I believe Lemire's point is that the compiler is still able to inline the function despite the generic definition, so at least in this somewhat narrow case generics are a zero-cost abstraction.