[bernhardmgruber]

• Use the new tuning API for detail::radix_sort::dispatch #7949
• Fix segmented radix sort benchmark segment size type #9039
• No SASS changes on cub.bench.segmented_radix_sort.keys.base for SM 75;80;90;100

Fixes: #8926

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[coderabbitai]

Note

Reviews paused

It 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 reviews.auto_review.auto_pause_after_reviewed_commits setting.

Use the following commands to manage reviews:

• @coderabbitai resume to resume automatic reviews.
• @coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

• ▶️ Resume reviews
• 🔍 Trigger review

📝 Walkthrough

Walkthrough

important: Migrates segmented-radix-sort dispatch to derive policy from a tuning environment; public overloads forward the environment; benchmark updated to call DeviceSegmentedRadixSort::SortKeys; adds tuning tests that assert selected block sizes.

Changes

Segmented Radix Sort Tuning API

Layer / File(s)	Summary
Dispatch API refactor to environment-driven policy selection cub/cub/device/dispatch/dispatch_segmented_radix_sort.cuh	Internal dispatch template accepts TuningEnvT (default ::cuda::std::execution::env<>), derives policy_selector_t from it, validates via static_assert, constructs kernel source/launcher internally, and computes active_policy by invoking the derived selector.
Public DeviceSegmentedRadixSort overloads forward tuning environment cub/cub/device/device_segmented_radix_sort.cuh	All environment-based overloads (SortPairs/SortPairsDescending and SortKeys/SortKeysDescending; pointer and DoubleBuffer variants) use dispatch_with_env and forward tuning_env plus a decomposer {} into segmented_radix_sort::dispatch.
Benchmark refactor to use public SortKeys API cub/benchmarks/bench/segmented_radix_sort/keys.cu	seg_radix_sort drops the named type-list parameter, removes unused segment/value aliases and value buffer wiring, simplifies key-pointer setup, and switches temp-storage sizing and execution to call DeviceSegmentedRadixSort::SortKeys directly.
Comprehensive tuning tests for block-size selection cub/test/catch2_test_device_segmented_radix_sort_env.cu	Adds #if TEST_LAUNCH == 0 tuning tests with a tuning-policy functor that forces threads_per_block; parameterized tests (64,128) verify the tuned blockDim.x for SortPairs/SortPairsDescending/SortKeys/SortKeysDescending and DoubleBuffer variants by recording d_block_size[0].

Assessment against linked issues

Objective	Addressed	Explanation
Use the new tuning API internally for detail::segmented_radix_sort::dispatch [#8926]	✅

Suggested reviewers

• pauleonix
• oleksandr-pavlyk

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 2

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: ba363737-cffc-471b-83a3-af69cfc27682

📥 Commits

Reviewing files that changed from the base of the PR and between db90838 and 3c3a5f9.

📒 Files selected for processing (4)

• cub/benchmarks/bench/segmented_radix_sort/keys.cu
• cub/cub/device/device_segmented_radix_sort.cuh
• cub/cub/device/dispatch/dispatch_segmented_radix_sort.cuh
• cub/test/catch2_test_device_segmented_radix_sort_env.cu

[bernhardmgruber]

This causes the SASS diffs and I think the benchmark definition is wrong. The public API only allows a segment size type of int32.

[github-actions]

🥳 CI Workflow Results

🟩 Finished in 2h 31m: Pass: 100%/285 | Total: 3d 13h | Max: 59m 30s | Hits: 65%/329558

See results here.