Functions exported from NVIDIA libraries like CUFFT or CUBLAS are not like functions from other shared libraries. Because they run on one or more GPUs, they have specific execution requirements that multi-threaded development environments like LabVIEW do not support w/out additional work.
Recently, a beta of the GPU Analysis Toolkit was made available for download from http://www.ni.com/beta. This toolkit includes wrappers for selecting and managing NVIDIA GPU devices and for calling many of the functions in NVIDIA's CUDA Toolkit, CUBLAS and CUFFT libraries. An example of a multi-channel 1D FFT is included and shows how to configure and call functions from the CUFFT library.
I updated the NILabs LabVIEW GPU Computing page with a brief introduction to the toolkit and a snapshot of how to request beta access.
