We consider the fast Fourier transform at nonequispaced nodes (NFFT) and give detailed information on the time and memory requirements of its building blocks. This manuscript reviews the state of the art approaches and focuses within the most successful scheme on the computational most involved part.
Beside a rigorous derivation of an lookup table technique, we compare a wide range of precomputation schemes which lead to substantially different computation times of the NFFT.
In particular, we show how to balance accuracy, memory usage, and computation time.