Chapel interops with GPU Code

[benjamin-robbins]

As a Chapel Programer I want to be able to take advantage of GPUs I want to be able to call into existing GPU code from Chapel, be it standard library like cuBLAS or my own custom lib because it's a way to take advange of GPU resources until Chapel can gen GPU code directly

[marcoscleison]

For me, this one of the most wanted feature. I am going to work with Deep learning stuff, but without GPU is hard.
I know that this is hard but what is needed to get started.

[buddha314]

Have any tasks been created for this epic?

[stonea]

This is already possible using standard Chapel interop features (although perhaps there are things we could do to make it more convenient?). We have some example code of having Chapel interop with cuBLAS and launching custom kernels here:

https://github.com/chapel-lang/chapel/tree/main/test/gpu/interop

I'm going to close this issue for now. If someone feels the current interop support is inadequate, feel free to revive this issue or create a new one. For the time being, however, our current efforts are primarily focused on adding
native GPUu support to Chapel. To learn about where we are with that see this tech note: https://chapel-lang.org/docs/technotes/gpu.html

Edit: For general discussion on GPU support feel free to add comments here: #18554