Dense Matrix Into CRS Format (For Pardiso)

[michaelmoc]

Hello!

I was searching for any routine in mkl to compress a dense matrix (stored in vector) into CRS form

used by PARDISO. Does anybody know such routine ?

Thanks for help.

[ArturGuzik]

I'm not aware of any (although I admit, I don't fully understand/know your dense matrix format). The MKL available routines are listed in Sparse BLAS Level 2 and Level 3 Routines Section. Table below.

A.

Matrix converters
mkl_ddnscsr	Converts a sparse matrix in the dense representation to the CSR format (3-array variation).
mkl_dcsrcoo	Converts a sparse matrix in the CSR format (3-array variation) to the coordinate format and vice versa.
mkl_dcsrbsr	Converts a sparse matrix in the CSR format to the BSR format (3-array variations) and vice versa.
mkl_dcsrcsc	Converts a sparse matrix in the CSR format to the CSC and vice versa (3-array variations).
mkl_dcsrdia	Converts a sparse matrix in the CSR format (3-array variation) to the diagonal format and vice versa.
mkl_dcsrsky	Converts a sparse matrix in the CSR format (3-array variation) to the sky line format and vice versa.

[michaelmoc]

My format is that the whole matrix is defined in 1 dimensional array. (I know its size so i am able to simulate 2D behavior). And i want such matrix convert to PARDISO format, which is CRS.

[Alexander_K_Intel2]

Hi,

If your store all elements of matrixconsistently column by column and want to call Pardiso for unsymmetric matrix you don't need any converters.To call pardiso you need 3 arrays: values, columns and row index. Values is yourvector, columns is array of n*n elements ={1,2,...,n,1,2,..,n,1,2,...n} and rowindexs is array of n+1 elemnts ={1,n+1,2n+1,..,n*n+1}, where n is size of matrix. So you need to create only two arrays in way I described before and after you could use Pardiso for unsymmetric matrix.

With best regards,

Alexander Kalinkin

[Gennady_F_Intel]

Michael,

I still believe that the most appropriate for you method is using mkl_ddnscsr function to convert from dense to CSR ( not CRS -:) ) format. Especially if you know the sizes of your matrixes.

--Gennady

[michaelmoc]

Gennady!

From what i know PARDISO is running on CRS storage format (http://netlib.org/linalg/html_templates/node91.html) . And thus i can't use CSR. (If i am wrong please correct me, i will gladly use CSR :) ).

Alexander

That is a nice idea but how about performance. It will probably be hit.

Although i will try it :). Thanks.

[michaelmoc]

Hello Gennady.

I have read once again CSR method compression documentation and i have noticed 3 array variation of this compression. This is what i really need :). Thank You for pointing this out.

Michael.