Use block-scoped atomics in BlockHistogramAtomic on SM 60+#8162
Open
NitishNaineni wants to merge 3 commits intoNVIDIA:mainfrom
Open
Use block-scoped atomics in BlockHistogramAtomic on SM 60+#8162NitishNaineni wants to merge 3 commits intoNVIDIA:mainfrom
NitishNaineni wants to merge 3 commits intoNVIDIA:mainfrom
Conversation
Contributor
miscco
requested changes
Mar 25, 2026
bernhardmgruber
approved these changes
Mar 25, 2026
NitishNaineni
force-pushed
the
histogram-atomic-scope
branch
from
784ceca to
652f279
Compare
simplify to just atomicAdd_block since CCCL only supports SM 75+ Co-authored-by: Michael Schellenberger Costa <miscco@nvidia.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Closes #3357
Summary
atomicAddwithatomicAdd_blockinBlockHistogramAtomic::Compositeon SM 60+ usingNV_IF_TARGET, with fallback toatomicAddon older architectures.agent_histogram.cuh(lines 312, 325, 350).Codegen verification
SASS confirms the scope change on SM 86 (RTX 3080):
RED.E.ADD.STRONG.GPURED.E.ADD.STRONG.SMPerformance was identical on Ampere in my benchmarks (1024 threads × 16 items, per-block global histograms, various bin counts and grid sizes).
Test plan
cub.test.block.histogram.bins_32— all 1512 assertions passedcub.test.block.histogram.bins_256— all 1512 assertions passedcub.test.block.histogram.bins_1024— all 1512 assertions passed