Skip to content

test: add CPU-only unit coverage for sharding helpers#3250

Open
Pritiks23 wants to merge 1 commit into
NVIDIA:mainfrom
Pritiks23:test/sharding-unit-tests
Open

test: add CPU-only unit coverage for sharding helpers#3250
Pritiks23 wants to merge 1 commit into
NVIDIA:mainfrom
Pritiks23:test/sharding-unit-tests

Conversation

@Pritiks23
Copy link
Copy Markdown

Summary

  • add unit tests for sharding utility internals in test/utils/test_sharding_unit.py
  • validate marker argument parsing/defaulting/validation in mark_values
  • validate context-aware predicate evaluation in params_in_context
  • validate default SM inheritance/precedence behavior in DefaultSMs

Why

test/utils/test_sharding.py contains core test-selection logic used by pytest sharding flows, but helper-level behavior had no direct unit coverage. These tests harden that logic without requiring CUDA or GPU runtime.

Validation

  • pytest -q test/utils/test_sharding_unit.py
  • result: 8 passed

Notes

  • CPU-only verification path
  • no CUTLASS kernel behavior changes

Signed-off-by: Pritika Vipin <65793273+Pritiks23@users.noreply.github.com>
@Pritiks23
Copy link
Copy Markdown
Author

This PR adds CPU-only unit coverage for sharding helper logic used during pytest test selection, without changing any kernel/runtime behavior.

What is covered:

Marker argument parsing/defaulting/validation via mark_values
Parameter-context predicate evaluation via params_in_context
Default SM inheritance and nearest-scope precedence via DefaultSMs
Why this matters:

The sharding helper code influences which tests run or skip in CI
These paths were previously untested at helper-unit level
Added coverage reduces regression risk in test selection behavior and keeps verification possible without CUDA/GPU

@Pritiks23
Copy link
Copy Markdown
Author

Hi maintainers — PR is ready for review. I’m happy to iterate on feedback.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant