Reproduce “Automatic Identification of Parallelizable Loops Using Transformer-Based Source Code Representations” 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
Izavan dos S. Correia, Henrique C. T. Santos, Tiago A. E. Ferreira
A Transformer-based method using DistilBERT automatically classifies the parallelization potential of C source code loops by distinguishing independent from undefined ones. The approach achieves a mean test accuracy of 99.600% and a mean False Positive Rate of 0.002872 across 10-fold cross-validation.
Key insights
- Lightweight Transformer models like DistilBERT are effective at learning contextual representations from raw source code, enabling robust classification of loop parallelizability without extensive feature engineering.
- Direct source code processing via subword tokenization significantly streamlines the analysis pipeline and enhances generalization capabilities compared to methods relying on structured code representations.
- The ability to maintain a consistently low False Positive Rate (FPR) is crucial for safe automatic parallelization, as incorrectly classifying a dependent loop as parallelizable can lead to critical errors.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.