September 2022 - mpiwg-hybridpm - mailman3.cels.anl.gov

Re: [mpiwg-hybridpm] Hybrid WG Meeting
by Jim Dinan 27 Sep '22

27 Sep '22

Hi All, Quick reminder that the Hybrid WG will meet from 10:30 - 12:00 on Wednesday, Sept 28. Please note, that we are *not* using our usual zoom meeting. This meeting will be hybrid in-person and remote using the MPI Forum's zoom session. Agenda for this meeting: (1 hour) Review proposals for integrating MPI communication with accelerators (1) MPIX_Queue as a new MPI object (Naveen) (2) MPIX_Queue as arguments to a function (Jim) (3) MPIX_Stream (Hui) (4) MPI Streams (Quincey) (15 min) Review questions captured and pro/con (15 min) Discuss approach we can use to reach a common proposal Common evaluation criteria (e.g. implementation targets, application usage models, etc.) Each speaker will have 15 minutes to present their approach. During this time, I would like to capture questions and pro/con so that we can use the final half hour for review and analysis. We will use a shared Google Document [1] to capture these notes and ask that we keep speaker interruptions to a minimum. Please open the document link below and click "Request Edit Access" if you would like to contribute to notes during the meeting. Look forward to tomorrow's discussion, ~Jim. [1] https://docs.google.com/document/d/1zXaNUI-ss_B1ptW6vc1_YTeff7plhe-3dd-4w8L… On Mon, Sep 26, 2022 at 11:03 AM Jim Dinan <james.dinan(a)gmail.com> wrote: > Hi All, > > The Hybrid WG will meet from 10:30 - 12:00 on Wednesday, Sept 28. I would > like to spend this meeting reviewing different approaches to Accelerator > stream/graph integration. At the moment, we have three approaches: > > (1) MPIX_Queue as a new MPI object (Naveen) > (2) MPIX_Queue as arguments to a function (Jim) > (3) MPIX_Stream (Hui) > > Please let me know if you have an approach that you would like to share. > The goal for this discussion will be to determine a path toward a unified > proposal. I don't expect we will eliminate any proposals during the > discussion, but would like to gather pro/con of each and additional > data/questions that we feel can help the WG narrow down the field of > proposals in future work. > > If time allows, I'd also like to touch briefly on accelerator triggered > partitioned communication with the same goal of mapping out a path to a > proposal for the MPI standard. > > In case you missed it, HACC also has a reading scheduled for this week: > https://github.com/mpi-forum/mpi-standard/pull/714 > > Thanks, > ~Jim. >

1 0

Re: [mpiwg-hybridpm] Hybrid WG Meeting
by Wes Bland 27 Sep '22

27 Sep '22

For those who are physically present in Chattanooga, we’ll have a breakout room for the WG upstairs. Also note that the 10:30am time is US Eastern (local time) instead of what we usually use (US Central). Thanks, Wes > On Sep 26, 2022, at 1:53 PM, Koziol, Quincey via mpiwg-hybridpm <mpiwg-hybridpm(a)lists.mpi-forum.org> wrote: > > Cool, thanks. Yes, I’m registered. > > Quincey > > >> On Sep 26, 2022, at 11:58 AM, Jim Dinan <james.dinan(a)gmail.com <mailto:[email protected]>> wrote: >> >> CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you can confirm the sender and know the content is safe. >> >> >> Hi Quincey, >> >> Thanks, I'll include you on the schedule. Meeting is eastern time. IIUC, you will need to register for the MPI Forum meeting in order to attend. >> >> ~Jim. >> >> On Mon, Sep 26, 2022 at 11:28 AM Koziol, Quincey <qkoziol(a)amazon.com <mailto:[email protected]>> wrote: >> Hi Jim, >> I’m working on a “stream” object for my async work also (along with graphs), so it would probably be worthwhile including in the list of approaches. Might be possible to make mine compatible with 1+ of the other ones, since my async work is designed to be hardware agnostic. >> >> What timezone is the 10:30am meeting in? >> >> Thanks, >> Quincey >> >> >>> On Sep 26, 2022, at 11:03 AM, Jim Dinan via mpiwg-hybridpm <mpiwg-hybridpm(a)lists.mpi-forum.org <mailto:[email protected]>> wrote: >>> >>> CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you can confirm the sender and know the content is safe. >>> >>> >>> Hi All, >>> >>> The Hybrid WG will meet from 10:30 - 12:00 on Wednesday, Sept 28. I would like to spend this meeting reviewing different approaches to Accelerator stream/graph integration. At the moment, we have three approaches: >>> >>> (1) MPIX_Queue as a new MPI object (Naveen) >>> (2) MPIX_Queue as arguments to a function (Jim) >>> (3) MPIX_Stream (Hui) >>> >>> Please let me know if you have an approach that you would like to share. The goal for this discussion will be to determine a path toward a unified proposal. I don't expect we will eliminate any proposals during the discussion, but would like to gather pro/con of each and additional data/questions that we feel can help the WG narrow down the field of proposals in future work. >>> >>> If time allows, I'd also like to touch briefly on accelerator triggered partitioned communication with the same goal of mapping out a path to a proposal for the MPI standard. >>> >>> In case you missed it, HACC also has a reading scheduled for this week: https://github.com/mpi-forum/mpi-standard/pull/714 <https://github.com/mpi-forum/mpi-standard/pull/714> >>> >>> Thanks, >>> ~Jim. >>> _______________________________________________ >>> mpiwg-hybridpm mailing list >>> mpiwg-hybridpm(a)lists.mpi-forum.org <mailto:[email protected]> >>> https://lists.mpi-forum.org/mailman/listinfo/mpiwg-hybridpm <https://lists.mpi-forum.org/mailman/listinfo/mpiwg-hybridpm> >> > > _______________________________________________ > mpiwg-hybridpm mailing list > mpiwg-hybridpm(a)lists.mpi-forum.org > https://lists.mpi-forum.org/mailman/listinfo/mpiwg-hybridpm

1 0

Re: [mpiwg-hybridpm] Hybrid WG Meeting
by Koziol, Quincey 26 Sep '22

26 Sep '22

Cool, thanks. Yes, I’m registered. Quincey On Sep 26, 2022, at 11:58 AM, Jim Dinan <james.dinan(a)gmail.com<mailto:[email protected]>> wrote: CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you can confirm the sender and know the content is safe. Hi Quincey, Thanks, I'll include you on the schedule. Meeting is eastern time. IIUC, you will need to register for the MPI Forum meeting in order to attend. ~Jim. On Mon, Sep 26, 2022 at 11:28 AM Koziol, Quincey <qkoziol(a)amazon.com<mailto:[email protected]>> wrote: Hi Jim, I’m working on a “stream” object for my async work also (along with graphs), so it would probably be worthwhile including in the list of approaches. Might be possible to make mine compatible with 1+ of the other ones, since my async work is designed to be hardware agnostic. What timezone is the 10:30am meeting in? Thanks, Quincey On Sep 26, 2022, at 11:03 AM, Jim Dinan via mpiwg-hybridpm <mpiwg-hybridpm(a)lists.mpi-forum.org<mailto:[email protected]>> wrote: CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you can confirm the sender and know the content is safe. Hi All, The Hybrid WG will meet from 10:30 - 12:00 on Wednesday, Sept 28. I would like to spend this meeting reviewing different approaches to Accelerator stream/graph integration. At the moment, we have three approaches: (1) MPIX_Queue as a new MPI object (Naveen) (2) MPIX_Queue as arguments to a function (Jim) (3) MPIX_Stream (Hui) Please let me know if you have an approach that you would like to share. The goal for this discussion will be to determine a path toward a unified proposal. I don't expect we will eliminate any proposals during the discussion, but would like to gather pro/con of each and additional data/questions that we feel can help the WG narrow down the field of proposals in future work. If time allows, I'd also like to touch briefly on accelerator triggered partitioned communication with the same goal of mapping out a path to a proposal for the MPI standard. In case you missed it, HACC also has a reading scheduled for this week: https://github.com/mpi-forum/mpi-standard/pull/714 Thanks, ~Jim. _______________________________________________ mpiwg-hybridpm mailing list mpiwg-hybridpm(a)lists.mpi-forum.org<mailto:[email protected]> https://lists.mpi-forum.org/mailman/listinfo/mpiwg-hybridpm

1 0

Re: [mpiwg-hybridpm] Hybrid WG Meeting
by Jim Dinan 26 Sep '22

26 Sep '22

Hi Quincey, Thanks, I'll include you on the schedule. Meeting is eastern time. IIUC, you will need to register for the MPI Forum meeting in order to attend. ~Jim. On Mon, Sep 26, 2022 at 11:28 AM Koziol, Quincey <qkoziol(a)amazon.com> wrote: > Hi Jim, > I’m working on a “stream” object for my async work also (along with > graphs), so it would probably be worthwhile including in the list of > approaches. Might be possible to make mine compatible with 1+ of the other > ones, since my async work is designed to be hardware agnostic. > > What timezone is the 10:30am meeting in? > > Thanks, > Quincey > > > On Sep 26, 2022, at 11:03 AM, Jim Dinan via mpiwg-hybridpm < > mpiwg-hybridpm(a)lists.mpi-forum.org> wrote: > > *CAUTION*: This email originated from outside of the organization. Do not > click links or open attachments unless you can confirm the sender and know > the content is safe. > > Hi All, > > The Hybrid WG will meet from 10:30 - 12:00 on Wednesday, Sept 28. I would > like to spend this meeting reviewing different approaches to Accelerator > stream/graph integration. At the moment, we have three approaches: > > (1) MPIX_Queue as a new MPI object (Naveen) > (2) MPIX_Queue as arguments to a function (Jim) > (3) MPIX_Stream (Hui) > > Please let me know if you have an approach that you would like to share. > The goal for this discussion will be to determine a path toward a unified > proposal. I don't expect we will eliminate any proposals during the > discussion, but would like to gather pro/con of each and additional > data/questions that we feel can help the WG narrow down the field of > proposals in future work. > > If time allows, I'd also like to touch briefly on accelerator triggered > partitioned communication with the same goal of mapping out a path to a > proposal for the MPI standard. > > In case you missed it, HACC also has a reading scheduled for this week: > https://github.com/mpi-forum/mpi-standard/pull/714 > > Thanks, > ~Jim. > _______________________________________________ > mpiwg-hybridpm mailing list > mpiwg-hybridpm(a)lists.mpi-forum.org > https://lists.mpi-forum.org/mailman/listinfo/mpiwg-hybridpm > > >

1 0

Re: [mpiwg-hybridpm] Hybrid WG Meeting
by Koziol, Quincey 26 Sep '22

26 Sep '22

Hi Jim, I’m working on a “stream” object for my async work also (along with graphs), so it would probably be worthwhile including in the list of approaches. Might be possible to make mine compatible with 1+ of the other ones, since my async work is designed to be hardware agnostic. What timezone is the 10:30am meeting in? Thanks, Quincey On Sep 26, 2022, at 11:03 AM, Jim Dinan via mpiwg-hybridpm <mpiwg-hybridpm(a)lists.mpi-forum.org<mailto:[email protected]>> wrote: CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you can confirm the sender and know the content is safe. Hi All, The Hybrid WG will meet from 10:30 - 12:00 on Wednesday, Sept 28. I would like to spend this meeting reviewing different approaches to Accelerator stream/graph integration. At the moment, we have three approaches: (1) MPIX_Queue as a new MPI object (Naveen) (2) MPIX_Queue as arguments to a function (Jim) (3) MPIX_Stream (Hui) Please let me know if you have an approach that you would like to share. The goal for this discussion will be to determine a path toward a unified proposal. I don't expect we will eliminate any proposals during the discussion, but would like to gather pro/con of each and additional data/questions that we feel can help the WG narrow down the field of proposals in future work. If time allows, I'd also like to touch briefly on accelerator triggered partitioned communication with the same goal of mapping out a path to a proposal for the MPI standard. In case you missed it, HACC also has a reading scheduled for this week: https://github.com/mpi-forum/mpi-standard/pull/714 Thanks, ~Jim. _______________________________________________ mpiwg-hybridpm mailing list mpiwg-hybridpm(a)lists.mpi-forum.org<mailto:[email protected]> https://lists.mpi-forum.org/mailman/listinfo/mpiwg-hybridpm

1 0

Hybrid WG Meeting
by Jim Dinan 26 Sep '22

26 Sep '22

Hi All, The Hybrid WG will meet from 10:30 - 12:00 on Wednesday, Sept 28. I would like to spend this meeting reviewing different approaches to Accelerator stream/graph integration. At the moment, we have three approaches: (1) MPIX_Queue as a new MPI object (Naveen) (2) MPIX_Queue as arguments to a function (Jim) (3) MPIX_Stream (Hui) Please let me know if you have an approach that you would like to share. The goal for this discussion will be to determine a path toward a unified proposal. I don't expect we will eliminate any proposals during the discussion, but would like to gather pro/con of each and additional data/questions that we feel can help the WG narrow down the field of proposals in future work. If time allows, I'd also like to touch briefly on accelerator triggered partitioned communication with the same goal of mapping out a path to a proposal for the MPI standard. In case you missed it, HACC also has a reading scheduled for this week: https://github.com/mpi-forum/mpi-standard/pull/714 Thanks, ~Jim.

1 0

Memory Allocator Info
by Jim Dinan 08 Sep '22

08 Sep '22

Hi All, I posted an initial draft of the memory allocator info proposal. I made the assert changes that we discussed yesterday. After reflecting on the semantics of info assertions, I felt that the two info key approach would be cleaner, so that's how I wrote it up. Look forward to your feedback! https://github.com/mpiwg-hybrid/mpi-standard/pull/2 Cheers, ~Jim.

1 0

Meeting Today
by Jim Dinan 07 Sep '22

07 Sep '22

Hi All, We will meet at the regular time today to discuss the memory allocation kinds info proposal and any other topics that folks would like to discuss. Best, ~Jim.

1 0