Abstract:

[Context] With the rapid development of high performance computers and the increasing requirement for reducing the difficulty of parallel programming, developing high performance programming frameworks for supercomputers has become one of the research directions in high-performance computing.[Methods] This article briefly introduces the motivation of developing a high-performance programming framework for numerical simulation, mainly focusing on the current status of the programming frameworks for structured mesh and unstructured mesh. [Conclusions] With the advent of the exascale computers, the programming framework software also needs to adjust the underlying data structure and core algorithms according to the architecture characteristics of the exascale computers, improving its scalability, and enhancing supports for heterogeneous computing. At the same time, it is necessary to explore new areas and new directions based on the needs of domain experts. and to develop new modules applied to more scientific computing fields.

Key words: High performance computing, programming framework, numerical simulation