Writing microbenchmarks is hard. Really hard.
Getting reliable, repeatable, statistically significant, statistically sound results from a microbenchmark requires an advanced level of understanding of statistics, an advanced level of understanding of all the possible implementations and their possible optimizations (and ideally even not-yet-existing implementations and future optimizations), an advanced level of understanding of the code involved, and lots of experience writing microbenchmarks. People who write benchmarks are typically specialized professionals who do nothing but write benchmarks all day, every day for years. And even then, they get it wrong. IIRC, there was a famous case of a SPEC benchmark (pretty much the organization specializing in benchmarking) which was supposed to test the performance of database drivers but actually ended up testing memory allocator performance.
And even if you manage to write a reliable, repeatable, statistically significant, statistically sound microbenchmark, you will end up with perfectly reliable, perfectly repeatable, statistically significant, statistically sound results which are also perfectly useless, because microbenchmarks pretty much by definition don't reflect your actual real-world use case, loads, and environment.
All that to say: There cannot possibly be any performance difference. The only way a benchmark can possibly claim otherwise is if the benchmark is broken. Aliases are literally implemented as two names for the same code; it does not matter which name you use.