The Smart AutoTuner project at SwAPP Lab is supported by NSF and focuses on profiling, performance analysis, and auto-tuning HPC programs using advanced machine learning, code representation, and embedding techniques. Smart AutoTuner automates optimizing the performance of high-performance computing (HPC) applications by adjusting its tuning parameters and configurations to achieve the best possible performance on heterogenous hardware platforms using code structure, semantics, and syntax information.
The Smart AutoTuner project aims to go beyond traditional auto-tuning approaches by utilizing advanced deep learning models, such as Graph Neural Networks (GNNs) and other AI-based and code analysis techniques, to automatically identify and adjust the application parameters and configurations. This approach is particularly useful when optimal tuning parameters are difficult to determine due to the large search space and the complex and heterogeneous hardware platforms. Smart AutoTuner aims to significantly improve the performance of HPC applications, reducing their execution time and improving their scalability by allowing developers to focus on the application's functionality rather than optimizing its performance.
Publications:
- Akash Dutta, Jordi Alcaraz, Ali TehraniJamsaz, Anna Sikora, Eduardo Cesar, Ali Jannesari: Performance Optimization using Multimodal Modeling and Heterogeneous GNN. In Proc. of the 32nd International Symposium on High-Performance Parallel and Distributed Computing (HPDC), Orlando, Florida, USA, pages 1–14, June 2023. (Accepted - arXiv preprint: 2304.12568)
PDF URL DOI BibTex - Akash Dutta, Jee Choi, Ali Jannesari: Power Constrained Autotuning using Graph Neural Networks. In Proc. of the 37th IEEE International Parallel and Distributed Processing Symposium (IPDPS), St. Petersburg, Florida, USA, pages 1–10, IEEE Computer Society, May 2023. (Accepted - arXiv preprint: 2302.11467)
PDF URL DOI BibTex - Ali TehraniJamsaz, Alok Mishra, Akash Dutta, Abid M. Malik, Barbara Chapman, Ali Jannesari: ParaGraph: Weighted Graph Representation for Performance Optimization of HPC Kernels. arXiv preprint: 2304.03487, pages 1–11, April 2023.
PDF URL DOI BibTex - Akash Dutta, Jordi Alcaraz, Ali TehraniJamsaz, Anna Sikora, Eduardo Cesar, Ali Jannesari: Pattern-based Autotuning of OpenMP Loops using Graph Neural Networks. In Proc. of the 3rd Workshop on Artificial Intelligence and Machine Learning for Scientific Applications (AI4S), co-located with SC 2022, Dallas, Texas, pages 1–6, November 2022.
PDF URL DOI BibTex - Ali TehraniJamsaz, Mihail Popov, Akash Dutta, Emmanuelle Saillard, Ali Jannesari: Learning Intermediate Representations using Graph Neural Networks for NUMA and Prefetchers Optimization. In Proc. of the 36th IEEE International Parallel and Distributed Processing Symposium (IPDPS), Lyon, France, pages 1–11, IEEE Computer Society, May 2022. (arXiv preprint: 2203.00611)
PDF URL DOI BibTex