Glibc error for dgesvd

Anonymous30 — Sun, 08 Dec 2013 07:50:59 GMT

Hi, I am using MKL to compute SVD of a real matrix. I use C++ code with std::vectors. Here is a portion of my code related to SVD

void Decompose_InterpolMatrix_To_SVD(int rows, int columns, std::vector<double> &Interp_matrix) {
        /* Locals */
        int m = rows, n = columns, lda = m, ldu = m, ldvt = m, info, lwork;
        int i,j,k;

        double wkopt;
        double* work;
        double *a,alpha=1.0e0,beta=0.0e0,tausq=1.0e-4;
        //double s[columns], u[rows*rows], vt[columns*columns];
        std::vector<double> u, s, vt;

        a = &(Interp_matrix[0]);
        s.resize(columns);
        u.resize(rows*rows);
        vt.resize(columns*columns);
        //inverse_s_ut.resize(columns*rows);

double *vtran;
vtran = &(vt[0]);

        /* Query and allocate the optimal workspace */
        lwork = -1;
        dgesvd( "All", "All", &m, &n, a, &lda, &(s[0]), &(u[0]), &ldu, &(vt[0]), &ldvt, &wkopt, &lwork,
         &info );
        lwork = (int)wkopt;
        work = (double*)malloc( lwork*sizeof(double) );
        /* Compute SVD */
        dgesvd( "All", "All", &m, &n, a, &lda, &(s[0]), &(u[0]), &ldu, &(vt[0]), &ldvt, work, &lwork,
         &info );
        /* Check for convergence */
        if( info > 0 ) {
                std::cerr << "The algorithm computing SVD failed to converge." << std::endl;
                exit( 1 );
        }

/* Free workspace */
free( (void*)work );

}

When I run this code I get :

*** glibc detected *** ./checkmkl1: munmap_chunk(): invalid pointer: 0x000000001a43e610 ***
======= Backtrace: =========
/lib64/libc.so.6(cfree+0x166)[0x3693672886]
./checkmkl1[0x40206a]
./checkmkl1(__gxx_personality_v0+0x29a)[0x401d22]
/lib64/libc.so.6(__libc_start_main+0xf4)[0x369361d994]
./checkmkl1(_ZNSt8ios_base4InitD1Ev+0x41)[0x401b29]

I do not get this error when I declare s,u,vt as proper arrays. This is strange as one can use vectors of primitive types as arrays. Any help is greatly appreciated.

Suman

Hi. Sorry for the false alarm

Anonymous30 — Sun, 08 Dec 2013 19:31:22 GMT

Hi. Sorry for the false alarm. I gave ldvt a value of m instead of n.

topic Hi. Sorry for the false alarm in Intel® oneAPI Math Kernel Library

Glibc error for dgesvd

Hi. Sorry for the false alarm