kernelfoundry.conftest

This is the conftest file for configuring pytest for kernelfoundry tasks.

Fixtures in this file are available in the task.py file and provide utilities to measure performance, switch between reference and kernel code, and use template arguments.

Functions

cleanup(request)	Session-scoped fixture to add cleanup code after all tests have run.
measure_runtime(request, profile_store)	Create a generic runtime measurement helper with shared output.
measure_runtime_torch(request, profile_store)	Create a torch runtime measurement helper with shared output.
profile_store(request)	This store is used to collect per-test performance results in a dictionary.
profiler_test_label(request)	Expose the current pytest node id to runtime helpers used during profiling.
pytest_addoption(parser)	Adds CLI options for performance measurements and templating behavior.
pytest_configure(config)	This function registers the "performance" custom marker and configures pytest logging.
template_args_wrapper(request)	Fixture to apply template arguments
torch_profile(measure_runtime_torch)	Deprecated.
use_reference(request)	Fixture to easily switch to the reference implementation.

kernelfoundry.conftest.pytest_configure(config)

This function registers the “performance” custom marker and configures pytest logging.

Note

This function is not intended to be called directly by test code.

kernelfoundry.conftest.pytest_addoption(parser)

Adds CLI options for performance measurements and templating behavior.

Note

This function is not intended to be called directly by test code.

kernelfoundry.conftest.use_reference(request) → bool

Fixture to easily switch to the reference implementation.

This can be used in tests to conditionally run the reference code instead of the kernel code, for example to collect performance data on the reference implementation or to verify that the reference implementation runs correctly.

Example

Run the test with the reference implementation:

python -m pytest --ref task.py

kernelfoundry.conftest.profile_store(request) → dict

This store is used to collect per-test performance results in a dictionary.

The keys of the dictionary are the test node ids, and the values are lists of measurements collected during the test. Tests can append to the list to collect multiple measurements.

Returns:

Mutable store keyed by test node id.

Return type:

dict

kernelfoundry.conftest.profiler_test_label(request)

Expose the current pytest node id to runtime helpers used during profiling.

kernelfoundry.conftest.measure_runtime_torch(request, profile_store) → Callable

Create a torch runtime measurement helper with shared output.

Parameters:

• request (pytest.FixtureRequest) – Pytest fixture request.

• profile_store (dict) – Mutable store for per-test measurements.

Returns:

Wrapped measure_runtime_torch callable.

The callable is kernelfoundry.eval_pipeline.utils.performance.measure_runtime_torch() with the use_itt and output parameters pre-configured.

Return type:

Callable

kernelfoundry.conftest.torch_profile(measure_runtime_torch) → Callable

Deprecated. Alias for the torch runtime measurement helper.

Attention

This fixture is deprecated. Please use measure_runtime_torch() instead.

Parameters:

measure_runtime_torch (Callable) – Runtime measurement helper.

Returns:

Same helper passed in.

Return type:

Callable

kernelfoundry.conftest.measure_runtime(request, profile_store) → Callable

Create a generic runtime measurement helper with shared output.

Parameters:

• request (pytest.FixtureRequest) – Pytest fixture request.

• profile_store (dict) – Mutable store for per-test measurements.

Returns:

Wrapped measure_runtime callable.

The callable is kernelfoundry.eval_pipeline.utils.performance.measure_runtime() with the use_itt and output parameters pre-configured.

Return type:

Callable

kernelfoundry.conftest.template_args_wrapper(request) → bool

Fixture to apply template arguments

kernelfoundry.conftest.cleanup(request)

Session-scoped fixture to add cleanup code after all tests have run.