数据与计算发展前沿 ›› 2025, Vol. 7 ›› Issue (6): 158-169.

CSTR: 32002.14.jfdc.CN10-1649/TP.2025.06.015

doi: 10.11871/jfdc.issn.2096-742X.2025.06.015

• 技术与应用 • 上一篇    下一篇

生命体系量子力学力场程序在新一代国产超算的千万核并行实现

郭剑萍1(),毛瑞超1,张邯1,张勇2,高军1,*()   

  1. 1.华中农业大学，信息学院，湖北 武汉 430070
    2.山东大学，青岛理论与计算科学研究院，山东 青岛 266237
  • 收稿日期:2025-02-20 出版日期:2025-12-20 发布日期:2025-12-17
  • 通讯作者: 高军
  • 作者简介:郭剑萍，华中农业大学，博士研究生，主要研究方向为量子力学力场程序开发，光合体系能量传递研究。
    本文承担工作为：模型设计，模型算法实现。
    GUO Jianping is a Ph.D. candidate at Huazhong Agricultural University. Her main research interests include the development of quantum mechanics force field programs and the study of energy transfer in photosynthetic systems.
    In this paper, she is mainly responsible for model design and the implementation of the model algorithms.
    E-mail: guojianping@webmail.hzau.edu.cn|高军，华中农业大学，博士生导师，主要研究方向为量子生物学与高性能并行计算等。
    本文承担工作为：指导优化模型和模型设计。
    GAO Jun is a Ph.D. supervisor at Huazhong Agricultural University. His main research interests include quantum biology, high-performance parallel computing, etc.
    In this paper, he is mainly responsible for providing guidance on model design and optimization.
    E-mail: gaojun@mail.hzau.edu.cn
  • 基金资助:
    国家重点研发计划(2017YFB0203405)

Parallel Implementation of the Quantum Mechanics Force Field Program of Biosystem Using Tens of Millions of Cores on the New-Generation Sunway Supercomputer

GUO Jianping1(),MAO Ruichao1,ZHANG Han1,ZHANG Yong2,GAO Jun1,*()   

  1. 1. College of Informatics, Huazhong Agricultural University, Wuhan, Hubei 430070, China
    2. Qingdao Institute of Theoretical and Computational Sciences, Shandong University, Qingdao, Shandong 266237, China
  • Received:2025-02-20 Online:2025-12-20 Published:2025-12-17
  • Contact: GAO Jun

RichHTML

2

PDF

0

摘要：

【目的】旨在新一代国产超算上实现量子力学力场方法X-Pol的千万核高效并行。【方法】开发了PyXPOL程序，该程序是量子力学力场方法的Python实现，采用MPI+Athread两级并行构架。通过引入邻居列表方法，实现了非键相互作用的并行计算。采用计算量S型排序优化方法，实现了负载均衡。结合SW26010-Pro处理器的特点，对X-Pol模型中的量子化学计算程序进行了优化。【结果】结果表明，PyXPOL程序实现了千万核规模的弱可扩展性计算。千万核相较于5万核的并行效率可达75%。此外，PyXPOL程序实现了十七万原子生物体系的大规模并行计算，300万核相对于5万核并行效率达82%。【结论】PyXPOL程序在新一代国产超算上实现了量子力学力场方法X-Pol的高效并行，显著提高了大规模生物分子体系的计算效率，为量子力学力场方法在大规模并行计算中的应用提供了有力工具。

关键词: 显性极化（X-Pol）模型, 两级并行, 量子化学力场

Abstract:

[Objective] This paper aims to achieve efficient parallelization of the explicit polarization (X-Pol) model, a quantum mechanical force field method, on a new generation of domestic supercomputers. [Methods] The PyXPOL program, a Python implementation of the quantum mechanical force field method, is developed. It adopts a two-level parallel architecture of MPI+Athread. By introducing the neighbor list method, the parallel computation of non-bonded interactions is achieved. An S-shaped sorting optimization method for computational workload is employed to achieve load balancing. Considering the characteristics of the SW26010-Pro processor, the quantum chemical calculation program in the X-Pol model is optimized. [Results] The results indicate that the PyXPOL program demonstrates weak scalability on ten million cores. Compared with the calculation using 50,050 cores, the parallel efficiency of the ten-million-core calculation can reach 75%. Moreover, it also achieves large-scale parallel calculations for a biological system with 170,788 atoms. The parallel efficiency of 3 million cores compared to 50,050 cores reaches 82%. [Conclusions] The PyXPOL program achieves efficient parallelization of the X-Pol quantum mechanical force field method on the new generation of domestic supercomputers, significantly improving the computational efficiency for large-scale biological molecular systems. It provides a powerful tool for the application of quantum mechanical force field methods in large-scale parallel computing.

Key words: Explicit polarization (X-Pol) model, two-level parallel, quantum mechanical force field