Use signed offset type for DevicePartition

[bernhardmgruber]

The env-overloads in cub::DevicePartition used unsigned offset types, while the non-env-overloads used signed ones. This was presumably an oversight in #7459.

Fixes: #8968

[coderabbitai]

📝 Walkthrough

Summary by CodeRabbit

• Bug Fixes
  • Improved offset computation in DevicePartition's Flagged and If APIs for better handling of large datasets.

Walkthrough

cub::DevicePartition environment-based Flagged and If overloads are updated to use signed offset type selection (detail::choose_signed_offset<NumItemsT>) instead of unsigned, matching the behavior of non-environment overloads and adding the ChooseOffsetT::is_exceeding_offset_type bounds check.

Changes

Offset type alignment

Layer / File(s)	Summary
Offset type switch in Flagged and If overloads cub/cub/device/device_partition.cuh	Flagged (lines 324–331) and If (lines 575–582) environment-based overloads switch from unsigned detail::choose_offset_t<NumItemsT> to signed detail::choose_signed_offset<NumItemsT>. Both now perform the exceeding-offset-type bounds check using the signed offset type, aligning with non-environment overloads.

Assessment against linked issues

Objective	Addressed	Explanation
Unify env-overloads to use signed offset types (#8968)	✅

---

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

[coderabbitai]

🧹 Nitpick comments (1)

cub/cub/device/device_partition.cuh (1)

324-331: ⚡ Quick win

suggestion: Add regression tests for env overload offset-bound behavior.
The signed offset switch and is_exceeding_offset_type guard look aligned with the non-env path; add targeted tests that cover boundary (max signed offset) and overflow (max + 1 for unsigned NumItemsT) inputs for both env Flagged and env If overloads to prevent regressions.

As per coding guidelines, "cub/**/*: Focus on algorithm correctness, temporary-storage protocol, dispatch/policy selection, stream behavior, CUDA error handling, synchronization, memory access safety, performance regressions, and test coverage."

Also applies to: 575-582

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 2d52e5ac-8795-4344-b77f-a99a02b6f277

📥 Commits

Reviewing files that changed from the base of the PR and between e49bdfa and 98092ec.

📒 Files selected for processing (1)

• cub/cub/device/device_partition.cuh

[github-actions]

🥳 CI Workflow Results

🟩 Finished in 3h 28m: Pass: 100%/283 | Total: 6d 16h | Max: 1h 53m | Hits: 53%/430269

See results here.

[github-actions]

Successfully created backport PR for branch/3.4.x:

• [Backport branch/3.4.x] Use signed offset type for DevicePartition #8976