C++ does it automatically, I think this is a call for Google to spend some more time optimizing floating point workloads.