RFR: Use BSC2::enqueue_useful_gc_barrier() to enqueue Shenandoah barriers

[Roman Kennke]

The webrev contained another change. Please look at this one:

http://cr.openjdk.java.net/~rkennke/enqueue-barriers/webrev.01/

Thanks,
Roman


> This moves the enqueuing of Shenandoah barrier (users) to go via the
> BarrierSetC2::enqueue_useful_gc_barrier() interface.
> 
> One little gotcha was that the interface takes the worklist, and we want
> to all all *users* of the node, via
> PhaseIterGVN::add_users_to_worklist(). I changed the interface (to be
> upstreamed) to take the PhaseIterGVN* instead of the actual worklist.
> This seems more flexible.
> 
> Close to the code that I touched I see:
> 
>             } else if (dead->Opcode() == Op_ShenandoahWBMemProj) {
>               assert(i == 0 && in->Opcode() ==
> Op_ShenandoahWriteBarrier, "broken graph");
>               _worklist.push(in);
>             }
> 
> I wonder if this can be handled in SBSC2::enqueue_useful_gc_barrier() as
> well. Problem is we don't get to see 'dead' but only the actual WB. Can
> we enqueue this wholesale, without checking for 'dead ==
> ShenandoahWBMemProj' ? Or can we use it somehow else?
> 
> I left it alone for now.
> 
> http://cr.openjdk.java.net/~rkennke/enqueue-barriers/webrev.00/
> 
> Testing: tier3_gc_shenandoah
> 
> Ok?
>