Re: Error of tan/tanf in libm.

["Tim Prince" <tprince at computer dot org>]

I haven't looked that deeply; I simply ran elefunt tests on the
functions used by cygwin.  I think the results were consistent with your
assessment of 3 ulps relative error.
----- Original Message -----
From: "Stephen L Moshier" <moshier@mediaone.net>
To: "Tim Prince" <tprince@computer.org>
Cc: <newlib@sources.redhat.com>
Sent: Saturday, April 14, 2001 7:37 PM
Subject: Re: Error of tan/tanf in libm.


>
> > The only libm functions in newlib which I've found to be substandard
> > are acosf() / asinf().
>
> newlib-1.9 has at least two versions: newlib/libm/math/ef_asin.c,
> newlib/libm/mathfp/sf_asine.c.  Are you referring to one of these?
> sf_asine.c seems to have about 2 ulps relative error, ef_asin.c
> has 3 ulps.
>
> The main problem with ef_asin.c is that the values
> of pio2_hi, pio2_lo, pio4_hi are chosen incorrectly for IEEE
> (strictly rounded) arithmetic.  If someone would like a patch
> to fix ef_asin.c, I can give you one written for the file e_asinf.c in
> glibc which I think is the same code.  The fix yields less than 1 ulp
> relative error.
>