6819085: G1: use larger and/or user settable region size

Instead of the region size being hard-coded, allow the user to set it.

Reviewed-by: jmasa, johnc, apetrusenko

hotspot/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp

@@ -25,6 +25,8 @@
 #include "incls/_precompiled.incl"
 #include "incls/_g1CollectedHeap.cpp.incl"
 
+size_t G1CollectedHeap::_humongous_object_threshold_in_words = 0;
+
 // turn it on so that the contents of the young list (scan-only /
 // to-be-collected) are printed at "strategic" points before / during
 // / after the collection --- this is useful for debugging
@@ -1394,6 +1396,9 @@ G1CollectedHeap::G1CollectedHeap(G1CollectorPolicy* policy_) :
   if (_process_strong_tasks == NULL || !_process_strong_tasks->valid()) {
     vm_exit_during_initialization("Failed necessary allocation.");
   }
+
+  _humongous_object_threshold_in_words = HeapRegion::GrainWords / 2;
+
   int n_queues = MAX2((int)ParallelGCThreads, 1);
   _task_queues = new RefToScanQueueSet(n_queues);
 
@@ -1546,9 +1551,10 @@ jint G1CollectedHeap::initialize() {
   const size_t max_region_idx = ((size_t)1 << (sizeof(RegionIdx_t)*BitsPerByte-1)) - 1;
   guarantee((max_regions() - 1) <= max_region_idx, "too many regions");
 
-  const size_t cards_per_region = HeapRegion::GrainBytes >> CardTableModRefBS::card_shift;
   size_t max_cards_per_region = ((size_t)1 << (sizeof(CardIdx_t)*BitsPerByte-1)) - 1;
-  guarantee(cards_per_region < max_cards_per_region, "too many cards per region");
+  guarantee(HeapRegion::CardsPerRegion > 0, "make sure it's initialized");
+  guarantee((size_t) HeapRegion::CardsPerRegion < max_cards_per_region,
+            "too many cards per region");
 
   _bot_shared = new G1BlockOffsetSharedArray(_reserved,
                                              heap_word_size(init_byte_size));

hotspot/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp

@@ -167,16 +167,11 @@ class G1CollectedHeap : public SharedHeap {
   friend class G1MarkSweep;
 
 private:
-  enum SomePrivateConstants {
-    VeryLargeInBytes = HeapRegion::GrainBytes/2,
-    VeryLargeInWords = VeryLargeInBytes/HeapWordSize,
-    MinHeapDeltaBytes = 10 * HeapRegion::GrainBytes,      // FIXME
-    NumAPIs = HeapRegion::MaxAge
-  };
-
   // The one and only G1CollectedHeap, so static functions can find it.
   static G1CollectedHeap* _g1h;
 
+  static size_t _humongous_object_threshold_in_words;
+
   // Storage for the G1 heap (excludes the permanent generation).
   VirtualSpace _g1_storage;
   MemRegion    _g1_reserved;
@@ -1021,7 +1016,7 @@ public:
 
   // Returns "true" iff the given word_size is "very large".
   static bool isHumongous(size_t word_size) {
-    return word_size >= VeryLargeInWords;
+    return word_size >= _humongous_object_threshold_in_words;
   }
 
   // Update mod union table with the set of dirty cards.

hotspot/src/share/vm/gc_implementation/g1/g1CollectorPolicy.cpp

@@ -201,6 +201,11 @@ G1CollectorPolicy::G1CollectorPolicy() :
   _survivors_age_table(true)
 
 {
+  // Set up the region size and associated fields. Given that the
+  // policy is created before the heap, we have to set this up here,
+  // so it's done as soon as possible.
+  HeapRegion::setup_heap_region_size(Arguments::min_heap_size());
+
   _recent_prev_end_times_for_all_gcs_sec->add(os::elapsedTime());
   _prev_collection_pause_end_ms = os::elapsedTime() * 1000.0;
 

hotspot/src/share/vm/gc_implementation/g1/g1CollectorPolicy.hpp

@@ -92,9 +92,7 @@ protected:
   int _parallel_gc_threads;
 
   enum SomePrivateConstants {
-    NumPrevPausesForHeuristics = 10,
+    NumPrevPausesForHeuristics = 10
-    NumPrevGCsForHeuristics = 10,
-    NumAPIs = HeapRegion::MaxAge
   };
 
   G1MMUTracker* _mmu_tracker;

hotspot/src/share/vm/gc_implementation/g1/g1_globals.hpp

@@ -37,7 +37,7 @@
   develop(intx, G1MarkingOverheadPercent, 0,                                \
           "Overhead of concurrent marking")                                 \
                                                                             \
-  product(intx, G1YoungGenSize, 0,                                          \
+  product(uintx, G1YoungGenSize, 0,                                         \
           "Size of the G1 young generation, 0 is the adaptive policy")      \
                                                                             \
   develop(bool, G1Gen, true,                                                \
@@ -246,6 +246,9 @@
           "If non-0 is the size of the G1 survivor space, "                 \
           "otherwise SurvivorRatio is used to determine the size")          \
                                                                             \
+  product(uintx, G1HeapRegionSize, 0,                                       \
+          "Size of the G1 regions.")                                        \
+                                                                            \
   experimental(bool, G1ParallelRSetUpdatingEnabled, false,                  \
           "Enables the parallelization of remembered set updating "         \
           "during evacuation pauses")                                       \

hotspot/src/share/vm/gc_implementation/g1/heapRegion.cpp

@@ -25,6 +25,12 @@
 #include "incls/_precompiled.incl"
 #include "incls/_heapRegion.cpp.incl"
 
+int HeapRegion::LogOfHRGrainBytes = 0;
+int HeapRegion::LogOfHRGrainWords = 0;
+int HeapRegion::GrainBytes        = 0;
+int HeapRegion::GrainWords        = 0;
+int HeapRegion::CardsPerRegion    = 0;
+
 HeapRegionDCTOC::HeapRegionDCTOC(G1CollectedHeap* g1,
                                  HeapRegion* hr, OopClosure* cl,
                                  CardTableModRefBS::PrecisionStyle precision,
@@ -231,6 +237,73 @@ void HeapRegionDCTOC::walk_mem_region_with_cl(MemRegion mr,
   }
 }
 
+// Minimum region size; we won't go lower than that.
+// We might want to decrease this in the future, to deal with small
+// heaps a bit more efficiently.
+#define MIN_REGION_SIZE  (      1024 * 1024 )
+
+// Maximum region size; we don't go higher than that. There's a good
+// reason for having an upper bound. We don't want regions to get too
+// large, otherwise cleanup's effectiveness would decrease as there
+// will be fewer opportunities to find totally empty regions after
+// marking.
+#define MAX_REGION_SIZE  ( 32 * 1024 * 1024 )
+
+// The automatic region size calculation will try to have around this
+// many regions in the heap (based on the min heap size).
+#define TARGET_REGION_NUMBER          2048
+
+void HeapRegion::setup_heap_region_size(uintx min_heap_size) {
+  // region_size in bytes
+  uintx region_size = G1HeapRegionSize;
+  if (FLAG_IS_DEFAULT(G1HeapRegionSize)) {
+    // We base the automatic calculation on the min heap size. This
+    // can be problematic if the spread between min and max is quite
+    // wide, imagine -Xms128m -Xmx32g. But, if we decided it based on
+    // the max size, the region size might be way too large for the
+    // min size. Either way, some users might have to set the region
+    // size manually for some -Xms / -Xmx combos.
+
+    region_size = MAX2(min_heap_size / TARGET_REGION_NUMBER,
+                       (uintx) MIN_REGION_SIZE);
+  }
+
+  int region_size_log = log2_long((jlong) region_size);
+  // Recalculate the region size to make sure it's a power of
+  // 2. This means that region_size is the largest power of 2 that's
+  // <= what we've calculated so far.
+  region_size = 1 << region_size_log;
+
+  // Now make sure that we don't go over or under our limits.
+  if (region_size < MIN_REGION_SIZE) {
+    region_size = MIN_REGION_SIZE;
+  } else if (region_size > MAX_REGION_SIZE) {
+    region_size = MAX_REGION_SIZE;
+  }
+
+  // And recalculate the log.
+  region_size_log = log2_long((jlong) region_size);
+
+  // Now, set up the globals.
+  guarantee(LogOfHRGrainBytes == 0, "we should only set it once");
+  LogOfHRGrainBytes = region_size_log;
+
+  guarantee(LogOfHRGrainWords == 0, "we should only set it once");
+  LogOfHRGrainWords = LogOfHRGrainBytes - LogHeapWordSize;
+
+  guarantee(GrainBytes == 0, "we should only set it once");
+  // The cast to int is safe, given that we've bounded region_size by
+  // MIN_REGION_SIZE and MAX_REGION_SIZE.
+  GrainBytes = (int) region_size;
+
+  guarantee(GrainWords == 0, "we should only set it once");
+  GrainWords = GrainBytes >> LogHeapWordSize;
+  guarantee(1 << LogOfHRGrainWords == GrainWords, "sanity");
+
+  guarantee(CardsPerRegion == 0, "we should only set it once");
+  CardsPerRegion = GrainBytes >> CardTableModRefBS::card_shift;
+}
+
 void HeapRegion::reset_after_compaction() {
   G1OffsetTableContigSpace::reset_after_compaction();
   // After a compaction the mark bitmap is invalid, so we must

hotspot/src/share/vm/gc_implementation/g1/heapRegion.hpp

@@ -297,15 +297,24 @@ class HeapRegion: public G1OffsetTableContigSpace {
   HeapRegion(G1BlockOffsetSharedArray* sharedOffsetArray,
              MemRegion mr, bool is_zeroed);
 
-  enum SomePublicConstants {
+  static int LogOfHRGrainBytes;
-    // HeapRegions are GrainBytes-aligned
+  static int LogOfHRGrainWords;
-    // and have sizes that are multiples of GrainBytes.
+  // The normal type of these should be size_t. However, they used to
-    LogOfHRGrainBytes = 20,
+  // be members of an enum before and they are assumed by the
-    LogOfHRGrainWords = LogOfHRGrainBytes - LogHeapWordSize,
+  // compilers to be ints. To avoid going and fixing all their uses,
-    GrainBytes = 1 << LogOfHRGrainBytes,
+  // I'm declaring them as ints. I'm not anticipating heap region
-    GrainWords = 1 <<LogOfHRGrainWords,
+  // sizes to reach anywhere near 2g, so using an int here is safe.
-    MaxAge = 2, NoOfAges = MaxAge+1
+  static int GrainBytes;
-  };
+  static int GrainWords;
+  static int CardsPerRegion;
+
+  // It sets up the heap region size (GrainBytes / GrainWords), as
+  // well as other related fields that are based on the heap region
+  // size (LogOfHRGrainBytes / LogOfHRGrainWords /
+  // CardsPerRegion). All those fields are considered constant
+  // throughout the JVM's execution, therefore they should only be set
+  // up once during initialization time.
+  static void setup_heap_region_size(uintx min_heap_size);
 
   enum ClaimValues {
     InitialClaimValue     = 0,

hotspot/src/share/vm/gc_implementation/g1/heapRegionRemSet.cpp

@@ -57,10 +57,6 @@ class PerRegionTable: public CHeapObj {
 
 #endif // _MSC_VER
 
-  enum SomePrivateConstants {
-    CardsPerRegion = HeapRegion::GrainBytes >> CardTableModRefBS::card_shift
-  };
-
 protected:
   // We need access in order to union things into the base table.
   BitMap* bm() { return &_bm; }
@@ -76,7 +72,7 @@ protected:
 #if PRT_COUNT_OCCUPIED
     _occupied(0),
 #endif
-    _bm(CardsPerRegion, false /* in-resource-area */)
+    _bm(HeapRegion::CardsPerRegion, false /* in-resource-area */)
   {}
 
   static void free(PerRegionTable* prt) {
@@ -144,7 +140,8 @@ protected:
       CardIdx_t from_card = (CardIdx_t)
           hw_offset >> (CardTableModRefBS::card_shift - LogHeapWordSize);
 
-      assert(0 <= from_card && from_card < CardsPerRegion, "Must be in range.");
+      assert(0 <= from_card && from_card < HeapRegion::CardsPerRegion,
+             "Must be in range.");
       add_card_work(from_card, par);
     }
   }
@@ -631,7 +628,7 @@ void OtherRegionsTable::add_reference(OopOrNarrowOopStar from, int tid) {
         uintptr_t(from_hr->bottom())
           >> CardTableModRefBS::card_shift;
       CardIdx_t card_index = from_card - from_hr_bot_card_index;
-      assert(0 <= card_index && card_index < PosParPRT::CardsPerRegion,
+      assert(0 <= card_index && card_index < HeapRegion::CardsPerRegion,
              "Must be in range.");
       if (G1HRRSUseSparseTable &&
           _sparse_table.add_card(from_hrs_ind, card_index)) {
@@ -922,7 +919,7 @@ size_t OtherRegionsTable::occ_fine() const {
 }
 
 size_t OtherRegionsTable::occ_coarse() const {
-  return (_n_coarse_entries * PosParPRT::CardsPerRegion);
+  return (_n_coarse_entries * HeapRegion::CardsPerRegion);
 }
 
 size_t OtherRegionsTable::occ_sparse() const {
@@ -1049,7 +1046,8 @@ bool OtherRegionsTable::contains_reference_locked(OopOrNarrowOopStar from) const
       uintptr_t(hr->bottom()) >> CardTableModRefBS::card_shift;
     assert(from_card >= hr_bot_card_index, "Inv");
     CardIdx_t card_index = from_card - hr_bot_card_index;
-    assert(0 <= card_index && card_index < PosParPRT::CardsPerRegion, "Must be in range.");
+    assert(0 <= card_index && card_index < HeapRegion::CardsPerRegion,
+           "Must be in range.");
     return _sparse_table.contains_card(hr_ind, card_index);
   }
 
@@ -1176,7 +1174,7 @@ void HeapRegionRemSetIterator::initialize(const HeapRegionRemSet* hrrs) {
   _is = Sparse;
   // Set these values so that we increment to the first region.
   _coarse_cur_region_index = -1;
-  _coarse_cur_region_cur_card = (PosParPRT::CardsPerRegion-1);;
+  _coarse_cur_region_cur_card = (HeapRegion::CardsPerRegion-1);;
 
   _cur_region_cur_card = 0;
 
@@ -1195,7 +1193,7 @@ bool HeapRegionRemSetIterator::coarse_has_next(size_t& card_index) {
   // Go to the next card.
   _coarse_cur_region_cur_card++;
   // Was the last the last card in the current region?
-  if (_coarse_cur_region_cur_card == PosParPRT::CardsPerRegion) {
+  if (_coarse_cur_region_cur_card == HeapRegion::CardsPerRegion) {
     // Yes: find the next region.  This may leave _coarse_cur_region_index
     // Set to the last index, in which case there are no more coarse
     // regions.
@@ -1232,7 +1230,7 @@ bool HeapRegionRemSetIterator::fine_has_next(size_t& card_index) {
       _fine_cur_prt->_bm.get_next_one_offset(_cur_region_cur_card + 1);
   }
   while (!fine_has_next()) {
-    if (_cur_region_cur_card == PosParPRT::CardsPerRegion) {
+    if (_cur_region_cur_card == (size_t) HeapRegion::CardsPerRegion) {
       _cur_region_cur_card = 0;
       _fine_cur_prt = _fine_cur_prt->next();
     }
@@ -1255,7 +1253,7 @@ bool HeapRegionRemSetIterator::fine_has_next(size_t& card_index) {
 bool HeapRegionRemSetIterator::fine_has_next() {
   return
     _fine_cur_prt != NULL &&
-    _cur_region_cur_card < PosParPRT::CardsPerRegion;
+    _cur_region_cur_card < (size_t) HeapRegion::CardsPerRegion;
 }
 
 bool HeapRegionRemSetIterator::has_next(size_t& card_index) {

hotspot/src/share/vm/gc_implementation/g1/sparsePRT.cpp

@@ -347,7 +347,7 @@ CardIdx_t /* RSHashTable:: */ RSHashTableIter::find_first_card_in_list() {
 size_t /* RSHashTable:: */ RSHashTableIter::compute_card_ind(CardIdx_t ci) {
   return
     _heap_bot_card_ind
-    + (_rsht->entry(_bl_ind)->r_ind() * CardsPerRegion)
+    + (_rsht->entry(_bl_ind)->r_ind() * HeapRegion::CardsPerRegion)
     + ci;
 }
 

hotspot/src/share/vm/gc_implementation/g1/sparsePRT.hpp

@@ -172,10 +172,6 @@ class RSHashTableIter VALUE_OBJ_CLASS_SPEC {
   RSHashTable* _rsht;
   size_t _heap_bot_card_ind;
 
-  enum SomePrivateConstants {
-    CardsPerRegion = HeapRegion::GrainBytes >> CardTableModRefBS::card_shift
-  };
-
   // If the bucket list pointed to by _bl_ind contains a card, sets
   // _bl_ind to the index of that entry, and returns the card.
   // Otherwise, returns SparseEntry::NullEntry.

hotspot/src/share/vm/gc_implementation/includeDB_gc_g1

@@ -165,6 +165,7 @@ g1CollectedHeap.inline.hpp              g1CollectedHeap.hpp
 g1CollectedHeap.inline.hpp              heapRegionSeq.hpp
 g1CollectedHeap.inline.hpp		taskqueue.hpp
 
+g1CollectorPolicy.cpp			arguments.hpp
 g1CollectorPolicy.cpp			concurrentG1Refine.hpp
 g1CollectorPolicy.cpp			concurrentMark.hpp
 g1CollectorPolicy.cpp			concurrentMarkThread.inline.hpp