https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116319
--- Comment #4 from SHIH YEN-TE <shihyente at hotmail dot com> --- I see, thank you. So, sounds like that if I don't implement the math by myself. For this case, we don't have a STL function to get the answer looks like infinite precision without doing rounding twice? In other word, the smallest c in bfloat16_t must be 2^2*2^-24 + 2^2*2^-(24+7) = 2^-22 + 2^-29.
