基于新型CPU/GPU混合架构的高精度隐式无网格并行算法研究

基于CPU/GPU混合架构发展新型并行算法是目前CFD等领域并行算法研究的前沿热点；但混合架构特有的多层众核和多样性存储器等特征，对算法的并行化提出了更加严苛的数据不相关性要求，这就为CFD并行算法，特别是本身数据依赖性更强的隐式算法的设计和实施带来了巨大挑战。本项目直面这一挑战，通过构造新型并行计算格式，寻求高精度隐式无网格算法数据依赖瓶颈问题的突破。具体工作涉及面向高精度算法的点云快速生成技术、兼容GPU架构特征的隐式格式构造方法、兼具负载平衡的CPU/GPU协同计算策略等关键点，通过创新突破，力求集格式高精度、无网格灵活性和异型架构高效率优势与一体，发展出基于CPU/GPU混合架构的新型高精度隐式无网格并行算法。研究将具体结合工程常用的Euler和Navier-Stokes方程求解问题展开，不但具有学术意义，而且具有十分重要的工程应用价值。

Developing new-type parallel algorithms on the CPU/GPU hybrid architecture has become a cutting-edge research topic in various engineering fields, especially in CFD. However, the hybrid architecture has obvious different characteristics of multilevel many-cores and multifarious memories, the comprehensive utilization of these characteristics raises stringent requirements of data independency and huge challenges for developing parallelized CFD algorithms, especially for implicit algorithms with inherent strong data dependency. Facing the challenge, this project is devoted to break the bottleneck of strong data dependency of high-resolution implicit meshless methods for solving the Euler and Navier-Stokes equations. To achieve this purpose, three key works, including the development of fast cloud generation technique oriented toward high-resolution algorithms, the design of new implicit schemes that compatible with GPU characteristics, and the construction of collaborative computing strategy between CPU and GPU, will be carried out to develop novel parallel algorithms of high-resolution implicit meshless methods on the CPU/GPU hybrid architecture. Due to the comprehensive utilization of high hardware performances as well as the careful concerning of high-resolution and flexibility of meshless methods, therefore, the project not only has important significances to the academic research, but also has very great engineering application values for accelerating the simulations of real complicated flows.