RFR: fix for alias analysis with ShenandoahBarriersForConst

[Aleksey Shipilev]

On 03/14/2018 05:16 PM, Roland Westrelin wrote:
> 
> http://cr.openjdk.java.net/~roland/shenandoah/ShenandoahBarriersForConst-alias-analysis/webrev.00/

*) I would prefer us to have large blocks of Shenandoah code in shenandoahSupport.?pp. For example,
can we move AddPNode::shenandoah_fix_type to ShenandoahBarrierNode::fix_type? Then we can make the
calls to it explicitly guarded with UseShenandoahGC:

  Type* r = tp->add_offset(txoffset);
  if (UseShenandoahGC) {
    r = ShenandoahBarrierNode::fix_type(r);
  }
  return r;

*) PhaseIdealLoop::place_near_use again: you mentioned this is a bugfix here
  http://mail.openjdk.java.net/pipermail/shenandoah-dev/2018-February/004982.html

 Do we want to split it out into a separate changeset? So we can e.g. backport it separately for
sh/jdk10? Otherwise it keeps getting reverted along with other bulk changes, like it did in sh/jdk10
when we reverted improved profile predicates...

Thanks,
-Aleksey