A DAG-based parallel Cholesky factorization for multicore systems
Jonathan D. Hogg · Science and Technology Facilities Council · 2008
Modern processors have multiple cores, making multiprocessing essential for competitive desktop linear algebra. Asynchronous processing with much inherent parallelism can be derived by using a directed acyclic graph (DAG) to represent the data dependencies between tasks. In this paper, we present our implementation of a DAG-based Cholesky factorization, using a last minute scheduling approach to prioritise tasks. Our factorization is implemented in Fortran 95 using OpenMP.