topic Difference: mkl_?cootrsv and mkl_?coosv in Intel® oneAPI Math Kernel Library

Difference: mkl_?cootrsv and mkl_?coosv

sunshinekid — Wed, 12 May 2010 17:53:03 GMT

Hello!

I have to solve a linear system of equations A*y=x.
My matrix A is a complex general matrix stored in coo.

When I try to call mkl_zcoosv the following way

[fortran]CALL mkl_zcoosv('N',700,1._8,'g  f',mat,row,column,i,workd(ipntr(1)),workd(ipntr(2)))[/fortran]

MKL says 'MKL ERROR: Parameter 4 was incorrect on entry to MKL_ZCOOSV'

In this thread: http://software.intel.com/en-us/forums/showthread.php?t=64052&o=a&s=lr
the first answer says that "for NIST like triangular solvers, matdescra(1) must be 'T' or 'D'".

But I thought mkl_?coosv isn't a triangular solver. And if so, what is the difference to mkl_?cootrsv ?

Regards,

S.

Difference: mkl_?cootrsv and mkl_?coosv

Victor_Gladkikh — Thu, 13 May 2010 07:48:08 GMT

Hi

mkl_?coosv is also triangular solver.

There is some difference between mkl_?coosv and mkl_?cootrsv/mkl_cspblas_?cootrsv.
The first is different interface of these functions.
The second is different functionality:
------------------------------------------------------------------------------------------------------------------
mkl_?cootrsv and mkl_cspblas_?cootrsv routines solve a system of linear equations with matrix-vector operations:

y = inv(A) * x or y = inv(A')*x,

where x and y are vectors, A is a sparse upper or lower triangular matrix with unit or non-unit main diagonal, A' is the transpose of A.

------------------------------------------------------------------------------------------------------------------
The mkl_?coosv routine solves a system of linear equations with matrix-vector operations:

y := alpha*inv(A)*x or y := alpha*inv(A')*x,

where alpha is scalar, x and y are vectors, A is a sparse upper or lower triangular matrix with unit or non-unit main diagonal, A' is the transpose of A.

------------------------------------------------------------------------------------------------------------------
Also mkl_?cootrsv is one-based, mkl_cspblas_?cootrsv is zero-based, and mkl_?coosv support both one-base and zero-base.

Difference: mkl_?cootrsv and mkl_?coosv

sunshinekid — Thu, 13 May 2010 10:19:50 GMT

Thank you very much!
I didn't get it from the reference manual.

Then I need another possibilty to solve my system... Maybe PARDISO or DSS... I'll see...

Difference: mkl_?cootrsv and mkl_?coosv

Sergey_Solovev__Inte — Fri, 14 May 2010 04:19:39 GMT

Hi,
the best way to solve sparse system is using PARDISO. It is flexible, has good performance and scalability. Please try to use PARDISO. If you have any questions you are welcome!

Difference: mkl_?cootrsv and mkl_?coosv

sunshinekid — Fri, 14 May 2010 21:30:03 GMT

Quoting Sergey Solovev (Intel)

Please try to use PARDISO.

I implemented PARDISO successfully, and I'm satisfied.
The performance is really good, even for small matrices of dimension 700x700.

Thank you for your efforts!

S.