topic Using LAPACK to invert sparse matrices in Intel® oneAPI Math Kernel Library

Using LAPACK to invert sparse matrices

thomasdd — Tue, 27 Oct 2009 14:29:21 GMT

I've some small square matrices (so around 100x100) that are sparse. I want to use the LAPACK functions to compute the inverse. To do this, I've tried to use the function dgetrf (in C++). But always the info return value is > 0. I've checked the matrix, but it has no zero values on the diagonals. Because the matrix is square, I can use n=m=lda for the input parameters. Is it right? Is there any trouble known when using sparse matrices with LAPACK? I've also check the matrices in Matlab, they are non-singular.

Thanks for any hints,

Thomas

Re: Using LAPACK to invert sparse matrices

Melvin_Robinson — Tue, 27 Oct 2009 23:28:36 GMT

Quoting - thomasdd

Can you post some example code so that I can see what you are doing?

Re: Using LAPACK to invert sparse matrices

thomasdd — Wed, 28 Oct 2009 07:32:15 GMT

Quoting - Melvin

Can you post some example code so that I can see what you are doing?

A very simple example:

int n = 4, m = 4, lda = 4;
int ipiv = 0, info = 0;
double matrix[16] = {1, 0, 0, 0,
0, 1, 0, 0,
0, 0, 1, 0,
0, 0, 0, 1};
dgetrf(&m, &n, matrix, &lda, &ipiv, &info);
std::cout << "INFO = " << info << std::endl;

In this case, info is 4! Is there anything wrong in the call of dgetrf?

Thomas

Re: Using LAPACK to invert sparse matrices

Gennady_F_Intel — Wed, 28 Oct 2009 15:56:18 GMT

Quoting - thomasdd

Quoting - Melvin

Can you post some example code so that I can see what you are doing?

A very simple example:

int n = 4, m = 4, lda = 4;
int ipiv = 0, info = 0;
double matrix[16] = {1, 0, 0, 0,
0, 1, 0, 0,
0, 0, 1, 0,
0, 0, 0, 1};
dgetrf(&m, &n, matrix, &lda, &ipiv, &info);
std::cout << "INFO = " << info << std::endl;

In this case, info is 4! Is there anything wrong in the call of dgetrf?

Thomas

Thomas,
ipiv is the array, but not integer
the dimention at least max(1,min(m,n)), therefore for your case ipiv[4].
Please refer to the mkl's reference manual.
--Gennady

Re: Using LAPACK to invert sparse matrices

thomasdd — Thu, 29 Oct 2009 07:27:10 GMT

Quoting - Gennady Fedorov (Intel)

Thomas,
ipiv is the array, but not integer
the dimention at least max(1,min(m,n)), therefore for your case ipiv[4].
Please refer to the mkl's reference manual.
--Gennady

Thank you very much! Now it works fine. I was too fast in reading the reference :)

Thomas