SIMD Vectorization for the Lennard-Jones Potential with AVX2 and AVX-512 instructions

TL;DR

This paper explores SIMD vectorization of Lennard-Jones force calculations using AVX2 and AVX-512, demonstrating significant performance improvements and analyzing data layout and optimization techniques across different Intel architectures.

Contribution

It provides a detailed analysis of data layout choices and optimization strategies for vectorizing Lennard-Jones force calculations with AVX2 and AVX-512.

Findings

AoS with padding outperforms SoA in vectorization.

Vectorization improves performance by up to 42%.

Performance varies across architectures and vectorization methods.

Abstract

This work describes the SIMD vectorization of the force calculation of the Lennard-Jones potential with Intel AVX2 and AVX-512 instruction sets. Since the force-calculation kernel of the molecular dynamics method involves indirect access to memory, the data layout is one of the most important factors in vectorization. We find that the Array of Structures (AoS) with padding exhibits better performance than Structure of Arrays (SoA) with appropriate vectorization and optimizations. In particular, AoS with 512-bit width exhibits the best performance among the architectures. While the difference in performance between AoS and SoA is significant for the vectorization with AVX2, that with AVX-512 is minor. The effect of other optimization techniques, such as software pipelining together with vectorization, is also discussed. We present results for benchmarks on three CPU architectures: Intel Haswell (HSW), Knights Landing (KNL), and Skylake (SKL). The performance gains by vectorization are about 42\% on HSW compared with the code optimized without vectorization. On KNL, the hand-vectorized codes exhibit 34\% better performance than the codes vectorized automatically by the Intel compiler. On SKL, the code vectorized with AVX2 exhibits slightly better performance than that with vectorized AVX-512.