RFR (11): [backport] 8221766: Load-reference barriers for Shenandoah

[Roman Kennke]

>> It reverts the part back to original jdk11 upstream because we no longer
>> need to insert into the control flow there. Is that not ok?
> 
> Ah ok. That change is good.
> 
>>> cfgnode.hpp
>>>
>>> That change doesn't seem needed
>>
>> Are you sure? We do have a corresponding call to step_over_gc_barrier()
>> in jdk/jdk too.
>>
>> http://hg.openjdk.java.net/jdk/jdk/file/ce8bab2c4185/src/hotspot/share/opto/cfgnode.cpp#l1451
> 
> cfgnode.*h*pp
> 
> The following change which removes:
> 
> 306   Node* dominated_by(Node* prev_dom, PhaseIterGVN* igvn);
> 
> and adds it back:
> 
> 393   Node* dominated_by(Node* prev_dom, PhaseIterGVN* igvn);

Right. This was there to make dominated_by() public, because we're using
it in ShBSC2::ideal_node(). However, we don't actually use that code
path and can remove it and revert the cfgnode.hpp change back:

Incremental:
http://cr.openjdk.java.net/~rkennke/backport-jdk11-JDK-8221766/webrev.02.diff/
Full:
http://cr.openjdk.java.net/~rkennke/backport-jdk11-JDK-8221766/webrev.02/

(Give it a minute or so)

Better?

Roman