Refereed article in conference proceedings (A4)

Towards a language framework for thick control flows




List of Authors: Mäkelä J., Forsell M., Leppänen V.

Conference name: IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW)

Publisher: Institute of Electrical and Electronics Engineers Inc.

Publication year: 2017

Book title *: Parallel and Distributed Processing Symposium Workshops (IPDPSW), 2017 IEEE International

Journal name in source: Proceedings - 2017 IEEE 31st International Parallel and Distributed Processing Symposium Workshops, IPDPSW 2017

Number of pages: 10

ISBN: 978-1-5386-3409-7

eISBN: 978-1-5386-3408-0

ISSN: 2164-7062

DOI: http://dx.doi.org/10.1109/IPDPSW.2017.119


Abstract

In the recent years the search for scalability in terms of computing power has led to very complex parallel computer architectures which require greater control of the storage and computation resources to utilize all the available hardware capacity for optimal performance. New solutions in the level of programming languages/models have increased the reliance and need for threads. A system with a huge number of threads can face problems with thread micro-management, smooth scaling between data and task parallelism, portability, and consistency. We present TCF++, a new concurrent C/C++ language extension generalizing on the idea of threads with so called thick control flows. Opposed to threading, thick control flows provide a way to orchestrate computation using lower number of independent actors, dynamically adapting to problem size. The language extension approach is chosen to support mixing with legacy code. We qualitatively analyze the new language's eligibility and explain its idiomatic use with a selection of core parallel algorithm kernels.


Last updated on 2021-24-06 at 08:59