Skip to content

Conversation

@JoelFilho
Copy link
Owner

First tentative for the pipeline idling interface. Closes #3.

Some questions need to be answered before merging:

  • Is it possible to have false positives on wait_until_idle(), between removing an item internally on the queue, and exiting a worker thread's idle state? (TODO: either demonstrate with a test case or show why it can't)
  • Does the O(1) lookup on the bitset makes it better than an array of atomic boolean? Is copy-and-swap much slower than basic false sharing? Is there a better lock-free alternative? (TODO: benchmark, analyze the possibility of race conditions and invalid states)
  • This incurs an overhead every iteration on each thread. Should it be opt-in only? (TODO: benchmark)

JoelFilho added 3 commits May 31, 2020 13:57
- preventing doctest target installation without relying on cache
Initial tentative for the idle interface
@JoelFilho JoelFilho added the work in progress Issue currently being worked on label Jun 8, 2020
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

work in progress Issue currently being worked on

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Pipeline idling

2 participants