您的位置:首页  > 论文页面

基于OpenMP+SIMD的波动方程数值模拟并行方法

发表时间:2013-11-15  浏览量:1125  下载量:368
全部作者: 曹丹平
作者单位: 中国石油大学(华东)地球科学与技术学院
摘 要: 地震勘探中的大规模并行计算在科研与生产中发挥重要作用,但是大规模并行中基于消息传递接口(message passing intexface, MPI)或图形处理器(graphic processing unit, GPU)的并行方法对编程的要求较高。以多核处理器中的小规模并行计算为例,在多核微机处理器上将支持多线程并行的OpenMP方法与支持数据并行的单指令流多数据流(single instruction multiple data,SIMD)指令相结合,针对二维波动方程在时间和空间网格计算中的循环嵌套特点,同时实现了基于OpenMP+SIMD并行的波动方程正演模拟方法。模型测试表明:在常规双核笔记本电脑上采用OpenMP+SIMD的并行方法即可达到6倍的加速比,从而在流行的多核处理器上实现计算效率的有效提高,为进一步提高大规模并行计算的效率奠定了基础。
关 键 词: 计算地球物理学;波动方程;数值模拟;多核并行;单指令流多数据流
Title: Parallel computing of wave equation numerical modeling based on the OpenMP+SIMD technique
Author: CAO Danping
Organization: School of Geosciences, China University of Petroleum
Abstract: The large scale parallel computing is very important in the seismic exploration industry and scientific research, which is a very professional skill for the message passing intexface (MPI) or graphic processing unit (GPU) programming. The small scale parallel computing problem is considered as an example in this paper, two parallel computing methods are combined to improve the compute efficiency. The OpenMP method that supports the multithread parallel computing is used to manage the multicore resources of the CPU, the single instruction multiple data (SIMD) method that supports the data parallel computing is used to compute more data with a single CPU instruction. OpenMP and SIMD are combined together to compute the wave equation parallel, the cycle characteristics of wave equation in time step and space step are considered. The model tests show that the parallel computing method of OpenMP+SIMD improves the compute efficiency significantly, the speedup of dual core processor is up to 6, which is very important to enhance the large scale parallel computing.
Key words: computational geophysics; wave equation; numerical modeling; multicore parallel computation; single instruction multiple data
发表期数: 2013年11月第21期
引用格式: 曹丹平. 基于OpenMP+SIMD的波动方程数值模拟并行方法[J]. 中国科技论文在线精品论文,2013,6(21):2104-2109.
 
0 评论数 0
暂无评论
友情链接