多核环境下矩阵库设计和优化

【Abstract】wWw.shuoshilunwen.com Since the advent of computer in 1946, numerical computation has been one of the main fields of computer application, which has been widely used in the human’s production and all aspects of life. In the numerical computation, the matrix is one of the most basic and often use operation. In a sense, the matrix is a common in numerical computation, the popularity of which is very important. This paper related matrix to multi-core and parallel computing, and expounded the realization of matrix library separately from the design of matrix library, matrix library parallelization and multi-core environment matrix optimization. The details are as following:(1) The design of matrix library Firstly, confirms the matrix library’ design principles from function, computing requirements, maintainability, object-oriented and parallelization; and then ge the organizational structure of matrix library when combined with matrix ’Mathematical definition, c++’ characteristics and the need of program later expansion. After making clear basic operations of matrix library, based on the platform Microsoft Visual C++ 2008 Express Edition propose solutions of parallel matrix library, and the paper ge the example of calculating determinant’ evaluation.(2) The Parallel research of matrix library After analyzing Message traner model, data parallel model and sharing storage parallel programming model three parallel model, through comparing and analysis emphatically introduces OpenMP which is an application programming interface (API) that supports multi-platform shared memory multi processing programming in C, C++, and Fortran on many architectures, and introduces it’s command, library function and the cycle selecting principles of Parallelization, then implements paralleli of the matrix library which has designed on the platform of Microsoft Visual c++ 2008 Express Edition. Meanwhile, the paper ge parallel calculating determinant evaluation.(3) Optimizing with Intel VTune Introduceing some existing optimization tools, the paper emphatically introduces Intel VTune which was developed for Intel cpu from Intel company, then using Intel VTune to analyse the program and collect informations. After know the actors which affect the efficiency, we improve the program.(4) According to all the operations of matrix such as construction, additional, substract, multiplication, division, equal, transposition, judge symmetrical and positive definite.matrix’ as sequence, inverse matrix, etc.,at the end, do Comparative experiments on existing platform.Based on the application of the basic operation of matrix and the analysis of the output, the calculation results accurate, procedures achieve to the original intention. And through the contrast can see the program after optimization of parallel performance has improved significantly.