Building representative and balanced datasets of OpenMP parallel regions

Alcaraz, Jordi; Sleder, Steven; TehraniJamsaz, Ali; Sikora, Anna; Jannesari, Ali; Sorribes, Joan; Eduardo, César

doi:10.1109/pdp52278.2021.00019

Cited by 8 publications

(6 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…3 Summary of the experimentation workflow -Phase 2: Validating the collection by showing that an ANN trained for classifying code regions using a balanced dataset is highly accurate, even for patterns that have not been considered in the creation of the dataset. It is worth noticing that [4] also shows that the accuracy of the trained model is significantly higher when using a balanced dataset than when using an imbalanced one.…”

Section: Pattern Detection Model Phase 1: Build Pattern Collectionmentioning

confidence: 95%

“…This section summarizes the methodology presented in [4] to determine whether a given OpenMP parallel region pattern covers a portion of the feature space that is not represented by the training set so far. Finding an effective solution to this issue is significant in ensuring a balanced and representative dataset.…”

Section: Generating Balanced Dataset Of Openmp Patternsmentioning

confidence: 99%

“…For tackling the problem of creating representative and balanced datasets we presented in [4] an approach for systematically deciding whether a given OpenMP parallel region pattern covers a unique portion of the search space that is not currently considered covered by in a training set. The search space is the N-dimensional space defined by the valid combinations of N hardware performance counters values.…”

Section: Introductionmentioning

confidence: 99%

“…Therefore, the objective of this paper is to present a holistic perspective for building representative and balanced datasets of OpenMP patterns and tackling their natural imbalance when used for tuning purposes. To fulfill this objective, we summarize the contribution made in [4] introducing a refinement for reducing the number of executions of the patterns, and present the study that has been done to determine which are the best approaches for dealing with naturally imbalanced datasets for tuning performance parameters.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Predicting number of threads using balanced datasets for openMP regions

et al. 2022

Self Cite

View full text Add to dashboard Cite

Incorporating machine learning into automatic performance analysis and tuning tools is a promising path to tackle the increasing heterogeneity of current HPC applications. However, this introduces the need for generating balanced datasets of parallel applications’ executions and for dealing with natural imbalances for optimizing performance parameters. This work proposes a holistic approach that integrates a methodology for building balanced datasets of OpenMP code-region patterns and a way to use such datasets for tuning performance parameters. The methodology uses hardware performance counters to characterize the execution of a given region and correlation analysis to determine whether it covers an unique part of the pattern input space. Nevertheless, a balanced dataset of region patterns may become naturally imbalanced when used for training a model for tuning any specific performance parameter. For this reason, we have explored several methods for dealing with naturally imbalanced datasets for finding the appropriated way of using them for tuning purposes. Experimentation shows that the proposed methodology can be used to build balanced datasets and that such datasets, plus a combination of Random Forest and binary classification, can be used to train a model able to accurately tune the number of threads of OpenMP parallel regions.

show abstract

Section: Pattern Detection Model Phase 1: Build Pattern Collectionmentioning

confidence: 95%

Section: Generating Balanced Dataset Of Openmp Patternsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Predicting number of threads using balanced datasets for openMP regions

et al. 2022

Self Cite

View full text Add to dashboard Cite

show abstract

“…Performance counters are commonly used to represent applications [10], [12], [20]. One drawback of this representation is its cost.…”

Section: A Graph Representation Of Programsmentioning

confidence: 99%

Learning Intermediate Representations using Graph Neural Networks for NUMA and Prefetchers Optimization

TehraniJamsaz,

Popov,

Dutta

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

There is a large space of NUMA and hardware prefetcher configurations that can significantly impact the performance of an application. Previous studies have demonstrated how a model can automatically select configurations based on the dynamic properties of the code to achieve speedups. This paper demonstrates how the static Intermediate Representation (IR) of the code can guide NUMA/prefetcher optimizations without the prohibitive cost of performance profiling. We propose a method to create a comprehensive dataset that includes a diverse set of intermediate representations along with optimum configurations. We then apply a graph neural network model in order to validate this dataset. We show that our static intermediate representation based model achieves 80% of the performance gains provided by expensive dynamic performance profiling based strategies. We further develop a hybrid model that uses both static and dynamic information. Our hybrid model achieves the same gains as the dynamic models but at a reduced cost by only profiling 30% of the programs.

show abstract