April 2019 - mpi-forum - mailman3.cels.anl.gov

Re: [Mpi-forum] Datatypes for all-to-all collectives on intercomms
by Michael Knobloch 30 Apr '19

30 Apr '19

Dan, thanks, that was my understanding as well, as otherwise it would make little sense. Would it make sense to clarify that in the standard? -Michael On 30.04.19 12:33, HOLMES Daniel wrote: > Hi Michael, > > Personally, I would read and summarise the current MPI text about this > as these two restrictions: > > The type map described by {sendcount, sendtype} in A must be consistent > with the type map described by {recvcount, recvtype} in B, e.g. {4, > MPI_INTEGER} in A and {1, strided_4_integers}. > > and, at the same time, > > The type map described by {sendcount, sendtype} in B must be consistent > with the type map described by {recvcount, recvtype} in A, e.g. {4, > MPI_DOUBLE} in B and {1, strided_4_doubles}. > > For an intra-communicator, or for processes in an inter-communicator > with dual-membership (erroneous but somehow supported according to > section 6.6.2), one of the send/recv communications for each process is > back to that same process. Thus, these two statements constrain the > operation to have only one type map, which is then applicable to both > directions at all processes. That type map could be specified using > different but consistent {count, type} parameters at each process. > > So, in short, yes. > > Cheers, > Dan. > — > Dr Daniel Holmes PhD > Applications Consultant in HPC Research > d.holmes(a)epcc.ed.ac.uk <mailto:[email protected]> > Phone: +44 (0) 131 651 3465 > Mobile: +44 (0) 7940 524 088 > Address: Room 2.09, Bayes Centre, 47 Potterrow, Central Area, Edinburgh, > EH8 9BT > — > The University of Edinburgh is a charitable body, registered in > Scotland, with registration number SC005336. > — > >> On 30 Apr 2019, at 10:07, Michael Knobloch via mpi-forum >> <mpi-forum(a)lists.mpi-forum.org <mailto:[email protected]>> >> wrote: >> >> Hi all, >> >> when looking at the semantics of all-to-all collectives on intercomms I >> came to a point that isn't entirely clear from the standard: Am I'm >> allowed to use different datatypes for each communication direction, >> i.e. send MPI_INTEGERS from group A to group B and MPI_DOUBLES from B >> to A? >> >> The standard says (page 168, line 38 for MPI_Alltoall, similar >> statements for the other all-to-all collectives): The type signature >> associated with sendcount, sendtype, at a process must be equal to >> the type signature associated with recvcount, recvtype at any other >> process. This implies that the amount of data sent must be equal to the >> amount of data received, pairwise between every pair of processes. As >> usual, however, the type maps may be different. >> >> This however is only true for an intracommunicator as the statement only >> holds for every pair of *communicating* processes, which in the case of >> an intercommunicator is not every pair of processes (though the name >> implies a different semantics). And the advice to users (page 169, line >> 23) contradicts that by saying that the counts may be different, but >> says nothing about the datatypes. >> >> -Michael >> -- >> Michael Knobloch >> Juelich Supercomputing Centre >> Institute for Advanced Simulation >> Forschungszentrum Juelich GmbH >> 52425 Juelich, Germany >> >> Phone: +49-2461-61-3546 >> Fax: +49-2461-61-6656 >> E-mail: m.knobloch(a)fz-juelich.de <mailto:[email protected]> >> WWW: http://www.fz-juelich.de/jsc/ >> >> _______________________________________________ >> mpi-forum mailing list >> mpi-forum(a)lists.mpi-forum.org >> https://lists.mpi-forum.org/mailman/listinfo/mpi-forum > -- Michael Knobloch Juelich Supercomputing Centre Institute for Advanced Simulation Forschungszentrum Juelich GmbH 52425 Juelich, Germany Phone: +49-2461-61-3546 Fax: +49-2461-61-6656 E-mail: m.knobloch(a)fz-juelich.de WWW: http://www.fz-juelich.de/jsc/

1 0

Re: [Mpi-forum] Datatypes for all-to-all collectives on intercomms
by HOLMES Daniel 30 Apr '19

30 Apr '19

Hi Michael, Personally, I would read and summarise the current MPI text about this as these two restrictions: The type map described by {sendcount, sendtype} in A must be consistent with the type map described by {recvcount, recvtype} in B, e.g. {4, MPI_INTEGER} in A and {1, strided_4_integers}. and, at the same time, The type map described by {sendcount, sendtype} in B must be consistent with the type map described by {recvcount, recvtype} in A, e.g. {4, MPI_DOUBLE} in B and {1, strided_4_doubles}. For an intra-communicator, or for processes in an inter-communicator with dual-membership (erroneous but somehow supported according to section 6.6.2), one of the send/recv communications for each process is back to that same process. Thus, these two statements constrain the operation to have only one type map, which is then applicable to both directions at all processes. That type map could be specified using different but consistent {count, type} parameters at each process. So, in short, yes. Cheers, Dan. — Dr Daniel Holmes PhD Applications Consultant in HPC Research d.holmes(a)epcc.ed.ac.uk<mailto:[email protected]> Phone: +44 (0) 131 651 3465 Mobile: +44 (0) 7940 524 088 Address: Room 2.09, Bayes Centre, 47 Potterrow, Central Area, Edinburgh, EH8 9BT — The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336. — On 30 Apr 2019, at 10:07, Michael Knobloch via mpi-forum <mpi-forum(a)lists.mpi-forum.org<mailto:[email protected]>> wrote: Hi all, when looking at the semantics of all-to-all collectives on intercomms I came to a point that isn't entirely clear from the standard: Am I'm allowed to use different datatypes for each communication direction, i.e. send MPI_INTEGERS from group A to group B and MPI_DOUBLES from B to A? The standard says (page 168, line 38 for MPI_Alltoall, similar statements for the other all-to-all collectives): The type signature associated with sendcount, sendtype, at a process must be equal to the type signature associated with recvcount, recvtype at any other process. This implies that the amount of data sent must be equal to the amount of data received, pairwise between every pair of processes. As usual, however, the type maps may be different. This however is only true for an intracommunicator as the statement only holds for every pair of *communicating* processes, which in the case of an intercommunicator is not every pair of processes (though the name implies a different semantics). And the advice to users (page 169, line 23) contradicts that by saying that the counts may be different, but says nothing about the datatypes. -Michael -- Michael Knobloch Juelich Supercomputing Centre Institute for Advanced Simulation Forschungszentrum Juelich GmbH 52425 Juelich, Germany Phone: +49-2461-61-3546 Fax: +49-2461-61-6656 E-mail: m.knobloch(a)fz-juelich.de<mailto:[email protected]> WWW: http://www.fz-juelich.de/jsc/ _______________________________________________ mpi-forum mailing list mpi-forum(a)lists.mpi-forum.org https://lists.mpi-forum.org/mailman/listinfo/mpi-forum

1 0

Datatypes for all-to-all collectives on intercomms
by Michael Knobloch 30 Apr '19

30 Apr '19

Hi all, when looking at the semantics of all-to-all collectives on intercomms I came to a point that isn't entirely clear from the standard: Am I'm allowed to use different datatypes for each communication direction, i.e. send MPI_INTEGERS from group A to group B and MPI_DOUBLES from B to A? The standard says (page 168, line 38 for MPI_Alltoall, similar statements for the other all-to-all collectives): The type signature associated with sendcount, sendtype, at a process must be equal to the type signature associated with recvcount, recvtype at any other process. This implies that the amount of data sent must be equal to the amount of data received, pairwise between every pair of processes. As usual, however, the type maps may be different. This however is only true for an intracommunicator as the statement only holds for every pair of *communicating* processes, which in the case of an intercommunicator is not every pair of processes (though the name implies a different semantics). And the advice to users (page 169, line 23) contradicts that by saying that the counts may be different, but says nothing about the datatypes. -Michael -- Michael Knobloch Juelich Supercomputing Centre Institute for Advanced Simulation Forschungszentrum Juelich GmbH 52425 Juelich, Germany Phone: +49-2461-61-3546 Fax: +49-2461-61-6656 E-mail: m.knobloch(a)fz-juelich.de WWW: http://www.fz-juelich.de/jsc/

1 0

Re: [Mpi-forum] Reminder: MPI Forum Virtual Meeting / Topologies
by Rolf Rabenseifner 24 Apr '19

24 Apr '19

For the first part of the upcoming virtual meeting about topologies, here the relevant links: pdf: https://github.com/mpi-forum/mpi-issues/files/3112266/mpi32-report-topol-20… Issue #120: https://github.com/mpi-forum/mpi-issues/issues/120 Pull request PR98: https://github.com/mpi-forum/mpi-standard/pull/98 Best regards Rolf ----- Original Message ----- > From: "Main MPI Forum mailing list" <mpi-forum(a)lists.mpi-forum.org> > To: "Main MPI Forum mailing list" <mpi-forum(a)lists.mpi-forum.org> > Cc: "Martin Schulz" <schulzm(a)in.tum.de> > Sent: Wednesday, April 24, 2019 10:49:07 AM > Subject: [Mpi-forum] Reminder: MPI Forum Virtual Meeting > Hi all, > > Just a reminder - the next MPI Forum meeting will be today, Wednesday 4/24 at > 10am CT. The dial-in information is linked from > > [ https://www.mpi-forum.org/meetings/ | https://www.mpi-forum.org/meetings/ ] > > The first hour will be on Topologies, the second on Sessions. > > As a reminder: we had to change the Webex Link a few weeks ago due to spammers - > if you have an old one bookmarked for this series, it won’t work anymore. You > will need to make sure you have requested access to the MPI Forum GitHub > repository before the meeting time in order to view the dial-in information. If > you haven't already done so, please email Wesley your GitHub user ID ASAP > (off-list). > > Thanks, > > Martin > > — > Prof. Dr. Martin Schulz, Chair of Computer Architecture and Parallel Systems > Department of Informatics, TU-Munich, Boltzmannstraße 3, D-85748 Garching > Member of the Board of Directors at the Leibniz Supercomputing Centre (LRZ) > [ mailto:[email protected] | Email: schulzm(a)in.tum.de ] > > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > https://lists.mpi-forum.org/mailman/listinfo/mpi-forum -- Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner(a)hlrs.de . High Performance Computing Center (HLRS) . phone ++49(0)711/685-65530 . University of Stuttgart . . . . . . . . .. fax ++49(0)711 / 685-65832 . Head of Dpmt Parallel Computing . . . www.hlrs.de/people/rabenseifner . Nobelstr. 19, D-70550 Stuttgart, Germany . . . . (Office: Room 1.307) .

1 0

Reminder: MPI Forum Virtual Meeting
by Martin Schulz 24 Apr '19

24 Apr '19

Hi all, Just a reminder - the next MPI Forum meeting will be today, Wednesday 4/24 at 10am CT. The dial-in information is linked from https://www.mpi-forum.org/meetings/ <https://www.mpi-forum.org/meetings/> The first hour will be on Topologies, the second on Sessions. As a reminder: we had to change the Webex Link a few weeks ago due to spammers - if you have an old one bookmarked for this series, it won’t work anymore. You will need to make sure you have requested access to the MPI Forum GitHub repository before the meeting time in order to view the dial-in information. If you haven't already done so, please email Wesley your GitHub user ID ASAP (off-list). Thanks, Martin — Prof. Dr. Martin Schulz, Chair of Computer Architecture and Parallel Systems Department of Informatics, TU-Munich, Boltzmannstraße 3, D-85748 Garching Member of the Board of Directors at the Leibniz Supercomputing Centre (LRZ) Email: schulzm(a)in.tum.de

1 0

Virtual Meeting Tomorrow - Finepoints
by Wesley Bland 16 Apr '19

16 Apr '19

Hi all, Just a reminder - tomorrow, Wednesday 4/17 will be our next virtual meeting for the MPI forum at 10am CT. The dial-in information (along with links to determine the time in your time zone and an iCal link to which you can subscribe) is linked from: https://www.mpi-forum.org/meetings/ <https://www.mpi-forum.org/meetings/> You will need to make sure you have requested access to the MPI Forum GitHub repository before the meeting time in order to view the dial-in information. If you haven't already done so, please email me your GitHub user ID ASAP (off-list). Tomorrow, we will have one topic and the call is scheduled for one hour: Finepoints. Originally there was also a discussion scheduled for the function pointers proposal, but that discussion is not ready yet and will be postponed. Thanks, Wesley

1 0

EuroMPI 2019, Zurich, Switzerland / Extended deadline / Call for papers
by Rolf Rabenseifner 12 Apr '19

12 Apr '19

EuroMPI 2019 Call for Papers ETH, Zurich, Switzerland, September 10-13, 2019. Extended deadline: May 6, 2019 The EuroMPI conference is since 1994 the preeminent meeting for users, developers and researchers to interact and discuss new developments and applications of message-passing parallel computing, in particular in and related to the Message Passing Interface (MPI). This includes parallel programming interfaces, libraries and langauges, architectures, networks, algorithms, tools, applications, and High Performance Computing with particular focus on quality, portability, performance and scalability. EuroMPI 2019 will continue to focus on not just MPI, but also extensions or alternative interfaces for high-performance homogeneous/heterogeneous/hybrid systems, benchmarks, tools, parallel I/O, fault tolerance, and parallel applications using MPI and other interfaces. Through the presentation of contributed papers, posters and invited talks, the meeting will provide ample opportunities for attendees to interact and share ideas and experiences to contribute to the improvement and furthering of message-passing and related parallel programming paradigms. We invite high-quality, full paper submissions on all topics related to message-passing parallel programming with MPI and related or competing models. Topics of interest include, but are not limited to: * Implementation Issues and Algorithms: Efficient and scalable implementations of message-passing constructs. * Architectures and systems: Towards exascale computing, efficient use of accelerators and other features of modern, large-scale systems, hardware-software interaction. * Programming models and paradigms for large-scale, parallel, distributed memory systems, hiearchical and hybrid models, partitioned global address space (PGAS) models. * Extensions to and shortcomings of MPI, alternative interfaces and solutions. * New, parallel (MPI-)I/O mechanisms optimizations. * Hybrid and heterogeneous programming with MPI and other interfaces. * Message passing interface support for data-intensive parallel applications. * Fault tolerance in message-passing implementations and systems. * MPI parallel programming in clouds and non-dedicated systems. * Applications and Performance * Performance evaluation for MPI and MPI-based applications. * Automatic performance tuning of applications and implementations. * Verification of message-passing applications and protocols. * Applications using message passing, in particular in computational science and scientific computing. * Parallel algorithms in the message-passing paradigm. Papers accepted for presentation as full papers will be published by ACM and appear in separate proceedings the ACM Digital Library. We also invite submissions of short papers on the same topics that are suitable for poster presentation at the conference. Papers that are not accepted for presentation as full papers may be recommended for poster presentation. Poster papers will be published informally but are not intended to appear in the regular ACM Digital Library proceedings. Papers will be reviewed for quality, originality, clarity, correctness, technical strength, and relevance to the topics of the meeting. Each paper will receive at least 3 reviews from members of the program committee. Reviews are blind (but not double-blind), the names of the authors will be seen by the program committee. Among the papers selected for presentation, a few of the best papers will, as has been the tradition for the last couple of years, be selected by the programme comittee and recommended for publication in a EuroMPI 2019 Special Issue of the Parallel Computing journal, see https://www.journals.elsevier.com/parallel-computing/. These papers will go through a separate round of reviews, and are intended as extended, more polished archival versions of the conference versions. Important Dates =============== Submission server opens: January 14th, 2019 Full paper Submission: May 6th, 2019 (AOE) (extended from April 15th) Notification: July 1st, 2019 Camera-ready: August 5th, 2019 Conference dates: September 11th-13th, 2019 See http://eurompi19.inf.ethz.ch/call for updates. Submission Guidelines ===================== To be considered, papers must be formatted according to the "sigconf" style in the ACM 2017 Template and up to 10 pages in length (shorter papers are welcome) in pdf-format. The recommended length of poster papers is two pages. Submissions (regular papers and posters) will be handled via easychair. The page limit includes figures, tables, and appendices, but does not include references, for which there is no page limit. Margins and font sizes should not be modified. Authors should submit their work through the Euro MPI 2018 Submission Site at easychair: https://easychair.org/conferences/?conf=eurompi2019 . Accepted papers will be published in the conference proceedings by ACM-ICPS. Orgnization =========== General Chair: Torsten Hoefler, ETH Zuerich, Switzerland Program Chair: Jesper Larsson Traeff, TU Wien, Austria Poster Chairs: Dan Holmes, EPCC, Scotland Local Organization: Timo Schneider --------------------------------------------------------------------- We would appreciate if you could forward this email to interested colleagues. If you receive double postings or you want to stop my postings, then please reply with "unsubscribe mpi-forum(a)lists.mpi-forum.org" in the subject line. --------------------------------------------------------------------- ------------------------------------------------------------------- . Dr. Rolf Rabenseifner . . . . . . . . .. email rabenseifner(a)hlrs.de . High Performance Computing Center (HLRS) phone ++49(0)711/685-65530 . University of Stuttgart . . . . . . www.hlrs.de/people/rabenseifner . Head of Dpmt Parallel Computing - Training and Application Services . Nobelstr. 19, D-70569 Stuttgart, Germany . . . (Office: Room 1.307) . ------------------------------------------------------------------- .

1 0

MPI Forum Virtual Meeting Reminder
by Wesley Bland 10 Apr '19

10 Apr '19

Hi all, Just a reminder - today, Wednesday 4/10 will be our next virtual meeting for the MPI forum at 10am CT. The dial-in information is linked from https://www.mpi-forum.org/meetings/ But please be aware that we had to change the Webex Link a few weeks ago due to spammers - if you have an old one bookmarked for this series, it won’t work anymore. You will need to make sure you have requested access to the MPI Forum GitHub repository before the meeting time in order to view the dial-in information. If you haven't already done so, please email me your GitHub user ID ASAP (off-list). Today, we will have one topic for one hour: Fault Tolerance. Thanks, Wesley

1 0

Re: [Mpi-forum] Broken links for Apr, 17 Virtual Meeting.
by Artem Polyakov 03 Apr '19

03 Apr '19

I see, I missed that. Thank you ---- Best regards, Artem Y. Polyakov, PhD Senior Architect, SW Mellanox Technologies ________________________________ От: Wesley Bland <work(a)wesbland.com> Отправлено: Wednesday, April 3, 2019 6:30:58 AM Кому: Main MPI Forum mailing list Копия: Artem Polyakov Тема: Re: [Mpi-forum] Broken links for Apr, 17 Virtual Meeting. Those links include all of the meetings, not just the individual ones they're next to. On Wed, Apr 3, 2019 at 4:01 AM Artem Polyakov via mpi-forum <mpi-forum(a)lists.mpi-forum.org<mailto:[email protected]>> wrote: Hello, FYI: For me an ICAL and Google Calendar links for Apr 17 virtual meeting are actually pointing to Apr, 3rd meeting. ---- Best regards, Artem Y. Polyakov, PhD Senior Architect, SW Mellanox Technologies _______________________________________________ mpi-forum mailing list mpi-forum(a)lists.mpi-forum.org<mailto:[email protected]> https://lists.mpi-forum.org/mailman/listinfo/mpi-forum<https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Flists.mpi…>

1 0

Re: [Mpi-forum] Broken links for Apr, 17 Virtual Meeting.
by Wesley Bland 03 Apr '19

03 Apr '19

Those links include all of the meetings, not just the individual ones they're next to. On Wed, Apr 3, 2019 at 4:01 AM Artem Polyakov via mpi-forum < mpi-forum(a)lists.mpi-forum.org> wrote: > Hello, > > FYI: For me an ICAL and Google Calendar links for Apr 17 virtual meeting > are actually pointing to Apr, 3rd meeting. > > > > ---- > > Best regards, > > Artem Y. Polyakov, PhD > > Senior Architect, SW > > Mellanox Technologies > > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > https://lists.mpi-forum.org/mailman/listinfo/mpi-forum >

1 0