April 2026 - mpiwg-hybridpm - mailman3.cels.anl.gov

Re: [mpiwg-hybridpm] HACC WG Meeting this week: stream-triggered operations
by Patrick Bridges 15 Apr '26

15 Apr '26

Thanks, Joseph. Here is the comment describing the state of this proposal that highlights just some of the issued to address. This PR is to the hybrid working group repository’s fork of the standard. I’ve also directly attached the relevant chapter cas a latex error I recently introduced and just fixed prevented the previous version from actually including the new queued operations chapter. This PR is an initial draft of a new chapter on MPI queues for supporting stream-triggered operations through a "queued communication" interface. The goal of this PR is to begin discussions on this topic. The current draft is not complete. Current issues include: * It includes text in the new chapter (chap-streams.tex) that will eventually need to be moved to other chapters. The idea is to keep these changes in the stream triggering chapter while we work out semantic details for the PR, and then move them to the appropriate other chapters. * It introduces a new //matched// semantic term for persistent operations that will need go into the semantic terms chapter. * It will include a new state for persistent operations (enqueued) which is neither active nor inactive that will need to go in to persistent operations chapter * It includes an MPI_Match operation that impacts the semantics of persistent operations that will need to go in persistent operations chapter. * It does not yet include discussion or a definition of a needed new state for persistent requests (//enqueued// is neither inactive not active) * It does not include changes to appendices for local/non-local for the new functions. * It needs to carefully define in MPI semantic terms what a successful MPI_Match on a persistent collective means. Intuitively, it means that all participants in the collective have completed initialization of the collective so that we know the final buffers into which data will be placed, but the MPI semantic terms around persistent collectives are complex and need to be dealt with carefully. On Apr 14, 2026, at 1:34 PM, Joseph Schuchart via mpiwg-hybridpm <mpiwg-hybridpm(a)lists.mpi-forum.org> wrote: [EXTERNAL] This Message Is From an External Sender This message came from outside your organization. Dear all, We'll meet this week Wednesday 10am EST to discuss the draft for Patrick's proposal on stream-triggered MPI operations [1]. The connection info is available at https://urldefense.us/v3/__https://github.com/mpiwg-hybrid/mpi-standard/wik…. Let me know if you want to join but cannot access the page. Cheers Joseph [1] https://urldefense.us/v3/__https://github.com/mpiwg-hybrid/mpi-standard/pul… _______________________________________________ mpiwg-hybridpm mailing list mpiwg-hybridpm(a)lists.mpi-forum.org<mailto:[email protected]> https://urldefense.us/v3/__https://lists.mpi-forum.org/mailman/listinfo/mpi…

1 0

HACC WG Meeting this week: stream-triggered operations
by Joseph Schuchart 14 Apr '26

14 Apr '26

1 0

Re: [mpiwg-hybridpm] Joint RMA & HACC WG meeting this week
by Joseph Schuchart 31 Mar '26

31 Mar '26

Dear all, Just a friendly reminder that we are meeting at 9am US Central this Wednesday to discuss the device-side RMA proposal. This is the normal time slot of the HACC WG. A shared Zoom room is listed at https://urldefense.us/v3/__https://github.com/mpiwg-rma/mpi-standard/wiki__… and https://urldefense.us/v3/__https://github.com/mpiwg-hybrid/mpi-standard/wik… . See you tomorrow, Joseph On Wed, Mar 25, 2026, 10:49 Joseph Schuchart < joseph.schuchart(a)stonybrook.edu> wrote: > Dear all, > > We had a short discussion on the call today but decided to reschedule to > next week, April 1, 9am US Central Time. The connection info will be the > same as today, found in the wiki links below. > > Topics we would like to discuss: > > * How can application specify devices in MPI? > * Introduction of a query mechanism for device IDs through MPI_INFO_ENV. > * ABI issues when supporting device-side MPI function calls. > * Portability issues at compile-time and runtime between different > implementations supporting device bindings. > * Memory space specification for allocation functions (see > https://urldefense.us/v3/__https://github.com/mpi-forum/mpi-issues/issues/1… ) > > Cheers > Joseph > >

1 0