[mpiwg-hybridpm] Hybrid WG Meeting
qkoziol at amazon.com
Mon Sep 26 10:15:35 CDT 2022
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?
On Sep 26, 2022, at 11:03 AM, Jim Dinan via mpiwg-hybridpm <mpiwg-hybridpm at lists.mpi-forum.org<mailto:mpiwg-hybridpm at 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.
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
mpiwg-hybridpm mailing list
mpiwg-hybridpm at lists.mpi-forum.org<mailto:mpiwg-hybridpm at lists.mpi-forum.org>
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the mpiwg-hybridpm