A heterogeneous parallel model of unstructured mesh finite element method based on CPU+GPU

Abstract

Most of the existing numerical simulation programs using the unstructured mesh finite element are based on the traditional multicore processor architecture. With the increase of the number of computing meshes, the computing time is increasing, which leads to the common multicore CPU cluster can’t meet the high computing demand of complex applications. In order to adapt to the trend of the heterogeneous development of high-performance computers, a heterogeneous parallel model of unstructured mesh finite element method is proposed in this paper. It can transplant the unstructured mesh finite element program framework to heterogeneous platform better and faster. The model realizes the efficient utilization of the multicore CPU by hierarchical parallelization, and realizes the efficient utilization of GPU by heterogeneous parallel rewriting for time-consuming computing hotspot. Finally, the model is applied to the parallel transplantation of CPU + GPU heterogeneous platform for the thermal radiation effect program. The results show that the model can reduce the programming difficulty and has good portability and extensibility.