[cuda.compute]: Add tests that depend on the system CTK instead of a pip installed one

[NaderAlAwar]

Description

closes #10060

Checklist

• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[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

Summary by CodeRabbit

• New Features
  • Added configurable CUDA toolkit modes for pinned, latest, and system-provided toolkits across Python and Windows CI workflows.
  • Added free-threaded Python test coverage across multiple CUDA toolkit versions.
  • Added system-toolkit dependency options for CUDA 12 and CUDA 13 environments.
• Bug Fixes
  • Improved dependency selection so tests install packages matching the selected toolkit mode.
  • Added validation for unsupported or empty toolkit mode values.

Walkthrough

Changes

The CI matrix now defines pinned, latest, and sysctk toolkit modes. Workflow generation forwards the mode to Unix and Windows tests. Test scripts select matching toolkit constraints and cuda_cccl extras. Sysctk package extras no longer request pip CUDA toolkit variants.

Python CTK modes

Layer / File(s)	Summary
Matrix mode propagation .github/actions/workflow-build/build-workflow.py, ci/matrix.yaml	Matrix entries define latest and sysctk lanes. Dispatch names and commands include py_ctk_mode. Two-stage producers remove this consumer-only tag.
CTK mode contracts and helpers ci/util/python/common_arg_parser.sh, ci/pyenv_helper.sh, ci/windows/build_common_python.psm1	Argument parsing, toolkit pinning, mode validation, and pip-extra flavor mapping now use explicit CTK modes.
Unix test dependency selection ci/test_cuda_*_python.sh	Unix tests pass ctk_mode to toolkit setup and install mode-specific cuda_cccl extras.
Windows test wiring and sysctk extras ci/windows/test_cuda_*_python.ps1, python/cuda_cccl/pyproject.toml	Windows tests accept ctk-mode and select matching extras. sysctk12 and sysctk13 use plain numba-cuda.

Assessment against linked issues

Objective	Addressed	Explanation
Add CI jobs for sysctk and minimal-sysctk extras, and use system-toolkit dependency selection [#10060]	✅
Remove [cu12] and [cu13] from the sysctk dependency definitions [#10060]	✅

Suggested reviewers: wmaxey

---

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

[coderabbitai]

Actionable comments posted: 1

🧹 Nitpick comments (1)

ci/test_cuda_cccl_examples_python.sh (1)

31-32: 🩺 Stability & Availability | 🔵 Trivial

suggestion: Run the targeted pinned, latest, and sysctk lanes, including the corresponding minimal/header/compute jobs, and run pre-commit run --all-files; no GPU/Windows validation results are included here. As per coding guidelines, changes must be validated with relevant builds/tests and pre-commit; as per path instructions, CI scripts should use targeted build/test behavior.

Sources: Coding guidelines, Path instructions

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 734bfd9a-f2a5-4d2d-92aa-b0ec6a938e40

📥 Commits

Reviewing files that changed from the base of the PR and between 2cb708a and 82914ca.

📒 Files selected for processing (14)

• .github/actions/workflow-build/build-workflow.py
• ci/matrix.yaml
• ci/pyenv_helper.sh
• ci/test_cuda_cccl_examples_python.sh
• ci/test_cuda_cccl_headers_python.sh
• ci/test_cuda_compute_minimal_python.sh
• ci/test_cuda_compute_python.sh
• ci/util/python/common_arg_parser.sh
• ci/windows/build_common_python.psm1
• ci/windows/test_cuda_cccl_examples_python.ps1
• ci/windows/test_cuda_cccl_headers_python.ps1
• ci/windows/test_cuda_compute_minimal_python.ps1
• ci/windows/test_cuda_compute_python.ps1
• python/cuda_cccl/pyproject.toml

[NaderAlAwar]

/ok to test d903dc4

[coderabbitai]

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

ci/matrix.yaml (1)

149-150: 🎯 Functional Correctness | 🟠 Major | 🏗️ Heavy lift

important: The global exclusion at Lines 451-452 removes every test job with msvc2022, so the full sysctk rows create no Windows coverage. The test_py_compute_minimal rows are not excluded and still request gpu: 'l4', although the matrix states that Windows GPU runners are unavailable. Use a Windows GPU-capable lane, or remove the unsupported Windows combinations. Apply the same platform rule to both full and minimal sysctk jobs.

• ci/matrix.yaml#L149-L150: Correct the pull-request sysctk Windows lane behavior.
• ci/matrix.yaml#L295-L296: Correct the nightly sysctk Windows lane behavior.
• ci/matrix.yaml#L392-L393: Correct the weekly sysctk Windows lane behavior.

As per path instructions, focus on matrix correctness and GPU availability assumptions.

Source: Path instructions

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 2c78a346-3dcb-4d73-98a5-4ebd20f46fb0

📥 Commits

Reviewing files that changed from the base of the PR and between 848ff7c and d903dc4.

📒 Files selected for processing (1)

• ci/matrix.yaml

[github-actions]

⏱️ CCCL compile-time benchmark comparison: Public headers compile-time bench

Result: 1 regression row(s), 10 improvement row(s) above threshold.

Run	Value
Config	public-headers-gcc13
Baseline	origin/main
Preset	all-dev
Targets	cub.headers.base, thrust.cpp.cuda.headers.base, libcudacxx.test.public_headers
GPU / launch args	rtx2080 / --cuda 13.3 --host gcc13

Artifacts: reports and traces

TU total compilation

-f total-compilation inclusive --sort total

🟢 TU total compilation — Improvements

Rank	Improvement impact	Selected Δ	Baseline	Current	Event	Matched traces
1	3.966160	-3.966160	13.315969	9.349809	Total Compilation Time: thrust/fill.h	1
2	3.427344	-3.427344	17.655111	14.227767	Total Compilation Time: thrust/device_vector.h	1
3	3.188338	-3.188338	19.081187	15.892849	Total Compilation Time: cub/cub.cuh	1
4	3.055851	-3.055851	11.009715	7.953864	Total Compilation Time: cub/device/dispatch/dispatch_select_if.cuh	1

Direct file processing

-f file-processing exclusive --sort total

🔴 Direct file processing — Regressions

Rank	Regression impact	Selected Δ	Baseline	Current	Event	Matched traces
1	1.053958	1.053958	21.197551	22.251509	Processing Header File: libcudacxx/include/cuda/__device/physical_device.h	94

🟢 Direct file processing — Improvements

Rank	Improvement impact	Selected Δ	Baseline	Current	Event	Matched traces
1	1.255461	-1.255461	9.962178	8.706717	Processing Header File: libcudacxx/include/cuda/std/__cccl/prologue.h	548
2	0.622186	-0.622186	3.654748	3.032562	Processing Header File: libcudacxx/include/cuda/std/__cccl/epilogue.h	548
3	0.393773	-0.393773	11.872543	11.478770	Processing Header File: libcudacxx/include/cuda/std/__iterator/concepts.h	436
4	0.236856	-0.236856	3.329322	3.092466	Processing Header File: libcudacxx/include/cuda/std/__ranges/concepts.h	404
5	0.229529	-0.229529	0.558630	0.329101	Processing Header File: libcudacxx/include/cuda/__memory_resource/any_resource.h	6
6	0.209039	-0.209039	1.778567	1.569528	Processing Header File: libcudacxx/include/cuda/std/__limits/numeric_limits_ext.h	407

[github-actions]

🥳 CI Workflow Results

🟩 Finished in 7h 40m: Pass: 100%/534 | Total: 17d 14h | Max: 2h 37m | Hits: 35%/1944623

See results here.