lenary wrote: What's the overall aim/trajectory here? I know @mgudim is working on making the liveness more accurate in the backend so RA can do some of the spilling, but that doesn't quite feel like what this work is. https://github.com/llvm/llvm-project/pull/180483