I think that's exactly the wrong takeaway here. Most of these have a well-defined result on a given platform (host + abi). You can measure that result. But it'll be different on a different platform. And the others don't have a well-defined result — a real-world compiler will produce a result, and you can measure that, but it might be different tomorrow. Unless you know the difference between platform-specific behavior and undefined behavior, you don't know which ones to avoid.
"Measuring" is exactly the wrong thing to do because often it is indicative of only your specific architecture/compiler.