Re: [PATCH] Remove multiple defination error

On Mon, Jul 13, 2020 at 12:48 AM Sebastian Huber <
[email protected]> wrote:

> On 13/07/2020 07:36, Sebastian Huber wrote:
> > On 11/07/2020 19:18, Eshan Dhawan wrote:
> >>
> >>
> >> On Sat, Jul 11, 2020 at 8:46 PM Sebastian Huber
> >> <[email protected]
> >> <mailto:[email protected]>> wrote:
> >>
> >>     On 11/07/2020 00:41, Eshan dhawan wrote:
> >>
> >>      > Signed-off-by: Eshan dhawan<[email protected]
> >>     <mailto:[email protected]>>
> >>      > ---
> >>      >   newlib/libm/machine/arm/fenv-vfp.c |  7 ++++-
> >>      >   newlib/libm/machine/arm/fenv.c     | 47
> >>     +++++++++++++++++++-----------
> >>      >   2 files changed, 36 insertions(+), 18 deletions(-)
> >>
> >>     What are the design goals of this implementation? Should it work
> with
> >>     some sort of run time VFP detection? Should it select the support
> >> using
> >>     builtin defines?
> >>
> >> It used the GCC flags to detect VPF
> >> if VPF is not selected all the symbols of vfp won't appear :)
> >>
> >>
> >>     I would use builtin defines only and then just use a single file.
> >>
> >> I tried converting into a single file but the file becomes too complex
> >> to handle and check for errors. :)
> >
> > Complexity is always a bit subjective.
> >
> > How did you test the ARM implementation? I doubt the soft-float support
> > works at all. FreeBSD ships its own soft-float library. Newlib uses the
> > one from libgcc by default. They are not compatible.
> >
> > I guess the MIPS soft-float fenv support has this issue as well.
> >
> > Having an GCC incompatible soft-float fenv support in Newlib makes no
> > sense to me. How can we fix this? Disable it completely or return some
> > run time errors?
> >
> > Maybe we should remove the ARM fenv support from Newlib until these
> > issues are fixed and a working implementation is available.
>
> I noticed that the PowerPC fenv support unconditionally assumes that a
> floating-point unit is available. Not all PowerPC processors support a
> floating-point unit and some applications may intentionally use
> soft-float. Using this code may result in an unimplemented exception if
> MSR[FP]=1, or an FP Unavailable exception if MSR[FP]=0.
>
> If this is a desirable behaviour, then we can use the same on ARM. Just
> provide the VFP implementation. If you use the stuff in other setups you
> get an exception.
>

The RISC-V would be a good model. Jim Wilson provided some default
behavior when there is no hard FPU.

--joel

>
> --
> Sebastian Huber, embedded brains GmbH
>
> Address : Dornierstr. 4, D-82178 Puchheim, Germany
> Phone   : +49 89 189 47 41-16
> Fax     : +49 89 189 47 41-09
> E-Mail  : [email protected]
> PGP     : Public key available on request.
>
> Diese Nachricht ist keine geschäftliche Mitteilung im Sinne des EHUG.
>