Accelerated Evaluation of Box Splines via a Parallel Inverse FFT

Abstract

Box splines are a multivariate extension of uniform univariate B-splines. Direct evaluation of a box spline basis function can he difficult but they have a relatively simple Fourier transform and can therefore be evaluated with an inverse FFT. Symmetry recursive evaluation of the coefficients, and parallelization can be used to improve absolute performance. A windowing function can also he used to reduce truncation artifacts. We explore all these options in the context of a high-performance parallel implementation. Our goal is the provision of an empirical touchstone for the inverse FFT evaluation of box spline basis functions, for eventual application to forward projection (splat-based) volume rendering.