基于OpenCL的归约算法优化研究.pdfVIP

基于OpenCL 的归约算法优化研究 1,2,3+ 1,2 1 颜深根 , 张云泉 , 龙国平 1( 中国科学院 软件研究所 并行软件与计算科学实验室.北京 100190) 2( 中国科学院 软件研究所 计算机科学国家重点实验室.北京 100190) 3( 中国科学院 研究生院.北京 100190) Reduction Algorithm Optimization Based on the OpenCL Shengen Yan1,2,3+, Yunquan Zhang 1,2, Guoping Long1 1(Laboratory of Parallel Software and Computational Science, Institute of Software, the Chinese Academy of Sciences, Beijing 100190, China) 1 2(State Key Laboratory of Computing Science, the Chinese Academy of Sciences, Beijing 100190, China) 3(Graduate University, the Chinese Academy of Sciences, Beijing 100190, China) + Corresponding author: E-mail : yanshengen@ Abstract : Reduction algorithm has a wide range application in areas such as scientific computing and image processing,this paper Systematicly studied the reduction algorithm optimization on the GPUs cross-platform performance optimization based on the OpenCL framework.Previously research has generally focused on a single hardware architecture, this paper based on the OpenCL,stydied various kinds of optimization methods,such as using vector 、on-chip memory bank conflict、threads organization、instruction selection and so on.we take the minMax function for example,dilatationed each optimization method for develep the performance,and detailed the reason.we tested the algorithm both on AMD GPU and NVIDIA GPU platforms. The test results show that the optimized algorithm on both platforms have achieved good performance. In the AMD ATI Radeon HD 5850 platform,Int and Float types of data bandwidth utilization up to 89%. In the NVIDIA GPU Tesla C2050 platform, the performance has reached 1.3 to 1.9 times compare to appropriate function version of CUDA. Key words: GPU;Parallel Reduction; OpenCL;CUDA 摘 要: 归约算法在科学计算和图像等领域有着广泛应用,本文系统研究了在 OpenCL 框架下,归约算法在 GPU 上的跨平台性能优化.已有研究工作一般只侧重单个硬件架构,本文基于 OpenCL 从向量化、片上存储