From owner-sc22wg5+sc22wg5-dom8=www.open-std.org@open-std.org Thu Jul 6 00:19:47 2017 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 4ADB7358758; Thu, 6 Jul 2017 00:19:46 +0200 (CEST) Delivered-To: sc22wg5@open-std.org Received: from rgout05.bt.lon5.cpcloud.co.uk (rgout0507.bt.lon5.cpcloud.co.uk [65.20.0.228]) by www.open-std.org (Postfix) with ESMTP id 8A3A93567EE for ; Thu, 6 Jul 2017 00:19:44 +0200 (CEST) X-OWM-Source-IP: 31.49.153.6 (GB) X-OWM-Env-Sender: d.t.muxworthy@btinternet.com X-Junkmail-Premium-Raw: score=10/50,refid=2.7.2:2017.7.5.214816:17:10.910,ip=,rules=__HAS_FROM, __CT, __CTYPE_MULTIPART_ALT, __CTYPE_HAS_BOUNDARY, __CTYPE_MULTIPART, __BOUNCE_CHALLENGE_SUBJ, __BOUNCE_NDR_SUBJ_EXEMPT, __REFERENCES, __TO_MALFORMED_2, __TO_NO_NAME, __HAS_MSGID, __SANE_MSGID, __MSGID_APPLEMAIL, __MIME_VERSION, __HAS_X_MAILER, __X_MAILER_APPLEMAIL, __TO_IN_SUBJECT2, MIME_LOWER_CASE, __MIME_TEXT_P2, __MIME_TEXT_H2, __ANY_URI, __URI_NO_WWW, __SUBJ_ALPHA_NEGATE, SUPERLONG_LINE, __HTML_AHREF_TAG, __FORWARDED_MSG, __URI_WITHOUT_PATH, __HAS_HTML, __HTML_TAG_DIV, HTML_NO_HTTP, BODYTEXTP_SIZE_3000_LESS, BODYTEXTH_SIZE_3000_MORE, BODYTEXTH_SIZE_10000_LESS, __MIME_TEXT_H1, __MIME_TEXT_P1, __MIME_HTML, __TAG_EXISTS_HTML, __URI_NS, HTML_70_90, __USER_AGENT_APPLEMAIL, MSG_THREAD, LEGITIMATE_SIGNS, TO_IN_SUBJECT, __MIME_TEXT_H, __MIME_TEXT_P, REFERENCES, NO_URI_HTTPS Received: from davids-imac.home (31.49.153.6) by rgout05.bt.lon5.cpcloud.co.uk (9.0.019.13-1) (authenticated as d.t.muxworthy@btinternet.com) id 5957E506006479D7 for sc22wg5@open-std.org; Wed, 5 Jul 2017 23:19:44 +0100 From: David Muxworthy Content-Type: multipart/alternative; boundary="Apple-Mail=_4AA88D1C-35E5-4DBF-803D-FDE6D3A047CE" Subject: Re: (SC22WG5.5898) 3 levels of parallelism? Date: Wed, 5 Jul 2017 23:19:43 +0100 References: To: sc22wg5@open-std.org Message-Id: <48C3219D-CC20-4EB0-8B39-3A5BEBFCFCE4@bcs.org.uk> Mime-Version: 1.0 (Mac OS X Mail 9.3 \(3124\)) X-Mailer: Apple Mail (2.3124) Sender: owner-sc22wg5@open-std.org Precedence: bulk --Apple-Mail=_4AA88D1C-35E5-4DBF-803D-FDE6D3A047CE Content-Transfer-Encoding: quoted-printable Content-Type: text/plain; charset=utf-8 Due to a glitch in the sc22wg5 mail server (it seems to have mislaid the = most recent whitelist) this message from Brian Friesen was incorrectly rejected. > Begin forwarded message: >=20 > From: Brian Friesen > Subject: Re: (j3.2006) (SC22WG5.5898) 3 levels of parallelism? > Date: 5 July 2017 at 23:05:34 BST > To: fortran standards email list for J3 > Cc: sc22wg5 > Reply-To: fortran standards email list for J3 = >=20 > On Wed, Jul 5, 2017 at 12:49 PM, Clune, Thomas L. (GSFC-6101) = > wrote: > I think the concern is that in practice, the user needs to be able to = control the division of work between the two levels to optimize = performance. A model where the compiler will make the decisions will = be easier to use, but =E2=80=A6 >=20 > The history of optimization in real codes is littered with the = difference between =E2=80=9Cin theory=E2=80=9D and =E2=80=9Cin = practice=E2=80=9D. >=20 > But isn't the point of the language to be descriptive, not = prescriptive? Controlling the division of work among different types of = parallelism seems like a job for programming models like OpenMP, = OpenACC, etc. They evolve rapidly and follow architecture trends. OpenMP = already supports both thread- and SIMD-level parallelism. (Don't know = about OpenACC.) >=20 > Or did you have something more subtle in mind? --Apple-Mail=_4AA88D1C-35E5-4DBF-803D-FDE6D3A047CE Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset=utf-8 Due to a glitch in the sc22wg5 mail server (it seems to have = mislaid the most recent whitelist) this message

from = Brian Friesen was incorrectly rejected.

Begin = forwarded message:

From: = Brian Friesen <bfriesen@lbl.gov>
Subject: = Re: (j3.2006) = (SC22WG5.5898) 3 levels of parallelism?
Date: = 5 July 2017 at 23:05:34 BST
To: = fortran standards email list = for J3 <j3@mailman.j3-fortran.org>
Cc: = sc22wg5 <sc22wg5@open-std.org>
Reply-To: = fortran standards email list = for J3 <j3@mailman.j3-fortran.org>

On Wed, = Jul 5, 2017 at 12:49 PM, Clune, Thomas L. (GSFC-6101) <thomas.l.clune@nasa.gov> = wrote:

I think the concern is that in practice, the user needs = to be able to control the division of work between the two levels to = optimize performance. A model where the compiler will make the = decisions will be easier to use, but =E2=80=A6

The history of optimization in real codes is littered = with the difference between =E2=80=9Cin theory=E2=80=9D and =E2=80=9Cin = practice=E2=80=9D.

But isn't the point of the language to = be descriptive, not prescriptive? Controlling the division of work among = different types of parallelism seems like a job for programming models = like OpenMP, OpenACC, etc. They evolve rapidly and follow architecture = trends. OpenMP already supports both thread- and SIMD-level parallelism. = (Don't know about OpenACC.)

Or did you have something more subtle in = mind?

= --Apple-Mail=_4AA88D1C-35E5-4DBF-803D-FDE6D3A047CE--