Merge "Include `nodes.h` in `nodes_shared.h` to help editing tools."
diff --git a/Android.mk b/Android.mk
index a1479e1..3740ed8 100644
--- a/Android.mk
+++ b/Android.mk
@@ -328,8 +328,6 @@
 	$(hide) $(call ART_TEST_PREREQ_FINISHED,$@)
 endif
 
-endif  # art_test_bother
-
 # Valgrind.
 .PHONY: valgrind-test-art-target
 valgrind-test-art-target: valgrind-test-art-target-gtest
@@ -343,6 +341,8 @@
 valgrind-test-art-target64: valgrind-test-art-target-gtest64
 	$(hide) $(call ART_TEST_PREREQ_FINISHED,$@)
 
+endif  # art_test_bother
+
 ########################################################################
 # oat-target and oat-target-sync rules
 
diff --git a/build/Android.common_path.mk b/build/Android.common_path.mk
index 00d29b9..e568ce2 100644
--- a/build/Android.common_path.mk
+++ b/build/Android.common_path.mk
@@ -35,13 +35,6 @@
 ART_TARGET_TEST_DIR := /data/art-test
 ART_TARGET_TEST_OUT := $(TARGET_OUT_DATA)/art-test
 
-# Directory used for temporary test files on the host.
-ifneq ($(TMPDIR),)
-ART_HOST_TEST_DIR := $(TMPDIR)/test-art-$(shell echo $$PPID)
-else
-ART_HOST_TEST_DIR := /tmp/$(USER)/test-art-$(shell echo $$PPID)
-endif
-
 # core.oat location on the device.
 TARGET_CORE_OAT := $(ART_TARGET_TEST_DIR)/$(DEX2OAT_TARGET_ARCH)/core.oat
 ifdef TARGET_2ND_ARCH
diff --git a/build/Android.common_test.mk b/build/Android.common_test.mk
index 6b7dc09..8124ca3 100644
--- a/build/Android.common_test.mk
+++ b/build/Android.common_test.mk
@@ -19,6 +19,13 @@
 
 include art/build/Android.common_path.mk
 
+# Directory used for temporary test files on the host.
+ifneq ($(TMPDIR),)
+ART_HOST_TEST_DIR := $(TMPDIR)/test-art-$(shell echo $$PPID)
+else
+ART_HOST_TEST_DIR := /tmp/$(USER)/test-art-$(shell echo $$PPID)
+endif
+
 # We need to set a define for the nativetest dir so that common_runtime_test will know the right
 # path. (The problem is being a 32b test on 64b device, which is still located in nativetest64).
 ART_TARGET_CFLAGS += -DART_TARGET_NATIVETEST_DIR=${ART_TARGET_NATIVETEST_DIR}
diff --git a/compiler/optimizing/code_generator_mips.cc b/compiler/optimizing/code_generator_mips.cc
index b767aa5..aa81bad 100644
--- a/compiler/optimizing/code_generator_mips.cc
+++ b/compiler/optimizing/code_generator_mips.cc
@@ -905,7 +905,7 @@
     } else {
       DCHECK(destination.IsStackSlot())
           << "Cannot move " << c->DebugName() << " to " << destination;
-      __ StoreConst32ToOffset(value, SP, destination.GetStackIndex(), TMP);
+      __ StoreConstToOffset(kStoreWord, value, SP, destination.GetStackIndex(), TMP);
     }
   } else if (c->IsLongConstant()) {
     // Move 64 bit constant.
@@ -917,7 +917,7 @@
     } else {
       DCHECK(destination.IsDoubleStackSlot())
           << "Cannot move " << c->DebugName() << " to " << destination;
-      __ StoreConst64ToOffset(value, SP, destination.GetStackIndex(), TMP);
+      __ StoreConstToOffset(kStoreDoubleword, value, SP, destination.GetStackIndex(), TMP);
     }
   } else if (c->IsFloatConstant()) {
     // Move 32 bit float constant.
@@ -927,7 +927,7 @@
     } else {
       DCHECK(destination.IsStackSlot())
           << "Cannot move " << c->DebugName() << " to " << destination;
-      __ StoreConst32ToOffset(value, SP, destination.GetStackIndex(), TMP);
+      __ StoreConstToOffset(kStoreWord, value, SP, destination.GetStackIndex(), TMP);
     }
   } else {
     // Move 64 bit double constant.
@@ -939,7 +939,7 @@
     } else {
       DCHECK(destination.IsDoubleStackSlot())
           << "Cannot move " << c->DebugName() << " to " << destination;
-      __ StoreConst64ToOffset(value, SP, destination.GetStackIndex(), TMP);
+      __ StoreConstToOffset(kStoreDoubleword, value, SP, destination.GetStackIndex(), TMP);
     }
   }
 }
@@ -1960,6 +1960,25 @@
   codegen_->MaybeRecordImplicitNullCheck(instruction);
 }
 
+Location LocationsBuilderMIPS::RegisterOrZeroConstant(HInstruction* instruction) {
+  return (instruction->IsConstant() && instruction->AsConstant()->IsZeroBitPattern())
+      ? Location::ConstantLocation(instruction->AsConstant())
+      : Location::RequiresRegister();
+}
+
+Location LocationsBuilderMIPS::FpuRegisterOrConstantForStore(HInstruction* instruction) {
+  // We can store 0.0 directly (from the ZERO register) without loading it into an FPU register.
+  // We can store a non-zero float or double constant without first loading it into the FPU,
+  // but we should only prefer this if the constant has a single use.
+  if (instruction->IsConstant() &&
+      (instruction->AsConstant()->IsZeroBitPattern() ||
+       instruction->GetUses().HasExactlyOneElement())) {
+    return Location::ConstantLocation(instruction->AsConstant());
+    // Otherwise fall through and require an FPU register for the constant.
+  }
+  return Location::RequiresFpuRegister();
+}
+
 void LocationsBuilderMIPS::VisitArraySet(HArraySet* instruction) {
   bool needs_runtime_call = instruction->NeedsTypeCheck();
   LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(
@@ -1974,9 +1993,9 @@
     locations->SetInAt(0, Location::RequiresRegister());
     locations->SetInAt(1, Location::RegisterOrConstant(instruction->InputAt(1)));
     if (Primitive::IsFloatingPointType(instruction->InputAt(2)->GetType())) {
-      locations->SetInAt(2, Location::RequiresFpuRegister());
+      locations->SetInAt(2, FpuRegisterOrConstantForStore(instruction->InputAt(2)));
     } else {
-      locations->SetInAt(2, Location::RequiresRegister());
+      locations->SetInAt(2, RegisterOrZeroConstant(instruction->InputAt(2)));
     }
   }
 }
@@ -1985,24 +2004,29 @@
   LocationSummary* locations = instruction->GetLocations();
   Register obj = locations->InAt(0).AsRegister<Register>();
   Location index = locations->InAt(1);
+  Location value_location = locations->InAt(2);
   Primitive::Type value_type = instruction->GetComponentType();
   bool needs_runtime_call = locations->WillCall();
   bool needs_write_barrier =
       CodeGenerator::StoreNeedsWriteBarrier(value_type, instruction->GetValue());
   auto null_checker = GetImplicitNullChecker(instruction);
+  Register base_reg = index.IsConstant() ? obj : TMP;
 
   switch (value_type) {
     case Primitive::kPrimBoolean:
     case Primitive::kPrimByte: {
       uint32_t data_offset = mirror::Array::DataOffset(sizeof(uint8_t)).Uint32Value();
-      Register value = locations->InAt(2).AsRegister<Register>();
       if (index.IsConstant()) {
-        size_t offset =
-            (index.GetConstant()->AsIntConstant()->GetValue() << TIMES_1) + data_offset;
-        __ StoreToOffset(kStoreByte, value, obj, offset, null_checker);
+        data_offset += index.GetConstant()->AsIntConstant()->GetValue() << TIMES_1;
       } else {
-        __ Addu(TMP, obj, index.AsRegister<Register>());
-        __ StoreToOffset(kStoreByte, value, TMP, data_offset, null_checker);
+        __ Addu(base_reg, obj, index.AsRegister<Register>());
+      }
+      if (value_location.IsConstant()) {
+        int32_t value = CodeGenerator::GetInt32ValueOf(value_location.GetConstant());
+        __ StoreConstToOffset(kStoreByte, value, base_reg, data_offset, TMP, null_checker);
+      } else {
+        Register value = value_location.AsRegister<Register>();
+        __ StoreToOffset(kStoreByte, value, base_reg, data_offset, null_checker);
       }
       break;
     }
@@ -2010,15 +2034,18 @@
     case Primitive::kPrimShort:
     case Primitive::kPrimChar: {
       uint32_t data_offset = mirror::Array::DataOffset(sizeof(uint16_t)).Uint32Value();
-      Register value = locations->InAt(2).AsRegister<Register>();
       if (index.IsConstant()) {
-        size_t offset =
-            (index.GetConstant()->AsIntConstant()->GetValue() << TIMES_2) + data_offset;
-        __ StoreToOffset(kStoreHalfword, value, obj, offset, null_checker);
+        data_offset += index.GetConstant()->AsIntConstant()->GetValue() << TIMES_2;
       } else {
-        __ Sll(TMP, index.AsRegister<Register>(), TIMES_2);
-        __ Addu(TMP, obj, TMP);
-        __ StoreToOffset(kStoreHalfword, value, TMP, data_offset, null_checker);
+        __ Sll(base_reg, index.AsRegister<Register>(), TIMES_2);
+        __ Addu(base_reg, obj, base_reg);
+      }
+      if (value_location.IsConstant()) {
+        int32_t value = CodeGenerator::GetInt32ValueOf(value_location.GetConstant());
+        __ StoreConstToOffset(kStoreHalfword, value, base_reg, data_offset, TMP, null_checker);
+      } else {
+        Register value = value_location.AsRegister<Register>();
+        __ StoreToOffset(kStoreHalfword, value, base_reg, data_offset, null_checker);
       }
       break;
     }
@@ -2027,20 +2054,23 @@
     case Primitive::kPrimNot: {
       if (!needs_runtime_call) {
         uint32_t data_offset = mirror::Array::DataOffset(sizeof(int32_t)).Uint32Value();
-        Register value = locations->InAt(2).AsRegister<Register>();
         if (index.IsConstant()) {
-          size_t offset =
-              (index.GetConstant()->AsIntConstant()->GetValue() << TIMES_4) + data_offset;
-          __ StoreToOffset(kStoreWord, value, obj, offset, null_checker);
+          data_offset += index.GetConstant()->AsIntConstant()->GetValue() << TIMES_4;
         } else {
-          DCHECK(index.IsRegister()) << index;
-          __ Sll(TMP, index.AsRegister<Register>(), TIMES_4);
-          __ Addu(TMP, obj, TMP);
-          __ StoreToOffset(kStoreWord, value, TMP, data_offset, null_checker);
+          __ Sll(base_reg, index.AsRegister<Register>(), TIMES_4);
+          __ Addu(base_reg, obj, base_reg);
         }
-        if (needs_write_barrier) {
-          DCHECK_EQ(value_type, Primitive::kPrimNot);
-          codegen_->MarkGCCard(obj, value);
+        if (value_location.IsConstant()) {
+          int32_t value = CodeGenerator::GetInt32ValueOf(value_location.GetConstant());
+          __ StoreConstToOffset(kStoreWord, value, base_reg, data_offset, TMP, null_checker);
+          DCHECK(!needs_write_barrier);
+        } else {
+          Register value = value_location.AsRegister<Register>();
+          __ StoreToOffset(kStoreWord, value, base_reg, data_offset, null_checker);
+          if (needs_write_barrier) {
+            DCHECK_EQ(value_type, Primitive::kPrimNot);
+            codegen_->MarkGCCard(obj, value);
+          }
         }
       } else {
         DCHECK_EQ(value_type, Primitive::kPrimNot);
@@ -2052,47 +2082,54 @@
 
     case Primitive::kPrimLong: {
       uint32_t data_offset = mirror::Array::DataOffset(sizeof(int64_t)).Uint32Value();
-      Register value = locations->InAt(2).AsRegisterPairLow<Register>();
       if (index.IsConstant()) {
-        size_t offset =
-            (index.GetConstant()->AsIntConstant()->GetValue() << TIMES_8) + data_offset;
-        __ StoreToOffset(kStoreDoubleword, value, obj, offset, null_checker);
+        data_offset += index.GetConstant()->AsIntConstant()->GetValue() << TIMES_8;
       } else {
-        __ Sll(TMP, index.AsRegister<Register>(), TIMES_8);
-        __ Addu(TMP, obj, TMP);
-        __ StoreToOffset(kStoreDoubleword, value, TMP, data_offset, null_checker);
+        __ Sll(base_reg, index.AsRegister<Register>(), TIMES_8);
+        __ Addu(base_reg, obj, base_reg);
+      }
+      if (value_location.IsConstant()) {
+        int64_t value = CodeGenerator::GetInt64ValueOf(value_location.GetConstant());
+        __ StoreConstToOffset(kStoreDoubleword, value, base_reg, data_offset, TMP, null_checker);
+      } else {
+        Register value = value_location.AsRegisterPairLow<Register>();
+        __ StoreToOffset(kStoreDoubleword, value, base_reg, data_offset, null_checker);
       }
       break;
     }
 
     case Primitive::kPrimFloat: {
       uint32_t data_offset = mirror::Array::DataOffset(sizeof(float)).Uint32Value();
-      FRegister value = locations->InAt(2).AsFpuRegister<FRegister>();
-      DCHECK(locations->InAt(2).IsFpuRegister());
       if (index.IsConstant()) {
-        size_t offset =
-            (index.GetConstant()->AsIntConstant()->GetValue() << TIMES_4) + data_offset;
-        __ StoreSToOffset(value, obj, offset, null_checker);
+        data_offset += index.GetConstant()->AsIntConstant()->GetValue() << TIMES_4;
       } else {
-        __ Sll(TMP, index.AsRegister<Register>(), TIMES_4);
-        __ Addu(TMP, obj, TMP);
-        __ StoreSToOffset(value, TMP, data_offset, null_checker);
+        __ Sll(base_reg, index.AsRegister<Register>(), TIMES_4);
+        __ Addu(base_reg, obj, base_reg);
+      }
+      if (value_location.IsConstant()) {
+        int32_t value = CodeGenerator::GetInt32ValueOf(value_location.GetConstant());
+        __ StoreConstToOffset(kStoreWord, value, base_reg, data_offset, TMP, null_checker);
+      } else {
+        FRegister value = value_location.AsFpuRegister<FRegister>();
+        __ StoreSToOffset(value, base_reg, data_offset, null_checker);
       }
       break;
     }
 
     case Primitive::kPrimDouble: {
       uint32_t data_offset = mirror::Array::DataOffset(sizeof(double)).Uint32Value();
-      FRegister value = locations->InAt(2).AsFpuRegister<FRegister>();
-      DCHECK(locations->InAt(2).IsFpuRegister());
       if (index.IsConstant()) {
-        size_t offset =
-            (index.GetConstant()->AsIntConstant()->GetValue() << TIMES_8) + data_offset;
-        __ StoreDToOffset(value, obj, offset, null_checker);
+        data_offset += index.GetConstant()->AsIntConstant()->GetValue() << TIMES_8;
       } else {
-        __ Sll(TMP, index.AsRegister<Register>(), TIMES_8);
-        __ Addu(TMP, obj, TMP);
-        __ StoreDToOffset(value, TMP, data_offset, null_checker);
+        __ Sll(base_reg, index.AsRegister<Register>(), TIMES_8);
+        __ Addu(base_reg, obj, base_reg);
+      }
+      if (value_location.IsConstant()) {
+        int64_t value = CodeGenerator::GetInt64ValueOf(value_location.GetConstant());
+        __ StoreConstToOffset(kStoreDoubleword, value, base_reg, data_offset, TMP, null_checker);
+      } else {
+        FRegister value = value_location.AsFpuRegister<FRegister>();
+        __ StoreDToOffset(value, base_reg, data_offset, null_checker);
       }
       break;
     }
@@ -3888,9 +3925,9 @@
     }
   } else {
     if (Primitive::IsFloatingPointType(field_type)) {
-      locations->SetInAt(1, Location::RequiresFpuRegister());
+      locations->SetInAt(1, FpuRegisterOrConstantForStore(instruction->InputAt(1)));
     } else {
-      locations->SetInAt(1, Location::RequiresRegister());
+      locations->SetInAt(1, RegisterOrZeroConstant(instruction->InputAt(1)));
     }
   }
 }
@@ -3901,6 +3938,7 @@
   Primitive::Type type = field_info.GetFieldType();
   LocationSummary* locations = instruction->GetLocations();
   Register obj = locations->InAt(0).AsRegister<Register>();
+  Location value_location = locations->InAt(1);
   StoreOperandType store_type = kStoreByte;
   bool is_volatile = field_info.IsVolatile();
   uint32_t offset = field_info.GetFieldOffset().Uint32Value();
@@ -3941,24 +3979,24 @@
     codegen_->RecordPcInfo(instruction, instruction->GetDexPc());
     if (type == Primitive::kPrimDouble) {
       // Pass FP parameters in core registers.
-      Location in = locations->InAt(1);
-      if (in.IsFpuRegister()) {
-        __ Mfc1(locations->GetTemp(1).AsRegister<Register>(), in.AsFpuRegister<FRegister>());
+      if (value_location.IsFpuRegister()) {
+        __ Mfc1(locations->GetTemp(1).AsRegister<Register>(),
+                value_location.AsFpuRegister<FRegister>());
         __ MoveFromFpuHigh(locations->GetTemp(2).AsRegister<Register>(),
-                           in.AsFpuRegister<FRegister>());
-      } else if (in.IsDoubleStackSlot()) {
+                           value_location.AsFpuRegister<FRegister>());
+      } else if (value_location.IsDoubleStackSlot()) {
         __ LoadFromOffset(kLoadWord,
                           locations->GetTemp(1).AsRegister<Register>(),
                           SP,
-                          in.GetStackIndex());
+                          value_location.GetStackIndex());
         __ LoadFromOffset(kLoadWord,
                           locations->GetTemp(2).AsRegister<Register>(),
                           SP,
-                          in.GetStackIndex() + 4);
+                          value_location.GetStackIndex() + 4);
       } else {
-        DCHECK(in.IsConstant());
-        DCHECK(in.GetConstant()->IsDoubleConstant());
-        int64_t value = bit_cast<int64_t, double>(in.GetConstant()->AsDoubleConstant()->GetValue());
+        DCHECK(value_location.IsConstant());
+        DCHECK(value_location.GetConstant()->IsDoubleConstant());
+        int64_t value = CodeGenerator::GetInt64ValueOf(value_location.GetConstant());
         __ LoadConst64(locations->GetTemp(2).AsRegister<Register>(),
                        locations->GetTemp(1).AsRegister<Register>(),
                        value);
@@ -3967,19 +4005,19 @@
     codegen_->InvokeRuntime(kQuickA64Store, instruction, dex_pc);
     CheckEntrypointTypes<kQuickA64Store, void, volatile int64_t *, int64_t>();
   } else {
-    if (!Primitive::IsFloatingPointType(type)) {
+    if (value_location.IsConstant()) {
+      int64_t value = CodeGenerator::GetInt64ValueOf(value_location.GetConstant());
+      __ StoreConstToOffset(store_type, value, obj, offset, TMP, null_checker);
+    } else if (!Primitive::IsFloatingPointType(type)) {
       Register src;
       if (type == Primitive::kPrimLong) {
-        DCHECK(locations->InAt(1).IsRegisterPair());
-        src = locations->InAt(1).AsRegisterPairLow<Register>();
+        src = value_location.AsRegisterPairLow<Register>();
       } else {
-        DCHECK(locations->InAt(1).IsRegister());
-        src = locations->InAt(1).AsRegister<Register>();
+        src = value_location.AsRegister<Register>();
       }
       __ StoreToOffset(store_type, src, obj, offset, null_checker);
     } else {
-      DCHECK(locations->InAt(1).IsFpuRegister());
-      FRegister src = locations->InAt(1).AsFpuRegister<FRegister>();
+      FRegister src = value_location.AsFpuRegister<FRegister>();
       if (type == Primitive::kPrimFloat) {
         __ StoreSToOffset(src, obj, offset, null_checker);
       } else {
@@ -3990,8 +4028,7 @@
 
   // TODO: memory barriers?
   if (CodeGenerator::StoreNeedsWriteBarrier(type, instruction->InputAt(1))) {
-    DCHECK(locations->InAt(1).IsRegister());
-    Register src = locations->InAt(1).AsRegister<Register>();
+    Register src = value_location.AsRegister<Register>();
     codegen_->MarkGCCard(obj, src);
   }
 
diff --git a/compiler/optimizing/code_generator_mips.h b/compiler/optimizing/code_generator_mips.h
index a42374f..553a7e6 100644
--- a/compiler/optimizing/code_generator_mips.h
+++ b/compiler/optimizing/code_generator_mips.h
@@ -191,6 +191,8 @@
   void HandleShift(HBinaryOperation* operation);
   void HandleFieldSet(HInstruction* instruction, const FieldInfo& field_info);
   void HandleFieldGet(HInstruction* instruction, const FieldInfo& field_info);
+  Location RegisterOrZeroConstant(HInstruction* instruction);
+  Location FpuRegisterOrConstantForStore(HInstruction* instruction);
 
   InvokeDexCallingConventionVisitorMIPS parameter_visitor_;
 
diff --git a/compiler/optimizing/induction_var_analysis.cc b/compiler/optimizing/induction_var_analysis.cc
index 129c2a9..c501ccf 100644
--- a/compiler/optimizing/induction_var_analysis.cc
+++ b/compiler/optimizing/induction_var_analysis.cc
@@ -714,10 +714,12 @@
     case kCondGE: op = kGE; break;
     default:      LOG(FATAL) << "CONDITION UNREACHABLE";
   }
+  // Associate trip count with control instruction, rather than the condition (even
+  // though it's its use) since former provides a convenient use-free placeholder.
+  HInstruction* control = loop->GetHeader()->GetLastInstruction();
   InductionInfo* taken_test = CreateInvariantOp(op, lower_expr, upper_expr);
-  AssignInfo(loop,
-             loop->GetHeader()->GetLastInstruction(),
-             CreateTripCount(tcKind, trip_count, taken_test, type));
+  DCHECK(control->IsIf());
+  AssignInfo(loop, control, CreateTripCount(tcKind, trip_count, taken_test, type));
 }
 
 bool HInductionVarAnalysis::IsTaken(InductionInfo* lower_expr,
diff --git a/compiler/optimizing/induction_var_analysis_test.cc b/compiler/optimizing/induction_var_analysis_test.cc
index 580d24b..292bc4e 100644
--- a/compiler/optimizing/induction_var_analysis_test.cc
+++ b/compiler/optimizing/induction_var_analysis_test.cc
@@ -157,6 +157,13 @@
         iva_->LookupInfo(loop_body_[d]->GetLoopInformation(), instruction));
   }
 
+  // Returns induction information of the trip-count of loop at depth d.
+  std::string GetTripCount(int d) {
+    HInstruction* control = loop_header_[d]->GetLastInstruction();
+    DCHECK(control->IsIf());
+    return GetInductionInfo(control, d);
+  }
+
   // Returns true if instructions have identical induction.
   bool HaveSameInduction(HInstruction* instruction1, HInstruction* instruction2) {
     return HInductionVarAnalysis::InductionEqual(
@@ -239,8 +246,7 @@
   EXPECT_FALSE(HaveSameInduction(store->InputAt(1), increment_[0]));
 
   // Trip-count.
-  EXPECT_STREQ("((100) (TC-loop) ((0) < (100)))",
-               GetInductionInfo(loop_header_[0]->GetLastInstruction(), 0).c_str());
+  EXPECT_STREQ("((100) (TC-loop) ((0) < (100)))", GetTripCount(0).c_str());
 }
 
 TEST_F(InductionVarAnalysisTest, FindDerivedInduction) {
@@ -579,8 +585,7 @@
     }
     EXPECT_STREQ("((1) * i + (1)):PrimInt", GetInductionInfo(increment_[d], d).c_str());
     // Trip-count.
-    EXPECT_STREQ("((100) (TC-loop) ((0) < (100)))",
-                 GetInductionInfo(loop_header_[d]->GetLastInstruction(), d).c_str());
+    EXPECT_STREQ("((100) (TC-loop) ((0) < (100)))", GetTripCount(d).c_str());
   }
 }
 
@@ -607,8 +612,7 @@
   EXPECT_FALSE(HaveSameInduction(store1->InputAt(1), store2->InputAt(1)));
 
   // Trip-count.
-  EXPECT_STREQ("((100) (TC-loop) ((0) < (100)))",
-               GetInductionInfo(loop_header_[0]->GetLastInstruction(), 0).c_str());
+  EXPECT_STREQ("((100) (TC-loop) ((0) < (100)))", GetTripCount(0).c_str());
 }
 
 TEST_F(InductionVarAnalysisTest, ByteLoopControl1) {
@@ -626,8 +630,7 @@
 
   EXPECT_STREQ("((1) * i + ((-128) + (1))):PrimByte", GetInductionInfo(increment_[0], 0).c_str());
   // Trip-count.
-  EXPECT_STREQ("(((127) - (-128)) (TC-loop) ((-128) < (127)))",
-               GetInductionInfo(loop_header_[0]->GetLastInstruction(), 0).c_str());
+  EXPECT_STREQ("(((127) - (-128)) (TC-loop) ((-128) < (127)))", GetTripCount(0).c_str());
 }
 
 TEST_F(InductionVarAnalysisTest, ByteLoopControl2) {
@@ -645,7 +648,7 @@
 
   EXPECT_STREQ("((1) * i + ((-128) + (1))):PrimByte", GetInductionInfo(increment_[0], 0).c_str());
   // Trip-count undefined.
-  EXPECT_STREQ("", GetInductionInfo(loop_header_[0]->GetLastInstruction(), 0).c_str());
+  EXPECT_STREQ("", GetTripCount(0).c_str());
 }
 
 TEST_F(InductionVarAnalysisTest, ShortLoopControl1) {
@@ -664,8 +667,7 @@
   EXPECT_STREQ("((1) * i + ((-32768) + (1))):PrimShort",
                GetInductionInfo(increment_[0], 0).c_str());
   // Trip-count.
-  EXPECT_STREQ("(((32767) - (-32768)) (TC-loop) ((-32768) < (32767)))",
-               GetInductionInfo(loop_header_[0]->GetLastInstruction(), 0).c_str());
+  EXPECT_STREQ("(((32767) - (-32768)) (TC-loop) ((-32768) < (32767)))", GetTripCount(0).c_str());
 }
 
 TEST_F(InductionVarAnalysisTest, ShortLoopControl2) {
@@ -684,7 +686,7 @@
   EXPECT_STREQ("((1) * i + ((-32768) + (1))):PrimShort",
                GetInductionInfo(increment_[0], 0).c_str());
   // Trip-count undefined.
-  EXPECT_STREQ("", GetInductionInfo(loop_header_[0]->GetLastInstruction(), 0).c_str());
+  EXPECT_STREQ("", GetTripCount(0).c_str());
 }
 
 TEST_F(InductionVarAnalysisTest, CharLoopControl1) {
@@ -701,8 +703,7 @@
 
   EXPECT_STREQ("((1) * i + (1)):PrimChar", GetInductionInfo(increment_[0], 0).c_str());
   // Trip-count.
-  EXPECT_STREQ("((65535) (TC-loop) ((0) < (65535)))",
-               GetInductionInfo(loop_header_[0]->GetLastInstruction(), 0).c_str());
+  EXPECT_STREQ("((65535) (TC-loop) ((0) < (65535)))", GetTripCount(0).c_str());
 }
 
 TEST_F(InductionVarAnalysisTest, CharLoopControl2) {
@@ -719,7 +720,7 @@
 
   EXPECT_STREQ("((1) * i + (1)):PrimChar", GetInductionInfo(increment_[0], 0).c_str());
   // Trip-count undefined.
-  EXPECT_STREQ("", GetInductionInfo(loop_header_[0]->GetLastInstruction(), 0).c_str());
+  EXPECT_STREQ("", GetTripCount(0).c_str());
 }
 
 }  // namespace art
diff --git a/compiler/optimizing/induction_var_range.cc b/compiler/optimizing/induction_var_range.cc
index 18e6f5c..cd8b7c7 100644
--- a/compiler/optimizing/induction_var_range.cc
+++ b/compiler/optimizing/induction_var_range.cc
@@ -106,6 +106,12 @@
   return instruction;
 }
 
+/** Helper method to obtain loop's control instruction. */
+static HInstruction* GetLoopControl(HLoopInformation* loop) {
+  DCHECK(loop != nullptr);
+  return loop->GetHeader()->GetLastInstruction();
+}
+
 //
 // Public class methods.
 //
@@ -179,7 +185,7 @@
                                       /*out*/HInstruction** lower,
                                       /*out*/HInstruction** upper) {
   bool is_last_value = false;
-  int64_t s = 0;
+  int64_t stride_value = 0;
   bool b1, b2;  // unused
   if (!GenerateCode(context,
                     instruction,
@@ -189,7 +195,7 @@
                     lower,
                     upper,
                     nullptr,
-                    &s,
+                    &stride_value,
                     &b1,
                     &b2)) {
     LOG(FATAL) << "Failed precondition: CanGenerateRange()";
@@ -232,7 +238,9 @@
                       nullptr,
                       nullptr,
                       nullptr,  // nothing generated yet
-                      &stride_value, &needs_finite_test, &needs_taken_test)
+                      &stride_value,
+                      &needs_finite_test,
+                      &needs_taken_test)
       && !needs_finite_test && !needs_taken_test;
 }
 
@@ -265,7 +273,10 @@
   for (HLoopInformation* lp = instruction->GetBlock()->GetLoopInformation();  // closest enveloping loop
        lp != nullptr;
        lp = lp->GetPreHeader()->GetLoopInformation()) {
+    // Update instruction's information.
     ReplaceInduction(induction_analysis_->LookupInfo(lp, instruction), fetch, replacement);
+    // Update loop's trip-count information.
+    ReplaceInduction(induction_analysis_->LookupInfo(lp, GetLoopControl(lp)), fetch, replacement);
   }
 }
 
@@ -308,13 +319,13 @@
     /*out*/ HLoopInformation** loop,
     /*out*/ HInductionVarAnalysis::InductionInfo** info,
     /*out*/ HInductionVarAnalysis::InductionInfo** trip) const {
-  HLoopInformation* l = context->GetBlock()->GetLoopInformation();  // closest enveloping loop
-  if (l != nullptr) {
-    HInductionVarAnalysis::InductionInfo* i = induction_analysis_->LookupInfo(l, instruction);
+  HLoopInformation* lp = context->GetBlock()->GetLoopInformation();  // closest enveloping loop
+  if (lp != nullptr) {
+    HInductionVarAnalysis::InductionInfo* i = induction_analysis_->LookupInfo(lp, instruction);
     if (i != nullptr) {
-      *loop = l;
+      *loop = lp;
       *info = i;
-      *trip = induction_analysis_->LookupInfo(l, l->GetHeader()->GetLastInstruction());
+      *trip = induction_analysis_->LookupInfo(lp, GetLoopControl(lp));
       return true;
     }
   }
@@ -878,7 +889,8 @@
               } else if (stride_value == -1) {
                 oper = new (graph->GetArena()) HSub(type, opb, opa);
               } else {
-                HInstruction* mul = new (graph->GetArena()) HMul(type, graph->GetIntConstant(stride_value), opa);
+                HInstruction* mul = new (graph->GetArena()) HMul(
+                    type, graph->GetIntConstant(stride_value), opa);
                 oper = new (graph->GetArena()) HAdd(type, Insert(block, mul), opb);
               }
               *result = Insert(block, oper);
diff --git a/compiler/optimizing/intrinsics_mips64.cc b/compiler/optimizing/intrinsics_mips64.cc
index be8eb51..1d153e2 100644
--- a/compiler/optimizing/intrinsics_mips64.cc
+++ b/compiler/optimizing/intrinsics_mips64.cc
@@ -1857,11 +1857,11 @@
   if (type == Primitive::kPrimLong) {
     __ Dclz(TMP, in);
     __ LoadConst64(AT, INT64_C(0x8000000000000000));
-    __ Dsrlv(out, AT, TMP);
+    __ Dsrlv(AT, AT, TMP);
   } else {
     __ Clz(TMP, in);
     __ LoadConst32(AT, 0x80000000);
-    __ Srlv(out, AT, TMP);
+    __ Srlv(AT, AT, TMP);
   }
   // For either value of "type", when "in" is zero, "out" should also
   // be zero. Without this extra "and" operation, when "in" is zero,
@@ -1869,7 +1869,7 @@
   // the MIPS logical shift operations "dsrlv", and "srlv" don't use
   // the shift amount (TMP) directly; they use either (TMP % 64) or
   // (TMP % 32), respectively.
-  __ And(out, out, in);
+  __ And(out, AT, in);
 }
 
 // int java.lang.Integer.highestOneBit(int)
diff --git a/compiler/utils/mips/assembler_mips.h b/compiler/utils/mips/assembler_mips.h
index 099620c..e1255f7 100644
--- a/compiler/utils/mips/assembler_mips.h
+++ b/compiler/utils/mips/assembler_mips.h
@@ -496,46 +496,61 @@
 
  public:
   template <typename ImplicitNullChecker = NoImplicitNullChecker>
-  void StoreConst32ToOffset(int32_t value,
-                            Register base,
-                            int32_t offset,
-                            Register temp,
-                            ImplicitNullChecker null_checker = NoImplicitNullChecker()) {
+  void StoreConstToOffset(StoreOperandType type,
+                          int64_t value,
+                          Register base,
+                          int32_t offset,
+                          Register temp,
+                          ImplicitNullChecker null_checker = NoImplicitNullChecker()) {
+    // We permit `base` and `temp` to coincide (however, we check that neither is AT),
+    // in which case the `base` register may be overwritten in the process.
     CHECK_NE(temp, AT);  // Must not use AT as temp, so as not to overwrite the adjusted base.
-    AdjustBaseAndOffset(base, offset, /* is_doubleword */ false);
-    if (value == 0) {
-      temp = ZERO;
-    } else {
-      LoadConst32(temp, value);
-    }
-    Sw(temp, base, offset);
-    null_checker();
-  }
-
-  template <typename ImplicitNullChecker = NoImplicitNullChecker>
-  void StoreConst64ToOffset(int64_t value,
-                            Register base,
-                            int32_t offset,
-                            Register temp,
-                            ImplicitNullChecker null_checker = NoImplicitNullChecker()) {
-    CHECK_NE(temp, AT);  // Must not use AT as temp, so as not to overwrite the adjusted base.
-    AdjustBaseAndOffset(base, offset, /* is_doubleword */ true);
+    AdjustBaseAndOffset(base, offset, /* is_doubleword */ (type == kStoreDoubleword));
     uint32_t low = Low32Bits(value);
     uint32_t high = High32Bits(value);
-    if (low == 0) {
-      Sw(ZERO, base, offset);
-    } else {
-      LoadConst32(temp, low);
-      Sw(temp, base, offset);
+    Register reg;
+    // If the adjustment left `base` unchanged and equal to `temp`, we can't use `temp`
+    // to load and hold the value but we can use AT instead as AT hasn't been used yet.
+    // Otherwise, `temp` can be used for the value. And if `temp` is the same as the
+    // original `base` (that is, `base` prior to the adjustment), the original `base`
+    // register will be overwritten.
+    if (base == temp) {
+      temp = AT;
     }
-    null_checker();
-    if (high == 0) {
-      Sw(ZERO, base, offset + kMipsWordSize);
+    if (low == 0) {
+      reg = ZERO;
     } else {
-      if (high != low) {
-        LoadConst32(temp, high);
-      }
-      Sw(temp, base, offset + kMipsWordSize);
+      reg = temp;
+      LoadConst32(reg, low);
+    }
+    switch (type) {
+      case kStoreByte:
+        Sb(reg, base, offset);
+        break;
+      case kStoreHalfword:
+        Sh(reg, base, offset);
+        break;
+      case kStoreWord:
+        Sw(reg, base, offset);
+        break;
+      case kStoreDoubleword:
+        Sw(reg, base, offset);
+        null_checker();
+        if (high == 0) {
+          reg = ZERO;
+        } else {
+          reg = temp;
+          if (high != low) {
+            LoadConst32(reg, high);
+          }
+        }
+        Sw(reg, base, offset + kMipsWordSize);
+        break;
+      default:
+        LOG(FATAL) << "UNREACHABLE";
+    }
+    if (type != kStoreDoubleword) {
+      null_checker();
     }
   }
 
diff --git a/compiler/utils/mips/assembler_mips_test.cc b/compiler/utils/mips/assembler_mips_test.cc
index a92455f..a9abf2f 100644
--- a/compiler/utils/mips/assembler_mips_test.cc
+++ b/compiler/utils/mips/assembler_mips_test.cc
@@ -1977,6 +1977,85 @@
   DriverStr(expected, "StoreDToOffset");
 }
 
+TEST_F(AssemblerMIPSTest, StoreConstToOffset) {
+  __ StoreConstToOffset(mips::kStoreByte, 0xFF, mips::A1, +0, mips::T8);
+  __ StoreConstToOffset(mips::kStoreHalfword, 0xFFFF, mips::A1, +0, mips::T8);
+  __ StoreConstToOffset(mips::kStoreWord, 0x12345678, mips::A1, +0, mips::T8);
+  __ StoreConstToOffset(mips::kStoreDoubleword, 0x123456789ABCDEF0, mips::A1, +0, mips::T8);
+
+  __ StoreConstToOffset(mips::kStoreByte, 0, mips::A1, +0, mips::T8);
+  __ StoreConstToOffset(mips::kStoreHalfword, 0, mips::A1, +0, mips::T8);
+  __ StoreConstToOffset(mips::kStoreWord, 0, mips::A1, +0, mips::T8);
+  __ StoreConstToOffset(mips::kStoreDoubleword, 0, mips::A1, +0, mips::T8);
+
+  __ StoreConstToOffset(mips::kStoreDoubleword, 0x1234567812345678, mips::A1, +0, mips::T8);
+  __ StoreConstToOffset(mips::kStoreDoubleword, 0x1234567800000000, mips::A1, +0, mips::T8);
+  __ StoreConstToOffset(mips::kStoreDoubleword, 0x0000000012345678, mips::A1, +0, mips::T8);
+
+  __ StoreConstToOffset(mips::kStoreWord, 0, mips::T8, +0, mips::T8);
+  __ StoreConstToOffset(mips::kStoreWord, 0x12345678, mips::T8, +0, mips::T8);
+
+  __ StoreConstToOffset(mips::kStoreWord, 0, mips::A1, -0xFFF0, mips::T8);
+  __ StoreConstToOffset(mips::kStoreWord, 0x12345678, mips::A1, +0xFFF0, mips::T8);
+
+  __ StoreConstToOffset(mips::kStoreWord, 0, mips::T8, -0xFFF0, mips::T8);
+  __ StoreConstToOffset(mips::kStoreWord, 0x12345678, mips::T8, +0xFFF0, mips::T8);
+
+  const char* expected =
+      "ori $t8, $zero, 0xFF\n"
+      "sb $t8, 0($a1)\n"
+      "ori $t8, $zero, 0xFFFF\n"
+      "sh $t8, 0($a1)\n"
+      "lui $t8, 0x1234\n"
+      "ori $t8, $t8, 0x5678\n"
+      "sw $t8, 0($a1)\n"
+      "lui $t8, 0x9ABC\n"
+      "ori $t8, $t8, 0xDEF0\n"
+      "sw $t8, 0($a1)\n"
+      "lui $t8, 0x1234\n"
+      "ori $t8, $t8, 0x5678\n"
+      "sw $t8, 4($a1)\n"
+
+      "sb $zero, 0($a1)\n"
+      "sh $zero, 0($a1)\n"
+      "sw $zero, 0($a1)\n"
+      "sw $zero, 0($a1)\n"
+      "sw $zero, 4($a1)\n"
+
+      "lui $t8, 0x1234\n"
+      "ori $t8, $t8, 0x5678\n"
+      "sw $t8, 0($a1)\n"
+      "sw $t8, 4($a1)\n"
+      "sw $zero, 0($a1)\n"
+      "lui $t8, 0x1234\n"
+      "ori $t8, $t8, 0x5678\n"
+      "sw $t8, 4($a1)\n"
+      "lui $t8, 0x1234\n"
+      "ori $t8, $t8, 0x5678\n"
+      "sw $t8, 0($a1)\n"
+      "sw $zero, 4($a1)\n"
+
+      "sw $zero, 0($t8)\n"
+      "lui $at, 0x1234\n"
+      "ori $at, $at, 0x5678\n"
+      "sw $at, 0($t8)\n"
+
+      "addiu $at, $a1, -0x7FF8\n"
+      "sw $zero, -0x7FF8($at)\n"
+      "addiu $at, $a1, 0x7FF8\n"
+      "lui $t8, 0x1234\n"
+      "ori $t8, $t8, 0x5678\n"
+      "sw $t8, 0x7FF8($at)\n"
+
+      "addiu $at, $t8, -0x7FF8\n"
+      "sw $zero, -0x7FF8($at)\n"
+      "addiu $at, $t8, 0x7FF8\n"
+      "lui $t8, 0x1234\n"
+      "ori $t8, $t8, 0x5678\n"
+      "sw $t8, 0x7FF8($at)\n";
+  DriverStr(expected, "StoreConstToOffset");
+}
+
 TEST_F(AssemblerMIPSTest, B) {
   mips::MipsLabel label1, label2;
   __ B(&label1);
diff --git a/disassembler/Android.bp b/disassembler/Android.bp
index b074d9f..8dfada2 100644
--- a/disassembler/Android.bp
+++ b/disassembler/Android.bp
@@ -38,7 +38,8 @@
     name: "libart-disassembler",
     defaults: ["libart-disassembler-defaults"],
     shared_libs: [
-        // For disassembler_arm64.
+        // For disassembler_arm*.
+        "libvixl-arm",
         "libvixl-arm64",
     ],
 }
@@ -50,7 +51,8 @@
         "art_debug_defaults",
     ],
     shared_libs: [
-        // For disassembler_arm64.
+        // For disassembler_arm*.
+        "libvixld-arm",
         "libvixld-arm64",
     ],
 }
diff --git a/disassembler/disassembler_arm.cc b/disassembler/disassembler_arm.cc
index c3e288d..925047f 100644
--- a/disassembler/disassembler_arm.cc
+++ b/disassembler/disassembler_arm.cc
@@ -16,1938 +16,232 @@
 
 #include "disassembler_arm.h"
 
-#include <inttypes.h>
-
-#include <ostream>
-#include <sstream>
+#include <memory>
+#include <string>
 
 #include "android-base/logging.h"
-#include "android-base/stringprintf.h"
 
 #include "arch/arm/registers_arm.h"
 #include "base/bit_utils.h"
 
-using android::base::StringPrintf;
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wshadow"
+#include "aarch32/instructions-aarch32.h"
+#include "aarch32/disasm-aarch32.h"
+#pragma GCC diagnostic pop
 
 namespace art {
 namespace arm {
 
-size_t DisassemblerArm::Dump(std::ostream& os, const uint8_t* begin) {
-  if ((reinterpret_cast<intptr_t>(begin) & 1) == 0) {
-    DumpArm(os, begin);
-    return 4;
-  } else {
-    // remove thumb specifier bits
-    begin = reinterpret_cast<const uint8_t*>(reinterpret_cast<uintptr_t>(begin) & ~1);
-    return DumpThumb16(os, begin);
-  }
-}
+using vixl::aarch32::MemOperand;
+using vixl::aarch32::PrintDisassembler;
+using vixl::aarch32::pc;
 
-void DisassemblerArm::Dump(std::ostream& os, const uint8_t* begin, const uint8_t* end) {
-  if ((reinterpret_cast<intptr_t>(begin) & 1) == 0) {
-    for (const uint8_t* cur = begin; cur < end; cur += 4) {
-      DumpArm(os, cur);
+static const vixl::aarch32::Register tr(TR);
+
+class DisassemblerArm::CustomDisassembler FINAL : public PrintDisassembler {
+  class CustomDisassemblerStream FINAL : public DisassemblerStream {
+   public:
+    CustomDisassemblerStream(std::ostream& os,
+                             const CustomDisassembler* disasm,
+                             const DisassemblerOptions* options)
+        : DisassemblerStream(os), disasm_(disasm), options_(options) {}
+
+    DisassemblerStream& operator<<(const PrintLabel& label) OVERRIDE {
+      const LocationType type = label.GetLocationType();
+
+      switch (type) {
+        case kLoadByteLocation:
+        case kLoadHalfWordLocation:
+        case kLoadWordLocation:
+        case kLoadDoubleWordLocation:
+        case kLoadSignedByteLocation:
+        case kLoadSignedHalfWordLocation:
+        case kLoadSinglePrecisionLocation:
+        case kLoadDoublePrecisionLocation:
+        case kVld1Location:
+        case kVld2Location:
+        case kVld3Location:
+        case kVld4Location: {
+          const uintptr_t pc_delta = disasm_->IsT32()
+              ? vixl::aarch32::kT32PcDelta
+              : vixl::aarch32::kA32PcDelta;
+          const int32_t offset = label.GetLabel()->GetLocation();
+
+          os() << "[pc, #" << offset - pc_delta << "]";
+          PrintLiteral(type, offset);
+          return *this;
+        }
+        default:
+          return DisassemblerStream::operator<<(label);
+      }
     }
-  } else {
-    // remove thumb specifier bits
-    begin = reinterpret_cast<const uint8_t*>(reinterpret_cast<uintptr_t>(begin) & ~1);
-    end = reinterpret_cast<const uint8_t*>(reinterpret_cast<uintptr_t>(end) & ~1);
-    for (const uint8_t* cur = begin; cur < end;) {
-      cur += DumpThumb16(os, cur);
-    }
-  }
-}
 
-static const char* kConditionCodeNames[] = {
-  "eq",  // 0000 - equal
-  "ne",  // 0001 - not-equal
-  "cs",  // 0010 - carry-set, greater than, equal or unordered
-  "cc",  // 0011 - carry-clear, less than
-  "mi",  // 0100 - minus, negative
-  "pl",  // 0101 - plus, positive or zero
-  "vs",  // 0110 - overflow
-  "vc",  // 0111 - no overflow
-  "hi",  // 1000 - unsigned higher
-  "ls",  // 1001 - unsigned lower or same
-  "ge",  // 1010 - signed greater than or equal
-  "lt",  // 1011 - signed less than
-  "gt",  // 1100 - signed greater than
-  "le",  // 1101 - signed less than or equal
-  "",    // 1110 - always
-  "nv",  // 1111 - never (mostly obsolete, but might be a clue that we're mistranslating)
-};
-
-void DisassemblerArm::DumpCond(std::ostream& os, uint32_t cond) {
-  if (cond < 15) {
-    os << kConditionCodeNames[cond];
-  } else {
-    os << "Unexpected condition: " << cond;
-  }
-}
-
-void DisassemblerArm::DumpMemoryDomain(std::ostream& os, uint32_t domain) {
-  switch (domain) {
-    case 15U /* 0b1111 */: os << "sy"; break;
-    case 14U /* 0b1110 */: os << "st"; break;
-    case 11U /* 0b1011 */: os << "ish"; break;
-    case 10U /* 0b1010 */: os << "ishst"; break;
-    case  7U /* 0b0111 */: os << "nsh"; break;
-    case  6U /* 0b0110 */: os << "nshst"; break;
-    case  3U /* 0b0011 */: os << "osh"; break;
-    case  2U /* 0b0010 */: os << "oshst"; break;
-  }
-}
-
-void DisassemblerArm::DumpBranchTarget(std::ostream& os, const uint8_t* instr_ptr, int32_t imm32) {
-  os << StringPrintf("%+d (", imm32) << FormatInstructionPointer(instr_ptr + imm32) << ")";
-}
-
-static uint32_t ReadU16(const uint8_t* ptr) {
-  return ptr[0] | (ptr[1] << 8);
-}
-
-static uint32_t ReadU32(const uint8_t* ptr) {
-  return ptr[0] | (ptr[1] << 8) | (ptr[2] << 16) | (ptr[3] << 24);
-}
-
-static const char* kDataProcessingOperations[] = {
-  "and", "eor", "sub", "rsb", "add", "adc", "sbc", "rsc",
-  "tst", "teq", "cmp", "cmn", "orr", "mov", "bic", "mvn",
-};
-
-static const char* kThumbDataProcessingOperations[] = {
-  "and", "eor", "lsl", "lsr", "asr", "adc", "sbc", "ror",
-  "tst", "rsb", "cmp", "cmn", "orr", "mul", "bic", "mvn",
-};
-
-static const char* const kThumb2ShiftOperations[] = {
-    "lsl", "lsr", "asr", "ror"
-};
-
-static const char* kThumbReverseOperations[] = {
-    "rev", "rev16", "rbit", "revsh"
-};
-
-struct ArmRegister {
-  explicit ArmRegister(uint32_t r_in) : r(r_in) { CHECK_LE(r_in, 15U); }
-  ArmRegister(uint32_t instruction, uint32_t at_bit) : r((instruction >> at_bit) & 0xf) {
-    CHECK_LE(r, 15U);
-  }
-  uint32_t r;
-};
-std::ostream& operator<<(std::ostream& os, const ArmRegister& r) {
-  if (r.r == 13) {
-    os << "sp";
-  } else if (r.r == 14) {
-    os << "lr";
-  } else if (r.r == 15) {
-    os << "pc";
-  } else {
-    os << "r" << r.r;
-  }
-  return os;
-}
-
-struct ThumbRegister : ArmRegister {
-  ThumbRegister(uint16_t instruction, uint16_t at_bit) : ArmRegister((instruction >> at_bit) & 0x7) {}
-};
-
-struct RmLslImm2 {
-  explicit RmLslImm2(uint32_t instr) : imm2((instr >> 4) & 0x3), rm(instr & 0xf) {}
-  uint32_t imm2;
-  ArmRegister rm;
-};
-std::ostream& operator<<(std::ostream& os, const RmLslImm2& r) {
-  os << r.rm;
-  if (r.imm2 != 0) {
-    os << ", lsl #" << r.imm2;
-  }
-  return os;
-}
-
-struct ShiftedImmediate {
-  explicit ShiftedImmediate(uint32_t instruction) {
-    uint32_t rotate = ((instruction >> 8) & 0xf);
-    uint32_t imm = (instruction & 0xff);
-    value = (imm >> (2 * rotate)) | (imm << (32 - (2 * rotate)));
-  }
-  uint32_t value;
-};
-std::ostream& operator<<(std::ostream& os, const ShiftedImmediate& rhs) {
-  os << "#" << rhs.value;
-  return os;
-}
-
-struct RegisterList {
-  explicit RegisterList(uint32_t instruction) : register_list(instruction & 0xffff) {}
-  uint32_t register_list;
-};
-std::ostream& operator<<(std::ostream& os, const RegisterList& rhs) {
-  if (rhs.register_list == 0) {
-    os << "<no register list?>";
-    return os;
-  }
-  os << "{";
-  bool first = true;
-  for (size_t i = 0; i < 16; i++) {
-    if ((rhs.register_list & (1 << i)) != 0) {
-      if (first) {
-        first = false;
+    DisassemblerStream& operator<<(const vixl::aarch32::Register reg) OVERRIDE {
+      if (reg.Is(tr)) {
+        os() << "tr";
+        return *this;
       } else {
-        os << ", ";
+        return DisassemblerStream::operator<<(reg);
       }
-      os << ArmRegister(i);
     }
-  }
-  os << "}";
-  return os;
-}
 
-struct FpRegister {
-  FpRegister(uint32_t instr, uint16_t at_bit, uint16_t extra_at_bit) {
-    size = (instr >> 8) & 1;
-    uint32_t Vn = (instr >> at_bit) & 0xF;
-    uint32_t N = (instr >> extra_at_bit) & 1;
-    r = (size != 0 ? ((N << 4) | Vn) : ((Vn << 1) | N));
-  }
-  FpRegister(uint32_t instr, uint16_t at_bit, uint16_t extra_at_bit, uint32_t forced_size) {
-    size = forced_size;
-    uint32_t Vn = (instr >> at_bit) & 0xF;
-    uint32_t N = (instr >> extra_at_bit) & 1;
-    r = (size != 0 ? ((N << 4) | Vn) : ((Vn << 1) | N));
-  }
-  FpRegister(const FpRegister& other, uint32_t offset)
-      : size(other.size), r(other.r + offset) {}
+    DisassemblerStream& operator<<(const MemOperand& operand) OVERRIDE {
+      // VIXL must use a PrintLabel object whenever the base register is PC;
+      // the following check verifies this invariant, and guards against bugs.
+      DCHECK(!operand.GetBaseRegister().Is(pc));
+      DisassemblerStream::operator<<(operand);
 
-  uint32_t size;  // 0 = f32, 1 = f64
-  uint32_t r;
-};
-std::ostream& operator<<(std::ostream& os, const FpRegister& rhs) {
-  return os << ((rhs.size != 0) ? "d" : "s") << rhs.r;
-}
+      if (operand.GetBaseRegister().Is(tr) && operand.IsImmediate()) {
+        os() << " ; ";
+        options_->thread_offset_name_function_(os(), operand.GetOffsetImmediate());
+      }
 
-struct FpRegisterRange {
-  explicit FpRegisterRange(uint32_t instr)
-      : first(instr, 12, 22), imm8(instr & 0xFF) {}
-  FpRegister first;
-  uint32_t imm8;
-};
-std::ostream& operator<<(std::ostream& os, const FpRegisterRange& rhs) {
-  os << "{" << rhs.first;
-  int count = (rhs.first.size != 0 ? ((rhs.imm8 + 1u) >> 1) : rhs.imm8);
-  if (count > 1) {
-    os << "-" << FpRegister(rhs.first, count - 1);
-  }
-  if (rhs.imm8 == 0) {
-    os << " (EMPTY)";
-  } else if (rhs.first.size != 0 && (rhs.imm8 & 1) != 0) {
-    os << rhs.first << " (HALF)";
-  }
-  os << "}";
-  return os;
-}
-
-void DisassemblerArm::DumpArm(std::ostream& os, const uint8_t* instr_ptr) {
-  uint32_t instruction = ReadU32(instr_ptr);
-  uint32_t cond = (instruction >> 28) & 0xf;
-  uint32_t op1 = (instruction >> 25) & 0x7;
-  std::string opcode;
-  std::string suffixes;
-  std::ostringstream args;
-  switch (op1) {
-    case 0:
-    case 1:  // Data processing instructions.
-      {
-        if ((instruction & 0x0ff000f0) == 0x01200070) {  // BKPT
-          opcode = "bkpt";
-          uint32_t imm12 = (instruction >> 8) & 0xfff;
-          uint32_t imm4 = (instruction & 0xf);
-          args << '#' << ((imm12 << 4) | imm4);
-          break;
-        }
-        if ((instruction & 0x0fffffd0) == 0x012fff10) {  // BX and BLX (register)
-          opcode = (((instruction >> 5) & 1) ? "blx" : "bx");
-          args << ArmRegister(instruction & 0xf);
-          break;
-        }
-        bool i = (instruction & (1 << 25)) != 0;
-        bool s = (instruction & (1 << 20)) != 0;
-        uint32_t op = (instruction >> 21) & 0xf;
-        opcode = kDataProcessingOperations[op];
-        bool implicit_s = ((op & ~3) == 8);  // TST, TEQ, CMP, and CMN.
-        bool is_mov = op == 13U /* 0b1101 */ || op == 15U /* 0b1111 */;
-        if (is_mov) {
-          // Show only Rd and Rm.
-          if (s) {
-             suffixes += 's';
-           }
-           args << ArmRegister(instruction, 12) << ", ";
-           if (i) {
-              args << ShiftedImmediate(instruction);
-            } else {
-              // TODO: Shifted register.
-              args << ArmRegister(instruction, 16) << ", " << ArmRegister(instruction, 0);
-            }
-        } else {
-          if (implicit_s) {
-            // Rd is unused (and not shown), and we don't show the 's' suffix either.
-          } else {
-            if (s) {
-              suffixes += 's';
-            }
-            args << ArmRegister(instruction, 12) << ", ";
-          }
-          if (i) {
-            args << ArmRegister(instruction, 16) << ", " << ShiftedImmediate(instruction);
-          } else {
-            // TODO: Shifted register.
-            args << ArmRegister(instruction, 16) << ", " << ArmRegister(instruction, 0);
-          }
-        }
-      }
-      break;
-    case 2:  // Load/store word and unsigned byte.
-      {
-        bool p = (instruction & (1 << 24)) != 0;
-        bool b = (instruction & (1 << 22)) != 0;
-        bool w = (instruction & (1 << 21)) != 0;
-        bool l = (instruction & (1 << 20)) != 0;
-        opcode = StringPrintf("%s%s", (l ? "ldr" : "str"), (b ? "b" : ""));
-        args << ArmRegister(instruction, 12) << ", ";
-        ArmRegister rn(instruction, 16);
-        if (rn.r == 0xf) {
-          UNIMPLEMENTED(FATAL) << "literals";
-        } else {
-          bool wback = !p || w;
-          uint32_t offset = (instruction & 0xfff);
-          if (p && !wback) {
-            args << "[" << rn << ", #" << offset << "]";
-          } else if (p && wback) {
-            args << "[" << rn << ", #" << offset << "]!";
-          } else if (!p && wback) {
-            args << "[" << rn << "], #" << offset;
-          } else {
-            LOG(FATAL) << p << " " << w;
-          }
-          if (rn.r == 9) {
-            args << "  ; ";
-            GetDisassemblerOptions()->thread_offset_name_function_(args, offset);
-          }
-        }
-      }
-      break;
-    case 4:  // Load/store multiple.
-      {
-        bool p = (instruction & (1 << 24)) != 0;
-        bool u = (instruction & (1 << 23)) != 0;
-        bool w = (instruction & (1 << 21)) != 0;
-        bool l = (instruction & (1 << 20)) != 0;
-        opcode = StringPrintf("%s%c%c", (l ? "ldm" : "stm"), (u ? 'i' : 'd'), (p ? 'b' : 'a'));
-        args << ArmRegister(instruction, 16) << (w ? "!" : "") << ", " << RegisterList(instruction);
-      }
-      break;
-    case 5:  // Branch/branch with link.
-      {
-        bool bl = (instruction & (1 << 24)) != 0;
-        opcode = (bl ? "bl" : "b");
-        int32_t imm26 = (instruction & 0xffffff) << 2;
-        int32_t imm32 = (imm26 << 6) >> 6;  // Sign extend.
-        DumpBranchTarget(args, instr_ptr + 8, imm32);
-      }
-      break;
-    default:
-      opcode = "???";
-      break;
+      return *this;
     }
-    opcode += kConditionCodeNames[cond];
-    opcode += suffixes;
-    // TODO: a more complete ARM disassembler could generate wider opcodes.
-    os << FormatInstructionPointer(instr_ptr)
-       << StringPrintf(": %08x\t%-7s ", instruction, opcode.c_str())
-       << args.str() << '\n';
-}
 
-int32_t ThumbExpand(int32_t imm12) {
-  if ((imm12 & 0xC00) == 0) {
-    switch ((imm12 >> 8) & 3) {
-      case 0:
-        return imm12 & 0xFF;
-      case 1:
-        return ((imm12 & 0xFF) << 16) | (imm12 & 0xFF);
-      case 2:
-        return ((imm12 & 0xFF) << 24) | ((imm12 & 0xFF) << 8);
-      default:  // 3
-        return ((imm12 & 0xFF) << 24) | ((imm12 & 0xFF) << 16) | ((imm12 & 0xFF) << 8) |
-            (imm12 & 0xFF);
+    DisassemblerStream& operator<<(const vixl::aarch32::AlignedMemOperand& operand) OVERRIDE {
+      // VIXL must use a PrintLabel object whenever the base register is PC;
+      // the following check verifies this invariant, and guards against bugs.
+      DCHECK(!operand.GetBaseRegister().Is(pc));
+      return DisassemblerStream::operator<<(operand);
     }
-  } else {
-    uint32_t val = 0x80 | (imm12 & 0x7F);
-    int32_t rotate = (imm12 >> 7) & 0x1F;
-    return (val >> rotate) | (val << (32 - rotate));
+
+   private:
+    void PrintLiteral(LocationType type, int32_t offset);
+
+    const CustomDisassembler* disasm_;
+    const DisassemblerOptions* options_;
+  };
+
+ public:
+  CustomDisassembler(std::ostream& os, const DisassemblerOptions* options)
+      // vixl::aarch32::Disassembler::~Disassembler() will delete the stream.
+      : PrintDisassembler(new CustomDisassemblerStream(os, this, options)) {}
+
+  void PrintPc(uint32_t prog_ctr) OVERRIDE {
+    os() << "0x" << std::hex << std::setw(8) << std::setfill('0') << prog_ctr << ": ";
   }
-}
 
-uint32_t VFPExpand32(uint32_t imm8) {
-  CHECK_EQ(imm8 & 0xffu, imm8);
-  uint32_t bit_a = (imm8 >> 7) & 1;
-  uint32_t bit_b = (imm8 >> 6) & 1;
-  uint32_t slice = imm8 & 0x3f;
-  return (bit_a << 31) | ((1 << 30) - (bit_b << 25)) | (slice << 19);
-}
+  bool IsT32() const {
+    return is_t32_;
+  }
 
-static uint64_t VFPExpand64(uint32_t imm8) {
-  CHECK_EQ(imm8 & 0xffu, imm8);
-  uint64_t bit_a = (imm8 >> 7) & 1;
-  uint64_t bit_b = (imm8 >> 6) & 1;
-  uint64_t slice = imm8 & 0x3f;
-  return (bit_a << 63) | ((UINT64_C(1) << 62) - (bit_b << 54)) | (slice << 48);
-}
+  void SetT32(bool is_t32) {
+    is_t32_ = is_t32;
+  }
 
-enum T2LitType {
-  kT2LitInvalid,
-  kT2LitUByte,
-  kT2LitSByte,
-  kT2LitUHalf,
-  kT2LitSHalf,
-  kT2LitUWord,
-  kT2LitSWord,
-  kT2LitHexWord,
-  kT2LitULong,
-  kT2LitSLong,
-  kT2LitHexLong,
+ private:
+  bool is_t32_;
 };
-std::ostream& operator<<(std::ostream& os, T2LitType type) {
-  return os << static_cast<int>(type);
-}
 
-void DumpThumb2Literal(std::ostream& args,
-                       const uint8_t* instr_ptr,
-                       const uintptr_t lo_adr,
-                       const uintptr_t hi_adr,
-                       uint32_t U,
-                       uint32_t imm32,
-                       T2LitType type) {
-  // Literal offsets (imm32) are not required to be aligned so we may need unaligned access.
+void DisassemblerArm::CustomDisassembler::CustomDisassemblerStream::PrintLiteral(LocationType type,
+                                                                                 int32_t offset) {
+  // Literal offsets are not required to be aligned, so we may need unaligned access.
   typedef const int16_t unaligned_int16_t __attribute__ ((aligned (1)));
   typedef const uint16_t unaligned_uint16_t __attribute__ ((aligned (1)));
   typedef const int32_t unaligned_int32_t __attribute__ ((aligned (1)));
-  typedef const uint32_t unaligned_uint32_t __attribute__ ((aligned (1)));
   typedef const int64_t unaligned_int64_t __attribute__ ((aligned (1)));
-  typedef const uint64_t unaligned_uint64_t __attribute__ ((aligned (1)));
+  typedef const float unaligned_float __attribute__ ((aligned (1)));
+  typedef const double unaligned_double __attribute__ ((aligned (1)));
 
-  // Get address of literal. Bail if not within expected buffer range to
-  // avoid trying to fetch invalid literals (we can encounter this when
-  // interpreting raw data as instructions).
-  uintptr_t pc = RoundDown(reinterpret_cast<intptr_t>(instr_ptr) + 4, 4);
-  uintptr_t lit_adr = U ? pc + imm32 : pc - imm32;
-  if (lit_adr < lo_adr || lit_adr >= hi_adr) {
-    args << "  ; (?)";
-    return;
+  // Zeros are used for the LocationType values this function does not care about.
+  const size_t literal_size[kVst4Location + 1] = {
+      0, 0, 0, 0, sizeof(uint8_t), sizeof(unaligned_uint16_t), sizeof(unaligned_int32_t),
+      sizeof(unaligned_int64_t), sizeof(int8_t), sizeof(unaligned_int16_t),
+      sizeof(unaligned_float), sizeof(unaligned_double)};
+  const uintptr_t begin = reinterpret_cast<uintptr_t>(options_->base_address_);
+  const uintptr_t end = reinterpret_cast<uintptr_t>(options_->end_address_);
+  uintptr_t literal_addr = RoundDown(disasm_->GetPc(), vixl::aarch32::kRegSizeInBytes) + offset;
+
+  if (!options_->absolute_addresses_) {
+    literal_addr += begin;
   }
 
-  args << "  ; ";
-  switch (type) {
-    case kT2LitUByte:
-      args << *reinterpret_cast<const uint8_t*>(lit_adr);
-      break;
-    case kT2LitSByte:
-      args << *reinterpret_cast<const int8_t*>(lit_adr);
-      break;
-    case kT2LitUHalf:
-      args << *reinterpret_cast<const unaligned_uint16_t*>(lit_adr);
-      break;
-    case kT2LitSHalf:
-      args << *reinterpret_cast<const unaligned_int16_t*>(lit_adr);
-      break;
-    case kT2LitUWord:
-      args << *reinterpret_cast<const unaligned_uint32_t*>(lit_adr);
-      break;
-    case kT2LitSWord:
-      args << *reinterpret_cast<const unaligned_int32_t*>(lit_adr);
-      break;
-    case kT2LitHexWord:
-      args << StringPrintf("0x%08x", *reinterpret_cast<const unaligned_uint32_t*>(lit_adr));
-      break;
-    case kT2LitULong:
-      args << *reinterpret_cast<const unaligned_uint64_t*>(lit_adr);
-      break;
-    case kT2LitSLong:
-      args << *reinterpret_cast<const unaligned_int64_t*>(lit_adr);
-      break;
-    case kT2LitHexLong:
-      args << StringPrintf("0x%" PRIx64, *reinterpret_cast<unaligned_int64_t*>(lit_adr));
-      break;
-    default:
-      LOG(FATAL) << "Invalid type: " << type;
-      break;
+  os() << "  ; ";
+
+  // Bail out if not within expected buffer range to avoid trying to fetch invalid literals
+  // (we can encounter them when interpreting raw data as instructions).
+  if (literal_addr < begin || literal_addr > end - literal_size[type]) {
+    os() << "(?)";
+  } else {
+    switch (type) {
+      case kLoadByteLocation:
+        os() << *reinterpret_cast<const uint8_t*>(literal_addr);
+        break;
+      case kLoadHalfWordLocation:
+        os() << *reinterpret_cast<unaligned_uint16_t*>(literal_addr);
+        break;
+      case kLoadWordLocation: {
+        const int32_t value = *reinterpret_cast<unaligned_int32_t*>(literal_addr);
+        os() << "0x" << std::hex << std::setw(8) << std::setfill('0') << value;
+        break;
+      }
+      case kLoadDoubleWordLocation: {
+        const int64_t value = *reinterpret_cast<unaligned_int64_t*>(literal_addr);
+        os() << "0x" << std::hex << std::setw(16) << std::setfill('0') << value;
+        break;
+      }
+      case kLoadSignedByteLocation:
+        os() << *reinterpret_cast<const int8_t*>(literal_addr);
+        break;
+      case kLoadSignedHalfWordLocation:
+        os() << *reinterpret_cast<unaligned_int16_t*>(literal_addr);
+        break;
+      case kLoadSinglePrecisionLocation:
+        os() << *reinterpret_cast<unaligned_float*>(literal_addr);
+        break;
+      case kLoadDoublePrecisionLocation:
+        os() << *reinterpret_cast<unaligned_double*>(literal_addr);
+        break;
+      default:
+        UNIMPLEMENTED(FATAL) << "Unexpected literal type: " << type;
+    }
   }
 }
 
-size_t DisassemblerArm::DumpThumb32(std::ostream& os, const uint8_t* instr_ptr) {
-  uint32_t instr = (ReadU16(instr_ptr) << 16) | ReadU16(instr_ptr + 2);
-  // |111|1 1|1000000|0000|1111110000000000|
-  // |5 3|2 1|0987654|3  0|5    0    5    0|
-  // |---|---|-------|----|----------------|
-  // |332|2 2|2222222|1111|1111110000000000|
-  // |1 9|8 7|6543210|9  6|5    0    5    0|
-  // |---|---|-------|----|----------------|
-  // |111|op1| op2   |    |                |
-  uint32_t op1 = (instr >> 27) & 3;
-  if (op1 == 0) {
-    return DumpThumb16(os, instr_ptr);
-  }
+DisassemblerArm::DisassemblerArm(DisassemblerOptions* options)
+    : Disassembler(options), disasm_(std::make_unique<CustomDisassembler>(output_, options)) {}
 
-  // Set valid address range of backing buffer.
-  const uintptr_t lo_adr = reinterpret_cast<intptr_t>(GetDisassemblerOptions()->base_address_);
-  const uintptr_t hi_adr = reinterpret_cast<intptr_t>(GetDisassemblerOptions()->end_address_);
+size_t DisassemblerArm::Dump(std::ostream& os, const uint8_t* begin) {
+  uintptr_t next;
+  // Remove the Thumb specifier bit; no effect if begin does not point to T32 code.
+  const uintptr_t instr_ptr = reinterpret_cast<uintptr_t>(begin) & ~1;
 
-  uint32_t op2 = (instr >> 20) & 0x7F;
-  std::ostringstream opcode;
-  std::ostringstream args;
-  switch (op1) {
-    case 0:
-      break;
-    case 1:
-      if ((op2 & 0x64) == 0) {  // 00x x0xx
-        // |111|11|10|00|0|00|0000|1111110000000000|
-        // |5 3|21|09|87|6|54|3  0|5    0    5    0|
-        // |---|--|--|--|-|--|----|----------------|
-        // |332|22|22|22|2|22|1111|1111110000000000|
-        // |1 9|87|65|43|2|10|9  6|5    0    5    0|
-        // |---|--|--|--|-|--|----|----------------|
-        // |111|01|00|op|0|WL| Rn |                |
-        // |111|01| op2      |    |                |
-        // STM - 111 01 00-01-0-W0 nnnn rrrrrrrrrrrrrrrr
-        // LDM - 111 01 00-01-0-W1 nnnn rrrrrrrrrrrrrrrr
-        // PUSH- 111 01 00-01-0-10 1101 0M0rrrrrrrrrrrrr
-        // POP - 111 01 00-01-0-11 1101 PM0rrrrrrrrrrrrr
-        uint32_t op = (instr >> 23) & 3;
-        uint32_t W = (instr >> 21) & 1;
-        uint32_t L = (instr >> 20) & 1;
-        ArmRegister Rn(instr, 16);
-        if (op == 1 || op == 2) {
-          if (op == 1) {
-            if (L == 0) {
-              opcode << "stm";
-              args << Rn << (W == 0 ? "" : "!") << ", ";
-            } else {
-              if (Rn.r != 13) {
-                opcode << "ldm";
-                args << Rn << (W == 0 ? "" : "!") << ", ";
-              } else {
-                opcode << "pop";
-              }
-            }
-          } else {
-            if (L == 0) {
-              if (Rn.r != 13) {
-                opcode << "stmdb";
-                args << Rn << (W == 0 ? "" : "!") << ", ";
-              } else {
-                opcode << "push";
-              }
-            } else {
-              opcode << "ldmdb";
-              args << Rn << (W == 0 ? "" : "!") << ", ";
-            }
-          }
-          args << RegisterList(instr);
-        }
-      } else if ((op2 & 0x64) == 4) {  // 00x x1xx
-        uint32_t op3 = (instr >> 23) & 3;
-        uint32_t op4 = (instr >> 20) & 3;
-        // uint32_t op5 = (instr >> 4) & 0xF;
-        ArmRegister Rn(instr, 16);
-        ArmRegister Rt(instr, 12);
-        ArmRegister Rd(instr, 8);
-        uint32_t imm8 = instr & 0xFF;
-        if ((op3 & 2) == 2) {     // 1x
-          int W = (instr >> 21) & 1;
-          int U = (instr >> 23) & 1;
-          int P = (instr >> 24) & 1;
+  disasm_->SetT32((reinterpret_cast<uintptr_t>(begin) & 1) != 0);
+  disasm_->JumpToPc(GetPc(instr_ptr));
 
-          if ((op4 & 1) == 1) {
-            opcode << "ldrd";
-          } else {
-            opcode << "strd";
-          }
-          args << Rt << "," << Rd << ", [" << Rn;
-          const char *sign = U ? "+" : "-";
-          if (P == 0 && W == 1) {
-            args << "], #" << sign << (imm8 << 2);
-          } else {
-            args << ", #" << sign << (imm8 << 2) << "]";
-            if (W == 1) {
-              args << "!";
-            }
-          }
-        } else {                  // 0x
-          switch (op4) {
-            case 0:
-              if (op3 == 0) {   // op3 is 00, op4 is 00
-                opcode << "strex";
-                args << Rd << ", " << Rt << ", [" << Rn << ", #" << (imm8 << 2) << "]";
-                if (Rd.r == 13 || Rd.r == 15 || Rt.r == 13 || Rt.r == 15 || Rn.r == 15 ||
-                    Rd.r == Rn.r || Rd.r == Rt.r) {
-                  args << " (UNPREDICTABLE)";
-                }
-              } else {          // op3 is 01, op4 is 00
-                // this is one of strexb, strexh or strexd
-                int op5 = (instr >> 4) & 0xf;
-                switch (op5) {
-                  case 4:
-                  case 5:
-                    opcode << ((op5 == 4) ? "strexb" : "strexh");
-                    Rd = ArmRegister(instr, 0);
-                    args << Rd << ", " << Rt << ", [" << Rn << "]";
-                    if (Rd.r == 13 || Rd.r == 15 || Rt.r == 13 || Rt.r == 15 || Rn.r == 15 ||
-                        Rd.r == Rn.r || Rd.r == Rt.r || (instr & 0xf00) != 0xf00) {
-                      args << " (UNPREDICTABLE)";
-                    }
-                    break;
-                  case 7:
-                    opcode << "strexd";
-                    ArmRegister Rt2 = Rd;
-                    Rd = ArmRegister(instr, 0);
-                    args << Rd << ", " << Rt << ", " << Rt2 << ", [" << Rn << "]";
-                    if (Rd.r == 13 || Rd.r == 15 || Rt.r == 13 || Rt.r == 15 ||
-                        Rt2.r == 13 || Rt2.r == 15 || Rn.r == 15 ||
-                        Rd.r == Rn.r || Rd.r == Rt.r || Rd.r == Rt2.r) {
-                      args << " (UNPREDICTABLE)";
-                    }
-                    break;
-                }
-              }
-              break;
-            case 1:
-              if (op3 == 0) {   // op3 is 00, op4 is 01
-                opcode << "ldrex";
-                args << Rt << ", [" << Rn << ", #" << (imm8 << 2) << "]";
-                if (Rt.r == 13 || Rt.r == 15 || Rn.r == 15 || (instr & 0xf00) != 0xf00) {
-                  args << " (UNPREDICTABLE)";
-                }
-              } else {          // op3 is 01, op4 is 01
-                // this is one of strexb, strexh or strexd
-                int op5 = (instr >> 4) & 0xf;
-                switch (op5) {
-                  case 0:
-                    opcode << "tbb";
-                    break;
-                  case 1:
-                    opcode << "tbh";
-                    break;
-                  case 4:
-                  case 5:
-                    opcode << ((op5 == 4) ? "ldrexb" : "ldrexh");
-                    args << Rt << ", [" << Rn << "]";
-                    if (Rt.r == 13 || Rt.r == 15 || Rn.r == 15 || (instr & 0xf0f) != 0xf0f) {
-                      args << " (UNPREDICTABLE)";
-                    }
-                    break;
-                  case 7:
-                    opcode << "ldrexd";
-                    args << Rt << ", " << Rd /* Rt2 */ << ", [" << Rn << "]";
-                    if (Rt.r == 13 || Rt.r == 15 || Rd.r == 13 /* Rt2 */ || Rd.r == 15 /* Rt2 */ ||
-                        Rn.r == 15 || (instr & 0x00f) != 0x00f) {
-                      args << " (UNPREDICTABLE)";
-                    }
-                    break;
-                }
-              }
-              break;
-            case 2:     // op3 is 0x, op4 is 10
-            case 3:   // op3 is 0x, op4 is 11
-              if (op4 == 2) {
-                opcode << "strd";
-              } else {
-                opcode << "ldrd";
-              }
-              int W = (instr >> 21) & 1;
-              int U = (instr >> 23) & 1;
-              int P = (instr >> 24) & 1;
-
-              args << Rt << "," << Rd << ", [" << Rn;
-              const char *sign = U ? "+" : "-";
-              if (P == 0 && W == 1) {
-                args << "], #" << sign << imm8;
-              } else {
-                args << ", #" << sign << imm8 << "]";
-                if (W == 1) {
-                  args << "!";
-                }
-              }
-              break;
-          }
-        }
-
-      } else if ((op2 & 0x60) == 0x20) {  // 01x xxxx
-        // Data-processing (shifted register)
-        // |111|1110|0000|0|0000|1111|1100|00|00|0000|
-        // |5 3|2109|8765|4|3  0|5   |10 8|7 |5 |3  0|
-        // |---|----|----|-|----|----|----|--|--|----|
-        // |332|2222|2222|2|1111|1111|1100|00|00|0000|
-        // |1 9|8765|4321|0|9  6|5   |10 8|7 |5 |3  0|
-        // |---|----|----|-|----|----|----|--|--|----|
-        // |111|0101| op3|S| Rn |imm3| Rd |i2|ty| Rm |
-        uint32_t op3 = (instr >> 21) & 0xF;
-        uint32_t S = (instr >> 20) & 1;
-        uint32_t imm3 = ((instr >> 12) & 0x7);
-        uint32_t imm2 = ((instr >> 6) & 0x3);
-        uint32_t imm5 = ((imm3 << 2) | imm2);
-        uint32_t shift_type = ((instr >> 4) & 0x3);
-        ArmRegister Rd(instr, 8);
-        ArmRegister Rn(instr, 16);
-        ArmRegister Rm(instr, 0);
-        switch (op3) {
-          case 0x0:
-            if (Rd.r != 0xF) {
-              opcode << "and";
-            } else {
-              if (S != 1U) {
-                opcode << "UNKNOWN TST-" << S;
-                break;
-              }
-              opcode << "tst";
-              S = 0;  // don't print 's'
-            }
-            break;
-          case 0x1: opcode << "bic"; break;
-          case 0x2:
-            if (Rn.r != 0xF) {
-              opcode << "orr";
-            } else {
-              // TODO: use canonical form if there is a shift (lsl, ...).
-              opcode << "mov";
-            }
-            break;
-          case 0x3:
-            if (Rn.r != 0xF) {
-              opcode << "orn";
-            } else {
-              opcode << "mvn";
-            }
-            break;
-          case 0x4:
-            if (Rd.r != 0xF) {
-              opcode << "eor";
-            } else {
-              if (S != 1U) {
-                opcode << "UNKNOWN TEQ-" << S;
-                break;
-              }
-              opcode << "teq";
-              S = 0;  // don't print 's'
-            }
-            break;
-          case 0x6: opcode << "pkh"; break;
-          case 0x8:
-            if (Rd.r != 0xF) {
-              opcode << "add";
-            } else {
-              if (S != 1U) {
-                opcode << "UNKNOWN CMN-" << S;
-                break;
-              }
-              opcode << "cmn";
-              S = 0;  // don't print 's'
-            }
-            break;
-          case 0xA: opcode << "adc"; break;
-          case 0xB: opcode << "sbc"; break;
-          case 0xD:
-            if (Rd.r != 0xF) {
-              opcode << "sub";
-            } else {
-              if (S != 1U) {
-                opcode << "UNKNOWN CMP-" << S;
-                break;
-              }
-              opcode << "cmp";
-              S = 0;  // don't print 's'
-            }
-            break;
-          case 0xE: opcode << "rsb"; break;
-          default: opcode << "UNKNOWN DPSR-" << op3; break;
-        }
-
-        if (S == 1) {
-          opcode << "s";
-        }
-        opcode << ".w";
-
-        if (Rd.r != 0xF) {
-          args << Rd << ", ";
-        }
-        if (Rn.r != 0xF) {
-          args << Rn << ", ";
-        }
-        args << Rm;
-
-        // Shift operand.
-        bool noShift = (imm5 == 0 && shift_type == 0x0);
-        if (!noShift) {
-          args << ", ";
-          if (shift_type == 0x3u && imm5 == 0u) {
-            args << "rrx";
-          } else {
-            args << kThumb2ShiftOperations[shift_type] << " #" << ((0 != imm5) ? imm5 : 32);
-          }
-        }
-
-      } else if ((op2 & 0x40) == 0x40) {  // 1xx xxxx
-        // Co-processor instructions
-        // |111|1|11|000000|0000|1111|1100|000|0  |0000|
-        // |5 3|2|10|987654|3  0|54 2|10 8|7 5|4  |   0|
-        // |---|-|--|------|----|----|----|---|---|----|
-        // |332|2|22|222222|1111|1111|1100|000|0  |0000|
-        // |1 9|8|76|543210|9  6|54 2|10 8|7 5|4  |   0|
-        // |---|-|--|------|----|----|----|---|---|----|
-        // |111| |11| op3  | Rn |    |copr|   |op4|    |
-        uint32_t op3 = (instr >> 20) & 0x3F;
-        uint32_t coproc = (instr >> 8) & 0xF;
-        uint32_t op4 = (instr >> 4) & 0x1;
-
-        if (coproc == 0xA || coproc == 0xB) {   // 101x
-          if (op3 < 0x20 && (op3 & ~5) != 0) {     // 0xxxxx and not 000x0x
-            // Extension register load/store instructions
-            // |1111|110|00000|0000|1111|110|0|00000000|
-            // |5  2|1 9|87654|3  0|5  2|1 9|8|7      0|
-            // |----|---|-----|----|----|---|-|--------|
-            // |3322|222|22222|1111|1111|110|0|00000000|
-            // |1  8|7 5|4   0|9  6|5  2|1 9|8|7      0|
-            // |----|---|-----|----|----|---|-|--------|
-            // |1110|110|PUDWL| Rn | Vd |101|S|  imm8  |
-            uint32_t P = (instr >> 24) & 1;
-            uint32_t U = (instr >> 23) & 1;
-            uint32_t W = (instr >> 21) & 1;
-            if (P == U && W == 1) {
-              opcode << "UNDEFINED";
-            } else {
-              uint32_t L = (instr >> 20) & 1;
-              uint32_t S = (instr >> 8) & 1;
-              ArmRegister Rn(instr, 16);
-              if (P == 1 && W == 0) {  // VLDR
-                FpRegister d(instr, 12, 22);
-                uint32_t imm8 = instr & 0xFF;
-                opcode << (L == 1 ? "vldr" : "vstr");
-                args << d << ", [" << Rn << ", #" << ((U == 1) ? "" : "-")
-                     << (imm8 << 2) << "]";
-                if (Rn.r == 15 && U == 1) {
-                  DumpThumb2Literal(args, instr_ptr, lo_adr, hi_adr, U, imm8 << 2, kT2LitHexLong);
-                }
-              } else if (Rn.r == 13 && W == 1 && U == L) {  // VPUSH/VPOP
-                opcode << (L == 1 ? "vpop" : "vpush");
-                args << FpRegisterRange(instr);
-              } else {  // VLDM
-                opcode << (L == 1 ? "vldm" : "vstm");
-                args << Rn << ((W == 1) ? "!" : "") << ", "
-                     << FpRegisterRange(instr);
-              }
-              opcode << (S == 1 ? ".f64" : ".f32");
-            }
-          } else if ((op3 >> 1) == 2) {      // 00010x
-            if ((instr & 0xD0) == 0x10) {
-              // 64bit transfers between ARM core and extension registers.
-              uint32_t L = (instr >> 20) & 1;
-              uint32_t S = (instr >> 8) & 1;
-              ArmRegister Rt2(instr, 16);
-              ArmRegister Rt(instr, 12);
-              FpRegister m(instr, 0, 5);
-              opcode << "vmov" << (S ? ".f64" : ".f32");
-              if (L == 1) {
-                args << Rt << ", " << Rt2 << ", ";
-              }
-              if (S) {
-                args << m;
-              } else {
-                args << m << ", " << FpRegister(m, 1);
-              }
-              if (L == 0) {
-                args << ", " << Rt << ", " << Rt2;
-              }
-              if (Rt.r == 15 || Rt.r == 13 || Rt2.r == 15 || Rt2.r == 13 ||
-                  (S == 0 && m.r == 31) || (L == 1 && Rt.r == Rt2.r)) {
-                args << " (UNPREDICTABLE)";
-              }
-            }
-          } else if ((op3 >> 4) == 2 && op4 == 0) {     // 10xxxx, op = 0
-            // fp data processing
-            // VMLA, VMLS, VMUL, VNMUL, VADD, VSUB, VDIV, VMOV, ...
-            // |1111|1100|0|0|00|0000|1111|110|0|0|0|0|0|0000|
-            // |5  2|1  8|7|6|54|3  0|5  2|1 9|8|7|6|5|4|3  0|
-            // |----|----|-|-|--|----|----|---|-|-|-|-|-|----|
-            // |3322|2222|2|2|22|1111|1111|110|0|0|0|0|0|0000|
-            // |1  8|7  4|3|2|10|9  6|5  2|1 9|8|7|6|5|4|3  0|
-            // |----|----|-|-|--|----|----|---|-|-|-|-|-|----|
-            // |1110|1110|  op3 | Vn | Vd |101|S|N|Q|M|0| Vm |
-            // |1110|1110|0|D|00| Vn | Vd |101|S|N|0|M|0| Vm | VMLA
-            // |1110|1110|0|D|00| Vn | Vd |101|S|N|1|M|0| Vm | VMLS
-            // |1110|1110|0|D|10| Vn | Vd |101|S|N|0|M|0| Vm | VMUL
-            // |1110|1110|0|D|10| Vn | Vd |101|S|N|1|M|0| Vm | VNMUL
-            // |1110|1110|0|D|11| Vn | Vd |101|S|N|0|M|0| Vm | VADD
-            // |1110|1110|0|D|11| Vn | Vd |101|S|N|1|M|0| Vm | VSUB
-            // |1110|1110|1|D|00| Vn | Vd |101|S|N|0|M|0| Vm | VDIV
-            // |1110|1110|1|D|11| iH | Vd |101|S|0|0|0|0| iL | VMOV (imm)
-            // |1110|1110|1|D|11|op5 | Vd |101|S|.|1|M|0| Vm | ... (see below)
-            uint32_t S = (instr >> 8) & 1;
-            uint32_t Q = (instr >> 6) & 1;
-            FpRegister d(instr, 12, 22);
-            FpRegister n(instr, 16, 7);
-            FpRegister m(instr, 0, 5);
-            if ((op3 & 0xB) == 0) {  // 100x00
-              opcode << (Q == 0 ? "vmla" : "vmls") << (S != 0 ? ".f64" : ".f32");
-              args << d << ", " << n << ", " << m;
-            } else if ((op3 & 0xB) == 0x2) {  // 100x10
-              opcode << (Q == 0 ? "vmul" : "vnmul") << (S != 0 ? ".f64" : ".f32");
-              args << d << ", " << n << ", " << m;
-            } else if ((op3 & 0xB) == 0x3) {  // 100x11
-              opcode << (Q == 0 ? "vadd" : "vsub") << (S != 0 ? ".f64" : ".f32");
-              args << d << ", " << n << ", " << m;
-            } else if ((op3 & 0xB) == 0x8 && Q == 0) {  // 101x00, Q == 0
-              opcode << "vdiv" << (S != 0 ? ".f64" : ".f32");
-              args << d << ", " << n << ", " << m;
-            } else if ((op3 & 0xB) == 0xB && Q == 0) {  // 101x11, Q == 0
-              uint32_t imm8 = ((instr & 0xf0000u) >> 12) | (instr & 0xfu);
-              opcode << "vmov" << (S != 0 ? ".f64" : ".f32");
-              args << d << ", " << (S != 0 ? StringPrintf("0x%016" PRIx64, VFPExpand64(imm8))
-                                           : StringPrintf("0x%08x", VFPExpand32(imm8)));
-              if ((instr & 0xa0) != 0) {
-                args << " (UNPREDICTABLE)";
-              }
-            } else if ((op3 & 0xB) == 0xB && Q == 1) {  // 101x11, Q == 1
-              // VNEG, VSQRT, VCMP, VCMPE, VCVT (floating-point conversion)
-              // |1111|1100|0|0|00|0000|1111|110|0|0 |0|0|0|0000|
-              // |5  2|1  8|7|6|54|3  0|5  2|1 9|8|7 |6|5|4|3  0|
-              // |----|----|-|-|--|----|----|---|-|- |-|-|-|----|
-              // |3322|2222|2|2|22|1111|1111|110|0|0 |0|0|0|0000|
-              // |1  8|7  4|3|2|10|9  6|5  2|1 9|8|7 |6|5|4|3  0|
-              // |----|----|-|-|--|----|----|---|-|- |-|-|-|----|
-              // |1110|1110|1|D|11|0000| Vd |101|S|0 |1|M|0| Vm | VMOV (reg)
-              // |1110|1110|1|D|11|0000| Vd |101|S|1 |1|M|0| Vm | VABS
-              // |1110|1110|1|D|11|0001| Vd |101|S|0 |1|M|0| Vm | VNEG
-              // |1110|1110|1|D|11|0001| Vd |101|S|1 |1|M|0| Vm | VSQRT
-              // |1110|1110|1|D|11|0100| Vd |101|S|op|1|M|0| Vm | VCMP
-              // |1110|1110|1|D|11|0101| Vd |101|S|op|1|0|0|0000| VCMPE
-              // |1110|1110|1|D|11|op5 | Vd |101|S|op|1|M|0| Vm | VCVT
-              uint32_t op5 = (instr >> 16) & 0xF;
-              uint32_t op = (instr >> 7) & 1;
-              // Register types in VCVT instructions rely on the combination of op5 and S.
-              FpRegister Dd(instr, 12, 22, 1);
-              FpRegister Sd(instr, 12, 22, 0);
-              FpRegister Dm(instr, 0, 5, 1);
-              FpRegister Sm(instr, 0, 5, 0);
-              if (op5 == 0) {
-                opcode << (op == 0 ? "vmov" : "vabs") << (S != 0 ? ".f64" : ".f32");
-                args << d << ", " << m;
-              } else if (op5 == 1) {
-                opcode << (op != 0 ? "vsqrt" : "vneg") << (S != 0 ? ".f64" : ".f32");
-                args << d << ", " << m;
-              } else if (op5 == 4) {
-                opcode << "vcmp" << ((op != 0) ? "e" : "") << (S != 0 ? ".f64" : ".f32");
-                args << d << ", " << m;
-              } else if (op5 == 5) {
-                opcode << "vcmp" << ((op != 0) ? "e" : "") << (S != 0 ? ".f64" : ".f32");
-                args << d << ", #0.0";
-                if ((instr & 0x2f) != 0) {
-                  args << " (UNPREDICTABLE)";
-                }
-              } else if (op5 == 0xD) {
-                if (S == 1) {
-                  // vcvt{r}.s32.f64
-                  opcode << "vcvt" << (op == 0 ? "r" : "") << ".s32.f64";
-                  args << Sd << ", " << Dm;
-                } else {
-                  // vcvt{r}.s32.f32
-                  opcode << "vcvt" << (op == 0 ? "r" : "") << ".s32.f32";
-                  args << Sd << ", " << Sm;
-                }
-              } else if (op5 == 0xC) {
-                if (S == 1) {
-                  // vcvt{r}.u32.f64
-                  opcode << "vcvt" << (op == 0 ? "r" : "") << ".u32.f64";
-                  args << Sd << ", " << Dm;
-                } else {
-                  // vcvt{r}.u32.f32
-                  opcode << "vcvt" << (op == 0 ? "r" : "") << ".u32.f32";
-                  args << Sd << ", " << Sm;
-                }
-              } else if (op5 == 0x8) {
-                if (S == 1) {
-                  // vcvt.f64.<Tm>
-                  opcode << "vcvt.f64." << (op == 0 ? "u" : "s") << "32";
-                  args << Dd << ", " << Sm;
-                } else {
-                  // vcvt.f32.<Tm>
-                  opcode << "vcvt.f32." << (op == 0 ? "u" : "s") << "32";
-                  args << Sd << ", " << Sm;
-                }
-              } else if (op5 == 0x7) {
-                if (op == 1) {
-                  if (S == 1) {
-                    // vcvt.f64.f32
-                    opcode << "vcvt.f64.f32";
-                    args << Dd << ", " << Sm;
-                  } else {
-                    // vcvt.f32.f64
-                    opcode << "vcvt.f32.f64";
-                    args << Sd << ", " << Dm;
-                  }
-                }
-              } else if ((op5 & 0xa) == 0xa) {
-                opcode << "vcvt";
-                args << "[undecoded: floating <-> fixed]";
-              }
-            }
-          } else if ((op3 >> 4) == 2 && op4 == 1) {     // 10xxxx, op = 1
-            if (coproc == 10 && (op3 & 0xE) == 0) {
-              // VMOV (between ARM core register and single-precision register)
-              // |1111|1100|000|0 |0000|1111|1100|0|00|0|0000|
-              // |5   |1  8|7 5|4 |3  0|5  2|1  8|7|65|4|3  0|
-              // |----|----|---|- |----|----|----|-|--|-|----|
-              // |3322|2222|222|2 |1111|1111|1100|0|00|0|0000|
-              // |1  8|7  4|3 1|0 |9  6|5  2|1  8|7|65|4|3  0|
-              // |----|----|---|- |----|----|----|-|--|-|----|
-              // |1110|1110|000|op| Vn | Rt |1010|N|00|1|0000|
-              uint32_t op = op3 & 1;
-              ArmRegister Rt(instr, 12);
-              FpRegister n(instr, 16, 7);
-              opcode << "vmov.f32";
-              if (op) {
-                args << Rt << ", " << n;
-              } else {
-                args << n << ", " << Rt;
-              }
-              if (Rt.r == 13 || Rt.r == 15 || (instr & 0x6F) != 0) {
-                args << " (UNPREDICTABLE)";
-              }
-            } else if (coproc == 10 && op3 == 0x2F) {
-              // VMRS
-              // |1111|11000000|0000|1111|1100|000|0|0000|
-              // |5   |1      4|3  0|5  2|1  8|7 5|4|3  0|
-              // |----|--------|----|----|----|---|-|----|
-              // |3322|22222222|1111|1111|1100|000|0|0000|
-              // |1  8|7      0|9  6|5  2|1  8|7 5|4|3  0|
-              // |----|--------|----|----|----|---|-|----|
-              // |1110|11101111|reg | Rt |1010|000|1|0000| - last 7 0s are (0)
-              uint32_t spec_reg = (instr >> 16) & 0xF;
-              ArmRegister Rt(instr, 12);
-              opcode << "vmrs";
-              if (spec_reg == 1) {
-                if (Rt.r == 15) {
-                  args << "APSR_nzcv, FPSCR";
-                } else if (Rt.r == 13) {
-                  args << Rt << ", FPSCR (UNPREDICTABLE)";
-                } else {
-                  args << Rt << ", FPSCR";
-                }
-              } else {
-                args << "(PRIVILEGED)";
-              }
-            } else if (coproc == 11 && (op3 & 0x9) != 8) {
-              // VMOV (ARM core register to scalar or vice versa; 8/16/32-bit)
-            }
-          }
-        }
-      }
-      break;
-    case 2:
-      if ((instr & 0x8000) == 0 && (op2 & 0x20) == 0) {
-        // Data-processing (modified immediate)
-        // |111|11|10|0000|0|0000|1|111|1100|00000000|
-        // |5 3|21|09|8765|4|3  0|5|4 2|10 8|7 5    0|
-        // |---|--|--|----|-|----|-|---|----|--------|
-        // |332|22|22|2222|2|1111|1|111|1100|00000000|
-        // |1 9|87|65|4321|0|9  6|5|4 2|10 8|7 5    0|
-        // |---|--|--|----|-|----|-|---|----|--------|
-        // |111|10|i0| op3|S| Rn |0|iii| Rd |iiiiiiii|
-        //  111 10 x0 xxxx x xxxx opxxx xxxx xxxxxxxx
-        uint32_t i = (instr >> 26) & 1;
-        uint32_t op3 = (instr >> 21) & 0xF;
-        uint32_t S = (instr >> 20) & 1;
-        ArmRegister Rn(instr, 16);
-        uint32_t imm3 = (instr >> 12) & 7;
-        ArmRegister Rd(instr, 8);
-        uint32_t imm8 = instr & 0xFF;
-        int32_t imm32 = (i << 11) | (imm3 << 8) | imm8;
-        if (Rn.r == 0xF && (op3 == 0x2 || op3 == 0x3)) {
-          if (op3 == 0x2) {
-            opcode << "mov";
-            if (S == 1) {
-              opcode << "s";
-            }
-            opcode << ".w";
-          } else {
-            opcode << "mvn";
-            if (S == 1) {
-              opcode << "s";
-            }
-          }
-          args << Rd << ", #" << ThumbExpand(imm32);
-        } else if (Rd.r == 0xF && S == 1 &&
-                   (op3 == 0x0 || op3 == 0x4 || op3 == 0x8 || op3 == 0xD)) {
-          if (op3 == 0x0) {
-            opcode << "tst";
-          } else if (op3 == 0x4) {
-            opcode << "teq";
-          } else if (op3 == 0x8) {
-            opcode << "cmn.w";
-          } else {
-            opcode << "cmp.w";
-          }
-          args << Rn << ", #" << ThumbExpand(imm32);
-        } else {
-          switch (op3) {
-            case 0x0: opcode << "and"; break;
-            case 0x1: opcode << "bic"; break;
-            case 0x2: opcode << "orr"; break;
-            case 0x3: opcode << "orn"; break;
-            case 0x4: opcode << "eor"; break;
-            case 0x8: opcode << "add"; break;
-            case 0xA: opcode << "adc"; break;
-            case 0xB: opcode << "sbc"; break;
-            case 0xD: opcode << "sub"; break;
-            case 0xE: opcode << "rsb"; break;
-            default: opcode << "UNKNOWN DPMI-" << op3; break;
-          }
-          if (S == 1) {
-            opcode << "s";
-          }
-          args << Rd << ", " << Rn << ", #" << ThumbExpand(imm32);
-        }
-      } else if ((instr & 0x8000) == 0 && (op2 & 0x20) != 0) {
-        // Data-processing (plain binary immediate)
-        // |111|11|10|00000|0000|1|111110000000000|
-        // |5 3|21|09|87654|3  0|5|4   0    5    0|
-        // |---|--|--|-----|----|-|---------------|
-        // |332|22|22|22222|1111|1|111110000000000|
-        // |1 9|87|65|43210|9  6|5|4   0    5    0|
-        // |---|--|--|-----|----|-|---------------|
-        // |111|10|x1| op3 | Rn |0|xxxxxxxxxxxxxxx|
-        uint32_t op3 = (instr >> 20) & 0x1F;
-        switch (op3) {
-          case 0x00: case 0x0A: {
-            // ADD/SUB.W Rd, Rn #imm12 - 111 10 i1 0101 0 nnnn 0 iii dddd iiiiiiii
-            ArmRegister Rd(instr, 8);
-            ArmRegister Rn(instr, 16);
-            uint32_t i = (instr >> 26) & 1;
-            uint32_t imm3 = (instr >> 12) & 0x7;
-            uint32_t imm8 = instr & 0xFF;
-            uint32_t imm12 = (i << 11) | (imm3 << 8) | imm8;
-            if (Rn.r != 0xF) {
-              opcode << (op3 == 0 ? "addw" : "subw");
-              args << Rd << ", " << Rn << ", #" << imm12;
-            } else {
-              opcode << "adr";
-              args << Rd << ", ";
-              DumpBranchTarget(args, instr_ptr + 4, (op3 == 0) ? imm12 : -imm12);
-            }
-            break;
-          }
-          case 0x04: case 0x0C: {
-            // MOVW/T Rd, #imm16     - 111 10 i0 0010 0 iiii 0 iii dddd iiiiiiii
-            ArmRegister Rd(instr, 8);
-            uint32_t i = (instr >> 26) & 1;
-            uint32_t imm3 = (instr >> 12) & 0x7;
-            uint32_t imm8 = instr & 0xFF;
-            uint32_t Rn = (instr >> 16) & 0xF;
-            uint32_t imm16 = (Rn << 12) | (i << 11) | (imm3 << 8) | imm8;
-            opcode << (op3 == 0x04 ? "movw" : "movt");
-            args << Rd << ", #" << imm16;
-            break;
-          }
-          case 0x16: case 0x14: case 0x1C: {
-            // BFI Rd, Rn, #lsb, #width - 111 10 0 11 011 0 nnnn 0 iii dddd ii 0 iiiii
-            // SBFX Rd, Rn, #lsb, #width - 111 10 0 11 010 0 nnnn 0 iii dddd ii 0 iiiii
-            // UBFX Rd, Rn, #lsb, #width - 111 10 0 11 110 0 nnnn 0 iii dddd ii 0 iiiii
-            ArmRegister Rd(instr, 8);
-            ArmRegister Rn(instr, 16);
-            uint32_t msb = instr & 0x1F;
-            uint32_t imm2 = (instr >> 6) & 0x3;
-            uint32_t imm3 = (instr >> 12) & 0x7;
-            uint32_t lsb = (imm3 << 2) | imm2;
-            uint32_t width = msb - lsb + 1;
-            if (op3 == 0x16) {
-              if (Rn.r != 0xF) {
-                opcode << "bfi";
-                args << Rd << ", " << Rn << ", #" << lsb << ", #" << width;
-              } else {
-                opcode << "bfc";
-                args << Rd << ", #" << lsb << ", #" << width;
-              }
-            } else {
-              opcode << ((op3 & 0x8) != 0u ? "ubfx" : "sbfx");
-              args << Rd << ", " << Rn << ", #" << lsb << ", #" << width;
-              if (Rd.r == 13 || Rd.r == 15 || Rn.r == 13 || Rn.r == 15 ||
-                  (instr & 0x04000020) != 0u) {
-                args << " (UNPREDICTABLE)";
-              }
-            }
-            break;
-          }
-          default:
-            break;
-        }
-      } else {
-        // Branches and miscellaneous control
-        // |111|11|1000000|0000|1|111|1100|00000000|
-        // |5 3|21|0987654|3  0|5|4 2|10 8|7 5    0|
-        // |---|--|-------|----|-|---|----|--------|
-        // |332|22|2222222|1111|1|111|1100|00000000|
-        // |1 9|87|6543210|9  6|5|4 2|10 8|7 5    0|
-        // |---|--|-------|----|-|---|----|--------|
-        // |111|10| op2   |    |1|op3|op4 |        |
-
-        uint32_t op3 = (instr >> 12) & 7;
-        // uint32_t op4 = (instr >> 8) & 0xF;
-        switch (op3) {
-          case 0:
-            if ((op2 & 0x38) != 0x38) {
-              // Conditional branch
-              // |111|11|1|0000|000000|1|1|1 |1|1 |10000000000|
-              // |5 3|21|0|9876|543  0|5|4|3 |2|1 |0    5    0|
-              // |---|--|-|----|------|-|-|--|-|--|-----------|
-              // |332|22|2|2222|221111|1|1|1 |1|1 |10000000000|
-              // |1 9|87|6|5432|109  6|5|4|3 |2|1 |0    5    0|
-              // |---|--|-|----|------|-|-|--|-|--|-----------|
-              // |111|10|S|cond| imm6 |1|0|J1|0|J2| imm11     |
-              uint32_t S = (instr >> 26) & 1;
-              uint32_t J2 = (instr >> 11) & 1;
-              uint32_t J1 = (instr >> 13) & 1;
-              uint32_t imm6 = (instr >> 16) & 0x3F;
-              uint32_t imm11 = instr & 0x7FF;
-              uint32_t cond = (instr >> 22) & 0xF;
-              int32_t imm32 = (S << 20) |  (J2 << 19) | (J1 << 18) | (imm6 << 12) | (imm11 << 1);
-              imm32 = (imm32 << 11) >> 11;  // sign extend 21bit immediate
-              opcode << "b";
-              DumpCond(opcode, cond);
-              opcode << ".w";
-              DumpBranchTarget(args, instr_ptr + 4, imm32);
-            } else if (op2 == 0x3B) {
-              // Miscellaneous control instructions
-              uint32_t op5 = (instr >> 4) & 0xF;
-              switch (op5) {
-                case 4: opcode << "dsb"; DumpMemoryDomain(args, instr & 0xF); break;
-                case 5: opcode << "dmb"; DumpMemoryDomain(args, instr & 0xF); break;
-                case 6: opcode << "isb"; DumpMemoryDomain(args, instr & 0xF); break;
-              }
-            }
-            break;
-          case 2:
-            if ((op2 & 0x38) == 0x38) {
-              if (op2 == 0x7F) {
-                opcode << "udf";
-              }
-              break;
-            }
-            FALLTHROUGH_INTENDED;  // Else deliberate fall-through to B.
-          case 1: case 3: {
-            // B
-            // |111|11|1|0000|000000|11|1 |1|1 |10000000000|
-            // |5 3|21|0|9876|543  0|54|3 |2|1 |0    5    0|
-            // |---|--|-|----|------|--|--|-|--|-----------|
-            // |332|22|2|2222|221111|11|1 |1|1 |10000000000|
-            // |1 9|87|6|5  2|10   6|54|3 |2|1 |0    5    0|
-            // |---|--|-|----|------|--|--|-|--|-----------|
-            // |111|10|S|cond| imm6 |10|J1|0|J2| imm11     |
-            // |111|10|S| imm10     |10|J1|1|J2| imm11     |
-            uint32_t S = (instr >> 26) & 1;
-            uint32_t cond = (instr >> 22) & 0xF;
-            uint32_t J2 = (instr >> 11) & 1;
-            uint32_t form = (instr >> 12) & 1;
-            uint32_t J1 = (instr >> 13) & 1;
-            uint32_t imm10 = (instr >> 16) & 0x3FF;
-            uint32_t imm6  = (instr >> 16) & 0x3F;
-            uint32_t imm11 = instr & 0x7FF;
-            opcode << "b";
-            int32_t imm32;
-            if (form == 0) {
-              DumpCond(opcode, cond);
-              imm32 = (S << 20) | (J2 << 19) | (J1 << 18) | (imm6 << 12) | (imm11 << 1);
-              imm32 = (imm32 << 11) >> 11;  // sign extend 21 bit immediate.
-            } else {
-              uint32_t I1 = (J1 ^ S) ^ 1;
-              uint32_t I2 = (J2 ^ S) ^ 1;
-              imm32 = (S << 24) | (I1 << 23) | (I2 << 22) | (imm10 << 12) | (imm11 << 1);
-              imm32 = (imm32 << 7) >> 7;  // sign extend 25 bit immediate.
-            }
-            opcode << ".w";
-            DumpBranchTarget(args, instr_ptr + 4, imm32);
-            break;
-          }
-          case 4: case 6: case 5: case 7: {
-            // BL, BLX (immediate)
-            // |111|11|1|0000000000|11|1 |1|1 |10000000000|
-            // |5 3|21|0|9876543  0|54|3 |2|1 |0    5    0|
-            // |---|--|-|----------|--|--|-|--|-----------|
-            // |332|22|2|2222221111|11|1 |1|1 |10000000000|
-            // |1 9|87|6|5    0   6|54|3 |2|1 |0    5    0|
-            // |---|--|-|----------|--|--|-|--|-----------|
-            // |111|10|S| imm10    |11|J1|L|J2| imm11     |
-            uint32_t S = (instr >> 26) & 1;
-            uint32_t J2 = (instr >> 11) & 1;
-            uint32_t L = (instr >> 12) & 1;
-            uint32_t J1 = (instr >> 13) & 1;
-            uint32_t imm10 = (instr >> 16) & 0x3FF;
-            uint32_t imm11 = instr & 0x7FF;
-            if (L == 0) {
-              opcode << "bx";
-            } else {
-              opcode << "blx";
-            }
-            uint32_t I1 = ~(J1 ^ S);
-            uint32_t I2 = ~(J2 ^ S);
-            int32_t imm32 = (S << 24) | (I1 << 23) | (I2 << 22) | (imm10 << 12) | (imm11 << 1);
-            imm32 = (imm32 << 8) >> 8;  // sign extend 24 bit immediate.
-            DumpBranchTarget(args, instr_ptr + 4, imm32);
-            break;
-          }
-        }
-      }
-      break;
-    case 3:
-      switch (op2) {
-        case 0x07: case 0x0F: case 0x17: case 0x1F: {  // Explicitly UNDEFINED, A6.3.
-          opcode << "UNDEFINED";
-          break;
-        }
-        case 0x06: case 0x0E: {  // "Store single data item" undefined opcodes, A6.3.10.
-          opcode << "UNDEFINED [store]";
-          break;
-        }
-        case 0x15: case 0x1D: {  // "Load word" undefined opcodes, A6.3.7.
-          opcode << "UNDEFINED [load]";
-          break;
-        }
-        case 0x10: case 0x12: case 0x14: case 0x16: case 0x18: case 0x1A: case 0x1C: case 0x1E: {
-          opcode << "UNKNOWN " << op2 << " [SIMD]";
-          break;
-        }
-        case 0x01: case 0x00: case 0x09: case 0x08:   // {LD,ST}RB{,T}
-        case 0x03: case 0x02: case 0x0B: case 0x0A:   // {LD,ST}RH{,T}
-        case 0x05: case 0x04: case 0x0D: case 0x0C:   // {LD,ST}R{,T}
-        case 0x11:            case 0x19:              // LDRSB{,T} (no signed store)
-        case 0x13:            case 0x1B: {            // LDRSH{,T} (no signed store)
-          // Load:
-          // (Store is the same except that l==0 and always s==0 below.)
-          //                       00s.whl (sign, word, half, load)
-          // LDR{S}B  imm12: 11111|00s1001| Rn | Rt |imm12             (0x09)
-          // LDR{S}B   imm8: 11111|00s0001| Rn | Rt |1PUW|imm8         (0x01)
-          // LDR{S}BT  imm8: 11111|00s0001| Rn | Rt |1110|imm8         (0x01)
-          // LDR{S}B    lit: 11111|00sU001|1111| Rt |imm12             (0x01/0x09)
-          // LDR{S}B    reg: 11111|00s0001| Rn | Rt |000000|imm2| Rm   (0x01)
-          // LDR{S}H  imm12: 11111|00s1011| Rn | Rt |imm12             (0x0B)
-          // LDR{S}H   imm8: 11111|00s0011| Rn | Rt |1PUW|imm8         (0x03)
-          // LDR{S}HT  imm8: 11111|00s0011| Rn | Rt |1110|imm8         (0x03)
-          // LDR{S}H    lit: 11111|00sU011|1111| Rt |imm12             (0x03/0x0B)
-          // LDR{S}H    reg: 11111|00s0011| Rn | Rt |000000|imm2| Rm   (0x03)
-          // LDR      imm12: 11111|0001101| Rn | Rt |imm12             (0x0D)
-          // LDR       imm8: 11111|0000101| Rn | Rt |1PUW|imm8         (0x05)
-          // LDRT      imm8: 11111|0000101| Rn | Rt |1110|imm8         (0x05)
-          // LDR        lit: 11111|000U101|1111| Rt |imm12             (0x05/0x0D)
-          // LDR        reg: 11111|0000101| Rn | Rt |000000|imm2| Rm   (0x05)
-          //
-          // If Rt == 15, instead of load we have preload:
-          // PLD{W}   imm12: 11111|00010W1| Rn |1111|imm12             (0x09/0x0B)
-          // PLD{W}    imm8: 11111|00000W1| Rn |1111|1100|imm8         (0x01/0x03); -imm8
-          // PLD        lit: 11111|000U001|1111|1111|imm12             (0x01/0x09)
-          // PLD{W}     reg: 11111|00000W1| Rn |1111|000000|imm2| Rm   (0x01/0x03)
-          // PLI      imm12: 11111|0011001| Rn |1111|imm12             (0x19)
-          // PLI       imm8: 11111|0010001| Rn |1111|1100|imm8         (0x11); -imm8
-          // PLI        lit: 11111|001U001|1111|1111|imm12             (0x01/0x09)
-          // PLI        reg: 11111|0010001| Rn |1111|000000|imm2| Rm   (0x01/0x03)
-
-          bool is_load = HasBitSet(instr, 20);
-          bool is_half = HasBitSet(instr, 21);  // W for PLD/PLDW.
-          bool is_word = HasBitSet(instr, 22);
-          bool is_signed = HasBitSet(instr, 24);
-          ArmRegister Rn(instr, 16);
-          ArmRegister Rt(instr, 12);
-          uint32_t imm12 = instr & 0xFFF;
-          uint32_t U = (instr >> 23) & 1;  // U for imm12
-          uint32_t imm8 = instr & 0xFF;
-          uint32_t op4 = (instr >> 8) & 0xF;  // 1PUW for imm8
-          if (Rt.r == PC && is_load && !is_word) {
-            // PLD, PLDW, PLI
-            const char* pld_pli = (is_signed ? "pli" : "pld");
-            const char* w = (is_half ? "w" : "");
-            if (is_signed && !is_half) {
-              opcode << "UNDEFINED [PLI+W]";
-            } else if (Rn.r == PC || U != 0u) {
-              opcode << pld_pli << w;
-              args << "[" << Rn << ", #" << (U != 0u ? "" : "-") << imm12 << "]";
-              if (Rn.r == PC && is_half) {
-                args << " (UNPREDICTABLE)";
-              }
-            } else if ((instr & 0xFC0) == 0) {
-              opcode << pld_pli << w;
-              RmLslImm2 Rm(instr);
-              args << "[" << Rn << ", " << Rm << "]";
-            } else if (op4 == 0xC) {
-              opcode << pld_pli << w;
-              args << "[" << Rn << ", #-" << imm8 << "]";
-            } else {
-              opcode << "UNDEFINED [~" << pld_pli << "]";
-            }
-            break;
-          }
-          const char* ldr_str = is_load ? "ldr" : "str";
-          const char* sign = is_signed ? "s" : "";
-          const char* type = is_word ? "" : is_half ? "h" : "b";
-          bool unpred = (Rt.r == SP && !is_word) || (Rt.r == PC && !is_load);
-          if (Rn.r == PC && !is_load) {
-            opcode << "UNDEFINED [STR-lit]";
-            unpred = false;
-          } else if (Rn.r == PC || U != 0u) {
-            // Load/store with imm12 (load literal if Rn.r == PC; there's no store literal).
-            opcode << ldr_str << sign << type << ".w";
-            args << Rt << ", [" << Rn << ", #" << (U != 0u ? "" : "-") << imm12 << "]";
-            if (Rn.r == TR && is_load) {
-              args << "  ; ";
-              GetDisassemblerOptions()->thread_offset_name_function_(args, imm12);
-            } else if (Rn.r == PC) {
-              T2LitType lit_type[] = {
-                  kT2LitUByte, kT2LitUHalf, kT2LitHexWord, kT2LitInvalid,
-                  kT2LitUByte, kT2LitUHalf, kT2LitHexWord, kT2LitInvalid,
-                  kT2LitSByte, kT2LitSHalf, kT2LitInvalid, kT2LitInvalid,
-                  kT2LitSByte, kT2LitSHalf, kT2LitInvalid, kT2LitInvalid,
-              };
-              DCHECK_LT(op2 >> 1, arraysize(lit_type));
-              DCHECK_NE(lit_type[op2 >> 1], kT2LitInvalid);
-              DumpThumb2Literal(args, instr_ptr, lo_adr, hi_adr, U, imm12, lit_type[op2 >> 1]);
-            }
-          } else if ((instr & 0xFC0) == 0) {
-            opcode << ldr_str << sign << type << ".w";
-            RmLslImm2 Rm(instr);
-            args << Rt << ", [" << Rn << ", " << Rm << "]";
-            unpred = unpred || (Rm.rm.r == SP) || (Rm.rm.r == PC);
-          } else if (is_word && Rn.r == SP && imm8 == 4 && op4 == (is_load ? 0xB : 0xD)) {
-            opcode << (is_load ? "pop" : "push") << ".w";
-            args << Rn;
-            unpred = unpred || (Rn.r == SP);
-          } else if ((op4 & 5) == 0) {
-            opcode << "UNDEFINED [P = W = 0 for " << ldr_str << "]";
-            unpred = false;
-          } else {
-            uint32_t P = (instr >> 10) & 1;
-            U = (instr >> 9) & 1;
-            uint32_t W = (instr >> 8) & 1;
-            bool pre_index = (P != 0 && W == 1);
-            bool post_index = (P == 0 && W == 1);
-            const char* t = (P != 0 && U != 0 && W == 0) ? "t" : "";  // Unprivileged load/store?
-            opcode << ldr_str << sign << type << t << ".w";
-            args << Rt << ", [" << Rn << (post_index ? "]" : "") << ", #" << (U != 0 ? "" : "-")
-                << imm8 << (post_index ? "" : "]") << (pre_index ? "!" : "");
-            unpred = (W != 0 && Rn.r == Rt.r);
-          }
-          if (unpred) {
-            args << " (UNPREDICTABLE)";
-          }
-          break;
-        }
-        case 0x29: {  // 0101001
-          // |111|11|1000000|0000|1111|1100|00|0 0|0000|
-          // |5 3|21|0     4|3  0|5  2|1  8|76|5 4|3  0|
-          // |---|--|-------|----|----|----|--|---|----|
-          // |332|22|2222222|1111|1111|1100|00|0 0|0000|
-          // |1 9|87|6     0|9  6|5  2|1  8|76|5 4|3  0|
-          // |---|--|-------|----|----|----|--|---|----|
-          // |111|11|0101001| Rm |1111| Rd |11|op3| Rm |
-          // REV   - 111 11 0101001 mmmm 1111 dddd 1000 mmmm
-          // REV16 - 111 11 0101001 mmmm 1111 dddd 1001 mmmm
-          // RBIT  - 111 11 0101001 mmmm 1111 dddd 1010 mmmm
-          // REVSH - 111 11 0101001 mmmm 1111 dddd 1011 mmmm
-          if ((instr & 0xf0c0) == 0xf080) {
-            uint32_t op3 = (instr >> 4) & 3;
-            opcode << kThumbReverseOperations[op3];
-            ArmRegister Rm(instr, 0);
-            ArmRegister Rd(instr, 8);
-            args << Rd << ", " << Rm;
-            ArmRegister Rm2(instr, 16);
-            if (Rm.r != Rm2.r || Rm.r == 13 || Rm.r == 15 || Rd.r == 13 || Rd.r == 15) {
-              args << " (UNPREDICTABLE)";
-            }
-          }  // else unknown instruction
-          break;
-        }
-        case 0x2B: {  // 0101011
-          //  CLZ - 111 11 0101011 mmmm 1111 dddd 1000 mmmm
-          if ((instr & 0xf0f0) == 0xf080) {
-            opcode << "clz";
-            ArmRegister Rm(instr, 0);
-            ArmRegister Rd(instr, 8);
-            args << Rd << ", " << Rm;
-            ArmRegister Rm2(instr, 16);
-            if (Rm.r != Rm2.r || Rm.r == 13 || Rm.r == 15 || Rd.r == 13 || Rd.r == 15) {
-              args << " (UNPREDICTABLE)";
-            }
-          }
-          break;
-        }
-        case 0x7B: case 0x7F: {
-          FpRegister d(instr, 12, 22);
-          FpRegister m(instr, 0, 5);
-          uint32_t sz = (instr >> 18) & 0x3;  // Decode size bits.
-          uint32_t size = (sz == 0) ? 8 : sz << 4;
-          uint32_t opc2 = (instr >> 7) & 0xF;
-          uint32_t Q = (instr >> 6) & 1;
-          if (Q == 0 && opc2 == 0xA && size == 8) {  // 1010, VCNT
-            opcode << "vcnt." << size;
-            args << d << ", " << m;
-          } else if (Q == 0 && (opc2 == 0x4 || opc2 == 0x5) && size <= 32) {  // 010x, VPADDL
-            bool op = HasBitSet(instr, 7);
-            opcode << "vpaddl." << (op ? "u" : "s") << size;
-            args << d << ", " << m;
-          } else {
-            opcode << "UNKNOWN " << op2;
-          }
-          break;
-        }
-        default:      // more formats
-          if ((op2 >> 4) == 2) {      // 010xxxx
-            // data processing (register)
-            if ((instr & 0x0080f0f0) == 0x0000f000) {
-              // LSL, LSR, ASR, ROR
-              uint32_t shift_op = (instr >> 21) & 3;
-              uint32_t S = (instr >> 20) & 1;
-              ArmRegister Rd(instr, 8);
-              ArmRegister Rn(instr, 16);
-              ArmRegister Rm(instr, 0);
-              opcode << kThumb2ShiftOperations[shift_op] << (S != 0 ? "s" : "");
-              args << Rd << ", " << Rn << ", " << Rm;
-            }
-          } else if ((op2 >> 3) == 6) {       // 0110xxx
-            // Multiply, multiply accumulate, and absolute difference
-            op1 = (instr >> 20) & 0x7;
-            op2 = (instr >> 4) & 0x1;
-            ArmRegister Ra(instr, 12);
-            ArmRegister Rn(instr, 16);
-            ArmRegister Rm(instr, 0);
-            ArmRegister Rd(instr, 8);
-            switch (op1) {
-            case 0:
-              if (op2 == 0) {
-                if (Ra.r == 0xf) {
-                  opcode << "mul";
-                  args << Rd << ", " << Rn << ", " << Rm;
-                } else {
-                  opcode << "mla";
-                  args << Rd << ", " << Rn << ", " << Rm << ", " << Ra;
-                }
-              } else {
-                opcode << "mls";
-                args << Rd << ", " << Rn << ", " << Rm << ", " << Ra;
-              }
-              break;
-            case 1:
-            case 2:
-            case 3:
-            case 4:
-            case 5:
-            case 6:
-                break;        // do these sometime
-            }
-          } else if ((op2 >> 3) == 7) {       // 0111xxx
-            // Long multiply, long multiply accumulate, and divide
-            op1 = (instr >> 20) & 0x7;
-            op2 = (instr >> 4) & 0xf;
-            ArmRegister Rn(instr, 16);
-            ArmRegister Rm(instr, 0);
-            ArmRegister Rd(instr, 8);
-            ArmRegister RdHi(instr, 8);
-            ArmRegister RdLo(instr, 12);
-            switch (op1) {
-            case 0:
-              opcode << "smull";
-              args << RdLo << ", " << RdHi << ", " << Rn << ", " << Rm;
-              break;
-            case 1:
-              opcode << "sdiv";
-              args << Rd << ", " << Rn << ", " << Rm;
-              break;
-            case 2:
-              opcode << "umull";
-              args << RdLo << ", " << RdHi << ", " << Rn << ", " << Rm;
-              break;
-            case 3:
-              opcode << "udiv";
-              args << Rd << ", " << Rn << ", " << Rm;
-              break;
-            case 4:
-            case 5:
-            case 6:
-              break;      // TODO: when we generate these...
-            }
-          }
-      }
-      break;
-    default:
-      break;
-  }
-
-  // Apply any IT-block conditions to the opcode if necessary.
-  if (!it_conditions_.empty()) {
-    opcode << it_conditions_.back();
-    it_conditions_.pop_back();
-  }
-  if (opcode.str().size() == 0) {
-    opcode << "UNKNOWN " << op2;
-  }
-
-  os << FormatInstructionPointer(instr_ptr)
-     << StringPrintf(": %08x\t%-7s ", instr, opcode.str().c_str())
-     << args.str() << '\n';
-  return 4;
-}  // NOLINT(readability/fn_size)
-
-size_t DisassemblerArm::DumpThumb16(std::ostream& os, const uint8_t* instr_ptr) {
-  uint16_t instr = ReadU16(instr_ptr);
-  bool is_32bit = ((instr & 0xF000) == 0xF000) || ((instr & 0xF800) == 0xE800);
-  if (is_32bit) {
-    return DumpThumb32(os, instr_ptr);
+  if (disasm_->IsT32()) {
+    const uint16_t* const ip = reinterpret_cast<const uint16_t*>(instr_ptr);
+    next = reinterpret_cast<uintptr_t>(disasm_->DecodeT32At(ip));
   } else {
-    std::ostringstream opcode;
-    std::ostringstream args;
-    uint16_t opcode1 = instr >> 10;
-    if (opcode1 < 0x10) {
-      // shift (immediate), add, subtract, move, and compare
-      uint16_t opcode2 = instr >> 9;
-      switch (opcode2) {
-        case 0x0: case 0x1: case 0x2: case 0x3: case 0x4: case 0x5: case 0x6: case 0x7:
-        case 0x8: case 0x9: case 0xA: case 0xB: {
-          // Logical shift left     - 00 000xx iii mmm ddd
-          // Logical shift right    - 00 001xx iii mmm ddd
-          // Arithmetic shift right - 00 010xx iii mmm ddd
-          uint16_t imm5 = (instr >> 6) & 0x1F;
-          ThumbRegister rm(instr, 3);
-          ThumbRegister Rd(instr, 0);
-          if (opcode2 <= 3) {
-            opcode << "lsls";
-          } else if (opcode2 <= 7) {
-            opcode << "lsrs";
-          } else {
-            opcode << "asrs";
-          }
-          args << Rd << ", " << rm << ", #" << imm5;
-          break;
-        }
-        case 0xC: case 0xD: case 0xE: case 0xF: {
-          // Add register        - 00 01100 mmm nnn ddd
-          // Sub register        - 00 01101 mmm nnn ddd
-          // Add 3-bit immediate - 00 01110 iii nnn ddd
-          // Sub 3-bit immediate - 00 01111 iii nnn ddd
-          uint16_t imm3_or_Rm = (instr >> 6) & 7;
-          ThumbRegister Rn(instr, 3);
-          ThumbRegister Rd(instr, 0);
-          if ((opcode2 & 2) != 0 && imm3_or_Rm == 0) {
-            opcode << "mov";
-          } else {
-            if ((opcode2 & 1) == 0) {
-              opcode << "adds";
-            } else {
-              opcode << "subs";
-            }
-          }
-          args << Rd << ", " << Rn;
-          if ((opcode2 & 2) == 0) {
-            ArmRegister Rm(imm3_or_Rm);
-            args << ", " << Rm;
-          } else if (imm3_or_Rm != 0) {
-            args << ", #" << imm3_or_Rm;
-          }
-          break;
-        }
-        case 0x10: case 0x11: case 0x12: case 0x13:
-        case 0x14: case 0x15: case 0x16: case 0x17:
-        case 0x18: case 0x19: case 0x1A: case 0x1B:
-        case 0x1C: case 0x1D: case 0x1E: case 0x1F: {
-          // MOVS Rd, #imm8 - 00100 ddd iiiiiiii
-          // CMP  Rn, #imm8 - 00101 nnn iiiiiiii
-          // ADDS Rn, #imm8 - 00110 nnn iiiiiiii
-          // SUBS Rn, #imm8 - 00111 nnn iiiiiiii
-          ThumbRegister Rn(instr, 8);
-          uint16_t imm8 = instr & 0xFF;
-          switch (opcode2 >> 2) {
-            case 4: opcode << "movs"; break;
-            case 5: opcode << "cmp"; break;
-            case 6: opcode << "adds"; break;
-            case 7: opcode << "subs"; break;
-          }
-          args << Rn << ", #" << imm8;
-          break;
-        }
-        default:
-          break;
-      }
-    } else if (opcode1 == 0x10) {
-      // Data-processing
-      uint16_t opcode2 = (instr >> 6) & 0xF;
-      ThumbRegister rm(instr, 3);
-      ThumbRegister rdn(instr, 0);
-      opcode << kThumbDataProcessingOperations[opcode2];
-      args << rdn << ", " << rm;
-    } else if (opcode1 == 0x11) {
-      // Special data instructions and branch and exchange
-      uint16_t opcode2 = (instr >> 6) & 0x0F;
-      switch (opcode2) {
-        case 0x0: case 0x1: case 0x2: case 0x3: {
-          // Add low registers  - 010001 0000 xxxxxx
-          // Add high registers - 010001 0001/001x xxxxxx
-          uint16_t DN = (instr >> 7) & 1;
-          ArmRegister rm(instr, 3);
-          uint16_t Rdn = instr & 7;
-          ArmRegister DN_Rdn((DN << 3) | Rdn);
-          opcode << "add";
-          args << DN_Rdn << ", " << rm;
-          break;
-        }
-        case 0x8: case 0x9: case 0xA: case 0xB: {
-          // Move low registers  - 010001 1000 xxxxxx
-          // Move high registers - 010001 1001/101x xxxxxx
-          uint16_t DN = (instr >> 7) & 1;
-          ArmRegister rm(instr, 3);
-          uint16_t Rdn = instr & 7;
-          ArmRegister DN_Rdn((DN << 3) | Rdn);
-          opcode << "mov";
-          args << DN_Rdn << ", " << rm;
-          break;
-        }
-        case 0x5: case 0x6: case 0x7: {
-          // Compare high registers - 010001 0101/011x xxxxxx
-          uint16_t N = (instr >> 7) & 1;
-          ArmRegister rm(instr, 3);
-          uint16_t Rn = instr & 7;
-          ArmRegister N_Rn((N << 3) | Rn);
-          opcode << "cmp";
-          args << N_Rn << ", " << rm;
-          break;
-        }
-        case 0xC: case 0xD: case 0xE: case 0xF: {
-          // Branch and exchange           - 010001 110x xxxxxx
-          // Branch with link and exchange - 010001 111x xxxxxx
-          ArmRegister rm(instr, 3);
-          opcode << ((opcode2 & 0x2) == 0 ? "bx" : "blx");
-          args << rm;
-          break;
-        }
-        default:
-          break;
-      }
-    } else if (opcode1 == 0x12 || opcode1 == 0x13) {  // 01001x
-      const uintptr_t lo_adr = reinterpret_cast<intptr_t>(GetDisassemblerOptions()->base_address_);
-      const uintptr_t hi_adr = reinterpret_cast<intptr_t>(GetDisassemblerOptions()->end_address_);
-      ThumbRegister Rt(instr, 8);
-      uint16_t imm8 = instr & 0xFF;
-      opcode << "ldr";
-      args << Rt << ", [pc, #" << (imm8 << 2) << "]";
-      DumpThumb2Literal(args, instr_ptr, lo_adr, hi_adr, /*U*/ 1u, imm8 << 2, kT2LitHexWord);
-    } else if ((opcode1 >= 0x14 && opcode1 <= 0x17) ||  // 0101xx
-               (opcode1 >= 0x18 && opcode1 <= 0x1f) ||  // 011xxx
-               (opcode1 >= 0x20 && opcode1 <= 0x27)) {  // 100xxx
-      // Load/store single data item
-      uint16_t opA = (instr >> 12) & 0xF;
-      if (opA == 0x5) {
-        uint16_t opB = (instr >> 9) & 0x7;
-        ThumbRegister Rm(instr, 6);
-        ThumbRegister Rn(instr, 3);
-        ThumbRegister Rt(instr, 0);
-        switch (opB) {
-          case 0: opcode << "str"; break;
-          case 1: opcode << "strh"; break;
-          case 2: opcode << "strb"; break;
-          case 3: opcode << "ldrsb"; break;
-          case 4: opcode << "ldr"; break;
-          case 5: opcode << "ldrh"; break;
-          case 6: opcode << "ldrb"; break;
-          case 7: opcode << "ldrsh"; break;
-        }
-        args << Rt << ", [" << Rn << ", " << Rm << "]";
-      } else if (opA == 9) {
-        uint16_t opB = (instr >> 11) & 1;
-        ThumbRegister Rt(instr, 8);
-        uint16_t imm8 = instr & 0xFF;
-        opcode << (opB == 0 ? "str" : "ldr");
-        args << Rt << ", [sp, #" << (imm8 << 2) << "]";
-      } else {
-        uint16_t imm5 = (instr >> 6) & 0x1F;
-        uint16_t opB = (instr >> 11) & 1;
-        ThumbRegister Rn(instr, 3);
-        ThumbRegister Rt(instr, 0);
-        switch (opA) {
-          case 6:
-            imm5 <<= 2;
-            opcode << (opB == 0 ? "str" : "ldr");
-            break;
-          case 7:
-            imm5 <<= 0;
-            opcode << (opB == 0 ? "strb" : "ldrb");
-            break;
-          case 8:
-            imm5 <<= 1;
-            opcode << (opB == 0 ? "strh" : "ldrh");
-            break;
-        }
-        args << Rt << ", [" << Rn << ", #" << imm5 << "]";
-      }
-    } else if (opcode1 >= 0x34 && opcode1 <= 0x37) {  // 1101xx
-      int8_t imm8 = instr & 0xFF;
-      uint32_t cond = (instr >> 8) & 0xF;
-      opcode << "b";
-      DumpCond(opcode, cond);
-      DumpBranchTarget(args, instr_ptr + 4, (imm8 << 1));
-    } else if ((instr & 0xF800) == 0xA800) {
-      // Generate SP-relative address
-      ThumbRegister rd(instr, 8);
-      int imm8 = instr & 0xFF;
-      opcode << "add";
-      args << rd << ", sp, #" << (imm8 << 2);
-    } else if ((instr & 0xF000) == 0xB000) {
-      // Miscellaneous 16-bit instructions
-      uint16_t opcode2 = (instr >> 5) & 0x7F;
-      switch (opcode2) {
-        case 0x00: case 0x01: case 0x02: case 0x03: case 0x04: case 0x05: case 0x06: case 0x07: {
-          // Add immediate to SP        - 1011 00000 ii iiiii
-          // Subtract immediate from SP - 1011 00001 ii iiiii
-          int imm7 = instr & 0x7F;
-          opcode << ((opcode2 & 4) == 0 ? "add" : "sub");
-          args << "sp, sp, #" << (imm7 << 2);
-          break;
-        }
-        case 0x08: case 0x09: case 0x0A: case 0x0B:  // 0001xxx
-        case 0x0C: case 0x0D: case 0x0E: case 0x0F:
-        case 0x18: case 0x19: case 0x1A: case 0x1B:  // 0011xxx
-        case 0x1C: case 0x1D: case 0x1E: case 0x1F:
-        case 0x48: case 0x49: case 0x4A: case 0x4B:  // 1001xxx
-        case 0x4C: case 0x4D: case 0x4E: case 0x4F:
-        case 0x58: case 0x59: case 0x5A: case 0x5B:  // 1011xxx
-        case 0x5C: case 0x5D: case 0x5E: case 0x5F: {
-          // CBNZ, CBZ
-          uint16_t op = (instr >> 11) & 1;
-          uint16_t i = (instr >> 9) & 1;
-          uint16_t imm5 = (instr >> 3) & 0x1F;
-          ThumbRegister Rn(instr, 0);
-          opcode << (op != 0 ? "cbnz" : "cbz");
-          uint32_t imm32 = (i << 6) | (imm5 << 1);
-          args << Rn << ", ";
-          DumpBranchTarget(args, instr_ptr + 4, imm32);
-          break;
-        }
-        case 0x20: case 0x21: case 0x22: case 0x23: case 0x24: case 0x25: case 0x26: case 0x27:
-        case 0x28: case 0x29: case 0x2A: case 0x2B: case 0x2C: case 0x2D: case 0x2E: case 0x2F: {
-          opcode << "push";
-          args << RegisterList((instr & 0xFF) | ((instr & 0x100) << 6));
-          break;
-        }
-        case 0x60: case 0x61: case 0x62: case 0x63: case 0x64: case 0x65: case 0x66: case 0x67:
-        case 0x68: case 0x69: case 0x6A: case 0x6B: case 0x6C: case 0x6D: case 0x6E: case 0x6F: {
-          opcode << "pop";
-          args << RegisterList((instr & 0xFF) | ((instr & 0x100) << 7));
-          break;
-        }
-        case 0x70: case 0x71: case 0x72: case 0x73: case 0x74: case 0x75: case 0x76: case 0x77: {
-          opcode << "bkpt";
-          args << "#" << (instr & 0xFF);
-          break;
-        }
-        case 0x50: case 0x51:    // 101000x
-        case 0x52: case 0x53:    // 101001x
-        case 0x56: case 0x57: {  // 101011x
-          uint16_t op = (instr >> 6) & 3;
-          opcode << kThumbReverseOperations[op];
-          ThumbRegister Rm(instr, 3);
-          ThumbRegister Rd(instr, 0);
-          args << Rd << ", " << Rm;
-          break;
-        }
-        case 0x78: case 0x79: case 0x7A: case 0x7B:  // 1111xxx
-        case 0x7C: case 0x7D: case 0x7E: case 0x7F: {
-          // If-Then, and hints
-          uint16_t opA = (instr >> 4) & 0xF;
-          uint16_t opB = instr & 0xF;
-          if (opB == 0) {
-            switch (opA) {
-              case 0: opcode << "nop"; break;
-              case 1: opcode << "yield"; break;
-              case 2: opcode << "wfe";  break;
-              case 3: opcode << "sev"; break;
-              default: break;
-            }
-          } else {
-            uint32_t first_cond = opA;
-            uint32_t mask = opB;
-            opcode << "it";
-
-            // Flesh out the base "it" opcode with the specific collection of 't's and 'e's,
-            // and store up the actual condition codes we'll want to add to the next few opcodes.
-            size_t count = 3 - CTZ(mask);
-            it_conditions_.resize(count + 2);  // Plus the implicit 't', plus the "" for the IT itself.
-            for (size_t i = 0; i < count; ++i) {
-              bool positive_cond = ((first_cond & 1) != 0);
-              bool positive_mask = ((mask & (1 << (3 - i))) != 0);
-              if (positive_mask == positive_cond) {
-                opcode << 't';
-                it_conditions_[i] = kConditionCodeNames[first_cond];
-              } else {
-                opcode << 'e';
-                it_conditions_[i] = kConditionCodeNames[first_cond ^ 1];
-              }
-            }
-            it_conditions_[count] = kConditionCodeNames[first_cond];  // The implicit 't'.
-
-            it_conditions_[count + 1] = "";  // No condition code for the IT itself...
-            DumpCond(args, first_cond);  // ...because it's considered an argument.
-          }
-          break;
-        }
-        default:
-          break;
-      }
-    } else if (((instr & 0xF000) == 0x5000) || ((instr & 0xE000) == 0x6000) ||
-        ((instr & 0xE000) == 0x8000)) {
-      // Load/store single data item
-      uint16_t opA = instr >> 12;
-      // uint16_t opB = (instr >> 9) & 7;
-      switch (opA) {
-        case 0x6: {
-          // STR Rt, [Rn, #imm] - 01100 iiiii nnn ttt
-          // LDR Rt, [Rn, #imm] - 01101 iiiii nnn ttt
-          uint16_t imm5 = (instr >> 6) & 0x1F;
-          ThumbRegister Rn(instr, 3);
-          ThumbRegister Rt(instr, 0);
-          opcode << ((instr & 0x800) == 0 ? "str" : "ldr");
-          args << Rt << ", [" << Rn << ", #" << (imm5 << 2) << "]";
-          break;
-        }
-        case 0x9: {
-          // STR Rt, [SP, #imm] - 01100 ttt iiiiiiii
-          // LDR Rt, [SP, #imm] - 01101 ttt iiiiiiii
-          uint16_t imm8 = instr & 0xFF;
-          ThumbRegister Rt(instr, 8);
-          opcode << ((instr & 0x800) == 0 ? "str" : "ldr");
-          args << Rt << ", [sp, #" << (imm8 << 2) << "]";
-          break;
-        }
-        default:
-          break;
-      }
-    } else if (opcode1 == 0x38 || opcode1 == 0x39) {
-      uint16_t imm11 = instr & 0x7FFF;
-      int32_t imm32 = imm11 << 1;
-      imm32 = (imm32 << 20) >> 20;  // sign extend 12 bit immediate
-      opcode << "b";
-      DumpBranchTarget(args, instr_ptr + 4, imm32);
-    }
-
-    // Apply any IT-block conditions to the opcode if necessary.
-    if (!it_conditions_.empty()) {
-      opcode << it_conditions_.back();
-      it_conditions_.pop_back();
-    }
-
-    os << FormatInstructionPointer(instr_ptr)
-       << StringPrintf(": %04x    \t%-7s ", instr, opcode.str().c_str())
-       << args.str() << '\n';
+    const uint32_t* const ip = reinterpret_cast<const uint32_t*>(instr_ptr);
+    next = reinterpret_cast<uintptr_t>(disasm_->DecodeA32At(ip));
   }
-  return 2;
+
+  os << output_.str();
+  output_.str(std::string());
+  return next - instr_ptr;
+}
+
+void DisassemblerArm::Dump(std::ostream& os, const uint8_t* begin, const uint8_t* end) {
+  DCHECK_LE(begin, end);
+
+  // Remove the Thumb specifier bit; no effect if begin does not point to T32 code.
+  const uintptr_t base = reinterpret_cast<uintptr_t>(begin) & ~1;
+
+  disasm_->SetT32((reinterpret_cast<uintptr_t>(begin) & 1) != 0);
+  disasm_->JumpToPc(GetPc(base));
+
+  if (disasm_->IsT32()) {
+    // The Thumb specifier bits cancel each other.
+    disasm_->DisassembleT32Buffer(reinterpret_cast<const uint16_t*>(base), end - begin);
+  } else {
+    disasm_->DisassembleA32Buffer(reinterpret_cast<const uint32_t*>(base), end - begin);
+  }
+
+  os << output_.str();
+  output_.str(std::string());
 }
 
 }  // namespace arm
diff --git a/disassembler/disassembler_arm.h b/disassembler/disassembler_arm.h
index f870e8e..237b577 100644
--- a/disassembler/disassembler_arm.h
+++ b/disassembler/disassembler_arm.h
@@ -17,32 +17,33 @@
 #ifndef ART_DISASSEMBLER_DISASSEMBLER_ARM_H_
 #define ART_DISASSEMBLER_DISASSEMBLER_ARM_H_
 
-#include <vector>
+#include <memory>
+#include <sstream>
 
+#include "base/macros.h"
 #include "disassembler.h"
 
 namespace art {
 namespace arm {
 
 class DisassemblerArm FINAL : public Disassembler {
+  class CustomDisassembler;
+
  public:
-  explicit DisassemblerArm(DisassemblerOptions* options) : Disassembler(options) {}
+  explicit DisassemblerArm(DisassemblerOptions* options);
 
   size_t Dump(std::ostream& os, const uint8_t* begin) OVERRIDE;
   void Dump(std::ostream& os, const uint8_t* begin, const uint8_t* end) OVERRIDE;
 
  private:
-  void DumpArm(std::ostream& os, const uint8_t* instr);
+  uintptr_t GetPc(uintptr_t instr_ptr) const {
+    return GetDisassemblerOptions()->absolute_addresses_
+        ? instr_ptr
+        : instr_ptr - reinterpret_cast<uintptr_t>(GetDisassemblerOptions()->base_address_);
+  }
 
-  // Returns the size of the instruction just decoded
-  size_t DumpThumb16(std::ostream& os, const uint8_t* instr);
-  size_t DumpThumb32(std::ostream& os, const uint8_t* instr_ptr);
-
-  void DumpBranchTarget(std::ostream& os, const uint8_t* instr_ptr, int32_t imm32);
-  void DumpCond(std::ostream& os, uint32_t cond);
-  void DumpMemoryDomain(std::ostream& os, uint32_t domain);
-
-  std::vector<const char*> it_conditions_;
+  std::ostringstream output_;
+  std::unique_ptr<CustomDisassembler> disasm_;
 
   DISALLOW_COPY_AND_ASSIGN(DisassemblerArm);
 };
diff --git a/runtime/common_runtime_test.cc b/runtime/common_runtime_test.cc
index 7234952..5bc4e88 100644
--- a/runtime/common_runtime_test.cc
+++ b/runtime/common_runtime_test.cc
@@ -409,6 +409,15 @@
   TearDownAndroidData(android_data_, true);
   dalvik_cache_.clear();
 
+  // icu4c has a fixed 10-element array "gCommonICUDataArray".
+  // If we run > 10 tests, we fill that array and u_setCommonData fails.
+  // There's a function to clear the array, but it's not public...
+  typedef void (*IcuCleanupFn)();
+  void* sym = dlsym(RTLD_DEFAULT, "u_cleanup_" U_ICU_VERSION_SHORT);
+  CHECK(sym != nullptr) << dlerror();
+  IcuCleanupFn icu_cleanup_fn = reinterpret_cast<IcuCleanupFn>(sym);
+  (*icu_cleanup_fn)();
+
   Runtime::Current()->GetHeap()->VerifyHeap();  // Check for heap corruption after the test
 }
 
diff --git a/test/538-checker-embed-constants/src/Main.java b/test/538-checker-embed-constants/src/Main.java
index 04a12fa..02c609e 100644
--- a/test/538-checker-embed-constants/src/Main.java
+++ b/test/538-checker-embed-constants/src/Main.java
@@ -37,7 +37,7 @@
   }
 
   /// CHECK-START-ARM: int Main.and511(int) disassembly (after)
-  /// CHECK:                movw {{r\d+}}, #511
+  /// CHECK:                mov {{r\d+}}, #511
   /// CHECK:                and{{(\.w)?}} {{r\d+}}, {{r\d+}}, {{r\d+}}
 
   public static int and511(int arg) {
@@ -61,7 +61,7 @@
   }
 
   /// CHECK-START-ARM: int Main.or511(int) disassembly (after)
-  /// CHECK:                movw {{r\d+}}, #511
+  /// CHECK:                mov {{r\d+}}, #511
   /// CHECK:                orr{{(\.w)?}} {{r\d+}}, {{r\d+}}, {{r\d+}}
 
   public static int or511(int arg) {
@@ -85,7 +85,7 @@
   }
 
   /// CHECK-START-ARM: int Main.xor511(int) disassembly (after)
-  /// CHECK:                movw {{r\d+}}, #511
+  /// CHECK:                mov {{r\d+}}, #511
   /// CHECK:                eor{{(\.w)?}} {{r\d+}}, {{r\d+}}, {{r\d+}}
 
   public static int xor511(int arg) {
@@ -114,7 +114,7 @@
   }
 
   /// CHECK-START-ARM: long Main.and511(long) disassembly (after)
-  /// CHECK:                movw {{r\d+}}, #511
+  /// CHECK:                mov {{r\d+}}, #511
   /// CHECK-NEXT:           movs {{r\d+}}, #0
   /// CHECK-NOT:            and{{(\.w)?}}
   /// CHECK-NOT:            bic{{(\.w)?}}
@@ -166,7 +166,7 @@
   }
 
   /// CHECK-START-ARM: long Main.or511(long) disassembly (after)
-  /// CHECK:                movw {{r\d+}}, #511
+  /// CHECK:                mov {{r\d+}}, #511
   /// CHECK-NEXT:           movs {{r\d+}}, #0
   /// CHECK-NOT:            orr{{(\.w)?}}
   /// CHECK-NOT:            orn
@@ -217,7 +217,7 @@
   }
 
   /// CHECK-START-ARM: long Main.xor511(long) disassembly (after)
-  /// CHECK:                movw {{r\d+}}, #511
+  /// CHECK:                mov {{r\d+}}, #511
   /// CHECK-NEXT:           movs {{r\d+}}, #0
   /// CHECK-NOT:            eor{{(\.w)?}}
   /// CHECK:                eor{{(\.w)?}} {{r\d+}}, {{r\d+}}, {{r\d+}}
@@ -230,7 +230,7 @@
 
   /// CHECK-START-ARM: long Main.xorNot15(long) disassembly (after)
   /// CHECK-DAG:            mvn {{r\d+}}, #15
-  /// CHECK-DAG:            mov.w {{r\d+}}, #-1
+  /// CHECK-DAG:            mov {{r\d+}}, #4294967295
   /// CHECK-NOT:            eor{{(\.w)?}}
   /// CHECK-DAG:            eor{{(\.w)?}} {{r\d+}}, {{r\d+}}, {{r\d+}}
   /// CHECK-DAG:            eor{{(\.w)?}} {{r\d+}}, {{r\d+}}, {{r\d+}}
@@ -258,7 +258,7 @@
   /// CHECK-NOT:            mov.w {{r\d+}}, #-268435456
   /// CHECK-NOT:            eor{{(\.w)?}}
   /// CHECK-DAG:            eor {{r\d+}}, {{r\d+}}, #15
-  /// CHECK-DAG:            eor {{r\d+}}, {{r\d+}}, #-268435456
+  /// CHECK-DAG:            eor {{r\d+}}, {{r\d+}}, #4026531840
   /// CHECK-NOT:            eor{{(\.w)?}}
 
   public static long xor0xf00000000000000f(long arg) {
@@ -285,7 +285,7 @@
 
   /// CHECK-START-ARM: long Main.shl2(long) disassembly (after)
   /// CHECK:                lsl{{s?|\.w}} <<oh:r\d+>>, {{r\d+}}, #2
-  /// CHECK:                orr.w <<oh>>, <<oh>>, <<low:r\d+>>, lsr #30
+  /// CHECK:                orr <<oh>>, <<low:r\d+>>, lsr #30
   /// CHECK:                lsl{{s?|\.w}} {{r\d+}}, <<low>>, #2
 
   /// CHECK-START-ARM: long Main.shl2(long) disassembly (after)
@@ -297,7 +297,7 @@
 
   /// CHECK-START-ARM: long Main.shl31(long) disassembly (after)
   /// CHECK:                lsl{{s?|\.w}} <<oh:r\d+>>, {{r\d+}}, #31
-  /// CHECK:                orr.w <<oh>>, <<oh>>, <<low:r\d+>>, lsr #1
+  /// CHECK:                orr <<oh>>, <<low:r\d+>>, lsr #1
   /// CHECK:                lsl{{s?|\.w}} {{r\d+}}, <<low>>, #31
 
   /// CHECK-START-ARM: long Main.shl31(long) disassembly (after)
@@ -342,7 +342,7 @@
 
   /// CHECK-START-ARM: long Main.shr1(long) disassembly (after)
   /// CHECK:                asrs{{(\.w)?}} {{r\d+}}, {{r\d+}}, #1
-  /// CHECK:                mov.w {{r\d+}}, {{r\d+}}, rrx
+  /// CHECK:                rrx {{r\d+}}, {{r\d+}}
 
   /// CHECK-START-ARM: long Main.shr1(long) disassembly (after)
   /// CHECK-NOT:            asr{{s?|\.w}} {{r\d+}}, {{r\d+}}, {{r\d+}}
@@ -353,7 +353,7 @@
 
   /// CHECK-START-ARM: long Main.shr2(long) disassembly (after)
   /// CHECK:                lsr{{s?|\.w}} <<ol:r\d+>>, {{r\d+}}, #2
-  /// CHECK:                orr.w <<ol>>, <<ol>>, <<high:r\d+>>, lsl #30
+  /// CHECK:                orr <<ol>>, <<high:r\d+>>, lsl #30
   /// CHECK-DAG:            asr{{s?|\.w}} {{r\d+}}, <<high>>, #2
 
   /// CHECK-START-ARM: long Main.shr2(long) disassembly (after)
@@ -365,7 +365,7 @@
 
   /// CHECK-START-ARM: long Main.shr31(long) disassembly (after)
   /// CHECK:                lsr{{s?|\.w}} <<ol:r\d+>>, {{r\d+}}, #31
-  /// CHECK:                orr.w <<ol>>, <<ol>>, <<high:r\d+>>, lsl #1
+  /// CHECK:                orr <<ol>>, <<high:r\d+>>, lsl #1
   /// CHECK:                asr{{s?|\.w}} {{r\d+}}, <<high>>, #31
 
   /// CHECK-START-ARM: long Main.shr31(long) disassembly (after)
@@ -411,7 +411,7 @@
 
   /// CHECK-START-ARM: long Main.ushr1(long) disassembly (after)
   /// CHECK:                lsrs{{|.w}} {{r\d+}}, {{r\d+}}, #1
-  /// CHECK:                mov.w {{r\d+}}, {{r\d+}}, rrx
+  /// CHECK:                rrx {{r\d+}}, {{r\d+}}
 
   /// CHECK-START-ARM: long Main.ushr1(long) disassembly (after)
   /// CHECK-NOT:            lsr{{s?|\.w}} {{r\d+}}, {{r\d+}}, {{r\d+}}
@@ -422,7 +422,7 @@
 
   /// CHECK-START-ARM: long Main.ushr2(long) disassembly (after)
   /// CHECK:                lsr{{s?|\.w}} <<ol:r\d+>>, {{r\d+}}, #2
-  /// CHECK:                orr.w <<ol>>, <<ol>>, <<high:r\d+>>, lsl #30
+  /// CHECK:                orr <<ol>>, <<high:r\d+>>, lsl #30
   /// CHECK-DAG:            lsr{{s?|\.w}} {{r\d+}}, <<high>>, #2
 
   /// CHECK-START-ARM: long Main.ushr2(long) disassembly (after)
@@ -434,7 +434,7 @@
 
   /// CHECK-START-ARM: long Main.ushr31(long) disassembly (after)
   /// CHECK:                lsr{{s?|\.w}} <<ol:r\d+>>, {{r\d+}}, #31
-  /// CHECK:                orr.w <<ol>>, <<ol>>, <<high:r\d+>>, lsl #1
+  /// CHECK:                orr <<ol>>, <<high:r\d+>>, lsl #1
   /// CHECK:                lsr{{s?|\.w}} {{r\d+}}, <<high>>, #31
 
   /// CHECK-START-ARM: long Main.ushr31(long) disassembly (after)
@@ -508,10 +508,10 @@
   /// CHECK:     <<ConstM1:j\d+>>   LongConstant -1
   /// CHECK:                        Add [<<Arg>>,<<ConstM1>>]
   /// CHECK-NEXT:                   subs r{{\d+}}, #1
-  /// CHECK-NEXT:                   adc r{{\d+}}, r{{\d+}}, #-1
+  /// CHECK-NEXT:                   adc r{{\d+}}, r{{\d+}}, #4294967295
   /// CHECK:                        Sub [<<Arg>>,<<ConstM1>>]
   /// CHECK-NEXT:                   adds r{{\d+}}, #1
-  /// CHECK-NEXT:                   adc r{{\d+}}, r{{\d+}}, #0
+  /// CHECK-NEXT:                   adc r{{\d+}}, #0
 
   public static long addM1(long arg) {
     return (arg + (-1)) | (arg - (-1));
@@ -542,10 +542,10 @@
   /// CHECK-NEXT:                   sbc r{{\d+}}, r{{\d+}}, #249561088
   /// CHECK:                        Add [<<Arg>>,<<ConstD>>]
   // There may or may not be a MOV here.
-  /// CHECK:                        addw r{{\d+}}, r{{\d+}}, #4095
+  /// CHECK:                        add r{{\d+}}, r{{\d+}}, #4095
   /// CHECK:                        Add [<<Arg>>,<<ConstE>>]
   // There may or may not be a MOV here.
-  /// CHECK:                        subw r{{\d+}}, r{{\d+}}, #2051
+  /// CHECK:                        sub r{{\d+}}, r{{\d+}}, #2051
   /// CHECK:                        Add [<<Arg>>,<<ConstF>>]
   /// CHECK-NEXT:                   adds{{(\.w)?}} r{{\d+}}, r{{\d+}}, r{{\d+}}
   /// CHECK-NEXT:                   adc{{(\.w)?}} r{{\d+}}, r{{\d+}}, r{{\d+}}
@@ -597,10 +597,10 @@
   /// CHECK-NEXT:                   adc r{{\d+}}, r{{\d+}}, #249561088
   /// CHECK:                        Sub [<<Arg>>,<<ConstD>>]
   // There may or may not be a MOV here.
-  /// CHECK:                        subw r{{\d+}}, r{{\d+}}, #4095
+  /// CHECK:                        sub r{{\d+}}, r{{\d+}}, #4095
   /// CHECK:                        Sub [<<Arg>>,<<ConstE>>]
   // There may or may not be a MOV here.
-  /// CHECK:                        addw r{{\d+}}, r{{\d+}}, #2051
+  /// CHECK:                        add r{{\d+}}, r{{\d+}}, #2051
   /// CHECK:                        Sub [<<Arg>>,<<ConstF>>]
   /// CHECK-NEXT:                   subs{{(\.w)?}} r{{\d+}}, r{{\d+}}, r{{\d+}}
   /// CHECK-NEXT:                   sbc{{(\.w)?}} r{{\d+}}, r{{\d+}}, r{{\d+}}
diff --git a/test/564-checker-negbitwise/src/Main.java b/test/564-checker-negbitwise/src/Main.java
index ccb8ff4..a047d21 100644
--- a/test/564-checker-negbitwise/src/Main.java
+++ b/test/564-checker-negbitwise/src/Main.java
@@ -74,7 +74,7 @@
   /// CHECK-NOT:                        And
 
   /// CHECK-START-ARM:   int Main.$opt$noinline$notAnd(int, int) disassembly (after)
-  /// CHECK:                            bic.w r{{\d+}}, r{{\d+}}, r{{\d+}}
+  /// CHECK:                            bic r{{\d+}}, r{{\d+}}, r{{\d+}}
 
   public static int $opt$noinline$notAnd(int base, int mask) {
     if (doThrow) throw new Error();
@@ -124,7 +124,7 @@
   /// CHECK-NOT:                        Or
 
   /// CHECK-START-ARM:   long Main.$opt$noinline$notOr(long, long) disassembly (after)
-  /// CHECK:                            orn.w r{{\d+}}, r{{\d+}}, r{{\d+}}
+  /// CHECK:                            orn r{{\d+}}, r{{\d+}}, r{{\d+}}
 
   public static long $opt$noinline$notOr(long base, long mask) {
     if (doThrow) throw new Error();
diff --git a/tools/ahat/Android.mk b/tools/ahat/Android.mk
index 60e0cd8..4003ee0 100644
--- a/tools/ahat/Android.mk
+++ b/tools/ahat/Android.mk
@@ -16,7 +16,7 @@
 
 LOCAL_PATH := $(call my-dir)
 
-include art/build/Android.common_test.mk
+include art/build/Android.common_path.mk
 
 # --- ahat.jar ----------------
 include $(CLEAR_VARS)