论文部分内容阅读
目前使用有限差分对波动方程进行数值模拟是比较流行的方法,在理论研究和工作生产中已经广泛被使用.本文给出了一种组合使用多种并行优化设计方法对算法核心进行深入优化的方法,它充分挖掘算法的并行性及程序各阶段运行的智能监控,通过采用减少同步、边界重划、线程和处理器绑定、零复制,空间局部性、时间局部性及Pthreads多线程编程模型等一系列并行优化设计方法,实现二维波动方程正演模拟的优化,从而大幅提升程序的性能,使正演数值模拟达到实时仿真的效果.通过对有限差分数值模拟的并行调优前后试验数据比较,验证了该一系列深入并行优化方法能够较好地提高有限差分二维正演模拟的计算效率.
At present, it is a popular method to simulate the wave equation by using the finite difference method, and it has been widely used in theory research and work production.This paper presents a method to optimize the core of the algorithm by using a variety of parallel optimization design methods , It fully exploits the parallelism of the algorithm and the intelligent monitoring of the operation of each stage of the program. Through the use of reduced synchronization, boundary rezoning, thread and processor binding, zero copy, spatial locality, temporal locality and Pthreads multithreaded programming model A series of parallel optimal design methods are used to optimize forward modeling of two-dimensional wave equation, so as to greatly improve the performance of the program and make the forward modeling achieve real-time simulation results.Through the comparison of experimental data before and after the parallel tuning of finite difference numerical simulation , And verified that a series of in-depth parallel optimization methods can better improve the computational efficiency of two-dimensional finite-difference forward modeling.