Prior art

• NumPy - https://numpy.org/doc/stable/reference/generated/numpy.kron.html
• CuPy - https://docs.cupy.dev/en/latest/reference/generated/cupy.kron.html
• Dask - Add NumPy's kron for Dask Arrays dask/dask#3657
• JAX - https://jax.readthedocs.io/en/latest/_autosummary/jax.numpy.kron.html
• PyTorch - https://pytorch.org/docs/stable/generated/torch.kron.html
• Tensorflow - https://github.com/tensorflow/tensorflow/blob/v2.16.1/tensorflow/python/ops/numpy_ops/np_math_ops.py#L436-L469

Motivation

We have an awkward function kron in scipy.linalg. We would like to deprecate the function and direct users to np.kron instead, like we have done for np.triu and np.tril. However, unlike those functions, xp.kron is not in the standard, hence we have been hesitant to deprecate in case it does not make it into the standard¹. Experiments done by @ilayn in scipy/scipy#20077 demonstrate that there are big performance gains to be had with a native array library implementation. Also, an array-agnostic implementation by @mdhaber in scipy/scipy#20077 (comment) demonstrates that at least a Python-level solution is easily implementable in standard array libraries.

Whether kron can be included in the standard or not, it would be very useful for us to know either way, so that we can plan whether we should deprecate scipy.linalg.kron or not.