jurahul wrote: The intent is avoid this encoding detail from leaking into client code that might want to change the intrinsic ID to another "compatible" one. https://github.com/llvm/llvm-project/pull/106236