Hello, I am a new user of Eigen. Previously, I code in Matlab and now I want to port my code to C/C++. I am actually looking for a good library for computing extensive linear algebra in C/C++ and interested in Eigen.
I have just downloaded Eigen3 (ver3.0.2) and did a small test of the computation speed of SVD function for matrix with data type double and size 225x225. However, the computation speed is not satisfying. I am wondering if there is a mistake in my code since I haven't read the documentation of Eigen3 completely. Below are the code and the comparison if I am using built-in svd function from OpenCV:

Code: Select all

#include <iostream>
#include <eigen/Dense>

using namespace Eigen;
int main(void)
{
    clock_t start;
    MatrixXd m1 = MatrixXd::Random(225,225);
    JacobiSVD<MatrixXd> svd;
    start = clock();
    svd.compute(m1);
    double speed = (double)(clock() - start)/CLOCKS_PER_SEC;
    std::cout << "speed is " << speed  << "seconds" << std::endl ;
    return 0;
}


The output:

Code: Select all

speed is 4.59441seconds


Compared with the OpenCV svd:

Code: Select all

#include "opencv2/core/core.hpp"
#include <iostream>

using namespace cv;

int main(void)
{
    clock_t start;
    Mat m1(225,225, CV_64F);
    cv::Mat w,u,v;
    randn(m1, Scalar::all(1000), Scalar::all(1000));
    start = clock();
    SVD::compute(m1, w,u,v);
    double speed = (double)(clock()-start)/CLOCKS_PER_SEC;
    std::cout << "speed is " << speed  << "seconds" << std::endl ;
    return 0;
}


The Output:

Code: Select all

speed is 0.147352seconds


The compiler:
i686-apple-darwin11-llvm-g++-4.2 (GCC) 4.2.1 (Based on Apple Inc. build 5658) (LLVM build 2335.15.00)

Any comment or suggestion? Thank you.

Eigen's SVD implements the Jacobi method which is more reliable than more standard implementations based on Householder reflexion. On the other hand it is also slower for large matrices. That said, a factor 30 for a 225^2 matrix does not make sense. Make sure you compiled you program with optimizations -O2.

I also recommend you to switch to macport' gcc 4.5 or 4.6. Default macosx compiler has a lot of issues.

Thank you for the reply. With the optimizations -O2, the speed improves alot:

Code: Select all

speed is 0.262015seconds