[cuda.compute]: add CI job for minimal cuda-cccl extra

[NaderAlAwar]

Description

closes #9433

This also decouples our tests from numba.cuda further since minimal does not install that package. We do this by switching to CuPy and cuda.core where necessary. Also, this ensures that each algorithm has at least one test that can run with the minimal extra

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

important:

Walkthrough

Adds a minimal Python CI job and runner script; applies no_numba markers and guarded imports across many compute tests; converts device memory movement to CuPy and switches compute-capability probes to cuda.core.Device(); and adds RawOp-based minimal tests for select and three-way-partition.

Changes

Python compute minimal CI and test execution path

Layer / File(s)	Summary
Minimal CI job wiring and runner script ci/matrix.yaml, ci/test_cuda_compute_minimal_python.sh	Adds a PR matrix/job test_py_compute_minimal and a script that prepares Python, resolves/installs the minimal wheel with CuPy/pytest, and runs the compute test subset with conditional marker selection.
No-numba markers and import guards python/cuda_cccl/tests/compute/*	Applies pytest.mark.no_numba at module or per-test level, moves eager numba imports into guarded branches or pytest.importorskip, and reduces the no-numba module to a fast-fail import test.
CuPy device buffers and Device capability queries python/cuda_cccl/tests/compute/*	Replaces numba.cuda.to_device/copy_to_host with cp.asarray/.get() and cp.empty, and switches compute-capability checks to cuda.core.Device().compute_capability across sorts/reduce/scan/unique/zip/iterators tests.
RawOp minimal predicate tests python/cuda_cccl/tests/compute/test_select.py, test_three_way_partition.py	Adds inline C++ compile helpers and RawOp predicates plus minimal tests exercising cuda.compute.select and cuda.compute.three_way_partition with explicit CuPy buffers.
Binary-search test adjustment python/cuda_cccl/tests/compute/test_binary_search.py	Marks the explicit OpKind.LESS comparator test with @pytest.mark.no_numba.

Assessment against linked issues

Objective	Addressed	Explanation
Add CI job for minimal cuda-cccl extra [#9433]	✅

Suggested reviewers

• jrhemstad
• davebayer
• miscco

---

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

[coderabbitai]

Actionable comments posted: 2

🧹 Nitpick comments (1)

ci/test_cuda_compute_minimal_python.sh (1)

22-23: 📐 Maintainability & Code Quality | ⚡ Quick win

suggestion: Line 22 and Line 23 hardcode /home/coder/cccl even though repo_root is already available; use ${repo_root} for artifact download and wheelhouse_dir to avoid CI workspace-path coupling.

As per coding guidelines, ci/**/* should focus on environment setup and clear failures; removing hardcoded workspace paths improves that reliability.

Source: Coding guidelines

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 7f0b31eb-d63b-4174-870b-b82feb8faae3

📥 Commits

Reviewing files that changed from the base of the PR and between 809e315 and 230bfad.

📒 Files selected for processing (26)

• ci/matrix.yaml
• ci/test_cuda_compute_minimal_python.sh
• python/cuda_cccl/cuda/compute/algorithms/_binary_search.py
• python/cuda_cccl/tests/compute/test_binary_search.py
• python/cuda_cccl/tests/compute/test_bindings.py
• python/cuda_cccl/tests/compute/test_deferred_annotations.py
• python/cuda_cccl/tests/compute/test_func_caching.py
• python/cuda_cccl/tests/compute/test_histogram.py
• python/cuda_cccl/tests/compute/test_iterators.py
• python/cuda_cccl/tests/compute/test_merge_sort.py
• python/cuda_cccl/tests/compute/test_nested_struct.py
• python/cuda_cccl/tests/compute/test_no_numba.py
• python/cuda_cccl/tests/compute/test_permutation_iterator.py
• python/cuda_cccl/tests/compute/test_radix_sort.py
• python/cuda_cccl/tests/compute/test_reduce.py
• python/cuda_cccl/tests/compute/test_scan.py
• python/cuda_cccl/tests/compute/test_segmented_reduce.py
• python/cuda_cccl/tests/compute/test_segmented_sort.py
• python/cuda_cccl/tests/compute/test_select.py
• python/cuda_cccl/tests/compute/test_shuffle_iterator.py
• python/cuda_cccl/tests/compute/test_struct_field_validation.py
• python/cuda_cccl/tests/compute/test_three_way_partition.py
• python/cuda_cccl/tests/compute/test_transform.py
• python/cuda_cccl/tests/compute/test_unique_by_key.py
• python/cuda_cccl/tests/compute/test_void_ptr_wrapper_validation.py
• python/cuda_cccl/tests/compute/test_zip_iterator.py

[NaderAlAwar]

pre-commit.ci autofix

[coderabbitai]

Actionable comments posted: 1

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 6ed5e7f1-eb57-4e74-9b43-32332ab11d65

📥 Commits

Reviewing files that changed from the base of the PR and between c564d3e and dd27462.

📒 Files selected for processing (1)

• python/cuda_cccl/tests/compute/test_binary_search.py

[NaderAlAwar]

pre-commit.ci autofix

[NaderAlAwar]

/ok to test c2e7c01

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[NaderAlAwar]

/ok to test b09c9c6

[github-actions]

😬 CI Workflow Results

🟥 Finished in 2h 29m: Pass: 96%/504 | Total: 4d 09h | Max: 51m 30s | Hits: 95%/667616

See results here.