Re: RFA: Build ARM eabi objects by default

[Richard Earnshaw <rearnsha at gcc dot gnu dot org>]

On Wed, 2005-03-16 at 17:26, Daniel Jacobowitz wrote:
> On Wed, Mar 16, 2005 at 05:16:22PM +0000, Richard Earnshaw wrote:
> > On Wed, 2005-03-16 at 16:54, Daniel Jacobowitz wrote:
> > > A customer of ours noticed that invoking arm-eabi-as generates binaries that
> > > aren't marked as EABI ver4.  It makes sense to me to try to get the default
> > > options right, if we're being invoked by hand instead of by gcc.
> > > 
> > > Is this patch OK?  Tested by running the gas testsuite on arm-elf,
> > > arm-linuxeabi, and arm-eabi; and by inspecting the default object flags on
> > > each of those platforms.
> > 
> > 2005-03-16  Daniel Jacobowitz  <dan@codesourcery.com>
> > 
> >         * configure.tgt: Set emulation for arm-*-eabi*.
> >         * config/tc-arm.c (meabi_flags): Check EABI_DEFAULT.
> >         * config/te-armeabi.h: New file.
> >         * config/te-armlinuxeabi.h (EABI_DEFAULT): Define.
> >         * config/te-symbian.h: Include "te-armeabi.h".
> > 
> > This is OK.  However,
> > 
> > diff -N config/te-armeabi.h
> > --- /dev/null   1 Jan 1970 00:00:00 -0000
> > +++ config/te-armeabi.h 16 Mar 2005 15:46:34 -0000
> > @@ -0,0 +1,8 @@
> > +/* The EABI requires the use of VFP.  */
> > +#define FPU_DEFAULT FPU_ARCH_VFP_V2
> > 
> > You might want to look at whether the code in md_begin() is doing the
> > right thing for you here.
> 
> Could you clarify what you mean by "the right thing"?  I think it will;
> for Linux we will use FPU_DEFAULT, and for arm-eabi we will try the
> processor and then fall back to FPU_DEFAULT if no processor is
> specified.
> 

It's a nightmare of legacy.  On platform ABIs the platform defines the
floating point format.  On non-platform builds (raw arm-elf, arm-eabi)
the floating-point format is determined by the CPU requested (for pre-v5
we default to FPA ordering, for v5 and above we default to VFP).  In
those environments FPU_DEFAULT only affects assembly when none of -mcpu,
-mfpu or -march is specified.

> Hmm, I see that may not work quite right in the arm-eabi case... man,
> this is twisty.

Yep.  Your in a little maze of twisty passages, all different.

R.