# HG changeset patch
# User dcubed
# Date 1370523393 25200
# Node ID f8c8cace25ad72da7b2f0748270c6a394b15b437
# Parent  04551f4dbdb9da1ecb6b9bb41c831ede07e2e18c# Parent  6bf8b8bb7c19316ce03748d10f40f254def85769
Merge

diff -r 6bf8b8bb7c19 -r f8c8cace25ad make/excludeSrc.make
--- a/make/excludeSrc.make	Wed Jun 05 14:12:49 2013 -0400
+++ b/make/excludeSrc.make	Thu Jun 06 05:56:33 2013 -0700
@@ -87,7 +87,7 @@
 	g1BlockOffsetTable.cpp g1CardCounts.cpp g1CollectedHeap.cpp g1CollectorPolicy.cpp \
 	g1ErgoVerbose.cpp g1GCPhaseTimes.cpp g1HRPrinter.cpp g1HotCardCache.cpp g1Log.cpp \
 	g1MMUTracker.cpp g1MarkSweep.cpp g1MemoryPool.cpp g1MonitoringSupport.cpp \
-	g1RemSet.cpp g1SATBCardTableModRefBS.cpp g1_globals.cpp heapRegion.cpp \
+	g1RemSet.cpp g1RemSetSummary.cpp g1SATBCardTableModRefBS.cpp g1_globals.cpp heapRegion.cpp \
 	heapRegionRemSet.cpp heapRegionSeq.cpp heapRegionSet.cpp heapRegionSets.cpp \
 	ptrQueue.cpp satbQueue.cpp sparsePRT.cpp survRateGroup.cpp vm_operations_g1.cpp \
 	adjoiningGenerations.cpp adjoiningVirtualSpaces.cpp asPSOldGen.cpp asPSYoungGen.cpp \
diff -r 6bf8b8bb7c19 -r f8c8cace25ad src/os/bsd/vm/os_bsd.cpp
--- a/src/os/bsd/vm/os_bsd.cpp	Wed Jun 05 14:12:49 2013 -0400
+++ b/src/os/bsd/vm/os_bsd.cpp	Thu Jun 06 05:56:33 2013 -0700
@@ -935,10 +935,10 @@
   return (1000 * 1000);
 }
 
-// XXX: For now, code this as if BSD does not support vtime.
-bool os::supports_vtime() { return false; }
+bool os::supports_vtime() { return true; }
 bool os::enable_vtime()   { return false; }
 bool os::vtime_enabled()  { return false; }
+
 double os::elapsedVTime() {
   // better than nothing, but not much
   return elapsedTime();
diff -r 6bf8b8bb7c19 -r f8c8cace25ad src/os/linux/vm/os_linux.cpp
--- a/src/os/linux/vm/os_linux.cpp	Wed Jun 05 14:12:49 2013 -0400
+++ b/src/os/linux/vm/os_linux.cpp	Thu Jun 06 05:56:33 2013 -0700
@@ -101,6 +101,12 @@
 # include <inttypes.h>
 # include <sys/ioctl.h>
 
+// if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
+// getrusage() is prepared to handle the associated failure.
+#ifndef RUSAGE_THREAD
+#define RUSAGE_THREAD   (1)               /* only the calling thread */
+#endif
+
 #define MAX_PATH    (2 * K)
 
 // for timer info max values which include all bits
@@ -1336,15 +1342,19 @@
   return (1000 * 1000);
 }
 
-// For now, we say that linux does not support vtime.  I have no idea
-// whether it can actually be made to (DLD, 9/13/05).
-
-bool os::supports_vtime() { return false; }
+bool os::supports_vtime() { return true; }
 bool os::enable_vtime()   { return false; }
 bool os::vtime_enabled()  { return false; }
+
 double os::elapsedVTime() {
-  // better than nothing, but not much
-  return elapsedTime();
+  struct rusage usage;
+  int retval = getrusage(RUSAGE_THREAD, &usage);
+  if (retval == 0) {
+    return (double) (usage.ru_utime.tv_sec + usage.ru_stime.tv_sec) + (double) (usage.ru_utime.tv_usec + usage.ru_stime.tv_usec) / (1000 * 1000);
+  } else {
+    // better than nothing, but not much
+    return elapsedTime();
+  }
 }
 
 jlong os::javaTimeMillis() {
diff -r 6bf8b8bb7c19 -r f8c8cace25ad src/os/windows/vm/os_windows.cpp
--- a/src/os/windows/vm/os_windows.cpp	Wed Jun 05 14:12:49 2013 -0400
+++ b/src/os/windows/vm/os_windows.cpp	Thu Jun 06 05:56:33 2013 -0700
@@ -813,15 +813,21 @@
   return result;
 }
 
-// For now, we say that Windows does not support vtime.  I have no idea
-// whether it can actually be made to (DLD, 9/13/05).
-
-bool os::supports_vtime() { return false; }
+bool os::supports_vtime() { return true; }
 bool os::enable_vtime() { return false; }
 bool os::vtime_enabled() { return false; }
+
 double os::elapsedVTime() {
-  // better than nothing, but not much
-  return elapsedTime();
+  FILETIME created;
+  FILETIME exited;
+  FILETIME kernel;
+  FILETIME user;
+  if (GetThreadTimes(GetCurrentThread(), &created, &exited, &kernel, &user) != 0) {
+    // the resolution of windows_to_java_time() should be sufficient (ms)
+    return (double) (windows_to_java_time(kernel) + windows_to_java_time(user)) / MILLIUNITS;
+  } else {
+    return elapsedTime();
+  }
 }
 
 jlong os::javaTimeMillis() {
diff -r 6bf8b8bb7c19 -r f8c8cace25ad src/share/vm/gc_implementation/g1/concurrentG1Refine.cpp
--- a/src/share/vm/gc_implementation/g1/concurrentG1Refine.cpp	Wed Jun 05 14:12:49 2013 -0400
+++ b/src/share/vm/gc_implementation/g1/concurrentG1Refine.cpp	Thu Jun 06 05:56:33 2013 -0700
@@ -114,6 +114,14 @@
   }
 }
 
+void ConcurrentG1Refine::worker_threads_do(ThreadClosure * tc) {
+  if (_threads != NULL) {
+    for (int i = 0; i < worker_thread_num(); i++) {
+      tc->do_thread(_threads[i]);
+    }
+  }
+}
+
 int ConcurrentG1Refine::thread_num() {
   int n_threads = (G1ConcRefinementThreads > 0) ? G1ConcRefinementThreads
                                                 : ParallelGCThreads;
@@ -126,3 +134,7 @@
     st->cr();
   }
 }
+
+ConcurrentG1RefineThread * ConcurrentG1Refine::sampling_thread() const {
+  return _threads[worker_thread_num()];
+}
diff -r 6bf8b8bb7c19 -r f8c8cace25ad src/share/vm/gc_implementation/g1/concurrentG1Refine.hpp
--- a/src/share/vm/gc_implementation/g1/concurrentG1Refine.hpp	Wed Jun 05 14:12:49 2013 -0400
+++ b/src/share/vm/gc_implementation/g1/concurrentG1Refine.hpp	Thu Jun 06 05:56:33 2013 -0700
@@ -35,6 +35,7 @@
 class G1CollectedHeap;
 class G1HotCardCache;
 class G1RemSet;
+class DirtyCardQueue;
 
 class ConcurrentG1Refine: public CHeapObj<mtGC> {
   ConcurrentG1RefineThread** _threads;
@@ -78,9 +79,15 @@
 
   void reinitialize_threads();
 
-  // Iterate over the conc refine threads
+  // Iterate over all concurrent refinement threads
   void threads_do(ThreadClosure *tc);
 
+  // Iterate over all worker refinement threads
+  void worker_threads_do(ThreadClosure * tc);
+
+  // The RS sampling thread
+  ConcurrentG1RefineThread * sampling_thread() const;
+
   static int thread_num();
 
   void print_worker_threads_on(outputStream* st) const;
diff -r 6bf8b8bb7c19 -r f8c8cace25ad src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp
--- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp	Wed Jun 05 14:12:49 2013 -0400
+++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp	Thu Jun 06 05:56:33 2013 -0700
@@ -1417,8 +1417,6 @@
 
       MemoryService::track_memory_usage();
 
-      verify_after_gc();
-
       assert(!ref_processor_stw()->discovery_enabled(), "Postcondition");
       ref_processor_stw()->verify_no_references_recorded();
 
@@ -1521,6 +1519,8 @@
       _hrs.verify_optional();
       verify_region_sets_optional();
 
+      verify_after_gc();
+
       // Start a new incremental collection set for the next pause
       assert(g1_policy()->collection_set() == NULL, "must be");
       g1_policy()->start_incremental_cset_building();
@@ -3539,6 +3539,14 @@
 }
 
 void G1CollectedHeap::gc_epilogue(bool full /* Ignored */) {
+
+  if (G1SummarizeRSetStats &&
+      (G1SummarizeRSetStatsPeriod > 0) &&
+      // we are at the end of the GC. Total collections has already been increased.
+      ((total_collections() - 1) % G1SummarizeRSetStatsPeriod == 0)) {
+    g1_rem_set()->print_periodic_summary_info();
+  }
+
   // FIXME: what is this about?
   // I'm ignoring the "fill_newgen()" call if "alloc_event_enabled"
   // is set.
@@ -4093,12 +4101,6 @@
     g1mm()->update_sizes();
   }
 
-  if (G1SummarizeRSetStats &&
-      (G1SummarizeRSetStatsPeriod > 0) &&
-      (total_collections() % G1SummarizeRSetStatsPeriod == 0)) {
-    g1_rem_set()->print_summary_info();
-  }
-
   // It should now be safe to tell the concurrent mark thread to start
   // without its logging output interfering with the logging output
   // that came from the pause.
diff -r 6bf8b8bb7c19 -r f8c8cace25ad src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp
--- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp	Wed Jun 05 14:12:49 2013 -0400
+++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp	Thu Jun 06 05:56:33 2013 -0700
@@ -593,11 +593,6 @@
   // may not be a humongous - it must fit into a single heap region.
   HeapWord* par_allocate_during_gc(GCAllocPurpose purpose, size_t word_size);
 
-  HeapWord* allocate_during_gc_slow(GCAllocPurpose purpose,
-                                    HeapRegion*    alloc_region,
-                                    bool           par,
-                                    size_t         word_size);
-
   // Ensure that no further allocations can happen in "r", bearing in mind
   // that parallel threads might be attempting allocations.
   void par_allocate_remaining_space(HeapRegion* r);
@@ -1733,6 +1728,95 @@
     ParGCAllocBuffer::retire(end_of_gc, retain);
     _retired = true;
   }
+
+  bool is_retired() {
+    return _retired;
+  }
+};
+
+class G1ParGCAllocBufferContainer {
+protected:
+  static int const _priority_max = 2;
+  G1ParGCAllocBuffer* _priority_buffer[_priority_max];
+
+public:
+  G1ParGCAllocBufferContainer(size_t gclab_word_size) {
+    for (int pr = 0; pr < _priority_max; ++pr) {
+      _priority_buffer[pr] = new G1ParGCAllocBuffer(gclab_word_size);
+    }
+  }
+
+  ~G1ParGCAllocBufferContainer() {
+    for (int pr = 0; pr < _priority_max; ++pr) {
+      assert(_priority_buffer[pr]->is_retired(), "alloc buffers should all retire at this point.");
+      delete _priority_buffer[pr];
+    }
+  }
+
+  HeapWord* allocate(size_t word_sz) {
+    HeapWord* obj;
+    for (int pr = 0; pr < _priority_max; ++pr) {
+      obj = _priority_buffer[pr]->allocate(word_sz);
+      if (obj != NULL) return obj;
+    }
+    return obj;
+  }
+
+  bool contains(void* addr) {
+    for (int pr = 0; pr < _priority_max; ++pr) {
+      if (_priority_buffer[pr]->contains(addr)) return true;
+    }
+    return false;
+  }
+
+  void undo_allocation(HeapWord* obj, size_t word_sz) {
+    bool finish_undo;
+    for (int pr = 0; pr < _priority_max; ++pr) {
+      if (_priority_buffer[pr]->contains(obj)) {
+        _priority_buffer[pr]->undo_allocation(obj, word_sz);
+        finish_undo = true;
+      }
+    }
+    if (!finish_undo) ShouldNotReachHere();
+  }
+
+  size_t words_remaining() {
+    size_t result = 0;
+    for (int pr = 0; pr < _priority_max; ++pr) {
+      result += _priority_buffer[pr]->words_remaining();
+    }
+    return result;
+  }
+
+  size_t words_remaining_in_retired_buffer() {
+    G1ParGCAllocBuffer* retired = _priority_buffer[0];
+    return retired->words_remaining();
+  }
+
+  void flush_stats_and_retire(PLABStats* stats, bool end_of_gc, bool retain) {
+    for (int pr = 0; pr < _priority_max; ++pr) {
+      _priority_buffer[pr]->flush_stats_and_retire(stats, end_of_gc, retain);
+    }
+  }
+
+  void update(bool end_of_gc, bool retain, HeapWord* buf, size_t word_sz) {
+    G1ParGCAllocBuffer* retired_and_set = _priority_buffer[0];
+    retired_and_set->retire(end_of_gc, retain);
+    retired_and_set->set_buf(buf);
+    retired_and_set->set_word_size(word_sz);
+    adjust_priority_order();
+  }
+
+private:
+  void adjust_priority_order() {
+    G1ParGCAllocBuffer* retired_and_set = _priority_buffer[0];
+
+    int last = _priority_max - 1;
+    for (int pr = 0; pr < last; ++pr) {
+      _priority_buffer[pr] = _priority_buffer[pr + 1];
+    }
+    _priority_buffer[last] = retired_and_set;
+  }
 };
 
 class G1ParScanThreadState : public StackObj {
@@ -1743,9 +1827,9 @@
   CardTableModRefBS* _ct_bs;
   G1RemSet* _g1_rem;
 
-  G1ParGCAllocBuffer  _surviving_alloc_buffer;
-  G1ParGCAllocBuffer  _tenured_alloc_buffer;
-  G1ParGCAllocBuffer* _alloc_buffers[GCAllocPurposeCount];
+  G1ParGCAllocBufferContainer  _surviving_alloc_buffer;
+  G1ParGCAllocBufferContainer  _tenured_alloc_buffer;
+  G1ParGCAllocBufferContainer* _alloc_buffers[GCAllocPurposeCount];
   ageTable            _age_table;
 
   size_t           _alloc_buffer_waste;
@@ -1809,7 +1893,7 @@
   RefToScanQueue*   refs()            { return _refs;             }
   ageTable*         age_table()       { return &_age_table;       }
 
-  G1ParGCAllocBuffer* alloc_buffer(GCAllocPurpose purpose) {
+  G1ParGCAllocBufferContainer* alloc_buffer(GCAllocPurpose purpose) {
     return _alloc_buffers[purpose];
   }
 
@@ -1839,15 +1923,13 @@
     HeapWord* obj = NULL;
     size_t gclab_word_size = _g1h->desired_plab_sz(purpose);
     if (word_sz * 100 < gclab_word_size * ParallelGCBufferWastePct) {
-      G1ParGCAllocBuffer* alloc_buf = alloc_buffer(purpose);
-      add_to_alloc_buffer_waste(alloc_buf->words_remaining());
-      alloc_buf->retire(false /* end_of_gc */, false /* retain */);
+      G1ParGCAllocBufferContainer* alloc_buf = alloc_buffer(purpose);
 
       HeapWord* buf = _g1h->par_allocate_during_gc(purpose, gclab_word_size);
       if (buf == NULL) return NULL; // Let caller handle allocation failure.
-      // Otherwise.
-      alloc_buf->set_word_size(gclab_word_size);
-      alloc_buf->set_buf(buf);
+
+      add_to_alloc_buffer_waste(alloc_buf->words_remaining_in_retired_buffer());
+      alloc_buf->update(false /* end_of_gc */, false /* retain */, buf, gclab_word_size);
 
       obj = alloc_buf->allocate(word_sz);
       assert(obj != NULL, "buffer was definitely big enough...");
@@ -1959,7 +2041,6 @@
     }
   }
 
-public:
   void trim_queue();
 };
 
diff -r 6bf8b8bb7c19 -r f8c8cace25ad src/share/vm/gc_implementation/g1/g1RemSet.cpp
--- a/src/share/vm/gc_implementation/g1/g1RemSet.cpp	Wed Jun 05 14:12:49 2013 -0400
+++ b/src/share/vm/gc_implementation/g1/g1RemSet.cpp	Thu Jun 06 05:56:33 2013 -0700
@@ -34,6 +34,7 @@
 #include "gc_implementation/g1/g1OopClosures.inline.hpp"
 #include "gc_implementation/g1/g1RemSet.inline.hpp"
 #include "gc_implementation/g1/heapRegionSeq.inline.hpp"
+#include "gc_implementation/g1/heapRegionRemSet.hpp"
 #include "memory/iterator.hpp"
 #include "oops/oop.inline.hpp"
 #include "utilities/intHisto.hpp"
@@ -73,7 +74,8 @@
     _ct_bs(ct_bs), _g1p(_g1->g1_policy()),
     _cg1r(g1->concurrent_g1_refine()),
     _cset_rs_update_cl(NULL),
-    _cards_scanned(NULL), _total_cards_scanned(0)
+    _cards_scanned(NULL), _total_cards_scanned(0),
+    _prev_period_summary()
 {
   _seq_task = new SubTasksDone(NumSeqTasks);
   guarantee(n_workers() > 0, "There should be some workers");
@@ -81,6 +83,7 @@
   for (uint i = 0; i < n_workers(); i++) {
     _cset_rs_update_cl[i] = NULL;
   }
+  _prev_period_summary.initialize(this, n_workers());
 }
 
 G1RemSet::~G1RemSet() {
@@ -697,47 +700,29 @@
   return has_refs_into_cset;
 }
 
-class HRRSStatsIter: public HeapRegionClosure {
-  size_t _occupied;
-  size_t _total_mem_sz;
-  size_t _max_mem_sz;
-  HeapRegion* _max_mem_sz_region;
-public:
-  HRRSStatsIter() :
-    _occupied(0),
-    _total_mem_sz(0),
-    _max_mem_sz(0),
-    _max_mem_sz_region(NULL)
-  {}
+void G1RemSet::print_periodic_summary_info() {
+  G1RemSetSummary current;
+  current.initialize(this, n_workers());
 
-  bool doHeapRegion(HeapRegion* r) {
-    if (r->continuesHumongous()) return false;
-    size_t mem_sz = r->rem_set()->mem_size();
-    if (mem_sz > _max_mem_sz) {
-      _max_mem_sz = mem_sz;
-      _max_mem_sz_region = r;
-    }
-    _total_mem_sz += mem_sz;
-    size_t occ = r->rem_set()->occupied();
-    _occupied += occ;
-    return false;
-  }
-  size_t total_mem_sz() { return _total_mem_sz; }
-  size_t max_mem_sz() { return _max_mem_sz; }
-  size_t occupied() { return _occupied; }
-  HeapRegion* max_mem_sz_region() { return _max_mem_sz_region; }
-};
+  _prev_period_summary.subtract_from(&current);
+  print_summary_info(&_prev_period_summary);
 
-class PrintRSThreadVTimeClosure : public ThreadClosure {
-public:
-  virtual void do_thread(Thread *t) {
-    ConcurrentG1RefineThread* crt = (ConcurrentG1RefineThread*) t;
-    gclog_or_tty->print("    %5.2f", crt->vtime_accum());
-  }
-};
+  _prev_period_summary.set(&current);
+}
 
 void G1RemSet::print_summary_info() {
-  G1CollectedHeap* g1 = G1CollectedHeap::heap();
+  G1RemSetSummary current;
+  current.initialize(this, n_workers());
+
+  print_summary_info(&current, " Cumulative RS summary");
+}
+
+void G1RemSet::print_summary_info(G1RemSetSummary * summary, const char * header) {
+  assert(summary != NULL, "just checking");
+
+  if (header != NULL) {
+    gclog_or_tty->print_cr("%s", header);
+  }
 
 #if CARD_REPEAT_HISTO
   gclog_or_tty->print_cr("\nG1 card_repeat count histogram: ");
@@ -745,52 +730,13 @@
   card_repeat_count.print_on(gclog_or_tty);
 #endif
 
-  gclog_or_tty->print_cr("\n Concurrent RS processed %d cards",
-                         _conc_refine_cards);
-  DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
-  jint tot_processed_buffers =
-    dcqs.processed_buffers_mut() + dcqs.processed_buffers_rs_thread();
-  gclog_or_tty->print_cr("  Of %d completed buffers:", tot_processed_buffers);
-  gclog_or_tty->print_cr("     %8d (%5.1f%%) by conc RS threads.",
-                dcqs.processed_buffers_rs_thread(),
-                100.0*(float)dcqs.processed_buffers_rs_thread()/
-                (float)tot_processed_buffers);
-  gclog_or_tty->print_cr("     %8d (%5.1f%%) by mutator threads.",
-                dcqs.processed_buffers_mut(),
-                100.0*(float)dcqs.processed_buffers_mut()/
-                (float)tot_processed_buffers);
-  gclog_or_tty->print_cr("  Conc RS threads times(s)");
-  PrintRSThreadVTimeClosure p;
-  gclog_or_tty->print("     ");
-  g1->concurrent_g1_refine()->threads_do(&p);
-  gclog_or_tty->print_cr("");
-
-  HRRSStatsIter blk;
-  g1->heap_region_iterate(&blk);
-  gclog_or_tty->print_cr("  Total heap region rem set sizes = "SIZE_FORMAT"K."
-                         "  Max = "SIZE_FORMAT"K.",
-                         blk.total_mem_sz()/K, blk.max_mem_sz()/K);
-  gclog_or_tty->print_cr("  Static structures = "SIZE_FORMAT"K,"
-                         " free_lists = "SIZE_FORMAT"K.",
-                         HeapRegionRemSet::static_mem_size() / K,
-                         HeapRegionRemSet::fl_mem_size() / K);
-  gclog_or_tty->print_cr("    "SIZE_FORMAT" occupied cards represented.",
-                         blk.occupied());
-  HeapRegion* max_mem_sz_region = blk.max_mem_sz_region();
-  HeapRegionRemSet* rem_set = max_mem_sz_region->rem_set();
-  gclog_or_tty->print_cr("    Max size region = "HR_FORMAT", "
-                         "size = "SIZE_FORMAT "K, occupied = "SIZE_FORMAT"K.",
-                         HR_FORMAT_PARAMS(max_mem_sz_region),
-                         (rem_set->mem_size() + K - 1)/K,
-                         (rem_set->occupied() + K - 1)/K);
-  gclog_or_tty->print_cr("    Did %d coarsenings.",
-                         HeapRegionRemSet::n_coarsenings());
+  summary->print_on(gclog_or_tty);
 }
 
 void G1RemSet::prepare_for_verify() {
   if (G1HRRSFlushLogBuffersOnVerify &&
       (VerifyBeforeGC || VerifyAfterGC)
-      &&  !_g1->full_collection()) {
+      &&  (!_g1->full_collection() || G1VerifyRSetsDuringFullGC)) {
     cleanupHRRS();
     _g1->set_refine_cte_cl_concurrency(false);
     if (SafepointSynchronize::is_at_safepoint()) {
diff -r 6bf8b8bb7c19 -r f8c8cace25ad src/share/vm/gc_implementation/g1/g1RemSet.hpp
--- a/src/share/vm/gc_implementation/g1/g1RemSet.hpp	Wed Jun 05 14:12:49 2013 -0400
+++ b/src/share/vm/gc_implementation/g1/g1RemSet.hpp	Thu Jun 06 05:56:33 2013 -0700
@@ -25,6 +25,8 @@
 #ifndef SHARE_VM_GC_IMPLEMENTATION_G1_G1REMSET_HPP
 #define SHARE_VM_GC_IMPLEMENTATION_G1_G1REMSET_HPP
 
+#include "gc_implementation/g1/g1RemSetSummary.hpp"
+
 // A G1RemSet provides ways of iterating over pointers into a selected
 // collection set.
 
@@ -37,9 +39,11 @@
 // so that they can be used to update the individual region remsets.
 
 class G1RemSet: public CHeapObj<mtGC> {
+private:
+  G1RemSetSummary _prev_period_summary;
 protected:
   G1CollectedHeap* _g1;
-  unsigned _conc_refine_cards;
+  size_t _conc_refine_cards;
   uint n_workers();
 
 protected:
@@ -66,6 +70,8 @@
   // references into the collection set.
   OopsInHeapRegionClosure** _cset_rs_update_cl;
 
+  // Print the given summary info
+  virtual void print_summary_info(G1RemSetSummary * summary, const char * header = NULL);
 public:
   // This is called to reset dual hash tables after the gc pause
   // is finished and the initial hash table is no longer being
@@ -123,11 +129,18 @@
                            int worker_i,
                            bool check_for_refs_into_cset);
 
-  // Print any relevant summary info.
+  // Print accumulated summary info from the start of the VM.
   virtual void print_summary_info();
 
+  // Print accumulated summary info from the last time called.
+  virtual void print_periodic_summary_info();
+
   // Prepare remembered set for verification.
   virtual void prepare_for_verify();
+
+  size_t conc_refine_cards() const {
+    return _conc_refine_cards;
+  }
 };
 
 class CountNonCleanMemRegionClosure: public MemRegionClosure {
diff -r 6bf8b8bb7c19 -r f8c8cace25ad src/share/vm/gc_implementation/g1/g1RemSetSummary.cpp
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/share/vm/gc_implementation/g1/g1RemSetSummary.cpp	Thu Jun 06 05:56:33 2013 -0700
@@ -0,0 +1,205 @@
+/*
+ * Copyright (c) 2013, Oracle and/or its affiliates. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ *
+ */
+
+#include "precompiled.hpp"
+#include "gc_implementation/g1/concurrentG1Refine.hpp"
+#include "gc_implementation/g1/concurrentG1RefineThread.hpp"
+#include "gc_implementation/g1/heapRegion.hpp"
+#include "gc_implementation/g1/g1CollectedHeap.inline.hpp"
+#include "gc_implementation/g1/g1RemSet.inline.hpp"
+#include "gc_implementation/g1/g1RemSetSummary.hpp"
+#include "gc_implementation/g1/heapRegionRemSet.hpp"
+#include "runtime/thread.inline.hpp"
+
+class GetRSThreadVTimeClosure : public ThreadClosure {
+private:
+  G1RemSetSummary* _summary;
+  uint _counter;
+
+public:
+  GetRSThreadVTimeClosure(G1RemSetSummary * summary) : ThreadClosure(), _summary(summary), _counter(0) {
+    assert(_summary != NULL, "just checking");
+  }
+
+  virtual void do_thread(Thread* t) {
+    ConcurrentG1RefineThread* crt = (ConcurrentG1RefineThread*) t;
+    _summary->set_rs_thread_vtime(_counter, crt->vtime_accum());
+    _counter++;
+  }
+};
+
+void G1RemSetSummary::update() {
+  _num_refined_cards = remset()->conc_refine_cards();
+  DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
+  _num_processed_buf_mutator = dcqs.processed_buffers_mut();
+  _num_processed_buf_rs_threads = dcqs.processed_buffers_rs_thread();
+
+  _num_coarsenings = HeapRegionRemSet::n_coarsenings();
+
+  ConcurrentG1Refine * cg1r = G1CollectedHeap::heap()->concurrent_g1_refine();
+  if (_rs_threads_vtimes != NULL) {
+    GetRSThreadVTimeClosure p(this);
+    cg1r->worker_threads_do(&p);
+  }
+  set_sampling_thread_vtime(cg1r->sampling_thread()->vtime_accum());
+}
+
+void G1RemSetSummary::set_rs_thread_vtime(uint thread, double value) {
+  assert(_rs_threads_vtimes != NULL, "just checking");
+  assert(thread < _num_vtimes, "just checking");
+  _rs_threads_vtimes[thread] = value;
+}
+
+double G1RemSetSummary::rs_thread_vtime(uint thread) const {
+  assert(_rs_threads_vtimes != NULL, "just checking");
+  assert(thread < _num_vtimes, "just checking");
+  return _rs_threads_vtimes[thread];
+}
+
+void G1RemSetSummary::initialize(G1RemSet* remset, uint num_workers) {
+  assert(_rs_threads_vtimes == NULL, "just checking");
+  assert(remset != NULL, "just checking");
+
+  _remset = remset;
+  _num_vtimes = num_workers;
+  _rs_threads_vtimes = NEW_C_HEAP_ARRAY(double, _num_vtimes, mtGC);
+  memset(_rs_threads_vtimes, 0, sizeof(double) * _num_vtimes);
+
+  update();
+}
+
+void G1RemSetSummary::set(G1RemSetSummary* other) {
+  assert(other != NULL, "just checking");
+  assert(remset() == other->remset(), "just checking");
+  assert(_num_vtimes == other->_num_vtimes, "just checking");
+
+  _num_refined_cards = other->num_concurrent_refined_cards();
+
+  _num_processed_buf_mutator = other->num_processed_buf_mutator();
+  _num_processed_buf_rs_threads = other->num_processed_buf_rs_threads();
+
+  _num_coarsenings = other->_num_coarsenings;
+
+  memcpy(_rs_threads_vtimes, other->_rs_threads_vtimes, sizeof(double) * _num_vtimes);
+
+  set_sampling_thread_vtime(other->sampling_thread_vtime());
+}
+
+void G1RemSetSummary::subtract_from(G1RemSetSummary* other) {
+  assert(other != NULL, "just checking");
+  assert(remset() == other->remset(), "just checking");
+  assert(_num_vtimes == other->_num_vtimes, "just checking");
+
+  _num_refined_cards = other->num_concurrent_refined_cards() - _num_refined_cards;
+
+  _num_processed_buf_mutator = other->num_processed_buf_mutator() - _num_processed_buf_mutator;
+  _num_processed_buf_rs_threads = other->num_processed_buf_rs_threads() - _num_processed_buf_rs_threads;
+
+  _num_coarsenings = other->num_coarsenings() - _num_coarsenings;
+
+  for (uint i = 0; i < _num_vtimes; i++) {
+    set_rs_thread_vtime(i, other->rs_thread_vtime(i) - rs_thread_vtime(i));
+  }
+
+  _sampling_thread_vtime = other->sampling_thread_vtime() - _sampling_thread_vtime;
+}
+
+class HRRSStatsIter: public HeapRegionClosure {
+  size_t _occupied;
+  size_t _total_mem_sz;
+  size_t _max_mem_sz;
+  HeapRegion* _max_mem_sz_region;
+public:
+  HRRSStatsIter() :
+    _occupied(0),
+    _total_mem_sz(0),
+    _max_mem_sz(0),
+    _max_mem_sz_region(NULL)
+  {}
+
+  bool doHeapRegion(HeapRegion* r) {
+    size_t mem_sz = r->rem_set()->mem_size();
+    if (mem_sz > _max_mem_sz) {
+      _max_mem_sz = mem_sz;
+      _max_mem_sz_region = r;
+    }
+    _total_mem_sz += mem_sz;
+    size_t occ = r->rem_set()->occupied();
+    _occupied += occ;
+    return false;
+  }
+  size_t total_mem_sz() { return _total_mem_sz; }
+  size_t max_mem_sz() { return _max_mem_sz; }
+  size_t occupied() { return _occupied; }
+  HeapRegion* max_mem_sz_region() { return _max_mem_sz_region; }
+};
+
+double calc_percentage(size_t numerator, size_t denominator) {
+  if (denominator != 0) {
+    return (double)numerator / denominator * 100.0;
+  } else {
+    return 0.0f;
+  }
+}
+
+void G1RemSetSummary::print_on(outputStream* out) {
+  out->print_cr("\n Concurrent RS processed "SIZE_FORMAT" cards",
+                num_concurrent_refined_cards());
+  out->print_cr("  Of %d completed buffers:", num_processed_buf_total());
+  out->print_cr("     %8d (%5.1f%%) by concurrent RS threads.",
+                num_processed_buf_total(),
+                calc_percentage(num_processed_buf_rs_threads(), num_processed_buf_total()));
+  out->print_cr("     %8d (%5.1f%%) by mutator threads.",
+                num_processed_buf_mutator(),
+                calc_percentage(num_processed_buf_mutator(), num_processed_buf_total()));
+  out->print_cr("  Concurrent RS threads times (s)");
+  out->print("     ");
+  for (uint i = 0; i < _num_vtimes; i++) {
+    out->print("    %5.2f", rs_thread_vtime(i));
+  }
+  out->cr();
+  out->print_cr("  Concurrent sampling threads times (s)");
+  out->print_cr("         %5.2f", sampling_thread_vtime());
+
+  HRRSStatsIter blk;
+  G1CollectedHeap::heap()->heap_region_iterate(&blk);
+  out->print_cr("  Total heap region rem set sizes = "SIZE_FORMAT"K."
+                "  Max = "SIZE_FORMAT"K.",
+                blk.total_mem_sz()/K, blk.max_mem_sz()/K);
+  out->print_cr("  Static structures = "SIZE_FORMAT"K,"
+                " free_lists = "SIZE_FORMAT"K.",
+                HeapRegionRemSet::static_mem_size() / K,
+                HeapRegionRemSet::fl_mem_size() / K);
+  out->print_cr("    "SIZE_FORMAT" occupied cards represented.",
+                blk.occupied());
+  HeapRegion* max_mem_sz_region = blk.max_mem_sz_region();
+  HeapRegionRemSet* rem_set = max_mem_sz_region->rem_set();
+  out->print_cr("    Max size region = "HR_FORMAT", "
+                "size = "SIZE_FORMAT "K, occupied = "SIZE_FORMAT"K.",
+                HR_FORMAT_PARAMS(max_mem_sz_region),
+                (rem_set->mem_size() + K - 1)/K,
+                (rem_set->occupied() + K - 1)/K);
+
+  out->print_cr("    Did %d coarsenings.", num_coarsenings());
+}
diff -r 6bf8b8bb7c19 -r f8c8cace25ad src/share/vm/gc_implementation/g1/g1RemSetSummary.hpp
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/share/vm/gc_implementation/g1/g1RemSetSummary.hpp	Thu Jun 06 05:56:33 2013 -0700
@@ -0,0 +1,118 @@
+/*
+ * Copyright (c) 2013, Oracle and/or its affiliates. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ *
+ */
+
+#ifndef SHARE_VM_GC_IMPLEMENTATION_G1_G1REMSETSUMMARY_HPP
+#define SHARE_VM_GC_IMPLEMENTATION_G1_G1REMSETSUMMARY_HPP
+
+#include "utilities/ostream.hpp"
+
+class G1RemSet;
+
+// A G1RemSetSummary manages statistical information about the G1RemSet
+
+class G1RemSetSummary VALUE_OBJ_CLASS_SPEC {
+private:
+  friend class GetRSThreadVTimeClosure;
+
+  G1RemSet* _remset;
+
+  G1RemSet* remset() const {
+    return _remset;
+  }
+
+  size_t _num_refined_cards;
+  size_t _num_processed_buf_mutator;
+  size_t _num_processed_buf_rs_threads;
+
+  size_t _num_coarsenings;
+
+  double* _rs_threads_vtimes;
+  size_t _num_vtimes;
+
+  double _sampling_thread_vtime;
+
+  void set_rs_thread_vtime(uint thread, double value);
+  void set_sampling_thread_vtime(double value) {
+    _sampling_thread_vtime = value;
+  }
+
+  void free_and_null() {
+    if (_rs_threads_vtimes) {
+      FREE_C_HEAP_ARRAY(double, _rs_threads_vtimes, mtGC);
+      _rs_threads_vtimes = NULL;
+      _num_vtimes = 0;
+    }
+  }
+
+  // update this summary with current data from various places
+  void update();
+
+public:
+  G1RemSetSummary() : _remset(NULL), _num_refined_cards(0),
+    _num_processed_buf_mutator(0), _num_processed_buf_rs_threads(0), _num_coarsenings(0),
+    _rs_threads_vtimes(NULL), _num_vtimes(0), _sampling_thread_vtime(0.0f) {
+  }
+
+  ~G1RemSetSummary() {
+    free_and_null();
+  }
+
+  // set the counters in this summary to the values of the others
+  void set(G1RemSetSummary* other);
+  // subtract all counters from the other summary, and set them in the current
+  void subtract_from(G1RemSetSummary* other);
+
+  // initialize and get the first sampling
+  void initialize(G1RemSet* remset, uint num_workers);
+
+  void print_on(outputStream* out);
+
+  double rs_thread_vtime(uint thread) const;
+
+  double sampling_thread_vtime() const {
+    return _sampling_thread_vtime;
+  }
+
+  size_t num_concurrent_refined_cards() const {
+    return _num_refined_cards;
+  }
+
+  size_t num_processed_buf_mutator() const {
+    return _num_processed_buf_mutator;
+  }
+
+  size_t num_processed_buf_rs_threads() const {
+    return _num_processed_buf_rs_threads;
+  }
+
+  size_t num_processed_buf_total() const {
+    return num_processed_buf_mutator() + num_processed_buf_rs_threads();
+  }
+
+  size_t num_coarsenings() const {
+    return _num_coarsenings;
+  }
+};
+
+#endif // SHARE_VM_GC_IMPLEMENTATION_G1_G1REMSETSUMMARY_HPP
diff -r 6bf8b8bb7c19 -r f8c8cace25ad src/share/vm/gc_implementation/g1/g1_globals.hpp
--- a/src/share/vm/gc_implementation/g1/g1_globals.hpp	Wed Jun 05 14:12:49 2013 -0400
+++ b/src/share/vm/gc_implementation/g1/g1_globals.hpp	Thu Jun 06 05:56:33 2013 -0700
@@ -329,7 +329,11 @@
                                                                             \
   develop(bool, G1EvacuationFailureALotDuringMixedGC, true,                 \
           "Force use of evacuation failure handling during mixed "          \
-          "evacuation pauses")
+          "evacuation pauses")                                              \
+                                                                            \
+  diagnostic(bool, G1VerifyRSetsDuringFullGC, false,                        \
+             "If true, perform verification of each heap region's "         \
+             "remembered set when verifying the heap during a full GC.")
 
 G1_FLAGS(DECLARE_DEVELOPER_FLAG, DECLARE_PD_DEVELOPER_FLAG, DECLARE_PRODUCT_FLAG, DECLARE_PD_PRODUCT_FLAG, DECLARE_DIAGNOSTIC_FLAG, DECLARE_EXPERIMENTAL_FLAG, DECLARE_NOTPRODUCT_FLAG, DECLARE_MANAGEABLE_FLAG, DECLARE_PRODUCT_RW_FLAG)
 
diff -r 6bf8b8bb7c19 -r f8c8cace25ad src/share/vm/gc_implementation/g1/heapRegion.cpp
--- a/src/share/vm/gc_implementation/g1/heapRegion.cpp	Wed Jun 05 14:12:49 2013 -0400
+++ b/src/share/vm/gc_implementation/g1/heapRegion.cpp	Thu Jun 06 05:56:33 2013 -0700
@@ -139,7 +139,7 @@
         _n_failures++;
       }
 
-      if (!_g1h->full_collection()) {
+      if (!_g1h->full_collection() || G1VerifyRSetsDuringFullGC) {
         HeapRegion* from = _g1h->heap_region_containing((HeapWord*)p);
         HeapRegion* to   = _g1h->heap_region_containing(obj);
         if (from != NULL && to != NULL &&
diff -r 6bf8b8bb7c19 -r f8c8cace25ad src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.hpp
--- a/src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.hpp	Wed Jun 05 14:12:49 2013 -0400
+++ b/src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.hpp	Thu Jun 06 05:56:33 2013 -0700
@@ -116,7 +116,7 @@
 
   // The alignment used for eden and survivors within the young gen
   // and for boundary between young gen and old gen.
-  size_t intra_heap_alignment() const { return 64 * K; }
+  size_t intra_heap_alignment() const { return 64 * K * HeapWordSize; }
 
   size_t capacity() const;
   size_t used() const;
diff -r 6bf8b8bb7c19 -r f8c8cace25ad src/share/vm/gc_implementation/parallelScavenge/psParallelCompact.cpp
--- a/src/share/vm/gc_implementation/parallelScavenge/psParallelCompact.cpp	Wed Jun 05 14:12:49 2013 -0400
+++ b/src/share/vm/gc_implementation/parallelScavenge/psParallelCompact.cpp	Thu Jun 06 05:56:33 2013 -0700
@@ -59,13 +59,25 @@
 #include <math.h>
 
 // All sizes are in HeapWords.
-const size_t ParallelCompactData::Log2RegionSize  = 9; // 512 words
+const size_t ParallelCompactData::Log2RegionSize  = 16; // 64K words
 const size_t ParallelCompactData::RegionSize      = (size_t)1 << Log2RegionSize;
 const size_t ParallelCompactData::RegionSizeBytes =
   RegionSize << LogHeapWordSize;
 const size_t ParallelCompactData::RegionSizeOffsetMask = RegionSize - 1;
 const size_t ParallelCompactData::RegionAddrOffsetMask = RegionSizeBytes - 1;
-const size_t ParallelCompactData::RegionAddrMask  = ~RegionAddrOffsetMask;
+const size_t ParallelCompactData::RegionAddrMask       = ~RegionAddrOffsetMask;
+
+const size_t ParallelCompactData::Log2BlockSize   = 7; // 128 words
+const size_t ParallelCompactData::BlockSize       = (size_t)1 << Log2BlockSize;
+const size_t ParallelCompactData::BlockSizeBytes  =
+  BlockSize << LogHeapWordSize;
+const size_t ParallelCompactData::BlockSizeOffsetMask = BlockSize - 1;
+const size_t ParallelCompactData::BlockAddrOffsetMask = BlockSizeBytes - 1;
+const size_t ParallelCompactData::BlockAddrMask       = ~BlockAddrOffsetMask;
+
+const size_t ParallelCompactData::BlocksPerRegion = RegionSize / BlockSize;
+const size_t ParallelCompactData::Log2BlocksPerRegion =
+  Log2RegionSize - Log2BlockSize;
 
 const ParallelCompactData::RegionData::region_sz_t
 ParallelCompactData::RegionData::dc_shift = 27;
@@ -359,6 +371,10 @@
   _reserved_byte_size = 0;
   _region_data = 0;
   _region_count = 0;
+
+  _block_vspace = 0;
+  _block_data = 0;
+  _block_count = 0;
 }
 
 bool ParallelCompactData::initialize(MemRegion covered_region)
@@ -372,8 +388,7 @@
   assert((region_size & RegionSizeOffsetMask) == 0,
          "region size not a multiple of RegionSize");
 
-  bool result = initialize_region_data(region_size);
-
+  bool result = initialize_region_data(region_size) && initialize_block_data();
   return result;
 }
 
@@ -418,17 +433,36 @@
   return false;
 }
 
+bool ParallelCompactData::initialize_block_data()
+{
+  assert(_region_count != 0, "region data must be initialized first");
+  const size_t count = _region_count << Log2BlocksPerRegion;
+  _block_vspace = create_vspace(count, sizeof(BlockData));
+  if (_block_vspace != 0) {
+    _block_data = (BlockData*)_block_vspace->reserved_low_addr();
+    _block_count = count;
+    return true;
+  }
+  return false;
+}
+
 void ParallelCompactData::clear()
 {
   memset(_region_data, 0, _region_vspace->committed_size());
+  memset(_block_data, 0, _block_vspace->committed_size());
 }
 
 void ParallelCompactData::clear_range(size_t beg_region, size_t end_region) {
   assert(beg_region <= _region_count, "beg_region out of range");
   assert(end_region <= _region_count, "end_region out of range");
+  assert(RegionSize % BlockSize == 0, "RegionSize not a multiple of BlockSize");
 
   const size_t region_cnt = end_region - beg_region;
   memset(_region_data + beg_region, 0, region_cnt * sizeof(RegionData));
+
+  const size_t beg_block = beg_region * BlocksPerRegion;
+  const size_t block_cnt = region_cnt * BlocksPerRegion;
+  memset(_block_data + beg_block, 0, block_cnt * sizeof(BlockData));
 }
 
 HeapWord* ParallelCompactData::partial_obj_end(size_t region_idx) const
@@ -707,49 +741,48 @@
 
 HeapWord* ParallelCompactData::calc_new_pointer(HeapWord* addr) {
   assert(addr != NULL, "Should detect NULL oop earlier");
-  assert(PSParallelCompact::gc_heap()->is_in(addr), "addr not in heap");
-#ifdef ASSERT
-  if (PSParallelCompact::mark_bitmap()->is_unmarked(addr)) {
-    gclog_or_tty->print_cr("calc_new_pointer:: addr " PTR_FORMAT, addr);
-  }
-#endif
-  assert(PSParallelCompact::mark_bitmap()->is_marked(addr), "obj not marked");
+  assert(PSParallelCompact::gc_heap()->is_in(addr), "not in heap");
+  assert(PSParallelCompact::mark_bitmap()->is_marked(addr), "not marked");
 
   // Region covering the object.
-  size_t region_index = addr_to_region_idx(addr);
-  const RegionData* const region_ptr = region(region_index);
-  HeapWord* const region_addr = region_align_down(addr);
-
-  assert(addr < region_addr + RegionSize, "Region does not cover object");
-  assert(addr_to_region_ptr(region_addr) == region_ptr, "sanity check");
-
+  RegionData* const region_ptr = addr_to_region_ptr(addr);
   HeapWord* result = region_ptr->destination();
 
-  // If all the data in the region is live, then the new location of the object
-  // can be calculated from the destination of the region plus the offset of the
-  // object in the region.
+  // If the entire Region is live, the new location is region->destination + the
+  // offset of the object within in the Region.
+
+  // Run some performance tests to determine if this special case pays off.  It
+  // is worth it for pointers into the dense prefix.  If the optimization to
+  // avoid pointer updates in regions that only point to the dense prefix is
+  // ever implemented, this should be revisited.
   if (region_ptr->data_size() == RegionSize) {
-    result += pointer_delta(addr, region_addr);
-    DEBUG_ONLY(PSParallelCompact::check_new_location(addr, result);)
+    result += region_offset(addr);
     return result;
   }
 
-  // The new location of the object is
-  //    region destination +
-  //    size of the partial object extending onto the region +
-  //    sizes of the live objects in the Region that are to the left of addr
-  const size_t partial_obj_size = region_ptr->partial_obj_size();
-  HeapWord* const search_start = region_addr + partial_obj_size;
+  // Otherwise, the new location is region->destination + block offset + the
+  // number of live words in the Block that are (a) to the left of addr and (b)
+  // due to objects that start in the Block.
+
+  // Fill in the block table if necessary.  This is unsynchronized, so multiple
+  // threads may fill the block table for a region (harmless, since it is
+  // idempotent).
+  if (!region_ptr->blocks_filled()) {
+    PSParallelCompact::fill_blocks(addr_to_region_idx(addr));
+    region_ptr->set_blocks_filled();
+  }
+
+  HeapWord* const search_start = block_align_down(addr);
+  const size_t block_offset = addr_to_block_ptr(addr)->offset();
 
   const ParMarkBitMap* bitmap = PSParallelCompact::mark_bitmap();
-  size_t live_to_left = bitmap->live_words_in_range(search_start, oop(addr));
-
-  result += partial_obj_size + live_to_left;
-  DEBUG_ONLY(PSParallelCompact::check_new_location(addr, result);)
+  const size_t live = bitmap->live_words_in_range(search_start, oop(addr));
+  result += block_offset + live;
+  DEBUG_ONLY(PSParallelCompact::check_new_location(addr, result));
   return result;
 }
 
-#ifdef  ASSERT
+#ifdef ASSERT
 void ParallelCompactData::verify_clear(const PSVirtualSpace* vspace)
 {
   const size_t* const beg = (const size_t*)vspace->committed_low_addr();
@@ -762,16 +795,10 @@
 void ParallelCompactData::verify_clear()
 {
   verify_clear(_region_vspace);
+  verify_clear(_block_vspace);
 }
 #endif  // #ifdef ASSERT
 
-#ifdef NOT_PRODUCT
-ParallelCompactData::RegionData* debug_region(size_t region_index) {
-  ParallelCompactData& sd = PSParallelCompact::summary_data();
-  return sd.region(region_index);
-}
-#endif
-
 elapsedTimer        PSParallelCompact::_accumulated_time;
 unsigned int        PSParallelCompact::_total_invocations = 0;
 unsigned int        PSParallelCompact::_maximum_compaction_gc_num = 0;
@@ -1961,11 +1988,6 @@
                                       maximum_heap_compaction);
 }
 
-bool ParallelCompactData::region_contains(size_t region_index, HeapWord* addr) {
-  size_t addr_region_index = addr_to_region_idx(addr);
-  return region_index == addr_region_index;
-}
-
 // This method contains no policy. You should probably
 // be calling invoke() instead.
 bool PSParallelCompact::invoke_no_policy(bool maximum_heap_compaction) {
@@ -2627,6 +2649,41 @@
   }
 }
 
+#ifdef ASSERT
+// Write a histogram of the number of times the block table was filled for a
+// region.
+void PSParallelCompact::write_block_fill_histogram(outputStream* const out)
+{
+  if (!TraceParallelOldGCCompactionPhase) return;
+
+  typedef ParallelCompactData::RegionData rd_t;
+  ParallelCompactData& sd = summary_data();
+
+  for (unsigned int id = old_space_id; id < last_space_id; ++id) {
+    MutableSpace* const spc = _space_info[id].space();
+    if (spc->bottom() != spc->top()) {
+      const rd_t* const beg = sd.addr_to_region_ptr(spc->bottom());
+      HeapWord* const top_aligned_up = sd.region_align_up(spc->top());
+      const rd_t* const end = sd.addr_to_region_ptr(top_aligned_up);
+
+      size_t histo[5] = { 0, 0, 0, 0, 0 };
+      const size_t histo_len = sizeof(histo) / sizeof(size_t);
+      const size_t region_cnt = pointer_delta(end, beg, sizeof(rd_t));
+
+      for (const rd_t* cur = beg; cur < end; ++cur) {
+        ++histo[MIN2(cur->blocks_filled_count(), histo_len - 1)];
+      }
+      out->print("%u %-4s" SIZE_FORMAT_W(5), id, space_names[id], region_cnt);
+      for (size_t i = 0; i < histo_len; ++i) {
+        out->print(" " SIZE_FORMAT_W(5) " %5.1f%%",
+                   histo[i], 100.0 * histo[i] / region_cnt);
+      }
+      out->cr();
+    }
+  }
+}
+#endif // #ifdef ASSERT
+
 void PSParallelCompact::compact() {
   // trace("5");
   TraceTime tm("compaction phase", print_phases(), true, gclog_or_tty);
@@ -2666,6 +2723,8 @@
       update_deferred_objects(cm, SpaceId(id));
     }
   }
+
+  DEBUG_ONLY(write_block_fill_histogram(gclog_or_tty));
 }
 
 #ifdef  ASSERT
@@ -3130,6 +3189,57 @@
   } while (true);
 }
 
+void PSParallelCompact::fill_blocks(size_t region_idx)
+{
+  // Fill in the block table elements for the specified region.  Each block
+  // table element holds the number of live words in the region that are to the
+  // left of the first object that starts in the block.  Thus only blocks in
+  // which an object starts need to be filled.
+  //
+  // The algorithm scans the section of the bitmap that corresponds to the
+  // region, keeping a running total of the live words.  When an object start is
+  // found, if it's the first to start in the block that contains it, the
+  // current total is written to the block table element.
+  const size_t Log2BlockSize = ParallelCompactData::Log2BlockSize;
+  const size_t Log2RegionSize = ParallelCompactData::Log2RegionSize;
+  const size_t RegionSize = ParallelCompactData::RegionSize;
+
+  ParallelCompactData& sd = summary_data();
+  const size_t partial_obj_size = sd.region(region_idx)->partial_obj_size();
+  if (partial_obj_size >= RegionSize) {
+    return; // No objects start in this region.
+  }
+
+  // Ensure the first loop iteration decides that the block has changed.
+  size_t cur_block = sd.block_count();
+
+  const ParMarkBitMap* const bitmap = mark_bitmap();
+
+  const size_t Log2BitsPerBlock = Log2BlockSize - LogMinObjAlignment;
+  assert((size_t)1 << Log2BitsPerBlock ==
+         bitmap->words_to_bits(ParallelCompactData::BlockSize), "sanity");
+
+  size_t beg_bit = bitmap->words_to_bits(region_idx << Log2RegionSize);
+  const size_t range_end = beg_bit + bitmap->words_to_bits(RegionSize);
+  size_t live_bits = bitmap->words_to_bits(partial_obj_size);
+  beg_bit = bitmap->find_obj_beg(beg_bit + live_bits, range_end);
+  while (beg_bit < range_end) {
+    const size_t new_block = beg_bit >> Log2BitsPerBlock;
+    if (new_block != cur_block) {
+      cur_block = new_block;
+      sd.block(cur_block)->set_offset(bitmap->bits_to_words(live_bits));
+    }
+
+    const size_t end_bit = bitmap->find_obj_end(beg_bit, range_end);
+    if (end_bit < range_end - 1) {
+      live_bits += end_bit - beg_bit + 1;
+      beg_bit = bitmap->find_obj_beg(end_bit + 1, range_end);
+    } else {
+      return;
+    }
+  }
+}
+
 void
 PSParallelCompact::move_and_update(ParCompactionManager* cm, SpaceId space_id) {
   const MutableSpace* sp = space(space_id);
diff -r 6bf8b8bb7c19 -r f8c8cace25ad src/share/vm/gc_implementation/parallelScavenge/psParallelCompact.hpp
--- a/src/share/vm/gc_implementation/parallelScavenge/psParallelCompact.hpp	Wed Jun 05 14:12:49 2013 -0400
+++ b/src/share/vm/gc_implementation/parallelScavenge/psParallelCompact.hpp	Thu Jun 06 05:56:33 2013 -0700
@@ -220,6 +220,17 @@
   // Mask for the bits in a pointer to get the address of the start of a region.
   static const size_t RegionAddrMask;
 
+  static const size_t Log2BlockSize;
+  static const size_t BlockSize;
+  static const size_t BlockSizeBytes;
+
+  static const size_t BlockSizeOffsetMask;
+  static const size_t BlockAddrOffsetMask;
+  static const size_t BlockAddrMask;
+
+  static const size_t BlocksPerRegion;
+  static const size_t Log2BlocksPerRegion;
+
   class RegionData
   {
   public:
@@ -272,6 +283,12 @@
     inline uint destination_count() const;
     inline uint destination_count_raw() const;
 
+    // Whether the block table for this region has been filled.
+    inline bool blocks_filled() const;
+
+    // Number of times the block table was filled.
+    DEBUG_ONLY(inline size_t blocks_filled_count() const;)
+
     // The location of the java heap data that corresponds to this region.
     inline HeapWord* data_location() const;
 
@@ -296,6 +313,7 @@
     void set_partial_obj_size(size_t words)    {
       _partial_obj_size = (region_sz_t) words;
     }
+    inline void set_blocks_filled();
 
     inline void set_destination_count(uint count);
     inline void set_live_obj_size(size_t words);
@@ -328,7 +346,11 @@
     HeapWord*            _partial_obj_addr;
     region_sz_t          _partial_obj_size;
     region_sz_t volatile _dc_and_los;
+    bool                 _blocks_filled;
+
 #ifdef ASSERT
+    size_t               _blocks_filled_count;   // Number of block table fills.
+
     // These enable optimizations that are only partially implemented.  Use
     // debug builds to prevent the code fragments from breaking.
     HeapWord*            _data_location;
@@ -337,11 +359,26 @@
 
 #ifdef ASSERT
    public:
-    uint            _pushed;   // 0 until region is pushed onto a worker's stack
+    uint                 _pushed;   // 0 until region is pushed onto a stack
    private:
 #endif
   };
 
+  // "Blocks" allow shorter sections of the bitmap to be searched.  Each Block
+  // holds an offset, which is the amount of live data in the Region to the left
+  // of the first live object that starts in the Block.
+  class BlockData
+  {
+  public:
+    typedef unsigned short int blk_ofs_t;
+
+    blk_ofs_t offset() const    { return _offset; }
+    void set_offset(size_t val) { _offset = (blk_ofs_t)val; }
+
+  private:
+    blk_ofs_t _offset;
+  };
+
 public:
   ParallelCompactData();
   bool initialize(MemRegion covered_region);
@@ -353,8 +390,9 @@
   inline RegionData* region(size_t region_idx) const;
   inline size_t     region(const RegionData* const region_ptr) const;
 
-  // Returns true if the given address is contained within the region
-  bool region_contains(size_t region_index, HeapWord* addr);
+  size_t block_count() const { return _block_count; }
+  inline BlockData* block(size_t block_idx) const;
+  inline size_t     block(const BlockData* block_ptr) const;
 
   void add_obj(HeapWord* addr, size_t len);
   void add_obj(oop p, size_t len) { add_obj((HeapWord*)p, len); }
@@ -394,11 +432,24 @@
   inline HeapWord*  region_align_up(HeapWord* addr) const;
   inline bool       is_region_aligned(HeapWord* addr) const;
 
+  // Analogous to region_offset() for blocks.
+  size_t     block_offset(const HeapWord* addr) const;
+  size_t     addr_to_block_idx(const HeapWord* addr) const;
+  size_t     addr_to_block_idx(const oop obj) const {
+    return addr_to_block_idx((HeapWord*) obj);
+  }
+  inline BlockData* addr_to_block_ptr(const HeapWord* addr) const;
+  inline HeapWord*  block_to_addr(size_t block) const;
+  inline size_t     region_to_block_idx(size_t region) const;
+
+  inline HeapWord*  block_align_down(HeapWord* addr) const;
+  inline HeapWord*  block_align_up(HeapWord* addr) const;
+  inline bool       is_block_aligned(HeapWord* addr) const;
+
   // Return the address one past the end of the partial object.
   HeapWord* partial_obj_end(size_t region_idx) const;
 
-  // Return the new location of the object p after the
-  // the compaction.
+  // Return the location of the object after compaction.
   HeapWord* calc_new_pointer(HeapWord* addr);
 
   HeapWord* calc_new_pointer(oop p) {
@@ -411,6 +462,7 @@
 #endif  // #ifdef ASSERT
 
 private:
+  bool initialize_block_data();
   bool initialize_region_data(size_t region_size);
   PSVirtualSpace* create_vspace(size_t count, size_t element_size);
 
@@ -424,6 +476,10 @@
   size_t          _reserved_byte_size;
   RegionData*     _region_data;
   size_t          _region_count;
+
+  PSVirtualSpace* _block_vspace;
+  BlockData*      _block_data;
+  size_t          _block_count;
 };
 
 inline uint
@@ -438,6 +494,28 @@
   return destination_count_raw() >> dc_shift;
 }
 
+inline bool
+ParallelCompactData::RegionData::blocks_filled() const
+{
+  return _blocks_filled;
+}
+
+#ifdef ASSERT
+inline size_t
+ParallelCompactData::RegionData::blocks_filled_count() const
+{
+  return _blocks_filled_count;
+}
+#endif // #ifdef ASSERT
+
+inline void
+ParallelCompactData::RegionData::set_blocks_filled()
+{
+  _blocks_filled = true;
+  // Debug builds count the number of times the table was filled.
+  DEBUG_ONLY(Atomic::inc_ptr(&_blocks_filled_count));
+}
+
 inline void
 ParallelCompactData::RegionData::set_destination_count(uint count)
 {
@@ -532,6 +610,12 @@
   return pointer_delta(region_ptr, _region_data, sizeof(RegionData));
 }
 
+inline ParallelCompactData::BlockData*
+ParallelCompactData::block(size_t n) const {
+  assert(n < block_count(), "bad arg");
+  return _block_data + n;
+}
+
 inline size_t
 ParallelCompactData::region_offset(const HeapWord* addr) const
 {
@@ -598,6 +682,63 @@
   return region_offset(addr) == 0;
 }
 
+inline size_t
+ParallelCompactData::block_offset(const HeapWord* addr) const
+{
+  assert(addr >= _region_start, "bad addr");
+  assert(addr <= _region_end, "bad addr");
+  return (size_t(addr) & BlockAddrOffsetMask) >> LogHeapWordSize;
+}
+
+inline size_t
+ParallelCompactData::addr_to_block_idx(const HeapWord* addr) const
+{
+  assert(addr >= _region_start, "bad addr");
+  assert(addr <= _region_end, "bad addr");
+  return pointer_delta(addr, _region_start) >> Log2BlockSize;
+}
+
+inline ParallelCompactData::BlockData*
+ParallelCompactData::addr_to_block_ptr(const HeapWord* addr) const
+{
+  return block(addr_to_block_idx(addr));
+}
+
+inline HeapWord*
+ParallelCompactData::block_to_addr(size_t block) const
+{
+  assert(block < _block_count, "block out of range");
+  return _region_start + (block << Log2BlockSize);
+}
+
+inline size_t
+ParallelCompactData::region_to_block_idx(size_t region) const
+{
+  return region << Log2BlocksPerRegion;
+}
+
+inline HeapWord*
+ParallelCompactData::block_align_down(HeapWord* addr) const
+{
+  assert(addr >= _region_start, "bad addr");
+  assert(addr < _region_end + RegionSize, "bad addr");
+  return (HeapWord*)(size_t(addr) & BlockAddrMask);
+}
+
+inline HeapWord*
+ParallelCompactData::block_align_up(HeapWord* addr) const
+{
+  assert(addr >= _region_start, "bad addr");
+  assert(addr <= _region_end, "bad addr");
+  return block_align_down(addr + BlockSizeOffsetMask);
+}
+
+inline bool
+ParallelCompactData::is_block_aligned(HeapWord* addr) const
+{
+  return block_offset(addr) == 0;
+}
+
 // Abstract closure for use with ParMarkBitMap::iterate(), which will invoke the
 // do_addr() method.
 //
@@ -775,6 +916,7 @@
   // Convenient access to type names.
   typedef ParMarkBitMap::idx_t idx_t;
   typedef ParallelCompactData::RegionData RegionData;
+  typedef ParallelCompactData::BlockData BlockData;
 
   typedef enum {
     old_space_id, eden_space_id,
@@ -962,6 +1104,8 @@
   // Adjust addresses in roots.  Does not adjust addresses in heap.
   static void adjust_roots();
 
+  DEBUG_ONLY(static void write_block_fill_histogram(outputStream* const out);)
+
   // Move objects to new locations.
   static void compact_perm(ParCompactionManager* cm);
   static void compact();
@@ -1128,6 +1272,9 @@
     fill_region(cm, region);
   }
 
+  // Fill in the block table for the specified region.
+  static void fill_blocks(size_t region_idx);
+
   // Update the deferred objects in the space.
   static void update_deferred_objects(ParCompactionManager* cm, SpaceId id);
 
diff -r 6bf8b8bb7c19 -r f8c8cace25ad src/share/vm/gc_implementation/parallelScavenge/psPromotionManager.inline.hpp
--- a/src/share/vm/gc_implementation/parallelScavenge/psPromotionManager.inline.hpp	Wed Jun 05 14:12:49 2013 -0400
+++ b/src/share/vm/gc_implementation/parallelScavenge/psPromotionManager.inline.hpp	Thu Jun 06 05:56:33 2013 -0700
@@ -42,7 +42,7 @@
     if (o->is_forwarded()) {
       o = o->forwardee();
       // Card mark
-      if (PSScavenge::is_obj_in_young((HeapWord*) o)) {
+      if (PSScavenge::is_obj_in_young(o)) {
         PSScavenge::card_table()->inline_write_ref_field_gc(p, o);
       }
       oopDesc::encode_store_heap_oop_not_null(p, o);
diff -r 6bf8b8bb7c19 -r f8c8cace25ad src/share/vm/gc_implementation/parallelScavenge/psScavenge.cpp
--- a/src/share/vm/gc_implementation/parallelScavenge/psScavenge.cpp	Wed Jun 05 14:12:49 2013 -0400
+++ b/src/share/vm/gc_implementation/parallelScavenge/psScavenge.cpp	Thu Jun 06 05:56:33 2013 -0700
@@ -61,6 +61,7 @@
 bool                       PSScavenge::_survivor_overflow = false;
 uint                       PSScavenge::_tenuring_threshold = 0;
 HeapWord*                  PSScavenge::_young_generation_boundary = NULL;
+uintptr_t                  PSScavenge::_young_generation_boundary_compressed = 0;
 elapsedTimer               PSScavenge::_accumulated_time;
 Stack<markOop, mtGC>       PSScavenge::_preserved_mark_stack;
 Stack<oop, mtGC>           PSScavenge::_preserved_oop_stack;
@@ -71,7 +72,7 @@
 class PSIsAliveClosure: public BoolObjectClosure {
 public:
   bool do_object_b(oop p) {
-    return (!PSScavenge::is_obj_in_young((HeapWord*) p)) || p->is_forwarded();
+    return (!PSScavenge::is_obj_in_young(p)) || p->is_forwarded();
   }
 };
 
@@ -815,7 +816,7 @@
   // Set boundary between young_gen and old_gen
   assert(old_gen->reserved().end() <= young_gen->eden_space()->bottom(),
          "old above young");
-  _young_generation_boundary = young_gen->eden_space()->bottom();
+  set_young_generation_boundary(young_gen->eden_space()->bottom());
 
   // Initialize ref handling object for scavenging.
   MemRegion mr = young_gen->reserved();
diff -r 6bf8b8bb7c19 -r f8c8cace25ad src/share/vm/gc_implementation/parallelScavenge/psScavenge.hpp
--- a/src/share/vm/gc_implementation/parallelScavenge/psScavenge.hpp	Wed Jun 05 14:12:49 2013 -0400
+++ b/src/share/vm/gc_implementation/parallelScavenge/psScavenge.hpp	Thu Jun 06 05:56:33 2013 -0700
@@ -62,19 +62,22 @@
 
  protected:
   // Flags/counters
-  static ReferenceProcessor* _ref_processor;        // Reference processor for scavenging.
-  static PSIsAliveClosure    _is_alive_closure;     // Closure used for reference processing
-  static CardTableExtension* _card_table;           // We cache the card table for fast access.
-  static bool                _survivor_overflow;    // Overflow this collection
-  static uint                _tenuring_threshold;   // tenuring threshold for next scavenge
-  static elapsedTimer        _accumulated_time;     // total time spent on scavenge
-  static HeapWord*           _young_generation_boundary; // The lowest address possible for the young_gen.
-                                                         // This is used to decide if an oop should be scavenged,
-                                                         // cards should be marked, etc.
+  static ReferenceProcessor*  _ref_processor;        // Reference processor for scavenging.
+  static PSIsAliveClosure     _is_alive_closure;     // Closure used for reference processing
+  static CardTableExtension*  _card_table;           // We cache the card table for fast access.
+  static bool                 _survivor_overflow;    // Overflow this collection
+  static uint                 _tenuring_threshold;   // tenuring threshold for next scavenge
+  static elapsedTimer         _accumulated_time;     // total time spent on scavenge
+  // The lowest address possible for the young_gen.
+  // This is used to decide if an oop should be scavenged,
+  // cards should be marked, etc.
+  static HeapWord*            _young_generation_boundary;
+  // Used to optimize compressed oops young gen boundary checking.
+  static uintptr_t            _young_generation_boundary_compressed;
   static Stack<markOop, mtGC> _preserved_mark_stack; // List of marks to be restored after failed promotion
   static Stack<oop, mtGC>     _preserved_oop_stack;  // List of oops that need their mark restored.
-  static CollectorCounters*      _counters;          // collector performance counters
-  static bool                    _promotion_failed;
+  static CollectorCounters*   _counters;             // collector performance counters
+  static bool                 _promotion_failed;
 
   static void clean_up_failed_promotion();
 
@@ -112,6 +115,9 @@
   // boundary moves, _young_generation_boundary must be reset
   static void set_young_generation_boundary(HeapWord* v) {
     _young_generation_boundary = v;
+    if (UseCompressedOops) {
+      _young_generation_boundary_compressed = (uintptr_t)oopDesc::encode_heap_oop((oop)v);
+    }
   }
 
   // Called by parallelScavengeHeap to init the tenuring threshold
@@ -140,11 +146,19 @@
   static void copy_and_push_safe_barrier_from_klass(PSPromotionManager* pm, oop* p);
 
   // Is an object in the young generation
-  // This assumes that the HeapWord argument is in the heap,
+  // This assumes that the 'o' is in the heap,
   // so it only checks one side of the complete predicate.
+
+  inline static bool is_obj_in_young(oop o) {
+    return (HeapWord*)o >= _young_generation_boundary;
+  }
+
+  inline static bool is_obj_in_young(narrowOop o) {
+    return (uintptr_t)o >= _young_generation_boundary_compressed;
+  }
+
   inline static bool is_obj_in_young(HeapWord* o) {
-    const bool result = (o >= _young_generation_boundary);
-    return result;
+    return o >= _young_generation_boundary;
   }
 };
 
diff -r 6bf8b8bb7c19 -r f8c8cace25ad src/share/vm/gc_implementation/parallelScavenge/psScavenge.inline.hpp
--- a/src/share/vm/gc_implementation/parallelScavenge/psScavenge.inline.hpp	Wed Jun 05 14:12:49 2013 -0400
+++ b/src/share/vm/gc_implementation/parallelScavenge/psScavenge.inline.hpp	Thu Jun 06 05:56:33 2013 -0700
@@ -39,9 +39,7 @@
 
 template <class T> inline bool PSScavenge::should_scavenge(T* p) {
   T heap_oop = oopDesc::load_heap_oop(p);
-  if (oopDesc::is_null(heap_oop)) return false;
-  oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
-  return PSScavenge::is_obj_in_young((HeapWord*)obj);
+  return PSScavenge::is_obj_in_young(heap_oop);
 }
 
 template <class T>
@@ -94,7 +92,7 @@
   // or from metadata.
   if ((!PSScavenge::is_obj_in_young((HeapWord*)p)) &&
       Universe::heap()->is_in_reserved(p)) {
-    if (PSScavenge::is_obj_in_young((HeapWord*)new_obj)) {
+    if (PSScavenge::is_obj_in_young(new_obj)) {
       card_table()->inline_write_ref_field_gc(p, new_obj);
     }
   }
@@ -147,7 +145,7 @@
       }
       oopDesc::encode_store_heap_oop_not_null(p, new_obj);
 
-      if (PSScavenge::is_obj_in_young((HeapWord*)new_obj)) {
+      if (PSScavenge::is_obj_in_young(new_obj)) {
         do_klass_barrier();
       }
     }
diff -r 6bf8b8bb7c19 -r f8c8cace25ad src/share/vm/gc_implementation/shared/parGCAllocBuffer.hpp
--- a/src/share/vm/gc_implementation/shared/parGCAllocBuffer.hpp	Wed Jun 05 14:12:49 2013 -0400
+++ b/src/share/vm/gc_implementation/shared/parGCAllocBuffer.hpp	Thu Jun 06 05:56:33 2013 -0700
@@ -158,7 +158,7 @@
   // Fills in the unallocated portion of the buffer with a garbage object.
   // If "end_of_gc" is TRUE, is after the last use in the GC.  IF "retain"
   // is true, attempt to re-use the unused portion in the next GC.
-  void retire(bool end_of_gc, bool retain);
+  virtual void retire(bool end_of_gc, bool retain);
 
   void print() PRODUCT_RETURN;
 };
diff -r 6bf8b8bb7c19 -r f8c8cace25ad test/gc/g1/TestSummarizeRSetStats.java
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/test/gc/g1/TestSummarizeRSetStats.java	Thu Jun 06 05:56:33 2013 -0700
@@ -0,0 +1,164 @@
+/*
+ * Copyright (c) 2013, Oracle and/or its affiliates. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ */
+
+/*
+ * @test TestSummarizeRSetStats.java
+ * @bug 8013895
+ * @library /testlibrary
+ * @build TestSummarizeRSetStats
+ * @summary Verify output of -XX:+G1SummarizeRSetStats
+ * @run main TestSummarizeRSetStats
+ *
+ * Test the output of G1SummarizeRSetStats in conjunction with G1SummarizeRSetStatsPeriod.
+ */
+
+import com.oracle.java.testlibrary.*;
+import java.lang.Thread;
+import java.util.ArrayList;
+import java.util.Arrays;
+
+class RunSystemGCs {
+    // 4M size, both are directly allocated into the old gen
+    static Object[] largeObject1 = new Object[1024 * 1024];
+    static Object[] largeObject2 = new Object[1024 * 1024];
+
+    static int[] temp;
+
+    public static void main(String[] args) {
+        // create some cross-references between these objects
+        for (int i = 0; i < largeObject1.length; i++) {
+            largeObject1[i] = largeObject2;
+        }
+
+        for (int i = 0; i < largeObject2.length; i++) {
+            largeObject2[i] = largeObject1;
+        }
+
+        int numGCs = Integer.parseInt(args[0]);
+
+        if (numGCs > 0) {
+            // try to force a minor collection: the young gen is 4M, the
+            // amount of data allocated below is roughly that (4*1024*1024 +
+            // some header data)
+            for (int i = 0; i < 1024 ; i++) {
+                temp = new int[1024];
+            }
+        }
+
+        for (int i = 0; i < numGCs - 1; i++) {
+            System.gc();
+        }
+    }
+}
+
+public class TestSummarizeRSetStats {
+
+    public static String runTest(String[] additionalArgs, int numGCs) throws Exception {
+        ArrayList<String> finalargs = new ArrayList<String>();
+        String[] defaultArgs = new String[] {
+            "-XX:+UseG1GC",
+            "-Xmn4m",
+            "-Xmx20m",
+            "-XX:InitiatingHeapOccupancyPercent=100", // we don't want the additional GCs due to initial marking
+            "-XX:+PrintGC",
+            "-XX:+UnlockDiagnosticVMOptions",
+            "-XX:G1HeapRegionSize=1M",
+        };
+
+        finalargs.addAll(Arrays.asList(defaultArgs));
+
+        if (additionalArgs != null) {
+            finalargs.addAll(Arrays.asList(additionalArgs));
+        }
+
+        finalargs.add(RunSystemGCs.class.getName());
+        finalargs.add(String.valueOf(numGCs));
+
+        ProcessBuilder pb = ProcessTools.createJavaProcessBuilder(
+            finalargs.toArray(new String[0]));
+        OutputAnalyzer output = new OutputAnalyzer(pb.start());
+
+        output.shouldHaveExitValue(0);
+
+        String result = output.getStdout();
+        return result;
+    }
+
+    private static void expectStatistics(String result, int expectedCumulative, int expectedPeriodic) throws Exception {
+        int actualTotal = result.split("Concurrent RS processed").length - 1;
+        int actualCumulative = result.split("Cumulative RS summary").length - 1;
+
+        if (expectedCumulative != actualCumulative) {
+            throw new Exception("Incorrect amount of RSet summaries at the end. Expected " + expectedCumulative + ", got " + actualCumulative);
+        }
+
+        if (expectedPeriodic != (actualTotal - actualCumulative)) {
+            throw new Exception("Incorrect amount of per-period RSet summaries at the end. Expected " + expectedPeriodic + ", got " + (actualTotal - actualCumulative));
+        }
+    }
+
+    public static void main(String[] args) throws Exception {
+        String result;
+
+        // no RSet statistics output
+        result = runTest(null, 0);
+        expectStatistics(result, 0, 0);
+
+        // no RSet statistics output
+        result = runTest(null, 2);
+        expectStatistics(result, 0, 0);
+
+        // no RSet statistics output
+        result = runTest(new String[] { "-XX:G1SummarizeRSetStatsPeriod=1" }, 3);
+        expectStatistics(result, 0, 0);
+
+        // single RSet statistics output at the end
+        result = runTest(new String[] { "-XX:+G1SummarizeRSetStats" }, 0);
+        expectStatistics(result, 1, 0);
+
+        // single RSet statistics output at the end
+        result = runTest(new String[] { "-XX:+G1SummarizeRSetStats" }, 2);
+        expectStatistics(result, 1, 0);
+
+        // single RSet statistics output
+        result = runTest(new String[] { "-XX:+G1SummarizeRSetStats", "-XX:G1SummarizeRSetStatsPeriod=1" }, 0);
+        expectStatistics(result, 1, 0);
+
+        // two times RSet statistics output
+        result = runTest(new String[] { "-XX:+G1SummarizeRSetStats", "-XX:G1SummarizeRSetStatsPeriod=1" }, 1);
+        expectStatistics(result, 1, 1);
+
+        // four times RSet statistics output
+        result = runTest(new String[] { "-XX:+G1SummarizeRSetStats", "-XX:G1SummarizeRSetStatsPeriod=1" }, 3);
+        expectStatistics(result, 1, 3);
+
+        // three times RSet statistics output
+        result = runTest(new String[] { "-XX:+G1SummarizeRSetStats", "-XX:G1SummarizeRSetStatsPeriod=2" }, 3);
+        expectStatistics(result, 1, 2);
+
+        // single RSet statistics output
+        result = runTest(new String[] { "-XX:+G1SummarizeRSetStats", "-XX:G1SummarizeRSetStatsPeriod=100" }, 3);
+        expectStatistics(result, 1, 1);
+    }
+}
+