From owner-freebsd-alpha  Mon Apr  2 13:14:48 2001
Delivered-To: freebsd-alpha@freebsd.org
Received: from duke.cs.duke.edu (duke.cs.duke.edu [152.3.140.1])
	by hub.freebsd.org (Postfix) with ESMTP id AE97137B71E
	for <freebsd-alpha@freebsd.org>; Mon,  2 Apr 2001 13:14:44 -0700 (PDT)
	(envelope-from gallatin@cs.duke.edu)
Received: from grasshopper.cs.duke.edu (grasshopper.cs.duke.edu [152.3.145.30])
	by duke.cs.duke.edu (8.9.3/8.9.3) with ESMTP id QAA18794;
	Mon, 2 Apr 2001 16:14:40 -0400 (EDT)
Received: (from gallatin@localhost)
	by grasshopper.cs.duke.edu (8.11.3/8.9.1) id f32KEAl94038;
	Mon, 2 Apr 2001 16:14:10 -0400 (EDT)
	(envelope-from gallatin@cs.duke.edu)
From: Andrew Gallatin <gallatin@cs.duke.edu>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Message-ID: <15048.56722.573307.129796@grasshopper.cs.duke.edu>
Date: Mon, 2 Apr 2001 16:14:10 -0400 (EDT)
To: "Alan L. Cox" <alc@imimic.com>
Cc: freebsd-alpha@freebsd.org
Subject: Re: do we care about performance yet?
In-Reply-To: <3AC63E6B.3D30B2B6@imimic.com>
References: <3AC63E6B.3D30B2B6@imimic.com>
X-Mailer: VM 6.75 under 21.1 (patch 12) "Channel Islands" XEmacs Lucid
Sender: owner-freebsd-alpha@FreeBSD.ORG
Precedence: bulk
X-Loop: FreeBSD.org


Alan L. Cox writes:
 > >    bzero                                    92338   6.5   2.6 
 > >    bcopy_samealign_lp                       71123   5.0   2.0 
 > 
 > I built a kernel last year using Compaq's implementation of these
 > routines.  A lot can be gained by using 21264-optimized
 > implementations.  (Unfortunately, the code I was given by Compaq was
 > under NDA.)  Alpha Processor (now API Networks) has, however, released
 > optimized implementations for Linux.  There are patches on their web
 > site.  Having talked to one hybrid business/technical person there, I
 > doubt that they would balk at dual-licensing these patches under a BSD
 > and GPL license.
 > 
 > Alan

Thanks for the pointer.  I'm currently talking to him about it.
Hopefully we'll be able to work something out.

I plugged his ev6 memcpy into lmbench and I see roughly a 40%
improvement for larger than cache copies on a UP1000.  I see about 25%
improvement on a DS10 & XP1000.  It bascially gives you a free upgrade
to the memory performance of the next-higher class of machine ;-)

For things that are entirely in the bcache or l1 cache, the
performance improvement is marginal, but still there.

Drew

To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-alpha" in the body of the message