Merge "Add ClassPreDefine hook."
diff --git a/build/Android.gtest.mk b/build/Android.gtest.mk
index d8b780a..c87075f 100644
--- a/build/Android.gtest.mk
+++ b/build/Android.gtest.mk
@@ -87,7 +87,7 @@
 ART_GTEST_dex2oat_environment_tests_DEX_DEPS := Main MainStripped MultiDex MultiDexModifiedSecondary Nested
 
 ART_GTEST_atomic_method_ref_map_test_DEX_DEPS := Interfaces
-ART_GTEST_class_linker_test_DEX_DEPS := ErroneousA ErroneousB Interfaces MethodTypes MultiDex MyClass Nested Statics StaticsFromCode
+ART_GTEST_class_linker_test_DEX_DEPS := AllFields ErroneousA ErroneousB Interfaces MethodTypes MultiDex MyClass Nested Statics StaticsFromCode
 ART_GTEST_class_table_test_DEX_DEPS := XandY
 ART_GTEST_compiler_driver_test_DEX_DEPS := AbstractMethod StaticLeafMethods ProfileTestMultiDex
 ART_GTEST_dex_cache_test_DEX_DEPS := Main Packages MethodTypes
diff --git a/compiler/debug/elf_debug_line_writer.h b/compiler/debug/elf_debug_line_writer.h
index 3db7306..18a9165 100644
--- a/compiler/debug/elf_debug_line_writer.h
+++ b/compiler/debug/elf_debug_line_writer.h
@@ -53,7 +53,8 @@
   // Write line table for given set of methods.
   // Returns the number of bytes written.
   size_t WriteCompilationUnit(ElfCompilationUnit& compilation_unit) {
-    const bool is64bit = Is64BitInstructionSet(builder_->GetIsa());
+    const InstructionSet isa = builder_->GetIsa();
+    const bool is64bit = Is64BitInstructionSet(isa);
     const Elf_Addr base_address = compilation_unit.is_code_address_text_relative
         ? builder_->GetText()->GetAddress()
         : 0;
@@ -66,7 +67,7 @@
     std::unordered_map<std::string, size_t> directories_map;
     int code_factor_bits_ = 0;
     int dwarf_isa = -1;
-    switch (builder_->GetIsa()) {
+    switch (isa) {
       case kArm:  // arm actually means thumb2.
       case kThumb2:
         code_factor_bits_ = 1;  // 16-bit instuctions
@@ -103,7 +104,7 @@
         for (uint32_t s = 0; s < code_info.GetNumberOfStackMaps(encoding); s++) {
           StackMap stack_map = code_info.GetStackMapAt(s, encoding);
           DCHECK(stack_map.IsValid());
-          const uint32_t pc = stack_map.GetNativePcOffset(encoding.stack_map_encoding);
+          const uint32_t pc = stack_map.GetNativePcOffset(encoding.stack_map_encoding, isa);
           const int32_t dex = stack_map.GetDexPc(encoding.stack_map_encoding);
           pc2dex_map.push_back({pc, dex});
           if (stack_map.HasDexRegisterMap(encoding.stack_map_encoding)) {
diff --git a/compiler/debug/elf_debug_loc_writer.h b/compiler/debug/elf_debug_loc_writer.h
index 9645643..bce5387 100644
--- a/compiler/debug/elf_debug_loc_writer.h
+++ b/compiler/debug/elf_debug_loc_writer.h
@@ -92,7 +92,8 @@
     bool is64bitValue,
     uint64_t compilation_unit_code_address,
     uint32_t dex_pc_low,
-    uint32_t dex_pc_high) {
+    uint32_t dex_pc_high,
+    InstructionSet isa) {
   std::vector<VariableLocation> variable_locations;
 
   // Get stack maps sorted by pc (they might not be sorted internally).
@@ -111,7 +112,7 @@
       // The main reason for this is to save space by avoiding undefined gaps.
       continue;
     }
-    const uint32_t pc_offset = stack_map.GetNativePcOffset(encoding.stack_map_encoding);
+    const uint32_t pc_offset = stack_map.GetNativePcOffset(encoding.stack_map_encoding, isa);
     DCHECK_LE(pc_offset, method_info->code_size);
     DCHECK_LE(compilation_unit_code_address, method_info->code_address);
     const uint32_t low_pc = dchecked_integral_cast<uint32_t>(
@@ -196,7 +197,8 @@
       is64bitValue,
       compilation_unit_code_address,
       dex_pc_low,
-      dex_pc_high);
+      dex_pc_high,
+      isa);
 
   // Write .debug_loc entries.
   dwarf::Writer<> debug_loc(debug_loc_buffer);
diff --git a/compiler/driver/compiler_driver.cc b/compiler/driver/compiler_driver.cc
index faf8b41..c03ffca 100644
--- a/compiler/driver/compiler_driver.cc
+++ b/compiler/driver/compiler_driver.cc
@@ -284,7 +284,7 @@
       verification_results_(verification_results),
       compiler_(Compiler::Create(this, compiler_kind)),
       compiler_kind_(compiler_kind),
-      instruction_set_(instruction_set == kArm ? kThumb2: instruction_set),
+      instruction_set_(instruction_set == kArm ? kThumb2 : instruction_set),
       instruction_set_features_(instruction_set_features),
       requires_constructor_barrier_lock_("constructor barrier lock"),
       compiled_classes_lock_("compiled classes lock"),
@@ -1251,7 +1251,7 @@
     }
   }
 
-  // java.lang.Reference visitor for VisitReferences.
+  // java.lang.ref.Reference visitor for VisitReferences.
   void operator()(ObjPtr<mirror::Class> klass ATTRIBUTE_UNUSED,
                   ObjPtr<mirror::Reference> ref ATTRIBUTE_UNUSED) const {}
 
diff --git a/compiler/exception_test.cc b/compiler/exception_test.cc
index f9e5cb9..eac46e5 100644
--- a/compiler/exception_test.cc
+++ b/compiler/exception_test.cc
@@ -61,7 +61,7 @@
 
     ArenaPool pool;
     ArenaAllocator allocator(&pool);
-    StackMapStream stack_maps(&allocator);
+    StackMapStream stack_maps(&allocator, kRuntimeISA);
     stack_maps.BeginStackMapEntry(/* dex_pc */ 3u,
                                   /* native_pc_offset */ 3u,
                                   /* register_mask */ 0u,
diff --git a/compiler/image_writer.cc b/compiler/image_writer.cc
index 4109345..459aca3 100644
--- a/compiler/image_writer.cc
+++ b/compiler/image_writer.cc
@@ -1166,9 +1166,9 @@
       // belongs.
       oat_index = GetOatIndexForDexCache(dex_cache);
       ImageInfo& image_info = GetImageInfo(oat_index);
-      {
-        // Note: This table is only accessed from the image writer, avoid locking to prevent lock
-        // order violations from root visiting.
+      if (!compile_app_image_) {
+        // Note: Avoid locking to prevent lock order violations from root visiting;
+        // image_info.class_table_ is only accessed from the image writer.
         image_info.class_table_->InsertWithoutLocks(as_klass);
       }
       for (LengthPrefixedArray<ArtField>* cur_fields : fields) {
@@ -1265,7 +1265,14 @@
       // class loader.
       mirror::ClassLoader* class_loader = obj->AsClassLoader();
       if (class_loader->GetClassTable() != nullptr) {
+        DCHECK(compile_app_image_);
+        DCHECK(class_loaders_.empty());
         class_loaders_.insert(class_loader);
+        ImageInfo& image_info = GetImageInfo(oat_index);
+        // Note: Avoid locking to prevent lock order violations from root visiting;
+        // image_info.class_table_ table is only accessed from the image writer
+        // and class_loader->GetClassTable() is iterated but not modified.
+        image_info.class_table_->CopyWithoutLocks(*class_loader->GetClassTable());
       }
     }
     AssignImageBinSlot(obj, oat_index);
diff --git a/compiler/jni/quick/x86_64/calling_convention_x86_64.cc b/compiler/jni/quick/x86_64/calling_convention_x86_64.cc
index 8ca0ffe..ba654f4 100644
--- a/compiler/jni/quick/x86_64/calling_convention_x86_64.cc
+++ b/compiler/jni/quick/x86_64/calling_convention_x86_64.cc
@@ -160,7 +160,7 @@
     while (HasNext()) {
       ManagedRegister in_reg = CurrentParamRegister();
       if (!in_reg.IsNoRegister()) {
-        int32_t size = IsParamALongOrDouble(itr_args_)? 8 : 4;
+        int32_t size = IsParamALongOrDouble(itr_args_) ? 8 : 4;
         int32_t spill_offset = CurrentParamStackOffset().Uint32Value();
         ManagedRegisterSpill spill(in_reg, size, spill_offset);
         entry_spills_.push_back(spill);
diff --git a/compiler/optimizing/code_generator.cc b/compiler/optimizing/code_generator.cc
index 70c2738..99427f0 100644
--- a/compiler/optimizing/code_generator.cc
+++ b/compiler/optimizing/code_generator.cc
@@ -839,8 +839,8 @@
     // last emitted is different than the native pc of the stack map just emitted.
     size_t number_of_stack_maps = stack_map_stream_.GetNumberOfStackMaps();
     if (number_of_stack_maps > 1) {
-      DCHECK_NE(stack_map_stream_.GetStackMap(number_of_stack_maps - 1).native_pc_offset,
-                stack_map_stream_.GetStackMap(number_of_stack_maps - 2).native_pc_offset);
+      DCHECK_NE(stack_map_stream_.GetStackMap(number_of_stack_maps - 1).native_pc_code_offset,
+                stack_map_stream_.GetStackMap(number_of_stack_maps - 2).native_pc_code_offset);
     }
   }
 }
@@ -848,7 +848,8 @@
 bool CodeGenerator::HasStackMapAtCurrentPc() {
   uint32_t pc = GetAssembler()->CodeSize();
   size_t count = stack_map_stream_.GetNumberOfStackMaps();
-  return count > 0 && stack_map_stream_.GetStackMap(count - 1).native_pc_offset == pc;
+  CodeOffset native_pc_offset = stack_map_stream_.GetStackMap(count - 1).native_pc_code_offset;
+  return (count > 0) && (native_pc_offset.Uint32Value(GetInstructionSet()) == pc);
 }
 
 void CodeGenerator::MaybeRecordNativeDebugInfo(HInstruction* instruction,
diff --git a/compiler/optimizing/code_generator.h b/compiler/optimizing/code_generator.h
index 38d532e..2d129af 100644
--- a/compiler/optimizing/code_generator.h
+++ b/compiler/optimizing/code_generator.h
@@ -608,7 +608,7 @@
         number_of_register_pairs_(number_of_register_pairs),
         core_callee_save_mask_(core_callee_save_mask),
         fpu_callee_save_mask_(fpu_callee_save_mask),
-        stack_map_stream_(graph->GetArena()),
+        stack_map_stream_(graph->GetArena(), graph->GetInstructionSet()),
         block_order_(nullptr),
         jit_string_roots_(StringReferenceValueComparator(),
                           graph->GetArena()->Adapter(kArenaAllocCodeGenerator)),
diff --git a/compiler/optimizing/code_generator_arm.cc b/compiler/optimizing/code_generator_arm.cc
index 9c9c604..f5b6ebe 100644
--- a/compiler/optimizing/code_generator_arm.cc
+++ b/compiler/optimizing/code_generator_arm.cc
@@ -1239,7 +1239,8 @@
 
   // Adjust native pc offsets in stack maps.
   for (size_t i = 0, num = stack_map_stream_.GetNumberOfStackMaps(); i != num; ++i) {
-    uint32_t old_position = stack_map_stream_.GetStackMap(i).native_pc_offset;
+    uint32_t old_position =
+        stack_map_stream_.GetStackMap(i).native_pc_code_offset.Uint32Value(kThumb2);
     uint32_t new_position = __ GetAdjustedPosition(old_position);
     stack_map_stream_.SetStackMapNativePcOffset(i, new_position);
   }
@@ -3331,7 +3332,7 @@
         InvokeRuntimeCallingConvention calling_convention;
         locations->SetInAt(0, Location::RegisterLocation(calling_convention.GetRegisterAt(0)));
         locations->SetInAt(1, Location::RegisterLocation(calling_convention.GetRegisterAt(1)));
-        // Note: divrem will compute both the quotient and the remainder as the pair R0 and R1, but
+        // Note: divmod will compute both the quotient and the remainder as the pair R0 and R1, but
         //       we only need the former.
         locations->SetOut(Location::RegisterLocation(R0));
       }
@@ -3458,7 +3459,7 @@
         InvokeRuntimeCallingConvention calling_convention;
         locations->SetInAt(0, Location::RegisterLocation(calling_convention.GetRegisterAt(0)));
         locations->SetInAt(1, Location::RegisterLocation(calling_convention.GetRegisterAt(1)));
-        // Note: divrem will compute both the quotient and the remainder as the pair R0 and R1, but
+        // Note: divmod will compute both the quotient and the remainder as the pair R0 and R1, but
         //       we only need the latter.
         locations->SetOut(Location::RegisterLocation(R1));
       }
@@ -7150,18 +7151,7 @@
 HInvokeStaticOrDirect::DispatchInfo CodeGeneratorARM::GetSupportedInvokeStaticOrDirectDispatch(
       const HInvokeStaticOrDirect::DispatchInfo& desired_dispatch_info,
       HInvokeStaticOrDirect* invoke ATTRIBUTE_UNUSED) {
-  HInvokeStaticOrDirect::DispatchInfo dispatch_info = desired_dispatch_info;
-  // We disable pc-relative load when there is an irreducible loop, as the optimization
-  // is incompatible with it.
-  // TODO: Create as many ArmDexCacheArraysBase instructions as needed for methods
-  // with irreducible loops.
-  if (GetGraph()->HasIrreducibleLoops() &&
-      (dispatch_info.method_load_kind ==
-          HInvokeStaticOrDirect::MethodLoadKind::kDexCachePcRelative)) {
-    dispatch_info.method_load_kind = HInvokeStaticOrDirect::MethodLoadKind::kDexCacheViaMethod;
-  }
-
-  return dispatch_info;
+  return desired_dispatch_info;
 }
 
 Register CodeGeneratorARM::GetInvokeStaticOrDirectExtraParameter(HInvokeStaticOrDirect* invoke,
@@ -7181,8 +7171,7 @@
   // save one load. However, since this is just an intrinsic slow path we prefer this
   // simple and more robust approach rather that trying to determine if that's the case.
   SlowPathCode* slow_path = GetCurrentSlowPath();
-  DCHECK(slow_path != nullptr);  // For intrinsified invokes the call is emitted on the slow path.
-  if (slow_path->IsCoreRegisterSaved(location.AsRegister<Register>())) {
+  if (slow_path != nullptr && slow_path->IsCoreRegisterSaved(location.AsRegister<Register>())) {
     int stack_offset = slow_path->GetStackOffsetOfCoreRegister(location.AsRegister<Register>());
     __ LoadFromOffset(kLoadWord, temp, SP, stack_offset);
     return temp;
@@ -7190,7 +7179,8 @@
   return location.AsRegister<Register>();
 }
 
-void CodeGeneratorARM::GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp) {
+Location CodeGeneratorARM::GenerateCalleeMethodStaticOrDirectCall(HInvokeStaticOrDirect* invoke,
+                                                                  Location temp) {
   Location callee_method = temp;  // For all kinds except kRecursive, callee will be in temp.
   switch (invoke->GetMethodLoadKind()) {
     case HInvokeStaticOrDirect::MethodLoadKind::kStringInit: {
@@ -7239,6 +7229,11 @@
       break;
     }
   }
+  return callee_method;
+}
+
+void CodeGeneratorARM::GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp) {
+  Location callee_method = GenerateCalleeMethodStaticOrDirectCall(invoke, temp);
 
   switch (invoke->GetCodePtrLocation()) {
     case HInvokeStaticOrDirect::CodePtrLocation::kCallSelf:
diff --git a/compiler/optimizing/code_generator_arm.h b/compiler/optimizing/code_generator_arm.h
index 52d1857..df2dbc7 100644
--- a/compiler/optimizing/code_generator_arm.h
+++ b/compiler/optimizing/code_generator_arm.h
@@ -456,6 +456,7 @@
       const HInvokeStaticOrDirect::DispatchInfo& desired_dispatch_info,
       HInvokeStaticOrDirect* invoke) OVERRIDE;
 
+  Location GenerateCalleeMethodStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp);
   void GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp) OVERRIDE;
   void GenerateVirtualCall(HInvokeVirtual* invoke, Location temp) OVERRIDE;
 
diff --git a/compiler/optimizing/code_generator_arm64.cc b/compiler/optimizing/code_generator_arm64.cc
index 68d0b86..9762ee8 100644
--- a/compiler/optimizing/code_generator_arm64.cc
+++ b/compiler/optimizing/code_generator_arm64.cc
@@ -3993,7 +3993,8 @@
   return desired_dispatch_info;
 }
 
-void CodeGeneratorARM64::GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp) {
+Location CodeGeneratorARM64::GenerateCalleeMethodStaticOrDirectCall(HInvokeStaticOrDirect* invoke,
+                                                                    Location temp) {
   // Make sure that ArtMethod* is passed in kArtMethodRegister as per the calling convention.
   Location callee_method = temp;  // For all kinds except kRecursive, callee will be in temp.
   switch (invoke->GetMethodLoadKind()) {
@@ -4047,6 +4048,12 @@
       break;
     }
   }
+  return callee_method;
+}
+
+void CodeGeneratorARM64::GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp) {
+  // All registers are assumed to be correctly set up.
+  Location callee_method = GenerateCalleeMethodStaticOrDirectCall(invoke, temp);
 
   switch (invoke->GetCodePtrLocation()) {
     case HInvokeStaticOrDirect::CodePtrLocation::kCallSelf:
@@ -4655,7 +4662,7 @@
 }
 
 void InstructionCodeGeneratorARM64::VisitMonitorOperation(HMonitorOperation* instruction) {
-  codegen_->InvokeRuntime(instruction->IsEnter() ? kQuickLockObject: kQuickUnlockObject,
+  codegen_->InvokeRuntime(instruction->IsEnter() ? kQuickLockObject : kQuickUnlockObject,
                           instruction,
                           instruction->GetDexPc());
   if (instruction->IsEnter()) {
diff --git a/compiler/optimizing/code_generator_arm64.h b/compiler/optimizing/code_generator_arm64.h
index a9dca92..7d3c655 100644
--- a/compiler/optimizing/code_generator_arm64.h
+++ b/compiler/optimizing/code_generator_arm64.h
@@ -527,6 +527,7 @@
       const HInvokeStaticOrDirect::DispatchInfo& desired_dispatch_info,
       HInvokeStaticOrDirect* invoke) OVERRIDE;
 
+  Location GenerateCalleeMethodStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp);
   void GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp) OVERRIDE;
   void GenerateVirtualCall(HInvokeVirtual* invoke, Location temp) OVERRIDE;
 
diff --git a/compiler/optimizing/code_generator_arm_vixl.cc b/compiler/optimizing/code_generator_arm_vixl.cc
index 592ee5a..ffaf18f 100644
--- a/compiler/optimizing/code_generator_arm_vixl.cc
+++ b/compiler/optimizing/code_generator_arm_vixl.cc
@@ -3336,7 +3336,7 @@
         InvokeRuntimeCallingConventionARMVIXL calling_convention;
         locations->SetInAt(0, LocationFrom(calling_convention.GetRegisterAt(0)));
         locations->SetInAt(1, LocationFrom(calling_convention.GetRegisterAt(1)));
-        // Note: divrem will compute both the quotient and the remainder as the pair R0 and R1, but
+        // Note: divmod will compute both the quotient and the remainder as the pair R0 and R1, but
         //       we only need the former.
         locations->SetOut(LocationFrom(r0));
       }
@@ -3450,7 +3450,7 @@
         InvokeRuntimeCallingConventionARMVIXL calling_convention;
         locations->SetInAt(0, LocationFrom(calling_convention.GetRegisterAt(0)));
         locations->SetInAt(1, LocationFrom(calling_convention.GetRegisterAt(1)));
-        // Note: divrem will compute both the quotient and the remainder as the pair R0 and R1, but
+        // Note: divmod will compute both the quotient and the remainder as the pair R0 and R1, but
         //       we only need the latter.
         locations->SetOut(LocationFrom(r1));
       }
@@ -7233,18 +7233,7 @@
 HInvokeStaticOrDirect::DispatchInfo CodeGeneratorARMVIXL::GetSupportedInvokeStaticOrDirectDispatch(
     const HInvokeStaticOrDirect::DispatchInfo& desired_dispatch_info,
     HInvokeStaticOrDirect* invoke ATTRIBUTE_UNUSED) {
-  HInvokeStaticOrDirect::DispatchInfo dispatch_info = desired_dispatch_info;
-  // We disable pc-relative load when there is an irreducible loop, as the optimization
-  // is incompatible with it.
-  // TODO: Create as many ArmDexCacheArraysBase instructions as needed for methods
-  // with irreducible loops.
-  if (GetGraph()->HasIrreducibleLoops() &&
-      (dispatch_info.method_load_kind ==
-          HInvokeStaticOrDirect::MethodLoadKind::kDexCachePcRelative)) {
-    dispatch_info.method_load_kind = HInvokeStaticOrDirect::MethodLoadKind::kDexCacheViaMethod;
-  }
-
-  return dispatch_info;
+  return desired_dispatch_info;
 }
 
 vixl32::Register CodeGeneratorARMVIXL::GetInvokeStaticOrDirectExtraParameter(
@@ -7273,7 +7262,7 @@
   return RegisterFrom(location);
 }
 
-void CodeGeneratorARMVIXL::GenerateStaticOrDirectCall(
+Location CodeGeneratorARMVIXL::GenerateCalleeMethodStaticOrDirectCall(
     HInvokeStaticOrDirect* invoke, Location temp) {
   Location callee_method = temp;  // For all kinds except kRecursive, callee will be in temp.
   switch (invoke->GetMethodLoadKind()) {
@@ -7324,6 +7313,12 @@
       break;
     }
   }
+  return callee_method;
+}
+
+void CodeGeneratorARMVIXL::GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke,
+                                                      Location temp) {
+  Location callee_method = GenerateCalleeMethodStaticOrDirectCall(invoke, temp);
 
   switch (invoke->GetCodePtrLocation()) {
     case HInvokeStaticOrDirect::CodePtrLocation::kCallSelf:
diff --git a/compiler/optimizing/code_generator_arm_vixl.h b/compiler/optimizing/code_generator_arm_vixl.h
index be65353..8ae3b7d 100644
--- a/compiler/optimizing/code_generator_arm_vixl.h
+++ b/compiler/optimizing/code_generator_arm_vixl.h
@@ -537,6 +537,7 @@
       const HInvokeStaticOrDirect::DispatchInfo& desired_dispatch_info,
       HInvokeStaticOrDirect* invoke) OVERRIDE;
 
+  Location GenerateCalleeMethodStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp);
   void GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp) OVERRIDE;
   void GenerateVirtualCall(HInvokeVirtual* invoke, Location temp) OVERRIDE;
 
diff --git a/compiler/optimizing/code_generator_mips.cc b/compiler/optimizing/code_generator_mips.cc
index a038382..76be74e 100644
--- a/compiler/optimizing/code_generator_mips.cc
+++ b/compiler/optimizing/code_generator_mips.cc
@@ -496,7 +496,8 @@
 
   // Adjust native pc offsets in stack maps.
   for (size_t i = 0, num = stack_map_stream_.GetNumberOfStackMaps(); i != num; ++i) {
-    uint32_t old_position = stack_map_stream_.GetStackMap(i).native_pc_offset;
+    uint32_t old_position =
+        stack_map_stream_.GetStackMap(i).native_pc_code_offset.Uint32Value(kMips);
     uint32_t new_position = __ GetAdjustedPosition(old_position);
     DCHECK_GE(new_position, old_position);
     stack_map_stream_.SetStackMapNativePcOffset(i, new_position);
diff --git a/compiler/optimizing/code_generator_mips64.cc b/compiler/optimizing/code_generator_mips64.cc
index 446dea6..192b4a5 100644
--- a/compiler/optimizing/code_generator_mips64.cc
+++ b/compiler/optimizing/code_generator_mips64.cc
@@ -450,7 +450,8 @@
 
   // Adjust native pc offsets in stack maps.
   for (size_t i = 0, num = stack_map_stream_.GetNumberOfStackMaps(); i != num; ++i) {
-    uint32_t old_position = stack_map_stream_.GetStackMap(i).native_pc_offset;
+    uint32_t old_position =
+        stack_map_stream_.GetStackMap(i).native_pc_code_offset.Uint32Value(kMips64);
     uint32_t new_position = __ GetAdjustedPosition(old_position);
     DCHECK_GE(new_position, old_position);
     stack_map_stream_.SetStackMapNativePcOffset(i, new_position);
diff --git a/compiler/optimizing/code_generator_x86.cc b/compiler/optimizing/code_generator_x86.cc
index 853c91f..5c561f5 100644
--- a/compiler/optimizing/code_generator_x86.cc
+++ b/compiler/optimizing/code_generator_x86.cc
@@ -1778,7 +1778,7 @@
         cond = X86Condition(condition->GetCondition());
       }
     } else {
-      // Must be a boolean condition, which needs to be compared to 0.
+      // Must be a Boolean condition, which needs to be compared to 0.
       Register cond_reg = locations->InAt(2).AsRegister<Register>();
       __ testl(cond_reg, cond_reg);
     }
diff --git a/compiler/optimizing/code_generator_x86_64.cc b/compiler/optimizing/code_generator_x86_64.cc
index 74c71cc..c4caf4b 100644
--- a/compiler/optimizing/code_generator_x86_64.cc
+++ b/compiler/optimizing/code_generator_x86_64.cc
@@ -1809,7 +1809,7 @@
         cond = X86_64IntegerCondition(condition->GetCondition());
       }
     } else {
-      // Must be a boolean condition, which needs to be compared to 0.
+      // Must be a Boolean condition, which needs to be compared to 0.
       CpuRegister cond_reg = locations->InAt(2).AsRegister<CpuRegister>();
       __ testl(cond_reg, cond_reg);
     }
@@ -4210,7 +4210,7 @@
 
 void CodeGeneratorX86_64::GenerateMemoryBarrier(MemBarrierKind kind) {
   /*
-   * According to the JSR-133 Cookbook, for x86 only StoreLoad/AnyAny barriers need memory fence.
+   * According to the JSR-133 Cookbook, for x86-64 only StoreLoad/AnyAny barriers need memory fence.
    * All other barriers (LoadAny, AnyStore, StoreStore) are nops due to the x86-64 memory model.
    * For those cases, all we need to ensure is that there is a scheduling barrier in place.
    */
diff --git a/compiler/optimizing/dex_cache_array_fixups_arm.cc b/compiler/optimizing/dex_cache_array_fixups_arm.cc
index 9ddcd56..cfcb276 100644
--- a/compiler/optimizing/dex_cache_array_fixups_arm.cc
+++ b/compiler/optimizing/dex_cache_array_fixups_arm.cc
@@ -65,7 +65,7 @@
     if (invoke->HasPcRelativeDexCache() &&
         !IsCallFreeIntrinsic<IntrinsicLocationsBuilderARMType>(invoke, codegen_)) {
       HArmDexCacheArraysBase* base =
-          GetOrCreateDexCacheArrayBase(invoke->GetDexFileForPcRelativeDexCache());
+          GetOrCreateDexCacheArrayBase(invoke, invoke->GetDexFileForPcRelativeDexCache());
       // Update the element offset in base.
       DexCacheArraysLayout layout(kArmPointerSize, &invoke->GetDexFileForPcRelativeDexCache());
       base->UpdateElementOffset(layout.MethodOffset(invoke->GetDexMethodIndex()));
@@ -75,21 +75,28 @@
     }
   }
 
-  HArmDexCacheArraysBase* GetOrCreateDexCacheArrayBase(const DexFile& dex_file) {
-    // Ensure we only initialize the pointer once for each dex file.
-    auto lb = dex_cache_array_bases_.lower_bound(&dex_file);
-    if (lb != dex_cache_array_bases_.end() &&
-        !dex_cache_array_bases_.key_comp()(&dex_file, lb->first)) {
-      return lb->second;
-    }
+  HArmDexCacheArraysBase* GetOrCreateDexCacheArrayBase(HInstruction* cursor,
+                                                       const DexFile& dex_file) {
+    if (GetGraph()->HasIrreducibleLoops()) {
+      HArmDexCacheArraysBase* base = new (GetGraph()->GetArena()) HArmDexCacheArraysBase(dex_file);
+      cursor->GetBlock()->InsertInstructionBefore(base, cursor);
+      return base;
+    } else {
+      // Ensure we only initialize the pointer once for each dex file.
+      auto lb = dex_cache_array_bases_.lower_bound(&dex_file);
+      if (lb != dex_cache_array_bases_.end() &&
+          !dex_cache_array_bases_.key_comp()(&dex_file, lb->first)) {
+        return lb->second;
+      }
 
-    // Insert the base at the start of the entry block, move it to a better
-    // position later in MoveBaseIfNeeded().
-    HArmDexCacheArraysBase* base = new (GetGraph()->GetArena()) HArmDexCacheArraysBase(dex_file);
-    HBasicBlock* entry_block = GetGraph()->GetEntryBlock();
-    entry_block->InsertInstructionBefore(base, entry_block->GetFirstInstruction());
-    dex_cache_array_bases_.PutBefore(lb, &dex_file, base);
-    return base;
+      // Insert the base at the start of the entry block, move it to a better
+      // position later in MoveBaseIfNeeded().
+      HArmDexCacheArraysBase* base = new (GetGraph()->GetArena()) HArmDexCacheArraysBase(dex_file);
+      HBasicBlock* entry_block = GetGraph()->GetEntryBlock();
+      entry_block->InsertInstructionBefore(base, entry_block->GetFirstInstruction());
+      dex_cache_array_bases_.PutBefore(lb, &dex_file, base);
+      return base;
+    }
   }
 
   CodeGeneratorARMType* codegen_;
@@ -100,11 +107,6 @@
 };
 
 void DexCacheArrayFixups::Run() {
-  if (graph_->HasIrreducibleLoops()) {
-    // Do not run this optimization, as irreducible loops do not work with an instruction
-    // that can be live-in at the irreducible loop header.
-    return;
-  }
   DexCacheArrayFixupsVisitor visitor(graph_, codegen_);
   visitor.VisitInsertionOrder();
   visitor.MoveBasesIfNeeded();
diff --git a/compiler/optimizing/gvn.cc b/compiler/optimizing/gvn.cc
index f5931a2..c93bc21 100644
--- a/compiler/optimizing/gvn.cc
+++ b/compiler/optimizing/gvn.cc
@@ -399,7 +399,7 @@
   ArenaVector<ValueSet*> sets_;
 
   // BitVector which serves as a fast-access map from block id to
-  // visited/unvisited boolean.
+  // visited/unvisited Boolean.
   ArenaBitVector visited_blocks_;
 
   DISALLOW_COPY_AND_ASSIGN(GlobalValueNumberer);
diff --git a/compiler/optimizing/instruction_builder.cc b/compiler/optimizing/instruction_builder.cc
index ef8d74d..cac385c 100644
--- a/compiler/optimizing/instruction_builder.cc
+++ b/compiler/optimizing/instruction_builder.cc
@@ -1,4 +1,3 @@
-
 /*
  * Copyright (C) 2016 The Android Open Source Project
  *
diff --git a/compiler/optimizing/intrinsics_arm.cc b/compiler/optimizing/intrinsics_arm.cc
index 8f64fae..c262cf9 100644
--- a/compiler/optimizing/intrinsics_arm.cc
+++ b/compiler/optimizing/intrinsics_arm.cc
@@ -2592,6 +2592,58 @@
   __ Lsr(out, out, 5);
 }
 
+void IntrinsicLocationsBuilderARM::VisitReferenceGetReferent(HInvoke* invoke) {
+  if (kEmitCompilerReadBarrier) {
+    // Do not intrinsify this call with the read barrier configuration.
+    return;
+  }
+  LocationSummary* locations = new (arena_) LocationSummary(invoke,
+                                                            LocationSummary::kCallOnSlowPath,
+                                                            kIntrinsified);
+  locations->SetInAt(0, Location::RequiresRegister());
+  locations->SetOut(Location::SameAsFirstInput());
+  locations->AddTemp(Location::RequiresRegister());
+}
+
+void IntrinsicCodeGeneratorARM::VisitReferenceGetReferent(HInvoke* invoke) {
+  DCHECK(!kEmitCompilerReadBarrier);
+  ArmAssembler* const assembler = GetAssembler();
+  LocationSummary* locations = invoke->GetLocations();
+
+  Register obj = locations->InAt(0).AsRegister<Register>();
+  Register out = locations->Out().AsRegister<Register>();
+
+  SlowPathCode* slow_path = new (GetAllocator()) IntrinsicSlowPathARM(invoke);
+  codegen_->AddSlowPath(slow_path);
+
+  // Load ArtMethod first.
+  HInvokeStaticOrDirect* invoke_direct = invoke->AsInvokeStaticOrDirect();
+  DCHECK(invoke_direct != nullptr);
+  Register temp = codegen_->GenerateCalleeMethodStaticOrDirectCall(
+      invoke_direct, locations->GetTemp(0)).AsRegister<Register>();
+
+  // Now get declaring class.
+  __ ldr(temp, Address(temp, ArtMethod::DeclaringClassOffset().Int32Value()));
+
+  uint32_t slow_path_flag_offset = codegen_->GetReferenceSlowFlagOffset();
+  uint32_t disable_flag_offset = codegen_->GetReferenceDisableFlagOffset();
+  DCHECK_NE(slow_path_flag_offset, 0u);
+  DCHECK_NE(disable_flag_offset, 0u);
+  DCHECK_NE(slow_path_flag_offset, disable_flag_offset);
+
+  // Check static flags that prevent using intrinsic.
+  __ ldr(IP, Address(temp, disable_flag_offset));
+  __ ldr(temp, Address(temp, slow_path_flag_offset));
+  __ orr(IP, IP, ShifterOperand(temp));
+  __ CompareAndBranchIfNonZero(IP, slow_path->GetEntryLabel());
+
+  // Fast path.
+  __ ldr(out, Address(obj, mirror::Reference::ReferentOffset().Int32Value()));
+  codegen_->MaybeRecordImplicitNullCheck(invoke);
+  __ MaybeUnpoisonHeapReference(out);
+  __ Bind(slow_path->GetExitLabel());
+}
+
 UNIMPLEMENTED_INTRINSIC(ARM, MathMinDoubleDouble)
 UNIMPLEMENTED_INTRINSIC(ARM, MathMinFloatFloat)
 UNIMPLEMENTED_INTRINSIC(ARM, MathMaxDoubleDouble)
@@ -2605,7 +2657,6 @@
 UNIMPLEMENTED_INTRINSIC(ARM, MathRoundFloat)    // Could be done by changing rounding mode, maybe?
 UNIMPLEMENTED_INTRINSIC(ARM, UnsafeCASLong)     // High register pressure.
 UNIMPLEMENTED_INTRINSIC(ARM, SystemArrayCopyChar)
-UNIMPLEMENTED_INTRINSIC(ARM, ReferenceGetReferent)
 UNIMPLEMENTED_INTRINSIC(ARM, IntegerHighestOneBit)
 UNIMPLEMENTED_INTRINSIC(ARM, LongHighestOneBit)
 UNIMPLEMENTED_INTRINSIC(ARM, IntegerLowestOneBit)
diff --git a/compiler/optimizing/intrinsics_arm64.cc b/compiler/optimizing/intrinsics_arm64.cc
index d8a896e..bbf826c 100644
--- a/compiler/optimizing/intrinsics_arm64.cc
+++ b/compiler/optimizing/intrinsics_arm64.cc
@@ -2773,7 +2773,65 @@
   GenIsInfinite(invoke->GetLocations(), /* is64bit */ true, GetVIXLAssembler());
 }
 
-UNIMPLEMENTED_INTRINSIC(ARM64, ReferenceGetReferent)
+void IntrinsicLocationsBuilderARM64::VisitReferenceGetReferent(HInvoke* invoke) {
+  if (kEmitCompilerReadBarrier) {
+    // Do not intrinsify this call with the read barrier configuration.
+    return;
+  }
+  LocationSummary* locations = new (arena_) LocationSummary(invoke,
+                                                            LocationSummary::kCallOnSlowPath,
+                                                            kIntrinsified);
+  locations->SetInAt(0, Location::RequiresRegister());
+  locations->SetOut(Location::SameAsFirstInput());
+  locations->AddTemp(Location::RequiresRegister());
+}
+
+void IntrinsicCodeGeneratorARM64::VisitReferenceGetReferent(HInvoke* invoke) {
+  DCHECK(!kEmitCompilerReadBarrier);
+  MacroAssembler* masm = GetVIXLAssembler();
+  LocationSummary* locations = invoke->GetLocations();
+
+  Register obj = InputRegisterAt(invoke, 0);
+  Register out = OutputRegister(invoke);
+
+  SlowPathCodeARM64* slow_path = new (GetAllocator()) IntrinsicSlowPathARM64(invoke);
+  codegen_->AddSlowPath(slow_path);
+
+  // Load ArtMethod first.
+  HInvokeStaticOrDirect* invoke_direct = invoke->AsInvokeStaticOrDirect();
+  DCHECK(invoke_direct != nullptr);
+  Register temp0 = XRegisterFrom(codegen_->GenerateCalleeMethodStaticOrDirectCall(
+                                 invoke_direct, locations->GetTemp(0)));
+
+  // Now get declaring class.
+  __ Ldr(temp0.W(), MemOperand(temp0, ArtMethod::DeclaringClassOffset().Int32Value()));
+
+  uint32_t slow_path_flag_offset = codegen_->GetReferenceSlowFlagOffset();
+  uint32_t disable_flag_offset = codegen_->GetReferenceDisableFlagOffset();
+  DCHECK_NE(slow_path_flag_offset, 0u);
+  DCHECK_NE(disable_flag_offset, 0u);
+  DCHECK_NE(slow_path_flag_offset, disable_flag_offset);
+
+  // Check static flags that prevent using intrinsic.
+  if (slow_path_flag_offset == disable_flag_offset + 1) {
+    // Load two adjacent flags in one 64-bit load.
+    __ Ldr(temp0, MemOperand(temp0, disable_flag_offset));
+  } else {
+    UseScratchRegisterScope temps(masm);
+    Register temp1 = temps.AcquireW();
+    __ Ldr(temp1.W(), MemOperand(temp0, disable_flag_offset));
+    __ Ldr(temp0.W(), MemOperand(temp0, slow_path_flag_offset));
+    __ Orr(temp0, temp1, temp0);
+  }
+  __ Cbnz(temp0, slow_path->GetEntryLabel());
+
+  // Fast path.
+  __ Ldr(out, HeapOperand(obj, mirror::Reference::ReferentOffset().Int32Value()));
+  codegen_->MaybeRecordImplicitNullCheck(invoke);
+  codegen_->GetAssembler()->MaybeUnpoisonHeapReference(out);
+  __ Bind(slow_path->GetExitLabel());
+}
+
 UNIMPLEMENTED_INTRINSIC(ARM64, IntegerHighestOneBit)
 UNIMPLEMENTED_INTRINSIC(ARM64, LongHighestOneBit)
 UNIMPLEMENTED_INTRINSIC(ARM64, IntegerLowestOneBit)
diff --git a/compiler/optimizing/intrinsics_arm_vixl.cc b/compiler/optimizing/intrinsics_arm_vixl.cc
index 85e84d8..68c2d2e 100644
--- a/compiler/optimizing/intrinsics_arm_vixl.cc
+++ b/compiler/optimizing/intrinsics_arm_vixl.cc
@@ -2688,6 +2688,60 @@
   __ Lsr(out, out, 5);
 }
 
+void IntrinsicLocationsBuilderARMVIXL::VisitReferenceGetReferent(HInvoke* invoke) {
+  if (kEmitCompilerReadBarrier) {
+    // Do not intrinsify this call with the read barrier configuration.
+    return;
+  }
+  LocationSummary* locations = new (arena_) LocationSummary(invoke,
+                                                            LocationSummary::kCallOnSlowPath,
+                                                            kIntrinsified);
+  locations->SetInAt(0, Location::RequiresRegister());
+  locations->SetOut(Location::SameAsFirstInput());
+  locations->AddTemp(Location::RequiresRegister());
+}
+
+void IntrinsicCodeGeneratorARMVIXL::VisitReferenceGetReferent(HInvoke* invoke) {
+  DCHECK(!kEmitCompilerReadBarrier);
+  ArmVIXLAssembler* assembler = GetAssembler();
+  LocationSummary* locations = invoke->GetLocations();
+
+  vixl32::Register obj = InputRegisterAt(invoke, 0);
+  vixl32::Register out = OutputRegister(invoke);
+
+  SlowPathCodeARMVIXL* slow_path = new (GetAllocator()) IntrinsicSlowPathARMVIXL(invoke);
+  codegen_->AddSlowPath(slow_path);
+
+  // Load ArtMethod first.
+  HInvokeStaticOrDirect* invoke_direct = invoke->AsInvokeStaticOrDirect();
+  DCHECK(invoke_direct != nullptr);
+  vixl32::Register temp0 = RegisterFrom(codegen_->GenerateCalleeMethodStaticOrDirectCall(
+      invoke_direct, locations->GetTemp(0)));
+
+  // Now get declaring class.
+  __ Ldr(temp0, MemOperand(temp0, ArtMethod::DeclaringClassOffset().Int32Value()));
+
+  uint32_t slow_path_flag_offset = codegen_->GetReferenceSlowFlagOffset();
+  uint32_t disable_flag_offset = codegen_->GetReferenceDisableFlagOffset();
+  DCHECK_NE(slow_path_flag_offset, 0u);
+  DCHECK_NE(disable_flag_offset, 0u);
+  DCHECK_NE(slow_path_flag_offset, disable_flag_offset);
+
+  // Check static flags that prevent using intrinsic.
+  UseScratchRegisterScope temps(assembler->GetVIXLAssembler());
+  vixl32::Register temp1 = temps.Acquire();
+  __ Ldr(temp1, MemOperand(temp0, disable_flag_offset));
+  __ Ldr(temp0, MemOperand(temp0, slow_path_flag_offset));
+  __ Orr(temp0, temp1, temp0);
+  __ CompareAndBranchIfNonZero(temp0, slow_path->GetEntryLabel());
+
+  // Fast path.
+  __ Ldr(out, MemOperand(obj, mirror::Reference::ReferentOffset().Int32Value()));
+  codegen_->MaybeRecordImplicitNullCheck(invoke);
+  assembler->MaybeUnpoisonHeapReference(out);
+  __ Bind(slow_path->GetExitLabel());
+}
+
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, MathMinDoubleDouble)
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, MathMinFloatFloat)
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, MathMaxDoubleDouble)
@@ -2701,7 +2755,6 @@
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, MathRoundFloat)    // Could be done by changing rounding mode, maybe?
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, UnsafeCASLong)     // High register pressure.
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, SystemArrayCopyChar)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, ReferenceGetReferent)
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, IntegerHighestOneBit)
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, LongHighestOneBit)
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, IntegerLowestOneBit)
diff --git a/compiler/optimizing/intrinsics_mips.cc b/compiler/optimizing/intrinsics_mips.cc
index f1ae549..6cf9b83 100644
--- a/compiler/optimizing/intrinsics_mips.cc
+++ b/compiler/optimizing/intrinsics_mips.cc
@@ -1878,7 +1878,7 @@
                         // If we use 'value' directly, we would lose 'value'
                         // in the case that the store fails.  Whether the
                         // store succeeds, or fails, it will load the
-                        // correct boolean value into the 'out' register.
+                        // correct Boolean value into the 'out' register.
   // This test isn't really necessary. We only support Primitive::kPrimInt,
   // Primitive::kPrimNot, and we already verified that we're working on one
   // of those two types. It's left here in case the code needs to support
diff --git a/compiler/optimizing/intrinsics_mips64.cc b/compiler/optimizing/intrinsics_mips64.cc
index 3022e97..00a1fa1 100644
--- a/compiler/optimizing/intrinsics_mips64.cc
+++ b/compiler/optimizing/intrinsics_mips64.cc
@@ -1477,7 +1477,7 @@
                         // If we use 'value' directly, we would lose 'value'
                         // in the case that the store fails.  Whether the
                         // store succeeds, or fails, it will load the
-                        // correct boolean value into the 'out' register.
+                        // correct Boolean value into the 'out' register.
   if (type == Primitive::kPrimLong) {
     __ Scd(out, TMP);
   } else {
diff --git a/compiler/optimizing/nodes.h b/compiler/optimizing/nodes.h
index a2980dc..f0ea9e2 100644
--- a/compiler/optimizing/nodes.h
+++ b/compiler/optimizing/nodes.h
@@ -565,7 +565,7 @@
   ArtMethod* GetArtMethod() const { return art_method_; }
   void SetArtMethod(ArtMethod* method) { art_method_ = method; }
 
-  // Returns an instruction with the opposite boolean value from 'cond'.
+  // Returns an instruction with the opposite Boolean value from 'cond'.
   // The instruction has been inserted into the graph, either as a constant, or
   // before cursor.
   HInstruction* InsertOppositeCondition(HInstruction* cond, HInstruction* cursor);
diff --git a/compiler/optimizing/stack_map_stream.cc b/compiler/optimizing/stack_map_stream.cc
index 6087e36..a9a1e6f 100644
--- a/compiler/optimizing/stack_map_stream.cc
+++ b/compiler/optimizing/stack_map_stream.cc
@@ -31,7 +31,7 @@
   DCHECK_EQ(0u, current_entry_.dex_pc) << "EndStackMapEntry not called after BeginStackMapEntry";
   DCHECK_NE(dex_pc, static_cast<uint32_t>(-1)) << "invalid dex_pc";
   current_entry_.dex_pc = dex_pc;
-  current_entry_.native_pc_offset = native_pc_offset;
+  current_entry_.native_pc_code_offset = CodeOffset::FromOffset(native_pc_offset, instruction_set_);
   current_entry_.register_mask = register_mask;
   current_entry_.sp_mask = sp_mask;
   current_entry_.num_dex_registers = num_dex_registers;
@@ -144,10 +144,10 @@
   current_inline_info_ = InlineInfoEntry();
 }
 
-uint32_t StackMapStream::ComputeMaxNativePcOffset() const {
-  uint32_t max_native_pc_offset = 0u;
+CodeOffset StackMapStream::ComputeMaxNativePcCodeOffset() const {
+  CodeOffset max_native_pc_offset;
   for (const StackMapEntry& entry : stack_maps_) {
-    max_native_pc_offset = std::max(max_native_pc_offset, entry.native_pc_offset);
+    max_native_pc_offset = std::max(max_native_pc_offset, entry.native_pc_code_offset);
   }
   return max_native_pc_offset;
 }
@@ -157,8 +157,9 @@
   dex_register_maps_size_ = ComputeDexRegisterMapsSize();
   ComputeInlineInfoEncoding();  // needs dex_register_maps_size_.
   inline_info_size_ = inline_infos_.size() * inline_info_encoding_.GetEntrySize();
-  uint32_t max_native_pc_offset = ComputeMaxNativePcOffset();
-  size_t stack_map_size = stack_map_encoding_.SetFromSizes(max_native_pc_offset,
+  CodeOffset max_native_pc_offset = ComputeMaxNativePcCodeOffset();
+  // The stack map contains compressed native offsets.
+  size_t stack_map_size = stack_map_encoding_.SetFromSizes(max_native_pc_offset.CompressedValue(),
                                                            dex_pc_max_,
                                                            dex_register_maps_size_,
                                                            inline_info_size_,
@@ -319,7 +320,7 @@
     StackMapEntry entry = stack_maps_[i];
 
     stack_map.SetDexPc(stack_map_encoding_, entry.dex_pc);
-    stack_map.SetNativePcOffset(stack_map_encoding_, entry.native_pc_offset);
+    stack_map.SetNativePcCodeOffset(stack_map_encoding_, entry.native_pc_code_offset);
     stack_map.SetRegisterMask(stack_map_encoding_, entry.register_mask);
     size_t number_of_stack_mask_bits = stack_map.GetNumberOfStackMaskBits(stack_map_encoding_);
     if (entry.sp_mask != nullptr) {
@@ -546,7 +547,8 @@
     StackMapEntry entry = stack_maps_[s];
 
     // Check main stack map fields.
-    DCHECK_EQ(stack_map.GetNativePcOffset(stack_map_encoding), entry.native_pc_offset);
+    DCHECK_EQ(stack_map.GetNativePcOffset(stack_map_encoding, instruction_set_),
+              entry.native_pc_code_offset.Uint32Value(instruction_set_));
     DCHECK_EQ(stack_map.GetDexPc(stack_map_encoding), entry.dex_pc);
     DCHECK_EQ(stack_map.GetRegisterMask(stack_map_encoding), entry.register_mask);
     size_t num_stack_mask_bits = stack_map.GetNumberOfStackMaskBits(stack_map_encoding);
diff --git a/compiler/optimizing/stack_map_stream.h b/compiler/optimizing/stack_map_stream.h
index d6f42b3..8fec472 100644
--- a/compiler/optimizing/stack_map_stream.h
+++ b/compiler/optimizing/stack_map_stream.h
@@ -59,8 +59,10 @@
  */
 class StackMapStream : public ValueObject {
  public:
-  explicit StackMapStream(ArenaAllocator* allocator)
+  explicit StackMapStream(ArenaAllocator* allocator,
+                          InstructionSet instruction_set)
       : allocator_(allocator),
+        instruction_set_(instruction_set),
         stack_maps_(allocator->Adapter(kArenaAllocStackMapStream)),
         location_catalog_entries_(allocator->Adapter(kArenaAllocStackMapStream)),
         location_catalog_entries_indices_(allocator->Adapter(kArenaAllocStackMapStream)),
@@ -95,7 +97,7 @@
   // See runtime/stack_map.h to know what these fields contain.
   struct StackMapEntry {
     uint32_t dex_pc;
-    uint32_t native_pc_offset;
+    CodeOffset native_pc_code_offset;
     uint32_t register_mask;
     BitVector* sp_mask;
     uint32_t num_dex_registers;
@@ -141,11 +143,9 @@
   }
 
   void SetStackMapNativePcOffset(size_t i, uint32_t native_pc_offset) {
-    stack_maps_[i].native_pc_offset = native_pc_offset;
+    stack_maps_[i].native_pc_code_offset = CodeOffset::FromOffset(native_pc_offset, instruction_set_);
   }
 
-  uint32_t ComputeMaxNativePcOffset() const;
-
   // Prepares the stream to fill in a memory region. Must be called before FillIn.
   // Returns the size (in bytes) needed to store this stream.
   size_t PrepareForFillIn();
@@ -158,6 +158,8 @@
   size_t ComputeDexRegisterMapsSize() const;
   void ComputeInlineInfoEncoding();
 
+  CodeOffset ComputeMaxNativePcCodeOffset() const;
+
   // Returns the index of an entry with the same dex register map as the current_entry,
   // or kNoSameDexMapFound if no such entry exists.
   size_t FindEntryWithTheSameDexMap();
@@ -175,6 +177,7 @@
   void CheckCodeInfo(MemoryRegion region) const;
 
   ArenaAllocator* allocator_;
+  const InstructionSet instruction_set_;
   ArenaVector<StackMapEntry> stack_maps_;
 
   // A catalog of unique [location_kind, register_value] pairs (per method).
diff --git a/compiler/optimizing/stack_map_test.cc b/compiler/optimizing/stack_map_test.cc
index 22810ea..f68695b 100644
--- a/compiler/optimizing/stack_map_test.cc
+++ b/compiler/optimizing/stack_map_test.cc
@@ -47,7 +47,7 @@
 TEST(StackMapTest, Test1) {
   ArenaPool pool;
   ArenaAllocator arena(&pool);
-  StackMapStream stream(&arena);
+  StackMapStream stream(&arena, kRuntimeISA);
 
   ArenaBitVector sp_mask(&arena, 0, false);
   size_t number_of_dex_registers = 2;
@@ -78,7 +78,7 @@
   ASSERT_TRUE(stack_map.Equals(code_info.GetStackMapForDexPc(0, encoding)));
   ASSERT_TRUE(stack_map.Equals(code_info.GetStackMapForNativePcOffset(64, encoding)));
   ASSERT_EQ(0u, stack_map.GetDexPc(encoding.stack_map_encoding));
-  ASSERT_EQ(64u, stack_map.GetNativePcOffset(encoding.stack_map_encoding));
+  ASSERT_EQ(64u, stack_map.GetNativePcOffset(encoding.stack_map_encoding, kRuntimeISA));
   ASSERT_EQ(0x3u, stack_map.GetRegisterMask(encoding.stack_map_encoding));
 
   ASSERT_TRUE(CheckStackMask(stack_map, encoding.stack_map_encoding, sp_mask));
@@ -128,7 +128,7 @@
 TEST(StackMapTest, Test2) {
   ArenaPool pool;
   ArenaAllocator arena(&pool);
-  StackMapStream stream(&arena);
+  StackMapStream stream(&arena, kRuntimeISA);
   ArtMethod art_method;
 
   ArenaBitVector sp_mask1(&arena, 0, true);
@@ -193,7 +193,7 @@
     ASSERT_TRUE(stack_map.Equals(code_info.GetStackMapForDexPc(0, encoding)));
     ASSERT_TRUE(stack_map.Equals(code_info.GetStackMapForNativePcOffset(64, encoding)));
     ASSERT_EQ(0u, stack_map.GetDexPc(encoding.stack_map_encoding));
-    ASSERT_EQ(64u, stack_map.GetNativePcOffset(encoding.stack_map_encoding));
+    ASSERT_EQ(64u, stack_map.GetNativePcOffset(encoding.stack_map_encoding, kRuntimeISA));
     ASSERT_EQ(0x3u, stack_map.GetRegisterMask(encoding.stack_map_encoding));
 
     ASSERT_TRUE(CheckStackMask(stack_map, encoding.stack_map_encoding, sp_mask1));
@@ -252,7 +252,7 @@
     ASSERT_TRUE(stack_map.Equals(code_info.GetStackMapForDexPc(1u, encoding)));
     ASSERT_TRUE(stack_map.Equals(code_info.GetStackMapForNativePcOffset(128u, encoding)));
     ASSERT_EQ(1u, stack_map.GetDexPc(encoding.stack_map_encoding));
-    ASSERT_EQ(128u, stack_map.GetNativePcOffset(encoding.stack_map_encoding));
+    ASSERT_EQ(128u, stack_map.GetNativePcOffset(encoding.stack_map_encoding, kRuntimeISA));
     ASSERT_EQ(0xFFu, stack_map.GetRegisterMask(encoding.stack_map_encoding));
 
     ASSERT_TRUE(CheckStackMask(stack_map, encoding.stack_map_encoding, sp_mask2));
@@ -306,7 +306,7 @@
     ASSERT_TRUE(stack_map.Equals(code_info.GetStackMapForDexPc(2u, encoding)));
     ASSERT_TRUE(stack_map.Equals(code_info.GetStackMapForNativePcOffset(192u, encoding)));
     ASSERT_EQ(2u, stack_map.GetDexPc(encoding.stack_map_encoding));
-    ASSERT_EQ(192u, stack_map.GetNativePcOffset(encoding.stack_map_encoding));
+    ASSERT_EQ(192u, stack_map.GetNativePcOffset(encoding.stack_map_encoding, kRuntimeISA));
     ASSERT_EQ(0xABu, stack_map.GetRegisterMask(encoding.stack_map_encoding));
 
     ASSERT_TRUE(CheckStackMask(stack_map, encoding.stack_map_encoding, sp_mask3));
@@ -360,7 +360,7 @@
     ASSERT_TRUE(stack_map.Equals(code_info.GetStackMapForDexPc(3u, encoding)));
     ASSERT_TRUE(stack_map.Equals(code_info.GetStackMapForNativePcOffset(256u, encoding)));
     ASSERT_EQ(3u, stack_map.GetDexPc(encoding.stack_map_encoding));
-    ASSERT_EQ(256u, stack_map.GetNativePcOffset(encoding.stack_map_encoding));
+    ASSERT_EQ(256u, stack_map.GetNativePcOffset(encoding.stack_map_encoding, kRuntimeISA));
     ASSERT_EQ(0xCDu, stack_map.GetRegisterMask(encoding.stack_map_encoding));
 
     ASSERT_TRUE(CheckStackMask(stack_map, encoding.stack_map_encoding, sp_mask4));
@@ -412,7 +412,7 @@
 TEST(StackMapTest, TestNonLiveDexRegisters) {
   ArenaPool pool;
   ArenaAllocator arena(&pool);
-  StackMapStream stream(&arena);
+  StackMapStream stream(&arena, kRuntimeISA);
 
   ArenaBitVector sp_mask(&arena, 0, false);
   uint32_t number_of_dex_registers = 2;
@@ -442,7 +442,7 @@
   ASSERT_TRUE(stack_map.Equals(code_info.GetStackMapForDexPc(0, encoding)));
   ASSERT_TRUE(stack_map.Equals(code_info.GetStackMapForNativePcOffset(64, encoding)));
   ASSERT_EQ(0u, stack_map.GetDexPc(encoding.stack_map_encoding));
-  ASSERT_EQ(64u, stack_map.GetNativePcOffset(encoding.stack_map_encoding));
+  ASSERT_EQ(64u, stack_map.GetNativePcOffset(encoding.stack_map_encoding, kRuntimeISA));
   ASSERT_EQ(0x3u, stack_map.GetRegisterMask(encoding.stack_map_encoding));
 
   ASSERT_TRUE(stack_map.HasDexRegisterMap(encoding.stack_map_encoding));
@@ -491,7 +491,7 @@
 TEST(StackMapTest, DexRegisterMapOffsetOverflow) {
   ArenaPool pool;
   ArenaAllocator arena(&pool);
-  StackMapStream stream(&arena);
+  StackMapStream stream(&arena, kRuntimeISA);
 
   ArenaBitVector sp_mask(&arena, 0, false);
   uint32_t number_of_dex_registers = 1024;
@@ -554,7 +554,7 @@
 TEST(StackMapTest, TestShareDexRegisterMap) {
   ArenaPool pool;
   ArenaAllocator arena(&pool);
-  StackMapStream stream(&arena);
+  StackMapStream stream(&arena, kRuntimeISA);
 
   ArenaBitVector sp_mask(&arena, 0, false);
   uint32_t number_of_dex_registers = 2;
@@ -612,7 +612,7 @@
 TEST(StackMapTest, TestNoDexRegisterMap) {
   ArenaPool pool;
   ArenaAllocator arena(&pool);
-  StackMapStream stream(&arena);
+  StackMapStream stream(&arena, kRuntimeISA);
 
   ArenaBitVector sp_mask(&arena, 0, false);
   uint32_t number_of_dex_registers = 0;
@@ -620,7 +620,7 @@
   stream.EndStackMapEntry();
 
   number_of_dex_registers = 1;
-  stream.BeginStackMapEntry(1, 67, 0x4, &sp_mask, number_of_dex_registers, 0);
+  stream.BeginStackMapEntry(1, 68, 0x4, &sp_mask, number_of_dex_registers, 0);
   stream.EndStackMapEntry();
 
   size_t size = stream.PrepareForFillIn();
@@ -641,7 +641,7 @@
   ASSERT_TRUE(stack_map.Equals(code_info.GetStackMapForDexPc(0, encoding)));
   ASSERT_TRUE(stack_map.Equals(code_info.GetStackMapForNativePcOffset(64, encoding)));
   ASSERT_EQ(0u, stack_map.GetDexPc(encoding.stack_map_encoding));
-  ASSERT_EQ(64u, stack_map.GetNativePcOffset(encoding.stack_map_encoding));
+  ASSERT_EQ(64u, stack_map.GetNativePcOffset(encoding.stack_map_encoding, kRuntimeISA));
   ASSERT_EQ(0x3u, stack_map.GetRegisterMask(encoding.stack_map_encoding));
 
   ASSERT_FALSE(stack_map.HasDexRegisterMap(encoding.stack_map_encoding));
@@ -649,9 +649,9 @@
 
   stack_map = code_info.GetStackMapAt(1, encoding);
   ASSERT_TRUE(stack_map.Equals(code_info.GetStackMapForDexPc(1, encoding)));
-  ASSERT_TRUE(stack_map.Equals(code_info.GetStackMapForNativePcOffset(67, encoding)));
+  ASSERT_TRUE(stack_map.Equals(code_info.GetStackMapForNativePcOffset(68, encoding)));
   ASSERT_EQ(1u, stack_map.GetDexPc(encoding.stack_map_encoding));
-  ASSERT_EQ(67u, stack_map.GetNativePcOffset(encoding.stack_map_encoding));
+  ASSERT_EQ(68u, stack_map.GetNativePcOffset(encoding.stack_map_encoding, kRuntimeISA));
   ASSERT_EQ(0x4u, stack_map.GetRegisterMask(encoding.stack_map_encoding));
 
   ASSERT_FALSE(stack_map.HasDexRegisterMap(encoding.stack_map_encoding));
@@ -661,7 +661,7 @@
 TEST(StackMapTest, InlineTest) {
   ArenaPool pool;
   ArenaAllocator arena(&pool);
-  StackMapStream stream(&arena);
+  StackMapStream stream(&arena, kRuntimeISA);
   ArtMethod art_method;
 
   ArenaBitVector sp_mask1(&arena, 0, true);
@@ -823,4 +823,20 @@
   }
 }
 
+TEST(StackMapTest, CodeOffsetTest) {
+  // Test minimum alignments, encoding, and decoding.
+  CodeOffset offset_thumb2 = CodeOffset::FromOffset(kThumb2InstructionAlignment, kThumb2);
+  CodeOffset offset_arm64 = CodeOffset::FromOffset(kArm64InstructionAlignment, kArm64);
+  CodeOffset offset_x86 = CodeOffset::FromOffset(kX86InstructionAlignment, kX86);
+  CodeOffset offset_x86_64 = CodeOffset::FromOffset(kX86_64InstructionAlignment, kX86_64);
+  CodeOffset offset_mips = CodeOffset::FromOffset(kMipsInstructionAlignment, kMips);
+  CodeOffset offset_mips64 = CodeOffset::FromOffset(kMips64InstructionAlignment, kMips64);
+  EXPECT_EQ(offset_thumb2.Uint32Value(kThumb2), kThumb2InstructionAlignment);
+  EXPECT_EQ(offset_arm64.Uint32Value(kArm64), kArm64InstructionAlignment);
+  EXPECT_EQ(offset_x86.Uint32Value(kX86), kX86InstructionAlignment);
+  EXPECT_EQ(offset_x86_64.Uint32Value(kX86_64), kX86_64InstructionAlignment);
+  EXPECT_EQ(offset_mips.Uint32Value(kMips), kMipsInstructionAlignment);
+  EXPECT_EQ(offset_mips64.Uint32Value(kMips64), kMips64InstructionAlignment);
+}
+
 }  // namespace art
diff --git a/compiler/utils/mips64/assembler_mips64_test.cc b/compiler/utils/mips64/assembler_mips64_test.cc
index f2cbebb..74b8f06 100644
--- a/compiler/utils/mips64/assembler_mips64_test.cc
+++ b/compiler/utils/mips64/assembler_mips64_test.cc
@@ -283,6 +283,38 @@
 // FP Operations //
 ///////////////////
 
+TEST_F(AssemblerMIPS64Test, AddS) {
+  DriverStr(RepeatFFF(&mips64::Mips64Assembler::AddS, "add.s ${reg1}, ${reg2}, ${reg3}"), "add.s");
+}
+
+TEST_F(AssemblerMIPS64Test, AddD) {
+  DriverStr(RepeatFFF(&mips64::Mips64Assembler::AddD, "add.d ${reg1}, ${reg2}, ${reg3}"), "add.d");
+}
+
+TEST_F(AssemblerMIPS64Test, SubS) {
+  DriverStr(RepeatFFF(&mips64::Mips64Assembler::SubS, "sub.s ${reg1}, ${reg2}, ${reg3}"), "sub.s");
+}
+
+TEST_F(AssemblerMIPS64Test, SubD) {
+  DriverStr(RepeatFFF(&mips64::Mips64Assembler::SubD, "sub.d ${reg1}, ${reg2}, ${reg3}"), "sub.d");
+}
+
+TEST_F(AssemblerMIPS64Test, MulS) {
+  DriverStr(RepeatFFF(&mips64::Mips64Assembler::MulS, "mul.s ${reg1}, ${reg2}, ${reg3}"), "mul.s");
+}
+
+TEST_F(AssemblerMIPS64Test, MulD) {
+  DriverStr(RepeatFFF(&mips64::Mips64Assembler::MulD, "mul.d ${reg1}, ${reg2}, ${reg3}"), "mul.d");
+}
+
+TEST_F(AssemblerMIPS64Test, DivS) {
+  DriverStr(RepeatFFF(&mips64::Mips64Assembler::DivS, "div.s ${reg1}, ${reg2}, ${reg3}"), "div.s");
+}
+
+TEST_F(AssemblerMIPS64Test, DivD) {
+  DriverStr(RepeatFFF(&mips64::Mips64Assembler::DivD, "div.d ${reg1}, ${reg2}, ${reg3}"), "div.d");
+}
+
 TEST_F(AssemblerMIPS64Test, SqrtS) {
   DriverStr(RepeatFF(&mips64::Mips64Assembler::SqrtS, "sqrt.s ${reg1}, ${reg2}"), "sqrt.s");
 }
@@ -567,6 +599,26 @@
   DriverStr(RepeatRF(&mips64::Mips64Assembler::Dmtc1, "dmtc1 ${reg1}, ${reg2}"), "Dmtc1");
 }
 
+TEST_F(AssemblerMIPS64Test, Lwc1) {
+  DriverStr(RepeatFRIb(&mips64::Mips64Assembler::Lwc1, -16, "lwc1 ${reg1}, {imm}(${reg2})"),
+            "lwc1");
+}
+
+TEST_F(AssemblerMIPS64Test, Ldc1) {
+  DriverStr(RepeatFRIb(&mips64::Mips64Assembler::Ldc1, -16, "ldc1 ${reg1}, {imm}(${reg2})"),
+            "ldc1");
+}
+
+TEST_F(AssemblerMIPS64Test, Swc1) {
+  DriverStr(RepeatFRIb(&mips64::Mips64Assembler::Swc1, -16, "swc1 ${reg1}, {imm}(${reg2})"),
+            "swc1");
+}
+
+TEST_F(AssemblerMIPS64Test, Sdc1) {
+  DriverStr(RepeatFRIb(&mips64::Mips64Assembler::Sdc1, -16, "sdc1 ${reg1}, {imm}(${reg2})"),
+            "sdc1");
+}
+
 ////////////////
 // CALL / JMP //
 ////////////////
@@ -850,6 +902,16 @@
   DriverStr(RepeatRIb(&mips64::Mips64Assembler::Ldpc, 18, code), "Ldpc");
 }
 
+TEST_F(AssemblerMIPS64Test, Auipc) {
+  DriverStr(RepeatRIb(&mips64::Mips64Assembler::Auipc, 16, "auipc ${reg}, {imm}"), "Auipc");
+}
+
+TEST_F(AssemblerMIPS64Test, Addiupc) {
+  // The comment from the Lwpc() test applies to this Addiupc() test as well.
+  const char* code = ".set imm, {imm}\naddiupc ${reg}, (imm - ((imm & 0x40000) << 1)) << 2";
+  DriverStr(RepeatRIb(&mips64::Mips64Assembler::Addiupc, 19, code), "Addiupc");
+}
+
 TEST_F(AssemblerMIPS64Test, LoadFarthestNearLabelAddress) {
   mips64::Mips64Label label;
   __ LoadLabelAddress(mips64::V0, &label);
@@ -1079,6 +1141,188 @@
   EXPECT_EQ(__ GetLabelLocation(literal->GetLabel()), (5 + kAdduCount) * 4);
 }
 
+TEST_F(AssemblerMIPS64Test, Addu) {
+  DriverStr(RepeatRRR(&mips64::Mips64Assembler::Addu, "addu ${reg1}, ${reg2}, ${reg3}"), "addu");
+}
+
+TEST_F(AssemblerMIPS64Test, Addiu) {
+  DriverStr(RepeatRRIb(&mips64::Mips64Assembler::Addiu, -16, "addiu ${reg1}, ${reg2}, {imm}"),
+            "addiu");
+}
+
+TEST_F(AssemblerMIPS64Test, Daddu) {
+  DriverStr(RepeatRRR(&mips64::Mips64Assembler::Daddu, "daddu ${reg1}, ${reg2}, ${reg3}"), "daddu");
+}
+
+TEST_F(AssemblerMIPS64Test, Daddiu) {
+  DriverStr(RepeatRRIb(&mips64::Mips64Assembler::Daddiu, -16, "daddiu ${reg1}, ${reg2}, {imm}"),
+            "daddiu");
+}
+
+TEST_F(AssemblerMIPS64Test, Subu) {
+  DriverStr(RepeatRRR(&mips64::Mips64Assembler::Subu, "subu ${reg1}, ${reg2}, ${reg3}"), "subu");
+}
+
+TEST_F(AssemblerMIPS64Test, Dsubu) {
+  DriverStr(RepeatRRR(&mips64::Mips64Assembler::Dsubu, "dsubu ${reg1}, ${reg2}, ${reg3}"), "dsubu");
+}
+
+TEST_F(AssemblerMIPS64Test, MulR6) {
+  DriverStr(RepeatRRR(&mips64::Mips64Assembler::MulR6, "mul ${reg1}, ${reg2}, ${reg3}"), "mulR6");
+}
+
+TEST_F(AssemblerMIPS64Test, DivR6) {
+  DriverStr(RepeatRRR(&mips64::Mips64Assembler::DivR6, "div ${reg1}, ${reg2}, ${reg3}"), "divR6");
+}
+
+TEST_F(AssemblerMIPS64Test, ModR6) {
+  DriverStr(RepeatRRR(&mips64::Mips64Assembler::ModR6, "mod ${reg1}, ${reg2}, ${reg3}"), "modR6");
+}
+
+TEST_F(AssemblerMIPS64Test, DivuR6) {
+  DriverStr(RepeatRRR(&mips64::Mips64Assembler::DivuR6, "divu ${reg1}, ${reg2}, ${reg3}"),
+            "divuR6");
+}
+
+TEST_F(AssemblerMIPS64Test, ModuR6) {
+  DriverStr(RepeatRRR(&mips64::Mips64Assembler::ModuR6, "modu ${reg1}, ${reg2}, ${reg3}"),
+            "moduR6");
+}
+
+TEST_F(AssemblerMIPS64Test, Dmul) {
+  DriverStr(RepeatRRR(&mips64::Mips64Assembler::Dmul, "dmul ${reg1}, ${reg2}, ${reg3}"), "dmul");
+}
+
+TEST_F(AssemblerMIPS64Test, Ddiv) {
+  DriverStr(RepeatRRR(&mips64::Mips64Assembler::Ddiv, "ddiv ${reg1}, ${reg2}, ${reg3}"), "ddiv");
+}
+
+TEST_F(AssemblerMIPS64Test, Dmod) {
+  DriverStr(RepeatRRR(&mips64::Mips64Assembler::Dmod, "dmod ${reg1}, ${reg2}, ${reg3}"), "dmod");
+}
+
+TEST_F(AssemblerMIPS64Test, Ddivu) {
+  DriverStr(RepeatRRR(&mips64::Mips64Assembler::Ddivu, "ddivu ${reg1}, ${reg2}, ${reg3}"), "ddivu");
+}
+
+TEST_F(AssemblerMIPS64Test, Dmodu) {
+  DriverStr(RepeatRRR(&mips64::Mips64Assembler::Dmodu, "dmodu ${reg1}, ${reg2}, ${reg3}"), "dmodu");
+}
+
+TEST_F(AssemblerMIPS64Test, And) {
+  DriverStr(RepeatRRR(&mips64::Mips64Assembler::And, "and ${reg1}, ${reg2}, ${reg3}"), "and");
+}
+
+TEST_F(AssemblerMIPS64Test, Andi) {
+  DriverStr(RepeatRRIb(&mips64::Mips64Assembler::Andi, 16, "andi ${reg1}, ${reg2}, {imm}"), "andi");
+}
+
+TEST_F(AssemblerMIPS64Test, Or) {
+  DriverStr(RepeatRRR(&mips64::Mips64Assembler::Or, "or ${reg1}, ${reg2}, ${reg3}"), "or");
+}
+
+TEST_F(AssemblerMIPS64Test, Ori) {
+  DriverStr(RepeatRRIb(&mips64::Mips64Assembler::Ori, 16, "ori ${reg1}, ${reg2}, {imm}"), "ori");
+}
+
+TEST_F(AssemblerMIPS64Test, Xor) {
+  DriverStr(RepeatRRR(&mips64::Mips64Assembler::Xor, "xor ${reg1}, ${reg2}, ${reg3}"), "xor");
+}
+
+TEST_F(AssemblerMIPS64Test, Xori) {
+  DriverStr(RepeatRRIb(&mips64::Mips64Assembler::Xori, 16, "xori ${reg1}, ${reg2}, {imm}"), "xori");
+}
+
+TEST_F(AssemblerMIPS64Test, Nor) {
+  DriverStr(RepeatRRR(&mips64::Mips64Assembler::Nor, "nor ${reg1}, ${reg2}, ${reg3}"), "nor");
+}
+
+TEST_F(AssemblerMIPS64Test, Lb) {
+  DriverStr(RepeatRRIb(&mips64::Mips64Assembler::Lb, -16, "lb ${reg1}, {imm}(${reg2})"), "lb");
+}
+
+TEST_F(AssemblerMIPS64Test, Lh) {
+  DriverStr(RepeatRRIb(&mips64::Mips64Assembler::Lh, -16, "lh ${reg1}, {imm}(${reg2})"), "lh");
+}
+
+TEST_F(AssemblerMIPS64Test, Lw) {
+  DriverStr(RepeatRRIb(&mips64::Mips64Assembler::Lw, -16, "lw ${reg1}, {imm}(${reg2})"), "lw");
+}
+
+TEST_F(AssemblerMIPS64Test, Ld) {
+  DriverStr(RepeatRRIb(&mips64::Mips64Assembler::Ld, -16, "ld ${reg1}, {imm}(${reg2})"), "ld");
+}
+
+TEST_F(AssemblerMIPS64Test, Lbu) {
+  DriverStr(RepeatRRIb(&mips64::Mips64Assembler::Lbu, -16, "lbu ${reg1}, {imm}(${reg2})"), "lbu");
+}
+
+TEST_F(AssemblerMIPS64Test, Lhu) {
+  DriverStr(RepeatRRIb(&mips64::Mips64Assembler::Lhu, -16, "lhu ${reg1}, {imm}(${reg2})"), "lhu");
+}
+
+TEST_F(AssemblerMIPS64Test, Lwu) {
+  DriverStr(RepeatRRIb(&mips64::Mips64Assembler::Lwu, -16, "lwu ${reg1}, {imm}(${reg2})"), "lwu");
+}
+
+TEST_F(AssemblerMIPS64Test, Lui) {
+  DriverStr(RepeatRIb(&mips64::Mips64Assembler::Lui, 16, "lui ${reg}, {imm}"), "lui");
+}
+
+TEST_F(AssemblerMIPS64Test, Dahi) {
+  DriverStr(RepeatRIb(&mips64::Mips64Assembler::Dahi, 16, "dahi ${reg}, ${reg}, {imm}"), "dahi");
+}
+
+TEST_F(AssemblerMIPS64Test, Dati) {
+  DriverStr(RepeatRIb(&mips64::Mips64Assembler::Dati, 16, "dati ${reg}, ${reg}, {imm}"), "dati");
+}
+
+TEST_F(AssemblerMIPS64Test, Sb) {
+  DriverStr(RepeatRRIb(&mips64::Mips64Assembler::Sb, -16, "sb ${reg1}, {imm}(${reg2})"), "sb");
+}
+
+TEST_F(AssemblerMIPS64Test, Sh) {
+  DriverStr(RepeatRRIb(&mips64::Mips64Assembler::Sh, -16, "sh ${reg1}, {imm}(${reg2})"), "sh");
+}
+
+TEST_F(AssemblerMIPS64Test, Sw) {
+  DriverStr(RepeatRRIb(&mips64::Mips64Assembler::Sw, -16, "sw ${reg1}, {imm}(${reg2})"), "sw");
+}
+
+TEST_F(AssemblerMIPS64Test, Sd) {
+  DriverStr(RepeatRRIb(&mips64::Mips64Assembler::Sd, -16, "sd ${reg1}, {imm}(${reg2})"), "sd");
+}
+
+TEST_F(AssemblerMIPS64Test, Slt) {
+  DriverStr(RepeatRRR(&mips64::Mips64Assembler::Slt, "slt ${reg1}, ${reg2}, ${reg3}"), "slt");
+}
+
+TEST_F(AssemblerMIPS64Test, Sltu) {
+  DriverStr(RepeatRRR(&mips64::Mips64Assembler::Sltu, "sltu ${reg1}, ${reg2}, ${reg3}"), "sltu");
+}
+
+TEST_F(AssemblerMIPS64Test, Slti) {
+  DriverStr(RepeatRRIb(&mips64::Mips64Assembler::Slti, -16, "slti ${reg1}, ${reg2}, {imm}"),
+            "slti");
+}
+
+TEST_F(AssemblerMIPS64Test, Sltiu) {
+  DriverStr(RepeatRRIb(&mips64::Mips64Assembler::Sltiu, -16, "sltiu ${reg1}, ${reg2}, {imm}"),
+            "sltiu");
+}
+
+TEST_F(AssemblerMIPS64Test, Move) {
+  DriverStr(RepeatRR(&mips64::Mips64Assembler::Move, "or ${reg1}, ${reg2}, $zero"), "move");
+}
+
+TEST_F(AssemblerMIPS64Test, Clear) {
+  DriverStr(RepeatR(&mips64::Mips64Assembler::Clear, "or ${reg}, $zero, $zero"), "clear");
+}
+
+TEST_F(AssemblerMIPS64Test, Not) {
+  DriverStr(RepeatRR(&mips64::Mips64Assembler::Not, "nor ${reg1}, ${reg2}, $zero"), "not");
+}
+
 TEST_F(AssemblerMIPS64Test, Bitswap) {
   DriverStr(RepeatRR(&mips64::Mips64Assembler::Bitswap, "bitswap ${reg1}, ${reg2}"), "bitswap");
 }
@@ -1230,6 +1474,18 @@
             "dsra32");
 }
 
+TEST_F(AssemblerMIPS64Test, Dsllv) {
+  DriverStr(RepeatRRR(&mips64::Mips64Assembler::Dsllv, "dsllv ${reg1}, ${reg2}, ${reg3}"), "dsllv");
+}
+
+TEST_F(AssemblerMIPS64Test, Dsrlv) {
+  DriverStr(RepeatRRR(&mips64::Mips64Assembler::Dsrlv, "dsrlv ${reg1}, ${reg2}, ${reg3}"), "dsrlv");
+}
+
+TEST_F(AssemblerMIPS64Test, Dsrav) {
+  DriverStr(RepeatRRR(&mips64::Mips64Assembler::Dsrav, "dsrav ${reg1}, ${reg2}, ${reg3}"), "dsrav");
+}
+
 TEST_F(AssemblerMIPS64Test, Sc) {
   DriverStr(RepeatRRIb(&mips64::Mips64Assembler::Sc, -9, "sc ${reg1}, {imm}(${reg2})"), "sc");
 }
diff --git a/dexlayout/dex_ir.h b/dexlayout/dex_ir.h
index a2d1190..e2ee940 100644
--- a/dexlayout/dex_ir.h
+++ b/dexlayout/dex_ir.h
@@ -741,7 +741,7 @@
   uint32_t GetAccessFlags() const { return access_flags_; }
   const TypeId* Superclass() const { return superclass_; }
   const TypeIdVector* Interfaces()
-      { return interfaces_ == nullptr ? nullptr: interfaces_->GetTypeList(); }
+      { return interfaces_ == nullptr ? nullptr : interfaces_->GetTypeList(); }
   uint32_t InterfacesOffset() { return interfaces_ == nullptr ? 0 : interfaces_->GetOffset(); }
   const StringId* SourceFile() const { return source_file_; }
   AnnotationsDirectoryItem* Annotations() const { return annotations_; }
diff --git a/oatdump/oatdump.cc b/oatdump/oatdump.cc
index 148ee88..69901c1 100644
--- a/oatdump/oatdump.cc
+++ b/oatdump/oatdump.cc
@@ -529,6 +529,12 @@
       }
     }
 
+    {
+      os << "OAT FILE STATS:\n";
+      VariableIndentationOutputStream vios(&os);
+      stats_.Dump(vios);
+    }
+
     os << std::flush;
     return success;
   }
@@ -574,6 +580,116 @@
     return nullptr;
   }
 
+  struct Stats {
+    enum ByteKind {
+      kByteKindCode,
+      kByteKindQuickMethodHeader,
+      kByteKindCodeInfoLocationCatalog,
+      kByteKindCodeInfoDexRegisterMap,
+      kByteKindCodeInfoInlineInfo,
+      kByteKindCodeInfoEncoding,
+      kByteKindCodeInfoOther,
+      kByteKindStackMapNativePc,
+      kByteKindStackMapDexPc,
+      kByteKindStackMapDexRegisterMap,
+      kByteKindStackMapInlineInfo,
+      kByteKindStackMapRegisterMask,
+      kByteKindStackMapMask,
+      kByteKindStackMapOther,
+      kByteKindCount,
+      kByteKindStackMapFirst = kByteKindCodeInfoOther,
+      kByteKindStackMapLast = kByteKindStackMapOther,
+    };
+    int64_t bits[kByteKindCount] = {};
+    // Since code has deduplication, seen tracks already seen pointers to avoid double counting
+    // deduplicated code and tables.
+    std::unordered_set<const void*> seen;
+
+    // Returns true if it was newly added.
+    bool AddBitsIfUnique(ByteKind kind, int64_t count, const void* address) {
+      if (seen.insert(address).second == true) {
+        // True means the address was not already in the set.
+        AddBits(kind, count);
+        return true;
+      }
+      return false;
+    }
+
+    void AddBits(ByteKind kind, int64_t count) {
+      bits[kind] += count;
+    }
+
+    void Dump(VariableIndentationOutputStream& os) {
+      const int64_t sum = std::accumulate(bits, bits + kByteKindCount, 0u);
+      os.Stream() << "Dumping cumulative use of " << sum / kBitsPerByte << " accounted bytes\n";
+      if (sum > 0) {
+        const int64_t stack_map_bits = std::accumulate(bits + kByteKindStackMapFirst,
+                                                       bits + kByteKindStackMapLast + 1,
+                                                       0u);
+        Dump(os, "Code                           ", bits[kByteKindCode], sum);
+        Dump(os, "QuickMethodHeader              ", bits[kByteKindQuickMethodHeader], sum);
+        Dump(os, "CodeInfoEncoding               ", bits[kByteKindCodeInfoEncoding], sum);
+        Dump(os, "CodeInfoLocationCatalog        ", bits[kByteKindCodeInfoLocationCatalog], sum);
+        Dump(os, "CodeInfoDexRegisterMap         ", bits[kByteKindCodeInfoDexRegisterMap], sum);
+        Dump(os, "CodeInfoInlineInfo             ", bits[kByteKindCodeInfoInlineInfo], sum);
+        Dump(os, "CodeInfoStackMap               ", stack_map_bits, sum);
+        {
+          ScopedIndentation indent1(&os);
+          Dump(os,
+               "StackMapNativePc             ",
+               bits[kByteKindStackMapNativePc],
+               stack_map_bits,
+               "stack map");
+          Dump(os,
+               "StackMapDexPcEncoding        ",
+               bits[kByteKindStackMapDexPc],
+               stack_map_bits,
+               "stack map");
+          Dump(os,
+               "StackMapDexRegisterMap       ",
+               bits[kByteKindStackMapDexRegisterMap],
+               stack_map_bits,
+               "stack map");
+          Dump(os,
+               "StackMapInlineInfo           ",
+               bits[kByteKindStackMapInlineInfo],
+               stack_map_bits,
+               "stack map");
+          Dump(os,
+               "StackMapRegisterMaskEncoding ",
+               bits[kByteKindStackMapRegisterMask],
+               stack_map_bits,
+               "stack map");
+          Dump(os,
+               "StackMapMask                 ",
+               bits[kByteKindStackMapMask],
+               stack_map_bits,
+               "stack map");
+          Dump(os,
+               "StackMapOther                ",
+               bits[kByteKindStackMapOther],
+               stack_map_bits,
+               "stack map");
+        }
+      }
+      os.Stream() << "\n" << std::flush;
+    }
+
+   private:
+    void Dump(VariableIndentationOutputStream& os,
+              const char* name,
+              int64_t size,
+              int64_t total,
+              const char* sum_of = "total") {
+      const double percent = (static_cast<double>(size) / static_cast<double>(total)) * 100;
+      os.Stream() << StringPrintf("%s = %8" PRId64 " (%2.0f%% of %s)\n",
+                                  name,
+                                  size / kBitsPerByte,
+                                  percent,
+                                  sum_of);
+    }
+  };
+
  private:
   void AddAllOffsets() {
     // We don't know the length of the code for each method, but we need to know where to stop
@@ -1046,7 +1162,9 @@
       vios->Stream() << "OatQuickMethodHeader ";
       uint32_t method_header_offset = oat_method.GetOatQuickMethodHeaderOffset();
       const OatQuickMethodHeader* method_header = oat_method.GetOatQuickMethodHeader();
-
+      stats_.AddBitsIfUnique(Stats::kByteKindQuickMethodHeader,
+                             sizeof(*method_header) * kBitsPerByte,
+                             method_header);
       if (options_.absolute_addresses_) {
         vios->Stream() << StringPrintf("%p ", method_header);
       }
@@ -1118,6 +1236,7 @@
         const void* code = oat_method.GetQuickCode();
         uint32_t aligned_code_begin = AlignCodeOffset(code_offset);
         uint64_t aligned_code_end = aligned_code_begin + code_size;
+        stats_.AddBitsIfUnique(Stats::kByteKindCode, code_size * kBitsPerByte, code);
 
         if (options_.absolute_addresses_) {
           vios->Stream() << StringPrintf("%p ", code);
@@ -1223,7 +1342,8 @@
     code_info.Dump(vios,
                    oat_method.GetCodeOffset(),
                    code_item.registers_size_,
-                   options_.dump_code_info_stack_maps_);
+                   options_.dump_code_info_stack_maps_,
+                   instruction_set_);
   }
 
   void DumpVregLocations(std::ostream& os, const OatFile::OatMethod& oat_method,
@@ -1329,21 +1449,22 @@
   // For identical native PCs, the order from the CodeInfo is preserved.
   class StackMapsHelper {
    public:
-    explicit StackMapsHelper(const uint8_t* raw_code_info)
+    explicit StackMapsHelper(const uint8_t* raw_code_info, InstructionSet instruction_set)
         : code_info_(raw_code_info),
           encoding_(code_info_.ExtractEncoding()),
           number_of_stack_maps_(code_info_.GetNumberOfStackMaps(encoding_)),
           indexes_(),
-          offset_(static_cast<size_t>(-1)),
-          stack_map_index_(0u) {
+          offset_(static_cast<uint32_t>(-1)),
+          stack_map_index_(0u),
+          instruction_set_(instruction_set) {
       if (number_of_stack_maps_ != 0u) {
         // Check if native PCs are ordered.
         bool ordered = true;
         StackMap last = code_info_.GetStackMapAt(0u, encoding_);
         for (size_t i = 1; i != number_of_stack_maps_; ++i) {
           StackMap current = code_info_.GetStackMapAt(i, encoding_);
-          if (last.GetNativePcOffset(encoding_.stack_map_encoding) >
-              current.GetNativePcOffset(encoding_.stack_map_encoding)) {
+          if (last.GetNativePcOffset(encoding_.stack_map_encoding, instruction_set) >
+              current.GetNativePcOffset(encoding_.stack_map_encoding, instruction_set)) {
             ordered = false;
             break;
           }
@@ -1359,14 +1480,17 @@
                     indexes_.end(),
                     [this](size_t lhs, size_t rhs) {
                       StackMap left = code_info_.GetStackMapAt(lhs, encoding_);
-                      uint32_t left_pc = left.GetNativePcOffset(encoding_.stack_map_encoding);
+                      uint32_t left_pc = left.GetNativePcOffset(encoding_.stack_map_encoding,
+                                                                instruction_set_);
                       StackMap right = code_info_.GetStackMapAt(rhs, encoding_);
-                      uint32_t right_pc = right.GetNativePcOffset(encoding_.stack_map_encoding);
+                      uint32_t right_pc = right.GetNativePcOffset(encoding_.stack_map_encoding,
+                                                                  instruction_set_);
                       // If the PCs are the same, compare indexes to preserve the original order.
                       return (left_pc < right_pc) || (left_pc == right_pc && lhs < rhs);
                     });
         }
-        offset_ = GetStackMapAt(0).GetNativePcOffset(encoding_.stack_map_encoding);
+        offset_ = GetStackMapAt(0).GetNativePcOffset(encoding_.stack_map_encoding,
+                                                     instruction_set_);
       }
     }
 
@@ -1378,7 +1502,7 @@
       return encoding_;
     }
 
-    size_t GetOffset() const {
+    uint32_t GetOffset() const {
       return offset_;
     }
 
@@ -1389,8 +1513,9 @@
     void Next() {
       ++stack_map_index_;
       offset_ = (stack_map_index_ == number_of_stack_maps_)
-          ? static_cast<size_t>(-1)
-          : GetStackMapAt(stack_map_index_).GetNativePcOffset(encoding_.stack_map_encoding);
+          ? static_cast<uint32_t>(-1)
+          : GetStackMapAt(stack_map_index_).GetNativePcOffset(encoding_.stack_map_encoding,
+                                                              instruction_set_);
     }
 
    private:
@@ -1406,8 +1531,9 @@
     const CodeInfoEncoding encoding_;
     const size_t number_of_stack_maps_;
     dchecked_vector<size_t> indexes_;  // Used if stack map native PCs are not ordered.
-    size_t offset_;
+    uint32_t offset_;
     size_t stack_map_index_;
+    const InstructionSet instruction_set_;
   };
 
   void DumpCode(VariableIndentationOutputStream* vios,
@@ -1423,7 +1549,61 @@
       return;
     } else if (!bad_input && IsMethodGeneratedByOptimizingCompiler(oat_method, code_item)) {
       // The optimizing compiler outputs its CodeInfo data in the vmap table.
-      StackMapsHelper helper(oat_method.GetVmapTable());
+      StackMapsHelper helper(oat_method.GetVmapTable(), instruction_set_);
+      {
+        CodeInfoEncoding encoding(helper.GetEncoding());
+        StackMapEncoding stack_map_encoding(encoding.stack_map_encoding);
+        // helper.GetCodeInfo().GetStackMapAt(0, encoding).;
+        const size_t num_stack_maps = encoding.number_of_stack_maps;
+        std::vector<uint8_t> size_vector;
+        encoding.Compress(&size_vector);
+        if (stats_.AddBitsIfUnique(Stats::kByteKindCodeInfoEncoding,
+                                   size_vector.size() * kBitsPerByte,
+                                   oat_method.GetVmapTable())) {
+          stats_.AddBits(
+              Stats::kByteKindStackMapNativePc,
+              stack_map_encoding.GetNativePcEncoding().BitSize() * num_stack_maps);
+          stats_.AddBits(
+              Stats::kByteKindStackMapDexPc,
+              stack_map_encoding.GetDexPcEncoding().BitSize() * num_stack_maps);
+          stats_.AddBits(
+              Stats::kByteKindStackMapDexRegisterMap,
+              stack_map_encoding.GetDexRegisterMapEncoding().BitSize() * num_stack_maps);
+          stats_.AddBits(
+              Stats::kByteKindStackMapInlineInfo,
+              stack_map_encoding.GetInlineInfoEncoding().BitSize() * num_stack_maps);
+          stats_.AddBits(
+              Stats::kByteKindStackMapRegisterMask,
+              stack_map_encoding.GetRegisterMaskEncoding().BitSize() * num_stack_maps);
+          const size_t stack_mask_bits = encoding.stack_map_size_in_bytes * kBitsPerByte -
+              stack_map_encoding.GetStackMaskBitOffset();
+          stats_.AddBits(
+              Stats::kByteKindStackMapMask,
+              stack_mask_bits * num_stack_maps);
+          const size_t stack_map_bits =
+              stack_map_encoding.GetStackMaskBitOffset() + stack_mask_bits;
+          stats_.AddBits(
+              Stats::kByteKindStackMapOther,
+              (encoding.stack_map_size_in_bytes * kBitsPerByte - stack_map_bits) * num_stack_maps);
+          const size_t stack_map_bytes = helper.GetCodeInfo().GetStackMapsSize(encoding);
+          const size_t location_catalog_bytes =
+              helper.GetCodeInfo().GetDexRegisterLocationCatalogSize(encoding);
+          stats_.AddBits(Stats::kByteKindCodeInfoLocationCatalog,
+                         kBitsPerByte * location_catalog_bytes);
+          const size_t dex_register_bytes =
+              helper.GetCodeInfo().GetDexRegisterMapsSize(encoding, code_item->registers_size_);
+          stats_.AddBits(
+              Stats::kByteKindCodeInfoDexRegisterMap,
+              kBitsPerByte * dex_register_bytes);
+          const size_t inline_info_bytes =
+              encoding.non_header_size -
+              stack_map_bytes -
+              location_catalog_bytes -
+              dex_register_bytes;
+          stats_.AddBits(Stats::kByteKindCodeInfoInlineInfo,
+                         inline_info_bytes * kBitsPerByte);
+        }
+      }
       const uint8_t* quick_native_pc = reinterpret_cast<const uint8_t*>(quick_code);
       size_t offset = 0;
       while (offset < code_size) {
@@ -1436,7 +1616,8 @@
                          helper.GetCodeInfo(),
                          helper.GetEncoding(),
                          oat_method.GetCodeOffset(),
-                         code_item->registers_size_);
+                         code_item->registers_size_,
+                         instruction_set_);
           do {
             helper.Next();
             // There may be multiple stack maps at a given PC. We display only the first one.
@@ -1460,6 +1641,7 @@
   const InstructionSet instruction_set_;
   std::set<uintptr_t> offsets_;
   Disassembler* disassembler_;
+  Stats stats_;
 };
 
 class ImageDumper {
@@ -1776,7 +1958,7 @@
         os << StringPrintf("%d (0x%x)\n", field->GetShort(obj), field->GetShort(obj));
         break;
       case Primitive::kPrimBoolean:
-        os << StringPrintf("%s (0x%x)\n", field->GetBoolean(obj)? "true" : "false",
+        os << StringPrintf("%s (0x%x)\n", field->GetBoolean(obj) ? "true" : "false",
             field->GetBoolean(obj));
         break;
       case Primitive::kPrimByte:
@@ -2132,7 +2314,6 @@
 
     size_t managed_code_bytes;
     size_t managed_code_bytes_ignoring_deduplication;
-    size_t managed_to_native_code_bytes;
     size_t native_to_managed_code_bytes;
     size_t class_initializer_code_bytes;
     size_t large_initializer_code_bytes;
@@ -2161,7 +2342,6 @@
           alignment_bytes(0),
           managed_code_bytes(0),
           managed_code_bytes_ignoring_deduplication(0),
-          managed_to_native_code_bytes(0),
           native_to_managed_code_bytes(0),
           class_initializer_code_bytes(0),
           large_initializer_code_bytes(0),
@@ -2359,7 +2539,6 @@
 
       os << StringPrintf("oat_file_bytes               = %8zd\n"
                          "managed_code_bytes           = %8zd (%2.0f%% of oat file bytes)\n"
-                         "managed_to_native_code_bytes = %8zd (%2.0f%% of oat file bytes)\n"
                          "native_to_managed_code_bytes = %8zd (%2.0f%% of oat file bytes)\n\n"
                          "class_initializer_code_bytes = %8zd (%2.0f%% of oat file bytes)\n"
                          "large_initializer_code_bytes = %8zd (%2.0f%% of oat file bytes)\n"
@@ -2367,8 +2546,6 @@
                          oat_file_bytes,
                          managed_code_bytes,
                          PercentOfOatBytes(managed_code_bytes),
-                         managed_to_native_code_bytes,
-                         PercentOfOatBytes(managed_to_native_code_bytes),
                          native_to_managed_code_bytes,
                          PercentOfOatBytes(native_to_managed_code_bytes),
                          class_initializer_code_bytes,
diff --git a/oatdump/oatdump_test.cc b/oatdump/oatdump_test.cc
index e77d03b..ba57d18 100644
--- a/oatdump/oatdump_test.cc
+++ b/oatdump/oatdump_test.cc
@@ -102,6 +102,7 @@
         // Code and dex code do not show up if list only.
         expected_prefixes.push_back("DEX CODE:");
         expected_prefixes.push_back("CODE:");
+        expected_prefixes.push_back("CodeInfoEncoding");
       }
       if (mode == kModeArt) {
         exec_argv.push_back("--image=" + core_art_location_);
diff --git a/runtime/arch/arm/quick_method_frame_info_arm.h b/runtime/arch/arm/quick_method_frame_info_arm.h
index 4b23c77..35f1948 100644
--- a/runtime/arch/arm/quick_method_frame_info_arm.h
+++ b/runtime/arch/arm/quick_method_frame_info_arm.h
@@ -62,7 +62,7 @@
 
 constexpr uint32_t ArmCalleeSaveFpSpills(Runtime::CalleeSaveType type) {
   return kArmCalleeSaveFpAlwaysSpills | kArmCalleeSaveFpRefSpills |
-      (type == Runtime::kSaveRefsAndArgs ? kArmCalleeSaveFpArgSpills: 0) |
+      (type == Runtime::kSaveRefsAndArgs ? kArmCalleeSaveFpArgSpills : 0) |
       (type == Runtime::kSaveAllCalleeSaves ? kArmCalleeSaveFpAllSpills : 0) |
       (type == Runtime::kSaveEverything ? kArmCalleeSaveFpEverythingSpills : 0);
 }
diff --git a/runtime/arch/arm64/instruction_set_features_arm64.cc b/runtime/arch/arm64/instruction_set_features_arm64.cc
index c598743..01bd177 100644
--- a/runtime/arch/arm64/instruction_set_features_arm64.cc
+++ b/runtime/arch/arm64/instruction_set_features_arm64.cc
@@ -33,7 +33,16 @@
     const std::string& variant, std::string* error_msg) {
   // Look for variants that need a fix for a53 erratum 835769.
   static const char* arm64_variants_with_a53_835769_bug[] = {
-      "default", "generic", "cortex-a53"  // Pessimistically assume all generic ARM64s are A53s.
+      // Pessimistically assume all generic CPUs are cortex-a53.
+      "default",
+      "generic",
+      "cortex-a53",
+      "cortex-a53.a57",
+      "cortex-a53.a72",
+      // Pessimistically assume all "big" cortex CPUs are paired with a cortex-a53.
+      "cortex-a57",
+      "cortex-a72",
+      "cortex-a73",
   };
   bool needs_a53_835769_fix = FindVariantInArray(arm64_variants_with_a53_835769_bug,
                                                  arraysize(arm64_variants_with_a53_835769_bug),
@@ -42,7 +51,10 @@
   if (!needs_a53_835769_fix) {
     // Check to see if this is an expected variant.
     static const char* arm64_known_variants[] = {
-        "denver64", "kryo", "exynos-m1"
+        "cortex-a35",
+        "exynos-m1",
+        "denver64",
+        "kryo"
     };
     if (!FindVariantInArray(arm64_known_variants, arraysize(arm64_known_variants), variant)) {
       std::ostringstream os;
diff --git a/runtime/arch/arm64/instruction_set_features_arm64_test.cc b/runtime/arch/arm64/instruction_set_features_arm64_test.cc
index cefa499..91cb58f 100644
--- a/runtime/arch/arm64/instruction_set_features_arm64_test.cc
+++ b/runtime/arch/arm64/instruction_set_features_arm64_test.cc
@@ -30,6 +30,40 @@
   EXPECT_TRUE(arm64_features->Equals(arm64_features.get()));
   EXPECT_STREQ("a53", arm64_features->GetFeatureString().c_str());
   EXPECT_EQ(arm64_features->AsBitmap(), 1U);
+
+  std::unique_ptr<const InstructionSetFeatures> cortex_a57_features(
+      InstructionSetFeatures::FromVariant(kArm64, "cortex-a57", &error_msg));
+  ASSERT_TRUE(cortex_a57_features.get() != nullptr) << error_msg;
+  EXPECT_EQ(cortex_a57_features->GetInstructionSet(), kArm64);
+  EXPECT_TRUE(cortex_a57_features->Equals(cortex_a57_features.get()));
+  EXPECT_STREQ("a53", cortex_a57_features->GetFeatureString().c_str());
+  EXPECT_EQ(cortex_a57_features->AsBitmap(), 1U);
+
+  std::unique_ptr<const InstructionSetFeatures> cortex_a73_features(
+      InstructionSetFeatures::FromVariant(kArm64, "cortex-a73", &error_msg));
+  ASSERT_TRUE(cortex_a73_features.get() != nullptr) << error_msg;
+  EXPECT_EQ(cortex_a73_features->GetInstructionSet(), kArm64);
+  EXPECT_TRUE(cortex_a73_features->Equals(cortex_a73_features.get()));
+  EXPECT_STREQ("a53", cortex_a73_features->GetFeatureString().c_str());
+  EXPECT_EQ(cortex_a73_features->AsBitmap(), 1U);
+
+  std::unique_ptr<const InstructionSetFeatures> cortex_a35_features(
+      InstructionSetFeatures::FromVariant(kArm64, "cortex-a35", &error_msg));
+  ASSERT_TRUE(cortex_a35_features.get() != nullptr) << error_msg;
+  EXPECT_EQ(cortex_a35_features->GetInstructionSet(), kArm64);
+  EXPECT_TRUE(cortex_a35_features->Equals(cortex_a35_features.get()));
+  EXPECT_STREQ("-a53", cortex_a35_features->GetFeatureString().c_str());
+  EXPECT_EQ(cortex_a35_features->AsBitmap(), 0U);
+
+  std::unique_ptr<const InstructionSetFeatures> kryo_features(
+      InstructionSetFeatures::FromVariant(kArm64, "kryo", &error_msg));
+  ASSERT_TRUE(kryo_features.get() != nullptr) << error_msg;
+  EXPECT_EQ(kryo_features->GetInstructionSet(), kArm64);
+  EXPECT_TRUE(kryo_features->Equals(kryo_features.get()));
+  EXPECT_TRUE(kryo_features->Equals(cortex_a35_features.get()));
+  EXPECT_FALSE(kryo_features->Equals(cortex_a57_features.get()));
+  EXPECT_STREQ("-a53", kryo_features->GetFeatureString().c_str());
+  EXPECT_EQ(kryo_features->AsBitmap(), 0U);
 }
 
 }  // namespace art
diff --git a/runtime/arch/arm64/quick_method_frame_info_arm64.h b/runtime/arch/arm64/quick_method_frame_info_arm64.h
index 36f283b..32d9d08 100644
--- a/runtime/arch/arm64/quick_method_frame_info_arm64.h
+++ b/runtime/arch/arm64/quick_method_frame_info_arm64.h
@@ -85,7 +85,7 @@
 
 constexpr uint32_t Arm64CalleeSaveFpSpills(Runtime::CalleeSaveType type) {
   return kArm64CalleeSaveFpAlwaysSpills | kArm64CalleeSaveFpRefSpills |
-      (type == Runtime::kSaveRefsAndArgs ? kArm64CalleeSaveFpArgSpills: 0) |
+      (type == Runtime::kSaveRefsAndArgs ? kArm64CalleeSaveFpArgSpills : 0) |
       (type == Runtime::kSaveAllCalleeSaves ? kArm64CalleeSaveFpAllSpills : 0) |
       (type == Runtime::kSaveEverything ? kArm64CalleeSaveFpEverythingSpills : 0);
 }
diff --git a/runtime/arch/code_offset.h b/runtime/arch/code_offset.h
new file mode 100644
index 0000000..ab04b1e
--- /dev/null
+++ b/runtime/arch/code_offset.h
@@ -0,0 +1,92 @@
+/*
+ * Copyright (C) 2017 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_ARCH_CODE_OFFSET_H_
+#define ART_RUNTIME_ARCH_CODE_OFFSET_H_
+
+#include <iosfwd>
+
+#include "base/bit_utils.h"
+#include "base/logging.h"
+#include "instruction_set.h"
+
+namespace art {
+
+// CodeOffset is a holder for compressed code offsets. Since some architectures have alignment
+// requirements it is possible to compress code offsets to reduce stack map sizes.
+class CodeOffset {
+ public:
+  ALWAYS_INLINE static CodeOffset FromOffset(uint32_t offset, InstructionSet isa = kRuntimeISA) {
+    return CodeOffset(offset / GetInstructionSetInstructionAlignment(isa));
+  }
+
+  ALWAYS_INLINE static CodeOffset FromCompressedOffset(uint32_t offset) {
+    return CodeOffset(offset);
+  }
+
+  ALWAYS_INLINE uint32_t Uint32Value(InstructionSet isa = kRuntimeISA) const {
+    uint32_t decoded = value_ * GetInstructionSetInstructionAlignment(isa);
+    DCHECK_GE(decoded, value_) << "Integer overflow";
+    return decoded;
+  }
+
+  // Return compressed internal value.
+  ALWAYS_INLINE uint32_t CompressedValue() const {
+    return value_;
+  }
+
+  ALWAYS_INLINE CodeOffset() = default;
+  ALWAYS_INLINE CodeOffset(const CodeOffset&) = default;
+  ALWAYS_INLINE CodeOffset& operator=(const CodeOffset&) = default;
+  ALWAYS_INLINE CodeOffset& operator=(CodeOffset&&) = default;
+
+ private:
+  ALWAYS_INLINE explicit CodeOffset(uint32_t value) : value_(value) {}
+
+  uint32_t value_ = 0u;
+};
+
+inline bool operator==(const CodeOffset& a, const CodeOffset& b) {
+  return a.CompressedValue() == b.CompressedValue();
+}
+
+inline bool operator!=(const CodeOffset& a, const CodeOffset& b) {
+  return !(a == b);
+}
+
+inline bool operator<(const CodeOffset& a, const CodeOffset& b) {
+  return a.CompressedValue() < b.CompressedValue();
+}
+
+inline bool operator<=(const CodeOffset& a, const CodeOffset& b) {
+  return a.CompressedValue() <= b.CompressedValue();
+}
+
+inline bool operator>(const CodeOffset& a, const CodeOffset& b) {
+  return a.CompressedValue() > b.CompressedValue();
+}
+
+inline bool operator>=(const CodeOffset& a, const CodeOffset& b) {
+  return a.CompressedValue() >= b.CompressedValue();
+}
+
+inline std::ostream& operator<<(std::ostream& os, const CodeOffset& offset) {
+  return os << offset.Uint32Value();
+}
+
+}  // namespace art
+
+#endif  // ART_RUNTIME_ARCH_CODE_OFFSET_H_
diff --git a/runtime/arch/instruction_set.h b/runtime/arch/instruction_set.h
index 4a8bea4..99aea62 100644
--- a/runtime/arch/instruction_set.h
+++ b/runtime/arch/instruction_set.h
@@ -75,6 +75,14 @@
 // X86 instruction alignment. This is the recommended alignment for maximum performance.
 static constexpr size_t kX86Alignment = 16;
 
+// Different than code alignment since code alignment is only first instruction of method.
+static constexpr size_t kThumb2InstructionAlignment = 2;
+static constexpr size_t kArm64InstructionAlignment = 4;
+static constexpr size_t kX86InstructionAlignment = 1;
+static constexpr size_t kX86_64InstructionAlignment = 1;
+static constexpr size_t kMipsInstructionAlignment = 2;
+static constexpr size_t kMips64InstructionAlignment = 2;
+
 const char* GetInstructionSetString(InstructionSet isa);
 
 // Note: Returns kNone when the string cannot be parsed to a known value.
@@ -106,6 +114,17 @@
   }
 }
 
+ALWAYS_INLINE static inline constexpr size_t GetInstructionSetInstructionAlignment(
+    InstructionSet isa) {
+  return (isa == kThumb2 || isa == kArm) ? kThumb2InstructionAlignment :
+         (isa == kArm64) ? kArm64InstructionAlignment :
+         (isa == kX86) ? kX86InstructionAlignment :
+         (isa == kX86_64) ? kX86_64InstructionAlignment :
+         (isa == kMips) ? kMipsInstructionAlignment :
+         (isa == kMips64) ? kMips64InstructionAlignment :
+         0;  // Invalid case, but constexpr doesn't support asserts.
+}
+
 static inline bool IsValidInstructionSet(InstructionSet isa) {
   switch (isa) {
     case kArm:
diff --git a/runtime/arch/instruction_set_test.cc b/runtime/arch/instruction_set_test.cc
index 5aae93a..b251b57 100644
--- a/runtime/arch/instruction_set_test.cc
+++ b/runtime/arch/instruction_set_test.cc
@@ -44,6 +44,15 @@
   EXPECT_STREQ("none", GetInstructionSetString(kNone));
 }
 
+TEST(InstructionSetTest, GetInstructionSetInstructionAlignment) {
+  EXPECT_EQ(GetInstructionSetInstructionAlignment(kThumb2), kThumb2InstructionAlignment);
+  EXPECT_EQ(GetInstructionSetInstructionAlignment(kArm64), kArm64InstructionAlignment);
+  EXPECT_EQ(GetInstructionSetInstructionAlignment(kX86), kX86InstructionAlignment);
+  EXPECT_EQ(GetInstructionSetInstructionAlignment(kX86_64), kX86_64InstructionAlignment);
+  EXPECT_EQ(GetInstructionSetInstructionAlignment(kMips), kMipsInstructionAlignment);
+  EXPECT_EQ(GetInstructionSetInstructionAlignment(kMips64), kMips64InstructionAlignment);
+}
+
 TEST(InstructionSetTest, TestRoundTrip) {
   EXPECT_EQ(kRuntimeISA, GetInstructionSetFromString(GetInstructionSetString(kRuntimeISA)));
 }
diff --git a/runtime/arch/mips64/quick_method_frame_info_mips64.h b/runtime/arch/mips64/quick_method_frame_info_mips64.h
index 397776e..d774473 100644
--- a/runtime/arch/mips64/quick_method_frame_info_mips64.h
+++ b/runtime/arch/mips64/quick_method_frame_info_mips64.h
@@ -78,7 +78,7 @@
 
 constexpr uint32_t Mips64CalleeSaveFpSpills(Runtime::CalleeSaveType type) {
   return kMips64CalleeSaveFpRefSpills |
-      (type == Runtime::kSaveRefsAndArgs ? kMips64CalleeSaveFpArgSpills: 0) |
+      (type == Runtime::kSaveRefsAndArgs ? kMips64CalleeSaveFpArgSpills : 0) |
       (type == Runtime::kSaveAllCalleeSaves ? kMips64CalleeSaveFpAllSpills : 0) |
       (type == Runtime::kSaveEverything ? kMips64CalleeSaveFpEverythingSpills : 0);
 }
diff --git a/runtime/class_linker.cc b/runtime/class_linker.cc
index ddb3245..b8ed530 100644
--- a/runtime/class_linker.cc
+++ b/runtime/class_linker.cc
@@ -1398,7 +1398,11 @@
         class_loader_(class_loader) {}
 
   bool operator()(ObjPtr<mirror::Class> klass) const REQUIRES_SHARED(Locks::mutator_lock_) {
-    klass->SetClassLoader(class_loader_);
+    // Do not update class loader for boot image classes where the app image
+    // class loader is only the initiating loader but not the defining loader.
+    if (klass->GetClassLoader() != nullptr) {
+      klass->SetClassLoader(class_loader_);
+    }
     return true;
   }
 
@@ -2458,10 +2462,8 @@
     return EnsureResolved(self, descriptor, klass);
   }
   // Class is not yet loaded.
-  if (descriptor[0] == '[') {
-    return CreateArrayClass(self, descriptor, hash, class_loader);
-  } else if (class_loader.Get() == nullptr) {
-    // The boot class loader, search the boot class path.
+  if (descriptor[0] != '[' && class_loader.Get() == nullptr) {
+    // Non-array class and the boot class loader, search the boot class path.
     ClassPathEntry pair = FindInClassPath(descriptor, hash, boot_class_path_);
     if (pair.second != nullptr) {
       return DefineClass(self,
@@ -2474,14 +2476,21 @@
       // The boot class loader is searched ahead of the application class loader, failures are
       // expected and will be wrapped in a ClassNotFoundException. Use the pre-allocated error to
       // trigger the chaining with a proper stack trace.
-      ObjPtr<mirror::Throwable> pre_allocated = Runtime::Current()->GetPreAllocatedNoClassDefFoundError();
+      ObjPtr<mirror::Throwable> pre_allocated =
+          Runtime::Current()->GetPreAllocatedNoClassDefFoundError();
       self->SetException(pre_allocated);
       return nullptr;
     }
+  }
+  ObjPtr<mirror::Class> result_ptr;
+  bool descriptor_equals;
+  if (descriptor[0] == '[') {
+    result_ptr = CreateArrayClass(self, descriptor, hash, class_loader);
+    DCHECK_EQ(result_ptr == nullptr, self->IsExceptionPending());
+    DCHECK(result_ptr == nullptr || result_ptr->DescriptorEquals(descriptor));
+    descriptor_equals = true;
   } else {
     ScopedObjectAccessUnchecked soa(self);
-    ObjPtr<mirror::Class> result_ptr;
-    bool descriptor_equals;
     bool known_hierarchy =
         FindClassInBaseDexClassLoader(soa, self, descriptor, hash, class_loader, &result_ptr);
     if (result_ptr != nullptr) {
@@ -2525,16 +2534,7 @@
                                                  WellKnownClasses::java_lang_ClassLoader_loadClass,
                                                  class_name_object.get()));
       }
-      if (self->IsExceptionPending()) {
-        // If the ClassLoader threw, pass that exception up.
-        // However, to comply with the RI behavior, first check if another thread succeeded.
-        result_ptr = LookupClass(self, descriptor, hash, class_loader.Get());
-        if (result_ptr != nullptr && !result_ptr->IsErroneous()) {
-          self->ClearException();
-          return EnsureResolved(self, descriptor, result_ptr);
-        }
-        return nullptr;
-      } else if (result.get() == nullptr) {
+      if (result.get() == nullptr && !self->IsExceptionPending()) {
         // broken loader - throw NPE to be compatible with Dalvik
         ThrowNullPointerException(StringPrintf("ClassLoader.loadClass returned null for %s",
                                                class_name_string.c_str()).c_str());
@@ -2542,50 +2542,60 @@
       }
       result_ptr = soa.Decode<mirror::Class>(result.get());
       // Check the name of the returned class.
-      descriptor_equals = result_ptr->DescriptorEquals(descriptor);
+      descriptor_equals = (result_ptr != nullptr) && result_ptr->DescriptorEquals(descriptor);
     }
-
-    // Try to insert the class to the class table, checking for mismatch.
-    ObjPtr<mirror::Class> old;
-    {
-      WriterMutexLock mu(self, *Locks::classlinker_classes_lock_);
-      ClassTable* const class_table = InsertClassTableForClassLoader(class_loader.Get());
-      old = class_table->Lookup(descriptor, hash);
-      if (old == nullptr) {
-        old = result_ptr;  // For the comparison below, after releasing the lock.
-        if (descriptor_equals) {
-          class_table->InsertWithHash(result_ptr.Ptr(), hash);
-          Runtime::Current()->GetHeap()->WriteBarrierEveryFieldOf(class_loader.Get());
-        }  // else throw below, after releasing the lock.
-      }
-    }
-    if (UNLIKELY(old != result_ptr)) {
-      // Return `old` (even if `!descriptor_equals`) to mimic the RI behavior for parallel
-      // capable class loaders.  (All class loaders are considered parallel capable on Android.)
-      mirror::Class* loader_class = class_loader->GetClass();
-      const char* loader_class_name =
-          loader_class->GetDexFile().StringByTypeIdx(loader_class->GetDexTypeIndex());
-      LOG(WARNING) << "Initiating class loader of type " << DescriptorToDot(loader_class_name)
-          << " is not well-behaved; it returned a different Class for racing loadClass(\""
-          << DescriptorToDot(descriptor) << "\").";
-      return EnsureResolved(self, descriptor, old);
-    }
-    if (UNLIKELY(!descriptor_equals)) {
-      std::string result_storage;
-      const char* result_name = result_ptr->GetDescriptor(&result_storage);
-      std::string loader_storage;
-      const char* loader_class_name = class_loader->GetClass()->GetDescriptor(&loader_storage);
-      ThrowNoClassDefFoundError(
-          "Initiating class loader of type %s returned class %s instead of %s.",
-          DescriptorToDot(loader_class_name).c_str(),
-          DescriptorToDot(result_name).c_str(),
-          DescriptorToDot(descriptor).c_str());
-      return nullptr;
-    }
-    // success, return mirror::Class*
-    return result_ptr.Ptr();
   }
-  UNREACHABLE();
+
+  if (self->IsExceptionPending()) {
+    // If the ClassLoader threw or array class allocation failed, pass that exception up.
+    // However, to comply with the RI behavior, first check if another thread succeeded.
+    result_ptr = LookupClass(self, descriptor, hash, class_loader.Get());
+    if (result_ptr != nullptr && !result_ptr->IsErroneous()) {
+      self->ClearException();
+      return EnsureResolved(self, descriptor, result_ptr);
+    }
+    return nullptr;
+  }
+
+  // Try to insert the class to the class table, checking for mismatch.
+  ObjPtr<mirror::Class> old;
+  {
+    WriterMutexLock mu(self, *Locks::classlinker_classes_lock_);
+    ClassTable* const class_table = InsertClassTableForClassLoader(class_loader.Get());
+    old = class_table->Lookup(descriptor, hash);
+    if (old == nullptr) {
+      old = result_ptr;  // For the comparison below, after releasing the lock.
+      if (descriptor_equals) {
+        class_table->InsertWithHash(result_ptr.Ptr(), hash);
+        Runtime::Current()->GetHeap()->WriteBarrierEveryFieldOf(class_loader.Get());
+      }  // else throw below, after releasing the lock.
+    }
+  }
+  if (UNLIKELY(old != result_ptr)) {
+    // Return `old` (even if `!descriptor_equals`) to mimic the RI behavior for parallel
+    // capable class loaders.  (All class loaders are considered parallel capable on Android.)
+    mirror::Class* loader_class = class_loader->GetClass();
+    const char* loader_class_name =
+        loader_class->GetDexFile().StringByTypeIdx(loader_class->GetDexTypeIndex());
+    LOG(WARNING) << "Initiating class loader of type " << DescriptorToDot(loader_class_name)
+        << " is not well-behaved; it returned a different Class for racing loadClass(\""
+        << DescriptorToDot(descriptor) << "\").";
+    return EnsureResolved(self, descriptor, old);
+  }
+  if (UNLIKELY(!descriptor_equals)) {
+    std::string result_storage;
+    const char* result_name = result_ptr->GetDescriptor(&result_storage);
+    std::string loader_storage;
+    const char* loader_class_name = class_loader->GetClass()->GetDescriptor(&loader_storage);
+    ThrowNoClassDefFoundError(
+        "Initiating class loader of type %s returned class %s instead of %s.",
+        DescriptorToDot(loader_class_name).c_str(),
+        DescriptorToDot(result_name).c_str(),
+        DescriptorToDot(descriptor).c_str());
+    return nullptr;
+  }
+  // success, return mirror::Class*
+  return result_ptr.Ptr();
 }
 
 mirror::Class* ClassLinker::DefineClass(Thread* self,
@@ -3507,7 +3517,8 @@
   // class to the hash table --- necessary because of possible races with
   // other threads.)
   if (class_loader.Get() != component_type->GetClassLoader()) {
-    ObjPtr<mirror::Class> new_class = LookupClass(self, descriptor, hash, component_type->GetClassLoader());
+    ObjPtr<mirror::Class> new_class =
+        LookupClass(self, descriptor, hash, component_type->GetClassLoader());
     if (new_class != nullptr) {
       return new_class.Ptr();
     }
@@ -7725,7 +7736,7 @@
       type = LookupClass(self, descriptor, hash, class_loader.Ptr());
     }
   }
-  if (type != nullptr || type->IsResolved()) {
+  if (type != nullptr && type->IsResolved()) {
     return type.Ptr();
   }
   return nullptr;
diff --git a/runtime/class_linker_test.cc b/runtime/class_linker_test.cc
index d98daa5..7b6c0dc 100644
--- a/runtime/class_linker_test.cc
+++ b/runtime/class_linker_test.cc
@@ -487,7 +487,7 @@
       // says AccessibleObject is 9 bytes but sizeof(AccessibleObject) is 12 bytes due to padding.
       // The RoundUp is to get around this case.
       static constexpr size_t kPackAlignment = 4;
-      size_t expected_size = RoundUp(is_static ? klass->GetClassSize(): klass->GetObjectSize(),
+      size_t expected_size = RoundUp(is_static ? klass->GetClassSize() : klass->GetObjectSize(),
           kPackAlignment);
       if (sizeof(T) != expected_size) {
         LOG(ERROR) << "Class size mismatch:"
@@ -906,6 +906,41 @@
       klass);
 }
 
+TEST_F(ClassLinkerTest, LookupResolvedTypeArray) {
+  ScopedObjectAccess soa(Thread::Current());
+  StackHandleScope<2> hs(soa.Self());
+  Handle<mirror::ClassLoader> class_loader(
+      hs.NewHandle(soa.Decode<mirror::ClassLoader>(LoadDex("AllFields"))));
+  // Get the AllFields class for the dex cache and dex file.
+  ObjPtr<mirror::Class> all_fields_klass
+      = class_linker_->FindClass(soa.Self(), "LAllFields;", class_loader);
+  ASSERT_OBJ_PTR_NE(all_fields_klass, ObjPtr<mirror::Class>(nullptr));
+  Handle<mirror::DexCache> dex_cache = hs.NewHandle(all_fields_klass->GetDexCache());
+  const DexFile& dex_file = *dex_cache->GetDexFile();
+  // Get the index of the array class we want to test.
+  const DexFile::TypeId* array_id = dex_file.FindTypeId("[Ljava/lang/Object;");
+  ASSERT_TRUE(array_id != nullptr);
+  dex::TypeIndex array_idx = dex_file.GetIndexForTypeId(*array_id);
+  // Check that the array class wasn't resolved yet.
+  EXPECT_OBJ_PTR_EQ(
+      class_linker_->LookupResolvedType(dex_file, array_idx, dex_cache.Get(), class_loader.Get()),
+      ObjPtr<mirror::Class>(nullptr));
+  // Resolve the array class we want to test.
+  ObjPtr<mirror::Class> array_klass
+      = class_linker_->FindClass(soa.Self(), "[Ljava/lang/Object;", class_loader);
+  ASSERT_OBJ_PTR_NE(array_klass, ObjPtr<mirror::Class>(nullptr));
+  // Test that LookupResolvedType() finds the array class.
+  EXPECT_OBJ_PTR_EQ(
+      class_linker_->LookupResolvedType(dex_file, array_idx, dex_cache.Get(), class_loader.Get()),
+      array_klass);
+  // Zero out the resolved type and make sure LookupResolvedType() still finds it.
+  dex_cache->SetResolvedType(array_idx, nullptr);
+  EXPECT_TRUE(dex_cache->GetResolvedType(array_idx) == nullptr);
+  EXPECT_OBJ_PTR_EQ(
+      class_linker_->LookupResolvedType(dex_file, array_idx, dex_cache.Get(), class_loader.Get()),
+      array_klass);
+}
+
 TEST_F(ClassLinkerTest, LibCore) {
   ScopedObjectAccess soa(Thread::Current());
   ASSERT_TRUE(java_lang_dex_file_ != nullptr);
diff --git a/runtime/class_table.cc b/runtime/class_table.cc
index 0f985c6..ff846a7 100644
--- a/runtime/class_table.cc
+++ b/runtime/class_table.cc
@@ -129,6 +129,19 @@
   classes_.back().InsertWithHash(TableSlot(klass, hash), hash);
 }
 
+void ClassTable::CopyWithoutLocks(const ClassTable& source_table) {
+  if (kIsDebugBuild) {
+    for (ClassSet& class_set : classes_) {
+      CHECK(class_set.Empty());
+    }
+  }
+  for (const ClassSet& class_set : source_table.classes_) {
+    for (const TableSlot& slot : class_set) {
+      classes_.back().Insert(slot);
+    }
+  }
+}
+
 void ClassTable::InsertWithoutLocks(ObjPtr<mirror::Class> klass) {
   const uint32_t hash = TableSlot::HashDescriptor(klass);
   classes_.back().InsertWithHash(TableSlot(klass, hash), hash);
diff --git a/runtime/class_table.h b/runtime/class_table.h
index f27d809..c8ec28e 100644
--- a/runtime/class_table.h
+++ b/runtime/class_table.h
@@ -240,6 +240,7 @@
   }
 
  private:
+  void CopyWithoutLocks(const ClassTable& source_table) NO_THREAD_SAFETY_ANALYSIS;
   void InsertWithoutLocks(ObjPtr<mirror::Class> klass) NO_THREAD_SAFETY_ANALYSIS;
 
   size_t CountDefiningLoaderClasses(ObjPtr<mirror::ClassLoader> defining_loader,
diff --git a/runtime/gc/heap.cc b/runtime/gc/heap.cc
index 34d8284..268cca0 100644
--- a/runtime/gc/heap.cc
+++ b/runtime/gc/heap.cc
@@ -360,7 +360,7 @@
     // If we are the zygote, the non moving space becomes the zygote space when we run
     // PreZygoteFork the first time. In this case, call the map "zygote space" since we can't
     // rename the mem map later.
-    const char* space_name = is_zygote ? kZygoteSpaceName: kNonMovingSpaceName;
+    const char* space_name = is_zygote ? kZygoteSpaceName : kNonMovingSpaceName;
     // Reserve the non moving mem map before the other two since it needs to be at a specific
     // address.
     non_moving_space_mem_map.reset(
diff --git a/runtime/gc/reference_processor.h b/runtime/gc/reference_processor.h
index b15544d..38b68cb 100644
--- a/runtime/gc/reference_processor.h
+++ b/runtime/gc/reference_processor.h
@@ -42,7 +42,7 @@
 
 class Heap;
 
-// Used to process java.lang.References concurrently or paused.
+// Used to process java.lang.ref.Reference instances concurrently or paused.
 class ReferenceProcessor {
  public:
   explicit ReferenceProcessor();
diff --git a/runtime/gc/scoped_gc_critical_section.h b/runtime/gc/scoped_gc_critical_section.h
index ec93bca..1271ff7 100644
--- a/runtime/gc/scoped_gc_critical_section.h
+++ b/runtime/gc/scoped_gc_critical_section.h
@@ -27,8 +27,8 @@
 
 namespace gc {
 
-// Wait until the GC is finished and then prevent GC from starting until the destructor. Used
-// to prevent deadlocks in places where we call ClassLinker::VisitClass with all th threads
+// Wait until the GC is finished and then prevent the GC from starting until the destructor. Used
+// to prevent deadlocks in places where we call ClassLinker::VisitClass with all the threads
 // suspended.
 class ScopedGCCriticalSection {
  public:
diff --git a/runtime/interpreter/mterp/arm64/op_iput_wide_quick.S b/runtime/interpreter/mterp/arm64/op_iput_wide_quick.S
index 6cec363..28e831a 100644
--- a/runtime/interpreter/mterp/arm64/op_iput_wide_quick.S
+++ b/runtime/interpreter/mterp/arm64/op_iput_wide_quick.S
@@ -4,7 +4,7 @@
     GET_VREG w2, w2                     // w2<- fp[B], the object pointer
     ubfx    w0, wINST, #8, #4           // w0<- A
     cbz     w2, common_errNullObject    // object was null
-    GET_VREG_WIDE x0, w0                // x0-< fp[A]
+    GET_VREG_WIDE x0, w0                // x0<- fp[A]
     FETCH_ADVANCE_INST 2                // advance rPC, load wINST
     str     x0, [x2, x3]                // obj.field<- x0
     GET_INST_OPCODE ip                  // extract opcode from wINST
diff --git a/runtime/interpreter/mterp/out/mterp_arm64.S b/runtime/interpreter/mterp/out/mterp_arm64.S
index 681790d..7d442c0 100644
--- a/runtime/interpreter/mterp/out/mterp_arm64.S
+++ b/runtime/interpreter/mterp/out/mterp_arm64.S
@@ -6593,7 +6593,7 @@
     GET_VREG w2, w2                     // w2<- fp[B], the object pointer
     ubfx    w0, wINST, #8, #4           // w0<- A
     cbz     w2, common_errNullObject    // object was null
-    GET_VREG_WIDE x0, w0                // x0-< fp[A]
+    GET_VREG_WIDE x0, w0                // x0<- fp[A]
     FETCH_ADVANCE_INST 2                // advance rPC, load wINST
     str     x0, [x2, x3]                // obj.field<- x0
     GET_INST_OPCODE ip                  // extract opcode from wINST
diff --git a/runtime/interpreter/unstarted_runtime.cc b/runtime/interpreter/unstarted_runtime.cc
index 7dd3d3d..feb6e08 100644
--- a/runtime/interpreter/unstarted_runtime.cc
+++ b/runtime/interpreter/unstarted_runtime.cc
@@ -1443,7 +1443,7 @@
 
   ObjPtr<mirror::Object> java_method_obj = shadow_frame->GetVRegReference(arg_offset);
   ScopedLocalRef<jobject> java_method(env,
-      java_method_obj == nullptr ? nullptr :env->AddLocalReference<jobject>(java_method_obj));
+      java_method_obj == nullptr ? nullptr : env->AddLocalReference<jobject>(java_method_obj));
 
   ObjPtr<mirror::Object> java_receiver_obj = shadow_frame->GetVRegReference(arg_offset + 1);
   ScopedLocalRef<jobject> java_receiver(env,
diff --git a/runtime/jit/jit.cc b/runtime/jit/jit.cc
index 2bb8819..6deb03d 100644
--- a/runtime/jit/jit.cc
+++ b/runtime/jit/jit.cc
@@ -514,7 +514,7 @@
       }
     }
 
-    native_pc = stack_map.GetNativePcOffset(encoding.stack_map_encoding) +
+    native_pc = stack_map.GetNativePcOffset(encoding.stack_map_encoding, kRuntimeISA) +
         osr_method->GetEntryPoint();
     VLOG(jit) << "Jumping to "
               << method_name
diff --git a/runtime/monitor.cc b/runtime/monitor.cc
index 9c09275..071b0e2 100644
--- a/runtime/monitor.cc
+++ b/runtime/monitor.cc
@@ -1394,6 +1394,12 @@
   }
 }
 
+size_t MonitorList::Size() {
+  Thread* self = Thread::Current();
+  MutexLock mu(self, monitor_list_lock_);
+  return list_.size();
+}
+
 class MonitorDeflateVisitor : public IsMarkedVisitor {
  public:
   MonitorDeflateVisitor() : self_(Thread::Current()), deflate_count_(0) {}
diff --git a/runtime/monitor.h b/runtime/monitor.h
index c3da563..1fa4682 100644
--- a/runtime/monitor.h
+++ b/runtime/monitor.h
@@ -331,6 +331,7 @@
   void BroadcastForNewMonitors() REQUIRES(!monitor_list_lock_);
   // Returns how many monitors were deflated.
   size_t DeflateMonitors() REQUIRES(!monitor_list_lock_) REQUIRES(Locks::mutator_lock_);
+  size_t Size() REQUIRES(!monitor_list_lock_);
 
   typedef std::list<Monitor*, TrackingAllocator<Monitor*, kAllocatorTagMonitorList>> Monitors;
 
diff --git a/runtime/oat.h b/runtime/oat.h
index 953b445..29821a2 100644
--- a/runtime/oat.h
+++ b/runtime/oat.h
@@ -32,7 +32,7 @@
 class PACKED(4) OatHeader {
  public:
   static constexpr uint8_t kOatMagic[] = { 'o', 'a', 't', '\n' };
-  static constexpr uint8_t kOatVersion[] = { '1', '0', '2', '\0' };  // Enabling CC
+  static constexpr uint8_t kOatVersion[] = { '1', '0', '3', '\0' };  // Native pc change
 
   static constexpr const char* kImageLocationKey = "image-location";
   static constexpr const char* kDex2OatCmdLineKey = "dex2oat-cmdline";
diff --git a/runtime/oat_quick_method_header.cc b/runtime/oat_quick_method_header.cc
index 9c2378d..fd84426 100644
--- a/runtime/oat_quick_method_header.cc
+++ b/runtime/oat_quick_method_header.cc
@@ -80,7 +80,7 @@
                                    : code_info.GetStackMapForDexPc(dex_pc, encoding);
   if (stack_map.IsValid()) {
     return reinterpret_cast<uintptr_t>(entry_point) +
-           stack_map.GetNativePcOffset(encoding.stack_map_encoding);
+           stack_map.GetNativePcOffset(encoding.stack_map_encoding, kRuntimeISA);
   }
   if (abort_on_failure) {
     ScopedObjectAccess soa(Thread::Current());
diff --git a/runtime/openjdkjvmti/Android.bp b/runtime/openjdkjvmti/Android.bp
index a731c17..976a1e7 100644
--- a/runtime/openjdkjvmti/Android.bp
+++ b/runtime/openjdkjvmti/Android.bp
@@ -22,6 +22,7 @@
            "OpenjdkJvmTi.cc",
            "ti_class.cc",
            "ti_class_definition.cc",
+           "ti_dump.cc",
            "ti_field.cc",
            "ti_heap.cc",
            "ti_jni.cc",
diff --git a/runtime/openjdkjvmti/OpenjdkJvmTi.cc b/runtime/openjdkjvmti/OpenjdkJvmTi.cc
index fcedd4e..417d104 100644
--- a/runtime/openjdkjvmti/OpenjdkJvmTi.cc
+++ b/runtime/openjdkjvmti/OpenjdkJvmTi.cc
@@ -49,6 +49,7 @@
 #include "thread-inl.h"
 #include "thread_list.h"
 #include "ti_class.h"
+#include "ti_dump.h"
 #include "ti_field.h"
 #include "ti_heap.h"
 #include "ti_jni.h"
@@ -1309,6 +1310,9 @@
     PhaseUtil::SetToOnLoad();
   }
   PhaseUtil::Register(&gEventHandler);
+  ThreadUtil::Register(&gEventHandler);
+  ClassUtil::Register(&gEventHandler);
+  DumpUtil::Register(&gEventHandler);
 
   runtime->GetJavaVM()->AddEnvironmentHook(GetEnvHandler);
   runtime->AddSystemWeakHolder(&gObjectTagTable);
@@ -1316,6 +1320,15 @@
   return true;
 }
 
+extern "C" bool ArtPlugin_Deinitialize() {
+  PhaseUtil::Unregister();
+  ThreadUtil::Unregister();
+  ClassUtil::Unregister();
+  DumpUtil::Unregister();
+
+  return true;
+}
+
 // The actual struct holding all of the entrypoints into the jvmti interface.
 const jvmtiInterface_1 gJvmtiInterface = {
   nullptr,  // reserved1
diff --git a/runtime/openjdkjvmti/events.cc b/runtime/openjdkjvmti/events.cc
index 7182055..d3f8001 100644
--- a/runtime/openjdkjvmti/events.cc
+++ b/runtime/openjdkjvmti/events.cc
@@ -212,7 +212,7 @@
                               thread.get(),
                               object.get(),
                               klass.get(),
-                              byte_count);
+                              static_cast<jlong>(byte_count));
     }
   }
 
diff --git a/runtime/openjdkjvmti/ti_class.cc b/runtime/openjdkjvmti/ti_class.cc
index abcc849..450f75a 100644
--- a/runtime/openjdkjvmti/ti_class.cc
+++ b/runtime/openjdkjvmti/ti_class.cc
@@ -31,16 +31,119 @@
 
 #include "ti_class.h"
 
+#include <mutex>
+#include <unordered_set>
+
 #include "art_jvmti.h"
+#include "base/macros.h"
 #include "class_table-inl.h"
 #include "class_linker.h"
+#include "events-inl.h"
+#include "handle.h"
+#include "jni_env_ext-inl.h"
 #include "jni_internal.h"
 #include "runtime.h"
+#include "runtime_callbacks.h"
+#include "ScopedLocalRef.h"
 #include "scoped_thread_state_change-inl.h"
 #include "thread-inl.h"
+#include "thread_list.h"
 
 namespace openjdkjvmti {
 
+struct ClassCallback : public art::ClassLoadCallback {
+  void ClassLoad(art::Handle<art::mirror::Class> klass) REQUIRES_SHARED(art::Locks::mutator_lock_) {
+    if (event_handler->IsEventEnabledAnywhere(ArtJvmtiEvent::kClassLoad)) {
+      art::Thread* thread = art::Thread::Current();
+      ScopedLocalRef<jclass> jklass(thread->GetJniEnv(),
+                                    thread->GetJniEnv()->AddLocalReference<jclass>(klass.Get()));
+      ScopedLocalRef<jclass> jthread(
+          thread->GetJniEnv(), thread->GetJniEnv()->AddLocalReference<jclass>(thread->GetPeer()));
+      {
+        art::ScopedThreadSuspension sts(thread, art::ThreadState::kNative);
+        event_handler->DispatchEvent(thread,
+                                     ArtJvmtiEvent::kClassLoad,
+                                     reinterpret_cast<JNIEnv*>(thread->GetJniEnv()),
+                                     jthread.get(),
+                                     jklass.get());
+      }
+      AddTempClass(thread, jklass.get());
+    }
+  }
+
+  void ClassPrepare(art::Handle<art::mirror::Class> temp_klass ATTRIBUTE_UNUSED,
+                    art::Handle<art::mirror::Class> klass)
+      REQUIRES_SHARED(art::Locks::mutator_lock_) {
+    if (event_handler->IsEventEnabledAnywhere(ArtJvmtiEvent::kClassPrepare)) {
+      art::Thread* thread = art::Thread::Current();
+      ScopedLocalRef<jclass> jklass(thread->GetJniEnv(),
+                                    thread->GetJniEnv()->AddLocalReference<jclass>(klass.Get()));
+      ScopedLocalRef<jclass> jthread(
+          thread->GetJniEnv(), thread->GetJniEnv()->AddLocalReference<jclass>(thread->GetPeer()));
+      art::ScopedThreadSuspension sts(thread, art::ThreadState::kNative);
+      event_handler->DispatchEvent(thread,
+                                   ArtJvmtiEvent::kClassPrepare,
+                                   reinterpret_cast<JNIEnv*>(thread->GetJniEnv()),
+                                   jthread.get(),
+                                   jklass.get());
+    }
+  }
+
+  void AddTempClass(art::Thread* self, jclass klass) {
+    std::unique_lock<std::mutex> mu(temp_classes_lock);
+    temp_classes.push_back(reinterpret_cast<jclass>(self->GetJniEnv()->NewGlobalRef(klass)));
+  }
+
+  void HandleTempClass(art::Handle<art::mirror::Class> temp_klass,
+                       art::Handle<art::mirror::Class> klass)
+      REQUIRES_SHARED(art::Locks::mutator_lock_) {
+    std::unique_lock<std::mutex> mu(temp_classes_lock);
+    if (temp_classes.empty()) {
+      return;
+    }
+
+    art::Thread* self = art::Thread::Current();
+    for (auto it = temp_classes.begin(); it != temp_classes.end(); ++it) {
+      if (temp_klass.Get() == art::ObjPtr<art::mirror::Class>::DownCast(self->DecodeJObject(*it))) {
+        temp_classes.erase(it);
+        FixupTempClass(temp_klass, klass);
+      }
+    }
+  }
+
+  void FixupTempClass(art::Handle<art::mirror::Class> temp_klass ATTRIBUTE_UNUSED,
+                      art::Handle<art::mirror::Class> klass ATTRIBUTE_UNUSED)
+     REQUIRES_SHARED(art::Locks::mutator_lock_) {
+    // TODO: Implement.
+  }
+
+  // A set of all the temp classes we have handed out. We have to fix up references to these.
+  // For simplicity, we store the temp classes as JNI global references in a vector. Normally a
+  // Prepare event will closely follow, so the vector should be small.
+  std::mutex temp_classes_lock;
+  std::vector<jclass> temp_classes;
+
+  EventHandler* event_handler = nullptr;
+};
+
+ClassCallback gClassCallback;
+
+void ClassUtil::Register(EventHandler* handler) {
+  gClassCallback.event_handler = handler;
+  art::ScopedThreadStateChange stsc(art::Thread::Current(),
+                                    art::ThreadState::kWaitingForDebuggerToAttach);
+  art::ScopedSuspendAll ssa("Add load callback");
+  art::Runtime::Current()->GetRuntimeCallbacks()->AddClassLoadCallback(&gClassCallback);
+}
+
+void ClassUtil::Unregister() {
+  art::ScopedThreadStateChange stsc(art::Thread::Current(),
+                                    art::ThreadState::kWaitingForDebuggerToAttach);
+  art::ScopedSuspendAll ssa("Remove thread callback");
+  art::Runtime* runtime = art::Runtime::Current();
+  runtime->GetRuntimeCallbacks()->RemoveClassLoadCallback(&gClassCallback);
+}
+
 jvmtiError ClassUtil::GetClassFields(jvmtiEnv* env,
                                      jclass jklass,
                                      jint* field_count_ptr,
@@ -200,7 +303,9 @@
   }
 
   // TODO: Support generic signature.
-  *generic_ptr = nullptr;
+  if (generic_ptr != nullptr) {
+    *generic_ptr = nullptr;
+  }
 
   // Everything is fine, release the buffers.
   sig_copy.release();
diff --git a/runtime/openjdkjvmti/ti_class.h b/runtime/openjdkjvmti/ti_class.h
index 9558894..aa2260f 100644
--- a/runtime/openjdkjvmti/ti_class.h
+++ b/runtime/openjdkjvmti/ti_class.h
@@ -37,8 +37,13 @@
 
 namespace openjdkjvmti {
 
+class EventHandler;
+
 class ClassUtil {
  public:
+  static void Register(EventHandler* event_handler);
+  static void Unregister();
+
   static jvmtiError GetClassFields(jvmtiEnv* env,
                                    jclass klass,
                                    jint* field_count_ptr,
diff --git a/runtime/openjdkjvmti/ti_dump.cc b/runtime/openjdkjvmti/ti_dump.cc
new file mode 100644
index 0000000..2ee5c40
--- /dev/null
+++ b/runtime/openjdkjvmti/ti_dump.cc
@@ -0,0 +1,74 @@
+/* Copyright (C) 2017 The Android Open Source Project
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This file implements interfaces from the file jvmti.h. This implementation
+ * is licensed under the same terms as the file jvmti.h.  The
+ * copyright and license information for the file jvmti.h follows.
+ *
+ * Copyright (c) 2003, 2011, Oracle and/or its affiliates. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.  Oracle designates this
+ * particular file as subject to the "Classpath" exception as provided
+ * by Oracle in the LICENSE file that accompanied this code.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ */
+
+#include "ti_dump.h"
+
+#include <limits>
+
+
+#include "art_jvmti.h"
+#include "base/mutex.h"
+#include "events-inl.h"
+#include "runtime_callbacks.h"
+#include "scoped_thread_state_change-inl.h"
+#include "thread-inl.h"
+#include "thread_list.h"
+
+namespace openjdkjvmti {
+
+struct DumpCallback : public art::RuntimeSigQuitCallback {
+  void SigQuit() OVERRIDE REQUIRES_SHARED(art::Locks::mutator_lock_) {
+    art::Thread* thread = art::Thread::Current();
+    art::ScopedThreadSuspension sts(thread, art::ThreadState::kNative);
+    event_handler->DispatchEvent(nullptr, ArtJvmtiEvent::kDataDumpRequest);
+  }
+
+  EventHandler* event_handler = nullptr;
+};
+
+static DumpCallback gDumpCallback;
+
+void DumpUtil::Register(EventHandler* handler) {
+  gDumpCallback.event_handler = handler;
+  art::ScopedThreadStateChange stsc(art::Thread::Current(),
+                                    art::ThreadState::kWaitingForDebuggerToAttach);
+  art::ScopedSuspendAll ssa("Add sigquit callback");
+  art::Runtime::Current()->GetRuntimeCallbacks()->AddRuntimeSigQuitCallback(&gDumpCallback);
+}
+
+void DumpUtil::Unregister() {
+  art::ScopedThreadStateChange stsc(art::Thread::Current(),
+                                    art::ThreadState::kWaitingForDebuggerToAttach);
+  art::ScopedSuspendAll ssa("Remove sigquit callback");
+  art::Runtime::Current()->GetRuntimeCallbacks()->RemoveRuntimeSigQuitCallback(&gDumpCallback);
+}
+
+}  // namespace openjdkjvmti
diff --git a/runtime/openjdkjvmti/ti_dump.h b/runtime/openjdkjvmti/ti_dump.h
new file mode 100644
index 0000000..67cb239
--- /dev/null
+++ b/runtime/openjdkjvmti/ti_dump.h
@@ -0,0 +1,50 @@
+/* Copyright (C) 2017 The Android Open Source Project
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This file implements interfaces from the file jvmti.h. This implementation
+ * is licensed under the same terms as the file jvmti.h.  The
+ * copyright and license information for the file jvmti.h follows.
+ *
+ * Copyright (c) 2003, 2011, Oracle and/or its affiliates. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.  Oracle designates this
+ * particular file as subject to the "Classpath" exception as provided
+ * by Oracle in the LICENSE file that accompanied this code.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ */
+
+#ifndef ART_RUNTIME_OPENJDKJVMTI_TI_DUMP_H_
+#define ART_RUNTIME_OPENJDKJVMTI_TI_DUMP_H_
+
+#include "jni.h"
+#include "jvmti.h"
+
+namespace openjdkjvmti {
+
+class EventHandler;
+
+class DumpUtil {
+ public:
+  static void Register(EventHandler* event_handler);
+  static void Unregister();
+};
+
+}  // namespace openjdkjvmti
+
+#endif  // ART_RUNTIME_OPENJDKJVMTI_TI_DUMP_H_
diff --git a/runtime/openjdkjvmti/ti_phase.cc b/runtime/openjdkjvmti/ti_phase.cc
index 85d6b72..154406a 100644
--- a/runtime/openjdkjvmti/ti_phase.cc
+++ b/runtime/openjdkjvmti/ti_phase.cc
@@ -55,19 +55,23 @@
     return soa.AddLocalReference<jthread>(soa.Self()->GetPeer());
   }
 
-  void NextRuntimePhase(RuntimePhase phase) OVERRIDE {
+  void NextRuntimePhase(RuntimePhase phase) REQUIRES_SHARED(art::Locks::mutator_lock_) OVERRIDE {
     // TODO: Events.
     switch (phase) {
       case RuntimePhase::kInitialAgents:
         PhaseUtil::current_phase_ = JVMTI_PHASE_PRIMORDIAL;
         break;
       case RuntimePhase::kStart:
-        event_handler->DispatchEvent(nullptr, ArtJvmtiEvent::kVmStart, GetJniEnv());
-        PhaseUtil::current_phase_ = JVMTI_PHASE_START;
+        {
+          art::ScopedThreadSuspension sts(art::Thread::Current(), art::ThreadState::kNative);
+          event_handler->DispatchEvent(nullptr, ArtJvmtiEvent::kVmStart, GetJniEnv());
+          PhaseUtil::current_phase_ = JVMTI_PHASE_START;
+        }
         break;
       case RuntimePhase::kInit:
         {
           ScopedLocalRef<jthread> thread(GetJniEnv(), GetCurrentJThread());
+          art::ScopedThreadSuspension sts(art::Thread::Current(), art::ThreadState::kNative);
           event_handler->DispatchEvent(nullptr,
                                        ArtJvmtiEvent::kVmInit,
                                        GetJniEnv(),
@@ -76,8 +80,11 @@
         }
         break;
       case RuntimePhase::kDeath:
-        event_handler->DispatchEvent(nullptr, ArtJvmtiEvent::kVmDeath, GetJniEnv());
-        PhaseUtil::current_phase_ = JVMTI_PHASE_DEAD;
+        {
+          art::ScopedThreadSuspension sts(art::Thread::Current(), art::ThreadState::kNative);
+          event_handler->DispatchEvent(nullptr, ArtJvmtiEvent::kVmDeath, GetJniEnv());
+          PhaseUtil::current_phase_ = JVMTI_PHASE_DEAD;
+        }
         // TODO: Block events now.
         break;
     }
@@ -125,5 +132,11 @@
   art::Runtime::Current()->GetRuntimeCallbacks()->AddRuntimePhaseCallback(&gPhaseCallback);
 }
 
+void PhaseUtil::Unregister() {
+  art::ScopedThreadStateChange stsc(art::Thread::Current(),
+                                    art::ThreadState::kWaitingForDebuggerToAttach);
+  art::ScopedSuspendAll ssa("Remove phase callback");
+  art::Runtime::Current()->GetRuntimeCallbacks()->RemoveRuntimePhaseCallback(&gPhaseCallback);
+}
 
 }  // namespace openjdkjvmti
diff --git a/runtime/openjdkjvmti/ti_phase.h b/runtime/openjdkjvmti/ti_phase.h
index 054652a..bd15fa6 100644
--- a/runtime/openjdkjvmti/ti_phase.h
+++ b/runtime/openjdkjvmti/ti_phase.h
@@ -44,6 +44,7 @@
   static jvmtiError GetPhase(jvmtiEnv* env, jvmtiPhase* phase_ptr);
 
   static void Register(EventHandler* event_handler);
+  static void Unregister();
 
   // Move the phase from unitialized to LOAD.
   static void SetToOnLoad();
diff --git a/runtime/openjdkjvmti/ti_thread.cc b/runtime/openjdkjvmti/ti_thread.cc
index 970cc24..9f81d6b 100644
--- a/runtime/openjdkjvmti/ti_thread.cc
+++ b/runtime/openjdkjvmti/ti_thread.cc
@@ -31,10 +31,12 @@
 
 #include "ti_thread.h"
 
+#include "android-base/strings.h"
 #include "art_field.h"
 #include "art_jvmti.h"
 #include "base/logging.h"
 #include "base/mutex.h"
+#include "events-inl.h"
 #include "gc/system_weak.h"
 #include "gc_root-inl.h"
 #include "jni_internal.h"
@@ -43,6 +45,8 @@
 #include "mirror/string.h"
 #include "obj_ptr.h"
 #include "runtime.h"
+#include "runtime_callbacks.h"
+#include "ScopedLocalRef.h"
 #include "scoped_thread_state_change-inl.h"
 #include "thread-inl.h"
 #include "thread_list.h"
@@ -50,6 +54,77 @@
 
 namespace openjdkjvmti {
 
+struct ThreadCallback : public art::ThreadLifecycleCallback, public art::RuntimePhaseCallback {
+  jthread GetThreadObject(art::Thread* self) REQUIRES_SHARED(art::Locks::mutator_lock_) {
+    if (self->GetPeer() == nullptr) {
+      return nullptr;
+    }
+    return self->GetJniEnv()->AddLocalReference<jthread>(self->GetPeer());
+  }
+  void Post(art::Thread* self, ArtJvmtiEvent type) REQUIRES_SHARED(art::Locks::mutator_lock_) {
+    DCHECK_EQ(self, art::Thread::Current());
+    ScopedLocalRef<jthread> thread(self->GetJniEnv(), GetThreadObject(self));
+    art::ScopedThreadSuspension sts(self, art::ThreadState::kNative);
+    event_handler->DispatchEvent(self, type, self->GetJniEnv(), thread.get());
+  }
+
+  void ThreadStart(art::Thread* self) OVERRIDE REQUIRES_SHARED(art::Locks::mutator_lock_) {
+    if (!started) {
+      // Runtime isn't started. We only expect at most the signal handler or JIT threads to be
+      // started here.
+      if (art::kIsDebugBuild) {
+        std::string name;
+        self->GetThreadName(name);
+        if (name != "Signal Catcher" && !android::base::StartsWith(name, "Jit thread pool")) {
+          LOG(FATAL) << "Unexpected thread before start: " << name;
+        }
+      }
+      return;
+    }
+    Post(self, ArtJvmtiEvent::kThreadStart);
+  }
+
+  void ThreadDeath(art::Thread* self) OVERRIDE REQUIRES_SHARED(art::Locks::mutator_lock_) {
+    Post(self, ArtJvmtiEvent::kThreadEnd);
+  }
+
+  void NextRuntimePhase(RuntimePhase phase) OVERRIDE REQUIRES_SHARED(art::Locks::mutator_lock_) {
+    if (phase == RuntimePhase::kInit) {
+      // We moved to VMInit. Report the main thread as started (it was attached early, and must
+      // not be reported until Init.
+      started = true;
+      Post(art::Thread::Current(), ArtJvmtiEvent::kThreadStart);
+    }
+  }
+
+  EventHandler* event_handler = nullptr;
+  bool started = false;
+};
+
+ThreadCallback gThreadCallback;
+
+void ThreadUtil::Register(EventHandler* handler) {
+  art::Runtime* runtime = art::Runtime::Current();
+
+  gThreadCallback.started = runtime->IsStarted();
+  gThreadCallback.event_handler = handler;
+
+  art::ScopedThreadStateChange stsc(art::Thread::Current(),
+                                    art::ThreadState::kWaitingForDebuggerToAttach);
+  art::ScopedSuspendAll ssa("Add thread callback");
+  runtime->GetRuntimeCallbacks()->AddThreadLifecycleCallback(&gThreadCallback);
+  runtime->GetRuntimeCallbacks()->AddRuntimePhaseCallback(&gThreadCallback);
+}
+
+void ThreadUtil::Unregister() {
+  art::ScopedThreadStateChange stsc(art::Thread::Current(),
+                                    art::ThreadState::kWaitingForDebuggerToAttach);
+  art::ScopedSuspendAll ssa("Remove thread callback");
+  art::Runtime* runtime = art::Runtime::Current();
+  runtime->GetRuntimeCallbacks()->RemoveThreadLifecycleCallback(&gThreadCallback);
+  runtime->GetRuntimeCallbacks()->RemoveRuntimePhaseCallback(&gThreadCallback);
+}
+
 jvmtiError ThreadUtil::GetCurrentThread(jvmtiEnv* env ATTRIBUTE_UNUSED, jthread* thread_ptr) {
   art::Thread* self = art::Thread::Current();
 
diff --git a/runtime/openjdkjvmti/ti_thread.h b/runtime/openjdkjvmti/ti_thread.h
index 5aaec58..f6f93ee 100644
--- a/runtime/openjdkjvmti/ti_thread.h
+++ b/runtime/openjdkjvmti/ti_thread.h
@@ -37,8 +37,13 @@
 
 namespace openjdkjvmti {
 
+class EventHandler;
+
 class ThreadUtil {
  public:
+  static void Register(EventHandler* event_handler);
+  static void Unregister();
+
   static jvmtiError GetAllThreads(jvmtiEnv* env, jint* threads_count_ptr, jthread** threads_ptr);
 
   static jvmtiError GetCurrentThread(jvmtiEnv* env, jthread* thread_ptr);
diff --git a/runtime/parsed_options.cc b/runtime/parsed_options.cc
index a72159b..d1ad77c 100644
--- a/runtime/parsed_options.cc
+++ b/runtime/parsed_options.cc
@@ -302,6 +302,9 @@
           .IntoKey(M::Plugins)
       .Define("-Xfully-deoptable")
           .IntoKey(M::FullyDeoptable)
+      .Define("-XX:ThreadSuspendTimeout=_")  // in ms
+          .WithType<MillisecondsToNanoseconds>()  // store as ns
+          .IntoKey(M::ThreadSuspendTimeout)
       .Ignore({
           "-ea", "-da", "-enableassertions", "-disableassertions", "--runtime-arg", "-esa",
           "-dsa", "-enablesystemassertions", "-disablesystemassertions", "-Xrs", "-Xint:_",
@@ -724,6 +727,7 @@
   UsageMessage(stream, "  -XX:MaxSpinsBeforeThinLockInflation=integervalue\n");
   UsageMessage(stream, "  -XX:LongPauseLogThreshold=integervalue\n");
   UsageMessage(stream, "  -XX:LongGCLogThreshold=integervalue\n");
+  UsageMessage(stream, "  -XX:ThreadSuspendTimeout=integervalue\n");
   UsageMessage(stream, "  -XX:DumpGCPerformanceOnShutdown\n");
   UsageMessage(stream, "  -XX:DumpJITInfoOnShutdown\n");
   UsageMessage(stream, "  -XX:IgnoreMaxFootprint\n");
diff --git a/runtime/runtime.cc b/runtime/runtime.cc
index 4936a2f..06cd7ff 100644
--- a/runtime/runtime.cc
+++ b/runtime/runtime.cc
@@ -735,6 +735,13 @@
                             GetInstructionSetString(kRuntimeISA));
   }
 
+  // Send the initialized phase event. Send it before starting daemons, as otherwise
+  // sending thread events becomes complicated.
+  {
+    ScopedObjectAccess soa(self);
+    callbacks_->NextRuntimePhase(RuntimePhaseCallback::RuntimePhase::kInit);
+  }
+
   StartDaemonThreads();
 
   {
@@ -756,12 +763,6 @@
                  0);
   }
 
-  // Send the initialized phase event.
-  {
-    ScopedObjectAccess soa(self);
-    callbacks_->NextRuntimePhase(RuntimePhaseCallback::RuntimePhase::kInit);
-  }
-
   return true;
 }
 
@@ -1045,7 +1046,7 @@
 
   monitor_list_ = new MonitorList;
   monitor_pool_ = MonitorPool::Create();
-  thread_list_ = new ThreadList;
+  thread_list_ = new ThreadList(runtime_options.GetOrDefault(Opt::ThreadSuspendTimeout));
   intern_table_ = new InternTable;
 
   verify_ = runtime_options.GetOrDefault(Opt::Verify);
diff --git a/runtime/runtime_options.cc b/runtime/runtime_options.cc
index e75481c..aa14719 100644
--- a/runtime/runtime_options.cc
+++ b/runtime/runtime_options.cc
@@ -21,6 +21,7 @@
 #include "gc/heap.h"
 #include "monitor.h"
 #include "runtime.h"
+#include "thread_list.h"
 #include "trace.h"
 #include "utils.h"
 #include "debugger.h"
diff --git a/runtime/runtime_options.def b/runtime/runtime_options.def
index ecabf9a..749a36e 100644
--- a/runtime/runtime_options.def
+++ b/runtime/runtime_options.def
@@ -60,6 +60,8 @@
                                           LongPauseLogThreshold,          gc::Heap::kDefaultLongPauseLogThreshold)
 RUNTIME_OPTIONS_KEY (MillisecondsToNanoseconds, \
                                           LongGCLogThreshold,             gc::Heap::kDefaultLongGCLogThreshold)
+RUNTIME_OPTIONS_KEY (MillisecondsToNanoseconds, \
+                                          ThreadSuspendTimeout,           ThreadList::kDefaultThreadSuspendTimeout)
 RUNTIME_OPTIONS_KEY (Unit,                DumpGCPerformanceOnShutdown)
 RUNTIME_OPTIONS_KEY (Unit,                DumpJITInfoOnShutdown)
 RUNTIME_OPTIONS_KEY (Unit,                IgnoreMaxFootprint)
diff --git a/runtime/stack_map.cc b/runtime/stack_map.cc
index 9ebf9a7..690b069 100644
--- a/runtime/stack_map.cc
+++ b/runtime/stack_map.cc
@@ -116,7 +116,8 @@
 void CodeInfo::Dump(VariableIndentationOutputStream* vios,
                     uint32_t code_offset,
                     uint16_t number_of_dex_registers,
-                    bool dump_stack_maps) const {
+                    bool dump_stack_maps,
+                    InstructionSet instruction_set) const {
   CodeInfoEncoding encoding = ExtractEncoding();
   size_t number_of_stack_maps = GetNumberOfStackMaps(encoding);
   vios->Stream()
@@ -139,6 +140,7 @@
                      encoding,
                      code_offset,
                      number_of_dex_registers,
+                     instruction_set,
                      " " + std::to_string(i));
     }
   }
@@ -188,14 +190,17 @@
                     const CodeInfoEncoding& encoding,
                     uint32_t code_offset,
                     uint16_t number_of_dex_registers,
+                    InstructionSet instruction_set,
                     const std::string& header_suffix) const {
   StackMapEncoding stack_map_encoding = encoding.stack_map_encoding;
+  const uint32_t pc_offset = GetNativePcOffset(stack_map_encoding, instruction_set);
   vios->Stream()
       << "StackMap" << header_suffix
       << std::hex
-      << " [native_pc=0x" << code_offset + GetNativePcOffset(stack_map_encoding) << "]"
+      << " [native_pc=0x" << code_offset + pc_offset << "]"
+      << " [entry_size=0x" << encoding.stack_map_size_in_bytes << "]"
       << " (dex_pc=0x" << GetDexPc(stack_map_encoding)
-      << ", native_pc_offset=0x" << GetNativePcOffset(stack_map_encoding)
+      << ", native_pc_offset=0x" << pc_offset
       << ", dex_register_map_offset=0x" << GetDexRegisterMapOffset(stack_map_encoding)
       << ", inline_info_offset=0x" << GetInlineDescriptorOffset(stack_map_encoding)
       << ", register_mask=0x" << GetRegisterMask(stack_map_encoding)
diff --git a/runtime/stack_map.h b/runtime/stack_map.h
index 13886f2..cd9a3f0 100644
--- a/runtime/stack_map.h
+++ b/runtime/stack_map.h
@@ -17,6 +17,7 @@
 #ifndef ART_RUNTIME_STACK_MAP_H_
 #define ART_RUNTIME_STACK_MAP_H_
 
+#include "arch/code_offset.h"
 #include "base/bit_vector.h"
 #include "base/bit_utils.h"
 #include "dex_file.h"
@@ -805,12 +806,16 @@
     encoding.GetDexPcEncoding().Store(region_, dex_pc);
   }
 
-  ALWAYS_INLINE uint32_t GetNativePcOffset(const StackMapEncoding& encoding) const {
-    return encoding.GetNativePcEncoding().Load(region_);
+  ALWAYS_INLINE uint32_t GetNativePcOffset(const StackMapEncoding& encoding,
+                                           InstructionSet instruction_set) const {
+    CodeOffset offset(
+        CodeOffset::FromCompressedOffset(encoding.GetNativePcEncoding().Load(region_)));
+    return offset.Uint32Value(instruction_set);
   }
 
-  ALWAYS_INLINE void SetNativePcOffset(const StackMapEncoding& encoding, uint32_t native_pc_offset) {
-    encoding.GetNativePcEncoding().Store(region_, native_pc_offset);
+  ALWAYS_INLINE void SetNativePcCodeOffset(const StackMapEncoding& encoding,
+                                           CodeOffset native_pc_offset) {
+    encoding.GetNativePcEncoding().Store(region_, native_pc_offset.CompressedValue());
   }
 
   ALWAYS_INLINE uint32_t GetDexRegisterMapOffset(const StackMapEncoding& encoding) const {
@@ -866,6 +871,7 @@
             const CodeInfoEncoding& encoding,
             uint32_t code_offset,
             uint16_t number_of_dex_registers,
+            InstructionSet instruction_set,
             const std::string& header_suffix = "") const;
 
   // Special (invalid) offset for the DexRegisterMapOffset field meaning
@@ -1176,6 +1182,17 @@
     }
   }
 
+  size_t GetDexRegisterMapsSize(const CodeInfoEncoding& encoding,
+                                uint32_t number_of_dex_registers) const {
+    size_t total = 0;
+    for (size_t i = 0, e = GetNumberOfStackMaps(encoding); i < e; ++i) {
+      StackMap stack_map = GetStackMapAt(i, encoding);
+      DexRegisterMap map(GetDexRegisterMapOf(stack_map, encoding, number_of_dex_registers));
+      total += map.Size();
+    }
+    return total;
+  }
+
   // Return the `DexRegisterMap` pointed by `inline_info` at depth `depth`.
   DexRegisterMap GetDexRegisterMapAtDepth(uint8_t depth,
                                           InlineInfo inline_info,
@@ -1234,15 +1251,16 @@
       if (stack_map.GetDexPc(stack_map_encoding) == dex_pc) {
         StackMap other = GetStackMapAt(i + 1, encoding);
         if (other.GetDexPc(stack_map_encoding) == dex_pc &&
-            other.GetNativePcOffset(stack_map_encoding) ==
-                stack_map.GetNativePcOffset(stack_map_encoding)) {
+            other.GetNativePcOffset(stack_map_encoding, kRuntimeISA) ==
+                stack_map.GetNativePcOffset(stack_map_encoding, kRuntimeISA)) {
           DCHECK_EQ(other.GetDexRegisterMapOffset(stack_map_encoding),
                     stack_map.GetDexRegisterMapOffset(stack_map_encoding));
           DCHECK(!stack_map.HasInlineInfo(stack_map_encoding));
           if (i < e - 2) {
             // Make sure there are not three identical stack maps following each other.
-            DCHECK_NE(stack_map.GetNativePcOffset(stack_map_encoding),
-                      GetStackMapAt(i + 2, encoding).GetNativePcOffset(stack_map_encoding));
+            DCHECK_NE(
+                stack_map.GetNativePcOffset(stack_map_encoding, kRuntimeISA),
+                GetStackMapAt(i + 2, encoding).GetNativePcOffset(stack_map_encoding, kRuntimeISA));
           }
           return stack_map;
         }
@@ -1258,7 +1276,8 @@
     //       we could do binary search.
     for (size_t i = 0, e = GetNumberOfStackMaps(encoding); i < e; ++i) {
       StackMap stack_map = GetStackMapAt(i, encoding);
-      if (stack_map.GetNativePcOffset(encoding.stack_map_encoding) == native_pc_offset) {
+      if (stack_map.GetNativePcOffset(encoding.stack_map_encoding, kRuntimeISA) ==
+          native_pc_offset) {
         return stack_map;
       }
     }
@@ -1273,7 +1292,8 @@
   void Dump(VariableIndentationOutputStream* vios,
             uint32_t code_offset,
             uint16_t number_of_dex_registers,
-            bool dump_stack_maps) const;
+            bool dump_stack_maps,
+            InstructionSet instruction_set) const;
 
   // Check that the code info has valid stack map and abort if it does not.
   void AssertValidStackMap(const CodeInfoEncoding& encoding) const {
diff --git a/runtime/thread_list.cc b/runtime/thread_list.cc
index c5c7e2c..01c940e 100644
--- a/runtime/thread_list.cc
+++ b/runtime/thread_list.cc
@@ -57,7 +57,6 @@
 using android::base::StringPrintf;
 
 static constexpr uint64_t kLongThreadSuspendThreshold = MsToNs(5);
-static constexpr uint64_t kThreadSuspendTimeoutMs = 30 * 1000;  // 30s.
 // Use 0 since we want to yield to prevent blocking for an unpredictable amount of time.
 static constexpr useconds_t kThreadSuspendInitialSleepUs = 0;
 static constexpr useconds_t kThreadSuspendMaxYieldUs = 3000;
@@ -68,12 +67,13 @@
 // Turned off again. b/29248079
 static constexpr bool kDumpUnattachedThreadNativeStackForSigQuit = false;
 
-ThreadList::ThreadList()
+ThreadList::ThreadList(uint64_t thread_suspend_timeout_ns)
     : suspend_all_count_(0),
       debug_suspend_all_count_(0),
       unregistering_count_(0),
       suspend_all_historam_("suspend all histogram", 16, 64),
       long_suspend_(false),
+      thread_suspend_timeout_ns_(thread_suspend_timeout_ns),
       empty_checkpoint_barrier_(new Barrier(0)) {
   CHECK(Monitor::IsValidLockWord(LockWord::FromThinLockId(kMaxThreadId, 1, 0U)));
 }
@@ -554,12 +554,14 @@
     // Make sure this thread grabs exclusive access to the mutator lock and its protected data.
 #if HAVE_TIMED_RWLOCK
     while (true) {
-      if (Locks::mutator_lock_->ExclusiveLockWithTimeout(self, kThreadSuspendTimeoutMs, 0)) {
+      if (Locks::mutator_lock_->ExclusiveLockWithTimeout(self,
+                                                         NsToMs(thread_suspend_timeout_ns_),
+                                                         0)) {
         break;
       } else if (!long_suspend_) {
         // Reading long_suspend without the mutator lock is slightly racy, in some rare cases, this
         // could result in a thread suspend timeout.
-        // Timeout if we wait more than kThreadSuspendTimeoutMs seconds.
+        // Timeout if we wait more than thread_suspend_timeout_ns_ nanoseconds.
         UnsafeLogFatalForThreadSuspendAllTimeout();
       }
     }
@@ -653,7 +655,7 @@
   // is done with a timeout so that we can detect problems.
 #if ART_USE_FUTEXES
   timespec wait_timeout;
-  InitTimeSpec(false, CLOCK_MONOTONIC, kIsDebugBuild ? 50000 : 10000, 0, &wait_timeout);
+  InitTimeSpec(false, CLOCK_MONOTONIC, NsToMs(thread_suspend_timeout_ns_), 0, &wait_timeout);
 #endif
   const uint64_t start_time = NanoTime();
   while (true) {
@@ -863,7 +865,7 @@
           return thread;
         }
         const uint64_t total_delay = NanoTime() - start_time;
-        if (total_delay >= MsToNs(kThreadSuspendTimeoutMs)) {
+        if (total_delay >= thread_suspend_timeout_ns_) {
           ThreadSuspendByPeerWarning(self,
                                      ::android::base::FATAL,
                                      "Thread suspension timed out",
@@ -969,7 +971,7 @@
           return thread;
         }
         const uint64_t total_delay = NanoTime() - start_time;
-        if (total_delay >= MsToNs(kThreadSuspendTimeoutMs)) {
+        if (total_delay >= thread_suspend_timeout_ns_) {
           ThreadSuspendByThreadIdWarning(::android::base::WARNING,
                                          "Thread suspension timed out",
                                          thread_id);
diff --git a/runtime/thread_list.h b/runtime/thread_list.h
index 658db00..b60fca1 100644
--- a/runtime/thread_list.h
+++ b/runtime/thread_list.h
@@ -20,6 +20,7 @@
 #include "barrier.h"
 #include "base/histogram.h"
 #include "base/mutex.h"
+#include "base/time_utils.h"
 #include "base/value_object.h"
 #include "gc_root.h"
 #include "jni.h"
@@ -41,11 +42,12 @@
 
 class ThreadList {
  public:
-  static const uint32_t kMaxThreadId = 0xFFFF;
-  static const uint32_t kInvalidThreadId = 0;
-  static const uint32_t kMainThreadId = 1;
+  static constexpr uint32_t kMaxThreadId = 0xFFFF;
+  static constexpr uint32_t kInvalidThreadId = 0;
+  static constexpr uint32_t kMainThreadId = 1;
+  static constexpr uint64_t kDefaultThreadSuspendTimeout = MsToNs(kIsDebugBuild ? 50000 : 10000);
 
-  explicit ThreadList();
+  explicit ThreadList(uint64_t thread_suspend_timeout_ns);
   ~ThreadList();
 
   void DumpForSigQuit(std::ostream& os)
@@ -219,6 +221,9 @@
   // Whether or not the current thread suspension is long.
   bool long_suspend_;
 
+  // Thread suspension timeout in nanoseconds.
+  const uint64_t thread_suspend_timeout_ns_;
+
   std::unique_ptr<Barrier> empty_checkpoint_barrier_;
 
   friend class Thread;
diff --git a/runtime/trace.cc b/runtime/trace.cc
index 9d9360e..2add955 100644
--- a/runtime/trace.cc
+++ b/runtime/trace.cc
@@ -54,6 +54,7 @@
     static_cast<size_t>(kTraceMethodActionMask));
 static constexpr uint8_t kOpNewMethod = 1U;
 static constexpr uint8_t kOpNewThread = 2U;
+static constexpr uint8_t kOpTraceSummary = 3U;
 
 class BuildStackTraceVisitor : public StackVisitor {
  public:
@@ -700,20 +701,19 @@
   std::string header(os.str());
 
   if (trace_output_mode_ == TraceOutputMode::kStreaming) {
-    File file(streaming_file_name_ + ".sec", O_CREAT | O_WRONLY, true);
-    if (!file.IsOpened()) {
-      LOG(WARNING) << "Could not open secondary trace file!";
-      return;
-    }
-    if (!file.WriteFully(header.c_str(), header.length())) {
-      file.Erase();
-      std::string detail(StringPrintf("Trace data write failed: %s", strerror(errno)));
-      PLOG(ERROR) << detail;
-      ThrowRuntimeException("%s", detail.c_str());
-    }
-    if (file.FlushCloseOrErase() != 0) {
-      PLOG(ERROR) << "Could not write secondary file";
-    }
+    MutexLock mu(Thread::Current(), *streaming_lock_);  // To serialize writing.
+    // Write a special token to mark the end of trace records and the start of
+    // trace summary.
+    uint8_t buf[7];
+    Append2LE(buf, 0);
+    buf[2] = kOpTraceSummary;
+    Append4LE(buf + 3, static_cast<uint32_t>(header.length()));
+    WriteToBuf(buf, sizeof(buf));
+    // Write the trace summary. The summary is identical to the file header when
+    // the output mode is not streaming (except for methods).
+    WriteToBuf(reinterpret_cast<const uint8_t*>(header.c_str()), header.length());
+    // Flush the buffer, which may include some trace records before the summary.
+    FlushBuf();
   } else {
     if (trace_file_.get() == nullptr) {
       iovec iov[2];
@@ -894,6 +894,14 @@
   memcpy(buf_.get() + old_offset, src, src_size);
 }
 
+void Trace::FlushBuf() {
+  int32_t offset = cur_offset_.LoadRelaxed();
+  if (!trace_file_->WriteFully(buf_.get(), offset)) {
+    PLOG(WARNING) << "Failed flush the remaining data in streaming.";
+  }
+  cur_offset_.StoreRelease(0);
+}
+
 void Trace::LogMethodTraceEvent(Thread* thread, ArtMethod* method,
                                 instrumentation::Instrumentation::InstrumentationEvent event,
                                 uint32_t thread_clock_diff, uint32_t wall_clock_diff) {
diff --git a/runtime/trace.h b/runtime/trace.h
index 824b150..485e9a1 100644
--- a/runtime/trace.h
+++ b/runtime/trace.h
@@ -202,7 +202,8 @@
       // This causes the negative annotations to incorrectly have a false positive. TODO: Figure out
       // how to annotate this.
       NO_THREAD_SAFETY_ANALYSIS;
-  void FinishTracing() REQUIRES_SHARED(Locks::mutator_lock_) REQUIRES(!*unique_methods_lock_);
+  void FinishTracing()
+      REQUIRES_SHARED(Locks::mutator_lock_) REQUIRES(!*unique_methods_lock_, !*streaming_lock_);
 
   void ReadClocks(Thread* thread, uint32_t* thread_clock_diff, uint32_t* wall_clock_diff);
 
@@ -229,6 +230,9 @@
   // annotation.
   void WriteToBuf(const uint8_t* src, size_t src_size)
       REQUIRES(streaming_lock_);
+  // Flush the main buffer to file. Used for streaming. Exposed here for lock annotation.
+  void FlushBuf()
+      REQUIRES(streaming_lock_);
 
   uint32_t EncodeTraceMethod(ArtMethod* method) REQUIRES(!*unique_methods_lock_);
   uint32_t EncodeTraceMethodAndAction(ArtMethod* method, TraceAction action)
diff --git a/test/596-monitor-inflation/expected.txt b/test/596-monitor-inflation/expected.txt
new file mode 100644
index 0000000..2add696
--- /dev/null
+++ b/test/596-monitor-inflation/expected.txt
@@ -0,0 +1,6 @@
+JNI_OnLoad called
+Monitor list grew by at least 4000 monitors
+Monitor list shrank correctly
+Finished first check
+Finished second check
+Total checks: 10000
diff --git a/test/596-monitor-inflation/info.txt b/test/596-monitor-inflation/info.txt
new file mode 100644
index 0000000..81dedb6
--- /dev/null
+++ b/test/596-monitor-inflation/info.txt
@@ -0,0 +1,5 @@
+A simple test that forces many monitors to be inflated, while checking
+that hashcodes are consistently maintained.
+
+This allocates more monitors and hence may exercise the monitor pool
+differently, and with more context, than the monitor_pool_test gtest.
diff --git a/test/596-monitor-inflation/monitor_inflation.cc b/test/596-monitor-inflation/monitor_inflation.cc
new file mode 100644
index 0000000..fb4275b
--- /dev/null
+++ b/test/596-monitor-inflation/monitor_inflation.cc
@@ -0,0 +1,35 @@
+/*
+ * Copyright (C) 2017 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.
+ */
+
+#include "gc/heap.h"
+#include "jni.h"
+#include "monitor.h"
+#include "runtime.h"
+#include "thread-inl.h"
+
+namespace art {
+namespace {
+
+extern "C" JNIEXPORT void JNICALL Java_Main_trim(JNIEnv*, jclass) {
+  Runtime::Current()->GetHeap()->Trim(Thread::Current());
+}
+
+extern "C" JNIEXPORT jint JNICALL Java_Main_monitorListSize(JNIEnv*, jclass) {
+  return Runtime::Current()->GetMonitorList()->Size();
+}
+
+}  // namespace
+}  // namespace art
diff --git a/test/596-monitor-inflation/src/Main.java b/test/596-monitor-inflation/src/Main.java
new file mode 100644
index 0000000..d97c766
--- /dev/null
+++ b/test/596-monitor-inflation/src/Main.java
@@ -0,0 +1,79 @@
+/*
+ * Copyright (C) 2016 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.
+ */
+import java.util.IdentityHashMap;
+import dalvik.system.VMRuntime;
+
+public class Main {
+  public static void main(String[] args) {
+    System.loadLibrary(args[0]);
+    int initialSize = monitorListSize();
+    IdentityHashMap<Object, Integer> all = new IdentityHashMap();
+    for (int i = 0; i < 5000; ++i) {
+      Object obj = new Object();
+      synchronized(obj) {
+        // Should force inflation.
+        all.put(obj, obj.hashCode());
+      }
+    }
+    // Since monitor deflation is delayed significantly, we believe that even with an intervening
+    // GC, monitors should remain inflated.  We allow some slop for unrelated concurrent runtime
+    // actions.
+    int inflatedSize = monitorListSize();
+    if (inflatedSize >= initialSize + 4000) {
+        System.out.println("Monitor list grew by at least 4000 monitors");
+    } else {
+        System.out.println("Monitor list did not grow as expected");
+    }
+    // Encourage monitor deflation.
+    // trim() (Heap::Trim()) deflates only in JANK_IMPERCEPTIBLE state.
+    // Some of this mirrors code in ActivityThread.java.
+    final int DALVIK_PROCESS_STATE_JANK_PERCEPTIBLE = 0;
+    final int DALVIK_PROCESS_STATE_JANK_IMPERCEPTIBLE = 1;
+    VMRuntime.getRuntime().updateProcessState(DALVIK_PROCESS_STATE_JANK_IMPERCEPTIBLE);
+    System.gc();
+    System.runFinalization();
+    trim();
+    VMRuntime.getRuntime().updateProcessState(DALVIK_PROCESS_STATE_JANK_PERCEPTIBLE);
+    int finalSize = monitorListSize();
+    if (finalSize > initialSize + 1000) {
+        System.out.println("Monitor list failed to shrink properly");
+    } else {
+        System.out.println("Monitor list shrank correctly");
+    }
+    int j = 0;
+    for (Object obj: all.keySet()) {
+      ++j;
+      if (obj.hashCode() != all.get(obj)) {
+        throw new AssertionError("Failed hashcode test!");
+      }
+    }
+    System.out.println("Finished first check");
+    for (Object obj: all.keySet()) {
+      ++j;
+      synchronized(obj) {
+        if (obj.hashCode() != all.get(obj)) {
+          throw new AssertionError("Failed hashcode test!");
+        }
+      }
+    }
+    System.out.println("Finished second check");
+    System.out.println("Total checks: " + j);
+  }
+
+  private static native void trim();
+
+  private static native int monitorListSize();
+}
diff --git a/test/901-hello-ti-agent/src/Main.java b/test/901-hello-ti-agent/src/Main.java
index faf2dc2..4d62ed3 100644
--- a/test/901-hello-ti-agent/src/Main.java
+++ b/test/901-hello-ti-agent/src/Main.java
@@ -16,8 +16,6 @@
 
 public class Main {
   public static void main(String[] args) {
-    System.loadLibrary(args[1]);
-
     System.out.println("Hello, world!");
 
     if (checkLivePhase()) {
diff --git a/test/902-hello-transformation/src/Main.java b/test/902-hello-transformation/src/Main.java
index ec47119..471c82b 100644
--- a/test/902-hello-transformation/src/Main.java
+++ b/test/902-hello-transformation/src/Main.java
@@ -49,7 +49,6 @@
     "AgAAABMCAAAAIAAAAQAAAB4CAAAAEAAAAQAAACwCAAA=");
 
   public static void main(String[] args) {
-    System.loadLibrary(args[1]);
     doTest(new Transform());
   }
 
diff --git a/test/903-hello-tagging/src/Main.java b/test/903-hello-tagging/src/Main.java
index a8aedb4..2f0365a 100644
--- a/test/903-hello-tagging/src/Main.java
+++ b/test/903-hello-tagging/src/Main.java
@@ -20,7 +20,6 @@
 
 public class Main {
   public static void main(String[] args) {
-    System.loadLibrary(args[1]);
     doTest();
     testGetTaggedObjects();
   }
diff --git a/test/904-object-allocation/src/Main.java b/test/904-object-allocation/src/Main.java
index fc8a112..df59179 100644
--- a/test/904-object-allocation/src/Main.java
+++ b/test/904-object-allocation/src/Main.java
@@ -18,8 +18,6 @@
 
 public class Main {
   public static void main(String[] args) throws Exception {
-    System.loadLibrary(args[1]);
-
     // Use a list to ensure objects must be allocated.
     ArrayList<Object> l = new ArrayList<>(100);
 
diff --git a/test/905-object-free/src/Main.java b/test/905-object-free/src/Main.java
index 16dec5d..e41e378 100644
--- a/test/905-object-free/src/Main.java
+++ b/test/905-object-free/src/Main.java
@@ -19,8 +19,6 @@
 
 public class Main {
   public static void main(String[] args) throws Exception {
-    System.loadLibrary(args[1]);
-
     doTest();
   }
 
diff --git a/test/906-iterate-heap/src/Main.java b/test/906-iterate-heap/src/Main.java
index 544a365..cab27be 100644
--- a/test/906-iterate-heap/src/Main.java
+++ b/test/906-iterate-heap/src/Main.java
@@ -19,8 +19,6 @@
 
 public class Main {
   public static void main(String[] args) throws Exception {
-    System.loadLibrary(args[1]);
-
     doTest();
   }
 
diff --git a/test/907-get-loaded-classes/src/Main.java b/test/907-get-loaded-classes/src/Main.java
index 468d037..370185a 100644
--- a/test/907-get-loaded-classes/src/Main.java
+++ b/test/907-get-loaded-classes/src/Main.java
@@ -20,8 +20,6 @@
 
 public class Main {
   public static void main(String[] args) throws Exception {
-    System.loadLibrary(args[1]);
-
     doTest();
   }
 
diff --git a/test/908-gc-start-finish/src/Main.java b/test/908-gc-start-finish/src/Main.java
index 2be0eea..05388c9 100644
--- a/test/908-gc-start-finish/src/Main.java
+++ b/test/908-gc-start-finish/src/Main.java
@@ -18,8 +18,6 @@
 
 public class Main {
   public static void main(String[] args) throws Exception {
-    System.loadLibrary(args[1]);
-
     doTest();
   }
 
diff --git a/test/910-methods/src/Main.java b/test/910-methods/src/Main.java
index bf25a0d..932a1ea 100644
--- a/test/910-methods/src/Main.java
+++ b/test/910-methods/src/Main.java
@@ -20,8 +20,6 @@
 
 public class Main {
   public static void main(String[] args) throws Exception {
-    System.loadLibrary(args[1]);
-
     doTest();
   }
 
diff --git a/test/911-get-stack-trace/expected.txt b/test/911-get-stack-trace/expected.txt
index dad08c9..2687f85 100644
--- a/test/911-get-stack-trace/expected.txt
+++ b/test/911-get-stack-trace/expected.txt
@@ -22,7 +22,7 @@
  bar (IIILControlData;)J 0 24
  foo (IIILControlData;)I 0 19
  doTest ()V 38 23
- main ([Ljava/lang/String;)V 6 21
+ main ([Ljava/lang/String;)V 3 21
 ---------
  print (Ljava/lang/Thread;II)V 0 34
  printOrWait (IILControlData;)V 6 39
@@ -42,7 +42,7 @@
  bar (IIILControlData;)J 0 24
  foo (IIILControlData;)I 0 19
  doTest ()V 42 24
- main ([Ljava/lang/String;)V 6 21
+ main ([Ljava/lang/String;)V 3 21
 ---------
  getStackTrace (Ljava/lang/Thread;II)[[Ljava/lang/String; -1 -2
  print (Ljava/lang/Thread;II)V 0 34
@@ -57,13 +57,13 @@
  baz (IIILControlData;)Ljava/lang/Object; 9 32
 From bottom
 ---------
- main ([Ljava/lang/String;)V 6 21
+ main ([Ljava/lang/String;)V 3 21
 ---------
  baz (IIILControlData;)Ljava/lang/Object; 9 32
  bar (IIILControlData;)J 0 24
  foo (IIILControlData;)I 0 19
  doTest ()V 65 30
- main ([Ljava/lang/String;)V 6 21
+ main ([Ljava/lang/String;)V 3 21
 ---------
  bar (IIILControlData;)J 0 24
  foo (IIILControlData;)I 0 19
@@ -358,7 +358,7 @@
  getAllStackTraces (I)[[Ljava/lang/Object; -1 -2
  printAll (I)V 0 73
  doTest ()V 102 57
- main ([Ljava/lang/String;)V 30 33
+ main ([Ljava/lang/String;)V 27 33
 
 ---------
 FinalizerDaemon
@@ -590,7 +590,7 @@
  getAllStackTraces (I)[[Ljava/lang/Object; -1 -2
  printAll (I)V 0 73
  doTest ()V 107 59
- main ([Ljava/lang/String;)V 30 33
+ main ([Ljava/lang/String;)V 27 33
 
 
 ########################################
@@ -659,7 +659,7 @@
  getThreadListStackTraces ([Ljava/lang/Thread;I)[[Ljava/lang/Object; -1 -2
  printList ([Ljava/lang/Thread;I)V 0 66
  doTest ()V 96 52
- main ([Ljava/lang/String;)V 38 37
+ main ([Ljava/lang/String;)V 35 37
 
 ---------
 Thread-14
@@ -771,7 +771,7 @@
  getThreadListStackTraces ([Ljava/lang/Thread;I)[[Ljava/lang/Object; -1 -2
  printList ([Ljava/lang/Thread;I)V 0 66
  doTest ()V 101 54
- main ([Ljava/lang/String;)V 38 37
+ main ([Ljava/lang/String;)V 35 37
 
 
 ###################
@@ -782,7 +782,7 @@
 [public static native java.lang.Object[] Frames.getFrameLocation(java.lang.Thread,int), ffffffff]
 [public static void Frames.doTestSameThread(), 38]
 [public static void Frames.doTest() throws java.lang.Exception, 0]
-[public static void Main.main(java.lang.String[]) throws java.lang.Exception, 2e]
+[public static void Main.main(java.lang.String[]) throws java.lang.Exception, 2b]
 JVMTI_ERROR_NO_MORE_FRAMES
 
 ################################
diff --git a/test/911-get-stack-trace/src/Main.java b/test/911-get-stack-trace/src/Main.java
index b199033..96a427d 100644
--- a/test/911-get-stack-trace/src/Main.java
+++ b/test/911-get-stack-trace/src/Main.java
@@ -16,7 +16,7 @@
 
 public class Main {
   public static void main(String[] args) throws Exception {
-    System.loadLibrary(args[1]);
+    bindTest911Classes();
 
     SameThread.doTest();
 
@@ -42,4 +42,6 @@
 
     System.out.println("Done");
   }
+
+  private static native void bindTest911Classes();
 }
diff --git a/test/911-get-stack-trace/stack_trace.cc b/test/911-get-stack-trace/stack_trace.cc
index d162e8a..68f6d8d 100644
--- a/test/911-get-stack-trace/stack_trace.cc
+++ b/test/911-get-stack-trace/stack_trace.cc
@@ -34,6 +34,14 @@
 
 using android::base::StringPrintf;
 
+extern "C" JNIEXPORT void JNICALL Java_Main_bindTest911Classes(
+    JNIEnv* env, jclass klass ATTRIBUTE_UNUSED) {
+  BindFunctions(jvmti_env, env, "AllTraces");
+  BindFunctions(jvmti_env, env, "Frames");
+  BindFunctions(jvmti_env, env, "PrintThread");
+  BindFunctions(jvmti_env, env, "ThreadListTraces");
+}
+
 static jint FindLineNumber(jint line_number_count,
                            jvmtiLineNumberEntry* line_number_table,
                            jlocation location) {
diff --git a/test/912-classes/classes.cc b/test/912-classes/classes.cc
index 29eeff6..d13436e 100644
--- a/test/912-classes/classes.cc
+++ b/test/912-classes/classes.cc
@@ -20,6 +20,7 @@
 #include "jni.h"
 #include "openjdkjvmti/jvmti.h"
 #include "ScopedLocalRef.h"
+#include "thread-inl.h"
 
 #include "ti-agent/common_helper.h"
 #include "ti-agent/common_load.h"
@@ -259,5 +260,120 @@
   return int_array;
 }
 
+static std::string GetClassName(jvmtiEnv* jenv, JNIEnv* jni_env, jclass klass) {
+  char* name;
+  jvmtiError result = jenv->GetClassSignature(klass, &name, nullptr);
+  if (result != JVMTI_ERROR_NONE) {
+    if (jni_env != nullptr) {
+      JvmtiErrorToException(jni_env, result);
+    } else {
+      printf("Failed to get class signature.\n");
+    }
+    return "";
+  }
+
+  std::string tmp(name);
+  jenv->Deallocate(reinterpret_cast<unsigned char*>(name));
+
+  return tmp;
+}
+
+static std::string GetThreadName(jvmtiEnv* jenv, JNIEnv* jni_env, jthread thread) {
+  jvmtiThreadInfo info;
+  jvmtiError result = jenv->GetThreadInfo(thread, &info);
+  if (result != JVMTI_ERROR_NONE) {
+    if (jni_env != nullptr) {
+      JvmtiErrorToException(jni_env, result);
+    } else {
+      printf("Failed to get thread name.\n");
+    }
+    return "";
+  }
+
+  std::string tmp(info.name);
+  jenv->Deallocate(reinterpret_cast<unsigned char*>(info.name));
+  jni_env->DeleteLocalRef(info.context_class_loader);
+  jni_env->DeleteLocalRef(info.thread_group);
+
+  return tmp;
+}
+
+static std::string GetThreadName(Thread* thread) {
+  std::string tmp;
+  thread->GetThreadName(tmp);
+  return tmp;
+}
+
+static void JNICALL ClassPrepareCallback(jvmtiEnv* jenv,
+                                         JNIEnv* jni_env,
+                                         jthread thread,
+                                         jclass klass) {
+  std::string name = GetClassName(jenv, jni_env, klass);
+  if (name == "") {
+    return;
+  }
+  std::string thread_name = GetThreadName(jenv, jni_env, thread);
+  if (thread_name == "") {
+    return;
+  }
+  std::string cur_thread_name = GetThreadName(Thread::Current());
+  printf("Prepare: %s on %s (cur=%s)\n",
+         name.c_str(),
+         thread_name.c_str(),
+         cur_thread_name.c_str());
+}
+
+static void JNICALL ClassLoadCallback(jvmtiEnv* jenv,
+                                      JNIEnv* jni_env,
+                                      jthread thread,
+                                      jclass klass) {
+  std::string name = GetClassName(jenv, jni_env, klass);
+  if (name == "") {
+    return;
+  }
+  std::string thread_name = GetThreadName(jenv, jni_env, thread);
+  if (thread_name == "") {
+    return;
+  }
+  printf("Load: %s on %s\n", name.c_str(), thread_name.c_str());
+}
+
+extern "C" JNIEXPORT void JNICALL Java_Main_enableClassLoadEvents(
+    JNIEnv* env, jclass Main_klass ATTRIBUTE_UNUSED, jboolean b) {
+  if (b == JNI_FALSE) {
+    jvmtiError ret = jvmti_env->SetEventNotificationMode(JVMTI_DISABLE,
+                                                         JVMTI_EVENT_CLASS_LOAD,
+                                                         nullptr);
+    if (JvmtiErrorToException(env, ret)) {
+      return;
+    }
+    ret = jvmti_env->SetEventNotificationMode(JVMTI_DISABLE,
+                                              JVMTI_EVENT_CLASS_PREPARE,
+                                              nullptr);
+    JvmtiErrorToException(env, ret);
+    return;
+  }
+
+  jvmtiEventCallbacks callbacks;
+  memset(&callbacks, 0, sizeof(jvmtiEventCallbacks));
+  callbacks.ClassLoad = ClassLoadCallback;
+  callbacks.ClassPrepare = ClassPrepareCallback;
+  jvmtiError ret = jvmti_env->SetEventCallbacks(&callbacks, sizeof(callbacks));
+  if (JvmtiErrorToException(env, ret)) {
+    return;
+  }
+
+  ret = jvmti_env->SetEventNotificationMode(JVMTI_ENABLE,
+                                            JVMTI_EVENT_CLASS_LOAD,
+                                            nullptr);
+  if (JvmtiErrorToException(env, ret)) {
+    return;
+  }
+  ret = jvmti_env->SetEventNotificationMode(JVMTI_ENABLE,
+                                            JVMTI_EVENT_CLASS_PREPARE,
+                                            nullptr);
+  JvmtiErrorToException(env, ret);
+}
+
 }  // namespace Test912Classes
 }  // namespace art
diff --git a/test/912-classes/expected.txt b/test/912-classes/expected.txt
index f3cb261..4976553 100644
--- a/test/912-classes/expected.txt
+++ b/test/912-classes/expected.txt
@@ -61,3 +61,35 @@
 [class A, class B, class java.lang.Object]
 
 [37, 0]
+
+B, false
+Load: LB; on main
+Prepare: LB; on main (cur=main)
+B, true
+Load: LB; on main
+Prepare: LB; on main (cur=main)
+C, false
+Load: LA; on main
+Prepare: LA; on main (cur=main)
+Load: LC; on main
+Prepare: LC; on main (cur=main)
+A, false
+C, true
+Load: LA; on main
+Prepare: LA; on main (cur=main)
+Load: LC; on main
+Prepare: LC; on main (cur=main)
+A, true
+A, true
+Load: LA; on main
+Prepare: LA; on main (cur=main)
+C, true
+Load: LC; on main
+Prepare: LC; on main (cur=main)
+Load: LMain$1; on main
+Prepare: LMain$1; on main (cur=main)
+C, true
+Load: LA; on TestRunner
+Prepare: LA; on TestRunner (cur=TestRunner)
+Load: LC; on TestRunner
+Prepare: LC; on TestRunner (cur=TestRunner)
diff --git a/test/912-classes/src-ex/A.java b/test/912-classes/src-ex/A.java
index 64acb2f..2c43cfb 100644
--- a/test/912-classes/src-ex/A.java
+++ b/test/912-classes/src-ex/A.java
@@ -15,4 +15,4 @@
  */
 
 public class A {
-}
\ No newline at end of file
+}
diff --git a/test/912-classes/src-ex/C.java b/test/912-classes/src-ex/C.java
new file mode 100644
index 0000000..97f8021
--- /dev/null
+++ b/test/912-classes/src-ex/C.java
@@ -0,0 +1,18 @@
+/*
+ * Copyright (C) 2017 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.
+ */
+
+public class C extends A {
+}
diff --git a/test/912-classes/src/B.java b/test/912-classes/src/B.java
index f1458c3..52ce4dd 100644
--- a/test/912-classes/src/B.java
+++ b/test/912-classes/src/B.java
@@ -15,4 +15,4 @@
  */
 
 public class B {
-}
\ No newline at end of file
+}
diff --git a/test/912-classes/src/Main.java b/test/912-classes/src/Main.java
index cbf2392..859f80c 100644
--- a/test/912-classes/src/Main.java
+++ b/test/912-classes/src/Main.java
@@ -21,8 +21,6 @@
 
 public class Main {
   public static void main(String[] args) throws Exception {
-    System.loadLibrary(args[1]);
-
     doTest();
   }
 
@@ -84,6 +82,10 @@
     System.out.println();
 
     testClassVersion();
+
+    System.out.println();
+
+    testClassEvents();
   }
 
   private static Class<?> proxyClass = null;
@@ -210,6 +212,60 @@
     System.out.println(Arrays.toString(getClassVersion(Main.class)));
   }
 
+  private static void testClassEvents() throws Exception {
+    ClassLoader cl = Main.class.getClassLoader();
+    while (cl.getParent() != null) {
+      cl = cl.getParent();
+    }
+    final ClassLoader boot = cl;
+
+    enableClassLoadEvents(true);
+
+    ClassLoader cl1 = create(boot, DEX1, DEX2);
+    System.out.println("B, false");
+    Class.forName("B", false, cl1);
+
+    ClassLoader cl2 = create(boot, DEX1, DEX2);
+    System.out.println("B, true");
+    Class.forName("B", true, cl2);
+
+    ClassLoader cl3 = create(boot, DEX1, DEX2);
+    System.out.println("C, false");
+    Class.forName("C", false, cl3);
+    System.out.println("A, false");
+    Class.forName("A", false, cl3);
+
+    ClassLoader cl4 = create(boot, DEX1, DEX2);
+    System.out.println("C, true");
+    Class.forName("C", true, cl4);
+    System.out.println("A, true");
+    Class.forName("A", true, cl4);
+
+    ClassLoader cl5 = create(boot, DEX1, DEX2);
+    System.out.println("A, true");
+    Class.forName("A", true, cl5);
+    System.out.println("C, true");
+    Class.forName("C", true, cl5);
+
+    Runnable r = new Runnable() {
+      @Override
+      public void run() {
+        try {
+          ClassLoader cl6 = create(boot, DEX1, DEX2);
+          System.out.println("C, true");
+          Class.forName("C", true, cl6);
+        } catch (Exception e) {
+          throw new RuntimeException(e);
+        }
+      }
+    };
+    Thread t = new Thread(r, "TestRunner");
+    t.start();
+    t.join();
+
+    enableClassLoadEvents(false);
+  }
+
   private static void printClassLoaderClasses(ClassLoader cl) {
     for (;;) {
       if (cl == null || !cl.getClass().getName().startsWith("dalvik.system")) {
@@ -272,6 +328,8 @@
 
   private static native int[] getClassVersion(Class<?> c);
 
+  private static native void enableClassLoadEvents(boolean b);
+
   private static class TestForNonInit {
     public static double dummy = Math.random();  // So it can't be compile-time initialized.
   }
diff --git a/test/913-heaps/src/Main.java b/test/913-heaps/src/Main.java
index 564596e..5a11a5b 100644
--- a/test/913-heaps/src/Main.java
+++ b/test/913-heaps/src/Main.java
@@ -21,8 +21,6 @@
 
 public class Main {
   public static void main(String[] args) throws Exception {
-    System.loadLibrary(args[1]);
-
     doTest();
     doFollowReferencesTest();
   }
diff --git a/test/914-hello-obsolescence/src/Main.java b/test/914-hello-obsolescence/src/Main.java
index 46266ef..8a14716 100644
--- a/test/914-hello-obsolescence/src/Main.java
+++ b/test/914-hello-obsolescence/src/Main.java
@@ -53,7 +53,6 @@
     "AAACIAAAEQAAAKIBAAADIAAAAgAAAJECAAAAIAAAAQAAAJ8CAAAAEAAAAQAAALACAAA=");
 
   public static void main(String[] args) {
-    System.loadLibrary(args[1]);
     doTest(new Transform());
   }
 
diff --git a/test/915-obsolete-2/src/Main.java b/test/915-obsolete-2/src/Main.java
index bbeb726..0e3145c 100644
--- a/test/915-obsolete-2/src/Main.java
+++ b/test/915-obsolete-2/src/Main.java
@@ -79,7 +79,6 @@
     "IAAAFwAAAD4CAAADIAAABAAAAAgEAAAAIAAAAQAAACYEAAAAEAAAAQAAADwEAAA=");
 
   public static void main(String[] args) {
-    System.loadLibrary(args[1]);
     doTest(new Transform());
   }
 
diff --git a/test/916-obsolete-jit/src/Main.java b/test/916-obsolete-jit/src/Main.java
index 1e43f7e..1b03200 100644
--- a/test/916-obsolete-jit/src/Main.java
+++ b/test/916-obsolete-jit/src/Main.java
@@ -113,7 +113,6 @@
   }
 
   public static void main(String[] args) {
-    System.loadLibrary(args[1]);
     doTest(new Transform(), new TestWatcher());
   }
 
diff --git a/test/917-fields-transformation/src/Main.java b/test/917-fields-transformation/src/Main.java
index 5378bb7..632a5c8 100644
--- a/test/917-fields-transformation/src/Main.java
+++ b/test/917-fields-transformation/src/Main.java
@@ -55,7 +55,6 @@
     "AAIgAAAMAAAAXAEAAAMgAAACAAAA4QEAAAAgAAABAAAA8AEAAAAQAAABAAAABAIAAA==");
 
   public static void main(String[] args) {
-    System.loadLibrary(args[1]);
     doTest(new Transform("Hello", "Goodbye"),
            new Transform("start", "end"));
   }
diff --git a/test/918-fields/src/Main.java b/test/918-fields/src/Main.java
index 8af6e7b..3ba535b 100644
--- a/test/918-fields/src/Main.java
+++ b/test/918-fields/src/Main.java
@@ -19,8 +19,6 @@
 
 public class Main {
   public static void main(String[] args) throws Exception {
-    System.loadLibrary(args[1]);
-
     doTest();
   }
 
diff --git a/test/919-obsolete-fields/src/Main.java b/test/919-obsolete-fields/src/Main.java
index 895c7a3..1d893f1 100644
--- a/test/919-obsolete-fields/src/Main.java
+++ b/test/919-obsolete-fields/src/Main.java
@@ -116,7 +116,6 @@
   }
 
   public static void main(String[] args) {
-    System.loadLibrary(args[1]);
     TestWatcher w = new TestWatcher();
     doTest(new Transform(w), w);
   }
diff --git a/test/921-hello-failure/src/Main.java b/test/921-hello-failure/src/Main.java
index 43d6e9e..67ca1e1 100644
--- a/test/921-hello-failure/src/Main.java
+++ b/test/921-hello-failure/src/Main.java
@@ -18,7 +18,6 @@
 public class Main {
 
   public static void main(String[] args) {
-    System.loadLibrary(args[1]);
     NewName.doTest(new Transform());
     DifferentAccess.doTest(new Transform());
     NewInterface.doTest(new Transform2());
diff --git a/test/922-properties/src/Main.java b/test/922-properties/src/Main.java
index 6cec6e9..8ad742f 100644
--- a/test/922-properties/src/Main.java
+++ b/test/922-properties/src/Main.java
@@ -19,8 +19,6 @@
 
 public class Main {
   public static void main(String[] args) throws Exception {
-    System.loadLibrary(args[1]);
-
     doTest();
   }
 
diff --git a/test/923-monitors/src/Main.java b/test/923-monitors/src/Main.java
index e35ce12..ef00728 100644
--- a/test/923-monitors/src/Main.java
+++ b/test/923-monitors/src/Main.java
@@ -21,8 +21,6 @@
 
 public class Main {
   public static void main(String[] args) throws Exception {
-    System.loadLibrary(args[1]);
-
     doTest();
   }
 
diff --git a/test/924-threads/expected.txt b/test/924-threads/expected.txt
index 3b7fb24..67d20eb 100644
--- a/test/924-threads/expected.txt
+++ b/test/924-threads/expected.txt
@@ -31,3 +31,7 @@
 [Thread[FinalizerDaemon,5,system], Thread[FinalizerWatchdogDaemon,5,system], Thread[HeapTaskDaemon,5,system], Thread[ReferenceQueueDaemon,5,system], Thread[Signal Catcher,5,system], Thread[main,5,main]]
 JVMTI_ERROR_THREAD_NOT_ALIVE
 JVMTI_ERROR_THREAD_NOT_ALIVE
+Constructed thread
+Thread(EventTestThread): start
+Thread(EventTestThread): end
+Thread joined
diff --git a/test/924-threads/src/Main.java b/test/924-threads/src/Main.java
index dec49a8..29c4aa3 100644
--- a/test/924-threads/src/Main.java
+++ b/test/924-threads/src/Main.java
@@ -25,8 +25,6 @@
 
 public class Main {
   public static void main(String[] args) throws Exception {
-    System.loadLibrary(args[1]);
-
     doTest();
   }
 
@@ -58,6 +56,8 @@
     doAllThreadsTests();
 
     doTLSTests();
+
+    doTestEvents();
   }
 
   private static class Holder {
@@ -228,6 +228,22 @@
     }
   }
 
+  private static void doTestEvents() throws Exception {
+    enableThreadEvents(true);
+
+    Thread t = new Thread("EventTestThread");
+
+    System.out.println("Constructed thread");
+    Thread.yield();
+
+    t.start();
+    t.join();
+
+    System.out.println("Thread joined");
+
+    enableThreadEvents(false);
+  }
+
   private final static Comparator<Thread> THREAD_COMP = new Comparator<Thread>() {
     public int compare(Thread o1, Thread o2) {
       return o1.getName().compareTo(o2.getName());
@@ -295,4 +311,5 @@
   private static native Thread[] getAllThreads();
   private static native void setTLS(Thread t, long l);
   private static native long getTLS(Thread t);
+  private static native void enableThreadEvents(boolean b);
 }
diff --git a/test/924-threads/threads.cc b/test/924-threads/threads.cc
index d35eaa8..0380433 100644
--- a/test/924-threads/threads.cc
+++ b/test/924-threads/threads.cc
@@ -137,5 +137,71 @@
   JvmtiErrorToException(env, result);
 }
 
+static void JNICALL ThreadEvent(jvmtiEnv* jvmti_env,
+                                JNIEnv* jni_env,
+                                jthread thread,
+                                bool is_start) {
+  jvmtiThreadInfo info;
+  jvmtiError result = jvmti_env->GetThreadInfo(thread, &info);
+  if (result != JVMTI_ERROR_NONE) {
+    printf("Error getting thread info");
+    return;
+  }
+  printf("Thread(%s): %s\n", info.name, is_start ? "start" : "end");
+
+  jvmti_env->Deallocate(reinterpret_cast<unsigned char*>(info.name));
+  jni_env->DeleteLocalRef(info.thread_group);
+  jni_env->DeleteLocalRef(info.context_class_loader);
+}
+
+static void JNICALL ThreadStart(jvmtiEnv* jvmti_env,
+                                JNIEnv* jni_env,
+                                jthread thread) {
+  ThreadEvent(jvmti_env, jni_env, thread, true);
+}
+
+static void JNICALL ThreadEnd(jvmtiEnv* jvmti_env,
+                              JNIEnv* jni_env,
+                              jthread thread) {
+  ThreadEvent(jvmti_env, jni_env, thread, false);
+}
+
+extern "C" JNIEXPORT void JNICALL Java_Main_enableThreadEvents(
+    JNIEnv* env, jclass Main_klass ATTRIBUTE_UNUSED, jboolean b) {
+  if (b == JNI_FALSE) {
+    jvmtiError ret = jvmti_env->SetEventNotificationMode(JVMTI_DISABLE,
+                                                         JVMTI_EVENT_THREAD_START,
+                                                         nullptr);
+    if (JvmtiErrorToException(env, ret)) {
+      return;
+    }
+    ret = jvmti_env->SetEventNotificationMode(JVMTI_DISABLE,
+                                              JVMTI_EVENT_THREAD_END,
+                                              nullptr);
+    JvmtiErrorToException(env, ret);
+    return;
+  }
+
+  jvmtiEventCallbacks callbacks;
+  memset(&callbacks, 0, sizeof(jvmtiEventCallbacks));
+  callbacks.ThreadStart = ThreadStart;
+  callbacks.ThreadEnd = ThreadEnd;
+  jvmtiError ret = jvmti_env->SetEventCallbacks(&callbacks, sizeof(callbacks));
+  if (JvmtiErrorToException(env, ret)) {
+    return;
+  }
+
+  ret = jvmti_env->SetEventNotificationMode(JVMTI_ENABLE,
+                                            JVMTI_EVENT_THREAD_START,
+                                            nullptr);
+  if (JvmtiErrorToException(env, ret)) {
+    return;
+  }
+  ret = jvmti_env->SetEventNotificationMode(JVMTI_ENABLE,
+                                            JVMTI_EVENT_THREAD_END,
+                                            nullptr);
+  JvmtiErrorToException(env, ret);
+}
+
 }  // namespace Test924Threads
 }  // namespace art
diff --git a/test/925-threadgroups/src/Main.java b/test/925-threadgroups/src/Main.java
index c59efe2..3d7a4ca 100644
--- a/test/925-threadgroups/src/Main.java
+++ b/test/925-threadgroups/src/Main.java
@@ -19,8 +19,6 @@
 
 public class Main {
   public static void main(String[] args) throws Exception {
-    System.loadLibrary(args[1]);
-
     doTest();
   }
 
diff --git a/test/926-multi-obsolescence/src/Main.java b/test/926-multi-obsolescence/src/Main.java
index 8a6cf84..6d9f96c 100644
--- a/test/926-multi-obsolescence/src/Main.java
+++ b/test/926-multi-obsolescence/src/Main.java
@@ -92,7 +92,6 @@
         "AACUAQAAAiAAABEAAACiAQAAAyAAAAIAAACXAgAAACAAAAEAAAClAgAAABAAAAEAAAC0AgAA"));
 
   public static void main(String[] args) {
-    System.loadLibrary(args[1]);
     doTest(new Transform(), new Transform2());
   }
 
diff --git a/test/927-timers/src/Main.java b/test/927-timers/src/Main.java
index 2f5c85c..b67f66d 100644
--- a/test/927-timers/src/Main.java
+++ b/test/927-timers/src/Main.java
@@ -18,8 +18,6 @@
 
 public class Main {
   public static void main(String[] args) throws Exception {
-    System.loadLibrary(args[1]);
-
     doTest();
   }
 
diff --git a/test/928-jni-table/src/Main.java b/test/928-jni-table/src/Main.java
index b0baea1..fd61b7d 100644
--- a/test/928-jni-table/src/Main.java
+++ b/test/928-jni-table/src/Main.java
@@ -16,8 +16,6 @@
 
 public class Main {
   public static void main(String[] args) throws Exception {
-    System.loadLibrary(args[1]);
-
     doJNITableTest();
 
     System.out.println("Done");
diff --git a/test/929-search/src/Main.java b/test/929-search/src/Main.java
index d253e6f..bbeb081 100644
--- a/test/929-search/src/Main.java
+++ b/test/929-search/src/Main.java
@@ -18,8 +18,6 @@
 
 public class Main {
   public static void main(String[] args) throws Exception {
-    System.loadLibrary(args[1]);
-
     doTest();
   }
 
diff --git a/test/930-hello-retransform/src/Main.java b/test/930-hello-retransform/src/Main.java
index 12194c3..0063c82 100644
--- a/test/930-hello-retransform/src/Main.java
+++ b/test/930-hello-retransform/src/Main.java
@@ -49,7 +49,6 @@
     "AgAAABMCAAAAIAAAAQAAAB4CAAAAEAAAAQAAACwCAAA=");
 
   public static void main(String[] args) {
-    System.loadLibrary(args[1]);
     doTest(new Transform());
   }
 
diff --git a/test/931-agent-thread/src/Main.java b/test/931-agent-thread/src/Main.java
index 6471bc8..a7639fb 100644
--- a/test/931-agent-thread/src/Main.java
+++ b/test/931-agent-thread/src/Main.java
@@ -18,8 +18,6 @@
 
 public class Main {
   public static void main(String[] args) throws Exception {
-    System.loadLibrary(args[1]);
-
     testAgentThread();
 
     System.out.println("Done");
diff --git a/test/932-transform-saves/src/Main.java b/test/932-transform-saves/src/Main.java
index d98ba6d..d960322 100644
--- a/test/932-transform-saves/src/Main.java
+++ b/test/932-transform-saves/src/Main.java
@@ -79,7 +79,6 @@
     "AgAAABMCAAAAIAAAAQAAAB4CAAAAEAAAAQAAACwCAAA=");
 
   public static void main(String[] args) {
-    System.loadLibrary(args[1]);
     doTest(new Transform());
   }
 
diff --git a/test/933-misc-events/build b/test/933-misc-events/build
new file mode 100755
index 0000000..898e2e5
--- /dev/null
+++ b/test/933-misc-events/build
@@ -0,0 +1,17 @@
+#!/bin/bash
+#
+# Copyright 2016 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.
+
+./default-build "$@" --experimental agents
diff --git a/test/933-misc-events/expected.txt b/test/933-misc-events/expected.txt
new file mode 100644
index 0000000..024c560
--- /dev/null
+++ b/test/933-misc-events/expected.txt
@@ -0,0 +1,2 @@
+Received dump request.
+Done
diff --git a/test/933-misc-events/info.txt b/test/933-misc-events/info.txt
new file mode 100644
index 0000000..875a5f6
--- /dev/null
+++ b/test/933-misc-events/info.txt
@@ -0,0 +1 @@
+Tests basic functions in the jvmti plugin.
diff --git a/test/933-misc-events/misc_events.cc b/test/933-misc-events/misc_events.cc
new file mode 100644
index 0000000..860d4b5
--- /dev/null
+++ b/test/933-misc-events/misc_events.cc
@@ -0,0 +1,72 @@
+/*
+ * Copyright (C) 2017 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.
+ */
+
+#include <atomic>
+#include <signal.h>
+#include <sys/types.h>
+
+#include "base/logging.h"
+#include "base/macros.h"
+#include "jni.h"
+#include "openjdkjvmti/jvmti.h"
+
+#include "ti-agent/common_helper.h"
+#include "ti-agent/common_load.h"
+
+namespace art {
+namespace Test933MiscEvents {
+
+static std::atomic<bool> saw_dump_request(false);
+
+static void DumpRequestCallback(jvmtiEnv* jenv ATTRIBUTE_UNUSED) {
+  printf("Received dump request.\n");
+  saw_dump_request.store(true, std::memory_order::memory_order_relaxed);
+}
+
+extern "C" JNIEXPORT void JNICALL Java_Main_testSigQuit(
+    JNIEnv* env, jclass Main_klass ATTRIBUTE_UNUSED) {
+  jvmtiEventCallbacks callbacks;
+  memset(&callbacks, 0, sizeof(jvmtiEventCallbacks));
+  callbacks.DataDumpRequest = DumpRequestCallback;
+  jvmtiError ret = jvmti_env->SetEventCallbacks(&callbacks, sizeof(callbacks));
+  if (JvmtiErrorToException(env, ret)) {
+    return;
+  }
+
+  ret = jvmti_env->SetEventNotificationMode(JVMTI_ENABLE,
+                                            JVMTI_EVENT_DATA_DUMP_REQUEST,
+                                            nullptr);
+  if (JvmtiErrorToException(env, ret)) {
+    return;
+  }
+
+  // Send sigquit to self.
+  kill(getpid(), SIGQUIT);
+
+  // Busy-wait for request.
+  for (;;) {
+    sleep(1);
+    if (saw_dump_request.load(std::memory_order::memory_order_relaxed)) {
+      break;
+    }
+  }
+
+  ret = jvmti_env->SetEventNotificationMode(JVMTI_DISABLE, JVMTI_EVENT_DATA_DUMP_REQUEST, nullptr);
+  JvmtiErrorToException(env, ret);
+}
+
+}  // namespace Test933MiscEvents
+}  // namespace art
diff --git a/test/933-misc-events/run b/test/933-misc-events/run
new file mode 100755
index 0000000..0a8d067
--- /dev/null
+++ b/test/933-misc-events/run
@@ -0,0 +1,23 @@
+#!/bin/bash
+#
+# Copyright 2016 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.
+
+# This test checks whether dex files can be injected into parent classloaders. App images preload
+# classes, which will make the injection moot. Turn off app images to avoid the issue.
+
+./default-run "$@" --experimental agents \
+                   --experimental runtime-plugins \
+                   --jvmti \
+                   --no-app-image
diff --git a/test/933-misc-events/src/Main.java b/test/933-misc-events/src/Main.java
new file mode 100644
index 0000000..89801a3
--- /dev/null
+++ b/test/933-misc-events/src/Main.java
@@ -0,0 +1,25 @@
+/*
+ * Copyright (C) 2017 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.
+ */
+
+public class Main {
+  public static void main(String[] args) throws Exception {
+    testSigQuit();
+
+    System.out.println("Done");
+  }
+
+  private static native void testSigQuit();
+}
diff --git a/test/Android.bp b/test/Android.bp
index be5bc59..89e4092 100644
--- a/test/Android.bp
+++ b/test/Android.bp
@@ -270,6 +270,7 @@
         "928-jni-table/jni_table.cc",
         "929-search/search.cc",
         "931-agent-thread/agent_thread.cc",
+        "933-misc-events/misc_events.cc",
     ],
     shared_libs: [
         "libbase",
@@ -326,6 +327,7 @@
         "570-checker-osr/osr.cc",
         "595-profile-saving/profile-saving.cc",
         "596-app-images/app_images.cc",
+        "596-monitor-inflation/monitor_inflation.cc",
         "597-deopt-new-string/deopt.cc",
         "626-const-class-linking/clear_dex_cache_types.cc",
     ],
diff --git a/test/Android.run-test.mk b/test/Android.run-test.mk
index 639996e..814f968 100644
--- a/test/Android.run-test.mk
+++ b/test/Android.run-test.mk
@@ -277,42 +277,6 @@
   147-stripped-dex-fallback \
   569-checker-pattern-replacement
 
-# These 9** tests are not supported in current form due to linker
-# restrictions. See b/31681198
-TEST_ART_BROKEN_TARGET_TESTS += \
-  901-hello-ti-agent \
-  902-hello-transformation \
-  903-hello-tagging \
-  904-object-allocation \
-  905-object-free \
-  906-iterate-heap \
-  907-get-loaded-classes \
-  908-gc-start-finish \
-  909-attach-agent \
-  910-methods \
-  911-get-stack-trace \
-  912-classes \
-  913-heaps \
-  914-hello-obsolescence \
-  915-obsolete-2 \
-  916-obsolete-jit \
-  917-fields-transformation \
-  918-fields \
-  919-obsolete-fields \
-  920-objects \
-  921-hello-failure \
-  922-properties \
-  923-monitors \
-  924-threads \
-  925-threadgroups \
-  926-multi-obsolescence \
-  927-timers \
-  928-jni-table \
-  929-search \
-  930-hello-retransform \
-  931-agent-thread \
-  932-transform-saves \
-
 ifneq (,$(filter target,$(TARGET_TYPES)))
   ART_TEST_KNOWN_BROKEN += $(call all-run-test-names,target,$(RUN_TYPES),$(PREBUILD_TYPES), \
       $(COMPILER_TYPES),$(RELOCATE_TYPES),$(TRACE_TYPES),$(GC_TYPES),$(JNI_TYPES), \
diff --git a/test/run-test b/test/run-test
index 9b17802..c78fa35 100755
--- a/test/run-test
+++ b/test/run-test
@@ -131,6 +131,7 @@
 multi_image_suffix=""
 android_root="/system"
 bisection_search="no"
+suspend_timeout="500000"
 # By default we will use optimizing.
 image_args=""
 image_suffix=""
@@ -219,6 +220,10 @@
         basic_verify="true"
         gc_stress="true"
         shift
+    elif [ "x$1" = "x--suspend-timeout" ]; then
+        shift
+        suspend_timeout="$1"
+        shift
     elif [ "x$1" = "x--image" ]; then
         shift
         image="$1"
@@ -402,6 +407,9 @@
 tmp_dir="`cd $oldwd ; python -c "import os; print os.path.realpath('$tmp_dir')"`"
 mkdir -p $tmp_dir
 
+# Add thread suspend timeout flag
+run_args="${run_args} --runtime-option -XX:ThreadSuspendTimeout=$suspend_timeout"
+
 if [ "$basic_verify" = "true" ]; then
   # Set HspaceCompactForOOMMinIntervalMs to zero to run hspace compaction for OOM more frequently in tests.
   run_args="${run_args} --runtime-option -Xgc:preverify --runtime-option -Xgc:postverify --runtime-option -XX:HspaceCompactForOOMMinIntervalMs=0"
@@ -649,6 +657,7 @@
         echo "    --quiet               Don't print anything except failure messages"
         echo "    --bisection-search    Perform bisection bug search."
         echo "    --vdex                Test using vdex as in input to dex2oat. Only works with --prebuild."
+        echo "    --suspend-timeout     Change thread suspend timeout ms (default 500000)."
     ) 1>&2  # Direct to stderr so usage is not printed if --quiet is set.
     exit 1
 fi
diff --git a/test/ti-agent/common_helper.cc b/test/ti-agent/common_helper.cc
index 4bceef5..80e1797 100644
--- a/test/ti-agent/common_helper.cc
+++ b/test/ti-agent/common_helper.cc
@@ -16,14 +16,18 @@
 
 #include "ti-agent/common_helper.h"
 
+#include <dlfcn.h>
 #include <stdio.h>
 #include <sstream>
 #include <deque>
 
+#include "android-base/stringprintf.h"
 #include "art_method.h"
 #include "jni.h"
+#include "jni_internal.h"
 #include "openjdkjvmti/jvmti.h"
 #include "scoped_thread_state_change-inl.h"
+#include "ScopedLocalRef.h"
 #include "stack.h"
 #include "ti-agent/common_load.h"
 #include "utils.h"
@@ -325,4 +329,123 @@
 
 }  // namespace common_retransform
 
+static void BindMethod(jvmtiEnv* jenv,
+                       JNIEnv* env,
+                       jclass klass,
+                       jmethodID method) {
+  char* name;
+  char* signature;
+  jvmtiError name_result = jenv->GetMethodName(method, &name, &signature, nullptr);
+  if (name_result != JVMTI_ERROR_NONE) {
+    LOG(FATAL) << "Could not get methods";
+  }
+
+  ArtMethod* m = jni::DecodeArtMethod(method);
+
+  std::string names[2];
+  {
+    ScopedObjectAccess soa(Thread::Current());
+    names[0] = m->JniShortName();
+    names[1] = m->JniLongName();
+  }
+  for (const std::string& mangled_name : names) {
+    void* sym = dlsym(RTLD_DEFAULT, mangled_name.c_str());
+    if (sym == nullptr) {
+      continue;
+    }
+
+    JNINativeMethod native_method;
+    native_method.fnPtr = sym;
+    native_method.name = name;
+    native_method.signature = signature;
+
+    env->RegisterNatives(klass, &native_method, 1);
+
+    jenv->Deallocate(reinterpret_cast<unsigned char*>(name));
+    jenv->Deallocate(reinterpret_cast<unsigned char*>(signature));
+    return;
+  }
+
+  LOG(FATAL) << "Could not find " << names[0];
+}
+
+static jclass FindClassWithSystemClassLoader(JNIEnv* env, const char* class_name) {
+  // Find the system classloader.
+  ScopedLocalRef<jclass> cl_klass(env, env->FindClass("java/lang/ClassLoader"));
+  if (cl_klass.get() == nullptr) {
+    return nullptr;
+  }
+  jmethodID getsystemclassloader_method = env->GetStaticMethodID(cl_klass.get(),
+                                                                 "getSystemClassLoader",
+                                                                 "()Ljava/lang/ClassLoader;");
+  if (getsystemclassloader_method == nullptr) {
+    return nullptr;
+  }
+  ScopedLocalRef<jobject> cl(env, env->CallStaticObjectMethod(cl_klass.get(),
+                                                              getsystemclassloader_method));
+  if (cl.get() == nullptr) {
+    return nullptr;
+  }
+
+  // Create a String of the name.
+  std::string descriptor = android::base::StringPrintf("L%s;", class_name);
+  std::string dot_name = DescriptorToDot(descriptor.c_str());
+  ScopedLocalRef<jstring> name_str(env, env->NewStringUTF(dot_name.c_str()));
+
+  // Call Class.forName with it.
+  ScopedLocalRef<jclass> c_klass(env, env->FindClass("java/lang/Class"));
+  if (c_klass.get() == nullptr) {
+    return nullptr;
+  }
+  jmethodID forname_method = env->GetStaticMethodID(
+      c_klass.get(),
+      "forName",
+      "(Ljava/lang/String;ZLjava/lang/ClassLoader;)Ljava/lang/Class;");
+  if (forname_method == nullptr) {
+    return nullptr;
+  }
+
+  return reinterpret_cast<jclass>(env->CallStaticObjectMethod(c_klass.get(),
+                                                              forname_method,
+                                                              name_str.get(),
+                                                              JNI_FALSE,
+                                                              cl.get()));
+}
+
+void BindFunctions(jvmtiEnv* jenv, JNIEnv* env, const char* class_name) {
+  // Use JNI to load the class.
+  ScopedLocalRef<jclass> klass(env, env->FindClass(class_name));
+  if (klass.get() == nullptr) {
+    // We may be called with the wrong classloader. Try explicitly using the system classloader.
+    env->ExceptionClear();
+    klass.reset(FindClassWithSystemClassLoader(env, class_name));
+    if (klass.get() == nullptr) {
+      LOG(FATAL) << "Could not load " << class_name;
+    }
+  }
+
+  // Use JVMTI to get the methods.
+  jint method_count;
+  jmethodID* methods;
+  jvmtiError methods_result = jenv->GetClassMethods(klass.get(), &method_count, &methods);
+  if (methods_result != JVMTI_ERROR_NONE) {
+    LOG(FATAL) << "Could not get methods";
+  }
+
+  // Check each method.
+  for (jint i = 0; i < method_count; ++i) {
+    jint modifiers;
+    jvmtiError mod_result = jenv->GetMethodModifiers(methods[i], &modifiers);
+    if (mod_result != JVMTI_ERROR_NONE) {
+      LOG(FATAL) << "Could not get methods";
+    }
+    constexpr jint kNative = static_cast<jint>(kAccNative);
+    if ((modifiers & kNative) != 0) {
+      BindMethod(jenv, env, klass.get(), methods[i]);
+    }
+  }
+
+  jenv->Deallocate(reinterpret_cast<unsigned char*>(methods));
+}
+
 }  // namespace art
diff --git a/test/ti-agent/common_helper.h b/test/ti-agent/common_helper.h
index 8599fc4..c60553d 100644
--- a/test/ti-agent/common_helper.h
+++ b/test/ti-agent/common_helper.h
@@ -71,6 +71,12 @@
 
 bool JvmtiErrorToException(JNIEnv* env, jvmtiError error);
 
+// Load the class through JNI. Inspect it, find all native methods. Construct the corresponding
+// mangled name, run dlsym and bind the method.
+//
+// This will abort on failure.
+void BindFunctions(jvmtiEnv* jvmti_env, JNIEnv* env, const char* class_name);
+
 }  // namespace art
 
 #endif  // ART_TEST_TI_AGENT_COMMON_HELPER_H_
diff --git a/test/ti-agent/common_load.cc b/test/ti-agent/common_load.cc
index f4ce4c3..8ed8e67 100644
--- a/test/ti-agent/common_load.cc
+++ b/test/ti-agent/common_load.cc
@@ -14,6 +14,8 @@
  * limitations under the License.
  */
 
+#include "common_load.h"
+
 #include <jni.h>
 #include <stdio.h>
 // TODO I don't know?
@@ -22,7 +24,6 @@
 #include "art_method-inl.h"
 #include "base/logging.h"
 #include "base/macros.h"
-#include "common_load.h"
 #include "common_helper.h"
 
 #include "901-hello-ti-agent/basics.h"
@@ -32,6 +33,8 @@
 
 jvmtiEnv* jvmti_env;
 
+namespace {
+
 using OnLoad   = jint (*)(JavaVM* vm, char* options, void* reserved);
 using OnAttach = jint (*)(JavaVM* vm, char* options, void* reserved);
 
@@ -41,11 +44,50 @@
   OnAttach attach;
 };
 
+static void JNICALL VMInitCallback(jvmtiEnv *jvmti_env,
+                                   JNIEnv* jni_env,
+                                   jthread thread ATTRIBUTE_UNUSED) {
+  // Bind Main native methods.
+  BindFunctions(jvmti_env, jni_env, "Main");
+}
+
+// Install a phase callback that will bind JNI functions on VMInit.
+bool InstallBindCallback(JavaVM* vm) {
+  // Use a new jvmtiEnv. Otherwise we might collide with table changes.
+  jvmtiEnv* install_env;
+  if (vm->GetEnv(reinterpret_cast<void**>(&install_env), JVMTI_VERSION_1_0) != 0) {
+    return false;
+  }
+  SetAllCapabilities(install_env);
+
+  {
+    jvmtiEventCallbacks callbacks;
+    memset(&callbacks, 0, sizeof(jvmtiEventCallbacks));
+    callbacks.VMInit = VMInitCallback;
+
+    jvmtiError install_error = install_env->SetEventCallbacks(&callbacks, sizeof(callbacks));
+    if (install_error != JVMTI_ERROR_NONE) {
+      return false;
+    }
+  }
+
+  {
+    jvmtiError enable_error = install_env->SetEventNotificationMode(JVMTI_ENABLE,
+                                                                    JVMTI_EVENT_VM_INIT,
+                                                                    nullptr);
+    if (enable_error != JVMTI_ERROR_NONE) {
+      return false;
+    }
+  }
+
+  return true;
+}
+
 // A trivial OnLoad implementation that only initializes the global jvmti_env.
 static jint MinimalOnLoad(JavaVM* vm,
                           char* options ATTRIBUTE_UNUSED,
                           void* reserved ATTRIBUTE_UNUSED) {
-  if (vm->GetEnv(reinterpret_cast<void**>(&jvmti_env), JVMTI_VERSION_1_0)) {
+  if (vm->GetEnv(reinterpret_cast<void**>(&jvmti_env), JVMTI_VERSION_1_0) != 0) {
     printf("Unable to get jvmti env!\n");
     return 1;
   }
@@ -55,7 +97,7 @@
 
 // A list of all non-standard the agents we have for testing. All other agents will use
 // MinimalOnLoad.
-AgentLib agents[] = {
+static AgentLib agents[] = {
   { "901-hello-ti-agent", Test901HelloTi::OnLoad, nullptr },
   { "902-hello-transformation", common_redefine::OnLoad, nullptr },
   { "909-attach-agent", nullptr, Test909AttachAgent::OnAttach },
@@ -101,6 +143,28 @@
   RuntimeIsJVM = strncmp(options, "jvm", 3) == 0;
 }
 
+static bool BindFunctionsAttached(JavaVM* vm, const char* class_name) {
+  // Get a JNIEnv. As the thread is attached, we must not destroy it.
+  JNIEnv* env;
+  if (vm->GetEnv(reinterpret_cast<void**>(&env), JNI_VERSION_1_6) != 0) {
+    printf("Unable to get JNI env!\n");
+    return false;
+  }
+
+  jvmtiEnv* jenv;
+  if (vm->GetEnv(reinterpret_cast<void**>(&jenv), JVMTI_VERSION_1_0) != 0) {
+    printf("Unable to get jvmti env!\n");
+    return false;
+  }
+  SetAllCapabilities(jenv);
+
+  BindFunctions(jenv, env, class_name);
+
+  return true;
+}
+
+}  // namespace
+
 extern "C" JNIEXPORT jint JNICALL Agent_OnLoad(JavaVM* vm, char* options, void* reserved) {
   char* remaining_options = nullptr;
   char* name_option = nullptr;
@@ -111,6 +175,10 @@
 
   SetIsJVM(remaining_options);
 
+  if (!InstallBindCallback(vm)) {
+    return 1;
+  }
+
   AgentLib* lib = FindAgent(name_option);
   OnLoad fn = nullptr;
   if (lib == nullptr) {
@@ -132,6 +200,9 @@
     printf("Unable to find agent name in options: %s\n", options);
     return -1;
   }
+
+  BindFunctionsAttached(vm, "Main");
+
   AgentLib* lib = FindAgent(name_option);
   if (lib == nullptr) {
     printf("Unable to find agent named: %s, add it to the list in test/ti-agent/common_load.cc\n",
diff --git a/test/ti-agent/common_load.h b/test/ti-agent/common_load.h
index fac94b4..d254421 100644
--- a/test/ti-agent/common_load.h
+++ b/test/ti-agent/common_load.h
@@ -17,6 +17,7 @@
 #ifndef ART_TEST_TI_AGENT_COMMON_LOAD_H_
 #define ART_TEST_TI_AGENT_COMMON_LOAD_H_
 
+#include "jni.h"
 #include "openjdkjvmti/jvmti.h"
 
 namespace art {
diff --git a/test/valgrind-suppressions.txt b/test/valgrind-suppressions.txt
index c148ad0..c775f98 100644
--- a/test/valgrind-suppressions.txt
+++ b/test/valgrind-suppressions.txt
@@ -46,3 +46,26 @@
    fun:_ZN14UnwindMapLocal5BuildEv
    fun:_ZN12BacktraceMap6CreateEib
 }
+{
+   BackTraceReading64
+   Memcheck:Addr8
+   fun:access_mem_unrestricted
+   fun:_Uelf64_memory_read
+   fun:_Uelf64_get_load_base
+   fun:map_create_list
+   fun:unw_map_local_create
+   fun:_ZN14UnwindMapLocal5BuildEv
+   fun:_ZN12BacktraceMap6CreateEib
+}
+{
+   BackTraceReading32
+   Memcheck:Addr4
+   fun:access_mem_unrestricted
+   fun:_Uelf32_memory_read
+   fun:_Uelf32_get_load_base
+   fun:map_create_list
+   fun:unw_map_local_create
+   fun:_ZN14UnwindMapLocal5BuildEv
+   fun:_ZN12BacktraceMap6CreateEib
+}
+
diff --git a/tools/dexfuzz/src/dexfuzz/Options.java b/tools/dexfuzz/src/dexfuzz/Options.java
index 99e03e8..af8a05c 100644
--- a/tools/dexfuzz/src/dexfuzz/Options.java
+++ b/tools/dexfuzz/src/dexfuzz/Options.java
@@ -50,6 +50,7 @@
   public static String deviceName = "";
   public static boolean usingSpecificDevice = false;
   public static int repeat = 1;
+  public static int divergenceRetry = 10;
   public static String executeDirectory = "/data/art-test";
   public static String androidRoot = "";
   public static String dumpMutationsFile = "mutations.dump";
@@ -118,6 +119,8 @@
     Log.always("    --repeat=<n>         : Fuzz N programs, executing each one.");
     Log.always("    --short-timeouts     : Shorten timeouts (faster; use if");
     Log.always("                           you want to focus on output divergences)");
+    Log.always("    --divergence-retry=<n> : Number of retries when checking if test is");
+    Log.always("                           self-divergent. (Default: 10)");
     Log.always("  --seed=<seed>          : RNG seed to use");
     Log.always("  --method-mutations=<n> : Maximum number of mutations to perform on each method.");
     Log.always("                           (Default: 3)");
@@ -239,6 +242,8 @@
       maxMethods = Integer.parseInt(value);
     } else if (key.equals("repeat")) {
       repeat = Integer.parseInt(value);
+    } else if (key.equals("divergence-retry")) {
+      divergenceRetry = Integer.parseInt(value);
     } else if (key.equals("log")) {
       Log.setLoggingLevel(LogTag.valueOf(value.toUpperCase()));
     } else if (key.equals("likelihoods")) {
@@ -360,6 +365,10 @@
       Log.error("--repeat must be at least 1!");
       return false;
     }
+    if (divergenceRetry < 0) {
+      Log.error("--divergence-retry cannot be negative!");
+      return false;
+    }
     if (usingProvidedSeed && repeat > 1) {
       Log.error("Cannot use --repeat with --seed");
       return false;
diff --git a/tools/dexfuzz/src/dexfuzz/fuzzers/Fuzzer.java b/tools/dexfuzz/src/dexfuzz/fuzzers/Fuzzer.java
index 1797d90..ccc426c 100644
--- a/tools/dexfuzz/src/dexfuzz/fuzzers/Fuzzer.java
+++ b/tools/dexfuzz/src/dexfuzz/fuzzers/Fuzzer.java
@@ -298,13 +298,13 @@
   }
 
   private boolean checkGoldenExecutorForSelfDivergence(String programName) {
-    // Run golden executor 5 times, make sure it always produces
+    // Run golden executor multiple times, make sure it always produces
     // the same output, otherwise report that it is self-divergent.
 
     // TODO: Instead, produce a list of acceptable outputs, and see if the divergent
     // outputs of the backends fall within this set of outputs.
     String seenOutput = null;
-    for (int i = 0; i < 5; i++) {
+    for (int i = 0; i < Options.divergenceRetry + 1; i++) {
       goldenExecutor.reset();
       goldenExecutor.execute(programName);
       String output = goldenExecutor.getResult().getFlattenedOutput();
diff --git a/tools/stream-trace-converter.py b/tools/stream-trace-converter.py
index 951b05b..7e341f2 100755
--- a/tools/stream-trace-converter.py
+++ b/tools/stream-trace-converter.py
@@ -124,12 +124,20 @@
     self._threads.append('%d\t%s\n' % (tid, str))
     print 'New thread: %d/%s' % (tid, str)
 
+  def ProcessTraceSummary(self, input):
+    summaryLength = ReadIntLE(input)
+    str = input.read(summaryLength)
+    self._summary = str
+    print 'Summary: \"%s\"' % str
+
   def ProcessSpecial(self, input):
     code = ord(input.read(1))
     if code == 1:
       self.ProcessMethod(input)
     elif code == 2:
       self.ProcessThread(input)
+    elif code == 3:
+      self.ProcessTraceSummary(input)
     else:
       raise MyException("Unknown special!")
 
@@ -147,9 +155,24 @@
       print 'Buffer underrun, file was probably truncated. Results should still be usable.'
 
   def Finalize(self, header):
-    header.write('*threads\n')
-    for t in self._threads:
-      header.write(t)
+    # If the summary is present in the input file, use it as the header except
+    # for the methods section which is emtpy in the input file. If not present,
+    # apppend header with the threads that are recorded in the input stream.
+    if (self._summary):
+      # Erase the contents that's already written earlier by PrintHeader.
+      header.seek(0)
+      header.truncate()
+      # Copy the lines from the input summary to the output header until
+      # the methods section is seen.
+      for line in self._summary.splitlines(True):
+        if line == "*methods\n":
+          break
+        else:
+          header.write(line)
+    else:
+      header.write('*threads\n')
+      for t in self._threads:
+        header.write(t)
     header.write('*methods\n')
     for m in self._methods:
       header.write(m)
@@ -166,6 +189,7 @@
 
     self._methods = []
     self._threads = []
+    self._summary = None
     self.Process(input, body)
 
     self.Finalize(header)