[LLVMdev] Crash in PBQP register allocator

[Lang Hames]

>> Bernhard and I devised a stronger normalization scheme...
> Is that the one described above?

Yes.

>> *I've got an example around here somewhere. It's more involved, but I
>> can dig it up if you'd like.
>>
> Yes, please.

Ahh it looks like my example was wrong. You're quite right - with the
stronger normalization scheme R1 and R2 should be safe for all nodes.
I'll update the PBQP solver to reflect this.

- Lang.