Automatic Tuning of Computation-Communication Overlap for Parallel 1-D FFT

摘要

This paper presents a method for the automatic tuning of the computation-communication overlap for a parallel one-dimensional fast Fourier transform (FFT). Parallel FFTs require intensive all-to-all communication, which affects their performance. Therefore, in the present study, an automatic tuning facility for selecting the optimal parameters of the computation-communication overlap, the radices, and the block size is implemented. The performance results demonstrate that the proposed implementation of a parallel one-dimensional FFT with automatic tuning provides superior performance.