Re: issue with tgammaf

["Keith Packard" <keithp@keithp.com>]

[-- Attachment #1: Type: text/plain, Size: 984 bytes --]

Paul Zimmermann <Paul.Zimmermann@inria.fr> writes:

> thank you Keith. For your information I have updated my comparison with
> Newlib 3.3.0:
>
> https://members.loria.fr/PZimmermann/papers/accuracy.pdf
>
> The differences with the previous (unknown) version of Newlib I used are
> that j1 is less accurate (largest error goes from 2.66e6 ulps to 1.68e7 ulps),
> and y1 is more accurate (largest error goes from 4.65e8 ulps to 6.18e6 ulps).
> (However, those huge errors simply tell us that j1f is completely off, in the
> first case for x=0x6.cfd78p+100 j1f returns 0 instead of -0x1p-52.)
>
> I am still unable to compile a program using lgammaf:

You might have more luck with lgammaf_r as that avoids use of
_impure_ptr in newlib.

Alternatively, you can use the picolibc fork of newlib which uses native
thread local storage instead of a giant struct full of library global
data. That has built-in support for compiling on x86_64 for testing.

-- 
-keith

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 832 bytes --]