From jwagener@amoco.com Thu May 18 12:50:37 1995 Received: from interlock.amoco.com by dkuug.dk with SMTP id AA26240 (5.65c8/IDA-1.4.4j for ); Fri, 19 May 1995 00:51:33 +0200 Received: by interlock.amoco.com id AA03760 (InterLock SMTP Gateway 3.0 for sc22wg5@dkuug.dk); Thu, 18 May 1995 17:51:21 -0500 Message-Id: <199505182251.AA03760@interlock.amoco.com> Received: by interlock.amoco.com (Protected-side Proxy Mail Agent-3); Thu, 18 May 1995 17:51:21 -0500 Received: by interlock.amoco.com (Protected-side Proxy Mail Agent-2); Thu, 18 May 1995 17:51:21 -0500 Received: by interlock.amoco.com (Protected-side Proxy Mail Agent-1); Thu, 18 May 1995 17:51:21 -0500 From: jwagener@amoco.com X-Openmail-Hops: 1 Date: Thu, 18 May 95 17:50:37 -0500 Subject: jlw informal report on the May 11-12 HPFF meeting To: sc22wg5@dkuug.dk X-Charset: ASCII X-Char-Esc: 29 Item Subject: Message text This was the first meeting of HPFF since the 1995 kickoff HPFF meeting following the Houston X3J3 meeting in January (many of you attended that kickoff meeting). This was a "working" meeting of the "core" HPFF group, and there were 26 individuals in attendance. One of the first orders of business was to establish the three functional subgroups that the HPFF organizers had brainstormed: group C "control flow related issues", group D "distribution related issues", and group E "external issues". The goal is to present HPF 2.0 at Supercomputing '96 (about the time that Fortran 95 should become official). Group C is to deal with things like asynchronous I/O, various kinds of task (or at least coarser-grained) parallelism, and possibly exception handling (Rob Schreiber, head). Group D is to deal with things like irregular mappings, mapping derived type components, and mapped I/O (Guy Steele, head). Group E is to deal with things like interfacing with debuggers and other tools, multi-language issues, and the HPF kernel (David Loveman, head). I volunteered for group E. The only other X3J3 member present, Henry Zongaro, volunteered for group D. All of the subgroups are to deal with CCI items pertaining to their functional areas, and all are to examine Fortran 95 relative to possibly making comments during the public review period - the next HPFF meeting is July 26-28, near the end of the expected public review period. (I gave a brief status of Fortran 95 early in the meeting, one slide of content summary and one slide of schedule, and encouraged HPFF to review F95, especially the HPF part, and make any appropriate comments.) From the various subgroup reports at the end of the meeting it would appear that at this meeting the subgroups focused mainly on CCI items and getting organized for the technical development in their areas. Many of the outstanding CCI items were resolved, and the subgroups seem to have made initial assignments relative to the new technical work. Group E (the groupies) spent about half its time completing its assigned CCI items and the other half on interoperability issues. Interoperability with C was chosen as the main discussion vehicle, partly because that's considered likely to be the most important interoperability area, and partially to help shed light on this area - HPFF is aware of the interoperability activities in WG5 and X3J3 and would like to coordinate closely with that work. To that end, I accepted the assignment of getting David Loveman and Jamie Shiers together - that is, to get to each the other's email address. Various forms of extrinsic interfaces were considered, but no clear direction emerged. Andy Meltzer was assigned the job of putting together a "first cut" for the July meeting, and all of the implementors in HPFF were encouraged to send Andy their "C interface rules". Meanwhile the groupies' discussion identified three approaches to using extrinsic interfaces for interoperability with C: (1) the "strict" approach (my choice of words), in which an F90 interface is used and any Fortran type/kind combination can be passed to exactly one (prespecified) C data type; (2) the "implicit permissive" approach, in which the extrinsic interface is essentially a C function prototype, and a Fortran type/kind combination may be prespecified to be an allowed actual argument for one or more C datatypes; and (3) the "explicit permissive" approach, which is similar to the implicit permissive one except the type/kind mapping is explicit (this requires an extrinsic interface that, effectively, contains both an F90 interface and a C function prototype). The meeting HPFF meeting schedule for the next year is as follows: July 26-28 Denver Sept 20-22 Dallas Nov 1-3 Dallas (chosen not to conflict with WG5/X3J3) Jan 9-12 Houston (this will be the 1996 kickoff) Mar 6-8 ? Jerry jwagener@amoco.com