About SVD running time in Matlab and in C code which produced by Matlab coder.

I use SVD like code below [U,S,V] = svd(PhiG,'econ'); which PhiG is double complex matrix with dimension 60*1024. In matlab code ,the SVD running time is about 0.04 second. However ,after I compile the matlab code to C code using matlab coder.The svd running time is about 4.0+s,which is about 100 times compared to the running time in Matlab code. I have noticed that the Matlab using lapack code to calculate SVD. So the difference in running time due to Fortran code beneath Matlab code? Or is there any other reason? And how can I improve the running time in C code? Thanks a lot for any feedback.

Answer 1

[Walter Roberson]

If you can use C++ then perhaps http://math.nist.gov/tnt/jama_doxygen/class_JAMA__SVD.html or something in http://lapackpp.sourceforge.net/

Your options are going to depend upon your target platform. For example if you are aiming at MS Windows then you may be able to call upon https://en.wikipedia.org/wiki/Math_Kernel_Library

[WillandSmith]

I am very appreciated for your reply.However, I want to compile my code to embedded system,especially ARM compatible.So C code will be the best choice. Also I know C++ code can run on ARM platform,but I have not much confident in whether these C++ lib/code can run with ARM. Could you give more advice?Thanks again.

[Walter Roberson]

Relevant information that I dug up a couple of months ago from a different context, about what is available for ARM: http://www.mathworks.com/matlabcentral/answers/122248#answer_179228

[WillandSmith]

Thank you,Walter.I will try all your advice in these days.