SUPERSONIC: Learning to Generate Source Code Optimizations in C/C++
Software optimization refines programs for resource efficiency while preserving functionality. Traditionally, it is a process done by developers and compilers. This paper introduces a third option, automated optimization at the source code level. We present Supersonic , a neural approach targeting minor source code modifications for optimization. Using a seq2seq model, Supersonic is trained on C/C++ program pairs ( xt , xt+1 ), where xt+1 is an optimized version of xt , and outputs a diff. Supersonic ’s performance is benchmarked against OpenAI’s GPT-3.5-Turbo and GPT-4 on competitive programming tasks. The experiments show that Supersonic not only outperforms both models on the code optimization task but also minimizes the extent of the change with a model more than 600x smaller than GPT-3.5-Turbo and 3700x smaller than GPT-4.
Wed 30 AprDisplayed time zone: Eastern Time (US & Canada) change
16:00 - 17:30 | Analysis 1Research Track / SE In Practice (SEIP) / Journal-first Papers at 215 Chair(s): Antonio Filieri AWS and Imperial College London | ||
16:00 15mTalk | SUPERSONIC: Learning to Generate Source Code Optimizations in C/C++ Journal-first Papers Zimin Chen KTH Royal Institute of Technology, Sen Fang North Carolina State University, Martin Monperrus KTH Royal Institute of Technology | ||
16:15 15mTalk | An Extensive Empirical Study of Nondeterministic Behavior in Static Analysis Tools Research Track Miao Miao The University of Texas at Dallas, Austin Mordahl University of Illinois Chicago, Dakota Soles The University of Texas at Dallas, Alice Beideck The University of Texas at Dallas, Shiyi Wei University of Texas at Dallas | ||
16:30 15mTalk | Interactive Cross-Language Pointer Analysis for Resolving Native Code in Java Programs Research Track Chenxi Zhang Nanjing University, Yufei Liang Nanjing University, Tian Tan Nanjing University, Chang Xu Nanjing University, Shuangxiang Kan UNSW, Yulei Sui University of New South Wales, Yue Li Nanjing University | ||
16:45 15mTalk | Execution Trace Reconstruction Using Diffusion-Based Generative Models Research Track Madeline Janecek Brock University, Naser Ezzati Jivan , Wahab Hamou-Lhadj Concordia University, Montreal, Canada | ||
17:00 15mTalk | Static Analysis of Remote Procedure Call in Java Programs Research Track Baoquan Cui Institute of Software at Chinese Academy of Sciences, China, RongQu State Key Laboratory of Computer Science, Institute of Software Chinese Academy of Sciences, University of Chinese Academy of Sciences, Beijing, China, Zhen Tang Key Laboratory of System Software (Chinese Academy of Sciences), State Key Laboratory of Computer Science, Institute of Software Chinese Academy of Sciences, University of Chinese Academy of Sciences, Beijing, China, Jian Zhang Institute of Software at Chinese Academy of Sciences; University of Chinese Academy of Sciences | ||
17:15 15mTalk | ArkAnalyzer: The Static Analysis Framework for OpenHarmony SE In Practice (SEIP) chenhaonan Beihang University, Daihang Chen Beihang University, Yizhuo Yang Beihang University, Lingyun Xu Huawei, Liang Gao Huawei, Mingyi Zhou Monash University, Chunming Hu Beihang University, Li Li Beihang University | ||