Sparse direct solver and preconditioner based on supernodal sparse LU factorization

Area: Mathematical libraries

CASS member: FASTMath

Description

SuperLU is a suite of libraries of sparse direct solvers using a DAG based supernodal approach. It can be used for any sparse linear systems and particularly efficient for non-symmetric sparsity. In addition to sparse LU and triangular solves, the library provides error bounds, condition estimation, and easy-to-use interface for repeated solutions with same sparsity pattern, etc. The Serial SuperLU package also contains numerical threshold-based ILU routines. It is written in C with MPI/OpenMP paralleliztion, and CUDA (NVIDIA) and HIP (AMD) GPU acceleration. It also contains bindings to Fortran, Julia and Python.

Target audience

SuperLU can be used by any applications that need to solve sparse linear systems. It is especially useful for the linear systems that are indefinite and/or ill-conditioned. These systems often arise from multiscale and multiphysics modeling and simulations. It can also be used in higher level solver libraries, such as coarse grid solvers in a multigrid solver, subdomain solvers in a domain decomposition solver, block diagonal preconditioners in a Krylov iterative solver.

Additional resources

Impact stories