Skip to content

Move CUDA backend to CUDA.jl

Churavy, Valentin requested to merge vc/gpu_arrays into master

Created by: vchuravy

In order to use KernelAbstractions to implement GPUArrays the GPU backends need to be implemented in CUDA.jl and ROCM.jl, otherwise we could get circular dependencies.

TODO:

  • Make tests work properly

Merge request reports