×

PLUMP

swMATH ID: 718
Software Authors: Beg, Ivan; Ling, Wu; M\"uller, Andreas; Przybyszewski, Piotr; R\"uhl, Roland; Sawyer, William
Description: The joint CSCS-ETH/NEC collaboration in parallel processing is creating a tool environment for porting large and complex codes to massively parallel computers. Along with a parallel debugger and a performance monitor, this environment provides a parallelization support tool (PST) to supplement the data-parallel programming language High Performance Fortran (HPF). Whereas HPF has only facilities for regular data decompositions, PST supports user-defined mappings of the global name space to individual processors, allowing for the parallelization of unstructured problems.
Since the additional directives of PST alone do not remove all of the complexity of programming parallel unstructured mesh applications, a parallel library for unstructured mesh problems (PLUMP) is currently being developed at CSCS to support the local refinement and dynamic repartioning of meshes distributed over a processor array. The constituent routines simplify the manipulation of the underlying dynamic data structures. The use of PLUMP in conjunction with PST can facilitate the design and implementation of a class of specific, but industrially important, applications.
In this paper we first specify the functionality of PLUMP, and then discuss its use in parallelizing two unstructured problems with dynamic aspects, namely a tight binding molecular dynamics code and a finite element package. Finally we indicate some possible future additions to the library and discuss other future applications.
Homepage: http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.48.3083
Keywords: parallel processing; data-parallel programming language; unstructured mesh; local refinement; molecular dynamics code; finite-element package; high performance Fortran
Related Software: PROGRES; SPARSKIT; ITSOL
Cited in: 14 Documents

Citations by Year