On Fri, Oct 14, 2022 at 04:26:50PM +0200, Aldy Hernandez via Gcc-patches wrote:
> Similar to what we do for NANs when !HONOR_NANS and Inf when
> flag_finite_math_only, we can remove -0.0 from the range at creation
> time.
>
> We were kinda sorta doing this because there is a bug in
> real_isdenormal that is causing flush_denormals_to_zero to saturate
> [x, -0.0] to [x, +0.0] when !HONOR_SIGNED_ZEROS. Fixing this bug
> (upcoming), causes us to leave -0.0 in places where we aren't
> expecting it (the intersection code).
>
> gcc/ChangeLog:
>
> * value-range.cc (frange::set): Drop -0.0 for !HONOR_SIGNED_ZEROS.
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].
> --- a/gcc/value-range.cc
> +++ b/gcc/value-range.cc
> @@ -324,6 +324,14 @@ frange::set (tree type,
> m_neg_nan = false;
> }
>
> + if (!HONOR_SIGNED_ZEROS (m_type))
> + {
> + if (real_iszero (&m_min, 1))
> + m_min.sign = 0;
> + if (real_iszero (&m_max, 1))
> + m_max.sign = 0;
> + }
> +
> // For -ffinite-math-only we can drop ranges outside the
> // representable numbers to min/max for the type.
> if (flag_finite_math_only)
> --
> 2.37.3
Jakub
