lucas-rami wrote: This should be correct now. These RP excess calculations are a bit messy; my next PR will make things much more simple by using `GCNRegPressure` objects to track RP in regions showing excess pressure. https://github.com/llvm/llvm-project/pull/140036