[Mpi-forum] Persistent Collective Communication Reading Request for February 2018 Meeting

Anthony Skjellum skjellum at gmail.com
Fri Nov 17 15:56:50 CST 2017


Please schedule a reading for issue #25 (pull request #29).

https://github.com/mpi-forum/mpi-issues/issues/25
<https://github.com/mpi-forum/mpi-issues/issues/25>
Adding Persistent Collective Communication · Issue #25 ...
<https://github.com/mpi-forum/mpi-issues/issues/25>
github.com
This ticket proposes non-blocking persistent collective operations be added
to the standard. The associated PR is at: mpi-forum/mpi-standard#29

Useful links:
https://github.com/mpi-forum/mpi-issues/issues/25
https://github.com/mpi-forum/mpi-standard/pull/29
https://github.com/mpi-forum/mpi-standard/files/1483740/mpi32-report.pdf

Summary of changes:

Point to point chapter
-----------------------------
3 .9 Clarification that persistent collective requests cannot be freed
while active.

Collective Communication Chapter
----------------------------------------------
5.13  Added Persistent collective communication functions
5.13.1 -- Added section to discuss Info Keys related to Persistent
Collectibrs
5.13.2-5.13.13 -- Added one-to-one a persistent collective for every
corresponding nonblocking collective

Process Topologies chapter
------------------------------------
Added Section 7.8 -- Persistent collective communication for communication
on process topologies;
added one-to-one a function for every nonblocking topological collective
operation

As noted previously, we have a working functional implementation (libPNBC)
as well.   This is the implementation that matches the proposal.

Please note that this will be a full reading.  We went over the material in
the September meeting, but determined we needed to do a new reading in
December after making clarifications on several points raised in Chicago.

In particular, we explain the semantics of MPI_Request_free for collective
requests. And, we explain the life-cycle of persistent collectives requests
as being valid for zero-or-more uses after their respective _Init.

Thank you,
Tony Skjellum, Puri Bangalore, and Dan Holmes










-- 
Anthony Skjellum, PhD
skjellum at gmail.com
Cell: +1-205-807-4968 <(205)%20807-4968>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mpi-forum.org/pipermail/mpi-forum/attachments/20171117/742e6fd3/attachment.html>


More information about the mpi-forum mailing list