计算机科学 ›› 2012, Vol. 39 ›› Issue (4): 296-298.
• 体系结构 • 上一篇 下一篇
张丹丹,徐 莹,徐 磊
出版日期:
发布日期:
Online:
Published:
摘要: 对CPU+GPU异构平台下的多种并行编程模式进行了研究,并针对格子Boltzmann方法实现了CUDA,MPI+CUDA, MPI+OpenMP+CUDA多级并行算法。结果表明,算法具有较好的加速性能;提出的根据计算量比例参数调节CPU和GPU之间负载均衡的方法,对于在异构平台上实现多级并行处理及资源的有效利用具有一定的参考和应用价值。
关键词: 异构平台,GPU,格子Boltzmann,并行
Abstract: This work investigated the implementation of lattice Boltzmann method(LBM) with CUDA on the CPU GPU heterogeneous platform with multiple parallel programming models,MPI+CUDA and MPI+OpenMP+CUDA,which shows good performance speedup. A method used to justify the computational load ratio was proposed in this paper to balance the computational time on CPU and GPUs, which provides insightful information about the multi-level parallclization and efficient usage of different computational resource on the CPU+GPU heterogeneous platform.
Key words: Heterogeneous platform, GPU, LBM, Parallel
张丹丹,徐 莹,徐 磊. 异构平台下格子Boltzmann方法实现及性能分析[J]. 计算机科学, 2012, 39(4): 296-298. https://doi.org/
0 / / 推荐
导出引用管理器 EndNote|Reference Manager|ProCite|BibTeX|RefWorks
链接本文: https://www.jsjkx.com/CN/
https://www.jsjkx.com/CN/Y2012/V39/I4/296
Cited