基于异构平台的通量分裂格式性能研究
DOI:
作者:
作者单位:

西南科技大学 计算机科学与技术学院

作者简介:

通讯作者:

中图分类号:

TP311

基金项目:

国家自然科学基金(61802320);四川省教育厅项目(18TD0021)。


Research on performance of flux split format based on heterogeneous platform
Author:
Affiliation:

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    通量分裂是在方程组条件下实现迎风特性的主要手段,为了实现典型通量分裂格式在CPU/GPU异构平台的性能分析。在NVIDIA GTX1660super上,使用统一设备计算架构(Compute Unified Device Architecture,CUDA)编程模型实现一维欧拉求解器。以激波管Riemann问题为算例,对矢通量分裂格式van leer、通量差分分裂格式Roe以及混合通量分裂AUSMPW +进行计算分析。数值结果表明,三种格式在异构计算体系能够得到合理且可用的计算结果;Roe格式激波分辨率最高且在CPU/GPU体系加速效果最好;Van Leer激波分辨率较低于Roe和AUSMPW+,计算效率高但其格式构造中存在大量判断分支,影响了加速性能;AUSMPW+格式激波分辨率与Roe相当,加速性能略好于Van Leer。

    Abstract:

    Flux splitting is the main method to realize the upwind characteristics under the condition of equations. In order to realize the performance analysis of typical flux splitting formats on CPU/GPU heterogeneous platforms. On NVIDIA GTX1660super, a one-dimensional Euler solver is implemented using the Compute Unified Device Architecture (CUDA) programming model. Taking the Riemann problem of shock tube as an example, van Leer, Roe and AUSMPW + are calculated and analyzed in vector flux splitting scheme, flux differential splitting scheme and mixed flux splitting scheme. The numerical results show that the three schemes can get reasonable and usable results in the heterogeneous computing system. Roe format has the highest resolution and the best acceleration effect in CPU/GPU system. The resolution of Van Leer shock wave is lower than Roe and AUSMPW+, and the calculation efficiency is high. However, there are a lot of judgment branches in the structure of Van Leer shock wave, which affects the acceleration performance. The AUSMPW+ format has a shock resolution comparable to Roe and a slightly better acceleration performance than Van Leer.

    参考文献
    相似文献
    引证文献
引用本文

梁正虹,黄俊,刘志勤,陈波,杨茂.基于异构平台的通量分裂格式性能研究计算机测量与控制[J].,2021,29(2):144-149.

复制
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2020-12-05
  • 最后修改日期:2020-12-24
  • 录用日期:2020-12-24
  • 在线发布日期: 2021-02-08
  • 出版日期: