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

rev 6140 : 8029075: String deduplication in G1
Implementation of JEP 192, http://openjdk.java.net/jeps/192

@@ -37,10 +37,11 @@
 #include "gc_implementation/g1/g1GCPhaseTimes.hpp"
 #include "gc_implementation/g1/g1Log.hpp"
 #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"
 #include "gc_implementation/g1/heapRegionSeq.inline.hpp"
 #include "gc_implementation/g1/vm_operations_g1.hpp"

@@ -2171,10 +2172,12 @@
 
   // Do create of the monitoring and management support so that
   // values in the heap have been properly initialized.
   _g1mm = new G1MonitoringSupport(this);
 
+  G1StringDedup::initialize();
+
   return JNI_OK;
 }
 
 size_t G1CollectedHeap::conservative_max_heap_alignment() {
   return HeapRegion::max_region_size();

@@ -3455,10 +3458,15 @@
       }
     }
     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
       // help us track down what went wrong. This is why we call
       // print_extended_on() instead of print_on().

@@ -3472,12 +3480,17 @@
 #endif
       gclog_or_tty->flush();
     }
     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(") ");
+    }
   }
 }
 
 void G1CollectedHeap::verify(bool silent) {
   verify(silent, VerifyOption_G1UsePrevMarking);

@@ -3566,18 +3579,24 @@
   }
   _cmThread->print_on(st);
   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 {
   if (G1CollectedHeap::use_parallel_gc_threads()) {
     workers()->threads_do(tc);
   }
   tc->do_thread(_cmThread);
   _cg1r->threads_do(tc);
+  if (G1StringDedup::is_enabled()) {
+    G1StringDedup::threads_do(tc);
+  }
 }
 
 void G1CollectedHeap::print_tracing_info() const {
   // We'll overload this to mean "trace GC pause statistics."
   if (TraceGen0Time || TraceGen1Time) {

@@ -4754,10 +4773,17 @@
       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

@@ -5217,10 +5243,14 @@
                            "strings: "SIZE_FORMAT" processed, "SIZE_FORMAT" removed, "
                            "symbols: "SIZE_FORMAT" processed, "SIZE_FORMAT" removed",
                            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 {
 public:
   bool do_card_ptr(jbyte* card_ptr, int worker_i) {

@@ -5840,10 +5870,13 @@
   // Weak root processing.
   {
     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);
   g1_rem_set()->cleanup_after_oops_into_collection_set_do();
 

@@ -6320,13 +6353,14 @@
 
   if (!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();
 }
 
 class RebuildRegionSetsClosure : public HeapRegionClosure {

@@ -6376,10 +6410,14 @@
 };
 
 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);
 
   if (!free_list_only) {
     _summary_bytes_used = cl.total_used();