Skip to Main content Skip to Navigation
Theses

Leveraging streaming for deterministic parallelization: an integrated language, compiler and runtime approach

Abstract : As single processing unit performance has reached a technological limit, the power wall, the past decade has seen a shift from the prevailing trend of increasing single-threaded performance to an exponentially growing number of processing units per chip. Higher performance returns on these newer architectures are contingent on the amount of parallelism that can be efficiently exploited in applications, either exposed through parallel programming or by parallelizing compilers. However, uncovering raw parallelism is insufficient if a host of cores vie for limited off-chip memory bandwidth. Mitigating the memory wall, the stream-computing model provides an important solution for exploiting upcoming architectures. This thesis explores streaming as a general-purpose parallel programming paradigm, rather than a model dedicated to a class of applications, by providing a highly expressive stream-computing extension to a de facto standard for shared memory programming, OpenMP. We rely on a new formal framework to investigate the properties of streaming programs, without the restrictions usually attached to dataflow models, and we prove that such programs benefit from deadlock and functional determinism, key assets in the productivity race. In a second part, we focus on the efficient exploitation of our model, with optimized runtime support and compiler optimizations, through an implementation in the GCC compiler.
Document type :
Theses
Complete list of metadatas

Cited literature [72 references]  Display  Hide  Download

https://pastel.archives-ouvertes.fr/pastel-00712006
Contributor : Malick Diop <>
Submitted on : Tuesday, June 26, 2012 - 11:27:57 AM
Last modification on : Friday, October 23, 2020 - 4:52:21 PM
Long-term archiving on: : Thursday, September 27, 2012 - 2:40:49 AM

Identifiers

  • HAL Id : pastel-00712006, version 1

Citation

Antoniu Pop. Leveraging streaming for deterministic parallelization: an integrated language, compiler and runtime approach. Automatic. École Nationale Supérieure des Mines de Paris, 2011. English. ⟨NNT : 2011ENMP0090⟩. ⟨pastel-00712006⟩

Share

Metrics

Record views

387

Files downloads

321