[PATCH] [ARM, Fix] Fix emitLeading/TrailingFence on old ARM processors

[JF Bastien]

>>! In D5304#7, @morisset wrote:
> Thanks jf for catching this forgotten test!
> 
> It turned out that this code is dead, as these subtargets completely avoid this
> code path anyway. I have replaced it by llvm_unreachable, and it is tested by
> the ARMv4 test in atomic-load-store.

Is there a test that demonstrates the result of the other code path? I can't find code that generates the barrier in the tests, so it would be nice to add (maybe in a different change).

http://reviews.llvm.org/D5304