When defined (and when using the C++ compiler) the math functions are overloaded with float and long double versions as specified by 26.5 paragraph 6 of the C++ standard. Disabling this flag removed the overloaded functions.