ScaLAPACK Solving Ax=B how to store A

[cfav]

Hi
I am completely new user of MPI and therefore of ScaLAPACK. I ve been trying to follow the tutorials provided in the ScaLAPACK web site, and I found my self a bit confused with the way the [A] matrix is given to ScaLAPACK.
One of the expamples presented in the Tuturial uses a 6x6 matrix, so it is reasonable to store in feed the subroutine with the complete matrix, but
my question arises when solving a 1 millon of sparse Eqs.(This is the reason why I am using parallel processing)
Would some body please explain me if there is some way to send [A] stored lets say for example as a 1D vector to avoid unnecessary zeros.
Any reply is very much welcome, thanks in advance.

[buttari]

Hi
I am completely new user of MPI and therefore of ScaLAPACK. I ve been trying to follow the tutorials provided in the ScaLAPACK web site, and I found my self a bit confused with the way the [A] matrix is given to ScaLAPACK.
One of the expamples presented in the Tuturial uses a 6x6 matrix, so it is reasonable to store in feed the subroutine with the complete matrix, but
my question arises when solving a 1 millon of sparse Eqs.(This is the reason why I am using parallel processing)
Would some body please explain me if there is some way to send [A] stored lets say for example as a 1D vector to avoid unnecessary zeros.
Any reply is very much welcome, thanks in advance.

Hi,
ScaLAPACK only works with dense data. For computations on sparse matrices you can use Trilinos, PETSc, PSBLAS,...
a list of freely available linear algebra libraries, where you can find what suits your needs best, is here:

http://www.netlib.org/utk/people/JackDo ... la-sw.html

Regards,

Alfredo

[cfav]

I am thankful to you for your reply.