Error in feast eigenvalue solver for sparse matrices

[aurora]

Hi,

We are trying to use the MKL FEAST 11.0.3.1 solver for symmetric sparse CSR matrices of doubles. The call we are using is something like:

--------------------------------------------

#include "mkl.h"

......

//declaring & preparing data...

....

 feastinit(&feastparam[0]);   dfeast_scsrev(&UPLO,&N,sa,ia,ja,feastparam,&epsout,&loop,&Emin,&Emax,&M0,E,X,&M,res,&info);

--------------------------------------------

We have no warnings at compiling time, but at runtime the feast call throws an exception like this:

First exception in 0x000007fee1b8249c in feast.exe: 0xC0000005: Invalid read in 0xffffffffffffffff.

Which could be our mistake? Input data? Bad linkage/compiler version-mkl version combination? We are not getting any compiling warnings.

Intel Composer XE 2011 Update 7 (package 258)

Thanks in advance!

Aurora

[Zhang_Z_Intel]

It's hard to see what's wrong in the a few lines of code you've shown. Please provide your entire test code. What is your OS? How do you compile/link the code?

Please first check the FEAST example in MKL: $MKLROOT/examples/solvers_eec/source/dexample_sparse_c.c. Does it work for you?

[SergeyKostrov]

>>Which could be our mistake? It is hard to answer without seeing your test case. >>Input data? See above. >>Bad linkage/compiler version-mkl version combination? No.

[aurora]

Hi again, 

The example that you told us worked fine. We are essentially doing the same. However we are still having problems on runtime. 

We are working in Windows 7 x64.

//!!!!!!!!!!!!!!! Matrix declaration variables !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

char          UPLO='F'; 

const MKL_INT N = matrix_size;

double * val = new double[size];

int * rows = new int [size_rows + 1];

int * cols = new int [size];

for(int i=0;i<size;i++){ val=matrix_CSR.pr; } 

for(int i=0;i<(size_rows+1);i++){ rows=matrix_CSR.jc; }

for(int i=0;i<size;i++){ cols=matrix_CSR.ir; } 

   //!!!!!!!!!!!!!!! Declaration of FEAST variables !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

    //!!!!!!!!!!!!!!! E - eigenvalues, X - eigenvectors, res - residual !!!!!!!!!!!!! 

    MKL_INT       fpm[128];

    double        Emin, Emax;

    double        epsout;

    MKL_INT       loop;

    MKL_INT       L = 20;

    MKL_INT       M0,M, info;

    double       * E =  new double ;

for(int i=0;i<N;i++){ E=0; }

    double       * X =  new double [N*N];

for(int i=0;i<N*N;i++){ X=0; }

    double       * res =  new double ;

for(int i=0;i<N;i++){ res=0; }

//!!!!!!!!!!!!!! search interval [Emin,Emax] including M eigenpairs !!!!!!!!!!!!

    Emin=(double)-0.5;

    Emax=(double)0.01;

    M0=L;

    M=M0;

    printf(" Search interval [ %.15e, %.15e  ]  \n", Emin, Emax);

    epsout=(double)0.0;

    loop=0;

    info=0;

    //         Task 1. Call  FEASTINIT to define the default values for the input 

    //         FEAST parameters. 

    feastinit(fpm);

    fpm[0]=1;

    //        Task 2. Solve the standard Ax=ex eigenvalue problem.

    dfeast_scsrev(&UPLO,&N,val,rows,cols,fpm,&epsout,&loop,&Emin,&Emax,&M0,E,X,&M,res,&info);

    if ( info != 0 ) return 1;

[Zhang_Z_Intel]

It could be a problem in your input data. How big is your matrix? What are the values for variables such as N, size, and size_row? Can you attach your matrix please?

[aurora]

Hi again, 

Find in the attachment the input data of our example. We have written the matrix values and the row and column info arrays of the CSR matrix.

Thank you!

[Zhang_Z_Intel]

Thanks for providing the test matrix. Please check a few things in your code:

• Make sure that row and column indices are 1-based instead of 0-based. This is an requirement of using MKL FEAST.
• Make sure the column indices for each row are sorted in increasing order.
• Check your file I/O. The exception you got seems to indicate a problem in reading data from file.

[aurora]

Thank you for your advice!

Now our example is working properly. The problem was that the row and column indices were 0-based instead of 1-based as MKL FEAST needs. 

Thanks again!

[SergeyKostrov]

>>>>First exception in 0x000007fee1b8249c in feast.exe: 0xC0000005: Invalid read in 0xffffffffffffffff... >> >>- Check your file I/O. The exception you got seems to indicate a problem in reading data from file 'Aurora' confirmed that actually it was an Out-Of-Bound problem ( memory related problem ) and it was not related to any I/O operations with files. Anyway, it is good to know that in some MKL processing indices are 1-based. Thanks for the note.

[Victor_N_]

Seems hard to learn how to use this routine at first, but if one goes through the example and runs it, it would become clearer.  Finally I got my problem resolved with the fixes for indices and column-oriented sparse matrix.