[PATCH v2 3/8] math: Use sinh from CORE-MATH

[Paul Zimmermann]

       Hi DJ,

these constants are error bounds, indeed they don't need full precision.
I have rounded them up to 13 bits, which corresponds to an increase of
at most 0.01%, which should be much less than the probability of failure
of the fast path.

Paul

> From: DJ Delorie <dj@redhat.com>
> Cc: adhemerval.zanella@linaro.org, libc-alpha@sourceware.org,
>         Wilco.Dijkstra@arm.com, josmyers@redhat.com
> Date: Wed, 11 Mar 2026 18:42:37 -0400
> 
> Paul Zimmermann <Paul.Zimmermann@inria.fr> writes:
> > thank you DJ, I changed to hexadecimal constants upstream:
> >
> > -      double e = 0.12e-18*th, lb = rh + (rl - e), ub = rh + (rl + e);
> > +      double e = 0x1.1b6p-63*th, lb = rh + (rl - e), ub = rh + (rl + e);
> >
> > -    double e = 0.122e-18*rh, lb = rh + (rl - e), ub = rh + (rl + e);
> > +    double e = 0x1.202p-63*rh, lb = rh + (rl - e), ub = rh + (rl + e);
> >
> > -    double e = 0.38e-18*rh, lb = rh + (rl - e), ub = rh + (rl + e);
> > +    double e = 0x1.c0ap-62*rh, lb = rh + (rl - e), ub = rh + (rl + e);
> 
> That change alone tells me these constants don't need the precision of a
> p format, because those values don't exactly match ;-)
> 
> I still think the consistency is worth the change though, because you
> can specify how many bits of useful precision you need.
> 
>