cuDNN support in CUDA Fortran?

Does PGI CUDA Fortran support cuDNN?

Since Fortran is so suitable for use in linear algebra, it seems like it’d be a great fit for deep learning.

We haven’t much experience with cuDNN, so making definite statements
may be risky. If it is yet another CUDA library, then we should be able to
interface to via CUDA Fortran or OpenACC in Fortran.

We can’t point you to any work that has been done in Fortran with cuDNN,
but if you have an example you think should work, but does not, please send it.