On Fri, Oct 14, 2022 at 5:00 PM Jakub Jelinek <ja...@redhat.com> wrote: > > On Fri, Oct 14, 2022 at 04:53:13PM +0200, Aldy Hernandez wrote: > > > This looks wrong to me. > > > !HONOR_NANS is different from !HONOR_SIGNED_ZEROS. > > > The former says that either NaNs aren't supported or if they appear, > > > it will be UB. > > > The latter says that either -0.0 doesn't exist, or user doesn't care > > > if -0.0 or 0.0 is used. > > > > > > So, what you do is ok for !MODE_HAS_SIGNED_ZEROS (TYPE_MODE (m_type)), > > > but otherwise we want to canonicalize [x, -0.0] to [x, 0.0] and > > > [0.0, y] to [-0.0, y]. > > > > If the user doesn't care, I would expect they'd be ok with it being > > +0.0, but I must say, this is way beyond my paygrade. > > Unlike the NaN case where they can (easily) arrange for NaNs not to appear > (say, avoid numerically undefined operations), for +/-0 if the hardware > supports it they don't have much choice, sometimes computation will yield > one, sometimes the other. The option is "I don't use anything that depends > on the zero sign, which is e.g. copysign from zero, signbit or poking at the > bit patterns". > > > How does this patch in testing look? > > LGTM (perhaps some comment would be nice though).
Ughh, can I trouble you for one, cause frankly I'm a bit lost on this one? Aldy
