Section: New Software and Platforms

High Performance Scientific Computing

PALMTREE

Participants : Lionel Lenôtre [correspondant] , Géraldine Pichot.

• Version: version 1.0, November 2013

• Programming language: C++

• Abstract: We present an easy-to-use package for the parallelization of Lagrangian methods for partial differential equations. In addition to the reduction of computation time, the code aims at satisfying three properties:

  • simplicity: the user just has to add the algorithm governing the behaviour of the particles.

  • portability: the possibility to use the package with any compiler and OS.

  • action-replay: the ability of the package to replay a selected batch of particles.

  The last property allows the user to replay and capture the whole sample path for selected particles of a batch. This feature is very useful for debugging and catching some relevant information.

• Current work: paper about performance results.

MUESLI

Participant : Édouard Canot [corresponding author] .

Muesli is a library designed to help in coding scientific problems in Fortran using a vector-oriented syntax like Matlab. One of its aims is therefore to speed-up the development process. It contains all the necessary materials to work numerically with a dynamic array (dynamic in size, shape, type, and storage structure), called mfArray. Muesli includes all or some parts of the following numerical libraries: Blas and Lapack, Arpack, Minpack, Slatec, Sparskit, SuiteSparse, Metis, Triangle, RngStreams, and other routines based on ACM algorithms.

The key points of Muesli is to efficiently solve large ODE/DAE systems (which come from, e.g., PDE problems after using the method of lines) or large non-linear minimization problems (where Jacobian matrices can be provided in a sparse format). The user can easily monitor the whole integration process and have access to tools to fix the singularity of the system of equations.

The lastest version of Muesli is 2.9.5 (2014-10-03). More information can be found at: http://people.irisa.fr/Edouard.Canot/muesli .

Zohour

Participant : Édouard Canot [correspondant] .

Zohour is a node-based adaptive 2D mesh algorithm, written in Fortran 2003. A basic rectangular, regular set of nodes is recursively refined. Then the cells come from the Voronoi tessellation. While the domain is currently limited to a rectangular shape, its strength is three-fold:

• first, computing the flux via a Finite Element or Finite Volume method is both simple and accurate because each cell-side of cells is the bisection of two nodes;

• second, the transition between zones of different levels of refinement is more progressive than other methods, leading to a smaller number of nodes for the whole mesh;

• third, during successive refinements when dealing with a transient problem, interpolation is needed only by the new nodes, limiting the numerical errors.

It is planned for use in the HeMaTiS code ( 5.4.1 ) in order to get a refined mesh zone around the phase change surface.

See also the web page http://people.irisa.fr/Edouard.Canot/zohour .