一种基于CPU-GPU异构计算的混合编程模型

Hybrid Programming Model Based on CPU-GPU Heterogeneous Computing

在分析基于CPU-GPU异构计算模式程序开发面临的主要挑战和当前解决途径的基础上,设计了一种适用于GPU集群环境的、综合利用MPI、OpenMP、CUDA以及OpenGL API开发技术的混合编程模型,重点分析了混合编程模型的两种编程方式的实现原理,搭建了相应的实验环境并进行了对比测试。实验结果表明,使用混合编程模型设计的程序具有良好的性能表现和扩展潜力。

The paper analyzes the main difficulties of programming based on CPU-GPU heterogeneous computing,summarizes the main approaches available,and designs a hybrid programming model for GPU cluster environment which comprehensively uses MPI,OpenMP,CUDA and OpenGL API.It focuses on analyzing two methods that implement the programming model,then builds experimentation environment and makes a comparison experiment.The result shows that programs built with hybrid programming model have good performance and scalability.