Sparse matrix eigenvalues in fortran

Open discussion for MAGMA library (Matrix Algebra on GPU and Multicore Architectures)
Post Reply
Posts: 9
Joined: Tue Jul 25, 2017 9:34 am

Sparse matrix eigenvalues in fortran

Post by Ólafur » Tue Aug 22, 2017 11:44 am

Hey all

I've been trying to leverage some new GPUs for a micromagnetics project and I'm currently looking into the 'magma_zlobpcg' routine which looks promising. The project's main computation is repeated calculation of eigenvalues for a large finite element Hamiltonian alternated with modifying said Hamiltonian. The wrinkle is that I'm working in Fortran so Magma usage is somewhat indirect. I've compiled the example_f.F90 and some other stuff with 'gfortran -Dmagma_devptr_t="integer(kind=8)"...' but I'm worried whether I can access other variable types, specifically magma_z_matrix. Also of concern is the usage of CUBLAS in the example rather than native Magma routines which suggests to me they might not be readily available. Though this might very well be my untrained eye and not really a problem.

If someone could suggest a bit of pseudocode that can point me in the right direction it would be much appreciated.

Posts: 911
Joined: Fri Jan 06, 2012 2:13 pm

Re: Sparse matrix eigenvalues in fortran

Post by mgates3 » Thu Aug 24, 2017 9:43 am

MAGMA BLAS functions that take a queue aren't currently accessible from Fortran. Recently we've been working on updating the Fortran interface to make those available. Hence using the old cuBLAS Fortran interface for those routines.

We haven't made a Fortran interface for sparse. Honestly, yours is the first such request. We'll discuss internally to see what would be involved.


Post Reply