面向众核处理器的水动力学CFD并行计算探索

doi:10.19423/j.cnki.31-1561/u.2021.04.015

船舶 ›› 2021, Vol. 32 ›› Issue (04): 15-23.DOI: 10.19423/j.cnki.31-1561/u.2021.04.015

面向众核处理器的水动力学CFD并行计算探索

张亚英¹, 吴乘胜¹, 王建春¹, 刘宏斌²

1.中国船舶科学研究中心无锡214082;
2.国家超级计算无锡中心无锡 214072

收稿日期:2021-01-06 修回日期:2021-03-05 出版日期:2021-08-25 发布日期:2021-09-13
作者简介:张亚英（1994-）,男,硕士,助理工程师。研究方向：CFD并行计算与优化设计。吴乘胜（1976-）,男,博士,研究员。研究方向：船舶水动力学CFD应用与开发。王建春（1989-）,男,硕士,工程师。研究方向：CFD并行计算与优化设计。刘宏斌（1993-）,男,硕士,工程师。研究方向：高性能计算、CUDA程序优化。
基金资助:
水动力预研项目（编号：50907070201）

Multi-Core Processor Oriented Parallel Computation for CFD in Hydrodynamics

ZHANG Ya-ying¹, WU Cheng-sheng¹, WANG Jian-chun¹, LIU Hong-bin²

1. China Ship Scientific Research Center, Wuxi 214082, China;
2. National Supercomputing Center in Wuxi, Wuxi 214072, China

Received:2021-01-06 Revised:2021-03-05 Online:2021-08-25 Published:2021-09-13

摘要/Abstract

摘要： 异构众核已成为超级计算机架构的发展趋势,而如何有效利用众核处理器的计算能力,是当前CFD界的一项重大挑战。该文面向异构众核处理器,基于自主开发代码,开展水动力学CFD并行计算探索研究。首先分析异构众核给CFD并行计算带来的挑战;针对国产申威26010处理器,对SIMPLE算法和人工压缩算法设计众核并行方案,并通过二维方腔驱动流算例,对计算结果和众核加速效果进行验证和测试;然后,针对SIMPLE算法热点分散的特点,采用循环融合的方法对其计算流程进行优化。研究结果表明：SIMPLE算法和人工压缩算法分别获得11倍和24倍的最高加速。该项研究工作,初步展现了众核处理器在水动力学CFD并行计算中的应用潜力。

关键词: 众核处理器, 并行计算, 不可压流动, SIMPLE算法, 人工压缩算法

Abstract: Heterogeneous multi-core technology has become the developing trend of the supercomputers architecture. The effective utilization of the computing power of multi-core processor is a grand challenge in the current CFD industry. The parallel computation for the CFD in hydrodynamics has been explored for heterogeneous multi-core processor based on self-developed code. Firstly, the challenges of the heterogeneous multi-core technology to the CFD parallel computation are analyzed. Secondly. The multi-core parallel computing scheme is designed for SIMPLE algorithm and pseudo-compressible algorithm for SIMPLE and the pseudo-compressible algorithms aiming at the SW26010 processor. The two-dimensional square cavity driven flow are calculated to verify and validate the calculation results and the many-core acceleration effect. Then, the calculation procedure is optimized by using the cyclic fusion method in view of the scattered hot spots of SIMPLE algorithm. The results show that the SIMPLE algorithm and Pseudo-compressible algorithm can achieve the maximum acceleration of 11 times and 24 times, respectively. It preliminarily presents the application potential of the multi-core processor in the parallel computation for the CFD in hydrodynamics.

Key words: multi-core processor, parallel computation, incompressible flow, SIMPLE algorithm, pseudo-compressible algorithm

中图分类号:

张亚英, 吴乘胜, 王建春, 刘宏斌. 面向众核处理器的水动力学CFD并行计算探索[J]. 船舶, 2021, 32(04): 15-23.

ZHANG Ya-ying, WU Cheng-sheng, WANG Jian-chun, LIU Hong-bin. Multi-Core Processor Oriented Parallel Computation for CFD in Hydrodynamics[J]. Ship & Boat, 2021, 32(04): 15-23.

面向众核处理器的水动力学CFD并行计算探索

Multi-Core Processor Oriented Parallel Computation for CFD in Hydrodynamics

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 1

编辑推荐

Metrics