ISO/IEC JTC1/SC22/WG5 N2068 Minutes of Meeting of ISO/IEC JTC1/SC22/WG5 Hosted by BSI, the UK Member Body for JTC1/SC22 in London, UK August 3-7, 2015 List of Participants: John Reid (JKR Associates, UK) convenor Dan Nagle (NCAR, USA) PL22.3 chair Reinhold Bader (Leibniz Supercomputing Centre, Germany and DIN) Daniel Chen (IBM, Canada) Tom Clune (NASA, USA) Malcolm Cohen (NAG, UK) Alessandro Fanfarillo (University of Rome, Italy) (Monday to Wednesday) Steve Lionel (Intel, USA) Bill Long (Cray, USA) Nick Maclaren (University of Cambridge, UK) Toon Moene (Gnu Fortran, Netherlands) David Muxworthy (British Standards Institution, UK) Damian Rouson (Sourcery Inc, USA) (Monday to Thursday) Bob Scollard (Nvidia Corporation, USA) Jane Sleightholme (BCS Fortran Special Group, UK) (Monday to Wednesday and Friday) Van Snyder (Caltech/JPL, USA) Masayuki Takata (Edogawa University, Japan) Notes: 1. This was a joint meeting of SC22/WG5 and INCITS/PL22.3. These minutes record only the parts of the meeting in WG5 mode. 2. WG5 papers are referenced as Nnnnn. Links to the papers are available via http://www.nag.co.uk/sc22wg5/docs.html 3. INCITS/PL22.3 is abbreviated throughout to PL22.3 and its papers are referenced as J3/15-nnn. They are available from http://www.j3-fortran.org/ 4. During the meeting TS 18508 "Additional Parallel Features in Fortran" was referenced by its informal name "Further Coarray TS" and is so recorded in these minutes. 1. Opening of the Meeting The meeting opened at 09:00 on 3rd August 2015. 2. Opening business 2.1 Introductory remarks from the Convenor The convenor welcomed participants, especially those attending WG5 for the first time, Tom Clune and Bob Scollard. He said that the principal objectives of the meeting were to construct the draft TS on further coarray facilities and to decide on new proposals for the revision of Fortran 2008. There was an open WG5 straw ballot on Fortran 2008 interpretations. 2.2 Welcome from the Host Jane Sleightholme welcomed participants on behalf of BSI and BCS. The co-host, Ian Chivers, was not present because of his wife's illness. The meeting sent best wishes for her rapid return to health. 2.3 Local arrangements Jane Sleightholme invited participants to dinner at the Salieri restaurant on the Wednesday evening. The BCS Fortran Specialist Group had contributed towards the cost of the dinner. 2.4 Appointments for this meeting The drafting committee was Reinhold Bader, Daniel Chen, Toon Moene, David Muxworthy (chair), Steve Lionel and Masayuki Takata. David Muxworthy acted as secretary and John Reid as librarian. 2.5 Adoption of the agenda [N2052] The preliminary agenda was adopted, subject to two minor amendments: at item 6 'DTS' was changed to 'TS' and at item 7 '15-006' was changed to '15-006r1'. 3. Matters arising from the minutes of the Las Vegas 2014 Meeting [N2020] There were no items not otherwise on the agenda. 4. Status of Las Vegas 2014 Resolutions [N2021] There was nothing to report. 5. Reports 5.1 SC22 Matters (Convener) John Reid had been re-appointed as convenor for the period to September 2017. He said that he would not stand for a further term. Document archive. The document archive had been re-arranged at NAG's request. Older documents (to N2019), and other WG5 pages, would remain on the NAG site but all new WG5 numbered documents would be on ISO's Livelink system. All documents would be accessible via the 'Complete Document Register'. Some members had had difficulty accessing documents without entering passwords. 5.2 National Activity Reports Canada: There was little standards activity related to Fortran. Germany: Reinhold Bader was now officially representing DIN. Japan: The Japanese Fortran Group had ten members and met once per month. Preparation for the Japanese version of Fortran 2008 continued but was progressing slowly because of lack of resources. Netherlands: Toon Moene was now an advisor to, rather than a member of, the Programming Languages Committee of NEN. UK: A report was in N2060. The main activity had been in participating in email discussion on development of the Further Coarray TS. US: The main activity was that of PL22.3. The US had voted approval in the SC22 ballot on the Further Coarray TS. There would be a US TAG meeting on the Thursday of the meeting. 5.3 Report from Primary Development Body (INCITS/PL22.3) Development had been proceeding to schedule. Some new proposals were available for consideration at this meeting. The Project Editor summarized items which had been incorporated in the latest Working Draft of the full language standard, j3/15-007r1. 5.4 Reports from other Development Bodies (Editors/Heads) Further Coarray TS: Bill Long The TS had been approved in the recent SC22 ballot and the comments from national bodies would be considered at this meeting. 5.5 Liaison Reports: NCITS/PL22.11 (C): Dan Nagle The CPLEX work, to provide parallel facilities in C, was continuing and was approaching completion. An estimated completion date was not yet available. MPI: Bill Long MPI 3.1 had been published. This was mainly a corrections update to 3.0. The plan for MPI 4.0 included deprecating mpif.h in favour of using the modules. It was planned to delete mpif.h in MPI 5.0. UPC: Damian Rouson UPC retained broad support with multiple vendors and academic groups producing compilers. However, there appeared to be no movement toward merging into the standard, which led to the technical issue that UPC must decide to "rebase" against each new C standard (see https://goo.gl/LGrKQC), which in turn raised the question of how potential future conflicts would be handled. A personal impression was that the path forward was clearer for UPC++, which had a very clear and well-worn path toward standardization. UPC++ was being developed in a "compiler-free" manner by simply defining a set of templates in a "upcxx" namespace. As is often the case with C++ growing by agglomeration, adding UPC++ to the C++ standard might be no more than changing the namespace to "std". We might do well to keep an eye on developments with UPC++. IFIP/WG2.5: Van Snyder Philip Sharp of the University of Auckland, New Zealand, had attended the recent WG2.5 meeting and had expressed interest in joining WG5 and possibly PL22.3. OpenMP: Bill Long OpenMP 4.1 had been released. It added support, compared to OpenMP 4.0 for - Allocatable enhancement - Finalization - Procedures bound by name to a type - Type extension Fortran 2003 features still missing from OpenMP 4.1: - IEEE Arithmetic issues covered in Fortran 2003 Section 14 - Parameterized derived types - The PASS attribute - Procedures bound to a type as operators - Overriding a type-bound procedure - Polymorphic entities - SELECT TYPE construct - Deferred bindings and abstract types - Controlling IEEE underflow - Another IEEE class value OpenMP 5.0 would try to finish support for Fortran 2003. There was an "interoperability" subgroup that was discussing interactions with other parallel models, but was mainly focused on shared-memory threading (C++), and not on distributed models like coarrays. WG23 (Vulnerabilities): Dan Nagle Eight new vulnerabilities had been defined, six of them related to parallelism. The Fortran Annex was being developed to incorporate these. OpenACC: Bob Scollard OpenACC now had several high profile success cases, particularly in the weather / climate community, which were largely Fortran code bases. The majority of these codes currently targeted systems with X86 host + NVIDIA Tesla GPUs. The committee was working on an OpenACC 2.5 specification, with several minor changes to the current version. The biggest item that users had requested was support for 'deep copy.' This is deferred until the next major release. Deep copy was similar to sourced allocation in Fortran. However, implementing this capability in systems with physically, logically, and virtually separated memories and address spaces was a significant challenge. Implementing deep copy in a way that was efficient, useful, and easy to understand was also a significant challenge. More information was available at http://www.openacc-standard.org/. The convenor said that from this point the following agenda items would proceed in parallel: 6. Construct the TS on Additional Parallel Features in Fortran. 7. Consider the Fortran defect reports (interpretations) in J3/15-006r1. 8. Confirm the list of deficiencies and discrepancies in Fortran 2008 to be addressed in the next revision. 9. Review of the current draft of the next revision. 10. WG5 Business and Strategic Plans [In fact no business was transacted for agenda items 7 and 9.] {agenda 8} Subgroups considered the new proposals to address deficiencies in Fortran 2008 which were listed in N2069. [PL22.3 plenary and subgroup sessions from 10:20 to 16:00] The following recommendations were made by subgroups: JoR subgroup: Approve: US-22 Purity enhancement US-23 Command line related intrinsic procedures US-25 Allow mixed kind in SIGN intrinsic US-26 POINTER arguments to SAME_TYPE_AS US-27 SELECT RANK construct UK-23 Remove unnecessary constraints on stop codes (possibly with an extension to allow a 'STOP' output message to be suppressed) UK-24 Remove anomaly in provision of error messages. Forward for further consideration: UV-03 Intrinsic assignment to a polymorphic array element without select type UV-06 TARGET for components Take no further action, for a variety of reasons: UV-01 select rank, subscript syntax UV-04 delete TBP override restrictions UV-05 delete 4 digit req for stat vars UV-07 expand FINDLOC w/ ORDERED UV-08 FINDALL intrinsic UV-09 MERGE_PTR intrinsic UV-10 AVAILABLE intrinsic UV-11 MOVE_ALLOC changes UV-12 PRESENT for nonoptional arg UV-13 FLUSH (*) for stdout. Data subgroup: Approve: US-24 Specification expressions US-28 Locality clauses in DO CONCURRENT. Take no further action: UV-02 delete TYPE(*) constraint It appeared that WG5 was intending to endorse all the items carried forward from 2014 and many of those put forward this year. Vendors present were asked for their reaction. Most were still working on Fortran 2008 and would start on Fortran 2015 only when that was completed, which could be up to three years in the future. gfortran, working in a different environment, was giving very low priority to user-defined derived-type input-output and parameterized derived types and would adopt Fortran 2015 features according to user demand. There was a straw vote on whether the enhancement to the display of the STOP code (UK-23) should be pursued. The result was: 9 yes - 1 no - 5 undecided. [PL22.3 plenary session from 16:55 to 17:00] Jane Sleightholme announced that London Underground would be on strike from 16:30 on Wednesday and for all of Thursday; she offered to advise participants on alternative travel arrangements. Buses and National Rail services would not be affected. The meeting adjourned at 17:02. ........................................................................... Tuesday {agenda 8} There was a brief presentation and straw vote on each of the proposals considered by subgroups the previous day. It was decided that taking country votes would not be appropriate. The vote would be on the concept, not necessarily on the detailed edits. In some cases there was discussion on amending the proposal but each vote was taken on the item as presented. The results were: US-22 Purity enhancement 15 - 0 - 2 US-23 Command line related intrinsic procedures 14 - 0 - 3 US-24 Specification expressions 15 - 0 - 2 US-25 Allow mixed kind in SIGN intrinsic 14 - 1 - 2 US-26 POINTER arguments to SAME_TYPE_AS 13 - 0 - 4 US-27 SELECT RANK construct 12 - 1 - 4 US-28 Locality clauses in DO CONCURRENT 12 - 1 - 4 UK-23 Remove unnecessary constraints on stop codes 16 - 0 - 1 UK-24 Remove anomaly in provision of error messages 16 - 0 - 1 UV-06 TARGET for components 3 - 4 - 10. There was a long discussion on item UV-03 Intrinsic assignment to a polymorphic array element without select type. A straw vote on whether to address the issue was: 1 - 7 - 9. {agenda 6} Paper j3/15-191 which gave partial replies to national member body comments in the SC22 ballot on the Further Coarray TS was discussed briefly but withdrawn before a vote. [PL22.3 plenary session from 11:10 to 11:50] Alessandro Fanfarillo gave a presentation on "Accelerated Keyword", written jointly with Damian Rouson, in which he discussed the effects of computer architecture changing significantly in the near future. The slides are in document N2072. This was followed by discussion. [PL22.3 subgroup sessions from 14:00 to 17:35] It was announced that documents giving draft responses to the comments in the SC22 ballot and a new draft strategic plan were available for discussion the following day. The meeting adjourned at 17:40. ........................................................................... Wednesday {agenda 8} There was a discussion on the proposed extension to UK-23 to allow suppression of output following execution of a STOP or ERROR STOP. A straw vote on whether the matter should be addressed was: 11 - 0 - 6. A vote on whether all output should be suppressed vs whether finer control was needed was: 11 - 4 - 2. [PL22.3 plenary session from 09:25 to 10:05] {agenda 6} A draft Disposition of Comments document for the SC22 ballot on TS 18508, was presented and the draft responses were explained. Minor rewordings were agreed during discussion. {agenda 10} A draft Strategic Plan for 2015 - 2018 was discussed. This showed the Further Coarray TS being completed one month earlier than previously scheduled and the full language revision five months later than previously scheduled, in July 2018. After discussion the plan remained unchanged. In 2013 Van Snyder had introduced a draft TS on Units of Measure in Fortran (ref N1969, N1970, N1977). It would be useful for his sponsors to know why WG5 had not adopted this project. The convenor agreed to provide a document with the reasons. [PL22.3 plenary and subgroup sessions from 11:00 to 17:30] Peter Crouch, chairman of the BCS Fortran Specialist Group, joined the meeting. He had played a major role in negotiating with BCS for free use of the rooms and IT facilities and for provision of refreshments. He was thanked and applauded by participants. The meeting adjourned at 17:40 and was followed by the meeting social event at the Salieri restaurant. ........................................................................... Thursday [PL22.3 plenary session from 09:00 to 09:50] {agenda 6} A revised, almost complete, draft Disposition of Comments document for the SC22 ballot on TS 18508, was presented and discussed. No amendments were proposed. {agenda 11} Dan Nagle described arrangements for the 2016 WG5 meeting in Boulder. The meeting would be in Mesa Lab and accommodation would be in the Millennium Hotel, some four miles away. There was a discussion on the best means to travel between the two and the best way to travel from Denver airport. Detailed information would be distributed in due course. Reinhold Bader asked WG5 for its preference for dates for the 2017 meeting. It was decided to hold it on June 26 to 30. Damian Rouson had been exploring possibilities for the 2018 WG5 meeting in Oakland, CA, with an old members' reunion to be held on the previous weekend. He was encouraged to continue investigations. [PL22.3 plenary, US TAG and subgroup sessions from 11:00 to 17:40] It was announced that new drafts of the Disposition of Comments document for the SC22 ballot on TS 18508, the convenor's report to SC22, a list of features to be included in the next revision and the resolutions were available for review. The meeting adjourned at 17:50. ........................................................................... Friday {agenda 6} The latest draft of the Disposition of Comments document for TS 18508 (N2066) was discussed. Members from the UK accepted the disposition of GB-1, the only remaining incomplete response. No changes were made to the document. {agenda 10} A revised Strategic Plan (N2070) was discussed and approved, the principal change being the addition of a statement that a future revision would not start before 2017. It was agreed that the committee needed to take a long-term strategic view and not continue with simply adding individual features. Also, more notice should be taken of vendors' resources. The Plan scheduled a new work item vote for the revision to take place in September to December 2016. This was queried, given the revised JTC1 Directives. The matter was not resolved at the meeting. [According to JTC1 Directive 2.3.1 (2015 version) a NWIP vote is not required for a revision.] It was noted that avoiding a vote would not affect the schedule. The Convenor's Report to SC22 (N2071) was discussed. Comments were welcomed but had to be made immediately. {agenda 8} Attention was drawn to a document (N2073) listing the new language features approved at this meeting. A draft of the resolutions (N2067) was discussed and minor edits suggested. [PL22.3 plenary session from 09:45 to 11:10] During discussion of paper j3/15-208 there was a WG5 straw vote on different possible responses to an interpretation. In the case of recursive USE within a submodule, (1) should accessing a PROTECTED item by a USE statement hide the host association?, or (2) should a submodule be allowed to USE its ancestor, PROTECTED having no effect?, or (3) should a submodule accessing its ancestor module by use association be forbidden? The vote was: 3 - 1 - 9 - 2. Also in discussion of paper j3/15-213 there was a question about nonadvancing input/output on a single unit during execution of a DO CONCURRENT construct. Should it be permitted for one iteration or not permitted at all. The vote was 3 - 8 - 4. {agenda 6} Nick Maclaren introduced paper j3/15-210 'What ordering is implied by collectives?' which was an attempt to achieve consensus on a consistency model. A straw vote on the principles involved was: 10 - 0 - 4. The principles would guide integration of TS 18508. {agenda 8} Paper j3/15-010, the PL22.3 Work Plan, was discussed and minor edits suggested. [PL22.3 plenary session from 11:40 to 11:55] 11. Closing Business 11.1 Future meetings These had been discussed and resolved the previous day (see above). 11.2 Any other business None was raised. 12. Adoption of Resolutions [N2067] Resolutions L1 to L4 and L11 to L13 were approved by unanimous acclaim. Resolutions L5 to L10 were approved by unanimous consent. During discussion of the document archive in resolution L1, members were encouraged to submit to the convenor electronic copies of any pre-1995 documents which they might possess and which are not already in the archive. 13. Adjournment The meeting adjourned at 12:10 on Friday, August 7, 2015.