RFA: fix bootstrap/25397

[Joern RENNECKE]

Ian Lance Taylor wrote:

>  
>
>Just remove the SUBREG case from the switch.  I don't see any reason
>to keep it around.
>
>OK if bootstrapped, etc.
>  
>
I can't bootstrap this because of the STACK_BOUNDARY problem on i686.  
The patch for that is still pending. See:
http://gcc.gnu.org/ml/gcc-patches/2005-12/msg01055.html .

Anyways I think the patch is so simple that it can be considered obvious 
from a stability standpoint.  What was unclear is if we wanted to patch 
the i686 backend instead to have a single movsi pattern, and/or if we 
wanted to do something so that we can again match SUBREGs of hard regs 
with equivalend hard regs.  Earlier versions of my patch did that but 
bernd thought
that we shoudn't see any SUBREGs of registers after reload.

So, to avoid a bootstrapping deadlock, I have checked in the patch to 
remve the SUBREG case.