https://github.com/nikic commented: A possible followup here would be to add a PM-level BatchAA analysis for backend use. Then it can be shared across all the SDAG/MIR backend passes. https://github.com/llvm/llvm-project/pull/123934