On Mon, Sep 5, 2022 at 11:28 AM Aldy Hernandez <al...@redhat.com> wrote: > > On Mon, Sep 5, 2022 at 11:14 AM Richard Biener > <richard.guent...@gmail.com> wrote: > > > > On Mon, Sep 5, 2022 at 11:06 AM Jakub Jelinek <ja...@redhat.com> wrote: > > > > > > On Mon, Sep 05, 2022 at 11:00:54AM +0200, Richard Biener wrote: > > > > On Mon, Sep 5, 2022 at 8:24 AM Aldy Hernandez via Gcc-patches > > > > <gcc-patches@gcc.gnu.org> wrote: > > > > > > > > > > Intersecting two ranges where one is a NAN is keeping the sign bit of > > > > > the NAN range. This is not correct as the sign bits may not match. > > > > > > > > > > I think the only time we're absolutely sure about the intersection of > > > > > a NAN and something else, is when both are a NAN with exactly the same > > > > > properties (sign bit). If we're intersecting two NANs of differing > > > > > sign, we can decide later whether that's undefined or just a NAN with > > > > > no known sign. For now I've done the latter. > > > > > > > > > > I'm still mentally working on intersections involving NANs, especially > > > > > if we want to keep track of signbits. For now, let's be extra careful > > > > > and only do things we're absolutely sure about. > > > > > > > > > > Later we may want to fold the intersect of [NAN,NAN] and say [3,5] > > > > > with the posibility of NAN, to a NAN, but I'm not 100% sure. > > > > > > > > The intersection of [NAN, NAN] and [3, 5] is empty. The intersection > > > > of [NAN, NAN] and VARYING is [NAN, NAN]. > > > > > > I think [3.0, 5.0] printed that way currently means U maybe NAN, > > > it would be [3.0, 5.0] !NAN if it was known not to be NAN. > > > > Uh, that's confusing. So [3, 5] U maybe NAN intersected with > > ][ NAN is ][ NAN. [3, 5] !NAN intersected with ][ NAN is ][ !NAN. > > I'm confused. What's ][ ??.
It's the empty range. > For clarity in the discussion, let's say ?NAN, NAN, and !NAN for the > NAN property. > > I would expect: > [3,5] ?NAN U NAN = [3,5] ?NAN > [3,5] !NAN U NAN = [3,5] ?NAN > [3,5] !NAN ^ NAN = [] > NAN !SIGN ^ NAN SIGN = [] (differing signs) > [3,5] ?NAN ^ NAN = NAN > [3,5] !NAN ^ NAN = [] > > Also, definite NANs must have a real_nan() on both sides of the > endpoints. They must be the same. And that real_nan() could have a > sign bit. So we could have: > [NAN, NAN] ?SIGN (sign unknown-- default) > [NAN, NAN] SIGN (negative NAN) > [NAN, NAN] !SIGN (positive NAN) > > The above is enforced by the setter and verify_range. I think having NAN in the endpoints is confusing since NANs do not behave with ordered compares. That is, [-NAN, +NAN] would not make sense. A value-range of either -INF or +INFs would be a two element [-INF, -INF] U [+INF, +INF] range. So a definite NAN should IMHO be ][ (empty range of non-NAN values) U NAN. And if there's a negative and a positive NAN then we probably should simply have two NAN flags, +NAN and -NAN. I'm not sure how tracking the sign bit separately is useful. "sign bit set" is simply [-INF, -0] U -NAN (if the value could be a NAN)? > Note, that setting the definite NAN property (fp_prop::YES) to a > range, makes it a NAN. That is, we will forcibly change the range to > [NAN, NAN]. There's also an assert making sure you're not setting > !NAN on a [NAN, NAN]. > > A varying has all the property bits set to unknown. So effectively > ?NAN and ?SIGN. > > Do you agree? No, I'm very confused about having [3, 5] ?NAN, [3, 5] NAN and [3, 5] !NAN. Richard. > Aldy > > > > > In fact [3, 5] U maybe NAN is just [3, 5] U NAN, there's no "maybe" ranges, > > if the value may be NAN then NAN is in the value-range. So it's either > > [3, 5] U NAN or [3, 5] (without U NAN). > > > > Richard. > > > > > > > > Jakub > > > > > >
