The phrase "premature optimization" gets thrown around a lot, but it seems to me we'd avoid this whole discussion (and benchmarking, and testing) if we could sa

The phrase "premature optimization" gets thrown around a lot, but it seems to me we'd avoid this whole discussion (and benchmarking, and testing) if we could safely and easily inline the assembly version of the function. After all, a machine instruction is just a library function burnt into hardware. Maybe a better term is "premature nonportability".