chan_vese: pass all constants to `_fused_variance_kernel2` as device scalars

rapidsai / cucim

cuCIM - RAPIDS GPU-accelerated image processing library

https://docs.rapids.ai/api/cucim/stable/

Apache License 2.0

356 stars 60 forks source link

chan_vese: pass all constants to `_fused_variance_kernel2` as device scalars #764

Closed grlee77 closed 3 months ago

grlee77 commented 3 months ago

This change avoids an issue with an internal numpy.can_cast call in CuPy's kernel fusion code when using NumPy 2.0. With this change all cuCIM tests passed using NumPy 2.0.1 and the CuPy 13.3dev branch.

Fixes https://github.com/rapidsai/cucim/issues/742

gigony commented 3 months ago

Thanks @grlee77 It looks good to me!

gigony commented 3 months ago

/merge

jakirkham commented 3 months ago

Thanks Greg and Gigon! 🙏