Instead of separate packages did we want to use extensions?

[ptiede]

I noticed that we are making a bunch of packages. Would it be easier to just use Julia package extensions for this?
This is described here https://pkgdocs.julialang.org/v1/creating-packages/#Conditional-loading-of-code-in-packages-(Extensions)

The benefit is that we can all put this into a single repo, and the code will be automatically loaded when a user, e.g., imports CUDA.
That way, they don't need to import several separate packages, and the registration procedure should be simplified for us.

[kazuakiyama]

This is a good idea indeed --- I didn't know about this capability so thanks! I would still keep AbstractFastHartleyTransform.jl separate from this particular one, but it is indeed a good idea to have CUDA (and perhaps AMDGPU one using rocFFT) inside this package in this way.

[kazuakiyama]

I added FastHartleyTransformCUDAExt in ext using weakdep, and then removed FastHartleyTransformCUDAExt.jl. I can add the rocFFT version by replacing CuArray/CUFFT to rocArray/rocFFT but I can't test because I don't have a AMD GPU now....

Anyway, it works ok with my local nvidia's GPU. I will move forward to registering this package.

[kazuakiyama]

I will close this issue as I have officially released the first version (which is now being registered to the Julia package repo). Feel free to open the issue again if needed.