Skip to Main content Skip to Navigation
Conference papers

AP+SOMT: Agent-Programming Combined with Self-Organized Multi-Threading

Yves Lhuillier 1 Olivier Temam 1
1 ALCHEMY - Architectures, Languages and Compilers to Harness the End of Moore Years
LRI - Laboratoire de Recherche en Informatique, UP11 - Université Paris-Sud - Paris 11, Inria Saclay - Ile de France, CNRS - Centre National de la Recherche Scientifique : UMR8623
Abstract : In order to scale up processors beyond ILP, we explore the exploitation of coarser-grain parallelism. We advocate that a slightly different programming approach, called agent programming (AP), can unveil a large amount of parallelism, considerably simplify the task of optimizing compilers and empower the architecture with the ability to exploit potential parallelism based on available resources. We show that an SMT, augmented with dynamic steering strategies and thread swapping features, is an appropriate solution for such self-organized architectures; self-organized SMT is called SOMT. Using a set of specially written agent-like programs corresponding to classic algorithms, we show that AP+SOMT exhibit better performance, stability and scalability for a large array of data sets, and makes compiler optimizations less necessary. Finally, we outline that the approach can be progressively adopted as a combination of a hardware add-on and C language extensions, much like multimedia support in current superscalar processors.
Complete list of metadata

Cited literature [28 references]  Display  Hide  Download
Contributor : Pierre Palatin Connect in order to contact the contributor
Submitted on : Thursday, February 9, 2006 - 11:26:27 AM
Last modification on : Friday, February 4, 2022 - 3:33:12 AM
Long-term archiving on: : Saturday, April 3, 2010 - 10:18:37 PM


  • HAL Id : inria-00001107, version 1


Yves Lhuillier, Olivier Temam. AP+SOMT: Agent-Programming Combined with Self-Organized Multi-Threading. Workshop on Complexity-Effective Design at ISCA 31, Jun 2004, Munich/Germany. ⟨inria-00001107⟩



Record views


Files downloads