ISO/IEC JTC/SC22/WG5-N1715 Minutes of Meeting of ISO/IEC JTC1/SC22/WG5 Hosted by the INCITS/J3 in Las Vegas, USA Feb 10-15 February, 2008 List of Attendees: WG5: John Reid (John Reid Associates, UK) Convenor Dan Nagle (J3 chair) Reinhold Bader (Leibniz Supercomputing Centre, Germany) Ian Chivers (Rhymney Consulting, UK) Malcolm Cohen (NAG, Oxford, UK) Aleksandar Donev (J3, LLNL, USA) Michael Ingrassia (J3, Sun Microsystems Inc., USA) Bill Long (J3, Cray Inc., USA) Jeanne Martin (J3, USA) Toon Moene (J3, Gnu Fortran, Netherlands) Steve Morgan (University of Liverpool, UK) Secretary David Muxworthy (BSI, UK) Craig Rasmussen (Los Alamos National Lab, USA) Jane Sleightholme (Kings College London, UK) Van Snyder (J3, Caltech/JPL, USA) Masayuki Takata (Edogawa University, Japan) Jerry Wagener (ex-J3 chair) John Wallin (George Mason University, USA) Stan Whitlock (Intel Inc., USA) Jim Xia (J3/IBM, Canada) No apologies for absence were received. 1. Opening of the Meeting The meeting opened at 0800, Sunday 10th Feb 2008. 2. Opening Business 2.1 Introductory Remarks from the Convenor John Reid welcomed everyone to the meeting and, in particular Reinhold Bader from Germany for his first meeting. 2.2 Welcome from the host Jim Xia welcomed everyone to Las Vegas. 2.3 Local arrangements A trip to a local park (Springs Reserve) and a dinner sponsored by J3 had been arranged. 2.4 Appointments for this meeting The drafting committee will consist of Reinhold Bader, David Muxworthy, Toon Moene, Van Snyder, Masayuki Takata, and Jim Xia with David Muxworthy as chairman. Steve Morgan would act as secretary. 2.5 Adoption of the agenda (N1699) 6a. Interoperability TR was added. There would be short tutorials by John Wallin and Dan Nagle. Otherwise the agenda was adopted as published. 2.6 Approval of the minutes of the London (N1691) Malcolm Cohen noted that he had not said he preferred meetings to start on Sunday. The minutes would be altered to reflect this and would be published as a new document (N1717). Otherwise the minutes were approved. 3. Matters arising from the minutes John Reid noted that he had not pursued the standards for special functions. Otherwise, there were no matters arising not covered by subsequent agenda items. 4. Status of the London Resolutions (N1690) Nothing to report. 5. Reports 5.1 SC22 Matters John Reid had already sent a report (N1703) to WG5. He reviewed its contents. There are no problems with producing further corrigenda (WG5 ballot underway, J3 ballot soon to finish). The TRs on Exceptions and Allocatables will be withdrawn at the SC22 level. Regarding the Interoperability TR, John Reid noted that he should not have applied for formal approval at the 2006 meeting of SC22 since the clock starts ticking immediately. An extension gives 18 months from Sept 2007. The first draft will be ready Feb/March 2009 (Bill Long Editor). A request for a delay of 10 months for the main standard was accepted by SC22. 5.2 National Activity Reports (Heads of delegations) N1716 US National Activity Report (Snyder) N1707 Japan National Activity Report (Takata) Germany - group has not met for some years, but hopefully will meet this year. N1707 UK National Activity Report (Muxworthy) N1713 Canada National Activity Report - only one member but seeking interest from academics. 5.3 Report from Primary development body (INCITS/J3 Chair) Dan Nagle noted that one set of interpretations had been balloted. Malcolm Cohen noted that he had produced document N1709 (N1705 with CAF removed) in a slightly changed format that reduced the total number of pages. He asked members to look at the document to see which format they prefer. J3/08-018 consists of Interp pages in a loose leaf form that can be inserted into the standard. It contains experimental use of underwave and strikeout. Comments on this were requested. J3/08-121 includes draft edits that had not yet been approved. 5.4 Reports from Other Development Bodies None. 5.5 Liaison Reports IEEE 754 revision - no comment IFIP/WG2.5: A meeting at Uppsala discussed interval arithmetic and floating point arithmetic (IEEE 754). There were no requests to WG5. OpenMP: No report (Matthijs van Waveren not present) OWG-V: Dan Nagle reported that very good progress was being made. There was a new draft available covering three main points: safety, security, and computer reliability. They are looking at floating point and parallel programming issues. C: The committee is producing new revision which will standardise p-threads. The UPC group was unsuccessful in its approach to get UPC into the C standard. They were pursuing a fast track separate standard. Aleksandar Donev requested liaison with the UPC committee - Dan Nagle may take this on. 6. Review the current working draft (J3/08-007) for the revision of Fortran 2003 (Tutorials on the need for parallel programming were given by Dan Nagle and John Wallin. Slides are available on the J3 server.) John Reid noted that one of the main aims of this meeting was to achieve consensus on the coarray issue. He gave a brief history to set the scene. At the 2005 Delft meeting WG5 had decided that coarrays (extended from the original proposals) should be part of the main standard. Subsequently at the Fairfax meeting in 2006 coarrays were made a category A, first priority item. In London 2007 there were three abstentions in the country vote and subsequently John Reid therefore invited the three heads of delegations to indicate what they required in order to vote yes on the standard. Two papers from David Muxworthy & John Reid (N1710) and from Jim Xia & John Reid (N1712) and comments from Masayuki Takata in the Japan National Report addressed this request. David Muxworthy presented a possible compromise from the UK (N1710). This suggested that CAF be put in a TR so that it and the main standard could converge on a longer timescale. There followed a long discussion. (Meeting adjourned at 1200. Reconvened at 1330) Straw vote: Would it make any difference to you if CAF were put in a TR or Part4? Yes-4 No-10 Undecided-2 Jim Xia presented his paper N1712 proposing reductions in the coarray feature. Straw vote: Take TEAM I/O out of the standard? Yes-8 No-2 Undecided 7 Straw vote: Reduce corank to 1? Yes-3 No-12 Undecided-2 Straw Vote: Assuming TEAM I/O has been removed do you favour moving coarrays to TR? Yes-8 No-9 Undecided-1 Straw Vote: Assuming TEAM I/O has been removed do you favour moving coarrays to Part 4? Yes-5 No-9 Undecided-4 (Adjourned at 1700) .................................................................. Monday 0800 John Reid again summarised the situation: He was very concerned at 3 abstentions in London so contacted heads of the three abstaining countries. Jim Xia (Canada) wanted reduction in size of coarray proposals. David Muxworthy (UK) wanted whole of coarrays in a TR. Masayuki Takata (Japan) wanted coarrays in Part 4. Following Sunday's discussions and straw votes John Reid had concluded that compromise was not possible at this stage. At this point the meeting moved to consider item J3-038 (math functions) of the Repository, N1649. Dan Nagle proposed that the work item be deleted. Straw Vote: Shall we eliminate item J3-038 (math functions) completely? Yes-1 No-6 Undecided 10 John Reid concluded that WG5 wanted the item to remain. It would be up to J3 to decide how to implement it (intrinsics or module or something else). The meeting then reverted to discussing the coarray issue again. David Muxworthy raised the issue of a possible Part 4 again. John Reid suggested that countries should caucus to discuss the issue and individual and country straw votes would be taken on Tuesday. Adjourned 0930. .................................................................. Tuesday 0800 The coarray issue was discussed again. John Reid asked all the heads of the delegations to state their positions. UK - following a caucus had decided that moving all of CAF to Part 4 was acceptable. Noted that lack of success of Parts 2 and 3 was not relevant to this case. This would satisfy Japan's concerns. Part 4 could be written to apply to both F2003 and F2008. Japan - very much against coarrays in Part 1. They feel that there may be better ways to handle parallelism. Wanted Part 4 (rather than TR) for legal reasons. Canada - would prefer to put coarrays in a TR. Netherlands - would stay with their decision in Delft to have coarrays in Part 1. Germany - perceived need for coarray functionality. Some features need further development. US position is the best way forward. At this point there was a break for a US caucus. On return US proposed a new position: keep core features of coarrays in main standard and develop a TR to cater for the other features plus possibly those suggested by public comment. John Reid then decided to take straw votes. Individual vote: Put whole of coarrays into part 1 of the standard Yes-No-Undecided: 9-8-2 Reduce coarrays in Part 1 and put the rest into a TR Yes-No-Undecided: 13-3-3 Move coarrays into a Part 4 of the standard Yes-No-Undecided: 9-6-4 Move coarrays to a TR Yes-No-Undecided: 5-10-2 Country Vote: Put whole of coarrays into part 1 of the standard US-yes, UK-No, Japan-No, Netherlands-Yes, Germany-Undecided, Canada-No (2-3-1) Reduce coarrays in Part 1 and put the rest into a TR US-yes, UK-Yes (reluctantly), Japan-No, Netherlands-Yes, Germany-Yes, Canada-Yes (5-1-0) Move coarrays into a Part 4 of the standard US-No, UK-Yes, Japan-Yes, Netherlands-Undecided, Germany-Undecided, Canada-Undecided (2-1-3) Move whole of coarrays to a TR US-No, UK-Yes, Japan-Yes, Netherlands-No, Germany-No, Canada-Yes (3-3-0) Following some clarification of the undecided country votes, John Reid declared that the best way forward (the way that would achieve the greatest consensus) would be to follow the US's revised position. John Reid asked for any objections to this and none were forthcoming. It was left to the HPC subgroup to produce a paper detailing the features of coarrays that would be included in the standard and those that would go into the TR. Straw Vote: Delete special functions (see J3/08-108r1): 0-14-4 [J3 decided later that they should become full intrinsics] Straw vote: Do you approve of the discussion part (not the edits) of J3/08-131 which gives the coarray items to be removed from Part 1. Yes-15 No-0 Undecided-4 Adjourned at 1640. .................................................................. Wednesday 0800 Discussion of N1693 (schedule). Malcolm intended to finish incorporating edits from this meeting into the draft document as soon as possible and certainly before April. There was some discussion about whether to delegate the checking to a subgroup. John Reid eventually declared that there would be a 15-day ballot. Dan, John and Malcolm would handle the final stage. There was some discussion about the logistics of interleaving the ISO ballot and J3 ballot (on main standard). Meeting adjourned. Wednesday 1330 John Reid had confirmed with ISO secretariat that the ballot would not be sent to Geneva, so there should be no delays. There followed a discussion on the logistics of conducting the ISO and J3 ballots to provide adequate time to process comments etc. A draft timetable was formed and this would appear in the resolutions (N1719). Adjourned 1355. .................................................................. Thursday 0800 John Reid had prepared a draft paper N1719 giving the schedule for the ISO and J3 public comment periods discussed yesterday. It was agreed that the Interoperability TR would be published in about a year's time. John Reid requested volunteers to act as the editor for the new Coarrays TR. Bill Long volunteered and (when the time comes) will be nominated by the US in the SC22 ballot for a new work. Bill noted that the TR would contain those items mentioned in J3/08-130 but would also clearly need to take account of public comment on the main draft. 7. Consider the Fortran defect reports (interpretations) in J3-006 There would be another WG5 ballot on a new corrigendum that John Reid intended to produce in about 2 months time (beginning of May). 8. WG5 Business & Strategic Plans The target date for F2008 will be 2010 (N1693). The next WG5 meeting will be 16-21 Nov 2008 in Tokyo, Japan. At this point John Reid expressed thanks on behalf of the WG5 members for the hospitality provided by J3 at the meeting and in particular for the organisation of the trip to the Springs Reserve and the evening meal at the Swiss Cafe. Dan Nagle noted that the main effort had been provided by Van Snyder. Meeting adjourned to subgroup at 0830. Thursday 1630 There was discussion (Bill Long) of the issue regarding optional allocatable and assumed shape dummy arguments and whether to allow the VALUE attribute. Most implementations use a null pointer to indicate a missing argument but at least one implementation doesn't. [Bill Long later said he would produce a paper on the issues and send it to the WG5 email list] Meeting was adjourned at c.1650. .................................................................. Friday 9. Closing Business 9.1 Future meetings Arrangements for the meeting in Tokyo - Nov 16-21, Tokyo, Japan, see N1720. 4 hotels - Shiba Park is the main one. Easy access from airport. Airport limousine available. Planning to have excursion (bus tour in Tokyo) and optional tour before or after meeting. There will probably be an evening tour on Wednesday probable return at 2200. Dinner at 1700. The meeting will take place from 0900-1700 Sunday to Thursday and 0900-1200 Friday. The meeting room will be available from 0830. An invitation from US to meet in May 2009 (N1693) was gratefully accepted. WG5 expressed preference for a Monday 0800 to Friday 1200 meeting without any daytime excursion. 9.2 Any other business New TR on coarrays would be published c.2Q 2011 (Stan Whitlock raised the issue). 10. Adoption of Resolutions (N1714) Votes on LV3: Individual Vote: 13-1-4 Country Vote: US-Y UK-A Japan-N Canada-Y Netherlands-Y Germany-Y (4-1-1) Resolutions LV1-LV2 & LV4-LV7 were passed with unanimous consent and resolutions LV8-LV9 were passed by unanimous acclaim. 11. Adjournment John Reid thanked the local hosts (J3) again for their excellent support during the week. The meeting adjourned at c. 1000.