26-28 Jun 2019 Bordeaux (France)

Talks (sorted by speaker) > Jacquelin Mathias

A 2D Task-based Fan-both Factorization Algorithm for Sparse Symmetric Matrices
Mathias Jacquelin  1@  , Esmond Ng  2  
1 : Lawrence Berkeley National Laboratory  (LBNL)  -  Website
Lawrence Berkeley National Lab MS 50F Cyclotron Rd., Berkeley, CA 94720 -  United States
2 : Lawrence Berkeley National Laboratory  (LBNL)

In this work, we present a new 2D data distribution of a sparse symmetric matrix, which balances the workload and the number of nonzero entries explicitly. We have incorporated low-overhead task-based Cholesky and LDLt algorithms in the symPACK solver using this new distribution, implemented using a custom runtime environment based on the UPC++ communication framework. We analyze the performance achieved during the numerical factorization as well as that of the solution phase.


Online user: 1