RMA WG 06 13 2019

Agenda

1. Review put-with-signal proposal

Attendees

• Did not take roll

Open Action Items

• None

New Action Items

• None

Notes

Naveen: reviewed put with signal discussion

• Discussed adding both signal wait and signal fetch operations

Naveen: nonblocking fetching AMOs

• Minor updates provided at the reading, will make the updates and prepare for next step
• Anshuman: what is the effect of a fence on a nonblocking AMO?
  • By default NV GPU atomics are not ordered. There are several ways to enforce ordering separately from completion.
  • Would want to order the atomic updates, but not necessarily the value returned to the caller.
  • AR: Anshuman, investigate performance potential on Nvidia GPU
  • Jim: Do we have similar performance opportunities on conventional processors (ordering of atomics)?
  • Anshuman: Can fetching atomics be reordered in x86?

Dave: wait/test vector operations

• Silent approval at FTF meeting (not much feedback)
• Will draft proposal and bring to WG

Jim: Reviewed wait/test clarification proposal

• Need to add advice to implementors, otherwise this is ready to vote on at the next opportunity

Jim: Reviewed symmetric pointer text

• Front matter updates are done (please review, if you have time)
• Need to implement api argument updates across the remaining functions in the spec

Anshuman: Get ordering