Merge "bce: add support to narrow two MonotonicValueRange's at the same time."
diff --git a/.gitignore b/.gitignore
index 1cdfed9..3d1658d 100644
--- a/.gitignore
+++ b/.gitignore
@@ -2,3 +2,4 @@
 USE_PORTABLE_COMPILER
 SMALL_ART
 SEA_IR_ART
+JIT_ART
diff --git a/Android.mk b/Android.mk
index 76c3aa5..c740a0d 100644
--- a/Android.mk
+++ b/Android.mk
@@ -202,6 +202,11 @@
 test-art-host-interpreter: test-art-host-run-test-interpreter
 	$(hide) $(call ART_TEST_PREREQ_FINISHED,$@)
 
+# All host tests that run solely on the jit.
+.PHONY: test-art-host-jit
+test-art-host-jit: test-art-host-run-test-jit
+	$(hide) $(call ART_TEST_PREREQ_FINISHED,$@)
+
 # Primary host architecture variants:
 .PHONY: test-art-host$(ART_PHONY_TEST_HOST_SUFFIX)
 test-art-host$(ART_PHONY_TEST_HOST_SUFFIX): test-art-host-gtest$(ART_PHONY_TEST_HOST_SUFFIX) \
@@ -220,6 +225,10 @@
 test-art-host-interpreter$(ART_PHONY_TEST_HOST_SUFFIX): test-art-host-run-test-interpreter$(ART_PHONY_TEST_HOST_SUFFIX)
 	$(hide) $(call ART_TEST_PREREQ_FINISHED,$@)
 
+.PHONY: test-art-host-jit$(ART_PHONY_TEST_HOST_SUFFIX)
+test-art-host-jit$(ART_PHONY_TEST_HOST_SUFFIX): test-art-host-run-test-jit$(ART_PHONY_TEST_HOST_SUFFIX)
+	$(hide) $(call ART_TEST_PREREQ_FINISHED,$@)
+
 # Secondary host architecture variants:
 ifneq ($(HOST_PREFER_32_BIT),true)
 .PHONY: test-art-host$(2ND_ART_PHONY_TEST_HOST_SUFFIX)
@@ -238,6 +247,10 @@
 .PHONY: test-art-host-interpreter$(2ND_ART_PHONY_TEST_HOST_SUFFIX)
 test-art-host-interpreter$(2ND_ART_PHONY_TEST_HOST_SUFFIX): test-art-host-run-test-interpreter$(2ND_ART_PHONY_TEST_HOST_SUFFIX)
 	$(hide) $(call ART_TEST_PREREQ_FINISHED,$@)
+
+.PHONY: test-art-host-jit$(2ND_ART_PHONY_TEST_HOST_SUFFIX)
+test-art-host-jit$(2ND_ART_PHONY_TEST_HOST_SUFFIX): test-art-host-run-test-jit$(2ND_ART_PHONY_TEST_HOST_SUFFIX)
+	$(hide) $(call ART_TEST_PREREQ_FINISHED,$@)
 endif
 
 # Valgrind. Currently only 32b gtests.
@@ -268,6 +281,11 @@
 test-art-target-interpreter: test-art-target-run-test-interpreter
 	$(hide) $(call ART_TEST_PREREQ_FINISHED,$@)
 
+# All target tests that run solely on the jit.
+.PHONY: test-art-target-jit
+test-art-target-jit: test-art-target-run-test-jit
+	$(hide) $(call ART_TEST_PREREQ_FINISHED,$@)
+
 # Primary target architecture variants:
 .PHONY: test-art-target$(ART_PHONY_TEST_TARGET_SUFFIX)
 test-art-target$(ART_PHONY_TEST_TARGET_SUFFIX): test-art-target-gtest$(ART_PHONY_TEST_TARGET_SUFFIX) \
@@ -286,6 +304,10 @@
 test-art-target-interpreter$(ART_PHONY_TEST_TARGET_SUFFIX): test-art-target-run-test-interpreter$(ART_PHONY_TEST_TARGET_SUFFIX)
 	$(hide) $(call ART_TEST_PREREQ_FINISHED,$@)
 
+.PHONY: test-art-target-jit$(ART_PHONY_TEST_TARGET_SUFFIX)
+test-art-target-jit$(ART_PHONY_TEST_TARGET_SUFFIX): test-art-target-run-test-jit$(ART_PHONY_TEST_TARGET_SUFFIX)
+	$(hide) $(call ART_TEST_PREREQ_FINISHED,$@)
+
 # Secondary target architecture variants:
 ifdef TARGET_2ND_ARCH
 .PHONY: test-art-target$(2ND_ART_PHONY_TEST_TARGET_SUFFIX)
@@ -304,6 +326,10 @@
 .PHONY: test-art-target-interpreter$(2ND_ART_PHONY_TEST_TARGET_SUFFIX)
 test-art-target-interpreter$(2ND_ART_PHONY_TEST_TARGET_SUFFIX): test-art-target-run-test-interpreter$(2ND_ART_PHONY_TEST_TARGET_SUFFIX)
 	$(hide) $(call ART_TEST_PREREQ_FINISHED,$@)
+
+.PHONY: test-art-target-jit$(2ND_ART_PHONY_TEST_TARGET_SUFFIX)
+test-art-target-jit$(2ND_ART_PHONY_TEST_TARGET_SUFFIX): test-art-target-run-test-jit$(2ND_ART_PHONY_TEST_TARGET_SUFFIX)
+	$(hide) $(call ART_TEST_PREREQ_FINISHED,$@)
 endif
 
 endif  # art_test_bother
diff --git a/build/Android.common_build.mk b/build/Android.common_build.mk
index 3000cdf..6a83e72 100644
--- a/build/Android.common_build.mk
+++ b/build/Android.common_build.mk
@@ -48,6 +48,18 @@
 endif
 
 #
+# Used to enable JIT
+#
+ART_JIT := false
+ifneq ($(wildcard art/JIT_ART),)
+$(info Enabling ART_JIT because of existence of art/JIT_ART)
+ART_JIT := true
+endif
+ifeq ($(WITH_ART_JIT), true)
+ART_JIT := true
+endif
+
+#
 # Used to enable smart mode
 #
 ART_SMALL_MODE := false
@@ -62,8 +74,8 @@
 #
 # Used to change the default GC. Valid values are CMS, SS, GSS. The default is CMS.
 #
-art_default_gc_type ?= CMS
-art_default_gc_type_cflags := -DART_DEFAULT_GC_TYPE_IS_$(art_default_gc_type)
+ART_DEFAULT_GC_TYPE ?= CMS
+art_default_gc_type_cflags := -DART_DEFAULT_GC_TYPE_IS_$(ART_DEFAULT_GC_TYPE)
 
 ART_HOST_CFLAGS :=
 ART_TARGET_CFLAGS :=
@@ -191,6 +203,7 @@
   -Wunreachable-code \
   -Wredundant-decls \
   -Wshadow \
+  -Wunused \
   -fvisibility=protected \
   $(art_default_gc_type_cflags)
 
diff --git a/build/Android.common_test.mk b/build/Android.common_test.mk
index da50d53..547e92e 100644
--- a/build/Android.common_test.mk
+++ b/build/Android.common_test.mk
@@ -46,6 +46,9 @@
 # Do you want interpreter tests run?
 ART_TEST_INTERPRETER ?= $(ART_TEST_FULL)
 
+# Do you want JIT tests run?
+ART_TEST_JIT ?= $(ART_TEST_FULL)
+
 # Do you want optimizing compiler tests run?
 ART_TEST_OPTIMIZING ?= $(ART_TEST_FULL)
 
diff --git a/build/Android.gtest.mk b/build/Android.gtest.mk
index c576d1b..7ab4d64 100644
--- a/build/Android.gtest.mk
+++ b/build/Android.gtest.mk
@@ -116,6 +116,7 @@
   runtime/entrypoints_order_test.cc \
   runtime/exception_test.cc \
   runtime/gc/accounting/card_table_test.cc \
+  runtime/gc/accounting/mod_union_table_test.cc \
   runtime/gc/accounting/space_bitmap_test.cc \
   runtime/gc/heap_test.cc \
   runtime/gc/reference_queue_test.cc \
diff --git a/build/Android.oat.mk b/build/Android.oat.mk
index 8d49565..4d2fa41 100644
--- a/build/Android.oat.mk
+++ b/build/Android.oat.mk
@@ -31,7 +31,7 @@
 endif
 
 # Use dex2oat debug version for better error reporting
-# $(1): compiler - default, optimizing or interpreter.
+# $(1): compiler - default, optimizing, jit or interpreter.
 # $(2): pic/no-pic
 # $(3): 2ND_ or undefined, 2ND_ for 32-bit host builds.
 # $(4): wrapper, e.g., valgrind.
@@ -67,9 +67,9 @@
   ifeq ($(1),default)
     # Default has no infix, no compile options.
   endif
-  ifneq ($(filter-out default interpreter optimizing,$(1)),)
+  ifneq ($(filter-out default interpreter jit optimizing,$(1)),)
     #Technically this test is not precise, but hopefully good enough.
-    $$(error found $(1) expected default, interpreter or optimizing)
+    $$(error found $(1) expected default, interpreter, jit or optimizing)
   endif
 
   ifeq ($(2),pic)
@@ -127,7 +127,7 @@
   core_pic_infix :=
 endef  # create-core-oat-host-rules
 
-# $(1): compiler - default, optimizing or interpreter.
+# $(1): compiler - default, optimizing, jit or interpreter.
 # $(2): wrapper.
 # $(3): dex2oat suffix.
 define create-core-oat-host-rule-combination
@@ -181,9 +181,9 @@
   ifeq ($(1),default)
     # Default has no infix, no compile options.
   endif
-  ifneq ($(filter-out default interpreter optimizing,$(1)),)
+  ifneq ($(filter-out default interpreter jit optimizing,$(1)),)
     # Technically this test is not precise, but hopefully good enough.
-    $$(error found $(1) expected default, interpreter or optimizing)
+    $$(error found $(1) expected default, interpreter, jit or optimizing)
   endif
 
   ifeq ($(2),pic)
@@ -245,7 +245,7 @@
   core_pic_infix :=
 endef  # create-core-oat-target-rules
 
-# $(1): compiler - default, optimizing or interpreter.
+# $(1): compiler - default, optimizing, jit or interpreter.
 # $(2): wrapper.
 # $(3): dex2oat suffix.
 define create-core-oat-target-rule-combination
diff --git a/cmdline/cmdline_parse_result.h b/cmdline/cmdline_parse_result.h
index d6ac341..717642f 100644
--- a/cmdline/cmdline_parse_result.h
+++ b/cmdline/cmdline_parse_result.h
@@ -117,9 +117,9 @@
   }
 
   // Make sure copying is allowed
-  CmdlineParseResult(const CmdlineParseResult& other) = default;
+  CmdlineParseResult(const CmdlineParseResult&) = default;
   // Make sure moving is cheap
-  CmdlineParseResult(CmdlineParseResult&& other) = default;
+  CmdlineParseResult(CmdlineParseResult&&) = default;
 
  private:
   explicit CmdlineParseResult(const T& value)
diff --git a/cmdline/cmdline_parser.h b/cmdline/cmdline_parser.h
index a555356..e4af4f9 100644
--- a/cmdline/cmdline_parser.h
+++ b/cmdline/cmdline_parser.h
@@ -490,9 +490,9 @@
   }
 
   // Ensure we have a default move constructor.
-  CmdlineParser(CmdlineParser&& other) = default;
+  CmdlineParser(CmdlineParser&&) = default;
   // Ensure we have a default move assignment operator.
-  CmdlineParser& operator=(CmdlineParser&& other) = default;
+  CmdlineParser& operator=(CmdlineParser&&) = default;
 
  private:
   friend struct Builder;
diff --git a/cmdline/cmdline_parser_test.cc b/cmdline/cmdline_parser_test.cc
index b740b41..18d8c7a 100644
--- a/cmdline/cmdline_parser_test.cc
+++ b/cmdline/cmdline_parser_test.cc
@@ -411,6 +411,26 @@
 }  // TEST_F
 
 /*
+* -Xjit, -Xnojit, -Xjitcodecachesize, Xjitcompilethreshold
+*/
+TEST_F(CmdlineParserTest, TestJitOptions) {
+ /*
+  * Test successes
+  */
+  {
+    EXPECT_SINGLE_PARSE_VALUE(true, "-Xjit", M::UseJIT);
+    EXPECT_SINGLE_PARSE_VALUE(false, "-Xnojit", M::UseJIT);
+  }
+  {
+    EXPECT_SINGLE_PARSE_VALUE(MemoryKiB(16 * KB), "-Xjitcodecachesize:16K", M::JITCodeCacheCapacity);
+    EXPECT_SINGLE_PARSE_VALUE(MemoryKiB(16 * MB), "-Xjitcodecachesize:16M", M::JITCodeCacheCapacity);
+  }
+  {
+    EXPECT_SINGLE_PARSE_VALUE(12345u, "-Xjitthreshold:12345", M::JITCompileThreshold);
+  }
+}  // TEST_F
+
+/*
 * -X-profile-*
 */
 TEST_F(CmdlineParserTest, TestProfilerOptions) {
@@ -495,9 +515,8 @@
       "-dsa", "-enablesystemassertions", "-disablesystemassertions", "-Xrs", "-Xint:abdef",
       "-Xdexopt:foobar", "-Xnoquithandler", "-Xjnigreflimit:ixnay", "-Xgenregmap", "-Xnogenregmap",
       "-Xverifyopt:never", "-Xcheckdexsum", "-Xincludeselectedop", "-Xjitop:noop",
-      "-Xincludeselectedmethod", "-Xjitthreshold:123", "-Xjitcodecachesize:12345",
-      "-Xjitblocking", "-Xjitmethod:_", "-Xjitclass:nosuchluck", "-Xjitoffset:none",
-      "-Xjitconfig:yes", "-Xjitcheckcg", "-Xjitverbose", "-Xjitprofile",
+      "-Xincludeselectedmethod", "-Xjitblocking", "-Xjitmethod:_", "-Xjitclass:nosuchluck",
+      "-Xjitoffset:none", "-Xjitconfig:yes", "-Xjitcheckcg", "-Xjitverbose", "-Xjitprofile",
       "-Xjitdisableopt", "-Xjitsuspendpoll", "-XX:mainThreadStackSize=1337"
   };
 
diff --git a/cmdline/cmdline_result.h b/cmdline/cmdline_result.h
index bf3a85d..963dfc1 100644
--- a/cmdline/cmdline_result.h
+++ b/cmdline/cmdline_result.h
@@ -65,9 +65,9 @@
     }
 
     // Make sure copying exists
-    CmdlineResult(const CmdlineResult& other) = default;
+    CmdlineResult(const CmdlineResult&) = default;
     // Make sure moving is cheap
-    CmdlineResult(CmdlineResult&& other) = default;
+    CmdlineResult(CmdlineResult&&) = default;
 
   private:
     const Status status_;
diff --git a/cmdline/cmdline_types.h b/cmdline/cmdline_types.h
index 180baec..de99278 100644
--- a/cmdline/cmdline_types.h
+++ b/cmdline/cmdline_types.h
@@ -337,8 +337,8 @@
 
   // Default constructors/copy-constructors.
   MillisecondsToNanoseconds() : nanoseconds_(0ul) {}
-  MillisecondsToNanoseconds(const MillisecondsToNanoseconds& rhs) = default;
-  MillisecondsToNanoseconds(MillisecondsToNanoseconds&& rhs) = default;
+  MillisecondsToNanoseconds(const MillisecondsToNanoseconds&) = default;
+  MillisecondsToNanoseconds(MillisecondsToNanoseconds&&) = default;
 
  private:
   uint64_t nanoseconds_;
@@ -421,8 +421,8 @@
   }
 
   ParseStringList() = default;
-  ParseStringList(const ParseStringList& rhs) = default;
-  ParseStringList(ParseStringList&& rhs) = default;
+  ParseStringList(const ParseStringList&) = default;
+  ParseStringList(ParseStringList&&) = default;
 
  private:
   std::vector<std::string> list_;
@@ -585,6 +585,8 @@
         log_verbosity.heap = true;
       } else if (verbose_options[j] == "jdwp") {
         log_verbosity.jdwp = true;
+      } else if (verbose_options[j] == "jit") {
+        log_verbosity.jit = true;
       } else if (verbose_options[j] == "jni") {
         log_verbosity.jni = true;
       } else if (verbose_options[j] == "monitor") {
@@ -651,8 +653,8 @@
     max_stack_depth_(0) {
   }
 
-  TestProfilerOptions(const TestProfilerOptions& other) = default;
-  TestProfilerOptions(TestProfilerOptions&& other) = default;
+  TestProfilerOptions(const TestProfilerOptions&) = default;
+  TestProfilerOptions(TestProfilerOptions&&) = default;
 };
 
 static inline std::ostream& operator<<(std::ostream& stream, const TestProfilerOptions& options) {
diff --git a/cmdline/token_range.h b/cmdline/token_range.h
index 50c54fe..5b54384 100644
--- a/cmdline/token_range.h
+++ b/cmdline/token_range.h
@@ -90,10 +90,10 @@
   }
 
   // Non-copying copy constructor.
-  TokenRange(const TokenRange& other) = default;
+  TokenRange(const TokenRange&) = default;
 
   // Non-copying move constructor.
-  TokenRange(TokenRange&& other) = default;
+  TokenRange(TokenRange&&) = default;
 
   // Non-copying constructor. Retains reference to an existing list of tokens, with offset.
   explicit TokenRange(std::shared_ptr<TokenList> token_list)
diff --git a/compiler/Android.mk b/compiler/Android.mk
index beb34dc..86a27c1 100644
--- a/compiler/Android.mk
+++ b/compiler/Android.mk
@@ -79,6 +79,7 @@
 	driver/compiler_driver.cc \
 	driver/compiler_options.cc \
 	driver/dex_compilation_unit.cc \
+	jit/jit_compiler.cc \
 	jni/quick/arm/calling_convention_arm.cc \
 	jni/quick/arm64/calling_convention_arm64.cc \
 	jni/quick/mips/calling_convention_mips.cc \
@@ -161,8 +162,7 @@
   driver/compiler_options.h \
   image_writer.h \
   optimizing/locations.h \
-  utils/arm/constants_arm.h \
-  utils/dex_instruction_utils.h
+  utils/arm/constants_arm.h
 
 # $(1): target or host
 # $(2): ndebug or debug
diff --git a/compiler/common_compiler_test.cc b/compiler/common_compiler_test.cc
index 1cd78f8..09be437 100644
--- a/compiler/common_compiler_test.cc
+++ b/compiler/common_compiler_test.cc
@@ -52,19 +52,19 @@
     const SwapVector<uint8_t>* code = compiled_method->GetQuickCode();
     uint32_t code_size = code->size();
     CHECK_NE(0u, code_size);
-    const SwapVector<uint8_t>& vmap_table = compiled_method->GetVmapTable();
-    uint32_t vmap_table_offset = vmap_table.empty() ? 0u
-        : sizeof(OatQuickMethodHeader) + vmap_table.size();
+    const SwapVector<uint8_t>* vmap_table = compiled_method->GetVmapTable();
+    uint32_t vmap_table_offset = vmap_table->empty() ? 0u
+        : sizeof(OatQuickMethodHeader) + vmap_table->size();
     const SwapVector<uint8_t>* mapping_table = compiled_method->GetMappingTable();
     bool mapping_table_used = mapping_table != nullptr && !mapping_table->empty();
     size_t mapping_table_size = mapping_table_used ? mapping_table->size() : 0U;
     uint32_t mapping_table_offset = !mapping_table_used ? 0u
-        : sizeof(OatQuickMethodHeader) + vmap_table.size() + mapping_table_size;
+        : sizeof(OatQuickMethodHeader) + vmap_table->size() + mapping_table_size;
     const SwapVector<uint8_t>* gc_map = compiled_method->GetGcMap();
     bool gc_map_used = gc_map != nullptr && !gc_map->empty();
     size_t gc_map_size = gc_map_used ? gc_map->size() : 0U;
     uint32_t gc_map_offset = !gc_map_used ? 0u
-        : sizeof(OatQuickMethodHeader) + vmap_table.size() + mapping_table_size + gc_map_size;
+        : sizeof(OatQuickMethodHeader) + vmap_table->size() + mapping_table_size + gc_map_size;
     OatQuickMethodHeader method_header(mapping_table_offset, vmap_table_offset, gc_map_offset,
                                        compiled_method->GetFrameSizeInBytes(),
                                        compiled_method->GetCoreSpillMask(),
@@ -72,14 +72,14 @@
 
     header_code_and_maps_chunks_.push_back(std::vector<uint8_t>());
     std::vector<uint8_t>* chunk = &header_code_and_maps_chunks_.back();
-    size_t size = sizeof(method_header) + code_size + vmap_table.size() + mapping_table_size +
+    size_t size = sizeof(method_header) + code_size + vmap_table->size() + mapping_table_size +
         gc_map_size;
     size_t code_offset = compiled_method->AlignCode(size - code_size);
     size_t padding = code_offset - (size - code_size);
     chunk->reserve(padding + size);
     chunk->resize(sizeof(method_header));
     memcpy(&(*chunk)[0], &method_header, sizeof(method_header));
-    chunk->insert(chunk->begin(), vmap_table.begin(), vmap_table.end());
+    chunk->insert(chunk->begin(), vmap_table->begin(), vmap_table->end());
     if (mapping_table_used) {
       chunk->insert(chunk->begin(), mapping_table->begin(), mapping_table->end());
     }
@@ -212,7 +212,7 @@
   CHECK(method != nullptr);
   TimingLogger timings("CommonTest::CompileMethod", false, false);
   TimingLogger::ScopedTiming t(__FUNCTION__, &timings);
-  compiler_driver_->CompileOne(method, &timings);
+  compiler_driver_->CompileOne(Thread::Current(), method, &timings);
   TimingLogger::ScopedTiming t2("MakeExecutable", &timings);
   MakeExecutable(method);
 }
@@ -253,6 +253,7 @@
                                                 (size_t)100 * 1024 * 1024,  // 100MB
                                                 PROT_NONE,
                                                 false /* no need for 4gb flag with fixed mmap*/,
+                                                false /* not reusing existing reservation */,
                                                 &error_msg));
   CHECK(image_reservation_.get() != nullptr) << error_msg;
 }
diff --git a/compiler/compiled_method.cc b/compiler/compiled_method.cc
index 22be28c..1849e7e 100644
--- a/compiler/compiled_method.cc
+++ b/compiler/compiled_method.cc
@@ -20,16 +20,29 @@
 namespace art {
 
 CompiledCode::CompiledCode(CompilerDriver* compiler_driver, InstructionSet instruction_set,
-                           const ArrayRef<const uint8_t>& quick_code)
+                           const ArrayRef<const uint8_t>& quick_code, bool owns_code_array)
     : compiler_driver_(compiler_driver), instruction_set_(instruction_set),
-      quick_code_(nullptr) {
+      owns_code_array_(owns_code_array), quick_code_(nullptr) {
   SetCode(&quick_code);
 }
 
 void CompiledCode::SetCode(const ArrayRef<const uint8_t>* quick_code) {
   if (quick_code != nullptr) {
     CHECK(!quick_code->empty());
-    quick_code_ = compiler_driver_->DeduplicateCode(*quick_code);
+    if (owns_code_array_) {
+      // If we are supposed to own the code, don't deduplicate it.
+      CHECK(quick_code_ == nullptr);
+      quick_code_ = new SwapVector<uint8_t>(quick_code->begin(), quick_code->end(),
+                                            compiler_driver_->GetSwapSpaceAllocator());
+    } else {
+      quick_code_ = compiler_driver_->DeduplicateCode(*quick_code);
+    }
+  }
+}
+
+CompiledCode::~CompiledCode() {
+  if (owns_code_array_) {
+    delete quick_code_;
   }
 }
 
@@ -46,11 +59,11 @@
   return (rhs.quick_code_ == nullptr);
 }
 
-uint32_t CompiledCode::AlignCode(uint32_t offset) const {
+size_t CompiledCode::AlignCode(size_t offset) const {
   return AlignCode(offset, instruction_set_);
 }
 
-uint32_t CompiledCode::AlignCode(uint32_t offset, InstructionSet instruction_set) {
+size_t CompiledCode::AlignCode(size_t offset, InstructionSet instruction_set) {
   return RoundUp(offset, GetInstructionSetAlignment(instruction_set));
 }
 
@@ -120,17 +133,39 @@
                                const ArrayRef<const uint8_t>& native_gc_map,
                                const ArrayRef<const uint8_t>& cfi_info,
                                const ArrayRef<LinkerPatch>& patches)
-    : CompiledCode(driver, instruction_set, quick_code), frame_size_in_bytes_(frame_size_in_bytes),
-      core_spill_mask_(core_spill_mask), fp_spill_mask_(fp_spill_mask),
-      src_mapping_table_(src_mapping_table == nullptr ?
-          driver->DeduplicateSrcMappingTable(ArrayRef<SrcMapElem>()) :
-          driver->DeduplicateSrcMappingTable(ArrayRef<SrcMapElem>(src_mapping_table->Arrange()))),
-      mapping_table_(mapping_table.data() == nullptr ?
-          nullptr : driver->DeduplicateMappingTable(mapping_table)),
-      vmap_table_(driver->DeduplicateVMapTable(vmap_table)),
-      gc_map_(native_gc_map.data() == nullptr ? nullptr : driver->DeduplicateGCMap(native_gc_map)),
-      cfi_info_(cfi_info.data() == nullptr ? nullptr : driver->DeduplicateCFIInfo(cfi_info)),
+    : CompiledCode(driver, instruction_set, quick_code, !driver->DedupeEnabled()),
+      owns_arrays_(!driver->DedupeEnabled()),
+      frame_size_in_bytes_(frame_size_in_bytes), core_spill_mask_(core_spill_mask),
+      fp_spill_mask_(fp_spill_mask),
       patches_(patches.begin(), patches.end(), driver->GetSwapSpaceAllocator()) {
+  if (owns_arrays_) {
+    if (src_mapping_table == nullptr) {
+      src_mapping_table_ = new SwapSrcMap(driver->GetSwapSpaceAllocator());
+    } else {
+      src_mapping_table->Arrange();
+      src_mapping_table_ = new SwapSrcMap(src_mapping_table->begin(), src_mapping_table->end(),
+                                          driver->GetSwapSpaceAllocator());
+    }
+    mapping_table_ = mapping_table.empty() ?
+        nullptr : new SwapVector<uint8_t>(mapping_table.begin(), mapping_table.end(),
+                                          driver->GetSwapSpaceAllocator());
+    vmap_table_ = new SwapVector<uint8_t>(vmap_table.begin(), vmap_table.end(),
+                                          driver->GetSwapSpaceAllocator());
+    gc_map_ = native_gc_map.empty() ? nullptr :
+        new SwapVector<uint8_t>(native_gc_map.begin(), native_gc_map.end(),
+                                driver->GetSwapSpaceAllocator());
+    cfi_info_ = cfi_info.empty() ? nullptr :
+        new SwapVector<uint8_t>(cfi_info.begin(), cfi_info.end(), driver->GetSwapSpaceAllocator());
+  } else {
+    src_mapping_table_ = src_mapping_table == nullptr ?
+        driver->DeduplicateSrcMappingTable(ArrayRef<SrcMapElem>()) :
+        driver->DeduplicateSrcMappingTable(ArrayRef<SrcMapElem>(src_mapping_table->Arrange()));
+    mapping_table_ = mapping_table.empty() ?
+        nullptr : driver->DeduplicateMappingTable(mapping_table);
+    vmap_table_ = driver->DeduplicateVMapTable(vmap_table);
+    gc_map_ = native_gc_map.empty() ? nullptr : driver->DeduplicateGCMap(native_gc_map);
+    cfi_info_ = cfi_info.empty() ? nullptr : driver->DeduplicateCFIInfo(cfi_info);
+  }
 }
 
 CompiledMethod* CompiledMethod::SwapAllocCompiledMethod(
@@ -194,4 +229,14 @@
   alloc.deallocate(m, 1);
 }
 
+CompiledMethod::~CompiledMethod() {
+  if (owns_arrays_) {
+    delete src_mapping_table_;
+    delete mapping_table_;
+    delete vmap_table_;
+    delete gc_map_;
+    delete cfi_info_;
+  }
+}
+
 }  // namespace art
diff --git a/compiler/compiled_method.h b/compiler/compiled_method.h
index 6013507..d6a07f6 100644
--- a/compiler/compiled_method.h
+++ b/compiler/compiled_method.h
@@ -27,10 +27,6 @@
 #include "utils/array_ref.h"
 #include "utils/swap_space.h"
 
-namespace llvm {
-  class Function;
-}  // namespace llvm
-
 namespace art {
 
 class CompilerDriver;
@@ -39,7 +35,9 @@
  public:
   // For Quick to supply an code blob
   CompiledCode(CompilerDriver* compiler_driver, InstructionSet instruction_set,
-               const ArrayRef<const uint8_t>& quick_code);
+               const ArrayRef<const uint8_t>& quick_code, bool owns_code_array);
+
+  virtual ~CompiledCode();
 
   InstructionSet GetInstructionSet() const {
     return instruction_set_;
@@ -56,8 +54,8 @@
   // To align an offset from a page-aligned value to make it suitable
   // for code storage. For example on ARM, to ensure that PC relative
   // valu computations work out as expected.
-  uint32_t AlignCode(uint32_t offset) const;
-  static uint32_t AlignCode(uint32_t offset, InstructionSet instruction_set);
+  size_t AlignCode(size_t offset) const;
+  static size_t AlignCode(size_t offset, InstructionSet instruction_set);
 
   // returns the difference between the code address and a usable PC.
   // mainly to cope with kThumb2 where the lower bit must be set.
@@ -78,6 +76,9 @@
 
   const InstructionSet instruction_set_;
 
+  // If we own the code array (means that we free in destructor).
+  const bool owns_code_array_;
+
   // Used to store the PIC code for Quick.
   SwapVector<uint8_t>* quick_code_;
 
@@ -122,6 +123,7 @@
   using std::vector<SrcMapElem, Allocator>::size;
 
   explicit SrcMap() {}
+  explicit SrcMap(const Allocator& alloc) : std::vector<SrcMapElem, Allocator>(alloc) {}
 
   template <class InputIt>
   SrcMap(InputIt first, InputIt last, const Allocator& alloc)
@@ -291,7 +293,7 @@
                  const ArrayRef<const uint8_t>& cfi_info,
                  const ArrayRef<LinkerPatch>& patches = ArrayRef<LinkerPatch>());
 
-  ~CompiledMethod() {}
+  virtual ~CompiledMethod();
 
   static CompiledMethod* SwapAllocCompiledMethod(
       CompilerDriver* driver,
@@ -347,9 +349,9 @@
     return mapping_table_;
   }
 
-  const SwapVector<uint8_t>& GetVmapTable() const {
+  const SwapVector<uint8_t>* GetVmapTable() const {
     DCHECK(vmap_table_ != nullptr);
-    return *vmap_table_;
+    return vmap_table_;
   }
 
   SwapVector<uint8_t> const* GetGcMap() const {
@@ -365,6 +367,8 @@
   }
 
  private:
+  // Whether or not the arrays are owned by the compiled method or dedupe sets.
+  const bool owns_arrays_;
   // For quick code, the size of the activation used by the code.
   const size_t frame_size_in_bytes_;
   // For quick code, a bit mask describing spilled GPR callee-save registers.
diff --git a/compiler/dex/global_value_numbering_test.cc b/compiler/dex/global_value_numbering_test.cc
index 54e34ea..b91c3ca 100644
--- a/compiler/dex/global_value_numbering_test.cc
+++ b/compiler/dex/global_value_numbering_test.cc
@@ -142,7 +142,7 @@
     cu_.mir_graph->ifield_lowering_infos_.reserve(count);
     for (size_t i = 0u; i != count; ++i) {
       const IFieldDef* def = &defs[i];
-      MirIFieldLoweringInfo field_info(def->field_idx, def->type);
+      MirIFieldLoweringInfo field_info(def->field_idx, def->type, false);
       if (def->declaring_dex_file != 0u) {
         field_info.declaring_dex_file_ = reinterpret_cast<const DexFile*>(def->declaring_dex_file);
         field_info.declaring_field_idx_ = def->declaring_field_idx;
diff --git a/compiler/dex/gvn_dead_code_elimination_test.cc b/compiler/dex/gvn_dead_code_elimination_test.cc
index 954e9f1..4d2b8b3 100644
--- a/compiler/dex/gvn_dead_code_elimination_test.cc
+++ b/compiler/dex/gvn_dead_code_elimination_test.cc
@@ -143,7 +143,7 @@
     cu_.mir_graph->ifield_lowering_infos_.reserve(count);
     for (size_t i = 0u; i != count; ++i) {
       const IFieldDef* def = &defs[i];
-      MirIFieldLoweringInfo field_info(def->field_idx, def->type);
+      MirIFieldLoweringInfo field_info(def->field_idx, def->type, false);
       if (def->declaring_dex_file != 0u) {
         field_info.declaring_dex_file_ = reinterpret_cast<const DexFile*>(def->declaring_dex_file);
         field_info.declaring_field_idx_ = def->declaring_field_idx;
diff --git a/compiler/dex/local_value_numbering.h b/compiler/dex/local_value_numbering.h
index 97ea05a..379c952 100644
--- a/compiler/dex/local_value_numbering.h
+++ b/compiler/dex/local_value_numbering.h
@@ -21,8 +21,8 @@
 
 #include "base/arena_object.h"
 #include "base/logging.h"
+#include "dex_instruction_utils.h"
 #include "global_value_numbering.h"
-#include "utils/dex_instruction_utils.h"
 
 namespace art {
 
diff --git a/compiler/dex/local_value_numbering_test.cc b/compiler/dex/local_value_numbering_test.cc
index d1c3a6b..566527a 100644
--- a/compiler/dex/local_value_numbering_test.cc
+++ b/compiler/dex/local_value_numbering_test.cc
@@ -96,7 +96,7 @@
     cu_.mir_graph->ifield_lowering_infos_.reserve(count);
     for (size_t i = 0u; i != count; ++i) {
       const IFieldDef* def = &defs[i];
-      MirIFieldLoweringInfo field_info(def->field_idx, def->type);
+      MirIFieldLoweringInfo field_info(def->field_idx, def->type, false);
       if (def->declaring_dex_file != 0u) {
         field_info.declaring_dex_file_ = reinterpret_cast<const DexFile*>(def->declaring_dex_file);
         field_info.declaring_field_idx_ = def->declaring_field_idx;
diff --git a/compiler/dex/mir_analysis.cc b/compiler/dex/mir_analysis.cc
index 31dbc60..a89b250 100644
--- a/compiler/dex/mir_analysis.cc
+++ b/compiler/dex/mir_analysis.cc
@@ -416,8 +416,8 @@
   // 72 INVOKE_INTERFACE {vD, vE, vF, vG, vA}
   kAnInvoke | kAnHeavyWeight,
 
-  // 73 UNUSED_73
-  kAnNone,
+  // 73 RETURN_VOID_BARRIER
+  kAnBranch,
 
   // 74 INVOKE_VIRTUAL_RANGE {vCCCC .. vNNNN}
   kAnInvoke | kAnHeavyWeight,
@@ -752,88 +752,88 @@
   // E2 USHR_INT_LIT8 vAA, vBB, #+CC
   kAnMath | kAnInt,
 
-  // E3 IGET_VOLATILE
+  // E3 IGET_QUICK
   kAnNone,
 
-  // E4 IPUT_VOLATILE
+  // E4 IGET_WIDE_QUICK
   kAnNone,
 
-  // E5 SGET_VOLATILE
+  // E5 IGET_OBJECT_QUICK
   kAnNone,
 
-  // E6 SPUT_VOLATILE
+  // E6 IPUT_QUICK
   kAnNone,
 
-  // E7 IGET_OBJECT_VOLATILE
+  // E7 IPUT_WIDE_QUICK
   kAnNone,
 
-  // E8 IGET_WIDE_VOLATILE
+  // E8 IPUT_OBJECT_QUICK
   kAnNone,
 
-  // E9 IPUT_WIDE_VOLATILE
-  kAnNone,
-
-  // EA SGET_WIDE_VOLATILE
-  kAnNone,
-
-  // EB SPUT_WIDE_VOLATILE
-  kAnNone,
-
-  // EC BREAKPOINT
-  kAnNone,
-
-  // ED THROW_VERIFICATION_ERROR
-  kAnHeavyWeight | kAnBranch,
-
-  // EE EXECUTE_INLINE
-  kAnNone,
-
-  // EF EXECUTE_INLINE_RANGE
-  kAnNone,
-
-  // F0 INVOKE_OBJECT_INIT_RANGE
+  // E9 INVOKE_VIRTUAL_QUICK
   kAnInvoke | kAnHeavyWeight,
 
-  // F1 RETURN_VOID_BARRIER
-  kAnBranch,
-
-  // F2 IGET_QUICK
-  kAnNone,
-
-  // F3 IGET_WIDE_QUICK
-  kAnNone,
-
-  // F4 IGET_OBJECT_QUICK
-  kAnNone,
-
-  // F5 IPUT_QUICK
-  kAnNone,
-
-  // F6 IPUT_WIDE_QUICK
-  kAnNone,
-
-  // F7 IPUT_OBJECT_QUICK
-  kAnNone,
-
-  // F8 INVOKE_VIRTUAL_QUICK
+  // EA INVOKE_VIRTUAL_RANGE_QUICK
   kAnInvoke | kAnHeavyWeight,
 
-  // F9 INVOKE_VIRTUAL_QUICK_RANGE
-  kAnInvoke | kAnHeavyWeight,
-
-  // FA INVOKE_SUPER_QUICK
-  kAnInvoke | kAnHeavyWeight,
-
-  // FB INVOKE_SUPER_QUICK_RANGE
-  kAnInvoke | kAnHeavyWeight,
-
-  // FC IPUT_OBJECT_VOLATILE
+  // EB IPUT_BOOLEAN_QUICK
   kAnNone,
 
-  // FD SGET_OBJECT_VOLATILE
+  // EC IPUT_BYTE_QUICK
   kAnNone,
 
-  // FE SPUT_OBJECT_VOLATILE
+  // ED IPUT_CHAR_QUICK
+  kAnNone,
+
+  // EE IPUT_SHORT_QUICK
+  kAnNone,
+
+  // EF IGET_BOOLEAN_QUICK
+  kAnNone,
+
+  // F0 IGET_BYTE_QUICK
+  kAnNone,
+
+  // F1 IGET_CHAR_QUICK
+  kAnNone,
+
+  // F2 IGET_SHORT_QUICK
+  kAnNone,
+
+  // F3 UNUSED_F3
+  kAnNone,
+
+  // F4 UNUSED_F4
+  kAnNone,
+
+  // F5 UNUSED_F5
+  kAnNone,
+
+  // F6 UNUSED_F6
+  kAnNone,
+
+  // F7 UNUSED_F7
+  kAnNone,
+
+  // F8 UNUSED_F8
+  kAnNone,
+
+  // F9 UNUSED_F9
+  kAnNone,
+
+  // FA UNUSED_FA
+  kAnNone,
+
+  // FB UNUSED_FB
+  kAnNone,
+
+  // FC UNUSED_FC
+  kAnNone,
+
+  // FD UNUSED_FD
+  kAnNone,
+
+  // FE UNUSED_FE
   kAnNone,
 
   // FF UNUSED_FF
@@ -1203,12 +1203,13 @@
 }
 
 void MIRGraph::DoCacheFieldLoweringInfo() {
+  static constexpr uint32_t kFieldIndexFlagQuickened = 0x80000000;
   // All IGET/IPUT/SGET/SPUT instructions take 2 code units and there must also be a RETURN.
   const uint32_t max_refs = (GetNumDalvikInsns() - 1u) / 2u;
   ScopedArenaAllocator allocator(&cu_->arena_stack);
-  uint16_t* field_idxs = allocator.AllocArray<uint16_t>(max_refs, kArenaAllocMisc);
-  DexMemAccessType* field_types = allocator.AllocArray<DexMemAccessType>(max_refs, kArenaAllocMisc);
-
+  auto* field_idxs = allocator.AllocArray<uint32_t>(max_refs, kArenaAllocMisc);
+  DexMemAccessType* field_types = allocator.AllocArray<DexMemAccessType>(
+      max_refs, kArenaAllocMisc);
   // Find IGET/IPUT/SGET/SPUT insns, store IGET/IPUT fields at the beginning, SGET/SPUT at the end.
   size_t ifield_pos = 0u;
   size_t sfield_pos = max_refs;
@@ -1221,23 +1222,36 @@
       // Get field index and try to find it among existing indexes. If found, it's usually among
       // the last few added, so we'll start the search from ifield_pos/sfield_pos. Though this
       // is a linear search, it actually performs much better than map based approach.
-      if (IsInstructionIGetOrIPut(mir->dalvikInsn.opcode)) {
-        uint16_t field_idx = mir->dalvikInsn.vC;
+      const bool is_iget_or_iput = IsInstructionIGetOrIPut(mir->dalvikInsn.opcode);
+      const bool is_iget_or_iput_quick = IsInstructionIGetQuickOrIPutQuick(mir->dalvikInsn.opcode);
+      if (is_iget_or_iput || is_iget_or_iput_quick) {
+        uint32_t field_idx;
+        DexMemAccessType access_type;
+        if (is_iget_or_iput) {
+          field_idx = mir->dalvikInsn.vC;
+          access_type = IGetOrIPutMemAccessType(mir->dalvikInsn.opcode);
+        } else {
+          DCHECK(is_iget_or_iput_quick);
+          // Set kFieldIndexFlagQuickened so that we don't deduplicate against non quickened field
+          // indexes.
+          field_idx = mir->offset | kFieldIndexFlagQuickened;
+          access_type = IGetQuickOrIPutQuickMemAccessType(mir->dalvikInsn.opcode);
+        }
         size_t i = ifield_pos;
         while (i != 0u && field_idxs[i - 1] != field_idx) {
           --i;
         }
         if (i != 0u) {
           mir->meta.ifield_lowering_info = i - 1;
-          DCHECK_EQ(field_types[i - 1], IGetOrIPutMemAccessType(mir->dalvikInsn.opcode));
+          DCHECK_EQ(field_types[i - 1], access_type);
         } else {
           mir->meta.ifield_lowering_info = ifield_pos;
           field_idxs[ifield_pos] = field_idx;
-          field_types[ifield_pos] = IGetOrIPutMemAccessType(mir->dalvikInsn.opcode);
+          field_types[ifield_pos] = access_type;
           ++ifield_pos;
         }
       } else if (IsInstructionSGetOrSPut(mir->dalvikInsn.opcode)) {
-        uint16_t field_idx = mir->dalvikInsn.vB;
+        auto field_idx = mir->dalvikInsn.vB;
         size_t i = sfield_pos;
         while (i != max_refs && field_idxs[i] != field_idx) {
           ++i;
@@ -1261,7 +1275,12 @@
     DCHECK_EQ(ifield_lowering_infos_.size(), 0u);
     ifield_lowering_infos_.reserve(ifield_pos);
     for (size_t pos = 0u; pos != ifield_pos; ++pos) {
-      ifield_lowering_infos_.push_back(MirIFieldLoweringInfo(field_idxs[pos], field_types[pos]));
+      const uint32_t field_idx = field_idxs[pos];
+      const bool is_quickened = (field_idx & kFieldIndexFlagQuickened) != 0;
+      const uint32_t masked_field_idx = field_idx & ~kFieldIndexFlagQuickened;
+      CHECK_LT(masked_field_idx, 1u << 16);
+      ifield_lowering_infos_.push_back(
+          MirIFieldLoweringInfo(masked_field_idx, field_types[pos], is_quickened));
     }
     MirIFieldLoweringInfo::Resolve(cu_->compiler_driver, GetCurrentDexCompilationUnit(),
                                    ifield_lowering_infos_.data(), ifield_pos);
@@ -1282,18 +1301,19 @@
 
 void MIRGraph::DoCacheMethodLoweringInfo() {
   static constexpr uint16_t invoke_types[] = { kVirtual, kSuper, kDirect, kStatic, kInterface };
+  static constexpr uint32_t kMethodIdxFlagQuickened = 0x80000000;
 
   // Embed the map value in the entry to avoid extra padding in 64-bit builds.
   struct MapEntry {
     // Map key: target_method_idx, invoke_type, devirt_target. Ordered to avoid padding.
     const MethodReference* devirt_target;
-    uint16_t target_method_idx;
+    uint32_t target_method_idx;
+    uint32_t vtable_idx;
     uint16_t invoke_type;
     // Map value.
     uint32_t lowering_info_index;
   };
 
-  // Sort INVOKEs by method index, then by opcode, then by devirtualization target.
   struct MapEntryComparator {
     bool operator()(const MapEntry& lhs, const MapEntry& rhs) const {
       if (lhs.target_method_idx != rhs.target_method_idx) {
@@ -1302,6 +1322,9 @@
       if (lhs.invoke_type != rhs.invoke_type) {
         return lhs.invoke_type < rhs.invoke_type;
       }
+      if (lhs.vtable_idx != rhs.vtable_idx) {
+        return lhs.vtable_idx < rhs.vtable_idx;
+      }
       if (lhs.devirt_target != rhs.devirt_target) {
         if (lhs.devirt_target == nullptr) {
           return true;
@@ -1319,7 +1342,7 @@
   ScopedArenaAllocator allocator(&cu_->arena_stack);
 
   // All INVOKE instructions take 3 code units and there must also be a RETURN.
-  uint32_t max_refs = (GetNumDalvikInsns() - 1u) / 3u;
+  const uint32_t max_refs = (GetNumDalvikInsns() - 1u) / 3u;
 
   // Map invoke key (see MapEntry) to lowering info index and vice versa.
   // The invoke_map and sequential entries are essentially equivalent to Boost.MultiIndex's
@@ -1330,28 +1353,43 @@
       allocator.AllocArray<const MapEntry*>(max_refs, kArenaAllocMisc);
 
   // Find INVOKE insns and their devirtualization targets.
+  const VerifiedMethod* verified_method = GetCurrentDexCompilationUnit()->GetVerifiedMethod();
   AllNodesIterator iter(this);
   for (BasicBlock* bb = iter.Next(); bb != nullptr; bb = iter.Next()) {
     if (bb->block_type != kDalvikByteCode) {
       continue;
     }
     for (MIR* mir = bb->first_mir_insn; mir != nullptr; mir = mir->next) {
-      if (IsInstructionInvoke(mir->dalvikInsn.opcode)) {
-        // Decode target method index and invoke type.
-        uint16_t target_method_idx = mir->dalvikInsn.vB;
-        DexInvokeType invoke_type_idx = InvokeInstructionType(mir->dalvikInsn.opcode);
-
+      const bool is_quick_invoke = IsInstructionQuickInvoke(mir->dalvikInsn.opcode);
+      const bool is_invoke = IsInstructionInvoke(mir->dalvikInsn.opcode);
+      if (is_quick_invoke || is_invoke) {
+        uint32_t vtable_index = 0;
+        uint32_t target_method_idx = 0;
+        uint32_t invoke_type_idx = 0;  // Default to virtual (in case of quickened).
+        DCHECK_EQ(invoke_types[invoke_type_idx], kVirtual);
+        if (is_quick_invoke) {
+          // We need to store the vtable index since we can't necessarily recreate it at resolve
+          // phase if the dequickening resolved to an interface method.
+          vtable_index = mir->dalvikInsn.vB;
+          // Fake up the method index by storing the mir offset so that we can read the dequicken
+          // info in resolve.
+          target_method_idx = mir->offset | kMethodIdxFlagQuickened;
+        } else {
+          DCHECK(is_invoke);
+          // Decode target method index and invoke type.
+          invoke_type_idx = InvokeInstructionType(mir->dalvikInsn.opcode);
+          target_method_idx = mir->dalvikInsn.vB;
+        }
         // Find devirtualization target.
         // TODO: The devirt map is ordered by the dex pc here. Is there a way to get INVOKEs
         // ordered by dex pc as well? That would allow us to keep an iterator to devirt targets
         // and increment it as needed instead of making O(log n) lookups.
-        const VerifiedMethod* verified_method = GetCurrentDexCompilationUnit()->GetVerifiedMethod();
         const MethodReference* devirt_target = verified_method->GetDevirtTarget(mir->offset);
-
         // Try to insert a new entry. If the insertion fails, we will have found an old one.
         MapEntry entry = {
             devirt_target,
             target_method_idx,
+            vtable_index,
             invoke_types[invoke_type_idx],
             static_cast<uint32_t>(invoke_map.size())
         };
@@ -1362,22 +1400,24 @@
       }
     }
   }
-
   if (invoke_map.empty()) {
     return;
   }
-
   // Prepare unique method infos, set method info indexes for their MIRs.
-  DCHECK_EQ(method_lowering_infos_.size(), 0u);
   const size_t count = invoke_map.size();
   method_lowering_infos_.reserve(count);
   for (size_t pos = 0u; pos != count; ++pos) {
     const MapEntry* entry = sequential_entries[pos];
-    MirMethodLoweringInfo method_info(entry->target_method_idx,
-                                      static_cast<InvokeType>(entry->invoke_type));
+    const bool is_quick = (entry->target_method_idx & kMethodIdxFlagQuickened) != 0;
+    const uint32_t masked_method_idx = entry->target_method_idx & ~kMethodIdxFlagQuickened;
+    MirMethodLoweringInfo method_info(masked_method_idx,
+                                      static_cast<InvokeType>(entry->invoke_type), is_quick);
     if (entry->devirt_target != nullptr) {
       method_info.SetDevirtualizationTarget(*entry->devirt_target);
     }
+    if (is_quick) {
+      method_info.SetVTableIndex(entry->vtable_idx);
+    }
     method_lowering_infos_.push_back(method_info);
   }
   MirMethodLoweringInfo::Resolve(cu_->compiler_driver, GetCurrentDexCompilationUnit(),
diff --git a/compiler/dex/mir_dataflow.cc b/compiler/dex/mir_dataflow.cc
index f9f7e22..dfaff6c 100644
--- a/compiler/dex/mir_dataflow.cc
+++ b/compiler/dex/mir_dataflow.cc
@@ -374,7 +374,7 @@
   // 72 INVOKE_INTERFACE {vD, vE, vF, vG, vA}
   DF_FORMAT_35C | DF_NULL_CHK_OUT0 | DF_UMS,
 
-  // 73 UNUSED_73
+  // 73 RETURN_VOID_BARRIER
   DF_NOP,
 
   // 74 INVOKE_VIRTUAL_RANGE {vCCCC .. vNNNN}
@@ -710,89 +710,89 @@
   // E2 USHR_INT_LIT8 vAA, vBB, #+CC
   DF_DA | DF_UB | DF_CORE_A | DF_CORE_B,
 
-  // E3 IGET_VOLATILE
+  // E3 IGET_QUICK
   DF_DA | DF_UB | DF_NULL_CHK_B | DF_REF_B | DF_IFIELD | DF_LVN,
 
-  // E4 IPUT_VOLATILE
-  DF_UA | DF_UB | DF_NULL_CHK_B | DF_REF_B | DF_IFIELD | DF_LVN,
-
-  // E5 SGET_VOLATILE
-  DF_DA | DF_SFIELD | DF_CLINIT | DF_UMS,
-
-  // E6 SPUT_VOLATILE
-  DF_UA | DF_SFIELD | DF_CLINIT | DF_UMS,
-
-  // E7 IGET_OBJECT_VOLATILE
-  DF_DA | DF_UB | DF_NULL_CHK_B | DF_REF_A | DF_REF_B | DF_IFIELD | DF_LVN,
-
-  // E8 IGET_WIDE_VOLATILE
+  // E4 IGET_WIDE_QUICK
   DF_DA | DF_A_WIDE | DF_UB | DF_NULL_CHK_B | DF_REF_B | DF_IFIELD | DF_LVN,
 
-  // E9 IPUT_WIDE_VOLATILE
+  // E5 IGET_OBJECT_QUICK
+  DF_DA | DF_UB | DF_NULL_CHK_B | DF_REF_A | DF_REF_B | DF_IFIELD | DF_LVN,
+
+  // E6 IPUT_QUICK
+  DF_UA | DF_UB | DF_NULL_CHK_B | DF_REF_B | DF_IFIELD | DF_LVN,
+
+  // E7 IPUT_WIDE_QUICK
   DF_UA | DF_A_WIDE | DF_UB | DF_NULL_CHK_B | DF_REF_B | DF_IFIELD | DF_LVN,
 
-  // EA SGET_WIDE_VOLATILE
-  DF_DA | DF_A_WIDE | DF_SFIELD | DF_CLINIT | DF_UMS,
-
-  // EB SPUT_WIDE_VOLATILE
-  DF_UA | DF_A_WIDE | DF_SFIELD | DF_CLINIT | DF_UMS,
-
-  // EC BREAKPOINT
-  DF_NOP,
-
-  // ED THROW_VERIFICATION_ERROR
-  DF_NOP | DF_UMS,
-
-  // EE EXECUTE_INLINE
-  DF_FORMAT_35C,
-
-  // EF EXECUTE_INLINE_RANGE
-  DF_FORMAT_3RC,
-
-  // F0 INVOKE_OBJECT_INIT_RANGE
-  DF_NOP,
-
-  // F1 RETURN_VOID_BARRIER
-  DF_NOP,
-
-  // F2 IGET_QUICK
-  DF_DA | DF_UB | DF_NULL_CHK_B | DF_IFIELD | DF_LVN,
-
-  // F3 IGET_WIDE_QUICK
-  DF_DA | DF_A_WIDE | DF_UB | DF_NULL_CHK_B | DF_IFIELD | DF_LVN,
-
-  // F4 IGET_OBJECT_QUICK
-  DF_DA | DF_UB | DF_NULL_CHK_B | DF_IFIELD | DF_LVN,
-
-  // F5 IPUT_QUICK
-  DF_UA | DF_UB | DF_NULL_CHK_B | DF_IFIELD | DF_LVN,
-
-  // F6 IPUT_WIDE_QUICK
-  DF_UA | DF_A_WIDE | DF_UB | DF_NULL_CHK_B | DF_IFIELD | DF_LVN,
-
-  // F7 IPUT_OBJECT_QUICK
-  DF_UA | DF_UB | DF_NULL_CHK_B | DF_IFIELD | DF_LVN,
-
-  // F8 INVOKE_VIRTUAL_QUICK
-  DF_FORMAT_35C | DF_NULL_CHK_OUT0 | DF_UMS,
-
-  // F9 INVOKE_VIRTUAL_QUICK_RANGE
-  DF_FORMAT_3RC | DF_NULL_CHK_OUT0 | DF_UMS,
-
-  // FA INVOKE_SUPER_QUICK
-  DF_FORMAT_35C | DF_NULL_CHK_OUT0 | DF_UMS,
-
-  // FB INVOKE_SUPER_QUICK_RANGE
-  DF_FORMAT_3RC | DF_NULL_CHK_OUT0 | DF_UMS,
-
-  // FC IPUT_OBJECT_VOLATILE
+  // E8 IPUT_OBJECT_QUICK
   DF_UA | DF_UB | DF_NULL_CHK_B | DF_REF_A | DF_REF_B | DF_IFIELD | DF_LVN,
 
-  // FD SGET_OBJECT_VOLATILE
-  DF_DA | DF_REF_A | DF_SFIELD | DF_CLINIT | DF_UMS,
+  // E9 INVOKE_VIRTUAL_QUICK
+  DF_FORMAT_35C | DF_NULL_CHK_OUT0 | DF_UMS,
 
-  // FE SPUT_OBJECT_VOLATILE
-  DF_UA | DF_REF_A | DF_SFIELD | DF_CLINIT | DF_UMS,
+  // EA INVOKE_VIRTUAL_RANGE_QUICK
+  DF_FORMAT_3RC | DF_NULL_CHK_OUT0 | DF_UMS,
+
+  // EB IPUT_BOOLEAN_QUICK vA, vB, index
+  DF_UA | DF_UB | DF_NULL_CHK_B | DF_REF_B | DF_IFIELD | DF_LVN,
+
+  // EC IPUT_BYTE_QUICK vA, vB, index
+  DF_UA | DF_UB | DF_NULL_CHK_B | DF_REF_B | DF_IFIELD | DF_LVN,
+
+  // ED IPUT_CHAR_QUICK vA, vB, index
+  DF_UA | DF_UB | DF_NULL_CHK_B | DF_REF_B | DF_IFIELD | DF_LVN,
+
+  // EE IPUT_SHORT_QUICK vA, vB, index
+  DF_UA | DF_UB | DF_NULL_CHK_B | DF_REF_B | DF_IFIELD | DF_LVN,
+
+  // EF IGET_BOOLEAN_QUICK vA, vB, index
+  DF_DA | DF_UB | DF_NULL_CHK_B | DF_REF_B | DF_IFIELD | DF_LVN,
+
+  // F0 IGET_BYTE_QUICK vA, vB, index
+  DF_DA | DF_UB | DF_NULL_CHK_B | DF_REF_B | DF_IFIELD | DF_LVN,
+
+  // F1 IGET_CHAR_QUICK vA, vB, index
+  DF_DA | DF_UB | DF_NULL_CHK_B | DF_REF_B | DF_IFIELD | DF_LVN,
+
+  // F2 IGET_SHORT_QUICK vA, vB, index
+  DF_DA | DF_UB | DF_NULL_CHK_B | DF_REF_B | DF_IFIELD | DF_LVN,
+
+  // F3 UNUSED_F3
+  DF_NOP,
+
+  // F4 UNUSED_F4
+  DF_NOP,
+
+  // F5 UNUSED_F5
+  DF_NOP,
+
+  // F6 UNUSED_F6
+  DF_NOP,
+
+  // F7 UNUSED_F7
+  DF_NOP,
+
+  // F8 UNUSED_F8
+  DF_NOP,
+
+  // F9 UNUSED_F9
+  DF_NOP,
+
+  // FA UNUSED_FA
+  DF_NOP,
+
+  // FB UNUSED_FB
+  DF_NOP,
+
+  // FC UNUSED_FC
+  DF_NOP,
+
+  // FD UNUSED_FD
+  DF_NOP,
+
+  // FE UNUSED_FE
+  DF_NOP,
 
   // FF UNUSED_FF
   DF_NOP,
diff --git a/compiler/dex/mir_field_info.cc b/compiler/dex/mir_field_info.cc
index 53afcad..d2079a2 100644
--- a/compiler/dex/mir_field_info.cc
+++ b/compiler/dex/mir_field_info.cc
@@ -35,8 +35,9 @@
     DCHECK(field_infos != nullptr);
     DCHECK_NE(count, 0u);
     for (auto it = field_infos, end = field_infos + count; it != end; ++it) {
-      MirIFieldLoweringInfo unresolved(it->field_idx_, it->MemAccessType());
-      DCHECK_EQ(memcmp(&unresolved, &*it, sizeof(*it)), 0);
+      MirIFieldLoweringInfo unresolved(it->field_idx_, it->MemAccessType(), it->IsQuickened());
+      unresolved.field_offset_ = it->field_offset_;
+      unresolved.CheckEquals(*it);
     }
   }
 
@@ -49,13 +50,30 @@
       hs.NewHandle(compiler_driver->GetClassLoader(soa, mUnit)));
   Handle<mirror::Class> referrer_class(hs.NewHandle(
       compiler_driver->ResolveCompilingMethodsClass(soa, dex_cache, class_loader, mUnit)));
+  const VerifiedMethod* const verified_method = mUnit->GetVerifiedMethod();
   // Even if the referrer class is unresolved (i.e. we're compiling a method without class
   // definition) we still want to resolve fields and record all available info.
-
   for (auto it = field_infos, end = field_infos + count; it != end; ++it) {
-    uint32_t field_idx = it->field_idx_;
-    mirror::ArtField* resolved_field =
-        compiler_driver->ResolveField(soa, dex_cache, class_loader, mUnit, field_idx, false);
+    uint32_t field_idx;
+    mirror::ArtField* resolved_field;
+    if (!it->IsQuickened()) {
+      field_idx = it->field_idx_;
+      resolved_field = compiler_driver->ResolveField(soa, dex_cache, class_loader, mUnit,
+                                                     field_idx, false);
+    } else {
+      const auto mir_offset = it->field_idx_;
+      // For quickened instructions, it->field_offset_ actually contains the mir offset.
+      // We need to use the de-quickening info to get dex file / field idx
+      auto* field_idx_ptr = verified_method->GetDequickenIndex(mir_offset);
+      CHECK(field_idx_ptr != nullptr);
+      field_idx = field_idx_ptr->index;
+      StackHandleScope<1> hs2(soa.Self());
+      auto h_dex_cache = hs2.NewHandle(compiler_driver->FindDexCache(field_idx_ptr->dex_file));
+      resolved_field = compiler_driver->ResolveFieldWithDexFile(
+          soa, h_dex_cache, class_loader, field_idx_ptr->dex_file, field_idx, false);
+      // Since we don't have a valid field index we can't go slow path later.
+      CHECK(resolved_field != nullptr);
+    }
     if (UNLIKELY(resolved_field == nullptr)) {
       continue;
     }
diff --git a/compiler/dex/mir_field_info.h b/compiler/dex/mir_field_info.h
index 98b2da8..ca56958 100644
--- a/compiler/dex/mir_field_info.h
+++ b/compiler/dex/mir_field_info.h
@@ -19,8 +19,8 @@
 
 #include "base/macros.h"
 #include "dex_file.h"
+#include "dex_instruction_utils.h"
 #include "offsets.h"
-#include "utils/dex_instruction_utils.h"
 
 namespace art {
 
@@ -39,6 +39,9 @@
   uint16_t FieldIndex() const {
     return field_idx_;
   }
+  void SetFieldIndex(uint16_t field_idx) {
+    field_idx_ = field_idx;
+  }
 
   bool IsStatic() const {
     return (flags_ & kFlagIsStatic) != 0u;
@@ -51,6 +54,9 @@
   const DexFile* DeclaringDexFile() const {
     return declaring_dex_file_;
   }
+  void SetDeclaringDexFile(const DexFile* dex_file) {
+    declaring_dex_file_ = dex_file;
+  }
 
   uint16_t DeclaringClassIndex() const {
     return declaring_class_idx_;
@@ -64,20 +70,35 @@
     return (flags_ & kFlagIsVolatile) != 0u;
   }
 
+  // IGET_QUICK, IGET_BYTE_QUICK, ...
+  bool IsQuickened() const {
+    return (flags_ & kFlagIsQuickened) != 0u;
+  }
+
   DexMemAccessType MemAccessType() const {
     return static_cast<DexMemAccessType>((flags_ >> kBitMemAccessTypeBegin) & kMemAccessTypeMask);
   }
 
+  void CheckEquals(const MirFieldInfo& other) const {
+    CHECK_EQ(field_idx_, other.field_idx_);
+    CHECK_EQ(flags_, other.flags_);
+    CHECK_EQ(declaring_field_idx_, other.declaring_field_idx_);
+    CHECK_EQ(declaring_class_idx_, other.declaring_class_idx_);
+    CHECK_EQ(declaring_dex_file_, other.declaring_dex_file_);
+  }
+
  protected:
   enum {
     kBitIsStatic = 0,
     kBitIsVolatile,
+    kBitIsQuickened,
     kBitMemAccessTypeBegin,
     kBitMemAccessTypeEnd = kBitMemAccessTypeBegin + 3,  // 3 bits for raw type.
     kFieldInfoBitEnd = kBitMemAccessTypeEnd
   };
   static constexpr uint16_t kFlagIsVolatile = 1u << kBitIsVolatile;
   static constexpr uint16_t kFlagIsStatic = 1u << kBitIsStatic;
+  static constexpr uint16_t kFlagIsQuickened = 1u << kBitIsQuickened;
   static constexpr uint16_t kMemAccessTypeMask = 7u;
   static_assert((1u << (kBitMemAccessTypeEnd - kBitMemAccessTypeBegin)) - 1u == kMemAccessTypeMask,
                 "Invalid raw type mask");
@@ -117,8 +138,10 @@
       LOCKS_EXCLUDED(Locks::mutator_lock_);
 
   // Construct an unresolved instance field lowering info.
-  explicit MirIFieldLoweringInfo(uint16_t field_idx, DexMemAccessType type)
-      : MirFieldInfo(field_idx, kFlagIsVolatile, type),  // Without kFlagIsStatic.
+  explicit MirIFieldLoweringInfo(uint16_t field_idx, DexMemAccessType type, bool is_quickened)
+      : MirFieldInfo(field_idx,
+                     kFlagIsVolatile | (is_quickened ? kFlagIsQuickened : 0u),
+                     type),  // Without kFlagIsStatic.
         field_offset_(0u) {
   }
 
@@ -134,6 +157,11 @@
     return field_offset_;
   }
 
+  void CheckEquals(const MirIFieldLoweringInfo& other) const {
+    MirFieldInfo::CheckEquals(other);
+    CHECK_EQ(field_offset_.Uint32Value(), other.field_offset_.Uint32Value());
+  }
+
  private:
   enum {
     kBitFastGet = kFieldInfoBitEnd,
diff --git a/compiler/dex/mir_graph.cc b/compiler/dex/mir_graph.cc
index 76b5e44..f354a49 100644
--- a/compiler/dex/mir_graph.cc
+++ b/compiler/dex/mir_graph.cc
@@ -1673,12 +1673,6 @@
   }
 }
 
-const char* MIRGraph::GetShortyFromTargetIdx(int target_idx) {
-  // TODO: for inlining support, use current code unit.
-  const DexFile::MethodId& method_id = cu_->dex_file->GetMethodId(target_idx);
-  return cu_->dex_file->GetShorty(method_id.proto_idx_);
-}
-
 const char* MIRGraph::GetShortyFromMethodReference(const MethodReference& target_method) {
   const DexFile::MethodId& method_id =
       target_method.dex_file->GetMethodId(target_method.dex_method_index);
@@ -1724,8 +1718,7 @@
  * high-word loc for wide arguments.  Also pull up any following
  * MOVE_RESULT and incorporate it into the invoke.
  */
-CallInfo* MIRGraph::NewMemCallInfo(BasicBlock* bb, MIR* mir, InvokeType type,
-                                  bool is_range) {
+CallInfo* MIRGraph::NewMemCallInfo(BasicBlock* bb, MIR* mir, InvokeType type, bool is_range) {
   CallInfo* info = static_cast<CallInfo*>(arena_->Alloc(sizeof(CallInfo),
                                                         kArenaAllocMisc));
   MIR* move_result_mir = FindMoveResult(bb, mir);
@@ -1744,6 +1737,13 @@
   info->opt_flags = mir->optimization_flags;
   info->type = type;
   info->is_range = is_range;
+  if (IsInstructionQuickInvoke(mir->dalvikInsn.opcode)) {
+    const auto& method_info = GetMethodLoweringInfo(mir);
+    info->method_ref = method_info.GetTargetMethod();
+  } else {
+    info->method_ref = MethodReference(GetCurrentDexCompilationUnit()->GetDexFile(),
+                                       mir->dalvikInsn.vB);
+  }
   info->index = mir->dalvikInsn.vB;
   info->offset = mir->offset;
   info->mir = mir;
diff --git a/compiler/dex/mir_graph.h b/compiler/dex/mir_graph.h
index e5abd3b..3dae5b4 100644
--- a/compiler/dex/mir_graph.h
+++ b/compiler/dex/mir_graph.h
@@ -504,6 +504,7 @@
   int opt_flags;
   InvokeType type;
   uint32_t dex_idx;
+  MethodReference method_ref;
   uint32_t index;         // Method idx for invokes, type idx for FilledNewArray.
   uintptr_t direct_code;
   uintptr_t direct_method;
@@ -687,7 +688,7 @@
 
   void DoCacheMethodLoweringInfo();
 
-  const MirMethodLoweringInfo& GetMethodLoweringInfo(MIR* mir) {
+  const MirMethodLoweringInfo& GetMethodLoweringInfo(MIR* mir) const {
     DCHECK_LT(mir->meta.method_lowering_info, method_lowering_infos_.size());
     return method_lowering_infos_[mir->meta.method_lowering_info];
   }
@@ -1132,7 +1133,6 @@
   std::string GetSSAName(int ssa_reg);
   std::string GetSSANameWithConst(int ssa_reg, bool singles_only);
   void GetBlockName(BasicBlock* bb, char* name);
-  const char* GetShortyFromTargetIdx(int);
   const char* GetShortyFromMethodReference(const MethodReference& target_method);
   void DumpMIRGraph();
   CallInfo* NewMemCallInfo(BasicBlock* bb, MIR* mir, InvokeType type, bool is_range);
diff --git a/compiler/dex/mir_method_info.cc b/compiler/dex/mir_method_info.cc
index b234950..3d3d979 100644
--- a/compiler/dex/mir_method_info.cc
+++ b/compiler/dex/mir_method_info.cc
@@ -33,51 +33,103 @@
     DCHECK(method_infos != nullptr);
     DCHECK_NE(count, 0u);
     for (auto it = method_infos, end = method_infos + count; it != end; ++it) {
-      MirMethodLoweringInfo unresolved(it->MethodIndex(), it->GetInvokeType());
+      MirMethodLoweringInfo unresolved(it->MethodIndex(), it->GetInvokeType(), it->IsQuickened());
+      unresolved.declaring_dex_file_ = it->declaring_dex_file_;
+      unresolved.vtable_idx_ = it->vtable_idx_;
       if (it->target_dex_file_ != nullptr) {
         unresolved.target_dex_file_ = it->target_dex_file_;
         unresolved.target_method_idx_ = it->target_method_idx_;
       }
-      DCHECK_EQ(memcmp(&unresolved, &*it, sizeof(*it)), 0);
+      if (kIsDebugBuild) {
+        unresolved.CheckEquals(*it);
+      }
     }
   }
 
   // We're going to resolve methods and check access in a tight loop. It's better to hold
   // the lock and needed references once than re-acquiring them again and again.
   ScopedObjectAccess soa(Thread::Current());
-  StackHandleScope<3> hs(soa.Self());
+  StackHandleScope<4> hs(soa.Self());
   Handle<mirror::DexCache> dex_cache(hs.NewHandle(compiler_driver->GetDexCache(mUnit)));
   Handle<mirror::ClassLoader> class_loader(
       hs.NewHandle(compiler_driver->GetClassLoader(soa, mUnit)));
   Handle<mirror::Class> referrer_class(hs.NewHandle(
       compiler_driver->ResolveCompilingMethodsClass(soa, dex_cache, class_loader, mUnit)));
+  auto current_dex_cache(hs.NewHandle<mirror::DexCache>(nullptr));
   // Even if the referrer class is unresolved (i.e. we're compiling a method without class
   // definition) we still want to resolve methods and record all available info.
+  const DexFile* const dex_file = mUnit->GetDexFile();
+  const bool use_jit = Runtime::Current()->UseJit();
+  const VerifiedMethod* const verified_method = mUnit->GetVerifiedMethod();
 
   for (auto it = method_infos, end = method_infos + count; it != end; ++it) {
+    // For quickened invokes, the dex method idx is actually the mir offset.
+    if (it->IsQuickened()) {
+      const auto* dequicken_ref = verified_method->GetDequickenIndex(it->method_idx_);
+      CHECK(dequicken_ref != nullptr);
+      it->target_dex_file_ = dequicken_ref->dex_file;
+      it->target_method_idx_ = dequicken_ref->index;
+    }
     // Remember devirtualized invoke target and set the called method to the default.
     MethodReference devirt_ref(it->target_dex_file_, it->target_method_idx_);
     MethodReference* devirt_target = (it->target_dex_file_ != nullptr) ? &devirt_ref : nullptr;
-    it->target_dex_file_ = mUnit->GetDexFile();
-    it->target_method_idx_ = it->MethodIndex();
-
     InvokeType invoke_type = it->GetInvokeType();
-    mirror::ArtMethod* resolved_method =
-        compiler_driver->ResolveMethod(soa, dex_cache, class_loader, mUnit, it->MethodIndex(),
-                                       invoke_type);
+    mirror::ArtMethod* resolved_method = nullptr;
+    if (!it->IsQuickened()) {
+      it->target_dex_file_ = dex_file;
+      it->target_method_idx_ = it->MethodIndex();
+      current_dex_cache.Assign(dex_cache.Get());
+      resolved_method = compiler_driver->ResolveMethod(soa, dex_cache, class_loader, mUnit,
+                                                       it->MethodIndex(), invoke_type);
+    } else {
+      // The method index is actually the dex PC in this case.
+      // Calculate the proper dex file and target method idx.
+      CHECK(use_jit);
+      CHECK_EQ(invoke_type, kVirtual);
+      // Don't devirt if we are in a different dex file since we can't have direct invokes in
+      // another dex file unless we always put a direct / patch pointer.
+      devirt_target = nullptr;
+      current_dex_cache.Assign(
+          Runtime::Current()->GetClassLinker()->FindDexCache(*it->target_dex_file_));
+      CHECK(current_dex_cache.Get() != nullptr);
+      DexCompilationUnit cu(
+          mUnit->GetCompilationUnit(), mUnit->GetClassLoader(), mUnit->GetClassLinker(),
+          *it->target_dex_file_, nullptr /* code_item not used */, 0u /* class_def_idx not used */,
+          it->target_method_idx_, 0u /* access_flags not used */,
+          nullptr /* verified_method not used */);
+      resolved_method = compiler_driver->ResolveMethod(soa, current_dex_cache, class_loader, &cu,
+                                                       it->target_method_idx_, invoke_type, false);
+      if (resolved_method != nullptr) {
+        // Since this was a dequickened virtual, it is guaranteed to be resolved. However, it may be
+        // resolved to an interface method. If this is the case then change the invoke type to
+        // interface with the assumption that sharp_type will be kVirtual.
+        if (resolved_method->GetInvokeType() == kInterface) {
+          it->flags_ = (it->flags_ & ~(kInvokeTypeMask << kBitInvokeTypeBegin)) |
+              (static_cast<uint16_t>(kInterface) << kBitInvokeTypeBegin);
+        }
+      }
+    }
     if (UNLIKELY(resolved_method == nullptr)) {
       continue;
     }
     compiler_driver->GetResolvedMethodDexFileLocation(resolved_method,
         &it->declaring_dex_file_, &it->declaring_class_idx_, &it->declaring_method_idx_);
-    it->vtable_idx_ = compiler_driver->GetResolvedMethodVTableIndex(resolved_method, invoke_type);
+    if (!it->IsQuickened()) {
+      // For quickened invoke virtuals we may have desharpened to an interface method which
+      // wont give us the right method index, in this case blindly dispatch or else we can't
+      // compile the method. Converting the invoke to interface dispatch doesn't work since we
+      // have no way to get the dex method index for quickened invoke virtuals in the interface
+      // trampolines.
+      it->vtable_idx_ =
+          compiler_driver->GetResolvedMethodVTableIndex(resolved_method, invoke_type);
+    }
 
-    MethodReference target_method(mUnit->GetDexFile(), it->MethodIndex());
+    MethodReference target_method(it->target_dex_file_, it->target_method_idx_);
     int fast_path_flags = compiler_driver->IsFastInvoke(
-        soa, dex_cache, class_loader, mUnit, referrer_class.Get(), resolved_method, &invoke_type,
-        &target_method, devirt_target, &it->direct_code_, &it->direct_method_);
-    bool is_referrers_class = (referrer_class.Get() == resolved_method->GetDeclaringClass());
-    bool is_class_initialized =
+        soa, dex_cache, class_loader, mUnit, referrer_class.Get(), resolved_method,
+        &invoke_type, &target_method, devirt_target, &it->direct_code_, &it->direct_method_);
+    const bool is_referrers_class = referrer_class.Get() == resolved_method->GetDeclaringClass();
+    const bool is_class_initialized =
         compiler_driver->IsMethodsClassInitialized(referrer_class.Get(), resolved_method);
     uint16_t other_flags = it->flags_ &
         ~(kFlagFastPath | kFlagClassIsInitialized | (kInvokeTypeMask << kBitSharpTypeBegin));
diff --git a/compiler/dex/mir_method_info.h b/compiler/dex/mir_method_info.h
index 08fb103..e131c96 100644
--- a/compiler/dex/mir_method_info.h
+++ b/compiler/dex/mir_method_info.h
@@ -46,6 +46,9 @@
   const DexFile* DeclaringDexFile() const {
     return declaring_dex_file_;
   }
+  void SetDeclaringDexFile(const DexFile* dex_file) {
+    declaring_dex_file_ = dex_file;
+  }
 
   uint16_t DeclaringClassIndex() const {
     return declaring_class_idx_;
@@ -98,11 +101,12 @@
                       MirMethodLoweringInfo* method_infos, size_t count)
       LOCKS_EXCLUDED(Locks::mutator_lock_);
 
-  MirMethodLoweringInfo(uint16_t method_idx, InvokeType type)
+  MirMethodLoweringInfo(uint16_t method_idx, InvokeType type, bool is_quickened)
       : MirMethodInfo(method_idx,
                       ((type == kStatic) ? kFlagIsStatic : 0u) |
                       (static_cast<uint16_t>(type) << kBitInvokeTypeBegin) |
-                      (static_cast<uint16_t>(type) << kBitSharpTypeBegin)),
+                      (static_cast<uint16_t>(type) << kBitSharpTypeBegin) |
+                      (is_quickened ? kFlagQuickened : 0u)),
         direct_code_(0u),
         direct_method_(0u),
         target_dex_file_(nullptr),
@@ -131,6 +135,11 @@
     return (flags_ & kFlagClassIsInitialized) != 0u;
   }
 
+  // Returns true iff the method invoke is INVOKE_VIRTUAL_QUICK or INVOKE_VIRTUAL_RANGE_QUICK.
+  bool IsQuickened() const {
+    return (flags_ & kFlagQuickened) != 0u;
+  }
+
   InvokeType GetInvokeType() const {
     return static_cast<InvokeType>((flags_ >> kBitInvokeTypeBegin) & kInvokeTypeMask);
   }
@@ -146,6 +155,9 @@
   uint16_t VTableIndex() const {
     return vtable_idx_;
   }
+  void SetVTableIndex(uint16_t index) {
+    vtable_idx_ = index;
+  }
 
   uintptr_t DirectCode() const {
     return direct_code_;
@@ -159,6 +171,20 @@
     return stats_flags_;
   }
 
+  void CheckEquals(const MirMethodLoweringInfo& info) const {
+    CHECK_EQ(method_idx_, info.method_idx_);
+    CHECK_EQ(flags_, info.flags_);
+    CHECK_EQ(declaring_method_idx_, info.declaring_method_idx_);
+    CHECK_EQ(declaring_class_idx_, info.declaring_class_idx_);
+    CHECK_EQ(declaring_dex_file_, info.declaring_dex_file_);
+    CHECK_EQ(direct_code_, info.direct_code_);
+    CHECK_EQ(direct_method_, info.direct_method_);
+    CHECK_EQ(target_dex_file_, info.target_dex_file_);
+    CHECK_EQ(target_method_idx_, info.target_method_idx_);
+    CHECK_EQ(vtable_idx_, info.vtable_idx_);
+    CHECK_EQ(stats_flags_, info.stats_flags_);
+  }
+
  private:
   enum {
     kBitFastPath = kMethodInfoBitEnd,
@@ -168,12 +194,14 @@
     kBitSharpTypeEnd = kBitSharpTypeBegin + 3,  // 3 bits for sharp type.
     kBitIsReferrersClass = kBitSharpTypeEnd,
     kBitClassIsInitialized,
+    kBitQuickened,
     kMethodLoweringInfoBitEnd
   };
   static_assert(kMethodLoweringInfoBitEnd <= 16, "Too many flags");
   static constexpr uint16_t kFlagFastPath = 1u << kBitFastPath;
   static constexpr uint16_t kFlagIsReferrersClass = 1u << kBitIsReferrersClass;
   static constexpr uint16_t kFlagClassIsInitialized = 1u << kBitClassIsInitialized;
+  static constexpr uint16_t kFlagQuickened = 1u << kBitQuickened;
   static constexpr uint16_t kInvokeTypeMask = 7u;
   static_assert((1u << (kBitInvokeTypeEnd - kBitInvokeTypeBegin)) - 1u == kInvokeTypeMask,
                 "assert invoke type bits failed");
diff --git a/compiler/dex/mir_optimization.cc b/compiler/dex/mir_optimization.cc
index fd67d4e..93749e4 100644
--- a/compiler/dex/mir_optimization.cc
+++ b/compiler/dex/mir_optimization.cc
@@ -1437,7 +1437,7 @@
       nullptr /* code_item not used */, 0u /* class_def_idx not used */, target.dex_method_index,
       0u /* access_flags not used */, nullptr /* verified_method not used */);
   DexMemAccessType type = IGetOrIPutMemAccessType(iget_or_iput->dalvikInsn.opcode);
-  MirIFieldLoweringInfo inlined_field_info(field_idx, type);
+  MirIFieldLoweringInfo inlined_field_info(field_idx, type, false);
   MirIFieldLoweringInfo::Resolve(cu_->compiler_driver, &inlined_unit, &inlined_field_info, 1u);
   DCHECK(inlined_field_info.IsResolved());
 
diff --git a/compiler/dex/mir_optimization_test.cc b/compiler/dex/mir_optimization_test.cc
index be05b80..9ce5ebb 100644
--- a/compiler/dex/mir_optimization_test.cc
+++ b/compiler/dex/mir_optimization_test.cc
@@ -254,7 +254,7 @@
     cu_.mir_graph->method_lowering_infos_.reserve(count);
     for (size_t i = 0u; i != count; ++i) {
       const MethodDef* def = &defs[i];
-      MirMethodLoweringInfo method_info(def->method_idx, def->invoke_type);
+      MirMethodLoweringInfo method_info(def->method_idx, def->invoke_type, false);
       if (def->declaring_dex_file != 0u) {
         method_info.declaring_dex_file_ = reinterpret_cast<const DexFile*>(def->declaring_dex_file);
         method_info.declaring_class_idx_ = def->declaring_class_idx;
@@ -407,7 +407,7 @@
     cu_.mir_graph->ifield_lowering_infos_.reserve(count);
     for (size_t i = 0u; i != count; ++i) {
       const IFieldDef* def = &defs[i];
-      MirIFieldLoweringInfo field_info(def->field_idx, def->type);
+      MirIFieldLoweringInfo field_info(def->field_idx, def->type, false);
       if (def->declaring_dex_file != 0u) {
         field_info.declaring_dex_file_ = reinterpret_cast<const DexFile*>(def->declaring_dex_file);
         field_info.declaring_class_idx_ = def->declaring_class_idx;
diff --git a/compiler/dex/quick/arm/assemble_arm.cc b/compiler/dex/quick/arm/assemble_arm.cc
index 3d64833..8833da3 100644
--- a/compiler/dex/quick/arm/assemble_arm.cc
+++ b/compiler/dex/quick/arm/assemble_arm.cc
@@ -117,11 +117,11 @@
                  "add", "!0C, !1C", 2, kFixupNone),
     ENCODING_MAP(kThumbAddPcRel,    0xa000,
                  kFmtBitBlt, 10, 8, kFmtBitBlt, 7, 0, kFmtUnused, -1, -1,
-                 kFmtUnused, -1, -1, IS_TERTIARY_OP | IS_BRANCH | NEEDS_FIXUP,
+                 kFmtUnused, -1, -1, IS_TERTIARY_OP | REG_DEF0 | REG_USE_PC | NEEDS_FIXUP,
                  "add", "!0C, pc, #!1E", 2, kFixupLoad),
     ENCODING_MAP(kThumbAddSpRel,    0xa800,
                  kFmtBitBlt, 10, 8, kFmtSkip, -1, -1, kFmtBitBlt, 7, 0,
-                 kFmtUnused, -1, -1, IS_TERTIARY_OP | REG_DEF_SP | REG_USE_SP,
+                 kFmtUnused, -1, -1, IS_TERTIARY_OP | REG_DEF0 | REG_USE_SP,
                  "add", "!0C, sp, #!2E", 2, kFixupNone),
     ENCODING_MAP(kThumbAddSpI7,      0xb000,
                  kFmtBitBlt, 6, 0, kFmtUnused, -1, -1, kFmtUnused, -1, -1,
@@ -182,7 +182,7 @@
                  "blx", "!0C", 2, kFixupNone),
     ENCODING_MAP(kThumbBx,            0x4700,
                  kFmtBitBlt, 6, 3, kFmtUnused, -1, -1, kFmtUnused, -1, -1,
-                 kFmtUnused, -1, -1, IS_UNARY_OP | IS_BRANCH,
+                 kFmtUnused, -1, -1, IS_UNARY_OP | REG_USE0 | IS_BRANCH,
                  "bx", "!0C", 2, kFixupNone),
     ENCODING_MAP(kThumbCmnRR,        0x42c0,
                  kFmtBitBlt, 2, 0, kFmtBitBlt, 5, 3, kFmtUnused, -1, -1,
@@ -693,7 +693,7 @@
     ENCODING_MAP(kThumb2AdcRRR,  0xeb500000, /* setflags encoding */
                  kFmtBitBlt, 11, 8, kFmtBitBlt, 19, 16, kFmtBitBlt, 3, 0,
                  kFmtShift, -1, -1,
-                 IS_QUAD_OP | REG_DEF0_USE12 | SETS_CCODES,
+                 IS_QUAD_OP | REG_DEF0_USE12 | SETS_CCODES | USES_CCODES,
                  "adcs", "!0C, !1C, !2C!3H", 4, kFixupNone),
     ENCODING_MAP(kThumb2AndRRR,  0xea000000,
                  kFmtBitBlt, 11, 8, kFmtBitBlt, 19, 16, kFmtBitBlt, 3, 0,
@@ -835,15 +835,15 @@
                  "it:!1b", "!0c", 2, kFixupNone),
     ENCODING_MAP(kThumb2Fmstat,  0xeef1fa10,
                  kFmtUnused, -1, -1, kFmtUnused, -1, -1, kFmtUnused, -1, -1,
-                 kFmtUnused, -1, -1, NO_OPERAND | SETS_CCODES,
+                 kFmtUnused, -1, -1, NO_OPERAND | SETS_CCODES | USES_CCODES,
                  "fmstat", "", 4, kFixupNone),
     ENCODING_MAP(kThumb2Vcmpd,        0xeeb40b40,
                  kFmtDfp, 22, 12, kFmtDfp, 5, 0, kFmtUnused, -1, -1,
-                 kFmtUnused, -1, -1, IS_BINARY_OP | REG_USE01,
+                 kFmtUnused, -1, -1, IS_BINARY_OP | REG_USE01 | SETS_CCODES,
                  "vcmp.f64", "!0S, !1S", 4, kFixupNone),
     ENCODING_MAP(kThumb2Vcmps,        0xeeb40a40,
                  kFmtSfp, 22, 12, kFmtSfp, 5, 0, kFmtUnused, -1, -1,
-                 kFmtUnused, -1, -1, IS_BINARY_OP | REG_USE01,
+                 kFmtUnused, -1, -1, IS_BINARY_OP | REG_USE01 | SETS_CCODES,
                  "vcmp.f32", "!0s, !1s", 4, kFixupNone),
     ENCODING_MAP(kThumb2LdrPcRel12,       0xf8df0000,
                  kFmtBitBlt, 15, 12, kFmtBitBlt, 11, 0, kFmtUnused, -1, -1,
diff --git a/compiler/dex/quick/arm/int_arm.cc b/compiler/dex/quick/arm/int_arm.cc
index 3159886..2a4d27b 100644
--- a/compiler/dex/quick/arm/int_arm.cc
+++ b/compiler/dex/quick/arm/int_arm.cc
@@ -1079,6 +1079,7 @@
 }
 
 LIR* ArmMir2Lir::OpPcRelLoad(RegStorage reg, LIR* target) {
+  ScopedMemRefType mem_ref_type(this, ResourceMask::kLiteral);
   return RawLIR(current_dalvik_offset_, kThumb2LdrPcRel12, reg.GetReg(), 0, 0, 0, 0, target);
 }
 
diff --git a/compiler/dex/quick/arm64/assemble_arm64.cc b/compiler/dex/quick/arm64/assemble_arm64.cc
index 806617b..aa5e5b4 100644
--- a/compiler/dex/quick/arm64/assemble_arm64.cc
+++ b/compiler/dex/quick/arm64/assemble_arm64.cc
@@ -111,7 +111,7 @@
 const A64EncodingMap Arm64Mir2Lir::EncodingMap[kA64Last] = {
     ENCODING_MAP(WIDE(kA64Adc3rrr), SF_VARIANTS(0x1a000000),
                  kFmtRegR, 4, 0, kFmtRegR, 9, 5, kFmtRegR, 20, 16,
-                 kFmtUnused, -1, -1, IS_TERTIARY_OP | REG_DEF0_USE12,
+                 kFmtUnused, -1, -1, IS_TERTIARY_OP | REG_DEF0_USE12 | USES_CCODES,
                  "adc", "!0r, !1r, !2r", kFixupNone),
     ENCODING_MAP(WIDE(kA64Add4RRdT), SF_VARIANTS(0x11000000),
                  kFmtRegROrSp, 4, 0, kFmtRegROrSp, 9, 5, kFmtBitBlt, 21, 10,
@@ -518,7 +518,7 @@
                  "ror", "!0r, !1r, !2r", kFixupNone),
     ENCODING_MAP(WIDE(kA64Sbc3rrr), SF_VARIANTS(0x5a000000),
                  kFmtRegR, 4, 0, kFmtRegR, 9, 5, kFmtRegR, 20, 16,
-                 kFmtUnused, -1, -1, IS_TERTIARY_OP | REG_DEF0_USE12,
+                 kFmtUnused, -1, -1, IS_TERTIARY_OP | REG_DEF0_USE12 | USES_CCODES,
                  "sbc", "!0r, !1r, !2r", kFixupNone),
     ENCODING_MAP(WIDE(kA64Sbfm4rrdd), SF_N_VARIANTS(0x13000000),
                  kFmtRegR, 4, 0, kFmtRegR, 9, 5, kFmtBitBlt, 21, 16,
diff --git a/compiler/dex/quick/dex_file_method_inliner.cc b/compiler/dex/quick/dex_file_method_inliner.cc
index 7245853..8e3f4ef 100644
--- a/compiler/dex/quick/dex_file_method_inliner.cc
+++ b/compiler/dex/quick/dex_file_method_inliner.cc
@@ -427,7 +427,7 @@
   InlineMethod intrinsic;
   {
     ReaderMutexLock mu(Thread::Current(), lock_);
-    auto it = inline_methods_.find(info->index);
+    auto it = inline_methods_.find(info->method_ref.dex_method_index);
     if (it == inline_methods_.end() || (it->second.flags & kInlineIntrinsic) == 0) {
       return false;
     }
@@ -718,7 +718,7 @@
     if (PrettyMethod(method_idx, *dex_file_) == "int java.lang.String.length()") {
       // TODO: String.length is both kIntrinsicIsEmptyOrLength and kInlineOpIGet.
     } else {
-      LOG(ERROR) << "Inliner: " << PrettyMethod(method_idx, *dex_file_) << " already inline";
+      LOG(WARNING) << "Inliner: " << PrettyMethod(method_idx, *dex_file_) << " already inline";
     }
     return false;
   }
diff --git a/compiler/dex/quick/gen_common.cc b/compiler/dex/quick/gen_common.cc
index 3c9b7a3..afae89d 100644
--- a/compiler/dex/quick/gen_common.cc
+++ b/compiler/dex/quick/gen_common.cc
@@ -865,7 +865,12 @@
 void Mir2Lir::GenIGet(MIR* mir, int opt_flags, OpSize size, Primitive::Type type,
                       RegLocation rl_dest, RegLocation rl_obj) {
   const MirIFieldLoweringInfo& field_info = mir_graph_->GetIFieldLoweringInfo(mir);
-  DCHECK_EQ(IGetMemAccessType(mir->dalvikInsn.opcode), field_info.MemAccessType());
+  if (kIsDebugBuild) {
+    auto mem_access_type = IsInstructionIGetQuickOrIPutQuick(mir->dalvikInsn.opcode) ?
+        IGetQuickOrIPutQuickMemAccessType(mir->dalvikInsn.opcode) :
+        IGetMemAccessType(mir->dalvikInsn.opcode);
+    DCHECK_EQ(mem_access_type, field_info.MemAccessType()) << mir->dalvikInsn.opcode;
+  }
   cu_->compiler_driver->ProcessedInstanceField(field_info.FastGet());
   if (!ForceSlowFieldPath(cu_) && field_info.FastGet()) {
     RegisterClass reg_class = RegClassForFieldLoadStore(size, field_info.IsVolatile());
@@ -939,7 +944,12 @@
 void Mir2Lir::GenIPut(MIR* mir, int opt_flags, OpSize size,
                       RegLocation rl_src, RegLocation rl_obj) {
   const MirIFieldLoweringInfo& field_info = mir_graph_->GetIFieldLoweringInfo(mir);
-  DCHECK_EQ(IPutMemAccessType(mir->dalvikInsn.opcode), field_info.MemAccessType());
+  if (kIsDebugBuild) {
+    auto mem_access_type = IsInstructionIGetQuickOrIPutQuick(mir->dalvikInsn.opcode) ?
+        IGetQuickOrIPutQuickMemAccessType(mir->dalvikInsn.opcode) :
+        IPutMemAccessType(mir->dalvikInsn.opcode);
+    DCHECK_EQ(mem_access_type, field_info.MemAccessType());
+  }
   cu_->compiler_driver->ProcessedInstanceField(field_info.FastPut());
   if (!ForceSlowFieldPath(cu_) && field_info.FastPut()) {
     RegisterClass reg_class = RegClassForFieldLoadStore(size, field_info.IsVolatile());
diff --git a/compiler/dex/quick/gen_invoke.cc b/compiler/dex/quick/gen_invoke.cc
index 8e3df7c..01f1d37 100755
--- a/compiler/dex/quick/gen_invoke.cc
+++ b/compiler/dex/quick/gen_invoke.cc
@@ -248,14 +248,16 @@
     if (arg0.wide == 0) {
       LoadValueDirectFixed(arg0, TargetReg(arg0.fp ? kFArg0 : kArg0, kNotWide));
       if (arg1.wide == 0) {
+        // For Mips, when the 1st arg is integral, then remaining arg are passed in core reg.
         if (cu_->instruction_set == kMips) {
-          LoadValueDirectFixed(arg1, TargetReg(arg1.fp ? kFArg2 : kArg1, kNotWide));
+          LoadValueDirectFixed(arg1, TargetReg((arg1.fp && arg0.fp) ? kFArg2 : kArg1, kNotWide));
         } else {
           LoadValueDirectFixed(arg1, TargetReg(arg1.fp ? kFArg1 : kArg1, kNotWide));
         }
       } else {
+        // For Mips, when the 1st arg is integral, then remaining arg are passed in core reg.
         if (cu_->instruction_set == kMips) {
-          LoadValueDirectWideFixed(arg1, TargetReg(arg1.fp ? kFArg2 : kArg2, kWide));
+          LoadValueDirectWideFixed(arg1, TargetReg((arg1.fp && arg0.fp) ? kFArg2 : kArg2, kWide));
         } else {
           LoadValueDirectWideFixed(arg1, TargetReg(arg1.fp ? kFArg1 : kArg1, kWide));
         }
@@ -263,9 +265,19 @@
     } else {
       LoadValueDirectWideFixed(arg0, TargetReg(arg0.fp ? kFArg0 : kArg0, kWide));
       if (arg1.wide == 0) {
-        LoadValueDirectFixed(arg1, TargetReg(arg1.fp ? kFArg2 : kArg2, kNotWide));
+        // For Mips, when the 1st arg is integral, then remaining arg are passed in core reg.
+        if (cu_->instruction_set == kMips) {
+          LoadValueDirectFixed(arg1, TargetReg((arg1.fp && arg0.fp) ? kFArg2 : kArg2, kNotWide));
+        } else {
+          LoadValueDirectFixed(arg1, TargetReg(arg1.fp ? kFArg2 : kArg2, kNotWide));
+        }
       } else {
-        LoadValueDirectWideFixed(arg1, TargetReg(arg1.fp ? kFArg2 : kArg2, kWide));
+        // For Mips, when the 1st arg is integral, then remaining arg are passed in core reg.
+        if (cu_->instruction_set == kMips) {
+          LoadValueDirectWideFixed(arg1, TargetReg((arg1.fp && arg0.fp) ? kFArg2 : kArg2, kWide));
+        } else {
+          LoadValueDirectWideFixed(arg1, TargetReg(arg1.fp ? kFArg2 : kArg2, kWide));
+        }
       }
     }
   }
@@ -863,11 +875,12 @@
   RegLocation res;
   if (info->result.location == kLocInvalid) {
     // If result is unused, return a sink target based on type of invoke target.
-    res = GetReturn(ShortyToRegClass(mir_graph_->GetShortyFromTargetIdx(info->index)[0]));
+    res = GetReturn(
+        ShortyToRegClass(mir_graph_->GetShortyFromMethodReference(info->method_ref)[0]));
   } else {
     res = info->result;
     DCHECK_EQ(LocToRegClass(res),
-              ShortyToRegClass(mir_graph_->GetShortyFromTargetIdx(info->index)[0]));
+              ShortyToRegClass(mir_graph_->GetShortyFromMethodReference(info->method_ref)[0]));
   }
   return res;
 }
@@ -876,11 +889,12 @@
   RegLocation res;
   if (info->result.location == kLocInvalid) {
     // If result is unused, return a sink target based on type of invoke target.
-    res = GetReturnWide(ShortyToRegClass(mir_graph_->GetShortyFromTargetIdx(info->index)[0]));
+    res = GetReturnWide(ShortyToRegClass(
+        mir_graph_->GetShortyFromMethodReference(info->method_ref)[0]));
   } else {
     res = info->result;
     DCHECK_EQ(LocToRegClass(res),
-              ShortyToRegClass(mir_graph_->GetShortyFromTargetIdx(info->index)[0]));
+              ShortyToRegClass(mir_graph_->GetShortyFromMethodReference(info->method_ref)[0]));
   }
   return res;
 }
@@ -1418,7 +1432,8 @@
 
 void Mir2Lir::GenInvoke(CallInfo* info) {
   DCHECK(cu_->compiler_driver->GetMethodInlinerMap() != nullptr);
-  if (cu_->compiler_driver->GetMethodInlinerMap()->GetMethodInliner(cu_->dex_file)
+  const DexFile* dex_file = info->method_ref.dex_file;
+  if (cu_->compiler_driver->GetMethodInlinerMap()->GetMethodInliner(dex_file)
       ->GenIntrinsic(this, info)) {
     return;
   }
@@ -1428,7 +1443,7 @@
 void Mir2Lir::GenInvokeNoInline(CallInfo* info) {
   int call_state = 0;
   LIR* null_ck;
-  LIR** p_null_ck = NULL;
+  LIR** p_null_ck = nullptr;
   NextCallInsn next_call_insn;
   FlushAllRegs();  /* Everything to home location */
   // Explicit register usage
@@ -1440,6 +1455,7 @@
   info->type = method_info.GetSharpType();
   bool fast_path = method_info.FastPath();
   bool skip_this;
+
   if (info->type == kInterface) {
     next_call_insn = fast_path ? NextInterfaceCallInsn : NextInterfaceCallInsnWithAccessCheck;
     skip_this = fast_path;
@@ -1469,7 +1485,8 @@
   // Finish up any of the call sequence not interleaved in arg loading
   while (call_state >= 0) {
     call_state = next_call_insn(cu_, info, call_state, target_method, method_info.VTableIndex(),
-                                method_info.DirectCode(), method_info.DirectMethod(), original_type);
+                                method_info.DirectCode(), method_info.DirectMethod(),
+                                original_type);
   }
   LIR* call_insn = GenCallInsn(method_info);
   MarkSafepointPC(call_insn);
diff --git a/compiler/dex/quick/mips/utility_mips.cc b/compiler/dex/quick/mips/utility_mips.cc
index ec6edab..2d26922 100644
--- a/compiler/dex/quick/mips/utility_mips.cc
+++ b/compiler/dex/quick/mips/utility_mips.cc
@@ -17,6 +17,7 @@
 #include "codegen_mips.h"
 
 #include "arch/mips/instruction_set_features_mips.h"
+#include "arch/mips/entrypoints_direct_mips.h"
 #include "base/logging.h"
 #include "dex/quick/mir_to_lir-inl.h"
 #include "dex/reg_storage_eq.h"
@@ -708,7 +709,18 @@
 }
 
 LIR* MipsMir2Lir::InvokeTrampoline(OpKind op, RegStorage r_tgt, QuickEntrypointEnum trampoline) {
-  UNUSED(trampoline);  // The address of the trampoline is already loaded into r_tgt.
+  if (IsDirectEntrypoint(trampoline)) {
+    // Reserve argument space on stack (for $a0-$a3) for
+    // entrypoints that directly reference native implementations.
+    // This is not safe in general, as it violates the frame size
+    // of the Quick method, but it is used here only for calling
+    // native functions, outside of the runtime.
+    OpRegImm(kOpSub, rs_rSP, 16);
+    LIR* retVal = OpReg(op, r_tgt);
+    OpRegImm(kOpAdd, rs_rSP, 16);
+    return retVal;
+  }
+
   return OpReg(op, r_tgt);
 }
 
diff --git a/compiler/dex/quick/mir_to_lir.cc b/compiler/dex/quick/mir_to_lir.cc
index 34e5e25..8348626 100644
--- a/compiler/dex/quick/mir_to_lir.cc
+++ b/compiler/dex/quick/mir_to_lir.cc
@@ -540,6 +540,7 @@
       GenMoveException(rl_dest);
       break;
 
+    case Instruction::RETURN_VOID_BARRIER:
     case Instruction::RETURN_VOID:
       if (((cu_->access_flags & kAccConstructor) != 0) &&
           cu_->compiler_driver->RequiresConstructorBarrier(Thread::Current(), cu_->dex_file,
@@ -586,6 +587,9 @@
     case Instruction::MOVE_FROM16:
     case Instruction::MOVE_OBJECT_FROM16:
       StoreValue(rl_dest, rl_src[0]);
+      if (rl_src[0].is_const && (mir_graph_->ConstantValue(rl_src[0]) == 0)) {
+        Workaround7250540(rl_dest, RegStorage::InvalidReg());
+      }
       break;
 
     case Instruction::MOVE_WIDE:
@@ -790,10 +794,12 @@
       GenArrayPut(opt_flags, kUnsignedByte, rl_src[1], rl_src[2], rl_src[0], 0, false);
       break;
 
+    case Instruction::IGET_OBJECT_QUICK:
     case Instruction::IGET_OBJECT:
       GenIGet(mir, opt_flags, kReference, Primitive::kPrimNot, rl_dest, rl_src[0]);
       break;
 
+    case Instruction::IGET_WIDE_QUICK:
     case Instruction::IGET_WIDE:
       // kPrimLong and kPrimDouble share the same entrypoints.
       if (rl_dest.fp) {
@@ -803,6 +809,7 @@
       }
       break;
 
+    case Instruction::IGET_QUICK:
     case Instruction::IGET:
       if (rl_dest.fp) {
         GenIGet(mir, opt_flags, kSingle, Primitive::kPrimFloat, rl_dest, rl_src[0]);
@@ -811,43 +818,54 @@
       }
       break;
 
+    case Instruction::IGET_CHAR_QUICK:
     case Instruction::IGET_CHAR:
       GenIGet(mir, opt_flags, kUnsignedHalf, Primitive::kPrimChar, rl_dest, rl_src[0]);
       break;
 
+    case Instruction::IGET_SHORT_QUICK:
     case Instruction::IGET_SHORT:
       GenIGet(mir, opt_flags, kSignedHalf, Primitive::kPrimShort, rl_dest, rl_src[0]);
       break;
 
+    case Instruction::IGET_BOOLEAN_QUICK:
     case Instruction::IGET_BOOLEAN:
       GenIGet(mir, opt_flags, kUnsignedByte, Primitive::kPrimBoolean, rl_dest, rl_src[0]);
       break;
 
+    case Instruction::IGET_BYTE_QUICK:
     case Instruction::IGET_BYTE:
       GenIGet(mir, opt_flags, kSignedByte, Primitive::kPrimByte, rl_dest, rl_src[0]);
       break;
 
+    case Instruction::IPUT_WIDE_QUICK:
     case Instruction::IPUT_WIDE:
       GenIPut(mir, opt_flags, rl_src[0].fp ? kDouble : k64, rl_src[0], rl_src[1]);
       break;
 
+    case Instruction::IPUT_OBJECT_QUICK:
     case Instruction::IPUT_OBJECT:
       GenIPut(mir, opt_flags, kReference, rl_src[0], rl_src[1]);
       break;
 
+    case Instruction::IPUT_QUICK:
     case Instruction::IPUT:
       GenIPut(mir, opt_flags, rl_src[0].fp ? kSingle : k32, rl_src[0], rl_src[1]);
       break;
 
+    case Instruction::IPUT_BYTE_QUICK:
+    case Instruction::IPUT_BOOLEAN_QUICK:
     case Instruction::IPUT_BYTE:
     case Instruction::IPUT_BOOLEAN:
       GenIPut(mir, opt_flags, kUnsignedByte, rl_src[0], rl_src[1]);
       break;
 
+    case Instruction::IPUT_CHAR_QUICK:
     case Instruction::IPUT_CHAR:
       GenIPut(mir, opt_flags, kUnsignedHalf, rl_src[0], rl_src[1]);
       break;
 
+    case Instruction::IPUT_SHORT_QUICK:
     case Instruction::IPUT_SHORT:
       GenIPut(mir, opt_flags, kSignedHalf, rl_src[0], rl_src[1]);
       break;
@@ -921,9 +939,12 @@
       GenInvoke(mir_graph_->NewMemCallInfo(bb, mir, kDirect, true));
       break;
 
+    case Instruction::INVOKE_VIRTUAL_QUICK:
     case Instruction::INVOKE_VIRTUAL:
       GenInvoke(mir_graph_->NewMemCallInfo(bb, mir, kVirtual, false));
       break;
+
+    case Instruction::INVOKE_VIRTUAL_RANGE_QUICK:
     case Instruction::INVOKE_VIRTUAL_RANGE:
       GenInvoke(mir_graph_->NewMemCallInfo(bb, mir, kVirtual, true));
       break;
diff --git a/compiler/dex/quick/quick_compiler.cc b/compiler/dex/quick/quick_compiler.cc
index 19c2a5a..fcf4716 100644
--- a/compiler/dex/quick/quick_compiler.cc
+++ b/compiler/dex/quick/quick_compiler.cc
@@ -542,6 +542,11 @@
 void QuickCompiler::InitCompilationUnit(CompilationUnit& cu) const {
   // Disable optimizations according to instruction set.
   cu.disable_opt |= kDisabledOptimizationsPerISA[cu.instruction_set];
+  if (Runtime::Current()->UseJit()) {
+    // Disable these optimizations for JIT until quickened byte codes are done being implemented.
+    // TODO: Find a cleaner way to do this.
+    cu.disable_opt |= 1u << kLocalValueNumbering;
+  }
 }
 
 void QuickCompiler::Init() {
diff --git a/compiler/dex/verification_results.cc b/compiler/dex/verification_results.cc
index 4ff173d..150bdac 100644
--- a/compiler/dex/verification_results.cc
+++ b/compiler/dex/verification_results.cc
@@ -66,11 +66,16 @@
     // 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);
-    DCHECK_EQ(it->second->GetDevirtMap().size(), verified_method->GetDevirtMap().size());
-    DCHECK_EQ(it->second->GetSafeCastSet().size(), verified_method->GetSafeCastSet().size());
+    if (!Runtime::Current()->UseJit()) {
+      DCHECK_EQ(it->second->GetDevirtMap().size(), verified_method->GetDevirtMap().size());
+      DCHECK_EQ(it->second->GetSafeCastSet().size(), verified_method->GetSafeCastSet().size());
+    }
     DCHECK_EQ(it->second->GetDexGcMap().size(), verified_method->GetDexGcMap().size());
-    delete it->second;
-    verified_methods_.erase(it);
+    // Delete the new verified method since there was already an existing one registered. It
+    // is unsafe to replace the existing one since the JIT may be using it to generate a
+    // native GC map.
+    delete verified_method;
+    return true;
   }
   verified_methods_.Put(ref, verified_method);
   DCHECK(verified_methods_.find(ref) != verified_methods_.end());
diff --git a/compiler/dex/verified_method.cc b/compiler/dex/verified_method.cc
index 21e965d..42d66be 100644
--- a/compiler/dex/verified_method.cc
+++ b/compiler/dex/verified_method.cc
@@ -24,6 +24,7 @@
 #include "base/stl_util.h"
 #include "dex_file.h"
 #include "dex_instruction-inl.h"
+#include "dex_instruction_utils.h"
 #include "mirror/art_method-inl.h"
 #include "mirror/class-inl.h"
 #include "mirror/dex_cache-inl.h"
@@ -52,6 +53,11 @@
     if (method_verifier->HasVirtualOrInterfaceInvokes()) {
       verified_method->GenerateDevirtMap(method_verifier);
     }
+
+    // Only need dequicken info for JIT so far.
+    if (Runtime::Current()->UseJit()) {
+      verified_method->GenerateDequickenMap(method_verifier);
+    }
   }
 
   if (method_verifier->HasCheckCasts()) {
@@ -65,6 +71,12 @@
   return (it != devirt_map_.end()) ? &it->second : nullptr;
 }
 
+const DexFileReference* VerifiedMethod::GetDequickenIndex(uint32_t dex_pc) const {
+  DCHECK(Runtime::Current()->UseJit());
+  auto it = dequicken_map_.find(dex_pc);
+  return (it != dequicken_map_.end()) ? &it->second : nullptr;
+}
+
 bool VerifiedMethod::IsSafeCast(uint32_t pc) const {
   return std::binary_search(safe_cast_set_.begin(), safe_cast_set_.end(), pc);
 }
@@ -182,7 +194,7 @@
   *log2_max_gc_pc = i;
 }
 
-void VerifiedMethod::GenerateDeQuickenMap(verifier::MethodVerifier* method_verifier) {
+void VerifiedMethod::GenerateDequickenMap(verifier::MethodVerifier* method_verifier) {
   if (method_verifier->HasFailures()) {
     return;
   }
@@ -196,13 +208,24 @@
     if (is_virtual_quick || is_range_quick) {
       uint32_t dex_pc = inst->GetDexPc(insns);
       verifier::RegisterLine* line = method_verifier->GetRegLine(dex_pc);
-      mirror::ArtMethod* method = method_verifier->GetQuickInvokedMethod(inst, line,
-                                                                         is_range_quick);
+      mirror::ArtMethod* method =
+          method_verifier->GetQuickInvokedMethod(inst, line, is_range_quick);
       CHECK(method != nullptr);
       // The verifier must know what the type of the object was or else we would have gotten a
       // failure. Put the dex method index in the dequicken map since we need this to get number of
       // arguments in the compiler.
-      dequicken_map_.Put(dex_pc, method->ToMethodReference());
+      dequicken_map_.Put(dex_pc, DexFileReference(method->GetDexFile(),
+                                                  method->GetDexMethodIndex()));
+    } else if (IsInstructionIGetQuickOrIPutQuick(inst->Opcode())) {
+      uint32_t dex_pc = inst->GetDexPc(insns);
+      verifier::RegisterLine* line = method_verifier->GetRegLine(dex_pc);
+      mirror::ArtField* field = method_verifier->GetQuickFieldAccess(inst, line);
+      CHECK(field != nullptr);
+      // The verifier must know what the type of the field was or else we would have gotten a
+      // failure. Put the dex field index in the dequicken map since we need this for lowering
+      // in the compiler.
+      // TODO: Putting a field index in a method reference is gross.
+      dequicken_map_.Put(dex_pc, DexFileReference(field->GetDexFile(), field->GetDexFieldIndex()));
     }
   }
 }
diff --git a/compiler/dex/verified_method.h b/compiler/dex/verified_method.h
index fe9dfd1..748bdcb 100644
--- a/compiler/dex/verified_method.h
+++ b/compiler/dex/verified_method.h
@@ -20,6 +20,7 @@
 #include <vector>
 
 #include "base/mutex.h"
+#include "dex_file.h"
 #include "method_reference.h"
 #include "safe_map.h"
 
@@ -39,6 +40,9 @@
   // Devirtualization map type maps dex offset to concrete method reference.
   typedef SafeMap<uint32_t, MethodReference> DevirtualizationMap;
 
+  // Devirtualization map type maps dex offset to field / method idx.
+  typedef SafeMap<uint32_t, DexFileReference> DequickenMap;
+
   static const VerifiedMethod* Create(verifier::MethodVerifier* method_verifier, bool compile)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
   ~VerifiedMethod() = default;
@@ -58,6 +62,10 @@
   // Returns the devirtualization target method, or nullptr if none.
   const MethodReference* GetDevirtTarget(uint32_t dex_pc) const;
 
+  // Returns the dequicken field / method for a quick invoke / field get. Returns null if there is
+  // no entry for that dex pc.
+  const DexFileReference* GetDequickenIndex(uint32_t dex_pc) const;
+
   // Returns true if the cast can statically be verified to be redundant
   // by using the check-cast elision peephole optimization in the verifier.
   bool IsSafeCast(uint32_t pc) const;
@@ -86,7 +94,7 @@
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   // Generate dequickening map into dequicken_map_.
-  void GenerateDeQuickenMap(verifier::MethodVerifier* method_verifier)
+  void GenerateDequickenMap(verifier::MethodVerifier* method_verifier)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   // Generate safe case set into safe_cast_set_.
@@ -95,9 +103,9 @@
 
   std::vector<uint8_t> dex_gc_map_;
   DevirtualizationMap devirt_map_;
-  // Dequicken map is required for having the compiler compiled quickened invokes. The quicken map
-  // enables us to get the dex method index so that we can get the required argument count.
-  DevirtualizationMap dequicken_map_;
+  // Dequicken map is required for compiling quickened byte codes. The quicken maps from
+  // dex PC to dex method index or dex field index based on the instruction.
+  DequickenMap dequicken_map_;
   SafeCastSet safe_cast_set_;
 };
 
diff --git a/compiler/dex/vreg_analysis.cc b/compiler/dex/vreg_analysis.cc
index b620969..2b78e38 100644
--- a/compiler/dex/vreg_analysis.cc
+++ b/compiler/dex/vreg_analysis.cc
@@ -19,6 +19,7 @@
 #include "compiler_ir.h"
 #include "dex/dataflow_iterator-inl.h"
 #include "dex_flags.h"
+#include "driver/dex_compilation_unit.h"
 
 namespace art {
 
@@ -259,8 +260,8 @@
     if ((flags & Instruction::kInvoke) &&
         (attrs & (DF_FORMAT_35C | DF_FORMAT_3RC))) {
       DCHECK_EQ(next, 0);
-      int target_idx = mir->dalvikInsn.vB;
-      const char* shorty = GetShortyFromTargetIdx(target_idx);
+      const auto& lowering_info = GetMethodLoweringInfo(mir);
+      const char* shorty = GetShortyFromMethodReference(lowering_info.GetTargetMethod());
       // Handle result type if floating point
       if ((shorty[0] == 'F') || (shorty[0] == 'D')) {
         MIR* move_result_mir = FindMoveResult(bb, mir);
diff --git a/compiler/driver/compiler_driver-inl.h b/compiler/driver/compiler_driver-inl.h
index 9948c82..4a35e9f 100644
--- a/compiler/driver/compiler_driver-inl.h
+++ b/compiler/driver/compiler_driver-inl.h
@@ -56,14 +56,13 @@
   return referrer_class;
 }
 
-inline mirror::ArtField* CompilerDriver::ResolveField(
+inline mirror::ArtField* CompilerDriver::ResolveFieldWithDexFile(
     const ScopedObjectAccess& soa, Handle<mirror::DexCache> dex_cache,
-    Handle<mirror::ClassLoader> class_loader, const DexCompilationUnit* mUnit,
+    Handle<mirror::ClassLoader> class_loader, const DexFile* dex_file,
     uint32_t field_idx, bool is_static) {
-  DCHECK_EQ(dex_cache->GetDexFile(), mUnit->GetDexFile());
-  DCHECK_EQ(class_loader.Get(), soa.Decode<mirror::ClassLoader*>(mUnit->GetClassLoader()));
-  mirror::ArtField* resolved_field = mUnit->GetClassLinker()->ResolveField(
-      *mUnit->GetDexFile(), field_idx, dex_cache, class_loader, is_static);
+  DCHECK_EQ(dex_cache->GetDexFile(), dex_file);
+  mirror::ArtField* resolved_field = Runtime::Current()->GetClassLinker()->ResolveField(
+      *dex_file, field_idx, dex_cache, class_loader, is_static);
   DCHECK_EQ(resolved_field == nullptr, soa.Self()->IsExceptionPending());
   if (UNLIKELY(resolved_field == nullptr)) {
     // Clean up any exception left by type resolution.
@@ -78,6 +77,19 @@
   return resolved_field;
 }
 
+inline mirror::DexCache* CompilerDriver::FindDexCache(const DexFile* dex_file) {
+  return Runtime::Current()->GetClassLinker()->FindDexCache(*dex_file);
+}
+
+inline mirror::ArtField* CompilerDriver::ResolveField(
+    const ScopedObjectAccess& soa, Handle<mirror::DexCache> dex_cache,
+    Handle<mirror::ClassLoader> class_loader, const DexCompilationUnit* mUnit,
+    uint32_t field_idx, bool is_static) {
+  DCHECK_EQ(class_loader.Get(), soa.Decode<mirror::ClassLoader*>(mUnit->GetClassLoader()));
+  return ResolveFieldWithDexFile(soa, dex_cache, class_loader, mUnit->GetDexFile(), field_idx,
+                                 is_static);
+}
+
 inline void CompilerDriver::GetResolvedFieldDexFileLocation(
     mirror::ArtField* resolved_field, const DexFile** declaring_dex_file,
     uint16_t* declaring_class_idx, uint16_t* declaring_field_idx) {
@@ -172,7 +184,7 @@
 inline mirror::ArtMethod* CompilerDriver::ResolveMethod(
     ScopedObjectAccess& soa, Handle<mirror::DexCache> dex_cache,
     Handle<mirror::ClassLoader> class_loader, const DexCompilationUnit* mUnit,
-    uint32_t method_idx, InvokeType invoke_type) {
+    uint32_t method_idx, InvokeType invoke_type, bool check_incompatible_class_change) {
   DCHECK_EQ(dex_cache->GetDexFile(), mUnit->GetDexFile());
   DCHECK_EQ(class_loader.Get(), soa.Decode<mirror::ClassLoader*>(mUnit->GetClassLoader()));
   mirror::ArtMethod* resolved_method = mUnit->GetClassLinker()->ResolveMethod(
@@ -184,7 +196,8 @@
     soa.Self()->ClearException();
     return nullptr;
   }
-  if (UNLIKELY(resolved_method->CheckIncompatibleClassChange(invoke_type))) {
+  if (check_incompatible_class_change &&
+      UNLIKELY(resolved_method->CheckIncompatibleClassChange(invoke_type))) {
     // Silently return nullptr on incompatible class change.
     return nullptr;
   }
@@ -227,14 +240,14 @@
                                                         target_method->dex_method_index))) {
     return 0;
   }
-
   // Sharpen a virtual call into a direct call when the target is known not to have been
   // overridden (ie is final).
-  bool can_sharpen_virtual_based_on_type =
+  const bool same_dex_file = target_method->dex_file == mUnit->GetDexFile();
+  bool can_sharpen_virtual_based_on_type = same_dex_file &&
       (*invoke_type == kVirtual) && (resolved_method->IsFinal() || methods_class->IsFinal());
   // For invoke-super, ensure the vtable index will be correct to dispatch in the vtable of
   // the super class.
-  bool can_sharpen_super_based_on_type = (*invoke_type == kSuper) &&
+  bool can_sharpen_super_based_on_type = same_dex_file && (*invoke_type == kSuper) &&
       (referrer_class != methods_class) && referrer_class->IsSubClass(methods_class) &&
       resolved_method->GetMethodIndex() < methods_class->GetVTableLength() &&
       (methods_class->GetVTableEntry(resolved_method->GetMethodIndex()) == resolved_method) &&
@@ -243,10 +256,10 @@
   if (can_sharpen_virtual_based_on_type || can_sharpen_super_based_on_type) {
     // Sharpen a virtual call into a direct call. The method_idx is into referrer's
     // dex cache, check that this resolved method is where we expect it.
-    CHECK(target_method->dex_file == mUnit->GetDexFile());
-    DCHECK(dex_cache.Get() == mUnit->GetClassLinker()->FindDexCache(*mUnit->GetDexFile()));
-    CHECK(referrer_class->GetDexCache()->GetResolvedMethod(target_method->dex_method_index) ==
-        resolved_method) << PrettyMethod(resolved_method);
+    CHECK_EQ(target_method->dex_file, mUnit->GetDexFile());
+    DCHECK_EQ(dex_cache.Get(), mUnit->GetClassLinker()->FindDexCache(*mUnit->GetDexFile()));
+    CHECK_EQ(referrer_class->GetDexCache()->GetResolvedMethod(target_method->dex_method_index),
+             resolved_method) << PrettyMethod(resolved_method);
     int stats_flags = kFlagMethodResolved;
     GetCodeAndMethodForDirectCall(/*out*/invoke_type,
                                   kDirect,  // Sharp type
diff --git a/compiler/driver/compiler_driver.cc b/compiler/driver/compiler_driver.cc
index b8a8936..029fd46 100644
--- a/compiler/driver/compiler_driver.cc
+++ b/compiler/driver/compiler_driver.cc
@@ -360,6 +360,7 @@
       classes_to_compile_(compiled_classes),
       thread_count_(thread_count),
       stats_(new AOTCompilationStats),
+      dedupe_enabled_(true),
       dump_stats_(dump_stats),
       dump_passes_(dump_passes),
       dump_cfg_file_name_(dump_cfg_file_name),
@@ -380,12 +381,7 @@
 
   compiler_->Init();
 
-  CHECK(!Runtime::Current()->IsStarted());
-  if (image_) {
-    CHECK(image_classes_.get() != nullptr);
-  } else {
-    CHECK(image_classes_.get() == nullptr);
-  }
+  CHECK_EQ(image_, image_classes_.get() != nullptr);
 
   // Read the profile file if one is provided.
   if (!profile_file.empty()) {
@@ -399,26 +395,32 @@
 }
 
 SwapVector<uint8_t>* CompilerDriver::DeduplicateCode(const ArrayRef<const uint8_t>& code) {
+  DCHECK(dedupe_enabled_);
   return dedupe_code_.Add(Thread::Current(), code);
 }
 
 SwapSrcMap* CompilerDriver::DeduplicateSrcMappingTable(const ArrayRef<SrcMapElem>& src_map) {
+  DCHECK(dedupe_enabled_);
   return dedupe_src_mapping_table_.Add(Thread::Current(), src_map);
 }
 
 SwapVector<uint8_t>* CompilerDriver::DeduplicateMappingTable(const ArrayRef<const uint8_t>& code) {
+  DCHECK(dedupe_enabled_);
   return dedupe_mapping_table_.Add(Thread::Current(), code);
 }
 
 SwapVector<uint8_t>* CompilerDriver::DeduplicateVMapTable(const ArrayRef<const uint8_t>& code) {
+  DCHECK(dedupe_enabled_);
   return dedupe_vmap_table_.Add(Thread::Current(), code);
 }
 
 SwapVector<uint8_t>* CompilerDriver::DeduplicateGCMap(const ArrayRef<const uint8_t>& code) {
+  DCHECK(dedupe_enabled_);
   return dedupe_gc_map_.Add(Thread::Current(), code);
 }
 
 SwapVector<uint8_t>* CompilerDriver::DeduplicateCFIInfo(const ArrayRef<const uint8_t>& cfi_info) {
+  DCHECK(dedupe_enabled_);
   return dedupe_cfi_info_.Add(Thread::Current(), cfi_info);
 }
 
@@ -491,8 +493,12 @@
 static DexToDexCompilationLevel GetDexToDexCompilationlevel(
     Thread* self, Handle<mirror::ClassLoader> class_loader, const DexFile& dex_file,
     const DexFile::ClassDef& class_def) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+  auto* const runtime = Runtime::Current();
+  if (runtime->UseJit()) {
+    return kDontDexToDexCompile;
+  }
   const char* descriptor = dex_file.GetClassDescriptor(class_def);
-  ClassLinker* class_linker = Runtime::Current()->GetClassLinker();
+  ClassLinker* class_linker = runtime->GetClassLinker();
   mirror::Class* klass = class_linker->FindClass(self, descriptor, class_loader);
   if (klass == nullptr) {
     CHECK(self->IsExceptionPending());
@@ -518,9 +524,8 @@
   }
 }
 
-void CompilerDriver::CompileOne(mirror::ArtMethod* method, TimingLogger* timings) {
+void CompilerDriver::CompileOne(Thread* self, mirror::ArtMethod* method, TimingLogger* timings) {
   DCHECK(!Runtime::Current()->IsStarted());
-  Thread* self = Thread::Current();
   jobject jclass_loader;
   const DexFile* dex_file;
   uint16_t class_def_idx;
@@ -529,9 +534,8 @@
   InvokeType invoke_type = method->GetInvokeType();
   {
     ScopedObjectAccessUnchecked soa(self);
-    ScopedLocalRef<jobject>
-      local_class_loader(soa.Env(),
-                    soa.AddLocalReference<jobject>(method->GetDeclaringClass()->GetClassLoader()));
+    ScopedLocalRef<jobject> local_class_loader(
+        soa.Env(), soa.AddLocalReference<jobject>(method->GetDeclaringClass()->GetClassLoader()));
     jclass_loader = soa.Env()->NewGlobalRef(local_class_loader.get());
     // Find the dex_file
     dex_file = method->GetDexFile();
@@ -549,7 +553,7 @@
   // Can we run DEX-to-DEX compiler on this class ?
   DexToDexCompilationLevel dex_to_dex_compilation_level = kDontDexToDexCompile;
   {
-    ScopedObjectAccess soa(Thread::Current());
+    ScopedObjectAccess soa(self);
     const DexFile::ClassDef& class_def = dex_file->GetClassDef(class_def_idx);
     StackHandleScope<1> hs(soa.Self());
     Handle<mirror::ClassLoader> class_loader(
@@ -557,14 +561,35 @@
     dex_to_dex_compilation_level = GetDexToDexCompilationlevel(self, class_loader, *dex_file,
                                                                class_def);
   }
-  CompileMethod(code_item, access_flags, invoke_type, class_def_idx, method_idx, jclass_loader,
-                *dex_file, dex_to_dex_compilation_level, true);
+  CompileMethod(self, code_item, access_flags, invoke_type, class_def_idx, method_idx,
+                jclass_loader, *dex_file, dex_to_dex_compilation_level, true);
 
   self->GetJniEnv()->DeleteGlobalRef(jclass_loader);
-
   self->TransitionFromSuspendedToRunnable();
 }
 
+CompiledMethod* CompilerDriver::CompileMethod(Thread* self, mirror::ArtMethod* method) {
+  const uint32_t method_idx = method->GetDexMethodIndex();
+  const uint32_t access_flags = method->GetAccessFlags();
+  const InvokeType invoke_type = method->GetInvokeType();
+  StackHandleScope<1> hs(self);
+  Handle<mirror::ClassLoader> class_loader(hs.NewHandle(
+      method->GetDeclaringClass()->GetClassLoader()));
+  jobject jclass_loader = class_loader.ToJObject();
+  const DexFile* dex_file = method->GetDexFile();
+  const uint16_t class_def_idx = method->GetClassDefIndex();
+  const DexFile::ClassDef& class_def = dex_file->GetClassDef(class_def_idx);
+  DexToDexCompilationLevel dex_to_dex_compilation_level =
+      GetDexToDexCompilationlevel(self, class_loader, *dex_file, class_def);
+  const DexFile::CodeItem* code_item = dex_file->GetCodeItem(method->GetCodeItemOffset());
+  self->TransitionFromRunnableToSuspended(kNative);
+  CompileMethod(self, code_item, access_flags, invoke_type, class_def_idx, method_idx,
+                jclass_loader, *dex_file, dex_to_dex_compilation_level, true);
+  auto* compiled_method = GetCompiledMethod(MethodReference(dex_file, method_idx));
+  self->TransitionFromSuspendedToRunnable();
+  return compiled_method;
+}
+
 void CompilerDriver::Resolve(jobject class_loader, const std::vector<const DexFile*>& dex_files,
                              ThreadPool* thread_pool, TimingLogger* timings) {
   for (size_t i = 0; i != dex_files.size(); ++i) {
@@ -1035,7 +1060,8 @@
                                         bool* is_type_initialized, bool* use_direct_type_ptr,
                                         uintptr_t* direct_type_ptr, bool* out_is_finalizable) {
   ScopedObjectAccess soa(Thread::Current());
-  mirror::DexCache* dex_cache = Runtime::Current()->GetClassLinker()->FindDexCache(dex_file);
+  Runtime* runtime = Runtime::Current();
+  mirror::DexCache* dex_cache = runtime->GetClassLinker()->FindDexCache(dex_file);
   mirror::Class* resolved_class = dex_cache->GetResolvedType(type_idx);
   if (resolved_class == nullptr) {
     return false;
@@ -1045,7 +1071,8 @@
     return false;
   }
   *out_is_finalizable = resolved_class->IsFinalizable();
-  const bool compiling_boot = Runtime::Current()->GetHeap()->IsCompilingBoot();
+  gc::Heap* heap = runtime->GetHeap();
+  const bool compiling_boot = heap->IsCompilingBoot();
   const bool support_boot_image_fixup = GetSupportBootImageFixup();
   if (compiling_boot) {
     // boot -> boot class pointers.
@@ -1061,10 +1088,15 @@
     } else {
       return false;
     }
+  } else if (runtime->UseJit() && !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;
+    *direct_type_ptr = reinterpret_cast<uintptr_t>(resolved_class);
+    return true;
   } else {
     // True if the class is in the image at app compiling time.
-    const bool class_in_image =
-        Runtime::Current()->GetHeap()->FindSpaceFromObject(resolved_class, false)->IsImageSpace();
+    const bool class_in_image = heap->FindSpaceFromObject(resolved_class, false)->IsImageSpace();
     if (class_in_image && support_boot_image_fixup) {
       // boot -> app class pointers.
       *is_type_initialized = resolved_class->IsInitialized();
@@ -1257,8 +1289,10 @@
   // invoked, so this can be passed to the out-of-line runtime support code.
   *direct_code = 0;
   *direct_method = 0;
+  Runtime* const runtime = Runtime::Current();
+  gc::Heap* const heap = runtime->GetHeap();
   bool use_dex_cache = GetCompilerOptions().GetCompilePic();  // Off by default
-  const bool compiling_boot = Runtime::Current()->GetHeap()->IsCompilingBoot();
+  const bool compiling_boot = heap->IsCompilingBoot();
   // TODO This is somewhat hacky. We should refactor all of this invoke codepath.
   const bool force_relocations = (compiling_boot ||
                                   GetCompilerOptions().GetIncludePatchInformation());
@@ -1267,20 +1301,29 @@
   }
   // TODO: support patching on all architectures.
   use_dex_cache = use_dex_cache || (force_relocations && !support_boot_image_fixup_);
-  bool method_code_in_boot = (method->GetDeclaringClass()->GetClassLoader() == nullptr);
+  mirror::Class* declaring_class = method->GetDeclaringClass();
+  bool method_code_in_boot = declaring_class->GetClassLoader() == nullptr;
   if (!use_dex_cache) {
     if (!method_code_in_boot) {
       use_dex_cache = true;
     } else {
       bool has_clinit_trampoline =
-          method->IsStatic() && !method->GetDeclaringClass()->IsInitialized();
-      if (has_clinit_trampoline && (method->GetDeclaringClass() != referrer_class)) {
+          method->IsStatic() && !declaring_class->IsInitialized();
+      if (has_clinit_trampoline && declaring_class != referrer_class) {
         // Ensure we run the clinit trampoline unless we are invoking a static method in the same
         // class.
         use_dex_cache = true;
       }
     }
   }
+  if (runtime->UseJit()) {
+    // 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 (runtime->GetClassLinker()->IsQuickToInterpreterBridge(
+        reinterpret_cast<const void*>(compiler_->GetEntryPointOf(method)))) {
+      use_dex_cache = true;
+    }
+  }
   if (method_code_in_boot) {
     *stats_flags |= kFlagDirectCallToBoot | kFlagDirectMethodToBoot;
   }
@@ -1302,7 +1345,9 @@
   // The method is defined not within this dex file. We need a dex cache slot within the current
   // dex file or direct pointers.
   bool must_use_direct_pointers = false;
-  if (target_method->dex_file == method->GetDeclaringClass()->GetDexCache()->GetDexFile()) {
+  mirror::DexCache* dex_cache = declaring_class->GetDexCache();
+  if (target_method->dex_file == dex_cache->GetDexFile() &&
+    !(runtime->UseJit() && dex_cache->GetResolvedMethod(method->GetDexMethodIndex()) == nullptr)) {
     target_method->dex_method_index = method->GetDexMethodIndex();
   } else {
     if (no_guarantee_of_dex_cache_entry) {
@@ -1315,7 +1360,7 @@
       } else {
         if (force_relocations && !use_dex_cache) {
           target_method->dex_method_index = method->GetDexMethodIndex();
-          target_method->dex_file = method->GetDeclaringClass()->GetDexCache()->GetDexFile();
+          target_method->dex_file = dex_cache->GetDexFile();
         }
         must_use_direct_pointers = true;
       }
@@ -1330,8 +1375,7 @@
       *type = sharp_type;
     }
   } else {
-    bool method_in_image =
-        Runtime::Current()->GetHeap()->FindSpaceFromObject(method, false)->IsImageSpace();
+    bool method_in_image = heap->FindSpaceFromObject(method, false)->IsImageSpace();
     if (method_in_image || compiling_boot) {
       // We know we must be able to get to the method in the image, so use that pointer.
       CHECK(!method->IsAbstract());
@@ -1926,8 +1970,7 @@
 
             if (!success) {
               CHECK(soa.Self()->IsExceptionPending());
-              ThrowLocation throw_location;
-              mirror::Throwable* exception = soa.Self()->GetException(&throw_location);
+              mirror::Throwable* exception = soa.Self()->GetException(nullptr);
               VLOG(compiler) << "Initialization of " << descriptor << " aborted because of "
                   << exception->Dump();
               std::ostream* file_log = manager->GetCompiler()->
@@ -2000,10 +2043,11 @@
   const DexFile::ClassDef& class_def = dex_file.GetClassDef(class_def_index);
   ClassLinker* class_linker = manager->GetClassLinker();
   jobject jclass_loader = manager->GetClassLoader();
+  Thread* self = Thread::Current();
   {
     // Use a scoped object access to perform to the quick SkipClass check.
     const char* descriptor = dex_file.GetClassDescriptor(class_def);
-    ScopedObjectAccess soa(Thread::Current());
+    ScopedObjectAccess soa(self);
     StackHandleScope<3> hs(soa.Self());
     Handle<mirror::ClassLoader> class_loader(
         hs.NewHandle(soa.Decode<mirror::ClassLoader*>(jclass_loader)));
@@ -2030,7 +2074,7 @@
   // Can we run DEX-to-DEX compiler on this class ?
   DexToDexCompilationLevel dex_to_dex_compilation_level = kDontDexToDexCompile;
   {
-    ScopedObjectAccess soa(Thread::Current());
+    ScopedObjectAccess soa(self);
     StackHandleScope<1> hs(soa.Self());
     Handle<mirror::ClassLoader> class_loader(
         hs.NewHandle(soa.Decode<mirror::ClassLoader*>(jclass_loader)));
@@ -2061,7 +2105,7 @@
       continue;
     }
     previous_direct_method_idx = method_idx;
-    driver->CompileMethod(it.GetMethodCodeItem(), it.GetMethodAccessFlags(),
+    driver->CompileMethod(self, it.GetMethodCodeItem(), it.GetMethodAccessFlags(),
                           it.GetMethodInvokeType(class_def), class_def_index,
                           method_idx, jclass_loader, dex_file, dex_to_dex_compilation_level,
                           compilation_enabled);
@@ -2078,7 +2122,7 @@
       continue;
     }
     previous_virtual_method_idx = method_idx;
-    driver->CompileMethod(it.GetMethodCodeItem(), it.GetMethodAccessFlags(),
+    driver->CompileMethod(self, it.GetMethodCodeItem(), it.GetMethodAccessFlags(),
                           it.GetMethodInvokeType(class_def), class_def_index,
                           method_idx, jclass_loader, dex_file, dex_to_dex_compilation_level,
                           compilation_enabled);
@@ -2111,10 +2155,10 @@
   }
 }
 
-void CompilerDriver::CompileMethod(const DexFile::CodeItem* code_item, uint32_t access_flags,
-                                   InvokeType invoke_type, uint16_t class_def_idx,
-                                   uint32_t method_idx, jobject class_loader,
-                                   const DexFile& dex_file,
+void CompilerDriver::CompileMethod(Thread* self, const DexFile::CodeItem* code_item,
+                                   uint32_t access_flags, InvokeType invoke_type,
+                                   uint16_t class_def_idx, uint32_t method_idx,
+                                   jobject class_loader, const DexFile& dex_file,
                                    DexToDexCompilationLevel dex_to_dex_compilation_level,
                                    bool compilation_enabled) {
   CompiledMethod* compiled_method = nullptr;
@@ -2162,7 +2206,6 @@
     }
   }
 
-  Thread* self = Thread::Current();
   if (compiled_method != nullptr) {
     // Count non-relative linker patches.
     size_t non_relative_linker_patch_count = 0u;
@@ -2194,6 +2237,21 @@
   }
 }
 
+void CompilerDriver::RemoveCompiledMethod(const MethodReference& method_ref) {
+  CompiledMethod* compiled_method = nullptr;
+  {
+    MutexLock mu(Thread::Current(), compiled_methods_lock_);
+    auto it = compiled_methods_.find(method_ref);
+    if (it != compiled_methods_.end()) {
+      compiled_method = it->second;
+      compiled_methods_.erase(it);
+    }
+  }
+  if (compiled_method != nullptr) {
+    CompiledMethod::ReleaseSwapAllocatedCompiledMethod(this, compiled_method);
+  }
+}
+
 CompiledClass* CompilerDriver::GetCompiledClass(ClassReference ref) const {
   MutexLock mu(Thread::Current(), compiled_classes_lock_);
   ClassTable::const_iterator it = compiled_classes_.find(ref);
diff --git a/compiler/driver/compiler_driver.h b/compiler/driver/compiler_driver.h
index b756244..24b6f17 100644
--- a/compiler/driver/compiler_driver.h
+++ b/compiler/driver/compiler_driver.h
@@ -45,6 +45,10 @@
 
 namespace art {
 
+namespace mirror {
+class DexCache;
+}  // namespace mirror
+
 namespace verifier {
 class MethodVerifier;
 }  // namespace verifier
@@ -107,8 +111,11 @@
                   TimingLogger* timings)
       LOCKS_EXCLUDED(Locks::mutator_lock_);
 
+  CompiledMethod* CompileMethod(Thread* self, mirror::ArtMethod*)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) WARN_UNUSED;
+
   // Compile a single Method.
-  void CompileOne(mirror::ArtMethod* method, TimingLogger* timings)
+  void CompileOne(Thread* self, mirror::ArtMethod* method, TimingLogger* timings)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   VerificationResults* GetVerificationResults() const {
@@ -172,6 +179,9 @@
   size_t GetNonRelativeLinkerPatchCount() const
       LOCKS_EXCLUDED(compiled_methods_lock_);
 
+  // Remove and delete a compiled method.
+  void RemoveCompiledMethod(const MethodReference& method_ref);
+
   void AddRequiresConstructorBarrier(Thread* self, const DexFile* dex_file,
                                      uint16_t class_def_index);
   bool RequiresConstructorBarrier(Thread* self, const DexFile* dex_file, uint16_t class_def_index);
@@ -226,6 +236,13 @@
       uint32_t field_idx, bool is_static)
     SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
+  // Resolve a field with a given dex file.
+  mirror::ArtField* ResolveFieldWithDexFile(
+      const ScopedObjectAccess& soa, Handle<mirror::DexCache> dex_cache,
+      Handle<mirror::ClassLoader> class_loader, const DexFile* dex_file,
+      uint32_t field_idx, bool is_static)
+    SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
   // Get declaration location of a resolved field.
   void GetResolvedFieldDexFileLocation(
       mirror::ArtField* resolved_field, const DexFile** declaring_dex_file,
@@ -235,6 +252,10 @@
   bool IsFieldVolatile(mirror::ArtField* field) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
   MemberOffset GetFieldOffset(mirror::ArtField* field) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
+  // Find a dex cache for a dex file.
+  inline mirror::DexCache* FindDexCache(const DexFile* dex_file)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
   // Can we fast-path an IGET/IPUT access to an instance field? If yes, compute the field offset.
   std::pair<bool, bool> IsFastInstanceField(
       mirror::DexCache* dex_cache, mirror::Class* referrer_class,
@@ -261,7 +282,7 @@
   mirror::ArtMethod* ResolveMethod(
       ScopedObjectAccess& soa, Handle<mirror::DexCache> dex_cache,
       Handle<mirror::ClassLoader> class_loader, const DexCompilationUnit* mUnit,
-      uint32_t method_idx, InvokeType invoke_type)
+      uint32_t method_idx, InvokeType invoke_type, bool check_incompatible_class_change = true)
     SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   // Get declaration location of a resolved field.
@@ -295,6 +316,13 @@
   void ProcessedStaticField(bool resolved, bool local);
   void ProcessedInvoke(InvokeType invoke_type, int flags);
 
+  void ComputeFieldInfo(uint32_t field_idx, const DexCompilationUnit* mUnit,
+                        const ScopedObjectAccess& soa, bool is_static,
+                        mirror::ArtField** resolved_field,
+                        mirror::Class** referrer_class,
+                        mirror::DexCache** dex_cache)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
   // Can we fast path instance field access? Computes field's offset and volatility.
   bool ComputeInstanceFieldInfo(uint32_t field_idx, const DexCompilationUnit* mUnit, bool is_put,
                                 MemberOffset* field_offset, bool* is_volatile)
@@ -380,6 +408,13 @@
     return timings_logger_;
   }
 
+  void SetDedupeEnabled(bool dedupe_enabled) {
+    dedupe_enabled_ = dedupe_enabled;
+  }
+  bool DedupeEnabled() const {
+    return dedupe_enabled_;
+  }
+
   // Checks if class specified by type_idx is one of the image_classes_
   bool IsImageClass(const char* descriptor) const;
 
@@ -484,7 +519,7 @@
                       const std::vector<const DexFile*>& dex_files,
                       ThreadPool* thread_pool, TimingLogger* timings)
       LOCKS_EXCLUDED(Locks::mutator_lock_);
-  void CompileMethod(const DexFile::CodeItem* code_item, uint32_t access_flags,
+  void CompileMethod(Thread* self, const DexFile::CodeItem* code_item, uint32_t access_flags,
                      InvokeType invoke_type, uint16_t class_def_idx, uint32_t method_idx,
                      jobject class_loader, const DexFile& dex_file,
                      DexToDexCompilationLevel dex_to_dex_compilation_level,
@@ -545,6 +580,7 @@
   class AOTCompilationStats;
   std::unique_ptr<AOTCompilationStats> stats_;
 
+  bool dedupe_enabled_;
   bool dump_stats_;
   const bool dump_passes_;
   const std::string& dump_cfg_file_name_;
diff --git a/compiler/elf_builder.h b/compiler/elf_builder.h
index 94268de..9ab3602 100644
--- a/compiler/elf_builder.h
+++ b/compiler/elf_builder.h
@@ -538,6 +538,8 @@
              Elf_Word rodata_size,
              Elf_Word text_relative_offset,
              Elf_Word text_size,
+             Elf_Word bss_relative_offset,
+             Elf_Word bss_size,
              const bool add_symbols,
              bool debug = false)
     : oat_writer_(oat_writer),
@@ -547,6 +549,7 @@
       text_builder_(".text", text_size, text_relative_offset, SHT_PROGBITS,
                     SHF_ALLOC | SHF_EXECINSTR),
       rodata_builder_(".rodata", rodata_size, rodata_relative_offset, SHT_PROGBITS, SHF_ALLOC),
+      bss_builder_(".bss", bss_size, bss_relative_offset, SHT_NOBITS, SHF_ALLOC),
       dynsym_builder_(".dynsym", SHT_DYNSYM, ".dynstr", SHT_STRTAB, true),
       symtab_builder_(".symtab", SHT_SYMTAB, ".strtab", SHT_STRTAB, false),
       hash_builder_(".hash", SHT_HASH, SHF_ALLOC, &dynsym_builder_, 0, sizeof(Elf_Word),
@@ -569,6 +572,11 @@
   }
 
   bool Init() {
+    // Since the .text section of an oat file contains relative references to .rodata
+    // and (optionally) .bss, we keep these 2 or 3 sections together. This creates
+    // a non-traditional layout where the .bss section is mapped independently of the
+    // .dynamic section and needs its own program header with LOAD RW.
+    //
     // The basic layout of the elf file. Order may be different in final output.
     // +-------------------------+
     // | Elf_Ehdr                |
@@ -576,6 +584,7 @@
     // | Elf_Phdr PHDR           |
     // | Elf_Phdr LOAD R         | .dynsym .dynstr .hash .rodata
     // | Elf_Phdr LOAD R X       | .text
+    // | Elf_Phdr LOAD RW        | .bss (Optional)
     // | Elf_Phdr LOAD RW        | .dynamic
     // | Elf_Phdr DYNAMIC        | .dynamic
     // +-------------------------+
@@ -584,6 +593,8 @@
     // | Elf_Sym  oatdata        |
     // | Elf_Sym  oatexec        |
     // | Elf_Sym  oatlastword    |
+    // | Elf_Sym  oatbss         | (Optional)
+    // | Elf_Sym  oatbsslastword | (Optional)
     // +-------------------------+
     // | .dynstr                 |
     // | \0                      |
@@ -631,6 +642,7 @@
     // | .hash\0                 |
     // | .rodata\0               |
     // | .text\0                 |
+    // | .bss\0                  |  (Optional)
     // | .shstrtab\0             |
     // | .symtab\0               |  (Optional)
     // | .strtab\0               |  (Optional)
@@ -654,8 +666,9 @@
     // | Elf_Shdr .dynsym        |
     // | Elf_Shdr .dynstr        |
     // | Elf_Shdr .hash          |
-    // | Elf_Shdr .text          |
     // | Elf_Shdr .rodata        |
+    // | Elf_Shdr .text          |
+    // | Elf_Shdr .bss           |  (Optional)
     // | Elf_Shdr .dynamic       |
     // | Elf_Shdr .shstrtab      |
     // | Elf_Shdr .debug_info    |  (Optional)
@@ -694,8 +707,11 @@
     program_headers_[PH_LOAD_R_X].p_type    = PT_LOAD;
     program_headers_[PH_LOAD_R_X].p_flags   = PF_R | PF_X;
 
-    program_headers_[PH_LOAD_RW_].p_type    = PT_LOAD;
-    program_headers_[PH_LOAD_RW_].p_flags   = PF_R | PF_W;
+    program_headers_[PH_LOAD_RW_BSS].p_type    = PT_LOAD;
+    program_headers_[PH_LOAD_RW_BSS].p_flags   = PF_R | PF_W;
+
+    program_headers_[PH_LOAD_RW_DYNAMIC].p_type    = PT_LOAD;
+    program_headers_[PH_LOAD_RW_DYNAMIC].p_flags   = PF_R | PF_W;
 
     program_headers_[PH_DYNAMIC].p_type    = PT_DYNAMIC;
     program_headers_[PH_DYNAMIC].p_flags   = PF_R | PF_W;
@@ -760,6 +776,14 @@
     text_builder_.SetSectionIndex(section_index_);
     section_index_++;
 
+    // Setup .bss
+    if (bss_builder_.GetSize() != 0u) {
+      section_ptrs_.push_back(bss_builder_.GetSection());
+      AssignSectionStr(&bss_builder_, &shstrtab_);
+      bss_builder_.SetSectionIndex(section_index_);
+      section_index_++;
+    }
+
     // Setup .dynamic
     section_ptrs_.push_back(dynamic_builder_.GetSection());
     AssignSectionStr(&dynamic_builder_, &shstrtab_);
@@ -820,10 +844,20 @@
     CHECK_ALIGNED(rodata_builder_.GetSection()->sh_offset +
                   rodata_builder_.GetSection()->sh_size, kPageSize);
 
+    // Get the layout of the .bss section.
+    bss_builder_.GetSection()->sh_offset =
+        NextOffset<Elf_Word, Elf_Shdr>(*bss_builder_.GetSection(),
+                                       *text_builder_.GetSection());
+    bss_builder_.GetSection()->sh_addr = bss_builder_.GetSection()->sh_offset;
+    bss_builder_.GetSection()->sh_size = bss_builder_.GetSize();
+    bss_builder_.GetSection()->sh_link = bss_builder_.GetLink();
+
     // Get the layout of the dynamic section.
-    dynamic_builder_.GetSection()->sh_offset =
-        NextOffset<Elf_Word, Elf_Shdr>(*dynamic_builder_.GetSection(), *text_builder_.GetSection());
-    dynamic_builder_.GetSection()->sh_addr = dynamic_builder_.GetSection()->sh_offset;
+    CHECK(IsAlignedParam(bss_builder_.GetSection()->sh_offset,
+                         dynamic_builder_.GetSection()->sh_addralign));
+    dynamic_builder_.GetSection()->sh_offset = bss_builder_.GetSection()->sh_offset;
+    dynamic_builder_.GetSection()->sh_addr =
+        NextOffset<Elf_Word, Elf_Shdr>(*dynamic_builder_.GetSection(), *bss_builder_.GetSection());
     dynamic_builder_.GetSection()->sh_size = dynamic_builder_.GetSize() * sizeof(Elf_Dyn);
     dynamic_builder_.GetSection()->sh_link = dynamic_builder_.GetLink();
 
@@ -987,16 +1021,23 @@
     program_headers_[PH_LOAD_R_X].p_memsz  = load_rx_size;
     program_headers_[PH_LOAD_R_X].p_align  = text_builder_.GetSection()->sh_addralign;
 
-    program_headers_[PH_LOAD_RW_].p_offset = dynamic_builder_.GetSection()->sh_offset;
-    program_headers_[PH_LOAD_RW_].p_vaddr  = dynamic_builder_.GetSection()->sh_offset;
-    program_headers_[PH_LOAD_RW_].p_paddr  = dynamic_builder_.GetSection()->sh_offset;
-    program_headers_[PH_LOAD_RW_].p_filesz = dynamic_builder_.GetSection()->sh_size;
-    program_headers_[PH_LOAD_RW_].p_memsz  = dynamic_builder_.GetSection()->sh_size;
-    program_headers_[PH_LOAD_RW_].p_align  = dynamic_builder_.GetSection()->sh_addralign;
+    program_headers_[PH_LOAD_RW_BSS].p_offset = bss_builder_.GetSection()->sh_offset;
+    program_headers_[PH_LOAD_RW_BSS].p_vaddr  = bss_builder_.GetSection()->sh_offset;
+    program_headers_[PH_LOAD_RW_BSS].p_paddr  = bss_builder_.GetSection()->sh_offset;
+    program_headers_[PH_LOAD_RW_BSS].p_filesz = 0;
+    program_headers_[PH_LOAD_RW_BSS].p_memsz  = bss_builder_.GetSection()->sh_size;
+    program_headers_[PH_LOAD_RW_BSS].p_align  = bss_builder_.GetSection()->sh_addralign;
+
+    program_headers_[PH_LOAD_RW_DYNAMIC].p_offset = dynamic_builder_.GetSection()->sh_offset;
+    program_headers_[PH_LOAD_RW_DYNAMIC].p_vaddr  = dynamic_builder_.GetSection()->sh_addr;
+    program_headers_[PH_LOAD_RW_DYNAMIC].p_paddr  = dynamic_builder_.GetSection()->sh_addr;
+    program_headers_[PH_LOAD_RW_DYNAMIC].p_filesz = dynamic_builder_.GetSection()->sh_size;
+    program_headers_[PH_LOAD_RW_DYNAMIC].p_memsz  = dynamic_builder_.GetSection()->sh_size;
+    program_headers_[PH_LOAD_RW_DYNAMIC].p_align  = dynamic_builder_.GetSection()->sh_addralign;
 
     program_headers_[PH_DYNAMIC].p_offset = dynamic_builder_.GetSection()->sh_offset;
-    program_headers_[PH_DYNAMIC].p_vaddr  = dynamic_builder_.GetSection()->sh_offset;
-    program_headers_[PH_DYNAMIC].p_paddr  = dynamic_builder_.GetSection()->sh_offset;
+    program_headers_[PH_DYNAMIC].p_vaddr  = dynamic_builder_.GetSection()->sh_addr;
+    program_headers_[PH_DYNAMIC].p_paddr  = dynamic_builder_.GetSection()->sh_addr;
     program_headers_[PH_DYNAMIC].p_filesz = dynamic_builder_.GetSection()->sh_size;
     program_headers_[PH_DYNAMIC].p_memsz  = dynamic_builder_.GetSection()->sh_size;
     program_headers_[PH_DYNAMIC].p_align  = dynamic_builder_.GetSection()->sh_addralign;
@@ -1004,15 +1045,29 @@
     // Finish setup of the Ehdr values.
     elf_header_.e_phoff = PHDR_OFFSET;
     elf_header_.e_shoff = sections_offset;
-    elf_header_.e_phnum = PH_NUM;
+    elf_header_.e_phnum = (bss_builder_.GetSection()->sh_size != 0u) ? PH_NUM : PH_NUM - 1;
     elf_header_.e_shnum = section_ptrs_.size();
     elf_header_.e_shstrndx = shstrtab_builder_.GetSectionIndex();
 
     // Add the rest of the pieces to the list.
     pieces.push_back(new ElfFileMemoryPiece<Elf_Word>("Elf Header", 0, &elf_header_,
                                                       sizeof(elf_header_)));
-    pieces.push_back(new ElfFileMemoryPiece<Elf_Word>("Program headers", PHDR_OFFSET,
-                                                      &program_headers_, sizeof(program_headers_)));
+    if (bss_builder_.GetSection()->sh_size != 0u) {
+      pieces.push_back(new ElfFileMemoryPiece<Elf_Word>("Program headers", PHDR_OFFSET,
+                                                        &program_headers_[0],
+                                                        elf_header_.e_phnum * sizeof(Elf_Phdr)));
+    } else {
+      // Skip PH_LOAD_RW_BSS.
+      Elf_Word part1_size = PH_LOAD_RW_BSS * sizeof(Elf_Phdr);
+      Elf_Word part2_size = (PH_NUM - PH_LOAD_RW_BSS - 1) * sizeof(Elf_Phdr);
+      CHECK_EQ(part1_size + part2_size, elf_header_.e_phnum * sizeof(Elf_Phdr));
+      pieces.push_back(new ElfFileMemoryPiece<Elf_Word>("Program headers", PHDR_OFFSET,
+                                                        &program_headers_[0], part1_size));
+      pieces.push_back(new ElfFileMemoryPiece<Elf_Word>("Program headers part 2",
+                                                        PHDR_OFFSET + part1_size,
+                                                        &program_headers_[PH_LOAD_RW_BSS + 1],
+                                                        part2_size));
+    }
     pieces.push_back(new ElfFileMemoryPiece<Elf_Word>(".dynamic",
                                                       dynamic_builder_.GetSection()->sh_offset,
                                                       dynamic.data(),
@@ -1175,6 +1230,12 @@
                               text_builder_.GetSize(), STB_GLOBAL, STT_OBJECT);
     dynsym_builder_.AddSymbol("oatlastword", &text_builder_, text_builder_.GetSize() - 4,
                               true, 4, STB_GLOBAL, STT_OBJECT);
+    if (bss_builder_.GetSize() != 0u) {
+      dynsym_builder_.AddSymbol("oatbss", &bss_builder_, 0, true,
+                                bss_builder_.GetSize(), STB_GLOBAL, STT_OBJECT);
+      dynsym_builder_.AddSymbol("oatbsslastword", &bss_builder_, bss_builder_.GetSize() - 4,
+                                true, 4, STB_GLOBAL, STT_OBJECT);
+    }
   }
 
   void AssignSectionStr(ElfSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr>* builder,
@@ -1213,12 +1274,13 @@
   // What phdr is.
   static const uint32_t PHDR_OFFSET = sizeof(Elf_Ehdr);
   enum : uint8_t {
-    PH_PHDR     = 0,
-        PH_LOAD_R__ = 1,
-        PH_LOAD_R_X = 2,
-        PH_LOAD_RW_ = 3,
-        PH_DYNAMIC  = 4,
-        PH_NUM      = 5,
+    PH_PHDR             = 0,
+    PH_LOAD_R__         = 1,
+    PH_LOAD_R_X         = 2,
+    PH_LOAD_RW_BSS      = 3,
+    PH_LOAD_RW_DYNAMIC  = 4,
+    PH_DYNAMIC          = 5,
+    PH_NUM              = 6,
   };
   static const uint32_t PHDR_SIZE = sizeof(Elf_Phdr) * PH_NUM;
   Elf_Phdr program_headers_[PH_NUM];
@@ -1236,6 +1298,7 @@
 
   ElfOatSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> text_builder_;
   ElfOatSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> rodata_builder_;
+  ElfOatSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> bss_builder_;
   ElfSymtabBuilder<Elf_Word, Elf_Sword, Elf_Addr, Elf_Sym, Elf_Shdr> dynsym_builder_;
   ElfSymtabBuilder<Elf_Word, Elf_Sword, Elf_Addr, Elf_Sym, Elf_Shdr> symtab_builder_;
   ElfSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> hash_builder_;
diff --git a/compiler/elf_writer_quick.cc b/compiler/elf_writer_quick.cc
index 401d5a9..a822b24 100644
--- a/compiler/elf_writer_quick.cc
+++ b/compiler/elf_writer_quick.cc
@@ -229,6 +229,7 @@
   const OatHeader& oat_header = oat_writer->GetOatHeader();
   Elf_Word oat_data_size = oat_header.GetExecutableOffset();
   uint32_t oat_exec_size = oat_writer->GetSize() - oat_data_size;
+  uint32_t oat_bss_size = oat_writer->GetBssSize();
 
   OatWriterWrapper wrapper(oat_writer);
 
@@ -243,6 +244,8 @@
           oat_data_size,
           oat_data_size,
           oat_exec_size,
+          RoundUp(oat_data_size + oat_exec_size, kPageSize),
+          oat_bss_size,
           compiler_driver_->GetCompilerOptions().GetIncludeDebugSymbols(),
           debug));
 
diff --git a/compiler/image_writer.cc b/compiler/image_writer.cc
index c588e1a..f5f9320 100644
--- a/compiler/image_writer.cc
+++ b/compiler/image_writer.cc
@@ -408,8 +408,8 @@
 bool ImageWriter::AllocMemory() {
   size_t length = RoundUp(Runtime::Current()->GetHeap()->GetTotalMemory(), kPageSize);
   std::string error_msg;
-  image_.reset(MemMap::MapAnonymous("image writer image", NULL, length, PROT_READ | PROT_WRITE,
-                                    false, &error_msg));
+  image_.reset(MemMap::MapAnonymous("image writer image", nullptr, length, PROT_READ | PROT_WRITE,
+                                    false, false, &error_msg));
   if (UNLIKELY(image_.get() == nullptr)) {
     LOG(ERROR) << "Failed to allocate memory for image file generation: " << error_msg;
     return false;
diff --git a/compiler/jit/jit_compiler.cc b/compiler/jit/jit_compiler.cc
new file mode 100644
index 0000000..0283791
--- /dev/null
+++ b/compiler/jit/jit_compiler.cc
@@ -0,0 +1,257 @@
+/*
+ * Copyright 2014 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "jit_compiler.h"
+
+#include "arch/instruction_set.h"
+#include "arch/instruction_set_features.h"
+#include "compiler_callbacks.h"
+#include "dex/pass_manager.h"
+#include "dex/quick_compiler_callbacks.h"
+#include "driver/compiler_driver.h"
+#include "driver/compiler_options.h"
+#include "jit/jit.h"
+#include "jit/jit_code_cache.h"
+#include "mirror/art_method-inl.h"
+#include "oat_file-inl.h"
+#include "object_lock.h"
+#include "thread_list.h"
+#include "verifier/method_verifier-inl.h"
+
+namespace art {
+namespace jit {
+
+JitCompiler* JitCompiler::Create() {
+  return new JitCompiler();
+}
+
+extern "C" void* jit_load(CompilerCallbacks** callbacks) {
+  VLOG(jit) << "loading jit compiler";
+  auto* const jit_compiler = JitCompiler::Create();
+  CHECK(jit_compiler != nullptr);
+  *callbacks = jit_compiler->GetCompilerCallbacks();
+  VLOG(jit) << "Done loading jit compiler";
+  return jit_compiler;
+}
+
+extern "C" void jit_unload(void* handle) {
+  DCHECK(handle != nullptr);
+  delete reinterpret_cast<JitCompiler*>(handle);
+}
+
+extern "C" bool jit_compile_method(void* handle, mirror::ArtMethod* method, Thread* self)
+    SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+  auto* jit_compiler = reinterpret_cast<JitCompiler*>(handle);
+  DCHECK(jit_compiler != nullptr);
+  return jit_compiler->CompileMethod(self, method);
+}
+
+JitCompiler::JitCompiler() : total_time_(0) {
+  auto* pass_manager_options = new PassManagerOptions;
+  pass_manager_options->SetDisablePassList("GVN,DCE");
+  compiler_options_.reset(new CompilerOptions(
+      CompilerOptions::kDefaultCompilerFilter,
+      CompilerOptions::kDefaultHugeMethodThreshold,
+      CompilerOptions::kDefaultLargeMethodThreshold,
+      CompilerOptions::kDefaultSmallMethodThreshold,
+      CompilerOptions::kDefaultTinyMethodThreshold,
+      CompilerOptions::kDefaultNumDexMethodsThreshold,
+      false,
+      false,
+      CompilerOptions::kDefaultTopKProfileThreshold,
+      false,
+      false,
+      false,
+      false,
+      false,  // pic
+      nullptr,
+      pass_manager_options,
+      nullptr));
+  const InstructionSet instruction_set = kRuntimeISA;
+  instruction_set_features_.reset(InstructionSetFeatures::FromCppDefines());
+  cumulative_logger_.reset(new CumulativeLogger("jit times"));
+  verification_results_.reset(new VerificationResults(compiler_options_.get()));
+  method_inliner_map_.reset(new DexFileToMethodInlinerMap);
+  callbacks_.reset(new QuickCompilerCallbacks(verification_results_.get(),
+                                              method_inliner_map_.get()));
+  compiler_driver_.reset(new CompilerDriver(
+      compiler_options_.get(), verification_results_.get(), method_inliner_map_.get(),
+      Compiler::kQuick, instruction_set, instruction_set_features_.get(), false,
+      nullptr, new std::set<std::string>, 1, false, true,
+      std::string(), cumulative_logger_.get(), -1, std::string()));
+  // Disable dedupe so we can remove compiled methods.
+  compiler_driver_->SetDedupeEnabled(false);
+  compiler_driver_->SetSupportBootImageFixup(false);
+}
+
+JitCompiler::~JitCompiler() {
+}
+
+bool JitCompiler::CompileMethod(Thread* self, mirror::ArtMethod* method) {
+  uint64_t start_time = NanoTime();
+  StackHandleScope<2> hs(self);
+  self->AssertNoPendingException();
+  Runtime* runtime = Runtime::Current();
+  Handle<mirror::ArtMethod> h_method(hs.NewHandle(method));
+  if (runtime->GetJit()->GetCodeCache()->ContainsMethod(method)) {
+    VLOG(jit) << "Already compiled " << PrettyMethod(method);
+    return true;  // Already compiled
+  }
+  Handle<mirror::Class> h_class(hs.NewHandle(h_method->GetDeclaringClass()));
+  if (!runtime->GetClassLinker()->EnsureInitialized(self, h_class, true, true)) {
+    VLOG(jit) << "JIT failed to initialize " << PrettyMethod(h_method.Get());
+    return false;
+  }
+  const DexFile* dex_file = h_class->GetDexCache()->GetDexFile();
+  MethodReference method_ref(dex_file, h_method->GetDexMethodIndex());
+  // Only verify if we don't already have verification results.
+  if (verification_results_->GetVerifiedMethod(method_ref) == nullptr) {
+    std::string error;
+    if (verifier::MethodVerifier::VerifyMethod(h_method.Get(), true, &error) ==
+        verifier::MethodVerifier::kHardFailure) {
+      VLOG(jit) << "Not compile method " << PrettyMethod(h_method.Get())
+          << " due to verification failure " << error;
+      return false;
+    }
+  }
+  CompiledMethod* compiled_method(compiler_driver_->CompileMethod(self, h_method.Get()));
+  if (compiled_method == nullptr) {
+    return false;
+  }
+  total_time_ += NanoTime() - start_time;
+  // Don't add the method if we are supposed to be deoptimized.
+  bool result = false;
+  if (!runtime->GetInstrumentation()->AreAllMethodsDeoptimized()) {
+    const void* code = Runtime::Current()->GetClassLinker()->GetOatMethodQuickCodeFor(
+        h_method.Get());
+    if (code != nullptr) {
+      // Already have some compiled code, just use this instead of linking.
+      // TODO: Fix recompilation.
+      h_method->SetEntryPointFromQuickCompiledCode(code);
+      result = true;
+    } else {
+      result = MakeExecutable(compiled_method, h_method.Get());
+    }
+  }
+  // Remove the compiled method to save memory.
+  compiler_driver_->RemoveCompiledMethod(method_ref);
+  return result;
+}
+
+CompilerCallbacks* JitCompiler::GetCompilerCallbacks() const {
+  return callbacks_.get();
+}
+
+uint8_t* JitCompiler::WriteMethodHeaderAndCode(const CompiledMethod* compiled_method,
+                                               uint8_t* reserve_begin, uint8_t* reserve_end,
+                                               const uint8_t* mapping_table,
+                                               const uint8_t* vmap_table,
+                                               const uint8_t* gc_map) {
+  reserve_begin += sizeof(OatQuickMethodHeader);
+  reserve_begin = reinterpret_cast<uint8_t*>(
+      compiled_method->AlignCode(reinterpret_cast<uintptr_t>(reserve_begin)));
+  const auto* quick_code = compiled_method->GetQuickCode();
+  CHECK_LE(reserve_begin, reserve_end);
+  CHECK_LE(quick_code->size(), static_cast<size_t>(reserve_end - reserve_begin));
+  auto* code_ptr = reserve_begin;
+  OatQuickMethodHeader* method_header = reinterpret_cast<OatQuickMethodHeader*>(code_ptr) - 1;
+  // Construct the header last.
+  const auto frame_size_in_bytes = compiled_method->GetFrameSizeInBytes();
+  const auto core_spill_mask = compiled_method->GetCoreSpillMask();
+  const auto fp_spill_mask = compiled_method->GetFpSpillMask();
+  const auto code_size = quick_code->size();
+  CHECK_NE(code_size, 0U);
+  std::copy(quick_code->data(), quick_code->data() + code_size, code_ptr);
+  // After we are done writing we need to update the method header.
+  // Write out the method header last.
+  method_header = new(method_header)OatQuickMethodHeader(
+      code_ptr - mapping_table, code_ptr - vmap_table, code_ptr - gc_map, frame_size_in_bytes,
+      core_spill_mask, fp_spill_mask, code_size);
+  // Return the code ptr.
+  return code_ptr;
+}
+
+bool JitCompiler::AddToCodeCache(mirror::ArtMethod* method, const CompiledMethod* compiled_method,
+                                 OatFile::OatMethod* out_method) {
+  Runtime* runtime = Runtime::Current();
+  JitCodeCache* const code_cache = runtime->GetJit()->GetCodeCache();
+  const auto* quick_code = compiled_method->GetQuickCode();
+  if (quick_code == nullptr) {
+    return false;
+  }
+  const auto code_size = quick_code->size();
+  Thread* const self = Thread::Current();
+  const uint8_t* base = code_cache->CodeCachePtr();
+  auto* const mapping_table = compiled_method->GetMappingTable();
+  auto* const vmap_table = compiled_method->GetVmapTable();
+  auto* const gc_map = compiled_method->GetGcMap();
+  // Write out pre-header stuff.
+  uint8_t* const mapping_table_ptr = code_cache->AddDataArray(
+      self, mapping_table->data(), mapping_table->data() + mapping_table->size());
+  if (mapping_table == nullptr) {
+    return false;  // Out of data cache.
+  }
+  uint8_t* const vmap_table_ptr = code_cache->AddDataArray(
+      self, vmap_table->data(), vmap_table->data() + vmap_table->size());
+  if (vmap_table == nullptr) {
+    return false;  // Out of data cache.
+  }
+  uint8_t* const gc_map_ptr = code_cache->AddDataArray(
+      self, gc_map->data(), gc_map->data() + gc_map->size());
+  if (gc_map == nullptr) {
+    return false;  // Out of data cache.
+  }
+  // Don't touch this until you protect / unprotect the code.
+  const size_t reserve_size = sizeof(OatQuickMethodHeader) + quick_code->size() + 32;
+  uint8_t* const code_reserve = code_cache->ReserveCode(self, reserve_size);
+  if (code_reserve == nullptr) {
+    return false;
+  }
+  auto* code_ptr = WriteMethodHeaderAndCode(
+      compiled_method, code_reserve, code_reserve + reserve_size, mapping_table_ptr,
+      vmap_table_ptr, gc_map_ptr);
+
+  const size_t thumb_offset = compiled_method->CodeDelta();
+  const uint32_t code_offset = code_ptr - base + thumb_offset;
+  *out_method = OatFile::OatMethod(base, code_offset);
+  DCHECK_EQ(out_method->GetGcMap(), gc_map_ptr);
+  DCHECK_EQ(out_method->GetMappingTable(), mapping_table_ptr);
+  DCHECK_EQ(out_method->GetVmapTable(), vmap_table_ptr);
+  DCHECK_EQ(out_method->GetFrameSizeInBytes(), compiled_method->GetFrameSizeInBytes());
+  DCHECK_EQ(out_method->GetCoreSpillMask(), compiled_method->GetCoreSpillMask());
+  DCHECK_EQ(out_method->GetFpSpillMask(), compiled_method->GetFpSpillMask());
+  VLOG(jit)  << "JIT added " << PrettyMethod(method) << "@" << method << " ccache_size="
+      << PrettySize(code_cache->CodeCacheSize()) << ": " << reinterpret_cast<void*>(code_ptr)
+      << "," << reinterpret_cast<void*>(code_ptr + code_size);
+  return true;
+}
+
+bool JitCompiler::MakeExecutable(CompiledMethod* compiled_method, mirror::ArtMethod* method) {
+  CHECK(method != nullptr);
+  CHECK(compiled_method != nullptr);
+  OatFile::OatMethod oat_method(nullptr, 0);
+  if (!AddToCodeCache(method, compiled_method, &oat_method)) {
+    return false;
+  }
+  // TODO: Flush instruction cache.
+  oat_method.LinkMethod(method);
+  CHECK(Runtime::Current()->GetJit()->GetCodeCache()->ContainsMethod(method))
+      << PrettyMethod(method);
+  return true;
+}
+
+}  // namespace jit
+}  // namespace art
diff --git a/compiler/jit/jit_compiler.h b/compiler/jit/jit_compiler.h
new file mode 100644
index 0000000..0876499
--- /dev/null
+++ b/compiler/jit/jit_compiler.h
@@ -0,0 +1,76 @@
+/*
+ * Copyright 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.
+ */
+
+#ifndef ART_COMPILER_JIT_JIT_COMPILER_H_
+#define ART_COMPILER_JIT_JIT_COMPILER_H_
+
+#include "base/mutex.h"
+#include "compiler_callbacks.h"
+#include "compiled_method.h"
+#include "dex/verification_results.h"
+#include "dex/quick/dex_file_to_method_inliner_map.h"
+#include "driver/compiler_driver.h"
+#include "driver/compiler_options.h"
+#include "oat_file.h"
+
+namespace art {
+
+class InstructionSetFeatures;
+
+namespace mirror {
+class ArtMethod;
+}
+
+namespace jit {
+
+class JitCompiler {
+ public:
+  static JitCompiler* Create();
+  virtual ~JitCompiler();
+  bool CompileMethod(Thread* self, mirror::ArtMethod* method)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  // This is in the compiler since the runtime doesn't have access to the compiled method
+  // structures.
+  bool AddToCodeCache(mirror::ArtMethod* method, const CompiledMethod* compiled_method,
+                      OatFile::OatMethod* out_method) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  CompilerCallbacks* GetCompilerCallbacks() const;
+  size_t GetTotalCompileTime() const {
+    return total_time_;
+  }
+
+ private:
+  uint64_t total_time_;
+  std::unique_ptr<CompilerOptions> compiler_options_;
+  std::unique_ptr<CumulativeLogger> cumulative_logger_;
+  std::unique_ptr<VerificationResults> verification_results_;
+  std::unique_ptr<DexFileToMethodInlinerMap> method_inliner_map_;
+  std::unique_ptr<CompilerCallbacks> callbacks_;
+  std::unique_ptr<CompilerDriver> compiler_driver_;
+  std::unique_ptr<const InstructionSetFeatures> instruction_set_features_;
+
+  explicit JitCompiler();
+  uint8_t* WriteMethodHeaderAndCode(
+      const CompiledMethod* compiled_method, uint8_t* reserve_begin, uint8_t* reserve_end,
+      const uint8_t* mapping_table, const uint8_t* vmap_table, const uint8_t* gc_map);
+  bool MakeExecutable(CompiledMethod* compiled_method, mirror::ArtMethod* method)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+};
+
+}  // namespace jit
+
+}  // namespace art
+
+#endif  // ART_COMPILER_JIT_JIT_COMPILER_H_
diff --git a/compiler/jni/quick/arm/calling_convention_arm.cc b/compiler/jni/quick/arm/calling_convention_arm.cc
index 669c3bb..d3690b2 100644
--- a/compiler/jni/quick/arm/calling_convention_arm.cc
+++ b/compiler/jni/quick/arm/calling_convention_arm.cc
@@ -31,6 +31,10 @@
   S0, S1, S2, S3, S4, S5, S6, S7, S8, S9, S10, S11, S12, S13, S14, S15
 };
 
+static const SRegister kHFSCalleeSaveRegisters[] = {
+  S16, S17, S18, S19, S20, S21, S22, S23, S24, S25, S26, S27, S28, S29, S30, S31
+};
+
 static const DRegister kHFDArgumentRegisters[] = {
   D0, D1, D2, D3, D4, D5, D6, D7
 };
@@ -226,6 +230,10 @@
   callee_save_regs_.push_back(ArmManagedRegister::FromCoreRegister(R8));
   callee_save_regs_.push_back(ArmManagedRegister::FromCoreRegister(R10));
   callee_save_regs_.push_back(ArmManagedRegister::FromCoreRegister(R11));
+
+  for (size_t i = 0; i < arraysize(kHFSCalleeSaveRegisters); ++i) {
+    callee_save_regs_.push_back(ArmManagedRegister::FromSRegister(kHFSCalleeSaveRegisters[i]));
+  }
 }
 
 uint32_t ArmJniCallingConvention::CoreSpillMask() const {
@@ -235,6 +243,14 @@
   return result;
 }
 
+uint32_t ArmJniCallingConvention::FpSpillMask() const {
+  uint32_t result = 0;
+  for (size_t i = 0; i < arraysize(kHFSCalleeSaveRegisters); ++i) {
+    result |= (1 << kHFSCalleeSaveRegisters[i]);
+  }
+  return result;
+}
+
 ManagedRegister ArmJniCallingConvention::ReturnScratchRegister() const {
   return ArmManagedRegister::FromCoreRegister(R2);
 }
diff --git a/compiler/jni/quick/arm/calling_convention_arm.h b/compiler/jni/quick/arm/calling_convention_arm.h
index 604ce1c..dbecb8e 100644
--- a/compiler/jni/quick/arm/calling_convention_arm.h
+++ b/compiler/jni/quick/arm/calling_convention_arm.h
@@ -63,9 +63,7 @@
   }
   ManagedRegister ReturnScratchRegister() const OVERRIDE;
   uint32_t CoreSpillMask() const OVERRIDE;
-  uint32_t FpSpillMask() const OVERRIDE {
-    return 0;  // Floats aren't spilled in JNI down call
-  }
+  uint32_t FpSpillMask() const OVERRIDE;
   bool IsCurrentParamInRegister() OVERRIDE;
   bool IsCurrentParamOnStack() OVERRIDE;
   ManagedRegister CurrentParamRegister() OVERRIDE;
diff --git a/compiler/jni/quick/arm64/calling_convention_arm64.cc b/compiler/jni/quick/arm64/calling_convention_arm64.cc
index b9c8178..05eb80a 100644
--- a/compiler/jni/quick/arm64/calling_convention_arm64.cc
+++ b/compiler/jni/quick/arm64/calling_convention_arm64.cc
@@ -38,6 +38,10 @@
   S0, S1, S2, S3, S4, S5, S6, S7
 };
 
+static const DRegister kDCalleeSaveRegisters[] = {
+  D8, D9, D10, D11, D12, D13, D14, D15
+};
+
 // Calling convention
 ManagedRegister Arm64ManagedRuntimeCallingConvention::InterproceduralScratchRegister() {
   return Arm64ManagedRegister::FromXRegister(X20);  // saved on entry restored on exit
@@ -166,6 +170,10 @@
   callee_save_regs_.push_back(Arm64ManagedRegister::FromXRegister(X28));
   callee_save_regs_.push_back(Arm64ManagedRegister::FromXRegister(X29));
   callee_save_regs_.push_back(Arm64ManagedRegister::FromXRegister(X30));
+
+  for (size_t i = 0; i < arraysize(kDCalleeSaveRegisters); ++i) {
+    callee_save_regs_.push_back(Arm64ManagedRegister::FromDRegister(kDCalleeSaveRegisters[i]));
+  }
 }
 
 uint32_t Arm64JniCallingConvention::CoreSpillMask() const {
@@ -184,10 +192,11 @@
 }
 
 uint32_t Arm64JniCallingConvention::FpSpillMask() const {
-  // Compute spill mask to agree with callee saves initialized in the constructor
-  // Note: All callee-save fp registers will be preserved by aapcs64. And they are not used
-  // in the jni method.
-  return 0;
+  uint32_t result = 0;
+  for (size_t i = 0; i < arraysize(kDCalleeSaveRegisters); ++i) {
+    result |= (1 << kDCalleeSaveRegisters[i]);
+  }
+  return result;
 }
 
 ManagedRegister Arm64JniCallingConvention::ReturnScratchRegister() const {
diff --git a/compiler/oat_writer.cc b/compiler/oat_writer.cc
index 9c0157e..c32a992 100644
--- a/compiler/oat_writer.cc
+++ b/compiler/oat_writer.cc
@@ -403,6 +403,7 @@
     image_writer_(image_writer),
     dex_files_(&dex_files),
     size_(0u),
+    bss_size_(0u),
     oat_data_offset_(0u),
     image_file_location_oat_checksum_(image_file_location_oat_checksum),
     image_file_location_oat_begin_(image_file_location_oat_begin),
@@ -549,7 +550,7 @@
 
 struct OatWriter::VmapTableDataAccess {
   static const SwapVector<uint8_t>* GetData(const CompiledMethod* compiled_method) ALWAYS_INLINE {
-    return &compiled_method->GetVmapTable();
+    return compiled_method->GetVmapTable();
   }
 
   static uint32_t GetOffset(OatClass* oat_class, size_t method_offsets_index) ALWAYS_INLINE {
diff --git a/compiler/oat_writer.h b/compiler/oat_writer.h
index e020d31..fd2ccae 100644
--- a/compiler/oat_writer.h
+++ b/compiler/oat_writer.h
@@ -96,6 +96,10 @@
     return size_;
   }
 
+  size_t GetBssSize() const {
+    return bss_size_;
+  }
+
   const std::vector<uintptr_t>& GetAbsolutePatchLocations() const {
     return absolute_patch_locations_;
   }
@@ -266,6 +270,9 @@
   // Size required for Oat data structures.
   size_t size_;
 
+  // The size of the required .bss section holding the DexCache data.
+  size_t bss_size_;
+
   // Offset of the oat data from the start of the mmapped region of the elf file.
   size_t oat_data_offset_;
 
@@ -341,8 +348,8 @@
       if (UNLIKELY(lhs->GetMappingTable() != rhs->GetMappingTable())) {
         return lhs->GetMappingTable() < rhs->GetMappingTable();
       }
-      if (UNLIKELY(&lhs->GetVmapTable() != &rhs->GetVmapTable())) {
-        return &lhs->GetVmapTable() < &rhs->GetVmapTable();
+      if (UNLIKELY(lhs->GetVmapTable() != rhs->GetVmapTable())) {
+        return lhs->GetVmapTable() < rhs->GetVmapTable();
       }
       if (UNLIKELY(lhs->GetGcMap() != rhs->GetGcMap())) {
         return lhs->GetGcMap() < rhs->GetGcMap();
diff --git a/compiler/optimizing/bounds_check_elimination.cc b/compiler/optimizing/bounds_check_elimination.cc
index bf0804a..070981e 100644
--- a/compiler/optimizing/bounds_check_elimination.cc
+++ b/compiler/optimizing/bounds_check_elimination.cc
@@ -904,6 +904,21 @@
     FindAndHandlePartialArrayLength(ushr);
   }
 
+  void VisitAnd(HAnd* instruction) {
+    if (instruction->GetRight()->IsIntConstant()) {
+      int32_t constant = instruction->GetRight()->AsIntConstant()->GetValue();
+      if (constant > 0) {
+        // constant serves as a mask so any number masked with it
+        // gets a [0, constant] value range.
+        ValueRange* range = new (GetGraph()->GetArena()) ValueRange(
+            GetGraph()->GetArena(),
+            ValueBound(nullptr, 0),
+            ValueBound(nullptr, constant));
+        GetValueRangeMap(instruction->GetBlock())->Overwrite(instruction->GetId(), range);
+      }
+    }
+  }
+
   void VisitNewArray(HNewArray* new_array) {
     HInstruction* len = new_array->InputAt(0);
     if (!len->IsIntConstant()) {
diff --git a/compiler/optimizing/bounds_check_elimination.h b/compiler/optimizing/bounds_check_elimination.h
index 05cb185..9e98ccf 100644
--- a/compiler/optimizing/bounds_check_elimination.h
+++ b/compiler/optimizing/bounds_check_elimination.h
@@ -23,10 +23,13 @@
 
 class BoundsCheckElimination : public HOptimization {
  public:
-  explicit BoundsCheckElimination(HGraph* graph) : HOptimization(graph, true, "BCE") {}
+  explicit BoundsCheckElimination(HGraph* graph)
+      : HOptimization(graph, true, kBoundsCheckEliminiationPassName) {}
 
   void Run() OVERRIDE;
 
+  static constexpr const char* kBoundsCheckEliminiationPassName = "BCE";
+
  private:
   DISALLOW_COPY_AND_ASSIGN(BoundsCheckElimination);
 };
diff --git a/compiler/optimizing/builder.h b/compiler/optimizing/builder.h
index 3e4a616..96196de 100644
--- a/compiler/optimizing/builder.h
+++ b/compiler/optimizing/builder.h
@@ -80,6 +80,8 @@
 
   bool BuildGraph(const DexFile::CodeItem& code);
 
+  static constexpr const char* kBuilderPassName = "builder";
+
  private:
   // Analyzes the dex instruction and adds HInstruction to the graph
   // to execute that instruction. Returns whether the instruction can
diff --git a/compiler/optimizing/code_generator.cc b/compiler/optimizing/code_generator.cc
index 2a57fdc..ba5f7d8 100644
--- a/compiler/optimizing/code_generator.cc
+++ b/compiler/optimizing/code_generator.cc
@@ -386,7 +386,9 @@
           compiler_options);
     }
     case kArm64: {
-      return new arm64::CodeGeneratorARM64(graph, compiler_options);
+      return new arm64::CodeGeneratorARM64(graph,
+          *isa_features.AsArm64InstructionSetFeatures(),
+          compiler_options);
     }
     case kMips:
       return nullptr;
diff --git a/compiler/optimizing/code_generator_arm64.cc b/compiler/optimizing/code_generator_arm64.cc
index 729bab7..c21084a 100644
--- a/compiler/optimizing/code_generator_arm64.cc
+++ b/compiler/optimizing/code_generator_arm64.cc
@@ -16,6 +16,7 @@
 
 #include "code_generator_arm64.h"
 
+#include "arch/arm64/instruction_set_features_arm64.h"
 #include "common_arm64.h"
 #include "entrypoints/quick/quick_entrypoints.h"
 #include "entrypoints/quick/quick_entrypoints_enum.h"
@@ -397,7 +398,9 @@
   return next_location;
 }
 
-CodeGeneratorARM64::CodeGeneratorARM64(HGraph* graph, const CompilerOptions& compiler_options)
+CodeGeneratorARM64::CodeGeneratorARM64(HGraph* graph,
+                                       const Arm64InstructionSetFeatures& isa_features,
+                                       const CompilerOptions& compiler_options)
     : CodeGenerator(graph,
                     kNumberOfAllocatableRegisters,
                     kNumberOfAllocatableFPRegisters,
@@ -408,7 +411,8 @@
       block_labels_(nullptr),
       location_builder_(graph, this),
       instruction_visitor_(graph, this),
-      move_resolver_(graph->GetArena(), this) {
+      move_resolver_(graph->GetArena(), this),
+      isa_features_(isa_features) {
   // Save the link register (containing the return address) to mimic Quick.
   AddAllocatedRegister(LocationFrom(lr));
 }
@@ -998,9 +1002,10 @@
   UseScratchRegisterScope temps(GetVIXLAssembler());
   Register temp = temps.AcquireW();
   size_t status_offset = mirror::Class::StatusOffset().SizeValue();
+  bool use_acquire_release = codegen_->GetInstructionSetFeatures().PreferAcquireRelease();
 
   // Even if the initialized flag is set, we need to ensure consistent memory ordering.
-  if (kUseAcquireRelease) {
+  if (use_acquire_release) {
     // TODO(vixl): Let the MacroAssembler handle MemOperand.
     __ Add(temp, class_reg, status_offset);
     __ Ldar(temp, HeapOperand(temp));
@@ -1689,9 +1694,10 @@
 
 void InstructionCodeGeneratorARM64::VisitInstanceFieldGet(HInstanceFieldGet* instruction) {
   MemOperand field = HeapOperand(InputRegisterAt(instruction, 0), instruction->GetFieldOffset());
+  bool use_acquire_release = codegen_->GetInstructionSetFeatures().PreferAcquireRelease();
 
   if (instruction->IsVolatile()) {
-    if (kUseAcquireRelease) {
+    if (use_acquire_release) {
       // NB: LoadAcquire will record the pc info if needed.
       codegen_->LoadAcquire(instruction, OutputCPURegister(instruction), field);
     } else {
@@ -1718,9 +1724,10 @@
   CPURegister value = InputCPURegisterAt(instruction, 1);
   Offset offset = instruction->GetFieldOffset();
   Primitive::Type field_type = instruction->GetFieldType();
+  bool use_acquire_release = codegen_->GetInstructionSetFeatures().PreferAcquireRelease();
 
   if (instruction->IsVolatile()) {
-    if (kUseAcquireRelease) {
+    if (use_acquire_release) {
       codegen_->StoreRelease(field_type, value, HeapOperand(obj, offset));
       codegen_->MaybeRecordImplicitNullCheck(instruction);
     } else {
@@ -2437,9 +2444,10 @@
 
 void InstructionCodeGeneratorARM64::VisitStaticFieldGet(HStaticFieldGet* instruction) {
   MemOperand field = HeapOperand(InputRegisterAt(instruction, 0), instruction->GetFieldOffset());
+  bool use_acquire_release = codegen_->GetInstructionSetFeatures().PreferAcquireRelease();
 
   if (instruction->IsVolatile()) {
-    if (kUseAcquireRelease) {
+    if (use_acquire_release) {
       // NB: LoadAcquire will record the pc info if needed.
       codegen_->LoadAcquire(instruction, OutputCPURegister(instruction), field);
     } else {
@@ -2464,9 +2472,10 @@
   CPURegister value = InputCPURegisterAt(instruction, 1);
   Offset offset = instruction->GetFieldOffset();
   Primitive::Type field_type = instruction->GetFieldType();
+  bool use_acquire_release = codegen_->GetInstructionSetFeatures().PreferAcquireRelease();
 
   if (instruction->IsVolatile()) {
-    if (kUseAcquireRelease) {
+    if (use_acquire_release) {
       codegen_->StoreRelease(field_type, value, HeapOperand(cls, offset));
     } else {
       GenerateMemoryBarrier(MemBarrierKind::kAnyStore);
diff --git a/compiler/optimizing/code_generator_arm64.h b/compiler/optimizing/code_generator_arm64.h
index afb7fc3..48961d6 100644
--- a/compiler/optimizing/code_generator_arm64.h
+++ b/compiler/optimizing/code_generator_arm64.h
@@ -32,10 +32,6 @@
 
 class CodeGeneratorARM64;
 
-// TODO: Tune the use of Load-Acquire, Store-Release vs Data Memory Barriers.
-// For now we prefer the use of load-acquire, store-release over explicit memory barriers.
-static constexpr bool kUseAcquireRelease = true;
-
 // Use a local definition to prevent copying mistakes.
 static constexpr size_t kArm64WordSize = kArm64PointerSize;
 
@@ -195,7 +191,9 @@
 
 class CodeGeneratorARM64 : public CodeGenerator {
  public:
-  CodeGeneratorARM64(HGraph* graph, const CompilerOptions& compiler_options);
+  CodeGeneratorARM64(HGraph* graph,
+                     const Arm64InstructionSetFeatures& isa_features,
+                     const CompilerOptions& compiler_options);
   virtual ~CodeGeneratorARM64() {}
 
   void GenerateFrameEntry() OVERRIDE;
@@ -273,6 +271,10 @@
     return InstructionSet::kArm64;
   }
 
+  const Arm64InstructionSetFeatures& GetInstructionSetFeatures() const {
+    return isa_features_;
+  }
+
   void Initialize() OVERRIDE {
     HGraph* graph = GetGraph();
     int length = graph->GetBlocks().Size();
@@ -317,6 +319,7 @@
   InstructionCodeGeneratorARM64 instruction_visitor_;
   ParallelMoveResolverARM64 move_resolver_;
   Arm64Assembler assembler_;
+  const Arm64InstructionSetFeatures& isa_features_;
 
   DISALLOW_COPY_AND_ASSIGN(CodeGeneratorARM64);
 };
diff --git a/compiler/optimizing/code_generator_x86.cc b/compiler/optimizing/code_generator_x86.cc
index 7b35cfd..116dd15 100644
--- a/compiler/optimizing/code_generator_x86.cc
+++ b/compiler/optimizing/code_generator_x86.cc
@@ -819,7 +819,7 @@
         // Materialized condition, compare against 0.
         Location lhs = if_instr->GetLocations()->InAt(0);
         if (lhs.IsRegister()) {
-          __ cmpl(lhs.AsRegister<Register>(), Immediate(0));
+          __ testl(lhs.AsRegister<Register>(), lhs.AsRegister<Register>());
         } else {
           __ cmpl(Address(ESP, lhs.GetStackIndex()), Immediate(0));
         }
@@ -836,9 +836,12 @@
       if (rhs.IsRegister()) {
         __ cmpl(lhs.AsRegister<Register>(), rhs.AsRegister<Register>());
       } else if (rhs.IsConstant()) {
-        HIntConstant* instruction = rhs.GetConstant()->AsIntConstant();
-        Immediate imm(instruction->AsIntConstant()->GetValue());
-        __ cmpl(lhs.AsRegister<Register>(), imm);
+        int32_t constant = rhs.GetConstant()->AsIntConstant()->GetValue();
+        if (constant == 0) {
+          __ testl(lhs.AsRegister<Register>(), lhs.AsRegister<Register>());
+        } else {
+          __ cmpl(lhs.AsRegister<Register>(), Immediate(constant));
+        }
       } else {
         __ cmpl(lhs.AsRegister<Register>(), Address(ESP, rhs.GetStackIndex()));
       }
@@ -914,16 +917,19 @@
     Register reg = locations->Out().AsRegister<Register>();
     // Clear register: setcc only sets the low byte.
     __ xorl(reg, reg);
-    if (locations->InAt(1).IsRegister()) {
-      __ cmpl(locations->InAt(0).AsRegister<Register>(),
-              locations->InAt(1).AsRegister<Register>());
-    } else if (locations->InAt(1).IsConstant()) {
-      HConstant* instruction = locations->InAt(1).GetConstant();
-      Immediate imm(CodeGenerator::GetInt32ValueOf(instruction));
-      __ cmpl(locations->InAt(0).AsRegister<Register>(), imm);
+    Location lhs = locations->InAt(0);
+    Location rhs = locations->InAt(1);
+    if (rhs.IsRegister()) {
+      __ cmpl(lhs.AsRegister<Register>(), rhs.AsRegister<Register>());
+    } else if (rhs.IsConstant()) {
+      int32_t constant = rhs.GetConstant()->AsIntConstant()->GetValue();
+      if (constant == 0) {
+        __ testl(lhs.AsRegister<Register>(), lhs.AsRegister<Register>());
+      } else {
+      __ cmpl(lhs.AsRegister<Register>(), Immediate(constant));
+      }
     } else {
-      __ cmpl(locations->InAt(0).AsRegister<Register>(),
-              Address(ESP, locations->InAt(1).GetStackIndex()));
+      __ cmpl(lhs.AsRegister<Register>(), Address(ESP, rhs.GetStackIndex()));
     }
     __ setb(X86Condition(comp->GetCondition()), reg);
   }
@@ -1798,7 +1804,13 @@
   LocationSummary* locations =
       new (GetGraph()->GetArena()) LocationSummary(add, LocationSummary::kNoCall);
   switch (add->GetResultType()) {
-    case Primitive::kPrimInt:
+    case Primitive::kPrimInt: {
+      locations->SetInAt(0, Location::RequiresRegister());
+      locations->SetInAt(1, Location::RegisterOrConstant(add->InputAt(1)));
+      locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
+      break;
+    }
+
     case Primitive::kPrimLong: {
       locations->SetInAt(0, Location::RequiresRegister());
       locations->SetInAt(1, Location::Any());
@@ -1824,15 +1836,26 @@
   LocationSummary* locations = add->GetLocations();
   Location first = locations->InAt(0);
   Location second = locations->InAt(1);
-  DCHECK(first.Equals(locations->Out()));
+  Location out = locations->Out();
+
   switch (add->GetResultType()) {
     case Primitive::kPrimInt: {
       if (second.IsRegister()) {
-        __ addl(first.AsRegister<Register>(), second.AsRegister<Register>());
+        if (out.AsRegister<Register>() == first.AsRegister<Register>()) {
+          __ addl(out.AsRegister<Register>(), second.AsRegister<Register>());
+        } else {
+          __ leal(out.AsRegister<Register>(), Address(
+              first.AsRegister<Register>(), second.AsRegister<Register>(), TIMES_1, 0));
+          }
       } else if (second.IsConstant()) {
-        __ addl(first.AsRegister<Register>(),
-                Immediate(second.GetConstant()->AsIntConstant()->GetValue()));
+        int32_t value = second.GetConstant()->AsIntConstant()->GetValue();
+        if (out.AsRegister<Register>() == first.AsRegister<Register>()) {
+          __ addl(out.AsRegister<Register>(), Immediate(value));
+        } else {
+          __ leal(out.AsRegister<Register>(), Address(first.AsRegister<Register>(), value));
+        }
       } else {
+        DCHECK(first.Equals(locations->Out()));
         __ addl(first.AsRegister<Register>(), Address(ESP, second.GetStackIndex()));
       }
       break;
@@ -3502,12 +3525,16 @@
   } else if (source.IsConstant()) {
     HConstant* constant = source.GetConstant();
     if (constant->IsIntConstant() || constant->IsNullConstant()) {
-      Immediate imm(CodeGenerator::GetInt32ValueOf(constant));
+      int32_t value = CodeGenerator::GetInt32ValueOf(constant);
       if (destination.IsRegister()) {
-        __ movl(destination.AsRegister<Register>(), imm);
+        if (value == 0) {
+          __ xorl(destination.AsRegister<Register>(), destination.AsRegister<Register>());
+        } else {
+          __ movl(destination.AsRegister<Register>(), Immediate(value));
+        }
       } else {
         DCHECK(destination.IsStackSlot()) << destination;
-        __ movl(Address(ESP, destination.GetStackIndex()), imm);
+        __ movl(Address(ESP, destination.GetStackIndex()), Immediate(value));
       }
     } else {
       DCHECK(constant->IsFloatConstant());
diff --git a/compiler/optimizing/code_generator_x86_64.cc b/compiler/optimizing/code_generator_x86_64.cc
index 74adb31..adc022a 100644
--- a/compiler/optimizing/code_generator_x86_64.cc
+++ b/compiler/optimizing/code_generator_x86_64.cc
@@ -1868,8 +1868,19 @@
 
     case Primitive::kPrimLong: {
       locations->SetInAt(0, Location::RequiresRegister());
-      locations->SetInAt(1, Location::RequiresRegister());
-      locations->SetOut(Location::SameAsFirstInput());
+      // We can use a leaq or addq if the constant can fit in an immediate.
+      HInstruction* rhs = add->InputAt(1);
+      bool is_int32_constant = false;
+      if (rhs->IsLongConstant()) {
+        int64_t value = rhs->AsLongConstant()->GetValue();
+        if (static_cast<int32_t>(value) == value) {
+          is_int32_constant = true;
+        }
+      }
+      locations->SetInAt(1,
+          is_int32_constant ? Location::RegisterOrConstant(rhs) :
+                              Location::RequiresRegister());
+      locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
       break;
     }
 
@@ -1917,7 +1928,25 @@
     }
 
     case Primitive::kPrimLong: {
-      __ addq(first.AsRegister<CpuRegister>(), second.AsRegister<CpuRegister>());
+      if (second.IsRegister()) {
+        if (out.AsRegister<Register>() == first.AsRegister<Register>()) {
+          __ addq(out.AsRegister<CpuRegister>(), second.AsRegister<CpuRegister>());
+        } else {
+          __ leaq(out.AsRegister<CpuRegister>(), Address(
+              first.AsRegister<CpuRegister>(), second.AsRegister<CpuRegister>(), TIMES_1, 0));
+        }
+      } else {
+        DCHECK(second.IsConstant());
+        int64_t value = second.GetConstant()->AsLongConstant()->GetValue();
+        int32_t int32_value = Low32Bits(value);
+        DCHECK_EQ(int32_value, value);
+        if (out.AsRegister<Register>() == first.AsRegister<Register>()) {
+          __ addq(out.AsRegister<CpuRegister>(), Immediate(int32_value));
+        } else {
+          __ leaq(out.AsRegister<CpuRegister>(), Address(
+              first.AsRegister<CpuRegister>(), int32_value));
+        }
+      }
       break;
     }
 
diff --git a/compiler/optimizing/codegen_test.cc b/compiler/optimizing/codegen_test.cc
index e0e0b4c..868fc5b 100644
--- a/compiler/optimizing/codegen_test.cc
+++ b/compiler/optimizing/codegen_test.cc
@@ -18,6 +18,7 @@
 
 #include "arch/instruction_set.h"
 #include "arch/arm/instruction_set_features_arm.h"
+#include "arch/arm64/instruction_set_features_arm64.h"
 #include "base/macros.h"
 #include "builder.h"
 #include "code_generator_arm.h"
@@ -115,9 +116,9 @@
     Run(allocator, codegenX86, has_result, expected);
   }
 
-  std::unique_ptr<const ArmInstructionSetFeatures> features(
+  std::unique_ptr<const ArmInstructionSetFeatures> features_arm(
       ArmInstructionSetFeatures::FromCppDefines());
-  TestCodeGeneratorARM codegenARM(graph, *features.get(), compiler_options);
+  TestCodeGeneratorARM codegenARM(graph, *features_arm.get(), compiler_options);
   codegenARM.CompileBaseline(&allocator, true);
   if (kRuntimeISA == kArm || kRuntimeISA == kThumb2) {
     Run(allocator, codegenARM, has_result, expected);
@@ -129,7 +130,9 @@
     Run(allocator, codegenX86_64, has_result, expected);
   }
 
-  arm64::CodeGeneratorARM64 codegenARM64(graph, compiler_options);
+  std::unique_ptr<const Arm64InstructionSetFeatures> features_arm64(
+      Arm64InstructionSetFeatures::FromCppDefines());
+  arm64::CodeGeneratorARM64 codegenARM64(graph, *features_arm64.get(), compiler_options);
   codegenARM64.CompileBaseline(&allocator, true);
   if (kRuntimeISA == kArm64) {
     Run(allocator, codegenARM64, has_result, expected);
@@ -166,7 +169,9 @@
                                     compiler_options);
     RunCodeOptimized(&codegenARM, graph, hook_before_codegen, has_result, expected);
   } else if (kRuntimeISA == kArm64) {
-    arm64::CodeGeneratorARM64 codegenARM64(graph, compiler_options);
+    arm64::CodeGeneratorARM64 codegenARM64(graph,
+                                           *Arm64InstructionSetFeatures::FromCppDefines(),
+                                           compiler_options);
     RunCodeOptimized(&codegenARM64, graph, hook_before_codegen, has_result, expected);
   } else if (kRuntimeISA == kX86) {
     x86::CodeGeneratorX86 codegenX86(graph, compiler_options);
diff --git a/compiler/optimizing/graph_visualizer.cc b/compiler/optimizing/graph_visualizer.cc
index c592737..cabfa48 100644
--- a/compiler/optimizing/graph_visualizer.cc
+++ b/compiler/optimizing/graph_visualizer.cc
@@ -17,8 +17,10 @@
 #include "graph_visualizer.h"
 
 #include "code_generator.h"
+#include "licm.h"
 #include "nodes.h"
 #include "optimization.h"
+#include "register_allocator.h"
 #include "ssa_liveness_analysis.h"
 
 namespace art {
@@ -188,6 +190,10 @@
     output_ << " " << phi->GetRegNumber();
   }
 
+  bool IsPass(const char* name) {
+    return strcmp(pass_name_, name) == 0;
+  }
+
   void PrintInstruction(HInstruction* instruction) {
     output_ << instruction->DebugName();
     instruction->Accept(this);
@@ -211,7 +217,7 @@
       }
       output_ << "])";
     }
-    if (pass_name_ == kLivenessPassName
+    if (IsPass(SsaLivenessAnalysis::kLivenessPassName)
         && is_after_pass_
         && instruction->GetLifetimePosition() != kNoLifetime) {
       output_ << " (liveness: " << instruction->GetLifetimePosition();
@@ -221,7 +227,7 @@
         interval.Dump(output_);
       }
       output_ << ")";
-    } else if (pass_name_ == kRegisterAllocatorPassName && is_after_pass_) {
+    } else if (IsPass(RegisterAllocator::kRegisterAllocatorPassName) && is_after_pass_) {
       LocationSummary* locations = instruction->GetLocations();
       if (locations != nullptr) {
         output_ << " ( ";
@@ -236,7 +242,7 @@
         }
       }
       output_ << " (liveness: " << instruction->GetLifetimePosition() << ")";
-    } else if (pass_name_ == kLoopInvariantCodeMotionPassName) {
+    } else if (IsPass(LICM::kLoopInvariantCodeMotionPassName)) {
       output_ << " ( loop_header:";
       HLoopInformation* info = instruction->GetBlock()->GetLoopInformation();
       if (info == nullptr) {
diff --git a/compiler/optimizing/gvn.h b/compiler/optimizing/gvn.h
index 57e0487..e74d969 100644
--- a/compiler/optimizing/gvn.h
+++ b/compiler/optimizing/gvn.h
@@ -27,10 +27,12 @@
 class GVNOptimization : public HOptimization {
  public:
   GVNOptimization(HGraph* graph, const SideEffectsAnalysis& side_effects)
-      : HOptimization(graph, true, "GVN"), side_effects_(side_effects) {}
+      : HOptimization(graph, true, kGlobalValueNumberingPassName), side_effects_(side_effects) {}
 
   void Run() OVERRIDE;
 
+  static constexpr const char* kGlobalValueNumberingPassName = "GVN";
+
  private:
   const SideEffectsAnalysis& side_effects_;
 
diff --git a/compiler/optimizing/inliner.h b/compiler/optimizing/inliner.h
index 8e9cf83..2b08d3d 100644
--- a/compiler/optimizing/inliner.h
+++ b/compiler/optimizing/inliner.h
@@ -35,13 +35,15 @@
            CompilerDriver* compiler_driver,
            OptimizingCompilerStats* stats,
            size_t depth = 0)
-      : HOptimization(outer_graph, true, "inliner", stats),
+      : HOptimization(outer_graph, true, kInlinerPassName, stats),
         outer_compilation_unit_(outer_compilation_unit),
         compiler_driver_(compiler_driver),
         depth_(depth) {}
 
   void Run() OVERRIDE;
 
+  static constexpr const char* kInlinerPassName = "inliner";
+
  private:
   bool TryInline(HInvoke* invoke_instruction, uint32_t method_index, InvokeType invoke_type) const;
 
diff --git a/compiler/optimizing/instruction_simplifier.h b/compiler/optimizing/instruction_simplifier.h
index a7ff755..0244620 100644
--- a/compiler/optimizing/instruction_simplifier.h
+++ b/compiler/optimizing/instruction_simplifier.h
@@ -30,9 +30,11 @@
  public:
   InstructionSimplifier(HGraph* graph,
                         OptimizingCompilerStats* stats = nullptr,
-                        const char* name = "instruction_simplifier")
+                        const char* name = kInstructionSimplifierPassName)
     : HOptimization(graph, true, name, stats) {}
 
+  static constexpr const char* kInstructionSimplifierPassName = "instruction_simplifier";
+
   void Run() OVERRIDE;
 };
 
diff --git a/compiler/optimizing/intrinsics.h b/compiler/optimizing/intrinsics.h
index 29cc8ef..dbb7cba 100644
--- a/compiler/optimizing/intrinsics.h
+++ b/compiler/optimizing/intrinsics.h
@@ -29,11 +29,13 @@
 class IntrinsicsRecognizer : public HOptimization {
  public:
   IntrinsicsRecognizer(HGraph* graph, const DexFile* dex_file, CompilerDriver* driver)
-      : HOptimization(graph, true, "intrinsics_recognition"),
+      : HOptimization(graph, true, kIntrinsicsRecognizerPassName),
         dex_file_(dex_file), driver_(driver) {}
 
   void Run() OVERRIDE;
 
+  static constexpr const char* kIntrinsicsRecognizerPassName = "intrinsics_recognition";
+
  private:
   const DexFile* dex_file_;
   CompilerDriver* driver_;
diff --git a/compiler/optimizing/intrinsics_arm64.cc b/compiler/optimizing/intrinsics_arm64.cc
index 8874edc..1ddff8a 100644
--- a/compiler/optimizing/intrinsics_arm64.cc
+++ b/compiler/optimizing/intrinsics_arm64.cc
@@ -16,6 +16,7 @@
 
 #include "intrinsics_arm64.h"
 
+#include "arch/arm64/instruction_set_features_arm64.h"
 #include "code_generator_arm64.h"
 #include "common_arm64.h"
 #include "entrypoints/quick/quick_entrypoints.h"
@@ -682,10 +683,11 @@
   Register base = WRegisterFrom(locations->InAt(1));    // Object pointer.
   Register offset = XRegisterFrom(locations->InAt(2));  // Long offset.
   Register trg = RegisterFrom(locations->Out(), type);
+  bool use_acquire_release = codegen->GetInstructionSetFeatures().PreferAcquireRelease();
 
   MemOperand mem_op(base.X(), offset);
   if (is_volatile) {
-    if (kUseAcquireRelease) {
+    if (use_acquire_release) {
       codegen->LoadAcquire(invoke, trg, mem_op);
     } else {
       codegen->Load(type, trg, mem_op);
@@ -792,11 +794,12 @@
   Register base = WRegisterFrom(locations->InAt(1));    // Object pointer.
   Register offset = XRegisterFrom(locations->InAt(2));  // Long offset.
   Register value = RegisterFrom(locations->InAt(3), type);
+  bool use_acquire_release = codegen->GetInstructionSetFeatures().PreferAcquireRelease();
 
   MemOperand mem_op(base.X(), offset);
 
   if (is_volatile || is_ordered) {
-    if (kUseAcquireRelease) {
+    if (use_acquire_release) {
       codegen->StoreRelease(type, value, mem_op);
     } else {
       __ Dmb(InnerShareable, BarrierAll);
@@ -856,10 +859,7 @@
 }
 
 static void GenCas(LocationSummary* locations, Primitive::Type type, CodeGeneratorARM64* codegen) {
-  // TODO: Currently we use acquire-release load-stores in the CAS loop. One could reasonably write
-  //       a version relying on simple exclusive load-stores and barriers instead.
-  static_assert(kUseAcquireRelease, "Non-acquire-release inlined CAS not implemented, yet.");
-
+  bool use_acquire_release = codegen->GetInstructionSetFeatures().PreferAcquireRelease();
   vixl::MacroAssembler* masm = codegen->GetAssembler()->vixl_masm_;
 
   Register out = WRegisterFrom(locations->Out());                  // Boolean result.
@@ -889,15 +889,23 @@
   // result = tmp_value != 0;
 
   vixl::Label loop_head, exit_loop;
-  __ Bind(&loop_head);
-
-  __ Ldaxr(tmp_value, MemOperand(tmp_ptr));
-  __ Cmp(tmp_value, expected);
-  __ B(&exit_loop, ne);
-
-  __ Stlxr(tmp_32, value, MemOperand(tmp_ptr));
-  __ Cbnz(tmp_32, &loop_head);
-
+  if (use_acquire_release) {
+    __ Bind(&loop_head);
+    __ Ldaxr(tmp_value, MemOperand(tmp_ptr));
+    __ Cmp(tmp_value, expected);
+    __ B(&exit_loop, ne);
+    __ Stlxr(tmp_32, value, MemOperand(tmp_ptr));
+    __ Cbnz(tmp_32, &loop_head);
+  } else {
+    __ Dmb(InnerShareable, BarrierWrites);
+    __ Bind(&loop_head);
+    __ Ldxr(tmp_value, MemOperand(tmp_ptr));
+    __ Cmp(tmp_value, expected);
+    __ B(&exit_loop, ne);
+    __ Stxr(tmp_32, value, MemOperand(tmp_ptr));
+    __ Cbnz(tmp_32, &loop_head);
+    __ Dmb(InnerShareable, BarrierAll);
+  }
   __ Bind(&exit_loop);
   __ Cset(out, eq);
 }
diff --git a/compiler/optimizing/licm.h b/compiler/optimizing/licm.h
index 4812394..cb6170e 100644
--- a/compiler/optimizing/licm.h
+++ b/compiler/optimizing/licm.h
@@ -31,6 +31,8 @@
 
   void Run() OVERRIDE;
 
+  static constexpr const char* kLoopInvariantCodeMotionPassName = "licm";
+
  private:
   const SideEffectsAnalysis& side_effects_;
 
diff --git a/compiler/optimizing/live_ranges_test.cc b/compiler/optimizing/live_ranges_test.cc
index 17914e8..c102c4f 100644
--- a/compiler/optimizing/live_ranges_test.cc
+++ b/compiler/optimizing/live_ranges_test.cc
@@ -399,11 +399,11 @@
   LiveInterval* interval = liveness.GetInstructionFromSsaIndex(0)->GetLiveInterval();
   LiveRange* range = interval->GetFirstRange();
   ASSERT_EQ(2u, range->GetStart());
-  ASSERT_EQ(16u, range->GetEnd());
+  ASSERT_EQ(17u, range->GetEnd());
   range = range->GetNext();
   ASSERT_TRUE(range != nullptr);
   ASSERT_EQ(20u, range->GetStart());
-  ASSERT_EQ(22u, range->GetEnd());
+  ASSERT_EQ(23u, range->GetEnd());
   ASSERT_TRUE(range->GetNext() == nullptr);
 
   // Test for the 4 constant.
diff --git a/compiler/optimizing/optimization.h b/compiler/optimizing/optimization.h
index af39e09..8b20281 100644
--- a/compiler/optimizing/optimization.h
+++ b/compiler/optimizing/optimization.h
@@ -22,12 +22,6 @@
 
 namespace art {
 
-static const char* kBuilderPassName = "builder";
-static const char* kSsaBuilderPassName = "ssa_builder";
-static const char* kLivenessPassName = "liveness";
-static const char* kRegisterAllocatorPassName = "register";
-static const char* kLoopInvariantCodeMotionPassName = "licm";
-
 /**
  * Abstraction to implement an optimization pass.
  */
diff --git a/compiler/optimizing/optimizing_compiler.cc b/compiler/optimizing/optimizing_compiler.cc
index 2fef8c7..eb98424 100644
--- a/compiler/optimizing/optimizing_compiler.cc
+++ b/compiler/optimizing/optimizing_compiler.cc
@@ -56,7 +56,7 @@
  */
 class CodeVectorAllocator FINAL : public CodeAllocator {
  public:
-  CodeVectorAllocator() {}
+  CodeVectorAllocator() : size_(0) {}
 
   virtual uint8_t* Allocate(size_t size) {
     size_ = size;
@@ -361,11 +361,11 @@
   PrepareForRegisterAllocation(graph).Run();
   SsaLivenessAnalysis liveness(*graph, codegen);
   {
-    PassInfo pass_info(kLivenessPassName, pass_info_printer);
+    PassInfo pass_info(SsaLivenessAnalysis::kLivenessPassName, pass_info_printer);
     liveness.Analyze();
   }
   {
-    PassInfo pass_info(kRegisterAllocatorPassName, pass_info_printer);
+    PassInfo pass_info(RegisterAllocator::kRegisterAllocatorPassName, pass_info_printer);
     RegisterAllocator(graph->GetArena(), codegen, liveness).AllocateRegisters();
   }
 
@@ -495,7 +495,7 @@
   VLOG(compiler) << "Building " << method_name;
 
   {
-    PassInfo pass_info(kBuilderPassName, &pass_info_printer);
+    PassInfo pass_info(HGraphBuilder::kBuilderPassName, &pass_info_printer);
     if (!builder.BuildGraph(*code_item)) {
       CHECK(!shouldCompile) << "Could not build graph in optimizing compiler";
       return nullptr;
@@ -508,7 +508,7 @@
     VLOG(compiler) << "Optimizing " << method_name;
 
     {
-      PassInfo pass_info(kSsaBuilderPassName, &pass_info_printer);
+      PassInfo pass_info(SsaBuilder::kSsaBuilderPassName, &pass_info_printer);
       if (!graph->TryBuildingSsa()) {
         // We could not transform the graph to SSA, bailout.
         LOG(INFO) << "Skipping compilation of " << method_name << ": it contains a non natural loop";
diff --git a/compiler/optimizing/reference_type_propagation.h b/compiler/optimizing/reference_type_propagation.h
index 815caab..733e18e 100644
--- a/compiler/optimizing/reference_type_propagation.h
+++ b/compiler/optimizing/reference_type_propagation.h
@@ -34,7 +34,7 @@
                            const DexFile& dex_file,
                            const DexCompilationUnit& dex_compilation_unit,
                            StackHandleScopeCollection* handles)
-    : HOptimization(graph, true, "reference_type_propagation"),
+    : HOptimization(graph, true, kReferenceTypePropagationPassName),
       dex_file_(dex_file),
       dex_compilation_unit_(dex_compilation_unit),
       handles_(handles),
@@ -42,6 +42,8 @@
 
   void Run() OVERRIDE;
 
+  static constexpr const char* kReferenceTypePropagationPassName = "reference_type_propagation";
+
  private:
   void VisitNewInstance(HNewInstance* new_instance);
   void VisitLoadClass(HLoadClass* load_class);
diff --git a/compiler/optimizing/register_allocator.h b/compiler/optimizing/register_allocator.h
index ff2f106..579f069 100644
--- a/compiler/optimizing/register_allocator.h
+++ b/compiler/optimizing/register_allocator.h
@@ -81,6 +81,8 @@
         + double_spill_slots_.Size();
   }
 
+  static constexpr const char* kRegisterAllocatorPassName = "register";
+
  private:
   // Main methods of the allocator.
   void LinearScan();
diff --git a/compiler/optimizing/register_allocator_test.cc b/compiler/optimizing/register_allocator_test.cc
index e5d06a9..b757a3b 100644
--- a/compiler/optimizing/register_allocator_test.cc
+++ b/compiler/optimizing/register_allocator_test.cc
@@ -322,9 +322,9 @@
 TEST(RegisterAllocatorTest, FirstRegisterUse) {
   const uint16_t data[] = THREE_REGISTERS_CODE_ITEM(
     Instruction::CONST_4 | 0 | 0,
-    Instruction::ADD_INT_LIT8 | 1 << 8, 1 << 8,
-    Instruction::ADD_INT_LIT8 | 0 << 8, 1 << 8,
-    Instruction::ADD_INT_LIT8 | 1 << 8, 1 << 8 | 1,
+    Instruction::XOR_INT_LIT8 | 1 << 8, 1 << 8,
+    Instruction::XOR_INT_LIT8 | 0 << 8, 1 << 8,
+    Instruction::XOR_INT_LIT8 | 1 << 8, 1 << 8 | 1,
     Instruction::RETURN_VOID);
 
   ArenaPool pool;
@@ -334,27 +334,27 @@
   SsaLivenessAnalysis liveness(*graph, &codegen);
   liveness.Analyze();
 
-  HAdd* first_add = graph->GetBlocks().Get(1)->GetFirstInstruction()->AsAdd();
-  HAdd* last_add = graph->GetBlocks().Get(1)->GetLastInstruction()->GetPrevious()->AsAdd();
-  ASSERT_EQ(last_add->InputAt(0), first_add);
-  LiveInterval* interval = first_add->GetLiveInterval();
-  ASSERT_EQ(interval->GetEnd(), last_add->GetLifetimePosition());
+  HXor* first_xor = graph->GetBlocks().Get(1)->GetFirstInstruction()->AsXor();
+  HXor* last_xor = graph->GetBlocks().Get(1)->GetLastInstruction()->GetPrevious()->AsXor();
+  ASSERT_EQ(last_xor->InputAt(0), first_xor);
+  LiveInterval* interval = first_xor->GetLiveInterval();
+  ASSERT_EQ(interval->GetEnd(), last_xor->GetLifetimePosition());
   ASSERT_TRUE(interval->GetNextSibling() == nullptr);
 
   // We need a register for the output of the instruction.
-  ASSERT_EQ(interval->FirstRegisterUse(), first_add->GetLifetimePosition());
+  ASSERT_EQ(interval->FirstRegisterUse(), first_xor->GetLifetimePosition());
 
   // Split at the next instruction.
-  interval = interval->SplitAt(first_add->GetLifetimePosition() + 2);
+  interval = interval->SplitAt(first_xor->GetLifetimePosition() + 2);
   // The user of the split is the last add.
-  ASSERT_EQ(interval->FirstRegisterUse(), last_add->GetLifetimePosition());
+  ASSERT_EQ(interval->FirstRegisterUse(), last_xor->GetLifetimePosition());
 
   // Split before the last add.
-  LiveInterval* new_interval = interval->SplitAt(last_add->GetLifetimePosition() - 1);
+  LiveInterval* new_interval = interval->SplitAt(last_xor->GetLifetimePosition() - 1);
   // Ensure the current interval has no register use...
   ASSERT_EQ(interval->FirstRegisterUse(), kNoLifetime);
   // And the new interval has it for the last add.
-  ASSERT_EQ(new_interval->FirstRegisterUse(), last_add->GetLifetimePosition());
+  ASSERT_EQ(new_interval->FirstRegisterUse(), last_xor->GetLifetimePosition());
 }
 
 TEST(RegisterAllocatorTest, DeadPhi) {
@@ -634,9 +634,9 @@
   }
 }
 
-static HGraph* BuildTwoAdds(ArenaAllocator* allocator,
-                            HInstruction** first_add,
-                            HInstruction** second_add) {
+static HGraph* BuildTwoSubs(ArenaAllocator* allocator,
+                            HInstruction** first_sub,
+                            HInstruction** second_sub) {
   HGraph* graph = new (allocator) HGraph(allocator);
   HBasicBlock* entry = new (allocator) HBasicBlock(graph);
   graph->AddBlock(entry);
@@ -652,10 +652,10 @@
   graph->AddBlock(block);
   entry->AddSuccessor(block);
 
-  *first_add = new (allocator) HAdd(Primitive::kPrimInt, parameter, constant1);
-  block->AddInstruction(*first_add);
-  *second_add = new (allocator) HAdd(Primitive::kPrimInt, *first_add, constant2);
-  block->AddInstruction(*second_add);
+  *first_sub = new (allocator) HSub(Primitive::kPrimInt, parameter, constant1);
+  block->AddInstruction(*first_sub);
+  *second_sub = new (allocator) HSub(Primitive::kPrimInt, *first_sub, constant2);
+  block->AddInstruction(*second_sub);
 
   block->AddInstruction(new (allocator) HExit());
   return graph;
@@ -664,10 +664,10 @@
 TEST(RegisterAllocatorTest, SameAsFirstInputHint) {
   ArenaPool pool;
   ArenaAllocator allocator(&pool);
-  HInstruction *first_add, *second_add;
+  HInstruction *first_sub, *second_sub;
 
   {
-    HGraph* graph = BuildTwoAdds(&allocator, &first_add, &second_add);
+    HGraph* graph = BuildTwoSubs(&allocator, &first_sub, &second_sub);
     x86::CodeGeneratorX86 codegen(graph, CompilerOptions());
     SsaLivenessAnalysis liveness(*graph, &codegen);
     liveness.Analyze();
@@ -676,27 +676,27 @@
     register_allocator.AllocateRegisters();
 
     // Sanity check that in normal conditions, the registers are the same.
-    ASSERT_EQ(first_add->GetLiveInterval()->GetRegister(), 1);
-    ASSERT_EQ(second_add->GetLiveInterval()->GetRegister(), 1);
+    ASSERT_EQ(first_sub->GetLiveInterval()->GetRegister(), 1);
+    ASSERT_EQ(second_sub->GetLiveInterval()->GetRegister(), 1);
   }
 
   {
-    HGraph* graph = BuildTwoAdds(&allocator, &first_add, &second_add);
+    HGraph* graph = BuildTwoSubs(&allocator, &first_sub, &second_sub);
     x86::CodeGeneratorX86 codegen(graph, CompilerOptions());
     SsaLivenessAnalysis liveness(*graph, &codegen);
     liveness.Analyze();
 
     // check that both adds get the same register.
     // Don't use UpdateOutput because output is already allocated.
-    first_add->InputAt(0)->GetLocations()->output_ = Location::RegisterLocation(2);
-    ASSERT_EQ(first_add->GetLocations()->Out().GetPolicy(), Location::kSameAsFirstInput);
-    ASSERT_EQ(second_add->GetLocations()->Out().GetPolicy(), Location::kSameAsFirstInput);
+    first_sub->InputAt(0)->GetLocations()->output_ = Location::RegisterLocation(2);
+    ASSERT_EQ(first_sub->GetLocations()->Out().GetPolicy(), Location::kSameAsFirstInput);
+    ASSERT_EQ(second_sub->GetLocations()->Out().GetPolicy(), Location::kSameAsFirstInput);
 
     RegisterAllocator register_allocator(&allocator, &codegen, liveness);
     register_allocator.AllocateRegisters();
 
-    ASSERT_EQ(first_add->GetLiveInterval()->GetRegister(), 2);
-    ASSERT_EQ(second_add->GetLiveInterval()->GetRegister(), 2);
+    ASSERT_EQ(first_sub->GetLiveInterval()->GetRegister(), 2);
+    ASSERT_EQ(second_sub->GetLiveInterval()->GetRegister(), 2);
   }
 }
 
diff --git a/compiler/optimizing/side_effects_analysis.h b/compiler/optimizing/side_effects_analysis.h
index f1c98ac..415d10c 100644
--- a/compiler/optimizing/side_effects_analysis.h
+++ b/compiler/optimizing/side_effects_analysis.h
@@ -25,7 +25,7 @@
 class SideEffectsAnalysis : public HOptimization {
  public:
   explicit SideEffectsAnalysis(HGraph* graph)
-      : HOptimization(graph, true, "SideEffects"),
+      : HOptimization(graph, true, kSideEffectsAnalysisPassName),
         graph_(graph),
         block_effects_(graph->GetArena(), graph->GetBlocks().Size(), SideEffects::None()),
         loop_effects_(graph->GetArena(), graph->GetBlocks().Size(), SideEffects::None()) {}
@@ -38,6 +38,8 @@
 
   bool HasRun() const { return has_run_; }
 
+  static constexpr const char* kSideEffectsAnalysisPassName = "SideEffects";
+
  private:
   void UpdateLoopEffects(HLoopInformation* info, SideEffects effects);
 
diff --git a/compiler/optimizing/ssa_builder.h b/compiler/optimizing/ssa_builder.h
index 148e959..f50da46 100644
--- a/compiler/optimizing/ssa_builder.h
+++ b/compiler/optimizing/ssa_builder.h
@@ -60,6 +60,8 @@
 
   static HInstruction* GetReferenceTypeEquivalent(HInstruction* instruction);
 
+  static constexpr const char* kSsaBuilderPassName = "ssa_builder";
+
  private:
   // Locals for the current block being visited.
   HEnvironment* current_locals_;
diff --git a/compiler/optimizing/ssa_liveness_analysis.cc b/compiler/optimizing/ssa_liveness_analysis.cc
index bebb73b..d009390 100644
--- a/compiler/optimizing/ssa_liveness_analysis.cc
+++ b/compiler/optimizing/ssa_liveness_analysis.cc
@@ -477,12 +477,12 @@
   }
 }
 
-Location LiveInterval::GetLocationAt(size_t position) const {
+Location LiveInterval::GetLocationAt(size_t position) {
   return GetIntervalAt(position).ToLocation();
 }
 
-const LiveInterval& LiveInterval::GetIntervalAt(size_t position) const {
-  const LiveInterval* current = this;
+const LiveInterval& LiveInterval::GetIntervalAt(size_t position) {
+  LiveInterval* current = this;
   while (!current->Covers(position)) {
     current = current->GetNextSibling();
     DCHECK(current != nullptr);
diff --git a/compiler/optimizing/ssa_liveness_analysis.h b/compiler/optimizing/ssa_liveness_analysis.h
index 0e68a61..9ff2f20 100644
--- a/compiler/optimizing/ssa_liveness_analysis.h
+++ b/compiler/optimizing/ssa_liveness_analysis.h
@@ -322,18 +322,8 @@
     return last_range_->GetEnd() <= position;
   }
 
-  bool Covers(size_t position) const {
-    if (IsDeadAt(position)) {
-      return false;
-    }
-    LiveRange* current = first_range_;
-    while (current != nullptr) {
-      if (position >= current->GetStart() && position < current->GetEnd()) {
-        return true;
-      }
-      current = current->GetNext();
-    }
-    return false;
+  bool Covers(size_t position) {
+    return !IsDeadAt(position) && FindRangeAt(position) != nullptr;
   }
 
   /**
@@ -345,19 +335,19 @@
     LiveRange* my_range = first_range_;
     LiveRange* other_range = other->first_range_;
     do {
-      if (my_range->IntersectsWith(*other_range)) {
-        return std::max(my_range->GetStart(), other_range->GetStart());
-      } else if (my_range->IsBefore(*other_range)) {
+      if (my_range->IsBefore(*other_range)) {
         my_range = my_range->GetNext();
         if (my_range == nullptr) {
           return kNoLifetime;
         }
-      } else {
-        DCHECK(other_range->IsBefore(*my_range));
+      } else if (other_range->IsBefore(*my_range)) {
         other_range = other_range->GetNext();
         if (other_range == nullptr) {
           return kNoLifetime;
         }
+      } else {
+        DCHECK(my_range->IntersectsWith(*other_range));
+        return std::max(my_range->GetStart(), other_range->GetStart());
       }
     } while (true);
   }
@@ -466,6 +456,7 @@
     new_interval->parent_ = parent_;
 
     new_interval->first_use_ = first_use_;
+    last_visited_range_ = nullptr;
     LiveRange* current = first_range_;
     LiveRange* previous = nullptr;
     // Iterate over the ranges, and either find a range that covers this position, or
@@ -569,10 +560,10 @@
   Location ToLocation() const;
 
   // Returns the location of the interval following its siblings at `position`.
-  Location GetLocationAt(size_t position) const;
+  Location GetLocationAt(size_t position);
 
   // Finds the interval that covers `position`.
-  const LiveInterval& GetIntervalAt(size_t position) const;
+  const LiveInterval& GetIntervalAt(size_t position);
 
   // Returns whether `other` and `this` share the same kind of register.
   bool SameRegisterKind(Location other) const;
@@ -698,6 +689,7 @@
       : allocator_(allocator),
         first_range_(nullptr),
         last_range_(nullptr),
+        last_visited_range_(nullptr),
         first_use_(nullptr),
         type_(type),
         next_sibling_(nullptr),
@@ -711,6 +703,41 @@
         high_or_low_interval_(nullptr),
         defined_by_(defined_by) {}
 
+  // Returns a LiveRange covering the given position or nullptr if no such range
+  // exists in the interval.
+  // This is a linear search optimized for multiple queries in a non-decreasing
+  // position order typical for linear scan register allocation.
+  LiveRange* FindRangeAt(size_t position) {
+    // Make sure operations on the interval didn't leave us with a cached result
+    // from a sibling.
+    if (kIsDebugBuild) {
+      if (last_visited_range_ != nullptr) {
+        DCHECK_GE(last_visited_range_->GetStart(), GetStart());
+        DCHECK_LE(last_visited_range_->GetEnd(), GetEnd());
+      }
+    }
+
+    // If this method was called earlier on a lower position, use that result as
+    // a starting point to save time. However, linear scan performs 3 scans:
+    // integers, floats, and resolution. Instead of resetting at the beginning
+    // of a scan, we do it here.
+    LiveRange* current;
+    if (last_visited_range_ != nullptr && position >= last_visited_range_->GetStart()) {
+      current = last_visited_range_;
+    } else {
+      current = first_range_;
+    }
+    while (current != nullptr && current->GetEnd() <= position) {
+      current = current->GetNext();
+    }
+    last_visited_range_ = current;
+    if (current != nullptr && position >= current->GetStart()) {
+      return current;
+    } else {
+      return nullptr;
+    }
+  }
+
   ArenaAllocator* const allocator_;
 
   // Ranges of this interval. We need a quick access to the last range to test
@@ -718,6 +745,10 @@
   LiveRange* first_range_;
   LiveRange* last_range_;
 
+  // Last visited range. This is a range search optimization leveraging the fact
+  // that the register allocator does a linear scan through the intervals.
+  LiveRange* last_visited_range_;
+
   // Uses of this interval. Note that this linked list is shared amongst siblings.
   UsePosition* first_use_;
 
@@ -816,6 +847,8 @@
     return number_of_ssa_values_;
   }
 
+  static constexpr const char* kLivenessPassName = "liveness";
+
  private:
   // Linearize the graph so that:
   // (1): a block is always after its dominator,
diff --git a/compiler/optimizing/ssa_phi_elimination.h b/compiler/optimizing/ssa_phi_elimination.h
index 88a5279..c4b63ab 100644
--- a/compiler/optimizing/ssa_phi_elimination.h
+++ b/compiler/optimizing/ssa_phi_elimination.h
@@ -29,7 +29,7 @@
 class SsaDeadPhiElimination : public HOptimization {
  public:
   explicit SsaDeadPhiElimination(HGraph* graph)
-      : HOptimization(graph, true, "dead_phi_elimination"),
+      : HOptimization(graph, true, kSsaDeadPhiEliminationPassName),
         worklist_(graph->GetArena(), kDefaultWorklistSize) {}
 
   void Run() OVERRIDE;
@@ -37,6 +37,8 @@
   void MarkDeadPhis();
   void EliminateDeadPhis();
 
+  static constexpr const char* kSsaDeadPhiEliminationPassName = "dead_phi_elimination";
+
  private:
   GrowableArray<HPhi*> worklist_;
 
@@ -54,11 +56,13 @@
 class SsaRedundantPhiElimination : public HOptimization {
  public:
   explicit SsaRedundantPhiElimination(HGraph* graph)
-      : HOptimization(graph, true, "redundant_phi_elimination"),
+      : HOptimization(graph, true, kSsaRedundantPhiEliminationPassName),
         worklist_(graph->GetArena(), kDefaultWorklistSize) {}
 
   void Run() OVERRIDE;
 
+  static constexpr const char* kSsaRedundantPhiEliminationPassName = "redundant_phi_elimination";
+
  private:
   GrowableArray<HPhi*> worklist_;
 
diff --git a/compiler/utils/arm/assembler_arm.cc b/compiler/utils/arm/assembler_arm.cc
index a52e6eb..a02191b 100644
--- a/compiler/utils/arm/assembler_arm.cc
+++ b/compiler/utils/arm/assembler_arm.cc
@@ -385,12 +385,24 @@
   // Push callee saves and link register.
   RegList push_list = 1 << LR;
   size_t pushed_values = 1;
+  int32_t min_s = kNumberOfSRegisters;
+  int32_t max_s = -1;
   for (size_t i = 0; i < callee_save_regs.size(); i++) {
-    Register reg = callee_save_regs.at(i).AsArm().AsCoreRegister();
-    push_list |= 1 << reg;
-    pushed_values++;
+    if (callee_save_regs.at(i).AsArm().IsCoreRegister()) {
+      Register reg = callee_save_regs.at(i).AsArm().AsCoreRegister();
+      push_list |= 1 << reg;
+      pushed_values++;
+    } else {
+      CHECK(callee_save_regs.at(i).AsArm().IsSRegister());
+      min_s = std::min(static_cast<int>(callee_save_regs.at(i).AsArm().AsSRegister()), min_s);
+      max_s = std::max(static_cast<int>(callee_save_regs.at(i).AsArm().AsSRegister()), max_s);
+    }
   }
   PushList(push_list);
+  if (max_s != -1) {
+    pushed_values += 1 + max_s - min_s;
+    vpushs(static_cast<SRegister>(min_s), 1 + max_s - min_s);
+  }
 
   // Increase frame to required size.
   CHECK_GT(frame_size, pushed_values * kFramePointerSize);  // Must at least have space for Method*.
@@ -427,10 +439,22 @@
   // Compute callee saves to pop and PC.
   RegList pop_list = 1 << PC;
   size_t pop_values = 1;
+  int32_t min_s = kNumberOfSRegisters;
+  int32_t max_s = -1;
   for (size_t i = 0; i < callee_save_regs.size(); i++) {
-    Register reg = callee_save_regs.at(i).AsArm().AsCoreRegister();
-    pop_list |= 1 << reg;
-    pop_values++;
+    if (callee_save_regs.at(i).AsArm().IsCoreRegister()) {
+      Register reg = callee_save_regs.at(i).AsArm().AsCoreRegister();
+      pop_list |= 1 << reg;
+      pop_values++;
+    } else {
+      CHECK(callee_save_regs.at(i).AsArm().IsSRegister());
+      min_s = std::min(static_cast<int>(callee_save_regs.at(i).AsArm().AsSRegister()), min_s);
+      max_s = std::max(static_cast<int>(callee_save_regs.at(i).AsArm().AsSRegister()), max_s);
+    }
+  }
+
+  if (max_s != -1) {
+    pop_values += 1 + max_s - min_s;
   }
 
   // Decrease frame to start of callee saves.
@@ -438,6 +462,10 @@
   size_t adjust = frame_size - (pop_values * kFramePointerSize);
   DecreaseFrameSize(adjust);
 
+  if (max_s != -1) {
+    vpops(static_cast<SRegister>(min_s), 1 + max_s - min_s);
+  }
+
   // Pop callee saves and PC.
   PopList(pop_list);
 }
diff --git a/compiler/utils/arm64/assembler_arm64.cc b/compiler/utils/arm64/assembler_arm64.cc
index 21014c8..58c7367 100644
--- a/compiler/utils/arm64/assembler_arm64.cc
+++ b/compiler/utils/arm64/assembler_arm64.cc
@@ -639,6 +639,7 @@
 }
 
 constexpr size_t kFramePointerSize = 8;
+constexpr unsigned int kJniRefSpillRegsSize = 11 + 8;
 
 void Arm64Assembler::BuildFrame(size_t frame_size, ManagedRegister method_reg,
                         const std::vector<ManagedRegister>& callee_save_regs,
@@ -648,7 +649,7 @@
 
   // TODO: *create APCS FP - end of FP chain;
   //       *add support for saving a different set of callee regs.
-  // For now we check that the size of callee regs vector is 11.
+  // For now we check that the size of callee regs vector is 11 core registers and 8 fp registers.
   CHECK_EQ(callee_save_regs.size(), kJniRefSpillRegsSize);
   // Increase frame to required size - must be at least space to push StackReference<Method>.
   CHECK_GT(frame_size, kJniRefSpillRegsSize * kFramePointerSize);
@@ -682,6 +683,23 @@
   reg_offset -= 8;
   StoreToOffset(X20, SP, reg_offset);
 
+  reg_offset -= 8;
+  StoreDToOffset(D15, SP, reg_offset);
+  reg_offset -= 8;
+  StoreDToOffset(D14, SP, reg_offset);
+  reg_offset -= 8;
+  StoreDToOffset(D13, SP, reg_offset);
+  reg_offset -= 8;
+  StoreDToOffset(D12, SP, reg_offset);
+  reg_offset -= 8;
+  StoreDToOffset(D11, SP, reg_offset);
+  reg_offset -= 8;
+  StoreDToOffset(D10, SP, reg_offset);
+  reg_offset -= 8;
+  StoreDToOffset(D9, SP, reg_offset);
+  reg_offset -= 8;
+  StoreDToOffset(D8, SP, reg_offset);
+
   // Move TR(Caller saved) to ETR(Callee saved). The original (ETR)X21 has been saved on stack.
   // This way we make sure that TR is not trashed by native code.
   ___ Mov(reg_x(ETR), reg_x(TR));
@@ -753,6 +771,23 @@
   reg_offset -= 8;
   LoadFromOffset(X20, SP, reg_offset);
 
+  reg_offset -= 8;
+  LoadDFromOffset(D15, SP, reg_offset);
+  reg_offset -= 8;
+  LoadDFromOffset(D14, SP, reg_offset);
+  reg_offset -= 8;
+  LoadDFromOffset(D13, SP, reg_offset);
+  reg_offset -= 8;
+  LoadDFromOffset(D12, SP, reg_offset);
+  reg_offset -= 8;
+  LoadDFromOffset(D11, SP, reg_offset);
+  reg_offset -= 8;
+  LoadDFromOffset(D10, SP, reg_offset);
+  reg_offset -= 8;
+  LoadDFromOffset(D9, SP, reg_offset);
+  reg_offset -= 8;
+  LoadDFromOffset(D8, SP, reg_offset);
+
   // Decrease frame size to start of callee saved regs.
   DecreaseFrameSize(frame_size);
 
diff --git a/compiler/utils/arm64/constants_arm64.h b/compiler/utils/arm64/constants_arm64.h
index ffb54d3..01e8be9 100644
--- a/compiler/utils/arm64/constants_arm64.h
+++ b/compiler/utils/arm64/constants_arm64.h
@@ -29,8 +29,6 @@
 namespace art {
 namespace arm64 {
 
-constexpr unsigned int kJniRefSpillRegsSize = 11;
-
 constexpr size_t kArm64BaseBufferSize = 4096;
 
 }  // namespace arm64
diff --git a/dex2oat/dex2oat.cc b/dex2oat/dex2oat.cc
index 0b1f14d..22665ea 100644
--- a/dex2oat/dex2oat.cc
+++ b/dex2oat/dex2oat.cc
@@ -1052,6 +1052,13 @@
     runtime_options.push_back(
         std::make_pair("imageinstructionset", GetInstructionSetString(instruction_set_)));
 
+    // Only allow no boot image for the runtime if we're compiling one. When we compile an app,
+    // we don't want fallback mode, it will abort as we do not push a boot classpath (it might
+    // have been stripped in preopting, anyways).
+    if (!image_) {
+      runtime_options.push_back(std::make_pair("-Xno-dex-file-fallback", nullptr));
+    }
+
     if (!CreateRuntime(runtime_options)) {
       return false;
     }
@@ -1637,9 +1644,13 @@
   }
 
   void LogCompletionTime() {
+    // Note: when creation of a runtime fails, e.g., when trying to compile an app but when there
+    //       is no image, there won't be a Runtime::Current().
     LOG(INFO) << "dex2oat took " << PrettyDuration(NanoTime() - start_ns_)
               << " (threads: " << thread_count_ << ") "
-              << driver_->GetMemoryUsageString(kIsDebugBuild || VLOG_IS_ON(compiler));
+              << ((Runtime::Current() != nullptr) ?
+                  driver_->GetMemoryUsageString(kIsDebugBuild || VLOG_IS_ON(compiler)) :
+                  "");
   }
 
   std::unique_ptr<CompilerOptions> compiler_options_;
diff --git a/disassembler/disassembler_mips.cc b/disassembler/disassembler_mips.cc
index 7442c70..3d8a567 100644
--- a/disassembler/disassembler_mips.cc
+++ b/disassembler/disassembler_mips.cc
@@ -143,27 +143,31 @@
   { kITypeMask, 61u << kOpcodeShift, "sdc1", "tO", },
 
   // Floating point.
-  { kFpMask,                kCop1 | 0, "add", "fdst" },
-  { kFpMask,                kCop1 | 1, "sub", "fdst" },
-  { kFpMask,                kCop1 | 2, "mul", "fdst" },
-  { kFpMask,                kCop1 | 3, "div", "fdst" },
-  { kFpMask | (0x1f << 16), kCop1 | 4, "sqrt", "fdst" },
-  { kFpMask | (0x1f << 16), kCop1 | 5, "abs", "fds" },
-  { kFpMask | (0x1f << 16), kCop1 | 6, "mov", "fds" },
-  { kFpMask | (0x1f << 16), kCop1 | 7, "neg", "fds" },
-  { kFpMask | (0x1f << 16), kCop1 | 8, "round.l", "fds" },
-  { kFpMask | (0x1f << 16), kCop1 | 9, "trunc.l", "fds" },
-  { kFpMask | (0x1f << 16), kCop1 | 10, "ceil.l", "fds" },
-  { kFpMask | (0x1f << 16), kCop1 | 11, "floor.l", "fds" },
-  { kFpMask | (0x1f << 16), kCop1 | 12, "round.w", "fds" },
-  { kFpMask | (0x1f << 16), kCop1 | 13, "trunc.w", "fds" },
-  { kFpMask | (0x1f << 16), kCop1 | 14, "ceil.w", "fds" },
-  { kFpMask | (0x1f << 16), kCop1 | 15, "floor.w", "fds" },
-  { kFpMask | (0x1f << 16), kCop1 | 32, "cvt.s", "fds" },
-  { kFpMask | (0x1f << 16), kCop1 | 33, "cvt.d", "fds" },
-  { kFpMask | (0x1f << 16), kCop1 | 36, "cvt.w", "fds" },
-  { kFpMask | (0x1f << 16), kCop1 | 37, "cvt.l", "fds" },
-  { kFpMask | (0x1f << 16), kCop1 | 38, "cvt.ps", "fds" },
+  { kFpMask | (0x1f << 21), kCop1 | (0x00 << 21) | 0, "mfc1", "Td" },
+  { kFpMask | (0x1f << 21), kCop1 | (0x03 << 21) | 0, "mfhc1", "Td" },
+  { kFpMask | (0x1f << 21), kCop1 | (0x04 << 21) | 0, "mtc1", "Td" },
+  { kFpMask | (0x1f << 21), kCop1 | (0x07 << 21) | 0, "mthc1", "Td" },
+  { kFpMask | (0x10 << 21), kCop1 | (0x10 << 21) | 0, "add", "fadt" },
+  { kFpMask | (0x10 << 21), kCop1 | (0x10 << 21) | 1, "sub", "fadt" },
+  { kFpMask | (0x10 << 21), kCop1 | (0x10 << 21) | 2, "mul", "fadt" },
+  { kFpMask | (0x10 << 21), kCop1 | (0x10 << 21) | 3, "div", "fadt" },
+  { kFpMask | (0x10 << 21), kCop1 | (0x10 << 21) | 4, "sqrt", "fad" },
+  { kFpMask | (0x21f << 16), kCop1 | (0x200 << 16) | 5, "abs", "fad" },
+  { kFpMask | (0x21f << 16), kCop1 | (0x200 << 16) | 6, "mov", "fad" },
+  { kFpMask | (0x21f << 16), kCop1 | (0x200 << 16) | 7, "neg", "fad" },
+  { kFpMask | (0x21f << 16), kCop1 | (0x200 << 16) | 8, "round.l", "fad" },
+  { kFpMask | (0x21f << 16), kCop1 | (0x200 << 16) | 9, "trunc.l", "fad" },
+  { kFpMask | (0x21f << 16), kCop1 | (0x200 << 16) | 10, "ceil.l", "fad" },
+  { kFpMask | (0x21f << 16), kCop1 | (0x200 << 16) | 11, "floor.l", "fad" },
+  { kFpMask | (0x21f << 16), kCop1 | (0x200 << 16) | 12, "round.w", "fad" },
+  { kFpMask | (0x21f << 16), kCop1 | (0x200 << 16) | 13, "trunc.w", "fad" },
+  { kFpMask | (0x21f << 16), kCop1 | (0x200 << 16) | 14, "ceil.w", "fad" },
+  { kFpMask | (0x21f << 16), kCop1 | (0x200 << 16) | 15, "floor.w", "fad" },
+  { kFpMask | (0x21f << 16), kCop1 | (0x200 << 16) | 32, "cvt.s", "fad" },
+  { kFpMask | (0x21f << 16), kCop1 | (0x200 << 16) | 33, "cvt.d", "fad" },
+  { kFpMask | (0x21f << 16), kCop1 | (0x200 << 16) | 36, "cvt.w", "fad" },
+  { kFpMask | (0x21f << 16), kCop1 | (0x200 << 16) | 37, "cvt.l", "fad" },
+  { kFpMask | (0x21f << 16), kCop1 | (0x200 << 16) | 38, "cvt.ps", "fad" },
 };
 
 static uint32_t ReadU32(const uint8_t* ptr) {
@@ -206,6 +210,7 @@
             break;
           case 'D': args << 'r' << rd; break;
           case 'd': args << 'f' << rd; break;
+          case 'a': args << 'f' << sa; break;
           case 'f':  // Floating point "fmt".
             {
               size_t fmt = (instruction >> 21) & 0x7;  // TODO: other fmts?
diff --git a/oatdump/oatdump.cc b/oatdump/oatdump.cc
index 11ccafb..aab4f8b 100644
--- a/oatdump/oatdump.cc
+++ b/oatdump/oatdump.cc
@@ -88,6 +88,7 @@
 
     uint32_t diff = static_cast<uint32_t>(oat_file_->End() - oat_file_->Begin());
     uint32_t oat_exec_size = diff - oat_data_size;
+    uint32_t oat_bss_size = oat_file_->BssSize();
 
     elf_output_ = OS::CreateEmptyFile(output_name_.c_str());
 
@@ -100,6 +101,8 @@
         oat_data_size,
         oat_data_size,
         oat_exec_size,
+        RoundUp(oat_data_size + oat_exec_size, kPageSize),
+        oat_bss_size,
         true,
         false));
 
@@ -311,13 +314,23 @@
                    bool dump_vmap,
                    bool disassemble_code,
                    bool absolute_addresses,
-                   const char* method_filter)
+                   const char* class_filter,
+                   const char* method_filter,
+                   bool list_classes,
+                   bool list_methods,
+                   const char* export_dex_location,
+                   uint32_t addr2instr)
     : dump_raw_mapping_table_(dump_raw_mapping_table),
       dump_raw_gc_map_(dump_raw_gc_map),
       dump_vmap_(dump_vmap),
       disassemble_code_(disassemble_code),
       absolute_addresses_(absolute_addresses),
+      class_filter_(class_filter),
       method_filter_(method_filter),
+      list_classes_(list_classes),
+      list_methods_(list_methods),
+      export_dex_location_(export_dex_location),
+      addr2instr_(addr2instr),
       class_loader_(nullptr) {}
 
   const bool dump_raw_mapping_table_;
@@ -325,27 +338,34 @@
   const bool dump_vmap_;
   const bool disassemble_code_;
   const bool absolute_addresses_;
+  const char* const class_filter_;
   const char* const method_filter_;
+  const bool list_classes_;
+  const bool list_methods_;
+  const char* const export_dex_location_;
+  uint32_t addr2instr_;
   Handle<mirror::ClassLoader>* class_loader_;
 };
 
 class OatDumper {
  public:
-  explicit OatDumper(const OatFile& oat_file, OatDumperOptions* options)
+  explicit OatDumper(const OatFile& oat_file, const OatDumperOptions& options)
     : oat_file_(oat_file),
       oat_dex_files_(oat_file.GetOatDexFiles()),
       options_(options),
+      resolved_addr2instr_(0),
       instruction_set_(oat_file_.GetOatHeader().GetInstructionSet()),
       disassembler_(Disassembler::Create(instruction_set_,
-                                         new DisassemblerOptions(options_->absolute_addresses_,
+                                         new DisassemblerOptions(options_.absolute_addresses_,
                                                                  oat_file.Begin(),
                                                                  true /* can_read_litals_ */))) {
-    CHECK(options_->class_loader_ != nullptr);
+    CHECK(options_.class_loader_ != nullptr);
+    CHECK(options_.class_filter_ != nullptr);
+    CHECK(options_.method_filter_ != nullptr);
     AddAllOffsets();
   }
 
   ~OatDumper() {
-    delete options_;
     delete disassembler_;
   }
 
@@ -380,7 +400,7 @@
 #define DUMP_OAT_HEADER_OFFSET(label, offset) \
     os << label " OFFSET:\n"; \
     os << StringPrintf("0x%08x", oat_header.offset()); \
-    if (oat_header.offset() != 0 && options_->absolute_addresses_) { \
+    if (oat_header.offset() != 0 && options_.absolute_addresses_) { \
       os << StringPrintf(" (%p)", oat_file_.Begin() + oat_header.offset()); \
     } \
     os << StringPrintf("\n\n");
@@ -426,7 +446,7 @@
       os << "\n";
     }
 
-    if (options_->absolute_addresses_) {
+    if (options_.absolute_addresses_) {
       os << "BEGIN:\n";
       os << reinterpret_cast<const void*>(oat_file_.Begin()) << "\n\n";
 
@@ -439,11 +459,26 @@
 
     os << std::flush;
 
+    // If set, adjust relative address to be searched
+    if (options_.addr2instr_ != 0) {
+      resolved_addr2instr_ = options_.addr2instr_ + oat_header.GetExecutableOffset();
+      os << "SEARCH ADDRESS (executable offset + input):\n";
+      os << StringPrintf("0x%08x\n\n", resolved_addr2instr_);
+    }
+
     for (size_t i = 0; i < oat_dex_files_.size(); i++) {
       const OatFile::OatDexFile* oat_dex_file = oat_dex_files_[i];
       CHECK(oat_dex_file != nullptr);
-      if (!DumpOatDexFile(os, *oat_dex_file)) {
-        success = false;
+
+      // If file export selected skip file analysis
+      if (options_.export_dex_location_) {
+        if (!ExportDexFile(os, *oat_dex_file)) {
+          success = false;
+        }
+      } else {
+        if (!DumpOatDexFile(os, *oat_dex_file)) {
+          success = false;
+        }
       }
     }
     os << std::flush;
@@ -553,6 +588,7 @@
 
   bool DumpOatDexFile(std::ostream& os, const OatFile::OatDexFile& oat_dex_file) {
     bool success = true;
+    bool stop_analysis = false;
     os << "OatDexFile:\n";
     os << StringPrintf("location: %s\n", oat_dex_file.GetDexFileLocation().c_str());
     os << StringPrintf("checksum: 0x%08x\n", oat_dex_file.GetDexFileLocationChecksum());
@@ -571,6 +607,12 @@
          class_def_index++) {
       const DexFile::ClassDef& class_def = dex_file->GetClassDef(class_def_index);
       const char* descriptor = dex_file->GetClassDescriptor(class_def);
+
+      // TODO: Support regex
+      if (DescriptorToDot(descriptor).find(options_.class_filter_) == std::string::npos) {
+        continue;
+      }
+
       uint32_t oat_class_offset = oat_dex_file.GetOatClassOffset(class_def_index);
       const OatFile::OatClass oat_class = oat_dex_file.GetOatClass(class_def_index);
       os << StringPrintf("%zd: %s (offset=0x%08x) (type_idx=%d)",
@@ -580,15 +622,98 @@
       // TODO: include bitmap here if type is kOatClassSomeCompiled?
       Indenter indent_filter(os.rdbuf(), kIndentChar, kIndentBy1Count);
       std::ostream indented_os(&indent_filter);
-      if (!DumpOatClass(indented_os, oat_class, *(dex_file.get()), class_def)) {
+      if (options_.list_classes_) continue;
+      if (!DumpOatClass(indented_os, oat_class, *(dex_file.get()), class_def, &stop_analysis)) {
         success = false;
       }
+      if (stop_analysis) {
+        os << std::flush;
+        return success;
+      }
     }
 
     os << std::flush;
     return success;
   }
 
+  bool ExportDexFile(std::ostream& os, const OatFile::OatDexFile& oat_dex_file) {
+    std::string error_msg;
+    std::string dex_file_location = oat_dex_file.GetDexFileLocation();
+
+    std::unique_ptr<const DexFile> dex_file(oat_dex_file.OpenDexFile(&error_msg));
+    if (dex_file == nullptr) {
+      os << "Failed to open dex file '" << dex_file_location << "': " << error_msg;
+      return false;
+    }
+    size_t fsize = oat_dex_file.FileSize();
+
+    // Some quick checks just in case
+    if (fsize == 0 || fsize < sizeof(DexFile::Header)) {
+      os << "Invalid dex file\n";
+      return false;
+    }
+
+    // Verify output directory exists
+    if (!OS::DirectoryExists(options_.export_dex_location_)) {
+      // TODO: Extend OS::DirectoryExists if symlink support is required
+      os << options_.export_dex_location_ << " output directory not found or symlink\n";
+      return false;
+    }
+
+    // Beautify path names
+    if (dex_file_location.size() > PATH_MAX || dex_file_location.size() <= 0) {
+      return false;
+    }
+
+    std::string dex_orig_name;
+    size_t dex_orig_pos = dex_file_location.rfind('/');
+    if (dex_orig_pos == std::string::npos)
+      dex_orig_name = dex_file_location;
+    else
+      dex_orig_name = dex_file_location.substr(dex_orig_pos + 1);
+
+    // A more elegant approach to efficiently name user installed apps is welcome
+    if (dex_orig_name.size() == 8 && !dex_orig_name.compare("base.apk")) {
+      dex_file_location.erase(dex_orig_pos, strlen("base.apk") + 1);
+      size_t apk_orig_pos = dex_file_location.rfind('/');
+      if (apk_orig_pos != std::string::npos) {
+        dex_orig_name = dex_file_location.substr(++apk_orig_pos);
+      }
+    }
+
+    std::string out_dex_path(options_.export_dex_location_);
+    if (out_dex_path.back() != '/') {
+      out_dex_path.append("/");
+    }
+    out_dex_path.append(dex_orig_name);
+    out_dex_path.append("_export.dex");
+    if (out_dex_path.length() > PATH_MAX) {
+      return false;
+    }
+
+    std::unique_ptr<File> file(OS::CreateEmptyFile(out_dex_path.c_str()));
+    if (file.get() == nullptr) {
+      os << "Failed to open output dex file " << out_dex_path;
+      return false;
+    }
+
+    if (!file->WriteFully(dex_file->Begin(), fsize)) {
+      os << "Failed to write dex file";
+      file->Erase();
+      return false;
+    }
+
+    if (file->FlushCloseOrErase() != 0) {
+      os << "Flush and close failed";
+      return false;
+    }
+
+    os << StringPrintf("Dex file exported at %s (%zd bytes)\n", out_dex_path.c_str(), fsize);
+    os << std::flush;
+
+    return true;
+  }
+
   static void SkipAllFields(ClassDataItemIterator& it) {
     while (it.HasNextStaticField()) {
       it.Next();
@@ -599,8 +724,9 @@
   }
 
   bool DumpOatClass(std::ostream& os, const OatFile::OatClass& oat_class, const DexFile& dex_file,
-                    const DexFile::ClassDef& class_def) {
+                    const DexFile::ClassDef& class_def, bool* stop_analysis) {
     bool success = true;
+    bool addr_found = false;
     const uint8_t* class_data = dex_file.GetClassData(class_def);
     if (class_data == nullptr) {  // empty class such as a marker interface?
       os << std::flush;
@@ -612,18 +738,26 @@
     while (it.HasNextDirectMethod()) {
       if (!DumpOatMethod(os, class_def, class_method_index, oat_class, dex_file,
                          it.GetMemberIndex(), it.GetMethodCodeItem(),
-                         it.GetRawMemberAccessFlags())) {
+                         it.GetRawMemberAccessFlags(), &addr_found)) {
         success = false;
       }
+      if (addr_found) {
+        *stop_analysis = true;
+        return success;
+      }
       class_method_index++;
       it.Next();
     }
     while (it.HasNextVirtualMethod()) {
       if (!DumpOatMethod(os, class_def, class_method_index, oat_class, dex_file,
                          it.GetMemberIndex(), it.GetMethodCodeItem(),
-                         it.GetRawMemberAccessFlags())) {
+                         it.GetRawMemberAccessFlags(), &addr_found)) {
         success = false;
       }
+      if (addr_found) {
+        *stop_analysis = true;
+        return success;
+      }
       class_method_index++;
       it.Next();
     }
@@ -641,20 +775,39 @@
                      uint32_t class_method_index,
                      const OatFile::OatClass& oat_class, const DexFile& dex_file,
                      uint32_t dex_method_idx, const DexFile::CodeItem* code_item,
-                     uint32_t method_access_flags) {
+                     uint32_t method_access_flags, bool* addr_found) {
     bool success = true;
-    std::string pretty_method = PrettyMethod(dex_method_idx, dex_file, true);
-    if (pretty_method.find(options_->method_filter_) == std::string::npos) {
+
+    // TODO: Support regex
+    std::string method_name = dex_file.GetMethodName(dex_file.GetMethodId(dex_method_idx));
+    if (method_name.find(options_.method_filter_) == std::string::npos) {
       return success;
     }
 
+    std::string pretty_method = PrettyMethod(dex_method_idx, dex_file, true);
     os << StringPrintf("%d: %s (dex_method_idx=%d)\n",
                        class_method_index, pretty_method.c_str(),
                        dex_method_idx);
+    if (options_.list_methods_) return success;
+
     Indenter indent1_filter(os.rdbuf(), kIndentChar, kIndentBy1Count);
     std::unique_ptr<std::ostream> indent1_os(new std::ostream(&indent1_filter));
     Indenter indent2_filter(indent1_os->rdbuf(), kIndentChar, kIndentBy1Count);
     std::unique_ptr<std::ostream> indent2_os(new std::ostream(&indent2_filter));
+
+    uint32_t oat_method_offsets_offset = oat_class.GetOatMethodOffsetsOffset(class_method_index);
+    const OatMethodOffsets* oat_method_offsets = oat_class.GetOatMethodOffsets(class_method_index);
+    const OatFile::OatMethod oat_method = oat_class.GetOatMethod(class_method_index);
+    uint32_t code_offset = oat_method.GetCodeOffset();
+    uint32_t code_size = oat_method.GetQuickCodeSize();
+    if (resolved_addr2instr_ != 0) {
+      if (resolved_addr2instr_ > code_offset + code_size) {
+        return success;
+      } else {
+        *addr_found = true;  // stop analyzing file at next iteration
+      }
+    }
+
     {
       *indent1_os << "DEX CODE:\n";
       DumpDexCode(*indent2_os, dex_file, code_item);
@@ -666,13 +819,9 @@
       verifier.reset(DumpVerifier(*indent2_os, dex_method_idx, &dex_file, class_def, code_item,
                                   method_access_flags));
     }
-
-    uint32_t oat_method_offsets_offset = oat_class.GetOatMethodOffsetsOffset(class_method_index);
-    const OatMethodOffsets* oat_method_offsets = oat_class.GetOatMethodOffsets(class_method_index);
-    const OatFile::OatMethod oat_method = oat_class.GetOatMethod(class_method_index);
     {
       *indent1_os << "OatMethodOffsets ";
-      if (options_->absolute_addresses_) {
+      if (options_.absolute_addresses_) {
         *indent1_os << StringPrintf("%p ", oat_method_offsets);
       }
       *indent1_os << StringPrintf("(offset=0x%08x)\n", oat_method_offsets_offset);
@@ -685,7 +834,6 @@
         return false;
       }
 
-      uint32_t code_offset = oat_method.GetCodeOffset();
       *indent2_os << StringPrintf("code_offset: 0x%08x ", code_offset);
       uint32_t aligned_code_begin = AlignCodeOffset(oat_method.GetCodeOffset());
       if (aligned_code_begin > oat_file_.Size()) {
@@ -697,7 +845,7 @@
       *indent2_os << "\n";
 
       *indent2_os << "gc_map: ";
-      if (options_->absolute_addresses_) {
+      if (options_.absolute_addresses_) {
         *indent2_os << StringPrintf("%p ", oat_method.GetGcMap());
       }
       uint32_t gc_map_offset = oat_method.GetGcMapOffset();
@@ -707,7 +855,7 @@
                                     "gc map table offset 0x%08x is past end of file 0x%08zx.\n",
                                     gc_map_offset, oat_file_.Size());
         success = false;
-      } else if (options_->dump_raw_gc_map_) {
+      } else if (options_.dump_raw_gc_map_) {
         Indenter indent3_filter(indent2_os->rdbuf(), kIndentChar, kIndentBy1Count);
         std::ostream indent3_os(&indent3_filter);
         DumpGcMap(indent3_os, oat_method, code_item);
@@ -718,7 +866,7 @@
       uint32_t method_header_offset = oat_method.GetOatQuickMethodHeaderOffset();
       const OatQuickMethodHeader* method_header = oat_method.GetOatQuickMethodHeader();
 
-      if (options_->absolute_addresses_) {
+      if (options_.absolute_addresses_) {
         *indent1_os << StringPrintf("%p ", method_header);
       }
       *indent1_os << StringPrintf("(offset=0x%08x)\n", method_header_offset);
@@ -732,7 +880,7 @@
       }
 
       *indent2_os << "mapping_table: ";
-      if (options_->absolute_addresses_) {
+      if (options_.absolute_addresses_) {
         *indent2_os << StringPrintf("%p ", oat_method.GetMappingTable());
       }
       uint32_t mapping_table_offset = oat_method.GetMappingTableOffset();
@@ -744,14 +892,14 @@
                                     mapping_table_offset, oat_file_.Size(),
                                     oat_method.GetMappingTableOffsetOffset());
         success = false;
-      } else if (options_->dump_raw_mapping_table_) {
+      } else if (options_.dump_raw_mapping_table_) {
         Indenter indent3_filter(indent2_os->rdbuf(), kIndentChar, kIndentBy1Count);
         std::ostream indent3_os(&indent3_filter);
         DumpMappingTable(indent3_os, oat_method);
       }
 
       *indent2_os << "vmap_table: ";
-      if (options_->absolute_addresses_) {
+      if (options_.absolute_addresses_) {
         *indent2_os << StringPrintf("%p ", oat_method.GetVmapTable());
       }
       uint32_t vmap_table_offset = oat_method.GetVmapTableOffset();
@@ -763,7 +911,7 @@
                                     vmap_table_offset, oat_file_.Size(),
                                     oat_method.GetVmapTableOffsetOffset());
         success = false;
-      } else if (options_->dump_vmap_) {
+      } else if (options_.dump_vmap_) {
         DumpVmapData(*indent2_os, oat_method, code_item);
       }
     }
@@ -794,12 +942,10 @@
         success = false;
       } else {
         const void* code = oat_method.GetQuickCode();
-        uint32_t code_size = oat_method.GetQuickCodeSize();
-        uint32_t code_offset = oat_method.GetCodeOffset();
         uint32_t aligned_code_begin = AlignCodeOffset(code_offset);
         uint64_t aligned_code_end = aligned_code_begin + code_size;
 
-        if (options_->absolute_addresses_) {
+        if (options_.absolute_addresses_) {
           *indent1_os << StringPrintf("%p ", code);
         }
         *indent1_os << StringPrintf("(code_offset=0x%08x size_offset=0x%08x size=%u)%s\n",
@@ -820,7 +966,7 @@
                                       aligned_code_end, oat_file_.Size(),
                                       code_size, code_size_offset);
           success = false;
-          if (options_->disassemble_code_) {
+          if (options_.disassemble_code_) {
             if (code_size_offset + kPrologueBytes <= oat_file_.Size()) {
               DumpCode(*indent2_os, verifier.get(), oat_method, code_item, true, kPrologueBytes);
             }
@@ -832,12 +978,12 @@
                                       code_size, kMaxCodeSize,
                                       code_size, code_size_offset);
           success = false;
-          if (options_->disassemble_code_) {
+          if (options_.disassemble_code_) {
             if (code_size_offset + kPrologueBytes <= oat_file_.Size()) {
               DumpCode(*indent2_os, verifier.get(), oat_method, code_item, true, kPrologueBytes);
             }
           }
-        } else if (options_->disassemble_code_) {
+        } else if (options_.disassemble_code_) {
           DumpCode(*indent2_os, verifier.get(), oat_method, code_item, !success, 0);
         }
       }
@@ -1175,7 +1321,8 @@
       size_t i = 0;
       while (i < code_item->insns_size_in_code_units_) {
         const Instruction* instruction = Instruction::At(&code_item->insns_[i]);
-        os << StringPrintf("0x%04zx: %s\n", i, instruction->DumpString(&dex_file).c_str());
+        os << StringPrintf("0x%04zx: ", i) << instruction->DumpHexLE(5)
+           << StringPrintf("\t| %s\n", instruction->DumpString(&dex_file).c_str());
         i += instruction->SizeInCodeUnits();
       }
     }
@@ -1191,10 +1338,10 @@
       StackHandleScope<1> hs(soa.Self());
       Handle<mirror::DexCache> dex_cache(
           hs.NewHandle(Runtime::Current()->GetClassLinker()->FindDexCache(*dex_file)));
-      DCHECK(options_->class_loader_ != nullptr);
+      DCHECK(options_.class_loader_ != nullptr);
       return verifier::MethodVerifier::VerifyMethodAndDump(soa.Self(), os, dex_method_idx, dex_file,
                                                            dex_cache,
-                                                           *options_->class_loader_,
+                                                           *options_.class_loader_,
                                                            &class_def, code_item,
                                                            NullHandle<mirror::ArtMethod>(),
                                                            method_access_flags);
@@ -1237,7 +1384,8 @@
 
   const OatFile& oat_file_;
   const std::vector<const OatFile::OatDexFile*> oat_dex_files_;
-  const OatDumperOptions* options_;
+  const OatDumperOptions& options_;
+  uint32_t resolved_addr2instr_;
   InstructionSet instruction_set_;
   std::set<uintptr_t> offsets_;
   Disassembler* disassembler_;
@@ -1335,7 +1483,7 @@
 
     stats_.oat_file_bytes = oat_file->Size();
 
-    oat_dumper_.reset(new OatDumper(*oat_file, oat_dumper_options_.release()));
+    oat_dumper_.reset(new OatDumper(*oat_file, *oat_dumper_options_));
 
     for (const OatFile::OatDexFile* oat_dex_file : oat_file->GetOatDexFiles()) {
       CHECK(oat_dex_file != nullptr);
@@ -2045,17 +2193,18 @@
       soa.Decode<mirror::ClassLoader*>(class_loader));
   options->class_loader_ = &loader_handle;
 
-  OatDumper oat_dumper(*oat_file, options);
+  OatDumper oat_dumper(*oat_file, *options);
   bool success = oat_dumper.Dump(*os);
   return (success) ? EXIT_SUCCESS : EXIT_FAILURE;
 }
 
 static int DumpOatWithoutRuntime(OatFile* oat_file, OatDumperOptions* options, std::ostream* os) {
+  CHECK(oat_file != nullptr && options != nullptr);
   // No image = no class loader.
   NullHandle<mirror::ClassLoader> null_class_loader;
   options->class_loader_ = &null_class_loader;
 
-  OatDumper oat_dumper(*oat_file, options);
+  OatDumper oat_dumper(*oat_file, *options);
   bool success = oat_dumper.Dump(*os);
   return (success) ? EXIT_SUCCESS : EXIT_FAILURE;
 }
@@ -2127,8 +2276,21 @@
     } else if (option.starts_with("--symbolize=")) {
       oat_filename_ = option.substr(strlen("--symbolize=")).data();
       symbolize_ = true;
+    } else if (option.starts_with("--class-filter=")) {
+      class_filter_ = option.substr(strlen("--class-filter=")).data();
     } else if (option.starts_with("--method-filter=")) {
       method_filter_ = option.substr(strlen("--method-filter=")).data();
+    } else if (option.starts_with("--list-classes")) {
+      list_classes_ = true;
+    } else if (option.starts_with("--list-methods")) {
+      list_methods_ = true;
+    } else if (option.starts_with("--export-dex-to=")) {
+      export_dex_location_ = option.substr(strlen("--export-dex-to=")).data();
+    } else if (option.starts_with("--addr2instr=")) {
+      if (!ParseUint(option.substr(strlen("--addr2instr=")).data(), &addr2instr_)) {
+        *error_msg = "Address conversion failed";
+        return kParseError;
+      }
     } else {
       return kParseUnknownArgument;
     }
@@ -2191,8 +2353,29 @@
         "  --no-disassemble may be used to disable disassembly.\n"
         "      Example: --no-disassemble\n"
         "\n"
+        "  --list-classes may be used to list target file classes (can be used with filters).\n"
+        "      Example: --list-classes\n"
+        "      Example: --list-classes --class-filter=com.example.foo\n"
+        "\n"
+        "  --list-methods may be used to list target file methods (can be used with filters).\n"
+        "      Example: --list-methods\n"
+        "      Example: --list-methods --class-filter=com.example --method-filter=foo\n"
+        "\n"
+        "  --symbolize=<file.oat>: output a copy of file.oat with elf symbols included.\n"
+        "      Example: --symbolize=/system/framework/boot.oat\n"
+        "\n"
+        "  --class-filter=<class name>: only dumps classes that contain the filter.\n"
+        "      Example: --class-filter=com.example.foo\n"
+        "\n"
         "  --method-filter=<method name>: only dumps methods that contain the filter.\n"
         "      Example: --method-filter=foo\n"
+        "\n"
+        "  --export-dex-to=<directory>: may be used to export oat embedded dex files.\n"
+        "      Example: --export-dex-to=/data/local/tmp\n"
+        "\n"
+        "  --addr2instr=<address>: output matching method disassembled code from relative\n"
+        "                          address (e.g. PC from crash dump)\n"
+        "      Example: --addr2instr=0x00001a3b\n"
         "\n";
 
     return usage;
@@ -2200,6 +2383,7 @@
 
  public:
   const char* oat_filename_ = nullptr;
+  const char* class_filter_ = "";
   const char* method_filter_ = "";
   const char* image_location_ = nullptr;
   std::string elf_filename_prefix_;
@@ -2208,6 +2392,10 @@
   bool dump_vmap_ = true;
   bool disassemble_code_ = true;
   bool symbolize_ = false;
+  bool list_classes_ = false;
+  bool list_methods_ = false;
+  uint32_t addr2instr_ = 0;
+  const char* export_dex_location_ = nullptr;
 };
 
 struct OatdumpMain : public CmdlineMain<OatdumpArgs> {
@@ -2223,7 +2411,12 @@
         args_->dump_vmap_,
         args_->disassemble_code_,
         absolute_addresses,
-        args_->method_filter_));
+        args_->class_filter_,
+        args_->method_filter_,
+        args_->list_classes_,
+        args_->list_methods_,
+        args_->export_dex_location_,
+        args_->addr2instr_));
 
     return (args_->boot_image_location_ != nullptr || args_->image_location_ != nullptr) &&
           !args_->symbolize_;
@@ -2240,7 +2433,7 @@
     } else {
       return DumpOat(nullptr,
                      args_->oat_filename_,
-                     oat_dumper_options_.release(),
+                     oat_dumper_options_.get(),
                      args_->os_) == EXIT_SUCCESS;
     }
   }
@@ -2251,11 +2444,11 @@
     if (args_->oat_filename_ != nullptr) {
       return DumpOat(runtime,
                      args_->oat_filename_,
-                     oat_dumper_options_.release(),
+                     oat_dumper_options_.get(),
                      args_->os_) == EXIT_SUCCESS;
     }
 
-    return DumpImage(runtime, args_->image_location_, oat_dumper_options_.release(), args_->os_)
+    return DumpImage(runtime, args_->image_location_, oat_dumper_options_.get(), args_->os_)
       == EXIT_SUCCESS;
   }
 
diff --git a/runtime/Android.mk b/runtime/Android.mk
index c647cc2..c5cf890 100644
--- a/runtime/Android.mk
+++ b/runtime/Android.mk
@@ -44,6 +44,7 @@
   elf_file.cc \
   gc/allocator/dlmalloc.cc \
   gc/allocator/rosalloc.cc \
+  gc/accounting/bitmap.cc \
   gc/accounting/card_table.cc \
   gc/accounting/heap_bitmap.cc \
   gc/accounting/mod_union_table.cc \
@@ -88,6 +89,9 @@
   jdwp/jdwp_socket.cc \
   jdwp/object_registry.cc \
   jni_env_ext.cc \
+  jit/jit.cc \
+  jit/jit_code_cache.cc \
+  jit/jit_instrumentation.cc \
   jni_internal.cc \
   jobject_comparator.cc \
   mem_map.cc \
@@ -298,6 +302,7 @@
   base/unix_file/fd_file.h \
   dex_file.h \
   dex_instruction.h \
+  dex_instruction_utils.h \
   gc_root.h \
   gc/allocator/rosalloc.h \
   gc/collector/gc_type.h \
diff --git a/runtime/arch/arm/quick_entrypoints_arm.S b/runtime/arch/arm/quick_entrypoints_arm.S
index fec1ce5..539b607 100644
--- a/runtime/arch/arm/quick_entrypoints_arm.S
+++ b/runtime/arch/arm/quick_entrypoints_arm.S
@@ -890,7 +890,7 @@
     // r1: pointer to the bottom of the used area of the alloca, can restore stack till there.
 
     // Check for error = 0.
-    cbz r0, .Lentry_error
+    cbz r0, .Lexception_in_native
 
     // Release part of the alloca.
     mov sp, r1
@@ -920,10 +920,6 @@
 
     blx artQuickGenericJniEndTrampoline
 
-    // Tear down the alloca.
-    mov sp, r10
-    .cfi_def_cfa_register sp
-
     // Restore self pointer.
     mov r9, r11
 
@@ -931,6 +927,10 @@
     ldr r2, [r9, #THREAD_EXCEPTION_OFFSET]  @ load Thread::Current()->exception_
     cbnz r2, .Lexception_in_native
 
+    // Tear down the alloca.
+    mov sp, r10
+    .cfi_def_cfa_register sp
+
     // Tear down the callee-save frame. Skip arg registers.
     add     sp, #FRAME_SIZE_REFS_AND_ARGS_CALLEE_SAVE-FRAME_SIZE_REFS_ONLY_CALLEE_SAVE
     .cfi_adjust_cfa_offset -(FRAME_SIZE_REFS_AND_ARGS_CALLEE_SAVE-FRAME_SIZE_REFS_ONLY_CALLEE_SAVE)
@@ -943,14 +943,11 @@
     .cfi_def_cfa_register r10
     .cfi_adjust_cfa_offset FRAME_SIZE_REFS_AND_ARGS_CALLEE_SAVE-FRAME_SIZE_REFS_ONLY_CALLEE_SAVE
 
-.Lentry_error:
-    mov sp, r10
-    .cfi_def_cfa_register sp
-    mov r9, r11
 .Lexception_in_native:
-    RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME
+    ldr sp, [r9, #THREAD_TOP_QUICK_FRAME_OFFSET]
+    .cfi_def_cfa_register sp
+    # This will create a new save-all frame, required by the runtime.
     DELIVER_PENDING_EXCEPTION
-
 END art_quick_generic_jni_trampoline
 
     .extern artQuickToInterpreterBridge
diff --git a/runtime/arch/arm64/instruction_set_features_arm64.h b/runtime/arch/arm64/instruction_set_features_arm64.h
index b0c66b3..f6bfee7 100644
--- a/runtime/arch/arm64/instruction_set_features_arm64.h
+++ b/runtime/arch/arm64/instruction_set_features_arm64.h
@@ -61,6 +61,15 @@
       return fix_cortex_a53_835769_;
   }
 
+  // TODO: Tune this on a per CPU basis. For now, we pessimistically assume
+  // that all ARM64 CPUs prefer explicit memory barriers over acquire-release.
+  //
+  // NOTE: This should not be the case! However we want to exercise the
+  // explicit memory barriers code paths in the Optimizing Compiler.
+  bool PreferAcquireRelease() const {
+    return false;
+  }
+
   virtual ~Arm64InstructionSetFeatures() {}
 
  protected:
diff --git a/runtime/arch/arm64/instruction_set_features_arm64_test.cc b/runtime/arch/arm64/instruction_set_features_arm64_test.cc
index 027e59c..753107b 100644
--- a/runtime/arch/arm64/instruction_set_features_arm64_test.cc
+++ b/runtime/arch/arm64/instruction_set_features_arm64_test.cc
@@ -30,6 +30,8 @@
   EXPECT_TRUE(arm64_features->Equals(arm64_features.get()));
   EXPECT_STREQ("smp,a53", arm64_features->GetFeatureString().c_str());
   EXPECT_EQ(arm64_features->AsBitmap(), 3U);
+  // See the comments in instruction_set_features_arm64.h.
+  EXPECT_FALSE(arm64_features->AsArm64InstructionSetFeatures()->PreferAcquireRelease());
 }
 
 }  // namespace art
diff --git a/runtime/arch/arm64/quick_entrypoints_arm64.S b/runtime/arch/arm64/quick_entrypoints_arm64.S
index 770073b5..ec25a33 100644
--- a/runtime/arch/arm64/quick_entrypoints_arm64.S
+++ b/runtime/arch/arm64/quick_entrypoints_arm64.S
@@ -1510,7 +1510,7 @@
     // x1: pointer to the bottom of the used area of the alloca, can restore stack till there.
 
     // Check for error = 0.
-    cbz x0, .Lentry_error
+    cbz x0, .Lexception_in_native
 
     // Release part of the alloca.
     mov sp, x1
@@ -1545,15 +1545,15 @@
 
     bl artQuickGenericJniEndTrampoline
 
+    // Pending exceptions possible.
+    // Use xETR as xSELF might be scratched by native code
+    ldr x2, [xETR, THREAD_EXCEPTION_OFFSET]
+    cbnz x2, .Lexception_in_native
+
     // Tear down the alloca.
     mov sp, x28
     .cfi_def_cfa_register sp
 
-    // Pending exceptions possible.
-    // Use xETR as xSELF might be scratched by native code
-    ldr x1, [xETR, THREAD_EXCEPTION_OFFSET]
-    cbnz x1, .Lexception_in_native
-
     // Tear down the callee-save frame.
     RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME
 
@@ -1561,13 +1561,15 @@
     fmov d0, x0
     ret
 
-.Lentry_error:
-    mov sp, x28
-    .cfi_def_cfa_register sp
 .Lexception_in_native:
-    RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME
+    // Restore xSELF. It might have been scratched by native code.
+    mov xSELF, xETR
+    // Move to x1 then sp to please assembler.
+    ldr x1, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET]
+    mov sp, x1
+    .cfi_def_cfa_register sp
+    # This will create a new save-all frame, required by the runtime.
     DELIVER_PENDING_EXCEPTION
-
 END art_quick_generic_jni_trampoline
 
 /*
diff --git a/runtime/arch/mips/entrypoints_direct_mips.h b/runtime/arch/mips/entrypoints_direct_mips.h
new file mode 100644
index 0000000..b1aa3ee
--- /dev/null
+++ b/runtime/arch/mips/entrypoints_direct_mips.h
@@ -0,0 +1,52 @@
+/*
+ * 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.
+ */
+
+#ifndef ART_RUNTIME_ARCH_MIPS_ENTRYPOINTS_DIRECT_MIPS_H_
+#define ART_RUNTIME_ARCH_MIPS_ENTRYPOINTS_DIRECT_MIPS_H_
+
+#include "entrypoints/quick/quick_entrypoints_enum.h"
+
+namespace art {
+
+/* Returns true if entrypoint contains direct reference to
+   native implementation. The list is required as direct
+   entrypoints need additional handling during invocation.*/
+static constexpr bool IsDirectEntrypoint(QuickEntrypointEnum entrypoint) {
+  return
+      entrypoint == kQuickInstanceofNonTrivial ||
+      entrypoint == kQuickA64Load ||
+      entrypoint == kQuickA64Store ||
+      entrypoint == kQuickFmod ||
+      entrypoint == kQuickFmodf ||
+      entrypoint == kQuickMemcpy ||
+      entrypoint == kQuickL2d ||
+      entrypoint == kQuickL2f ||
+      entrypoint == kQuickD2iz ||
+      entrypoint == kQuickF2iz ||
+      entrypoint == kQuickD2l ||
+      entrypoint == kQuickF2l ||
+      entrypoint == kQuickLdiv ||
+      entrypoint == kQuickLmod ||
+      entrypoint == kQuickLmul ||
+      entrypoint == kQuickCmpgDouble ||
+      entrypoint == kQuickCmpgFloat ||
+      entrypoint == kQuickCmplDouble ||
+      entrypoint == kQuickCmplFloat;
+}
+
+}  // namespace art
+
+#endif  // ART_RUNTIME_ARCH_MIPS_ENTRYPOINTS_DIRECT_MIPS_H_
diff --git a/runtime/arch/mips/entrypoints_init_mips.cc b/runtime/arch/mips/entrypoints_init_mips.cc
index 1a661c4..e3ec27c 100644
--- a/runtime/arch/mips/entrypoints_init_mips.cc
+++ b/runtime/arch/mips/entrypoints_init_mips.cc
@@ -23,6 +23,7 @@
 #include "entrypoints/entrypoint_utils.h"
 #include "entrypoints/math_entrypoints.h"
 #include "entrypoints/runtime_asm_entrypoints.h"
+#include "entrypoints_direct_mips.h"
 #include "interpreter/interpreter.h"
 
 namespace art {
@@ -72,83 +73,155 @@
 
   // Cast
   qpoints->pInstanceofNonTrivial = artIsAssignableFromCode;
+  static_assert(IsDirectEntrypoint(kQuickInstanceofNonTrivial), "Direct C stub not marked direct.");
   qpoints->pCheckCast = art_quick_check_cast;
+  static_assert(!IsDirectEntrypoint(kQuickCheckCast), "Non-direct C stub marked direct.");
 
   // DexCache
   qpoints->pInitializeStaticStorage = art_quick_initialize_static_storage;
+  static_assert(!IsDirectEntrypoint(kQuickInitializeStaticStorage),
+                "Non-direct C stub marked direct.");
   qpoints->pInitializeTypeAndVerifyAccess = art_quick_initialize_type_and_verify_access;
+  static_assert(!IsDirectEntrypoint(kQuickInitializeTypeAndVerifyAccess),
+                "Non-direct C stub marked direct.");
   qpoints->pInitializeType = art_quick_initialize_type;
+  static_assert(!IsDirectEntrypoint(kQuickInitializeType), "Non-direct C stub marked direct.");
   qpoints->pResolveString = art_quick_resolve_string;
+  static_assert(!IsDirectEntrypoint(kQuickResolveString), "Non-direct C stub marked direct.");
 
   // Field
   qpoints->pSet8Instance = art_quick_set8_instance;
+  static_assert(!IsDirectEntrypoint(kQuickSet8Instance), "Non-direct C stub marked direct.");
   qpoints->pSet8Static = art_quick_set8_static;
+  static_assert(!IsDirectEntrypoint(kQuickSet8Static), "Non-direct C stub marked direct.");
   qpoints->pSet16Instance = art_quick_set16_instance;
+  static_assert(!IsDirectEntrypoint(kQuickSet16Instance), "Non-direct C stub marked direct.");
   qpoints->pSet16Static = art_quick_set16_static;
+  static_assert(!IsDirectEntrypoint(kQuickSet16Static), "Non-direct C stub marked direct.");
   qpoints->pSet32Instance = art_quick_set32_instance;
+  static_assert(!IsDirectEntrypoint(kQuickSet32Instance), "Non-direct C stub marked direct.");
   qpoints->pSet32Static = art_quick_set32_static;
+  static_assert(!IsDirectEntrypoint(kQuickSet32Static), "Non-direct C stub marked direct.");
   qpoints->pSet64Instance = art_quick_set64_instance;
+  static_assert(!IsDirectEntrypoint(kQuickSet64Instance), "Non-direct C stub marked direct.");
   qpoints->pSet64Static = art_quick_set64_static;
+  static_assert(!IsDirectEntrypoint(kQuickSet64Static), "Non-direct C stub marked direct.");
   qpoints->pSetObjInstance = art_quick_set_obj_instance;
+  static_assert(!IsDirectEntrypoint(kQuickSetObjInstance), "Non-direct C stub marked direct.");
   qpoints->pSetObjStatic = art_quick_set_obj_static;
+  static_assert(!IsDirectEntrypoint(kQuickSetObjStatic), "Non-direct C stub marked direct.");
   qpoints->pGetBooleanInstance = art_quick_get_boolean_instance;
+  static_assert(!IsDirectEntrypoint(kQuickGetBooleanInstance), "Non-direct C stub marked direct.");
   qpoints->pGetByteInstance = art_quick_get_byte_instance;
+  static_assert(!IsDirectEntrypoint(kQuickGetByteInstance), "Non-direct C stub marked direct.");
   qpoints->pGetCharInstance = art_quick_get_char_instance;
+  static_assert(!IsDirectEntrypoint(kQuickGetCharInstance), "Non-direct C stub marked direct.");
   qpoints->pGetShortInstance = art_quick_get_short_instance;
+  static_assert(!IsDirectEntrypoint(kQuickGetShortInstance), "Non-direct C stub marked direct.");
   qpoints->pGet32Instance = art_quick_get32_instance;
+  static_assert(!IsDirectEntrypoint(kQuickGet32Instance), "Non-direct C stub marked direct.");
   qpoints->pGet64Instance = art_quick_get64_instance;
+  static_assert(!IsDirectEntrypoint(kQuickGet64Instance), "Non-direct C stub marked direct.");
   qpoints->pGetObjInstance = art_quick_get_obj_instance;
+  static_assert(!IsDirectEntrypoint(kQuickGetObjInstance), "Non-direct C stub marked direct.");
   qpoints->pGetBooleanStatic = art_quick_get_boolean_static;
+  static_assert(!IsDirectEntrypoint(kQuickGetBooleanStatic), "Non-direct C stub marked direct.");
   qpoints->pGetByteStatic = art_quick_get_byte_static;
+  static_assert(!IsDirectEntrypoint(kQuickGetByteStatic), "Non-direct C stub marked direct.");
   qpoints->pGetCharStatic = art_quick_get_char_static;
+  static_assert(!IsDirectEntrypoint(kQuickGetCharStatic), "Non-direct C stub marked direct.");
   qpoints->pGetShortStatic = art_quick_get_short_static;
+  static_assert(!IsDirectEntrypoint(kQuickGetShortStatic), "Non-direct C stub marked direct.");
   qpoints->pGet32Static = art_quick_get32_static;
+  static_assert(!IsDirectEntrypoint(kQuickGet32Static), "Non-direct C stub marked direct.");
   qpoints->pGet64Static = art_quick_get64_static;
+  static_assert(!IsDirectEntrypoint(kQuickGet64Static), "Non-direct C stub marked direct.");
   qpoints->pGetObjStatic = art_quick_get_obj_static;
+  static_assert(!IsDirectEntrypoint(kQuickGetObjStatic), "Non-direct C stub marked direct.");
 
   // Array
   qpoints->pAputObjectWithNullAndBoundCheck = art_quick_aput_obj_with_null_and_bound_check;
+  static_assert(!IsDirectEntrypoint(kQuickAputObjectWithNullAndBoundCheck),
+                "Non-direct C stub marked direct.");
   qpoints->pAputObjectWithBoundCheck = art_quick_aput_obj_with_bound_check;
+  static_assert(!IsDirectEntrypoint(kQuickAputObjectWithBoundCheck),
+                "Non-direct C stub marked direct.");
   qpoints->pAputObject = art_quick_aput_obj;
+  static_assert(!IsDirectEntrypoint(kQuickAputObject), "Non-direct C stub marked direct.");
   qpoints->pHandleFillArrayData = art_quick_handle_fill_data;
+  static_assert(!IsDirectEntrypoint(kQuickHandleFillArrayData), "Non-direct C stub marked direct.");
 
   // JNI
   qpoints->pJniMethodStart = JniMethodStart;
+  static_assert(!IsDirectEntrypoint(kQuickJniMethodStart), "Non-direct C stub marked direct.");
   qpoints->pJniMethodStartSynchronized = JniMethodStartSynchronized;
+  static_assert(!IsDirectEntrypoint(kQuickJniMethodStartSynchronized),
+                "Non-direct C stub marked direct.");
   qpoints->pJniMethodEnd = JniMethodEnd;
+  static_assert(!IsDirectEntrypoint(kQuickJniMethodEnd), "Non-direct C stub marked direct.");
   qpoints->pJniMethodEndSynchronized = JniMethodEndSynchronized;
+  static_assert(!IsDirectEntrypoint(kQuickJniMethodEndSynchronized),
+                "Non-direct C stub marked direct.");
   qpoints->pJniMethodEndWithReference = JniMethodEndWithReference;
+  static_assert(!IsDirectEntrypoint(kQuickJniMethodEndWithReference),
+                "Non-direct C stub marked direct.");
   qpoints->pJniMethodEndWithReferenceSynchronized = JniMethodEndWithReferenceSynchronized;
+  static_assert(!IsDirectEntrypoint(kQuickJniMethodEndWithReferenceSynchronized),
+                "Non-direct C stub marked direct.");
   qpoints->pQuickGenericJniTrampoline = art_quick_generic_jni_trampoline;
+  static_assert(!IsDirectEntrypoint(kQuickQuickGenericJniTrampoline),
+                "Non-direct C stub marked direct.");
 
   // Locks
   qpoints->pLockObject = art_quick_lock_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
   qpoints->pCmpgDouble = CmpgDouble;
+  static_assert(IsDirectEntrypoint(kQuickCmpgDouble), "Direct C stub not marked direct.");
   qpoints->pCmpgFloat = CmpgFloat;
+  static_assert(IsDirectEntrypoint(kQuickCmpgFloat), "Direct C stub not marked direct.");
   qpoints->pCmplDouble = CmplDouble;
+  static_assert(IsDirectEntrypoint(kQuickCmplDouble), "Direct C stub not marked direct.");
   qpoints->pCmplFloat = CmplFloat;
+  static_assert(IsDirectEntrypoint(kQuickCmplFloat), "Direct C stub not marked direct.");
   qpoints->pFmod = fmod;
+  static_assert(IsDirectEntrypoint(kQuickFmod), "Direct C stub not marked direct.");
   qpoints->pL2d = art_l2d;
+  static_assert(IsDirectEntrypoint(kQuickL2d), "Direct C stub not marked direct.");
   qpoints->pFmodf = fmodf;
+  static_assert(IsDirectEntrypoint(kQuickFmodf), "Direct C stub not marked direct.");
   qpoints->pL2f = art_l2f;
+  static_assert(IsDirectEntrypoint(kQuickL2f), "Direct C stub not marked direct.");
   qpoints->pD2iz = art_d2i;
+  static_assert(IsDirectEntrypoint(kQuickD2iz), "Direct C stub not marked direct.");
   qpoints->pF2iz = art_f2i;
+  static_assert(IsDirectEntrypoint(kQuickF2iz), "Direct C stub not marked direct.");
   qpoints->pIdivmod = NULL;
   qpoints->pD2l = art_d2l;
+  static_assert(IsDirectEntrypoint(kQuickD2l), "Direct C stub not marked direct.");
   qpoints->pF2l = art_f2l;
+  static_assert(IsDirectEntrypoint(kQuickF2l), "Direct C stub not marked direct.");
   qpoints->pLdiv = artLdiv;
+  static_assert(IsDirectEntrypoint(kQuickLdiv), "Direct C stub not marked direct.");
   qpoints->pLmod = artLmod;
+  static_assert(IsDirectEntrypoint(kQuickLmod), "Direct C stub not marked direct.");
   qpoints->pLmul = artLmul;
+  static_assert(IsDirectEntrypoint(kQuickLmul), "Direct C stub not marked direct.");
   qpoints->pShlLong = art_quick_shl_long;
+  static_assert(!IsDirectEntrypoint(kQuickShlLong), "Non-direct C stub marked direct.");
   qpoints->pShrLong = art_quick_shr_long;
+  static_assert(!IsDirectEntrypoint(kQuickShrLong), "Non-direct C stub marked direct.");
   qpoints->pUshrLong = art_quick_ushr_long;
+  static_assert(!IsDirectEntrypoint(kQuickUshrLong), "Non-direct C stub marked direct.");
 
   // Intrinsics
   qpoints->pIndexOf = art_quick_indexof;
+  static_assert(!IsDirectEntrypoint(kQuickIndexOf), "Non-direct C stub marked direct.");
   qpoints->pStringCompareTo = art_quick_string_compareto;
+  static_assert(!IsDirectEntrypoint(kQuickStringCompareTo), "Non-direct C stub marked direct.");
   qpoints->pMemcpy = memcpy;
 
   // Invocation
@@ -156,25 +229,44 @@
   qpoints->pQuickResolutionTrampoline = art_quick_resolution_trampoline;
   qpoints->pQuickToInterpreterBridge = art_quick_to_interpreter_bridge;
   qpoints->pInvokeDirectTrampolineWithAccessCheck = art_quick_invoke_direct_trampoline_with_access_check;
+  static_assert(!IsDirectEntrypoint(kQuickInvokeDirectTrampolineWithAccessCheck),
+                "Non-direct C stub marked direct.");
   qpoints->pInvokeInterfaceTrampolineWithAccessCheck = art_quick_invoke_interface_trampoline_with_access_check;
+  static_assert(!IsDirectEntrypoint(kQuickInvokeInterfaceTrampolineWithAccessCheck),
+                "Non-direct C stub marked direct.");
   qpoints->pInvokeStaticTrampolineWithAccessCheck = art_quick_invoke_static_trampoline_with_access_check;
+  static_assert(!IsDirectEntrypoint(kQuickInvokeStaticTrampolineWithAccessCheck),
+                "Non-direct C stub marked direct.");
   qpoints->pInvokeSuperTrampolineWithAccessCheck = art_quick_invoke_super_trampoline_with_access_check;
+  static_assert(!IsDirectEntrypoint(kQuickInvokeSuperTrampolineWithAccessCheck),
+                "Non-direct C stub marked direct.");
   qpoints->pInvokeVirtualTrampolineWithAccessCheck = art_quick_invoke_virtual_trampoline_with_access_check;
+  static_assert(!IsDirectEntrypoint(kQuickInvokeVirtualTrampolineWithAccessCheck),
+                "Non-direct C stub marked direct.");
 
   // Thread
   qpoints->pTestSuspend = art_quick_test_suspend;
+  static_assert(!IsDirectEntrypoint(kQuickTestSuspend), "Non-direct C stub marked direct.");
 
   // Throws
   qpoints->pDeliverException = art_quick_deliver_exception;
+  static_assert(!IsDirectEntrypoint(kQuickDeliverException), "Non-direct C stub marked direct.");
   qpoints->pThrowArrayBounds = art_quick_throw_array_bounds;
+  static_assert(!IsDirectEntrypoint(kQuickThrowArrayBounds), "Non-direct C stub marked direct.");
   qpoints->pThrowDivZero = art_quick_throw_div_zero;
+  static_assert(!IsDirectEntrypoint(kQuickThrowDivZero), "Non-direct C stub marked direct.");
   qpoints->pThrowNoSuchMethod = art_quick_throw_no_such_method;
+  static_assert(!IsDirectEntrypoint(kQuickThrowNoSuchMethod), "Non-direct C stub marked direct.");
   qpoints->pThrowNullPointer = art_quick_throw_null_pointer_exception;
+  static_assert(!IsDirectEntrypoint(kQuickThrowNullPointer), "Non-direct C stub marked direct.");
   qpoints->pThrowStackOverflow = art_quick_throw_stack_overflow;
+  static_assert(!IsDirectEntrypoint(kQuickThrowStackOverflow), "Non-direct C stub marked direct.");
 
   // Atomic 64-bit load/store
   qpoints->pA64Load = QuasiAtomic::Read64;
+  static_assert(IsDirectEntrypoint(kQuickA64Load), "Non-direct C stub marked direct.");
   qpoints->pA64Store = QuasiAtomic::Write64;
+  static_assert(IsDirectEntrypoint(kQuickA64Store), "Non-direct C stub marked direct.");
 };
 
 }  // namespace art
diff --git a/runtime/arch/mips/quick_entrypoints_mips.S b/runtime/arch/mips/quick_entrypoints_mips.S
index df2feb7..16f0e70 100644
--- a/runtime/arch/mips/quick_entrypoints_mips.S
+++ b/runtime/arch/mips/quick_entrypoints_mips.S
@@ -1142,10 +1142,10 @@
     addiu   $sp, $sp, -24          # reserve arg slots
     jal     artQuickGenericJniEndTrampoline
     s.d     $f0, 16($sp)           # pass result_f
-    addiu   $sp, $sp, 24           # remove arg slots
 
     lw      $t0, THREAD_EXCEPTION_OFFSET(rSELF) # load Thread::Current()->exception_
-    bne     $t0, $zero, 2f         # check for pending exceptions
+    bne     $t0, $zero, 1f         # check for pending exceptions
+
     move    $sp, $s8               # tear down the alloca
 
     # tear dpown the callee-save frame
@@ -1156,9 +1156,8 @@
     nop
 
 1:
-    move    $sp, $s8               # tear down the alloca
-2:
-    RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME
+    lw $sp, THREAD_TOP_QUICK_FRAME_OFFSET(rSELF)
+    # This will create a new save-all frame, required by the runtime.
     DELIVER_PENDING_EXCEPTION
 END art_quick_generic_jni_trampoline
 
diff --git a/runtime/arch/mips64/quick_entrypoints_mips64.S b/runtime/arch/mips64/quick_entrypoints_mips64.S
index 60e692b..6f1b826 100644
--- a/runtime/arch/mips64/quick_entrypoints_mips64.S
+++ b/runtime/arch/mips64/quick_entrypoints_mips64.S
@@ -849,7 +849,7 @@
     dmfc1   $a2, $f0
 
     ld      $t0, THREAD_EXCEPTION_OFFSET(rSELF) # load Thread::Current()->exception_
-    bne     $t0, $zero, 2f         # check for pending exceptions
+    bne     $t0, $zero, 1f         # check for pending exceptions
     move    $sp, $s8               # tear down the alloca
 
     # tear dpown the callee-save frame
@@ -859,9 +859,8 @@
     dmtc1   $v0, $f0               # place return value to FP return value
 
 1:
-    move    $sp, $s8               # tear down the alloca
-2:
-    RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME
+    ld      $sp, THREAD_TOP_QUICK_FRAME_OFFSET(rSELF)
+    # This will create a new save-all frame, required by the runtime.
     DELIVER_PENDING_EXCEPTION
 END art_quick_generic_jni_trampoline
 
diff --git a/runtime/arch/stub_test.cc b/runtime/arch/stub_test.cc
index 6acc2a7..0d41a8f 100644
--- a/runtime/arch/stub_test.cc
+++ b/runtime/arch/stub_test.cc
@@ -278,6 +278,7 @@
           "memory");  // clobber all
     // TODO: Should we clobber the other registers?
 #else
+    UNUSED(arg0, arg1, arg2, code, referrer);
     LOG(WARNING) << "Was asked to invoke for an architecture I do not understand.";
     result = 0;
 #endif
@@ -503,6 +504,7 @@
           "memory");  // clobber all
     // TODO: Should we clobber the other registers?
 #else
+    UNUSED(arg0, arg1, arg2, code, referrer, hidden);
     LOG(WARNING) << "Was asked to invoke for an architecture I do not understand.";
     result = 0;
 #endif
@@ -792,6 +794,7 @@
 
   // Test done.
 #else
+  UNUSED(test);
   LOG(INFO) << "Skipping unlock_object as I don't know how to do that on " << kRuntimeISA;
   // Force-print to std::cout so it's also outside the logcat.
   std::cout << "Skipping unlock_object as I don't know how to do that on " << kRuntimeISA << std::endl;
@@ -1326,6 +1329,7 @@
     EXPECT_EQ(values[i], static_cast<uint8_t>(res)) << "Iteration " << i;
   }
 #else
+  UNUSED(f, self, referrer, test);
   LOG(INFO) << "Skipping set_boolean_static as I don't know how to do that on " << kRuntimeISA;
   // Force-print to std::cout so it's also outside the logcat.
   std::cout << "Skipping set_boolean_static as I don't know how to do that on " << kRuntimeISA << std::endl;
@@ -1353,6 +1357,7 @@
     EXPECT_EQ(values[i], static_cast<int8_t>(res)) << "Iteration " << i;
   }
 #else
+  UNUSED(f, self, referrer, test);
   LOG(INFO) << "Skipping set_byte_static as I don't know how to do that on " << kRuntimeISA;
   // Force-print to std::cout so it's also outside the logcat.
   std::cout << "Skipping set_byte_static as I don't know how to do that on " << kRuntimeISA << std::endl;
@@ -1388,6 +1393,7 @@
     EXPECT_EQ(res, static_cast<uint8_t>(res2));
   }
 #else
+  UNUSED(obj, f, self, referrer, test);
   LOG(INFO) << "Skipping set_boolean_instance as I don't know how to do that on " << kRuntimeISA;
   // Force-print to std::cout so it's also outside the logcat.
   std::cout << "Skipping set_boolean_instance as I don't know how to do that on " << kRuntimeISA << std::endl;
@@ -1420,6 +1426,7 @@
     EXPECT_EQ(res, static_cast<int8_t>(res2));
   }
 #else
+  UNUSED(obj, f, self, referrer, test);
   LOG(INFO) << "Skipping set_byte_instance as I don't know how to do that on " << kRuntimeISA;
   // Force-print to std::cout so it's also outside the logcat.
   std::cout << "Skipping set_byte_instance as I don't know how to do that on " << kRuntimeISA << std::endl;
@@ -1449,6 +1456,7 @@
     EXPECT_EQ(values[i], static_cast<uint16_t>(res)) << "Iteration " << i;
   }
 #else
+  UNUSED(f, self, referrer, test);
   LOG(INFO) << "Skipping set_char_static as I don't know how to do that on " << kRuntimeISA;
   // Force-print to std::cout so it's also outside the logcat.
   std::cout << "Skipping set_char_static as I don't know how to do that on " << kRuntimeISA << std::endl;
@@ -1477,6 +1485,7 @@
     EXPECT_EQ(static_cast<int16_t>(res), values[i]) << "Iteration " << i;
   }
 #else
+  UNUSED(f, self, referrer, test);
   LOG(INFO) << "Skipping set_short_static as I don't know how to do that on " << kRuntimeISA;
   // Force-print to std::cout so it's also outside the logcat.
   std::cout << "Skipping set_short_static as I don't know how to do that on " << kRuntimeISA << std::endl;
@@ -1510,6 +1519,7 @@
     EXPECT_EQ(res, static_cast<uint16_t>(res2));
   }
 #else
+  UNUSED(obj, f, self, referrer, test);
   LOG(INFO) << "Skipping set_char_instance as I don't know how to do that on " << kRuntimeISA;
   // Force-print to std::cout so it's also outside the logcat.
   std::cout << "Skipping set_char_instance as I don't know how to do that on " << kRuntimeISA << std::endl;
@@ -1542,6 +1552,7 @@
     EXPECT_EQ(res, static_cast<int16_t>(res2));
   }
 #else
+  UNUSED(obj, f, self, referrer, test);
   LOG(INFO) << "Skipping set_short_instance as I don't know how to do that on " << kRuntimeISA;
   // Force-print to std::cout so it's also outside the logcat.
   std::cout << "Skipping set_short_instance as I don't know how to do that on " << kRuntimeISA << std::endl;
@@ -1571,6 +1582,7 @@
     EXPECT_EQ(res, values[i]) << "Iteration " << i;
   }
 #else
+  UNUSED(f, self, referrer, test);
   LOG(INFO) << "Skipping set32static as I don't know how to do that on " << kRuntimeISA;
   // Force-print to std::cout so it's also outside the logcat.
   std::cout << "Skipping set32static as I don't know how to do that on " << kRuntimeISA << std::endl;
@@ -1607,6 +1619,7 @@
     EXPECT_EQ(res, static_cast<int32_t>(res2));
   }
 #else
+  UNUSED(obj, f, self, referrer, test);
   LOG(INFO) << "Skipping set32instance as I don't know how to do that on " << kRuntimeISA;
   // Force-print to std::cout so it's also outside the logcat.
   std::cout << "Skipping set32instance as I don't know how to do that on " << kRuntimeISA << std::endl;
@@ -1648,6 +1661,7 @@
 
   set_and_check_static((*f)->GetDexFieldIndex(), nullptr, self, referrer, test);
 #else
+  UNUSED(f, self, referrer, test);
   LOG(INFO) << "Skipping setObjstatic as I don't know how to do that on " << kRuntimeISA;
   // Force-print to std::cout so it's also outside the logcat.
   std::cout << "Skipping setObjstatic as I don't know how to do that on " << kRuntimeISA << std::endl;
@@ -1692,6 +1706,7 @@
 
   set_and_check_instance(f, obj->Get(), nullptr, self, referrer, test);
 #else
+  UNUSED(obj, f, self, referrer, test);
   LOG(INFO) << "Skipping setObjinstance as I don't know how to do that on " << kRuntimeISA;
   // Force-print to std::cout so it's also outside the logcat.
   std::cout << "Skipping setObjinstance as I don't know how to do that on " << kRuntimeISA << std::endl;
diff --git a/runtime/arch/x86/quick_entrypoints_x86.S b/runtime/arch/x86/quick_entrypoints_x86.S
index c2acdd1..47bc5ea 100644
--- a/runtime/arch/x86/quick_entrypoints_x86.S
+++ b/runtime/arch/x86/quick_entrypoints_x86.S
@@ -1349,7 +1349,7 @@
 
     // Check for error = 0.
     test %eax, %eax
-    jz .Lentry_error
+    jz .Lexception_in_native
 
     // Release part of the alloca.
     movl %edx, %esp
@@ -1371,15 +1371,16 @@
     pushl %fs:THREAD_SELF_OFFSET  // Pass Thread::Current().
     call SYMBOL(artQuickGenericJniEndTrampoline)
 
-    // Tear down the alloca.
-    movl %ebp, %esp
-    CFI_DEF_CFA_REGISTER(esp)
-
     // Pending exceptions possible.
     mov %fs:THREAD_EXCEPTION_OFFSET, %ebx
     testl %ebx, %ebx
     jnz .Lexception_in_native
 
+    // Tear down the alloca.
+    movl %ebp, %esp
+    CFI_DEF_CFA_REGISTER(esp)
+
+
     // Tear down the callee-save frame.
     // Remove space for FPR args and EAX
     addl LITERAL(4 + 4 * 8), %esp
@@ -1397,11 +1398,11 @@
     movd %edx, %xmm1
     punpckldq %xmm1, %xmm0
     ret
-.Lentry_error:
-    movl %ebp, %esp
-    CFI_DEF_CFA_REGISTER(esp)
 .Lexception_in_native:
-    RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME
+    movl %fs:THREAD_TOP_QUICK_FRAME_OFFSET, %esp
+    // Do a call to push a new save-all frame required by the runtime.
+    call .Lexception_call
+.Lexception_call:
     DELIVER_PENDING_EXCEPTION
 END_FUNCTION art_quick_generic_jni_trampoline
 
diff --git a/runtime/arch/x86_64/quick_entrypoints_x86_64.S b/runtime/arch/x86_64/quick_entrypoints_x86_64.S
index c865541..406126b 100644
--- a/runtime/arch/x86_64/quick_entrypoints_x86_64.S
+++ b/runtime/arch/x86_64/quick_entrypoints_x86_64.S
@@ -1361,7 +1361,7 @@
 
     // Check for error = 0.
     test %rax, %rax
-    jz .Lentry_error
+    jz .Lexception_in_native
 
     // Release part of the alloca.
     movq %rdx, %rsp
@@ -1398,16 +1398,16 @@
     movq %xmm0, %rdx
     call SYMBOL(artQuickGenericJniEndTrampoline)
 
-    // Tear down the alloca.
-    movq %rbp, %rsp
-    CFI_DEF_CFA_REGISTER(rsp)
-
     // Pending exceptions possible.
     // TODO: use cmpq, needs direct encoding because of gas bug
     movq %gs:THREAD_EXCEPTION_OFFSET, %rcx
     test %rcx, %rcx
     jnz .Lexception_in_native
 
+    // Tear down the alloca.
+    movq %rbp, %rsp
+    CFI_DEF_CFA_REGISTER(rsp)
+
     // Tear down the callee-save frame.
     // Load FPRs.
     // movq %xmm0, 16(%rsp)         // doesn't make sense!!!
@@ -1440,40 +1440,12 @@
     // store into fpr, for when it's a fpr return...
     movq %rax, %xmm0
     ret
-.Lentry_error:
-    movq %rbp, %rsp
-    CFI_DEF_CFA_REGISTER(rsp)
 .Lexception_in_native:
-    // TODO: the handle scope contains the this pointer which is used by the debugger for exception
-    //       delivery.
-    movq %xmm0, 16(%rsp)         // doesn't make sense!!!
-    movq 24(%rsp), %xmm1            // neither does this!!!
-    movq 32(%rsp), %xmm2
-    movq 40(%rsp), %xmm3
-    movq 48(%rsp), %xmm4
-    movq 56(%rsp), %xmm5
-    movq 64(%rsp), %xmm6
-    movq 72(%rsp), %xmm7
-    movq 80(%rsp), %xmm12
-    movq 88(%rsp), %xmm13
-    movq 96(%rsp), %xmm14
-    movq 104(%rsp), %xmm15
-    // was 80 + 32 bytes
-    addq LITERAL(80 + 4*8), %rsp
-    CFI_ADJUST_CFA_OFFSET(-80 - 4*8)
-    // Save callee and GPR args, mixed together to agree with core spills bitmap.
-    POP rcx  // Arg.
-    POP rdx  // Arg.
-    POP rbx  // Callee save.
-    POP rbp  // Callee save.
-    POP rsi  // Arg.
-    POP r8   // Arg.
-    POP r9   // Arg.
-    POP r12  // Callee save.
-    POP r13  // Callee save.
-    POP r14  // Callee save.
-    POP r15  // Callee save.
-
+    movq %gs:THREAD_TOP_QUICK_FRAME_OFFSET, %rsp
+    CFI_DEF_CFA_REGISTER(rsp)
+    // Do a call to push a new save-all frame required by the runtime.
+    call .Lexception_call
+.Lexception_call:
     DELIVER_PENDING_EXCEPTION
 END_FUNCTION art_quick_generic_jni_trampoline
 
diff --git a/runtime/base/arena_allocator.cc b/runtime/base/arena_allocator.cc
index b3f812e..e6380bf 100644
--- a/runtime/base/arena_allocator.cc
+++ b/runtime/base/arena_allocator.cc
@@ -129,7 +129,7 @@
       next_(nullptr) {
   if (kUseMemMap) {
     std::string error_msg;
-    map_ = MemMap::MapAnonymous("dalvik-arena", NULL, size, PROT_READ | PROT_WRITE, false,
+    map_ = MemMap::MapAnonymous("dalvik-arena", nullptr, size, PROT_READ | PROT_WRITE, false, false,
                                 &error_msg);
     CHECK(map_ != nullptr) << error_msg;
     memory_ = map_->Begin();
diff --git a/runtime/base/arena_containers.h b/runtime/base/arena_containers.h
index 162eb16..ceff6e8 100644
--- a/runtime/base/arena_containers.h
+++ b/runtime/base/arena_containers.h
@@ -66,8 +66,8 @@
 class ArenaAllocatorAdapterKindImpl<false> {
  public:
   // Not tracking allocations, ignore the supplied kind and arbitrarily provide kArenaAllocSTL.
-  explicit ArenaAllocatorAdapterKindImpl(ArenaAllocKind kind) { UNUSED(kind); }
-  ArenaAllocatorAdapterKindImpl& operator=(const ArenaAllocatorAdapterKindImpl& other) = default;
+  explicit ArenaAllocatorAdapterKindImpl(ArenaAllocKind kind ATTRIBUTE_UNUSED) {}
+  ArenaAllocatorAdapterKindImpl& operator=(const ArenaAllocatorAdapterKindImpl&) = default;
   ArenaAllocKind Kind() { return kArenaAllocSTL; }
 };
 
@@ -75,7 +75,7 @@
 class ArenaAllocatorAdapterKindImpl {
  public:
   explicit ArenaAllocatorAdapterKindImpl(ArenaAllocKind kind) : kind_(kind) { }
-  ArenaAllocatorAdapterKindImpl& operator=(const ArenaAllocatorAdapterKindImpl& other) = default;
+  ArenaAllocatorAdapterKindImpl& operator=(const ArenaAllocatorAdapterKindImpl&) = default;
   ArenaAllocKind Kind() { return kind_; }
 
  private:
@@ -109,8 +109,8 @@
         ArenaAllocatorAdapterKind(other),
         arena_allocator_(other.arena_allocator_) {
   }
-  ArenaAllocatorAdapter(const ArenaAllocatorAdapter& other) = default;
-  ArenaAllocatorAdapter& operator=(const ArenaAllocatorAdapter& other) = default;
+  ArenaAllocatorAdapter(const ArenaAllocatorAdapter&) = default;
+  ArenaAllocatorAdapter& operator=(const ArenaAllocatorAdapter&) = default;
   ~ArenaAllocatorAdapter() = default;
 
  private:
@@ -147,8 +147,8 @@
         ArenaAllocatorAdapterKind(other),
         arena_allocator_(other.arena_allocator_) {
   }
-  ArenaAllocatorAdapter(const ArenaAllocatorAdapter& other) = default;
-  ArenaAllocatorAdapter& operator=(const ArenaAllocatorAdapter& other) = default;
+  ArenaAllocatorAdapter(const ArenaAllocatorAdapter&) = default;
+  ArenaAllocatorAdapter& operator=(const ArenaAllocatorAdapter&) = default;
   ~ArenaAllocatorAdapter() = default;
 
   size_type max_size() const {
diff --git a/runtime/base/logging.h b/runtime/base/logging.h
index cc1a4a1..3d007ba 100644
--- a/runtime/base/logging.h
+++ b/runtime/base/logging.h
@@ -42,6 +42,7 @@
   bool gc;
   bool heap;
   bool jdwp;
+  bool jit;
   bool jni;
   bool monitor;
   bool profiler;
diff --git a/runtime/base/mutex.h b/runtime/base/mutex.h
index 745b209..45d2347 100644
--- a/runtime/base/mutex.h
+++ b/runtime/base/mutex.h
@@ -67,6 +67,7 @@
   kReferenceQueueWeakReferencesLock,
   kReferenceQueueClearedReferencesLock,
   kReferenceProcessorLock,
+  kJitCodeCacheLock,
   kRosAllocGlobalLock,
   kRosAllocBracketLock,
   kRosAllocBulkFreeLock,
diff --git a/runtime/base/scoped_arena_containers.h b/runtime/base/scoped_arena_containers.h
index 664a909..df79085 100644
--- a/runtime/base/scoped_arena_containers.h
+++ b/runtime/base/scoped_arena_containers.h
@@ -85,8 +85,8 @@
         ArenaAllocatorAdapterKind(other),
         arena_stack_(other.arena_stack_) {
   }
-  ScopedArenaAllocatorAdapter(const ScopedArenaAllocatorAdapter& other) = default;
-  ScopedArenaAllocatorAdapter& operator=(const ScopedArenaAllocatorAdapter& other) = default;
+  ScopedArenaAllocatorAdapter(const ScopedArenaAllocatorAdapter&) = default;
+  ScopedArenaAllocatorAdapter& operator=(const ScopedArenaAllocatorAdapter&) = default;
   ~ScopedArenaAllocatorAdapter() = default;
 
  private:
@@ -128,8 +128,8 @@
         ArenaAllocatorAdapterKind(other),
         arena_stack_(other.arena_stack_) {
   }
-  ScopedArenaAllocatorAdapter(const ScopedArenaAllocatorAdapter& other) = default;
-  ScopedArenaAllocatorAdapter& operator=(const ScopedArenaAllocatorAdapter& other) = default;
+  ScopedArenaAllocatorAdapter(const ScopedArenaAllocatorAdapter&) = default;
+  ScopedArenaAllocatorAdapter& operator=(const ScopedArenaAllocatorAdapter&) = default;
   ~ScopedArenaAllocatorAdapter() = default;
 
   size_type max_size() const {
diff --git a/runtime/base/variant_map.h b/runtime/base/variant_map.h
index c9718fc..8655a9e 100644
--- a/runtime/base/variant_map.h
+++ b/runtime/base/variant_map.h
@@ -120,8 +120,8 @@
 
    protected:
     // Avoid the object slicing problem; use Clone() instead.
-    VariantMapKeyRaw(const VariantMapKeyRaw& other) = default;
-    VariantMapKeyRaw(VariantMapKeyRaw&& other) = default;
+    VariantMapKeyRaw(const VariantMapKeyRaw&) = default;
+    VariantMapKeyRaw(VariantMapKeyRaw&&) = default;
 
    private:
     size_t key_counter_;  // Runtime type ID. Unique each time a new type is reified.
@@ -174,8 +174,8 @@
     deleter(reinterpret_cast<TValue*>(value));
   }
 
-  VariantMapKey(const VariantMapKey& other) = default;
-  VariantMapKey(VariantMapKey&& other) = default;
+  VariantMapKey(const VariantMapKey&) = default;
+  VariantMapKey(VariantMapKey&&) = default;
 
   template <typename Base, template <typename TV> class TKey> friend struct VariantMap;
 
diff --git a/runtime/class_linker.cc b/runtime/class_linker.cc
index 3278751..2989b8c 100644
--- a/runtime/class_linker.cc
+++ b/runtime/class_linker.cc
@@ -43,6 +43,8 @@
 #include "handle_scope.h"
 #include "intern_table.h"
 #include "interpreter/interpreter.h"
+#include "jit/jit.h"
+#include "jit/jit_code_cache.h"
 #include "leb128.h"
 #include "oat.h"
 #include "oat_file.h"
@@ -91,15 +93,14 @@
   // a NoClassDefFoundError (v2 2.17.5).  The exception to this rule is if we
   // failed in verification, in which case v2 5.4.1 says we need to re-throw
   // the previous error.
-  Runtime* runtime = Runtime::Current();
-  bool is_compiler = runtime->IsCompiler();
-  if (!is_compiler) {  // Give info if this occurs at runtime.
+  Runtime* const runtime = Runtime::Current();
+  if (!runtime->IsAotCompiler()) {  // Give info if this occurs at runtime.
     LOG(INFO) << "Rejecting re-init on previously-failed class " << PrettyClass(c);
   }
 
   CHECK(c->IsErroneous()) << PrettyClass(c) << " " << c->GetStatus();
   Thread* self = Thread::Current();
-  if (is_compiler) {
+  if (runtime->IsAotCompiler()) {
     // At compile time, accurate errors and NCDFE are disabled to speed compilation.
     mirror::Throwable* pre_allocated = runtime->GetPreAllocatedNoClassDefFoundError();
     self->SetException(ThrowLocation(), pre_allocated);
@@ -428,7 +429,7 @@
   // Set up GenericJNI entrypoint. That is mainly a hack for common_compiler_test.h so that
   // we do not need friend classes or a publicly exposed setter.
   quick_generic_jni_trampoline_ = GetQuickGenericJniStub();
-  if (!runtime->IsCompiler()) {
+  if (!runtime->IsAotCompiler()) {
     // We need to set up the generic trampolines since we don't have an image.
     quick_resolution_trampoline_ = GetQuickResolutionStub();
     quick_imt_conflict_trampoline_ = GetQuickImtConflictStub();
@@ -1037,8 +1038,7 @@
                                                                const char* oat_location,
                                                                std::string* error_msg) {
   std::unique_ptr<OatFile> oat_file(OatFile::Open(oat_location, oat_location, nullptr, nullptr,
-                                            !Runtime::Current()->IsCompiler(),
-                                            error_msg));
+                                                  !Runtime::Current()->IsAotCompiler(), error_msg));
   if (oat_file.get() == nullptr) {
     *error_msg = StringPrintf("Failed to find existing oat file at %s: %s", oat_location,
                               error_msg->c_str());
@@ -1109,8 +1109,8 @@
     return nullptr;
   }
   std::unique_ptr<OatFile> oat_file(OatFile::Open(oat_location, oat_location, nullptr, nullptr,
-                                            !Runtime::Current()->IsCompiler(),
-                                            &error_msg));
+                                                  !Runtime::Current()->IsAotCompiler(),
+                                                  &error_msg));
   if (oat_file.get() == nullptr) {
     std::string compound_msg = StringPrintf("\nFailed to open generated oat file '%s': %s",
                                             oat_location, error_msg.c_str());
@@ -1350,7 +1350,7 @@
   *already_opened = false;
   const Runtime* runtime = Runtime::Current();
   CHECK(runtime != nullptr);
-  bool executable = !runtime->IsCompiler();
+  bool executable = !runtime->IsAotCompiler();
 
   std::string odex_error_msg;
   bool should_patch_system = false;
@@ -1518,7 +1518,7 @@
   bool success = Exec(argv, error_msg);
   if (success) {
     std::unique_ptr<OatFile> output(OatFile::Open(output_oat, output_oat, nullptr, nullptr,
-                                                  !runtime->IsCompiler(), error_msg));
+                                                  !runtime->IsAotCompiler(), error_msg));
     bool checksum_verified = false;
     if (output.get() != nullptr && CheckOatFile(runtime, output.get(), isa, &checksum_verified,
                                                 error_msg)) {
@@ -1532,7 +1532,7 @@
                                 "but was unable to open output file '%s': %s",
                                 input_oat.c_str(), output_oat.c_str(), error_msg->c_str());
     }
-  } else if (!runtime->IsCompiler()) {
+  } else if (!runtime->IsAotCompiler()) {
     // patchoat failed which means we probably don't have enough room to place the output oat file,
     // instead of failing we should just run the interpreter from the dex files in the input oat.
     LOG(WARNING) << "Patching of oat file '" << input_oat << "' failed. Attempting to use oat file "
@@ -1619,22 +1619,20 @@
   if (oat_file != nullptr) {
     return oat_file;
   }
-
-  return OatFile::Open(oat_location, oat_location, nullptr, nullptr, !Runtime::Current()->IsCompiler(),
-                       error_msg);
+  return OatFile::Open(oat_location, oat_location, nullptr, nullptr,
+                       !Runtime::Current()->IsAotCompiler(), error_msg);
 }
 
 void ClassLinker::InitFromImageInterpretOnlyCallback(mirror::Object* obj, void* arg) {
   ClassLinker* class_linker = reinterpret_cast<ClassLinker*>(arg);
   DCHECK(obj != nullptr);
   DCHECK(class_linker != nullptr);
-  size_t pointer_size = class_linker->image_pointer_size_;
-
   if (obj->IsArtMethod()) {
     mirror::ArtMethod* method = obj->AsArtMethod();
     if (!method->IsNative()) {
+      const size_t pointer_size = class_linker->image_pointer_size_;
       method->SetEntryPointFromInterpreterPtrSize(artInterpreterToInterpreterBridge, pointer_size);
-      if (method != Runtime::Current()->GetResolutionMethod()) {
+      if (!method->IsRuntimeMethod() && method != Runtime::Current()->GetResolutionMethod()) {
         method->SetEntryPointFromQuickCompiledCodePtrSize(GetQuickToInterpreterBridge(),
                                                           pointer_size);
       }
@@ -1703,8 +1701,8 @@
   // bitmap walk.
   mirror::ArtMethod::SetClass(GetClassRoot(kJavaLangReflectArtMethod));
   size_t art_method_object_size = mirror::ArtMethod::GetJavaLangReflectArtMethod()->GetObjectSize();
-  if (!Runtime::Current()->IsCompiler()) {
-    // Compiler supports having an image with a different pointer size than the runtime. This
+  if (!Runtime::Current()->IsAotCompiler()) {
+    // Aot compiler supports having an image with a different pointer size than the runtime. This
     // happens on the host for compile 32 bit tests since we use a 64 bit libart compiler. We may
     // also use 32 bit dex2oat on a system with 64 bit apps.
     CHECK_EQ(art_method_object_size, mirror::ArtMethod::InstanceSize(sizeof(void*)))
@@ -1719,7 +1717,7 @@
 
   // Set entry point to interpreter if in InterpretOnly mode.
   Runtime* runtime = Runtime::Current();
-  if (!runtime->IsCompiler() && runtime->GetInstrumentation()->InterpretOnly()) {
+  if (!runtime->IsAotCompiler() && runtime->GetInstrumentation()->InterpretOnly()) {
     heap->VisitObjects(InitFromImageInterpretOnlyCallback, this);
   }
 
@@ -2135,15 +2133,16 @@
           }
           mirror::Object* dex_file = dex_file_field->GetObject(element);
           if (dex_file != nullptr) {
-            const uint64_t cookie = cookie_field->GetLong(dex_file);
-            auto* dex_files =
-                reinterpret_cast<std::vector<const DexFile*>*>(static_cast<uintptr_t>(cookie));
-            if (dex_files == 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 for " << descriptor;
               break;
             }
-            for (const DexFile* cp_dex_file : *dex_files) {
+            int32_t long_array_size = long_array->GetLength();
+            for (int32_t j = 0; j < long_array_size; ++j) {
+              const DexFile* cp_dex_file = reinterpret_cast<const DexFile*>(static_cast<uintptr_t>(
+                  long_array->GetWithoutChecks(j)));
               const DexFile::ClassDef* dex_class_def = cp_dex_file->FindClassDef(descriptor, hash);
               if (dex_class_def != nullptr) {
                 RegisterDexFile(*cp_dex_file);
@@ -2523,21 +2522,24 @@
   }
   bool found;
   OatFile::OatMethod oat_method = FindOatMethodFor(method, &found);
-  const void* result = nullptr;
   if (found) {
-    result = oat_method.GetQuickCode();
-  }
-
-  if (result == nullptr) {
-    if (method->IsNative()) {
-      // No code and native? Use generic trampoline.
-      result = GetQuickGenericJniStub();
-    } else {
-      // No code? You must mean to go into the interpreter.
-      result = GetQuickToInterpreterBridge();
+    auto* code = oat_method.GetQuickCode();
+    if (code != nullptr) {
+      return code;
     }
   }
-  return result;
+  jit::Jit* const jit = Runtime::Current()->GetJit();
+  if (jit != nullptr) {
+    auto* code = jit->GetCodeCache()->GetCodeFor(method);
+    if (code != nullptr) {
+      return code;
+    }
+  }
+  if (method->IsNative()) {
+    // No code and native? Use generic trampoline.
+    return GetQuickGenericJniStub();
+  }
+  return GetQuickToInterpreterBridge();
 }
 
 const void* ClassLinker::GetOatMethodQuickCodeFor(mirror::ArtMethod* method) {
@@ -2546,7 +2548,17 @@
   }
   bool found;
   OatFile::OatMethod oat_method = FindOatMethodFor(method, &found);
-  return found ? oat_method.GetQuickCode() : nullptr;
+  if (found) {
+    return oat_method.GetQuickCode();
+  }
+  jit::Jit* jit = Runtime::Current()->GetJit();
+  if (jit != nullptr) {
+    auto* code = jit->GetCodeCache()->GetCodeFor(method);
+    if (code != nullptr) {
+      return code;
+    }
+  }
+  return nullptr;
 }
 
 const void* ClassLinker::GetQuickOatCodeFor(const DexFile& dex_file, uint16_t class_def_idx,
@@ -2582,7 +2594,7 @@
   }
   Runtime* runtime = Runtime::Current();
   if (!runtime->IsStarted() || runtime->UseCompileTimeClassPath()) {
-    if (runtime->IsCompiler() || runtime->GetHeap()->HasImageSpace()) {
+    if (runtime->IsAotCompiler() || runtime->GetHeap()->HasImageSpace()) {
       return;  // OAT file unavailable.
     }
   }
@@ -2635,7 +2647,7 @@
                            const OatFile::OatClass* oat_class,
                            uint32_t class_def_method_index) {
   Runtime* runtime = Runtime::Current();
-  if (runtime->IsCompiler()) {
+  if (runtime->IsAotCompiler()) {
     // The following code only applies to a non-compiler runtime.
     return;
   }
@@ -3474,7 +3486,7 @@
     EnsurePreverifiedMethods(klass);
     return;
   }
-  if (klass->IsCompileTimeVerified() && Runtime::Current()->IsCompiler()) {
+  if (klass->IsCompileTimeVerified() && Runtime::Current()->IsAotCompiler()) {
     return;
   }
 
@@ -3490,7 +3502,7 @@
   } else {
     CHECK_EQ(klass->GetStatus(), mirror::Class::kStatusRetryVerificationAtRuntime)
         << PrettyClass(klass.Get());
-    CHECK(!Runtime::Current()->IsCompiler());
+    CHECK(!Runtime::Current()->IsAotCompiler());
     klass->SetStatus(mirror::Class::kStatusVerifyingAtRuntime, self);
   }
 
@@ -3526,7 +3538,7 @@
         self->GetException(nullptr)->SetCause(cause.Get());
       }
       ClassReference ref(klass->GetDexCache()->GetDexFile(), klass->GetDexClassDefIndex());
-      if (Runtime::Current()->IsCompiler()) {
+      if (Runtime::Current()->IsAotCompiler()) {
         Runtime::Current()->GetCompilerCallbacks()->ClassRejected(ref);
       }
       klass->SetStatus(mirror::Class::kStatusError, self);
@@ -3551,7 +3563,7 @@
   std::string error_msg;
   if (!preverified) {
     verifier_failure = verifier::MethodVerifier::VerifyClass(self, klass.Get(),
-                                                             Runtime::Current()->IsCompiler(),
+                                                             Runtime::Current()->IsAotCompiler(),
                                                              &error_msg);
   }
   if (preverified || verifier_failure != verifier::MethodVerifier::kHardFailure) {
@@ -3579,7 +3591,7 @@
       // Soft failures at compile time should be retried at runtime. Soft
       // failures at runtime will be handled by slow paths in the generated
       // code. Set status accordingly.
-      if (Runtime::Current()->IsCompiler()) {
+      if (Runtime::Current()->IsAotCompiler()) {
         klass->SetStatus(mirror::Class::kStatusRetryVerificationAtRuntime, self);
       } else {
         klass->SetStatus(mirror::Class::kStatusVerified, self);
@@ -3620,7 +3632,7 @@
   // we are not compiling the image or if the class we're verifying is not part of
   // the app.  In other words, we will only check for preverification of bootclasspath
   // classes.
-  if (Runtime::Current()->IsCompiler()) {
+  if (Runtime::Current()->IsAotCompiler()) {
     // Are we compiling the bootclasspath?
     if (!Runtime::Current()->UseCompileTimeClassPath()) {
       return false;
@@ -3646,7 +3658,7 @@
   // image (that we just failed loading), and the verifier can't be run on quickened opcodes when
   // the runtime isn't started. On the other hand, app classes can be re-verified even if they are
   // already pre-opted, as then the runtime is started.
-  if (!Runtime::Current()->IsCompiler() &&
+  if (!Runtime::Current()->IsAotCompiler() &&
       !Runtime::Current()->GetHeap()->HasImageSpace() &&
       klass->GetClassLoader() != nullptr) {
     return false;
@@ -4094,7 +4106,7 @@
           CHECK(self->IsExceptionPending());
           VlogClassInitializationFailure(klass);
         } else {
-          CHECK(Runtime::Current()->IsCompiler());
+          CHECK(Runtime::Current()->IsAotCompiler());
           CHECK_EQ(klass->GetStatus(), mirror::Class::kStatusRetryVerificationAtRuntime);
         }
         return false;
@@ -4229,8 +4241,8 @@
     } else if (Runtime::Current()->IsTransactionAborted()) {
       // The exception thrown when the transaction aborted has been caught and cleared
       // so we need to throw it again now.
-      LOG(WARNING) << "Return from class initializer of " << PrettyDescriptor(klass.Get())
-                   << " without exception while transaction was aborted: re-throw it now.";
+      VLOG(compiler) << "Return from class initializer of " << PrettyDescriptor(klass.Get())
+                     << " without exception while transaction was aborted: re-throw it now.";
       Runtime::Current()->ThrowInternalErrorForAbortedTransaction(self);
       klass->SetStatus(mirror::Class::kStatusError, self);
       success = false;
@@ -4275,7 +4287,8 @@
     if (klass->GetStatus() == mirror::Class::kStatusInitializing) {
       continue;
     }
-    if (klass->GetStatus() == mirror::Class::kStatusVerified && Runtime::Current()->IsCompiler()) {
+    if (klass->GetStatus() == mirror::Class::kStatusVerified &&
+        Runtime::Current()->IsAotCompiler()) {
       // Compile time initialization failed.
       return false;
     }
diff --git a/runtime/debugger.cc b/runtime/debugger.cc
index 13bbdeb..c0dd197 100644
--- a/runtime/debugger.cc
+++ b/runtime/debugger.cc
@@ -287,6 +287,13 @@
     Dbg::PostException(throw_location, catch_method, catch_dex_pc, exception_object);
   }
 
+  // We only care about how many backward branches were executed in the Jit.
+  void BackwardBranch(Thread* /*thread*/, mirror::ArtMethod* method, int32_t dex_pc_offset)
+      OVERRIDE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+    LOG(ERROR) << "Unexpected backward branch event in debugger " << PrettyMethod(method)
+               << " " << dex_pc_offset;
+  }
+
  private:
   DISALLOW_COPY_AND_ASSIGN(DebugInstrumentationListener);
 } gDebugInstrumentationListener;
diff --git a/runtime/dex_file.h b/runtime/dex_file.h
index e121a08..c8ede48 100644
--- a/runtime/dex_file.h
+++ b/runtime/dex_file.h
@@ -1028,6 +1028,13 @@
   // from an oat file, e.g., directly from an apk.
   const OatFile* oat_file_;
 };
+
+struct DexFileReference {
+  DexFileReference(const DexFile* file, uint32_t idx) : dex_file(file), index(idx) { }
+  const DexFile* dex_file;
+  uint32_t index;
+};
+
 std::ostream& operator<<(std::ostream& os, const DexFile& dex_file);
 
 // Iterate over a dex file's ProtoId's paramters
diff --git a/runtime/dex_instruction.cc b/runtime/dex_instruction.cc
index a802759..69fe874 100644
--- a/runtime/dex_instruction.cc
+++ b/runtime/dex_instruction.cc
@@ -134,6 +134,23 @@
   return os.str();
 }
 
+std::string Instruction::DumpHexLE(size_t instr_code_units) const {
+  size_t inst_length = SizeInCodeUnits();
+  if (inst_length > instr_code_units) {
+    inst_length = instr_code_units;
+  }
+  std::ostringstream os;
+  const uint16_t* insn = reinterpret_cast<const uint16_t*>(this);
+  for (size_t i = 0; i < inst_length; i++) {
+    os << StringPrintf("%02x%02x", static_cast<uint8_t>(insn[i] & 0x00FF),
+                       static_cast<uint8_t>((insn[i] & 0xFF00) >> 8)) << " ";
+  }
+  for (size_t i = inst_length; i < instr_code_units; i++) {
+    os << "     ";
+  }
+  return os.str();
+}
+
 std::string Instruction::DumpString(const DexFile* file) const {
   std::ostringstream os;
   const char* opcode = kInstructionNames[Opcode()];
diff --git a/runtime/dex_instruction.h b/runtime/dex_instruction.h
index af5d9d0..d3b9eb4 100644
--- a/runtime/dex_instruction.h
+++ b/runtime/dex_instruction.h
@@ -525,6 +525,10 @@
   // Dump code_units worth of this instruction, padding to code_units for shorter instructions
   std::string DumpHex(size_t code_units) const;
 
+  // Little-endian dump code_units worth of this instruction, padding to code_units for
+  // shorter instructions
+  std::string DumpHexLE(size_t instr_code_units) const;
+
   uint16_t Fetch16(size_t offset) const {
     const uint16_t* insns = reinterpret_cast<const uint16_t*>(this);
     return insns[offset];
diff --git a/compiler/utils/dex_instruction_utils.h b/runtime/dex_instruction_utils.h
similarity index 82%
rename from compiler/utils/dex_instruction_utils.h
rename to runtime/dex_instruction_utils.h
index bb2c592..1a671c5 100644
--- a/compiler/utils/dex_instruction_utils.h
+++ b/runtime/dex_instruction_utils.h
@@ -14,8 +14,8 @@
  * limitations under the License.
  */
 
-#ifndef ART_COMPILER_UTILS_DEX_INSTRUCTION_UTILS_H_
-#define ART_COMPILER_UTILS_DEX_INSTRUCTION_UTILS_H_
+#ifndef ART_RUNTIME_DEX_INSTRUCTION_UTILS_H_
+#define ART_RUNTIME_DEX_INSTRUCTION_UTILS_H_
 
 #include "dex_instruction.h"
 
@@ -58,6 +58,11 @@
       opcode != Instruction::RETURN_VOID_BARRIER;
 }
 
+constexpr bool IsInstructionQuickInvoke(Instruction::Code opcode) {
+  return opcode == Instruction::INVOKE_VIRTUAL_QUICK ||
+      opcode == Instruction::INVOKE_VIRTUAL_RANGE_QUICK;
+}
+
 constexpr bool IsInstructionInvokeStatic(Instruction::Code opcode) {
   return opcode == Instruction::INVOKE_STATIC || opcode == Instruction::INVOKE_STATIC_RANGE;
 }
@@ -102,6 +107,11 @@
   return Instruction::IGET <= code && code <= Instruction::IPUT_SHORT;
 }
 
+constexpr bool IsInstructionIGetQuickOrIPutQuick(Instruction::Code code) {
+  return (code >= Instruction::IGET_QUICK && code <= Instruction::IPUT_OBJECT_QUICK) ||
+      (code >= Instruction::IPUT_BOOLEAN_QUICK && code <= Instruction::IGET_SHORT_QUICK);
+}
+
 constexpr bool IsInstructionSGetOrSPut(Instruction::Code code) {
   return Instruction::SGET <= code && code <= Instruction::SPUT_SHORT;
 }
@@ -181,6 +191,29 @@
   return (code >= Instruction::IPUT) ? IPutMemAccessType(code) : IGetMemAccessType(code);
 }
 
+static inline DexMemAccessType IGetQuickOrIPutQuickMemAccessType(Instruction::Code code) {
+  DCHECK(IsInstructionIGetQuickOrIPutQuick(code));
+  switch (code) {
+    case Instruction::IGET_QUICK: case Instruction::IPUT_QUICK:
+      return kDexMemAccessWord;
+    case Instruction::IGET_WIDE_QUICK: case Instruction::IPUT_WIDE_QUICK:
+      return kDexMemAccessWide;
+    case Instruction::IGET_OBJECT_QUICK: case Instruction::IPUT_OBJECT_QUICK:
+      return kDexMemAccessObject;
+    case Instruction::IGET_BOOLEAN_QUICK: case Instruction::IPUT_BOOLEAN_QUICK:
+      return kDexMemAccessBoolean;
+    case Instruction::IGET_BYTE_QUICK: case Instruction::IPUT_BYTE_QUICK:
+      return kDexMemAccessByte;
+    case Instruction::IGET_CHAR_QUICK: case Instruction::IPUT_CHAR_QUICK:
+      return kDexMemAccessChar;
+    case Instruction::IGET_SHORT_QUICK: case Instruction::IPUT_SHORT_QUICK:
+      return kDexMemAccessShort;
+    default:
+      LOG(FATAL) << code;
+      UNREACHABLE();
+  }
+}
+
 constexpr DexMemAccessType SGetOrSPutMemAccessType(Instruction::Code code) {
 #if __cplusplus >= 201402  // C++14 allows the DCHECK() in constexpr functions.
   DCHECK(IsInstructionSGetOrSPut(opcode));
@@ -197,4 +230,4 @@
 
 }  // namespace art
 
-#endif  // ART_COMPILER_UTILS_DEX_INSTRUCTION_UTILS_H_
+#endif  // ART_RUNTIME_DEX_INSTRUCTION_UTILS_H_
diff --git a/runtime/elf_file.cc b/runtime/elf_file.cc
index a22e274..3490bcf 100644
--- a/runtime/elf_file.cc
+++ b/runtime/elf_file.cc
@@ -1370,7 +1370,7 @@
       reservation_name += file_->GetPath();
       std::unique_ptr<MemMap> reserve(MemMap::MapAnonymous(reservation_name.c_str(),
                                                            reserve_base_override,
-                                                           GetLoadedSize(), PROT_NONE, false,
+                                                           GetLoadedSize(), PROT_NONE, false, false,
                                                            error_msg));
       if (reserve.get() == nullptr) {
         *error_msg = StringPrintf("Failed to allocate %s: %s",
@@ -1411,32 +1411,72 @@
     } else {
       flags |= MAP_PRIVATE;
     }
-    if (file_length < (program_header->p_offset + program_header->p_memsz)) {
-      *error_msg = StringPrintf("File size of %zd bytes not large enough to contain ELF segment "
-                                "%d of %" PRIu64 " bytes: '%s'", file_length, i,
-                                static_cast<uint64_t>(program_header->p_offset + program_header->p_memsz),
+    if (program_header->p_filesz > program_header->p_memsz) {
+      *error_msg = StringPrintf("Invalid p_filesz > p_memsz (%" PRIu64 " > %" PRIu64 "): %s",
+                                static_cast<uint64_t>(program_header->p_filesz),
+                                static_cast<uint64_t>(program_header->p_memsz),
                                 file_->GetPath().c_str());
       return false;
     }
-    std::unique_ptr<MemMap> segment(MemMap::MapFileAtAddress(p_vaddr,
-                                                       program_header->p_memsz,
-                                                       prot, flags, file_->Fd(),
-                                                       program_header->p_offset,
-                                                       true,  // implies MAP_FIXED
-                                                       file_->GetPath().c_str(),
-                                                       error_msg));
-    if (segment.get() == nullptr) {
-      *error_msg = StringPrintf("Failed to map ELF file segment %d from %s: %s",
-                                i, file_->GetPath().c_str(), error_msg->c_str());
+    if (program_header->p_filesz < program_header->p_memsz &&
+        !IsAligned<kPageSize>(program_header->p_filesz)) {
+      *error_msg = StringPrintf("Unsupported unaligned p_filesz < p_memsz (%" PRIu64
+                                " < %" PRIu64 "): %s",
+                                static_cast<uint64_t>(program_header->p_filesz),
+                                static_cast<uint64_t>(program_header->p_memsz),
+                                file_->GetPath().c_str());
       return false;
     }
-    if (segment->Begin() != p_vaddr) {
-      *error_msg = StringPrintf("Failed to map ELF file segment %d from %s at expected address %p, "
-                                "instead mapped to %p",
-                                i, file_->GetPath().c_str(), p_vaddr, segment->Begin());
+    if (file_length < (program_header->p_offset + program_header->p_filesz)) {
+      *error_msg = StringPrintf("File size of %zd bytes not large enough to contain ELF segment "
+                                "%d of %" PRIu64 " bytes: '%s'", file_length, i,
+                                static_cast<uint64_t>(program_header->p_offset + program_header->p_filesz),
+                                file_->GetPath().c_str());
       return false;
     }
-    segments_.push_back(segment.release());
+    if (program_header->p_filesz != 0u) {
+      std::unique_ptr<MemMap> segment(
+          MemMap::MapFileAtAddress(p_vaddr,
+                                   program_header->p_filesz,
+                                   prot, flags, file_->Fd(),
+                                   program_header->p_offset,
+                                   true,  // implies MAP_FIXED
+                                   file_->GetPath().c_str(),
+                                   error_msg));
+      if (segment.get() == nullptr) {
+        *error_msg = StringPrintf("Failed to map ELF file segment %d from %s: %s",
+                                  i, file_->GetPath().c_str(), error_msg->c_str());
+        return false;
+      }
+      if (segment->Begin() != p_vaddr) {
+        *error_msg = StringPrintf("Failed to map ELF file segment %d from %s at expected address %p, "
+                                  "instead mapped to %p",
+                                  i, file_->GetPath().c_str(), p_vaddr, segment->Begin());
+        return false;
+      }
+      segments_.push_back(segment.release());
+    }
+    if (program_header->p_filesz < program_header->p_memsz) {
+      std::string name = StringPrintf("Zero-initialized segment %" PRIu64 " of ELF file %s",
+                                      static_cast<uint64_t>(i), file_->GetPath().c_str());
+      std::unique_ptr<MemMap> segment(
+          MemMap::MapAnonymous(name.c_str(),
+                               p_vaddr + program_header->p_filesz,
+                               program_header->p_memsz - program_header->p_filesz,
+                               prot, false, true /* reuse */, error_msg));
+      if (segment == nullptr) {
+        *error_msg = StringPrintf("Failed to map zero-initialized ELF file segment %d from %s: %s",
+                                  i, file_->GetPath().c_str(), error_msg->c_str());
+        return false;
+      }
+      if (segment->Begin() != p_vaddr) {
+        *error_msg = StringPrintf("Failed to map zero-initialized ELF file segment %d from %s "
+                                  "at expected address %p, instead mapped to %p",
+                                  i, file_->GetPath().c_str(), p_vaddr, segment->Begin());
+        return false;
+      }
+      segments_.push_back(segment.release());
+    }
   }
 
   // Now that we are done loading, .dynamic should be in memory to find .dynstr, .dynsym, .hash
diff --git a/runtime/gc/accounting/atomic_stack.h b/runtime/gc/accounting/atomic_stack.h
index 72734e9..5224d64 100644
--- a/runtime/gc/accounting/atomic_stack.h
+++ b/runtime/gc/accounting/atomic_stack.h
@@ -236,8 +236,8 @@
   // Size in number of elements.
   void Init() {
     std::string error_msg;
-    mem_map_.reset(MemMap::MapAnonymous(name_.c_str(), NULL, capacity_ * sizeof(begin_[0]),
-                                        PROT_READ | PROT_WRITE, false, &error_msg));
+    mem_map_.reset(MemMap::MapAnonymous(name_.c_str(), nullptr, capacity_ * sizeof(begin_[0]),
+                                        PROT_READ | PROT_WRITE, false, false, &error_msg));
     CHECK(mem_map_.get() != NULL) << "couldn't allocate mark stack.\n" << error_msg;
     uint8_t* addr = mem_map_->Begin();
     CHECK(addr != NULL);
diff --git a/runtime/gc/accounting/bitmap-inl.h b/runtime/gc/accounting/bitmap-inl.h
new file mode 100644
index 0000000..e87a0c0
--- /dev/null
+++ b/runtime/gc/accounting/bitmap-inl.h
@@ -0,0 +1,152 @@
+/*
+ * 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.
+ */
+
+#ifndef ART_RUNTIME_GC_ACCOUNTING_BITMAP_INL_H_
+#define ART_RUNTIME_GC_ACCOUNTING_BITMAP_INL_H_
+
+#include "bitmap.h"
+
+#include <memory>
+
+#include "atomic.h"
+#include "base/logging.h"
+#include "utils.h"
+
+namespace art {
+namespace gc {
+namespace accounting {
+
+inline bool Bitmap::AtomicTestAndSetBit(uintptr_t bit_index) {
+  CheckValidBitIndex(bit_index);
+  const size_t word_index = BitIndexToWordIndex(bit_index);
+  const uintptr_t word_mask = BitIndexToMask(bit_index);
+  auto* atomic_entry = reinterpret_cast<Atomic<uintptr_t>*>(&bitmap_begin_[word_index]);
+  uintptr_t old_word;
+  do {
+    old_word = atomic_entry->LoadRelaxed();
+    // Fast path: The bit is already set.
+    if ((old_word & word_mask) != 0) {
+      DCHECK(TestBit(bit_index));
+      return true;
+    }
+  } while (!atomic_entry->CompareExchangeWeakSequentiallyConsistent(old_word,
+                                                                    old_word | word_mask));
+  DCHECK(TestBit(bit_index));
+  return false;
+}
+
+inline bool Bitmap::TestBit(uintptr_t bit_index) const {
+  CheckValidBitIndex(bit_index);
+  return (bitmap_begin_[BitIndexToWordIndex(bit_index)] & BitIndexToMask(bit_index)) != 0;
+}
+
+template<typename Visitor>
+inline void Bitmap::VisitSetBits(uintptr_t bit_start, uintptr_t bit_end, const Visitor& visitor)
+    const {
+  DCHECK_LE(bit_start, bit_end);
+  CheckValidBitIndex(bit_start);
+  const uintptr_t index_start = BitIndexToWordIndex(bit_start);
+  const uintptr_t index_end = BitIndexToWordIndex(bit_end);
+  if (bit_start != bit_end) {
+    CheckValidBitIndex(bit_end - 1);
+  }
+
+  // Index(begin)  ...    Index(end)
+  // [xxxxx???][........][????yyyy]
+  //      ^                   ^
+  //      |                   #---- Bit of visit_end
+  //      #---- Bit of visit_begin
+  //
+
+  // Left edge.
+  uintptr_t left_edge = bitmap_begin_[index_start];
+  // Clear the lower bits that are not in range.
+  left_edge &= ~((static_cast<uintptr_t>(1) << (bit_start % kBitsPerBitmapWord)) - 1);
+
+  // Right edge. Either unique, or left_edge.
+  uintptr_t right_edge;
+
+  if (index_start < index_end) {
+    // Left edge != right edge.
+
+    // Traverse left edge.
+    if (left_edge != 0) {
+      const uintptr_t ptr_base = WordIndexToBitIndex(index_start);
+      do {
+        const size_t shift = CTZ(left_edge);
+        visitor(ptr_base + shift);
+        left_edge ^= static_cast<uintptr_t>(1) << shift;
+      } while (left_edge != 0);
+    }
+
+    // Traverse the middle, full part.
+    for (size_t i = index_start + 1; i < index_end; ++i) {
+      uintptr_t w = bitmap_begin_[i];
+      if (w != 0) {
+        const uintptr_t ptr_base = WordIndexToBitIndex(i);
+        do {
+          const size_t shift = CTZ(w);
+          visitor(ptr_base + shift);
+          w ^= static_cast<uintptr_t>(1) << shift;
+        } while (w != 0);
+      }
+    }
+
+    // Right edge is unique.
+    // But maybe we don't have anything to do: visit_end starts in a new word...
+    if (bit_end == 0) {
+      // Do not read memory, as it could be after the end of the bitmap.
+      right_edge = 0;
+    } else {
+      right_edge = bitmap_begin_[index_end];
+    }
+  } else {
+    right_edge = left_edge;
+  }
+
+  // Right edge handling.
+  right_edge &= ((static_cast<uintptr_t>(1) << (bit_end % kBitsPerBitmapWord)) - 1);
+  if (right_edge != 0) {
+    const uintptr_t ptr_base = WordIndexToBitIndex(index_end);
+    do {
+      const size_t shift = CTZ(right_edge);
+      visitor(ptr_base + shift);
+      right_edge ^= (static_cast<uintptr_t>(1)) << shift;
+    } while (right_edge != 0);
+  }
+}
+
+template<bool kSetBit>
+inline bool Bitmap::ModifyBit(uintptr_t bit_index) {
+  CheckValidBitIndex(bit_index);
+  const size_t word_index = BitIndexToWordIndex(bit_index);
+  const uintptr_t word_mask = BitIndexToMask(bit_index);
+  uintptr_t* address = &bitmap_begin_[word_index];
+  uintptr_t old_word = *address;
+  if (kSetBit) {
+    *address = old_word | word_mask;
+  } else {
+    *address = old_word & ~word_mask;
+  }
+  DCHECK_EQ(TestBit(bit_index), kSetBit);
+  return (old_word & word_mask) != 0;
+}
+
+}  // namespace accounting
+}  // namespace gc
+}  // namespace art
+
+#endif  // ART_RUNTIME_GC_ACCOUNTING_BITMAP_INL_H_
diff --git a/runtime/gc/accounting/bitmap.cc b/runtime/gc/accounting/bitmap.cc
new file mode 100644
index 0000000..20984fd
--- /dev/null
+++ b/runtime/gc/accounting/bitmap.cc
@@ -0,0 +1,93 @@
+/*
+ * 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 "bitmap-inl.h"
+
+#include "card_table.h"
+#include "mem_map.h"
+
+namespace art {
+namespace gc {
+namespace accounting {
+
+Bitmap* Bitmap::CreateFromMemMap(MemMap* mem_map, size_t num_bits) {
+  CHECK(mem_map != nullptr);
+  return new Bitmap(mem_map, num_bits);
+}
+
+Bitmap::Bitmap(MemMap* mem_map, size_t bitmap_size)
+    : mem_map_(mem_map), bitmap_begin_(reinterpret_cast<uintptr_t*>(mem_map->Begin())),
+      bitmap_size_(bitmap_size) {
+  CHECK(bitmap_begin_ != nullptr);
+  CHECK_NE(bitmap_size, 0U);
+}
+
+MemMap* Bitmap::AllocateMemMap(const std::string& name, size_t num_bits) {
+  const size_t bitmap_size = RoundUp(
+      RoundUp(num_bits, kBitsPerBitmapWord) / kBitsPerBitmapWord * sizeof(uintptr_t), kPageSize);
+  std::string error_msg;
+  std::unique_ptr<MemMap> mem_map(MemMap::MapAnonymous(name.c_str(), nullptr, bitmap_size,
+                                                       PROT_READ | PROT_WRITE, false, false,
+                                                       &error_msg));
+  if (UNLIKELY(mem_map.get() == nullptr)) {
+    LOG(ERROR) << "Failed to allocate bitmap " << name << ": " << error_msg;
+    return nullptr;
+  }
+  return mem_map.release();
+}
+
+Bitmap* Bitmap::Create(const std::string& name, size_t num_bits) {
+  auto* const mem_map = AllocateMemMap(name, num_bits);
+  if (mem_map == nullptr) {
+    return nullptr;
+  }
+  return CreateFromMemMap(mem_map, num_bits);
+}
+
+void Bitmap::Clear() {
+  if (bitmap_begin_ != nullptr) {
+    mem_map_->MadviseDontNeedAndZero();
+  }
+}
+
+void Bitmap::CopyFrom(Bitmap* source_bitmap) {
+  DCHECK_EQ(BitmapSize(), source_bitmap->BitmapSize());
+  std::copy(source_bitmap->Begin(),
+            source_bitmap->Begin() + BitmapSize() / kBitsPerBitmapWord, Begin());
+}
+
+template<size_t kAlignment>
+MemoryRangeBitmap<kAlignment>* MemoryRangeBitmap<kAlignment>::Create(
+    const std::string& name, uintptr_t cover_begin, uintptr_t cover_end) {
+  CHECK_ALIGNED(cover_begin, kAlignment);
+  CHECK_ALIGNED(cover_end, kAlignment);
+  const size_t num_bits = (cover_end - cover_begin) / kAlignment;
+  auto* const mem_map = Bitmap::AllocateMemMap(name, num_bits);
+  return CreateFromMemMap(mem_map, cover_begin, num_bits);
+}
+
+template<size_t kAlignment>
+MemoryRangeBitmap<kAlignment>* MemoryRangeBitmap<kAlignment>::CreateFromMemMap(
+    MemMap* mem_map, uintptr_t begin, size_t num_bits) {
+  return new MemoryRangeBitmap(mem_map, begin, num_bits);
+}
+
+template class MemoryRangeBitmap<CardTable::kCardSize>;
+
+}  // namespace accounting
+}  // namespace gc
+}  // namespace art
+
diff --git a/runtime/gc/accounting/bitmap.h b/runtime/gc/accounting/bitmap.h
new file mode 100644
index 0000000..cf2c293
--- /dev/null
+++ b/runtime/gc/accounting/bitmap.h
@@ -0,0 +1,192 @@
+/*
+ * 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.
+ */
+
+#ifndef ART_RUNTIME_GC_ACCOUNTING_BITMAP_H_
+#define ART_RUNTIME_GC_ACCOUNTING_BITMAP_H_
+
+#include <limits.h>
+#include <stdint.h>
+#include <memory>
+#include <set>
+#include <vector>
+
+#include "base/mutex.h"
+#include "globals.h"
+#include "object_callbacks.h"
+
+namespace art {
+
+class MemMap;
+
+namespace gc {
+namespace accounting {
+
+// TODO: Use this code to implement SpaceBitmap.
+class Bitmap {
+ public:
+  // Create and initialize a bitmap with size num_bits. Storage is allocated with a MemMap.
+  static Bitmap* Create(const std::string& name, size_t num_bits);
+
+  // Initialize a space bitmap using the provided mem_map as the live bits. Takes ownership of the
+  // mem map. The address range covered starts at heap_begin and is of size equal to heap_capacity.
+  // Objects are kAlignement-aligned.
+  static Bitmap* CreateFromMemMap(MemMap* mem_map, size_t num_bits);
+
+  // offset is the difference from base to a index.
+  static ALWAYS_INLINE constexpr size_t BitIndexToWordIndex(uintptr_t offset) {
+    return offset / kBitsPerBitmapWord;
+  }
+
+  template<typename T>
+  static ALWAYS_INLINE constexpr T WordIndexToBitIndex(T word_index) {
+    return static_cast<T>(word_index * kBitsPerBitmapWord);
+  }
+
+  static ALWAYS_INLINE constexpr uintptr_t BitIndexToMask(uintptr_t bit_index) {
+    return static_cast<uintptr_t>(1) << (bit_index % kBitsPerBitmapWord);
+  }
+
+  ALWAYS_INLINE bool SetBit(size_t bit_index) {
+    return ModifyBit<true>(bit_index);
+  }
+
+  ALWAYS_INLINE bool ClearBit(size_t bit_index) {
+    return ModifyBit<false>(bit_index);
+  }
+
+  ALWAYS_INLINE bool TestBit(size_t bit_index) const;
+
+  // Returns true if the bit_index was previously set.
+  ALWAYS_INLINE bool AtomicTestAndSetBit(size_t bit_index);
+
+  // Fill the bitmap with zeroes.  Returns the bitmap's memory to the system as a side-effect.
+  void Clear();
+
+  // Visit the all the set bits range [visit_begin, visit_end) where visit_begin and visit_end are
+  // bit indices visitor is called with the index of each set bit.
+  template <typename Visitor>
+  void VisitSetBits(uintptr_t visit_begin, size_t visit_end, const Visitor& visitor) const;
+
+  void CopyFrom(Bitmap* source_bitmap);
+
+  // Starting address of our internal storage.
+  uintptr_t* Begin() {
+    return bitmap_begin_;
+  }
+
+  // Size of our bitmap in bits.
+  size_t BitmapSize() const {
+    return bitmap_size_;
+  }
+
+  // Check that a bit index is valid with a DCHECK.
+  ALWAYS_INLINE void CheckValidBitIndex(size_t bit_index) const {
+    DCHECK_LT(bit_index, BitmapSize());
+  }
+
+  std::string Dump() const;
+
+ protected:
+  static constexpr size_t kBitsPerBitmapWord = sizeof(uintptr_t) * kBitsPerByte;
+
+  Bitmap(MemMap* mem_map, size_t bitmap_size);
+
+  // Allocate the mem-map for a bitmap based on how many bits are required.
+  static MemMap* AllocateMemMap(const std::string& name, size_t num_bits);
+
+  template<bool kSetBit>
+  ALWAYS_INLINE bool ModifyBit(uintptr_t bit_index);
+
+  // Backing storage for bitmap.
+  std::unique_ptr<MemMap> mem_map_;
+
+  // This bitmap itself, word sized for efficiency in scanning.
+  uintptr_t* const bitmap_begin_;
+
+  // Number of bits in the bitmap.
+  const size_t bitmap_size_;
+
+ private:
+  DISALLOW_COPY_AND_ASSIGN(Bitmap);
+};
+
+// One bit per kAlignment in range (start, end]
+template<size_t kAlignment>
+class MemoryRangeBitmap : public Bitmap {
+ public:
+  static MemoryRangeBitmap* Create(const std::string& name, uintptr_t cover_begin,
+                                   uintptr_t cover_end);
+  static MemoryRangeBitmap* CreateFromMemMap(MemMap* mem_map, uintptr_t cover_begin,
+                                             size_t num_bits);
+
+  // Beginning of the memory range that the bitmap covers.
+  ALWAYS_INLINE uintptr_t CoverBegin() const {
+    return cover_begin_;
+  }
+
+  // End of the memory range that the bitmap covers.
+  ALWAYS_INLINE uintptr_t CoverEnd() const {
+    return cover_end_;
+  }
+
+  // Return the address associated with a bit index.
+  ALWAYS_INLINE uintptr_t AddrFromBitIndex(size_t bit_index) const {
+    const uintptr_t addr = CoverBegin() + bit_index * kAlignment;
+    DCHECK_EQ(BitIndexFromAddr(addr), bit_index);
+    return addr;
+  }
+
+  // Return the bit index associated with an address .
+  ALWAYS_INLINE uintptr_t BitIndexFromAddr(uintptr_t addr) const {
+    DCHECK(HasAddress(addr)) << CoverBegin() << " <= " <<  addr << " < " << CoverEnd();
+    return (addr - CoverBegin()) / kAlignment;
+  }
+
+  ALWAYS_INLINE bool HasAddress(const uintptr_t addr) const {
+    return cover_begin_ <= addr && addr < cover_end_;
+  }
+
+  ALWAYS_INLINE bool Set(uintptr_t addr) {
+    return SetBit(BitIndexFromAddr(addr));
+  }
+
+  ALWAYS_INLINE bool Clear(size_t addr) {
+    return ClearBit(BitIndexFromAddr(addr));
+  }
+
+  ALWAYS_INLINE bool Test(size_t addr) const {
+    return TestBit(BitIndexFromAddr(addr));
+  }
+
+  // Returns true if the object was previously set.
+  ALWAYS_INLINE bool AtomicTestAndSet(size_t addr) {
+    return AtomicTestAndSetBit(BitIndexFromAddr(addr));
+  }
+
+ private:
+  MemoryRangeBitmap(MemMap* mem_map, uintptr_t begin, size_t num_bits)
+     : Bitmap(mem_map, num_bits), cover_begin_(begin), cover_end_(begin + kAlignment * num_bits) {
+  }
+
+  uintptr_t const cover_begin_;
+  uintptr_t const cover_end_;
+};
+
+}  // namespace accounting
+}  // namespace gc
+}  // namespace art
+
+#endif  // ART_RUNTIME_GC_ACCOUNTING_BITMAP_H_
diff --git a/runtime/gc/accounting/card_table.cc b/runtime/gc/accounting/card_table.cc
index ca1e7c1..ad1f192 100644
--- a/runtime/gc/accounting/card_table.cc
+++ b/runtime/gc/accounting/card_table.cc
@@ -62,7 +62,7 @@
   std::string error_msg;
   std::unique_ptr<MemMap> mem_map(
       MemMap::MapAnonymous("card table", nullptr, capacity + 256, PROT_READ | PROT_WRITE,
-                           false, &error_msg));
+                           false, false, &error_msg));
   CHECK(mem_map.get() != NULL) << "couldn't allocate card table: " << error_msg;
   // All zeros is the correct initial value; all clean. Anonymous mmaps are initialized to zero, we
   // don't clear the card table to avoid unnecessary pages being allocated
diff --git a/runtime/gc/accounting/mod_union_table.cc b/runtime/gc/accounting/mod_union_table.cc
index b1ccc0b..a3fac58 100644
--- a/runtime/gc/accounting/mod_union_table.cc
+++ b/runtime/gc/accounting/mod_union_table.cc
@@ -19,6 +19,7 @@
 #include <memory>
 
 #include "base/stl_util.h"
+#include "bitmap-inl.h"
 #include "card_table-inl.h"
 #include "heap_bitmap.h"
 #include "gc/accounting/space_bitmap-inl.h"
@@ -40,14 +41,14 @@
 namespace gc {
 namespace accounting {
 
-class ModUnionClearCardSetVisitor {
+class ModUnionAddToCardSetVisitor {
  public:
-  explicit ModUnionClearCardSetVisitor(ModUnionTable::CardSet* const cleared_cards)
-    : cleared_cards_(cleared_cards) {
+  explicit ModUnionAddToCardSetVisitor(ModUnionTable::CardSet* const cleared_cards)
+      : cleared_cards_(cleared_cards) {
   }
 
-  inline void operator()(uint8_t* card, uint8_t expected_value, uint8_t new_value) const {
-    UNUSED(new_value);
+  inline void operator()(uint8_t* card, uint8_t expected_value,
+                         uint8_t new_value ATTRIBUTE_UNUSED) const {
     if (expected_value == CardTable::kCardDirty) {
       cleared_cards_->insert(card);
     }
@@ -57,18 +58,38 @@
   ModUnionTable::CardSet* const cleared_cards_;
 };
 
-class ModUnionClearCardVisitor {
+class ModUnionAddToCardBitmapVisitor {
  public:
-  explicit ModUnionClearCardVisitor(std::vector<uint8_t*>* cleared_cards)
-    : cleared_cards_(cleared_cards) {
+  explicit ModUnionAddToCardBitmapVisitor(ModUnionTable::CardBitmap* bitmap,
+                                          CardTable* card_table)
+      : bitmap_(bitmap), card_table_(card_table) {
   }
 
-  void operator()(uint8_t* card, uint8_t expected_card, uint8_t new_card) const {
-    UNUSED(new_card);
+  inline void operator()(uint8_t* card, uint8_t expected_value,
+                         uint8_t new_value ATTRIBUTE_UNUSED) const {
+    if (expected_value == CardTable::kCardDirty) {
+      // We want the address the card represents, not the address of the card.
+      bitmap_->Set(reinterpret_cast<uintptr_t>(card_table_->AddrFromCard(card)));
+    }
+  }
+
+ private:
+  ModUnionTable::CardBitmap* const bitmap_;
+  CardTable* const card_table_;
+};
+
+class ModUnionAddToCardVectorVisitor {
+ public:
+  explicit ModUnionAddToCardVectorVisitor(std::vector<uint8_t*>* cleared_cards)
+      : cleared_cards_(cleared_cards) {
+  }
+
+  void operator()(uint8_t* card, uint8_t expected_card, uint8_t new_card ATTRIBUTE_UNUSED) const {
     if (expected_card == CardTable::kCardDirty) {
       cleared_cards_->push_back(card);
     }
   }
+
  private:
   std::vector<uint8_t*>* const cleared_cards_;
 };
@@ -77,19 +98,19 @@
  public:
   ModUnionUpdateObjectReferencesVisitor(MarkHeapReferenceCallback* callback, void* arg,
                                         space::ContinuousSpace* from_space,
-                                        space::ImageSpace* image_space,
+                                        space::ContinuousSpace* immune_space,
                                         bool* contains_reference_to_other_space)
-    : callback_(callback), arg_(arg), from_space_(from_space), image_space_(image_space),
+    : callback_(callback), arg_(arg), from_space_(from_space), immune_space_(immune_space),
       contains_reference_to_other_space_(contains_reference_to_other_space) {
   }
 
   // Extra parameters are required since we use this same visitor signature for checking objects.
-  void operator()(Object* obj, MemberOffset offset, bool /*is_static*/) const
+  void operator()(Object* obj, MemberOffset offset, bool is_static ATTRIBUTE_UNUSED) const
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     // Only add the reference if it is non null and fits our criteria.
-    mirror::HeapReference<Object>* obj_ptr = obj->GetFieldObjectReferenceAddr(offset);
+    mirror::HeapReference<Object>* const obj_ptr = obj->GetFieldObjectReferenceAddr(offset);
     mirror::Object* ref = obj_ptr->AsMirrorPtr();
-    if (ref != nullptr && !from_space_->HasAddress(ref) && !image_space_->HasAddress(ref)) {
+    if (ref != nullptr && !from_space_->HasAddress(ref) && !immune_space_->HasAddress(ref)) {
       *contains_reference_to_other_space_ = true;
       callback_(obj_ptr, arg_);
     }
@@ -97,27 +118,30 @@
 
  private:
   MarkHeapReferenceCallback* const callback_;
-  void* arg_;
+  void* const arg_;
   // Space which we are scanning
   space::ContinuousSpace* const from_space_;
-  space::ImageSpace* const image_space_;
+  space::ContinuousSpace* const immune_space_;
   // Set if we have any references to another space.
   bool* const contains_reference_to_other_space_;
 };
 
 class ModUnionScanImageRootVisitor {
  public:
+  // Immune space is any other space which we don't care about references to. Currently this is
+  // the image space in the case of the zygote mod union table.
   ModUnionScanImageRootVisitor(MarkHeapReferenceCallback* callback, void* arg,
-                               space::ContinuousSpace* from_space, space::ImageSpace* image_space,
+                               space::ContinuousSpace* from_space,
+                               space::ContinuousSpace* immune_space,
                                bool* contains_reference_to_other_space)
-      : callback_(callback), arg_(arg), from_space_(from_space), image_space_(image_space),
+      : callback_(callback), arg_(arg), from_space_(from_space), immune_space_(immune_space),
         contains_reference_to_other_space_(contains_reference_to_other_space) {}
 
   void operator()(Object* root) const
       EXCLUSIVE_LOCKS_REQUIRED(Locks::heap_bitmap_lock_)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    DCHECK(root != NULL);
-    ModUnionUpdateObjectReferencesVisitor ref_visitor(callback_, arg_, from_space_, image_space_,
+    DCHECK(root != nullptr);
+    ModUnionUpdateObjectReferencesVisitor ref_visitor(callback_, arg_, from_space_, immune_space_,
                                                       contains_reference_to_other_space_);
     root->VisitReferences<kMovingClasses>(ref_visitor, VoidFunctor());
   }
@@ -127,14 +151,14 @@
   void* const arg_;
   // Space which we are scanning
   space::ContinuousSpace* const from_space_;
-  space::ImageSpace* const image_space_;
+  space::ContinuousSpace* const immune_space_;
   // Set if we have any references to another space.
   bool* const contains_reference_to_other_space_;
 };
 
 void ModUnionTableReferenceCache::ClearCards() {
   CardTable* card_table = GetHeap()->GetCardTable();
-  ModUnionClearCardSetVisitor visitor(&cleared_cards_);
+  ModUnionAddToCardSetVisitor visitor(&cleared_cards_);
   // Clear dirty cards in the this space and update the corresponding mod-union bits.
   card_table->ModifyCardsAtomic(space_->Begin(), space_->End(), AgeCardVisitor(), visitor);
 }
@@ -324,9 +348,54 @@
   }
 }
 
+ModUnionTableCardCache::ModUnionTableCardCache(const std::string& name, Heap* heap,
+                                               space::ContinuousSpace* space)
+    : ModUnionTable(name, heap, space) {
+  // Normally here we could use End() instead of Limit(), but for testing we may want to have a
+  // mod-union table for a space which can still grow.
+  if (!space->IsImageSpace()) {
+    CHECK_ALIGNED(reinterpret_cast<uintptr_t>(space->Limit()), CardTable::kCardSize);
+  }
+  card_bitmap_.reset(CardBitmap::Create(
+      "mod union bitmap", reinterpret_cast<uintptr_t>(space->Begin()),
+      RoundUp(reinterpret_cast<uintptr_t>(space->Limit()), CardTable::kCardSize)));
+}
+
+class CardBitVisitor {
+ public:
+  CardBitVisitor(MarkHeapReferenceCallback* callback, void* arg, space::ContinuousSpace* space,
+                 space::ContinuousSpace* immune_space, ModUnionTable::CardBitmap* card_bitmap)
+      : callback_(callback), arg_(arg), space_(space), immune_space_(immune_space),
+        bitmap_(space->GetLiveBitmap()), card_bitmap_(card_bitmap) {
+    DCHECK(immune_space_ != nullptr);
+  }
+
+  void operator()(size_t bit_index) const {
+    const uintptr_t start = card_bitmap_->AddrFromBitIndex(bit_index);
+    DCHECK(space_->HasAddress(reinterpret_cast<mirror::Object*>(start)))
+        << start << " " << *space_;
+    bool reference_to_other_space = false;
+    ModUnionScanImageRootVisitor scan_visitor(callback_, arg_, space_, immune_space_,
+                                              &reference_to_other_space);
+    bitmap_->VisitMarkedRange(start, start + CardTable::kCardSize, scan_visitor);
+    if (!reference_to_other_space) {
+      // No non null reference to another space, clear the bit.
+      card_bitmap_->ClearBit(bit_index);
+    }
+  }
+
+ private:
+  MarkHeapReferenceCallback* const callback_;
+  void* const arg_;
+  space::ContinuousSpace* const space_;
+  space::ContinuousSpace* const immune_space_;
+  ContinuousSpaceBitmap* const bitmap_;
+  ModUnionTable::CardBitmap* const card_bitmap_;
+};
+
 void ModUnionTableCardCache::ClearCards() {
-  CardTable* card_table = GetHeap()->GetCardTable();
-  ModUnionClearCardSetVisitor visitor(&cleared_cards_);
+  CardTable* const card_table = GetHeap()->GetCardTable();
+  ModUnionAddToCardBitmapVisitor visitor(card_bitmap_.get(), card_table);
   // Clear dirty cards in the this space and update the corresponding mod-union bits.
   card_table->ModifyCardsAtomic(space_->Begin(), space_->End(), AgeCardVisitor(), visitor);
 }
@@ -334,46 +403,51 @@
 // Mark all references to the alloc space(s).
 void ModUnionTableCardCache::UpdateAndMarkReferences(MarkHeapReferenceCallback* callback,
                                                      void* arg) {
-  CardTable* card_table = heap_->GetCardTable();
-  space::ImageSpace* image_space = heap_->GetImageSpace();
-  ContinuousSpaceBitmap* bitmap = space_->GetLiveBitmap();
-  bool reference_to_other_space = false;
-  ModUnionScanImageRootVisitor scan_visitor(callback, arg, space_, image_space,
-                                            &reference_to_other_space);
-  for (auto it = cleared_cards_.begin(), end = cleared_cards_.end(); it != end; ) {
-    uintptr_t start = reinterpret_cast<uintptr_t>(card_table->AddrFromCard(*it));
-    DCHECK(space_->HasAddress(reinterpret_cast<Object*>(start)));
-    reference_to_other_space = false;
-    bitmap->VisitMarkedRange(start, start + CardTable::kCardSize, scan_visitor);
-    if (!reference_to_other_space) {
-      // No non null reference to another space, remove the card.
-      it = cleared_cards_.erase(it);
-    } else {
-      ++it;
-    }
-  }
+  auto* image_space = heap_->GetImageSpace();
+  // If we don't have an image space, just pass in space_ as the immune space. Pass in the same
+  // space_ instead of image_space to avoid a null check in ModUnionUpdateObjectReferencesVisitor.
+  CardBitVisitor visitor(callback, arg, space_, image_space != nullptr ? image_space : space_,
+      card_bitmap_.get());
+  card_bitmap_->VisitSetBits(
+      0, RoundUp(space_->Size(), CardTable::kCardSize) / CardTable::kCardSize, visitor);
 }
 
 void ModUnionTableCardCache::Dump(std::ostream& os) {
-  CardTable* card_table = heap_->GetCardTable();
   os << "ModUnionTable dirty cards: [";
-  for (const uint8_t* card_addr : cleared_cards_) {
-    auto start = reinterpret_cast<uintptr_t>(card_table->AddrFromCard(card_addr));
-    auto end = start + CardTable::kCardSize;
-    os << reinterpret_cast<void*>(start) << "-" << reinterpret_cast<void*>(end) << "\n";
+  // TODO: Find cleaner way of doing this.
+  for (uint8_t* addr = space_->Begin(); addr < AlignUp(space_->End(), CardTable::kCardSize);
+      addr += CardTable::kCardSize) {
+    if (card_bitmap_->Test(reinterpret_cast<uintptr_t>(addr))) {
+      os << reinterpret_cast<void*>(addr) << "-"
+         << reinterpret_cast<void*>(addr + CardTable::kCardSize) << "\n";
+    }
   }
   os << "]";
 }
 
 void ModUnionTableCardCache::SetCards() {
-  CardTable* card_table = heap_->GetCardTable();
+  // Only clean up to the end since there cannot be any objects past the End() of the space.
   for (uint8_t* addr = space_->Begin(); addr < AlignUp(space_->End(), CardTable::kCardSize);
        addr += CardTable::kCardSize) {
-    cleared_cards_.insert(card_table->CardFromAddr(addr));
+    card_bitmap_->Set(reinterpret_cast<uintptr_t>(addr));
   }
 }
 
+bool ModUnionTableCardCache::ContainsCardFor(uintptr_t addr) {
+  return card_bitmap_->Test(addr);
+}
+
 void ModUnionTableReferenceCache::SetCards() {
+  for (uint8_t* addr = space_->Begin(); addr < AlignUp(space_->End(), CardTable::kCardSize);
+       addr += CardTable::kCardSize) {
+    cleared_cards_.insert(heap_->GetCardTable()->CardFromAddr(reinterpret_cast<void*>(addr)));
+  }
+}
+
+bool ModUnionTableReferenceCache::ContainsCardFor(uintptr_t addr) {
+  auto* card_ptr = heap_->GetCardTable()->CardFromAddr(reinterpret_cast<void*>(addr));
+  return cleared_cards_.find(card_ptr) != cleared_cards_.end() ||
+      references_.find(card_ptr) != references_.end();
 }
 
 }  // namespace accounting
diff --git a/runtime/gc/accounting/mod_union_table.h b/runtime/gc/accounting/mod_union_table.h
index d6342cf..2e232ca 100644
--- a/runtime/gc/accounting/mod_union_table.h
+++ b/runtime/gc/accounting/mod_union_table.h
@@ -17,7 +17,9 @@
 #ifndef ART_RUNTIME_GC_ACCOUNTING_MOD_UNION_TABLE_H_
 #define ART_RUNTIME_GC_ACCOUNTING_MOD_UNION_TABLE_H_
 
+#include "bitmap.h"
 #include "base/allocator.h"
+#include "card_table.h"
 #include "globals.h"
 #include "object_callbacks.h"
 #include "safe_map.h"
@@ -44,6 +46,7 @@
 
 namespace accounting {
 
+class Bitmap;
 class HeapBitmap;
 
 // The mod-union table is the union of modified cards. It is used to allow the card table to be
@@ -52,6 +55,7 @@
  public:
   typedef std::set<uint8_t*, std::less<uint8_t*>,
                    TrackingAllocator<uint8_t*, kAllocatorTagModUnionCardSet>> CardSet;
+  typedef MemoryRangeBitmap<CardTable::kCardSize> CardBitmap;
 
   explicit ModUnionTable(const std::string& name, Heap* heap, space::ContinuousSpace* space)
       : name_(name),
@@ -80,6 +84,10 @@
   // bitmap or not.
   virtual void Verify() EXCLUSIVE_LOCKS_REQUIRED(Locks::heap_bitmap_lock_) = 0;
 
+  // Returns true if a card is marked inside the mod union table. Used for testing. The address
+  // doesn't need to be aligned.
+  virtual bool ContainsCardFor(uintptr_t addr) = 0;
+
   virtual void Dump(std::ostream& os) = 0;
   space::ContinuousSpace* GetSpace() {
     return space_;
@@ -106,25 +114,27 @@
   virtual ~ModUnionTableReferenceCache() {}
 
   // Clear and store cards for a space.
-  void ClearCards();
+  void ClearCards() OVERRIDE;
 
   // Update table based on cleared cards and mark all references to the other spaces.
-  void UpdateAndMarkReferences(MarkHeapReferenceCallback* callback, void* arg)
+  void UpdateAndMarkReferences(MarkHeapReferenceCallback* callback, void* arg) OVERRIDE
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_)
       EXCLUSIVE_LOCKS_REQUIRED(Locks::heap_bitmap_lock_);
 
   // Exclusive lock is required since verify uses SpaceBitmap::VisitMarkedRange and
   // VisitMarkedRange can't know if the callback will modify the bitmap or not.
-  void Verify()
+  void Verify() OVERRIDE
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_)
       EXCLUSIVE_LOCKS_REQUIRED(Locks::heap_bitmap_lock_);
 
   // Function that tells whether or not to add a reference to the table.
   virtual bool ShouldAddReference(const mirror::Object* ref) const = 0;
 
-  void Dump(std::ostream& os) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  virtual bool ContainsCardFor(uintptr_t addr) OVERRIDE;
 
-  void SetCards() OVERRIDE;
+  virtual void Dump(std::ostream& os) OVERRIDE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
+  virtual void SetCards() OVERRIDE;
 
  protected:
   // Cleared card array, used to update the mod-union table.
@@ -138,28 +148,32 @@
 // Card caching implementation. Keeps track of which cards we cleared and only this information.
 class ModUnionTableCardCache : public ModUnionTable {
  public:
-  explicit ModUnionTableCardCache(const std::string& name, Heap* heap, space::ContinuousSpace* space)
-      : ModUnionTable(name, heap, space) {}
+  // Note: There is assumption that the space End() doesn't change.
+  explicit ModUnionTableCardCache(const std::string& name, Heap* heap,
+                                  space::ContinuousSpace* space);
   virtual ~ModUnionTableCardCache() {}
 
   // Clear and store cards for a space.
-  void ClearCards();
+  virtual void ClearCards() OVERRIDE;
 
   // Mark all references to the alloc space(s).
-  void UpdateAndMarkReferences(MarkHeapReferenceCallback* callback, void* arg)
+  virtual void UpdateAndMarkReferences(MarkHeapReferenceCallback* callback, void* arg) OVERRIDE
       EXCLUSIVE_LOCKS_REQUIRED(Locks::heap_bitmap_lock_)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   // Nothing to verify.
-  void Verify() {}
+  virtual void Verify() OVERRIDE {}
 
-  void Dump(std::ostream& os);
+  virtual void Dump(std::ostream& os) OVERRIDE;
 
-  void SetCards() OVERRIDE;
+  virtual bool ContainsCardFor(uintptr_t addr) OVERRIDE;
+
+  // Sets all the cards in the mod union table to be marked.
+  virtual void SetCards() OVERRIDE;
 
  protected:
-  // Cleared card array, used to update the mod-union table.
-  CardSet cleared_cards_;
+  // Cleared card bitmap, used to update the mod-union table.
+  std::unique_ptr<CardBitmap> card_bitmap_;
 };
 
 }  // namespace accounting
diff --git a/runtime/gc/accounting/mod_union_table_test.cc b/runtime/gc/accounting/mod_union_table_test.cc
new file mode 100644
index 0000000..87ce166
--- /dev/null
+++ b/runtime/gc/accounting/mod_union_table_test.cc
@@ -0,0 +1,242 @@
+/*
+ * 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 "mod_union_table-inl.h"
+
+#include "common_runtime_test.h"
+#include "gc/space/space-inl.h"
+#include "mirror/array-inl.h"
+#include "space_bitmap-inl.h"
+#include "thread-inl.h"
+
+namespace art {
+namespace gc {
+namespace accounting {
+
+class ModUnionTableFactory {
+ public:
+  enum TableType {
+    kTableTypeCardCache,
+    kTableTypeReferenceCache,
+    kTableTypeCount,  // Number of values in the enum.
+  };
+
+  // Target space is ignored for the card cache implementation.
+  static ModUnionTable* Create(
+      TableType type, space::ContinuousSpace* space, space::ContinuousSpace* target_space);
+};
+
+class ModUnionTableTest : public CommonRuntimeTest {
+ public:
+  ModUnionTableTest() : java_lang_object_array_(nullptr) {
+  }
+  mirror::ObjectArray<mirror::Object>* AllocObjectArray(
+      Thread* self, space::ContinuousMemMapAllocSpace* space, size_t component_count)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+    auto* klass = GetObjectArrayClass(self, space);
+    const size_t size = ComputeArraySize(self, klass, component_count, 2);
+    size_t bytes_allocated = 0;
+    auto* obj = down_cast<mirror::ObjectArray<mirror::Object>*>(
+        space->Alloc(self, size, &bytes_allocated, nullptr));
+    if (obj != nullptr) {
+      obj->SetClass(klass);
+      obj->SetLength(static_cast<int32_t>(component_count));
+      space->GetLiveBitmap()->Set(obj);
+      EXPECT_GE(bytes_allocated, size);
+    }
+    return obj;
+  }
+  void ResetClass() {
+    java_lang_object_array_ = nullptr;
+  }
+  void RunTest(ModUnionTableFactory::TableType type);
+
+ private:
+  mirror::Class* GetObjectArrayClass(Thread* self, space::ContinuousMemMapAllocSpace* space)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+    if (java_lang_object_array_ == nullptr) {
+      java_lang_object_array_ =
+          Runtime::Current()->GetClassLinker()->GetClassRoot(ClassLinker::kObjectArrayClass);
+      // Since the test doesn't have an image, the class of the object array keeps cards live
+      // inside the card cache mod-union table and causes the check
+      // ASSERT_FALSE(table->ContainsCardFor(reinterpret_cast<uintptr_t>(obj3)));
+      // to fail since the class ends up keeping the card dirty. To get around this, we make a fake
+      // copy of the class in the same space that we are allocating in.
+      DCHECK(java_lang_object_array_ != nullptr);
+      const size_t class_size = java_lang_object_array_->GetClassSize();
+      size_t bytes_allocated = 0;
+      auto* klass = down_cast<mirror::Class*>(space->Alloc(self, class_size, &bytes_allocated,
+                                                           nullptr));
+      DCHECK(klass != nullptr);
+      memcpy(klass, java_lang_object_array_, class_size);
+      Runtime::Current()->GetHeap()->GetCardTable()->MarkCard(klass);
+      java_lang_object_array_ = klass;
+    }
+    return java_lang_object_array_;
+  }
+  mirror::Class* java_lang_object_array_;
+};
+
+// Collect visited objects into container.
+static void CollectVisitedCallback(mirror::HeapReference<mirror::Object>* ref, void* arg)
+    SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+  DCHECK(ref != nullptr);
+  DCHECK(arg != nullptr);
+  reinterpret_cast<std::set<mirror::Object*>*>(arg)->insert(ref->AsMirrorPtr());
+}
+
+// A mod union table that only holds references to a specified target space.
+class ModUnionTableRefCacheToSpace : public ModUnionTableReferenceCache {
+ public:
+  explicit ModUnionTableRefCacheToSpace(
+      const std::string& name, Heap* heap, space::ContinuousSpace* space,
+      space::ContinuousSpace* target_space)
+      : ModUnionTableReferenceCache(name, heap, space), target_space_(target_space) {}
+
+  bool ShouldAddReference(const mirror::Object* ref) const OVERRIDE {
+    return target_space_->HasAddress(ref);
+  }
+
+ private:
+  space::ContinuousSpace* const target_space_;
+};
+
+std::ostream& operator<<(std::ostream& oss, ModUnionTableFactory::TableType type) {
+  switch (type) {
+    case ModUnionTableFactory::kTableTypeCardCache: {
+      oss << "CardCache";
+      break;
+    }
+    case ModUnionTableFactory::kTableTypeReferenceCache: {
+      oss << "ReferenceCache";
+      break;
+    }
+    default: {
+      UNIMPLEMENTED(FATAL) << static_cast<size_t>(type);
+    }
+  }
+  return oss;
+}
+
+ModUnionTable* ModUnionTableFactory::Create(
+    TableType type, space::ContinuousSpace* space, space::ContinuousSpace* target_space) {
+  std::ostringstream name;
+  name << "Mod union table: " << type;
+  switch (type) {
+    case kTableTypeCardCache: {
+      return new ModUnionTableCardCache(name.str(), Runtime::Current()->GetHeap(), space);
+    }
+    case kTableTypeReferenceCache: {
+      return new ModUnionTableRefCacheToSpace(name.str(), Runtime::Current()->GetHeap(), space,
+                                              target_space);
+    }
+    default: {
+      UNIMPLEMENTED(FATAL) << "Invalid type " << type;
+    }
+  }
+  return nullptr;
+}
+
+TEST_F(ModUnionTableTest, TestCardCache) {
+  RunTest(ModUnionTableFactory::kTableTypeCardCache);
+}
+
+TEST_F(ModUnionTableTest, TestReferenceCache) {
+  RunTest(ModUnionTableFactory::kTableTypeReferenceCache);
+}
+
+void ModUnionTableTest::RunTest(ModUnionTableFactory::TableType type) {
+  Thread* const self = Thread::Current();
+  ScopedObjectAccess soa(self);
+  Runtime* const runtime = Runtime::Current();
+  gc::Heap* const heap = runtime->GetHeap();
+  // Use non moving space since moving GC don't necessarily have a primary free list space.
+  auto* space = heap->GetNonMovingSpace();
+  ResetClass();
+  // Create another space that we can put references in.
+  std::unique_ptr<space::DlMallocSpace> other_space(space::DlMallocSpace::Create(
+      "other space", 128 * KB, 4 * MB, 4 * MB, nullptr, false));
+  ASSERT_TRUE(other_space.get() != nullptr);
+  heap->AddSpace(other_space.get());
+  std::unique_ptr<ModUnionTable> table(ModUnionTableFactory::Create(
+      type, space, other_space.get()));
+  ASSERT_TRUE(table.get() != nullptr);
+  // Create some fake objects and put the main space and dirty cards in the non moving space.
+  auto* obj1 = AllocObjectArray(self, space, CardTable::kCardSize);
+  ASSERT_TRUE(obj1 != nullptr);
+  auto* obj2 = AllocObjectArray(self, space, CardTable::kCardSize);
+  ASSERT_TRUE(obj2 != nullptr);
+  auto* obj3 = AllocObjectArray(self, space, CardTable::kCardSize);
+  ASSERT_TRUE(obj3 != nullptr);
+  auto* obj4 = AllocObjectArray(self, space, CardTable::kCardSize);
+  ASSERT_TRUE(obj4 != nullptr);
+  // Dirty some cards.
+  obj1->Set(0, obj2);
+  obj2->Set(0, obj3);
+  obj3->Set(0, obj4);
+  obj4->Set(0, obj1);
+  // Dirty some more cards to objects in another space.
+  auto* other_space_ref1 = AllocObjectArray(self, other_space.get(), CardTable::kCardSize);
+  ASSERT_TRUE(other_space_ref1 != nullptr);
+  auto* other_space_ref2 = AllocObjectArray(self, other_space.get(), CardTable::kCardSize);
+  ASSERT_TRUE(other_space_ref2 != nullptr);
+  obj1->Set(1, other_space_ref1);
+  obj2->Set(3, other_space_ref2);
+  table->ClearCards();
+  std::set<mirror::Object*> visited;
+  table->UpdateAndMarkReferences(&CollectVisitedCallback, &visited);
+  // Check that we visited all the references in other spaces only.
+  ASSERT_GE(visited.size(), 2u);
+  ASSERT_TRUE(visited.find(other_space_ref1) != visited.end());
+  ASSERT_TRUE(visited.find(other_space_ref2) != visited.end());
+  // Verify that all the other references were visited.
+  // obj1, obj2 cards should still be in mod union table since they have references to other
+  // spaces.
+  ASSERT_TRUE(table->ContainsCardFor(reinterpret_cast<uintptr_t>(obj1)));
+  ASSERT_TRUE(table->ContainsCardFor(reinterpret_cast<uintptr_t>(obj2)));
+  // obj3, obj4 don't have a reference to any object in the other space, their cards should have
+  // been removed from the mod union table during UpdateAndMarkReferences.
+  ASSERT_FALSE(table->ContainsCardFor(reinterpret_cast<uintptr_t>(obj3)));
+  ASSERT_FALSE(table->ContainsCardFor(reinterpret_cast<uintptr_t>(obj4)));
+  {
+    // Currently no-op, make sure it still works however.
+    ReaderMutexLock mu(self, *Locks::heap_bitmap_lock_);
+    table->Verify();
+  }
+  // Verify that dump doesn't crash.
+  std::ostringstream oss;
+  table->Dump(oss);
+  // Set all the cards, then verify.
+  table->SetCards();
+  // TODO: Check that the cards are actually set.
+  for (auto* ptr = space->Begin(); ptr < AlignUp(space->End(), CardTable::kCardSize);
+      ptr += CardTable::kCardSize) {
+    ASSERT_TRUE(table->ContainsCardFor(reinterpret_cast<uintptr_t>(ptr)));
+  }
+  // Visit again and make sure the cards got cleared back to their sane state.
+  visited.clear();
+  table->UpdateAndMarkReferences(&CollectVisitedCallback, &visited);
+  // Verify that the dump matches what we saw earlier.
+  std::ostringstream oss2;
+  table->Dump(oss2);
+  ASSERT_EQ(oss.str(), oss2.str());
+  // Remove the space we added so it doesn't persist to the next test.
+  heap->RemoveSpace(other_space.get());
+}
+
+}  // namespace accounting
+}  // namespace gc
+}  // namespace art
diff --git a/runtime/gc/accounting/read_barrier_table.h b/runtime/gc/accounting/read_barrier_table.h
index 84d5da3..bb9aae7 100644
--- a/runtime/gc/accounting/read_barrier_table.h
+++ b/runtime/gc/accounting/read_barrier_table.h
@@ -37,7 +37,7 @@
               static_cast<uint64_t>(static_cast<size_t>(kHeapCapacity / kRegionSize)));
     std::string error_msg;
     MemMap* mem_map = MemMap::MapAnonymous("read barrier table", nullptr, capacity,
-                                           PROT_READ | PROT_WRITE, false, &error_msg);
+                                           PROT_READ | PROT_WRITE, false, false, &error_msg);
     CHECK(mem_map != nullptr && mem_map->Begin() != nullptr)
         << "couldn't allocate read barrier table: " << error_msg;
     mem_map_.reset(mem_map);
diff --git a/runtime/gc/accounting/space_bitmap.cc b/runtime/gc/accounting/space_bitmap.cc
index f5d3b47..ad8d988 100644
--- a/runtime/gc/accounting/space_bitmap.cc
+++ b/runtime/gc/accounting/space_bitmap.cc
@@ -63,7 +63,8 @@
   const size_t bitmap_size = ComputeBitmapSize(heap_capacity);
   std::string error_msg;
   std::unique_ptr<MemMap> mem_map(MemMap::MapAnonymous(name.c_str(), nullptr, bitmap_size,
-                                                       PROT_READ | PROT_WRITE, false, &error_msg));
+                                                       PROT_READ | PROT_WRITE, false, false,
+                                                       &error_msg));
   if (UNLIKELY(mem_map.get() == nullptr)) {
     LOG(ERROR) << "Failed to allocate bitmap " << name << ": " << error_msg;
     return nullptr;
diff --git a/runtime/gc/accounting/space_bitmap.h b/runtime/gc/accounting/space_bitmap.h
index 7bc83ef..d6b3ed4 100644
--- a/runtime/gc/accounting/space_bitmap.h
+++ b/runtime/gc/accounting/space_bitmap.h
@@ -188,13 +188,6 @@
 
   std::string Dump() const;
 
-  const void* GetObjectWordAddress(const mirror::Object* obj) const {
-    uintptr_t addr = reinterpret_cast<uintptr_t>(obj);
-    const uintptr_t offset = addr - heap_begin_;
-    const size_t index = OffsetToIndex(offset);
-    return &bitmap_begin_[index];
-  }
-
  private:
   // TODO: heap_end_ is initialized so that the heap bitmap is empty, this doesn't require the -1,
   // however, we document that this is expected on heap_end_
diff --git a/runtime/gc/allocator/rosalloc.cc b/runtime/gc/allocator/rosalloc.cc
index 72aacf5..f51093a 100644
--- a/runtime/gc/allocator/rosalloc.cc
+++ b/runtime/gc/allocator/rosalloc.cc
@@ -80,8 +80,9 @@
   size_t num_of_pages = footprint_ / kPageSize;
   size_t max_num_of_pages = max_capacity_ / kPageSize;
   std::string error_msg;
-  page_map_mem_map_.reset(MemMap::MapAnonymous("rosalloc page map", NULL, RoundUp(max_num_of_pages, kPageSize),
-                                               PROT_READ | PROT_WRITE, false, &error_msg));
+  page_map_mem_map_.reset(MemMap::MapAnonymous("rosalloc page map", nullptr,
+                                               RoundUp(max_num_of_pages, kPageSize),
+                                               PROT_READ | PROT_WRITE, false, false, &error_msg));
   CHECK(page_map_mem_map_.get() != nullptr) << "Couldn't allocate the page map : " << error_msg;
   page_map_ = page_map_mem_map_->Begin();
   page_map_size_ = num_of_pages;
diff --git a/runtime/gc/collector/mark_sweep.cc b/runtime/gc/collector/mark_sweep.cc
index cd63d26..8aac484 100644
--- a/runtime/gc/collector/mark_sweep.cc
+++ b/runtime/gc/collector/mark_sweep.cc
@@ -106,7 +106,7 @@
   MemMap* mem_map = MemMap::MapAnonymous(
       "mark sweep sweep array free buffer", nullptr,
       RoundUp(kSweepArrayChunkFreeSize * sizeof(mirror::Object*), kPageSize),
-      PROT_READ | PROT_WRITE, false, &error_msg);
+      PROT_READ | PROT_WRITE, false, false, &error_msg);
   CHECK(mem_map != nullptr) << "Couldn't allocate sweep array free buffer: " << error_msg;
   sweep_array_free_buffer_mem_map_.reset(mem_map);
 }
diff --git a/runtime/gc/heap.cc b/runtime/gc/heap.cc
index 419d555..9343622 100644
--- a/runtime/gc/heap.cc
+++ b/runtime/gc/heap.cc
@@ -284,7 +284,8 @@
     // address.
     non_moving_space_mem_map.reset(
         MemMap::MapAnonymous(space_name, requested_alloc_space_begin,
-                             non_moving_space_capacity, PROT_READ | PROT_WRITE, true, &error_str));
+                             non_moving_space_capacity, PROT_READ | PROT_WRITE, true, false,
+                             &error_str));
     CHECK(non_moving_space_mem_map != nullptr) << error_str;
     // Try to reserve virtual memory at a lower address if we have a separate non moving space.
     request_begin = reinterpret_cast<uint8_t*>(300 * MB);
@@ -398,14 +399,14 @@
     rb_table_.reset(new accounting::ReadBarrierTable());
     DCHECK(rb_table_->IsAllCleared());
   }
-
-  // Card cache for now since it makes it easier for us to update the references to the copying
-  // spaces.
-  accounting::ModUnionTable* mod_union_table =
-      new accounting::ModUnionTableToZygoteAllocspace("Image mod-union table", this,
-                                                      GetImageSpace());
-  CHECK(mod_union_table != nullptr) << "Failed to create image mod-union table";
-  AddModUnionTable(mod_union_table);
+  if (GetImageSpace() != nullptr) {
+    // Don't add the image mod union table if we are running without an image, this can crash if
+    // we use the CardCache implementation.
+    accounting::ModUnionTable* mod_union_table = new accounting::ModUnionTableToZygoteAllocspace(
+        "Image mod-union table", this, GetImageSpace());
+    CHECK(mod_union_table != nullptr) << "Failed to create image mod-union table";
+    AddModUnionTable(mod_union_table);
+  }
   if (collector::SemiSpace::kUseRememberedSet && non_moving_space_ != main_space_) {
     accounting::RememberedSet* non_moving_space_rem_set =
         new accounting::RememberedSet("Non-moving space remembered set", this, non_moving_space_);
@@ -476,7 +477,7 @@
                                            size_t capacity, std::string* out_error_str) {
   while (true) {
     MemMap* map = MemMap::MapAnonymous(name, request_begin, capacity,
-                                       PROT_READ | PROT_WRITE, true, out_error_str);
+                                       PROT_READ | PROT_WRITE, true, false, out_error_str);
     if (map != nullptr || request_begin == nullptr) {
       return map;
     }
@@ -655,7 +656,7 @@
 }
 
 bool Heap::IsCompilingBoot() const {
-  if (!Runtime::Current()->IsCompiler()) {
+  if (!Runtime::Current()->IsAotCompiler()) {
     return false;
   }
   for (const auto& space : continuous_spaces_) {
@@ -1675,7 +1676,8 @@
   AddSpace(to_space);
   // Make sure that we will have enough room to copy.
   CHECK_GE(to_space->GetFootprintLimit(), from_space->GetFootprintLimit());
-  Compact(to_space, from_space, kGcCauseHomogeneousSpaceCompact);
+  collector::GarbageCollector* collector = Compact(to_space, from_space,
+                                                   kGcCauseHomogeneousSpaceCompact);
   const uint64_t space_size_after_compaction = to_space->Size();
   main_space_ = to_space;
   main_space_backup_.reset(from_space);
@@ -1694,6 +1696,7 @@
   // Finish GC.
   reference_processor_.EnqueueClearedReferences(self);
   GrowForUtilization(semi_space_collector_);
+  LogGC(kGcCauseHomogeneousSpaceCompact, collector);
   FinishGC(self, collector::kGcTypeFull);
   return HomogeneousSpaceCompactResult::kSuccess;
 }
@@ -1744,6 +1747,7 @@
     FinishGC(self, collector::kGcTypeNone);
     return;
   }
+  collector::GarbageCollector* collector = nullptr;
   tl->SuspendAll();
   switch (collector_type) {
     case kCollectorTypeSS: {
@@ -1758,7 +1762,7 @@
         bump_pointer_space_ = space::BumpPointerSpace::CreateFromMemMap("Bump pointer space",
                                                                         mem_map.release());
         AddSpace(bump_pointer_space_);
-        Compact(bump_pointer_space_, main_space_, kGcCauseCollectorTransition);
+        collector = Compact(bump_pointer_space_, main_space_, kGcCauseCollectorTransition);
         // Use the now empty main space mem map for the bump pointer temp space.
         mem_map.reset(main_space_->ReleaseMemMap());
         // Unset the pointers just in case.
@@ -1795,7 +1799,7 @@
         mem_map.release();
         // Compact to the main space from the bump pointer space, don't need to swap semispaces.
         AddSpace(main_space_);
-        Compact(main_space_, bump_pointer_space_, kGcCauseCollectorTransition);
+        collector = Compact(main_space_, bump_pointer_space_, kGcCauseCollectorTransition);
         mem_map.reset(bump_pointer_space_->ReleaseMemMap());
         RemoveSpace(bump_pointer_space_);
         bump_pointer_space_ = nullptr;
@@ -1826,6 +1830,8 @@
   reference_processor_.EnqueueClearedReferences(self);
   uint64_t duration = NanoTime() - start_time;
   GrowForUtilization(semi_space_collector_);
+  DCHECK(collector != nullptr);
+  LogGC(kGcCauseCollectorTransition, collector);
   FinishGC(self, collector::kGcTypeFull);
   int32_t after_allocated = num_bytes_allocated_.LoadSequentiallyConsistent();
   int32_t delta_allocated = before_allocated - after_allocated;
@@ -2166,9 +2172,9 @@
   std::swap(bump_pointer_space_, temp_space_);
 }
 
-void Heap::Compact(space::ContinuousMemMapAllocSpace* target_space,
-                   space::ContinuousMemMapAllocSpace* source_space,
-                   GcCause gc_cause) {
+collector::GarbageCollector* Heap::Compact(space::ContinuousMemMapAllocSpace* target_space,
+                                           space::ContinuousMemMapAllocSpace* source_space,
+                                           GcCause gc_cause) {
   CHECK(kMovingCollector);
   if (target_space != source_space) {
     // Don't swap spaces since this isn't a typical semi space collection.
@@ -2176,11 +2182,13 @@
     semi_space_collector_->SetFromSpace(source_space);
     semi_space_collector_->SetToSpace(target_space);
     semi_space_collector_->Run(gc_cause, false);
+    return semi_space_collector_;
   } else {
     CHECK(target_space->IsBumpPointerSpace())
         << "In-place compaction is only supported for bump pointer spaces";
     mark_compact_collector_->SetSpace(target_space->AsBumpPointerSpace());
     mark_compact_collector_->Run(kGcCauseCollectorTransition, false);
+    return mark_compact_collector_;
   }
 }
 
@@ -2291,6 +2299,14 @@
   reference_processor_.EnqueueClearedReferences(self);
   // Grow the heap so that we know when to perform the next GC.
   GrowForUtilization(collector, bytes_allocated_before_gc);
+  LogGC(gc_cause, collector);
+  FinishGC(self, gc_type);
+  // Inform DDMS that a GC completed.
+  Dbg::GcDidFinish();
+  return gc_type;
+}
+
+void Heap::LogGC(GcCause gc_cause, collector::GarbageCollector* collector) {
   const size_t duration = GetCurrentGcIteration()->GetDurationNs();
   const std::vector<uint64_t>& pause_times = GetCurrentGcIteration()->GetPauseTimes();
   // Print the GC if it is an explicit GC (e.g. Runtime.gc()) or a slow GC
@@ -2310,8 +2326,8 @@
     const size_t total_memory = GetTotalMemory();
     std::ostringstream pause_string;
     for (size_t i = 0; i < pause_times.size(); ++i) {
-        pause_string << PrettyDuration((pause_times[i] / 1000) * 1000)
-                     << ((i != pause_times.size() - 1) ? "," : "");
+      pause_string << PrettyDuration((pause_times[i] / 1000) * 1000)
+                   << ((i != pause_times.size() - 1) ? "," : "");
     }
     LOG(INFO) << gc_cause << " " << collector->GetName()
               << " GC freed "  << current_gc_iteration_.GetFreedObjects() << "("
@@ -2323,10 +2339,6 @@
               << " total " << PrettyDuration((duration / 1000) * 1000);
     VLOG(heap) << Dumpable<TimingLogger>(*current_gc_iteration_.GetTimings());
   }
-  FinishGC(self, gc_type);
-  // Inform DDMS that a GC completed.
-  Dbg::GcDidFinish();
-  return gc_type;
 }
 
 void Heap::FinishGC(Thread* self, collector::GcType gc_type) {
diff --git a/runtime/gc/heap.h b/runtime/gc/heap.h
index 57c1460..b2478e6 100644
--- a/runtime/gc/heap.h
+++ b/runtime/gc/heap.h
@@ -666,12 +666,13 @@
   class CollectorTransitionTask;
   class HeapTrimTask;
 
-  // Compact source space to target space.
-  void Compact(space::ContinuousMemMapAllocSpace* target_space,
-               space::ContinuousMemMapAllocSpace* source_space,
-               GcCause gc_cause)
+  // Compact source space to target space. Returns the collector used.
+  collector::GarbageCollector* Compact(space::ContinuousMemMapAllocSpace* target_space,
+                                       space::ContinuousMemMapAllocSpace* source_space,
+                                       GcCause gc_cause)
       EXCLUSIVE_LOCKS_REQUIRED(Locks::mutator_lock_);
 
+  void LogGC(GcCause gc_cause, collector::GarbageCollector* collector);
   void FinishGC(Thread* self, collector::GcType gc_type) LOCKS_EXCLUDED(gc_complete_lock_);
 
   // Create a mem map with a preferred base address.
diff --git a/runtime/gc/space/bump_pointer_space.cc b/runtime/gc/space/bump_pointer_space.cc
index 9675ba6..fbfc449 100644
--- a/runtime/gc/space/bump_pointer_space.cc
+++ b/runtime/gc/space/bump_pointer_space.cc
@@ -29,7 +29,8 @@
   capacity = RoundUp(capacity, kPageSize);
   std::string error_msg;
   std::unique_ptr<MemMap> mem_map(MemMap::MapAnonymous(name.c_str(), requested_begin, capacity,
-                                                 PROT_READ | PROT_WRITE, true, &error_msg));
+                                                       PROT_READ | PROT_WRITE, true, false,
+                                                       &error_msg));
   if (mem_map.get() == nullptr) {
     LOG(ERROR) << "Failed to allocate pages for alloc space (" << name << ") of size "
         << PrettySize(capacity) << " with message " << error_msg;
diff --git a/runtime/gc/space/image_space.cc b/runtime/gc/space/image_space.cc
index d873e6d..14f770d 100644
--- a/runtime/gc/space/image_space.cc
+++ b/runtime/gc/space/image_space.cc
@@ -788,7 +788,7 @@
 
   OatFile* oat_file = OatFile::Open(oat_filename, oat_filename, image_header.GetOatDataBegin(),
                                     image_header.GetOatFileBegin(),
-                                    !Runtime::Current()->IsCompiler(), error_msg);
+                                    !Runtime::Current()->IsAotCompiler(), error_msg);
   if (oat_file == NULL) {
     *error_msg = StringPrintf("Failed to open oat file '%s' referenced from image %s: %s",
                               oat_filename.c_str(), GetName(), error_msg->c_str());
diff --git a/runtime/gc/space/large_object_space.cc b/runtime/gc/space/large_object_space.cc
index c0c6444..7523de5 100644
--- a/runtime/gc/space/large_object_space.cc
+++ b/runtime/gc/space/large_object_space.cc
@@ -110,8 +110,8 @@
 mirror::Object* LargeObjectMapSpace::Alloc(Thread* self, size_t num_bytes,
                                            size_t* bytes_allocated, size_t* usable_size) {
   std::string error_msg;
-  MemMap* mem_map = MemMap::MapAnonymous("large object space allocation", NULL, num_bytes,
-                                         PROT_READ | PROT_WRITE, true, &error_msg);
+  MemMap* mem_map = MemMap::MapAnonymous("large object space allocation", nullptr, num_bytes,
+                                         PROT_READ | PROT_WRITE, true, false, &error_msg);
   if (UNLIKELY(mem_map == NULL)) {
     LOG(WARNING) << "Large object allocation failed: " << error_msg;
     return NULL;
@@ -291,7 +291,7 @@
   CHECK_EQ(size % kAlignment, 0U);
   std::string error_msg;
   MemMap* mem_map = MemMap::MapAnonymous(name.c_str(), requested_begin, size,
-                                         PROT_READ | PROT_WRITE, true, &error_msg);
+                                         PROT_READ | PROT_WRITE, true, false, &error_msg);
   CHECK(mem_map != NULL) << "Failed to allocate large object space mem map: " << error_msg;
   return new FreeListSpace(name, mem_map, mem_map->Begin(), mem_map->End());
 }
@@ -305,9 +305,10 @@
   CHECK_ALIGNED(space_capacity, kAlignment);
   const size_t alloc_info_size = sizeof(AllocationInfo) * (space_capacity / kAlignment);
   std::string error_msg;
-  allocation_info_map_.reset(MemMap::MapAnonymous("large object free list space allocation info map",
-                                                  nullptr, alloc_info_size, PROT_READ | PROT_WRITE,
-                                                  false, &error_msg));
+  allocation_info_map_.reset(
+      MemMap::MapAnonymous("large object free list space allocation info map",
+                           nullptr, alloc_info_size, PROT_READ | PROT_WRITE,
+                           false, false, &error_msg));
   CHECK(allocation_info_map_.get() != nullptr) << "Failed to allocate allocation info map"
       << error_msg;
   allocation_info_ = reinterpret_cast<AllocationInfo*>(allocation_info_map_->Begin());
diff --git a/runtime/gc/space/malloc_space.cc b/runtime/gc/space/malloc_space.cc
index 9bbbb3c..67e8847 100644
--- a/runtime/gc/space/malloc_space.cc
+++ b/runtime/gc/space/malloc_space.cc
@@ -90,7 +90,7 @@
 
   std::string error_msg;
   MemMap* mem_map = MemMap::MapAnonymous(name.c_str(), requested_begin, *capacity,
-                                         PROT_READ | PROT_WRITE, true, &error_msg);
+                                         PROT_READ | PROT_WRITE, true, false, &error_msg);
   if (mem_map == nullptr) {
     LOG(ERROR) << "Failed to allocate pages for alloc space (" << name << ") of size "
                << PrettySize(*capacity) << ": " << error_msg;
diff --git a/runtime/gc/space/region_space.cc b/runtime/gc/space/region_space.cc
index 2c556d9..8bb73d6 100644
--- a/runtime/gc/space/region_space.cc
+++ b/runtime/gc/space/region_space.cc
@@ -33,7 +33,8 @@
   capacity = RoundUp(capacity, kRegionSize);
   std::string error_msg;
   std::unique_ptr<MemMap> mem_map(MemMap::MapAnonymous(name.c_str(), requested_begin, capacity,
-                                                       PROT_READ | PROT_WRITE, true, &error_msg));
+                                                       PROT_READ | PROT_WRITE, true, false,
+                                                       &error_msg));
   if (mem_map.get() == nullptr) {
     LOG(ERROR) << "Failed to allocate pages for alloc space (" << name << ") of size "
         << PrettySize(capacity) << " with message " << error_msg;
diff --git a/runtime/indirect_reference_table.cc b/runtime/indirect_reference_table.cc
index aa2a6b5..1a3f107 100644
--- a/runtime/indirect_reference_table.cc
+++ b/runtime/indirect_reference_table.cc
@@ -74,7 +74,7 @@
   std::string error_str;
   const size_t table_bytes = maxCount * sizeof(IrtEntry);
   table_mem_map_.reset(MemMap::MapAnonymous("indirect ref table", nullptr, table_bytes,
-                                            PROT_READ | PROT_WRITE, false, &error_str));
+                                            PROT_READ | PROT_WRITE, false, false, &error_str));
   CHECK(table_mem_map_.get() != nullptr) << error_str;
   CHECK_EQ(table_mem_map_->Size(), table_bytes);
   table_ = reinterpret_cast<IrtEntry*>(table_mem_map_->Begin());
diff --git a/runtime/instrumentation.cc b/runtime/instrumentation.cc
index 90115c3..a054462 100644
--- a/runtime/instrumentation.cc
+++ b/runtime/instrumentation.cc
@@ -31,6 +31,8 @@
 #include "entrypoints/runtime_asm_entrypoints.h"
 #include "gc_root-inl.h"
 #include "interpreter/interpreter.h"
+#include "jit/jit.h"
+#include "jit/jit_code_cache.h"
 #include "mirror/art_method-inl.h"
 #include "mirror/class-inl.h"
 #include "mirror/dex_cache.h"
@@ -92,6 +94,16 @@
 
 static void UpdateEntrypoints(mirror::ArtMethod* method, const void* quick_code)
     SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+  Runtime* const runtime = Runtime::Current();
+  jit::Jit* jit = runtime->GetJit();
+  if (jit != nullptr) {
+    const void* old_code_ptr = method->GetEntryPointFromQuickCompiledCode();
+    jit::JitCodeCache* code_cache = jit->GetCodeCache();
+    if (code_cache->ContainsCodePtr(old_code_ptr)) {
+      // Save the old compiled code since we need it to implement ClassLinker::GetQuickOatCodeFor.
+      code_cache->SaveCompiledCode(method, old_code_ptr);
+    }
+  }
   method->SetEntryPointFromQuickCompiledCode(quick_code);
   if (!method->IsResolutionMethod()) {
     ClassLinker* class_linker = Runtime::Current()->GetClassLinker();
@@ -120,7 +132,8 @@
   }
   const void* new_quick_code;
   bool uninstall = !entry_exit_stubs_installed_ && !interpreter_stubs_installed_;
-  ClassLinker* class_linker = Runtime::Current()->GetClassLinker();
+  Runtime* const runtime = Runtime::Current();
+  ClassLinker* const class_linker = runtime->GetClassLinker();
   bool is_class_initialized = method->GetDeclaringClass()->IsInitialized();
   if (uninstall) {
     if ((forced_interpret_only_ || IsDeoptimized(method)) && !method->IsNative()) {
@@ -143,7 +156,6 @@
           new_quick_code = GetQuickInstrumentationEntryPoint();
         } else {
           new_quick_code = class_linker->GetQuickOatCodeFor(method);
-          DCHECK(!class_linker->IsQuickToInterpreterBridge(new_quick_code));
         }
       } else {
         new_quick_code = GetQuickResolutionStub();
@@ -397,6 +409,10 @@
     method_unwind_listeners_.push_back(listener);
     have_method_unwind_listeners_ = true;
   }
+  if ((events & kBackwardBranch) != 0) {
+    backward_branch_listeners_.push_back(listener);
+    have_backward_branch_listeners_ = true;
+  }
   if ((events & kDexPcMoved) != 0) {
     std::list<InstrumentationListener*>* modified;
     if (have_dex_pc_listeners_) {
@@ -904,6 +920,13 @@
   }
 }
 
+void Instrumentation::BackwardBranchImpl(Thread* thread, mirror::ArtMethod* method,
+                                         int32_t offset) const {
+  for (InstrumentationListener* listener : backward_branch_listeners_) {
+    listener->BackwardBranch(thread, method, offset);
+  }
+}
+
 void Instrumentation::FieldReadEventImpl(Thread* thread, mirror::Object* this_object,
                                          mirror::ArtMethod* method, uint32_t dex_pc,
                                          mirror::ArtField* field) const {
diff --git a/runtime/instrumentation.h b/runtime/instrumentation.h
index cea0388..b667a40 100644
--- a/runtime/instrumentation.h
+++ b/runtime/instrumentation.h
@@ -94,6 +94,10 @@
                                mirror::ArtMethod* catch_method, uint32_t catch_dex_pc,
                                mirror::Throwable* exception_object)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) = 0;
+
+  // Call-back for when we get a backward branch.
+  virtual void BackwardBranch(Thread* thread, mirror::ArtMethod* method, int32_t dex_pc_offset)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) = 0;
 };
 
 // Instrumentation is a catch-all for when extra information is required from the runtime. The
@@ -103,13 +107,14 @@
 class Instrumentation {
  public:
   enum InstrumentationEvent {
-    kMethodEntered   = 1,  // 1 << 0
-    kMethodExited    = 2,  // 1 << 1
-    kMethodUnwind    = 4,  // 1 << 2
-    kDexPcMoved      = 8,  // 1 << 3
-    kFieldRead       = 16,  // 1 << 4,
-    kFieldWritten    = 32,  // 1 << 5
-    kExceptionCaught = 64,  // 1 << 6
+    kMethodEntered = 0x1,
+    kMethodExited = 0x2,
+    kMethodUnwind = 0x4,
+    kDexPcMoved = 0x8,
+    kFieldRead = 0x10,
+    kFieldWritten = 0x20,
+    kExceptionCaught = 0x40,
+    kBackwardBranch = 0x80,
   };
 
   Instrumentation();
@@ -244,6 +249,10 @@
     return have_exception_caught_listeners_;
   }
 
+  bool HasBackwardBranchListeners() const SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+    return have_backward_branch_listeners_;
+  }
+
   bool IsActive() const SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     return have_dex_pc_listeners_ || have_method_entry_listeners_ || have_method_exit_listeners_ ||
         have_field_read_listeners_ || have_field_write_listeners_ ||
@@ -284,6 +293,14 @@
     }
   }
 
+  // Inform listeners that a backward branch has been taken (only supported by the interpreter).
+  void BackwardBranch(Thread* thread, mirror::ArtMethod* method, int32_t offset) const
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+    if (UNLIKELY(HasBackwardBranchListeners())) {
+      BackwardBranchImpl(thread, method, offset);
+    }
+  }
+
   // Inform listeners that we read a field (only supported by the interpreter).
   void FieldReadEvent(Thread* thread, mirror::Object* this_object,
                       mirror::ArtMethod* method, uint32_t dex_pc,
@@ -361,6 +378,8 @@
   void DexPcMovedEventImpl(Thread* thread, mirror::Object* this_object,
                            mirror::ArtMethod* method, uint32_t dex_pc) const
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  void BackwardBranchImpl(Thread* thread, mirror::ArtMethod* method, int32_t offset) const
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
   void FieldReadEventImpl(Thread* thread, mirror::Object* this_object,
                            mirror::ArtMethod* method, uint32_t dex_pc,
                            mirror::ArtField* field) const
@@ -429,10 +448,14 @@
   // Do we have any exception caught listeners? Short-cut to avoid taking the instrumentation_lock_.
   bool have_exception_caught_listeners_ GUARDED_BY(Locks::mutator_lock_);
 
+  // Do we have any backward branch listeners? Short-cut to avoid taking the instrumentation_lock_.
+  bool have_backward_branch_listeners_ GUARDED_BY(Locks::mutator_lock_);
+
   // The event listeners, written to with the mutator_lock_ exclusively held.
   std::list<InstrumentationListener*> method_entry_listeners_ GUARDED_BY(Locks::mutator_lock_);
   std::list<InstrumentationListener*> method_exit_listeners_ GUARDED_BY(Locks::mutator_lock_);
   std::list<InstrumentationListener*> method_unwind_listeners_ GUARDED_BY(Locks::mutator_lock_);
+  std::list<InstrumentationListener*> backward_branch_listeners_ GUARDED_BY(Locks::mutator_lock_);
   std::shared_ptr<std::list<InstrumentationListener*>> dex_pc_listeners_
       GUARDED_BY(Locks::mutator_lock_);
   std::shared_ptr<std::list<InstrumentationListener*>> field_read_listeners_
diff --git a/runtime/interpreter/interpreter_common.cc b/runtime/interpreter/interpreter_common.cc
index a29558e..3ab7f30 100644
--- a/runtime/interpreter/interpreter_common.cc
+++ b/runtime/interpreter/interpreter_common.cc
@@ -16,6 +16,8 @@
 
 #include "interpreter_common.h"
 
+#include <cmath>
+
 #include "mirror/array-inl.h"
 
 namespace art {
@@ -839,6 +841,23 @@
   result->SetL(found);
 }
 
+// Common helper for class-loading cutouts in an unstarted runtime. We call Runtime methods that
+// rely on Java code to wrap errors in the correct exception class (i.e., NoClassDefFoundError into
+// ClassNotFoundException), so need to do the same. The only exception is if the exception is
+// actually InternalError. This must not be wrapped, as it signals an initialization abort.
+static void CheckExceptionGenerateClassNotFound(Thread* self)
+    SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+  if (self->IsExceptionPending()) {
+    // If it is not an InternalError, wrap it.
+    std::string type(PrettyTypeOf(self->GetException(nullptr)));
+    if (type != "java.lang.InternalError") {
+      self->ThrowNewWrappedException(self->GetCurrentLocationForThrow(),
+                                     "Ljava/lang/ClassNotFoundException;",
+                                     "ClassNotFoundException");
+    }
+  }
+}
+
 static void UnstartedRuntimeInvoke(Thread* self,  const DexFile::CodeItem* code_item,
                                    ShadowFrame* shadow_frame,
                                    JValue* result, size_t arg_offset) {
@@ -846,18 +865,34 @@
   // problems in core libraries.
   std::string name(PrettyMethod(shadow_frame->GetMethod()));
   if (name == "java.lang.Class java.lang.Class.forName(java.lang.String)") {
-    // TODO: Support for the other variants that take more arguments should also be added.
     mirror::String* class_name = shadow_frame->GetVRegReference(arg_offset)->AsString();
     StackHandleScope<1> hs(self);
     Handle<mirror::String> h_class_name(hs.NewHandle(class_name));
     UnstartedRuntimeFindClass(self, h_class_name, NullHandle<mirror::ClassLoader>(), result, name,
-                              true, true);
-  } else if (name == "java.lang.Class java.lang.VMClassLoader.loadClass(java.lang.String, boolean)") {
+                              true, false);
+    CheckExceptionGenerateClassNotFound(self);
+  } else if (name == "java.lang.Class java.lang.Class.forName(java.lang.String, boolean, java.lang.ClassLoader)") {
     mirror::String* class_name = shadow_frame->GetVRegReference(arg_offset)->AsString();
-    StackHandleScope<1> hs(self);
+    bool initialize_class = shadow_frame->GetVReg(arg_offset + 1) != 0;
+    mirror::ClassLoader* class_loader =
+        down_cast<mirror::ClassLoader*>(shadow_frame->GetVRegReference(arg_offset + 2));
+    StackHandleScope<2> hs(self);
     Handle<mirror::String> h_class_name(hs.NewHandle(class_name));
-    UnstartedRuntimeFindClass(self, h_class_name, NullHandle<mirror::ClassLoader>(), result, name,
-                              false, true);
+    Handle<mirror::ClassLoader> h_class_loader(hs.NewHandle(class_loader));
+    UnstartedRuntimeFindClass(self, h_class_name, h_class_loader, result, name, initialize_class,
+                              false);
+    CheckExceptionGenerateClassNotFound(self);
+  } else if (name == "java.lang.Class java.lang.Class.classForName(java.lang.String, boolean, java.lang.ClassLoader)") {
+    mirror::String* class_name = shadow_frame->GetVRegReference(arg_offset)->AsString();
+    bool initialize_class = shadow_frame->GetVReg(arg_offset + 1) != 0;
+    mirror::ClassLoader* class_loader =
+        down_cast<mirror::ClassLoader*>(shadow_frame->GetVRegReference(arg_offset + 2));
+    StackHandleScope<2> hs(self);
+    Handle<mirror::String> h_class_name(hs.NewHandle(class_name));
+    Handle<mirror::ClassLoader> h_class_loader(hs.NewHandle(class_loader));
+    UnstartedRuntimeFindClass(self, h_class_name, h_class_loader, result, name, initialize_class,
+                              false);
+    CheckExceptionGenerateClassNotFound(self);
   } else if (name == "java.lang.Class java.lang.VMClassLoader.findLoadedClass(java.lang.ClassLoader, java.lang.String)") {
     mirror::String* class_name = shadow_frame->GetVRegReference(arg_offset + 1)->AsString();
     mirror::ClassLoader* class_loader =
@@ -866,17 +901,47 @@
     Handle<mirror::String> h_class_name(hs.NewHandle(class_name));
     Handle<mirror::ClassLoader> h_class_loader(hs.NewHandle(class_loader));
     UnstartedRuntimeFindClass(self, h_class_name, h_class_loader, result, name, false, false);
+    // This might have an error pending. But semantics are to just return null.
+    if (self->IsExceptionPending()) {
+      // If it is an InternalError, keep it. See CheckExceptionGenerateClassNotFound.
+      std::string type(PrettyTypeOf(self->GetException(nullptr)));
+      if (type != "java.lang.InternalError") {
+        self->ClearException();
+      }
+    }
   } else if (name == "java.lang.Class java.lang.Void.lookupType()") {
     result->SetL(Runtime::Current()->GetClassLinker()->FindPrimitiveClass('V'));
   } else if (name == "java.lang.Object java.lang.Class.newInstance()") {
+    StackHandleScope<2> hs(self);
     Class* klass = shadow_frame->GetVRegReference(arg_offset)->AsClass();
-    ArtMethod* c = klass->FindDeclaredDirectMethod("<init>", "()V");
-    CHECK(c != NULL);
-    StackHandleScope<1> hs(self);
-    Handle<Object> obj(hs.NewHandle(klass->AllocObject(self)));
-    CHECK(obj.Get() != NULL);
-    EnterInterpreterFromInvoke(self, c, obj.Get(), NULL, NULL);
-    result->SetL(obj.Get());
+    Handle<Class> h_klass(hs.NewHandle(klass));
+    // There are two situations in which we'll abort this run.
+    //  1) If the class isn't yet initialized and initialization fails.
+    //  2) If we can't find the default constructor. We'll postpone the exception to runtime.
+    // Note that 2) could likely be handled here, but for safety abort the transaction.
+    bool ok = false;
+    if (Runtime::Current()->GetClassLinker()->EnsureInitialized(self, h_klass, true, true)) {
+      ArtMethod* c = h_klass->FindDeclaredDirectMethod("<init>", "()V");
+      if (c != nullptr) {
+        Handle<Object> obj(hs.NewHandle(klass->AllocObject(self)));
+        CHECK(obj.Get() != nullptr);  // We don't expect OOM at compile-time.
+        EnterInterpreterFromInvoke(self, c, obj.Get(), nullptr, nullptr);
+        result->SetL(obj.Get());
+        ok = true;
+      } else {
+        self->ThrowNewExceptionF(self->GetCurrentLocationForThrow(), "Ljava/lang/InternalError;",
+                                 "Could not find default constructor for '%s'",
+                                 PrettyClass(h_klass.Get()).c_str());
+      }
+    }
+    if (!ok) {
+      std::string error_msg = StringPrintf("Failed in Class.newInstance for '%s' with %s",
+                                           PrettyClass(h_klass.Get()).c_str(),
+                                           PrettyTypeOf(self->GetException(nullptr)).c_str());
+      self->ThrowNewWrappedException(self->GetCurrentLocationForThrow(),
+                                     "Ljava/lang/InternalError;",
+                                     error_msg.c_str());
+    }
   } else if (name == "java.lang.reflect.Field java.lang.Class.getDeclaredField(java.lang.String)") {
     // Special managed code cut-out to allow field lookup in a un-started runtime that'd fail
     // going the reflective Dex way.
@@ -949,12 +1014,67 @@
                                "Unimplemented System.arraycopy for type '%s'",
                                PrettyDescriptor(ctype).c_str());
     }
-  } else  if (name == "java.lang.Object java.lang.ThreadLocal.get()") {
+  } else if (name == "long java.lang.Double.doubleToRawLongBits(double)") {
+    double in = shadow_frame->GetVRegDouble(arg_offset);
+    result->SetJ(bit_cast<int64_t>(in));
+  } else if (name == "double java.lang.Math.ceil(double)") {
+    double in = shadow_frame->GetVRegDouble(arg_offset);
+    double out;
+    // Special cases:
+    // 1) NaN, infinity, +0, -0 -> out := in. All are guaranteed by cmath.
+    // -1 < in < 0 -> out := -0.
+    if (-1.0 < in && in < 0) {
+      out = -0.0;
+    } else {
+      out = ceil(in);
+    }
+    result->SetD(out);
+  } else if (name == "java.lang.Object java.lang.ThreadLocal.get()") {
     std::string caller(PrettyMethod(shadow_frame->GetLink()->GetMethod()));
+    bool ok = false;
     if (caller == "java.lang.String java.lang.IntegralToString.convertInt(java.lang.AbstractStringBuilder, int)") {
       // Allocate non-threadlocal buffer.
       result->SetL(mirror::CharArray::Alloc(self, 11));
-    } else {
+      ok = true;
+    } else if (caller == "java.lang.RealToString java.lang.RealToString.getInstance()") {
+      // Note: RealToString is implemented and used in a different fashion than IntegralToString.
+      // Conversion is done over an actual object of RealToString (the conversion method is an
+      // instance method). This means it is not as clear whether it is correct to return a new
+      // object each time. The caller needs to be inspected by hand to see whether it (incorrectly)
+      // stores the object for later use.
+      // See also b/19548084 for a possible rewrite and bringing it in line with IntegralToString.
+      if (shadow_frame->GetLink()->GetLink() != nullptr) {
+        std::string caller2(PrettyMethod(shadow_frame->GetLink()->GetLink()->GetMethod()));
+        if (caller2 == "java.lang.String java.lang.Double.toString(double)") {
+          // Allocate new object.
+          mirror::Class* real_to_string_class =
+              shadow_frame->GetLink()->GetMethod()->GetDeclaringClass();
+          mirror::Object* real_to_string_obj = real_to_string_class->AllocObject(self);
+          if (real_to_string_obj != nullptr) {
+            mirror::ArtMethod* init_method =
+                real_to_string_class->FindDirectMethod("<init>", "()V");
+            if (init_method == nullptr) {
+              real_to_string_class->DumpClass(LOG(FATAL), mirror::Class::kDumpClassFullDetail);
+            }
+            JValue invoke_result;
+            // One arg, this.
+            uint32_t args = static_cast<uint32_t>(reinterpret_cast<uintptr_t>(real_to_string_obj));
+            init_method->Invoke(self, &args, 4, &invoke_result, init_method->GetShorty());
+            if (!self->IsExceptionPending()) {
+              result->SetL(real_to_string_obj);
+              ok = true;
+            }
+          }
+
+          if (!ok) {
+            // We'll abort, so clear exception.
+            self->ClearException();
+          }
+        }
+      }
+    }
+
+    if (!ok) {
       self->ThrowNewException(self->GetCurrentLocationForThrow(), "Ljava/lang/InternalError;",
                               "Unimplemented ThreadLocal.get");
     }
diff --git a/runtime/interpreter/interpreter_goto_table_impl.cc b/runtime/interpreter/interpreter_goto_table_impl.cc
index e4b3247..37324ea 100644
--- a/runtime/interpreter/interpreter_goto_table_impl.cc
+++ b/runtime/interpreter/interpreter_goto_table_impl.cc
@@ -54,6 +54,12 @@
 #define UPDATE_HANDLER_TABLE() \
   currentHandlersTable = handlersTable[Runtime::Current()->GetInstrumentation()->GetInterpreterHandlerTable()]
 
+#define BACKWARD_BRANCH_INSTRUMENTATION(offset) \
+  do { \
+    instrumentation::Instrumentation* instrumentation = Runtime::Current()->GetInstrumentation(); \
+    instrumentation->BackwardBranch(self, shadow_frame.GetMethod(), offset); \
+  } while (false)
+
 #define UNREACHABLE_CODE_CHECK()                \
   do {                                          \
     if (kIsDebugBuild) {                        \
@@ -135,7 +141,7 @@
     }
   };
 
-  const bool do_assignability_check = do_access_check;
+  constexpr bool do_assignability_check = do_access_check;
   if (UNLIKELY(!shadow_frame.HasReferenceArray())) {
     LOG(FATAL) << "Invalid shadow frame for interpreter use";
     return JValue();
@@ -608,6 +614,7 @@
   HANDLE_INSTRUCTION_START(GOTO) {
     int8_t offset = inst->VRegA_10t(inst_data);
     if (IsBackwardBranch(offset)) {
+      BACKWARD_BRANCH_INSTRUMENTATION(offset);
       if (UNLIKELY(self->TestAllFlags())) {
         self->CheckSuspend();
         UPDATE_HANDLER_TABLE();
@@ -620,6 +627,7 @@
   HANDLE_INSTRUCTION_START(GOTO_16) {
     int16_t offset = inst->VRegA_20t();
     if (IsBackwardBranch(offset)) {
+      BACKWARD_BRANCH_INSTRUMENTATION(offset);
       if (UNLIKELY(self->TestAllFlags())) {
         self->CheckSuspend();
         UPDATE_HANDLER_TABLE();
@@ -632,6 +640,7 @@
   HANDLE_INSTRUCTION_START(GOTO_32) {
     int32_t offset = inst->VRegA_30t();
     if (IsBackwardBranch(offset)) {
+      BACKWARD_BRANCH_INSTRUMENTATION(offset);
       if (UNLIKELY(self->TestAllFlags())) {
         self->CheckSuspend();
         UPDATE_HANDLER_TABLE();
@@ -644,6 +653,7 @@
   HANDLE_INSTRUCTION_START(PACKED_SWITCH) {
     int32_t offset = DoPackedSwitch(inst, shadow_frame, inst_data);
     if (IsBackwardBranch(offset)) {
+      BACKWARD_BRANCH_INSTRUMENTATION(offset);
       if (UNLIKELY(self->TestAllFlags())) {
         self->CheckSuspend();
         UPDATE_HANDLER_TABLE();
@@ -656,6 +666,7 @@
   HANDLE_INSTRUCTION_START(SPARSE_SWITCH) {
     int32_t offset = DoSparseSwitch(inst, shadow_frame, inst_data);
     if (IsBackwardBranch(offset)) {
+      BACKWARD_BRANCH_INSTRUMENTATION(offset);
       if (UNLIKELY(self->TestAllFlags())) {
         self->CheckSuspend();
         UPDATE_HANDLER_TABLE();
@@ -758,6 +769,7 @@
     if (shadow_frame.GetVReg(inst->VRegA_22t(inst_data)) == shadow_frame.GetVReg(inst->VRegB_22t(inst_data))) {
       int16_t offset = inst->VRegC_22t();
       if (IsBackwardBranch(offset)) {
+        BACKWARD_BRANCH_INSTRUMENTATION(offset);
         if (UNLIKELY(self->TestAllFlags())) {
           self->CheckSuspend();
           UPDATE_HANDLER_TABLE();
@@ -774,6 +786,7 @@
     if (shadow_frame.GetVReg(inst->VRegA_22t(inst_data)) != shadow_frame.GetVReg(inst->VRegB_22t(inst_data))) {
       int16_t offset = inst->VRegC_22t();
       if (IsBackwardBranch(offset)) {
+        BACKWARD_BRANCH_INSTRUMENTATION(offset);
         if (UNLIKELY(self->TestAllFlags())) {
           self->CheckSuspend();
           UPDATE_HANDLER_TABLE();
@@ -790,6 +803,7 @@
     if (shadow_frame.GetVReg(inst->VRegA_22t(inst_data)) < shadow_frame.GetVReg(inst->VRegB_22t(inst_data))) {
       int16_t offset = inst->VRegC_22t();
       if (IsBackwardBranch(offset)) {
+        BACKWARD_BRANCH_INSTRUMENTATION(offset);
         if (UNLIKELY(self->TestAllFlags())) {
           self->CheckSuspend();
           UPDATE_HANDLER_TABLE();
@@ -806,6 +820,7 @@
     if (shadow_frame.GetVReg(inst->VRegA_22t(inst_data)) >= shadow_frame.GetVReg(inst->VRegB_22t(inst_data))) {
       int16_t offset = inst->VRegC_22t();
       if (IsBackwardBranch(offset)) {
+        BACKWARD_BRANCH_INSTRUMENTATION(offset);
         if (UNLIKELY(self->TestAllFlags())) {
           self->CheckSuspend();
           UPDATE_HANDLER_TABLE();
@@ -822,6 +837,7 @@
     if (shadow_frame.GetVReg(inst->VRegA_22t(inst_data)) > shadow_frame.GetVReg(inst->VRegB_22t(inst_data))) {
       int16_t offset = inst->VRegC_22t();
       if (IsBackwardBranch(offset)) {
+        BACKWARD_BRANCH_INSTRUMENTATION(offset);
         if (UNLIKELY(self->TestAllFlags())) {
           self->CheckSuspend();
           UPDATE_HANDLER_TABLE();
@@ -838,6 +854,7 @@
     if (shadow_frame.GetVReg(inst->VRegA_22t(inst_data)) <= shadow_frame.GetVReg(inst->VRegB_22t(inst_data))) {
       int16_t offset = inst->VRegC_22t();
       if (IsBackwardBranch(offset)) {
+        BACKWARD_BRANCH_INSTRUMENTATION(offset);
         if (UNLIKELY(self->TestAllFlags())) {
           self->CheckSuspend();
           UPDATE_HANDLER_TABLE();
@@ -854,6 +871,7 @@
     if (shadow_frame.GetVReg(inst->VRegA_21t(inst_data)) == 0) {
       int16_t offset = inst->VRegB_21t();
       if (IsBackwardBranch(offset)) {
+        BACKWARD_BRANCH_INSTRUMENTATION(offset);
         if (UNLIKELY(self->TestAllFlags())) {
           self->CheckSuspend();
           UPDATE_HANDLER_TABLE();
@@ -870,6 +888,7 @@
     if (shadow_frame.GetVReg(inst->VRegA_21t(inst_data)) != 0) {
       int16_t offset = inst->VRegB_21t();
       if (IsBackwardBranch(offset)) {
+        BACKWARD_BRANCH_INSTRUMENTATION(offset);
         if (UNLIKELY(self->TestAllFlags())) {
           self->CheckSuspend();
           UPDATE_HANDLER_TABLE();
@@ -886,6 +905,7 @@
     if (shadow_frame.GetVReg(inst->VRegA_21t(inst_data)) < 0) {
       int16_t offset = inst->VRegB_21t();
       if (IsBackwardBranch(offset)) {
+        BACKWARD_BRANCH_INSTRUMENTATION(offset);
         if (UNLIKELY(self->TestAllFlags())) {
           self->CheckSuspend();
           UPDATE_HANDLER_TABLE();
@@ -902,6 +922,7 @@
     if (shadow_frame.GetVReg(inst->VRegA_21t(inst_data)) >= 0) {
       int16_t offset = inst->VRegB_21t();
       if (IsBackwardBranch(offset)) {
+        BACKWARD_BRANCH_INSTRUMENTATION(offset);
         if (UNLIKELY(self->TestAllFlags())) {
           self->CheckSuspend();
           UPDATE_HANDLER_TABLE();
@@ -918,6 +939,7 @@
     if (shadow_frame.GetVReg(inst->VRegA_21t(inst_data)) > 0) {
       int16_t offset = inst->VRegB_21t();
       if (IsBackwardBranch(offset)) {
+        BACKWARD_BRANCH_INSTRUMENTATION(offset);
         if (UNLIKELY(self->TestAllFlags())) {
           self->CheckSuspend();
           UPDATE_HANDLER_TABLE();
@@ -934,6 +956,7 @@
     if (shadow_frame.GetVReg(inst->VRegA_21t(inst_data)) <= 0) {
       int16_t offset = inst->VRegB_21t();
       if (IsBackwardBranch(offset)) {
+        BACKWARD_BRANCH_INSTRUMENTATION(offset);
         if (UNLIKELY(self->TestAllFlags())) {
           self->CheckSuspend();
           UPDATE_HANDLER_TABLE();
diff --git a/runtime/java_vm_ext.cc b/runtime/java_vm_ext.cc
index ea7c192..08332d3 100644
--- a/runtime/java_vm_ext.cc
+++ b/runtime/java_vm_ext.cc
@@ -337,7 +337,8 @@
       thread_group = args->group;
     }
 
-    if (!runtime->AttachCurrentThread(thread_name, as_daemon, thread_group, !runtime->IsCompiler())) {
+    if (!runtime->AttachCurrentThread(thread_name, as_daemon, thread_group,
+                                      !runtime->IsAotCompiler())) {
       *p_env = nullptr;
       return JNI_ERR;
     } else {
diff --git a/runtime/jdwp/jdwp_event.cc b/runtime/jdwp/jdwp_event.cc
index b71f6cd..fc08d23 100644
--- a/runtime/jdwp/jdwp_event.cc
+++ b/runtime/jdwp/jdwp_event.cc
@@ -633,7 +633,11 @@
     AcquireJdwpTokenForEvent(threadId);
   }
   EventFinish(pReq);
-  SuspendByPolicy(suspend_policy, thread_self_id);
+  {
+    // Before suspending, we change our state to kSuspended so the debugger sees us as RUNNING.
+    ScopedThreadStateChange stsc(self, kSuspended);
+    SuspendByPolicy(suspend_policy, thread_self_id);
+  }
   self->TransitionFromSuspendedToRunnable();
 }
 
diff --git a/runtime/jit/jit.cc b/runtime/jit/jit.cc
new file mode 100644
index 0000000..539c181
--- /dev/null
+++ b/runtime/jit/jit.cc
@@ -0,0 +1,160 @@
+/*
+ * Copyright 2014 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "jit.h"
+
+#include <dlfcn.h>
+
+#include "entrypoints/runtime_asm_entrypoints.h"
+#include "interpreter/interpreter.h"
+#include "jit_code_cache.h"
+#include "jit_instrumentation.h"
+#include "mirror/art_method-inl.h"
+#include "runtime.h"
+#include "runtime_options.h"
+#include "thread_list.h"
+#include "utils.h"
+
+namespace art {
+namespace jit {
+
+JitOptions* JitOptions::CreateFromRuntimeArguments(const RuntimeArgumentMap& options) {
+  if (!options.GetOrDefault(RuntimeArgumentMap::UseJIT)) {
+    return nullptr;
+  }
+  auto* jit_options = new JitOptions;
+  jit_options->code_cache_capacity_ =
+      options.GetOrDefault(RuntimeArgumentMap::JITCodeCacheCapacity);
+  jit_options->compile_threshold_ =
+      options.GetOrDefault(RuntimeArgumentMap::JITCompileThreshold);
+  return jit_options;
+}
+
+Jit::Jit()
+    : jit_library_handle_(nullptr), jit_compiler_handle_(nullptr), jit_load_(nullptr),
+      jit_compile_method_(nullptr) {
+}
+
+Jit* Jit::Create(JitOptions* options, std::string* error_msg) {
+  std::unique_ptr<Jit> jit(new Jit);
+  if (!jit->LoadCompiler(error_msg)) {
+    return nullptr;
+  }
+  jit->code_cache_.reset(JitCodeCache::Create(options->GetCodeCacheCapacity(), error_msg));
+  if (jit->GetCodeCache() == nullptr) {
+    return nullptr;
+  }
+  LOG(INFO) << "JIT created with code_cache_capacity="
+      << PrettySize(options->GetCodeCacheCapacity())
+      << " compile_threshold=" << options->GetCompileThreshold();
+  return jit.release();
+}
+
+bool Jit::LoadCompiler(std::string* error_msg) {
+  jit_library_handle_ = dlopen(
+      kIsDebugBuild ? "libartd-compiler.so" : "libart-compiler.so", RTLD_NOW);
+  if (jit_library_handle_ == nullptr) {
+    std::ostringstream oss;
+    oss << "JIT could not load libart-compiler.so: " << dlerror();
+    *error_msg = oss.str();
+    return false;
+  }
+  jit_load_ = reinterpret_cast<void* (*)(CompilerCallbacks**)>(
+      dlsym(jit_library_handle_, "jit_load"));
+  if (jit_load_ == nullptr) {
+    dlclose(jit_library_handle_);
+    *error_msg = "JIT couldn't find jit_load entry point";
+    return false;
+  }
+  jit_unload_ = reinterpret_cast<void (*)(void*)>(
+      dlsym(jit_library_handle_, "jit_unload"));
+  if (jit_unload_ == nullptr) {
+    dlclose(jit_library_handle_);
+    *error_msg = "JIT couldn't find jit_unload entry point";
+    return false;
+  }
+  jit_compile_method_ = reinterpret_cast<bool (*)(void*, mirror::ArtMethod*, Thread*)>(
+      dlsym(jit_library_handle_, "jit_compile_method"));
+  if (jit_compile_method_ == nullptr) {
+    dlclose(jit_library_handle_);
+    *error_msg = "JIT couldn't find jit_compile_method entry point";
+    return false;
+  }
+  CompilerCallbacks* callbacks = nullptr;
+  VLOG(jit) << "Calling JitLoad interpreter_only="
+      << Runtime::Current()->GetInstrumentation()->InterpretOnly();
+  jit_compiler_handle_ = (jit_load_)(&callbacks);
+  if (jit_compiler_handle_ == nullptr) {
+    dlclose(jit_library_handle_);
+    *error_msg = "JIT couldn't load compiler";
+    return false;
+  }
+  if (callbacks == nullptr) {
+    dlclose(jit_library_handle_);
+    *error_msg = "JIT compiler callbacks were not set";
+    jit_compiler_handle_ = nullptr;
+    return false;
+  }
+  compiler_callbacks_ = callbacks;
+  return true;
+}
+
+bool Jit::CompileMethod(mirror::ArtMethod* method, Thread* self) {
+  DCHECK(!method->IsRuntimeMethod());
+  const bool result = jit_compile_method_(jit_compiler_handle_, method, self);
+  if (result) {
+    method->SetEntryPointFromInterpreter(artInterpreterToCompiledCodeBridge);
+  }
+  return result;
+}
+
+void Jit::CreateThreadPool() {
+  CHECK(instrumentation_cache_.get() != nullptr);
+  instrumentation_cache_->CreateThreadPool();
+}
+
+void Jit::DeleteThreadPool() {
+  if (instrumentation_cache_.get() != nullptr) {
+    instrumentation_cache_->DeleteThreadPool();
+  }
+}
+
+Jit::~Jit() {
+  DeleteThreadPool();
+  if (jit_compiler_handle_ != nullptr) {
+    jit_unload_(jit_compiler_handle_);
+  }
+  if (jit_library_handle_ != nullptr) {
+    dlclose(jit_library_handle_);
+  }
+}
+
+void Jit::CreateInstrumentationCache(size_t compile_threshold) {
+  CHECK_GT(compile_threshold, 0U);
+  Runtime* const runtime = Runtime::Current();
+  runtime->GetThreadList()->SuspendAll();
+  // Add Jit interpreter instrumentation, tells the interpreter when to notify the jit to compile
+  // something.
+  instrumentation_cache_.reset(new jit::JitInstrumentationCache(compile_threshold));
+  runtime->GetInstrumentation()->AddListener(
+      new jit::JitInstrumentationListener(instrumentation_cache_.get()),
+      instrumentation::Instrumentation::kMethodEntered |
+      instrumentation::Instrumentation::kBackwardBranch);
+  runtime->GetThreadList()->ResumeAll();
+}
+
+}  // namespace jit
+}  // namespace art
diff --git a/runtime/jit/jit.h b/runtime/jit/jit.h
new file mode 100644
index 0000000..b80015f
--- /dev/null
+++ b/runtime/jit/jit.h
@@ -0,0 +1,102 @@
+/*
+ * Copyright 2014 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef ART_RUNTIME_JIT_JIT_H_
+#define ART_RUNTIME_JIT_JIT_H_
+
+#include <unordered_map>
+
+#include "instrumentation.h"
+
+#include "atomic.h"
+#include "base/macros.h"
+#include "base/mutex.h"
+#include "gc_root.h"
+#include "jni.h"
+#include "object_callbacks.h"
+#include "thread_pool.h"
+
+namespace art {
+
+class CompilerCallbacks;
+struct RuntimeArgumentMap;
+
+namespace jit {
+
+class JitCodeCache;
+class JitInstrumentationCache;
+class JitOptions;
+
+class Jit {
+ public:
+  static constexpr bool kStressMode = kIsDebugBuild;
+  static constexpr size_t kDefaultCompileThreshold = kStressMode ? 1 : 1000;
+
+  virtual ~Jit();
+  static Jit* Create(JitOptions* options, std::string* error_msg);
+  bool CompileMethod(mirror::ArtMethod* method, Thread* self)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  void CreateInstrumentationCache(size_t compile_threshold);
+  void CreateThreadPool();
+  CompilerCallbacks* GetCompilerCallbacks() {
+    return compiler_callbacks_;
+  }
+  const JitCodeCache* GetCodeCache() const {
+    return code_cache_.get();
+  }
+  JitCodeCache* GetCodeCache() {
+    return code_cache_.get();
+  }
+  void DeleteThreadPool();
+
+ private:
+  Jit();
+  bool LoadCompiler(std::string* error_msg);
+
+  // JIT compiler
+  void* jit_library_handle_;
+  void* jit_compiler_handle_;
+  void* (*jit_load_)(CompilerCallbacks**);
+  void (*jit_unload_)(void*);
+  bool (*jit_compile_method_)(void*, mirror::ArtMethod*, Thread*);
+
+  std::unique_ptr<jit::JitInstrumentationCache> instrumentation_cache_;
+  std::unique_ptr<jit::JitCodeCache> code_cache_;
+  CompilerCallbacks* compiler_callbacks_;  // Owned by the jit compiler.
+};
+
+class JitOptions {
+ public:
+  static JitOptions* CreateFromRuntimeArguments(const RuntimeArgumentMap& options);
+  size_t GetCompileThreshold() const {
+    return compile_threshold_;
+  }
+  size_t GetCodeCacheCapacity() const {
+    return code_cache_capacity_;
+  }
+
+ private:
+  size_t code_cache_capacity_;
+  size_t compile_threshold_;
+
+  JitOptions() : code_cache_capacity_(0), compile_threshold_(0) {
+  }
+};
+
+}  // namespace jit
+}  // namespace art
+
+#endif  // ART_RUNTIME_JIT_JIT_H_
diff --git a/runtime/jit/jit_code_cache.cc b/runtime/jit/jit_code_cache.cc
new file mode 100644
index 0000000..4ae4d57
--- /dev/null
+++ b/runtime/jit/jit_code_cache.cc
@@ -0,0 +1,121 @@
+/*
+ * Copyright 2014 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "jit_code_cache.h"
+
+#include <sstream>
+
+#include "mem_map.h"
+#include "mirror/art_method-inl.h"
+#include "oat_file-inl.h"
+
+namespace art {
+namespace jit {
+
+JitCodeCache* JitCodeCache::Create(size_t capacity, std::string* error_msg) {
+  CHECK_GT(capacity, 0U);
+  CHECK_LT(capacity, kMaxCapacity);
+  std::string error_str;
+  // Map name specific for android_os_Debug.cpp accounting.
+  MemMap* map = MemMap::MapAnonymous("jit-code-cache", nullptr, capacity,
+                                     PROT_READ | PROT_WRITE | PROT_EXEC, false, false, &error_str);
+  if (map == nullptr) {
+    std::ostringstream oss;
+    oss << "Failed to create read write execute cache: " << error_str << " size=" << capacity;
+    *error_msg = oss.str();
+    return nullptr;
+  }
+  return new JitCodeCache(map);
+}
+
+JitCodeCache::JitCodeCache(MemMap* mem_map)
+    : lock_("Jit code cache", kJitCodeCacheLock), num_methods_(0) {
+  VLOG(jit) << "Created jit code cache size=" << PrettySize(mem_map->Size());
+  mem_map_.reset(mem_map);
+  uint8_t* divider = mem_map->Begin() + RoundUp(mem_map->Size() / 4, kPageSize);
+  // Data cache is 1 / 4 of the map. TODO: Make this variable?
+  // Put data at the start.
+  data_cache_ptr_ = mem_map->Begin();
+  data_cache_end_ = divider;
+  data_cache_begin_ = data_cache_ptr_;
+  mprotect(data_cache_ptr_, data_cache_end_ - data_cache_begin_, PROT_READ | PROT_WRITE);
+  // Code cache after.
+  code_cache_begin_ = divider;
+  code_cache_ptr_ = divider;
+  code_cache_end_ = mem_map->End();
+}
+
+bool JitCodeCache::ContainsMethod(mirror::ArtMethod* method) const {
+  return ContainsCodePtr(method->GetEntryPointFromQuickCompiledCode());
+}
+
+bool JitCodeCache::ContainsCodePtr(const void* ptr) const {
+  return ptr >= code_cache_begin_ && ptr < code_cache_end_;
+}
+
+void JitCodeCache::FlushInstructionCache() {
+  UNIMPLEMENTED(FATAL);
+  // TODO: Investigate if we need to do this.
+  // __clear_cache(reinterpret_cast<char*>(code_cache_begin_), static_cast<int>(CodeCacheSize()));
+}
+
+uint8_t* JitCodeCache::ReserveCode(Thread* self, size_t size) {
+  MutexLock mu(self, lock_);
+  if (size > CodeCacheRemain()) {
+    return nullptr;
+  }
+  code_cache_ptr_ += size;
+  return code_cache_ptr_ - size;
+}
+
+uint8_t* JitCodeCache::AddDataArray(Thread* self, const uint8_t* begin, const uint8_t* end) {
+  MutexLock mu(self, lock_);
+  const size_t size = end - begin;
+  if (size > DataCacheRemain()) {
+    return nullptr;  // Out of space in the data cache.
+  }
+  std::copy(begin, end, data_cache_ptr_);
+  data_cache_ptr_ += size;
+  return data_cache_ptr_ - size;
+}
+
+const void* JitCodeCache::GetCodeFor(mirror::ArtMethod* method) {
+  const void* code = method->GetEntryPointFromQuickCompiledCode();
+  if (ContainsCodePtr(code)) {
+    return code;
+  }
+  MutexLock mu(Thread::Current(), lock_);
+  auto it = method_code_map_.find(method);
+  if (it != method_code_map_.end()) {
+    return it->second;
+  }
+  return nullptr;
+}
+
+void JitCodeCache::SaveCompiledCode(mirror::ArtMethod* method, const void* old_code_ptr) {
+  DCHECK_EQ(method->GetEntryPointFromQuickCompiledCode(), old_code_ptr);
+  DCHECK(ContainsCodePtr(old_code_ptr)) << PrettyMethod(method) << " old_code_ptr="
+      << old_code_ptr;
+  MutexLock mu(Thread::Current(), lock_);
+  auto it = method_code_map_.find(method);
+  if (it != method_code_map_.end()) {
+    return;
+  }
+  method_code_map_.Put(method, old_code_ptr);
+}
+
+}  // namespace jit
+}  // namespace art
diff --git a/runtime/jit/jit_code_cache.h b/runtime/jit/jit_code_cache.h
new file mode 100644
index 0000000..aa8c717
--- /dev/null
+++ b/runtime/jit/jit_code_cache.h
@@ -0,0 +1,118 @@
+/*
+ * Copyright 2014 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef ART_RUNTIME_JIT_JIT_CODE_CACHE_H_
+#define ART_RUNTIME_JIT_JIT_CODE_CACHE_H_
+
+#include "instrumentation.h"
+
+#include "atomic.h"
+#include "base/macros.h"
+#include "base/mutex.h"
+#include "gc_root.h"
+#include "jni.h"
+#include "oat_file.h"
+#include "object_callbacks.h"
+#include "safe_map.h"
+#include "thread_pool.h"
+
+namespace art {
+
+class CompiledMethod;
+class CompilerCallbacks;
+
+namespace mirror {
+class ArtMethod;
+}  // namespcae mirror
+
+namespace jit {
+
+class JitInstrumentationCache;
+
+class JitCodeCache {
+ public:
+  static constexpr size_t kMaxCapacity = 1 * GB;
+  static constexpr size_t kDefaultCapacity = 2 * MB;
+
+  static JitCodeCache* Create(size_t capacity, std::string* error_msg);
+
+  const uint8_t* CodeCachePtr() const {
+    return code_cache_ptr_;
+  }
+  size_t CodeCacheSize() const {
+    return code_cache_ptr_ - code_cache_begin_;
+  }
+  size_t CodeCacheRemain() const {
+    return code_cache_end_ - code_cache_ptr_;
+  }
+  size_t DataCacheSize() const {
+    return data_cache_ptr_ - data_cache_begin_;
+  }
+  size_t DataCacheRemain() const {
+    return data_cache_end_ - data_cache_ptr_;
+  }
+  size_t NumMethods() const {
+    return num_methods_;
+  }
+
+  bool ContainsMethod(mirror::ArtMethod* method) const
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  bool ContainsCodePtr(const void* ptr) const;
+
+  uint8_t* ReserveCode(Thread* self, size_t size) LOCKS_EXCLUDED(lock_);
+
+  uint8_t* AddDataArray(Thread* self, const uint8_t* begin, const uint8_t* end)
+      LOCKS_EXCLUDED(lock_);
+
+  // Get code for a method, returns null if it is not in the jit cache.
+  const void* GetCodeFor(mirror::ArtMethod* method)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) LOCKS_EXCLUDED(lock_);
+
+  void SaveCompiledCode(mirror::ArtMethod* method, const void* old_code_ptr)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) LOCKS_EXCLUDED(lock_);
+
+ private:
+  // Takes ownership of code_mem_map.
+  explicit JitCodeCache(MemMap* code_mem_map);
+  void FlushInstructionCache();
+
+  Mutex lock_;
+  // Mem map which holds code and data. We do this since we need to have 32 bit offsets from method
+  // headers in code cache which point to things in the data cache. If the maps are more than 4GB
+  // apart, having multiple maps wouldn't work.
+  std::unique_ptr<MemMap> mem_map_;
+  // Code cache section.
+  uint8_t* code_cache_ptr_;
+  const uint8_t* code_cache_begin_;
+  const uint8_t* code_cache_end_;
+  // Data cache section.
+  uint8_t* data_cache_ptr_;
+  const uint8_t* data_cache_begin_;
+  const uint8_t* data_cache_end_;
+  size_t num_methods_;
+  // TODO: This relies on methods not moving.
+  // This map holds code for methods if they were deoptimized by the instrumentation stubs. This is
+  // required since we have to implement ClassLinker::GetQuickOatCodeFor for walking stacks.
+  SafeMap<mirror::ArtMethod*, const void*> method_code_map_;
+
+  DISALLOW_COPY_AND_ASSIGN(JitCodeCache);
+};
+
+
+}  // namespace jit
+}  // namespace art
+
+#endif  // ART_RUNTIME_JIT_JIT_CODE_CACHE_H_
diff --git a/runtime/jit/jit_instrumentation.cc b/runtime/jit/jit_instrumentation.cc
new file mode 100644
index 0000000..160e678
--- /dev/null
+++ b/runtime/jit/jit_instrumentation.cc
@@ -0,0 +1,117 @@
+/*
+ * Copyright 2014 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "jit_instrumentation.h"
+
+#include "jit.h"
+#include "jit_code_cache.h"
+#include "mirror/art_method-inl.h"
+#include "scoped_thread_state_change.h"
+
+namespace art {
+namespace jit {
+
+class JitCompileTask : public Task {
+ public:
+  explicit JitCompileTask(mirror::ArtMethod* method, JitInstrumentationCache* cache)
+      : method_(method), cache_(cache) {
+  }
+
+  virtual void Run(Thread* self) OVERRIDE {
+    ScopedObjectAccess soa(self);
+    VLOG(jit) << "JitCompileTask compiling method " << PrettyMethod(method_);
+    if (Runtime::Current()->GetJit()->CompileMethod(method_, self)) {
+      cache_->SignalCompiled(self, method_);
+    } else {
+      VLOG(jit) << "Failed to compile method " << PrettyMethod(method_);
+    }
+  }
+
+  virtual void Finalize() OVERRIDE {
+    delete this;
+  }
+
+ private:
+  mirror::ArtMethod* const method_;
+  JitInstrumentationCache* const cache_;
+};
+
+JitInstrumentationCache::JitInstrumentationCache(size_t hot_method_threshold)
+    : lock_("jit instrumentation lock"), hot_method_threshold_(hot_method_threshold) {
+}
+
+void JitInstrumentationCache::CreateThreadPool() {
+  thread_pool_.reset(new ThreadPool("Jit thread pool", 1));
+}
+
+void JitInstrumentationCache::DeleteThreadPool() {
+  thread_pool_.reset();
+}
+
+void JitInstrumentationCache::SignalCompiled(Thread* self, mirror::ArtMethod* method) {
+  ScopedObjectAccessUnchecked soa(self);
+  jmethodID method_id = soa.EncodeMethod(method);
+  MutexLock mu(self, lock_);
+  auto it = samples_.find(method_id);
+  if (it != samples_.end()) {
+    samples_.erase(it);
+  }
+}
+
+void JitInstrumentationCache::AddSamples(Thread* self, mirror::ArtMethod* method, size_t count) {
+  ScopedObjectAccessUnchecked soa(self);
+  // Since we don't have on-stack replacement, some methods can remain in the interpreter longer
+  // than we want resulting in samples even after the method is compiled.
+  if (method->IsClassInitializer() ||
+      Runtime::Current()->GetJit()->GetCodeCache()->ContainsMethod(method)) {
+    return;
+  }
+  jmethodID method_id = soa.EncodeMethod(method);
+  bool is_hot = false;
+  {
+    MutexLock mu(self, lock_);
+    size_t sample_count = 0;
+    auto it = samples_.find(method_id);
+    if (it != samples_.end()) {
+      it->second += count;
+      sample_count = it->second;
+    } else {
+      sample_count = count;
+      samples_.insert(std::make_pair(method_id, count));
+    }
+    // If we have enough samples, mark as hot and request Jit compilation.
+    if (sample_count >= hot_method_threshold_ && sample_count - count < hot_method_threshold_) {
+      is_hot = true;
+    }
+  }
+  if (is_hot) {
+    if (thread_pool_.get() != nullptr) {
+      thread_pool_->AddTask(self, new JitCompileTask(method->GetInterfaceMethodIfProxy(), this));
+      thread_pool_->StartWorkers(self);
+    } else {
+      VLOG(jit) << "Compiling hot method " << PrettyMethod(method);
+      Runtime::Current()->GetJit()->CompileMethod(method->GetInterfaceMethodIfProxy(), self);
+    }
+  }
+}
+
+JitInstrumentationListener::JitInstrumentationListener(JitInstrumentationCache* cache)
+    : instrumentation_cache_(cache) {
+  CHECK(instrumentation_cache_ != nullptr);
+}
+
+}  // namespace jit
+}  // namespace art
diff --git a/runtime/jit/jit_instrumentation.h b/runtime/jit/jit_instrumentation.h
new file mode 100644
index 0000000..9576f4b
--- /dev/null
+++ b/runtime/jit/jit_instrumentation.h
@@ -0,0 +1,107 @@
+/*
+ * Copyright 2014 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef ART_RUNTIME_JIT_JIT_INSTRUMENTATION_H_
+#define ART_RUNTIME_JIT_JIT_INSTRUMENTATION_H_
+
+#include <unordered_map>
+
+#include "instrumentation.h"
+
+#include "atomic.h"
+#include "base/macros.h"
+#include "base/mutex.h"
+#include "gc_root.h"
+#include "jni.h"
+#include "object_callbacks.h"
+#include "thread_pool.h"
+
+namespace art {
+namespace mirror {
+  class ArtField;
+  class ArtMethod;
+  class Class;
+  class Object;
+  class Throwable;
+}  // namespace mirror
+union JValue;
+class Thread;
+class ThrowLocation;
+
+namespace jit {
+
+// Keeps track of which methods are hot.
+class JitInstrumentationCache {
+ public:
+  explicit JitInstrumentationCache(size_t hot_method_threshold);
+  void AddSamples(Thread* self, mirror::ArtMethod* method, size_t samples)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  void SignalCompiled(Thread* self, mirror::ArtMethod* method)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  void CreateThreadPool();
+  void DeleteThreadPool();
+
+ private:
+  Mutex lock_;
+  std::unordered_map<jmethodID, size_t> samples_;
+  size_t hot_method_threshold_;
+  std::unique_ptr<ThreadPool> thread_pool_;
+};
+
+class JitInstrumentationListener : public instrumentation::InstrumentationListener {
+ public:
+  explicit JitInstrumentationListener(JitInstrumentationCache* cache);
+
+  virtual void MethodEntered(Thread* thread, mirror::Object* /*this_object*/,
+                             mirror::ArtMethod* method, uint32_t /*dex_pc*/)
+      OVERRIDE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+    instrumentation_cache_->AddSamples(thread, method, 1);
+  }
+  virtual void MethodExited(Thread* /*thread*/, mirror::Object* /*this_object*/,
+                            mirror::ArtMethod* /*method*/, uint32_t /*dex_pc*/,
+                            const JValue& /*return_value*/)
+      OVERRIDE { }
+  virtual void MethodUnwind(Thread* /*thread*/, mirror::Object* /*this_object*/,
+                            mirror::ArtMethod* /*method*/, uint32_t /*dex_pc*/) OVERRIDE { }
+  virtual void FieldRead(Thread* /*thread*/, mirror::Object* /*this_object*/,
+                         mirror::ArtMethod* /*method*/, uint32_t /*dex_pc*/,
+                         mirror::ArtField* /*field*/) OVERRIDE { }
+  virtual void FieldWritten(Thread* /*thread*/, mirror::Object* /*this_object*/,
+                            mirror::ArtMethod* /*method*/, uint32_t /*dex_pc*/,
+                            mirror::ArtField* /*field*/, const JValue& /*field_value*/)
+      OVERRIDE { }
+  virtual void ExceptionCaught(Thread* /*thread*/, const ThrowLocation& /*throw_location*/,
+                               mirror::ArtMethod* /*catch_method*/, uint32_t /*catch_dex_pc*/,
+                               mirror::Throwable* /*exception_object*/) OVERRIDE { }
+
+  virtual void DexPcMoved(Thread* /*self*/, mirror::Object* /*this_object*/,
+                          mirror::ArtMethod* /*method*/, uint32_t /*new_dex_pc*/) OVERRIDE { }
+
+  // We only care about how many dex instructions were executed in the Jit.
+  virtual void BackwardBranch(Thread* thread, mirror::ArtMethod* method, int32_t dex_pc_offset)
+      OVERRIDE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+    CHECK_LE(dex_pc_offset, 0);
+    instrumentation_cache_->AddSamples(thread, method, 1);
+  }
+
+ private:
+  JitInstrumentationCache* const instrumentation_cache_;
+};
+
+}  // namespace jit
+}  // namespace art
+
+#endif  // ART_RUNTIME_JIT_JIT_INSTRUMENTATION_H_
diff --git a/runtime/mem_map.cc b/runtime/mem_map.cc
index a722813..588615f 100644
--- a/runtime/mem_map.cc
+++ b/runtime/mem_map.cc
@@ -138,9 +138,10 @@
 #endif
 
 // Return true if the address range is contained in a single /proc/self/map entry.
-static bool ContainedWithinExistingMap(uintptr_t begin,
-                                       uintptr_t end,
+static bool ContainedWithinExistingMap(uint8_t* ptr, size_t size,
                                        std::string* error_msg) {
+  uintptr_t begin = reinterpret_cast<uintptr_t>(ptr);
+  uintptr_t end = begin + size;
   std::unique_ptr<BacktraceMap> map(BacktraceMap::Create(getpid(), true));
   if (map.get() == nullptr) {
     *error_msg = StringPrintf("Failed to build process map");
@@ -152,11 +153,9 @@
       return true;
     }
   }
-  std::string maps;
-  ReadFileToString("/proc/self/maps", &maps);
+  PrintFileToLog("/proc/self/maps", LogSeverity::WARNING);
   *error_msg = StringPrintf("Requested region 0x%08" PRIxPTR "-0x%08" PRIxPTR " does not overlap "
-                            "any existing map:\n%s\n",
-                            begin, end, maps.c_str());
+                            "any existing map. See process maps in the log.", begin, end);
   return false;
 }
 
@@ -242,7 +241,7 @@
 }
 
 MemMap* MemMap::MapAnonymous(const char* name, uint8_t* expected_ptr, size_t byte_count, int prot,
-                             bool low_4gb, std::string* error_msg) {
+                             bool low_4gb, bool reuse, std::string* error_msg) {
 #ifndef __LP64__
   UNUSED(low_4gb);
 #endif
@@ -252,6 +251,15 @@
   size_t page_aligned_byte_count = RoundUp(byte_count, kPageSize);
 
   int flags = MAP_PRIVATE | MAP_ANONYMOUS;
+  if (reuse) {
+    // reuse means it is okay that it overlaps an existing page mapping.
+    // Only use this if you actually made the page reservation yourself.
+    CHECK(expected_ptr != nullptr);
+
+    DCHECK(ContainedWithinExistingMap(expected_ptr, byte_count, error_msg)) << error_msg;
+    flags |= MAP_FIXED;
+  }
+
   ScopedFd fd(-1);
 
 #ifdef USE_ASHMEM
@@ -275,7 +283,7 @@
       *error_msg = StringPrintf("ashmem_create_region failed for '%s': %s", name, strerror(errno));
       return nullptr;
     }
-    flags = MAP_PRIVATE;
+    flags &= ~MAP_ANONYMOUS;
   }
 #endif
 
@@ -375,12 +383,11 @@
 #endif
 
   if (actual == MAP_FAILED) {
-    std::string maps;
-    ReadFileToString("/proc/self/maps", &maps);
+    PrintFileToLog("/proc/self/maps", LogSeverity::WARNING);
 
-    *error_msg = StringPrintf("Failed anonymous mmap(%p, %zd, 0x%x, 0x%x, %d, 0): %s\n%s",
-                              expected_ptr, page_aligned_byte_count, prot, flags, fd.get(),
-                              strerror(saved_errno), maps.c_str());
+    *error_msg = StringPrintf("Failed anonymous mmap(%p, %zd, 0x%x, 0x%x, %d, 0): %s. See process "
+                              "maps in the log.", expected_ptr, page_aligned_byte_count, prot,
+                              flags, fd.get(), strerror(saved_errno));
     return nullptr;
   }
   std::ostringstream check_map_request_error_msg;
@@ -396,8 +403,6 @@
                                  std::string* error_msg) {
   CHECK_NE(0, prot);
   CHECK_NE(0, flags & (MAP_SHARED | MAP_PRIVATE));
-  uintptr_t expected = reinterpret_cast<uintptr_t>(expected_ptr);
-  uintptr_t limit = expected + byte_count;
 
   // Note that we do not allow MAP_FIXED unless reuse == true, i.e we
   // expect his mapping to be contained within an existing map.
@@ -406,7 +411,7 @@
     // Only use this if you actually made the page reservation yourself.
     CHECK(expected_ptr != nullptr);
 
-    DCHECK(ContainedWithinExistingMap(expected, limit, error_msg));
+    DCHECK(ContainedWithinExistingMap(expected_ptr, byte_count, error_msg)) << error_msg;
     flags |= MAP_FIXED;
   } else {
     CHECK_EQ(0, flags & MAP_FIXED);
@@ -435,14 +440,13 @@
   if (actual == MAP_FAILED) {
     auto saved_errno = errno;
 
-    std::string maps;
-    ReadFileToString("/proc/self/maps", &maps);
+    PrintFileToLog("/proc/self/maps", LogSeverity::WARNING);
 
     *error_msg = StringPrintf("mmap(%p, %zd, 0x%x, 0x%x, %d, %" PRId64
-                              ") of file '%s' failed: %s\n%s",
+                              ") of file '%s' failed: %s. See process maps in the log.",
                               page_aligned_expected, page_aligned_byte_count, prot, flags, fd,
                               static_cast<int64_t>(page_aligned_offset), filename,
-                              strerror(saved_errno), maps.c_str());
+                              strerror(saved_errno));
     return nullptr;
   }
   std::ostringstream check_map_request_error_msg;
@@ -544,11 +548,9 @@
   // Unmap/map the tail region.
   int result = munmap(tail_base_begin, tail_base_size);
   if (result == -1) {
-    std::string maps;
-    ReadFileToString("/proc/self/maps", &maps);
-    *error_msg = StringPrintf("munmap(%p, %zd) failed for '%s'\n%s",
-                              tail_base_begin, tail_base_size, name_.c_str(),
-                              maps.c_str());
+    PrintFileToLog("/proc/self/maps", LogSeverity::WARNING);
+    *error_msg = StringPrintf("munmap(%p, %zd) failed for '%s'. See process maps in the log.",
+                              tail_base_begin, tail_base_size, name_.c_str());
     return nullptr;
   }
   // Don't cause memory allocation between the munmap and the mmap
@@ -558,11 +560,10 @@
   uint8_t* actual = reinterpret_cast<uint8_t*>(mmap(tail_base_begin, tail_base_size, tail_prot,
                                               flags, fd.get(), 0));
   if (actual == MAP_FAILED) {
-    std::string maps;
-    ReadFileToString("/proc/self/maps", &maps);
-    *error_msg = StringPrintf("anonymous mmap(%p, %zd, 0x%x, 0x%x, %d, 0) failed\n%s",
-                              tail_base_begin, tail_base_size, tail_prot, flags, fd.get(),
-                              maps.c_str());
+    PrintFileToLog("/proc/self/maps", LogSeverity::WARNING);
+    *error_msg = StringPrintf("anonymous mmap(%p, %zd, 0x%x, 0x%x, %d, 0) failed. See process "
+                              "maps in the log.", tail_base_begin, tail_base_size, tail_prot, flags,
+                              fd.get());
     return nullptr;
   }
   return new MemMap(tail_name, actual, tail_size, actual, tail_base_size, tail_prot, false);
diff --git a/runtime/mem_map.h b/runtime/mem_map.h
index dc337e0..11b2569 100644
--- a/runtime/mem_map.h
+++ b/runtime/mem_map.h
@@ -54,6 +54,7 @@
  public:
   // Request an anonymous region of length 'byte_count' and a requested base address.
   // Use NULL as the requested base address if you don't care.
+  // "reuse" allows re-mapping an address range from an existing mapping.
   //
   // The word "anonymous" in this context means "not backed by a file". The supplied
   // 'ashmem_name' will be used -- on systems that support it -- to give the mapping
@@ -61,7 +62,7 @@
   //
   // On success, returns returns a MemMap instance.  On failure, returns a NULL;
   static MemMap* MapAnonymous(const char* ashmem_name, uint8_t* addr, size_t byte_count, int prot,
-                              bool low_4gb, std::string* error_msg);
+                              bool low_4gb, bool reuse, std::string* error_msg);
 
   // Map part of a file, taking care of non-page aligned offsets.  The
   // "start" offset is absolute, not relative.
diff --git a/runtime/mem_map_test.cc b/runtime/mem_map_test.cc
index 14a72b9..f635b5d 100644
--- a/runtime/mem_map_test.cc
+++ b/runtime/mem_map_test.cc
@@ -43,6 +43,7 @@
                                       2 * page_size,
                                       PROT_READ | PROT_WRITE,
                                       low_4gb,
+                                      false,
                                       &error_msg);
     // Check its state and write to it.
     uint8_t* base0 = m0->Begin();
@@ -129,11 +130,12 @@
   CommonInit();
   std::string error_msg;
   std::unique_ptr<MemMap> map(MemMap::MapAnonymous("MapAnonymousEmpty",
-                                             nullptr,
-                                             0,
-                                             PROT_READ,
-                                             false,
-                                             &error_msg));
+                                                   nullptr,
+                                                   0,
+                                                   PROT_READ,
+                                                   false,
+                                                   false,
+                                                   &error_msg));
   ASSERT_TRUE(map.get() != nullptr) << error_msg;
   ASSERT_TRUE(error_msg.empty());
   map.reset(MemMap::MapAnonymous("MapAnonymousEmpty",
@@ -141,6 +143,7 @@
                                  kPageSize,
                                  PROT_READ | PROT_WRITE,
                                  false,
+                                 false,
                                  &error_msg));
   ASSERT_TRUE(map.get() != nullptr) << error_msg;
   ASSERT_TRUE(error_msg.empty());
@@ -151,11 +154,12 @@
   CommonInit();
   std::string error_msg;
   std::unique_ptr<MemMap> map(MemMap::MapAnonymous("MapAnonymousEmpty",
-                                             nullptr,
-                                             kPageSize,
-                                             PROT_READ | PROT_WRITE,
-                                             true,
-                                             &error_msg));
+                                                   nullptr,
+                                                   kPageSize,
+                                                   PROT_READ | PROT_WRITE,
+                                                   true,
+                                                   false,
+                                                   &error_msg));
   ASSERT_TRUE(map.get() != nullptr) << error_msg;
   ASSERT_TRUE(error_msg.empty());
   ASSERT_LT(reinterpret_cast<uintptr_t>(BaseBegin(map.get())), 1ULL << 32);
@@ -167,31 +171,34 @@
   std::string error_msg;
   // Map at an address that should work, which should succeed.
   std::unique_ptr<MemMap> map0(MemMap::MapAnonymous("MapAnonymous0",
-                                              reinterpret_cast<uint8_t*>(ART_BASE_ADDRESS),
-                                              kPageSize,
-                                              PROT_READ | PROT_WRITE,
-                                              false,
-                                              &error_msg));
+                                                    reinterpret_cast<uint8_t*>(ART_BASE_ADDRESS),
+                                                    kPageSize,
+                                                    PROT_READ | PROT_WRITE,
+                                                    false,
+                                                    false,
+                                                    &error_msg));
   ASSERT_TRUE(map0.get() != nullptr) << error_msg;
   ASSERT_TRUE(error_msg.empty());
   ASSERT_TRUE(map0->BaseBegin() == reinterpret_cast<void*>(ART_BASE_ADDRESS));
   // Map at an unspecified address, which should succeed.
   std::unique_ptr<MemMap> map1(MemMap::MapAnonymous("MapAnonymous1",
-                                              nullptr,
-                                              kPageSize,
-                                              PROT_READ | PROT_WRITE,
-                                              false,
-                                              &error_msg));
+                                                    nullptr,
+                                                    kPageSize,
+                                                    PROT_READ | PROT_WRITE,
+                                                    false,
+                                                    false,
+                                                    &error_msg));
   ASSERT_TRUE(map1.get() != nullptr) << error_msg;
   ASSERT_TRUE(error_msg.empty());
   ASSERT_TRUE(map1->BaseBegin() != nullptr);
   // Attempt to map at the same address, which should fail.
   std::unique_ptr<MemMap> map2(MemMap::MapAnonymous("MapAnonymous2",
-                                              reinterpret_cast<uint8_t*>(map1->BaseBegin()),
-                                              kPageSize,
-                                              PROT_READ | PROT_WRITE,
-                                              false,
-                                              &error_msg));
+                                                    reinterpret_cast<uint8_t*>(map1->BaseBegin()),
+                                                    kPageSize,
+                                                    PROT_READ | PROT_WRITE,
+                                                    false,
+                                                    false,
+                                                    &error_msg));
   ASSERT_TRUE(map2.get() == nullptr) << error_msg;
   ASSERT_TRUE(!error_msg.empty());
 }
@@ -217,6 +224,7 @@
                                                      0x21000000,
                                                      PROT_READ | PROT_WRITE,
                                                      true,
+                                                     false,
                                                      &error_msg));
     ASSERT_TRUE(map.get() != nullptr) << error_msg;
     ASSERT_TRUE(error_msg.empty());
@@ -230,11 +238,12 @@
   uintptr_t ptr = 0;
   ptr -= kPageSize;  // Now it's close to the top.
   std::unique_ptr<MemMap> map(MemMap::MapAnonymous("MapAnonymousOverflow",
-                                             reinterpret_cast<uint8_t*>(ptr),
-                                             2 * kPageSize,  // brings it over the top.
-                                             PROT_READ | PROT_WRITE,
-                                             false,
-                                             &error_msg));
+                                                   reinterpret_cast<uint8_t*>(ptr),
+                                                   2 * kPageSize,  // brings it over the top.
+                                                   PROT_READ | PROT_WRITE,
+                                                   false,
+                                                   false,
+                                                   &error_msg));
   ASSERT_EQ(nullptr, map.get());
   ASSERT_FALSE(error_msg.empty());
 }
@@ -243,12 +252,14 @@
 TEST_F(MemMapTest, MapAnonymousLow4GBExpectedTooHigh) {
   CommonInit();
   std::string error_msg;
-  std::unique_ptr<MemMap> map(MemMap::MapAnonymous("MapAnonymousLow4GBExpectedTooHigh",
-                                             reinterpret_cast<uint8_t*>(UINT64_C(0x100000000)),
-                                             kPageSize,
-                                             PROT_READ | PROT_WRITE,
-                                             true,
-                                             &error_msg));
+  std::unique_ptr<MemMap> map(
+      MemMap::MapAnonymous("MapAnonymousLow4GBExpectedTooHigh",
+                           reinterpret_cast<uint8_t*>(UINT64_C(0x100000000)),
+                           kPageSize,
+                           PROT_READ | PROT_WRITE,
+                           true,
+                           false,
+                           &error_msg));
   ASSERT_EQ(nullptr, map.get());
   ASSERT_FALSE(error_msg.empty());
 }
@@ -257,16 +268,40 @@
   CommonInit();
   std::string error_msg;
   std::unique_ptr<MemMap> map(MemMap::MapAnonymous("MapAnonymousLow4GBRangeTooHigh",
-                                             reinterpret_cast<uint8_t*>(0xF0000000),
-                                             0x20000000,
-                                             PROT_READ | PROT_WRITE,
-                                             true,
-                                             &error_msg));
+                                                   reinterpret_cast<uint8_t*>(0xF0000000),
+                                                   0x20000000,
+                                                   PROT_READ | PROT_WRITE,
+                                                   true,
+                                                   false,
+                                                   &error_msg));
   ASSERT_EQ(nullptr, map.get());
   ASSERT_FALSE(error_msg.empty());
 }
 #endif
 
+TEST_F(MemMapTest, MapAnonymousReuse) {
+  CommonInit();
+  std::string error_msg;
+  std::unique_ptr<MemMap> map(MemMap::MapAnonymous("MapAnonymousReserve",
+                                                   nullptr,
+                                                   0x20000,
+                                                   PROT_READ | PROT_WRITE,
+                                                   false,
+                                                   false,
+                                                   &error_msg));
+  ASSERT_NE(nullptr, map.get());
+  ASSERT_TRUE(error_msg.empty());
+  std::unique_ptr<MemMap> map2(MemMap::MapAnonymous("MapAnonymousReused",
+                                                    reinterpret_cast<uint8_t*>(map->BaseBegin()),
+                                                    0x10000,
+                                                    PROT_READ | PROT_WRITE,
+                                                    false,
+                                                    true,
+                                                    &error_msg));
+  ASSERT_NE(nullptr, map2.get());
+  ASSERT_TRUE(error_msg.empty());
+}
+
 TEST_F(MemMapTest, CheckNoGaps) {
   CommonInit();
   std::string error_msg;
@@ -277,6 +312,7 @@
                                                    kPageSize * kNumPages,
                                                    PROT_READ | PROT_WRITE,
                                                    false,
+                                                   false,
                                                    &error_msg));
   ASSERT_TRUE(map.get() != nullptr) << error_msg;
   ASSERT_TRUE(error_msg.empty());
@@ -292,6 +328,7 @@
                                                     kPageSize,
                                                     PROT_READ | PROT_WRITE,
                                                     false,
+                                                    false,
                                                     &error_msg));
   ASSERT_TRUE(map0.get() != nullptr) << error_msg;
   ASSERT_TRUE(error_msg.empty());
@@ -300,6 +337,7 @@
                                                     kPageSize,
                                                     PROT_READ | PROT_WRITE,
                                                     false,
+                                                    false,
                                                     &error_msg));
   ASSERT_TRUE(map1.get() != nullptr) << error_msg;
   ASSERT_TRUE(error_msg.empty());
@@ -308,6 +346,7 @@
                                                     kPageSize,
                                                     PROT_READ | PROT_WRITE,
                                                     false,
+                                                    false,
                                                     &error_msg));
   ASSERT_TRUE(map2.get() != nullptr) << error_msg;
   ASSERT_TRUE(error_msg.empty());
diff --git a/runtime/mirror/art_field.cc b/runtime/mirror/art_field.cc
index 5a4ebd1..3cea4a1 100644
--- a/runtime/mirror/art_field.cc
+++ b/runtime/mirror/art_field.cc
@@ -44,7 +44,7 @@
 
 void ArtField::SetOffset(MemberOffset num_bytes) {
   DCHECK(GetDeclaringClass()->IsLoaded() || GetDeclaringClass()->IsErroneous());
-  if (kIsDebugBuild && Runtime::Current()->IsCompiler() &&
+  if (kIsDebugBuild && Runtime::Current()->IsAotCompiler() &&
       !Runtime::Current()->UseCompileTimeClassPath()) {
     Primitive::Type type = GetTypeAsPrimitiveType();
     if (type == Primitive::kPrimDouble || type == Primitive::kPrimLong) {
diff --git a/runtime/mirror/art_method-inl.h b/runtime/mirror/art_method-inl.h
index 7d31148..c27c6e9 100644
--- a/runtime/mirror/art_method-inl.h
+++ b/runtime/mirror/art_method-inl.h
@@ -147,7 +147,10 @@
 
 inline uint32_t ArtMethod::GetCodeSize() {
   DCHECK(!IsRuntimeMethod() && !IsProxyMethod()) << PrettyMethod(this);
-  const void* code = EntryPointToCodePointer(GetEntryPointFromQuickCompiledCode());
+  return GetCodeSize(EntryPointToCodePointer(GetEntryPointFromQuickCompiledCode()));
+}
+
+inline uint32_t ArtMethod::GetCodeSize(const void* code) {
   if (code == nullptr) {
     return 0u;
   }
diff --git a/runtime/mirror/art_method.cc b/runtime/mirror/art_method.cc
index b2016dc..26f6f34 100644
--- a/runtime/mirror/art_method.cc
+++ b/runtime/mirror/art_method.cc
@@ -27,6 +27,8 @@
 #include "entrypoints/runtime_asm_entrypoints.h"
 #include "gc/accounting/card_table-inl.h"
 #include "interpreter/interpreter.h"
+#include "jit/jit.h"
+#include "jit/jit_code_cache.h"
 #include "jni_internal.h"
 #include "mapping_table.h"
 #include "object_array-inl.h"
@@ -229,6 +231,7 @@
   if (abort_on_failure) {
       LOG(FATAL) << "Failed to find Dex offset for PC offset " << reinterpret_cast<void*>(sought_offset)
              << "(PC " << reinterpret_cast<void*>(pc) << ", entry_point=" << entry_point
+             << " current entry_point=" << GetQuickOatEntryPoint(sizeof(void*))
              << ") in " << PrettyMethod(this);
   }
   return DexFile::kDexNoIndex;
@@ -329,6 +332,13 @@
       class_linker->IsQuickResolutionStub(code)) {
     return;
   }
+  // 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()->ContainsCodePtr(reinterpret_cast<const void*>(code))) {
+    return;
+  }
   /*
    * During a stack walk, a return PC may point past-the-end of the code
    * in the case that the last instruction is a call that isn't expected to
@@ -336,11 +346,12 @@
    *
    * NOTE: For Thumb both pc and code are offset by 1 indicating the Thumb state.
    */
-  CHECK(PcIsWithinQuickCode(pc))
+  CHECK(PcIsWithinQuickCode(reinterpret_cast<uintptr_t>(code), pc))
       << PrettyMethod(this)
       << " pc=" << std::hex << pc
       << " code=" << code
-      << " size=" << GetCodeSize();
+      << " size=" << GetCodeSize(
+          EntryPointToCodePointer(reinterpret_cast<const void*>(code)));
 }
 
 bool ArtMethod::IsEntrypointInterpreter() {
@@ -410,7 +421,8 @@
       }
 
       // Ensure that we won't be accidentally calling quick compiled code when -Xint.
-      if (kIsDebugBuild && Runtime::Current()->GetInstrumentation()->IsForcedInterpretOnly()) {
+      if (kIsDebugBuild && runtime->GetInstrumentation()->IsForcedInterpretOnly()) {
+        DCHECK(!runtime->UseJit());
         CHECK(IsEntrypointInterpreter())
             << "Don't call compiled code when -Xint " << PrettyMethod(this);
       }
diff --git a/runtime/mirror/art_method.h b/runtime/mirror/art_method.h
index f33ca94..d878f25 100644
--- a/runtime/mirror/art_method.h
+++ b/runtime/mirror/art_method.h
@@ -305,18 +305,8 @@
   // quick entrypoint. This code isn't robust for instrumentation, etc. and is only used for
   // debug purposes.
   bool PcIsWithinQuickCode(uintptr_t pc) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    uintptr_t code = reinterpret_cast<uintptr_t>(GetEntryPointFromQuickCompiledCode());
-    if (code == 0) {
-      return pc == 0;
-    }
-    /*
-     * During a stack walk, a return PC may point past-the-end of the code
-     * in the case that the last instruction is a call that isn't expected to
-     * return.  Thus, we check <= code + GetCodeSize().
-     *
-     * NOTE: For Thumb both pc and code are offset by 1 indicating the Thumb state.
-     */
-    return code <= pc && pc <= code + GetCodeSize();
+    return PcIsWithinQuickCode(
+        reinterpret_cast<uintptr_t>(GetEntryPointFromQuickCompiledCode()), pc);
   }
 
   void AssertPcIsWithinQuickCode(uintptr_t pc) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
@@ -622,6 +612,24 @@
     return offset;
   }
 
+  // Code points to the start of the quick code.
+  static uint32_t GetCodeSize(const void* code);
+
+  static bool PcIsWithinQuickCode(uintptr_t code, uintptr_t pc) {
+    if (code == 0) {
+      return pc == 0;
+    }
+    /*
+     * During a stack walk, a return PC may point past-the-end of the code
+     * in the case that the last instruction is a call that isn't expected to
+     * return.  Thus, we check <= code + GetCodeSize().
+     *
+     * NOTE: For Thumb both pc and code are offset by 1 indicating the Thumb state.
+     */
+    return code <= pc && pc <= code + GetCodeSize(
+        EntryPointToCodePointer(reinterpret_cast<const void*>(code)));
+  }
+
   friend struct art::ArtMethodOffsets;  // for verifying offset information
   DISALLOW_IMPLICIT_CONSTRUCTORS(ArtMethod);
 };
diff --git a/runtime/native/dalvik_system_DexFile.cc b/runtime/native/dalvik_system_DexFile.cc
index 037072d..e1fe3eb 100644
--- a/runtime/native/dalvik_system_DexFile.cc
+++ b/runtime/native/dalvik_system_DexFile.cc
@@ -58,6 +58,70 @@
 
 namespace art {
 
+static std::unique_ptr<std::vector<const DexFile*>>
+ConvertJavaArrayToNative(JNIEnv* env, jobject arrayObject) {
+  jarray array = reinterpret_cast<jarray>(arrayObject);
+
+  jsize array_size = env->GetArrayLength(array);
+  if (env->ExceptionCheck() == JNI_TRUE) {
+    return std::unique_ptr<std::vector<const DexFile*>>();
+  }
+
+  // TODO: Optimize. On 32bit we can use an int array.
+  jboolean is_long_data_copied;
+  jlong* long_data = env->GetLongArrayElements(reinterpret_cast<jlongArray>(array),
+                                               &is_long_data_copied);
+  if (env->ExceptionCheck() == JNI_TRUE) {
+    return std::unique_ptr<std::vector<const DexFile*>>();
+  }
+
+  std::unique_ptr<std::vector<const DexFile*>> ret(new std::vector<const DexFile*>());
+  ret->reserve(array_size);
+  for (jsize i = 0; i < array_size; ++i) {
+    ret->push_back(reinterpret_cast<const DexFile*>(static_cast<uintptr_t>(*(long_data + i))));
+  }
+
+  env->ReleaseLongArrayElements(reinterpret_cast<jlongArray>(array), long_data, JNI_ABORT);
+  if (env->ExceptionCheck() == JNI_TRUE) {
+    return std::unique_ptr<std::vector<const DexFile*>>();
+  }
+
+  return ret;
+}
+
+static jlongArray ConvertNativeToJavaArray(JNIEnv* env,
+                                           std::vector<std::unique_ptr<const DexFile>>& vec) {
+  size_t vec_size = vec.size();
+  jlongArray long_array = env->NewLongArray(static_cast<jsize>(vec_size));
+  if (env->ExceptionCheck() == JNI_TRUE) {
+    return nullptr;
+  }
+
+  jboolean is_long_data_copied;
+  jlong* long_data = env->GetLongArrayElements(long_array, &is_long_data_copied);
+  if (env->ExceptionCheck() == JNI_TRUE) {
+    return nullptr;
+  }
+
+  jlong* tmp = long_data;
+  for (auto& dex_file : vec) {
+    *tmp = reinterpret_cast<uintptr_t>(dex_file.get());
+    tmp++;
+  }
+
+  env->ReleaseLongArrayElements(long_array, long_data, 0);
+  if (env->ExceptionCheck() == JNI_TRUE) {
+    return nullptr;
+  }
+
+  // Now release all the unique_ptrs.
+  for (auto& dex_file : vec) {
+    dex_file.release();
+  }
+
+  return long_array;
+}
+
 // A smart pointer that provides read-only access to a Java string's UTF chars.
 // Unlike libcore's NullableScopedUtfChars, this will *not* throw NullPointerException if
 // passed a null jstring. The correct idiom is:
@@ -104,7 +168,7 @@
   void operator=(const NullableScopedUtfChars&);
 };
 
-static jlong DexFile_openDexFileNative(JNIEnv* env, jclass, jstring javaSourceName, jstring javaOutputName, jint) {
+static jobject DexFile_openDexFileNative(JNIEnv* env, jclass, jstring javaSourceName, jstring javaOutputName, jint) {
   ScopedUtfChars sourceName(env, javaSourceName);
   if (sourceName.c_str() == NULL) {
     return 0;
@@ -115,20 +179,26 @@
   }
 
   ClassLinker* linker = Runtime::Current()->GetClassLinker();
-  std::unique_ptr<std::vector<std::unique_ptr<const DexFile>>> dex_files(
-      new std::vector<std::unique_ptr<const DexFile>>());
+  std::vector<std::unique_ptr<const DexFile>> dex_files;
   std::vector<std::string> error_msgs;
 
   bool success = linker->OpenDexFilesFromOat(sourceName.c_str(), outputName.c_str(), &error_msgs,
-                                             dex_files.get());
+                                             &dex_files);
 
-  if (success || !dex_files->empty()) {
-    // In the case of non-success, we have not found or could not generate the oat file.
-    // But we may still have found a dex file that we can use.
-    return static_cast<jlong>(reinterpret_cast<uintptr_t>(dex_files.release()));
+  if (success || !dex_files.empty()) {
+    jlongArray array = ConvertNativeToJavaArray(env, dex_files);
+    if (array == nullptr) {
+      ScopedObjectAccess soa(env);
+      for (auto& dex_file : dex_files) {
+        if (Runtime::Current()->GetClassLinker()->IsDexFileRegistered(*dex_file)) {
+          dex_file.release();
+        }
+      }
+    }
+    return array;
   } else {
     // The vector should be empty after a failed loading attempt.
-    DCHECK_EQ(0U, dex_files->size());
+    DCHECK_EQ(0U, dex_files.size());
 
     ScopedObjectAccess soa(env);
     CHECK(!error_msgs.empty());
@@ -140,27 +210,17 @@
       ThrowWrappedIOException("%s", it->c_str());
     }
 
-    return 0;
+    return nullptr;
   }
 }
 
-static std::vector<std::unique_ptr<const DexFile>>*
-toDexFiles(jlong dex_file_address, JNIEnv* env) {
-  std::vector<std::unique_ptr<const DexFile>>* dex_files
-    = reinterpret_cast<std::vector<std::unique_ptr<const DexFile>>*>(
-        static_cast<uintptr_t>(dex_file_address));
-  if (UNLIKELY(dex_files == nullptr)) {
-    ScopedObjectAccess soa(env);
-    ThrowNullPointerException(NULL, "dex_file == null");
-  }
-  return dex_files;
-}
-
-static void DexFile_closeDexFile(JNIEnv* env, jclass, jlong cookie) {
-  std::unique_ptr<std::vector<std::unique_ptr<const DexFile>>> dex_files(toDexFiles(cookie, env));
+static void DexFile_closeDexFile(JNIEnv* env, jclass, jobject cookie) {
+  std::unique_ptr<std::vector<const DexFile*>> dex_files = ConvertJavaArrayToNative(env, cookie);
   if (dex_files.get() == nullptr) {
+    DCHECK(env->ExceptionCheck());
     return;
   }
+
   ScopedObjectAccess soa(env);
 
   // The Runtime currently never unloads classes, which means any registered
@@ -171,19 +231,21 @@
   // TODO: The Runtime should support unloading of classes and freeing of the
   // dex files for those unloaded classes rather than leaking dex files here.
   for (auto& dex_file : *dex_files) {
-    if (Runtime::Current()->GetClassLinker()->IsDexFileRegistered(*dex_file)) {
-      dex_file.release();
+    if (!Runtime::Current()->GetClassLinker()->IsDexFileRegistered(*dex_file)) {
+      delete dex_file;
     }
   }
 }
 
 static jclass DexFile_defineClassNative(JNIEnv* env, jclass, jstring javaName, jobject javaLoader,
-                                        jlong cookie) {
-  std::vector<std::unique_ptr<const DexFile>>* dex_files = toDexFiles(cookie, env);
-  if (dex_files == NULL) {
+                                        jobject cookie) {
+  std::unique_ptr<std::vector<const DexFile*>> dex_files = ConvertJavaArrayToNative(env, cookie);
+  if (dex_files.get() == nullptr) {
     VLOG(class_linker) << "Failed to find dex_file";
-    return NULL;
+    DCHECK(env->ExceptionCheck());
+    return nullptr;
   }
+
   ScopedUtfChars class_name(env, javaName);
   if (class_name.c_str() == NULL) {
     VLOG(class_linker) << "Failed to find class_name";
@@ -221,36 +283,38 @@
 };
 
 // Note: this can be an expensive call, as we sort out duplicates in MultiDex files.
-static jobjectArray DexFile_getClassNameList(JNIEnv* env, jclass, jlong cookie) {
-  jobjectArray result = nullptr;
-  std::vector<std::unique_ptr<const DexFile>>* dex_files = toDexFiles(cookie, env);
+static jobjectArray DexFile_getClassNameList(JNIEnv* env, jclass, jobject cookie) {
+  std::unique_ptr<std::vector<const DexFile*>> dex_files = ConvertJavaArrayToNative(env, cookie);
+  if (dex_files.get() == nullptr) {
+    DCHECK(env->ExceptionCheck());
+    return nullptr;
+  }
 
-  if (dex_files != nullptr) {
-    // Push all class descriptors into a set. Use set instead of unordered_set as we want to
-    // retrieve all in the end.
-    std::set<const char*, CharPointerComparator> descriptors;
-    for (auto& dex_file : *dex_files) {
-      for (size_t i = 0; i < dex_file->NumClassDefs(); ++i) {
-        const DexFile::ClassDef& class_def = dex_file->GetClassDef(i);
-        const char* descriptor = dex_file->GetClassDescriptor(class_def);
-        descriptors.insert(descriptor);
-      }
+  // Push all class descriptors into a set. Use set instead of unordered_set as we want to
+  // retrieve all in the end.
+  std::set<const char*, CharPointerComparator> descriptors;
+  for (auto& dex_file : *dex_files) {
+    for (size_t i = 0; i < dex_file->NumClassDefs(); ++i) {
+      const DexFile::ClassDef& class_def = dex_file->GetClassDef(i);
+      const char* descriptor = dex_file->GetClassDescriptor(class_def);
+      descriptors.insert(descriptor);
     }
+  }
 
-    // Now create output array and copy the set into it.
-    result = env->NewObjectArray(descriptors.size(), WellKnownClasses::java_lang_String, nullptr);
-    if (result != nullptr) {
-      auto it = descriptors.begin();
-      auto it_end = descriptors.end();
-      jsize i = 0;
-      for (; it != it_end; it++, ++i) {
-        std::string descriptor(DescriptorToDot(*it));
-        ScopedLocalRef<jstring> jdescriptor(env, env->NewStringUTF(descriptor.c_str()));
-        if (jdescriptor.get() == nullptr) {
-          return nullptr;
-        }
-        env->SetObjectArrayElement(result, i, jdescriptor.get());
+  // Now create output array and copy the set into it.
+  jobjectArray result = env->NewObjectArray(descriptors.size(), WellKnownClasses::java_lang_String,
+                                            nullptr);
+  if (result != nullptr) {
+    auto it = descriptors.begin();
+    auto it_end = descriptors.end();
+    jsize i = 0;
+    for (; it != it_end; it++, ++i) {
+      std::string descriptor(DescriptorToDot(*it));
+      ScopedLocalRef<jstring> jdescriptor(env, env->NewStringUTF(descriptor.c_str()));
+      if (jdescriptor.get() == nullptr) {
+        return nullptr;
       }
+      env->SetObjectArrayElement(result, i, jdescriptor.get());
     }
   }
   return result;
@@ -620,12 +684,12 @@
 
 
 static JNINativeMethod gMethods[] = {
-  NATIVE_METHOD(DexFile, closeDexFile, "(J)V"),
-  NATIVE_METHOD(DexFile, defineClassNative, "(Ljava/lang/String;Ljava/lang/ClassLoader;J)Ljava/lang/Class;"),
-  NATIVE_METHOD(DexFile, getClassNameList, "(J)[Ljava/lang/String;"),
+  NATIVE_METHOD(DexFile, closeDexFile, "(Ljava/lang/Object;)V"),
+  NATIVE_METHOD(DexFile, defineClassNative, "(Ljava/lang/String;Ljava/lang/ClassLoader;Ljava/lang/Object;)Ljava/lang/Class;"),
+  NATIVE_METHOD(DexFile, getClassNameList, "(Ljava/lang/Object;)[Ljava/lang/String;"),
   NATIVE_METHOD(DexFile, isDexOptNeeded, "(Ljava/lang/String;)Z"),
   NATIVE_METHOD(DexFile, isDexOptNeededInternal, "(Ljava/lang/String;Ljava/lang/String;Ljava/lang/String;Z)B"),
-  NATIVE_METHOD(DexFile, openDexFileNative, "(Ljava/lang/String;Ljava/lang/String;I)J"),
+  NATIVE_METHOD(DexFile, openDexFileNative, "(Ljava/lang/String;Ljava/lang/String;I)Ljava/lang/Object;"),
 };
 
 void register_dalvik_system_DexFile(JNIEnv* env) {
diff --git a/runtime/native/dalvik_system_DexFile.h b/runtime/native/dalvik_system_DexFile.h
index 487df05..7585ab9 100644
--- a/runtime/native/dalvik_system_DexFile.h
+++ b/runtime/native/dalvik_system_DexFile.h
@@ -21,6 +21,8 @@
 
 namespace art {
 
+class DexFile;
+
 void register_dalvik_system_DexFile(JNIEnv* env);
 
 }  // namespace art
diff --git a/runtime/oat_file.cc b/runtime/oat_file.cc
index 9061bb3..356e3d2 100644
--- a/runtime/oat_file.cc
+++ b/runtime/oat_file.cc
@@ -52,17 +52,7 @@
   CHECK(has_section);
   oat_file->begin_ = elf_file->Begin() + offset;
   oat_file->end_ = elf_file->Begin() + size + offset;
-  return oat_file->Setup(error_msg) ? oat_file.release() : nullptr;
-}
-
-OatFile* OatFile::OpenMemory(std::vector<uint8_t>& oat_contents,
-                             const std::string& location,
-                             std::string* error_msg) {
-  CHECK(!oat_contents.empty()) << location;
-  CheckLocation(location);
-  std::unique_ptr<OatFile> oat_file(new OatFile(location, false));
-  oat_file->begin_ = &oat_contents[0];
-  oat_file->end_ = &oat_contents[oat_contents.size()];
+  // Ignore the optional .bss section when opening non-executable.
   return oat_file->Setup(error_msg) ? oat_file.release() : nullptr;
 }
 
@@ -108,18 +98,6 @@
   return OpenElfFile(file, location, nullptr, nullptr, false, false, error_msg);
 }
 
-OatFile* OatFile::OpenDlopen(const std::string& elf_filename,
-                             const std::string& location,
-                             uint8_t* requested_base,
-                             std::string* error_msg) {
-  std::unique_ptr<OatFile> oat_file(new OatFile(location, true));
-  bool success = oat_file->Dlopen(elf_filename, requested_base, error_msg);
-  if (!success) {
-    return nullptr;
-  }
-  return oat_file.release();
-}
-
 OatFile* OatFile::OpenElfFile(File* file,
                               const std::string& location,
                               uint8_t* requested_base,
@@ -138,8 +116,8 @@
 }
 
 OatFile::OatFile(const std::string& location, bool is_executable)
-    : location_(location), begin_(NULL), end_(NULL), is_executable_(is_executable),
-      dlopen_handle_(NULL),
+    : location_(location), begin_(NULL), end_(NULL), bss_begin_(nullptr), bss_end_(nullptr),
+      is_executable_(is_executable), dlopen_handle_(NULL),
       secondary_lookup_lock_("OatFile secondary lookup lock", kOatFileSecondaryLookupLock) {
   CHECK(!location_.empty());
 }
@@ -151,43 +129,6 @@
   }
 }
 
-bool OatFile::Dlopen(const std::string& elf_filename, uint8_t* requested_base,
-                     std::string* error_msg) {
-  char* absolute_path = realpath(elf_filename.c_str(), NULL);
-  if (absolute_path == NULL) {
-    *error_msg = StringPrintf("Failed to find absolute path for '%s'", elf_filename.c_str());
-    return false;
-  }
-  dlopen_handle_ = dlopen(absolute_path, RTLD_NOW);
-  free(absolute_path);
-  if (dlopen_handle_ == NULL) {
-    *error_msg = StringPrintf("Failed to dlopen '%s': %s", elf_filename.c_str(), dlerror());
-    return false;
-  }
-  begin_ = reinterpret_cast<uint8_t*>(dlsym(dlopen_handle_, "oatdata"));
-  if (begin_ == NULL) {
-    *error_msg = StringPrintf("Failed to find oatdata symbol in '%s': %s", elf_filename.c_str(),
-                              dlerror());
-    return false;
-  }
-  if (requested_base != NULL && begin_ != requested_base) {
-    *error_msg = StringPrintf("Failed to find oatdata symbol at expected address: "
-                              "oatdata=%p != expected=%p /proc/self/maps:\n",
-                              begin_, requested_base);
-    ReadFileToString("/proc/self/maps", error_msg);
-    return false;
-  }
-  end_ = reinterpret_cast<uint8_t*>(dlsym(dlopen_handle_, "oatlastword"));
-  if (end_ == NULL) {
-    *error_msg = StringPrintf("Failed to find oatlastword symbol in '%s': %s", elf_filename.c_str(),
-                              dlerror());
-    return false;
-  }
-  // Readjust to be non-inclusive upper bound.
-  end_ += sizeof(uint32_t);
-  return Setup(error_msg);
-}
-
 bool OatFile::ElfFileOpen(File* file, uint8_t* requested_base, uint8_t* oat_file_begin,
                           bool writable, bool executable,
                           std::string* error_msg) {
@@ -209,10 +150,10 @@
     return false;
   }
   if (requested_base != NULL && begin_ != requested_base) {
+    PrintFileToLog("/proc/self/maps", LogSeverity::WARNING);
     *error_msg = StringPrintf("Failed to find oatdata symbol at expected address: "
-                              "oatdata=%p != expected=%p /proc/self/maps:\n",
+                              "oatdata=%p != expected=%p. See process maps in the log.",
                               begin_, requested_base);
-    ReadFileToString("/proc/self/maps", error_msg);
     return false;
   }
   end_ = elf_file_->FindDynamicSymbolAddress("oatlastword");
@@ -222,6 +163,23 @@
   }
   // Readjust to be non-inclusive upper bound.
   end_ += sizeof(uint32_t);
+
+  bss_begin_ = elf_file_->FindDynamicSymbolAddress("oatbss");
+  if (bss_begin_ == nullptr) {
+    // No .bss section. Clear dlerror().
+    bss_end_ = nullptr;
+    dlerror();
+  } else {
+    bss_end_ = elf_file_->FindDynamicSymbolAddress("oatbsslastword");
+    if (bss_end_ == nullptr) {
+      *error_msg = StringPrintf("Failed to find oatbasslastword symbol in '%s'",
+                                file->GetPath().c_str());
+      return false;
+    }
+    // Readjust to be non-inclusive upper bound.
+    bss_end_ += sizeof(uint32_t);
+  }
+
   return Setup(error_msg);
 }
 
@@ -363,6 +321,14 @@
   return end_;
 }
 
+const uint8_t* OatFile::BssBegin() const {
+  return bss_begin_;
+}
+
+const uint8_t* OatFile::BssEnd() const {
+  return bss_end_;
+}
+
 const OatFile::OatDexFile* OatFile::GetOatDexFile(const char* dex_location,
                                                   const uint32_t* dex_location_checksum,
                                                   bool warn_if_not_found) const {
@@ -577,12 +543,12 @@
   }
   if (oat_file_->IsExecutable() ||
       Runtime::Current() == nullptr ||        // This case applies for oatdump.
-      Runtime::Current()->IsCompiler()) {
+      Runtime::Current()->IsAotCompiler()) {
     return OatMethod(oat_file_->Begin(), oat_method_offsets->code_offset_);
-  } else {
-    // We aren't allowed to use the compiled code. We just force it down the interpreted version.
-    return OatMethod(oat_file_->Begin(), 0);
   }
+  // We aren't allowed to use the compiled code. We just force it down the interpreted / jit
+  // version.
+  return OatMethod(oat_file_->Begin(), 0);
 }
 
 void OatFile::OatMethod::LinkMethod(mirror::ArtMethod* method) const {
diff --git a/runtime/oat_file.h b/runtime/oat_file.h
index 6ae3c3e..564185c 100644
--- a/runtime/oat_file.h
+++ b/runtime/oat_file.h
@@ -62,11 +62,6 @@
   // Opens an oat file from an already opened File. Maps it PROT_READ, MAP_PRIVATE.
   static OatFile* OpenReadable(File* file, const std::string& location, std::string* error_msg);
 
-  // Open an oat file backed by a std::vector with the given location.
-  static OatFile* OpenMemory(std::vector<uint8_t>& oat_contents,
-                             const std::string& location,
-                             std::string* error_msg);
-
   ~OatFile();
 
   bool IsExecutable() const {
@@ -147,8 +142,8 @@
       return reinterpret_cast<T>(begin_ + offset);
     }
 
-    const uint8_t* const begin_;
-    const uint32_t code_offset_;
+    const uint8_t* begin_;
+    uint32_t code_offset_;
 
     friend class OatClass;
   };
@@ -274,17 +269,19 @@
     return End() - Begin();
   }
 
+  size_t BssSize() const {
+    return BssEnd() - BssBegin();
+  }
+
   const uint8_t* Begin() const;
   const uint8_t* End() const;
 
+  const uint8_t* BssBegin() const;
+  const uint8_t* BssEnd() const;
+
  private:
   static void CheckLocation(const std::string& location);
 
-  static OatFile* OpenDlopen(const std::string& elf_filename,
-                             const std::string& location,
-                             uint8_t* requested_base,
-                             std::string* error_msg);
-
   static OatFile* OpenElfFile(File* file,
                               const std::string& location,
                               uint8_t* requested_base,
@@ -294,7 +291,6 @@
                               std::string* error_msg);
 
   explicit OatFile(const std::string& filename, bool executable);
-  bool Dlopen(const std::string& elf_filename, uint8_t* requested_base, std::string* error_msg);
   bool ElfFileOpen(File* file, uint8_t* requested_base,
                    uint8_t* oat_file_begin,  // Override where the file is loaded to if not null
                    bool writable, bool executable,
@@ -312,6 +308,12 @@
   // Pointer to end of oat region for bounds checking.
   const uint8_t* end_;
 
+  // Pointer to the .bss section, if present, otherwise nullptr.
+  const uint8_t* bss_begin_;
+
+  // Pointer to the end of the .bss section, if present, otherwise nullptr.
+  const uint8_t* bss_end_;
+
   // Was this oat_file loaded executable?
   const bool is_executable_;
 
diff --git a/runtime/object_lock.cc b/runtime/object_lock.cc
index f7accc0..749fb5d 100644
--- a/runtime/object_lock.cc
+++ b/runtime/object_lock.cc
@@ -47,6 +47,7 @@
   obj_->NotifyAll(self_);
 }
 
+template class ObjectLock<mirror::ArtMethod>;
 template class ObjectLock<mirror::Class>;
 template class ObjectLock<mirror::Object>;
 
diff --git a/runtime/parsed_options.cc b/runtime/parsed_options.cc
index c0c7baa..9d87ed7 100644
--- a/runtime/parsed_options.cc
+++ b/runtime/parsed_options.cc
@@ -146,6 +146,15 @@
       .Define({"-XX:EnableHSpaceCompactForOOM", "-XX:DisableHSpaceCompactForOOM"})
           .WithValues({true, false})
           .IntoKey(M::EnableHSpaceCompactForOOM)
+      .Define({"-Xjit", "-Xnojit"})
+          .WithValues({true, false})
+          .IntoKey(M::UseJIT)
+      .Define("-Xjitcodecachesize:_")
+          .WithType<MemoryKiB>()
+          .IntoKey(M::JITCodeCacheCapacity)
+      .Define("-Xjitthreshold:_")
+          .WithType<unsigned int>()
+          .IntoKey(M::JITCompileThreshold)
       .Define("-XX:HspaceCompactForOOMMinIntervalMs=_")  // in ms
           .WithType<MillisecondsToNanoseconds>()  // store as ns
           .IntoKey(M::HSpaceCompactForOOMMinIntervalsMs)
@@ -248,7 +257,7 @@
           "-dsa", "-enablesystemassertions", "-disablesystemassertions", "-Xrs", "-Xint:_",
           "-Xdexopt:_", "-Xnoquithandler", "-Xjnigreflimit:_", "-Xgenregmap", "-Xnogenregmap",
           "-Xverifyopt:_", "-Xcheckdexsum", "-Xincludeselectedop", "-Xjitop:_",
-          "-Xincludeselectedmethod", "-Xjitthreshold:_", "-Xjitcodecachesize:_",
+          "-Xincludeselectedmethod", "-Xjitthreshold:_",
           "-Xjitblocking", "-Xjitmethod:_", "-Xjitclass:_", "-Xjitoffset:_",
           "-Xjitconfig:_", "-Xjitcheckcg", "-Xjitverbose", "-Xjitprofile",
           "-Xjitdisableopt", "-Xjitsuspendpoll", "-XX:mainThreadStackSize=_"})
@@ -353,19 +362,20 @@
 
 bool ParsedOptions::Parse(const RuntimeOptions& options, bool ignore_unrecognized,
                           RuntimeArgumentMap* runtime_options) {
-//  gLogVerbosity.class_linker = true;  // TODO: don't check this in!
-//  gLogVerbosity.compiler = true;  // TODO: don't check this in!
-//  gLogVerbosity.gc = true;  // TODO: don't check this in!
-//  gLogVerbosity.heap = true;  // TODO: don't check this in!
-//  gLogVerbosity.jdwp = true;  // TODO: don't check this in!
-//  gLogVerbosity.jni = true;  // TODO: don't check this in!
-//  gLogVerbosity.monitor = true;  // TODO: don't check this in!
-//  gLogVerbosity.profiler = true;  // TODO: don't check this in!
-//  gLogVerbosity.signals = true;  // TODO: don't check this in!
-//  gLogVerbosity.startup = true;  // TODO: don't check this in!
-//  gLogVerbosity.third_party_jni = true;  // TODO: don't check this in!
-//  gLogVerbosity.threads = true;  // TODO: don't check this in!
-//  gLogVerbosity.verifier = true;  // TODO: don't check this in!
+  //  gLogVerbosity.class_linker = true;  // TODO: don't check this in!
+  //  gLogVerbosity.compiler = true;  // TODO: don't check this in!
+  //  gLogVerbosity.gc = true;  // TODO: don't check this in!
+  //  gLogVerbosity.heap = true;  // TODO: don't check this in!
+  //  gLogVerbosity.jdwp = true;  // TODO: don't check this in!
+  //  gLogVerbosity.jit = true;  // TODO: don't check this in!
+  //  gLogVerbosity.jni = true;  // TODO: don't check this in!
+  //  gLogVerbosity.monitor = true;  // TODO: don't check this in!
+  //  gLogVerbosity.profiler = true;  // TODO: don't check this in!
+  //  gLogVerbosity.signals = true;  // TODO: don't check this in!
+  //  gLogVerbosity.startup = true;  // TODO: don't check this in!
+  //  gLogVerbosity.third_party_jni = true;  // TODO: don't check this in!
+  //  gLogVerbosity.threads = true;  // TODO: don't check this in!
+  //  gLogVerbosity.verifier = true;  // TODO: don't check this in!
 
   for (size_t i = 0; i < options.size(); ++i) {
     if (true && options[0].first == "-Xzygote") {
@@ -560,7 +570,7 @@
   UsageMessage(stream, "The following standard options are supported:\n");
   UsageMessage(stream, "  -classpath classpath (-cp classpath)\n");
   UsageMessage(stream, "  -Dproperty=value\n");
-  UsageMessage(stream, "  -verbose:tag ('gc', 'jni', or 'class')\n");
+  UsageMessage(stream, "  -verbose:tag ('gc', 'jit', 'jni', or 'class')\n");
   UsageMessage(stream, "  -showversion\n");
   UsageMessage(stream, "  -help\n");
   UsageMessage(stream, "  -agentlib:jdwp=options\n");
@@ -590,6 +600,8 @@
   UsageMessage(stream, "  -XX:ForegroundHeapGrowthMultiplier=doublevalue\n");
   UsageMessage(stream, "  -XX:LowMemoryMode\n");
   UsageMessage(stream, "  -Xprofile:{threadcpuclock,wallclock,dualclock}\n");
+  UsageMessage(stream, "  -Xjitcodecachesize:N\n");
+  UsageMessage(stream, "  -Xjitthreshold:integervalue\n");
   UsageMessage(stream, "\n");
 
   UsageMessage(stream, "The following unique to ART options are supported:\n");
@@ -630,6 +642,8 @@
   UsageMessage(stream, "  -Xcompiler-option dex2oat-option\n");
   UsageMessage(stream, "  -Ximage-compiler-option dex2oat-option\n");
   UsageMessage(stream, "  -Xpatchoat:filename\n");
+  UsageMessage(stream, "  -Xjit\n");
+  UsageMessage(stream, "  -Xnojit\n");
   UsageMessage(stream, "  -X[no]relocate\n");
   UsageMessage(stream, "  -X[no]dex2oat (Whether to invoke dex2oat on the application)\n");
   UsageMessage(stream, "  -X[no]image-dex2oat (Whether to create and use a boot image)\n");
@@ -659,8 +673,6 @@
   UsageMessage(stream, "  -Xincludeselectedop\n");
   UsageMessage(stream, "  -Xjitop:hexopvalue[-endvalue][,hexopvalue[-endvalue]]*\n");
   UsageMessage(stream, "  -Xincludeselectedmethod\n");
-  UsageMessage(stream, "  -Xjitthreshold:integervalue\n");
-  UsageMessage(stream, "  -Xjitcodecachesize:decimalvalueofkbytes\n");
   UsageMessage(stream, "  -Xjitblocking\n");
   UsageMessage(stream, "  -Xjitmethod:signature[,signature]* (eg Ljava/lang/String\\;replace)\n");
   UsageMessage(stream, "  -Xjitclass:classname[,classname]*\n");
diff --git a/runtime/profiler.cc b/runtime/profiler.cc
index c3bdcb1..db372c3 100644
--- a/runtime/profiler.cc
+++ b/runtime/profiler.cc
@@ -161,7 +161,7 @@
 
 
   CHECK(runtime->AttachCurrentThread("Profiler", true, runtime->GetSystemThreadGroup(),
-                                      !runtime->IsCompiler()));
+                                      !runtime->IsAotCompiler()));
 
   Thread* self = Thread::Current();
 
diff --git a/runtime/quick/inline_method_analyser.cc b/runtime/quick/inline_method_analyser.cc
index d65b2d5..44e2844 100644
--- a/runtime/quick/inline_method_analyser.cc
+++ b/runtime/quick/inline_method_analyser.cc
@@ -77,7 +77,9 @@
                                              InlineMethod* method) {
   DCHECK(verifier != nullptr);
   DCHECK_EQ(Runtime::Current()->IsCompiler(), method != nullptr);
-  DCHECK_EQ(verifier->CanLoadClasses(), method != nullptr);
+  if (!Runtime::Current()->UseJit()) {
+    DCHECK_EQ(verifier->CanLoadClasses(), method != nullptr);
+  }
   // We currently support only plain return or 2-instruction methods.
 
   const DexFile::CodeItem* code_item = verifier->CodeItem();
@@ -110,6 +112,10 @@
     case Instruction::IGET_CHAR:
     case Instruction::IGET_SHORT:
     case Instruction::IGET_WIDE:
+    // TODO: Add handling for JIT.
+    // case Instruction::IGET_QUICK:
+    // case Instruction::IGET_WIDE_QUICK:
+    // case Instruction::IGET_OBJECT_QUICK:
       return AnalyseIGetMethod(verifier, method);
     case Instruction::IPUT:
     case Instruction::IPUT_OBJECT:
@@ -118,6 +124,10 @@
     case Instruction::IPUT_CHAR:
     case Instruction::IPUT_SHORT:
     case Instruction::IPUT_WIDE:
+      // TODO: Add handling for JIT.
+    // case Instruction::IPUT_QUICK:
+    // case Instruction::IPUT_WIDE_QUICK:
+    // case Instruction::IPUT_OBJECT_QUICK:
       return AnalyseIPutMethod(verifier, method);
     default:
       return false;
diff --git a/runtime/runtime.cc b/runtime/runtime.cc
index 8cd9e24..383308c 100644
--- a/runtime/runtime.cc
+++ b/runtime/runtime.cc
@@ -65,6 +65,8 @@
 #include "image.h"
 #include "instrumentation.h"
 #include "intern_table.h"
+#include "interpreter/interpreter.h"
+#include "jit/jit.h"
 #include "jni_internal.h"
 #include "mirror/array.h"
 #include "mirror/art_field-inl.h"
@@ -229,6 +231,12 @@
   // Make sure to let the GC complete if it is running.
   heap_->WaitForGcToComplete(gc::kGcCauseBackground, self);
   heap_->DeleteThreadPool();
+  if (jit_.get() != nullptr) {
+    VLOG(jit) << "Deleting jit thread pool";
+    // Delete thread pool before the thread list since we don't want to wait forever on the
+    // JIT compiler threads.
+    jit_->DeleteThreadPool();
+  }
 
   // Make sure our internal threads are dead before we start tearing down things they're using.
   Dbg::StopJdwp();
@@ -237,6 +245,13 @@
   // Make sure all other non-daemon threads have terminated, and all daemon threads are suspended.
   delete thread_list_;
 
+  // Delete the JIT after thread list to ensure that there is no remaining threads which could be
+  // accessing the instrumentation when we delete it.
+  if (jit_.get() != nullptr) {
+    VLOG(jit) << "Deleting jit";
+    jit_.reset(nullptr);
+  }
+
   // Shutdown the fault manager if it was initialized.
   fault_manager.Shutdown();
 
@@ -461,17 +476,24 @@
 
   started_ = true;
 
-  // Use !IsCompiler so that we get test coverage, tests are never the zygote.
-  if (!IsCompiler()) {
+  // Use !IsAotCompiler so that we get test coverage, tests are never the zygote.
+  if (!IsAotCompiler()) {
     ScopedObjectAccess soa(self);
     gc::space::ImageSpace* image_space = heap_->GetImageSpace();
     if (image_space != nullptr) {
-      Runtime::Current()->GetInternTable()->AddImageStringsToTable(image_space);
-      Runtime::Current()->GetClassLinker()->MoveImageClassesToClassTable();
+      GetInternTable()->AddImageStringsToTable(image_space);
+      GetClassLinker()->MoveImageClassesToClassTable();
     }
   }
 
-  if (!IsImageDex2OatEnabled() || !Runtime::Current()->GetHeap()->HasImageSpace()) {
+  // 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.
+  if (!IsZygote() && jit_.get() != nullptr) {
+    jit_->CreateInstrumentationCache(jit_options_->GetCompileThreshold());
+    jit_->CreateThreadPool();
+  }
+
+  if (!IsImageDex2OatEnabled() || !GetHeap()->HasImageSpace()) {
     ScopedObjectAccess soa(self);
     StackHandleScope<1> hs(soa.Self());
     auto klass(hs.NewHandle<mirror::Class>(mirror::Class::GetJavaLangClass()));
@@ -590,8 +612,14 @@
     }
   }
 
-  // Create the thread pool.
+  // Create the thread pools.
   heap_->CreateThreadPool();
+  if (jit_options_.get() != nullptr && jit_.get() == nullptr) {
+    // Create the JIT if the flag is set and we haven't already create it (happens for run-tests).
+    CreateJit();
+    jit_->CreateInstrumentationCache(jit_options_->GetCompileThreshold());
+    jit_->CreateThreadPool();
+  }
 
   StartSignalCatcher();
 
@@ -818,6 +846,17 @@
     Dbg::ConfigureJdwp(runtime_options.GetOrDefault(Opt::JdwpOptions));
   }
 
+  if (!IsCompiler()) {
+    // If we are already the compiler at this point, we must be dex2oat. Don't create the jit in
+    // this case.
+    // If runtime_options doesn't have UseJIT set to true then CreateFromRuntimeArguments returns
+    // nullptr and we don't create the jit.
+    jit_options_.reset(jit::JitOptions::CreateFromRuntimeArguments(runtime_options));
+  }
+  if (!IsZygote() && jit_options_.get() != nullptr) {
+    CreateJit();
+  }
+
   BlockSignals();
   InitPlatformSignalHandlers();
 
@@ -1066,26 +1105,26 @@
       env->NewGlobalRef(env->GetStaticObjectField(
           WellKnownClasses::java_lang_ThreadGroup,
           WellKnownClasses::java_lang_ThreadGroup_mainThreadGroup));
-  CHECK(main_thread_group_ != NULL || IsCompiler());
+  CHECK(main_thread_group_ != NULL || IsAotCompiler());
   system_thread_group_ =
       env->NewGlobalRef(env->GetStaticObjectField(
           WellKnownClasses::java_lang_ThreadGroup,
           WellKnownClasses::java_lang_ThreadGroup_systemThreadGroup));
-  CHECK(system_thread_group_ != NULL || IsCompiler());
+  CHECK(system_thread_group_ != NULL || IsAotCompiler());
 }
 
 jobject Runtime::GetMainThreadGroup() const {
-  CHECK(main_thread_group_ != NULL || IsCompiler());
+  CHECK(main_thread_group_ != NULL || IsAotCompiler());
   return main_thread_group_;
 }
 
 jobject Runtime::GetSystemThreadGroup() const {
-  CHECK(system_thread_group_ != NULL || IsCompiler());
+  CHECK(system_thread_group_ != NULL || IsAotCompiler());
   return system_thread_group_;
 }
 
 jobject Runtime::GetSystemClassLoader() const {
-  CHECK(system_class_loader_ != NULL || IsCompiler());
+  CHECK(system_class_loader_ != NULL || IsAotCompiler());
   return system_class_loader_;
 }
 
@@ -1341,7 +1380,7 @@
   // TODO: use a special method for imt conflict method saves.
   method->SetDexMethodIndex(DexFile::kDexNoIndex);
   // When compiling, the code pointer will get set later when the image is loaded.
-  if (runtime->IsCompiler()) {
+  if (runtime->IsAotCompiler()) {
     size_t pointer_size = GetInstructionSetPointerSize(instruction_set_);
     method->SetEntryPointFromQuickCompiledCodePtrSize(nullptr, pointer_size);
   } else {
@@ -1350,6 +1389,10 @@
   return method.Get();
 }
 
+void Runtime::SetImtConflictMethod(mirror::ArtMethod* method) {
+  imt_conflict_method_ = GcRoot<mirror::ArtMethod>(method);
+}
+
 mirror::ArtMethod* Runtime::CreateResolutionMethod() {
   Thread* self = Thread::Current();
   Runtime* runtime = Runtime::Current();
@@ -1360,7 +1403,7 @@
   // TODO: use a special method for resolution method saves
   method->SetDexMethodIndex(DexFile::kDexNoIndex);
   // When compiling, the code pointer will get set later when the image is loaded.
-  if (runtime->IsCompiler()) {
+  if (runtime->IsAotCompiler()) {
     size_t pointer_size = GetInstructionSetPointerSize(instruction_set_);
     method->SetEntryPointFromQuickCompiledCodePtrSize(nullptr, pointer_size);
   } else {
@@ -1491,14 +1534,14 @@
 
 // Transaction support.
 void Runtime::EnterTransactionMode(Transaction* transaction) {
-  DCHECK(IsCompiler());
+  DCHECK(IsAotCompiler());
   DCHECK(transaction != nullptr);
   DCHECK(!IsActiveTransaction());
   preinitialization_transaction_ = transaction;
 }
 
 void Runtime::ExitTransactionMode() {
-  DCHECK(IsCompiler());
+  DCHECK(IsAotCompiler());
   DCHECK(IsActiveTransaction());
   preinitialization_transaction_ = nullptr;
 }
@@ -1517,15 +1560,17 @@
                                                     const std::string& abort_message) {
   DCHECK(IsCompiler());
   DCHECK(IsActiveTransaction());
+  // Throwing an exception may cause its class initialization. If we mark the transaction
+  // aborted before that, we may warn with a false alarm. Throwing the exception before
+  // marking the transaction aborted avoids that.
+  preinitialization_transaction_->ThrowInternalError(self, false);
   preinitialization_transaction_->Abort(abort_message);
-  ThrowInternalErrorForAbortedTransaction(self);
 }
 
 void Runtime::ThrowInternalErrorForAbortedTransaction(Thread* self) {
   DCHECK(IsCompiler());
   DCHECK(IsActiveTransaction());
-  DCHECK(IsTransactionAborted());
-  preinitialization_transaction_->ThrowInternalError(self);
+  preinitialization_transaction_->ThrowInternalError(self, true);
 }
 
 void Runtime::RecordWriteFieldBoolean(mirror::Object* obj, MemberOffset field_offset,
@@ -1558,51 +1603,51 @@
 
 void Runtime::RecordWriteField32(mirror::Object* obj, MemberOffset field_offset,
                                  uint32_t value, bool is_volatile) const {
-  DCHECK(IsCompiler());
+  DCHECK(IsAotCompiler());
   DCHECK(IsActiveTransaction());
   preinitialization_transaction_->RecordWriteField32(obj, field_offset, value, is_volatile);
 }
 
 void Runtime::RecordWriteField64(mirror::Object* obj, MemberOffset field_offset,
                                  uint64_t value, bool is_volatile) const {
-  DCHECK(IsCompiler());
+  DCHECK(IsAotCompiler());
   DCHECK(IsActiveTransaction());
   preinitialization_transaction_->RecordWriteField64(obj, field_offset, value, is_volatile);
 }
 
 void Runtime::RecordWriteFieldReference(mirror::Object* obj, MemberOffset field_offset,
                                         mirror::Object* value, bool is_volatile) const {
-  DCHECK(IsCompiler());
+  DCHECK(IsAotCompiler());
   DCHECK(IsActiveTransaction());
   preinitialization_transaction_->RecordWriteFieldReference(obj, field_offset, value, is_volatile);
 }
 
 void Runtime::RecordWriteArray(mirror::Array* array, size_t index, uint64_t value) const {
-  DCHECK(IsCompiler());
+  DCHECK(IsAotCompiler());
   DCHECK(IsActiveTransaction());
   preinitialization_transaction_->RecordWriteArray(array, index, value);
 }
 
 void Runtime::RecordStrongStringInsertion(mirror::String* s) const {
-  DCHECK(IsCompiler());
+  DCHECK(IsAotCompiler());
   DCHECK(IsActiveTransaction());
   preinitialization_transaction_->RecordStrongStringInsertion(s);
 }
 
 void Runtime::RecordWeakStringInsertion(mirror::String* s) const {
-  DCHECK(IsCompiler());
+  DCHECK(IsAotCompiler());
   DCHECK(IsActiveTransaction());
   preinitialization_transaction_->RecordWeakStringInsertion(s);
 }
 
 void Runtime::RecordStrongStringRemoval(mirror::String* s) const {
-  DCHECK(IsCompiler());
+  DCHECK(IsAotCompiler());
   DCHECK(IsActiveTransaction());
   preinitialization_transaction_->RecordStrongStringRemoval(s);
 }
 
 void Runtime::RecordWeakStringRemoval(mirror::String* s) const {
-  DCHECK(IsCompiler());
+  DCHECK(IsAotCompiler());
   DCHECK(IsActiveTransaction());
   preinitialization_transaction_->RecordWeakStringRemoval(s);
 }
@@ -1614,7 +1659,7 @@
 
 void Runtime::AddCurrentRuntimeFeaturesAsDex2OatArguments(std::vector<std::string>* argv)
     const {
-  if (GetInstrumentation()->InterpretOnly()) {
+  if (GetInstrumentation()->InterpretOnly() || UseJit()) {
     argv->push_back("--compiler-filter=interpret-only");
   }
 
@@ -1634,4 +1679,16 @@
 void Runtime::UpdateProfilerState(int state) {
   VLOG(profiler) << "Profiler state updated to " << state;
 }
+
+void Runtime::CreateJit() {
+  CHECK(jit_options_.get() != nullptr);
+  std::string error_msg;
+  jit_.reset(jit::Jit::Create(jit_options_.get(), &error_msg));
+  if (jit_.get() != nullptr) {
+    compiler_callbacks_ = jit_->GetCompilerCallbacks();
+  } else {
+    LOG(WARNING) << "Failed to create JIT " << error_msg;
+  }
+}
+
 }  // namespace art
diff --git a/runtime/runtime.h b/runtime/runtime.h
index 944c8bd..5078b7f 100644
--- a/runtime/runtime.h
+++ b/runtime/runtime.h
@@ -48,6 +48,12 @@
     class GarbageCollector;
   }  // namespace collector
 }  // namespace gc
+
+namespace jit {
+  class Jit;
+  class JitOptions;
+}  // namespace jit
+
 namespace mirror {
   class ArtMethod;
   class ClassLoader;
@@ -95,12 +101,18 @@
   static bool Create(const RuntimeOptions& options, bool ignore_unrecognized)
       SHARED_TRYLOCK_FUNCTION(true, Locks::mutator_lock_);
 
+  // IsAotCompiler for compilers that don't have a running runtime. Only dex2oat currently.
+  bool IsAotCompiler() const {
+    return !UseJit() && IsCompiler();
+  }
+
+  // IsCompiler is any runtime which has a running compiler, either dex2oat or JIT.
   bool IsCompiler() const {
     return compiler_callbacks_ != nullptr;
   }
 
   bool CanRelocate() const {
-    return !IsCompiler() || compiler_callbacks_->IsRelocationPossible();
+    return !IsAotCompiler() || compiler_callbacks_->IsRelocationPossible();
   }
 
   bool ShouldRelocate() const {
@@ -339,9 +351,7 @@
     return !imt_conflict_method_.IsNull();
   }
 
-  void SetImtConflictMethod(mirror::ArtMethod* method) {
-    imt_conflict_method_ = GcRoot<mirror::ArtMethod>(method);
-  }
+  void SetImtConflictMethod(mirror::ArtMethod* method);
   void SetImtUnimplementedMethod(mirror::ArtMethod* method) {
     imt_unimplemented_method_ = GcRoot<mirror::ArtMethod>(method);
   }
@@ -421,6 +431,14 @@
     kUnload,
     kInitialize
   };
+
+  jit::Jit* GetJit() {
+    return jit_.get();
+  }
+  bool UseJit() const {
+    return jit_.get() != nullptr;
+  }
+
   void PreZygoteFork();
   bool InitZygote();
   void DidForkFromZygote(JNIEnv* env, NativeBridgeAction action, const char* isa);
@@ -525,6 +543,8 @@
     return zygote_max_failed_boots_;
   }
 
+  void CreateJit();
+
  private:
   static void InitPlatformSignalHandlers();
 
@@ -604,6 +624,9 @@
 
   JavaVMExt* java_vm_;
 
+  std::unique_ptr<jit::Jit> jit_;
+  std::unique_ptr<jit::JitOptions> jit_options_;
+
   // Fault message, printed when we get a SIGSEGV.
   Mutex fault_message_lock_ DEFAULT_MUTEX_ACQUIRED_AFTER;
   std::string fault_message_ GUARDED_BY(fault_message_lock_);
diff --git a/runtime/runtime_options.def b/runtime/runtime_options.def
index d9cc4d4..d072ffa 100644
--- a/runtime/runtime_options.def
+++ b/runtime/runtime_options.def
@@ -63,6 +63,9 @@
 RUNTIME_OPTIONS_KEY (Unit,                LowMemoryMode)
 RUNTIME_OPTIONS_KEY (bool,                UseTLAB,                        false)
 RUNTIME_OPTIONS_KEY (bool,                EnableHSpaceCompactForOOM,      true)
+RUNTIME_OPTIONS_KEY (bool,                UseJIT,      false)
+RUNTIME_OPTIONS_KEY (unsigned int,        JITCompileThreshold, jit::Jit::kDefaultCompileThreshold)
+RUNTIME_OPTIONS_KEY (MemoryKiB,           JITCodeCacheCapacity, jit::JitCodeCache::kDefaultCapacity)
 RUNTIME_OPTIONS_KEY (MillisecondsToNanoseconds, \
                                           HSpaceCompactForOOMMinIntervalsMs,\
                                                                           MsToNs(100 * 1000))  // 100s
diff --git a/runtime/runtime_options.h b/runtime/runtime_options.h
index ebd52d7..7e59000 100644
--- a/runtime/runtime_options.h
+++ b/runtime/runtime_options.h
@@ -26,6 +26,8 @@
 #include "runtime/base/logging.h"
 #include "cmdline/unit.h"
 #include "jdwp/jdwp.h"
+#include "jit/jit.h"
+#include "jit/jit_code_cache.h"
 #include "gc/collector_type.h"
 #include "gc/space/large_object_space.h"
 #include "profiler_options.h"
diff --git a/runtime/signal_catcher.cc b/runtime/signal_catcher.cc
index e377542..26bf655 100644
--- a/runtime/signal_catcher.cc
+++ b/runtime/signal_catcher.cc
@@ -180,7 +180,7 @@
 
   Runtime* runtime = Runtime::Current();
   CHECK(runtime->AttachCurrentThread("Signal Catcher", true, runtime->GetSystemThreadGroup(),
-                                     !runtime->IsCompiler()));
+                                     !runtime->IsAotCompiler()));
 
   Thread* self = Thread::Current();
   DCHECK_NE(self->GetState(), kRunnable);
diff --git a/runtime/stack.cc b/runtime/stack.cc
index b39aebf..97a8d01 100644
--- a/runtime/stack.cc
+++ b/runtime/stack.cc
@@ -163,42 +163,10 @@
   if (cur_quick_frame_ != nullptr) {
     DCHECK(context_ != nullptr);  // You can't reliably read registers without a context.
     DCHECK(m == GetMethod());
-    const void* code_pointer = m->GetQuickOatCodePointer(sizeof(void*));
-    DCHECK(code_pointer != nullptr);
-    const VmapTable vmap_table(m->GetVmapTable(code_pointer, sizeof(void*)));
-    QuickMethodFrameInfo frame_info = m->GetQuickFrameInfo(code_pointer);
-    uint32_t vmap_offset;
-    // TODO: IsInContext stops before spotting floating point registers.
-    if (vmap_table.IsInContext(vreg, kind, &vmap_offset)) {
-      bool is_float = (kind == kFloatVReg) || (kind == kDoubleLoVReg) || (kind == kDoubleHiVReg);
-      uint32_t spill_mask = is_float ? frame_info.FpSpillMask() : frame_info.CoreSpillMask();
-      uint32_t reg = vmap_table.ComputeRegister(spill_mask, vmap_offset, kind);
-      if (!IsAccessibleRegister(reg, is_float)) {
-        return false;
-      }
-      uintptr_t ptr_val = GetRegister(reg, is_float);
-      bool target64 = Is64BitInstructionSet(kRuntimeISA);
-      if (target64) {
-        bool wide_lo = (kind == kLongLoVReg) || (kind == kDoubleLoVReg);
-        bool wide_hi = (kind == kLongHiVReg) || (kind == kDoubleHiVReg);
-        int64_t value_long = static_cast<int64_t>(ptr_val);
-        if (wide_lo) {
-          ptr_val = static_cast<uintptr_t>(value_long & 0xFFFFFFFF);
-        } else if (wide_hi) {
-          ptr_val = static_cast<uintptr_t>(value_long >> 32);
-        }
-      }
-      *val = ptr_val;
-      return true;
+    if (m->IsOptimized(sizeof(void*))) {
+      return GetVRegFromOptimizedCode(m, vreg, kind, val);
     } else {
-      const DexFile::CodeItem* code_item = m->GetCodeItem();
-      DCHECK(code_item != nullptr) << PrettyMethod(m);  // Can't be NULL or how would we compile
-                                                        // its instructions?
-      uint32_t* addr = GetVRegAddr(cur_quick_frame_, code_item, frame_info.CoreSpillMask(),
-                                   frame_info.FpSpillMask(), frame_info.FrameSizeInBytes(), vreg);
-      DCHECK(addr != nullptr);
-      *val = *addr;
-      return true;
+      return GetVRegFromQuickCode(m, vreg, kind, val);
     }
   } else {
     DCHECK(cur_shadow_frame_ != nullptr);
@@ -207,6 +175,86 @@
   }
 }
 
+bool StackVisitor::GetVRegFromQuickCode(mirror::ArtMethod* m, uint16_t vreg, VRegKind kind,
+                                        uint32_t* val) const {
+  const void* code_pointer = m->GetQuickOatCodePointer(sizeof(void*));
+  DCHECK(code_pointer != nullptr);
+  const VmapTable vmap_table(m->GetVmapTable(code_pointer, sizeof(void*)));
+  QuickMethodFrameInfo frame_info = m->GetQuickFrameInfo(code_pointer);
+  uint32_t vmap_offset;
+  // TODO: IsInContext stops before spotting floating point registers.
+  if (vmap_table.IsInContext(vreg, kind, &vmap_offset)) {
+    bool is_float = (kind == kFloatVReg) || (kind == kDoubleLoVReg) || (kind == kDoubleHiVReg);
+    uint32_t spill_mask = is_float ? frame_info.FpSpillMask() : frame_info.CoreSpillMask();
+    uint32_t reg = vmap_table.ComputeRegister(spill_mask, vmap_offset, kind);
+    return GetRegisterIfAccessible(reg, kind, val);
+  } else {
+    const DexFile::CodeItem* code_item = m->GetCodeItem();
+    DCHECK(code_item != nullptr) << PrettyMethod(m);  // Can't be NULL or how would we compile
+                                                      // its instructions?
+    *val = *GetVRegAddr(cur_quick_frame_, code_item, frame_info.CoreSpillMask(),
+                        frame_info.FpSpillMask(), frame_info.FrameSizeInBytes(), vreg);
+    return true;
+  }
+}
+
+bool StackVisitor::GetVRegFromOptimizedCode(mirror::ArtMethod* m, uint16_t vreg, VRegKind kind,
+                                            uint32_t* val) const {
+  const void* code_pointer = m->GetQuickOatCodePointer(sizeof(void*));
+  DCHECK(code_pointer != nullptr);
+  uint32_t native_pc_offset = m->NativeQuickPcOffset(cur_quick_frame_pc_);
+  CodeInfo code_info = m->GetOptimizedCodeInfo();
+  StackMap stack_map = code_info.GetStackMapForNativePcOffset(native_pc_offset);
+  const DexFile::CodeItem* code_item = m->GetCodeItem();
+  DCHECK(code_item != nullptr) << PrettyMethod(m);  // Can't be NULL or how would we compile
+                                                    // its instructions?
+  DCHECK_LT(vreg, code_item->registers_size_);
+  DexRegisterMap dex_register_map = code_info.GetDexRegisterMapOf(stack_map,
+                                                                  code_item->registers_size_);
+  DexRegisterMap::LocationKind location_kind = dex_register_map.GetLocationKind(vreg);
+  switch (location_kind) {
+    case DexRegisterMap::kInStack: {
+      const int32_t offset = dex_register_map.GetStackOffsetInBytes(vreg);
+      const uint8_t* addr = reinterpret_cast<const uint8_t*>(cur_quick_frame_) + offset;
+      *val = *reinterpret_cast<const uint32_t*>(addr);
+      return true;
+    }
+    case DexRegisterMap::kInRegister:
+    case DexRegisterMap::kInFpuRegister: {
+      uint32_t reg = dex_register_map.GetMachineRegister(vreg);
+      return GetRegisterIfAccessible(reg, kind, val);
+    }
+    case DexRegisterMap::kConstant:
+      *val = dex_register_map.GetConstant(vreg);
+      return true;
+    case DexRegisterMap::kNone:
+      return false;
+  }
+  UNREACHABLE();
+  return false;
+}
+
+bool StackVisitor::GetRegisterIfAccessible(uint32_t reg, VRegKind kind, uint32_t* val) const {
+  const bool is_float = (kind == kFloatVReg) || (kind == kDoubleLoVReg) || (kind == kDoubleHiVReg);
+  if (!IsAccessibleRegister(reg, is_float)) {
+    return false;
+  }
+  uintptr_t ptr_val = GetRegister(reg, is_float);
+  const bool target64 = Is64BitInstructionSet(kRuntimeISA);
+  if (target64) {
+    const bool wide_lo = (kind == kLongLoVReg) || (kind == kDoubleLoVReg);
+    const bool wide_hi = (kind == kLongHiVReg) || (kind == kDoubleHiVReg);
+    int64_t value_long = static_cast<int64_t>(ptr_val);
+    if (wide_lo) {
+      ptr_val = static_cast<uintptr_t>(Low32Bits(value_long));
+    } else if (wide_hi) {
+      ptr_val = static_cast<uintptr_t>(High32Bits(value_long));
+    }
+  }
+  *val = ptr_val;
+  return true;
+}
+
 bool StackVisitor::GetVRegPair(mirror::ArtMethod* m, uint16_t vreg, VRegKind kind_lo,
                                VRegKind kind_hi, uint64_t* val) const {
   if (kind_lo == kLongLoVReg) {
@@ -215,45 +263,15 @@
     DCHECK_EQ(kind_hi, kDoubleHiVReg);
   } else {
     LOG(FATAL) << "Expected long or double: kind_lo=" << kind_lo << ", kind_hi=" << kind_hi;
+    UNREACHABLE();
   }
   if (cur_quick_frame_ != nullptr) {
     DCHECK(context_ != nullptr);  // You can't reliably read registers without a context.
     DCHECK(m == GetMethod());
-    const void* code_pointer = m->GetQuickOatCodePointer(sizeof(void*));
-    DCHECK(code_pointer != nullptr);
-    const VmapTable vmap_table(m->GetVmapTable(code_pointer, sizeof(void*)));
-    QuickMethodFrameInfo frame_info = m->GetQuickFrameInfo(code_pointer);
-    uint32_t vmap_offset_lo, vmap_offset_hi;
-    // TODO: IsInContext stops before spotting floating point registers.
-    if (vmap_table.IsInContext(vreg, kind_lo, &vmap_offset_lo) &&
-        vmap_table.IsInContext(vreg + 1, kind_hi, &vmap_offset_hi)) {
-      bool is_float = (kind_lo == kDoubleLoVReg);
-      uint32_t spill_mask = is_float ? frame_info.FpSpillMask() : frame_info.CoreSpillMask();
-      uint32_t reg_lo = vmap_table.ComputeRegister(spill_mask, vmap_offset_lo, kind_lo);
-      uint32_t reg_hi = vmap_table.ComputeRegister(spill_mask, vmap_offset_hi, kind_hi);
-      if (!IsAccessibleRegister(reg_lo, is_float) || !IsAccessibleRegister(reg_hi, is_float)) {
-        return false;
-      }
-      uintptr_t ptr_val_lo = GetRegister(reg_lo, is_float);
-      uintptr_t ptr_val_hi = GetRegister(reg_hi, is_float);
-      bool target64 = Is64BitInstructionSet(kRuntimeISA);
-      if (target64) {
-        int64_t value_long_lo = static_cast<int64_t>(ptr_val_lo);
-        int64_t value_long_hi = static_cast<int64_t>(ptr_val_hi);
-        ptr_val_lo = static_cast<uintptr_t>(value_long_lo & 0xFFFFFFFF);
-        ptr_val_hi = static_cast<uintptr_t>(value_long_hi >> 32);
-      }
-      *val = (static_cast<uint64_t>(ptr_val_hi) << 32) | static_cast<uint32_t>(ptr_val_lo);
-      return true;
+    if (m->IsOptimized(sizeof(void*))) {
+      return GetVRegPairFromOptimizedCode(m, vreg, kind_lo, kind_hi, val);
     } else {
-      const DexFile::CodeItem* code_item = m->GetCodeItem();
-      DCHECK(code_item != nullptr) << PrettyMethod(m);  // Can't be NULL or how would we compile
-                                                        // its instructions?
-      uint32_t* addr = GetVRegAddr(cur_quick_frame_, code_item, frame_info.CoreSpillMask(),
-                                   frame_info.FpSpillMask(), frame_info.FrameSizeInBytes(), vreg);
-      DCHECK(addr != nullptr);
-      *val = *reinterpret_cast<uint64_t*>(addr);
-      return true;
+      return GetVRegPairFromQuickCode(m, vreg, kind_lo, kind_hi, val);
     }
   } else {
     DCHECK(cur_shadow_frame_ != nullptr);
@@ -262,61 +280,185 @@
   }
 }
 
+bool StackVisitor::GetVRegPairFromQuickCode(mirror::ArtMethod* m, uint16_t vreg, VRegKind kind_lo,
+                                            VRegKind kind_hi, uint64_t* val) const {
+  const void* code_pointer = m->GetQuickOatCodePointer(sizeof(void*));
+  DCHECK(code_pointer != nullptr);
+  const VmapTable vmap_table(m->GetVmapTable(code_pointer, sizeof(void*)));
+  QuickMethodFrameInfo frame_info = m->GetQuickFrameInfo(code_pointer);
+  uint32_t vmap_offset_lo, vmap_offset_hi;
+  // TODO: IsInContext stops before spotting floating point registers.
+  if (vmap_table.IsInContext(vreg, kind_lo, &vmap_offset_lo) &&
+      vmap_table.IsInContext(vreg + 1, kind_hi, &vmap_offset_hi)) {
+    bool is_float = (kind_lo == kDoubleLoVReg);
+    uint32_t spill_mask = is_float ? frame_info.FpSpillMask() : frame_info.CoreSpillMask();
+    uint32_t reg_lo = vmap_table.ComputeRegister(spill_mask, vmap_offset_lo, kind_lo);
+    uint32_t reg_hi = vmap_table.ComputeRegister(spill_mask, vmap_offset_hi, kind_hi);
+    return GetRegisterPairIfAccessible(reg_lo, reg_hi, kind_lo, val);
+  } else {
+    const DexFile::CodeItem* code_item = m->GetCodeItem();
+    DCHECK(code_item != nullptr) << PrettyMethod(m);  // Can't be NULL or how would we compile
+                                                      // its instructions?
+    uint32_t* addr = GetVRegAddr(cur_quick_frame_, code_item, frame_info.CoreSpillMask(),
+                                 frame_info.FpSpillMask(), frame_info.FrameSizeInBytes(), vreg);
+    *val = *reinterpret_cast<uint64_t*>(addr);
+    return true;
+  }
+}
+
+bool StackVisitor::GetVRegPairFromOptimizedCode(mirror::ArtMethod* m, uint16_t vreg,
+                                                VRegKind kind_lo, VRegKind kind_hi,
+                                                uint64_t* val) const {
+  uint32_t low_32bits;
+  uint32_t high_32bits;
+  bool success = GetVRegFromOptimizedCode(m, vreg, kind_lo, &low_32bits);
+  success &= GetVRegFromOptimizedCode(m, vreg + 1, kind_hi, &high_32bits);
+  if (success) {
+    *val = (static_cast<uint64_t>(high_32bits) << 32) | static_cast<uint64_t>(low_32bits);
+  }
+  return success;
+}
+
+bool StackVisitor::GetRegisterPairIfAccessible(uint32_t reg_lo, uint32_t reg_hi,
+                                               VRegKind kind_lo, uint64_t* val) const {
+  const bool is_float = (kind_lo == kDoubleLoVReg);
+  if (!IsAccessibleRegister(reg_lo, is_float) || !IsAccessibleRegister(reg_hi, is_float)) {
+    return false;
+  }
+  uintptr_t ptr_val_lo = GetRegister(reg_lo, is_float);
+  uintptr_t ptr_val_hi = GetRegister(reg_hi, is_float);
+  bool target64 = Is64BitInstructionSet(kRuntimeISA);
+  if (target64) {
+    int64_t value_long_lo = static_cast<int64_t>(ptr_val_lo);
+    int64_t value_long_hi = static_cast<int64_t>(ptr_val_hi);
+    ptr_val_lo = static_cast<uintptr_t>(Low32Bits(value_long_lo));
+    ptr_val_hi = static_cast<uintptr_t>(High32Bits(value_long_hi));
+  }
+  *val = (static_cast<uint64_t>(ptr_val_hi) << 32) | static_cast<uint32_t>(ptr_val_lo);
+  return true;
+}
+
 bool StackVisitor::SetVReg(mirror::ArtMethod* m, uint16_t vreg, uint32_t new_value,
                            VRegKind kind) {
   if (cur_quick_frame_ != nullptr) {
-    DCHECK(context_ != nullptr);  // You can't reliably write registers without a context.
-    DCHECK(m == GetMethod());
-    const void* code_pointer = m->GetQuickOatCodePointer(sizeof(void*));
-    DCHECK(code_pointer != nullptr);
-    const VmapTable vmap_table(m->GetVmapTable(code_pointer, sizeof(void*)));
-    QuickMethodFrameInfo frame_info = m->GetQuickFrameInfo(code_pointer);
-    uint32_t vmap_offset;
-    // TODO: IsInContext stops before spotting floating point registers.
-    if (vmap_table.IsInContext(vreg, kind, &vmap_offset)) {
-      bool is_float = (kind == kFloatVReg) || (kind == kDoubleLoVReg) || (kind == kDoubleHiVReg);
-      uint32_t spill_mask = is_float ? frame_info.FpSpillMask() : frame_info.CoreSpillMask();
-      const uint32_t reg = vmap_table.ComputeRegister(spill_mask, vmap_offset, kind);
-      if (!IsAccessibleRegister(reg, is_float)) {
-        return false;
+      DCHECK(context_ != nullptr);  // You can't reliably write registers without a context.
+      DCHECK(m == GetMethod());
+      if (m->IsOptimized(sizeof(void*))) {
+        return SetVRegFromOptimizedCode(m, vreg, new_value, kind);
+      } else {
+        return SetVRegFromQuickCode(m, vreg, new_value, kind);
       }
-      bool target64 = Is64BitInstructionSet(kRuntimeISA);
-      // Deal with 32 or 64-bit wide registers in a way that builds on all targets.
-      if (target64) {
-        bool wide_lo = (kind == kLongLoVReg) || (kind == kDoubleLoVReg);
-        bool wide_hi = (kind == kLongHiVReg) || (kind == kDoubleHiVReg);
-        if (wide_lo || wide_hi) {
-          uintptr_t old_reg_val = GetRegister(reg, is_float);
-          uint64_t new_vreg_portion = static_cast<uint64_t>(new_value);
-          uint64_t old_reg_val_as_wide = static_cast<uint64_t>(old_reg_val);
-          uint64_t mask = 0xffffffff;
-          if (wide_lo) {
-            mask = mask << 32;
-          } else {
-            new_vreg_portion = new_vreg_portion << 32;
-          }
-          new_value = static_cast<uintptr_t>((old_reg_val_as_wide & mask) | new_vreg_portion);
-        }
-      }
-      SetRegister(reg, new_value, is_float);
-      return true;
     } else {
-      const DexFile::CodeItem* code_item = m->GetCodeItem();
-      DCHECK(code_item != nullptr) << PrettyMethod(m);  // Can't be NULL or how would we compile
-                                                        // its instructions?
-      uint32_t* addr = GetVRegAddr(cur_quick_frame_, code_item, frame_info.CoreSpillMask(),
-                                   frame_info.FpSpillMask(), frame_info.FrameSizeInBytes(), vreg);
-      DCHECK(addr != nullptr);
-      *addr = new_value;
+      cur_shadow_frame_->SetVReg(vreg, new_value);
       return true;
     }
+}
+
+bool StackVisitor::SetVRegFromQuickCode(mirror::ArtMethod* m, uint16_t vreg, uint32_t new_value,
+                                        VRegKind kind) {
+  DCHECK(context_ != nullptr);  // You can't reliably write registers without a context.
+  DCHECK(m == GetMethod());
+  const void* code_pointer = m->GetQuickOatCodePointer(sizeof(void*));
+  DCHECK(code_pointer != nullptr);
+  const VmapTable vmap_table(m->GetVmapTable(code_pointer, sizeof(void*)));
+  QuickMethodFrameInfo frame_info = m->GetQuickFrameInfo(code_pointer);
+  uint32_t vmap_offset;
+  // TODO: IsInContext stops before spotting floating point registers.
+  if (vmap_table.IsInContext(vreg, kind, &vmap_offset)) {
+    bool is_float = (kind == kFloatVReg) || (kind == kDoubleLoVReg) || (kind == kDoubleHiVReg);
+    uint32_t spill_mask = is_float ? frame_info.FpSpillMask() : frame_info.CoreSpillMask();
+    uint32_t reg = vmap_table.ComputeRegister(spill_mask, vmap_offset, kind);
+    return SetRegisterIfAccessible(reg, new_value, kind);
   } else {
-    DCHECK(cur_shadow_frame_ != nullptr);
-    cur_shadow_frame_->SetVReg(vreg, new_value);
+    const DexFile::CodeItem* code_item = m->GetCodeItem();
+    DCHECK(code_item != nullptr) << PrettyMethod(m);  // Can't be NULL or how would we compile
+                                                      // its instructions?
+    uint32_t* addr = GetVRegAddr(cur_quick_frame_, code_item, frame_info.CoreSpillMask(),
+                                 frame_info.FpSpillMask(), frame_info.FrameSizeInBytes(), vreg);
+    *addr = new_value;
     return true;
   }
 }
 
+bool StackVisitor::SetVRegFromOptimizedCode(mirror::ArtMethod* m, uint16_t vreg, uint32_t new_value,
+                                            VRegKind kind) {
+  const void* code_pointer = m->GetQuickOatCodePointer(sizeof(void*));
+  DCHECK(code_pointer != nullptr);
+  uint32_t native_pc_offset = m->NativeQuickPcOffset(cur_quick_frame_pc_);
+  CodeInfo code_info = m->GetOptimizedCodeInfo();
+  StackMap stack_map = code_info.GetStackMapForNativePcOffset(native_pc_offset);
+  const DexFile::CodeItem* code_item = m->GetCodeItem();
+  DCHECK(code_item != nullptr) << PrettyMethod(m);  // Can't be NULL or how would we compile
+                                                    // its instructions?
+  DCHECK_LT(vreg, code_item->registers_size_);
+  DexRegisterMap dex_register_map = code_info.GetDexRegisterMapOf(stack_map,
+                                                                  code_item->registers_size_);
+  DexRegisterMap::LocationKind location_kind = dex_register_map.GetLocationKind(vreg);
+  uint32_t dex_pc = m->ToDexPc(cur_quick_frame_pc_, false);
+  switch (location_kind) {
+    case DexRegisterMap::kInStack: {
+      const int32_t offset = dex_register_map.GetStackOffsetInBytes(vreg);
+      uint8_t* addr = reinterpret_cast<uint8_t*>(cur_quick_frame_) + offset;
+      *reinterpret_cast<uint32_t*>(addr) = new_value;
+      return true;
+    }
+    case DexRegisterMap::kInRegister:
+    case DexRegisterMap::kInFpuRegister: {
+      uint32_t reg = dex_register_map.GetMachineRegister(vreg);
+      return SetRegisterIfAccessible(reg, new_value, kind);
+    }
+    case DexRegisterMap::kConstant:
+      LOG(ERROR) << StringPrintf("Cannot change value of DEX register v%u used as a constant at "
+                                 "DEX pc 0x%x (native pc 0x%x) of method %s",
+                                 vreg, dex_pc, native_pc_offset,
+                                 PrettyMethod(cur_quick_frame_->AsMirrorPtr()).c_str());
+      return false;
+    case DexRegisterMap::kNone:
+      LOG(ERROR) << StringPrintf("No location for DEX register v%u at DEX pc 0x%x "
+                                 "(native pc 0x%x) of method %s",
+                                 vreg, dex_pc, native_pc_offset,
+                                 PrettyMethod(cur_quick_frame_->AsMirrorPtr()).c_str());
+      return false;
+    default:
+      LOG(FATAL) << StringPrintf("Unknown location for DEX register v%u at DEX pc 0x%x "
+                                 "(native pc 0x%x) of method %s",
+                                 vreg, dex_pc, native_pc_offset,
+                                 PrettyMethod(cur_quick_frame_->AsMirrorPtr()).c_str());
+      UNREACHABLE();
+  }
+}
+
+bool StackVisitor::SetRegisterIfAccessible(uint32_t reg, uint32_t new_value, VRegKind kind) {
+  const bool is_float = (kind == kFloatVReg) || (kind == kDoubleLoVReg) || (kind == kDoubleHiVReg);
+  if (!IsAccessibleRegister(reg, is_float)) {
+    return false;
+  }
+  const bool target64 = Is64BitInstructionSet(kRuntimeISA);
+
+  // Create a new value that can hold both low 32 and high 32 bits, in
+  // case we are running 64 bits.
+  uintptr_t full_new_value = new_value;
+  // Deal with 32 or 64-bit wide registers in a way that builds on all targets.
+  if (target64) {
+    bool wide_lo = (kind == kLongLoVReg) || (kind == kDoubleLoVReg);
+    bool wide_hi = (kind == kLongHiVReg) || (kind == kDoubleHiVReg);
+    if (wide_lo || wide_hi) {
+      uintptr_t old_reg_val = GetRegister(reg, is_float);
+      uint64_t new_vreg_portion = static_cast<uint64_t>(new_value);
+      uint64_t old_reg_val_as_wide = static_cast<uint64_t>(old_reg_val);
+      uint64_t mask = 0xffffffff;
+      if (wide_lo) {
+        mask = mask << 32;
+      } else {
+        new_vreg_portion = new_vreg_portion << 32;
+      }
+      full_new_value = static_cast<uintptr_t>((old_reg_val_as_wide & mask) | new_vreg_portion);
+    }
+  }
+  SetRegister(reg, full_new_value, is_float);
+  return true;
+}
+
 bool StackVisitor::SetVRegPair(mirror::ArtMethod* m, uint16_t vreg, uint64_t new_value,
                                VRegKind kind_lo, VRegKind kind_hi) {
   if (kind_lo == kLongLoVReg) {
@@ -329,49 +471,10 @@
   if (cur_quick_frame_ != nullptr) {
     DCHECK(context_ != nullptr);  // You can't reliably write registers without a context.
     DCHECK(m == GetMethod());
-    const void* code_pointer = m->GetQuickOatCodePointer(sizeof(void*));
-    DCHECK(code_pointer != nullptr);
-    const VmapTable vmap_table(m->GetVmapTable(code_pointer, sizeof(void*)));
-    QuickMethodFrameInfo frame_info = m->GetQuickFrameInfo(code_pointer);
-    uint32_t vmap_offset_lo, vmap_offset_hi;
-    // TODO: IsInContext stops before spotting floating point registers.
-    if (vmap_table.IsInContext(vreg, kind_lo, &vmap_offset_lo) &&
-        vmap_table.IsInContext(vreg + 1, kind_hi, &vmap_offset_hi)) {
-      bool is_float = (kind_lo == kDoubleLoVReg);
-      uint32_t spill_mask = is_float ? frame_info.FpSpillMask() : frame_info.CoreSpillMask();
-      uint32_t reg_lo = vmap_table.ComputeRegister(spill_mask, vmap_offset_lo, kind_lo);
-      uint32_t reg_hi = vmap_table.ComputeRegister(spill_mask, vmap_offset_hi, kind_hi);
-      if (!IsAccessibleRegister(reg_lo, is_float) || !IsAccessibleRegister(reg_hi, is_float)) {
-        return false;
-      }
-      uintptr_t new_value_lo = static_cast<uintptr_t>(new_value & 0xFFFFFFFF);
-      uintptr_t new_value_hi = static_cast<uintptr_t>(new_value >> 32);
-      bool target64 = Is64BitInstructionSet(kRuntimeISA);
-      // Deal with 32 or 64-bit wide registers in a way that builds on all targets.
-      if (target64) {
-        uintptr_t old_reg_val_lo = GetRegister(reg_lo, is_float);
-        uintptr_t old_reg_val_hi = GetRegister(reg_hi, is_float);
-        uint64_t new_vreg_portion_lo = static_cast<uint64_t>(new_value_lo);
-        uint64_t new_vreg_portion_hi = static_cast<uint64_t>(new_value_hi) << 32;
-        uint64_t old_reg_val_lo_as_wide = static_cast<uint64_t>(old_reg_val_lo);
-        uint64_t old_reg_val_hi_as_wide = static_cast<uint64_t>(old_reg_val_hi);
-        uint64_t mask_lo = static_cast<uint64_t>(0xffffffff) << 32;
-        uint64_t mask_hi = 0xffffffff;
-        new_value_lo = static_cast<uintptr_t>((old_reg_val_lo_as_wide & mask_lo) | new_vreg_portion_lo);
-        new_value_hi = static_cast<uintptr_t>((old_reg_val_hi_as_wide & mask_hi) | new_vreg_portion_hi);
-      }
-      SetRegister(reg_lo, new_value_lo, is_float);
-      SetRegister(reg_hi, new_value_hi, is_float);
-      return true;
+    if (m->IsOptimized(sizeof(void*))) {
+      return SetVRegPairFromOptimizedCode(m, vreg, new_value, kind_lo, kind_hi);
     } else {
-      const DexFile::CodeItem* code_item = m->GetCodeItem();
-      DCHECK(code_item != nullptr) << PrettyMethod(m);  // Can't be NULL or how would we compile
-                                                        // its instructions?
-      uint32_t* addr = GetVRegAddr(cur_quick_frame_, code_item, frame_info.CoreSpillMask(),
-                                   frame_info.FpSpillMask(), frame_info.FrameSizeInBytes(), vreg);
-      DCHECK(addr != nullptr);
-      *reinterpret_cast<uint64_t*>(addr) = new_value;
-      return true;
+      return SetVRegPairFromQuickCode(m, vreg, new_value, kind_lo, kind_hi);
     }
   } else {
     DCHECK(cur_shadow_frame_ != nullptr);
@@ -380,6 +483,60 @@
   }
 }
 
+bool StackVisitor::SetVRegPairFromQuickCode(mirror::ArtMethod* m, uint16_t vreg, uint64_t new_value,
+                                            VRegKind kind_lo, VRegKind kind_hi) {
+  const void* code_pointer = m->GetQuickOatCodePointer(sizeof(void*));
+  DCHECK(code_pointer != nullptr);
+  const VmapTable vmap_table(m->GetVmapTable(code_pointer, sizeof(void*)));
+  QuickMethodFrameInfo frame_info = m->GetQuickFrameInfo(code_pointer);
+  uint32_t vmap_offset_lo, vmap_offset_hi;
+  // TODO: IsInContext stops before spotting floating point registers.
+  if (vmap_table.IsInContext(vreg, kind_lo, &vmap_offset_lo) &&
+      vmap_table.IsInContext(vreg + 1, kind_hi, &vmap_offset_hi)) {
+    bool is_float = (kind_lo == kDoubleLoVReg);
+    uint32_t spill_mask = is_float ? frame_info.FpSpillMask() : frame_info.CoreSpillMask();
+    uint32_t reg_lo = vmap_table.ComputeRegister(spill_mask, vmap_offset_lo, kind_lo);
+    uint32_t reg_hi = vmap_table.ComputeRegister(spill_mask, vmap_offset_hi, kind_hi);
+    return SetRegisterPairIfAccessible(reg_lo, reg_hi, new_value, is_float);
+  } else {
+    const DexFile::CodeItem* code_item = m->GetCodeItem();
+    DCHECK(code_item != nullptr) << PrettyMethod(m);  // Can't be NULL or how would we compile
+                                                      // its instructions?
+    uint32_t* addr = GetVRegAddr(cur_quick_frame_, code_item, frame_info.CoreSpillMask(),
+                                 frame_info.FpSpillMask(), frame_info.FrameSizeInBytes(), vreg);
+    *reinterpret_cast<uint64_t*>(addr) = new_value;
+    return true;
+  }
+}
+
+bool StackVisitor::SetVRegPairFromOptimizedCode(mirror::ArtMethod* m, uint16_t vreg, uint64_t new_value,
+                                                VRegKind kind_lo, VRegKind kind_hi) {
+  uint32_t low_32bits = Low32Bits(new_value);
+  uint32_t high_32bits = High32Bits(new_value);
+  bool success = SetVRegFromOptimizedCode(m, vreg, low_32bits, kind_lo);
+  success &= SetVRegFromOptimizedCode(m, vreg + 1, high_32bits, kind_hi);
+  return success;
+}
+
+bool StackVisitor::SetRegisterPairIfAccessible(uint32_t reg_lo, uint32_t reg_hi,
+                                               uint64_t new_value, bool is_float) {
+  if (!IsAccessibleRegister(reg_lo, is_float) || !IsAccessibleRegister(reg_hi, is_float)) {
+    return false;
+  }
+  uintptr_t new_value_lo = static_cast<uintptr_t>(new_value & 0xFFFFFFFF);
+  uintptr_t new_value_hi = static_cast<uintptr_t>(new_value >> 32);
+  bool target64 = Is64BitInstructionSet(kRuntimeISA);
+  // Deal with 32 or 64-bit wide registers in a way that builds on all targets.
+  if (target64) {
+    DCHECK_EQ(reg_lo, reg_hi);
+    SetRegister(reg_lo, new_value, is_float);
+  } else {
+    SetRegister(reg_lo, new_value_lo, is_float);
+    SetRegister(reg_hi, new_value_hi, is_float);
+  }
+  return true;
+}
+
 bool StackVisitor::IsAccessibleGPR(uint32_t reg) const {
   DCHECK(context_ != nullptr);
   return context_->IsAccessibleGPR(reg);
diff --git a/runtime/stack.h b/runtime/stack.h
index 5a86ca1..b495f03 100644
--- a/runtime/stack.h
+++ b/runtime/stack.h
@@ -492,7 +492,8 @@
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     uint32_t val;
     bool success = GetVReg(m, vreg, kind, &val);
-    CHECK(success) << "Failed to read vreg " << vreg << " of kind " << kind;
+    CHECK(success) << "Failed to read v" << vreg << " of kind " << kind << " in method "
+                   << PrettyMethod(m);
     return val;
   }
 
@@ -505,7 +506,8 @@
     uint64_t val;
     bool success = GetVRegPair(m, vreg, kind_lo, kind_hi, &val);
     CHECK(success) << "Failed to read vreg pair " << vreg
-                   << " of kind [" << kind_lo << "," << kind_hi << "]";
+                   << " of kind [" << kind_lo << "," << kind_hi << "] in method "
+                   << PrettyMethod(m);
     return val;
   }
 
@@ -673,6 +675,45 @@
   uintptr_t GetFPR(uint32_t reg) const;
   void SetFPR(uint32_t reg, uintptr_t value);
 
+  bool GetVRegFromQuickCode(mirror::ArtMethod* m, uint16_t vreg, VRegKind kind,
+                            uint32_t* val) const
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  bool GetVRegFromOptimizedCode(mirror::ArtMethod* m, uint16_t vreg, VRegKind kind,
+                                uint32_t* val) const
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  bool GetRegisterIfAccessible(uint32_t reg, VRegKind kind, uint32_t* val) const
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
+  bool GetVRegPairFromQuickCode(mirror::ArtMethod* m, uint16_t vreg, VRegKind kind_lo,
+                                VRegKind kind_hi, uint64_t* val) const
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  bool GetVRegPairFromOptimizedCode(mirror::ArtMethod* m, uint16_t vreg,
+                                    VRegKind kind_lo, VRegKind kind_hi,
+                                    uint64_t* val) const
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  bool GetRegisterPairIfAccessible(uint32_t reg_lo, uint32_t reg_hi, VRegKind kind_lo,
+                                   uint64_t* val) const
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
+  bool SetVRegFromQuickCode(mirror::ArtMethod* m, uint16_t vreg, uint32_t new_value,
+                            VRegKind kind)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  bool SetVRegFromOptimizedCode(mirror::ArtMethod* m, uint16_t vreg, uint32_t new_value,
+                                VRegKind kind)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  bool SetRegisterIfAccessible(uint32_t reg, uint32_t new_value, VRegKind kind)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
+  bool SetVRegPairFromQuickCode(mirror::ArtMethod* m, uint16_t vreg, uint64_t new_value,
+                                VRegKind kind_lo, VRegKind kind_hi)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  bool SetVRegPairFromOptimizedCode(mirror::ArtMethod* m, uint16_t vreg, uint64_t new_value,
+                                    VRegKind kind_lo, VRegKind kind_hi)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  bool SetRegisterPairIfAccessible(uint32_t reg_lo, uint32_t reg_hi, uint64_t new_value,
+                                   bool is_float)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
   void SanityCheckFrame() const SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   Thread* const thread_;
diff --git a/runtime/stack_map.h b/runtime/stack_map.h
index fd22361..ec37699 100644
--- a/runtime/stack_map.h
+++ b/runtime/stack_map.h
@@ -104,10 +104,9 @@
         return "in fpu register";
       case kConstant:
         return "as constant";
-      default:
-        LOG(FATAL) << "Invalid location kind " << static_cast<int>(kind);
-        return nullptr;
     }
+    UNREACHABLE();
+    return nullptr;
   }
 
   LocationKind GetLocationKind(uint16_t register_index) const {
@@ -126,6 +125,23 @@
         kFixedSize + sizeof(LocationKind) + register_index * SingleEntrySize());
   }
 
+  int32_t GetStackOffsetInBytes(uint16_t register_index) const {
+    DCHECK(GetLocationKind(register_index) == kInStack);
+    // We currently encode the offset in bytes.
+    return GetValue(register_index);
+  }
+
+  int32_t GetConstant(uint16_t register_index) const {
+    DCHECK(GetLocationKind(register_index) == kConstant);
+    return GetValue(register_index);
+  }
+
+  int32_t GetMachineRegister(uint16_t register_index) const {
+    DCHECK(GetLocationKind(register_index) == kInRegister
+        || GetLocationKind(register_index) == kInFpuRegister);
+    return GetValue(register_index);
+  }
+
   static size_t SingleEntrySize() {
     return sizeof(LocationKind) + sizeof(int32_t);
   }
diff --git a/runtime/thread.cc b/runtime/thread.cc
index 3b48f49..79d0066 100644
--- a/runtime/thread.cc
+++ b/runtime/thread.cc
@@ -541,7 +541,7 @@
   // Set stack_end_ to the bottom of the stack saving space of stack overflows
 
   Runtime* runtime = Runtime::Current();
-  bool implicit_stack_check = !runtime->ExplicitStackOverflowChecks() && !runtime->IsCompiler();
+  bool implicit_stack_check = !runtime->ExplicitStackOverflowChecks() && !runtime->IsAotCompiler();
   ResetDefaultStackEnd();
 
   // Install the protected region if we are doing implicit overflow checks.
diff --git a/runtime/thread_list.cc b/runtime/thread_list.cc
index d0f014a..83c5ffb 100644
--- a/runtime/thread_list.cc
+++ b/runtime/thread_list.cc
@@ -1107,9 +1107,11 @@
     Locks::thread_list_lock_->ExclusiveLock(self);
     bool removed = true;
     if (!Contains(self)) {
+      std::string thread_name;
+      self->GetThreadName(thread_name);
       std::ostringstream os;
       DumpNativeStack(os, GetTid(), "  native: ", nullptr);
-      LOG(ERROR) << "Request to unregister unattached thread\n" << os.str();
+      LOG(ERROR) << "Request to unregister unattached thread " << thread_name << "\n" << os.str();
     } else {
       Locks::thread_suspend_count_lock_->ExclusiveLock(self);
       if (!self->IsSuspended()) {
diff --git a/runtime/thread_pool.cc b/runtime/thread_pool.cc
index 587eb32..2a82285 100644
--- a/runtime/thread_pool.cc
+++ b/runtime/thread_pool.cc
@@ -31,7 +31,7 @@
       name_(name) {
   std::string error_msg;
   stack_.reset(MemMap::MapAnonymous(name.c_str(), nullptr, stack_size, PROT_READ | PROT_WRITE,
-                                    false, &error_msg));
+                                    false, false, &error_msg));
   CHECK(stack_.get() != nullptr) << error_msg;
   const char* reason = "new thread pool worker thread";
   pthread_attr_t attr;
diff --git a/runtime/trace.cc b/runtime/trace.cc
index 0950abeb..93b3877 100644
--- a/runtime/trace.cc
+++ b/runtime/trace.cc
@@ -298,7 +298,7 @@
   intptr_t interval_us = reinterpret_cast<intptr_t>(arg);
   CHECK_GE(interval_us, 0);
   CHECK(runtime->AttachCurrentThread("Sampling Profiler", true, runtime->GetSystemThreadGroup(),
-                                     !runtime->IsCompiler()));
+                                     !runtime->IsAotCompiler()));
 
   while (true) {
     usleep(interval_us);
@@ -627,6 +627,12 @@
   LOG(ERROR) << "Unexpected exception caught event in tracing";
 }
 
+void Trace::BackwardBranch(Thread* /*thread*/, mirror::ArtMethod* method,
+                           int32_t /*dex_pc_offset*/)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+  LOG(ERROR) << "Unexpected backward branch event in tracing" << PrettyMethod(method);
+}
+
 void Trace::ReadClocks(Thread* thread, uint32_t* thread_clock_diff, uint32_t* wall_clock_diff) {
   if (UseThreadCpuClock()) {
     uint64_t clock_base = thread->GetTraceClockBase();
diff --git a/runtime/trace.h b/runtime/trace.h
index ead1c29..9ba30d5 100644
--- a/runtime/trace.h
+++ b/runtime/trace.h
@@ -99,7 +99,8 @@
                        mirror::ArtMethod* catch_method, uint32_t catch_dex_pc,
                        mirror::Throwable* exception_object)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) OVERRIDE;
-
+  void BackwardBranch(Thread* thread, mirror::ArtMethod* method, int32_t dex_pc_offset)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) OVERRIDE;
   // Reuse an old stack trace if it exists, otherwise allocate a new one.
   static std::vector<mirror::ArtMethod*>* AllocStackTrace();
   // Clear and store an old stack trace for later use.
diff --git a/runtime/transaction.cc b/runtime/transaction.cc
index 7e2e0a6..c0fd7a5 100644
--- a/runtime/transaction.cc
+++ b/runtime/transaction.cc
@@ -32,7 +32,7 @@
 
 Transaction::Transaction()
   : log_lock_("transaction log lock", kTransactionLogLock), aborted_(false) {
-  CHECK(Runtime::Current()->IsCompiler());
+  CHECK(Runtime::Current()->IsAotCompiler());
 }
 
 Transaction::~Transaction() {
@@ -70,8 +70,10 @@
   }
 }
 
-void Transaction::ThrowInternalError(Thread* self) {
-  DCHECK(IsAborted());
+void Transaction::ThrowInternalError(Thread* self, bool rethrow) {
+  if (kIsDebugBuild && rethrow) {
+    CHECK(IsAborted()) << "Rethrow InternalError while transaction is not aborted";
+  }
   std::string abort_msg(GetAbortMessage());
   self->ThrowNewException(self->GetCurrentLocationForThrow(), "Ljava/lang/InternalError;",
                           abort_msg.c_str());
diff --git a/runtime/transaction.h b/runtime/transaction.h
index be614f9..e1b93c9 100644
--- a/runtime/transaction.h
+++ b/runtime/transaction.h
@@ -45,7 +45,7 @@
   void Abort(const std::string& abort_message)
       LOCKS_EXCLUDED(log_lock_)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
-  void ThrowInternalError(Thread* self)
+  void ThrowInternalError(Thread* self, bool rethrow)
       LOCKS_EXCLUDED(log_lock_)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
   bool IsAborted() LOCKS_EXCLUDED(log_lock_);
diff --git a/runtime/utils.cc b/runtime/utils.cc
index 85c9340..851eceb 100644
--- a/runtime/utils.cc
+++ b/runtime/utils.cc
@@ -133,14 +133,14 @@
 }
 
 bool ReadFileToString(const std::string& file_name, std::string* result) {
-  std::unique_ptr<File> file(new File);
-  if (!file->Open(file_name, O_RDONLY)) {
+  File file;
+  if (!file.Open(file_name, O_RDONLY)) {
     return false;
   }
 
   std::vector<char> buf(8 * KB);
   while (true) {
-    int64_t n = TEMP_FAILURE_RETRY(read(file->Fd(), &buf[0], buf.size()));
+    int64_t n = TEMP_FAILURE_RETRY(read(file.Fd(), &buf[0], buf.size()));
     if (n == -1) {
       return false;
     }
@@ -151,6 +151,59 @@
   }
 }
 
+bool PrintFileToLog(const std::string& file_name, LogSeverity level) {
+  File file;
+  if (!file.Open(file_name, O_RDONLY)) {
+    return false;
+  }
+
+  constexpr size_t kBufSize = 256;  // Small buffer. Avoid stack overflow and stack size warnings.
+  char buf[kBufSize + 1];           // +1 for terminator.
+  size_t filled_to = 0;
+  while (true) {
+    DCHECK_LT(filled_to, kBufSize);
+    int64_t n = TEMP_FAILURE_RETRY(read(file.Fd(), &buf[filled_to], kBufSize - filled_to));
+    if (n <= 0) {
+      // Print the rest of the buffer, if it exists.
+      if (filled_to > 0) {
+        buf[filled_to] = 0;
+        LOG(level) << buf;
+      }
+      return n == 0;
+    }
+    // Scan for '\n'.
+    size_t i = filled_to;
+    bool found_newline = false;
+    for (; i < filled_to + n; ++i) {
+      if (buf[i] == '\n') {
+        // Found a line break, that's something to print now.
+        buf[i] = 0;
+        LOG(level) << buf;
+        // Copy the rest to the front.
+        if (i + 1 < filled_to + n) {
+          memmove(&buf[0], &buf[i + 1], filled_to + n - i - 1);
+          filled_to = filled_to + n - i - 1;
+        } else {
+          filled_to = 0;
+        }
+        found_newline = true;
+        break;
+      }
+    }
+    if (found_newline) {
+      continue;
+    } else {
+      filled_to += n;
+      // Check if we must flush now.
+      if (filled_to == kBufSize) {
+        buf[kBufSize] = 0;
+        LOG(level) << buf;
+        filled_to = 0;
+      }
+    }
+  }
+}
+
 std::string GetIsoDate() {
   time_t now = time(NULL);
   tm tmbuf;
@@ -1211,14 +1264,6 @@
     return;
   }
 
-#if !defined(HAVE_ANDROID_OS)
-  if (GetTid() != tid) {
-    // TODO: dumping of other threads is disabled to avoid crashes during stress testing.
-    //       b/15446488.
-    return;
-  }
-#endif
-
   std::unique_ptr<Backtrace> backtrace(Backtrace::Create(BACKTRACE_CURRENT_PROCESS, tid));
   if (!backtrace->Unwind(0, reinterpret_cast<ucontext*>(ucontext_ptr))) {
     os << prefix << "(backtrace::Unwind failed for thread " << tid << ")\n";
@@ -1262,9 +1307,9 @@
           os << "+" << it->func_offset;
         }
         try_addr2line = true;
-      } else if (current_method != nullptr &&
-                 Locks::mutator_lock_->IsSharedHeld(Thread::Current()) &&
-                 current_method->PcIsWithinQuickCode(it->pc)) {
+      } else if (
+          current_method != nullptr && Locks::mutator_lock_->IsSharedHeld(Thread::Current()) &&
+          current_method->PcIsWithinQuickCode(it->pc)) {
         const void* start_of_code = current_method->GetEntryPointFromQuickCompiledCode();
         os << JniLongName(current_method) << "+"
            << (it->pc - reinterpret_cast<uintptr_t>(start_of_code));
@@ -1466,14 +1511,16 @@
 std::string DexFilenameToOdexFilename(const std::string& location, const InstructionSet isa) {
   // location = /foo/bar/baz.jar
   // odex_location = /foo/bar/<isa>/baz.odex
-
-  CHECK_GE(location.size(), 4U) << location;  // must be at least .123
   std::string odex_location(location);
   InsertIsaDirectory(isa, &odex_location);
-  size_t dot_index = odex_location.size() - 3 - 1;  // 3=dex or zip or apk
-  CHECK_EQ('.', odex_location[dot_index]) << location;
+  size_t dot_index = odex_location.rfind('.');
+
+  // The location must have an extension, otherwise it's not clear what we
+  // should return.
+  CHECK_NE(dot_index, std::string::npos) << odex_location;
+  CHECK_EQ(std::string::npos, odex_location.find('/', dot_index)) << odex_location;
+
   odex_location.resize(dot_index + 1);
-  CHECK_EQ('.', odex_location[odex_location.size()-1]) << location << " " << odex_location;
   odex_location += "odex";
   return odex_location;
 }
diff --git a/runtime/utils.h b/runtime/utils.h
index 3191e7d..9d04d35 100644
--- a/runtime/utils.h
+++ b/runtime/utils.h
@@ -410,6 +410,7 @@
     SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
 bool ReadFileToString(const std::string& file_name, std::string* result);
+bool PrintFileToLog(const std::string& file_name, LogSeverity level);
 
 // Returns the current date in ISO yyyy-mm-dd hh:mm:ss format.
 std::string GetIsoDate();
@@ -515,8 +516,9 @@
 // Returns the system location for an image
 std::string GetSystemImageFilename(const char* location, InstructionSet isa);
 
-// Returns an .odex file name next adjacent to the dex location.
+// Returns an .odex file name adjacent to the dex location.
 // For example, for "/foo/bar/baz.jar", return "/foo/bar/<isa>/baz.odex".
+// The dex location must include a directory component and have an extension.
 // Note: does not support multidex location strings.
 std::string DexFilenameToOdexFilename(const std::string& location, InstructionSet isa);
 
diff --git a/runtime/utils_test.cc b/runtime/utils_test.cc
index a3dd13c..5465762 100644
--- a/runtime/utils_test.cc
+++ b/runtime/utils_test.cc
@@ -374,6 +374,8 @@
 TEST_F(UtilsTest, DexFilenameToOdexFilename) {
   EXPECT_STREQ("/foo/bar/arm/baz.odex",
                DexFilenameToOdexFilename("/foo/bar/baz.jar", kArm).c_str());
+  EXPECT_STREQ("/foo/bar/arm/baz.odex",
+               DexFilenameToOdexFilename("/foo/bar/baz.funnyext", kArm).c_str());
 }
 
 TEST_F(UtilsTest, ExecSuccess) {
diff --git a/runtime/verifier/method_verifier.cc b/runtime/verifier/method_verifier.cc
index 474a066..87a29ed 100644
--- a/runtime/verifier/method_verifier.cc
+++ b/runtime/verifier/method_verifier.cc
@@ -24,6 +24,7 @@
 #include "compiler_callbacks.h"
 #include "dex_file-inl.h"
 #include "dex_instruction-inl.h"
+#include "dex_instruction_utils.h"
 #include "dex_instruction_visitor.h"
 #include "gc/accounting/card_table-inl.h"
 #include "indenter.h"
@@ -111,6 +112,20 @@
   reg_line->MarkAllRegistersAsConflicts(verifier);
 }
 
+MethodVerifier::FailureKind MethodVerifier::VerifyMethod(
+    mirror::ArtMethod* method, bool allow_soft_failures, std::string* error ATTRIBUTE_UNUSED) {
+  Thread* self = Thread::Current();
+  StackHandleScope<3> hs(self);
+  mirror::Class* klass = method->GetDeclaringClass();
+  auto h_dex_cache(hs.NewHandle(klass->GetDexCache()));
+  auto h_class_loader(hs.NewHandle(klass->GetClassLoader()));
+  auto h_method = hs.NewHandle(method);
+  return VerifyMethod(self, method->GetDexMethodIndex(), method->GetDexFile(), h_dex_cache,
+                      h_class_loader, klass->GetClassDef(), method->GetCodeItem(), h_method,
+                      method->GetAccessFlags(), allow_soft_failures, false);
+}
+
+
 MethodVerifier::FailureKind MethodVerifier::VerifyClass(Thread* self,
                                                         mirror::Class* klass,
                                                         bool allow_soft_failures,
@@ -136,7 +151,7 @@
   }
   if (early_failure) {
     *error = "Verifier rejected class " + PrettyDescriptor(klass) + failure_message;
-    if (Runtime::Current()->IsCompiler()) {
+    if (Runtime::Current()->IsAotCompiler()) {
       ClassReference ref(&dex_file, klass->GetDexClassDefIndex());
       Runtime::Current()->GetCompilerCallbacks()->ClassRejected(ref);
     }
@@ -544,7 +559,7 @@
     case VERIFY_ERROR_ACCESS_METHOD:
     case VERIFY_ERROR_INSTANTIATION:
     case VERIFY_ERROR_CLASS_CHANGE:
-      if (Runtime::Current()->IsCompiler() || !can_load_classes_) {
+      if (Runtime::Current()->IsAotCompiler() || !can_load_classes_) {
         // If we're optimistically running verification at compile time, turn NO_xxx, ACCESS_xxx,
         // class change and instantiation errors into soft verification errors so that we re-verify
         // at runtime. We may fail to find or to agree on access because of not yet available class
@@ -568,7 +583,7 @@
       // Hard verification failures at compile time will still fail at runtime, so the class is
       // marked as rejected to prevent it from being compiled.
     case VERIFY_ERROR_BAD_CLASS_HARD: {
-      if (Runtime::Current()->IsCompiler()) {
+      if (Runtime::Current()->IsAotCompiler()) {
         ClassReference ref(dex_file_, dex_file_->GetIndexForClassDef(*class_def_));
         Runtime::Current()->GetCompilerCallbacks()->ClassRejected(ref);
       }
@@ -844,7 +859,7 @@
       result = false;
       break;
   }
-  if (inst->GetVerifyIsRuntimeOnly() && Runtime::Current()->IsCompiler() && !verify_to_dump_) {
+  if (inst->GetVerifyIsRuntimeOnly() && Runtime::Current()->IsAotCompiler() && !verify_to_dump_) {
     Fail(VERIFY_ERROR_BAD_CLASS_HARD) << "opcode only expected at runtime " << inst->Name();
     result = false;
   }
@@ -2812,8 +2827,8 @@
   }
 
   if (have_pending_hard_failure_) {
-    if (Runtime::Current()->IsCompiler()) {
-      /* When compiling, check that the last failure is a hard failure */
+    if (Runtime::Current()->IsAotCompiler()) {
+      /* When AOT compiling, check that the last failure is a hard failure */
       CHECK_EQ(failures_[failures_.size() - 1], VERIFY_ERROR_BAD_CLASS_HARD);
     }
     /* immediate failure, reject class */
@@ -3941,28 +3956,16 @@
 
 mirror::ArtField* MethodVerifier::GetQuickFieldAccess(const Instruction* inst,
                                                       RegisterLine* reg_line) {
-  DCHECK(inst->Opcode() == Instruction::IGET_QUICK ||
-         inst->Opcode() == Instruction::IGET_WIDE_QUICK ||
-         inst->Opcode() == Instruction::IGET_OBJECT_QUICK ||
-         inst->Opcode() == Instruction::IGET_BOOLEAN_QUICK ||
-         inst->Opcode() == Instruction::IGET_BYTE_QUICK ||
-         inst->Opcode() == Instruction::IGET_CHAR_QUICK ||
-         inst->Opcode() == Instruction::IGET_SHORT_QUICK ||
-         inst->Opcode() == Instruction::IPUT_QUICK ||
-         inst->Opcode() == Instruction::IPUT_WIDE_QUICK ||
-         inst->Opcode() == Instruction::IPUT_OBJECT_QUICK ||
-         inst->Opcode() == Instruction::IPUT_BOOLEAN_QUICK ||
-         inst->Opcode() == Instruction::IPUT_BYTE_QUICK ||
-         inst->Opcode() == Instruction::IPUT_CHAR_QUICK ||
-         inst->Opcode() == Instruction::IPUT_SHORT_QUICK);
+  DCHECK(IsInstructionIGetQuickOrIPutQuick(inst->Opcode())) << inst->Opcode();
   const RegType& object_type = reg_line->GetRegisterType(this, inst->VRegB_22c());
   if (!object_type.HasClass()) {
     VLOG(verifier) << "Failed to get mirror::Class* from '" << object_type << "'";
     return nullptr;
   }
   uint32_t field_offset = static_cast<uint32_t>(inst->VRegC_22c());
-  mirror::ArtField* f = mirror::ArtField::FindInstanceFieldWithOffset(object_type.GetClass(),
-                                                                      field_offset);
+  mirror::ArtField* const f = mirror::ArtField::FindInstanceFieldWithOffset(object_type.GetClass(),
+                                                                            field_offset);
+  DCHECK_EQ(f->GetOffset().Uint32Value(), field_offset);
   if (f == nullptr) {
     VLOG(verifier) << "Failed to find instance field at offset '" << field_offset
                    << "' from '" << PrettyDescriptor(object_type.GetClass()) << "'";
diff --git a/runtime/verifier/method_verifier.h b/runtime/verifier/method_verifier.h
index b83e647..bdd6259 100644
--- a/runtime/verifier/method_verifier.h
+++ b/runtime/verifier/method_verifier.h
@@ -156,6 +156,9 @@
                                              uint32_t method_access_flags)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
+  static FailureKind VerifyMethod(mirror::ArtMethod* method, bool allow_soft_failures,
+                                  std::string* error) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
   uint8_t EncodePcToReferenceMapData() const;
 
   uint32_t DexFileVersion() const {
@@ -239,10 +242,14 @@
   bool HasFailures() const;
   const RegType& ResolveCheckedClass(uint32_t class_idx)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
-  mirror::ArtMethod* GetQuickInvokedMethod(const Instruction* inst,
-                                           RegisterLine* reg_line,
+  // Returns the method of a quick invoke or nullptr if it cannot be found.
+  mirror::ArtMethod* GetQuickInvokedMethod(const Instruction* inst, RegisterLine* reg_line,
                                            bool is_range)
-        SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  // Returns the access field of a quick field access (iget/iput-quick) or nullptr
+  // if it cannot be found.
+  mirror::ArtField* GetQuickFieldAccess(const Instruction* inst, RegisterLine* reg_line)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   // Is the method being verified a constructor?
   bool IsConstructor() const {
@@ -532,11 +539,6 @@
                            bool is_primitive, bool is_static)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
-  // Returns the access field of a quick field access (iget/iput-quick) or nullptr
-  // if it cannot be found.
-  mirror::ArtField* GetQuickFieldAccess(const Instruction* inst, RegisterLine* reg_line)
-      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
-
   template <FieldAccessType kAccType>
   void VerifyQuickFieldAccess(const Instruction* inst, const RegType& insn_type, bool is_primitive)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
diff --git a/runtime/verifier/reg_type_cache.cc b/runtime/verifier/reg_type_cache.cc
index 1dfbe51..c248565 100644
--- a/runtime/verifier/reg_type_cache.cc
+++ b/runtime/verifier/reg_type_cache.cc
@@ -580,8 +580,9 @@
 }
 
 void RegTypeCache::VisitRoots(RootCallback* callback, void* arg) {
-  for (const RegType* entry : entries_) {
-    entry->VisitRoots(callback, arg);
+  // Exclude the static roots that are visited by VisitStaticRoots().
+  for (size_t i = primitive_count_; i < entries_.size(); ++i) {
+    entries_[i]->VisitRoots(callback, arg);
   }
 }
 
diff --git a/runtime/well_known_classes.cc b/runtime/well_known_classes.cc
index e368d2c..78185bf 100644
--- a/runtime/well_known_classes.cc
+++ b/runtime/well_known_classes.cc
@@ -222,7 +222,7 @@
   org_apache_harmony_dalvik_ddmc_DdmServer_broadcast = CacheMethod(env, org_apache_harmony_dalvik_ddmc_DdmServer, true, "broadcast", "(I)V");
   org_apache_harmony_dalvik_ddmc_DdmServer_dispatch = CacheMethod(env, org_apache_harmony_dalvik_ddmc_DdmServer, true, "dispatch", "(I[BII)Lorg/apache/harmony/dalvik/ddmc/Chunk;");
 
-  dalvik_system_DexFile_cookie = CacheField(env, dalvik_system_DexFile, false, "mCookie", "J");
+  dalvik_system_DexFile_cookie = CacheField(env, dalvik_system_DexFile, false, "mCookie", "Ljava/lang/Object;");
   dalvik_system_PathClassLoader_pathList = CacheField(env, dalvik_system_PathClassLoader, false, "pathList", "Ldalvik/system/DexPathList;");
   dalvik_system_DexPathList_dexElements = CacheField(env, dalvik_system_DexPathList, false, "dexElements", "[Ldalvik/system/DexPathList$Element;");
   dalvik_system_DexPathList__Element_dexFile = CacheField(env, dalvik_system_DexPathList__Element, false, "dexFile", "Ldalvik/system/DexFile;");
diff --git a/runtime/zip_archive.cc b/runtime/zip_archive.cc
index 63bfc44..ffab674 100644
--- a/runtime/zip_archive.cc
+++ b/runtime/zip_archive.cc
@@ -57,7 +57,8 @@
   name += zip_filename;
   std::unique_ptr<MemMap> map(MemMap::MapAnonymous(name.c_str(),
                                                    NULL, GetUncompressedLength(),
-                                                   PROT_READ | PROT_WRITE, false, error_msg));
+                                                   PROT_READ | PROT_WRITE, false, false,
+                                                   error_msg));
   if (map.get() == nullptr) {
     DCHECK(!error_msg->empty());
     return nullptr;
diff --git a/test/004-SignalTest/signaltest.cc b/test/004-SignalTest/signaltest.cc
index 31371f6..876d27e 100644
--- a/test/004-SignalTest/signaltest.cc
+++ b/test/004-SignalTest/signaltest.cc
@@ -65,6 +65,8 @@
 #elif defined(__i386__) || defined(__x86_64__)
   struct ucontext *uc = reinterpret_cast<struct ucontext*>(context);
   uc->CTX_EIP += 3;
+#else
+  UNUSED(context);
 #endif
 }
 
diff --git a/test/134-reg-promotion/expected.txt b/test/134-reg-promotion/expected.txt
new file mode 100644
index 0000000..e69de29
--- /dev/null
+++ b/test/134-reg-promotion/expected.txt
diff --git a/test/134-reg-promotion/info.txt b/test/134-reg-promotion/info.txt
new file mode 100644
index 0000000..6eff7eb
--- /dev/null
+++ b/test/134-reg-promotion/info.txt
@@ -0,0 +1,4 @@
+Test that a vreg value that was defined by a const 0 and is used is both ref
+and float operations is flushed to all home location.
+
+See: b/19417710, b/7250540 & b.android.com/147187
diff --git a/test/134-reg-promotion/smali/Test.smali b/test/134-reg-promotion/smali/Test.smali
new file mode 100644
index 0000000..6a35c45
--- /dev/null
+++ b/test/134-reg-promotion/smali/Test.smali
@@ -0,0 +1,38 @@
+#
+# 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.
+
+.class public LTest;
+
+.super Ljava/lang/Object;
+
+.method public static run()V
+   .registers 3
+   new-instance v2, Ljava/lang/String;
+   invoke-direct {v2}, Ljava/lang/String;-><init>()V
+   const/4 v0, 0
+   move v1, v0
+   :start
+   invoke-static {}, LMain;->blowup()V
+   if-ne v1, v0, :end
+   const/4 v2, 1
+   invoke-static {v2}, Ljava/lang/Integer;->toString(I)Ljava/lang/String;
+   move v2, v0
+   # The call makes v2 float type.
+   invoke-static {v2}, Ljava/lang/Float;->isNaN(F)Z
+   const/4 v1, 1
+   goto :start
+   :end
+   return-void
+.end method
diff --git a/test/134-reg-promotion/src/Main.java b/test/134-reg-promotion/src/Main.java
new file mode 100644
index 0000000..d45ec66
--- /dev/null
+++ b/test/134-reg-promotion/src/Main.java
@@ -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.
+ */
+
+import java.lang.reflect.Method;
+
+public class Main {
+    static char [][] holder;
+    static boolean sawOome;
+
+    static void blowup() {
+        try {
+            for (int i = 0; i < holder.length; ++i) {
+                holder[i] = new char[1024 * 1024];
+            }
+        } catch (OutOfMemoryError oome) {
+            sawOome = true;
+        }
+    }
+
+    public static void main(String args[]) throws Exception {
+        Class<?> c = Class.forName("Test");
+        Method m = c.getMethod("run", (Class[]) null);
+        for (int i = 0; i < 10; i++) {
+            holder = new char[128 * 1024][];
+            m.invoke(null, (Object[]) null);
+            holder = null;
+        }
+    }
+}
diff --git a/test/449-checker-bce/src/Main.java b/test/449-checker-bce/src/Main.java
index 9391533..30aa870 100644
--- a/test/449-checker-bce/src/Main.java
+++ b/test/449-checker-bce/src/Main.java
@@ -314,6 +314,29 @@
     array[10] = 1; // Bounds check can't be eliminated.
   }
 
+
+  static byte readData() {
+    return 1;
+  }
+
+  // CHECK-START: void Main.circularBufferProducer() BCE (before)
+  // CHECK: BoundsCheck
+  // CHECK: ArraySet
+
+  // CHECK-START: void Main.circularBufferProducer() BCE (after)
+  // CHECK-NOT: BoundsCheck
+  // CHECK: ArraySet
+
+  static void circularBufferProducer() {
+    byte[] array = new byte[4096];
+    int i = 0;
+    while (true) {
+      array[i & (array.length - 1)] = readData();
+      i++;
+    }
+  }
+
+
   // CHECK-START: void Main.pyramid1(int[]) BCE (before)
   // CHECK: BoundsCheck
   // CHECK: ArraySet
diff --git a/test/454-get-vreg/expected.txt b/test/454-get-vreg/expected.txt
new file mode 100644
index 0000000..e69de29
--- /dev/null
+++ b/test/454-get-vreg/expected.txt
diff --git a/test/454-get-vreg/get_vreg_jni.cc b/test/454-get-vreg/get_vreg_jni.cc
new file mode 100644
index 0000000..937d2fe
--- /dev/null
+++ b/test/454-get-vreg/get_vreg_jni.cc
@@ -0,0 +1,122 @@
+/*
+ * 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 "arch/context.h"
+#include "jni.h"
+#include "mirror/art_method-inl.h"
+#include "scoped_thread_state_change.h"
+#include "stack.h"
+#include "thread.h"
+
+namespace art {
+
+namespace {
+
+class TestVisitor : public StackVisitor {
+ public:
+  TestVisitor(Thread* thread, Context* context, mirror::Object* this_value)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_)
+      : StackVisitor(thread, context), this_value_(this_value), found_method_index_(0) {}
+
+  bool VisitFrame() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+    mirror::ArtMethod* m = GetMethod();
+    std::string m_name(m->GetName());
+
+    if (m_name.compare("testSimpleVReg") == 0) {
+      found_method_index_ = 1;
+      uint32_t value = 0;
+
+      CHECK(GetVReg(m, 0, kIntVReg, &value));
+      CHECK_EQ(value, 42u);
+
+      bool success = GetVReg(m, 1, kIntVReg, &value);
+      if (m->IsOptimized(sizeof(void*))) CHECK(!success);
+
+      success = GetVReg(m, 2, kIntVReg, &value);
+      if (m->IsOptimized(sizeof(void*))) CHECK(!success);
+
+      CHECK(GetVReg(m, 3, kReferenceVReg, &value));
+      CHECK_EQ(reinterpret_cast<mirror::Object*>(value), this_value_);
+
+      CHECK(GetVReg(m, 4, kIntVReg, &value));
+      CHECK_EQ(value, 1u);
+
+      CHECK(GetVReg(m, 5, kFloatVReg, &value));
+      uint32_t cast = bit_cast<float, uint32_t>(1.0f);
+      CHECK_EQ(value, cast);
+
+      CHECK(GetVReg(m, 6, kIntVReg, &value));
+      CHECK_EQ(value, 2u);
+
+      CHECK(GetVReg(m, 7, kIntVReg, &value));
+      CHECK_EQ(value, true);
+
+      CHECK(GetVReg(m, 8, kIntVReg, &value));
+      CHECK_EQ(value, 3u);
+
+      CHECK(GetVReg(m, 9, kIntVReg, &value));
+      CHECK_EQ(value, static_cast<uint32_t>('c'));
+    } else if (m_name.compare("testPairVReg") == 0) {
+      found_method_index_ = 2;
+      uint64_t value = 0;
+      CHECK(GetVRegPair(m, 0, kLongLoVReg, kLongHiVReg, &value));
+      CHECK_EQ(value, 42u);
+
+      bool success = GetVRegPair(m, 2, kLongLoVReg, kLongHiVReg, &value);
+      if (m->IsOptimized(sizeof(void*))) CHECK(!success);
+
+      success = GetVRegPair(m, 4, kLongLoVReg, kLongHiVReg, &value);
+      if (m->IsOptimized(sizeof(void*))) CHECK(!success);
+
+      uint32_t value32 = 0;
+      CHECK(GetVReg(m, 6, kReferenceVReg, &value32));
+      CHECK_EQ(reinterpret_cast<mirror::Object*>(value32), this_value_);
+
+      CHECK(GetVRegPair(m, 7, kLongLoVReg, kLongHiVReg, &value));
+      CHECK_EQ(static_cast<int64_t>(value), std::numeric_limits<int64_t>::min());
+
+      CHECK(GetVRegPair(m, 9, kLongLoVReg, kLongHiVReg, &value));
+      CHECK_EQ(static_cast<int64_t>(value), std::numeric_limits<int64_t>::max());
+
+      CHECK(GetVRegPair(m, 11, kLongLoVReg, kLongHiVReg, &value));
+      CHECK_EQ(value, 0u);
+
+      CHECK(GetVRegPair(m, 13, kDoubleLoVReg, kDoubleHiVReg, &value));
+      uint64_t cast = bit_cast<double, uint64_t>(2.0);
+      CHECK_EQ(value, cast);
+    }
+
+    return true;
+  }
+
+  mirror::Object* this_value_;
+
+  // Value returned to Java to ensure the methods testSimpleVReg and testPairVReg
+  // have been found and tested.
+  jint found_method_index_;
+};
+
+extern "C" JNIEXPORT jint JNICALL Java_Main_doNativeCall(JNIEnv*, jobject value) {
+  ScopedObjectAccess soa(Thread::Current());
+  std::unique_ptr<Context> context(Context::Create());
+  TestVisitor visitor(soa.Self(), context.get(), soa.Decode<mirror::Object*>(value));
+  visitor.WalkStack();
+  return visitor.found_method_index_;
+}
+
+}  // namespace
+
+}  // namespace art
diff --git a/test/454-get-vreg/info.txt b/test/454-get-vreg/info.txt
new file mode 100644
index 0000000..20df0b5
--- /dev/null
+++ b/test/454-get-vreg/info.txt
@@ -0,0 +1 @@
+Tests for inspecting DEX registers in a Java method.
diff --git a/test/454-get-vreg/src/Main.java b/test/454-get-vreg/src/Main.java
new file mode 100644
index 0000000..df07d44
--- /dev/null
+++ b/test/454-get-vreg/src/Main.java
@@ -0,0 +1,53 @@
+/*
+ * 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.
+ */
+
+public class Main {
+  public Main() {
+  }
+
+  int testSimpleVReg(int a, float f, short s, boolean z, byte b, char c) {
+    int e = doCall();
+    int g = doNativeCall();
+    return e + g;
+  }
+
+  long testPairVReg(long a, long b, long c, double e) {
+    long f = doCall();
+    long g = doNativeCall();
+    return f + g;
+  }
+
+  native int doNativeCall();
+
+  int doCall() {
+    return 42;
+  }
+
+  static {
+    System.loadLibrary("arttest");
+  }
+
+  public static void main(String[] args) {
+    Main rm = new Main();
+    if (rm.testSimpleVReg(1, 1.0f, (short)2, true, (byte)3, 'c') != 43) {
+      throw new Error("Expected 43");
+    }
+
+    if (rm.testPairVReg(Long.MIN_VALUE, Long.MAX_VALUE, 0, 2.0) != 44) {
+      throw new Error("Expected 44");
+    }
+  }
+}
diff --git a/test/455-set-vreg/expected.txt b/test/455-set-vreg/expected.txt
new file mode 100644
index 0000000..e69de29
--- /dev/null
+++ b/test/455-set-vreg/expected.txt
diff --git a/test/455-set-vreg/info.txt b/test/455-set-vreg/info.txt
new file mode 100644
index 0000000..e8c57b5
--- /dev/null
+++ b/test/455-set-vreg/info.txt
@@ -0,0 +1 @@
+Tests for setting DEX registers in a Java method.
diff --git a/test/455-set-vreg/set_vreg_jni.cc b/test/455-set-vreg/set_vreg_jni.cc
new file mode 100644
index 0000000..24d7832
--- /dev/null
+++ b/test/455-set-vreg/set_vreg_jni.cc
@@ -0,0 +1,95 @@
+/*
+ * 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 "arch/context.h"
+#include "jni.h"
+#include "mirror/art_method-inl.h"
+#include "scoped_thread_state_change.h"
+#include "stack.h"
+#include "thread.h"
+
+namespace art {
+
+namespace {
+
+class TestVisitor : public StackVisitor {
+ public:
+  TestVisitor(Thread* thread, Context* context, mirror::Object* this_value)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_)
+      : StackVisitor(thread, context), this_value_(this_value) {}
+
+  bool VisitFrame() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+    mirror::ArtMethod* m = GetMethod();
+    std::string m_name(m->GetName());
+
+    if (m_name.compare("testIntVReg") == 0) {
+      uint32_t value = 0;
+      CHECK(GetVReg(m, 1, kReferenceVReg, &value));
+      CHECK_EQ(reinterpret_cast<mirror::Object*>(value), this_value_);
+
+      CHECK(SetVReg(m, 2, 5, kIntVReg));
+      CHECK(SetVReg(m, 3, 4, kIntVReg));
+      CHECK(SetVReg(m, 4, 3, kIntVReg));
+      CHECK(SetVReg(m, 5, 2, kIntVReg));
+      CHECK(SetVReg(m, 6, 1, kIntVReg));
+    } else if (m_name.compare("testLongVReg") == 0) {
+      uint32_t value = 0;
+      CHECK(GetVReg(m, 3, kReferenceVReg, &value));
+      CHECK_EQ(reinterpret_cast<mirror::Object*>(value), this_value_);
+
+      CHECK(SetVRegPair(m, 4, std::numeric_limits<int64_t>::max(), kLongLoVReg, kLongHiVReg));
+      CHECK(SetVRegPair(m, 6, 4, kLongLoVReg, kLongHiVReg));
+      CHECK(SetVRegPair(m, 8, 3, kLongLoVReg, kLongHiVReg));
+      CHECK(SetVRegPair(m, 10, 2, kLongLoVReg, kLongHiVReg));
+      CHECK(SetVRegPair(m, 12, 1, kLongLoVReg, kLongHiVReg));
+    } else if (m_name.compare("testFloatVReg") == 0) {
+      uint32_t value = 0;
+      CHECK(GetVReg(m, 1, kReferenceVReg, &value));
+      CHECK_EQ(reinterpret_cast<mirror::Object*>(value), this_value_);
+
+      CHECK(SetVReg(m, 2, bit_cast<float, uint32_t>(5.0f), kFloatVReg));
+      CHECK(SetVReg(m, 3, bit_cast<float, uint32_t>(4.0f), kFloatVReg));
+      CHECK(SetVReg(m, 4, bit_cast<float, uint32_t>(3.0f), kFloatVReg));
+      CHECK(SetVReg(m, 5, bit_cast<float, uint32_t>(2.0f), kFloatVReg));
+      CHECK(SetVReg(m, 6, bit_cast<float, uint32_t>(1.0f), kFloatVReg));
+    } else if (m_name.compare("testDoubleVReg") == 0) {
+      uint32_t value = 0;
+      CHECK(GetVReg(m, 3, kReferenceVReg, &value));
+      CHECK_EQ(reinterpret_cast<mirror::Object*>(value), this_value_);
+
+      CHECK(SetVRegPair(m, 4, bit_cast<double, uint64_t>(5.0), kDoubleLoVReg, kDoubleHiVReg));
+      CHECK(SetVRegPair(m, 6, bit_cast<double, uint64_t>(4.0), kDoubleLoVReg, kDoubleHiVReg));
+      CHECK(SetVRegPair(m, 8, bit_cast<double, uint64_t>(3.0), kDoubleLoVReg, kDoubleHiVReg));
+      CHECK(SetVRegPair(m, 10, bit_cast<double, uint64_t>(2.0), kDoubleLoVReg, kDoubleHiVReg));
+      CHECK(SetVRegPair(m, 12, bit_cast<double, uint64_t>(1.0), kDoubleLoVReg, kDoubleHiVReg));
+    }
+
+    return true;
+  }
+
+  mirror::Object* this_value_;
+};
+
+extern "C" JNIEXPORT void JNICALL Java_Main_doNativeCallSetVReg(JNIEnv*, jobject value) {
+  ScopedObjectAccess soa(Thread::Current());
+  std::unique_ptr<Context> context(Context::Create());
+  TestVisitor visitor(soa.Self(), context.get(), soa.Decode<mirror::Object*>(value));
+  visitor.WalkStack();
+}
+
+}  // namespace
+
+}  // namespace art
diff --git a/test/455-set-vreg/src/Main.java b/test/455-set-vreg/src/Main.java
new file mode 100644
index 0000000..2172d92
--- /dev/null
+++ b/test/455-set-vreg/src/Main.java
@@ -0,0 +1,73 @@
+/*
+ * 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.
+ */
+
+public class Main {
+  public Main() {
+  }
+
+  int testIntVReg(int a, int b, int c, int d, int e) {
+    doNativeCallSetVReg();
+    return a - b - c - d - e;
+  }
+
+  long testLongVReg(long a, long b, long c, long d, long e) {
+    doNativeCallSetVReg();
+    return a - b - c - d - e;
+  }
+
+  float testFloatVReg(float a, float b, float c, float d, float e) {
+    doNativeCallSetVReg();
+    return a - b - c - d - e;
+  }
+
+  double testDoubleVReg(double a, double b, double c, double d, double e) {
+    doNativeCallSetVReg();
+    return a - b - c - d - e;
+  }
+
+  native void doNativeCallSetVReg();
+
+  static {
+    System.loadLibrary("arttest");
+  }
+
+  public static void main(String[] args) {
+    Main rm = new Main();
+    int intExpected = 5 - 4 - 3 - 2 - 1;
+    int intResult = rm.testIntVReg(0, 0, 0, 0, 0);
+    if (intResult != intExpected) {
+      throw new Error("Expected " + intExpected + ", got " + intResult);
+    }
+
+    long longExpected = Long.MAX_VALUE - 4 - 3 - 2 - 1;
+    long longResult = rm.testLongVReg(0, 0, 0, 0, 0);
+    if (longResult != longExpected) {
+      throw new Error("Expected " + longExpected + ", got " + longResult);
+    }
+
+    float floatExpected = 5.0f - 4.0f - 3.0f - 2.0f - 1.0f;
+    float floatResult = rm.testFloatVReg(0.0f, 0.0f, 0.0f, 0.0f, 0.0f);
+    if (floatResult != floatExpected) {
+      throw new Error("Expected " + floatExpected + ", got " + floatResult);
+    }
+
+    double doubleExpected = 5.0 - 4.0 - 3.0 - 2.0 - 1.0;
+    double doubleResult = rm.testDoubleVReg(0.0, 0.0, 0.0, 0.0, 0.0);
+    if (doubleResult != doubleExpected) {
+      throw new Error("Expected " + doubleExpected + ", got " + doubleResult);
+    }
+  }
+}
diff --git a/test/Android.libarttest.mk b/test/Android.libarttest.mk
index e64df5c..75c5d72 100644
--- a/test/Android.libarttest.mk
+++ b/test/Android.libarttest.mk
@@ -27,7 +27,9 @@
   051-thread/thread_test.cc \
   116-nodex2oat/nodex2oat.cc \
   117-nopatchoat/nopatchoat.cc \
-  118-noimage-dex2oat/noimage-dex2oat.cc
+  118-noimage-dex2oat/noimage-dex2oat.cc \
+  454-get-vreg/get_vreg_jni.cc \
+  455-set-vreg/set_vreg_jni.cc
 
 ART_TARGET_LIBARTTEST_$(ART_PHONY_TEST_TARGET_SUFFIX) += $(ART_TARGET_TEST_OUT)/$(TARGET_ARCH)/libarttest.so
 ifdef TARGET_2ND_ARCH
diff --git a/test/Android.run-test.mk b/test/Android.run-test.mk
index 8e4b46b..10c422e 100644
--- a/test/Android.run-test.mk
+++ b/test/Android.run-test.mk
@@ -84,6 +84,9 @@
 ifeq ($(ART_TEST_INTERPRETER),true)
   COMPILER_TYPES += interpreter
 endif
+ifeq ($(ART_TEST_JIT),true)
+  COMPILER_TYPES += jit
+endif
 ifeq ($(ART_TEST_OPTIMIZING),true)
   COMPILER_TYPES += optimizing
 endif
@@ -301,6 +304,8 @@
   118-noimage-dex2oat \
   119-noimage-patchoat \
   131-structural-change \
+  454-get-vreg \
+  455-set-vreg \
 
 ifneq (,$(filter ndebug,$(RUN_TYPES)))
   ART_TEST_KNOWN_BROKEN += $(call all-run-test-names,$(TARGET_TYPES),ndebug,$(PREBUILD_TYPES), \
@@ -434,7 +439,7 @@
 
 # Create a rule to build and run a tests following the form:
 # test-art-{1: host or target}-run-test-{2: debug ndebug}-{3: prebuild no-prebuild no-dex2oat}-
-#    {4: interpreter default optimizing}-{5: relocate no-relocate relocate-no-patchoat}-
+#    {4: interpreter default optimizing jit}-{5: relocate no-relocate relocate-no-patchoat}-
 #    {6: trace or no-trace}-{7: gcstress gcverify cms}-{8: forcecopy checkjni jni}-
 #    {9: no-image image picimage}-{10: pictest nopictest}-{11: test name}{12: 32 or 64}
 define define-test-art-run-test
@@ -497,7 +502,12 @@
         test_groups += ART_RUN_TEST_$$(uc_host_or_target)_DEFAULT_RULES
         run_test_options += --quick
       else
-        $$(error found $(4) expected $(COMPILER_TYPES))
+        ifeq ($(4),jit)
+          test_groups += ART_RUN_TEST_$$(uc_host_or_target)_JIT_RULES
+          run_test_options += --jit
+        else
+          $$(error found $(4) expected $(COMPILER_TYPES))
+        endif
       endif
     endif
   endif
@@ -561,32 +571,38 @@
       endif
     endif
   endif
+  ifeq ($(4),jit)
+    # Use interpreter image for JIT.
+    image_suffix := interpreter
+  else
+    image_suffix := $(4)
+  endif
   ifeq ($(9),no-image)
     test_groups += ART_RUN_TEST_$$(uc_host_or_target)_NO_IMAGE_RULES
     run_test_options += --no-image
     # Add the core dependency. This is required for pre-building.
     ifeq ($(1),host)
-      prereq_rule += $(HOST_CORE_IMAGE_$(4)_no-pic_$(12))
+      prereq_rule += $$(HOST_CORE_IMAGE_$$(image_suffix)_no-pic_$(12))
     else
-      prereq_rule += $(TARGET_CORE_IMAGE_$(4)_no-pic_$(12))
+      prereq_rule += $$(TARGET_CORE_IMAGE_$$(image_suffix)_no-pic_$(12))
     endif
   else
     ifeq ($(9),image)
       test_groups += ART_RUN_TEST_$$(uc_host_or_target)_IMAGE_RULES
       # Add the core dependency.
       ifeq ($(1),host)
-        prereq_rule += $(HOST_CORE_IMAGE_$(4)_no-pic_$(12))
+        prereq_rule += $$(HOST_CORE_IMAGE_$$(image_suffix)_no-pic_$(12))
       else
-        prereq_rule += $(TARGET_CORE_IMAGE_$(4)_no-pic_$(12))
+        prereq_rule += $$(TARGET_CORE_IMAGE_$$(image_suffix)_no-pic_$(12))
       endif
     else
       ifeq ($(9),picimage)
         test_groups += ART_RUN_TEST_$$(uc_host_or_target)_PICIMAGE_RULES
         run_test_options += --pic-image
         ifeq ($(1),host)
-          prereq_rule += $(HOST_CORE_IMAGE_$(4)_pic_$(12))
+          prereq_rule += $$(HOST_CORE_IMAGE_$$(image_suffix)_pic_$(12))
         else
-          prereq_rule += $(TARGET_CORE_IMAGE_$(4)_pic_$(12))
+          prereq_rule += $$(TARGET_CORE_IMAGE_$$(image_suffix)_pic_$(12))
         endif
       else
         $$(error found $(9) expected $(IMAGE_TYPES))
diff --git a/test/etc/run-test-jar b/test/etc/run-test-jar
index f64756b..7a2ad1c 100755
--- a/test/etc/run-test-jar
+++ b/test/etc/run-test-jar
@@ -26,6 +26,7 @@
 HAVE_IMAGE="y"
 HOST="n"
 INTERPRETER="n"
+JIT="n"
 INVOKE_WITH=""
 ISA=x86
 LIBRARY_DIRECTORY="lib"
@@ -127,6 +128,9 @@
     elif [ "x$1" = "x--interpreter" ]; then
         INTERPRETER="y"
         shift
+    elif [ "x$1" = "x--jit" ]; then
+        JIT="y"
+        shift
     elif [ "x$1" = "x--jvm" ]; then
         USE_JVM="y"
         shift
@@ -260,6 +264,16 @@
     fi
 fi
 
+if [ "$JIT" = "y" ]; then
+    INT_OPTS="-Xjit"
+    if [ "$VERIFY" = "y" ] ; then
+      COMPILE_FLAGS="${COMPILE_FLAGS} --compiler-filter=interpret-only"
+    else
+      COMPILE_FLAGS="${COMPILE_FLAGS} --compiler-filter=verify-none"
+      DEX_VERIFY="${DEX_VERIFY} -Xverify:none"
+    fi
+fi
+
 JNI_OPTS="-Xjnigreflimit:512 -Xcheck:jni"
 
 if [ "$RELOCATE" = "y" ]; then
@@ -347,8 +361,8 @@
              export ANDROID_DATA=$DEX_LOCATION && \
              export DEX_LOCATION=$DEX_LOCATION && \
              export ANDROID_ROOT=$ANDROID_ROOT && \
-             export LD_LIBRARY_PATH=$LD_LIBRARY_PATH && \
              $mkdir_cmdline && \
+             export LD_LIBRARY_PATH=$LD_LIBRARY_PATH && \
              $dex2oat_cmdline && \
              $dalvikvm_cmdline"
 
diff --git a/test/run-all-tests b/test/run-all-tests
index 318a0de..d0b3cf9 100755
--- a/test/run-all-tests
+++ b/test/run-all-tests
@@ -53,6 +53,9 @@
     elif [ "x$1" = "x--interpreter" ]; then
         run_args="${run_args} --interpreter"
         shift
+    elif [ "x$1" = "x--jit" ]; then
+        run_args="${run_args} --jit"
+        shift
     elif [ "x$1" = "x--no-verify" ]; then
         run_args="${run_args} --no-verify"
         shift
@@ -126,7 +129,7 @@
         echo "  $prog [options]  Run all tests with the given options."
         echo "  Options are all passed to run-test; refer to that for " \
              "further documentation:"
-        echo "    --debug --dev --host --interpreter --jvm --no-optimize"
+        echo "    --debug --dev --host --interpreter --jit --jvm --no-optimize"
         echo "    --no-verify -O --update --valgrind --zygote --64 --relocate"
         echo "    --prebuild --always-clean --gcstress --gcverify --trace"
         echo "    --no-patchoat --no-dex2oat"
diff --git a/test/run-test b/test/run-test
index 8c47663..52f5e0c 100755
--- a/test/run-test
+++ b/test/run-test
@@ -193,6 +193,10 @@
         run_args="${run_args} --interpreter"
         image_suffix="-interpreter"
         shift
+    elif [ "x$1" = "x--jit" ]; then
+        run_args="${run_args} --jit"
+        image_suffix="-interpreter"
+        shift
     elif [ "x$1" = "x--optimizing" ]; then
         run_args="${run_args} -Xcompiler-option --compiler-backend=Optimizing"
         image_suffix="-optimizing"
@@ -430,6 +434,7 @@
         echo "    --gdb                 Run under gdb; incompatible with some tests."
         echo "    --build-only          Build test files only (off by default)."
         echo "    --interpreter         Enable interpreter only mode (off by default)."
+        echo "    --jit                 Enable jit (off by default)."
         echo "    --optimizing          Enable optimizing compiler (off by default)."
         echo "    --quick               Use Quick compiler (default)."
         echo "    --no-verify           Turn off verification (on by default)."
diff --git a/tools/libcore_failures.txt b/tools/libcore_failures.txt
index 4502c02..41d814a 100644
--- a/tools/libcore_failures.txt
+++ b/tools/libcore_failures.txt
@@ -63,15 +63,20 @@
           "org.apache.harmony.tests.java.text.SimpleDateFormatTest#test_parseLjava_lang_StringLjava_text_ParsePosition"]
 },
 {
-  description: "Failing due to switched off network stack on volantisg.",
+  description: "Failing due to missing localhost on volantis.",
   result: EXEC_FAILED,
   modes: [device],
-  names: ["libcore.javax.crypto.CipherTest#testCipherInitWithCertificate",
-          "org.apache.harmony.luni.tests.internal.net.www.protocol.http.HttpURLConnectionTest",
+  names: ["org.apache.harmony.luni.tests.internal.net.www.protocol.http.HttpURLConnectionTest",
           "org.apache.harmony.luni.tests.internal.net.www.protocol.https.HttpsURLConnectionTest",
           "org.apache.harmony.luni.tests.java.net.URLConnectionTest"]
 },
 {
+  description: "Failing due to missing localhost on hammerhead and volantis.",
+  result: EXEC_FAILED,
+  modes: [device],
+  names: ["libcore.javax.crypto.CipherTest#testCipherInitWithCertificate"]
+},
+{
   description: "Test timeouts",
   result: EXEC_TIMEOUT,
   modes: [device],
@@ -94,5 +99,12 @@
   result: EXEC_FAILED,
   name: "org.apache.harmony.security.tests.java.security.Signature2Test#test_verify$BII",
   bug: 18869265
+},
+{
+  description: "Test sometimes timeouts on volantis",
+  result: EXEC_TIMEOUT,
+  modes_variants: [[device,X64]],
+  names: ["libcore.java.lang.SystemTest#testArrayCopyConcurrentModification"],
+  bug: 19165288
 }
 ]