Towards parallelism detection of sequential programs with graph neural network

Shen, Yuanyuan; Peng, Manman; Wang, Shiling; Wu, Qiang

doi:10.1016/j.future.2021.07.001

Cited by 6 publications

(12 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our multigraph learning method achieved higher improvements in all metrics than our previous single-representation learning method. 20 Our method performed slightly worse than AdaBoost DT in terms of scores when identifying non-parallelizable loops. However, it achieved higher accuracy, and the scores for identifying parallelizable loops are higher than those of the other models.…”

Section: Quantitative Analysismentioning

confidence: 84%

“…We constructed an extended dataset for the parallelism discovery task based on a dataset generator 20 to improve the performance of the model.…”

Section: Datasetmentioning

confidence: 99%

“…17 Previous machine learning methods using manual feature extraction models have proved the validity of discovering the potential parallelism of sequential programs. 18,19 Recently, a data-driven method using graph neural networks demonstrated its accuracy and efficiency 20 in this field. We can divide the studies into two approach categories: static and dynamic.…”

Section: Introductionmentioning

confidence: 99%

“…However, this dynamic method triggers a high runtime overhead, particularly when analyzing complex programs. From the static perspective, the parallelism discovery method is efficient because there is no need to execute the programs. In paper, 20 we proposed an end‐to‐end neural network framework based on contextual flow graphs (XFG) 17 that does not require the manual extraction of dynamic features to discover parallelism. However, because the training dataset is generated based on Pluto, some programs are discarded because their parallelisms are not currently determined by Pluto which cannot extract the polyhedron dependencies from non‐affine programs.…”

Section: Introductionmentioning

confidence: 99%

“…We exploit the greatest strengths of the multigraph learning 33,34 method to learn each graph representation that contributes to the parallelism pattern on its own, rather than based on explicit manually designed rules or hand‐crafted features. The contributions of this study are as follows.Considering that the dataset used in paper 20 only contains affine programs that reduced the parallelism diversity of the training dataset, we generate a newly extended dataset that includes both affine and non‐affine programs to demonstrate the generality of the data‐driven method. We aim to train a more comprehensive set of potential parallelism features in sequential programs to improve the model. Considering that the information from multiple graphs of the source code can comprehensively enrich its heterogeneous features, we propose a multigraph representation strategy to represent codes in a complementary manner.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Multigraph learning for parallelism discovery in sequential programs

Shen

Peng

et al. 2023

Concurrency and Computation

Self Cite

View full text Add to dashboard Cite

Summary Parallelization is an optimization technique that is playing an increasingly vital role in software applications. Discovering the potential parallelism in sequential programs is of primary importance for parallelization. To this end, various tools have been created to obtain parallelizable targets by analyzing the dependencies in sequential programs. However, the majority are constructed based on manually designed dependence analysis rules. Consequently, they must follow some constraints, which limit the scope of their program analysis capability. Furthermore, constructing a dependency analysis rule is a complicated and highly sophisticated endeavor. In this study, we exploit the fact that neural network models can learn the potential features of programs to tackle the parallelism discovery task in an end‐to‐end manner. Specifically, we developed a multigraph learning architecture on top of multiple abstract code representations in a complementary manner. In the architecture, we equip code representations with different neural network models in a targeted manner, including a deep convolutional neural network for analyzing both control and data flows and a novel neural network for learning the abstract syntax trees. Experimental results obtained on a common parallelism discovery dataset indicate that the developed multigraph learning architecture can learn potential parallelism patterns in sequential programs with high accuracy and efficiency.

show abstract

Section: Quantitative Analysismentioning

confidence: 84%

“…We constructed an extended dataset for the parallelism discovery task based on a dataset generator 20 to improve the performance of the model.…”

Section: Datasetmentioning

confidence: 99%