Automatic Identification of Parallelizable Loops Using Transformer-Based Source Code Representations

Automatic parallelization remains a challenging problem in software engineering. This work proposes a Transformer-based approach to classify the parallelization potential of source code, focusing on distinguishing independent (parallelizable) loops from undefined ones. The approach adopts DistilBERT to process source code sequences using subword tokenization, capturing contextual syntactic and semantic patterns without handcrafted features. Results show consistently high performance, highlighting the potential of lightweight Transformer models for practical identification of parallelization opportunities at the loop level.