changeset 20223:b0c374311c4e

8035400: Move G1ParScanThreadState into its own files Summary: Extract the G1ParScanThreadState class from G1CollectedHeap.?pp into its own files. Reviewed-by: brutisso, mgerdin
author tschatzl
date Mon, 21 Jul 2014 09:41:04 +0200
parents 0abcece2ee27
children a2328cbebb23
files src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp src/share/vm/gc_implementation/g1/g1CollectedHeap.inline.hpp src/share/vm/gc_implementation/g1/g1OopClosures.cpp src/share/vm/gc_implementation/g1/g1OopClosures.inline.hpp src/share/vm/gc_implementation/g1/g1ParScanThreadState.cpp src/share/vm/gc_implementation/g1/g1ParScanThreadState.hpp src/share/vm/gc_implementation/g1/g1ParScanThreadState.inline.hpp src/share/vm/gc_implementation/g1/g1RemSet.inline.hpp
diffstat 9 files changed, 667 insertions(+), 562 deletions(-) [+]
line wrap: on
line diff
--- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp	Mon Jul 21 09:40:19 2014 +0200
+++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp	Mon Jul 21 09:41:04 2014 +0200
@@ -42,6 +42,7 @@
 #include "gc_implementation/g1/g1Log.hpp"
 #include "gc_implementation/g1/g1MarkSweep.hpp"
 #include "gc_implementation/g1/g1OopClosures.inline.hpp"
+#include "gc_implementation/g1/g1ParScanThreadState.inline.hpp"
 #include "gc_implementation/g1/g1RemSet.inline.hpp"
 #include "gc_implementation/g1/g1StringDedup.hpp"
 #include "gc_implementation/g1/g1YCTypes.hpp"
@@ -60,10 +61,8 @@
 #include "memory/referenceProcessor.hpp"
 #include "oops/oop.inline.hpp"
 #include "oops/oop.pcgc.inline.hpp"
-#include "runtime/prefetch.inline.hpp"
 #include "runtime/orderAccess.inline.hpp"
 #include "runtime/vmThread.hpp"
-#include "utilities/ticks.hpp"
 
 size_t G1CollectedHeap::_humongous_object_threshold_in_words = 0;
 
@@ -4568,126 +4567,6 @@
 G1ParGCAllocBuffer::G1ParGCAllocBuffer(size_t gclab_word_size) :
   ParGCAllocBuffer(gclab_word_size), _retired(true) { }
 
-G1ParScanThreadState::G1ParScanThreadState(G1CollectedHeap* g1h, uint queue_num, ReferenceProcessor* rp)
-  : _g1h(g1h),
-    _refs(g1h->task_queue(queue_num)),
-    _dcq(&g1h->dirty_card_queue_set()),
-    _ct_bs(g1h->g1_barrier_set()),
-    _g1_rem(g1h->g1_rem_set()),
-    _hash_seed(17), _queue_num(queue_num),
-    _term_attempts(0),
-    _surviving_alloc_buffer(g1h->desired_plab_sz(GCAllocForSurvived)),
-    _tenured_alloc_buffer(g1h->desired_plab_sz(GCAllocForTenured)),
-    _age_table(false), _scanner(g1h, this, rp),
-    _strong_roots_time(0), _term_time(0),
-    _alloc_buffer_waste(0), _undo_waste(0) {
-  // we allocate G1YoungSurvRateNumRegions plus one entries, since
-  // we "sacrifice" entry 0 to keep track of surviving bytes for
-  // non-young regions (where the age is -1)
-  // We also add a few elements at the beginning and at the end in
-  // an attempt to eliminate cache contention
-  uint real_length = 1 + _g1h->g1_policy()->young_cset_region_length();
-  uint array_length = PADDING_ELEM_NUM +
-                      real_length +
-                      PADDING_ELEM_NUM;
-  _surviving_young_words_base = NEW_C_HEAP_ARRAY(size_t, array_length, mtGC);
-  if (_surviving_young_words_base == NULL)
-    vm_exit_out_of_memory(array_length * sizeof(size_t), OOM_MALLOC_ERROR,
-                          "Not enough space for young surv histo.");
-  _surviving_young_words = _surviving_young_words_base + PADDING_ELEM_NUM;
-  memset(_surviving_young_words, 0, (size_t) real_length * sizeof(size_t));
-
-  _alloc_buffers[GCAllocForSurvived] = &_surviving_alloc_buffer;
-  _alloc_buffers[GCAllocForTenured]  = &_tenured_alloc_buffer;
-
-  _start = os::elapsedTime();
-}
-
-void
-G1ParScanThreadState::print_termination_stats_hdr(outputStream* const st)
-{
-  st->print_raw_cr("GC Termination Stats");
-  st->print_raw_cr("     elapsed  --strong roots-- -------termination-------"
-                   " ------waste (KiB)------");
-  st->print_raw_cr("thr     ms        ms      %        ms      %    attempts"
-                   "  total   alloc    undo");
-  st->print_raw_cr("--- --------- --------- ------ --------- ------ --------"
-                   " ------- ------- -------");
-}
-
-void
-G1ParScanThreadState::print_termination_stats(int i,
-                                              outputStream* const st) const
-{
-  const double elapsed_ms = elapsed_time() * 1000.0;
-  const double s_roots_ms = strong_roots_time() * 1000.0;
-  const double term_ms    = term_time() * 1000.0;
-  st->print_cr("%3d %9.2f %9.2f %6.2f "
-               "%9.2f %6.2f " SIZE_FORMAT_W(8) " "
-               SIZE_FORMAT_W(7) " " SIZE_FORMAT_W(7) " " SIZE_FORMAT_W(7),
-               i, elapsed_ms, s_roots_ms, s_roots_ms * 100 / elapsed_ms,
-               term_ms, term_ms * 100 / elapsed_ms, term_attempts(),
-               (alloc_buffer_waste() + undo_waste()) * HeapWordSize / K,
-               alloc_buffer_waste() * HeapWordSize / K,
-               undo_waste() * HeapWordSize / K);
-}
-
-#ifdef ASSERT
-bool G1ParScanThreadState::verify_ref(narrowOop* ref) const {
-  assert(ref != NULL, "invariant");
-  assert(UseCompressedOops, "sanity");
-  assert(!has_partial_array_mask(ref), err_msg("ref=" PTR_FORMAT, ref));
-  oop p = oopDesc::load_decode_heap_oop(ref);
-  assert(_g1h->is_in_g1_reserved(p),
-         err_msg("ref=" PTR_FORMAT " p=" PTR_FORMAT, ref, (void *)p));
-  return true;
-}
-
-bool G1ParScanThreadState::verify_ref(oop* ref) const {
-  assert(ref != NULL, "invariant");
-  if (has_partial_array_mask(ref)) {
-    // Must be in the collection set--it's already been copied.
-    oop p = clear_partial_array_mask(ref);
-    assert(_g1h->obj_in_cs(p),
-           err_msg("ref=" PTR_FORMAT " p=" PTR_FORMAT, ref, (void *)p));
-  } else {
-    oop p = oopDesc::load_decode_heap_oop(ref);
-    assert(_g1h->is_in_g1_reserved(p),
-           err_msg("ref=" PTR_FORMAT " p=" PTR_FORMAT, ref, (void *)p));
-  }
-  return true;
-}
-
-bool G1ParScanThreadState::verify_task(StarTask ref) const {
-  if (ref.is_narrow()) {
-    return verify_ref((narrowOop*) ref);
-  } else {
-    return verify_ref((oop*) ref);
-  }
-}
-#endif // ASSERT
-
-void G1ParScanThreadState::trim_queue() {
-  assert(_evac_failure_cl != NULL, "not set");
-
-  StarTask ref;
-  do {
-    // Drain the overflow stack first, so other threads can steal.
-    while (refs()->pop_overflow(ref)) {
-      deal_with_reference(ref);
-    }
-
-    while (refs()->pop_local(ref)) {
-      deal_with_reference(ref);
-    }
-  } while (!refs()->is_empty());
-}
-
-G1ParClosureSuper::G1ParClosureSuper(G1CollectedHeap* g1,
-                                     G1ParScanThreadState* par_scan_state) :
-  _g1(g1), _par_scan_state(par_scan_state),
-  _worker_id(par_scan_state->queue_num()) { }
-
 void G1ParCopyHelper::mark_object(oop obj) {
 #ifdef ASSERT
   HeapRegion* hr = _g1->heap_region_containing(obj);
@@ -4721,107 +4600,6 @@
   _cm->grayRoot(to_obj, (size_t) from_obj->size(), _worker_id);
 }
 
-oop G1ParScanThreadState::copy_to_survivor_space(oop const old) {
-  size_t word_sz = old->size();
-  HeapRegion* from_region = _g1h->heap_region_containing_raw(old);
-  // +1 to make the -1 indexes valid...
-  int       young_index = from_region->young_index_in_cset()+1;
-  assert( (from_region->is_young() && young_index >  0) ||
-         (!from_region->is_young() && young_index == 0), "invariant" );
-  G1CollectorPolicy* g1p = _g1h->g1_policy();
-  markOop m = old->mark();
-  int age = m->has_displaced_mark_helper() ? m->displaced_mark_helper()->age()
-                                           : m->age();
-  GCAllocPurpose alloc_purpose = g1p->evacuation_destination(from_region, age,
-                                                             word_sz);
-  HeapWord* obj_ptr = allocate(alloc_purpose, word_sz);
-#ifndef PRODUCT
-  // Should this evacuation fail?
-  if (_g1h->evacuation_should_fail()) {
-    if (obj_ptr != NULL) {
-      undo_allocation(alloc_purpose, obj_ptr, word_sz);
-      obj_ptr = NULL;
-    }
-  }
-#endif // !PRODUCT
-
-  if (obj_ptr == NULL) {
-    // This will either forward-to-self, or detect that someone else has
-    // installed a forwarding pointer.
-    return _g1h->handle_evacuation_failure_par(this, old);
-  }
-
-  oop obj = oop(obj_ptr);
-
-  // We're going to allocate linearly, so might as well prefetch ahead.
-  Prefetch::write(obj_ptr, PrefetchCopyIntervalInBytes);
-
-  oop forward_ptr = old->forward_to_atomic(obj);
-  if (forward_ptr == NULL) {
-    Copy::aligned_disjoint_words((HeapWord*) old, obj_ptr, word_sz);
-
-    // alloc_purpose is just a hint to allocate() above, recheck the type of region
-    // we actually allocated from and update alloc_purpose accordingly
-    HeapRegion* to_region = _g1h->heap_region_containing_raw(obj_ptr);
-    alloc_purpose = to_region->is_young() ? GCAllocForSurvived : GCAllocForTenured;
-
-    if (g1p->track_object_age(alloc_purpose)) {
-      // We could simply do obj->incr_age(). However, this causes a
-      // performance issue. obj->incr_age() will first check whether
-      // the object has a displaced mark by checking its mark word;
-      // getting the mark word from the new location of the object
-      // stalls. So, given that we already have the mark word and we
-      // are about to install it anyway, it's better to increase the
-      // age on the mark word, when the object does not have a
-      // displaced mark word. We're not expecting many objects to have
-      // a displaced marked word, so that case is not optimized
-      // further (it could be...) and we simply call obj->incr_age().
-
-      if (m->has_displaced_mark_helper()) {
-        // in this case, we have to install the mark word first,
-        // otherwise obj looks to be forwarded (the old mark word,
-        // which contains the forward pointer, was copied)
-        obj->set_mark(m);
-        obj->incr_age();
-      } else {
-        m = m->incr_age();
-        obj->set_mark(m);
-      }
-      age_table()->add(obj, word_sz);
-    } else {
-      obj->set_mark(m);
-    }
-
-    if (G1StringDedup::is_enabled()) {
-      G1StringDedup::enqueue_from_evacuation(from_region->is_young(),
-                                             to_region->is_young(),
-                                             queue_num(),
-                                             obj);
-    }
-
-    size_t* surv_young_words = surviving_young_words();
-    surv_young_words[young_index] += word_sz;
-
-    if (obj->is_objArray() && arrayOop(obj)->length() >= ParGCArrayScanChunk) {
-      // We keep track of the next start index in the length field of
-      // the to-space object. The actual length can be found in the
-      // length field of the from-space object.
-      arrayOop(obj)->set_length(0);
-      oop* old_p = set_partial_array_mask(old);
-      push_on_queue(old_p);
-    } else {
-      // No point in using the slower heap_region_containing() method,
-      // given that we know obj is in the heap.
-      _scanner.set_region(_g1h->heap_region_containing_raw(obj));
-      obj->oop_iterate_backwards(&_scanner);
-    }
-  } else {
-    undo_allocation(alloc_purpose, obj_ptr, word_sz);
-    obj = forward_ptr;
-  }
-  return obj;
-}
-
 template <class T>
 void G1ParCopyHelper::do_klass_barrier(T* p, oop new_obj) {
   if (_g1->heap_region_containing_raw(new_obj)->is_young()) {
--- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp	Mon Jul 21 09:40:19 2014 +0200
+++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp	Mon Jul 21 09:41:04 2014 +0200
@@ -31,7 +31,6 @@
 #include "gc_implementation/g1/g1BiasedArray.hpp"
 #include "gc_implementation/g1/g1HRPrinter.hpp"
 #include "gc_implementation/g1/g1MonitoringSupport.hpp"
-#include "gc_implementation/g1/g1RemSet.hpp"
 #include "gc_implementation/g1/g1SATBCardTableModRefBS.hpp"
 #include "gc_implementation/g1/g1YCTypes.hpp"
 #include "gc_implementation/g1/heapRegionSeq.hpp"
@@ -1709,256 +1708,4 @@
   }
 };
 
-class G1ParScanThreadState : public StackObj {
-protected:
-  G1CollectedHeap* _g1h;
-  RefToScanQueue*  _refs;
-  DirtyCardQueue   _dcq;
-  G1SATBCardTableModRefBS* _ct_bs;
-  G1RemSet* _g1_rem;
-
-  G1ParGCAllocBuffer  _surviving_alloc_buffer;
-  G1ParGCAllocBuffer  _tenured_alloc_buffer;
-  G1ParGCAllocBuffer* _alloc_buffers[GCAllocPurposeCount];
-  ageTable            _age_table;
-
-  G1ParScanClosure    _scanner;
-
-  size_t           _alloc_buffer_waste;
-  size_t           _undo_waste;
-
-  OopsInHeapRegionClosure*      _evac_failure_cl;
-
-  int  _hash_seed;
-  uint _queue_num;
-
-  size_t _term_attempts;
-
-  double _start;
-  double _start_strong_roots;
-  double _strong_roots_time;
-  double _start_term;
-  double _term_time;
-
-  // Map from young-age-index (0 == not young, 1 is youngest) to
-  // surviving words. base is what we get back from the malloc call
-  size_t* _surviving_young_words_base;
-  // this points into the array, as we use the first few entries for padding
-  size_t* _surviving_young_words;
-
-#define PADDING_ELEM_NUM (DEFAULT_CACHE_LINE_SIZE / sizeof(size_t))
-
-  void   add_to_alloc_buffer_waste(size_t waste) { _alloc_buffer_waste += waste; }
-
-  void   add_to_undo_waste(size_t waste)         { _undo_waste += waste; }
-
-  DirtyCardQueue& dirty_card_queue()             { return _dcq;  }
-  G1SATBCardTableModRefBS* ctbs()                { return _ct_bs; }
-
-  template <class T> inline void immediate_rs_update(HeapRegion* from, T* p, int tid);
-
-  template <class T> void deferred_rs_update(HeapRegion* from, T* p, int tid) {
-    // If the new value of the field points to the same region or
-    // is the to-space, we don't need to include it in the Rset updates.
-    if (!from->is_in_reserved(oopDesc::load_decode_heap_oop(p)) && !from->is_survivor()) {
-      size_t card_index = ctbs()->index_for(p);
-      // If the card hasn't been added to the buffer, do it.
-      if (ctbs()->mark_card_deferred(card_index)) {
-        dirty_card_queue().enqueue((jbyte*)ctbs()->byte_for_index(card_index));
-      }
-    }
-  }
-
-public:
-  G1ParScanThreadState(G1CollectedHeap* g1h, uint queue_num, ReferenceProcessor* rp);
-
-  ~G1ParScanThreadState() {
-    retire_alloc_buffers();
-    FREE_C_HEAP_ARRAY(size_t, _surviving_young_words_base, mtGC);
-  }
-
-  RefToScanQueue*   refs()            { return _refs;             }
-  ageTable*         age_table()       { return &_age_table;       }
-
-  G1ParGCAllocBuffer* alloc_buffer(GCAllocPurpose purpose) {
-    return _alloc_buffers[purpose];
-  }
-
-  size_t alloc_buffer_waste() const              { return _alloc_buffer_waste; }
-  size_t undo_waste() const                      { return _undo_waste; }
-
-#ifdef ASSERT
-  bool verify_ref(narrowOop* ref) const;
-  bool verify_ref(oop* ref) const;
-  bool verify_task(StarTask ref) const;
-#endif // ASSERT
-
-  template <class T> void push_on_queue(T* ref) {
-    assert(verify_ref(ref), "sanity");
-    refs()->push(ref);
-  }
-
-  template <class T> inline void update_rs(HeapRegion* from, T* p, int tid);
-
-  HeapWord* allocate_slow(GCAllocPurpose purpose, size_t word_sz) {
-    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 */);
-
-      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);
-
-      obj = alloc_buf->allocate(word_sz);
-      assert(obj != NULL, "buffer was definitely big enough...");
-    } else {
-      obj = _g1h->par_allocate_during_gc(purpose, word_sz);
-    }
-    return obj;
-  }
-
-  HeapWord* allocate(GCAllocPurpose purpose, size_t word_sz) {
-    HeapWord* obj = alloc_buffer(purpose)->allocate(word_sz);
-    if (obj != NULL) return obj;
-    return allocate_slow(purpose, word_sz);
-  }
-
-  void undo_allocation(GCAllocPurpose purpose, HeapWord* obj, size_t word_sz) {
-    if (alloc_buffer(purpose)->contains(obj)) {
-      assert(alloc_buffer(purpose)->contains(obj + word_sz - 1),
-             "should contain whole object");
-      alloc_buffer(purpose)->undo_allocation(obj, word_sz);
-    } else {
-      CollectedHeap::fill_with_object(obj, word_sz);
-      add_to_undo_waste(word_sz);
-    }
-  }
-
-  void set_evac_failure_closure(OopsInHeapRegionClosure* evac_failure_cl) {
-    _evac_failure_cl = evac_failure_cl;
-  }
-  OopsInHeapRegionClosure* evac_failure_closure() {
-    return _evac_failure_cl;
-  }
-
-  int* hash_seed() { return &_hash_seed; }
-  uint queue_num() { return _queue_num; }
-
-  size_t term_attempts() const  { return _term_attempts; }
-  void note_term_attempt() { _term_attempts++; }
-
-  void start_strong_roots() {
-    _start_strong_roots = os::elapsedTime();
-  }
-  void end_strong_roots() {
-    _strong_roots_time += (os::elapsedTime() - _start_strong_roots);
-  }
-  double strong_roots_time() const { return _strong_roots_time; }
-
-  void start_term_time() {
-    note_term_attempt();
-    _start_term = os::elapsedTime();
-  }
-  void end_term_time() {
-    _term_time += (os::elapsedTime() - _start_term);
-  }
-  double term_time() const { return _term_time; }
-
-  double elapsed_time() const {
-    return os::elapsedTime() - _start;
-  }
-
-  static void
-    print_termination_stats_hdr(outputStream* const st = gclog_or_tty);
-  void
-    print_termination_stats(int i, outputStream* const st = gclog_or_tty) const;
-
-  size_t* surviving_young_words() {
-    // We add on to hide entry 0 which accumulates surviving words for
-    // age -1 regions (i.e. non-young ones)
-    return _surviving_young_words;
-  }
-
-private:
-  void retire_alloc_buffers() {
-    for (int ap = 0; ap < GCAllocPurposeCount; ++ap) {
-      size_t waste = _alloc_buffers[ap]->words_remaining();
-      add_to_alloc_buffer_waste(waste);
-      _alloc_buffers[ap]->flush_stats_and_retire(_g1h->stats_for_purpose((GCAllocPurpose)ap),
-                                                 true /* end_of_gc */,
-                                                 false /* retain */);
-    }
-  }
-
-#define G1_PARTIAL_ARRAY_MASK 0x2
-
-  inline bool has_partial_array_mask(oop* ref) const {
-    return ((uintptr_t)ref & G1_PARTIAL_ARRAY_MASK) == G1_PARTIAL_ARRAY_MASK;
-  }
-
-  // We never encode partial array oops as narrowOop*, so return false immediately.
-  // This allows the compiler to create optimized code when popping references from
-  // the work queue.
-  inline bool has_partial_array_mask(narrowOop* ref) const {
-    assert(((uintptr_t)ref & G1_PARTIAL_ARRAY_MASK) != G1_PARTIAL_ARRAY_MASK, "Partial array oop reference encoded as narrowOop*");
-    return false;
-  }
-
-  // Only implement set_partial_array_mask() for regular oops, not for narrowOops.
-  // We always encode partial arrays as regular oop, to allow the
-  // specialization for has_partial_array_mask() for narrowOops above.
-  // This means that unintentional use of this method with narrowOops are caught
-  // by the compiler.
-  inline oop* set_partial_array_mask(oop obj) const {
-    assert(((uintptr_t)(void *)obj & G1_PARTIAL_ARRAY_MASK) == 0, "Information loss!");
-    return (oop*) ((uintptr_t)(void *)obj | G1_PARTIAL_ARRAY_MASK);
-  }
-
-  inline oop clear_partial_array_mask(oop* ref) const {
-    return cast_to_oop((intptr_t)ref & ~G1_PARTIAL_ARRAY_MASK);
-  }
-
-  inline void do_oop_partial_array(oop* p);
-
-  // This method is applied to the fields of the objects that have just been copied.
-  template <class T> void do_oop_evac(T* p, HeapRegion* from) {
-    assert(!oopDesc::is_null(oopDesc::load_decode_heap_oop(p)),
-           "Reference should not be NULL here as such are never pushed to the task queue.");
-    oop obj = oopDesc::load_decode_heap_oop_not_null(p);
-
-    // Although we never intentionally push references outside of the collection
-    // set, due to (benign) races in the claim mechanism during RSet scanning more
-    // than one thread might claim the same card. So the same card may be
-    // processed multiple times. So redo this check.
-    if (_g1h->in_cset_fast_test(obj)) {
-      oop forwardee;
-      if (obj->is_forwarded()) {
-        forwardee = obj->forwardee();
-      } else {
-        forwardee = copy_to_survivor_space(obj);
-      }
-      assert(forwardee != NULL, "forwardee should not be NULL");
-      oopDesc::encode_store_heap_oop(p, forwardee);
-    }
-
-    assert(obj != NULL, "Must be");
-    update_rs(from, p, queue_num());
-  }
-public:
-
-  oop copy_to_survivor_space(oop const obj);
-
-  template <class T> inline void deal_with_reference(T* ref_to_scan);
-
-  inline void deal_with_reference(StarTask ref);
-
-public:
-  void trim_queue();
-};
-
 #endif // SHARE_VM_GC_IMPLEMENTATION_G1_G1COLLECTEDHEAP_HPP
--- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.inline.hpp	Mon Jul 21 09:40:19 2014 +0200
+++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.inline.hpp	Mon Jul 21 09:41:04 2014 +0200
@@ -29,7 +29,6 @@
 #include "gc_implementation/g1/g1CollectedHeap.hpp"
 #include "gc_implementation/g1/g1AllocRegion.inline.hpp"
 #include "gc_implementation/g1/g1CollectorPolicy.hpp"
-#include "gc_implementation/g1/g1RemSet.inline.hpp"
 #include "gc_implementation/g1/g1SATBCardTableModRefBS.hpp"
 #include "gc_implementation/g1/heapRegionSet.inline.hpp"
 #include "gc_implementation/g1/heapRegionSeq.inline.hpp"
@@ -291,89 +290,4 @@
   else return is_obj_ill(obj, hr);
 }
 
-template <class T> inline void G1ParScanThreadState::immediate_rs_update(HeapRegion* from, T* p, int tid) {
-  if (!from->is_survivor()) {
-    _g1_rem->par_write_ref(from, p, tid);
-  }
-}
-
-template <class T> void G1ParScanThreadState::update_rs(HeapRegion* from, T* p, int tid) {
-  if (G1DeferredRSUpdate) {
-    deferred_rs_update(from, p, tid);
-  } else {
-    immediate_rs_update(from, p, tid);
-  }
-}
-
-
-inline void G1ParScanThreadState::do_oop_partial_array(oop* p) {
-  assert(has_partial_array_mask(p), "invariant");
-  oop from_obj = clear_partial_array_mask(p);
-
-  assert(Universe::heap()->is_in_reserved(from_obj), "must be in heap.");
-  assert(from_obj->is_objArray(), "must be obj array");
-  objArrayOop from_obj_array = objArrayOop(from_obj);
-  // The from-space object contains the real length.
-  int length                 = from_obj_array->length();
-
-  assert(from_obj->is_forwarded(), "must be forwarded");
-  oop to_obj                 = from_obj->forwardee();
-  assert(from_obj != to_obj, "should not be chunking self-forwarded objects");
-  objArrayOop to_obj_array   = objArrayOop(to_obj);
-  // We keep track of the next start index in the length field of the
-  // to-space object.
-  int next_index             = to_obj_array->length();
-  assert(0 <= next_index && next_index < length,
-         err_msg("invariant, next index: %d, length: %d", next_index, length));
-
-  int start                  = next_index;
-  int end                    = length;
-  int remainder              = end - start;
-  // We'll try not to push a range that's smaller than ParGCArrayScanChunk.
-  if (remainder > 2 * ParGCArrayScanChunk) {
-    end = start + ParGCArrayScanChunk;
-    to_obj_array->set_length(end);
-    // Push the remainder before we process the range in case another
-    // worker has run out of things to do and can steal it.
-    oop* from_obj_p = set_partial_array_mask(from_obj);
-    push_on_queue(from_obj_p);
-  } else {
-    assert(length == end, "sanity");
-    // We'll process the final range for this object. Restore the length
-    // so that the heap remains parsable in case of evacuation failure.
-    to_obj_array->set_length(end);
-  }
-  _scanner.set_region(_g1h->heap_region_containing_raw(to_obj));
-  // Process indexes [start,end). It will also process the header
-  // along with the first chunk (i.e., the chunk with start == 0).
-  // Note that at this point the length field of to_obj_array is not
-  // correct given that we are using it to keep track of the next
-  // start index. oop_iterate_range() (thankfully!) ignores the length
-  // field and only relies on the start / end parameters.  It does
-  // however return the size of the object which will be incorrect. So
-  // we have to ignore it even if we wanted to use it.
-  to_obj_array->oop_iterate_range(&_scanner, start, end);
-}
-
-template <class T> inline void G1ParScanThreadState::deal_with_reference(T* ref_to_scan) {
-  if (!has_partial_array_mask(ref_to_scan)) {
-    // Note: we can use "raw" versions of "region_containing" because
-    // "obj_to_scan" is definitely in the heap, and is not in a
-    // humongous region.
-    HeapRegion* r = _g1h->heap_region_containing_raw(ref_to_scan);
-    do_oop_evac(ref_to_scan, r);
-  } else {
-    do_oop_partial_array((oop*)ref_to_scan);
-  }
-}
-
-inline void G1ParScanThreadState::deal_with_reference(StarTask ref) {
-  assert(verify_task(ref), "sanity");
-  if (ref.is_narrow()) {
-    deal_with_reference((narrowOop*)ref);
-  } else {
-    deal_with_reference((oop*)ref);
-  }
-}
-
 #endif // SHARE_VM_GC_IMPLEMENTATION_G1_G1COLLECTEDHEAP_INLINE_HPP
--- a/src/share/vm/gc_implementation/g1/g1OopClosures.cpp	Mon Jul 21 09:40:19 2014 +0200
+++ b/src/share/vm/gc_implementation/g1/g1OopClosures.cpp	Mon Jul 21 09:41:04 2014 +0200
@@ -29,3 +29,7 @@
 G1ParCopyHelper::G1ParCopyHelper(G1CollectedHeap* g1,  G1ParScanThreadState* par_scan_state) :
   G1ParClosureSuper(g1, par_scan_state), _scanned_klass(NULL),
   _cm(_g1->concurrent_mark()) {}
+
+G1ParClosureSuper::G1ParClosureSuper(G1CollectedHeap* g1, G1ParScanThreadState* par_scan_state) :
+  _g1(g1), _par_scan_state(par_scan_state),
+  _worker_id(par_scan_state->queue_num()) { }
--- a/src/share/vm/gc_implementation/g1/g1OopClosures.inline.hpp	Mon Jul 21 09:40:19 2014 +0200
+++ b/src/share/vm/gc_implementation/g1/g1OopClosures.inline.hpp	Mon Jul 21 09:41:04 2014 +0200
@@ -28,6 +28,7 @@
 #include "gc_implementation/g1/concurrentMark.inline.hpp"
 #include "gc_implementation/g1/g1CollectedHeap.hpp"
 #include "gc_implementation/g1/g1OopClosures.hpp"
+#include "gc_implementation/g1/g1ParScanThreadState.inline.hpp"
 #include "gc_implementation/g1/g1RemSet.hpp"
 #include "gc_implementation/g1/g1RemSet.inline.hpp"
 #include "gc_implementation/g1/heapRegionRemSet.hpp"
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/share/vm/gc_implementation/g1/g1ParScanThreadState.cpp	Mon Jul 21 09:41:04 2014 +0200
@@ -0,0 +1,251 @@
+/*
+ * Copyright (c) 2014, 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/g1CollectedHeap.inline.hpp"
+#include "gc_implementation/g1/g1OopClosures.inline.hpp"
+#include "gc_implementation/g1/g1ParScanThreadState.inline.hpp"
+#include "oops/oop.inline.hpp"
+#include "oops/oop.pcgc.inline.hpp"
+#include "runtime/prefetch.inline.hpp"
+
+#ifdef _MSC_VER // the use of 'this' below gets a warning, make it go away
+#pragma warning( disable:4355 ) // 'this' : used in base member initializer list
+#endif // _MSC_VER
+
+G1ParScanThreadState::G1ParScanThreadState(G1CollectedHeap* g1h, uint queue_num, ReferenceProcessor* rp)
+  : _g1h(g1h),
+    _refs(g1h->task_queue(queue_num)),
+    _dcq(&g1h->dirty_card_queue_set()),
+    _ct_bs(g1h->g1_barrier_set()),
+    _g1_rem(g1h->g1_rem_set()),
+    _hash_seed(17), _queue_num(queue_num),
+    _term_attempts(0),
+    _surviving_alloc_buffer(g1h->desired_plab_sz(GCAllocForSurvived)),
+    _tenured_alloc_buffer(g1h->desired_plab_sz(GCAllocForTenured)),
+    _age_table(false), _scanner(g1h, this, rp),
+    _strong_roots_time(0), _term_time(0),
+    _alloc_buffer_waste(0), _undo_waste(0) {
+  // we allocate G1YoungSurvRateNumRegions plus one entries, since
+  // we "sacrifice" entry 0 to keep track of surviving bytes for
+  // non-young regions (where the age is -1)
+  // We also add a few elements at the beginning and at the end in
+  // an attempt to eliminate cache contention
+  uint real_length = 1 + _g1h->g1_policy()->young_cset_region_length();
+  uint array_length = PADDING_ELEM_NUM +
+                      real_length +
+                      PADDING_ELEM_NUM;
+  _surviving_young_words_base = NEW_C_HEAP_ARRAY(size_t, array_length, mtGC);
+  if (_surviving_young_words_base == NULL)
+    vm_exit_out_of_memory(array_length * sizeof(size_t), OOM_MALLOC_ERROR,
+                          "Not enough space for young surv histo.");
+  _surviving_young_words = _surviving_young_words_base + PADDING_ELEM_NUM;
+  memset(_surviving_young_words, 0, (size_t) real_length * sizeof(size_t));
+
+  _alloc_buffers[GCAllocForSurvived] = &_surviving_alloc_buffer;
+  _alloc_buffers[GCAllocForTenured]  = &_tenured_alloc_buffer;
+
+  _start = os::elapsedTime();
+}
+
+void
+G1ParScanThreadState::print_termination_stats_hdr(outputStream* const st)
+{
+  st->print_raw_cr("GC Termination Stats");
+  st->print_raw_cr("     elapsed  --strong roots-- -------termination-------"
+                   " ------waste (KiB)------");
+  st->print_raw_cr("thr     ms        ms      %        ms      %    attempts"
+                   "  total   alloc    undo");
+  st->print_raw_cr("--- --------- --------- ------ --------- ------ --------"
+                   " ------- ------- -------");
+}
+
+void
+G1ParScanThreadState::print_termination_stats(int i,
+                                              outputStream* const st) const
+{
+  const double elapsed_ms = elapsed_time() * 1000.0;
+  const double s_roots_ms = strong_roots_time() * 1000.0;
+  const double term_ms    = term_time() * 1000.0;
+  st->print_cr("%3d %9.2f %9.2f %6.2f "
+               "%9.2f %6.2f " SIZE_FORMAT_W(8) " "
+               SIZE_FORMAT_W(7) " " SIZE_FORMAT_W(7) " " SIZE_FORMAT_W(7),
+               i, elapsed_ms, s_roots_ms, s_roots_ms * 100 / elapsed_ms,
+               term_ms, term_ms * 100 / elapsed_ms, term_attempts(),
+               (alloc_buffer_waste() + undo_waste()) * HeapWordSize / K,
+               alloc_buffer_waste() * HeapWordSize / K,
+               undo_waste() * HeapWordSize / K);
+}
+
+#ifdef ASSERT
+bool G1ParScanThreadState::verify_ref(narrowOop* ref) const {
+  assert(ref != NULL, "invariant");
+  assert(UseCompressedOops, "sanity");
+  assert(!has_partial_array_mask(ref), err_msg("ref=" PTR_FORMAT, p2i(ref)));
+  oop p = oopDesc::load_decode_heap_oop(ref);
+  assert(_g1h->is_in_g1_reserved(p),
+         err_msg("ref=" PTR_FORMAT " p=" PTR_FORMAT, p2i(ref), p2i(p)));
+  return true;
+}
+
+bool G1ParScanThreadState::verify_ref(oop* ref) const {
+  assert(ref != NULL, "invariant");
+  if (has_partial_array_mask(ref)) {
+    // Must be in the collection set--it's already been copied.
+    oop p = clear_partial_array_mask(ref);
+    assert(_g1h->obj_in_cs(p),
+           err_msg("ref=" PTR_FORMAT " p=" PTR_FORMAT, p2i(ref), p2i(p)));
+  } else {
+    oop p = oopDesc::load_decode_heap_oop(ref);
+    assert(_g1h->is_in_g1_reserved(p),
+           err_msg("ref=" PTR_FORMAT " p=" PTR_FORMAT, p2i(ref), p2i(p)));
+  }
+  return true;
+}
+
+bool G1ParScanThreadState::verify_task(StarTask ref) const {
+  if (ref.is_narrow()) {
+    return verify_ref((narrowOop*) ref);
+  } else {
+    return verify_ref((oop*) ref);
+  }
+}
+#endif // ASSERT
+
+void G1ParScanThreadState::trim_queue() {
+  assert(_evac_failure_cl != NULL, "not set");
+
+  StarTask ref;
+  do {
+    // Drain the overflow stack first, so other threads can steal.
+    while (refs()->pop_overflow(ref)) {
+      deal_with_reference(ref);
+    }
+
+    while (refs()->pop_local(ref)) {
+      deal_with_reference(ref);
+    }
+  } while (!refs()->is_empty());
+}
+
+oop G1ParScanThreadState::copy_to_survivor_space(oop const old) {
+  size_t word_sz = old->size();
+  HeapRegion* from_region = _g1h->heap_region_containing_raw(old);
+  // +1 to make the -1 indexes valid...
+  int       young_index = from_region->young_index_in_cset()+1;
+  assert( (from_region->is_young() && young_index >  0) ||
+         (!from_region->is_young() && young_index == 0), "invariant" );
+  G1CollectorPolicy* g1p = _g1h->g1_policy();
+  markOop m = old->mark();
+  int age = m->has_displaced_mark_helper() ? m->displaced_mark_helper()->age()
+                                           : m->age();
+  GCAllocPurpose alloc_purpose = g1p->evacuation_destination(from_region, age,
+                                                             word_sz);
+  HeapWord* obj_ptr = allocate(alloc_purpose, word_sz);
+#ifndef PRODUCT
+  // Should this evacuation fail?
+  if (_g1h->evacuation_should_fail()) {
+    if (obj_ptr != NULL) {
+      undo_allocation(alloc_purpose, obj_ptr, word_sz);
+      obj_ptr = NULL;
+    }
+  }
+#endif // !PRODUCT
+
+  if (obj_ptr == NULL) {
+    // This will either forward-to-self, or detect that someone else has
+    // installed a forwarding pointer.
+    return _g1h->handle_evacuation_failure_par(this, old);
+  }
+
+  oop obj = oop(obj_ptr);
+
+  // We're going to allocate linearly, so might as well prefetch ahead.
+  Prefetch::write(obj_ptr, PrefetchCopyIntervalInBytes);
+
+  oop forward_ptr = old->forward_to_atomic(obj);
+  if (forward_ptr == NULL) {
+    Copy::aligned_disjoint_words((HeapWord*) old, obj_ptr, word_sz);
+
+    // alloc_purpose is just a hint to allocate() above, recheck the type of region
+    // we actually allocated from and update alloc_purpose accordingly
+    HeapRegion* to_region = _g1h->heap_region_containing_raw(obj_ptr);
+    alloc_purpose = to_region->is_young() ? GCAllocForSurvived : GCAllocForTenured;
+
+    if (g1p->track_object_age(alloc_purpose)) {
+      // We could simply do obj->incr_age(). However, this causes a
+      // performance issue. obj->incr_age() will first check whether
+      // the object has a displaced mark by checking its mark word;
+      // getting the mark word from the new location of the object
+      // stalls. So, given that we already have the mark word and we
+      // are about to install it anyway, it's better to increase the
+      // age on the mark word, when the object does not have a
+      // displaced mark word. We're not expecting many objects to have
+      // a displaced marked word, so that case is not optimized
+      // further (it could be...) and we simply call obj->incr_age().
+
+      if (m->has_displaced_mark_helper()) {
+        // in this case, we have to install the mark word first,
+        // otherwise obj looks to be forwarded (the old mark word,
+        // which contains the forward pointer, was copied)
+        obj->set_mark(m);
+        obj->incr_age();
+      } else {
+        m = m->incr_age();
+        obj->set_mark(m);
+      }
+      age_table()->add(obj, word_sz);
+    } else {
+      obj->set_mark(m);
+    }
+
+    if (G1StringDedup::is_enabled()) {
+      G1StringDedup::enqueue_from_evacuation(from_region->is_young(),
+                                             to_region->is_young(),
+                                             queue_num(),
+                                             obj);
+    }
+
+    size_t* surv_young_words = surviving_young_words();
+    surv_young_words[young_index] += word_sz;
+
+    if (obj->is_objArray() && arrayOop(obj)->length() >= ParGCArrayScanChunk) {
+      // We keep track of the next start index in the length field of
+      // the to-space object. The actual length can be found in the
+      // length field of the from-space object.
+      arrayOop(obj)->set_length(0);
+      oop* old_p = set_partial_array_mask(old);
+      push_on_queue(old_p);
+    } else {
+      // No point in using the slower heap_region_containing() method,
+      // given that we know obj is in the heap.
+      _scanner.set_region(_g1h->heap_region_containing_raw(obj));
+      obj->oop_iterate_backwards(&_scanner);
+    }
+  } else {
+    undo_allocation(alloc_purpose, obj_ptr, word_sz);
+    obj = forward_ptr;
+  }
+  return obj;
+}
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/share/vm/gc_implementation/g1/g1ParScanThreadState.hpp	Mon Jul 21 09:41:04 2014 +0200
@@ -0,0 +1,292 @@
+/*
+ * Copyright (c) 2014, 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_G1PARSCANTHREADSTATE_HPP
+#define SHARE_VM_GC_IMPLEMENTATION_G1_G1PARSCANTHREADSTATE_HPP
+
+#include "gc_implementation/g1/dirtyCardQueue.hpp"
+#include "gc_implementation/g1/g1SATBCardTableModRefBS.hpp"
+#include "gc_implementation/g1/g1CollectedHeap.hpp"
+#include "gc_implementation/g1/g1CollectorPolicy.hpp"
+#include "gc_implementation/g1/g1OopClosures.hpp"
+#include "gc_implementation/g1/g1RemSet.hpp"
+#include "gc_implementation/shared/ageTable.hpp"
+#include "memory/allocation.hpp"
+#include "oops/oop.hpp"
+
+class HeapRegion;
+class outputStream;
+
+class G1ParScanThreadState : public StackObj {
+protected:
+  G1CollectedHeap* _g1h;
+  RefToScanQueue*  _refs;
+  DirtyCardQueue   _dcq;
+  G1SATBCardTableModRefBS* _ct_bs;
+  G1RemSet* _g1_rem;
+
+  G1ParGCAllocBuffer  _surviving_alloc_buffer;
+  G1ParGCAllocBuffer  _tenured_alloc_buffer;
+  G1ParGCAllocBuffer* _alloc_buffers[GCAllocPurposeCount];
+  ageTable            _age_table;
+
+  G1ParScanClosure    _scanner;
+
+  size_t           _alloc_buffer_waste;
+  size_t           _undo_waste;
+
+  OopsInHeapRegionClosure*      _evac_failure_cl;
+
+  int  _hash_seed;
+  uint _queue_num;
+
+  size_t _term_attempts;
+
+  double _start;
+  double _start_strong_roots;
+  double _strong_roots_time;
+  double _start_term;
+  double _term_time;
+
+  // Map from young-age-index (0 == not young, 1 is youngest) to
+  // surviving words. base is what we get back from the malloc call
+  size_t* _surviving_young_words_base;
+  // this points into the array, as we use the first few entries for padding
+  size_t* _surviving_young_words;
+
+#define PADDING_ELEM_NUM (DEFAULT_CACHE_LINE_SIZE / sizeof(size_t))
+
+  void   add_to_alloc_buffer_waste(size_t waste) { _alloc_buffer_waste += waste; }
+
+  void   add_to_undo_waste(size_t waste)         { _undo_waste += waste; }
+
+  DirtyCardQueue& dirty_card_queue()             { return _dcq;  }
+  G1SATBCardTableModRefBS* ctbs()                { return _ct_bs; }
+
+  template <class T> inline void immediate_rs_update(HeapRegion* from, T* p, int tid);
+
+  template <class T> void deferred_rs_update(HeapRegion* from, T* p, int tid) {
+    // If the new value of the field points to the same region or
+    // is the to-space, we don't need to include it in the Rset updates.
+    if (!from->is_in_reserved(oopDesc::load_decode_heap_oop(p)) && !from->is_survivor()) {
+      size_t card_index = ctbs()->index_for(p);
+      // If the card hasn't been added to the buffer, do it.
+      if (ctbs()->mark_card_deferred(card_index)) {
+        dirty_card_queue().enqueue((jbyte*)ctbs()->byte_for_index(card_index));
+      }
+    }
+  }
+
+public:
+  G1ParScanThreadState(G1CollectedHeap* g1h, uint queue_num, ReferenceProcessor* rp);
+  ~G1ParScanThreadState() {
+    retire_alloc_buffers();
+    FREE_C_HEAP_ARRAY(size_t, _surviving_young_words_base, mtGC);
+  }
+
+  RefToScanQueue*   refs()            { return _refs;             }
+  ageTable*         age_table()       { return &_age_table;       }
+
+  G1ParGCAllocBuffer* alloc_buffer(GCAllocPurpose purpose) {
+    return _alloc_buffers[purpose];
+  }
+
+  size_t alloc_buffer_waste() const              { return _alloc_buffer_waste; }
+  size_t undo_waste() const                      { return _undo_waste; }
+
+#ifdef ASSERT
+  bool verify_ref(narrowOop* ref) const;
+  bool verify_ref(oop* ref) const;
+  bool verify_task(StarTask ref) const;
+#endif // ASSERT
+
+  template <class T> void push_on_queue(T* ref) {
+    assert(verify_ref(ref), "sanity");
+    refs()->push(ref);
+  }
+
+  template <class T> inline void update_rs(HeapRegion* from, T* p, int tid);
+
+  HeapWord* allocate_slow(GCAllocPurpose purpose, size_t word_sz) {
+    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 */);
+
+      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);
+
+      obj = alloc_buf->allocate(word_sz);
+      assert(obj != NULL, "buffer was definitely big enough...");
+    } else {
+      obj = _g1h->par_allocate_during_gc(purpose, word_sz);
+    }
+    return obj;
+  }
+
+  HeapWord* allocate(GCAllocPurpose purpose, size_t word_sz) {
+    HeapWord* obj = alloc_buffer(purpose)->allocate(word_sz);
+    if (obj != NULL) return obj;
+    return allocate_slow(purpose, word_sz);
+  }
+
+  void undo_allocation(GCAllocPurpose purpose, HeapWord* obj, size_t word_sz) {
+    if (alloc_buffer(purpose)->contains(obj)) {
+      assert(alloc_buffer(purpose)->contains(obj + word_sz - 1),
+             "should contain whole object");
+      alloc_buffer(purpose)->undo_allocation(obj, word_sz);
+    } else {
+      CollectedHeap::fill_with_object(obj, word_sz);
+      add_to_undo_waste(word_sz);
+    }
+  }
+
+  void set_evac_failure_closure(OopsInHeapRegionClosure* evac_failure_cl) {
+    _evac_failure_cl = evac_failure_cl;
+  }
+  OopsInHeapRegionClosure* evac_failure_closure() {
+    return _evac_failure_cl;
+  }
+
+  int* hash_seed() { return &_hash_seed; }
+  uint queue_num() { return _queue_num; }
+
+  size_t term_attempts() const  { return _term_attempts; }
+  void note_term_attempt() { _term_attempts++; }
+
+  void start_strong_roots() {
+    _start_strong_roots = os::elapsedTime();
+  }
+  void end_strong_roots() {
+    _strong_roots_time += (os::elapsedTime() - _start_strong_roots);
+  }
+  double strong_roots_time() const { return _strong_roots_time; }
+
+  void start_term_time() {
+    note_term_attempt();
+    _start_term = os::elapsedTime();
+  }
+  void end_term_time() {
+    _term_time += (os::elapsedTime() - _start_term);
+  }
+  double term_time() const { return _term_time; }
+
+  double elapsed_time() const {
+    return os::elapsedTime() - _start;
+  }
+
+  static void
+    print_termination_stats_hdr(outputStream* const st = gclog_or_tty);
+  void
+    print_termination_stats(int i, outputStream* const st = gclog_or_tty) const;
+
+  size_t* surviving_young_words() {
+    // We add on to hide entry 0 which accumulates surviving words for
+    // age -1 regions (i.e. non-young ones)
+    return _surviving_young_words;
+  }
+
+ private:
+  void retire_alloc_buffers() {
+    for (int ap = 0; ap < GCAllocPurposeCount; ++ap) {
+      size_t waste = _alloc_buffers[ap]->words_remaining();
+      add_to_alloc_buffer_waste(waste);
+      _alloc_buffers[ap]->flush_stats_and_retire(_g1h->stats_for_purpose((GCAllocPurpose)ap),
+                                                 true /* end_of_gc */,
+                                                 false /* retain */);
+    }
+  }
+
+  #define G1_PARTIAL_ARRAY_MASK 0x2
+
+  inline bool has_partial_array_mask(oop* ref) const {
+    return ((uintptr_t)ref & G1_PARTIAL_ARRAY_MASK) == G1_PARTIAL_ARRAY_MASK;
+  }
+
+  // We never encode partial array oops as narrowOop*, so return false immediately.
+  // This allows the compiler to create optimized code when popping references from
+  // the work queue.
+  inline bool has_partial_array_mask(narrowOop* ref) const {
+    assert(((uintptr_t)ref & G1_PARTIAL_ARRAY_MASK) != G1_PARTIAL_ARRAY_MASK, "Partial array oop reference encoded as narrowOop*");
+    return false;
+  }
+
+  // Only implement set_partial_array_mask() for regular oops, not for narrowOops.
+  // We always encode partial arrays as regular oop, to allow the
+  // specialization for has_partial_array_mask() for narrowOops above.
+  // This means that unintentional use of this method with narrowOops are caught
+  // by the compiler.
+  inline oop* set_partial_array_mask(oop obj) const {
+    assert(((uintptr_t)(void *)obj & G1_PARTIAL_ARRAY_MASK) == 0, "Information loss!");
+    return (oop*) ((uintptr_t)(void *)obj | G1_PARTIAL_ARRAY_MASK);
+  }
+
+  inline oop clear_partial_array_mask(oop* ref) const {
+    return cast_to_oop((intptr_t)ref & ~G1_PARTIAL_ARRAY_MASK);
+  }
+
+  inline void do_oop_partial_array(oop* p);
+
+  // This method is applied to the fields of the objects that have just been copied.
+  template <class T> void do_oop_evac(T* p, HeapRegion* from) {
+    assert(!oopDesc::is_null(oopDesc::load_decode_heap_oop(p)),
+           "Reference should not be NULL here as such are never pushed to the task queue.");
+    oop obj = oopDesc::load_decode_heap_oop_not_null(p);
+
+    // Although we never intentionally push references outside of the collection
+    // set, due to (benign) races in the claim mechanism during RSet scanning more
+    // than one thread might claim the same card. So the same card may be
+    // processed multiple times. So redo this check.
+    if (_g1h->in_cset_fast_test(obj)) {
+      oop forwardee;
+      if (obj->is_forwarded()) {
+        forwardee = obj->forwardee();
+      } else {
+        forwardee = copy_to_survivor_space(obj);
+      }
+      assert(forwardee != NULL, "forwardee should not be NULL");
+      oopDesc::encode_store_heap_oop(p, forwardee);
+    }
+
+    assert(obj != NULL, "Must be");
+    update_rs(from, p, queue_num());
+  }
+public:
+
+  oop copy_to_survivor_space(oop const obj);
+
+  template <class T> inline void deal_with_reference(T* ref_to_scan);
+
+  inline void deal_with_reference(StarTask ref);
+
+public:
+  void trim_queue();
+};
+
+#endif // SHARE_VM_GC_IMPLEMENTATION_G1_G1PARSCANTHREADSTATE_HPP
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/share/vm/gc_implementation/g1/g1ParScanThreadState.inline.hpp	Mon Jul 21 09:41:04 2014 +0200
@@ -0,0 +1,117 @@
+/*
+ * Copyright (c) 2014, 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_G1PARSCANTHREADSTATE_INLINE_HPP
+#define SHARE_VM_GC_IMPLEMENTATION_G1_G1PARSCANTHREADSTATE_INLINE_HPP
+
+#include "gc_implementation/g1/g1ParScanThreadState.hpp"
+#include "gc_implementation/g1/g1RemSet.inline.hpp"
+#include "oops/oop.inline.hpp"
+
+template <class T> inline void G1ParScanThreadState::immediate_rs_update(HeapRegion* from, T* p, int tid) {
+  if (!from->is_survivor()) {
+    _g1_rem->par_write_ref(from, p, tid);
+  }
+}
+
+template <class T> void G1ParScanThreadState::update_rs(HeapRegion* from, T* p, int tid) {
+  if (G1DeferredRSUpdate) {
+    deferred_rs_update(from, p, tid);
+  } else {
+    immediate_rs_update(from, p, tid);
+  }
+}
+
+inline void G1ParScanThreadState::do_oop_partial_array(oop* p) {
+  assert(has_partial_array_mask(p), "invariant");
+  oop from_obj = clear_partial_array_mask(p);
+
+  assert(Universe::heap()->is_in_reserved(from_obj), "must be in heap.");
+  assert(from_obj->is_objArray(), "must be obj array");
+  objArrayOop from_obj_array = objArrayOop(from_obj);
+  // The from-space object contains the real length.
+  int length                 = from_obj_array->length();
+
+  assert(from_obj->is_forwarded(), "must be forwarded");
+  oop to_obj                 = from_obj->forwardee();
+  assert(from_obj != to_obj, "should not be chunking self-forwarded objects");
+  objArrayOop to_obj_array   = objArrayOop(to_obj);
+  // We keep track of the next start index in the length field of the
+  // to-space object.
+  int next_index             = to_obj_array->length();
+  assert(0 <= next_index && next_index < length,
+         err_msg("invariant, next index: %d, length: %d", next_index, length));
+
+  int start                  = next_index;
+  int end                    = length;
+  int remainder              = end - start;
+  // We'll try not to push a range that's smaller than ParGCArrayScanChunk.
+  if (remainder > 2 * ParGCArrayScanChunk) {
+    end = start + ParGCArrayScanChunk;
+    to_obj_array->set_length(end);
+    // Push the remainder before we process the range in case another
+    // worker has run out of things to do and can steal it.
+    oop* from_obj_p = set_partial_array_mask(from_obj);
+    push_on_queue(from_obj_p);
+  } else {
+    assert(length == end, "sanity");
+    // We'll process the final range for this object. Restore the length
+    // so that the heap remains parsable in case of evacuation failure.
+    to_obj_array->set_length(end);
+  }
+  _scanner.set_region(_g1h->heap_region_containing_raw(to_obj));
+  // Process indexes [start,end). It will also process the header
+  // along with the first chunk (i.e., the chunk with start == 0).
+  // Note that at this point the length field of to_obj_array is not
+  // correct given that we are using it to keep track of the next
+  // start index. oop_iterate_range() (thankfully!) ignores the length
+  // field and only relies on the start / end parameters.  It does
+  // however return the size of the object which will be incorrect. So
+  // we have to ignore it even if we wanted to use it.
+  to_obj_array->oop_iterate_range(&_scanner, start, end);
+}
+
+template <class T> inline void G1ParScanThreadState::deal_with_reference(T* ref_to_scan) {
+  if (!has_partial_array_mask(ref_to_scan)) {
+    // Note: we can use "raw" versions of "region_containing" because
+    // "obj_to_scan" is definitely in the heap, and is not in a
+    // humongous region.
+    HeapRegion* r = _g1h->heap_region_containing_raw(ref_to_scan);
+    do_oop_evac(ref_to_scan, r);
+  } else {
+    do_oop_partial_array((oop*)ref_to_scan);
+  }
+}
+
+inline void G1ParScanThreadState::deal_with_reference(StarTask ref) {
+  assert(verify_task(ref), "sanity");
+  if (ref.is_narrow()) {
+    deal_with_reference((narrowOop*)ref);
+  } else {
+    deal_with_reference((oop*)ref);
+  }
+}
+
+#endif /* SHARE_VM_GC_IMPLEMENTATION_G1_G1PARSCANTHREADSTATE_INLINE_HPP */
+
--- a/src/share/vm/gc_implementation/g1/g1RemSet.inline.hpp	Mon Jul 21 09:40:19 2014 +0200
+++ b/src/share/vm/gc_implementation/g1/g1RemSet.inline.hpp	Mon Jul 21 09:41:04 2014 +0200
@@ -26,6 +26,7 @@
 #define SHARE_VM_GC_IMPLEMENTATION_G1_G1REMSET_INLINE_HPP
 
 #include "gc_implementation/g1/g1RemSet.hpp"
+#include "gc_implementation/g1/heapRegion.hpp"
 #include "gc_implementation/g1/heapRegionRemSet.hpp"
 #include "oops/oop.inline.hpp"