Vectorization Study for FFT Algorithm Based on ARM SVE
LI Feng-jiao,GU Nai-jie,QI Dong-sheng,SU Jun-jie
(School of Computer Scienceand Technology,University of Science and Technology of China,Hefei 230027,China)(Anhui Province Key Laboratory of Computing and Communication Software,University of Science and Technology of China,Hefei 230027,China)
Abstract：Fast Fourier Transform (FFT) is animportant research tool in signal processing,image processing,etc. Scalable Vector Extension (SVE) is the next-generation SIMD instruction set for ARMv8-A architecture. It supports a vector register length between 128 and 2048 bits and Vector Length Agnostic (VLA) programming model,which allows it well vector parallelism and software portability. SVEis designed for high performance computing,machine learning,etc.The current research of FFT algorithm based on ARM SVE has not fully utilized the architecture features and computing resources.This paper focuses on 1d complex FFT of power of 2 and improves FFT algorithm based on SVE features of predicate-driven loop control,non-linear data accesses,complex operations,etc.The result shows that,this algorithm has significant performance improvement compared to FFTW library for ARMNEON. Appointing the vector length as 1024 bits,the average improvementis 5.83 times,and the highest improvement can reach to 9.22 times.
李凤娇,顾乃杰,齐东升,苏俊杰. 基于ARM SVE的FFT算法向量化研究[J]. 小型微型计算机系统, 2022, 43(10): 2017-2021.
LI Feng-jiao,GU Nai-jie,QI Dong-sheng,SU Jun-jie. Vectorization Study for FFT Algorithm Based on ARM SVE. Journal of Chinese Computer Systems, 2022, 43(10): 2017-2021.