论文部分内容阅读
LICOM是中国科学院大气物理研究所开发的全球海洋环流模式,广泛应用于海洋研究和气候预测。随着分辨率的提高,LICOM对计算的要求呈几何级数的增长。多核或众核已成为当前主流的高性能计算体系结构,原LICOM程序只使用了MPI并行,无法发挥混合架构的高性能。本文介绍了如何在分析LICOM特征基础之上,应用Open MP及其它优化手段对LICOM进行MIC移植和并行优化,详细介绍了模式优化的实现过程。并通过数值试验证明了优化前后的计算效果有一定提高。
LICOM is a global ocean circulation model developed by Institute of Atmospheric Physics, Chinese Academy of Sciences and widely used in marine research and climate prediction. As the resolution increases, LICOM’s computational requirements increase geometrically. Multi-core or multi-core has become the mainstream of high-performance computing architecture, the original LICOM program only uses MPI parallel, unable to play the high performance of hybrid architecture. In this paper, we introduce how to transplant LICOM and optimize it in parallel by using Open MP and other optimization methods based on the analysis of LICOM features. The implementation of pattern optimization is introduced in detail. Numerical experiments show that the computational results before and after optimization are improved to some extent.