are most floating point cases in tree_call_nonnegative_warnv_p() wrong for HONOR_NANS?

[Aldy Hernandez]

I suppose most of the calls to tree_expr_nonnegative_p that apply to
floats in match.pd and tree-ssa-math-opts.cc are guarded by
!HONOR_NANS || tree_expr_nonnegative_p.  Still feels kinda delicate...

Aldy

On Sun, Nov 13, 2022 at 4:56 PM Aldy Hernandez <aldyh@redhat.com> wrote:
>
> Based on discussions in the last few weeks, aren't most of the cases in
> tree_call_nonnegative_warnv_p() wrong when honoring NANS?
>
> For example:
>      CASE_CFN_ACOS:
>      CASE_CFN_ACOS_FN:
>      CASE_CFN_ACOSH:
>      CASE_CFN_ACOSH_FN:
> ...
> ...
>        /* Always true.  */
>        return true;
>
> But are we guaranteed a +NAN for any NAN input?  I thought we were only
> guaranteed the NAN sign for abs, copysign, assignment, etc?  Similarly
> for most other cases in this function.
>
> Hmmm.  I really think a good chunk of fold-const.cc should live in
> range-ops.  It seems we're duplicating a lot of functionality.
> Similarly to bit-CCP as I've mentioned.
>
> Aldy