CUDA平台的非结构网格DSMC并行算法

王志; 王学德

doi:10.13224/j.cnki.jasp.20250348

CUDA平台的非结构网格DSMC并行算法

doi: 10.13224/j.cnki.jasp.20250348

王志,
王学德^*, ,

南京理工大学能源与动力工程学院，南京 210094

详细信息

作者简介:
王志（2000-），男，硕士生，主要从事稀薄气体动力学研究。E-mail：wangzhinjust@qq.com

通讯作者:
王学德（1977-），男，副教授，博士，主要从事稀薄气体动力学研究。E-mail：wangxuede2000@njust.edu.cn

中图分类号: V411.4
计量
- 文章访问数: 234
- HTML浏览量: 140
- PDF量: 14
- 被引次数: 0
出版历程
- 收稿日期: 2025-07-23
- 网络出版日期: 2025-11-27

Parallel DSMC algorithm for unstructured grids on CUDA platform

WANG Zhi,
WANG Xuede^{*
, ,}

School of Energy and Power Engineering，Nanjing University of Science and Technology，Nanjing 210094，China

摘要

摘要:
为提升稀薄气体流动模拟的计算效率，针对非结构网格在复杂几何建模中的适应性特点，构建了一种基于统一计算设备架构（CUDA）平台的非结构网格直接模拟蒙特卡洛（DSMC）并行算法。建立通用的DSMC求解框架，设计高效的网格搜索与粒子追踪机制，并在CUDA平台上对流场初始化、粒子推进、碰撞、排序及结果取样等核心模块进行了并行优化。通过合理划分线程与存储资源，解决了图形处理器（GPU）计算中数据一致性、线程发散与负载不均衡等问题。以超声速圆柱绕流为算例进行验证，结果表明：该算法在单卡GPU上整体加速比达到52.5；基于网格并行策略的各模块取得了24.7～53.7的加速比，基于分子并行策略的各模块取得了47～68倍的加速比。该算法显著提升了非结构网格DSMC模拟的并行性能，为高马赫数稀薄气体流动的高效数值模拟提供了一种可行方案。
- 直接模拟蒙特卡洛（DSMC）方法 /
- 非结构网格 /
- 图形处理器（GPU）并行计算 /
- 统一计算设备架构（CUDA）平台 /
- 稀薄气体动力学
Abstract:
To enhance the computational efficiency of rarefied gas flow simulations, a compute unified device architecture (CUDA) platform-based parallel direct simulation Monte Carlo (DSMC) algorithm for unstructured grids was developed considering the adaptability of unstructured meshes to complex geometries. A general DSMC framework was established, and efficient grid-search and particle-tracking mechanisms were designed. The core computational modules—including flow-field initialization, particle movement, collision, sorting, and sampling—were parallelized and optimized on the CUDA platform. By rationally allocating threads and memory resources, issues such as data consistency, thread divergence, and load imbalance in graphics processing unit (GPU) computing were effectively addressed. The algorithm was validated through simulation of supersonic flow over a circular cylinder. Results showed that the overall speedup on a single GPU reached 52.5; the speedups of individual modules based on the grid-parallel strategy ranged from 24.7 to 53.7, while those based on the molecule-parallel strategy ranged from 47 to 68. The proposed algorithm significantly improved the parallel performance of DSMC simulations on unstructured grids, thus providing a feasible approach for efficient numerical simulation of high-Mach-number rarefied gas flows.

HTML

图 1 网格搜索示意图

Figure 1. Schematic diagram of grid search

下载: 全尺寸图片幻灯片

图 2 GPU编程模型

Figure 2. GPU programing model

下载: 全尺寸图片幻灯片

图 3 线程与网格映射关系图

Figure 3. Thread and grid allusion relationship diagram

下载: 全尺寸图片幻灯片

图 4 粒子运动模块串行算法流程图

Figure 4. Flowchart of serial algorithm for molecular motion module

下载: 全尺寸图片幻灯片

图 5 粒子运动模块GPU并行算法流程图

Figure 5. Molecular motion module GPU parallel algorithm flowchart

下载: 全尺寸图片幻灯片

图 6 串行算法并行算法密度和温度比较

Figure 6. Comparison of density and temperature between Serial and Parallel Algorithms

下载: 全尺寸图片幻灯片

图 7 沿驻点线密度和温度比较

Figure 7. Comparison of density and temperature along the stagnation line

下载: 全尺寸图片幻灯片

图 8 圆柱壁面特征值比较

Figure 8. Comparison of wall characteristics for cylindrical walls

下载: 全尺寸图片幻灯片

图 9 各子模块及并行模块整体加速比随计算规模的变化

Figure 9. Changes in acceleration of parallel modules and overall with computing scale

下载: 全尺寸图片幻灯片

表 1 各模块可并行部分比例

Table 1. Proportion of parallelizable parts for each module

模块名称	比例 $ Q $/%	核心可并行部分
模块名称	比例 $ Q $/%	子模块名称	子模块功能
初始化	92.6	Volume	网格面积计算
		Subcell	子网格编号计算
		Reassign	取样赋值初始化
粒子运动	23.6	Move	粒子坐标计算
粒子排序	94.1	Index	粒子排序
粒子碰撞	10.1	Colnsel	碰撞对数目计算
取样	100	Sample	网格信息取样

下载: 导出CSV

表 2 算法测试环境

Table 2. Algorithm testing environment

类别	名称	配置
硬件环境	CPU	Intel（R） Core（TM） i7-14700KF
	GPU	NVIDIA GeForce RTX 4070 SUPER
	内存	48 G DDR5
软件环境	操作系统	Ubuntu 22.04
软件环境	编译器	NVFORTRAN 25.3

下载: 导出CSV

参考文献(25)

[1]	RADTKE J, KEBSCHULL C, STOLL E. Interactions of the space debris environment with mega constellations: using the example of the OneWeb constellation[J]. Acta Astronautica, 2017, 131: 55-68. doi: 10.1016/j.actaastro.2016.11.021
[2]	BIRD G A. Molecular gas dynamics and the direct simulation of gas flows[M]. Oxford: Oxford University Press, 1994.
[3]	许啸, 王学德, 谭俊杰. 近空间高超声速流场通量分裂型DSMC-IP方法[J]. 航空动力学报, 2015, 30(2): 315-323. XU Xiao, WANG Xuede, TAN Junjie. Flux splitting DSMC-IP method in near space hypersonic flow field[J]. Journal of Aerospace Power, 2015, 30(2): 315-323. (in Chinese XU Xiao, WANG Xuede, TAN Junjie. Flux splitting DSMC-IP method in near space hypersonic flow field[J]. Journal of Aerospace Power, 2015, 30(2): 315-323. (in Chinese)
[4]	王保国, 黄伟光, 钱耕, 等. 再入飞行中DSMC与Navier-Stokes两种模型的计算与分析[J]. 航空动力学报, 2011, 26(5): 961-976. WANG Baoguo, HUANG Weiguang, QIAN Geng, et al. Computation and analysis of DSMC and Navier-Stokes model in reentry flight[J]. Journal of Aerospace Power, 2011, 26(5): 961-976. (in Chinese WANG Baoguo, HUANG Weiguang, QIAN Geng, et al. Computation and analysis of DSMC and Navier-Stokes model in reentry flight[J]. Journal of Aerospace Power, 2011, 26(5): 961-976. (in Chinese)
[5]	朱荣丽, 曹义华, 李栋, 等. 高超声速飞行器复杂流场过渡区DSMC数值模拟的一种新方案[J]. 宇航学报, 2006, 27(2): 167-171. ZHU Rongli, CAO Yihua, LI Dong, et al. A new version of hypersonic vehicle numerical simulation in direct simulation of Monte Carlo method of three dimensional flow in transitional regime[J]. Journal of Astronautics, 2006, 27(2): 167-171. (in Chinese ZHU Rongli, CAO Yihua, LI Dong, et al. A new version of hypersonic vehicle numerical simulation in direct simulation of Monte Carlo method of three dimensional flow in transitional regime[J]. Journal of Astronautics, 2006, 27(2): 167-171. (in Chinese)
[6]	李中华, 李志辉, 李海燕, 等. 过渡流区N-S/DSMC耦合计算研究[J]. 空气动力学学报, 2013, 31(3): 282-287. LI Zhonghua, LI Zhihui, LI Haiyan, et al. Study on N-S/DSMC coupling calculation in transition flow region[J]. Acta Aerodynamica Sinica, 2013, 31(3): 282-287. (in Chinese LI Zhonghua, LI Zhihui, LI Haiyan, et al. Study on N-S/DSMC coupling calculation in transition flow region[J]. Acta Aerodynamica Sinica, 2013, 31(3): 282-287. (in Chinese)
[7]	王保国, 李学东, 刘淑艳. 高温高速稀薄流的DSMC算法与流场传热分析[J]. 航空动力学报, 2010, 25(6): 1203-1220. WANG Baoguo, LI Xuedong, LIU Shuyan. DSMC algorithm and heat transfer analysis of high temperature and high velocity rarefied gas flow[J]. Journal of Aerospace Power, 2010, 25(6): 1203-1220. (in Chinese WANG Baoguo, LI Xuedong, LIU Shuyan. DSMC algorithm and heat transfer analysis of high temperature and high velocity rarefied gas flow[J]. Journal of Aerospace Power, 2010, 25(6): 1203-1220. (in Chinese)
[8]	梁杰, 阎超, 李志辉, 等. 稀薄过渡流区横向喷流干扰效应数值模拟研究[J]. 空气动力学学报, 2013, 31(1): 27-33. LIANG Jie, YAN Chao, LI Zhihui, et al. Numerical investigation of lateral jet interaction effects in rarefied transition flow regime[J]. Acta Aerodynamica Sinica, 2013, 31(1): 27-33. (in Chinese LIANG Jie, YAN Chao, LI Zhihui, et al. Numerical investigation of lateral jet interaction effects in rarefied transition flow regime[J]. Acta Aerodynamica Sinica, 2013, 31(1): 27-33. (in Chinese)
[9]	王保国, 李耀华, 钱耕. 四种飞行器绕流的三维DSMC计算与传热分析[J]. 航空动力学报, 2011, 26(1): 1-20. WANG Baoguo, LI Yaohua, QIAN Geng. Three-dimensional DSMC calculation and heat transfer analysis of four capsules for hypersonic rarefied conditions[J]. Journal of Aerospace Power, 2011, 26(1): 1-20. (in Chinese WANG Baoguo, LI Yaohua, QIAN Geng. Three-dimensional DSMC calculation and heat transfer analysis of four capsules for hypersonic rarefied conditions[J]. Journal of Aerospace Power, 2011, 26(1): 1-20. (in Chinese)
[10]	蔡国飙, 刘世俭, 王慧玉, 等. 真空羽流场的DSMC并行数值模拟[J]. 航空动力学报, 1999, 14(2): 113-118. CAI Guobiao, LIU Shijian, WANG Huiyu, et al. DSMC parallel numerical simulation of vacuum plume flow field[J]. Journal of Aerospace Power, 1999, 14(2): 113-118. (in Chinese CAI Guobiao, LIU Shijian, WANG Huiyu, et al. DSMC parallel numerical simulation of vacuum plume flow field[J]. Journal of Aerospace Power, 1999, 14(2): 113-118. (in Chinese)
[11]	杜松蔚, 王学德. 一种基于双重空间网格结合的高效DSMC实现方法[J]. 航空动力学报, 2022, 37(9): 1846-1854. DU Songwei, WANG Xuede. High-efficiency DSMC implement method based on combination of dual spatial grids[J]. Journal of Aerospace Power, 2022, 37(9): 1846-1854. (in Chinese DU Songwei, WANG Xuede. High-efficiency DSMC implement method based on combination of dual spatial grids[J]. Journal of Aerospace Power, 2022, 37(9): 1846-1854. (in Chinese)
[12]	张哲铭. 非结构网格DSMC的大规模并行计算研究[D]. 成都: 电子科技大学, 2022. ZHANG Zheming. Research on large-scale parallel computing of unstructured grid DSMC applications[D]. Chengdu: University of Electronic Science and Technology of China, 2022. (in Chinese ZHANG Zheming. Research on large-scale parallel computing of unstructured grid DSMC applications[D]. Chengdu: University of Electronic Science and Technology of China, 2022. (in Chinese)
[13]	LUSK E, HUSS S, SAPHIR B, et al. MPI: a message-passing interface standard[J]. International Journal of Supercomputer Applications, 2009, 8(3/4): 623.
[14]	王学德. 高超声速稀薄气流非结构网格DSMC及并行算法研究[D]. 南京: 南京航空航天大学, 2006. WANG Xuede. DSMC method on unstructured grids for hypersonic rarefied gas flow and its parallelization[D]. Nanjing: Nanjing University of Aeronautics and Astronautics, 2006. (in Chinese WANG Xuede. DSMC method on unstructured grids for hypersonic rarefied gas flow and its parallelization[D]. Nanjing: Nanjing University of Aeronautics and Astronautics, 2006. (in Chinese)
[15]	PassMark Software. Top CPU performance to date [EB/OL]. (2025-06-28)[2025-07-06]. https://www.cpubenchmark.net/year-on-year.html.
[16]	NVIDIA. NVIDIA CUDA toolkit documentation [EB/OL]. (2025-06-09) [2025-07-06]. https://developer.nvidia.com/cuda-faq.
[17]	CHEN Bojun. High performance parallel computing of DSMC method based on CUDA[EB/OL]. (2009-12-21) [2025-07-06]. https://wenku.baidu.com/view/f8f381ccbb4cf7ec4afed0cd.html?_wkts_=1763396744283
[18]	严立, 戴欣怡, 陈佳洛, 等. 基于计算统一设备架物Fortran的直接模拟蒙特卡洛方法并行优化[J]. 上海交通大学学报, 2013, 47(8): 1198-1204. YAN Li, DAI Xinyi, CHEN Jialuo, et al. Parallel optimization of direct simulation Monte Carlo method using compute unified device architecture fortran[J]. Journal of Shanghai Jiao Tong University, 2013, 47(8): 1198-1204. (in Chinese YAN Li, DAI Xinyi, CHEN Jialuo, et al. Parallel optimization of direct simulation Monte Carlo method using compute unified device architecture fortran[J]. Journal of Shanghai Jiao Tong University, 2013, 47(8): 1198-1204. (in Chinese)
[19]	邱天林. 直角坐标网格下DSMC方法的GPU并行研究[D]. 南京: 南京航空航天大学, 2017. QIU Tianlin. Parallel research on GPU-based DSMC method using Cartesian grid[D]. Nanjing: Nanjing University of Aeronautics and Astronautics, 2017. (in Chinese QIU Tianlin. Parallel research on GPU-based DSMC method using Cartesian grid[D]. Nanjing: Nanjing University of Aeronautics and Astronautics, 2017. (in Chinese)
[20]	陈飞同, 王学德. 三维复杂界面非结构网格N-S/DSMC耦合方法[J]. 航空动力学报, 2025, 40(9): 20240329. CHEN Feitong, WANG Xuede. N-S/DSMC coupling method using three-dimensional unstructured mesh for complex interfaces[J]. Journal of Aerospace Power, 2025, 40(9): 20240329. (in Chinese CHEN Feitong, WANG Xuede. N-S/DSMC coupling method using three-dimensional unstructured mesh for complex interfaces[J]. Journal of Aerospace Power, 2025, 40(9): 20240329. (in Chinese)
[21]	AMDAHL G M. Validity of the single processor approach to achieving large scale computing capabilities[C]//Proceedings of the Spring Joint Computer Conference. New York: Association for Computing Machinery, 1967: 483-485.
[22]	王学德, 伍贻兆, 夏健, 等. 三维非结构网格DSMC并行算法及应用研究[J]. 宇航学报, 2007, 28(6): 1500-1505. WANG Xuede, WU Yizhao, XIA Jian, et al. A parallel algorithm of 3D unstructured DSMC method and its application[J]. Journal of Astronautics, 2007, 28(6): 1500-1505. (in Chinese WANG Xuede, WU Yizhao, XIA Jian, et al. A parallel algorithm of 3D unstructured DSMC method and its application[J]. Journal of Astronautics, 2007, 28(6): 1500-1505. (in Chinese)
[23]	白洪涛. 基于GPU的高性能并行算法研究[D]. 长春: 吉林大学, 2010. BAI Hongtao. Research on high performance parallel algorithms based on GPU[D]. Changchun: Jilin University, 2010. (in Chinese BAI Hongtao. Research on high performance parallel algorithms based on GPU[D]. Changchun: Jilin University, 2010. (in Chinese)
[24]	傅萌. 基于CPU+GPU并行计算加速的电力系统状态估计技术[D]. 南京: 东南大学, 2023. FU Meng. Power system state estimation technology based on CPU+GPU parallel computing acceleration[D]. Nanjing: Southeast University, 2023. (in Chinese FU Meng. Power system state estimation technology based on CPU+GPU parallel computing acceleration[D]. Nanjing: Southeast University, 2023. (in Chinese)
[25]	鞠鹏飞, 宁方飞. GPU平台上的叶轮机械CFD加速计算[J]. 航空动力学报, 2014, 29(5): 1154-1162. JU Pengfei, NING Fangfei. Accelerated CFD computing of turbomachinery on GPU platform[J]. Journal of Aerospace Power, 2014, 29(5): 1154-1162. (in Chinese JU Pengfei, NING Fangfei. Accelerated CFD computing of turbomachinery on GPU platform[J]. Journal of Aerospace Power, 2014, 29(5): 1154-1162. (in Chinese)