Repository navigation
Optimize cuda::device::warp_shuffle - #10035
Merged
Merged
Conversation
Merged
Contributor
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
📝 WalkthroughSummary by CodeRabbit
WalkthroughChangesWarp shuffle operations now share packing, precondition, and result-construction paths. Runtime tests cover additional types and predicates. New architecture-specific SASS tests validate emitted instructions and wrapper/native register usage. Warp shuffle behavior
Possibly related PRs
Suggested reviewers: Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 1
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Enterprise
Run ID: 5655ce3a-9e53-4e59-8c1c-d112ca368704
📒 Files selected for processing (8)
CMakePresets.jsonlibcudacxx/include/cuda/__warp/warp_shuffle.hlibcudacxx/test/CMakeLists.txtlibcudacxx/test/libcudacxx/cuda/warp/warp_shuffle.pass.cpplibcudacxx/test/warp_codegen/CMakeLists.txtlibcudacxx/test/warp_codegen/compare_register_usage.pylibcudacxx/test/warp_codegen/warp_shuffle.culibcudacxx/test/warp_codegen/warp_shuffle_stress.cu
pauleonix
reviewed
Jul 21, 2026
This comment has been minimized.
This comment has been minimized.
Contributor
|
Do we have/need a benchmark for |
davebayer
reviewed
Jul 21, 2026
Contributor
There was a problem hiding this comment.
Actionable comments posted: 1
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Enterprise
Run ID: d5bf2479-0e91-445c-87bb-e2dd2fca66a0
📒 Files selected for processing (2)
libcudacxx/include/cuda/__warp/warp_shuffle.hlibcudacxx/test/libcudacxx/cuda/warp/warp_shuffle.pass.cpp
🚧 Files skipped from review as they are similar to previous changes (1)
- libcudacxx/test/libcudacxx/cuda/warp/warp_shuffle.pass.cpp
Contributor
Author
I think SASS codegen tests are enough. The goal is to match or outperform standard shuffle intrinsics. WarpBitonicSort is a very good candidate to check for potential perf/SASS issues |
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
elstehle
reviewed
Aug 27, 2026
Contributor
There was a problem hiding this comment.
To be on the safe side, we should be expanding testing a bit:
- I think we are missing test coverage for 2-byte types in
warp_shuffle.pass.cpp. Also, we should be shuffling distinct values for the 1 byte case, to confirm values are shuffled correctly. - We should have 128-bit codegen tests
Jacobfaib
approved these changes
Aug 31, 2026
davebayer
requested changes
Aug 31, 2026
fbusato
enabled auto-merge (squash)
August 31, 2026 16:24
This comment has been minimized.
This comment has been minimized.
davebayer
reviewed
Aug 31, 2026
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
Contributor
This comment has been minimized.
This comment has been minimized.
davebayer
reviewed
Sep 1, 2026
Contributor
🔬 CUB benchmark SASS comparisonHow to request a benchmark run
Targets with a SASS change
|
Contributor
🥳 CI Workflow Results🟩 Finished in 2h 57m: Pass: 100%/609 | Total: 21d 12h | Max: 2h 56m | Hits: 36%/2283719See results here. AI failure analysis1. RAPIDS conda environment: rapidsai-nightly repodata HTTP 503 · 4 jobsExplanation: All four jobs fail before compilation while creating the shared RAPIDS conda environment because libmamba cannot retrieve both rapidsai-nightly repodata subdirectories. The identical signatures within seconds across disjoint matrices, combined with no RAPIDS or CI files in the PR diff, indicate an external channel availability failure; subsequent fork and exit errors are derivative. Evidence: Copy this prompt into a coding agentJobs: |
davebayer
approved these changes
Sep 3, 2026
davebayer
disabled auto-merge
September 3, 2026 17:13
1 task done
charan-003
pushed a commit
to charan-003/cccl
that referenced
this pull request
Sep 15, 2026
* draft * add codegen tests * __shuffle_cast * code simplification * address comments * static_cast * fix CI failures * fix CI * Update libcudacxx/include/cuda/__warp/warp_shuffle.h Co-authored-by: Elias Stehle <3958403+elstehle@users.noreply.github.com> * Update libcudacxx/include/cuda/__warp/warp_shuffle.h Co-authored-by: Elias Stehle <3958403+elstehle@users.noreply.github.com> * address comments * fix cmakePresents * Update libcudacxx/include/cuda/__warp/warp_shuffle.h Co-authored-by: Michael Schellenberger Costa <miscco@nvidia.com> * Update libcudacxx/include/cuda/__warp/warp_shuffle.h Co-authored-by: Michael Schellenberger Costa <miscco@nvidia.com> * [bench-only] * use "mandates" * add _CCCL_NODEBUG_ALIAS * use CodegenTest.cmake facilities * use _CCCL_NODEBUG * use int128 shift * remove uniform delta check * remove int128 specialization * revert bench.yaml * use shift for 64-bit values --------- Co-authored-by: Elias Stehle <3958403+elstehle@users.noreply.github.com> Co-authored-by: Michael Schellenberger Costa <miscco@nvidia.com>
19 of 27 tasks
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.
Description
The work related to the PR
highlighted that
cuda::device::warp_shufflegenerates code that is inefficient compared to plain__shfl_<op>_syncinstrinsics. This behavior is caused by additional instructions and far greater register pressure.The PR addresses the following issues:
movinstructions instead ofmemcpyto map generic types to their unsigned types when possible.memcpyto avoid UB values that significantly incremented the register pressure._idxand_xorfunctions (implicit).