ak.to_kernels

Defined in awkward.operations.convert on line 790.

ak.to_kernels(array, kernels, highlevel=True, behavior=None)
Parameters

  • array – Data to convert to a specified kernels set.

  • kernels ("cpu" or "cuda") – If "cpu", the array structure is recursively copied (if need be) to main memory for use with the default libawkward-cpu-kernels.so; if "cuda", the structure is copied to the GPU(s) for use with libawkward-cuda-kernels.so.

  • highlevel (bool) – If True, return an ak.Array; otherwise, return a low-level ak.layout.Content subclass.

  • behavior (None or dict) – Custom ak.behavior for the output array, if high-level.

Converts an array from "cpu", "cuda", or "mixed" kernels to "cpu" or "cuda".

An array is "mixed" if some components are set to use "cpu" kernels and others are set to use "cuda" kernels. Mixed arrays can’t be used in any operations, and two arrays set to different kernels can’t be used in the same operation.

Any components that are already in the desired kernels library are viewed, rather than copied, so this operation can be an inexpensive way to ensure that an array is ready for a particular library.

To use "cuda", the package awkward-cuda-kernels be installed, either by

pip install awkward-cuda-kernels

or as an optional dependency with

pip install awkward[cuda] --upgrade

It is only available for Linux as a binary wheel, and only supports Nvidia GPUs (it is written in CUDA).

See ak.kernels.