8296954: G1: Enable parallel scanning for heap region remset

Reviewed-by: tschatzl, ayang

src/hotspot/share/gc/g1/g1CardSet.cpp

@@ -35,6 +35,7 @@
 #include "runtime/java.hpp"
 #include "utilities/bitMap.inline.hpp"
 #include "utilities/concurrentHashTable.inline.hpp"
+#include "utilities/concurrentHashTableTasks.inline.hpp"
 #include "utilities/globalDefinitions.hpp"
 
 G1CardSet::ContainerPtr G1CardSet::FullCardSet = (G1CardSet::ContainerPtr)-1;
@@ -240,12 +241,15 @@ void G1CardSetCoarsenStats::print_on(outputStream* out) {
 
 class G1CardSetHashTable : public CHeapObj<mtGCCardSet> {
   using ContainerPtr = G1CardSet::ContainerPtr;
+  using CHTScanTask = CardSetHash::ScanTask;
 
+  const static uint BucketClaimSize = 16;
   // Did we insert at least one card in the table?
   bool volatile _inserted_card;
 
   G1CardSetMemoryManager* _mm;
   CardSetHash _table;
+  CHTScanTask _table_scanner;
 
   class G1CardSetHashTableLookUp : public StackObj {
     uint _region_idx;
@@ -291,7 +295,8 @@ public:
                      size_t initial_log_table_size = InitialLogTableSize) :
     _inserted_card(false),
     _mm(mm),
-    _table(mm, initial_log_table_size, false) {
+    _table(mm, initial_log_table_size, false),
+    _table_scanner(&_table, BucketClaimSize) {
   }
 
   ~G1CardSetHashTable() {
@@ -332,7 +337,7 @@ public:
 
   void iterate_safepoint(G1CardSet::ContainerPtrClosure* cl2) {
     G1CardSetHashTableScan cl(cl2);
-    _table.do_safepoint_scan(cl);
+    _table_scanner.do_safepoint_scan(cl);
   }
 
   void iterate(G1CardSet::ContainerPtrClosure* cl2) {
@@ -347,6 +352,10 @@ public:
     }
   }
 
+  void reset_table_scanner() {
+    _table_scanner.set(&_table, BucketClaimSize);
+  }
+
   void grow() {
     size_t new_limit = _table.get_size_log2(Thread::current()) + 1;
     _table.grow(Thread::current(), new_limit);
@@ -1029,3 +1038,7 @@ void G1CardSet::clear() {
   _num_occupied = 0;
   _mm->flush();
 }
+
+void G1CardSet::reset_table_scanner() {
+  _table->reset_table_scanner();
+}

src/hotspot/share/gc/g1/g1CardSet.hpp

@@ -378,6 +378,8 @@ public:
   // Clear the entire contents of this remembered set.
   void clear();
 
+  void reset_table_scanner();
+
   // Iterate over the container, calling a method on every card or card range contained
   // in the card container.
   // For every container, first calls

src/hotspot/share/gc/g1/g1RemSet.cpp

@@ -916,6 +916,8 @@ void G1RemSet::assert_scan_top_is_null(uint hrm_index) {
 void G1RemSet::prepare_region_for_scan(HeapRegion* r) {
   uint hrm_index = r->hrm_index();
 
+  r->prepare_remset_for_scan();
+
   // Only update non-collection set old regions, others must have already been set
   // to NULL (don't scan) in the initialization.
   if (r->in_collection_set()) {
@@ -1358,7 +1360,7 @@ public:
         G1ClearBitmapClosure clear(g1h);
         G1CombinedClosure combined(&merge, &clear);
 
-        g1h->collection_set_iterate_increment_from(&combined, &_hr_claimer, worker_id);
+        g1h->collection_set_iterate_increment_from(&combined, nullptr, worker_id);
         stats = merge.stats();
       }
 

src/hotspot/share/gc/g1/heapRegion.cpp

@@ -220,6 +220,10 @@ void HeapRegion::clear_humongous() {
   _humongous_start_region = NULL;
 }
 
+void HeapRegion::prepare_remset_for_scan() {
+  return _rem_set->reset_table_scanner();
+}
+
 HeapRegion::HeapRegion(uint hrm_index,
                        G1BlockOffsetTable* bot,
                        MemRegion mr,

src/hotspot/share/gc/g1/heapRegion.hpp

@@ -457,6 +457,8 @@ public:
 
   inline bool in_collection_set() const;
 
+  void prepare_remset_for_scan();
+
   // Methods used by the HeapRegionSetBase class and subclasses.
 
   // Getter and setter for the next and prev fields used to link regions into

src/hotspot/share/gc/g1/heapRegionRemSet.cpp

@@ -83,6 +83,10 @@ void HeapRegionRemSet::clear_locked(bool only_cardset) {
   assert(occupied() == 0, "Should be clear.");
 }
 
+void HeapRegionRemSet::reset_table_scanner() {
+  _card_set.reset_table_scanner();
+}
+
 G1MonotonicArenaMemoryStats HeapRegionRemSet::card_set_memory_stats() const {
   return _card_set_mm.memory_stats();
 }

src/hotspot/share/gc/g1/heapRegionRemSet.hpp

@@ -122,6 +122,8 @@ public:
   void clear(bool only_cardset = false);
   void clear_locked(bool only_cardset = false);
 
+  void reset_table_scanner();
+
   G1MonotonicArenaMemoryStats card_set_memory_stats() const;
 
   // The actual # of bytes this hr_remset takes up. Also includes the code

src/hotspot/share/utilities/concurrentHashTableTasks.inline.hpp

@@ -56,6 +56,7 @@ public:
 
     void set(size_t claim_size, InternalTable* table) {
       assert(table != nullptr, "precondition");
+      _next = 0;
       _limit = table->_size;
       _size  = MIN2(claim_size, _limit);
     }