In this paper, we present a new parallel implementation of a fully three-dimensional particle-in-cell solver for the numerical simulation of plasma dynamics. The interaction of injected particles with the background plasma in a rectangular domain is modeled in Cartesian coordinates. The code employs a hybrid approach: the ion component is described kinetically, while the electron component is governed by magnetohydrodynamic equations. Written in Fortran and parallelized using MPI, the solver is designed for high-performance computing. We analyze its computational performance, comparing vectorized and non-vectorized implementations, and assess memory usage. Additionally, we present a performance evaluation of the code on AVX2- and AVX512-compatible processors. The paper is organized as follows. Section 1 contains the introduction. Section 2 is devoted to the mathematical model of the plasma physics problem. In Sect. 3, we briefly describe the numerical method. Section 4 contains the results of the numerical simulation of plasma confinement in a magnetic trap. In Sect. 5, we provide the performance analysis method and the comparison of the performance between unvectorized and vectorized codes. The conclusion summarizes the study and points directions for further work.

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Performance Analysis of a New Three-Dimensional Plasma Physics Solver

  • Igor Chernykh,
  • Igor Kulikov,
  • Ivan Chernoshtanov,
  • Tatyana Liseykina,
  • Vitaly Vshivkov,
  • Marina Boronina

摘要

In this paper, we present a new parallel implementation of a fully three-dimensional particle-in-cell solver for the numerical simulation of plasma dynamics. The interaction of injected particles with the background plasma in a rectangular domain is modeled in Cartesian coordinates. The code employs a hybrid approach: the ion component is described kinetically, while the electron component is governed by magnetohydrodynamic equations. Written in Fortran and parallelized using MPI, the solver is designed for high-performance computing. We analyze its computational performance, comparing vectorized and non-vectorized implementations, and assess memory usage. Additionally, we present a performance evaluation of the code on AVX2- and AVX512-compatible processors. The paper is organized as follows. Section 1 contains the introduction. Section 2 is devoted to the mathematical model of the plasma physics problem. In Sect. 3, we briefly describe the numerical method. Section 4 contains the results of the numerical simulation of plasma confinement in a magnetic trap. In Sect. 5, we provide the performance analysis method and the comparison of the performance between unvectorized and vectorized codes. The conclusion summarizes the study and points directions for further work.