Skip to content

Concurrency/cuda stream Improvements#566

Draft
ptheywood wants to merge 1 commit into
masterfrom
cineca_concurrency
Draft

Concurrency/cuda stream Improvements#566
ptheywood wants to merge 1 commit into
masterfrom
cineca_concurrency

Conversation

@ptheywood

Copy link
Copy Markdown
Member

Additional use of concurrency / improvments implemented during the cineca hackathon

  • Async/streamed id generation related functions (for birth)
  • Message output post processing
  • Agent birth post processing
  • Agent death post processing
  • Batched memcpy
  • Pinned memory.
  • Stream ownership / access.

@ptheywood
ptheywood force-pushed the cineca_concurrency branch from 4eb1eb2 to fbb0893 Compare June 24, 2021 10:48
@ptheywood
ptheywood force-pushed the cineca_concurrency branch 2 times, most recently from 1563c73 to 098cf61 Compare September 21, 2021 14:30
@ptheywood

Copy link
Copy Markdown
Member Author

A decent chunk of the todo's in this have been covered by @Robadob since this was made, however there are still uses of non-async default stream cudaMemcpy in getDeviceNextID, so I won't close this yet (but won't resume working on this as a priority either)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants