Skip to content

Conversation

@ashwinvis
Copy link
Contributor

@ashwinvis ashwinvis commented Nov 26, 2025

FYI, @stepas-toliautas , this is a follow up since paths were changed in #113

@ashwinvis ashwinvis merged commit 7a16ad1 into main Nov 26, 2025
4 checks passed
@stepas-toliautas
Copy link
Collaborator

@ashwinvis : thanks for the updates!
Do I understand correctly that

  • the CuPy version directly uses CUDA kernels from _cuda subfolder, and
  • the Jax version has potential to use vendor-agnostic Python GPU kernels (as in SYCL for C++), but code still needs to be finished/ polished?

The latter would be really nice, but I won't have time to integrate Jax variant into existing codebase until tomorrow. Still, maybe a comment on the possibility during the lesson could be worth it.

@ashwinvis
Copy link
Contributor Author

  • CuPy version uses the kernel written inline as strings in the file, where we define the evolve kernel
  • Jax needs a lot of polish certainly. It can be illuminating to show how it forces us to write code in a certain way. When it works, Jax is the most vendor agnostic of all.

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.

3 participants