计算机工程与应用 ›› 2014, Vol. 50 ›› Issue (7): 35-40.

• 理论研究、研发设计 • 上一篇    下一篇

基于GPU的重启PGMRES并行算法研究

陈  华,史悦戎   

  1. 中国石油大学(华东) 理学院,山东 青岛 266580
  • 出版日期:2014-04-01 发布日期:2014-04-25

Study on restarted PGMRES parallel algorithm with GPU

CHEN Hua, SHI Yuerong   

  1. College of Science, China University of Petroleum(East China), Qingdao, Shandong 266580, China
  • Online:2014-04-01 Published:2014-04-25

摘要: 重启的PGMRES算法是求解稀疏线性方程组高效的迭代方法之一,计算过程也比较稳定。为加快大规模稀疏线性方程组的求解速度,对重启PGMRES算法使用GPU并行方式进行并行算法实现。提出了ELL压缩存储格式的新存取方式,并依据问题规模和SM数目提出了动态分配线程策略。实验结果表明,该算法可有效提高SM资源利用率,获得3~10倍的加速比。

关键词: 重启PGMRES, 统一计算设备架构(CUDA), ELL压缩存储格式

Abstract: Restarted PGMRES algorithm is one efficient iterative method for solving the sparse linear systems, its calculation process is relatively stable. To accelerate the solving speed of the large sparse linear equations, a parallel PGMRES algorithm is implemented in GPU. This paper presents a new access mode for ELL compression storage format, and proposes a new dynamic allocation strategy of threads based on the problem size and number of SM. Experimental results show that the new algorithm can effectively improve the utilization of SM resource and get 3 to 10 times speedup.

Key words: restarted PGMRES, Compute Unified Device Architecture(CUDA), ELL compression storage format