matlab效率（matlab效率）.doc

matlab效率（matlab效率） Students Bowen, benefit from the collection here. - Ive been talking about theory before, and to avoid being a pure theory blog, talk about the most experimental part of this: coding. Matlab is the most widely used software in numerical computation and simulation. It is convenient and efficient, and you can usually convert pseudo code algorithms into matlab code directly. But at the same time, matlab also has a lot of performance traps. For example, for statements that are critical. In MATLAB, there are at least two ways to add an array of each element to 1., one from the generic programming language for, i=1:length (a), a (I), =a (I), +1, end, and the other to learn from the mathematical notation a=a+1. But in MATLAB, the performance of the two implementation is very different. Although the MATLAB in the continuous optimization of the implementation of for, but even matlab 2010b on my laptop (Core 2 Duo 2.4 GHz, 4GB, Win7), for a one-dimensional array of 100 thousand elements, the first to achieve 5 microseconds, while the second only 0.4 delicate. The latter is 10 times faster. In fact, using MATLAB, youll find that MATLABs own functions usually have very high performance. For example, just mentioned a=a+1, or sorting sort (a), as well as commendable matrix multiplication, SVD decomposition, and so on. In fact, for two matrices A and B, if you simply implement with three layers of for loops (using C), the performance will definitely be much worse than matlab. Although the Matlab interface is implemented with java to facilitate cross platform, performance is not satisfactory, but the kernel is implemented with C. Try the command matlab -nojvm. For matrix operations, Matlab is called a highly optimized BLAS/LAPACK library. For example, the general matrix multiplication provided by BLAS level3, it will be split into many small pieces of the matrix, make full use of memory continuity, CPUs cache architecture, multi-threaded, etc. to optimize the imp