cuda/tests: fix array manipulation tests for cudasim and grid-stride correctness

[CodersAcademy006]

Summary

This PR fixes multiple correctness and consistency issues in
numba/cuda/tests/test_array_manipulation.py for the CUDA backend.

The changes focus on aligning test behavior with actual CUDA execution
semantics and ensuring tests are skipped appropriately when running
under the CUDA simulator (cudasim).

Changes

• Implement a true grid-stride loop in test_grid_stride_correctness
  using cuda.grid(1) and cuda.gridsize(1) to match the test’s intent.
• Add @skip_on_cudasim to tests that rely on real device memory
  semantics:
  • test_dtype_transitions
  • test_shape_semantics_in_kernel
  • test_memory_lifetime
  • test_grid_stride_correctness
  • test_cpu_parity_reference
• Remove the duplicate inline kernel definition in
  test_cpu_parity_reference and reuse the module-level add_kernel
  for consistency and maintainability.
• Improve overall test clarity and avoid misleading behavior under
  cudasim.

Rationale

Several tests relied on CUDA device semantics that are not supported
by the CUDA simulator, leading to misleading failures. Additionally,
the grid-stride test previously did not implement a true grid-stride
pattern, making the test name inaccurate.

These changes ensure:

• Correct CUDA semantics are exercised where intended.
• Simulator runs are skipped when behavior cannot be faithfully tested.
• Kernel reuse and test intent are clear and maintainable.

If these changes are ready to be merged, I will continue by adding more
test files listed in the tracking issue and work toward completing the
entire test coverage list.
Fixes a part of #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 Overview

Greptile Summary

Adds comprehensive array manipulation test coverage for CUDA backend. The implementation correctly uses grid-stride loops with cuda.grid(1) and cuda.gridsize(1), properly skips tests on cudasim that require real device memory semantics, and marks unsupported operations with @pytest.mark.xfail. Test kernels are well-structured and reuse module-level kernel definitions for consistency.

Confidence Score: 4/5

• This PR is safe to merge with minimal risk
• The new test file follows established patterns in the codebase for grid-stride loops and cudasim handling. All test implementations are straightforward and correct. Minor note: some tests use uninitialized arrays but only check specific indices they write to, which is acceptable for testing purposes.
• No files require special attention

Important Files Changed

File Analysis

Filename	Score	Overview
numba/cuda/tests/test_array_manipulation.py	4/5	New comprehensive test file for array manipulation - well-structured with correct grid-stride patterns and appropriate cudasim skips

[greptile-apps]

Greptile Summary

This PR adds a new comprehensive test file for CUDA array manipulation functionality, addressing part of issue #515. The changes include:

• Implementation of true grid-stride loop pattern in test_grid_stride_correctness using cuda.grid(1) and cuda.gridsize(1), matching existing patterns in the codebase (e.g., test_datetime.py)
• Appropriate use of @skip_on_cudasim decorator for tests requiring real device memory semantics, consistent with other test files
• Kernel reuse via module-level add_kernel definition in test_cpu_parity_reference
• Multiple test cases covering basic operations, indexing, dtype transitions, and expected failures for unsupported operations
• All xfail tests properly document unsupported CUDA device array operations

Confidence Score: 5/5

• This PR is safe to merge with minimal risk
• The implementation follows established patterns from the codebase (grid-stride loops match test_datetime.py, @skip_on_cudasim usage matches test_array.py), adds comprehensive test coverage without modifying existing functionality, and all tests follow CUDA best practices for device memory handling and kernel execution
• No files require special attention

Important Files Changed

Filename	Overview
numba/cuda/tests/test_array_manipulation.py	New comprehensive test file for CUDA array manipulation with proper grid-stride loops, cudasim skipping, and kernel reuse

[greptile-apps]

Greptile's behavior is changing!

From now on, if a review finishes with no comments, we will not post an additional "statistics" comment to confirm that our review found nothing to comment on. However, you can confirm that we reviewed your changes in the status check section.

_{This feature can be toggled off in your Code Review Settings by deselecting "Create a status check for each PR".}