Sources to port fenv.h to ARM and AARCH64

Sat Mar 21 21:19:06 UTC 2020

On Sat, Mar 21, 2020 at 4:05 PM Aditya Upadhyay <aadit0402 at gmail.com> wrote:

> On Sun, Mar 22, 2020 at 1:56 AM Joel Sherrill <joel at rtems.org> wrote:
> >
> >
> >
> > On Sat, Mar 21, 2020, 3:16 PM Aditya Upadhyay <aadit0402 at gmail.com>
> wrote:
> >>
> >> On Sun, Mar 22, 2020 at 1:30 AM Joel Sherrill <joel at rtems.org> wrote:
> >> >
> >> >
> >> >
> >> > On Sat, Mar 21, 2020 at 3:03 AM Aditya Upadhyay <aadit0402 at gmail.com>
> wrote:
> >> >>
> >> >>
> >> >>
> >> >> On Sat, 21 Mar 2020, 10:29 Eshan Dhawan, <eshandhawan51 at gmail.com>
> wrote:
> >> >>>
> >> >>> I went through the FreeBSD and NetBSD sources for implementation
> >> >>> In FreeBSD there is an architecture-specific implementation with
> different header file for each architecture it supports .
> >> >>> Whereas in NetBSD there is a single fenv.h defined but each
> architecture has its own C file to implement the functions.
> >> >>> Also FreeBSD has soft-float for ARM
> >> >>> So, I think FreeBSD would be a better option.
> >> >>
> >> >>
> >> >> Look into this discussion on mail thread:
> >> >> https://sourceware.org/legacy-ml/newlib/2017/msg00818.html
> >> >>
> >> >> and this patch series on newlib mailing list.
> >> >>
> >> >> https://sourceware.org/legacy-ml/newlib/2019/msg00418.html .
> >> >>
> >> >> It will help to understand the fenv-stub code.
> >> >
> >> >
> >> > I am having trouble finding the implementation of any fenv method in
> the NetBSD code. I only am finding the weak alias mappings. Where is an
> example of the body of one of the methods in any architecture?
> >> >
> >>
> >> I have cloned this NetBSD Sources. As I can see those
> >> architecture-specific implemented methods here. Please correct me if I
> >> am looking into different than what you are looking for.
> >>
> >> ./lib/libm/arch/powerpc/fenv.c
> >> ./lib/libm/arch/sh/fenv.c
> >> ./lib/libm/arch/mips64/fenv.c
> >> ./lib/libm/arch/amd64/fenv.c
> >> ./lib/libm/arch/arm/fenv.c
> >> ./lib/libm/arch/i387/fenv.c
> >> ./lib/libm/arch/hppa/fenv.c
> >> ./lib/libm/arch/m88k/fenv.c
> >> ./lib/libm/arch/alpha/fenv.c
> >> ./lib/libm/arch/sparc64/fenv.c
> >> ./lib/libm/arch/aarch64/fenv.c
> >> ./regress/lib/libm/fenv/fenv.c
> >
> >
> > That's what I see browsing the repo.
> >
> > The single file NetBSD looks easy to integrate. Would be similar to the
> x86_64 or i386 there now on file layout.
> >
> > Does the freebsd support more hardware variants?
>
> FreeBSD does not have fenv support for some architecture like RISCV,
> M88k, alpha architecture, I can see netbsd support more
> hardware variants. NetBSD includes fenv support for these architecture
> also.
>

+1

NetBSD has more architectures and looks easier to integrate. But we need to
be
sure the NetBSD implementation for an specific architecture covers all the
cases
that FreeBSD does. If equal, pick the more that is easier to integrate.

--joel

>
> >>
> >>
> >> > --joel
> >> >>
> >> >>
> >> >>
> >> >>>
> >> >>> On Sat, Mar 21, 2020 at 2:37 AM Joel Sherrill <joel at rtems.org>
> wrote:
> >> >>>>
> >> >>>>
> >> >>>>
> >> >>>> On Fri, Mar 20, 2020 at 3:33 PM Eshan Dhawan <
> eshandhawan51 at gmail.com> wrote:
> >> >>>>>
> >> >>>>> thanks, dr Joel
> >> >>>>>
> >> >>>>> I had gone through the musl-libc library but it doesn't have much
> architecture specific support for ARM as well as AARCH64
> >> >>>>> It has support for s390x, m68k, powerpc64
> >> >>>>
> >> >>>>
> >> >>>> This type of evaluation is important.  The architecture may be
> supported in only one implementation or one may be more complete than
> another.
> >> >>>>
> >> >>>> Ignoring the license requirements of course.
> >> >>>>>
> >> >>>>>
> >> >>>>>
> >> >>>>> On Sat, Mar 21, 2020 at 1:32 AM Joel Sherrill <joel at rtems.org>
> wrote:
> >> >>>>>>
> >> >>>>>>
> >> >>>>>>
> >> >>>>>> On Fri, Mar 20, 2020 at 2:43 PM Eshan Dhawan <
> eshandhawan51 at gmail.com> wrote:
> >> >>>>>>>
> >> >>>>>>> What would be the preferred source to port fenv.h to ARM and
> AARCH64
> >> >>>>>>> its implementation is present in both FreeBSD as well AS NetBSD
> >> >>>>>>> -> ARM
> >> >>>>>>> ---FreeBSD Source
> >> >>>>>>> # https://github.com/freebsd/freebsd/tree/master/lib/msun/arm
> >> >>>>>>> ---NetBSD Source
> >> >>>>>>> # https://github.com/NetBSD/src/tree/trunk/lib/libm/arch/arm
> >> >>>>>>>
> >> >>>>>>> ->AARCH64
> >> >>>>>>> ---FreeBSD Source
> >> >>>>>>> #
> https://github.com/freebsd/freebsd/tree/master/lib/msun/aarch64
> >> >>>>>>> ---NetBSD Source
> >> >>>>>>> #
> https://github.com/NetBSD/src/tree/trunk/lib/libm/arch/aarch64
> >> >>>>>>
> >> >>>>>>
> >> >>>>>> Don't forget MUSL-C Library which has a lot of architectures and
> >> >>>>>> is appropriately licensed.
> >> >>>>>>
> >> >>>>>> https://git.musl-libc.org/cgit/musl/tree/src/fenv
> >> >>>>>>
> >> >>>>>> I think our the order is going to be FreeBSD, NetBSD, then other
> places.
> >> >>>>>>
> >> >>>>>> The code drops into newlib's libm in a particular way which may
> require
> >> >>>>>> some refactoring. fenv.h is shared across all ports and
> machine/fenv.h is
> >> >>>>>> where port code goes. There must be an architecture specific
> file for each
> >> >>>>>> method. But the entire implementation could go in one file and
> the others
> >> >>>>>> be stubs. The i386 does this.
> >> >>>>>>
> >> >>>>>> --joel
> >> >>>>>>>
> >> >>>>>>>
> >> >>>>>>>
> >> >>>>>>>
> >> >>>>>>> Thanks
> >> >>>>>>> -Eshan
> >> >>>>>>> _______________________________________________
> >> >>>>>>> devel mailing list
> >> >>>>>>> devel at rtems.org
> >> >>>>>>> http://lists.rtems.org/mailman/listinfo/devel
> >> >>>
> >> >>> _______________________________________________
> >> >>> devel mailing list
> >> >>> devel at rtems.org
> >> >>> http://lists.rtems.org/mailman/listinfo/devel
> >> >>
> >> >> _______________________________________________
> >> >> devel mailing list
> >> >> devel at rtems.org
> >> >> http://lists.rtems.org/mailman/listinfo/devel
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rtems.org/pipermail/devel/attachments/20200321/c1b12423/attachment.html>