I don't remember the original reasoning for having the separate capi/float.cpp,
and all the other runtime/ files have their CAPI methods inline.