truffle: src/share/vm/memory/referenceProcessor.cpp comparison

comparison src/share/vm/memory/referenceProcessor.cpp @ 1833:8b10f48633dc

6984287: Regularize how GC parallel workers are specified. Summary: Associate number of GC workers with the workgang as opposed to the task. Reviewed-by: johnc, ysr

author	jmasa
date	Mon, 20 Sep 2010 14:38:38 -0700
parents	c18cbe5936b8
children	f95d63e2154a

comparison

equal deleted inserted replaced

-:97fbf5beff7b
+:8b10f48633dc
 /*
-* Copyright (c) 2001, 2009, Oracle and/or its affiliates. All rights reserved.
+* Copyright (c) 2001, 2010, 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.
 {
 _span = span;
 _discovery_is_atomic = atomic_discovery;
 _discovery_is_mt     = mt_discovery;
 _num_q               = mt_degree;
-_discoveredSoftRefs  = NEW_C_HEAP_ARRAY(DiscoveredList, _num_q * subclasses_of_ref);
+_max_num_q           = mt_degree;
+_discoveredSoftRefs  = NEW_C_HEAP_ARRAY(DiscoveredList, _max_num_q * subclasses_of_ref);
 if (_discoveredSoftRefs == NULL) {
 vm_exit_during_initialization("Could not allocated RefProc Array");
 }
-_discoveredWeakRefs    = &_discoveredSoftRefs[_num_q];
+_discoveredWeakRefs    = &_discoveredSoftRefs[_max_num_q];
-_discoveredFinalRefs   = &_discoveredWeakRefs[_num_q];
+_discoveredFinalRefs   = &_discoveredWeakRefs[_max_num_q];
-_discoveredPhantomRefs = &_discoveredFinalRefs[_num_q];
+_discoveredPhantomRefs = &_discoveredFinalRefs[_max_num_q];
 assert(sentinel_ref() != NULL, "_sentinelRef is NULL");
 // Initialized all entries to _sentinelRef
-for (int i = 0; i < _num_q * subclasses_of_ref; i++) {
+for (int i = 0; i < _max_num_q * subclasses_of_ref; i++) {
 _discoveredSoftRefs[i].set_head(sentinel_ref());
 _discoveredSoftRefs[i].set_length(0);
 }
 // If we do barreirs, cache a copy of the barrier set.
 if (discovered_list_needs_barrier) {
 }
 #ifndef PRODUCT
 void ReferenceProcessor::verify_no_references_recorded() {
 guarantee(!_discovering_refs, "Discovering refs?");
-for (int i = 0; i < _num_q * subclasses_of_ref; i++) {
+for (int i = 0; i < _max_num_q * subclasses_of_ref; i++) {
 guarantee(_discoveredSoftRefs[i].empty(),
 "Found non-empty discovered list");
 }
 }
 #endif
 void ReferenceProcessor::weak_oops_do(OopClosure* f) {
-for (int i = 0; i < _num_q * subclasses_of_ref; i++) {
+// Should this instead be
+// for (int i = 0; i < subclasses_of_ref; i++_ {
+//   for (int j = 0; j < _num_q; j++) {
+//     int index = i * _max_num_q + j;
+for (int i = 0; i < _max_num_q * subclasses_of_ref; i++) {
 if (UseCompressedOops) {
 f->do_oop((narrowOop*)_discoveredSoftRefs[i].adr_head());
 } else {
 f->do_oop((oop*)_discoveredSoftRefs[i].adr_head());
 }
 virtual void work(unsigned int work_id) {
 assert(work_id < (unsigned int)_ref_processor.num_q(), "Index out-of-bounds");
 // Simplest first cut: static partitioning.
 int index = work_id;
-for (int j = 0; j < subclasses_of_ref; j++, index += _n_queues) {
+// The increment on "index" must correspond to the maximum number of queues
+// (n_queues) with which that ReferenceProcessor was created.  That
+// is because of the "clever" way the discovered references lists were
+// allocated and are indexed into.  That number is ParallelGCThreads
+// currently.  Assert that.
+assert(_n_queues == (int) ParallelGCThreads, "Different number not expected");
+for (int j = 0;
+j < subclasses_of_ref;
+j++, index += _n_queues) {
 _ref_processor.enqueue_discovered_reflist(
 _refs_lists[index], _pending_list_addr);
 _refs_lists[index].set_head(_sentinel_ref);
 _refs_lists[index].set_length(0);
 }
 void ReferenceProcessor::enqueue_discovered_reflists(HeapWord* pending_list_addr,
 AbstractRefProcTaskExecutor* task_executor) {
 if (_processing_is_mt && task_executor != NULL) {
 // Parallel code
 RefProcEnqueueTask tsk(*this, _discoveredSoftRefs,
-pending_list_addr, sentinel_ref(), _num_q);
+pending_list_addr, sentinel_ref(), _max_num_q);
 task_executor->execute(tsk);
 } else {
 // Serial code: call the parent class's implementation
-for (int i = 0; i < _num_q * subclasses_of_ref; i++) {
+for (int i = 0; i < _max_num_q * subclasses_of_ref; i++) {
 enqueue_discovered_reflist(_discoveredSoftRefs[i], pending_list_addr);
 _discoveredSoftRefs[i].set_head(sentinel_ref());
 _discoveredSoftRefs[i].set_length(0);
 }
 }
 }
 // Close the reachable set
 complete_gc->do_void();
 NOT_PRODUCT(
 if (PrintGCDetails && TraceReferenceGC) {
-gclog_or_tty->print(" Dropped %d dead Refs out of %d "
+gclog_or_tty->print_cr(" Dropped %d dead Refs out of %d "
-"discovered Refs by policy ", iter.removed(), iter.processed());
+"discovered Refs by policy  list " INTPTR_FORMAT,
+iter.removed(), iter.processed(), (address)refs_list.head());
 }
 )
 }
 // Traverse the list and remove any Refs that are not active, or
 iter.next();
 }
 }
 NOT_PRODUCT(
 if (PrintGCDetails && TraceReferenceGC) {
-gclog_or_tty->print(" Dropped %d active Refs out of %d "
+gclog_or_tty->print_cr(" Dropped %d active Refs out of %d "
-"Refs in discovered list ", iter.removed(), iter.processed());
+"Refs in discovered list " INTPTR_FORMAT,
+iter.removed(), iter.processed(), (address)refs_list.head());
 }
 )
 }
 void
 }
 // Now close the newly reachable set
 complete_gc->do_void();
 NOT_PRODUCT(
 if (PrintGCDetails && TraceReferenceGC) {
-gclog_or_tty->print(" Dropped %d active Refs out of %d "
+gclog_or_tty->print_cr(" Dropped %d active Refs out of %d "
-"Refs in discovered list ", iter.removed(), iter.processed());
+"Refs in discovered list " INTPTR_FORMAT,
+iter.removed(), iter.processed(), (address)refs_list.head());
 }
 )
 }
 // Traverse the list and process the referents, by either
 ReferenceProcessor::process_phase3(DiscoveredList&    refs_list,
 bool               clear_referent,
 BoolObjectClosure* is_alive,
 OopClosure*        keep_alive,
 VoidClosure*       complete_gc) {
+ResourceMark rm;
 DiscoveredListIterator iter(refs_list, keep_alive, is_alive);
 while (iter.has_next()) {
 iter.update_discovered();
 iter.load_ptrs(DEBUG_ONLY(false /* allow_null_referent */));
 if (clear_referent) {
 refs_list.set_length(0);
 }
 void ReferenceProcessor::abandon_partial_discovery() {
 // loop over the lists
-for (int i = 0; i < _num_q * subclasses_of_ref; i++) {
+for (int i = 0; i < _max_num_q * subclasses_of_ref; i++) {
-if (TraceReferenceGC && PrintGCDetails && ((i % _num_q) == 0)) {
+if (TraceReferenceGC && PrintGCDetails && ((i % _max_num_q) == 0)) {
 gclog_or_tty->print_cr(
 "\nAbandoning %s discovered list",
 list_name(i));
 }
 abandon_partial_discovered_list(_discoveredSoftRefs[i]);
 { }
 virtual void work(unsigned int i, BoolObjectClosure& is_alive,
 OopClosure& keep_alive,
 VoidClosure& complete_gc)
 {
-_ref_processor.process_phase1(_refs_lists[i], _policy,
+Thread* thr = Thread::current();
+int refs_list_index = ((WorkerThread*)thr)->id();
+_ref_processor.process_phase1(_refs_lists[refs_list_index], _policy,
 &is_alive, &keep_alive, &complete_gc);
 }
 private:
 ReferencePolicy* _policy;
 };
 { }
 virtual void work(unsigned int i, BoolObjectClosure& is_alive,
 OopClosure& keep_alive,
 VoidClosure& complete_gc)
 {
+// Don't use "refs_list_index" calculated in this way because
+// balance_queues() has moved the Ref's into the first n queues.
+// Thread* thr = Thread::current();
+// int refs_list_index = ((WorkerThread*)thr)->id();
+// _ref_processor.process_phase3(_refs_lists[refs_list_index], _clear_referent,
 _ref_processor.process_phase3(_refs_lists[i], _clear_referent,
 &is_alive, &keep_alive, &complete_gc);
 }
 private:
 bool _clear_referent;
 };
 // Balances reference queues.
+// Move entries from all queues[0, 1, ..., _max_num_q-1] to
+// queues[0, 1, ..., _num_q-1] because only the first _num_q
+// corresponding to the active workers will be processed.
 void ReferenceProcessor::balance_queues(DiscoveredList ref_lists[])
 {
 // calculate total length
 size_t total_refs = 0;
-for (int i = 0; i < _num_q; ++i) {
+if (TraceReferenceGC && PrintGCDetails) {
+gclog_or_tty->print_cr("\nBalance ref_lists ");
+}
+for (int i = 0; i < _max_num_q; ++i) {
 total_refs += ref_lists[i].length();
+if (TraceReferenceGC && PrintGCDetails) {
+gclog_or_tty->print("%d ", ref_lists[i].length());
+}
+}
+if (TraceReferenceGC && PrintGCDetails) {
+gclog_or_tty->print_cr(" = %d", total_refs);
 }
 size_t avg_refs = total_refs / _num_q + 1;
 int to_idx = 0;
-for (int from_idx = 0; from_idx < _num_q; from_idx++) {
+for (int from_idx = 0; from_idx < _max_num_q; from_idx++) {
-while (ref_lists[from_idx].length() > avg_refs) {
+bool move_all = false;
+if (from_idx >= _num_q) {
+move_all = ref_lists[from_idx].length() > 0;
+}
+while ((ref_lists[from_idx].length() > avg_refs) ||
+move_all) {
 assert(to_idx < _num_q, "Sanity Check!");
 if (ref_lists[to_idx].length() < avg_refs) {
 // move superfluous refs
-size_t refs_to_move =
+size_t refs_to_move;
-MIN2(ref_lists[from_idx].length() - avg_refs,
+// Move all the Ref's if the from queue will not be processed.
-avg_refs - ref_lists[to_idx].length());
+if (move_all) {
+refs_to_move = MIN2(ref_lists[from_idx].length(),
+avg_refs - ref_lists[to_idx].length());
+} else {
+refs_to_move = MIN2(ref_lists[from_idx].length() - avg_refs,
+avg_refs - ref_lists[to_idx].length());
+}
 oop move_head = ref_lists[from_idx].head();
 oop move_tail = move_head;
 oop new_head  = move_head;
 // find an element to split the list on
 for (size_t j = 0; j < refs_to_move; ++j) {
 java_lang_ref_Reference::set_discovered(move_tail, ref_lists[to_idx].head());
 ref_lists[to_idx].set_head(move_head);
 ref_lists[to_idx].inc_length(refs_to_move);
 ref_lists[from_idx].set_head(new_head);
 ref_lists[from_idx].dec_length(refs_to_move);
+if (ref_lists[from_idx].length() == 0) {
+break;
+}
 } else {
-++to_idx;
+to_idx = (to_idx + 1) % _num_q;
 }
 }
 }
+#ifdef ASSERT
+size_t balanced_total_refs = 0;
+for (int i = 0; i < _max_num_q; ++i) {
+balanced_total_refs += ref_lists[i].length();
+if (TraceReferenceGC && PrintGCDetails) {
+gclog_or_tty->print("%d ", ref_lists[i].length());
+}
+}
+if (TraceReferenceGC && PrintGCDetails) {
+gclog_or_tty->print_cr(" = %d", balanced_total_refs);
+gclog_or_tty->flush();
+}
+assert(total_refs == balanced_total_refs, "Balancing was incomplete");
+#endif
+}
+void ReferenceProcessor::balance_all_queues() {
+balance_queues(_discoveredSoftRefs);
+balance_queues(_discoveredWeakRefs);
+balance_queues(_discoveredFinalRefs);
+balance_queues(_discoveredPhantomRefs);
 }
 void
 ReferenceProcessor::process_discovered_reflist(
 DiscoveredList               refs_lists[],
 BoolObjectClosure*           is_alive,
 OopClosure*                  keep_alive,
 VoidClosure*                 complete_gc,
 AbstractRefProcTaskExecutor* task_executor)
 {
-bool mt = task_executor != NULL && _processing_is_mt;
+bool mt_processing = task_executor != NULL && _processing_is_mt;
-if (mt && ParallelRefProcBalancingEnabled) {
+// If discovery used MT and a dynamic number of GC threads, then
+// the queues must be balanced for correctness if fewer than the
+// maximum number of queues were used.  The number of queue used
+// during discovery may be different than the number to be used
+// for processing so don't depend of _num_q < _max_num_q as part
+// of the test.
+bool must_balance = _discovery_is_mt;
+if ((mt_processing && ParallelRefProcBalancingEnabled) ||
+must_balance) {
 balance_queues(refs_lists);
 }
 if (PrintReferenceGC && PrintGCDetails) {
 size_t total = 0;
 for (int i = 0; i < _num_q; ++i) {
 // . Traverse the list and remove any SoftReferences whose
 //   referents are not alive, but that should be kept alive for
 //   policy reasons. Keep alive the transitive closure of all
 //   such referents.
 if (policy != NULL) {
-if (mt) {
+if (mt_processing) {
 RefProcPhase1Task phase1(*this, refs_lists, policy, true /*marks_oops_alive*/);
 task_executor->execute(phase1);
 } else {
 for (int i = 0; i < _num_q; i++) {
 process_phase1(refs_lists[i], policy,
 "Policy must be specified for soft references.");
 }
 // Phase 2:
 // . Traverse the list and remove any refs whose referents are alive.
-if (mt) {
+if (mt_processing) {
 RefProcPhase2Task phase2(*this, refs_lists, !discovery_is_atomic() /*marks_oops_alive*/);
 task_executor->execute(phase2);
 } else {
 for (int i = 0; i < _num_q; i++) {
 process_phase2(refs_lists[i], is_alive, keep_alive, complete_gc);
 }
 }
 // Phase 3:
 // . Traverse the list and process referents as appropriate.
-if (mt) {
+if (mt_processing) {
 RefProcPhase3Task phase3(*this, refs_lists, clear_referent, true /*marks_oops_alive*/);
 task_executor->execute(phase3);
 } else {
 for (int i = 0; i < _num_q; i++) {
 process_phase3(refs_lists[i], clear_referent,
 }
 }
 void ReferenceProcessor::clean_up_discovered_references() {
 // loop over the lists
-for (int i = 0; i < _num_q * subclasses_of_ref; i++) {
+// Should this instead be
+// for (int i = 0; i < subclasses_of_ref; i++_ {
+//   for (int j = 0; j < _num_q; j++) {
+//     int index = i * _max_num_q + j;
+for (int i = 0; i < _max_num_q * subclasses_of_ref; i++) {
 if (TraceReferenceGC && PrintGCDetails && ((i % _num_q) == 0)) {
 gclog_or_tty->print_cr(
 "\nScrubbing %s discovered list of Null referents",
 list_name(i));
 }
 // fashion to each of the lists.
 if (_processing_is_mt) {
 id = next_id();
 }
 }
-assert(0 <= id && id < _num_q, "Id is out-of-bounds (call Freud?)");
+assert(0 <= id && id < _max_num_q, "Id is out-of-bounds (call Freud?)");
 // Get the discovered queue to which we will add
 DiscoveredList* list = NULL;
 switch (rt) {
 case REF_OTHER:
 break;
 case REF_NONE:
 // we should not reach here if we are an instanceRefKlass
 default:
 ShouldNotReachHere();
+}
+if (TraceReferenceGC && PrintGCDetails) {
+gclog_or_tty->print_cr("Thread %d gets list " INTPTR_FORMAT,
+id, list);
 }
 return list;
 }
 inline void
 #endif
 // Soft references
 {
 TraceTime tt("Preclean SoftReferences", PrintGCDetails && PrintReferenceGC,
 false, gclog_or_tty);
-for (int i = 0; i < _num_q; i++) {
+for (int i = 0; i < _max_num_q; i++) {
 if (yield->should_return()) {
 return;
 }
 preclean_discovered_reflist(_discoveredSoftRefs[i], is_alive,
 keep_alive, complete_gc, yield);
 // Close the reachable set
 complete_gc->do_void();
 NOT_PRODUCT(
 if (PrintGCDetails && PrintReferenceGC) {
-gclog_or_tty->print(" Dropped %d Refs out of %d "
+gclog_or_tty->print_cr(" Dropped %d Refs out of %d "
-"Refs in discovered list ", iter.removed(), iter.processed());
+"Refs in discovered list " INTPTR_FORMAT,
+iter.removed(), iter.processed(), (address)refs_list.head());
 }
 )
 }
 const char* ReferenceProcessor::list_name(int i) {
-assert(i >= 0 && i <= _num_q * subclasses_of_ref, "Out of bounds index");
+assert(i >= 0 && i <= _max_num_q * subclasses_of_ref, "Out of bounds index");
-int j = i / _num_q;
+int j = i / _max_num_q;
 switch (j) {
 case 0: return "SoftRef";
 case 1: return "WeakRef";
 case 2: return "FinalRef";
 case 3: return "PhantomRef";
 }
 #ifndef PRODUCT
 void ReferenceProcessor::clear_discovered_references() {
 guarantee(!_discovering_refs, "Discovering refs?");
-for (int i = 0; i < _num_q * subclasses_of_ref; i++) {
+for (int i = 0; i < _max_num_q * subclasses_of_ref; i++) {
 oop obj = _discoveredSoftRefs[i].head();
 while (obj != sentinel_ref()) {
 oop next = java_lang_ref_Reference::discovered(obj);
 java_lang_ref_Reference::set_discovered(obj, (oop) NULL);
 obj = next;

Mercurial > hg > truffle

comparison src/share/vm/memory/referenceProcessor.cpp @ 1833:8b10f48633dc