| The solution of large-scale sparse linear equations plays an important role in the finite element program analysis of rock and soil mass.A large number of experiments have proved that the iterative method is used in the solution process,the maximum time cost of the entire program operation process will be spent on the calculation of matrix vector multiplication,so how to improve the efficiency of matrix vector multiplication calculation will reduce the program operation time to the greatest extent.This paper starts from the algorithm and storage method required by the finite element analysis program of rock and soil mass,and studies the MPI message transfer interface and Open MP parallel computing instruction in parallel computing.The main research contents are as follows:(1)In MPI messaging interface,write parallel programs about matrix vector multiplication and vector dot multiplication,and study the running efficiency of parallel programs.By comparison,it is found that when the matrix dimension reaches 10000 magnitude and the number of parallel CPUS is 8,the running time of parallel programs is about 70% less than that of serial programs,and the computing efficiency of matrix vector multiplication is up to 3 times higher.(2)In Open MP application programming interface,taking preprocessing conjugate gradient algorithm as an example,the plane strain problem is analyzed by finite element method.The computational efficiency of parallel programs is tested by using compressed sparse column(CSC)storage method,cell to cell(EBE)storage method and edge-based storage method respectively.By comparison,when the number of threads is equal to the number of CPU cores,the parallel program is the most efficient.Then select the appropriate number of threads to run the parallel program to do finite element calculation of three-dimensional foundation bearing capacity,and study the improvement of calculation efficiency.(3)In order to improve the running efficiency of the finite element program,the matrix vector multiplication calculation part of the program is mixed and parallelized.Open MP parallel computing instruction is added into the single MPI parallel program for matrix vector multiplication operation to make it a hybrid parallel program.Then,by changing the size of the matrix,the calculation time of matrix vector multiplication in the mixed parallel program,the single MPI parallel program and the serial program is compared,and the improvement of the calculation efficiency of the mixed parallel mode is studied,which makes a contribution to improve the running efficiency of the finite element program.The purpose of this paper is to improve the efficiency of finite element analysis program of rock and soil by studying different parallel computing interfaces. |
PDF Full Text Request