Optimizing Code Runtime Performance through Context-Aware Retrieval-Augmented Generation
Optimizing software performance through automated code refinement presents a significant opportunity for enhancing execution speed and efficiency. This study explores a Retrieval-Augmented Generation (RAG) approach for large language models (LLMs) to perform C++ code optimization by providing contextually relevant information for in-context learning. We leverage Control Flow Graph (CFG) analysis and CFG differences between original and optimized code versions to identify performance bottlenecks, eliminate redundancies, and streamline execution paths. Our approach retrieves CFG-based insights and optimization examples as context, guiding the LLM to generate improved versions of input code by mimicking patterns of structural and algorithmic refinement. This research demonstrates how targeted retrieval of CFG-related information and context-appropriate examples can significantly improve the model’s ability to apply efficient transformations, highlighting the potential of RAG-enhanced LLMs in automated software performance optimization.
Sun 27 AprDisplayed time zone: Eastern Time (US & Canada) change
14:00 - 15:30 | Education, Debugging, Dynamic AnalysisResearch Track / Early Research Achievements (ERA) / Replications and Negative Results (RENE) / Tool Demonstration at 205 Chair(s): Simone Scalabrino University of Molise, Coen De Roover Vrije Universiteit Brussel, Gema Rodríguez-Pérez Department of Computer Science, Mathematics, Physics and Statistics, University of British Columbia, Okanagan Campus | ||
14:00 10mTalk | JavaWiz: A Trace-Based Graphical Debugger for Software Development Education Research Track Markus Weninger JKU Linz, Simon Grünbacher Institute for System Software; Johannes Kepler University Linz, Austria, Herbert Prähofer Johannes Kepler University Linz Pre-print | ||
14:10 10mTalk | Pinpointing the Learning Obstacles of an Interactive Theorem Prover Research Track Sára Juhošová Delft University of Technology, Andy Zaidman TU Delft, Jesper Cockx Delft University of Technology Pre-print | ||
14:20 10mTalk | AI-based automated grading of source code of introductory programming assignments Research Track Jayant Havare Indian Institute of technology - Bombay, Varsha Apte Indian Institute of technology - Bombay, Kaushikraj Maharajan Indian Institute of technology - Bombay, Nithin Chandra Gupta Samudrala Indian Institute of technology - Bombay, Ganesh Ramakrishnan Indian Institute of technology - Bombay, Srikanth Tamilselvam IBM Research, Sainath Vavilapalli Indian Institute of Technology - Bombay | ||
14:30 10mTalk | An Analysis of Students' Program Comprehension Processes in a Large Code Base Research Track Anshul Shah University of California, San Diego, Thanh Tong University of California, San Diego, Elena Tomson University of California, San Diego, Steven Shi University of California, San Diego, William G. Griswold University of California San Diego, Gerald Soosairaj University of California, San Diego | ||
14:40 6mTalk | OVERLORD: A C++ overloading inspector Tool Demonstration Botond Horváth ELTE Eötvös Loránd University, Budapest, Hungary, Richárd Szalay Eötvös Loránd University, Faculty of Informatics, Department of Programming Languages and Compilers, Zoltán Porkoláb ELTE Eötvös Loránd University, Budapest, Hungary Pre-print | ||
14:46 6mTalk | Optimizing Code Runtime Performance through Context-Aware Retrieval-Augmented Generation Early Research Achievements (ERA) Manish Acharya Vanderbilt University, Yifan Zhang Vanderbilt University, Kevin Leach Vanderbilt University, Yu Huang Vanderbilt University | ||
14:52 6mTalk | Investigating Execution-Aware Language Models for Code Optimization Replications and Negative Results (RENE) Federico Di Menna University of L'Aquila, Luca Traini University of L'Aquila, Gabriele Bavota Software Institute @ Università della Svizzera Italiana, Vittorio Cortellessa University of L'Aquila Pre-print | ||
14:58 6mTalk | Understanding Data Access in Microservices Applications Using Interactive Treemaps Early Research Achievements (ERA) Maxime ANDRÉ Namur Digital Institute, University of Namur, Marco Raglianti Software Institute - USI, Lugano, Anthony Cleve University of Namur, Michele Lanza Software Institute - USI, Lugano Pre-print | ||
15:04 6mTalk | Divergence-Driven Debugging: Understanding Behavioral Changes Between Two Program Versions Early Research Achievements (ERA) Rémi Dufloer Univ. Lille, Inria, CNRS, Centrale Lille, UMR 9189 CRIStAL, F-59000 Lille, France, Imen Sayar Univ. Lille, CNRS, Inria, Centrale Lille, UMR 9189 CRIStAL, F-59000 Lille, France, Anne Etien University of Lille, Lille, France, Steven Costiou INRIA Lille | ||
15:10 10mTalk | Effectively Modeling UI Transition Graphs for Android Apps via Reinforcement Learning Research Track Wunan Guo School of Optical-Electrical and Computer Engineering, University of Shanghai for Science and Technology, Zhen Dong Fudan University, Liwei Shen Fudan University, Daihong Zhou School of Computer Science and Information Engineering, Shanghai Institute of Technology, Bin Hu Fudan University, Chen Zhang Fudan University, Hai Xue University of Shanghai for Science and Technology | ||
15:20 10mLive Q&A | Session's Discussion: "Education, Debugging, Dynamic Analysis" Research Track |