Performance study of AVX instructions for the FDTD method

Abstract

This paper studies a novel acceleration technique for the FDTD method using new Intel instruction extensions - Intel Advanced Vector Extensions (AVX). We compare two configurations of FDTD algorithm (one enabled with AVX and a second without AVX) and present a performance evolution of the two settings. The numerical results show that the FDTD with AVX gives over 7 times speedup ratio compared to the FDTD without AVX.