Matrix Multiplication Based On NoC Multi-core System

With the rapid expansion of array signal scale,the performance requirements of digital signal system for large-scale data operation continue to improve.It is of great practical significance and engineering value to realize a high-performance parallel computing system.In the intensive computing,the traditional single core processing system has been difficult to meet the requirements of computing performance.Therefore,using multi-core technology and parallel algorithm to achieve the intensive computing is one of the issues that must be solved under the background of the times.Matrix operation is one of the core problems of intensive computing.In this paper,the relationship between parallel algorithm and multi-core system in matrix operation is studied,and the design method of multi-core system is preliminarily explored.Taking the matrix of large dimension as the research object,the parallel algorithm of matrix multiplication is analyzed in the multi-core system,and the task scheduling and resource allocation of matrix multiplication in multi-core system are realized,and the parallel computing ability of multi-core system is fully brought into play.The main work of this paper is as follows:Firstly,the parallel algorithms of matrix multiplication are analyzed.According to the parallelism of the algorithm and the feasibility of hardware implementation,the algorithm based on systolic array is selected as the parallel computing method of matrix multiplication,and the storage mode,data transmission and operation mode of multi-core system are constructed.According to the storage,communication and operation model of matrix multiplication and parallel algorithm,2D-Mesh NoC isomorphic multi-core system is designed.The separation of control layer and data transmission layer is realized through three-layer network,and the communication efficiency of the whole system is improved.Finally,the algorithm mapping of matrix multiplication of different dimensions in NoC multi-core system is designed under the limited bandwidth resources and operation resources.The operation results of each algorithm in the system are verified,compared,analyzed and summarized,and the further optimization direction is proposed.