[head tinderbox] failure on arm/arm

[Nicholas Clark]

On Mon, Nov 13, 2006 at 12:44:12AM +0100, Olivier Houchard wrote:
> On Sun, Nov 12, 2006 at 11:27:42PM +0000, Nicholas Clark wrote:
> > On Mon, Nov 13, 2006 at 12:34:34AM +0100, Olivier Houchard wrote:
> > 
> > > No, we're not using the mixed endian IEEE 64bits representation. We're
> > > defaulting to softfloat VFP. What would be te point of switching ?
> > 
> > >From my limited understanding of these things (mostly observing on the
> > ARM Linux lists) absolutely none. The mixed endian IEEE representation
> > is a complete pain, I'm unaware of any reason why it was chosen over a
> > conventional little endian representation (probably back some time in
> > 1987).
> > 
> 
> I thought so :)
> I think FPA is used for historical reasons, because that's what some older 
> arm cpus used when they had a FPU. And of course using a kernel FPE was a
> great idea for linux too.

But I think that the FPA (the external floating point unit) only was only
actually produced to used with the 25 Mhz ARM 3s (*), which I think was 1990
or so, which I think would be the first time that mixed endian layout was
set in silicon. But the mixed endian layout would have had to have been
chosen before RISC OS 2 shipped in 1989, as it had bundled applications
written in C, which uses the (emulated) floating point instructions.
Hence as far as I know there was no hardware reason to choose that layout -
hardware came later.

Nicholas Clark


* because at the time I upgraded my parents' Archimedes to an ARM 3, I asked
  about this, and the 30 Mhz CPU-on-a-daughterboards then available would not
  suitable for adding an FPA, not that it mattered, as the run of FPAs was
  now all sold)