leave comp-fortran-90' to [log in to unmask]
X-List-Unsubscribe: <mailto:[log in to unmask]>
Reply-To: "David Vowles" <[log in to unmask]>
Sender: [log in to unmask]
Errors-To: [log in to unmask]
Precedence: list
Content-Transfer-Encoding: 7BIT

As a matter of interest, I have found that the use of the LAHEY (LF90 4.x) 
equivalent of NAG's -float-store (-ap) while being detrimental to performance 
is infact essential to the correct operation of the widely used LAPACK library 
which makes extensive use of floating point tests for equality. If the contents 
of 80 bit registers are not first stored then equality tests may (and do) yield 
unexpected results. 

Of course, robustness requires that floating point equality tests not be 
performed as: IF (X.EQ.Y) THEN but rather: IF (ABS(X-Y) .LE. TOL) THEN or 
variations on this theme. However, the practicality is that many widely used 
libraries do not adopt such defensive schemes, and so for me prudence dictates 
the use of -float-store (or their equivalents) unless I am certain that it is 
unnecessary.


Date sent:      	Tue, 28 Sep 1999 13:57:26 -0600
Organization:   	Los Alamos National Laboratory,  Group T-1
Subject:        	Linux F90 comparisons at Polyhedron
From:           	Neil Carlson <[log in to unmask]>
To:             	[log in to unmask]
Send reply to:  	Neil Carlson <[log in to unmask]>

> I checked out the Intel/Linux F90 compiler comparisons at Polyhedron
> (http://www.polyhedron.co.uk/compare/linux/f90bench.html) and one of
> the compiler switches for the NAG compiler looks really fishy to me.
> It reports using "-float-store".  This switch prevents floating point
> variables from being stored in registers and can result in *much*
> worse performance.  (This ensures that intermediate 80 bit fp results
> get truncated to 64 bit values.)  Was the equivalent switch thrown on
> the other compilers?  These benchmark results are completely out of
> line with benchmarks of my own.
> 
> -Neil
> ------------------------------------------------------------------------
> Neil N. Carlson                      [log in to unmask] (work)
> Theoretical Division                 [log in to unmask] (non-work)
> Group T-1, Mailstop B221
> Los Alamos National Laboratory	     505-665-1220 (voice)
> Los Alamos, NM 87545		     505-665-5757 (fax)


Regards,
David.

----------------------------------------------------------

David Vowles
Research Officer
Department of Electrical and Electronic Engineering
The University of Adelaide
Australia 5005

Voice:     +61 8 8303 5416
Fax:       +61 8 8303 4360
Email:     [log in to unmask]
Home Page: http://www.eleceng.adelaide.edu.au/Personal/dvowles/home.html

>From [log in to unmask] Wed Sep 29 06:22:51 1999


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%