Use of extern in in the implementation of double floats from an external library

[MichaelMeissner]

But when you get to double precision done in hardware, it often is the same speed or faster to do double precision than single precision, for most calculations other than division and square roots.