[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20260803172445.GQ27423@brightrain.aerifal.cx>
Date: Mon, 3 Aug 2026 13:24:45 -0400
From: Rich Felker <dalias@...c.org>
To: musl@...ts.openwall.com
Subject: Re: fix acosh for x<0
On Mon, Aug 03, 2026 at 06:41:12PM +0200, Szabolcs Nagy wrote:
> * Rich Felker <dalias@...c.org> [2026-08-03 12:00:25 -0400]:
> > On Mon, Aug 03, 2026 at 02:10:37PM +0200, Szabolcs Nagy wrote:
> > > >From 84bf2a51418196f96192d7fc8f11cf268b02e1b1 Mon Sep 17 00:00:00 2001
> > > From: Szabolcs Nagy <nsz@...t70.net>
> > > Date: Mon, 3 Aug 2026 06:41:28 +0000
> > > Subject: [PATCH 1/2] math: fix acosh for x<0
> > >
> > > acosh(-0x1.8p15) returned -3.7534177368329567 instead of nan.
> > >
> > > the same issue got fixed for acoshf and acoshl in commits
> > > c4c38e6364323b6d83ba3428464e19987b981d7a and
> > > 6d10102709df4bc966d2846c1c45cd667e5048e5 here we follow the latter.
> > >
> > > reported by Paul Zimmermann.
> > > ---
> > > src/math/acosh.c | 13 +++++++------
> > > 1 file changed, 7 insertions(+), 6 deletions(-)
> > >
> > > diff --git a/src/math/acosh.c b/src/math/acosh.c
> > > index badbf908..75ce95aa 100644
> > > --- a/src/math/acosh.c
> > > +++ b/src/math/acosh.c
> > > @@ -9,16 +9,17 @@
> > > double acosh(double x)
> > > {
> > > union {double f; uint64_t i;} u = {.f = x};
> > > - unsigned e = u.i >> 52 & 0x7ff;
> > > -
> > > - /* x < 1 domain error is handled in the called functions */
> > > + unsigned e = u.i >> 52;
> > >
> > > if (e < 0x3ff + 1)
> > > - /* |x| < 2, up to 2ulp error in [1,1.125] */
> > > + /* 0 <= x < 2, up to 2ulp error in [1,1.125] */
> > > return log1p(x-1 + sqrt((x-1)*(x-1)+2*(x-1)));
> > > if (e < 0x3ff + 26)
> > > - /* |x| < 0x1p26 */
> > > + /* 2 <= x < 0x1p26 */
> > > return log(2*x - 1/(x+sqrt(x*x-1)));
> > > - /* |x| >= 0x1p26 or nan */
> > > + if (e & 0x800)
> > > + /* x < 0 or x = -0, invalid */
> > > + return (x-x)/(x-x);
> > > + /* x >= 0x1p26 or nan */
> > > return log(x) + 0.693147180559945309417232121458176568;
> > > }
> >
> > I don't think you need the special case for negative. The final case
> > of passing a negative x to log will already produce a nan.
>
> unfortunately it is needed for -0
> log(-0)=-inf instead of nan
OK, let's go with your fix then.
Rich
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
