topic Hello, in Intel® oneAPI Math Kernel Library

About linear solver dgetrs

FortCpp — Wed, 27 Nov 2013 00:39:56 GMT

Dear MKL developers and users,

I am learning and coding with MKL linear solver. I went to some tutorials online such as the tutorials for eigensolvers. I am using MSVS 2010 and intel cluster studio. So I think using MKL linear solver dgetrs in a c++ program would like this:

[cpp]

#include <mkl.h>
#include <iostream>

int _tmain(int argc, _TCHAR* argv[])
{
   char trans='N';
    int n = 3, nrhs = 1, lda = 3, info = 0, ldb = 3;
    int *ipiv = new int[3]();
   double a[] = {1.,0.,0.,0.,1.,0.,0.,0.,1.};
   double b[] = {3.,4.,5.};
    dgetrs( &trans, &n, &nrhs, a, &lda, ipiv, b, &ldb, &info );
   std::cout << b[0] << " " << b[1] << " " << b[2] << " " << b[3] << std::endl;

}

[/cpp]

But the linear solver doesn't work correctly. Since matrix "a" is Identity matrix (row/column major doesn't matter here), I should be able to get 3, 4, 5 as the solution. However, I got two errors:

1. After executed, the numbers in b are

       [0]   -9.2559631349317831e+061   double
       [1]   3.0000000000000000   double
       [2]   4.0000000000000000   double

2. On exit, there is a run-time error: Run-Time Check Failure #2 - Stack around the variable 'b' was corrupted.

How do I fix these errors?

Thanks.

Hello,

Chao_Y_Intel — Wed, 27 Nov 2013 01:17:35 GMT

Hello,

dgetrs solver the LU-factored square matrix. It needs first call dgetrf to factor the matrix, then call the function.

Thanks,
Chao

Hi Chao,

FortCpp — Wed, 27 Nov 2013 01:45:00 GMT

Hi Chao,

Thanks for the response, dgetrs solved both of them.

Two more questions: 1. does dsytry and dsytrs work in the same way (dsytry goes before dsytrs.)? Does dsytrs take an upper/lower triangle matrix as input or a full matrix? (I tried a bit. But I still cannot make it right.)

Hello,

Chao_Y_Intel — Wed, 27 Nov 2013 08:50:50 GMT

Hello,

dsytrf and dsytrs work in the same way: dsytrf goes before dsytrs.

dsytrf/dsytrs take ither the upper or the lower triangular part of the matrix A, depends on the 'uplo' settings, not the full matrix.

Thanks,
Chao

Thanks Chao. I'll try it.

FortCpp — Thu, 28 Nov 2013 00:30:15 GMT

Thanks Chao. I'll try it.