jevopt pairs Clang's optimizer with Jev's read of the LLVM IR and source code at each call site, to decide whether inlining a function will shrink or grow the compiled binary.
Tested across all 19 Embench 1.0 programs, it beat clang -Oz's binary size on 7 of them, though it trailed by about 7.9% in aggregate, showing promise without yet beating a mature heuristic.
Ships as a Python CLI plus an LLVM plugin and a web dashboard for exploring the decisions it made on a run.
What you get
- Full Python and LLVM plugin source code, GPL-3.0 licensed
- Web dashboard for exploring inlining decisions and benchmark results






