Contemporary View of Fft Algorithms

A large number of fast Fourier transform (FFT) algorithms exist for efficient computation of the discrete Fourier transform (DFT). In most previous benchmarking efforts, only the computation speed or the operations count have been used for assessing the efficiency of these algorithms. In this paper we provide a comprehensive comparison of several contemporary FFT algorithms on state-of-the-art processors. The criterion used are the operations count, CPU time, memory usage, processor, and compiler. The processors that were evaluated include the DEC Alpha, Intel Pentium Pro and Sun UltraSparc. Preliminary work on quantifying the effects of compilers on these algorithms is also presented. The most efficient algorithm was found to be the fast Hartley transform. For some algorithms, the results presented here differ significantly from theoretical predictions, thereby stressing the need for such contemporary benchmarks.