Reproduce “NPS: A Framework for Accurate Program Sampling Using Graph Neural Network” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yuanwei Fang, Zihao Liu, Yanheng Lu, Jiawei Liu, Jiajie Li, Yi Jin, Jian Chen, Yenkuang Chen and 2 more
This work presents NPS, a framework for accurate program sampling that leverages Graph Neural Networks to learn rich execution embeddings. NPS reduces average sampling error by 38% compared to the standard SimPoint approach, significantly improving accuracy and robustness for microprocessor performance evaluation.
Key insights
- Program execution behavior, including control-flow, dataflow, and dynamic values, can be learned as high-resolution execution embeddings using Graph Neural Networks.
- A self-supervised data prefetching task, which requires predicting bitwise-accurate consecutive memory addresses, effectively trains the GNN model to capture complex execution rules.
- Learning execution invariants allows for generalized program characterization and sampling across different applications, enabling direct comparison of workload behaviors.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.