Re: [rfa/arm] Fix the types of ARM FPA registers

[Daniel Jacobowitz <drow@false.org>]

On Thu, Mar 25, 2004 at 02:53:13PM +0000, Richard Earnshaw wrote:
> > This has apparently been broken for a while... build an arm-elf GDB, connect
> > to the simulator, and say 'print $f0'.  You get an internal error because
> > the virtual size is 8 but the raw size is 12, and we don't define
> > DEPRECATED_REGISTER_CONVERTIBLE.
> > 
> > I don't think any of the baggage is necessary any more.  We can simply
> > expose the correct type.  After this patch, we have:
> > 
> > (gdb) ptype $f0
> > type = builtin_type_arm_ext_littlebyte_bigword
> > (gdb) p sizeof($f0)
> > $1 = 12
> > (gdb) p $f0
> > $2 = 0
> > 
> > I also got to whack five deprecated methods.
> > 
> > No change in test results on arm-sim, and it fixes the internal error.  I
> > don't have any hard-FPA targets to try it on (all my Linux setups are
> > soft-FPA or soft-VFP).
> > 
> 
> I get a few testsuite differences, but these appear to be race conditions 
> of some sort, and not related to floating point values in any way.  No FP 
> tests change.

Thanks.  I've checked this in for HEAD/6.1.

Any differences in the threads tests are ignorable.  I have patches
locally that need a little more testing which will improve the
situation.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer