高性能应用软件并行优化策略的研究

On the Parallelization Optimization Strategy for High Performance Computing Software

多核时代的来临对现有的应用软件提出了严重挑战,串行代码难以充分发挥硬件资源的性能;软件的并行优化成为亟待解决的重要问题。本文综合了MPI,OpenMP,众核编程模型CUDA三个编程模型进行研究,讨论了适用于不同软件并行优化的方法,提出了适用于企业级应用的软件并行优化策略,最后总结和展望了软件并行优化的挑战和前景。

With the arrival of the multi-core era, traditional software can barely utilize the peak performance of the hardware. Parallelization and optimization of the industrial code is a tough problem the HPC community is facing. In this paper, we present our parallel im-plementation using MPI, OpenMP and CUDA programming models. Different methods, implementation skills and optimization strategies are introduced. At the end, the challenge and the vision of the future work of our optimization strategy are discussed.