Toward an Object-Oriented Core of the PPM Library

Abstract

A tensorial approach to computational continuum mechanics using object-oriented techniques Comput. Phys. 12, 620 (1998) Abstract. As high-performance computing (HPC) machines become increasingly complex, middleware-based programming paradigms have been particularly successful in reducing code development time and increasing simulation efficiency. The parallel particle-mesh (PPM) library is a state-of-the-art HPC middleware for parallel particle-mesh simulations. It is based on a concise set of six data and operation abstractions. The present paper describes the architecture of the new PPM library core. This new core architecture enables several simplifications in the library's user interface and supports for the first time the implementation of multi-resolution simulations using PPM. We further demonstrate the competitive performance of the new core architecture compared to the previous version of the PPM library.

A Pthreads Wrapper for Fortran 2003

“…We showed the versatility of forthreads in three examples using different design patterns. All examples extended the PPM library [Awile et al 2010;Sbalzarini et al 2006] using forthreads in order to provide new functionality and multicore support. First, we have extended the existing particle-mesh interpolation routines to spawn one thread per subdomain, executed on separate processor cores.…”

“…The PPM library [Awile et al 2010;Sbalzarini et al 2006] is a programming middleware for parallel hybrid particle-mesh methods. It provides abstract data structures and operations allowing the programmer to reason in terms of particles, meshes, and the associated compute operations [Sbalzarini 2010].…”

“…Our examples extend the Parallel Particle-Mesh (PPM) library, a middleware for distributed-memory parallel particle-mesh simulations [Awile et al 2010;Sbalzarini et al 2006] with shared-memory threads support. The PPM library provides an abstraction layer consisting of a set of high-level data types and operations for hybrid particle-mesh methods [Sbalzarini 2010].…”

A parallel particle method for solving the EEG source localization forward problem

“…The results show that the method converges with the proper order of accuracy. The simulation is implemented as a client to the PPM Library [16,1], which allows it to run in parallel on multiple processors. We have benchmarked the implementation on up to 64 distributed-memory processor cores, showing parallel efficiencies better than 80%.…”

“…The simulation is implemented as a client to the parallelization middleware PPM [16,1] and has been tested on up to 64 distributed-memory processor cores. Figure 4 shows the parallel efficiency achieved by the present implementation of the method for weak scaling.…”