question on cmath (gcc 4.5.1)

[Neal Becker]

cmath has patterns like this (taking cos as an example):

  using ::cos;

  inline float
  cos(float __x)
  { return __builtin_cosf(__x); }

  inline long double
  cos(long double __x)
  { return __builtin_cosl(__x); }

  template<typename _Tp>
    inline typename __gnu_cxx::__enable_if<__is_integer<_Tp>::__value, 
					   double>::__type
    cos(_Tp __x)
    { return __builtin_cos(__x); }

Do I understand correctly, that this means float, long double will use 
__builtin_cos(xxx), but plain double will not use __builtin_cos???

I've tried tracing the generated code, and that's the way it looks.  It 
appears to just call the cos function in glibc.

Is this correct and if so, would not __builtin_cos be faster?