@inproceedings{mastoras23a,
author = {Mastoras, Aristeidis and Yzelman, Albert-Jan N.},
title = {Studying the Expressiveness and Performance of Parallelization Abstractions for Linear Pipelines},
year = {2023},
isbn = {9798400701153},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3582514.3582522},
doi = {10.1145/3582514.3582522},
abstract = {Semi-automatic parallelization provides abstractions that simplify the programming effort and allow the user to make decisions that cannot be made by tools. However, abstractions for general-purpose systems usually do not carry sufficient knowledge about the structure of the program, and thus parallelization with them may lead to poor performance.In this paper, we present a popular class of programs, called linear pipelines, that cannot be easily and efficiently parallelized with general-purpose abstractions. We discuss the difficulties and inefficiencies of parallelizing linear pipelines with general-purpose abstractions, and we explain how pattern-specific abstractions overcome these problems. We present the properties of linear pipelines that should be described with pattern-specific abstractions and how these properties are exploited by the state of the art. In addition, we discuss the importance of exposing the performance parameters and how they are combined by pattern-specific knowledge. We claim that designing pattern-specific abstractions for general-purpose programming models is one way to simplify parallel programming and improve performance without sacrificing any expressive power. Consequently, we propose possible pattern-specific extensions to general-purpose parallel programming models, e.g., {OpenMP}, to support easy and efficient parallelization of linear pipelines.},
booktitle = {Proceedings of the 14th International Workshop on Programming Models and Applications for Multicores and Manycores},
pages = {29–38},
numpages = {10},
keywords = {linear pipelines, parallel programming models, performance parameters, parallelization abstractions, pattern-specific knowledge},
location = {Montreal, QC, Canada},
series = {PMAM'23}
}