search

rapidsai / cuspatial

CUDA-accelerated GIS and spatiotemporal algorithms
https://docs.rapids.ai/api/cuspatial/stable/
Apache License 2.0
608 stars 153 forks source link

[BUG]: Extensive unit test warnings need a clean up. #1183

Open thomcom opened 1 year ago

thomcom commented 1 year ago

Version

23.08

On which installation method(s) does this occur?

Docker, Conda, Source, Rapids-Compose

Describe the issue

pytest unit test are throwing quite a few warnings during execution. They need to be triaged and eliminated where possible.

Minimum reproducible example

No response

Relevant log output

No response

Environment details

No response

Other/Misc.

No response

thomcom commented 1 year ago

Numba has a CUDA_LOW_OCCUPANCY_WARNINGS that raises a warning whenever a binary predicate is called with a small input size (ie for testing). I added NUMBA_CUDA_LOW_OCCUPANCY_WARNINGS=0 to my environment which reduced the number of test errors from 984 to 71.

harrism commented 1 year ago

What binary predicates use Numba?

thomcom commented 1 year ago

Nothing in the binary predicates directly uses numba. My only guess is that cudf functionality is being used that produces that warning, but haven't come up with a way of tracking down exactly what is emitting the warnings yet.

harrism commented 1 year ago

I don't feel comfortable suppressing warnings without understanding the cause.