https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125127
--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> --- The fact that it has double type doesn't imply in which precision it is evaluated. See e.g. C23 5.2.4.2.2/19, ia32 is FLT_EVAL_METHOD 2 target, so 2 evaluate all operations and constants to the range and precision of the long double type See PR323 for some more details.
