基于GPU的GRAPES模型并行加速及性能优化  

Parallel Acceleration and Performance Optimization for GRAPES Model Based on GPU

作　　者：王卓薇[1] 许先斌[2] 赵武清[3] 何水兵[2,4] 张玉萍[2] 

机构地区：[1]广东工业大学计算学院,广州510006 [2]武汉大学计算机学院,武汉430072 [3]广东电网公司信息中心,广州510000 [4]高可信软件技术教育部重点实验室(北京大学),北京100871

出　　处：《计算机研究与发展》2013年第2期401-411,共11页Journal of Computer Research and Development

基　　金：中央高校基本科研业务费专项基金项目(3101012);高可信软件技术教育部重点实验室开放课题基金项目(HCST201104)

摘　　要：GRAPES(global/regional assimilation and prediction system)数值天气预报模式作为地球大气一个典型的非线性化离散系统,计算量非常巨大,因此利用低成本、低功耗和高性能的GPU对GRAPES模式进行并行加速成为目前的研究热点.首先通过实现GRAPES模式在GPU中的并行加速,发现系统性能提升并不理想.在此基础上,提出了性能优化策略,包括缓解数据传输时间、降低设备内存加载和存储的数量和避免线程控制流分支,实验结果表明,利用GPU的性能优化策略有效地提升了GRAPES系统性能.GRAPES（global/regional assimilation and prediction system） is a typical non-linear discrete system of the Earth＇s atmosphere developed for numericai weather prediction. There are heavy computations involved in GRAPES. Researchers have recently paid a lot of attentions to the parallel acceleration of the GRAPES model by low-cost, low-power, and high-performance GPUs. In this paper, we implement the parallel acceleration for the GRAPES model in GPUs. But the experimental results show that its performance is not efficient as supposed. Therefore, based on this, we further propose some strategies for optimizing the system performance, including reducing the data transmission time, decreasing the amount of device memory avoiding the branches of thread optimizations on GPUs improve control flows. The experiment GRAPES system performance e loaded and stored equipment, and al results show that the performance ffectively.

关 键 词：GRAPES GPU 数据传输时间 设备内存加载和存储 线程控制流 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]