From owner-sc22wg5+sc22wg5-dom8=www.open-std.org@open-std.org Fri Apr 5 22:39:28 2013 Return-Path: X-Original-To: sc22wg5-dom8 Delivered-To: sc22wg5-dom8@www.open-std.org Received: by www.open-std.org (Postfix, from userid 521) id B2E2C356C92; Fri, 5 Apr 2013 22:39:28 +0200 (CEST) Delivered-To: sc22wg5@open-std.org Received: from exprod6og120.obsmtp.com (exprod6og120.obsmtp.com [64.18.1.236]) by www.open-std.org (Postfix) with ESMTP id 31F9A356C92 for ; Fri, 5 Apr 2013 22:39:21 +0200 (CEST) Received: from CFWEX01.americas.cray.com ([136.162.34.11]) (using TLSv1) by exprod6ob120.postini.com ([64.18.5.12]) with SMTP ID DSNKUV82ePwgc5V1R05bUha/SXE7E2VWJM0g@postini.com; Fri, 05 Apr 2013 13:39:26 PDT Received: from fortran.us.cray.com (172.31.19.200) by CFWEX01.americas.cray.com (172.30.88.25) with Microsoft SMTP Server id 14.2.342.3; Fri, 5 Apr 2013 15:30:55 -0500 Message-ID: <515F3505.3000806@cray.com> Date: Fri, 5 Apr 2013 15:33:09 -0500 From: Bill Long Reply-To: Organization: Cray Inc. User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.6; rv:17.0) Gecko/20130328 Thunderbird/17.0.5 MIME-Version: 1.0 To: sc22wg5 Subject: Re: (j3.2006) (SC22WG5.4958) [ukfortran] AW: AW: WG5 ballot on first draft TS 18508, Additional Parallel Features in Fortran (Update) References: <20130329104945.2C28A356BB3@www.open-std.org> <20130329232930.A368A356DC2@www.open-std.org> <20130330132208.D7468356DDA@www.open-std.org> <20130330141442.761E2356D8C@www.open-std.org> In-Reply-To: <20130330141442.761E2356D8C@www.open-std.org> Content-Type: text/plain; charset="ISO-8859-1"; format=flowed Content-Transfer-Encoding: 7bit Sender: owner-sc22wg5@open-std.org Precedence: bulk On 3/30/13 9:14 AM, N.M. Maclaren wrote: > On Mar 30 2013, Bader, Reinhold wrote: >>> >>> If we added this, it would be CO_PRODUCT since the local one is PRODUCT. >>> The previous proposal had CO_PRODUCT. It was removed because the >>> corresponding MPI_REDUCE for that operation almost never occurs in real >>> codes. Is there any common usage of this operation? Also, is there >>> hardware support in network hardware for a multiply reduction? >> >> According to Mellanox' FCA documentation, all commonly used numeric data >> types (except complex) support MPI_Reduce and MPI_Allreduce; I would >> assume that this includes the argument variants MPI_SUM and MPI_PRODUCT. Not sure about that. But the decision on whether to add CO_PRODUCT should be based on whether it is sufficiently useful to warrant the added clutter. We voted on this once before at WG5, but perhaps need to discuss and vote again in June. In terms of edits disruption to the TS, adding CO_PRODUCT is (hopefully) not significant. > > Complex summation isn't exactly hard to build on top of real summation > in the software interface :-) Complex product is more interesting. > >> I must however admit that a significant (if not the bulk) part of the FCA >> optimization is concerned with topology awareness, so will also be >> applicable to the general reduction facility. > > How much of that gets through the driver layer is less clear. When I > last looked, OpenIB didn't support RDMA, which makes a lot of the fancy > MPI one-sided and PGAS stuff irrelevant. Not sure about OpenIB, but the Mellanox IB hardware certainly supports RDMA. Cheers, Bill > > > Regards, > Nick Maclaren. > > _______________________________________________ > J3 mailing list > J3@mailman.j3-fortran.org > http://mailman.j3-fortran.org/mailman/listinfo/j3 > -- Bill Long longb@cray.com Fortran Technical Support & voice: 651-605-9024 Bioinformatics Software Development fax: 651-605-9142 Cray Inc./Cray Plaza, Suite 210/380 Jackson St./St. Paul, MN 55101