Repository navigation
parallel_reduce_sum: drop caller scratch, keep default pool cached - #10
Merged
Merged
Conversation
cub's env DeviceReduce::Sum now owns the scratch (cudaMallocAsync/ cudaFreeAsync per call), so the parallel_reduce_sum_bytes API is gone. The default mem pool trims to 0 bytes at every sync, which made a synchronized reduce cost ~0.8 ms per call. retain_default_pool() raises the release threshold once, bringing it back to caller-scratch speed.
shsahiti
force-pushed
the
reduce-per-call-scratch
branch
from
September 28, 2026 17:17
f6b2569 to
7ceaef4
Compare
karl-kes
approved these changes
Sep 30, 2026
karl-kes
left a comment
karl-kes
left a comment
Contributor
There was a problem hiding this comment.
Excellent reduction in code
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
2 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
cub's env DeviceReduce::Sum now owns the scratch (cudaMallocAsync/ cudaFreeAsync per call), so the parallel_reduce_sum_bytes API is gone.
The default mem pool trims to 0 bytes at every sync, which made a synchronized reduce cost ~0.8 ms per call. retain_default_pool() raises the release threshold once, bringing it back to caller-scratch speed.
bench/reduce_scratch.cu compares against caller scratch and naive cudaMalloc per call.