GPU异构系统中的存储层次和负载均衡策略研究

Research on Memory Hierarchy and Load Balance Strategy in Heterogeneous System Based on GPU

GPU体系结构的革新和相应开发平台的发展使得GPU广泛地应用于科学计算领域。通过深入地分析GPU体系结构和存储层次的优缺点以及GPU上的关键性能特征,阐明了GPU体系结构、编程模型和存储层次之间的关系。针对GPU异构系统上的应用映射提出三种基本负载均衡优化策略:预取、流化、任务划分。试验结果揭示了不同的优化因子与优化效率之间的具体关联。

Owing to the revolution of GPU architecture and improvement of developing platforms, GPU is widely used in scientific computing nowadays. Relationships among GPU architecture, programming model and memory hierarchy are illustrated by analyzing memory hierarchy and exploring key performance featmes of GPU. Three basic load balance strategies on mapping applications onto GPU are presented： Prefetch, stream computing, task division. The effective relationships among different factors and optimization efficiency are tested and exposed by experiments.