From mboxrd@z Thu Jan  1 00:00:00 1970
From: Ricardo Wurmus <rekado@elephly.net>
Subject: Re: OpenBLAS and performance
Date: Wed, 20 Dec 2017 21:00:46 +0100
Message-ID: <87bmitxj81.fsf@elephly.net>
References: <20171219104956.GB806@thebird.nl>
	<87tvwl7h4w.fsf@albion.it.manchester.ac.uk>
	<87h8sl78vp.fsf@albion.it.manchester.ac.uk>
	<20171220172215.GA7926@thebird.nl> <87d139xo3v.fsf@elephly.net>
	<20171220192802.GA8426@thebird.nl>
Mime-Version: 1.0
Content-Type: text/plain
Return-path: <guix-devel-bounces+gcggd-guix-devel=m.gmane.org@gnu.org>
Received: from eggs.gnu.org ([2001:4830:134:3::10]:53409)
	by lists.gnu.org with esmtp (Exim 4.71)
	(envelope-from <rekado@elephly.net>) id 1eSTMl-0001eW-Ft
	for guix-devel@gnu.org; Fri, 22 Dec 2017 14:52:24 -0500
Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71)
	(envelope-from <rekado@elephly.net>) id 1eSTMh-0005X9-MT
	for guix-devel@gnu.org; Fri, 22 Dec 2017 14:52:23 -0500
Received: from sender-of-o51.zoho.com ([135.84.80.216]:21141)
	by eggs.gnu.org with esmtps (TLS1.0:RSA_AES_256_CBC_SHA1:32)
	(Exim 4.71) (envelope-from <rekado@elephly.net>) id 1eSTMh-0005WL-E4
	for guix-devel@gnu.org; Fri, 22 Dec 2017 14:52:19 -0500
In-reply-to: <20171220192802.GA8426@thebird.nl>
List-Id: "Development of GNU Guix and the GNU System distribution."
	<guix-devel.gnu.org>
List-Unsubscribe: <https://lists.gnu.org/mailman/options/guix-devel>,
	<mailto:guix-devel-request@gnu.org?subject=unsubscribe>
List-Archive: <http://lists.gnu.org/archive/html/guix-devel/>
List-Post: <mailto:guix-devel@gnu.org>
List-Help: <mailto:guix-devel-request@gnu.org?subject=help>
List-Subscribe: <https://lists.gnu.org/mailman/listinfo/guix-devel>,
	<mailto:guix-devel-request@gnu.org?subject=subscribe>
Errors-To: guix-devel-bounces+gcggd-guix-devel=m.gmane.org@gnu.org
Sender: "Guix-devel" <guix-devel-bounces+gcggd-guix-devel=m.gmane.org@gnu.org>
To: Pjotr Prins <pjotr.public12@thebird.nl>
Cc: guix-devel@gnu.org, Dave Love <fx@gnu.org>


Pjotr Prins <pjotr.public12@thebird.nl> writes:

>> > If I compile for a target it
>> > makes a large difference.
>> 
>> The FAQ document[1] says this:
>> 
>>   The environment variable which control the kernel selection is
>>   OPENBLAS_CORETYPE (see driver/others/dynamic.c) e.g. export
>>   OPENBLAS_CORETYPE=Haswell. And the function char*
>>   openblas_get_corename() returns the used target.
>> 
>> [1]: https://github.com/xianyi/OpenBLAS/wiki/Faq
>> 
>> Have you tried this and compared the performance?
>
> About 10x difference on 24+ cores for matrix multiplication (my
> version vs what comes with Guix).
>
> I do think we need to default to a conservative openblas for general
> use. Question is how we make it fly on dedicated hardware.

Have you tried preloading the special library with LD_PRELOAD?

-- 
Ricardo

GPG: BCA6 89B6 3655 3801 C3C6  2150 197A 5888 235F ACAC
https://elephly.net