Methods and High Performance Computing Optimizations of 5D interpolation

Kai Zhuang, Daniel O. Trad

We implement 5D interpolation with the Non-Uniform exact location DFT operator on GPUs to address binning-related errors in far offset approximations without compromising computational efficiency. This approach aims to accelerate the relatively slow Non-Uniform DFT operator using GPU computation. Our findings indicate that the GPU implementation effectively resolves both issues. Although the computation of the Non-Uniform DFT is more expensive than the FFT, transitioning to GPU computation significantly reduces runtime compared to the standard CPU implementation of DFT interpolation.