8029075: String deduplication in G1

Implementation of JEP 192, http://openjdk.java.net/jeps/192

Reviewed-by: brutisso, tschatzl, coleenp

Author: pliden

hotspot/make/excludeSrc.make

@@ -87,7 +87,8 @@ ifeq ($(INCLUDE_ALL_GCS), false)
 	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 g1OopClosures.cpp \
-	g1RemSet.cpp g1RemSetSummary.cpp g1SATBCardTableModRefBS.cpp g1_globals.cpp heapRegion.cpp \
+	g1RemSet.cpp g1RemSetSummary.cpp g1SATBCardTableModRefBS.cpp g1StringDedup.cpp g1StringDedupStat.cpp \
+	g1StringDedupTable.cpp g1StringDedupThread.cpp g1StringDedupQueue.cpp g1_globals.cpp heapRegion.cpp \
 	g1BiasedArray.cpp heapRegionRemSet.cpp heapRegionSeq.cpp heapRegionSet.cpp heapRegionSets.cpp \
 	ptrQueue.cpp satbQueue.cpp sparsePRT.cpp survRateGroup.cpp vm_operations_g1.cpp g1CodeCacheRemSet.cpp \
 	adjoiningGenerations.cpp adjoiningVirtualSpaces.cpp asPSOldGen.cpp asPSYoungGen.cpp \

hotspot/src/share/vm/classfile/javaClasses.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 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
@@ -61,10 +61,6 @@ class java_lang_String : AllStatic {
 
   static Handle basic_create(int length, TRAPS);
 
-  static void set_value( oop string, typeArrayOop buffer) {
-    assert(initialized, "Must be initialized");
-    string->obj_field_put(value_offset,  (oop)buffer);
-  }
   static void set_offset(oop string, int offset) {
     assert(initialized, "Must be initialized");
     if (offset_offset > 0) {
@@ -122,12 +118,26 @@ class java_lang_String : AllStatic {
     return hash_offset;
   }
 
+  static void set_value(oop string, typeArrayOop buffer) {
+    assert(initialized && (value_offset > 0), "Must be initialized");
+    string->obj_field_put(value_offset, (oop)buffer);
+  }
+  static void set_hash(oop string, unsigned int hash) {
+    assert(initialized && (hash_offset > 0), "Must be initialized");
+    string->int_field_put(hash_offset, hash);
+  }
+
   // Accessors
   static typeArrayOop value(oop java_string) {
     assert(initialized && (value_offset > 0), "Must be initialized");
     assert(is_instance(java_string), "must be java_string");
     return (typeArrayOop) java_string->obj_field(value_offset);
   }
+  static unsigned int hash(oop java_string) {
+    assert(initialized && (hash_offset > 0), "Must be initialized");
+    assert(is_instance(java_string), "must be java_string");
+    return java_string->int_field(hash_offset);
+  }
   static int offset(oop java_string) {
     assert(initialized, "Must be initialized");
     assert(is_instance(java_string), "must be java_string");

hotspot/src/share/vm/classfile/symbolTable.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 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
@@ -35,6 +35,9 @@
 #include "oops/oop.inline2.hpp"
 #include "runtime/mutexLocker.hpp"
 #include "utilities/hashtable.inline.hpp"
+#if INCLUDE_ALL_GCS
+#include "gc_implementation/g1/g1StringDedup.hpp"
+#endif
 
 // --------------------------------------------------------------------------
 
@@ -728,6 +731,15 @@ oop StringTable::intern(Handle string_or_null, jchar* name,
     string = java_lang_String::create_from_unicode(name, len, CHECK_NULL);
   }
 
+#if INCLUDE_ALL_GCS
+  if (G1StringDedup::is_enabled()) {
+    // Deduplicate the string before it is interned. Note that we should never
+    // deduplicate a string after it has been interned. Doing so will counteract
+    // compiler optimizations done on e.g. interned string literals.
+    G1StringDedup::deduplicate(string());
+  }
+#endif
+
   // Grab the StringTable_lock before getting the_table() because it could
   // change at safepoint.
   MutexLocker ml(StringTable_lock, THREAD);

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

@@ -39,6 +39,7 @@
 #include "gc_implementation/g1/g1MarkSweep.hpp"
 #include "gc_implementation/g1/g1OopClosures.inline.hpp"
 #include "gc_implementation/g1/g1RemSet.inline.hpp"
+#include "gc_implementation/g1/g1StringDedup.hpp"
 #include "gc_implementation/g1/g1YCTypes.hpp"
 #include "gc_implementation/g1/heapRegion.inline.hpp"
 #include "gc_implementation/g1/heapRegionRemSet.hpp"
@@ -2172,6 +2173,8 @@ jint G1CollectedHeap::initialize() {
   // values in the heap have been properly initialized.
   _g1mm = new G1MonitoringSupport(this);
 
+  G1StringDedup::initialize();
+
   return JNI_OK;
 }
 
@@ -3456,6 +3459,11 @@ void G1CollectedHeap::verify(bool silent, VerifyOption vo) {
     if (!silent) gclog_or_tty->print("RemSet ");
     rem_set()->verify();
 
+    if (G1StringDedup::is_enabled()) {
+      if (!silent) gclog_or_tty->print("StrDedup ");
+      G1StringDedup::verify();
+    }
+
     if (failures) {
       gclog_or_tty->print_cr("Heap:");
       // It helps to have the per-region information in the output to
@@ -3473,8 +3481,13 @@ void G1CollectedHeap::verify(bool silent, VerifyOption vo) {
     }
     guarantee(!failures, "there should not have been any failures");
   } else {
-    if (!silent)
-      gclog_or_tty->print("(SKIPPING roots, heapRegionSets, heapRegions, remset) ");
+    if (!silent) {
+      gclog_or_tty->print("(SKIPPING Roots, HeapRegionSets, HeapRegions, RemSet");
+      if (G1StringDedup::is_enabled()) {
+        gclog_or_tty->print(", StrDedup");
+      }
+      gclog_or_tty->print(") ");
+    }
   }
 }
 
@@ -3567,6 +3580,9 @@ void G1CollectedHeap::print_gc_threads_on(outputStream* st) const {
   st->cr();
   _cm->print_worker_threads_on(st);
   _cg1r->print_worker_threads_on(st);
+  if (G1StringDedup::is_enabled()) {
+    G1StringDedup::print_worker_threads_on(st);
+  }
 }
 
 void G1CollectedHeap::gc_threads_do(ThreadClosure* tc) const {
@@ -3575,6 +3591,9 @@ void G1CollectedHeap::gc_threads_do(ThreadClosure* tc) const {
   }
   tc->do_thread(_cmThread);
   _cg1r->threads_do(tc);
+  if (G1StringDedup::is_enabled()) {
+    G1StringDedup::threads_do(tc);
+  }
 }
 
 void G1CollectedHeap::print_tracing_info() const {
@@ -4755,6 +4774,13 @@ oop G1ParScanThreadState::copy_to_survivor_space(oop const old) {
       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;
 
@@ -5218,6 +5244,10 @@ void G1CollectedHeap::unlink_string_and_symbol_table(BoolObjectClosure* is_alive
                            g1_unlink_task.strings_processed(), g1_unlink_task.strings_removed(),
                            g1_unlink_task.symbols_processed(), g1_unlink_task.symbols_removed());
   }
+
+  if (G1StringDedup::is_enabled()) {
+    G1StringDedup::unlink(is_alive);
+  }
 }
 
 class RedirtyLoggedCardTableEntryFastClosure : public CardTableEntryClosure {
@@ -5841,6 +5871,9 @@ void G1CollectedHeap::evacuate_collection_set(EvacuationInfo& evacuation_info) {
     G1STWIsAliveClosure is_alive(this);
     G1KeepAliveClosure keep_alive(this);
     JNIHandles::weak_oops_do(&is_alive, &keep_alive);
+    if (G1StringDedup::is_enabled()) {
+      G1StringDedup::unlink_or_oops_do(&is_alive, &keep_alive);
+    }
   }
 
   release_gc_alloc_regions(n_workers, evacuation_info);
@@ -6321,9 +6354,10 @@ void G1CollectedHeap::tear_down_region_sets(bool free_list_only) {
     TearDownRegionSetsClosure cl(&_old_set);
     heap_region_iterate(&cl);
 
-    // Need to do this after the heap iteration to be able to
-    // recognize the young regions and ignore them during the iteration.
-    _young_list->empty_list();
+    // Note that emptying the _young_list is postponed and instead done as
+    // the first step when rebuilding the regions sets again. The reason for
+    // this is that during a full GC string deduplication needs to know if
+    // a collected region was young or old when the full GC was initiated.
   }
   _free_list.remove_all();
 }
@@ -6377,6 +6411,10 @@ class RebuildRegionSetsClosure : public HeapRegionClosure {
 void G1CollectedHeap::rebuild_region_sets(bool free_list_only) {
   assert_at_safepoint(true /* should_be_vm_thread */);
 
+  if (!free_list_only) {
+    _young_list->empty_list();
+  }
+
   RebuildRegionSetsClosure cl(free_list_only, &_old_set, &_free_list);
   heap_region_iterate(&cl);
 

hotspot/src/share/vm/gc_implementation/g1/g1GCPhaseTimes.cpp

@@ -27,6 +27,7 @@
 #include "gc_implementation/g1/g1CollectedHeap.inline.hpp"
 #include "gc_implementation/g1/g1GCPhaseTimes.hpp"
 #include "gc_implementation/g1/g1Log.hpp"
+#include "gc_implementation/g1/g1StringDedup.hpp"
 
 // Helper class for avoiding interleaved logging
 class LineBuffer: public StackObj {
@@ -168,7 +169,9 @@ G1GCPhaseTimes::G1GCPhaseTimes(uint max_gc_threads) :
   _last_termination_attempts(_max_gc_threads, SIZE_FORMAT),
   _last_gc_worker_end_times_ms(_max_gc_threads, "%.1lf", false),
   _last_gc_worker_times_ms(_max_gc_threads, "%.1lf"),
-  _last_gc_worker_other_times_ms(_max_gc_threads, "%.1lf")
+  _last_gc_worker_other_times_ms(_max_gc_threads, "%.1lf"),
+  _cur_string_dedup_queue_fixup_worker_times_ms(_max_gc_threads, "%.1lf"),
+  _cur_string_dedup_table_fixup_worker_times_ms(_max_gc_threads, "%.1lf")
 {
   assert(max_gc_threads > 0, "Must have some GC threads");
 }
@@ -229,6 +232,16 @@ void G1GCPhaseTimes::note_gc_end() {
   _last_gc_worker_other_times_ms.verify();
 }
 
+void G1GCPhaseTimes::note_string_dedup_fixup_start() {
+  _cur_string_dedup_queue_fixup_worker_times_ms.reset();
+  _cur_string_dedup_table_fixup_worker_times_ms.reset();
+}
+
+void G1GCPhaseTimes::note_string_dedup_fixup_end() {
+  _cur_string_dedup_queue_fixup_worker_times_ms.verify();
+  _cur_string_dedup_table_fixup_worker_times_ms.verify();
+}
+
 void G1GCPhaseTimes::print_stats(int level, const char* str, double value) {
   LineBuffer(level).append_and_print_cr("[%s: %.1lf ms]", str, value);
 }
@@ -253,6 +266,11 @@ double G1GCPhaseTimes::accounted_time_ms() {
     // Strong code root purge time
     misc_time_ms += _cur_strong_code_root_purge_time_ms;
 
+    if (G1StringDedup::is_enabled()) {
+      // String dedup fixup time
+      misc_time_ms += _cur_string_dedup_fixup_time_ms;
+    }
+
     // Subtract the time taken to clean the card table from the
     // current value of "other time"
     misc_time_ms += _cur_clear_ct_time_ms;
@@ -303,6 +321,11 @@ void G1GCPhaseTimes::print(double pause_time_sec) {
   print_stats(1, "Code Root Fixup", _cur_collection_code_root_fixup_time_ms);
   print_stats(1, "Code Root Migration", _cur_strong_code_root_migration_time_ms);
   print_stats(1, "Code Root Purge", _cur_strong_code_root_purge_time_ms);
+  if (G1StringDedup::is_enabled()) {
+    print_stats(1, "String Dedup Fixup", _cur_string_dedup_fixup_time_ms, _active_gc_threads);
+    _cur_string_dedup_queue_fixup_worker_times_ms.print(2, "Queue Fixup (ms)");
+    _cur_string_dedup_table_fixup_worker_times_ms.print(2, "Table Fixup (ms)");
+  }
   print_stats(1, "Clear CT", _cur_clear_ct_time_ms);
   double misc_time_ms = pause_time_sec * MILLIUNITS - accounted_time_ms();
   print_stats(1, "Other", misc_time_ms);

hotspot/src/share/vm/gc_implementation/g1/g1GCPhaseTimes.hpp

@@ -137,6 +137,10 @@ class G1GCPhaseTimes : public CHeapObj<mtGC> {
   double _cur_evac_fail_restore_remsets;
   double _cur_evac_fail_remove_self_forwards;
 
+  double                  _cur_string_dedup_fixup_time_ms;
+  WorkerDataArray<double> _cur_string_dedup_queue_fixup_worker_times_ms;
+  WorkerDataArray<double> _cur_string_dedup_table_fixup_worker_times_ms;
+
   double _cur_clear_ct_time_ms;
   double _cur_ref_proc_time_ms;
   double _cur_ref_enq_time_ms;
@@ -246,6 +250,21 @@ class G1GCPhaseTimes : public CHeapObj<mtGC> {
     _cur_evac_fail_remove_self_forwards = ms;
   }
 
+  void note_string_dedup_fixup_start();
+  void note_string_dedup_fixup_end();
+
+  void record_string_dedup_fixup_time(double ms) {
+    _cur_string_dedup_fixup_time_ms = ms;
+  }
+
+  void record_string_dedup_queue_fixup_worker_time(uint worker_id, double ms) {
+    _cur_string_dedup_queue_fixup_worker_times_ms.set(worker_id, ms);
+  }
+
+  void record_string_dedup_table_fixup_worker_time(uint worker_id, double ms) {
+    _cur_string_dedup_table_fixup_worker_times_ms.set(worker_id, ms);
+  }
+
   void record_ref_proc_time(double ms) {
     _cur_ref_proc_time_ms = ms;
   }

hotspot/src/share/vm/gc_implementation/g1/g1MarkSweep.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2001, 2013, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 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
@@ -31,6 +31,7 @@
 #include "code/icBuffer.hpp"
 #include "gc_implementation/g1/g1Log.hpp"
 #include "gc_implementation/g1/g1MarkSweep.hpp"
+#include "gc_implementation/g1/g1StringDedup.hpp"
 #include "gc_implementation/shared/gcHeapSummary.hpp"
 #include "gc_implementation/shared/gcTimer.hpp"
 #include "gc_implementation/shared/gcTrace.hpp"
@@ -316,6 +317,10 @@ void G1MarkSweep::mark_sweep_phase3() {
   // have been cleared if they pointed to non-surviving objects.)
   sh->process_weak_roots(&GenMarkSweep::adjust_pointer_closure);
 
+  if (G1StringDedup::is_enabled()) {
+    G1StringDedup::oops_do(&GenMarkSweep::adjust_pointer_closure);
+  }
+
   GenMarkSweep::adjust_marks();
 
   G1AdjustPointersClosure blk;