Cuda casting tests

[CodersAcademy006]

This PR adds kernel-based tests for device-side casting between common numeric and boolean dtypes, mirroring a subset of the CPU-side test_casting logic. The tests validate CUDA device semantics for:

• Numeric widening and narrowing (e.g., int32 ↔ int64, float32 ↔ float64)
• Float ↔ int (with explicit truncation semantics)
• Boolean ↔ int
• Only dtypes and conversions natively supported by CUDA
• Tests are skipped under cudasim to avoid simulator/Python semantic mismatches. This work references and addresses issue Incomplete Test Coverage in Numba-CUDA #515.

[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.

[greptile-apps]

Greptile Summary

This PR adds a new comprehensive test suite for CUDA device-side casting operations, addressing issue #515 to improve test coverage for Numba-CUDA. The implementation validates casting between numeric types (int32, int64, float32, float64) and boolean conversions, with special attention to CUDA's C-style truncation semantics for float-to-int conversions.

The test suite is well-structured with:

• A clean 1D kernel that performs implicit type casting during assignment
• A helper method that handles device memory allocation, kernel invocation, and validation
• Comprehensive test cases for widening, narrowing, and cross-type conversions
• Proper use of @skip_on_cudasim to avoid simulator semantic mismatches
• Efficient device memory management using cuda.device_array()
• Clear documentation via comments explaining CUDA truncation behavior

Confidence Score: 5/5

• This PR is safe to merge. The new test code is well-structured, follows established patterns, contains no syntax errors, and properly validates CUDA casting behavior.
• Full confidence is warranted because: (1) The code is syntactically correct and follows Python/CUDA best practices, (2) The kernel implementation is simple and correct with proper boundary checking, (3) Test coverage is comprehensive with good dtype variety, (4) Device memory management is efficient using cuda.device_array(), (5) The decorator usage correctly skips tests on cudasim to avoid false failures, (6) Grid/block dimension calculations are correct, and (7) The PR directly addresses a tracked issue (Incomplete Test Coverage in Numba-CUDA #515) for improving test coverage in Numba-CUDA.
• No files require special attention

Important Files Changed

Filename	Overview
numba_cuda/numba/cuda/tests/test_casting.py	New comprehensive CUDA casting test suite that validates device-side type conversions across numeric and boolean dtypes. Tests properly use the CUDA kernel architecture with correct grid/block calculations, efficient device memory allocation via device_array, and comprehensive test coverage for int/float/bool conversions including proper handling of CUDA truncation semantics.

[greptile-apps]

_{1 file reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{1 file reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{1 file reviewed, 2 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{1 file reviewed, 3 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{1 file reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{1 file reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}