Localized parallel computing architecture and method based on MPICH + OPENMP

Abstract

The invention relates to a localized parallel computing architecture and method based on MPICH + OPENMP, and belongs to the field of parallel computing. The method is based on MPICH + OpenMP hybrid architecture construction, and the effect of improving the calculation efficiency is achieved by achieving hybrid parallel acceleration of a calculation algorithm. Coarse grain parallel based on the MPICH parallel architecture realizes parallel computing among nodes, and has the functions of task distribution among the nodes, data collection and synchronization and the like. And parallel computing of an algorithm level between CPU cores in the nodes is realized based on fine grain parallel computing of the OpenMP parallel architecture. The parallel acceleration platform can be flexibly applied to domestic hardware platforms such as Feiteng and Loongson and domestic operating systems such as Neokylin and Kylin. According to the method, the number of the parallel threads can be automatically adjusted according to the parallel computing node system resources, the parallel threads are distributed, the maximum utilization of computing power is achieved, compared with traditional MPI parallel, better stability and flexibility are achieved, and meanwhile the defect that OpenMP only supports a single computing node is overcome.

Description

technical field [0001] The invention belongs to the field of parallel computing, in particular to a localized parallel computing architecture and method based on MPICH+OPENMP. Background technique [0002] From 1986 to 2002, the performance of microprocessors increased rapidly at an average rate of 50% per year. Since 2002, the performance improvement rate has dropped to 20%. In recent years, the upper limit of clock frequency has made the speed increase of a single processor very limited as the heat dissipation capacity of cooling integrated circuits with air has reached its limit. The development of multi-core processors has become the choice of most mainstream CPU manufacturers. [0003] Traditional serial computing programs cannot fully utilize multiple processor cores. Therefore, special computing programs need to be written so that the computing power of each processor core can be fully utilized. [0004] Since the Prism Gate, the state has increased its support fo...

AI Technical Summary

Problems solved by technology

[0006] The technical problem to be solved by the present invention is how to provide a localized parallel computing architecture and method based on MPICH+OPENMP to solve the problem of poor parallel stability and flexibility of traditional MPI, while OpenMP only supports a single computing node

Images