Computer Engineering and Applications ›› 2012, Vol. 48 ›› Issue (36): 9-13.

Previous Articles     Next Articles

Multicore parallel programming of acoustic equation numerical modeling

CAO Danping   

  1. School of Geosciences, China University of Petroleum(Huadong), Qingdao, Shandong 266555, China
  • Online:2012-12-21 Published:2012-12-21

声波数值模拟中的多核并行方法研究

曹丹平   

  1. 中国石油大学(华东) 地球科学与技术学院,山东 青岛 266555

Abstract: The highly intensive computation cost is typical for the wave equation numerical modeling. The parallel computation that is integrated with the characteristics of the wave equation finite difference methods is the inevitable trend to adapt the development of the multiprocessor. According to the multi loop nesting structure of the wave equation modeling and the superiority of the OpenMP multithread programming, making sure the parallel order of the circulation body, reducing the serial programming, combining the cycle body, setting the parallel directive to the proper loops and binding the OpenMP threads to the physical cores are used to enhance the computing power of the multiprocessor. These optimized methods are useful to improve the computation efficiency of the multicore processor computer and meaningful to enhance the parallel efficiency of the MPI+OpenMP in the clusters.

Key words: acoustic wave equation, finite difference, numerical modeling, multicore parallel computation

摘要: 波动方程数值模拟普遍存在计算量大的问题,如何根据波动方程有限差分方法的特点开展并行化方法研究是适应微机多核发展的必然趋势。结合波动方程数值模拟中的多层循环嵌套问题和OpenMP的特点,通过确定循环体并行顺序、减少串行环节、合并循环体、准确设置制导语句以及线程绑定优化等方法有助于实现微机多核的高效并行。针对波动方程特点的多核并行不仅有助于提高单机计算效率,对于提高计算机集群上常用的MPI+OpenMP混合并行效率也具有重要意义。

关键词: 声波方程, 有限差分, 数值模拟, 多核并行