> This is a pure performance optimization, not a correctness requirement. Removing the initial pre-alignment loop doesn't change any results.

Somebody needs to test their code on a CPU that requires alignment before loading 64bit values.