Code Transformation

Code transformation research focuses on automatically altering code's structure and/or semantics while preserving functionality, aiming to improve performance, understand code semantics, or generate new code. Current efforts leverage machine learning, particularly deep neural networks and contrastive learning, often within the framework of polyhedral compilation or using large language models, to guide the selection and application of transformations. This field is significant because it promises to automate complex software optimization tasks, leading to more efficient and robust software systems, and enhancing our understanding of code's underlying structure and meaning.