Closed
Description
The current test suite needs several improvements:
- Move from Python unittest to pytest to allow parameterization.
- Run pytest in conda recipe (Run pytest in conda recipe #222 )
- Organize tests logically. At the least, we should split them into
dppy_kernel_tests
andnjit_tests
. - Tests should run for realistic problem sizes.
- We need to run the test cases for multiple
dpctl.device_context
(opencl:gpu
,opencl:cpu
,level0:gpu
). - The test cases should be run for multiple data types. To start with maybe
float32, float64
are sufficient, but later we should test for integer types, complex types, and reduced precision types. -
Add tests for numpy broadcast operations. (Are there existing Numba tests that we can repurpose?)