RFR: Concurrent uncommit should be recorded as GC event + perf

[Zhengyu Gu]

Is this right?

      ShenandoahHeapRegion* r = get_region(i);
      if (r->is_empty_committed() && (r->empty_time() < shrink_before)) {
+      ShenandoahHeapLocker locker(lock());
+      if (r->is_empty_uncommitted()) {
        r->make_uncommitted();
        count++;

should it be?

+      if (r->is_empty_committed()) {
        r->make_uncommitted();
        count++;

Thanks,

-Zhengyu

On 07/06/2018 01:10 PM, Aleksey Shipilev wrote:
> http://cr.openjdk.java.net/~shade/shenandoah/uncommit-events/webrev.01/
> 
> This implements two easy improvements in heap uncommit: recording the currently hidden event
> properly in statistics to observe the real time concurrent thread spend dealing with uncommits, and
> simple lock interleaving with SpinPause, like we do in FreeSet::recycle_trash.
> 
> Testing: tier3_gc_shenandoah, ad-hoc benchmarks
> 
> Thanks,
> -Aleksey
>