From jwagener@ionet.net Sat Nov 4 17:32:23 1995 Received: from ion3.ionet.net (ion3.ionet.net [204.96.200.8]) by dkuug.dk (8.6.12/8.6.12) with ESMTP id RAA22425 for ; Sat, 4 Nov 1995 17:32:19 +0100 Received: from erehwon (tsip48.ionet.net [206.28.164.57]) by ion3.ionet.net (8.6.12/8.6.12) with SMTP id KAA10126; Sat, 4 Nov 1995 10:32:01 -0600 X-Mailer: InterCon TCP/Connect II 2.2.1 MIME-Version: 1.0 Message-Id: <9511041137.AA14617@erehwon> Date: Sat, 4 Nov 1995 11:37:14 -0500 From: "Jerrold L. Wagener" To: sc22wg5@dkuug.dk Cc: jwagener@ionet.net Subject: JLW informal report on the Nov 1-3 HPFF meeting Content-Type: Multipart/Mixed;boundary=part_ACC100E400004F1E00000001 --part_ACC100E400004F1E00000001 Content-Transfer-Encoding: Quoted-Printable Content-Type: Text/Plain; charset=US-ASCII Content-Disposition: Inline This meeting took place in Arlington, TX, with 26 attendees, 22 = voting members. There were two "procedural" decisions of importance: = (1) the kernel replaces the subset and becomes HPF-2, with the = full language becoming HPF 2000, and (2) all future technical votes = require 2/3 majority. The main distinction between the kernel and = the subset is that whereas the subset contained all of the = distribution directives and part of F90, the kernel contains (the = highest performance) part of the distribution directives and all of F90 = (or F95). Features gaining acceptance on second (final) readings at this = meeting include: interoperability with C, asynchronous I/O, = generalized transpose, indirect mappings (analogous to = vector-valued subscripts), mapping to processor subsets, ability to = map structure components, and user-defined reduction operators. = The tasking proposal, which has been around for a while in various = forms, has still not progressed to the second reading stage. = Initial discussions on an MPI binding and calling HPF from C++ = received go-ahead green lights (a straw vote on "an MPI binding = is important" came out 24-0-2). Of these various items, probably those = of the most immediate interest to WG5/X3J3 are interoperability = and asynchronous I/O, and possibly generalized transpose and an MPI binding. The approved interoperability with C facility is very much like the = one presented in the September HPFF meeting (distributed to SC22WG5 = in September). In a straw vote HPFF strongly favored "@" over "Loc()" = for the address operator, but wants X3J3's counsel on this. = (Initial thoughts were against "&", which C uses for this purpose, = because of the continuation role "&" plays in F90; however, on = reflection there does not appear to be a true ambiguity with using "&" = for this purpose in Fortran as well.) Missing pieces from = the interoperability facility: accommodating structures and a map_to_= C function outside the interface block. The approved asynchronous I/O facility is also similar to the one = presented in September, the main difference being that ID=3D is required = in the WAIT statement. Recall that the two principal features of = this asynchronous I/O facility are addition of the ID=3D spec in the = READ/WRITE statement and the addition of the WAIT statement; ID=3D = causes the READ/WRITE statement to be asynchronous and WAIT = provides synchronization when needed. HPFF proposes this asynchronous I/= O facility for unformatted direct I/O (that's the critical I/O for = high performance), though it could be extended to other forms of I/O = as well. Though HPFF believes an MPI binding is important, what is needed is an = F90 (or F95) binding to MPI, not an HPF/MPI binding per se. The next HPFF meeting is the 1996 kickoff January 9-12 in Houston, = with subsequent meetings in March and May in Dallas to complete the HPF-= 2/HPF 2000 document. Jerry --part_ACC100E400004F1E00000001 Content-Type: Text/Plain; charset=US-ASCII Content-Disposition: Inline ========================================== Jerrold L. Wagener jwagener@ionet.net 6 E 5th, Suite 308 918-660-3978 Tulsa, OK 74103 918-742-2566 --part_ACC100E400004F1E00000001--