huge error for pow

huge error for pow

[Paul Zimmermann]

       Hi,

while testing Newlib 4.3.0, I noticed the following huge error:

zimmerma@salade:~/svn/tbd/20/src/binary64$ VERBOSE=-v ./doitb.newlib pow 1000
Checking pow with newlib-4.3.0.20230120
Using seed 1619945
pow 0 -1 -0x1.647ff80007ff8p-576,-0x1.3d018267f12fp+48 [inf] inf inf
libm gives inf
mpfr gives -inf

The exponent y in x^y is a large odd integer, thus since x is negative we
should get a negative number as result.

This issue was apparently already present in 3.3.0, but we only discovered
it now.

Best regards,
Paul

Re: huge error for pow

[Jeff Johnston]

[-- Attachment #1: Type: text/plain, Size: 726 bytes --]

Thanks Paul,

Do you think you can put together a patch?

-- Jeff J.

On Tue, Feb 7, 2023 at 9:27 AM Paul Zimmermann <Paul.Zimmermann@inria.fr>
wrote:

>        Hi,
>
> while testing Newlib 4.3.0, I noticed the following huge error:
>
> zimmerma@salade:~/svn/tbd/20/src/binary64$ VERBOSE=-v ./doitb.newlib pow
> 1000
> Checking pow with newlib-4.3.0.20230120
> Using seed 1619945
> pow 0 -1 -0x1.647ff80007ff8p-576,-0x1.3d018267f12fp+48 [inf] inf inf
> libm gives inf
> mpfr gives -inf
>
> The exponent y in x^y is a large odd integer, thus since x is negative we
> should get a negative number as result.
>
> This issue was apparently already present in 3.3.0, but we only discovered
> it now.
>
> Best regards,
> Paul
>
>
>

Re: huge error for pow

[Paul Zimmermann]

       Hi Jeff,

this would not make sense on our side, since in CORE-MATH we are currently
completing a correctly rounded pow function, which will be available in a
few weeks, and can then be integrated into newlib.

Best regards,
Paul

> From: Jeff Johnston <jjohnstn@redhat.com>
> Date: Tue, 7 Feb 2023 16:48:14 -0500
> Cc: newlib@sourceware.org, vincenzo.innocente@cern.ch
> 
> 
> [1:text/plain Show]
> 
> 
> [2:text/html Hide Save:noname (1kB)]
> 
> Thanks Paul,
> 
> Do you think you can put together a patch?
> 
> -- Jeff J.
> 
> On Tue, Feb 7, 2023 at 9:27 AM Paul Zimmermann <Paul.Zimmermann@inria.fr>
> wrote:
> 
>         Hi,
> 
>  while testing Newlib 4.3.0, I noticed the following huge error:
> 
>  zimmerma@salade:~/svn/tbd/20/src/binary64$ VERBOSE=-v ./doitb.newlib pow
>  1000
>  Checking pow with newlib-4.3.0.20230120
>  Using seed 1619945
>  pow 0 -1 -0x1.647ff80007ff8p-576,-0x1.3d018267f12fp+48 [inf] inf inf
>  libm gives inf
>  mpfr gives -inf
> 
>  The exponent y in x^y is a large odd integer, thus since x is negative
>  we
>  should get a negative number as result.
> 
>  This issue was apparently already present in 3.3.0, but we only
>  discovered
>  it now.
> 
>  Best regards,
>  Paul

Re: huge error for pow

[Paul Zimmermann]

       Dear Corinna,

a correctly rounded pow function will soon be available in CORE-MATH,
stay tuned: https://core-math.gitlabpages.inria.fr/

Best regards,
Paul

> Date: Tue, 11 Apr 2023 11:18:06 +0200
> From: Corinna Vinschen <vinschen@redhat.com>
> 
> On Apr  8 08:00, Paul Zimmermann wrote:
> >        Hi,
> > 
> > a gentle reminder for this issue reported two months ago:
> > 
> > https://sourceware.org/pipermail/newlib/2023/020164.html
> > 
> > Paul
> 
> The patch is missing ¯\_(ツ)_/¯
> 
> 
> Corinna
> 
> 

Re: huge error for pow

[Corinna Vinschen]

On Apr  8 08:00, Paul Zimmermann wrote:
>        Hi,
> 
> a gentle reminder for this issue reported two months ago:
> 
> https://sourceware.org/pipermail/newlib/2023/020164.html
> 
> Paul

The patch is missing ¯\_(ツ)_/¯


Corinna

huge error for pow

[Paul Zimmermann]

       Hi,

a gentle reminder for this issue reported two months ago:

https://sourceware.org/pipermail/newlib/2023/020164.html

Paul