JITVis: A Phase-based Visualization for Intermediate Representations in JIT Compilers

Abstract

Web browsers rely on Just-In-Time (JIT) compilers to optimize JavaScript execution. JIT compilers improve the performance of input bytecode through a sequence of optimization phases. JIT compilers often use an intermediate representation (IR), such as an IR graph, to track code dependencies in these optimization phases. Compiler bugs can corrupt these IR graphs, risking accuracy and security. While current debugging tools visualize single optimization phases, they struggle with multi-phase, cross-run analysis. We propose JITVis, a phase-based visualization tool that enables direct comparison of two phases and an overview of all phases in all code runs. We validate our prototype through a case study and are iteratively refining the tool with domain experts.