ARM patch: Fix Thumb addsi3_cbranch pattern

[Richard Earnshaw]

On Tue, 2010-03-23 at 14:25 +0100, Bernd Schmidt wrote:
> On 03/23/2010 02:05 PM, Richard Earnshaw wrote:
> > 
> > On Tue, 2010-03-23 at 12:13 +0100, Bernd Schmidt wrote:
> >>> 	* config/arm/arm.md (addsi3_cbranch): If destination is a high
> >>> 	register, inputs must be low registers and we need a low register
> >>> 	scratch.  Handle alternative 2 like alternative 3.
> > Is there a PR for this?
> 
> No.  I discovered it while testing a different patch, but that only
> perturbed register allocation sufficiently to expose the problem.
> 
> >  At this stage we really should be focused on closing out critical bugs.
> 
> It's a wrong-code bug, so I'd consider it critical.
> 
> 
> Bernd

This is OK for trunk.

Before putting this on release branches, I think there should be a PR
describing the bug in a bit more detail (and ideally a testcase); the PR
should also explain that it could potentially also affect the release
branches (even if it's hard to reproduce).  Once we have that, we I can
then approve it for those branches as well.

R.