diff --git a/compiler/image_test.cc b/compiler/image_test.cc
index 05d6693..7c5741b 100644
--- a/compiler/image_test.cc
+++ b/compiler/image_test.cc
@@ -164,7 +164,7 @@
       EXPECT_TRUE(reinterpret_cast<byte*>(klass) >= image_end ||
                   reinterpret_cast<byte*>(klass) < image_begin) << descriptor;
     }
-    EXPECT_TRUE(Monitor::IsValidLockWord(klass->GetLockWord()));
+    EXPECT_TRUE(Monitor::IsValidLockWord(klass->GetLockWord(false)));
   }
 }
 
diff --git a/compiler/image_writer.cc b/compiler/image_writer.cc
index f76587a..3400b01 100644
--- a/compiler/image_writer.cc
+++ b/compiler/image_writer.cc
@@ -179,7 +179,7 @@
   image_bitmap_->Set(obj);
   // Before we stomp over the lock word, save the hash code for later.
   Monitor::Deflate(Thread::Current(), object);;
-  LockWord lw(object->GetLockWord());
+  LockWord lw(object->GetLockWord(false));
   switch (lw.GetState()) {
     case LockWord::kFatLocked: {
       LOG(FATAL) << "Fat locked object " << obj << " found during object copy";
@@ -199,7 +199,7 @@
       LOG(FATAL) << "Unreachable.";
       break;
   }
-  object->SetLockWord(LockWord::FromForwardingAddress(offset));
+  object->SetLockWord(LockWord::FromForwardingAddress(offset), false);
   DCHECK(IsImageOffsetAssigned(object));
 }
 
@@ -212,13 +212,13 @@
 
 bool ImageWriter::IsImageOffsetAssigned(mirror::Object* object) const {
   DCHECK(object != nullptr);
-  return object->GetLockWord().GetState() == LockWord::kForwardingAddress;
+  return object->GetLockWord(false).GetState() == LockWord::kForwardingAddress;
 }
 
 size_t ImageWriter::GetImageOffset(mirror::Object* object) const {
   DCHECK(object != nullptr);
   DCHECK(IsImageOffsetAssigned(object));
-  LockWord lock_word = object->GetLockWord();
+  LockWord lock_word = object->GetLockWord(false);
   size_t offset = lock_word.ForwardingAddress();
   DCHECK_LT(offset, image_end_);
   return offset;
@@ -555,15 +555,15 @@
   heap->VisitObjects(CopyAndFixupObjectsCallback, this);
   // Fix up the object previously had hash codes.
   for (const std::pair<mirror::Object*, uint32_t>& hash_pair : saved_hashes_) {
-    hash_pair.first->SetLockWord(LockWord::FromHashCode(hash_pair.second));
+    hash_pair.first->SetLockWord(LockWord::FromHashCode(hash_pair.second), false);
   }
   saved_hashes_.clear();
   self->EndAssertNoThreadSuspension(old_cause);
 }
 
 void ImageWriter::CopyAndFixupObjectsCallback(Object* obj, void* arg) {
-  DCHECK(obj != NULL);
-  DCHECK(arg != NULL);
+  DCHECK(obj != nullptr);
+  DCHECK(arg != nullptr);
   ImageWriter* image_writer = reinterpret_cast<ImageWriter*>(arg);
   // see GetLocalAddress for similar computation
   size_t offset = image_writer->GetImageOffset(obj);
@@ -575,7 +575,7 @@
   Object* copy = reinterpret_cast<Object*>(dst);
   // Write in a hash code of objects which have inflated monitors or a hash code in their monitor
   // word.
-  copy->SetLockWord(LockWord());
+  copy->SetLockWord(LockWord(), false);
   image_writer->FixupObject(obj, copy);
 }
 
diff --git a/runtime/gc/collector/semi_space-inl.h b/runtime/gc/collector/semi_space-inl.h
index 8a9611f..d03baf1 100644
--- a/runtime/gc/collector/semi_space-inl.h
+++ b/runtime/gc/collector/semi_space-inl.h
@@ -28,7 +28,7 @@
 
 inline mirror::Object* SemiSpace::GetForwardingAddressInFromSpace(mirror::Object* obj) const {
   DCHECK(from_space_->HasAddress(obj));
-  LockWord lock_word = obj->GetLockWord();
+  LockWord lock_word = obj->GetLockWord(false);
   if (lock_word.GetState() != LockWord::kForwardingAddress) {
     return nullptr;
   }
@@ -58,8 +58,8 @@
         DCHECK(forward_address != nullptr);
         // Make sure to only update the forwarding address AFTER you copy the object so that the
         // monitor word doesn't get stomped over.
-        obj->SetLockWord(LockWord::FromForwardingAddress(
-            reinterpret_cast<size_t>(forward_address)));
+        obj->SetLockWord(
+            LockWord::FromForwardingAddress(reinterpret_cast<size_t>(forward_address)), false);
         // Push the object onto the mark stack for later processing.
         MarkStackPush(forward_address);
       }
diff --git a/runtime/gc/collector/semi_space.cc b/runtime/gc/collector/semi_space.cc
index c0e172e..4a1bf18 100644
--- a/runtime/gc/collector/semi_space.cc
+++ b/runtime/gc/collector/semi_space.cc
@@ -117,6 +117,8 @@
   immune_region_.Reset();
   is_large_object_space_immune_ = false;
   saved_bytes_ = 0;
+  bytes_moved_ = 0;
+  objects_moved_ = 0;
   self_ = Thread::Current();
   // Do any pre GC verification.
   timings_.NewSplit("PreGcVerification");
@@ -382,9 +384,9 @@
   }
   // Record freed memory.
   uint64_t from_bytes = from_space_->GetBytesAllocated();
-  uint64_t to_bytes = to_space_->GetBytesAllocated();
+  uint64_t to_bytes = bytes_moved_;
   uint64_t from_objects = from_space_->GetObjectsAllocated();
-  uint64_t to_objects = to_space_->GetObjectsAllocated();
+  uint64_t to_objects = objects_moved_;
   CHECK_LE(to_objects, from_objects);
   int64_t freed_bytes = from_bytes - to_bytes;
   int64_t freed_objects = from_objects - to_objects;
@@ -521,15 +523,13 @@
     // If it's allocated before the last GC (older), move
     // (pseudo-promote) it to the main free list space (as sort
     // of an old generation.)
-    size_t bytes_promoted;
     space::MallocSpace* promo_dest_space = GetHeap()->GetPrimaryFreeListSpace();
-    forward_address = promo_dest_space->Alloc(self_, object_size, &bytes_promoted, nullptr);
-    if (forward_address == nullptr) {
+    forward_address = promo_dest_space->Alloc(self_, object_size, &bytes_allocated, nullptr);
+    if (UNLIKELY(forward_address == nullptr)) {
       // If out of space, fall back to the to-space.
       forward_address = to_space_->Alloc(self_, object_size, &bytes_allocated, nullptr);
     } else {
-      GetHeap()->num_bytes_allocated_.FetchAndAdd(bytes_promoted);
-      bytes_promoted_ += bytes_promoted;
+      bytes_promoted_ += bytes_allocated;
       // Dirty the card at the destionation as it may contain
       // references (including the class pointer) to the bump pointer
       // space.
@@ -573,6 +573,8 @@
     // If it's allocated after the last GC (younger), copy it to the to-space.
     forward_address = to_space_->Alloc(self_, object_size, &bytes_allocated, nullptr);
   }
+  ++objects_moved_;
+  bytes_moved_ += bytes_allocated;
   // Copy over the object and add it to the mark stack since we still need to update its
   // references.
   saved_bytes_ +=
@@ -619,10 +621,9 @@
 
 // Marks all objects in the root set.
 void SemiSpace::MarkRoots() {
-  timings_.StartSplit("MarkRoots");
+  timings_.NewSplit("MarkRoots");
   // TODO: Visit up image roots as well?
   Runtime::Current()->VisitRoots(MarkRootCallback, this);
-  timings_.EndSplit();
 }
 
 mirror::Object* SemiSpace::MarkedForwardingAddressCallback(mirror::Object* object, void* arg) {
@@ -641,7 +642,7 @@
 
 void SemiSpace::Sweep(bool swap_bitmaps) {
   DCHECK(mark_stack_->IsEmpty());
-  TimingLogger::ScopedSplit("Sweep", &timings_);
+  TimingLogger::ScopedSplit split("Sweep", &timings_);
   for (const auto& space : GetHeap()->GetContinuousSpaces()) {
     if (space->IsContinuousMemMapAllocSpace()) {
       space::ContinuousMemMapAllocSpace* alloc_space = space->AsContinuousMemMapAllocSpace();
@@ -665,13 +666,13 @@
 
 void SemiSpace::SweepLargeObjects(bool swap_bitmaps) {
   DCHECK(!is_large_object_space_immune_);
-  TimingLogger::ScopedSplit("SweepLargeObjects", &timings_);
+  TimingLogger::ScopedSplit split("SweepLargeObjects", &timings_);
   size_t freed_objects = 0;
   size_t freed_bytes = 0;
-  GetHeap()->GetLargeObjectsSpace()->Sweep(swap_bitmaps, &freed_objects, &freed_bytes);
+  heap_->GetLargeObjectsSpace()->Sweep(swap_bitmaps, &freed_objects, &freed_bytes);
   freed_large_objects_.FetchAndAdd(freed_objects);
   freed_large_object_bytes_.FetchAndAdd(freed_bytes);
-  GetHeap()->RecordFree(freed_objects, freed_bytes);
+  heap_->RecordFree(freed_objects, freed_bytes);
 }
 
 // Process the "referent" field in a java.lang.ref.Reference.  If the referent has not yet been
diff --git a/runtime/gc/collector/semi_space.h b/runtime/gc/collector/semi_space.h
index 4169ca9..b6726b2 100644
--- a/runtime/gc/collector/semi_space.h
+++ b/runtime/gc/collector/semi_space.h
@@ -231,6 +231,11 @@
   // whole_heap_collection_ once per interval.
   int whole_heap_collection_interval_counter_;
 
+  // How many objects and bytes we moved, used so that we don't need to get the size of the
+  // to_space_ when calculating how many objects and bytes we freed.
+  size_t bytes_moved_;
+  size_t objects_moved_;
+
   // How many bytes we avoided dirtying.
   size_t saved_bytes_;
 
diff --git a/runtime/mirror/object-inl.h b/runtime/mirror/object-inl.h
index a6db387..2505855 100644
--- a/runtime/mirror/object-inl.h
+++ b/runtime/mirror/object-inl.h
@@ -51,13 +51,14 @@
       OFFSET_OF_OBJECT_MEMBER(Object, klass_), new_klass, false);
 }
 
-inline LockWord Object::GetLockWord() {
-  return LockWord(GetField32(OFFSET_OF_OBJECT_MEMBER(Object, monitor_), true));
+inline LockWord Object::GetLockWord(bool is_volatile) {
+  return LockWord(GetField32(OFFSET_OF_OBJECT_MEMBER(Object, monitor_), is_volatile));
 }
 
-inline void Object::SetLockWord(LockWord new_val) {
+inline void Object::SetLockWord(LockWord new_val, bool is_volatile) {
   // Force use of non-transactional mode and do not check.
-  SetField32<false, false>(OFFSET_OF_OBJECT_MEMBER(Object, monitor_), new_val.GetValue(), true);
+  SetField32<false, false>(OFFSET_OF_OBJECT_MEMBER(Object, monitor_), new_val.GetValue(),
+                           is_volatile);
 }
 
 inline bool Object::CasLockWord(LockWord old_val, LockWord new_val) {
diff --git a/runtime/mirror/object.cc b/runtime/mirror/object.cc
index be7e9f2..766bbc9 100644
--- a/runtime/mirror/object.cc
+++ b/runtime/mirror/object.cc
@@ -150,7 +150,7 @@
 int32_t Object::IdentityHashCode() const {
   mirror::Object* current_this = const_cast<mirror::Object*>(this);
   while (true) {
-    LockWord lw = current_this->GetLockWord();
+    LockWord lw = current_this->GetLockWord(false);
     switch (lw.GetState()) {
       case LockWord::kUnlocked: {
         // Try to compare and swap in a new hash, if we succeed we will return the hash on the next
diff --git a/runtime/mirror/object.h b/runtime/mirror/object.h
index f652202..fd31dfb 100644
--- a/runtime/mirror/object.h
+++ b/runtime/mirror/object.h
@@ -100,8 +100,10 @@
     return OFFSET_OF_OBJECT_MEMBER(Object, monitor_);
   }
 
-  LockWord GetLockWord();
-  void SetLockWord(LockWord new_val);
+  // 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);
   bool CasLockWord(LockWord old_val, LockWord new_val) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
   uint32_t GetLockOwnerThreadId();
 
diff --git a/runtime/monitor.cc b/runtime/monitor.cc
index bbc7dd0..38b77d1 100644
--- a/runtime/monitor.cc
+++ b/runtime/monitor.cc
@@ -111,7 +111,7 @@
   MutexLock mu(self, monitor_lock_);  // Uncontended mutex acquisition as monitor isn't yet public.
   CHECK(owner_ == nullptr || owner_ == self || owner_->IsSuspended());
   // Propagate the lock state.
-  LockWord lw(obj_->GetLockWord());
+  LockWord lw(obj_->GetLockWord(false));
   switch (lw.GetState()) {
     case LockWord::kThinLocked: {
       CHECK_EQ(owner_->GetThreadId(), lw.ThinLockOwner());
@@ -574,7 +574,8 @@
 
 bool Monitor::Deflate(Thread* self, mirror::Object* obj) {
   DCHECK(obj != nullptr);
-  LockWord lw(obj->GetLockWord());
+  // Don't need volatile since we only deflate with mutators suspended.
+  LockWord lw(obj->GetLockWord(false));
   // If the lock isn't an inflated monitor, then we don't need to deflate anything.
   if (lw.GetState() == LockWord::kFatLocked) {
     Monitor* monitor = lw.FatLockMonitor();
@@ -595,14 +596,15 @@
         return false;
       }
       // Deflate to a thin lock.
-      obj->SetLockWord(LockWord::FromThinLockId(owner->GetThreadId(), monitor->lock_count_));
-      VLOG(monitor) << "Deflated " << obj << " to thin lock " << owner->GetTid() << " / " << monitor->lock_count_;
+      obj->SetLockWord(LockWord::FromThinLockId(owner->GetThreadId(), monitor->lock_count_), false);
+      VLOG(monitor) << "Deflated " << obj << " to thin lock " << owner->GetTid() << " / "
+          << monitor->lock_count_;
     } else if (monitor->HasHashCode()) {
-      obj->SetLockWord(LockWord::FromHashCode(monitor->GetHashCode()));
+      obj->SetLockWord(LockWord::FromHashCode(monitor->GetHashCode()), false);
       VLOG(monitor) << "Deflated " << obj << " to hash monitor " << monitor->GetHashCode();
     } else {
       // No lock and no hash, just put an empty lock word inside the object.
-      obj->SetLockWord(LockWord());
+      obj->SetLockWord(LockWord(), false);
       VLOG(monitor) << "Deflated" << obj << " to empty lock word";
     }
     // The monitor is deflated, mark the object as nullptr so that we know to delete it during the
@@ -626,7 +628,7 @@
     VLOG(monitor) << "monitor: thread " << owner->GetThreadId()
                     << " created monitor " << m.get() << " for object " << obj;
     Runtime::Current()->GetMonitorList()->Add(m.release());
-    CHECK_EQ(obj->GetLockWord().GetState(), LockWord::kFatLocked);
+    CHECK_EQ(obj->GetLockWord(true).GetState(), LockWord::kFatLocked);
   }
 }
 
@@ -642,12 +644,12 @@
     // Suspend the owner, inflate. First change to blocked and give up mutator_lock_.
     ScopedThreadStateChange tsc(self, kBlocked);
     self->SetMonitorEnterObject(obj.get());
-    if (lock_word == obj->GetLockWord()) {  // If lock word hasn't changed.
+    if (lock_word == obj->GetLockWord(true)) {  // If lock word hasn't changed.
       bool timed_out;
       Thread* owner = thread_list->SuspendThreadByThreadId(owner_thread_id, false, &timed_out);
       if (owner != nullptr) {
         // We succeeded in suspending the thread, check the lock's status didn't change.
-        lock_word = obj->GetLockWord();
+        lock_word = obj->GetLockWord(true);
         if (lock_word.GetState() == LockWord::kThinLocked &&
             lock_word.ThinLockOwner() == owner_thread_id) {
           // Go ahead and inflate the lock.
@@ -680,7 +682,7 @@
   size_t contention_count = 0;
   SirtRef<mirror::Object> sirt_obj(self, obj);
   while (true) {
-    LockWord lock_word = sirt_obj->GetLockWord();
+    LockWord lock_word = sirt_obj->GetLockWord(true);
     switch (lock_word.GetState()) {
       case LockWord::kUnlocked: {
         LockWord thin_locked(LockWord::FromThinLockId(thread_id, 0));
@@ -697,7 +699,7 @@
           uint32_t new_count = lock_word.ThinLockCount() + 1;
           if (LIKELY(new_count <= LockWord::kThinLockMaxCount)) {
             LockWord thin_locked(LockWord::FromThinLockId(thread_id, new_count));
-            sirt_obj->SetLockWord(thin_locked);
+            sirt_obj->SetLockWord(thin_locked, true);
             return sirt_obj.get();  // Success!
           } else {
             // We'd overflow the recursion count, so inflate the monitor.
@@ -737,13 +739,13 @@
   DCHECK(self != NULL);
   DCHECK(obj != NULL);
   obj = FakeUnlock(obj);
-  LockWord lock_word = obj->GetLockWord();
+  LockWord lock_word = obj->GetLockWord(true);
   SirtRef<mirror::Object> sirt_obj(self, obj);
   switch (lock_word.GetState()) {
     case LockWord::kHashCode:
       // Fall-through.
     case LockWord::kUnlocked:
-      FailedUnlock(sirt_obj.get(), self, NULL, NULL);
+      FailedUnlock(sirt_obj.get(), self, nullptr, nullptr);
       return false;  // Failure.
     case LockWord::kThinLocked: {
       uint32_t thread_id = self->GetThreadId();
@@ -752,16 +754,16 @@
         // TODO: there's a race here with the owner dying while we unlock.
         Thread* owner =
             Runtime::Current()->GetThreadList()->FindThreadByThreadId(lock_word.ThinLockOwner());
-        FailedUnlock(sirt_obj.get(), self, owner, NULL);
+        FailedUnlock(sirt_obj.get(), self, owner, nullptr);
         return false;  // Failure.
       } else {
         // We own the lock, decrease the recursion count.
         if (lock_word.ThinLockCount() != 0) {
           uint32_t new_count = lock_word.ThinLockCount() - 1;
           LockWord thin_locked(LockWord::FromThinLockId(thread_id, new_count));
-          sirt_obj->SetLockWord(thin_locked);
+          sirt_obj->SetLockWord(thin_locked, true);
         } else {
-          sirt_obj->SetLockWord(LockWord());
+          sirt_obj->SetLockWord(LockWord(), true);
         }
         return true;  // Success!
       }
@@ -782,10 +784,9 @@
  */
 void Monitor::Wait(Thread* self, mirror::Object *obj, int64_t ms, int32_t ns,
                    bool interruptShouldThrow, ThreadState why) {
-  DCHECK(self != NULL);
-  DCHECK(obj != NULL);
-
-  LockWord lock_word = obj->GetLockWord();
+  DCHECK(self != nullptr);
+  DCHECK(obj != nullptr);
+  LockWord lock_word = obj->GetLockWord(true);
   switch (lock_word.GetState()) {
     case LockWord::kHashCode:
       // Fall-through.
@@ -801,7 +802,7 @@
       } else {
         // We own the lock, inflate to enqueue ourself on the Monitor.
         Inflate(self, self, obj, 0);
-        lock_word = obj->GetLockWord();
+        lock_word = obj->GetLockWord(true);
       }
       break;
     }
@@ -817,10 +818,9 @@
 }
 
 void Monitor::DoNotify(Thread* self, mirror::Object* obj, bool notify_all) {
-  DCHECK(self != NULL);
-  DCHECK(obj != NULL);
-
-  LockWord lock_word = obj->GetLockWord();
+  DCHECK(self != nullptr);
+  DCHECK(obj != nullptr);
+  LockWord lock_word = obj->GetLockWord(true);
   switch (lock_word.GetState()) {
     case LockWord::kHashCode:
       // Fall-through.
@@ -855,9 +855,8 @@
 }
 
 uint32_t Monitor::GetLockOwnerThreadId(mirror::Object* obj) {
-  DCHECK(obj != NULL);
-
-  LockWord lock_word = obj->GetLockWord();
+  DCHECK(obj != nullptr);
+  LockWord lock_word = obj->GetLockWord(true);
   switch (lock_word.GetState()) {
     case LockWord::kHashCode:
       // Fall-through.
@@ -902,7 +901,7 @@
     if (pretty_object == nullptr) {
       os << wait_message << "an unknown object";
     } else {
-      if ((pretty_object->GetLockWord().GetState() == LockWord::kThinLocked) &&
+      if ((pretty_object->GetLockWord(true).GetState() == LockWord::kThinLocked) &&
           Locks::mutator_lock_->IsExclusiveHeld(Thread::Current())) {
         // Getting the identity hashcode here would result in lock inflation and suspension of the
         // current thread, which isn't safe if this is the only runnable thread.
@@ -1112,7 +1111,7 @@
 static mirror::Object* MonitorDeflateCallback(mirror::Object* object, void* arg)
     SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
   if (Monitor::Deflate(reinterpret_cast<Thread*>(arg), object)) {
-    DCHECK_NE(object->GetLockWord().GetState(), LockWord::kFatLocked);
+    DCHECK_NE(object->GetLockWord(true).GetState(), LockWord::kFatLocked);
     // If we deflated, return nullptr so that the monitor gets removed from the array.
     return nullptr;
   }
@@ -1126,9 +1125,8 @@
 }
 
 MonitorInfo::MonitorInfo(mirror::Object* obj) : owner_(NULL), entry_count_(0) {
-  DCHECK(obj != NULL);
-
-  LockWord lock_word = obj->GetLockWord();
+  DCHECK(obj != nullptr);
+  LockWord lock_word = obj->GetLockWord(true);
   switch (lock_word.GetState()) {
     case LockWord::kUnlocked:
       // Fall-through.
diff --git a/runtime/thread.cc b/runtime/thread.cc
index b3d14f0..8691dec 100644
--- a/runtime/thread.cc
+++ b/runtime/thread.cc
@@ -876,7 +876,7 @@
     if (o == nullptr) {
       os << "an unknown object";
     } else {
-      if ((o->GetLockWord().GetState() == LockWord::kThinLocked) &&
+      if ((o->GetLockWord(false).GetState() == LockWord::kThinLocked) &&
           Locks::mutator_lock_->IsExclusiveHeld(Thread::Current())) {
         // Getting the identity hashcode here would result in lock inflation and suspension of the
         // current thread, which isn't safe if this is the only runnable thread.
diff --git a/runtime/transaction_test.cc b/runtime/transaction_test.cc
index 76b6f27..1dc2da0 100644
--- a/runtime/transaction_test.cc
+++ b/runtime/transaction_test.cc
@@ -55,18 +55,18 @@
 
   // Lock object's monitor outside the transaction.
   sirt_obj->MonitorEnter(soa.Self());
-  uint32_t old_lock_word = sirt_obj->GetLockWord().GetValue();
+  uint32_t old_lock_word = sirt_obj->GetLockWord(false).GetValue();
 
   Transaction transaction;
   Runtime::Current()->EnterTransactionMode(&transaction);
   // Unlock object's monitor inside the transaction.
   sirt_obj->MonitorExit(soa.Self());
-  uint32_t new_lock_word = sirt_obj->GetLockWord().GetValue();
+  uint32_t new_lock_word = sirt_obj->GetLockWord(false).GetValue();
   Runtime::Current()->ExitTransactionMode();
 
   // Aborting transaction must not clear the Object::class field.
   transaction.Abort();
-  uint32_t aborted_lock_word = sirt_obj->GetLockWord().GetValue();
+  uint32_t aborted_lock_word = sirt_obj->GetLockWord(false).GetValue();
   EXPECT_NE(old_lock_word, new_lock_word);
   EXPECT_EQ(aborted_lock_word, new_lock_word);
 }