| The name of this matrix library is COS_matrix,which is designed and developed to make the COStream developed by the laboratory convenient for matrix operation.Since the host language COStream is a data flow programming language based on static data flow scheduling,the workload of the matrix operation contained in the matrix library must can be estimated by COStream.Because the existing matrix libraries fail to meet this requirement,I chose to develop this library instead of using external libraries.COS_matrix library implements many common matrix operations,such as matrix multiplication,matrix inverse,etc.Since COStream is generally used in a multi-core environment,in order to make full use of the performance of multi-core and ensure the high performance of the matrix library,the COS_matrix library uses C++and OpenMP for library development,parallelizes the library functions suitable for parallel,and according to the difference of matrix partitioning methods,three kinds of parallel algorithms are summarized,that is,the parallel algorithm of chessboard partitioning,the parallel algorithm of column partitioning and the parallel algorithm of row partitioning.In addition,the COS_matrix library has a compile-time checking module added to the COStream front end for the case where the COStream language compile-time checking mechanism is still not perfect.Used to check the spelling of library function names,the number of function parameters and function parameter types,and the repeated definition of variables and those variables which not defined but still use,and after the error is found,output detailed error messages to allow users to locate errors more quickly in order to correct them and improve programming efficiency.In order to verify the effectiveness of using OpenMP combined with the parallel algorithm designed for matrix operation,these parallelized library functions are tested sequentially on the laboratory server,and a better speedup result is obtained. |
PDF Full Text Request