Passing OpenCV Mat into cuFFT program

I am trying to create a cuda program that accepts images and performs a FFT on them. The images that are captured from Point Grey camera are converted to OpenCV Mat with a 16-bit floating point pixel format.

The cuFFT library uses floating point data types as inputs, but I am not sure if they have to be in a C/C++ floating point array or if they can simply be passed in as an OpenCV Mat with a floating point pixel format.

If anyone knows about this detail, or knows of a simple way to pass images into cuFFT programs please let me know.

cufft expects the data to be in an array, not a OpenCV object.

http://docs.nvidia.com/cuda/cufft/index.html#abstract