A DAG-based sparse Cholesky solver for multicore architectures
Jonathan D. Hogg, John K. Reid, J. A. Scott · Science and Technology Facilities Council · 2009
In this paper, we describe the design and development of a new code for the solution of sparse symmetric positive-definite linear systems aimed primarily at multicore architectures. Our new Fortran 95/OpenMP code, HSL_MA67, is available as part of the sofware library HSL and extends to the sparse case the task DAG-based approach that is becoming popular in the design and development of dense linear algebra kernels. Comparisons are made with existing parallel solvers, using problems arising from a range of practical applications. We demonstrate thta HSL_MA87 obtains good serial and parallel times on an 8-core machine.