Repository navigation
GPU Metal: run the tracklet constructor with 128 threads per group - #16113
Merged
Merged
Conversation
The tracklet constructor needs the most registers a Metal thread can have, which leaves room for 384 threads on a core of an M1 Max, so with 256 threads per group only one group fits. With 128 threads per group two to three groups do, and the kernel takes 20% less time: 516 ms -> 415 ms on a quarter of a 30 kHz Pb-Pb time frame (--debug 1, interleaved runs), with the same number of tracks. 96 and 192 threads are within 2% of 128; 64, 160, 256 and 384 are slower. The kernel loops over its start hits with a grid stride and its shared memory does not depend on the group size.
Member
Author
|
Metal only, not checked. |
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The tracklet constructor needs the most registers a Metal thread can
have, which leaves room for 384 threads on a core of an M1 Max, so with
256 threads per group only one group fits. With 128 threads per group
two to three groups do, and the kernel takes 20% less time: 516 ms ->
415 ms on a quarter of a 30 kHz Pb-Pb time frame (--debug 1, interleaved
runs), with the same number of tracks. 96 and 192 threads are within
2% of 128; 64, 160, 256 and 384 are slower. The kernel loops over its
start hits with a grid stride and its shared memory does not depend on
the group size.