数据与计算发展前沿 ›› 2021, Vol. 3 ›› Issue (4): 93-103.

doi: 10.11871/jfdc.issn.2096-742X.2021.04.008

• 技术与应用 • 上一篇    下一篇

面向国产加速器的CFD核心算法并行优化

曹义魁1,2(),陆忠华1,*(),张鉴1(),刘夏真1,2(),袁武1,2(),梁姗1()   

  1. 1.中国科学院计算机网络信息中心,北京 100190
    2.中国科学院大学,北京 100049
  • 收稿日期:2021-03-03 出版日期:2021-08-20 发布日期:2021-08-30
  • 通讯作者: 陆忠华
  • 作者简介:曹义魁, 中国科学院计算机网络信息中心,在读硕士生,主要研究方向为高性能计算与应用。
    本文承担工作为:负责CCFD V3.0程序在国产加速器上的移植、优化与应用测试。
    CAO Yikui is a master student at CNIC. His activities mainly focus on high perfor mance computition.
    In this paper, he is mainly responsible for transplantation, optimization and application test of CCFD V3.0 program on domestic processors.
    E-mail: caoyikui@cnic.cn|陆忠华,中国科学院计算机网络信息中心,博士,研究员,主要研究方向为网格计算、高性能计算与应用。
    本文承担工作为:CCFD V3.0程序开发指导。
    LU Zhonghua, Ph.D., is a reserch fellow at CNIC. Her activities mainly focus on grid computing and high performance computition.
    In this paper, she is mainly responsible for development gui-dance of CCFD V3.0 program.
    E-mail: zhlu@sccas.cn|张鉴,中国科学院计算机网络信息中心,博士,研究员,主要研究方向为科学计算、高性能计算。
    本文承担工作为:CCFD V3.0程序移植、优化指导。
    ZHANG Jian, Ph.D., is a reserch fellow at CNIC. His activities mainly focus on scientific computing and high performance computition.
    In this paper, he is mainly responsible for transplantation and optimization guidance of CCFD V3.0
    E-mail: zhangjian@sccas.cn|刘夏真,中国科学院计算机网络信息中心,在读博士,主要研究方向为高性能计算,计算流体力学。
    本文承担工作为:CCFD V3.0程序移植与优化指导,正确性验证。
    LIU Xiazhen is Ph.D. candidate at CN-IC. He works in high performance com-putition and computational fluid dyna-mics.
    In this paper, he is mainly responsible for transplantation and optimization guidance of CCFD V3.0 and the correctness veri-fication.
    E-mail: liuxz@sccas.cn|袁武,中国科学院计算机网络信息中心,博士,副研究员,主要研究方向为高性能计算,计算流体力学。
    本文承担工作为:CCFD V3.0程序移植、优化指导。
    YUAN Wu, Ph.D., is an associate research fellow at CNIC. He works in high perfor-mance computition and computational fluid dynamics.
    In this paper, he is mainly responsible for transplantation and optimization guidance of CCFD V3.0.
    E-mail: yuanwu@sccas.cn|梁姗,中国科学院计算机网络信息中心,博士,高级工程师,主要研究方向为高性能计算,计算流体力学。
    本文承担工作为:CCFD V3.0程序移植、优化指导。
    LIANG Shan, Ph.D., is a senior engin-eer at CNIC. Her activities mainly focus on high performance computition and computational fluid dynamics.
    In this paper, she is mainly responsible for transplantation and optimization guidance of CCFD V3.0.
    E-mail: liangshan@sccas.cn
  • 基金资助:
    国家重点研发计划(2017YFB0202803)

Parallel Optimization of CFD Core Algorithms Based on Domestic Processor

CAO Yikui1,2(),LU Zhonghua1,*(),ZHANG Jian1(),LIU Xiazhen1,2(),YUAN Wu1,2(),LIANG Shan1()   

  1. 1. Computer Network Information Center, Chinese Academy of Sciences, Beijing 100190, China
    2. University of Chinese Academy of Sciences, Beijing 100049, China
  • Received:2021-03-03 Online:2021-08-20 Published:2021-08-30
  • Contact: LU Zhonghua

摘要:

【目的】为了加快国产CFD软件的计算速度,本文设计并实现了基于国产加速器的加速版本。【方法】基于CCFD V3.0版本,将软件的核心算法移植到国产加速器,并采用多种方法进行优化。【结果】使用128*128*128大小的网格进行实验,移植后的程序模拟结果与原CPU版本基本一致,单加速卡相比于单CPU核心,对流项计算部分取得了166倍的加速,ADI迭代计算部分取得了59倍的加速。【局限】由于CFD软件模块较多,未对整个程序进行移植优化,未来会将软件都移植到国产加速器上进行加速。【结论】本文实现了基于国产加速器的CFD核心算法并行优化,取得了较好的加速效果,为以后CFD软件的移植与优化工作提供了经验和参考。

关键词: 国产加速器, CFD, 移植, 高性能计算, 并行计算, 优化

Abstract:

[Objective] In order to accelerate the calculation of domestic CFD software, this paper designs and implements an accelerated version of CFD core algorithms based on the domestic processor. [Methods] Based on the CCFD V3.0 version, the core algorithms of the software were ported to the domestic processor and optimized by various methods. [Results] Using a 128*128*128 grid for experiments, the simulation results of the ported program are basically the same as the original CPU version. Compared with a single CPU core, one acceleration card has achieved 166 times acceleration on convection calculations, and 59 times acceleration on ADI iterative calculations. [Limitations] Because there are many modules in CFD software, the target CFD program has not been ported and optimized entirely. In the future, the software will be ported completely to the domestic processors for acceleration. [Conclusions] This article has realized the parallel optimization of the CFD core algorithms based on domestic processors and achieved good acceleration results, which provides experience and reference for porting and optimizing CFD software in the future.

Key words: domestic processors, CFD, transplantation, high-performance computing, parallel computing, optimization