https://github.com/fhahn commented: Did you get a chance to evaluate the performance impact of the change? Form the changed tests, it is not clear that they actually show cases where there's a gain from the additional vectorization. https://github.com/llvm/llvm-project/pull/192056