Mankov Chain Optimizations on Shenwei Processor

In this paper, we focus on the optimizations of Markov chain on Shenwei processor. In order to increase the speed of simulation calculations using Markov chain, we present a series of optimizations for Shenwei processor according to its structural and instruction characteristics. First, we use the Athread thread library to accelerate and optimize the slave cores to achieve a fine-grained division of the slave core arrays, and based on this, the MPI technology is used to expand the processor from a single core group to a multi-core group for parallel computing. Second, for the limited LDM space, access optimization is performed to achieve the optimal LDM space allocation scheme. Finally, for the problem of increasing communication frequency and communication ratio caused by parallel computing of multi-core group, we adopted DMA data multiplexing and register communication techniques to improve the communication performance.