Lieu : salle George Boole 2
Programme
9h30 : Salli Moustapha. Shared Memory Parallelism for 3D Cartesian Discrete Ordinates Solver
10h30 : Pause café
10h45 : Jérémie Gaidamour. A parallel multigrid framework for implementing grid transfer operators based on constrained minimization
11h45 : Point d’équipe
12h00 : Fin des hostilités
Title: Shared Memory Parallelism for 3D Cartesian Discrete Ordinates Solver
Speaker: Salli Moustapha
In a preliminary work, a 3D Cartesian SN solver (DOMINO) has been designed and implemented using two nested levels of parallelism (multicore+SIMD) on shared memory computation nodes. DOMINO is written in C++, a multi-paradigm programming language that enables the use of powerful and generic parallel programming tools such as Intel TBB and Eigen. These two libraries allow us to combine multi-thread parallelism with vector operations in an efficient and yet portable way. As a result, DOMINO can exploit the full power of modern multi-core processors and is able to tackle very large simulations, that usually require large HPC clusters, using a single computing node. The very high Flops/Watt ratio of DOMINO makes it a very interesting building block for a future many-nodes nuclear simulation tool.
Title: A parallel multigrid framework for implementing grid transfer operators based on constrained minimization
Speaker: Jérémie Gaidamour
Multigrid methods can be applied successfully to many linear systems arising from the discretisation of partial differential equations. However, choosing suitable components for multigrid algorithms is challenging, especially for multi-physics simulations involving boundary layers, stretching, or discontinuities. As part of the Sandia Trilinos project, we propose a parallel multigrid framework called MueLu. This new library allows for great flexibility on the choice of inter-grid operators (geometric, aggregate-based, F/C) and can be tuned to choose strategies that are tailored for specific problems. In particular, the library is intended to make available new multigrid methods based on energy minimization. These methods ensure an accurate representation of near nullspace modes on the coarse grids while providing an explicit control over the grid transfer operators’ sparsity patterns. In this talk, we present a configuration of the constrained minimization problem suitable for 3D linear elasticity. We also discuss the integration of MueLu into Trilinos and the lessons learned from the transition to production code as part of an important ASC milestone.
