Compilers are essential, foundational tools for engineering software. Indeed, we would all want to have a perfect compiler—one that is fully trustworthy and produces the best-performing code. This is a lofty goal and a long-standing challenge. Toward this goal, this talk introduces and advocates a novel shift in perspective by exploring the question: How well do modern compilers perform versus that hypothetical perfect compiler? Doing so is scientifically interesting and can lead to deep, systematic understandings of modern compilers’ capabilities, and significant fundamental and practical advances in compilers. This talk will introduce our vision and recent work on this quest.

Zhendong Su is a Professor in the Department of Computer Science at ETH Zurich. He is passionate about fundamental and practical innovations for building software. His research spans programming languages and compilers, software engineering, computer security, deep learning, and education technologies. He is a Member of the Academia Europaea, a Fellow of the ACM and of the IEEE, and participates in the ACM Distinguished Speaker program.