A threaded SPIKE algorithm for solving general banded systems

A parallel SPIKE algorithm for solving banded diagonally-dominant and non-diagonally dominant linear systems, has been implemented for shared-memory architectures. The numerical experiments demonstrate that the SPIKE-OpenMP solvers offer a highly competitive threaded alternative strategy to the LAPACK BLAS-threaded LU model, for solving banded linear systems on current multicore architectures.

► The SPIKE banded algorithm is implemented using OpenMP on shared-memory architecture.
► The SPIKE-OpenMP solver is compared to the LAPACK BLAS-threaded LU model.
► Significant speed-up improvements are reported for SPIKE on multicore architecture.