Change check cast entrypoint to check instance of

Reduces code size since we do not need to reload class before
calling slow path.

TODO: Delete read barriers in the check cast code since the slow
path will retry with the proper read barriers if the check fails.

Bug: 12687968
Bug: 29516974

Test: test-art-host + test-art-target with CC

Change-Id: Ia4eb9bbe3fe2d2016e44523cf0451210828d7b88
diff --git a/compiler/optimizing/code_generator_arm.cc b/compiler/optimizing/code_generator_arm.cc
index 7cab97d..08d22f8 100644
--- a/compiler/optimizing/code_generator_arm.cc
+++ b/compiler/optimizing/code_generator_arm.cc
@@ -489,8 +489,14 @@
 
   void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
     LocationSummary* locations = instruction_->GetLocations();
-    Location object_class = instruction_->IsCheckCast() ? locations->GetTemp(0)
-                                                        : locations->Out();
+    Location arg0, arg1;
+    if (instruction_->IsInstanceOf()) {
+      arg0 = locations->InAt(1);
+      arg1 = locations->Out();
+    } else {
+      arg0 = locations->InAt(0);
+      arg1 = locations->InAt(1);
+    }
     DCHECK(instruction_->IsCheckCast()
            || !locations->GetLiveRegisters()->ContainsCoreRegister(locations->Out().reg()));
 
@@ -504,26 +510,26 @@
     // We're moving two locations to locations that could overlap, so we need a parallel
     // move resolver.
     InvokeRuntimeCallingConvention calling_convention;
-    codegen->EmitParallelMoves(
-        locations->InAt(1),
-        Location::RegisterLocation(calling_convention.GetRegisterAt(0)),
-        Primitive::kPrimNot,
-        object_class,
-        Location::RegisterLocation(calling_convention.GetRegisterAt(1)),
-        Primitive::kPrimNot);
-
+    codegen->EmitParallelMoves(arg0,
+                               Location::RegisterLocation(calling_convention.GetRegisterAt(0)),
+                               Primitive::kPrimNot,
+                               arg1,
+                               Location::RegisterLocation(calling_convention.GetRegisterAt(1)),
+                               Primitive::kPrimNot);
     if (instruction_->IsInstanceOf()) {
       arm_codegen->InvokeRuntime(kQuickInstanceofNonTrivial,
                                  instruction_,
                                  instruction_->GetDexPc(),
                                  this);
-      CheckEntrypointTypes<
-          kQuickInstanceofNonTrivial, size_t, const mirror::Class*, const mirror::Class*>();
+      CheckEntrypointTypes<kQuickInstanceofNonTrivial, size_t, mirror::Class*, mirror::Class*>();
       arm_codegen->Move32(locations->Out(), Location::RegisterLocation(R0));
     } else {
       DCHECK(instruction_->IsCheckCast());
-      arm_codegen->InvokeRuntime(kQuickCheckCast, instruction_, instruction_->GetDexPc(), this);
-      CheckEntrypointTypes<kQuickCheckCast, void, const mirror::Class*, const mirror::Class*>();
+      arm_codegen->InvokeRuntime(kQuickCheckInstanceOf,
+                                 instruction_,
+                                 instruction_->GetDexPc(),
+                                 this);
+      CheckEntrypointTypes<kQuickCheckInstanceOf, void, mirror::Object*, mirror::Class*>();
     }
 
     if (!is_fatal_) {
@@ -6297,26 +6303,16 @@
     case TypeCheckKind::kAbstractClassCheck: {
       // If the class is abstract, we eagerly fetch the super class of the
       // object to avoid doing a comparison we know will fail.
-      Label loop, compare_classes;
+      Label loop;
       __ Bind(&loop);
       // /* HeapReference<Class> */ temp = temp->super_class_
       GenerateReferenceLoadOneRegister(instruction, temp_loc, super_offset, maybe_temp2_loc);
 
-      // If the class reference currently in `temp` is not null, jump
-      // to the `compare_classes` label to compare it with the checked
-      // class.
-      __ CompareAndBranchIfNonZero(temp, &compare_classes);
-      // Otherwise, jump to the slow path to throw the exception.
-      //
-      // But before, move back the object's class into `temp` before
-      // going into the slow path, as it has been overwritten in the
-      // meantime.
-      // /* HeapReference<Class> */ temp = obj->klass_
-      GenerateReferenceLoadTwoRegisters(
-          instruction, temp_loc, obj_loc, class_offset, maybe_temp2_loc);
-      __ b(type_check_slow_path->GetEntryLabel());
+      // If the class reference currently in `temp` is null, jump to the slow path to throw the
+      // exception.
+      __ CompareAndBranchIfZero(temp, type_check_slow_path->GetEntryLabel());
 
-      __ Bind(&compare_classes);
+      // Otherwise, compare the classes.
       __ cmp(temp, ShifterOperand(cls));
       __ b(&loop, NE);
       break;
@@ -6332,55 +6328,29 @@
       // /* HeapReference<Class> */ temp = temp->super_class_
       GenerateReferenceLoadOneRegister(instruction, temp_loc, super_offset, maybe_temp2_loc);
 
-      // If the class reference currently in `temp` is not null, jump
-      // back at the beginning of the loop.
-      __ CompareAndBranchIfNonZero(temp, &loop);
-      // Otherwise, jump to the slow path to throw the exception.
-      //
-      // But before, move back the object's class into `temp` before
-      // going into the slow path, as it has been overwritten in the
-      // meantime.
-      // /* HeapReference<Class> */ temp = obj->klass_
-      GenerateReferenceLoadTwoRegisters(
-          instruction, temp_loc, obj_loc, class_offset, maybe_temp2_loc);
-      __ b(type_check_slow_path->GetEntryLabel());
+      // If the class reference currently in `temp` is null, jump to the slow path to throw the
+      // exception.
+      __ CompareAndBranchIfZero(temp, type_check_slow_path->GetEntryLabel());
+      // Otherwise, jump to the beginning of the loop.
+      __ b(&loop);
       break;
     }
 
     case TypeCheckKind::kArrayObjectCheck: {
       // Do an exact check.
-      Label check_non_primitive_component_type;
       __ cmp(temp, ShifterOperand(cls));
       __ b(&done, EQ);
 
       // Otherwise, we need to check that the object's class is a non-primitive array.
       // /* HeapReference<Class> */ temp = temp->component_type_
       GenerateReferenceLoadOneRegister(instruction, temp_loc, component_offset, maybe_temp2_loc);
-
-      // If the component type is not null (i.e. the object is indeed
-      // an array), jump to label `check_non_primitive_component_type`
-      // to further check that this component type is not a primitive
-      // type.
-      __ CompareAndBranchIfNonZero(temp, &check_non_primitive_component_type);
-      // Otherwise, jump to the slow path to throw the exception.
-      //
-      // But before, move back the object's class into `temp` before
-      // going into the slow path, as it has been overwritten in the
-      // meantime.
-      // /* HeapReference<Class> */ temp = obj->klass_
-      GenerateReferenceLoadTwoRegisters(
-          instruction, temp_loc, obj_loc, class_offset, maybe_temp2_loc);
-      __ b(type_check_slow_path->GetEntryLabel());
-
-      __ Bind(&check_non_primitive_component_type);
+      // If the component type is null, jump to the slow path to throw the exception.
+      __ CompareAndBranchIfZero(temp, type_check_slow_path->GetEntryLabel());
+      // Otherwise,the object is indeed an array, jump to label `check_non_primitive_component_type`
+      // to further check that this component type is not a primitive type.
       __ LoadFromOffset(kLoadUnsignedHalfword, temp, temp, primitive_offset);
       static_assert(Primitive::kPrimNot == 0, "Expected 0 for art::Primitive::kPrimNot");
-      __ CompareAndBranchIfZero(temp, &done);
-      // Same comment as above regarding `temp` and the slow path.
-      // /* HeapReference<Class> */ temp = obj->klass_
-      GenerateReferenceLoadTwoRegisters(
-          instruction, temp_loc, obj_loc, class_offset, maybe_temp2_loc);
-      __ b(type_check_slow_path->GetEntryLabel());
+      __ CompareAndBranchIfNonZero(temp, type_check_slow_path->GetEntryLabel());
       break;
     }
 
@@ -6396,13 +6366,6 @@
       // instruction (following the runtime calling convention), which
       // might be cluttered by the potential first read barrier
       // emission at the beginning of this method.
-      //
-      // TODO: Introduce a new runtime entry point taking the object
-      // to test (instead of its class) as argument, and let it deal
-      // with the read barrier issues. This will let us refactor this
-      // case of the `switch` code as it was previously (with a direct
-      // call to the runtime not using a type checking slow path).
-      // This should also be beneficial for the other cases above.
       __ b(type_check_slow_path->GetEntryLabel());
       break;
   }
diff --git a/compiler/optimizing/code_generator_arm64.cc b/compiler/optimizing/code_generator_arm64.cc
index d868984..f5119df 100644
--- a/compiler/optimizing/code_generator_arm64.cc
+++ b/compiler/optimizing/code_generator_arm64.cc
@@ -459,9 +459,15 @@
 
   void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
     LocationSummary* locations = instruction_->GetLocations();
-    Location class_to_check = locations->InAt(1);
-    Location object_class = instruction_->IsCheckCast() ? locations->GetTemp(0)
-                                                        : locations->Out();
+    Location arg0, arg1;
+    if (instruction_->IsInstanceOf()) {
+      arg0 = locations->InAt(1);
+      arg1 = locations->Out();
+    } else {
+      arg0 = locations->InAt(0);
+      arg1 = locations->InAt(1);
+    }
+
     DCHECK(instruction_->IsCheckCast()
            || !locations->GetLiveRegisters()->ContainsCoreRegister(locations->Out().reg()));
     CodeGeneratorARM64* arm64_codegen = down_cast<CodeGeneratorARM64*>(codegen);
@@ -476,21 +482,22 @@
     // We're moving two locations to locations that could overlap, so we need a parallel
     // move resolver.
     InvokeRuntimeCallingConvention calling_convention;
-    codegen->EmitParallelMoves(
-        class_to_check, LocationFrom(calling_convention.GetRegisterAt(0)), Primitive::kPrimNot,
-        object_class, LocationFrom(calling_convention.GetRegisterAt(1)), Primitive::kPrimNot);
-
+    codegen->EmitParallelMoves(arg0,
+                               LocationFrom(calling_convention.GetRegisterAt(0)),
+                               Primitive::kPrimNot,
+                               arg1,
+                               LocationFrom(calling_convention.GetRegisterAt(1)),
+                               Primitive::kPrimNot);
     if (instruction_->IsInstanceOf()) {
       arm64_codegen->InvokeRuntime(kQuickInstanceofNonTrivial, instruction_, dex_pc, this);
-      CheckEntrypointTypes<kQuickInstanceofNonTrivial, size_t,
-                           const mirror::Class*, const mirror::Class*>();
+      CheckEntrypointTypes<kQuickInstanceofNonTrivial, size_t, mirror::Class*, mirror::Class*>();
       Primitive::Type ret_type = instruction_->GetType();
       Location ret_loc = calling_convention.GetReturnLocation(ret_type);
       arm64_codegen->MoveLocation(locations->Out(), ret_loc, ret_type);
     } else {
       DCHECK(instruction_->IsCheckCast());
-      arm64_codegen->InvokeRuntime(kQuickCheckCast, instruction_, dex_pc, this);
-      CheckEntrypointTypes<kQuickCheckCast, void, const mirror::Class*, const mirror::Class*>();
+      arm64_codegen->InvokeRuntime(kQuickCheckInstanceOf, instruction_, dex_pc, this);
+      CheckEntrypointTypes<kQuickCheckInstanceOf, void, mirror::Object*, mirror::Class*>();
     }
 
     if (!is_fatal_) {
@@ -3594,26 +3601,15 @@
     case TypeCheckKind::kAbstractClassCheck: {
       // If the class is abstract, we eagerly fetch the super class of the
       // object to avoid doing a comparison we know will fail.
-      vixl::aarch64::Label loop, compare_classes;
+      vixl::aarch64::Label loop;
       __ Bind(&loop);
       // /* HeapReference<Class> */ temp = temp->super_class_
       GenerateReferenceLoadOneRegister(instruction, temp_loc, super_offset, maybe_temp2_loc);
 
-      // If the class reference currently in `temp` is not null, jump
-      // to the `compare_classes` label to compare it with the checked
-      // class.
-      __ Cbnz(temp, &compare_classes);
-      // Otherwise, jump to the slow path to throw the exception.
-      //
-      // But before, move back the object's class into `temp` before
-      // going into the slow path, as it has been overwritten in the
-      // meantime.
-      // /* HeapReference<Class> */ temp = obj->klass_
-      GenerateReferenceLoadTwoRegisters(
-          instruction, temp_loc, obj_loc, class_offset, maybe_temp2_loc);
-      __ B(type_check_slow_path->GetEntryLabel());
-
-      __ Bind(&compare_classes);
+      // If the class reference currently in `temp` is null, jump to the slow path to throw the
+      // exception.
+      __ Cbz(temp, type_check_slow_path->GetEntryLabel());
+      // Otherwise, compare classes.
       __ Cmp(temp, cls);
       __ B(ne, &loop);
       break;
@@ -3633,20 +3629,12 @@
       // back at the beginning of the loop.
       __ Cbnz(temp, &loop);
       // Otherwise, jump to the slow path to throw the exception.
-      //
-      // But before, move back the object's class into `temp` before
-      // going into the slow path, as it has been overwritten in the
-      // meantime.
-      // /* HeapReference<Class> */ temp = obj->klass_
-      GenerateReferenceLoadTwoRegisters(
-          instruction, temp_loc, obj_loc, class_offset, maybe_temp2_loc);
       __ B(type_check_slow_path->GetEntryLabel());
       break;
     }
 
     case TypeCheckKind::kArrayObjectCheck: {
       // Do an exact check.
-      vixl::aarch64::Label check_non_primitive_component_type;
       __ Cmp(temp, cls);
       __ B(eq, &done);
 
@@ -3654,30 +3642,13 @@
       // /* HeapReference<Class> */ temp = temp->component_type_
       GenerateReferenceLoadOneRegister(instruction, temp_loc, component_offset, maybe_temp2_loc);
 
-      // If the component type is not null (i.e. the object is indeed
-      // an array), jump to label `check_non_primitive_component_type`
-      // to further check that this component type is not a primitive
-      // type.
-      __ Cbnz(temp, &check_non_primitive_component_type);
-      // Otherwise, jump to the slow path to throw the exception.
-      //
-      // But before, move back the object's class into `temp` before
-      // going into the slow path, as it has been overwritten in the
-      // meantime.
-      // /* HeapReference<Class> */ temp = obj->klass_
-      GenerateReferenceLoadTwoRegisters(
-          instruction, temp_loc, obj_loc, class_offset, maybe_temp2_loc);
-      __ B(type_check_slow_path->GetEntryLabel());
-
-      __ Bind(&check_non_primitive_component_type);
+      // If the component type is null, jump to the slow path to throw the exception.
+      __ Cbz(temp, type_check_slow_path->GetEntryLabel());
+      // Otherwise, the object is indeed an array. Further check that this component type is not a
+      // primitive type.
       __ Ldrh(temp, HeapOperand(temp, primitive_offset));
       static_assert(Primitive::kPrimNot == 0, "Expected 0 for kPrimNot");
-      __ Cbz(temp, &done);
-      // Same comment as above regarding `temp` and the slow path.
-      // /* HeapReference<Class> */ temp = obj->klass_
-      GenerateReferenceLoadTwoRegisters(
-          instruction, temp_loc, obj_loc, class_offset, maybe_temp2_loc);
-      __ B(type_check_slow_path->GetEntryLabel());
+      __ Cbnz(temp, type_check_slow_path->GetEntryLabel());
       break;
     }
 
@@ -3693,13 +3664,6 @@
       // instruction (following the runtime calling convention), which
       // might be cluttered by the potential first read barrier
       // emission at the beginning of this method.
-      //
-      // TODO: Introduce a new runtime entry point taking the object
-      // to test (instead of its class) as argument, and let it deal
-      // with the read barrier issues. This will let us refactor this
-      // case of the `switch` code as it was previously (with a direct
-      // call to the runtime not using a type checking slow path).
-      // This should also be beneficial for the other cases above.
       __ B(type_check_slow_path->GetEntryLabel());
       break;
   }
diff --git a/compiler/optimizing/code_generator_arm_vixl.cc b/compiler/optimizing/code_generator_arm_vixl.cc
index e69528e..87f6c89 100644
--- a/compiler/optimizing/code_generator_arm_vixl.cc
+++ b/compiler/optimizing/code_generator_arm_vixl.cc
@@ -443,8 +443,14 @@
 
   void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
     LocationSummary* locations = instruction_->GetLocations();
-    Location object_class = instruction_->IsCheckCast() ? locations->GetTemp(0)
-                                                        : locations->Out();
+    Location arg0, arg1;
+    if (instruction_->IsInstanceOf()) {
+      arg0 = locations->InAt(1);
+      arg1 = locations->Out();
+    } else {
+      arg0 = locations->InAt(0);
+      arg1 = locations->InAt(1);
+    }
     DCHECK(instruction_->IsCheckCast()
            || !locations->GetLiveRegisters()->ContainsCoreRegister(locations->Out().reg()));
 
@@ -458,20 +464,22 @@
     // We're moving two locations to locations that could overlap, so we need a parallel
     // move resolver.
     InvokeRuntimeCallingConventionARMVIXL calling_convention;
-    codegen->EmitParallelMoves(
-        locations->InAt(1),
-        LocationFrom(calling_convention.GetRegisterAt(0)),
-        Primitive::kPrimNot,
-        object_class,
-        LocationFrom(calling_convention.GetRegisterAt(1)),
-        Primitive::kPrimNot);
 
+    codegen->EmitParallelMoves(arg0,
+                               LocationFrom(calling_convention.GetRegisterAt(0)),
+                               Primitive::kPrimNot,
+                               arg1,
+                               LocationFrom(calling_convention.GetRegisterAt(1)),
+                               Primitive::kPrimNot);
     if (instruction_->IsInstanceOf()) {
       TODO_VIXL32(FATAL);
     } else {
       DCHECK(instruction_->IsCheckCast());
-      arm_codegen->InvokeRuntime(kQuickCheckCast, instruction_, instruction_->GetDexPc(), this);
-      CheckEntrypointTypes<kQuickCheckCast, void, const mirror::Class*, const mirror::Class*>();
+      arm_codegen->InvokeRuntime(kQuickCheckInstanceOf,
+                                 instruction_,
+                                 instruction_->GetDexPc(),
+                                 this);
+      CheckEntrypointTypes<kQuickCheckInstanceOf, void, mirror::Object*, mirror::Class*>();
     }
 
     if (!is_fatal_) {
diff --git a/compiler/optimizing/code_generator_mips.cc b/compiler/optimizing/code_generator_mips.cc
index 12b1ab9..6e9fbd2 100644
--- a/compiler/optimizing/code_generator_mips.cc
+++ b/compiler/optimizing/code_generator_mips.cc
@@ -378,7 +378,14 @@
 
   void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
     LocationSummary* locations = instruction_->GetLocations();
-    Location object_class = instruction_->IsCheckCast() ? locations->GetTemp(0) : locations->Out();
+    Location arg0, arg1;
+    if (instruction_->IsInstanceOf()) {
+      arg0 = locations->InAt(1);
+      arg1 = locations->Out();
+    } else {
+      arg0 = locations->InAt(0);
+      arg1 = locations->InAt(1);
+    }
     uint32_t dex_pc = instruction_->GetDexPc();
     DCHECK(instruction_->IsCheckCast()
            || !locations->GetLiveRegisters()->ContainsCoreRegister(locations->Out().reg()));
@@ -390,24 +397,22 @@
     // We're moving two locations to locations that could overlap, so we need a parallel
     // move resolver.
     InvokeRuntimeCallingConvention calling_convention;
-    codegen->EmitParallelMoves(locations->InAt(1),
+    codegen->EmitParallelMoves(arg0,
                                Location::RegisterLocation(calling_convention.GetRegisterAt(0)),
                                Primitive::kPrimNot,
-                               object_class,
+                               arg1,
                                Location::RegisterLocation(calling_convention.GetRegisterAt(1)),
                                Primitive::kPrimNot);
-
     if (instruction_->IsInstanceOf()) {
       mips_codegen->InvokeRuntime(kQuickInstanceofNonTrivial, instruction_, dex_pc, this);
-      CheckEntrypointTypes<
-          kQuickInstanceofNonTrivial, size_t, const mirror::Class*, const mirror::Class*>();
+      CheckEntrypointTypes<kQuickInstanceofNonTrivial, size_t, mirror::Class*, mirror::Class*>();
       Primitive::Type ret_type = instruction_->GetType();
       Location ret_loc = calling_convention.GetReturnLocation(ret_type);
       mips_codegen->MoveLocation(locations->Out(), ret_loc, ret_type);
     } else {
       DCHECK(instruction_->IsCheckCast());
-      mips_codegen->InvokeRuntime(kQuickCheckCast, instruction_, dex_pc, this);
-      CheckEntrypointTypes<kQuickCheckCast, void, const mirror::Class*, const mirror::Class*>();
+      mips_codegen->InvokeRuntime(kQuickCheckInstanceOf, instruction_, dex_pc, this);
+      CheckEntrypointTypes<kQuickCheckInstanceOf, void, mirror::Object*, mirror::Class*>();
     }
 
     RestoreLiveRegisters(codegen, locations);
diff --git a/compiler/optimizing/code_generator_mips64.cc b/compiler/optimizing/code_generator_mips64.cc
index 010bf24..7598740 100644
--- a/compiler/optimizing/code_generator_mips64.cc
+++ b/compiler/optimizing/code_generator_mips64.cc
@@ -322,7 +322,15 @@
 
   void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
     LocationSummary* locations = instruction_->GetLocations();
-    Location object_class = instruction_->IsCheckCast() ? locations->GetTemp(0) : locations->Out();
+    Location arg0, arg1;
+    if (instruction_->IsInstanceOf()) {
+      arg0 = locations->InAt(1);
+      arg1 = locations->Out();
+    } else {
+      arg0 = locations->InAt(0);
+      arg1 = locations->InAt(1);
+    }
+
     uint32_t dex_pc = instruction_->GetDexPc();
     DCHECK(instruction_->IsCheckCast()
            || !locations->GetLiveRegisters()->ContainsCoreRegister(locations->Out().reg()));
@@ -334,24 +342,23 @@
     // We're moving two locations to locations that could overlap, so we need a parallel
     // move resolver.
     InvokeRuntimeCallingConvention calling_convention;
-    codegen->EmitParallelMoves(locations->InAt(1),
+    codegen->EmitParallelMoves(arg0,
                                Location::RegisterLocation(calling_convention.GetRegisterAt(0)),
                                Primitive::kPrimNot,
-                               object_class,
+                               arg1,
                                Location::RegisterLocation(calling_convention.GetRegisterAt(1)),
                                Primitive::kPrimNot);
-
     if (instruction_->IsInstanceOf()) {
       mips64_codegen->InvokeRuntime(kQuickInstanceofNonTrivial, instruction_, dex_pc, this);
       CheckEntrypointTypes<
-          kQuickInstanceofNonTrivial, size_t, const mirror::Class*, const mirror::Class*>();
+          kQuickInstanceofNonTrivial, size_t, mirror::Class*, mirror::Class*>();
       Primitive::Type ret_type = instruction_->GetType();
       Location ret_loc = calling_convention.GetReturnLocation(ret_type);
       mips64_codegen->MoveLocation(locations->Out(), ret_loc, ret_type);
     } else {
       DCHECK(instruction_->IsCheckCast());
-      mips64_codegen->InvokeRuntime(kQuickCheckCast, instruction_, dex_pc, this);
-      CheckEntrypointTypes<kQuickCheckCast, void, const mirror::Class*, const mirror::Class*>();
+      mips64_codegen->InvokeRuntime(kQuickCheckInstanceOf, instruction_, dex_pc, this);
+      CheckEntrypointTypes<kQuickCheckInstanceOf, void, mirror::Object*, mirror::Class*>();
     }
 
     RestoreLiveRegisters(codegen, locations);
diff --git a/compiler/optimizing/code_generator_x86.cc b/compiler/optimizing/code_generator_x86.cc
index 2f946e4..c902279 100644
--- a/compiler/optimizing/code_generator_x86.cc
+++ b/compiler/optimizing/code_generator_x86.cc
@@ -312,8 +312,14 @@
 
   void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
     LocationSummary* locations = instruction_->GetLocations();
-    Location object_class = instruction_->IsCheckCast() ? locations->GetTemp(0)
-                                                        : locations->Out();
+    Location arg0, arg1;
+    if (instruction_->IsInstanceOf()) {
+      arg0 = locations->InAt(1);
+      arg1 = locations->Out();
+    } else {
+      arg0 = locations->InAt(0);
+      arg1 = locations->InAt(1);
+    }
     DCHECK(instruction_->IsCheckCast()
            || !locations->GetLiveRegisters()->ContainsCoreRegister(locations->Out().reg()));
 
@@ -327,25 +333,25 @@
     // We're moving two locations to locations that could overlap, so we need a parallel
     // move resolver.
     InvokeRuntimeCallingConvention calling_convention;
-    x86_codegen->EmitParallelMoves(
-        locations->InAt(1),
-        Location::RegisterLocation(calling_convention.GetRegisterAt(0)),
-        Primitive::kPrimNot,
-        object_class,
-        Location::RegisterLocation(calling_convention.GetRegisterAt(1)),
-        Primitive::kPrimNot);
-
+    x86_codegen->EmitParallelMoves(arg0,
+                                   Location::RegisterLocation(calling_convention.GetRegisterAt(0)),
+                                   Primitive::kPrimNot,
+                                   arg1,
+                                   Location::RegisterLocation(calling_convention.GetRegisterAt(1)),
+                                   Primitive::kPrimNot);
     if (instruction_->IsInstanceOf()) {
       x86_codegen->InvokeRuntime(kQuickInstanceofNonTrivial,
                                  instruction_,
                                  instruction_->GetDexPc(),
                                  this);
-      CheckEntrypointTypes<
-          kQuickInstanceofNonTrivial, size_t, const mirror::Class*, const mirror::Class*>();
+      CheckEntrypointTypes<kQuickInstanceofNonTrivial, size_t, mirror::Class*, mirror::Class*>();
     } else {
       DCHECK(instruction_->IsCheckCast());
-      x86_codegen->InvokeRuntime(kQuickCheckCast, instruction_, instruction_->GetDexPc(), this);
-      CheckEntrypointTypes<kQuickCheckCast, void, const mirror::Class*, const mirror::Class*>();
+      x86_codegen->InvokeRuntime(kQuickCheckInstanceOf,
+                                 instruction_,
+                                 instruction_->GetDexPc(),
+                                 this);
+      CheckEntrypointTypes<kQuickCheckInstanceOf, void, mirror::Object*, mirror::Class*>();
     }
 
     if (!is_fatal_) {
@@ -6645,26 +6651,17 @@
     case TypeCheckKind::kAbstractClassCheck: {
       // If the class is abstract, we eagerly fetch the super class of the
       // object to avoid doing a comparison we know will fail.
-      NearLabel loop, compare_classes;
+      NearLabel loop;
       __ Bind(&loop);
       // /* HeapReference<Class> */ temp = temp->super_class_
       GenerateReferenceLoadOneRegister(instruction, temp_loc, super_offset, maybe_temp2_loc);
 
-      // If the class reference currently in `temp` is not null, jump
-      // to the `compare_classes` label to compare it with the checked
-      // class.
+      // If the class reference currently in `temp` is null, jump to the slow path to throw the
+      // exception.
       __ testl(temp, temp);
-      __ j(kNotEqual, &compare_classes);
-      // Otherwise, jump to the slow path to throw the exception.
-      //
-      // But before, move back the object's class into `temp` before
-      // going into the slow path, as it has been overwritten in the
-      // meantime.
-      // /* HeapReference<Class> */ temp = obj->klass_
-      GenerateReferenceLoadTwoRegisters(instruction, temp_loc, obj_loc, class_offset);
-      __ jmp(type_check_slow_path->GetEntryLabel());
+      __ j(kZero, type_check_slow_path->GetEntryLabel());
 
-      __ Bind(&compare_classes);
+      // Otherwise, compare the classes
       if (cls.IsRegister()) {
         __ cmpl(temp, cls.AsRegister<Register>());
       } else {
@@ -6693,21 +6690,14 @@
       // If the class reference currently in `temp` is not null, jump
       // back at the beginning of the loop.
       __ testl(temp, temp);
-      __ j(kNotEqual, &loop);
-      // Otherwise, jump to the slow path to throw the exception.
-      //
-      // But before, move back the object's class into `temp` before
-      // going into the slow path, as it has been overwritten in the
-      // meantime.
-      // /* HeapReference<Class> */ temp = obj->klass_
-      GenerateReferenceLoadTwoRegisters(instruction, temp_loc, obj_loc, class_offset);
+      __ j(kNotZero, &loop);
+      // Otherwise, jump to the slow path to throw the exception.;
       __ jmp(type_check_slow_path->GetEntryLabel());
       break;
     }
 
     case TypeCheckKind::kArrayObjectCheck: {
       // Do an exact check.
-      NearLabel check_non_primitive_component_type;
       if (cls.IsRegister()) {
         __ cmpl(temp, cls.AsRegister<Register>());
       } else {
@@ -6720,28 +6710,13 @@
       // /* HeapReference<Class> */ temp = temp->component_type_
       GenerateReferenceLoadOneRegister(instruction, temp_loc, component_offset, maybe_temp2_loc);
 
-      // If the component type is not null (i.e. the object is indeed
-      // an array), jump to label `check_non_primitive_component_type`
-      // to further check that this component type is not a primitive
-      // type.
+      // If the component type is null (i.e. the object not an array),  jump to the slow path to
+      // throw the exception. Otherwise proceed with the check.
       __ testl(temp, temp);
-      __ j(kNotEqual, &check_non_primitive_component_type);
-      // Otherwise, jump to the slow path to throw the exception.
-      //
-      // But before, move back the object's class into `temp` before
-      // going into the slow path, as it has been overwritten in the
-      // meantime.
-      // /* HeapReference<Class> */ temp = obj->klass_
-      GenerateReferenceLoadTwoRegisters(instruction, temp_loc, obj_loc, class_offset);
-      __ jmp(type_check_slow_path->GetEntryLabel());
+      __ j(kZero, type_check_slow_path->GetEntryLabel());
 
-      __ Bind(&check_non_primitive_component_type);
       __ cmpw(Address(temp, primitive_offset), Immediate(Primitive::kPrimNot));
-      __ j(kEqual, &done);
-      // Same comment as above regarding `temp` and the slow path.
-      // /* HeapReference<Class> */ temp = obj->klass_
-      GenerateReferenceLoadTwoRegisters(instruction, temp_loc, obj_loc, class_offset);
-      __ jmp(type_check_slow_path->GetEntryLabel());
+      __ j(kNotEqual, type_check_slow_path->GetEntryLabel());
       break;
     }
 
diff --git a/compiler/optimizing/code_generator_x86_64.cc b/compiler/optimizing/code_generator_x86_64.cc
index 232c3b3..89b16d3 100644
--- a/compiler/optimizing/code_generator_x86_64.cc
+++ b/compiler/optimizing/code_generator_x86_64.cc
@@ -332,8 +332,14 @@
 
   void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
     LocationSummary* locations = instruction_->GetLocations();
-    Location object_class = instruction_->IsCheckCast() ? locations->GetTemp(0)
-                                                        : locations->Out();
+    Location arg0, arg1;
+    if (instruction_->IsInstanceOf()) {
+      arg0 = locations->InAt(1);
+      arg1 = locations->Out();
+    } else {
+      arg0 = locations->InAt(0);
+      arg1 = locations->InAt(1);
+    }
     uint32_t dex_pc = instruction_->GetDexPc();
     DCHECK(instruction_->IsCheckCast()
            || !locations->GetLiveRegisters()->ContainsCoreRegister(locations->Out().reg()));
@@ -348,22 +354,19 @@
     // We're moving two locations to locations that could overlap, so we need a parallel
     // move resolver.
     InvokeRuntimeCallingConvention calling_convention;
-    codegen->EmitParallelMoves(
-        locations->InAt(1),
-        Location::RegisterLocation(calling_convention.GetRegisterAt(0)),
-        Primitive::kPrimNot,
-        object_class,
-        Location::RegisterLocation(calling_convention.GetRegisterAt(1)),
-        Primitive::kPrimNot);
-
+    codegen->EmitParallelMoves(arg0,
+                               Location::RegisterLocation(calling_convention.GetRegisterAt(0)),
+                               Primitive::kPrimNot,
+                               arg1,
+                               Location::RegisterLocation(calling_convention.GetRegisterAt(1)),
+                               Primitive::kPrimNot);
     if (instruction_->IsInstanceOf()) {
       x86_64_codegen->InvokeRuntime(kQuickInstanceofNonTrivial, instruction_, dex_pc, this);
-      CheckEntrypointTypes<
-          kQuickInstanceofNonTrivial, size_t, const mirror::Class*, const mirror::Class*>();
+      CheckEntrypointTypes<kQuickInstanceofNonTrivial, size_t, mirror::Class*, mirror::Class*>();
     } else {
       DCHECK(instruction_->IsCheckCast());
-      x86_64_codegen->InvokeRuntime(kQuickCheckCast, instruction_, dex_pc, this);
-      CheckEntrypointTypes<kQuickCheckCast, void, const mirror::Class*, const mirror::Class*>();
+      x86_64_codegen->InvokeRuntime(kQuickCheckInstanceOf, instruction_, dex_pc, this);
+      CheckEntrypointTypes<kQuickCheckInstanceOf, void, mirror::Object*, mirror::Class*>();
     }
 
     if (!is_fatal_) {
@@ -6100,30 +6103,16 @@
                                         kEmitCompilerReadBarrier);
       // If the class is abstract, we eagerly fetch the super class of the
       // object to avoid doing a comparison we know will fail.
-      NearLabel loop, compare_classes;
+      NearLabel loop;
       __ Bind(&loop);
       // /* HeapReference<Class> */ temp = temp->super_class_
       GenerateReferenceLoadOneRegister(instruction, temp_loc, super_offset, maybe_temp2_loc);
 
-      // If the class reference currently in `temp` is not null, jump
-      // to the `compare_classes` label to compare it with the checked
-      // class.
+      // If the class reference currently in `temp` is null, jump to the slow path to throw the
+      // exception.
       __ testl(temp, temp);
-      __ j(kNotEqual, &compare_classes);
-      // Otherwise, jump to the slow path to throw the exception.
-      //
-      // But before, move back the object's class into `temp` before
-      // going into the slow path, as it has been overwritten in the
-      // meantime.
-      // /* HeapReference<Class> */ temp = obj->klass_
-      GenerateReferenceLoadTwoRegisters(instruction,
-                                        temp_loc,
-                                        obj_loc,
-                                        class_offset,
-                                        kEmitCompilerReadBarrier);
-      __ jmp(type_check_slow_path->GetEntryLabel());
-
-      __ Bind(&compare_classes);
+      // Otherwise, compare the classes.
+      __ j(kZero, type_check_slow_path->GetEntryLabel());
       if (cls.IsRegister()) {
         __ cmpl(temp, cls.AsRegister<CpuRegister>());
       } else {
@@ -6166,18 +6155,8 @@
       // If the class reference currently in `temp` is not null, jump
       // back at the beginning of the loop.
       __ testl(temp, temp);
-      __ j(kNotEqual, &loop);
+      __ j(kNotZero, &loop);
       // Otherwise, jump to the slow path to throw the exception.
-      //
-      // But before, move back the object's class into `temp` before
-      // going into the slow path, as it has been overwritten in the
-      // meantime.
-      // /* HeapReference<Class> */ temp = obj->klass_
-      GenerateReferenceLoadTwoRegisters(instruction,
-                                        temp_loc,
-                                        obj_loc,
-                                        class_offset,
-                                        kEmitCompilerReadBarrier);
       __ jmp(type_check_slow_path->GetEntryLabel());
       __ Bind(&done);
       break;
@@ -6220,31 +6199,10 @@
       // to further check that this component type is not a primitive
       // type.
       __ testl(temp, temp);
-      __ j(kNotEqual, &check_non_primitive_component_type);
       // Otherwise, jump to the slow path to throw the exception.
-      //
-      // But before, move back the object's class into `temp` before
-      // going into the slow path, as it has been overwritten in the
-      // meantime.
-      // /* HeapReference<Class> */ temp = obj->klass_
-      GenerateReferenceLoadTwoRegisters(instruction,
-                                        temp_loc,
-                                        obj_loc,
-                                        class_offset,
-                                        kEmitCompilerReadBarrier);
-      __ jmp(type_check_slow_path->GetEntryLabel());
-
-      __ Bind(&check_non_primitive_component_type);
+      __ j(kZero, type_check_slow_path->GetEntryLabel());
       __ cmpw(Address(temp, primitive_offset), Immediate(Primitive::kPrimNot));
-      __ j(kEqual, &done);
-      // Same comment as above regarding `temp` and the slow path.
-      // /* HeapReference<Class> */ temp = obj->klass_
-      GenerateReferenceLoadTwoRegisters(instruction,
-                                        temp_loc,
-                                        obj_loc,
-                                        class_offset,
-                                        kEmitCompilerReadBarrier);
-      __ jmp(type_check_slow_path->GetEntryLabel());
+      __ j(kNotEqual, type_check_slow_path->GetEntryLabel());
       __ Bind(&done);
       break;
     }
@@ -6316,13 +6274,6 @@
         __ j(kNotZero, &start_loop);
         __ Bind(&is_null);
       }
-
-      // Since we clobbered temp_loc holding the class, we need to reload it.
-      GenerateReferenceLoadTwoRegisters(instruction,
-                                        temp_loc,
-                                        obj_loc,
-                                        class_offset,
-                                        kEmitCompilerReadBarrier);
       __ jmp(type_check_slow_path->GetEntryLabel());
       __ Bind(&done);
       break;