diff --git a/compiler/compiler.h b/compiler/compiler.h
index 1d5fc24..2357297 100644
--- a/compiler/compiler.h
+++ b/compiler/compiler.h
@@ -74,7 +74,8 @@
                                      uint32_t method_idx,
                                      const DexFile& dex_file) const = 0;
 
-  virtual uintptr_t GetEntryPointOf(mirror::ArtMethod* method) const = 0;
+  virtual uintptr_t GetEntryPointOf(mirror::ArtMethod* method) const
+     SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) = 0;
 
   virtual bool WriteElf(art::File* file,
                         OatWriter* oat_writer,
diff --git a/compiler/compilers.h b/compiler/compilers.h
index 892a6bd..255dd23 100644
--- a/compiler/compilers.h
+++ b/compiler/compilers.h
@@ -43,7 +43,8 @@
                              uint32_t method_idx,
                              const DexFile& dex_file) const OVERRIDE;
 
-  uintptr_t GetEntryPointOf(mirror::ArtMethod* method) const OVERRIDE;
+  uintptr_t GetEntryPointOf(mirror::ArtMethod* method) const OVERRIDE
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   bool WriteElf(art::File* file,
                 OatWriter* oat_writer,
diff --git a/compiler/image_writer.cc b/compiler/image_writer.cc
index c35d400..e6d983f 100644
--- a/compiler/image_writer.cc
+++ b/compiler/image_writer.cc
@@ -45,6 +45,7 @@
 #include "mirror/dex_cache-inl.h"
 #include "mirror/object-inl.h"
 #include "mirror/object_array-inl.h"
+#include "mirror/string-inl.h"
 #include "oat.h"
 #include "oat_file.h"
 #include "object_utils.h"
@@ -448,7 +449,7 @@
   for (size_t i = 0; i < num_reference_fields; ++i) {
     mirror::ArtField* field = sirt_class->GetInstanceField(i);
     MemberOffset field_offset = field->GetOffset();
-    mirror::Object* value = obj->GetFieldObject<mirror::Object>(field_offset, false);
+    mirror::Object* value = obj->GetFieldObject<mirror::Object>(field_offset);
     if (value != nullptr) {
       WalkFieldsInOrder(value);
     }
@@ -471,7 +472,7 @@
       for (size_t i = 0; i < num_static_fields; ++i) {
         mirror::ArtField* field = klass->GetStaticField(i);
         MemberOffset field_offset = field->GetOffset();
-        mirror::Object* value = sirt_obj->GetFieldObject<mirror::Object>(field_offset, false);
+        mirror::Object* value = sirt_obj->GetFieldObject<mirror::Object>(field_offset);
         if (value != nullptr) {
           WalkFieldsInOrder(value);
         }
@@ -586,11 +587,11 @@
 
   void operator()(Object* obj, MemberOffset offset, bool /*is_static*/) const
       EXCLUSIVE_LOCKS_REQUIRED(Locks::mutator_lock_, Locks::heap_bitmap_lock_) {
-    Object* ref = obj->GetFieldObject<Object, kVerifyNone>(offset, false);
+    Object* ref = obj->GetFieldObject<Object, kVerifyNone, false>(offset);
     // Use SetFieldObjectWithoutWriteBarrier to avoid card marking since we are writing to the
     // image.
     copy_->SetFieldObjectWithoutWriteBarrier<false, true, kVerifyNone>(
-        offset, image_writer_->GetImageAddress(ref), false);
+        offset, image_writer_->GetImageAddress(ref));
   }
 
   // java.lang.ref.Reference visitor.
@@ -598,8 +599,7 @@
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_)
       EXCLUSIVE_LOCKS_REQUIRED(Locks::heap_bitmap_lock_) {
     copy_->SetFieldObjectWithoutWriteBarrier<false, true, kVerifyNone>(
-        mirror::Reference::ReferentOffset(), image_writer_->GetImageAddress(ref->GetReferent()),
-        false);
+        mirror::Reference::ReferentOffset(), image_writer_->GetImageAddress(ref->GetReferent()));
   }
 
  private:
diff --git a/compiler/image_writer.h b/compiler/image_writer.h
index ee241cb..7e22a96 100644
--- a/compiler/image_writer.h
+++ b/compiler/image_writer.h
@@ -66,17 +66,20 @@
   void AssignImageOffset(mirror::Object* object) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
   void SetImageOffset(mirror::Object* object, size_t offset)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
-  bool IsImageOffsetAssigned(mirror::Object* object) const;
-  size_t GetImageOffset(mirror::Object* object) const;
+  bool IsImageOffsetAssigned(mirror::Object* object) const
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  size_t GetImageOffset(mirror::Object* object) const SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
-  mirror::Object* GetImageAddress(mirror::Object* object) const {
+  mirror::Object* GetImageAddress(mirror::Object* object) const
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     if (object == NULL) {
       return NULL;
     }
     return reinterpret_cast<mirror::Object*>(image_begin_ + GetImageOffset(object));
   }
 
-  mirror::Object* GetLocalAddress(mirror::Object* object) const {
+  mirror::Object* GetLocalAddress(mirror::Object* object) const
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     size_t offset = GetImageOffset(object);
     byte* dst = image_->Begin() + offset;
     return reinterpret_cast<mirror::Object*>(dst);
@@ -108,7 +111,7 @@
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   // Wire dex cache resolved strings to strings in the image to avoid runtime resolution.
-  void ComputeEagerResolvedStrings();
+  void ComputeEagerResolvedStrings() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
   static void ComputeEagerResolvedStringsCallback(mirror::Object* obj, void* arg)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
diff --git a/compiler/optimizing/code_generator_arm.cc b/compiler/optimizing/code_generator_arm.cc
index a446701..be51232 100644
--- a/compiler/optimizing/code_generator_arm.cc
+++ b/compiler/optimizing/code_generator_arm.cc
@@ -19,8 +19,10 @@
 #include "utils/arm/assembler_arm.h"
 #include "utils/arm/managed_register_arm.h"
 
+#include "entrypoints/quick/quick_entrypoints.h"
 #include "mirror/array.h"
 #include "mirror/art_method.h"
+#include "thread.h"
 
 #define __ reinterpret_cast<ArmAssembler*>(GetAssembler())->
 
diff --git a/compiler/optimizing/code_generator_x86.cc b/compiler/optimizing/code_generator_x86.cc
index fbb054a..e4f95c7 100644
--- a/compiler/optimizing/code_generator_x86.cc
+++ b/compiler/optimizing/code_generator_x86.cc
@@ -19,8 +19,10 @@
 #include "utils/x86/assembler_x86.h"
 #include "utils/x86/managed_register_x86.h"
 
+#include "entrypoints/quick/quick_entrypoints.h"
 #include "mirror/array.h"
 #include "mirror/art_method.h"
+#include "thread.h"
 
 #define __ reinterpret_cast<X86Assembler*>(GetAssembler())->
 
diff --git a/runtime/check_jni.cc b/runtime/check_jni.cc
index b52941b..3df050e 100644
--- a/runtime/check_jni.cc
+++ b/runtime/check_jni.cc
@@ -29,6 +29,7 @@
 #include "mirror/class-inl.h"
 #include "mirror/object-inl.h"
 #include "mirror/object_array-inl.h"
+#include "mirror/string-inl.h"
 #include "mirror/throwable.h"
 #include "object_utils.h"
 #include "runtime.h"
diff --git a/runtime/class_linker.cc b/runtime/class_linker.cc
index 338133c..583e5e5 100644
--- a/runtime/class_linker.cc
+++ b/runtime/class_linker.cc
@@ -4053,8 +4053,8 @@
         LOG(INFO) << "LinkFields: " << (is_static ? "static" : "instance")
                     << " class=" << PrettyClass(klass.get())
                     << " field=" << PrettyField(field)
-                    << " offset=" << field->GetField32(MemberOffset(mirror::ArtField::OffsetOffset()),
-                                                       false);
+                    << " offset="
+                    << field->GetField32(MemberOffset(mirror::ArtField::OffsetOffset()));
       }
       FieldHelper fh(field);
       Primitive::Type type = fh.GetTypeAsPrimitiveType();
diff --git a/runtime/class_linker.h b/runtime/class_linker.h
index 9771318..a23add0 100644
--- a/runtime/class_linker.h
+++ b/runtime/class_linker.h
@@ -251,7 +251,7 @@
       LOCKS_EXCLUDED(dex_lock_)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
   bool IsDexFileRegistered(const DexFile& dex_file) const
-      LOCKS_EXCLUDED(dex_lock_);
+      LOCKS_EXCLUDED(dex_lock_) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
   void FixupDexCaches(mirror::ArtMethod* resolution_method) const
       LOCKS_EXCLUDED(dex_lock_)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
@@ -462,7 +462,8 @@
   void RegisterDexFileLocked(const DexFile& dex_file, const SirtRef<mirror::DexCache>& dex_cache)
       EXCLUSIVE_LOCKS_REQUIRED(dex_lock_)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
-  bool IsDexFileRegisteredLocked(const DexFile& dex_file) const SHARED_LOCKS_REQUIRED(dex_lock_);
+  bool IsDexFileRegisteredLocked(const DexFile& dex_file) const
+      SHARED_LOCKS_REQUIRED(dex_lock_, Locks::mutator_lock_);
 
   bool InitializeClass(const SirtRef<mirror::Class>& klass, bool can_run_clinit,
                        bool can_init_parents)
diff --git a/runtime/debugger.cc b/runtime/debugger.cc
index 07d3a2a..1efd2e0 100644
--- a/runtime/debugger.cc
+++ b/runtime/debugger.cc
@@ -36,6 +36,7 @@
 #include "mirror/class_loader.h"
 #include "mirror/object-inl.h"
 #include "mirror/object_array-inl.h"
+#include "mirror/string-inl.h"
 #include "mirror/throwable.h"
 #include "object_utils.h"
 #include "quick/inline_method_analyser.h"
diff --git a/runtime/dex_file.cc b/runtime/dex_file.cc
index e5bc19c..6adfc1f 100644
--- a/runtime/dex_file.cc
+++ b/runtime/dex_file.cc
@@ -293,7 +293,6 @@
       location_(location),
       location_checksum_(location_checksum),
       mem_map_(mem_map),
-      modification_lock("DEX modification lock"),
       header_(reinterpret_cast<const Header*>(base)),
       string_ids_(reinterpret_cast<const StringId*>(base + header_->string_ids_off_)),
       type_ids_(reinterpret_cast<const TypeId*>(base + header_->type_ids_off_)),
diff --git a/runtime/dex_file.h b/runtime/dex_file.h
index 70baeed..c782ab1 100644
--- a/runtime/dex_file.h
+++ b/runtime/dex_file.h
@@ -21,7 +21,7 @@
 #include <vector>
 
 #include "base/logging.h"
-#include "base/mutex.h"
+#include "base/mutex.h"  // For Locks::mutator_lock_.
 #include "globals.h"
 #include "invoke_type.h"
 #include "jni.h"
@@ -386,10 +386,6 @@
     return *header_;
   }
 
-  Mutex& GetModificationLock() {
-    return modification_lock;
-  }
-
   // Decode the dex magic version
   uint32_t GetVersion() const;
 
@@ -877,11 +873,6 @@
   // Manages the underlying memory allocation.
   UniquePtr<MemMap> mem_map_;
 
-  // The DEX-to-DEX compiler uses this lock to ensure thread safety when
-  // enabling write access to a read-only DEX file.
-  // TODO: move to Locks::dex_file_modification_lock.
-  Mutex modification_lock;
-
   // Points to the header section.
   const Header* const header_;
 
diff --git a/runtime/gc/accounting/mod_union_table.cc b/runtime/gc/accounting/mod_union_table.cc
index d744dee..7cddaf4 100644
--- a/runtime/gc/accounting/mod_union_table.cc
+++ b/runtime/gc/accounting/mod_union_table.cc
@@ -101,7 +101,7 @@
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     DCHECK(root != NULL);
     ModUnionUpdateObjectReferencesVisitor ref_visitor(callback_, arg_);
-    root->VisitReferences<kMovingClasses>(ref_visitor);
+    root->VisitReferences<kMovingClasses>(ref_visitor, VoidFunctor());
   }
 
  private:
@@ -153,7 +153,7 @@
     // We don't have an early exit since we use the visitor pattern, an early
     // exit should significantly speed this up.
     AddToReferenceArrayVisitor visitor(mod_union_table_, references_);
-    obj->VisitReferences<kMovingClasses>(visitor);
+    obj->VisitReferences<kMovingClasses>(visitor, VoidFunctor());
   }
  private:
   ModUnionTableReferenceCache* const mod_union_table_;
@@ -171,7 +171,7 @@
   // Extra parameters are required since we use this same visitor signature for checking objects.
   void operator()(Object* obj, MemberOffset offset, bool /*is_static*/) const
       SHARED_LOCKS_REQUIRED(Locks::heap_bitmap_lock_, Locks::mutator_lock_) {
-    mirror::Object* ref = obj->GetFieldObject<mirror::Object>(offset, false);
+    mirror::Object* ref = obj->GetFieldObject<mirror::Object>(offset);
     if (ref != nullptr && mod_union_table_->ShouldAddReference(ref) &&
         references_.find(ref) == references_.end()) {
       Heap* heap = mod_union_table_->GetHeap();
@@ -205,7 +205,7 @@
   void operator()(Object* obj) const NO_THREAD_SAFETY_ANALYSIS {
     Locks::heap_bitmap_lock_->AssertSharedHeld(Thread::Current());
     CheckReferenceVisitor visitor(mod_union_table_, references_);
-    obj->VisitReferences<kMovingClasses>(visitor);
+    obj->VisitReferences<kMovingClasses>(visitor, VoidFunctor());
   }
 
  private:
diff --git a/runtime/gc/allocator_type.h b/runtime/gc/allocator_type.h
new file mode 100644
index 0000000..938b0f1
--- /dev/null
+++ b/runtime/gc/allocator_type.h
@@ -0,0 +1,36 @@
+/*
+ * Copyright (C) 2014 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef ART_RUNTIME_GC_ALLOCATOR_TYPE_H_
+#define ART_RUNTIME_GC_ALLOCATOR_TYPE_H_
+
+namespace art {
+namespace gc {
+
+// Different types of allocators.
+enum AllocatorType {
+  kAllocatorTypeBumpPointer,  // Use BumpPointer allocator, has entrypoints.
+  kAllocatorTypeTLAB,  // Use TLAB allocator, has entrypoints.
+  kAllocatorTypeRosAlloc,  // Use RosAlloc allocator, has entrypoints.
+  kAllocatorTypeDlMalloc,  // Use dlmalloc allocator, has entrypoints.
+  kAllocatorTypeNonMoving,  // Special allocator for non moving objects, doesn't have entrypoints.
+  kAllocatorTypeLOS,  // Large object space, also doesn't have entrypoints.
+};
+
+}  // namespace gc
+}  // namespace art
+
+#endif  // ART_RUNTIME_GC_ALLOCATOR_TYPE_H_
diff --git a/runtime/gc/collector/mark_sweep.cc b/runtime/gc/collector/mark_sweep.cc
index 9cd740e..e225d5a 100644
--- a/runtime/gc/collector/mark_sweep.cc
+++ b/runtime/gc/collector/mark_sweep.cc
@@ -596,7 +596,7 @@
 
     void operator()(Object* obj, MemberOffset offset, bool /* static */) const ALWAYS_INLINE
         SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-      mirror::Object* ref = obj->GetFieldObject<mirror::Object>(offset, false);
+      mirror::Object* ref = obj->GetFieldObject<mirror::Object>(offset);
       if (ref != nullptr && mark_sweep_->MarkObjectParallel(ref)) {
         if (kUseFinger) {
           android_memory_barrier();
@@ -1190,7 +1190,7 @@
       Locks::mutator_lock_->AssertSharedHeld(Thread::Current());
       Locks::heap_bitmap_lock_->AssertExclusiveHeld(Thread::Current());
     }
-    mark_sweep_->MarkObject(obj->GetFieldObject<mirror::Object>(offset, false));
+    mark_sweep_->MarkObject(obj->GetFieldObject<mirror::Object>(offset));
   }
 
  private:
diff --git a/runtime/gc/collector/mark_sweep.h b/runtime/gc/collector/mark_sweep.h
index 0c5a0da..bfc70d1 100644
--- a/runtime/gc/collector/mark_sweep.h
+++ b/runtime/gc/collector/mark_sweep.h
@@ -22,7 +22,7 @@
 #include "base/macros.h"
 #include "base/mutex.h"
 #include "garbage_collector.h"
-#include "gc/accounting/space_bitmap.h"
+#include "gc/accounting/heap_bitmap.h"
 #include "immune_region.h"
 #include "object_callbacks.h"
 #include "offsets.h"
diff --git a/runtime/gc/collector/semi_space.cc b/runtime/gc/collector/semi_space.cc
index 65bbbd2..0b26019 100644
--- a/runtime/gc/collector/semi_space.cc
+++ b/runtime/gc/collector/semi_space.cc
@@ -352,7 +352,7 @@
 
   void operator()(Object* obj, MemberOffset offset, bool /* is_static */) const
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) ALWAYS_INLINE {
-    mirror::Object* ref = obj->GetFieldObject<mirror::Object>(offset, false);
+    mirror::Object* ref = obj->GetFieldObject<mirror::Object>(offset);
     if (from_space_->HasAddress(ref)) {
       Runtime::Current()->GetHeap()->DumpObject(LOG(INFO), obj);
       LOG(FATAL) << ref << " found in from space";
@@ -365,7 +365,7 @@
 void SemiSpace::VerifyNoFromSpaceReferences(Object* obj) {
   DCHECK(!from_space_->HasAddress(obj)) << "Scanning object " << obj << " in from space";
   SemiSpaceVerifyNoFromSpaceReferencesVisitor visitor(from_space_);
-  obj->VisitReferences<kMovingClasses>(visitor);
+  obj->VisitReferences<kMovingClasses>(visitor, VoidFunctor());
 }
 
 class SemiSpaceVerifyNoFromSpaceReferencesObjectVisitor {
diff --git a/runtime/gc/collector/semi_space.h b/runtime/gc/collector/semi_space.h
index d468561..9b6df16 100644
--- a/runtime/gc/collector/semi_space.h
+++ b/runtime/gc/collector/semi_space.h
@@ -21,7 +21,7 @@
 #include "base/macros.h"
 #include "base/mutex.h"
 #include "garbage_collector.h"
-#include "gc/accounting/space_bitmap.h"
+#include "gc/accounting/heap_bitmap.h"
 #include "immune_region.h"
 #include "object_callbacks.h"
 #include "offsets.h"
@@ -191,7 +191,8 @@
   void ProcessMarkStack()
       EXCLUSIVE_LOCKS_REQUIRED(Locks::mutator_lock_, Locks::heap_bitmap_lock_);
 
-  inline mirror::Object* GetForwardingAddressInFromSpace(mirror::Object* obj) const;
+  inline mirror::Object* GetForwardingAddressInFromSpace(mirror::Object* obj) const
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   // Revoke all the thread-local buffers.
   void RevokeAllThreadLocalBuffers();
diff --git a/runtime/gc/heap.cc b/runtime/gc/heap.cc
index b57fc69..4d074f1 100644
--- a/runtime/gc/heap.cc
+++ b/runtime/gc/heap.cc
@@ -1136,8 +1136,7 @@
     return;
   }
   CHECK(IsAligned<kObjectAlignment>(obj)) << "Object isn't aligned: " << obj;
-  mirror::Class* c = obj->GetFieldObject<mirror::Class, kVerifyNone>(
-      mirror::Object::ClassOffset(), false);
+  mirror::Class* c = obj->GetFieldObject<mirror::Class, kVerifyNone>(mirror::Object::ClassOffset());
   CHECK(c != nullptr) << "Null class in object " << obj;
   CHECK(IsAligned<kObjectAlignment>(c)) << "Class " << c << " not aligned in object " << obj;
   CHECK(VerifyClassClass(c));
@@ -1378,13 +1377,13 @@
   // TODO: Fix lock analysis to not use NO_THREAD_SAFETY_ANALYSIS, requires support for
   // annotalysis on visitors.
   void operator()(mirror::Object* o) const NO_THREAD_SAFETY_ANALYSIS {
-    o->VisitReferences<true>(*this);
+    o->VisitReferences<true>(*this, VoidFunctor());
   }
 
   // For Object::VisitReferences.
   void operator()(mirror::Object* obj, MemberOffset offset, bool /* is_static */) const
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    mirror::Object* ref = obj->GetFieldObject<mirror::Object>(offset, false);
+    mirror::Object* ref = obj->GetFieldObject<mirror::Object>(offset);
     if (ref == object_ && (max_count_ == 0 || referring_objects_.size() < max_count_)) {
       referring_objects_.push_back(obj);
     }
@@ -1990,7 +1989,7 @@
 
   void operator()(mirror::Object* obj, MemberOffset offset, bool /*is_static*/) const
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    this->operator()(obj, obj->GetFieldObject<mirror::Object>(offset, false), offset);
+    this->operator()(obj, obj->GetFieldObject<mirror::Object>(offset), offset);
   }
 
   // TODO: Fix the no thread safety analysis.
@@ -2182,7 +2181,7 @@
   // annotalysis on visitors.
   void operator()(mirror::Object* obj, MemberOffset offset, bool is_static) const
       NO_THREAD_SAFETY_ANALYSIS {
-    mirror::Object* ref = obj->GetFieldObject<mirror::Object>(offset, false);
+    mirror::Object* ref = obj->GetFieldObject<mirror::Object>(offset);
     // Filter out class references since changing an object's class does not mark the card as dirty.
     // Also handles large objects, since the only reference they hold is a class reference.
     if (ref != nullptr && !ref->IsClass()) {
@@ -2252,7 +2251,7 @@
   void operator()(mirror::Object* obj) const
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_, Locks::heap_bitmap_lock_) {
     VerifyReferenceCardVisitor visitor(heap_, const_cast<bool*>(&failed_));
-    obj->VisitReferences<true>(visitor);
+    obj->VisitReferences<true>(visitor, VoidFunctor());
   }
 
   bool Failed() const {
diff --git a/runtime/gc/heap.h b/runtime/gc/heap.h
index 631397b..c631372 100644
--- a/runtime/gc/heap.h
+++ b/runtime/gc/heap.h
@@ -21,6 +21,7 @@
 #include <string>
 #include <vector>
 
+#include "allocator_type.h"
 #include "atomic.h"
 #include "base/timing_logger.h"
 #include "gc/accounting/atomic_stack.h"
@@ -90,16 +91,6 @@
   }
 };
 
-// Different types of allocators.
-enum AllocatorType {
-  kAllocatorTypeBumpPointer,  // Use BumpPointer allocator, has entrypoints.
-  kAllocatorTypeTLAB,  // Use TLAB allocator, has entrypoints.
-  kAllocatorTypeRosAlloc,  // Use RosAlloc allocator, has entrypoints.
-  kAllocatorTypeDlMalloc,  // Use dlmalloc allocator, has entrypoints.
-  kAllocatorTypeNonMoving,  // Special allocator for non moving objects, doesn't have entrypoints.
-  kAllocatorTypeLOS,  // Large object space, also doesn't have entrypoints.
-};
-
 // If true, use rosalloc/RosAllocSpace instead of dlmalloc/DlMallocSpace
 static constexpr bool kUseRosAlloc = true;
 
diff --git a/runtime/gc/space/space-inl.h b/runtime/gc/space/space-inl.h
index 3a715ab..3ea68cf 100644
--- a/runtime/gc/space/space-inl.h
+++ b/runtime/gc/space/space-inl.h
@@ -19,6 +19,7 @@
 
 #include "space.h"
 
+#include "base/casts.h"
 #include "dlmalloc_space.h"
 #include "image_space.h"
 #include "large_object_space.h"
diff --git a/runtime/interpreter/interpreter.cc b/runtime/interpreter/interpreter.cc
index e3f3cd0..3c6c225 100644
--- a/runtime/interpreter/interpreter.cc
+++ b/runtime/interpreter/interpreter.cc
@@ -15,8 +15,11 @@
  */
 
 #include "interpreter_common.h"
+
 #include <limits>
 
+#include "mirror/string-inl.h"
+
 namespace art {
 namespace interpreter {
 
@@ -85,7 +88,7 @@
     Object* obj = reinterpret_cast<Object*>(args[0]);
     jlong offset = (static_cast<uint64_t>(args[2]) << 32) | args[1];
     Object* newValue = reinterpret_cast<Object*>(args[3]);
-    obj->SetFieldObject<true>(MemberOffset(offset), newValue, false);
+    obj->SetFieldObject<true>(MemberOffset(offset), newValue);
   } else if (name == "int sun.misc.Unsafe.getArrayBaseOffsetForComponentType(java.lang.Class)") {
     mirror::Class* component = reinterpret_cast<Object*>(args[0])->AsClass();
     Primitive::Type primitive_type = component->GetPrimitiveType();
diff --git a/runtime/interpreter/interpreter_common.h b/runtime/interpreter/interpreter_common.h
index cc1fa0c..ce3346e 100644
--- a/runtime/interpreter/interpreter_common.h
+++ b/runtime/interpreter/interpreter_common.h
@@ -228,17 +228,17 @@
     instrumentation->FieldReadEvent(Thread::Current(), obj, shadow_frame.GetMethod(),
                                     shadow_frame.GetDexPC(), f);
   }
-  const bool is_volatile = false;  // iget-x-quick only on non volatile fields.
+  // Note: iget-x-quick instructions are only for non-volatile fields.
   const uint32_t vregA = inst->VRegA_22c(inst_data);
   switch (field_type) {
     case Primitive::kPrimInt:
-      shadow_frame.SetVReg(vregA, static_cast<int32_t>(obj->GetField32(field_offset, is_volatile)));
+      shadow_frame.SetVReg(vregA, static_cast<int32_t>(obj->GetField32(field_offset)));
       break;
     case Primitive::kPrimLong:
-      shadow_frame.SetVRegLong(vregA, static_cast<int64_t>(obj->GetField64(field_offset, is_volatile)));
+      shadow_frame.SetVRegLong(vregA, static_cast<int64_t>(obj->GetField64(field_offset)));
       break;
     case Primitive::kPrimNot:
-      shadow_frame.SetVRegReference(vregA, obj->GetFieldObject<mirror::Object>(field_offset, is_volatile));
+      shadow_frame.SetVRegReference(vregA, obj->GetFieldObject<mirror::Object>(field_offset));
       break;
     default:
       LOG(FATAL) << "Unreachable: " << field_type;
@@ -382,18 +382,16 @@
     instrumentation->FieldWriteEvent(Thread::Current(), obj, shadow_frame.GetMethod(),
                                      shadow_frame.GetDexPC(), f, field_value);
   }
-  const bool is_volatile = false;  // iput-x-quick only on non volatile fields.
+  // Note: iput-x-quick instructions are only for non-volatile fields.
   switch (field_type) {
     case Primitive::kPrimInt:
-      obj->SetField32<transaction_active>(field_offset, shadow_frame.GetVReg(vregA), is_volatile);
+      obj->SetField32<transaction_active>(field_offset, shadow_frame.GetVReg(vregA));
       break;
     case Primitive::kPrimLong:
-      obj->SetField64<transaction_active>(field_offset, shadow_frame.GetVRegLong(vregA),
-                                          is_volatile);
+      obj->SetField64<transaction_active>(field_offset, shadow_frame.GetVRegLong(vregA));
       break;
     case Primitive::kPrimNot:
-      obj->SetFieldObject<transaction_active>(field_offset, shadow_frame.GetVRegReference(vregA),
-                                              is_volatile);
+      obj->SetFieldObject<transaction_active>(field_offset, shadow_frame.GetVRegReference(vregA));
       break;
     default:
       LOG(FATAL) << "Unreachable: " << field_type;
diff --git a/runtime/interpreter/interpreter_switch_impl.cc b/runtime/interpreter/interpreter_switch_impl.cc
index abee1db..77e2a82 100644
--- a/runtime/interpreter/interpreter_switch_impl.cc
+++ b/runtime/interpreter/interpreter_switch_impl.cc
@@ -835,7 +835,7 @@
         }
         int32_t index = shadow_frame.GetVReg(inst->VRegC_23x());
         BooleanArray* array = a->AsBooleanArray();
-        if (LIKELY(array->CheckIsValidIndex(index))) {
+        if (array->CheckIsValidIndex(index)) {
           shadow_frame.SetVReg(inst->VRegA_23x(inst_data), array->GetWithoutChecks(index));
           inst = inst->Next_2xx();
         } else {
@@ -853,7 +853,7 @@
         }
         int32_t index = shadow_frame.GetVReg(inst->VRegC_23x());
         ByteArray* array = a->AsByteArray();
-        if (LIKELY(array->CheckIsValidIndex(index))) {
+        if (array->CheckIsValidIndex(index)) {
           shadow_frame.SetVReg(inst->VRegA_23x(inst_data), array->GetWithoutChecks(index));
           inst = inst->Next_2xx();
         } else {
@@ -871,7 +871,7 @@
         }
         int32_t index = shadow_frame.GetVReg(inst->VRegC_23x());
         CharArray* array = a->AsCharArray();
-        if (LIKELY(array->CheckIsValidIndex(index))) {
+        if (array->CheckIsValidIndex(index)) {
           shadow_frame.SetVReg(inst->VRegA_23x(inst_data), array->GetWithoutChecks(index));
           inst = inst->Next_2xx();
         } else {
@@ -889,7 +889,7 @@
         }
         int32_t index = shadow_frame.GetVReg(inst->VRegC_23x());
         ShortArray* array = a->AsShortArray();
-        if (LIKELY(array->CheckIsValidIndex(index))) {
+        if (array->CheckIsValidIndex(index)) {
           shadow_frame.SetVReg(inst->VRegA_23x(inst_data), array->GetWithoutChecks(index));
           inst = inst->Next_2xx();
         } else {
@@ -907,7 +907,7 @@
         }
         int32_t index = shadow_frame.GetVReg(inst->VRegC_23x());
         IntArray* array = a->AsIntArray();
-        if (LIKELY(array->CheckIsValidIndex(index))) {
+        if (array->CheckIsValidIndex(index)) {
           shadow_frame.SetVReg(inst->VRegA_23x(inst_data), array->GetWithoutChecks(index));
           inst = inst->Next_2xx();
         } else {
@@ -925,7 +925,7 @@
         }
         int32_t index = shadow_frame.GetVReg(inst->VRegC_23x());
         LongArray* array = a->AsLongArray();
-        if (LIKELY(array->CheckIsValidIndex(index))) {
+        if (array->CheckIsValidIndex(index)) {
           shadow_frame.SetVRegLong(inst->VRegA_23x(inst_data), array->GetWithoutChecks(index));
           inst = inst->Next_2xx();
         } else {
@@ -943,7 +943,7 @@
         }
         int32_t index = shadow_frame.GetVReg(inst->VRegC_23x());
         ObjectArray<Object>* array = a->AsObjectArray<Object>();
-        if (LIKELY(array->CheckIsValidIndex(index))) {
+        if (array->CheckIsValidIndex(index)) {
           shadow_frame.SetVRegReference(inst->VRegA_23x(inst_data), array->GetWithoutChecks(index));
           inst = inst->Next_2xx();
         } else {
@@ -962,7 +962,7 @@
         uint8_t val = shadow_frame.GetVReg(inst->VRegA_23x(inst_data));
         int32_t index = shadow_frame.GetVReg(inst->VRegC_23x());
         BooleanArray* array = a->AsBooleanArray();
-        if (LIKELY(array->CheckIsValidIndex(index))) {
+        if (array->CheckIsValidIndex(index)) {
           array->SetWithoutChecks<transaction_active>(index, val);
           inst = inst->Next_2xx();
         } else {
@@ -981,7 +981,7 @@
         int8_t val = shadow_frame.GetVReg(inst->VRegA_23x(inst_data));
         int32_t index = shadow_frame.GetVReg(inst->VRegC_23x());
         ByteArray* array = a->AsByteArray();
-        if (LIKELY(array->CheckIsValidIndex(index))) {
+        if (array->CheckIsValidIndex(index)) {
           array->SetWithoutChecks<transaction_active>(index, val);
           inst = inst->Next_2xx();
         } else {
@@ -1000,7 +1000,7 @@
         uint16_t val = shadow_frame.GetVReg(inst->VRegA_23x(inst_data));
         int32_t index = shadow_frame.GetVReg(inst->VRegC_23x());
         CharArray* array = a->AsCharArray();
-        if (LIKELY(array->CheckIsValidIndex(index))) {
+        if (array->CheckIsValidIndex(index)) {
           array->SetWithoutChecks<transaction_active>(index, val);
           inst = inst->Next_2xx();
         } else {
@@ -1019,7 +1019,7 @@
         int16_t val = shadow_frame.GetVReg(inst->VRegA_23x(inst_data));
         int32_t index = shadow_frame.GetVReg(inst->VRegC_23x());
         ShortArray* array = a->AsShortArray();
-        if (LIKELY(array->CheckIsValidIndex(index))) {
+        if (array->CheckIsValidIndex(index)) {
           array->SetWithoutChecks<transaction_active>(index, val);
           inst = inst->Next_2xx();
         } else {
@@ -1038,7 +1038,7 @@
         int32_t val = shadow_frame.GetVReg(inst->VRegA_23x(inst_data));
         int32_t index = shadow_frame.GetVReg(inst->VRegC_23x());
         IntArray* array = a->AsIntArray();
-        if (LIKELY(array->CheckIsValidIndex(index))) {
+        if (array->CheckIsValidIndex(index)) {
           array->SetWithoutChecks<transaction_active>(index, val);
           inst = inst->Next_2xx();
         } else {
@@ -1057,7 +1057,7 @@
         int64_t val = shadow_frame.GetVRegLong(inst->VRegA_23x(inst_data));
         int32_t index = shadow_frame.GetVReg(inst->VRegC_23x());
         LongArray* array = a->AsLongArray();
-        if (LIKELY(array->CheckIsValidIndex(index))) {
+        if (array->CheckIsValidIndex(index)) {
           array->SetWithoutChecks<transaction_active>(index, val);
           inst = inst->Next_2xx();
         } else {
@@ -1076,7 +1076,7 @@
         int32_t index = shadow_frame.GetVReg(inst->VRegC_23x());
         Object* val = shadow_frame.GetVRegReference(inst->VRegA_23x(inst_data));
         ObjectArray<Object>* array = a->AsObjectArray<Object>();
-        if (LIKELY(array->CheckIsValidIndex(index) && array->CheckAssignable(val))) {
+        if (array->CheckIsValidIndex(index) && array->CheckAssignable(val)) {
           array->SetWithoutChecks<transaction_active>(index, val);
           inst = inst->Next_2xx();
         } else {
diff --git a/runtime/jni_internal.cc b/runtime/jni_internal.cc
index fd9c40b..e6a35d0 100644
--- a/runtime/jni_internal.cc
+++ b/runtime/jni_internal.cc
@@ -37,6 +37,7 @@
 #include "mirror/class_loader.h"
 #include "mirror/object-inl.h"
 #include "mirror/object_array-inl.h"
+#include "mirror/string-inl.h"
 #include "mirror/throwable.h"
 #include "object_utils.h"
 #include "parsed_options.h"
diff --git a/runtime/mirror/array-inl.h b/runtime/mirror/array-inl.h
index 7f974d0..bc8d348 100644
--- a/runtime/mirror/array-inl.h
+++ b/runtime/mirror/array-inl.h
@@ -40,6 +40,16 @@
   return header_size + data_size;
 }
 
+template<VerifyObjectFlags kVerifyFlags>
+inline bool Array::CheckIsValidIndex(int32_t index) {
+  if (UNLIKELY(static_cast<uint32_t>(index) >=
+               static_cast<uint32_t>(GetLength<kVerifyFlags>()))) {
+    ThrowArrayIndexOutOfBoundsException(index);
+    return false;
+  }
+  return true;
+}
+
 static inline size_t ComputeArraySize(Thread* self, Class* array_class, int32_t component_count,
                                       size_t component_size)
     SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
@@ -164,6 +174,46 @@
   return down_cast<PrimitiveArray<T>*>(raw_array);
 }
 
+template<typename T>
+inline T PrimitiveArray<T>::Get(int32_t i) {
+  if (!CheckIsValidIndex(i)) {
+    DCHECK(Thread::Current()->IsExceptionPending());
+    return T(0);
+  }
+  return GetWithoutChecks(i);
+}
+
+template<typename T>
+inline void PrimitiveArray<T>::Set(int32_t i, T value) {
+  if (Runtime::Current()->IsActiveTransaction()) {
+    Set<true>(i, value);
+  } else {
+    Set<false>(i, value);
+  }
+}
+
+template<typename T>
+template<bool kTransactionActive, bool kCheckTransaction>
+inline void PrimitiveArray<T>::Set(int32_t i, T value) {
+  if (CheckIsValidIndex(i)) {
+    SetWithoutChecks<kTransactionActive, kCheckTransaction>(i, value);
+  } else {
+    DCHECK(Thread::Current()->IsExceptionPending());
+  }
+}
+
+template<typename T>
+template<bool kTransactionActive, bool kCheckTransaction>
+inline void PrimitiveArray<T>::SetWithoutChecks(int32_t i, T value) {
+  if (kCheckTransaction) {
+    DCHECK_EQ(kTransactionActive, Runtime::Current()->IsActiveTransaction());
+  }
+  if (kTransactionActive) {
+    Runtime::Current()->RecordWriteArray(this, i, GetWithoutChecks(i));
+  }
+  DCHECK(CheckIsValidIndex(i));
+  GetData()[i] = value;
+}
 // Backward copy where elements are of aligned appropriately for T. Count is in T sized units.
 // Copies are guaranteed not to tear when the sizeof T is less-than 64bit.
 template<typename T>
diff --git a/runtime/mirror/array.h b/runtime/mirror/array.h
index 6bfd5c8..92f0e67 100644
--- a/runtime/mirror/array.h
+++ b/runtime/mirror/array.h
@@ -17,13 +17,14 @@
 #ifndef ART_RUNTIME_MIRROR_ARRAY_H_
 #define ART_RUNTIME_MIRROR_ARRAY_H_
 
+#include "gc/allocator_type.h"
 #include "object.h"
 #include "object_callbacks.h"
-#include "gc/heap.h"
-#include "runtime.h"
-#include "thread.h"
 
 namespace art {
+
+template<class T> class SirtRef;
+
 namespace mirror {
 
 class MANAGED Array : public Object {
@@ -45,14 +46,14 @@
   size_t SizeOf() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
   template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
   int32_t GetLength() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    return GetField32<kVerifyFlags>(OFFSET_OF_OBJECT_MEMBER(Array, length_), false);
+    return GetField32<kVerifyFlags>(OFFSET_OF_OBJECT_MEMBER(Array, length_));
   }
 
   void SetLength(int32_t length) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     CHECK_GE(length, 0);
     // We use non transactional version since we can't undo this write. We also disable checking
     // since it would fail during a transaction.
-    SetField32<false, false, kVerifyNone>(OFFSET_OF_OBJECT_MEMBER(Array, length_), length, false);
+    SetField32<false, false, kVerifyNone>(OFFSET_OF_OBJECT_MEMBER(Array, length_), length);
   }
 
   static MemberOffset LengthOffset() {
@@ -84,14 +85,7 @@
   // Returns true if the index is valid. If not, throws an ArrayIndexOutOfBoundsException and
   // returns false.
   template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
-  bool CheckIsValidIndex(int32_t index) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    if (UNLIKELY(static_cast<uint32_t>(index) >=
-                 static_cast<uint32_t>(GetLength<kVerifyFlags>()))) {
-      ThrowArrayIndexOutOfBoundsException(index);
-      return false;
-    }
-    return true;
-  }
+  bool CheckIsValidIndex(int32_t index) ALWAYS_INLINE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
  protected:
   void ThrowArrayStoreException(Object* object) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
@@ -108,7 +102,7 @@
   DISALLOW_IMPLICIT_CONSTRUCTORS(Array);
 };
 
-template<class T>
+template<typename T>
 class MANAGED PrimitiveArray : public Array {
  public:
   typedef T ElementType;
@@ -116,59 +110,32 @@
   static PrimitiveArray<T>* Alloc(Thread* self, size_t length)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
-  const T* GetData() const SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+  const T* GetData() const ALWAYS_INLINE  SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     return reinterpret_cast<const T*>(GetRawData(sizeof(T), 0));
   }
 
-  T* GetData() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+  T* GetData() ALWAYS_INLINE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     return reinterpret_cast<T*>(GetRawData(sizeof(T), 0));
   }
 
-  T Get(int32_t i) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    if (UNLIKELY(!CheckIsValidIndex(i))) {
-      DCHECK(Thread::Current()->IsExceptionPending());
-      return T(0);
-    }
-    return GetWithoutChecks(i);
-  }
+  T Get(int32_t i) ALWAYS_INLINE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
-  T GetWithoutChecks(int32_t i) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+  T GetWithoutChecks(int32_t i) ALWAYS_INLINE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     DCHECK(CheckIsValidIndex(i));
     return GetData()[i];
   }
 
-  void Set(int32_t i, T value) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    if (Runtime::Current()->IsActiveTransaction()) {
-      Set<true>(i, value);
-    } else {
-      Set<false>(i, value);
-    }
-  }
+  void Set(int32_t i, T value) ALWAYS_INLINE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   // TODO fix thread safety analysis broken by the use of template. This should be
   // SHARED_LOCKS_REQUIRED(Locks::mutator_lock_).
   template<bool kTransactionActive, bool kCheckTransaction = true>
-  void Set(int32_t i, T value) NO_THREAD_SAFETY_ANALYSIS {
-    if (LIKELY(CheckIsValidIndex(i))) {
-      SetWithoutChecks<kTransactionActive, kCheckTransaction>(i, value);
-    } else {
-      DCHECK(Thread::Current()->IsExceptionPending());
-    }
-  }
+  void Set(int32_t i, T value) ALWAYS_INLINE NO_THREAD_SAFETY_ANALYSIS;
 
   // TODO fix thread safety analysis broken by the use of template. This should be
   // SHARED_LOCKS_REQUIRED(Locks::mutator_lock_).
   template<bool kTransactionActive, bool kCheckTransaction = true>
-  void SetWithoutChecks(int32_t i, T value) NO_THREAD_SAFETY_ANALYSIS {
-    if (kCheckTransaction) {
-      DCHECK_EQ(kTransactionActive, Runtime::Current()->IsActiveTransaction());
-    }
-    if (kTransactionActive) {
-      Runtime::Current()->RecordWriteArray(this, i, GetWithoutChecks(i));
-    }
-    DCHECK(CheckIsValidIndex(i));
-    GetData()[i] = value;
-  }
+  void SetWithoutChecks(int32_t i, T value) ALWAYS_INLINE NO_THREAD_SAFETY_ANALYSIS;
 
   /*
    * Works like memmove(), except we guarantee not to allow tearing of array values (ie using
diff --git a/runtime/mirror/art_field-inl.h b/runtime/mirror/art_field-inl.h
index 6253edd..ad24d0a 100644
--- a/runtime/mirror/art_field-inl.h
+++ b/runtime/mirror/art_field-inl.h
@@ -30,69 +30,89 @@
 namespace mirror {
 
 inline Class* ArtField::GetDeclaringClass() {
-  Class* result = GetFieldObject<Class>(OFFSET_OF_OBJECT_MEMBER(ArtField, declaring_class_), false);
+  Class* result = GetFieldObject<Class>(OFFSET_OF_OBJECT_MEMBER(ArtField, declaring_class_));
   DCHECK(result != NULL);
   DCHECK(result->IsLoaded() || result->IsErroneous());
   return result;
 }
 
 inline void ArtField::SetDeclaringClass(Class *new_declaring_class) {
-  SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(ArtField, declaring_class_),
-                        new_declaring_class, false);
+  SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(ArtField, declaring_class_), new_declaring_class);
 }
 
 inline uint32_t ArtField::GetAccessFlags() {
   DCHECK(GetDeclaringClass()->IsLoaded() || GetDeclaringClass()->IsErroneous());
-  return GetField32(OFFSET_OF_OBJECT_MEMBER(ArtField, access_flags_), false);
+  return GetField32(OFFSET_OF_OBJECT_MEMBER(ArtField, access_flags_));
 }
 
 inline MemberOffset ArtField::GetOffset() {
   DCHECK(GetDeclaringClass()->IsResolved() || GetDeclaringClass()->IsErroneous());
-  return MemberOffset(GetField32(OFFSET_OF_OBJECT_MEMBER(ArtField, offset_), false));
+  return MemberOffset(GetField32(OFFSET_OF_OBJECT_MEMBER(ArtField, offset_)));
 }
 
 inline MemberOffset ArtField::GetOffsetDuringLinking() {
   DCHECK(GetDeclaringClass()->IsLoaded() || GetDeclaringClass()->IsErroneous());
-  return MemberOffset(GetField32(OFFSET_OF_OBJECT_MEMBER(ArtField, offset_), false));
+  return MemberOffset(GetField32(OFFSET_OF_OBJECT_MEMBER(ArtField, offset_)));
 }
 
 inline uint32_t ArtField::Get32(Object* object) {
-  DCHECK(object != NULL) << PrettyField(this);
+  DCHECK(object != nullptr) << PrettyField(this);
   DCHECK(!IsStatic() || (object == GetDeclaringClass()) || !Runtime::Current()->IsStarted());
-  return object->GetField32(GetOffset(), IsVolatile());
+  if (UNLIKELY(IsVolatile())) {
+    return object->GetField32Volatile(GetOffset());
+  }
+  return object->GetField32(GetOffset());
 }
 
 template<bool kTransactionActive>
 inline void ArtField::Set32(Object* object, uint32_t new_value) {
-  DCHECK(object != NULL) << PrettyField(this);
+  DCHECK(object != nullptr) << PrettyField(this);
   DCHECK(!IsStatic() || (object == GetDeclaringClass()) || !Runtime::Current()->IsStarted());
-  object->SetField32<kTransactionActive>(GetOffset(), new_value, IsVolatile());
+  if (UNLIKELY(IsVolatile())) {
+    object->SetField32Volatile<kTransactionActive>(GetOffset(), new_value);
+  } else {
+    object->SetField32<kTransactionActive>(GetOffset(), new_value);
+  }
 }
 
 inline uint64_t ArtField::Get64(Object* object) {
   DCHECK(object != NULL) << PrettyField(this);
   DCHECK(!IsStatic() || (object == GetDeclaringClass()) || !Runtime::Current()->IsStarted());
-  return object->GetField64(GetOffset(), IsVolatile());
+  if (UNLIKELY(IsVolatile())) {
+    return object->GetField64Volatile(GetOffset());
+  }
+  return object->GetField64(GetOffset());
 }
 
 template<bool kTransactionActive>
 inline void ArtField::Set64(Object* object, uint64_t new_value) {
   DCHECK(object != NULL) << PrettyField(this);
   DCHECK(!IsStatic() || (object == GetDeclaringClass()) || !Runtime::Current()->IsStarted());
-  object->SetField64<kTransactionActive>(GetOffset(), new_value, IsVolatile());
+  if (UNLIKELY(IsVolatile())) {
+    object->SetField64Volatile<kTransactionActive>(GetOffset(), new_value);
+  } else {
+    object->SetField64<kTransactionActive>(GetOffset(), new_value);
+  }
 }
 
 inline Object* ArtField::GetObj(Object* object) {
   DCHECK(object != NULL) << PrettyField(this);
   DCHECK(!IsStatic() || (object == GetDeclaringClass()) || !Runtime::Current()->IsStarted());
-  return object->GetFieldObject<Object>(GetOffset(), IsVolatile());
+  if (UNLIKELY(IsVolatile())) {
+    return object->GetFieldObjectVolatile<Object>(GetOffset());
+  }
+  return object->GetFieldObject<Object>(GetOffset());
 }
 
 template<bool kTransactionActive>
 inline void ArtField::SetObj(Object* object, Object* new_value) {
   DCHECK(object != NULL) << PrettyField(this);
   DCHECK(!IsStatic() || (object == GetDeclaringClass()) || !Runtime::Current()->IsStarted());
-  object->SetFieldObject<kTransactionActive>(GetOffset(), new_value, IsVolatile());
+  if (UNLIKELY(IsVolatile())) {
+    object->SetFieldObjectVolatile<kTransactionActive>(GetOffset(), new_value);
+  } else {
+    object->SetFieldObject<kTransactionActive>(GetOffset(), new_value);
+  }
 }
 
 inline bool ArtField::GetBoolean(Object* object) {
diff --git a/runtime/mirror/art_field.cc b/runtime/mirror/art_field.cc
index 7b0b94c..8eb30f9 100644
--- a/runtime/mirror/art_field.cc
+++ b/runtime/mirror/art_field.cc
@@ -60,7 +60,7 @@
     }
   }
   // Not called within a transaction.
-  SetField32<false>(OFFSET_OF_OBJECT_MEMBER(ArtField, offset_), num_bytes.Uint32Value(), false);
+  SetField32<false>(OFFSET_OF_OBJECT_MEMBER(ArtField, offset_), num_bytes.Uint32Value());
 }
 
 void ArtField::VisitRoots(RootCallback* callback, void* arg) {
diff --git a/runtime/mirror/art_field.h b/runtime/mirror/art_field.h
index ba70cc6..029bd5a 100644
--- a/runtime/mirror/art_field.h
+++ b/runtime/mirror/art_field.h
@@ -17,6 +17,8 @@
 #ifndef ART_RUNTIME_MIRROR_ART_FIELD_H_
 #define ART_RUNTIME_MIRROR_ART_FIELD_H_
 
+#include <jni.h>
+
 #include "class.h"
 #include "modifiers.h"
 #include "object.h"
@@ -43,7 +45,7 @@
 
   void SetAccessFlags(uint32_t new_access_flags) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     // Not called within a transaction.
-    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(ArtField, access_flags_), new_access_flags, false);
+    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(ArtField, access_flags_), new_access_flags);
   }
 
   bool IsPublic() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
@@ -59,12 +61,12 @@
   }
 
   uint32_t GetDexFieldIndex() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    return GetField32(OFFSET_OF_OBJECT_MEMBER(ArtField, field_dex_idx_), false);
+    return GetField32(OFFSET_OF_OBJECT_MEMBER(ArtField, field_dex_idx_));
   }
 
   void SetDexFieldIndex(uint32_t new_idx) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     // Not called within a transaction.
-    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(ArtField, field_dex_idx_), new_idx, false);
+    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(ArtField, field_dex_idx_), new_idx);
   }
 
   // Offset to field within an Object.
diff --git a/runtime/mirror/art_method-inl.h b/runtime/mirror/art_method-inl.h
index 6e1f062..fb9a09a 100644
--- a/runtime/mirror/art_method-inl.h
+++ b/runtime/mirror/art_method-inl.h
@@ -29,8 +29,7 @@
 namespace mirror {
 
 inline Class* ArtMethod::GetDeclaringClass() {
-  Class* result = GetFieldObject<Class>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, declaring_class_),
-                                        false);
+  Class* result = GetFieldObject<Class>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, declaring_class_));
   DCHECK(result != NULL) << this;
   DCHECK(result->IsIdxLoaded() || result->IsErroneous()) << this;
   return result;
@@ -38,17 +37,17 @@
 
 inline void ArtMethod::SetDeclaringClass(Class *new_declaring_class) {
   SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, declaring_class_),
-                        new_declaring_class, false);
+                        new_declaring_class);
 }
 
 inline uint32_t ArtMethod::GetAccessFlags() {
   DCHECK(GetDeclaringClass()->IsIdxLoaded() || GetDeclaringClass()->IsErroneous());
-  return GetField32(OFFSET_OF_OBJECT_MEMBER(ArtMethod, access_flags_), false);
+  return GetField32(OFFSET_OF_OBJECT_MEMBER(ArtMethod, access_flags_));
 }
 
 inline uint16_t ArtMethod::GetMethodIndex() {
   DCHECK(GetDeclaringClass()->IsResolved() || GetDeclaringClass()->IsErroneous());
-  return GetField32(OFFSET_OF_OBJECT_MEMBER(ArtMethod, method_index_), false);
+  return GetField32(OFFSET_OF_OBJECT_MEMBER(ArtMethod, method_index_));
 }
 
 inline uint32_t ArtMethod::GetDexMethodIndex() {
@@ -58,22 +57,22 @@
 #else
   DCHECK(GetDeclaringClass()->IsLoaded() || GetDeclaringClass()->IsErroneous());
 #endif
-  return GetField32(OFFSET_OF_OBJECT_MEMBER(ArtMethod, dex_method_index_), false);
+  return GetField32(OFFSET_OF_OBJECT_MEMBER(ArtMethod, dex_method_index_));
 }
 
 inline ObjectArray<String>* ArtMethod::GetDexCacheStrings() {
   return GetFieldObject<ObjectArray<String> >(
-      OFFSET_OF_OBJECT_MEMBER(ArtMethod, dex_cache_strings_), false);
+      OFFSET_OF_OBJECT_MEMBER(ArtMethod, dex_cache_strings_));
 }
 
 inline ObjectArray<ArtMethod>* ArtMethod::GetDexCacheResolvedMethods() {
   return GetFieldObject<ObjectArray<ArtMethod> >(
-      OFFSET_OF_OBJECT_MEMBER(ArtMethod, dex_cache_resolved_methods_), false);
+      OFFSET_OF_OBJECT_MEMBER(ArtMethod, dex_cache_resolved_methods_));
 }
 
 inline ObjectArray<Class>* ArtMethod::GetDexCacheResolvedTypes() {
   return GetFieldObject<ObjectArray<Class> >(
-      OFFSET_OF_OBJECT_MEMBER(ArtMethod, dex_cache_resolved_types_), false);
+      OFFSET_OF_OBJECT_MEMBER(ArtMethod, dex_cache_resolved_types_));
 }
 
 inline uint32_t ArtMethod::GetCodeSize() {
@@ -199,7 +198,7 @@
 template<VerifyObjectFlags kVerifyFlags>
 inline void ArtMethod::SetNativeMethod(const void* native_method) {
   SetFieldPtr<false, true, kVerifyFlags>(
-      OFFSET_OF_OBJECT_MEMBER(ArtMethod, entry_point_from_jni_), native_method, false);
+      OFFSET_OF_OBJECT_MEMBER(ArtMethod, entry_point_from_jni_), native_method);
 }
 
 }  // namespace mirror
diff --git a/runtime/mirror/art_method.cc b/runtime/mirror/art_method.cc
index 90bcbab..7453d4d 100644
--- a/runtime/mirror/art_method.cc
+++ b/runtime/mirror/art_method.cc
@@ -90,17 +90,17 @@
 
 void ArtMethod::SetDexCacheStrings(ObjectArray<String>* new_dex_cache_strings) {
   SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, dex_cache_strings_),
-                        new_dex_cache_strings, false);
+                        new_dex_cache_strings);
 }
 
 void ArtMethod::SetDexCacheResolvedMethods(ObjectArray<ArtMethod>* new_dex_cache_methods) {
   SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, dex_cache_resolved_methods_),
-                        new_dex_cache_methods, false);
+                        new_dex_cache_methods);
 }
 
 void ArtMethod::SetDexCacheResolvedTypes(ObjectArray<Class>* new_dex_cache_classes) {
   SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, dex_cache_resolved_types_),
-                        new_dex_cache_classes, false);
+                        new_dex_cache_classes);
 }
 
 size_t ArtMethod::NumArgRegisters(const StringPiece& shorty) {
@@ -345,7 +345,7 @@
 
 bool ArtMethod::IsRegistered() {
   void* native_method =
-      GetFieldPtr<void*>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, entry_point_from_jni_), false);
+      GetFieldPtr<void*>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, entry_point_from_jni_));
   CHECK(native_method != nullptr);
   void* jni_stub = GetJniDlsymLookupStub();
   return native_method != jni_stub;
diff --git a/runtime/mirror/art_method.h b/runtime/mirror/art_method.h
index b3b9ca7..f61a01d 100644
--- a/runtime/mirror/art_method.h
+++ b/runtime/mirror/art_method.h
@@ -21,7 +21,6 @@
 #include "dex_file.h"
 #include "invoke_type.h"
 #include "modifiers.h"
-#include "oat.h"
 #include "object.h"
 #include "object_callbacks.h"
 
@@ -49,7 +48,7 @@
   static ArtMethod* FromReflectedMethod(const ScopedObjectAccess& soa, jobject jlr_method)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
-  Class* GetDeclaringClass() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  Class* GetDeclaringClass() ALWAYS_INLINE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   void SetDeclaringClass(Class *new_declaring_class) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
@@ -61,7 +60,7 @@
 
   void SetAccessFlags(uint32_t new_access_flags) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     // Not called within a transaction.
-    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, access_flags_), new_access_flags, false);
+    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, access_flags_), new_access_flags);
   }
 
   // Approximate what kind of method call would be used for this method.
@@ -162,7 +161,7 @@
 
   void SetMethodIndex(uint16_t new_method_index) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     // Not called within a transaction.
-    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, method_index_), new_method_index, false);
+    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, method_index_), new_method_index);
   }
 
   static MemberOffset MethodIndexOffset() {
@@ -170,12 +169,12 @@
   }
 
   uint32_t GetCodeItemOffset() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    return GetField32(OFFSET_OF_OBJECT_MEMBER(ArtMethod, dex_code_item_offset_), false);
+    return GetField32(OFFSET_OF_OBJECT_MEMBER(ArtMethod, dex_code_item_offset_));
   }
 
-  void SetCodeItemOffset(uint32_t new_code_off) {
+  void SetCodeItemOffset(uint32_t new_code_off) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     // Not called within a transaction.
-    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, dex_code_item_offset_), new_code_off, false);
+    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, dex_code_item_offset_), new_code_off);
   }
 
   // Number of 32bit registers that would be required to hold all the arguments
@@ -183,9 +182,9 @@
 
   uint32_t GetDexMethodIndex() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
-  void SetDexMethodIndex(uint32_t new_idx) {
+  void SetDexMethodIndex(uint32_t new_idx) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     // Not called within a transaction.
-    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, dex_method_index_), new_idx, false);
+    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, dex_method_index_), new_idx);
   }
 
   ObjectArray<String>* GetDexCacheStrings() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
@@ -219,16 +218,18 @@
               const char* shorty) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
-  EntryPointFromInterpreter* GetEntryPointFromInterpreter() {
+  EntryPointFromInterpreter* GetEntryPointFromInterpreter()
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     return GetFieldPtr<EntryPointFromInterpreter*, kVerifyFlags>(
-               OFFSET_OF_OBJECT_MEMBER(ArtMethod, entry_point_from_interpreter_), false);
+               OFFSET_OF_OBJECT_MEMBER(ArtMethod, entry_point_from_interpreter_));
   }
 
   template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
-  void SetEntryPointFromInterpreter(EntryPointFromInterpreter* entry_point_from_interpreter) {
+  void SetEntryPointFromInterpreter(EntryPointFromInterpreter* entry_point_from_interpreter)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     SetFieldPtr<false, true, kVerifyFlags>(
         OFFSET_OF_OBJECT_MEMBER(ArtMethod, entry_point_from_interpreter_),
-        entry_point_from_interpreter, false);
+        entry_point_from_interpreter);
   }
 
   static MemberOffset EntryPointFromPortableCompiledCodeOffset() {
@@ -236,15 +237,16 @@
   }
 
   template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
-  const void* GetEntryPointFromPortableCompiledCode() {
+  const void* GetEntryPointFromPortableCompiledCode() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     return GetFieldPtr<const void*, kVerifyFlags>(
-        EntryPointFromPortableCompiledCodeOffset(), false);
+        EntryPointFromPortableCompiledCodeOffset());
   }
 
   template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
-  void SetEntryPointFromPortableCompiledCode(const void* entry_point_from_portable_compiled_code) {
+  void SetEntryPointFromPortableCompiledCode(const void* entry_point_from_portable_compiled_code)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     SetFieldPtr<false, true, kVerifyFlags>(
-        EntryPointFromPortableCompiledCodeOffset(), entry_point_from_portable_compiled_code, false);
+        EntryPointFromPortableCompiledCodeOffset(), entry_point_from_portable_compiled_code);
   }
 
   static MemberOffset EntryPointFromQuickCompiledCodeOffset() {
@@ -252,14 +254,15 @@
   }
 
   template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
-  const void* GetEntryPointFromQuickCompiledCode() {
-    return GetFieldPtr<const void*, kVerifyFlags>(EntryPointFromQuickCompiledCodeOffset(), false);
+  const void* GetEntryPointFromQuickCompiledCode() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+    return GetFieldPtr<const void*, kVerifyFlags>(EntryPointFromQuickCompiledCodeOffset());
   }
 
   template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
-  void SetEntryPointFromQuickCompiledCode(const void* entry_point_from_quick_compiled_code) {
+  void SetEntryPointFromQuickCompiledCode(const void* entry_point_from_quick_compiled_code)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     SetFieldPtr<false, true, kVerifyFlags>(
-        EntryPointFromQuickCompiledCodeOffset(), entry_point_from_quick_compiled_code, false);
+        EntryPointFromQuickCompiledCodeOffset(), entry_point_from_quick_compiled_code);
   }
 
   uint32_t GetCodeSize() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
@@ -281,10 +284,10 @@
 
   void AssertPcIsWithinQuickCode(uintptr_t pc) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
-  uint32_t GetQuickOatCodeOffset();
-  uint32_t GetPortableOatCodeOffset();
-  void SetQuickOatCodeOffset(uint32_t code_offset);
-  void SetPortableOatCodeOffset(uint32_t code_offset);
+  uint32_t GetQuickOatCodeOffset() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  uint32_t GetPortableOatCodeOffset() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  void SetQuickOatCodeOffset(uint32_t code_offset) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  void SetPortableOatCodeOffset(uint32_t code_offset) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   static const void* EntryPointToCodePointer(const void* entry_point) ALWAYS_INLINE {
     uintptr_t code = reinterpret_cast<uintptr_t>(entry_point);
@@ -301,36 +304,35 @@
   // Callers should wrap the uint8_t* in a VmapTable instance for convenient access.
   const uint8_t* GetVmapTable() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
-  const uint8_t* GetNativeGcMap() {
-    return GetFieldPtr<uint8_t*>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, gc_map_), false);
+  const uint8_t* GetNativeGcMap() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+    return GetFieldPtr<uint8_t*>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, gc_map_));
   }
   template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
-  void SetNativeGcMap(const uint8_t* data) {
-    SetFieldPtr<false, true, kVerifyFlags>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, gc_map_), data,
-                                           false);
+  void SetNativeGcMap(const uint8_t* data) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+    SetFieldPtr<false, true, kVerifyFlags>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, gc_map_), data);
   }
 
   // When building the oat need a convenient place to stuff the offset of the native GC map.
-  void SetOatNativeGcMapOffset(uint32_t gc_map_offset);
-  uint32_t GetOatNativeGcMapOffset();
+  void SetOatNativeGcMapOffset(uint32_t gc_map_offset) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  uint32_t GetOatNativeGcMapOffset() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   template <bool kCheckFrameSize = true>
-  uint32_t GetFrameSizeInBytes() {
-    uint32_t result = GetField32(OFFSET_OF_OBJECT_MEMBER(ArtMethod, quick_frame_size_in_bytes_),
-                                 false);
+  uint32_t GetFrameSizeInBytes() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+    uint32_t result = GetField32(OFFSET_OF_OBJECT_MEMBER(ArtMethod, quick_frame_size_in_bytes_));
     if (kCheckFrameSize) {
       DCHECK_LE(static_cast<size_t>(kStackAlignment), result);
     }
     return result;
   }
 
-  void SetFrameSizeInBytes(size_t new_frame_size_in_bytes) {
+  void SetFrameSizeInBytes(size_t new_frame_size_in_bytes)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     // Not called within a transaction.
     SetField32<false>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, quick_frame_size_in_bytes_),
-                      new_frame_size_in_bytes, false);
+                      new_frame_size_in_bytes);
   }
 
-  size_t GetReturnPcOffsetInBytes() {
+  size_t GetReturnPcOffsetInBytes() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     return GetFrameSizeInBytes() - kPointerSize;
   }
 
@@ -338,7 +340,7 @@
     return kPointerSize;
   }
 
-  bool IsRegistered();
+  bool IsRegistered() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   void RegisterNative(Thread* self, const void* native_method, bool is_fast)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
@@ -349,35 +351,35 @@
     return OFFSET_OF_OBJECT_MEMBER(ArtMethod, entry_point_from_jni_);
   }
 
-  const void* GetNativeMethod() {
-    return GetFieldPtr<const void*>(NativeMethodOffset(), false);
+  const void* GetNativeMethod() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+    return GetFieldPtr<const void*>(NativeMethodOffset());
   }
 
   template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
-  void SetNativeMethod(const void*);
+  void SetNativeMethod(const void*) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   static MemberOffset GetMethodIndexOffset() {
     return OFFSET_OF_OBJECT_MEMBER(ArtMethod, method_index_);
   }
 
-  uint32_t GetCoreSpillMask() {
-    return GetField32(OFFSET_OF_OBJECT_MEMBER(ArtMethod, quick_core_spill_mask_), false);
+  uint32_t GetCoreSpillMask() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+    return GetField32(OFFSET_OF_OBJECT_MEMBER(ArtMethod, quick_core_spill_mask_));
   }
 
-  void SetCoreSpillMask(uint32_t core_spill_mask) {
+  void SetCoreSpillMask(uint32_t core_spill_mask) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     // Computed during compilation.
     // Not called within a transaction.
-    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, quick_core_spill_mask_), core_spill_mask, false);
+    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, quick_core_spill_mask_), core_spill_mask);
   }
 
-  uint32_t GetFpSpillMask() {
-    return GetField32(OFFSET_OF_OBJECT_MEMBER(ArtMethod, quick_fp_spill_mask_), false);
+  uint32_t GetFpSpillMask() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+    return GetField32(OFFSET_OF_OBJECT_MEMBER(ArtMethod, quick_fp_spill_mask_));
   }
 
-  void SetFpSpillMask(uint32_t fp_spill_mask) {
+  void SetFpSpillMask(uint32_t fp_spill_mask) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     // Computed during compilation.
     // Not called within a transaction.
-    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, quick_fp_spill_mask_), fp_spill_mask, false);
+    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(ArtMethod, quick_fp_spill_mask_), fp_spill_mask);
   }
 
   // Is this a CalleSaveMethod or ResolutionMethod and therefore doesn't adhere to normal
diff --git a/runtime/mirror/class-inl.h b/runtime/mirror/class-inl.h
index 3c02aa0..8a1f383 100644
--- a/runtime/mirror/class-inl.h
+++ b/runtime/mirror/class-inl.h
@@ -35,38 +35,36 @@
 
 inline uint32_t Class::GetObjectSize() {
   DCHECK(!IsVariableSize()) << " class=" << PrettyTypeOf(this);
-  return GetField32(OFFSET_OF_OBJECT_MEMBER(Class, object_size_), false);
+  return GetField32(OFFSET_OF_OBJECT_MEMBER(Class, object_size_));
 }
 
 inline Class* Class::GetSuperClass() {
   // Can only get super class for loaded classes (hack for when runtime is
   // initializing)
   DCHECK(IsLoaded() || IsErroneous() || !Runtime::Current()->IsStarted()) << IsLoaded();
-  return GetFieldObject<Class>(OFFSET_OF_OBJECT_MEMBER(Class, super_class_), false);
+  return GetFieldObject<Class>(OFFSET_OF_OBJECT_MEMBER(Class, super_class_));
 }
 
 inline ClassLoader* Class::GetClassLoader() {
-  return GetFieldObject<ClassLoader>(OFFSET_OF_OBJECT_MEMBER(Class, class_loader_), false);
+  return GetFieldObject<ClassLoader>(OFFSET_OF_OBJECT_MEMBER(Class, class_loader_));
 }
 
 template<VerifyObjectFlags kVerifyFlags>
 inline DexCache* Class::GetDexCache() {
-  return GetFieldObject<DexCache, kVerifyFlags>(OFFSET_OF_OBJECT_MEMBER(Class, dex_cache_), false);
+  return GetFieldObject<DexCache, kVerifyFlags>(OFFSET_OF_OBJECT_MEMBER(Class, dex_cache_));
 }
 
 inline ObjectArray<ArtMethod>* Class::GetDirectMethods() {
   DCHECK(IsLoaded() || IsErroneous());
-  return GetFieldObject<ObjectArray<ArtMethod> >(
-      OFFSET_OF_OBJECT_MEMBER(Class, direct_methods_), false);
+  return GetFieldObject<ObjectArray<ArtMethod> >(OFFSET_OF_OBJECT_MEMBER(Class, direct_methods_));
 }
 
 inline void Class::SetDirectMethods(ObjectArray<ArtMethod>* new_direct_methods)
     SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
   DCHECK(NULL == GetFieldObject<ObjectArray<ArtMethod> >(
-      OFFSET_OF_OBJECT_MEMBER(Class, direct_methods_), false));
+      OFFSET_OF_OBJECT_MEMBER(Class, direct_methods_)));
   DCHECK_NE(0, new_direct_methods->GetLength());
-  SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Class, direct_methods_),
-                        new_direct_methods, false);
+  SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Class, direct_methods_), new_direct_methods);
 }
 
 inline ArtMethod* Class::GetDirectMethod(int32_t i) {
@@ -76,8 +74,7 @@
 inline void Class::SetDirectMethod(uint32_t i, ArtMethod* f)  // TODO: uint16_t
     SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
   ObjectArray<ArtMethod>* direct_methods =
-      GetFieldObject<ObjectArray<ArtMethod> >(
-          OFFSET_OF_OBJECT_MEMBER(Class, direct_methods_), false);
+      GetFieldObject<ObjectArray<ArtMethod> >(OFFSET_OF_OBJECT_MEMBER(Class, direct_methods_));
   direct_methods->Set<false>(i, f);
 }
 
@@ -89,16 +86,14 @@
 template<VerifyObjectFlags kVerifyFlags>
 inline ObjectArray<ArtMethod>* Class::GetVirtualMethods() {
   DCHECK(IsLoaded() || IsErroneous());
-  return GetFieldObject<ObjectArray<ArtMethod> >(
-      OFFSET_OF_OBJECT_MEMBER(Class, virtual_methods_), false);
+  return GetFieldObject<ObjectArray<ArtMethod> >(OFFSET_OF_OBJECT_MEMBER(Class, virtual_methods_));
 }
 
 inline void Class::SetVirtualMethods(ObjectArray<ArtMethod>* new_virtual_methods) {
   // TODO: we reassign virtual methods to grow the table for miranda
   // methods.. they should really just be assigned once.
   DCHECK_NE(0, new_virtual_methods->GetLength());
-  SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Class, virtual_methods_),
-                        new_virtual_methods, false);
+  SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Class, virtual_methods_), new_virtual_methods);
 }
 
 inline uint32_t Class::NumVirtualMethods() {
@@ -119,31 +114,30 @@
 inline void Class::SetVirtualMethod(uint32_t i, ArtMethod* f)  // TODO: uint16_t
     SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
   ObjectArray<ArtMethod>* virtual_methods =
-      GetFieldObject<ObjectArray<ArtMethod> >(
-          OFFSET_OF_OBJECT_MEMBER(Class, virtual_methods_), false);
+      GetFieldObject<ObjectArray<ArtMethod> >(OFFSET_OF_OBJECT_MEMBER(Class, virtual_methods_));
   virtual_methods->Set<false>(i, f);
 }
 
 inline ObjectArray<ArtMethod>* Class::GetVTable() {
   DCHECK(IsResolved() || IsErroneous());
-  return GetFieldObject<ObjectArray<ArtMethod> >(OFFSET_OF_OBJECT_MEMBER(Class, vtable_), false);
+  return GetFieldObject<ObjectArray<ArtMethod> >(OFFSET_OF_OBJECT_MEMBER(Class, vtable_));
 }
 
 inline ObjectArray<ArtMethod>* Class::GetVTableDuringLinking() {
   DCHECK(IsLoaded() || IsErroneous());
-  return GetFieldObject<ObjectArray<ArtMethod> >(OFFSET_OF_OBJECT_MEMBER(Class, vtable_), false);
+  return GetFieldObject<ObjectArray<ArtMethod> >(OFFSET_OF_OBJECT_MEMBER(Class, vtable_));
 }
 
 inline void Class::SetVTable(ObjectArray<ArtMethod>* new_vtable) {
-  SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Class, vtable_), new_vtable, false);
+  SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Class, vtable_), new_vtable);
 }
 
 inline ObjectArray<ArtMethod>* Class::GetImTable() {
-  return GetFieldObject<ObjectArray<ArtMethod> >(OFFSET_OF_OBJECT_MEMBER(Class, imtable_), false);
+  return GetFieldObject<ObjectArray<ArtMethod> >(OFFSET_OF_OBJECT_MEMBER(Class, imtable_));
 }
 
 inline void Class::SetImTable(ObjectArray<ArtMethod>* new_imtable) {
-  SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Class, imtable_), new_imtable, false);
+  SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Class, imtable_), new_imtable);
 }
 
 inline bool Class::Implements(Class* klass) {
@@ -338,7 +332,7 @@
 }
 
 inline IfTable* Class::GetIfTable() {
-  return GetFieldObject<IfTable>(OFFSET_OF_OBJECT_MEMBER(Class, iftable_), false);
+  return GetFieldObject<IfTable>(OFFSET_OF_OBJECT_MEMBER(Class, iftable_));
 }
 
 inline int32_t Class::GetIfTableCount() {
@@ -350,46 +344,45 @@
 }
 
 inline void Class::SetIfTable(IfTable* new_iftable) {
-  SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Class, iftable_), new_iftable, false);
+  SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Class, iftable_), new_iftable);
 }
 
 inline ObjectArray<ArtField>* Class::GetIFields() {
   DCHECK(IsLoaded() || IsErroneous());
-  return GetFieldObject<ObjectArray<ArtField>>(OFFSET_OF_OBJECT_MEMBER(Class, ifields_), false);
+  return GetFieldObject<ObjectArray<ArtField>>(OFFSET_OF_OBJECT_MEMBER(Class, ifields_));
 }
 
 inline void Class::SetIFields(ObjectArray<ArtField>* new_ifields)
     SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-  DCHECK(NULL == GetFieldObject<ObjectArray<ArtField> >(
-      OFFSET_OF_OBJECT_MEMBER(Class, ifields_), false));
-  SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Class, ifields_), new_ifields, false);
+  DCHECK(NULL == GetFieldObject<ObjectArray<ArtField> >(OFFSET_OF_OBJECT_MEMBER(Class, ifields_)));
+  SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Class, ifields_), new_ifields);
 }
 
 inline ObjectArray<ArtField>* Class::GetSFields() {
   DCHECK(IsLoaded() || IsErroneous());
-  return GetFieldObject<ObjectArray<ArtField> >(OFFSET_OF_OBJECT_MEMBER(Class, sfields_), false);
+  return GetFieldObject<ObjectArray<ArtField> >(OFFSET_OF_OBJECT_MEMBER(Class, sfields_));
 }
 
 inline void Class::SetSFields(ObjectArray<ArtField>* new_sfields)
     SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-  DCHECK(NULL == GetFieldObject<ObjectArray<ArtField> >(
-      OFFSET_OF_OBJECT_MEMBER(Class, sfields_), false));
-  SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Class, sfields_), new_sfields, false);
+  DCHECK(NULL == GetFieldObject<ObjectArray<ArtField> >(OFFSET_OF_OBJECT_MEMBER(Class, sfields_)));
+  SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Class, sfields_), new_sfields);
 }
 
 inline uint32_t Class::NumStaticFields() {
   return (GetSFields() != NULL) ? GetSFields()->GetLength() : 0;
 }
 
+
 inline ArtField* Class::GetStaticField(uint32_t i)  // TODO: uint16_t
     SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-  return GetSFields()->Get(i);
+  return GetSFields()->GetWithoutChecks(i);
 }
 
 inline void Class::SetStaticField(uint32_t i, ArtField* f)  // TODO: uint16_t
     SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
   ObjectArray<ArtField>* sfields= GetFieldObject<ObjectArray<ArtField> >(
-      OFFSET_OF_OBJECT_MEMBER(Class, sfields_), false);
+      OFFSET_OF_OBJECT_MEMBER(Class, sfields_));
   sfields->Set<false>(i, f);
 }
 
@@ -399,22 +392,36 @@
 
 inline ArtField* Class::GetInstanceField(uint32_t i) {  // TODO: uint16_t
   DCHECK_NE(NumInstanceFields(), 0U);
-  return GetIFields()->Get(i);
+  return GetIFields()->GetWithoutChecks(i);
 }
 
 inline void Class::SetInstanceField(uint32_t i, ArtField* f)  // TODO: uint16_t
     SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
   ObjectArray<ArtField>* ifields= GetFieldObject<ObjectArray<ArtField> >(
-      OFFSET_OF_OBJECT_MEMBER(Class, ifields_), false);
+      OFFSET_OF_OBJECT_MEMBER(Class, ifields_));
   ifields->Set<false>(i, f);
 }
 
+template<VerifyObjectFlags kVerifyFlags>
+inline uint32_t Class::GetReferenceInstanceOffsets() {
+  DCHECK(IsResolved<kVerifyFlags>() || IsErroneous<kVerifyFlags>());
+  return GetField32<kVerifyFlags>(OFFSET_OF_OBJECT_MEMBER(Class, reference_instance_offsets_));
+}
+
+inline void Class::SetClinitThreadId(pid_t new_clinit_thread_id) {
+  if (Runtime::Current()->IsActiveTransaction()) {
+    SetField32<true>(OFFSET_OF_OBJECT_MEMBER(Class, clinit_thread_id_), new_clinit_thread_id);
+  } else {
+    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(Class, clinit_thread_id_), new_clinit_thread_id);
+  }
+}
+
 inline void Class::SetVerifyErrorClass(Class* klass) {
   CHECK(klass != NULL) << PrettyClass(this);
   if (Runtime::Current()->IsActiveTransaction()) {
-    SetFieldObject<true>(OFFSET_OF_OBJECT_MEMBER(Class, verify_error_class_), klass, false);
+    SetFieldObject<true>(OFFSET_OF_OBJECT_MEMBER(Class, verify_error_class_), klass);
   } else {
-    SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Class, verify_error_class_), klass, false);
+    SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Class, verify_error_class_), klass);
   }
 }
 
@@ -427,20 +434,27 @@
          this == String::GetJavaLangString() ||
          this == ArtField::GetJavaLangReflectArtField() ||
          this == ArtMethod::GetJavaLangReflectArtMethod());
-  return GetField32<kVerifyFlags>(OFFSET_OF_OBJECT_MEMBER(Class, access_flags_), false);
+  return GetField32<kVerifyFlags>(OFFSET_OF_OBJECT_MEMBER(Class, access_flags_));
 }
 
 inline String* Class::GetName() {
-  return GetFieldObject<String>(OFFSET_OF_OBJECT_MEMBER(Class, name_), false);
+  return GetFieldObject<String>(OFFSET_OF_OBJECT_MEMBER(Class, name_));
 }
 inline void Class::SetName(String* name) {
   if (Runtime::Current()->IsActiveTransaction()) {
-    SetFieldObject<true>(OFFSET_OF_OBJECT_MEMBER(Class, name_), name, false);
+    SetFieldObject<true>(OFFSET_OF_OBJECT_MEMBER(Class, name_), name);
   } else {
-    SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Class, name_), name, false);
+    SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Class, name_), name);
   }
 }
 
+template<VerifyObjectFlags kVerifyFlags>
+inline Primitive::Type Class::GetPrimitiveType() {
+  DCHECK_EQ(sizeof(Primitive::Type), sizeof(int32_t));
+  return static_cast<Primitive::Type>(
+      GetField32<kVerifyFlags>(OFFSET_OF_OBJECT_MEMBER(Class, primitive_type_)));
+}
+
 inline void Class::CheckObjectAlloc() {
   DCHECK(!IsArrayClass())
       << PrettyClass(this)
diff --git a/runtime/mirror/class.cc b/runtime/mirror/class.cc
index ad86e1f..64a849b 100644
--- a/runtime/mirror/class.cc
+++ b/runtime/mirror/class.cc
@@ -114,9 +114,9 @@
   }
   CHECK(sizeof(Status) == sizeof(uint32_t)) << PrettyClass(this);
   if (Runtime::Current()->IsActiveTransaction()) {
-    SetField32<true>(OFFSET_OF_OBJECT_MEMBER(Class, status_), new_status, false);
+    SetField32<true>(OFFSET_OF_OBJECT_MEMBER(Class, status_), new_status);
   } else {
-    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(Class, status_), new_status, false);
+    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(Class, status_), new_status);
   }
   // Classes that are being resolved or initialized need to notify waiters that the class status
   // changed. See ClassLinker::EnsureResolved and ClassLinker::WaitForInitializeClass.
@@ -127,7 +127,7 @@
 }
 
 void Class::SetDexCache(DexCache* new_dex_cache) {
-  SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Class, dex_cache_), new_dex_cache, false);
+  SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Class, dex_cache_), new_dex_cache);
 }
 
 void Class::SetClassSize(uint32_t new_class_size) {
@@ -136,7 +136,7 @@
     CHECK_GE(new_class_size, GetClassSize()) << " class=" << PrettyTypeOf(this);
   }
   // Not called within a transaction.
-  SetField32<false>(OFFSET_OF_OBJECT_MEMBER(Class, class_size_), new_class_size, false);
+  SetField32<false>(OFFSET_OF_OBJECT_MEMBER(Class, class_size_), new_class_size);
 }
 
 // Return the class' name. The exact format is bizarre, but it's the specified behavior for
@@ -261,7 +261,7 @@
   }
   // Not called within a transaction.
   SetField32<false>(OFFSET_OF_OBJECT_MEMBER(Class, reference_instance_offsets_),
-                    new_reference_offsets, false);
+                    new_reference_offsets);
 }
 
 void Class::SetReferenceStaticOffsets(uint32_t new_reference_offsets) {
@@ -273,7 +273,7 @@
   }
   // Not called within a transaction.
   SetField32<false>(OFFSET_OF_OBJECT_MEMBER(Class, reference_static_offsets_),
-                    new_reference_offsets, false);
+                    new_reference_offsets);
 }
 
 bool Class::IsInSamePackage(const StringPiece& descriptor1, const StringPiece& descriptor2) {
@@ -330,9 +330,9 @@
 
 void Class::SetClassLoader(ClassLoader* new_class_loader) {
   if (Runtime::Current()->IsActiveTransaction()) {
-    SetFieldObject<true>(OFFSET_OF_OBJECT_MEMBER(Class, class_loader_), new_class_loader, false);
+    SetFieldObject<true>(OFFSET_OF_OBJECT_MEMBER(Class, class_loader_), new_class_loader);
   } else {
-    SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Class, class_loader_), new_class_loader, false);
+    SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Class, class_loader_), new_class_loader);
   }
 }
 
diff --git a/runtime/mirror/class.h b/runtime/mirror/class.h
index 226dee0..23211c2 100644
--- a/runtime/mirror/class.h
+++ b/runtime/mirror/class.h
@@ -17,9 +17,11 @@
 #ifndef ART_RUNTIME_MIRROR_CLASS_H_
 #define ART_RUNTIME_MIRROR_CLASS_H_
 
+#include "gc/allocator_type.h"
 #include "invoke_type.h"
 #include "modifiers.h"
 #include "object.h"
+#include "object_callbacks.h"
 #include "primitive.h"
 
 /*
@@ -122,8 +124,7 @@
   template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
   Status GetStatus() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     COMPILE_ASSERT(sizeof(Status) == sizeof(uint32_t), size_of_status_not_uint32);
-    return static_cast<Status>(GetField32<kVerifyFlags>(OFFSET_OF_OBJECT_MEMBER(Class, status_),
-                                                       true));
+    return static_cast<Status>(GetField32<kVerifyFlags>(OFFSET_OF_OBJECT_MEMBER(Class, status_)));
   }
 
   void SetStatus(Status new_status, Thread* self) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
@@ -185,7 +186,7 @@
 
   void SetAccessFlags(uint32_t new_access_flags) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     // Not called within a transaction.
-    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(Class, access_flags_), new_access_flags, false);
+    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(Class, access_flags_), new_access_flags);
   }
 
   // Returns true if the class is an interface.
@@ -208,7 +209,7 @@
   }
 
   void SetFinalizable() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    uint32_t flags = GetField32(OFFSET_OF_OBJECT_MEMBER(Class, access_flags_), false);
+    uint32_t flags = GetField32(OFFSET_OF_OBJECT_MEMBER(Class, access_flags_));
     SetAccessFlags(flags | kAccClassIsFinalizable);
   }
 
@@ -280,21 +281,16 @@
     // Read access flags without using getter as whether something is a proxy can be check in
     // any loaded state
     // TODO: switch to a check if the super class is java.lang.reflect.Proxy?
-    uint32_t access_flags = GetField32<kVerifyFlags>(OFFSET_OF_OBJECT_MEMBER(Class, access_flags_),
-                                                     false);
+    uint32_t access_flags = GetField32<kVerifyFlags>(OFFSET_OF_OBJECT_MEMBER(Class, access_flags_));
     return (access_flags & kAccClassIsProxy) != 0;
   }
 
   template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
-  Primitive::Type GetPrimitiveType() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    DCHECK_EQ(sizeof(Primitive::Type), sizeof(int32_t));
-    return static_cast<Primitive::Type>(
-        GetField32<kVerifyFlags>(OFFSET_OF_OBJECT_MEMBER(Class, primitive_type_), false));
-  }
+  Primitive::Type GetPrimitiveType() ALWAYS_INLINE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   void SetPrimitiveType(Primitive::Type new_type) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     DCHECK_EQ(sizeof(Primitive::Type), sizeof(int32_t));
-    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(Class, primitive_type_), new_type, false);
+    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(Class, primitive_type_), new_type);
   }
 
   // Returns true if the class is a primitive type.
@@ -387,14 +383,14 @@
 
   template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags, bool kDoReadBarrier = true>
   Class* GetComponentType() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    return GetFieldObject<Class, kVerifyFlags, kDoReadBarrier>(ComponentTypeOffset(), false);
+    return GetFieldObject<Class, kVerifyFlags, kDoReadBarrier>(ComponentTypeOffset());
   }
 
   void SetComponentType(Class* new_component_type) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     DCHECK(GetComponentType() == NULL);
     DCHECK(new_component_type != NULL);
     // Component type is invariant: use non-transactional mode without check.
-    SetFieldObject<false, false>(ComponentTypeOffset(), new_component_type, false);
+    SetFieldObject<false, false>(ComponentTypeOffset(), new_component_type);
   }
 
   template<bool kDoReadBarrier = true>
@@ -433,12 +429,12 @@
 
   template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags, bool kDoReadBarrier = true>
   uint32_t SizeOf() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    return GetField32<kVerifyFlags>(OFFSET_OF_OBJECT_MEMBER(Class, class_size_), false);
+    return GetField32<kVerifyFlags>(OFFSET_OF_OBJECT_MEMBER(Class, class_size_));
   }
 
   template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
   uint32_t GetClassSize() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    return GetField32<kVerifyFlags>(OFFSET_OF_OBJECT_MEMBER(Class, class_size_), false);
+    return GetField32<kVerifyFlags>(OFFSET_OF_OBJECT_MEMBER(Class, class_size_));
   }
 
   void SetClassSize(uint32_t new_class_size)
@@ -449,7 +445,7 @@
   void SetObjectSize(uint32_t new_object_size) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     DCHECK(!IsVariableSize());
     // Not called within a transaction.
-    return SetField32<false>(OFFSET_OF_OBJECT_MEMBER(Class, object_size_), new_object_size, false);
+    return SetField32<false>(OFFSET_OF_OBJECT_MEMBER(Class, object_size_), new_object_size);
   }
 
   // Returns true if this class is in the same packages as that class.
@@ -538,11 +534,10 @@
 
   void SetSuperClass(Class *new_super_class) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     // Super class is assigned once, except during class linker initialization.
-    Class* old_super_class = GetFieldObject<Class>(OFFSET_OF_OBJECT_MEMBER(Class, super_class_),
-                                                   false);
+    Class* old_super_class = GetFieldObject<Class>(OFFSET_OF_OBJECT_MEMBER(Class, super_class_));
     DCHECK(old_super_class == nullptr || old_super_class == new_super_class);
     DCHECK(new_super_class != nullptr);
-    SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Class, super_class_), new_super_class, false);
+    SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Class, super_class_), new_super_class);
   }
 
   bool HasSuperClass() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
@@ -553,7 +548,7 @@
     return MemberOffset(OFFSETOF_MEMBER(Class, super_class_));
   }
 
-  ClassLoader* GetClassLoader() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  ClassLoader* GetClassLoader() ALWAYS_INLINE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   void SetClassLoader(ClassLoader* new_cl) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
@@ -709,26 +704,21 @@
   // Returns the number of instance fields containing reference types.
   uint32_t NumReferenceInstanceFields() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     DCHECK(IsResolved() || IsErroneous());
-    return GetField32(OFFSET_OF_OBJECT_MEMBER(Class, num_reference_instance_fields_), false);
+    return GetField32(OFFSET_OF_OBJECT_MEMBER(Class, num_reference_instance_fields_));
   }
 
   uint32_t NumReferenceInstanceFieldsDuringLinking() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     DCHECK(IsLoaded() || IsErroneous());
-    return GetField32(OFFSET_OF_OBJECT_MEMBER(Class, num_reference_instance_fields_), false);
+    return GetField32(OFFSET_OF_OBJECT_MEMBER(Class, num_reference_instance_fields_));
   }
 
-  void SetNumReferenceInstanceFields(uint32_t new_num) {
+  void SetNumReferenceInstanceFields(uint32_t new_num) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     // Not called within a transaction.
-    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(Class, num_reference_instance_fields_), new_num,
-                      false);
+    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(Class, num_reference_instance_fields_), new_num);
   }
 
   template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
-  uint32_t GetReferenceInstanceOffsets() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    DCHECK(IsResolved<kVerifyFlags>() || IsErroneous<kVerifyFlags>());
-    return GetField32<kVerifyFlags>(OFFSET_OF_OBJECT_MEMBER(Class, reference_instance_offsets_),
-                                   false);
-  }
+  uint32_t GetReferenceInstanceOffsets() ALWAYS_INLINE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   void SetReferenceInstanceOffsets(uint32_t new_reference_offsets)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
@@ -741,17 +731,17 @@
   // Returns the number of static fields containing reference types.
   uint32_t NumReferenceStaticFields() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     DCHECK(IsResolved() || IsErroneous());
-    return GetField32(OFFSET_OF_OBJECT_MEMBER(Class, num_reference_static_fields_), false);
+    return GetField32(OFFSET_OF_OBJECT_MEMBER(Class, num_reference_static_fields_));
   }
 
   uint32_t NumReferenceStaticFieldsDuringLinking() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     DCHECK(IsLoaded() || IsErroneous());
-    return GetField32(OFFSET_OF_OBJECT_MEMBER(Class, num_reference_static_fields_), false);
+    return GetField32(OFFSET_OF_OBJECT_MEMBER(Class, num_reference_static_fields_));
   }
 
-  void SetNumReferenceStaticFields(uint32_t new_num) {
+  void SetNumReferenceStaticFields(uint32_t new_num) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     // Not called within a transaction.
-    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(Class, num_reference_static_fields_), new_num, false);
+    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(Class, num_reference_static_fields_), new_num);
   }
 
   // Gets the static fields of the class.
@@ -769,8 +759,7 @@
 
   template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
   uint32_t GetReferenceStaticOffsets() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    return GetField32<kVerifyFlags>(OFFSET_OF_OBJECT_MEMBER(Class, reference_static_offsets_),
-                                   false);
+    return GetField32<kVerifyFlags>(OFFSET_OF_OBJECT_MEMBER(Class, reference_static_offsets_));
   }
 
   void SetReferenceStaticOffsets(uint32_t new_reference_offsets)
@@ -812,40 +801,32 @@
 
   pid_t GetClinitThreadId() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     DCHECK(IsIdxLoaded() || IsErroneous());
-    return GetField32(OFFSET_OF_OBJECT_MEMBER(Class, clinit_thread_id_), false);
+    return GetField32(OFFSET_OF_OBJECT_MEMBER(Class, clinit_thread_id_));
   }
 
-  void SetClinitThreadId(pid_t new_clinit_thread_id) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    if (Runtime::Current()->IsActiveTransaction()) {
-      SetField32<true>(OFFSET_OF_OBJECT_MEMBER(Class, clinit_thread_id_), new_clinit_thread_id,
-                       false);
-    } else {
-      SetField32<false>(OFFSET_OF_OBJECT_MEMBER(Class, clinit_thread_id_), new_clinit_thread_id,
-                        false);
-    }
-  }
+  void SetClinitThreadId(pid_t new_clinit_thread_id) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   Class* GetVerifyErrorClass() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     // DCHECK(IsErroneous());
-    return GetFieldObject<Class>(OFFSET_OF_OBJECT_MEMBER(Class, verify_error_class_), false);
+    return GetFieldObject<Class>(OFFSET_OF_OBJECT_MEMBER(Class, verify_error_class_));
   }
 
   uint16_t GetDexClassDefIndex() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    return GetField32(OFFSET_OF_OBJECT_MEMBER(Class, dex_class_def_idx_), false);
+    return GetField32(OFFSET_OF_OBJECT_MEMBER(Class, dex_class_def_idx_));
   }
 
   void SetDexClassDefIndex(uint16_t class_def_idx) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     // Not called within a transaction.
-    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(Class, dex_class_def_idx_), class_def_idx, false);
+    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(Class, dex_class_def_idx_), class_def_idx);
   }
 
   uint16_t GetDexTypeIndex() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    return GetField32(OFFSET_OF_OBJECT_MEMBER(Class, dex_type_idx_), false);
+    return GetField32(OFFSET_OF_OBJECT_MEMBER(Class, dex_type_idx_));
   }
 
   void SetDexTypeIndex(uint16_t type_idx) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     // Not called within a transaction.
-    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(Class, dex_type_idx_), type_idx, false);
+    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(Class, dex_type_idx_), type_idx);
   }
 
   static Class* GetJavaLangClass() {
diff --git a/runtime/mirror/class_loader.h b/runtime/mirror/class_loader.h
index 69accf5..74dae38 100644
--- a/runtime/mirror/class_loader.h
+++ b/runtime/mirror/class_loader.h
@@ -17,9 +17,6 @@
 #ifndef ART_RUNTIME_MIRROR_CLASS_LOADER_H_
 #define ART_RUNTIME_MIRROR_CLASS_LOADER_H_
 
-#include <vector>
-
-#include "dex_file.h"
 #include "mirror/object.h"
 
 namespace art {
diff --git a/runtime/mirror/dex_cache.cc b/runtime/mirror/dex_cache.cc
index 0a77db3..d6c11e8 100644
--- a/runtime/mirror/dex_cache.cc
+++ b/runtime/mirror/dex_cache.cc
@@ -44,12 +44,12 @@
   CHECK(resolved_methods != nullptr);
   CHECK(resolved_fields != nullptr);
 
-  SetFieldPtr<false>(OFFSET_OF_OBJECT_MEMBER(DexCache, dex_file_), dex_file, false);
-  SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(DexCache, location_), location, false);
-  SetFieldObject<false>(StringsOffset(), strings, false);
-  SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(DexCache, resolved_types_), resolved_types, false);
-  SetFieldObject<false>(ResolvedMethodsOffset(), resolved_methods, false);
-  SetFieldObject<false>(ResolvedFieldsOffset(), resolved_fields, false);
+  SetFieldPtr<false>(OFFSET_OF_OBJECT_MEMBER(DexCache, dex_file_), dex_file);
+  SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(DexCache, location_), location);
+  SetFieldObject<false>(StringsOffset(), strings);
+  SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(DexCache, resolved_types_), resolved_types);
+  SetFieldObject<false>(ResolvedMethodsOffset(), resolved_methods);
+  SetFieldObject<false>(ResolvedFieldsOffset(), resolved_fields);
 
   Runtime* runtime = Runtime::Current();
   if (runtime->HasResolutionMethod()) {
diff --git a/runtime/mirror/dex_cache.h b/runtime/mirror/dex_cache.h
index 843f860..11a4002 100644
--- a/runtime/mirror/dex_cache.h
+++ b/runtime/mirror/dex_cache.h
@@ -53,7 +53,7 @@
   void Fixup(ArtMethod* trampoline) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   String* GetLocation() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    return GetFieldObject<String>(OFFSET_OF_OBJECT_MEMBER(DexCache, location_), false);
+    return GetFieldObject<String>(OFFSET_OF_OBJECT_MEMBER(DexCache, location_));
   }
 
   static MemberOffset StringsOffset() {
@@ -88,17 +88,18 @@
     return GetStrings()->Get(string_idx);
   }
 
-  void SetResolvedString(uint32_t string_idx, String* resolved)
+  void SetResolvedString(uint32_t string_idx, String* resolved) ALWAYS_INLINE
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     // TODO default transaction support.
     GetStrings()->Set(string_idx, resolved);
   }
 
-  Class* GetResolvedType(uint32_t type_idx) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+  Class* GetResolvedType(uint32_t type_idx) ALWAYS_INLINE
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     return GetResolvedTypes()->Get(type_idx);
   }
 
-  void SetResolvedType(uint32_t type_idx, Class* resolved)
+  void SetResolvedType(uint32_t type_idx, Class* resolved) ALWAYS_INLINE
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     // TODO default transaction support.
     GetResolvedTypes()->Set(type_idx, resolved);
@@ -106,43 +107,47 @@
 
   ArtMethod* GetResolvedMethod(uint32_t method_idx) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
-  void SetResolvedMethod(uint32_t method_idx, ArtMethod* resolved)
+  void SetResolvedMethod(uint32_t method_idx, ArtMethod* resolved) ALWAYS_INLINE
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     GetResolvedMethods()->Set(method_idx, resolved);
   }
 
-  ArtField* GetResolvedField(uint32_t field_idx) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+  ArtField* GetResolvedField(uint32_t field_idx) ALWAYS_INLINE
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     return GetResolvedFields()->Get(field_idx);
   }
 
-  void SetResolvedField(uint32_t field_idx, ArtField* resolved)
+  void SetResolvedField(uint32_t field_idx, ArtField* resolved) ALWAYS_INLINE
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     GetResolvedFields()->Set(field_idx, resolved);
   }
 
-  ObjectArray<String>* GetStrings() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    return GetFieldObject< ObjectArray<String> >(StringsOffset(), false);
+  ObjectArray<String>* GetStrings() ALWAYS_INLINE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+    return GetFieldObject< ObjectArray<String> >(StringsOffset());
   }
 
-  ObjectArray<Class>* GetResolvedTypes() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+  ObjectArray<Class>* GetResolvedTypes() ALWAYS_INLINE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     return GetFieldObject<ObjectArray<Class> >(
-        OFFSET_OF_OBJECT_MEMBER(DexCache, resolved_types_), false);
+        OFFSET_OF_OBJECT_MEMBER(DexCache, resolved_types_));
   }
 
-  ObjectArray<ArtMethod>* GetResolvedMethods() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    return GetFieldObject< ObjectArray<ArtMethod> >(ResolvedMethodsOffset(), false);
+  ObjectArray<ArtMethod>* GetResolvedMethods() ALWAYS_INLINE
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+    return GetFieldObject< ObjectArray<ArtMethod> >(ResolvedMethodsOffset());
   }
 
-  ObjectArray<ArtField>* GetResolvedFields() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    return GetFieldObject<ObjectArray<ArtField> >(ResolvedFieldsOffset(), false);
+  ObjectArray<ArtField>* GetResolvedFields() ALWAYS_INLINE
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+    return GetFieldObject<ObjectArray<ArtField> >(ResolvedFieldsOffset());
   }
 
-  const DexFile* GetDexFile() {
-    return GetFieldPtr<const DexFile*>(OFFSET_OF_OBJECT_MEMBER(DexCache, dex_file_), false);
+  const DexFile* GetDexFile() ALWAYS_INLINE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+    return GetFieldPtr<const DexFile*>(OFFSET_OF_OBJECT_MEMBER(DexCache, dex_file_));
   }
 
-  void SetDexFile(const DexFile* dex_file) {
-    return SetFieldPtr<false>(OFFSET_OF_OBJECT_MEMBER(DexCache, dex_file_), dex_file, false);
+  void SetDexFile(const DexFile* dex_file) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_)
+      ALWAYS_INLINE {
+    return SetFieldPtr<false>(OFFSET_OF_OBJECT_MEMBER(DexCache, dex_file_), dex_file);
   }
 
  private:
diff --git a/runtime/mirror/iftable.h b/runtime/mirror/iftable.h
index bb4cd41..ad312ed 100644
--- a/runtime/mirror/iftable.h
+++ b/runtime/mirror/iftable.h
@@ -17,6 +17,7 @@
 #ifndef ART_RUNTIME_MIRROR_IFTABLE_H_
 #define ART_RUNTIME_MIRROR_IFTABLE_H_
 
+#include "base/casts.h"
 #include "object_array.h"
 
 namespace art {
diff --git a/runtime/mirror/object-inl.h b/runtime/mirror/object-inl.h
index 04517ec..c70a08d 100644
--- a/runtime/mirror/object-inl.h
+++ b/runtime/mirror/object-inl.h
@@ -37,7 +37,7 @@
 template<VerifyObjectFlags kVerifyFlags, bool kDoReadBarrier>
 inline Class* Object::GetClass() {
   return GetFieldObject<Class, kVerifyFlags, kDoReadBarrier>(
-      OFFSET_OF_OBJECT_MEMBER(Object, klass_), false);
+      OFFSET_OF_OBJECT_MEMBER(Object, klass_));
 }
 
 template<VerifyObjectFlags kVerifyFlags>
@@ -49,17 +49,23 @@
   // we may run in transaction mode here.
   SetFieldObjectWithoutWriteBarrier<false, false,
       static_cast<VerifyObjectFlags>(kVerifyFlags & ~kVerifyThis)>(
-      OFFSET_OF_OBJECT_MEMBER(Object, klass_), new_klass, false);
+      OFFSET_OF_OBJECT_MEMBER(Object, klass_), new_klass);
 }
 
 inline LockWord Object::GetLockWord(bool as_volatile) {
-  return LockWord(GetField32(OFFSET_OF_OBJECT_MEMBER(Object, monitor_), as_volatile));
+  if (as_volatile) {
+    return LockWord(GetField32Volatile(OFFSET_OF_OBJECT_MEMBER(Object, monitor_)));
+  }
+  return LockWord(GetField32(OFFSET_OF_OBJECT_MEMBER(Object, monitor_)));
 }
 
 inline void Object::SetLockWord(LockWord new_val, bool as_volatile) {
   // Force use of non-transactional mode and do not check.
-  SetField32<false, false>(OFFSET_OF_OBJECT_MEMBER(Object, monitor_), new_val.GetValue(),
-                           as_volatile);
+  if (as_volatile) {
+    SetField32Volatile<false, false>(OFFSET_OF_OBJECT_MEMBER(Object, monitor_), new_val.GetValue());
+  } else {
+    SetField32<false, false>(OFFSET_OF_OBJECT_MEMBER(Object, monitor_), new_val.GetValue());
+  }
 }
 
 inline bool Object::CasLockWord(LockWord old_val, LockWord new_val) {
@@ -387,14 +393,14 @@
   return result;
 }
 
-template<VerifyObjectFlags kVerifyFlags>
-inline int32_t Object::GetField32(MemberOffset field_offset, bool is_volatile) {
+template<VerifyObjectFlags kVerifyFlags, bool kIsVolatile>
+inline int32_t Object::GetField32(MemberOffset field_offset) {
   if (kVerifyFlags & kVerifyThis) {
     VerifyObject(this);
   }
   const byte* raw_addr = reinterpret_cast<const byte*>(this) + field_offset.Int32Value();
   const int32_t* word_addr = reinterpret_cast<const int32_t*>(raw_addr);
-  if (UNLIKELY(is_volatile)) {
+  if (UNLIKELY(kIsVolatile)) {
     int32_t result = *(reinterpret_cast<volatile int32_t*>(const_cast<int32_t*>(word_addr)));
     QuasiAtomic::MembarLoadLoad();  // Ensure volatile loads don't re-order.
     return result;
@@ -403,21 +409,28 @@
   }
 }
 
-template<bool kTransactionActive, bool kCheckTransaction, VerifyObjectFlags kVerifyFlags>
-inline void Object::SetField32(MemberOffset field_offset, int32_t new_value, bool is_volatile) {
+template<VerifyObjectFlags kVerifyFlags>
+inline int32_t Object::GetField32Volatile(MemberOffset field_offset) {
+  return GetField32<kVerifyFlags, true>(field_offset);
+}
+
+template<bool kTransactionActive, bool kCheckTransaction, VerifyObjectFlags kVerifyFlags,
+    bool kIsVolatile>
+inline void Object::SetField32(MemberOffset field_offset, int32_t new_value) {
   if (kCheckTransaction) {
     DCHECK_EQ(kTransactionActive, Runtime::Current()->IsActiveTransaction());
   }
   if (kTransactionActive) {
-    Runtime::Current()->RecordWriteField32(this, field_offset, GetField32(field_offset, is_volatile),
-                                           is_volatile);
+    Runtime::Current()->RecordWriteField32(this, field_offset,
+                                           GetField32<kVerifyFlags, kIsVolatile>(field_offset),
+                                           kIsVolatile);
   }
   if (kVerifyFlags & kVerifyThis) {
     VerifyObject(this);
   }
   byte* raw_addr = reinterpret_cast<byte*>(this) + field_offset.Int32Value();
   int32_t* word_addr = reinterpret_cast<int32_t*>(raw_addr);
-  if (UNLIKELY(is_volatile)) {
+  if (kIsVolatile) {
     QuasiAtomic::MembarStoreStore();  // Ensure this store occurs after others in the queue.
     *word_addr = new_value;
     QuasiAtomic::MembarStoreLoad();  // Ensure this store occurs before any volatile loads.
@@ -427,6 +440,11 @@
 }
 
 template<bool kTransactionActive, bool kCheckTransaction, VerifyObjectFlags kVerifyFlags>
+inline void Object::SetField32Volatile(MemberOffset field_offset, int32_t new_value) {
+  SetField32<kTransactionActive, kCheckTransaction, kVerifyFlags, true>(field_offset, new_value);
+}
+
+template<bool kTransactionActive, bool kCheckTransaction, VerifyObjectFlags kVerifyFlags>
 inline bool Object::CasField32(MemberOffset field_offset, int32_t old_value, int32_t new_value) {
   if (kCheckTransaction) {
     DCHECK_EQ(kTransactionActive, Runtime::Current()->IsActiveTransaction());
@@ -442,14 +460,14 @@
   return __sync_bool_compare_and_swap(addr, old_value, new_value);
 }
 
-template<VerifyObjectFlags kVerifyFlags>
-inline int64_t Object::GetField64(MemberOffset field_offset, bool is_volatile) {
+template<VerifyObjectFlags kVerifyFlags, bool kIsVolatile>
+inline int64_t Object::GetField64(MemberOffset field_offset) {
   if (kVerifyFlags & kVerifyThis) {
     VerifyObject(this);
   }
   const byte* raw_addr = reinterpret_cast<const byte*>(this) + field_offset.Int32Value();
   const int64_t* addr = reinterpret_cast<const int64_t*>(raw_addr);
-  if (UNLIKELY(is_volatile)) {
+  if (kIsVolatile) {
     int64_t result = QuasiAtomic::Read64(addr);
     QuasiAtomic::MembarLoadLoad();  // Ensure volatile loads don't re-order.
     return result;
@@ -458,21 +476,28 @@
   }
 }
 
-template<bool kTransactionActive, bool kCheckTransaction, VerifyObjectFlags kVerifyFlags>
-inline void Object::SetField64(MemberOffset field_offset, int64_t new_value, bool is_volatile) {
+template<VerifyObjectFlags kVerifyFlags>
+inline int64_t Object::GetField64Volatile(MemberOffset field_offset) {
+  return GetField64<kVerifyFlags, true>(field_offset);
+}
+
+template<bool kTransactionActive, bool kCheckTransaction, VerifyObjectFlags kVerifyFlags,
+    bool kIsVolatile>
+inline void Object::SetField64(MemberOffset field_offset, int64_t new_value) {
   if (kCheckTransaction) {
     DCHECK_EQ(kTransactionActive, Runtime::Current()->IsActiveTransaction());
   }
   if (kTransactionActive) {
-    Runtime::Current()->RecordWriteField64(this, field_offset, GetField64(field_offset, is_volatile),
-                                           is_volatile);
+    Runtime::Current()->RecordWriteField64(this, field_offset,
+                                           GetField64<kVerifyFlags, kIsVolatile>(field_offset),
+                                           kIsVolatile);
   }
   if (kVerifyFlags & kVerifyThis) {
     VerifyObject(this);
   }
   byte* raw_addr = reinterpret_cast<byte*>(this) + field_offset.Int32Value();
   int64_t* addr = reinterpret_cast<int64_t*>(raw_addr);
-  if (UNLIKELY(is_volatile)) {
+  if (kIsVolatile) {
     QuasiAtomic::MembarStoreStore();  // Ensure this store occurs after others in the queue.
     QuasiAtomic::Write64(addr, new_value);
     if (!QuasiAtomic::LongAtomicsUseMutexes()) {
@@ -486,6 +511,12 @@
 }
 
 template<bool kTransactionActive, bool kCheckTransaction, VerifyObjectFlags kVerifyFlags>
+inline void Object::SetField64Volatile(MemberOffset field_offset, int64_t new_value) {
+  return SetField64<kTransactionActive, kCheckTransaction, kVerifyFlags, true>(field_offset,
+                                                                               new_value);
+}
+
+template<bool kTransactionActive, bool kCheckTransaction, VerifyObjectFlags kVerifyFlags>
 inline bool Object::CasField64(MemberOffset field_offset, int64_t old_value, int64_t new_value) {
   if (kCheckTransaction) {
     DCHECK_EQ(kTransactionActive, Runtime::Current()->IsActiveTransaction());
@@ -501,15 +532,15 @@
   return QuasiAtomic::Cas64(old_value, new_value, addr);
 }
 
-template<class T, VerifyObjectFlags kVerifyFlags, bool kDoReadBarrier>
-inline T* Object::GetFieldObject(MemberOffset field_offset, bool is_volatile) {
+template<class T, VerifyObjectFlags kVerifyFlags, bool kDoReadBarrier, bool kIsVolatile>
+inline T* Object::GetFieldObject(MemberOffset field_offset) {
   if (kVerifyFlags & kVerifyThis) {
     VerifyObject(this);
   }
   byte* raw_addr = reinterpret_cast<byte*>(this) + field_offset.Int32Value();
   HeapReference<T>* objref_addr = reinterpret_cast<HeapReference<T>*>(raw_addr);
   T* result = ReadBarrier::Barrier<T, kDoReadBarrier>(this, field_offset, objref_addr);
-  if (UNLIKELY(is_volatile)) {
+  if (kIsVolatile) {
     QuasiAtomic::MembarLoadLoad();  // Ensure loads don't re-order.
   }
   if (kVerifyFlags & kVerifyReads) {
@@ -518,16 +549,26 @@
   return result;
 }
 
-template<bool kTransactionActive, bool kCheckTransaction, VerifyObjectFlags kVerifyFlags>
-inline void Object::SetFieldObjectWithoutWriteBarrier(MemberOffset field_offset, Object* new_value,
-                                                      bool is_volatile) {
+template<class T, VerifyObjectFlags kVerifyFlags, bool kDoReadBarrier>
+inline T* Object::GetFieldObjectVolatile(MemberOffset field_offset) {
+  return GetFieldObject<T, kVerifyFlags, kDoReadBarrier, true>(field_offset);
+}
+
+template<bool kTransactionActive, bool kCheckTransaction, VerifyObjectFlags kVerifyFlags,
+    bool kIsVolatile>
+inline void Object::SetFieldObjectWithoutWriteBarrier(MemberOffset field_offset,
+                                                      Object* new_value) {
   if (kCheckTransaction) {
     DCHECK_EQ(kTransactionActive, Runtime::Current()->IsActiveTransaction());
   }
   if (kTransactionActive) {
-    Runtime::Current()->RecordWriteFieldReference(this, field_offset,
-                                                  GetFieldObject<Object>(field_offset, is_volatile),
-                                                  true);
+    mirror::Object* obj;
+    if (kIsVolatile) {
+      obj = GetFieldObjectVolatile<Object>(field_offset);
+    } else {
+      obj = GetFieldObject<Object>(field_offset);
+    }
+    Runtime::Current()->RecordWriteFieldReference(this, field_offset, obj, true);
   }
   if (kVerifyFlags & kVerifyThis) {
     VerifyObject(this);
@@ -537,7 +578,7 @@
   }
   byte* raw_addr = reinterpret_cast<byte*>(this) + field_offset.Int32Value();
   HeapReference<Object>* objref_addr = reinterpret_cast<HeapReference<Object>*>(raw_addr);
-  if (UNLIKELY(is_volatile)) {
+  if (kIsVolatile) {
     QuasiAtomic::MembarStoreStore();  // Ensure this store occurs after others in the queue.
     objref_addr->Assign(new_value);
     QuasiAtomic::MembarStoreLoad();  // Ensure this store occurs before any loads.
@@ -546,16 +587,23 @@
   }
 }
 
-template<bool kTransactionActive, bool kCheckTransaction, VerifyObjectFlags kVerifyFlags>
-inline void Object::SetFieldObject(MemberOffset field_offset, Object* new_value, bool is_volatile) {
-  SetFieldObjectWithoutWriteBarrier<kTransactionActive, kCheckTransaction, kVerifyFlags>(
-      field_offset, new_value, is_volatile);
+template<bool kTransactionActive, bool kCheckTransaction, VerifyObjectFlags kVerifyFlags,
+    bool kIsVolatile>
+inline void Object::SetFieldObject(MemberOffset field_offset, Object* new_value) {
+  SetFieldObjectWithoutWriteBarrier<kTransactionActive, kCheckTransaction, kVerifyFlags,
+      kIsVolatile>(field_offset, new_value);
   if (new_value != nullptr) {
     CheckFieldAssignment(field_offset, new_value);
     Runtime::Current()->GetHeap()->WriteBarrierField(this, field_offset, new_value);
   }
 }
 
+template<bool kTransactionActive, bool kCheckTransaction, VerifyObjectFlags kVerifyFlags>
+inline void Object::SetFieldObjectVolatile(MemberOffset field_offset, Object* new_value) {
+  SetFieldObject<kTransactionActive, kCheckTransaction, kVerifyFlags, true>(field_offset,
+                                                                            new_value);
+}
+
 template <VerifyObjectFlags kVerifyFlags>
 inline HeapReference<Object>* Object::GetFieldObjectReferenceAddr(MemberOffset field_offset) {
   if (kVerifyFlags & kVerifyThis) {
@@ -618,8 +666,7 @@
       size_t num_reference_fields =
           kIsStatic ? klass->NumReferenceStaticFields() : klass->NumReferenceInstanceFields();
       for (size_t i = 0; i < num_reference_fields; ++i) {
-        mirror::ArtField* field = kIsStatic ? klass->GetStaticField(i)
-            : klass->GetInstanceField(i);
+        mirror::ArtField* field = kIsStatic ? klass->GetStaticField(i) : klass->GetInstanceField(i);
         MemberOffset field_offset = field->GetOffset();
         // TODO: Do a simpler check?
         if (!kVisitClass && UNLIKELY(field_offset.Uint32Value() == ClassOffset().Uint32Value())) {
@@ -648,14 +695,16 @@
 inline void Object::VisitReferences(const Visitor& visitor,
                                     const JavaLangRefVisitor& ref_visitor) {
   mirror::Class* klass = GetClass<kVerifyFlags>();
-  if (UNLIKELY(klass == Class::GetJavaLangClass())) {
-    DCHECK_EQ(klass->GetClass<kVerifyNone>(), Class::GetJavaLangClass());
-    AsClass<kVerifyNone>()->VisitReferences<kVisitClass>(klass, visitor);
-  } else if (UNLIKELY(klass->IsArrayClass<kVerifyFlags>())) {
-    if (klass->IsObjectArrayClass<kVerifyNone>()) {
-      AsObjectArray<mirror::Object, kVerifyNone>()->VisitReferences<kVisitClass>(visitor);
-    } else if (kVisitClass) {
-      visitor(this, ClassOffset(), false);
+  if (klass->IsVariableSize()) {
+    if (klass->IsClassClass()) {
+      AsClass<kVerifyNone>()->VisitReferences<kVisitClass>(klass, visitor);
+    } else {
+      DCHECK(klass->IsArrayClass<kVerifyFlags>());
+      if (klass->IsObjectArrayClass<kVerifyNone>()) {
+        AsObjectArray<mirror::Object, kVerifyNone>()->VisitReferences<kVisitClass>(visitor);
+      } else if (kVisitClass) {
+        visitor(this, ClassOffset(), false);
+      }
     }
   } else {
     VisitInstanceFieldsReferences<kVisitClass>(klass, visitor);
diff --git a/runtime/mirror/object.cc b/runtime/mirror/object.cc
index 766bbc9..2cd71a0 100644
--- a/runtime/mirror/object.cc
+++ b/runtime/mirror/object.cc
@@ -47,10 +47,10 @@
   void operator()(Object* obj, MemberOffset offset, bool /* is_static */) const
       ALWAYS_INLINE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     // GetFieldObject() contains a RB.
-    Object* ref = obj->GetFieldObject<Object>(offset, false);
+    Object* ref = obj->GetFieldObject<Object>(offset);
     // No WB here as a large object space does not have a card table
     // coverage. Instead, cards will be marked separately.
-    dest_obj_->SetFieldObjectWithoutWriteBarrier<false, false>(offset, ref, false);
+    dest_obj_->SetFieldObjectWithoutWriteBarrier<false, false>(offset, ref);
   }
 
   void operator()(mirror::Class* klass, mirror::Reference* ref) const
diff --git a/runtime/mirror/object.h b/runtime/mirror/object.h
index 370b3b8..cf28b18 100644
--- a/runtime/mirror/object.h
+++ b/runtime/mirror/object.h
@@ -17,23 +17,18 @@
 #ifndef ART_RUNTIME_MIRROR_OBJECT_H_
 #define ART_RUNTIME_MIRROR_OBJECT_H_
 
-#include "base/casts.h"
-#include "base/logging.h"
-#include "base/macros.h"
-#include "cutils/atomic-inline.h"
-#include "monitor.h"
 #include "object_reference.h"
 #include "offsets.h"
-#include "runtime.h"
 #include "verify_object.h"
 
 namespace art {
 
 class ImageWriter;
 class LockWord;
+class Monitor;
 struct ObjectOffsets;
 class Thread;
-template <typename T> class SirtRef;
+class VoidFunctor;
 
 namespace mirror {
 
@@ -73,7 +68,7 @@
   }
 
   template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags, bool kDoReadBarrier = true>
-  Class* GetClass() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  Class* GetClass() ALWAYS_INLINE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
   void SetClass(Class* new_klass) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
@@ -104,8 +99,8 @@
 
   // As volatile can be false if the mutators are suspended. This is an optimization since it
   // avoids the barriers.
-  LockWord GetLockWord(bool as_volatile);
-  void SetLockWord(LockWord new_val, bool as_volatile);
+  LockWord GetLockWord(bool as_volatile) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  void SetLockWord(LockWord new_val, bool as_volatile) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
   bool CasLockWord(LockWord old_val, LockWord new_val) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
   uint32_t GetLockOwnerThreadId();
 
@@ -189,18 +184,31 @@
   bool IsPhantomReferenceInstance() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   // Accessor for Java type fields.
-  template<class T, VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags, bool kDoReadBarrier = true>
-  T* GetFieldObject(MemberOffset field_offset, bool is_volatile)
+  template<class T, VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags,
+      bool kDoReadBarrier = true, bool kIsVolatile = false>
+  T* GetFieldObject(MemberOffset field_offset) ALWAYS_INLINE
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
+  template<class T, VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags,
+      bool kDoReadBarrier = true>
+  T* GetFieldObjectVolatile(MemberOffset field_offset) ALWAYS_INLINE
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
+  template<bool kTransactionActive, bool kCheckTransaction = true,
+      VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags, bool kIsVolatile = false>
+  void SetFieldObjectWithoutWriteBarrier(MemberOffset field_offset, Object* new_value)
+      ALWAYS_INLINE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
+  template<bool kTransactionActive, bool kCheckTransaction = true,
+      VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags, bool kIsVolatile = false>
+  void SetFieldObject(MemberOffset field_offset, Object* new_value) ALWAYS_INLINE
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
   template<bool kTransactionActive, bool kCheckTransaction = true,
       VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
-  void SetFieldObjectWithoutWriteBarrier(MemberOffset field_offset, Object* new_value,
-                                         bool is_volatile)
+  void SetFieldObjectVolatile(MemberOffset field_offset, Object* new_value) ALWAYS_INLINE
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
-  template<bool kTransactionActive, bool kCheckTransaction = true,
-      VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
-  void SetFieldObject(MemberOffset field_offset, Object* new_value, bool is_volatile)
-      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
   template<bool kTransactionActive, bool kCheckTransaction = true,
       VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
   bool CasFieldObject(MemberOffset field_offset, Object* old_value, Object* new_value)
@@ -209,23 +217,46 @@
   template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
   HeapReference<Object>* GetFieldObjectReferenceAddr(MemberOffset field_offset);
 
-  template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
-  int32_t GetField32(MemberOffset field_offset, bool is_volatile)
-      NO_THREAD_SAFETY_ANALYSIS;
-
-  template<bool kTransactionActive, bool kCheckTransaction = true,
-      VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
-  void SetField32(MemberOffset field_offset, int32_t new_value, bool is_volatile);
-  template<bool kTransactionActive, bool kCheckTransaction = true,
-      VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
-  bool CasField32(MemberOffset field_offset, int32_t old_value, int32_t new_value)
+  template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags, bool kIsVolatile = false>
+  int32_t GetField32(MemberOffset field_offset) ALWAYS_INLINE
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
-  int64_t GetField64(MemberOffset field_offset, bool is_volatile);
+  int32_t GetField32Volatile(MemberOffset field_offset) ALWAYS_INLINE
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
+  template<bool kTransactionActive, bool kCheckTransaction = true,
+      VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags, bool kIsVolatile = false>
+  void SetField32(MemberOffset field_offset, int32_t new_value) ALWAYS_INLINE
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
   template<bool kTransactionActive, bool kCheckTransaction = true,
       VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
-  void SetField64(MemberOffset field_offset, int64_t new_value, bool is_volatile);
+  void SetField32Volatile(MemberOffset field_offset, int32_t new_value) ALWAYS_INLINE
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
+  template<bool kTransactionActive, bool kCheckTransaction = true,
+      VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
+  bool CasField32(MemberOffset field_offset, int32_t old_value, int32_t new_value) ALWAYS_INLINE
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
+  template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags, bool kIsVolatile = false>
+  int64_t GetField64(MemberOffset field_offset) ALWAYS_INLINE
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
+  template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
+  int64_t GetField64Volatile(MemberOffset field_offset) ALWAYS_INLINE
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
+  template<bool kTransactionActive, bool kCheckTransaction = true,
+      VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags, bool kIsVolatile = false>
+  void SetField64(MemberOffset field_offset, int64_t new_value) ALWAYS_INLINE
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
+  template<bool kTransactionActive, bool kCheckTransaction = true,
+      VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
+  void SetField64Volatile(MemberOffset field_offset, int64_t new_value) ALWAYS_INLINE
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   template<bool kTransactionActive, bool kCheckTransaction = true,
       VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
@@ -234,13 +265,14 @@
 
   template<bool kTransactionActive, bool kCheckTransaction = true,
       VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags, typename T>
-  void SetFieldPtr(MemberOffset field_offset, T new_value, bool is_volatile) {
+  void SetFieldPtr(MemberOffset field_offset, T new_value)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
 #ifndef __LP64__
     SetField32<kTransactionActive, kCheckTransaction, kVerifyFlags>(
-        field_offset, reinterpret_cast<int32_t>(new_value), is_volatile);
+        field_offset, reinterpret_cast<int32_t>(new_value));
 #else
     SetField64<kTransactionActive, kCheckTransaction, kVerifyFlags>(
-        field_offset, reinterpret_cast<int64_t>(new_value), is_volatile);
+        field_offset, reinterpret_cast<int64_t>(new_value));
 #endif
   }
 
@@ -248,30 +280,30 @@
   // SHARED_LOCKS_REQUIRED(Locks::mutator_lock_).
   template <const bool kVisitClass, VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags,
       typename Visitor, typename JavaLangRefVisitor = VoidFunctor>
-  void VisitReferences(const Visitor& visitor,
-                       const JavaLangRefVisitor& ref_visitor = VoidFunctor())
+  void VisitReferences(const Visitor& visitor, const JavaLangRefVisitor& ref_visitor)
       NO_THREAD_SAFETY_ANALYSIS;
 
  protected:
   // Accessors for non-Java type fields
-  template<class T, VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
-  T GetFieldPtr(MemberOffset field_offset, bool is_volatile) NO_THREAD_SAFETY_ANALYSIS {
+  template<class T, VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags, bool kIsVolatile = false>
+  T GetFieldPtr(MemberOffset field_offset)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
 #ifndef __LP64__
-    return reinterpret_cast<T>(GetField32<kVerifyFlags>(field_offset, is_volatile));
+    return reinterpret_cast<T>(GetField32<kVerifyFlags, kIsVolatile>(field_offset));
 #else
-    return reinterpret_cast<T>(GetField64<kVerifyFlags>(field_offset, is_volatile));
+    return reinterpret_cast<T>(GetField64<kVerifyFlags, kIsVolatile>(field_offset));
 #endif
   }
 
   // TODO: Fixme when anotatalysis works with visitors.
   template<bool kVisitClass, bool kIsStatic, typename Visitor>
-  void VisitFieldsReferences(uint32_t ref_offsets, const Visitor& visitor)
+  void VisitFieldsReferences(uint32_t ref_offsets, const Visitor& visitor) HOT_ATTR
       NO_THREAD_SAFETY_ANALYSIS;
   template<bool kVisitClass, typename Visitor>
-  void VisitInstanceFieldsReferences(mirror::Class* klass, const Visitor& visitor)
+  void VisitInstanceFieldsReferences(mirror::Class* klass, const Visitor& visitor) HOT_ATTR
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
   template<bool kVisitClass, typename Visitor>
-  void VisitStaticFieldsReferences(mirror::Class* klass, const Visitor& visitor)
+  void VisitStaticFieldsReferences(mirror::Class* klass, const Visitor& visitor) HOT_ATTR
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
  private:
diff --git a/runtime/mirror/object_array-inl.h b/runtime/mirror/object_array-inl.h
index e0c14c3..203a6b2 100644
--- a/runtime/mirror/object_array-inl.h
+++ b/runtime/mirror/object_array-inl.h
@@ -51,11 +51,11 @@
 
 template<class T>
 inline T* ObjectArray<T>::Get(int32_t i) {
-  if (UNLIKELY(!CheckIsValidIndex(i))) {
+  if (!CheckIsValidIndex(i)) {
     DCHECK(Thread::Current()->IsExceptionPending());
     return NULL;
   }
-  return GetFieldObject<T>(OffsetOfElement(i), false);
+  return GetFieldObject<T>(OffsetOfElement(i));
 }
 
 template<class T> template<VerifyObjectFlags kVerifyFlags>
@@ -82,9 +82,8 @@
 template<class T>
 template<bool kTransactionActive, bool kCheckTransaction, VerifyObjectFlags kVerifyFlags>
 inline void ObjectArray<T>::Set(int32_t i, T* object) {
-  if (LIKELY(CheckIsValidIndex(i) && CheckAssignable<kVerifyFlags>(object))) {
-    SetFieldObject<kTransactionActive, kCheckTransaction, kVerifyFlags>(OffsetOfElement(i), object,
-                                                                        false);
+  if (CheckIsValidIndex(i) && CheckAssignable<kVerifyFlags>(object)) {
+    SetFieldObject<kTransactionActive, kCheckTransaction, kVerifyFlags>(OffsetOfElement(i), object);
   } else {
     DCHECK(Thread::Current()->IsExceptionPending());
   }
@@ -95,8 +94,7 @@
 inline void ObjectArray<T>::SetWithoutChecks(int32_t i, T* object) {
   DCHECK(CheckIsValidIndex<kVerifyFlags>(i));
   DCHECK(CheckAssignable<static_cast<VerifyObjectFlags>(kVerifyFlags & ~kVerifyThis)>(object));
-  SetFieldObject<kTransactionActive, kCheckTransaction, kVerifyFlags>(OffsetOfElement(i), object,
-                                                                      false);
+  SetFieldObject<kTransactionActive, kCheckTransaction, kVerifyFlags>(OffsetOfElement(i), object);
 }
 
 template<class T>
@@ -106,13 +104,13 @@
   // TODO:  enable this check. It fails when writing the image in ImageWriter::FixupObjectArray.
   // DCHECK(CheckAssignable(object));
   SetFieldObjectWithoutWriteBarrier<kTransactionActive, kCheckTransaction, kVerifyFlags>(
-      OffsetOfElement(i), object, false);
+      OffsetOfElement(i), object);
 }
 
 template<class T>
 inline T* ObjectArray<T>::GetWithoutChecks(int32_t i) {
   DCHECK(CheckIsValidIndex(i));
-  return GetFieldObject<T>(OffsetOfElement(i), false);
+  return GetFieldObject<T>(OffsetOfElement(i));
 }
 
 template<class T>
diff --git a/runtime/mirror/object_array.h b/runtime/mirror/object_array.h
index 26b1fd1..54d1240 100644
--- a/runtime/mirror/object_array.h
+++ b/runtime/mirror/object_array.h
@@ -18,7 +18,6 @@
 #define ART_RUNTIME_MIRROR_OBJECT_ARRAY_H_
 
 #include "array.h"
-#include "gc/heap.h"
 
 namespace art {
 namespace mirror {
@@ -33,7 +32,7 @@
   static ObjectArray<T>* Alloc(Thread* self, Class* object_array_class, int32_t length)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
-  T* Get(int32_t i) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  T* Get(int32_t i) ALWAYS_INLINE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   // Returns true if the object can be stored into the array. If not, throws
   // an ArrayStoreException and returns false.
@@ -41,11 +40,11 @@
   template<VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
   bool CheckAssignable(T* object) NO_THREAD_SAFETY_ANALYSIS;
 
-  void Set(int32_t i, T* object) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  void Set(int32_t i, T* object) ALWAYS_INLINE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
   // TODO fix thread safety analysis: should be SHARED_LOCKS_REQUIRED(Locks::mutator_lock_).
   template<bool kTransactionActive, bool kCheckTransaction = true,
       VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
-  void Set(int32_t i, T* object) NO_THREAD_SAFETY_ANALYSIS;
+  void Set(int32_t i, T* object) ALWAYS_INLINE NO_THREAD_SAFETY_ANALYSIS;
 
   // Set element without bound and element type checks, to be used in limited
   // circumstances, such as during boot image writing.
@@ -53,14 +52,15 @@
   // SHARED_LOCKS_REQUIRED(Locks::mutator_lock_).
   template<bool kTransactionActive, bool kCheckTransaction = true,
       VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
-  void SetWithoutChecks(int32_t i, T* object) NO_THREAD_SAFETY_ANALYSIS;
+  void SetWithoutChecks(int32_t i, T* object) ALWAYS_INLINE NO_THREAD_SAFETY_ANALYSIS;
   // TODO fix thread safety analysis broken by the use of template. This should be
   // SHARED_LOCKS_REQUIRED(Locks::mutator_lock_).
   template<bool kTransactionActive, bool kCheckTransaction = true,
       VerifyObjectFlags kVerifyFlags = kDefaultVerifyFlags>
-  void SetWithoutChecksAndWriteBarrier(int32_t i, T* object) NO_THREAD_SAFETY_ANALYSIS;
+  void SetWithoutChecksAndWriteBarrier(int32_t i, T* object) ALWAYS_INLINE
+      NO_THREAD_SAFETY_ANALYSIS;
 
-  T* GetWithoutChecks(int32_t i) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  T* GetWithoutChecks(int32_t i) ALWAYS_INLINE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   // Copy src into this array (dealing with overlaps as memmove does) without assignability checks.
   void AssignableMemmove(int32_t dst_pos, ObjectArray<T>* src, int32_t src_pos,
diff --git a/runtime/mirror/object_reference.h b/runtime/mirror/object_reference.h
index 72f281d..b63d13d 100644
--- a/runtime/mirror/object_reference.h
+++ b/runtime/mirror/object_reference.h
@@ -17,7 +17,7 @@
 #ifndef ART_RUNTIME_MIRROR_OBJECT_REFERENCE_H_
 #define ART_RUNTIME_MIRROR_OBJECT_REFERENCE_H_
 
-#include "base/mutex.h"
+#include "base/mutex.h"  // For Locks::mutator_lock_.
 #include "globals.h"
 
 namespace art {
diff --git a/runtime/mirror/object_test.cc b/runtime/mirror/object_test.cc
index 32f30c3..c494f13 100644
--- a/runtime/mirror/object_test.cc
+++ b/runtime/mirror/object_test.cc
@@ -35,6 +35,7 @@
 #include "object-inl.h"
 #include "object_array-inl.h"
 #include "sirt_ref.h"
+#include "string-inl.h"
 #include "UniquePtr.h"
 
 namespace art {
diff --git a/runtime/mirror/proxy.h b/runtime/mirror/proxy.h
index ff019c6..5f9cceb 100644
--- a/runtime/mirror/proxy.h
+++ b/runtime/mirror/proxy.h
@@ -17,7 +17,7 @@
 #ifndef ART_RUNTIME_MIRROR_PROXY_H_
 #define ART_RUNTIME_MIRROR_PROXY_H_
 
-#include "mirror/object.h"
+#include "object.h"
 
 namespace art {
 
@@ -31,14 +31,12 @@
  public:
   ObjectArray<Class>* GetInterfaces() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     return GetFieldObject<ObjectArray<Class> >(OFFSET_OF_OBJECT_MEMBER(SynthesizedProxyClass,
-                                                                       interfaces_),
-                                               false);
+                                                                       interfaces_));
   }
 
   ObjectArray<ObjectArray<Class> >* GetThrows()  SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     return GetFieldObject<ObjectArray<ObjectArray<Class> > >(OFFSET_OF_OBJECT_MEMBER(SynthesizedProxyClass,
-                                                                                     throws_),
-                                               false);
+                                                                                     throws_));
   }
 
  private:
diff --git a/runtime/mirror/reference-inl.h b/runtime/mirror/reference-inl.h
index 0f76f77..43767c8 100644
--- a/runtime/mirror/reference-inl.h
+++ b/runtime/mirror/reference-inl.h
@@ -25,8 +25,8 @@
 inline bool Reference::IsEnqueuable() {
   // Not using volatile reads as an optimization since this is only called with all the mutators
   // suspended.
-  const Object* queue = GetFieldObject<mirror::Object>(QueueOffset(), false);
-  const Object* queue_next = GetFieldObject<mirror::Object>(QueueNextOffset(), false);
+  const Object* queue = GetFieldObject<mirror::Object>(QueueOffset());
+  const Object* queue_next = GetFieldObject<mirror::Object>(QueueNextOffset());
   return queue != nullptr && queue_next == nullptr;
 }
 
diff --git a/runtime/mirror/reference.h b/runtime/mirror/reference.h
index c2a83ff..cf65d20 100644
--- a/runtime/mirror/reference.h
+++ b/runtime/mirror/reference.h
@@ -43,26 +43,26 @@
   }
 
   Object* GetReferent() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    return GetFieldObject<Object>(ReferentOffset(), true);
+    return GetFieldObjectVolatile<Object>(ReferentOffset());
   }
   template<bool kTransactionActive>
   void SetReferent(Object* referent) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    SetFieldObject<kTransactionActive>(ReferentOffset(), referent, true);
+    SetFieldObjectVolatile<kTransactionActive>(ReferentOffset(), referent);
   }
   template<bool kTransactionActive>
   void ClearReferent() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    SetFieldObject<kTransactionActive>(ReferentOffset(), nullptr, true);
+    SetFieldObjectVolatile<kTransactionActive>(ReferentOffset(), nullptr);
   }
 
   // Volatile read/write is not necessary since the java pending next is only accessed from
   // the java threads for cleared references. Once these cleared references have a null referent,
   // we never end up reading their pending next from the GC again.
   Reference* GetPendingNext() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    return GetFieldObject<Reference>(PendingNextOffset(), false);
+    return GetFieldObject<Reference>(PendingNextOffset());
   }
   template<bool kTransactionActive>
   void SetPendingNext(Reference* pending_next) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    SetFieldObject<kTransactionActive>(PendingNextOffset(), pending_next, false);
+    SetFieldObject<kTransactionActive>(PendingNextOffset(), pending_next);
   }
 
   bool IsEnqueued() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
@@ -93,10 +93,10 @@
 
   template<bool kTransactionActive>
   void SetZombie(Object* zombie) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    return SetFieldObject<kTransactionActive>(ZombieOffset(), zombie, true);
+    return SetFieldObjectVolatile<kTransactionActive>(ZombieOffset(), zombie);
   }
   Object* GetZombie() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    return GetFieldObject<Object>(ZombieOffset(), true);
+    return GetFieldObjectVolatile<Object>(ZombieOffset());
   }
 
  private:
diff --git a/runtime/mirror/stack_trace_element.cc b/runtime/mirror/stack_trace_element.cc
index 5217e5e..f220039 100644
--- a/runtime/mirror/stack_trace_element.cc
+++ b/runtime/mirror/stack_trace_element.cc
@@ -60,13 +60,13 @@
 void StackTraceElement::Init(SirtRef<String>& declaring_class, SirtRef<String>& method_name,
                              SirtRef<String>& file_name, int32_t line_number) {
   SetFieldObject<kTransactionActive>(OFFSET_OF_OBJECT_MEMBER(StackTraceElement, declaring_class_),
-                                     declaring_class.get(), false);
+                                     declaring_class.get());
   SetFieldObject<kTransactionActive>(OFFSET_OF_OBJECT_MEMBER(StackTraceElement, method_name_),
-                                     method_name.get(), false);
+                                     method_name.get());
   SetFieldObject<kTransactionActive>(OFFSET_OF_OBJECT_MEMBER(StackTraceElement, file_name_),
-                                     file_name.get(), false);
+                                     file_name.get());
   SetField32<kTransactionActive>(OFFSET_OF_OBJECT_MEMBER(StackTraceElement, line_number_),
-                                 line_number, false);
+                                 line_number);
 }
 
 void StackTraceElement::VisitRoots(RootCallback* callback, void* arg) {
diff --git a/runtime/mirror/stack_trace_element.h b/runtime/mirror/stack_trace_element.h
index 9e023c7..1acbbb0 100644
--- a/runtime/mirror/stack_trace_element.h
+++ b/runtime/mirror/stack_trace_element.h
@@ -18,6 +18,7 @@
 #define ART_RUNTIME_MIRROR_STACK_TRACE_ELEMENT_H_
 
 #include "object.h"
+#include "object_callbacks.h"
 
 namespace art {
 
@@ -30,22 +31,19 @@
 class MANAGED StackTraceElement : public Object {
  public:
   String* GetDeclaringClass() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    return GetFieldObject<String>(OFFSET_OF_OBJECT_MEMBER(StackTraceElement, declaring_class_),
-                                  false);
+    return GetFieldObject<String>(OFFSET_OF_OBJECT_MEMBER(StackTraceElement, declaring_class_));
   }
 
   String* GetMethodName() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    return GetFieldObject<String>(OFFSET_OF_OBJECT_MEMBER(StackTraceElement, method_name_),
-                                  false);
+    return GetFieldObject<String>(OFFSET_OF_OBJECT_MEMBER(StackTraceElement, method_name_));
   }
 
   String* GetFileName() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    return GetFieldObject<String>(OFFSET_OF_OBJECT_MEMBER(StackTraceElement, file_name_),
-                                  false);
+    return GetFieldObject<String>(OFFSET_OF_OBJECT_MEMBER(StackTraceElement, file_name_));
   }
 
   int32_t GetLineNumber() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    return GetField32(OFFSET_OF_OBJECT_MEMBER(StackTraceElement, line_number_), false);
+    return GetField32(OFFSET_OF_OBJECT_MEMBER(StackTraceElement, line_number_));
   }
 
   static StackTraceElement* Alloc(Thread* self,
diff --git a/runtime/mirror/string-inl.h b/runtime/mirror/string-inl.h
new file mode 100644
index 0000000..315f7b1
--- /dev/null
+++ b/runtime/mirror/string-inl.h
@@ -0,0 +1,67 @@
+/*
+ * Copyright (C) 2011 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef ART_RUNTIME_MIRROR_STRING_INL_H_
+#define ART_RUNTIME_MIRROR_STRING_INL_H_
+
+#include "array.h"
+#include "intern_table.h"
+#include "runtime.h"
+#include "string.h"
+#include "thread.h"
+
+namespace art {
+namespace mirror {
+
+inline CharArray* String::GetCharArray() {
+  return GetFieldObject<CharArray>(ValueOffset());
+}
+
+inline int32_t String::GetLength() {
+  int32_t result = GetField32(OFFSET_OF_OBJECT_MEMBER(String, count_));
+  DCHECK(result >= 0 && result <= GetCharArray()->GetLength());
+  return result;
+}
+
+inline void String::SetArray(CharArray* new_array) {
+  // Array is invariant so use non-transactional mode. Also disable check as we may run inside
+  // a transaction.
+  DCHECK(new_array != NULL);
+  SetFieldObject<false, false>(OFFSET_OF_OBJECT_MEMBER(String, array_), new_array);
+}
+
+inline String* String::Intern() {
+  return Runtime::Current()->GetInternTable()->InternWeak(this);
+}
+
+inline uint16_t String::CharAt(int32_t index) {
+  // TODO: do we need this? Equals is the only caller, and could
+  // bounds check itself.
+  DCHECK_GE(count_, 0);  // ensures the unsigned comparison is safe.
+  if (UNLIKELY(static_cast<uint32_t>(index) >= static_cast<uint32_t>(count_))) {
+    Thread* self = Thread::Current();
+    ThrowLocation throw_location = self->GetCurrentLocationForThrow();
+    self->ThrowNewExceptionF(throw_location, "Ljava/lang/StringIndexOutOfBoundsException;",
+                             "length=%i; index=%i", count_, index);
+    return 0;
+  }
+  return GetCharArray()->Get(index + GetOffset());
+}
+
+}  // namespace mirror
+}  // namespace art
+
+#endif  // ART_RUNTIME_MIRROR_STRING_INL_H_
diff --git a/runtime/mirror/string.cc b/runtime/mirror/string.cc
index 88a8e6f..6a0c225 100644
--- a/runtime/mirror/string.cc
+++ b/runtime/mirror/string.cc
@@ -14,7 +14,7 @@
  * limitations under the License.
  */
 
-#include "string.h"
+#include "string-inl.h"
 
 #include "array.h"
 #include "class-inl.h"
@@ -29,17 +29,8 @@
 namespace art {
 namespace mirror {
 
-CharArray* String::GetCharArray() {
-  return GetFieldObject<CharArray>(ValueOffset(), false);
-}
-
-void String::ComputeHashCode() {
-  SetHashCode(ComputeUtf16Hash(GetCharArray(), GetOffset(), GetLength()));
-}
-
-int32_t String::GetUtfLength() {
-  return CountUtf8Bytes(GetCharArray()->GetData() + GetOffset(), GetLength());
-}
+// TODO: get global references for these
+Class* String::java_lang_String_ = NULL;
 
 int32_t String::FastIndexOf(int32_t ch, int32_t start) {
   int32_t count = GetLength();
@@ -59,16 +50,6 @@
   return -1;
 }
 
-void String::SetArray(CharArray* new_array) {
-  // Array is invariant so use non-transactional mode. Also disable check as we may run inside
-  // a transaction.
-  DCHECK(new_array != NULL);
-  SetFieldObject<false, false>(OFFSET_OF_OBJECT_MEMBER(String, array_), new_array, false);
-}
-
-// TODO: get global references for these
-Class* String::java_lang_String_ = NULL;
-
 void String::SetClass(Class* java_lang_String) {
   CHECK(java_lang_String_ == NULL);
   CHECK(java_lang_String != NULL);
@@ -80,39 +61,23 @@
   java_lang_String_ = NULL;
 }
 
-String* String::Intern() {
-  return Runtime::Current()->GetInternTable()->InternWeak(this);
-}
-
 int32_t String::GetHashCode() {
-  int32_t result = GetField32(OFFSET_OF_OBJECT_MEMBER(String, hash_code_), false);
-  if (result == 0) {
+  int32_t result = GetField32(OFFSET_OF_OBJECT_MEMBER(String, hash_code_));
+  if (UNLIKELY(result == 0)) {
     ComputeHashCode();
   }
-  result = GetField32(OFFSET_OF_OBJECT_MEMBER(String, hash_code_), false);
+  result = GetField32(OFFSET_OF_OBJECT_MEMBER(String, hash_code_));
   DCHECK(result != 0 || ComputeUtf16Hash(GetCharArray(), GetOffset(), GetLength()) == 0)
           << ToModifiedUtf8() << " " << result;
   return result;
 }
 
-int32_t String::GetLength() {
-  int32_t result = GetField32(OFFSET_OF_OBJECT_MEMBER(String, count_), false);
-  DCHECK(result >= 0 && result <= GetCharArray()->GetLength());
-  return result;
+void String::ComputeHashCode() {
+  SetHashCode(ComputeUtf16Hash(GetCharArray(), GetOffset(), GetLength()));
 }
 
-uint16_t String::CharAt(int32_t index) {
-  // TODO: do we need this? Equals is the only caller, and could
-  // bounds check itself.
-  DCHECK_GE(count_, 0);  // ensures the unsigned comparison is safe.
-  if (UNLIKELY(static_cast<uint32_t>(index) >= static_cast<uint32_t>(count_))) {
-    Thread* self = Thread::Current();
-    ThrowLocation throw_location = self->GetCurrentLocationForThrow();
-    self->ThrowNewExceptionF(throw_location, "Ljava/lang/StringIndexOutOfBoundsException;",
-                             "length=%i; index=%i", count_, index);
-    return 0;
-  }
-  return GetCharArray()->Get(index + GetOffset());
+int32_t String::GetUtfLength() {
+  return CountUtf8Bytes(GetCharArray()->GetData() + GetOffset(), GetLength());
 }
 
 String* String::AllocFromUtf16(Thread* self,
diff --git a/runtime/mirror/string.h b/runtime/mirror/string.h
index de9e4c4..f97308e 100644
--- a/runtime/mirror/string.h
+++ b/runtime/mirror/string.h
@@ -17,12 +17,14 @@
 #ifndef ART_RUNTIME_MIRROR_STRING_H_
 #define ART_RUNTIME_MIRROR_STRING_H_
 
+#include <gtest/gtest.h>
+
 #include "class.h"
-#include "gtest/gtest.h"
 #include "object_callbacks.h"
 
 namespace art {
 
+template<class T> class SirtRef;
 struct StringClassOffsets;
 struct StringOffsets;
 class StringPiece;
@@ -47,7 +49,7 @@
   CharArray* GetCharArray() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   int32_t GetOffset() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    int32_t result = GetField32(OffsetOffset(), false);
+    int32_t result = GetField32(OffsetOffset());
     DCHECK_LE(0, result);
     return result;
   }
@@ -111,25 +113,25 @@
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
  private:
-  void SetHashCode(int32_t new_hash_code) {
+  void SetHashCode(int32_t new_hash_code) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     // Hash code is invariant so use non-transactional mode. Also disable check as we may run inside
     // a transaction.
-    DCHECK_EQ(0, GetField32(OFFSET_OF_OBJECT_MEMBER(String, hash_code_), false));
-    SetField32<false, false>(OFFSET_OF_OBJECT_MEMBER(String, hash_code_), new_hash_code, false);
+    DCHECK_EQ(0, GetField32(OFFSET_OF_OBJECT_MEMBER(String, hash_code_)));
+    SetField32<false, false>(OFFSET_OF_OBJECT_MEMBER(String, hash_code_), new_hash_code);
   }
 
-  void SetCount(int32_t new_count) {
+  void SetCount(int32_t new_count) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     // Count is invariant so use non-transactional mode. Also disable check as we may run inside
     // a transaction.
     DCHECK_LE(0, new_count);
-    SetField32<false, false>(OFFSET_OF_OBJECT_MEMBER(String, count_), new_count, false);
+    SetField32<false, false>(OFFSET_OF_OBJECT_MEMBER(String, count_), new_count);
   }
 
   void SetOffset(int32_t new_offset) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     // Offset is only used during testing so use non-transactional mode.
     DCHECK_LE(0, new_offset);
     DCHECK_GE(GetLength(), new_offset);
-    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(String, offset_), new_offset, false);
+    SetField32<false>(OFFSET_OF_OBJECT_MEMBER(String, offset_), new_offset);
   }
 
   static String* Alloc(Thread* self, int32_t utf16_length)
diff --git a/runtime/mirror/throwable.cc b/runtime/mirror/throwable.cc
index 4c53993..d393a13 100644
--- a/runtime/mirror/throwable.cc
+++ b/runtime/mirror/throwable.cc
@@ -32,16 +32,24 @@
 
 Class* Throwable::java_lang_Throwable_ = NULL;
 
+void Throwable::SetDetailMessage(String* new_detail_message) {
+  if (Runtime::Current()->IsActiveTransaction()) {
+    SetFieldObject<true>(OFFSET_OF_OBJECT_MEMBER(Throwable, detail_message_), new_detail_message);
+  } else {
+    SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Throwable, detail_message_),
+                          new_detail_message);
+  }
+}
+
 void Throwable::SetCause(Throwable* cause) {
   CHECK(cause != nullptr);
   CHECK(cause != this);
-  Throwable* current_cause = GetFieldObject<Throwable>(OFFSET_OF_OBJECT_MEMBER(Throwable, cause_),
-                                                       false);
+  Throwable* current_cause = GetFieldObject<Throwable>(OFFSET_OF_OBJECT_MEMBER(Throwable, cause_));
   CHECK(current_cause == NULL || current_cause == this);
   if (Runtime::Current()->IsActiveTransaction()) {
-    SetFieldObject<true>(OFFSET_OF_OBJECT_MEMBER(Throwable, cause_), cause, false);
+    SetFieldObject<true>(OFFSET_OF_OBJECT_MEMBER(Throwable, cause_), cause);
   } else {
-    SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Throwable, cause_), cause, false);
+    SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Throwable, cause_), cause);
   }
 }
 
@@ -78,7 +86,7 @@
                              source_file, line_number);
     }
   }
-  Throwable* cause = GetFieldObject<Throwable>(OFFSET_OF_OBJECT_MEMBER(Throwable, cause_), false);
+  Throwable* cause = GetFieldObject<Throwable>(OFFSET_OF_OBJECT_MEMBER(Throwable, cause_));
   if (cause != NULL && cause != this) {  // Constructor makes cause == this by default.
     result += "Caused by: ";
     result += cause->Dump();
diff --git a/runtime/mirror/throwable.h b/runtime/mirror/throwable.h
index c1438d7..950b5e7 100644
--- a/runtime/mirror/throwable.h
+++ b/runtime/mirror/throwable.h
@@ -30,18 +30,12 @@
 // C++ mirror of java.lang.Throwable
 class MANAGED Throwable : public Object {
  public:
-  void SetDetailMessage(String* new_detail_message) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    if (Runtime::Current()->IsActiveTransaction()) {
-      SetFieldObject<true>(OFFSET_OF_OBJECT_MEMBER(Throwable, detail_message_), new_detail_message,
-                           false);
-    } else {
-      SetFieldObject<false>(OFFSET_OF_OBJECT_MEMBER(Throwable, detail_message_), new_detail_message,
-                            false);
-    }
-  }
+  void SetDetailMessage(String* new_detail_message) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
   String* GetDetailMessage() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    return GetFieldObject<String>(OFFSET_OF_OBJECT_MEMBER(Throwable, detail_message_), false);
+    return GetFieldObject<String>(OFFSET_OF_OBJECT_MEMBER(Throwable, detail_message_));
   }
+
   std::string Dump() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   // This is a runtime version of initCause, you shouldn't use it if initCause may have been
@@ -62,7 +56,7 @@
 
  private:
   Object* GetStackState() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    return GetFieldObject<Object>(OFFSET_OF_OBJECT_MEMBER(Throwable, stack_state_), true);
+    return GetFieldObjectVolatile<Object>(OFFSET_OF_OBJECT_MEMBER(Throwable, stack_state_));
   }
 
   // Field order required by test "ValidateFieldOrderOfJavaCppUnionClasses".
diff --git a/runtime/monitor.h b/runtime/monitor.h
index 0b80892..15620d5 100644
--- a/runtime/monitor.h
+++ b/runtime/monitor.h
@@ -31,18 +31,16 @@
 
 namespace art {
 
+class LockWord;
 template<class T> class SirtRef;
+class Thread;
+class StackVisitor;
+typedef uint32_t MonitorId;
 
 namespace mirror {
   class ArtMethod;
   class Object;
 }  // namespace mirror
-class LockWord;
-template<class T> class SirtRef;
-class Thread;
-class StackVisitor;
-
-typedef uint32_t MonitorId;
 
 class Monitor {
  public:
diff --git a/runtime/monitor_pool.h b/runtime/monitor_pool.h
index 82d0fee..32e1553 100644
--- a/runtime/monitor_pool.h
+++ b/runtime/monitor_pool.h
@@ -17,11 +17,12 @@
 #ifndef ART_RUNTIME_MONITOR_POOL_H_
 #define ART_RUNTIME_MONITOR_POOL_H_
 
+#include "monitor.h"
+
 #ifdef __LP64__
 #include <bitset>
 #include <stdint.h>
 
-#include "monitor.h"
 #include "runtime.h"
 #include "safe_map.h"
 #endif
diff --git a/runtime/native/dalvik_system_DexFile.cc b/runtime/native/dalvik_system_DexFile.cc
index 2231070..953d3a6 100644
--- a/runtime/native/dalvik_system_DexFile.cc
+++ b/runtime/native/dalvik_system_DexFile.cc
@@ -158,6 +158,7 @@
   if (dex_file == nullptr) {
     return;
   }
+  ScopedObjectAccess soa(env);
   if (Runtime::Current()->GetClassLinker()->IsDexFileRegistered(*dex_file)) {
     return;
   }
diff --git a/runtime/native/java_lang_String.cc b/runtime/native/java_lang_String.cc
index f95664b..d6b47eb 100644
--- a/runtime/native/java_lang_String.cc
+++ b/runtime/native/java_lang_String.cc
@@ -16,7 +16,7 @@
 
 #include "common_throws.h"
 #include "jni_internal.h"
-#include "mirror/string.h"
+#include "mirror/string-inl.h"
 #include "scoped_fast_native_object_access.h"
 #include "scoped_thread_state_change.h"
 #include "ScopedLocalRef.h"
diff --git a/runtime/native/sun_misc_Unsafe.cc b/runtime/native/sun_misc_Unsafe.cc
index ad0f317..764db5e 100644
--- a/runtime/native/sun_misc_Unsafe.cc
+++ b/runtime/native/sun_misc_Unsafe.cc
@@ -55,20 +55,20 @@
 static jint Unsafe_getInt(JNIEnv* env, jobject, jobject javaObj, jlong offset) {
   ScopedFastNativeObjectAccess soa(env);
   mirror::Object* obj = soa.Decode<mirror::Object*>(javaObj);
-  return obj->GetField32(MemberOffset(offset), false);
+  return obj->GetField32(MemberOffset(offset));
 }
 
 static jint Unsafe_getIntVolatile(JNIEnv* env, jobject, jobject javaObj, jlong offset) {
   ScopedFastNativeObjectAccess soa(env);
   mirror::Object* obj = soa.Decode<mirror::Object*>(javaObj);
-  return obj->GetField32(MemberOffset(offset), true);
+  return obj->GetField32Volatile(MemberOffset(offset));
 }
 
 static void Unsafe_putInt(JNIEnv* env, jobject, jobject javaObj, jlong offset, jint newValue) {
   ScopedFastNativeObjectAccess soa(env);
   mirror::Object* obj = soa.Decode<mirror::Object*>(javaObj);
   // JNI must use non transactional mode.
-  obj->SetField32<false>(MemberOffset(offset), newValue, false);
+  obj->SetField32<false>(MemberOffset(offset), newValue);
 }
 
 static void Unsafe_putIntVolatile(JNIEnv* env, jobject, jobject javaObj, jlong offset,
@@ -76,7 +76,7 @@
   ScopedFastNativeObjectAccess soa(env);
   mirror::Object* obj = soa.Decode<mirror::Object*>(javaObj);
   // JNI must use non transactional mode.
-  obj->SetField32<false>(MemberOffset(offset), newValue, true);
+  obj->SetField32Volatile<false>(MemberOffset(offset), newValue);
 }
 
 static void Unsafe_putOrderedInt(JNIEnv* env, jobject, jobject javaObj, jlong offset,
@@ -85,26 +85,26 @@
   mirror::Object* obj = soa.Decode<mirror::Object*>(javaObj);
   QuasiAtomic::MembarStoreStore();
   // JNI must use non transactional mode.
-  obj->SetField32<false>(MemberOffset(offset), newValue, false);
+  obj->SetField32<false>(MemberOffset(offset), newValue);
 }
 
 static jlong Unsafe_getLong(JNIEnv* env, jobject, jobject javaObj, jlong offset) {
   ScopedFastNativeObjectAccess soa(env);
   mirror::Object* obj = soa.Decode<mirror::Object*>(javaObj);
-  return obj->GetField64(MemberOffset(offset), false);
+  return obj->GetField64(MemberOffset(offset));
 }
 
 static jlong Unsafe_getLongVolatile(JNIEnv* env, jobject, jobject javaObj, jlong offset) {
   ScopedFastNativeObjectAccess soa(env);
   mirror::Object* obj = soa.Decode<mirror::Object*>(javaObj);
-  return obj->GetField64(MemberOffset(offset), true);
+  return obj->GetField64Volatile(MemberOffset(offset));
 }
 
 static void Unsafe_putLong(JNIEnv* env, jobject, jobject javaObj, jlong offset, jlong newValue) {
   ScopedFastNativeObjectAccess soa(env);
   mirror::Object* obj = soa.Decode<mirror::Object*>(javaObj);
   // JNI must use non transactional mode.
-  obj->SetField64<false>(MemberOffset(offset), newValue, false);
+  obj->SetField64<false>(MemberOffset(offset), newValue);
 }
 
 static void Unsafe_putLongVolatile(JNIEnv* env, jobject, jobject javaObj, jlong offset,
@@ -112,7 +112,7 @@
   ScopedFastNativeObjectAccess soa(env);
   mirror::Object* obj = soa.Decode<mirror::Object*>(javaObj);
   // JNI must use non transactional mode.
-  obj->SetField64<false>(MemberOffset(offset), newValue, true);
+  obj->SetField64Volatile<false>(MemberOffset(offset), newValue);
 }
 
 static void Unsafe_putOrderedLong(JNIEnv* env, jobject, jobject javaObj, jlong offset,
@@ -121,20 +121,20 @@
   mirror::Object* obj = soa.Decode<mirror::Object*>(javaObj);
   QuasiAtomic::MembarStoreStore();
   // JNI must use non transactional mode.
-  obj->SetField64<false>(MemberOffset(offset), newValue, false);
+  obj->SetField64<false>(MemberOffset(offset), newValue);
 }
 
 static jobject Unsafe_getObjectVolatile(JNIEnv* env, jobject, jobject javaObj, jlong offset) {
   ScopedFastNativeObjectAccess soa(env);
   mirror::Object* obj = soa.Decode<mirror::Object*>(javaObj);
-  mirror::Object* value = obj->GetFieldObject<mirror::Object>(MemberOffset(offset), true);
+  mirror::Object* value = obj->GetFieldObjectVolatile<mirror::Object>(MemberOffset(offset));
   return soa.AddLocalReference<jobject>(value);
 }
 
 static jobject Unsafe_getObject(JNIEnv* env, jobject, jobject javaObj, jlong offset) {
   ScopedFastNativeObjectAccess soa(env);
   mirror::Object* obj = soa.Decode<mirror::Object*>(javaObj);
-  mirror::Object* value = obj->GetFieldObject<mirror::Object>(MemberOffset(offset), false);
+  mirror::Object* value = obj->GetFieldObject<mirror::Object>(MemberOffset(offset));
   return soa.AddLocalReference<jobject>(value);
 }
 
@@ -144,7 +144,7 @@
   mirror::Object* obj = soa.Decode<mirror::Object*>(javaObj);
   mirror::Object* newValue = soa.Decode<mirror::Object*>(javaNewValue);
   // JNI must use non transactional mode.
-  obj->SetFieldObject<false>(MemberOffset(offset), newValue, false);
+  obj->SetFieldObject<false>(MemberOffset(offset), newValue);
 }
 
 static void Unsafe_putObjectVolatile(JNIEnv* env, jobject, jobject javaObj, jlong offset,
@@ -153,7 +153,7 @@
   mirror::Object* obj = soa.Decode<mirror::Object*>(javaObj);
   mirror::Object* newValue = soa.Decode<mirror::Object*>(javaNewValue);
   // JNI must use non transactional mode.
-  obj->SetFieldObject<false>(MemberOffset(offset), newValue, true);
+  obj->SetFieldObjectVolatile<false>(MemberOffset(offset), newValue);
 }
 
 static void Unsafe_putOrderedObject(JNIEnv* env, jobject, jobject javaObj, jlong offset,
@@ -163,7 +163,7 @@
   mirror::Object* newValue = soa.Decode<mirror::Object*>(javaNewValue);
   QuasiAtomic::MembarStoreStore();
   // JNI must use non transactional mode.
-  obj->SetFieldObject<false>(MemberOffset(offset), newValue, false);
+  obj->SetFieldObject<false>(MemberOffset(offset), newValue);
 }
 
 static jint Unsafe_getArrayBaseOffsetForComponentType(JNIEnv* env, jclass, jobject component_class) {
diff --git a/runtime/oat_file.h b/runtime/oat_file.h
index 5f6cb1e..b358a00 100644
--- a/runtime/oat_file.h
+++ b/runtime/oat_file.h
@@ -73,7 +73,7 @@
 
   class OatMethod {
    public:
-    void LinkMethod(mirror::ArtMethod* method) const;
+    void LinkMethod(mirror::ArtMethod* method) const SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
     uint32_t GetCodeOffset() const {
       return code_offset_;
diff --git a/runtime/reference_table.cc b/runtime/reference_table.cc
index a3119bb..11527fa 100644
--- a/runtime/reference_table.cc
+++ b/runtime/reference_table.cc
@@ -23,7 +23,7 @@
 #include "mirror/class.h"
 #include "mirror/class-inl.h"
 #include "mirror/object-inl.h"
-#include "mirror/string.h"
+#include "mirror/string-inl.h"
 #include "thread.h"
 #include "utils.h"
 
diff --git a/runtime/stack_indirect_reference_table.h b/runtime/stack_indirect_reference_table.h
index b113129..3b632e7 100644
--- a/runtime/stack_indirect_reference_table.h
+++ b/runtime/stack_indirect_reference_table.h
@@ -20,6 +20,7 @@
 #include "base/logging.h"
 #include "base/macros.h"
 #include "stack.h"
+#include "utils.h"
 
 namespace art {
 namespace mirror {
diff --git a/runtime/thread.cc b/runtime/thread.cc
index 0fafbfa..7470670 100644
--- a/runtime/thread.cc
+++ b/runtime/thread.cc
@@ -1400,7 +1400,8 @@
       return true;  // Ignore runtime frames (in particular callee save).
     }
     method_trace_->Set<kTransactionActive>(count_, m);
-    dex_pc_trace_->Set<kTransactionActive>(count_, m->IsProxyMethod() ? DexFile::kDexNoIndex : GetDexPc());
+    dex_pc_trace_->Set<kTransactionActive>(count_,
+        m->IsProxyMethod() ? DexFile::kDexNoIndex : GetDexPc());
     ++count_;
     return true;
   }
diff --git a/runtime/transaction.cc b/runtime/transaction.cc
index e18cf04..cc02a8d 100644
--- a/runtime/transaction.cc
+++ b/runtime/transaction.cc
@@ -282,17 +282,29 @@
   constexpr bool kCheckTransaction = true;
   switch (field_value.kind) {
     case k32Bits:
-      obj->SetField32<false, kCheckTransaction>(field_offset, static_cast<uint32_t>(field_value.value),
-                                                field_value.is_volatile);
+      if (UNLIKELY(field_value.is_volatile)) {
+        obj->SetField32Volatile<false, kCheckTransaction>(field_offset,
+                                                          static_cast<uint32_t>(field_value.value));
+      } else {
+        obj->SetField32<false, kCheckTransaction>(field_offset,
+                                                  static_cast<uint32_t>(field_value.value));
+      }
       break;
     case k64Bits:
-      obj->SetField64<false, kCheckTransaction>(field_offset, field_value.value,
-                                                field_value.is_volatile);
+      if (UNLIKELY(field_value.is_volatile)) {
+        obj->SetField64Volatile<false, kCheckTransaction>(field_offset, field_value.value);
+      } else {
+        obj->SetField64<false, kCheckTransaction>(field_offset, field_value.value);
+      }
       break;
     case kReference:
-      obj->SetFieldObject<false, kCheckTransaction>(field_offset,
-                                                    reinterpret_cast<mirror::Object*>(field_value.value),
-                                                    field_value.is_volatile);
+      if (UNLIKELY(field_value.is_volatile)) {
+        obj->SetFieldObjectVolatile<false, kCheckTransaction>(field_offset,
+                                                              reinterpret_cast<mirror::Object*>(field_value.value));
+      } else {
+        obj->SetFieldObject<false, kCheckTransaction>(field_offset,
+                                                      reinterpret_cast<mirror::Object*>(field_value.value));
+      }
       break;
     default:
       LOG(FATAL) << "Unknown value kind " << field_value.kind;