How to run NiftyReg on GPU

[koegl]

How can I run NiftyReg on my GPU?

I built it from the branch rearchitect-and-reinstate-gpu (I set cmake -USE_CUDA=ON).

When I run reg_f3d on an example .nii pair, the version from the main branch takes 12s and the one from the rearchitect-and-reinstate-gpu takes 15s - so I think it is not running on the GPU.

Is there something else that I have to do to enable the GPU?