Performance Tuning of a Parallel 3-D FFT Package OpenFFT

TL;DR

This paper presents performance analysis and auto-tuning of the OpenFFT package, optimizing communication strategies for large-scale 3-D FFT computations across various parallel computing platforms.

Contribution

It introduces six communication methods and an auto-tuning mechanism to enhance OpenFFT's performance on diverse hardware and scales.

Findings

Optimized communication methods improve performance.

Auto-tuning selects best method at runtime.

OpenFFT outperforms some state-of-the-art packages.

Abstract

The fast Fourier transform (FFT) is a primitive kernel in numerous fields of science and engineering. OpenFFT is an open-source parallel package for 3-D FFTs, built on a communication-optimal domain decomposition method for achieving minimal volume of communication. In this paper, we analyze and tune the performance of OpenFFT, paying a particular attention to tuning of communication that dominates the run time of large-scale calculations. We first analyze its performance on different machines for an understanding of the behaviors of the package and machines. Based on the performance analysis, we develop six communication methods for performing communication with the aim of covering varied calculation scales on a variety of computational platforms. OpenFFT is then augmented with an auto-tuning of communication to select the best method in run time depending on their performance. Numerical results demonstrate that the optimized OpenFFT is able to deliver relatively good performance in comparison with other state-of-the-art packages at different computational scales on a number of parallel machines.