With PDL, you'd want to use PDL::FFTW3, which as its name suggests uses v3 of FFTW.

That itself might be considered not-the-latest thing, since like the rest of current PDL, it doesn't (last I checked, and again just now) have the capability to use GPUs. For API-compatible options for GPU, NVidia's CUFFT is probably a good starting point.