feat(rocprim): Implement Kernel Tuner for Sort Algorithms - #11678
Draft
NguyenNhuDi wants to merge 42 commits into
Draft
NguyenNhuDi wants to merge 42 commits into
NguyenNhuDi wants to merge 42 commits into
Conversation
NguyenNhuDi
force-pushed
the
users/NguyenNhuDi/rocprim-kernel-tuner-sort-algos
branch
from
September 3, 2026 19:19
1e5198b to
1b3fc03
Compare
NguyenNhuDi
changed the title
✅ All Checks Passed — Ready for Review
📖 Need help? See the Policy FAQ for details on every check and how to fix failures. |
|
🎉 All checks passed! This PR is ready for review. |
1 task
…totuning I followed the instructions in projects/rocprim/tuning/README.md to confirm that the generated default config now matches the fastest specialization. The reason this problem was hidden is because the autotuning system always picked the existing default config, which was already the fastest specialization. By passing `--exclude-default-config` to `run_tuning.py` the autotuning system before this fix commit would output terribly slow configs.
In rocPRIM there are several benchmarks which call `state.run()` multiple times during tuning, like `projects/rocprim/benchmark/benchmark_device_segmented_reduce.hpp` calling it for 5 different segment sizes. This caused the bytes/sec that was returned to Kernel Tuner to be inaccurate, and for the same specialization to be printed 5x in the terminal, which was very confusing.
…NhuDi/rocprim-kernel-tuner-search-algos
NguyenNhuDi
force-pushed
the
users/NguyenNhuDi/rocprim-kernel-tuner-sort-algos
branch
from
September 11, 2026 19:12
fa2b90f to
ad31b6f
Compare
1 task
amd-jmahovsky
left a comment
amd-jmahovsky
left a comment
Contributor
There was a problem hiding this comment.
I think the config headers look OK.
I found a few small things.
amd-jmahovsky
requested changes
Sep 11, 2026
amd-jmahovsky
left a comment
amd-jmahovsky
left a comment
Contributor
There was a problem hiding this comment.
Two potential issues: block_size and ipt metadata swapped in one file, and CONFIG entries missing from another.
NguyenNhuDi
force-pushed
the
users/NguyenNhuDi/rocprim-kernel-tuner-sort-algos
branch
from
September 11, 2026 22:20
74c9d0f to
feae0dd
Compare
NguyenNhuDi
marked this pull request as draft
September 15, 2026 17:01
Contributor
Author
|
Moving back to draft because major changes are occuring |
1 task
NguyenNhuDi
force-pushed
the
users/NguyenNhuDi/rocprim-kernel-tuner-sort-algos
branch
from
September 16, 2026 21:11
ae456d9 to
be7588d
Compare
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
4 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.
Motivation
Implement Kernel Tuner for Sort (merge_sort_block_merge, merge_sort_block_sort, radix_sort_block_sort, radix_sort_onesweep) algorithms.
JIRA ID: AIPRIMS-236 & EXSWSTRHPC-322
Technical Details
The rocPRIM config changes are due to running migrate.py (this updates the legacy config to be ready for Kernel Tuner)
This PR is built on top of #11617, #12119 and
#11687, do not merge until the mentioned PR has been mergedRecommended way to review this PR is to go to "Commit" and review the files per commit.
Test Plan
Local running of kernel tuner, making sure output matches legacy tuners and that there are no errors.
Test Result
Generated config files compiles, and all unit tests were passing locally
Submission Checklist