https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126240
--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> --- Note, if frange is a union of a few sub-ranges, I think the m_type/m_pos_nan/m_neg_nan bits can be there just once, not per subrange, so it would probably be just a mattter of turning m_min/m_max into array of a few of them (I think we don't need 255 like in irange, 2 or 4 would be good enough), plus a count how many are present.
