A High-Performance Implementation of Atomistic Spin Dynamics Simulations on x86 CPUs

TL;DR

This paper presents a highly optimized implementation of atomistic spin dynamics simulations on x86 CPUs, significantly improving performance by leveraging GEMM routines and kernel fusion to efficiently compute dynamical correlation functions.

Contribution

The work introduces a novel, optimized approach using GEMM and kernel fusion to accelerate spin dynamics simulations on standard CPUs, enabling larger system analysis.

Findings

Performance improvements of 44% - 71% over existing methods

Efficient utilization of hardware through GEMM-based calculations

Enabling simulation of larger spin systems with reduced computational cost

Abstract

Atomistic spin dynamics simulations provide valuable information about the energy spectrum of magnetic materials in different phases, allowing one to identify instabilities and the nature of their excitations. However, the time cost of evaluating the dynamical correlation function $S(\mathbf{q}, t)$ increases quadratically as the number of spins $N$, leading to significant computational effort, making the simulation of large spin systems very challenging. In this work, we propose to use a highly optimized general matrix multiply (GEMM) subroutine to calculate the dynamical spin-spin correlation function that can achieve near-optimal hardware utilization. Furthermore, we fuse the element-wise operations in the calculation of $S(\mathbf{q}, t)$ into the in-house GEMM kernel, which results in further performance improvements of 44\% - 71\% on several relatively large lattice sizes when compared to the implementation that uses the GEMM subroutine in OpenBLAS, which is the state-of-the-art open source library for Basic Linear Algebra Subroutine (BLAS).