海洋环流模式NEMO的代码现代化  

Code Modernization Optimization of Ocean General Circulation Model NEMO

作　　者：周生昌 刘卫国 宋振亚 杨晓丹 ZHOU Sheng-chang;LIU Wei-guo;SONG Zhen-ya;YANG Xiao-dan(School of Software,Shandong University,Jinan 250101,China;First Institute of Oceanography,Ministry of Natural Resources,Qingdao 266061,China;Laboratory for Regional Oceanography and Numerical Modeling,Qingdao Pilot National Laboratory for Marine Science and Technology,Qingdao 266237,China;Key Laboratory of Marine Science and Numerical Modeling,Ministry of Natural Resources,Qingdao 266061,China)

机构地区：[1]山东大学软件学院,山东济南250101 [2]自然资源部第一海洋研究所,山东青岛266061 [3]青岛海洋科学与技术试点国家实验室区域海洋动力学与数值模拟功能实验室,山东青岛266237 [4]海洋环境科学和数值模拟自然资源部重点实验室,山东青岛266061

出　　处：《海洋科学进展》2021年第1期62-72,共11页Advances in Marine Science

基　　金：国家自然科学基金项目--海量数据驱动下的高分辨率海洋数值模式关键算法研究(U1806206)和新型海洋与气候模式的发展(41821004);自然资源部基本科研业务费专项资金项目--ENSO,PDO和AMO的非线性调制机理研究(GY0219Q08)

摘　　要：海洋数值模式是精准海洋环境预报的核心手段。随着计算分辨率的不断提高,海洋数值模式对计算性能的要求也越来越高。为了提高模式计算性能,充分发挥现代计算机的特点,选取海洋环流模式NEMO开展了代码现代化优化方案在海洋环流模式中的应用研究。首先使用Intel性能分析工具对模式的计算性能进行了分析;随后,针对热点函数,开展了编译选项优化、标量串行代码优化、SIMD优化、内存带宽优化以及多进程扩展。结果显示:经过以上优化步骤,在不增加任何硬件成本的前提下,模式整体性能可提升31%,且在多进程下表现出了较好的负载均衡性。这表明,本研究采用的优化策略是一种切实可行的方法。在此基础上,进一步对代码现代化过程中出现的显著影响计算效率的问题,如大量指针的使用阻止矢量化、循环嵌套过多、内存带宽占用过高等,开展了分析和讨论,为未来模式的设计和改进提供了参考和建议。The ocean general circulation model(OGCM)is the key tool for ocean environment simulation and forecast.With the ocean resolution finer,the demand for improving computational performance is more and more urgent.To improve the calculation performance of OGCM by taking full advantage of modern computers,a code modernization optimization scheme is carried out in this paper using an OGCN named NEMO as an example.The Intel performance analysis tool is used to evaluate the computing performance of the model at first.Then,several optimization steps,which are compiler options,serial and scalar optimization,SIMD,memory bandwidth optimization and extending to multi-cores,are applied to hotspot functions.After optimization,the model s overall performance can be improved by 31%without increasing any hardware cost and load balance have a good performance in multi-process.The results indicate that the optimization strategy used in this study is very effective and useful.Furthermore,the problems that significantly affect the computational efficiency in the model,such as the heavy use of pointers that can prevent vectorization,multiple loop nesting,high memory bandwidth usage,are discussed in this paper,to provide reference and suggestion for the OGCM in the future design and improvement.

关 键 词：代码现代化 海洋环流模式 NEMO Intel性能分析工具 SIMD 

分 类 号：P73[天文地球—海洋科学]