IRHash - Efficient Multi-Language Compiler Caching by IR-Level Hashing

Learn about IRHash, an innovative intermediate representation (IR)-level compiler caching system for LLVM that addresses the limitations of existing compilation caches in this 16-minute conference presentation from USENIX ATC '25. Discover how researchers from Leibniz Universität Hannover and Technische Universität Braunschweig developed a solution that extends beyond traditional C/C++ language support to enable multi-language compiler caching for all LLVM-supported languages including C, C++, Fortran, and Haskell. Explore the conceptual framework behind choosing the optimal lookup level for compiler caches, moving from source code and AST-level hashing to IR-level hashing for improved accuracy and broader language compatibility. Examine the performance evaluation results showing IRHash's superior end-to-end build time savings of 19% compared to existing tools like Ccache (10%) and cHash (16%) across 16 open-source projects. Understand how IR-level hashing achieves higher cache hit accuracy while supporting the diverse ecosystem of LLVM backend languages, making it a significant advancement in compilation optimization technology for modern multi-language development environments.