RMA WG 03 15 2018

Agenda

1. Collect opens, assign note taker
2. Bob C. (Cray): Follow-up on Cray put with signal #206
3. Nick (DoD): Follow-up on Wait-and-then-reset / SHMEM-IFTTT #208
4. Dave O. (Intel): SHMEM Wait/test-all/some/any #207

Attendees

Manju, Ferrol, Thomas, Bob, Dave, Jim, Megan, Naveen, Min, Tony SBU, Wasi

Actions

• Follow-up from 3/1: Get feedback/input from users on shmem_put_signal w.r.t. the type of the signal word (e.g., size_t, uint64_t, something else?) and the type of the signal operation (e.g., atomic write, atomic add).
• Present nonblocking AMOs for next week.

Minutes

• shmem_wait_until_some:

  • Wasi: Do we need to know how many elements have been updated ? The user can use until for specifically waiting on a ivar.
  • Dave: No extra overhead to tell user. Runtime already has the information.
  • Jim: To distinguish old / new completions in a loop, we need an array indicating completion; or user ensures the completed ivar is reset.

• Do we need all the wait_until_all|some|any routines ? Can we implement all using some | any ?

  • Jim: Use test routines.
  • Wasi: Does it make sense to only have any and all ?

• test_any:

  • Dave: make sense to return indexes instead of 0|1.
  • Toni: count == 0 means no match, and count > 0 means the number of matched conditions.

• test_some:

  • Toni: for a big ivars array, does count array need to be the same size ?

Questions:

• waitsome_until or wailt_until_some ?
• Is some all we need ?
• start|stride: No real usage, multi dimensional ?