结构材料辐照损伤的分子动力学程序GPU并行化及优化

计算物理 ›› 2017, Vol. 34 ›› Issue (4): 461-467.

结构材料辐照损伤的分子动力学程序GPU并行化及优化

祁美玲, 杨琼, 王苍龙, 田园, 杨磊

中国科学院近代物理研究所, 兰州 730000

收稿日期:2016-11-10 修回日期:2016-12-16 出版日期:2017-07-25 发布日期:2017-07-25
通讯作者: 王苍龙,E-mail:clwang@impcas.ac.cn
作者简介:祁美玲(1986-),女,硕士,助理研究员,从事高性能计算研究;E-mail:qimeiling@impcas.ac.cn
基金资助:
国家自然科学基金（11304324）及中国科学院战略性先导科技专项（XDA03030100）资助项目

Parallel Algorithm on GPU and Optimization for Molecular Dynamics Simulation of Irradiation Damage of Structure Materials

QI Meiling, YANG Qiong, WANG Canglong, TIAN Yuan, YANG Lei

Institute of Modern Physics, Chinese Academy of Sciences, Lanzhou 730000, China

Received:2016-11-10 Revised:2016-12-16 Online:2017-07-25 Published:2017-07-25

摘要/Abstract

摘要： 基于NIVIDIA公司的CUDA架构对结构材料辐照损伤的分子动力学程序在单个GPU上进行并行化，并对影响程序运行效率的相关因素进行分析和测试.经过一系列优化，当粒子数为两百万时，对比单CPU的执行时间，优化后的GPU程序其双精度加速比可达112倍，单精度加速比达到了三百倍，为后续扩展多GPU结构材料辐照损伤的分子动力学程序奠定基础.

关键词: 辐照效应, 分子动力学模拟, GPU, 并行算法

Abstract: We implemented parallel algorithms on single GPU for molecular dynamics simulation of irradiation damage of structure materials based on CUDA framework of NIVIDIA. The program was effectively optimized with analyzing related factors. Compared with a single CPU implementation, GPU implementation achieved 112 speedup in double precision, 300 speedup in single precision as number of particle is 2 000 000. The implementation is helpful for development on parallel algorithm on multi-GPU.

Key words: irradiation effects, molecular dynamics simulation, GPU, parallel algorithms

中图分类号:

TP391.9

祁美玲, 杨琼, 王苍龙, 田园, 杨磊. 结构材料辐照损伤的分子动力学程序GPU并行化及优化[J]. 计算物理, 2017, 34(4): 461-467.

QI Meiling, YANG Qiong, WANG Canglong, TIAN Yuan, YANG Lei. Parallel Algorithm on GPU and Optimization for Molecular Dynamics Simulation of Irradiation Damage of Structure Materials[J]. CHINESE JOURNAL OF COMPUTATIONAL PHYSICS, 2017, 34(4): 461-467.

参考文献

[1] 费辉,张云泉,王可,等. 基于GPU的分子动力学模拟并行化及实现[J]. 计算机科学, 2011, 38(9):276-278.
[2] VERLET L. Computer experiments on classical fluids. I. Thermodynamical properties of Lennard-Jones molecules[J]. Physical Review, 1967, 159(1):98-103.
[3] MAYO P, MESIROV J P, BOGHOSIAN B M. Parallel approaches to short range molecular dynamics simulations[C]//Proceedings of Conference on Supercomputing, New York, USA:ACM Press,1991:462-470.
[4] 刘青昆,滕人达,刘凤,等. 多核并行技术在分子动力学模拟中的应用[J]. 计算机工程与设计,2011, 32(10):3395.
[5] LIANG Jie, LI Kenli, SHI Lin, et al. Accelerating solidification process simulation for large-sized system of liquid metal atoms using GPU with CUDA[J]. Journal of Computational Physics, 2014, 257:521-535.
[6] HOU Qing, LI Min, ZHOU Yulu, et al. Molecular dynamics simulations with many-body potentials on multiple GPUs-The implementation, package and performance[J]. Comput Phys Comm, 2013, 184:2091-2101
[7] BROWNA W Michael, WANG Peng, PLIMPTON Steven J, et al. Implementing molecular dynamics on hybrid high performance computers-short range forces[J]. Comput Phys Comm, 2011, 182:898-911.
[8] QI J, LI K C, JIANG H, et al. GPU-accelerated DEM implementation with CUDA[J]. Int J Computational Science and Engineering, 2015,11(3):330-337.
[9] 徐骥,葛蔚,任瑛,等. Particle-mesh ewald (PME)算法的GPU加速[J]. 计算物理, 2010,27(4):548-554.
[10] 丁科,谭营. GPU通用计算及其在计算智能领域的应用[J]. 智能系统学报, 2015,10(1):1-11.
[11] 韩苗苗. 4H-SiC辐照损伤分子动力学模拟初步研究[D]. 哈尔滨:哈尔滨工程大学, 2013.
[12] RAPAPORT D C. Art of molecular dynamics simulation[M]. Cambridge University Press, 1995:1-30.
[13] 张邦维,胡望宇,舒小林. 嵌入原子方法理论及其在材料科学中的应用——原子尺度材料设计理论[M]. 长沙:湖南大学出版社,2002.
[14] 邹岩,杨志义,张凯龙. CUDA并行程序的内存访问优化技术研究[J]. 计算机测量与控制, 2009,17(12):2504-2506.
[15] SANDERS Jason, KANDRAT Edward. GPU高性能编程CUDA实战[M]. 聂雪军,译. 北京:机械工业出版社, 2011.