changeset 5124:c1604d5885a6 hs25-b47

Merge
author amurillo
date Fri, 23 Aug 2013 03:01:16 -0700
parents c93e0a210e1b 8009adb44523
children acac3bde66b2
files
diffstat 138 files changed, 5069 insertions(+), 2417 deletions(-) [+]
line wrap: on
line diff
--- a/agent/src/share/classes/sun/jvm/hotspot/oops/InstanceKlass.java	Thu Aug 22 09:10:01 2013 -0700
+++ b/agent/src/share/classes/sun/jvm/hotspot/oops/InstanceKlass.java	Fri Aug 23 03:01:16 2013 -0700
@@ -75,19 +75,19 @@
     javaFieldsCount      = new CIntField(type.getCIntegerField("_java_fields_count"), 0);
     constants            = new MetadataField(type.getAddressField("_constants"), 0);
     classLoaderData      = type.getAddressField("_class_loader_data");
-    sourceFileName       = type.getAddressField("_source_file_name");
     sourceDebugExtension = type.getAddressField("_source_debug_extension");
     innerClasses         = type.getAddressField("_inner_classes");
+    sourceFileNameIndex  = new CIntField(type.getCIntegerField("_source_file_name_index"), 0);
     nonstaticFieldSize   = new CIntField(type.getCIntegerField("_nonstatic_field_size"), 0);
     staticFieldSize      = new CIntField(type.getCIntegerField("_static_field_size"), 0);
-    staticOopFieldCount   = new CIntField(type.getCIntegerField("_static_oop_field_count"), 0);
+    staticOopFieldCount  = new CIntField(type.getCIntegerField("_static_oop_field_count"), 0);
     nonstaticOopMapSize  = new CIntField(type.getCIntegerField("_nonstatic_oop_map_size"), 0);
     isMarkedDependent    = new CIntField(type.getCIntegerField("_is_marked_dependent"), 0);
     initState            = new CIntField(type.getCIntegerField("_init_state"), 0);
     vtableLen            = new CIntField(type.getCIntegerField("_vtable_len"), 0);
     itableLen            = new CIntField(type.getCIntegerField("_itable_len"), 0);
     breakpoints          = type.getAddressField("_breakpoints");
-    genericSignature     = type.getAddressField("_generic_signature");
+    genericSignatureIndex = new CIntField(type.getCIntegerField("_generic_signature_index"), 0);
     majorVersion         = new CIntField(type.getCIntegerField("_major_version"), 0);
     minorVersion         = new CIntField(type.getCIntegerField("_minor_version"), 0);
     headerSize           = Oop.alignObjectOffset(type.getSize());
@@ -134,9 +134,9 @@
   private static CIntField javaFieldsCount;
   private static MetadataField constants;
   private static AddressField  classLoaderData;
-  private static AddressField  sourceFileName;
   private static AddressField  sourceDebugExtension;
   private static AddressField  innerClasses;
+  private static CIntField sourceFileNameIndex;
   private static CIntField nonstaticFieldSize;
   private static CIntField staticFieldSize;
   private static CIntField staticOopFieldCount;
@@ -146,7 +146,7 @@
   private static CIntField vtableLen;
   private static CIntField itableLen;
   private static AddressField breakpoints;
-  private static AddressField  genericSignature;
+  private static CIntField genericSignatureIndex;
   private static CIntField majorVersion;
   private static CIntField minorVersion;
 
@@ -346,7 +346,7 @@
   public ConstantPool getConstants()        { return (ConstantPool) constants.getValue(this); }
   public ClassLoaderData getClassLoaderData() { return                ClassLoaderData.instantiateWrapperFor(classLoaderData.getValue(getAddress())); }
   public Oop       getClassLoader()         { return                getClassLoaderData().getClassLoader(); }
-  public Symbol    getSourceFileName()      { return getSymbol(sourceFileName); }
+  public Symbol    getSourceFileName()      { return                getConstants().getSymbolAt(sourceFileNameIndex.getValue(this)); }
   public String    getSourceDebugExtension(){ return                CStringUtilities.getString(sourceDebugExtension.getValue(getAddress())); }
   public long      getNonstaticFieldSize()  { return                nonstaticFieldSize.getValue(this); }
   public long      getStaticOopFieldCount() { return                staticOopFieldCount.getValue(this); }
@@ -354,7 +354,7 @@
   public boolean   getIsMarkedDependent()   { return                isMarkedDependent.getValue(this) != 0; }
   public long      getVtableLen()           { return                vtableLen.getValue(this); }
   public long      getItableLen()           { return                itableLen.getValue(this); }
-  public Symbol    getGenericSignature()    { return getSymbol(genericSignature); }
+  public Symbol    getGenericSignature()    { return                getConstants().getSymbolAt(genericSignatureIndex.getValue(this)); }
   public long      majorVersion()           { return                majorVersion.getValue(this); }
   public long      minorVersion()           { return                minorVersion.getValue(this); }
 
--- a/agent/src/share/classes/sun/jvm/hotspot/tools/jcore/ClassDump.java	Thu Aug 22 09:10:01 2013 -0700
+++ b/agent/src/share/classes/sun/jvm/hotspot/tools/jcore/ClassDump.java	Fri Aug 23 03:01:16 2013 -0700
@@ -92,8 +92,13 @@
                     System.err.println("Warning: Can not create class filter!");
                 }
             }
-            String outputDirectory = System.getProperty("sun.jvm.hotspot.tools.jcore.outputDir", ".");
-            setOutputDirectory(outputDirectory);
+
+            // outputDirectory and jarStream are alternatives: setting one closes the other.
+            // If neither is set, use outputDirectory from the System property:
+            if (outputDirectory == null && jarStream == null) {
+                String dirName = System.getProperty("sun.jvm.hotspot.tools.jcore.outputDir", ".");
+                setOutputDirectory(dirName);
+            }
 
             // walk through the system dictionary
             SystemDictionary dict = VM.getVM().getSystemDictionary();
--- a/make/bsd/makefiles/gcc.make	Thu Aug 22 09:10:01 2013 -0700
+++ b/make/bsd/makefiles/gcc.make	Fri Aug 23 03:01:16 2013 -0700
@@ -247,7 +247,7 @@
 # Not yet supported by clang in Xcode 4.6.2
 #  WARNINGS_ARE_ERRORS += -Wno-tautological-constant-out-of-range-compare
   WARNINGS_ARE_ERRORS += -Wno-delete-non-virtual-dtor -Wno-deprecated -Wno-format -Wno-dynamic-class-memaccess
-  WARNINGS_ARE_ERRORS += -Wno-return-type -Wno-empty-body
+  WARNINGS_ARE_ERRORS += -Wno-empty-body
 endif
 
 WARNING_FLAGS = -Wpointer-arith -Wsign-compare -Wundef
--- a/make/hotspot_version	Thu Aug 22 09:10:01 2013 -0700
+++ b/make/hotspot_version	Fri Aug 23 03:01:16 2013 -0700
@@ -35,7 +35,7 @@
 
 HS_MAJOR_VER=25
 HS_MINOR_VER=0
-HS_BUILD_NUMBER=46
+HS_BUILD_NUMBER=47
 
 JDK_MAJOR_VER=1
 JDK_MINOR_VER=8
--- a/src/cpu/sparc/vm/macroAssembler_sparc.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/sparc/vm/macroAssembler_sparc.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2012, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2013, 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
@@ -29,6 +29,7 @@
 #include "interpreter/interpreter.hpp"
 #include "memory/cardTableModRefBS.hpp"
 #include "memory/resourceArea.hpp"
+#include "memory/universe.hpp"
 #include "prims/methodHandles.hpp"
 #include "runtime/biasedLocking.hpp"
 #include "runtime/interfaceSupport.hpp"
@@ -1145,7 +1146,7 @@
   assert(oop_recorder() != NULL, "this assembler needs an OopRecorder");
   int klass_index = oop_recorder()->find_index(k);
   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
-  narrowOop encoded_k = oopDesc::encode_klass(k);
+  narrowOop encoded_k = Klass::encode_klass(k);
 
   assert_not_delayed();
   // Relocation with special format (see relocInfo_sparc.hpp).
@@ -1419,7 +1420,6 @@
   load_klass(O0_obj, O0_obj);
   // assert((klass != NULL)
   br_null_short(O0_obj, pn, fail);
-  // TODO: Future assert that klass is lower 4g memory for UseCompressedKlassPointers
 
   wrccr( O5_save_flags ); // Restore CCR's
 
@@ -4089,52 +4089,91 @@
 }
 
 void MacroAssembler::encode_klass_not_null(Register r) {
-  assert(Metaspace::is_initialized(), "metaspace should be initialized");
   assert (UseCompressedKlassPointers, "must be compressed");
-  assert (LogKlassAlignmentInBytes == Universe::narrow_klass_shift(), "decode alg wrong");
-  if (Universe::narrow_klass_base() != NULL)
-    sub(r, G6_heapbase, r);
-  srlx(r, LogKlassAlignmentInBytes, r);
+  assert(Universe::narrow_klass_base() != NULL, "narrow_klass_base should be initialized");
+  assert(r != G6_heapbase, "bad register choice");
+  set((intptr_t)Universe::narrow_klass_base(), G6_heapbase);
+  sub(r, G6_heapbase, r);
+  if (Universe::narrow_klass_shift() != 0) {
+    assert (LogKlassAlignmentInBytes == Universe::narrow_klass_shift(), "decode alg wrong");
+    srlx(r, LogKlassAlignmentInBytes, r);
+  }
+  reinit_heapbase();
 }
 
 void MacroAssembler::encode_klass_not_null(Register src, Register dst) {
-  assert(Metaspace::is_initialized(), "metaspace should be initialized");
-  assert (UseCompressedKlassPointers, "must be compressed");
-  assert (LogKlassAlignmentInBytes == Universe::narrow_klass_shift(), "decode alg wrong");
-  if (Universe::narrow_klass_base() == NULL) {
-    srlx(src, LogKlassAlignmentInBytes, dst);
+  if (src == dst) {
+    encode_klass_not_null(src);
   } else {
-    sub(src, G6_heapbase, dst);
-    srlx(dst, LogKlassAlignmentInBytes, dst);
+    assert (UseCompressedKlassPointers, "must be compressed");
+    assert(Universe::narrow_klass_base() != NULL, "narrow_klass_base should be initialized");
+    set((intptr_t)Universe::narrow_klass_base(), dst);
+    sub(src, dst, dst);
+    if (Universe::narrow_klass_shift() != 0) {
+      srlx(dst, LogKlassAlignmentInBytes, dst);
+    }
   }
 }
 
+// Function instr_size_for_decode_klass_not_null() counts the instructions
+// generated by decode_klass_not_null() and reinit_heapbase().  Hence, if
+// the instructions they generate change, then this method needs to be updated.
+int MacroAssembler::instr_size_for_decode_klass_not_null() {
+  assert (UseCompressedKlassPointers, "only for compressed klass ptrs");
+  // set + add + set
+  int num_instrs = insts_for_internal_set((intptr_t)Universe::narrow_klass_base()) + 1 +
+    insts_for_internal_set((intptr_t)Universe::narrow_ptrs_base());
+  if (Universe::narrow_klass_shift() == 0) {
+    return num_instrs * BytesPerInstWord;
+  } else { // sllx
+    return (num_instrs + 1) * BytesPerInstWord;
+  }
+}
+
+// !!! If the instructions that get generated here change then function
+// instr_size_for_decode_klass_not_null() needs to get updated.
 void  MacroAssembler::decode_klass_not_null(Register r) {
-  assert(Metaspace::is_initialized(), "metaspace should be initialized");
   // Do not add assert code to this unless you change vtableStubs_sparc.cpp
   // pd_code_size_limit.
   assert (UseCompressedKlassPointers, "must be compressed");
-  assert (LogKlassAlignmentInBytes == Universe::narrow_klass_shift(), "decode alg wrong");
-  sllx(r, LogKlassAlignmentInBytes, r);
-  if (Universe::narrow_klass_base() != NULL)
-    add(r, G6_heapbase, r);
+  assert(Universe::narrow_klass_base() != NULL, "narrow_klass_base should be initialized");
+  assert(r != G6_heapbase, "bad register choice");
+  set((intptr_t)Universe::narrow_klass_base(), G6_heapbase);
+  if (Universe::narrow_klass_shift() != 0)
+    sllx(r, LogKlassAlignmentInBytes, r);
+  add(r, G6_heapbase, r);
+  reinit_heapbase();
 }
 
 void  MacroAssembler::decode_klass_not_null(Register src, Register dst) {
-  assert(Metaspace::is_initialized(), "metaspace should be initialized");
-  // Do not add assert code to this unless you change vtableStubs_sparc.cpp
-  // pd_code_size_limit.
-  assert (UseCompressedKlassPointers, "must be compressed");
-  assert (LogKlassAlignmentInBytes == Universe::narrow_klass_shift(), "decode alg wrong");
-  sllx(src, LogKlassAlignmentInBytes, dst);
-  if (Universe::narrow_klass_base() != NULL)
-    add(dst, G6_heapbase, dst);
+  if (src == dst) {
+    decode_klass_not_null(src);
+  } else {
+    // Do not add assert code to this unless you change vtableStubs_sparc.cpp
+    // pd_code_size_limit.
+    assert (UseCompressedKlassPointers, "must be compressed");
+    assert(Universe::narrow_klass_base() != NULL, "narrow_klass_base should be initialized");
+    if (Universe::narrow_klass_shift() != 0) {
+      assert((src != G6_heapbase) && (dst != G6_heapbase), "bad register choice");
+      set((intptr_t)Universe::narrow_klass_base(), G6_heapbase);
+      sllx(src, LogKlassAlignmentInBytes, dst);
+      add(dst, G6_heapbase, dst);
+      reinit_heapbase();
+    } else {
+      set((intptr_t)Universe::narrow_klass_base(), dst);
+      add(src, dst, dst);
+    }
+  }
 }
 
 void MacroAssembler::reinit_heapbase() {
   if (UseCompressedOops || UseCompressedKlassPointers) {
-    AddressLiteral base(Universe::narrow_ptrs_base_addr());
-    load_ptr_contents(base, G6_heapbase);
+    if (Universe::heap() != NULL) {
+      set((intptr_t)Universe::narrow_ptrs_base(), G6_heapbase);
+    } else {
+      AddressLiteral base(Universe::narrow_ptrs_base_addr());
+      load_ptr_contents(base, G6_heapbase);
+    }
   }
 }
 
--- a/src/cpu/sparc/vm/macroAssembler_sparc.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/sparc/vm/macroAssembler_sparc.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -1177,6 +1177,9 @@
   void push_CPU_state();
   void pop_CPU_state();
 
+  // Returns the byte size of the instructions generated by decode_klass_not_null().
+  static int instr_size_for_decode_klass_not_null();
+
   // if heap base register is used - reinit it with the correct value
   void reinit_heapbase();
 
--- a/src/cpu/sparc/vm/relocInfo_sparc.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/sparc/vm/relocInfo_sparc.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1998, 2012, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1998, 2013, 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
@@ -97,7 +97,7 @@
     guarantee(Assembler::inv_op2(inst)==Assembler::sethi_op2, "must be sethi");
     if (format() != 0) {
       assert(type() == relocInfo::oop_type || type() == relocInfo::metadata_type, "only narrow oops or klasses case");
-      jint np = type() == relocInfo::oop_type ? oopDesc::encode_heap_oop((oop)x) : oopDesc::encode_klass((Klass*)x);
+      jint np = type() == relocInfo::oop_type ? oopDesc::encode_heap_oop((oop)x) : Klass::encode_klass((Klass*)x);
       inst &= ~Assembler::hi22(-1);
       inst |=  Assembler::hi22((intptr_t)np);
       if (verify_only) {
--- a/src/cpu/sparc/vm/sparc.ad	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/sparc/vm/sparc.ad	Fri Aug 23 03:01:16 2013 -0700
@@ -559,10 +559,7 @@
     int klass_load_size;
     if (UseCompressedKlassPointers) {
       assert(Universe::heap() != NULL, "java heap should be initialized");
-      if (Universe::narrow_klass_base() == NULL)
-        klass_load_size = 2*BytesPerInstWord; // see MacroAssembler::load_klass()
-      else
-        klass_load_size = 3*BytesPerInstWord;
+      klass_load_size = MacroAssembler::instr_size_for_decode_klass_not_null() + 1*BytesPerInstWord;
     } else {
       klass_load_size = 1*BytesPerInstWord;
     }
@@ -1663,9 +1660,12 @@
   if (UseCompressedKlassPointers) {
     assert(Universe::heap() != NULL, "java heap should be initialized");
     st->print_cr("\tLDUW   [R_O0 + oopDesc::klass_offset_in_bytes],R_G5\t! Inline cache check - compressed klass");
-    st->print_cr("\tSLL    R_G5,3,R_G5");
-    if (Universe::narrow_klass_base() != NULL)
-      st->print_cr("\tADD    R_G5,R_G6_heap_base,R_G5");
+    st->print_cr("\tSET    Universe::narrow_klass_base,R_G6_heap_base");
+    if (Universe::narrow_klass_shift() != 0) {
+      st->print_cr("\tSLL    R_G5,3,R_G5");
+    }
+    st->print_cr("\tADD    R_G5,R_G6_heap_base,R_G5");
+    st->print_cr("\tSET    Universe::narrow_ptrs_base,R_G6_heap_base");
   } else {
     st->print_cr("\tLDX    [R_O0 + oopDesc::klass_offset_in_bytes],R_G5\t! Inline cache check");
   }
@@ -2563,10 +2563,7 @@
       int klass_load_size;
       if (UseCompressedKlassPointers) {
         assert(Universe::heap() != NULL, "java heap should be initialized");
-        if (Universe::narrow_klass_base() == NULL)
-          klass_load_size = 2*BytesPerInstWord;
-        else
-          klass_load_size = 3*BytesPerInstWord;
+        klass_load_size = MacroAssembler::instr_size_for_decode_klass_not_null() + 1*BytesPerInstWord;
       } else {
         klass_load_size = 1*BytesPerInstWord;
       }
--- a/src/cpu/sparc/vm/vtableStubs_sparc.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/sparc/vm/vtableStubs_sparc.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2012, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2013, 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
@@ -219,13 +219,13 @@
       const int basic = 5*BytesPerInstWord +
                         // shift;add for load_klass (only shift with zero heap based)
                         (UseCompressedKlassPointers ?
-                         ((Universe::narrow_klass_base() == NULL) ? BytesPerInstWord : 2*BytesPerInstWord) : 0);
+                          MacroAssembler::instr_size_for_decode_klass_not_null() : 0);
       return basic + slop;
     } else {
       const int basic = (28 LP64_ONLY(+ 6)) * BytesPerInstWord +
                         // shift;add for load_klass (only shift with zero heap based)
                         (UseCompressedKlassPointers ?
-                         ((Universe::narrow_klass_base() == NULL) ? BytesPerInstWord : 2*BytesPerInstWord) : 0);
+                          MacroAssembler::instr_size_for_decode_klass_not_null() : 0);
       return (basic + slop);
     }
   }
--- a/src/cpu/x86/vm/macroAssembler_x86.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/x86/vm/macroAssembler_x86.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -30,6 +30,7 @@
 #include "interpreter/interpreter.hpp"
 #include "memory/cardTableModRefBS.hpp"
 #include "memory/resourceArea.hpp"
+#include "memory/universe.hpp"
 #include "prims/methodHandles.hpp"
 #include "runtime/biasedLocking.hpp"
 #include "runtime/interfaceSupport.hpp"
@@ -4810,23 +4811,8 @@
 }
 
 void MacroAssembler::load_prototype_header(Register dst, Register src) {
-#ifdef _LP64
-  if (UseCompressedKlassPointers) {
-    assert (Universe::heap() != NULL, "java heap should be initialized");
-    movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
-    if (Universe::narrow_klass_shift() != 0) {
-      assert(LogKlassAlignmentInBytes == Universe::narrow_klass_shift(), "decode alg wrong");
-      assert(LogKlassAlignmentInBytes == Address::times_8, "klass not aligned on 64bits?");
-      movq(dst, Address(r12_heapbase, dst, Address::times_8, Klass::prototype_header_offset()));
-    } else {
-      movq(dst, Address(dst, Klass::prototype_header_offset()));
-    }
-  } else
-#endif
-  {
-    movptr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
-    movptr(dst, Address(dst, Klass::prototype_header_offset()));
-  }
+  load_klass(dst, src);
+  movptr(dst, Address(dst, Klass::prototype_header_offset()));
 }
 
 void MacroAssembler::store_klass(Register dst, Register src) {
@@ -4914,7 +4900,7 @@
 
 #ifdef ASSERT
 void MacroAssembler::verify_heapbase(const char* msg) {
-  assert (UseCompressedOops || UseCompressedKlassPointers, "should be compressed");
+  assert (UseCompressedOops, "should be compressed");
   assert (Universe::heap() != NULL, "java heap should be initialized");
   if (CheckCompressedOops) {
     Label ok;
@@ -5058,69 +5044,80 @@
 }
 
 void MacroAssembler::encode_klass_not_null(Register r) {
-  assert(Metaspace::is_initialized(), "metaspace should be initialized");
-#ifdef ASSERT
-  verify_heapbase("MacroAssembler::encode_klass_not_null: heap base corrupted?");
-#endif
-  if (Universe::narrow_klass_base() != NULL) {
-    subq(r, r12_heapbase);
-  }
+  assert(Universe::narrow_klass_base() != NULL, "Base should be initialized");
+  // Use r12 as a scratch register in which to temporarily load the narrow_klass_base.
+  assert(r != r12_heapbase, "Encoding a klass in r12");
+  mov64(r12_heapbase, (int64_t)Universe::narrow_klass_base());
+  subq(r, r12_heapbase);
   if (Universe::narrow_klass_shift() != 0) {
     assert (LogKlassAlignmentInBytes == Universe::narrow_klass_shift(), "decode alg wrong");
     shrq(r, LogKlassAlignmentInBytes);
   }
+  reinit_heapbase();
 }
 
 void MacroAssembler::encode_klass_not_null(Register dst, Register src) {
-  assert(Metaspace::is_initialized(), "metaspace should be initialized");
-#ifdef ASSERT
-  verify_heapbase("MacroAssembler::encode_klass_not_null2: heap base corrupted?");
-#endif
-  if (dst != src) {
-    movq(dst, src);
-  }
-  if (Universe::narrow_klass_base() != NULL) {
-    subq(dst, r12_heapbase);
-  }
-  if (Universe::narrow_klass_shift() != 0) {
-    assert (LogKlassAlignmentInBytes == Universe::narrow_klass_shift(), "decode alg wrong");
-    shrq(dst, LogKlassAlignmentInBytes);
-  }
-}
-
+  if (dst == src) {
+    encode_klass_not_null(src);
+  } else {
+    mov64(dst, (int64_t)Universe::narrow_klass_base());
+    negq(dst);
+    addq(dst, src);
+    if (Universe::narrow_klass_shift() != 0) {
+      assert (LogKlassAlignmentInBytes == Universe::narrow_klass_shift(), "decode alg wrong");
+      shrq(dst, LogKlassAlignmentInBytes);
+    }
+  }
+}
+
+// Function instr_size_for_decode_klass_not_null() counts the instructions
+// generated by decode_klass_not_null(register r) and reinit_heapbase(),
+// when (Universe::heap() != NULL).  Hence, if the instructions they
+// generate change, then this method needs to be updated.
+int MacroAssembler::instr_size_for_decode_klass_not_null() {
+  assert (UseCompressedKlassPointers, "only for compressed klass ptrs");
+  // mov64 + addq + shlq? + mov64  (for reinit_heapbase()).
+  return (Universe::narrow_klass_shift() == 0 ? 20 : 24);
+}
+
+// !!! If the instructions that get generated here change then function
+// instr_size_for_decode_klass_not_null() needs to get updated.
 void  MacroAssembler::decode_klass_not_null(Register r) {
-  assert(Metaspace::is_initialized(), "metaspace should be initialized");
   // Note: it will change flags
+  assert(Universe::narrow_klass_base() != NULL, "Base should be initialized");
   assert (UseCompressedKlassPointers, "should only be used for compressed headers");
+  assert(r != r12_heapbase, "Decoding a klass in r12");
   // Cannot assert, unverified entry point counts instructions (see .ad file)
   // vtableStubs also counts instructions in pd_code_size_limit.
   // Also do not verify_oop as this is called by verify_oop.
   if (Universe::narrow_klass_shift() != 0) {
     assert(LogKlassAlignmentInBytes == Universe::narrow_klass_shift(), "decode alg wrong");
     shlq(r, LogKlassAlignmentInBytes);
-    if (Universe::narrow_klass_base() != NULL) {
-      addq(r, r12_heapbase);
-    }
+  }
+  // Use r12 as a scratch register in which to temporarily load the narrow_klass_base.
+  mov64(r12_heapbase, (int64_t)Universe::narrow_klass_base());
+  addq(r, r12_heapbase);
+  reinit_heapbase();
+}
+
+void  MacroAssembler::decode_klass_not_null(Register dst, Register src) {
+  // Note: it will change flags
+  assert(Universe::narrow_klass_base() != NULL, "Base should be initialized");
+  assert (UseCompressedKlassPointers, "should only be used for compressed headers");
+  if (dst == src) {
+    decode_klass_not_null(dst);
   } else {
-    assert (Universe::narrow_klass_base() == NULL, "sanity");
-  }
-}
-
-void  MacroAssembler::decode_klass_not_null(Register dst, Register src) {
-  assert(Metaspace::is_initialized(), "metaspace should be initialized");
-  // Note: it will change flags
-  assert (UseCompressedKlassPointers, "should only be used for compressed headers");
-  // Cannot assert, unverified entry point counts instructions (see .ad file)
-  // vtableStubs also counts instructions in pd_code_size_limit.
-  // Also do not verify_oop as this is called by verify_oop.
-  if (Universe::narrow_klass_shift() != 0) {
-    assert(LogKlassAlignmentInBytes == Universe::narrow_klass_shift(), "decode alg wrong");
-    assert(LogKlassAlignmentInBytes == Address::times_8, "klass not aligned on 64bits?");
-    leaq(dst, Address(r12_heapbase, src, Address::times_8, 0));
-  } else {
-    assert (Universe::narrow_klass_base() == NULL, "sanity");
-    if (dst != src) {
-      movq(dst, src);
+    // Cannot assert, unverified entry point counts instructions (see .ad file)
+    // vtableStubs also counts instructions in pd_code_size_limit.
+    // Also do not verify_oop as this is called by verify_oop.
+
+    mov64(dst, (int64_t)Universe::narrow_klass_base());
+    if (Universe::narrow_klass_shift() != 0) {
+      assert(LogKlassAlignmentInBytes == Universe::narrow_klass_shift(), "decode alg wrong");
+      assert(LogKlassAlignmentInBytes == Address::times_8, "klass not aligned on 64bits?");
+      leaq(dst, Address(dst, src, Address::times_8, 0));
+    } else {
+      addq(dst, src);
     }
   }
 }
@@ -5148,7 +5145,7 @@
   assert (oop_recorder() != NULL, "this assembler needs an OopRecorder");
   int klass_index = oop_recorder()->find_index(k);
   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
-  mov_narrow_oop(dst, oopDesc::encode_klass(k), rspec);
+  mov_narrow_oop(dst, Klass::encode_klass(k), rspec);
 }
 
 void  MacroAssembler::set_narrow_klass(Address dst, Klass* k) {
@@ -5156,7 +5153,7 @@
   assert (oop_recorder() != NULL, "this assembler needs an OopRecorder");
   int klass_index = oop_recorder()->find_index(k);
   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
-  mov_narrow_oop(dst, oopDesc::encode_klass(k), rspec);
+  mov_narrow_oop(dst, Klass::encode_klass(k), rspec);
 }
 
 void  MacroAssembler::cmp_narrow_oop(Register dst, jobject obj) {
@@ -5182,7 +5179,7 @@
   assert (oop_recorder() != NULL, "this assembler needs an OopRecorder");
   int klass_index = oop_recorder()->find_index(k);
   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
-  Assembler::cmp_narrow_oop(dst, oopDesc::encode_klass(k), rspec);
+  Assembler::cmp_narrow_oop(dst, Klass::encode_klass(k), rspec);
 }
 
 void  MacroAssembler::cmp_narrow_klass(Address dst, Klass* k) {
@@ -5190,14 +5187,23 @@
   assert (oop_recorder() != NULL, "this assembler needs an OopRecorder");
   int klass_index = oop_recorder()->find_index(k);
   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
-  Assembler::cmp_narrow_oop(dst, oopDesc::encode_klass(k), rspec);
+  Assembler::cmp_narrow_oop(dst, Klass::encode_klass(k), rspec);
 }
 
 void MacroAssembler::reinit_heapbase() {
   if (UseCompressedOops || UseCompressedKlassPointers) {
-    movptr(r12_heapbase, ExternalAddress((address)Universe::narrow_ptrs_base_addr()));
-  }
-}
+    if (Universe::heap() != NULL) {
+      if (Universe::narrow_oop_base() == NULL) {
+        MacroAssembler::xorptr(r12_heapbase, r12_heapbase);
+      } else {
+        mov64(r12_heapbase, (int64_t)Universe::narrow_ptrs_base());
+      }
+    } else {
+      movptr(r12_heapbase, ExternalAddress((address)Universe::narrow_ptrs_base_addr()));
+    }
+  }
+}
+
 #endif // _LP64
 
 
--- a/src/cpu/x86/vm/macroAssembler_x86.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/x86/vm/macroAssembler_x86.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -371,6 +371,10 @@
   void cmp_narrow_klass(Register dst, Klass* k);
   void cmp_narrow_klass(Address dst, Klass* k);
 
+  // Returns the byte size of the instructions generated by decode_klass_not_null()
+  // when compressed klass pointers are being used.
+  static int instr_size_for_decode_klass_not_null();
+
   // if heap base register is used - reinit it with the correct value
   void reinit_heapbase();
 
--- a/src/cpu/x86/vm/relocInfo_x86.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/x86/vm/relocInfo_x86.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1998, 2012, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1998, 2013, 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
@@ -55,9 +55,9 @@
     }
   } else {
       if (verify_only) {
-        assert(*(uint32_t*) disp == oopDesc::encode_klass((Klass*)x), "instructions must match");
+        assert(*(uint32_t*) disp == Klass::encode_klass((Klass*)x), "instructions must match");
       } else {
-        *(int32_t*) disp = oopDesc::encode_klass((Klass*)x);
+        *(int32_t*) disp = Klass::encode_klass((Klass*)x);
       }
     }
   } else {
--- a/src/cpu/x86/vm/stubGenerator_x86_32.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/x86/vm/stubGenerator_x86_32.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -675,7 +675,6 @@
     __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass
     __ testptr(rax, rax);
     __ jcc(Assembler::zero, error);              // if klass is NULL it is broken
-    // TODO: Future assert that klass is lower 4g memory for UseCompressedKlassPointers
 
     // return if everything seems ok
     __ bind(exit);
--- a/src/cpu/x86/vm/stubGenerator_x86_64.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/x86/vm/stubGenerator_x86_64.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -1021,7 +1021,6 @@
     __ load_klass(rax, rax);  // get klass
     __ testptr(rax, rax);
     __ jcc(Assembler::zero, error); // if klass is NULL it is broken
-    // TODO: Future assert that klass is lower 4g memory for UseCompressedKlassPointers
 
     // return if everything seems ok
     __ bind(exit);
--- a/src/cpu/x86/vm/templateInterpreter_x86_64.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/x86/vm/templateInterpreter_x86_64.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -849,9 +849,9 @@
     address entry = __ pc();
 
     // rbx,: Method*
-    // rsi: senderSP must preserved for slow path, set SP to it on fast path
-    // rdx: scratch
-    // rdi: scratch
+    // r13: senderSP must preserved for slow path, set SP to it on fast path
+    // c_rarg0: scratch (rdi on non-Win64, rcx on Win64)
+    // c_rarg1: scratch (rsi on non-Win64, rdx on Win64)
 
     Label slow_path;
     // If we need a safepoint check, generate full interpreter entry.
@@ -865,8 +865,8 @@
 
     // Load parameters
     const Register crc = rax;  // crc
-    const Register val = rdx;  // source java byte value
-    const Register tbl = rdi;  // scratch
+    const Register val = c_rarg0;  // source java byte value
+    const Register tbl = c_rarg1;  // scratch
 
     // Arguments are reversed on java expression stack
     __ movl(val, Address(rsp,   wordSize)); // byte value
@@ -880,7 +880,7 @@
 
     // _areturn
     __ pop(rdi);                // get return address
-    __ mov(rsp, rsi);           // set sp to sender sp
+    __ mov(rsp, r13);           // set sp to sender sp
     __ jmp(rdi);
 
     // generate a vanilla native entry as the slow path
@@ -919,20 +919,24 @@
     const Register crc = c_rarg0;  // crc
     const Register buf = c_rarg1;  // source java byte array address
     const Register len = c_rarg2;  // length
+    const Register off = len;      // offset (never overlaps with 'len')
 
     // Arguments are reversed on java expression stack
-    __ movl(len,   Address(rsp,   wordSize)); // Length
     // Calculate address of start element
     if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
       __ movptr(buf, Address(rsp, 3*wordSize)); // long buf
-      __ addptr(buf, Address(rsp, 2*wordSize)); // + offset
+      __ movl2ptr(off, Address(rsp, 2*wordSize)); // offset
+      __ addq(buf, off); // + offset
       __ movl(crc,   Address(rsp, 5*wordSize)); // Initial CRC
     } else {
       __ movptr(buf, Address(rsp, 3*wordSize)); // byte[] array
       __ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
-      __ addptr(buf, Address(rsp, 2*wordSize)); // + offset
+      __ movl2ptr(off, Address(rsp, 2*wordSize)); // offset
+      __ addq(buf, off); // + offset
       __ movl(crc,   Address(rsp, 4*wordSize)); // Initial CRC
     }
+    // Can now load 'len' since we're finished with 'off'
+    __ movl(len, Address(rsp, wordSize)); // Length
 
     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()), crc, buf, len);
     // result in rax
--- a/src/cpu/x86/vm/vtableStubs_x86_64.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/x86/vm/vtableStubs_x86_64.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -211,11 +211,11 @@
   if (is_vtable_stub) {
     // Vtable stub size
     return (DebugVtables ? 512 : 24) + (CountCompiledCalls ? 13 : 0) +
-           (UseCompressedKlassPointers ? 16 : 0);  // 1 leaq can be 3 bytes + 1 long
+           (UseCompressedKlassPointers ?  MacroAssembler::instr_size_for_decode_klass_not_null() : 0);
   } else {
     // Itable stub size
     return (DebugVtables ? 512 : 74) + (CountCompiledCalls ? 13 : 0) +
-           (UseCompressedKlassPointers ? 32 : 0);  // 2 leaqs
+           (UseCompressedKlassPointers ?  MacroAssembler::instr_size_for_decode_klass_not_null() : 0);
   }
   // In order to tune these parameters, run the JVM with VM options
   // +PrintMiscellaneous and +WizardMode to see information about
--- a/src/cpu/x86/vm/x86_64.ad	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/x86/vm/x86_64.ad	Fri Aug 23 03:01:16 2013 -0700
@@ -1393,9 +1393,7 @@
 {
   if (UseCompressedKlassPointers) {
     st->print_cr("movl    rscratch1, [j_rarg0 + oopDesc::klass_offset_in_bytes()]\t# compressed klass");
-    if (Universe::narrow_klass_shift() != 0) {
-      st->print_cr("\tdecode_klass_not_null rscratch1, rscratch1");
-    }
+    st->print_cr("\tdecode_klass_not_null rscratch1, rscratch1");
     st->print_cr("\tcmpq    rax, rscratch1\t # Inline cache check");
   } else {
     st->print_cr("\tcmpq    rax, [j_rarg0 + oopDesc::klass_offset_in_bytes()]\t"
@@ -4035,146 +4033,6 @@
   %}
 %}
 
-operand indirectNarrowKlass(rRegN reg)
-%{
-  predicate(Universe::narrow_klass_shift() == 0);
-  constraint(ALLOC_IN_RC(ptr_reg));
-  match(DecodeNKlass reg);
-
-  format %{ "[$reg]" %}
-  interface(MEMORY_INTER) %{
-    base($reg);
-    index(0x4);
-    scale(0x0);
-    disp(0x0);
-  %}
-%}
-
-operand indOffset8NarrowKlass(rRegN reg, immL8 off)
-%{
-  predicate(Universe::narrow_klass_shift() == 0);
-  constraint(ALLOC_IN_RC(ptr_reg));
-  match(AddP (DecodeNKlass reg) off);
-
-  format %{ "[$reg + $off (8-bit)]" %}
-  interface(MEMORY_INTER) %{
-    base($reg);
-    index(0x4);
-    scale(0x0);
-    disp($off);
-  %}
-%}
-
-operand indOffset32NarrowKlass(rRegN reg, immL32 off)
-%{
-  predicate(Universe::narrow_klass_shift() == 0);
-  constraint(ALLOC_IN_RC(ptr_reg));
-  match(AddP (DecodeNKlass reg) off);
-
-  format %{ "[$reg + $off (32-bit)]" %}
-  interface(MEMORY_INTER) %{
-    base($reg);
-    index(0x4);
-    scale(0x0);
-    disp($off);
-  %}
-%}
-
-operand indIndexOffsetNarrowKlass(rRegN reg, rRegL lreg, immL32 off)
-%{
-  predicate(Universe::narrow_klass_shift() == 0);
-  constraint(ALLOC_IN_RC(ptr_reg));
-  match(AddP (AddP (DecodeNKlass reg) lreg) off);
-
-  op_cost(10);
-  format %{"[$reg + $off + $lreg]" %}
-  interface(MEMORY_INTER) %{
-    base($reg);
-    index($lreg);
-    scale(0x0);
-    disp($off);
-  %}
-%}
-
-operand indIndexNarrowKlass(rRegN reg, rRegL lreg)
-%{
-  predicate(Universe::narrow_klass_shift() == 0);
-  constraint(ALLOC_IN_RC(ptr_reg));
-  match(AddP (DecodeNKlass reg) lreg);
-
-  op_cost(10);
-  format %{"[$reg + $lreg]" %}
-  interface(MEMORY_INTER) %{
-    base($reg);
-    index($lreg);
-    scale(0x0);
-    disp(0x0);
-  %}
-%}
-
-operand indIndexScaleNarrowKlass(rRegN reg, rRegL lreg, immI2 scale)
-%{
-  predicate(Universe::narrow_klass_shift() == 0);
-  constraint(ALLOC_IN_RC(ptr_reg));
-  match(AddP (DecodeNKlass reg) (LShiftL lreg scale));
-
-  op_cost(10);
-  format %{"[$reg + $lreg << $scale]" %}
-  interface(MEMORY_INTER) %{
-    base($reg);
-    index($lreg);
-    scale($scale);
-    disp(0x0);
-  %}
-%}
-
-operand indIndexScaleOffsetNarrowKlass(rRegN reg, immL32 off, rRegL lreg, immI2 scale)
-%{
-  predicate(Universe::narrow_klass_shift() == 0);
-  constraint(ALLOC_IN_RC(ptr_reg));
-  match(AddP (AddP (DecodeNKlass reg) (LShiftL lreg scale)) off);
-
-  op_cost(10);
-  format %{"[$reg + $off + $lreg << $scale]" %}
-  interface(MEMORY_INTER) %{
-    base($reg);
-    index($lreg);
-    scale($scale);
-    disp($off);
-  %}
-%}
-
-operand indCompressedKlassOffset(rRegN reg, immL32 off) %{
-  predicate(UseCompressedKlassPointers && (Universe::narrow_klass_shift() == Address::times_8));
-  constraint(ALLOC_IN_RC(ptr_reg));
-  match(AddP (DecodeNKlass reg) off);
-
-  op_cost(10);
-  format %{"[R12 + $reg << 3 + $off] (compressed klass addressing)" %}
-  interface(MEMORY_INTER) %{
-    base(0xc); // R12
-    index($reg);
-    scale(0x3);
-    disp($off);
-  %}
-%}
-
-operand indPosIndexScaleOffsetNarrowKlass(rRegN reg, immL32 off, rRegI idx, immI2 scale)
-%{
-  constraint(ALLOC_IN_RC(ptr_reg));
-  predicate(Universe::narrow_klass_shift() == 0 && n->in(2)->in(3)->in(1)->as_Type()->type()->is_long()->_lo >= 0);
-  match(AddP (AddP (DecodeNKlass reg) (LShiftL (ConvI2L idx) scale)) off);
-
-  op_cost(10);
-  format %{"[$reg + $off + $idx << $scale]" %}
-  interface(MEMORY_INTER) %{
-    base($reg);
-    index($idx);
-    scale($scale);
-    disp($off);
-  %}
-%}
-
 //----------Special Memory Operands--------------------------------------------
 // Stack Slot Operand - This operand is used for loading and storing temporary
 //                      values on the stack where a match requires a value to
@@ -4345,11 +4203,7 @@
                indCompressedOopOffset,
                indirectNarrow, indOffset8Narrow, indOffset32Narrow,
                indIndexOffsetNarrow, indIndexNarrow, indIndexScaleNarrow,
-               indIndexScaleOffsetNarrow, indPosIndexScaleOffsetNarrow,
-               indCompressedKlassOffset,
-               indirectNarrowKlass, indOffset8NarrowKlass, indOffset32NarrowKlass,
-               indIndexOffsetNarrowKlass, indIndexNarrowKlass, indIndexScaleNarrowKlass,
-               indIndexScaleOffsetNarrowKlass, indPosIndexScaleOffsetNarrowKlass);
+               indIndexScaleOffsetNarrow, indPosIndexScaleOffsetNarrow);
 
 //----------PIPELINE-----------------------------------------------------------
 // Rules which define the behavior of the target architectures pipeline.
@@ -6665,7 +6519,7 @@
 instruct encodeKlass_not_null(rRegN dst, rRegP src, rFlagsReg cr) %{
   match(Set dst (EncodePKlass src));
   effect(KILL cr);
-  format %{ "encode_heap_oop_not_null $dst,$src" %}
+  format %{ "encode_klass_not_null $dst,$src" %}
   ins_encode %{
     __ encode_klass_not_null($dst$$Register, $src$$Register);
   %}
@@ -6675,7 +6529,7 @@
 instruct decodeKlass_not_null(rRegP dst, rRegN src, rFlagsReg cr) %{
   match(Set dst (DecodeNKlass src));
   effect(KILL cr);
-  format %{ "decode_heap_oop_not_null $dst,$src" %}
+  format %{ "decode_klass_not_null $dst,$src" %}
   ins_encode %{
     Register s = $src$$Register;
     Register d = $dst$$Register;
--- a/src/cpu/zero/vm/assembler_zero.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/zero/vm/assembler_zero.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -50,6 +50,7 @@
 #ifdef ASSERT
 bool AbstractAssembler::pd_check_instruction_mark() {
   ShouldNotCallThis();
+  return false;
 }
 #endif
 
@@ -73,6 +74,7 @@
 RegisterOrConstant MacroAssembler::delayed_value_impl(
   intptr_t* delayed_value_addr, Register tmpl, int offset) {
   ShouldNotCallThis();
+  return RegisterOrConstant();
 }
 
 void MacroAssembler::store_oop(jobject obj) {
--- a/src/cpu/zero/vm/cppInterpreter_zero.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/zero/vm/cppInterpreter_zero.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -1008,6 +1008,7 @@
 
 address CppInterpreter::return_entry(TosState state, int length) {
   ShouldNotCallThis();
+  return NULL;
 }
 
 address CppInterpreter::deopt_entry(TosState state, int length) {
--- a/src/cpu/zero/vm/frame_zero.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/zero/vm/frame_zero.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -116,6 +116,7 @@
 
 bool frame::safe_for_sender(JavaThread *thread) {
   ShouldNotCallThis();
+  return false;
 }
 
 void frame::pd_gc_epilog() {
@@ -123,6 +124,7 @@
 
 bool frame::is_interpreted_frame_valid(JavaThread *thread) const {
   ShouldNotCallThis();
+  return false;
 }
 
 BasicType frame::interpreter_frame_result(oop* oop_result,
@@ -184,9 +186,8 @@
 int frame::frame_size(RegisterMap* map) const {
 #ifdef PRODUCT
   ShouldNotCallThis();
-#else
+#endif // PRODUCT
   return 0; // make javaVFrame::print_value work
-#endif // PRODUCT
 }
 
 intptr_t* frame::interpreter_frame_tos_at(jint offset) const {
--- a/src/cpu/zero/vm/frame_zero.inline.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/zero/vm/frame_zero.inline.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -36,7 +36,7 @@
   _deopt_state = unknown;
 }
 
-inline address  frame::sender_pc()           const { ShouldNotCallThis();  }
+inline address  frame::sender_pc()           const { ShouldNotCallThis(); return NULL; }
 
 inline frame::frame(ZeroFrame* zf, intptr_t* sp) {
   _zeroframe = zf;
@@ -89,6 +89,7 @@
 
 inline intptr_t* frame::link() const {
   ShouldNotCallThis();
+  return NULL;
 }
 
 #ifdef CC_INTERP
@@ -151,14 +152,17 @@
 
 inline oop frame::saved_oop_result(RegisterMap* map) const {
   ShouldNotCallThis();
+  return NULL;
 }
 
 inline bool frame::is_older(intptr_t* id) const {
   ShouldNotCallThis();
+  return false;
 }
 
 inline intptr_t* frame::entry_frame_argument_at(int offset) const {
   ShouldNotCallThis();
+  return NULL;
 }
 
 inline intptr_t* frame::unextended_sp() const {
--- a/src/cpu/zero/vm/icBuffer_zero.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/zero/vm/icBuffer_zero.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -49,8 +49,10 @@
 address InlineCacheBuffer::ic_buffer_entry_point(address code_begin) {
   // NB ic_stub_code_size() must return the size of the code we generate
   ShouldNotCallThis();
+  return NULL;
 }
 
 void* InlineCacheBuffer::ic_buffer_cached_value(address code_begin) {
   ShouldNotCallThis();
+  return NULL;
 }
--- a/src/cpu/zero/vm/interp_masm_zero.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/zero/vm/interp_masm_zero.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -40,6 +40,7 @@
                                         Register  tmp,
                                         int       offset) {
     ShouldNotCallThis();
+    return RegisterOrConstant();
   }
 };
 
--- a/src/cpu/zero/vm/interpreter_zero.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/zero/vm/interpreter_zero.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -64,6 +64,7 @@
     return NULL;
 
   Unimplemented();
+  return NULL;
 }
 
 address InterpreterGenerator::generate_abstract_entry() {
--- a/src/cpu/zero/vm/nativeInst_zero.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/zero/vm/nativeInst_zero.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -51,15 +51,18 @@
  public:
   bool is_jump() {
     ShouldNotCallThis();
+    return false;
   }
 
   bool is_safepoint_poll() {
     ShouldNotCallThis();
+    return false;
   }
 };
 
 inline NativeInstruction* nativeInstruction_at(address address) {
   ShouldNotCallThis();
+  return NULL;
 }
 
 class NativeCall : public NativeInstruction {
@@ -70,18 +73,22 @@
 
   address instruction_address() const {
     ShouldNotCallThis();
+    return NULL;
   }
 
   address next_instruction_address() const {
     ShouldNotCallThis();
+    return NULL;
   }
 
   address return_address() const {
     ShouldNotCallThis();
+    return NULL;
   }
 
   address destination() const {
     ShouldNotCallThis();
+    return NULL;
   }
 
   void set_destination_mt_safe(address dest) {
@@ -98,25 +105,30 @@
 
   static bool is_call_before(address return_address) {
     ShouldNotCallThis();
+    return false;
   }
 };
 
 inline NativeCall* nativeCall_before(address return_address) {
   ShouldNotCallThis();
+  return NULL;
 }
 
 inline NativeCall* nativeCall_at(address address) {
   ShouldNotCallThis();
+  return NULL;
 }
 
 class NativeMovConstReg : public NativeInstruction {
  public:
   address next_instruction_address() const {
     ShouldNotCallThis();
+    return NULL;
   }
 
   intptr_t data() const {
     ShouldNotCallThis();
+    return 0;
   }
 
   void set_data(intptr_t x) {
@@ -126,12 +138,14 @@
 
 inline NativeMovConstReg* nativeMovConstReg_at(address address) {
   ShouldNotCallThis();
+  return NULL;
 }
 
 class NativeMovRegMem : public NativeInstruction {
  public:
   int offset() const {
     ShouldNotCallThis();
+    return 0;
   }
 
   void set_offset(intptr_t x) {
@@ -145,6 +159,7 @@
 
 inline NativeMovRegMem* nativeMovRegMem_at(address address) {
   ShouldNotCallThis();
+  return NULL;
 }
 
 class NativeJump : public NativeInstruction {
@@ -155,6 +170,7 @@
 
   address jump_destination() const {
     ShouldNotCallThis();
+    return NULL;
   }
 
   void set_jump_destination(address dest) {
@@ -172,12 +188,14 @@
 
 inline NativeJump* nativeJump_at(address address) {
   ShouldNotCallThis();
+  return NULL;
 }
 
 class NativeGeneralJump : public NativeInstruction {
  public:
   address jump_destination() const {
     ShouldNotCallThis();
+    return NULL;
   }
 
   static void insert_unconditional(address code_pos, address entry) {
@@ -191,6 +209,7 @@
 
 inline NativeGeneralJump* nativeGeneralJump_at(address address) {
   ShouldNotCallThis();
+  return NULL;
 }
 
 #endif // CPU_ZERO_VM_NATIVEINST_ZERO_HPP
--- a/src/cpu/zero/vm/register_zero.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/zero/vm/register_zero.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -32,8 +32,10 @@
 
 const char* RegisterImpl::name() const {
   ShouldNotCallThis();
+  return NULL;
 }
 
 const char* FloatRegisterImpl::name() const {
   ShouldNotCallThis();
+  return NULL;
 }
--- a/src/cpu/zero/vm/relocInfo_zero.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/zero/vm/relocInfo_zero.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -37,6 +37,7 @@
 
 address Relocation::pd_call_destination(address orig_addr) {
   ShouldNotCallThis();
+  return NULL;
 }
 
 void Relocation::pd_set_call_destination(address x) {
@@ -45,6 +46,7 @@
 
 address Relocation::pd_get_address_from_code() {
   ShouldNotCallThis();
+  return NULL;
 }
 
 address* Relocation::pd_address_in_code() {
--- a/src/cpu/zero/vm/sharedRuntime_zero.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/zero/vm/sharedRuntime_zero.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -89,6 +89,7 @@
                                                             ret_type);
 #else
   ShouldNotCallThis();
+  return NULL;
 #endif // SHARK
 }
 
@@ -99,6 +100,7 @@
 
 uint SharedRuntime::out_preserve_stack_slots() {
   ShouldNotCallThis();
+  return 0;
 }
 
 JRT_LEAF(void, zero_stub())
@@ -135,4 +137,5 @@
                                          VMRegPair *regs,
                                          int total_args_passed) {
   ShouldNotCallThis();
+  return 0;
 }
--- a/src/cpu/zero/vm/vtableStubs_zero.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/cpu/zero/vm/vtableStubs_zero.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -39,16 +39,20 @@
 
 VtableStub* VtableStubs::create_vtable_stub(int vtable_index) {
   ShouldNotCallThis();
+  return NULL;
 }
 
 VtableStub* VtableStubs::create_itable_stub(int vtable_index) {
   ShouldNotCallThis();
+  return NULL;
 }
 
 int VtableStub::pd_code_size_limit(bool is_vtable_stub) {
   ShouldNotCallThis();
+  return 0;
 }
 
 int VtableStub::pd_code_alignment() {
   ShouldNotCallThis();
+  return 0;
 }
--- a/src/os_cpu/bsd_x86/vm/orderAccess_bsd_x86.inline.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/os_cpu/bsd_x86/vm/orderAccess_bsd_x86.inline.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -190,7 +190,7 @@
 inline void     OrderAccess::release_store_fence(volatile julong*  p, julong  v) { release_store_fence((volatile jlong*)p,  (jlong)v);  }
 
 inline void     OrderAccess::release_store_fence(volatile jfloat*  p, jfloat  v) { *p = v; fence(); }
-inline void     OrderAccess::release_store_fence(volatile jdouble* p, jdouble v) { release_store_fence((volatile jlong*)p, jdouble_cast(v)); }
+inline void     OrderAccess::release_store_fence(volatile jdouble* p, jdouble v) { release_store_fence((volatile jlong*)p, jlong_cast(v)); }
 
 inline void     OrderAccess::release_store_ptr_fence(volatile intptr_t* p, intptr_t v) {
 #ifdef AMD64
--- a/src/os_cpu/bsd_x86/vm/os_bsd_x86.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/os_cpu/bsd_x86/vm/os_bsd_x86.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -715,6 +715,7 @@
   err.report_and_die();
 
   ShouldNotReachHere();
+  return false;
 }
 
 // From solaris_i486.s ported to bsd_i486.s
--- a/src/os_cpu/bsd_zero/vm/os_bsd_zero.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/os_cpu/bsd_zero/vm/os_bsd_zero.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -66,6 +66,7 @@
 
 frame os::get_sender_for_C_frame(frame* fr) {
   ShouldNotCallThis();
+  return frame();
 }
 
 frame os::current_frame() {
@@ -103,16 +104,19 @@
 
 address os::Bsd::ucontext_get_pc(ucontext_t* uc) {
   ShouldNotCallThis();
+  return NULL;
 }
 
 ExtendedPC os::fetch_frame_from_context(void* ucVoid,
                                         intptr_t** ret_sp,
                                         intptr_t** ret_fp) {
   ShouldNotCallThis();
+  return ExtendedPC();
 }
 
 frame os::fetch_frame_from_context(void* ucVoid) {
   ShouldNotCallThis();
+  return frame();
 }
 
 extern "C" JNIEXPORT int
@@ -240,6 +244,7 @@
 
   sprintf(buf, fmt, sig, info->si_addr);
   fatal(buf);
+  return false;
 }
 
 void os::Bsd::init_thread_fpu_state(void) {
@@ -373,17 +378,7 @@
 
 extern "C" {
   int SpinPause() {
-  }
-
-  int SafeFetch32(int *adr, int errValue) {
-    int value = errValue;
-    value = *adr;
-    return value;
-  }
-  intptr_t SafeFetchN(intptr_t *adr, intptr_t errValue) {
-    intptr_t value = errValue;
-    value = *adr;
-    return value;
+    return 1;
   }
 
   void _Copy_conjoint_jshorts_atomic(jshort* from, jshort* to, size_t count) {
--- a/src/os_cpu/bsd_zero/vm/thread_bsd_zero.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/os_cpu/bsd_zero/vm/thread_bsd_zero.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -110,6 +110,7 @@
                                            void* ucontext,
                                            bool isInJava) {
     ShouldNotCallThis();
+    return false;
   }
 
   // These routines are only used on cpu architectures that
--- a/src/share/vm/c1/c1_Runtime1.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/c1/c1_Runtime1.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -915,16 +915,6 @@
     // Return to the now deoptimized frame.
   }
 
-  // If we are patching in a non-perm oop, make sure the nmethod
-  // is on the right list.
-  if (ScavengeRootsInCode && mirror.not_null() && mirror()->is_scavengable()) {
-    MutexLockerEx ml_code (CodeCache_lock, Mutex::_no_safepoint_check_flag);
-    nmethod* nm = CodeCache::find_nmethod(caller_frame.pc());
-    guarantee(nm != NULL, "only nmethods can contain non-perm oops");
-    if (!nm->on_scavenge_root_list())
-      CodeCache::add_scavenge_root_nmethod(nm);
-  }
-
   // Now copy code back
 
   {
@@ -1125,6 +1115,21 @@
       }
     }
   }
+
+  // If we are patching in a non-perm oop, make sure the nmethod
+  // is on the right list.
+  if (ScavengeRootsInCode && mirror.not_null() && mirror()->is_scavengable()) {
+    MutexLockerEx ml_code (CodeCache_lock, Mutex::_no_safepoint_check_flag);
+    nmethod* nm = CodeCache::find_nmethod(caller_frame.pc());
+    guarantee(nm != NULL, "only nmethods can contain non-perm oops");
+    if (!nm->on_scavenge_root_list()) {
+      CodeCache::add_scavenge_root_nmethod(nm);
+    }
+
+    // Since we've patched some oops in the nmethod,
+    // (re)register it with the heap.
+    Universe::heap()->register_nmethod(nm);
+  }
 JRT_END
 
 //
--- a/src/share/vm/classfile/classFileParser.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/classfile/classFileParser.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -2590,7 +2590,7 @@
     valid_symbol_at(sourcefile_index),
     "Invalid SourceFile attribute at constant pool index %u in class file %s",
     sourcefile_index, CHECK);
-  set_class_sourcefile(_cp->symbol_at(sourcefile_index));
+  set_class_sourcefile_index(sourcefile_index);
 }
 
 
@@ -2728,7 +2728,7 @@
     valid_symbol_at(signature_index),
     "Invalid constant pool index %u in Signature attribute in class file %s",
     signature_index, CHECK);
-  set_class_generic_signature(_cp->symbol_at(signature_index));
+  set_class_generic_signature_index(signature_index);
 }
 
 void ClassFileParser::parse_classfile_bootstrap_methods_attribute(u4 attribute_byte_length, TRAPS) {
@@ -2975,13 +2975,11 @@
 void ClassFileParser::apply_parsed_class_attributes(instanceKlassHandle k) {
   if (_synthetic_flag)
     k->set_is_synthetic();
-  if (_sourcefile != NULL) {
-    _sourcefile->increment_refcount();
-    k->set_source_file_name(_sourcefile);
+  if (_sourcefile_index != 0) {
+    k->set_source_file_name_index(_sourcefile_index);
   }
-  if (_generic_signature != NULL) {
-    _generic_signature->increment_refcount();
-    k->set_generic_signature(_generic_signature);
+  if (_generic_signature_index != 0) {
+    k->set_generic_signature_index(_generic_signature_index);
   }
   if (_sde_buffer != NULL) {
     k->set_source_debug_extension(_sde_buffer, _sde_length);
--- a/src/share/vm/classfile/classFileParser.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/classfile/classFileParser.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -62,8 +62,8 @@
   bool       _synthetic_flag;
   int        _sde_length;
   char*      _sde_buffer;
-  Symbol*    _sourcefile;
-  Symbol*    _generic_signature;
+  u2         _sourcefile_index;
+  u2         _generic_signature_index;
 
   // Metadata created before the instance klass is created.  Must be deallocated
   // if not transferred to the InstanceKlass upon successful class loading
@@ -81,16 +81,16 @@
   Array<AnnotationArray*>* _fields_type_annotations;
   InstanceKlass*   _klass;  // InstanceKlass once created.
 
-  void set_class_synthetic_flag(bool x)           { _synthetic_flag = x; }
-  void set_class_sourcefile(Symbol* x)            { _sourcefile = x; }
-  void set_class_generic_signature(Symbol* x)     { _generic_signature = x; }
-  void set_class_sde_buffer(char* x, int len)     { _sde_buffer = x; _sde_length = len; }
+  void set_class_synthetic_flag(bool x)        { _synthetic_flag = x; }
+  void set_class_sourcefile_index(u2 x)        { _sourcefile_index = x; }
+  void set_class_generic_signature_index(u2 x) { _generic_signature_index = x; }
+  void set_class_sde_buffer(char* x, int len)  { _sde_buffer = x; _sde_length = len; }
 
   void init_parsed_class_attributes(ClassLoaderData* loader_data) {
     _loader_data = loader_data;
     _synthetic_flag = false;
-    _sourcefile = NULL;
-    _generic_signature = NULL;
+    _sourcefile_index = 0;
+    _generic_signature_index = 0;
     _sde_buffer = NULL;
     _sde_length = 0;
     // initialize the other flags too:
--- a/src/share/vm/code/nmethod.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/code/nmethod.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -687,6 +687,7 @@
     code_buffer->copy_values_to(this);
     if (ScavengeRootsInCode && detect_scavenge_root_oops()) {
       CodeCache::add_scavenge_root_nmethod(this);
+      Universe::heap()->register_nmethod(this);
     }
     debug_only(verify_scavenge_root_oops());
     CodeCache::commit(this);
@@ -881,6 +882,7 @@
     dependencies->copy_to(this);
     if (ScavengeRootsInCode && detect_scavenge_root_oops()) {
       CodeCache::add_scavenge_root_nmethod(this);
+      Universe::heap()->register_nmethod(this);
     }
     debug_only(verify_scavenge_root_oops());
 
@@ -1300,6 +1302,13 @@
   methodHandle the_method(method());
   No_Safepoint_Verifier nsv;
 
+  // during patching, depending on the nmethod state we must notify the GC that
+  // code has been unloaded, unregistering it. We cannot do this right while
+  // holding the Patching_lock because we need to use the CodeCache_lock. This
+  // would be prone to deadlocks.
+  // This flag is used to remember whether we need to later lock and unregister.
+  bool nmethod_needs_unregister = false;
+
   {
     // invalidate osr nmethod before acquiring the patching lock since
     // they both acquire leaf locks and we don't want a deadlock.
@@ -1332,6 +1341,13 @@
       inc_decompile_count();
     }
 
+    // If the state is becoming a zombie, signal to unregister the nmethod with
+    // the heap.
+    // This nmethod may have already been unloaded during a full GC.
+    if ((state == zombie) && !is_unloaded()) {
+      nmethod_needs_unregister = true;
+    }
+
     // Change state
     _state = state;
 
@@ -1367,6 +1383,9 @@
       // safepoint can sneak in, otherwise the oops used by the
       // dependency logic could have become stale.
       MutexLockerEx mu(CodeCache_lock, Mutex::_no_safepoint_check_flag);
+      if (nmethod_needs_unregister) {
+        Universe::heap()->unregister_nmethod(this);
+      }
       flush_dependencies(NULL);
     }
 
@@ -1817,21 +1836,10 @@
   if (_method != NULL) f(_method);
 }
 
-
-// This method is called twice during GC -- once while
-// tracing the "active" nmethods on thread stacks during
-// the (strong) marking phase, and then again when walking
-// the code cache contents during the weak roots processing
-// phase. The two uses are distinguished by means of the
-// 'do_strong_roots_only' flag, which is true in the first
-// case. We want to walk the weak roots in the nmethod
-// only in the second case. The weak roots in the nmethod
-// are the oops in the ExceptionCache and the InlineCache
-// oops.
-void nmethod::oops_do(OopClosure* f, bool do_strong_roots_only) {
+void nmethod::oops_do(OopClosure* f, bool allow_zombie) {
   // make sure the oops ready to receive visitors
-  assert(!is_zombie() && !is_unloaded(),
-         "should not call follow on zombie or unloaded nmethod");
+  assert(allow_zombie || !is_zombie(), "should not call follow on zombie nmethod");
+  assert(!is_unloaded(), "should not call follow on unloaded nmethod");
 
   // If the method is not entrant or zombie then a JMP is plastered over the
   // first few bytes.  If an oop in the old code was there, that oop
--- a/src/share/vm/code/nmethod.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/code/nmethod.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -566,7 +566,7 @@
   void preserve_callee_argument_oops(frame fr, const RegisterMap *reg_map,
                                      OopClosure* f);
   void oops_do(OopClosure* f) { oops_do(f, false); }
-  void oops_do(OopClosure* f, bool do_strong_roots_only);
+  void oops_do(OopClosure* f, bool allow_zombie);
   bool detect_scavenge_root_oops();
   void verify_scavenge_root_oops() PRODUCT_RETURN;
 
--- a/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -5478,40 +5478,42 @@
   HandleMark   hm;
 
   SequentialSubTasksDone* pst = space->par_seq_tasks();
-  assert(pst->valid(), "Uninitialized use?");
 
   uint nth_task = 0;
   uint n_tasks  = pst->n_tasks();
 
-  HeapWord *start, *end;
-  while (!pst->is_task_claimed(/* reference */ nth_task)) {
-    // We claimed task # nth_task; compute its boundaries.
-    if (chunk_top == 0) {  // no samples were taken
-      assert(nth_task == 0 && n_tasks == 1, "Can have only 1 EdenSpace task");
-      start = space->bottom();
-      end   = space->top();
-    } else if (nth_task == 0) {
-      start = space->bottom();
-      end   = chunk_array[nth_task];
-    } else if (nth_task < (uint)chunk_top) {
-      assert(nth_task >= 1, "Control point invariant");
-      start = chunk_array[nth_task - 1];
-      end   = chunk_array[nth_task];
-    } else {
-      assert(nth_task == (uint)chunk_top, "Control point invariant");
-      start = chunk_array[chunk_top - 1];
-      end   = space->top();
-    }
-    MemRegion mr(start, end);
-    // Verify that mr is in space
-    assert(mr.is_empty() || space->used_region().contains(mr),
-           "Should be in space");
-    // Verify that "start" is an object boundary
-    assert(mr.is_empty() || oop(mr.start())->is_oop(),
-           "Should be an oop");
-    space->par_oop_iterate(mr, cl);
-  }
-  pst->all_tasks_completed();
+  if (n_tasks > 0) {
+    assert(pst->valid(), "Uninitialized use?");
+    HeapWord *start, *end;
+    while (!pst->is_task_claimed(/* reference */ nth_task)) {
+      // We claimed task # nth_task; compute its boundaries.
+      if (chunk_top == 0) {  // no samples were taken
+        assert(nth_task == 0 && n_tasks == 1, "Can have only 1 EdenSpace task");
+        start = space->bottom();
+        end   = space->top();
+      } else if (nth_task == 0) {
+        start = space->bottom();
+        end   = chunk_array[nth_task];
+      } else if (nth_task < (uint)chunk_top) {
+        assert(nth_task >= 1, "Control point invariant");
+        start = chunk_array[nth_task - 1];
+        end   = chunk_array[nth_task];
+      } else {
+        assert(nth_task == (uint)chunk_top, "Control point invariant");
+        start = chunk_array[chunk_top - 1];
+        end   = space->top();
+      }
+      MemRegion mr(start, end);
+      // Verify that mr is in space
+      assert(mr.is_empty() || space->used_region().contains(mr),
+             "Should be in space");
+      // Verify that "start" is an object boundary
+      assert(mr.is_empty() || oop(mr.start())->is_oop(),
+             "Should be an oop");
+      space->par_oop_iterate(mr, cl);
+    }
+    pst->all_tasks_completed();
+  }
 }
 
 void
@@ -5788,7 +5790,7 @@
   DefNewGeneration* dng = (DefNewGeneration*)_young_gen;
 
   // Eden space
-  {
+  if (!dng->eden()->is_empty()) {
     SequentialSubTasksDone* pst = dng->eden()->par_seq_tasks();
     assert(!pst->valid(), "Clobbering existing data?");
     // Each valid entry in [0, _eden_chunk_index) represents a task.
--- a/src/share/vm/gc_implementation/g1/concurrentMark.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/gc_implementation/g1/concurrentMark.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -4529,7 +4529,7 @@
     _total_prev_live_bytes(0), _total_next_live_bytes(0),
     _hum_used_bytes(0), _hum_capacity_bytes(0),
     _hum_prev_live_bytes(0), _hum_next_live_bytes(0),
-    _total_remset_bytes(0) {
+    _total_remset_bytes(0), _total_strong_code_roots_bytes(0) {
   G1CollectedHeap* g1h = G1CollectedHeap::heap();
   MemRegion g1_committed = g1h->g1_committed();
   MemRegion g1_reserved = g1h->g1_reserved();
@@ -4553,9 +4553,11 @@
                 G1PPRL_BYTE_H_FORMAT
                 G1PPRL_BYTE_H_FORMAT
                 G1PPRL_DOUBLE_H_FORMAT
+                G1PPRL_BYTE_H_FORMAT
                 G1PPRL_BYTE_H_FORMAT,
                 "type", "address-range",
-                "used", "prev-live", "next-live", "gc-eff", "remset");
+                "used", "prev-live", "next-live", "gc-eff",
+                "remset", "code-roots");
   _out->print_cr(G1PPRL_LINE_PREFIX
                 G1PPRL_TYPE_H_FORMAT
                 G1PPRL_ADDR_BASE_H_FORMAT
@@ -4563,9 +4565,11 @@
                 G1PPRL_BYTE_H_FORMAT
                 G1PPRL_BYTE_H_FORMAT
                 G1PPRL_DOUBLE_H_FORMAT
+                G1PPRL_BYTE_H_FORMAT
                 G1PPRL_BYTE_H_FORMAT,
                 "", "",
-                "(bytes)", "(bytes)", "(bytes)", "(bytes/ms)", "(bytes)");
+                "(bytes)", "(bytes)", "(bytes)", "(bytes/ms)",
+                "(bytes)", "(bytes)");
 }
 
 // It takes as a parameter a reference to one of the _hum_* fields, it
@@ -4608,6 +4612,8 @@
   size_t next_live_bytes = r->next_live_bytes();
   double gc_eff          = r->gc_efficiency();
   size_t remset_bytes    = r->rem_set()->mem_size();
+  size_t strong_code_roots_bytes = r->rem_set()->strong_code_roots_mem_size();
+
   if (r->used() == 0) {
     type = "FREE";
   } else if (r->is_survivor()) {
@@ -4642,6 +4648,7 @@
   _total_prev_live_bytes += prev_live_bytes;
   _total_next_live_bytes += next_live_bytes;
   _total_remset_bytes    += remset_bytes;
+  _total_strong_code_roots_bytes += strong_code_roots_bytes;
 
   // Print a line for this particular region.
   _out->print_cr(G1PPRL_LINE_PREFIX
@@ -4651,9 +4658,11 @@
                  G1PPRL_BYTE_FORMAT
                  G1PPRL_BYTE_FORMAT
                  G1PPRL_DOUBLE_FORMAT
+                 G1PPRL_BYTE_FORMAT
                  G1PPRL_BYTE_FORMAT,
                  type, bottom, end,
-                 used_bytes, prev_live_bytes, next_live_bytes, gc_eff , remset_bytes);
+                 used_bytes, prev_live_bytes, next_live_bytes, gc_eff,
+                 remset_bytes, strong_code_roots_bytes);
 
   return false;
 }
@@ -4669,7 +4678,8 @@
                  G1PPRL_SUM_MB_PERC_FORMAT("used")
                  G1PPRL_SUM_MB_PERC_FORMAT("prev-live")
                  G1PPRL_SUM_MB_PERC_FORMAT("next-live")
-                 G1PPRL_SUM_MB_FORMAT("remset"),
+                 G1PPRL_SUM_MB_FORMAT("remset")
+                 G1PPRL_SUM_MB_FORMAT("code-roots"),
                  bytes_to_mb(_total_capacity_bytes),
                  bytes_to_mb(_total_used_bytes),
                  perc(_total_used_bytes, _total_capacity_bytes),
@@ -4677,6 +4687,7 @@
                  perc(_total_prev_live_bytes, _total_capacity_bytes),
                  bytes_to_mb(_total_next_live_bytes),
                  perc(_total_next_live_bytes, _total_capacity_bytes),
-                 bytes_to_mb(_total_remset_bytes));
+                 bytes_to_mb(_total_remset_bytes),
+                 bytes_to_mb(_total_strong_code_roots_bytes));
   _out->cr();
 }
--- a/src/share/vm/gc_implementation/g1/concurrentMark.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/gc_implementation/g1/concurrentMark.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -1257,6 +1257,9 @@
   // Accumulator for the remembered set size
   size_t _total_remset_bytes;
 
+  // Accumulator for strong code roots memory size
+  size_t _total_strong_code_roots_bytes;
+
   static double perc(size_t val, size_t total) {
     if (total == 0) {
       return 0.0;
--- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -23,6 +23,7 @@
  */
 
 #include "precompiled.hpp"
+#include "code/codeCache.hpp"
 #include "code/icBuffer.hpp"
 #include "gc_implementation/g1/bufferingOopClosure.hpp"
 #include "gc_implementation/g1/concurrentG1Refine.hpp"
@@ -1176,20 +1177,27 @@
   ModRefBarrierSet* _mr_bs;
 public:
   PostMCRemSetClearClosure(G1CollectedHeap* g1h, ModRefBarrierSet* mr_bs) :
-    _g1h(g1h), _mr_bs(mr_bs) { }
+    _g1h(g1h), _mr_bs(mr_bs) {}
+
   bool doHeapRegion(HeapRegion* r) {
+    HeapRegionRemSet* hrrs = r->rem_set();
+
     if (r->continuesHumongous()) {
+      // We'll assert that the strong code root list and RSet is empty
+      assert(hrrs->strong_code_roots_list_length() == 0, "sanity");
+      assert(hrrs->occupied() == 0, "RSet should be empty");
       return false;
     }
+
     _g1h->reset_gc_time_stamps(r);
-    HeapRegionRemSet* hrrs = r->rem_set();
-    if (hrrs != NULL) hrrs->clear();
+    hrrs->clear();
     // You might think here that we could clear just the cards
     // corresponding to the used region.  But no: if we leave a dirty card
     // in a region we might allocate into, then it would prevent that card
     // from being enqueued, and cause it to be missed.
     // Re: the performance cost: we shouldn't be doing full GC anyway!
     _mr_bs->clear(MemRegion(r->bottom(), r->end()));
+
     return false;
   }
 };
@@ -1269,30 +1277,6 @@
   heap_region_iterate(&cl);
 }
 
-double G1CollectedHeap::verify(bool guard, const char* msg) {
-  double verify_time_ms = 0.0;
-
-  if (guard && total_collections() >= VerifyGCStartAt) {
-    double verify_start = os::elapsedTime();
-    HandleMark hm;  // Discard invalid handles created during verification
-    prepare_for_verify();
-    Universe::verify(VerifyOption_G1UsePrevMarking, msg);
-    verify_time_ms = (os::elapsedTime() - verify_start) * 1000;
-  }
-
-  return verify_time_ms;
-}
-
-void G1CollectedHeap::verify_before_gc() {
-  double verify_time_ms = verify(VerifyBeforeGC, " VerifyBeforeGC:");
-  g1_policy()->phase_times()->record_verify_before_time_ms(verify_time_ms);
-}
-
-void G1CollectedHeap::verify_after_gc() {
-  double verify_time_ms = verify(VerifyAfterGC, " VerifyAfterGC:");
-  g1_policy()->phase_times()->record_verify_after_time_ms(verify_time_ms);
-}
-
 bool G1CollectedHeap::do_collection(bool explicit_gc,
                                     bool clear_all_soft_refs,
                                     size_t word_size) {
@@ -1433,7 +1417,7 @@
 
       // Delete metaspaces for unloaded class loaders and clean up loader_data graph
       ClassLoaderDataGraph::purge();
-    MetaspaceAux::verify_metrics();
+      MetaspaceAux::verify_metrics();
 
       // Note: since we've just done a full GC, concurrent
       // marking is no longer active. Therefore we need not
@@ -1504,6 +1488,9 @@
         heap_region_iterate(&rebuild_rs);
       }
 
+      // Rebuild the strong code root lists for each region
+      rebuild_strong_code_roots();
+
       if (true) { // FIXME
         MetaspaceGC::compute_new_size();
       }
@@ -3109,6 +3096,145 @@
   return NULL; // keep some compilers happy
 }
 
+// TODO: VerifyRootsClosure extends OopsInGenClosure so that we can
+//       pass it as the perm_blk to SharedHeap::process_strong_roots.
+//       When process_strong_roots stop calling perm_blk->younger_refs_iterate
+//       we can change this closure to extend the simpler OopClosure.
+class VerifyRootsClosure: public OopsInGenClosure {
+private:
+  G1CollectedHeap* _g1h;
+  VerifyOption     _vo;
+  bool             _failures;
+public:
+  // _vo == UsePrevMarking -> use "prev" marking information,
+  // _vo == UseNextMarking -> use "next" marking information,
+  // _vo == UseMarkWord    -> use mark word from object header.
+  VerifyRootsClosure(VerifyOption vo) :
+    _g1h(G1CollectedHeap::heap()),
+    _vo(vo),
+    _failures(false) { }
+
+  bool failures() { return _failures; }
+
+  template <class T> void do_oop_nv(T* p) {
+    T heap_oop = oopDesc::load_heap_oop(p);
+    if (!oopDesc::is_null(heap_oop)) {
+      oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
+      if (_g1h->is_obj_dead_cond(obj, _vo)) {
+        gclog_or_tty->print_cr("Root location "PTR_FORMAT" "
+                              "points to dead obj "PTR_FORMAT, p, (void*) obj);
+        if (_vo == VerifyOption_G1UseMarkWord) {
+          gclog_or_tty->print_cr("  Mark word: "PTR_FORMAT, (void*)(obj->mark()));
+        }
+        obj->print_on(gclog_or_tty);
+        _failures = true;
+      }
+    }
+  }
+
+  void do_oop(oop* p)       { do_oop_nv(p); }
+  void do_oop(narrowOop* p) { do_oop_nv(p); }
+};
+
+class G1VerifyCodeRootOopClosure: public OopsInGenClosure {
+  G1CollectedHeap* _g1h;
+  OopClosure* _root_cl;
+  nmethod* _nm;
+  VerifyOption _vo;
+  bool _failures;
+
+  template <class T> void do_oop_work(T* p) {
+    // First verify that this root is live
+    _root_cl->do_oop(p);
+
+    if (!G1VerifyHeapRegionCodeRoots) {
+      // We're not verifying the code roots attached to heap region.
+      return;
+    }
+
+    // Don't check the code roots during marking verification in a full GC
+    if (_vo == VerifyOption_G1UseMarkWord) {
+      return;
+    }
+
+    // Now verify that the current nmethod (which contains p) is
+    // in the code root list of the heap region containing the
+    // object referenced by p.
+
+    T heap_oop = oopDesc::load_heap_oop(p);
+    if (!oopDesc::is_null(heap_oop)) {
+      oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
+
+      // Now fetch the region containing the object
+      HeapRegion* hr = _g1h->heap_region_containing(obj);
+      HeapRegionRemSet* hrrs = hr->rem_set();
+      // Verify that the strong code root list for this region
+      // contains the nmethod
+      if (!hrrs->strong_code_roots_list_contains(_nm)) {
+        gclog_or_tty->print_cr("Code root location "PTR_FORMAT" "
+                              "from nmethod "PTR_FORMAT" not in strong "
+                              "code roots for region ["PTR_FORMAT","PTR_FORMAT")",
+                              p, _nm, hr->bottom(), hr->end());
+        _failures = true;
+      }
+    }
+  }
+
+public:
+  G1VerifyCodeRootOopClosure(G1CollectedHeap* g1h, OopClosure* root_cl, VerifyOption vo):
+    _g1h(g1h), _root_cl(root_cl), _vo(vo), _nm(NULL), _failures(false) {}
+
+  void do_oop(oop* p) { do_oop_work(p); }
+  void do_oop(narrowOop* p) { do_oop_work(p); }
+
+  void set_nmethod(nmethod* nm) { _nm = nm; }
+  bool failures() { return _failures; }
+};
+
+class G1VerifyCodeRootBlobClosure: public CodeBlobClosure {
+  G1VerifyCodeRootOopClosure* _oop_cl;
+
+public:
+  G1VerifyCodeRootBlobClosure(G1VerifyCodeRootOopClosure* oop_cl):
+    _oop_cl(oop_cl) {}
+
+  void do_code_blob(CodeBlob* cb) {
+    nmethod* nm = cb->as_nmethod_or_null();
+    if (nm != NULL) {
+      _oop_cl->set_nmethod(nm);
+      nm->oops_do(_oop_cl);
+    }
+  }
+};
+
+class YoungRefCounterClosure : public OopClosure {
+  G1CollectedHeap* _g1h;
+  int              _count;
+ public:
+  YoungRefCounterClosure(G1CollectedHeap* g1h) : _g1h(g1h), _count(0) {}
+  void do_oop(oop* p)       { if (_g1h->is_in_young(*p)) { _count++; } }
+  void do_oop(narrowOop* p) { ShouldNotReachHere(); }
+
+  int count() { return _count; }
+  void reset_count() { _count = 0; };
+};
+
+class VerifyKlassClosure: public KlassClosure {
+  YoungRefCounterClosure _young_ref_counter_closure;
+  OopClosure *_oop_closure;
+ public:
+  VerifyKlassClosure(G1CollectedHeap* g1h, OopClosure* cl) : _young_ref_counter_closure(g1h), _oop_closure(cl) {}
+  void do_klass(Klass* k) {
+    k->oops_do(_oop_closure);
+
+    _young_ref_counter_closure.reset_count();
+    k->oops_do(&_young_ref_counter_closure);
+    if (_young_ref_counter_closure.count() > 0) {
+      guarantee(k->has_modified_oops(), err_msg("Klass %p, has young refs but is not dirty.", k));
+    }
+  }
+};
+
 class VerifyLivenessOopClosure: public OopClosure {
   G1CollectedHeap* _g1h;
   VerifyOption _vo;
@@ -3242,75 +3368,7 @@
   }
 };
 
-class YoungRefCounterClosure : public OopClosure {
-  G1CollectedHeap* _g1h;
-  int              _count;
- public:
-  YoungRefCounterClosure(G1CollectedHeap* g1h) : _g1h(g1h), _count(0) {}
-  void do_oop(oop* p)       { if (_g1h->is_in_young(*p)) { _count++; } }
-  void do_oop(narrowOop* p) { ShouldNotReachHere(); }
-
-  int count() { return _count; }
-  void reset_count() { _count = 0; };
-};
-
-class VerifyKlassClosure: public KlassClosure {
-  YoungRefCounterClosure _young_ref_counter_closure;
-  OopClosure *_oop_closure;
- public:
-  VerifyKlassClosure(G1CollectedHeap* g1h, OopClosure* cl) : _young_ref_counter_closure(g1h), _oop_closure(cl) {}
-  void do_klass(Klass* k) {
-    k->oops_do(_oop_closure);
-
-    _young_ref_counter_closure.reset_count();
-    k->oops_do(&_young_ref_counter_closure);
-    if (_young_ref_counter_closure.count() > 0) {
-      guarantee(k->has_modified_oops(), err_msg("Klass %p, has young refs but is not dirty.", k));
-    }
-  }
-};
-
-// TODO: VerifyRootsClosure extends OopsInGenClosure so that we can
-//       pass it as the perm_blk to SharedHeap::process_strong_roots.
-//       When process_strong_roots stop calling perm_blk->younger_refs_iterate
-//       we can change this closure to extend the simpler OopClosure.
-class VerifyRootsClosure: public OopsInGenClosure {
-private:
-  G1CollectedHeap* _g1h;
-  VerifyOption     _vo;
-  bool             _failures;
-public:
-  // _vo == UsePrevMarking -> use "prev" marking information,
-  // _vo == UseNextMarking -> use "next" marking information,
-  // _vo == UseMarkWord    -> use mark word from object header.
-  VerifyRootsClosure(VerifyOption vo) :
-    _g1h(G1CollectedHeap::heap()),
-    _vo(vo),
-    _failures(false) { }
-
-  bool failures() { return _failures; }
-
-  template <class T> void do_oop_nv(T* p) {
-    T heap_oop = oopDesc::load_heap_oop(p);
-    if (!oopDesc::is_null(heap_oop)) {
-      oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
-      if (_g1h->is_obj_dead_cond(obj, _vo)) {
-        gclog_or_tty->print_cr("Root location "PTR_FORMAT" "
-                              "points to dead obj "PTR_FORMAT, p, (void*) obj);
-        if (_vo == VerifyOption_G1UseMarkWord) {
-          gclog_or_tty->print_cr("  Mark word: "PTR_FORMAT, (void*)(obj->mark()));
-        }
-        obj->print_on(gclog_or_tty);
-        _failures = true;
-      }
-    }
-  }
-
-  void do_oop(oop* p)       { do_oop_nv(p); }
-  void do_oop(narrowOop* p) { do_oop_nv(p); }
-};
-
-// This is the task used for parallel heap verification.
+// This is the task used for parallel verification of the heap regions
 
 class G1ParVerifyTask: public AbstractGangTask {
 private:
@@ -3344,20 +3402,15 @@
   }
 };
 
-void G1CollectedHeap::verify(bool silent) {
-  verify(silent, VerifyOption_G1UsePrevMarking);
-}
-
-void G1CollectedHeap::verify(bool silent,
-                             VerifyOption vo) {
+void G1CollectedHeap::verify(bool silent, VerifyOption vo) {
   if (SafepointSynchronize::is_at_safepoint()) {
+    assert(Thread::current()->is_VM_thread(),
+           "Expected to be executed serially by the VM thread at this point");
+
     if (!silent) { gclog_or_tty->print("Roots "); }
     VerifyRootsClosure rootsCl(vo);
-
-    assert(Thread::current()->is_VM_thread(),
-           "Expected to be executed serially by the VM thread at this point");
-
-    CodeBlobToOopClosure blobsCl(&rootsCl, /*do_marking=*/ false);
+    G1VerifyCodeRootOopClosure codeRootsCl(this, &rootsCl, vo);
+    G1VerifyCodeRootBlobClosure blobsCl(&codeRootsCl);
     VerifyKlassClosure klassCl(this, &rootsCl);
 
     // We apply the relevant closures to all the oops in the
@@ -3376,7 +3429,7 @@
                          &klassCl
                          );
 
-    bool failures = rootsCl.failures();
+    bool failures = rootsCl.failures() || codeRootsCl.failures();
 
     if (vo != VerifyOption_G1UseMarkWord) {
       // If we're verifying during a full GC then the region sets
@@ -3445,6 +3498,34 @@
   }
 }
 
+void G1CollectedHeap::verify(bool silent) {
+  verify(silent, VerifyOption_G1UsePrevMarking);
+}
+
+double G1CollectedHeap::verify(bool guard, const char* msg) {
+  double verify_time_ms = 0.0;
+
+  if (guard && total_collections() >= VerifyGCStartAt) {
+    double verify_start = os::elapsedTime();
+    HandleMark hm;  // Discard invalid handles created during verification
+    prepare_for_verify();
+    Universe::verify(VerifyOption_G1UsePrevMarking, msg);
+    verify_time_ms = (os::elapsedTime() - verify_start) * 1000;
+  }
+
+  return verify_time_ms;
+}
+
+void G1CollectedHeap::verify_before_gc() {
+  double verify_time_ms = verify(VerifyBeforeGC, " VerifyBeforeGC:");
+  g1_policy()->phase_times()->record_verify_before_time_ms(verify_time_ms);
+}
+
+void G1CollectedHeap::verify_after_gc() {
+  double verify_time_ms = verify(VerifyAfterGC, " VerifyAfterGC:");
+  g1_policy()->phase_times()->record_verify_after_time_ms(verify_time_ms);
+}
+
 class PrintRegionClosure: public HeapRegionClosure {
   outputStream* _st;
 public:
@@ -3866,8 +3947,9 @@
       append_secondary_free_list_if_not_empty_with_lock();
     }
 
-    assert(check_young_list_well_formed(),
-      "young list should be well formed");
+    assert(check_young_list_well_formed(), "young list should be well formed");
+    assert(check_heap_region_claim_values(HeapRegion::InitialClaimValue),
+           "sanity check");
 
     // Don't dynamically change the number of GC threads this early.  A value of
     // 0 is used to indicate serial work.  When parallel work is done,
@@ -4987,7 +5069,11 @@
 
       G1ParPushHeapRSClosure          push_heap_rs_cl(_g1h, &pss);
 
-      int so = SharedHeap::SO_AllClasses | SharedHeap::SO_Strings | SharedHeap::SO_CodeCache;
+      // Don't scan the scavengable methods in the code cache as part
+      // of strong root scanning. The code roots that point into a
+      // region in the collection set are scanned when we scan the
+      // region's RSet.
+      int so = SharedHeap::SO_AllClasses | SharedHeap::SO_Strings;
 
       pss.start_strong_roots();
       _g1h->g1_process_strong_roots(/* is scavenging */ true,
@@ -5029,67 +5115,6 @@
 
 // *** Common G1 Evacuation Stuff
 
-// Closures that support the filtering of CodeBlobs scanned during
-// external root scanning.
-
-// Closure applied to reference fields in code blobs (specifically nmethods)
-// to determine whether an nmethod contains references that point into
-// the collection set. Used as a predicate when walking code roots so
-// that only nmethods that point into the collection set are added to the
-// 'marked' list.
-
-class G1FilteredCodeBlobToOopClosure : public CodeBlobToOopClosure {
-
-  class G1PointsIntoCSOopClosure : public OopClosure {
-    G1CollectedHeap* _g1;
-    bool _points_into_cs;
-  public:
-    G1PointsIntoCSOopClosure(G1CollectedHeap* g1) :
-      _g1(g1), _points_into_cs(false) { }
-
-    bool points_into_cs() const { return _points_into_cs; }
-
-    template <class T>
-    void do_oop_nv(T* p) {
-      if (!_points_into_cs) {
-        T heap_oop = oopDesc::load_heap_oop(p);
-        if (!oopDesc::is_null(heap_oop) &&
-            _g1->in_cset_fast_test(oopDesc::decode_heap_oop_not_null(heap_oop))) {
-          _points_into_cs = true;
-        }
-      }
-    }
-
-    virtual void do_oop(oop* p)        { do_oop_nv(p); }
-    virtual void do_oop(narrowOop* p)  { do_oop_nv(p); }
-  };
-
-  G1CollectedHeap* _g1;
-
-public:
-  G1FilteredCodeBlobToOopClosure(G1CollectedHeap* g1, OopClosure* cl) :
-    CodeBlobToOopClosure(cl, true), _g1(g1) { }
-
-  virtual void do_code_blob(CodeBlob* cb) {
-    nmethod* nm = cb->as_nmethod_or_null();
-    if (nm != NULL && !(nm->test_oops_do_mark())) {
-      G1PointsIntoCSOopClosure predicate_cl(_g1);
-      nm->oops_do(&predicate_cl);
-
-      if (predicate_cl.points_into_cs()) {
-        // At least one of the reference fields or the oop relocations
-        // in the nmethod points into the collection set. We have to
-        // 'mark' this nmethod.
-        // Note: Revisit the following if CodeBlobToOopClosure::do_code_blob()
-        // or MarkingCodeBlobClosure::do_code_blob() change.
-        if (!nm->test_set_oops_do_mark()) {
-          do_newly_marked_nmethod(nm);
-        }
-      }
-    }
-  }
-};
-
 // This method is run in a GC worker.
 
 void
@@ -5107,9 +5132,10 @@
 
   BufferingOopClosure buf_scan_non_heap_roots(scan_non_heap_roots);
 
-  // Walk the code cache w/o buffering, because StarTask cannot handle
-  // unaligned oop locations.
-  G1FilteredCodeBlobToOopClosure eager_scan_code_roots(this, scan_non_heap_roots);
+  assert(so & SO_CodeCache || scan_rs != NULL, "must scan code roots somehow");
+  // Walk the code cache/strong code roots w/o buffering, because StarTask
+  // cannot handle unaligned oop locations.
+  CodeBlobToOopClosure eager_scan_code_roots(scan_non_heap_roots, true /* do_marking */);
 
   process_strong_roots(false, // no scoping; this is parallel code
                        is_scavenging, so,
@@ -5154,9 +5180,22 @@
   }
   g1_policy()->phase_times()->record_satb_filtering_time(worker_i, satb_filtering_ms);
 
+  // If this is an initial mark pause, and we're not scanning
+  // the entire code cache, we need to mark the oops in the
+  // strong code root lists for the regions that are not in
+  // the collection set.
+  // Note all threads participate in this set of root tasks.
+  double mark_strong_code_roots_ms = 0.0;
+  if (g1_policy()->during_initial_mark_pause() && !(so & SO_CodeCache)) {
+    double mark_strong_roots_start = os::elapsedTime();
+    mark_strong_code_roots(worker_i);
+    mark_strong_code_roots_ms = (os::elapsedTime() - mark_strong_roots_start) * 1000.0;
+  }
+  g1_policy()->phase_times()->record_strong_code_root_mark_time(worker_i, mark_strong_code_roots_ms);
+
   // Now scan the complement of the collection set.
   if (scan_rs != NULL) {
-    g1_rem_set()->oops_into_collection_set_do(scan_rs, worker_i);
+    g1_rem_set()->oops_into_collection_set_do(scan_rs, &eager_scan_code_roots, worker_i);
   }
   _process_strong_tasks->all_tasks_completed();
 }
@@ -5774,9 +5813,6 @@
   process_discovered_references(n_workers);
 
   // Weak root processing.
-  // Note: when JSR 292 is enabled and code blobs can contain
-  // non-perm oops then we will need to process the code blobs
-  // here too.
   {
     G1STWIsAliveClosure is_alive(this);
     G1KeepAliveClosure keep_alive(this);
@@ -5792,6 +5828,17 @@
   hot_card_cache->reset_hot_cache();
   hot_card_cache->set_use_cache(true);
 
+  // Migrate the strong code roots attached to each region in
+  // the collection set. Ideally we would like to do this
+  // after we have finished the scanning/evacuation of the
+  // strong code roots for a particular heap region.
+  migrate_strong_code_roots();
+
+  if (g1_policy()->during_initial_mark_pause()) {
+    // Reset the claim values set during marking the strong code roots
+    reset_heap_region_claim_values();
+  }
+
   finalize_for_evac_failure();
 
   if (evacuation_failed()) {
@@ -6588,3 +6635,208 @@
   _humongous_set.verify_end();
   _free_list.verify_end();
 }
+
+// Optimized nmethod scanning
+
+class RegisterNMethodOopClosure: public OopClosure {
+  G1CollectedHeap* _g1h;
+  nmethod* _nm;
+
+  template <class T> void do_oop_work(T* p) {
+    T heap_oop = oopDesc::load_heap_oop(p);
+    if (!oopDesc::is_null(heap_oop)) {
+      oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
+      HeapRegion* hr = _g1h->heap_region_containing(obj);
+      assert(!hr->isHumongous(), "code root in humongous region?");
+
+      // HeapRegion::add_strong_code_root() avoids adding duplicate
+      // entries but having duplicates is  OK since we "mark" nmethods
+      // as visited when we scan the strong code root lists during the GC.
+      hr->add_strong_code_root(_nm);
+      assert(hr->rem_set()->strong_code_roots_list_contains(_nm), "add failed?");
+    }
+  }
+
+public:
+  RegisterNMethodOopClosure(G1CollectedHeap* g1h, nmethod* nm) :
+    _g1h(g1h), _nm(nm) {}
+
+  void do_oop(oop* p)       { do_oop_work(p); }
+  void do_oop(narrowOop* p) { do_oop_work(p); }
+};
+
+class UnregisterNMethodOopClosure: public OopClosure {
+  G1CollectedHeap* _g1h;
+  nmethod* _nm;
+
+  template <class T> void do_oop_work(T* p) {
+    T heap_oop = oopDesc::load_heap_oop(p);
+    if (!oopDesc::is_null(heap_oop)) {
+      oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
+      HeapRegion* hr = _g1h->heap_region_containing(obj);
+      assert(!hr->isHumongous(), "code root in humongous region?");
+      hr->remove_strong_code_root(_nm);
+      assert(!hr->rem_set()->strong_code_roots_list_contains(_nm), "remove failed?");
+    }
+  }
+
+public:
+  UnregisterNMethodOopClosure(G1CollectedHeap* g1h, nmethod* nm) :
+    _g1h(g1h), _nm(nm) {}
+
+  void do_oop(oop* p)       { do_oop_work(p); }
+  void do_oop(narrowOop* p) { do_oop_work(p); }
+};
+
+void G1CollectedHeap::register_nmethod(nmethod* nm) {
+  CollectedHeap::register_nmethod(nm);
+
+  guarantee(nm != NULL, "sanity");
+  RegisterNMethodOopClosure reg_cl(this, nm);
+  nm->oops_do(&reg_cl);
+}
+
+void G1CollectedHeap::unregister_nmethod(nmethod* nm) {
+  CollectedHeap::unregister_nmethod(nm);
+
+  guarantee(nm != NULL, "sanity");
+  UnregisterNMethodOopClosure reg_cl(this, nm);
+  nm->oops_do(&reg_cl, true);
+}
+
+class MigrateCodeRootsHeapRegionClosure: public HeapRegionClosure {
+public:
+  bool doHeapRegion(HeapRegion *hr) {
+    assert(!hr->isHumongous(), "humongous region in collection set?");
+    hr->migrate_strong_code_roots();
+    return false;
+  }
+};
+
+void G1CollectedHeap::migrate_strong_code_roots() {
+  MigrateCodeRootsHeapRegionClosure cl;
+  double migrate_start = os::elapsedTime();
+  collection_set_iterate(&cl);
+  double migration_time_ms = (os::elapsedTime() - migrate_start) * 1000.0;
+  g1_policy()->phase_times()->record_strong_code_root_migration_time(migration_time_ms);
+}
+
+// Mark all the code roots that point into regions *not* in the
+// collection set.
+//
+// Note we do not want to use a "marking" CodeBlobToOopClosure while
+// walking the the code roots lists of regions not in the collection
+// set. Suppose we have an nmethod (M) that points to objects in two
+// separate regions - one in the collection set (R1) and one not (R2).
+// Using a "marking" CodeBlobToOopClosure here would result in "marking"
+// nmethod M when walking the code roots for R1. When we come to scan
+// the code roots for R2, we would see that M is already marked and it
+// would be skipped and the objects in R2 that are referenced from M
+// would not be evacuated.
+
+class MarkStrongCodeRootCodeBlobClosure: public CodeBlobClosure {
+
+  class MarkStrongCodeRootOopClosure: public OopClosure {
+    ConcurrentMark* _cm;
+    HeapRegion* _hr;
+    uint _worker_id;
+
+    template <class T> void do_oop_work(T* p) {
+      T heap_oop = oopDesc::load_heap_oop(p);
+      if (!oopDesc::is_null(heap_oop)) {
+        oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
+        // Only mark objects in the region (which is assumed
+        // to be not in the collection set).
+        if (_hr->is_in(obj)) {
+          _cm->grayRoot(obj, (size_t) obj->size(), _worker_id);
+        }
+      }
+    }
+
+  public:
+    MarkStrongCodeRootOopClosure(ConcurrentMark* cm, HeapRegion* hr, uint worker_id) :
+      _cm(cm), _hr(hr), _worker_id(worker_id) {
+      assert(!_hr->in_collection_set(), "sanity");
+    }
+
+    void do_oop(narrowOop* p) { do_oop_work(p); }
+    void do_oop(oop* p)       { do_oop_work(p); }
+  };
+
+  MarkStrongCodeRootOopClosure _oop_cl;
+
+public:
+  MarkStrongCodeRootCodeBlobClosure(ConcurrentMark* cm, HeapRegion* hr, uint worker_id):
+    _oop_cl(cm, hr, worker_id) {}
+
+  void do_code_blob(CodeBlob* cb) {
+    nmethod* nm = (cb == NULL) ? NULL : cb->as_nmethod_or_null();
+    if (nm != NULL) {
+      nm->oops_do(&_oop_cl);
+    }
+  }
+};
+
+class MarkStrongCodeRootsHRClosure: public HeapRegionClosure {
+  G1CollectedHeap* _g1h;
+  uint _worker_id;
+
+public:
+  MarkStrongCodeRootsHRClosure(G1CollectedHeap* g1h, uint worker_id) :
+    _g1h(g1h), _worker_id(worker_id) {}
+
+  bool doHeapRegion(HeapRegion *hr) {
+    HeapRegionRemSet* hrrs = hr->rem_set();
+    if (hr->isHumongous()) {
+      // Code roots should never be attached to a humongous region
+      assert(hrrs->strong_code_roots_list_length() == 0, "sanity");
+      return false;
+    }
+
+    if (hr->in_collection_set()) {
+      // Don't mark code roots into regions in the collection set here.
+      // They will be marked when we scan them.
+      return false;
+    }
+
+    MarkStrongCodeRootCodeBlobClosure cb_cl(_g1h->concurrent_mark(), hr, _worker_id);
+    hr->strong_code_roots_do(&cb_cl);
+    return false;
+  }
+};
+
+void G1CollectedHeap::mark_strong_code_roots(uint worker_id) {
+  MarkStrongCodeRootsHRClosure cl(this, worker_id);
+  if (G1CollectedHeap::use_parallel_gc_threads()) {
+    heap_region_par_iterate_chunked(&cl,
+                                    worker_id,
+                                    workers()->active_workers(),
+                                    HeapRegion::ParMarkRootClaimValue);
+  } else {
+    heap_region_iterate(&cl);
+  }
+}
+
+class RebuildStrongCodeRootClosure: public CodeBlobClosure {
+  G1CollectedHeap* _g1h;
+
+public:
+  RebuildStrongCodeRootClosure(G1CollectedHeap* g1h) :
+    _g1h(g1h) {}
+
+  void do_code_blob(CodeBlob* cb) {
+    nmethod* nm = (cb != NULL) ? cb->as_nmethod_or_null() : NULL;
+    if (nm == NULL) {
+      return;
+    }
+
+    if (ScavengeRootsInCode && nm->detect_scavenge_root_oops()) {
+      _g1h->register_nmethod(nm);
+    }
+  }
+};
+
+void G1CollectedHeap::rebuild_strong_code_roots() {
+  RebuildStrongCodeRootClosure blob_cl(this);
+  CodeCache::blobs_do(&blob_cl);
+}
--- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -46,6 +46,7 @@
 // may combine concurrent marking with parallel, incremental compaction of
 // heap subsets that will yield large amounts of garbage.
 
+// Forward declarations
 class HeapRegion;
 class HRRSCleanupTask;
 class GenerationSpec;
@@ -69,6 +70,7 @@
 class G1NewTracer;
 class G1OldTracer;
 class EvacuationFailedInfo;
+class nmethod;
 
 typedef OverflowTaskQueue<StarTask, mtGC>         RefToScanQueue;
 typedef GenericTaskQueueSet<RefToScanQueue, mtGC> RefToScanQueueSet;
@@ -163,18 +165,6 @@
     : G1AllocRegion("Mutator Alloc Region", false /* bot_updates */) { }
 };
 
-// The G1 STW is alive closure.
-// An instance is embedded into the G1CH and used as the
-// (optional) _is_alive_non_header closure in the STW
-// reference processor. It is also extensively used during
-// reference processing during STW evacuation pauses.
-class G1STWIsAliveClosure: public BoolObjectClosure {
-  G1CollectedHeap* _g1;
-public:
-  G1STWIsAliveClosure(G1CollectedHeap* g1) : _g1(g1) {}
-  bool do_object_b(oop p);
-};
-
 class SurvivorGCAllocRegion : public G1AllocRegion {
 protected:
   virtual HeapRegion* allocate_new_region(size_t word_size, bool force);
@@ -193,6 +183,18 @@
   : G1AllocRegion("Old GC Alloc Region", true /* bot_updates */) { }
 };
 
+// The G1 STW is alive closure.
+// An instance is embedded into the G1CH and used as the
+// (optional) _is_alive_non_header closure in the STW
+// reference processor. It is also extensively used during
+// reference processing during STW evacuation pauses.
+class G1STWIsAliveClosure: public BoolObjectClosure {
+  G1CollectedHeap* _g1;
+public:
+  G1STWIsAliveClosure(G1CollectedHeap* g1) : _g1(g1) {}
+  bool do_object_b(oop p);
+};
+
 class RefineCardTableEntryClosure;
 
 class G1CollectedHeap : public SharedHeap {
@@ -1549,42 +1551,6 @@
 
   virtual jlong millis_since_last_gc();
 
-  // Perform any cleanup actions necessary before allowing a verification.
-  virtual void prepare_for_verify();
-
-  // Perform verification.
-
-  // vo == UsePrevMarking  -> use "prev" marking information,
-  // vo == UseNextMarking -> use "next" marking information
-  // vo == UseMarkWord    -> use the mark word in the object header
-  //
-  // NOTE: Only the "prev" marking information is guaranteed to be
-  // consistent most of the time, so most calls to this should use
-  // vo == UsePrevMarking.
-  // Currently, there is only one case where this is called with
-  // vo == UseNextMarking, which is to verify the "next" marking
-  // information at the end of remark.
-  // Currently there is only one place where this is called with
-  // vo == UseMarkWord, which is to verify the marking during a
-  // full GC.
-  void verify(bool silent, VerifyOption vo);
-
-  // Override; it uses the "prev" marking information
-  virtual void verify(bool silent);
-
-  virtual void print_on(outputStream* st) const;
-  virtual void print_extended_on(outputStream* st) const;
-  virtual void print_on_error(outputStream* st) const;
-
-  virtual void print_gc_threads_on(outputStream* st) const;
-  virtual void gc_threads_do(ThreadClosure* tc) const;
-
-  // Override
-  void print_tracing_info() const;
-
-  // The following two methods are helpful for debugging RSet issues.
-  void print_cset_rsets() PRODUCT_RETURN;
-  void print_all_rsets() PRODUCT_RETURN;
 
   // Convenience function to be used in situations where the heap type can be
   // asserted to be this type.
@@ -1661,13 +1627,86 @@
     else return is_obj_ill(obj, hr);
   }
 
+  bool allocated_since_marking(oop obj, HeapRegion* hr, VerifyOption vo);
+  HeapWord* top_at_mark_start(HeapRegion* hr, VerifyOption vo);
+  bool is_marked(oop obj, VerifyOption vo);
+  const char* top_at_mark_start_str(VerifyOption vo);
+
+  ConcurrentMark* concurrent_mark() const { return _cm; }
+
+  // Refinement
+
+  ConcurrentG1Refine* concurrent_g1_refine() const { return _cg1r; }
+
+  // The dirty cards region list is used to record a subset of regions
+  // whose cards need clearing. The list if populated during the
+  // remembered set scanning and drained during the card table
+  // cleanup. Although the methods are reentrant, population/draining
+  // phases must not overlap. For synchronization purposes the last
+  // element on the list points to itself.
+  HeapRegion* _dirty_cards_region_list;
+  void push_dirty_cards_region(HeapRegion* hr);
+  HeapRegion* pop_dirty_cards_region();
+
+  // Optimized nmethod scanning support routines
+
+  // Register the given nmethod with the G1 heap
+  virtual void register_nmethod(nmethod* nm);
+
+  // Unregister the given nmethod from the G1 heap
+  virtual void unregister_nmethod(nmethod* nm);
+
+  // Migrate the nmethods in the code root lists of the regions
+  // in the collection set to regions in to-space. In the event
+  // of an evacuation failure, nmethods that reference objects
+  // that were not successfullly evacuated are not migrated.
+  void migrate_strong_code_roots();
+
+  // During an initial mark pause, mark all the code roots that
+  // point into regions *not* in the collection set.
+  void mark_strong_code_roots(uint worker_id);
+
+  // Rebuild the stong code root lists for each region
+  // after a full GC
+  void rebuild_strong_code_roots();
+
+  // Verification
+
+  // The following is just to alert the verification code
+  // that a full collection has occurred and that the
+  // remembered sets are no longer up to date.
+  bool _full_collection;
+  void set_full_collection() { _full_collection = true;}
+  void clear_full_collection() {_full_collection = false;}
+  bool full_collection() {return _full_collection;}
+
+  // Perform any cleanup actions necessary before allowing a verification.
+  virtual void prepare_for_verify();
+
+  // Perform verification.
+
+  // vo == UsePrevMarking  -> use "prev" marking information,
+  // vo == UseNextMarking -> use "next" marking information
+  // vo == UseMarkWord    -> use the mark word in the object header
+  //
+  // NOTE: Only the "prev" marking information is guaranteed to be
+  // consistent most of the time, so most calls to this should use
+  // vo == UsePrevMarking.
+  // Currently, there is only one case where this is called with
+  // vo == UseNextMarking, which is to verify the "next" marking
+  // information at the end of remark.
+  // Currently there is only one place where this is called with
+  // vo == UseMarkWord, which is to verify the marking during a
+  // full GC.
+  void verify(bool silent, VerifyOption vo);
+
+  // Override; it uses the "prev" marking information
+  virtual void verify(bool silent);
+
   // The methods below are here for convenience and dispatch the
   // appropriate method depending on value of the given VerifyOption
-  // parameter. The options for that parameter are:
-  //
-  // vo == UsePrevMarking -> use "prev" marking information,
-  // vo == UseNextMarking -> use "next" marking information,
-  // vo == UseMarkWord    -> use mark word from object header
+  // parameter. The values for that parameter, and their meanings,
+  // are the same as those above.
 
   bool is_obj_dead_cond(const oop obj,
                         const HeapRegion* hr,
@@ -1692,31 +1731,21 @@
     return false; // keep some compilers happy
   }
 
-  bool allocated_since_marking(oop obj, HeapRegion* hr, VerifyOption vo);
-  HeapWord* top_at_mark_start(HeapRegion* hr, VerifyOption vo);
-  bool is_marked(oop obj, VerifyOption vo);
-  const char* top_at_mark_start_str(VerifyOption vo);
+  // Printing
 
-  // The following is just to alert the verification code
-  // that a full collection has occurred and that the
-  // remembered sets are no longer up to date.
-  bool _full_collection;
-  void set_full_collection() { _full_collection = true;}
-  void clear_full_collection() {_full_collection = false;}
-  bool full_collection() {return _full_collection;}
+  virtual void print_on(outputStream* st) const;
+  virtual void print_extended_on(outputStream* st) const;
+  virtual void print_on_error(outputStream* st) const;
 
-  ConcurrentMark* concurrent_mark() const { return _cm; }
-  ConcurrentG1Refine* concurrent_g1_refine() const { return _cg1r; }
+  virtual void print_gc_threads_on(outputStream* st) const;
+  virtual void gc_threads_do(ThreadClosure* tc) const;
 
-  // The dirty cards region list is used to record a subset of regions
-  // whose cards need clearing. The list if populated during the
-  // remembered set scanning and drained during the card table
-  // cleanup. Although the methods are reentrant, population/draining
-  // phases must not overlap. For synchronization purposes the last
-  // element on the list points to itself.
-  HeapRegion* _dirty_cards_region_list;
-  void push_dirty_cards_region(HeapRegion* hr);
-  HeapRegion* pop_dirty_cards_region();
+  // Override
+  void print_tracing_info() const;
+
+  // The following two methods are helpful for debugging RSet issues.
+  void print_cset_rsets() PRODUCT_RETURN;
+  void print_all_rsets() PRODUCT_RETURN;
 
 public:
   void stop_conc_gc_threads();
--- a/src/share/vm/gc_implementation/g1/g1GCPhaseTimes.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/gc_implementation/g1/g1GCPhaseTimes.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -161,6 +161,8 @@
   _last_update_rs_times_ms(_max_gc_threads, "%.1lf"),
   _last_update_rs_processed_buffers(_max_gc_threads, "%d"),
   _last_scan_rs_times_ms(_max_gc_threads, "%.1lf"),
+  _last_strong_code_root_scan_times_ms(_max_gc_threads, "%.1lf"),
+  _last_strong_code_root_mark_times_ms(_max_gc_threads, "%.1lf"),
   _last_obj_copy_times_ms(_max_gc_threads, "%.1lf"),
   _last_termination_times_ms(_max_gc_threads, "%.1lf"),
   _last_termination_attempts(_max_gc_threads, SIZE_FORMAT),
@@ -182,6 +184,8 @@
   _last_update_rs_times_ms.reset();
   _last_update_rs_processed_buffers.reset();
   _last_scan_rs_times_ms.reset();
+  _last_strong_code_root_scan_times_ms.reset();
+  _last_strong_code_root_mark_times_ms.reset();
   _last_obj_copy_times_ms.reset();
   _last_termination_times_ms.reset();
   _last_termination_attempts.reset();
@@ -197,6 +201,8 @@
   _last_update_rs_times_ms.verify();
   _last_update_rs_processed_buffers.verify();
   _last_scan_rs_times_ms.verify();
+  _last_strong_code_root_scan_times_ms.verify();
+  _last_strong_code_root_mark_times_ms.verify();
   _last_obj_copy_times_ms.verify();
   _last_termination_times_ms.verify();
   _last_termination_attempts.verify();
@@ -210,6 +216,8 @@
                                _last_satb_filtering_times_ms.get(i) +
                                _last_update_rs_times_ms.get(i) +
                                _last_scan_rs_times_ms.get(i) +
+                               _last_strong_code_root_scan_times_ms.get(i) +
+                               _last_strong_code_root_mark_times_ms.get(i) +
                                _last_obj_copy_times_ms.get(i) +
                                _last_termination_times_ms.get(i);
 
@@ -239,6 +247,9 @@
     // Now subtract the time taken to fix up roots in generated code
     misc_time_ms += _cur_collection_code_root_fixup_time_ms;
 
+    // Strong code root migration time
+    misc_time_ms += _cur_strong_code_root_migration_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;
@@ -257,9 +268,13 @@
     if (_last_satb_filtering_times_ms.sum() > 0.0) {
       _last_satb_filtering_times_ms.print(2, "SATB Filtering (ms)");
     }
+    if (_last_strong_code_root_mark_times_ms.sum() > 0.0) {
+     _last_strong_code_root_mark_times_ms.print(2, "Code Root Marking (ms)");
+    }
     _last_update_rs_times_ms.print(2, "Update RS (ms)");
       _last_update_rs_processed_buffers.print(3, "Processed Buffers");
     _last_scan_rs_times_ms.print(2, "Scan RS (ms)");
+    _last_strong_code_root_scan_times_ms.print(2, "Code Root Scanning (ms)");
     _last_obj_copy_times_ms.print(2, "Object Copy (ms)");
     _last_termination_times_ms.print(2, "Termination (ms)");
     if (G1Log::finest()) {
@@ -273,12 +288,17 @@
     if (_last_satb_filtering_times_ms.sum() > 0.0) {
       _last_satb_filtering_times_ms.print(1, "SATB Filtering (ms)");
     }
+    if (_last_strong_code_root_mark_times_ms.sum() > 0.0) {
+      _last_strong_code_root_mark_times_ms.print(1, "Code Root Marking (ms)");
+    }
     _last_update_rs_times_ms.print(1, "Update RS (ms)");
       _last_update_rs_processed_buffers.print(2, "Processed Buffers");
     _last_scan_rs_times_ms.print(1, "Scan RS (ms)");
+    _last_strong_code_root_scan_times_ms.print(1, "Code Root Scanning (ms)");
     _last_obj_copy_times_ms.print(1, "Object Copy (ms)");
   }
   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, "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);
--- a/src/share/vm/gc_implementation/g1/g1GCPhaseTimes.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/gc_implementation/g1/g1GCPhaseTimes.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -119,6 +119,8 @@
   WorkerDataArray<double> _last_update_rs_times_ms;
   WorkerDataArray<int>    _last_update_rs_processed_buffers;
   WorkerDataArray<double> _last_scan_rs_times_ms;
+  WorkerDataArray<double> _last_strong_code_root_scan_times_ms;
+  WorkerDataArray<double> _last_strong_code_root_mark_times_ms;
   WorkerDataArray<double> _last_obj_copy_times_ms;
   WorkerDataArray<double> _last_termination_times_ms;
   WorkerDataArray<size_t> _last_termination_attempts;
@@ -128,6 +130,7 @@
 
   double _cur_collection_par_time_ms;
   double _cur_collection_code_root_fixup_time_ms;
+  double _cur_strong_code_root_migration_time_ms;
 
   double _cur_clear_ct_time_ms;
   double _cur_ref_proc_time_ms;
@@ -179,6 +182,14 @@
     _last_scan_rs_times_ms.set(worker_i, ms);
   }
 
+  void record_strong_code_root_scan_time(uint worker_i, double ms) {
+    _last_strong_code_root_scan_times_ms.set(worker_i, ms);
+  }
+
+  void record_strong_code_root_mark_time(uint worker_i, double ms) {
+    _last_strong_code_root_mark_times_ms.set(worker_i, ms);
+  }
+
   void record_obj_copy_time(uint worker_i, double ms) {
     _last_obj_copy_times_ms.set(worker_i, ms);
   }
@@ -208,6 +219,10 @@
     _cur_collection_code_root_fixup_time_ms = ms;
   }
 
+  void record_strong_code_root_migration_time(double ms) {
+    _cur_strong_code_root_migration_time_ms = ms;
+  }
+
   void record_ref_proc_time(double ms) {
     _cur_ref_proc_time_ms = ms;
   }
@@ -294,6 +309,14 @@
     return _last_scan_rs_times_ms.average();
   }
 
+  double average_last_strong_code_root_scan_time(){
+    return _last_strong_code_root_scan_times_ms.average();
+  }
+
+  double average_last_strong_code_root_mark_time(){
+    return _last_strong_code_root_mark_times_ms.average();
+  }
+
   double average_last_obj_copy_time() {
     return _last_obj_copy_times_ms.average();
   }
--- a/src/share/vm/gc_implementation/g1/g1MonitoringSupport.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/gc_implementation/g1/g1MonitoringSupport.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -262,6 +262,7 @@
     old_collection_counters()->update_all();
     young_collection_counters()->update_all();
     MetaspaceCounters::update_performance_counters();
+    CompressedClassSpaceCounters::update_performance_counters();
   }
 }
 
--- a/src/share/vm/gc_implementation/g1/g1RemSet.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/gc_implementation/g1/g1RemSet.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -104,15 +104,25 @@
 class ScanRSClosure : public HeapRegionClosure {
   size_t _cards_done, _cards;
   G1CollectedHeap* _g1h;
+
   OopsInHeapRegionClosure* _oc;
+  CodeBlobToOopClosure* _code_root_cl;
+
   G1BlockOffsetSharedArray* _bot_shared;
   CardTableModRefBS *_ct_bs;
-  int _worker_i;
-  int _block_size;
-  bool _try_claimed;
+
+  double _strong_code_root_scan_time_sec;
+  int    _worker_i;
+  int    _block_size;
+  bool   _try_claimed;
+
 public:
-  ScanRSClosure(OopsInHeapRegionClosure* oc, int worker_i) :
+  ScanRSClosure(OopsInHeapRegionClosure* oc,
+                CodeBlobToOopClosure* code_root_cl,
+                int worker_i) :
     _oc(oc),
+    _code_root_cl(code_root_cl),
+    _strong_code_root_scan_time_sec(0.0),
     _cards(0),
     _cards_done(0),
     _worker_i(worker_i),
@@ -160,6 +170,12 @@
                            card_start, card_start + G1BlockOffsetSharedArray::N_words);
   }
 
+  void scan_strong_code_roots(HeapRegion* r) {
+    double scan_start = os::elapsedTime();
+    r->strong_code_roots_do(_code_root_cl);
+    _strong_code_root_scan_time_sec += (os::elapsedTime() - scan_start);
+  }
+
   bool doHeapRegion(HeapRegion* r) {
     assert(r->in_collection_set(), "should only be called on elements of CS.");
     HeapRegionRemSet* hrrs = r->rem_set();
@@ -173,6 +189,7 @@
     //   _try_claimed || r->claim_iter()
     // is true: either we're supposed to work on claimed-but-not-complete
     // regions, or we successfully claimed the region.
+
     HeapRegionRemSetIterator iter(hrrs);
     size_t card_index;
 
@@ -205,30 +222,43 @@
       }
     }
     if (!_try_claimed) {
+      // Scan the strong code root list attached to the current region
+      scan_strong_code_roots(r);
+
       hrrs->set_iter_complete();
     }
     return false;
   }
+
+  double strong_code_root_scan_time_sec() {
+    return _strong_code_root_scan_time_sec;
+  }
+
   size_t cards_done() { return _cards_done;}
   size_t cards_looked_up() { return _cards;}
 };
 
-void G1RemSet::scanRS(OopsInHeapRegionClosure* oc, int worker_i) {
+void G1RemSet::scanRS(OopsInHeapRegionClosure* oc,
+                      CodeBlobToOopClosure* code_root_cl,
+                      int worker_i) {
   double rs_time_start = os::elapsedTime();
   HeapRegion *startRegion = _g1->start_cset_region_for_worker(worker_i);
 
-  ScanRSClosure scanRScl(oc, worker_i);
+  ScanRSClosure scanRScl(oc, code_root_cl, worker_i);
 
   _g1->collection_set_iterate_from(startRegion, &scanRScl);
   scanRScl.set_try_claimed();
   _g1->collection_set_iterate_from(startRegion, &scanRScl);
 
-  double scan_rs_time_sec = os::elapsedTime() - rs_time_start;
+  double scan_rs_time_sec = (os::elapsedTime() - rs_time_start)
+                            - scanRScl.strong_code_root_scan_time_sec();
 
-  assert( _cards_scanned != NULL, "invariant" );
+  assert(_cards_scanned != NULL, "invariant");
   _cards_scanned[worker_i] = scanRScl.cards_done();
 
   _g1p->phase_times()->record_scan_rs_time(worker_i, scan_rs_time_sec * 1000.0);
+  _g1p->phase_times()->record_strong_code_root_scan_time(worker_i,
+                                                         scanRScl.strong_code_root_scan_time_sec() * 1000.0);
 }
 
 // Closure used for updating RSets and recording references that
@@ -288,7 +318,8 @@
 }
 
 void G1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc,
-                                             int worker_i) {
+                                           CodeBlobToOopClosure* code_root_cl,
+                                           int worker_i) {
 #if CARD_REPEAT_HISTO
   ct_freq_update_histo_and_reset();
 #endif
@@ -328,7 +359,7 @@
     _g1p->phase_times()->record_update_rs_time(worker_i, 0.0);
   }
   if (G1UseParallelRSetScanning || (worker_i == 0)) {
-    scanRS(oc, worker_i);
+    scanRS(oc, code_root_cl, worker_i);
   } else {
     _g1p->phase_times()->record_scan_rs_time(worker_i, 0.0);
   }
--- a/src/share/vm/gc_implementation/g1/g1RemSet.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/gc_implementation/g1/g1RemSet.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -81,14 +81,23 @@
   G1RemSet(G1CollectedHeap* g1, CardTableModRefBS* ct_bs);
   ~G1RemSet();
 
-  // Invoke "blk->do_oop" on all pointers into the CS in objects in regions
-  // outside the CS (having invoked "blk->set_region" to set the "from"
-  // region correctly beforehand.) The "worker_i" param is for the
-  // parallel case where the number of the worker thread calling this
-  // function can be helpful in partitioning the work to be done. It
-  // should be the same as the "i" passed to the calling thread's
-  // work(i) function. In the sequential case this param will be ingored.
-  void oops_into_collection_set_do(OopsInHeapRegionClosure* blk, int worker_i);
+  // Invoke "blk->do_oop" on all pointers into the collection set
+  // from objects in regions outside the collection set (having
+  // invoked "blk->set_region" to set the "from" region correctly
+  // beforehand.)
+  //
+  // Invoke code_root_cl->do_code_blob on the unmarked nmethods
+  // on the strong code roots list for each region in the
+  // collection set.
+  //
+  // The "worker_i" param is for the parallel case where the id
+  // of the worker thread calling this function can be helpful in
+  // partitioning the work to be done. It should be the same as
+  // the "i" passed to the calling thread's work(i) function.
+  // In the sequential case this param will be ignored.
+  void oops_into_collection_set_do(OopsInHeapRegionClosure* blk,
+                                   CodeBlobToOopClosure* code_root_cl,
+                                   int worker_i);
 
   // Prepare for and cleanup after an oops_into_collection_set_do
   // call.  Must call each of these once before and after (in sequential
@@ -98,7 +107,10 @@
   void prepare_for_oops_into_collection_set_do();
   void cleanup_after_oops_into_collection_set_do();
 
-  void scanRS(OopsInHeapRegionClosure* oc, int worker_i);
+  void scanRS(OopsInHeapRegionClosure* oc,
+              CodeBlobToOopClosure* code_root_cl,
+              int worker_i);
+
   void updateRS(DirtyCardQueue* into_cset_dcq, int worker_i);
 
   CardTableModRefBS* ct_bs() { return _ct_bs; }
--- a/src/share/vm/gc_implementation/g1/g1RemSetSummary.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/gc_implementation/g1/g1RemSetSummary.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -127,32 +127,55 @@
 
 class HRRSStatsIter: public HeapRegionClosure {
   size_t _occupied;
-  size_t _total_mem_sz;
-  size_t _max_mem_sz;
-  HeapRegion* _max_mem_sz_region;
+
+  size_t _total_rs_mem_sz;
+  size_t _max_rs_mem_sz;
+  HeapRegion* _max_rs_mem_sz_region;
+
+  size_t _total_code_root_mem_sz;
+  size_t _max_code_root_mem_sz;
+  HeapRegion* _max_code_root_mem_sz_region;
 public:
   HRRSStatsIter() :
     _occupied(0),
-    _total_mem_sz(0),
-    _max_mem_sz(0),
-    _max_mem_sz_region(NULL)
+    _total_rs_mem_sz(0),
+    _max_rs_mem_sz(0),
+    _max_rs_mem_sz_region(NULL),
+    _total_code_root_mem_sz(0),
+    _max_code_root_mem_sz(0),
+    _max_code_root_mem_sz_region(NULL)
   {}
 
   bool doHeapRegion(HeapRegion* r) {
-    size_t mem_sz = r->rem_set()->mem_size();
-    if (mem_sz > _max_mem_sz) {
-      _max_mem_sz = mem_sz;
-      _max_mem_sz_region = r;
+    HeapRegionRemSet* hrrs = r->rem_set();
+
+    // HeapRegionRemSet::mem_size() includes the
+    // size of the strong code roots
+    size_t rs_mem_sz = hrrs->mem_size();
+    if (rs_mem_sz > _max_rs_mem_sz) {
+      _max_rs_mem_sz = rs_mem_sz;
+      _max_rs_mem_sz_region = r;
     }
-    _total_mem_sz += mem_sz;
-    size_t occ = r->rem_set()->occupied();
+    _total_rs_mem_sz += rs_mem_sz;
+
+    size_t code_root_mem_sz = hrrs->strong_code_roots_mem_size();
+    if (code_root_mem_sz > _max_code_root_mem_sz) {
+      _max_code_root_mem_sz = code_root_mem_sz;
+      _max_code_root_mem_sz_region = r;
+    }
+    _total_code_root_mem_sz += code_root_mem_sz;
+
+    size_t occ = hrrs->occupied();
     _occupied += occ;
     return false;
   }
-  size_t total_mem_sz() { return _total_mem_sz; }
-  size_t max_mem_sz() { return _max_mem_sz; }
+  size_t total_rs_mem_sz() { return _total_rs_mem_sz; }
+  size_t max_rs_mem_sz() { return _max_rs_mem_sz; }
+  HeapRegion* max_rs_mem_sz_region() { return _max_rs_mem_sz_region; }
+  size_t total_code_root_mem_sz() { return _total_code_root_mem_sz; }
+  size_t max_code_root_mem_sz() { return _max_code_root_mem_sz; }
+  HeapRegion* max_code_root_mem_sz_region() { return _max_code_root_mem_sz_region; }
   size_t occupied() { return _occupied; }
-  HeapRegion* max_mem_sz_region() { return _max_mem_sz_region; }
 };
 
 double calc_percentage(size_t numerator, size_t denominator) {
@@ -184,22 +207,33 @@
 
   HRRSStatsIter blk;
   G1CollectedHeap::heap()->heap_region_iterate(&blk);
+  // RemSet stats
   out->print_cr("  Total heap region rem set sizes = "SIZE_FORMAT"K."
                 "  Max = "SIZE_FORMAT"K.",
-                blk.total_mem_sz()/K, blk.max_mem_sz()/K);
+                blk.total_rs_mem_sz()/K, blk.max_rs_mem_sz()/K);
   out->print_cr("  Static structures = "SIZE_FORMAT"K,"
                 " free_lists = "SIZE_FORMAT"K.",
                 HeapRegionRemSet::static_mem_size() / K,
                 HeapRegionRemSet::fl_mem_size() / K);
   out->print_cr("    "SIZE_FORMAT" occupied cards represented.",
                 blk.occupied());
-  HeapRegion* max_mem_sz_region = blk.max_mem_sz_region();
-  HeapRegionRemSet* rem_set = max_mem_sz_region->rem_set();
+  HeapRegion* max_rs_mem_sz_region = blk.max_rs_mem_sz_region();
+  HeapRegionRemSet* max_rs_rem_set = max_rs_mem_sz_region->rem_set();
   out->print_cr("    Max size region = "HR_FORMAT", "
                 "size = "SIZE_FORMAT "K, occupied = "SIZE_FORMAT"K.",
-                HR_FORMAT_PARAMS(max_mem_sz_region),
-                (rem_set->mem_size() + K - 1)/K,
-                (rem_set->occupied() + K - 1)/K);
-
+                HR_FORMAT_PARAMS(max_rs_mem_sz_region),
+                (max_rs_rem_set->mem_size() + K - 1)/K,
+                (max_rs_rem_set->occupied() + K - 1)/K);
   out->print_cr("    Did %d coarsenings.", num_coarsenings());
+  // Strong code root stats
+  out->print_cr("  Total heap region code-root set sizes = "SIZE_FORMAT"K."
+                "  Max = "SIZE_FORMAT"K.",
+                blk.total_code_root_mem_sz()/K, blk.max_code_root_mem_sz()/K);
+  HeapRegion* max_code_root_mem_sz_region = blk.max_code_root_mem_sz_region();
+  HeapRegionRemSet* max_code_root_rem_set = max_code_root_mem_sz_region->rem_set();
+  out->print_cr("    Max size region = "HR_FORMAT", "
+                "size = "SIZE_FORMAT "K, num_elems = "SIZE_FORMAT".",
+                HR_FORMAT_PARAMS(max_code_root_mem_sz_region),
+                (max_code_root_rem_set->strong_code_roots_mem_size() + K - 1)/K,
+                (max_code_root_rem_set->strong_code_roots_list_length()));
 }
--- a/src/share/vm/gc_implementation/g1/g1_globals.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/gc_implementation/g1/g1_globals.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -319,7 +319,10 @@
                                                                             \
   diagnostic(bool, G1VerifyRSetsDuringFullGC, false,                        \
              "If true, perform verification of each heap region's "         \
-             "remembered set when verifying the heap during a full GC.")
+             "remembered set when verifying the heap during a full GC.")    \
+                                                                            \
+  diagnostic(bool, G1VerifyHeapRegionCodeRoots, false,                      \
+             "Verify the code root lists attached to each heap region.")
 
 G1_FLAGS(DECLARE_DEVELOPER_FLAG, DECLARE_PD_DEVELOPER_FLAG, DECLARE_PRODUCT_FLAG, DECLARE_PD_PRODUCT_FLAG, DECLARE_DIAGNOSTIC_FLAG, DECLARE_EXPERIMENTAL_FLAG, DECLARE_NOTPRODUCT_FLAG, DECLARE_MANAGEABLE_FLAG, DECLARE_PRODUCT_RW_FLAG)
 
--- a/src/share/vm/gc_implementation/g1/heapRegion.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/gc_implementation/g1/heapRegion.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -23,6 +23,7 @@
  */
 
 #include "precompiled.hpp"
+#include "code/nmethod.hpp"
 #include "gc_implementation/g1/g1BlockOffsetTable.inline.hpp"
 #include "gc_implementation/g1/g1CollectedHeap.inline.hpp"
 #include "gc_implementation/g1/g1OopClosures.inline.hpp"
@@ -50,144 +51,6 @@
                                                    OopClosure* oc) :
   _r_bottom(r->bottom()), _r_end(r->end()), _oc(oc) { }
 
-class VerifyLiveClosure: public OopClosure {
-private:
-  G1CollectedHeap* _g1h;
-  CardTableModRefBS* _bs;
-  oop _containing_obj;
-  bool _failures;
-  int _n_failures;
-  VerifyOption _vo;
-public:
-  // _vo == UsePrevMarking -> use "prev" marking information,
-  // _vo == UseNextMarking -> use "next" marking information,
-  // _vo == UseMarkWord    -> use mark word from object header.
-  VerifyLiveClosure(G1CollectedHeap* g1h, VerifyOption vo) :
-    _g1h(g1h), _bs(NULL), _containing_obj(NULL),
-    _failures(false), _n_failures(0), _vo(vo)
-  {
-    BarrierSet* bs = _g1h->barrier_set();
-    if (bs->is_a(BarrierSet::CardTableModRef))
-      _bs = (CardTableModRefBS*)bs;
-  }
-
-  void set_containing_obj(oop obj) {
-    _containing_obj = obj;
-  }
-
-  bool failures() { return _failures; }
-  int n_failures() { return _n_failures; }
-
-  virtual void do_oop(narrowOop* p) { do_oop_work(p); }
-  virtual void do_oop(      oop* p) { do_oop_work(p); }
-
-  void print_object(outputStream* out, oop obj) {
-#ifdef PRODUCT
-    Klass* k = obj->klass();
-    const char* class_name = InstanceKlass::cast(k)->external_name();
-    out->print_cr("class name %s", class_name);
-#else // PRODUCT
-    obj->print_on(out);
-#endif // PRODUCT
-  }
-
-  template <class T>
-  void do_oop_work(T* p) {
-    assert(_containing_obj != NULL, "Precondition");
-    assert(!_g1h->is_obj_dead_cond(_containing_obj, _vo),
-           "Precondition");
-    T heap_oop = oopDesc::load_heap_oop(p);
-    if (!oopDesc::is_null(heap_oop)) {
-      oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
-      bool failed = false;
-      if (!_g1h->is_in_closed_subset(obj) || _g1h->is_obj_dead_cond(obj, _vo)) {
-        MutexLockerEx x(ParGCRareEvent_lock,
-                        Mutex::_no_safepoint_check_flag);
-
-        if (!_failures) {
-          gclog_or_tty->print_cr("");
-          gclog_or_tty->print_cr("----------");
-        }
-        if (!_g1h->is_in_closed_subset(obj)) {
-          HeapRegion* from = _g1h->heap_region_containing((HeapWord*)p);
-          gclog_or_tty->print_cr("Field "PTR_FORMAT
-                                 " of live obj "PTR_FORMAT" in region "
-                                 "["PTR_FORMAT", "PTR_FORMAT")",
-                                 p, (void*) _containing_obj,
-                                 from->bottom(), from->end());
-          print_object(gclog_or_tty, _containing_obj);
-          gclog_or_tty->print_cr("points to obj "PTR_FORMAT" not in the heap",
-                                 (void*) obj);
-        } else {
-          HeapRegion* from = _g1h->heap_region_containing((HeapWord*)p);
-          HeapRegion* to   = _g1h->heap_region_containing((HeapWord*)obj);
-          gclog_or_tty->print_cr("Field "PTR_FORMAT
-                                 " of live obj "PTR_FORMAT" in region "
-                                 "["PTR_FORMAT", "PTR_FORMAT")",
-                                 p, (void*) _containing_obj,
-                                 from->bottom(), from->end());
-          print_object(gclog_or_tty, _containing_obj);
-          gclog_or_tty->print_cr("points to dead obj "PTR_FORMAT" in region "
-                                 "["PTR_FORMAT", "PTR_FORMAT")",
-                                 (void*) obj, to->bottom(), to->end());
-          print_object(gclog_or_tty, obj);
-        }
-        gclog_or_tty->print_cr("----------");
-        gclog_or_tty->flush();
-        _failures = true;
-        failed = true;
-        _n_failures++;
-      }
-
-      if (!_g1h->full_collection() || G1VerifyRSetsDuringFullGC) {
-        HeapRegion* from = _g1h->heap_region_containing((HeapWord*)p);
-        HeapRegion* to   = _g1h->heap_region_containing(obj);
-        if (from != NULL && to != NULL &&
-            from != to &&
-            !to->isHumongous()) {
-          jbyte cv_obj = *_bs->byte_for_const(_containing_obj);
-          jbyte cv_field = *_bs->byte_for_const(p);
-          const jbyte dirty = CardTableModRefBS::dirty_card_val();
-
-          bool is_bad = !(from->is_young()
-                          || to->rem_set()->contains_reference(p)
-                          || !G1HRRSFlushLogBuffersOnVerify && // buffers were not flushed
-                              (_containing_obj->is_objArray() ?
-                                  cv_field == dirty
-                               : cv_obj == dirty || cv_field == dirty));
-          if (is_bad) {
-            MutexLockerEx x(ParGCRareEvent_lock,
-                            Mutex::_no_safepoint_check_flag);
-
-            if (!_failures) {
-              gclog_or_tty->print_cr("");
-              gclog_or_tty->print_cr("----------");
-            }
-            gclog_or_tty->print_cr("Missing rem set entry:");
-            gclog_or_tty->print_cr("Field "PTR_FORMAT" "
-                                   "of obj "PTR_FORMAT", "
-                                   "in region "HR_FORMAT,
-                                   p, (void*) _containing_obj,
-                                   HR_FORMAT_PARAMS(from));
-            _containing_obj->print_on(gclog_or_tty);
-            gclog_or_tty->print_cr("points to obj "PTR_FORMAT" "
-                                   "in region "HR_FORMAT,
-                                   (void*) obj,
-                                   HR_FORMAT_PARAMS(to));
-            obj->print_on(gclog_or_tty);
-            gclog_or_tty->print_cr("Obj head CTE = %d, field CTE = %d.",
-                          cv_obj, cv_field);
-            gclog_or_tty->print_cr("----------");
-            gclog_or_tty->flush();
-            _failures = true;
-            if (!failed) _n_failures++;
-          }
-        }
-      }
-    }
-  }
-};
-
 template<class ClosureType>
 HeapWord* walk_mem_region_loop(ClosureType* cl, G1CollectedHeap* g1h,
                                HeapRegion* hr,
@@ -368,7 +231,7 @@
   if (!par) {
     // If this is parallel, this will be done later.
     HeapRegionRemSet* hrrs = rem_set();
-    if (hrrs != NULL) hrrs->clear();
+    hrrs->clear();
     _claimed = InitialClaimValue;
   }
   zero_marked_bytes();
@@ -505,6 +368,7 @@
     _rem_set(NULL), _recorded_rs_length(0), _predicted_elapsed_time_ms(0),
     _predicted_bytes_to_copy(0)
 {
+  _rem_set = new HeapRegionRemSet(sharedOffsetArray, this);
   _orig_end = mr.end();
   // Note that initialize() will set the start of the unmarked area of the
   // region.
@@ -512,8 +376,6 @@
   set_top(bottom());
   set_saved_mark();
 
-  _rem_set =  new HeapRegionRemSet(sharedOffsetArray, this);
-
   assert(HeapRegionRemSet::num_par_rem_sets() > 0, "Invariant.");
 }
 
@@ -733,6 +595,160 @@
   return NULL;
 }
 
+// Code roots support
+
+void HeapRegion::add_strong_code_root(nmethod* nm) {
+  HeapRegionRemSet* hrrs = rem_set();
+  hrrs->add_strong_code_root(nm);
+}
+
+void HeapRegion::remove_strong_code_root(nmethod* nm) {
+  HeapRegionRemSet* hrrs = rem_set();
+  hrrs->remove_strong_code_root(nm);
+}
+
+void HeapRegion::migrate_strong_code_roots() {
+  assert(in_collection_set(), "only collection set regions");
+  assert(!isHumongous(), "not humongous regions");
+
+  HeapRegionRemSet* hrrs = rem_set();
+  hrrs->migrate_strong_code_roots();
+}
+
+void HeapRegion::strong_code_roots_do(CodeBlobClosure* blk) const {
+  HeapRegionRemSet* hrrs = rem_set();
+  hrrs->strong_code_roots_do(blk);
+}
+
+class VerifyStrongCodeRootOopClosure: public OopClosure {
+  const HeapRegion* _hr;
+  nmethod* _nm;
+  bool _failures;
+  bool _has_oops_in_region;
+
+  template <class T> void do_oop_work(T* p) {
+    T heap_oop = oopDesc::load_heap_oop(p);
+    if (!oopDesc::is_null(heap_oop)) {
+      oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
+
+      // Note: not all the oops embedded in the nmethod are in the
+      // current region. We only look at those which are.
+      if (_hr->is_in(obj)) {
+        // Object is in the region. Check that its less than top
+        if (_hr->top() <= (HeapWord*)obj) {
+          // Object is above top
+          gclog_or_tty->print_cr("Object "PTR_FORMAT" in region "
+                                 "["PTR_FORMAT", "PTR_FORMAT") is above "
+                                 "top "PTR_FORMAT,
+                                 obj, _hr->bottom(), _hr->end(), _hr->top());
+          _failures = true;
+          return;
+        }
+        // Nmethod has at least one oop in the current region
+        _has_oops_in_region = true;
+      }
+    }
+  }
+
+public:
+  VerifyStrongCodeRootOopClosure(const HeapRegion* hr, nmethod* nm):
+    _hr(hr), _failures(false), _has_oops_in_region(false) {}
+
+  void do_oop(narrowOop* p) { do_oop_work(p); }
+  void do_oop(oop* p)       { do_oop_work(p); }
+
+  bool failures()           { return _failures; }
+  bool has_oops_in_region() { return _has_oops_in_region; }
+};
+
+class VerifyStrongCodeRootCodeBlobClosure: public CodeBlobClosure {
+  const HeapRegion* _hr;
+  bool _failures;
+public:
+  VerifyStrongCodeRootCodeBlobClosure(const HeapRegion* hr) :
+    _hr(hr), _failures(false) {}
+
+  void do_code_blob(CodeBlob* cb) {
+    nmethod* nm = (cb == NULL) ? NULL : cb->as_nmethod_or_null();
+    if (nm != NULL) {
+      // Verify that the nemthod is live
+      if (!nm->is_alive()) {
+        gclog_or_tty->print_cr("region ["PTR_FORMAT","PTR_FORMAT"] has dead nmethod "
+                               PTR_FORMAT" in its strong code roots",
+                               _hr->bottom(), _hr->end(), nm);
+        _failures = true;
+      } else {
+        VerifyStrongCodeRootOopClosure oop_cl(_hr, nm);
+        nm->oops_do(&oop_cl);
+        if (!oop_cl.has_oops_in_region()) {
+          gclog_or_tty->print_cr("region ["PTR_FORMAT","PTR_FORMAT"] has nmethod "
+                                 PTR_FORMAT" in its strong code roots "
+                                 "with no pointers into region",
+                                 _hr->bottom(), _hr->end(), nm);
+          _failures = true;
+        } else if (oop_cl.failures()) {
+          gclog_or_tty->print_cr("region ["PTR_FORMAT","PTR_FORMAT"] has other "
+                                 "failures for nmethod "PTR_FORMAT,
+                                 _hr->bottom(), _hr->end(), nm);
+          _failures = true;
+        }
+      }
+    }
+  }
+
+  bool failures()       { return _failures; }
+};
+
+void HeapRegion::verify_strong_code_roots(VerifyOption vo, bool* failures) const {
+  if (!G1VerifyHeapRegionCodeRoots) {
+    // We're not verifying code roots.
+    return;
+  }
+  if (vo == VerifyOption_G1UseMarkWord) {
+    // Marking verification during a full GC is performed after class
+    // unloading, code cache unloading, etc so the strong code roots
+    // attached to each heap region are in an inconsistent state. They won't
+    // be consistent until the strong code roots are rebuilt after the
+    // actual GC. Skip verifying the strong code roots in this particular
+    // time.
+    assert(VerifyDuringGC, "only way to get here");
+    return;
+  }
+
+  HeapRegionRemSet* hrrs = rem_set();
+  int strong_code_roots_length = hrrs->strong_code_roots_list_length();
+
+  // if this region is empty then there should be no entries
+  // on its strong code root list
+  if (is_empty()) {
+    if (strong_code_roots_length > 0) {
+      gclog_or_tty->print_cr("region ["PTR_FORMAT","PTR_FORMAT"] is empty "
+                             "but has "INT32_FORMAT" code root entries",
+                             bottom(), end(), strong_code_roots_length);
+      *failures = true;
+    }
+    return;
+  }
+
+  // An H-region should have an empty strong code root list
+  if (isHumongous()) {
+    if (strong_code_roots_length > 0) {
+      gclog_or_tty->print_cr("region ["PTR_FORMAT","PTR_FORMAT"] is humongous "
+                             "but has "INT32_FORMAT" code root entries",
+                             bottom(), end(), strong_code_roots_length);
+      *failures = true;
+    }
+    return;
+  }
+
+  VerifyStrongCodeRootCodeBlobClosure cb_cl(this);
+  strong_code_roots_do(&cb_cl);
+
+  if (cb_cl.failures()) {
+    *failures = true;
+  }
+}
+
 void HeapRegion::print() const { print_on(gclog_or_tty); }
 void HeapRegion::print_on(outputStream* st) const {
   if (isHumongous()) {
@@ -761,10 +777,143 @@
   G1OffsetTableContigSpace::print_on(st);
 }
 
-void HeapRegion::verify() const {
-  bool dummy = false;
-  verify(VerifyOption_G1UsePrevMarking, /* failures */ &dummy);
-}
+class VerifyLiveClosure: public OopClosure {
+private:
+  G1CollectedHeap* _g1h;
+  CardTableModRefBS* _bs;
+  oop _containing_obj;
+  bool _failures;
+  int _n_failures;
+  VerifyOption _vo;
+public:
+  // _vo == UsePrevMarking -> use "prev" marking information,
+  // _vo == UseNextMarking -> use "next" marking information,
+  // _vo == UseMarkWord    -> use mark word from object header.
+  VerifyLiveClosure(G1CollectedHeap* g1h, VerifyOption vo) :
+    _g1h(g1h), _bs(NULL), _containing_obj(NULL),
+    _failures(false), _n_failures(0), _vo(vo)
+  {
+    BarrierSet* bs = _g1h->barrier_set();
+    if (bs->is_a(BarrierSet::CardTableModRef))
+      _bs = (CardTableModRefBS*)bs;
+  }
+
+  void set_containing_obj(oop obj) {
+    _containing_obj = obj;
+  }
+
+  bool failures() { return _failures; }
+  int n_failures() { return _n_failures; }
+
+  virtual void do_oop(narrowOop* p) { do_oop_work(p); }
+  virtual void do_oop(      oop* p) { do_oop_work(p); }
+
+  void print_object(outputStream* out, oop obj) {
+#ifdef PRODUCT
+    Klass* k = obj->klass();
+    const char* class_name = InstanceKlass::cast(k)->external_name();
+    out->print_cr("class name %s", class_name);
+#else // PRODUCT
+    obj->print_on(out);
+#endif // PRODUCT
+  }
+
+  template <class T>
+  void do_oop_work(T* p) {
+    assert(_containing_obj != NULL, "Precondition");
+    assert(!_g1h->is_obj_dead_cond(_containing_obj, _vo),
+           "Precondition");
+    T heap_oop = oopDesc::load_heap_oop(p);
+    if (!oopDesc::is_null(heap_oop)) {
+      oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
+      bool failed = false;
+      if (!_g1h->is_in_closed_subset(obj) || _g1h->is_obj_dead_cond(obj, _vo)) {
+        MutexLockerEx x(ParGCRareEvent_lock,
+                        Mutex::_no_safepoint_check_flag);
+
+        if (!_failures) {
+          gclog_or_tty->print_cr("");
+          gclog_or_tty->print_cr("----------");
+        }
+        if (!_g1h->is_in_closed_subset(obj)) {
+          HeapRegion* from = _g1h->heap_region_containing((HeapWord*)p);
+          gclog_or_tty->print_cr("Field "PTR_FORMAT
+                                 " of live obj "PTR_FORMAT" in region "
+                                 "["PTR_FORMAT", "PTR_FORMAT")",
+                                 p, (void*) _containing_obj,
+                                 from->bottom(), from->end());
+          print_object(gclog_or_tty, _containing_obj);
+          gclog_or_tty->print_cr("points to obj "PTR_FORMAT" not in the heap",
+                                 (void*) obj);
+        } else {
+          HeapRegion* from = _g1h->heap_region_containing((HeapWord*)p);
+          HeapRegion* to   = _g1h->heap_region_containing((HeapWord*)obj);
+          gclog_or_tty->print_cr("Field "PTR_FORMAT
+                                 " of live obj "PTR_FORMAT" in region "
+                                 "["PTR_FORMAT", "PTR_FORMAT")",
+                                 p, (void*) _containing_obj,
+                                 from->bottom(), from->end());
+          print_object(gclog_or_tty, _containing_obj);
+          gclog_or_tty->print_cr("points to dead obj "PTR_FORMAT" in region "
+                                 "["PTR_FORMAT", "PTR_FORMAT")",
+                                 (void*) obj, to->bottom(), to->end());
+          print_object(gclog_or_tty, obj);
+        }
+        gclog_or_tty->print_cr("----------");
+        gclog_or_tty->flush();
+        _failures = true;
+        failed = true;
+        _n_failures++;
+      }
+
+      if (!_g1h->full_collection() || G1VerifyRSetsDuringFullGC) {
+        HeapRegion* from = _g1h->heap_region_containing((HeapWord*)p);
+        HeapRegion* to   = _g1h->heap_region_containing(obj);
+        if (from != NULL && to != NULL &&
+            from != to &&
+            !to->isHumongous()) {
+          jbyte cv_obj = *_bs->byte_for_const(_containing_obj);
+          jbyte cv_field = *_bs->byte_for_const(p);
+          const jbyte dirty = CardTableModRefBS::dirty_card_val();
+
+          bool is_bad = !(from->is_young()
+                          || to->rem_set()->contains_reference(p)
+                          || !G1HRRSFlushLogBuffersOnVerify && // buffers were not flushed
+                              (_containing_obj->is_objArray() ?
+                                  cv_field == dirty
+                               : cv_obj == dirty || cv_field == dirty));
+          if (is_bad) {
+            MutexLockerEx x(ParGCRareEvent_lock,
+                            Mutex::_no_safepoint_check_flag);
+
+            if (!_failures) {
+              gclog_or_tty->print_cr("");
+              gclog_or_tty->print_cr("----------");
+            }
+            gclog_or_tty->print_cr("Missing rem set entry:");
+            gclog_or_tty->print_cr("Field "PTR_FORMAT" "
+                                   "of obj "PTR_FORMAT", "
+                                   "in region "HR_FORMAT,
+                                   p, (void*) _containing_obj,
+                                   HR_FORMAT_PARAMS(from));
+            _containing_obj->print_on(gclog_or_tty);
+            gclog_or_tty->print_cr("points to obj "PTR_FORMAT" "
+                                   "in region "HR_FORMAT,
+                                   (void*) obj,
+                                   HR_FORMAT_PARAMS(to));
+            obj->print_on(gclog_or_tty);
+            gclog_or_tty->print_cr("Obj head CTE = %d, field CTE = %d.",
+                          cv_obj, cv_field);
+            gclog_or_tty->print_cr("----------");
+            gclog_or_tty->flush();
+            _failures = true;
+            if (!failed) _n_failures++;
+          }
+        }
+      }
+    }
+  }
+};
 
 // This really ought to be commoned up into OffsetTableContigSpace somehow.
 // We would need a mechanism to make that code skip dead objects.
@@ -904,6 +1053,13 @@
     *failures = true;
     return;
   }
+
+  verify_strong_code_roots(vo, failures);
+}
+
+void HeapRegion::verify() const {
+  bool dummy = false;
+  verify(VerifyOption_G1UsePrevMarking, /* failures */ &dummy);
 }
 
 // G1OffsetTableContigSpace code; copied from space.cpp.  Hope this can go
--- a/src/share/vm/gc_implementation/g1/heapRegion.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/gc_implementation/g1/heapRegion.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -52,6 +52,7 @@
 class HeapRegionRemSetIterator;
 class HeapRegion;
 class HeapRegionSetBase;
+class nmethod;
 
 #define HR_FORMAT "%u:(%s)["PTR_FORMAT","PTR_FORMAT","PTR_FORMAT"]"
 #define HR_FORMAT_PARAMS(_hr_) \
@@ -371,7 +372,8 @@
     RebuildRSClaimValue        = 5,
     ParEvacFailureClaimValue   = 6,
     AggregateCountClaimValue   = 7,
-    VerifyCountClaimValue      = 8
+    VerifyCountClaimValue      = 8,
+    ParMarkRootClaimValue      = 9
   };
 
   inline HeapWord* par_allocate_no_bot_updates(size_t word_size) {
@@ -796,6 +798,25 @@
 
   virtual void reset_after_compaction();
 
+  // Routines for managing a list of code roots (attached to the
+  // this region's RSet) that point into this heap region.
+  void add_strong_code_root(nmethod* nm);
+  void remove_strong_code_root(nmethod* nm);
+
+  // During a collection, migrate the successfully evacuated
+  // strong code roots that referenced into this region to the
+  // new regions that they now point into. Unsuccessfully
+  // evacuated code roots are not migrated.
+  void migrate_strong_code_roots();
+
+  // Applies blk->do_code_blob() to each of the entries in
+  // the strong code roots list for this region
+  void strong_code_roots_do(CodeBlobClosure* blk) const;
+
+  // Verify that the entries on the strong code root list for this
+  // region are live and include at least one pointer into this region.
+  void verify_strong_code_roots(VerifyOption vo, bool* failures) const;
+
   void print() const;
   void print_on(outputStream* st) const;
 
--- a/src/share/vm/gc_implementation/g1/heapRegionRemSet.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/gc_implementation/g1/heapRegionRemSet.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -33,6 +33,7 @@
 #include "oops/oop.inline.hpp"
 #include "utilities/bitMap.inline.hpp"
 #include "utilities/globalDefinitions.hpp"
+#include "utilities/growableArray.hpp"
 
 class PerRegionTable: public CHeapObj<mtGC> {
   friend class OtherRegionsTable;
@@ -849,7 +850,7 @@
 
 HeapRegionRemSet::HeapRegionRemSet(G1BlockOffsetSharedArray* bosa,
                                    HeapRegion* hr)
-  : _bosa(bosa), _other_regions(hr) {
+  : _bosa(bosa), _strong_code_roots_list(NULL), _other_regions(hr) {
   reset_for_par_iteration();
 }
 
@@ -908,6 +909,12 @@
 }
 
 void HeapRegionRemSet::clear() {
+  if (_strong_code_roots_list != NULL) {
+    delete _strong_code_roots_list;
+  }
+  _strong_code_roots_list = new (ResourceObj::C_HEAP, mtGC)
+                                GrowableArray<nmethod*>(10, 0, NULL, true);
+
   _other_regions.clear();
   assert(occupied() == 0, "Should be clear.");
   reset_for_par_iteration();
@@ -925,6 +932,121 @@
   _other_regions.scrub(ctbs, region_bm, card_bm);
 }
 
+
+// Code roots support
+
+void HeapRegionRemSet::add_strong_code_root(nmethod* nm) {
+  assert(nm != NULL, "sanity");
+  // Search for the code blob from the RHS to avoid
+  // duplicate entries as much as possible
+  if (_strong_code_roots_list->find_from_end(nm) < 0) {
+    // Code blob isn't already in the list
+    _strong_code_roots_list->push(nm);
+  }
+}
+
+void HeapRegionRemSet::remove_strong_code_root(nmethod* nm) {
+  assert(nm != NULL, "sanity");
+  int idx = _strong_code_roots_list->find(nm);
+  if (idx >= 0) {
+    _strong_code_roots_list->remove_at(idx);
+  }
+  // Check that there were no duplicates
+  guarantee(_strong_code_roots_list->find(nm) < 0, "duplicate entry found");
+}
+
+class NMethodMigrationOopClosure : public OopClosure {
+  G1CollectedHeap* _g1h;
+  HeapRegion* _from;
+  nmethod* _nm;
+
+  uint _num_self_forwarded;
+
+  template <class T> void do_oop_work(T* p) {
+    T heap_oop = oopDesc::load_heap_oop(p);
+    if (!oopDesc::is_null(heap_oop)) {
+      oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
+      if (_from->is_in(obj)) {
+        // Reference still points into the source region.
+        // Since roots are immediately evacuated this means that
+        // we must have self forwarded the object
+        assert(obj->is_forwarded(),
+               err_msg("code roots should be immediately evacuated. "
+                       "Ref: "PTR_FORMAT", "
+                       "Obj: "PTR_FORMAT", "
+                       "Region: "HR_FORMAT,
+                       p, (void*) obj, HR_FORMAT_PARAMS(_from)));
+        assert(obj->forwardee() == obj,
+               err_msg("not self forwarded? obj = "PTR_FORMAT, (void*)obj));
+
+        // The object has been self forwarded.
+        // Note, if we're during an initial mark pause, there is
+        // no need to explicitly mark object. It will be marked
+        // during the regular evacuation failure handling code.
+        _num_self_forwarded++;
+      } else {
+        // The reference points into a promotion or to-space region
+        HeapRegion* to = _g1h->heap_region_containing(obj);
+        to->rem_set()->add_strong_code_root(_nm);
+      }
+    }
+  }
+
+public:
+  NMethodMigrationOopClosure(G1CollectedHeap* g1h, HeapRegion* from, nmethod* nm):
+    _g1h(g1h), _from(from), _nm(nm), _num_self_forwarded(0) {}
+
+  void do_oop(narrowOop* p) { do_oop_work(p); }
+  void do_oop(oop* p)       { do_oop_work(p); }
+
+  uint retain() { return _num_self_forwarded > 0; }
+};
+
+void HeapRegionRemSet::migrate_strong_code_roots() {
+  assert(hr()->in_collection_set(), "only collection set regions");
+  assert(!hr()->isHumongous(), "not humongous regions");
+
+  ResourceMark rm;
+
+  // List of code blobs to retain for this region
+  GrowableArray<nmethod*> to_be_retained(10);
+  G1CollectedHeap* g1h = G1CollectedHeap::heap();
+
+  while (_strong_code_roots_list->is_nonempty()) {
+    nmethod *nm = _strong_code_roots_list->pop();
+    if (nm != NULL) {
+      NMethodMigrationOopClosure oop_cl(g1h, hr(), nm);
+      nm->oops_do(&oop_cl);
+      if (oop_cl.retain()) {
+        to_be_retained.push(nm);
+      }
+    }
+  }
+
+  // Now push any code roots we need to retain
+  assert(to_be_retained.is_empty() || hr()->evacuation_failed(),
+         "Retained nmethod list must be empty or "
+         "evacuation of this region failed");
+
+  while (to_be_retained.is_nonempty()) {
+    nmethod* nm = to_be_retained.pop();
+    assert(nm != NULL, "sanity");
+    add_strong_code_root(nm);
+  }
+}
+
+void HeapRegionRemSet::strong_code_roots_do(CodeBlobClosure* blk) const {
+  for (int i = 0; i < _strong_code_roots_list->length(); i += 1) {
+    nmethod* nm = _strong_code_roots_list->at(i);
+    blk->do_code_blob(nm);
+  }
+}
+
+size_t HeapRegionRemSet::strong_code_roots_mem_size() {
+  return sizeof(GrowableArray<nmethod*>) +
+         _strong_code_roots_list->max_length() * sizeof(nmethod*);
+}
+
 //-------------------- Iteration --------------------
 
 HeapRegionRemSetIterator:: HeapRegionRemSetIterator(const HeapRegionRemSet* hrrs) :
--- a/src/share/vm/gc_implementation/g1/heapRegionRemSet.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/gc_implementation/g1/heapRegionRemSet.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -37,6 +37,7 @@
 class HeapRegionRemSetIterator;
 class PerRegionTable;
 class SparsePRT;
+class nmethod;
 
 // Essentially a wrapper around SparsePRTCleanupTask. See
 // sparsePRT.hpp for more details.
@@ -191,6 +192,10 @@
   G1BlockOffsetSharedArray* _bosa;
   G1BlockOffsetSharedArray* bosa() const { return _bosa; }
 
+  // A list of code blobs (nmethods) whose code contains pointers into
+  // the region that owns this RSet.
+  GrowableArray<nmethod*>* _strong_code_roots_list;
+
   OtherRegionsTable _other_regions;
 
   enum ParIterState { Unclaimed, Claimed, Complete };
@@ -282,11 +287,13 @@
   }
 
   // The actual # of bytes this hr_remset takes up.
+  // Note also includes the strong code root set.
   size_t mem_size() {
     return _other_regions.mem_size()
       // This correction is necessary because the above includes the second
       // part.
-      + sizeof(this) - sizeof(OtherRegionsTable);
+      + (sizeof(this) - sizeof(OtherRegionsTable))
+      + strong_code_roots_mem_size();
   }
 
   // Returns the memory occupancy of all static data structures associated
@@ -304,6 +311,37 @@
   bool contains_reference(OopOrNarrowOopStar from) const {
     return _other_regions.contains_reference(from);
   }
+
+  // Routines for managing the list of code roots that point into
+  // the heap region that owns this RSet.
+  void add_strong_code_root(nmethod* nm);
+  void remove_strong_code_root(nmethod* nm);
+
+  // During a collection, migrate the successfully evacuated strong
+  // code roots that referenced into the region that owns this RSet
+  // to the RSets of the new regions that they now point into.
+  // Unsuccessfully evacuated code roots are not migrated.
+  void migrate_strong_code_roots();
+
+  // Applies blk->do_code_blob() to each of the entries in
+  // the strong code roots list
+  void strong_code_roots_do(CodeBlobClosure* blk) const;
+
+  // Returns the number of elements in the strong code roots list
+  int strong_code_roots_list_length() {
+    return _strong_code_roots_list->length();
+  }
+
+  // Returns true if the strong code roots contains the given
+  // nmethod.
+  bool strong_code_roots_list_contains(nmethod* nm) {
+    return _strong_code_roots_list->contains(nm);
+  }
+
+  // Returns the amount of memory, in bytes, currently
+  // consumed by the strong code roots.
+  size_t strong_code_roots_mem_size();
+
   void print() const;
 
   // Called during a stop-world phase to perform any deferred cleanups.
--- a/src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -216,6 +216,7 @@
   young_gen()->update_counters();
   old_gen()->update_counters();
   MetaspaceCounters::update_performance_counters();
+  CompressedClassSpaceCounters::update_performance_counters();
 }
 
 size_t ParallelScavengeHeap::capacity() const {
--- a/src/share/vm/gc_interface/collectedHeap.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/gc_interface/collectedHeap.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -118,6 +118,14 @@
   }
 }
 
+void CollectedHeap::register_nmethod(nmethod* nm) {
+  assert_locked_or_safepoint(CodeCache_lock);
+}
+
+void CollectedHeap::unregister_nmethod(nmethod* nm) {
+  assert_locked_or_safepoint(CodeCache_lock);
+}
+
 void CollectedHeap::trace_heap(GCWhen::Type when, GCTracer* gc_tracer) {
   const GCHeapSummary& heap_summary = create_heap_summary();
   const MetaspaceSummary& metaspace_summary = create_metaspace_summary();
--- a/src/share/vm/gc_interface/collectedHeap.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/gc_interface/collectedHeap.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -49,6 +49,7 @@
 class Thread;
 class ThreadClosure;
 class VirtualSpaceSummary;
+class nmethod;
 
 class GCMessage : public FormatBuffer<1024> {
  public:
@@ -603,6 +604,11 @@
   void print_heap_before_gc();
   void print_heap_after_gc();
 
+  // Registering and unregistering an nmethod (compiled code) with the heap.
+  // Override with specific mechanism for each specialized heap type.
+  virtual void register_nmethod(nmethod* nm);
+  virtual void unregister_nmethod(nmethod* nm);
+
   void trace_heap_before_gc(GCTracer* gc_tracer);
   void trace_heap_after_gc(GCTracer* gc_tracer);
 
--- a/src/share/vm/memory/filemap.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/memory/filemap.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -362,15 +362,12 @@
 ReservedSpace FileMapInfo::reserve_shared_memory() {
   struct FileMapInfo::FileMapHeader::space_info* si = &_header._space[0];
   char* requested_addr = si->_base;
-  size_t alignment = os::vm_allocation_granularity();
 
-  size_t size = align_size_up(SharedReadOnlySize + SharedReadWriteSize +
-                              SharedMiscDataSize + SharedMiscCodeSize,
-                              alignment);
+  size_t size = FileMapInfo::shared_spaces_size();
 
   // Reserve the space first, then map otherwise map will go right over some
   // other reserved memory (like the code cache).
-  ReservedSpace rs(size, alignment, false, requested_addr);
+  ReservedSpace rs(size, os::vm_allocation_granularity(), false, requested_addr);
   if (!rs.is_reserved()) {
     fail_continue(err_msg("Unable to reserve shared space at required address " INTPTR_FORMAT, requested_addr));
     return rs;
@@ -559,3 +556,19 @@
                         si->_base, si->_base + si->_used);
   }
 }
+
+// Unmap mapped regions of shared space.
+void FileMapInfo::stop_sharing_and_unmap(const char* msg) {
+  FileMapInfo *map_info = FileMapInfo::current_info();
+  if (map_info) {
+    map_info->fail_continue(msg);
+    for (int i = 0; i < MetaspaceShared::n_regions; i++) {
+      if (map_info->_header._space[i]._base != NULL) {
+        map_info->unmap_region(i);
+        map_info->_header._space[i]._base = NULL;
+      }
+    }
+  } else if (DumpSharedSpaces) {
+    fail_stop(msg, NULL);
+  }
+}
--- a/src/share/vm/memory/filemap.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/memory/filemap.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2003, 2012, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2003, 2013, 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
@@ -150,6 +150,15 @@
   // Return true if given address is in the mapped shared space.
   bool is_in_shared_space(const void* p) NOT_CDS_RETURN_(false);
   void print_shared_spaces() NOT_CDS_RETURN;
+
+  static size_t shared_spaces_size() {
+    return align_size_up(SharedReadOnlySize + SharedReadWriteSize +
+                         SharedMiscDataSize + SharedMiscCodeSize,
+                         os::vm_allocation_granularity());
+  }
+
+  // Stop CDS sharing and unmap CDS regions.
+  static void stop_sharing_and_unmap(const char* msg);
 };
 
 #endif // SHARE_VM_MEMORY_FILEMAP_HPP
--- a/src/share/vm/memory/genCollectedHeap.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/memory/genCollectedHeap.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -1211,6 +1211,7 @@
   }
 
   MetaspaceCounters::update_performance_counters();
+  CompressedClassSpaceCounters::update_performance_counters();
 
   always_do_update_barrier = UseConcMarkSweepGC;
 };
--- a/src/share/vm/memory/heap.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/memory/heap.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2012, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2013, 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
@@ -118,9 +118,12 @@
   _number_of_committed_segments = size_to_segments(_memory.committed_size());
   _number_of_reserved_segments  = size_to_segments(_memory.reserved_size());
   assert(_number_of_reserved_segments >= _number_of_committed_segments, "just checking");
+  const size_t reserved_segments_alignment = MAX2((size_t)os::vm_page_size(), granularity);
+  const size_t reserved_segments_size = align_size_up(_number_of_reserved_segments, reserved_segments_alignment);
+  const size_t committed_segments_size = align_to_page_size(_number_of_committed_segments);
 
   // reserve space for _segmap
-  if (!_segmap.initialize(align_to_page_size(_number_of_reserved_segments), align_to_page_size(_number_of_committed_segments))) {
+  if (!_segmap.initialize(reserved_segments_size, committed_segments_size)) {
     return false;
   }
 
--- a/src/share/vm/memory/iterator.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/memory/iterator.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -64,7 +64,7 @@
 }
 
 void CodeBlobToOopClosure::do_newly_marked_nmethod(nmethod* nm) {
-  nm->oops_do(_cl, /*do_strong_roots_only=*/ true);
+  nm->oops_do(_cl, /*allow_zombie=*/ false);
 }
 
 void CodeBlobToOopClosure::do_code_blob(CodeBlob* cb) {
--- a/src/share/vm/memory/metaspace.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/memory/metaspace.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -35,6 +35,7 @@
 #include "memory/resourceArea.hpp"
 #include "memory/universe.hpp"
 #include "runtime/globals.hpp"
+#include "runtime/java.hpp"
 #include "runtime/mutex.hpp"
 #include "runtime/orderAccess.hpp"
 #include "services/memTracker.hpp"
@@ -54,6 +55,8 @@
 
 MetaWord* last_allocated = 0;
 
+size_t Metaspace::_class_metaspace_size;
+
 // Used in declarations in SpaceManager and ChunkManager
 enum ChunkIndex {
   ZeroIndex = 0,
@@ -261,10 +264,6 @@
   // count of chunks contained in this VirtualSpace
   uintx _container_count;
 
-  // Convenience functions for logical bottom and end
-  MetaWord* bottom() const { return (MetaWord*) _virtual_space.low(); }
-  MetaWord* end() const { return (MetaWord*) _virtual_space.high(); }
-
   // Convenience functions to access the _virtual_space
   char* low()  const { return virtual_space()->low(); }
   char* high() const { return virtual_space()->high(); }
@@ -284,6 +283,10 @@
   VirtualSpaceNode(ReservedSpace rs) : _top(NULL), _next(NULL), _rs(rs), _container_count(0) {}
   ~VirtualSpaceNode();
 
+  // Convenience functions for logical bottom and end
+  MetaWord* bottom() const { return (MetaWord*) _virtual_space.low(); }
+  MetaWord* end() const { return (MetaWord*) _virtual_space.high(); }
+
   // address of next available space in _virtual_space;
   // Accessors
   VirtualSpaceNode* next() { return _next; }
@@ -1313,7 +1316,8 @@
 
   // Class virtual space should always be expanded.  Call GC for the other
   // metadata virtual space.
-  if (vsl == Metaspace::class_space_list()) return true;
+  if (Metaspace::using_class_space() &&
+      (vsl == Metaspace::class_space_list())) return true;
 
   // If this is part of an allocation after a GC, expand
   // unconditionally.
@@ -2257,7 +2261,7 @@
   size_t raw_word_size = get_raw_word_size(word_size);
   size_t min_size = TreeChunk<Metablock, FreeList>::min_size();
   assert(raw_word_size >= min_size,
-    err_msg("Should not deallocate dark matter " SIZE_FORMAT, word_size));
+         err_msg("Should not deallocate dark matter " SIZE_FORMAT "<" SIZE_FORMAT, word_size, min_size));
   block_freelists()->return_block(p, raw_word_size);
 }
 
@@ -2374,7 +2378,7 @@
   if (result == NULL) {
     result = grow_and_allocate(word_size);
   }
-  if (result > 0) {
+  if (result != 0) {
     inc_used_metrics(word_size);
     assert(result != (MetaWord*) chunks_in_use(MediumIndex),
            "Head of the list is being allocated");
@@ -2476,15 +2480,13 @@
 size_t MetaspaceAux::_allocated_capacity_words[] = {0, 0};
 size_t MetaspaceAux::_allocated_used_words[] = {0, 0};
 
+size_t MetaspaceAux::free_bytes(Metaspace::MetadataType mdtype) {
+  VirtualSpaceList* list = Metaspace::get_space_list(mdtype);
+  return list == NULL ? 0 : list->free_bytes();
+}
+
 size_t MetaspaceAux::free_bytes() {
-  size_t result = 0;
-  if (Metaspace::class_space_list() != NULL) {
-    result = result + Metaspace::class_space_list()->free_bytes();
-  }
-  if (Metaspace::space_list() != NULL) {
-    result = result + Metaspace::space_list()->free_bytes();
-  }
-  return result;
+  return free_bytes(Metaspace::ClassType) + free_bytes(Metaspace::NonClassType);
 }
 
 void MetaspaceAux::dec_capacity(Metaspace::MetadataType mdtype, size_t words) {
@@ -2549,6 +2551,9 @@
 }
 
 size_t MetaspaceAux::capacity_bytes_slow(Metaspace::MetadataType mdtype) {
+  if ((mdtype == Metaspace::ClassType) && !Metaspace::using_class_space()) {
+    return 0;
+  }
   // Don't count the space in the freelists.  That space will be
   // added to the capacity calculation as needed.
   size_t capacity = 0;
@@ -2563,18 +2568,18 @@
 }
 
 size_t MetaspaceAux::reserved_in_bytes(Metaspace::MetadataType mdtype) {
-  size_t reserved = (mdtype == Metaspace::ClassType) ?
-                       Metaspace::class_space_list()->virtual_space_total() :
-                       Metaspace::space_list()->virtual_space_total();
-  return reserved * BytesPerWord;
+  VirtualSpaceList* list = Metaspace::get_space_list(mdtype);
+  return list == NULL ? 0 : list->virtual_space_total();
 }
 
 size_t MetaspaceAux::min_chunk_size() { return Metaspace::first_chunk_word_size(); }
 
 size_t MetaspaceAux::free_chunks_total(Metaspace::MetadataType mdtype) {
-  ChunkManager* chunk = (mdtype == Metaspace::ClassType) ?
-                            Metaspace::class_space_list()->chunk_manager() :
-                            Metaspace::space_list()->chunk_manager();
+  VirtualSpaceList* list = Metaspace::get_space_list(mdtype);
+  if (list == NULL) {
+    return 0;
+  }
+  ChunkManager* chunk = list->chunk_manager();
   chunk->slow_verify();
   return chunk->free_chunks_total();
 }
@@ -2615,7 +2620,6 @@
 
 // This is printed when PrintGCDetails
 void MetaspaceAux::print_on(outputStream* out) {
-  Metaspace::MetadataType ct = Metaspace::ClassType;
   Metaspace::MetadataType nct = Metaspace::NonClassType;
 
   out->print_cr(" Metaspace total "
@@ -2629,12 +2633,15 @@
                 allocated_capacity_bytes(nct)/K,
                 allocated_used_bytes(nct)/K,
                 reserved_in_bytes(nct)/K);
-  out->print_cr("  class space    "
-                SIZE_FORMAT "K, used " SIZE_FORMAT "K,"
-                " reserved " SIZE_FORMAT "K",
-                allocated_capacity_bytes(ct)/K,
-                allocated_used_bytes(ct)/K,
-                reserved_in_bytes(ct)/K);
+  if (Metaspace::using_class_space()) {
+    Metaspace::MetadataType ct = Metaspace::ClassType;
+    out->print_cr("  class space    "
+                  SIZE_FORMAT "K, used " SIZE_FORMAT "K,"
+                  " reserved " SIZE_FORMAT "K",
+                  allocated_capacity_bytes(ct)/K,
+                  allocated_used_bytes(ct)/K,
+                  reserved_in_bytes(ct)/K);
+  }
 }
 
 // Print information for class space and data space separately.
@@ -2659,13 +2666,37 @@
   assert(!SafepointSynchronize::is_at_safepoint() || used_and_free == capacity_bytes, "Accounting is wrong");
 }
 
-// Print total fragmentation for class and data metaspaces separately
+// Print total fragmentation for class metaspaces
+void MetaspaceAux::print_class_waste(outputStream* out) {
+  assert(Metaspace::using_class_space(), "class metaspace not used");
+  size_t cls_specialized_waste = 0, cls_small_waste = 0, cls_medium_waste = 0;
+  size_t cls_specialized_count = 0, cls_small_count = 0, cls_medium_count = 0, cls_humongous_count = 0;
+  ClassLoaderDataGraphMetaspaceIterator iter;
+  while (iter.repeat()) {
+    Metaspace* msp = iter.get_next();
+    if (msp != NULL) {
+      cls_specialized_waste += msp->class_vsm()->sum_waste_in_chunks_in_use(SpecializedIndex);
+      cls_specialized_count += msp->class_vsm()->sum_count_in_chunks_in_use(SpecializedIndex);
+      cls_small_waste += msp->class_vsm()->sum_waste_in_chunks_in_use(SmallIndex);
+      cls_small_count += msp->class_vsm()->sum_count_in_chunks_in_use(SmallIndex);
+      cls_medium_waste += msp->class_vsm()->sum_waste_in_chunks_in_use(MediumIndex);
+      cls_medium_count += msp->class_vsm()->sum_count_in_chunks_in_use(MediumIndex);
+      cls_humongous_count += msp->class_vsm()->sum_count_in_chunks_in_use(HumongousIndex);
+    }
+  }
+  out->print_cr(" class: " SIZE_FORMAT " specialized(s) " SIZE_FORMAT ", "
+                SIZE_FORMAT " small(s) " SIZE_FORMAT ", "
+                SIZE_FORMAT " medium(s) " SIZE_FORMAT ", "
+                "large count " SIZE_FORMAT,
+                cls_specialized_count, cls_specialized_waste,
+                cls_small_count, cls_small_waste,
+                cls_medium_count, cls_medium_waste, cls_humongous_count);
+}
+
+// Print total fragmentation for data and class metaspaces separately
 void MetaspaceAux::print_waste(outputStream* out) {
-
   size_t specialized_waste = 0, small_waste = 0, medium_waste = 0;
   size_t specialized_count = 0, small_count = 0, medium_count = 0, humongous_count = 0;
-  size_t cls_specialized_waste = 0, cls_small_waste = 0, cls_medium_waste = 0;
-  size_t cls_specialized_count = 0, cls_small_count = 0, cls_medium_count = 0, cls_humongous_count = 0;
 
   ClassLoaderDataGraphMetaspaceIterator iter;
   while (iter.repeat()) {
@@ -2678,14 +2709,6 @@
       medium_waste += msp->vsm()->sum_waste_in_chunks_in_use(MediumIndex);
       medium_count += msp->vsm()->sum_count_in_chunks_in_use(MediumIndex);
       humongous_count += msp->vsm()->sum_count_in_chunks_in_use(HumongousIndex);
-
-      cls_specialized_waste += msp->class_vsm()->sum_waste_in_chunks_in_use(SpecializedIndex);
-      cls_specialized_count += msp->class_vsm()->sum_count_in_chunks_in_use(SpecializedIndex);
-      cls_small_waste += msp->class_vsm()->sum_waste_in_chunks_in_use(SmallIndex);
-      cls_small_count += msp->class_vsm()->sum_count_in_chunks_in_use(SmallIndex);
-      cls_medium_waste += msp->class_vsm()->sum_waste_in_chunks_in_use(MediumIndex);
-      cls_medium_count += msp->class_vsm()->sum_count_in_chunks_in_use(MediumIndex);
-      cls_humongous_count += msp->class_vsm()->sum_count_in_chunks_in_use(HumongousIndex);
     }
   }
   out->print_cr("Total fragmentation waste (words) doesn't count free space");
@@ -2695,13 +2718,9 @@
                         "large count " SIZE_FORMAT,
              specialized_count, specialized_waste, small_count,
              small_waste, medium_count, medium_waste, humongous_count);
-  out->print_cr(" class: " SIZE_FORMAT " specialized(s) " SIZE_FORMAT ", "
-                           SIZE_FORMAT " small(s) " SIZE_FORMAT ", "
-                           SIZE_FORMAT " medium(s) " SIZE_FORMAT ", "
-                           "large count " SIZE_FORMAT,
-             cls_specialized_count, cls_specialized_waste,
-             cls_small_count, cls_small_waste,
-             cls_medium_count, cls_medium_waste, cls_humongous_count);
+  if (Metaspace::using_class_space()) {
+    print_class_waste(out);
+  }
 }
 
 // Dump global metaspace things from the end of ClassLoaderDataGraph
@@ -2714,7 +2733,9 @@
 
 void MetaspaceAux::verify_free_chunks() {
   Metaspace::space_list()->chunk_manager()->verify();
-  Metaspace::class_space_list()->chunk_manager()->verify();
+  if (Metaspace::using_class_space()) {
+    Metaspace::class_space_list()->chunk_manager()->verify();
+  }
 }
 
 void MetaspaceAux::verify_capacity() {
@@ -2776,7 +2797,9 @@
 
 Metaspace::~Metaspace() {
   delete _vsm;
-  delete _class_vsm;
+  if (using_class_space()) {
+    delete _class_vsm;
+  }
 }
 
 VirtualSpaceList* Metaspace::_space_list = NULL;
@@ -2784,9 +2807,123 @@
 
 #define VIRTUALSPACEMULTIPLIER 2
 
+#ifdef _LP64
+void Metaspace::set_narrow_klass_base_and_shift(address metaspace_base, address cds_base) {
+  // Figure out the narrow_klass_base and the narrow_klass_shift.  The
+  // narrow_klass_base is the lower of the metaspace base and the cds base
+  // (if cds is enabled).  The narrow_klass_shift depends on the distance
+  // between the lower base and higher address.
+  address lower_base;
+  address higher_address;
+  if (UseSharedSpaces) {
+    higher_address = MAX2((address)(cds_base + FileMapInfo::shared_spaces_size()),
+                          (address)(metaspace_base + class_metaspace_size()));
+    lower_base = MIN2(metaspace_base, cds_base);
+  } else {
+    higher_address = metaspace_base + class_metaspace_size();
+    lower_base = metaspace_base;
+  }
+  Universe::set_narrow_klass_base(lower_base);
+  if ((uint64_t)(higher_address - lower_base) < (uint64_t)max_juint) {
+    Universe::set_narrow_klass_shift(0);
+  } else {
+    assert(!UseSharedSpaces, "Cannot shift with UseSharedSpaces");
+    Universe::set_narrow_klass_shift(LogKlassAlignmentInBytes);
+  }
+}
+
+// Return TRUE if the specified metaspace_base and cds_base are close enough
+// to work with compressed klass pointers.
+bool Metaspace::can_use_cds_with_metaspace_addr(char* metaspace_base, address cds_base) {
+  assert(cds_base != 0 && UseSharedSpaces, "Only use with CDS");
+  assert(UseCompressedKlassPointers, "Only use with CompressedKlassPtrs");
+  address lower_base = MIN2((address)metaspace_base, cds_base);
+  address higher_address = MAX2((address)(cds_base + FileMapInfo::shared_spaces_size()),
+                                (address)(metaspace_base + class_metaspace_size()));
+  return ((uint64_t)(higher_address - lower_base) < (uint64_t)max_juint);
+}
+
+// Try to allocate the metaspace at the requested addr.
+void Metaspace::allocate_metaspace_compressed_klass_ptrs(char* requested_addr, address cds_base) {
+  assert(using_class_space(), "called improperly");
+  assert(UseCompressedKlassPointers, "Only use with CompressedKlassPtrs");
+  assert(class_metaspace_size() < KlassEncodingMetaspaceMax,
+         "Metaspace size is too big");
+
+  ReservedSpace metaspace_rs = ReservedSpace(class_metaspace_size(),
+                                             os::vm_allocation_granularity(),
+                                             false, requested_addr, 0);
+  if (!metaspace_rs.is_reserved()) {
+    if (UseSharedSpaces) {
+      // Keep trying to allocate the metaspace, increasing the requested_addr
+      // by 1GB each time, until we reach an address that will no longer allow
+      // use of CDS with compressed klass pointers.
+      char *addr = requested_addr;
+      while (!metaspace_rs.is_reserved() && (addr + 1*G > addr) &&
+             can_use_cds_with_metaspace_addr(addr + 1*G, cds_base)) {
+        addr = addr + 1*G;
+        metaspace_rs = ReservedSpace(class_metaspace_size(),
+                                     os::vm_allocation_granularity(), false, addr, 0);
+      }
+    }
+
+    // If no successful allocation then try to allocate the space anywhere.  If
+    // that fails then OOM doom.  At this point we cannot try allocating the
+    // metaspace as if UseCompressedKlassPointers is off because too much
+    // initialization has happened that depends on UseCompressedKlassPointers.
+    // So, UseCompressedKlassPointers cannot be turned off at this point.
+    if (!metaspace_rs.is_reserved()) {
+      metaspace_rs = ReservedSpace(class_metaspace_size(),
+                                   os::vm_allocation_granularity(), false);
+      if (!metaspace_rs.is_reserved()) {
+        vm_exit_during_initialization(err_msg("Could not allocate metaspace: %d bytes",
+                                              class_metaspace_size()));
+      }
+    }
+  }
+
+  // If we got here then the metaspace got allocated.
+  MemTracker::record_virtual_memory_type((address)metaspace_rs.base(), mtClass);
+
+  // Verify that we can use shared spaces.  Otherwise, turn off CDS.
+  if (UseSharedSpaces && !can_use_cds_with_metaspace_addr(metaspace_rs.base(), cds_base)) {
+    FileMapInfo::stop_sharing_and_unmap(
+        "Could not allocate metaspace at a compatible address");
+  }
+
+  set_narrow_klass_base_and_shift((address)metaspace_rs.base(),
+                                  UseSharedSpaces ? (address)cds_base : 0);
+
+  initialize_class_space(metaspace_rs);
+
+  if (PrintCompressedOopsMode || (PrintMiscellaneous && Verbose)) {
+    gclog_or_tty->print_cr("Narrow klass base: " PTR_FORMAT ", Narrow klass shift: " SIZE_FORMAT,
+                            Universe::narrow_klass_base(), Universe::narrow_klass_shift());
+    gclog_or_tty->print_cr("Metaspace Size: " SIZE_FORMAT " Address: " PTR_FORMAT " Req Addr: " PTR_FORMAT,
+                           class_metaspace_size(), metaspace_rs.base(), requested_addr);
+  }
+}
+
+// For UseCompressedKlassPointers the class space is reserved above the top of
+// the Java heap.  The argument passed in is at the base of the compressed space.
+void Metaspace::initialize_class_space(ReservedSpace rs) {
+  // The reserved space size may be bigger because of alignment, esp with UseLargePages
+  assert(rs.size() >= ClassMetaspaceSize,
+         err_msg(SIZE_FORMAT " != " UINTX_FORMAT, rs.size(), ClassMetaspaceSize));
+  assert(using_class_space(), "Must be using class space");
+  _class_space_list = new VirtualSpaceList(rs);
+}
+
+#endif
+
 void Metaspace::global_initialize() {
   // Initialize the alignment for shared spaces.
   int max_alignment = os::vm_page_size();
+  size_t cds_total = 0;
+
+  set_class_metaspace_size(align_size_up(ClassMetaspaceSize,
+                                         os::vm_allocation_granularity()));
+
   MetaspaceShared::set_max_alignment(max_alignment);
 
   if (DumpSharedSpaces) {
@@ -2798,15 +2935,31 @@
     // Initialize with the sum of the shared space sizes.  The read-only
     // and read write metaspace chunks will be allocated out of this and the
     // remainder is the misc code and data chunks.
-    size_t total = align_size_up(SharedReadOnlySize + SharedReadWriteSize +
-                                 SharedMiscDataSize + SharedMiscCodeSize,
-                                 os::vm_allocation_granularity());
-    size_t word_size = total/wordSize;
-    _space_list = new VirtualSpaceList(word_size);
+    cds_total = FileMapInfo::shared_spaces_size();
+    _space_list = new VirtualSpaceList(cds_total/wordSize);
+
+#ifdef _LP64
+    // Set the compressed klass pointer base so that decoding of these pointers works
+    // properly when creating the shared archive.
+    assert(UseCompressedOops && UseCompressedKlassPointers,
+      "UseCompressedOops and UseCompressedKlassPointers must be set");
+    Universe::set_narrow_klass_base((address)_space_list->current_virtual_space()->bottom());
+    if (TraceMetavirtualspaceAllocation && Verbose) {
+      gclog_or_tty->print_cr("Setting_narrow_klass_base to Address: " PTR_FORMAT,
+                             _space_list->current_virtual_space()->bottom());
+    }
+
+    // Set the shift to zero.
+    assert(class_metaspace_size() < (uint64_t)(max_juint) - cds_total,
+           "CDS region is too large");
+    Universe::set_narrow_klass_shift(0);
+#endif
+
   } else {
     // If using shared space, open the file that contains the shared space
     // and map in the memory before initializing the rest of metaspace (so
     // the addresses don't conflict)
+    address cds_address = NULL;
     if (UseSharedSpaces) {
       FileMapInfo* mapinfo = new FileMapInfo();
       memset(mapinfo, 0, sizeof(FileMapInfo));
@@ -2821,8 +2974,22 @@
         assert(!mapinfo->is_open() && !UseSharedSpaces,
                "archive file not closed or shared spaces not disabled.");
       }
+      cds_total = FileMapInfo::shared_spaces_size();
+      cds_address = (address)mapinfo->region_base(0);
     }
 
+#ifdef _LP64
+    // If UseCompressedKlassPointers is set then allocate the metaspace area
+    // above the heap and above the CDS area (if it exists).
+    if (using_class_space()) {
+      if (UseSharedSpaces) {
+        allocate_metaspace_compressed_klass_ptrs((char *)(cds_address + cds_total), cds_address);
+      } else {
+        allocate_metaspace_compressed_klass_ptrs((char *)CompressedKlassPointersBase, 0);
+      }
+    }
+#endif
+
     // Initialize these before initializing the VirtualSpaceList
     _first_chunk_word_size = InitialBootClassLoaderMetaspaceSize / BytesPerWord;
     _first_chunk_word_size = align_word_size_up(_first_chunk_word_size);
@@ -2840,39 +3007,28 @@
   }
 }
 
-// For UseCompressedKlassPointers the class space is reserved as a piece of the
-// Java heap because the compression algorithm is the same for each.  The
-// argument passed in is at the top of the compressed space
-void Metaspace::initialize_class_space(ReservedSpace rs) {
-  // The reserved space size may be bigger because of alignment, esp with UseLargePages
-  assert(rs.size() >= ClassMetaspaceSize,
-         err_msg(SIZE_FORMAT " != " UINTX_FORMAT, rs.size(), ClassMetaspaceSize));
-  _class_space_list = new VirtualSpaceList(rs);
-}
-
-void Metaspace::initialize(Mutex* lock,
-                           MetaspaceType type) {
+void Metaspace::initialize(Mutex* lock, MetaspaceType type) {
 
   assert(space_list() != NULL,
     "Metadata VirtualSpaceList has not been initialized");
 
-  _vsm = new SpaceManager(Metaspace::NonClassType, lock, space_list());
+  _vsm = new SpaceManager(NonClassType, lock, space_list());
   if (_vsm == NULL) {
     return;
   }
   size_t word_size;
   size_t class_word_size;
-  vsm()->get_initial_chunk_sizes(type,
-                                 &word_size,
-                                 &class_word_size);
-
-  assert(class_space_list() != NULL,
-    "Class VirtualSpaceList has not been initialized");
-
-  // Allocate SpaceManager for classes.
-  _class_vsm = new SpaceManager(Metaspace::ClassType, lock, class_space_list());
-  if (_class_vsm == NULL) {
-    return;
+  vsm()->get_initial_chunk_sizes(type, &word_size, &class_word_size);
+
+  if (using_class_space()) {
+    assert(class_space_list() != NULL,
+      "Class VirtualSpaceList has not been initialized");
+
+    // Allocate SpaceManager for classes.
+    _class_vsm = new SpaceManager(ClassType, lock, class_space_list());
+    if (_class_vsm == NULL) {
+      return;
+    }
   }
 
   MutexLockerEx cl(SpaceManager::expand_lock(), Mutex::_no_safepoint_check_flag);
@@ -2888,11 +3044,13 @@
   }
 
   // Allocate chunk for class metadata objects
-  Metachunk* class_chunk =
-     class_space_list()->get_initialization_chunk(class_word_size,
-                                                  class_vsm()->medium_chunk_bunch());
-  if (class_chunk != NULL) {
-    class_vsm()->add_chunk(class_chunk, true);
+  if (using_class_space()) {
+    Metachunk* class_chunk =
+       class_space_list()->get_initialization_chunk(class_word_size,
+                                                    class_vsm()->medium_chunk_bunch());
+    if (class_chunk != NULL) {
+      class_vsm()->add_chunk(class_chunk, true);
+    }
   }
 
   _alloc_record_head = NULL;
@@ -2906,7 +3064,8 @@
 
 MetaWord* Metaspace::allocate(size_t word_size, MetadataType mdtype) {
   // DumpSharedSpaces doesn't use class metadata area (yet)
-  if (mdtype == ClassType && !DumpSharedSpaces) {
+  // Also, don't use class_vsm() unless UseCompressedKlassPointers is true.
+  if (mdtype == ClassType && using_class_space()) {
     return  class_vsm()->allocate(word_size);
   } else {
     return  vsm()->allocate(word_size);
@@ -2937,14 +3096,19 @@
 }
 
 size_t Metaspace::used_words_slow(MetadataType mdtype) const {
-  // return vsm()->allocated_used_words();
-  return mdtype == ClassType ? class_vsm()->sum_used_in_chunks_in_use() :
-                               vsm()->sum_used_in_chunks_in_use();  // includes overhead!
+  if (mdtype == ClassType) {
+    return using_class_space() ? class_vsm()->sum_used_in_chunks_in_use() : 0;
+  } else {
+    return vsm()->sum_used_in_chunks_in_use();  // includes overhead!
+  }
 }
 
 size_t Metaspace::free_words(MetadataType mdtype) const {
-  return mdtype == ClassType ? class_vsm()->sum_free_in_chunks_in_use() :
-                               vsm()->sum_free_in_chunks_in_use();
+  if (mdtype == ClassType) {
+    return using_class_space() ? class_vsm()->sum_free_in_chunks_in_use() : 0;
+  } else {
+    return vsm()->sum_free_in_chunks_in_use();
+  }
 }
 
 // Space capacity in the Metaspace.  It includes
@@ -2953,8 +3117,11 @@
 // in the space available in the dictionary which
 // is already counted in some chunk.
 size_t Metaspace::capacity_words_slow(MetadataType mdtype) const {
-  return mdtype == ClassType ? class_vsm()->sum_capacity_in_chunks_in_use() :
-                               vsm()->sum_capacity_in_chunks_in_use();
+  if (mdtype == ClassType) {
+    return using_class_space() ? class_vsm()->sum_capacity_in_chunks_in_use() : 0;
+  } else {
+    return vsm()->sum_capacity_in_chunks_in_use();
+  }
 }
 
 size_t Metaspace::used_bytes_slow(MetadataType mdtype) const {
@@ -2977,8 +3144,8 @@
 #endif
       return;
     }
-    if (is_class) {
-       class_vsm()->deallocate(ptr, word_size);
+    if (is_class && using_class_space()) {
+      class_vsm()->deallocate(ptr, word_size);
     } else {
       vsm()->deallocate(ptr, word_size);
     }
@@ -2992,7 +3159,7 @@
 #endif
       return;
     }
-    if (is_class) {
+    if (is_class && using_class_space()) {
       class_vsm()->deallocate(ptr, word_size);
     } else {
       vsm()->deallocate(ptr, word_size);
@@ -3101,14 +3268,18 @@
   MutexLockerEx cl(SpaceManager::expand_lock(),
                    Mutex::_no_safepoint_check_flag);
   space_list()->purge();
-  class_space_list()->purge();
+  if (using_class_space()) {
+    class_space_list()->purge();
+  }
 }
 
 void Metaspace::print_on(outputStream* out) const {
   // Print both class virtual space counts and metaspace.
   if (Verbose) {
-      vsm()->print_on(out);
+    vsm()->print_on(out);
+    if (using_class_space()) {
       class_vsm()->print_on(out);
+    }
   }
 }
 
@@ -3122,17 +3293,21 @@
   // be needed.  Note, locking this can cause inversion problems with the
   // caller in MetaspaceObj::is_metadata() function.
   return space_list()->contains(ptr) ||
-         class_space_list()->contains(ptr);
+         (using_class_space() && class_space_list()->contains(ptr));
 }
 
 void Metaspace::verify() {
   vsm()->verify();
-  class_vsm()->verify();
+  if (using_class_space()) {
+    class_vsm()->verify();
+  }
 }
 
 void Metaspace::dump(outputStream* const out) const {
   out->print_cr("\nVirtual space manager: " INTPTR_FORMAT, vsm());
   vsm()->dump(out);
-  out->print_cr("\nClass space manager: " INTPTR_FORMAT, class_vsm());
-  class_vsm()->dump(out);
+  if (using_class_space()) {
+    out->print_cr("\nClass space manager: " INTPTR_FORMAT, class_vsm());
+    class_vsm()->dump(out);
+  }
 }
--- a/src/share/vm/memory/metaspace.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/memory/metaspace.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -105,6 +105,16 @@
   // Align up the word size to the allocation word size
   static size_t align_word_size_up(size_t);
 
+  // Aligned size of the metaspace.
+  static size_t _class_metaspace_size;
+
+  static size_t class_metaspace_size() {
+    return _class_metaspace_size;
+  }
+  static void set_class_metaspace_size(size_t metaspace_size) {
+    _class_metaspace_size = metaspace_size;
+  }
+
   static size_t _first_chunk_word_size;
   static size_t _first_class_chunk_word_size;
 
@@ -126,11 +136,26 @@
 
   static VirtualSpaceList* space_list()       { return _space_list; }
   static VirtualSpaceList* class_space_list() { return _class_space_list; }
+  static VirtualSpaceList* get_space_list(MetadataType mdtype) {
+    assert(mdtype != MetadataTypeCount, "MetadaTypeCount can't be used as mdtype");
+    return mdtype == ClassType ? class_space_list() : space_list();
+  }
 
   // This is used by DumpSharedSpaces only, where only _vsm is used. So we will
   // maintain a single list for now.
   void record_allocation(void* ptr, MetaspaceObj::Type type, size_t word_size);
 
+#ifdef _LP64
+  static void set_narrow_klass_base_and_shift(address metaspace_base, address cds_base);
+
+  // Returns true if can use CDS with metaspace allocated as specified address.
+  static bool can_use_cds_with_metaspace_addr(char* metaspace_base, address cds_base);
+
+  static void allocate_metaspace_compressed_klass_ptrs(char* requested_addr, address cds_base);
+
+  static void initialize_class_space(ReservedSpace rs);
+#endif
+
   class AllocRecord : public CHeapObj<mtClass> {
   public:
     AllocRecord(address ptr, MetaspaceObj::Type type, int byte_size)
@@ -151,7 +176,6 @@
 
   // Initialize globals for Metaspace
   static void global_initialize();
-  static void initialize_class_space(ReservedSpace rs);
 
   static size_t first_chunk_word_size() { return _first_chunk_word_size; }
   static size_t first_class_chunk_word_size() { return _first_class_chunk_word_size; }
@@ -172,8 +196,6 @@
   MetaWord* expand_and_allocate(size_t size,
                                 MetadataType mdtype);
 
-  static bool is_initialized() { return _class_space_list != NULL; }
-
   static bool contains(const void *ptr);
   void dump(outputStream* const out) const;
 
@@ -190,11 +212,16 @@
   };
 
   void iterate(AllocRecordClosure *closure);
+
+  // Return TRUE only if UseCompressedKlassPointers is True and DumpSharedSpaces is False.
+  static bool using_class_space() {
+    return NOT_LP64(false) LP64_ONLY(UseCompressedKlassPointers && !DumpSharedSpaces);
+  }
+
 };
 
 class MetaspaceAux : AllStatic {
   static size_t free_chunks_total(Metaspace::MetadataType mdtype);
-  static size_t free_chunks_total_in_bytes(Metaspace::MetadataType mdtype);
 
  public:
   // Statistics for class space and data space in metaspace.
@@ -238,13 +265,15 @@
   // Used by MetaspaceCounters
   static size_t free_chunks_total();
   static size_t free_chunks_total_in_bytes();
+  static size_t free_chunks_total_in_bytes(Metaspace::MetadataType mdtype);
 
   static size_t allocated_capacity_words(Metaspace::MetadataType mdtype) {
     return _allocated_capacity_words[mdtype];
   }
   static size_t allocated_capacity_words() {
-    return _allocated_capacity_words[Metaspace::ClassType] +
-           _allocated_capacity_words[Metaspace::NonClassType];
+    return _allocated_capacity_words[Metaspace::NonClassType] +
+           (Metaspace::using_class_space() ?
+           _allocated_capacity_words[Metaspace::ClassType] : 0);
   }
   static size_t allocated_capacity_bytes(Metaspace::MetadataType mdtype) {
     return allocated_capacity_words(mdtype) * BytesPerWord;
@@ -257,8 +286,9 @@
     return _allocated_used_words[mdtype];
   }
   static size_t allocated_used_words() {
-    return _allocated_used_words[Metaspace::ClassType] +
-           _allocated_used_words[Metaspace::NonClassType];
+    return _allocated_used_words[Metaspace::NonClassType] +
+           (Metaspace::using_class_space() ?
+           _allocated_used_words[Metaspace::ClassType] : 0);
   }
   static size_t allocated_used_bytes(Metaspace::MetadataType mdtype) {
     return allocated_used_words(mdtype) * BytesPerWord;
@@ -268,6 +298,7 @@
   }
 
   static size_t free_bytes();
+  static size_t free_bytes(Metaspace::MetadataType mdtype);
 
   // Total capacity in all Metaspaces
   static size_t capacity_bytes_slow() {
@@ -300,6 +331,7 @@
   static void print_on(outputStream * out);
   static void print_on(outputStream * out, Metaspace::MetadataType mdtype);
 
+  static void print_class_waste(outputStream* out);
   static void print_waste(outputStream* out);
   static void dump(outputStream* out);
   static void verify_free_chunks();
--- a/src/share/vm/memory/metaspaceCounters.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/memory/metaspaceCounters.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -25,11 +25,47 @@
 #include "precompiled.hpp"
 #include "memory/metaspaceCounters.hpp"
 #include "memory/resourceArea.hpp"
+#include "runtime/globals.hpp"
+#include "runtime/perfData.hpp"
 #include "utilities/exceptions.hpp"
 
-MetaspaceCounters* MetaspaceCounters::_metaspace_counters = NULL;
+class MetaspacePerfCounters: public CHeapObj<mtInternal> {
+  friend class VMStructs;
+  PerfVariable*      _capacity;
+  PerfVariable*      _used;
+  PerfVariable*      _max_capacity;
 
-size_t MetaspaceCounters::calc_total_capacity() {
+  PerfVariable* create_variable(const char *ns, const char *name, size_t value, TRAPS) {
+    const char *path = PerfDataManager::counter_name(ns, name);
+    return PerfDataManager::create_variable(SUN_GC, path, PerfData::U_Bytes, value, THREAD);
+  }
+
+  void create_constant(const char *ns, const char *name, size_t value, TRAPS) {
+    const char *path = PerfDataManager::counter_name(ns, name);
+    PerfDataManager::create_constant(SUN_GC, path, PerfData::U_Bytes, value, THREAD);
+  }
+
+ public:
+  MetaspacePerfCounters(const char* ns, size_t min_capacity, size_t curr_capacity, size_t max_capacity, size_t used) {
+    EXCEPTION_MARK;
+    ResourceMark rm;
+
+    create_constant(ns, "minCapacity", min_capacity, THREAD);
+    _capacity = create_variable(ns, "capacity", curr_capacity, THREAD);
+    _max_capacity = create_variable(ns, "maxCapacity", max_capacity, THREAD);
+    _used = create_variable(ns, "used", used, THREAD);
+  }
+
+  void update(size_t capacity, size_t max_capacity, size_t used) {
+    _capacity->set_value(capacity);
+    _max_capacity->set_value(max_capacity);
+    _used->set_value(used);
+  }
+};
+
+MetaspacePerfCounters* MetaspaceCounters::_perf_counters = NULL;
+
+size_t MetaspaceCounters::calculate_capacity() {
   // The total capacity is the sum of
   //   1) capacity of Metachunks in use by all Metaspaces
   //   2) unused space at the end of each Metachunk
@@ -39,95 +75,65 @@
   return total_capacity;
 }
 
-MetaspaceCounters::MetaspaceCounters() :
-    _capacity(NULL),
-    _used(NULL),
-    _max_capacity(NULL) {
+void MetaspaceCounters::initialize_performance_counters() {
   if (UsePerfData) {
+    assert(_perf_counters == NULL, "Should only be initialized once");
+
     size_t min_capacity = MetaspaceAux::min_chunk_size();
+    size_t capacity = calculate_capacity();
     size_t max_capacity = MetaspaceAux::reserved_in_bytes();
-    size_t curr_capacity = calc_total_capacity();
     size_t used = MetaspaceAux::allocated_used_bytes();
 
-    initialize(min_capacity, max_capacity, curr_capacity, used);
-  }
-}
-
-static PerfVariable* create_ms_variable(const char *ns,
-                                        const char *name,
-                                        size_t value,
-                                        TRAPS) {
-  const char *path = PerfDataManager::counter_name(ns, name);
-  PerfVariable *result =
-      PerfDataManager::create_variable(SUN_GC, path, PerfData::U_Bytes, value,
-                                       CHECK_NULL);
-  return result;
-}
-
-static void create_ms_constant(const char *ns,
-                               const char *name,
-                               size_t value,
-                               TRAPS) {
-  const char *path = PerfDataManager::counter_name(ns, name);
-  PerfDataManager::create_constant(SUN_GC, path, PerfData::U_Bytes, value, CHECK);
-}
-
-void MetaspaceCounters::initialize(size_t min_capacity,
-                                   size_t max_capacity,
-                                   size_t curr_capacity,
-                                   size_t used) {
-
-  if (UsePerfData) {
-    EXCEPTION_MARK;
-    ResourceMark rm;
-
-    const char *ms = "metaspace";
-
-    create_ms_constant(ms, "minCapacity", min_capacity, CHECK);
-    _max_capacity = create_ms_variable(ms, "maxCapacity", max_capacity, CHECK);
-    _capacity = create_ms_variable(ms, "capacity", curr_capacity, CHECK);
-    _used = create_ms_variable(ms, "used", used, CHECK);
-  }
-}
-
-void MetaspaceCounters::update_capacity() {
-  assert(UsePerfData, "Should not be called unless being used");
-  size_t total_capacity = calc_total_capacity();
-  _capacity->set_value(total_capacity);
-}
-
-void MetaspaceCounters::update_used() {
-  assert(UsePerfData, "Should not be called unless being used");
-  size_t used_in_bytes = MetaspaceAux::allocated_used_bytes();
-  _used->set_value(used_in_bytes);
-}
-
-void MetaspaceCounters::update_max_capacity() {
-  assert(UsePerfData, "Should not be called unless being used");
-  assert(_max_capacity != NULL, "Should be initialized");
-  size_t reserved_in_bytes = MetaspaceAux::reserved_in_bytes();
-  _max_capacity->set_value(reserved_in_bytes);
-}
-
-void MetaspaceCounters::update_all() {
-  if (UsePerfData) {
-    update_used();
-    update_capacity();
-    update_max_capacity();
-  }
-}
-
-void MetaspaceCounters::initialize_performance_counters() {
-  if (UsePerfData) {
-    assert(_metaspace_counters == NULL, "Should only be initialized once");
-    _metaspace_counters = new MetaspaceCounters();
+    _perf_counters = new MetaspacePerfCounters("metaspace", min_capacity, capacity, max_capacity, used);
   }
 }
 
 void MetaspaceCounters::update_performance_counters() {
   if (UsePerfData) {
-    assert(_metaspace_counters != NULL, "Should be initialized");
-    _metaspace_counters->update_all();
+    assert(_perf_counters != NULL, "Should be initialized");
+
+    size_t capacity = calculate_capacity();
+    size_t max_capacity = MetaspaceAux::reserved_in_bytes();
+    size_t used = MetaspaceAux::allocated_used_bytes();
+
+    _perf_counters->update(capacity, max_capacity, used);
   }
 }
 
+MetaspacePerfCounters* CompressedClassSpaceCounters::_perf_counters = NULL;
+
+size_t CompressedClassSpaceCounters::calculate_capacity() {
+    return MetaspaceAux::allocated_capacity_bytes(_class_type) +
+           MetaspaceAux::free_bytes(_class_type) +
+           MetaspaceAux::free_chunks_total_in_bytes(_class_type);
+}
+
+void CompressedClassSpaceCounters::update_performance_counters() {
+  if (UsePerfData && UseCompressedKlassPointers) {
+    assert(_perf_counters != NULL, "Should be initialized");
+
+    size_t capacity = calculate_capacity();
+    size_t max_capacity = MetaspaceAux::reserved_in_bytes(_class_type);
+    size_t used = MetaspaceAux::allocated_used_bytes(_class_type);
+
+    _perf_counters->update(capacity, max_capacity, used);
+  }
+}
+
+void CompressedClassSpaceCounters::initialize_performance_counters() {
+  if (UsePerfData) {
+    assert(_perf_counters == NULL, "Should only be initialized once");
+    const char* ns = "compressedclassspace";
+
+    if (UseCompressedKlassPointers) {
+      size_t min_capacity = MetaspaceAux::min_chunk_size();
+      size_t capacity = calculate_capacity();
+      size_t max_capacity = MetaspaceAux::reserved_in_bytes(_class_type);
+      size_t used = MetaspaceAux::allocated_used_bytes(_class_type);
+
+      _perf_counters = new MetaspacePerfCounters(ns, min_capacity, capacity, max_capacity, used);
+    } else {
+      _perf_counters = new MetaspacePerfCounters(ns, 0, 0, 0, 0);
+    }
+  }
+}
--- a/src/share/vm/memory/metaspaceCounters.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/memory/metaspaceCounters.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -25,31 +25,27 @@
 #ifndef SHARE_VM_MEMORY_METASPACECOUNTERS_HPP
 #define SHARE_VM_MEMORY_METASPACECOUNTERS_HPP
 
-#include "runtime/perfData.hpp"
+#include "memory/metaspace.hpp"
 
-class MetaspaceCounters: public CHeapObj<mtClass> {
-  friend class VMStructs;
-  PerfVariable*      _capacity;
-  PerfVariable*      _used;
-  PerfVariable*      _max_capacity;
-  static MetaspaceCounters* _metaspace_counters;
-  void initialize(size_t min_capacity,
-                  size_t max_capacity,
-                  size_t curr_capacity,
-                  size_t used);
-  size_t calc_total_capacity();
+class MetaspacePerfCounters;
+
+class MetaspaceCounters: public AllStatic {
+  static MetaspacePerfCounters* _perf_counters;
+  static size_t calculate_capacity();
+
  public:
-  MetaspaceCounters();
-  ~MetaspaceCounters();
-
-  void update_capacity();
-  void update_used();
-  void update_max_capacity();
-
-  void update_all();
-
   static void initialize_performance_counters();
   static void update_performance_counters();
+};
 
+class CompressedClassSpaceCounters: public AllStatic {
+  static MetaspacePerfCounters* _perf_counters;
+  static size_t calculate_capacity();
+  static const Metaspace::MetadataType _class_type = Metaspace::ClassType;
+
+ public:
+  static void initialize_performance_counters();
+  static void update_performance_counters();
 };
+
 #endif // SHARE_VM_MEMORY_METASPACECOUNTERS_HPP
--- a/src/share/vm/memory/metaspaceShared.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/memory/metaspaceShared.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -52,7 +52,6 @@
   int tag = 0;
   soc->do_tag(--tag);
 
-  assert(!UseCompressedOops, "UseCompressedOops doesn't work with shared archive");
   // Verify the sizes of various metadata in the system.
   soc->do_tag(sizeof(Method));
   soc->do_tag(sizeof(ConstMethod));
--- a/src/share/vm/memory/universe.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/memory/universe.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -145,8 +145,6 @@
 NarrowPtrStruct Universe::_narrow_klass = { NULL, 0, true };
 address Universe::_narrow_ptrs_base;
 
-size_t          Universe::_class_metaspace_size;
-
 void Universe::basic_type_classes_do(void f(Klass*)) {
   f(boolArrayKlassObj());
   f(byteArrayKlassObj());
@@ -641,6 +639,8 @@
     return status;
   }
 
+  Metaspace::global_initialize();
+
   // Create memory for metadata.  Must be after initializing heap for
   // DumpSharedSpaces.
   ClassLoaderData::init_null_class_loader_data();
@@ -693,13 +693,9 @@
     if (!FLAG_IS_DEFAULT(HeapBaseMinAddress) && (mode == UnscaledNarrowOop)) {
       base = HeapBaseMinAddress;
 
-    // If the total size and the metaspace size are small enough to allow
-    // UnscaledNarrowOop then just use UnscaledNarrowOop.
-    } else if ((total_size <= OopEncodingHeapMax) && (mode != HeapBasedNarrowOop) &&
-        (!UseCompressedKlassPointers ||
-          (((OopEncodingHeapMax - heap_size) + Universe::class_metaspace_size()) <= KlassEncodingMetaspaceMax))) {
-      // We don't need to check the metaspace size here because it is always smaller
-      // than total_size.
+    // If the total size is small enough to allow UnscaledNarrowOop then
+    // just use UnscaledNarrowOop.
+    } else if ((total_size <= OopEncodingHeapMax) && (mode != HeapBasedNarrowOop)) {
       if ((total_size <= NarrowOopHeapMax) && (mode == UnscaledNarrowOop) &&
           (Universe::narrow_oop_shift() == 0)) {
         // Use 32-bits oops without encoding and
@@ -716,13 +712,6 @@
           base = (OopEncodingHeapMax - heap_size);
         }
       }
-
-    // See if ZeroBaseNarrowOop encoding will work for a heap based at
-    // (KlassEncodingMetaspaceMax - class_metaspace_size()).
-    } else if (UseCompressedKlassPointers && (mode != HeapBasedNarrowOop) &&
-        (Universe::class_metaspace_size() + HeapBaseMinAddress <= KlassEncodingMetaspaceMax) &&
-        (KlassEncodingMetaspaceMax + heap_size - Universe::class_metaspace_size() <= OopEncodingHeapMax)) {
-      base = (KlassEncodingMetaspaceMax - Universe::class_metaspace_size());
     } else {
       // UnscaledNarrowOop encoding didn't work, and no base was found for ZeroBasedOops or
       // HeapBasedNarrowOop encoding was requested.  So, can't reserve below 32Gb.
@@ -732,8 +721,7 @@
     // Set narrow_oop_base and narrow_oop_use_implicit_null_checks
     // used in ReservedHeapSpace() constructors.
     // The final values will be set in initialize_heap() below.
-    if ((base != 0) && ((base + heap_size) <= OopEncodingHeapMax) &&
-        (!UseCompressedKlassPointers || (base + Universe::class_metaspace_size()) <= KlassEncodingMetaspaceMax)) {
+    if ((base != 0) && ((base + heap_size) <= OopEncodingHeapMax)) {
       // Use zero based compressed oops
       Universe::set_narrow_oop_base(NULL);
       // Don't need guard page for implicit checks in indexed
@@ -816,9 +804,7 @@
       tty->print("heap address: " PTR_FORMAT ", size: " SIZE_FORMAT " MB",
                  Universe::heap()->base(), Universe::heap()->reserved_region().byte_size()/M);
     }
-    if (((uint64_t)Universe::heap()->reserved_region().end() > OopEncodingHeapMax) ||
-        (UseCompressedKlassPointers &&
-        ((uint64_t)Universe::heap()->base() + Universe::class_metaspace_size() > KlassEncodingMetaspaceMax))) {
+    if (((uint64_t)Universe::heap()->reserved_region().end() > OopEncodingHeapMax)) {
       // Can't reserve heap below 32Gb.
       // keep the Universe::narrow_oop_base() set in Universe::reserve_heap()
       Universe::set_narrow_oop_shift(LogMinObjAlignmentInBytes);
@@ -849,20 +835,16 @@
         }
       }
     }
+
     if (verbose) {
       tty->cr();
       tty->cr();
     }
-    if (UseCompressedKlassPointers) {
-      Universe::set_narrow_klass_base(Universe::narrow_oop_base());
-      Universe::set_narrow_klass_shift(MIN2(Universe::narrow_oop_shift(), LogKlassAlignmentInBytes));
-    }
     Universe::set_narrow_ptrs_base(Universe::narrow_oop_base());
   }
-  // Universe::narrow_oop_base() is one page below the metaspace
-  // base. The actual metaspace base depends on alignment constraints
-  // so we don't know its exact location here.
-  assert((intptr_t)Universe::narrow_oop_base() <= (intptr_t)(Universe::heap()->base() - os::vm_page_size() - ClassMetaspaceSize) ||
+  // Universe::narrow_oop_base() is one page below the heap.
+  assert((intptr_t)Universe::narrow_oop_base() <= (intptr_t)(Universe::heap()->base() -
+         os::vm_page_size()) ||
          Universe::narrow_oop_base() == NULL, "invalid value");
   assert(Universe::narrow_oop_shift() == LogMinObjAlignmentInBytes ||
          Universe::narrow_oop_shift() == 0, "invalid value");
@@ -882,12 +864,7 @@
 
 // Reserve the Java heap, which is now the same for all GCs.
 ReservedSpace Universe::reserve_heap(size_t heap_size, size_t alignment) {
-  // Add in the class metaspace area so the classes in the headers can
-  // be compressed the same as instances.
-  // Need to round class space size up because it's below the heap and
-  // the actual alignment depends on its size.
-  Universe::set_class_metaspace_size(align_size_up(ClassMetaspaceSize, alignment));
-  size_t total_reserved = align_size_up(heap_size + Universe::class_metaspace_size(), alignment);
+  size_t total_reserved = align_size_up(heap_size, alignment);
   assert(!UseCompressedOops || (total_reserved <= (OopEncodingHeapMax - os::vm_page_size())),
       "heap size is too big for compressed oops");
   char* addr = Universe::preferred_heap_base(total_reserved, Universe::UnscaledNarrowOop);
@@ -923,28 +900,17 @@
     return total_rs;
   }
 
-  // Split the reserved space into main Java heap and a space for
-  // classes so that they can be compressed using the same algorithm
-  // as compressed oops. If compress oops and compress klass ptrs are
-  // used we need the meta space first: if the alignment used for
-  // compressed oops is greater than the one used for compressed klass
-  // ptrs, a metadata space on top of the heap could become
-  // unreachable.
-  ReservedSpace class_rs = total_rs.first_part(Universe::class_metaspace_size());
-  ReservedSpace heap_rs = total_rs.last_part(Universe::class_metaspace_size(), alignment);
-  Metaspace::initialize_class_space(class_rs);
-
   if (UseCompressedOops) {
     // Universe::initialize_heap() will reset this to NULL if unscaled
     // or zero-based narrow oops are actually used.
     address base = (address)(total_rs.base() - os::vm_page_size());
     Universe::set_narrow_oop_base(base);
   }
-  return heap_rs;
+  return total_rs;
 }
 
 
-// It's the caller's repsonsibility to ensure glitch-freedom
+// It's the caller's responsibility to ensure glitch-freedom
 // (if required).
 void Universe::update_heap_info_at_gc() {
   _heap_capacity_at_last_gc = heap()->capacity();
@@ -1135,6 +1101,8 @@
 
   // Initialize performance counters for metaspaces
   MetaspaceCounters::initialize_performance_counters();
+  CompressedClassSpaceCounters::initialize_performance_counters();
+
   MemoryService::add_metaspace_memory_pools();
 
   GC_locker::unlock();  // allow gc after bootstrapping
--- a/src/share/vm/memory/universe.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/memory/universe.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -75,10 +75,10 @@
 };
 
 
-// For UseCompressedOops and UseCompressedKlassPointers.
+// For UseCompressedOops.
 struct NarrowPtrStruct {
-  // Base address for oop/klass-within-java-object materialization.
-  // NULL if using wide oops/klasses or zero based narrow oops/klasses.
+  // Base address for oop-within-java-object materialization.
+  // NULL if using wide oops or zero based narrow oops.
   address _base;
   // Number of shift bits for encoding/decoding narrow ptrs.
   // 0 if using wide ptrs or zero based unscaled narrow ptrs,
@@ -106,6 +106,7 @@
   friend class SystemDictionary;
   friend class VMStructs;
   friend class VM_PopulateDumpSharedSpace;
+  friend class Metaspace;
 
   friend jint  universe_init();
   friend void  universe2_init();
@@ -184,9 +185,6 @@
   static struct NarrowPtrStruct _narrow_klass;
   static address _narrow_ptrs_base;
 
-  // Aligned size of the metaspace.
-  static size_t _class_metaspace_size;
-
   // array of dummy objects used with +FullGCAlot
   debug_only(static objArrayOop _fullgc_alot_dummy_array;)
   // index of next entry to clear
@@ -238,15 +236,6 @@
     assert(UseCompressedOops, "no compressed ptrs?");
     _narrow_oop._use_implicit_null_checks   = use;
   }
-  static bool     reserve_metaspace_helper(bool with_base = false);
-  static ReservedHeapSpace reserve_heap_metaspace(size_t heap_size, size_t alignment, bool& contiguous);
-
-  static size_t  class_metaspace_size() {
-    return _class_metaspace_size;
-  }
-  static void    set_class_metaspace_size(size_t metaspace_size) {
-    _class_metaspace_size = metaspace_size;
-  }
 
   // Debugging
   static int _verify_count;                           // number of verifies done
--- a/src/share/vm/oops/instanceKlass.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/oops/instanceKlass.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -269,7 +269,7 @@
   set_fields(NULL, 0);
   set_constants(NULL);
   set_class_loader_data(NULL);
-  set_source_file_name(NULL);
+  set_source_file_name_index(0);
   set_source_debug_extension(NULL, 0);
   set_array_name(NULL);
   set_inner_classes(NULL);
@@ -284,7 +284,7 @@
   set_osr_nmethods_head(NULL);
   set_breakpoints(NULL);
   init_previous_versions();
-  set_generic_signature(NULL);
+  set_generic_signature_index(0);
   release_set_methods_jmethod_ids(NULL);
   release_set_methods_cached_itable_indices(NULL);
   set_annotations(NULL);
@@ -2368,18 +2368,12 @@
   // unreference array name derived from this class name (arrays of an unloaded
   // class can't be referenced anymore).
   if (_array_name != NULL)  _array_name->decrement_refcount();
-  if (_source_file_name != NULL) _source_file_name->decrement_refcount();
   if (_source_debug_extension != NULL) FREE_C_HEAP_ARRAY(char, _source_debug_extension, mtClass);
 
   assert(_total_instanceKlass_count >= 1, "Sanity check");
   Atomic::dec(&_total_instanceKlass_count);
 }
 
-void InstanceKlass::set_source_file_name(Symbol* n) {
-  _source_file_name = n;
-  if (_source_file_name != NULL) _source_file_name->increment_refcount();
-}
-
 void InstanceKlass::set_source_debug_extension(char* array, int length) {
   if (array == NULL) {
     _source_debug_extension = NULL;
--- a/src/share/vm/oops/instanceKlass.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/oops/instanceKlass.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -201,14 +201,10 @@
   // number_of_inner_classes * 4 + enclosing_method_attribute_size.
   Array<jushort>* _inner_classes;
 
-  // Name of source file containing this klass, NULL if not specified.
-  Symbol*         _source_file_name;
   // the source debug extension for this klass, NULL if not specified.
   // Specified as UTF-8 string without terminating zero byte in the classfile,
   // it is stored in the instanceklass as a NULL-terminated UTF-8 string
   char*           _source_debug_extension;
-  // Generic signature, or null if none.
-  Symbol*         _generic_signature;
   // Array name derived from this class which needs unreferencing
   // if this class is unloaded.
   Symbol*         _array_name;
@@ -217,6 +213,12 @@
   // (including inherited fields but after header_size()).
   int             _nonstatic_field_size;
   int             _static_field_size;    // number words used by static fields (oop and non-oop) in this klass
+  // Constant pool index to the utf8 entry of the Generic signature,
+  // or 0 if none.
+  u2              _generic_signature_index;
+  // Constant pool index to the utf8 entry for the name of source file
+  // containing this klass, 0 if not specified.
+  u2              _source_file_name_index;
   u2              _static_oop_field_count;// number of static oop fields in this klass
   u2              _java_fields_count;    // The number of declared Java fields
   int             _nonstatic_oop_map_size;// size in words of nonstatic oop map blocks
@@ -570,8 +572,16 @@
   }
 
   // source file name
-  Symbol* source_file_name() const         { return _source_file_name; }
-  void set_source_file_name(Symbol* n);
+  Symbol* source_file_name() const               {
+    return (_source_file_name_index == 0) ?
+      (Symbol*)NULL : _constants->symbol_at(_source_file_name_index);
+  }
+  u2 source_file_name_index() const              {
+    return _source_file_name_index;
+  }
+  void set_source_file_name_index(u2 sourcefile_index) {
+    _source_file_name_index = sourcefile_index;
+  }
 
   // minor and major version numbers of class file
   u2 minor_version() const                 { return _minor_version; }
@@ -648,8 +658,16 @@
   void set_initial_method_idnum(u2 value)             { _idnum_allocated_count = value; }
 
   // generics support
-  Symbol* generic_signature() const                   { return _generic_signature; }
-  void set_generic_signature(Symbol* sig)             { _generic_signature = sig; }
+  Symbol* generic_signature() const                   {
+    return (_generic_signature_index == 0) ?
+      (Symbol*)NULL : _constants->symbol_at(_generic_signature_index);
+  }
+  u2 generic_signature_index() const                  {
+    return _generic_signature_index;
+  }
+  void set_generic_signature_index(u2 sig_index)      {
+    _generic_signature_index = sig_index;
+  }
 
   u2 enclosing_method_data(int offset);
   u2 enclosing_method_class_index() {
--- a/src/share/vm/oops/klass.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/oops/klass.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -352,7 +352,8 @@
   static int layout_helper_log2_element_size(jint lh) {
     assert(lh < (jint)_lh_neutral_value, "must be array");
     int l2esz = (lh >> _lh_log2_element_size_shift) & _lh_log2_element_size_mask;
-    assert(l2esz <= LogBitsPerLong, "sanity");
+    assert(l2esz <= LogBitsPerLong,
+        err_msg("sanity. l2esz: 0x%x for lh: 0x%x", (uint)l2esz, (uint)lh));
     return l2esz;
   }
   static jint array_layout_helper(jint tag, int hsize, BasicType etype, int log2_esize) {
@@ -703,6 +704,16 @@
 
   virtual void oop_verify_on(oop obj, outputStream* st);
 
+  static bool is_null(narrowKlass obj);
+  static bool is_null(Klass* obj);
+
+  // klass encoding for klass pointer in objects.
+  static narrowKlass encode_klass_not_null(Klass* v);
+  static narrowKlass encode_klass(Klass* v);
+
+  static Klass* decode_klass_not_null(narrowKlass v);
+  static Klass* decode_klass(narrowKlass v);
+
  private:
   // barriers used by klass_oop_store
   void klass_update_barrier_set(oop v);
--- a/src/share/vm/oops/klass.inline.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/oops/klass.inline.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2005, 2013, 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
@@ -25,6 +25,7 @@
 #ifndef SHARE_VM_OOPS_KLASS_INLINE_HPP
 #define SHARE_VM_OOPS_KLASS_INLINE_HPP
 
+#include "memory/universe.hpp"
 #include "oops/klass.hpp"
 #include "oops/markOop.hpp"
 
@@ -33,4 +34,41 @@
   _prototype_header = header;
 }
 
+inline bool Klass::is_null(Klass* obj)  { return obj == NULL; }
+inline bool Klass::is_null(narrowKlass obj) { return obj == 0; }
+
+// Encoding and decoding for klass field.
+
+inline bool check_klass_alignment(Klass* obj) {
+  return (intptr_t)obj % KlassAlignmentInBytes == 0;
+}
+
+inline narrowKlass Klass::encode_klass_not_null(Klass* v) {
+  assert(!is_null(v), "klass value can never be zero");
+  assert(check_klass_alignment(v), "Address not aligned");
+  int    shift = Universe::narrow_klass_shift();
+  uint64_t pd = (uint64_t)(pointer_delta((void*)v, Universe::narrow_klass_base(), 1));
+  assert(KlassEncodingMetaspaceMax > pd, "change encoding max if new encoding");
+  uint64_t result = pd >> shift;
+  assert((result & CONST64(0xffffffff00000000)) == 0, "narrow klass pointer overflow");
+  assert(decode_klass(result) == v, "reversibility");
+  return (narrowKlass)result;
+}
+
+inline narrowKlass Klass::encode_klass(Klass* v) {
+  return is_null(v) ? (narrowKlass)0 : encode_klass_not_null(v);
+}
+
+inline Klass* Klass::decode_klass_not_null(narrowKlass v) {
+  assert(!is_null(v), "narrow klass value can never be zero");
+  int    shift = Universe::narrow_klass_shift();
+  Klass* result = (Klass*)(void*)((uintptr_t)Universe::narrow_klass_base() + ((uintptr_t)v << shift));
+  assert(check_klass_alignment(result), err_msg("address not aligned: " PTR_FORMAT, (void*) result));
+  return result;
+}
+
+inline Klass* Klass::decode_klass(narrowKlass v) {
+  return is_null(v) ? (Klass*)NULL : decode_klass_not_null(v);
+}
+
 #endif // SHARE_VM_OOPS_KLASS_INLINE_HPP
--- a/src/share/vm/oops/method.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/oops/method.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -747,6 +747,7 @@
       set_not_c2_compilable();
   }
   CompilationPolicy::policy()->disable_compilation(this);
+  assert(!CompilationPolicy::can_be_compiled(this, comp_level), "sanity check");
 }
 
 bool Method::is_not_osr_compilable(int comp_level) const {
@@ -773,6 +774,7 @@
       set_not_c2_osr_compilable();
   }
   CompilationPolicy::policy()->disable_compilation(this);
+  assert(!CompilationPolicy::can_be_osr_compiled(this, comp_level), "sanity check");
 }
 
 // Revert to using the interpreter and clear out the nmethod
--- a/src/share/vm/oops/oop.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/oops/oop.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2012, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2013, 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
@@ -62,7 +62,7 @@
   volatile markOop  _mark;
   union _metadata {
     Klass*      _klass;
-    narrowOop       _compressed_klass;
+    narrowKlass _compressed_klass;
   } _metadata;
 
   // Fast access to barrier set.  Must be initialized.
@@ -84,7 +84,7 @@
   Klass* klass() const;
   Klass* klass_or_null() const volatile;
   Klass** klass_addr();
-  narrowOop* compressed_klass_addr();
+  narrowKlass* compressed_klass_addr();
 
   void set_klass(Klass* k);
 
@@ -189,13 +189,6 @@
                                          oop compare_value,
                                          bool prebarrier = false);
 
-  // klass encoding for klass pointer in objects.
-  static narrowOop encode_klass_not_null(Klass* v);
-  static narrowOop encode_klass(Klass* v);
-
-  static Klass* decode_klass_not_null(narrowOop v);
-  static Klass* decode_klass(narrowOop v);
-
   // Access to fields in a instanceOop through these methods.
   oop obj_field(int offset) const;
   volatile oop obj_field_volatile(int offset) const;
--- a/src/share/vm/oops/oop.inline.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/oops/oop.inline.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -35,7 +35,7 @@
 #include "memory/specialized_oop_closures.hpp"
 #include "oops/arrayKlass.hpp"
 #include "oops/arrayOop.hpp"
-#include "oops/klass.hpp"
+#include "oops/klass.inline.hpp"
 #include "oops/markOop.inline.hpp"
 #include "oops/oop.hpp"
 #include "runtime/atomic.hpp"
@@ -70,7 +70,7 @@
 
 inline Klass* oopDesc::klass() const {
   if (UseCompressedKlassPointers) {
-    return decode_klass_not_null(_metadata._compressed_klass);
+    return Klass::decode_klass_not_null(_metadata._compressed_klass);
   } else {
     return _metadata._klass;
   }
@@ -79,7 +79,7 @@
 inline Klass* oopDesc::klass_or_null() const volatile {
   // can be NULL in CMS
   if (UseCompressedKlassPointers) {
-    return decode_klass(_metadata._compressed_klass);
+    return Klass::decode_klass(_metadata._compressed_klass);
   } else {
     return _metadata._klass;
   }
@@ -87,7 +87,7 @@
 
 inline int oopDesc::klass_gap_offset_in_bytes() {
   assert(UseCompressedKlassPointers, "only applicable to compressed klass pointers");
-  return oopDesc::klass_offset_in_bytes() + sizeof(narrowOop);
+  return oopDesc::klass_offset_in_bytes() + sizeof(narrowKlass);
 }
 
 inline Klass** oopDesc::klass_addr() {
@@ -97,9 +97,9 @@
   return (Klass**) &_metadata._klass;
 }
 
-inline narrowOop* oopDesc::compressed_klass_addr() {
+inline narrowKlass* oopDesc::compressed_klass_addr() {
   assert(UseCompressedKlassPointers, "only called by compressed klass pointers");
-  return (narrowOop*) &_metadata._compressed_klass;
+  return &_metadata._compressed_klass;
 }
 
 inline void oopDesc::set_klass(Klass* k) {
@@ -107,7 +107,7 @@
   assert(Universe::is_bootstrapping() || k != NULL, "must be a real Klass*");
   assert(Universe::is_bootstrapping() || k->is_klass(), "not a Klass*");
   if (UseCompressedKlassPointers) {
-    *compressed_klass_addr() = encode_klass_not_null(k);
+    *compressed_klass_addr() = Klass::encode_klass_not_null(k);
   } else {
     *klass_addr() = k;
   }
@@ -127,7 +127,7 @@
   // This is only to be used during GC, for from-space objects, so no
   // barrier is needed.
   if (UseCompressedKlassPointers) {
-    _metadata._compressed_klass = encode_heap_oop(k);  // may be null (parnew overflow handling)
+    _metadata._compressed_klass = (narrowKlass)encode_heap_oop(k);  // may be null (parnew overflow handling)
   } else {
     _metadata._klass = (Klass*)(address)k;
   }
@@ -136,7 +136,7 @@
 inline oop oopDesc::list_ptr_from_klass() {
   // This is only to be used during GC, for from-space objects.
   if (UseCompressedKlassPointers) {
-    return decode_heap_oop(_metadata._compressed_klass);
+    return decode_heap_oop((narrowOop)_metadata._compressed_klass);
   } else {
     // Special case for GC
     return (oop)(address)_metadata._klass;
@@ -176,7 +176,6 @@
 // the right type and inlines the appopriate code).
 
 inline bool oopDesc::is_null(oop obj)       { return obj == NULL; }
-inline bool oopDesc::is_null(Klass* obj)  { return obj == NULL; }
 inline bool oopDesc::is_null(narrowOop obj) { return obj == 0; }
 
 // Algorithm for encoding and decoding oops from 64 bit pointers to 32 bit
@@ -186,9 +185,6 @@
 inline bool check_obj_alignment(oop obj) {
   return (intptr_t)obj % MinObjAlignmentInBytes == 0;
 }
-inline bool check_klass_alignment(Klass* obj) {
-  return (intptr_t)obj % KlassAlignmentInBytes == 0;
-}
 
 inline narrowOop oopDesc::encode_heap_oop_not_null(oop v) {
   assert(!is_null(v), "oop value can never be zero");
@@ -224,39 +220,6 @@
 inline oop oopDesc::decode_heap_oop_not_null(oop v) { return v; }
 inline oop oopDesc::decode_heap_oop(oop v)  { return v; }
 
-// Encoding and decoding for klass field.  It is copied code, but someday
-// might not be the same as oop.
-
-inline narrowOop oopDesc::encode_klass_not_null(Klass* v) {
-  assert(!is_null(v), "klass value can never be zero");
-  assert(check_klass_alignment(v), "Address not aligned");
-  address base = Universe::narrow_klass_base();
-  int    shift = Universe::narrow_klass_shift();
-  uint64_t  pd = (uint64_t)(pointer_delta((void*)v, (void*)base, 1));
-  assert(KlassEncodingMetaspaceMax > pd, "change encoding max if new encoding");
-  uint64_t result = pd >> shift;
-  assert((result & CONST64(0xffffffff00000000)) == 0, "narrow klass pointer overflow");
-  assert(decode_klass(result) == v, "reversibility");
-  return (narrowOop)result;
-}
-
-inline narrowOop oopDesc::encode_klass(Klass* v) {
-  return (is_null(v)) ? (narrowOop)0 : encode_klass_not_null(v);
-}
-
-inline Klass* oopDesc::decode_klass_not_null(narrowOop v) {
-  assert(!is_null(v), "narrow oop value can never be zero");
-  address base = Universe::narrow_klass_base();
-  int    shift = Universe::narrow_klass_shift();
-  Klass* result = (Klass*)(void*)((uintptr_t)base + ((uintptr_t)v << shift));
-  assert(check_klass_alignment(result), err_msg("address not aligned: " PTR_FORMAT, (void*) result));
-  return result;
-}
-
-inline Klass* oopDesc::decode_klass(narrowOop v) {
-  return is_null(v) ? (Klass*)NULL : decode_klass_not_null(v);
-}
-
 // Load an oop out of the Java heap as is without decoding.
 // Called by GC to check for null before decoding.
 inline oop       oopDesc::load_heap_oop(oop* p)          { return *p; }
--- a/src/share/vm/oops/oopsHierarchy.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/oops/oopsHierarchy.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2012, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2013, 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
@@ -33,6 +33,10 @@
 // of B, A's representation is a prefix of B's representation.
 
 typedef juint narrowOop; // Offset instead of address for an oop within a java object
+
+// If compressed klass pointers then use narrowKlass.
+typedef juint  narrowKlass;
+
 typedef void* OopOrNarrowOopStar;
 typedef class   markOopDesc*                markOop;
 
--- a/src/share/vm/opto/block.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/opto/block.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -35,10 +35,6 @@
 #include "opto/rootnode.hpp"
 #include "utilities/copy.hpp"
 
-// Optimization - Graph Style
-
-
-//-----------------------------------------------------------------------------
 void Block_Array::grow( uint i ) {
   assert(i >= Max(), "must be an overflow");
   debug_only(_limit = i+1);
@@ -54,7 +50,6 @@
   Copy::zero_to_bytes( &_blocks[old], (_size-old)*sizeof(Block*) );
 }
 
-//=============================================================================
 void Block_List::remove(uint i) {
   assert(i < _cnt, "index out of bounds");
   Copy::conjoint_words_to_lower((HeapWord*)&_blocks[i+1], (HeapWord*)&_blocks[i], ((_cnt-i-1)*sizeof(Block*)));
@@ -76,8 +71,6 @@
 }
 #endif
 
-//=============================================================================
-
 uint Block::code_alignment() {
   // Check for Root block
   if (_pre_order == 0) return CodeEntryAlignment;
@@ -113,7 +106,6 @@
   return unit_sz; // no particular alignment
 }
 
-//-----------------------------------------------------------------------------
 // Compute the size of first 'inst_cnt' instructions in this block.
 // Return the number of instructions left to compute if the block has
 // less then 'inst_cnt' instructions. Stop, and return 0 if sum_size
@@ -138,7 +130,6 @@
   return inst_cnt;
 }
 
-//-----------------------------------------------------------------------------
 uint Block::find_node( const Node *n ) const {
   for( uint i = 0; i < _nodes.size(); i++ ) {
     if( _nodes[i] == n )
@@ -153,7 +144,6 @@
   _nodes.remove(find_node(n));
 }
 
-//------------------------------is_Empty---------------------------------------
 // Return empty status of a block.  Empty blocks contain only the head, other
 // ideal nodes, and an optional trailing goto.
 int Block::is_Empty() const {
@@ -192,7 +182,6 @@
   return not_empty;
 }
 
-//------------------------------has_uncommon_code------------------------------
 // Return true if the block's code implies that it is likely to be
 // executed infrequently.  Check to see if the block ends in a Halt or
 // a low probability call.
@@ -218,7 +207,6 @@
   return op == Op_Halt;
 }
 
-//------------------------------is_uncommon------------------------------------
 // True if block is low enough frequency or guarded by a test which
 // mostly does not go here.
 bool Block::is_uncommon(PhaseCFG* cfg) const {
@@ -271,7 +259,6 @@
   return false;
 }
 
-//------------------------------dump-------------------------------------------
 #ifndef PRODUCT
 void Block::dump_bidx(const Block* orig, outputStream* st) const {
   if (_pre_order) st->print("B%d",_pre_order);
@@ -364,13 +351,12 @@
 }
 #endif
 
-//=============================================================================
-//------------------------------PhaseCFG---------------------------------------
 PhaseCFG::PhaseCFG(Arena* arena, RootNode* root, Matcher& matcher)
 : Phase(CFG)
 , _block_arena(arena)
+, _root(root)
+, _matcher(matcher)
 , _node_to_block_mapping(arena)
-, _root(root)
 , _node_latency(NULL)
 #ifndef PRODUCT
 , _trace_opto_pipelining(TraceOptoPipelining || C->method_has_option("TraceOptoPipelining"))
@@ -390,11 +376,10 @@
   _goto->set_req(0,_goto);
 
   // Build the CFG in Reverse Post Order
-  _num_blocks = build_cfg();
-  _broot = get_block_for_node(_root);
+  _number_of_blocks = build_cfg();
+  _root_block = get_block_for_node(_root);
 }
 
-//------------------------------build_cfg--------------------------------------
 // Build a proper looking CFG.  Make every block begin with either a StartNode
 // or a RegionNode.  Make every block end with either a Goto, If or Return.
 // The RootNode both starts and ends it's own block.  Do this with a recursive
@@ -496,13 +481,12 @@
   return sum;
 }
 
-//------------------------------insert_goto_at---------------------------------
 // Inserts a goto & corresponding basic block between
 // block[block_no] and its succ_no'th successor block
 void PhaseCFG::insert_goto_at(uint block_no, uint succ_no) {
   // get block with block_no
-  assert(block_no < _num_blocks, "illegal block number");
-  Block* in  = _blocks[block_no];
+  assert(block_no < number_of_blocks(), "illegal block number");
+  Block* in  = get_block(block_no);
   // get successor block succ_no
   assert(succ_no < in->_num_succs, "illegal successor number");
   Block* out = in->_succs[succ_no];
@@ -537,11 +521,9 @@
   // Set the frequency of the new block
   block->_freq = freq;
   // add new basic block to basic block list
-  _blocks.insert(block_no + 1, block);
-  _num_blocks++;
+  add_block_at(block_no + 1, block);
 }
 
-//------------------------------no_flip_branch---------------------------------
 // Does this block end in a multiway branch that cannot have the default case
 // flipped for another case?
 static bool no_flip_branch( Block *b ) {
@@ -560,7 +542,6 @@
   return false;
 }
 
-//------------------------------convert_NeverBranch_to_Goto--------------------
 // Check for NeverBranch at block end.  This needs to become a GOTO to the
 // true target.  NeverBranch are treated as a conditional branch that always
 // goes the same direction for most of the optimizer and are used to give a
@@ -598,7 +579,6 @@
     dead->_nodes[k]->del_req(j);
 }
 
-//------------------------------move_to_next-----------------------------------
 // Helper function to move block bx to the slot following b_index. Return
 // true if the move is successful, otherwise false
 bool PhaseCFG::move_to_next(Block* bx, uint b_index) {
@@ -606,20 +586,22 @@
 
   // Return false if bx is already scheduled.
   uint bx_index = bx->_pre_order;
-  if ((bx_index <= b_index) && (_blocks[bx_index] == bx)) {
+  if ((bx_index <= b_index) && (get_block(bx_index) == bx)) {
     return false;
   }
 
   // Find the current index of block bx on the block list
   bx_index = b_index + 1;
-  while( bx_index < _num_blocks && _blocks[bx_index] != bx ) bx_index++;
-  assert(_blocks[bx_index] == bx, "block not found");
+  while (bx_index < number_of_blocks() && get_block(bx_index) != bx) {
+    bx_index++;
+  }
+  assert(get_block(bx_index) == bx, "block not found");
 
   // If the previous block conditionally falls into bx, return false,
   // because moving bx will create an extra jump.
   for(uint k = 1; k < bx->num_preds(); k++ ) {
     Block* pred = get_block_for_node(bx->pred(k));
-    if (pred == _blocks[bx_index-1]) {
+    if (pred == get_block(bx_index - 1)) {
       if (pred->_num_succs != 1) {
         return false;
       }
@@ -632,7 +614,6 @@
   return true;
 }
 
-//------------------------------move_to_end------------------------------------
 // Move empty and uncommon blocks to the end.
 void PhaseCFG::move_to_end(Block *b, uint i) {
   int e = b->is_Empty();
@@ -650,31 +631,31 @@
   _blocks.push(b);
 }
 
-//---------------------------set_loop_alignment--------------------------------
 // Set loop alignment for every block
 void PhaseCFG::set_loop_alignment() {
-  uint last = _num_blocks;
-  assert( _blocks[0] == _broot, "" );
+  uint last = number_of_blocks();
+  assert(get_block(0) == get_root_block(), "");
 
-  for (uint i = 1; i < last; i++ ) {
-    Block *b = _blocks[i];
-    if (b->head()->is_Loop()) {
-      b->set_loop_alignment(b);
+  for (uint i = 1; i < last; i++) {
+    Block* block = get_block(i);
+    if (block->head()->is_Loop()) {
+      block->set_loop_alignment(block);
     }
   }
 }
 
-//-----------------------------remove_empty------------------------------------
 // Make empty basic blocks to be "connector" blocks, Move uncommon blocks
 // to the end.
-void PhaseCFG::remove_empty() {
+void PhaseCFG::remove_empty_blocks() {
   // Move uncommon blocks to the end
-  uint last = _num_blocks;
-  assert( _blocks[0] == _broot, "" );
+  uint last = number_of_blocks();
+  assert(get_block(0) == get_root_block(), "");
 
   for (uint i = 1; i < last; i++) {
-    Block *b = _blocks[i];
-    if (b->is_connector()) break;
+    Block* block = get_block(i);
+    if (block->is_connector()) {
+      break;
+    }
 
     // Check for NeverBranch at block end.  This needs to become a GOTO to the
     // true target.  NeverBranch are treated as a conditional branch that
@@ -682,124 +663,127 @@
     // to give a fake exit path to infinite loops.  At this late stage they
     // need to turn into Goto's so that when you enter the infinite loop you
     // indeed hang.
-    if( b->_nodes[b->end_idx()]->Opcode() == Op_NeverBranch )
-      convert_NeverBranch_to_Goto(b);
+    if (block->_nodes[block->end_idx()]->Opcode() == Op_NeverBranch) {
+      convert_NeverBranch_to_Goto(block);
+    }
 
     // Look for uncommon blocks and move to end.
     if (!C->do_freq_based_layout()) {
-      if (b->is_uncommon(this)) {
-        move_to_end(b, i);
+      if (block->is_uncommon(this)) {
+        move_to_end(block, i);
         last--;                   // No longer check for being uncommon!
-        if( no_flip_branch(b) ) { // Fall-thru case must follow?
-          b = _blocks[i];         // Find the fall-thru block
-          move_to_end(b, i);
+        if (no_flip_branch(block)) { // Fall-thru case must follow?
+          // Find the fall-thru block
+          block = get_block(i);
+          move_to_end(block, i);
           last--;
         }
-        i--;                      // backup block counter post-increment
+        // backup block counter post-increment
+        i--;
       }
     }
   }
 
   // Move empty blocks to the end
-  last = _num_blocks;
+  last = number_of_blocks();
   for (uint i = 1; i < last; i++) {
-    Block *b = _blocks[i];
-    if (b->is_Empty() != Block::not_empty) {
-      move_to_end(b, i);
+    Block* block = get_block(i);
+    if (block->is_Empty() != Block::not_empty) {
+      move_to_end(block, i);
       last--;
       i--;
     }
   } // End of for all blocks
 }
 
-//-----------------------------fixup_flow--------------------------------------
 // Fix up the final control flow for basic blocks.
 void PhaseCFG::fixup_flow() {
   // Fixup final control flow for the blocks.  Remove jump-to-next
   // block.  If neither arm of a IF follows the conditional branch, we
   // have to add a second jump after the conditional.  We place the
   // TRUE branch target in succs[0] for both GOTOs and IFs.
-  for (uint i=0; i < _num_blocks; i++) {
-    Block *b = _blocks[i];
-    b->_pre_order = i;          // turn pre-order into block-index
+  for (uint i = 0; i < number_of_blocks(); i++) {
+    Block* block = get_block(i);
+    block->_pre_order = i;          // turn pre-order into block-index
 
     // Connector blocks need no further processing.
-    if (b->is_connector()) {
-      assert((i+1) == _num_blocks || _blocks[i+1]->is_connector(),
-             "All connector blocks should sink to the end");
+    if (block->is_connector()) {
+      assert((i+1) == number_of_blocks() || get_block(i + 1)->is_connector(), "All connector blocks should sink to the end");
       continue;
     }
-    assert(b->is_Empty() != Block::completely_empty,
-           "Empty blocks should be connectors");
+    assert(block->is_Empty() != Block::completely_empty, "Empty blocks should be connectors");
 
-    Block *bnext = (i < _num_blocks-1) ? _blocks[i+1] : NULL;
-    Block *bs0 = b->non_connector_successor(0);
+    Block* bnext = (i < number_of_blocks() - 1) ? get_block(i + 1) : NULL;
+    Block* bs0 = block->non_connector_successor(0);
 
     // Check for multi-way branches where I cannot negate the test to
     // exchange the true and false targets.
-    if( no_flip_branch( b ) ) {
+    if (no_flip_branch(block)) {
       // Find fall through case - if must fall into its target
-      int branch_idx = b->_nodes.size() - b->_num_succs;
-      for (uint j2 = 0; j2 < b->_num_succs; j2++) {
-        const ProjNode* p = b->_nodes[branch_idx + j2]->as_Proj();
+      int branch_idx = block->_nodes.size() - block->_num_succs;
+      for (uint j2 = 0; j2 < block->_num_succs; j2++) {
+        const ProjNode* p = block->_nodes[branch_idx + j2]->as_Proj();
         if (p->_con == 0) {
           // successor j2 is fall through case
-          if (b->non_connector_successor(j2) != bnext) {
+          if (block->non_connector_successor(j2) != bnext) {
             // but it is not the next block => insert a goto
             insert_goto_at(i, j2);
           }
           // Put taken branch in slot 0
-          if( j2 == 0 && b->_num_succs == 2) {
+          if (j2 == 0 && block->_num_succs == 2) {
             // Flip targets in succs map
-            Block *tbs0 = b->_succs[0];
-            Block *tbs1 = b->_succs[1];
-            b->_succs.map( 0, tbs1 );
-            b->_succs.map( 1, tbs0 );
+            Block *tbs0 = block->_succs[0];
+            Block *tbs1 = block->_succs[1];
+            block->_succs.map(0, tbs1);
+            block->_succs.map(1, tbs0);
           }
           break;
         }
       }
+
       // Remove all CatchProjs
-      for (uint j1 = 0; j1 < b->_num_succs; j1++) b->_nodes.pop();
+      for (uint j = 0; j < block->_num_succs; j++) {
+        block->_nodes.pop();
+      }
 
-    } else if (b->_num_succs == 1) {
+    } else if (block->_num_succs == 1) {
       // Block ends in a Goto?
       if (bnext == bs0) {
         // We fall into next block; remove the Goto
-        b->_nodes.pop();
+        block->_nodes.pop();
       }
 
-    } else if( b->_num_succs == 2 ) { // Block ends in a If?
+    } else if(block->_num_succs == 2) { // Block ends in a If?
       // Get opcode of 1st projection (matches _succs[0])
       // Note: Since this basic block has 2 exits, the last 2 nodes must
       //       be projections (in any order), the 3rd last node must be
       //       the IfNode (we have excluded other 2-way exits such as
       //       CatchNodes already).
-      MachNode *iff   = b->_nodes[b->_nodes.size()-3]->as_Mach();
-      ProjNode *proj0 = b->_nodes[b->_nodes.size()-2]->as_Proj();
-      ProjNode *proj1 = b->_nodes[b->_nodes.size()-1]->as_Proj();
+      MachNode* iff   = block->_nodes[block->_nodes.size() - 3]->as_Mach();
+      ProjNode* proj0 = block->_nodes[block->_nodes.size() - 2]->as_Proj();
+      ProjNode* proj1 = block->_nodes[block->_nodes.size() - 1]->as_Proj();
 
       // Assert that proj0 and succs[0] match up. Similarly for proj1 and succs[1].
-      assert(proj0->raw_out(0) == b->_succs[0]->head(), "Mismatch successor 0");
-      assert(proj1->raw_out(0) == b->_succs[1]->head(), "Mismatch successor 1");
+      assert(proj0->raw_out(0) == block->_succs[0]->head(), "Mismatch successor 0");
+      assert(proj1->raw_out(0) == block->_succs[1]->head(), "Mismatch successor 1");
 
-      Block *bs1 = b->non_connector_successor(1);
+      Block* bs1 = block->non_connector_successor(1);
 
       // Check for neither successor block following the current
       // block ending in a conditional. If so, move one of the
       // successors after the current one, provided that the
       // successor was previously unscheduled, but moveable
       // (i.e., all paths to it involve a branch).
-      if( !C->do_freq_based_layout() && bnext != bs0 && bnext != bs1 ) {
+      if (!C->do_freq_based_layout() && bnext != bs0 && bnext != bs1) {
         // Choose the more common successor based on the probability
         // of the conditional branch.
-        Block *bx = bs0;
-        Block *by = bs1;
+        Block* bx = bs0;
+        Block* by = bs1;
 
         // _prob is the probability of taking the true path. Make
         // p the probability of taking successor #1.
         float p = iff->as_MachIf()->_prob;
-        if( proj0->Opcode() == Op_IfTrue ) {
+        if (proj0->Opcode() == Op_IfTrue) {
           p = 1.0 - p;
         }
 
@@ -826,14 +810,16 @@
       // succs[1].
       if (bnext == bs0) {
         // Fall-thru case in succs[0], so flip targets in succs map
-        Block *tbs0 = b->_succs[0];
-        Block *tbs1 = b->_succs[1];
-        b->_succs.map( 0, tbs1 );
-        b->_succs.map( 1, tbs0 );
+        Block* tbs0 = block->_succs[0];
+        Block* tbs1 = block->_succs[1];
+        block->_succs.map(0, tbs1);
+        block->_succs.map(1, tbs0);
         // Flip projection for each target
-        { ProjNode *tmp = proj0; proj0 = proj1; proj1 = tmp; }
+        ProjNode* tmp = proj0;
+        proj0 = proj1;
+        proj1 = tmp;
 
-      } else if( bnext != bs1 ) {
+      } else if(bnext != bs1) {
         // Need a double-branch
         // The existing conditional branch need not change.
         // Add a unconditional branch to the false target.
@@ -843,12 +829,12 @@
       }
 
       // Make sure we TRUE branch to the target
-      if( proj0->Opcode() == Op_IfFalse ) {
+      if (proj0->Opcode() == Op_IfFalse) {
         iff->as_MachIf()->negate();
       }
 
-      b->_nodes.pop();          // Remove IfFalse & IfTrue projections
-      b->_nodes.pop();
+      block->_nodes.pop();          // Remove IfFalse & IfTrue projections
+      block->_nodes.pop();
 
     } else {
       // Multi-exit block, e.g. a switch statement
@@ -858,7 +844,6 @@
 }
 
 
-//------------------------------dump-------------------------------------------
 #ifndef PRODUCT
 void PhaseCFG::_dump_cfg( const Node *end, VectorSet &visited  ) const {
   const Node *x = end->is_block_proj();
@@ -884,10 +869,11 @@
 }
 
 void PhaseCFG::dump( ) const {
-  tty->print("\n--- CFG --- %d BBs\n",_num_blocks);
+  tty->print("\n--- CFG --- %d BBs\n", number_of_blocks());
   if (_blocks.size()) {        // Did we do basic-block layout?
-    for (uint i = 0; i < _num_blocks; i++) {
-      _blocks[i]->dump(this);
+    for (uint i = 0; i < number_of_blocks(); i++) {
+      const Block* block = get_block(i);
+      block->dump(this);
     }
   } else {                      // Else do it with a DFS
     VectorSet visited(_block_arena);
@@ -896,27 +882,26 @@
 }
 
 void PhaseCFG::dump_headers() {
-  for( uint i = 0; i < _num_blocks; i++ ) {
-    if (_blocks[i]) {
-      _blocks[i]->dump_head(this);
+  for (uint i = 0; i < number_of_blocks(); i++) {
+    Block* block = get_block(i);
+    if (block != NULL) {
+      block->dump_head(this);
     }
   }
 }
 
-void PhaseCFG::verify( ) const {
+void PhaseCFG::verify() const {
 #ifdef ASSERT
   // Verify sane CFG
-  for (uint i = 0; i < _num_blocks; i++) {
-    Block *b = _blocks[i];
-    uint cnt = b->_nodes.size();
+  for (uint i = 0; i < number_of_blocks(); i++) {
+    Block* block = get_block(i);
+    uint cnt = block->_nodes.size();
     uint j;
     for (j = 0; j < cnt; j++)  {
-      Node *n = b->_nodes[j];
-      assert(get_block_for_node(n) == b, "");
-      if (j >= 1 && n->is_Mach() &&
-          n->as_Mach()->ideal_Opcode() == Op_CreateEx) {
-        assert(j == 1 || b->_nodes[j-1]->is_Phi(),
-               "CreateEx must be first instruction in block");
+      Node *n = block->_nodes[j];
+      assert(get_block_for_node(n) == block, "");
+      if (j >= 1 && n->is_Mach() && n->as_Mach()->ideal_Opcode() == Op_CreateEx) {
+        assert(j == 1 || block->_nodes[j-1]->is_Phi(), "CreateEx must be first instruction in block");
       }
       for (uint k = 0; k < n->req(); k++) {
         Node *def = n->in(k);
@@ -926,8 +911,7 @@
           // Uses must follow their definition if they are at the same block.
           // Mostly done to check that MachSpillCopy nodes are placed correctly
           // when CreateEx node is moved in build_ifg_physical().
-          if (get_block_for_node(def) == b &&
-              !(b->head()->is_Loop() && n->is_Phi()) &&
+          if (get_block_for_node(def) == block && !(block->head()->is_Loop() && n->is_Phi()) &&
               // See (+++) comment in reg_split.cpp
               !(n->jvms() != NULL && n->jvms()->is_monitor_use(k))) {
             bool is_loop = false;
@@ -939,29 +923,29 @@
                 }
               }
             }
-            assert(is_loop || b->find_node(def) < j, "uses must follow definitions");
+            assert(is_loop || block->find_node(def) < j, "uses must follow definitions");
           }
         }
       }
     }
 
-    j = b->end_idx();
-    Node *bp = (Node*)b->_nodes[b->_nodes.size()-1]->is_block_proj();
-    assert( bp, "last instruction must be a block proj" );
-    assert( bp == b->_nodes[j], "wrong number of successors for this block" );
+    j = block->end_idx();
+    Node* bp = (Node*)block->_nodes[block->_nodes.size() - 1]->is_block_proj();
+    assert(bp, "last instruction must be a block proj");
+    assert(bp == block->_nodes[j], "wrong number of successors for this block");
     if (bp->is_Catch()) {
-      while (b->_nodes[--j]->is_MachProj()) ;
-      assert(b->_nodes[j]->is_MachCall(), "CatchProj must follow call");
+      while (block->_nodes[--j]->is_MachProj()) {
+        ;
+      }
+      assert(block->_nodes[j]->is_MachCall(), "CatchProj must follow call");
     } else if (bp->is_Mach() && bp->as_Mach()->ideal_Opcode() == Op_If) {
-      assert(b->_num_succs == 2, "Conditional branch must have two targets");
+      assert(block->_num_succs == 2, "Conditional branch must have two targets");
     }
   }
 #endif
 }
 #endif
 
-//=============================================================================
-//------------------------------UnionFind--------------------------------------
 UnionFind::UnionFind( uint max ) : _cnt(max), _max(max), _indices(NEW_RESOURCE_ARRAY(uint,max)) {
   Copy::zero_to_bytes( _indices, sizeof(uint)*max );
 }
@@ -986,7 +970,6 @@
   for( uint i=0; i<max; i++ ) map(i,i);
 }
 
-//------------------------------Find_compress----------------------------------
 // Straight out of Tarjan's union-find algorithm
 uint UnionFind::Find_compress( uint idx ) {
   uint cur  = idx;
@@ -1006,7 +989,6 @@
   return idx;
 }
 
-//------------------------------Find_const-------------------------------------
 // Like Find above, but no path compress, so bad asymptotic behavior
 uint UnionFind::Find_const( uint idx ) const {
   if( idx == 0 ) return idx;    // Ignore the zero idx
@@ -1021,7 +1003,6 @@
   return next;
 }
 
-//------------------------------Union------------------------------------------
 // union 2 sets together.
 void UnionFind::Union( uint idx1, uint idx2 ) {
   uint src = Find(idx1);
@@ -1070,9 +1051,6 @@
 }
 #endif
 
-//=============================================================================
-
-//------------------------------edge_order-------------------------------------
 // Comparison function for edges
 static int edge_order(CFGEdge **e0, CFGEdge **e1) {
   float freq0 = (*e0)->freq();
@@ -1087,7 +1065,6 @@
   return dist1 - dist0;
 }
 
-//------------------------------trace_frequency_order--------------------------
 // Comparison function for edges
 extern "C" int trace_frequency_order(const void *p0, const void *p1) {
   Trace *tr0 = *(Trace **) p0;
@@ -1113,17 +1090,15 @@
   return diff;
 }
 
-//------------------------------find_edges-------------------------------------
 // Find edges of interest, i.e, those which can fall through. Presumes that
 // edges which don't fall through are of low frequency and can be generally
 // ignored.  Initialize the list of traces.
-void PhaseBlockLayout::find_edges()
-{
+void PhaseBlockLayout::find_edges() {
   // Walk the blocks, creating edges and Traces
   uint i;
   Trace *tr = NULL;
-  for (i = 0; i < _cfg._num_blocks; i++) {
-    Block *b = _cfg._blocks[i];
+  for (i = 0; i < _cfg.number_of_blocks(); i++) {
+    Block* b = _cfg.get_block(i);
     tr = new Trace(b, next, prev);
     traces[tr->id()] = tr;
 
@@ -1147,7 +1122,7 @@
       if (n->num_preds() != 1) break;
 
       i++;
-      assert(n = _cfg._blocks[i], "expecting next block");
+      assert(n = _cfg.get_block(i), "expecting next block");
       tr->append(n);
       uf->map(n->_pre_order, tr->id());
       traces[n->_pre_order] = NULL;
@@ -1171,8 +1146,8 @@
   }
 
   // Group connector blocks into one trace
-  for (i++; i < _cfg._num_blocks; i++) {
-    Block *b = _cfg._blocks[i];
+  for (i++; i < _cfg.number_of_blocks(); i++) {
+    Block *b = _cfg.get_block(i);
     assert(b->is_connector(), "connector blocks at the end");
     tr->append(b);
     uf->map(b->_pre_order, tr->id());
@@ -1180,10 +1155,8 @@
   }
 }
 
-//------------------------------union_traces----------------------------------
 // Union two traces together in uf, and null out the trace in the list
-void PhaseBlockLayout::union_traces(Trace* updated_trace, Trace* old_trace)
-{
+void PhaseBlockLayout::union_traces(Trace* updated_trace, Trace* old_trace) {
   uint old_id = old_trace->id();
   uint updated_id = updated_trace->id();
 
@@ -1207,10 +1180,8 @@
   traces[hi_id] = NULL;
 }
 
-//------------------------------grow_traces-------------------------------------
 // Append traces together via the most frequently executed edges
-void PhaseBlockLayout::grow_traces()
-{
+void PhaseBlockLayout::grow_traces() {
   // Order the edges, and drive the growth of Traces via the most
   // frequently executed edges.
   edges->sort(edge_order);
@@ -1252,11 +1223,9 @@
   }
 }
 
-//------------------------------merge_traces-----------------------------------
 // Embed one trace into another, if the fork or join points are sufficiently
 // balanced.
-void PhaseBlockLayout::merge_traces(bool fall_thru_only)
-{
+void PhaseBlockLayout::merge_traces(bool fall_thru_only) {
   // Walk the edge list a another time, looking at unprocessed edges.
   // Fold in diamonds
   for (int i = 0; i < edges->length(); i++) {
@@ -1310,7 +1279,7 @@
         src_trace->insert_after(src_block, targ_trace);
         union_traces(src_trace, targ_trace);
       } else if (src_at_tail) {
-        if (src_trace != trace(_cfg._broot)) {
+        if (src_trace != trace(_cfg.get_root_block())) {
           e->set_state(CFGEdge::connected);
           targ_trace->insert_before(targ_block, src_trace);
           union_traces(targ_trace, src_trace);
@@ -1319,7 +1288,7 @@
     } else if (e->state() == CFGEdge::open) {
       // Append traces, even without a fall-thru connection.
       // But leave root entry at the beginning of the block list.
-      if (targ_trace != trace(_cfg._broot)) {
+      if (targ_trace != trace(_cfg.get_root_block())) {
         e->set_state(CFGEdge::connected);
         src_trace->append(targ_trace);
         union_traces(src_trace, targ_trace);
@@ -1328,11 +1297,9 @@
   }
 }
 
-//----------------------------reorder_traces-----------------------------------
 // Order the sequence of the traces in some desirable way, and fixup the
 // jumps at the end of each block.
-void PhaseBlockLayout::reorder_traces(int count)
-{
+void PhaseBlockLayout::reorder_traces(int count) {
   ResourceArea *area = Thread::current()->resource_area();
   Trace ** new_traces = NEW_ARENA_ARRAY(area, Trace *, count);
   Block_List worklist;
@@ -1347,15 +1314,14 @@
   }
 
   // The entry block should be first on the new trace list.
-  Trace *tr = trace(_cfg._broot);
+  Trace *tr = trace(_cfg.get_root_block());
   assert(tr == new_traces[0], "entry trace misplaced");
 
   // Sort the new trace list by frequency
   qsort(new_traces + 1, new_count - 1, sizeof(new_traces[0]), trace_frequency_order);
 
   // Patch up the successor blocks
-  _cfg._blocks.reset();
-  _cfg._num_blocks = 0;
+  _cfg.clear_blocks();
   for (int i = 0; i < new_count; i++) {
     Trace *tr = new_traces[i];
     if (tr != NULL) {
@@ -1364,17 +1330,15 @@
   }
 }
 
-//------------------------------PhaseBlockLayout-------------------------------
 // Order basic blocks based on frequency
-PhaseBlockLayout::PhaseBlockLayout(PhaseCFG &cfg) :
-  Phase(BlockLayout),
-  _cfg(cfg)
-{
+PhaseBlockLayout::PhaseBlockLayout(PhaseCFG &cfg)
+: Phase(BlockLayout)
+, _cfg(cfg) {
   ResourceMark rm;
   ResourceArea *area = Thread::current()->resource_area();
 
   // List of traces
-  int size = _cfg._num_blocks + 1;
+  int size = _cfg.number_of_blocks() + 1;
   traces = NEW_ARENA_ARRAY(area, Trace *, size);
   memset(traces, 0, size*sizeof(Trace*));
   next = NEW_ARENA_ARRAY(area, Block *, size);
@@ -1407,11 +1371,10 @@
   // Re-order all the remaining traces by frequency
   reorder_traces(size);
 
-  assert(_cfg._num_blocks >= (uint) (size - 1), "number of blocks can not shrink");
+  assert(_cfg.number_of_blocks() >= (uint) (size - 1), "number of blocks can not shrink");
 }
 
 
-//------------------------------backedge---------------------------------------
 // Edge e completes a loop in a trace. If the target block is head of the
 // loop, rotate the loop block so that the loop ends in a conditional branch.
 bool Trace::backedge(CFGEdge *e) {
@@ -1463,14 +1426,12 @@
   return loop_rotated;
 }
 
-//------------------------------fixup_blocks-----------------------------------
 // push blocks onto the CFG list
 // ensure that blocks have the correct two-way branch sense
 void Trace::fixup_blocks(PhaseCFG &cfg) {
   Block *last = last_block();
   for (Block *b = first_block(); b != NULL; b = next(b)) {
-    cfg._blocks.push(b);
-    cfg._num_blocks++;
+    cfg.add_block(b);
     if (!b->is_connector()) {
       int nfallthru = b->num_fall_throughs();
       if (b != last) {
--- a/src/share/vm/opto/block.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/opto/block.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -348,20 +348,77 @@
 class PhaseCFG : public Phase {
   friend class VMStructs;
  private:
+
+  // Root of whole program
+  RootNode* _root;
+
+  // The block containing the root node
+  Block* _root_block;
+
+  // List of basic blocks that are created during CFG creation
+  Block_List _blocks;
+
+  // Count of basic blocks
+  uint _number_of_blocks;
+
   // Arena for the blocks to be stored in
   Arena* _block_arena;
 
+  // The matcher for this compilation
+  Matcher& _matcher;
+
   // Map nodes to owning basic block
   Block_Array _node_to_block_mapping;
 
+  // Loop from the root
+  CFGLoop* _root_loop;
+
+  // Outmost loop frequency
+  float _outer_loop_frequency;
+
+  // Per node latency estimation, valid only during GCM
+  GrowableArray<uint>* _node_latency;
+
   // Build a proper looking cfg.  Return count of basic blocks
   uint build_cfg();
 
-  // Perform DFS search.
+  // Build the dominator tree so that we know where we can move instructions
+  void build_dominator_tree();
+
+  // Estimate block frequencies based on IfNode probabilities, so that we know where we want to move instructions
+  void estimate_block_frequency();
+
+  // Global Code Motion.  See Click's PLDI95 paper.  Place Nodes in specific
+  // basic blocks; i.e. _node_to_block_mapping now maps _idx for all Nodes to some Block.
+  // Move nodes to ensure correctness from GVN and also try to move nodes out of loops.
+  void global_code_motion();
+
+  // Schedule Nodes early in their basic blocks.
+  bool schedule_early(VectorSet &visited, Node_List &roots);
+
+  // For each node, find the latest block it can be scheduled into
+  // and then select the cheapest block between the latest and earliest
+  // block to place the node.
+  void schedule_late(VectorSet &visited, Node_List &stack);
+
+  // Compute the (backwards) latency of a node from a single use
+  int latency_from_use(Node *n, const Node *def, Node *use);
+
+  // Compute the (backwards) latency of a node from the uses of this instruction
+  void partial_latency_of_defs(Node *n);
+
+  // Compute the instruction global latency with a backwards walk
+  void compute_latencies_backwards(VectorSet &visited, Node_List &stack);
+
+  // Pick a block between early and late that is a cheaper alternative
+  // to late. Helper for schedule_late.
+  Block* hoist_to_cheaper_block(Block* LCA, Block* early, Node* self);
+
+  // Perform a Depth First Search (DFS).
   // Setup 'vertex' as DFS to vertex mapping.
   // Setup 'semi' as vertex to DFS mapping.
   // Set 'parent' to DFS parent.
-  uint DFS( Tarjan *tarjan );
+  uint do_DFS(Tarjan* tarjan, uint rpo_counter);
 
   // Helper function to insert a node into a block
   void schedule_node_into_block( Node *n, Block *b );
@@ -372,7 +429,8 @@
   void schedule_pinned_nodes( VectorSet &visited );
 
   // I'll need a few machine-specific GotoNodes.  Clone from this one.
-  MachNode *_goto;
+  // Used when building the CFG and creating end nodes for blocks.
+  MachNode* _goto;
 
   Block* insert_anti_dependences(Block* LCA, Node* load, bool verify = false);
   void verify_anti_dependences(Block* LCA, Node* load) {
@@ -380,17 +438,77 @@
     insert_anti_dependences(LCA, load, true);
   }
 
+  bool move_to_next(Block* bx, uint b_index);
+  void move_to_end(Block* bx, uint b_index);
+
+  void insert_goto_at(uint block_no, uint succ_no);
+
+  // Check for NeverBranch at block end.  This needs to become a GOTO to the
+  // true target.  NeverBranch are treated as a conditional branch that always
+  // goes the same direction for most of the optimizer and are used to give a
+  // fake exit path to infinite loops.  At this late stage they need to turn
+  // into Goto's so that when you enter the infinite loop you indeed hang.
+  void convert_NeverBranch_to_Goto(Block *b);
+
+  CFGLoop* create_loop_tree();
+
+  #ifndef PRODUCT
+  bool _trace_opto_pipelining;  // tracing flag
+  #endif
+
  public:
   PhaseCFG(Arena* arena, RootNode* root, Matcher& matcher);
 
-  uint _num_blocks;             // Count of basic blocks
-  Block_List _blocks;           // List of basic blocks
-  RootNode *_root;              // Root of whole program
-  Block *_broot;                // Basic block of root
-  uint _rpo_ctr;
-  CFGLoop* _root_loop;
-  float _outer_loop_freq;       // Outmost loop frequency
+  void set_latency_for_node(Node* node, int latency) {
+    _node_latency->at_put_grow(node->_idx, latency);
+  }
 
+  uint get_latency_for_node(Node* node) {
+    return _node_latency->at_grow(node->_idx);
+  }
+
+  // Get the outer most frequency
+  float get_outer_loop_frequency() const {
+    return _outer_loop_frequency;
+  }
+
+  // Get the root node of the CFG
+  RootNode* get_root_node() const {
+    return _root;
+  }
+
+  // Get the block of the root node
+  Block* get_root_block() const {
+    return _root_block;
+  }
+
+  // Add a block at a position and moves the later ones one step
+  void add_block_at(uint pos, Block* block) {
+    _blocks.insert(pos, block);
+    _number_of_blocks++;
+  }
+
+  // Adds a block to the top of the block list
+  void add_block(Block* block) {
+    _blocks.push(block);
+    _number_of_blocks++;
+  }
+
+  // Clear the list of blocks
+  void clear_blocks() {
+    _blocks.reset();
+    _number_of_blocks = 0;
+  }
+
+  // Get the block at position pos in _blocks
+  Block* get_block(uint pos) const {
+    return _blocks[pos];
+  }
+
+  // Number of blocks
+  uint number_of_blocks() const {
+    return _number_of_blocks;
+  }
 
   // set which block this node should reside in
   void map_node_to_block(const Node* node, Block* block) {
@@ -412,72 +530,26 @@
     return (_node_to_block_mapping.lookup(node->_idx) != NULL);
   }
 
-  // Per node latency estimation, valid only during GCM
-  GrowableArray<uint> *_node_latency;
-
-#ifndef PRODUCT
-  bool _trace_opto_pipelining;  // tracing flag
-#endif
-
 #ifdef ASSERT
   Unique_Node_List _raw_oops;
 #endif
 
-  // Build dominators
-  void Dominators();
-
-  // Estimate block frequencies based on IfNode probabilities
-  void Estimate_Block_Frequency();
-
-  // Global Code Motion.  See Click's PLDI95 paper.  Place Nodes in specific
-  // basic blocks; i.e. _node_to_block_mapping now maps _idx for all Nodes to some Block.
-  void GlobalCodeMotion( Matcher &m, uint unique, Node_List &proj_list );
+  // Do global code motion by first building dominator tree and estimate block frequency
+  // Returns true on success
+  bool do_global_code_motion();
 
   // Compute the (backwards) latency of a node from the uses
   void latency_from_uses(Node *n);
 
-  // Compute the (backwards) latency of a node from a single use
-  int latency_from_use(Node *n, const Node *def, Node *use);
-
-  // Compute the (backwards) latency of a node from the uses of this instruction
-  void partial_latency_of_defs(Node *n);
-
-  // Schedule Nodes early in their basic blocks.
-  bool schedule_early(VectorSet &visited, Node_List &roots);
-
-  // For each node, find the latest block it can be scheduled into
-  // and then select the cheapest block between the latest and earliest
-  // block to place the node.
-  void schedule_late(VectorSet &visited, Node_List &stack);
-
-  // Pick a block between early and late that is a cheaper alternative
-  // to late. Helper for schedule_late.
-  Block* hoist_to_cheaper_block(Block* LCA, Block* early, Node* self);
-
-  // Compute the instruction global latency with a backwards walk
-  void ComputeLatenciesBackwards(VectorSet &visited, Node_List &stack);
-
   // Set loop alignment
   void set_loop_alignment();
 
   // Remove empty basic blocks
-  void remove_empty();
+  void remove_empty_blocks();
   void fixup_flow();
-  bool move_to_next(Block* bx, uint b_index);
-  void move_to_end(Block* bx, uint b_index);
-  void insert_goto_at(uint block_no, uint succ_no);
 
-  // Check for NeverBranch at block end.  This needs to become a GOTO to the
-  // true target.  NeverBranch are treated as a conditional branch that always
-  // goes the same direction for most of the optimizer and are used to give a
-  // fake exit path to infinite loops.  At this late stage they need to turn
-  // into Goto's so that when you enter the infinite loop you indeed hang.
-  void convert_NeverBranch_to_Goto(Block *b);
-
-  CFGLoop* create_loop_tree();
-
-  // Insert a node into a block, and update the _bbs
-  void insert( Block *b, uint idx, Node *n ) {
+  // Insert a node into a block at index and map the node to the block
+  void insert(Block *b, uint idx, Node *n) {
     b->_nodes.insert( idx, n );
     map_node_to_block(n, b);
   }
--- a/src/share/vm/opto/buildOopMap.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/opto/buildOopMap.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -87,7 +87,6 @@
 // OptoReg::Bad for not-callee-saved.
 
 
-//------------------------------OopFlow----------------------------------------
 // Structure to pass around
 struct OopFlow : public ResourceObj {
   short *_callees;              // Array mapping register to callee-saved
@@ -119,7 +118,6 @@
   OopMap *build_oop_map( Node *n, int max_reg, PhaseRegAlloc *regalloc, int* live );
 };
 
-//------------------------------compute_reach----------------------------------
 // Given reaching-defs for this block start, compute it for this block end
 void OopFlow::compute_reach( PhaseRegAlloc *regalloc, int max_reg, Dict *safehash ) {
 
@@ -177,7 +175,6 @@
   }
 }
 
-//------------------------------merge------------------------------------------
 // Merge the given flow into the 'this' flow
 void OopFlow::merge( OopFlow *flow, int max_reg ) {
   assert( _b == NULL, "merging into a happy flow" );
@@ -197,14 +194,12 @@
 
 }
 
-//------------------------------clone------------------------------------------
 void OopFlow::clone( OopFlow *flow, int max_size ) {
   _b = flow->_b;
   memcpy( _callees, flow->_callees, sizeof(short)*max_size);
   memcpy( _defs   , flow->_defs   , sizeof(Node*)*max_size);
 }
 
-//------------------------------make-------------------------------------------
 OopFlow *OopFlow::make( Arena *A, int max_size, Compile* C ) {
   short *callees = NEW_ARENA_ARRAY(A,short,max_size+1);
   Node **defs    = NEW_ARENA_ARRAY(A,Node*,max_size+1);
@@ -215,7 +210,6 @@
   return flow;
 }
 
-//------------------------------bit twiddlers----------------------------------
 static int get_live_bit( int *live, int reg ) {
   return live[reg>>LogBitsPerInt] &   (1<<(reg&(BitsPerInt-1))); }
 static void set_live_bit( int *live, int reg ) {
@@ -223,7 +217,6 @@
 static void clr_live_bit( int *live, int reg ) {
          live[reg>>LogBitsPerInt] &= ~(1<<(reg&(BitsPerInt-1))); }
 
-//------------------------------build_oop_map----------------------------------
 // Build an oopmap from the current flow info
 OopMap *OopFlow::build_oop_map( Node *n, int max_reg, PhaseRegAlloc *regalloc, int* live ) {
   int framesize = regalloc->_framesize;
@@ -412,19 +405,18 @@
   return omap;
 }
 
-//------------------------------do_liveness------------------------------------
 // Compute backwards liveness on registers
-static void do_liveness( PhaseRegAlloc *regalloc, PhaseCFG *cfg, Block_List *worklist, int max_reg_ints, Arena *A, Dict *safehash ) {
-  int *live = NEW_ARENA_ARRAY(A, int, (cfg->_num_blocks+1) * max_reg_ints);
-  int *tmp_live = &live[cfg->_num_blocks * max_reg_ints];
-  Node *root = cfg->C->root();
+static void do_liveness(PhaseRegAlloc* regalloc, PhaseCFG* cfg, Block_List* worklist, int max_reg_ints, Arena* A, Dict* safehash) {
+  int* live = NEW_ARENA_ARRAY(A, int, (cfg->number_of_blocks() + 1) * max_reg_ints);
+  int* tmp_live = &live[cfg->number_of_blocks() * max_reg_ints];
+  Node* root = cfg->get_root_node();
   // On CISC platforms, get the node representing the stack pointer  that regalloc
   // used for spills
   Node *fp = NodeSentinel;
   if (UseCISCSpill && root->req() > 1) {
     fp = root->in(1)->in(TypeFunc::FramePtr);
   }
-  memset( live, 0, cfg->_num_blocks * (max_reg_ints<<LogBytesPerInt) );
+  memset(live, 0, cfg->number_of_blocks() * (max_reg_ints << LogBytesPerInt));
   // Push preds onto worklist
   for (uint i = 1; i < root->req(); i++) {
     Block* block = cfg->get_block_for_node(root->in(i));
@@ -549,29 +541,32 @@
     // Scan for any missing safepoints.  Happens to infinite loops
     // ala ZKM.jar
     uint i;
-    for( i=1; i<cfg->_num_blocks; i++ ) {
-      Block *b = cfg->_blocks[i];
+    for (i = 1; i < cfg->number_of_blocks(); i++) {
+      Block* block = cfg->get_block(i);
       uint j;
-      for( j=1; j<b->_nodes.size(); j++ )
-        if( b->_nodes[j]->jvms() &&
-            (*safehash)[b->_nodes[j]] == NULL )
+      for (j = 1; j < block->_nodes.size(); j++) {
+        if (block->_nodes[j]->jvms() && (*safehash)[block->_nodes[j]] == NULL) {
            break;
-      if( j<b->_nodes.size() ) break;
+        }
+      }
+      if (j < block->_nodes.size()) {
+        break;
+      }
     }
-    if( i == cfg->_num_blocks )
+    if (i == cfg->number_of_blocks()) {
       break;                    // Got 'em all
+    }
 #ifndef PRODUCT
     if( PrintOpto && Verbose )
       tty->print_cr("retripping live calc");
 #endif
     // Force the issue (expensively): recheck everybody
-    for( i=1; i<cfg->_num_blocks; i++ )
-      worklist->push(cfg->_blocks[i]);
+    for (i = 1; i < cfg->number_of_blocks(); i++) {
+      worklist->push(cfg->get_block(i));
+    }
   }
-
 }
 
-//------------------------------BuildOopMaps-----------------------------------
 // Collect GC mask info - where are all the OOPs?
 void Compile::BuildOopMaps() {
   NOT_PRODUCT( TracePhase t3("bldOopMaps", &_t_buildOopMaps, TimeCompiler); )
@@ -592,12 +587,12 @@
   OopFlow *free_list = NULL;    // Free, unused
 
   // Array mapping blocks to completed oopflows
-  OopFlow **flows = NEW_ARENA_ARRAY(A, OopFlow*, _cfg->_num_blocks);
-  memset( flows, 0, _cfg->_num_blocks*sizeof(OopFlow*) );
+  OopFlow **flows = NEW_ARENA_ARRAY(A, OopFlow*, _cfg->number_of_blocks());
+  memset( flows, 0, _cfg->number_of_blocks() * sizeof(OopFlow*) );
 
 
   // Do the first block 'by hand' to prime the worklist
-  Block *entry = _cfg->_blocks[1];
+  Block *entry = _cfg->get_block(1);
   OopFlow *rootflow = OopFlow::make(A,max_reg,this);
   // Initialize to 'bottom' (not 'top')
   memset( rootflow->_callees, OptoReg::Bad, max_reg*sizeof(short) );
@@ -623,7 +618,9 @@
 
     Block *b = worklist.pop();
     // Ignore root block
-    if( b == _cfg->_broot ) continue;
+    if (b == _cfg->get_root_block()) {
+      continue;
+    }
     // Block is already done?  Happens if block has several predecessors,
     // he can get on the worklist more than once.
     if( flows[b->_pre_order] ) continue;
--- a/src/share/vm/opto/chaitin.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/opto/chaitin.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -40,10 +40,8 @@
 #include "opto/opcodes.hpp"
 #include "opto/rootnode.hpp"
 
-//=============================================================================
-
 #ifndef PRODUCT
-void LRG::dump( ) const {
+void LRG::dump() const {
   ttyLocker ttyl;
   tty->print("%d ",num_regs());
   _mask.dump();
@@ -94,7 +92,6 @@
 }
 #endif
 
-//------------------------------score------------------------------------------
 // Compute score from cost and area.  Low score is best to spill.
 static double raw_score( double cost, double area ) {
   return cost - (area*RegisterCostAreaRatio) * 1.52588e-5;
@@ -125,7 +122,6 @@
   return score;
 }
 
-//------------------------------LRG_List---------------------------------------
 LRG_List::LRG_List( uint max ) : _cnt(max), _max(max), _lidxs(NEW_RESOURCE_ARRAY(uint,max)) {
   memset( _lidxs, 0, sizeof(uint)*max );
 }
@@ -211,7 +207,6 @@
   return next;
 }
 
-//------------------------------Chaitin----------------------------------------
 PhaseChaitin::PhaseChaitin(uint unique, PhaseCFG &cfg, Matcher &matcher)
   : PhaseRegAlloc(unique, cfg, matcher,
 #ifndef PRODUCT
@@ -232,31 +227,31 @@
 {
   NOT_PRODUCT( Compile::TracePhase t3("ctorChaitin", &_t_ctorChaitin, TimeCompiler); )
 
-  _high_frequency_lrg = MIN2(float(OPTO_LRG_HIGH_FREQ), _cfg._outer_loop_freq);
+  _high_frequency_lrg = MIN2(float(OPTO_LRG_HIGH_FREQ), _cfg.get_outer_loop_frequency());
 
   // Build a list of basic blocks, sorted by frequency
-  _blks = NEW_RESOURCE_ARRAY( Block *, _cfg._num_blocks );
+  _blks = NEW_RESOURCE_ARRAY(Block *, _cfg.number_of_blocks());
   // Experiment with sorting strategies to speed compilation
   double  cutoff = BLOCK_FREQUENCY(1.0); // Cutoff for high frequency bucket
   Block **buckets[NUMBUCKS];             // Array of buckets
   uint    buckcnt[NUMBUCKS];             // Array of bucket counters
   double  buckval[NUMBUCKS];             // Array of bucket value cutoffs
   for (uint i = 0; i < NUMBUCKS; i++) {
-    buckets[i] = NEW_RESOURCE_ARRAY(Block *, _cfg._num_blocks);
+    buckets[i] = NEW_RESOURCE_ARRAY(Block *, _cfg.number_of_blocks());
     buckcnt[i] = 0;
     // Bump by three orders of magnitude each time
     cutoff *= 0.001;
     buckval[i] = cutoff;
-    for (uint j = 0; j < _cfg._num_blocks; j++) {
+    for (uint j = 0; j < _cfg.number_of_blocks(); j++) {
       buckets[i][j] = NULL;
     }
   }
   // Sort blocks into buckets
-  for (uint i = 0; i < _cfg._num_blocks; i++) {
+  for (uint i = 0; i < _cfg.number_of_blocks(); i++) {
     for (uint j = 0; j < NUMBUCKS; j++) {
-      if ((j == NUMBUCKS - 1) || (_cfg._blocks[i]->_freq > buckval[j])) {
+      if ((j == NUMBUCKS - 1) || (_cfg.get_block(i)->_freq > buckval[j])) {
         // Assign block to end of list for appropriate bucket
-        buckets[j][buckcnt[j]++] = _cfg._blocks[i];
+        buckets[j][buckcnt[j]++] = _cfg.get_block(i);
         break; // kick out of inner loop
       }
     }
@@ -269,10 +264,9 @@
     }
   }
 
-  assert(blkcnt == _cfg._num_blocks, "Block array not totally filled");
+  assert(blkcnt == _cfg.number_of_blocks(), "Block array not totally filled");
 }
 
-//------------------------------Union------------------------------------------
 // union 2 sets together.
 void PhaseChaitin::Union( const Node *src_n, const Node *dst_n ) {
   uint src = _lrg_map.find(src_n);
@@ -285,7 +279,6 @@
   _lrg_map.uf_map(dst, src);
 }
 
-//------------------------------new_lrg----------------------------------------
 void PhaseChaitin::new_lrg(const Node *x, uint lrg) {
   // Make the Node->LRG mapping
   _lrg_map.extend(x->_idx,lrg);
@@ -294,24 +287,28 @@
 }
 
 
-bool PhaseChaitin::clone_projs_shared(Block *b, uint idx, Node *con, Node *copy, uint max_lrg_id) {
-  Block* bcon = _cfg.get_block_for_node(con);
-  uint cindex = bcon->find_node(con);
-  Node *con_next = bcon->_nodes[cindex+1];
-  if (con_next->in(0) != con || !con_next->is_MachProj()) {
-    return false;               // No MachProj's follow
+int PhaseChaitin::clone_projs(Block* b, uint idx, Node* orig, Node* copy, uint& max_lrg_id) {
+  assert(b->find_node(copy) == (idx - 1), "incorrect insert index for copy kill projections");
+  DEBUG_ONLY( Block* borig = _cfg.get_block_for_node(orig); )
+  int found_projs = 0;
+  uint cnt = orig->outcnt();
+  for (uint i = 0; i < cnt; i++) {
+    Node* proj = orig->raw_out(i);
+    if (proj->is_MachProj()) {
+      assert(proj->outcnt() == 0, "only kill projections are expected here");
+      assert(_cfg.get_block_for_node(proj) == borig, "incorrect block for kill projections");
+      found_projs++;
+      // Copy kill projections after the cloned node
+      Node* kills = proj->clone();
+      kills->set_req(0, copy);
+      b->_nodes.insert(idx++, kills);
+      _cfg.map_node_to_block(kills, b);
+      new_lrg(kills, max_lrg_id++);
+    }
   }
-
-  // Copy kills after the cloned constant
-  Node *kills = con_next->clone();
-  kills->set_req(0, copy);
-  b->_nodes.insert(idx, kills);
-  _cfg.map_node_to_block(kills, b);
-  new_lrg(kills, max_lrg_id);
-  return true;
+  return found_projs;
 }
 
-//------------------------------compact----------------------------------------
 // Renumber the live ranges to compact them.  Makes the IFG smaller.
 void PhaseChaitin::compact() {
   // Current the _uf_map contains a series of short chains which are headed
@@ -677,20 +674,19 @@
   C->set_indexSet_arena(NULL);  // ResourceArea is at end of scope
 }
 
-//------------------------------de_ssa-----------------------------------------
 void PhaseChaitin::de_ssa() {
   // Set initial Names for all Nodes.  Most Nodes get the virtual register
   // number.  A few get the ZERO live range number.  These do not
   // get allocated, but instead rely on correct scheduling to ensure that
   // only one instance is simultaneously live at a time.
   uint lr_counter = 1;
-  for( uint i = 0; i < _cfg._num_blocks; i++ ) {
-    Block *b = _cfg._blocks[i];
-    uint cnt = b->_nodes.size();
+  for( uint i = 0; i < _cfg.number_of_blocks(); i++ ) {
+    Block* block = _cfg.get_block(i);
+    uint cnt = block->_nodes.size();
 
     // Handle all the normal Nodes in the block
     for( uint j = 0; j < cnt; j++ ) {
-      Node *n = b->_nodes[j];
+      Node *n = block->_nodes[j];
       // Pre-color to the zero live range, or pick virtual register
       const RegMask &rm = n->out_RegMask();
       _lrg_map.map(n->_idx, rm.is_NotEmpty() ? lr_counter++ : 0);
@@ -701,52 +697,55 @@
 }
 
 
-//------------------------------gather_lrg_masks-------------------------------
 // Gather LiveRanGe information, including register masks.  Modification of
 // cisc spillable in_RegMasks should not be done before AggressiveCoalesce.
 void PhaseChaitin::gather_lrg_masks( bool after_aggressive ) {
 
   // Nail down the frame pointer live range
-  uint fp_lrg = _lrg_map.live_range_id(_cfg._root->in(1)->in(TypeFunc::FramePtr));
+  uint fp_lrg = _lrg_map.live_range_id(_cfg.get_root_node()->in(1)->in(TypeFunc::FramePtr));
   lrgs(fp_lrg)._cost += 1e12;   // Cost is infinite
 
   // For all blocks
-  for( uint i = 0; i < _cfg._num_blocks; i++ ) {
-    Block *b = _cfg._blocks[i];
+  for (uint i = 0; i < _cfg.number_of_blocks(); i++) {
+    Block* block = _cfg.get_block(i);
 
     // For all instructions
-    for( uint j = 1; j < b->_nodes.size(); j++ ) {
-      Node *n = b->_nodes[j];
+    for (uint j = 1; j < block->_nodes.size(); j++) {
+      Node* n = block->_nodes[j];
       uint input_edge_start =1; // Skip control most nodes
-      if( n->is_Mach() ) input_edge_start = n->as_Mach()->oper_input_base();
+      if (n->is_Mach()) {
+        input_edge_start = n->as_Mach()->oper_input_base();
+      }
       uint idx = n->is_Copy();
 
       // Get virtual register number, same as LiveRanGe index
       uint vreg = _lrg_map.live_range_id(n);
-      LRG &lrg = lrgs(vreg);
-      if( vreg ) {              // No vreg means un-allocable (e.g. memory)
+      LRG& lrg = lrgs(vreg);
+      if (vreg) {              // No vreg means un-allocable (e.g. memory)
 
         // Collect has-copy bit
-        if( idx ) {
+        if (idx) {
           lrg._has_copy = 1;
           uint clidx = _lrg_map.live_range_id(n->in(idx));
-          LRG &copy_src = lrgs(clidx);
+          LRG& copy_src = lrgs(clidx);
           copy_src._has_copy = 1;
         }
 
         // Check for float-vs-int live range (used in register-pressure
         // calculations)
         const Type *n_type = n->bottom_type();
-        if (n_type->is_floatingpoint())
+        if (n_type->is_floatingpoint()) {
           lrg._is_float = 1;
+        }
 
         // Check for twice prior spilling.  Once prior spilling might have
         // spilled 'soft', 2nd prior spill should have spilled 'hard' and
         // further spilling is unlikely to make progress.
-        if( _spilled_once.test(n->_idx) ) {
+        if (_spilled_once.test(n->_idx)) {
           lrg._was_spilled1 = 1;
-          if( _spilled_twice.test(n->_idx) )
+          if (_spilled_twice.test(n->_idx)) {
             lrg._was_spilled2 = 1;
+          }
         }
 
 #ifndef PRODUCT
@@ -783,16 +782,18 @@
 
         // Check for bound register masks
         const RegMask &lrgmask = lrg.mask();
-        if (lrgmask.is_bound(ireg))
+        if (lrgmask.is_bound(ireg)) {
           lrg._is_bound = 1;
+        }
 
         // Check for maximum frequency value
-        if (lrg._maxfreq < b->_freq)
-          lrg._maxfreq = b->_freq;
+        if (lrg._maxfreq < block->_freq) {
+          lrg._maxfreq = block->_freq;
+        }
 
         // Check for oop-iness, or long/double
         // Check for multi-kill projection
-        switch( ireg ) {
+        switch (ireg) {
         case MachProjNode::fat_proj:
           // Fat projections have size equal to number of registers killed
           lrg.set_num_regs(rm.Size());
@@ -962,7 +963,7 @@
         // AggressiveCoalesce.  This effectively pre-virtual-splits
         // around uncommon uses of common defs.
         const RegMask &rm = n->in_RegMask(k);
-        if (!after_aggressive && _cfg.get_block_for_node(n->in(k))->_freq > 1000 * b->_freq) {
+        if (!after_aggressive && _cfg.get_block_for_node(n->in(k))->_freq > 1000 * block->_freq) {
           // Since we are BEFORE aggressive coalesce, leave the register
           // mask untrimmed by the call.  This encourages more coalescing.
           // Later, AFTER aggressive, this live range will have to spill
@@ -1006,8 +1007,9 @@
         }
 
         // Check for maximum frequency value
-        if( lrg._maxfreq < b->_freq )
-          lrg._maxfreq = b->_freq;
+        if (lrg._maxfreq < block->_freq) {
+          lrg._maxfreq = block->_freq;
+        }
 
       } // End for all allocated inputs
     } // end for all instructions
@@ -1029,7 +1031,6 @@
   }
 }
 
-//------------------------------set_was_low------------------------------------
 // Set the was-lo-degree bit.  Conservative coalescing should not change the
 // colorability of the graph.  If any live range was of low-degree before
 // coalescing, it should Simplify.  This call sets the was-lo-degree bit.
@@ -1066,7 +1067,6 @@
 
 #define REGISTER_CONSTRAINED 16
 
-//------------------------------cache_lrg_info---------------------------------
 // Compute cost/area ratio, in case we spill.  Build the lo-degree list.
 void PhaseChaitin::cache_lrg_info( ) {
 
@@ -1100,7 +1100,6 @@
   }
 }
 
-//------------------------------Pre-Simplify-----------------------------------
 // Simplify the IFG by removing LRGs of low degree that have NO copies
 void PhaseChaitin::Pre_Simplify( ) {
 
@@ -1151,7 +1150,6 @@
   // No more lo-degree no-copy live ranges to simplify
 }
 
-//------------------------------Simplify---------------------------------------
 // Simplify the IFG by removing LRGs of low degree.
 void PhaseChaitin::Simplify( ) {
 
@@ -1288,7 +1286,6 @@
 
 }
 
-//------------------------------is_legal_reg-----------------------------------
 // Is 'reg' register legal for 'lrg'?
 static bool is_legal_reg(LRG &lrg, OptoReg::Name reg, int chunk) {
   if (reg >= chunk && reg < (chunk + RegMask::CHUNK_SIZE) &&
@@ -1315,7 +1312,6 @@
   return false;
 }
 
-//------------------------------bias_color-------------------------------------
 // Choose a color using the biasing heuristic
 OptoReg::Name PhaseChaitin::bias_color( LRG &lrg, int chunk ) {
 
@@ -1377,7 +1373,6 @@
   return OptoReg::add( reg, chunk );
 }
 
-//------------------------------choose_color-----------------------------------
 // Choose a color in the current chunk
 OptoReg::Name PhaseChaitin::choose_color( LRG &lrg, int chunk ) {
   assert( C->in_preserve_stack_slots() == 0 || chunk != 0 || lrg._is_bound || lrg.mask().is_bound1() || !lrg.mask().Member(OptoReg::Name(_matcher._old_SP-1)), "must not allocate stack0 (inside preserve area)");
@@ -1399,7 +1394,6 @@
   return lrg.mask().find_last_elem();
 }
 
-//------------------------------Select-----------------------------------------
 // Select colors by re-inserting LRGs back into the IFG.  LRGs are re-inserted
 // in reverse order of removal.  As long as nothing of hi-degree was yanked,
 // everything going back is guaranteed a color.  Select that color.  If some
@@ -1574,8 +1568,6 @@
   return spill_reg-LRG::SPILL_REG;      // Return number of spills
 }
 
-
-//------------------------------copy_was_spilled-------------------------------
 // Copy 'was_spilled'-edness from the source Node to the dst Node.
 void PhaseChaitin::copy_was_spilled( Node *src, Node *dst ) {
   if( _spilled_once.test(src->_idx) ) {
@@ -1588,14 +1580,12 @@
   }
 }
 
-//------------------------------set_was_spilled--------------------------------
 // Set the 'spilled_once' or 'spilled_twice' flag on a node.
 void PhaseChaitin::set_was_spilled( Node *n ) {
   if( _spilled_once.test_set(n->_idx) )
     _spilled_twice.set(n->_idx);
 }
 
-//------------------------------fixup_spills-----------------------------------
 // Convert Ideal spill instructions into proper FramePtr + offset Loads and
 // Stores.  Use-def chains are NOT preserved, but Node->LRG->reg maps are.
 void PhaseChaitin::fixup_spills() {
@@ -1605,16 +1595,16 @@
   NOT_PRODUCT( Compile::TracePhase t3("fixupSpills", &_t_fixupSpills, TimeCompiler); )
 
   // Grab the Frame Pointer
-  Node *fp = _cfg._broot->head()->in(1)->in(TypeFunc::FramePtr);
+  Node *fp = _cfg.get_root_block()->head()->in(1)->in(TypeFunc::FramePtr);
 
   // For all blocks
-  for( uint i = 0; i < _cfg._num_blocks; i++ ) {
-    Block *b = _cfg._blocks[i];
+  for (uint i = 0; i < _cfg.number_of_blocks(); i++) {
+    Block* block = _cfg.get_block(i);
 
     // For all instructions in block
-    uint last_inst = b->end_idx();
-    for( uint j = 1; j <= last_inst; j++ ) {
-      Node *n = b->_nodes[j];
+    uint last_inst = block->end_idx();
+    for (uint j = 1; j <= last_inst; j++) {
+      Node* n = block->_nodes[j];
 
       // Dead instruction???
       assert( n->outcnt() != 0 ||// Nothing dead after post alloc
@@ -1651,7 +1641,7 @@
             assert( cisc->oper_input_base() == 2, "Only adding one edge");
             cisc->ins_req(1,src);         // Requires a memory edge
           }
-          b->_nodes.map(j,cisc);          // Insert into basic block
+          block->_nodes.map(j,cisc);          // Insert into basic block
           n->subsume_by(cisc, C); // Correct graph
           //
           ++_used_cisc_instructions;
@@ -1677,7 +1667,6 @@
   } // End of for all blocks
 }
 
-//------------------------------find_base_for_derived--------------------------
 // Helper to stretch above; recursively discover the base Node for a
 // given derived Node.  Easy for AddP-related machine nodes, but needs
 // to be recursive for derived Phis.
@@ -1707,7 +1696,7 @@
       // Initialize it once and make it shared:
       // set control to _root and place it into Start block
       // (where top() node is placed).
-      base->init_req(0, _cfg._root);
+      base->init_req(0, _cfg.get_root_node());
       Block *startb = _cfg.get_block_for_node(C->top());
       startb->_nodes.insert(startb->find_node(C->top()), base );
       _cfg.map_node_to_block(base, startb);
@@ -1716,7 +1705,7 @@
     if (_lrg_map.live_range_id(base) == 0) {
       new_lrg(base, maxlrg++);
     }
-    assert(base->in(0) == _cfg._root && _cfg.get_block_for_node(base) == _cfg.get_block_for_node(C->top()), "base NULL should be shared");
+    assert(base->in(0) == _cfg.get_root_node() && _cfg.get_block_for_node(base) == _cfg.get_block_for_node(C->top()), "base NULL should be shared");
     derived_base_map[derived->_idx] = base;
     return base;
   }
@@ -1779,8 +1768,6 @@
   return base;
 }
 
-
-//------------------------------stretch_base_pointer_live_ranges---------------
 // At each Safepoint, insert extra debug edges for each pair of derived value/
 // base pointer that is live across the Safepoint for oopmap building.  The
 // edge pairs get added in after sfpt->jvmtail()->oopoff(), but are in the
@@ -1792,14 +1779,14 @@
   memset( derived_base_map, 0, sizeof(Node*)*C->unique() );
 
   // For all blocks in RPO do...
-  for( uint i=0; i<_cfg._num_blocks; i++ ) {
-    Block *b = _cfg._blocks[i];
+  for (uint i = 0; i < _cfg.number_of_blocks(); i++) {
+    Block* block = _cfg.get_block(i);
     // Note use of deep-copy constructor.  I cannot hammer the original
     // liveout bits, because they are needed by the following coalesce pass.
-    IndexSet liveout(_live->live(b));
+    IndexSet liveout(_live->live(block));
 
-    for( uint j = b->end_idx() + 1; j > 1; j-- ) {
-      Node *n = b->_nodes[j-1];
+    for (uint j = block->end_idx() + 1; j > 1; j--) {
+      Node* n = block->_nodes[j - 1];
 
       // Pre-split compares of loop-phis.  Loop-phis form a cycle we would
       // like to see in the same register.  Compare uses the loop-phi and so
@@ -1814,7 +1801,7 @@
         Node *phi = n->in(1);
         if( phi->is_Phi() && phi->as_Phi()->region()->is_Loop() ) {
           Block *phi_block = _cfg.get_block_for_node(phi);
-          if (_cfg.get_block_for_node(phi_block->pred(2)) == b) {
+          if (_cfg.get_block_for_node(phi_block->pred(2)) == block) {
             const RegMask *mask = C->matcher()->idealreg2spillmask[Op_RegI];
             Node *spill = new (C) MachSpillCopyNode( phi, *mask, *mask );
             insert_proj( phi_block, 1, spill, maxlrg++ );
@@ -1868,7 +1855,7 @@
             if ((_lrg_map.live_range_id(base) >= _lrg_map.max_lrg_id() || // (Brand new base (hence not live) or
                  !liveout.member(_lrg_map.live_range_id(base))) && // not live) AND
                  (_lrg_map.live_range_id(base) > 0) && // not a constant
-                 _cfg.get_block_for_node(base) != b) { // base not def'd in blk)
+                 _cfg.get_block_for_node(base) != block) { // base not def'd in blk)
               // Base pointer is not currently live.  Since I stretched
               // the base pointer to here and it crosses basic-block
               // boundaries, the global live info is now incorrect.
@@ -1903,15 +1890,12 @@
   return must_recompute_live != 0;
 }
 
-
-//------------------------------add_reference----------------------------------
 // Extend the node to LRG mapping
 
 void PhaseChaitin::add_reference(const Node *node, const Node *old_node) {
   _lrg_map.extend(node->_idx, _lrg_map.live_range_id(old_node));
 }
 
-//------------------------------dump-------------------------------------------
 #ifndef PRODUCT
 void PhaseChaitin::dump(const Node *n) const {
   uint r = (n->_idx < _lrg_map.size()) ? _lrg_map.find_const(n) : 0;
@@ -2017,8 +2001,9 @@
               _matcher._new_SP, _framesize );
 
   // For all blocks
-  for( uint i = 0; i < _cfg._num_blocks; i++ )
-    dump(_cfg._blocks[i]);
+  for (uint i = 0; i < _cfg.number_of_blocks(); i++) {
+    dump(_cfg.get_block(i));
+  }
   // End of per-block dump
   tty->print("\n");
 
@@ -2059,7 +2044,6 @@
   tty->print_cr("");
 }
 
-//------------------------------dump_degree_lists------------------------------
 void PhaseChaitin::dump_degree_lists() const {
   // Dump lo-degree list
   tty->print("Lo degree: ");
@@ -2080,7 +2064,6 @@
   tty->print_cr("");
 }
 
-//------------------------------dump_simplified--------------------------------
 void PhaseChaitin::dump_simplified() const {
   tty->print("Simplified: ");
   for( uint i = _simplified; i; i = lrgs(i)._next )
@@ -2099,7 +2082,6 @@
   return buf+strlen(buf);
 }
 
-//------------------------------dump_register----------------------------------
 // Dump a register name into a buffer.  Be intelligent if we get called
 // before allocation is complete.
 char *PhaseChaitin::dump_register( const Node *n, char *buf  ) const {
@@ -2133,7 +2115,6 @@
   return buf+strlen(buf);
 }
 
-//----------------------dump_for_spill_split_recycle--------------------------
 void PhaseChaitin::dump_for_spill_split_recycle() const {
   if( WizardMode && (PrintCompilation || PrintOpto) ) {
     // Display which live ranges need to be split and the allocator's state
@@ -2149,7 +2130,6 @@
   }
 }
 
-//------------------------------dump_frame------------------------------------
 void PhaseChaitin::dump_frame() const {
   const char *fp = OptoReg::regname(OptoReg::c_frame_pointer);
   const TypeTuple *domain = C->tf()->domain();
@@ -2255,17 +2235,16 @@
   tty->print_cr("#");
 }
 
-//------------------------------dump_bb----------------------------------------
 void PhaseChaitin::dump_bb( uint pre_order ) const {
   tty->print_cr("---dump of B%d---",pre_order);
-  for( uint i = 0; i < _cfg._num_blocks; i++ ) {
-    Block *b = _cfg._blocks[i];
-    if( b->_pre_order == pre_order )
-      dump(b);
+  for (uint i = 0; i < _cfg.number_of_blocks(); i++) {
+    Block* block = _cfg.get_block(i);
+    if (block->_pre_order == pre_order) {
+      dump(block);
+    }
   }
 }
 
-//------------------------------dump_lrg---------------------------------------
 void PhaseChaitin::dump_lrg( uint lidx, bool defs_only ) const {
   tty->print_cr("---dump of L%d---",lidx);
 
@@ -2287,17 +2266,17 @@
     tty->cr();
   }
   // For all blocks
-  for( uint i = 0; i < _cfg._num_blocks; i++ ) {
-    Block *b = _cfg._blocks[i];
+  for (uint i = 0; i < _cfg.number_of_blocks(); i++) {
+    Block* block = _cfg.get_block(i);
     int dump_once = 0;
 
     // For all instructions
-    for( uint j = 0; j < b->_nodes.size(); j++ ) {
-      Node *n = b->_nodes[j];
+    for( uint j = 0; j < block->_nodes.size(); j++ ) {
+      Node *n = block->_nodes[j];
       if (_lrg_map.find_const(n) == lidx) {
         if (!dump_once++) {
           tty->cr();
-          b->dump_head(&_cfg);
+          block->dump_head(&_cfg);
         }
         dump(n);
         continue;
@@ -2312,7 +2291,7 @@
           if (_lrg_map.find_const(m) == lidx) {
             if (!dump_once++) {
               tty->cr();
-              b->dump_head(&_cfg);
+              block->dump_head(&_cfg);
             }
             dump(n);
           }
@@ -2324,7 +2303,6 @@
 }
 #endif // not PRODUCT
 
-//------------------------------print_chaitin_statistics-------------------------------
 int PhaseChaitin::_final_loads  = 0;
 int PhaseChaitin::_final_stores = 0;
 int PhaseChaitin::_final_memoves= 0;
--- a/src/share/vm/opto/chaitin.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/opto/chaitin.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -412,33 +412,22 @@
   uint split_DEF( Node *def, Block *b, int loc, uint max, Node **Reachblock, Node **debug_defs, GrowableArray<uint> splits, int slidx );
   uint split_USE( Node *def, Block *b, Node *use, uint useidx, uint max, bool def_down, bool cisc_sp, GrowableArray<uint> splits, int slidx );
 
-  bool clone_projs(Block *b, uint idx, Node *con, Node *copy, LiveRangeMap &lrg_map) {
-    bool found_projs = clone_projs_shared(b, idx, con, copy, lrg_map.max_lrg_id());
-
-    if(found_projs) {
-      uint max_lrg_id = lrg_map.max_lrg_id();
-      lrg_map.set_max_lrg_id(max_lrg_id + 1);
-    }
-
-    return found_projs;
-  }
-
   //------------------------------clone_projs------------------------------------
   // After cloning some rematerialized instruction, clone any MachProj's that
   // follow it.  Example: Intel zero is XOR, kills flags.  Sparc FP constants
   // use G3 as an address temp.
-  bool clone_projs(Block *b, uint idx, Node *con, Node *copy, uint &max_lrg_id) {
-    bool found_projs = clone_projs_shared(b, idx, con, copy, max_lrg_id);
+  int clone_projs(Block* b, uint idx, Node* orig, Node* copy, uint& max_lrg_id);
 
-    if(found_projs) {
-      max_lrg_id++;
+  int clone_projs(Block* b, uint idx, Node* orig, Node* copy, LiveRangeMap& lrg_map) {
+    uint max_lrg_id = lrg_map.max_lrg_id();
+    int found_projs = clone_projs(b, idx, orig, copy, max_lrg_id);
+    if (found_projs > 0) {
+      // max_lrg_id is updated during call above
+      lrg_map.set_max_lrg_id(max_lrg_id);
     }
-
     return found_projs;
   }
 
-  bool clone_projs_shared(Block *b, uint idx, Node *con, Node *copy, uint max_lrg_id);
-
   Node *split_Rematerialize(Node *def, Block *b, uint insidx, uint &maxlrg, GrowableArray<uint> splits,
                             int slidx, uint *lrg2reach, Node **Reachblock, bool walkThru);
   // True if lidx is used before any real register is def'd in the block
--- a/src/share/vm/opto/coalesce.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/opto/coalesce.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -34,8 +34,6 @@
 #include "opto/matcher.hpp"
 #include "opto/regmask.hpp"
 
-//=============================================================================
-//------------------------------Dump-------------------------------------------
 #ifndef PRODUCT
 void PhaseCoalesce::dump(Node *n) const {
   // Being a const function means I cannot use 'Find'
@@ -43,12 +41,11 @@
   tty->print("L%d/N%d ",r,n->_idx);
 }
 
-//------------------------------dump-------------------------------------------
 void PhaseCoalesce::dump() const {
   // I know I have a block layout now, so I can print blocks in a loop
-  for( uint i=0; i<_phc._cfg._num_blocks; i++ ) {
+  for( uint i=0; i<_phc._cfg.number_of_blocks(); i++ ) {
     uint j;
-    Block *b = _phc._cfg._blocks[i];
+    Block* b = _phc._cfg.get_block(i);
     // Print a nice block header
     tty->print("B%d: ",b->_pre_order);
     for( j=1; j<b->num_preds(); j++ )
@@ -85,7 +82,6 @@
 }
 #endif
 
-//------------------------------combine_these_two------------------------------
 // Combine the live ranges def'd by these 2 Nodes.  N2 is an input to N1.
 void PhaseCoalesce::combine_these_two(Node *n1, Node *n2) {
   uint lr1 = _phc._lrg_map.find(n1);
@@ -127,18 +123,15 @@
   }
 }
 
-//------------------------------coalesce_driver--------------------------------
 // Copy coalescing
-void PhaseCoalesce::coalesce_driver( ) {
-
+void PhaseCoalesce::coalesce_driver() {
   verify();
   // Coalesce from high frequency to low
-  for( uint i=0; i<_phc._cfg._num_blocks; i++ )
-    coalesce( _phc._blks[i] );
-
+  for (uint i = 0; i < _phc._cfg.number_of_blocks(); i++) {
+    coalesce(_phc._blks[i]);
+  }
 }
 
-//------------------------------insert_copy_with_overlap-----------------------
 // I am inserting copies to come out of SSA form.  In the general case, I am
 // doing a parallel renaming.  I'm in the Named world now, so I can't do a
 // general parallel renaming.  All the copies now use  "names" (live-ranges)
@@ -216,7 +209,6 @@
   b->_nodes.insert(last_use_idx+1,copy);
 }
 
-//------------------------------insert_copies----------------------------------
 void PhaseAggressiveCoalesce::insert_copies( Matcher &matcher ) {
   // We do LRGs compressing and fix a liveout data only here since the other
   // place in Split() is guarded by the assert which we never hit.
@@ -225,8 +217,8 @@
   for (uint lrg = 1; lrg < _phc._lrg_map.max_lrg_id(); lrg++) {
     uint compressed_lrg = _phc._lrg_map.find(lrg);
     if (lrg != compressed_lrg) {
-      for (uint bidx = 0; bidx < _phc._cfg._num_blocks; bidx++) {
-        IndexSet *liveout = _phc._live->live(_phc._cfg._blocks[bidx]);
+      for (uint bidx = 0; bidx < _phc._cfg.number_of_blocks(); bidx++) {
+        IndexSet *liveout = _phc._live->live(_phc._cfg.get_block(bidx));
         if (liveout->member(lrg)) {
           liveout->remove(lrg);
           liveout->insert(compressed_lrg);
@@ -239,10 +231,10 @@
   // Nodes with index less than '_unique' are original, non-virtual Nodes.
   _unique = C->unique();
 
-  for( uint i=0; i<_phc._cfg._num_blocks; i++ ) {
+  for (uint i = 0; i < _phc._cfg.number_of_blocks(); i++) {
     C->check_node_count(NodeLimitFudgeFactor, "out of nodes in coalesce");
     if (C->failing()) return;
-    Block *b = _phc._cfg._blocks[i];
+    Block *b = _phc._cfg.get_block(i);
     uint cnt = b->num_preds();  // Number of inputs to the Phi
 
     for( uint l = 1; l<b->_nodes.size(); l++ ) {
@@ -330,9 +322,7 @@
               copy = m->clone();
               // Insert the copy in the basic block, just before us
               b->_nodes.insert(l++, copy);
-              if(_phc.clone_projs(b, l, m, copy, _phc._lrg_map)) {
-                l++;
-              }
+              l += _phc.clone_projs(b, l, m, copy, _phc._lrg_map);
             } else {
               const RegMask *rm = C->matcher()->idealreg2spillmask[m->ideal_reg()];
               copy = new (C) MachSpillCopyNode(m, *rm, *rm);
@@ -403,8 +393,7 @@
   } // End of for all blocks
 }
 
-//=============================================================================
-//------------------------------coalesce---------------------------------------
+
 // Aggressive (but pessimistic) copy coalescing of a single block
 
 // The following coalesce pass represents a single round of aggressive
@@ -464,20 +453,16 @@
   } // End of for all instructions in block
 }
 
-//=============================================================================
-//------------------------------PhaseConservativeCoalesce----------------------
 PhaseConservativeCoalesce::PhaseConservativeCoalesce(PhaseChaitin &chaitin) : PhaseCoalesce(chaitin) {
   _ulr.initialize(_phc._lrg_map.max_lrg_id());
 }
 
-//------------------------------verify-----------------------------------------
 void PhaseConservativeCoalesce::verify() {
 #ifdef ASSERT
   _phc.set_was_low();
 #endif
 }
 
-//------------------------------union_helper-----------------------------------
 void PhaseConservativeCoalesce::union_helper( Node *lr1_node, Node *lr2_node, uint lr1, uint lr2, Node *src_def, Node *dst_copy, Node *src_copy, Block *b, uint bindex ) {
   // Join live ranges.  Merge larger into smaller.  Union lr2 into lr1 in the
   // union-find tree
@@ -520,7 +505,6 @@
   }
 }
 
-//------------------------------compute_separating_interferences---------------
 // Factored code from copy_copy that computes extra interferences from
 // lengthening a live range by double-coalescing.
 uint PhaseConservativeCoalesce::compute_separating_interferences(Node *dst_copy, Node *src_copy, Block *b, uint bindex, RegMask &rm, uint reg_degree, uint rm_size, uint lr1, uint lr2 ) {
@@ -586,7 +570,6 @@
   return reg_degree;
 }
 
-//------------------------------update_ifg-------------------------------------
 void PhaseConservativeCoalesce::update_ifg(uint lr1, uint lr2, IndexSet *n_lr1, IndexSet *n_lr2) {
   // Some original neighbors of lr1 might have gone away
   // because the constrained register mask prevented them.
@@ -616,7 +599,6 @@
       lrgs(neighbor).inc_degree( lrg1.compute_degree(lrgs(neighbor)) );
 }
 
-//------------------------------record_bias------------------------------------
 static void record_bias( const PhaseIFG *ifg, int lr1, int lr2 ) {
   // Tag copy bias here
   if( !ifg->lrgs(lr1)._copy_bias )
@@ -625,7 +607,6 @@
     ifg->lrgs(lr2)._copy_bias = lr1;
 }
 
-//------------------------------copy_copy--------------------------------------
 // See if I can coalesce a series of multiple copies together.  I need the
 // final dest copy and the original src copy.  They can be the same Node.
 // Compute the compatible register masks.
@@ -785,7 +766,6 @@
   return true;
 }
 
-//------------------------------coalesce---------------------------------------
 // Conservative (but pessimistic) copy coalescing of a single block
 void PhaseConservativeCoalesce::coalesce( Block *b ) {
   // Bail out on infrequent blocks
--- a/src/share/vm/opto/compile.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/opto/compile.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -2136,7 +2136,9 @@
 //------------------------------Code_Gen---------------------------------------
 // Given a graph, generate code for it
 void Compile::Code_Gen() {
-  if (failing())  return;
+  if (failing()) {
+    return;
+  }
 
   // Perform instruction selection.  You might think we could reclaim Matcher
   // memory PDQ, but actually the Matcher is used in generating spill code.
@@ -2148,12 +2150,11 @@
   // nodes.  Mapping is only valid at the root of each matched subtree.
   NOT_PRODUCT( verify_graph_edges(); )
 
-  Node_List proj_list;
-  Matcher m(proj_list);
-  _matcher = &m;
+  Matcher matcher;
+  _matcher = &matcher;
   {
     TracePhase t2("matcher", &_t_matcher, true);
-    m.match();
+    matcher.match();
   }
   // In debug mode can dump m._nodes.dump() for mapping of ideal to machine
   // nodes.  Mapping is only valid at the root of each matched subtree.
@@ -2161,31 +2162,26 @@
 
   // If you have too many nodes, or if matching has failed, bail out
   check_node_count(0, "out of nodes matching instructions");
-  if (failing())  return;
+  if (failing()) {
+    return;
+  }
 
   // Build a proper-looking CFG
-  PhaseCFG cfg(node_arena(), root(), m);
+  PhaseCFG cfg(node_arena(), root(), matcher);
   _cfg = &cfg;
   {
     NOT_PRODUCT( TracePhase t2("scheduler", &_t_scheduler, TimeCompiler); )
-    cfg.Dominators();
-    if (failing())  return;
-
+    bool success = cfg.do_global_code_motion();
+    if (!success) {
+      return;
+    }
+
+    print_method(PHASE_GLOBAL_CODE_MOTION, 2);
     NOT_PRODUCT( verify_graph_edges(); )
-
-    cfg.Estimate_Block_Frequency();
-    cfg.GlobalCodeMotion(m,unique(),proj_list);
-    if (failing())  return;
-
-    print_method(PHASE_GLOBAL_CODE_MOTION, 2);
-
-    NOT_PRODUCT( verify_graph_edges(); )
-
     debug_only( cfg.verify(); )
   }
-  NOT_PRODUCT( verify_graph_edges(); )
-
-  PhaseChaitin regalloc(unique(), cfg, m);
+
+  PhaseChaitin regalloc(unique(), cfg, matcher);
   _regalloc = &regalloc;
   {
     TracePhase t2("regalloc", &_t_registerAllocation, true);
@@ -2206,7 +2202,7 @@
   // can now safely remove it.
   {
     NOT_PRODUCT( TracePhase t2("blockOrdering", &_t_blockOrdering, TimeCompiler); )
-    cfg.remove_empty();
+    cfg.remove_empty_blocks();
     if (do_freq_based_layout()) {
       PhaseBlockLayout layout(cfg);
     } else {
@@ -2253,38 +2249,50 @@
   _regalloc->dump_frame();
 
   Node *n = NULL;
-  for( uint i=0; i<_cfg->_num_blocks; i++ ) {
-    if (VMThread::should_terminate()) { cut_short = true; break; }
-    Block *b = _cfg->_blocks[i];
-    if (b->is_connector() && !Verbose) continue;
-    n = b->_nodes[0];
-    if (pcs && n->_idx < pc_limit)
+  for (uint i = 0; i < _cfg->number_of_blocks(); i++) {
+    if (VMThread::should_terminate()) {
+      cut_short = true;
+      break;
+    }
+    Block* block = _cfg->get_block(i);
+    if (block->is_connector() && !Verbose) {
+      continue;
+    }
+    n = block->_nodes[0];
+    if (pcs && n->_idx < pc_limit) {
       tty->print("%3.3x   ", pcs[n->_idx]);
-    else
+    } else {
       tty->print("      ");
-    b->dump_head(_cfg);
-    if (b->is_connector()) {
+    }
+    block->dump_head(_cfg);
+    if (block->is_connector()) {
       tty->print_cr("        # Empty connector block");
-    } else if (b->num_preds() == 2 && b->pred(1)->is_CatchProj() && b->pred(1)->as_CatchProj()->_con == CatchProjNode::fall_through_index) {
+    } else if (block->num_preds() == 2 && block->pred(1)->is_CatchProj() && block->pred(1)->as_CatchProj()->_con == CatchProjNode::fall_through_index) {
       tty->print_cr("        # Block is sole successor of call");
     }
 
     // For all instructions
     Node *delay = NULL;
-    for( uint j = 0; j<b->_nodes.size(); j++ ) {
-      if (VMThread::should_terminate()) { cut_short = true; break; }
-      n = b->_nodes[j];
+    for (uint j = 0; j < block->_nodes.size(); j++) {
+      if (VMThread::should_terminate()) {
+        cut_short = true;
+        break;
+      }
+      n = block->_nodes[j];
       if (valid_bundle_info(n)) {
-        Bundle *bundle = node_bundling(n);
+        Bundle* bundle = node_bundling(n);
         if (bundle->used_in_unconditional_delay()) {
           delay = n;
           continue;
         }
-        if (bundle->starts_bundle())
+        if (bundle->starts_bundle()) {
           starts_bundle = '+';
+        }
       }
 
-      if (WizardMode) n->dump();
+      if (WizardMode) {
+        n->dump();
+      }
 
       if( !n->is_Region() &&    // Dont print in the Assembly
           !n->is_Phi() &&       // a few noisely useless nodes
--- a/src/share/vm/opto/domgraph.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/opto/domgraph.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -32,9 +32,6 @@
 
 // Portions of code courtesy of Clifford Click
 
-// Optimization - Graph Style
-
-//------------------------------Tarjan-----------------------------------------
 // A data structure that holds all the information needed to find dominators.
 struct Tarjan {
   Block *_block;                // Basic block for this info
@@ -60,23 +57,21 @@
 
 };
 
-//------------------------------Dominator--------------------------------------
 // Compute the dominator tree of the CFG.  The CFG must already have been
 // constructed.  This is the Lengauer & Tarjan O(E-alpha(E,V)) algorithm.
-void PhaseCFG::Dominators( ) {
+void PhaseCFG::build_dominator_tree() {
   // Pre-grow the blocks array, prior to the ResourceMark kicking in
-  _blocks.map(_num_blocks,0);
+  _blocks.map(number_of_blocks(), 0);
 
   ResourceMark rm;
   // Setup mappings from my Graph to Tarjan's stuff and back
   // Note: Tarjan uses 1-based arrays
-  Tarjan *tarjan = NEW_RESOURCE_ARRAY(Tarjan,_num_blocks+1);
+  Tarjan* tarjan = NEW_RESOURCE_ARRAY(Tarjan, number_of_blocks() + 1);
 
   // Tarjan's algorithm, almost verbatim:
   // Step 1:
-  _rpo_ctr = _num_blocks;
-  uint dfsnum = DFS( tarjan );
-  if( dfsnum-1 != _num_blocks ) {// Check for unreachable loops!
+  uint dfsnum = do_DFS(tarjan, number_of_blocks());
+  if (dfsnum - 1 != number_of_blocks()) { // Check for unreachable loops!
     // If the returned dfsnum does not match the number of blocks, then we
     // must have some unreachable loops.  These can be made at any time by
     // IterGVN.  They are cleaned up by CCP or the loop opts, but the last
@@ -93,14 +88,13 @@
     C->record_method_not_compilable("unreachable loop");
     return;
   }
-  _blocks._cnt = _num_blocks;
+  _blocks._cnt = number_of_blocks();
 
   // Tarjan is using 1-based arrays, so these are some initialize flags
   tarjan[0]._size = tarjan[0]._semi = 0;
   tarjan[0]._label = &tarjan[0];
 
-  uint i;
-  for( i=_num_blocks; i>=2; i-- ) { // For all vertices in DFS order
+  for (uint i = number_of_blocks(); i >= 2; i--) { // For all vertices in DFS order
     Tarjan *w = &tarjan[i];     // Get vertex from DFS
 
     // Step 2:
@@ -130,19 +124,19 @@
   }
 
   // Step 4:
-  for( i=2; i <= _num_blocks; i++ ) {
+  for (uint i = 2; i <= number_of_blocks(); i++) {
     Tarjan *w = &tarjan[i];
     if( w->_dom != &tarjan[w->_semi] )
       w->_dom = w->_dom->_dom;
     w->_dom_next = w->_dom_child = NULL;  // Initialize for building tree later
   }
   // No immediate dominator for the root
-  Tarjan *w = &tarjan[_broot->_pre_order];
+  Tarjan *w = &tarjan[get_root_block()->_pre_order];
   w->_dom = NULL;
   w->_dom_next = w->_dom_child = NULL;  // Initialize for building tree later
 
   // Convert the dominator tree array into my kind of graph
-  for( i=1; i<=_num_blocks;i++){// For all Tarjan vertices
+  for(uint i = 1; i <= number_of_blocks(); i++){ // For all Tarjan vertices
     Tarjan *t = &tarjan[i];     // Handy access
     Tarjan *tdom = t->_dom;     // Handy access to immediate dominator
     if( tdom )  {               // Root has no immediate dominator
@@ -152,11 +146,10 @@
     } else
       t->_block->_idom = NULL;  // Root
   }
-  w->setdepth( _num_blocks+1 ); // Set depth in dominator tree
+  w->setdepth(number_of_blocks() + 1); // Set depth in dominator tree
 
 }
 
-//----------------------------Block_Stack--------------------------------------
 class Block_Stack {
   private:
     struct Block_Descr {
@@ -214,7 +207,6 @@
     }
 };
 
-//-------------------------most_frequent_successor-----------------------------
 // Find the index into the b->succs[] array of the most frequent successor.
 uint Block_Stack::most_frequent_successor( Block *b ) {
   uint freq_idx = 0;
@@ -258,40 +250,38 @@
   return freq_idx;
 }
 
-//------------------------------DFS--------------------------------------------
 // Perform DFS search.  Setup 'vertex' as DFS to vertex mapping.  Setup
 // 'semi' as vertex to DFS mapping.  Set 'parent' to DFS parent.
-uint PhaseCFG::DFS( Tarjan *tarjan ) {
-  Block *b = _broot;
+uint PhaseCFG::do_DFS(Tarjan *tarjan, uint rpo_counter) {
+  Block* root_block = get_root_block();
   uint pre_order = 1;
-  // Allocate stack of size _num_blocks+1 to avoid frequent realloc
-  Block_Stack bstack(tarjan, _num_blocks+1);
+  // Allocate stack of size number_of_blocks() + 1 to avoid frequent realloc
+  Block_Stack bstack(tarjan, number_of_blocks() + 1);
 
   // Push on stack the state for the first block
-  bstack.push(pre_order, b);
+  bstack.push(pre_order, root_block);
   ++pre_order;
 
   while (bstack.is_nonempty()) {
     if (!bstack.last_successor()) {
       // Walk over all successors in pre-order (DFS).
-      Block *s = bstack.next_successor();
-      if (s->_pre_order == 0) { // Check for no-pre-order, not-visited
+      Block* next_block = bstack.next_successor();
+      if (next_block->_pre_order == 0) { // Check for no-pre-order, not-visited
         // Push on stack the state of successor
-        bstack.push(pre_order, s);
+        bstack.push(pre_order, next_block);
         ++pre_order;
       }
     }
     else {
       // Build a reverse post-order in the CFG _blocks array
       Block *stack_top = bstack.pop();
-      stack_top->_rpo = --_rpo_ctr;
+      stack_top->_rpo = --rpo_counter;
       _blocks.map(stack_top->_rpo, stack_top);
     }
   }
   return pre_order;
 }
 
-//------------------------------COMPRESS---------------------------------------
 void Tarjan::COMPRESS()
 {
   assert( _ancestor != 0, "" );
@@ -303,14 +293,12 @@
   }
 }
 
-//------------------------------EVAL-------------------------------------------
 Tarjan *Tarjan::EVAL() {
   if( !_ancestor ) return _label;
   COMPRESS();
   return (_ancestor->_label->_semi >= _label->_semi) ? _label : _ancestor->_label;
 }
 
-//------------------------------LINK-------------------------------------------
 void Tarjan::LINK( Tarjan *w, Tarjan *tarjan0 ) {
   Tarjan *s = w;
   while( w->_label->_semi < s->_child->_label->_semi ) {
@@ -333,7 +321,6 @@
   }
 }
 
-//------------------------------setdepth---------------------------------------
 void Tarjan::setdepth( uint stack_size ) {
   Tarjan **top  = NEW_RESOURCE_ARRAY(Tarjan*, stack_size);
   Tarjan **next = top;
@@ -362,8 +349,7 @@
   } while (last < top);
 }
 
-//*********************** DOMINATORS ON THE SEA OF NODES***********************
-//------------------------------NTarjan----------------------------------------
+// Compute dominators on the Sea of Nodes form
 // A data structure that holds all the information needed to find dominators.
 struct NTarjan {
   Node *_control;               // Control node associated with this info
@@ -396,7 +382,6 @@
 #endif
 };
 
-//------------------------------Dominator--------------------------------------
 // Compute the dominator tree of the sea of nodes.  This version walks all CFG
 // nodes (using the is_CFG() call) and places them in a dominator tree.  Thus,
 // it needs a count of the CFG nodes for the mapping table. This is the
@@ -517,7 +502,6 @@
   }
 }
 
-//------------------------------DFS--------------------------------------------
 // Perform DFS search.  Setup 'vertex' as DFS to vertex mapping.  Setup
 // 'semi' as vertex to DFS mapping.  Set 'parent' to DFS parent.
 int NTarjan::DFS( NTarjan *ntarjan, VectorSet &visited, PhaseIdealLoop *pil, uint *dfsorder) {
@@ -560,7 +544,6 @@
   return dfsnum;
 }
 
-//------------------------------COMPRESS---------------------------------------
 void NTarjan::COMPRESS()
 {
   assert( _ancestor != 0, "" );
@@ -572,14 +555,12 @@
   }
 }
 
-//------------------------------EVAL-------------------------------------------
 NTarjan *NTarjan::EVAL() {
   if( !_ancestor ) return _label;
   COMPRESS();
   return (_ancestor->_label->_semi >= _label->_semi) ? _label : _ancestor->_label;
 }
 
-//------------------------------LINK-------------------------------------------
 void NTarjan::LINK( NTarjan *w, NTarjan *ntarjan0 ) {
   NTarjan *s = w;
   while( w->_label->_semi < s->_child->_label->_semi ) {
@@ -602,7 +583,6 @@
   }
 }
 
-//------------------------------setdepth---------------------------------------
 void NTarjan::setdepth( uint stack_size, uint *dom_depth ) {
   NTarjan **top  = NEW_RESOURCE_ARRAY(NTarjan*, stack_size);
   NTarjan **next = top;
@@ -631,7 +611,6 @@
   } while (last < top);
 }
 
-//------------------------------dump-------------------------------------------
 #ifndef PRODUCT
 void NTarjan::dump(int offset) const {
   // Dump the data from this node
--- a/src/share/vm/opto/gcm.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/opto/gcm.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -121,27 +121,30 @@
 
 //------------------------------schedule_pinned_nodes--------------------------
 // Set the basic block for Nodes pinned into blocks
-void PhaseCFG::schedule_pinned_nodes( VectorSet &visited ) {
+void PhaseCFG::schedule_pinned_nodes(VectorSet &visited) {
   // Allocate node stack of size C->unique()+8 to avoid frequent realloc
-  GrowableArray <Node *> spstack(C->unique()+8);
+  GrowableArray <Node *> spstack(C->unique() + 8);
   spstack.push(_root);
-  while ( spstack.is_nonempty() ) {
-    Node *n = spstack.pop();
-    if( !visited.test_set(n->_idx) ) { // Test node and flag it as visited
-      if( n->pinned() && !has_block(n)) {  // Pinned?  Nail it down!
-        assert( n->in(0), "pinned Node must have Control" );
+  while (spstack.is_nonempty()) {
+    Node* node = spstack.pop();
+    if (!visited.test_set(node->_idx)) { // Test node and flag it as visited
+      if (node->pinned() && !has_block(node)) {  // Pinned?  Nail it down!
+        assert(node->in(0), "pinned Node must have Control");
         // Before setting block replace block_proj control edge
-        replace_block_proj_ctrl(n);
-        Node *input = n->in(0);
+        replace_block_proj_ctrl(node);
+        Node* input = node->in(0);
         while (!input->is_block_start()) {
           input = input->in(0);
         }
-        Block *b = get_block_for_node(input); // Basic block of controlling input
-        schedule_node_into_block(n, b);
+        Block* block = get_block_for_node(input); // Basic block of controlling input
+        schedule_node_into_block(node, block);
       }
-      for( int i = n->req() - 1; i >= 0; --i ) {  // For all inputs
-        if( n->in(i) != NULL )
-          spstack.push(n->in(i));
+
+      // process all inputs that are non NULL
+      for (int i = node->req() - 1; i >= 0; --i) {
+        if (node->in(i) != NULL) {
+          spstack.push(node->in(i));
+        }
       }
     }
   }
@@ -205,32 +208,29 @@
 // which all their inputs occur.
 bool PhaseCFG::schedule_early(VectorSet &visited, Node_List &roots) {
   // Allocate stack with enough space to avoid frequent realloc
-  Node_Stack nstack(roots.Size() + 8); // (unique >> 1) + 24 from Java2D stats
-  // roots.push(_root); _root will be processed among C->top() inputs
+  Node_Stack nstack(roots.Size() + 8);
+  // _root will be processed among C->top() inputs
   roots.push(C->top());
   visited.set(C->top()->_idx);
 
   while (roots.size() != 0) {
     // Use local variables nstack_top_n & nstack_top_i to cache values
     // on stack's top.
-    Node *nstack_top_n = roots.pop();
-    uint  nstack_top_i = 0;
-//while_nstack_nonempty:
+    Node* parent_node = roots.pop();
+    uint  input_index = 0;
+
     while (true) {
-      // Get parent node and next input's index from stack's top.
-      Node *n = nstack_top_n;
-      uint  i = nstack_top_i;
-
-      if (i == 0) {
+      if (input_index == 0) {
         // Fixup some control.  Constants without control get attached
         // to root and nodes that use is_block_proj() nodes should be attached
         // to the region that starts their block.
-        const Node *in0 = n->in(0);
-        if (in0 != NULL) {              // Control-dependent?
-          replace_block_proj_ctrl(n);
-        } else {               // n->in(0) == NULL
-          if (n->req() == 1) { // This guy is a constant with NO inputs?
-            n->set_req(0, _root);
+        const Node* control_input = parent_node->in(0);
+        if (control_input != NULL) {
+          replace_block_proj_ctrl(parent_node);
+        } else {
+          // Is a constant with NO inputs?
+          if (parent_node->req() == 1) {
+            parent_node->set_req(0, _root);
           }
         }
       }
@@ -239,37 +239,47 @@
       // input is already in a block we quit following inputs (to avoid
       // cycles). Instead we put that Node on a worklist to be handled
       // later (since IT'S inputs may not have a block yet).
-      bool done = true;              // Assume all n's inputs will be processed
-      while (i < n->len()) {         // For all inputs
-        Node *in = n->in(i);         // Get input
-        ++i;
-        if (in == NULL) continue;    // Ignore NULL, missing inputs
+
+      // Assume all n's inputs will be processed
+      bool done = true;
+
+      while (input_index < parent_node->len()) {
+        Node* in = parent_node->in(input_index++);
+        if (in == NULL) {
+          continue;
+        }
+
         int is_visited = visited.test_set(in->_idx);
-        if (!has_block(in)) { // Missing block selection?
+        if (!has_block(in)) {
           if (is_visited) {
-            // assert( !visited.test(in->_idx), "did not schedule early" );
             return false;
           }
-          nstack.push(n, i);         // Save parent node and next input's index.
-          nstack_top_n = in;         // Process current input now.
-          nstack_top_i = 0;
-          done = false;              // Not all n's inputs processed.
-          break; // continue while_nstack_nonempty;
-        } else if (!is_visited) {    // Input not yet visited?
-          roots.push(in);            // Visit this guy later, using worklist
+          // Save parent node and next input's index.
+          nstack.push(parent_node, input_index);
+          // Process current input now.
+          parent_node = in;
+          input_index = 0;
+          // Not all n's inputs processed.
+          done = false;
+          break;
+        } else if (!is_visited) {
+          // Visit this guy later, using worklist
+          roots.push(in);
         }
       }
+
       if (done) {
         // All of n's inputs have been processed, complete post-processing.
 
         // Some instructions are pinned into a block.  These include Region,
         // Phi, Start, Return, and other control-dependent instructions and
         // any projections which depend on them.
-        if (!n->pinned()) {
+        if (!parent_node->pinned()) {
           // Set earliest legal block.
-          map_node_to_block(n, find_deepest_input(n, this));
+          Block* earliest_block = find_deepest_input(parent_node, this);
+          map_node_to_block(parent_node, earliest_block);
         } else {
-          assert(get_block_for_node(n) == get_block_for_node(n->in(0)), "Pinned Node should be at the same block as its control edge");
+          assert(get_block_for_node(parent_node) == get_block_for_node(parent_node->in(0)), "Pinned Node should be at the same block as its control edge");
         }
 
         if (nstack.is_empty()) {
@@ -278,12 +288,12 @@
           break;
         }
         // Get saved parent node and next input's index.
-        nstack_top_n = nstack.node();
-        nstack_top_i = nstack.index();
+        parent_node = nstack.node();
+        input_index = nstack.index();
         nstack.pop();
-      } //    if (done)
-    }   // while (true)
-  }     // while (roots.size() != 0)
+      }
+    }
+  }
   return true;
 }
 
@@ -847,7 +857,7 @@
 
 //------------------------------ComputeLatenciesBackwards----------------------
 // Compute the latency of all the instructions.
-void PhaseCFG::ComputeLatenciesBackwards(VectorSet &visited, Node_List &stack) {
+void PhaseCFG::compute_latencies_backwards(VectorSet &visited, Node_List &stack) {
 #ifndef PRODUCT
   if (trace_opto_pipelining())
     tty->print("\n#---- ComputeLatenciesBackwards ----\n");
@@ -870,31 +880,34 @@
   // Set the latency for this instruction
 #ifndef PRODUCT
   if (trace_opto_pipelining()) {
-    tty->print("# latency_to_inputs: node_latency[%d] = %d for node",
-               n->_idx, _node_latency->at_grow(n->_idx));
+    tty->print("# latency_to_inputs: node_latency[%d] = %d for node", n->_idx, get_latency_for_node(n));
     dump();
   }
 #endif
 
-  if (n->is_Proj())
+  if (n->is_Proj()) {
     n = n->in(0);
+  }
 
-  if (n->is_Root())
+  if (n->is_Root()) {
     return;
+  }
 
   uint nlen = n->len();
-  uint use_latency = _node_latency->at_grow(n->_idx);
+  uint use_latency = get_latency_for_node(n);
   uint use_pre_order = get_block_for_node(n)->_pre_order;
 
-  for ( uint j=0; j<nlen; j++ ) {
+  for (uint j = 0; j < nlen; j++) {
     Node *def = n->in(j);
 
-    if (!def || def == n)
+    if (!def || def == n) {
       continue;
+    }
 
     // Walk backwards thru projections
-    if (def->is_Proj())
+    if (def->is_Proj()) {
       def = def->in(0);
+    }
 
 #ifndef PRODUCT
     if (trace_opto_pipelining()) {
@@ -907,22 +920,20 @@
     Block *def_block = get_block_for_node(def);
     uint def_pre_order = def_block ? def_block->_pre_order : 0;
 
-    if ( (use_pre_order <  def_pre_order) ||
-         (use_pre_order == def_pre_order && n->is_Phi()) )
+    if ((use_pre_order <  def_pre_order) || (use_pre_order == def_pre_order && n->is_Phi())) {
       continue;
+    }
 
     uint delta_latency = n->latency(j);
     uint current_latency = delta_latency + use_latency;
 
-    if (_node_latency->at_grow(def->_idx) < current_latency) {
-      _node_latency->at_put_grow(def->_idx, current_latency);
+    if (get_latency_for_node(def) < current_latency) {
+      set_latency_for_node(def, current_latency);
     }
 
 #ifndef PRODUCT
     if (trace_opto_pipelining()) {
-      tty->print_cr("#      %d + edge_latency(%d) == %d -> %d, node_latency[%d] = %d",
-                    use_latency, j, delta_latency, current_latency, def->_idx,
-                    _node_latency->at_grow(def->_idx));
+      tty->print_cr("#      %d + edge_latency(%d) == %d -> %d, node_latency[%d] = %d", use_latency, j, delta_latency, current_latency, def->_idx, get_latency_for_node(def));
     }
 #endif
   }
@@ -957,7 +968,7 @@
       return 0;
 
     uint nlen = use->len();
-    uint nl = _node_latency->at_grow(use->_idx);
+    uint nl = get_latency_for_node(use);
 
     for ( uint j=0; j<nlen; j++ ) {
       if (use->in(j) == n) {
@@ -992,8 +1003,7 @@
   // Set the latency for this instruction
 #ifndef PRODUCT
   if (trace_opto_pipelining()) {
-    tty->print("# latency_from_outputs: node_latency[%d] = %d for node",
-               n->_idx, _node_latency->at_grow(n->_idx));
+    tty->print("# latency_from_outputs: node_latency[%d] = %d for node", n->_idx, get_latency_for_node(n));
     dump();
   }
 #endif
@@ -1006,7 +1016,7 @@
     if (latency < l) latency = l;
   }
 
-  _node_latency->at_put_grow(n->_idx, latency);
+  set_latency_for_node(n, latency);
 }
 
 //------------------------------hoist_to_cheaper_block-------------------------
@@ -1016,9 +1026,9 @@
   const double delta = 1+PROB_UNLIKELY_MAG(4);
   Block* least       = LCA;
   double least_freq  = least->_freq;
-  uint target        = _node_latency->at_grow(self->_idx);
-  uint start_latency = _node_latency->at_grow(LCA->_nodes[0]->_idx);
-  uint end_latency   = _node_latency->at_grow(LCA->_nodes[LCA->end_idx()]->_idx);
+  uint target        = get_latency_for_node(self);
+  uint start_latency = get_latency_for_node(LCA->_nodes[0]);
+  uint end_latency   = get_latency_for_node(LCA->_nodes[LCA->end_idx()]);
   bool in_latency    = (target <= start_latency);
   const Block* root_block = get_block_for_node(_root);
 
@@ -1035,8 +1045,7 @@
 
 #ifndef PRODUCT
   if (trace_opto_pipelining()) {
-    tty->print("# Find cheaper block for latency %d: ",
-      _node_latency->at_grow(self->_idx));
+    tty->print("# Find cheaper block for latency %d: ", get_latency_for_node(self));
     self->dump();
     tty->print_cr("#   B%d: start latency for [%4d]=%d, end latency for [%4d]=%d, freq=%g",
       LCA->_pre_order,
@@ -1065,9 +1074,9 @@
     if (mach && LCA == root_block)
       break;
 
-    uint start_lat = _node_latency->at_grow(LCA->_nodes[0]->_idx);
+    uint start_lat = get_latency_for_node(LCA->_nodes[0]);
     uint end_idx   = LCA->end_idx();
-    uint end_lat   = _node_latency->at_grow(LCA->_nodes[end_idx]->_idx);
+    uint end_lat   = get_latency_for_node(LCA->_nodes[end_idx]);
     double LCA_freq = LCA->_freq;
 #ifndef PRODUCT
     if (trace_opto_pipelining()) {
@@ -1109,7 +1118,7 @@
       tty->print_cr("#  Change latency for [%4d] from %d to %d", self->_idx, target, end_latency);
     }
 #endif
-    _node_latency->at_put_grow(self->_idx, end_latency);
+    set_latency_for_node(self, end_latency);
     partial_latency_of_defs(self);
   }
 
@@ -1255,7 +1264,7 @@
 } // end ScheduleLate
 
 //------------------------------GlobalCodeMotion-------------------------------
-void PhaseCFG::GlobalCodeMotion( Matcher &matcher, uint unique, Node_List &proj_list ) {
+void PhaseCFG::global_code_motion() {
   ResourceMark rm;
 
 #ifndef PRODUCT
@@ -1265,21 +1274,22 @@
 #endif
 
   // Initialize the node to block mapping for things on the proj_list
-  for (uint i = 0; i < proj_list.size(); i++) {
-    unmap_node_from_block(proj_list[i]);
+  for (uint i = 0; i < _matcher.number_of_projections(); i++) {
+    unmap_node_from_block(_matcher.get_projection(i));
   }
 
   // Set the basic block for Nodes pinned into blocks
-  Arena *a = Thread::current()->resource_area();
-  VectorSet visited(a);
-  schedule_pinned_nodes( visited );
+  Arena* arena = Thread::current()->resource_area();
+  VectorSet visited(arena);
+  schedule_pinned_nodes(visited);
 
   // Find the earliest Block any instruction can be placed in.  Some
   // instructions are pinned into Blocks.  Unpinned instructions can
   // appear in last block in which all their inputs occur.
   visited.Clear();
-  Node_List stack(a);
-  stack.map( (unique >> 1) + 16, NULL); // Pre-grow the list
+  Node_List stack(arena);
+  // Pre-grow the list
+  stack.map((C->unique() >> 1) + 16, NULL);
   if (!schedule_early(visited, stack)) {
     // Bailout without retry
     C->record_method_not_compilable("early schedule failed");
@@ -1287,29 +1297,25 @@
   }
 
   // Build Def-Use edges.
-  proj_list.push(_root);        // Add real root as another root
-  proj_list.pop();
-
   // Compute the latency information (via backwards walk) for all the
   // instructions in the graph
   _node_latency = new GrowableArray<uint>(); // resource_area allocation
 
-  if( C->do_scheduling() )
-    ComputeLatenciesBackwards(visited, stack);
+  if (C->do_scheduling()) {
+    compute_latencies_backwards(visited, stack);
+  }
 
   // Now schedule all codes as LATE as possible.  This is the LCA in the
   // dominator tree of all USES of a value.  Pick the block with the least
   // loop nesting depth that is lowest in the dominator tree.
   // ( visited.Clear() called in schedule_late()->Node_Backward_Iterator() )
   schedule_late(visited, stack);
-  if( C->failing() ) {
+  if (C->failing()) {
     // schedule_late fails only when graph is incorrect.
     assert(!VerifyGraphEdges, "verification should have failed");
     return;
   }
 
-  unique = C->unique();
-
 #ifndef PRODUCT
   if (trace_opto_pipelining()) {
     tty->print("\n---- Detect implicit null checks ----\n");
@@ -1332,10 +1338,11 @@
     // By reversing the loop direction we get a very minor gain on mpegaudio.
     // Feel free to revert to a forward loop for clarity.
     // for( int i=0; i < (int)matcher._null_check_tests.size(); i+=2 ) {
-    for( int i= matcher._null_check_tests.size()-2; i>=0; i-=2 ) {
-      Node *proj = matcher._null_check_tests[i  ];
-      Node *val  = matcher._null_check_tests[i+1];
-      get_block_for_node(proj)->implicit_null_check(this, proj, val, allowed_reasons);
+    for (int i = _matcher._null_check_tests.size() - 2; i >= 0; i -= 2) {
+      Node* proj = _matcher._null_check_tests[i];
+      Node* val  = _matcher._null_check_tests[i + 1];
+      Block* block = get_block_for_node(proj);
+      block->implicit_null_check(this, proj, val, allowed_reasons);
       // The implicit_null_check will only perform the transformation
       // if the null branch is truly uncommon, *and* it leads to an
       // uncommon trap.  Combined with the too_many_traps guards
@@ -1352,11 +1359,11 @@
 
   // Schedule locally.  Right now a simple topological sort.
   // Later, do a real latency aware scheduler.
-  uint max_idx = C->unique();
-  GrowableArray<int> ready_cnt(max_idx, max_idx, -1);
+  GrowableArray<int> ready_cnt(C->unique(), C->unique(), -1);
   visited.Clear();
-  for (uint i = 0; i < _num_blocks; i++) {
-    if (!_blocks[i]->schedule_local(this, matcher, ready_cnt, visited)) {
+  for (uint i = 0; i < number_of_blocks(); i++) {
+    Block* block = get_block(i);
+    if (!block->schedule_local(this, _matcher, ready_cnt, visited)) {
       if (!C->failure_reason_is(C2Compiler::retry_no_subsuming_loads())) {
         C->record_method_not_compilable("local schedule failed");
       }
@@ -1366,15 +1373,17 @@
 
   // If we inserted any instructions between a Call and his CatchNode,
   // clone the instructions on all paths below the Catch.
-  for (uint i = 0; i < _num_blocks; i++) {
-    _blocks[i]->call_catch_cleanup(this, C);
+  for (uint i = 0; i < number_of_blocks(); i++) {
+    Block* block = get_block(i);
+    block->call_catch_cleanup(this, C);
   }
 
 #ifndef PRODUCT
   if (trace_opto_pipelining()) {
     tty->print("\n---- After GlobalCodeMotion ----\n");
-    for (uint i = 0; i < _num_blocks; i++) {
-      _blocks[i]->dump();
+    for (uint i = 0; i < number_of_blocks(); i++) {
+      Block* block = get_block(i);
+      block->dump();
     }
   }
 #endif
@@ -1382,10 +1391,29 @@
   _node_latency = (GrowableArray<uint> *)0xdeadbeef;
 }
 
+bool PhaseCFG::do_global_code_motion() {
+
+  build_dominator_tree();
+  if (C->failing()) {
+    return false;
+  }
+
+  NOT_PRODUCT( C->verify_graph_edges(); )
+
+  estimate_block_frequency();
+
+  global_code_motion();
+
+  if (C->failing()) {
+    return false;
+  }
+
+  return true;
+}
 
 //------------------------------Estimate_Block_Frequency-----------------------
 // Estimate block frequencies based on IfNode probabilities.
-void PhaseCFG::Estimate_Block_Frequency() {
+void PhaseCFG::estimate_block_frequency() {
 
   // Force conditional branches leading to uncommon traps to be unlikely,
   // not because we get to the uncommon_trap with less relative frequency,
@@ -1393,7 +1421,7 @@
   // there once.
   if (C->do_freq_based_layout()) {
     Block_List worklist;
-    Block* root_blk = _blocks[0];
+    Block* root_blk = get_block(0);
     for (uint i = 1; i < root_blk->num_preds(); i++) {
       Block *pb = get_block_for_node(root_blk->pred(i));
       if (pb->has_uncommon_code()) {
@@ -1402,7 +1430,9 @@
     }
     while (worklist.size() > 0) {
       Block* uct = worklist.pop();
-      if (uct == _broot) continue;
+      if (uct == get_root_block()) {
+        continue;
+      }
       for (uint i = 1; i < uct->num_preds(); i++) {
         Block *pb = get_block_for_node(uct->pred(i));
         if (pb->_num_succs == 1) {
@@ -1426,12 +1456,12 @@
   _root_loop->scale_freq();
 
   // Save outmost loop frequency for LRG frequency threshold
-  _outer_loop_freq = _root_loop->outer_loop_freq();
+  _outer_loop_frequency = _root_loop->outer_loop_freq();
 
   // force paths ending at uncommon traps to be infrequent
   if (!C->do_freq_based_layout()) {
     Block_List worklist;
-    Block* root_blk = _blocks[0];
+    Block* root_blk = get_block(0);
     for (uint i = 1; i < root_blk->num_preds(); i++) {
       Block *pb = get_block_for_node(root_blk->pred(i));
       if (pb->has_uncommon_code()) {
@@ -1451,8 +1481,8 @@
   }
 
 #ifdef ASSERT
-  for (uint i = 0; i < _num_blocks; i++ ) {
-    Block *b = _blocks[i];
+  for (uint i = 0; i < number_of_blocks(); i++) {
+    Block* b = get_block(i);
     assert(b->_freq >= MIN_BLOCK_FREQUENCY, "Register Allocator requires meaningful block frequency");
   }
 #endif
@@ -1476,16 +1506,16 @@
 CFGLoop* PhaseCFG::create_loop_tree() {
 
 #ifdef ASSERT
-  assert( _blocks[0] == _broot, "" );
-  for (uint i = 0; i < _num_blocks; i++ ) {
-    Block *b = _blocks[i];
+  assert(get_block(0) == get_root_block(), "first block should be root block");
+  for (uint i = 0; i < number_of_blocks(); i++) {
+    Block* block = get_block(i);
     // Check that _loop field are clear...we could clear them if not.
-    assert(b->_loop == NULL, "clear _loop expected");
+    assert(block->_loop == NULL, "clear _loop expected");
     // Sanity check that the RPO numbering is reflected in the _blocks array.
     // It doesn't have to be for the loop tree to be built, but if it is not,
     // then the blocks have been reordered since dom graph building...which
     // may question the RPO numbering
-    assert(b->_rpo == i, "unexpected reverse post order number");
+    assert(block->_rpo == i, "unexpected reverse post order number");
   }
 #endif
 
@@ -1495,11 +1525,11 @@
   Block_List worklist;
 
   // Assign blocks to loops
-  for(uint i = _num_blocks - 1; i > 0; i-- ) { // skip Root block
-    Block *b = _blocks[i];
+  for(uint i = number_of_blocks() - 1; i > 0; i-- ) { // skip Root block
+    Block* block = get_block(i);
 
-    if (b->head()->is_Loop()) {
-      Block* loop_head = b;
+    if (block->head()->is_Loop()) {
+      Block* loop_head = block;
       assert(loop_head->num_preds() - 1 == 2, "loop must have 2 predecessors");
       Node* tail_n = loop_head->pred(LoopNode::LoopBackControl);
       Block* tail = get_block_for_node(tail_n);
@@ -1533,23 +1563,23 @@
 
   // Create a member list for each loop consisting
   // of both blocks and (immediate child) loops.
-  for (uint i = 0; i < _num_blocks; i++) {
-    Block *b = _blocks[i];
-    CFGLoop* lp = b->_loop;
+  for (uint i = 0; i < number_of_blocks(); i++) {
+    Block* block = get_block(i);
+    CFGLoop* lp = block->_loop;
     if (lp == NULL) {
       // Not assigned to a loop. Add it to the method's pseudo loop.
-      b->_loop = root_loop;
+      block->_loop = root_loop;
       lp = root_loop;
     }
-    if (lp == root_loop || b != lp->head()) { // loop heads are already members
-      lp->add_member(b);
+    if (lp == root_loop || block != lp->head()) { // loop heads are already members
+      lp->add_member(block);
     }
     if (lp != root_loop) {
       if (lp->parent() == NULL) {
         // Not a nested loop. Make it a child of the method's pseudo loop.
         root_loop->add_nested_loop(lp);
       }
-      if (b == lp->head()) {
+      if (block == lp->head()) {
         // Add nested loop to member list of parent loop.
         lp->parent()->add_member(lp);
       }
--- a/src/share/vm/opto/idealGraphPrinter.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/opto/idealGraphPrinter.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -416,7 +416,7 @@
     if (C->cfg() != NULL) {
       Block* block = C->cfg()->get_block_for_node(node);
       if (block == NULL) {
-        print_prop("block", C->cfg()->_blocks[0]->_pre_order);
+        print_prop("block", C->cfg()->get_block(0)->_pre_order);
       } else {
         print_prop("block", block->_pre_order);
       }
@@ -637,10 +637,10 @@
   if (C->cfg() != NULL) {
     // once we have a CFG there are some nodes that aren't really
     // reachable but are in the CFG so add them here.
-    for (uint i = 0; i < C->cfg()->_blocks.size(); i++) {
-      Block *b = C->cfg()->_blocks[i];
-      for (uint s = 0; s < b->_nodes.size(); s++) {
-        nodeStack.push(b->_nodes[s]);
+    for (uint i = 0; i < C->cfg()->number_of_blocks(); i++) {
+      Block* block = C->cfg()->get_block(i);
+      for (uint s = 0; s < block->_nodes.size(); s++) {
+        nodeStack.push(block->_nodes[s]);
       }
     }
   }
@@ -698,24 +698,24 @@
   tail(EDGES_ELEMENT);
   if (C->cfg() != NULL) {
     head(CONTROL_FLOW_ELEMENT);
-    for (uint i = 0; i < C->cfg()->_blocks.size(); i++) {
-      Block *b = C->cfg()->_blocks[i];
+    for (uint i = 0; i < C->cfg()->number_of_blocks(); i++) {
+      Block* block = C->cfg()->get_block(i);
       begin_head(BLOCK_ELEMENT);
-      print_attr(BLOCK_NAME_PROPERTY, b->_pre_order);
+      print_attr(BLOCK_NAME_PROPERTY, block->_pre_order);
       end_head();
 
       head(SUCCESSORS_ELEMENT);
-      for (uint s = 0; s < b->_num_succs; s++) {
+      for (uint s = 0; s < block->_num_succs; s++) {
         begin_elem(SUCCESSOR_ELEMENT);
-        print_attr(BLOCK_NAME_PROPERTY, b->_succs[s]->_pre_order);
+        print_attr(BLOCK_NAME_PROPERTY, block->_succs[s]->_pre_order);
         end_elem();
       }
       tail(SUCCESSORS_ELEMENT);
 
       head(NODES_ELEMENT);
-      for (uint s = 0; s < b->_nodes.size(); s++) {
+      for (uint s = 0; s < block->_nodes.size(); s++) {
         begin_elem(NODE_ELEMENT);
-        print_attr(NODE_ID_PROPERTY, get_node_id(b->_nodes[s]));
+        print_attr(NODE_ID_PROPERTY, get_node_id(block->_nodes[s]));
         end_elem();
       }
       tail(NODES_ELEMENT);
--- a/src/share/vm/opto/ifg.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/opto/ifg.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -37,12 +37,9 @@
 #include "opto/memnode.hpp"
 #include "opto/opcodes.hpp"
 
-//=============================================================================
-//------------------------------IFG--------------------------------------------
 PhaseIFG::PhaseIFG( Arena *arena ) : Phase(Interference_Graph), _arena(arena) {
 }
 
-//------------------------------init-------------------------------------------
 void PhaseIFG::init( uint maxlrg ) {
   _maxlrg = maxlrg;
   _yanked = new (_arena) VectorSet(_arena);
@@ -59,7 +56,6 @@
   }
 }
 
-//------------------------------add--------------------------------------------
 // Add edge between vertices a & b.  These are sorted (triangular matrix),
 // then the smaller number is inserted in the larger numbered array.
 int PhaseIFG::add_edge( uint a, uint b ) {
@@ -71,7 +67,6 @@
   return _adjs[a].insert( b );
 }
 
-//------------------------------add_vector-------------------------------------
 // Add an edge between 'a' and everything in the vector.
 void PhaseIFG::add_vector( uint a, IndexSet *vec ) {
   // IFG is triangular, so do the inserts where 'a' < 'b'.
@@ -86,7 +81,6 @@
   }
 }
 
-//------------------------------test-------------------------------------------
 // Is there an edge between a and b?
 int PhaseIFG::test_edge( uint a, uint b ) const {
   // Sort a and b, so that a is larger
@@ -95,7 +89,6 @@
   return _adjs[a].member(b);
 }
 
-//------------------------------SquareUp---------------------------------------
 // Convert triangular matrix to square matrix
 void PhaseIFG::SquareUp() {
   assert( !_is_square, "only on triangular" );
@@ -111,7 +104,6 @@
   _is_square = true;
 }
 
-//------------------------------Compute_Effective_Degree-----------------------
 // Compute effective degree in bulk
 void PhaseIFG::Compute_Effective_Degree() {
   assert( _is_square, "only on square" );
@@ -120,7 +112,6 @@
     lrgs(i).set_degree(effective_degree(i));
 }
 
-//------------------------------test_edge_sq-----------------------------------
 int PhaseIFG::test_edge_sq( uint a, uint b ) const {
   assert( _is_square, "only on square" );
   // Swap, so that 'a' has the lesser count.  Then binary search is on
@@ -130,7 +121,6 @@
   return _adjs[a].member(b);
 }
 
-//------------------------------Union------------------------------------------
 // Union edges of B into A
 void PhaseIFG::Union( uint a, uint b ) {
   assert( _is_square, "only on square" );
@@ -146,7 +136,6 @@
   }
 }
 
-//------------------------------remove_node------------------------------------
 // Yank a Node and all connected edges from the IFG.  Return a
 // list of neighbors (edges) yanked.
 IndexSet *PhaseIFG::remove_node( uint a ) {
@@ -165,7 +154,6 @@
   return neighbors(a);
 }
 
-//------------------------------re_insert--------------------------------------
 // Re-insert a yanked Node.
 void PhaseIFG::re_insert( uint a ) {
   assert( _is_square, "only on square" );
@@ -180,7 +168,6 @@
   }
 }
 
-//------------------------------compute_degree---------------------------------
 // Compute the degree between 2 live ranges.  If both live ranges are
 // aligned-adjacent powers-of-2 then we use the MAX size.  If either is
 // mis-aligned (or for Fat-Projections, not-adjacent) then we have to
@@ -196,7 +183,6 @@
   return tmp;
 }
 
-//------------------------------effective_degree-------------------------------
 // Compute effective degree for this live range.  If both live ranges are
 // aligned-adjacent powers-of-2 then we use the MAX size.  If either is
 // mis-aligned (or for Fat-Projections, not-adjacent) then we have to
@@ -221,7 +207,6 @@
 
 
 #ifndef PRODUCT
-//------------------------------dump-------------------------------------------
 void PhaseIFG::dump() const {
   tty->print_cr("-- Interference Graph --%s--",
                 _is_square ? "square" : "triangular" );
@@ -260,7 +245,6 @@
   tty->print("\n");
 }
 
-//------------------------------stats------------------------------------------
 void PhaseIFG::stats() const {
   ResourceMark rm;
   int *h_cnt = NEW_RESOURCE_ARRAY(int,_maxlrg*2);
@@ -276,7 +260,6 @@
   tty->print_cr("");
 }
 
-//------------------------------verify-----------------------------------------
 void PhaseIFG::verify( const PhaseChaitin *pc ) const {
   // IFG is square, sorted and no need for Find
   for( uint i = 0; i < _maxlrg; i++ ) {
@@ -298,7 +281,6 @@
 }
 #endif
 
-//------------------------------interfere_with_live----------------------------
 // Interfere this register with everything currently live.  Use the RegMasks
 // to trim the set of possible interferences. Return a count of register-only
 // interferences as an estimate of register pressure.
@@ -315,7 +297,6 @@
       _ifg->add_edge( r, l );
 }
 
-//------------------------------build_ifg_virtual------------------------------
 // Actually build the interference graph.  Uses virtual registers only, no
 // physical register masks.  This allows me to be very aggressive when
 // coalescing copies.  Some of this aggressiveness will have to be undone
@@ -325,9 +306,9 @@
 void PhaseChaitin::build_ifg_virtual( ) {
 
   // For all blocks (in any order) do...
-  for( uint i=0; i<_cfg._num_blocks; i++ ) {
-    Block *b = _cfg._blocks[i];
-    IndexSet *liveout = _live->live(b);
+  for (uint i = 0; i < _cfg.number_of_blocks(); i++) {
+    Block* block = _cfg.get_block(i);
+    IndexSet* liveout = _live->live(block);
 
     // The IFG is built by a single reverse pass over each basic block.
     // Starting with the known live-out set, we remove things that get
@@ -337,8 +318,8 @@
     // The defined value interferes with everything currently live.  The
     // value is then removed from the live-ness set and it's inputs are
     // added to the live-ness set.
-    for( uint j = b->end_idx() + 1; j > 1; j-- ) {
-      Node *n = b->_nodes[j-1];
+    for (uint j = block->end_idx() + 1; j > 1; j--) {
+      Node* n = block->_nodes[j - 1];
 
       // Get value being defined
       uint r = _lrg_map.live_range_id(n);
@@ -408,7 +389,6 @@
   } // End of forall blocks
 }
 
-//------------------------------count_int_pressure-----------------------------
 uint PhaseChaitin::count_int_pressure( IndexSet *liveout ) {
   IndexSetIterator elements(liveout);
   uint lidx;
@@ -424,7 +404,6 @@
   return cnt;
 }
 
-//------------------------------count_float_pressure---------------------------
 uint PhaseChaitin::count_float_pressure( IndexSet *liveout ) {
   IndexSetIterator elements(liveout);
   uint lidx;
@@ -438,7 +417,6 @@
   return cnt;
 }
 
-//------------------------------lower_pressure---------------------------------
 // Adjust register pressure down by 1.  Capture last hi-to-low transition,
 static void lower_pressure( LRG *lrg, uint where, Block *b, uint *pressure, uint *hrp_index ) {
   if (lrg->mask().is_UP() && lrg->mask_size()) {
@@ -460,40 +438,41 @@
   }
 }
 
-//------------------------------build_ifg_physical-----------------------------
 // Build the interference graph using physical registers when available.
 // That is, if 2 live ranges are simultaneously alive but in their acceptable
 // register sets do not overlap, then they do not interfere.
 uint PhaseChaitin::build_ifg_physical( ResourceArea *a ) {
   NOT_PRODUCT( Compile::TracePhase t3("buildIFG", &_t_buildIFGphysical, TimeCompiler); )
 
-  uint spill_reg = LRG::SPILL_REG;
   uint must_spill = 0;
 
   // For all blocks (in any order) do...
-  for( uint i = 0; i < _cfg._num_blocks; i++ ) {
-    Block *b = _cfg._blocks[i];
+  for (uint i = 0; i < _cfg.number_of_blocks(); i++) {
+    Block* block = _cfg.get_block(i);
     // Clone (rather than smash in place) the liveout info, so it is alive
     // for the "collect_gc_info" phase later.
-    IndexSet liveout(_live->live(b));
-    uint last_inst = b->end_idx();
+    IndexSet liveout(_live->live(block));
+    uint last_inst = block->end_idx();
     // Compute first nonphi node index
     uint first_inst;
-    for( first_inst = 1; first_inst < last_inst; first_inst++ )
-      if( !b->_nodes[first_inst]->is_Phi() )
+    for (first_inst = 1; first_inst < last_inst; first_inst++) {
+      if (!block->_nodes[first_inst]->is_Phi()) {
         break;
+      }
+    }
 
     // Spills could be inserted before CreateEx node which should be
     // first instruction in block after Phis. Move CreateEx up.
-    for( uint insidx = first_inst; insidx < last_inst; insidx++ ) {
-      Node *ex = b->_nodes[insidx];
-      if( ex->is_SpillCopy() ) continue;
-      if( insidx > first_inst && ex->is_Mach() &&
-          ex->as_Mach()->ideal_Opcode() == Op_CreateEx ) {
+    for (uint insidx = first_inst; insidx < last_inst; insidx++) {
+      Node *ex = block->_nodes[insidx];
+      if (ex->is_SpillCopy()) {
+        continue;
+      }
+      if (insidx > first_inst && ex->is_Mach() && ex->as_Mach()->ideal_Opcode() == Op_CreateEx) {
         // If the CreateEx isn't above all the MachSpillCopies
         // then move it to the top.
-        b->_nodes.remove(insidx);
-        b->_nodes.insert(first_inst, ex);
+        block->_nodes.remove(insidx);
+        block->_nodes.insert(first_inst, ex);
       }
       // Stop once a CreateEx or any other node is found
       break;
@@ -503,12 +482,12 @@
     uint pressure[2], hrp_index[2];
     pressure[0] = pressure[1] = 0;
     hrp_index[0] = hrp_index[1] = last_inst+1;
-    b->_reg_pressure = b->_freg_pressure = 0;
+    block->_reg_pressure = block->_freg_pressure = 0;
     // Liveout things are presumed live for the whole block.  We accumulate
     // 'area' accordingly.  If they get killed in the block, we'll subtract
     // the unused part of the block from the area.
     int inst_count = last_inst - first_inst;
-    double cost = (inst_count <= 0) ? 0.0 : b->_freq * double(inst_count);
+    double cost = (inst_count <= 0) ? 0.0 : block->_freq * double(inst_count);
     assert(!(cost < 0.0), "negative spill cost" );
     IndexSetIterator elements(&liveout);
     uint lidx;
@@ -519,13 +498,15 @@
       if (lrg.mask().is_UP() && lrg.mask_size()) {
         if (lrg._is_float || lrg._is_vector) {   // Count float pressure
           pressure[1] += lrg.reg_pressure();
-          if( pressure[1] > b->_freg_pressure )
-            b->_freg_pressure = pressure[1];
+          if (pressure[1] > block->_freg_pressure) {
+            block->_freg_pressure = pressure[1];
+          }
           // Count int pressure, but do not count the SP, flags
-        } else if( lrgs(lidx).mask().overlap(*Matcher::idealreg2regmask[Op_RegI]) ) {
+        } else if(lrgs(lidx).mask().overlap(*Matcher::idealreg2regmask[Op_RegI])) {
           pressure[0] += lrg.reg_pressure();
-          if( pressure[0] > b->_reg_pressure )
-            b->_reg_pressure = pressure[0];
+          if (pressure[0] > block->_reg_pressure) {
+            block->_reg_pressure = pressure[0];
+          }
         }
       }
     }
@@ -541,8 +522,8 @@
     // value is then removed from the live-ness set and it's inputs are added
     // to the live-ness set.
     uint j;
-    for( j = last_inst + 1; j > 1; j-- ) {
-      Node *n = b->_nodes[j - 1];
+    for (j = last_inst + 1; j > 1; j--) {
+      Node* n = block->_nodes[j - 1];
 
       // Get value being defined
       uint r = _lrg_map.live_range_id(n);
@@ -551,7 +532,7 @@
       if(r) {
         // A DEF normally costs block frequency; rematerialized values are
         // removed from the DEF sight, so LOWER costs here.
-        lrgs(r)._cost += n->rematerialize() ? 0 : b->_freq;
+        lrgs(r)._cost += n->rematerialize() ? 0 : block->_freq;
 
         // If it is not live, then this instruction is dead.  Probably caused
         // by spilling and rematerialization.  Who cares why, yank this baby.
@@ -560,7 +541,7 @@
           if( !n->is_Proj() ||
               // Could also be a flags-projection of a dead ADD or such.
               (_lrg_map.live_range_id(def) && !liveout.member(_lrg_map.live_range_id(def)))) {
-            b->_nodes.remove(j - 1);
+            block->_nodes.remove(j - 1);
             if (lrgs(r)._def == n) {
               lrgs(r)._def = 0;
             }
@@ -580,21 +561,21 @@
             RegMask itmp = lrgs(r).mask();
             itmp.AND(*Matcher::idealreg2regmask[Op_RegI]);
             int iregs = itmp.Size();
-            if( pressure[0]+iregs > b->_reg_pressure )
-              b->_reg_pressure = pressure[0]+iregs;
-            if( pressure[0]       <= (uint)INTPRESSURE &&
-                pressure[0]+iregs >  (uint)INTPRESSURE ) {
-              hrp_index[0] = j-1;
+            if (pressure[0]+iregs > block->_reg_pressure) {
+              block->_reg_pressure = pressure[0] + iregs;
+            }
+            if (pressure[0] <= (uint)INTPRESSURE && pressure[0] + iregs > (uint)INTPRESSURE) {
+              hrp_index[0] = j - 1;
             }
             // Count the float-only registers
             RegMask ftmp = lrgs(r).mask();
             ftmp.AND(*Matcher::idealreg2regmask[Op_RegD]);
             int fregs = ftmp.Size();
-            if( pressure[1]+fregs > b->_freg_pressure )
-              b->_freg_pressure = pressure[1]+fregs;
-            if( pressure[1]       <= (uint)FLOATPRESSURE &&
-                pressure[1]+fregs >  (uint)FLOATPRESSURE ) {
-              hrp_index[1] = j-1;
+            if (pressure[1] + fregs > block->_freg_pressure) {
+              block->_freg_pressure = pressure[1] + fregs;
+            }
+            if(pressure[1] <= (uint)FLOATPRESSURE && pressure[1]+fregs > (uint)FLOATPRESSURE) {
+              hrp_index[1] = j - 1;
             }
           }
 
@@ -607,7 +588,7 @@
           if( n->is_SpillCopy()
               && lrgs(r).is_singledef()        // MultiDef live range can still split
               && n->outcnt() == 1              // and use must be in this block
-              && _cfg.get_block_for_node(n->unique_out()) == b ) {
+              && _cfg.get_block_for_node(n->unique_out()) == block) {
             // All single-use MachSpillCopy(s) that immediately precede their
             // use must color early.  If a longer live range steals their
             // color, the spill copy will split and may push another spill copy
@@ -617,14 +598,16 @@
             //
 
             Node *single_use = n->unique_out();
-            assert( b->find_node(single_use) >= j, "Use must be later in block");
+            assert(block->find_node(single_use) >= j, "Use must be later in block");
             // Use can be earlier in block if it is a Phi, but then I should be a MultiDef
 
             // Find first non SpillCopy 'm' that follows the current instruction
             // (j - 1) is index for current instruction 'n'
             Node *m = n;
-            for( uint i = j; i <= last_inst && m->is_SpillCopy(); ++i ) { m = b->_nodes[i]; }
-            if( m == single_use ) {
+            for (uint i = j; i <= last_inst && m->is_SpillCopy(); ++i) {
+              m = block->_nodes[i];
+            }
+            if (m == single_use) {
               lrgs(r)._area = 0.0;
             }
           }
@@ -633,7 +616,7 @@
           if( liveout.remove(r) ) {
             // Adjust register pressure.
             // Capture last hi-to-lo pressure transition
-            lower_pressure( &lrgs(r), j-1, b, pressure, hrp_index );
+            lower_pressure(&lrgs(r), j - 1, block, pressure, hrp_index);
             assert( pressure[0] == count_int_pressure  (&liveout), "" );
             assert( pressure[1] == count_float_pressure(&liveout), "" );
           }
@@ -646,7 +629,7 @@
             if (liveout.remove(x)) {
               lrgs(x)._area -= cost;
               // Adjust register pressure.
-              lower_pressure(&lrgs(x), j-1, b, pressure, hrp_index);
+              lower_pressure(&lrgs(x), j - 1, block, pressure, hrp_index);
               assert( pressure[0] == count_int_pressure  (&liveout), "" );
               assert( pressure[1] == count_float_pressure(&liveout), "" );
             }
@@ -718,7 +701,7 @@
 
       // Area remaining in the block
       inst_count--;
-      cost = (inst_count <= 0) ? 0.0 : b->_freq * double(inst_count);
+      cost = (inst_count <= 0) ? 0.0 : block->_freq * double(inst_count);
 
       // Make all inputs live
       if( !n->is_Phi() ) {      // Phi function uses come from prior block
@@ -743,7 +726,7 @@
           if (k < debug_start) {
             // A USE costs twice block frequency (once for the Load, once
             // for a Load-delay).  Rematerialized uses only cost once.
-            lrg._cost += (def->rematerialize() ? b->_freq : (b->_freq + b->_freq));
+            lrg._cost += (def->rematerialize() ? block->_freq : (block->_freq + block->_freq));
           }
           // It is live now
           if (liveout.insert(x)) {
@@ -753,12 +736,14 @@
             if (lrg.mask().is_UP() && lrg.mask_size()) {
               if (lrg._is_float || lrg._is_vector) {
                 pressure[1] += lrg.reg_pressure();
-                if( pressure[1] > b->_freg_pressure )
-                  b->_freg_pressure = pressure[1];
+                if (pressure[1] > block->_freg_pressure)  {
+                  block->_freg_pressure = pressure[1];
+                }
               } else if( lrg.mask().overlap(*Matcher::idealreg2regmask[Op_RegI]) ) {
                 pressure[0] += lrg.reg_pressure();
-                if( pressure[0] > b->_reg_pressure )
-                  b->_reg_pressure = pressure[0];
+                if (pressure[0] > block->_reg_pressure) {
+                  block->_reg_pressure = pressure[0];
+                }
               }
             }
             assert( pressure[0] == count_int_pressure  (&liveout), "" );
@@ -772,44 +757,47 @@
     // If we run off the top of the block with high pressure and
     // never see a hi-to-low pressure transition, just record that
     // the whole block is high pressure.
-    if( pressure[0] > (uint)INTPRESSURE   ) {
+    if (pressure[0] > (uint)INTPRESSURE) {
       hrp_index[0] = 0;
-      if( pressure[0] > b->_reg_pressure )
-        b->_reg_pressure = pressure[0];
+      if (pressure[0] > block->_reg_pressure) {
+        block->_reg_pressure = pressure[0];
+      }
     }
-    if( pressure[1] > (uint)FLOATPRESSURE ) {
+    if (pressure[1] > (uint)FLOATPRESSURE) {
       hrp_index[1] = 0;
-      if( pressure[1] > b->_freg_pressure )
-        b->_freg_pressure = pressure[1];
+      if (pressure[1] > block->_freg_pressure) {
+        block->_freg_pressure = pressure[1];
+      }
     }
 
     // Compute high pressure indice; avoid landing in the middle of projnodes
     j = hrp_index[0];
-    if( j < b->_nodes.size() && j < b->end_idx()+1 ) {
-      Node *cur = b->_nodes[j];
-      while( cur->is_Proj() || (cur->is_MachNullCheck()) || cur->is_Catch() ) {
+    if (j < block->_nodes.size() && j < block->end_idx() + 1) {
+      Node* cur = block->_nodes[j];
+      while (cur->is_Proj() || (cur->is_MachNullCheck()) || cur->is_Catch()) {
         j--;
-        cur = b->_nodes[j];
+        cur = block->_nodes[j];
       }
     }
-    b->_ihrp_index = j;
+    block->_ihrp_index = j;
     j = hrp_index[1];
-    if( j < b->_nodes.size() && j < b->end_idx()+1 ) {
-      Node *cur = b->_nodes[j];
-      while( cur->is_Proj() || (cur->is_MachNullCheck()) || cur->is_Catch() ) {
+    if (j < block->_nodes.size() && j < block->end_idx() + 1) {
+      Node* cur = block->_nodes[j];
+      while (cur->is_Proj() || (cur->is_MachNullCheck()) || cur->is_Catch()) {
         j--;
-        cur = b->_nodes[j];
+        cur = block->_nodes[j];
       }
     }
-    b->_fhrp_index = j;
+    block->_fhrp_index = j;
 
 #ifndef PRODUCT
     // Gather Register Pressure Statistics
     if( PrintOptoStatistics ) {
-      if( b->_reg_pressure > (uint)INTPRESSURE || b->_freg_pressure > (uint)FLOATPRESSURE )
+      if (block->_reg_pressure > (uint)INTPRESSURE || block->_freg_pressure > (uint)FLOATPRESSURE) {
         _high_pressure++;
-      else
+      } else {
         _low_pressure++;
+      }
     }
 #endif
   } // End of for all blocks
--- a/src/share/vm/opto/lcm.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/opto/lcm.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -501,7 +501,7 @@
       n_choice = 1;
     }
 
-    uint n_latency = cfg->_node_latency->at_grow(n->_idx);
+    uint n_latency = cfg->get_latency_for_node(n);
     uint n_score   = n->req();   // Many inputs get high score to break ties
 
     // Keep best latency found
@@ -797,7 +797,7 @@
         Node     *n = _nodes[j];
         int     idx = n->_idx;
         tty->print("#   ready cnt:%3d  ", ready_cnt.at(idx));
-        tty->print("latency:%3d  ", cfg->_node_latency->at_grow(idx));
+        tty->print("latency:%3d  ", cfg->get_latency_for_node(n));
         tty->print("%4d: %s\n", idx, n->Name());
       }
     }
@@ -825,7 +825,7 @@
 #ifndef PRODUCT
     if (cfg->trace_opto_pipelining()) {
       tty->print("#    select %d: %s", n->_idx, n->Name());
-      tty->print(", latency:%d", cfg->_node_latency->at_grow(n->_idx));
+      tty->print(", latency:%d", cfg->get_latency_for_node(n));
       n->dump();
       if (Verbose) {
         tty->print("#   ready list:");
--- a/src/share/vm/opto/library_call.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/opto/library_call.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -213,6 +213,7 @@
   void insert_pre_barrier(Node* base_oop, Node* offset, Node* pre_val, bool need_mem_bar);
   bool inline_unsafe_access(bool is_native_ptr, bool is_store, BasicType type, bool is_volatile);
   bool inline_unsafe_prefetch(bool is_native_ptr, bool is_store, bool is_static);
+  static bool klass_needs_init_guard(Node* kls);
   bool inline_unsafe_allocate();
   bool inline_unsafe_copyMemory();
   bool inline_native_currentThread();
@@ -2892,8 +2893,21 @@
   }
 }
 
+bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
+  if (!kls->is_Con()) {
+    return true;
+  }
+  const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
+  if (klsptr == NULL) {
+    return true;
+  }
+  ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
+  // don't need a guard for a klass that is already initialized
+  return !ik->is_initialized();
+}
+
 //----------------------------inline_unsafe_allocate---------------------------
-// public native Object sun.mics.Unsafe.allocateInstance(Class<?> cls);
+// public native Object sun.misc.Unsafe.allocateInstance(Class<?> cls);
 bool LibraryCallKit::inline_unsafe_allocate() {
   if (callee()->is_static())  return false;  // caller must have the capability!
 
@@ -2905,16 +2919,19 @@
   kls = null_check(kls);
   if (stopped())  return true;  // argument was like int.class
 
-  // Note:  The argument might still be an illegal value like
-  // Serializable.class or Object[].class.   The runtime will handle it.
-  // But we must make an explicit check for initialization.
-  Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
-  // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
-  // can generate code to load it as unsigned byte.
-  Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN);
-  Node* bits = intcon(InstanceKlass::fully_initialized);
-  Node* test = _gvn.transform(new (C) SubINode(inst, bits));
-  // The 'test' is non-zero if we need to take a slow path.
+  Node* test = NULL;
+  if (LibraryCallKit::klass_needs_init_guard(kls)) {
+    // Note:  The argument might still be an illegal value like
+    // Serializable.class or Object[].class.   The runtime will handle it.
+    // But we must make an explicit check for initialization.
+    Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
+    // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
+    // can generate code to load it as unsigned byte.
+    Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN);
+    Node* bits = intcon(InstanceKlass::fully_initialized);
+    test = _gvn.transform(new (C) SubINode(inst, bits));
+    // The 'test' is non-zero if we need to take a slow path.
+  }
 
   Node* obj = new_instance(kls, test);
   set_result(obj);
--- a/src/share/vm/opto/live.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/opto/live.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -30,9 +30,6 @@
 #include "opto/machnode.hpp"
 
 
-
-//=============================================================================
-//------------------------------PhaseLive--------------------------------------
 // Compute live-in/live-out.  We use a totally incremental algorithm.  The LIVE
 // problem is monotonic.  The steady-state solution looks like this: pull a
 // block from the worklist.  It has a set of delta's - values which are newly
@@ -53,9 +50,9 @@
 
   // Init the sparse live arrays.  This data is live on exit from here!
   // The _live info is the live-out info.
-  _live = (IndexSet*)_arena->Amalloc(sizeof(IndexSet)*_cfg._num_blocks);
+  _live = (IndexSet*)_arena->Amalloc(sizeof(IndexSet) * _cfg.number_of_blocks());
   uint i;
-  for( i=0; i<_cfg._num_blocks; i++ ) {
+  for (i = 0; i < _cfg.number_of_blocks(); i++) {
     _live[i].initialize(_maxlrg);
   }
 
@@ -65,14 +62,14 @@
   // Does the memory used by _defs and _deltas get reclaimed?  Does it matter?  TT
 
   // Array of values defined locally in blocks
-  _defs = NEW_RESOURCE_ARRAY(IndexSet,_cfg._num_blocks);
-  for( i=0; i<_cfg._num_blocks; i++ ) {
+  _defs = NEW_RESOURCE_ARRAY(IndexSet,_cfg.number_of_blocks());
+  for (i = 0; i < _cfg.number_of_blocks(); i++) {
     _defs[i].initialize(_maxlrg);
   }
 
   // Array of delta-set pointers, indexed by block pre_order-1.
-  _deltas = NEW_RESOURCE_ARRAY(IndexSet*,_cfg._num_blocks);
-  memset( _deltas, 0, sizeof(IndexSet*)* _cfg._num_blocks);
+  _deltas = NEW_RESOURCE_ARRAY(IndexSet*,_cfg.number_of_blocks());
+  memset( _deltas, 0, sizeof(IndexSet*)* _cfg.number_of_blocks());
 
   _free_IndexSet = NULL;
 
@@ -80,31 +77,32 @@
   VectorSet first_pass(Thread::current()->resource_area());
 
   // Outer loop: must compute local live-in sets and push into predecessors.
-  uint iters = _cfg._num_blocks;        // stat counters
-  for( uint j=_cfg._num_blocks; j>0; j-- ) {
-    Block *b = _cfg._blocks[j-1];
+  for (uint j = _cfg.number_of_blocks(); j > 0; j--) {
+    Block* block = _cfg.get_block(j - 1);
 
     // Compute the local live-in set.  Start with any new live-out bits.
-    IndexSet *use = getset( b );
-    IndexSet *def = &_defs[b->_pre_order-1];
+    IndexSet* use = getset(block);
+    IndexSet* def = &_defs[block->_pre_order-1];
     DEBUG_ONLY(IndexSet *def_outside = getfreeset();)
     uint i;
-    for( i=b->_nodes.size(); i>1; i-- ) {
-      Node *n = b->_nodes[i-1];
-      if( n->is_Phi() ) break;
+    for (i = block->_nodes.size(); i > 1; i--) {
+      Node* n = block->_nodes[i-1];
+      if (n->is_Phi()) {
+        break;
+      }
 
       uint r = _names[n->_idx];
       assert(!def_outside->member(r), "Use of external LRG overlaps the same LRG defined in this block");
       def->insert( r );
       use->remove( r );
       uint cnt = n->req();
-      for( uint k=1; k<cnt; k++ ) {
+      for (uint k = 1; k < cnt; k++) {
         Node *nk = n->in(k);
         uint nkidx = nk->_idx;
-        if (_cfg.get_block_for_node(nk) != b) {
+        if (_cfg.get_block_for_node(nk) != block) {
           uint u = _names[nkidx];
-          use->insert( u );
-          DEBUG_ONLY(def_outside->insert( u );)
+          use->insert(u);
+          DEBUG_ONLY(def_outside->insert(u);)
         }
       }
     }
@@ -113,41 +111,38 @@
     _free_IndexSet = def_outside;     // Drop onto free list
 #endif
     // Remove anything defined by Phis and the block start instruction
-    for( uint k=i; k>0; k-- ) {
-      uint r = _names[b->_nodes[k-1]->_idx];
-      def->insert( r );
-      use->remove( r );
+    for (uint k = i; k > 0; k--) {
+      uint r = _names[block->_nodes[k - 1]->_idx];
+      def->insert(r);
+      use->remove(r);
     }
 
     // Push these live-in things to predecessors
-    for( uint l=1; l<b->num_preds(); l++ ) {
-      Block *p = _cfg.get_block_for_node(b->pred(l));
-      add_liveout( p, use, first_pass );
+    for (uint l = 1; l < block->num_preds(); l++) {
+      Block* p = _cfg.get_block_for_node(block->pred(l));
+      add_liveout(p, use, first_pass);
 
       // PhiNode uses go in the live-out set of prior blocks.
-      for( uint k=i; k>0; k-- )
-        add_liveout( p, _names[b->_nodes[k-1]->in(l)->_idx], first_pass );
+      for (uint k = i; k > 0; k--) {
+        add_liveout(p, _names[block->_nodes[k-1]->in(l)->_idx], first_pass);
+      }
     }
-    freeset( b );
-    first_pass.set(b->_pre_order);
+    freeset(block);
+    first_pass.set(block->_pre_order);
 
     // Inner loop: blocks that picked up new live-out values to be propagated
-    while( _worklist->size() ) {
-        // !!!!!
-// #ifdef ASSERT
-      iters++;
-// #endif
-      Block *b = _worklist->pop();
-      IndexSet *delta = getset(b);
+    while (_worklist->size()) {
+      Block* block = _worklist->pop();
+      IndexSet *delta = getset(block);
       assert( delta->count(), "missing delta set" );
 
       // Add new-live-in to predecessors live-out sets
-      for (uint l = 1; l < b->num_preds(); l++) {
-        Block* block = _cfg.get_block_for_node(b->pred(l));
-        add_liveout(block, delta, first_pass);
+      for (uint l = 1; l < block->num_preds(); l++) {
+        Block* predecessor = _cfg.get_block_for_node(block->pred(l));
+        add_liveout(predecessor, delta, first_pass);
       }
 
-      freeset(b);
+      freeset(block);
     } // End of while-worklist-not-empty
 
   } // End of for-all-blocks-outer-loop
@@ -155,7 +150,7 @@
   // We explicitly clear all of the IndexSets which we are about to release.
   // This allows us to recycle their internal memory into IndexSet's free list.
 
-  for( i=0; i<_cfg._num_blocks; i++ ) {
+  for (i = 0; i < _cfg.number_of_blocks(); i++) {
     _defs[i].clear();
     if (_deltas[i]) {
       // Is this always true?
@@ -171,13 +166,11 @@
 
 }
 
-//------------------------------stats------------------------------------------
 #ifndef PRODUCT
 void PhaseLive::stats(uint iters) const {
 }
 #endif
 
-//------------------------------getset-----------------------------------------
 // Get an IndexSet for a block.  Return existing one, if any.  Make a new
 // empty one if a prior one does not exist.
 IndexSet *PhaseLive::getset( Block *p ) {
@@ -188,7 +181,6 @@
   return delta;                 // Return set of new live-out items
 }
 
-//------------------------------getfreeset-------------------------------------
 // Pull from free list, or allocate.  Internal allocation on the returned set
 // is always from thread local storage.
 IndexSet *PhaseLive::getfreeset( ) {
@@ -207,7 +199,6 @@
   return f;
 }
 
-//------------------------------freeset----------------------------------------
 // Free an IndexSet from a block.
 void PhaseLive::freeset( const Block *p ) {
   IndexSet *f = _deltas[p->_pre_order-1];
@@ -216,7 +207,6 @@
   _deltas[p->_pre_order-1] = NULL;
 }
 
-//------------------------------add_liveout------------------------------------
 // Add a live-out value to a given blocks live-out set.  If it is new, then
 // also add it to the delta set and stick the block on the worklist.
 void PhaseLive::add_liveout( Block *p, uint r, VectorSet &first_pass ) {
@@ -233,8 +223,6 @@
   }
 }
 
-
-//------------------------------add_liveout------------------------------------
 // Add a vector of live-out values to a given blocks live-out set.
 void PhaseLive::add_liveout( Block *p, IndexSet *lo, VectorSet &first_pass ) {
   IndexSet *live = &_live[p->_pre_order-1];
@@ -262,7 +250,6 @@
 }
 
 #ifndef PRODUCT
-//------------------------------dump-------------------------------------------
 // Dump the live-out set for a block
 void PhaseLive::dump( const Block *b ) const {
   tty->print("Block %d: ",b->_pre_order);
@@ -275,18 +262,19 @@
   tty->print("\n");
 }
 
-//------------------------------verify_base_ptrs-------------------------------
 // Verify that base pointers and derived pointers are still sane.
 void PhaseChaitin::verify_base_ptrs( ResourceArea *a ) const {
 #ifdef ASSERT
   Unique_Node_List worklist(a);
-  for( uint i = 0; i < _cfg._num_blocks; i++ ) {
-    Block *b = _cfg._blocks[i];
-    for( uint j = b->end_idx() + 1; j > 1; j-- ) {
-      Node *n = b->_nodes[j-1];
-      if( n->is_Phi() ) break;
+  for (uint i = 0; i < _cfg.number_of_blocks(); i++) {
+    Block* block = _cfg.get_block(i);
+    for (uint j = block->end_idx() + 1; j > 1; j--) {
+      Node* n = block->_nodes[j-1];
+      if (n->is_Phi()) {
+        break;
+      }
       // Found a safepoint?
-      if( n->is_MachSafePoint() ) {
+      if (n->is_MachSafePoint()) {
         MachSafePointNode *sfpt = n->as_MachSafePoint();
         JVMState* jvms = sfpt->jvms();
         if (jvms != NULL) {
@@ -358,7 +346,6 @@
 #endif
 }
 
-//------------------------------verify-------------------------------------
 // Verify that graphs and base pointers are still sane.
 void PhaseChaitin::verify( ResourceArea *a, bool verify_ifg ) const {
 #ifdef ASSERT
--- a/src/share/vm/opto/matcher.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/opto/matcher.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -67,8 +67,8 @@
 const uint Matcher::_end_rematerialize   = _END_REMATERIALIZE;
 
 //---------------------------Matcher-------------------------------------------
-Matcher::Matcher( Node_List &proj_list ) :
-  PhaseTransform( Phase::Ins_Select ),
+Matcher::Matcher()
+: PhaseTransform( Phase::Ins_Select ),
 #ifdef ASSERT
   _old2new_map(C->comp_arena()),
   _new2old_map(C->comp_arena()),
@@ -78,7 +78,7 @@
   _swallowed(swallowed),
   _begin_inst_chain_rule(_BEGIN_INST_CHAIN_RULE),
   _end_inst_chain_rule(_END_INST_CHAIN_RULE),
-  _must_clone(must_clone), _proj_list(proj_list),
+  _must_clone(must_clone),
   _register_save_policy(register_save_policy),
   _c_reg_save_policy(c_reg_save_policy),
   _register_save_type(register_save_type),
@@ -1304,8 +1304,9 @@
       for (int i = begin_out_arg_area; i < out_arg_limit_per_call; i++)
         proj->_rout.Insert(OptoReg::Name(i));
     }
-    if( proj->_rout.is_NotEmpty() )
-      _proj_list.push(proj);
+    if (proj->_rout.is_NotEmpty()) {
+      push_projection(proj);
+    }
   }
   // Transfer the safepoint information from the call to the mcall
   // Move the JVMState list
@@ -1685,14 +1686,15 @@
   }
 
   // If the _leaf is an AddP, insert the base edge
-  if( leaf->is_AddP() )
+  if (leaf->is_AddP()) {
     mach->ins_req(AddPNode::Base,leaf->in(AddPNode::Base));
+  }
 
-  uint num_proj = _proj_list.size();
+  uint number_of_projections_prior = number_of_projections();
 
   // Perform any 1-to-many expansions required
-  MachNode *ex = mach->Expand(s,_proj_list, mem);
-  if( ex != mach ) {
+  MachNode *ex = mach->Expand(s, _projection_list, mem);
+  if (ex != mach) {
     assert(ex->ideal_reg() == mach->ideal_reg(), "ideal types should match");
     if( ex->in(1)->is_Con() )
       ex->in(1)->set_req(0, C->root());
@@ -1713,7 +1715,7 @@
   // generated belatedly during spill code generation.
   if (_allocation_started) {
     guarantee(ex == mach, "no expand rules during spill generation");
-    guarantee(_proj_list.size() == num_proj, "no allocation during spill generation");
+    guarantee(number_of_projections_prior == number_of_projections(), "no allocation during spill generation");
   }
 
   if (leaf->is_Con() || leaf->is_DecodeNarrowPtr()) {
--- a/src/share/vm/opto/matcher.hpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/opto/matcher.hpp	Fri Aug 23 03:01:16 2013 -0700
@@ -88,7 +88,7 @@
 
   Node *transform( Node *dummy );
 
-  Node_List &_proj_list;        // For Machine nodes killing many values
+  Node_List _projection_list;        // For Machine nodes killing many values
 
   Node_Array _shared_nodes;
 
@@ -183,10 +183,30 @@
   void collect_null_checks( Node *proj, Node *orig_proj );
   void validate_null_checks( );
 
-  Matcher( Node_List &proj_list );
+  Matcher();
+
+  // Get a projection node at position pos
+  Node* get_projection(uint pos) {
+    return _projection_list[pos];
+  }
+
+  // Push a projection node onto the projection list
+  void push_projection(Node* node) {
+    _projection_list.push(node);
+  }
+
+  Node* pop_projection() {
+    return _projection_list.pop();
+  }
+
+  // Number of nodes in the projection list
+  uint number_of_projections() const {
+    return _projection_list.size();
+  }
 
   // Select instructions for entire method
-  void  match( );
+  void match();
+
   // Helper for match
   OptoReg::Name warp_incoming_stk_arg( VMReg reg );
 
--- a/src/share/vm/opto/output.cpp	Thu Aug 22 09:10:01 2013 -0700
+++ b/src/share/vm/opto/output.cpp	Fri Aug 23 03:01:16 2013 -0700
@@ -54,11 +54,10 @@
 extern int emit_exception_handler(CodeBuffer &cbuf);
 extern int emit_deopt_handler(CodeBuffer &cbuf);
 
-//------------------------------Output-----------------------------------------
 // Convert Nodes to instruction bits and pass off to the VM
 void Compile::Output() {
   // RootNode goes
-  assert( _cfg->_broot->_nodes.size() == 0, "" );
+  assert( _cfg->get_root_block()->_nodes.size() == 0, "" );
 
   // The number of new nodes (mostly MachNop) is proportional to
   // the number of java calls and inner loops which are aligned.
@@ -68,8 +67,8 @@
     return;
   }
   // Make sure I can find the Start Node
-  Block *entry = _cfg->_blocks[1];
-  Block *broot = _cfg->_broot;
+  Block *entry = _cfg->get_block(1);
+  Block *broot = _cfg->get_root_block();
 
   const StartNode *start = entry->_nodes[0]->as_Start();
 
@@ -109,40 +108,44 @@
   }
 
   // Insert epilogs before every return
-  for( uint i=0; i<_cfg->_num_blocks; i++ ) {
-    Block *b = _cfg->_blocks[i];
-    if( !b->is_connector() && b->non_connector_successor(0) == _cfg->_broot ) { // Found a program exit point?
-      Node *m = b->end();
-      if( m->is_Mach() && m->as_Mach()->ideal_Opcode() != Op_Halt ) {
-        MachEpilogNode *epilog = new (this) MachEpilogNode(m->as_Mach()->ideal_Opcode() == Op_Return);
-        b->add_inst( epilog );
-        _cfg->map_node_to_block(epilog, b);
+  for (uint i = 0; i < _cfg->number_of_blocks(); i++) {
+    Block* block = _cfg->get_block(i);
+    if (!block->is_connector() && block->non_connector_successor(0) == _cfg->get_root_block()) { // Found a program exit point?
+      Node* m = block->end();
+      if (m->is_Mach() && m->as_Mach()->ideal_Opcode() != Op_Halt) {
+        MachEpilogNode* epilog = new (this) MachEpilogNode(m->as_Mach()->ideal_Opcode() == Op_Return);
+        block->add_inst(epilog);
+        _cfg->map_node_to_block(epilog, block);
       }
     }
   }
 
 # ifdef ENABLE_ZAP_DEAD_LOCALS
-  if ( ZapDeadCompiledLocals )  Insert_zap_nodes();
+  if (ZapDeadCompiledLocals) {
+    Insert_zap_nodes();
+  }
 # endif
 
-  uint* blk_starts = NEW_RESOURCE_ARRAY(uint,_cfg->_num_blocks+1);
-  blk_starts[0]    = 0;
+  uint* blk_starts = NEW_RESOURCE_ARRAY(uint, _cfg->number_of_blocks() + 1);
+  blk_starts[0] = 0;
 
   // Initialize code buffer and process short branches.
   CodeBuffer* cb = init_buffer(blk_starts);
 
-  if (cb == NULL || failing())  return;
+  if (cb == NULL || failing()) {
+    return;
+  }
 
   ScheduleAndBundle();
 
 #ifndef PRODUCT
   if (trace_opto_output()) {
     tty->print("\n---- After ScheduleAndBundle ----\n");
-    for (uint i = 0; i < _cfg->_num_blocks; i++) {
+    for (uint i = 0; i < _cfg->number_of_blocks(); i++) {
       tty->print("\nBB#%03d:\n", i);
-      Block *bb = _cfg->_blocks[i];
-      for (uint j = 0; j < bb->_nodes.size(); j++) {
-        Node *n = bb->_nodes[j];
+      Block* block = _cfg->get_block(i);
+      for (uint j = 0; j < block->_nodes.size(); j++) {
+        Node* n = block->_nodes[j];
         OptoReg::Name reg = _regalloc->get_reg_first(n);
         tty->print(" %-6s ", reg >= 0 && reg < REG_COUNT ? Matcher::regName[reg] : "");
         n->dump();
@@ -151,11 +154,15 @@
   }
 #endif
 
-  if (failing())  return;
+  if (failing()) {
+    return;
+  }
 
   BuildOopMaps();
 
-  if (failing())  return;
+  if (failing())  {
+    return;
+  }
 
   fill_buffer(cb, blk_starts);
 }
@@ -217,8 +224,8 @@
     return; // no safepoints/oopmaps emitted for calls in stubs,so we don't care
 
   // Insert call to zap runtime stub before every node with an oop map
-  for( uint i=0; i<_cfg->_num_blocks; i++ ) {
-    Block *b = _cfg->_blocks[i];
+  for( uint i=0; i<_cfg->number_of_blocks(); i++ ) {
+    Block *b = _cfg->get_block(i);
     for ( uint j = 0;  j < b->_nodes.size();  ++j ) {
       Node *n = b->_nodes[j];
 
@@ -275,7 +282,6 @@
   return _matcher->match_sfpt(ideal_node);
 }
 
-//------------------------------is_node_getting_a_safepoint--------------------
 bool Compile::is_node_getting_a_safepoint( Node* n) {
   // This code duplicates the logic prior to the call of add_safepoint
   // below in this file.
@@ -285,7 +291,6 @@
 
 # endif // ENABLE_ZAP_DEAD_LOCALS
 
-//------------------------------compute_loop_first_inst_sizes------------------
 // Compute the size of first NumberOfLoopInstrToAlign instructions at the top
 // of a loop. When aligning a loop we need to provide enough instructions
 // in cpu's fetch buffer to feed decoders. The loop alignment could be
@@ -302,42 +307,39 @@
   // or alignment padding is larger then MaxLoopPad. By default, MaxLoopPad
   // is equal to OptoLoopAlignment-1 except on new Intel cpus, where it is
   // equal to 11 bytes which is the largest address NOP instruction.
-  if( MaxLoopPad < OptoLoopAlignment-1 ) {
-    uint last_block = _cfg->_num_blocks-1;
-    for( uint i=1; i <= last_block; i++ ) {
-      Block *b = _cfg->_blocks[i];
+  if (MaxLoopPad < OptoLoopAlignment - 1) {
+    uint last_block = _cfg->number_of_blocks() - 1;
+    for (uint i = 1; i <= last_block; i++) {
+      Block* block = _cfg->get_block(i);
       // Check the first loop's block which requires an alignment.
-      if( b->loop_alignment() > (uint)relocInfo::addr_unit() ) {
+      if (block->loop_alignment() > (uint)relocInfo::addr_unit()) {
         uint sum_size = 0;
         uint inst_cnt = NumberOfLoopInstrToAlign;
-        inst_cnt = b->compute_first_inst_size(sum_size, inst_cnt, _regalloc);
+        inst_cnt = block->compute_first_inst_size(sum_size, inst_cnt, _regalloc);
 
         // Check subsequent fallthrough blocks if the loop's first
         // block(s) does not have enough instructions.
-        Block *nb = b;
-        while( inst_cnt > 0 &&
-               i < last_block &&
-               !_cfg->_blocks[i+1]->has_loop_alignment() &&
-               !nb->has_successor(b) ) {
+        Block *nb = block;
+        while(inst_cnt > 0 &&
+              i < last_block &&
+              !_cfg->get_block(i + 1)->has_loop_alignment() &&
+              !nb->has_successor(block)) {
           i++;
-          nb = _cfg->_blocks[i];
+          nb = _cfg->get_block(i);
           inst_cnt  = nb->compute_first_inst_size(sum_size, inst_cnt, _regalloc);
         } // while( inst_cnt > 0 && i < last_block  )
 
-        b->set_first_inst_size(sum_size);
+        block->set_first_inst_size(sum_size);
       } // f( b->head()->is_Loop() )
     } // for( i <= last_block )
   } // if( MaxLoopPad < OptoLoopAlignment-1 )
 }
 
-//----------------------shorten_branches---------------------------------------
 // The architecture description provides short branch variants for some long
 // branch instructions. Replace eligible long branches with short branches.
 void Compile::shorten_branches(uint* blk_starts, int& code_size, int& reloc_size, int& stub_size) {
-
-  // ------------------
   // Compute size of each block, method size, and relocation information size
-  uint nblocks  = _cfg->_num_blocks;
+  uint nblocks  = _cfg->number_of_blocks();
 
   uint*      jmp_offset = NEW_RESOURCE_ARRAY(uint,nblocks);
   uint*      jmp_size   = NEW_RESOURCE_ARRAY(uint,nblocks);
@@ -364,7 +366,7 @@
   uint last_avoid_back_to_back_adr = max_uint;
   uint nop_size = (new (this) MachNopNode())->size(_regalloc);
   for (uint i = 0; i < nblocks; i++) { // For all blocks
-    Block *b = _cfg->_blocks[i];
+    Block* block = _cfg->get_block(i);
 
     // During short branch replacement, we store the relative (to blk_starts)
     // offset of jump in jmp_offset, rather than the absolute offset of jump.
@@ -377,10 +379,10 @@
     DEBUG_ONLY( jmp_rule[i]   = 0; )
 
     // Sum all instruction sizes to compute block size
-    uint last_inst = b->_nodes.size();
+    uint last_inst = block->_nodes.size();
     uint blk_size = 0;
     for (uint j = 0; j < last_inst; j++) {
-      Node* nj = b->_nodes[j];
+      Node* nj = block->_nodes[j];
       // Handle machine instruction nodes
       if (nj->is_Mach()) {
         MachNode *mach = nj->as_Mach();
@@ -441,8 +443,8 @@
     // When the next block starts a loop, we may insert pad NOP
     // instructions.  Since we cannot know our future alignment,
     // assume the worst.
-    if (i< nblocks-1) {
-      Block *nb = _cfg->_blocks[i+1];
+    if (i < nblocks - 1) {
+      Block* nb = _cfg->get_block(i + 1);
       int max_loop_pad = nb->code_alignment()-relocInfo::addr_unit();
       if (max_loop_pad > 0) {
         assert(is_power_of_2(max_loop_pad+relocInfo::addr_unit()), "");
@@ -473,26 +475,26 @@
     has_short_branch_candidate = false;
     int adjust_block_start = 0;
     for (uint i = 0; i < nblocks; i++) {
-      Block *b = _cfg->_blocks[i];
+      Block* block = _cfg->get_block(i);
       int idx = jmp_nidx[i];
-      MachNode* mach = (idx == -1) ? NULL: b->_nodes[idx]->as_Mach();
+      MachNode* mach = (idx == -1) ? NULL: block->_nodes[idx]->as_Mach();
       if (mach != NULL && mach->may_be_short_branch()) {
 #ifdef ASSERT
         assert(jmp_size[i] > 0 && mach->is_MachBranch(), "sanity");
         int j;
         // Find the branch; ignore trailing NOPs.
-        for (j = b->_nodes.size()-1; j>=0; j--) {
-          Node* n = b->_nodes[j];
+        for (j = block->_nodes.size()-1; j>=0; j--) {
+          Node* n = block->_nodes[j];
           if (!n->is_Mach() || n->as_Mach()->ideal_Opcode() != Op_Con)
             break;
         }
-        assert(j >= 0 && j == idx && b->_nodes[j] == (Node*)mach, "sanity");
+        assert(j >= 0 && j == idx && block->_nodes[j] == (Node*)mach, "sanity");
 #endif
         int br_size = jmp_size[i];
         int br_offs = blk_starts[i] + jmp_offset[i];
 
         // This requires the TRUE branch target be in succs[0]
-        uint bnum = b->non_connector_successor(0)->_pre_order;
+        uint bnum = block->non_connector_successor(0)->_pre_order;
         int offset = blk_starts[bnum] - br_offs;
         if (bnum > i) { // adjust following block's offset
           offset -= adjust_block_start;
@@ -520,7 +522,7 @@
             diff -= nop_size;
           }
           adjust_block_start += diff;
-          b->_nodes.map(idx, replacement);
+          block->_nodes.map(idx, replacement);
           mach->subsume_by(replacement, C);
           mach = replacement;
           progress = true;
@@ -1083,8 +1085,8 @@
   if (has_mach_constant_base_node()) {
     // Fill the constant table.
     // Note:  This must happen before shorten_branches.
-    for (uint i = 0; i < _cfg->_num_blocks; i++) {
-      Block* b = _cfg->_blocks[i];
+    for (uint i = 0; i < _cfg->number_of_blocks(); i++) {
+      Block* b = _cfg->get_block(i);
 
       for (uint j = 0; j < b->_nodes.size(); j++) {
         Node* n = b->_nodes[j];
@@ -1170,7 +1172,7 @@
   // !!!!! This preserves old handling of oopmaps for now
   debug_info()->set_oopmaps(_oop_map_set);
 
-  uint nblocks  = _cfg->_num_blocks;
+  uint nblocks  = _cfg->number_of_blocks();
   // Count and start of implicit null check instructions
   uint inct_cnt = 0;
   uint *inct_starts = NEW_RESOURCE_ARRAY(uint, nblocks+1);
@@ -1218,21 +1220,21 @@
   // Now fill in the code buffer
   Node *delay_slot = NULL;
 
-  for (uint i=0; i < nblocks; i++) {
-    Block *b = _cfg->_blocks[i];
-
-    Node *head = b->head();
+  for (uint i = 0; i < nblocks; i++) {
+    Block* block = _cfg->get_block(i);
+    Node* head = block->head();
 
     // If this block needs to start aligned (i.e, can be reached other
     // than by falling-thru from the previous block), then force the
     // start of a new bundle.
-    if (Pipeline::requires_bundling() && starts_bundle(head))
+    if (Pipeline::requires_bundling() && starts_bundle(head)) {
       cb->flush_bundle(true);
+    }
 
 #ifdef ASSERT
-    if (!b->is_connector()) {
+    if (!block->is_connector()) {
       stringStream st;
-      b->dump_head(_cfg, &st);
+      block->dump_head(_cfg, &st);
       MacroAssembler(cb).block_comment(st.as_string());
     }
     jmp_target[i] = 0;
@@ -1243,16 +1245,16 @@
     int blk_offset = current_offset;
 
     // Define the label at the beginning of the basic block
-    MacroAssembler(cb).bind(blk_labels[b->_pre_order]);
-
-    uint last_inst = b->_nodes.size();
+    MacroAssembler(cb).bind(blk_labels[block->_pre_order]);
+
+    uint last_inst = block->_nodes.size();
 
     // Emit block normally, except for last instruction.
     // Emit means "dump code bits into code buffer".
     for (uint j = 0; j<last_inst; j++) {
 
       // Get the node
-      Node* n = b->_nodes[j];
+      Node* n = block->_nodes[j];
 
       // See if delay slots are supported
       if (valid_bundle_info(n) &&
@@ -1306,9 +1308,9 @@
           assert((padding % nop_size) == 0, "padding is not a multiple of NOP size");
           int nops_cnt = padding / nop_size;
           MachNode *nop = new (this) MachNopNode(nops_cnt);
-          b->_nodes.insert(j++, nop);
+          block->_nodes.insert(j++, nop);
           last_inst++;
-          _cfg->map_node_to_block(nop, b);
+          _cfg->map_node_to_block(nop, block);
           nop->emit(*cb, _regalloc);
           cb->flush_bundle(true);
           current_offset = cb->insts_size();
@@ -1322,7 +1324,7 @@
           mcall->method_set((intptr_t)mcall->entry_point());
 
           // Save the return address
-          call_returns[b->_pre_order] = current_offset + mcall->ret_addr_offset();
+          call_returns[block->_pre_order] = current_offset + mcall->ret_addr_offset();
 
           if (mcall->is_MachCallLeaf()) {
             is_mcall = false;
@@ -1359,7 +1361,7 @@
         // If this is a branch, then fill in the label with the target BB's label
         else if (mach->is_MachBranch()) {
           // This requires the TRUE branch target be in succs[0]
-          uint block_num = b->non_connector_successor(0)->_pre_order;
+          uint block_num = block->non_connector_successor(0)->_pre_order;
 
           // Try to replace long branch if delay slot is not used,
           // it is mostly for back branches since forward branch's
@@ -1392,8 +1394,8 @@
               // Insert padding between avoid_back_to_back branches.
               if (needs_padding && replacement->avoid_back_to_back()) {
                 MachNode *nop = new (this) MachNopNode();
-                b->_nodes.insert(j++, nop);
-                _cfg->map_node_to_block(nop, b);
+                block->_nodes.insert(j++, nop);
+                _cfg->map_node_to_block(nop, block);
                 last_inst++;
                 nop->emit(*cb, _regalloc);
                 cb->flush_bundle(true);
@@ -1405,7 +1407,7 @@
               jmp_size[i]   = new_size;
               jmp_rule[i]   = mach->rule();
 #endif
-              b->_nodes.map(j, replacement);
+              block->_nodes.map(j, replacement);
               mach->subsume_by(replacement, C);
               n    = replacement;
               mach = replacement;
@@ -1413,8 +1415,8 @@
           }
           mach->as_MachBranch()->label_set( &blk_labels[block_num], block_num );