Implement invoke virtual in optimizing compiler.

Also refactor 004 tests to make them work with both Quick and
Optimizing.

Change-Id: I87e275cb0ae0258fc3bb32b612140000b1d2adf8
diff --git a/compiler/optimizing/builder.cc b/compiler/optimizing/builder.cc
index a03588f..33b00d2 100644
--- a/compiler/optimizing/builder.cc
+++ b/compiler/optimizing/builder.cc
@@ -331,18 +331,61 @@
                                 bool is_range,
                                 uint32_t* args,
                                 uint32_t register_index) {
+  Instruction::Code opcode = instruction.Opcode();
+  InvokeType invoke_type;
+  switch (opcode) {
+    case Instruction::INVOKE_STATIC:
+    case Instruction::INVOKE_STATIC_RANGE:
+      invoke_type = kStatic;
+      break;
+    case Instruction::INVOKE_DIRECT:
+    case Instruction::INVOKE_DIRECT_RANGE:
+      invoke_type = kDirect;
+      break;
+    case Instruction::INVOKE_VIRTUAL:
+    case Instruction::INVOKE_VIRTUAL_RANGE:
+      invoke_type = kVirtual;
+      break;
+    case Instruction::INVOKE_INTERFACE:
+    case Instruction::INVOKE_INTERFACE_RANGE:
+      invoke_type = kInterface;
+      break;
+    case Instruction::INVOKE_SUPER_RANGE:
+    case Instruction::INVOKE_SUPER:
+      invoke_type = kSuper;
+      break;
+    default:
+      LOG(FATAL) << "Unexpected invoke op: " << opcode;
+      return false;
+  }
+
   const DexFile::MethodId& method_id = dex_file_->GetMethodId(method_idx);
   const DexFile::ProtoId& proto_id = dex_file_->GetProtoId(method_id.proto_idx_);
   const char* descriptor = dex_file_->StringDataByIdx(proto_id.shorty_idx_);
   Primitive::Type return_type = Primitive::GetType(descriptor[0]);
-  bool is_instance_call =
-      instruction.Opcode() != Instruction::INVOKE_STATIC
-      && instruction.Opcode() != Instruction::INVOKE_STATIC_RANGE;
+  bool is_instance_call = invoke_type != kStatic;
   const size_t number_of_arguments = strlen(descriptor) - (is_instance_call ? 0 : 1);
 
-  // Treat invoke-direct like static calls for now.
-  HInvoke* invoke = new (arena_) HInvokeStatic(
-      arena_, number_of_arguments, return_type, dex_offset, method_idx);
+  HInvoke* invoke = nullptr;
+  if (invoke_type == kVirtual) {
+    MethodReference target_method(dex_file_, method_idx);
+    uintptr_t direct_code;
+    uintptr_t direct_method;
+    int vtable_index;
+    // TODO: Add devirtualization support.
+    compiler_driver_->ComputeInvokeInfo(dex_compilation_unit_, dex_offset, true, true,
+                                        &invoke_type, &target_method, &vtable_index,
+                                        &direct_code, &direct_method);
+    if (vtable_index == -1) {
+      return false;
+    }
+    invoke = new (arena_) HInvokeVirtual(
+        arena_, number_of_arguments, return_type, dex_offset, vtable_index);
+  } else {
+    // Treat invoke-direct like static calls for now.
+    invoke = new (arena_) HInvokeStatic(
+        arena_, number_of_arguments, return_type, dex_offset, method_idx);
+  }
 
   size_t start_index = 0;
   Temporaries temps(graph_, is_instance_call ? 1 : 0);
@@ -620,7 +663,8 @@
     }
 
     case Instruction::INVOKE_STATIC:
-    case Instruction::INVOKE_DIRECT: {
+    case Instruction::INVOKE_DIRECT:
+    case Instruction::INVOKE_VIRTUAL: {
       uint32_t method_idx = instruction.VRegB_35c();
       uint32_t number_of_vreg_arguments = instruction.VRegA_35c();
       uint32_t args[5];
@@ -632,7 +676,8 @@
     }
 
     case Instruction::INVOKE_STATIC_RANGE:
-    case Instruction::INVOKE_DIRECT_RANGE: {
+    case Instruction::INVOKE_DIRECT_RANGE:
+    case Instruction::INVOKE_VIRTUAL_RANGE: {
       uint32_t method_idx = instruction.VRegB_3rc();
       uint32_t number_of_vreg_arguments = instruction.VRegA_3rc();
       uint32_t register_index = instruction.VRegC();
diff --git a/compiler/optimizing/code_generator_arm.cc b/compiler/optimizing/code_generator_arm.cc
index 9903092..ad62279 100644
--- a/compiler/optimizing/code_generator_arm.cc
+++ b/compiler/optimizing/code_generator_arm.cc
@@ -20,6 +20,7 @@
 #include "gc/accounting/card_table.h"
 #include "mirror/array.h"
 #include "mirror/art_method.h"
+#include "mirror/class.h"
 #include "thread.h"
 #include "utils/assembler.h"
 #include "utils/arm/assembler_arm.h"
@@ -818,6 +819,47 @@
 }
 
 void LocationsBuilderARM::VisitInvokeStatic(HInvokeStatic* invoke) {
+  HandleInvoke(invoke);
+}
+
+void InstructionCodeGeneratorARM::LoadCurrentMethod(Register reg) {
+  __ ldr(reg, Address(SP, kCurrentMethodStackOffset));
+}
+
+void InstructionCodeGeneratorARM::VisitInvokeStatic(HInvokeStatic* invoke) {
+  Register temp = invoke->GetLocations()->GetTemp(0).AsArm().AsCoreRegister();
+  uint32_t heap_reference_size = sizeof(mirror::HeapReference<mirror::Object>);
+  size_t index_in_cache = mirror::Array::DataOffset(heap_reference_size).Int32Value() +
+      invoke->GetIndexInDexCache() * kArmWordSize;
+
+  // TODO: Implement all kinds of calls:
+  // 1) boot -> boot
+  // 2) app -> boot
+  // 3) app -> app
+  //
+  // Currently we implement the app -> app logic, which looks up in the resolve cache.
+
+  // temp = method;
+  LoadCurrentMethod(temp);
+  // temp = temp->dex_cache_resolved_methods_;
+  __ ldr(temp, Address(temp, mirror::ArtMethod::DexCacheResolvedMethodsOffset().Int32Value()));
+  // temp = temp[index_in_cache]
+  __ ldr(temp, Address(temp, index_in_cache));
+  // LR = temp[offset_of_quick_compiled_code]
+  __ ldr(LR, Address(temp,
+                     mirror::ArtMethod::EntryPointFromQuickCompiledCodeOffset().Int32Value()));
+  // LR()
+  __ blx(LR);
+
+  codegen_->RecordPcInfo(invoke, invoke->GetDexPc());
+  DCHECK(!codegen_->IsLeafMethod());
+}
+
+void LocationsBuilderARM::VisitInvokeVirtual(HInvokeVirtual* invoke) {
+  HandleInvoke(invoke);
+}
+
+void LocationsBuilderARM::HandleInvoke(HInvoke* invoke) {
   LocationSummary* locations =
       new (GetGraph()->GetArena()) LocationSummary(invoke, LocationSummary::kCall);
   locations->AddTemp(ArmCoreLocation(R0));
@@ -852,37 +894,30 @@
   }
 }
 
-void InstructionCodeGeneratorARM::LoadCurrentMethod(Register reg) {
-  __ ldr(reg, Address(SP, kCurrentMethodStackOffset));
-}
 
-void InstructionCodeGeneratorARM::VisitInvokeStatic(HInvokeStatic* invoke) {
+void InstructionCodeGeneratorARM::VisitInvokeVirtual(HInvokeVirtual* invoke) {
   Register temp = invoke->GetLocations()->GetTemp(0).AsArm().AsCoreRegister();
-  uint32_t heap_reference_size = sizeof(mirror::HeapReference<mirror::Object>);
-  size_t index_in_cache = mirror::Array::DataOffset(heap_reference_size).Int32Value() +
-      invoke->GetIndexInDexCache() * kArmWordSize;
-
-  // TODO: Implement all kinds of calls:
-  // 1) boot -> boot
-  // 2) app -> boot
-  // 3) app -> app
-  //
-  // Currently we implement the app -> app logic, which looks up in the resolve cache.
-
-  // temp = method;
-  LoadCurrentMethod(temp);
-  // temp = temp->dex_cache_resolved_methods_;
-  __ ldr(temp, Address(temp, mirror::ArtMethod::DexCacheResolvedMethodsOffset().Int32Value()));
-  // temp = temp[index_in_cache]
-  __ ldr(temp, Address(temp, index_in_cache));
-  // LR = temp[offset_of_quick_compiled_code]
-  __ ldr(LR, Address(temp,
-                     mirror::ArtMethod::EntryPointFromQuickCompiledCodeOffset().Int32Value()));
-  // LR()
+  uint32_t method_offset = mirror::Class::EmbeddedVTableOffset().Uint32Value() +
+          invoke->GetVTableIndex() * sizeof(mirror::Class::VTableEntry);
+  LocationSummary* locations = invoke->GetLocations();
+  Location receiver = locations->InAt(0);
+  uint32_t class_offset = mirror::Object::ClassOffset().Int32Value();
+  // temp = object->GetClass();
+  if (receiver.IsStackSlot()) {
+    __ ldr(temp, Address(SP, receiver.GetStackIndex()));
+    __ ldr(temp, Address(temp, class_offset));
+  } else {
+    __ ldr(temp, Address(receiver.AsArm().AsCoreRegister(), class_offset));
+  }
+  // temp = temp->GetMethodAt(method_offset);
+  uint32_t entry_point = mirror::ArtMethod::EntryPointFromQuickCompiledCodeOffset().Int32Value();
+  __ ldr(temp, Address(temp, method_offset));
+  // LR = temp->GetEntryPoint();
+  __ ldr(LR, Address(temp, entry_point));
+  // LR();
   __ blx(LR);
-
-  codegen_->RecordPcInfo(invoke, invoke->GetDexPc());
   DCHECK(!codegen_->IsLeafMethod());
+  codegen_->RecordPcInfo(invoke, invoke->GetDexPc());
 }
 
 void LocationsBuilderARM::VisitAdd(HAdd* add) {
diff --git a/compiler/optimizing/code_generator_arm.h b/compiler/optimizing/code_generator_arm.h
index 660294b..2480960 100644
--- a/compiler/optimizing/code_generator_arm.h
+++ b/compiler/optimizing/code_generator_arm.h
@@ -93,6 +93,8 @@
 
 #undef DECLARE_VISIT_INSTRUCTION
 
+  void HandleInvoke(HInvoke* invoke);
+
  private:
   CodeGeneratorARM* const codegen_;
   InvokeDexCallingConventionVisitor parameter_visitor_;
diff --git a/compiler/optimizing/code_generator_x86.cc b/compiler/optimizing/code_generator_x86.cc
index 3dd9b37..3383cb2 100644
--- a/compiler/optimizing/code_generator_x86.cc
+++ b/compiler/optimizing/code_generator_x86.cc
@@ -20,6 +20,7 @@
 #include "gc/accounting/card_table.h"
 #include "mirror/array.h"
 #include "mirror/art_method.h"
+#include "mirror/class.h"
 #include "thread.h"
 #include "utils/assembler.h"
 #include "utils/stack_checks.h"
@@ -763,6 +764,40 @@
 }
 
 void LocationsBuilderX86::VisitInvokeStatic(HInvokeStatic* invoke) {
+  HandleInvoke(invoke);
+}
+
+void InstructionCodeGeneratorX86::VisitInvokeStatic(HInvokeStatic* invoke) {
+  Register temp = invoke->GetLocations()->GetTemp(0).AsX86().AsCpuRegister();
+  uint32_t heap_reference_size = sizeof(mirror::HeapReference<mirror::Object>);
+  size_t index_in_cache = mirror::Array::DataOffset(heap_reference_size).Int32Value() +
+      invoke->GetIndexInDexCache() * kX86WordSize;
+
+  // TODO: Implement all kinds of calls:
+  // 1) boot -> boot
+  // 2) app -> boot
+  // 3) app -> app
+  //
+  // Currently we implement the app -> app logic, which looks up in the resolve cache.
+
+  // temp = method;
+  LoadCurrentMethod(temp);
+  // temp = temp->dex_cache_resolved_methods_;
+  __ movl(temp, Address(temp, mirror::ArtMethod::DexCacheResolvedMethodsOffset().Int32Value()));
+  // temp = temp[index_in_cache]
+  __ movl(temp, Address(temp, index_in_cache));
+  // (temp + offset_of_quick_compiled_code)()
+  __ call(Address(temp, mirror::ArtMethod::EntryPointFromQuickCompiledCodeOffset().Int32Value()));
+
+  DCHECK(!codegen_->IsLeafMethod());
+  codegen_->RecordPcInfo(invoke, invoke->GetDexPc());
+}
+
+void LocationsBuilderX86::VisitInvokeVirtual(HInvokeVirtual* invoke) {
+  HandleInvoke(invoke);
+}
+
+void LocationsBuilderX86::HandleInvoke(HInvoke* invoke) {
   LocationSummary* locations =
       new (GetGraph()->GetArena()) LocationSummary(invoke, LocationSummary::kCall);
   locations->AddTemp(X86CpuLocation(EAX));
@@ -799,26 +834,23 @@
   invoke->SetLocations(locations);
 }
 
-void InstructionCodeGeneratorX86::VisitInvokeStatic(HInvokeStatic* invoke) {
+void InstructionCodeGeneratorX86::VisitInvokeVirtual(HInvokeVirtual* invoke) {
   Register temp = invoke->GetLocations()->GetTemp(0).AsX86().AsCpuRegister();
-  uint32_t heap_reference_size = sizeof(mirror::HeapReference<mirror::Object>);
-  size_t index_in_cache = mirror::Array::DataOffset(heap_reference_size).Int32Value() +
-      invoke->GetIndexInDexCache() * kX86WordSize;
-
-  // TODO: Implement all kinds of calls:
-  // 1) boot -> boot
-  // 2) app -> boot
-  // 3) app -> app
-  //
-  // Currently we implement the app -> app logic, which looks up in the resolve cache.
-
-  // temp = method;
-  LoadCurrentMethod(temp);
-  // temp = temp->dex_cache_resolved_methods_;
-  __ movl(temp, Address(temp, mirror::ArtMethod::DexCacheResolvedMethodsOffset().Int32Value()));
-  // temp = temp[index_in_cache]
-  __ movl(temp, Address(temp, index_in_cache));
-  // (temp + offset_of_quick_compiled_code)()
+  uint32_t method_offset = mirror::Class::EmbeddedVTableOffset().Uint32Value() +
+          invoke->GetVTableIndex() * sizeof(mirror::Class::VTableEntry);
+  LocationSummary* locations = invoke->GetLocations();
+  Location receiver = locations->InAt(0);
+  uint32_t class_offset = mirror::Object::ClassOffset().Int32Value();
+  // temp = object->GetClass();
+  if (receiver.IsStackSlot()) {
+    __ movl(temp, Address(ESP, receiver.GetStackIndex()));
+    __ movl(temp, Address(temp, class_offset));
+  } else {
+    __ movl(temp, Address(receiver.AsX86().AsCpuRegister(), class_offset));
+  }
+  // temp = temp->GetMethodAt(method_offset);
+  __ movl(temp, Address(temp, method_offset));
+  // call temp->GetEntryPoint();
   __ call(Address(temp, mirror::ArtMethod::EntryPointFromQuickCompiledCodeOffset().Int32Value()));
 
   DCHECK(!codegen_->IsLeafMethod());
diff --git a/compiler/optimizing/code_generator_x86.h b/compiler/optimizing/code_generator_x86.h
index 7c50204..f1be0ad 100644
--- a/compiler/optimizing/code_generator_x86.h
+++ b/compiler/optimizing/code_generator_x86.h
@@ -94,6 +94,8 @@
 
 #undef DECLARE_VISIT_INSTRUCTION
 
+  void HandleInvoke(HInvoke* invoke);
+
  private:
   CodeGeneratorX86* const codegen_;
   InvokeDexCallingConventionVisitor parameter_visitor_;
diff --git a/compiler/optimizing/code_generator_x86_64.cc b/compiler/optimizing/code_generator_x86_64.cc
index 2f352e0..ca03af8 100644
--- a/compiler/optimizing/code_generator_x86_64.cc
+++ b/compiler/optimizing/code_generator_x86_64.cc
@@ -20,6 +20,7 @@
 #include "gc/accounting/card_table.h"
 #include "mirror/array.h"
 #include "mirror/art_method.h"
+#include "mirror/class.h"
 #include "mirror/object_reference.h"
 #include "thread.h"
 #include "utils/assembler.h"
@@ -709,12 +710,46 @@
 }
 
 void LocationsBuilderX86_64::VisitInvokeStatic(HInvokeStatic* invoke) {
+  HandleInvoke(invoke);
+}
+
+void InstructionCodeGeneratorX86_64::VisitInvokeStatic(HInvokeStatic* invoke) {
+  CpuRegister temp = invoke->GetLocations()->GetTemp(0).AsX86_64().AsCpuRegister();
+  uint32_t heap_reference_size = sizeof(mirror::HeapReference<mirror::Object>);
+  size_t index_in_cache = mirror::Array::DataOffset(heap_reference_size).SizeValue() +
+      invoke->GetIndexInDexCache() * heap_reference_size;
+
+  // TODO: Implement all kinds of calls:
+  // 1) boot -> boot
+  // 2) app -> boot
+  // 3) app -> app
+  //
+  // Currently we implement the app -> app logic, which looks up in the resolve cache.
+
+  // temp = method;
+  LoadCurrentMethod(temp);
+  // temp = temp->dex_cache_resolved_methods_;
+  __ movl(temp, Address(temp, mirror::ArtMethod::DexCacheResolvedMethodsOffset().SizeValue()));
+  // temp = temp[index_in_cache]
+  __ movl(temp, Address(temp, index_in_cache));
+  // (temp + offset_of_quick_compiled_code)()
+  __ call(Address(temp, mirror::ArtMethod::EntryPointFromQuickCompiledCodeOffset().SizeValue()));
+
+  DCHECK(!codegen_->IsLeafMethod());
+  codegen_->RecordPcInfo(invoke, invoke->GetDexPc());
+}
+
+void LocationsBuilderX86_64::VisitInvokeVirtual(HInvokeVirtual* invoke) {
+  HandleInvoke(invoke);
+}
+
+void LocationsBuilderX86_64::HandleInvoke(HInvoke* invoke) {
   LocationSummary* locations =
       new (GetGraph()->GetArena()) LocationSummary(invoke, LocationSummary::kCall);
   locations->AddTemp(X86_64CpuLocation(RDI));
 
   InvokeDexCallingConventionVisitor calling_convention_visitor;
-  for (size_t i = 0; i < invoke->InputCount(); ++i) {
+  for (size_t i = 0; i < invoke->InputCount(); i++) {
     HInstruction* input = invoke->InputAt(i);
     locations->SetInAt(i, calling_convention_visitor.GetNextLocation(input->GetType()));
   }
@@ -740,26 +775,23 @@
   }
 }
 
-void InstructionCodeGeneratorX86_64::VisitInvokeStatic(HInvokeStatic* invoke) {
+void InstructionCodeGeneratorX86_64::VisitInvokeVirtual(HInvokeVirtual* invoke) {
   CpuRegister temp = invoke->GetLocations()->GetTemp(0).AsX86_64().AsCpuRegister();
-  uint32_t heap_reference_size = sizeof(mirror::HeapReference<mirror::Object>);
-  size_t index_in_cache = mirror::Array::DataOffset(heap_reference_size).SizeValue() +
-      invoke->GetIndexInDexCache() * heap_reference_size;
-
-  // TODO: Implement all kinds of calls:
-  // 1) boot -> boot
-  // 2) app -> boot
-  // 3) app -> app
-  //
-  // Currently we implement the app -> app logic, which looks up in the resolve cache.
-
-  // temp = method;
-  LoadCurrentMethod(temp);
-  // temp = temp->dex_cache_resolved_methods_;
-  __ movl(temp, Address(temp, mirror::ArtMethod::DexCacheResolvedMethodsOffset().SizeValue()));
-  // temp = temp[index_in_cache]
-  __ movl(temp, Address(temp, index_in_cache));
-  // (temp + offset_of_quick_compiled_code)()
+  size_t method_offset = mirror::Class::EmbeddedVTableOffset().SizeValue() +
+          invoke->GetVTableIndex() * sizeof(mirror::Class::VTableEntry);
+  LocationSummary* locations = invoke->GetLocations();
+  Location receiver = locations->InAt(0);
+  size_t class_offset = mirror::Object::ClassOffset().SizeValue();
+  // temp = object->GetClass();
+  if (receiver.IsStackSlot()) {
+    __ movq(temp, Address(CpuRegister(RSP), receiver.GetStackIndex()));
+    __ movq(temp, Address(temp, class_offset));
+  } else {
+    __ movq(temp, Address(receiver.AsX86_64().AsCpuRegister(), class_offset));
+  }
+  // temp = temp->GetMethodAt(method_offset);
+  __ movl(temp, Address(temp, method_offset));
+  // call temp->GetEntryPoint();
   __ call(Address(temp, mirror::ArtMethod::EntryPointFromQuickCompiledCodeOffset().SizeValue()));
 
   DCHECK(!codegen_->IsLeafMethod());
diff --git a/compiler/optimizing/code_generator_x86_64.h b/compiler/optimizing/code_generator_x86_64.h
index 44552ea..78b60fe 100644
--- a/compiler/optimizing/code_generator_x86_64.h
+++ b/compiler/optimizing/code_generator_x86_64.h
@@ -91,6 +91,8 @@
 
 #undef DECLARE_VISIT_INSTRUCTION
 
+  void HandleInvoke(HInvoke* invoke);
+
  private:
   CodeGeneratorX86_64* const codegen_;
   InvokeDexCallingConventionVisitor parameter_visitor_;
diff --git a/compiler/optimizing/nodes.h b/compiler/optimizing/nodes.h
index ed6dd93..d6dfeae 100644
--- a/compiler/optimizing/nodes.h
+++ b/compiler/optimizing/nodes.h
@@ -422,6 +422,7 @@
   M(If)                                                    \
   M(IntConstant)                                           \
   M(InvokeStatic)                                          \
+  M(InvokeVirtual)                                         \
   M(LoadLocal)                                             \
   M(Local)                                                 \
   M(LongConstant)                                          \
@@ -1272,6 +1273,26 @@
   DISALLOW_COPY_AND_ASSIGN(HInvokeStatic);
 };
 
+class HInvokeVirtual : public HInvoke {
+ public:
+  HInvokeVirtual(ArenaAllocator* arena,
+                 uint32_t number_of_arguments,
+                 Primitive::Type return_type,
+                 uint32_t dex_pc,
+                 uint32_t vtable_index)
+      : HInvoke(arena, number_of_arguments, return_type, dex_pc),
+        vtable_index_(vtable_index) {}
+
+  uint32_t GetVTableIndex() const { return vtable_index_; }
+
+  DECLARE_INSTRUCTION(InvokeVirtual);
+
+ private:
+  const uint32_t vtable_index_;
+
+  DISALLOW_COPY_AND_ASSIGN(HInvokeVirtual);
+};
+
 class HNewInstance : public HExpression<0> {
  public:
   HNewInstance(uint32_t dex_pc, uint16_t type_index)