RFR: Load reference barriers

[Roman Kennke]

>> http://cr.openjdk.java.net/~rkennke/load-ref-barriers/webrev.00/
> 
> In ShenandoahBarrierSetC2::final_graph_reshaping() for case
> Op_ShenandoahLoadReferenceBarrier: shouldn't we have:
> 
> assert(false, "should have been expanded already");
> 
> ?

Correct! Fixed!

> ShenandoahBarrierC2Support::call_wb_stub(): should that be renamed?
> wb_mem seems useless?

Right! And alias too! Fixed.

Do you think it's still useful to have this code:
  Node* base = find_bottom_mem(ctrl, phase);
  MergeMemNode* mm = MergeMemNode::make(base);
  mm->set_memory_at(Compile::AliasIdxRaw, raw_mem);

or would mm already contain the raw_mem?

> Otherwise looks good to me.

Thanks. Updated webrev that contains the C2 changes and the whitespace
stuff that Aleksey asked for:

Incremental:
http://cr.openjdk.java.net/~rkennke/load-ref-barriers/webrev.02.diff/
Full:
http://cr.openjdk.java.net/~rkennke/load-ref-barriers/webrev.02/

(give it some mins to upload)

Ok to push?

Roman