Open ZhongyuXie921 opened 1 month ago
Could you please consider this in the next version?
It's a good feature request, but this is very unlikely to happen. If you rely on this functionality, it's best to try and implement it yourself. Some pointers: CUSPARSE does not support Kronecker product, so this needs a native implementation. cupy has one: https://github.com/cupy/cupy/blob/4c9821bf219026340e976d090b259e370e3ecddd/cupyx/scipy/sparse/_construct.py#L496-L546.
I'm trying to convert my code from CPU implementation to GPU implementation. However, my code require lots of kron products for sparse arrays. I could not find implementation of kron in CUDA.jl. Could you please consider this in the next version? Many thanks