SVD of a complex matrix in Matlab's MEX interface

Hello, I have been trying to implement a MEX function to calculate the pseudo inverse of a complex matrix and when looking for the best technique to implement this I came upon the proposed C algorithm by JULIEN LANGOU in the following thread

http://icl.cs.utk.edu/lapack-forum/viewtopic.php?f=5&t=2712&p=7404&hilit=pseudo+inverse#p7404

the above proposed algorithm in my understanding comprises of three steps: 1) Calculate the SVD. 2) calculating the inverse of S 3) calculating the inverse which is (V * S^(-1) * U^H)

I encountered a problem while trying to calculate the SVD using the "zgesvd" function as the pre-allocated random valued matrices S,VT,U values' don't change after calling the "zgesvd" function, I don't fully understand why this happens as this is actually my first time using C-MEX. my code is as follows:

if true
  void mexFunction(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[])
{

    double  *A, *PINV; /* pointers to input & output matrices*/

 int M,N;      /* matrix dimensions */
    M = (int)mxGetM(prhs[0]);
    N = (int)mxGetN(prhs[0]);
    A = (double*)mat2fort(prhs[0], M, N);
    PINV = mxMalloc(2*N*M*sizeof(*PINV));
    /* dimensions of input matrices */

    int  K, L, LDA, LDAxN;
    int LDU, LDVT, LWORK, INFO;
    char JOBU, JOBVT;
    mxArray *temp1, *temp2, *temp3;
    double *S, *U, *VT, *WORK, *RWORK ;

    K     = M < N ? M : N;
    L     = M > N ? M : N;
    LDA   = M;
    LDAxN = LDA * N;
    LDU   = M;
    LDVT  = K;
    JOBU  = 'S';
    JOBVT = 'S';
    LWORK = 1 > 2*K+L ? 1 : 2*K+L;

    S = (double * )mxMalloc(1*K*sizeof(double));

    U = (double * )mxMalloc(2*LDU*K*sizeof(double));

    VT = (double * )mxMalloc(2*LDVT*K*sizeof(double));

    WORK = (double * )mxMalloc(2*1*LWORK*sizeof(double));

    RWORK = (double * )mxMalloc(1*5*K*sizeof(double));

     /* Compute the SVD of input matrix */
    zgesvd( &JOBU, &JOBVT, &M, &N, A, &LDA, S, U, &LDU, VT, &LDVT, WORK, &LWORK, RWORK, &INFO );

    temp1 = fort2mat(U, LDU, LDU, K);

   temp2 = mxCreateDoubleMatrix(1,K,mxREAL);
    double * temp2R = mxGetPr(temp2);
     for( int i=0; i< K; i++ ) {
         *temp2R++ = *S++;
     }

    temp3 = fort2mat(VT, LDVT, LDVT, K);

    plhs[0] = temp1;
    plhs[1] = temp2;
    plhs[2] = temp3;

} end

Any help regarding this problem would be much appreciated and also if you have any other suggestions to implement the pseudo inverse in MEX and many thanks in advance.

Answer 1

[Mustafa]

So apparently the problem was that when calling Lapack functions in MEX all the input indices should be of type "mwSignedIndex" instead of "int".This successfully solved the problem. I found this while trying to find a solution, however I don't know exactly why can someone explain ?!