Execution of dataflow process networks on OpenCL platforms

A4 Conference proceedings


Internal Authors/Editors


Publication Details

List of Authors: Wictor Lund, Sudeep Kanur, Johan Ersfolk, Leonidas Tsiopoulos, Johan Lilius, Joakim Haldin, Ulf Falk
Editors: Masoud Daneshtalab, Marco Aldinucci, Ville Leppänen, Johan Lilius, and Mats Brorsson
Publication year: 2015
Publisher: IEEE
Book title: Parallel, Distributed and Network-Based Processing (PDP), 2015 23rd Euromicro International Conference on
Title of series: Euromicro International Conference on Parallel, Distributed, and Network-Based Processing
Number in series: 23
Start page: 618
End page: 625
ISBN: 978-1-4799-8492-3
ISSN: 1066-6192


Abstract

The trend in computing systems is to combine
various kinds of processing elements (PEs) to build more parallel
architectures. This trend leads to more heterogeneous computing
systems, for which abstractions are needed to efficiently program
the systems without increasing the programming cost. This has
lead to new programming languages and application program-
ming interfaces (APIs). Parallel programming has always been
a holy grail in computer science and dataflow programming
promises a way to automatically provide parallel constructs for
the programmer. This paper provides an approach to translate
dataflow process networks (DPNs) into programs running some
of the computations on the Open Computing Language (OpenCL)
platform, supporting running programs on massively parallel
hardware such as graphics processing units (GPUs). We show
that certain DPN programs could run very efficiently on data-
parallel architectures but also that there are certain patterns in
DPN programs that prove problematic.

The trend in computing systems is to combine various kinds of processing elements (PEs) to build more parallel architectures. This trend leads to more heterogeneous computing systems, for which abstractions are needed to efficiently program the systems without increasing the programming cost. This has lead to new programming languages and application programming interfaces (APIs). Parallel programming has always been a holy grail in computer science and dataflow programming promises a way to automatically provide parallel constructs for the programmer. This paper provides an approach to translate dataflow process networks (DPNs) into programs running some of the computations on the Open Computing Language (OpenCL) platform, supporting running programs on massively parallel hardware such as graphics processing units (GPUs). We show that certain DPN programs could run very efficiently on dataparallel architectures but also that there are certain patterns in DPN programs that prove problematic.


Last updated on 2019-15-09 at 07:20