culagesvd() in CULA

Question

I want to use the culadgesvd() function to compute SVD for a matrix. It is not clear for me how to actually use it in C/C++ with the info in the documentation. Could anyone give me a complete small C program, a template, to show how basically the function is used? Just a couple of lines (with stuff like culaInitialize() and culaShutdown()) will do, I just need to see how this function can run without error.

Answer 1

I thought that SVD is not included in the free version (it is though). Below is an example for a QR decomposition. Put this in a .cpp file:

#include<cula.h>
#include<iostream>

int main()
{       
    float x[] = {1, 2, 3, 4};
    int n_rows = 2, n_cols = 2;
    float scale[n_cols];

    culaInitialize();   
    culaSgeqrf(n_rows, n_cols, &(x[0]), n_rows, &(scale[0]));
    culaShutdown();

    for(int ii = 1; ii < n_rows; ii++)
    {
        for(int jj = 0; jj < n_cols; jj++)
        {
            if(ii > jj) { x[ii + jj * n_rows] *= scale[jj]; }
        }
    }

    for(int ii = 0; ii < n_rows * n_cols; ii++)
    {
        std::cout << x[ii] << std::endl;
    }

    return 0;
}

and compile it using:

g++ -fPIC -c -I/usr/local/cula/include -Wl,-rpath,/usr/local/cula/lib64 -L/usr/local/cula/lib64 -lcula_lapack_basic gpuQR.cpp
g++ -o gpuQR gpuQR.o -Wl,-rpath,/usr/local/cula/lib64 -L/usr/local/cula/lib64 -lcula_lapack_basic

Then call the program using:

./gpuQR
-2.23607
0.894427
-4.91935
-0.894427

Check my post about CULA and Rcpp here.