RMA WG 10 18 2018 - openshmem-org/specification GitHub Wiki

Agenda

  1. Review update to put-and-notify proposal #218, (Naveen)
  2. Reading of wait/test proposal #214, (Dave Ozog)

Attendees

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

Notes

Review update to put-and-notify proposal #218, (Naveen):

  • (Naveen) The main update: addition of semantic check for overlapping the destination buffer and signal register buffer (sig_addr and dest may not be same or overlapping data objects).
  • (Jim) Data object is not defined in a precise way. Could use "multiple bytes" or "non-overlapping".
  • (Jim) Not a good idea to follow the broadcast spec descriptions.
  • (Dave) "Overlapping" is also used for communication/computation so might add "memory" with it.

Reading of wait/test proposal #214, (Dave Ozog):

  • (Dave) "restrict" keyword is not in C++ standard. Might have to remove it.
  • (Jim) Or specify its limitation.
  • (Dave) Will wait for Nick's opinion as he proposed "restrict" initially.
  • (Anshuman) Suggest to use template.
  • (Jim) SOS has used generic binding for c++.
  • (Dave) (Reading the spec).
  • (Jim) (Some minor suggestions on the text:) variable declaration outside a loop, a comma position, etc.
  • (Dave) Wait set might appear new and strange to a reader in a later paragraph.
  • (Jim) Reader should be able to read it twice to understand it.
  • (Jim) There is an inconsistency between API description and return value description of shmem_test_any regarding the return value.
  • (Dave) Will do reading in the next spec meeting.