https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119131
--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> --- But e.g. 0.0e-12df shouldn't be treated like that. As can be seen on _Decimal32 a = 0.0e-12df; _Decimal32 b = 0.0e-98df; _Decimal32 c = 0.0e-99df; _Decimal32 d = 0.0e-100df; _Decimal32 e = 0.0e-101df; _Decimal32 f = 0.0e+89df; _Decimal32 g = 0.0e+90df; _Decimal32 h = 0.0e+91df; _Decimal32 i = 0.0e+92df; _Decimal32 j = 0.0e+93df; d and e already are represented as 32-bit zero (and anything smaller than that), while i and j are equal to h in value. So, for decimal it is best for mantissa 0 (at least if sign is not negative) to check if it reinterprets as integral zero.
