Parallel Optimization of Accelerator Toolbox by OpenMP and MPI

Abstract

Parallel computing adopted in the simulation program of accelerators improves the computational efficiency, especially tracking with multi-particle and multi-turn which always takes a lot of time. Through investigating the operational principle and running flow of Accelerator Toolbox(AT) embeded in Matlab for accelerator design and simulation, Open Multi-Processing (OpenMP) and Message Passing Interface (MPI) were studied on and adopted in the parallel computing. The key code of particle tracking in AT has been transplanted and test with OpenMP-MPI as well as estimated reasonably, which improve the calculation efficiency largely. It is indicated that OpenMP-MPI hybrid parallel structure is in good agreement with AT programs by avoiding the internal communication and improving load balance. A multi-core computing platform based on OpenMP-MPI has been developed and contributed to the deep optimization of accelerators. And it shows good extensibility, which could speed up by adding computing nodes.