Implementation of parallel convolution based on MPI
Jin Lu, Kaisheng Zhang, Ming Chen, Ke Ma · 2013
Convolution computing plays an important role in scientific computing. However, traditional Message Passing Interface (MPI) model has the disadvantages such as massive message passing and load-unbalancing. According to these problems, this paper propose a new parallel convolution based on MPI model, which is able to effectively balance the load as well as bring great reduction on message passing. By analyzing the speedup ratio and efficiency, conclusions can be drawn that the new MPI parallel programming, which obtains obvious promotions on parallel efficiency and large-scale sequence's speed, can better exert the advantages of parallel computing and distributional memories between nodes. And it is shown by the experimental results that the proposed method is a feasible and effective parallel strategy.