https://github.com/mshockwave commented: This is a pretty neat optimization as the original version might allocated more bits than we needed for `Scratch`. Thank you! I only have a few comments. https://github.com/llvm/llvm-project/pull/69898