Particle-in-Cell Laser-Plasma Simulation on Xeon Phi Coprocessors

TL;DR

This paper presents optimized Particle-in-Cell simulation code for Xeon Phi coprocessors, achieving significant speedups over CPUs through data locality, parallelization, and vectorization improvements.

Contribution

It demonstrates effective porting and optimization techniques for Particle-in-Cell plasma simulations on Xeon Phi architecture, resulting in substantial performance gains.

Findings

3.75x speedup on CPU after optimization

7.5x speedup on Xeon Phi after optimization

Xeon Phi achieves 1.6x speedup over CPU on real laser ion acceleration problem

Abstract

This paper concerns development of a high-performance implementation of the Particle-in-Cell method for plasma simulation on Intel Xeon Phi coprocessors. We discuss suitability of the method for Xeon Phi architecture and present our experience of porting and optimization of the existing parallel Particle-in-Cell code PICADOR. Direct porting with no code modification gives performance on Xeon Phi close to 8-core CPU on a benchmark problem with 50 particles per cell. We demonstrate step-by-step application of optimization techniques such as improving data locality, enhancing parallelization efficiency and vectorization that leads to 3.75 x speedup on CPU and 7.5 x on Xeon Phi. The optimized version achieves 18.8 ns per particle update on Intel Xeon E5-2660 CPU and 9.3 ns per particle update on Intel Xeon Phi 5110P. On a real problem of laser ion acceleration in targets with surface grating that requires a large number of macroparticles per cell the speedup of Xeon Phi compared to CPU is 1.6 x.