Circuits, Systems, and Signal Processing
Parallel functional languages and compilers
Computational frameworks for the fast Fourier transform
Computational frameworks for the fast Fourier transform
Retire Fortran?: a debate rekindled
Communications of the ACM
Fast digit-reversal algorithms on a shared-memory machine
Parallel Computing
Scientific Programming - Special issue: high-performance functional computing
An Adaptation of the Fast Fourier Transform for Parallel Processing
Journal of the ACM (JACM)
Hi-index | 0.00 |
We show that each member of a well-known family of radix-r FFTs can be expressed as the composition of functions chosen from a set of four "basis" functions. Thus, a complete library of radix-r FFTs can be developed by programming the four basis functions and composing them to obtain various FFTs. We illustrate this by developing implementations in the functional language Sisal. We compare performance of these FFTs on a Cray C-90. Based on observations of these results we develop an implementation of an FFT of size 2k which is seemingly optimal for the Cray C-90. We discuss language features of particular benefit for programming FFTs and suggest some enhancements that would further streamline the programming.