Jul 11 2016

Focus on a joint research project: FASTLA

FASTLA (2012-2017)

Fast and Scalable Hierarchical Algorithms for Computational Linear Algebra

Principal Investigators :

  • Dr. Olivier Coulaud, HIEPACS project-team, Inria Bordeaux Sud Ouest
  • Prof. Eric Darve, Stanford University
  • Dr. X.S. Li, Lawrence Berkeley National Lab

Research objectives:

FASTLA studies fast and scalable hierarchical numerical kernels and their implementations on heterogeneous manycore platforms for two major computational kernels in intensive challenging applications, namely, fast multipole methods (FMM) and sparse linear solvers that appear in many intensive numerical simulations in computational sciences. For the solution of large linear systems, the ultimate goal is to design parallel scalable methods that rely on efficient sparse and dense direct methods using H-matrix arithmetic. Finally, the innovative algorithmic design will be essentially focused on heterogeneous manycore platforms by using task based runtime systems.

Scientific achievements:

  • Fast Multipole Method: The focus is on developing and on improving methods based on interpolation formulations (Chebyschev, equispaced points) and their efficiency and ease of use in the context of dislocation kernels.
  • Sparse Linear Solvers: Study the use of low rank techniques (H-matrix like) to design fast direct and hybrid solvers able to compute data-sparse approximation of Schur complements.
  • Improved parallelism for modern computers for heterogeneous manycores: Improvement of the parallel performance and scalability using hybrid parallelism and a task based programming model.

Publications and Awards:

  • 9 Journal articles
  • 12 Conference and Workshop papers.

Selected publication:

E. Agullo, B. Bramas, O. Coulaud, E. Darve, M. Messner, T. Takahashi, Taskbased FMM for Multicore Architectures, SIAM Journal on Scientific Computing. vol 36, num 1, 2014.

Follow up :

The team was renewed for 3 years in 2015 to pursue its collaboration along the aforementioned topics involving many PhD and post-docs in addition to the senior researchers of each group.