[PAL/vm-common] Check the "break" variable less frequently in delay()
#42
Conversation
This file contains 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
Previously, `delay()` function accessed the "break out of loop early" variable `continue_gate` basically on every CPU cycle. This variable is typically a global variable causing high contention on multi-core workloads. This e.g. manifested in the Candle Quantized LLaMA app. This commit fixes this by checking the variable less frequently. The current heuristic is to check it every 1 ms. Signed-off-by: Dmitrii Kuvaiskii <[email protected]>
dimakuv
force-pushed
the
dimakuv/fix-candle-rust-perf-32-threads
branch
from
4a973e6 to
1d7aee2
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
Description of the changes
Previously,
delay()function accessed the "break out of loop early" variablecontinue_gatebasically on every CPU cycle. This variable is typically a global variable causing high contention on multi-core workloads. This e.g. manifested in the Candle Quantized LLaMA app.This PR fixes this by checking the variable less frequently. The current heuristic is to check it every 1 ms.
How to test this PR?
Run the Candle example; it shows better scalability.
TODO: Run other benchmarks to see if they are also ok.
This change is