Application of GPU parallelization for non-uniform Fourier interpolation

Kai Zhuang, Daniel O. Trad

We look at the computation of the naive DFT interpolation on GPUs in an attempt to solve two main issues commonly seen in the application of 5D interpolation, compute times and far offset approximations. Our results show that the implementation on GPUs help solve both problems, although the computation of the naive DFT is significantly more expensive than the FFT, the switch to GPU computation for the dataflow decreases the runtime compared to standard CPU implementation of DFT interpolation.