How to use `copy!`(or other array operations) inside a CUDA kernel?

[maleadt]

@mateuszbaran

I don’t know why I can’t pass a HybridMatrix directly to the kernel, it’s a very simple wrapper.

CUDA.jl uses Adapt.jl to convert wrappers. If Adapt doesn’t know about HybridMatrix, which it doesn’t (the package should define Adapt rules), you’ll end up with a CuArray on the device instead of a CuDeviceArray.