Page 1 of 1

MKL's Direct Sparse Solver alternative

Posted: Sat Jun 07, 2014 7:05 pm
by jribeiro
Hi!

I just found out about magma and i am investigating the possibility of porting a linear solver in a sparse matrix which is now implemented with this MKL routine:

https://software.intel.com/sites/produc ... 43ABC8.htm

to my Fermi GPU. But from my readings i understood that Magma doesn't offert support for Sparse Matrices, is this true or is there an alternative?

Thank you very much and good work!!

Re: MKL's Direct Sparse Solver alternative

Posted: Mon Jun 09, 2014 9:35 am
by mgates3
The latest beta of MAGMA (1.5.0 beta 2) includes sparse support. This is the first release, so the interface and features may change as we refine it. We would appreciate any feedback to help in that process.

-mark

Re: MKL's Direct Sparse Solver alternative

Posted: Tue Jun 17, 2014 6:42 am
by jribeiro
Is there any documentation where i can see how could I solve this linear system with a CSR Matrix? The last documentation I found is from the 1.2 version...

Re: MKL's Direct Sparse Solver alternative

Posted: Tue Jun 17, 2014 1:17 pm
by mgates3
The online docs are actually from 1.4.0. The link that listed it as 1.2.0 was outdated.
http://icl.cs.utk.edu/magma/docs/

Updated documentation is in the MAGMA release,
magma-1.5.0-beta2/docs/html/index.html

But those docs don't yet contain the sparse solvers. The best thing to do is look at the examples in sparse-iter/testing/ and the documentation in the related files, for instance
magma-1.5.0-beta2/sparse-iter/src/dgmres.cpp

-mark