forum.alglib.net

I see that the commercial version of ALGLIB supports multithreading. I am looking for a multithreaded sparse direct matrix solver. The matrix may have up to 50000 real (double precision) entries and matrices with the same structure and similar numerical values are repeated solved. The Harwell Subroutine Library routines MA48 works great for this situation. But it is single threaded. I see ALGLIB has a sparse nonsymmetric direct solver called sparsesolv which utilizes the sparse LU decomposition with dynamic pivoting. Is it multithreaded by chance? Anyone have experience with it?

I got a reply from Sergey Bochkanov which reads:

"Yes, the commercial version comes with many algorithms multithreaded and that specific one was parallelized.

Specific speed-up depends on the sparsity pattern. For example, if you matrix is band diagonal, it is impossible to parallelize no matter how large, because band diagonal structure promotes sequential interdependencies. So, I think that we should just try and see how it works for your case!"