Re: clang+libc++ using missing powl

On Fri, Sep 6, 2013 at 11:59 AM, Steve Kargl <
sgk_at_troutmask.apl.washington.edu> wrote:

> On Fri, Sep 06, 2013 at 07:54:46AM +0100, David Chisnall wrote:
> >
> > On 5 Sep 2013, at 22:09, Steve Kargl <sgk_at_troutmask.apl.washington.edu>
> wrote:
> >
> > > On Thu, Sep 05, 2013 at 09:52:13AM +0100, David Chisnall wrote:
> > >> On 4 Sep 2013, at 23:38, Baptiste Daroussin <bapt_at_freebsd.org> wrote:
> > >>
> > >>> As a result we have a lot of fallouts of ports complaining about:
> > >>> undefined reference to `powl'
> > >>>
> > >>> It seems like libc++ is relying on a function we don't have yet
> > >>> in libm, am I missing something?
> > >>
> > >> I've attached a diff that I'd like to commit to msun
> > >
> > > Why not disable libc++ from exposing these functions?
> >
> > Because they're in the C++ spec and they should work.
> > I consider this a placeholder until we have the real versions
> > in the tree (the wiki says you have coshl, sinhl, and tanhl,
> > in progress, any idea of an ETA for them)?
>
> Well, your commit has pre-empted any discussion on whether
> there would have been a better kludge.  Oh well.
>
> Concerning coshl, sinhl, and tanhl.  I had integrated bde's
> code into msun and prepared a patch to commit over a week ago.
> Unfortunately, my testing on sparc64 revealed a few issues
> with tanhl, and Bruce and I are still discussing the fix.
>
> PS: I have working erfl and erfcl for ld80 archs.  I'm still
> testing and refining the code.  It turns out that computing
> the needed rational approximation is fairly difficult (at least
> for me).
>
> --
> Steve
>


Not only for you , because function values are not very "smooth" , and
representation by an approximation is very difficult . During many weeks ,
I am trying to obtain good approximations to cumulative normal distribution
and its inverse , without very much success .

Approximation to
inverse cumulative normal distribution is much more difficult than
cumulative normal distribution .

I am using Fortran double precision . I think , it is necessary to use an
arbitrary precision package for optimization , but I do not have any one .

Thank you very much .


Mehmet Erol Sanliturk