Re: [PATCH 2 of 2] cc/gcc: 4.4.5, add arm-softfloat-libgcc patch from buildroot

[Bryan Hundven <bryanhundven at gmail dot com>]

Yann, Khem, All,
On Sat, Nov 13, 2010 at 10:56 AM, Yann E. MORIN
<yann.morin.1998@anciens.enib.fr> wrote:
> Bryan, Khem, All,
>
> On Saturday 13 November 2010 19:09:18 Bryan Hundven wrote:
>> > The patch is needed for a toolchain configured with old ABI (oabi) and
>> > soft float
>> > as old abi is stale and obsolete upstream gcc has no interest in this
>> > patch. that said
>> > you dont need it if you do not build oabi toolchains. Most of stuff
>> > these days use eabi
>> > and eabi toolchains dont require it. Does ct-ng supprt oabi toolchains ?
>
> Yes, crostool-NG still has the ability to build OABI toolchains.
> I know that OABI is deprecated, but there are still systems out
> there that use OABI, and for which upgrading to EABI is simply
> not feasible.
>
>> It may be interesting to know now if Michel's ctng config had eabi turned
>> off.
>
> The tuple for Michael's toolchain was arm-cortexa8-linux-gnu which means
> it is OABI.
>
>> I know that eabi is a boolean option, that defaults to 'y', but can be disabled.
>> So I would assume that oabi is still possible with crosstool-ng.
>
> Yes.
>
>> I have a question about the "-lfloat" in the gcc/config/arm/linux-elf.h.
>> I thought that libfloat was provided by the netwinder floating point
>> emulation, but on modern (gcc > 4.1) toolchains, I don't see any
>> reference to libfloat except in gcc/config/arm/linux-elf.h. Am I
>> missing something?
>
> Googling for "netwinder floating point emulation" gives:
> Âhttp://netwinder.osuosl.org/users/s/scottb/public_html/notes/FP-Notes-all.html
>
> Which sends us to:
> Âhttp://www.eecs.berkeley.edu/~jhauser/arithmetic/softfloat.html
> which is now 404.
>
> Anyway, the latest release of the SoftFloat library was release 2b, dating
> back to May 2002:
> Âhttp://www.jhauser.us/arithmetic/SoftFloat.html
>
> For my part, I wonder why gcc still has reference to -lfloat, when there are
> available softfloat functions bundled with gcc. Are we not supposed to use
> them? If so, why are they present in the first place? Well...

Looking through the gcc commit log, I found CL 113056 where soft-fp
was added from glibc by Joseph Myers (originally only rs6000 was
supported) and CL 132837 Paul Brook added soft-fp support for arm in
t-arm-softfp, but disabled for every platform except ARMv6-M(?). I
don't see any newer commits that actually enable the gcc's soft-fp for
any other arm architecture.

So if I understood that correctly, it sounds like all other arm
architectures are getting soft float from libm, or are expected to
have libfloat from(?):

http://netwinder.osuosl.org/users/s/scottb/libfloat/
(I believe this was the source you were looking for and not "The
SoftFloat Library", although I might be wrong. See:
http://netwinder.osuosl.org/users/s/scottb/public_html/notes/FP-Notes-6.html)
...which is 11 years old!

Anyways, I hope this info is somewhat close to accurate and useful to someone.

It might be useful to ask these questions again on gcc-help, as I'm
sure someone with an oabi arm toolchain is going to bring this up
again in the future.

> Regards,
> Yann E. MORIN.
>
> --
> .-----------------.--------------------.------------------.--------------------.
> | ÂYann E. MORIN Â| Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
> | +33 662 376 056 | Software ÂDesigner | \ / CAMPAIGN Â Â | Â___ Â Â Â Â Â Â Â |
> | +33 223 225 172 `------------.-------: ÂX ÂAGAINST Â Â Â| Â\e/ ÂThere is no Â|
> | http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL  Â|  v  conspiracy. Â|
> '------------------------------^-------^------------------^--------------------'
>
>
>

-Bryan

--
For unsubscribe information see http://sourceware.org/lists.html#faq