From: Treehugger Robot <treehugger-gerrit@google.com>
Date: Thu, 19 May 2016 12:19:53 +0000 (+0000)
Subject: Merge "Allow libcore and JDWP tests to be executed without JIT."
X-Git-Url: http://git.osdn.net/view?a=commitdiff_plain;h=6c313600e995ac81a6e68e84b64761a10fb1311c;hp=5db109bc8030858f2830f4217333768c3c77095f;p=android-x86%2Fart.git

Merge "Allow libcore and JDWP tests to be executed without JIT."
---

diff --git a/Android.mk b/Android.mk
index 25796a017..bb1334a05 100644
--- a/Android.mk
+++ b/Android.mk
@@ -93,6 +93,7 @@ include $(art_path)/tools/ahat/Android.mk
 include $(art_path)/tools/dexfuzz/Android.mk
 include $(art_path)/tools/dmtracedump/Android.mk
 include $(art_path)/sigchainlib/Android.mk
+include $(art_path)/libart_fake/Android.mk
 
 
 # ART_HOST_DEPENDENCIES depends on Android.executable.mk above for ART_HOST_EXECUTABLES
diff --git a/benchmark/jni-perf/src/JniPerfBenchmark.java b/benchmark/jni-perf/src/JniPerfBenchmark.java
index b1b21ce0b..1e7cc2bf4 100644
--- a/benchmark/jni-perf/src/JniPerfBenchmark.java
+++ b/benchmark/jni-perf/src/JniPerfBenchmark.java
@@ -14,9 +14,7 @@
  * limitations under the License.
  */
 
-import com.google.caliper.SimpleBenchmark;
-
-public class JniPerfBenchmark extends SimpleBenchmark {
+public class JniPerfBenchmark {
   private static final String MSG = "ABCDE";
 
   native void perfJniEmptyCall();
diff --git a/benchmark/jobject-benchmark/src/JObjectBenchmark.java b/benchmark/jobject-benchmark/src/JObjectBenchmark.java
index f4c059c58..90a53b399 100644
--- a/benchmark/jobject-benchmark/src/JObjectBenchmark.java
+++ b/benchmark/jobject-benchmark/src/JObjectBenchmark.java
@@ -14,9 +14,7 @@
  * limitations under the License.
  */
 
-import com.google.caliper.SimpleBenchmark;
-
-public class JObjectBenchmark extends SimpleBenchmark {
+public class JObjectBenchmark {
   public JObjectBenchmark() {
     // Make sure to link methods before benchmark starts.
     System.loadLibrary("artbenchmark");
diff --git a/benchmark/scoped-primitive-array/src/ScopedPrimitiveArrayBenchmark.java b/benchmark/scoped-primitive-array/src/ScopedPrimitiveArrayBenchmark.java
index be276fe48..0ad9c3695 100644
--- a/benchmark/scoped-primitive-array/src/ScopedPrimitiveArrayBenchmark.java
+++ b/benchmark/scoped-primitive-array/src/ScopedPrimitiveArrayBenchmark.java
@@ -14,9 +14,7 @@
  * limitations under the License.
  */
 
-import com.google.caliper.SimpleBenchmark; 
-
-public class ScopedPrimitiveArrayBenchmark extends SimpleBenchmark {
+public class ScopedPrimitiveArrayBenchmark {
   // Measure adds the first and last element of the array by using ScopedPrimitiveArray.
   static native long measureByteArray(int reps, byte[] arr);
   static native long measureShortArray(int reps, short[] arr);
diff --git a/build/Android.common_build.mk b/build/Android.common_build.mk
index dde3cdb1d..0235a308f 100644
--- a/build/Android.common_build.mk
+++ b/build/Android.common_build.mk
@@ -363,11 +363,20 @@ ART_HOST_ASFLAGS += $(art_asflags)
 ifndef LIBART_IMG_TARGET_BASE_ADDRESS
   $(error LIBART_IMG_TARGET_BASE_ADDRESS unset)
 endif
+
+ART_TARGET_CFLAGS += $(art_cflags) -DART_TARGET \
+                     -DART_BASE_ADDRESS=$(LIBART_IMG_TARGET_BASE_ADDRESS) \
+
+ifeq ($(ART_TARGET_LINUX),true)
+# Setting ART_TARGET_LINUX to true compiles art/ assuming that the target device
+# will be running linux rather than android.
+ART_TARGET_CFLAGS += -DART_TARGET_LINUX
+else
 # The ART_TARGET_ANDROID macro is passed to target builds, which check
 # against it instead of against __ANDROID__ (which is provided by target
 # toolchains).
-ART_TARGET_CFLAGS += $(art_cflags) -DART_TARGET -DART_TARGET_ANDROID \
-                     -DART_BASE_ADDRESS=$(LIBART_IMG_TARGET_BASE_ADDRESS) \
+ART_TARGET_CFLAGS += -DART_TARGET_ANDROID
+endif
 
 ART_TARGET_CFLAGS += $(art_target_cflags)
 ART_TARGET_ASFLAGS += $(art_asflags)
diff --git a/build/Android.gtest.mk b/build/Android.gtest.mk
index cd463ecc7..3b459c3ad 100644
--- a/build/Android.gtest.mk
+++ b/build/Android.gtest.mk
@@ -650,11 +650,11 @@ endef  # define-art-gtest
 
 ifeq ($(ART_BUILD_TARGET),true)
   $(foreach file,$(RUNTIME_GTEST_TARGET_SRC_FILES), $(eval $(call define-art-gtest,target,$(file),,libbacktrace)))
-  $(foreach file,$(COMPILER_GTEST_TARGET_SRC_FILES), $(eval $(call define-art-gtest,target,$(file),art/compiler,libartd-compiler libbacktrace)))
+  $(foreach file,$(COMPILER_GTEST_TARGET_SRC_FILES), $(eval $(call define-art-gtest,target,$(file),art/compiler,libartd-compiler libbacktrace libnativeloader)))
 endif
 ifeq ($(ART_BUILD_HOST),true)
   $(foreach file,$(RUNTIME_GTEST_HOST_SRC_FILES), $(eval $(call define-art-gtest,host,$(file),,libbacktrace)))
-  $(foreach file,$(COMPILER_GTEST_HOST_SRC_FILES), $(eval $(call define-art-gtest,host,$(file),art/compiler,libartd-compiler libbacktrace)))
+  $(foreach file,$(COMPILER_GTEST_HOST_SRC_FILES), $(eval $(call define-art-gtest,host,$(file),art/compiler,libartd-compiler libbacktrace libnativeloader)))
 endif
 
 # Used outside the art project to get a list of the current tests
diff --git a/cmdline/cmdline_parser_test.cc b/cmdline/cmdline_parser_test.cc
index 81b854e9c..7c53e01c4 100644
--- a/cmdline/cmdline_parser_test.cc
+++ b/cmdline/cmdline_parser_test.cc
@@ -461,8 +461,8 @@ TEST_F(CmdlineParserTest, TestJitOptions) {
   * Test successes
   */
   {
-    EXPECT_SINGLE_PARSE_VALUE(true, "-Xusejit:true", M::UseJIT);
-    EXPECT_SINGLE_PARSE_VALUE(false, "-Xusejit:false", M::UseJIT);
+    EXPECT_SINGLE_PARSE_VALUE(true, "-Xusejit:true", M::UseJitCompilation);
+    EXPECT_SINGLE_PARSE_VALUE(false, "-Xusejit:false", M::UseJitCompilation);
   }
   {
     EXPECT_SINGLE_PARSE_VALUE(
diff --git a/cmdline/cmdline_types.h b/cmdline/cmdline_types.h
index c0a00cce7..4797540c3 100644
--- a/cmdline/cmdline_types.h
+++ b/cmdline/cmdline_types.h
@@ -620,6 +620,8 @@ struct CmdlineType<LogVerbosity> : CmdlineTypeParser<LogVerbosity> {
         log_verbosity.verifier = true;
       } else if (verbose_options[j] == "image") {
         log_verbosity.image = true;
+      } else if (verbose_options[j] == "systrace-locks") {
+        log_verbosity.systrace_lock_logging = true;
       } else {
         return Result::Usage(std::string("Unknown -verbose option ") + verbose_options[j]);
       }
diff --git a/compiler/cfi_test.h b/compiler/cfi_test.h
index 230cb9aee..f8b746093 100644
--- a/compiler/cfi_test.h
+++ b/compiler/cfi_test.h
@@ -55,7 +55,9 @@ class CFITest : public dwarf::DwarfTest {
                     kCFIFormat, 0, &debug_frame_data_, &debug_frame_patches);
     ReformatCfi(Objdump(false, "-W"), &lines);
     // Pretty-print assembly.
-    auto* opts = new DisassemblerOptions(false, actual_asm.data(), true);
+    const uint8_t* asm_base = actual_asm.data();
+    const uint8_t* asm_end = asm_base + actual_asm.size();
+    auto* opts = new DisassemblerOptions(false, asm_base, asm_end, true);
     std::unique_ptr<Disassembler> disasm(Disassembler::Create(isa, opts));
     std::stringstream stream;
     const uint8_t* base = actual_asm.data() + (isa == kThumb2 ? 1 : 0);
diff --git a/compiler/common_compiler_test.cc b/compiler/common_compiler_test.cc
index f75a252df..bf29e1c31 100644
--- a/compiler/common_compiler_test.cc
+++ b/compiler/common_compiler_test.cc
@@ -180,6 +180,7 @@ void CommonCompilerTest::CreateCompilerDriver(Compiler::Kind kind,
                                             isa,
                                             instruction_set_features_.get(),
                                             /* boot_image */ true,
+                                            /* app_image */ false,
                                             GetImageClasses(),
                                             GetCompiledClasses(),
                                             GetCompiledMethods(),
diff --git a/compiler/dex/verification_results.cc b/compiler/dex/verification_results.cc
index 1491a183f..606302bd7 100644
--- a/compiler/dex/verification_results.cc
+++ b/compiler/dex/verification_results.cc
@@ -60,7 +60,7 @@ void VerificationResults::ProcessVerifiedMethod(verifier::MethodVerifier* method
     // TODO: Investigate why are we doing the work again for this method and try to avoid it.
     LOG(WARNING) << "Method processed more than once: "
         << PrettyMethod(ref.dex_method_index, *ref.dex_file);
-    if (!Runtime::Current()->UseJit()) {
+    if (!Runtime::Current()->UseJitCompilation()) {
       DCHECK_EQ(it->second->GetDevirtMap().size(), verified_method->GetDevirtMap().size());
       DCHECK_EQ(it->second->GetSafeCastSet().size(), verified_method->GetSafeCastSet().size());
     }
diff --git a/compiler/dex/verified_method.cc b/compiler/dex/verified_method.cc
index 5c0253c29..bace01471 100644
--- a/compiler/dex/verified_method.cc
+++ b/compiler/dex/verified_method.cc
@@ -54,7 +54,8 @@ const VerifiedMethod* VerifiedMethod::Create(verifier::MethodVerifier* method_ve
     }
 
     // Only need dequicken info for JIT so far.
-    if (Runtime::Current()->UseJit() && !verified_method->GenerateDequickenMap(method_verifier)) {
+    if (Runtime::Current()->UseJitCompilation() &&
+        !verified_method->GenerateDequickenMap(method_verifier)) {
       return nullptr;
     }
   }
@@ -72,7 +73,7 @@ const MethodReference* VerifiedMethod::GetDevirtTarget(uint32_t dex_pc) const {
 }
 
 const DexFileReference* VerifiedMethod::GetDequickenIndex(uint32_t dex_pc) const {
-  DCHECK(Runtime::Current()->UseJit());
+  DCHECK(Runtime::Current()->UseJitCompilation());
   auto it = dequicken_map_.find(dex_pc);
   return (it != dequicken_map_.end()) ? &it->second : nullptr;
 }
diff --git a/compiler/driver/compiled_method_storage_test.cc b/compiler/driver/compiled_method_storage_test.cc
index 9e0c22c68..6863f42d1 100644
--- a/compiler/driver/compiled_method_storage_test.cc
+++ b/compiler/driver/compiled_method_storage_test.cc
@@ -36,6 +36,7 @@ TEST(CompiledMethodStorage, Deduplicate) {
                         /* instruction_set_ */ kNone,
                         /* instruction_set_features */ nullptr,
                         /* boot_image */ false,
+                        /* app_image */ false,
                         /* image_classes */ nullptr,
                         /* compiled_classes */ nullptr,
                         /* compiled_methods */ nullptr,
diff --git a/compiler/driver/compiler_driver.cc b/compiler/driver/compiler_driver.cc
index be82956e7..1ab1d31f0 100644
--- a/compiler/driver/compiler_driver.cc
+++ b/compiler/driver/compiler_driver.cc
@@ -341,6 +341,7 @@ CompilerDriver::CompilerDriver(
     InstructionSet instruction_set,
     const InstructionSetFeatures* instruction_set_features,
     bool boot_image,
+    bool app_image,
     std::unordered_set<std::string>* image_classes,
     std::unordered_set<std::string>* compiled_classes,
     std::unordered_set<std::string>* compiled_methods,
@@ -363,6 +364,7 @@ CompilerDriver::CompilerDriver(
       compiled_methods_(MethodTable::key_compare()),
       non_relative_linker_patch_count_(0u),
       boot_image_(boot_image),
+      app_image_(app_image),
       image_classes_(image_classes),
       classes_to_compile_(compiled_classes),
       methods_to_compile_(compiled_methods),
@@ -473,7 +475,7 @@ static optimizer::DexToDexCompilationLevel GetDexToDexCompilationLevel(
     const DexFile& dex_file, const DexFile::ClassDef& class_def)
     SHARED_REQUIRES(Locks::mutator_lock_) {
   auto* const runtime = Runtime::Current();
-  if (runtime->UseJit() || driver.GetCompilerOptions().VerifyAtRuntime()) {
+  if (runtime->UseJitCompilation() || driver.GetCompilerOptions().VerifyAtRuntime()) {
     // Verify at runtime shouldn't dex to dex since we didn't resolve of verify.
     return optimizer::DexToDexCompilationLevel::kDontDexToDexCompile;
   }
@@ -945,7 +947,7 @@ bool CompilerDriver::ShouldVerifyClassBasedOnProfile(const DexFile& dex_file,
 
 class ResolveCatchBlockExceptionsClassVisitor : public ClassVisitor {
  public:
-  ResolveCatchBlockExceptionsClassVisitor(
+  explicit ResolveCatchBlockExceptionsClassVisitor(
       std::set<std::pair<uint16_t, const DexFile*>>& exceptions_to_resolve)
      : exceptions_to_resolve_(exceptions_to_resolve) {}
 
@@ -1268,7 +1270,7 @@ void CompilerDriver::UpdateImageClasses(TimingLogger* timings) {
 bool CompilerDriver::CanAssumeClassIsLoaded(mirror::Class* klass) {
   Runtime* runtime = Runtime::Current();
   if (!runtime->IsAotCompiler()) {
-    DCHECK(runtime->UseJit());
+    DCHECK(runtime->UseJitCompilation());
     // Having the klass reference here implies that the klass is already loaded.
     return true;
   }
@@ -1289,7 +1291,7 @@ bool CompilerDriver::CanAssumeTypeIsPresentInDexCache(Handle<mirror::DexCache> d
   if ((IsBootImage() &&
        IsImageClass(dex_cache->GetDexFile()->StringDataByIdx(
            dex_cache->GetDexFile()->GetTypeId(type_idx).descriptor_idx_))) ||
-      Runtime::Current()->UseJit()) {
+      Runtime::Current()->UseJitCompilation()) {
     mirror::Class* resolved_class = dex_cache->GetResolvedType(type_idx);
     result = (resolved_class != nullptr);
   }
@@ -1307,7 +1309,7 @@ bool CompilerDriver::CanAssumeStringIsPresentInDexCache(const DexFile& dex_file,
   // See also Compiler::ResolveDexFile
 
   bool result = false;
-  if (IsBootImage() || Runtime::Current()->UseJit()) {
+  if (IsBootImage() || Runtime::Current()->UseJitCompilation()) {
     ScopedObjectAccess soa(Thread::Current());
     StackHandleScope<1> hs(soa.Self());
     ClassLinker* const class_linker = Runtime::Current()->GetClassLinker();
@@ -1319,7 +1321,7 @@ bool CompilerDriver::CanAssumeStringIsPresentInDexCache(const DexFile& dex_file,
       result = true;
     } else {
       // Just check whether the dex cache already has the string.
-      DCHECK(Runtime::Current()->UseJit());
+      DCHECK(Runtime::Current()->UseJitCompilation());
       result = (dex_cache->GetResolvedString(string_idx) != nullptr);
     }
   }
@@ -1427,7 +1429,7 @@ bool CompilerDriver::CanEmbedTypeInCode(const DexFile& dex_file, uint32_t type_i
     } else {
       return false;
     }
-  } else if (runtime->UseJit() && !heap->IsMovableObject(resolved_class)) {
+  } else if (runtime->UseJitCompilation() && !heap->IsMovableObject(resolved_class)) {
     *is_type_initialized = resolved_class->IsInitialized();
     // If the class may move around, then don't embed it as a direct pointer.
     *use_direct_type_ptr = true;
@@ -1604,7 +1606,7 @@ void CompilerDriver::GetCodeAndMethodForDirectCall(InvokeType* type, InvokeType
       }
     }
   }
-  if (runtime->UseJit()) {
+  if (runtime->UseJitCompilation()) {
     // If we are the JIT, then don't allow a direct call to the interpreter bridge since this will
     // never be updated even after we compile the method.
     if (cl->IsQuickToInterpreterBridge(
@@ -1636,7 +1638,7 @@ void CompilerDriver::GetCodeAndMethodForDirectCall(InvokeType* type, InvokeType
   bool must_use_direct_pointers = false;
   mirror::DexCache* dex_cache = declaring_class->GetDexCache();
   if (target_method->dex_file == dex_cache->GetDexFile() &&
-    !(runtime->UseJit() && dex_cache->GetResolvedMethod(
+    !(runtime->UseJitCompilation() && dex_cache->GetResolvedMethod(
         method->GetDexMethodIndex(), pointer_size) == nullptr)) {
     target_method->dex_method_index = method->GetDexMethodIndex();
   } else {
@@ -1673,7 +1675,7 @@ void CompilerDriver::GetCodeAndMethodForDirectCall(InvokeType* type, InvokeType
         break;
       }
     }
-    if (method_in_image || compiling_boot || runtime->UseJit()) {
+    if (method_in_image || compiling_boot || runtime->UseJitCompilation()) {
       // We know we must be able to get to the method in the image, so use that pointer.
       // In the case where we are the JIT, we can always use direct pointers since we know where
       // the method and its code are / will be. We don't sharpen to interpreter bridge since we
@@ -2440,9 +2442,12 @@ void CompilerDriver::InitializeClasses(jobject jni_class_loader,
   context.ForAll(0, dex_file.NumClassDefs(), &visitor, init_thread_count);
 }
 
-class InitializeArrayClassVisitor : public ClassVisitor {
+class InitializeArrayClassesAndCreateConflictTablesVisitor : public ClassVisitor {
  public:
   virtual bool operator()(mirror::Class* klass) OVERRIDE SHARED_REQUIRES(Locks::mutator_lock_) {
+    if (Runtime::Current()->GetHeap()->ObjectIsInBootImageSpace(klass)) {
+      return true;
+    }
     if (klass->IsArrayClass()) {
       StackHandleScope<1> hs(Thread::Current());
       Runtime::Current()->GetClassLinker()->EnsureInitialized(hs.Self(),
@@ -2450,6 +2455,10 @@ class InitializeArrayClassVisitor : public ClassVisitor {
                                                               true,
                                                               true);
     }
+    // Create the conflict tables.
+    if (!klass->IsTemp() && klass->ShouldHaveEmbeddedImtAndVTable()) {
+      Runtime::Current()->GetClassLinker()->FillIMTAndConflictTables(klass);
+    }
     return true;
   }
 };
@@ -2462,13 +2471,15 @@ void CompilerDriver::InitializeClasses(jobject class_loader,
     CHECK(dex_file != nullptr);
     InitializeClasses(class_loader, *dex_file, dex_files, timings);
   }
-  {
+  if (boot_image_ || app_image_) {
     // Make sure that we call EnsureIntiailized on all the array classes to call
     // SetVerificationAttempted so that the access flags are set. If we do not do this they get
     // changed at runtime resulting in more dirty image pages.
+    // Also create conflict tables.
+    // Only useful if we are compiling an image (image_classes_ is not null).
     ScopedObjectAccess soa(Thread::Current());
-    InitializeArrayClassVisitor visitor;
-    Runtime::Current()->GetClassLinker()->VisitClasses(&visitor);
+    InitializeArrayClassesAndCreateConflictTablesVisitor visitor;
+    Runtime::Current()->GetClassLinker()->VisitClassesWithoutClassesLock(&visitor);
   }
   if (IsBootImage()) {
     // Prune garbage objects created during aborted transactions.
diff --git a/compiler/driver/compiler_driver.h b/compiler/driver/compiler_driver.h
index d63dffa49..19a1ecc49 100644
--- a/compiler/driver/compiler_driver.h
+++ b/compiler/driver/compiler_driver.h
@@ -92,6 +92,7 @@ class CompilerDriver {
                  InstructionSet instruction_set,
                  const InstructionSetFeatures* instruction_set_features,
                  bool boot_image,
+                 bool app_image,
                  std::unordered_set<std::string>* image_classes,
                  std::unordered_set<std::string>* compiled_classes,
                  std::unordered_set<std::string>* compiled_methods,
@@ -652,6 +653,7 @@ class CompilerDriver {
   size_t non_relative_linker_patch_count_ GUARDED_BY(compiled_methods_lock_);
 
   const bool boot_image_;
+  const bool app_image_;
 
   // If image_ is true, specifies the classes that will be included in the image.
   // Note if image_classes_ is null, all classes are included in the image.
diff --git a/compiler/driver/compiler_options.cc b/compiler/driver/compiler_options.cc
index 1bd4c3ad8..f20dba34a 100644
--- a/compiler/driver/compiler_options.cc
+++ b/compiler/driver/compiler_options.cc
@@ -21,7 +21,7 @@
 namespace art {
 
 CompilerOptions::CompilerOptions()
-    : compiler_filter_(kDefaultCompilerFilter),
+    : compiler_filter_(CompilerFilter::kDefaultCompilerFilter),
       huge_method_threshold_(kDefaultHugeMethodThreshold),
       large_method_threshold_(kDefaultLargeMethodThreshold),
       small_method_threshold_(kDefaultSmallMethodThreshold),
diff --git a/compiler/driver/compiler_options.h b/compiler/driver/compiler_options.h
index c67ab6ef1..6bbd3c5a1 100644
--- a/compiler/driver/compiler_options.h
+++ b/compiler/driver/compiler_options.h
@@ -31,7 +31,6 @@ namespace art {
 class CompilerOptions FINAL {
  public:
   // Guide heuristics to determine whether to compile method if profile data not available.
-  static const CompilerFilter::Filter kDefaultCompilerFilter = CompilerFilter::kSpeed;
   static const size_t kDefaultHugeMethodThreshold = 10000;
   static const size_t kDefaultLargeMethodThreshold = 600;
   static const size_t kDefaultSmallMethodThreshold = 60;
diff --git a/compiler/image_writer.cc b/compiler/image_writer.cc
index 8bb462c66..be720ad2f 100644
--- a/compiler/image_writer.cc
+++ b/compiler/image_writer.cc
@@ -653,8 +653,7 @@ bool ImageWriter::AllocMemory() {
   for (ImageInfo& image_info : image_infos_) {
     ImageSection unused_sections[ImageHeader::kSectionCount];
     const size_t length = RoundUp(
-        image_info.CreateImageSections(target_ptr_size_, unused_sections),
-        kPageSize);
+        image_info.CreateImageSections(unused_sections), kPageSize);
 
     std::string error_msg;
     image_info.image_.reset(MemMap::MapAnonymous("image writer image",
@@ -1029,6 +1028,9 @@ ObjectArray<Object>* ImageWriter::CreateImageRoots(size_t oat_index) const {
     for (const ClassLinker::DexCacheData& data : class_linker->GetDexCachesData()) {
       mirror::DexCache* dex_cache =
           down_cast<mirror::DexCache*>(self->DecodeJObject(data.weak_root));
+      if (dex_cache == nullptr) {
+        continue;
+      }
       const DexFile* dex_file = dex_cache->GetDexFile();
       if (!IsInBootImage(dex_cache)) {
         dex_cache_count += image_dex_files.find(dex_file) != image_dex_files.end() ? 1u : 0u;
@@ -1045,6 +1047,9 @@ ObjectArray<Object>* ImageWriter::CreateImageRoots(size_t oat_index) const {
     for (const ClassLinker::DexCacheData& data : class_linker->GetDexCachesData()) {
       mirror::DexCache* dex_cache =
           down_cast<mirror::DexCache*>(self->DecodeJObject(data.weak_root));
+      if (dex_cache == nullptr) {
+        continue;
+      }
       const DexFile* dex_file = dex_cache->GetDexFile();
       if (!IsInBootImage(dex_cache)) {
         non_image_dex_caches += image_dex_files.find(dex_file) != image_dex_files.end() ? 1u : 0u;
@@ -1056,6 +1061,9 @@ ObjectArray<Object>* ImageWriter::CreateImageRoots(size_t oat_index) const {
     for (const ClassLinker::DexCacheData& data : class_linker->GetDexCachesData()) {
       mirror::DexCache* dex_cache =
           down_cast<mirror::DexCache*>(self->DecodeJObject(data.weak_root));
+      if (dex_cache == nullptr) {
+        continue;
+      }
       const DexFile* dex_file = dex_cache->GetDexFile();
       if (!IsInBootImage(dex_cache) && image_dex_files.find(dex_file) != image_dex_files.end()) {
         dex_caches->Set<false>(i, dex_cache);
@@ -1215,6 +1223,19 @@ void ImageWriter::WalkFieldsInOrder(mirror::Object* obj) {
         }
         (any_dirty ? dirty_methods_ : clean_methods_) += num_methods;
       }
+      // Assign offsets for all runtime methods in the IMT since these may hold conflict tables
+      // live.
+      if (as_klass->ShouldHaveEmbeddedImtAndVTable()) {
+        for (size_t i = 0; i < mirror::Class::kImtSize; ++i) {
+          ArtMethod* imt_method = as_klass->GetEmbeddedImTableEntry(i, target_ptr_size_);
+          DCHECK(imt_method != nullptr);
+          if (imt_method->IsRuntimeMethod() &&
+              !IsInBootImage(imt_method) &&
+              !NativeRelocationAssigned(imt_method)) {
+            AssignMethodOffset(imt_method, kNativeObjectRelocationTypeRuntimeMethod, oat_index);
+          }
+        }
+      }
     } else if (h_obj->IsObjectArray()) {
       // Walk elements of an object array.
       int32_t length = h_obj->AsObjectArray<mirror::Object>()->GetLength();
@@ -1237,13 +1258,37 @@ void ImageWriter::WalkFieldsInOrder(mirror::Object* obj) {
   }
 }
 
+bool ImageWriter::NativeRelocationAssigned(void* ptr) const {
+  return native_object_relocations_.find(ptr) != native_object_relocations_.end();
+}
+
+void ImageWriter::TryAssignConflictTableOffset(ImtConflictTable* table, size_t oat_index) {
+  // No offset, or already assigned.
+  if (table == nullptr || NativeRelocationAssigned(table)) {
+    return;
+  }
+  CHECK(!IsInBootImage(table));
+  // If the method is a conflict method we also want to assign the conflict table offset.
+  ImageInfo& image_info = GetImageInfo(oat_index);
+  const size_t size = table->ComputeSize(target_ptr_size_);
+  native_object_relocations_.emplace(
+      table,
+      NativeObjectRelocation {
+          oat_index,
+          image_info.bin_slot_sizes_[kBinIMTConflictTable],
+          kNativeObjectRelocationTypeIMTConflictTable});
+  image_info.bin_slot_sizes_[kBinIMTConflictTable] += size;
+}
+
 void ImageWriter::AssignMethodOffset(ArtMethod* method,
                                      NativeObjectRelocationType type,
                                      size_t oat_index) {
   DCHECK(!IsInBootImage(method));
-  auto it = native_object_relocations_.find(method);
-  CHECK(it == native_object_relocations_.end()) << "Method " << method << " already assigned "
+  CHECK(!NativeRelocationAssigned(method)) << "Method " << method << " already assigned "
       << PrettyMethod(method);
+  if (method->IsRuntimeMethod()) {
+    TryAssignConflictTableOffset(method->GetImtConflictTable(target_ptr_size_), oat_index);
+  }
   ImageInfo& image_info = GetImageInfo(oat_index);
   size_t& offset = image_info.bin_slot_sizes_[BinTypeForNativeRelocationType(type)];
   native_object_relocations_.emplace(method, NativeObjectRelocation { oat_index, offset, type });
@@ -1292,8 +1337,7 @@ void ImageWriter::CalculateNewObjectOffsets() {
   // know where image_roots is going to end up
   image_objects_offset_begin_ = RoundUp(sizeof(ImageHeader), kObjectAlignment);  // 64-bit-alignment
 
-  // Clear any pre-existing monitors which may have been in the monitor words, assign bin slots.
-  heap->VisitObjects(WalkFieldsCallback, this);
+  const size_t method_alignment = ArtMethod::Alignment(target_ptr_size_);
   // Write the image runtime methods.
   image_methods_[ImageHeader::kResolutionMethod] = runtime->GetResolutionMethod();
   image_methods_[ImageHeader::kImtConflictMethod] = runtime->GetImtConflictMethod();
@@ -1303,31 +1347,19 @@ void ImageWriter::CalculateNewObjectOffsets() {
       runtime->GetCalleeSaveMethod(Runtime::kRefsOnly);
   image_methods_[ImageHeader::kRefsAndArgsSaveMethod] =
       runtime->GetCalleeSaveMethod(Runtime::kRefsAndArgs);
-
-  // Add room for fake length prefixed array for holding the image methods.
-  const auto image_method_type = kNativeObjectRelocationTypeArtMethodArrayClean;
-  auto it = native_object_relocations_.find(&image_method_array_);
-  CHECK(it == native_object_relocations_.end());
-  ImageInfo& default_image_info = GetImageInfo(GetDefaultOatIndex());
-  size_t& offset =
-      default_image_info.bin_slot_sizes_[BinTypeForNativeRelocationType(image_method_type)];
-  if (!compile_app_image_) {
-    native_object_relocations_.emplace(&image_method_array_,
-        NativeObjectRelocation { GetDefaultOatIndex(), offset, image_method_type });
-  }
-  size_t method_alignment = ArtMethod::Alignment(target_ptr_size_);
-  const size_t array_size = LengthPrefixedArray<ArtMethod>::ComputeSize(
-      0, ArtMethod::Size(target_ptr_size_), method_alignment);
-  CHECK_ALIGNED_PARAM(array_size, method_alignment);
-  offset += array_size;
+  // Visit image methods first to have the main runtime methods in the first image.
   for (auto* m : image_methods_) {
     CHECK(m != nullptr);
     CHECK(m->IsRuntimeMethod());
     DCHECK_EQ(compile_app_image_, IsInBootImage(m)) << "Trampolines should be in boot image";
     if (!IsInBootImage(m)) {
-      AssignMethodOffset(m, kNativeObjectRelocationTypeArtMethodClean, GetDefaultOatIndex());
+      AssignMethodOffset(m, kNativeObjectRelocationTypeRuntimeMethod, GetDefaultOatIndex());
     }
   }
+
+  // Clear any pre-existing monitors which may have been in the monitor words, assign bin slots.
+  heap->VisitObjects(WalkFieldsCallback, this);
+
   // Calculate size of the dex cache arrays slot and prepare offsets.
   PrepareDexCacheArraySlots();
 
@@ -1346,15 +1378,22 @@ void ImageWriter::CalculateNewObjectOffsets() {
   for (ImageInfo& image_info : image_infos_) {
     size_t bin_offset = image_objects_offset_begin_;
     for (size_t i = 0; i != kBinSize; ++i) {
+      switch (i) {
+        case kBinArtMethodClean:
+        case kBinArtMethodDirty: {
+          bin_offset = RoundUp(bin_offset, method_alignment);
+          break;
+        }
+        case kBinIMTConflictTable: {
+          bin_offset = RoundUp(bin_offset, target_ptr_size_);
+          break;
+        }
+        default: {
+          // Normal alignment.
+        }
+      }
       image_info.bin_slot_offsets_[i] = bin_offset;
       bin_offset += image_info.bin_slot_sizes_[i];
-      if (i == kBinArtField) {
-        static_assert(kBinArtField + 1 == kBinArtMethodClean, "Methods follow fields.");
-        static_assert(alignof(ArtField) == 4u, "ArtField alignment is 4.");
-        DCHECK_ALIGNED(bin_offset, 4u);
-        DCHECK(method_alignment == 4u || method_alignment == 8u);
-        bin_offset = RoundUp(bin_offset, method_alignment);
-      }
     }
     // NOTE: There may be additional padding between the bin slots and the intern table.
     DCHECK_EQ(image_info.image_end_,
@@ -1367,9 +1406,7 @@ void ImageWriter::CalculateNewObjectOffsets() {
     image_info.image_begin_ = global_image_begin_ + image_offset;
     image_info.image_offset_ = image_offset;
     ImageSection unused_sections[ImageHeader::kSectionCount];
-    image_info.image_size_ = RoundUp(
-        image_info.CreateImageSections(target_ptr_size_, unused_sections),
-        kPageSize);
+    image_info.image_size_ = RoundUp(image_info.CreateImageSections(unused_sections), kPageSize);
     // There should be no gaps until the next image.
     image_offset += image_info.image_size_;
   }
@@ -1396,42 +1433,52 @@ void ImageWriter::CalculateNewObjectOffsets() {
   // Note that image_info.image_end_ is left at end of used mirror object section.
 }
 
-size_t ImageWriter::ImageInfo::CreateImageSections(size_t target_ptr_size,
-                                                   ImageSection* out_sections) const {
+size_t ImageWriter::ImageInfo::CreateImageSections(ImageSection* out_sections) const {
   DCHECK(out_sections != nullptr);
+
+  // Do not round up any sections here that are represented by the bins since it will break
+  // offsets.
+
   // Objects section
-  auto* objects_section = &out_sections[ImageHeader::kSectionObjects];
+  ImageSection* objects_section = &out_sections[ImageHeader::kSectionObjects];
   *objects_section = ImageSection(0u, image_end_);
-  size_t cur_pos = objects_section->End();
+
   // Add field section.
-  auto* field_section = &out_sections[ImageHeader::kSectionArtFields];
-  *field_section = ImageSection(cur_pos, bin_slot_sizes_[kBinArtField]);
+  ImageSection* field_section = &out_sections[ImageHeader::kSectionArtFields];
+  *field_section = ImageSection(bin_slot_offsets_[kBinArtField], bin_slot_sizes_[kBinArtField]);
   CHECK_EQ(bin_slot_offsets_[kBinArtField], field_section->Offset());
-  cur_pos = field_section->End();
-  // Round up to the alignment the required by the method section.
-  cur_pos = RoundUp(cur_pos, ArtMethod::Alignment(target_ptr_size));
+
   // Add method section.
-  auto* methods_section = &out_sections[ImageHeader::kSectionArtMethods];
-  *methods_section = ImageSection(cur_pos,
-                                  bin_slot_sizes_[kBinArtMethodClean] +
-                                      bin_slot_sizes_[kBinArtMethodDirty]);
-  CHECK_EQ(bin_slot_offsets_[kBinArtMethodClean], methods_section->Offset());
-  cur_pos = methods_section->End();
+  ImageSection* methods_section = &out_sections[ImageHeader::kSectionArtMethods];
+  *methods_section = ImageSection(
+      bin_slot_offsets_[kBinArtMethodClean],
+      bin_slot_sizes_[kBinArtMethodClean] + bin_slot_sizes_[kBinArtMethodDirty]);
+
+  // Conflict tables section.
+  ImageSection* imt_conflict_tables_section = &out_sections[ImageHeader::kSectionIMTConflictTables];
+  *imt_conflict_tables_section = ImageSection(bin_slot_offsets_[kBinIMTConflictTable],
+                                              bin_slot_sizes_[kBinIMTConflictTable]);
+
+  // Runtime methods section.
+  ImageSection* runtime_methods_section = &out_sections[ImageHeader::kSectionRuntimeMethods];
+  *runtime_methods_section = ImageSection(bin_slot_offsets_[kBinRuntimeMethod],
+                                          bin_slot_sizes_[kBinRuntimeMethod]);
+
   // Add dex cache arrays section.
-  auto* dex_cache_arrays_section = &out_sections[ImageHeader::kSectionDexCacheArrays];
-  *dex_cache_arrays_section = ImageSection(cur_pos, bin_slot_sizes_[kBinDexCacheArray]);
-  CHECK_EQ(bin_slot_offsets_[kBinDexCacheArray], dex_cache_arrays_section->Offset());
-  cur_pos = dex_cache_arrays_section->End();
+  ImageSection* dex_cache_arrays_section = &out_sections[ImageHeader::kSectionDexCacheArrays];
+  *dex_cache_arrays_section = ImageSection(bin_slot_offsets_[kBinDexCacheArray],
+                                           bin_slot_sizes_[kBinDexCacheArray]);
+
   // Round up to the alignment the string table expects. See HashSet::WriteToMemory.
-  cur_pos = RoundUp(cur_pos, sizeof(uint64_t));
+  size_t cur_pos = RoundUp(dex_cache_arrays_section->End(), sizeof(uint64_t));
   // Calculate the size of the interned strings.
-  auto* interned_strings_section = &out_sections[ImageHeader::kSectionInternedStrings];
+  ImageSection* interned_strings_section = &out_sections[ImageHeader::kSectionInternedStrings];
   *interned_strings_section = ImageSection(cur_pos, intern_table_bytes_);
   cur_pos = interned_strings_section->End();
   // Round up to the alignment the class table expects. See HashSet::WriteToMemory.
   cur_pos = RoundUp(cur_pos, sizeof(uint64_t));
   // Calculate the size of the class table section.
-  auto* class_table_section = &out_sections[ImageHeader::kSectionClassTable];
+  ImageSection* class_table_section = &out_sections[ImageHeader::kSectionClassTable];
   *class_table_section = ImageSection(cur_pos, class_table_bytes_);
   cur_pos = class_table_section->End();
   // Image end goes right before the start of the image bitmap.
@@ -1446,7 +1493,7 @@ void ImageWriter::CreateHeader(size_t oat_index) {
 
   // Create the image sections.
   ImageSection sections[ImageHeader::kSectionCount];
-  const size_t image_end = image_info.CreateImageSections(target_ptr_size_, sections);
+  const size_t image_end = image_info.CreateImageSections(sections);
 
   // Finally bitmap section.
   const size_t bitmap_bytes = image_info.image_bitmap_->Size();
@@ -1531,8 +1578,20 @@ class FixupRootVisitor : public RootVisitor {
   ImageWriter* const image_writer_;
 };
 
+void ImageWriter::CopyAndFixupImtConflictTable(ImtConflictTable* orig, ImtConflictTable* copy) {
+  const size_t count = orig->NumEntries(target_ptr_size_);
+  for (size_t i = 0; i < count; ++i) {
+    ArtMethod* interface_method = orig->GetInterfaceMethod(i, target_ptr_size_);
+    ArtMethod* implementation_method = orig->GetImplementationMethod(i, target_ptr_size_);
+    copy->SetInterfaceMethod(i, target_ptr_size_, NativeLocationInImage(interface_method));
+    copy->SetImplementationMethod(i,
+                                  target_ptr_size_,
+                                  NativeLocationInImage(implementation_method));
+  }
+}
+
 void ImageWriter::CopyAndFixupNativeData(size_t oat_index) {
-  ImageInfo& image_info = GetImageInfo(oat_index);
+  const ImageInfo& image_info = GetImageInfo(oat_index);
   // Copy ArtFields and methods to their locations and update the array for convenience.
   for (auto& pair : native_object_relocations_) {
     NativeObjectRelocation& relocation = pair.second;
@@ -1550,6 +1609,7 @@ void ImageWriter::CopyAndFixupNativeData(size_t oat_index) {
             GetImageAddress(reinterpret_cast<ArtField*>(pair.first)->GetDeclaringClass()));
         break;
       }
+      case kNativeObjectRelocationTypeRuntimeMethod:
       case kNativeObjectRelocationTypeArtMethodClean:
       case kNativeObjectRelocationTypeArtMethodDirty: {
         CopyAndFixupMethod(reinterpret_cast<ArtMethod*>(pair.first),
@@ -1575,26 +1635,22 @@ void ImageWriter::CopyAndFixupNativeData(size_t oat_index) {
       case kNativeObjectRelocationTypeDexCacheArray:
         // Nothing to copy here, everything is done in FixupDexCache().
         break;
+      case kNativeObjectRelocationTypeIMTConflictTable: {
+        auto* orig_table = reinterpret_cast<ImtConflictTable*>(pair.first);
+        CopyAndFixupImtConflictTable(
+            orig_table,
+            new(dest)ImtConflictTable(orig_table->NumEntries(target_ptr_size_), target_ptr_size_));
+        break;
+      }
     }
   }
   // Fixup the image method roots.
   auto* image_header = reinterpret_cast<ImageHeader*>(image_info.image_->Begin());
-  const ImageSection& methods_section = image_header->GetMethodsSection();
   for (size_t i = 0; i < ImageHeader::kImageMethodsCount; ++i) {
     ArtMethod* method = image_methods_[i];
     CHECK(method != nullptr);
-    // Only place runtime methods in the image of the default oat file.
-    if (method->IsRuntimeMethod() && oat_index != GetDefaultOatIndex()) {
-      continue;
-    }
     if (!IsInBootImage(method)) {
-      auto it = native_object_relocations_.find(method);
-      CHECK(it != native_object_relocations_.end()) << "No forwarding for " << PrettyMethod(method);
-      NativeObjectRelocation& relocation = it->second;
-      CHECK(methods_section.Contains(relocation.offset)) << relocation.offset << " not in "
-          << methods_section;
-      CHECK(relocation.IsArtMethodRelocation()) << relocation.type;
-      method = reinterpret_cast<ArtMethod*>(global_image_begin_ + it->second.offset);
+      method = NativeLocationInImage(method);
     }
     image_header->SetImageMethod(static_cast<ImageHeader::ImageMethod>(i), method);
   }
@@ -2057,24 +2113,28 @@ void ImageWriter::CopyAndFixupMethod(ArtMethod* orig,
 
   // The resolution method has a special trampoline to call.
   Runtime* runtime = Runtime::Current();
-  if (UNLIKELY(orig == runtime->GetResolutionMethod())) {
-    copy->SetEntryPointFromQuickCompiledCodePtrSize(
-        GetOatAddress(kOatAddressQuickResolutionTrampoline), target_ptr_size_);
-  } else if (UNLIKELY(orig == runtime->GetImtConflictMethod() ||
-                      orig == runtime->GetImtUnimplementedMethod())) {
-    copy->SetEntryPointFromQuickCompiledCodePtrSize(
-        GetOatAddress(kOatAddressQuickIMTConflictTrampoline), target_ptr_size_);
-  } else if (UNLIKELY(orig->IsRuntimeMethod())) {
-    bool found_one = false;
-    for (size_t i = 0; i < static_cast<size_t>(Runtime::kLastCalleeSaveType); ++i) {
-      auto idx = static_cast<Runtime::CalleeSaveType>(i);
-      if (runtime->HasCalleeSaveMethod(idx) && runtime->GetCalleeSaveMethod(idx) == orig) {
-        found_one = true;
-        break;
+  if (orig->IsRuntimeMethod()) {
+    ImtConflictTable* orig_table = orig->GetImtConflictTable(target_ptr_size_);
+    if (orig_table != nullptr) {
+      // Special IMT conflict method, normal IMT conflict method or unimplemented IMT method.
+      copy->SetEntryPointFromQuickCompiledCodePtrSize(
+          GetOatAddress(kOatAddressQuickIMTConflictTrampoline), target_ptr_size_);
+      copy->SetImtConflictTable(NativeLocationInImage(orig_table), target_ptr_size_);
+    } else if (UNLIKELY(orig == runtime->GetResolutionMethod())) {
+      copy->SetEntryPointFromQuickCompiledCodePtrSize(
+          GetOatAddress(kOatAddressQuickResolutionTrampoline), target_ptr_size_);
+    } else {
+      bool found_one = false;
+      for (size_t i = 0; i < static_cast<size_t>(Runtime::kLastCalleeSaveType); ++i) {
+        auto idx = static_cast<Runtime::CalleeSaveType>(i);
+        if (runtime->HasCalleeSaveMethod(idx) && runtime->GetCalleeSaveMethod(idx) == orig) {
+          found_one = true;
+          break;
+        }
       }
+      CHECK(found_one) << "Expected to find callee save method but got " << PrettyMethod(orig);
+      CHECK(copy->IsRuntimeMethod());
     }
-    CHECK(found_one) << "Expected to find callee save method but got " << PrettyMethod(orig);
-    CHECK(copy->IsRuntimeMethod());
   } else {
     // We assume all methods have code. If they don't currently then we set them to the use the
     // resolution trampoline. Abstract methods never have code and so we need to make sure their
@@ -2141,6 +2201,10 @@ ImageWriter::Bin ImageWriter::BinTypeForNativeRelocationType(NativeObjectRelocat
       return kBinArtMethodDirty;
     case kNativeObjectRelocationTypeDexCacheArray:
       return kBinDexCacheArray;
+    case kNativeObjectRelocationTypeRuntimeMethod:
+      return kBinRuntimeMethod;
+    case kNativeObjectRelocationTypeIMTConflictTable:
+      return kBinIMTConflictTable;
   }
   UNREACHABLE();
 }
@@ -2242,7 +2306,6 @@ ImageWriter::ImageWriter(
       compile_app_image_(compile_app_image),
       target_ptr_size_(InstructionSetPointerSize(compiler_driver_.GetInstructionSet())),
       image_infos_(oat_filenames.size()),
-      image_method_array_(ImageHeader::kImageMethodsCount),
       dirty_methods_(0u),
       clean_methods_(0u),
       image_storage_mode_(image_storage_mode),
diff --git a/compiler/image_writer.h b/compiler/image_writer.h
index 0cb6aea9b..51976c511 100644
--- a/compiler/image_writer.h
+++ b/compiler/image_writer.h
@@ -169,6 +169,10 @@ class ImageWriter FINAL {
     // ArtMethods may be dirty if the class has native methods or a declaring class that isn't
     // initialized.
     kBinArtMethodDirty,
+    // Conflict tables (clean).
+    kBinIMTConflictTable,
+    // Runtime methods (always clean, do not have a length prefix array).
+    kBinRuntimeMethod,
     // Dex cache arrays have a special slot for PC-relative addressing. Since they are
     // huge, and as such their dirtiness is not important for the clean/dirty separation,
     // we arbitrarily keep them at the end of the native data.
@@ -186,6 +190,8 @@ class ImageWriter FINAL {
     kNativeObjectRelocationTypeArtMethodArrayClean,
     kNativeObjectRelocationTypeArtMethodDirty,
     kNativeObjectRelocationTypeArtMethodArrayDirty,
+    kNativeObjectRelocationTypeRuntimeMethod,
+    kNativeObjectRelocationTypeIMTConflictTable,
     kNativeObjectRelocationTypeDexCacheArray,
   };
   friend std::ostream& operator<<(std::ostream& stream, const NativeObjectRelocationType& type);
@@ -240,7 +246,7 @@ class ImageWriter FINAL {
 
     // Create the image sections into the out sections variable, returns the size of the image
     // excluding the bitmap.
-    size_t CreateImageSections(size_t target_ptr_size, ImageSection* out_sections) const;
+    size_t CreateImageSections(ImageSection* out_sections) const;
 
     std::unique_ptr<MemMap> image_;  // Memory mapped for generating the image.
 
@@ -395,6 +401,8 @@ class ImageWriter FINAL {
   void CopyAndFixupObject(mirror::Object* obj) SHARED_REQUIRES(Locks::mutator_lock_);
   void CopyAndFixupMethod(ArtMethod* orig, ArtMethod* copy, const ImageInfo& image_info)
       SHARED_REQUIRES(Locks::mutator_lock_);
+  void CopyAndFixupImtConflictTable(ImtConflictTable* orig, ImtConflictTable* copy)
+      SHARED_REQUIRES(Locks::mutator_lock_);
   void FixupClass(mirror::Class* orig, mirror::Class* copy)
       SHARED_REQUIRES(Locks::mutator_lock_);
   void FixupObject(mirror::Object* orig, mirror::Object* copy)
@@ -425,6 +433,11 @@ class ImageWriter FINAL {
                           size_t oat_index)
       SHARED_REQUIRES(Locks::mutator_lock_);
 
+  // Assign the offset for an IMT conflict table. Does nothing if the table already has a native
+  // relocation.
+  void TryAssignConflictTableOffset(ImtConflictTable* table, size_t oat_index)
+      SHARED_REQUIRES(Locks::mutator_lock_);
+
   // Return true if klass is loaded by the boot class loader but not in the boot image.
   bool IsBootClassLoaderNonImageClass(mirror::Class* klass) SHARED_REQUIRES(Locks::mutator_lock_);
 
@@ -481,6 +494,9 @@ class ImageWriter FINAL {
   // remove duplicates in the multi image and app image case.
   mirror::String* FindInternedString(mirror::String* string) SHARED_REQUIRES(Locks::mutator_lock_);
 
+  // Return true if there already exists a native allocation for an object.
+  bool NativeRelocationAssigned(void* ptr) const;
+
   const CompilerDriver& compiler_driver_;
 
   // Beginning target image address for the first image.
@@ -517,16 +533,14 @@ class ImageWriter FINAL {
 
     bool IsArtMethodRelocation() const {
       return type == kNativeObjectRelocationTypeArtMethodClean ||
-          type == kNativeObjectRelocationTypeArtMethodDirty;
+          type == kNativeObjectRelocationTypeArtMethodDirty ||
+          type == kNativeObjectRelocationTypeRuntimeMethod;
     }
   };
   std::unordered_map<void*, NativeObjectRelocation> native_object_relocations_;
 
   // Runtime ArtMethods which aren't reachable from any Class but need to be copied into the image.
   ArtMethod* image_methods_[ImageHeader::kImageMethodsCount];
-  // Fake length prefixed array for image methods. This array does not contain the actual
-  // ArtMethods. We only use it for the header and relocation addresses.
-  LengthPrefixedArray<ArtMethod> image_method_array_;
 
   // Counters for measurements, used for logging only.
   uint64_t dirty_methods_;
diff --git a/compiler/jit/jit_compiler.cc b/compiler/jit/jit_compiler.cc
index 5de9842d9..178533849 100644
--- a/compiler/jit/jit_compiler.cc
+++ b/compiler/jit/jit_compiler.cc
@@ -88,7 +88,7 @@ NO_RETURN static void Usage(const char* fmt, ...) {
 
 JitCompiler::JitCompiler() {
   compiler_options_.reset(new CompilerOptions(
-      CompilerOptions::kDefaultCompilerFilter,
+      CompilerFilter::kDefaultCompilerFilter,
       CompilerOptions::kDefaultHugeMethodThreshold,
       CompilerOptions::kDefaultLargeMethodThreshold,
       CompilerOptions::kDefaultSmallMethodThreshold,
@@ -155,7 +155,8 @@ JitCompiler::JitCompiler() {
       Compiler::kOptimizing,
       instruction_set,
       instruction_set_features_.get(),
-      /* image */ false,
+      /* boot_image */ false,
+      /* app_image */ false,
       /* image_classes */ nullptr,
       /* compiled_classes */ nullptr,
       /* compiled_methods */ nullptr,
diff --git a/compiler/jni/jni_compiler_test.cc b/compiler/jni/jni_compiler_test.cc
index cf836a9c9..c4c2399cc 100644
--- a/compiler/jni/jni_compiler_test.cc
+++ b/compiler/jni/jni_compiler_test.cc
@@ -31,6 +31,7 @@
 #include "mirror/object_array-inl.h"
 #include "mirror/object-inl.h"
 #include "mirror/stack_trace_element.h"
+#include "nativeloader/native_loader.h"
 #include "runtime.h"
 #include "ScopedLocalRef.h"
 #include "scoped_thread_state_change.h"
@@ -53,6 +54,11 @@ class JniCompilerTest : public CommonCompilerTest {
     check_generic_jni_ = false;
   }
 
+  void TearDown() OVERRIDE {
+    android::ResetNativeLoader();
+    CommonCompilerTest::TearDown();
+  }
+
   void SetCheckGenericJni(bool generic) {
     check_generic_jni_ = generic;
   }
@@ -92,11 +98,13 @@ class JniCompilerTest : public CommonCompilerTest {
       CompileForTest(class_loader_, direct, method_name, method_sig);
       // Start runtime.
       Thread::Current()->TransitionFromSuspendedToRunnable();
+      android::InitializeNativeLoader();
       bool started = runtime_->Start();
       CHECK(started);
     }
     // JNI operations after runtime start.
     env_ = Thread::Current()->GetJniEnv();
+    library_search_path_ = env_->NewStringUTF("");
     jklass_ = env_->FindClass("MyClassNatives");
     ASSERT_TRUE(jklass_ != nullptr) << method_name << " " << method_sig;
 
@@ -168,6 +176,7 @@ class JniCompilerTest : public CommonCompilerTest {
   void StackArgsSignExtendedMips64Impl();
 
   JNIEnv* env_;
+  jstring library_search_path_;
   jmethodID jmethod_;
   bool check_generic_jni_;
 };
@@ -220,7 +229,7 @@ void JniCompilerTest::CompileAndRunIntMethodThroughStubImpl() {
 
   std::string reason;
   ASSERT_TRUE(Runtime::Current()->GetJavaVM()->
-                  LoadNativeLibrary(env_, "", class_loader_, nullptr, &reason))
+                  LoadNativeLibrary(env_, "", class_loader_, library_search_path_, &reason))
       << reason;
 
   jint result = env_->CallNonvirtualIntMethod(jobj_, jklass_, jmethod_, 24);
@@ -235,7 +244,7 @@ void JniCompilerTest::CompileAndRunStaticIntMethodThroughStubImpl() {
 
   std::string reason;
   ASSERT_TRUE(Runtime::Current()->GetJavaVM()->
-                  LoadNativeLibrary(env_, "", class_loader_, nullptr, &reason))
+                  LoadNativeLibrary(env_, "", class_loader_, library_search_path_, &reason))
       << reason;
 
   jint result = env_->CallStaticIntMethod(jklass_, jmethod_, 42);
diff --git a/compiler/linker/relative_patcher_test.h b/compiler/linker/relative_patcher_test.h
index c07de7998..ec69107d8 100644
--- a/compiler/linker/relative_patcher_test.h
+++ b/compiler/linker/relative_patcher_test.h
@@ -51,6 +51,7 @@ class RelativePatcherTest : public testing::Test {
                 instruction_set,
                 /* instruction_set_features*/ nullptr,
                 /* boot_image */ false,
+                /* app_image */ false,
                 /* image_classes */ nullptr,
                 /* compiled_classes */ nullptr,
                 /* compiled_methods */ nullptr,
diff --git a/compiler/oat_test.cc b/compiler/oat_test.cc
index 73b16d5b4..5b192846b 100644
--- a/compiler/oat_test.cc
+++ b/compiler/oat_test.cc
@@ -112,6 +112,7 @@ class OatTest : public CommonCompilerTest {
                                               insn_set,
                                               insn_features_.get(),
                                               /* boot_image */ false,
+                                              /* app_image */ false,
                                               /* image_classes */ nullptr,
                                               /* compiled_classes */ nullptr,
                                               /* compiled_methods */ nullptr,
diff --git a/compiler/oat_writer.cc b/compiler/oat_writer.cc
index e804beef0..8da9f06dd 100644
--- a/compiler/oat_writer.cc
+++ b/compiler/oat_writer.cc
@@ -1127,17 +1127,23 @@ class OatWriter::WriteCodeMethodVisitor : public OatDexMethodVisitor {
     return target_offset;
   }
 
-  mirror::Class* GetTargetType(const LinkerPatch& patch) SHARED_REQUIRES(Locks::mutator_lock_) {
-    mirror::DexCache* dex_cache = (dex_file_ == patch.TargetTypeDexFile())
+  mirror::DexCache* GetDexCache(const DexFile* target_dex_file)
+      SHARED_REQUIRES(Locks::mutator_lock_) {
+    return (target_dex_file == dex_file_)
         ? dex_cache_
-        : class_linker_->FindDexCache(Thread::Current(), *patch.TargetTypeDexFile());
+        : class_linker_->FindDexCache(Thread::Current(), *target_dex_file);
+  }
+
+  mirror::Class* GetTargetType(const LinkerPatch& patch) SHARED_REQUIRES(Locks::mutator_lock_) {
+    mirror::DexCache* dex_cache = GetDexCache(patch.TargetTypeDexFile());
     mirror::Class* type = dex_cache->GetResolvedType(patch.TargetTypeIndex());
     CHECK(type != nullptr);
     return type;
   }
 
   mirror::String* GetTargetString(const LinkerPatch& patch) SHARED_REQUIRES(Locks::mutator_lock_) {
-    mirror::String* string = dex_cache_->GetResolvedString(patch.TargetStringIndex());
+    mirror::DexCache* dex_cache = GetDexCache(patch.TargetStringDexFile());
+    mirror::String* string = dex_cache->GetResolvedString(patch.TargetStringIndex());
     DCHECK(string != nullptr);
     DCHECK(writer_->HasBootImage() ||
            Runtime::Current()->GetHeap()->ObjectIsInBootImageSpace(string));
diff --git a/compiler/optimizing/bounds_check_elimination.cc b/compiler/optimizing/bounds_check_elimination.cc
index 659c6f849..6c6e5af0b 100644
--- a/compiler/optimizing/bounds_check_elimination.cc
+++ b/compiler/optimizing/bounds_check_elimination.cc
@@ -552,7 +552,11 @@ class BCEVisitor : public HGraphVisitor {
     DCHECK(!IsAddedBlock(block));
     first_index_bounds_check_map_.clear();
     HGraphVisitor::VisitBasicBlock(block);
-    AddComparesWithDeoptimization(block);
+    // We should never deoptimize from an osr method, otherwise we might wrongly optimize
+    // code dominated by the deoptimization.
+    if (!GetGraph()->IsCompilingOsr()) {
+      AddComparesWithDeoptimization(block);
+    }
   }
 
   void Finish() {
@@ -796,6 +800,27 @@ class BCEVisitor : public HGraphVisitor {
             ValueRange(GetGraph()->GetArena(), ValueBound::Min(), new_upper);
         ApplyRangeFromComparison(left, block, false_successor, new_range);
       }
+    } else if (cond == kCondNE || cond == kCondEQ) {
+      if (left->IsArrayLength() && lower.IsConstant() && upper.IsConstant()) {
+        // Special case:
+        //   length == [c,d] yields [c, d] along true
+        //   length != [c,d] yields [c, d] along false
+        if (!lower.Equals(ValueBound::Min()) || !upper.Equals(ValueBound::Max())) {
+          ValueRange* new_range = new (GetGraph()->GetArena())
+              ValueRange(GetGraph()->GetArena(), lower, upper);
+          ApplyRangeFromComparison(
+              left, block, cond == kCondEQ ? true_successor : false_successor, new_range);
+        }
+        // In addition:
+        //   length == 0 yields [1, max] along false
+        //   length != 0 yields [1, max] along true
+        if (lower.GetConstant() == 0 && upper.GetConstant() == 0) {
+          ValueRange* new_range = new (GetGraph()->GetArena())
+              ValueRange(GetGraph()->GetArena(), ValueBound(nullptr, 1), ValueBound::Max());
+          ApplyRangeFromComparison(
+              left, block, cond == kCondEQ ? false_successor : true_successor, new_range);
+        }
+      }
     }
   }
 
@@ -951,13 +976,7 @@ class BCEVisitor : public HGraphVisitor {
   void VisitIf(HIf* instruction) OVERRIDE {
     if (instruction->InputAt(0)->IsCondition()) {
       HCondition* cond = instruction->InputAt(0)->AsCondition();
-      IfCondition cmp = cond->GetCondition();
-      if (cmp == kCondGT || cmp == kCondGE ||
-          cmp == kCondLT || cmp == kCondLE) {
-        HInstruction* left = cond->GetLeft();
-        HInstruction* right = cond->GetRight();
-        HandleIf(instruction, left, right, cmp);
-      }
+      HandleIf(instruction, cond->GetLeft(), cond->GetRight(), cond->GetCondition());
     }
   }
 
@@ -1358,6 +1377,11 @@ class BCEVisitor : public HGraphVisitor {
       if (loop->IsIrreducible()) {
         return false;
       }
+      // We should never deoptimize from an osr method, otherwise we might wrongly optimize
+      // code dominated by the deoptimization.
+      if (GetGraph()->IsCompilingOsr()) {
+        return false;
+      }
       // A try boundary preheader is hard to handle.
       // TODO: remove this restriction.
       if (loop->GetPreHeader()->GetLastInstruction()->IsTryBoundary()) {
diff --git a/compiler/optimizing/code_generator.cc b/compiler/optimizing/code_generator.cc
index e7fa4e472..51fbaea51 100644
--- a/compiler/optimizing/code_generator.cc
+++ b/compiler/optimizing/code_generator.cc
@@ -50,6 +50,7 @@
 #include "mirror/array-inl.h"
 #include "mirror/object_array-inl.h"
 #include "mirror/object_reference.h"
+#include "mirror/string.h"
 #include "parallel_move_resolver.h"
 #include "ssa_liveness_analysis.h"
 #include "utils/assembler.h"
@@ -139,6 +140,12 @@ size_t CodeGenerator::GetCachePointerOffset(uint32_t index) {
   return pointer_size * index;
 }
 
+uint32_t CodeGenerator::GetArrayLengthOffset(HArrayLength* array_length) {
+  return array_length->IsStringLength()
+      ? mirror::String::CountOffset().Uint32Value()
+      : mirror::Array::LengthOffset().Uint32Value();
+}
+
 bool CodeGenerator::GoesToNextBlock(HBasicBlock* current, HBasicBlock* next) const {
   DCHECK_EQ((*block_order_)[current_block_index_], current);
   return GetNextBlockToEmit() == FirstNonEmptyBlock(next);
diff --git a/compiler/optimizing/code_generator.h b/compiler/optimizing/code_generator.h
index d69c41055..6e75e3bb2 100644
--- a/compiler/optimizing/code_generator.h
+++ b/compiler/optimizing/code_generator.h
@@ -340,6 +340,11 @@ class CodeGenerator : public DeletableArenaObject<kArenaAllocCodeGenerator> {
   // Pointer variant for ArtMethod and ArtField arrays.
   size_t GetCachePointerOffset(uint32_t index);
 
+  // Helper that returns the offset of the array's length field.
+  // Note: Besides the normal arrays, we also use the HArrayLength for
+  // accessing the String's `count` field in String intrinsics.
+  static uint32_t GetArrayLengthOffset(HArrayLength* array_length);
+
   void EmitParallelMoves(Location from1,
                          Location to1,
                          Primitive::Type type1,
diff --git a/compiler/optimizing/code_generator_arm.cc b/compiler/optimizing/code_generator_arm.cc
index 45d23fe51..e0106628c 100644
--- a/compiler/optimizing/code_generator_arm.cc
+++ b/compiler/optimizing/code_generator_arm.cc
@@ -4742,7 +4742,7 @@ void LocationsBuilderARM::VisitArrayLength(HArrayLength* instruction) {
 
 void InstructionCodeGeneratorARM::VisitArrayLength(HArrayLength* instruction) {
   LocationSummary* locations = instruction->GetLocations();
-  uint32_t offset = mirror::Array::LengthOffset().Uint32Value();
+  uint32_t offset = CodeGenerator::GetArrayLengthOffset(instruction);
   Register obj = locations->InAt(0).AsRegister<Register>();
   Register out = locations->Out().AsRegister<Register>();
   __ LoadFromOffset(kLoadWord, out, obj, offset);
@@ -5183,10 +5183,10 @@ HLoadString::LoadKind CodeGeneratorARM::GetSupportedLoadStringKind(
     case HLoadString::LoadKind::kBootImageAddress:
       break;
     case HLoadString::LoadKind::kDexCacheAddress:
-      DCHECK(Runtime::Current()->UseJit());
+      DCHECK(Runtime::Current()->UseJitCompilation());
       break;
     case HLoadString::LoadKind::kDexCachePcRelative:
-      DCHECK(!Runtime::Current()->UseJit());
+      DCHECK(!Runtime::Current()->UseJitCompilation());
       // 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
diff --git a/compiler/optimizing/code_generator_arm64.cc b/compiler/optimizing/code_generator_arm64.cc
index e8e6b6897..261c04f06 100644
--- a/compiler/optimizing/code_generator_arm64.cc
+++ b/compiler/optimizing/code_generator_arm64.cc
@@ -2118,9 +2118,9 @@ void LocationsBuilderARM64::VisitArrayLength(HArrayLength* instruction) {
 }
 
 void InstructionCodeGeneratorARM64::VisitArrayLength(HArrayLength* instruction) {
+  uint32_t offset = CodeGenerator::GetArrayLengthOffset(instruction);
   BlockPoolsScope block_pools(GetVIXLAssembler());
-  __ Ldr(OutputRegister(instruction),
-         HeapOperand(InputRegisterAt(instruction, 0), mirror::Array::LengthOffset()));
+  __ Ldr(OutputRegister(instruction), HeapOperand(InputRegisterAt(instruction, 0), offset));
   codegen_->MaybeRecordImplicitNullCheck(instruction);
 }
 
@@ -4010,10 +4010,10 @@ HLoadString::LoadKind CodeGeneratorARM64::GetSupportedLoadStringKind(
     case HLoadString::LoadKind::kBootImageAddress:
       break;
     case HLoadString::LoadKind::kDexCacheAddress:
-      DCHECK(Runtime::Current()->UseJit());
+      DCHECK(Runtime::Current()->UseJitCompilation());
       break;
     case HLoadString::LoadKind::kDexCachePcRelative:
-      DCHECK(!Runtime::Current()->UseJit());
+      DCHECK(!Runtime::Current()->UseJitCompilation());
       break;
     case HLoadString::LoadKind::kDexCacheViaMethod:
       break;
diff --git a/compiler/optimizing/code_generator_mips.cc b/compiler/optimizing/code_generator_mips.cc
index 12d1164d0..fb50680c9 100644
--- a/compiler/optimizing/code_generator_mips.cc
+++ b/compiler/optimizing/code_generator_mips.cc
@@ -1803,7 +1803,7 @@ void LocationsBuilderMIPS::VisitArrayLength(HArrayLength* instruction) {
 
 void InstructionCodeGeneratorMIPS::VisitArrayLength(HArrayLength* instruction) {
   LocationSummary* locations = instruction->GetLocations();
-  uint32_t offset = mirror::Array::LengthOffset().Uint32Value();
+  uint32_t offset = CodeGenerator::GetArrayLengthOffset(instruction);
   Register obj = locations->InAt(0).AsRegister<Register>();
   Register out = locations->Out().AsRegister<Register>();
   __ LoadFromOffset(kLoadWord, out, obj, offset);
diff --git a/compiler/optimizing/code_generator_mips64.cc b/compiler/optimizing/code_generator_mips64.cc
index 56ac38ef8..e67d8d0dc 100644
--- a/compiler/optimizing/code_generator_mips64.cc
+++ b/compiler/optimizing/code_generator_mips64.cc
@@ -1426,7 +1426,7 @@ void LocationsBuilderMIPS64::VisitArrayLength(HArrayLength* instruction) {
 
 void InstructionCodeGeneratorMIPS64::VisitArrayLength(HArrayLength* instruction) {
   LocationSummary* locations = instruction->GetLocations();
-  uint32_t offset = mirror::Array::LengthOffset().Uint32Value();
+  uint32_t offset = CodeGenerator::GetArrayLengthOffset(instruction);
   GpuRegister obj = locations->InAt(0).AsRegister<GpuRegister>();
   GpuRegister out = locations->Out().AsRegister<GpuRegister>();
   __ LoadFromOffset(kLoadWord, out, obj, offset);
diff --git a/compiler/optimizing/code_generator_x86.cc b/compiler/optimizing/code_generator_x86.cc
index e73e88030..50892a9d4 100644
--- a/compiler/optimizing/code_generator_x86.cc
+++ b/compiler/optimizing/code_generator_x86.cc
@@ -3303,17 +3303,6 @@ void InstructionCodeGeneratorX86::GenerateDivRemWithAnyConstant(HBinaryOperation
   int shift;
   CalculateMagicAndShiftForDivRem(imm, false /* is_long */, &magic, &shift);
 
-  NearLabel ndiv;
-  NearLabel end;
-  // If numerator is 0, the result is 0, no computation needed.
-  __ testl(eax, eax);
-  __ j(kNotEqual, &ndiv);
-
-  __ xorl(out, out);
-  __ jmp(&end);
-
-  __ Bind(&ndiv);
-
   // Save the numerator.
   __ movl(num, eax);
 
@@ -3348,7 +3337,6 @@ void InstructionCodeGeneratorX86::GenerateDivRemWithAnyConstant(HBinaryOperation
   } else {
     __ movl(eax, edx);
   }
-  __ Bind(&end);
 }
 
 void InstructionCodeGeneratorX86::GenerateDivRemIntegral(HBinaryOperation* instruction) {
@@ -5492,7 +5480,7 @@ void LocationsBuilderX86::VisitArrayLength(HArrayLength* instruction) {
 
 void InstructionCodeGeneratorX86::VisitArrayLength(HArrayLength* instruction) {
   LocationSummary* locations = instruction->GetLocations();
-  uint32_t offset = mirror::Array::LengthOffset().Uint32Value();
+  uint32_t offset = CodeGenerator::GetArrayLengthOffset(instruction);
   Register obj = locations->InAt(0).AsRegister<Register>();
   Register out = locations->Out().AsRegister<Register>();
   __ movl(out, Address(obj, offset));
@@ -5977,7 +5965,7 @@ HLoadString::LoadKind CodeGeneratorX86::GetSupportedLoadStringKind(
       DCHECK(GetCompilerOptions().GetCompilePic());
       FALLTHROUGH_INTENDED;
     case HLoadString::LoadKind::kDexCachePcRelative:
-      DCHECK(!Runtime::Current()->UseJit());  // Note: boot image is also non-JIT.
+      DCHECK(!Runtime::Current()->UseJitCompilation());  // Note: boot image is also non-JIT.
       // We disable pc-relative load when there is an irreducible loop, as the optimization
       // is incompatible with it.
       // TODO: Create as many X86ComputeBaseMethodAddress instructions as needed for methods
@@ -5989,7 +5977,7 @@ HLoadString::LoadKind CodeGeneratorX86::GetSupportedLoadStringKind(
     case HLoadString::LoadKind::kBootImageAddress:
       break;
     case HLoadString::LoadKind::kDexCacheAddress:
-      DCHECK(Runtime::Current()->UseJit());
+      DCHECK(Runtime::Current()->UseJitCompilation());
       break;
     case HLoadString::LoadKind::kDexCacheViaMethod:
       break;
diff --git a/compiler/optimizing/code_generator_x86_64.cc b/compiler/optimizing/code_generator_x86_64.cc
index 5576d839c..56c5b0694 100644
--- a/compiler/optimizing/code_generator_x86_64.cc
+++ b/compiler/optimizing/code_generator_x86_64.cc
@@ -3390,16 +3390,6 @@ void InstructionCodeGeneratorX86_64::GenerateDivRemWithAnyConstant(HBinaryOperat
 
     __ movl(numerator, eax);
 
-    NearLabel no_div;
-    NearLabel end;
-    __ testl(eax, eax);
-    __ j(kNotEqual, &no_div);
-
-    __ xorl(out, out);
-    __ jmp(&end);
-
-    __ Bind(&no_div);
-
     __ movl(eax, Immediate(magic));
     __ imull(numerator);
 
@@ -3425,7 +3415,6 @@ void InstructionCodeGeneratorX86_64::GenerateDivRemWithAnyConstant(HBinaryOperat
     } else {
       __ movl(eax, edx);
     }
-    __ Bind(&end);
   } else {
     int64_t imm = second.GetConstant()->AsLongConstant()->GetValue();
 
@@ -4967,7 +4956,7 @@ void LocationsBuilderX86_64::VisitArrayLength(HArrayLength* instruction) {
 
 void InstructionCodeGeneratorX86_64::VisitArrayLength(HArrayLength* instruction) {
   LocationSummary* locations = instruction->GetLocations();
-  uint32_t offset = mirror::Array::LengthOffset().Uint32Value();
+  uint32_t offset = CodeGenerator::GetArrayLengthOffset(instruction);
   CpuRegister obj = locations->InAt(0).AsRegister<CpuRegister>();
   CpuRegister out = locations->Out().AsRegister<CpuRegister>();
   __ movl(out, Address(obj, offset));
@@ -5413,10 +5402,10 @@ HLoadString::LoadKind CodeGeneratorX86_64::GetSupportedLoadStringKind(
     case HLoadString::LoadKind::kBootImageAddress:
       break;
     case HLoadString::LoadKind::kDexCacheAddress:
-      DCHECK(Runtime::Current()->UseJit());
+      DCHECK(Runtime::Current()->UseJitCompilation());
       break;
     case HLoadString::LoadKind::kDexCachePcRelative:
-      DCHECK(!Runtime::Current()->UseJit());
+      DCHECK(!Runtime::Current()->UseJitCompilation());
       break;
     case HLoadString::LoadKind::kDexCacheViaMethod:
       break;
diff --git a/compiler/optimizing/dead_code_elimination.cc b/compiler/optimizing/dead_code_elimination.cc
index 5f1102499..49cfff46d 100644
--- a/compiler/optimizing/dead_code_elimination.cc
+++ b/compiler/optimizing/dead_code_elimination.cc
@@ -23,7 +23,7 @@
 namespace art {
 
 static void MarkReachableBlocks(HGraph* graph, ArenaBitVector* visited) {
-  ArenaVector<HBasicBlock*> worklist(graph->GetArena()->Adapter());
+  ArenaVector<HBasicBlock*> worklist(graph->GetArena()->Adapter(kArenaAllocDCE));
   constexpr size_t kDefaultWorlistSize = 8;
   worklist.reserve(kDefaultWorlistSize);
   visited->SetBit(graph->GetEntryBlock()->GetBlockId());
diff --git a/compiler/optimizing/graph_checker.cc b/compiler/optimizing/graph_checker.cc
index 96837a826..968e26724 100644
--- a/compiler/optimizing/graph_checker.cc
+++ b/compiler/optimizing/graph_checker.cc
@@ -258,6 +258,15 @@ void GraphChecker::VisitBoundsCheck(HBoundsCheck* check) {
   VisitInstruction(check);
 }
 
+void GraphChecker::VisitDeoptimize(HDeoptimize* deopt) {
+  if (GetGraph()->IsCompilingOsr()) {
+    AddError(StringPrintf("A graph compiled OSR cannot have a HDeoptimize instruction"));
+  }
+
+  // Perform the instruction base checks too.
+  VisitInstruction(deopt);
+}
+
 void GraphChecker::VisitTryBoundary(HTryBoundary* try_boundary) {
   ArrayRef<HBasicBlock* const> handlers = try_boundary->GetExceptionHandlers();
 
diff --git a/compiler/optimizing/graph_checker.h b/compiler/optimizing/graph_checker.h
index 83b198474..3060c8007 100644
--- a/compiler/optimizing/graph_checker.h
+++ b/compiler/optimizing/graph_checker.h
@@ -57,6 +57,7 @@ class GraphChecker : public HGraphDelegateVisitor {
   void VisitCheckCast(HCheckCast* check) OVERRIDE;
   void VisitCondition(HCondition* op) OVERRIDE;
   void VisitConstant(HConstant* instruction) OVERRIDE;
+  void VisitDeoptimize(HDeoptimize* instruction) OVERRIDE;
   void VisitIf(HIf* instruction) OVERRIDE;
   void VisitInstanceOf(HInstanceOf* check) OVERRIDE;
   void VisitInvokeStaticOrDirect(HInvokeStaticOrDirect* invoke) OVERRIDE;
diff --git a/compiler/optimizing/graph_visualizer.cc b/compiler/optimizing/graph_visualizer.cc
index 9efc13f61..6aec46354 100644
--- a/compiler/optimizing/graph_visualizer.cc
+++ b/compiler/optimizing/graph_visualizer.cc
@@ -98,7 +98,9 @@ typedef Disassembler* create_disasm_prototype(InstructionSet instruction_set,
                                               DisassemblerOptions* options);
 class HGraphVisualizerDisassembler {
  public:
-  HGraphVisualizerDisassembler(InstructionSet instruction_set, const uint8_t* base_address)
+  HGraphVisualizerDisassembler(InstructionSet instruction_set,
+                               const uint8_t* base_address,
+                               const uint8_t* end_address)
       : instruction_set_(instruction_set), disassembler_(nullptr) {
     libart_disassembler_handle_ =
         dlopen(kIsDebugBuild ? "libartd-disassembler.so" : "libart-disassembler.so", RTLD_NOW);
@@ -119,6 +121,7 @@ class HGraphVisualizerDisassembler {
             instruction_set,
             new DisassemblerOptions(/* absolute_addresses */ false,
                                     base_address,
+                                    end_address,
                                     /* can_read_literals */ true)));
   }
 
@@ -174,7 +177,9 @@ class HGraphVisualizerPrinter : public HGraphDelegateVisitor {
         disassembler_(disasm_info_ != nullptr
                       ? new HGraphVisualizerDisassembler(
                             codegen_.GetInstructionSet(),
-                            codegen_.GetAssembler().CodeBufferBaseAddress())
+                            codegen_.GetAssembler().CodeBufferBaseAddress(),
+                            codegen_.GetAssembler().CodeBufferBaseAddress()
+                                + codegen_.GetAssembler().CodeSize())
                       : nullptr),
         indent_(0) {}
 
@@ -389,6 +394,11 @@ class HGraphVisualizerPrinter : public HGraphDelegateVisitor {
         << instance_of->MustDoNullCheck() << std::noboolalpha;
   }
 
+  void VisitArrayLength(HArrayLength* array_length) OVERRIDE {
+    StartAttributeStream("is_string_length") << std::boolalpha
+        << array_length->IsStringLength() << std::noboolalpha;
+  }
+
   void VisitArraySet(HArraySet* array_set) OVERRIDE {
     StartAttributeStream("value_can_be_null") << std::boolalpha
         << array_set->GetValueCanBeNull() << std::noboolalpha;
@@ -544,26 +554,19 @@ class HGraphVisualizerPrinter : public HGraphDelegateVisitor {
       }
     }
 
-    if (IsPass(LICM::kLoopInvariantCodeMotionPassName)
-        || IsPass(HDeadCodeElimination::kFinalDeadCodeEliminationPassName)
-        || IsPass(HDeadCodeElimination::kInitialDeadCodeEliminationPassName)
-        || IsPass(BoundsCheckElimination::kBoundsCheckEliminationPassName)
-        || IsPass(RegisterAllocator::kRegisterAllocatorPassName)
-        || IsPass(HGraphBuilder::kBuilderPassName)) {
-      HLoopInformation* info = instruction->GetBlock()->GetLoopInformation();
-      if (info == nullptr) {
-        StartAttributeStream("loop") << "none";
+    HLoopInformation* loop_info = instruction->GetBlock()->GetLoopInformation();
+    if (loop_info == nullptr) {
+      StartAttributeStream("loop") << "none";
+    } else {
+      StartAttributeStream("loop") << "B" << loop_info->GetHeader()->GetBlockId();
+      HLoopInformation* outer = loop_info->GetPreHeader()->GetLoopInformation();
+      if (outer != nullptr) {
+        StartAttributeStream("outer_loop") << "B" << outer->GetHeader()->GetBlockId();
       } else {
-        StartAttributeStream("loop") << "B" << info->GetHeader()->GetBlockId();
-        HLoopInformation* outer = info->GetPreHeader()->GetLoopInformation();
-        if (outer != nullptr) {
-          StartAttributeStream("outer_loop") << "B" << outer->GetHeader()->GetBlockId();
-        } else {
-          StartAttributeStream("outer_loop") << "none";
-        }
-        StartAttributeStream("irreducible")
-            << std::boolalpha << info->IsIrreducible() << std::noboolalpha;
+        StartAttributeStream("outer_loop") << "none";
       }
+      StartAttributeStream("irreducible")
+          << std::boolalpha << loop_info->IsIrreducible() << std::noboolalpha;
     }
 
     if ((IsPass(HGraphBuilder::kBuilderPassName)
diff --git a/compiler/optimizing/gvn.cc b/compiler/optimizing/gvn.cc
index d0d52bf6c..1e86b7507 100644
--- a/compiler/optimizing/gvn.cc
+++ b/compiler/optimizing/gvn.cc
@@ -454,11 +454,16 @@ void GlobalValueNumberer::VisitBasicBlock(HBasicBlock* block) {
 
     if (!set->IsEmpty()) {
       if (block->IsLoopHeader()) {
-        if (block->GetLoopInformation()->IsIrreducible()) {
+        if (block->GetLoopInformation()->ContainsIrreducibleLoop()) {
           // To satisfy our linear scan algorithm, no instruction should flow in an irreducible
-          // loop header.
+          // loop header. We clear the set at entry of irreducible loops and any loop containing
+          // an irreducible loop, as in both cases, GVN can extend the liveness of an instruction
+          // across the irreducible loop.
+          // Note that, if we're not compiling OSR, we could still do GVN and introduce
+          // phis at irreducible loop headers. We decided it was not worth the complexity.
           set->Clear();
         } else {
+          DCHECK(!block->GetLoopInformation()->IsIrreducible());
           DCHECK_EQ(block->GetDominator(), block->GetLoopInformation()->GetPreHeader());
           set->Kill(side_effects_.GetLoopEffects(block));
         }
diff --git a/compiler/optimizing/inliner.cc b/compiler/optimizing/inliner.cc
index ff4b9a765..59de89518 100644
--- a/compiler/optimizing/inliner.cc
+++ b/compiler/optimizing/inliner.cc
@@ -308,7 +308,7 @@ bool HInliner::TryInline(HInvoke* invoke_instruction) {
 
   // Check if we can use an inline cache.
   ArtMethod* caller = graph_->GetArtMethod();
-  if (Runtime::Current()->UseJit()) {
+  if (Runtime::Current()->UseJitCompilation()) {
     // Under JIT, we should always know the caller.
     DCHECK(caller != nullptr);
     ScopedProfilingInfoInlineUse spiis(caller, soa.Self());
@@ -322,7 +322,13 @@ bool HInliner::TryInline(HInvoke* invoke_instruction) {
         return false;
       } else if (ic.IsMonomorphic()) {
         MaybeRecordStat(kMonomorphicCall);
-        return TryInlineMonomorphicCall(invoke_instruction, resolved_method, ic);
+        if (outermost_graph_->IsCompilingOsr()) {
+          // If we are compiling OSR, we pretend this call is polymorphic, as we may come from the
+          // interpreter and it may have seen different receiver types.
+          return TryInlinePolymorphicCall(invoke_instruction, resolved_method, ic);
+        } else {
+          return TryInlineMonomorphicCall(invoke_instruction, resolved_method, ic);
+        }
       } else if (ic.IsPolymorphic()) {
         MaybeRecordStat(kPolymorphicCall);
         return TryInlinePolymorphicCall(invoke_instruction, resolved_method, ic);
@@ -510,6 +516,11 @@ bool HInliner::TryInlinePolymorphicCall(HInvoke* invoke_instruction,
       bool deoptimize = all_targets_inlined &&
           (i != InlineCache::kIndividualCacheSize - 1) &&
           (ic.GetTypeAt(i + 1) == nullptr);
+
+      if (outermost_graph_->IsCompilingOsr()) {
+        // We do not support HDeoptimize in OSR methods.
+        deoptimize = false;
+      }
       HInstruction* compare = AddTypeGuard(
           receiver, cursor, bb_cursor, class_index, is_referrer, invoke_instruction, deoptimize);
       if (deoptimize) {
@@ -623,7 +634,7 @@ bool HInliner::TryInlinePolymorphicCallToSameTarget(HInvoke* invoke_instruction,
                                                     ArtMethod* resolved_method,
                                                     const InlineCache& ic) {
   // This optimization only works under JIT for now.
-  DCHECK(Runtime::Current()->UseJit());
+  DCHECK(Runtime::Current()->UseJitCompilation());
   if (graph_->GetInstructionSet() == kMips64) {
     // TODO: Support HClassTableGet for mips64.
     return false;
@@ -672,7 +683,8 @@ bool HInliner::TryInlinePolymorphicCallToSameTarget(HInvoke* invoke_instruction,
   HInstruction* cursor = invoke_instruction->GetPrevious();
   HBasicBlock* bb_cursor = invoke_instruction->GetBlock();
 
-  if (!TryInlineAndReplace(invoke_instruction, actual_method, /* do_rtp */ false)) {
+  HInstruction* return_replacement = nullptr;
+  if (!TryBuildAndInline(invoke_instruction, actual_method, &return_replacement)) {
     return false;
   }
 
@@ -701,9 +713,6 @@ bool HInliner::TryInlinePolymorphicCallToSameTarget(HInvoke* invoke_instruction,
   }
 
   HNotEqual* compare = new (graph_->GetArena()) HNotEqual(class_table_get, constant);
-  HDeoptimize* deoptimize = new (graph_->GetArena()) HDeoptimize(
-      compare, invoke_instruction->GetDexPc());
-  // TODO: Extend reference type propagation to understand the guard.
   if (cursor != nullptr) {
     bb_cursor->InsertInstructionAfter(receiver_class, cursor);
   } else {
@@ -711,8 +720,20 @@ bool HInliner::TryInlinePolymorphicCallToSameTarget(HInvoke* invoke_instruction,
   }
   bb_cursor->InsertInstructionAfter(class_table_get, receiver_class);
   bb_cursor->InsertInstructionAfter(compare, class_table_get);
-  bb_cursor->InsertInstructionAfter(deoptimize, compare);
-  deoptimize->CopyEnvironmentFrom(invoke_instruction->GetEnvironment());
+
+  if (outermost_graph_->IsCompilingOsr()) {
+    CreateDiamondPatternForPolymorphicInline(compare, return_replacement, invoke_instruction);
+  } else {
+    // TODO: Extend reference type propagation to understand the guard.
+    HDeoptimize* deoptimize = new (graph_->GetArena()) HDeoptimize(
+        compare, invoke_instruction->GetDexPc());
+    bb_cursor->InsertInstructionAfter(deoptimize, compare);
+    deoptimize->CopyEnvironmentFrom(invoke_instruction->GetEnvironment());
+    if (return_replacement != nullptr) {
+      invoke_instruction->ReplaceWith(return_replacement);
+    }
+    invoke_instruction->GetBlock()->RemoveInstruction(invoke_instruction);
+  }
 
   // Run type propagation to get the guard typed.
   ReferenceTypePropagation rtp_fixup(graph_,
@@ -744,6 +765,12 @@ bool HInliner::TryBuildAndInline(HInvoke* invoke_instruction,
                                  HInstruction** return_replacement) {
   const DexFile& caller_dex_file = *caller_compilation_unit_.GetDexFile();
 
+  if (method->IsProxyMethod()) {
+    VLOG(compiler) << "Method " << PrettyMethod(method)
+                   << " is not inlined because of unimplemented inline support for proxy methods.";
+    return false;
+  }
+
   // Check whether we're allowed to inline. The outermost compilation unit is the relevant
   // dex file here (though the transitivity of an inline chain would allow checking the calller).
   if (!compiler_driver_->MayInline(method->GetDexFile(),
@@ -802,7 +829,7 @@ bool HInliner::TryBuildAndInline(HInvoke* invoke_instruction,
 
   if (!method->GetDeclaringClass()->IsVerified()) {
     uint16_t class_def_idx = method->GetDeclaringClass()->GetDexClassDefIndex();
-    if (Runtime::Current()->UseJit() ||
+    if (Runtime::Current()->UseJitCompilation() ||
         !compiler_driver_->IsMethodVerifiedWithoutFailures(
             method->GetDexMethodIndex(), class_def_idx, *method->GetDexFile())) {
       VLOG(compiler) << "Method " << PrettyMethod(method_index, caller_dex_file)
@@ -1265,6 +1292,8 @@ bool HInliner::TryBuildAndInlineHelper(HInvoke* invoke_instruction,
 size_t HInliner::RunOptimizations(HGraph* callee_graph,
                                   const DexFile::CodeItem* code_item,
                                   const DexCompilationUnit& dex_compilation_unit) {
+  // Note: if the outermost_graph_ is being compiled OSR, we should not run any
+  // optimization that could lead to a HDeoptimize. The following optimizations do not.
   HDeadCodeElimination dce(callee_graph, stats_);
   HConstantFolding fold(callee_graph);
   HSharpening sharpening(callee_graph, codegen_, dex_compilation_unit, compiler_driver_);
diff --git a/compiler/optimizing/instruction_builder.cc b/compiler/optimizing/instruction_builder.cc
index a834216d0..aaddc01f1 100644
--- a/compiler/optimizing/instruction_builder.cc
+++ b/compiler/optimizing/instruction_builder.cc
@@ -721,6 +721,11 @@ ArtMethod* HInstructionBuilder::ResolveMethod(uint16_t method_idx, InvokeType in
       DCHECK(Runtime::Current()->IsAotCompiler());
       return nullptr;
     }
+    if (!methods_class->IsAssignableFrom(compiling_class.Get())) {
+      // We cannot statically determine the target method. The runtime will throw a
+      // NoSuchMethodError on this one.
+      return nullptr;
+    }
     ArtMethod* actual_method;
     if (methods_class->IsInterface()) {
       actual_method = methods_class->FindVirtualMethodForInterfaceSuper(
diff --git a/compiler/optimizing/instruction_simplifier.cc b/compiler/optimizing/instruction_simplifier.cc
index d7b3856bf..fd79901ff 100644
--- a/compiler/optimizing/instruction_simplifier.cc
+++ b/compiler/optimizing/instruction_simplifier.cc
@@ -101,6 +101,7 @@ class InstructionSimplifierVisitor : public HGraphDelegateVisitor {
   void SimplifyCompare(HInvoke* invoke, bool is_signum, Primitive::Type type);
   void SimplifyIsNaN(HInvoke* invoke);
   void SimplifyFP2Int(HInvoke* invoke);
+  void SimplifyStringIsEmptyOrLength(HInvoke* invoke);
   void SimplifyMemBarrier(HInvoke* invoke, MemBarrierKind barrier_kind);
 
   OptimizingCompilerStats* stats_;
@@ -1673,6 +1674,27 @@ void InstructionSimplifierVisitor::SimplifyFP2Int(HInvoke* invoke) {
   invoke->ReplaceWithExceptInReplacementAtIndex(select, 0);  // false at index 0
 }
 
+void InstructionSimplifierVisitor::SimplifyStringIsEmptyOrLength(HInvoke* invoke) {
+  HInstruction* str = invoke->InputAt(0);
+  uint32_t dex_pc = invoke->GetDexPc();
+  // We treat String as an array to allow DCE and BCE to seamlessly work on strings,
+  // so create the HArrayLength.
+  HArrayLength* length = new (GetGraph()->GetArena()) HArrayLength(str, dex_pc);
+  length->MarkAsStringLength();
+  HInstruction* replacement;
+  if (invoke->GetIntrinsic() == Intrinsics::kStringIsEmpty) {
+    // For String.isEmpty(), create the `HEqual` representing the `length == 0`.
+    invoke->GetBlock()->InsertInstructionBefore(length, invoke);
+    HIntConstant* zero = GetGraph()->GetIntConstant(0);
+    HEqual* equal = new (GetGraph()->GetArena()) HEqual(length, zero, dex_pc);
+    replacement = equal;
+  } else {
+    DCHECK_EQ(invoke->GetIntrinsic(), Intrinsics::kStringLength);
+    replacement = length;
+  }
+  invoke->GetBlock()->ReplaceAndRemoveInstructionWith(invoke, replacement);
+}
+
 void InstructionSimplifierVisitor::SimplifyMemBarrier(HInvoke* invoke, MemBarrierKind barrier_kind) {
   uint32_t dex_pc = invoke->GetDexPc();
   HMemoryBarrier* mem_barrier = new (GetGraph()->GetArena()) HMemoryBarrier(barrier_kind, dex_pc);
@@ -1719,6 +1741,10 @@ void InstructionSimplifierVisitor::VisitInvoke(HInvoke* instruction) {
     case Intrinsics::kDoubleDoubleToLongBits:
       SimplifyFP2Int(instruction);
       break;
+    case Intrinsics::kStringIsEmpty:
+    case Intrinsics::kStringLength:
+      SimplifyStringIsEmptyOrLength(instruction);
+      break;
     case Intrinsics::kUnsafeLoadFence:
       SimplifyMemBarrier(instruction, MemBarrierKind::kLoadAny);
       break;
diff --git a/compiler/optimizing/intrinsics.cc b/compiler/optimizing/intrinsics.cc
index 5d4c4e295..418d59c6c 100644
--- a/compiler/optimizing/intrinsics.cc
+++ b/compiler/optimizing/intrinsics.cc
@@ -388,10 +388,8 @@ static Intrinsics GetIntrinsic(InlineMethod method) {
     case kIntrinsicGetCharsNoCheck:
       return Intrinsics::kStringGetCharsNoCheck;
     case kIntrinsicIsEmptyOrLength:
-      // The inliner can handle these two cases - and this is the preferred approach
-      // since after inlining the call is no longer visible (as opposed to waiting
-      // until codegen to handle intrinsic).
-      return Intrinsics::kNone;
+      return ((method.d.data & kIntrinsicFlagIsEmpty) == 0) ?
+          Intrinsics::kStringLength : Intrinsics::kStringIsEmpty;
     case kIntrinsicIndexOf:
       return ((method.d.data & kIntrinsicFlagBase0) == 0) ?
           Intrinsics::kStringIndexOfAfter : Intrinsics::kStringIndexOf;
diff --git a/compiler/optimizing/intrinsics.h b/compiler/optimizing/intrinsics.h
index 863dd1c6f..214250f33 100644
--- a/compiler/optimizing/intrinsics.h
+++ b/compiler/optimizing/intrinsics.h
@@ -30,6 +30,10 @@ class DexFile;
 // Temporary measure until we have caught up with the Java 7 definition of Math.round. b/26327751
 static constexpr bool kRoundIsPlusPointFive = false;
 
+// Positive floating-point infinities.
+static constexpr uint32_t kPositiveInfinityFloat = 0x7f800000U;
+static constexpr uint64_t kPositiveInfinityDouble = UINT64_C(0x7ff0000000000000);
+
 // Recognize intrinsics from HInvoke nodes.
 class IntrinsicsRecognizer : public HOptimization {
  public:
@@ -235,6 +239,8 @@ UNREACHABLE_INTRINSIC(Arch, IntegerCompare)         \
 UNREACHABLE_INTRINSIC(Arch, LongCompare)            \
 UNREACHABLE_INTRINSIC(Arch, IntegerSignum)          \
 UNREACHABLE_INTRINSIC(Arch, LongSignum)             \
+UNREACHABLE_INTRINSIC(Arch, StringIsEmpty)          \
+UNREACHABLE_INTRINSIC(Arch, StringLength)           \
 UNREACHABLE_INTRINSIC(Arch, UnsafeLoadFence)        \
 UNREACHABLE_INTRINSIC(Arch, UnsafeStoreFence)       \
 UNREACHABLE_INTRINSIC(Arch, UnsafeFullFence)
diff --git a/compiler/optimizing/intrinsics_arm.cc b/compiler/optimizing/intrinsics_arm.cc
index 86b7bc138..de04175e3 100644
--- a/compiler/optimizing/intrinsics_arm.cc
+++ b/compiler/optimizing/intrinsics_arm.cc
@@ -1115,16 +1115,16 @@ static void GenerateVisitStringIndexOf(HInvoke* invoke,
                                        ArenaAllocator* allocator,
                                        bool start_at_zero) {
   LocationSummary* locations = invoke->GetLocations();
-  Register tmp_reg = locations->GetTemp(0).AsRegister<Register>();
 
   // Note that the null check must have been done earlier.
   DCHECK(!invoke->CanDoImplicitNullCheckOn(invoke->InputAt(0)));
 
   // Check for code points > 0xFFFF. Either a slow-path check when we don't know statically,
-  // or directly dispatch if we have a constant.
+  // or directly dispatch for a large constant, or omit slow-path for a small constant or a char.
   SlowPathCode* slow_path = nullptr;
-  if (invoke->InputAt(1)->IsIntConstant()) {
-    if (static_cast<uint32_t>(invoke->InputAt(1)->AsIntConstant()->GetValue()) >
+  HInstruction* code_point = invoke->InputAt(1);
+  if (code_point->IsIntConstant()) {
+    if (static_cast<uint32_t>(code_point->AsIntConstant()->GetValue()) >
         std::numeric_limits<uint16_t>::max()) {
       // Always needs the slow-path. We could directly dispatch to it, but this case should be
       // rare, so for simplicity just put the full slow-path down and branch unconditionally.
@@ -1134,16 +1134,18 @@ static void GenerateVisitStringIndexOf(HInvoke* invoke,
       __ Bind(slow_path->GetExitLabel());
       return;
     }
-  } else {
+  } else if (code_point->GetType() != Primitive::kPrimChar) {
     Register char_reg = locations->InAt(1).AsRegister<Register>();
-    __ LoadImmediate(tmp_reg, std::numeric_limits<uint16_t>::max());
-    __ cmp(char_reg, ShifterOperand(tmp_reg));
+    // 0xffff is not modified immediate but 0x10000 is, so use `>= 0x10000` instead of `> 0xffff`.
+    __ cmp(char_reg,
+           ShifterOperand(static_cast<uint32_t>(std::numeric_limits<uint16_t>::max()) + 1));
     slow_path = new (allocator) IntrinsicSlowPathARM(invoke);
     codegen->AddSlowPath(slow_path);
-    __ b(slow_path->GetEntryLabel(), HI);
+    __ b(slow_path->GetEntryLabel(), HS);
   }
 
   if (start_at_zero) {
+    Register tmp_reg = locations->GetTemp(0).AsRegister<Register>();
     DCHECK_EQ(tmp_reg, R2);
     // Start-index = 0.
     __ LoadImmediate(tmp_reg, 0);
@@ -1170,7 +1172,7 @@ void IntrinsicLocationsBuilderARM::VisitStringIndexOf(HInvoke* invoke) {
   locations->SetInAt(1, Location::RegisterLocation(calling_convention.GetRegisterAt(1)));
   locations->SetOut(Location::RegisterLocation(R0));
 
-  // Need a temp for slow-path codepoint compare, and need to send start-index=0.
+  // Need to send start-index=0.
   locations->AddTemp(Location::RegisterLocation(calling_convention.GetRegisterAt(2)));
 }
 
@@ -1190,9 +1192,6 @@ void IntrinsicLocationsBuilderARM::VisitStringIndexOfAfter(HInvoke* invoke) {
   locations->SetInAt(1, Location::RegisterLocation(calling_convention.GetRegisterAt(1)));
   locations->SetInAt(2, Location::RegisterLocation(calling_convention.GetRegisterAt(2)));
   locations->SetOut(Location::RegisterLocation(R0));
-
-  // Need a temp for slow-path codepoint compare.
-  locations->AddTemp(Location::RequiresRegister());
 }
 
 void IntrinsicCodeGeneratorARM::VisitStringIndexOfAfter(HInvoke* invoke) {
@@ -1985,6 +1984,56 @@ void IntrinsicCodeGeneratorARM::VisitStringGetCharsNoCheck(HInvoke* invoke) {
   __ Bind(&done);
 }
 
+void IntrinsicLocationsBuilderARM::VisitFloatIsInfinite(HInvoke* invoke) {
+  CreateFPToIntLocations(arena_, invoke);
+}
+
+void IntrinsicCodeGeneratorARM::VisitFloatIsInfinite(HInvoke* invoke) {
+  ArmAssembler* const assembler = GetAssembler();
+  LocationSummary* const locations = invoke->GetLocations();
+  const Register out = locations->Out().AsRegister<Register>();
+  // Shifting left by 1 bit makes the value encodable as an immediate operand;
+  // we don't care about the sign bit anyway.
+  constexpr uint32_t infinity = kPositiveInfinityFloat << 1U;
+
+  __ vmovrs(out, locations->InAt(0).AsFpuRegister<SRegister>());
+  // We don't care about the sign bit, so shift left.
+  __ Lsl(out, out, 1);
+  __ eor(out, out, ShifterOperand(infinity));
+  // If the result is 0, then it has 32 leading zeros, and less than that otherwise.
+  __ clz(out, out);
+  // Any number less than 32 logically shifted right by 5 bits results in 0;
+  // the same operation on 32 yields 1.
+  __ Lsr(out, out, 5);
+}
+
+void IntrinsicLocationsBuilderARM::VisitDoubleIsInfinite(HInvoke* invoke) {
+  CreateFPToIntLocations(arena_, invoke);
+}
+
+void IntrinsicCodeGeneratorARM::VisitDoubleIsInfinite(HInvoke* invoke) {
+  ArmAssembler* const assembler = GetAssembler();
+  LocationSummary* const locations = invoke->GetLocations();
+  const Register out = locations->Out().AsRegister<Register>();
+  // The highest 32 bits of double precision positive infinity separated into
+  // two constants encodable as immediate operands.
+  constexpr uint32_t infinity_high  = 0x7f000000U;
+  constexpr uint32_t infinity_high2 = 0x00f00000U;
+
+  static_assert((infinity_high | infinity_high2) == static_cast<uint32_t>(kPositiveInfinityDouble >> 32U),
+                "The constants do not add up to the high 32 bits of double precision positive infinity.");
+  __ vmovrrd(IP, out, FromLowSToD(locations->InAt(0).AsFpuRegisterPairLow<SRegister>()));
+  __ eor(out, out, ShifterOperand(infinity_high));
+  __ eor(out, out, ShifterOperand(infinity_high2));
+  // We don't care about the sign bit, so shift left.
+  __ orr(out, IP, ShifterOperand(out, LSL, 1));
+  // If the result is 0, then it has 32 leading zeros, and less than that otherwise.
+  __ clz(out, out);
+  // Any number less than 32 logically shifted right by 5 bits results in 0;
+  // the same operation on 32 yields 1.
+  __ Lsr(out, out, 5);
+}
+
 UNIMPLEMENTED_INTRINSIC(ARM, IntegerBitCount)
 UNIMPLEMENTED_INTRINSIC(ARM, LongBitCount)
 UNIMPLEMENTED_INTRINSIC(ARM, MathMinDoubleDouble)
@@ -2001,8 +2050,6 @@ UNIMPLEMENTED_INTRINSIC(ARM, MathRoundFloat)    // Could be done by changing rou
 UNIMPLEMENTED_INTRINSIC(ARM, UnsafeCASLong)     // High register pressure.
 UNIMPLEMENTED_INTRINSIC(ARM, SystemArrayCopyChar)
 UNIMPLEMENTED_INTRINSIC(ARM, ReferenceGetReferent)
-UNIMPLEMENTED_INTRINSIC(ARM, FloatIsInfinite)
-UNIMPLEMENTED_INTRINSIC(ARM, DoubleIsInfinite)
 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 04ae3a673..6cd1726eb 100644
--- a/compiler/optimizing/intrinsics_arm64.cc
+++ b/compiler/optimizing/intrinsics_arm64.cc
@@ -47,6 +47,7 @@ using helpers::SRegisterFrom;
 using helpers::WRegisterFrom;
 using helpers::XRegisterFrom;
 using helpers::InputRegisterAt;
+using helpers::OutputRegister;
 
 namespace {
 
@@ -1173,31 +1174,118 @@ void IntrinsicCodeGeneratorARM64::VisitStringCharAt(HInvoke* invoke) {
 
 void IntrinsicLocationsBuilderARM64::VisitStringCompareTo(HInvoke* invoke) {
   LocationSummary* locations = new (arena_) LocationSummary(invoke,
-                                                            LocationSummary::kCall,
+                                                            invoke->InputAt(1)->CanBeNull()
+                                                                ? LocationSummary::kCallOnSlowPath
+                                                                : LocationSummary::kNoCall,
                                                             kIntrinsified);
-  InvokeRuntimeCallingConvention calling_convention;
-  locations->SetInAt(0, LocationFrom(calling_convention.GetRegisterAt(0)));
-  locations->SetInAt(1, LocationFrom(calling_convention.GetRegisterAt(1)));
-  locations->SetOut(calling_convention.GetReturnLocation(Primitive::kPrimInt));
+  locations->SetInAt(0, Location::RequiresRegister());
+  locations->SetInAt(1, Location::RequiresRegister());
+  locations->AddTemp(Location::RequiresRegister());
+  locations->AddTemp(Location::RequiresRegister());
+  locations->AddTemp(Location::RequiresRegister());
+  locations->SetOut(Location::RequiresRegister(), Location::kOutputOverlap);
 }
 
 void IntrinsicCodeGeneratorARM64::VisitStringCompareTo(HInvoke* invoke) {
   vixl::MacroAssembler* masm = GetVIXLAssembler();
   LocationSummary* locations = invoke->GetLocations();
 
+  Register str = XRegisterFrom(locations->InAt(0));
+  Register arg = XRegisterFrom(locations->InAt(1));
+  Register out = OutputRegister(invoke);
+
+  Register temp0 = WRegisterFrom(locations->GetTemp(0));
+  Register temp1 = WRegisterFrom(locations->GetTemp(1));
+  Register temp2 = WRegisterFrom(locations->GetTemp(2));
+
+  vixl::Label loop;
+  vixl::Label find_char_diff;
+  vixl::Label end;
+
+  // Get offsets of count and value fields within a string object.
+  const int32_t count_offset = mirror::String::CountOffset().Int32Value();
+  const int32_t value_offset = mirror::String::ValueOffset().Int32Value();
+
   // Note that the null check must have been done earlier.
   DCHECK(!invoke->CanDoImplicitNullCheckOn(invoke->InputAt(0)));
 
-  Register argument = WRegisterFrom(locations->InAt(1));
-  __ Cmp(argument, 0);
-  SlowPathCodeARM64* slow_path = new (GetAllocator()) IntrinsicSlowPathARM64(invoke);
-  codegen_->AddSlowPath(slow_path);
-  __ B(eq, slow_path->GetEntryLabel());
+  // Take slow path and throw if input can be and is null.
+  SlowPathCodeARM64* slow_path = nullptr;
+  const bool can_slow_path = invoke->InputAt(1)->CanBeNull();
+  if (can_slow_path) {
+    slow_path = new (GetAllocator()) IntrinsicSlowPathARM64(invoke);
+    codegen_->AddSlowPath(slow_path);
+    __ Cbz(arg, slow_path->GetEntryLabel());
+  }
 
-  __ Ldr(
-      lr, MemOperand(tr, QUICK_ENTRYPOINT_OFFSET(kArm64WordSize, pStringCompareTo).Int32Value()));
-  __ Blr(lr);
-  __ Bind(slow_path->GetExitLabel());
+  // Reference equality check, return 0 if same reference.
+  __ Subs(out, str, arg);
+  __ B(&end, eq);
+  // Load lengths of this and argument strings.
+  __ Ldr(temp0, MemOperand(str.X(), count_offset));
+  __ Ldr(temp1, MemOperand(arg.X(), count_offset));
+  // Return zero if both strings are empty.
+  __ Orr(out, temp0, temp1);
+  __ Cbz(out, &end);
+  // out = length diff.
+  __ Subs(out, temp0, temp1);
+  // temp2 = min(len(str), len(arg)).
+  __ Csel(temp2, temp1, temp0, ge);
+  // Shorter string is empty?
+  __ Cbz(temp2, &end);
+
+  // Store offset of string value in preparation for comparison loop.
+  __ Mov(temp1, value_offset);
+
+  UseScratchRegisterScope scratch_scope(masm);
+  Register temp4 = scratch_scope.AcquireX();
+
+  // Assertions that must hold in order to compare strings 4 characters at a time.
+  DCHECK_ALIGNED(value_offset, 8);
+  static_assert(IsAligned<8>(kObjectAlignment), "String of odd length is not zero padded");
+
+  const size_t char_size = Primitive::ComponentSize(Primitive::kPrimChar);
+  DCHECK_EQ(char_size, 2u);
+
+  // Promote temp0 to an X reg, ready for LDR.
+  temp0 = temp0.X();
+
+  // Loop to compare 4x16-bit characters at a time (ok because of string data alignment).
+  __ Bind(&loop);
+  __ Ldr(temp4, MemOperand(str.X(), temp1));
+  __ Ldr(temp0, MemOperand(arg.X(), temp1));
+  __ Cmp(temp4, temp0);
+  __ B(ne, &find_char_diff);
+  __ Add(temp1, temp1, char_size * 4);
+  __ Subs(temp2, temp2, 4);
+  __ B(gt, &loop);
+  __ B(&end);
+
+  // Promote temp1 to an X reg, ready for EOR.
+  temp1 = temp1.X();
+
+  // Find the single 16-bit character difference.
+  __ Bind(&find_char_diff);
+  // Get the bit position of the first character that differs.
+  __ Eor(temp1, temp0, temp4);
+  __ Rbit(temp1, temp1);
+  __ Clz(temp1, temp1);
+  __ Bic(temp1, temp1, 0xf);
+  // If the number of 16-bit chars remaining <= the index where the difference occurs (0-3), then
+  // the difference occurs outside the remaining string data, so just return length diff (out).
+  __ Cmp(temp2, Operand(temp1, LSR, 4));
+  __ B(le, &end);
+  // Extract the characters and calculate the difference.
+  __ Lsr(temp0, temp0, temp1);
+  __ Lsr(temp4, temp4, temp1);
+  __ And(temp4, temp4, 0xffff);
+  __ Sub(out, temp4, Operand(temp0, UXTH));
+
+  __ Bind(&end);
+
+  if (can_slow_path) {
+    __ Bind(slow_path->GetExitLabel());
+  }
 }
 
 void IntrinsicLocationsBuilderARM64::VisitStringEquals(HInvoke* invoke) {
@@ -1302,16 +1390,16 @@ static void GenerateVisitStringIndexOf(HInvoke* invoke,
                                        ArenaAllocator* allocator,
                                        bool start_at_zero) {
   LocationSummary* locations = invoke->GetLocations();
-  Register tmp_reg = WRegisterFrom(locations->GetTemp(0));
 
   // Note that the null check must have been done earlier.
   DCHECK(!invoke->CanDoImplicitNullCheckOn(invoke->InputAt(0)));
 
   // Check for code points > 0xFFFF. Either a slow-path check when we don't know statically,
-  // or directly dispatch if we have a constant.
+  // or directly dispatch for a large constant, or omit slow-path for a small constant or a char.
   SlowPathCodeARM64* slow_path = nullptr;
-  if (invoke->InputAt(1)->IsIntConstant()) {
-    if (static_cast<uint32_t>(invoke->InputAt(1)->AsIntConstant()->GetValue()) > 0xFFFFU) {
+  HInstruction* code_point = invoke->InputAt(1);
+  if (code_point->IsIntConstant()) {
+    if (static_cast<uint32_t>(code_point->AsIntConstant()->GetValue()) > 0xFFFFU) {
       // Always needs the slow-path. We could directly dispatch to it, but this case should be
       // rare, so for simplicity just put the full slow-path down and branch unconditionally.
       slow_path = new (allocator) IntrinsicSlowPathARM64(invoke);
@@ -1320,17 +1408,17 @@ static void GenerateVisitStringIndexOf(HInvoke* invoke,
       __ Bind(slow_path->GetExitLabel());
       return;
     }
-  } else {
+  } else if (code_point->GetType() != Primitive::kPrimChar) {
     Register char_reg = WRegisterFrom(locations->InAt(1));
-    __ Mov(tmp_reg, 0xFFFF);
-    __ Cmp(char_reg, Operand(tmp_reg));
+    __ Tst(char_reg, 0xFFFF0000);
     slow_path = new (allocator) IntrinsicSlowPathARM64(invoke);
     codegen->AddSlowPath(slow_path);
-    __ B(hi, slow_path->GetEntryLabel());
+    __ B(ne, slow_path->GetEntryLabel());
   }
 
   if (start_at_zero) {
     // Start-index = 0.
+    Register tmp_reg = WRegisterFrom(locations->GetTemp(0));
     __ Mov(tmp_reg, 0);
   }
 
@@ -1354,7 +1442,7 @@ void IntrinsicLocationsBuilderARM64::VisitStringIndexOf(HInvoke* invoke) {
   locations->SetInAt(1, LocationFrom(calling_convention.GetRegisterAt(1)));
   locations->SetOut(calling_convention.GetReturnLocation(Primitive::kPrimInt));
 
-  // Need a temp for slow-path codepoint compare, and need to send start_index=0.
+  // Need to send start_index=0.
   locations->AddTemp(LocationFrom(calling_convention.GetRegisterAt(2)));
 }
 
@@ -1374,9 +1462,6 @@ void IntrinsicLocationsBuilderARM64::VisitStringIndexOfAfter(HInvoke* invoke) {
   locations->SetInAt(1, LocationFrom(calling_convention.GetRegisterAt(1)));
   locations->SetInAt(2, LocationFrom(calling_convention.GetRegisterAt(2)));
   locations->SetOut(calling_convention.GetReturnLocation(Primitive::kPrimInt));
-
-  // Need a temp for slow-path codepoint compare.
-  locations->AddTemp(Location::RequiresRegister());
 }
 
 void IntrinsicCodeGeneratorARM64::VisitStringIndexOfAfter(HInvoke* invoke) {
@@ -2201,9 +2286,46 @@ void IntrinsicCodeGeneratorARM64::VisitSystemArrayCopy(HInvoke* invoke) {
   __ Bind(slow_path->GetExitLabel());
 }
 
+static void GenIsInfinite(LocationSummary* locations,
+                          bool is64bit,
+                          vixl::MacroAssembler* masm) {
+  Operand infinity;
+  Register out;
+
+  if (is64bit) {
+    infinity = kPositiveInfinityDouble;
+    out = XRegisterFrom(locations->Out());
+  } else {
+    infinity = kPositiveInfinityFloat;
+    out = WRegisterFrom(locations->Out());
+  }
+
+  const Register zero = vixl::Assembler::AppropriateZeroRegFor(out);
+
+  MoveFPToInt(locations, is64bit, masm);
+  __ Eor(out, out, infinity);
+  // We don't care about the sign bit, so shift left.
+  __ Cmp(zero, Operand(out, LSL, 1));
+  __ Cset(out, eq);
+}
+
+void IntrinsicLocationsBuilderARM64::VisitFloatIsInfinite(HInvoke* invoke) {
+  CreateFPToIntLocations(arena_, invoke);
+}
+
+void IntrinsicCodeGeneratorARM64::VisitFloatIsInfinite(HInvoke* invoke) {
+  GenIsInfinite(invoke->GetLocations(), /* is64bit */ false, GetVIXLAssembler());
+}
+
+void IntrinsicLocationsBuilderARM64::VisitDoubleIsInfinite(HInvoke* invoke) {
+  CreateFPToIntLocations(arena_, invoke);
+}
+
+void IntrinsicCodeGeneratorARM64::VisitDoubleIsInfinite(HInvoke* invoke) {
+  GenIsInfinite(invoke->GetLocations(), /* is64bit */ true, GetVIXLAssembler());
+}
+
 UNIMPLEMENTED_INTRINSIC(ARM64, ReferenceGetReferent)
-UNIMPLEMENTED_INTRINSIC(ARM64, FloatIsInfinite)
-UNIMPLEMENTED_INTRINSIC(ARM64, DoubleIsInfinite)
 UNIMPLEMENTED_INTRINSIC(ARM64, IntegerHighestOneBit)
 UNIMPLEMENTED_INTRINSIC(ARM64, LongHighestOneBit)
 UNIMPLEMENTED_INTRINSIC(ARM64, IntegerLowestOneBit)
diff --git a/compiler/optimizing/intrinsics_list.h b/compiler/optimizing/intrinsics_list.h
index dd9294d48..db60238fb 100644
--- a/compiler/optimizing/intrinsics_list.h
+++ b/compiler/optimizing/intrinsics_list.h
@@ -107,6 +107,8 @@
   V(StringGetCharsNoCheck, kDirect, kNeedsEnvironmentOrCache, kReadSideEffects, kCanThrow) \
   V(StringIndexOf, kDirect, kNeedsEnvironmentOrCache, kReadSideEffects, kCanThrow) \
   V(StringIndexOfAfter, kDirect, kNeedsEnvironmentOrCache, kReadSideEffects, kCanThrow) \
+  V(StringIsEmpty, kDirect, kNeedsEnvironmentOrCache, kReadSideEffects, kNoThrow) \
+  V(StringLength, kDirect, kNeedsEnvironmentOrCache, kReadSideEffects, kNoThrow) \
   V(StringNewStringFromBytes, kStatic, kNeedsEnvironmentOrCache, kAllSideEffects, kCanThrow) \
   V(StringNewStringFromChars, kStatic, kNeedsEnvironmentOrCache, kAllSideEffects, kCanThrow) \
   V(StringNewStringFromString, kStatic, kNeedsEnvironmentOrCache, kAllSideEffects, kCanThrow) \
diff --git a/compiler/optimizing/intrinsics_mips.cc b/compiler/optimizing/intrinsics_mips.cc
index 19c6a225a..fa250a306 100644
--- a/compiler/optimizing/intrinsics_mips.cc
+++ b/compiler/optimizing/intrinsics_mips.cc
@@ -2067,11 +2067,12 @@ static void GenerateStringIndexOf(HInvoke* invoke,
   // Note that the null check must have been done earlier.
   DCHECK(!invoke->CanDoImplicitNullCheckOn(invoke->InputAt(0)));
 
-  // Check for code points > 0xFFFF. Either a slow-path check when we
-  // don't know statically, or directly dispatch if we have a constant.
+  // Check for code points > 0xFFFF. Either a slow-path check when we don't know statically,
+  // or directly dispatch for a large constant, or omit slow-path for a small constant or a char.
   SlowPathCodeMIPS* slow_path = nullptr;
-  if (invoke->InputAt(1)->IsIntConstant()) {
-    if (!IsUint<16>(invoke->InputAt(1)->AsIntConstant()->GetValue())) {
+  HInstruction* code_point = invoke->InputAt(1);
+  if (code_point->IsIntConstant()) {
+    if (!IsUint<16>(code_point->AsIntConstant()->GetValue())) {
       // Always needs the slow-path. We could directly dispatch to it,
       // but this case should be rare, so for simplicity just put the
       // full slow-path down and branch unconditionally.
@@ -2081,7 +2082,7 @@ static void GenerateStringIndexOf(HInvoke* invoke,
       __ Bind(slow_path->GetExitLabel());
       return;
     }
-  } else {
+  } else if (code_point->GetType() != Primitive::kPrimChar) {
     Register char_reg = locations->InAt(1).AsRegister<Register>();
     // The "bltu" conditional branch tests to see if the character value
     // fits in a valid 16-bit (MIPS halfword) value. If it doesn't then
@@ -2283,10 +2284,10 @@ static void GenIsInfinite(LocationSummary* locations,
     // If one, or more, of the exponent bits is zero, then the number can't be infinite.
     if (type == Primitive::kPrimDouble) {
       __ MoveFromFpuHigh(TMP, in);
-      __ LoadConst32(AT, 0x7FF00000);
+      __ LoadConst32(AT, High32Bits(kPositiveInfinityDouble));
     } else {
       __ Mfc1(TMP, in);
-      __ LoadConst32(AT, 0x7F800000);
+      __ LoadConst32(AT, kPositiveInfinityFloat);
     }
     __ Xor(TMP, TMP, AT);
 
diff --git a/compiler/optimizing/intrinsics_mips64.cc b/compiler/optimizing/intrinsics_mips64.cc
index 1524e1e01..6c4e64e4b 100644
--- a/compiler/optimizing/intrinsics_mips64.cc
+++ b/compiler/optimizing/intrinsics_mips64.cc
@@ -1563,11 +1563,12 @@ static void GenerateStringIndexOf(HInvoke* invoke,
   // Note that the null check must have been done earlier.
   DCHECK(!invoke->CanDoImplicitNullCheckOn(invoke->InputAt(0)));
 
-  // Check for code points > 0xFFFF. Either a slow-path check when we
-  // don't know statically, or directly dispatch if we have a constant.
+  // Check for code points > 0xFFFF. Either a slow-path check when we don't know statically,
+  // or directly dispatch for a large constant, or omit slow-path for a small constant or a char.
   SlowPathCodeMIPS64* slow_path = nullptr;
-  if (invoke->InputAt(1)->IsIntConstant()) {
-    if (!IsUint<16>(invoke->InputAt(1)->AsIntConstant()->GetValue())) {
+  HInstruction* code_point = invoke->InputAt(1);
+  if (code_point->IsIntConstant()) {
+    if (!IsUint<16>(code_point->AsIntConstant()->GetValue())) {
       // Always needs the slow-path. We could directly dispatch to it,
       // but this case should be rare, so for simplicity just put the
       // full slow-path down and branch unconditionally.
@@ -1577,7 +1578,7 @@ static void GenerateStringIndexOf(HInvoke* invoke,
       __ Bind(slow_path->GetExitLabel());
       return;
     }
-  } else {
+  } else if (code_point->GetType() != Primitive::kPrimChar) {
     GpuRegister char_reg = locations->InAt(1).AsRegister<GpuRegister>();
     __ LoadConst32(tmp_reg, std::numeric_limits<uint16_t>::max());
     slow_path = new (allocator) IntrinsicSlowPathMIPS64(invoke);
diff --git a/compiler/optimizing/intrinsics_x86.cc b/compiler/optimizing/intrinsics_x86.cc
index 4aab3e276..99bc40e71 100644
--- a/compiler/optimizing/intrinsics_x86.cc
+++ b/compiler/optimizing/intrinsics_x86.cc
@@ -1418,10 +1418,11 @@ static void GenerateStringIndexOf(HInvoke* invoke,
   DCHECK_EQ(out, EDI);
 
   // Check for code points > 0xFFFF. Either a slow-path check when we don't know statically,
-  // or directly dispatch if we have a constant.
+  // or directly dispatch for a large constant, or omit slow-path for a small constant or a char.
   SlowPathCode* slow_path = nullptr;
-  if (invoke->InputAt(1)->IsIntConstant()) {
-    if (static_cast<uint32_t>(invoke->InputAt(1)->AsIntConstant()->GetValue()) >
+  HInstruction* code_point = invoke->InputAt(1);
+  if (code_point->IsIntConstant()) {
+    if (static_cast<uint32_t>(code_point->AsIntConstant()->GetValue()) >
     std::numeric_limits<uint16_t>::max()) {
       // Always needs the slow-path. We could directly dispatch to it, but this case should be
       // rare, so for simplicity just put the full slow-path down and branch unconditionally.
@@ -1431,7 +1432,7 @@ static void GenerateStringIndexOf(HInvoke* invoke,
       __ Bind(slow_path->GetExitLabel());
       return;
     }
-  } else {
+  } else if (code_point->GetType() != Primitive::kPrimChar) {
     __ cmpl(search_value, Immediate(std::numeric_limits<uint16_t>::max()));
     slow_path = new (allocator) IntrinsicSlowPathX86(invoke);
     codegen->AddSlowPath(slow_path);
diff --git a/compiler/optimizing/intrinsics_x86_64.cc b/compiler/optimizing/intrinsics_x86_64.cc
index 1d32dc7bc..06e9cc2b2 100644
--- a/compiler/optimizing/intrinsics_x86_64.cc
+++ b/compiler/optimizing/intrinsics_x86_64.cc
@@ -1517,10 +1517,11 @@ static void GenerateStringIndexOf(HInvoke* invoke,
   DCHECK_EQ(out.AsRegister(), RDI);
 
   // Check for code points > 0xFFFF. Either a slow-path check when we don't know statically,
-  // or directly dispatch if we have a constant.
+  // or directly dispatch for a large constant, or omit slow-path for a small constant or a char.
   SlowPathCode* slow_path = nullptr;
-  if (invoke->InputAt(1)->IsIntConstant()) {
-    if (static_cast<uint32_t>(invoke->InputAt(1)->AsIntConstant()->GetValue()) >
+  HInstruction* code_point = invoke->InputAt(1);
+  if (code_point->IsIntConstant()) {
+    if (static_cast<uint32_t>(code_point->AsIntConstant()->GetValue()) >
     std::numeric_limits<uint16_t>::max()) {
       // Always needs the slow-path. We could directly dispatch to it, but this case should be
       // rare, so for simplicity just put the full slow-path down and branch unconditionally.
@@ -1530,7 +1531,7 @@ static void GenerateStringIndexOf(HInvoke* invoke,
       __ Bind(slow_path->GetExitLabel());
       return;
     }
-  } else {
+  } else if (code_point->GetType() != Primitive::kPrimChar) {
     __ cmpl(search_value, Immediate(std::numeric_limits<uint16_t>::max()));
     slow_path = new (allocator) IntrinsicSlowPathX86_64(invoke);
     codegen->AddSlowPath(slow_path);
diff --git a/compiler/optimizing/licm.cc b/compiler/optimizing/licm.cc
index 5a0b89c90..7543cd6c5 100644
--- a/compiler/optimizing/licm.cc
+++ b/compiler/optimizing/licm.cc
@@ -101,16 +101,6 @@ void LICM::Run() {
     SideEffects loop_effects = side_effects_.GetLoopEffects(block);
     HBasicBlock* pre_header = loop_info->GetPreHeader();
 
-    bool contains_irreducible_loop = false;
-    if (graph_->HasIrreducibleLoops()) {
-      for (HBlocksInLoopIterator it_loop(*loop_info); !it_loop.Done(); it_loop.Advance()) {
-        if (it_loop.Current()->GetLoopInformation()->IsIrreducible()) {
-          contains_irreducible_loop = true;
-          break;
-        }
-      }
-    }
-
     for (HBlocksInLoopIterator it_loop(*loop_info); !it_loop.Done(); it_loop.Advance()) {
       HBasicBlock* inner = it_loop.Current();
       DCHECK(inner->IsInLoop());
@@ -123,11 +113,12 @@ void LICM::Run() {
         visited->SetBit(inner->GetBlockId());
       }
 
-      if (contains_irreducible_loop) {
+      if (loop_info->ContainsIrreducibleLoop()) {
         // We cannot licm in an irreducible loop, or in a natural loop containing an
         // irreducible loop.
         continue;
       }
+      DCHECK(!loop_info->IsIrreducible());
 
       // We can move an instruction that can throw only if it is the first
       // throwing instruction in the loop. Note that the first potentially
diff --git a/compiler/optimizing/load_store_elimination.cc b/compiler/optimizing/load_store_elimination.cc
index 2de41580b..8a75a90cf 100644
--- a/compiler/optimizing/load_store_elimination.cc
+++ b/compiler/optimizing/load_store_elimination.cc
@@ -733,19 +733,14 @@ class LSEVisitor : public HGraphVisitor {
       if (Primitive::PrimitiveKind(heap_value->GetType())
               != Primitive::PrimitiveKind(instruction->GetType())) {
         // The only situation where the same heap location has different type is when
-        // we do an array get from a null constant. In order to stay properly typed
-        // we do not merge the array gets.
+        // we do an array get on an instruction that originates from the null constant
+        // (the null could be behind a field access, an array access, a null check or
+        // a bound type).
+        // In order to stay properly typed on primitive types, we do not eliminate
+        // the array gets.
         if (kIsDebugBuild) {
           DCHECK(heap_value->IsArrayGet()) << heap_value->DebugName();
           DCHECK(instruction->IsArrayGet()) << instruction->DebugName();
-          HInstruction* array = instruction->AsArrayGet()->GetArray();
-          DCHECK(array->IsNullCheck()) << array->DebugName();
-          HInstruction* input = HuntForOriginalReference(array->InputAt(0));
-          DCHECK(input->IsNullConstant()) << input->DebugName();
-          array = heap_value->AsArrayGet()->GetArray();
-          DCHECK(array->IsNullCheck()) << array->DebugName();
-          input = HuntForOriginalReference(array->InputAt(0));
-          DCHECK(input->IsNullConstant()) << input->DebugName();
         }
         return;
       }
diff --git a/compiler/optimizing/nodes.cc b/compiler/optimizing/nodes.cc
index fe75451ad..60329ccff 100644
--- a/compiler/optimizing/nodes.cc
+++ b/compiler/optimizing/nodes.cc
@@ -56,9 +56,11 @@ void HGraph::FindBackEdges(ArenaBitVector* visited) {
   // Nodes that we're currently visiting, indexed by block id.
   ArenaBitVector visiting(arena_, blocks_.size(), false, kArenaAllocGraphBuilder);
   // Number of successors visited from a given node, indexed by block id.
-  ArenaVector<size_t> successors_visited(blocks_.size(), 0u, arena_->Adapter());
+  ArenaVector<size_t> successors_visited(blocks_.size(),
+                                         0u,
+                                         arena_->Adapter(kArenaAllocGraphBuilder));
   // Stack of nodes that we're currently visiting (same as marked in "visiting" above).
-  ArenaVector<HBasicBlock*> worklist(arena_->Adapter());
+  ArenaVector<HBasicBlock*> worklist(arena_->Adapter(kArenaAllocGraphBuilder));
   constexpr size_t kDefaultWorklistSize = 8;
   worklist.reserve(kDefaultWorklistSize);
   visited->SetBit(entry_block_->GetBlockId());
@@ -206,17 +208,35 @@ HInstruction* HBasicBlock::GetFirstInstructionDisregardMoves() const {
   return instruction;
 }
 
+static bool UpdateDominatorOfSuccessor(HBasicBlock* block, HBasicBlock* successor) {
+  DCHECK(ContainsElement(block->GetSuccessors(), successor));
+
+  HBasicBlock* old_dominator = successor->GetDominator();
+  HBasicBlock* new_dominator =
+      (old_dominator == nullptr) ? block
+                                 : CommonDominator::ForPair(old_dominator, block);
+
+  if (old_dominator == new_dominator) {
+    return false;
+  } else {
+    successor->SetDominator(new_dominator);
+    return true;
+  }
+}
+
 void HGraph::ComputeDominanceInformation() {
   DCHECK(reverse_post_order_.empty());
   reverse_post_order_.reserve(blocks_.size());
   reverse_post_order_.push_back(entry_block_);
 
   // Number of visits of a given node, indexed by block id.
-  ArenaVector<size_t> visits(blocks_.size(), 0u, arena_->Adapter());
+  ArenaVector<size_t> visits(blocks_.size(), 0u, arena_->Adapter(kArenaAllocGraphBuilder));
   // Number of successors visited from a given node, indexed by block id.
-  ArenaVector<size_t> successors_visited(blocks_.size(), 0u, arena_->Adapter());
+  ArenaVector<size_t> successors_visited(blocks_.size(),
+                                         0u,
+                                         arena_->Adapter(kArenaAllocGraphBuilder));
   // Nodes for which we need to visit successors.
-  ArenaVector<HBasicBlock*> worklist(arena_->Adapter());
+  ArenaVector<HBasicBlock*> worklist(arena_->Adapter(kArenaAllocGraphBuilder));
   constexpr size_t kDefaultWorklistSize = 8;
   worklist.reserve(kDefaultWorklistSize);
   worklist.push_back(entry_block_);
@@ -228,15 +248,7 @@ void HGraph::ComputeDominanceInformation() {
       worklist.pop_back();
     } else {
       HBasicBlock* successor = current->GetSuccessors()[successors_visited[current_id]++];
-
-      if (successor->GetDominator() == nullptr) {
-        successor->SetDominator(current);
-      } else {
-        // The CommonDominator can work for multiple blocks as long as the
-        // domination information doesn't change. However, since we're changing
-        // that information here, we can use the finder only for pairs of blocks.
-        successor->SetDominator(CommonDominator::ForPair(successor->GetDominator(), current));
-      }
+      UpdateDominatorOfSuccessor(current, successor);
 
       // Once all the forward edges have been visited, we know the immediate
       // dominator of the block. We can then start visiting its successors.
@@ -248,6 +260,44 @@ void HGraph::ComputeDominanceInformation() {
     }
   }
 
+  // Check if the graph has back edges not dominated by their respective headers.
+  // If so, we need to update the dominators of those headers and recursively of
+  // their successors. We do that with a fix-point iteration over all blocks.
+  // The algorithm is guaranteed to terminate because it loops only if the sum
+  // of all dominator chains has decreased in the current iteration.
+  bool must_run_fix_point = false;
+  for (HBasicBlock* block : blocks_) {
+    if (block != nullptr &&
+        block->IsLoopHeader() &&
+        block->GetLoopInformation()->HasBackEdgeNotDominatedByHeader()) {
+      must_run_fix_point = true;
+      break;
+    }
+  }
+  if (must_run_fix_point) {
+    bool update_occurred = true;
+    while (update_occurred) {
+      update_occurred = false;
+      for (HReversePostOrderIterator it(*this); !it.Done(); it.Advance()) {
+        HBasicBlock* block = it.Current();
+        for (HBasicBlock* successor : block->GetSuccessors()) {
+          update_occurred |= UpdateDominatorOfSuccessor(block, successor);
+        }
+      }
+    }
+  }
+
+  // Make sure that there are no remaining blocks whose dominator information
+  // needs to be updated.
+  if (kIsDebugBuild) {
+    for (HReversePostOrderIterator it(*this); !it.Done(); it.Advance()) {
+      HBasicBlock* block = it.Current();
+      for (HBasicBlock* successor : block->GetSuccessors()) {
+        DCHECK(!UpdateDominatorOfSuccessor(block, successor));
+      }
+    }
+  }
+
   // Populate `dominated_blocks_` information after computing all dominators.
   // The potential presence of irreducible loops requires to do it after.
   for (HReversePostOrderIterator it(*this); !it.Done(); it.Advance()) {
@@ -396,8 +446,10 @@ void HGraph::SimplifyCFG() {
 }
 
 GraphAnalysisResult HGraph::AnalyzeLoops() const {
-  // Order does not matter.
-  for (HReversePostOrderIterator it(*this); !it.Done(); it.Advance()) {
+  // We iterate post order to ensure we visit inner loops before outer loops.
+  // `PopulateRecursive` needs this guarantee to know whether a natural loop
+  // contains an irreducible loop.
+  for (HPostOrderIterator it(*this); !it.Done(); it.Advance()) {
     HBasicBlock* block = it.Current();
     if (block->IsLoopHeader()) {
       if (block->IsCatchBlock()) {
@@ -530,6 +582,14 @@ void HLoopInformation::PopulateRecursive(HBasicBlock* block) {
 
   blocks_.SetBit(block->GetBlockId());
   block->SetInLoop(this);
+  if (block->IsLoopHeader()) {
+    // We're visiting loops in post-order, so inner loops must have been
+    // populated already.
+    DCHECK(block->GetLoopInformation()->IsPopulated());
+    if (block->GetLoopInformation()->IsIrreducible()) {
+      contains_irreducible_loop_ = true;
+    }
+  }
   for (HBasicBlock* predecessor : block->GetPredecessors()) {
     PopulateRecursive(predecessor);
   }
@@ -595,20 +655,16 @@ void HLoopInformation::Populate() {
   blocks_.SetBit(header_->GetBlockId());
   header_->SetInLoop(this);
 
-  bool is_irreducible_loop = false;
-  for (HBasicBlock* back_edge : GetBackEdges()) {
-    DCHECK(back_edge->GetDominator() != nullptr);
-    if (!header_->Dominates(back_edge)) {
-      is_irreducible_loop = true;
-      break;
-    }
-  }
+  bool is_irreducible_loop = HasBackEdgeNotDominatedByHeader();
 
   if (is_irreducible_loop) {
     ArenaBitVector visited(graph->GetArena(),
                            graph->GetBlocks().size(),
                            /* expandable */ false,
                            kArenaAllocGraphBuilder);
+    // Stop marking blocks at the loop header.
+    visited.SetBit(header_->GetBlockId());
+
     for (HBasicBlock* back_edge : GetBackEdges()) {
       PopulateIrreducibleRecursive(back_edge, &visited);
     }
@@ -637,6 +693,7 @@ void HLoopInformation::Populate() {
   }
   if (is_irreducible_loop) {
     irreducible_ = true;
+    contains_irreducible_loop_ = true;
     graph->SetHasIrreducibleLoops(true);
   }
 }
@@ -667,6 +724,16 @@ size_t HLoopInformation::GetLifetimeEnd() const {
   return last_position;
 }
 
+bool HLoopInformation::HasBackEdgeNotDominatedByHeader() const {
+  for (HBasicBlock* back_edge : GetBackEdges()) {
+    DCHECK(back_edge->GetDominator() != nullptr);
+    if (!header_->Dominates(back_edge)) {
+      return true;
+    }
+  }
+  return false;
+}
+
 bool HBasicBlock::Dominates(HBasicBlock* other) const {
   // Walk up the dominator tree from `other`, to find out if `this`
   // is an ancestor.
@@ -2218,6 +2285,7 @@ ReferenceTypeInfo ReferenceTypeInfo::Create(TypeHandle type_handle, bool is_exac
     ScopedObjectAccess soa(Thread::Current());
     DCHECK(IsValidHandle(type_handle));
     DCHECK(!type_handle->IsErroneous());
+    DCHECK(!type_handle->IsArrayClass() || !type_handle->GetComponentType()->IsErroneous());
     if (!is_exact) {
       DCHECK(!type_handle->CannotBeAssignedFromOtherTypes())
           << "Callers of ReferenceTypeInfo::Create should ensure is_exact is properly computed";
@@ -2395,6 +2463,7 @@ void HLoadString::SetLoadKindInternal(LoadKind load_kind) {
   }
   if (!NeedsEnvironment()) {
     RemoveEnvironment();
+    SetSideEffects(SideEffects::None());
   }
 }
 
diff --git a/compiler/optimizing/nodes.h b/compiler/optimizing/nodes.h
index afb995d43..c08323a0c 100644
--- a/compiler/optimizing/nodes.h
+++ b/compiler/optimizing/nodes.h
@@ -650,6 +650,7 @@ class HLoopInformation : public ArenaObject<kArenaAllocLoopInfo> {
       : header_(header),
         suspend_check_(nullptr),
         irreducible_(false),
+        contains_irreducible_loop_(false),
         back_edges_(graph->GetArena()->Adapter(kArenaAllocLoopInfoBackEdges)),
         // Make bit vector growable, as the number of blocks may change.
         blocks_(graph->GetArena(), graph->GetBlocks().size(), true, kArenaAllocLoopInfoBackEdges) {
@@ -657,6 +658,7 @@ class HLoopInformation : public ArenaObject<kArenaAllocLoopInfo> {
   }
 
   bool IsIrreducible() const { return irreducible_; }
+  bool ContainsIrreducibleLoop() const { return contains_irreducible_loop_; }
 
   void Dump(std::ostream& os);
 
@@ -725,6 +727,12 @@ class HLoopInformation : public ArenaObject<kArenaAllocLoopInfo> {
     blocks_.ClearAllBits();
   }
 
+  bool HasBackEdgeNotDominatedByHeader() const;
+
+  bool IsPopulated() const {
+    return blocks_.GetHighestBitSet() != -1;
+  }
+
  private:
   // Internal recursive implementation of `Populate`.
   void PopulateRecursive(HBasicBlock* block);
@@ -733,6 +741,7 @@ class HLoopInformation : public ArenaObject<kArenaAllocLoopInfo> {
   HBasicBlock* header_;
   HSuspendCheck* suspend_check_;
   bool irreducible_;
+  bool contains_irreducible_loop_;
   ArenaVector<HBasicBlock*> back_edges_;
   ArenaBitVector blocks_;
 
@@ -2281,7 +2290,7 @@ class HExpression : public HTemplateInstruction<N> {
 
 // Represents dex's RETURN_VOID opcode. A HReturnVoid is a control flow
 // instruction that branches to the exit block.
-class HReturnVoid : public HTemplateInstruction<0> {
+class HReturnVoid FINAL : public HTemplateInstruction<0> {
  public:
   explicit HReturnVoid(uint32_t dex_pc = kNoDexPc)
       : HTemplateInstruction(SideEffects::None(), dex_pc) {}
@@ -2296,7 +2305,7 @@ class HReturnVoid : public HTemplateInstruction<0> {
 
 // Represents dex's RETURN opcodes. A HReturn is a control flow
 // instruction that branches to the exit block.
-class HReturn : public HTemplateInstruction<1> {
+class HReturn FINAL : public HTemplateInstruction<1> {
  public:
   explicit HReturn(HInstruction* value, uint32_t dex_pc = kNoDexPc)
       : HTemplateInstruction(SideEffects::None(), dex_pc) {
@@ -2311,7 +2320,7 @@ class HReturn : public HTemplateInstruction<1> {
   DISALLOW_COPY_AND_ASSIGN(HReturn);
 };
 
-class HPhi : public HInstruction {
+class HPhi FINAL : public HInstruction {
  public:
   HPhi(ArenaAllocator* arena,
        uint32_t reg_number,
@@ -2415,7 +2424,7 @@ class HPhi : public HInstruction {
 // The exit instruction is the only instruction of the exit block.
 // Instructions aborting the method (HThrow and HReturn) must branch to the
 // exit block.
-class HExit : public HTemplateInstruction<0> {
+class HExit FINAL : public HTemplateInstruction<0> {
  public:
   explicit HExit(uint32_t dex_pc = kNoDexPc) : HTemplateInstruction(SideEffects::None(), dex_pc) {}
 
@@ -2428,7 +2437,7 @@ class HExit : public HTemplateInstruction<0> {
 };
 
 // Jumps from one block to another.
-class HGoto : public HTemplateInstruction<0> {
+class HGoto FINAL : public HTemplateInstruction<0> {
  public:
   explicit HGoto(uint32_t dex_pc = kNoDexPc) : HTemplateInstruction(SideEffects::None(), dex_pc) {}
 
@@ -2468,7 +2477,7 @@ class HConstant : public HExpression<0> {
   DISALLOW_COPY_AND_ASSIGN(HConstant);
 };
 
-class HNullConstant : public HConstant {
+class HNullConstant FINAL : public HConstant {
  public:
   bool InstructionDataEquals(HInstruction* other ATTRIBUTE_UNUSED) const OVERRIDE {
     return true;
@@ -2492,7 +2501,7 @@ class HNullConstant : public HConstant {
 
 // Constants of the type int. Those can be from Dex instructions, or
 // synthesized (for example with the if-eqz instruction).
-class HIntConstant : public HConstant {
+class HIntConstant FINAL : public HConstant {
  public:
   int32_t GetValue() const { return value_; }
 
@@ -2533,7 +2542,7 @@ class HIntConstant : public HConstant {
   DISALLOW_COPY_AND_ASSIGN(HIntConstant);
 };
 
-class HLongConstant : public HConstant {
+class HLongConstant FINAL : public HConstant {
  public:
   int64_t GetValue() const { return value_; }
 
@@ -2563,7 +2572,7 @@ class HLongConstant : public HConstant {
   DISALLOW_COPY_AND_ASSIGN(HLongConstant);
 };
 
-class HFloatConstant : public HConstant {
+class HFloatConstant FINAL : public HConstant {
  public:
   float GetValue() const { return value_; }
 
@@ -2616,7 +2625,7 @@ class HFloatConstant : public HConstant {
   DISALLOW_COPY_AND_ASSIGN(HFloatConstant);
 };
 
-class HDoubleConstant : public HConstant {
+class HDoubleConstant FINAL : public HConstant {
  public:
   double GetValue() const { return value_; }
 
@@ -2669,7 +2678,7 @@ class HDoubleConstant : public HConstant {
 
 // Conditional branch. A block ending with an HIf instruction must have
 // two successors.
-class HIf : public HTemplateInstruction<1> {
+class HIf FINAL : public HTemplateInstruction<1> {
  public:
   explicit HIf(HInstruction* input, uint32_t dex_pc = kNoDexPc)
       : HTemplateInstruction(SideEffects::None(), dex_pc) {
@@ -2698,7 +2707,7 @@ class HIf : public HTemplateInstruction<1> {
 // non-exceptional control flow.
 // Normal-flow successor is stored at index zero, exception handlers under
 // higher indices in no particular order.
-class HTryBoundary : public HTemplateInstruction<0> {
+class HTryBoundary FINAL : public HTemplateInstruction<0> {
  public:
   enum class BoundaryKind {
     kEntry,
@@ -2756,7 +2765,7 @@ class HTryBoundary : public HTemplateInstruction<0> {
 };
 
 // Deoptimize to interpreter, upon checking a condition.
-class HDeoptimize : public HTemplateInstruction<1> {
+class HDeoptimize FINAL : public HTemplateInstruction<1> {
  public:
   // We set CanTriggerGC to prevent any intermediate address to be live
   // at the point of the `HDeoptimize`.
@@ -2781,7 +2790,7 @@ class HDeoptimize : public HTemplateInstruction<1> {
 // Represents the ArtMethod that was passed as a first argument to
 // the method. It is used by instructions that depend on it, like
 // instructions that work with the dex cache.
-class HCurrentMethod : public HExpression<0> {
+class HCurrentMethod FINAL : public HExpression<0> {
  public:
   explicit HCurrentMethod(Primitive::Type type, uint32_t dex_pc = kNoDexPc)
       : HExpression(type, SideEffects::None(), dex_pc) {}
@@ -2794,7 +2803,7 @@ class HCurrentMethod : public HExpression<0> {
 
 // Fetches an ArtMethod from the virtual table or the interface method table
 // of a class.
-class HClassTableGet : public HExpression<1> {
+class HClassTableGet FINAL : public HExpression<1> {
  public:
   enum class TableKind {
     kVTable,
@@ -2841,7 +2850,7 @@ class HClassTableGet : public HExpression<1> {
 // PackedSwitch (jump table). A block ending with a PackedSwitch instruction will
 // have one successor for each entry in the switch table, and the final successor
 // will be the block containing the next Dex opcode.
-class HPackedSwitch : public HTemplateInstruction<1> {
+class HPackedSwitch FINAL : public HTemplateInstruction<1> {
  public:
   HPackedSwitch(int32_t start_value,
                 uint32_t num_entries,
@@ -3086,7 +3095,7 @@ class HCondition : public HBinaryOperation {
 };
 
 // Instruction to check if two inputs are equal to each other.
-class HEqual : public HCondition {
+class HEqual FINAL : public HCondition {
  public:
   HEqual(HInstruction* first, HInstruction* second, uint32_t dex_pc = kNoDexPc)
       : HCondition(first, second, dex_pc) {}
@@ -3130,7 +3139,7 @@ class HEqual : public HCondition {
   DISALLOW_COPY_AND_ASSIGN(HEqual);
 };
 
-class HNotEqual : public HCondition {
+class HNotEqual FINAL : public HCondition {
  public:
   HNotEqual(HInstruction* first, HInstruction* second, uint32_t dex_pc = kNoDexPc)
       : HCondition(first, second, dex_pc) {}
@@ -3173,7 +3182,7 @@ class HNotEqual : public HCondition {
   DISALLOW_COPY_AND_ASSIGN(HNotEqual);
 };
 
-class HLessThan : public HCondition {
+class HLessThan FINAL : public HCondition {
  public:
   HLessThan(HInstruction* first, HInstruction* second, uint32_t dex_pc = kNoDexPc)
       : HCondition(first, second, dex_pc) {}
@@ -3210,7 +3219,7 @@ class HLessThan : public HCondition {
   DISALLOW_COPY_AND_ASSIGN(HLessThan);
 };
 
-class HLessThanOrEqual : public HCondition {
+class HLessThanOrEqual FINAL : public HCondition {
  public:
   HLessThanOrEqual(HInstruction* first, HInstruction* second, uint32_t dex_pc = kNoDexPc)
       : HCondition(first, second, dex_pc) {}
@@ -3247,7 +3256,7 @@ class HLessThanOrEqual : public HCondition {
   DISALLOW_COPY_AND_ASSIGN(HLessThanOrEqual);
 };
 
-class HGreaterThan : public HCondition {
+class HGreaterThan FINAL : public HCondition {
  public:
   HGreaterThan(HInstruction* first, HInstruction* second, uint32_t dex_pc = kNoDexPc)
       : HCondition(first, second, dex_pc) {}
@@ -3284,7 +3293,7 @@ class HGreaterThan : public HCondition {
   DISALLOW_COPY_AND_ASSIGN(HGreaterThan);
 };
 
-class HGreaterThanOrEqual : public HCondition {
+class HGreaterThanOrEqual FINAL : public HCondition {
  public:
   HGreaterThanOrEqual(HInstruction* first, HInstruction* second, uint32_t dex_pc = kNoDexPc)
       : HCondition(first, second, dex_pc) {}
@@ -3321,7 +3330,7 @@ class HGreaterThanOrEqual : public HCondition {
   DISALLOW_COPY_AND_ASSIGN(HGreaterThanOrEqual);
 };
 
-class HBelow : public HCondition {
+class HBelow FINAL : public HCondition {
  public:
   HBelow(HInstruction* first, HInstruction* second, uint32_t dex_pc = kNoDexPc)
       : HCondition(first, second, dex_pc) {}
@@ -3361,7 +3370,7 @@ class HBelow : public HCondition {
   DISALLOW_COPY_AND_ASSIGN(HBelow);
 };
 
-class HBelowOrEqual : public HCondition {
+class HBelowOrEqual FINAL : public HCondition {
  public:
   HBelowOrEqual(HInstruction* first, HInstruction* second, uint32_t dex_pc = kNoDexPc)
       : HCondition(first, second, dex_pc) {}
@@ -3401,7 +3410,7 @@ class HBelowOrEqual : public HCondition {
   DISALLOW_COPY_AND_ASSIGN(HBelowOrEqual);
 };
 
-class HAbove : public HCondition {
+class HAbove FINAL : public HCondition {
  public:
   HAbove(HInstruction* first, HInstruction* second, uint32_t dex_pc = kNoDexPc)
       : HCondition(first, second, dex_pc) {}
@@ -3441,7 +3450,7 @@ class HAbove : public HCondition {
   DISALLOW_COPY_AND_ASSIGN(HAbove);
 };
 
-class HAboveOrEqual : public HCondition {
+class HAboveOrEqual FINAL : public HCondition {
  public:
   HAboveOrEqual(HInstruction* first, HInstruction* second, uint32_t dex_pc = kNoDexPc)
       : HCondition(first, second, dex_pc) {}
@@ -3483,7 +3492,7 @@ class HAboveOrEqual : public HCondition {
 
 // Instruction to check how two inputs compare to each other.
 // Result is 0 if input0 == input1, 1 if input0 > input1, or -1 if input0 < input1.
-class HCompare : public HBinaryOperation {
+class HCompare FINAL : public HBinaryOperation {
  public:
   // Note that `comparison_type` is the type of comparison performed
   // between the comparison's inputs, not the type of the instantiated
@@ -3572,7 +3581,7 @@ class HCompare : public HBinaryOperation {
   DISALLOW_COPY_AND_ASSIGN(HCompare);
 };
 
-class HNewInstance : public HExpression<2> {
+class HNewInstance FINAL : public HExpression<2> {
  public:
   HNewInstance(HInstruction* cls,
                HCurrentMethod* current_method,
@@ -3775,7 +3784,7 @@ class HInvoke : public HInstruction {
   DISALLOW_COPY_AND_ASSIGN(HInvoke);
 };
 
-class HInvokeUnresolved : public HInvoke {
+class HInvokeUnresolved FINAL : public HInvoke {
  public:
   HInvokeUnresolved(ArenaAllocator* arena,
                     uint32_t number_of_arguments,
@@ -3798,7 +3807,7 @@ class HInvokeUnresolved : public HInvoke {
   DISALLOW_COPY_AND_ASSIGN(HInvokeUnresolved);
 };
 
-class HInvokeStaticOrDirect : public HInvoke {
+class HInvokeStaticOrDirect FINAL : public HInvoke {
  public:
   // Requirements of this method call regarding the class
   // initialization (clinit) check of its declaring class.
@@ -4087,7 +4096,7 @@ class HInvokeStaticOrDirect : public HInvoke {
 std::ostream& operator<<(std::ostream& os, HInvokeStaticOrDirect::MethodLoadKind rhs);
 std::ostream& operator<<(std::ostream& os, HInvokeStaticOrDirect::ClinitCheckRequirement rhs);
 
-class HInvokeVirtual : public HInvoke {
+class HInvokeVirtual FINAL : public HInvoke {
  public:
   HInvokeVirtual(ArenaAllocator* arena,
                  uint32_t number_of_arguments,
@@ -4113,7 +4122,7 @@ class HInvokeVirtual : public HInvoke {
   DISALLOW_COPY_AND_ASSIGN(HInvokeVirtual);
 };
 
-class HInvokeInterface : public HInvoke {
+class HInvokeInterface FINAL : public HInvoke {
  public:
   HInvokeInterface(ArenaAllocator* arena,
                    uint32_t number_of_arguments,
@@ -4140,7 +4149,7 @@ class HInvokeInterface : public HInvoke {
   DISALLOW_COPY_AND_ASSIGN(HInvokeInterface);
 };
 
-class HNeg : public HUnaryOperation {
+class HNeg FINAL : public HUnaryOperation {
  public:
   HNeg(Primitive::Type result_type, HInstruction* input, uint32_t dex_pc = kNoDexPc)
       : HUnaryOperation(result_type, input, dex_pc) {
@@ -4168,7 +4177,7 @@ class HNeg : public HUnaryOperation {
   DISALLOW_COPY_AND_ASSIGN(HNeg);
 };
 
-class HNewArray : public HExpression<2> {
+class HNewArray FINAL : public HExpression<2> {
  public:
   HNewArray(HInstruction* length,
             HCurrentMethod* current_method,
@@ -4207,7 +4216,7 @@ class HNewArray : public HExpression<2> {
   DISALLOW_COPY_AND_ASSIGN(HNewArray);
 };
 
-class HAdd : public HBinaryOperation {
+class HAdd FINAL : public HBinaryOperation {
  public:
   HAdd(Primitive::Type result_type,
        HInstruction* left,
@@ -4242,7 +4251,7 @@ class HAdd : public HBinaryOperation {
   DISALLOW_COPY_AND_ASSIGN(HAdd);
 };
 
-class HSub : public HBinaryOperation {
+class HSub FINAL : public HBinaryOperation {
  public:
   HSub(Primitive::Type result_type,
        HInstruction* left,
@@ -4275,7 +4284,7 @@ class HSub : public HBinaryOperation {
   DISALLOW_COPY_AND_ASSIGN(HSub);
 };
 
-class HMul : public HBinaryOperation {
+class HMul FINAL : public HBinaryOperation {
  public:
   HMul(Primitive::Type result_type,
        HInstruction* left,
@@ -4310,7 +4319,7 @@ class HMul : public HBinaryOperation {
   DISALLOW_COPY_AND_ASSIGN(HMul);
 };
 
-class HDiv : public HBinaryOperation {
+class HDiv FINAL : public HBinaryOperation {
  public:
   HDiv(Primitive::Type result_type,
        HInstruction* left,
@@ -4362,7 +4371,7 @@ class HDiv : public HBinaryOperation {
   DISALLOW_COPY_AND_ASSIGN(HDiv);
 };
 
-class HRem : public HBinaryOperation {
+class HRem FINAL : public HBinaryOperation {
  public:
   HRem(Primitive::Type result_type,
        HInstruction* left,
@@ -4413,7 +4422,7 @@ class HRem : public HBinaryOperation {
   DISALLOW_COPY_AND_ASSIGN(HRem);
 };
 
-class HDivZeroCheck : public HExpression<1> {
+class HDivZeroCheck FINAL : public HExpression<1> {
  public:
   // `HDivZeroCheck` can trigger GC, as it may call the `ArithmeticException`
   // constructor.
@@ -4439,7 +4448,7 @@ class HDivZeroCheck : public HExpression<1> {
   DISALLOW_COPY_AND_ASSIGN(HDivZeroCheck);
 };
 
-class HShl : public HBinaryOperation {
+class HShl FINAL : public HBinaryOperation {
  public:
   HShl(Primitive::Type result_type,
        HInstruction* value,
@@ -4485,7 +4494,7 @@ class HShl : public HBinaryOperation {
   DISALLOW_COPY_AND_ASSIGN(HShl);
 };
 
-class HShr : public HBinaryOperation {
+class HShr FINAL : public HBinaryOperation {
  public:
   HShr(Primitive::Type result_type,
        HInstruction* value,
@@ -4531,7 +4540,7 @@ class HShr : public HBinaryOperation {
   DISALLOW_COPY_AND_ASSIGN(HShr);
 };
 
-class HUShr : public HBinaryOperation {
+class HUShr FINAL : public HBinaryOperation {
  public:
   HUShr(Primitive::Type result_type,
         HInstruction* value,
@@ -4579,7 +4588,7 @@ class HUShr : public HBinaryOperation {
   DISALLOW_COPY_AND_ASSIGN(HUShr);
 };
 
-class HAnd : public HBinaryOperation {
+class HAnd FINAL : public HBinaryOperation {
  public:
   HAnd(Primitive::Type result_type,
        HInstruction* left,
@@ -4616,7 +4625,7 @@ class HAnd : public HBinaryOperation {
   DISALLOW_COPY_AND_ASSIGN(HAnd);
 };
 
-class HOr : public HBinaryOperation {
+class HOr FINAL : public HBinaryOperation {
  public:
   HOr(Primitive::Type result_type,
       HInstruction* left,
@@ -4653,7 +4662,7 @@ class HOr : public HBinaryOperation {
   DISALLOW_COPY_AND_ASSIGN(HOr);
 };
 
-class HXor : public HBinaryOperation {
+class HXor FINAL : public HBinaryOperation {
  public:
   HXor(Primitive::Type result_type,
        HInstruction* left,
@@ -4690,7 +4699,7 @@ class HXor : public HBinaryOperation {
   DISALLOW_COPY_AND_ASSIGN(HXor);
 };
 
-class HRor : public HBinaryOperation {
+class HRor FINAL : public HBinaryOperation {
  public:
   HRor(Primitive::Type result_type, HInstruction* value, HInstruction* distance)
     : HBinaryOperation(result_type, value, distance) {
@@ -4743,7 +4752,7 @@ class HRor : public HBinaryOperation {
 
 // The value of a parameter in this method. Its location depends on
 // the calling convention.
-class HParameterValue : public HExpression<0> {
+class HParameterValue FINAL : public HExpression<0> {
  public:
   HParameterValue(const DexFile& dex_file,
                   uint16_t type_index,
@@ -4785,7 +4794,7 @@ class HParameterValue : public HExpression<0> {
   DISALLOW_COPY_AND_ASSIGN(HParameterValue);
 };
 
-class HNot : public HUnaryOperation {
+class HNot FINAL : public HUnaryOperation {
  public:
   HNot(Primitive::Type result_type, HInstruction* input, uint32_t dex_pc = kNoDexPc)
       : HUnaryOperation(result_type, input, dex_pc) {}
@@ -4818,7 +4827,7 @@ class HNot : public HUnaryOperation {
   DISALLOW_COPY_AND_ASSIGN(HNot);
 };
 
-class HBooleanNot : public HUnaryOperation {
+class HBooleanNot FINAL : public HUnaryOperation {
  public:
   explicit HBooleanNot(HInstruction* input, uint32_t dex_pc = kNoDexPc)
       : HUnaryOperation(Primitive::Type::kPrimBoolean, input, dex_pc) {}
@@ -4855,7 +4864,7 @@ class HBooleanNot : public HUnaryOperation {
   DISALLOW_COPY_AND_ASSIGN(HBooleanNot);
 };
 
-class HTypeConversion : public HExpression<1> {
+class HTypeConversion FINAL : public HExpression<1> {
  public:
   // Instantiate a type conversion of `input` to `result_type`.
   HTypeConversion(Primitive::Type result_type, HInstruction* input, uint32_t dex_pc)
@@ -4898,7 +4907,7 @@ class HTypeConversion : public HExpression<1> {
 
 static constexpr uint32_t kNoRegNumber = -1;
 
-class HNullCheck : public HExpression<1> {
+class HNullCheck FINAL : public HExpression<1> {
  public:
   // `HNullCheck` can trigger GC, as it may call the `NullPointerException`
   // constructor.
@@ -4960,7 +4969,7 @@ class FieldInfo : public ValueObject {
   const Handle<mirror::DexCache> dex_cache_;
 };
 
-class HInstanceFieldGet : public HExpression<1> {
+class HInstanceFieldGet FINAL : public HExpression<1> {
  public:
   HInstanceFieldGet(HInstruction* value,
                     Primitive::Type field_type,
@@ -5012,7 +5021,7 @@ class HInstanceFieldGet : public HExpression<1> {
   DISALLOW_COPY_AND_ASSIGN(HInstanceFieldGet);
 };
 
-class HInstanceFieldSet : public HTemplateInstruction<2> {
+class HInstanceFieldSet FINAL : public HTemplateInstruction<2> {
  public:
   HInstanceFieldSet(HInstruction* object,
                     HInstruction* value,
@@ -5063,7 +5072,7 @@ class HInstanceFieldSet : public HTemplateInstruction<2> {
   DISALLOW_COPY_AND_ASSIGN(HInstanceFieldSet);
 };
 
-class HArrayGet : public HExpression<2> {
+class HArrayGet FINAL : public HExpression<2> {
  public:
   HArrayGet(HInstruction* array, HInstruction* index, Primitive::Type type, uint32_t dex_pc)
       : HExpression(type, SideEffects::ArrayReadOfType(type), dex_pc) {
@@ -5109,7 +5118,7 @@ class HArrayGet : public HExpression<2> {
   DISALLOW_COPY_AND_ASSIGN(HArrayGet);
 };
 
-class HArraySet : public HTemplateInstruction<3> {
+class HArraySet FINAL : public HTemplateInstruction<3> {
  public:
   HArraySet(HInstruction* array,
             HInstruction* index,
@@ -5209,7 +5218,7 @@ class HArraySet : public HTemplateInstruction<3> {
   DISALLOW_COPY_AND_ASSIGN(HArraySet);
 };
 
-class HArrayLength : public HExpression<1> {
+class HArrayLength FINAL : public HExpression<1> {
  public:
   HArrayLength(HInstruction* array, uint32_t dex_pc)
       : HExpression(Primitive::kPrimInt, SideEffects::None(), dex_pc) {
@@ -5226,13 +5235,26 @@ class HArrayLength : public HExpression<1> {
     return obj == InputAt(0);
   }
 
+  void MarkAsStringLength() { SetPackedFlag<kFlagIsStringLength>(); }
+  bool IsStringLength() const { return GetPackedFlag<kFlagIsStringLength>(); }
+
   DECLARE_INSTRUCTION(ArrayLength);
 
  private:
+  // We treat a String as an array, creating the HArrayLength from String.length()
+  // or String.isEmpty() intrinsic in the instruction simplifier. We can always
+  // determine whether a particular HArrayLength is actually a String.length() by
+  // looking at the type of the input but that requires holding the mutator lock, so
+  // we prefer to use a flag, so that code generators don't need to do the locking.
+  static constexpr size_t kFlagIsStringLength = kNumberOfExpressionPackedBits;
+  static constexpr size_t kNumberOfArrayLengthPackedBits = kFlagIsStringLength + 1;
+  static_assert(kNumberOfArrayLengthPackedBits <= HInstruction::kMaxNumberOfPackedBits,
+                "Too many packed fields.");
+
   DISALLOW_COPY_AND_ASSIGN(HArrayLength);
 };
 
-class HBoundsCheck : public HExpression<2> {
+class HBoundsCheck FINAL : public HExpression<2> {
  public:
   // `HBoundsCheck` can trigger GC, as it may call the `IndexOutOfBoundsException`
   // constructor.
@@ -5260,7 +5282,7 @@ class HBoundsCheck : public HExpression<2> {
   DISALLOW_COPY_AND_ASSIGN(HBoundsCheck);
 };
 
-class HSuspendCheck : public HTemplateInstruction<0> {
+class HSuspendCheck FINAL : public HTemplateInstruction<0> {
  public:
   explicit HSuspendCheck(uint32_t dex_pc = kNoDexPc)
       : HTemplateInstruction(SideEffects::CanTriggerGC(), dex_pc), slow_path_(nullptr) {}
@@ -5302,7 +5324,7 @@ class HNativeDebugInfo : public HTemplateInstruction<0> {
 /**
  * Instruction to load a Class object.
  */
-class HLoadClass : public HExpression<1> {
+class HLoadClass FINAL : public HExpression<1> {
  public:
   HLoadClass(HCurrentMethod* current_method,
              uint16_t type_index,
@@ -5406,7 +5428,7 @@ class HLoadClass : public HExpression<1> {
   DISALLOW_COPY_AND_ASSIGN(HLoadClass);
 };
 
-class HLoadString : public HExpression<1> {
+class HLoadString FINAL : public HExpression<1> {
  public:
   // Determines how to load the String.
   enum class LoadKind {
@@ -5530,6 +5552,7 @@ class HLoadString : public HExpression<1> {
     SetPackedFlag<kFlagIsInDexCache>(true);
     DCHECK(!NeedsEnvironment());
     RemoveEnvironment();
+    SetSideEffects(SideEffects::None());
   }
 
   size_t InputCount() const OVERRIDE {
@@ -5607,7 +5630,7 @@ inline void HLoadString::AddSpecialInput(HInstruction* special_input) {
 /**
  * Performs an initialization check on its Class object input.
  */
-class HClinitCheck : public HExpression<1> {
+class HClinitCheck FINAL : public HExpression<1> {
  public:
   HClinitCheck(HLoadClass* constant, uint32_t dex_pc)
       : HExpression(
@@ -5637,7 +5660,7 @@ class HClinitCheck : public HExpression<1> {
   DISALLOW_COPY_AND_ASSIGN(HClinitCheck);
 };
 
-class HStaticFieldGet : public HExpression<1> {
+class HStaticFieldGet FINAL : public HExpression<1> {
  public:
   HStaticFieldGet(HInstruction* cls,
                   Primitive::Type field_type,
@@ -5686,7 +5709,7 @@ class HStaticFieldGet : public HExpression<1> {
   DISALLOW_COPY_AND_ASSIGN(HStaticFieldGet);
 };
 
-class HStaticFieldSet : public HTemplateInstruction<2> {
+class HStaticFieldSet FINAL : public HTemplateInstruction<2> {
  public:
   HStaticFieldSet(HInstruction* cls,
                   HInstruction* value,
@@ -5734,7 +5757,7 @@ class HStaticFieldSet : public HTemplateInstruction<2> {
   DISALLOW_COPY_AND_ASSIGN(HStaticFieldSet);
 };
 
-class HUnresolvedInstanceFieldGet : public HExpression<1> {
+class HUnresolvedInstanceFieldGet FINAL : public HExpression<1> {
  public:
   HUnresolvedInstanceFieldGet(HInstruction* obj,
                               Primitive::Type field_type,
@@ -5759,7 +5782,7 @@ class HUnresolvedInstanceFieldGet : public HExpression<1> {
   DISALLOW_COPY_AND_ASSIGN(HUnresolvedInstanceFieldGet);
 };
 
-class HUnresolvedInstanceFieldSet : public HTemplateInstruction<2> {
+class HUnresolvedInstanceFieldSet FINAL : public HTemplateInstruction<2> {
  public:
   HUnresolvedInstanceFieldSet(HInstruction* obj,
                               HInstruction* value,
@@ -5797,7 +5820,7 @@ class HUnresolvedInstanceFieldSet : public HTemplateInstruction<2> {
   DISALLOW_COPY_AND_ASSIGN(HUnresolvedInstanceFieldSet);
 };
 
-class HUnresolvedStaticFieldGet : public HExpression<0> {
+class HUnresolvedStaticFieldGet FINAL : public HExpression<0> {
  public:
   HUnresolvedStaticFieldGet(Primitive::Type field_type,
                             uint32_t field_index,
@@ -5820,7 +5843,7 @@ class HUnresolvedStaticFieldGet : public HExpression<0> {
   DISALLOW_COPY_AND_ASSIGN(HUnresolvedStaticFieldGet);
 };
 
-class HUnresolvedStaticFieldSet : public HTemplateInstruction<1> {
+class HUnresolvedStaticFieldSet FINAL : public HTemplateInstruction<1> {
  public:
   HUnresolvedStaticFieldSet(HInstruction* value,
                             Primitive::Type field_type,
@@ -5857,7 +5880,7 @@ class HUnresolvedStaticFieldSet : public HTemplateInstruction<1> {
 };
 
 // Implement the move-exception DEX instruction.
-class HLoadException : public HExpression<0> {
+class HLoadException FINAL : public HExpression<0> {
  public:
   explicit HLoadException(uint32_t dex_pc = kNoDexPc)
       : HExpression(Primitive::kPrimNot, SideEffects::None(), dex_pc) {}
@@ -5872,7 +5895,7 @@ class HLoadException : public HExpression<0> {
 
 // Implicit part of move-exception which clears thread-local exception storage.
 // Must not be removed because the runtime expects the TLS to get cleared.
-class HClearException : public HTemplateInstruction<0> {
+class HClearException FINAL : public HTemplateInstruction<0> {
  public:
   explicit HClearException(uint32_t dex_pc = kNoDexPc)
       : HTemplateInstruction(SideEffects::AllWrites(), dex_pc) {}
@@ -5883,7 +5906,7 @@ class HClearException : public HTemplateInstruction<0> {
   DISALLOW_COPY_AND_ASSIGN(HClearException);
 };
 
-class HThrow : public HTemplateInstruction<1> {
+class HThrow FINAL : public HTemplateInstruction<1> {
  public:
   HThrow(HInstruction* exception, uint32_t dex_pc)
       : HTemplateInstruction(SideEffects::CanTriggerGC(), dex_pc) {
@@ -5920,7 +5943,7 @@ enum class TypeCheckKind {
 
 std::ostream& operator<<(std::ostream& os, TypeCheckKind rhs);
 
-class HInstanceOf : public HExpression<2> {
+class HInstanceOf FINAL : public HExpression<2> {
  public:
   HInstanceOf(HInstruction* object,
               HLoadClass* constant,
@@ -5974,7 +5997,7 @@ class HInstanceOf : public HExpression<2> {
   DISALLOW_COPY_AND_ASSIGN(HInstanceOf);
 };
 
-class HBoundType : public HExpression<1> {
+class HBoundType FINAL : public HExpression<1> {
  public:
   HBoundType(HInstruction* input, uint32_t dex_pc = kNoDexPc)
       : HExpression(Primitive::kPrimNot, SideEffects::None(), dex_pc),
@@ -6018,7 +6041,7 @@ class HBoundType : public HExpression<1> {
   DISALLOW_COPY_AND_ASSIGN(HBoundType);
 };
 
-class HCheckCast : public HTemplateInstruction<2> {
+class HCheckCast FINAL : public HTemplateInstruction<2> {
  public:
   HCheckCast(HInstruction* object,
              HLoadClass* constant,
@@ -6063,7 +6086,7 @@ class HCheckCast : public HTemplateInstruction<2> {
   DISALLOW_COPY_AND_ASSIGN(HCheckCast);
 };
 
-class HMemoryBarrier : public HTemplateInstruction<0> {
+class HMemoryBarrier FINAL : public HTemplateInstruction<0> {
  public:
   explicit HMemoryBarrier(MemBarrierKind barrier_kind, uint32_t dex_pc = kNoDexPc)
       : HTemplateInstruction(
@@ -6088,7 +6111,7 @@ class HMemoryBarrier : public HTemplateInstruction<0> {
   DISALLOW_COPY_AND_ASSIGN(HMemoryBarrier);
 };
 
-class HMonitorOperation : public HTemplateInstruction<1> {
+class HMonitorOperation FINAL : public HTemplateInstruction<1> {
  public:
   enum class OperationKind {
     kEnter,
@@ -6133,7 +6156,7 @@ class HMonitorOperation : public HTemplateInstruction<1> {
   DISALLOW_COPY_AND_ASSIGN(HMonitorOperation);
 };
 
-class HSelect : public HExpression<3> {
+class HSelect FINAL : public HExpression<3> {
  public:
   HSelect(HInstruction* condition,
           HInstruction* true_value,
@@ -6246,7 +6269,7 @@ std::ostream& operator<<(std::ostream& os, const MoveOperands& rhs);
 
 static constexpr size_t kDefaultNumberOfMoves = 4;
 
-class HParallelMove : public HTemplateInstruction<0> {
+class HParallelMove FINAL : public HTemplateInstruction<0> {
  public:
   explicit HParallelMove(ArenaAllocator* arena, uint32_t dex_pc = kNoDexPc)
       : HTemplateInstruction(SideEffects::None(), dex_pc),
diff --git a/compiler/optimizing/nodes_arm.h b/compiler/optimizing/nodes_arm.h
index 6a1dbb9e7..371e8ef6b 100644
--- a/compiler/optimizing/nodes_arm.h
+++ b/compiler/optimizing/nodes_arm.h
@@ -19,7 +19,7 @@
 
 namespace art {
 
-class HArmDexCacheArraysBase : public HExpression<0> {
+class HArmDexCacheArraysBase FINAL : public HExpression<0> {
  public:
   explicit HArmDexCacheArraysBase(const DexFile& dex_file)
       : HExpression(Primitive::kPrimInt, SideEffects::None(), kNoDexPc),
diff --git a/compiler/optimizing/nodes_arm64.h b/compiler/optimizing/nodes_arm64.h
index 173852a55..737aece9c 100644
--- a/compiler/optimizing/nodes_arm64.h
+++ b/compiler/optimizing/nodes_arm64.h
@@ -21,7 +21,7 @@
 
 namespace art {
 
-class HArm64DataProcWithShifterOp : public HExpression<2> {
+class HArm64DataProcWithShifterOp FINAL : public HExpression<2> {
  public:
   enum OpKind {
     kLSL,   // Logical shift left.
@@ -97,7 +97,7 @@ std::ostream& operator<<(std::ostream& os, const HArm64DataProcWithShifterOp::Op
 // This instruction computes an intermediate address pointing in the 'middle' of an object. The
 // result pointer cannot be handled by GC, so extra care is taken to make sure that this value is
 // never used across anything that can trigger GC.
-class HArm64IntermediateAddress : public HExpression<2> {
+class HArm64IntermediateAddress FINAL : public HExpression<2> {
  public:
   HArm64IntermediateAddress(HInstruction* base_address, HInstruction* offset, uint32_t dex_pc)
       : HExpression(Primitive::kPrimNot, SideEffects::DependsOnGC(), dex_pc) {
diff --git a/compiler/optimizing/nodes_shared.h b/compiler/optimizing/nodes_shared.h
index c10c718ff..bdcf54a6f 100644
--- a/compiler/optimizing/nodes_shared.h
+++ b/compiler/optimizing/nodes_shared.h
@@ -19,7 +19,7 @@
 
 namespace art {
 
-class HMultiplyAccumulate : public HExpression<3> {
+class HMultiplyAccumulate FINAL : public HExpression<3> {
  public:
   HMultiplyAccumulate(Primitive::Type type,
                       InstructionKind op,
@@ -53,7 +53,7 @@ class HMultiplyAccumulate : public HExpression<3> {
   DISALLOW_COPY_AND_ASSIGN(HMultiplyAccumulate);
 };
 
-class HBitwiseNegatedRight : public HBinaryOperation {
+class HBitwiseNegatedRight FINAL : public HBinaryOperation {
  public:
   HBitwiseNegatedRight(Primitive::Type result_type,
                             InstructionKind op,
diff --git a/compiler/optimizing/nodes_x86.h b/compiler/optimizing/nodes_x86.h
index 0b3a84d3d..c3696b593 100644
--- a/compiler/optimizing/nodes_x86.h
+++ b/compiler/optimizing/nodes_x86.h
@@ -20,7 +20,7 @@
 namespace art {
 
 // Compute the address of the method for X86 Constant area support.
-class HX86ComputeBaseMethodAddress : public HExpression<0> {
+class HX86ComputeBaseMethodAddress FINAL : public HExpression<0> {
  public:
   // Treat the value as an int32_t, but it is really a 32 bit native pointer.
   HX86ComputeBaseMethodAddress()
@@ -33,7 +33,7 @@ class HX86ComputeBaseMethodAddress : public HExpression<0> {
 };
 
 // Load a constant value from the constant table.
-class HX86LoadFromConstantTable : public HExpression<2> {
+class HX86LoadFromConstantTable FINAL : public HExpression<2> {
  public:
   HX86LoadFromConstantTable(HX86ComputeBaseMethodAddress* method_base,
                             HConstant* constant)
@@ -57,7 +57,7 @@ class HX86LoadFromConstantTable : public HExpression<2> {
 };
 
 // Version of HNeg with access to the constant table for FP types.
-class HX86FPNeg : public HExpression<2> {
+class HX86FPNeg FINAL : public HExpression<2> {
  public:
   HX86FPNeg(Primitive::Type result_type,
             HInstruction* input,
@@ -76,7 +76,7 @@ class HX86FPNeg : public HExpression<2> {
 };
 
 // X86 version of HPackedSwitch that holds a pointer to the base method address.
-class HX86PackedSwitch : public HTemplateInstruction<2> {
+class HX86PackedSwitch FINAL : public HTemplateInstruction<2> {
  public:
   HX86PackedSwitch(int32_t start_value,
                    int32_t num_entries,
diff --git a/compiler/optimizing/reference_type_propagation.cc b/compiler/optimizing/reference_type_propagation.cc
index 04c9ff9d6..f2394f605 100644
--- a/compiler/optimizing/reference_type_propagation.cc
+++ b/compiler/optimizing/reference_type_propagation.cc
@@ -46,6 +46,13 @@ static inline ReferenceTypeInfo::TypeHandle GetRootHandle(StackHandleScopeCollec
   return *cache;
 }
 
+// Returns true if klass is admissible to the propagation: non-null and non-erroneous.
+// For an array type, we also check if the component type is admissible.
+static bool IsAdmissible(mirror::Class* klass) SHARED_REQUIRES(Locks::mutator_lock_) {
+  return klass != nullptr && !klass->IsErroneous() &&
+      (!klass->IsArrayClass() || IsAdmissible(klass->GetComponentType()));
+}
+
 ReferenceTypeInfo::TypeHandle ReferenceTypePropagation::HandleCache::GetObjectClassHandle() {
   return GetRootHandle(handles_, ClassLinker::kJavaLangObject, &object_class_handle_);
 }
@@ -453,15 +460,10 @@ void ReferenceTypePropagation::RTPVisitor::SetClassAsTypeInfo(HInstruction* inst
     }
     instr->SetReferenceTypeInfo(
         ReferenceTypeInfo::Create(handle_cache_->GetStringClassHandle(), /* is_exact */ true));
-  } else if (klass != nullptr) {
-    if (klass->IsErroneous()) {
-      // Set inexact object type for erroneous types.
-      instr->SetReferenceTypeInfo(instr->GetBlock()->GetGraph()->GetInexactObjectRti());
-    } else {
-      ReferenceTypeInfo::TypeHandle handle = handle_cache_->NewHandle(klass);
-      is_exact = is_exact || handle->CannotBeAssignedFromOtherTypes();
-      instr->SetReferenceTypeInfo(ReferenceTypeInfo::Create(handle, is_exact));
-    }
+  } else if (IsAdmissible(klass)) {
+    ReferenceTypeInfo::TypeHandle handle = handle_cache_->NewHandle(klass);
+    is_exact = is_exact || handle->CannotBeAssignedFromOtherTypes();
+    instr->SetReferenceTypeInfo(ReferenceTypeInfo::Create(handle, is_exact));
   } else {
     instr->SetReferenceTypeInfo(instr->GetBlock()->GetGraph()->GetInexactObjectRti());
   }
@@ -563,7 +565,7 @@ void ReferenceTypePropagation::RTPVisitor::VisitLoadClass(HLoadClass* instr) {
                                                        instr->GetDexFile(),
                                                        instr->GetTypeIndex(),
                                                        hint_dex_cache_);
-  if (resolved_class != nullptr && !resolved_class->IsErroneous()) {
+  if (IsAdmissible(resolved_class)) {
     instr->SetLoadedClassRTI(ReferenceTypeInfo::Create(
         handle_cache_->NewHandle(resolved_class), /* is_exact */ true));
   }
@@ -664,12 +666,6 @@ void ReferenceTypePropagation::VisitPhi(HPhi* phi) {
   }
 
   if (phi->GetBlock()->IsLoopHeader()) {
-    if (!is_first_run_ && graph_->IsCompilingOsr()) {
-      // Don't update the type of a loop phi when compiling OSR: we may have done
-      // speculative optimizations dominating that phi, that do not hold at the
-      // point the interpreter jumps to that loop header.
-      return;
-    }
     // Set the initial type for the phi. Use the non back edge input for reaching
     // a fixed point faster.
     HInstruction* first_input = phi->InputAt(0);
@@ -742,7 +738,7 @@ void ReferenceTypePropagation::UpdateArrayGet(HArrayGet* instr, HandleCache* han
   }
 
   Handle<mirror::Class> handle = parent_rti.GetTypeHandle();
-  if (handle->IsObjectArrayClass() && !handle->GetComponentType()->IsErroneous()) {
+  if (handle->IsObjectArrayClass() && IsAdmissible(handle->GetComponentType())) {
     ReferenceTypeInfo::TypeHandle component_handle =
         handle_cache->NewHandle(handle->GetComponentType());
     bool is_exact = component_handle->CannotBeAssignedFromOtherTypes();
diff --git a/compiler/optimizing/register_allocator.cc b/compiler/optimizing/register_allocator.cc
index b1f9cbcdf..4405b803e 100644
--- a/compiler/optimizing/register_allocator.cc
+++ b/compiler/optimizing/register_allocator.cc
@@ -1773,7 +1773,9 @@ void RegisterAllocator::ConnectSplitSiblings(LiveInterval* interval,
     // therefore will not have a location for that instruction for `to`.
     // Because the instruction is a constant or the ArtMethod, we don't need to
     // do anything: it will be materialized in the irreducible loop.
-    DCHECK(IsMaterializableEntryBlockInstructionOfGraphWithIrreducibleLoop(defined_by));
+    DCHECK(IsMaterializableEntryBlockInstructionOfGraphWithIrreducibleLoop(defined_by))
+        << defined_by->DebugName() << ":" << defined_by->GetId()
+        << " " << from->GetBlockId() << " -> " << to->GetBlockId();
     return;
   }
 
diff --git a/compiler/optimizing/sharpening.cc b/compiler/optimizing/sharpening.cc
index 7a1bb316e..08bd35f14 100644
--- a/compiler/optimizing/sharpening.cc
+++ b/compiler/optimizing/sharpening.cc
@@ -99,7 +99,7 @@ void HSharpening::ProcessInvokeStaticOrDirect(HInvokeStaticOrDirect* invoke) {
     if (direct_method != 0u) {  // Should we use a direct pointer to the method?
       // Note: For JIT, kDirectAddressWithFixup doesn't make sense at all and while
       // kDirectAddress would be fine for image methods, we don't support it at the moment.
-      DCHECK(!Runtime::Current()->UseJit());
+      DCHECK(!Runtime::Current()->UseJitCompilation());
       if (direct_method != static_cast<uintptr_t>(-1)) {  // Is the method pointer known now?
         method_load_kind = HInvokeStaticOrDirect::MethodLoadKind::kDirectAddress;
         method_load_data = direct_method;
@@ -109,7 +109,7 @@ void HSharpening::ProcessInvokeStaticOrDirect(HInvokeStaticOrDirect* invoke) {
     } else {  // Use dex cache.
       DCHECK_EQ(target_method.dex_file, &graph_->GetDexFile());
       if (use_pc_relative_instructions) {  // Can we use PC-relative access to the dex cache arrays?
-        DCHECK(!Runtime::Current()->UseJit());
+        DCHECK(!Runtime::Current()->UseJitCompilation());
         method_load_kind = HInvokeStaticOrDirect::MethodLoadKind::kDexCachePcRelative;
         DexCacheArraysLayout layout(GetInstructionSetPointerSize(codegen_->GetInstructionSet()),
                                     &graph_->GetDexFile());
@@ -121,7 +121,7 @@ void HSharpening::ProcessInvokeStaticOrDirect(HInvokeStaticOrDirect* invoke) {
     if (direct_code != 0u) {  // Should we use a direct pointer to the code?
       // Note: For JIT, kCallPCRelative and kCallDirectWithFixup don't make sense at all and
       // while kCallDirect would be fine for image methods, we don't support it at the moment.
-      DCHECK(!Runtime::Current()->UseJit());
+      DCHECK(!Runtime::Current()->UseJitCompilation());
       if (direct_code != static_cast<uintptr_t>(-1)) {  // Is the code pointer known now?
         code_ptr_location = HInvokeStaticOrDirect::CodePtrLocation::kCallDirect;
         direct_code_ptr = direct_code;
@@ -174,7 +174,7 @@ void HSharpening::ProcessLoadString(HLoadString* load_string) {
 
     if (compiler_driver_->IsBootImage()) {
       // Compiling boot image. Resolve the string and allocate it if needed.
-      DCHECK(!runtime->UseJit());
+      DCHECK(!runtime->UseJitCompilation());
       mirror::String* string = class_linker->ResolveString(dex_file, string_index, dex_cache);
       CHECK(string != nullptr);
       if (!compiler_driver_->GetSupportBootImageFixup()) {
@@ -187,7 +187,7 @@ void HSharpening::ProcessLoadString(HLoadString* load_string) {
             ? HLoadString::LoadKind::kBootImageLinkTimePcRelative
             : HLoadString::LoadKind::kBootImageLinkTimeAddress;
       }
-    } else if (runtime->UseJit()) {
+    } else if (runtime->UseJitCompilation()) {
       // TODO: Make sure we don't set the "compile PIC" flag for JIT as that's bogus.
       // DCHECK(!codegen_->GetCompilerOptions().GetCompilePic());
       mirror::String* string = dex_cache->GetResolvedString(string_index);
diff --git a/compiler/optimizing/ssa_builder.cc b/compiler/optimizing/ssa_builder.cc
index c2aa0c007..f96ca321c 100644
--- a/compiler/optimizing/ssa_builder.cc
+++ b/compiler/optimizing/ssa_builder.cc
@@ -233,7 +233,7 @@ bool SsaBuilder::UpdatePrimitiveType(HPhi* phi, ArenaVector<HPhi*>* worklist) {
 }
 
 void SsaBuilder::RunPrimitiveTypePropagation() {
-  ArenaVector<HPhi*> worklist(graph_->GetArena()->Adapter());
+  ArenaVector<HPhi*> worklist(graph_->GetArena()->Adapter(kArenaAllocGraphBuilder));
 
   for (HReversePostOrderIterator it(*graph_); !it.Done(); it.Advance()) {
     HBasicBlock* block = it.Current();
@@ -319,7 +319,7 @@ bool SsaBuilder::FixAmbiguousArrayOps() {
   // uses (because they are untyped) and environment uses (if --debuggable).
   // After resolving all ambiguous ArrayGets, we will re-run primitive type
   // propagation on the Phis which need to be updated.
-  ArenaVector<HPhi*> worklist(graph_->GetArena()->Adapter());
+  ArenaVector<HPhi*> worklist(graph_->GetArena()->Adapter(kArenaAllocGraphBuilder));
 
   {
     ScopedObjectAccess soa(Thread::Current());
diff --git a/compiler/optimizing/ssa_liveness_analysis.cc b/compiler/optimizing/ssa_liveness_analysis.cc
index 5534aeac2..36e0d993d 100644
--- a/compiler/optimizing/ssa_liveness_analysis.cc
+++ b/compiler/optimizing/ssa_liveness_analysis.cc
@@ -309,17 +309,8 @@ void SsaLivenessAnalysis::ComputeLiveRanges() {
     }
 
     if (block->IsLoopHeader()) {
-      if (kIsDebugBuild && block->GetLoopInformation()->IsIrreducible()) {
-        // To satisfy our liveness algorithm, we need to ensure loop headers of
-        // irreducible loops do not have any live-in instructions, except constants
-        // and the current method, which can be trivially re-materialized.
-        for (uint32_t idx : live_in->Indexes()) {
-          HInstruction* instruction = GetInstructionFromSsaIndex(idx);
-          DCHECK(instruction->GetBlock()->IsEntryBlock()) << instruction->DebugName();
-          DCHECK(!instruction->IsParameterValue());
-          DCHECK(instruction->IsCurrentMethod() || instruction->IsConstant())
-              << instruction->DebugName();
-        }
+      if (kIsDebugBuild) {
+        CheckNoLiveInIrreducibleLoop(*block);
       }
       size_t last_position = block->GetLoopInformation()->GetLifetimeEnd();
       // For all live_in instructions at the loop header, we need to create a range
@@ -344,6 +335,9 @@ void SsaLivenessAnalysis::ComputeLiveInAndLiveOutSets() {
       // change in this loop), and the live_out set.  If the live_out
       // set does not change, there is no need to update the live_in set.
       if (UpdateLiveOut(block) && UpdateLiveIn(block)) {
+        if (kIsDebugBuild) {
+          CheckNoLiveInIrreducibleLoop(block);
+        }
         changed = true;
       }
     }
diff --git a/compiler/optimizing/ssa_liveness_analysis.h b/compiler/optimizing/ssa_liveness_analysis.h
index 40dab74a2..1fcba8bc7 100644
--- a/compiler/optimizing/ssa_liveness_analysis.h
+++ b/compiler/optimizing/ssa_liveness_analysis.h
@@ -1003,6 +1003,15 @@ class LiveInterval : public ArenaObject<kArenaAllocSsaLiveness> {
 
   void AddBackEdgeUses(const HBasicBlock& block_at_use) {
     DCHECK(block_at_use.IsInLoop());
+    if (block_at_use.GetGraph()->HasIrreducibleLoops()) {
+      // Linear order may not be well formed when irreducible loops are present,
+      // i.e. loop blocks may not be adjacent and a back edge may not be last,
+      // which violates assumptions made in this method.
+      return;
+    }
+
+    DCHECK(IsLinearOrderWellFormed(*block_at_use.GetGraph()));
+
     // Add synthesized uses at the back edge of loops to help the register allocator.
     // Note that this method is called in decreasing liveness order, to faciliate adding
     // uses at the head of the `first_use_` linked list. Because below
@@ -1027,30 +1036,12 @@ class LiveInterval : public ArenaObject<kArenaAllocSsaLiveness> {
       if ((first_use_ != nullptr) && (first_use_->GetPosition() <= back_edge_use_position)) {
         // There was a use already seen in this loop. Therefore the previous call to `AddUse`
         // already inserted the backedge use. We can stop going outward.
-        if (kIsDebugBuild) {
-          if (!HasSynthesizeUseAt(back_edge_use_position)) {
-            // There exists a use prior to `back_edge_use_position` but there is
-            // no synthesized use at the back edge. This can happen in the presence
-            // of irreducible loops, when blocks of the loop are not adjacent in
-            // linear order, i.e. when there is an out-of-loop block between
-            // `block_at_use` and `back_edge_position` that uses this interval.
-            DCHECK(block_at_use.GetGraph()->HasIrreducibleLoops());
-            DCHECK(!IsLinearOrderWellFormed(*block_at_use.GetGraph()));
-          }
-        }
+        DCHECK(HasSynthesizeUseAt(back_edge_use_position));
         break;
       }
 
-      if (last_in_new_list != nullptr &&
-          back_edge_use_position <= last_in_new_list->GetPosition()) {
-        // Loops are not properly nested in the linear order, i.e. the back edge
-        // of an outer loop preceeds blocks of an inner loop. This can happen
-        // in the presence of irreducible loops.
-        DCHECK(block_at_use.GetGraph()->HasIrreducibleLoops());
-        DCHECK(!IsLinearOrderWellFormed(*block_at_use.GetGraph()));
-        // We must bail out, otherwise we would generate an unsorted use list.
-        break;
-      }
+      DCHECK(last_in_new_list == nullptr ||
+             back_edge_use_position > last_in_new_list->GetPosition());
 
       UsePosition* new_use = new (allocator_) UsePosition(
           /* user */ nullptr,
@@ -1269,6 +1260,23 @@ class SsaLivenessAnalysis : public ValueObject {
     return instruction->GetType() == Primitive::kPrimNot;
   }
 
+  void CheckNoLiveInIrreducibleLoop(const HBasicBlock& block) const {
+    if (!block.IsLoopHeader() || !block.GetLoopInformation()->IsIrreducible()) {
+      return;
+    }
+    BitVector* live_in = GetLiveInSet(block);
+    // To satisfy our liveness algorithm, we need to ensure loop headers of
+    // irreducible loops do not have any live-in instructions, except constants
+    // and the current method, which can be trivially re-materialized.
+    for (uint32_t idx : live_in->Indexes()) {
+      HInstruction* instruction = GetInstructionFromSsaIndex(idx);
+      DCHECK(instruction->GetBlock()->IsEntryBlock()) << instruction->DebugName();
+      DCHECK(!instruction->IsParameterValue());
+      DCHECK(instruction->IsCurrentMethod() || instruction->IsConstant())
+          << instruction->DebugName();
+    }
+  }
+
   HGraph* const graph_;
   CodeGenerator* const codegen_;
   ArenaVector<BlockInfo*> block_infos_;
diff --git a/compiler/optimizing/ssa_phi_elimination.cc b/compiler/optimizing/ssa_phi_elimination.cc
index aeb31094d..c67612e65 100644
--- a/compiler/optimizing/ssa_phi_elimination.cc
+++ b/compiler/optimizing/ssa_phi_elimination.cc
@@ -17,6 +17,7 @@
 #include "ssa_phi_elimination.h"
 
 #include "base/arena_containers.h"
+#include "base/arena_bit_vector.h"
 #include "base/bit_vector-inl.h"
 
 namespace art {
@@ -30,7 +31,7 @@ void SsaDeadPhiElimination::MarkDeadPhis() {
   // Phis are constructed live and should not be revived if previously marked
   // dead. This algorithm temporarily breaks that invariant but we DCHECK that
   // only phis which were initially live are revived.
-  ArenaSet<HPhi*> initially_live(graph_->GetArena()->Adapter());
+  ArenaSet<HPhi*> initially_live(graph_->GetArena()->Adapter(kArenaAllocSsaPhiElimination));
 
   // Add to the worklist phis referenced by non-phi instructions.
   for (HReversePostOrderIterator it(*graph_); !it.Done(); it.Advance()) {
@@ -127,8 +128,11 @@ void SsaRedundantPhiElimination::Run() {
     }
   }
 
-  ArenaSet<uint32_t> visited_phis_in_cycle(graph_->GetArena()->Adapter());
-  ArenaVector<HPhi*> cycle_worklist(graph_->GetArena()->Adapter());
+  ArenaBitVector visited_phis_in_cycle(graph_->GetArena(),
+                                       graph_->GetCurrentInstructionId(),
+                                       /* expandable */ false,
+                                       kArenaAllocSsaPhiElimination);
+  ArenaVector<HPhi*> cycle_worklist(graph_->GetArena()->Adapter(kArenaAllocSsaPhiElimination));
 
   while (!worklist_.empty()) {
     HPhi* phi = worklist_.back();
@@ -139,17 +143,18 @@ void SsaRedundantPhiElimination::Run() {
       continue;
     }
 
-    if (phi->InputCount() == 0) {
-      DCHECK(phi->IsDead());
+    // If the phi is dead, we know we won't revive it and it will be removed,
+    // so don't process it.
+    if (phi->IsDead()) {
       continue;
     }
 
     HInstruction* candidate = nullptr;
-    visited_phis_in_cycle.clear();
+    visited_phis_in_cycle.ClearAllBits();
     cycle_worklist.clear();
 
     cycle_worklist.push_back(phi);
-    visited_phis_in_cycle.insert(phi->GetId());
+    visited_phis_in_cycle.SetBit(phi->GetId());
     bool catch_phi_in_cycle = phi->IsCatchPhi();
     bool irreducible_loop_phi_in_cycle = phi->IsIrreducibleLoopHeaderPhi();
 
@@ -181,9 +186,9 @@ void SsaRedundantPhiElimination::Run() {
           if (input == current) {
             continue;
           } else if (input->IsPhi()) {
-            if (!ContainsElement(visited_phis_in_cycle, input->GetId())) {
+            if (!visited_phis_in_cycle.IsBitSet(input->GetId())) {
               cycle_worklist.push_back(input->AsPhi());
-              visited_phis_in_cycle.insert(input->GetId());
+              visited_phis_in_cycle.SetBit(input->GetId());
               catch_phi_in_cycle |= input->AsPhi()->IsCatchPhi();
               irreducible_loop_phi_in_cycle |= input->IsIrreducibleLoopHeaderPhi();
             } else {
@@ -232,7 +237,7 @@ void SsaRedundantPhiElimination::Run() {
       // for elimination. Add phis that use this phi to the worklist.
       for (const HUseListNode<HInstruction*>& use : current->GetUses()) {
         HInstruction* user = use.GetUser();
-        if (user->IsPhi() && !ContainsElement(visited_phis_in_cycle, user->GetId())) {
+        if (user->IsPhi() && !visited_phis_in_cycle.IsBitSet(user->GetId())) {
           worklist_.push_back(user->AsPhi());
         }
       }
diff --git a/compiler/utils/x86/assembler_x86.cc b/compiler/utils/x86/assembler_x86.cc
index 2203646e7..84cdb7d4d 100644
--- a/compiler/utils/x86/assembler_x86.cc
+++ b/compiler/utils/x86/assembler_x86.cc
@@ -1030,6 +1030,14 @@ void X86Assembler::xchgl(Register reg, const Address& address) {
 }
 
 
+void X86Assembler::cmpb(const Address& address, const Immediate& imm) {
+  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  EmitUint8(0x80);
+  EmitOperand(7, address);
+  EmitUint8(imm.value() & 0xFF);
+}
+
+
 void X86Assembler::cmpw(const Address& address, const Immediate& imm) {
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
   EmitUint8(0x66);
diff --git a/compiler/utils/x86/assembler_x86.h b/compiler/utils/x86/assembler_x86.h
index 8567ad2a1..bc46e9f7c 100644
--- a/compiler/utils/x86/assembler_x86.h
+++ b/compiler/utils/x86/assembler_x86.h
@@ -479,6 +479,7 @@ class X86Assembler FINAL : public Assembler {
   void xchgl(Register dst, Register src);
   void xchgl(Register reg, const Address& address);
 
+  void cmpb(const Address& address, const Immediate& imm);
   void cmpw(const Address& address, const Immediate& imm);
 
   void cmpl(Register reg, const Immediate& imm);
diff --git a/compiler/utils/x86/assembler_x86_test.cc b/compiler/utils/x86/assembler_x86_test.cc
index 1d1df6e44..28043c938 100644
--- a/compiler/utils/x86/assembler_x86_test.cc
+++ b/compiler/utils/x86/assembler_x86_test.cc
@@ -389,4 +389,10 @@ TEST_F(AssemblerX86Test, NearLabel) {
   DriverStr(expected, "near_label");
 }
 
+TEST_F(AssemblerX86Test, Cmpb) {
+  GetAssembler()->cmpb(x86::Address(x86::EDI, 128), x86::Immediate(0));
+  const char* expected = "cmpb $0, 128(%EDI)\n";
+  DriverStr(expected, "cmpb");
+}
+
 }  // namespace art
diff --git a/compiler/utils/x86_64/assembler_x86_64.cc b/compiler/utils/x86_64/assembler_x86_64.cc
index 32eb4a37b..5e7b587e4 100644
--- a/compiler/utils/x86_64/assembler_x86_64.cc
+++ b/compiler/utils/x86_64/assembler_x86_64.cc
@@ -1224,6 +1224,16 @@ void X86_64Assembler::xchgl(CpuRegister reg, const Address& address) {
 }
 
 
+void X86_64Assembler::cmpb(const Address& address, const Immediate& imm) {
+  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  CHECK(imm.is_int32());
+  EmitOptionalRex32(address);
+  EmitUint8(0x80);
+  EmitOperand(7, address);
+  EmitUint8(imm.value() & 0xFF);
+}
+
+
 void X86_64Assembler::cmpw(const Address& address, const Immediate& imm) {
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
   CHECK(imm.is_int32());
diff --git a/compiler/utils/x86_64/assembler_x86_64.h b/compiler/utils/x86_64/assembler_x86_64.h
index 92c7d0ab9..720a402b5 100644
--- a/compiler/utils/x86_64/assembler_x86_64.h
+++ b/compiler/utils/x86_64/assembler_x86_64.h
@@ -506,6 +506,7 @@ class X86_64Assembler FINAL : public Assembler {
   void xchgq(CpuRegister dst, CpuRegister src);
   void xchgl(CpuRegister reg, const Address& address);
 
+  void cmpb(const Address& address, const Immediate& imm);
   void cmpw(const Address& address, const Immediate& imm);
 
   void cmpl(CpuRegister reg, const Immediate& imm);
diff --git a/compiler/utils/x86_64/assembler_x86_64_test.cc b/compiler/utils/x86_64/assembler_x86_64_test.cc
index afe9207eb..9dccc9f21 100644
--- a/compiler/utils/x86_64/assembler_x86_64_test.cc
+++ b/compiler/utils/x86_64/assembler_x86_64_test.cc
@@ -1637,4 +1637,11 @@ TEST_F(AssemblerX86_64Test, Repecmpsq) {
   DriverStr(expected, "Repecmpsq");
 }
 
+TEST_F(AssemblerX86_64Test, Cmpb) {
+  GetAssembler()->cmpb(x86_64::Address(x86_64::CpuRegister(x86_64::RDI), 128),
+                       x86_64::Immediate(0));
+  const char* expected = "cmpb $0, 128(%RDI)\n";
+  DriverStr(expected, "cmpb");
+}
+
 }  // namespace art
diff --git a/dex2oat/dex2oat.cc b/dex2oat/dex2oat.cc
index 370583e3b..cb274dcc0 100644
--- a/dex2oat/dex2oat.cc
+++ b/dex2oat/dex2oat.cc
@@ -1312,7 +1312,7 @@ class Dex2Oat FINAL {
     if (IsBootImage() && image_filenames_.size() > 1) {
       // If we're compiling the boot image, store the boot classpath into the Key-Value store.
       // We need this for the multi-image case.
-      key_value_store_->Put(OatHeader::kBootClassPath, GetMultiImageBootClassPath());
+      key_value_store_->Put(OatHeader::kBootClassPathKey, GetMultiImageBootClassPath());
     }
 
     if (!IsBootImage()) {
@@ -1348,12 +1348,22 @@ class Dex2Oat FINAL {
       // Open dex files for class path.
       const std::vector<std::string> class_path_locations =
           GetClassPathLocations(runtime_->GetClassPathString());
-      OpenClassPathFiles(class_path_locations, &class_path_files_);
+      OpenClassPathFiles(class_path_locations,
+                         &class_path_files_,
+                         &opened_oat_files_,
+                         runtime_->GetInstructionSet());
 
       // Store the classpath we have right now.
       std::vector<const DexFile*> class_path_files = MakeNonOwningPointerVector(class_path_files_);
-      key_value_store_->Put(OatHeader::kClassPathKey,
-                            OatFile::EncodeDexFileDependencies(class_path_files));
+      std::string encoded_class_path;
+      if (class_path_locations.size() == 1 &&
+          class_path_locations[0] == OatFile::kSpecialSharedLibrary) {
+        // When passing the special shared library as the classpath, it is the only path.
+        encoded_class_path = OatFile::kSpecialSharedLibrary;
+      } else {
+        encoded_class_path = OatFile::EncodeDexFileDependencies(class_path_files);
+      }
+      key_value_store_->Put(OatHeader::kClassPathKey, encoded_class_path);
     }
 
     // Now that we have finalized key_value_store_, start writing the oat file.
@@ -1526,6 +1536,7 @@ class Dex2Oat FINAL {
                                      instruction_set_,
                                      instruction_set_features_.get(),
                                      IsBootImage(),
+                                     IsAppImage(),
                                      image_classes_.release(),
                                      compiled_classes_.release(),
                                      /* compiled_methods */ nullptr,
@@ -1963,14 +1974,37 @@ class Dex2Oat FINAL {
     return parsed;
   }
 
-  // Opens requested class path files and appends them to opened_dex_files.
+  // Opens requested class path files and appends them to opened_dex_files. If the dex files have
+  // been stripped, this opens them from their oat files and appends them to opened_oat_files.
   static void OpenClassPathFiles(const std::vector<std::string>& class_path_locations,
-                                 std::vector<std::unique_ptr<const DexFile>>* opened_dex_files) {
-    DCHECK(opened_dex_files != nullptr) << "OpenClassPathFiles out-param is nullptr";
+                                 std::vector<std::unique_ptr<const DexFile>>* opened_dex_files,
+                                 std::vector<std::unique_ptr<OatFile>>* opened_oat_files,
+                                 InstructionSet isa) {
+    DCHECK(opened_dex_files != nullptr) << "OpenClassPathFiles dex out-param is nullptr";
+    DCHECK(opened_oat_files != nullptr) << "OpenClassPathFiles oat out-param is nullptr";
     for (const std::string& location : class_path_locations) {
+      // Stop early if we detect the special shared library, which may be passed as the classpath
+      // for dex2oat when we want to skip the shared libraries check.
+      if (location == OatFile::kSpecialSharedLibrary) {
+        break;
+      }
       std::string error_msg;
       if (!DexFile::Open(location.c_str(), location.c_str(), &error_msg, opened_dex_files)) {
-        LOG(WARNING) << "Failed to open dex file '" << location << "': " << error_msg;
+        // If we fail to open the dex file because it's been stripped, try to open the dex file
+        // from its corresponding oat file.
+        OatFileAssistant oat_file_assistant(location.c_str(), isa, false, false);
+        std::unique_ptr<OatFile> oat_file(oat_file_assistant.GetBestOatFile());
+        if (oat_file == nullptr) {
+          LOG(WARNING) << "Failed to open dex file and associated oat file for '" << location
+                       << "': " << error_msg;
+        } else {
+          std::vector<std::unique_ptr<const DexFile>> oat_dex_files =
+              oat_file_assistant.LoadDexFiles(*oat_file, location.c_str());
+          opened_oat_files->push_back(std::move(oat_file));
+          opened_dex_files->insert(opened_dex_files->end(),
+                                   std::make_move_iterator(oat_dex_files.begin()),
+                                   std::make_move_iterator(oat_dex_files.end()));
+        }
       }
     }
   }
@@ -2440,6 +2474,7 @@ class Dex2Oat FINAL {
   std::unique_ptr<CompilerDriver> driver_;
 
   std::vector<std::unique_ptr<MemMap>> opened_dex_files_maps_;
+  std::vector<std::unique_ptr<OatFile>> opened_oat_files_;
   std::vector<std::unique_ptr<const DexFile>> opened_dex_files_;
 
   std::vector<const DexFile*> no_inline_from_dex_files_;
diff --git a/disassembler/disassembler.h b/disassembler/disassembler.h
index b99e5c2df..b08031587 100644
--- a/disassembler/disassembler.h
+++ b/disassembler/disassembler.h
@@ -31,16 +31,23 @@ class DisassemblerOptions {
   // Should the disassembler print absolute or relative addresses.
   const bool absolute_addresses_;
 
-  // Base addess for calculating relative code offsets when absolute_addresses_ is false.
+  // Base address for calculating relative code offsets when absolute_addresses_ is false.
   const uint8_t* const base_address_;
 
+  // End address (exclusive);
+  const uint8_t* const end_address_;
+
   // If set, the disassembler is allowed to look at load targets in literal
   // pools.
   const bool can_read_literals_;
 
-  DisassemblerOptions(bool absolute_addresses, const uint8_t* base_address,
+  DisassemblerOptions(bool absolute_addresses,
+                      const uint8_t* base_address,
+                      const uint8_t* end_address,
                       bool can_read_literals)
-      : absolute_addresses_(absolute_addresses), base_address_(base_address),
+      : absolute_addresses_(absolute_addresses),
+        base_address_(base_address),
+        end_address_(end_address),
         can_read_literals_(can_read_literals) {}
 
  private:
diff --git a/disassembler/disassembler_arm.cc b/disassembler/disassembler_arm.cc
index bcb043883..286faf215 100644
--- a/disassembler/disassembler_arm.cc
+++ b/disassembler/disassembler_arm.cc
@@ -418,7 +418,12 @@ std::ostream& operator<<(std::ostream& os, T2LitType type) {
   return os << static_cast<int>(type);
 }
 
-void DumpThumb2Literal(std::ostream& args, const uint8_t* instr_ptr, uint32_t U, uint32_t imm32,
+void DumpThumb2Literal(std::ostream& args,
+                       const uint8_t* instr_ptr,
+                       const uintptr_t lo_adr,
+                       const uintptr_t hi_adr,
+                       uint32_t U,
+                       uint32_t imm32,
                        T2LitType type) {
   // Literal offsets (imm32) are not required to be aligned so we may need unaligned access.
   typedef const int16_t unaligned_int16_t __attribute__ ((aligned (1)));
@@ -428,8 +433,16 @@ void DumpThumb2Literal(std::ostream& args, const uint8_t* instr_ptr, uint32_t U,
   typedef const int64_t unaligned_int64_t __attribute__ ((aligned (1)));
   typedef const uint64_t unaligned_uint64_t __attribute__ ((aligned (1)));
 
+  // Get address of literal. Bail if not within expected buffer range to
+  // avoid trying to fetch invalid literals (we can encounter this when
+  // interpreting raw data as instructions).
   uintptr_t pc = RoundDown(reinterpret_cast<intptr_t>(instr_ptr) + 4, 4);
   uintptr_t lit_adr = U ? pc + imm32 : pc - imm32;
+  if (lit_adr < lo_adr || lit_adr >= hi_adr) {
+    args << "  ; (?)";
+    return;
+  }
+
   args << "  ; ";
   switch (type) {
     case kT2LitUByte:
@@ -482,6 +495,10 @@ size_t DisassemblerArm::DumpThumb32(std::ostream& os, const uint8_t* instr_ptr)
     return DumpThumb16(os, instr_ptr);
   }
 
+  // Set valid address range of backing buffer.
+  const uintptr_t lo_adr = reinterpret_cast<intptr_t>(GetDisassemblerOptions()->base_address_);
+  const uintptr_t hi_adr = reinterpret_cast<intptr_t>(GetDisassemblerOptions()->end_address_);
+
   uint32_t op2 = (instr >> 20) & 0x7F;
   std::ostringstream opcode;
   std::ostringstream args;
@@ -824,7 +841,7 @@ size_t DisassemblerArm::DumpThumb32(std::ostream& os, const uint8_t* instr_ptr)
                 args << d << ", [" << Rn << ", #" << ((U == 1) ? "" : "-")
                      << (imm8 << 2) << "]";
                 if (Rn.r == 15 && U == 1) {
-                  DumpThumb2Literal(args, instr_ptr, U, imm8 << 2, kT2LitHexLong);
+                  DumpThumb2Literal(args, instr_ptr, lo_adr, hi_adr, U, imm8 << 2, kT2LitHexLong);
                 }
               } else if (Rn.r == 13 && W == 1 && U == L) {  // VPUSH/VPOP
                 opcode << (L == 1 ? "vpop" : "vpush");
@@ -1410,7 +1427,7 @@ size_t DisassemblerArm::DumpThumb32(std::ostream& os, const uint8_t* instr_ptr)
               };
               DCHECK_LT(op2 >> 1, arraysize(lit_type));
               DCHECK_NE(lit_type[op2 >> 1], kT2LitInvalid);
-              DumpThumb2Literal(args, instr_ptr, U, imm12, lit_type[op2 >> 1]);
+              DumpThumb2Literal(args, instr_ptr, lo_adr, hi_adr, U, imm12, lit_type[op2 >> 1]);
             }
           } else if ((instr & 0xFC0) == 0) {
             opcode << ldr_str << sign << type << ".w";
@@ -1711,10 +1728,13 @@ size_t DisassemblerArm::DumpThumb16(std::ostream& os, const uint8_t* instr_ptr)
           break;
       }
     } else if (opcode1 == 0x12 || opcode1 == 0x13) {  // 01001x
+      const uintptr_t lo_adr = reinterpret_cast<intptr_t>(GetDisassemblerOptions()->base_address_);
+      const uintptr_t hi_adr = reinterpret_cast<intptr_t>(GetDisassemblerOptions()->end_address_);
       ThumbRegister Rt(instr, 8);
       uint16_t imm8 = instr & 0xFF;
       opcode << "ldr";
       args << Rt << ", [pc, #" << (imm8 << 2) << "]";
+      DumpThumb2Literal(args, instr_ptr, lo_adr, hi_adr, /*U*/ 1u, imm8 << 2, kT2LitHexWord);
     } else if ((opcode1 >= 0x14 && opcode1 <= 0x17) ||  // 0101xx
                (opcode1 >= 0x18 && opcode1 <= 0x1f) ||  // 011xxx
                (opcode1 >= 0x20 && opcode1 <= 0x27)) {  // 100xxx
diff --git a/disassembler/disassembler_arm64.cc b/disassembler/disassembler_arm64.cc
index 5f8871470..6a9afe574 100644
--- a/disassembler/disassembler_arm64.cc
+++ b/disassembler/disassembler_arm64.cc
@@ -63,9 +63,17 @@ void CustomDisassembler::VisitLoadLiteral(const vixl::Instruction* instr) {
     return;
   }
 
+  // Get address of literal. Bail if not within expected buffer range to
+  // avoid trying to fetch invalid literals (we can encounter this when
+  // interpreting raw data as instructions).
   void* data_address = instr->LiteralAddress<void*>();
-  vixl::Instr op = instr->Mask(vixl::LoadLiteralMask);
+  if (data_address < base_address_ || data_address >= end_address_) {
+    AppendToOutput(" (?)");
+    return;
+  }
 
+  // Output information on literal.
+  vixl::Instr op = instr->Mask(vixl::LoadLiteralMask);
   switch (op) {
     case vixl::LDR_w_lit:
     case vixl::LDR_x_lit:
diff --git a/disassembler/disassembler_arm64.h b/disassembler/disassembler_arm64.h
index 44fa53f9f..a4e5ee8a4 100644
--- a/disassembler/disassembler_arm64.h
+++ b/disassembler/disassembler_arm64.h
@@ -30,8 +30,11 @@ namespace arm64 {
 
 class CustomDisassembler FINAL : public vixl::Disassembler {
  public:
-  explicit CustomDisassembler(DisassemblerOptions* options) :
-      vixl::Disassembler(), read_literals_(options->can_read_literals_) {
+  explicit CustomDisassembler(DisassemblerOptions* options)
+      : vixl::Disassembler(),
+        read_literals_(options->can_read_literals_),
+        base_address_(options->base_address_),
+        end_address_(options->end_address_) {
     if (!options->absolute_addresses_) {
       MapCodeAddress(0, reinterpret_cast<const vixl::Instruction*>(options->base_address_));
     }
@@ -55,6 +58,10 @@ class CustomDisassembler FINAL : public vixl::Disassembler {
   //           true | 0x72681558: 1c000acb  ldr s11, pc+344 (addr 0x726816b0)
   //          false | 0x72681558: 1c000acb  ldr s11, pc+344 (addr 0x726816b0) (3.40282e+38)
   const bool read_literals_;
+
+  // Valid address range: [base_address_, end_address_)
+  const void* const base_address_;
+  const void* const end_address_;
 };
 
 class DisassemblerArm64 FINAL : public Disassembler {
diff --git a/libart_fake/Android.mk b/libart_fake/Android.mk
new file mode 100644
index 000000000..ed868a5bd
--- /dev/null
+++ b/libart_fake/Android.mk
@@ -0,0 +1,34 @@
+# 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.
+#
+
+LOCAL_PATH := $(my-dir)
+
+include $(CLEAR_VARS)
+
+LOCAL_MODULE := libart_fake
+LOCAL_INSTALLED_MODULE_STEM := libart.so
+LOCAL_SDK_VERSION := 9
+LOCAL_CPP_EXTENSION := .cc
+LOCAL_SRC_FILES := fake.cc
+LOCAL_SHARED_LIBRARIES := liblog
+
+ifdef TARGET_2ND_ARCH
+    LOCAL_MODULE_PATH_32 := $(TARGET_OUT)/fake-libs
+    LOCAL_MODULE_PATH_64 := $(TARGET_OUT)/fake-libs64
+else
+    LOCAL_MODULE_PATH := $(TARGET_OUT)/fake-libs
+endif
+
+include $(BUILD_SHARED_LIBRARY)
diff --git a/libart_fake/README.md b/libart_fake/README.md
new file mode 100644
index 000000000..6e3621e55
--- /dev/null
+++ b/libart_fake/README.md
@@ -0,0 +1,5 @@
+libart_fake
+====
+
+A fake libart made to satisfy some misbehaving apps that will attempt to link
+against libart.so.
diff --git a/libart_fake/fake.cc b/libart_fake/fake.cc
new file mode 100644
index 000000000..884242101
--- /dev/null
+++ b/libart_fake/fake.cc
@@ -0,0 +1,46 @@
+/*
+ * 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.
+ */
+
+#define LOG_TAG "libart_fake"
+
+#include <android/log.h>
+
+#define LOGIT(...) __android_log_print(ANDROID_LOG_ERROR, LOG_TAG, __VA_ARGS__)
+namespace art {
+class Dbg {
+ public:
+  void SuspendVM();
+  void ResumeVM();
+};
+
+class FaultManager {
+ public:
+  void EnsureArtActionInFrontOfSignalChain();
+};
+
+void Dbg::SuspendVM() {
+  LOGIT("Linking to and calling into libart.so internal functions is not supported. "
+        "This call to '%s' is being ignored.", __func__);
+}
+void Dbg::ResumeVM() {
+  LOGIT("Linking to and calling into libart.so internal functions is not supported. "
+        "This call to '%s' is being ignored.", __func__);
+}
+void FaultManager::EnsureArtActionInFrontOfSignalChain() {
+  LOGIT("Linking to and calling into libart.so internal functions is not supported. "
+        "This call to '%s' is being ignored.", __func__);
+}
+};  // namespace art
diff --git a/oatdump/oatdump.cc b/oatdump/oatdump.cc
index 3c6a05d97..f5458c067 100644
--- a/oatdump/oatdump.cc
+++ b/oatdump/oatdump.cc
@@ -59,6 +59,7 @@
 #include "stack_map.h"
 #include "ScopedLocalRef.h"
 #include "thread_list.h"
+#include "type_lookup_table.h"
 #include "verifier/method_verifier.h"
 #include "well_known_classes.h"
 
@@ -334,6 +335,7 @@ class OatDumper {
       disassembler_(Disassembler::Create(instruction_set_,
                                          new DisassemblerOptions(options_.absolute_addresses_,
                                                                  oat_file.Begin(),
+                                                                 oat_file.End(),
                                                                  true /* can_read_literals_ */))) {
     CHECK(options_.class_loader_ != nullptr);
     CHECK(options_.class_filter_ != nullptr);
@@ -573,8 +575,15 @@ class OatDumper {
     os << StringPrintf("location: %s\n", oat_dex_file.GetDexFileLocation().c_str());
     os << StringPrintf("checksum: 0x%08x\n", oat_dex_file.GetDexFileLocationChecksum());
 
-    // Create the verifier early.
+    // Print embedded dex file data range.
+    const uint8_t* const oat_file_begin = oat_dex_file.GetOatFile()->Begin();
+    const uint8_t* const dex_file_pointer = oat_dex_file.GetDexFilePointer();
+    uint32_t dex_offset = dchecked_integral_cast<uint32_t>(dex_file_pointer - oat_file_begin);
+    os << StringPrintf("dex-file: 0x%08x..0x%08x\n",
+                       dex_offset,
+                       dchecked_integral_cast<uint32_t>(dex_offset + oat_dex_file.FileSize() - 1));
 
+    // Create the dex file early. A lot of print-out things depend on it.
     std::string error_msg;
     const DexFile* const dex_file = OpenDexFile(&oat_dex_file, &error_msg);
     if (dex_file == nullptr) {
@@ -583,6 +592,16 @@ class OatDumper {
       return false;
     }
 
+    // Print lookup table, if it exists.
+    if (oat_dex_file.GetLookupTableData() != nullptr) {
+      uint32_t table_offset = dchecked_integral_cast<uint32_t>(
+          oat_dex_file.GetLookupTableData() - oat_file_begin);
+      uint32_t table_size = TypeLookupTable::RawDataLength(*dex_file);
+      os << StringPrintf("type-table: 0x%08x..0x%08x\n",
+                         table_offset,
+                         table_offset + table_size - 1);
+    }
+
     VariableIndentationOutputStream vios(&os);
     ScopedIndentation indent1(&vios);
     for (size_t class_def_index = 0;
@@ -1416,11 +1435,10 @@ class ImageDumper {
       indent_os << "\n";
       // TODO: Dump fields.
       // Dump methods after.
-      const auto& methods_section = image_header_.GetMethodsSection();
       DumpArtMethodVisitor visitor(this);
-      methods_section.VisitPackedArtMethods(&visitor,
-                                            image_space_.Begin(),
-                                            image_header_.GetPointerSize());
+      image_header_.VisitPackedArtMethods(&visitor,
+                                          image_space_.Begin(),
+                                          image_header_.GetPointerSize());
       // Dump the large objects separately.
       heap->GetLargeObjectsSpace()->GetLiveBitmap()->Walk(ImageDumper::Callback, this);
       indent_os << "\n";
@@ -1779,6 +1797,7 @@ class ImageDumper {
     DCHECK(method != nullptr);
     const void* quick_oat_code_begin = GetQuickOatCodeBegin(method);
     const void* quick_oat_code_end = GetQuickOatCodeEnd(method);
+    const size_t pointer_size = image_header_.GetPointerSize();
     OatQuickMethodHeader* method_header = reinterpret_cast<OatQuickMethodHeader*>(
         reinterpret_cast<uintptr_t>(quick_oat_code_begin) - sizeof(OatQuickMethodHeader));
     if (method->IsNative()) {
@@ -1792,13 +1811,16 @@ class ImageDumper {
           image_header_.GetPointerSize())) {
         indent_os << StringPrintf("OAT CODE: %p\n", quick_oat_code_begin);
       }
-    } else if (method->IsAbstract() ||
-               method->IsCalleeSaveMethod() ||
-               method->IsResolutionMethod() ||
-               (method == Runtime::Current()->GetImtConflictMethod()) ||
-               method->IsImtUnimplementedMethod() ||
-               method->IsClassInitializer()) {
+    } else if (method->IsAbstract() || method->IsClassInitializer()) {
       // Don't print information for these.
+    } else if (method->IsRuntimeMethod()) {
+      ImtConflictTable* table = method->GetImtConflictTable(image_header_.GetPointerSize());
+      if (table != nullptr) {
+        indent_os << "IMT conflict table " << table << " method: ";
+        for (size_t i = 0, count = table->NumEntries(pointer_size); i < count; ++i) {
+          indent_os << PrettyMethod(table->GetImplementationMethod(i, pointer_size)) << " ";
+        }
+      }
     } else {
       const DexFile::CodeItem* code_item = method->GetCodeItem();
       size_t dex_instruction_bytes = code_item->insns_size_in_code_units_ * 2;
diff --git a/patchoat/patchoat.cc b/patchoat/patchoat.cc
index 93e40afea..0a7ffda3b 100644
--- a/patchoat/patchoat.cc
+++ b/patchoat/patchoat.cc
@@ -472,8 +472,7 @@ class PatchOatArtFieldVisitor : public ArtFieldVisitor {
 
 void PatchOat::PatchArtFields(const ImageHeader* image_header) {
   PatchOatArtFieldVisitor visitor(this);
-  const auto& section = image_header->GetImageSection(ImageHeader::kSectionArtFields);
-  section.VisitPackedArtFields(&visitor, heap_->Begin());
+  image_header->VisitPackedArtFields(&visitor, heap_->Begin());
 }
 
 class PatchOatArtMethodVisitor : public ArtMethodVisitor {
@@ -490,10 +489,20 @@ class PatchOatArtMethodVisitor : public ArtMethodVisitor {
 };
 
 void PatchOat::PatchArtMethods(const ImageHeader* image_header) {
-  const auto& section = image_header->GetMethodsSection();
   const size_t pointer_size = InstructionSetPointerSize(isa_);
   PatchOatArtMethodVisitor visitor(this);
-  section.VisitPackedArtMethods(&visitor, heap_->Begin(), pointer_size);
+  image_header->VisitPackedArtMethods(&visitor, heap_->Begin(), pointer_size);
+}
+
+void PatchOat::PatchImtConflictTables(const ImageHeader* image_header) {
+  const size_t pointer_size = InstructionSetPointerSize(isa_);
+  // We can safely walk target image since the conflict tables are independent.
+  image_header->VisitPackedImtConflictTables(
+      [this](ArtMethod* method) {
+        return RelocatedAddressOfPointer(method);
+      },
+      image_->Begin(),
+      pointer_size);
 }
 
 class FixupRootVisitor : public RootVisitor {
@@ -627,6 +636,7 @@ bool PatchOat::PatchImage(bool primary_image) {
 
   PatchArtFields(image_header);
   PatchArtMethods(image_header);
+  PatchImtConflictTables(image_header);
   PatchInternedStrings(image_header);
   PatchClassTable(image_header);
   // Patch dex file int/long arrays which point to ArtFields.
@@ -725,6 +735,7 @@ void PatchOat::FixupMethod(ArtMethod* object, ArtMethod* copy) {
       RelocatedAddressOfPointer(object->GetDexCacheResolvedTypes(pointer_size)), pointer_size);
   copy->SetEntryPointFromQuickCompiledCodePtrSize(RelocatedAddressOfPointer(
       object->GetEntryPointFromQuickCompiledCodePtrSize(pointer_size)), pointer_size);
+  // No special handling for IMT conflict table since all pointers are moved by the same offset.
   copy->SetEntryPointFromJniPtrSize(RelocatedAddressOfPointer(
       object->GetEntryPointFromJniPtrSize(pointer_size)), pointer_size);
 }
diff --git a/patchoat/patchoat.h b/patchoat/patchoat.h
index 510ff1e5b..3ef837fde 100644
--- a/patchoat/patchoat.h
+++ b/patchoat/patchoat.h
@@ -117,6 +117,8 @@ class PatchOat {
   bool PatchImage(bool primary_image) SHARED_REQUIRES(Locks::mutator_lock_);
   void PatchArtFields(const ImageHeader* image_header) SHARED_REQUIRES(Locks::mutator_lock_);
   void PatchArtMethods(const ImageHeader* image_header) SHARED_REQUIRES(Locks::mutator_lock_);
+  void PatchImtConflictTables(const ImageHeader* image_header)
+      SHARED_REQUIRES(Locks::mutator_lock_);
   void PatchInternedStrings(const ImageHeader* image_header)
       SHARED_REQUIRES(Locks::mutator_lock_);
   void PatchClassTable(const ImageHeader* image_header)
diff --git a/profman/profman.cc b/profman/profman.cc
index 3e632bce8..4d9276f15 100644
--- a/profman/profman.cc
+++ b/profman/profman.cc
@@ -189,7 +189,6 @@ class ProfMan FINAL {
       return -1;
     }
     std::string dump = info.DumpInfo(/*dex_files*/ nullptr);
-    info.Save(fd);
     std::cout << dump << "\n";
     return 0;
   }
@@ -216,7 +215,11 @@ class ProfMan FINAL {
   }
 
   void LogCompletionTime() {
-    LOG(INFO) << "profman took " << PrettyDuration(NanoTime() - start_ns_);
+    static constexpr uint64_t kLogThresholdTime = MsToNs(100);  // 100ms
+    uint64_t time_taken = NanoTime() - start_ns_;
+    if (time_taken > kLogThresholdTime) {
+      LOG(WARNING) << "profman took " << PrettyDuration(time_taken);
+    }
   }
 
   std::vector<std::string> profile_files_;
diff --git a/runtime/arch/arm/entrypoints_init_arm.cc b/runtime/arch/arm/entrypoints_init_arm.cc
index e358ff879..f0e9ac517 100644
--- a/runtime/arch/arm/entrypoints_init_arm.cc
+++ b/runtime/arch/arm/entrypoints_init_arm.cc
@@ -17,6 +17,7 @@
 #include "entrypoints/jni/jni_entrypoints.h"
 #include "entrypoints/quick/quick_alloc_entrypoints.h"
 #include "entrypoints/quick/quick_default_externs.h"
+#include "entrypoints/quick/quick_default_init_entrypoints.h"
 #include "entrypoints/quick/quick_entrypoints.h"
 #include "entrypoints/entrypoint_utils.h"
 #include "entrypoints/math_entrypoints.h"
@@ -47,67 +48,12 @@ extern "C" int __aeabi_idivmod(int32_t, int32_t);  // [DIV|REM]_INT[_2ADDR|_LIT8
 extern "C" int64_t __aeabi_ldivmod(int64_t, int64_t);
 
 void InitEntryPoints(JniEntryPoints* jpoints, QuickEntryPoints* qpoints) {
-  // JNI
-  jpoints->pDlsymLookup = art_jni_dlsym_lookup_stub;
-
-  // Alloc
-  ResetQuickAllocEntryPoints(qpoints);
+  DefaultInitEntryPoints(jpoints, qpoints);
 
   // Cast
   qpoints->pInstanceofNonTrivial = artIsAssignableFromCode;
   qpoints->pCheckCast = art_quick_check_cast;
 
-  // DexCache
-  qpoints->pInitializeStaticStorage = art_quick_initialize_static_storage;
-  qpoints->pInitializeTypeAndVerifyAccess = art_quick_initialize_type_and_verify_access;
-  qpoints->pInitializeType = art_quick_initialize_type;
-  qpoints->pResolveString = art_quick_resolve_string;
-
-  // Field
-  qpoints->pSet8Instance = art_quick_set8_instance;
-  qpoints->pSet8Static = art_quick_set8_static;
-  qpoints->pSet16Instance = art_quick_set16_instance;
-  qpoints->pSet16Static = art_quick_set16_static;
-  qpoints->pSet32Instance = art_quick_set32_instance;
-  qpoints->pSet32Static = art_quick_set32_static;
-  qpoints->pSet64Instance = art_quick_set64_instance;
-  qpoints->pSet64Static = art_quick_set64_static;
-  qpoints->pSetObjInstance = art_quick_set_obj_instance;
-  qpoints->pSetObjStatic = art_quick_set_obj_static;
-  qpoints->pGetByteInstance = art_quick_get_byte_instance;
-  qpoints->pGetBooleanInstance = art_quick_get_boolean_instance;
-  qpoints->pGetShortInstance = art_quick_get_short_instance;
-  qpoints->pGetCharInstance = art_quick_get_char_instance;
-  qpoints->pGet32Instance = art_quick_get32_instance;
-  qpoints->pGet64Instance = art_quick_get64_instance;
-  qpoints->pGetObjInstance = art_quick_get_obj_instance;
-  qpoints->pGetByteStatic = art_quick_get_byte_static;
-  qpoints->pGetBooleanStatic = art_quick_get_boolean_static;
-  qpoints->pGetShortStatic = art_quick_get_short_static;
-  qpoints->pGetCharStatic = art_quick_get_char_static;
-  qpoints->pGet32Static = art_quick_get32_static;
-  qpoints->pGet64Static = art_quick_get64_static;
-  qpoints->pGetObjStatic = art_quick_get_obj_static;
-
-  // Array
-  qpoints->pAputObjectWithNullAndBoundCheck = art_quick_aput_obj_with_null_and_bound_check;
-  qpoints->pAputObjectWithBoundCheck = art_quick_aput_obj_with_bound_check;
-  qpoints->pAputObject = art_quick_aput_obj;
-  qpoints->pHandleFillArrayData = art_quick_handle_fill_data;
-
-  // JNI
-  qpoints->pJniMethodStart = JniMethodStart;
-  qpoints->pJniMethodStartSynchronized = JniMethodStartSynchronized;
-  qpoints->pJniMethodEnd = JniMethodEnd;
-  qpoints->pJniMethodEndSynchronized = JniMethodEndSynchronized;
-  qpoints->pJniMethodEndWithReference = JniMethodEndWithReference;
-  qpoints->pJniMethodEndWithReferenceSynchronized = JniMethodEndWithReferenceSynchronized;
-  qpoints->pQuickGenericJniTrampoline = art_quick_generic_jni_trampoline;
-
-  // Locks
-  qpoints->pLockObject = art_quick_lock_object;
-  qpoints->pUnlockObject = art_quick_unlock_object;
-
   // Math
   qpoints->pIdivmod = __aeabi_idivmod;
   qpoints->pLdiv = __aeabi_ldivmod;
@@ -154,35 +100,6 @@ void InitEntryPoints(JniEntryPoints* jpoints, QuickEntryPoints* qpoints) {
   qpoints->pStringCompareTo = art_quick_string_compareto;
   qpoints->pMemcpy = memcpy;
 
-  // Invocation
-  qpoints->pQuickImtConflictTrampoline = art_quick_imt_conflict_trampoline;
-  qpoints->pQuickResolutionTrampoline = art_quick_resolution_trampoline;
-  qpoints->pQuickToInterpreterBridge = art_quick_to_interpreter_bridge;
-  qpoints->pInvokeDirectTrampolineWithAccessCheck =
-      art_quick_invoke_direct_trampoline_with_access_check;
-  qpoints->pInvokeInterfaceTrampolineWithAccessCheck =
-      art_quick_invoke_interface_trampoline_with_access_check;
-  qpoints->pInvokeStaticTrampolineWithAccessCheck =
-      art_quick_invoke_static_trampoline_with_access_check;
-  qpoints->pInvokeSuperTrampolineWithAccessCheck =
-      art_quick_invoke_super_trampoline_with_access_check;
-  qpoints->pInvokeVirtualTrampolineWithAccessCheck =
-      art_quick_invoke_virtual_trampoline_with_access_check;
-
-  // Thread
-  qpoints->pTestSuspend = art_quick_test_suspend;
-
-  // Throws
-  qpoints->pDeliverException = art_quick_deliver_exception;
-  qpoints->pThrowArrayBounds = art_quick_throw_array_bounds;
-  qpoints->pThrowDivZero = art_quick_throw_div_zero;
-  qpoints->pThrowNoSuchMethod = art_quick_throw_no_such_method;
-  qpoints->pThrowNullPointer = art_quick_throw_null_pointer_exception;
-  qpoints->pThrowStackOverflow = art_quick_throw_stack_overflow;
-
-  // Deoptimization from compiled code.
-  qpoints->pDeoptimize = art_quick_deoptimize_from_compiled_code;
-
   // Read barrier.
   qpoints->pReadBarrierJni = ReadBarrierJni;
   qpoints->pReadBarrierMark = artReadBarrierMark;
diff --git a/runtime/arch/arm/quick_entrypoints_arm.S b/runtime/arch/arm/quick_entrypoints_arm.S
index e6ff0aa13..321b9d217 100644
--- a/runtime/arch/arm/quick_entrypoints_arm.S
+++ b/runtime/arch/arm/quick_entrypoints_arm.S
@@ -544,6 +544,15 @@ ENTRY art_quick_lock_object
     DELIVER_PENDING_EXCEPTION
 END art_quick_lock_object
 
+ENTRY art_quick_lock_object_no_inline
+    SETUP_REFS_ONLY_CALLEE_SAVE_FRAME r1, r2  @ save callee saves in case we block
+    mov    r1, r9                     @ pass Thread::Current
+    bl     artLockObjectFromCode      @ (Object* obj, Thread*)
+    RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME
+    RETURN_IF_RESULT_IS_ZERO
+    DELIVER_PENDING_EXCEPTION
+END art_quick_lock_object_no_inline
+
     /*
      * Entry from managed code that calls artUnlockObjectFromCode and delivers exception on failure.
      * r0 holds the possibly null object to lock.
@@ -601,6 +610,16 @@ ENTRY art_quick_unlock_object
     DELIVER_PENDING_EXCEPTION
 END art_quick_unlock_object
 
+ENTRY art_quick_unlock_object_no_inline
+    @ save callee saves in case exception allocation triggers GC
+    SETUP_REFS_ONLY_CALLEE_SAVE_FRAME r1, r2
+    mov    r1, r9                     @ pass Thread::Current
+    bl     artUnlockObjectFromCode    @ (Object* obj, Thread*)
+    RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME
+    RETURN_IF_RESULT_IS_ZERO
+    DELIVER_PENDING_EXCEPTION
+END art_quick_unlock_object_no_inline
+
     /*
      * Entry from managed code that calls artIsAssignableFromCode and on failure calls
      * artThrowClassCastException.
diff --git a/runtime/arch/arm64/entrypoints_init_arm64.cc b/runtime/arch/arm64/entrypoints_init_arm64.cc
index 4db941174..bf0f6470d 100644
--- a/runtime/arch/arm64/entrypoints_init_arm64.cc
+++ b/runtime/arch/arm64/entrypoints_init_arm64.cc
@@ -17,6 +17,7 @@
 #include "entrypoints/jni/jni_entrypoints.h"
 #include "entrypoints/quick/quick_alloc_entrypoints.h"
 #include "entrypoints/quick/quick_default_externs.h"
+#include "entrypoints/quick/quick_default_init_entrypoints.h"
 #include "entrypoints/quick/quick_entrypoints.h"
 #include "entrypoints/entrypoint_utils.h"
 #include "entrypoints/math_entrypoints.h"
@@ -30,67 +31,12 @@ extern "C" uint32_t artIsAssignableFromCode(const mirror::Class* klass,
                                             const mirror::Class* ref_class);
 
 void InitEntryPoints(JniEntryPoints* jpoints, QuickEntryPoints* qpoints) {
-  // JNI
-  jpoints->pDlsymLookup = art_jni_dlsym_lookup_stub;
-
-  // Alloc
-  ResetQuickAllocEntryPoints(qpoints);
+  DefaultInitEntryPoints(jpoints, qpoints);
 
   // Cast
   qpoints->pInstanceofNonTrivial = artIsAssignableFromCode;
   qpoints->pCheckCast = art_quick_check_cast;
 
-  // DexCache
-  qpoints->pInitializeStaticStorage = art_quick_initialize_static_storage;
-  qpoints->pInitializeTypeAndVerifyAccess = art_quick_initialize_type_and_verify_access;
-  qpoints->pInitializeType = art_quick_initialize_type;
-  qpoints->pResolveString = art_quick_resolve_string;
-
-  // Field
-  qpoints->pSet8Instance = art_quick_set8_instance;
-  qpoints->pSet8Static = art_quick_set8_static;
-  qpoints->pSet16Instance = art_quick_set16_instance;
-  qpoints->pSet16Static = art_quick_set16_static;
-  qpoints->pSet32Instance = art_quick_set32_instance;
-  qpoints->pSet32Static = art_quick_set32_static;
-  qpoints->pSet64Instance = art_quick_set64_instance;
-  qpoints->pSet64Static = art_quick_set64_static;
-  qpoints->pSetObjInstance = art_quick_set_obj_instance;
-  qpoints->pSetObjStatic = art_quick_set_obj_static;
-  qpoints->pGetBooleanInstance = art_quick_get_boolean_instance;
-  qpoints->pGetByteInstance = art_quick_get_byte_instance;
-  qpoints->pGetCharInstance = art_quick_get_char_instance;
-  qpoints->pGetShortInstance = art_quick_get_short_instance;
-  qpoints->pGet32Instance = art_quick_get32_instance;
-  qpoints->pGet64Instance = art_quick_get64_instance;
-  qpoints->pGetObjInstance = art_quick_get_obj_instance;
-  qpoints->pGetBooleanStatic = art_quick_get_boolean_static;
-  qpoints->pGetByteStatic = art_quick_get_byte_static;
-  qpoints->pGetCharStatic = art_quick_get_char_static;
-  qpoints->pGetShortStatic = art_quick_get_short_static;
-  qpoints->pGet32Static = art_quick_get32_static;
-  qpoints->pGet64Static = art_quick_get64_static;
-  qpoints->pGetObjStatic = art_quick_get_obj_static;
-
-  // Array
-  qpoints->pAputObjectWithNullAndBoundCheck = art_quick_aput_obj_with_null_and_bound_check;
-  qpoints->pAputObjectWithBoundCheck = art_quick_aput_obj_with_bound_check;
-  qpoints->pAputObject = art_quick_aput_obj;
-  qpoints->pHandleFillArrayData = art_quick_handle_fill_data;
-
-  // JNI
-  qpoints->pJniMethodStart = JniMethodStart;
-  qpoints->pJniMethodStartSynchronized = JniMethodStartSynchronized;
-  qpoints->pJniMethodEnd = JniMethodEnd;
-  qpoints->pJniMethodEndSynchronized = JniMethodEndSynchronized;
-  qpoints->pJniMethodEndWithReference = JniMethodEndWithReference;
-  qpoints->pJniMethodEndWithReferenceSynchronized = JniMethodEndWithReferenceSynchronized;
-  qpoints->pQuickGenericJniTrampoline = art_quick_generic_jni_trampoline;
-
-  // Locks
-  qpoints->pLockObject = art_quick_lock_object;
-  qpoints->pUnlockObject = art_quick_unlock_object;
-
   // Math
   // TODO null entrypoints not needed for ARM64 - generate inline.
   qpoints->pCmpgDouble = nullptr;
@@ -134,38 +80,10 @@ void InitEntryPoints(JniEntryPoints* jpoints, QuickEntryPoints* qpoints) {
 
   // Intrinsics
   qpoints->pIndexOf = art_quick_indexof;
-  qpoints->pStringCompareTo = art_quick_string_compareto;
+  // The ARM64 StringCompareTo intrinsic does not call the runtime.
+  qpoints->pStringCompareTo = nullptr;
   qpoints->pMemcpy = memcpy;
 
-  // Invocation
-  qpoints->pQuickImtConflictTrampoline = art_quick_imt_conflict_trampoline;
-  qpoints->pQuickResolutionTrampoline = art_quick_resolution_trampoline;
-  qpoints->pQuickToInterpreterBridge = art_quick_to_interpreter_bridge;
-  qpoints->pInvokeDirectTrampolineWithAccessCheck =
-      art_quick_invoke_direct_trampoline_with_access_check;
-  qpoints->pInvokeInterfaceTrampolineWithAccessCheck =
-      art_quick_invoke_interface_trampoline_with_access_check;
-  qpoints->pInvokeStaticTrampolineWithAccessCheck =
-      art_quick_invoke_static_trampoline_with_access_check;
-  qpoints->pInvokeSuperTrampolineWithAccessCheck =
-      art_quick_invoke_super_trampoline_with_access_check;
-  qpoints->pInvokeVirtualTrampolineWithAccessCheck =
-      art_quick_invoke_virtual_trampoline_with_access_check;
-
-  // Thread
-  qpoints->pTestSuspend = art_quick_test_suspend;
-
-  // Throws
-  qpoints->pDeliverException = art_quick_deliver_exception;
-  qpoints->pThrowArrayBounds = art_quick_throw_array_bounds;
-  qpoints->pThrowDivZero = art_quick_throw_div_zero;
-  qpoints->pThrowNoSuchMethod = art_quick_throw_no_such_method;
-  qpoints->pThrowNullPointer = art_quick_throw_null_pointer_exception;
-  qpoints->pThrowStackOverflow = art_quick_throw_stack_overflow;
-
-  // Deoptimization from compiled code.
-  qpoints->pDeoptimize = art_quick_deoptimize_from_compiled_code;
-
   // Read barrier.
   qpoints->pReadBarrierJni = ReadBarrierJni;
   qpoints->pReadBarrierMark = artReadBarrierMark;
diff --git a/runtime/arch/arm64/instruction_set_features_arm64.cc b/runtime/arch/arm64/instruction_set_features_arm64.cc
index 613bb5c76..cad13b29d 100644
--- a/runtime/arch/arm64/instruction_set_features_arm64.cc
+++ b/runtime/arch/arm64/instruction_set_features_arm64.cc
@@ -39,7 +39,7 @@ const Arm64InstructionSetFeatures* Arm64InstructionSetFeatures::FromVariant(
   if (!needs_a53_835769_fix) {
     // Check to see if this is an expected variant.
     static const char* arm64_known_variants[] = {
-        "denver64", "kryo"
+        "denver64", "kryo", "exynos-m1"
     };
     if (!FindVariantInArray(arm64_known_variants, arraysize(arm64_known_variants), variant)) {
       std::ostringstream os;
diff --git a/runtime/arch/arm64/quick_entrypoints_arm64.S b/runtime/arch/arm64/quick_entrypoints_arm64.S
index 1cdda2d19..1fba09bae 100644
--- a/runtime/arch/arm64/quick_entrypoints_arm64.S
+++ b/runtime/arch/arm64/quick_entrypoints_arm64.S
@@ -1113,6 +1113,14 @@ ENTRY art_quick_lock_object
     RETURN_IF_W0_IS_ZERO_OR_DELIVER
 END art_quick_lock_object
 
+ENTRY art_quick_lock_object_no_inline
+    SETUP_REFS_ONLY_CALLEE_SAVE_FRAME  // save callee saves in case we block
+    mov    x1, xSELF                  // pass Thread::Current
+    bl     artLockObjectFromCode      // (Object* obj, Thread*)
+    RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME
+    RETURN_IF_W0_IS_ZERO_OR_DELIVER
+END art_quick_lock_object_no_inline
+
     /*
      * Entry from managed code that calls artUnlockObjectFromCode and delivers exception on failure.
      * x0 holds the possibly null object to lock.
@@ -1171,6 +1179,14 @@ ENTRY art_quick_unlock_object
     RETURN_IF_W0_IS_ZERO_OR_DELIVER
 END art_quick_unlock_object
 
+ENTRY art_quick_unlock_object_no_inline
+    SETUP_REFS_ONLY_CALLEE_SAVE_FRAME  // save callee saves in case exception allocation triggers GC
+    mov    x1, xSELF                  // pass Thread::Current
+    bl     artUnlockObjectFromCode    // (Object* obj, Thread*)
+    RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME
+    RETURN_IF_W0_IS_ZERO_OR_DELIVER
+END art_quick_unlock_object_no_inline
+
     /*
      * Entry from managed code that calls artIsAssignableFromCode and on failure calls
      * artThrowClassCastException.
@@ -2187,108 +2203,3 @@ ENTRY art_quick_indexof
     asr   x0, x0, #1
     ret
 END art_quick_indexof
-
-   /*
-     * String's compareTo.
-     *
-     * TODO: Not very optimized.
-     *
-     * On entry:
-     *    x0:   this object pointer
-     *    x1:   comp object pointer
-     *
-     */
-    .extern __memcmp16
-ENTRY art_quick_string_compareto
-    mov    x2, x0         // x0 is return, use x2 for first input.
-    sub    x0, x2, x1     // Same string object?
-    cbnz   x0,1f
-    ret
-1:                        // Different string objects.
-
-    ldr    w4, [x2, #MIRROR_STRING_COUNT_OFFSET]
-    ldr    w3, [x1, #MIRROR_STRING_COUNT_OFFSET]
-    add    x2, x2, #MIRROR_STRING_VALUE_OFFSET
-    add    x1, x1, #MIRROR_STRING_VALUE_OFFSET
-
-    /*
-     * Now:           Data*  Count
-     *    first arg    x2      w4
-     *   second arg    x1      w3
-     */
-
-    // x0 := str1.length(w4) - str2.length(w3). ldr zero-extended w3/w4 into x3/x4.
-    subs x0, x4, x3
-    // Min(count1, count2) into w3.
-    csel x3, x3, x4, ge
-
-    // TODO: Tune this value.
-    // Check for long string, do memcmp16 for them.
-    cmp w3, #28  // Constant from arm32.
-    bgt .Ldo_memcmp16
-
-    /*
-     * Now:
-     *   x2: *first string data
-     *   x1: *second string data
-     *   w3: iteration count
-     *   x0: return value if comparison equal
-     *   x4, x5, x6, x7: free
-     */
-
-    // Do a simple unrolled loop.
-.Lloop:
-    // At least two more elements?
-    subs w3, w3, #2
-    b.lt .Lremainder_or_done
-
-    ldrh w4, [x2], #2
-    ldrh w5, [x1], #2
-
-    ldrh w6, [x2], #2
-    ldrh w7, [x1], #2
-
-    subs w4, w4, w5
-    b.ne .Lw4_result
-
-    subs w6, w6, w7
-    b.ne .Lw6_result
-
-    b .Lloop
-
-.Lremainder_or_done:
-    adds w3, w3, #1
-    b.eq .Lremainder
-    ret
-
-.Lremainder:
-    ldrh w4, [x2], #2
-    ldrh w5, [x1], #2
-    subs w4, w4, w5
-    b.ne .Lw4_result
-    ret
-
-// Result is in w4
-.Lw4_result:
-    sxtw x0, w4
-    ret
-
-// Result is in w6
-.Lw6_result:
-    sxtw x0, w6
-    ret
-
-.Ldo_memcmp16:
-    mov x14, x0                  // Save x0 and LR. __memcmp16 does not use these temps.
-    mov x15, xLR                 //                 TODO: Codify and check that?
-
-    mov x0, x2
-    uxtw x2, w3
-    bl __memcmp16
-
-    mov xLR, x15                 // Restore LR.
-
-    cmp x0, #0                   // Check the memcmp difference.
-    csel x0, x0, x14, ne         // x0 := x0 != 0 ? x14(prev x0=length diff) : x1.
-    ret
-END art_quick_string_compareto
diff --git a/runtime/arch/mips/entrypoints_init_mips.cc b/runtime/arch/mips/entrypoints_init_mips.cc
index 51eb77f40..45e33a850 100644
--- a/runtime/arch/mips/entrypoints_init_mips.cc
+++ b/runtime/arch/mips/entrypoints_init_mips.cc
@@ -59,6 +59,9 @@ extern "C" int64_t __divdi3(int64_t, int64_t);
 extern "C" int64_t __moddi3(int64_t, int64_t);
 
 void InitEntryPoints(JniEntryPoints* jpoints, QuickEntryPoints* qpoints) {
+  // Note: MIPS has asserts checking for the type of entrypoint. Don't move it
+  //       to InitDefaultEntryPoints().
+
   // JNI
   jpoints->pDlsymLookup = art_jni_dlsym_lookup_stub;
 
@@ -167,9 +170,14 @@ void InitEntryPoints(JniEntryPoints* jpoints, QuickEntryPoints* qpoints) {
                 "Non-direct C stub marked direct.");
 
   // Locks
-  qpoints->pLockObject = art_quick_lock_object;
+  if (UNLIKELY(VLOG_IS_ON(systrace_lock_logging))) {
+    qpoints->pLockObject = art_quick_lock_object_no_inline;
+    qpoints->pUnlockObject = art_quick_unlock_object_no_inline;
+  } else {
+    qpoints->pLockObject = art_quick_lock_object;
+    qpoints->pUnlockObject = art_quick_unlock_object;
+  }
   static_assert(!IsDirectEntrypoint(kQuickLockObject), "Non-direct C stub marked direct.");
-  qpoints->pUnlockObject = art_quick_unlock_object;
   static_assert(!IsDirectEntrypoint(kQuickUnlockObject), "Non-direct C stub marked direct.");
 
   // Math
diff --git a/runtime/arch/mips/quick_entrypoints_mips.S b/runtime/arch/mips/quick_entrypoints_mips.S
index 8939a488e..3ee26afc4 100644
--- a/runtime/arch/mips/quick_entrypoints_mips.S
+++ b/runtime/arch/mips/quick_entrypoints_mips.S
@@ -906,6 +906,16 @@ ENTRY art_quick_lock_object
     RETURN_IF_ZERO
 END art_quick_lock_object
 
+ENTRY art_quick_lock_object_no_inline
+    beqz    $a0, .Lart_quick_throw_null_pointer_exception_gp_set
+    nop
+    SETUP_REFS_ONLY_CALLEE_SAVE_FRAME     # save callee saves in case we block
+    la      $t9, artLockObjectFromCode
+    jalr    $t9                           # (Object* obj, Thread*)
+    move    $a1, rSELF                    # pass Thread::Current
+    RETURN_IF_ZERO
+END art_quick_lock_object_no_inline
+
     /*
      * Entry from managed code that calls artUnlockObjectFromCode and delivers exception on failure.
      */
@@ -920,6 +930,16 @@ ENTRY art_quick_unlock_object
     RETURN_IF_ZERO
 END art_quick_unlock_object
 
+ENTRY art_quick_unlock_object_no_inline
+    beqz    $a0, .Lart_quick_throw_null_pointer_exception_gp_set
+    nop
+    SETUP_REFS_ONLY_CALLEE_SAVE_FRAME # save callee saves in case exception allocation triggers GC
+    la      $t9, artUnlockObjectFromCode
+    jalr    $t9                       # (Object* obj, Thread*)
+    move    $a1, rSELF                # pass Thread::Current
+    RETURN_IF_ZERO
+END art_quick_unlock_object_no_inline
+
     /*
      * Entry from managed code that calls artCheckCastFromCode and delivers exception on failure.
      */
diff --git a/runtime/arch/mips64/entrypoints_init_mips64.cc b/runtime/arch/mips64/entrypoints_init_mips64.cc
index 4bdb38e51..030c12707 100644
--- a/runtime/arch/mips64/entrypoints_init_mips64.cc
+++ b/runtime/arch/mips64/entrypoints_init_mips64.cc
@@ -18,6 +18,7 @@
 #include "entrypoints/jni/jni_entrypoints.h"
 #include "entrypoints/quick/quick_alloc_entrypoints.h"
 #include "entrypoints/quick/quick_default_externs.h"
+#include "entrypoints/quick/quick_default_init_entrypoints.h"
 #include "entrypoints/quick/quick_entrypoints.h"
 #include "entrypoints/entrypoint_utils.h"
 #include "entrypoints/math_entrypoints.h"
@@ -57,67 +58,12 @@ extern "C" int64_t __divdi3(int64_t, int64_t);
 extern "C" int64_t __moddi3(int64_t, int64_t);
 
 void InitEntryPoints(JniEntryPoints* jpoints, QuickEntryPoints* qpoints) {
-  // JNI
-  jpoints->pDlsymLookup = art_jni_dlsym_lookup_stub;
-
-  // Alloc
-  ResetQuickAllocEntryPoints(qpoints);
+  DefaultInitEntryPoints(jpoints, qpoints);
 
   // Cast
   qpoints->pInstanceofNonTrivial = artIsAssignableFromCode;
   qpoints->pCheckCast = art_quick_check_cast;
 
-  // DexCache
-  qpoints->pInitializeStaticStorage = art_quick_initialize_static_storage;
-  qpoints->pInitializeTypeAndVerifyAccess = art_quick_initialize_type_and_verify_access;
-  qpoints->pInitializeType = art_quick_initialize_type;
-  qpoints->pResolveString = art_quick_resolve_string;
-
-  // Field
-  qpoints->pSet8Instance = art_quick_set8_instance;
-  qpoints->pSet8Static = art_quick_set8_static;
-  qpoints->pSet16Instance = art_quick_set16_instance;
-  qpoints->pSet16Static = art_quick_set16_static;
-  qpoints->pSet32Instance = art_quick_set32_instance;
-  qpoints->pSet32Static = art_quick_set32_static;
-  qpoints->pSet64Instance = art_quick_set64_instance;
-  qpoints->pSet64Static = art_quick_set64_static;
-  qpoints->pSetObjInstance = art_quick_set_obj_instance;
-  qpoints->pSetObjStatic = art_quick_set_obj_static;
-  qpoints->pGetBooleanInstance = art_quick_get_boolean_instance;
-  qpoints->pGetByteInstance = art_quick_get_byte_instance;
-  qpoints->pGetCharInstance = art_quick_get_char_instance;
-  qpoints->pGetShortInstance = art_quick_get_short_instance;
-  qpoints->pGet32Instance = art_quick_get32_instance;
-  qpoints->pGet64Instance = art_quick_get64_instance;
-  qpoints->pGetObjInstance = art_quick_get_obj_instance;
-  qpoints->pGetBooleanStatic = art_quick_get_boolean_static;
-  qpoints->pGetByteStatic = art_quick_get_byte_static;
-  qpoints->pGetCharStatic = art_quick_get_char_static;
-  qpoints->pGetShortStatic = art_quick_get_short_static;
-  qpoints->pGet32Static = art_quick_get32_static;
-  qpoints->pGet64Static = art_quick_get64_static;
-  qpoints->pGetObjStatic = art_quick_get_obj_static;
-
-  // Array
-  qpoints->pAputObjectWithNullAndBoundCheck = art_quick_aput_obj_with_null_and_bound_check;
-  qpoints->pAputObjectWithBoundCheck = art_quick_aput_obj_with_bound_check;
-  qpoints->pAputObject = art_quick_aput_obj;
-  qpoints->pHandleFillArrayData = art_quick_handle_fill_data;
-
-  // JNI
-  qpoints->pJniMethodStart = JniMethodStart;
-  qpoints->pJniMethodStartSynchronized = JniMethodStartSynchronized;
-  qpoints->pJniMethodEnd = JniMethodEnd;
-  qpoints->pJniMethodEndSynchronized = JniMethodEndSynchronized;
-  qpoints->pJniMethodEndWithReference = JniMethodEndWithReference;
-  qpoints->pJniMethodEndWithReferenceSynchronized = JniMethodEndWithReferenceSynchronized;
-  qpoints->pQuickGenericJniTrampoline = art_quick_generic_jni_trampoline;
-
-  // Locks
-  qpoints->pLockObject = art_quick_lock_object;
-  qpoints->pUnlockObject = art_quick_unlock_object;
-
   // Math
   qpoints->pCmpgDouble = CmpgDouble;
   qpoints->pCmpgFloat = CmpgFloat;
@@ -144,35 +90,6 @@ void InitEntryPoints(JniEntryPoints* jpoints, QuickEntryPoints* qpoints) {
   qpoints->pStringCompareTo = art_quick_string_compareto;
   qpoints->pMemcpy = memcpy;
 
-  // Invocation
-  qpoints->pQuickImtConflictTrampoline = art_quick_imt_conflict_trampoline;
-  qpoints->pQuickResolutionTrampoline = art_quick_resolution_trampoline;
-  qpoints->pQuickToInterpreterBridge = art_quick_to_interpreter_bridge;
-  qpoints->pInvokeDirectTrampolineWithAccessCheck =
-      art_quick_invoke_direct_trampoline_with_access_check;
-  qpoints->pInvokeInterfaceTrampolineWithAccessCheck =
-      art_quick_invoke_interface_trampoline_with_access_check;
-  qpoints->pInvokeStaticTrampolineWithAccessCheck =
-      art_quick_invoke_static_trampoline_with_access_check;
-  qpoints->pInvokeSuperTrampolineWithAccessCheck =
-      art_quick_invoke_super_trampoline_with_access_check;
-  qpoints->pInvokeVirtualTrampolineWithAccessCheck =
-      art_quick_invoke_virtual_trampoline_with_access_check;
-
-  // Thread
-  qpoints->pTestSuspend = art_quick_test_suspend;
-
-  // Throws
-  qpoints->pDeliverException = art_quick_deliver_exception;
-  qpoints->pThrowArrayBounds = art_quick_throw_array_bounds;
-  qpoints->pThrowDivZero = art_quick_throw_div_zero;
-  qpoints->pThrowNoSuchMethod = art_quick_throw_no_such_method;
-  qpoints->pThrowNullPointer = art_quick_throw_null_pointer_exception;
-  qpoints->pThrowStackOverflow = art_quick_throw_stack_overflow;
-
-  // Deoptimization from compiled code.
-  qpoints->pDeoptimize = art_quick_deoptimize_from_compiled_code;
-
   // TODO - use lld/scd instructions for Mips64
   // Atomic 64-bit load/store
   qpoints->pA64Load = QuasiAtomic::Read64;
diff --git a/runtime/arch/mips64/quick_entrypoints_mips64.S b/runtime/arch/mips64/quick_entrypoints_mips64.S
index 5d0c94c63..8f1a35a69 100644
--- a/runtime/arch/mips64/quick_entrypoints_mips64.S
+++ b/runtime/arch/mips64/quick_entrypoints_mips64.S
@@ -971,6 +971,15 @@ ENTRY art_quick_lock_object
     RETURN_IF_ZERO
 END art_quick_lock_object
 
+ENTRY art_quick_lock_object_no_inline
+    beq     $a0, $zero, .Lart_quick_throw_null_pointer_exception_gp_set
+    nop
+    SETUP_REFS_ONLY_CALLEE_SAVE_FRAME     # save callee saves in case we block
+    jal     artLockObjectFromCode         # (Object* obj, Thread*)
+    move    $a1, rSELF                    # pass Thread::Current
+    RETURN_IF_ZERO
+END art_quick_lock_object_no_inline
+
     /*
      * Entry from managed code that calls artUnlockObjectFromCode and delivers exception on failure.
      */
@@ -984,6 +993,15 @@ ENTRY art_quick_unlock_object
     RETURN_IF_ZERO
 END art_quick_unlock_object
 
+ENTRY art_quick_unlock_object_no_inline
+    beq     $a0, $zero, .Lart_quick_throw_null_pointer_exception_gp_set
+    nop
+    SETUP_REFS_ONLY_CALLEE_SAVE_FRAME  # save callee saves in case exception allocation triggers GC
+    jal     artUnlockObjectFromCode    # (Object* obj, Thread*)
+    move    $a1, rSELF                 # pass Thread::Current
+    RETURN_IF_ZERO
+END art_quick_unlock_object_no_inline
+
     /*
      * Entry from managed code that calls artCheckCastFromCode and delivers exception on failure.
      */
diff --git a/runtime/arch/stub_test.cc b/runtime/arch/stub_test.cc
index 75d9073cf..02629e819 100644
--- a/runtime/arch/stub_test.cc
+++ b/runtime/arch/stub_test.cc
@@ -1205,7 +1205,8 @@ TEST_F(StubTest, AllocObjectArray) {
 
 
 TEST_F(StubTest, StringCompareTo) {
-#if defined(__i386__) || defined(__arm__) || defined(__aarch64__) || \
+  // There is no StringCompareTo runtime entrypoint for __aarch64__.
+#if defined(__i386__) || defined(__arm__) || \
     defined(__mips__) || (defined(__x86_64__) && !defined(__APPLE__))
   // TODO: Check the "Unresolved" allocation stubs
 
@@ -2010,14 +2011,14 @@ TEST_F(StubTest, DISABLED_IMT) {
   // that will create it: the runtime stub expects to be called by compiled code.
   LinearAlloc* linear_alloc = Runtime::Current()->GetLinearAlloc();
   ArtMethod* conflict_method = Runtime::Current()->CreateImtConflictMethod(linear_alloc);
-  static ImtConflictTable::Entry empty_entry = { nullptr, nullptr };
-  ImtConflictTable* empty_conflict_table = reinterpret_cast<ImtConflictTable*>(&empty_entry);
+  ImtConflictTable* empty_conflict_table =
+      Runtime::Current()->GetClassLinker()->CreateImtConflictTable(/*count*/0u, linear_alloc);
   void* data = linear_alloc->Alloc(
       self,
-      ImtConflictTable::ComputeSizeWithOneMoreEntry(empty_conflict_table));
+      ImtConflictTable::ComputeSizeWithOneMoreEntry(empty_conflict_table, sizeof(void*)));
   ImtConflictTable* new_table = new (data) ImtConflictTable(
-      empty_conflict_table, inf_contains, contains_amethod);
-  conflict_method->SetImtConflictTable(new_table);
+      empty_conflict_table, inf_contains, contains_amethod, sizeof(void*));
+  conflict_method->SetImtConflictTable(new_table, sizeof(void*));
 
   size_t result =
       Invoke3WithReferrerAndHidden(reinterpret_cast<size_t>(conflict_method),
diff --git a/runtime/arch/x86/entrypoints_init_x86.cc b/runtime/arch/x86/entrypoints_init_x86.cc
index e593f39fd..15a857146 100644
--- a/runtime/arch/x86/entrypoints_init_x86.cc
+++ b/runtime/arch/x86/entrypoints_init_x86.cc
@@ -17,6 +17,7 @@
 #include "entrypoints/jni/jni_entrypoints.h"
 #include "entrypoints/quick/quick_alloc_entrypoints.h"
 #include "entrypoints/quick/quick_default_externs.h"
+#include "entrypoints/quick/quick_default_init_entrypoints.h"
 #include "entrypoints/quick/quick_entrypoints.h"
 #include "entrypoints/runtime_asm_entrypoints.h"
 #include "interpreter/interpreter.h"
@@ -33,67 +34,12 @@ extern "C" mirror::Object* art_quick_read_barrier_slow(mirror::Object*, mirror::
 extern "C" mirror::Object* art_quick_read_barrier_for_root_slow(GcRoot<mirror::Object>*);
 
 void InitEntryPoints(JniEntryPoints* jpoints, QuickEntryPoints* qpoints) {
-  // JNI
-  jpoints->pDlsymLookup = art_jni_dlsym_lookup_stub;
-
-  // Alloc
-  ResetQuickAllocEntryPoints(qpoints);
+  DefaultInitEntryPoints(jpoints, qpoints);
 
   // Cast
   qpoints->pInstanceofNonTrivial = art_quick_is_assignable;
   qpoints->pCheckCast = art_quick_check_cast;
 
-  // DexCache
-  qpoints->pInitializeStaticStorage = art_quick_initialize_static_storage;
-  qpoints->pInitializeTypeAndVerifyAccess = art_quick_initialize_type_and_verify_access;
-  qpoints->pInitializeType = art_quick_initialize_type;
-  qpoints->pResolveString = art_quick_resolve_string;
-
-  // Field
-  qpoints->pSet8Instance = art_quick_set8_instance;
-  qpoints->pSet8Static = art_quick_set8_static;
-  qpoints->pSet16Instance = art_quick_set16_instance;
-  qpoints->pSet16Static = art_quick_set16_static;
-  qpoints->pSet32Instance = art_quick_set32_instance;
-  qpoints->pSet32Static = art_quick_set32_static;
-  qpoints->pSet64Instance = art_quick_set64_instance;
-  qpoints->pSet64Static = art_quick_set64_static;
-  qpoints->pSetObjInstance = art_quick_set_obj_instance;
-  qpoints->pSetObjStatic = art_quick_set_obj_static;
-  qpoints->pGetByteInstance = art_quick_get_byte_instance;
-  qpoints->pGetBooleanInstance = art_quick_get_boolean_instance;
-  qpoints->pGetShortInstance = art_quick_get_short_instance;
-  qpoints->pGetCharInstance = art_quick_get_char_instance;
-  qpoints->pGet32Instance = art_quick_get32_instance;
-  qpoints->pGet64Instance = art_quick_get64_instance;
-  qpoints->pGetObjInstance = art_quick_get_obj_instance;
-  qpoints->pGetByteStatic = art_quick_get_byte_static;
-  qpoints->pGetBooleanStatic = art_quick_get_boolean_static;
-  qpoints->pGetShortStatic = art_quick_get_short_static;
-  qpoints->pGetCharStatic = art_quick_get_char_static;
-  qpoints->pGet32Static = art_quick_get32_static;
-  qpoints->pGet64Static = art_quick_get64_static;
-  qpoints->pGetObjStatic = art_quick_get_obj_static;
-
-  // Array
-  qpoints->pAputObjectWithNullAndBoundCheck = art_quick_aput_obj_with_null_and_bound_check;
-  qpoints->pAputObjectWithBoundCheck = art_quick_aput_obj_with_bound_check;
-  qpoints->pAputObject = art_quick_aput_obj;
-  qpoints->pHandleFillArrayData = art_quick_handle_fill_data;
-
-  // JNI
-  qpoints->pJniMethodStart = JniMethodStart;
-  qpoints->pJniMethodStartSynchronized = JniMethodStartSynchronized;
-  qpoints->pJniMethodEnd = JniMethodEnd;
-  qpoints->pJniMethodEndSynchronized = JniMethodEndSynchronized;
-  qpoints->pJniMethodEndWithReference = JniMethodEndWithReference;
-  qpoints->pJniMethodEndWithReferenceSynchronized = JniMethodEndWithReferenceSynchronized;
-  qpoints->pQuickGenericJniTrampoline = art_quick_generic_jni_trampoline;
-
-  // Locks
-  qpoints->pLockObject = art_quick_lock_object;
-  qpoints->pUnlockObject = art_quick_unlock_object;
-
   // More math.
   qpoints->pCos = cos;
   qpoints->pSin = sin;
@@ -128,35 +74,6 @@ void InitEntryPoints(JniEntryPoints* jpoints, QuickEntryPoints* qpoints) {
   qpoints->pStringCompareTo = art_quick_string_compareto;
   qpoints->pMemcpy = art_quick_memcpy;
 
-  // Invocation
-  qpoints->pQuickImtConflictTrampoline = art_quick_imt_conflict_trampoline;
-  qpoints->pQuickResolutionTrampoline = art_quick_resolution_trampoline;
-  qpoints->pQuickToInterpreterBridge = art_quick_to_interpreter_bridge;
-  qpoints->pInvokeDirectTrampolineWithAccessCheck =
-      art_quick_invoke_direct_trampoline_with_access_check;
-  qpoints->pInvokeInterfaceTrampolineWithAccessCheck =
-      art_quick_invoke_interface_trampoline_with_access_check;
-  qpoints->pInvokeStaticTrampolineWithAccessCheck =
-      art_quick_invoke_static_trampoline_with_access_check;
-  qpoints->pInvokeSuperTrampolineWithAccessCheck =
-      art_quick_invoke_super_trampoline_with_access_check;
-  qpoints->pInvokeVirtualTrampolineWithAccessCheck =
-      art_quick_invoke_virtual_trampoline_with_access_check;
-
-  // Thread
-  qpoints->pTestSuspend = art_quick_test_suspend;
-
-  // Throws
-  qpoints->pDeliverException = art_quick_deliver_exception;
-  qpoints->pThrowArrayBounds = art_quick_throw_array_bounds;
-  qpoints->pThrowDivZero = art_quick_throw_div_zero;
-  qpoints->pThrowNoSuchMethod = art_quick_throw_no_such_method;
-  qpoints->pThrowNullPointer = art_quick_throw_null_pointer_exception;
-  qpoints->pThrowStackOverflow = art_quick_throw_stack_overflow;
-
-  // Deoptimize
-  qpoints->pDeoptimize = art_quick_deoptimize_from_compiled_code;
-
   // Read barrier.
   qpoints->pReadBarrierJni = ReadBarrierJni;
   qpoints->pReadBarrierMark = art_quick_read_barrier_mark;
diff --git a/runtime/arch/x86/quick_entrypoints_x86.S b/runtime/arch/x86/quick_entrypoints_x86.S
index 4f9b3f787..485da9fe3 100644
--- a/runtime/arch/x86/quick_entrypoints_x86.S
+++ b/runtime/arch/x86/quick_entrypoints_x86.S
@@ -1075,6 +1075,22 @@ DEFINE_FUNCTION art_quick_lock_object
     RETURN_IF_EAX_ZERO
 END_FUNCTION art_quick_lock_object
 
+DEFINE_FUNCTION art_quick_lock_object_no_inline
+    SETUP_REFS_ONLY_CALLEE_SAVE_FRAME  ebx, ebx  // save ref containing registers for GC
+    // Outgoing argument set up
+    subl LITERAL(8), %esp                 // alignment padding
+    CFI_ADJUST_CFA_OFFSET(8)
+    pushl %fs:THREAD_SELF_OFFSET          // pass Thread::Current()
+    CFI_ADJUST_CFA_OFFSET(4)
+    PUSH eax                              // pass object
+    call SYMBOL(artLockObjectFromCode)    // artLockObjectFromCode(object, Thread*)
+    addl LITERAL(16), %esp                // pop arguments
+    CFI_ADJUST_CFA_OFFSET(-16)
+    RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME   // restore frame up to return address
+    RETURN_IF_EAX_ZERO
+END_FUNCTION art_quick_lock_object_no_inline
+
+
 DEFINE_FUNCTION art_quick_unlock_object
     testl %eax, %eax                      // null check object/eax
     jz   .Lslow_unlock
@@ -1130,6 +1146,21 @@ DEFINE_FUNCTION art_quick_unlock_object
     RETURN_IF_EAX_ZERO
 END_FUNCTION art_quick_unlock_object
 
+DEFINE_FUNCTION art_quick_unlock_object_no_inline
+    SETUP_REFS_ONLY_CALLEE_SAVE_FRAME  ebx, ebx  // save ref containing registers for GC
+    // Outgoing argument set up
+    subl LITERAL(8), %esp                 // alignment padding
+    CFI_ADJUST_CFA_OFFSET(8)
+    pushl %fs:THREAD_SELF_OFFSET          // pass Thread::Current()
+    CFI_ADJUST_CFA_OFFSET(4)
+    PUSH eax                              // pass object
+    call SYMBOL(artUnlockObjectFromCode)  // artUnlockObjectFromCode(object, Thread*)
+    addl LITERAL(16), %esp                // pop arguments
+    CFI_ADJUST_CFA_OFFSET(-16)
+    RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME   // restore frame up to return address
+    RETURN_IF_EAX_ZERO
+END_FUNCTION art_quick_unlock_object_no_inline
+
 DEFINE_FUNCTION art_quick_is_assignable
     PUSH eax                              // alignment padding
     PUSH ecx                              // pass arg2 - obj->klass
diff --git a/runtime/arch/x86_64/entrypoints_init_x86_64.cc b/runtime/arch/x86_64/entrypoints_init_x86_64.cc
index 0a5d14a16..bd6df700d 100644
--- a/runtime/arch/x86_64/entrypoints_init_x86_64.cc
+++ b/runtime/arch/x86_64/entrypoints_init_x86_64.cc
@@ -17,6 +17,9 @@
 #include "entrypoints/jni/jni_entrypoints.h"
 #include "entrypoints/quick/quick_alloc_entrypoints.h"
 #include "entrypoints/quick/quick_default_externs.h"
+#if !defined(__APPLE__)
+#include "entrypoints/quick/quick_default_init_entrypoints.h"
+#endif
 #include "entrypoints/quick/quick_entrypoints.h"
 #include "entrypoints/math_entrypoints.h"
 #include "entrypoints/runtime_asm_entrypoints.h"
@@ -38,67 +41,12 @@ void InitEntryPoints(JniEntryPoints* jpoints, QuickEntryPoints* qpoints) {
   UNUSED(jpoints, qpoints);
   UNIMPLEMENTED(FATAL);
 #else
-  // JNI
-  jpoints->pDlsymLookup = art_jni_dlsym_lookup_stub;
-
-  // Alloc
-  ResetQuickAllocEntryPoints(qpoints);
+  DefaultInitEntryPoints(jpoints, qpoints);
 
   // Cast
   qpoints->pInstanceofNonTrivial = art_quick_assignable_from_code;
   qpoints->pCheckCast = art_quick_check_cast;
 
-  // DexCache
-  qpoints->pInitializeStaticStorage = art_quick_initialize_static_storage;
-  qpoints->pInitializeTypeAndVerifyAccess = art_quick_initialize_type_and_verify_access;
-  qpoints->pInitializeType = art_quick_initialize_type;
-  qpoints->pResolveString = art_quick_resolve_string;
-
-  // Field
-  qpoints->pSet8Instance = art_quick_set8_instance;
-  qpoints->pSet8Static = art_quick_set8_static;
-  qpoints->pSet16Instance = art_quick_set16_instance;
-  qpoints->pSet16Static = art_quick_set16_static;
-  qpoints->pSet32Instance = art_quick_set32_instance;
-  qpoints->pSet32Static = art_quick_set32_static;
-  qpoints->pSet64Instance = art_quick_set64_instance;
-  qpoints->pSet64Static = art_quick_set64_static;
-  qpoints->pSetObjInstance = art_quick_set_obj_instance;
-  qpoints->pSetObjStatic = art_quick_set_obj_static;
-  qpoints->pGetByteInstance = art_quick_get_byte_instance;
-  qpoints->pGetBooleanInstance = art_quick_get_boolean_instance;
-  qpoints->pGetShortInstance = art_quick_get_short_instance;
-  qpoints->pGetCharInstance = art_quick_get_char_instance;
-  qpoints->pGet32Instance = art_quick_get32_instance;
-  qpoints->pGet64Instance = art_quick_get64_instance;
-  qpoints->pGetObjInstance = art_quick_get_obj_instance;
-  qpoints->pGetByteStatic = art_quick_get_byte_static;
-  qpoints->pGetBooleanStatic = art_quick_get_boolean_static;
-  qpoints->pGetShortStatic = art_quick_get_short_static;
-  qpoints->pGetCharStatic = art_quick_get_char_static;
-  qpoints->pGet32Static = art_quick_get32_static;
-  qpoints->pGet64Static = art_quick_get64_static;
-  qpoints->pGetObjStatic = art_quick_get_obj_static;
-
-  // Array
-  qpoints->pAputObjectWithNullAndBoundCheck = art_quick_aput_obj_with_null_and_bound_check;
-  qpoints->pAputObjectWithBoundCheck = art_quick_aput_obj_with_bound_check;
-  qpoints->pAputObject = art_quick_aput_obj;
-  qpoints->pHandleFillArrayData = art_quick_handle_fill_data;
-
-  // JNI
-  qpoints->pJniMethodStart = JniMethodStart;
-  qpoints->pJniMethodStartSynchronized = JniMethodStartSynchronized;
-  qpoints->pJniMethodEnd = JniMethodEnd;
-  qpoints->pJniMethodEndSynchronized = JniMethodEndSynchronized;
-  qpoints->pJniMethodEndWithReference = JniMethodEndWithReference;
-  qpoints->pJniMethodEndWithReferenceSynchronized = JniMethodEndWithReferenceSynchronized;
-  qpoints->pQuickGenericJniTrampoline = art_quick_generic_jni_trampoline;
-
-  // Locks
-  qpoints->pLockObject = art_quick_lock_object;
-  qpoints->pUnlockObject = art_quick_unlock_object;
-
   // More math.
   qpoints->pCos = cos;
   qpoints->pSin = sin;
@@ -132,35 +80,6 @@ void InitEntryPoints(JniEntryPoints* jpoints, QuickEntryPoints* qpoints) {
   qpoints->pStringCompareTo = art_quick_string_compareto;
   qpoints->pMemcpy = art_quick_memcpy;
 
-  // Invocation
-  qpoints->pQuickImtConflictTrampoline = art_quick_imt_conflict_trampoline;
-  qpoints->pQuickResolutionTrampoline = art_quick_resolution_trampoline;
-  qpoints->pQuickToInterpreterBridge = art_quick_to_interpreter_bridge;
-  qpoints->pInvokeDirectTrampolineWithAccessCheck =
-      art_quick_invoke_direct_trampoline_with_access_check;
-  qpoints->pInvokeInterfaceTrampolineWithAccessCheck =
-      art_quick_invoke_interface_trampoline_with_access_check;
-  qpoints->pInvokeStaticTrampolineWithAccessCheck =
-      art_quick_invoke_static_trampoline_with_access_check;
-  qpoints->pInvokeSuperTrampolineWithAccessCheck =
-      art_quick_invoke_super_trampoline_with_access_check;
-  qpoints->pInvokeVirtualTrampolineWithAccessCheck =
-      art_quick_invoke_virtual_trampoline_with_access_check;
-
-  // Thread
-  qpoints->pTestSuspend = art_quick_test_suspend;
-
-  // Throws
-  qpoints->pDeliverException = art_quick_deliver_exception;
-  qpoints->pThrowArrayBounds = art_quick_throw_array_bounds;
-  qpoints->pThrowDivZero = art_quick_throw_div_zero;
-  qpoints->pThrowNoSuchMethod = art_quick_throw_no_such_method;
-  qpoints->pThrowNullPointer = art_quick_throw_null_pointer_exception;
-  qpoints->pThrowStackOverflow = art_quick_throw_stack_overflow;
-
-  // Deoptimize
-  qpoints->pDeoptimize = art_quick_deoptimize_from_compiled_code;
-
   // Read barrier.
   qpoints->pReadBarrierJni = ReadBarrierJni;
   qpoints->pReadBarrierMark = art_quick_read_barrier_mark;
diff --git a/runtime/arch/x86_64/quick_entrypoints_x86_64.S b/runtime/arch/x86_64/quick_entrypoints_x86_64.S
index 26e668e7a..8064ed696 100644
--- a/runtime/arch/x86_64/quick_entrypoints_x86_64.S
+++ b/runtime/arch/x86_64/quick_entrypoints_x86_64.S
@@ -894,57 +894,107 @@ DEFINE_FUNCTION art_quick_alloc_object_rosalloc
     RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER                   // return or deliver exception
 END_FUNCTION art_quick_alloc_object_rosalloc
 
-// A handle-written override for GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT(_tlab, TLAB).
-DEFINE_FUNCTION art_quick_alloc_object_tlab
-    // Fast path tlab allocation.
-    // RDI: uint32_t type_idx, RSI: ArtMethod*
-    // RDX, RCX, R8, R9: free. RAX: return val.
-    // TODO: Add read barrier when this function is used.
-    // Note this function can/should implement read barrier fast path only
-    // (no read barrier slow path) because this is the fast path of tlab allocation.
-    // We can fall back to the allocation slow path to do the read barrier slow path.
-#if defined(USE_READ_BARRIER)
-    int3
-    int3
-#endif
-    // Might need a special macro since rsi and edx is 32b/64b mismatched.
-    movq ART_METHOD_DEX_CACHE_TYPES_OFFSET_64(%rsi), %rdx  // Load dex cache resolved types array
-    // TODO: Add read barrier when this function is used.
-    // Might need to break down into multiple instructions to get the base address in a register.
-                                                               // Load the class
-    movl 0(%rdx, %rdi, COMPRESSED_REFERENCE_SIZE), %edx
+// The common fast path code for art_quick_alloc_object_tlab and art_quick_alloc_object_region_tlab.
+//
+// RDI: type_idx, RSI: ArtMethod*, RDX/EDX: the class, RAX: return value.
+// RCX: scratch, r8: Thread::Current().
+MACRO1(ALLOC_OBJECT_TLAB_FAST_PATH, slowPathLabel)
     testl %edx, %edx                                           // Check null class
-    jz   .Lart_quick_alloc_object_tlab_slow_path
+    jz   RAW_VAR(slowPathLabel)
                                                                // Check class status.
     cmpl LITERAL(MIRROR_CLASS_STATUS_INITIALIZED), MIRROR_CLASS_STATUS_OFFSET(%rdx)
-    jne  .Lart_quick_alloc_object_tlab_slow_path
-                                                               // Check access flags has kAccClassIsFinalizable
+    jne  RAW_VAR(slowPathLabel)
+                                                               // No fake dependence needed on x86
+                                                               // between status and flags load,
+                                                               // since each load is a load-acquire,
+                                                               // no loads reordering.
+                                                               // Check access flags has
+                                                               // kAccClassIsFinalizable
     testl LITERAL(ACCESS_FLAGS_CLASS_IS_FINALIZABLE), MIRROR_CLASS_ACCESS_FLAGS_OFFSET(%rdx)
-    jnz  .Lart_quick_alloc_object_tlab_slow_path
+    jnz  RAW_VAR(slowPathLabel)
+    movq %gs:THREAD_SELF_OFFSET, %r8                           // r8 = thread
+    movq THREAD_LOCAL_END_OFFSET(%r8), %rax                    // Load thread_local_end.
+    subq THREAD_LOCAL_POS_OFFSET(%r8), %rax                    // Compute the remaining buffer size.
     movl MIRROR_CLASS_OBJECT_SIZE_OFFSET(%rdx), %ecx           // Load the object size.
+    cmpq %rax, %rcx                                            // Check if it fits. OK to do this
+                                                               // before rounding up the object size
+                                                               // assuming the buf size alignment.
+    ja   RAW_VAR(slowPathLabel)
     addl LITERAL(OBJECT_ALIGNMENT_MASK), %ecx                  // Align the size by 8. (addr + 7) & ~7.
     andl LITERAL(OBJECT_ALIGNMENT_MASK_TOGGLED), %ecx
-    movq %gs:THREAD_SELF_OFFSET, %r8                           // r8 = thread
-    movq THREAD_LOCAL_POS_OFFSET(%r8), %rax                    // Load thread_local_pos.
+    movq THREAD_LOCAL_POS_OFFSET(%r8), %rax                    // Load thread_local_pos
+                                                               // as allocated object.
     addq %rax, %rcx                                            // Add the object size.
-    cmpq THREAD_LOCAL_END_OFFSET(%r8), %rcx                    // Check if it fits.
-    ja   .Lart_quick_alloc_object_tlab_slow_path
     movq %rcx, THREAD_LOCAL_POS_OFFSET(%r8)                    // Update thread_local_pos.
-    addq LITERAL(1), THREAD_LOCAL_OBJECTS_OFFSET(%r8)          // Increment thread_local_objects.
+    addq LITERAL(1), THREAD_LOCAL_OBJECTS_OFFSET(%r8)          // Increase thread_local_objects.
                                                                // Store the class pointer in the header.
                                                                // No fence needed for x86.
+    POISON_HEAP_REF edx
     movl %edx, MIRROR_OBJECT_CLASS_OFFSET(%rax)
     ret                                                        // Fast path succeeded.
-.Lart_quick_alloc_object_tlab_slow_path:
+END_MACRO
+
+// The common slow path code for art_quick_alloc_object_tlab and art_quick_alloc_object_region_tlab.
+MACRO1(ALLOC_OBJECT_TLAB_SLOW_PATH, cxx_name)
     SETUP_REFS_ONLY_CALLEE_SAVE_FRAME                          // save ref containing registers for GC
     // Outgoing argument set up
     movq %gs:THREAD_SELF_OFFSET, %rdx                          // pass Thread::Current()
-    call SYMBOL(artAllocObjectFromCodeTLAB)                    // cxx_name(arg0, arg1, Thread*)
+    call VAR(cxx_name)                                         // cxx_name(arg0, arg1, Thread*)
     RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME                        // restore frame up to return address
     RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER                    // return or deliver exception
+END_MACRO
+
+// A hand-written override for GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT(_tlab, TLAB).
+DEFINE_FUNCTION art_quick_alloc_object_tlab
+    // Fast path tlab allocation.
+    // RDI: uint32_t type_idx, RSI: ArtMethod*
+    // RDX, RCX, R8, R9: free. RAX: return val.
+#if defined(USE_READ_BARRIER)
+    int3
+    int3
+#endif
+    // Might need a special macro since rsi and edx is 32b/64b mismatched.
+    movq ART_METHOD_DEX_CACHE_TYPES_OFFSET_64(%rsi), %rdx  // Load dex cache resolved types array
+    // Might need to break down into multiple instructions to get the base address in a register.
+                                                               // Load the class
+    movl 0(%rdx, %rdi, COMPRESSED_REFERENCE_SIZE), %edx
+    ALLOC_OBJECT_TLAB_FAST_PATH .Lart_quick_alloc_object_tlab_slow_path
+.Lart_quick_alloc_object_tlab_slow_path:
+    ALLOC_OBJECT_TLAB_SLOW_PATH artAllocObjectFromCodeTLAB
 END_FUNCTION art_quick_alloc_object_tlab
 
-GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT(_region_tlab, RegionTLAB)
+// A hand-written override for GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT(_region_tlab, RegionTLAB).
+DEFINE_FUNCTION art_quick_alloc_object_region_tlab
+    // Fast path region tlab allocation.
+    // RDI: uint32_t type_idx, RSI: ArtMethod*
+    // RDX, RCX, R8, R9: free. RAX: return val.
+#if !defined(USE_READ_BARRIER)
+    int3
+    int3
+#endif
+    // Might need a special macro since rsi and edx is 32b/64b mismatched.
+    movq ART_METHOD_DEX_CACHE_TYPES_OFFSET_64(%rsi), %rdx  // Load dex cache resolved types array
+    // Might need to break down into multiple instructions to get the base address in a register.
+                                                               // Load the class
+    movl 0(%rdx, %rdi, COMPRESSED_REFERENCE_SIZE), %edx
+    cmpl LITERAL(0), %gs:THREAD_IS_GC_MARKING_OFFSET
+    jne .Lart_quick_alloc_object_region_tlab_class_load_read_barrier_slow_path
+.Lart_quick_alloc_object_region_tlab_class_load_read_barrier_slow_path_exit:
+    ALLOC_OBJECT_TLAB_FAST_PATH .Lart_quick_alloc_object_region_tlab_slow_path
+.Lart_quick_alloc_object_region_tlab_class_load_read_barrier_slow_path:
+    // The read barrier slow path. Mark the class.
+    PUSH rdi
+    PUSH rsi
+    // Outgoing argument set up
+    movq %rdx, %rdi                                            // Pass the class as the first param.
+    call SYMBOL(artReadBarrierMark)                            // cxx_name(mirror::Object* obj)
+    movq %rax, %rdx
+    POP rsi
+    POP rdi
+    jmp .Lart_quick_alloc_object_region_tlab_class_load_read_barrier_slow_path_exit
+.Lart_quick_alloc_object_region_tlab_slow_path:
+    ALLOC_OBJECT_TLAB_SLOW_PATH artAllocObjectFromCodeRegionTLAB
+END_FUNCTION art_quick_alloc_object_region_tlab
 
 ONE_ARG_DOWNCALL art_quick_resolve_string, artResolveStringFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER
 ONE_ARG_DOWNCALL art_quick_initialize_static_storage, artInitializeStaticStorageFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER
@@ -994,6 +1044,14 @@ DEFINE_FUNCTION art_quick_lock_object
     RETURN_IF_EAX_ZERO
 END_FUNCTION art_quick_lock_object
 
+DEFINE_FUNCTION art_quick_lock_object_no_inline
+    SETUP_REFS_ONLY_CALLEE_SAVE_FRAME
+    movq %gs:THREAD_SELF_OFFSET, %rsi     // pass Thread::Current()
+    call SYMBOL(artLockObjectFromCode)    // artLockObjectFromCode(object, Thread*)
+    RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME   // restore frame up to return address
+    RETURN_IF_EAX_ZERO
+END_FUNCTION art_quick_lock_object_no_inline
+
 DEFINE_FUNCTION art_quick_unlock_object
     testl %edi, %edi                      // null check object/edi
     jz   .Lslow_unlock
@@ -1037,6 +1095,14 @@ DEFINE_FUNCTION art_quick_unlock_object
     RETURN_IF_EAX_ZERO
 END_FUNCTION art_quick_unlock_object
 
+DEFINE_FUNCTION art_quick_unlock_object_no_inline
+    SETUP_REFS_ONLY_CALLEE_SAVE_FRAME
+    movq %gs:THREAD_SELF_OFFSET, %rsi     // pass Thread::Current()
+    call SYMBOL(artUnlockObjectFromCode)  // artUnlockObjectFromCode(object, Thread*)
+    RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME   // restore frame up to return address
+    RETURN_IF_EAX_ZERO
+END_FUNCTION art_quick_unlock_object_no_inline
+
 DEFINE_FUNCTION art_quick_check_cast
     PUSH rdi                          // Save args for exc
     PUSH rsi
diff --git a/runtime/art_method-inl.h b/runtime/art_method-inl.h
index 6449efad7..7647ad6e5 100644
--- a/runtime/art_method-inl.h
+++ b/runtime/art_method-inl.h
@@ -456,13 +456,18 @@ void ArtMethod::VisitRoots(RootVisitorType& visitor, size_t pointer_size) {
       interface_method->VisitRoots(visitor, pointer_size);
     }
     visitor.VisitRoot(declaring_class_.AddressWithoutBarrier());
-    // Runtime methods and native methods use the same field as the profiling info for
-    // storing their own data (jni entrypoint for native methods, and ImtConflictTable for
-    // some runtime methods).
-    if (!IsNative() && !IsRuntimeMethod()) {
-      ProfilingInfo* profiling_info = GetProfilingInfo(pointer_size);
-      if (profiling_info != nullptr) {
-        profiling_info->VisitRoots(visitor);
+    // We know we don't have profiling information if the class hasn't been verified. Note
+    // that this check also ensures the IsNative call can be made, as IsNative expects a fully
+    // created class (and not a retired one).
+    if (klass->IsVerified()) {
+      // Runtime methods and native methods use the same field as the profiling info for
+      // storing their own data (jni entrypoint for native methods, and ImtConflictTable for
+      // some runtime methods).
+      if (!IsNative() && !IsRuntimeMethod()) {
+        ProfilingInfo* profiling_info = GetProfilingInfo(pointer_size);
+        if (profiling_info != nullptr) {
+          profiling_info->VisitRoots(visitor);
+        }
       }
     }
   }
diff --git a/runtime/art_method.cc b/runtime/art_method.cc
index 34d19d151..1790df6be 100644
--- a/runtime/art_method.cc
+++ b/runtime/art_method.cc
@@ -253,14 +253,17 @@ void ArtMethod::Invoke(Thread* self, uint32_t* args, uint32_t args_size, JValue*
   Runtime* runtime = Runtime::Current();
   // Call the invoke stub, passing everything as arguments.
   // If the runtime is not yet started or it is required by the debugger, then perform the
-  // Invocation by the interpreter.
+  // Invocation by the interpreter, explicitly forcing interpretation over JIT to prevent
+  // cycling around the various JIT/Interpreter methods that handle method invocation.
   if (UNLIKELY(!runtime->IsStarted() || Dbg::IsForcedInterpreterNeededForCalling(self, this))) {
     if (IsStatic()) {
-      art::interpreter::EnterInterpreterFromInvoke(self, this, nullptr, args, result);
+      art::interpreter::EnterInterpreterFromInvoke(
+          self, this, nullptr, args, result, /*stay_in_interpreter*/ true);
     } else {
       mirror::Object* receiver =
           reinterpret_cast<StackReference<mirror::Object>*>(&args[0])->AsMirrorPtr();
-      art::interpreter::EnterInterpreterFromInvoke(self, this, receiver, args + 1, result);
+      art::interpreter::EnterInterpreterFromInvoke(
+          self, this, receiver, args + 1, result, /*stay_in_interpreter*/ true);
     }
   } else {
     DCHECK_EQ(runtime->GetClassLinker()->GetImagePointerSize(), sizeof(void*));
@@ -276,7 +279,7 @@ void ArtMethod::Invoke(Thread* self, uint32_t* args, uint32_t args_size, JValue*
 
       // Ensure that we won't be accidentally calling quick compiled code when -Xint.
       if (kIsDebugBuild && runtime->GetInstrumentation()->IsForcedInterpretOnly()) {
-        CHECK(!runtime->UseJit());
+        CHECK(!runtime->UseJitCompilation());
         const void* oat_quick_code = runtime->GetClassLinker()->GetOatMethodQuickCodeFor(this);
         CHECK(oat_quick_code == nullptr || oat_quick_code != GetEntryPointFromQuickCompiledCode())
             << "Don't call compiled code when -Xint " << PrettyMethod(this);
@@ -481,7 +484,7 @@ void ArtMethod::CopyFrom(ArtMethod* src, size_t image_pointer_size) {
   // to the JIT code, but this would require taking the JIT code cache lock to notify
   // it, which we do not want at this level.
   Runtime* runtime = Runtime::Current();
-  if (runtime->GetJit() != nullptr) {
+  if (runtime->UseJitCompilation()) {
     if (runtime->GetJit()->GetCodeCache()->ContainsPc(GetEntryPointFromQuickCompiledCode())) {
       SetEntryPointFromQuickCompiledCodePtrSize(GetQuickToInterpreterBridge(), image_pointer_size);
     }
diff --git a/runtime/art_method.h b/runtime/art_method.h
index 08f02852e..a012a5a9c 100644
--- a/runtime/art_method.h
+++ b/runtime/art_method.h
@@ -41,6 +41,7 @@ class ShadowFrame;
 namespace mirror {
 class Array;
 class Class;
+class IfTable;
 class PointerArray;
 }  // namespace mirror
 
@@ -50,66 +51,151 @@ class PointerArray;
 // with the last entry being null to make an assembly implementation of a lookup
 // faster.
 class ImtConflictTable {
+  enum MethodIndex {
+    kMethodInterface,
+    kMethodImplementation,
+    kMethodCount,  // Number of elements in enum.
+  };
+
  public:
   // Build a new table copying `other` and adding the new entry formed of
   // the pair { `interface_method`, `implementation_method` }
   ImtConflictTable(ImtConflictTable* other,
                    ArtMethod* interface_method,
-                   ArtMethod* implementation_method) {
-    size_t index = 0;
-    while (other->entries_[index].interface_method != nullptr) {
-      entries_[index] = other->entries_[index];
-      index++;
+                   ArtMethod* implementation_method,
+                   size_t pointer_size) {
+    const size_t count = other->NumEntries(pointer_size);
+    for (size_t i = 0; i < count; ++i) {
+      SetInterfaceMethod(i, pointer_size, other->GetInterfaceMethod(i, pointer_size));
+      SetImplementationMethod(i, pointer_size, other->GetImplementationMethod(i, pointer_size));
     }
-    entries_[index].interface_method = interface_method;
-    entries_[index].implementation_method = implementation_method;
+    SetInterfaceMethod(count, pointer_size, interface_method);
+    SetImplementationMethod(count, pointer_size, implementation_method);
     // Add the null marker.
-    entries_[index + 1].interface_method = nullptr;
-    entries_[index + 1].implementation_method = nullptr;
+    SetInterfaceMethod(count + 1, pointer_size, nullptr);
+    SetImplementationMethod(count + 1, pointer_size, nullptr);
+  }
+
+  // num_entries excludes the header.
+  ImtConflictTable(size_t num_entries, size_t pointer_size) {
+    SetInterfaceMethod(num_entries, pointer_size, nullptr);
+    SetImplementationMethod(num_entries, pointer_size, nullptr);
+  }
+
+  // Set an entry at an index.
+  void SetInterfaceMethod(size_t index, size_t pointer_size, ArtMethod* method) {
+    SetMethod(index * kMethodCount + kMethodInterface, pointer_size, method);
+  }
+
+  void SetImplementationMethod(size_t index, size_t pointer_size, ArtMethod* method) {
+    SetMethod(index * kMethodCount + kMethodImplementation, pointer_size, method);
+  }
+
+  ArtMethod* GetInterfaceMethod(size_t index, size_t pointer_size) const {
+    return GetMethod(index * kMethodCount + kMethodInterface, pointer_size);
+  }
+
+  ArtMethod* GetImplementationMethod(size_t index, size_t pointer_size) const {
+    return GetMethod(index * kMethodCount + kMethodImplementation, pointer_size);
+  }
+
+  // Visit all of the entries.
+  // NO_THREAD_SAFETY_ANALYSIS for calling with held locks. Visitor is passed a pair of ArtMethod*
+  // and also returns one. The order is <interface, implementation>.
+  template<typename Visitor>
+  void Visit(const Visitor& visitor, size_t pointer_size) NO_THREAD_SAFETY_ANALYSIS {
+    uint32_t table_index = 0;
+    for (;;) {
+      ArtMethod* interface_method = GetInterfaceMethod(table_index, pointer_size);
+      if (interface_method == nullptr) {
+        break;
+      }
+      ArtMethod* implementation_method = GetImplementationMethod(table_index, pointer_size);
+      auto input = std::make_pair(interface_method, implementation_method);
+      std::pair<ArtMethod*, ArtMethod*> updated = visitor(input);
+      if (input.first != updated.first) {
+        SetInterfaceMethod(table_index, pointer_size, updated.first);
+      }
+      if (input.second != updated.second) {
+        SetImplementationMethod(table_index, pointer_size, updated.second);
+      }
+      ++table_index;
+    }
   }
 
   // Lookup the implementation ArtMethod associated to `interface_method`. Return null
   // if not found.
-  ArtMethod* Lookup(ArtMethod* interface_method) const {
+  ArtMethod* Lookup(ArtMethod* interface_method, size_t pointer_size) const {
     uint32_t table_index = 0;
-    ArtMethod* current_interface_method;
-    while ((current_interface_method = entries_[table_index].interface_method) != nullptr) {
+    for (;;) {
+      ArtMethod* current_interface_method = GetInterfaceMethod(table_index, pointer_size);
+      if (current_interface_method == nullptr) {
+        break;
+      }
       if (current_interface_method == interface_method) {
-        return entries_[table_index].implementation_method;
+        return GetImplementationMethod(table_index, pointer_size);
       }
-      table_index++;
+      ++table_index;
     }
     return nullptr;
   }
 
-  // Compute the size in bytes taken by this table.
-  size_t ComputeSize() const {
+  // Compute the number of entries in this table.
+  size_t NumEntries(size_t pointer_size) const {
     uint32_t table_index = 0;
-    size_t total_size = 0;
-    while ((entries_[table_index].interface_method) != nullptr) {
-      total_size += sizeof(Entry);
-      table_index++;
+    while (GetInterfaceMethod(table_index, pointer_size) != nullptr) {
+      ++table_index;
     }
+    return table_index;
+  }
+
+  // Compute the size in bytes taken by this table.
+  size_t ComputeSize(size_t pointer_size) const {
     // Add the end marker.
-    return total_size + sizeof(Entry);
+    return ComputeSize(NumEntries(pointer_size), pointer_size);
   }
 
   // Compute the size in bytes needed for copying the given `table` and add
   // one more entry.
-  static size_t ComputeSizeWithOneMoreEntry(ImtConflictTable* table) {
-    return table->ComputeSize() + sizeof(Entry);
+  static size_t ComputeSizeWithOneMoreEntry(ImtConflictTable* table, size_t pointer_size) {
+    return table->ComputeSize(pointer_size) + EntrySize(pointer_size);
   }
 
-  struct Entry {
-    ArtMethod* interface_method;
-    ArtMethod* implementation_method;
-  };
+  // Compute size with a fixed number of entries.
+  static size_t ComputeSize(size_t num_entries, size_t pointer_size) {
+    return (num_entries + 1) * EntrySize(pointer_size);  // Add one for null terminator.
+  }
+
+  static size_t EntrySize(size_t pointer_size) {
+    return pointer_size * static_cast<size_t>(kMethodCount);
+  }
 
  private:
+  ArtMethod* GetMethod(size_t index, size_t pointer_size) const {
+    if (pointer_size == 8) {
+      return reinterpret_cast<ArtMethod*>(static_cast<uintptr_t>(data64_[index]));
+    } else {
+      DCHECK_EQ(pointer_size, 4u);
+      return reinterpret_cast<ArtMethod*>(static_cast<uintptr_t>(data32_[index]));
+    }
+  }
+
+  void SetMethod(size_t index, size_t pointer_size, ArtMethod* method) {
+    if (pointer_size == 8) {
+      data64_[index] = dchecked_integral_cast<uint64_t>(reinterpret_cast<uintptr_t>(method));
+    } else {
+      DCHECK_EQ(pointer_size, 4u);
+      data32_[index] = dchecked_integral_cast<uint32_t>(reinterpret_cast<uintptr_t>(method));
+    }
+  }
+
   // Array of entries that the assembly stubs will iterate over. Note that this is
   // not fixed size, and we allocate data prior to calling the constructor
   // of ImtConflictTable.
-  Entry entries_[0];
+  union {
+    uint32_t data32_[0];
+    uint64_t data64_[0];
+  };
 
   DISALLOW_COPY_AND_ASSIGN(ImtConflictTable);
 };
@@ -265,6 +351,12 @@ class ArtMethod FINAL {
     SetAccessFlags(GetAccessFlags() | kAccSkipAccessChecks);
   }
 
+  // Should this method be run in the interpreter and count locks (e.g., failed structured-
+  // locking verification)?
+  bool MustCountLocks() {
+    return (GetAccessFlags() & kAccMustCountLocks) != 0;
+  }
+
   // Returns true if this method could be overridden by a default method.
   bool IsOverridableByDefaultMethod() SHARED_REQUIRES(Locks::mutator_lock_);
 
@@ -351,7 +443,6 @@ class ArtMethod FINAL {
 
   // Find the method that this method overrides.
   ArtMethod* FindOverriddenMethod(size_t pointer_size)
-      REQUIRES(Roles::uninterruptible_)
       SHARED_REQUIRES(Locks::mutator_lock_);
 
   // Find the method index for this method within other_dexfile. If this method isn't present then
@@ -417,8 +508,8 @@ class ArtMethod FINAL {
     return reinterpret_cast<ImtConflictTable*>(GetEntryPointFromJniPtrSize(pointer_size));
   }
 
-  ALWAYS_INLINE void SetImtConflictTable(ImtConflictTable* table) {
-    SetEntryPointFromJniPtrSize(table, sizeof(void*));
+  ALWAYS_INLINE void SetImtConflictTable(ImtConflictTable* table, size_t pointer_size) {
+    SetEntryPointFromJniPtrSize(table, pointer_size);
   }
 
   ALWAYS_INLINE void SetProfilingInfo(ProfilingInfo* info) {
diff --git a/runtime/base/histogram.h b/runtime/base/histogram.h
index bcb7b3b76..0e3bc8e1b 100644
--- a/runtime/base/histogram.h
+++ b/runtime/base/histogram.h
@@ -85,6 +85,10 @@ template <class Value> class Histogram {
     return max_value_added_;
   }
 
+  Value BucketWidth() const {
+    return bucket_width_;
+  }
+
   const std::string& Name() const {
     return name_;
   }
diff --git a/runtime/base/logging.h b/runtime/base/logging.h
index 97280c3a0..3b5b8b54a 100644
--- a/runtime/base/logging.h
+++ b/runtime/base/logging.h
@@ -56,6 +56,7 @@ struct LogVerbosity {
   bool threads;
   bool verifier;
   bool image;
+  bool systrace_lock_logging;  // Enabled with "-verbose:sys-locks".
 };
 
 // Global log verbosity setting, initialized by InitLogging.
diff --git a/runtime/check_jni.cc b/runtime/check_jni.cc
index beabce36f..639f913e8 100644
--- a/runtime/check_jni.cc
+++ b/runtime/check_jni.cc
@@ -1176,14 +1176,16 @@ class ScopedCheck {
       return false;
     }
 
-    // Get the *correct* JNIEnv by going through our TLS pointer.
+    // Get the current thread's JNIEnv by going through our TLS pointer.
     JNIEnvExt* threadEnv = self->GetJniEnv();
 
     // Verify that the current thread is (a) attached and (b) associated with
     // this particular instance of JNIEnv.
     if (env != threadEnv) {
+      // Get the thread owning the JNIEnv that's being used.
+      Thread* envThread = reinterpret_cast<JNIEnvExt*>(env)->self;
       AbortF("thread %s using JNIEnv* from thread %s",
-             ToStr<Thread>(*self).c_str(), ToStr<Thread>(*self).c_str());
+             ToStr<Thread>(*self).c_str(), ToStr<Thread>(*envThread).c_str());
       return false;
     }
 
diff --git a/runtime/class_linker.cc b/runtime/class_linker.cc
index fa0107af8..8fcb6b25f 100644
--- a/runtime/class_linker.cc
+++ b/runtime/class_linker.cc
@@ -687,6 +687,9 @@ bool ClassLinker::InitWithoutImage(std::vector<std::unique_ptr<const DexFile>> b
     self->AssertNoPendingException();
   }
 
+  // Create conflict tables that depend on the class linker.
+  runtime->FixupConflictTables();
+
   FinishInit(self);
 
   VLOG(startup) << "ClassLinker::InitFromCompiler exiting";
@@ -773,9 +776,13 @@ static void SanityCheckArtMethod(ArtMethod* m,
     bool contains = false;
     for (gc::space::ImageSpace* space : spaces) {
       auto& header = space->GetImageHeader();
-      auto& methods = header.GetMethodsSection();
-      auto offset = reinterpret_cast<uint8_t*>(m) - space->Begin();
-      contains |= methods.Contains(offset);
+      size_t offset = reinterpret_cast<uint8_t*>(m) - space->Begin();
+
+      const ImageSection& methods = header.GetMethodsSection();
+      contains = contains || methods.Contains(offset);
+
+      const ImageSection& runtime_methods = header.GetRuntimeMethodsSection();
+      contains = contains || runtime_methods.Contains(offset);
     }
     CHECK(contains) << m << " not found";
   }
@@ -1056,9 +1063,8 @@ bool ClassLinker::InitFromBootImage(std::string* error_msg) {
   return true;
 }
 
-static bool IsBootClassLoader(ScopedObjectAccessAlreadyRunnable& soa,
-                              mirror::ClassLoader* class_loader)
-    SHARED_REQUIRES(Locks::mutator_lock_) {
+bool ClassLinker::IsBootClassLoader(ScopedObjectAccessAlreadyRunnable& soa,
+                                    mirror::ClassLoader* class_loader) {
   return class_loader == nullptr ||
       class_loader->GetClass() ==
           soa.Decode<mirror::Class*>(WellKnownClasses::java_lang_BootClassLoader);
@@ -1099,7 +1105,7 @@ static bool FlattenPathClassLoader(mirror::ClassLoader* class_loader,
       soa.DecodeField(WellKnownClasses::dalvik_system_DexPathList_dexElements);
   CHECK(dex_path_list_field != nullptr);
   CHECK(dex_elements_field != nullptr);
-  while (!IsBootClassLoader(soa, class_loader)) {
+  while (!ClassLinker::IsBootClassLoader(soa, class_loader)) {
     if (class_loader->GetClass() !=
         soa.Decode<mirror::Class*>(WellKnownClasses::dalvik_system_PathClassLoader)) {
       *error_msg = StringPrintf("Unknown class loader type %s", PrettyTypeOf(class_loader).c_str());
@@ -1438,20 +1444,14 @@ bool ClassLinker::UpdateAppImageClassLoadersAndDexCaches(
   if (*out_forward_dex_cache_array) {
     ScopedTrace timing("Fixup ArtMethod dex cache arrays");
     FixupArtMethodArrayVisitor visitor(header);
-    header.GetImageSection(ImageHeader::kSectionArtMethods).VisitPackedArtMethods(
-        &visitor,
-        space->Begin(),
-        sizeof(void*));
+    header.VisitPackedArtMethods(&visitor, space->Begin(), sizeof(void*));
     Runtime::Current()->GetHeap()->WriteBarrierEveryFieldOf(class_loader.Get());
   }
   if (kVerifyArtMethodDeclaringClasses) {
     ScopedTrace timing("Verify declaring classes");
     ReaderMutexLock rmu(self, *Locks::heap_bitmap_lock_);
     VerifyDeclaringClassVisitor visitor;
-    header.GetImageSection(ImageHeader::kSectionArtMethods).VisitPackedArtMethods(
-        &visitor,
-        space->Begin(),
-        sizeof(void*));
+    header.VisitPackedArtMethods(&visitor, space->Begin(), sizeof(void*));
   }
   return true;
 }
@@ -1729,9 +1729,8 @@ bool ClassLinker::AddImageSpace(
 
   // Set entry point to interpreter if in InterpretOnly mode.
   if (!runtime->IsAotCompiler() && runtime->GetInstrumentation()->InterpretOnly()) {
-    const ImageSection& methods = header.GetMethodsSection();
     SetInterpreterEntrypointArtMethodVisitor visitor(image_pointer_size_);
-    methods.VisitPackedArtMethods(&visitor, space->Begin(), image_pointer_size_);
+    header.VisitPackedArtMethods(&visitor, space->Begin(), image_pointer_size_);
   }
 
   ClassTable* class_table = nullptr;
@@ -1800,10 +1799,7 @@ bool ClassLinker::AddImageSpace(
     // This verification needs to happen after the classes have been added to the class loader.
     // Since it ensures classes are in the class table.
     VerifyClassInTableArtMethodVisitor visitor2(class_table);
-    header.GetImageSection(ImageHeader::kSectionArtMethods).VisitPackedArtMethods(
-        &visitor2,
-        space->Begin(),
-        sizeof(void*));
+    header.VisitPackedArtMethods(&visitor2, space->Begin(), sizeof(void*));
   }
   VLOG(class_linker) << "Adding image space took " << PrettyDuration(NanoTime() - start_time);
   return true;
@@ -2035,6 +2031,7 @@ void ClassLinker::DeleteClassLoader(Thread* self, const ClassLoaderData& data) {
   Runtime* const runtime = Runtime::Current();
   JavaVMExt* const vm = runtime->GetJavaVM();
   vm->DeleteWeakGlobalRef(self, data.weak_root);
+  // Notify the JIT that we need to remove the methods and/or profiling info.
   if (runtime->GetJit() != nullptr) {
     jit::JitCodeCache* code_cache = runtime->GetJit()->GetCodeCache();
     if (code_cache != nullptr) {
@@ -2752,7 +2749,7 @@ bool ClassLinker::ShouldUseInterpreterEntrypoint(ArtMethod* method, const void*
   }
 
   if (runtime->IsNativeDebuggable()) {
-    DCHECK(runtime->UseJit() && runtime->GetJit()->JitAtFirstUse());
+    DCHECK(runtime->UseJitCompilation() && runtime->GetJit()->JitAtFirstUse());
     // If we are doing native debugging, ignore application's AOT code,
     // since we want to JIT it with extra stackmaps for native debugging.
     // On the other hand, keep all AOT code from the boot image, since the
@@ -5319,6 +5316,19 @@ bool ClassLinker::LoadSuperAndInterfaces(Handle<mirror::Class> klass, const DexF
   const DexFile::ClassDef& class_def = dex_file.GetClassDef(klass->GetDexClassDefIndex());
   uint16_t super_class_idx = class_def.superclass_idx_;
   if (super_class_idx != DexFile::kDexNoIndex16) {
+    // Check that a class does not inherit from itself directly.
+    //
+    // TODO: This is a cheap check to detect the straightforward case
+    // of a class extending itself (b/28685551), but we should do a
+    // proper cycle detection on loaded classes, to detect all cases
+    // of class circularity errors (b/28830038).
+    if (super_class_idx == class_def.class_idx_) {
+      ThrowClassCircularityError(klass.Get(),
+                                 "Class %s extends itself",
+                                 PrettyDescriptor(klass.Get()).c_str());
+      return false;
+    }
+
     mirror::Class* super_class = ResolveType(dex_file, super_class_idx, klass.Get());
     if (super_class == nullptr) {
       DCHECK(Thread::Current()->IsExceptionPending());
@@ -5961,16 +5971,49 @@ ClassLinker::DefaultMethodSearchResult ClassLinker::FindDefaultMethodImplementat
   }
 }
 
-// Sets imt_ref appropriately for LinkInterfaceMethods.
-// If there is no method in the imt location of imt_ref it will store the given method there.
-// Otherwise it will set the conflict method which will figure out which method to use during
-// runtime.
-static void SetIMTRef(ArtMethod* unimplemented_method,
-                      ArtMethod* imt_conflict_method,
-                      size_t image_pointer_size,
-                      ArtMethod* current_method,
-                      /*out*/ArtMethod** imt_ref)
-    SHARED_REQUIRES(Locks::mutator_lock_) {
+ArtMethod* ClassLinker::AddMethodToConflictTable(mirror::Class* klass,
+                                                 ArtMethod* conflict_method,
+                                                 ArtMethod* interface_method,
+                                                 ArtMethod* method,
+                                                 bool force_new_conflict_method) {
+  ImtConflictTable* current_table = conflict_method->GetImtConflictTable(sizeof(void*));
+  Runtime* const runtime = Runtime::Current();
+  LinearAlloc* linear_alloc = GetAllocatorForClassLoader(klass->GetClassLoader());
+  bool new_entry = conflict_method == runtime->GetImtConflictMethod() || force_new_conflict_method;
+
+  // Create a new entry if the existing one is the shared conflict method.
+  ArtMethod* new_conflict_method = new_entry
+      ? runtime->CreateImtConflictMethod(linear_alloc)
+      : conflict_method;
+
+  // Allocate a new table. Note that we will leak this table at the next conflict,
+  // but that's a tradeoff compared to making the table fixed size.
+  void* data = linear_alloc->Alloc(
+      Thread::Current(), ImtConflictTable::ComputeSizeWithOneMoreEntry(current_table,
+                                                                       image_pointer_size_));
+  if (data == nullptr) {
+    LOG(ERROR) << "Failed to allocate conflict table";
+    return conflict_method;
+  }
+  ImtConflictTable* new_table = new (data) ImtConflictTable(current_table,
+                                                            interface_method,
+                                                            method,
+                                                            image_pointer_size_);
+
+  // Do a fence to ensure threads see the data in the table before it is assigned
+  // to the conflict method.
+  // Note that there is a race in the presence of multiple threads and we may leak
+  // memory from the LinearAlloc, but that's a tradeoff compared to using
+  // atomic operations.
+  QuasiAtomic::ThreadFenceRelease();
+  new_conflict_method->SetImtConflictTable(new_table, image_pointer_size_);
+  return new_conflict_method;
+}
+
+void ClassLinker::SetIMTRef(ArtMethod* unimplemented_method,
+                            ArtMethod* imt_conflict_method,
+                            ArtMethod* current_method,
+                            /*out*/ArtMethod** imt_ref) {
   // Place method in imt if entry is empty, place conflict otherwise.
   if (*imt_ref == unimplemented_method) {
     *imt_ref = current_method;
@@ -5980,9 +6023,9 @@ static void SetIMTRef(ArtMethod* unimplemented_method,
     // Note that we have checked IsRuntimeMethod, as there may be multiple different
     // conflict methods.
     MethodNameAndSignatureComparator imt_comparator(
-        (*imt_ref)->GetInterfaceMethodIfProxy(image_pointer_size));
+        (*imt_ref)->GetInterfaceMethodIfProxy(image_pointer_size_));
     if (imt_comparator.HasSameNameAndSignature(
-          current_method->GetInterfaceMethodIfProxy(image_pointer_size))) {
+          current_method->GetInterfaceMethodIfProxy(image_pointer_size_))) {
       *imt_ref = current_method;
     } else {
       *imt_ref = imt_conflict_method;
@@ -5995,6 +6038,151 @@ static void SetIMTRef(ArtMethod* unimplemented_method,
   }
 }
 
+void ClassLinker::FillIMTAndConflictTables(mirror::Class* klass) {
+  DCHECK(klass->ShouldHaveEmbeddedImtAndVTable()) << PrettyClass(klass);
+  DCHECK(!klass->IsTemp()) << PrettyClass(klass);
+  ArtMethod* imt[mirror::Class::kImtSize];
+  Runtime* const runtime = Runtime::Current();
+  ArtMethod* const unimplemented_method = runtime->GetImtUnimplementedMethod();
+  ArtMethod* const conflict_method = runtime->GetImtConflictMethod();
+  std::fill_n(imt, arraysize(imt), unimplemented_method);
+  if (klass->GetIfTable() != nullptr) {
+    FillIMTFromIfTable(klass->GetIfTable(),
+                       unimplemented_method,
+                       conflict_method,
+                       klass,
+                       true,
+                       false,
+                       &imt[0]);
+  }
+  for (size_t i = 0; i < mirror::Class::kImtSize; ++i) {
+    klass->SetEmbeddedImTableEntry(i, imt[i], image_pointer_size_);
+  }
+}
+
+static inline uint32_t GetIMTIndex(ArtMethod* interface_method)
+    SHARED_REQUIRES(Locks::mutator_lock_) {
+  return interface_method->GetDexMethodIndex() % mirror::Class::kImtSize;
+}
+
+ImtConflictTable* ClassLinker::CreateImtConflictTable(size_t count,
+                                                      LinearAlloc* linear_alloc,
+                                                      size_t image_pointer_size) {
+  void* data = linear_alloc->Alloc(Thread::Current(),
+                                   ImtConflictTable::ComputeSize(count,
+                                                                 image_pointer_size));
+  return (data != nullptr) ? new (data) ImtConflictTable(count, image_pointer_size) : nullptr;
+}
+
+ImtConflictTable* ClassLinker::CreateImtConflictTable(size_t count, LinearAlloc* linear_alloc) {
+  return CreateImtConflictTable(count, linear_alloc, image_pointer_size_);
+}
+
+void ClassLinker::FillIMTFromIfTable(mirror::IfTable* if_table,
+                                     ArtMethod* unimplemented_method,
+                                     ArtMethod* imt_conflict_method,
+                                     mirror::Class* klass,
+                                     bool create_conflict_tables,
+                                     bool ignore_copied_methods,
+                                     ArtMethod** imt) {
+  uint32_t conflict_counts[mirror::Class::kImtSize] = {};
+  for (size_t i = 0, length = if_table->Count(); i < length; ++i) {
+    mirror::Class* interface = if_table->GetInterface(i);
+    const size_t num_virtuals = interface->NumVirtualMethods();
+    const size_t method_array_count = if_table->GetMethodArrayCount(i);
+    // Virtual methods can be larger than the if table methods if there are default methods.
+    DCHECK_GE(num_virtuals, method_array_count);
+    if (kIsDebugBuild) {
+      if (klass->IsInterface()) {
+        DCHECK_EQ(method_array_count, 0u);
+      } else {
+        DCHECK_EQ(interface->NumDeclaredVirtualMethods(), method_array_count);
+      }
+    }
+    if (method_array_count == 0) {
+      continue;
+    }
+    auto* method_array = if_table->GetMethodArray(i);
+    for (size_t j = 0; j < method_array_count; ++j) {
+      ArtMethod* implementation_method =
+          method_array->GetElementPtrSize<ArtMethod*>(j, image_pointer_size_);
+      if (ignore_copied_methods && implementation_method->IsCopied()) {
+        continue;
+      }
+      DCHECK(implementation_method != nullptr);
+      // Miranda methods cannot be used to implement an interface method, but they are safe to put
+      // in the IMT since their entrypoint is the interface trampoline. If we put any copied methods
+      // or interface methods in the IMT here they will not create extra conflicts since we compare
+      // names and signatures in SetIMTRef.
+      ArtMethod* interface_method = interface->GetVirtualMethod(j, image_pointer_size_);
+      const uint32_t imt_index = GetIMTIndex(interface_method);
+
+      // There is only any conflicts if all of the interface methods for an IMT slot don't have
+      // the same implementation method, keep track of this to avoid creating a conflict table in
+      // this case.
+
+      // Conflict table size for each IMT slot.
+      ++conflict_counts[imt_index];
+
+      SetIMTRef(unimplemented_method,
+                imt_conflict_method,
+                implementation_method,
+                /*out*/&imt[imt_index]);
+    }
+  }
+
+  if (create_conflict_tables) {
+    // Create the conflict tables.
+    LinearAlloc* linear_alloc = GetAllocatorForClassLoader(klass->GetClassLoader());
+    for (size_t i = 0; i < mirror::Class::kImtSize; ++i) {
+      size_t conflicts = conflict_counts[i];
+      if (imt[i] == imt_conflict_method) {
+        ImtConflictTable* new_table = CreateImtConflictTable(conflicts, linear_alloc);
+        if (new_table != nullptr) {
+          ArtMethod* new_conflict_method =
+              Runtime::Current()->CreateImtConflictMethod(linear_alloc);
+          new_conflict_method->SetImtConflictTable(new_table, image_pointer_size_);
+          imt[i] = new_conflict_method;
+        } else {
+          LOG(ERROR) << "Failed to allocate conflict table";
+          imt[i] = imt_conflict_method;
+        }
+      } else {
+        DCHECK_NE(imt[i], imt_conflict_method);
+      }
+    }
+
+    for (size_t i = 0, length = if_table->Count(); i < length; ++i) {
+      mirror::Class* interface = if_table->GetInterface(i);
+      const size_t method_array_count = if_table->GetMethodArrayCount(i);
+      // Virtual methods can be larger than the if table methods if there are default methods.
+      if (method_array_count == 0) {
+        continue;
+      }
+      auto* method_array = if_table->GetMethodArray(i);
+      for (size_t j = 0; j < method_array_count; ++j) {
+        ArtMethod* implementation_method =
+            method_array->GetElementPtrSize<ArtMethod*>(j, image_pointer_size_);
+        if (ignore_copied_methods && implementation_method->IsCopied()) {
+          continue;
+        }
+        DCHECK(implementation_method != nullptr);
+        ArtMethod* interface_method = interface->GetVirtualMethod(j, image_pointer_size_);
+        const uint32_t imt_index = GetIMTIndex(interface_method);
+        if (!imt[imt_index]->IsRuntimeMethod() ||
+            imt[imt_index] == unimplemented_method ||
+            imt[imt_index] == imt_conflict_method) {
+          continue;
+        }
+        ImtConflictTable* table = imt[imt_index]->GetImtConflictTable(image_pointer_size_);
+        const size_t num_entries = table->NumEntries(image_pointer_size_);
+        table->SetInterfaceMethod(num_entries, image_pointer_size_, interface_method);
+        table->SetImplementationMethod(num_entries, image_pointer_size_, implementation_method);
+      }
+    }
+  }
+}
+
 // Simple helper function that checks that no subtypes of 'val' are contained within the 'classes'
 // set.
 static bool NotSubinterfaceOfAny(const std::unordered_set<mirror::Class*>& classes,
@@ -6230,48 +6418,28 @@ static void SanityCheckVTable(Handle<mirror::Class> klass, uint32_t pointer_size
   }
 }
 
-static void FillImtFromSuperClass(Handle<mirror::Class> klass,
-                                  Handle<mirror::IfTable> iftable,
-                                  ArtMethod* unimplemented_method,
-                                  ArtMethod* imt_conflict_method,
-                                  ArtMethod** out_imt,
-                                  size_t pointer_size) SHARED_REQUIRES(Locks::mutator_lock_) {
+void ClassLinker::FillImtFromSuperClass(Handle<mirror::Class> klass,
+                                        ArtMethod* unimplemented_method,
+                                        ArtMethod* imt_conflict_method,
+                                        ArtMethod** imt) {
   DCHECK(klass->HasSuperClass());
   mirror::Class* super_class = klass->GetSuperClass();
   if (super_class->ShouldHaveEmbeddedImtAndVTable()) {
     for (size_t i = 0; i < mirror::Class::kImtSize; ++i) {
-      out_imt[i] = super_class->GetEmbeddedImTableEntry(i, pointer_size);
+      imt[i] = super_class->GetEmbeddedImTableEntry(i, image_pointer_size_);
     }
   } else {
     // No imt in the super class, need to reconstruct from the iftable.
     mirror::IfTable* if_table = super_class->GetIfTable();
-    const size_t length = super_class->GetIfTableCount();
-    for (size_t i = 0; i < length; ++i) {
-      mirror::Class* interface = iftable->GetInterface(i);
-      const size_t num_virtuals = interface->NumDeclaredVirtualMethods();
-      const size_t method_array_count = if_table->GetMethodArrayCount(i);
-      DCHECK_EQ(num_virtuals, method_array_count);
-      if (method_array_count == 0) {
-        continue;
-      }
-      auto* method_array = if_table->GetMethodArray(i);
-      for (size_t j = 0; j < num_virtuals; ++j) {
-        auto method = method_array->GetElementPtrSize<ArtMethod*>(j, pointer_size);
-        DCHECK(method != nullptr) << PrettyClass(super_class);
-        // Miranda methods cannot be used to implement an interface method and defaults should be
-        // skipped in case we override it.
-        if (method->IsDefault() || method->IsMiranda()) {
-          continue;
-        }
-        ArtMethod* interface_method = interface->GetVirtualMethod(j, pointer_size);
-        uint32_t imt_index = interface_method->GetDexMethodIndex() % mirror::Class::kImtSize;
-        auto** imt_ref = &out_imt[imt_index];
-        if (*imt_ref == unimplemented_method) {
-          *imt_ref = method;
-        } else if (*imt_ref != imt_conflict_method) {
-          *imt_ref = imt_conflict_method;
-        }
-      }
+    if (if_table != nullptr) {
+      // Ignore copied methods since we will handle these in LinkInterfaceMethods.
+      FillIMTFromIfTable(if_table,
+                         unimplemented_method,
+                         imt_conflict_method,
+                         klass.Get(),
+                         /*create_conflict_table*/false,
+                         /*ignore_copied_methods*/true,
+                         /*out*/imt);
     }
   }
 }
@@ -6314,13 +6482,10 @@ bool ClassLinker::LinkInterfaceMethods(
   const bool extend_super_iftable = has_superclass;
   if (has_superclass && fill_tables) {
     FillImtFromSuperClass(klass,
-                          iftable,
                           unimplemented_method,
                           imt_conflict_method,
-                          out_imt,
-                          image_pointer_size_);
+                          out_imt);
   }
-
   // Allocate method arrays before since we don't want miss visiting miranda method roots due to
   // thread suspension.
   if (fill_tables) {
@@ -6404,7 +6569,7 @@ bool ClassLinker::LinkInterfaceMethods(
         auto* interface_method = iftable->GetInterface(i)->GetVirtualMethod(j, image_pointer_size_);
         MethodNameAndSignatureComparator interface_name_comparator(
             interface_method->GetInterfaceMethodIfProxy(image_pointer_size_));
-        uint32_t imt_index = interface_method->GetDexMethodIndex() % mirror::Class::kImtSize;
+        uint32_t imt_index = GetIMTIndex(interface_method);
         ArtMethod** imt_ptr = &out_imt[imt_index];
         // For each method listed in the interface's method list, find the
         // matching method in our class's method list.  We want to favor the
@@ -6449,7 +6614,6 @@ bool ClassLinker::LinkInterfaceMethods(
                 // Place method in imt if entry is empty, place conflict otherwise.
                 SetIMTRef(unimplemented_method,
                           imt_conflict_method,
-                          image_pointer_size_,
                           vtable_method,
                           /*out*/imt_ptr);
               }
@@ -6483,6 +6647,17 @@ bool ClassLinker::LinkInterfaceMethods(
             // The method is not overridable by a default method (i.e. it is directly implemented
             // in some class). Therefore move onto the next interface method.
             continue;
+          } else {
+            // If the super-classes method is override-able by a default method we need to keep
+            // track of it since though it is override-able it is not guaranteed to be 'overridden'.
+            // If it turns out not to be overridden and we did not keep track of it we might add it
+            // to the vtable twice, causing corruption (vtable entries having inconsistent and
+            // illegal states, incorrect vtable size, and incorrect or inconsistent iftable entries)
+            // in this class and any subclasses.
+            DCHECK(vtable_impl == nullptr || vtable_impl == supers_method)
+                << "vtable_impl was " << PrettyMethod(vtable_impl) << " and not 'nullptr' or "
+                << PrettyMethod(supers_method) << " as expected. IFTable appears to be corrupt!";
+            vtable_impl = supers_method;
           }
         }
         // If we haven't found it yet we should search through the interfaces for default methods.
@@ -6581,7 +6756,6 @@ bool ClassLinker::LinkInterfaceMethods(
             method_array->SetElementPtrSize(j, current_method, image_pointer_size_);
             SetIMTRef(unimplemented_method,
                       imt_conflict_method,
-                      image_pointer_size_,
                       current_method,
                       /*out*/imt_ptr);
           }
@@ -6808,6 +6982,7 @@ bool ClassLinker::LinkInterfaceMethods(
     }
     // Put some random garbage in old methods to help find stale pointers.
     if (methods != old_methods && old_methods != nullptr) {
+      WriterMutexLock mu(self, ClassTableForClassLoader(klass->GetClassLoader())->GetLock());
       memset(old_methods, 0xFEu, old_size);
     }
   } else {
@@ -7652,7 +7827,8 @@ const char* ClassLinker::GetClassRootDescriptor(ClassRoot class_root) {
   return descriptor;
 }
 
-jobject ClassLinker::CreatePathClassLoader(Thread* self, std::vector<const DexFile*>& dex_files) {
+jobject ClassLinker::CreatePathClassLoader(Thread* self,
+                                           const std::vector<const DexFile*>& dex_files) {
   // SOAAlreadyRunnable is protected, and we need something to add a global reference.
   // We could move the jobject to the callers, but all call-sites do this...
   ScopedObjectAccessUnchecked soa(self);
@@ -7835,6 +8011,7 @@ std::set<DexCacheResolvedClasses> ClassLinker::GetResolvedClasses(bool ignore_bo
     VLOG(class_linker) << "Collecting class profile for dex file " << location
                        << " types=" << num_types << " class_defs=" << num_class_defs;
     DexCacheResolvedClasses resolved_classes(dex_file->GetLocation(),
+                                             dex_file->GetBaseLocation(),
                                              dex_file->GetLocationChecksum());
     size_t num_resolved = 0;
     std::unordered_set<uint16_t> class_set;
diff --git a/runtime/class_linker.h b/runtime/class_linker.h
index 97a10367f..f6ce545a1 100644
--- a/runtime/class_linker.h
+++ b/runtime/class_linker.h
@@ -53,6 +53,7 @@ namespace mirror {
   class StackTraceElement;
 }  // namespace mirror
 
+class ImtConflictTable;
 template<class T> class Handle;
 template<class T> class MutableHandle;
 class InternTable;
@@ -254,7 +255,7 @@ class ClassLinker {
       SHARED_REQUIRES(Locks::mutator_lock_);
 
   // Resolve a Type with the given index from the DexFile, storing the
-  // result in the DexCache. The referrer is used to identity the
+  // result in the DexCache. The referrer is used to identify the
   // target DexCache and ClassLoader to use for resolution.
   mirror::Class* ResolveType(const DexFile& dex_file, uint16_t type_idx, mirror::Class* referrer)
       SHARED_REQUIRES(Locks::mutator_lock_)
@@ -559,7 +560,7 @@ class ClassLinker {
 
   // Creates a GlobalRef PathClassLoader that can be used to load classes from the given dex files.
   // Note: the objects are not completely set up. Do not use this outside of tests and the compiler.
-  jobject CreatePathClassLoader(Thread* self, std::vector<const DexFile*>& dex_files)
+  jobject CreatePathClassLoader(Thread* self, const std::vector<const DexFile*>& dex_files)
       SHARED_REQUIRES(Locks::mutator_lock_)
       REQUIRES(!dex_lock_);
 
@@ -610,6 +611,30 @@ class ClassLinker {
       const std::set<DexCacheResolvedClasses>& classes)
       REQUIRES(!dex_lock_);
 
+  static bool IsBootClassLoader(ScopedObjectAccessAlreadyRunnable& soa,
+                                mirror::ClassLoader* class_loader)
+      SHARED_REQUIRES(Locks::mutator_lock_);
+
+  ArtMethod* AddMethodToConflictTable(mirror::Class* klass,
+                                      ArtMethod* conflict_method,
+                                      ArtMethod* interface_method,
+                                      ArtMethod* method,
+                                      bool force_new_conflict_method)
+      SHARED_REQUIRES(Locks::mutator_lock_);
+
+  // Create a conflict table with a specified capacity.
+  ImtConflictTable* CreateImtConflictTable(size_t count, LinearAlloc* linear_alloc);
+
+  // Static version for when the class linker is not yet created.
+  static ImtConflictTable* CreateImtConflictTable(size_t count,
+                                                  LinearAlloc* linear_alloc,
+                                                  size_t pointer_size);
+
+
+  // Create the IMT and conflict tables for a class.
+  void FillIMTAndConflictTables(mirror::Class* klass) SHARED_REQUIRES(Locks::mutator_lock_);
+
+
   struct DexCacheData {
     // Weak root to the DexCache. Note: Do not decode this unnecessarily or else class unloading may
     // not work properly.
@@ -999,7 +1024,7 @@ class ClassLinker {
 
   // Returns null if not found.
   ClassTable* ClassTableForClassLoader(mirror::ClassLoader* class_loader)
-      SHARED_REQUIRES(Locks::mutator_lock_, Locks::classlinker_classes_lock_);
+      SHARED_REQUIRES(Locks::mutator_lock_);
 
   // Insert a new class table if not found.
   ClassTable* InsertClassTableForClassLoader(mirror::ClassLoader* class_loader)
@@ -1057,6 +1082,28 @@ class ClassLinker {
       REQUIRES(!dex_lock_)
       SHARED_REQUIRES(Locks::mutator_lock_);
 
+  // Sets imt_ref appropriately for LinkInterfaceMethods.
+  // If there is no method in the imt location of imt_ref it will store the given method there.
+  // Otherwise it will set the conflict method which will figure out which method to use during
+  // runtime.
+  void SetIMTRef(ArtMethod* unimplemented_method,
+                 ArtMethod* imt_conflict_method,
+                 ArtMethod* current_method,
+                 /*out*/ArtMethod** imt_ref) SHARED_REQUIRES(Locks::mutator_lock_);
+
+  void FillIMTFromIfTable(mirror::IfTable* if_table,
+                          ArtMethod* unimplemented_method,
+                          ArtMethod* imt_conflict_method,
+                          mirror::Class* klass,
+                          bool create_conflict_tables,
+                          bool ignore_copied_methods,
+                          ArtMethod** imt) SHARED_REQUIRES(Locks::mutator_lock_);
+
+  void FillImtFromSuperClass(Handle<mirror::Class> klass,
+                             ArtMethod* unimplemented_method,
+                             ArtMethod* imt_conflict_method,
+                             ArtMethod** imt) SHARED_REQUIRES(Locks::mutator_lock_);
+
   std::vector<const DexFile*> boot_class_path_;
   std::vector<std::unique_ptr<const DexFile>> boot_dex_files_;
 
diff --git a/runtime/class_table.h b/runtime/class_table.h
index eb784b5c7..686381d35 100644
--- a/runtime/class_table.h
+++ b/runtime/class_table.h
@@ -153,6 +153,10 @@ class ClassTable {
       REQUIRES(!lock_)
       SHARED_REQUIRES(Locks::mutator_lock_);
 
+  ReaderWriterMutex& GetLock() {
+    return lock_;
+  }
+
  private:
   // Lock to guard inserting and removing.
   mutable ReaderWriterMutex lock_;
diff --git a/runtime/common_throws.cc b/runtime/common_throws.cc
index b4208fe05..75cce424e 100644
--- a/runtime/common_throws.cc
+++ b/runtime/common_throws.cc
@@ -137,13 +137,21 @@ void ThrowClassCircularityError(mirror::Class* c) {
   ThrowException("Ljava/lang/ClassCircularityError;", c, msg.str().c_str());
 }
 
+void ThrowClassCircularityError(mirror::Class* c, const char* fmt, ...) {
+  va_list args;
+  va_start(args, fmt);
+  ThrowException("Ljava/lang/ClassCircularityError;", c, fmt, &args);
+  va_end(args);
+}
+
 // ClassFormatError
 
 void ThrowClassFormatError(mirror::Class* referrer, const char* fmt, ...) {
   va_list args;
   va_start(args, fmt);
   ThrowException("Ljava/lang/ClassFormatError;", referrer, fmt, &args);
-  va_end(args);}
+  va_end(args);
+}
 
 // IllegalAccessError
 
diff --git a/runtime/common_throws.h b/runtime/common_throws.h
index 39c4e52b1..c3a1f09db 100644
--- a/runtime/common_throws.h
+++ b/runtime/common_throws.h
@@ -58,6 +58,9 @@ void ThrowArrayStoreException(mirror::Class* element_class, mirror::Class* array
 void ThrowClassCircularityError(mirror::Class* c)
     SHARED_REQUIRES(Locks::mutator_lock_) COLD_ATTR;
 
+void ThrowClassCircularityError(mirror::Class* c, const char* fmt, ...)
+    SHARED_REQUIRES(Locks::mutator_lock_) COLD_ATTR;
+
 // ClassCastException
 
 void ThrowClassCastException(mirror::Class* dest_type, mirror::Class* src_type)
diff --git a/runtime/compiler_filter.h b/runtime/compiler_filter.h
index 6289d8a22..e8d74dd9e 100644
--- a/runtime/compiler_filter.h
+++ b/runtime/compiler_filter.h
@@ -44,6 +44,8 @@ class CompilerFilter FINAL {
     kEverything,          // Compile everything capable of being compiled.
   };
 
+  static const Filter kDefaultCompilerFilter = kSpeed;
+
   // Returns true if an oat file with this compiler filter contains
   // compiled executable code.
   static bool IsCompilationEnabled(Filter filter);
diff --git a/runtime/debugger.cc b/runtime/debugger.cc
index 55f68d3f2..80056423a 100644
--- a/runtime/debugger.cc
+++ b/runtime/debugger.cc
@@ -582,7 +582,7 @@ class UpdateEntryPointsClassVisitor : public ClassVisitor {
       if (Runtime::Current()->GetHeap()->IsInBootImageOatFile(code) &&
           !m.IsNative() &&
           !m.IsProxyMethod()) {
-        instrumentation_->UpdateMethodsCode(&m, GetQuickToInterpreterBridge());
+        instrumentation_->UpdateMethodsCodeFromDebugger(&m, GetQuickToInterpreterBridge());
       }
     }
     return true;
diff --git a/runtime/dex_cache_resolved_classes.h b/runtime/dex_cache_resolved_classes.h
index 80c12cb64..0febbedf0 100644
--- a/runtime/dex_cache_resolved_classes.h
+++ b/runtime/dex_cache_resolved_classes.h
@@ -26,8 +26,11 @@ namespace art {
 // Data structure for passing around which classes belonging to a dex cache / dex file are resolved.
 class DexCacheResolvedClasses {
  public:
-  DexCacheResolvedClasses(const std::string& dex_location, uint32_t location_checksum)
+  DexCacheResolvedClasses(const std::string& dex_location,
+                          const std::string& base_location,
+                          uint32_t location_checksum)
       : dex_location_(dex_location),
+        base_location_(base_location),
         location_checksum_(location_checksum) {}
 
   // Only compare the key elements, ignore the resolved classes.
@@ -35,6 +38,7 @@ class DexCacheResolvedClasses {
     if (location_checksum_ != other.location_checksum_) {
       return static_cast<int>(location_checksum_ - other.location_checksum_);
     }
+    // Don't need to compare base_location_ since dex_location_ has more info.
     return dex_location_.compare(other.dex_location_);
   }
 
@@ -47,6 +51,10 @@ class DexCacheResolvedClasses {
     return dex_location_;
   }
 
+  const std::string& GetBaseLocation() const {
+    return base_location_;
+  }
+
   uint32_t GetLocationChecksum() const {
     return location_checksum_;
   }
@@ -57,6 +65,7 @@ class DexCacheResolvedClasses {
 
  private:
   const std::string dex_location_;
+  const std::string base_location_;
   const uint32_t location_checksum_;
   // Array of resolved class def indexes.
   mutable std::unordered_set<uint16_t> classes_;
diff --git a/runtime/dex_file.cc b/runtime/dex_file.cc
index 63f3f08ba..af12abfc0 100644
--- a/runtime/dex_file.cc
+++ b/runtime/dex_file.cc
@@ -518,7 +518,7 @@ const DexFile::ClassDef* DexFile::FindClassDef(const char* descriptor, size_t ha
     return (class_def_idx != DexFile::kDexNoIndex) ? &GetClassDef(class_def_idx) : nullptr;
   }
 
-  // Fast path for rate no class defs case.
+  // Fast path for rare no class defs case.
   const uint32_t num_class_defs = NumClassDefs();
   if (num_class_defs == 0) {
     return nullptr;
@@ -548,8 +548,8 @@ const DexFile::ClassDef* DexFile::FindClassDef(uint16_t type_idx) const {
 }
 
 const DexFile::FieldId* DexFile::FindFieldId(const DexFile::TypeId& declaring_klass,
-                                              const DexFile::StringId& name,
-                                              const DexFile::TypeId& type) const {
+                                             const DexFile::StringId& name,
+                                             const DexFile::TypeId& type) const {
   // Binary search MethodIds knowing that they are sorted by class_idx, name_idx then proto_idx
   const uint16_t class_idx = GetIndexForTypeId(declaring_klass);
   const uint32_t name_idx = GetIndexForStringId(name);
diff --git a/runtime/dex_file_verifier.cc b/runtime/dex_file_verifier.cc
index 3df4e98c8..bbffbbb7b 100644
--- a/runtime/dex_file_verifier.cc
+++ b/runtime/dex_file_verifier.cc
@@ -251,6 +251,14 @@ bool DexFileVerifier::CheckValidOffsetAndSize(uint32_t offset,
   return true;
 }
 
+bool DexFileVerifier::CheckSizeLimit(uint32_t size, uint32_t limit, const char* label) {
+  if (size > limit) {
+    ErrorStringPrintf("Size(%u) should not exceed limit(%u) for %s.", size, limit, label);
+    return false;
+  }
+  return true;
+}
+
 bool DexFileVerifier::CheckHeader() {
   // Check file size from the header.
   uint32_t expected_size = header_->file_size_;
@@ -298,10 +306,12 @@ bool DexFileVerifier::CheckHeader() {
                               header_->type_ids_size_,
                               4,
                               "type-ids") &&
+      CheckSizeLimit(header_->type_ids_size_, DexFile::kDexNoIndex16, "type-ids") &&
       CheckValidOffsetAndSize(header_->proto_ids_off_,
                               header_->proto_ids_size_,
                               4,
                               "proto-ids") &&
+      CheckSizeLimit(header_->proto_ids_size_, DexFile::kDexNoIndex16, "proto-ids") &&
       CheckValidOffsetAndSize(header_->field_ids_off_,
                               header_->field_ids_size_,
                               4,
@@ -1786,13 +1796,8 @@ bool DexFileVerifier::CheckInterProtoIdItem() {
       while (curr_it.HasNext() && prev_it.HasNext()) {
         uint16_t prev_idx = prev_it.GetTypeIdx();
         uint16_t curr_idx = curr_it.GetTypeIdx();
-        if (prev_idx == DexFile::kDexNoIndex16) {
-          break;
-        }
-        if (UNLIKELY(curr_idx == DexFile::kDexNoIndex16)) {
-          ErrorStringPrintf("Out-of-order proto_id arguments");
-          return false;
-        }
+        DCHECK_NE(prev_idx, DexFile::kDexNoIndex16);
+        DCHECK_NE(curr_idx, DexFile::kDexNoIndex16);
 
         if (prev_idx < curr_idx) {
           break;
@@ -1804,6 +1809,12 @@ bool DexFileVerifier::CheckInterProtoIdItem() {
         prev_it.Next();
         curr_it.Next();
       }
+      if (!curr_it.HasNext()) {
+        // Either a duplicate ProtoId or a ProtoId with a shorter argument list follows
+        // a ProtoId with a longer one. Both cases are forbidden by the specification.
+        ErrorStringPrintf("Out-of-order proto_id arguments");
+        return false;
+      }
     }
   }
 
@@ -2358,7 +2369,8 @@ static bool CheckAtMostOneOfPublicProtectedPrivate(uint32_t flags) {
 static std::string GetStringOrError(const uint8_t* const begin,
                                     const DexFile::Header* const header,
                                     uint32_t string_idx) {
-  if (header->string_ids_size_ < string_idx) {
+  // The `string_idx` is not guaranteed to be valid yet.
+  if (header->string_ids_size_ <= string_idx) {
     return "(error)";
   }
 
@@ -2375,9 +2387,11 @@ static std::string GetStringOrError(const uint8_t* const begin,
 static std::string GetClassOrError(const uint8_t* const begin,
                                    const DexFile::Header* const header,
                                    uint32_t class_idx) {
-  if (header->type_ids_size_ < class_idx) {
-    return "(error)";
-  }
+  // The `class_idx` is either `FieldId::class_idx_` or `MethodId::class_idx_` and
+  // it has already been checked in `DexFileVerifier::CheckClassDataItemField()`
+  // or `DexFileVerifier::CheckClassDataItemMethod()`, respectively, to match
+  // a valid defining class.
+  CHECK_LT(class_idx, header->type_ids_size_);
 
   const DexFile::TypeId* type_id =
       reinterpret_cast<const DexFile::TypeId*>(begin + header->type_ids_off_) + class_idx;
@@ -2390,9 +2404,8 @@ static std::string GetClassOrError(const uint8_t* const begin,
 static std::string GetFieldDescriptionOrError(const uint8_t* const begin,
                                               const DexFile::Header* const header,
                                               uint32_t idx) {
-  if (header->field_ids_size_ < idx) {
-    return "(error)";
-  }
+  // The `idx` has already been checked in `DexFileVerifier::CheckClassDataItemField()`.
+  CHECK_LT(idx, header->field_ids_size_);
 
   const DexFile::FieldId* field_id =
       reinterpret_cast<const DexFile::FieldId*>(begin + header->field_ids_off_) + idx;
@@ -2408,9 +2421,8 @@ static std::string GetFieldDescriptionOrError(const uint8_t* const begin,
 static std::string GetMethodDescriptionOrError(const uint8_t* const begin,
                                                const DexFile::Header* const header,
                                                uint32_t idx) {
-  if (header->method_ids_size_ < idx) {
-    return "(error)";
-  }
+  // The `idx` has already been checked in `DexFileVerifier::CheckClassDataItemMethod()`.
+  CHECK_LT(idx, header->method_ids_size_);
 
   const DexFile::MethodId* method_id =
       reinterpret_cast<const DexFile::MethodId*>(begin + header->method_ids_off_) + idx;
@@ -2608,7 +2620,13 @@ bool DexFileVerifier::CheckMethodAccessFlags(uint32_t method_index,
       *error_msg = StringPrintf("Constructor %" PRIu32 "(%s) is not flagged correctly wrt/ static.",
                                 method_index,
                                 GetMethodDescriptionOrError(begin_, header_, method_index).c_str());
-      return false;
+      if (header_->GetVersion() >= DexFile::kDefaultMethodsVersion) {
+        return false;
+      } else {
+        // Allow in older versions, but warn.
+        LOG(WARNING) << "This dex file is invalid and will be rejected in the future. Error is: "
+                     << *error_msg;
+      }
     }
   }
   // Check that static and private methods, as well as constructors, are in the direct methods list,
@@ -2662,7 +2680,13 @@ bool DexFileVerifier::CheckMethodAccessFlags(uint32_t method_index,
       *error_msg = StringPrintf("Constructor %u(%s) must not be abstract or native",
                                 method_index,
                                 GetMethodDescriptionOrError(begin_, header_, method_index).c_str());
-      return false;
+      if (header_->GetVersion() >= DexFile::kDefaultMethodsVersion) {
+        return false;
+      } else {
+        // Allow in older versions, but warn.
+        LOG(WARNING) << "This dex file is invalid and will be rejected in the future. Error is: "
+                      << *error_msg;
+      }
     }
     if ((method_access_flags & kAccAbstract) != 0) {
       // Abstract methods are not allowed to have the following flags.
diff --git a/runtime/dex_file_verifier.h b/runtime/dex_file_verifier.h
index be0e6d83f..90409db44 100644
--- a/runtime/dex_file_verifier.h
+++ b/runtime/dex_file_verifier.h
@@ -49,6 +49,8 @@ class DexFileVerifier {
   // Checks whether the offset is zero (when size is zero) or that the offset falls within the area
   // claimed by the file.
   bool CheckValidOffsetAndSize(uint32_t offset, uint32_t size, size_t alignment, const char* label);
+  // Checks whether the size is less than the limit.
+  bool CheckSizeLimit(uint32_t size, uint32_t limit, const char* label);
   bool CheckIndex(uint32_t field, uint32_t limit, const char* label);
 
   bool CheckHeader();
diff --git a/runtime/dex_file_verifier_test.cc b/runtime/dex_file_verifier_test.cc
index 344d186ad..3741c1e76 100644
--- a/runtime/dex_file_verifier_test.cc
+++ b/runtime/dex_file_verifier_test.cc
@@ -57,7 +57,14 @@ static const uint8_t kBase64Map[256] = {
   255, 255, 255, 255
 };
 
-static inline uint8_t* DecodeBase64(const char* src, size_t* dst_size) {
+// Make the Dex file version 37.
+static void MakeDexVersion37(DexFile* dex_file) {
+  size_t offset = OFFSETOF_MEMBER(DexFile::Header, magic_) + 6;
+  CHECK_EQ(*(dex_file->Begin() + offset), '5');
+  *(const_cast<uint8_t*>(dex_file->Begin()) + offset) = '7';
+}
+
+static inline std::unique_ptr<uint8_t[]> DecodeBase64(const char* src, size_t* dst_size) {
   std::vector<uint8_t> tmp;
   uint32_t t = 0, y = 0;
   int g = 3;
@@ -100,7 +107,7 @@ static inline uint8_t* DecodeBase64(const char* src, size_t* dst_size) {
     *dst_size = 0;
   }
   std::copy(tmp.begin(), tmp.end(), dst.get());
-  return dst.release();
+  return dst;
 }
 
 static void FixUpChecksum(uint8_t* dex_file) {
@@ -113,25 +120,18 @@ static void FixUpChecksum(uint8_t* dex_file) {
   header->checksum_ = adler_checksum;
 }
 
-// Custom deleter. Necessary to clean up the memory we use (to be able to mutate).
-struct DexFileDeleter {
-  void operator()(DexFile* in) {
-    if (in != nullptr) {
-      delete[] in->Begin();
-      delete in;
-    }
-  }
-};
-
-using DexFileUniquePtr = std::unique_ptr<DexFile, DexFileDeleter>;
-
 class DexFileVerifierTest : public CommonRuntimeTest {
  protected:
   void VerifyModification(const char* dex_file_base64_content,
                           const char* location,
                           std::function<void(DexFile*)> f,
                           const char* expected_error) {
-    DexFileUniquePtr dex_file(WrapAsDexFile(dex_file_base64_content));
+    size_t length;
+    std::unique_ptr<uint8_t[]> dex_bytes = DecodeBase64(dex_file_base64_content, &length);
+    CHECK(dex_bytes != nullptr);
+    // Note: `dex_file` will be destroyed before `dex_bytes`.
+    std::unique_ptr<DexFile> dex_file(
+        new DexFile(dex_bytes.get(), length, "tmp", 0, nullptr, nullptr));
     f(dex_file.get());
     FixUpChecksum(const_cast<uint8_t*>(dex_file->Begin()));
 
@@ -150,15 +150,6 @@ class DexFileVerifierTest : public CommonRuntimeTest {
       }
     }
   }
-
- private:
-  static DexFile* WrapAsDexFile(const char* dex_file_content_in_base_64) {
-    // Decode base64.
-    size_t length;
-    uint8_t* dex_bytes = DecodeBase64(dex_file_content_in_base_64, &length);
-    CHECK(dex_bytes != nullptr);
-    return new DexFile(dex_bytes, length, "tmp", 0, nullptr, nullptr);
-  }
 };
 
 static std::unique_ptr<const DexFile> OpenDexFileBase64(const char* base64,
@@ -290,7 +281,9 @@ static const char kMethodFlagsTestDex[] =
 // Find the method data for the first method with the given name (from class 0). Note: the pointer
 // is to the access flags, so that the caller doesn't have to handle the leb128-encoded method-index
 // delta.
-static const uint8_t* FindMethodData(const DexFile* dex_file, const char* name) {
+static const uint8_t* FindMethodData(const DexFile* dex_file,
+                                     const char* name,
+                                     /*out*/ uint32_t* method_idx = nullptr) {
   const DexFile::ClassDef& class_def = dex_file->GetClassDef(0);
   const uint8_t* class_data = dex_file->GetClassData(class_def);
 
@@ -316,6 +309,9 @@ static const uint8_t* FindMethodData(const DexFile* dex_file, const char* name)
     const DexFile::StringId& string_id = dex_file->GetStringId(name_index);
     const char* str = dex_file->GetStringData(string_id);
     if (strcmp(name, str) == 0) {
+      if (method_idx != nullptr) {
+        *method_idx = method_index;
+      }
       DecodeUnsignedLeb128(&trailing);
       return trailing;
     }
@@ -449,6 +445,7 @@ TEST_F(DexFileVerifierTest, MethodAccessFlagsConstructors) {
         kMethodFlagsTestDex,
         "method_flags_constructor_native_nocode",
         [&](DexFile* dex_file) {
+          MakeDexVersion37(dex_file);
           ApplyMaskToMethodFlags(dex_file, "foo", ~kAccDeclaredSynchronized);
           ApplyMaskToMethodFlags(dex_file, "bar", ~kAccDeclaredSynchronized);
 
@@ -461,6 +458,7 @@ TEST_F(DexFileVerifierTest, MethodAccessFlagsConstructors) {
         kMethodFlagsTestDex,
         "method_flags_constructor_abstract_nocode",
         [&](DexFile* dex_file) {
+          MakeDexVersion37(dex_file);
           ApplyMaskToMethodFlags(dex_file, "foo", ~kAccDeclaredSynchronized);
           ApplyMaskToMethodFlags(dex_file, "bar", ~kAccDeclaredSynchronized);
 
@@ -521,6 +519,7 @@ TEST_F(DexFileVerifierTest, MethodAccessFlagsConstructors) {
       kMethodFlagsTestDex,
       "init_not_allowed_flags",
       [&](DexFile* dex_file) {
+        MakeDexVersion37(dex_file);
         ApplyMaskToMethodFlags(dex_file, "foo", ~kAccDeclaredSynchronized);
         ApplyMaskToMethodFlags(dex_file, "bar", ~kAccDeclaredSynchronized);
 
@@ -683,6 +682,22 @@ TEST_F(DexFileVerifierTest, MethodAccessFlagsIgnoredOK) {
   }
 }
 
+TEST_F(DexFileVerifierTest, B28552165) {
+  // Regression test for bad error string retrieval in different situations.
+  // Using invalid access flags to trigger the error.
+  VerifyModification(
+      kMethodFlagsTestDex,
+      "b28552165",
+      [](DexFile* dex_file) {
+        OrMaskToMethodFlags(dex_file, "foo", kAccPublic | kAccProtected);
+        uint32_t method_idx;
+        FindMethodData(dex_file, "foo", &method_idx);
+        auto* method_id = const_cast<DexFile::MethodId*>(&dex_file->GetMethodId(method_idx));
+        method_id->name_idx_ = dex_file->NumStringIds();
+      },
+      "Method may have only one of public/protected/private, LMethodFlags;.(error)");
+}
+
 // Set of dex files for interface method tests. As it's not as easy to mutate method names, it's
 // just easier to break up bad cases.
 
@@ -725,13 +740,6 @@ static uint32_t ApplyMaskShifted(uint32_t src_value, uint32_t mask) {
   return result;
 }
 
-// Make the Dex file version 37.
-static void MakeDexVersion37(DexFile* dex_file) {
-  size_t offset = OFFSETOF_MEMBER(DexFile::Header, magic_) + 6;
-  CHECK_EQ(*(dex_file->Begin() + offset), '5');
-  *(const_cast<uint8_t*>(dex_file->Begin()) + offset) = '7';
-}
-
 TEST_F(DexFileVerifierTest, MethodAccessFlagsInterfaces) {
   VerifyModification(
       kMethodFlagsInterface,
@@ -1436,4 +1444,81 @@ TEST_F(DexFileVerifierTest, SectionAlignment) {
   }
 }
 
+// Generated from
+//
+// .class LOverloading;
+//
+// .super Ljava/lang/Object;
+//
+// .method public static foo()V
+// .registers 1
+//     return-void
+// .end method
+//
+// .method public static foo(I)V
+// .registers 1
+//     return-void
+// .end method
+static const char kProtoOrderingTestDex[] =
+    "ZGV4CjAzNQA1L+ABE6voQ9Lr4Ci//efB53oGnDr5PinsAQAAcAAAAHhWNBIAAAAAAAAAAFgBAAAG"
+    "AAAAcAAAAAQAAACIAAAAAgAAAJgAAAAAAAAAAAAAAAIAAACwAAAAAQAAAMAAAAAMAQAA4AAAAOAA"
+    "AADjAAAA8gAAAAYBAAAJAQAADQEAAAAAAAABAAAAAgAAAAMAAAADAAAAAwAAAAAAAAAEAAAAAwAA"
+    "ABQBAAABAAAABQAAAAEAAQAFAAAAAQAAAAAAAAACAAAAAAAAAP////8AAAAASgEAAAAAAAABSQAN"
+    "TE92ZXJsb2FkaW5nOwASTGphdmEvbGFuZy9PYmplY3Q7AAFWAAJWSQADZm9vAAAAAQAAAAAAAAAA"
+    "AAAAAAAAAAEAAAAAAAAAAAAAAAEAAAAOAAAAAQABAAAAAAAAAAAAAQAAAA4AAAACAAAJpAIBCbgC"
+    "AAAMAAAAAAAAAAEAAAAAAAAAAQAAAAYAAABwAAAAAgAAAAQAAACIAAAAAwAAAAIAAACYAAAABQAA"
+    "AAIAAACwAAAABgAAAAEAAADAAAAAAiAAAAYAAADgAAAAARAAAAEAAAAUAQAAAxAAAAIAAAAcAQAA"
+    "ASAAAAIAAAAkAQAAACAAAAEAAABKAQAAABAAAAEAAABYAQAA";
+
+TEST_F(DexFileVerifierTest, ProtoOrdering) {
+  {
+    // The input dex file should be good before modification.
+    ScratchFile tmp;
+    std::string error_msg;
+    std::unique_ptr<const DexFile> raw(OpenDexFileBase64(kProtoOrderingTestDex,
+                                                         tmp.GetFilename().c_str(),
+                                                         &error_msg));
+    ASSERT_TRUE(raw.get() != nullptr) << error_msg;
+  }
+
+  // Modify the order of the ProtoIds for two overloads of "foo" with the
+  // same return type and one having longer parameter list than the other.
+  for (size_t i = 0; i != 2; ++i) {
+    VerifyModification(
+        kProtoOrderingTestDex,
+        "proto_ordering",
+        [i](DexFile* dex_file) {
+          uint32_t method_idx;
+          const uint8_t* data = FindMethodData(dex_file, "foo", &method_idx);
+          CHECK(data != nullptr);
+          // There should be 2 methods called "foo".
+          CHECK_LT(method_idx + 1u, dex_file->NumMethodIds());
+          CHECK_EQ(dex_file->GetMethodId(method_idx).name_idx_,
+                   dex_file->GetMethodId(method_idx + 1).name_idx_);
+          CHECK_EQ(dex_file->GetMethodId(method_idx).proto_idx_ + 1u,
+                   dex_file->GetMethodId(method_idx + 1).proto_idx_);
+          // Their return types should be the same.
+          uint32_t proto1_idx = dex_file->GetMethodId(method_idx).proto_idx_;
+          const DexFile::ProtoId& proto1 = dex_file->GetProtoId(proto1_idx);
+          const DexFile::ProtoId& proto2 = dex_file->GetProtoId(proto1_idx + 1u);
+          CHECK_EQ(proto1.return_type_idx_, proto2.return_type_idx_);
+          // And the first should not have any parameters while the second should have some.
+          CHECK(!DexFileParameterIterator(*dex_file, proto1).HasNext());
+          CHECK(DexFileParameterIterator(*dex_file, proto2).HasNext());
+          if (i == 0) {
+            // Swap the proto parameters and shorties to break the ordering.
+            std::swap(const_cast<uint32_t&>(proto1.parameters_off_),
+                      const_cast<uint32_t&>(proto2.parameters_off_));
+            std::swap(const_cast<uint32_t&>(proto1.shorty_idx_),
+                      const_cast<uint32_t&>(proto2.shorty_idx_));
+          } else {
+            // Copy the proto parameters and shorty to create duplicate proto id.
+            const_cast<uint32_t&>(proto1.parameters_off_) = proto2.parameters_off_;
+            const_cast<uint32_t&>(proto1.shorty_idx_) = proto2.shorty_idx_;
+          }
+        },
+        "Out-of-order proto_id arguments");
+  }
+}
+
 }  // namespace art
diff --git a/runtime/entrypoints/entrypoint_utils-inl.h b/runtime/entrypoints/entrypoint_utils-inl.h
index 16fbfaad3..fc6257302 100644
--- a/runtime/entrypoints/entrypoint_utils-inl.h
+++ b/runtime/entrypoints/entrypoint_utils-inl.h
@@ -514,12 +514,18 @@ inline ArtMethod* FindMethodFromCode(uint32_t method_idx, mirror::Object** this_
         CHECK(self->IsExceptionPending());
         return nullptr;
       } else if (!method_reference_class->IsInterface()) {
-        // It is not an interface.
-        mirror::Class* super_class = referring_class->GetSuperClass();
+        // It is not an interface. If the referring class is in the class hierarchy of the
+        // referenced class in the bytecode, we use its super class. Otherwise, we throw
+        // a NoSuchMethodError.
+        mirror::Class* super_class = nullptr;
+        if (method_reference_class->IsAssignableFrom(referring_class)) {
+          super_class = referring_class->GetSuperClass();
+        }
         uint16_t vtable_index = resolved_method->GetMethodIndex();
         if (access_check) {
           // Check existence of super class.
-          if (super_class == nullptr || !super_class->HasVTable() ||
+          if (super_class == nullptr ||
+              !super_class->HasVTable() ||
               vtable_index >= static_cast<uint32_t>(super_class->GetVTableLength())) {
             // Behavior to agree with that of the verifier.
             ThrowNoSuchMethodError(type, resolved_method->GetDeclaringClass(),
@@ -693,8 +699,13 @@ inline ArtMethod* FindMethodFast(uint32_t method_idx, mirror::Object* this_objec
       // Need to do full type resolution...
       return nullptr;
     } else if (!method_reference_class->IsInterface()) {
-      // It is not an interface.
-      mirror::Class* super_class = referrer->GetDeclaringClass()->GetSuperClass();
+      // It is not an interface. If the referring class is in the class hierarchy of the
+      // referenced class in the bytecode, we use its super class. Otherwise, we cannot
+      // resolve the method.
+      if (!method_reference_class->IsAssignableFrom(referring_class)) {
+        return nullptr;
+      }
+      mirror::Class* super_class = referring_class->GetSuperClass();
       if (resolved_method->GetMethodIndex() >= super_class->GetVTableLength()) {
         // The super class does not have the method.
         return nullptr;
diff --git a/runtime/entrypoints/quick/quick_default_externs.h b/runtime/entrypoints/quick/quick_default_externs.h
index 4e01d8031..f3a0d2f3e 100644
--- a/runtime/entrypoints/quick/quick_default_externs.h
+++ b/runtime/entrypoints/quick/quick_default_externs.h
@@ -77,6 +77,10 @@ extern "C" void art_quick_handle_fill_data(void*, void*);
 extern "C" void art_quick_lock_object(art::mirror::Object*);
 extern "C" void art_quick_unlock_object(art::mirror::Object*);
 
+// Lock entrypoints that do not inline any behavior (e.g., thin-locks).
+extern "C" void art_quick_lock_object_no_inline(art::mirror::Object*);
+extern "C" void art_quick_unlock_object_no_inline(art::mirror::Object*);
+
 // Math entrypoints.
 extern "C" int64_t art_quick_d2l(double);
 extern "C" int64_t art_quick_f2l(float);
diff --git a/runtime/entrypoints/quick/quick_default_init_entrypoints.h b/runtime/entrypoints/quick/quick_default_init_entrypoints.h
new file mode 100644
index 000000000..5dafa8b59
--- /dev/null
+++ b/runtime/entrypoints/quick/quick_default_init_entrypoints.h
@@ -0,0 +1,124 @@
+/*
+ * 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.
+ */
+
+#ifndef ART_RUNTIME_ENTRYPOINTS_QUICK_QUICK_DEFAULT_INIT_ENTRYPOINTS_H_
+#define ART_RUNTIME_ENTRYPOINTS_QUICK_QUICK_DEFAULT_INIT_ENTRYPOINTS_H_
+
+#include "base/logging.h"
+#include "entrypoints/jni/jni_entrypoints.h"
+#include "entrypoints/runtime_asm_entrypoints.h"
+#include "quick_alloc_entrypoints.h"
+#include "quick_default_externs.h"
+#include "quick_entrypoints.h"
+
+namespace art {
+
+void DefaultInitEntryPoints(JniEntryPoints* jpoints, QuickEntryPoints* qpoints) {
+  // JNI
+  jpoints->pDlsymLookup = art_jni_dlsym_lookup_stub;
+
+  // Alloc
+  ResetQuickAllocEntryPoints(qpoints);
+
+  // DexCache
+  qpoints->pInitializeStaticStorage = art_quick_initialize_static_storage;
+  qpoints->pInitializeTypeAndVerifyAccess = art_quick_initialize_type_and_verify_access;
+  qpoints->pInitializeType = art_quick_initialize_type;
+  qpoints->pResolveString = art_quick_resolve_string;
+
+  // Field
+  qpoints->pSet8Instance = art_quick_set8_instance;
+  qpoints->pSet8Static = art_quick_set8_static;
+  qpoints->pSet16Instance = art_quick_set16_instance;
+  qpoints->pSet16Static = art_quick_set16_static;
+  qpoints->pSet32Instance = art_quick_set32_instance;
+  qpoints->pSet32Static = art_quick_set32_static;
+  qpoints->pSet64Instance = art_quick_set64_instance;
+  qpoints->pSet64Static = art_quick_set64_static;
+  qpoints->pSetObjInstance = art_quick_set_obj_instance;
+  qpoints->pSetObjStatic = art_quick_set_obj_static;
+  qpoints->pGetByteInstance = art_quick_get_byte_instance;
+  qpoints->pGetBooleanInstance = art_quick_get_boolean_instance;
+  qpoints->pGetShortInstance = art_quick_get_short_instance;
+  qpoints->pGetCharInstance = art_quick_get_char_instance;
+  qpoints->pGet32Instance = art_quick_get32_instance;
+  qpoints->pGet64Instance = art_quick_get64_instance;
+  qpoints->pGetObjInstance = art_quick_get_obj_instance;
+  qpoints->pGetByteStatic = art_quick_get_byte_static;
+  qpoints->pGetBooleanStatic = art_quick_get_boolean_static;
+  qpoints->pGetShortStatic = art_quick_get_short_static;
+  qpoints->pGetCharStatic = art_quick_get_char_static;
+  qpoints->pGet32Static = art_quick_get32_static;
+  qpoints->pGet64Static = art_quick_get64_static;
+  qpoints->pGetObjStatic = art_quick_get_obj_static;
+
+  // Array
+  qpoints->pAputObjectWithNullAndBoundCheck = art_quick_aput_obj_with_null_and_bound_check;
+  qpoints->pAputObjectWithBoundCheck = art_quick_aput_obj_with_bound_check;
+  qpoints->pAputObject = art_quick_aput_obj;
+  qpoints->pHandleFillArrayData = art_quick_handle_fill_data;
+
+  // JNI
+  qpoints->pJniMethodStart = JniMethodStart;
+  qpoints->pJniMethodStartSynchronized = JniMethodStartSynchronized;
+  qpoints->pJniMethodEnd = JniMethodEnd;
+  qpoints->pJniMethodEndSynchronized = JniMethodEndSynchronized;
+  qpoints->pJniMethodEndWithReference = JniMethodEndWithReference;
+  qpoints->pJniMethodEndWithReferenceSynchronized = JniMethodEndWithReferenceSynchronized;
+  qpoints->pQuickGenericJniTrampoline = art_quick_generic_jni_trampoline;
+
+  // Locks
+  if (UNLIKELY(VLOG_IS_ON(systrace_lock_logging))) {
+    qpoints->pLockObject = art_quick_lock_object_no_inline;
+    qpoints->pUnlockObject = art_quick_unlock_object_no_inline;
+  } else {
+    qpoints->pLockObject = art_quick_lock_object;
+    qpoints->pUnlockObject = art_quick_unlock_object;
+  }
+
+  // Invocation
+  qpoints->pQuickImtConflictTrampoline = art_quick_imt_conflict_trampoline;
+  qpoints->pQuickResolutionTrampoline = art_quick_resolution_trampoline;
+  qpoints->pQuickToInterpreterBridge = art_quick_to_interpreter_bridge;
+  qpoints->pInvokeDirectTrampolineWithAccessCheck =
+      art_quick_invoke_direct_trampoline_with_access_check;
+  qpoints->pInvokeInterfaceTrampolineWithAccessCheck =
+      art_quick_invoke_interface_trampoline_with_access_check;
+  qpoints->pInvokeStaticTrampolineWithAccessCheck =
+      art_quick_invoke_static_trampoline_with_access_check;
+  qpoints->pInvokeSuperTrampolineWithAccessCheck =
+      art_quick_invoke_super_trampoline_with_access_check;
+  qpoints->pInvokeVirtualTrampolineWithAccessCheck =
+      art_quick_invoke_virtual_trampoline_with_access_check;
+
+  // Thread
+  qpoints->pTestSuspend = art_quick_test_suspend;
+
+  // Throws
+  qpoints->pDeliverException = art_quick_deliver_exception;
+  qpoints->pThrowArrayBounds = art_quick_throw_array_bounds;
+  qpoints->pThrowDivZero = art_quick_throw_div_zero;
+  qpoints->pThrowNoSuchMethod = art_quick_throw_no_such_method;
+  qpoints->pThrowNullPointer = art_quick_throw_null_pointer_exception;
+  qpoints->pThrowStackOverflow = art_quick_throw_stack_overflow;
+
+  // Deoptimize
+  qpoints->pDeoptimize = art_quick_deoptimize_from_compiled_code;
+};
+
+}  // namespace art
+
+#endif  // ART_RUNTIME_ENTRYPOINTS_QUICK_QUICK_DEFAULT_INIT_ENTRYPOINTS_H_
diff --git a/runtime/entrypoints/quick/quick_trampoline_entrypoints.cc b/runtime/entrypoints/quick/quick_trampoline_entrypoints.cc
index da6af724a..e9cdbb743 100644
--- a/runtime/entrypoints/quick/quick_trampoline_entrypoints.cc
+++ b/runtime/entrypoints/quick/quick_trampoline_entrypoints.cc
@@ -2174,7 +2174,8 @@ extern "C" TwoWordReturn artInvokeInterfaceTrampoline(uint32_t deadbeef ATTRIBUT
         imt_index % mirror::Class::kImtSize, sizeof(void*));
     if (LIKELY(conflict_method->IsRuntimeMethod())) {
       ImtConflictTable* current_table = conflict_method->GetImtConflictTable(sizeof(void*));
-      method = current_table->Lookup(interface_method);
+      DCHECK(current_table != nullptr);
+      method = current_table->Lookup(interface_method, sizeof(void*));
     } else {
       // It seems we aren't really a conflict method!
       method = cls->FindVirtualMethodForInterface(interface_method, sizeof(void*));
@@ -2225,34 +2226,13 @@ extern "C" TwoWordReturn artInvokeInterfaceTrampoline(uint32_t deadbeef ATTRIBUT
   ArtMethod* conflict_method = cls->GetEmbeddedImTableEntry(
       imt_index % mirror::Class::kImtSize, sizeof(void*));
   if (conflict_method->IsRuntimeMethod()) {
-    ImtConflictTable* current_table = conflict_method->GetImtConflictTable(sizeof(void*));
-    Runtime* runtime = Runtime::Current();
-    LinearAlloc* linear_alloc = (cls->GetClassLoader() == nullptr)
-        ? runtime->GetLinearAlloc()
-        : cls->GetClassLoader()->GetAllocator();
-    bool is_new_entry = (conflict_method == runtime->GetImtConflictMethod());
-
-    // Create a new entry if the existing one is the shared conflict method.
-    ArtMethod* new_conflict_method = is_new_entry
-        ? runtime->CreateImtConflictMethod(linear_alloc)
-        : conflict_method;
-
-    // Allocate a new table. Note that we will leak this table at the next conflict,
-    // but that's a tradeoff compared to making the table fixed size.
-    void* data = linear_alloc->Alloc(
-        self, ImtConflictTable::ComputeSizeWithOneMoreEntry(current_table));
-    CHECK(data != nullptr) << "Out of memory";
-    ImtConflictTable* new_table = new (data) ImtConflictTable(
-        current_table, interface_method, method);
-
-    // Do a fence to ensure threads see the data in the table before it is assigned
-    // to the conlict method.
-    // Note that there is a race in the presence of multiple threads and we may leak
-    // memory from the LinearAlloc, but that's a tradeoff compared to using
-    // atomic operations.
-    QuasiAtomic::ThreadFenceRelease();
-    new_conflict_method->SetImtConflictTable(new_table);
-    if (is_new_entry) {
+    ArtMethod* new_conflict_method = Runtime::Current()->GetClassLinker()->AddMethodToConflictTable(
+        cls.Get(),
+        conflict_method,
+        interface_method,
+        method,
+        /*force_new_conflict_method*/false);
+    if (new_conflict_method != conflict_method) {
       // Update the IMT if we create a new conflict method. No fence needed here, as the
       // data is consistent.
       cls->SetEmbeddedImTableEntry(imt_index % mirror::Class::kImtSize,
diff --git a/runtime/gc/heap.cc b/runtime/gc/heap.cc
index c2f772f87..cdd5f2e12 100644
--- a/runtime/gc/heap.cc
+++ b/runtime/gc/heap.cc
@@ -119,6 +119,8 @@ static constexpr uint32_t kAllocSpaceBeginForDeterministicAoT = 0x40000000;
 // Dump the rosalloc stats on SIGQUIT.
 static constexpr bool kDumpRosAllocStatsOnSigQuit = false;
 
+static constexpr size_t kNativeAllocationHistogramBuckets = 16;
+
 static inline bool CareAboutPauseTimes() {
   return Runtime::Current()->InJankPerceptibleProcessState();
 }
@@ -186,6 +188,11 @@ Heap::Heap(size_t initial_size,
       total_objects_freed_ever_(0),
       num_bytes_allocated_(0),
       native_bytes_allocated_(0),
+      native_histogram_lock_("Native allocation lock"),
+      native_allocation_histogram_("Native allocation sizes",
+                                   1U,
+                                   kNativeAllocationHistogramBuckets),
+      native_free_histogram_("Native free sizes", 1U, kNativeAllocationHistogramBuckets),
       num_bytes_freed_revoke_(0),
       verify_missing_card_marks_(false),
       verify_system_weaks_(false),
@@ -304,7 +311,7 @@ Heap::Heap(size_t initial_size,
 
           const OatHeader& boot_oat_header = boot_oat_file->GetOatHeader();
           const char* boot_classpath =
-              boot_oat_header.GetStoreValueByKey(OatHeader::kBootClassPath);
+              boot_oat_header.GetStoreValueByKey(OatHeader::kBootClassPathKey);
           if (boot_classpath == nullptr) {
             continue;
           }
@@ -1185,6 +1192,20 @@ void Heap::DumpGcPerformanceInfo(std::ostream& os) {
     rosalloc_space_->DumpStats(os);
   }
 
+  {
+    MutexLock mu(Thread::Current(), native_histogram_lock_);
+    if (native_allocation_histogram_.SampleSize() > 0u) {
+      os << "Histogram of native allocation ";
+      native_allocation_histogram_.DumpBins(os);
+      os << " bucket size " << native_allocation_histogram_.BucketWidth() << "\n";
+    }
+    if (native_free_histogram_.SampleSize() > 0u) {
+      os << "Histogram of native free ";
+      native_free_histogram_.DumpBins(os);
+      os << " bucket size " << native_free_histogram_.BucketWidth() << "\n";
+    }
+  }
+
   BaseMutex::DumpAll(os);
 }
 
@@ -2687,8 +2708,8 @@ collector::GcType Heap::CollectGarbageInternal(collector::GcType gc_type,
     concurrent_start_bytes_ = std::numeric_limits<size_t>::max();
   }
 
-  if ((gc_type == collector::kGcTypeFull) && runtime->UseJit()) {
-    // It's time to clear all inline caches, in case some classes can be unloaded.
+  // It's time to clear all inline caches, in case some classes can be unloaded.
+  if ((gc_type == collector::kGcTypeFull) && (runtime->GetJit() != nullptr)) {
     runtime->GetJit()->GetCodeCache()->ClearGcRootsInInlineCaches(self);
   }
 
@@ -3848,6 +3869,10 @@ void Heap::RunFinalization(JNIEnv* env, uint64_t timeout) {
 
 void Heap::RegisterNativeAllocation(JNIEnv* env, size_t bytes) {
   Thread* self = ThreadForEnv(env);
+  {
+    MutexLock mu(self, native_histogram_lock_);
+    native_allocation_histogram_.AddValue(bytes);
+  }
   if (native_need_to_run_finalization_) {
     RunFinalization(env, kNativeAllocationFinalizeTimeout);
     UpdateMaxNativeFootprint();
@@ -3892,6 +3917,10 @@ void Heap::RegisterNativeAllocation(JNIEnv* env, size_t bytes) {
 
 void Heap::RegisterNativeFree(JNIEnv* env, size_t bytes) {
   size_t expected_size;
+  {
+    MutexLock mu(Thread::Current(), native_histogram_lock_);
+    native_free_histogram_.AddValue(bytes);
+  }
   do {
     expected_size = native_bytes_allocated_.LoadRelaxed();
     if (UNLIKELY(bytes > expected_size)) {
diff --git a/runtime/gc/heap.h b/runtime/gc/heap.h
index fada1a221..2a1a4a17a 100644
--- a/runtime/gc/heap.h
+++ b/runtime/gc/heap.h
@@ -241,9 +241,9 @@ class Heap {
       SHARED_REQUIRES(Locks::mutator_lock_);
 
   void RegisterNativeAllocation(JNIEnv* env, size_t bytes)
-      REQUIRES(!*gc_complete_lock_, !*pending_task_lock_);
+      REQUIRES(!*gc_complete_lock_, !*pending_task_lock_, !native_histogram_lock_);
   void RegisterNativeFree(JNIEnv* env, size_t bytes)
-      REQUIRES(!*gc_complete_lock_, !*pending_task_lock_);
+      REQUIRES(!*gc_complete_lock_, !*pending_task_lock_, !native_histogram_lock_);
 
   // Change the allocator, updates entrypoints.
   void ChangeAllocator(AllocatorType allocator)
@@ -532,7 +532,7 @@ class Heap {
   space::Space* FindSpaceFromObject(const mirror::Object*, bool fail_ok) const
       SHARED_REQUIRES(Locks::mutator_lock_);
 
-  void DumpForSigQuit(std::ostream& os) REQUIRES(!*gc_complete_lock_);
+  void DumpForSigQuit(std::ostream& os) REQUIRES(!*gc_complete_lock_, !native_histogram_lock_);
 
   // Do a pending collector transition.
   void DoPendingCollectorTransition() REQUIRES(!*gc_complete_lock_);
@@ -654,7 +654,8 @@ class Heap {
   std::string SafePrettyTypeOf(mirror::Object* obj) NO_THREAD_SAFETY_ANALYSIS;
 
   // GC performance measuring
-  void DumpGcPerformanceInfo(std::ostream& os) REQUIRES(!*gc_complete_lock_);
+  void DumpGcPerformanceInfo(std::ostream& os)
+      REQUIRES(!*gc_complete_lock_, !native_histogram_lock_);
   void ResetGcPerformanceInfo() REQUIRES(!*gc_complete_lock_);
 
   // Thread pool.
@@ -1156,6 +1157,11 @@ class Heap {
   // Bytes which are allocated and managed by native code but still need to be accounted for.
   Atomic<size_t> native_bytes_allocated_;
 
+  // Native allocation stats.
+  Mutex native_histogram_lock_;
+  Histogram<uint64_t> native_allocation_histogram_;
+  Histogram<uint64_t> native_free_histogram_;
+
   // Number of bytes freed by thread local buffer revokes. This will
   // cancel out the ahead-of-time bulk counting of bytes allocated in
   // rosalloc thread-local buffers.  It is temporarily accumulated
diff --git a/runtime/gc/space/image_space.cc b/runtime/gc/space/image_space.cc
index d386c7435..78c570fa9 100644
--- a/runtime/gc/space/image_space.cc
+++ b/runtime/gc/space/image_space.cc
@@ -880,7 +880,7 @@ class FixupObjectVisitor : public FixupVisitor {
 
 class ForwardObjectAdapter {
  public:
-  ALWAYS_INLINE ForwardObjectAdapter(const FixupVisitor* visitor) : visitor_(visitor) {}
+  ALWAYS_INLINE explicit ForwardObjectAdapter(const FixupVisitor* visitor) : visitor_(visitor) {}
 
   template <typename T>
   ALWAYS_INLINE T* operator()(T* src) const {
@@ -893,7 +893,7 @@ class ForwardObjectAdapter {
 
 class ForwardCodeAdapter {
  public:
-  ALWAYS_INLINE ForwardCodeAdapter(const FixupVisitor* visitor)
+  ALWAYS_INLINE explicit ForwardCodeAdapter(const FixupVisitor* visitor)
       : visitor_(visitor) {}
 
   template <typename T>
@@ -914,10 +914,26 @@ class FixupArtMethodVisitor : public FixupVisitor, public ArtMethodVisitor {
         pointer_size_(pointer_size) {}
 
   virtual void Visit(ArtMethod* method) NO_THREAD_SAFETY_ANALYSIS {
-    if (fixup_heap_objects_) {
-      method->UpdateObjectsForImageRelocation(ForwardObjectAdapter(this), pointer_size_);
+    // TODO: Separate visitor for runtime vs normal methods.
+    if (UNLIKELY(method->IsRuntimeMethod())) {
+      ImtConflictTable* table = method->GetImtConflictTable(pointer_size_);
+      if (table != nullptr) {
+        ImtConflictTable* new_table = ForwardObject(table);
+        if (table != new_table) {
+          method->SetImtConflictTable(new_table, pointer_size_);
+        }
+      }
+      const void* old_code = method->GetEntryPointFromQuickCompiledCodePtrSize(pointer_size_);
+      const void* new_code = ForwardCode(old_code);
+      if (old_code != new_code) {
+        method->SetEntryPointFromQuickCompiledCodePtrSize(new_code, pointer_size_);
+      }
+    } else {
+      if (fixup_heap_objects_) {
+        method->UpdateObjectsForImageRelocation(ForwardObjectAdapter(this), pointer_size_);
+      }
+      method->UpdateEntrypoints<kWithoutReadBarrier>(ForwardCodeAdapter(this), pointer_size_);
     }
-    method->UpdateEntrypoints<kWithoutReadBarrier>(ForwardCodeAdapter(this), pointer_size_);
   }
 
  private:
@@ -1018,6 +1034,7 @@ static bool RelocateInPlace(ImageHeader& image_header,
   const ImageSection& objects_section = image_header.GetImageSection(ImageHeader::kSectionObjects);
   uintptr_t objects_begin = reinterpret_cast<uintptr_t>(target_base + objects_section.Offset());
   uintptr_t objects_end = reinterpret_cast<uintptr_t>(target_base + objects_section.End());
+  FixupObjectAdapter fixup_adapter(boot_image, boot_oat, app_image, app_oat);
   if (fixup_image) {
     // Two pass approach, fix up all classes first, then fix up non class-objects.
     // The visited bitmap is used to ensure that pointer arrays are not forwarded twice.
@@ -1037,7 +1054,6 @@ static bool RelocateInPlace(ImageHeader& image_header,
     ScopedObjectAccess soa(Thread::Current());
     timing.NewTiming("Fixup objects");
     bitmap->VisitMarkedRange(objects_begin, objects_end, fixup_object_visitor);
-    FixupObjectAdapter fixup_adapter(boot_image, boot_oat, app_image, app_oat);
     // Fixup image roots.
     CHECK(app_image.InSource(reinterpret_cast<uintptr_t>(
         image_header.GetImageRoots<kWithoutReadBarrier>())));
@@ -1104,19 +1120,18 @@ static bool RelocateInPlace(ImageHeader& image_header,
                                          boot_oat,
                                          app_image,
                                          app_oat);
-    image_header.GetImageSection(ImageHeader::kSectionArtMethods).VisitPackedArtMethods(
-        &method_visitor,
-        target_base,
-        pointer_size);
+    image_header.VisitPackedArtMethods(&method_visitor, target_base, pointer_size);
   }
   if (fixup_image) {
     {
       // Only touches objects in the app image, no need for mutator lock.
       TimingLogger::ScopedTiming timing("Fixup fields", &logger);
       FixupArtFieldVisitor field_visitor(boot_image, boot_oat, app_image, app_oat);
-      image_header.GetImageSection(ImageHeader::kSectionArtFields).VisitPackedArtFields(
-          &field_visitor,
-          target_base);
+      image_header.VisitPackedArtFields(&field_visitor, target_base);
+    }
+    {
+      TimingLogger::ScopedTiming timing("Fixup conflict tables", &logger);
+      image_header.VisitPackedImtConflictTables(fixup_adapter, target_base, pointer_size);
     }
     // In the app image case, the image methods are actually in the boot image.
     image_header.RelocateImageMethods(boot_image.Delta());
diff --git a/runtime/globals.h b/runtime/globals.h
index e7ea6f378..477cbdf5d 100644
--- a/runtime/globals.h
+++ b/runtime/globals.h
@@ -51,11 +51,31 @@ static constexpr bool kIsDebugBuild = false;
 static constexpr bool kIsDebugBuild = true;
 #endif
 
-// Whether or not this is a target (vs host) build. Useful in conditionals where ART_TARGET isn't.
+// ART_TARGET - Defined for target builds of ART.
+// ART_TARGET_LINUX - Defined for target Linux builds of ART.
+// ART_TARGET_ANDROID - Defined for target Android builds of ART.
+// Note: Either ART_TARGET_LINUX or ART_TARGET_ANDROID need to be set when ART_TARGET is set.
+// Note: When ART_TARGET_LINUX is defined mem_map.h will not be using Ashmem for memory mappings
+// (usually only available on Android kernels).
 #if defined(ART_TARGET)
+// Useful in conditionals where ART_TARGET isn't.
 static constexpr bool kIsTargetBuild = true;
+#if defined(ART_TARGET_LINUX)
+static constexpr bool kIsTargetLinux = true;
+#elif defined(ART_TARGET_ANDROID)
+static constexpr bool kIsTargetLinux = false;
+#else
+#error "Either ART_TARGET_LINUX or ART_TARGET_ANDROID needs to be defined for target builds."
+#endif
 #else
 static constexpr bool kIsTargetBuild = false;
+#if defined(ART_TARGET_LINUX)
+#error "ART_TARGET_LINUX defined for host build."
+#elif defined(ART_TARGET_ANDROID)
+#error "ART_TARGET_ANDROID defined for host build."
+#else
+static constexpr bool kIsTargetLinux = false;
+#endif
 #endif
 
 // Garbage collector constants.
diff --git a/runtime/hprof/hprof.cc b/runtime/hprof/hprof.cc
index 3885c605b..989539516 100644
--- a/runtime/hprof/hprof.cc
+++ b/runtime/hprof/hprof.cc
@@ -505,6 +505,7 @@ class Hprof : public SingleRootVisitor {
     // Walk the roots and the heap.
     output_->StartNewRecord(HPROF_TAG_HEAP_DUMP_SEGMENT, kHprofTime);
 
+    simple_roots_.clear();
     runtime->VisitRoots(this);
     runtime->VisitImageRoots(this);
     runtime->GetHeap()->VisitObjectsPaused(VisitObjectCallback, this);
@@ -884,6 +885,14 @@ class Hprof : public SingleRootVisitor {
                      gc::EqAllocRecordTypesPtr<gc::AllocRecordStackTraceElement>> frames_;
   std::unordered_map<const mirror::Object*, const gc::AllocRecordStackTrace*> allocation_records_;
 
+  // Set used to keep track of what simple root records we have already
+  // emitted, to avoid emitting duplicate entries. The simple root records are
+  // those that contain no other information than the root type and the object
+  // id. A pair of root type and object id is packed into a uint64_t, with
+  // the root type in the upper 32 bits and the object id in the lower 32
+  // bits.
+  std::unordered_set<uint64_t> simple_roots_;
+
   friend class GcRootVisitor;
   DISALLOW_COPY_AND_ASSIGN(Hprof);
 };
@@ -962,10 +971,14 @@ void Hprof::MarkRootObject(const mirror::Object* obj, jobject jni_obj, HprofHeap
     case HPROF_ROOT_MONITOR_USED:
     case HPROF_ROOT_INTERNED_STRING:
     case HPROF_ROOT_DEBUGGER:
-    case HPROF_ROOT_VM_INTERNAL:
-      __ AddU1(heap_tag);
-      __ AddObjectId(obj);
+    case HPROF_ROOT_VM_INTERNAL: {
+      uint64_t key = (static_cast<uint64_t>(heap_tag) << 32) | PointerToLowMemUInt32(obj);
+      if (simple_roots_.insert(key).second) {
+        __ AddU1(heap_tag);
+        __ AddObjectId(obj);
+      }
       break;
+    }
 
       // ID: object ID
       // ID: JNI global ref ID
diff --git a/runtime/image-inl.h b/runtime/image-inl.h
index e3307d87b..ea75a622c 100644
--- a/runtime/image-inl.h
+++ b/runtime/image-inl.h
@@ -19,6 +19,8 @@
 
 #include "image.h"
 
+#include "art_method.h"
+
 namespace art {
 
 template <ReadBarrierOption kReadBarrierOption>
@@ -42,6 +44,20 @@ inline mirror::ObjectArray<mirror::Object>* ImageHeader::GetImageRoots() const {
   return image_roots;
 }
 
+template <typename Visitor>
+inline void ImageHeader::VisitPackedImtConflictTables(const Visitor& visitor,
+                                                      uint8_t* base,
+                                                      size_t pointer_size) const {
+  const ImageSection& section = GetImageSection(kSectionIMTConflictTables);
+  for (size_t pos = 0; pos < section.Size(); ) {
+    auto* table = reinterpret_cast<ImtConflictTable*>(base + section.Offset() + pos);
+    table->Visit([&visitor](const std::pair<ArtMethod*, ArtMethod*>& methods) {
+      return std::make_pair(visitor(methods.first), visitor(methods.second));
+    }, pointer_size);
+    pos += table->ComputeSize(pointer_size);
+  }
+}
+
 }  // namespace art
 
 #endif  // ART_RUNTIME_IMAGE_INL_H_
diff --git a/runtime/image.cc b/runtime/image.cc
index 1f54e3e6a..a9552c27d 100644
--- a/runtime/image.cc
+++ b/runtime/image.cc
@@ -24,7 +24,7 @@
 namespace art {
 
 const uint8_t ImageHeader::kImageMagic[] = { 'a', 'r', 't', '\n' };
-const uint8_t ImageHeader::kImageVersion[] = { '0', '2', '7', '\0' };
+const uint8_t ImageHeader::kImageVersion[] = { '0', '2', '9', '\0' };
 
 ImageHeader::ImageHeader(uint32_t image_begin,
                          uint32_t image_size,
@@ -147,9 +147,10 @@ std::ostream& operator<<(std::ostream& os, const ImageSection& section) {
   return os << "size=" << section.Size() << " range=" << section.Offset() << "-" << section.End();
 }
 
-void ImageSection::VisitPackedArtFields(ArtFieldVisitor* visitor, uint8_t* base) const {
-  for (size_t pos = 0; pos < Size(); ) {
-    auto* array = reinterpret_cast<LengthPrefixedArray<ArtField>*>(base + Offset() + pos);
+void ImageHeader::VisitPackedArtFields(ArtFieldVisitor* visitor, uint8_t* base) const {
+  const ImageSection& fields = GetFieldsSection();
+  for (size_t pos = 0; pos < fields.Size(); ) {
+    auto* array = reinterpret_cast<LengthPrefixedArray<ArtField>*>(base + fields.Offset() + pos);
     for (size_t i = 0; i < array->size(); ++i) {
       visitor->Visit(&array->At(i, sizeof(ArtField)));
     }
@@ -157,18 +158,25 @@ void ImageSection::VisitPackedArtFields(ArtFieldVisitor* visitor, uint8_t* base)
   }
 }
 
-void ImageSection::VisitPackedArtMethods(ArtMethodVisitor* visitor,
-                                         uint8_t* base,
-                                         size_t pointer_size) const {
+void ImageHeader::VisitPackedArtMethods(ArtMethodVisitor* visitor,
+                                        uint8_t* base,
+                                        size_t pointer_size) const {
   const size_t method_alignment = ArtMethod::Alignment(pointer_size);
   const size_t method_size = ArtMethod::Size(pointer_size);
-  for (size_t pos = 0; pos < Size(); ) {
-    auto* array = reinterpret_cast<LengthPrefixedArray<ArtMethod>*>(base + Offset() + pos);
+  const ImageSection& methods = GetMethodsSection();
+  for (size_t pos = 0; pos < methods.Size(); ) {
+    auto* array = reinterpret_cast<LengthPrefixedArray<ArtMethod>*>(base + methods.Offset() + pos);
     for (size_t i = 0; i < array->size(); ++i) {
       visitor->Visit(&array->At(i, method_size, method_alignment));
     }
     pos += array->ComputeSize(array->size(), method_size, method_alignment);
   }
+  const ImageSection& runtime_methods = GetRuntimeMethodsSection();
+  for (size_t pos = 0; pos < runtime_methods.Size(); ) {
+    auto* method = reinterpret_cast<ArtMethod*>(base + runtime_methods.Offset() + pos);
+    visitor->Visit(method);
+    pos += method_size;
+  }
 }
 
 }  // namespace art
diff --git a/runtime/image.h b/runtime/image.h
index 8e5dbad57..2ea9af772 100644
--- a/runtime/image.h
+++ b/runtime/image.h
@@ -64,12 +64,6 @@ class PACKED(4) ImageSection {
     return offset - offset_ < size_;
   }
 
-  // Visit ArtMethods in the section starting at base.
-  void VisitPackedArtMethods(ArtMethodVisitor* visitor, uint8_t* base, size_t pointer_size) const;
-
-  // Visit ArtMethods in the section starting at base.
-  void VisitPackedArtFields(ArtFieldVisitor* visitor, uint8_t* base) const;
-
  private:
   uint32_t offset_;
   uint32_t size_;
@@ -200,6 +194,8 @@ class PACKED(4) ImageHeader {
     kSectionObjects,
     kSectionArtFields,
     kSectionArtMethods,
+    kSectionRuntimeMethods,
+    kSectionIMTConflictTables,
     kSectionDexCacheArrays,
     kSectionInternedStrings,
     kSectionClassTable,
@@ -211,10 +207,19 @@ class PACKED(4) ImageHeader {
   void SetImageMethod(ImageMethod index, ArtMethod* method);
 
   const ImageSection& GetImageSection(ImageSections index) const;
+
   const ImageSection& GetMethodsSection() const {
     return GetImageSection(kSectionArtMethods);
   }
 
+  const ImageSection& GetRuntimeMethodsSection() const {
+    return GetImageSection(kSectionRuntimeMethods);
+  }
+
+  const ImageSection& GetFieldsSection() const {
+    return GetImageSection(ImageHeader::kSectionArtFields);
+  }
+
   template <ReadBarrierOption kReadBarrierOption = kWithReadBarrier>
   mirror::Object* GetImageRoot(ImageRoot image_root) const
       SHARED_REQUIRES(Locks::mutator_lock_);
@@ -265,6 +270,19 @@ class PACKED(4) ImageHeader {
     return boot_image_size_ != 0u;
   }
 
+  // Visit ArtMethods in the section starting at base. Includes runtime methods.
+  // TODO: Delete base parameter if it is always equal to GetImageBegin.
+  void VisitPackedArtMethods(ArtMethodVisitor* visitor, uint8_t* base, size_t pointer_size) const;
+
+  // Visit ArtMethods in the section starting at base.
+  // TODO: Delete base parameter if it is always equal to GetImageBegin.
+  void VisitPackedArtFields(ArtFieldVisitor* visitor, uint8_t* base) const;
+
+  template <typename Visitor>
+  void VisitPackedImtConflictTables(const Visitor& visitor,
+                                    uint8_t* base,
+                                    size_t pointer_size) const;
+
  private:
   static const uint8_t kImageMagic[4];
   static const uint8_t kImageVersion[4];
diff --git a/runtime/instrumentation.cc b/runtime/instrumentation.cc
index 34bc45857..61119f849 100644
--- a/runtime/instrumentation.cc
+++ b/runtime/instrumentation.cc
@@ -687,8 +687,7 @@ void Instrumentation::ResetQuickAllocEntryPoints() {
   }
 }
 
-void Instrumentation::UpdateMethodsCode(ArtMethod* method, const void* quick_code) {
-  DCHECK(method->GetDeclaringClass()->IsResolved());
+void Instrumentation::UpdateMethodsCodeImpl(ArtMethod* method, const void* quick_code) {
   const void* new_quick_code;
   if (LIKELY(!instrumentation_stubs_installed_)) {
     new_quick_code = quick_code;
@@ -710,6 +709,18 @@ void Instrumentation::UpdateMethodsCode(ArtMethod* method, const void* quick_cod
   UpdateEntrypoints(method, new_quick_code);
 }
 
+void Instrumentation::UpdateMethodsCode(ArtMethod* method, const void* quick_code) {
+  DCHECK(method->GetDeclaringClass()->IsResolved());
+  UpdateMethodsCodeImpl(method, quick_code);
+}
+
+void Instrumentation::UpdateMethodsCodeFromDebugger(ArtMethod* method, const void* quick_code) {
+  // When debugger attaches, we may update the entry points of all methods of a class
+  // to the interpreter bridge. A method's declaring class might not be in resolved
+  // state yet in that case.
+  UpdateMethodsCodeImpl(method, quick_code);
+}
+
 bool Instrumentation::AddDeoptimizedMethod(ArtMethod* method) {
   if (IsDeoptimizedMethod(method)) {
     // Already in the map. Return.
diff --git a/runtime/instrumentation.h b/runtime/instrumentation.h
index a4c3d4153..ce6ead453 100644
--- a/runtime/instrumentation.h
+++ b/runtime/instrumentation.h
@@ -227,6 +227,10 @@ class Instrumentation {
   void UpdateMethodsCode(ArtMethod* method, const void* quick_code)
       SHARED_REQUIRES(Locks::mutator_lock_) REQUIRES(!deoptimized_methods_lock_);
 
+  // Update the code of a method respecting any installed stubs from debugger.
+  void UpdateMethodsCodeFromDebugger(ArtMethod* method, const void* quick_code)
+      SHARED_REQUIRES(Locks::mutator_lock_) REQUIRES(!deoptimized_methods_lock_);
+
   // Get the quick code for the given method. More efficient than asking the class linker as it
   // will short-cut to GetCode if instrumentation and static method resolution stubs aren't
   // installed.
@@ -493,6 +497,9 @@ class Instrumentation {
       SHARED_REQUIRES(Locks::mutator_lock_, deoptimized_methods_lock_);
   bool IsDeoptimizedMethodsEmpty() const
       SHARED_REQUIRES(Locks::mutator_lock_, deoptimized_methods_lock_);
+  void UpdateMethodsCodeImpl(ArtMethod* method, const void* quick_code)
+      SHARED_REQUIRES(Locks::mutator_lock_) REQUIRES(!deoptimized_methods_lock_);
+
 
   // Have we hijacked ArtMethod::code_ so that it calls instrumentation/interpreter code?
   bool instrumentation_stubs_installed_;
diff --git a/runtime/interpreter/interpreter.cc b/runtime/interpreter/interpreter.cc
index 97dbe5d21..1d0e600e4 100644
--- a/runtime/interpreter/interpreter.cc
+++ b/runtime/interpreter/interpreter.cc
@@ -264,12 +264,12 @@ JValue ExecuteGotoImpl<false, true>(Thread* self, const DexFile::CodeItem* code_
                                     ShadowFrame& shadow_frame, JValue result_register);
 #endif
 
-static JValue Execute(Thread* self, const DexFile::CodeItem* code_item, ShadowFrame& shadow_frame,
-                      JValue result_register)
-    SHARED_REQUIRES(Locks::mutator_lock_);
-
-static inline JValue Execute(Thread* self, const DexFile::CodeItem* code_item,
-                             ShadowFrame& shadow_frame, JValue result_register) {
+static inline JValue Execute(
+    Thread* self,
+    const DexFile::CodeItem* code_item,
+    ShadowFrame& shadow_frame,
+    JValue result_register,
+    bool stay_in_interpreter = false) SHARED_REQUIRES(Locks::mutator_lock_) {
   DCHECK(!shadow_frame.GetMethod()->IsAbstract());
   DCHECK(!shadow_frame.GetMethod()->IsNative());
   if (LIKELY(shadow_frame.GetDexPC() == 0)) {  // Entering the method, but not via deoptimization.
@@ -284,27 +284,34 @@ static inline JValue Execute(Thread* self, const DexFile::CodeItem* code_item,
                                         method, 0);
     }
 
-    jit::Jit* jit = Runtime::Current()->GetJit();
-    if (jit != nullptr) {
-      jit->MethodEntered(self, shadow_frame.GetMethod());
-      if (jit->CanInvokeCompiledCode(method)) {
-        JValue result;
+    if (!stay_in_interpreter) {
+      jit::Jit* jit = Runtime::Current()->GetJit();
+      if (jit != nullptr) {
+        jit->MethodEntered(self, shadow_frame.GetMethod());
+        if (jit->CanInvokeCompiledCode(method)) {
+          JValue result;
 
-        // Pop the shadow frame before calling into compiled code.
-        self->PopShadowFrame();
-        ArtInterpreterToCompiledCodeBridge(self, nullptr, code_item, &shadow_frame, &result);
-        // Push the shadow frame back as the caller will expect it.
-        self->PushShadowFrame(&shadow_frame);
+          // Pop the shadow frame before calling into compiled code.
+          self->PopShadowFrame();
+          ArtInterpreterToCompiledCodeBridge(self, nullptr, code_item, &shadow_frame, &result);
+          // Push the shadow frame back as the caller will expect it.
+          self->PushShadowFrame(&shadow_frame);
 
-        return result;
+          return result;
+        }
       }
     }
   }
 
   shadow_frame.GetMethod()->GetDeclaringClass()->AssertInitializedOrInitializingInThread(self);
 
+  // Lock counting is a special version of accessibility checks, and for simplicity and
+  // reduction of template parameters, we gate it behind access-checks mode.
+  ArtMethod* method = shadow_frame.GetMethod();
+  DCHECK(!method->SkipAccessChecks() || !method->MustCountLocks());
+
   bool transaction_active = Runtime::Current()->IsActiveTransaction();
-  if (LIKELY(shadow_frame.GetMethod()->SkipAccessChecks())) {
+  if (LIKELY(method->SkipAccessChecks())) {
     // Enter the "without access check" interpreter.
     if (kInterpreterImplKind == kMterpImplKind) {
       if (transaction_active) {
@@ -382,7 +389,8 @@ static inline JValue Execute(Thread* self, const DexFile::CodeItem* code_item,
 }
 
 void EnterInterpreterFromInvoke(Thread* self, ArtMethod* method, Object* receiver,
-                                uint32_t* args, JValue* result) {
+                                uint32_t* args, JValue* result,
+                                bool stay_in_interpreter) {
   DCHECK_EQ(self, Thread::Current());
   bool implicit_check = !Runtime::Current()->ExplicitStackOverflowChecks();
   if (UNLIKELY(__builtin_frame_address(0) < self->GetStackEndForInterpreter(implicit_check))) {
@@ -457,7 +465,7 @@ void EnterInterpreterFromInvoke(Thread* self, ArtMethod* method, Object* receive
     }
   }
   if (LIKELY(!method->IsNative())) {
-    JValue r = Execute(self, code_item, *shadow_frame, JValue());
+    JValue r = Execute(self, code_item, *shadow_frame, JValue(), stay_in_interpreter);
     if (result != nullptr) {
       *result = r;
     }
@@ -487,6 +495,10 @@ void EnterInterpreterFromDeoptimize(Thread* self,
   // Are we executing the first shadow frame?
   bool first = true;
   while (shadow_frame != nullptr) {
+    // We do not want to recover lock state for lock counting when deoptimizing. Currently,
+    // the compiler should not have compiled a method that failed structured-locking checks.
+    DCHECK(!shadow_frame->GetMethod()->MustCountLocks());
+
     self->SetTopOfShadowStack(shadow_frame);
     const DexFile::CodeItem* code_item = shadow_frame->GetMethod()->GetCodeItem();
     const uint32_t dex_pc = shadow_frame->GetDexPC();
@@ -506,8 +518,24 @@ void EnterInterpreterFromDeoptimize(Thread* self,
       // instruction, as it already executed.
       // TODO: should be tested more once b/17586779 is fixed.
       const Instruction* instr = Instruction::At(&code_item->insns_[dex_pc]);
-      DCHECK(instr->IsInvoke());
-      new_dex_pc = dex_pc + instr->SizeInCodeUnits();
+      if (instr->IsInvoke()) {
+        new_dex_pc = dex_pc + instr->SizeInCodeUnits();
+      } else if (instr->Opcode() == Instruction::NEW_INSTANCE) {
+        // It's possible to deoptimize at a NEW_INSTANCE dex instruciton that's for a
+        // java string, which is turned into a call into StringFactory.newEmptyString();
+        if (kIsDebugBuild) {
+          ClassLinker* class_linker = Runtime::Current()->GetClassLinker();
+          mirror::Class* klass = class_linker->ResolveType(
+              instr->VRegB_21c(), shadow_frame->GetMethod());
+          DCHECK(klass->IsStringClass());
+        }
+        // Skip the dex instruction since we essentially come back from an invocation.
+        new_dex_pc = dex_pc + instr->SizeInCodeUnits();
+      } else {
+        DCHECK(false) << "Unexpected instruction opcode " << instr->Opcode()
+                      << " at dex_pc " << dex_pc
+                      << " of method: " << PrettyMethod(shadow_frame->GetMethod(), false);
+      }
     } else {
       // Nothing to do, the dex_pc is the one at which the code requested
       // the deoptimization.
diff --git a/runtime/interpreter/interpreter.h b/runtime/interpreter/interpreter.h
index 6353a9b7b..bf4bcff85 100644
--- a/runtime/interpreter/interpreter.h
+++ b/runtime/interpreter/interpreter.h
@@ -33,8 +33,11 @@ class Thread;
 namespace interpreter {
 
 // Called by ArtMethod::Invoke, shadow frames arguments are taken from the args array.
+// The optional stay_in_interpreter parameter (false by default) can be used by clients to
+// explicitly force interpretation in the remaining path that implements method invocation.
 extern void EnterInterpreterFromInvoke(Thread* self, ArtMethod* method,
-                                       mirror::Object* receiver, uint32_t* args, JValue* result)
+                                       mirror::Object* receiver, uint32_t* args, JValue* result,
+                                       bool stay_in_interpreter = false)
     SHARED_REQUIRES(Locks::mutator_lock_);
 
 // 'from_code' denotes whether the deoptimization was explicitly triggered by compiled code.
diff --git a/runtime/interpreter/interpreter_common.h b/runtime/interpreter/interpreter_common.h
index e5b89e2f9..69376fd7a 100644
--- a/runtime/interpreter/interpreter_common.h
+++ b/runtime/interpreter/interpreter_common.h
@@ -95,7 +95,9 @@ static inline void DoMonitorEnter(Thread* self,
   StackHandleScope<1> hs(self);
   Handle<Object> h_ref(hs.NewHandle(ref));
   h_ref->MonitorEnter(self);
-  frame->GetLockCountData().AddMonitor<kMonitorCounting>(self, h_ref.Get());
+  if (kMonitorCounting && frame->GetMethod()->MustCountLocks()) {
+    frame->GetLockCountData().AddMonitor(self, h_ref.Get());
+  }
 }
 
 template <bool kMonitorCounting>
@@ -107,7 +109,19 @@ static inline void DoMonitorExit(Thread* self,
   StackHandleScope<1> hs(self);
   Handle<Object> h_ref(hs.NewHandle(ref));
   h_ref->MonitorExit(self);
-  frame->GetLockCountData().RemoveMonitorOrThrow<kMonitorCounting>(self, h_ref.Get());
+  if (kMonitorCounting && frame->GetMethod()->MustCountLocks()) {
+    frame->GetLockCountData().RemoveMonitorOrThrow(self, h_ref.Get());
+  }
+}
+
+template <bool kMonitorCounting>
+static inline bool DoMonitorCheckOnExit(Thread* self, ShadowFrame* frame)
+    NO_THREAD_SAFETY_ANALYSIS
+    REQUIRES(!Roles::uninterruptible_) {
+  if (kMonitorCounting && frame->GetMethod()->MustCountLocks()) {
+    return frame->GetLockCountData().CheckAllMonitorsReleasedOrThrow(self);
+  }
+  return true;
 }
 
 void AbortTransactionF(Thread* self, const char* fmt, ...)
diff --git a/runtime/interpreter/interpreter_goto_table_impl.cc b/runtime/interpreter/interpreter_goto_table_impl.cc
index 13cfb9877..f03036b6a 100644
--- a/runtime/interpreter/interpreter_goto_table_impl.cc
+++ b/runtime/interpreter/interpreter_goto_table_impl.cc
@@ -104,8 +104,7 @@ namespace interpreter {
   } HANDLE_INSTRUCTION_END();
 
 #define HANDLE_MONITOR_CHECKS()                                                                   \
-  if (!shadow_frame.GetLockCountData().                                                           \
-          CheckAllMonitorsReleasedOrThrow<do_assignability_check>(self)) {                        \
+  if (!DoMonitorCheckOnExit<do_assignability_check>(self, &shadow_frame)) {                       \
     HANDLE_PENDING_EXCEPTION();                                                                   \
   }
 
@@ -2584,7 +2583,7 @@ JValue ExecuteGotoImpl(Thread* self, const DexFile::CodeItem* code_item, ShadowF
                                                                   instrumentation);
     if (found_dex_pc == DexFile::kDexNoIndex) {
       // Structured locking is to be enforced for abnormal termination, too.
-      shadow_frame.GetLockCountData().CheckAllMonitorsReleasedOrThrow<do_assignability_check>(self);
+      DoMonitorCheckOnExit<do_assignability_check>(self, &shadow_frame);
       return JValue(); /* Handled in caller. */
     } else {
       int32_t displacement = static_cast<int32_t>(found_dex_pc) - static_cast<int32_t>(dex_pc);
diff --git a/runtime/interpreter/interpreter_switch_impl.cc b/runtime/interpreter/interpreter_switch_impl.cc
index 4323d4f42..18330babe 100644
--- a/runtime/interpreter/interpreter_switch_impl.cc
+++ b/runtime/interpreter/interpreter_switch_impl.cc
@@ -34,8 +34,7 @@ namespace interpreter {
                                                                   instrumentation);             \
     if (found_dex_pc == DexFile::kDexNoIndex) {                                                 \
       /* Structured locking is to be enforced for abnormal termination, too. */                 \
-      shadow_frame.GetLockCountData().                                                          \
-          CheckAllMonitorsReleasedOrThrow<do_assignability_check>(self);                        \
+      DoMonitorCheckOnExit<do_assignability_check>(self, &shadow_frame);                        \
       if (interpret_one_instruction) {                                                          \
         /* Signal mterp to return to caller */                                                  \
         shadow_frame.SetDexPC(DexFile::kDexNoIndex);                                            \
@@ -57,8 +56,7 @@ namespace interpreter {
   } while (false)
 
 #define HANDLE_MONITOR_CHECKS()                                                                   \
-  if (!shadow_frame.GetLockCountData().                                                           \
-          CheckAllMonitorsReleasedOrThrow<do_assignability_check>(self)) {                        \
+  if (!DoMonitorCheckOnExit<do_assignability_check>(self, &shadow_frame)) {                       \
     HANDLE_PENDING_EXCEPTION();                                                                   \
   }
 
diff --git a/runtime/interpreter/mterp/out/mterp_x86.S b/runtime/interpreter/mterp/out/mterp_x86.S
index e46f9cde9..f78e1bc41 100644
--- a/runtime/interpreter/mterp/out/mterp_x86.S
+++ b/runtime/interpreter/mterp/out/mterp_x86.S
@@ -12985,6 +12985,7 @@ MterpCommonTakenBranch:
  * not-taken path.  All Dalvik not-taken conditional branch offsets are 2.
  */
 .L_check_not_taken_osr:
+    EXPORT_PC
     movl    rSELF, %eax
     movl    %eax, OUT_ARG0(%esp)
     leal    OFF_FP_SHADOWFRAME(rFP), %ecx
diff --git a/runtime/interpreter/mterp/out/mterp_x86_64.S b/runtime/interpreter/mterp/out/mterp_x86_64.S
index 62dce6e77..031cec823 100644
--- a/runtime/interpreter/mterp/out/mterp_x86_64.S
+++ b/runtime/interpreter/mterp/out/mterp_x86_64.S
@@ -11961,6 +11961,7 @@ MterpCommonTakenBranch:
  * not-taken path.  All Dalvik not-taken conditional branch offsets are 2.
  */
 .L_check_not_taken_osr:
+    EXPORT_PC
     movq    rSELF, OUT_ARG0
     leaq    OFF_FP_SHADOWFRAME(rFP), OUT_ARG1
     movl    $2, OUT_32_ARG2
diff --git a/runtime/interpreter/mterp/x86/footer.S b/runtime/interpreter/mterp/x86/footer.S
index fa03e78fe..e8c8ca8d7 100644
--- a/runtime/interpreter/mterp/x86/footer.S
+++ b/runtime/interpreter/mterp/x86/footer.S
@@ -234,6 +234,7 @@ MterpCommonTakenBranch:
  * not-taken path.  All Dalvik not-taken conditional branch offsets are 2.
  */
 .L_check_not_taken_osr:
+    EXPORT_PC
     movl    rSELF, %eax
     movl    %eax, OUT_ARG0(%esp)
     leal    OFF_FP_SHADOWFRAME(rFP), %ecx
diff --git a/runtime/interpreter/mterp/x86_64/footer.S b/runtime/interpreter/mterp/x86_64/footer.S
index 54d0cb1ce..f78f16357 100644
--- a/runtime/interpreter/mterp/x86_64/footer.S
+++ b/runtime/interpreter/mterp/x86_64/footer.S
@@ -213,6 +213,7 @@ MterpCommonTakenBranch:
  * not-taken path.  All Dalvik not-taken conditional branch offsets are 2.
  */
 .L_check_not_taken_osr:
+    EXPORT_PC
     movq    rSELF, OUT_ARG0
     leaq    OFF_FP_SHADOWFRAME(rFP), OUT_ARG1
     movl    $$2, OUT_32_ARG2
diff --git a/runtime/java_vm_ext.cc b/runtime/java_vm_ext.cc
index a41fd4504..d983a9fa1 100644
--- a/runtime/java_vm_ext.cc
+++ b/runtime/java_vm_ext.cc
@@ -318,6 +318,7 @@ class JII {
     }
     JavaVMExt* raw_vm = reinterpret_cast<JavaVMExt*>(vm);
     delete raw_vm->GetRuntime();
+    android::ResetNativeLoader();
     return JNI_OK;
   }
 
@@ -942,6 +943,11 @@ extern "C" jint JNI_CreateJavaVM(JavaVM** p_vm, JNIEnv** p_env, void* vm_args) {
   if (!Runtime::Create(options, ignore_unrecognized)) {
     return JNI_ERR;
   }
+
+  // Initialize native loader. This step makes sure we have
+  // everything set up before we start using JNI.
+  android::InitializeNativeLoader();
+
   Runtime* runtime = Runtime::Current();
   bool started = runtime->Start();
   if (!started) {
@@ -950,6 +956,7 @@ extern "C" jint JNI_CreateJavaVM(JavaVM** p_vm, JNIEnv** p_env, void* vm_args) {
     LOG(WARNING) << "CreateJavaVM failed";
     return JNI_ERR;
   }
+
   *p_env = Thread::Current()->GetJniEnv();
   *p_vm = runtime->GetJavaVM();
   return JNI_OK;
diff --git a/runtime/jit/jit.cc b/runtime/jit/jit.cc
index c36543f1f..b6b7eb184 100644
--- a/runtime/jit/jit.cc
+++ b/runtime/jit/jit.cc
@@ -51,7 +51,7 @@ bool Jit::generate_debug_info_ = false;
 
 JitOptions* JitOptions::CreateFromRuntimeArguments(const RuntimeArgumentMap& options) {
   auto* jit_options = new JitOptions;
-  jit_options->use_jit_ = options.GetOrDefault(RuntimeArgumentMap::UseJIT);
+  jit_options->use_jit_compilation_ = options.GetOrDefault(RuntimeArgumentMap::UseJitCompilation);
 
   jit_options->code_cache_initial_capacity_ =
       options.GetOrDefault(RuntimeArgumentMap::JITCodeCacheInitialCapacity);
@@ -102,14 +102,26 @@ JitOptions* JitOptions::CreateFromRuntimeArguments(const RuntimeArgumentMap& opt
         static_cast<size_t>(1));
   }
 
+  if (options.Exists(RuntimeArgumentMap::JITInvokeTransitionWeight)) {
+    jit_options->invoke_transition_weight_ =
+        *options.Get(RuntimeArgumentMap::JITInvokeTransitionWeight);
+    if (jit_options->invoke_transition_weight_ > jit_options->warmup_threshold_) {
+      LOG(FATAL) << "Invoke transition weight is above the warmup threshold.";
+    } else if (jit_options->invoke_transition_weight_  == 0) {
+      LOG(FATAL) << "Invoke transition weight cannot be 0.";
+    }
+  } else {
+    jit_options->invoke_transition_weight_ = std::max(
+        jit_options->warmup_threshold_ / Jit::kDefaultInvokeTransitionWeightRatio,
+        static_cast<size_t>(1));;
+  }
+
   return jit_options;
 }
 
 bool Jit::ShouldUsePriorityThreadWeight() {
-  // TODO(calin): verify that IsSensitiveThread covers only the cases we are interested on.
-  // In particular if apps can set StrictMode policies for any of their threads, case in which
-  // we need to find another way to track sensitive threads.
-  return Runtime::Current()->InJankPerceptibleProcessState() && Thread::IsSensitiveThread();
+  return Runtime::Current()->InJankPerceptibleProcessState()
+      && Thread::Current()->IsJitSensitiveThread();
 }
 
 void Jit::DumpInfo(std::ostream& os) {
@@ -132,9 +144,11 @@ Jit::Jit() : dump_info_on_shutdown_(false),
              cumulative_timings_("JIT timings"),
              memory_use_("Memory used for compilation", 16),
              lock_("JIT memory use lock"),
+             use_jit_compilation_(true),
              save_profiling_info_(false) {}
 
 Jit* Jit::Create(JitOptions* options, std::string* error_msg) {
+  DCHECK(options->UseJitCompilation() || options->GetSaveProfilingInfo());
   std::unique_ptr<Jit> jit(new Jit);
   jit->dump_info_on_shutdown_ = options->DumpJitInfoOnShutdown();
   if (jit_compiler_handle_ == nullptr && !LoadCompiler(error_msg)) {
@@ -148,6 +162,7 @@ Jit* Jit::Create(JitOptions* options, std::string* error_msg) {
   if (jit->GetCodeCache() == nullptr) {
     return nullptr;
   }
+  jit->use_jit_compilation_ = options->UseJitCompilation();
   jit->save_profiling_info_ = options->GetSaveProfilingInfo();
   VLOG(jit) << "JIT created with initial_capacity="
       << PrettySize(options->GetCodeCacheInitialCapacity())
@@ -160,8 +175,7 @@ Jit* Jit::Create(JitOptions* options, std::string* error_msg) {
   jit->warm_method_threshold_ = options->GetWarmupThreshold();
   jit->osr_method_threshold_ = options->GetOsrThreshold();
   jit->priority_thread_weight_ = options->GetPriorityThreadWeight();
-  jit->transition_weight_ = std::max(
-      jit->warm_method_threshold_ / kDefaultTransitionRatio, static_cast<size_t>(1));
+  jit->invoke_transition_weight_ = options->GetInvokeTransitionWeight();
 
   jit->CreateThreadPool();
 
@@ -227,6 +241,7 @@ bool Jit::LoadCompiler(std::string* error_msg) {
 }
 
 bool Jit::CompileMethod(ArtMethod* method, Thread* self, bool osr) {
+  DCHECK(Runtime::Current()->UseJitCompilation());
   DCHECK(!method->IsRuntimeMethod());
 
   // Don't compile the method if it has breakpoints.
@@ -331,8 +346,12 @@ Jit::~Jit() {
 }
 
 void Jit::NewTypeLoadedIfUsingJit(mirror::Class* type) {
+  if (!Runtime::Current()->UseJitCompilation()) {
+    // No need to notify if we only use the JIT to save profiles.
+    return;
+  }
   jit::Jit* jit = Runtime::Current()->GetJit();
-  if (jit != nullptr && jit->generate_debug_info_) {
+  if (jit->generate_debug_info_) {
     DCHECK(jit->jit_types_loaded_ != nullptr);
     jit->jit_types_loaded_(jit->jit_compiler_handle_, &type, 1);
   }
@@ -546,6 +565,7 @@ class JitCompileTask FINAL : public Task {
         VLOG(jit) << "Start profiling " << PrettyMethod(method_);
       }
     }
+    ProfileSaver::NotifyJitActivity();
   }
 
   void Finalize() OVERRIDE {
@@ -606,22 +626,24 @@ void Jit::AddSamples(Thread* self, ArtMethod* method, uint16_t count, bool with_
     }
     // Avoid jumping more than one state at a time.
     new_count = std::min(new_count, hot_method_threshold_ - 1);
-  } else if (starting_count < hot_method_threshold_) {
-    if ((new_count >= hot_method_threshold_) &&
-        !code_cache_->ContainsPc(method->GetEntryPointFromQuickCompiledCode())) {
-      DCHECK(thread_pool_ != nullptr);
-      thread_pool_->AddTask(self, new JitCompileTask(method, JitCompileTask::kCompile));
-    }
-    // Avoid jumping more than one state at a time.
-    new_count = std::min(new_count, osr_method_threshold_ - 1);
-  } else if (starting_count < osr_method_threshold_) {
-    if (!with_backedges) {
-      // If the samples don't contain any back edge, we don't increment the hotness.
-      return;
-    }
-    if ((new_count >= osr_method_threshold_) &&  !code_cache_->IsOsrCompiled(method)) {
-      DCHECK(thread_pool_ != nullptr);
-      thread_pool_->AddTask(self, new JitCompileTask(method, JitCompileTask::kCompileOsr));
+  } else if (use_jit_compilation_) {
+    if (starting_count < hot_method_threshold_) {
+      if ((new_count >= hot_method_threshold_) &&
+          !code_cache_->ContainsPc(method->GetEntryPointFromQuickCompiledCode())) {
+        DCHECK(thread_pool_ != nullptr);
+        thread_pool_->AddTask(self, new JitCompileTask(method, JitCompileTask::kCompile));
+      }
+      // Avoid jumping more than one state at a time.
+      new_count = std::min(new_count, osr_method_threshold_ - 1);
+    } else if (starting_count < osr_method_threshold_) {
+      if (!with_backedges) {
+        // If the samples don't contain any back edge, we don't increment the hotness.
+        return;
+      }
+      if ((new_count >= osr_method_threshold_) &&  !code_cache_->IsOsrCompiled(method)) {
+        DCHECK(thread_pool_ != nullptr);
+        thread_pool_->AddTask(self, new JitCompileTask(method, JitCompileTask::kCompileOsr));
+      }
     }
   }
   // Update hotness counter
@@ -629,7 +651,8 @@ void Jit::AddSamples(Thread* self, ArtMethod* method, uint16_t count, bool with_
 }
 
 void Jit::MethodEntered(Thread* thread, ArtMethod* method) {
-  if (UNLIKELY(Runtime::Current()->GetJit()->JitAtFirstUse())) {
+  Runtime* runtime = Runtime::Current();
+  if (UNLIKELY(runtime->UseJitCompilation() && runtime->GetJit()->JitAtFirstUse())) {
     // The compiler requires a ProfilingInfo object.
     ProfilingInfo::Create(thread, method, /* retry_allocation */ true);
     JitCompileTask compile_task(method, JitCompileTask::kCompile);
diff --git a/runtime/jit/jit.h b/runtime/jit/jit.h
index 8198c18eb..f3a6240e8 100644
--- a/runtime/jit/jit.h
+++ b/runtime/jit/jit.h
@@ -44,7 +44,7 @@ class Jit {
   static constexpr bool kStressMode = kIsDebugBuild;
   static constexpr size_t kDefaultCompileThreshold = kStressMode ? 2 : 10000;
   static constexpr size_t kDefaultPriorityThreadWeightRatio = 1000;
-  static constexpr size_t kDefaultTransitionRatio = 100;
+  static constexpr size_t kDefaultInvokeTransitionWeightRatio = 500;
 
   virtual ~Jit();
   static Jit* Create(JitOptions* options, std::string* error_msg);
@@ -87,6 +87,15 @@ class Jit {
     return priority_thread_weight_;
   }
 
+  // Returns false if we only need to save profile information and not compile methods.
+  bool UseJitCompilation() const {
+    return use_jit_compilation_;
+  }
+
+  bool SaveProfilingInfo() const {
+    return save_profiling_info_;
+  }
+
   // Wait until there is no more pending compilation tasks.
   void WaitForCompilationToFinish(Thread* self);
 
@@ -106,12 +115,12 @@ class Jit {
 
   void NotifyInterpreterToCompiledCodeTransition(Thread* self, ArtMethod* caller)
       SHARED_REQUIRES(Locks::mutator_lock_) {
-    AddSamples(self, caller, transition_weight_, false);
+    AddSamples(self, caller, invoke_transition_weight_, false);
   }
 
   void NotifyCompiledCodeToInterpreterTransition(Thread* self, ArtMethod* callee)
       SHARED_REQUIRES(Locks::mutator_lock_) {
-    AddSamples(self, callee, transition_weight_, false);
+    AddSamples(self, callee, invoke_transition_weight_, false);
   }
 
   // Starts the profile saver if the config options allow profile recording.
@@ -179,13 +188,14 @@ class Jit {
 
   std::unique_ptr<jit::JitCodeCache> code_cache_;
 
+  bool use_jit_compilation_;
   bool save_profiling_info_;
   static bool generate_debug_info_;
   uint16_t hot_method_threshold_;
   uint16_t warm_method_threshold_;
   uint16_t osr_method_threshold_;
   uint16_t priority_thread_weight_;
-  uint16_t transition_weight_;
+  uint16_t invoke_transition_weight_;
   std::unique_ptr<ThreadPool> thread_pool_;
 
   DISALLOW_COPY_AND_ASSIGN(Jit);
@@ -206,6 +216,9 @@ class JitOptions {
   uint16_t GetPriorityThreadWeight() const {
     return priority_thread_weight_;
   }
+  size_t GetInvokeTransitionWeight() const {
+    return invoke_transition_weight_;
+  }
   size_t GetCodeCacheInitialCapacity() const {
     return code_cache_initial_capacity_;
   }
@@ -218,33 +231,34 @@ class JitOptions {
   bool GetSaveProfilingInfo() const {
     return save_profiling_info_;
   }
-  bool UseJIT() const {
-    return use_jit_;
+  bool UseJitCompilation() const {
+    return use_jit_compilation_;
   }
-  void SetUseJIT(bool b) {
-    use_jit_ = b;
+  void SetUseJitCompilation(bool b) {
+    use_jit_compilation_ = b;
   }
   void SetSaveProfilingInfo(bool b) {
     save_profiling_info_ = b;
   }
   void SetJitAtFirstUse() {
-    use_jit_ = true;
+    use_jit_compilation_ = true;
     compile_threshold_ = 0;
   }
 
  private:
-  bool use_jit_;
+  bool use_jit_compilation_;
   size_t code_cache_initial_capacity_;
   size_t code_cache_max_capacity_;
   size_t compile_threshold_;
   size_t warmup_threshold_;
   size_t osr_threshold_;
   uint16_t priority_thread_weight_;
+  size_t invoke_transition_weight_;
   bool dump_info_on_shutdown_;
   bool save_profiling_info_;
 
   JitOptions()
-      : use_jit_(false),
+      : use_jit_compilation_(false),
         code_cache_initial_capacity_(0),
         code_cache_max_capacity_(0),
         compile_threshold_(0),
diff --git a/runtime/jit/offline_profiling_info.cc b/runtime/jit/offline_profiling_info.cc
index a3078bc1b..c99d3636a 100644
--- a/runtime/jit/offline_profiling_info.cc
+++ b/runtime/jit/offline_profiling_info.cc
@@ -645,7 +645,8 @@ std::set<DexCacheResolvedClasses> ProfileCompilationInfo::GetResolvedClasses() c
   for (auto&& pair : info_) {
     const std::string& profile_key = pair.first;
     const DexFileData& data = pair.second;
-    DexCacheResolvedClasses classes(profile_key, data.checksum);
+    // TODO: Is it OK to use the same location for both base and dex location here?
+    DexCacheResolvedClasses classes(profile_key, profile_key, data.checksum);
     classes.AddClasses(data.class_set.begin(), data.class_set.end());
     ret.insert(classes);
   }
diff --git a/runtime/jit/profile_saver.cc b/runtime/jit/profile_saver.cc
index 7a9d2506d..8358ce360 100644
--- a/runtime/jit/profile_saver.cc
+++ b/runtime/jit/profile_saver.cc
@@ -33,14 +33,15 @@ namespace art {
 // TODO: read the constants from ProfileOptions,
 // Add a random delay each time we go to sleep so that we don't hammer the CPU
 // with all profile savers running at the same time.
-static constexpr const uint64_t kRandomDelayMaxMs = 30 * 1000;  // 30 seconds
-static constexpr const uint64_t kMaxBackoffMs = 10 * 60 * 1000;  // 10 minutes
-static constexpr const uint64_t kSavePeriodMs = 20 * 1000;  // 20 seconds
+static constexpr const uint64_t kMinSavePeriodNs = MsToNs(20 * 1000);  // 20 seconds
 static constexpr const uint64_t kSaveResolvedClassesDelayMs = 2 * 1000;  // 2 seconds
-static constexpr const double kBackoffCoef = 2.0;
 
 static constexpr const uint32_t kMinimumNumberOfMethodsToSave = 10;
 static constexpr const uint32_t kMinimumNumberOfClassesToSave = 10;
+static constexpr const uint32_t kMinimumNumberOfNotificationBeforeWake =
+    kMinimumNumberOfMethodsToSave;
+static constexpr const uint32_t kMaximumNumberOfNotificationBeforeWake = 50;
+
 
 ProfileSaver* ProfileSaver::instance_ = nullptr;
 pthread_t ProfileSaver::profiler_pthread_ = 0U;
@@ -55,6 +56,8 @@ ProfileSaver::ProfileSaver(const std::string& output_filename,
       shutting_down_(false),
       last_save_number_of_methods_(0),
       last_save_number_of_classes_(0),
+      last_time_ns_saver_woke_up_(0),
+      jit_activity_notifications_(0),
       wait_lock_("ProfileSaver wait lock"),
       period_condition_("ProfileSaver period condition", wait_lock_),
       total_bytes_written_(0),
@@ -65,7 +68,9 @@ ProfileSaver::ProfileSaver(const std::string& output_filename,
       total_ms_of_sleep_(0),
       total_ns_of_work_(0),
       total_number_of_foreign_dex_marks_(0),
-      max_number_of_profile_entries_cached_(0) {
+      max_number_of_profile_entries_cached_(0),
+      total_number_of_hot_spikes_(0),
+      total_number_of_wake_ups_(0) {
   AddTrackedLocations(output_filename, app_data_dir, code_paths);
   if (!app_data_dir.empty()) {
     // The application directory is used to determine which dex files are owned by app.
@@ -83,55 +88,89 @@ ProfileSaver::ProfileSaver(const std::string& output_filename,
 }
 
 void ProfileSaver::Run() {
-  srand(MicroTime() * getpid());
   Thread* self = Thread::Current();
 
-  uint64_t save_period_ms = kSavePeriodMs;
-  VLOG(profiler) << "Save profiling information every " << save_period_ms << " ms";
-  bool cache_resolved_classes = true;
+  // Fetch the resolved classes for the app images after sleeping for
+  // kSaveResolvedClassesDelayMs.
+  // TODO(calin) This only considers the case of the primary profile file.
+  // Anything that gets loaded in the same VM will not have their resolved
+  // classes save (unless they started before the initial saving was done).
+  {
+    MutexLock mu(self, wait_lock_);
+    period_condition_.TimedWait(self, kSaveResolvedClassesDelayMs, 0);
+    total_ms_of_sleep_ += kSaveResolvedClassesDelayMs;
+  }
+  FetchAndCacheResolvedClasses();
+
+  // Loop for the profiled methods.
   while (!ShuttingDown(self)) {
-    uint64_t sleep_time_ms;
-    if (cache_resolved_classes) {
-      // Sleep less long for the first iteration since we want to record loaded classes shortly
-      // after app launch.
-      sleep_time_ms = kSaveResolvedClassesDelayMs;
-    } else {
-      const uint64_t random_sleep_delay_ms = rand() % kRandomDelayMaxMs;
-      sleep_time_ms = save_period_ms + random_sleep_delay_ms;
-    }
+    uint64_t sleep_start = NanoTime();
     {
       MutexLock mu(self, wait_lock_);
-      period_condition_.TimedWait(self, sleep_time_ms, 0);
+      period_condition_.Wait(self);
+      total_number_of_wake_ups_++;
+      // We might have been woken up by a huge number of notifications to guarantee saving.
+      // If we didn't meet the minimum saving period go back to sleep (only if missed by
+      // a reasonable margin).
+      uint64_t sleep_time = NanoTime() - last_time_ns_saver_woke_up_;
+      while (kMinSavePeriodNs - sleep_time > (kMinSavePeriodNs / 10)) {
+        period_condition_.TimedWait(self, NsToMs(kMinSavePeriodNs - sleep_time), 0);
+        total_number_of_wake_ups_++;
+        sleep_time = NanoTime() - last_time_ns_saver_woke_up_;
+      }
     }
-    total_ms_of_sleep_ += sleep_time_ms;
+    total_ms_of_sleep_ += NsToMs(NanoTime() - sleep_start);
+
     if (ShuttingDown(self)) {
       break;
     }
 
-    uint64_t start = NanoTime();
-    if (cache_resolved_classes) {
-      // TODO(calin) This only considers the case of the primary profile file.
-      // Anything that gets loaded in the same VM will not have their resolved
-      // classes save (unless they started before the initial saving was done).
-      FetchAndCacheResolvedClasses();
-    } else {
-      bool profile_saved_to_disk = ProcessProfilingInfo();
-      if (profile_saved_to_disk) {
-        // Reset the period to the initial value as it's highly likely to JIT again.
-        save_period_ms = kSavePeriodMs;
-        VLOG(profiler) << "Profile saver: saved something, period reset to: " << save_period_ms;
-      } else {
-        // If we don't need to save now it is less likely that we will need to do
-        // so in the future. Increase the time between saves according to the
-        // kBackoffCoef, but make it no larger than kMaxBackoffMs.
-        save_period_ms = std::min(kMaxBackoffMs,
-                                  static_cast<uint64_t>(kBackoffCoef * save_period_ms));
-        VLOG(profiler) << "Profile saver: nothing to save, delaying period to: " << save_period_ms;
-      }
+    uint16_t new_methods = 0;
+    uint64_t start_work = NanoTime();
+    bool profile_saved_to_disk = ProcessProfilingInfo(&new_methods);
+    // Update the notification counter based on result. Note that there might be contention on this
+    // but we don't care about to be 100% precise.
+    if (!profile_saved_to_disk) {
+      // If we didn't save to disk it may be because we didn't have enough new methods.
+      // Set the jit activity notifications to new_methods so we can wake up earlier if needed.
+      jit_activity_notifications_ = new_methods;
     }
-    cache_resolved_classes = false;
+    total_ns_of_work_ += NanoTime() - start_work;
+  }
+}
+
+void ProfileSaver::NotifyJitActivity() {
+  MutexLock mu(Thread::Current(), *Locks::profiler_lock_);
+  if (instance_ == nullptr || instance_->shutting_down_) {
+    return;
+  }
+  instance_->NotifyJitActivityInternal();
+}
 
-    total_ns_of_work_ += (NanoTime() - start);
+void ProfileSaver::WakeUpSaver() {
+  jit_activity_notifications_ = 0;
+  last_time_ns_saver_woke_up_ = NanoTime();
+  period_condition_.Signal(Thread::Current());
+}
+
+void ProfileSaver::NotifyJitActivityInternal() {
+  // Unlikely to overflow but if it happens,
+  // we would have waken up the saver long before that.
+  jit_activity_notifications_++;
+  // Note that we are not as precise as we could be here but we don't want to wake the saver
+  // every time we see a hot method.
+  if (jit_activity_notifications_ > kMinimumNumberOfNotificationBeforeWake) {
+    MutexLock wait_mutex(Thread::Current(), wait_lock_);
+    if ((NanoTime() - last_time_ns_saver_woke_up_) > kMinSavePeriodNs) {
+      WakeUpSaver();
+    }
+  } else if (jit_activity_notifications_ > kMaximumNumberOfNotificationBeforeWake) {
+    // Make sure to wake up the saver if we see a spike in the number of notifications.
+    // This is a precaution to avoid "loosing" a big number of methods in case
+    // this is a spike with no jit after.
+    total_number_of_hot_spikes_++;
+    MutexLock wait_mutex(Thread::Current(), wait_lock_);
+    WakeUpSaver();
   }
 }
 
@@ -145,20 +184,25 @@ ProfileCompilationInfo* ProfileSaver::GetCachedProfiledInfo(const std::string& f
 
 void ProfileSaver::FetchAndCacheResolvedClasses() {
   ScopedTrace trace(__PRETTY_FUNCTION__);
-
   ClassLinker* const class_linker = Runtime::Current()->GetClassLinker();
   std::set<DexCacheResolvedClasses> resolved_classes =
       class_linker->GetResolvedClasses(/*ignore boot classes*/ true);
   MutexLock mu(Thread::Current(), *Locks::profiler_lock_);
   uint64_t total_number_of_profile_entries_cached = 0;
+
   for (const auto& it : tracked_dex_base_locations_) {
-      std::set<DexCacheResolvedClasses> resolved_classes_for_location;
+    std::set<DexCacheResolvedClasses> resolved_classes_for_location;
     const std::string& filename = it.first;
     const std::set<std::string>& locations = it.second;
 
     for (const DexCacheResolvedClasses& classes : resolved_classes) {
-      if (locations.find(classes.GetDexLocation()) != locations.end()) {
+      if (locations.find(classes.GetBaseLocation()) != locations.end()) {
+        VLOG(profiler) << "Added classes for location " << classes.GetBaseLocation()
+                       << " (" << classes.GetDexLocation() << ")";
         resolved_classes_for_location.insert(classes);
+      } else {
+        VLOG(profiler) << "Location not found " << classes.GetBaseLocation()
+                       << " (" << classes.GetDexLocation() << ")";
       }
     }
     ProfileCompilationInfo* info = GetCachedProfiledInfo(filename);
@@ -170,7 +214,7 @@ void ProfileSaver::FetchAndCacheResolvedClasses() {
       total_number_of_profile_entries_cached);
 }
 
-bool ProfileSaver::ProcessProfilingInfo() {
+bool ProfileSaver::ProcessProfilingInfo(uint16_t* new_methods) {
   ScopedTrace trace(__PRETTY_FUNCTION__);
   SafeMap<std::string, std::set<std::string>> tracked_locations;
   {
@@ -181,6 +225,8 @@ bool ProfileSaver::ProcessProfilingInfo() {
 
   bool profile_file_saved = false;
   uint64_t total_number_of_profile_entries_cached = 0;
+  *new_methods = 0;
+
   for (const auto& it : tracked_locations) {
     if (ShuttingDown(Thread::Current())) {
       return true;
@@ -211,6 +257,7 @@ bool ProfileSaver::ProcessProfilingInfo() {
       total_number_of_skipped_writes_++;
       continue;
     }
+    *new_methods = std::max(static_cast<uint16_t>(delta_number_of_methods), *new_methods);
     uint64_t bytes_written;
     // Force the save. In case the profile data is corrupted or the the profile
     // has the wrong version this will "fix" the file to the correct format.
@@ -247,12 +294,17 @@ bool ProfileSaver::ProcessProfilingInfo() {
 
 void* ProfileSaver::RunProfileSaverThread(void* arg) {
   Runtime* runtime = Runtime::Current();
-  ProfileSaver* profile_saver = reinterpret_cast<ProfileSaver*>(arg);
 
-  CHECK(runtime->AttachCurrentThread("Profile Saver",
-                                     /*as_daemon*/true,
-                                     runtime->GetSystemThreadGroup(),
-                                     /*create_peer*/true));
+  bool attached = runtime->AttachCurrentThread("Profile Saver",
+                                               /*as_daemon*/true,
+                                               runtime->GetSystemThreadGroup(),
+                                               /*create_peer*/true);
+  if (!attached) {
+    CHECK(runtime->IsShuttingDown(Thread::Current()));
+    return nullptr;
+  }
+
+  ProfileSaver* profile_saver = reinterpret_cast<ProfileSaver*>(arg);
   profile_saver->Run();
 
   runtime->DetachCurrentThread();
@@ -285,7 +337,7 @@ void ProfileSaver::Start(const std::string& output_filename,
                          const std::vector<std::string>& code_paths,
                          const std::string& foreign_dex_profile_path,
                          const std::string& app_data_dir) {
-  DCHECK(Runtime::Current()->UseJit());
+  DCHECK(Runtime::Current()->SaveProfileInfo());
   DCHECK(!output_filename.empty());
   DCHECK(jit_code_cache != nullptr);
 
@@ -480,17 +532,20 @@ bool ProfileSaver::MaybeRecordDexUseInternal(
   // frameworks/base/services/core/java/com/android/server/pm/PackageDexOptimizer.java)
   std::replace(dex_location_real_path_str.begin(), dex_location_real_path_str.end(), '/', '@');
   std::string flag_path = foreign_dex_profile_path + "/" + dex_location_real_path_str;
-  // No need to give any sort of access to flag_path. The system has enough permissions
-  // to test for its existence.
-  int fd = TEMP_FAILURE_RETRY(open(flag_path.c_str(), O_CREAT | O_EXCL, 0));
+  // We use O_RDONLY as the access mode because we must supply some access
+  // mode, and there is no access mode that means 'create but do not read' the
+  // file. We will not not actually read from the file.
+  int fd = TEMP_FAILURE_RETRY(open(flag_path.c_str(),
+        O_CREAT | O_RDONLY | O_EXCL | O_CLOEXEC | O_NOFOLLOW, 0));
   if (fd != -1) {
     if (close(fd) != 0) {
       PLOG(WARNING) << "Could not close file after flagging foreign dex use " << flag_path;
     }
     return true;
   } else {
-    if (errno != EEXIST) {
-      // Another app could have already created the file.
+    if (errno != EEXIST && errno != EACCES) {
+      // Another app could have already created the file, and selinux may not
+      // allow the read access to the file implied by the call to open.
       PLOG(WARNING) << "Could not create foreign dex use mark " << flag_path;
       return false;
     }
@@ -517,7 +572,38 @@ void ProfileSaver::DumpInfo(std::ostream& os) {
      << "ProfileSaver total_number_of_foreign_dex_marks="
      << total_number_of_foreign_dex_marks_ << '\n'
      << "ProfileSaver max_number_profile_entries_cached="
-    << max_number_of_profile_entries_cached_ << '\n';
+     << max_number_of_profile_entries_cached_ << '\n'
+     << "ProfileSaver total_number_of_hot_spikes=" << total_number_of_hot_spikes_ << '\n'
+     << "ProfileSaver total_number_of_wake_ups=" << total_number_of_wake_ups_ << '\n';
+}
+
+
+void ProfileSaver::ForceProcessProfiles() {
+  ProfileSaver* saver = nullptr;
+  {
+    MutexLock mu(Thread::Current(), *Locks::profiler_lock_);
+    saver = instance_;
+  }
+  // TODO(calin): this is not actually thread safe as the instance_ may have been deleted,
+  // but we only use this in testing when we now this won't happen.
+  // Refactor the way we handle the instance so that we don't end up in this situation.
+  if (saver != nullptr) {
+    uint16_t new_methods;
+    saver->ProcessProfilingInfo(&new_methods);
+  }
+}
+
+bool ProfileSaver::HasSeenMethod(const std::string& profile,
+                                 const DexFile* dex_file,
+                                 uint16_t method_idx) {
+  MutexLock mu(Thread::Current(), *Locks::profiler_lock_);
+  if (instance_ != nullptr) {
+    ProfileCompilationInfo* info = instance_->GetCachedProfiledInfo(profile);
+    if (info != nullptr) {
+      return info->ContainsMethod(MethodReference(dex_file, method_idx));
+    }
+  }
+  return false;
 }
 
 }   // namespace art
diff --git a/runtime/jit/profile_saver.h b/runtime/jit/profile_saver.h
index 0a222bfdc..c6da95931 100644
--- a/runtime/jit/profile_saver.h
+++ b/runtime/jit/profile_saver.h
@@ -49,6 +49,17 @@ class ProfileSaver {
   // If the profile saver is running, dumps statistics to the `os`. Otherwise it does nothing.
   static void DumpInstanceInfo(std::ostream& os);
 
+  // NO_THREAD_SAFETY_ANALYSIS for static function calling into member function with excludes lock.
+  static void NotifyJitActivity()
+      REQUIRES(!Locks::profiler_lock_, !wait_lock_)
+      NO_THREAD_SAFETY_ANALYSIS;
+
+  // Just for testing purpose.
+  static void ForceProcessProfiles();
+  static bool HasSeenMethod(const std::string& profile,
+                            const DexFile* dex_file,
+                            uint16_t method_idx);
+
  private:
   ProfileSaver(const std::string& output_filename,
                jit::JitCodeCache* jit_code_cache,
@@ -65,7 +76,13 @@ class ProfileSaver {
   void Run() REQUIRES(!Locks::profiler_lock_, !wait_lock_);
   // Processes the existing profiling info from the jit code cache and returns
   // true if it needed to be saved to disk.
-  bool ProcessProfilingInfo();
+  bool ProcessProfilingInfo(uint16_t* new_methods)
+    REQUIRES(!Locks::profiler_lock_)
+    REQUIRES(!Locks::mutator_lock_);
+
+  void NotifyJitActivityInternal() REQUIRES(!wait_lock_);
+  void WakeUpSaver() REQUIRES(wait_lock_);
+
   // Returns true if the saver is shutting down (ProfileSaver::Stop() has been called).
   bool ShuttingDown(Thread* self) REQUIRES(!Locks::profiler_lock_);
 
@@ -112,6 +129,8 @@ class ProfileSaver {
   bool shutting_down_ GUARDED_BY(Locks::profiler_lock_);
   uint32_t last_save_number_of_methods_;
   uint32_t last_save_number_of_classes_;
+  uint64_t last_time_ns_saver_woke_up_ GUARDED_BY(wait_lock_);
+  uint32_t jit_activity_notifications_;
 
   // A local cache for the profile information. Maps each tracked file to its
   // profile information. The size of this cache is usually very small and tops
@@ -133,6 +152,8 @@ class ProfileSaver {
   uint64_t total_number_of_foreign_dex_marks_;
   // TODO(calin): replace with an actual size.
   uint64_t max_number_of_profile_entries_cached_;
+  uint64_t total_number_of_hot_spikes_;
+  uint64_t total_number_of_wake_ups_;
 
   DISALLOW_COPY_AND_ASSIGN(ProfileSaver);
 };
diff --git a/runtime/mem_map.cc b/runtime/mem_map.cc
index 5d89c2180..771f8ed29 100644
--- a/runtime/mem_map.cc
+++ b/runtime/mem_map.cc
@@ -302,8 +302,9 @@ MemMap* MemMap::MapAnonymous(const char* name,
 
   if (use_ashmem) {
     if (!kIsTargetBuild) {
-      // When not on Android ashmem is faked using files in /tmp. Ensure that such files won't
-      // fail due to ulimit restrictions. If they will then use a regular mmap.
+      // When not on Android (either host or assuming a linux target) ashmem is faked using
+      // files in /tmp. Ensure that such files won't fail due to ulimit restrictions. If they
+      // will then use a regular mmap.
       struct rlimit rlimit_fsize;
       CHECK_EQ(getrlimit(RLIMIT_FSIZE, &rlimit_fsize), 0);
       use_ashmem = (rlimit_fsize.rlim_cur == RLIM_INFINITY) ||
diff --git a/runtime/mem_map.h b/runtime/mem_map.h
index 3eaf57684..597f0d46e 100644
--- a/runtime/mem_map.h
+++ b/runtime/mem_map.h
@@ -68,7 +68,7 @@ class MemMap {
                               bool low_4gb,
                               bool reuse,
                               std::string* error_msg,
-                              bool use_ashmem = true);
+                              bool use_ashmem = !kIsTargetLinux);
 
   // Create placeholder for a region allocated by direct call to mmap.
   // This is useful when we do not have control over the code calling mmap,
@@ -172,7 +172,7 @@ class MemMap {
                      const char* tail_name,
                      int tail_prot,
                      std::string* error_msg,
-                     bool use_ashmem = true);
+                     bool use_ashmem = !kIsTargetLinux);
 
   static bool CheckNoGaps(MemMap* begin_map, MemMap* end_map)
       REQUIRES(!Locks::mem_maps_lock_);
diff --git a/runtime/mirror/class.h b/runtime/mirror/class.h
index 5b6ded162..8c20fa680 100644
--- a/runtime/mirror/class.h
+++ b/runtime/mirror/class.h
@@ -673,7 +673,7 @@ class MANAGED Class FINAL : public Object {
   // `This` and `klass` must be classes.
   Class* GetCommonSuperClass(Handle<Class> klass) SHARED_REQUIRES(Locks::mutator_lock_);
 
-  void SetSuperClass(Class *new_super_class) SHARED_REQUIRES(Locks::mutator_lock_) {
+  void SetSuperClass(Class* new_super_class) SHARED_REQUIRES(Locks::mutator_lock_) {
     // Super class is assigned once, except during class linker initialization.
     Class* old_super_class = GetFieldObject<Class>(OFFSET_OF_OBJECT_MEMBER(Class, super_class_));
     DCHECK(old_super_class == nullptr || old_super_class == new_super_class);
diff --git a/runtime/modifiers.h b/runtime/modifiers.h
index 6dd182a11..fd7a125bc 100644
--- a/runtime/modifiers.h
+++ b/runtime/modifiers.h
@@ -60,10 +60,13 @@ static constexpr uint32_t kAccDefault =               0x00400000;  // method (ru
 // This is set by the class linker during LinkInterfaceMethods. Prior to that point we do not know
 // if any particular method needs to be a default conflict. Used to figure out at runtime if
 // invoking this method will throw an exception.
-static constexpr uint32_t kAccDefaultConflict =      0x00800000;  // method (runtime)
+static constexpr uint32_t kAccDefaultConflict =       0x00800000;  // method (runtime)
 
 // Set by the verifier for a method we do not want the compiler to compile.
-static constexpr uint32_t kAccCompileDontBother =    0x01000000;  // method (runtime)
+static constexpr uint32_t kAccCompileDontBother =     0x01000000;  // method (runtime)
+
+// Set by the verifier for a method that could not be verified to follow structured locking.
+static constexpr uint32_t kAccMustCountLocks =        0x02000000;  // method (runtime)
 
 // Special runtime-only flags.
 // Interface and all its super-interfaces with default methods have been recursively initialized.
diff --git a/runtime/monitor.cc b/runtime/monitor.cc
index 3680c7831..71c866f3d 100644
--- a/runtime/monitor.cc
+++ b/runtime/monitor.cc
@@ -215,6 +215,85 @@ void Monitor::SetObject(mirror::Object* object) {
   obj_ = GcRoot<mirror::Object>(object);
 }
 
+// Note: Adapted from CurrentMethodVisitor in thread.cc. We must not resolve here.
+
+struct NthCallerWithDexPcVisitor FINAL : public StackVisitor {
+  explicit NthCallerWithDexPcVisitor(Thread* thread, size_t frame)
+      SHARED_REQUIRES(Locks::mutator_lock_)
+      : StackVisitor(thread, nullptr, StackVisitor::StackWalkKind::kIncludeInlinedFramesNoResolve),
+        method_(nullptr),
+        dex_pc_(0),
+        current_frame_number_(0),
+        wanted_frame_number_(frame) {}
+  bool VisitFrame() OVERRIDE SHARED_REQUIRES(Locks::mutator_lock_) {
+    ArtMethod* m = GetMethod();
+    if (m == nullptr || m->IsRuntimeMethod()) {
+      // Runtime method, upcall, or resolution issue. Skip.
+      return true;
+    }
+
+    // Is this the requested frame?
+    if (current_frame_number_ == wanted_frame_number_) {
+      method_ = m;
+      dex_pc_ = GetDexPc(false /* abort_on_error*/);
+      return false;
+    }
+
+    // Look for more.
+    current_frame_number_++;
+    return true;
+  }
+
+  ArtMethod* method_;
+  uint32_t dex_pc_;
+
+ private:
+  size_t current_frame_number_;
+  const size_t wanted_frame_number_;
+};
+
+// This function is inlined and just helps to not have the VLOG and ATRACE check at all the
+// potential tracing points.
+void Monitor::AtraceMonitorLock(Thread* self, mirror::Object* obj, bool is_wait) {
+  if (UNLIKELY(VLOG_IS_ON(systrace_lock_logging) && ATRACE_ENABLED())) {
+    AtraceMonitorLockImpl(self, obj, is_wait);
+  }
+}
+
+void Monitor::AtraceMonitorLockImpl(Thread* self, mirror::Object* obj, bool is_wait) {
+  // Wait() requires a deeper call stack to be useful. Otherwise you'll see "Waiting at
+  // Object.java". Assume that we'll wait a nontrivial amount, so it's OK to do a longer
+  // stack walk than if !is_wait.
+  NthCallerWithDexPcVisitor visitor(self, is_wait ? 1U : 0U);
+  visitor.WalkStack(false);
+  const char* prefix = is_wait ? "Waiting on " : "Locking ";
+
+  const char* filename;
+  int32_t line_number;
+  TranslateLocation(visitor.method_, visitor.dex_pc_, &filename, &line_number);
+
+  // It would be nice to have a stable "ID" for the object here. However, the only stable thing
+  // would be the identity hashcode. But we cannot use IdentityHashcode here: For one, there are
+  // times when it is unsafe to make that call (see stack dumping for an explanation). More
+  // importantly, we would have to give up on thin-locking when adding systrace locks, as the
+  // identity hashcode is stored in the lockword normally (so can't be used with thin-locks).
+  //
+  // Because of thin-locks we also cannot use the monitor id (as there is no monitor). Monitor ids
+  // also do not have to be stable, as the monitor may be deflated.
+  std::string tmp = StringPrintf("%s %d at %s:%d",
+      prefix,
+      (obj == nullptr ? -1 : static_cast<int32_t>(reinterpret_cast<uintptr_t>(obj))),
+      (filename != nullptr ? filename : "null"),
+      line_number);
+  ATRACE_BEGIN(tmp.c_str());
+}
+
+void Monitor::AtraceMonitorUnlock() {
+  if (UNLIKELY(VLOG_IS_ON(systrace_lock_logging))) {
+    ATRACE_END();
+  }
+}
+
 std::string Monitor::PrettyContentionInfo(const std::string& owner_name,
                                           pid_t owner_tid,
                                           ArtMethod* owners_method,
@@ -228,8 +307,8 @@ std::string Monitor::PrettyContentionInfo(const std::string& owner_name,
   std::ostringstream oss;
   oss << "monitor contention with owner " << owner_name << " (" << owner_tid << ")";
   if (owners_method != nullptr) {
-    oss << " owner method=" << PrettyMethod(owners_method);
-    oss << " from " << owners_filename << ":" << owners_line_number;
+    oss << " at " << PrettyMethod(owners_method);
+    oss << "(" << owners_filename << ":" << owners_line_number << ")";
   }
   oss << " waiters=" << num_waiters;
   return oss.str();
@@ -246,10 +325,10 @@ void Monitor::Lock(Thread* self) {
       if (lock_profiling_threshold_ != 0) {
         locking_method_ = self->GetCurrentMethod(&locking_dex_pc_);
       }
-      return;
+      break;
     } else if (owner_ == self) {  // Recursive.
       lock_count_++;
-      return;
+      break;
     }
     // Contended.
     const bool log_contention = (lock_profiling_threshold_ != 0);
@@ -284,8 +363,9 @@ void Monitor::Lock(Thread* self) {
             const char* filename;
             int32_t line_number;
             TranslateLocation(m, pc, &filename, &line_number);
-            oss << " blocking from " << (filename != nullptr ? filename : "null")
-                << ":" << line_number;
+            oss << " blocking from "
+                << PrettyMethod(m) << "(" << (filename != nullptr ? filename : "null") << ":"
+                << line_number << ")";
             ATRACE_BEGIN(oss.str().c_str());
           }
           monitor_contenders_.Wait(self);  // Still contended so wait.
@@ -318,6 +398,8 @@ void Monitor::Lock(Thread* self) {
             }
             if (sample_percent != 0 && (static_cast<uint32_t>(rand() % 100) < sample_percent)) {
               if (wait_ms > kLongWaitMs && owners_method != nullptr) {
+                uint32_t pc;
+                ArtMethod* m = self->GetCurrentMethod(&pc);
                 // TODO: We should maybe check that original_owner is still a live thread.
                 LOG(WARNING) << "Long "
                     << PrettyContentionInfo(original_owner_name,
@@ -325,7 +407,7 @@ void Monitor::Lock(Thread* self) {
                                             owners_method,
                                             owners_dex_pc,
                                             num_waiters)
-                    << " for " << PrettyDuration(MsToNs(wait_ms));
+                    << " in " << PrettyMethod(m) << " for " << PrettyDuration(MsToNs(wait_ms));
               }
               const char* owners_filename;
               int32_t owners_line_number;
@@ -348,6 +430,8 @@ void Monitor::Lock(Thread* self) {
     monitor_lock_.Lock(self);  // Reacquire locks in order.
     --num_waiters_;
   }
+
+  AtraceMonitorLock(self, GetObject(), false /* is_wait */);
 }
 
 static void ThrowIllegalMonitorStateExceptionF(const char* fmt, ...)
@@ -457,6 +541,7 @@ bool Monitor::Unlock(Thread* self) {
     }
     if (owner == self) {
       // We own the monitor, so nobody else can be in here.
+      AtraceMonitorUnlock();
       if (lock_count_ == 0) {
         owner_ = nullptr;
         locking_method_ = nullptr;
@@ -523,6 +608,11 @@ void Monitor::Wait(Thread* self, int64_t ms, int32_t ns,
   uintptr_t saved_dex_pc = locking_dex_pc_;
   locking_dex_pc_ = 0;
 
+  AtraceMonitorUnlock();  // For the implict Unlock() just above. This will only end the deepest
+                          // nesting, but that is enough for the visualization, and corresponds to
+                          // the single Lock() we do afterwards.
+  AtraceMonitorLock(self, GetObject(), true /* is_wait */);
+
   bool was_interrupted = false;
   {
     // Update thread state. If the GC wakes up, it'll ignore us, knowing
@@ -586,6 +676,8 @@ void Monitor::Wait(Thread* self, int64_t ms, int32_t ns,
     self->ThrowNewException("Ljava/lang/InterruptedException;", nullptr);
   }
 
+  AtraceMonitorUnlock();  // End Wait().
+
   // Re-acquire the monitor and lock.
   Lock(self);
   monitor_lock_.Lock(self);
@@ -775,6 +867,7 @@ mirror::Object* Monitor::MonitorEnter(Thread* self, mirror::Object* obj) {
       case LockWord::kUnlocked: {
         LockWord thin_locked(LockWord::FromThinLockId(thread_id, 0, lock_word.ReadBarrierState()));
         if (h_obj->CasLockWordWeakSequentiallyConsistent(lock_word, thin_locked)) {
+          AtraceMonitorLock(self, h_obj.Get(), false /* is_wait */);
           // CasLockWord enforces more than the acquire ordering we need here.
           return h_obj.Get();  // Success!
         }
@@ -790,10 +883,12 @@ mirror::Object* Monitor::MonitorEnter(Thread* self, mirror::Object* obj) {
                                                           lock_word.ReadBarrierState()));
             if (!kUseReadBarrier) {
               h_obj->SetLockWord(thin_locked, true);
+              AtraceMonitorLock(self, h_obj.Get(), false /* is_wait */);
               return h_obj.Get();  // Success!
             } else {
               // Use CAS to preserve the read barrier state.
               if (h_obj->CasLockWordWeakSequentiallyConsistent(lock_word, thin_locked)) {
+                AtraceMonitorLock(self, h_obj.Get(), false /* is_wait */);
                 return h_obj.Get();  // Success!
               }
             }
@@ -830,7 +925,7 @@ mirror::Object* Monitor::MonitorEnter(Thread* self, mirror::Object* obj) {
         continue;  // Start from the beginning.
       default: {
         LOG(FATAL) << "Invalid monitor state " << lock_word.GetState();
-        return h_obj.Get();
+        UNREACHABLE();
       }
     }
   }
@@ -869,11 +964,13 @@ bool Monitor::MonitorExit(Thread* self, mirror::Object* obj) {
           if (!kUseReadBarrier) {
             DCHECK_EQ(new_lw.ReadBarrierState(), 0U);
             h_obj->SetLockWord(new_lw, true);
+            AtraceMonitorUnlock();
             // Success!
             return true;
           } else {
             // Use CAS to preserve the read barrier state.
             if (h_obj->CasLockWordWeakSequentiallyConsistent(lock_word, new_lw)) {
+              AtraceMonitorUnlock();
               // Success!
               return true;
             }
diff --git a/runtime/monitor.h b/runtime/monitor.h
index 8c7496b52..7b4b8f946 100644
--- a/runtime/monitor.h
+++ b/runtime/monitor.h
@@ -250,6 +250,17 @@ class Monitor {
 
   uint32_t GetOwnerThreadId() REQUIRES(!monitor_lock_);
 
+  // Support for systrace output of monitor operations.
+  ALWAYS_INLINE static void AtraceMonitorLock(Thread* self,
+                                              mirror::Object* obj,
+                                              bool is_wait)
+      SHARED_REQUIRES(Locks::mutator_lock_);
+  static void AtraceMonitorLockImpl(Thread* self,
+                                    mirror::Object* obj,
+                                    bool is_wait)
+      SHARED_REQUIRES(Locks::mutator_lock_);
+  ALWAYS_INLINE static void AtraceMonitorUnlock();
+
   static uint32_t lock_profiling_threshold_;
 
   Mutex monitor_lock_ DEFAULT_MUTEX_ACQUIRED_AFTER;
diff --git a/runtime/monitor_pool.cc b/runtime/monitor_pool.cc
index ce38e4f10..a47a4b2cf 100644
--- a/runtime/monitor_pool.cc
+++ b/runtime/monitor_pool.cc
@@ -28,7 +28,11 @@ namespace mirror {
 }  // namespace mirror
 
 MonitorPool::MonitorPool()
-    : num_chunks_(0), capacity_(0), first_free_(nullptr) {
+    : current_chunk_list_index_(0), num_chunks_(0), current_chunk_list_capacity_(0),
+    first_free_(nullptr) {
+  for (size_t i = 0; i < kMaxChunkLists; ++i) {
+    monitor_chunks_[i] = nullptr;  // Not absolutely required, but ...
+  }
   AllocateChunk();  // Get our first chunk.
 }
 
@@ -37,24 +41,19 @@ MonitorPool::MonitorPool()
 void MonitorPool::AllocateChunk() {
   DCHECK(first_free_ == nullptr);
 
-  // Do we need to resize?
-  if (num_chunks_ == capacity_) {
-    if (capacity_ == 0U) {
-      // Initialization.
-      capacity_ = kInitialChunkStorage;
-      uintptr_t* new_backing = new uintptr_t[capacity_]();
-      DCHECK(monitor_chunks_.LoadRelaxed() == nullptr);
-      monitor_chunks_.StoreRelaxed(new_backing);
-    } else {
-      size_t new_capacity = 2 * capacity_;
-      uintptr_t* new_backing = new uintptr_t[new_capacity]();
-      uintptr_t* old_backing = monitor_chunks_.LoadRelaxed();
-      memcpy(new_backing, old_backing, sizeof(uintptr_t) * capacity_);
-      monitor_chunks_.StoreRelaxed(new_backing);
-      capacity_ = new_capacity;
-      old_chunk_arrays_.push_back(std::unique_ptr<uintptr_t[]>(old_backing));
-      VLOG(monitor) << "Resizing to capacity " << capacity_;
-    }
+  // Do we need to allocate another chunk list?
+  if (num_chunks_ == current_chunk_list_capacity_) {
+    if (current_chunk_list_capacity_ != 0U) {
+      ++current_chunk_list_index_;
+      CHECK_LT(current_chunk_list_index_, kMaxChunkLists) << "Out of space for inflated monitors";
+      VLOG(monitor) << "Expanding to capacity "
+          << 2 * ChunkListCapacity(current_chunk_list_index_) - kInitialChunkStorage;
+    }  // else we're initializing
+    current_chunk_list_capacity_ = ChunkListCapacity(current_chunk_list_index_);
+    uintptr_t* new_list = new uintptr_t[current_chunk_list_capacity_]();
+    DCHECK(monitor_chunks_[current_chunk_list_index_] == nullptr);
+    monitor_chunks_[current_chunk_list_index_] = new_list;
+    num_chunks_ = 0;
   }
 
   // Allocate the chunk.
@@ -65,7 +64,7 @@ void MonitorPool::AllocateChunk() {
   CHECK_EQ(0U, reinterpret_cast<uintptr_t>(chunk) % kMonitorAlignment);
 
   // Add the chunk.
-  *(monitor_chunks_.LoadRelaxed() + num_chunks_) = reinterpret_cast<uintptr_t>(chunk);
+  monitor_chunks_[current_chunk_list_index_][num_chunks_] = reinterpret_cast<uintptr_t>(chunk);
   num_chunks_++;
 
   // Set up the free list
@@ -73,8 +72,8 @@ void MonitorPool::AllocateChunk() {
                                              (kChunkCapacity - 1) * kAlignedMonitorSize);
   last->next_free_ = nullptr;
   // Eagerly compute id.
-  last->monitor_id_ = OffsetToMonitorId((num_chunks_ - 1) * kChunkSize +
-                                        (kChunkCapacity - 1) * kAlignedMonitorSize);
+  last->monitor_id_ = OffsetToMonitorId(current_chunk_list_index_* (kMaxListSize * kChunkSize)
+      + (num_chunks_ - 1) * kChunkSize + (kChunkCapacity - 1) * kAlignedMonitorSize);
   for (size_t i = 0; i < kChunkCapacity - 1; ++i) {
     Monitor* before = reinterpret_cast<Monitor*>(reinterpret_cast<uintptr_t>(last) -
                                                  kAlignedMonitorSize);
@@ -91,21 +90,19 @@ void MonitorPool::AllocateChunk() {
 
 void MonitorPool::FreeInternal() {
   // This is on shutdown with NO_THREAD_SAFETY_ANALYSIS, can't/don't need to lock.
-  uintptr_t* backing = monitor_chunks_.LoadRelaxed();
-  DCHECK(backing != nullptr);
-  DCHECK_GT(capacity_, 0U);
-  DCHECK_GT(num_chunks_, 0U);
-
-  for (size_t i = 0; i < capacity_; ++i) {
-    if (i < num_chunks_) {
-      DCHECK_NE(backing[i], 0U);
-      allocator_.deallocate(reinterpret_cast<uint8_t*>(backing[i]), kChunkSize);
-    } else {
-      DCHECK_EQ(backing[i], 0U);
+  DCHECK_NE(current_chunk_list_capacity_, 0UL);
+  for (size_t i = 0; i <= current_chunk_list_index_; ++i) {
+    DCHECK_NE(monitor_chunks_[i], static_cast<uintptr_t*>(nullptr));
+    for (size_t j = 0; j < ChunkListCapacity(i); ++j) {
+      if (i < current_chunk_list_index_ || j < num_chunks_) {
+        DCHECK_NE(monitor_chunks_[i][j], 0U);
+        allocator_.deallocate(reinterpret_cast<uint8_t*>(monitor_chunks_[i][j]), kChunkSize);
+      } else {
+        DCHECK_EQ(monitor_chunks_[i][j], 0U);
+      }
     }
+    delete[] monitor_chunks_[i];
   }
-
-  delete[] backing;
 }
 
 Monitor* MonitorPool::CreateMonitorInPool(Thread* self, Thread* owner, mirror::Object* obj,
diff --git a/runtime/monitor_pool.h b/runtime/monitor_pool.h
index 875b3fe73..99810e0c8 100644
--- a/runtime/monitor_pool.h
+++ b/runtime/monitor_pool.h
@@ -128,12 +128,17 @@ class MonitorPool {
   void ReleaseMonitorToPool(Thread* self, Monitor* monitor);
   void ReleaseMonitorsToPool(Thread* self, MonitorList::Monitors* monitors);
 
-  // Note: This is safe as we do not ever move chunks.
+  // Note: This is safe as we do not ever move chunks.  All needed entries in the monitor_chunks_
+  // data structure are read-only once we get here.  Updates happen-before this call because
+  // the lock word was stored with release semantics and we read it with acquire semantics to
+  // retrieve the id.
   Monitor* LookupMonitor(MonitorId mon_id) {
     size_t offset = MonitorIdToOffset(mon_id);
     size_t index = offset / kChunkSize;
+    size_t top_index = index / kMaxListSize;
+    size_t list_index = index % kMaxListSize;
     size_t offset_in_chunk = offset % kChunkSize;
-    uintptr_t base = *(monitor_chunks_.LoadRelaxed()+index);
+    uintptr_t base = monitor_chunks_[top_index][list_index];
     return reinterpret_cast<Monitor*>(base + offset_in_chunk);
   }
 
@@ -142,28 +147,37 @@ class MonitorPool {
     return base_addr <= mon_ptr && (mon_ptr - base_addr < kChunkSize);
   }
 
-  // Note: This is safe as we do not ever move chunks.
   MonitorId ComputeMonitorIdInPool(Monitor* mon, Thread* self) {
     MutexLock mu(self, *Locks::allocated_monitor_ids_lock_);
-    for (size_t index = 0; index < num_chunks_; ++index) {
-      uintptr_t chunk_addr = *(monitor_chunks_.LoadRelaxed() + index);
-      if (IsInChunk(chunk_addr, mon)) {
-        return OffsetToMonitorId(
-            reinterpret_cast<uintptr_t>(mon) - chunk_addr + index * kChunkSize);
+    for (size_t i = 0; i <= current_chunk_list_index_; ++i) {
+      for (size_t j = 0; j < ChunkListCapacity(i); ++j) {
+        if (j >= num_chunks_ && i == current_chunk_list_index_) {
+          break;
+        }
+        uintptr_t chunk_addr = monitor_chunks_[i][j];
+        if (IsInChunk(chunk_addr, mon)) {
+          return OffsetToMonitorId(
+              reinterpret_cast<uintptr_t>(mon) - chunk_addr
+              + i * (kMaxListSize * kChunkSize) + j * kChunkSize);
+        }
       }
     }
     LOG(FATAL) << "Did not find chunk that contains monitor.";
     return 0;
   }
 
-  static size_t MonitorIdToOffset(MonitorId id) {
+  static constexpr size_t MonitorIdToOffset(MonitorId id) {
     return id << 3;
   }
 
-  static MonitorId OffsetToMonitorId(size_t offset) {
+  static constexpr MonitorId OffsetToMonitorId(size_t offset) {
     return static_cast<MonitorId>(offset >> 3);
   }
 
+  static constexpr size_t ChunkListCapacity(size_t index) {
+    return kInitialChunkStorage << index;
+  }
+
   // TODO: There are assumptions in the code that monitor addresses are 8B aligned (>>3).
   static constexpr size_t kMonitorAlignment = 8;
   // Size of a monitor, rounded up to a multiple of alignment.
@@ -174,20 +188,47 @@ class MonitorPool {
   // Chunk size that is referenced in the id. We can collapse this to the actually used storage
   // in a chunk, i.e., kChunkCapacity * kAlignedMonitorSize, but this will mean proper divisions.
   static constexpr size_t kChunkSize = kPageSize;
-  // The number of initial chunks storable in monitor_chunks_. The number is large enough to make
-  // resizing unlikely, but small enough to not waste too much memory.
-  static constexpr size_t kInitialChunkStorage = 8U;
-
-  // List of memory chunks. Each chunk is kChunkSize.
-  Atomic<uintptr_t*> monitor_chunks_;
-  // Number of chunks stored.
+  static_assert(IsPowerOfTwo(kChunkSize), "kChunkSize must be power of 2");
+  // The number of chunks of storage that can be referenced by the initial chunk list.
+  // The total number of usable monitor chunks is typically 255 times this number, so it
+  // should be large enough that we don't run out. We run out of address bits if it's > 512.
+  // Currently we set it a bit smaller, to save half a page per process.  We make it tiny in
+  // debug builds to catch growth errors. The only value we really expect to tune.
+  static constexpr size_t kInitialChunkStorage = kIsDebugBuild ? 1U : 256U;
+  static_assert(IsPowerOfTwo(kInitialChunkStorage), "kInitialChunkStorage must be power of 2");
+  // The number of lists, each containing pointers to storage chunks.
+  static constexpr size_t kMaxChunkLists = 8;  //  Dictated by 3 bit index. Don't increase above 8.
+  static_assert(IsPowerOfTwo(kMaxChunkLists), "kMaxChunkLists must be power of 2");
+  static constexpr size_t kMaxListSize = kInitialChunkStorage << (kMaxChunkLists - 1);
+  // We lose 3 bits in monitor id due to 3 bit monitor_chunks_ index, and gain it back from
+  // the 3 bit alignment constraint on monitors:
+  static_assert(kMaxListSize * kChunkSize < (1 << LockWord::kMonitorIdSize),
+      "Monitor id bits don't fit");
+  static_assert(IsPowerOfTwo(kMaxListSize), "kMaxListSize must be power of 2");
+
+  // Array of pointers to lists (again arrays) of pointers to chunks containing monitors.
+  // Zeroth entry points to a list (array) of kInitialChunkStorage pointers to chunks.
+  // Each subsequent list as twice as large as the preceding one.
+  // Monitor Ids are interpreted as follows:
+  //     Top 3 bits (of 28): index into monitor_chunks_.
+  //     Next 16 bits: index into the chunk list, i.e. monitor_chunks_[i].
+  //     Last 9 bits: offset within chunk, expressed as multiple of kMonitorAlignment.
+  // If we set kInitialChunkStorage to 512, this would allow us to use roughly 128K chunks of
+  // monitors, which is 0.5GB of monitors.  With this maximum setting, the largest chunk list
+  // contains 64K entries, and we make full use of the available index space. With a
+  // kInitialChunkStorage value of 256, this is proportionately reduced to 0.25GB of monitors.
+  // Updates to monitor_chunks_ are guarded by allocated_monitor_ids_lock_ .
+  // No field in this entire data structure is ever updated once a monitor id whose lookup
+  // requires it has been made visible to another thread.  Thus readers never race with
+  // updates, in spite of the fact that they acquire no locks.
+  uintptr_t* monitor_chunks_[kMaxChunkLists];  //  uintptr_t is really a Monitor* .
+  // Highest currently used index in monitor_chunks_ . Used for newly allocated chunks.
+  size_t current_chunk_list_index_ GUARDED_BY(Locks::allocated_monitor_ids_lock_);
+  // Number of chunk pointers stored in monitor_chunks_[current_chunk_list_index_] so far.
   size_t num_chunks_ GUARDED_BY(Locks::allocated_monitor_ids_lock_);
-  // Number of chunks storable.
-  size_t capacity_ GUARDED_BY(Locks::allocated_monitor_ids_lock_);
-
-  // To avoid race issues when resizing, we keep all the previous arrays.
-  std::vector<std::unique_ptr<uintptr_t[]>> old_chunk_arrays_
-      GUARDED_BY(Locks::allocated_monitor_ids_lock_);
+  // After the initial allocation, this is always equal to
+  // ChunkListCapacity(current_chunk_list_index_).
+  size_t current_chunk_list_capacity_ GUARDED_BY(Locks::allocated_monitor_ids_lock_);
 
   typedef TrackingAllocator<uint8_t, kAllocatorTagMonitorPool> Allocator;
   Allocator allocator_;
diff --git a/runtime/native/dalvik_system_DexFile.cc b/runtime/native/dalvik_system_DexFile.cc
index 0abe39d87..0126b4d0a 100644
--- a/runtime/native/dalvik_system_DexFile.cc
+++ b/runtime/native/dalvik_system_DexFile.cc
@@ -16,6 +16,8 @@
 
 #include "dalvik_system_DexFile.h"
 
+#include <sstream>
+
 #include "base/logging.h"
 #include "base/stl_util.h"
 #include "base/stringprintf.h"
@@ -27,6 +29,7 @@
 #include "mirror/class_loader.h"
 #include "mirror/object-inl.h"
 #include "mirror/string.h"
+#include "oat_file.h"
 #include "oat_file_assistant.h"
 #include "oat_file_manager.h"
 #include "os.h"
@@ -387,6 +390,61 @@ static jint GetDexOptNeeded(JNIEnv* env,
   return oat_file_assistant.GetDexOptNeeded(filter);
 }
 
+static jstring DexFile_getDexFileStatus(JNIEnv* env,
+                                        jclass,
+                                        jstring javaFilename,
+                                        jstring javaInstructionSet) {
+  ScopedUtfChars filename(env, javaFilename);
+  if (env->ExceptionCheck()) {
+    return nullptr;
+  }
+
+  ScopedUtfChars instruction_set(env, javaInstructionSet);
+  if (env->ExceptionCheck()) {
+    return nullptr;
+  }
+
+  const InstructionSet target_instruction_set = GetInstructionSetFromString(
+      instruction_set.c_str());
+  if (target_instruction_set == kNone) {
+    ScopedLocalRef<jclass> iae(env, env->FindClass("java/lang/IllegalArgumentException"));
+    std::string message(StringPrintf("Instruction set %s is invalid.", instruction_set.c_str()));
+    env->ThrowNew(iae.get(), message.c_str());
+    return nullptr;
+  }
+
+  OatFileAssistant oat_file_assistant(filename.c_str(), target_instruction_set,
+                                      false /* profile_changed */,
+                                      false /* load_executable */);
+
+  std::ostringstream status;
+  bool oat_file_exists = false;
+  bool odex_file_exists = false;
+  if (oat_file_assistant.OatFileExists()) {
+    oat_file_exists = true;
+    status << *oat_file_assistant.OatFileName() << " [compilation_filter=";
+    status << CompilerFilter::NameOfFilter(oat_file_assistant.OatFileCompilerFilter());
+    status << ", status=" << oat_file_assistant.OatFileStatus();
+  }
+
+  if (oat_file_assistant.OdexFileExists()) {
+    odex_file_exists = true;
+    if (oat_file_exists) {
+      status << "] ";
+    }
+    status << *oat_file_assistant.OdexFileName() << " [compilation_filter=";
+    status << CompilerFilter::NameOfFilter(oat_file_assistant.OdexFileCompilerFilter());
+    status << ", status=" << oat_file_assistant.OdexFileStatus();
+  }
+
+  if (!oat_file_exists && !odex_file_exists) {
+    status << "invalid[";
+  }
+
+  status << "]";
+  return env->NewStringUTF(status.str().c_str());
+}
+
 static jint DexFile_getDexOptNeeded(JNIEnv* env,
                                     jclass,
                                     jstring javaFilename,
@@ -481,6 +539,16 @@ static jstring DexFile_getNonProfileGuidedCompilerFilter(JNIEnv* env,
   return env->NewStringUTF(new_filter_str.c_str());
 }
 
+static jboolean DexFile_isBackedByOatFile(JNIEnv* env, jclass, jobject cookie) {
+  const OatFile* oat_file = nullptr;
+  std::vector<const DexFile*> dex_files;
+  if (!ConvertJavaArrayToDexFiles(env, cookie, /*out */ dex_files, /* out */ oat_file)) {
+    DCHECK(env->ExceptionCheck());
+    return false;
+  }
+  return oat_file != nullptr;
+}
+
 static JNINativeMethod gMethods[] = {
   NATIVE_METHOD(DexFile, closeDexFile, "(Ljava/lang/Object;)Z"),
   NATIVE_METHOD(DexFile,
@@ -506,6 +574,9 @@ static JNINativeMethod gMethods[] = {
   NATIVE_METHOD(DexFile,
                 getNonProfileGuidedCompilerFilter,
                 "(Ljava/lang/String;)Ljava/lang/String;"),
+  NATIVE_METHOD(DexFile, isBackedByOatFile, "(Ljava/lang/Object;)Z"),
+  NATIVE_METHOD(DexFile, getDexFileStatus,
+                "(Ljava/lang/String;Ljava/lang/String;)Ljava/lang/String;")
 };
 
 void register_dalvik_system_DexFile(JNIEnv* env) {
diff --git a/runtime/native/dalvik_system_VMRuntime.cc b/runtime/native/dalvik_system_VMRuntime.cc
index f355c2a94..6c943dc17 100644
--- a/runtime/native/dalvik_system_VMRuntime.cc
+++ b/runtime/native/dalvik_system_VMRuntime.cc
@@ -145,6 +145,10 @@ static jboolean VMRuntime_isDebuggerActive(JNIEnv*, jobject) {
   return Dbg::IsDebuggerActive();
 }
 
+static jboolean VMRuntime_isNativeDebuggable(JNIEnv*, jobject) {
+  return Runtime::Current()->IsNativeDebuggable();
+}
+
 static jobjectArray VMRuntime_properties(JNIEnv* env, jobject) {
   return toStringArray(env, Runtime::Current()->GetProperties());
 }
@@ -212,6 +216,10 @@ static void VMRuntime_registerNativeAllocation(JNIEnv* env, jobject, jint bytes)
   Runtime::Current()->GetHeap()->RegisterNativeAllocation(env, static_cast<size_t>(bytes));
 }
 
+static void VMRuntime_registerSensitiveThread(JNIEnv*, jobject) {
+  Runtime::Current()->RegisterSensitiveThread();
+}
+
 static void VMRuntime_registerNativeFree(JNIEnv* env, jobject, jint bytes) {
   if (UNLIKELY(bytes < 0)) {
     ScopedObjectAccess soa(env);
@@ -637,12 +645,14 @@ static JNINativeMethod gMethods[] = {
   NATIVE_METHOD(VMRuntime, disableJitCompilation, "()V"),
   NATIVE_METHOD(VMRuntime, getTargetHeapUtilization, "()F"),
   NATIVE_METHOD(VMRuntime, isDebuggerActive, "!()Z"),
+  NATIVE_METHOD(VMRuntime, isNativeDebuggable, "!()Z"),
   NATIVE_METHOD(VMRuntime, nativeSetTargetHeapUtilization, "(F)V"),
   NATIVE_METHOD(VMRuntime, newNonMovableArray, "!(Ljava/lang/Class;I)Ljava/lang/Object;"),
   NATIVE_METHOD(VMRuntime, newUnpaddedArray, "!(Ljava/lang/Class;I)Ljava/lang/Object;"),
   NATIVE_METHOD(VMRuntime, properties, "()[Ljava/lang/String;"),
   NATIVE_METHOD(VMRuntime, setTargetSdkVersionNative, "(I)V"),
   NATIVE_METHOD(VMRuntime, registerNativeAllocation, "(I)V"),
+  NATIVE_METHOD(VMRuntime, registerSensitiveThread, "()V"),
   NATIVE_METHOD(VMRuntime, registerNativeFree, "(I)V"),
   NATIVE_METHOD(VMRuntime, requestConcurrentGC, "()V"),
   NATIVE_METHOD(VMRuntime, requestHeapTrim, "()V"),
diff --git a/runtime/native/java_lang_Class.cc b/runtime/native/java_lang_Class.cc
index 6b7ca40be..0624da38c 100644
--- a/runtime/native/java_lang_Class.cc
+++ b/runtime/native/java_lang_Class.cc
@@ -238,12 +238,13 @@ static mirror::Field* GetPublicFieldRecursive(
   DCHECK(name != nullptr);
   DCHECK(self != nullptr);
 
-  StackHandleScope<1> hs(self);
+  StackHandleScope<2> hs(self);
   MutableHandle<mirror::Class> h_clazz(hs.NewHandle(clazz));
+  Handle<mirror::String> h_name(hs.NewHandle(name));
 
   // We search the current class, its direct interfaces then its superclass.
   while (h_clazz.Get() != nullptr) {
-    mirror::Field* result = GetDeclaredField(self, h_clazz.Get(), name);
+    mirror::Field* result = GetDeclaredField(self, h_clazz.Get(), h_name.Get());
     if ((result != nullptr) && (result->GetAccessFlags() & kAccPublic)) {
       return result;
     } else if (UNLIKELY(self->IsExceptionPending())) {
@@ -258,7 +259,7 @@ static mirror::Field* GetPublicFieldRecursive(
         self->AssertPendingException();
         return nullptr;
       }
-      result = GetPublicFieldRecursive(self, iface, name);
+      result = GetPublicFieldRecursive(self, iface, h_name.Get());
       if (result != nullptr) {
         DCHECK(result->GetAccessFlags() & kAccPublic);
         return result;
diff --git a/runtime/native/java_lang_reflect_Constructor.cc b/runtime/native/java_lang_reflect_Constructor.cc
index ddcaadefa..54b8afd1f 100644
--- a/runtime/native/java_lang_reflect_Constructor.cc
+++ b/runtime/native/java_lang_reflect_Constructor.cc
@@ -34,20 +34,38 @@ static jobject Constructor_getAnnotationNative(JNIEnv* env, jobject javaMethod,
   ScopedFastNativeObjectAccess soa(env);
   StackHandleScope<1> hs(soa.Self());
   ArtMethod* method = ArtMethod::FromReflectedMethod(soa, javaMethod);
-  Handle<mirror::Class> klass(hs.NewHandle(soa.Decode<mirror::Class*>(annotationType)));
-  return soa.AddLocalReference<jobject>(
-      method->GetDexFile()->GetAnnotationForMethod(method, klass));
+  if (method->IsProxyMethod()) {
+    return nullptr;
+  } else {
+    Handle<mirror::Class> klass(hs.NewHandle(soa.Decode<mirror::Class*>(annotationType)));
+    return soa.AddLocalReference<jobject>(
+        method->GetDexFile()->GetAnnotationForMethod(method, klass));
+  }
 }
 
 static jobjectArray Constructor_getDeclaredAnnotations(JNIEnv* env, jobject javaMethod) {
   ScopedFastNativeObjectAccess soa(env);
   ArtMethod* method = ArtMethod::FromReflectedMethod(soa, javaMethod);
-  return soa.AddLocalReference<jobjectArray>(method->GetDexFile()->GetAnnotationsForMethod(method));
+  if (method->IsProxyMethod()) {
+    mirror::Class* class_class = mirror::Class::GetJavaLangClass();
+    mirror::Class* class_array_class =
+        Runtime::Current()->GetClassLinker()->FindArrayClass(soa.Self(), &class_class);
+    if (class_array_class == nullptr) {
+      return nullptr;
+    }
+    mirror::ObjectArray<mirror::Class>* empty_array =
+        mirror::ObjectArray<mirror::Class>::Alloc(soa.Self(), class_array_class, 0);
+    return soa.AddLocalReference<jobjectArray>(empty_array);
+  } else {
+    return soa.AddLocalReference<jobjectArray>(
+        method->GetDexFile()->GetAnnotationsForMethod(method));
+  }
 }
 
 static jobjectArray Constructor_getExceptionTypes(JNIEnv* env, jobject javaMethod) {
   ScopedFastNativeObjectAccess soa(env);
-  ArtMethod* method = ArtMethod::FromReflectedMethod(soa, javaMethod);
+  ArtMethod* method = ArtMethod::FromReflectedMethod(soa, javaMethod)
+      ->GetInterfaceMethodIfProxy(sizeof(void*));
   mirror::ObjectArray<mirror::Class>* result_array =
       method->GetDexFile()->GetExceptionTypesForMethod(method);
   if (result_array == nullptr) {
@@ -69,7 +87,12 @@ static jobjectArray Constructor_getExceptionTypes(JNIEnv* env, jobject javaMetho
 static jobjectArray Constructor_getParameterAnnotationsNative(JNIEnv* env, jobject javaMethod) {
   ScopedFastNativeObjectAccess soa(env);
   ArtMethod* method = ArtMethod::FromReflectedMethod(soa, javaMethod);
-  return soa.AddLocalReference<jobjectArray>(method->GetDexFile()->GetParameterAnnotations(method));
+  if (method->IsProxyMethod()) {
+    return nullptr;
+  } else {
+    return soa.AddLocalReference<jobjectArray>(
+        method->GetDexFile()->GetParameterAnnotations(method));
+  }
 }
 
 static jboolean Constructor_isAnnotationPresentNative(JNIEnv* env, jobject javaMethod,
@@ -77,6 +100,10 @@ static jboolean Constructor_isAnnotationPresentNative(JNIEnv* env, jobject javaM
   ScopedFastNativeObjectAccess soa(env);
   StackHandleScope<1> hs(soa.Self());
   ArtMethod* method = ArtMethod::FromReflectedMethod(soa, javaMethod);
+  if (method->IsProxyMethod()) {
+    // Proxies have no annotations.
+    return false;
+  }
   Handle<mirror::Class> klass(hs.NewHandle(soa.Decode<mirror::Class*>(annotationType)));
   return method->GetDexFile()->IsMethodAnnotationPresent(method, klass);
 }
diff --git a/runtime/oat.h b/runtime/oat.h
index 543d99f2a..57675dc73 100644
--- a/runtime/oat.h
+++ b/runtime/oat.h
@@ -43,7 +43,7 @@ class PACKED(4) OatHeader {
   static constexpr const char* kNativeDebuggableKey = "native-debuggable";
   static constexpr const char* kCompilerFilter = "compiler-filter";
   static constexpr const char* kClassPathKey = "classpath";
-  static constexpr const char* kBootClassPath = "bootclasspath";
+  static constexpr const char* kBootClassPathKey = "bootclasspath";
 
   static constexpr const char kTrueValue[] = "true";
   static constexpr const char kFalseValue[] = "false";
diff --git a/runtime/oat_file.h b/runtime/oat_file.h
index 11a9d76da..aa727ff45 100644
--- a/runtime/oat_file.h
+++ b/runtime/oat_file.h
@@ -48,6 +48,9 @@ class DummyOatFile;
 
 class OatFile {
  public:
+  // Special classpath that skips shared library check.
+  static constexpr const char* kSpecialSharedLibrary = "&";
+
   typedef art::OatDexFile OatDexFile;
 
   // Opens an oat file contained within the given elf file. This is always opened as
@@ -370,6 +373,10 @@ class OatDexFile FINAL {
     return lookup_table_data_;
   }
 
+  const uint8_t* GetDexFilePointer() const {
+    return dex_file_pointer_;
+  }
+
   ~OatDexFile();
 
  private:
diff --git a/runtime/oat_file_assistant.cc b/runtime/oat_file_assistant.cc
index 3f95772b4..fba10ca01 100644
--- a/runtime/oat_file_assistant.cc
+++ b/runtime/oat_file_assistant.cc
@@ -30,6 +30,7 @@
 
 #include "base/logging.h"
 #include "base/stringprintf.h"
+#include "compiler_filter.h"
 #include "class_linker.h"
 #include "gc/heap.h"
 #include "gc/space/image_space.h"
@@ -43,6 +44,24 @@
 
 namespace art {
 
+std::ostream& operator << (std::ostream& stream, const OatFileAssistant::OatStatus status) {
+  switch (status) {
+    case OatFileAssistant::kOatOutOfDate:
+      stream << "kOatOutOfDate";
+      break;
+    case OatFileAssistant::kOatUpToDate:
+      stream << "kOatUpToDate";
+      break;
+    case OatFileAssistant::kOatNeedsRelocation:
+      stream << "kOatNeedsRelocation";
+      break;
+    default:
+      UNREACHABLE();
+  }
+
+  return stream;
+}
+
 OatFileAssistant::OatFileAssistant(const char* dex_location,
                                    const InstructionSet isa,
                                    bool profile_changed,
@@ -179,11 +198,38 @@ OatFileAssistant::DexOptNeeded OatFileAssistant::GetDexOptNeeded(CompilerFilter:
   return HasOriginalDexFiles() ? kDex2OatNeeded : kNoDexOptNeeded;
 }
 
+// Figure out the currently specified compile filter option in the runtime.
+// Returns true on success, false if the compiler filter is invalid, in which
+// case error_msg describes the problem.
+static bool GetRuntimeCompilerFilterOption(CompilerFilter::Filter* filter,
+                                           std::string* error_msg) {
+  CHECK(filter != nullptr);
+  CHECK(error_msg != nullptr);
+
+  *filter = CompilerFilter::kDefaultCompilerFilter;
+  for (StringPiece option : Runtime::Current()->GetCompilerOptions()) {
+    if (option.starts_with("--compiler-filter=")) {
+      const char* compiler_filter_string = option.substr(strlen("--compiler-filter=")).data();
+      if (!CompilerFilter::ParseCompilerFilter(compiler_filter_string, filter)) {
+        *error_msg = std::string("Unknown --compiler-filter value: ")
+                   + std::string(compiler_filter_string);
+        return false;
+      }
+    }
+  }
+  return true;
+}
+
 OatFileAssistant::ResultOfAttemptToUpdate
-OatFileAssistant::MakeUpToDate(CompilerFilter::Filter target, std::string* error_msg) {
+OatFileAssistant::MakeUpToDate(std::string* error_msg) {
+  CompilerFilter::Filter target;
+  if (!GetRuntimeCompilerFilterOption(&target, error_msg)) {
+    return kUpdateNotAttempted;
+  }
+
   switch (GetDexOptNeeded(target)) {
     case kNoDexOptNeeded: return kUpdateSucceeded;
-    case kDex2OatNeeded: return GenerateOatFile(target, error_msg);
+    case kDex2OatNeeded: return GenerateOatFile(error_msg);
     case kPatchOatNeeded: return RelocateOatFile(OdexFileName(), error_msg);
     case kSelfPatchOatNeeded: return RelocateOatFile(OatFileName(), error_msg);
   }
@@ -350,6 +396,12 @@ bool OatFileAssistant::OdexFileIsUpToDate() {
   return cached_odex_file_is_up_to_date_;
 }
 
+CompilerFilter::Filter OatFileAssistant::OdexFileCompilerFilter() {
+  const OatFile* odex_file = GetOdexFile();
+  CHECK(odex_file != nullptr);
+
+  return odex_file->GetCompilerFilter();
+}
 std::string OatFileAssistant::ArtFileName(const OatFile* oat_file) const {
   const std::string oat_file_location = oat_file->GetLocation();
   // Replace extension with .art
@@ -428,6 +480,13 @@ bool OatFileAssistant::OatFileIsUpToDate() {
   return cached_oat_file_is_up_to_date_;
 }
 
+CompilerFilter::Filter OatFileAssistant::OatFileCompilerFilter() {
+  const OatFile* oat_file = GetOatFile();
+  CHECK(oat_file != nullptr);
+
+  return oat_file->GetCompilerFilter();
+}
+
 OatFileAssistant::OatStatus OatFileAssistant::GivenOatFileStatus(const OatFile& file) {
   // TODO: This could cause GivenOatFileIsOutOfDate to be called twice, which
   // is more work than we need to do. If performance becomes a concern, and
@@ -634,7 +693,7 @@ OatFileAssistant::RelocateOatFile(const std::string* input_file, std::string* er
 }
 
 OatFileAssistant::ResultOfAttemptToUpdate
-OatFileAssistant::GenerateOatFile(CompilerFilter::Filter target, std::string* error_msg) {
+OatFileAssistant::GenerateOatFile(std::string* error_msg) {
   CHECK(error_msg != nullptr);
 
   Runtime* runtime = Runtime::Current();
@@ -678,7 +737,6 @@ OatFileAssistant::GenerateOatFile(CompilerFilter::Filter target, std::string* er
   args.push_back("--dex-file=" + dex_location_);
   args.push_back("--oat-fd=" + std::to_string(oat_file->Fd()));
   args.push_back("--oat-location=" + oat_file_name);
-  args.push_back("--compiler-filter=" + CompilerFilter::NameOfFilter(target));
 
   if (!Dex2Oat(args, error_msg)) {
     // Manually delete the file. This ensures there is no garbage left over if
@@ -713,7 +771,11 @@ bool OatFileAssistant::Dex2Oat(const std::vector<std::string>& args,
   argv.push_back("--runtime-arg");
   argv.push_back("-classpath");
   argv.push_back("--runtime-arg");
-  argv.push_back(runtime->GetClassPathString());
+  std::string class_path = runtime->GetClassPathString();
+  if (class_path == "") {
+    class_path = OatFile::kSpecialSharedLibrary;
+  }
+  argv.push_back(class_path);
   if (runtime->IsDebuggable()) {
     argv.push_back("--debuggable");
   }
diff --git a/runtime/oat_file_assistant.h b/runtime/oat_file_assistant.h
index d3228deac..f48cdf343 100644
--- a/runtime/oat_file_assistant.h
+++ b/runtime/oat_file_assistant.h
@@ -19,6 +19,7 @@
 
 #include <cstdint>
 #include <memory>
+#include <sstream>
 #include <string>
 
 #include "arch/instruction_set.h"
@@ -159,15 +160,12 @@ class OatFileAssistant {
   };
 
   // Attempts to generate or relocate the oat file as needed to make it up to
-  // date with in a way that is at least as good as an oat file generated with
-  // the given compiler filter.
-  // Returns the result of attempting to update the code.
+  // date based on the current runtime and compiler options.
   //
   // If the result is not kUpdateSucceeded, the value of error_msg will be set
   // to a string describing why there was a failure or the update was not
   // attempted. error_msg must not be null.
-  ResultOfAttemptToUpdate MakeUpToDate(CompilerFilter::Filter target_compiler_filter,
-                                       std::string* error_msg);
+  ResultOfAttemptToUpdate MakeUpToDate(std::string* error_msg);
 
   // Returns an oat file that can be used for loading dex files.
   // Returns null if no suitable oat file was found.
@@ -214,6 +212,9 @@ class OatFileAssistant {
   bool OdexFileIsOutOfDate();
   bool OdexFileNeedsRelocation();
   bool OdexFileIsUpToDate();
+  // Must only be called if the associated odex file exists, i.e, if
+  // |OdexFileExists() == true|.
+  CompilerFilter::Filter OdexFileCompilerFilter();
 
   // When the dex files is compiled on the target device, the oat file is the
   // result. The oat file will have been relocated to some
@@ -230,6 +231,9 @@ class OatFileAssistant {
   bool OatFileIsOutOfDate();
   bool OatFileNeedsRelocation();
   bool OatFileIsUpToDate();
+  // Must only be called if the associated oat file exists, i.e, if
+  // |OatFileExists() == true|.
+  CompilerFilter::Filter OatFileCompilerFilter();
 
   // Return image file name. Does not cache since it relies on the oat file.
   std::string ArtFileName(const OatFile* oat_file) const;
@@ -250,14 +254,15 @@ class OatFileAssistant {
   // attempted. error_msg must not be null.
   ResultOfAttemptToUpdate RelocateOatFile(const std::string* input_file, std::string* error_msg);
 
-  // Generate the oat file from the dex file using the given compiler filter.
+  // Generate the oat file from the dex file using the current runtime
+  // compiler options.
   // This does not check the current status before attempting to generate the
   // oat file.
   //
   // If the result is not kUpdateSucceeded, the value of error_msg will be set
   // to a string describing why there was a failure or the update was not
   // attempted. error_msg must not be null.
-  ResultOfAttemptToUpdate GenerateOatFile(CompilerFilter::Filter filter, std::string* error_msg);
+  ResultOfAttemptToUpdate GenerateOatFile(std::string* error_msg);
 
   // Executes dex2oat using the current runtime configuration overridden with
   // the given arguments. This does not check to see if dex2oat is enabled in
@@ -438,6 +443,8 @@ class OatFileAssistant {
   DISALLOW_COPY_AND_ASSIGN(OatFileAssistant);
 };
 
+std::ostream& operator << (std::ostream& stream, const OatFileAssistant::OatStatus status);
+
 }  // namespace art
 
 #endif  // ART_RUNTIME_OAT_FILE_ASSISTANT_H_
diff --git a/runtime/oat_file_assistant_test.cc b/runtime/oat_file_assistant_test.cc
index f50d1cb74..15a1aa4d1 100644
--- a/runtime/oat_file_assistant_test.cc
+++ b/runtime/oat_file_assistant_test.cc
@@ -453,8 +453,7 @@ TEST_F(OatFileAssistantTest, NoDexNoOat) {
 
   // Trying to make the oat file up to date should not fail or crash.
   std::string error_msg;
-  EXPECT_EQ(OatFileAssistant::kUpdateSucceeded,
-      oat_file_assistant.MakeUpToDate(CompilerFilter::kSpeed, &error_msg));
+  EXPECT_EQ(OatFileAssistant::kUpdateSucceeded, oat_file_assistant.MakeUpToDate(&error_msg));
 
   // Trying to get the best oat file should fail, but not crash.
   std::unique_ptr<OatFile> oat_file = oat_file_assistant.GetBestOatFile();
@@ -705,8 +704,9 @@ TEST_F(OatFileAssistantTest, StrippedDexOdexNoOat) {
 
   // Make the oat file up to date.
   std::string error_msg;
+  Runtime::Current()->AddCompilerOption("--compiler-filter=speed");
   ASSERT_EQ(OatFileAssistant::kUpdateSucceeded,
-      oat_file_assistant.MakeUpToDate(CompilerFilter::kSpeed, &error_msg)) << error_msg;
+      oat_file_assistant.MakeUpToDate(&error_msg)) << error_msg;
 
   EXPECT_EQ(OatFileAssistant::kNoDexOptNeeded,
       oat_file_assistant.GetDexOptNeeded(CompilerFilter::kSpeed));
@@ -768,8 +768,9 @@ TEST_F(OatFileAssistantTest, StrippedDexOdexOat) {
 
   // Make the oat file up to date.
   std::string error_msg;
+  Runtime::Current()->AddCompilerOption("--compiler-filter=speed");
   ASSERT_EQ(OatFileAssistant::kUpdateSucceeded,
-      oat_file_assistant.MakeUpToDate(CompilerFilter::kSpeed, &error_msg)) << error_msg;
+      oat_file_assistant.MakeUpToDate(&error_msg)) << error_msg;
 
   EXPECT_EQ(OatFileAssistant::kNoDexOptNeeded,
       oat_file_assistant.GetDexOptNeeded(CompilerFilter::kSpeed));
@@ -825,8 +826,9 @@ TEST_F(OatFileAssistantTest, ResourceOnlyDex) {
 
   // Make the oat file up to date. This should have no effect.
   std::string error_msg;
+  Runtime::Current()->AddCompilerOption("--compiler-filter=speed");
   EXPECT_EQ(OatFileAssistant::kUpdateSucceeded,
-      oat_file_assistant.MakeUpToDate(CompilerFilter::kSpeed, &error_msg)) << error_msg;
+      oat_file_assistant.MakeUpToDate(&error_msg)) << error_msg;
 
   EXPECT_EQ(OatFileAssistant::kNoDexOptNeeded,
       oat_file_assistant.GetDexOptNeeded(CompilerFilter::kSpeed));
@@ -876,8 +878,9 @@ TEST_F(OatFileAssistantTest, SelfRelocation) {
 
   // Make the oat file up to date.
   std::string error_msg;
+  Runtime::Current()->AddCompilerOption("--compiler-filter=speed");
   ASSERT_EQ(OatFileAssistant::kUpdateSucceeded,
-      oat_file_assistant.MakeUpToDate(CompilerFilter::kSpeed, &error_msg)) << error_msg;
+      oat_file_assistant.MakeUpToDate(&error_msg)) << error_msg;
 
   EXPECT_EQ(OatFileAssistant::kNoDexOptNeeded,
       oat_file_assistant.GetDexOptNeeded(CompilerFilter::kSpeed));
@@ -920,8 +923,9 @@ TEST_F(OatFileAssistantTest, NoSelfRelocation) {
 
   // Make the oat file up to date.
   std::string error_msg;
+  Runtime::Current()->AddCompilerOption("--compiler-filter=speed");
   ASSERT_EQ(OatFileAssistant::kUpdateSucceeded,
-      oat_file_assistant.MakeUpToDate(CompilerFilter::kSpeed, &error_msg)) << error_msg;
+      oat_file_assistant.MakeUpToDate(&error_msg)) << error_msg;
   EXPECT_EQ(OatFileAssistant::kNoDexOptNeeded,
       oat_file_assistant.GetDexOptNeeded(CompilerFilter::kSpeed));
 
@@ -1100,8 +1104,9 @@ TEST_F(OatFileAssistantTest, LoadDexNoAlternateOat) {
   OatFileAssistant oat_file_assistant(
       dex_location.c_str(), oat_location.c_str(), kRuntimeISA, false, true);
   std::string error_msg;
+  Runtime::Current()->AddCompilerOption("--compiler-filter=speed");
   ASSERT_EQ(OatFileAssistant::kUpdateSucceeded,
-      oat_file_assistant.MakeUpToDate(CompilerFilter::kSpeed, &error_msg)) << error_msg;
+      oat_file_assistant.MakeUpToDate(&error_msg)) << error_msg;
 
   std::unique_ptr<OatFile> oat_file = oat_file_assistant.GetBestOatFile();
   ASSERT_TRUE(oat_file.get() != nullptr);
@@ -1131,8 +1136,9 @@ TEST_F(OatFileAssistantTest, LoadDexUnwriteableAlternateOat) {
   OatFileAssistant oat_file_assistant(
       dex_location.c_str(), oat_location.c_str(), kRuntimeISA, false, true);
   std::string error_msg;
+  Runtime::Current()->AddCompilerOption("--compiler-filter=speed");
   ASSERT_EQ(OatFileAssistant::kUpdateNotAttempted,
-      oat_file_assistant.MakeUpToDate(CompilerFilter::kSpeed, &error_msg));
+      oat_file_assistant.MakeUpToDate(&error_msg));
 
   std::unique_ptr<OatFile> oat_file = oat_file_assistant.GetBestOatFile();
   ASSERT_TRUE(oat_file.get() == nullptr);
@@ -1147,8 +1153,9 @@ TEST_F(OatFileAssistantTest, GenNoDex) {
   OatFileAssistant oat_file_assistant(
       dex_location.c_str(), oat_location.c_str(), kRuntimeISA, false, true);
   std::string error_msg;
+  Runtime::Current()->AddCompilerOption("--compiler-filter=speed");
   EXPECT_EQ(OatFileAssistant::kUpdateNotAttempted,
-      oat_file_assistant.GenerateOatFile(CompilerFilter::kSpeed, &error_msg));
+      oat_file_assistant.GenerateOatFile(&error_msg));
 }
 
 // Turn an absolute path into a path relative to the current working
@@ -1227,8 +1234,9 @@ TEST_F(OatFileAssistantTest, ShortDexLocation) {
 
   // Trying to make it up to date should have no effect.
   std::string error_msg;
+  Runtime::Current()->AddCompilerOption("--compiler-filter=speed");
   EXPECT_EQ(OatFileAssistant::kUpdateSucceeded,
-      oat_file_assistant.MakeUpToDate(CompilerFilter::kSpeed, &error_msg));
+      oat_file_assistant.MakeUpToDate(&error_msg));
   EXPECT_TRUE(error_msg.empty());
 }
 
@@ -1256,8 +1264,7 @@ TEST_F(OatFileAssistantTest, LongDexExtension) {
 class RaceGenerateTask : public Task {
  public:
   explicit RaceGenerateTask(const std::string& dex_location, const std::string& oat_location)
-    : dex_location_(dex_location), oat_location_(oat_location),
-      loaded_oat_file_(nullptr)
+    : dex_location_(dex_location), oat_location_(oat_location), loaded_oat_file_(nullptr)
   {}
 
   void Run(Thread* self ATTRIBUTE_UNUSED) {
@@ -1368,6 +1375,34 @@ TEST_F(OatFileAssistantNoDex2OatTest, LoadMultiDexOdexNoOat) {
   EXPECT_EQ(2u, dex_files.size());
 }
 
+TEST_F(OatFileAssistantTest, RuntimeCompilerFilterOptionUsed) {
+  std::string dex_location = GetScratchDir() + "/RuntimeCompilerFilterOptionUsed.jar";
+  Copy(GetDexSrc1(), dex_location);
+
+  OatFileAssistant oat_file_assistant(dex_location.c_str(), kRuntimeISA, false, false);
+
+  std::string error_msg;
+  Runtime::Current()->AddCompilerOption("--compiler-filter=interpret-only");
+  EXPECT_EQ(OatFileAssistant::kUpdateSucceeded,
+      oat_file_assistant.MakeUpToDate(&error_msg)) << error_msg;
+  EXPECT_EQ(OatFileAssistant::kNoDexOptNeeded,
+      oat_file_assistant.GetDexOptNeeded(CompilerFilter::kInterpretOnly));
+  EXPECT_EQ(OatFileAssistant::kDex2OatNeeded,
+      oat_file_assistant.GetDexOptNeeded(CompilerFilter::kSpeed));
+
+  Runtime::Current()->AddCompilerOption("--compiler-filter=speed");
+  EXPECT_EQ(OatFileAssistant::kUpdateSucceeded,
+      oat_file_assistant.MakeUpToDate(&error_msg)) << error_msg;
+  EXPECT_EQ(OatFileAssistant::kNoDexOptNeeded,
+      oat_file_assistant.GetDexOptNeeded(CompilerFilter::kInterpretOnly));
+  EXPECT_EQ(OatFileAssistant::kNoDexOptNeeded,
+      oat_file_assistant.GetDexOptNeeded(CompilerFilter::kSpeed));
+
+  Runtime::Current()->AddCompilerOption("--compiler-filter=bogus");
+  EXPECT_EQ(OatFileAssistant::kUpdateNotAttempted,
+      oat_file_assistant.MakeUpToDate(&error_msg));
+}
+
 TEST(OatFileAssistantUtilsTest, DexFilenameToOdexFilename) {
   std::string error_msg;
   std::string odex_file;
diff --git a/runtime/oat_file_manager.cc b/runtime/oat_file_manager.cc
index 9ab0072ea..0af6716af 100644
--- a/runtime/oat_file_manager.cc
+++ b/runtime/oat_file_manager.cc
@@ -36,16 +36,9 @@
 
 namespace art {
 
-// For b/21333911.
-// Only enabled for debug builds to prevent bit rot. There are too many performance regressions for
-// normal builds.
-static constexpr bool kDuplicateClassesCheck = kIsDebugBuild;
-
 // If true, then we attempt to load the application image if it exists.
 static constexpr bool kEnableAppImage = true;
 
-CompilerFilter::Filter OatFileManager::filter_ = CompilerFilter::Filter::kSpeed;
-
 const OatFile* OatFileManager::RegisterOatFile(std::unique_ptr<const OatFile> oat_file) {
   WriterMutexLock mu(Thread::Current(), *Locks::oat_file_manager_lock_);
   DCHECK(oat_file != nullptr);
@@ -175,7 +168,7 @@ class DexFileAndClassPair : ValueObject {
 
   void Next() {
     ++current_class_index_;
-    cached_descriptor_ = GetClassDescriptor(dex_file_.get(), current_class_index_);
+    cached_descriptor_ = GetClassDescriptor(dex_file_, current_class_index_);
   }
 
   size_t GetCurrentClassIndex() const {
@@ -187,7 +180,12 @@ class DexFileAndClassPair : ValueObject {
   }
 
   const DexFile* GetDexFile() const {
-    return dex_file_.get();
+    return dex_file_;
+  }
+
+  void DeleteDexFile() {
+    delete dex_file_;
+    dex_file_ = nullptr;
   }
 
  private:
@@ -198,7 +196,7 @@ class DexFileAndClassPair : ValueObject {
   }
 
   const char* cached_descriptor_;
-  std::shared_ptr<const DexFile> dex_file_;
+  const DexFile* dex_file_;
   size_t current_class_index_;
   bool from_loaded_oat_;  // We only need to compare mismatches between what we load now
                           // and what was loaded before. Any old duplicates must have been
@@ -221,53 +219,299 @@ static void AddDexFilesFromOat(const OatFile* oat_file,
 }
 
 static void AddNext(/*inout*/DexFileAndClassPair* original,
-                    /*inout*/std::priority_queue<DexFileAndClassPair>* heap) {
+                    /*inout*/std::priority_queue<DexFileAndClassPair>* heap,
+                    bool owning_dex_files) {
   if (original->DexFileHasMoreClasses()) {
     original->Next();
     heap->push(std::move(*original));
+  } else if (owning_dex_files) {
+    original->DeleteDexFile();
+  }
+}
+
+static void FreeDexFilesInHeap(std::priority_queue<DexFileAndClassPair>* heap,
+                               bool owning_dex_files) {
+  if (owning_dex_files) {
+    while (!heap->empty()) {
+      delete heap->top().GetDexFile();
+      heap->pop();
+    }
+  }
+}
+
+static void IterateOverJavaDexFile(mirror::Object* dex_file,
+                                   ArtField* const cookie_field,
+                                   std::function<bool(const DexFile*)> fn)
+    SHARED_REQUIRES(Locks::mutator_lock_) {
+  if (dex_file != nullptr) {
+    mirror::LongArray* long_array = cookie_field->GetObject(dex_file)->AsLongArray();
+    if (long_array == nullptr) {
+      // This should never happen so log a warning.
+      LOG(WARNING) << "Null DexFile::mCookie";
+      return;
+    }
+    int32_t long_array_size = long_array->GetLength();
+    // Start from 1 to skip the oat file.
+    for (int32_t j = 1; j < long_array_size; ++j) {
+      const DexFile* cp_dex_file = reinterpret_cast<const DexFile*>(static_cast<uintptr_t>(
+          long_array->GetWithoutChecks(j)));
+      if (!fn(cp_dex_file)) {
+        return;
+      }
+    }
+  }
+}
+
+static void IterateOverPathClassLoader(
+    ScopedObjectAccessAlreadyRunnable& soa,
+    Handle<mirror::ClassLoader> class_loader,
+    MutableHandle<mirror::ObjectArray<mirror::Object>> dex_elements,
+    std::function<bool(const DexFile*)> fn) SHARED_REQUIRES(Locks::mutator_lock_) {
+  // Handle this step.
+  // Handle as if this is the child PathClassLoader.
+  // The class loader is a PathClassLoader which inherits from BaseDexClassLoader.
+  // We need to get the DexPathList and loop through it.
+  ArtField* const cookie_field = soa.DecodeField(WellKnownClasses::dalvik_system_DexFile_cookie);
+  ArtField* const dex_file_field =
+      soa.DecodeField(WellKnownClasses::dalvik_system_DexPathList__Element_dexFile);
+  mirror::Object* dex_path_list =
+      soa.DecodeField(WellKnownClasses::dalvik_system_PathClassLoader_pathList)->
+      GetObject(class_loader.Get());
+  if (dex_path_list != nullptr && dex_file_field != nullptr && cookie_field != nullptr) {
+    // DexPathList has an array dexElements of Elements[] which each contain a dex file.
+    mirror::Object* dex_elements_obj =
+        soa.DecodeField(WellKnownClasses::dalvik_system_DexPathList_dexElements)->
+        GetObject(dex_path_list);
+    // Loop through each dalvik.system.DexPathList$Element's dalvik.system.DexFile and look
+    // at the mCookie which is a DexFile vector.
+    if (dex_elements_obj != nullptr) {
+      dex_elements.Assign(dex_elements_obj->AsObjectArray<mirror::Object>());
+      for (int32_t i = 0; i < dex_elements->GetLength(); ++i) {
+        mirror::Object* element = dex_elements->GetWithoutChecks(i);
+        if (element == nullptr) {
+          // Should never happen, fall back to java code to throw a NPE.
+          break;
+        }
+        mirror::Object* dex_file = dex_file_field->GetObject(element);
+        IterateOverJavaDexFile(dex_file, cookie_field, fn);
+      }
+    }
+  }
+}
+
+static bool GetDexFilesFromClassLoader(
+    ScopedObjectAccessAlreadyRunnable& soa,
+    mirror::ClassLoader* class_loader,
+    std::priority_queue<DexFileAndClassPair>* queue) SHARED_REQUIRES(Locks::mutator_lock_) {
+  if (ClassLinker::IsBootClassLoader(soa, class_loader)) {
+    // The boot class loader. We don't load any of these files, as we know we compiled against
+    // them correctly.
+    return true;
+  }
+
+  // Unsupported class-loader?
+  if (class_loader->GetClass() !=
+      soa.Decode<mirror::Class*>(WellKnownClasses::dalvik_system_PathClassLoader)) {
+    VLOG(class_linker) << "Unsupported class-loader " << PrettyClass(class_loader->GetClass());
+    return false;
+  }
+
+  bool recursive_result = GetDexFilesFromClassLoader(soa, class_loader->GetParent(), queue);
+  if (!recursive_result) {
+    // Something wrong up the chain.
+    return false;
+  }
+
+  // Collect all the dex files.
+  auto GetDexFilesFn = [&] (const DexFile* cp_dex_file)
+            SHARED_REQUIRES(Locks::mutator_lock_) {
+    if (cp_dex_file->NumClassDefs() > 0) {
+      queue->emplace(cp_dex_file, 0U, true);
+    }
+    return true;  // Continue looking.
+  };
+
+  // Handle for dex-cache-element.
+  StackHandleScope<3> hs(soa.Self());
+  MutableHandle<mirror::ObjectArray<mirror::Object>> dex_elements(
+      hs.NewHandle<mirror::ObjectArray<mirror::Object>>(nullptr));
+  Handle<mirror::ClassLoader> h_class_loader(hs.NewHandle(class_loader));
+
+  IterateOverPathClassLoader(soa, h_class_loader, dex_elements, GetDexFilesFn);
+
+  return true;
+}
+
+static void GetDexFilesFromDexElementsArray(
+    ScopedObjectAccessAlreadyRunnable& soa,
+    Handle<mirror::ObjectArray<mirror::Object>> dex_elements,
+    std::priority_queue<DexFileAndClassPair>* queue) SHARED_REQUIRES(Locks::mutator_lock_) {
+  if (dex_elements.Get() == nullptr) {
+    // Nothing to do.
+    return;
+  }
+
+  ArtField* const cookie_field = soa.DecodeField(WellKnownClasses::dalvik_system_DexFile_cookie);
+  ArtField* const dex_file_field =
+      soa.DecodeField(WellKnownClasses::dalvik_system_DexPathList__Element_dexFile);
+  const mirror::Class* const element_class = soa.Decode<mirror::Class*>(
+      WellKnownClasses::dalvik_system_DexPathList__Element);
+  const mirror::Class* const dexfile_class = soa.Decode<mirror::Class*>(
+        WellKnownClasses::dalvik_system_DexFile);
+
+  // Collect all the dex files.
+  auto GetDexFilesFn = [&] (const DexFile* cp_dex_file)
+      SHARED_REQUIRES(Locks::mutator_lock_) {
+    if (cp_dex_file != nullptr && cp_dex_file->NumClassDefs() > 0) {
+      queue->emplace(cp_dex_file, 0U, true);
+    }
+    return true;  // Continue looking.
+  };
+
+  for (int32_t i = 0; i < dex_elements->GetLength(); ++i) {
+    mirror::Object* element = dex_elements->GetWithoutChecks(i);
+    if (element == nullptr) {
+      continue;
+    }
+
+    // We support this being dalvik.system.DexPathList$Element and dalvik.system.DexFile.
+
+    mirror::Object* dex_file;
+    if (element->GetClass() == element_class) {
+      dex_file = dex_file_field->GetObject(element);
+    } else if (element->GetClass() == dexfile_class) {
+      dex_file = element;
+    } else {
+      LOG(WARNING) << "Unsupported element in dex_elements: " << PrettyClass(element->GetClass());
+      continue;
+    }
+
+    IterateOverJavaDexFile(dex_file, cookie_field, GetDexFilesFn);
+  }
+}
+
+static bool AreSharedLibrariesOk(const std::string shared_libraries,
+                                 std::priority_queue<DexFileAndClassPair>& queue) {
+  if (shared_libraries.empty()) {
+    if (queue.empty()) {
+      // No shared libraries or oat files, as expected.
+      return true;
+    }
+  } else {
+    if (shared_libraries.compare(OatFile::kSpecialSharedLibrary) == 0) {
+      // If we find the special shared library, skip the shared libraries check.
+      return true;
+    }
+    // Shared libraries is a series of dex file paths and their checksums, each separated by '*'.
+    std::vector<std::string> shared_libraries_split;
+    Split(shared_libraries, '*', &shared_libraries_split);
+
+    size_t index = 0;
+    std::priority_queue<DexFileAndClassPair> temp = queue;
+    while (!temp.empty() && index < shared_libraries_split.size() - 1) {
+      DexFileAndClassPair pair(temp.top());
+      const DexFile* dex_file = pair.GetDexFile();
+      std::string dex_filename(dex_file->GetLocation());
+      uint32_t dex_checksum = dex_file->GetLocationChecksum();
+      if (dex_filename != shared_libraries_split[index] ||
+          dex_checksum != std::stoul(shared_libraries_split[index + 1])) {
+        break;
+      }
+      temp.pop();
+      index += 2;
+    }
+
+    // Check is successful if it made it through the queue and all the shared libraries.
+    return temp.empty() && index == shared_libraries_split.size();
   }
+  return false;
 }
 
 // Check for class-def collisions in dex files.
 //
-// This works by maintaining a heap with one class from each dex file, sorted by the class
-// descriptor. Then a dex-file/class pair is continually removed from the heap and compared
+// This first walks the class loader chain, getting all the dex files from the class loader. If
+// the class loader is null or one of the class loaders in the chain is unsupported, we collect
+// dex files from all open non-boot oat files to be safe.
+//
+// This first checks whether the shared libraries are in the expected order and the oat files
+// have the expected checksums. If so, we exit early. Otherwise, we do the collision check.
+//
+// The collision check works by maintaining a heap with one class from each dex file, sorted by the
+// class descriptor. Then a dex-file/class pair is continually removed from the heap and compared
 // against the following top element. If the descriptor is the same, it is now checked whether
 // the two elements agree on whether their dex file was from an already-loaded oat-file or the
 // new oat file. Any disagreement indicates a collision.
 bool OatFileManager::HasCollisions(const OatFile* oat_file,
+                                   jobject class_loader,
+                                   jobjectArray dex_elements,
                                    std::string* error_msg /*out*/) const {
   DCHECK(oat_file != nullptr);
   DCHECK(error_msg != nullptr);
-  if (!kDuplicateClassesCheck) {
-    return false;
+
+  std::priority_queue<DexFileAndClassPair> queue;
+  bool owning_dex_files = false;
+
+  // Try to get dex files from the given class loader. If the class loader is null, or we do
+  // not support one of the class loaders in the chain, conservatively compare against all
+  // (non-boot) oat files.
+  bool class_loader_ok = false;
+  {
+    ScopedObjectAccess soa(Thread::Current());
+    StackHandleScope<2> hs(Thread::Current());
+    Handle<mirror::ClassLoader> h_class_loader =
+        hs.NewHandle(soa.Decode<mirror::ClassLoader*>(class_loader));
+    Handle<mirror::ObjectArray<mirror::Object>> h_dex_elements =
+        hs.NewHandle(soa.Decode<mirror::ObjectArray<mirror::Object>*>(dex_elements));
+    if (h_class_loader.Get() != nullptr &&
+        GetDexFilesFromClassLoader(soa, h_class_loader.Get(), &queue)) {
+      class_loader_ok = true;
+
+      // In this case, also take into account the dex_elements array, if given. We don't need to
+      // read it otherwise, as we'll compare against all open oat files anyways.
+      GetDexFilesFromDexElementsArray(soa, h_dex_elements, &queue);
+    } else if (h_class_loader.Get() != nullptr) {
+      VLOG(class_linker) << "Something unsupported with "
+                         << PrettyClass(h_class_loader->GetClass());
+    }
   }
 
   // Dex files are registered late - once a class is actually being loaded. We have to compare
   // against the open oat files. Take the oat_file_manager_lock_ that protects oat_files_ accesses.
   ReaderMutexLock mu(Thread::Current(), *Locks::oat_file_manager_lock_);
 
-  std::priority_queue<DexFileAndClassPair> queue;
+  if (!class_loader_ok) {
+    // Add dex files from already loaded oat files, but skip boot.
 
-  // Add dex files from already loaded oat files, but skip boot.
-  std::vector<const OatFile*> boot_oat_files = GetBootOatFiles();
-  // The same OatFile can be loaded multiple times at different addresses. In this case, we don't
-  // need to check both against each other since they would have resolved the same way at compile
-  // time.
-  std::unordered_set<std::string> unique_locations;
-  for (const std::unique_ptr<const OatFile>& loaded_oat_file : oat_files_) {
-    DCHECK_NE(loaded_oat_file.get(), oat_file);
-    const std::string& location = loaded_oat_file->GetLocation();
-    if (std::find(boot_oat_files.begin(), boot_oat_files.end(), loaded_oat_file.get()) ==
-        boot_oat_files.end() && location != oat_file->GetLocation() &&
-        unique_locations.find(location) == unique_locations.end()) {
-      unique_locations.insert(location);
-      AddDexFilesFromOat(loaded_oat_file.get(), /*already_loaded*/true, &queue);
+    // Clean up the queue.
+    while (!queue.empty()) {
+      queue.pop();
+    }
+
+    // Anything we load now is something we own and must be released later.
+    owning_dex_files = true;
+
+    std::vector<const OatFile*> boot_oat_files = GetBootOatFiles();
+    // The same OatFile can be loaded multiple times at different addresses. In this case, we don't
+    // need to check both against each other since they would have resolved the same way at compile
+    // time.
+    std::unordered_set<std::string> unique_locations;
+    for (const std::unique_ptr<const OatFile>& loaded_oat_file : oat_files_) {
+      DCHECK_NE(loaded_oat_file.get(), oat_file);
+      const std::string& location = loaded_oat_file->GetLocation();
+      if (std::find(boot_oat_files.begin(), boot_oat_files.end(), loaded_oat_file.get()) ==
+          boot_oat_files.end() && location != oat_file->GetLocation() &&
+          unique_locations.find(location) == unique_locations.end()) {
+        unique_locations.insert(location);
+        AddDexFilesFromOat(loaded_oat_file.get(), /*already_loaded*/true, &queue);
+      }
     }
   }
 
-  if (queue.empty()) {
-    // No other oat files, return early.
+  // Exit if shared libraries are ok. Do a full duplicate classes check otherwise.
+  const std::string
+      shared_libraries(oat_file->GetOatHeader().GetStoreValueByKey(OatHeader::kClassPathKey));
+  if (AreSharedLibrariesOk(shared_libraries, queue)) {
+    FreeDexFilesInHeap(&queue, owning_dex_files);
     return false;
   }
 
@@ -292,16 +536,17 @@ bool OatFileManager::HasCollisions(const OatFile* oat_file,
                            compare_pop.GetCachedDescriptor(),
                            compare_pop.GetDexFile()->GetLocation().c_str(),
                            top.GetDexFile()->GetLocation().c_str());
+          FreeDexFilesInHeap(&queue, owning_dex_files);
           return true;
         }
         queue.pop();
-        AddNext(&top, &queue);
+        AddNext(&top, &queue, owning_dex_files);
       } else {
         // Something else. Done here.
         break;
       }
     }
-    AddNext(&compare_pop, &queue);
+    AddNext(&compare_pop, &queue, owning_dex_files);
   }
 
   return false;
@@ -341,9 +586,10 @@ std::vector<std::unique_ptr<const DexFile>> OatFileManager::OpenDexFilesFromOat(
 
   const OatFile* source_oat_file = nullptr;
 
-  // Update the oat file on disk if we can. This may fail, but that's okay.
-  // Best effort is all that matters here.
-  switch (oat_file_assistant.MakeUpToDate(filter_, /*out*/ &error_msg)) {
+  // Update the oat file on disk if we can, based on the --compiler-filter
+  // option derived from the current runtime options.
+  // This may fail, but that's okay. Best effort is all that matters here.
+  switch (oat_file_assistant.MakeUpToDate(/*out*/ &error_msg)) {
     case OatFileAssistant::kUpdateFailed:
       LOG(WARNING) << error_msg;
       break;
@@ -364,7 +610,8 @@ std::vector<std::unique_ptr<const DexFile>> OatFileManager::OpenDexFilesFromOat(
 
   if (oat_file != nullptr) {
     // Take the file only if it has no collisions, or we must take it because of preopting.
-    bool accept_oat_file = !HasCollisions(oat_file.get(), /*out*/ &error_msg);
+    bool accept_oat_file =
+        !HasCollisions(oat_file.get(), class_loader, dex_elements, /*out*/ &error_msg);
     if (!accept_oat_file) {
       // Failed the collision check. Print warning.
       if (Runtime::Current()->IsDexFileFallbackEnabled()) {
diff --git a/runtime/oat_file_manager.h b/runtime/oat_file_manager.h
index f98102e84..a1d1275e6 100644
--- a/runtime/oat_file_manager.h
+++ b/runtime/oat_file_manager.h
@@ -25,7 +25,6 @@
 
 #include "base/macros.h"
 #include "base/mutex.h"
-#include "compiler_filter.h"
 #include "jni.h"
 
 namespace art {
@@ -116,14 +115,17 @@ class OatFileManager {
 
   void DumpForSigQuit(std::ostream& os);
 
-  static void SetCompilerFilter(CompilerFilter::Filter filter) {
-    filter_ = filter;
-  }
-
  private:
-  // Check for duplicate class definitions of the given oat file against all open oat files.
+  // Check that the shared libraries in the given oat file match those in the given class loader and
+  // dex elements. If the class loader is null or we do not support one of the class loaders in the
+  // chain, compare against all non-boot oat files instead. If the shared libraries are not ok,
+  // check for duplicate class definitions of the given oat file against the oat files (either from
+  // the class loader and dex elements if possible or all non-boot oat files otherwise).
   // Return true if there are any class definition collisions in the oat_file.
-  bool HasCollisions(const OatFile* oat_file, /*out*/std::string* error_msg) const
+  bool HasCollisions(const OatFile* oat_file,
+                     jobject class_loader,
+                     jobjectArray dex_elements,
+                     /*out*/ std::string* error_msg) const
       REQUIRES(!Locks::oat_file_manager_lock_);
 
   const OatFile* FindOpenedOatFileFromOatLocationLocked(const std::string& oat_location) const
@@ -133,9 +135,6 @@ class OatFileManager {
   std::unordered_map<std::string, size_t> oat_file_count_ GUARDED_BY(Locks::oat_file_count_lock_);
   bool have_non_pic_oat_file_;
 
-  // The compiler filter used for oat files loaded by the oat file manager.
-  static CompilerFilter::Filter filter_;
-
   DISALLOW_COPY_AND_ASSIGN(OatFileManager);
 };
 
diff --git a/runtime/parsed_options.cc b/runtime/parsed_options.cc
index c8d429158..eac5b43ff 100644
--- a/runtime/parsed_options.cc
+++ b/runtime/parsed_options.cc
@@ -153,7 +153,7 @@ std::unique_ptr<RuntimeParser> ParsedOptions::MakeParser(bool ignore_unrecognize
       .Define("-Xusejit:_")
           .WithType<bool>()
           .WithValueMap({{"false", false}, {"true", true}})
-          .IntoKey(M::UseJIT)
+          .IntoKey(M::UseJitCompilation)
       .Define("-Xjitinitialsize:_")
           .WithType<MemoryKiB>()
           .IntoKey(M::JITCodeCacheInitialCapacity)
@@ -172,6 +172,9 @@ std::unique_ptr<RuntimeParser> ParsedOptions::MakeParser(bool ignore_unrecognize
       .Define("-Xjitprithreadweight:_")
           .WithType<unsigned int>()
           .IntoKey(M::JITPriorityThreadWeight)
+      .Define("-Xjittransitionweight:_")
+          .WithType<unsigned int>()
+          .IntoKey(M::JITInvokeTransitionWeight)
       .Define("-Xjitsaveprofilinginfo")
           .WithValue(true)
           .IntoKey(M::JITSaveProfilingInfo)
@@ -289,9 +292,6 @@ std::unique_ptr<RuntimeParser> ParsedOptions::MakeParser(bool ignore_unrecognize
           .IntoKey(M::Experimental)
       .Define("-Xforce-nb-testing")
           .IntoKey(M::ForceNativeBridge)
-      .Define("-XOatFileManagerCompilerFilter:_")
-          .WithType<std::string>()
-          .IntoKey(M::OatFileManagerCompilerFilter)
       .Ignore({
           "-ea", "-da", "-enableassertions", "-disableassertions", "--runtime-arg", "-esa",
           "-dsa", "-enablesystemassertions", "-disablesystemassertions", "-Xrs", "-Xint:_",
@@ -471,6 +471,11 @@ bool ParsedOptions::DoParse(const RuntimeOptions& options,
     LOG(INFO) << "setting boot class path to " << *args.Get(M::BootClassPath);
   }
 
+  if (args.GetOrDefault(M::UseJitCompilation) && args.GetOrDefault(M::Interpret)) {
+    Usage("-Xusejit:true and -Xint cannot be specified together");
+    Exit(0);
+  }
+
   // Set a default boot class path if we didn't get an explicit one via command line.
   if (getenv("BOOTCLASSPATH") != nullptr) {
     args.SetIfMissing(M::BootClassPath, std::string(getenv("BOOTCLASSPATH")));
diff --git a/runtime/quick/inline_method_analyser.cc b/runtime/quick/inline_method_analyser.cc
index c7ccee212..1dea562b5 100644
--- a/runtime/quick/inline_method_analyser.cc
+++ b/runtime/quick/inline_method_analyser.cc
@@ -434,7 +434,7 @@ static_assert(InlineMethodAnalyser::IGetVariant(Instruction::IGET_SHORT) ==
 bool InlineMethodAnalyser::AnalyseMethodCode(verifier::MethodVerifier* verifier,
                                              InlineMethod* result) {
   DCHECK(verifier != nullptr);
-  if (!Runtime::Current()->UseJit()) {
+  if (!Runtime::Current()->UseJitCompilation()) {
     DCHECK_EQ(verifier->CanLoadClasses(), result != nullptr);
   }
 
diff --git a/runtime/quick_exception_handler.cc b/runtime/quick_exception_handler.cc
index a785ecba3..a3e1f0020 100644
--- a/runtime/quick_exception_handler.cc
+++ b/runtime/quick_exception_handler.cc
@@ -509,7 +509,7 @@ void QuickExceptionHandler::DeoptimizeSingleFrame() {
   // Compiled code made an explicit deoptimization.
   ArtMethod* deopt_method = visitor.GetSingleFrameDeoptMethod();
   DCHECK(deopt_method != nullptr);
-  if (Runtime::Current()->UseJit()) {
+  if (Runtime::Current()->UseJitCompilation()) {
     Runtime::Current()->GetJit()->GetCodeCache()->InvalidateCompiledCodeFor(
         deopt_method, visitor.GetSingleFrameDeoptQuickMethodHeader());
   } else {
@@ -611,7 +611,7 @@ void QuickExceptionHandler::DoLongJump(bool smash_caller_saves) {
 // Prints out methods with their type of frame.
 class DumpFramesWithTypeStackVisitor FINAL : public StackVisitor {
  public:
-  DumpFramesWithTypeStackVisitor(Thread* self, bool show_details = false)
+  explicit DumpFramesWithTypeStackVisitor(Thread* self, bool show_details = false)
       SHARED_REQUIRES(Locks::mutator_lock_)
       : StackVisitor(self, nullptr, StackVisitor::StackWalkKind::kIncludeInlinedFrames),
         show_details_(show_details) {}
diff --git a/runtime/runtime.cc b/runtime/runtime.cc
index a4d31ef12..63976d0b1 100644
--- a/runtime/runtime.cc
+++ b/runtime/runtime.cc
@@ -60,7 +60,6 @@
 #include "base/unix_file/fd_file.h"
 #include "class_linker-inl.h"
 #include "compiler_callbacks.h"
-#include "compiler_filter.h"
 #include "debugger.h"
 #include "elf_file.h"
 #include "entrypoints/runtime_asm_entrypoints.h"
@@ -558,23 +557,34 @@ bool Runtime::Start() {
 
   started_ = true;
 
-  if (jit_options_->UseJIT()) {
+  // Create the JIT either if we have to use JIT compilation or save profiling info.
+  // TODO(calin): We use the JIT class as a proxy for JIT compilation and for
+  // recoding profiles. Maybe we should consider changing the name to be more clear it's
+  // not only about compiling. b/28295073.
+  if (jit_options_->UseJitCompilation() || jit_options_->GetSaveProfilingInfo()) {
     std::string error_msg;
     if (!IsZygote()) {
     // If we are the zygote then we need to wait until after forking to create the code cache
     // due to SELinux restrictions on r/w/x memory regions.
       CreateJit();
-    } else if (!jit::Jit::LoadCompilerLibrary(&error_msg)) {
-      // Try to load compiler pre zygote to reduce PSS. b/27744947
-      LOG(WARNING) << "Failed to load JIT compiler with error " << error_msg;
+    } else if (jit_options_->UseJitCompilation()) {
+      if (!jit::Jit::LoadCompilerLibrary(&error_msg)) {
+        // Try to load compiler pre zygote to reduce PSS. b/27744947
+        LOG(WARNING) << "Failed to load JIT compiler with error " << error_msg;
+      }
     }
   }
 
   if (!IsImageDex2OatEnabled() || !GetHeap()->HasBootImageSpace()) {
     ScopedObjectAccess soa(self);
-    StackHandleScope<1> hs(soa.Self());
-    auto klass(hs.NewHandle<mirror::Class>(mirror::Class::GetJavaLangClass()));
-    class_linker_->EnsureInitialized(soa.Self(), klass, true, true);
+    StackHandleScope<2> hs(soa.Self());
+
+    auto class_class(hs.NewHandle<mirror::Class>(mirror::Class::GetJavaLangClass()));
+    auto field_class(hs.NewHandle<mirror::Class>(mirror::Field::StaticClass()));
+
+    class_linker_->EnsureInitialized(soa.Self(), class_class, true, true);
+    // Field class is needed for register_java_net_InetAddress in libcore, b/28153851.
+    class_linker_->EnsureInitialized(soa.Self(), field_class, true, true);
   }
 
   // InitNativeMethods needs to be after started_ so that the classes
@@ -713,7 +723,11 @@ void Runtime::InitNonZygoteOrPostFork(
   // before fork aren't attributed to an app.
   heap_->ResetGcPerformanceInfo();
 
-  if (!is_system_server && !safe_mode_ && jit_options_->UseJIT() && jit_.get() == nullptr) {
+
+  if (!is_system_server &&
+      !safe_mode_ &&
+      (jit_options_->UseJitCompilation() || jit_options_->GetSaveProfilingInfo()) &&
+      jit_.get() == nullptr) {
     // Note that when running ART standalone (not zygote, nor zygote fork),
     // the jit may have already been created.
     CreateJit();
@@ -842,7 +856,7 @@ static bool OpenDexFilesFromImage(const std::string& image_location,
     if (index == 0) {
       // First file. See if this is a multi-image environment, and if so, enqueue the other images.
       const OatHeader& boot_oat_header = oat_file->GetOatHeader();
-      const char* boot_cp = boot_oat_header.GetStoreValueByKey(OatHeader::kBootClassPath);
+      const char* boot_cp = boot_oat_header.GetStoreValueByKey(OatHeader::kBootClassPathKey);
       if (boot_cp != nullptr) {
         gc::space::ImageSpace::CreateMultiImageLocations(image_locations[0],
                                                          boot_cp,
@@ -957,16 +971,6 @@ bool Runtime::Init(RuntimeArgumentMap&& runtime_options_in) {
   experimental_flags_ = runtime_options.GetOrDefault(Opt::Experimental);
   is_low_memory_mode_ = runtime_options.Exists(Opt::LowMemoryMode);
 
-  {
-    CompilerFilter::Filter filter;
-    std::string filter_str = runtime_options.GetOrDefault(Opt::OatFileManagerCompilerFilter);
-    if (!CompilerFilter::ParseCompilerFilter(filter_str.c_str(), &filter)) {
-      LOG(ERROR) << "Cannot parse compiler filter " << filter_str;
-      return false;
-    }
-    OatFileManager::SetCompilerFilter(filter);
-  }
-
   XGcOption xgc_option = runtime_options.GetOrDefault(Opt::GcOption);
   heap_ = new gc::Heap(runtime_options.GetOrDefault(Opt::MemoryInitialSize),
                        runtime_options.GetOrDefault(Opt::HeapGrowthLimit),
@@ -1016,7 +1020,8 @@ bool Runtime::Init(RuntimeArgumentMap&& runtime_options_in) {
     // this case.
     // If runtime_options doesn't have UseJIT set to true then CreateFromRuntimeArguments returns
     // null and we don't create the jit.
-    jit_options_->SetUseJIT(false);
+    jit_options_->SetUseJitCompilation(false);
+    jit_options_->SetSaveProfilingInfo(false);
   }
 
   // Allocate a global table of boxed lambda objects <-> closures.
@@ -1613,18 +1618,19 @@ void Runtime::VisitImageRoots(RootVisitor* visitor) {
   }
 }
 
-static ImtConflictTable::Entry empty_entry = { nullptr, nullptr };
-
 ArtMethod* Runtime::CreateImtConflictMethod(LinearAlloc* linear_alloc) {
-  auto* method = Runtime::Current()->GetClassLinker()->CreateRuntimeMethod(linear_alloc);
+  ClassLinker* const class_linker = GetClassLinker();
+  ArtMethod* method = class_linker->CreateRuntimeMethod(linear_alloc);
   // When compiling, the code pointer will get set later when the image is loaded.
+  const size_t pointer_size = GetInstructionSetPointerSize(instruction_set_);
   if (IsAotCompiler()) {
-    size_t pointer_size = GetInstructionSetPointerSize(instruction_set_);
     method->SetEntryPointFromQuickCompiledCodePtrSize(nullptr, pointer_size);
   } else {
     method->SetEntryPointFromQuickCompiledCode(GetQuickImtConflictStub());
-    method->SetImtConflictTable(reinterpret_cast<ImtConflictTable*>(&empty_entry));
   }
+  // Create empty conflict table.
+  method->SetImtConflictTable(class_linker->CreateImtConflictTable(/*count*/0u, linear_alloc),
+                              pointer_size);
   return method;
 }
 
@@ -1632,9 +1638,6 @@ void Runtime::SetImtConflictMethod(ArtMethod* method) {
   CHECK(method != nullptr);
   CHECK(method->IsRuntimeMethod());
   imt_conflict_method_ = method;
-  if (!IsAotCompiler()) {
-    method->SetImtConflictTable(reinterpret_cast<ImtConflictTable*>(&empty_entry));
-  }
 }
 
 ArtMethod* Runtime::CreateResolutionMethod() {
@@ -1915,9 +1918,8 @@ void Runtime::AddCurrentRuntimeFeaturesAsDex2OatArguments(std::vector<std::strin
 
 void Runtime::CreateJit() {
   CHECK(!IsAotCompiler());
-  if (GetInstrumentation()->IsForcedInterpretOnly()) {
-    // Don't create JIT if forced interpret only.
-    return;
+  if (kIsDebugBuild && GetInstrumentation()->IsForcedInterpretOnly()) {
+    DCHECK(!jit_options_->UseJitCompilation());
   }
   std::string error_msg;
   jit_.reset(jit::Jit::Create(jit_options_.get(), &error_msg));
@@ -1944,8 +1946,20 @@ void Runtime::SetImtUnimplementedMethod(ArtMethod* method) {
   CHECK(method != nullptr);
   CHECK(method->IsRuntimeMethod());
   imt_unimplemented_method_ = method;
-  if (!IsAotCompiler()) {
-    method->SetImtConflictTable(reinterpret_cast<ImtConflictTable*>(&empty_entry));
+}
+
+void Runtime::FixupConflictTables() {
+  // We can only do this after the class linker is created.
+  const size_t pointer_size = GetClassLinker()->GetImagePointerSize();
+  if (imt_unimplemented_method_->GetImtConflictTable(pointer_size) == nullptr) {
+    imt_unimplemented_method_->SetImtConflictTable(
+        ClassLinker::CreateImtConflictTable(/*count*/0u, GetLinearAlloc(), pointer_size),
+        pointer_size);
+  }
+  if (imt_conflict_method_->GetImtConflictTable(pointer_size) == nullptr) {
+    imt_conflict_method_->SetImtConflictTable(
+          ClassLinker::CreateImtConflictTable(/*count*/0u, GetLinearAlloc(), pointer_size),
+          pointer_size);
   }
 }
 
@@ -1981,4 +1995,18 @@ void Runtime::UpdateProcessState(ProcessState process_state) {
   GetHeap()->UpdateProcessState(old_process_state, process_state);
 }
 
+void Runtime::RegisterSensitiveThread() const {
+  Thread::SetJitSensitiveThread();
+}
+
+// Returns true if JIT compilations are enabled. GetJit() will be not null in this case.
+bool Runtime::UseJitCompilation() const {
+  return (jit_ != nullptr) && jit_->UseJitCompilation();
+}
+
+// Returns true if profile saving is enabled. GetJit() will be not null in this case.
+bool Runtime::SaveProfileInfo() const {
+  return (jit_ != nullptr) && jit_->SaveProfilingInfo();
+}
+
 }  // namespace art
diff --git a/runtime/runtime.h b/runtime/runtime.h
index ae25dd1c6..1394462fd 100644
--- a/runtime/runtime.h
+++ b/runtime/runtime.h
@@ -127,7 +127,7 @@ class Runtime {
 
   // IsAotCompiler for compilers that don't have a running runtime. Only dex2oat currently.
   bool IsAotCompiler() const {
-    return !UseJit() && IsCompiler();
+    return !UseJitCompilation() && IsCompiler();
   }
 
   // IsCompiler is any runtime which has a running compiler, either dex2oat or JIT.
@@ -383,6 +383,7 @@ class Runtime {
     return imt_conflict_method_ != nullptr;
   }
 
+  void FixupConflictTables();
   void SetImtConflictMethod(ArtMethod* method) SHARED_REQUIRES(Locks::mutator_lock_);
   void SetImtUnimplementedMethod(ArtMethod* method) SHARED_REQUIRES(Locks::mutator_lock_);
 
@@ -451,9 +452,11 @@ class Runtime {
   jit::Jit* GetJit() {
     return jit_.get();
   }
-  bool UseJit() const {
-    return jit_.get() != nullptr;
-  }
+
+  // Returns true if JIT compilations are enabled. GetJit() will be not null in this case.
+  bool UseJitCompilation() const;
+  // Returns true if profile saving is enabled. GetJit() will be not null in this case.
+  bool SaveProfileInfo() const;
 
   void PreZygoteFork();
   bool InitZygote();
@@ -635,6 +638,8 @@ class Runtime {
     return process_state_ == kProcessStateJankPerceptible;
   }
 
+  void RegisterSensitiveThread() const;
+
   void SetZygoteNoThreadSection(bool val) {
     zygote_no_threads_ = val;
   }
diff --git a/runtime/runtime_options.def b/runtime/runtime_options.def
index 6433c3352..635ff5169 100644
--- a/runtime/runtime_options.def
+++ b/runtime/runtime_options.def
@@ -66,12 +66,13 @@ RUNTIME_OPTIONS_KEY (Unit,                IgnoreMaxFootprint)
 RUNTIME_OPTIONS_KEY (Unit,                LowMemoryMode)
 RUNTIME_OPTIONS_KEY (bool,                UseTLAB,                        (kUseTlab || kUseReadBarrier))
 RUNTIME_OPTIONS_KEY (bool,                EnableHSpaceCompactForOOM,      true)
-RUNTIME_OPTIONS_KEY (bool,                UseJIT,                         false)
+RUNTIME_OPTIONS_KEY (bool,                UseJitCompilation,              false)
 RUNTIME_OPTIONS_KEY (bool,                DumpNativeStackOnSigQuit,       true)
 RUNTIME_OPTIONS_KEY (unsigned int,        JITCompileThreshold,            jit::Jit::kDefaultCompileThreshold)
 RUNTIME_OPTIONS_KEY (unsigned int,        JITWarmupThreshold)
 RUNTIME_OPTIONS_KEY (unsigned int,        JITOsrThreshold)
 RUNTIME_OPTIONS_KEY (unsigned int,        JITPriorityThreadWeight)
+RUNTIME_OPTIONS_KEY (unsigned int,        JITInvokeTransitionWeight)
 RUNTIME_OPTIONS_KEY (MemoryKiB,           JITCodeCacheInitialCapacity,    jit::JitCodeCache::kInitialCapacity)
 RUNTIME_OPTIONS_KEY (MemoryKiB,           JITCodeCacheMaxCapacity,        jit::JitCodeCache::kMaxCapacity)
 RUNTIME_OPTIONS_KEY (bool,                JITSaveProfilingInfo,           false)
@@ -128,11 +129,12 @@ RUNTIME_OPTIONS_KEY (CompilerCallbacks*,  CompilerCallbacksPtr)  // TDOO: make u
 RUNTIME_OPTIONS_KEY (bool (*)(),          HookIsSensitiveThread)
 RUNTIME_OPTIONS_KEY (int32_t (*)(FILE* stream, const char* format, va_list ap), \
                                           HookVfprintf,                   vfprintf)
+// Use _exit instead of exit so that we won't get DCHECK failures in global data
+// destructors. b/28106055.
 RUNTIME_OPTIONS_KEY (void (*)(int32_t status), \
-                                          HookExit,                       exit)
+                                          HookExit,                       _exit)
                                                                           // We don't call abort(3) by default; see
                                                                           // Runtime::Abort.
 RUNTIME_OPTIONS_KEY (void (*)(),          HookAbort,                      nullptr)
-RUNTIME_OPTIONS_KEY (std::string,         OatFileManagerCompilerFilter,   "speed")
 
 #undef RUNTIME_OPTIONS_KEY
diff --git a/runtime/stack.cc b/runtime/stack.cc
index 56ef5aaa9..a5ca527aa 100644
--- a/runtime/stack.cc
+++ b/runtime/stack.cc
@@ -637,8 +637,8 @@ static void AssertPcIsWithinQuickCode(ArtMethod* method, uintptr_t pc)
 
   // If we are the JIT then we may have just compiled the method after the
   // IsQuickToInterpreterBridge check.
-  jit::Jit* const jit = Runtime::Current()->GetJit();
-  if (jit != nullptr && jit->GetCodeCache()->ContainsPc(code)) {
+  Runtime* runtime = Runtime::Current();
+  if (runtime->UseJitCompilation() && runtime->GetJit()->GetCodeCache()->ContainsPc(code)) {
     return;
   }
 
@@ -678,8 +678,10 @@ void StackVisitor::SanityCheckFrame() const {
           if (space->IsImageSpace()) {
             auto* image_space = space->AsImageSpace();
             const auto& header = image_space->GetImageHeader();
-            const auto* methods = &header.GetMethodsSection();
-            if (methods->Contains(reinterpret_cast<const uint8_t*>(method) - image_space->Begin())) {
+            const ImageSection& methods = header.GetMethodsSection();
+            const ImageSection& runtime_methods = header.GetRuntimeMethodsSection();
+            const size_t offset =  reinterpret_cast<const uint8_t*>(method) - image_space->Begin();
+            if (methods.Contains(offset) || runtime_methods.Contains(offset)) {
               in_image = true;
               break;
             }
@@ -948,7 +950,7 @@ int StackVisitor::GetVRegOffsetFromQuickCode(const DexFile::CodeItem* code_item,
   }
 }
 
-void LockCountData::AddMonitorInternal(Thread* self, mirror::Object* obj) {
+void LockCountData::AddMonitor(Thread* self, mirror::Object* obj) {
   if (obj == nullptr) {
     return;
   }
@@ -965,7 +967,7 @@ void LockCountData::AddMonitorInternal(Thread* self, mirror::Object* obj) {
   monitors_->push_back(obj);
 }
 
-void LockCountData::RemoveMonitorInternal(Thread* self, const mirror::Object* obj) {
+void LockCountData::RemoveMonitorOrThrow(Thread* self, const mirror::Object* obj) {
   if (obj == nullptr) {
     return;
   }
@@ -998,7 +1000,7 @@ void MonitorExitHelper(Thread* self, mirror::Object* obj) NO_THREAD_SAFETY_ANALY
   obj->MonitorExit(self);
 }
 
-bool LockCountData::CheckAllMonitorsReleasedInternal(Thread* self) {
+bool LockCountData::CheckAllMonitorsReleasedOrThrow(Thread* self) {
   DCHECK(self != nullptr);
   if (monitors_ != nullptr) {
     if (!monitors_->empty()) {
diff --git a/runtime/stack.h b/runtime/stack.h
index 7301184a9..e77ab4647 100644
--- a/runtime/stack.h
+++ b/runtime/stack.h
@@ -80,39 +80,18 @@ class LockCountData {
  public:
   // Add the given object to the list of monitors, that is, objects that have been locked. This
   // will not throw (but be skipped if there is an exception pending on entry).
-  template <bool kLockCounting>
-  void AddMonitor(Thread* self, mirror::Object* obj) SHARED_REQUIRES(Locks::mutator_lock_) {
-    DCHECK(self != nullptr);
-    if (!kLockCounting) {
-      return;
-    }
-    AddMonitorInternal(self, obj);
-  }
+  void AddMonitor(Thread* self, mirror::Object* obj) SHARED_REQUIRES(Locks::mutator_lock_);
 
   // Try to remove the given object from the monitor list, indicating an unlock operation.
   // This will throw an IllegalMonitorStateException (clearing any already pending exception), in
   // case that there wasn't a lock recorded for the object.
-  template <bool kLockCounting>
   void RemoveMonitorOrThrow(Thread* self,
-                            const mirror::Object* obj) SHARED_REQUIRES(Locks::mutator_lock_) {
-    DCHECK(self != nullptr);
-    if (!kLockCounting) {
-      return;
-    }
-    RemoveMonitorInternal(self, obj);
-  }
+                            const mirror::Object* obj) SHARED_REQUIRES(Locks::mutator_lock_);
 
   // Check whether all acquired monitors have been released. This will potentially throw an
   // IllegalMonitorStateException, clearing any already pending exception. Returns true if the
   // check shows that everything is OK wrt/ lock counting, false otherwise.
-  template <bool kLockCounting>
-  bool CheckAllMonitorsReleasedOrThrow(Thread* self) SHARED_REQUIRES(Locks::mutator_lock_) {
-    DCHECK(self != nullptr);
-    if (!kLockCounting) {
-      return true;
-    }
-    return CheckAllMonitorsReleasedInternal(self);
-  }
+  bool CheckAllMonitorsReleasedOrThrow(Thread* self) SHARED_REQUIRES(Locks::mutator_lock_);
 
   template <typename T, typename... Args>
   void VisitMonitors(T visitor, Args&&... args) SHARED_REQUIRES(Locks::mutator_lock_) {
@@ -125,12 +104,6 @@ class LockCountData {
   }
 
  private:
-  // Internal implementations.
-  void AddMonitorInternal(Thread* self, mirror::Object* obj) SHARED_REQUIRES(Locks::mutator_lock_);
-  void RemoveMonitorInternal(Thread* self, const mirror::Object* obj)
-      SHARED_REQUIRES(Locks::mutator_lock_);
-  bool CheckAllMonitorsReleasedInternal(Thread* self) SHARED_REQUIRES(Locks::mutator_lock_);
-
   // Stores references to the locked-on objects. As noted, this should be visited during thread
   // marking.
   std::unique_ptr<std::vector<mirror::Object*>> monitors_;
diff --git a/runtime/thread.cc b/runtime/thread.cc
index 7922b6096..424894439 100644
--- a/runtime/thread.cc
+++ b/runtime/thread.cc
@@ -89,6 +89,7 @@ pthread_key_t Thread::pthread_key_self_;
 ConditionVariable* Thread::resume_cond_ = nullptr;
 const size_t Thread::kStackOverflowImplicitCheckSize = GetStackOverflowReservedBytes(kRuntimeISA);
 bool (*Thread::is_sensitive_thread_hook_)() = nullptr;
+Thread* Thread::jit_sensitive_thread_ = nullptr;
 
 static constexpr bool kVerifyImageObjectsMarked = kIsDebugBuild;
 
@@ -512,14 +513,20 @@ static size_t FixStackSize(size_t stack_size) {
   return stack_size;
 }
 
+// Return the nearest page-aligned address below the current stack top.
+NO_INLINE
+static uint8_t* FindStackTop() {
+  return reinterpret_cast<uint8_t*>(
+      AlignDown(__builtin_frame_address(0), kPageSize));
+}
+
 // Install a protected region in the stack.  This is used to trigger a SIGSEGV if a stack
 // overflow is detected.  It is located right below the stack_begin_.
 ATTRIBUTE_NO_SANITIZE_ADDRESS
 void Thread::InstallImplicitProtection() {
   uint8_t* pregion = tlsPtr_.stack_begin - kStackOverflowProtectedSize;
-  uint8_t* stack_himem = tlsPtr_.stack_end;
-  uint8_t* stack_top = reinterpret_cast<uint8_t*>(reinterpret_cast<uintptr_t>(&stack_himem) &
-      ~(kPageSize - 1));    // Page containing current top of stack.
+  // Page containing current top of stack.
+  uint8_t* stack_top = FindStackTop();
 
   // Try to directly protect the stack.
   VLOG(threads) << "installing stack protected region at " << std::hex <<
@@ -739,7 +746,7 @@ Thread* Thread::Attach(const char* thread_name, bool as_daemon, jobject thread_g
   {
     MutexLock mu(nullptr, *Locks::runtime_shutdown_lock_);
     if (runtime->IsShuttingDownLocked()) {
-      LOG(ERROR) << "Thread attaching while runtime is shutting down: " << thread_name;
+      LOG(WARNING) << "Thread attaching while runtime is shutting down: " << thread_name;
       return nullptr;
     } else {
       Runtime::Current()->StartThreadBirth();
@@ -931,8 +938,7 @@ bool Thread::InitStackHwm() {
   }
 
   // Sanity check.
-  int stack_variable;
-  CHECK_GT(&stack_variable, reinterpret_cast<void*>(tlsPtr_.stack_end));
+  CHECK_GT(FindStackTop(), reinterpret_cast<void*>(tlsPtr_.stack_end));
 
   return true;
 }
@@ -2763,7 +2769,7 @@ class ReferenceMapVisitor : public StackVisitor {
     VisitDeclaringClass(m);
 
     // Process register map (which native and runtime methods don't have)
-    if (!m->IsNative() && !m->IsRuntimeMethod() && !m->IsProxyMethod()) {
+    if (!m->IsNative() && !m->IsRuntimeMethod() && (!m->IsProxyMethod() || m->IsConstructor())) {
       const OatQuickMethodHeader* method_header = GetCurrentOatQuickMethodHeader();
       DCHECK(method_header->IsOptimized());
       auto* vreg_base = reinterpret_cast<StackReference<mirror::Object>*>(
@@ -3010,7 +3016,6 @@ size_t Thread::NumberOfHeldMutexes() const {
   return count;
 }
 
-
 void Thread::DeoptimizeWithDeoptimizationException(JValue* result) {
   DCHECK_EQ(GetException(), Thread::GetDeoptimizationException());
   ClearException();
@@ -3031,4 +3036,11 @@ void Thread::DeoptimizeWithDeoptimizationException(JValue* result) {
   interpreter::EnterInterpreterFromDeoptimize(this, shadow_frame, from_code, result);
 }
 
+void Thread::SetException(mirror::Throwable* new_exception) {
+  CHECK(new_exception != nullptr);
+  // TODO: DCHECK(!IsExceptionPending());
+  tlsPtr_.exception = new_exception;
+  // LOG(ERROR) << new_exception->Dump();
+}
+
 }  // namespace art
diff --git a/runtime/thread.h b/runtime/thread.h
index ed42e462a..582a0cdbd 100644
--- a/runtime/thread.h
+++ b/runtime/thread.h
@@ -363,12 +363,7 @@ class Thread {
   void AssertNoPendingException() const;
   void AssertNoPendingExceptionForNewException(const char* msg) const;
 
-  void SetException(mirror::Throwable* new_exception)
-      SHARED_REQUIRES(Locks::mutator_lock_) {
-    CHECK(new_exception != nullptr);
-    // TODO: DCHECK(!IsExceptionPending());
-    tlsPtr_.exception = new_exception;
-  }
+  void SetException(mirror::Throwable* new_exception) SHARED_REQUIRES(Locks::mutator_lock_);
 
   void ClearException() SHARED_REQUIRES(Locks::mutator_lock_) {
     tlsPtr_.exception = nullptr;
@@ -1098,6 +1093,12 @@ class Thread {
     return debug_disallow_read_barrier_;
   }
 
+  // Returns true if the current thread is the jit sensitive thread.
+  bool IsJitSensitiveThread() const {
+    return this == jit_sensitive_thread_;
+  }
+
+  // Returns true if StrictMode events are traced for the current thread.
   static bool IsSensitiveThread() {
     if (is_sensitive_thread_hook_ != nullptr) {
       return (*is_sensitive_thread_hook_)();
@@ -1180,6 +1181,16 @@ class Thread {
   ALWAYS_INLINE void PassActiveSuspendBarriers()
       REQUIRES(!Locks::thread_suspend_count_lock_, !Roles::uninterruptible_);
 
+  // Registers the current thread as the jit sensitive thread. Should be called just once.
+  static void SetJitSensitiveThread() {
+    if (jit_sensitive_thread_ == nullptr) {
+      jit_sensitive_thread_ = Thread::Current();
+    } else {
+      LOG(WARNING) << "Attempt to set the sensitive thread twice. Tid:"
+          << Thread::Current()->GetTid();
+    }
+  }
+
   static void SetSensitiveThreadHook(bool (*is_sensitive_thread_hook)()) {
     is_sensitive_thread_hook_ = is_sensitive_thread_hook;
   }
@@ -1229,6 +1240,8 @@ class Thread {
   // Hook passed by framework which returns true
   // when StrictMode events are traced for the current thread.
   static bool (*is_sensitive_thread_hook_)();
+  // Stores the jit sensitive thread (which for now is the UI thread).
+  static Thread* jit_sensitive_thread_;
 
   /***********************************************************************************************/
   // Thread local storage. Fields are grouped by size to enable 32 <-> 64 searching to account for
diff --git a/runtime/verifier/method_verifier.cc b/runtime/verifier/method_verifier.cc
index 8802e6243..2b96328f8 100644
--- a/runtime/verifier/method_verifier.cc
+++ b/runtime/verifier/method_verifier.cc
@@ -401,8 +401,13 @@ MethodVerifier::FailureData MethodVerifier::VerifyMethod(Thread* self,
         method->SetAccessFlags(method->GetAccessFlags() | kAccCompileDontBother);
       }
     }
-    if (method != nullptr && verifier.HasInstructionThatWillThrow()) {
-      method->SetAccessFlags(method->GetAccessFlags() | kAccCompileDontBother);
+    if (method != nullptr) {
+      if (verifier.HasInstructionThatWillThrow()) {
+        method->SetAccessFlags(method->GetAccessFlags() | kAccCompileDontBother);
+      }
+      if ((verifier.encountered_failure_types_ & VerifyError::VERIFY_ERROR_LOCKING) != 0) {
+        method->SetAccessFlags(method->GetAccessFlags() | kAccMustCountLocks);
+      }
     }
   } else {
     // Bad method data.
@@ -4096,8 +4101,8 @@ ArtMethod* MethodVerifier::VerifyInvocationArgs(
                                     << " to super " << PrettyMethod(res_method);
         return nullptr;
       }
-      mirror::Class* super_klass = super.GetClass();
-      if (res_method->GetMethodIndex() >= super_klass->GetVTableLength()) {
+      if (!reference_class->IsAssignableFrom(GetDeclaringClass().GetClass()) ||
+          (res_method->GetMethodIndex() >= super.GetClass()->GetVTableLength())) {
         Fail(VERIFY_ERROR_NO_METHOD) << "invalid invoke-super from "
                                     << PrettyMethod(dex_method_idx_, *dex_file_)
                                     << " to super " << super
diff --git a/test/004-JniTest/jni_test.cc b/test/004-JniTest/jni_test.cc
index 2bdf8d1e7..8619ff7f3 100644
--- a/test/004-JniTest/jni_test.cc
+++ b/test/004-JniTest/jni_test.cc
@@ -14,23 +14,22 @@
  * limitations under the License.
  */
 
-#include <assert.h>
 #include <iostream>
 #include <pthread.h>
 #include <stdio.h>
 #include <vector>
 
+#include "art_method-inl.h"
+#include "base/logging.h"
 #include "jni.h"
 
-#if defined(NDEBUG)
-#error test code compiled without NDEBUG
-#endif
+namespace art {
 
 static JavaVM* jvm = nullptr;
 
 extern "C" JNIEXPORT jint JNI_OnLoad(JavaVM *vm, void*) {
-  assert(vm != nullptr);
-  assert(jvm == nullptr);
+  CHECK(vm != nullptr);
+  CHECK(jvm == nullptr);
   jvm = vm;
   std::cout << "JNI_OnLoad called" << std::endl;
   return JNI_VERSION_1_6;
@@ -39,24 +38,24 @@ extern "C" JNIEXPORT jint JNI_OnLoad(JavaVM *vm, void*) {
 extern "C" JNIEXPORT void JNI_OnUnload(JavaVM*, void*) {
   // std::cout since LOG(INFO) adds extra stuff like pid.
   std::cout << "JNI_OnUnload called" << std::endl;
-  // Clear jvm for assert in test 004-JniTest.
+  // Clear jvm for CHECK in test 004-JniTest.
   jvm = nullptr;
 }
 
 static void* AttachHelper(void* arg) {
-  assert(jvm != nullptr);
+  CHECK(jvm != nullptr);
 
   JNIEnv* env = nullptr;
   JavaVMAttachArgs args = { JNI_VERSION_1_6, __FUNCTION__, nullptr };
   int attach_result = jvm->AttachCurrentThread(&env, &args);
-  assert(attach_result == 0);
+  CHECK_EQ(attach_result, 0);
 
   typedef void (*Fn)(JNIEnv*);
   Fn fn = reinterpret_cast<Fn>(arg);
   fn(env);
 
   int detach_result = jvm->DetachCurrentThread();
-  assert(detach_result == 0);
+  CHECK_EQ(detach_result, 0);
   return nullptr;
 }
 
@@ -64,19 +63,19 @@ static void PthreadHelper(void (*fn)(JNIEnv*)) {
   pthread_t pthread;
   int pthread_create_result = pthread_create(&pthread, nullptr, AttachHelper,
                                              reinterpret_cast<void*>(fn));
-  assert(pthread_create_result == 0);
+  CHECK_EQ(pthread_create_result, 0);
   int pthread_join_result = pthread_join(pthread, nullptr);
-  assert(pthread_join_result == 0);
+  CHECK_EQ(pthread_join_result, 0);
 }
 
 static void testFindClassOnAttachedNativeThread(JNIEnv* env) {
   jclass clazz = env->FindClass("Main");
-  assert(clazz != nullptr);
-  assert(!env->ExceptionCheck());
+  CHECK(clazz != nullptr);
+  CHECK(!env->ExceptionCheck());
 
   jobjectArray array = env->NewObjectArray(0, clazz, nullptr);
-  assert(array != nullptr);
-  assert(!env->ExceptionCheck());
+  CHECK(array != nullptr);
+  CHECK(!env->ExceptionCheck());
 }
 
 // http://b/10994325
@@ -86,12 +85,12 @@ extern "C" JNIEXPORT void JNICALL Java_Main_testFindClassOnAttachedNativeThread(
 
 static void testFindFieldOnAttachedNativeThread(JNIEnv* env) {
   jclass clazz = env->FindClass("Main");
-  assert(clazz != nullptr);
-  assert(!env->ExceptionCheck());
+  CHECK(clazz != nullptr);
+  CHECK(!env->ExceptionCheck());
 
   jfieldID field = env->GetStaticFieldID(clazz, "testFindFieldOnAttachedNativeThreadField", "Z");
-  assert(field != nullptr);
-  assert(!env->ExceptionCheck());
+  CHECK(field != nullptr);
+  CHECK(!env->ExceptionCheck());
 
   env->SetStaticBooleanField(clazz, field, JNI_TRUE);
 }
@@ -103,38 +102,38 @@ extern "C" JNIEXPORT void JNICALL Java_Main_testFindFieldOnAttachedNativeThreadN
 
 static void testReflectFieldGetFromAttachedNativeThread(JNIEnv* env) {
   jclass clazz = env->FindClass("Main");
-  assert(clazz != nullptr);
-  assert(!env->ExceptionCheck());
+  CHECK(clazz != nullptr);
+  CHECK(!env->ExceptionCheck());
 
   jclass class_clazz = env->FindClass("java/lang/Class");
-  assert(class_clazz != nullptr);
-  assert(!env->ExceptionCheck());
+  CHECK(class_clazz != nullptr);
+  CHECK(!env->ExceptionCheck());
 
   jmethodID getFieldMetodId = env->GetMethodID(class_clazz, "getField",
                                                "(Ljava/lang/String;)Ljava/lang/reflect/Field;");
-  assert(getFieldMetodId != nullptr);
-  assert(!env->ExceptionCheck());
+  CHECK(getFieldMetodId != nullptr);
+  CHECK(!env->ExceptionCheck());
 
   jstring field_name = env->NewStringUTF("testReflectFieldGetFromAttachedNativeThreadField");
-  assert(field_name != nullptr);
-  assert(!env->ExceptionCheck());
+  CHECK(field_name != nullptr);
+  CHECK(!env->ExceptionCheck());
 
   jobject field = env->CallObjectMethod(clazz, getFieldMetodId, field_name);
-  assert(field != nullptr);
-  assert(!env->ExceptionCheck());
+  CHECK(field != nullptr);
+  CHECK(!env->ExceptionCheck());
 
   jclass field_clazz = env->FindClass("java/lang/reflect/Field");
-  assert(field_clazz != nullptr);
-  assert(!env->ExceptionCheck());
+  CHECK(field_clazz != nullptr);
+  CHECK(!env->ExceptionCheck());
 
   jmethodID getBooleanMetodId = env->GetMethodID(field_clazz, "getBoolean",
                                                  "(Ljava/lang/Object;)Z");
-  assert(getBooleanMetodId != nullptr);
-  assert(!env->ExceptionCheck());
+  CHECK(getBooleanMetodId != nullptr);
+  CHECK(!env->ExceptionCheck());
 
   jboolean value = env->CallBooleanMethod(field, getBooleanMetodId, /* ignored */ clazz);
-  assert(value == false);
-  assert(!env->ExceptionCheck());
+  CHECK(value == false);
+  CHECK(!env->ExceptionCheck());
 }
 
 // http://b/15539150
@@ -148,22 +147,22 @@ extern "C" JNIEXPORT void JNICALL Java_Main_testReflectFieldGetFromAttachedNativ
 extern "C" JNIEXPORT void JNICALL Java_Main_testCallStaticVoidMethodOnSubClassNative(JNIEnv* env,
                                                                                      jclass) {
   jclass super_class = env->FindClass("Main$testCallStaticVoidMethodOnSubClass_SuperClass");
-  assert(super_class != nullptr);
+  CHECK(super_class != nullptr);
 
   jmethodID execute = env->GetStaticMethodID(super_class, "execute", "()V");
-  assert(execute != nullptr);
+  CHECK(execute != nullptr);
 
   jclass sub_class = env->FindClass("Main$testCallStaticVoidMethodOnSubClass_SubClass");
-  assert(sub_class != nullptr);
+  CHECK(sub_class != nullptr);
 
   env->CallStaticVoidMethod(sub_class, execute);
 }
 
 extern "C" JNIEXPORT jobject JNICALL Java_Main_testGetMirandaMethodNative(JNIEnv* env, jclass) {
   jclass abstract_class = env->FindClass("Main$testGetMirandaMethod_MirandaAbstract");
-  assert(abstract_class != nullptr);
+  CHECK(abstract_class != nullptr);
   jmethodID miranda_method = env->GetMethodID(abstract_class, "inInterface", "()Z");
-  assert(miranda_method != nullptr);
+  CHECK(miranda_method != nullptr);
   return env->ToReflectedMethod(abstract_class, miranda_method, JNI_FALSE);
 }
 
@@ -171,11 +170,11 @@ extern "C" JNIEXPORT jobject JNICALL Java_Main_testGetMirandaMethodNative(JNIEnv
 extern "C" void JNICALL Java_Main_testZeroLengthByteBuffers(JNIEnv* env, jclass) {
   std::vector<uint8_t> buffer(1);
   jobject byte_buffer = env->NewDirectByteBuffer(&buffer[0], 0);
-  assert(byte_buffer != nullptr);
-  assert(!env->ExceptionCheck());
+  CHECK(byte_buffer != nullptr);
+  CHECK(!env->ExceptionCheck());
 
-  assert(env->GetDirectBufferAddress(byte_buffer) == &buffer[0]);
-  assert(env->GetDirectBufferCapacity(byte_buffer) == 0);
+  CHECK_EQ(env->GetDirectBufferAddress(byte_buffer), &buffer[0]);
+  CHECK_EQ(env->GetDirectBufferCapacity(byte_buffer), 0);
 }
 
 constexpr size_t kByteReturnSize = 7;
@@ -185,18 +184,18 @@ extern "C" jbyte JNICALL Java_Main_byteMethod(JNIEnv*, jclass, jbyte b1, jbyte b
                                               jbyte b3, jbyte b4, jbyte b5, jbyte b6,
                                               jbyte b7, jbyte b8, jbyte b9, jbyte b10) {
   // We use b1 to drive the output.
-  assert(b2 == 2);
-  assert(b3 == -3);
-  assert(b4 == 4);
-  assert(b5 == -5);
-  assert(b6 == 6);
-  assert(b7 == -7);
-  assert(b8 == 8);
-  assert(b9 == -9);
-  assert(b10 == 10);
-
-  assert(0 <= b1);
-  assert(b1 < static_cast<jbyte>(kByteReturnSize));
+  CHECK_EQ(b2, 2);
+  CHECK_EQ(b3, -3);
+  CHECK_EQ(b4, 4);
+  CHECK_EQ(b5, -5);
+  CHECK_EQ(b6, 6);
+  CHECK_EQ(b7, -7);
+  CHECK_EQ(b8, 8);
+  CHECK_EQ(b9, -9);
+  CHECK_EQ(b10, 10);
+
+  CHECK_LE(0, b1);
+  CHECK_LT(b1, static_cast<jbyte>(kByteReturnSize));
 
   return byte_returns[b1];
 }
@@ -210,18 +209,18 @@ extern "C" jshort JNICALL Java_Main_shortMethod(JNIEnv*, jclass, jshort s1, jsho
                                                 jshort s3, jshort s4, jshort s5, jshort s6,
                                                 jshort s7, jshort s8, jshort s9, jshort s10) {
   // We use s1 to drive the output.
-  assert(s2 == 2);
-  assert(s3 == -3);
-  assert(s4 == 4);
-  assert(s5 == -5);
-  assert(s6 == 6);
-  assert(s7 == -7);
-  assert(s8 == 8);
-  assert(s9 == -9);
-  assert(s10 == 10);
-
-  assert(0 <= s1);
-  assert(s1 < static_cast<jshort>(kShortReturnSize));
+  CHECK_EQ(s2, 2);
+  CHECK_EQ(s3, -3);
+  CHECK_EQ(s4, 4);
+  CHECK_EQ(s5, -5);
+  CHECK_EQ(s6, 6);
+  CHECK_EQ(s7, -7);
+  CHECK_EQ(s8, 8);
+  CHECK_EQ(s9, -9);
+  CHECK_EQ(s10, 10);
+
+  CHECK_LE(0, s1);
+  CHECK_LT(s1, static_cast<jshort>(kShortReturnSize));
 
   return short_returns[s1];
 }
@@ -231,17 +230,17 @@ extern "C" jboolean JNICALL Java_Main_booleanMethod(JNIEnv*, jclass, jboolean b1
                                                     jboolean b5, jboolean b6, jboolean b7,
                                                     jboolean b8, jboolean b9, jboolean b10) {
   // We use b1 to drive the output.
-  assert(b2 == JNI_TRUE);
-  assert(b3 == JNI_FALSE);
-  assert(b4 == JNI_TRUE);
-  assert(b5 == JNI_FALSE);
-  assert(b6 == JNI_TRUE);
-  assert(b7 == JNI_FALSE);
-  assert(b8 == JNI_TRUE);
-  assert(b9 == JNI_FALSE);
-  assert(b10 == JNI_TRUE);
-
-  assert(b1 == JNI_TRUE || b1 == JNI_FALSE);
+  CHECK_EQ(b2, JNI_TRUE);
+  CHECK_EQ(b3, JNI_FALSE);
+  CHECK_EQ(b4, JNI_TRUE);
+  CHECK_EQ(b5, JNI_FALSE);
+  CHECK_EQ(b6, JNI_TRUE);
+  CHECK_EQ(b7, JNI_FALSE);
+  CHECK_EQ(b8, JNI_TRUE);
+  CHECK_EQ(b9, JNI_FALSE);
+  CHECK_EQ(b10, JNI_TRUE);
+
+  CHECK(b1 == JNI_TRUE || b1 == JNI_FALSE);
   return b1;
 }
 
@@ -252,17 +251,17 @@ extern "C" jchar JNICALL Java_Main_charMethod(JNIEnv*, jclass, jchar c1, jchar c
                                               jchar c3, jchar c4, jchar c5, jchar c6, jchar c7,
                                               jchar c8, jchar c9, jchar c10) {
   // We use c1 to drive the output.
-  assert(c2 == 'a');
-  assert(c3 == 'b');
-  assert(c4 == 'c');
-  assert(c5 == '0');
-  assert(c6 == '1');
-  assert(c7 == '2');
-  assert(c8 == 1234);
-  assert(c9 == 2345);
-  assert(c10 == 3456);
-
-  assert(c1 < static_cast<jchar>(kCharReturnSize));
+  CHECK_EQ(c2, 'a');
+  CHECK_EQ(c3, 'b');
+  CHECK_EQ(c4, 'c');
+  CHECK_EQ(c5, '0');
+  CHECK_EQ(c6, '1');
+  CHECK_EQ(c7, '2');
+  CHECK_EQ(c8, 1234);
+  CHECK_EQ(c9, 2345);
+  CHECK_EQ(c10, 3456);
+
+  CHECK_LT(c1, static_cast<jchar>(kCharReturnSize));
 
   return char_returns[c1];
 }
@@ -281,39 +280,39 @@ static void testShallowGetCallingClassLoader(JNIEnv* env) {
   // Test direct call.
   {
     jclass vmstack_clazz = env->FindClass("dalvik/system/VMStack");
-    assert(vmstack_clazz != nullptr);
-    assert(!env->ExceptionCheck());
+    CHECK(vmstack_clazz != nullptr);
+    CHECK(!env->ExceptionCheck());
 
     jmethodID getCallingClassLoaderMethodId = env->GetStaticMethodID(vmstack_clazz,
                                                                      "getCallingClassLoader",
                                                                      "()Ljava/lang/ClassLoader;");
-    assert(getCallingClassLoaderMethodId != nullptr);
-    assert(!env->ExceptionCheck());
+    CHECK(getCallingClassLoaderMethodId != nullptr);
+    CHECK(!env->ExceptionCheck());
 
     jobject class_loader = env->CallStaticObjectMethod(vmstack_clazz,
                                                        getCallingClassLoaderMethodId);
-    assert(class_loader == nullptr);
-    assert(!env->ExceptionCheck());
+    CHECK(class_loader == nullptr);
+    CHECK(!env->ExceptionCheck());
   }
 
   // Test one-level call. Use System.loadLibrary().
   {
     jclass system_clazz = env->FindClass("java/lang/System");
-    assert(system_clazz != nullptr);
-    assert(!env->ExceptionCheck());
+    CHECK(system_clazz != nullptr);
+    CHECK(!env->ExceptionCheck());
 
     jmethodID loadLibraryMethodId = env->GetStaticMethodID(system_clazz, "loadLibrary",
                                                            "(Ljava/lang/String;)V");
-    assert(loadLibraryMethodId != nullptr);
-    assert(!env->ExceptionCheck());
+    CHECK(loadLibraryMethodId != nullptr);
+    CHECK(!env->ExceptionCheck());
 
     // Create a string object.
     jobject library_string = env->NewStringUTF("non_existing_library");
-    assert(library_string != nullptr);
-    assert(!env->ExceptionCheck());
+    CHECK(library_string != nullptr);
+    CHECK(!env->ExceptionCheck());
 
     env->CallStaticVoidMethod(system_clazz, loadLibraryMethodId, library_string);
-    assert(env->ExceptionCheck());
+    CHECK(env->ExceptionCheck());
 
     // We expect UnsatisfiedLinkError.
     jthrowable thrown = env->ExceptionOccurred();
@@ -321,7 +320,7 @@ static void testShallowGetCallingClassLoader(JNIEnv* env) {
 
     jclass unsatisfied_link_error_clazz = env->FindClass("java/lang/UnsatisfiedLinkError");
     jclass thrown_class = env->GetObjectClass(thrown);
-    assert(env->IsSameObject(unsatisfied_link_error_clazz, thrown_class));
+    CHECK(env->IsSameObject(unsatisfied_link_error_clazz, thrown_class));
   }
 }
 
@@ -333,31 +332,31 @@ extern "C" JNIEXPORT void JNICALL Java_Main_nativeTestShallowGetCallingClassLoad
 
 static void testShallowGetStackClass2(JNIEnv* env) {
   jclass vmstack_clazz = env->FindClass("dalvik/system/VMStack");
-  assert(vmstack_clazz != nullptr);
-  assert(!env->ExceptionCheck());
+  CHECK(vmstack_clazz != nullptr);
+  CHECK(!env->ExceptionCheck());
 
   // Test direct call.
   {
     jmethodID getStackClass2MethodId = env->GetStaticMethodID(vmstack_clazz, "getStackClass2",
                                                               "()Ljava/lang/Class;");
-    assert(getStackClass2MethodId != nullptr);
-    assert(!env->ExceptionCheck());
+    CHECK(getStackClass2MethodId != nullptr);
+    CHECK(!env->ExceptionCheck());
 
     jobject caller_class = env->CallStaticObjectMethod(vmstack_clazz, getStackClass2MethodId);
-    assert(caller_class == nullptr);
-    assert(!env->ExceptionCheck());
+    CHECK(caller_class == nullptr);
+    CHECK(!env->ExceptionCheck());
   }
 
   // Test one-level call. Use VMStack.getStackClass1().
   {
     jmethodID getStackClass1MethodId = env->GetStaticMethodID(vmstack_clazz, "getStackClass1",
                                                               "()Ljava/lang/Class;");
-    assert(getStackClass1MethodId != nullptr);
-    assert(!env->ExceptionCheck());
+    CHECK(getStackClass1MethodId != nullptr);
+    CHECK(!env->ExceptionCheck());
 
     jobject caller_class = env->CallStaticObjectMethod(vmstack_clazz, getStackClass1MethodId);
-    assert(caller_class == nullptr);
-    assert(!env->ExceptionCheck());
+    CHECK(caller_class == nullptr);
+    CHECK(!env->ExceptionCheck());
   }
 
   // For better testing we would need to compile against libcore and have a two-deep stack
@@ -416,8 +415,8 @@ class JniCallNonvirtualVoidMethodTest {
       env_->ExceptionDescribe();
       env_->FatalError(__FUNCTION__);
     }
-    assert(!env_->ExceptionCheck());
-    assert(c != nullptr);
+    CHECK(!env_->ExceptionCheck());
+    CHECK(c != nullptr);
     return c;
   }
 
@@ -429,7 +428,7 @@ class JniCallNonvirtualVoidMethodTest {
       env_->ExceptionDescribe();
       env_->FatalError(__FUNCTION__);
     }
-    assert(m != nullptr);
+    CHECK(m != nullptr);
     return m;
   }
 
@@ -439,7 +438,7 @@ class JniCallNonvirtualVoidMethodTest {
       env_->ExceptionDescribe();
       env_->FatalError(__FUNCTION__);
     }
-    assert(o != nullptr);
+    CHECK(o != nullptr);
     return o;
   }
 
@@ -467,7 +466,7 @@ class JniCallNonvirtualVoidMethodTest {
       env_->ExceptionDescribe();
       env_->FatalError(__FUNCTION__);
     }
-    assert(m != nullptr);
+    CHECK(m != nullptr);
     return m;
   }
 
@@ -508,21 +507,21 @@ class JniCallNonvirtualVoidMethodTest {
     jobject sub_super = CallConstructor(sub_, super_constructor_);
     jobject sub_sub = CallConstructor(sub_, sub_constructor_);
 
-    assert(env_->IsInstanceOf(super_super, super_));
-    assert(!env_->IsInstanceOf(super_super, sub_));
+    CHECK(env_->IsInstanceOf(super_super, super_));
+    CHECK(!env_->IsInstanceOf(super_super, sub_));
 
     // Note that even though we called (and ran) the subclass
     // constructor, we are not the subclass.
-    assert(env_->IsInstanceOf(super_sub, super_));
-    assert(!env_->IsInstanceOf(super_sub, sub_));
+    CHECK(env_->IsInstanceOf(super_sub, super_));
+    CHECK(!env_->IsInstanceOf(super_sub, sub_));
 
     // Note that even though we called the superclass constructor, we
     // are still the subclass.
-    assert(env_->IsInstanceOf(sub_super, super_));
-    assert(env_->IsInstanceOf(sub_super, sub_));
+    CHECK(env_->IsInstanceOf(sub_super, super_));
+    CHECK(env_->IsInstanceOf(sub_super, sub_));
 
-    assert(env_->IsInstanceOf(sub_sub, super_));
-    assert(env_->IsInstanceOf(sub_sub, sub_));
+    CHECK(env_->IsInstanceOf(sub_sub, super_));
+    CHECK(env_->IsInstanceOf(sub_sub, sub_));
   }
 
   void TestnonstaticCallNonvirtualMethod(bool super_object, bool super_class, bool super_method, const char* test_case) {
@@ -542,8 +541,8 @@ class JniCallNonvirtualVoidMethodTest {
     CallMethod(o, c, m, true, test_case);
     jboolean super_field = GetBooleanField(o, super_field_);
     jboolean sub_field = GetBooleanField(o, sub_field_);
-    assert(super_field == super_method);
-    assert(sub_field != super_method);
+    CHECK_EQ(super_field, super_method);
+    CHECK_NE(sub_field, super_method);
   }
 
   void TestnonstaticCallNonvirtualMethod() {
@@ -565,20 +564,20 @@ extern "C" void JNICALL Java_Main_testCallNonvirtual(JNIEnv* env, jclass) {
 
 extern "C" JNIEXPORT void JNICALL Java_Main_testNewStringObject(JNIEnv* env, jclass) {
   jclass c = env->FindClass("java/lang/String");
-  assert(c != nullptr);
+  CHECK(c != nullptr);
 
   jmethodID mid1 = env->GetMethodID(c, "<init>", "()V");
-  assert(mid1 != nullptr);
-  assert(!env->ExceptionCheck());
+  CHECK(mid1 != nullptr);
+  CHECK(!env->ExceptionCheck());
   jmethodID mid2 = env->GetMethodID(c, "<init>", "([B)V");
-  assert(mid2 != nullptr);
-  assert(!env->ExceptionCheck());
+  CHECK(mid2 != nullptr);
+  CHECK(!env->ExceptionCheck());
   jmethodID mid3 = env->GetMethodID(c, "<init>", "([C)V");
-  assert(mid3 != nullptr);
-  assert(!env->ExceptionCheck());
+  CHECK(mid3 != nullptr);
+  CHECK(!env->ExceptionCheck());
   jmethodID mid4 = env->GetMethodID(c, "<init>", "(Ljava/lang/String;)V");
-  assert(mid4 != nullptr);
-  assert(!env->ExceptionCheck());
+  CHECK(mid4 != nullptr);
+  CHECK(!env->ExceptionCheck());
 
   const char* test_array = "Test";
   int byte_array_length = strlen(test_array);
@@ -587,22 +586,22 @@ extern "C" JNIEXPORT void JNICALL Java_Main_testNewStringObject(JNIEnv* env, jcl
 
   // Test NewObject
   jstring s = reinterpret_cast<jstring>(env->NewObject(c, mid2, byte_array));
-  assert(s != nullptr);
-  assert(env->GetStringLength(s) == byte_array_length);
-  assert(env->GetStringUTFLength(s) == byte_array_length);
+  CHECK(s != nullptr);
+  CHECK_EQ(env->GetStringLength(s), byte_array_length);
+  CHECK_EQ(env->GetStringUTFLength(s), byte_array_length);
   const char* chars = env->GetStringUTFChars(s, nullptr);
-  assert(strcmp(test_array, chars) == 0);
+  CHECK_EQ(strcmp(test_array, chars), 0);
   env->ReleaseStringUTFChars(s, chars);
 
   // Test AllocObject and Call(Nonvirtual)VoidMethod
   jstring s1 = reinterpret_cast<jstring>(env->AllocObject(c));
-  assert(s1 != nullptr);
+  CHECK(s1 != nullptr);
   jstring s2 = reinterpret_cast<jstring>(env->AllocObject(c));
-  assert(s2 != nullptr);
+  CHECK(s2 != nullptr);
   jstring s3 = reinterpret_cast<jstring>(env->AllocObject(c));
-  assert(s3 != nullptr);
+  CHECK(s3 != nullptr);
   jstring s4 = reinterpret_cast<jstring>(env->AllocObject(c));
-  assert(s4 != nullptr);
+  CHECK(s4 != nullptr);
 
   jcharArray char_array = env->NewCharArray(5);
   jstring string_arg = env->NewStringUTF("helloworld");
@@ -621,18 +620,18 @@ extern "C" JNIEXPORT void JNICALL Java_Main_testNewStringObject(JNIEnv* env, jcl
 
   // Test with global and weak global references
   jstring s5 = reinterpret_cast<jstring>(env->AllocObject(c));
-  assert(s5 != nullptr);
+  CHECK(s5 != nullptr);
   s5 = reinterpret_cast<jstring>(env->NewGlobalRef(s5));
   jstring s6 = reinterpret_cast<jstring>(env->AllocObject(c));
-  assert(s6 != nullptr);
+  CHECK(s6 != nullptr);
   s6 = reinterpret_cast<jstring>(env->NewWeakGlobalRef(s6));
 
   env->CallVoidMethod(s5, mid1);
   env->CallNonvirtualVoidMethod(s6, c, mid2, byte_array);
-  assert(env->GetStringLength(s5) == 0);
-  assert(env->GetStringLength(s6) == byte_array_length);
+  CHECK_EQ(env->GetStringLength(s5), 0);
+  CHECK_EQ(env->GetStringLength(s6), byte_array_length);
   const char* chars6 = env->GetStringUTFChars(s6, nullptr);
-  assert(strcmp(test_array, chars6) == 0);
+  CHECK_EQ(strcmp(test_array, chars6), 0);
   env->ReleaseStringUTFChars(s6, chars6);
 }
 
@@ -664,8 +663,8 @@ class JniCallDefaultMethodsTest {
  public:
   explicit JniCallDefaultMethodsTest(JNIEnv* env)
       : env_(env), concrete_class_(env_->FindClass("ConcreteClass")) {
-    assert(!env_->ExceptionCheck());
-    assert(concrete_class_ != nullptr);
+    CHECK(!env_->ExceptionCheck());
+    CHECK(concrete_class_ != nullptr);
   }
 
   void Test() {
@@ -688,14 +687,14 @@ class JniCallDefaultMethodsTest {
   void TestCalls(const char* declaring_class, std::vector<const char*> methods) {
     jmethodID new_method = env_->GetMethodID(concrete_class_, "<init>", "()V");
     jobject obj = env_->NewObject(concrete_class_, new_method);
-    assert(!env_->ExceptionCheck());
-    assert(obj != nullptr);
+    CHECK(!env_->ExceptionCheck());
+    CHECK(obj != nullptr);
     jclass decl_class = env_->FindClass(declaring_class);
-    assert(!env_->ExceptionCheck());
-    assert(decl_class != nullptr);
+    CHECK(!env_->ExceptionCheck());
+    CHECK(decl_class != nullptr);
     for (const char* method : methods) {
       jmethodID method_id = env_->GetMethodID(decl_class, method, "()V");
-      assert(!env_->ExceptionCheck());
+      CHECK(!env_->ExceptionCheck());
       printf("Calling method %s->%s on object of type ConcreteClass\n", declaring_class, method);
       env_->CallVoidMethod(obj, method_id);
       if (env_->ExceptionCheck()) {
@@ -704,10 +703,10 @@ class JniCallDefaultMethodsTest {
         jmethodID to_string = env_->GetMethodID(
             env_->FindClass("java/lang/Object"), "toString", "()Ljava/lang/String;");
         jstring exception_string = (jstring) env_->CallObjectMethod(thrown, to_string);
-        assert(!env_->ExceptionCheck());
+        CHECK(!env_->ExceptionCheck());
         const char* exception_string_utf8 = env_->GetStringUTFChars(exception_string, nullptr);
-        assert(!env_->ExceptionCheck());
-        assert(exception_string_utf8 != nullptr);
+        CHECK(!env_->ExceptionCheck());
+        CHECK(exception_string_utf8 != nullptr);
         printf("EXCEPTION OCCURED: %s\n", exception_string_utf8);
         env_->ReleaseStringUTFChars(exception_string, exception_string_utf8);
       }
@@ -724,12 +723,12 @@ extern "C" JNIEXPORT void JNICALL Java_Main_testCallDefaultMethods(JNIEnv* env)
 
 static void InvokeSpecificMethod(JNIEnv* env, jobject obj, const char* method) {
   jclass lambda_class = env->FindClass("LambdaInterface");
-  assert(!env->ExceptionCheck());
-  assert(lambda_class != nullptr);
+  CHECK(!env->ExceptionCheck());
+  CHECK(lambda_class != nullptr);
   jmethodID method_id = env->GetMethodID(lambda_class, method, "()V");
-  assert(!env->ExceptionCheck());
+  CHECK(!env->ExceptionCheck());
   env->CallVoidMethod(obj, method_id);
-  assert(!env->ExceptionCheck());
+  CHECK(!env->ExceptionCheck());
 }
 
 extern "C" JNIEXPORT void JNICALL Java_Main_testInvokeLambdaDefaultMethod(
@@ -740,3 +739,6 @@ extern "C" JNIEXPORT void JNICALL Java_Main_testInvokeLambdaDefaultMethod(
 extern "C" JNIEXPORT void JNICALL Java_Main_testInvokeLambdaMethod(JNIEnv* e, jclass, jobject l) {
   InvokeSpecificMethod(e, l, "sayHi");
 }
+
+}  // namespace art
+
diff --git a/test/004-ReferenceMap/stack_walk_refmap_jni.cc b/test/004-ReferenceMap/stack_walk_refmap_jni.cc
index 284e5544f..5304590ad 100644
--- a/test/004-ReferenceMap/stack_walk_refmap_jni.cc
+++ b/test/004-ReferenceMap/stack_walk_refmap_jni.cc
@@ -14,6 +14,7 @@
  * limitations under the License.
  */
 
+#include "art_method-inl.h"
 #include "check_reference_map_visitor.h"
 #include "jni.h"
 
diff --git a/test/004-StackWalk/stack_walk_jni.cc b/test/004-StackWalk/stack_walk_jni.cc
index 51bb68f24..420224dd2 100644
--- a/test/004-StackWalk/stack_walk_jni.cc
+++ b/test/004-StackWalk/stack_walk_jni.cc
@@ -14,6 +14,7 @@
  * limitations under the License.
  */
 
+#include "art_method-inl.h"
 #include "check_reference_map_visitor.h"
 #include "jni.h"
 
diff --git a/test/044-proxy/expected.txt b/test/044-proxy/expected.txt
index be7023e49..2a5f0b90d 100644
--- a/test/044-proxy/expected.txt
+++ b/test/044-proxy/expected.txt
@@ -95,3 +95,5 @@ Proxy narrowed invocation return type passed
 5.8
 JNI_OnLoad called
 callback
+Found constructor.
+Found constructors with 0 exceptions
diff --git a/test/044-proxy/src/ConstructorProxy.java b/test/044-proxy/src/ConstructorProxy.java
new file mode 100644
index 000000000..95d150cbb
--- /dev/null
+++ b/test/044-proxy/src/ConstructorProxy.java
@@ -0,0 +1,53 @@
+/*
+ * 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.
+ */
+
+import java.lang.reflect.Constructor;
+import java.lang.reflect.InvocationHandler;
+import java.lang.reflect.Method;
+import java.lang.reflect.Proxy;
+
+/**
+ * Tests proxies when used with constructor methods.
+ */
+class ConstructorProxy implements InvocationHandler {
+  public static void main() {
+    try {
+      new ConstructorProxy().runTest();
+    } catch (Exception e) {
+      System.out.println("Unexpected failure occured");
+      e.printStackTrace();
+    }
+  }
+
+  public void runTest() throws Exception {
+    Class<?> proxyClass = Proxy.getProxyClass(
+            getClass().getClassLoader(),
+            new Class<?>[] { Runnable.class }
+    );
+    Constructor<?> constructor = proxyClass.getConstructor(InvocationHandler.class);
+    System.out.println("Found constructor.");
+    // We used to crash when asking the exception types of the constructor, because the runtime was
+    // not using the non-proxy ArtMethod
+    Object[] exceptions = constructor.getExceptionTypes();
+    System.out.println("Found constructors with " + exceptions.length + " exceptions");
+  }
+
+  @Override
+  public Object invoke(Object proxy, Method method, Object[] args) throws Throwable {
+    return args[0];
+  }
+}
+
diff --git a/test/044-proxy/src/Main.java b/test/044-proxy/src/Main.java
index 1f23b95cf..9dadb7c6e 100644
--- a/test/044-proxy/src/Main.java
+++ b/test/044-proxy/src/Main.java
@@ -31,6 +31,7 @@ public class Main {
         NarrowingTest.main(null);
         FloatSelect.main(null);
         NativeProxy.main(args);
+        ConstructorProxy.main();
     }
 
     // The following code maps from the actual proxy class names (eg $Proxy2) to their test output
diff --git a/test/136-daemon-jni-shutdown/daemon_jni_shutdown.cc b/test/136-daemon-jni-shutdown/daemon_jni_shutdown.cc
index 54879fbad..c9110a905 100644
--- a/test/136-daemon-jni-shutdown/daemon_jni_shutdown.cc
+++ b/test/136-daemon-jni-shutdown/daemon_jni_shutdown.cc
@@ -14,6 +14,7 @@
  * limitations under the License.
  */
 
+#include <dlfcn.h>
 #include <iostream>
 
 #include "base/casts.h"
@@ -45,6 +46,10 @@ extern "C" JNIEXPORT void JNICALL Java_Main_destroyJavaVMAndExit(JNIEnv* env, jc
   self->SetTopOfShadowStack(nullptr);
   JavaVM* vm = down_cast<JNIEnvExt*>(env)->vm;
   vm->DetachCurrentThread();
+  // Open ourself again to make sure the native library does not get unloaded from
+  // underneath us due to DestroyJavaVM. b/28406866
+  void* handle = dlopen(kIsDebugBuild ? "libarttestd.so" : "libarttest.so", RTLD_NOW);
+  CHECK(handle != nullptr);
   vm->DestroyJavaVM();
   vm_was_shutdown.store(true);
   // Give threads some time to get stuck in ExceptionCheck.
diff --git a/test/138-duplicate-classes-check/src/FancyLoader.java b/test/138-duplicate-classes-check/src/FancyLoader.java
deleted file mode 100644
index 03ec94876..000000000
--- a/test/138-duplicate-classes-check/src/FancyLoader.java
+++ /dev/null
@@ -1,229 +0,0 @@
-/*
- * Copyright (C) 2008 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.io.File;
-import java.io.FileNotFoundException;
-import java.io.IOException;
-import java.io.RandomAccessFile;
-import java.lang.reflect.Constructor;
-import java.lang.reflect.Method;
-import java.lang.reflect.InvocationTargetException;
-
-/**
- * A class loader with atypical behavior: we try to load a private
- * class implementation before asking the system or boot loader.  This
- * is used to create multiple classes with identical names in a single VM.
- *
- * If DexFile is available, we use that; if not, we assume we're not in
- * Dalvik and instantiate the class with defineClass().
- *
- * The location of the DEX files and class data is dependent upon the
- * test framework.
- */
-public class FancyLoader extends ClassLoader {
-    /* this is where the "alternate" .class files live */
-    static final String CLASS_PATH = "classes-ex/";
-
-    /* this is the "alternate" DEX/Jar file */
-    static final String DEX_FILE = System.getenv("DEX_LOCATION") +
-            "/138-duplicate-classes-check-ex.jar";
-
-    /* on Dalvik, this is a DexFile; otherwise, it's null */
-    private Class mDexClass;
-
-    private Object mDexFile;
-
-    /**
-     * Construct FancyLoader, grabbing a reference to the DexFile class
-     * if we're running under Dalvik.
-     */
-    public FancyLoader(ClassLoader parent) {
-        super(parent);
-
-        try {
-            mDexClass = parent.loadClass("dalvik.system.DexFile");
-        } catch (ClassNotFoundException cnfe) {
-            // ignore -- not running Dalvik
-        }
-    }
-
-    /**
-     * Finds the class with the specified binary name.
-     *
-     * We search for a file in CLASS_PATH or pull an entry from DEX_FILE.
-     * If we don't find a match, we throw an exception.
-     */
-    protected Class<?> findClass(String name) throws ClassNotFoundException
-    {
-        if (mDexClass != null) {
-            return findClassDalvik(name);
-        } else {
-            return findClassNonDalvik(name);
-        }
-    }
-
-    /**
-     * Finds the class with the specified binary name, from a DEX file.
-     */
-    private Class<?> findClassDalvik(String name)
-        throws ClassNotFoundException {
-
-        if (mDexFile == null) {
-            synchronized (FancyLoader.class) {
-                Constructor ctor;
-                /*
-                 * Construct a DexFile object through reflection.
-                 */
-                try {
-                    ctor = mDexClass.getConstructor(new Class[] {String.class});
-                } catch (NoSuchMethodException nsme) {
-                    throw new ClassNotFoundException("getConstructor failed",
-                        nsme);
-                }
-
-                try {
-                    mDexFile = ctor.newInstance(DEX_FILE);
-                } catch (InstantiationException ie) {
-                    throw new ClassNotFoundException("newInstance failed", ie);
-                } catch (IllegalAccessException iae) {
-                    throw new ClassNotFoundException("newInstance failed", iae);
-                } catch (InvocationTargetException ite) {
-                    throw new ClassNotFoundException("newInstance failed", ite);
-                }
-            }
-        }
-
-        /*
-         * Call DexFile.loadClass(String, ClassLoader).
-         */
-        Method meth;
-
-        try {
-            meth = mDexClass.getMethod("loadClass",
-                    new Class[] { String.class, ClassLoader.class });
-        } catch (NoSuchMethodException nsme) {
-            throw new ClassNotFoundException("getMethod failed", nsme);
-        }
-
-        try {
-            meth.invoke(mDexFile, name, this);
-        } catch (IllegalAccessException iae) {
-            throw new ClassNotFoundException("loadClass failed", iae);
-        } catch (InvocationTargetException ite) {
-            throw new ClassNotFoundException("loadClass failed",
-                ite.getCause());
-        }
-
-        return null;
-    }
-
-    /**
-     * Finds the class with the specified binary name, from .class files.
-     */
-    private Class<?> findClassNonDalvik(String name)
-        throws ClassNotFoundException {
-
-        String pathName = CLASS_PATH + name + ".class";
-        //System.out.println("--- Fancy: looking for " + pathName);
-
-        File path = new File(pathName);
-        RandomAccessFile raf;
-
-        try {
-            raf = new RandomAccessFile(path, "r");
-        } catch (FileNotFoundException fnfe) {
-            throw new ClassNotFoundException("Not found: " + pathName);
-        }
-
-        /* read the entire file in */
-        byte[] fileData;
-        try {
-            fileData = new byte[(int) raf.length()];
-            raf.readFully(fileData);
-        } catch (IOException ioe) {
-            throw new ClassNotFoundException("Read error: " + pathName);
-        } finally {
-            try {
-                raf.close();
-            } catch (IOException ioe) {
-                // drop
-            }
-        }
-
-        /* create the class */
-        //System.out.println("--- Fancy: defining " + name);
-        try {
-            return defineClass(name, fileData, 0, fileData.length);
-        } catch (Throwable th) {
-            throw new ClassNotFoundException("defineClass failed", th);
-        }
-    }
-
-    /**
-     * Load a class.
-     *
-     * Normally a class loader wouldn't override this, but we want our
-     * version of the class to take precedence over an already-loaded
-     * version.
-     *
-     * We still want the system classes (e.g. java.lang.Object) from the
-     * bootstrap class loader.
-     */
-    protected Class<?> loadClass(String name, boolean resolve)
-        throws ClassNotFoundException
-    {
-        Class res;
-
-        /*
-         * 1. Invoke findLoadedClass(String) to check if the class has
-         * already been loaded.
-         *
-         * This doesn't change.
-         */
-        res = findLoadedClass(name);
-        if (res != null) {
-            System.out.println("FancyLoader.loadClass: "
-                + name + " already loaded");
-            if (resolve)
-                resolveClass(res);
-            return res;
-        }
-
-        /*
-         * 3. Invoke the findClass(String) method to find the class.
-         */
-        try {
-            res = findClass(name);
-            if (resolve)
-                resolveClass(res);
-        }
-        catch (ClassNotFoundException e) {
-            // we couldn't find it, so eat the exception and keep going
-        }
-
-        /*
-         * 2. Invoke the loadClass method on the parent class loader.  If
-         * the parent loader is null the class loader built-in to the
-         * virtual machine is used, instead.
-         *
-         * (Since we're not in java.lang, we can't actually invoke the
-         * parent's loadClass() method, but we passed our parent to the
-         * super-class which can take care of it for us.)
-         */
-        res = super.loadClass(name, resolve);   // returns class or throws
-        return res;
-    }
-}
diff --git a/test/138-duplicate-classes-check/src/Main.java b/test/138-duplicate-classes-check/src/Main.java
index a9b5bb04e..a2ef28193 100644
--- a/test/138-duplicate-classes-check/src/Main.java
+++ b/test/138-duplicate-classes-check/src/Main.java
@@ -14,6 +14,7 @@
  * limitations under the License.
  */
 
+import dalvik.system.DexClassLoader;
 import java.io.File;
 import java.lang.reflect.Method;
 
@@ -30,7 +31,11 @@ public class Main {
 
         // Now run the class from the -ex file.
 
-        FancyLoader loader = new FancyLoader(getClass().getClassLoader());
+        String dexPath = System.getenv("DEX_LOCATION") + "/138-duplicate-classes-check-ex.jar";
+        String optimizedDirectory = System.getenv("DEX_LOCATION");
+        String librarySearchPath = null;
+        DexClassLoader loader = new DexClassLoader(dexPath, optimizedDirectory, librarySearchPath,
+                getClass().getClassLoader());
 
         try {
             Class testEx = loader.loadClass("TestEx");
diff --git a/test/141-class-unload/src/Main.java b/test/141-class-unload/src/Main.java
index 15683b0b1..17a6049db 100644
--- a/test/141-class-unload/src/Main.java
+++ b/test/141-class-unload/src/Main.java
@@ -23,6 +23,7 @@ import java.lang.reflect.Method;
 
 public class Main {
     static final String DEX_FILE = System.getenv("DEX_LOCATION") + "/141-class-unload-ex.jar";
+    static final String LIBRARY_SEARCH_PATH = System.getProperty("java.library.path");
     static String nativeLibraryName;
 
     public static void main(String[] args) throws Exception {
@@ -32,7 +33,7 @@ public class Main {
             throw new AssertionError("Couldn't find path class loader class");
         }
         Constructor constructor =
-            pathClassLoader.getDeclaredConstructor(String.class, ClassLoader.class);
+            pathClassLoader.getDeclaredConstructor(String.class, String.class, ClassLoader.class);
         try {
             testUnloadClass(constructor);
             testUnloadLoader(constructor);
@@ -49,7 +50,7 @@ public class Main {
             // Test that the oat files are unloaded.
             testOatFilesUnloaded(getPid());
         } catch (Exception e) {
-            System.out.println(e);
+            e.printStackTrace();
         }
     }
 
@@ -118,7 +119,7 @@ public class Main {
     private static void testNoUnloadInvoke(Constructor constructor) throws Exception {
         WeakReference<ClassLoader> loader =
             new WeakReference((ClassLoader) constructor.newInstance(
-                DEX_FILE, ClassLoader.getSystemClassLoader()));
+                DEX_FILE, LIBRARY_SEARCH_PATH, ClassLoader.getSystemClassLoader()));
         WeakReference<Class> intHolder = new WeakReference(loader.get().loadClass("IntHolder"));
         intHolder.get().getDeclaredMethod("runGC").invoke(intHolder.get());
         boolean isNull = loader.get() == null;
@@ -128,7 +129,7 @@ public class Main {
     private static void testNoUnloadInstance(Constructor constructor) throws Exception {
         WeakReference<ClassLoader> loader =
             new WeakReference((ClassLoader) constructor.newInstance(
-                DEX_FILE, ClassLoader.getSystemClassLoader()));
+                DEX_FILE, LIBRARY_SEARCH_PATH, ClassLoader.getSystemClassLoader()));
         WeakReference<Class> intHolder = new WeakReference(loader.get().loadClass("IntHolder"));
         Object o = intHolder.get().newInstance();
         Runtime.getRuntime().gc();
@@ -138,7 +139,7 @@ public class Main {
 
     private static WeakReference<Class> setUpUnloadClass(Constructor constructor) throws Exception {
         ClassLoader loader = (ClassLoader) constructor.newInstance(
-            DEX_FILE, ClassLoader.getSystemClassLoader());
+            DEX_FILE, LIBRARY_SEARCH_PATH, ClassLoader.getSystemClassLoader());
         Class intHolder = loader.loadClass("IntHolder");
         Method getValue = intHolder.getDeclaredMethod("getValue");
         Method setValue = intHolder.getDeclaredMethod("setValue", Integer.TYPE);
@@ -155,7 +156,7 @@ public class Main {
                                                                 boolean waitForCompilation)
         throws Exception {
         ClassLoader loader = (ClassLoader) constructor.newInstance(
-            DEX_FILE, ClassLoader.getSystemClassLoader());
+            DEX_FILE, LIBRARY_SEARCH_PATH, ClassLoader.getSystemClassLoader());
         Class intHolder = loader.loadClass("IntHolder");
         Method setValue = intHolder.getDeclaredMethod("setValue", Integer.TYPE);
         setValue.invoke(intHolder, 2);
@@ -177,7 +178,7 @@ public class Main {
     private static WeakReference<ClassLoader> setUpLoadLibrary(Constructor constructor)
         throws Exception {
         ClassLoader loader = (ClassLoader) constructor.newInstance(
-            DEX_FILE, ClassLoader.getSystemClassLoader());
+            DEX_FILE, LIBRARY_SEARCH_PATH, ClassLoader.getSystemClassLoader());
         Class intHolder = loader.loadClass("IntHolder");
         Method loadLibrary = intHolder.getDeclaredMethod("loadLibrary", String.class);
         loadLibrary.invoke(intHolder, nativeLibraryName);
diff --git a/test/148-multithread-gc-annotations/check b/test/148-multithread-gc-annotations/check
new file mode 100755
index 000000000..842bdc6ae
--- /dev/null
+++ b/test/148-multithread-gc-annotations/check
@@ -0,0 +1,22 @@
+#!/bin/bash
+#
+# Copyright (C) 2015 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.
+
+# Check that the string "error" isn't present
+if grep error "$2"; then
+    exit 1
+else
+    exit 0
+fi
diff --git a/test/148-multithread-gc-annotations/expected.txt b/test/148-multithread-gc-annotations/expected.txt
new file mode 100644
index 000000000..e69de29bb
diff --git a/test/148-multithread-gc-annotations/gc_coverage.cc b/test/148-multithread-gc-annotations/gc_coverage.cc
new file mode 100644
index 000000000..263eefd3a
--- /dev/null
+++ b/test/148-multithread-gc-annotations/gc_coverage.cc
@@ -0,0 +1,42 @@
+/*
+ * Copyright (C) 2015 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 "runtime.h"
+#include "scoped_thread_state_change.h"
+#include "thread-inl.h"
+
+namespace art {
+namespace {
+
+extern "C" JNIEXPORT jboolean JNICALL Java_MovingGCThread_performHomogeneousSpaceCompact(JNIEnv*, jclass) {
+  return Runtime::Current()->GetHeap()->PerformHomogeneousSpaceCompact() == gc::kSuccess ?
+      JNI_TRUE : JNI_FALSE;
+}
+
+extern "C" JNIEXPORT jboolean JNICALL Java_MovingGCThread_supportHomogeneousSpaceCompact(JNIEnv*, jclass) {
+  return Runtime::Current()->GetHeap()->SupportHomogeneousSpaceCompactAndCollectorTransitions() ?
+      JNI_TRUE : JNI_FALSE;
+}
+
+extern "C" JNIEXPORT jlong JNICALL Java_MovingGCThread_objectAddress(JNIEnv* env, jclass, jobject object) {
+  ScopedObjectAccess soa(env);
+  return reinterpret_cast<jlong>(soa.Decode<mirror::Object*>(object));
+}
+
+}  // namespace
+}  // namespace art
diff --git a/test/148-multithread-gc-annotations/info.txt b/test/148-multithread-gc-annotations/info.txt
new file mode 100644
index 000000000..c62e544e0
--- /dev/null
+++ b/test/148-multithread-gc-annotations/info.txt
@@ -0,0 +1 @@
+Tests that getting annotations works during moving gc.
diff --git a/test/148-multithread-gc-annotations/src/AnnoClass1.java b/test/148-multithread-gc-annotations/src/AnnoClass1.java
new file mode 100644
index 000000000..b82c61fd5
--- /dev/null
+++ b/test/148-multithread-gc-annotations/src/AnnoClass1.java
@@ -0,0 +1,23 @@
+/*
+ * 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.lang.annotation.*;
+
+@Retention(RetentionPolicy.RUNTIME)
+@Target(ElementType.TYPE)
+public @interface AnnoClass1 {
+    Class value();
+}
diff --git a/test/148-multithread-gc-annotations/src/AnnoClass2.java b/test/148-multithread-gc-annotations/src/AnnoClass2.java
new file mode 100644
index 000000000..c75d950e2
--- /dev/null
+++ b/test/148-multithread-gc-annotations/src/AnnoClass2.java
@@ -0,0 +1,23 @@
+/*
+ * 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.lang.annotation.*;
+
+@Retention(RetentionPolicy.RUNTIME)
+@Target(ElementType.TYPE)
+public @interface AnnoClass2 {
+    Class value();
+}
diff --git a/test/148-multithread-gc-annotations/src/AnnoClass3.java b/test/148-multithread-gc-annotations/src/AnnoClass3.java
new file mode 100644
index 000000000..5b4a37809
--- /dev/null
+++ b/test/148-multithread-gc-annotations/src/AnnoClass3.java
@@ -0,0 +1,23 @@
+/*
+ * 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.lang.annotation.*;
+
+@Retention(RetentionPolicy.RUNTIME)
+@Target(ElementType.TYPE)
+public @interface AnnoClass3 {
+    Class value();
+}
diff --git a/test/148-multithread-gc-annotations/src/AnnotationThread.java b/test/148-multithread-gc-annotations/src/AnnotationThread.java
new file mode 100644
index 000000000..ebc14e96b
--- /dev/null
+++ b/test/148-multithread-gc-annotations/src/AnnotationThread.java
@@ -0,0 +1,32 @@
+/*
+ * 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.lang.annotation.*;
+
+@AnnoClass1(AnnoClass2.class)
+@AnnoClass2(AnnoClass3.class)
+@AnnoClass3(AnnoClass1.class)
+public class AnnotationThread implements Runnable {
+    public void run() {
+        for (int i = 0; i < 20; i++) {
+            Annotation[] annotations = AnnotationThread.class.getAnnotations();
+            if (annotations == null) {
+                System.out.println("error: AnnotationThread class has no annotations");
+                return;
+            }
+        }
+    }
+}
diff --git a/test/148-multithread-gc-annotations/src/Main.java b/test/148-multithread-gc-annotations/src/Main.java
new file mode 100644
index 000000000..b652ed651
--- /dev/null
+++ b/test/148-multithread-gc-annotations/src/Main.java
@@ -0,0 +1,32 @@
+/*
+ * 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.
+ */
+
+public class Main {
+    public static void main(String[] args) {
+        System.loadLibrary(args[0]);
+        Thread annoThread = new Thread(new AnnotationThread(), "Annotation thread");
+        Thread gcThread = new Thread(new MovingGCThread(), "Moving GC thread");
+        annoThread.start();
+        gcThread.start();
+        try {
+            annoThread.join();
+            gcThread.join();
+        } catch (InterruptedException e) {
+            System.out.println("error: " + e);
+        }
+        System.out.println("Done.");
+    }
+}
diff --git a/test/148-multithread-gc-annotations/src/MovingGCThread.java b/test/148-multithread-gc-annotations/src/MovingGCThread.java
new file mode 100644
index 000000000..87de9f4ec
--- /dev/null
+++ b/test/148-multithread-gc-annotations/src/MovingGCThread.java
@@ -0,0 +1,62 @@
+/*
+ * 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.TreeMap;
+
+public class MovingGCThread implements Runnable {
+    private static TreeMap treeMap = new TreeMap();
+
+    public void run() {
+        for (int i = 0; i < 20; i++) {
+            testHomogeneousCompaction();
+        }
+    }
+
+    public static void testHomogeneousCompaction() {
+        final boolean supportHSC = supportHomogeneousSpaceCompact();
+        if (!supportHSC) {
+            return;
+        }
+        Object o = new Object();
+        long addressBefore = objectAddress(o);
+        allocateStuff();
+        final boolean success = performHomogeneousSpaceCompact();
+        allocateStuff();
+        if (!success) {
+            System.out.println("error: Expected " + supportHSC + " but got " + success);
+        }
+        allocateStuff();
+        long addressAfter = objectAddress(o);
+        // This relies on the compaction copying from one space to another space and there being
+        // no overlap.
+        if (addressBefore == addressAfter) {
+            System.out.println("error: Expected different adddress " + addressBefore + " vs " +
+                    addressAfter);
+        }
+    }
+
+    private static void allocateStuff() {
+        for (int i = 0; i < 1000; ++i) {
+            Object o = new Object();
+            treeMap.put(o.hashCode(), o);
+        }
+    }
+
+    // Methods to get access to ART internals.
+    private static native boolean supportHomogeneousSpaceCompact();
+    private static native boolean performHomogeneousSpaceCompact();
+    private static native long objectAddress(Object object);
+}
diff --git a/test/449-checker-bce/src/Main.java b/test/449-checker-bce/src/Main.java
index 66e1d92cc..41771b52c 100644
--- a/test/449-checker-bce/src/Main.java
+++ b/test/449-checker-bce/src/Main.java
@@ -927,6 +927,32 @@ public class Main {
     }
   }
 
+  /// CHECK-START: void Main.nonzeroLength(int[]) BCE (before)
+  /// CHECK-DAG: BoundsCheck
+  //
+  /// CHECK-START: void Main.nonzeroLength(int[]) BCE (after)
+  /// CHECK-NOT: BoundsCheck
+  /// CHECK-NOT: Deoptimize
+  public static void nonzeroLength(int[] a) {
+    if (a.length != 0) {
+      a[0] = 112;
+    }
+  }
+
+  /// CHECK-START: void Main.knownLength(int[]) BCE (before)
+  /// CHECK-DAG: BoundsCheck
+  /// CHECK-DAG: BoundsCheck
+  //
+  /// CHECK-START: void Main.knownLength(int[]) BCE (after)
+  /// CHECK-NOT: BoundsCheck
+  /// CHECK-NOT: Deoptimize
+  public static void knownLength(int[] a) {
+    if (a.length == 2) {
+      a[0] = -1;
+      a[1] = -2;
+    }
+  }
+
   static int[][] mA;
 
   /// CHECK-START: void Main.dynamicBCEAndIntrinsic(int) BCE (before)
@@ -1586,6 +1612,26 @@ public class Main {
       }
     }
 
+    nonzeroLength(array);
+    if (array[0] != 112) {
+      System.out.println("nonzero length failed!");
+    }
+
+    knownLength(array);
+    if (array[0] != 112 || array[1] != 1) {
+      System.out.println("nonzero length failed!");
+    }
+    array = new int[2];
+    knownLength(array);
+    if (array[0] != -1 || array[1] != -2) {
+      System.out.println("nonzero length failed!");
+    }
+
+    // Zero length array does not break.
+    array = new int[0];
+    nonzeroLength(array);
+    knownLength(array);
+
     mA = new int[4][4];
     for (int i = 0; i < 4; i++) {
       for (int j = 0; j < 4; j++) {
diff --git a/test/530-checker-loops2/src/Main.java b/test/530-checker-loops2/src/Main.java
index c644692f0..b12fbd609 100644
--- a/test/530-checker-loops2/src/Main.java
+++ b/test/530-checker-loops2/src/Main.java
@@ -710,8 +710,8 @@ public class Main {
         // making them a candidate for deoptimization based on constant indices.
         // Compiler should ensure the array loads are not subsequently hoisted
         // "above" the deoptimization "barrier" on the bounds.
-        a[0][i] = 1;
-        a[1][i] = 2;
+        a[1][i] = 1;
+        a[2][i] = 2;
         a[99][i] = 3;
       }
     }
@@ -1042,11 +1042,11 @@ public class Main {
     a = new int[100][10];
     expectEquals(55, dynamicBCEAndConstantIndices(x, a, 0, 10));
     for (int i = 0; i < 10; i++) {
-      expectEquals((i % 10) != 0 ? 1 : 0, a[0][i]);
-      expectEquals((i % 10) != 0 ? 2 : 0, a[1][i]);
+      expectEquals((i % 10) != 0 ? 1 : 0, a[1][i]);
+      expectEquals((i % 10) != 0 ? 2 : 0, a[2][i]);
       expectEquals((i % 10) != 0 ? 3 : 0, a[99][i]);
     }
-    a = new int[2][10];
+    a = new int[3][10];
     sResult = 0;
     try {
       expectEquals(55, dynamicBCEAndConstantIndices(x, a, 0, 10));
@@ -1054,8 +1054,8 @@ public class Main {
       sResult = 1;
     }
     expectEquals(1, sResult);
-    expectEquals(a[0][1], 1);
-    expectEquals(a[1][1], 2);
+    expectEquals(a[1][1], 1);
+    expectEquals(a[2][1], 2);
 
     // Dynamic BCE combined with constant indices of all types.
     boolean[] x1 = { true };
diff --git a/test/536-checker-intrinsic-optimization/src/Main.java b/test/536-checker-intrinsic-optimization/src/Main.java
index be666e94f..15a9504ac 100644
--- a/test/536-checker-intrinsic-optimization/src/Main.java
+++ b/test/536-checker-intrinsic-optimization/src/Main.java
@@ -16,9 +16,69 @@
 
 
 public class Main {
+  public static boolean doThrow = false;
+
+  public static void assertIntEquals(int expected, int result) {
+    if (expected != result) {
+      throw new Error("Expected: " + expected + ", found: " + result);
+    }
+  }
+
+  public static void assertBooleanEquals(boolean expected, boolean result) {
+    if (expected != result) {
+      throw new Error("Expected: " + expected + ", found: " + result);
+    }
+  }
+
   public static void main(String[] args) {
     stringEqualsSame();
     stringArgumentNotNull("Foo");
+
+    assertIntEquals(0, $opt$noinline$getStringLength(""));
+    assertIntEquals(3, $opt$noinline$getStringLength("abc"));
+    assertIntEquals(10, $opt$noinline$getStringLength("0123456789"));
+
+    assertBooleanEquals(true, $opt$noinline$isStringEmpty(""));
+    assertBooleanEquals(false, $opt$noinline$isStringEmpty("abc"));
+    assertBooleanEquals(false, $opt$noinline$isStringEmpty("0123456789"));
+  }
+
+  /// CHECK-START: int Main.$opt$noinline$getStringLength(java.lang.String) instruction_simplifier (before)
+  /// CHECK-DAG:  <<Length:i\d+>>   InvokeVirtual intrinsic:StringLength
+  /// CHECK-DAG:                    Return [<<Length>>]
+
+  /// CHECK-START: int Main.$opt$noinline$getStringLength(java.lang.String) instruction_simplifier (after)
+  /// CHECK-DAG:  <<String:l\d+>>   ParameterValue
+  /// CHECK-DAG:  <<NullCk:l\d+>>   NullCheck [<<String>>]
+  /// CHECK-DAG:  <<Length:i\d+>>   ArrayLength [<<NullCk>>] is_string_length:true
+  /// CHECK-DAG:                    Return [<<Length>>]
+
+  /// CHECK-START: int Main.$opt$noinline$getStringLength(java.lang.String) instruction_simplifier (after)
+  /// CHECK-NOT:                    InvokeVirtual intrinsic:StringLength
+
+  static public int $opt$noinline$getStringLength(String s) {
+    if (doThrow) { throw new Error(); }
+    return s.length();
+  }
+
+  /// CHECK-START: boolean Main.$opt$noinline$isStringEmpty(java.lang.String) instruction_simplifier (before)
+  /// CHECK-DAG:  <<IsEmpty:z\d+>>  InvokeVirtual intrinsic:StringIsEmpty
+  /// CHECK-DAG:                    Return [<<IsEmpty>>]
+
+  /// CHECK-START: boolean Main.$opt$noinline$isStringEmpty(java.lang.String) instruction_simplifier (after)
+  /// CHECK-DAG:  <<String:l\d+>>   ParameterValue
+  /// CHECK-DAG:  <<Const0:i\d+>>   IntConstant 0
+  /// CHECK-DAG:  <<NullCk:l\d+>>   NullCheck [<<String>>]
+  /// CHECK-DAG:  <<Length:i\d+>>   ArrayLength [<<NullCk>>] is_string_length:true
+  /// CHECK-DAG:  <<IsEmpty:z\d+>>  Equal [<<Length>>,<<Const0>>]
+  /// CHECK-DAG:                    Return [<<IsEmpty>>]
+
+  /// CHECK-START: boolean Main.$opt$noinline$isStringEmpty(java.lang.String) instruction_simplifier (after)
+  /// CHECK-NOT:                    InvokeVirtual intrinsic:StringIsEmpty
+
+  static public boolean $opt$noinline$isStringEmpty(String s) {
+    if (doThrow) { throw new Error(); }
+    return s.isEmpty();
   }
 
   /// CHECK-START: boolean Main.stringEqualsSame() instruction_simplifier (before)
diff --git a/test/551-checker-shifter-operand/src/Main.java b/test/551-checker-shifter-operand/src/Main.java
index edb8a68b4..a4561b83d 100644
--- a/test/551-checker-shifter-operand/src/Main.java
+++ b/test/551-checker-shifter-operand/src/Main.java
@@ -500,9 +500,9 @@ public class Main {
     assertIntEquals(a + $noinline$IntShl(b, 16),  a + (b << 16));
     assertIntEquals(a + $noinline$IntShl(b, 30),  a + (b << 30));
     assertIntEquals(a + $noinline$IntShl(b, 31),  a + (b << 31));
-    assertIntEquals(a + $noinline$IntShl(b, 32),  a + (b << 32));
-    assertIntEquals(a + $noinline$IntShl(b, 62),  a + (b << 62));
-    assertIntEquals(a + $noinline$IntShl(b, 63),  a + (b << 63));
+    assertIntEquals(a + $noinline$IntShl(b, 32),  a + (b << $opt$inline$IntConstant32()));
+    assertIntEquals(a + $noinline$IntShl(b, 62),  a + (b << $opt$inline$IntConstant62()));
+    assertIntEquals(a + $noinline$IntShl(b, 63),  a + (b << $opt$inline$IntConstant63()));
 
     assertIntEquals(a - $noinline$IntShr(b, 1),   a - (b >>  1));
     assertIntEquals(a - $noinline$IntShr(b, 6),   a - (b >>  6));
@@ -513,9 +513,9 @@ public class Main {
     assertIntEquals(a - $noinline$IntShr(b, 16),  a - (b >> 16));
     assertIntEquals(a - $noinline$IntShr(b, 30),  a - (b >> 30));
     assertIntEquals(a - $noinline$IntShr(b, 31),  a - (b >> 31));
-    assertIntEquals(a - $noinline$IntShr(b, 32),  a - (b >> 32));
-    assertIntEquals(a - $noinline$IntShr(b, 62),  a - (b >> 62));
-    assertIntEquals(a - $noinline$IntShr(b, 63),  a - (b >> 63));
+    assertIntEquals(a - $noinline$IntShr(b, 32),  a - (b >> $opt$inline$IntConstant32()));
+    assertIntEquals(a - $noinline$IntShr(b, 62),  a - (b >> $opt$inline$IntConstant62()));
+    assertIntEquals(a - $noinline$IntShr(b, 63),  a - (b >> $opt$inline$IntConstant63()));
 
     assertIntEquals(a ^ $noinline$IntUshr(b, 1),   a ^ (b >>>  1));
     assertIntEquals(a ^ $noinline$IntUshr(b, 6),   a ^ (b >>>  6));
@@ -526,11 +526,17 @@ public class Main {
     assertIntEquals(a ^ $noinline$IntUshr(b, 16),  a ^ (b >>> 16));
     assertIntEquals(a ^ $noinline$IntUshr(b, 30),  a ^ (b >>> 30));
     assertIntEquals(a ^ $noinline$IntUshr(b, 31),  a ^ (b >>> 31));
-    assertIntEquals(a ^ $noinline$IntUshr(b, 32),  a ^ (b >>> 32));
-    assertIntEquals(a ^ $noinline$IntUshr(b, 62),  a ^ (b >>> 62));
-    assertIntEquals(a ^ $noinline$IntUshr(b, 63),  a ^ (b >>> 63));
+    assertIntEquals(a ^ $noinline$IntUshr(b, 32),  a ^ (b >>> $opt$inline$IntConstant32()));
+    assertIntEquals(a ^ $noinline$IntUshr(b, 62),  a ^ (b >>> $opt$inline$IntConstant62()));
+    assertIntEquals(a ^ $noinline$IntUshr(b, 63),  a ^ (b >>> $opt$inline$IntConstant63()));
   }
 
+  // Hiding constants outside the range [0, 32) used for int shifts from Jack.
+  // (Jack extracts only the low 5 bits.)
+  public static int $opt$inline$IntConstant32() { return 32; }
+  public static int $opt$inline$IntConstant62() { return 62; }
+  public static int $opt$inline$IntConstant63() { return 63; }
+
 
   static long $noinline$LongShl(long b, long c) {
     if (doThrow) throw new Error();
diff --git a/test/557-checker-instruction-simplifier-ror/src/Main.java b/test/557-checker-instruction-simplifier-ror/src/Main.java
index 310611bba..6d8b74d1e 100644
--- a/test/557-checker-instruction-simplifier-ror/src/Main.java
+++ b/test/557-checker-instruction-simplifier-ror/src/Main.java
@@ -175,28 +175,32 @@ public class Main {
 
   //  (i >>> #distance) | (i << #-distance)
 
-  /// CHECK-START: int Main.ror_int_constant_c_negc(int) instruction_simplifier (before)
+  /// CHECK-START: int Main.ror_int_constant_c_negc(int) instruction_simplifier_after_bce (before)
   /// CHECK:          <<ArgValue:i\d+>>     ParameterValue
   /// CHECK:          <<Const2:i\d+>>       IntConstant 2
-  /// CHECK:          <<ConstNeg2:i\d+>>    IntConstant {{-2|30}}
+  /// CHECK:          <<ConstNeg2:i\d+>>    IntConstant -2
   /// CHECK-DAG:      <<UShr:i\d+>>         UShr [<<ArgValue>>,<<Const2>>]
   /// CHECK-DAG:      <<Shl:i\d+>>          Shl [<<ArgValue>>,<<ConstNeg2>>]
   /// CHECK:          <<Or:i\d+>>           Or [<<UShr>>,<<Shl>>]
   /// CHECK:                                Return [<<Or>>]
 
-  /// CHECK-START: int Main.ror_int_constant_c_negc(int) instruction_simplifier (after)
+  /// CHECK-START: int Main.ror_int_constant_c_negc(int) instruction_simplifier_after_bce (after)
   /// CHECK:          <<ArgValue:i\d+>>     ParameterValue
   /// CHECK:          <<Const2:i\d+>>       IntConstant 2
   /// CHECK:          <<Ror:i\d+>>          Ror [<<ArgValue>>,<<Const2>>]
   /// CHECK:                                Return [<<Ror>>]
 
-  /// CHECK-START: int Main.ror_int_constant_c_negc(int) instruction_simplifier (after)
+  /// CHECK-START: int Main.ror_int_constant_c_negc(int) instruction_simplifier_after_bce (after)
   /// CHECK-NOT:      UShr
   /// CHECK-NOT:      Shl
   public static int ror_int_constant_c_negc(int value) {
-    return (value >>> 2) | (value << -2);
+    return (value >>> 2) | (value << $opt$inline$IntConstantM2());
   }
 
+  // Hiding constants outside the range [0, 32) used for int shifts from Jack.
+  // (Jack extracts only the low 5 bits.)
+  public static int $opt$inline$IntConstantM2() { return -2; }
+
   //  (j >>> #distance) | (j << #-distance)
 
   /// CHECK-START: long Main.ror_long_constant_c_negc(long) instruction_simplifier (before)
diff --git a/test/566-polymorphic-inlining/polymorphic_inline.cc b/test/566-polymorphic-inlining/polymorphic_inline.cc
index 167a5757f..7b2c6cbcd 100644
--- a/test/566-polymorphic-inlining/polymorphic_inline.cc
+++ b/test/566-polymorphic-inlining/polymorphic_inline.cc
@@ -60,6 +60,7 @@ extern "C" JNIEXPORT void JNICALL Java_Main_ensureJittedAndPolymorphicInline(JNI
 
   do_checks(cls, "testInvokeVirtual");
   do_checks(cls, "testInvokeInterface");
+  do_checks(cls, "$noinline$testInlineToSameTarget");
 }
 
 }  // namespace art
diff --git a/test/566-polymorphic-inlining/src/Main.java b/test/566-polymorphic-inlining/src/Main.java
index 7283e8622..a59ce5b34 100644
--- a/test/566-polymorphic-inlining/src/Main.java
+++ b/test/566-polymorphic-inlining/src/Main.java
@@ -25,6 +25,12 @@ public class Main implements Itf {
     }
   }
 
+  public static void assertEquals(int expected, int actual) {
+    if (expected != actual) {
+      throw new Error("Expected " + expected  + ", got " + actual);
+    }
+  }
+
   public static void main(String[] args) throws Exception {
     System.loadLibrary(args[0]);
     Main[] mains = new Main[3];
@@ -41,6 +47,8 @@ public class Main implements Itf {
       testInvokeVirtual(mains[1]);
       testInvokeInterface(itfs[0]);
       testInvokeInterface(itfs[1]);
+      $noinline$testInlineToSameTarget(mains[0]);
+      $noinline$testInlineToSameTarget(mains[1]);
     }
 
     ensureJittedAndPolymorphicInline();
@@ -56,6 +64,10 @@ public class Main implements Itf {
     // This will trigger a deoptimization of the compiled code.
     assertEquals(OtherSubclass.class, testInvokeVirtual(mains[2]));
     assertEquals(OtherSubclass.class, testInvokeInterface(itfs[2]));
+
+    // Run this once to make sure we execute the JITted code.
+    $noinline$testInlineToSameTarget(mains[0]);
+    assertEquals(20001, counter);
   }
 
   public Class sameInvokeVirtual() {
@@ -76,9 +88,21 @@ public class Main implements Itf {
     return m.sameInvokeVirtual();
   }
 
+  public static void $noinline$testInlineToSameTarget(Main m) {
+    if (doThrow) throw new Error("");
+    m.increment();
+  }
+
   public Object field = new Object();
 
   public static native void ensureJittedAndPolymorphicInline();
+
+  public void increment() {
+    field.getClass(); // null check to ensure we get an inlined frame in the CodeInfo
+    counter++;
+  }
+  public static int counter = 0;
+  public static boolean doThrow = false;
 }
 
 class Subclass extends Main {
diff --git a/test/570-checker-osr/osr.cc b/test/570-checker-osr/osr.cc
index 2a5b2c954..2fa5800e5 100644
--- a/test/570-checker-osr/osr.cc
+++ b/test/570-checker-osr/osr.cc
@@ -14,7 +14,7 @@
  * limitations under the License.
  */
 
-#include "art_method.h"
+#include "art_method-inl.h"
 #include "jit/jit.h"
 #include "jit/jit_code_cache.h"
 #include "jit/profiling_info.h"
@@ -75,7 +75,7 @@ extern "C" JNIEXPORT jboolean JNICALL Java_Main_isInOsrCode(JNIEnv* env,
 extern "C" JNIEXPORT jboolean JNICALL Java_Main_isInInterpreter(JNIEnv* env,
                                                                 jclass,
                                                                 jstring method_name) {
-  if (!Runtime::Current()->UseJit()) {
+  if (!Runtime::Current()->UseJitCompilation()) {
     // The return value is irrelevant if we're not using JIT.
     return false;
   }
@@ -111,7 +111,7 @@ class ProfilingInfoVisitor : public StackVisitor {
 extern "C" JNIEXPORT void JNICALL Java_Main_ensureHasProfilingInfo(JNIEnv* env,
                                                                    jclass,
                                                                    jstring method_name) {
-  if (!Runtime::Current()->UseJit()) {
+  if (!Runtime::Current()->UseJitCompilation()) {
     return;
   }
   ScopedUtfChars chars(env, method_name);
@@ -151,7 +151,7 @@ class OsrCheckVisitor : public StackVisitor {
 extern "C" JNIEXPORT void JNICALL Java_Main_ensureHasOsrCode(JNIEnv* env,
                                                              jclass,
                                                              jstring method_name) {
-  if (!Runtime::Current()->UseJit()) {
+  if (!Runtime::Current()->UseJitCompilation()) {
     return;
   }
   ScopedUtfChars chars(env, method_name);
diff --git a/test/570-checker-osr/src/Main.java b/test/570-checker-osr/src/Main.java
index 200b54a44..15c232d6a 100644
--- a/test/570-checker-osr/src/Main.java
+++ b/test/570-checker-osr/src/Main.java
@@ -61,6 +61,18 @@ public class Main {
       throw new Error("Unexpected return value");
     }
 
+    $noinline$inlineCache2(new Main(), /* isSecondInvocation */ false);
+    if ($noinline$inlineCache2(new SubMain(), /* isSecondInvocation */ true) != SubMain.class) {
+      throw new Error("Unexpected return value");
+    }
+
+    // Test polymorphic inline cache to the same target (inlineCache3).
+    $noinline$inlineCache3(new Main(), /* isSecondInvocation */ false);
+    $noinline$inlineCache3(new SubMain(), /* isSecondInvocation */ false);
+    if ($noinline$inlineCache3(new SubMain(), /* isSecondInvocation */ true) != null) {
+      throw new Error("Unexpected return value");
+    }
+
     $noinline$stackOverflow(new Main(), /* isSecondInvocation */ false);
     $noinline$stackOverflow(new SubMain(), /* isSecondInvocation */ true);
 
@@ -147,10 +159,76 @@ public class Main {
     return other.returnClass();
   }
 
+  public static Class $noinline$inlineCache2(Main m, boolean isSecondInvocation) {
+    // If we are running in non-JIT mode, or were unlucky enough to get this method
+    // already JITted, just return the expected value.
+    if (!isInInterpreter("$noinline$inlineCache2")) {
+      return SubMain.class;
+    }
+
+    ensureHasProfilingInfo("$noinline$inlineCache2");
+
+    // Ensure that we have OSR code to jump to.
+    if (isSecondInvocation) {
+      ensureHasOsrCode("$noinline$inlineCache2");
+    }
+
+    // This call will be optimized in the OSR compiled code
+    // to check and deoptimize if m is not of type 'Main'.
+    Main other = m.inlineCache2();
+
+    // Jump to OSR compiled code. The second run
+    // of this method will have 'm' as a SubMain, and the compiled
+    // code we are jumping to will have wrongly optimize other as being null.
+    if (isSecondInvocation) {
+      while (!isInOsrCode("$noinline$inlineCache2")) {}
+    }
+
+    // We used to wrongly optimize this code and assume 'other' was always null.
+    return (other == null) ? null : other.returnClass();
+  }
+
+  public static Class $noinline$inlineCache3(Main m, boolean isSecondInvocation) {
+    // If we are running in non-JIT mode, or were unlucky enough to get this method
+    // already JITted, just return the expected value.
+    if (!isInInterpreter("$noinline$inlineCache3")) {
+      return null;
+    }
+
+    ensureHasProfilingInfo("$noinline$inlineCache3");
+
+    // Ensure that we have OSR code to jump to.
+    if (isSecondInvocation) {
+      ensureHasOsrCode("$noinline$inlineCache3");
+    }
+
+    // This call will be optimized in the OSR compiled code
+    // to check and deoptimize if m is not of type 'Main'.
+    Main other = m.inlineCache3();
+
+    // Jump to OSR compiled code. The second run
+    // of this method will have 'm' as a SubMain, and the compiled
+    // code we are jumping to will have wrongly optimize other as being null.
+    if (isSecondInvocation) {
+      while (!isInOsrCode("$noinline$inlineCache3")) {}
+    }
+
+    // We used to wrongly optimize this code and assume 'other' was always null.
+    return (other == null) ? null : other.returnClass();
+  }
+
   public Main inlineCache() {
     return new Main();
   }
 
+  public Main inlineCache2() {
+    return null;
+  }
+
+  public Main inlineCache3() {
+    return null;
+  }
+
   public Class returnClass() {
     return Main.class;
   }
@@ -235,6 +313,10 @@ class SubMain extends Main {
     return new SubMain();
   }
 
+  public Main inlineCache2() {
+    return new SubMain();
+  }
+
   public void otherInlineCache() {
     return;
   }
diff --git a/test/586-checker-null-array-get/src/Main.java b/test/586-checker-null-array-get/src/Main.java
index 332cfb0a5..e0782bc84 100644
--- a/test/586-checker-null-array-get/src/Main.java
+++ b/test/586-checker-null-array-get/src/Main.java
@@ -14,10 +14,20 @@
  * limitations under the License.
  */
 
+class Test1 {
+  int[] iarr;
+}
+
+class Test2 {
+  float[] farr;
+}
+
 public class Main {
   public static Object[] getObjectArray() { return null; }
   public static long[] getLongArray() { return null; }
   public static Object getNull() { return null; }
+  public static Test1 getNullTest1() { return null; }
+  public static Test2 getNullTest2() { return null; }
 
   public static void main(String[] args) {
     try {
@@ -26,13 +36,25 @@ public class Main {
     } catch (NullPointerException e) {
       // Expected.
     }
+    try {
+      bar();
+      throw new Error("Expected NullPointerException");
+    } catch (NullPointerException e) {
+      // Expected.
+    }
+    try {
+      test1();
+      throw new Error("Expected NullPointerException");
+    } catch (NullPointerException e) {
+      // Expected.
+    }
   }
 
   /// CHECK-START: void Main.foo() load_store_elimination (after)
-  /// CHECK-DAG: <<Null:l\d+>>  NullConstant
-  /// CHECK-DAG: <<Check:l\d+>> NullCheck [<<Null>>]
-  /// CHECK-DAG: <<Get1:j\d+>>  ArrayGet [<<Check>>,{{i\d+}}]
-  /// CHECK-DAG: <<Get2:l\d+>>  ArrayGet [<<Check>>,{{i\d+}}]
+  /// CHECK-DAG: <<Null:l\d+>>   NullConstant
+  /// CHECK-DAG: <<Check:l\d+>>  NullCheck [<<Null>>]
+  /// CHECK-DAG: <<Get1:j\d+>>   ArrayGet [<<Check>>,{{i\d+}}]
+  /// CHECK-DAG: <<Get2:l\d+>>   ArrayGet [<<Check>>,{{i\d+}}]
   public static void foo() {
     longField = getLongArray()[0];
     objectField = getObjectArray()[0];
@@ -56,7 +78,7 @@ public class Main {
     // elimination pass to add a HDeoptimize. Not having the bounds check helped
     // the load store elimination think it could merge two ArrayGet with different
     // types.
-    String[] array = ((String[])getNull());
+    String[] array = (String[])getNull();
     objectField = array[0];
     objectField = array[1];
     objectField = array[2];
@@ -68,6 +90,23 @@ public class Main {
     longField = longArray[3];
   }
 
+  /// CHECK-START: float Main.test1() load_store_elimination (after)
+  /// CHECK-DAG: <<Null:l\d+>>       NullConstant
+  /// CHECK-DAG: <<Check1:l\d+>>     NullCheck [<<Null>>]
+  /// CHECK-DAG: <<FieldGet1:l\d+>>  InstanceFieldGet [<<Check1>>] field_name:Test1.iarr
+  /// CHECK-DAG: <<Check2:l\d+>>     NullCheck [<<FieldGet1>>]
+  /// CHECK-DAG: <<ArrayGet1:i\d+>>  ArrayGet [<<Check2>>,{{i\d+}}]
+  /// CHECK-DAG: <<ArrayGet2:f\d+>>  ArrayGet [<<Check2>>,{{i\d+}}]
+  /// CHECK-DAG:                     Return [<<ArrayGet2>>]
+  public static float test1() {
+    Test1 test1 = getNullTest1();
+    Test2 test2 = getNullTest2();;
+    int[] iarr = test1.iarr;
+    float[] farr = test2.farr;
+    iarr[0] = iarr[1];
+    return farr[0];
+  }
+
   public static long longField;
   public static Object objectField;
 }
diff --git a/test/594-checker-irreducible-linorder/smali/IrreducibleLoop.smali b/test/594-checker-irreducible-linorder/smali/IrreducibleLoop.smali
index 366c7b9b6..ef53ee867 100644
--- a/test/594-checker-irreducible-linorder/smali/IrreducibleLoop.smali
+++ b/test/594-checker-irreducible-linorder/smali/IrreducibleLoop.smali
@@ -67,3 +67,57 @@
    return p3
 
 .end method
+
+## CHECK-START: int IrreducibleLoop.liveness2(boolean, boolean, boolean, int) builder (after)
+## CHECK-DAG:     Mul loop:<<Loop:B\d+>>
+## CHECK-DAG:     Not loop:<<Loop>>
+
+## CHECK-START: int IrreducibleLoop.liveness2(boolean, boolean, boolean, int) liveness (after)
+## CHECK-DAG:     Mul liveness:<<LPreEntry2:\d+>>
+## CHECK-DAG:     Not liveness:<<LBackEdge1:\d+>>
+## CHECK-EVAL:    <<LBackEdge1>> < <<LPreEntry2>>
+
+.method public liveness2(ZZZI)I
+    .registers 10
+
+    const v1, 1
+
+    :header1
+    if-eqz p0, :body1
+
+    :exit
+    return p3
+
+    :body1
+    # The test will generate an incorrect linear order when the following IF swaps
+    # its successors. To do that, load a boolean value and compare NotEqual to 1.
+    sget-boolean v2, LIrreducibleLoop;->f:Z
+    const v3, 1
+    if-ne v2, v3, :pre_header2
+
+    :pre_entry2
+    # This constant has a use in a phi in :back_edge2 and a back edge use in
+    # :back_edge1. Because the linear order is wrong, the back edge use has
+    # a lower liveness than the phi use.
+    const v0, 42
+    mul-int/2addr p3, p3
+    goto :back_edge2
+
+    :back_edge2
+    add-int/2addr p3, v0
+    add-int/2addr v0, v1
+    goto :header2
+
+    :header2
+    if-eqz p2, :back_edge2
+
+    :back_edge1
+    not-int p3, p3
+    goto :header1
+
+    :pre_header2
+    const v0, 42
+    goto :header2
+.end method
+
+.field public static f:Z
diff --git a/test/594-invoke-super/expected.txt b/test/594-invoke-super/expected.txt
new file mode 100644
index 000000000..de26026c9
--- /dev/null
+++ b/test/594-invoke-super/expected.txt
@@ -0,0 +1,7 @@
+new A
+I am A's foo
+new B
+I am B's foo
+new A
+new B
+passed
diff --git a/test/594-invoke-super/info.txt b/test/594-invoke-super/info.txt
new file mode 100644
index 000000000..440d8b8c6
--- /dev/null
+++ b/test/594-invoke-super/info.txt
@@ -0,0 +1 @@
+Invoke-super on various references.
diff --git a/test/594-invoke-super/smali/invoke-super.smali b/test/594-invoke-super/smali/invoke-super.smali
new file mode 100644
index 000000000..6f787dd17
--- /dev/null
+++ b/test/594-invoke-super/smali/invoke-super.smali
@@ -0,0 +1,31 @@
+#
+# 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.
+
+.class public LZ;
+.super LA;
+
+.method public constructor <init>()V
+.registers 1
+    invoke-direct {v0}, LA;-><init>()V
+    return-void
+.end method
+
+.method public foo()V
+.registers 3
+    new-instance v0, LY;
+    invoke-direct {v0}, LY;-><init>()V
+    invoke-super {v0}, LY;->foo()V
+    return-void
+.end method
diff --git a/test/594-invoke-super/src/Main.java b/test/594-invoke-super/src/Main.java
new file mode 100644
index 000000000..53f2bbf67
--- /dev/null
+++ b/test/594-invoke-super/src/Main.java
@@ -0,0 +1,80 @@
+/*
+ * 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.lang.reflect.InvocationTargetException;
+import java.lang.reflect.Method;
+
+//
+// Two classes A and B with method foo().
+//
+
+class A {
+  A() { System.out.println("new A"); }
+
+  public void foo() { System.out.println("I am A's foo"); }
+
+  // We previously used to invoke this method with a Y instance, due
+  // to invoke-super underspecified behavior.
+  public void bar() { System.out.println("I am A's bar"); }
+}
+
+class B {
+  B() { System.out.println("new B"); }
+
+  public void foo() { System.out.println("I am B's foo"); }
+}
+
+//
+// Two subclasses X and Y that call foo() on super.
+//
+
+class X extends A {
+  public void foo() { super.foo(); }
+}
+
+class Y extends B {
+  public void foo() { super.foo(); }
+}
+
+//
+// Driver class.
+//
+
+public class Main {
+
+  public static void main(String[] args) throws Exception {
+    // The normal stuff, X's super goes to A, Y's super goes to B.
+    new X().foo();
+    new Y().foo();
+
+    // And now it gets interesting.
+
+    // In bytecode, we define a class Z that is a subclass of A, and we call
+    // invoke-super on an instance of Y.
+    Class<?> z = Class.forName("Z");
+    Method m = z.getMethod("foo");
+    try {
+      m.invoke(z.newInstance());
+      throw new Error("Expected InvocationTargetException");
+    } catch (InvocationTargetException e) {
+      if (!(e.getCause() instanceof NoSuchMethodError)) {
+        throw new Error("Expected NoSuchMethodError");
+      }
+    }
+
+    System.out.println("passed");
+  }
+}
diff --git a/test/594-load-string-regression/expected.txt b/test/594-load-string-regression/expected.txt
new file mode 100644
index 000000000..365b0e168
--- /dev/null
+++ b/test/594-load-string-regression/expected.txt
@@ -0,0 +1 @@
+String: ""
diff --git a/test/594-load-string-regression/info.txt b/test/594-load-string-regression/info.txt
new file mode 100644
index 000000000..6a07ace81
--- /dev/null
+++ b/test/594-load-string-regression/info.txt
@@ -0,0 +1,2 @@
+Regression test for LoadString listing side effects when it doesn't have any
+and triggering a DCHECK() failure when merging ClinitCheck into NewInstance.
diff --git a/test/594-load-string-regression/src/Main.java b/test/594-load-string-regression/src/Main.java
new file mode 100644
index 000000000..0b9f7b52a
--- /dev/null
+++ b/test/594-load-string-regression/src/Main.java
@@ -0,0 +1,77 @@
+/*
+ * 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.
+ */
+
+public class Main {
+  static boolean doThrow = false;
+
+  // Note: We're not doing checker tests as we cannot do them specifically for a non-PIC
+  // configuration. The check here would be "prepare_for_register_allocation (before)"
+  //     CHECK:         LoadClass
+  //     CHECK-NEXT:    ClinitCheck
+  //     CHECK-NEXT:    LoadString load_kind:BootImageAddress
+  //     CHECK-NEXT:    NewInstance
+  // and "prepare_for_register_allocation (after)"
+  //     CHECK:         LoadString
+  //     CHECK-NEXT:    NewInstance
+  // but the order of instructions for non-PIC mode is different.
+  public static int $noinline$test() {
+    if (doThrow) { throw new Error(); }
+
+    int r = 0x12345678;
+    do {
+      // LICM pulls the LoadClass and ClinitCheck out of the loop, leaves NewInstance in the loop.
+      Helper h = new Helper();
+      // For non-PIC mode, LICM pulls the boot image LoadString out of the loop.
+      // (For PIC mode, the LoadString can throw and will not be moved out of the loop.)
+      String s = "";  // Empty string is known to be in the boot image.
+      r = r ^ (r >> 5);
+      h.$noinline$printString(s);
+      // During DCE after inlining, the loop back-edge disappears and the pre-header is
+      // merged with the body, leaving consecutive LoadClass, ClinitCheck, LoadString
+      // and NewInstance in non-PIC mode. The prepare_for_register_allocation pass
+      // merges the LoadClass and ClinitCheck with the NewInstance and checks that
+      // there are no instructions with side effects in between. This check used to
+      // fail because LoadString was always listing SideEffects::CanTriggerGC() even
+      // when it doesn't really have any side effects, i.e. for direct references to
+      // boot image Strings or for Strings known to be in the dex cache.
+    } while ($inline$shouldContinue());
+    return r;
+  }
+
+  static boolean $inline$shouldContinue() {
+    return false;
+  }
+
+  public static void main(String[] args) {
+    assertIntEquals(0x12345678 ^ (0x12345678 >> 5), $noinline$test());
+  }
+
+  public static void assertIntEquals(int expected, int result) {
+    if (expected != result) {
+      throw new Error("Expected: " + expected + ", found: " + result);
+    }
+  }
+}
+
+class Helper {
+  static boolean doThrow = false;
+
+  public void $noinline$printString(String s) {
+    if (doThrow) { throw new Error(); }
+
+    System.out.println("String: \"" + s + "\"");
+  }
+}
diff --git a/test/595-error-class/expected.txt b/test/595-error-class/expected.txt
new file mode 100644
index 000000000..b0aad4deb
--- /dev/null
+++ b/test/595-error-class/expected.txt
@@ -0,0 +1 @@
+passed
diff --git a/test/595-error-class/info.txt b/test/595-error-class/info.txt
new file mode 100644
index 000000000..a58b8b31b
--- /dev/null
+++ b/test/595-error-class/info.txt
@@ -0,0 +1 @@
+Regression test on merging array type with error component type.
diff --git a/test/595-error-class/smali/error.smali b/test/595-error-class/smali/error.smali
new file mode 100644
index 000000000..925c34b29
--- /dev/null
+++ b/test/595-error-class/smali/error.smali
@@ -0,0 +1,23 @@
+# 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.
+
+.class public final LAnError;
+
+.super LSuperOfAnError;
+
+# Override a final method to put this class in the error state.
+.method public foo()V
+  .registers 1
+  return-void
+.end method
diff --git a/test/595-error-class/smali/merge.smali b/test/595-error-class/smali/merge.smali
new file mode 100644
index 000000000..2f8b41504
--- /dev/null
+++ b/test/595-error-class/smali/merge.smali
@@ -0,0 +1,31 @@
+#
+# 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.
+
+.class public LMerge;
+.super Ljava/lang/Object;
+
+# Method that selects between x = new Integer[] or new AnError[],
+# Reference type propagation should correctly see error in component type.
+.method public static select(Z)Ljava/lang/Object;
+    .registers 2
+    const/16 v0, 10
+    if-eqz v1, :Skip
+    new-array v0, v0, [LAnError;
+    goto :Done
+:Skip
+    new-array v0, v0, [Ljava/lang/Integer;
+:Done
+    return-object v0
+.end method
diff --git a/test/595-error-class/smali/super.smali b/test/595-error-class/smali/super.smali
new file mode 100644
index 000000000..da7467d16
--- /dev/null
+++ b/test/595-error-class/smali/super.smali
@@ -0,0 +1,22 @@
+# 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.
+
+.class public LSuperOfAnError;
+
+.super Ljava/lang/Object;
+
+.method public final foo()V
+  .registers 1
+  return-void
+.end method
diff --git a/test/595-error-class/src/Main.java b/test/595-error-class/src/Main.java
new file mode 100644
index 000000000..655fa4336
--- /dev/null
+++ b/test/595-error-class/src/Main.java
@@ -0,0 +1,30 @@
+/*
+ * 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.lang.reflect.*;
+
+public class Main {
+
+  public static void main(String args[]) throws Throwable {
+    Class<?> c = Class.forName("Merge");
+    Method m = c.getMethod("select", boolean.class);
+    Object x = m.invoke(null, true);
+    if (x == null) {
+      throw new Error("Did not get array");
+    }
+    System.out.println("passed");
+  }
+}
diff --git a/test/595-profile-saving/expected.txt b/test/595-profile-saving/expected.txt
new file mode 100644
index 000000000..6a5618ebc
--- /dev/null
+++ b/test/595-profile-saving/expected.txt
@@ -0,0 +1 @@
+JNI_OnLoad called
diff --git a/test/595-profile-saving/info.txt b/test/595-profile-saving/info.txt
new file mode 100644
index 000000000..5d318f5b1
--- /dev/null
+++ b/test/595-profile-saving/info.txt
@@ -0,0 +1 @@
+Check that profile recording works even when JIT compilation is not enabled.
diff --git a/test/595-profile-saving/profile-saving.cc b/test/595-profile-saving/profile-saving.cc
new file mode 100644
index 000000000..0d26f454c
--- /dev/null
+++ b/test/595-profile-saving/profile-saving.cc
@@ -0,0 +1,84 @@
+/*
+ * 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.
+ */
+
+#include "dex_file.h"
+
+#include "art_method-inl.h"
+#include "jit/offline_profiling_info.h"
+#include "jit/profile_saver.h"
+#include "jni.h"
+#include "method_reference.h"
+#include "mirror/class-inl.h"
+#include "oat_file_assistant.h"
+#include "oat_file_manager.h"
+#include "scoped_thread_state_change.h"
+#include "ScopedUtfChars.h"
+#include "thread.h"
+
+namespace art {
+namespace {
+
+class CreateProfilingInfoVisitor : public StackVisitor {
+ public:
+  explicit CreateProfilingInfoVisitor(Thread* thread, const char* method_name)
+      SHARED_REQUIRES(Locks::mutator_lock_)
+      : StackVisitor(thread, nullptr, StackVisitor::StackWalkKind::kIncludeInlinedFrames),
+        method_name_(method_name) {}
+
+  bool VisitFrame() SHARED_REQUIRES(Locks::mutator_lock_) {
+    ArtMethod* m = GetMethod();
+    std::string m_name(m->GetName());
+
+    if (m_name.compare(method_name_) == 0) {
+      ProfilingInfo::Create(Thread::Current(), m, /* retry_allocation */ true);
+      method_index_ = m->GetDexMethodIndex();
+      return false;
+    }
+    return true;
+  }
+
+  int method_index_ = -1;
+  const char* const method_name_;
+};
+
+extern "C" JNIEXPORT jint JNICALL Java_Main_ensureProfilingInfo(JNIEnv* env,
+                                                                jclass,
+                                                                jstring method_name) {
+  ScopedUtfChars chars(env, method_name);
+  CHECK(chars.c_str() != nullptr);
+  ScopedObjectAccess soa(Thread::Current());
+  CreateProfilingInfoVisitor visitor(soa.Self(), chars.c_str());
+  visitor.WalkStack();
+  return visitor.method_index_;
+}
+
+extern "C" JNIEXPORT void JNICALL Java_Main_ensureProfileProcessing(JNIEnv*, jclass) {
+  ProfileSaver::ForceProcessProfiles();
+}
+
+extern "C" JNIEXPORT jboolean JNICALL Java_Main_presentInProfile(
+      JNIEnv* env, jclass cls, jstring filename, jint method_index) {
+  ScopedUtfChars filename_chars(env, filename);
+  CHECK(filename_chars.c_str() != nullptr);
+  ScopedObjectAccess soa(Thread::Current());
+  const DexFile* dex_file = soa.Decode<mirror::Class*>(cls)->GetDexCache()->GetDexFile();
+  return ProfileSaver::HasSeenMethod(std::string(filename_chars.c_str()),
+                                     dex_file,
+                                     static_cast<uint16_t>(method_index));
+}
+
+}  // namespace
+}  // namespace art
diff --git a/test/595-profile-saving/run b/test/595-profile-saving/run
new file mode 100644
index 000000000..068ad03ce
--- /dev/null
+++ b/test/595-profile-saving/run
@@ -0,0 +1,27 @@
+#!/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.
+
+# Use
+# --compiler-filter=interpret-only to make sure that the test is not compiled AOT
+# and to make sure the test is not compiled  when loaded (by PathClassLoader)
+# -Xjitsaveprofilinginfo to enable profile saving
+# -Xusejit:false to disable jit and only test profiles.
+exec ${RUN} \
+  -Xcompiler-option --compiler-filter=interpret-only \
+  --runtime-option '-Xcompiler-option --compiler-filter=interpret-only' \
+  --runtime-option -Xjitsaveprofilinginfo \
+  --runtime-option -Xusejit:false \
+  "${@}"
diff --git a/test/595-profile-saving/src/Main.java b/test/595-profile-saving/src/Main.java
new file mode 100644
index 000000000..039503f7a
--- /dev/null
+++ b/test/595-profile-saving/src/Main.java
@@ -0,0 +1,99 @@
+/*
+ * 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.io.File;
+import java.io.IOException;
+import java.lang.reflect.Method;
+
+public class Main {
+
+  public static void main(String[] args) throws Exception {
+    System.loadLibrary(args[0]);
+
+    File file = null;
+    try {
+      file = createTempFile();
+      // String codePath = getDexBaseLocation();
+      String codePath = System.getenv("DEX_LOCATION") + "/595-profile-saving.jar";
+      VMRuntime.registerAppInfo(file.getPath(),
+                                System.getenv("DEX_LOCATION"),
+                                new String[] {codePath},
+                                /* foreignProfileDir */ null);
+
+      int methodIdx = $opt$noinline$testProfile();
+      ensureProfileProcessing();
+      if (!presentInProfile(file.getPath(), methodIdx)) {
+        throw new RuntimeException("Method with index " + methodIdx + " not in the profile");
+      }
+    } finally {
+      if (file != null) {
+        file.delete();
+      }
+    }
+  }
+
+  public static int $opt$noinline$testProfile() {
+    if (doThrow) throw new Error();
+    // Make sure we have a profile info for this method without the need to loop.
+    return ensureProfilingInfo("$opt$noinline$testProfile");
+  }
+
+  // Return the dex method index.
+  public static native int ensureProfilingInfo(String methodName);
+  // Ensures the profile saver does its usual processing.
+  public static native void ensureProfileProcessing();
+  // Checks if the profiles saver knows about the method.
+  public static native boolean presentInProfile(String profile, int methodIdx);
+
+  public static boolean doThrow = false;
+  private static final String TEMP_FILE_NAME_PREFIX = "dummy";
+  private static final String TEMP_FILE_NAME_SUFFIX = "-file";
+
+  static native String getProfileInfoDump(
+      String filename);
+
+  private static File createTempFile() throws Exception {
+    try {
+      return File.createTempFile(TEMP_FILE_NAME_PREFIX, TEMP_FILE_NAME_SUFFIX);
+    } catch (IOException e) {
+      System.setProperty("java.io.tmpdir", "/data/local/tmp");
+      try {
+        return File.createTempFile(TEMP_FILE_NAME_PREFIX, TEMP_FILE_NAME_SUFFIX);
+      } catch (IOException e2) {
+        System.setProperty("java.io.tmpdir", "/sdcard");
+        return File.createTempFile(TEMP_FILE_NAME_PREFIX, TEMP_FILE_NAME_SUFFIX);
+      }
+    }
+  }
+
+  private static class VMRuntime {
+    private static final Method registerAppInfoMethod;
+    static {
+      try {
+        Class<? extends Object> c = Class.forName("dalvik.system.VMRuntime");
+        registerAppInfoMethod = c.getDeclaredMethod("registerAppInfo",
+            String.class, String.class, String[].class, String.class);
+      } catch (Exception e) {
+        throw new RuntimeException(e);
+      }
+    }
+
+    public static void registerAppInfo(String profile, String appDir,
+                                       String[] codePaths, String foreignDir) throws Exception {
+      registerAppInfoMethod.invoke(null, profile, appDir, codePaths, foreignDir);
+    }
+  }
+}
diff --git a/test/596-app-images/app_images.cc b/test/596-app-images/app_images.cc
new file mode 100644
index 000000000..a5bbf5fba
--- /dev/null
+++ b/test/596-app-images/app_images.cc
@@ -0,0 +1,68 @@
+/*
+ * 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.
+ */
+
+#include <iostream>
+#include <pthread.h>
+#include <stdio.h>
+#include <vector>
+
+#include "gc/heap.h"
+#include "gc/space/image_space.h"
+#include "gc/space/space-inl.h"
+#include "image.h"
+#include "jni.h"
+#include "mirror/class.h"
+#include "runtime.h"
+#include "scoped_thread_state_change.h"
+
+namespace art {
+
+namespace {
+
+extern "C" JNIEXPORT jboolean JNICALL Java_Main_checkAppImageLoaded(JNIEnv*, jclass) {
+  ScopedObjectAccess soa(Thread::Current());
+  for (auto* space : Runtime::Current()->GetHeap()->GetContinuousSpaces()) {
+    if (space->IsImageSpace()) {
+      auto* image_space = space->AsImageSpace();
+      const auto& image_header = image_space->GetImageHeader();
+      if (image_header.IsAppImage()) {
+        return JNI_TRUE;
+      }
+    }
+  }
+  return JNI_FALSE;
+}
+
+extern "C" JNIEXPORT jboolean JNICALL Java_Main_checkAppImageContains(JNIEnv*, jclass, jclass c) {
+  ScopedObjectAccess soa(Thread::Current());
+  mirror::Class* klass_ptr = soa.Decode<mirror::Class*>(c);
+  for (auto* space : Runtime::Current()->GetHeap()->GetContinuousSpaces()) {
+    if (space->IsImageSpace()) {
+      auto* image_space = space->AsImageSpace();
+      const auto& image_header = image_space->GetImageHeader();
+      if (image_header.IsAppImage()) {
+        if (image_space->HasAddress(klass_ptr)) {
+          return JNI_TRUE;
+        }
+      }
+    }
+  }
+  return JNI_FALSE;
+}
+
+}  // namespace
+
+}  // namespace art
diff --git a/test/596-app-images/expected.txt b/test/596-app-images/expected.txt
new file mode 100644
index 000000000..6a5618ebc
--- /dev/null
+++ b/test/596-app-images/expected.txt
@@ -0,0 +1 @@
+JNI_OnLoad called
diff --git a/test/596-app-images/info.txt b/test/596-app-images/info.txt
new file mode 100644
index 000000000..a3d5e7ea7
--- /dev/null
+++ b/test/596-app-images/info.txt
@@ -0,0 +1 @@
+Tests that app-images are loaded and used.
diff --git a/test/596-app-images/src/Main.java b/test/596-app-images/src/Main.java
new file mode 100644
index 000000000..75b31b806
--- /dev/null
+++ b/test/596-app-images/src/Main.java
@@ -0,0 +1,33 @@
+/*
+ * 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.
+ */
+
+class Main {
+  static class Inner {
+    public static int abc = 0;
+  }
+
+  public static void main(String[] args) {
+    System.loadLibrary(args[0]);
+    if (!checkAppImageLoaded()) {
+      System.out.println("App image is not loaded!");
+    } else if (!checkAppImageContains(Inner.class)) {
+      System.out.println("App image does not contain Inner!");
+    }
+  }
+
+  public static native boolean checkAppImageLoaded();
+  public static native boolean checkAppImageContains(Class<?> klass);
+}
diff --git a/test/596-checker-dead-phi/expected.txt b/test/596-checker-dead-phi/expected.txt
new file mode 100644
index 000000000..d81cc0710
--- /dev/null
+++ b/test/596-checker-dead-phi/expected.txt
@@ -0,0 +1 @@
+42
diff --git a/test/596-checker-dead-phi/info.txt b/test/596-checker-dead-phi/info.txt
new file mode 100644
index 000000000..7f7cf0f9e
--- /dev/null
+++ b/test/596-checker-dead-phi/info.txt
@@ -0,0 +1,2 @@
+Regression test for optimizing where we used to replace a dead loop
+phi with its first incoming input.
diff --git a/test/596-checker-dead-phi/smali/IrreducibleLoop.smali b/test/596-checker-dead-phi/smali/IrreducibleLoop.smali
new file mode 100644
index 000000000..bab2ba99c
--- /dev/null
+++ b/test/596-checker-dead-phi/smali/IrreducibleLoop.smali
@@ -0,0 +1,74 @@
+# 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.
+
+.class public LIrreducibleLoop;
+
+.super Ljava/lang/Object;
+
+# Test case where liveness analysis produces linear order where loop blocks are
+# not adjacent. This revealed a bug in our SSA builder, where a dead loop phi would
+# be replaced by its incoming input during SsaRedundantPhiElimination.
+
+# Check that the outer loop suspend check environment only has the parameter vreg.
+## CHECK-START: int IrreducibleLoop.liveness(int) builder (after)
+## CHECK-DAG:     <<Phi:i\d+>> Phi reg:4 loop:{{B\d+}} irreducible:false
+## CHECK-DAG:     SuspendCheck env:[[_,_,_,_,<<Phi>>]] loop:{{B\d+}} irreducible:false
+
+# Check that the linear order has non-adjacent loop blocks.
+## CHECK-START: int IrreducibleLoop.liveness(int) liveness (after)
+## CHECK-DAG:     Mul liveness:<<LPreEntry2:\d+>>
+## CHECK-DAG:     Add liveness:<<LBackEdge1:\d+>>
+## CHECK-EVAL:    <<LBackEdge1>> < <<LPreEntry2>>
+
+.method public static liveness(I)I
+    .registers 5
+
+    const-string v1, "MyString"
+
+    :header1
+    if-eqz p0, :body1
+
+    :exit
+    return p0
+
+    :body1
+    # The test will generate an incorrect linear order when the following IF swaps
+    # its successors. To do that, load a boolean value and compare NotEqual to 1.
+    sget-boolean v2, LIrreducibleLoop;->f:Z
+    const v3, 1
+    if-ne v2, v3, :pre_header2
+
+    :pre_entry2
+    # Add a marker on the irreducible loop entry.
+    mul-int/2addr p0, p0
+    goto :back_edge2
+
+    :back_edge2
+    goto :header2
+
+    :header2
+    if-eqz p0, :back_edge2
+
+    :back_edge1
+    # Add a marker on the outer loop back edge.
+    add-int/2addr p0, p0
+    # Set a wide register, to have v1 undefined at the back edge.
+    const-wide/16 v0, 0x1
+    goto :header1
+
+    :pre_header2
+    goto :header2
+.end method
+
+.field public static f:Z
diff --git a/test/596-checker-dead-phi/src/Main.java b/test/596-checker-dead-phi/src/Main.java
new file mode 100644
index 000000000..5a3fffc8f
--- /dev/null
+++ b/test/596-checker-dead-phi/src/Main.java
@@ -0,0 +1,32 @@
+/*
+ * 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.lang.reflect.Method;
+
+public class Main {
+  // Workaround for b/18051191.
+  class InnerClass {}
+
+  public static void main(String[] args) throws Exception {
+    Class<?> c = Class.forName("IrreducibleLoop");
+    // Note that we don't actually enter the loops in the 'liveness'
+    // method, so this is just a sanity check that part of the code we
+    // generated for that method is correct.
+    Method m = c.getMethod("liveness", int.class);
+    Object[] arguments = { 42 };
+    System.out.println(m.invoke(null, arguments));
+  }
+}
diff --git a/test/597-deopt-new-string/deopt.cc b/test/597-deopt-new-string/deopt.cc
new file mode 100644
index 000000000..844a786e1
--- /dev/null
+++ b/test/597-deopt-new-string/deopt.cc
@@ -0,0 +1,59 @@
+/*
+ * 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.
+ */
+
+#include "jni.h"
+#include "mirror/class-inl.h"
+#include "runtime.h"
+#include "thread_list.h"
+#include "thread_state.h"
+#include "gc/gc_cause.h"
+#include "gc/scoped_gc_critical_section.h"
+
+namespace art {
+
+extern "C" JNIEXPORT void JNICALL Java_Main_deoptimizeAll(
+    JNIEnv* env,
+    jclass cls ATTRIBUTE_UNUSED) {
+  ScopedObjectAccess soa(env);
+  ScopedThreadSuspension sts(Thread::Current(), kWaitingForDeoptimization);
+  gc::ScopedGCCriticalSection gcs(Thread::Current(),
+                                  gc::kGcCauseInstrumentation,
+                                  gc::kCollectorTypeInstrumentation);
+  // We need to suspend mutator threads first.
+  ScopedSuspendAll ssa(__FUNCTION__);
+  static bool first = true;
+  if (first) {
+    // We need to enable deoptimization once in order to call DeoptimizeEverything().
+    Runtime::Current()->GetInstrumentation()->EnableDeoptimization();
+    first = false;
+  }
+  Runtime::Current()->GetInstrumentation()->DeoptimizeEverything("test");
+}
+
+extern "C" JNIEXPORT void JNICALL Java_Main_undeoptimizeAll(
+    JNIEnv* env,
+    jclass cls ATTRIBUTE_UNUSED) {
+  ScopedObjectAccess soa(env);
+  ScopedThreadSuspension sts(Thread::Current(), kWaitingForDeoptimization);
+  gc::ScopedGCCriticalSection gcs(Thread::Current(),
+                                  gc::kGcCauseInstrumentation,
+                                  gc::kCollectorTypeInstrumentation);
+  // We need to suspend mutator threads first.
+  ScopedSuspendAll ssa(__FUNCTION__);
+  Runtime::Current()->GetInstrumentation()->UndeoptimizeEverything("test");
+}
+
+}  // namespace art
diff --git a/test/597-deopt-new-string/expected.txt b/test/597-deopt-new-string/expected.txt
new file mode 100644
index 000000000..f993efcda
--- /dev/null
+++ b/test/597-deopt-new-string/expected.txt
@@ -0,0 +1,2 @@
+JNI_OnLoad called
+Finishing
diff --git a/test/597-deopt-new-string/info.txt b/test/597-deopt-new-string/info.txt
new file mode 100644
index 000000000..1bd1f79c0
--- /dev/null
+++ b/test/597-deopt-new-string/info.txt
@@ -0,0 +1 @@
+Regression test for b/28555675
diff --git a/test/597-deopt-new-string/run b/test/597-deopt-new-string/run
new file mode 100644
index 000000000..9776ab3eb
--- /dev/null
+++ b/test/597-deopt-new-string/run
@@ -0,0 +1,18 @@
+#!/bin/bash
+#
+# 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.
+
+# We want to run in debuggable mode which keeps the call into StringFactory.newEmptyString().
+exec ${RUN} -Xcompiler-option --debuggable "${@}"
diff --git a/test/597-deopt-new-string/src/Main.java b/test/597-deopt-new-string/src/Main.java
new file mode 100644
index 000000000..1224e407b
--- /dev/null
+++ b/test/597-deopt-new-string/src/Main.java
@@ -0,0 +1,75 @@
+/*
+ * 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.
+ */
+
+public class Main implements Runnable {
+    static final int numberOfThreads = 2;
+    static final int totalOperations = 40000;
+    static boolean sFlag = false;
+    static volatile boolean done = false;
+    int threadIndex;
+
+    public static native void deoptimizeAll();
+    public static native void undeoptimizeAll();
+
+    Main(int index) {
+        threadIndex = index;
+    }
+
+    public static void main(String[] args) throws Exception {
+        System.loadLibrary(args[0]);
+
+        final Thread[] threads = new Thread[numberOfThreads];
+        for (int t = 0; t < threads.length; t++) {
+            threads[t] = new Thread(new Main(t));
+            threads[t].start();
+        }
+        for (Thread t : threads) {
+            t.join();
+        }
+        System.out.println("Finishing");
+    }
+
+    public String $noinline$run0() {
+        // Prevent inlining.
+        if (sFlag) {
+            throw new Error();
+        }
+        char[] arr = {'a', 'b', 'c'};
+        return new String(arr, 0, arr.length);
+    }
+
+    public void run() {
+        if (threadIndex == 0) {
+            // This thread keeps doing deoptimization of all threads.
+            // Hopefully that will trigger one deoptimization when returning from
+            // StringFactory.newEmptyString() in one of the other threads.
+            for (int i = 0; i < totalOperations; ++i) {
+                if (i % 50 == 0) {
+                    deoptimizeAll();
+                }
+                if (i % 50 == 25) {
+                    undeoptimizeAll();
+                }
+            }
+            done = true;
+        } else {
+            // This thread keeps doing new String() from a char array.
+            while (!done) {
+                $noinline$run0();
+            }
+        }
+    }
+}
diff --git a/test/598-checker-irreducible-dominance/expected.txt b/test/598-checker-irreducible-dominance/expected.txt
new file mode 100644
index 000000000..e69de29bb
diff --git a/test/598-checker-irreducible-dominance/info.txt b/test/598-checker-irreducible-dominance/info.txt
new file mode 100644
index 000000000..8ca4e63be
--- /dev/null
+++ b/test/598-checker-irreducible-dominance/info.txt
@@ -0,0 +1,2 @@
+Regression test for HGraphBuilder which would compute wrong dominance information
+in the presence of irreducible loops.
\ No newline at end of file
diff --git a/test/598-checker-irreducible-dominance/smali/IrreducibleLoop.smali b/test/598-checker-irreducible-dominance/smali/IrreducibleLoop.smali
new file mode 100644
index 000000000..4d8b5152f
--- /dev/null
+++ b/test/598-checker-irreducible-dominance/smali/IrreducibleLoop.smali
@@ -0,0 +1,52 @@
+# 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.
+
+.class public LIrreducibleLoop;
+.super Ljava/lang/Object;
+
+# Test case in which `inner_back_edge` is not dominated by `inner_header` and
+# causes `outer_back_edge` to not be dominated by `outer_header`. HGraphBuilder
+# not do a fix-point iteration and would miss the path to `outer_back_edge`
+# through `inner_back_edge` and incorrectly label the outer loop non-irreducible.
+
+## CHECK-START: int IrreducibleLoop.dominance(int) builder (after)
+## CHECK:         Add irreducible:true
+
+.method public static dominance(I)I
+    .registers 2
+
+    if-eqz p0, :outer_header
+    goto :inner_back_edge
+
+    :outer_header
+    if-eqz p0, :inner_header
+
+    :outer_branch_exit
+    if-eqz p0, :outer_merge
+    return p0
+
+    :inner_header
+    goto :outer_merge
+
+    :inner_back_edge
+    goto :inner_header
+
+    :outer_merge
+    if-eqz p0, :inner_back_edge
+
+    :outer_back_edge
+    add-int/2addr p0, p0
+    goto :outer_header
+
+.end method
diff --git a/test/598-checker-irreducible-dominance/src/Main.java b/test/598-checker-irreducible-dominance/src/Main.java
new file mode 100644
index 000000000..38b2ab438
--- /dev/null
+++ b/test/598-checker-irreducible-dominance/src/Main.java
@@ -0,0 +1,25 @@
+/*
+ * 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.
+ */
+
+public class Main {
+  // Workaround for b/18051191.
+  class InnerClass {}
+
+  public static void main(String[] args) {
+    // Nothing to run. This regression test merely makes sure the smali test
+    // case successfully compiles.
+  }
+}
diff --git a/test/599-checker-irreducible-loop/expected.txt b/test/599-checker-irreducible-loop/expected.txt
new file mode 100644
index 000000000..573541ac9
--- /dev/null
+++ b/test/599-checker-irreducible-loop/expected.txt
@@ -0,0 +1 @@
+0
diff --git a/test/599-checker-irreducible-loop/info.txt b/test/599-checker-irreducible-loop/info.txt
new file mode 100644
index 000000000..1e0dd0228
--- /dev/null
+++ b/test/599-checker-irreducible-loop/info.txt
@@ -0,0 +1,2 @@
+Regression test for optimizing in the presence of
+an irreducible loop.
diff --git a/test/599-checker-irreducible-loop/smali/IrreducibleLoop.smali b/test/599-checker-irreducible-loop/smali/IrreducibleLoop.smali
new file mode 100644
index 000000000..5331fd6a3
--- /dev/null
+++ b/test/599-checker-irreducible-loop/smali/IrreducibleLoop.smali
@@ -0,0 +1,56 @@
+# 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.
+
+.class public LIrreducibleLoop;
+
+.super Ljava/lang/Object;
+
+## CHECK-START: int IrreducibleLoop.test(int) GVN (before)
+## CHECK-DAG:                     LoadClass loop:none
+## CHECK-DAG:                     LoadClass loop:{{B\d+}} outer_loop:none
+
+## CHECK-START: int IrreducibleLoop.test(int) GVN (after)
+## CHECK-DAG:                     LoadClass loop:none
+## CHECK-DAG:                     LoadClass loop:{{B\d+}} outer_loop:none
+.method public static test(I)I
+   .registers 2
+
+   sget v0, LIrreducibleLoop;->field1:I
+   sput v0, LIrreducibleLoop;->field2:I
+
+   if-eqz p0, :loop_entry
+   goto :exit
+
+   :loop_entry
+   if-eqz p0, :irreducible_loop_entry
+   sget v0, LIrreducibleLoop;->field2:I
+   sput v0, LIrreducibleLoop;->field1:I
+   if-eqz v0, :exit
+   goto :irreducible_other_loop_entry
+
+   :irreducible_loop_entry
+   if-eqz p0, :loop_back_edge
+   :irreducible_other_loop_entry
+   if-eqz v0, :loop_back_edge
+   goto :irreducible_loop_entry
+
+   :loop_back_edge
+   goto :loop_entry
+
+   :exit
+   return v0
+.end method
+
+.field public static field1:I
+.field public static field2:I
diff --git a/test/599-checker-irreducible-loop/src/Main.java b/test/599-checker-irreducible-loop/src/Main.java
new file mode 100644
index 000000000..b47721f72
--- /dev/null
+++ b/test/599-checker-irreducible-loop/src/Main.java
@@ -0,0 +1,30 @@
+/*
+ * 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.lang.reflect.Method;
+
+public class Main {
+  // Workaround for b/18051191.
+  class InnerClass {}
+
+  public static void main(String[] args) throws Exception {
+    Class<?> c = Class.forName("IrreducibleLoop");
+    Method m = c.getMethod("test", int.class);
+    Object[] arguments = { 42 };
+    // Invoke the code just for sanity checking.
+    System.out.println(m.invoke(null, arguments));
+  }
+}
diff --git a/test/800-smali/smali/b_28187158.smali b/test/800-smali/smali/b_28187158.smali
index 14d5cec7e..47e5ef64f 100644
--- a/test/800-smali/smali/b_28187158.smali
+++ b/test/800-smali/smali/b_28187158.smali
@@ -9,4 +9,3 @@
    iget v0, p0, Ljava/lang/System;->in:Ljava/io/InputStream;
    return-void
 .end method
-
diff --git a/test/800-smali/src/Main.java b/test/800-smali/src/Main.java
index c883b7f0f..b2fc00562 100644
--- a/test/800-smali/src/Main.java
+++ b/test/800-smali/src/Main.java
@@ -174,7 +174,7 @@ public class Main {
         testCases.add(new TestCase("b/27799205 (5)", "B27799205Helper", "run5", null,
                 new VerifyError(), null));
         testCases.add(new TestCase("b/27799205 (6)", "B27799205Helper", "run6", null, null, null));
-        testCases.add(new TestCase("b/28187158", "B28187158", "run", new Object[] { null} ,
+        testCases.add(new TestCase("b/28187158", "B28187158", "run", new Object[] { null },
                 new VerifyError(), null));
     }
 
diff --git a/test/803-no-super/expected.txt b/test/803-no-super/expected.txt
new file mode 100644
index 000000000..503699139
--- /dev/null
+++ b/test/803-no-super/expected.txt
@@ -0,0 +1,2 @@
+java.lang.ClassNotFoundException: NoSuper1
+Done!
diff --git a/test/803-no-super/info.txt b/test/803-no-super/info.txt
new file mode 100644
index 000000000..0178a446e
--- /dev/null
+++ b/test/803-no-super/info.txt
@@ -0,0 +1,3 @@
+Regression test that temp (erroneous) classes don't get conflict tables created.
+
+Obviously needs to run under Dalvik or ART.
diff --git a/test/803-no-super/smali/nosuper1.smali b/test/803-no-super/smali/nosuper1.smali
new file mode 100644
index 000000000..df2eaa5ca
--- /dev/null
+++ b/test/803-no-super/smali/nosuper1.smali
@@ -0,0 +1,3 @@
+.class public LNoSuper1;
+
+.super LNoClass;
diff --git a/test/803-no-super/src/Main.java b/test/803-no-super/src/Main.java
new file mode 100644
index 000000000..a07e042c3
--- /dev/null
+++ b/test/803-no-super/src/Main.java
@@ -0,0 +1,29 @@
+/*
+ * 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.
+ */
+
+/**
+ * Attempt to load class with no superclass.
+ */
+public class Main {
+    public static void main(String[] args) throws Exception {
+        try {
+            Class<?> c = Class.forName("NoSuper1");
+        } catch (Exception e) {
+            System.out.println(e);
+        }
+        System.out.println("Done!");
+    }
+}
diff --git a/test/804-class-extends-itself/expected.txt b/test/804-class-extends-itself/expected.txt
new file mode 100644
index 000000000..b98f963ce
--- /dev/null
+++ b/test/804-class-extends-itself/expected.txt
@@ -0,0 +1,2 @@
+Caught ClassCircularityError
+Done!
diff --git a/test/804-class-extends-itself/info.txt b/test/804-class-extends-itself/info.txt
new file mode 100644
index 000000000..c48934c21
--- /dev/null
+++ b/test/804-class-extends-itself/info.txt
@@ -0,0 +1 @@
+Exercise class linker check for classes extending themselves (b/28685551).
diff --git a/test/804-class-extends-itself/smali/Main.smali b/test/804-class-extends-itself/smali/Main.smali
new file mode 100644
index 000000000..5c349edcc
--- /dev/null
+++ b/test/804-class-extends-itself/smali/Main.smali
@@ -0,0 +1,57 @@
+# 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.
+
+# We cannot implement Main in Java, as this would require to run
+# dexmerger (to merge the Dex file produced from Smali code and the
+# Dex file produced from Java code), which loops indefinitely when
+# processing class B28685551, as this class inherits from itself.  As
+# a workaround, implement Main using Smali (we could also have used
+# multidex, but this requires a custom build script).
+
+.class public LMain;
+.super Ljava/lang/Object;
+
+.method public static main([Ljava/lang/String;)V
+    .registers 3
+    .param p0, "args"
+
+    invoke-static {}, LMain;->test()V
+    sget-object v0, Ljava/lang/System;->out:Ljava/io/PrintStream;
+    const-string v1, "Done!"
+    invoke-virtual {v0, v1}, Ljava/io/PrintStream;->println(Ljava/lang/String;)V
+    return-void
+.end method
+
+.method static test()V
+    .registers 4
+
+    :try_start
+    const-string v2, "B28685551"
+    invoke-static {v2}, Ljava/lang/Class;->forName(Ljava/lang/String;)Ljava/lang/Class;
+    :try_end
+    .catch Ljava/lang/ClassCircularityError; {:try_start .. :try_end} :catch
+
+    move-result-object v0
+
+    :goto_7
+    return-void
+
+    :catch
+    move-exception v1
+    .local v1, "e":Ljava/lang/ClassCircularityError;
+    sget-object v2, Ljava/lang/System;->out:Ljava/io/PrintStream;
+    const-string v3, "Caught ClassCircularityError"
+    invoke-virtual {v2, v3}, Ljava/io/PrintStream;->println(Ljava/lang/String;)V
+    goto :goto_7
+.end method
diff --git a/test/804-class-extends-itself/smali/b_28685551.smali b/test/804-class-extends-itself/smali/b_28685551.smali
new file mode 100644
index 000000000..d98c6e3b3
--- /dev/null
+++ b/test/804-class-extends-itself/smali/b_28685551.smali
@@ -0,0 +1,18 @@
+# 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.
+
+# Regression test for a class inheriting from itself.
+
+.class public LB28685551;
+.super LB28685551;
diff --git a/test/960-default-smali/expected.txt b/test/960-default-smali/expected.txt
index 7671eed5d..f3db93f87 100644
--- a/test/960-default-smali/expected.txt
+++ b/test/960-default-smali/expected.txt
@@ -82,3 +82,19 @@ J-virtual           A.SayHiTwice()='Hi Hi '
 J-interface   Greeter.SayHiTwice()='Hi Hi '
 J-virtual           J.SayHiTwice()='Hi Hi '
 End testing for type J
+Testing for type K
+K-interface       Foo.bar()='foobar'
+K-virtual           K.bar()='foobar'
+End testing for type K
+Testing for type L
+L-interface       Foo.bar()='foobar'
+L-virtual           K.bar()='foobar'
+L-virtual           L.bar()='foobar'
+End testing for type L
+Testing for type M
+M-interface       Foo.bar()='BAZ!'
+M-interface     Fooer.bar()='BAZ!'
+M-virtual           K.bar()='BAZ!'
+M-virtual           L.bar()='BAZ!'
+M-virtual           M.bar()='BAZ!'
+End testing for type M
diff --git a/test/960-default-smali/src/Foo.java b/test/960-default-smali/src/Foo.java
new file mode 100644
index 000000000..ed5b35f47
--- /dev/null
+++ b/test/960-default-smali/src/Foo.java
@@ -0,0 +1,20 @@
+/*
+ * 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.
+ */
+interface Foo {
+  public default String bar() {
+    return "foobar";
+  }
+}
diff --git a/test/960-default-smali/src/Fooer.java b/test/960-default-smali/src/Fooer.java
new file mode 100644
index 000000000..d8a5f6163
--- /dev/null
+++ b/test/960-default-smali/src/Fooer.java
@@ -0,0 +1,19 @@
+/*
+ * 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.
+ */
+
+interface Fooer extends Foo {
+  public String bar();
+}
diff --git a/test/960-default-smali/src/K.java b/test/960-default-smali/src/K.java
new file mode 100644
index 000000000..4426be719
--- /dev/null
+++ b/test/960-default-smali/src/K.java
@@ -0,0 +1,17 @@
+/*
+ * 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.
+ */
+
+class K implements Foo { }
diff --git a/test/960-default-smali/src/L.java b/test/960-default-smali/src/L.java
new file mode 100644
index 000000000..c08ab72a9
--- /dev/null
+++ b/test/960-default-smali/src/L.java
@@ -0,0 +1,17 @@
+/*
+ * 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.
+ */
+
+class L extends K { }
diff --git a/test/960-default-smali/src/M.java b/test/960-default-smali/src/M.java
new file mode 100644
index 000000000..affe7e9c9
--- /dev/null
+++ b/test/960-default-smali/src/M.java
@@ -0,0 +1,21 @@
+/*
+ * 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.
+ */
+
+class M extends L implements Fooer {
+  public String bar() {
+    return "BAZ!";
+  }
+}
diff --git a/test/960-default-smali/src/classes.xml b/test/960-default-smali/src/classes.xml
index 0aa41f7fb..f3e50c570 100644
--- a/test/960-default-smali/src/classes.xml
+++ b/test/960-default-smali/src/classes.xml
@@ -81,6 +81,27 @@
       <implements> </implements>
       <methods> </methods>
     </class>
+
+    <class name="K" super="java/lang/Object">
+      <implements>
+        <item>Foo</item>
+      </implements>
+      <methods> </methods>
+    </class>
+
+    <class name="L" super="K">
+      <implements> </implements>
+      <methods> </methods>
+    </class>
+
+    <class name="M" super="L">
+      <implements>
+        <item>Fooer</item>
+      </implements>
+      <methods>
+        <method>bar</method>
+      </methods>
+    </class>
   </classes>
 
   <interfaces>
@@ -123,5 +144,22 @@
         <method type="abstract">GetPlace</method>
       </methods>
     </interface>
+
+    <interface name="Foo" super="java/lang/Object">
+      <implements>
+      </implements>
+      <methods>
+        <method type="default">bar</method>
+      </methods>
+    </interface>
+
+    <interface name="Fooer" super="java/lang/Object">
+      <implements>
+        <item>Foo</item>
+      </implements>
+      <methods>
+        <method type="abstract">bar</method>
+      </methods>
+    </interface>
   </interfaces>
 </data>
diff --git a/test/976-conflict-no-methods/expected.txt b/test/976-conflict-no-methods/expected.txt
new file mode 100644
index 000000000..656dfc57d
--- /dev/null
+++ b/test/976-conflict-no-methods/expected.txt
@@ -0,0 +1 @@
+Pass
diff --git a/test/976-conflict-no-methods/info.txt b/test/976-conflict-no-methods/info.txt
new file mode 100644
index 000000000..cdc314903
--- /dev/null
+++ b/test/976-conflict-no-methods/info.txt
@@ -0,0 +1 @@
+Regression test for classes that have conflict tables but no methods. b/28707801
\ No newline at end of file
diff --git a/test/976-conflict-no-methods/smali/Iface.smali b/test/976-conflict-no-methods/smali/Iface.smali
new file mode 100644
index 000000000..aa4ec3768
--- /dev/null
+++ b/test/976-conflict-no-methods/smali/Iface.smali
@@ -0,0 +1,281 @@
+# /*
+#  * 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.
+#  */
+#
+# public interface Iface2 {
+#    public void abstractMethod0();
+#    public void abstractMethod1();
+#    public void abstractMethod2();
+#    public void abstractMethod3();
+#    public void abstractMethod4();
+#    public void abstractMethod5();
+#    public void abstractMethod6();
+#    public void abstractMethod7();
+#    public void abstractMethod8();
+#    public void abstractMethod9();
+#    public void abstractMethod10();
+#    public void abstractMethod11();
+#    public void abstractMethod12();
+#    public void abstractMethod13();
+#    public void abstractMethod14();
+#    public void abstractMethod15();
+#    public void abstractMethod16();
+#    public void abstractMethod17();
+#    public void abstractMethod18();
+#    public void abstractMethod19();
+#    public void abstractMethod20();
+#    public void abstractMethod21();
+#    public void abstractMethod22();
+#    public void abstractMethod23();
+#    public void abstractMethod24();
+#    public void abstractMethod25();
+#    public void abstractMethod26();
+#    public void abstractMethod27();
+#    public void abstractMethod28();
+#    public void abstractMethod29();
+#    public void abstractMethod30();
+#    public void abstractMethod31();
+#    public void abstractMethod32();
+#    public void abstractMethod33();
+#    public void abstractMethod34();
+#    public void abstractMethod35();
+#    public void abstractMethod36();
+#    public void abstractMethod37();
+#    public void abstractMethod38();
+#    public void abstractMethod39();
+#    public void abstractMethod40();
+#    public void abstractMethod41();
+#    public void abstractMethod42();
+#    public void abstractMethod43();
+#    public void abstractMethod44();
+#    public void abstractMethod45();
+#    public void abstractMethod46();
+#    public void abstractMethod47();
+#    public void abstractMethod48();
+#    public void abstractMethod49();
+#    public void abstractMethod50();
+#    public void abstractMethod51();
+#    public void abstractMethod52();
+#    public void abstractMethod53();
+#    public void abstractMethod54();
+#    public void abstractMethod55();
+#    public void abstractMethod56();
+#    public void abstractMethod57();
+#    public void abstractMethod58();
+#    public void abstractMethod59();
+#    public void abstractMethod60();
+#    public void abstractMethod61();
+#    public void abstractMethod62();
+#    public void abstractMethod63();
+#    public void abstractMethod64();
+# }
+
+.class public abstract interface LIface;
+.super Ljava/lang/Object;
+
+.method public abstract abstractMethod0()V
+.end method
+
+.method public abstract abstractMethod1()V
+.end method
+
+.method public abstract abstractMethod2()V
+.end method
+
+.method public abstract abstractMethod3()V
+.end method
+
+.method public abstract abstractMethod4()V
+.end method
+
+.method public abstract abstractMethod5()V
+.end method
+
+.method public abstract abstractMethod6()V
+.end method
+
+.method public abstract abstractMethod7()V
+.end method
+
+.method public abstract abstractMethod8()V
+.end method
+
+.method public abstract abstractMethod9()V
+.end method
+
+.method public abstract abstractMethod10()V
+.end method
+
+.method public abstract abstractMethod11()V
+.end method
+
+.method public abstract abstractMethod12()V
+.end method
+
+.method public abstract abstractMethod13()V
+.end method
+
+.method public abstract abstractMethod14()V
+.end method
+
+.method public abstract abstractMethod15()V
+.end method
+
+.method public abstract abstractMethod16()V
+.end method
+
+.method public abstract abstractMethod17()V
+.end method
+
+.method public abstract abstractMethod18()V
+.end method
+
+.method public abstract abstractMethod19()V
+.end method
+
+.method public abstract abstractMethod20()V
+.end method
+
+.method public abstract abstractMethod21()V
+.end method
+
+.method public abstract abstractMethod22()V
+.end method
+
+.method public abstract abstractMethod23()V
+.end method
+
+.method public abstract abstractMethod24()V
+.end method
+
+.method public abstract abstractMethod25()V
+.end method
+
+.method public abstract abstractMethod26()V
+.end method
+
+.method public abstract abstractMethod27()V
+.end method
+
+.method public abstract abstractMethod28()V
+.end method
+
+.method public abstract abstractMethod29()V
+.end method
+
+.method public abstract abstractMethod30()V
+.end method
+
+.method public abstract abstractMethod31()V
+.end method
+
+.method public abstract abstractMethod32()V
+.end method
+
+.method public abstract abstractMethod33()V
+.end method
+
+.method public abstract abstractMethod34()V
+.end method
+
+.method public abstract abstractMethod35()V
+.end method
+
+.method public abstract abstractMethod36()V
+.end method
+
+.method public abstract abstractMethod37()V
+.end method
+
+.method public abstract abstractMethod38()V
+.end method
+
+.method public abstract abstractMethod39()V
+.end method
+
+.method public abstract abstractMethod40()V
+.end method
+
+.method public abstract abstractMethod41()V
+.end method
+
+.method public abstract abstractMethod42()V
+.end method
+
+.method public abstract abstractMethod43()V
+.end method
+
+.method public abstract abstractMethod44()V
+.end method
+
+.method public abstract abstractMethod45()V
+.end method
+
+.method public abstract abstractMethod46()V
+.end method
+
+.method public abstract abstractMethod47()V
+.end method
+
+.method public abstract abstractMethod48()V
+.end method
+
+.method public abstract abstractMethod49()V
+.end method
+
+.method public abstract abstractMethod50()V
+.end method
+
+.method public abstract abstractMethod51()V
+.end method
+
+.method public abstract abstractMethod52()V
+.end method
+
+.method public abstract abstractMethod53()V
+.end method
+
+.method public abstract abstractMethod54()V
+.end method
+
+.method public abstract abstractMethod55()V
+.end method
+
+.method public abstract abstractMethod56()V
+.end method
+
+.method public abstract abstractMethod57()V
+.end method
+
+.method public abstract abstractMethod58()V
+.end method
+
+.method public abstract abstractMethod59()V
+.end method
+
+.method public abstract abstractMethod60()V
+.end method
+
+.method public abstract abstractMethod61()V
+.end method
+
+.method public abstract abstractMethod62()V
+.end method
+
+.method public abstract abstractMethod63()V
+.end method
+
+.method public abstract abstractMethod64()V
+.end method
diff --git a/test/976-conflict-no-methods/smali/Main.smali b/test/976-conflict-no-methods/smali/Main.smali
new file mode 100644
index 000000000..7dd11605b
--- /dev/null
+++ b/test/976-conflict-no-methods/smali/Main.smali
@@ -0,0 +1,358 @@
+# /*
+#  * 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.
+#  */
+#
+.class public LMain;
+.super Ljava/lang/Object;
+.implements LIface;
+
+.method public constructor <init>()V
+    .registers 1
+    invoke-direct {p0}, Ljava/lang/Object;-><init>()V
+    return-void
+.end method
+
+.method public static main([Ljava/lang/String;)V
+    .locals 2
+    sget-object v0, Ljava/lang/System;->out:Ljava/io/PrintStream;
+    const-string v1, "Pass"
+    invoke-virtual {v0, v1}, Ljava/io/PrintStream;->println(Ljava/lang/Object;)V
+    return-void
+.end method
+
+.method public abstractMethod0()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod1()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod2()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod3()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod4()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod5()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod6()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod7()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod8()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod9()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod10()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod11()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod12()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod13()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod14()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod15()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod16()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod17()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod18()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod19()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod20()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod21()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod22()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod23()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod24()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod25()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod26()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod27()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod28()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod29()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod30()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod31()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod32()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod33()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod34()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod35()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod36()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod37()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod38()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod39()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod40()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod41()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod42()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod43()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod44()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod45()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod46()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod47()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod48()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod49()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod50()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod51()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod52()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod53()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod54()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod55()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod56()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod57()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod58()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod59()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod60()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod61()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod62()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod63()V
+    .locals 0
+    return-void
+.end method
+
+.method public abstractMethod64()V
+    .locals 0
+    return-void
+.end method
diff --git a/test/976-conflict-no-methods/smali/NoMethods.smali b/test/976-conflict-no-methods/smali/NoMethods.smali
new file mode 100644
index 000000000..787e34a42
--- /dev/null
+++ b/test/976-conflict-no-methods/smali/NoMethods.smali
@@ -0,0 +1,19 @@
+# /*
+#  * 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.
+#  */
+#
+
+.class public LNoMethods;
+.super LMain;
diff --git a/test/Android.libarttest.mk b/test/Android.libarttest.mk
index e547c72c0..97204d34c 100644
--- a/test/Android.libarttest.mk
+++ b/test/Android.libarttest.mk
@@ -34,6 +34,7 @@ LIBARTTEST_COMMON_SRC_FILES := \
   137-cfi/cfi.cc \
   139-register-natives/regnative.cc \
   141-class-unload/jni_unload.cc \
+  148-multithread-gc-annotations/gc_coverage.cc \
   454-get-vreg/get_vreg_jni.cc \
   457-regs/regs_jni.cc \
   461-get-reference-vreg/get_reference_vreg_jni.cc \
@@ -41,7 +42,10 @@ LIBARTTEST_COMMON_SRC_FILES := \
   497-inlining-and-class-loader/clear_dex_cache.cc \
   543-env-long-ref/env_long_ref.cc \
   566-polymorphic-inlining/polymorphic_inline.cc \
-  570-checker-osr/osr.cc
+  570-checker-osr/osr.cc \
+  595-profile-saving/profile-saving.cc \
+  596-app-images/app_images.cc \
+  597-deopt-new-string/deopt.cc
 
 ART_TARGET_LIBARTTEST_$(ART_PHONY_TEST_TARGET_SUFFIX) += $(ART_TARGET_TEST_OUT)/$(TARGET_ARCH)/libarttest.so
 ART_TARGET_LIBARTTEST_$(ART_PHONY_TEST_TARGET_SUFFIX) += $(ART_TARGET_TEST_OUT)/$(TARGET_ARCH)/libarttestd.so
@@ -90,7 +94,12 @@ define build-libarttest
     include $(BUILD_SHARED_LIBRARY)
   else # host
     LOCAL_CLANG := $(ART_HOST_CLANG)
-    LOCAL_CFLAGS := $(ART_HOST_CFLAGS) $(ART_HOST_DEBUG_CFLAGS)
+    LOCAL_CFLAGS := $(ART_HOST_CFLAGS)
+    ifeq ($$(suffix),d)
+      LOCAL_CFLAGS += $(ART_HOST_DEBUG_CFLAGS)
+    else
+      LOCAL_CFLAGS += $(ART_HOST_NON_DEBUG_CFLAGS)
+    endif
     LOCAL_ASFLAGS := $(ART_HOST_ASFLAGS)
     LOCAL_LDLIBS := $(ART_HOST_LDLIBS) -ldl -lpthread
     LOCAL_IS_HOST_MODULE := true
diff --git a/test/Android.run-test.mk b/test/Android.run-test.mk
index f3cda479d..ee651b549 100644
--- a/test/Android.run-test.mk
+++ b/test/Android.run-test.mk
@@ -563,6 +563,13 @@ endif
 TEST_ART_BROKEN_OPTIMIZING_READ_BARRIER_RUN_TESTS :=
 TEST_ART_BROKEN_JIT_READ_BARRIER_RUN_TESTS :=
 
+TEST_ART_BROKEN_NPIC_RUN_TESTS := 596-app-images
+ifneq (,$(filter npictest,$(PICTEST_TYPES)))
+  ART_TEST_KNOWN_BROKEN += $(call all-run-test-names,$(TARGET_TYPES),$(RUN_TYPES),$(PREBUILD_TYPES), \
+      ${COMPILER_TYPES},$(RELOCATE_TYPES),$(TRACE_TYPES),$(GC_TYPES),$(JNI_TYPES), \
+      $(IMAGE_TYPES),npictest,$(DEBUGGABLE_TYPES),$(TEST_ART_BROKEN_NPIC_RUN_TESTS),$(ALL_ADDRESS_SIZES))
+endif
+
 # Tests that should fail in the heap poisoning configuration with the Optimizing compiler.
 # 055: Exceeds run time limits due to heap poisoning instrumentation (on ARM and ARM64 devices).
 TEST_ART_BROKEN_OPTIMIZING_HEAP_POISONING_RUN_TESTS := \
diff --git a/test/etc/run-test-jar b/test/etc/run-test-jar
index d61fc8f8f..aa45d40cb 100755
--- a/test/etc/run-test-jar
+++ b/test/etc/run-test-jar
@@ -323,11 +323,14 @@ fi
 if [ "$INTERPRETER" = "y" ]; then
     INT_OPTS="-Xint"
     if [ "$VERIFY" = "y" ] ; then
+      INT_OPTS="${INT_OPTS} -Xcompiler-option --compiler-filter=interpret-only"
       COMPILE_FLAGS="${COMPILE_FLAGS} --compiler-filter=interpret-only"
     elif [ "$VERIFY" = "s" ]; then
+      INT_OPTS="${INT_OPTS} -Xcompiler-option --compiler-filter=verify-at-runtime"
       COMPILE_FLAGS="${COMPILE_FLAGS} --compiler-filter=verify-at-runtime"
       DEX_VERIFY="${DEX_VERIFY} -Xverify:softfail"
     else # VERIFY = "n"
+      INT_OPTS="${INT_OPTS} -Xcompiler-option --compiler-filter=verify-none"
       COMPILE_FLAGS="${COMPILE_FLAGS} --compiler-filter=verify-none"
       DEX_VERIFY="${DEX_VERIFY} -Xverify:none"
     fi
@@ -336,18 +339,12 @@ fi
 if [ "$JIT" = "y" ]; then
     INT_OPTS="-Xusejit:true"
     if [ "$VERIFY" = "y" ] ; then
+      INT_OPTS="${INT_OPTS} -Xcompiler-option --compiler-filter=verify-at-runtime"
       COMPILE_FLAGS="${COMPILE_FLAGS} --compiler-filter=verify-at-runtime"
-      if [ "$PREBUILD" = "n" ]; then
-        # Make sure that if we have noprebuild we still JIT as DexClassLoader will
-        # try to compile the dex file.
-        INT_OPTS="${INT_OPTS} -Xcompiler-option --compiler-filter=verify-at-runtime"
-      fi
     else
+      INT_OPTS="${INT_OPTS} -Xcompiler-option --compiler-filter=verify-none"
       COMPILE_FLAGS="${COMPILE_FLAGS} --compiler-filter=verify-none"
       DEX_VERIFY="${DEX_VERIFY} -Xverify:none"
-      if [ "$PREBUILD" = "n" ]; then
-        INT_OPTS="${INT_OPTS} -Xcompiler-option --compiler-filter=verify-none"
-      fi
     fi
 fi
 
@@ -476,11 +473,14 @@ if [ "$HOST" = "n" ]; then
       LD_LIBRARY_PATH=$ANDROID_ROOT/$LIBRARY_DIRECTORY
     fi
 
+    PUBLIC_LIBS=libart.so:libartd.so
+
     # Create a script with the command. The command can get longer than the longest
     # allowed adb command and there is no way to get the exit status from a adb shell
     # command.
     cmdline="cd $DEX_LOCATION && \
              export ANDROID_DATA=$DEX_LOCATION && \
+             export ANDROID_ADDITIONAL_PUBLIC_LIBRARIES=$PUBLIC_LIBS && \
              export DEX_LOCATION=$DEX_LOCATION && \
              export ANDROID_ROOT=$ANDROID_ROOT && \
              $mkdir_cmdline && \
diff --git a/test/run-test b/test/run-test
index fc57d0914..2710ea32b 100755
--- a/test/run-test
+++ b/test/run-test
@@ -122,10 +122,12 @@ never_clean="no"
 have_dex2oat="yes"
 have_patchoat="yes"
 have_image="yes"
-image_suffix=""
 pic_image_suffix=""
 multi_image_suffix=""
 android_root="/system"
+# By default we will use optimizing.
+image_args=""
+image_suffix="-optimizing"
 
 while true; do
     if [ "x$1" = "x--host" ]; then
@@ -148,6 +150,7 @@ while true; do
     elif [ "x$1" = "x--jvm" ]; then
         target_mode="no"
         runtime="jvm"
+        image_args=""
         prebuild_mode="no"
         NEED_DEX="false"
         USE_JACK="false"
@@ -244,22 +247,22 @@ while true; do
         run_args="${run_args} --zygote"
         shift
     elif [ "x$1" = "x--interpreter" ]; then
-        run_args="${run_args} --interpreter --runtime-option -XOatFileManagerCompilerFilter:verify-at-runtime"
+        run_args="${run_args} --interpreter"
         image_suffix="-interpreter"
         shift
     elif [ "x$1" = "x--jit" ]; then
-        run_args="${run_args} --jit --runtime-option -XOatFileManagerCompilerFilter:verify-at-runtime"
+        image_args="--jit"
         image_suffix="-jit"
         shift
     elif [ "x$1" = "x--optimizing" ]; then
-        run_args="${run_args} -Xcompiler-option --compiler-backend=Optimizing"
+        image_args="-Xcompiler-option --compiler-backend=Optimizing"
         image_suffix="-optimizing"
         shift
     elif [ "x$1" = "x--no-verify" ]; then
-        run_args="${run_args} --no-verify --runtime-option -XOatFileManagerCompilerFilter:verify-none"
+        run_args="${run_args} --no-verify"
         shift
     elif [ "x$1" = "x--verify-soft-fail" ]; then
-        run_args="${run_args} --verify-soft-fail --runtime-option -XOatFileManagerCompilerFilter:verify-at-runtime"
+        image_args="--verify-soft-fail"
         image_suffix="-interp-ac"
         shift
     elif [ "x$1" = "x--no-optimize" ]; then
@@ -348,6 +351,7 @@ while true; do
     fi
 done
 
+run_args="${run_args} ${image_args}"
 # Allocate file descriptor real_stderr and redirect it to the shell's error
 # output (fd 2).
 if [ ${BASH_VERSINFO[1]} -ge 4 ] && [ ${BASH_VERSINFO[2]} -ge 1 ]; then
@@ -467,7 +471,7 @@ elif [ "$runtime" = "art" ]; then
         run_args="${run_args} --runtime-option -Djava.library.path=${ANDROID_HOST_OUT}/lib${suffix64}"
     else
         guess_target_arch_name
-        run_args="${run_args} --runtime-option -Djava.library.path=/data/art-test/${target_arch_name}"
+        run_args="${run_args} --runtime-option -Djava.library.path=/data/art-test/${target_arch_name}:/system/lib${suffix64}"
         run_args="${run_args} --boot /data/art-test/core${image_suffix}${pic_image_suffix}${multi_image_suffix}.art"
     fi
     if [ "$relocate" = "yes" ]; then
diff --git a/tools/ahat/test-dump/Main.java b/tools/ahat/test-dump/Main.java
index d61a98da5..3936f296d 100644
--- a/tools/ahat/test-dump/Main.java
+++ b/tools/ahat/test-dump/Main.java
@@ -50,7 +50,8 @@ public class Main {
         bigArray[i] = (byte)((i*i) & 0xFF);
       }
 
-      NativeAllocationRegistry registry = new NativeAllocationRegistry(0x12345, 42);
+      NativeAllocationRegistry registry = new NativeAllocationRegistry(
+          Main.class.getClassLoader(), 0x12345, 42);
       registry.registerNativeAllocation(anObject, 0xABCDABCD);
     }
   }
diff --git a/tools/buildbot-build.sh b/tools/buildbot-build.sh
index 2eb52bcad..304c2a939 100755
--- a/tools/buildbot-build.sh
+++ b/tools/buildbot-build.sh
@@ -46,9 +46,14 @@ while true; do
 done
 
 if [[ $mode == "host" ]]; then
-  make_command="make $j_arg $showcommands build-art-host-tests $common_targets ${out_dir}/host/linux-x86/lib/libjavacoretests.so ${out_dir}/host/linux-x86/lib64/libjavacoretests.so"
+  make_command="make $j_arg $showcommands build-art-host-tests $common_targets"
+  make_command+=" ${out_dir}/host/linux-x86/lib/libjavacoretests.so "
+  make_command+=" ${out_dir}/host/linux-x86/lib64/libjavacoretests.so"
 elif [[ $mode == "target" ]]; then
-  make_command="make $j_arg $showcommands build-art-target-tests $common_targets libjavacrypto libjavacoretests linker toybox toolbox sh ${out_dir}/host/linux-x86/bin/adb libstdc++"
+  make_command="make $j_arg $showcommands build-art-target-tests $common_targets"
+  make_command+=" libjavacrypto libjavacoretests linker toybox toolbox sh"
+  make_command+=" ${out_dir}/host/linux-x86/bin/adb libstdc++ "
+  make_command+=" ${out_dir}/target/product/${TARGET_PRODUCT}/system/etc/public.libraries.txt"
 fi
 
 echo "Executing $make_command"
diff --git a/tools/dmtracedump/tracedump.cc b/tools/dmtracedump/tracedump.cc
index f70e2c220..3afee6fdc 100644
--- a/tools/dmtracedump/tracedump.cc
+++ b/tools/dmtracedump/tracedump.cc
@@ -512,10 +512,10 @@ int32_t compareUniqueExclusive(const void* a, const void* b) {
 void freeDataKeys(DataKeys* pKeys) {
   if (pKeys == nullptr) return;
 
-  free(pKeys->fileData);
-  free(pKeys->threads);
-  free(pKeys->methods);
-  free(pKeys);
+  delete[] pKeys->fileData;
+  delete[] pKeys->threads;
+  delete[] pKeys->methods;
+  delete pKeys;
 }
 
 /*
@@ -822,8 +822,8 @@ void sortMethodList(DataKeys* pKeys) {
 DataKeys* parseKeys(FILE* fp, int32_t verbose) {
   int64_t offset;
   DataKeys* pKeys = new DataKeys();
-  memset(pKeys, 0, sizeof(DataKeys));
   if (pKeys == nullptr) return nullptr;
+  memset(pKeys, 0, sizeof(DataKeys));
 
   /*
    * We load the entire file into memory.  We do this, rather than memory-
@@ -865,9 +865,13 @@ DataKeys* parseKeys(FILE* fp, int32_t verbose) {
     return nullptr;
   }
 
-  /* Reduce our allocation now that we know where the end of the key section is. */
-  pKeys->fileData = reinterpret_cast<char*>(realloc(pKeys->fileData, offset));
-  pKeys->fileLen = offset;
+  /*
+   * Although it is tempting to reduce our allocation now that we know where the
+   * end of the key section is, there is a pitfall. The method names and
+   * signatures in the method list contain pointers into the fileData area.
+   * Realloc or free will result in corruption.
+   */
+
   /* Leave fp pointing to the beginning of the data section. */
   fseek(fp, offset, SEEK_SET);
 
@@ -2607,7 +2611,7 @@ int32_t main(int32_t argc, char** argv) {
     if (gOptions.graphFileName != nullptr) {
       createInclusiveProfileGraphNew(dataKeys);
     }
-    free(methods);
+    delete[] methods;
   }
 
   freeDataKeys(dataKeys);
diff --git a/tools/libcore_failures.txt b/tools/libcore_failures.txt
index 38b6ea60f..bdcf86d85 100644
--- a/tools/libcore_failures.txt
+++ b/tools/libcore_failures.txt
@@ -243,51 +243,14 @@
           "org.apache.harmony.tests.java.util.prefs.FilePreferencesImplTest#testPutGet"]
 },
 {
-  description: "libnativehelper_compat_libc++ loading issue",
-  result: EXEC_FAILED,
-  modes: [device],
-  names: ["dalvik.system.JniTest#testGetSuperclass",
-          "dalvik.system.JniTest#testPassingBooleans",
-          "dalvik.system.JniTest#testPassingBytes",
-          "dalvik.system.JniTest#testPassingChars",
-          "dalvik.system.JniTest#testPassingClass",
-          "dalvik.system.JniTest#testPassingDoubles",
-          "dalvik.system.JniTest#testPassingFloats",
-          "dalvik.system.JniTest#testPassingInts",
-          "dalvik.system.JniTest#testPassingLongs",
-          "dalvik.system.JniTest#testPassingObjectReferences",
-          "dalvik.system.JniTest#testPassingShorts",
-          "dalvik.system.JniTest#testPassingThis",
-          "libcore.util.NativeAllocationRegistryTest#testBadSize",
-          "libcore.util.NativeAllocationRegistryTest#testEarlyFree",
-          "libcore.util.NativeAllocationRegistryTest#testNativeAllocationAllocatorAndNoSharedRegistry",
-          "libcore.util.NativeAllocationRegistryTest#testNativeAllocationAllocatorAndSharedRegistry",
-          "libcore.util.NativeAllocationRegistryTest#testNativeAllocationNoAllocatorAndNoSharedRegistry",
-          "libcore.util.NativeAllocationRegistryTest#testNativeAllocationNoAllocatorAndSharedRegistry",
-          "libcore.util.NativeAllocationRegistryTest#testNullArguments"]
-},
-{
-  description: "libnativehelper_compat_libc++.so not found by dlopen on ARM64",
+  description: "Only work with --mode=activity",
   result: EXEC_FAILED,
-  modes: [device],
-  bug: 28082914,
-  names: ["libcore.java.lang.ThreadTest#testContextClassLoaderIsInherited",
-          "libcore.java.lang.ThreadTest#testContextClassLoaderIsNotNull",
-          "libcore.java.lang.ThreadTest#testGetAllStackTracesIncludesAllGroups",
-          "libcore.java.lang.ThreadTest#testGetStackTrace",
-          "libcore.java.lang.ThreadTest#testJavaContextClassLoader",
-          "libcore.java.lang.ThreadTest#testLeakingStartedThreads",
-          "libcore.java.lang.ThreadTest#testLeakingUnstartedThreads",
-          "libcore.java.lang.ThreadTest#testNativeThreadNames",
-          "libcore.java.lang.ThreadTest#testThreadInterrupted",
-          "libcore.java.lang.ThreadTest#testThreadSleep",
-          "libcore.java.lang.ThreadTest#testThreadSleepIllegalArguments",
-          "libcore.java.lang.ThreadTest#testThreadWakeup"]
+  names: [ "libcore.java.io.FileTest#testJavaIoTmpdirMutable" ]
 },
 {
-  description: "Only work with --mode=activity",
+  description: "Investigate whether the test is making a wrong assumption with the newly enforced classpath.",
   result: EXEC_FAILED,
-  names: [ "libcore.java.io.FileTest#testJavaIoTmpdirMutable" ]
+  names: ["dalvik.system.DexClassLoaderTest#testDexThenPathClassLoader"]
 },
 {
   description: "Made for extending, shouldn't be run",
diff --git a/tools/public.libraries.buildbot.txt b/tools/public.libraries.buildbot.txt
new file mode 100644
index 000000000..4b01796a0
--- /dev/null
+++ b/tools/public.libraries.buildbot.txt
@@ -0,0 +1,8 @@
+libart.so
+libartd.so
+libbacktrace.so
+libc.so
+libc++.so
+libdl.so
+libm.so
+libnativehelper.so