RMA WG 10 11 2018 - openshmem-org/specification GitHub Wiki

Agenda

  1. Put-with-signal Proposal Update, 10 min. (Naveen Ravichandrasekaran)
  2. Reading of wait/test proposal #214, 40 min. (Dave Ozog)

Attendees

  • Jim, Dave, Wasiur (Intel)
  • Manju (Mellanox)
  • Naveen, Bob (Cray)
  • Min, Huansong (ANL)
  • Swaroop (ORNL)
  • Anshuman (Nvidia)

Naveen’s put-with-signal proposal:

  • (Naveen) Several main changes: (1) removed strict qualifier for sig_addr, (2) add the API summary to the API description, (3) change from broadcast to a more useful case in example code.
  • (Jim) Suggestions for the example code: (1) switch the if-else conditions; (2) add synchronization operation afterwards (or some other ways).
  • (Naveen) Do we need to specify that target buffer and signal cannot overlap?
  • (Min) We may propose some top-level change to define the overlapping of buffers.
  • (Naveen) This should be handled separately.
  • (Manju & Naveen) Even in overlapping scenarios data and signal can still be sent in one package.
  • (Anshuman) This API has two symmetric objects which can be an argument for adding special statements.
  • (Jim) It is sufficient to fix this particular API now, but later similar problem for other APIs needs to be addressed too.
  • (Naveen) After this particular fix, will create another issue for not allowing source & target, and source & signal buffers overlapping when target PE is the same.

Dave’s practice reading of wait/test proposal #214:

  • (Dave) Read the changes.
  • (Jim) nelem should < SIZE_MAX.
  • (Anshuman) Could use ssize_t.
  • (Jim) Could return a number larger than or equal to nelem.
  • (Jim) The next RMA WG meeting will be in the next week to wrap up the current issues.