refactoring: have latches tickle registries directly#691
Closed
nikomatsakis wants to merge 3 commits intorayon-rs:masterfrom
Closed
refactoring: have latches tickle registries directly#691nikomatsakis wants to merge 3 commits intorayon-rs:masterfrom
nikomatsakis wants to merge 3 commits intorayon-rs:masterfrom
Conversation
This eliminates the need for a TickleLatch, since the spin latch can just track the registry that it needs to wake up directly.
The latches now tickle the appropriate registry directly, instead.
nikomatsakis
force-pushed
the
latch-tickle
branch
from
49b473e to
714ceb4
Compare
|
Tried this patch and it does seem to reduce our cpu load immensely. Thanks! |
Member
|
Is this still relevant / worth reviewing on its own, in relation to rayon-rs/rfcs#5? |
Member
Author
|
closing in favor of #746 |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
As a start to improving our sleep system, this PR does a refactoring to how tickle works. We used to tickle the registry after each job executed: the idea was that executing a job would set some latch, and we needed to wake up any threads that might be blocked, waiting on that latch. However, this was a bit overapproximated: for example, the latch might've been a
LockLatch, which indicates a thread from outside the pool, in which case there is no need to tickle threads at all.In this PR, the latches themselves track the registry that they must tickle when they are set. In the case of a
Countdownlatch, they are given it from the outside; this is because countdown latches are sometimes owned by the registry itself, so it would be impossible for them to have a reference to the registry.I'd like to work towards a scenario where we know not only the registry that must be awoken but the exact helper thread. This could avoid needless wakeups. But this refactoring is as far as I got for now.