[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20250819151716.GB1827@brightrain.aerifal.cx>
Date: Tue, 19 Aug 2025 11:17:17 -0400
From: Rich Felker <dalias@...c.org>
To: Damian McGuckin <damianm@....com.au>
Cc: MUSL <musl@...ts.openwall.com>
Subject: Re: Should asinf(x) raise underflow for subnormal 'x'
On Tue, Aug 19, 2025 at 04:59:33PM +1000, Damian McGuckin wrote:
> On Mon, 18 Aug 2025, Rich Felker wrote:
>
> > On Mon, Aug 18, 2025 at 04:43:46PM +1000, Damian McGuckin wrote:
> > >
> > > The routine correctly returns 'x' for subnormal 'x'.
> > >
> > > For comparison, atan(x) for subnormal x returns 'x' and raises underflow.
> > >
> > > Should not asinf(x) also do the same?
> > >
> > > Also, should not both of these raise inexact when the result is small but
> > > still normal, i.e. the return value is just 'x'.
> >
> > This all sounds right and like it's an oversight.
>
> The code as it is written would satisfy IEEE 754 1985 but not the later
> variants which tightened that up. And that code predates the later variants.
Sounds right. I'd welcome a fix for this. I think the double and ld
versions might be affected too.
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.
