Parallel Programming using FastFlow

Sep 2, 2014, 11:40 AM
40m
Aula (FTU)

Aula

FTU

Speaker

Dr Massimo Torquati (University of Pisa)

Description

FastFlow is an open-source C++ research framework to support the development of multi-threaded applications in modern multi/many-core heterogeneous platforms. The framework provides well-known stream-based algorithm skeleton constructs such as pipeline, task-farm and loop that are used to build more complex and powerful pattern: parallel_for, map, reduce, macro data-flow interpreter, genetic-computation, etc. During the talk we introduce the structured parallel programming framework FastFlow and we discuss problems and issues related to the run-time implementation of the patterns. In particular we will discuss: - algorithmic skeleton approaches and the associated static (template based) or dynamic (macro-data-flow based) implementation - management of non functional features, with particular focus on performance - different optimisations aimed at targeting clusters of multi-core - heterogeneous architecture targeting (including GPGPUs, Intel Xeon PHI and Tilera Tile64)

Presentation materials