Current embedded computing architectures are moving to many-core concepts in order to sustain ever growing computing requirements within complexity and power budgets. Programming many-core architectures not only needs parallel programming skills, but also efficient exploitation of the parallelism at both the architecture and runtime levels. This paper presents a reactive tasks management (RTM) technique that is suitable for fine grain parallelism. Exploiting fine-grain parallelism eases the work of the developer since, most of the time, it is a form of parallelism which is naturally present in applications and doesn't require heavy algorithm rewriting. The RTM API leverages both hardware and software support to efficiently exploit fine-grain parallelism at the lowest possible cost. Simulation on the VC-1 decoding application showed that only 3.5% overhead is induced by this API which makes it truly suitable for fine grain tasks scheduling.
Titolo: | Synchronous Reactive Fine Grain Tasks Management for Homogeneous Many-Core Architectures |
Autore/i: | Ojail M.; David R.; Chehida K. B.; Lhuillier Y.; BENINI, LUCA |
Autore/i Unibo: | |
Anno: | 2011 |
Titolo del libro: | ARCS 2011, 24th International Conference on Architecture of Computing Systems 2011, Workshop Proceedings |
Pagina iniziale: | 144 |
Pagina finale: | 150 |
Abstract: | Current embedded computing architectures are moving to many-core concepts in order to sustain ever growing computing requirements within complexity and power budgets. Programming many-core architectures not only needs parallel programming skills, but also efficient exploitation of the parallelism at both the architecture and runtime levels. This paper presents a reactive tasks management (RTM) technique that is suitable for fine grain parallelism. Exploiting fine-grain parallelism eases the work of the developer since, most of the time, it is a form of parallelism which is naturally present in applications and doesn't require heavy algorithm rewriting. The RTM API leverages both hardware and software support to efficiently exploit fine-grain parallelism at the lowest possible cost. Simulation on the VC-1 decoding application showed that only 3.5% overhead is induced by this API which makes it truly suitable for fine grain tasks scheduling. |
Data prodotto definitivo in UGOV: | 6-ott-2014 |
Appare nelle tipologie: | 4.01 Contributo in Atti di convegno |