jryans wrote: > AI-generated test cases that apparently still optimize under noipa + noinline. Aha, interesting, thanks... I'll take a look at these cases as well. https://github.com/llvm/llvm-project/pull/203304