GPU/CPU concurrent heterogeneous parallel MOC calculation with asynchronous communication scheme

Recently, method of characteristics (MOC) has been widely developed as the most promise method to process the whole-core transport calculation. Meanwhile, GPU/CPU heterogeneous parallel calculation has been widely used and great performance has been achieved. In this research, GPU/CPU concurrent heterogeneous parallel MOC calculation is implemented to exploit all the computational resource in the heterogeneous high performance computer (HPC) while the asynchronous communication scheme is introduced in to improve the parallel efficiency. In order to accomplish this scheme, both MPI, OpenMP, CUDA protocols are introduced. The spatial domain decomposition (SDD) technique provides the coarse-grained parallelism with the MPI protocol while the fine-grained parallelism is exploited through OpenMP (in CPU calculated domain) and CUDA (in GPU calculated domain) based on the ray parallelization. Numerical results indicate that both the concurrent heterogeneous parallel calculation and the asynchronous communication scheme are effective to improve the performance of the parallel MOC calculation. Moreover, the CPUs/GPUs heterogeneous clusters significantly outperform the CPUs clusters, which makes the large-scale whole-core transport calculation is more practicable. (author)