[gccsdk] GCC upstream possibly deprecating FPA support?

John Tytgat John.Tytgat at aaug.net
Mon Jul 26 14:23:13 PDT 2010


In message <20100725205946.GO27483 at chiark.greenend.org.uk>
          Theo Markettos <theo at markettos.org.uk> wrote:

> I don't know if anyone (John?) reads the main GCC list, but I've just come
> across this thread from a few weeks ago:
> 
> http://www.mail-archive.com/gcc@gcc.gnu.org/msg51554.html
> 
> Various people are suggesting pulling FPA support out of GCC 4.7.  I don't
> know if anyone here has a view on that, but if you do you may wish to get
> involved in the discussion.

Yes, I saw that discussion but refrained from giving a public reaction. My
reasoning is based on the question why this would still be a matter
to us ?

First of all, with the 4.1.1/4.1.2 port we currently have, we're not using
FPA in compiled code at all when UnixLib is used (as this is soft-float
based).  FPA instructions are only emitted for SCL based code (-mlibscl or
-mmodule option).

Secondly, assuming the port 4.6 results in a succes, we can use this gcc
codebase in the same way like we're doing today and this for the next
several years.

During that period I would rather see a consensus on adopting and
deploying a new ABI which we could use on RISC OS with UnixLib and SCL
which is a bit more efficient than APCS-R/APCS-32 and allows the use
of NEON/VFP (either hardfloat, either softfp).  Basically one of the
AAPCS flavours which the rest of the world is now using.

When we would jump on that train, and have support for this ABI in SCL,
FPA would completely be a thing of the past for us as well (note, that
does not mean removing APCS-32 support from SCL as it is still useful
for the legacy programs).

John.
-- 
John Tytgat, in his comfy chair at home                                 BASS
John.Tytgat at aaug.net                             ARM powered, RISC OS driven




More information about the gcc mailing list