[Bug tree-optimization/107569] [13 Regression] Failure to optimize std::isfinite since r13-3596

["aldyh at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107569

--- Comment #29 from Aldy Hernandez <aldyh at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #28)
> (In reply to Aldy Hernandez from comment #27)
> > > As for signed zeros in -fsigned-zeros (default) mode, wonder if we e.g. don't
> > > say sqrt is nonnegative (even when sqrt (-0.0) is -0.0).
> > 
> > It seems tree_call_nonnegative_warnv_p is already doing the right thing for
> > sqrt?
> > 
> >     CASE_CFN_SQRT:
> >     CASE_CFN_SQRT_FN:
> >       /* sqrt(-0.0) is -0.0.  */
> >       if (!HONOR_SIGNED_ZEROS (type))
> > 	return true;
> >       return RECURSE (arg0);
> 
> Ah, ok.  So, if we during stage3 go through all the cases and convince
> ourselves it is good enough, it might be ok for signed zeros then.  Still
> the sign of NAN is separate question from that, and there we should set +NAN
> only when we see
> CASE_CFN_FABS{,_FN} or ABS_EXPR, toggle the sign of NEGATE_EXPR, inherit
> from second argument for CASE_CFN_COPYSIGN{,_FN} and copy over for copies.

From what I've seen, ABS is represented in the IL directly without the need for
a builtin, at least by evrp time.  Is this not always the case?

Negate, copysign, and copy are already handled specially so no need to do
anything there.

I have a patch I'm testing for the generic case.