I will note also that GCC makes __float128 and _Float128 the same type in
C, but it mangles __float128 the same as (128-bit) long double in C++.
C++'s std::float128(_t) cannot be the same as __float128 (as it is). So the
relationship between __float128 and the standardized types may need to
differ between GNU C and GNU C++.