RFR: Streamline+cleanup ShenandoahBarrierSetC1 and related code

[Roland Westrelin]

> http://cr.openjdk.java.net/~rkennke/c1-bs-cleanups/webrev.00/

So you reverted back to upstream code for c1_LIRGenerator_xxx.cpp? The
change to LIR_Opr LIRGenerator::atomic_cmpxchg() on x86 should be kept I
think because it fixes an inconsistency.

Isn't there some missing aarch64 code:
Wouldn't the changeshenandoahBarrierSetC1_x86.cpp needed on aarch64 too?
In shenandoahBarrierSetC1.cpp,
ShenandoahBarrierSetC1::atomic_xchg_at_resolved(), the logic you use
seems to be a copy of the one for x86 but the aarch64 is different and
needs a temp register?

Roland.