Support for the log() on General Unitary Matrices.

[zazabap]

Is your feature request related to a problem? Please describe.

Currently I am work on Manifold Optimization and hope to compute on the GPU for the General Unitary Matrices, not only the Hermitian ones on log().

Describe the solution you'd like

Direct API support.

Describe alternatives you've considered

Nothing alternative.

Additional context

Could check PR 856 for the discussion JuliaManifolds/Manifolds.jl#856

[mateuszbaran]

I think it would be better to implement this in ManifoldsGPU using for example eigenvalue decomposition that already exists in CUDA.jl instead of asking for a new feature here. You have the right idea in the original post here: JuliaManifolds/Manifolds.jl#856 (comment) (see log_safe!)

[mateuszbaran]

You could just directly copy the part related to log_safe! and exp on GeneralUnitaryMatrices in that PR and submit it to ManifoldsGPU,jl -- these are fine.

[zazabap]

@mateuszbaran Ok I will just copy and integrate it into a new PR.