ARM s_register_operand

[Paolo Bonzini]

On 06/14/2010 01:31 PM, Bernd Schmidt wrote:
> Is there any point to the s_register_operand predicate used in arm.md?
> Whenever I make changes to a pattern it makes me uneasy because I don't
> understand why we're not simply using register_operand like every other
> port.
>
> So, I propose the following patch:
>    s/s_register_operand/register_operand/g
> (and also removing the definition of the predicate)
>
> I've not been able to find code generation differences in either ARM or
> Thumb1.  A testsuite run is in progress.  Ok?

I had the impression that it is meant to avoid matching (subreg (mem))?

Paolo