Exploring GPU acceleration using CuPy for enhanced performance

[zjzjwang]

Thank you for your excellent work.

I am considering whether it would be feasible to further accelerate the process using a GPU (such as CuPy).

As mentioned in #63 , using CuPy should speed up some operators, but my implementation did not observe a significant speed increase (in fact, it significantly decreased). （I simply replaced numpy with cupy in aggregate_numpy.py )

Do you have some ideas on implementations on GPU?

[ml31415]

Hi @zjzjwang We only had this PR so far, you probably saw it: #63 . The problem with this algorithm is, that it is hard to parallelize. Which is, what the GPU would be good for. See here for approaches for further parallelization: https://github.com/xarray-contrib/flox

[zjzjwang]

Thanks for the feedback, @ml31415 .

I'll check out the link you provided. We can close this issue now.