[Code Stubs] Convert FastNewClosureStub to a TurboFanCodeStub.

src/code-stubs.cc

 // Copyright 2012 the V8 project authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "src/code-stubs.h"
 
 #include <sstream>
 
 #include "src/bootstrapper.h"
 #include "src/code-factory.h"
@@ skipping 4171 matching lines @@
     const {
   return VectorStoreTransitionDescriptor(isolate());
 }
 
 
 CallInterfaceDescriptor
 ElementsTransitionAndStoreStub::GetCallInterfaceDescriptor() const {
   return VectorStoreTransitionDescriptor(isolate());
 }
 
-void FastNewClosureStub::InitializeDescriptor(CodeStubDescriptor* descriptor) {}
-
 void FastNewContextStub::InitializeDescriptor(CodeStubDescriptor* d) {}
 
-
 void TypeofStub::InitializeDescriptor(CodeStubDescriptor* descriptor) {}
 
-
 void NumberToStringStub::InitializeDescriptor(CodeStubDescriptor* descriptor) {
   descriptor->Initialize(
       Runtime::FunctionForId(Runtime::kNumberToString)->entry);
 }
 
 
 void FastCloneRegExpStub::InitializeDescriptor(CodeStubDescriptor* descriptor) {
   FastCloneRegExpDescriptor call_descriptor(isolate());
   descriptor->Initialize(
       Runtime::FunctionForId(Runtime::kCreateRegExpLiteral)->entry);
@@ skipping 208 matching lines @@
   {
     result.Bind(
         assembler->CallRuntime(Runtime::kHasProperty, context, key, object));
     assembler->Goto(&end);
   }
 
   assembler->Bind(&end);
   return result.value();
 }
 
+// static
+compiler::Node* FastNewClosureStub::Generate(CodeStubAssembler* assembler,
+                                             compiler::Node* shared_info,
+                                             compiler::Node* context) {
+  typedef compiler::Node Node;
+  typedef compiler::CodeAssembler::Label Label;
+  typedef compiler::CodeAssembler::Variable Variable;
+
+  Isolate* isolate = assembler->isolate();
+  Factory* factory = assembler->isolate()->factory();
+  assembler->IncrementCounter(isolate->counters()->fast_new_closure_total(), 1);
+
+  // Create a new closure from the given function info in new space
+  Node* result = assembler->Allocate(JSFunction::kSize);
+
+  // Calculate the index of the map we should install on the function based on
+  // the FunctionKind and LanguageMode of the function.
+  // Note: Must be kept in sync with Context::FunctionMapIndex
+  Node* compiler_hints = assembler->LoadObjectField(
+      shared_info, SharedFunctionInfo::kCompilerHintsOffset,
+      MachineType::Uint32());
+  Node* is_strict = assembler->Word32And(
+      compiler_hints,
+      assembler->Int32Constant(1 << SharedFunctionInfo::kStrictModeBit));
+
+  Label if_normal(assembler), if_generator(assembler), if_async(assembler),
+      if_class_constructor(assembler), if_function_without_prototype(assembler),
+      load_map(assembler);
+  Variable map_index(assembler, MachineRepresentation::kTagged);
+
+  Node* is_not_normal = assembler->Word32And(
+      compiler_hints,
+      assembler->Int32Constant(SharedFunctionInfo::FunctionKindBits::kMask));
+  assembler->GotoUnless(is_not_normal, &if_normal);
+
+  Node* is_generator = assembler->Word32And(
+      compiler_hints,
+      assembler->Int32Constant(1 << SharedFunctionInfo::kIsGeneratorBit));
+  assembler->GotoIf(is_generator, &if_generator);
+
+  Node* is_async = assembler->Word32And(
+      compiler_hints,
+      assembler->Int32Constant(1 << SharedFunctionInfo::kIsAsyncFunctionBit));
+  assembler->GotoIf(is_async, &if_async);
+
+  Node* is_class_constructor = assembler->Word32And(
+      compiler_hints,
+      assembler->Int32Constant(SharedFunctionInfo::kClassConstructorBits));
+  assembler->GotoIf(is_class_constructor, &if_class_constructor);
+
+  if (FLAG_debug_code) {
+    // Function must be a function without a prototype.
+    assembler->Assert(assembler->Word32And(
+        compiler_hints, assembler->Int32Constant(
+                            SharedFunctionInfo::kAccessorFunctionBits |
+                            (1 << SharedFunctionInfo::kIsArrowBit) |
+                            (1 << SharedFunctionInfo::kIsConciseMethodBit))));
+  }
+  assembler->Goto(&if_function_without_prototype);
+
+  assembler->Bind(&if_normal);
+  {
+    map_index.Bind(assembler->Select(
+        is_strict, assembler->Int32Constant(Context::STRICT_FUNCTION_MAP_INDEX),
+        assembler->Int32Constant(Context::SLOPPY_FUNCTION_MAP_INDEX)));
+    assembler->Goto(&load_map);
+  }
+
+  assembler->Bind(&if_generator);
+  {
+    map_index.Bind(assembler->Select(
+        is_strict,
+        assembler->Int32Constant(Context::STRICT_GENERATOR_FUNCTION_MAP_INDEX),
+        assembler->Int32Constant(
+            Context::SLOPPY_GENERATOR_FUNCTION_MAP_INDEX)));
+    assembler->Goto(&load_map);
+  }
+
+  assembler->Bind(&if_async);
+  {
+    map_index.Bind(assembler->Select(
+        is_strict,
+        assembler->Int32Constant(Context::STRICT_ASYNC_FUNCTION_MAP_INDEX),
+        assembler->Int32Constant(Context::SLOPPY_ASYNC_FUNCTION_MAP_INDEX)));
+    assembler->Goto(&load_map);
+  }
+
+  assembler->Bind(&if_class_constructor);
+  {
+    map_index.Bind(
+        assembler->Int32Constant(Context::STRICT_FUNCTION_MAP_INDEX));
+    assembler->Goto(&load_map);
+  }
+
+  assembler->Bind(&if_function_without_prototype);
+  {
+    map_index.Bind(assembler->Int32Constant(
+        Context::STRICT_FUNCTION_WITHOUT_PROTOTYPE_MAP_INDEX));
+    assembler->Goto(&load_map);
+  }
+
+  assembler->Bind(&load_map);
+
+  // Get the function map in the current native context and set that
+  // as the map of the allocated object.
+  Node* native_context = assembler->LoadNativeContext(context);
+  Node* map_slot_value =
+      assembler->LoadFixedArrayElement(native_context, map_index.value());
+  assembler->StoreMapNoWriteBarrier(result, map_slot_value);
+
+  // Initialize the rest of the function.
+  Node* empty_fixed_array =
+      assembler->HeapConstant(factory->empty_fixed_array());
+  Node* empty_literals_array =
+      assembler->HeapConstant(factory->empty_literals_array());
+  assembler->StoreObjectFieldNoWriteBarrier(result, JSObject::kPropertiesOffset,
+                                            empty_fixed_array);
+  assembler->StoreObjectFieldNoWriteBarrier(result, JSObject::kElementsOffset,
+                                            empty_fixed_array);
+  assembler->StoreObjectFieldNoWriteBarrier(result, JSFunction::kLiteralsOffset,
+                                            empty_literals_array);
+  assembler->StoreObjectFieldNoWriteBarrier(
+      result, JSFunction::kPrototypeOrInitialMapOffset,
+      assembler->TheHoleConstant());
+  assembler->StoreObjectFieldNoWriteBarrier(
+      result, JSFunction::kSharedFunctionInfoOffset, shared_info);
+  assembler->StoreObjectFieldNoWriteBarrier(result, JSFunction::kContextOffset,
+                                            context);
+
+  // TODO(rmcilroy): Should we set the code entry from the SharedFunctionInfo

[Benedikt Meurer, 2016/06/29 17:42:16]

We want to go to optimized code if it's already available in the optimized code
map. Installing the baseline code would force us to baseline first, and might
even result in unnecessary OSR.

[rmcilroy, 2016/06/29 21:11:48]

Ahh yes, of course. Removed this TODO.

Is there any thought that we could enable generated code to look up the
optimized code map and avoid the roundtrip to the runtime for new functions, or
is this likely not worth the effort?

[Benedikt Meurer, 2016/06/30 03:30:15]

CompileLazy doesn't do a round trip to the runtime if code is already available.

+  // instead? For eager compilation this would seem preferable.
+  Handle<Code> lazy_builtin_handle(
+      assembler->isolate()->builtins()->builtin(Builtins::kCompileLazy));
+  Node* lazy_builtin = assembler->HeapConstant(lazy_builtin_handle);
+  Node* lazy_builtin_entry = assembler->IntPtrAdd(
+      lazy_builtin,
+      assembler->IntPtrConstant(Code::kHeaderSize - kHeapObjectTag));
+  assembler->StoreObjectFieldNoWriteBarrier(
+      result, JSFunction::kCodeEntryOffset, lazy_builtin_entry);
+  assembler->StoreObjectFieldNoWriteBarrier(result,
+                                            JSFunction::kNextFunctionLinkOffset,
+                                            assembler->UndefinedConstant());
+
+  return result;
+}
+
+void FastNewClosureStub::GenerateAssembly(CodeStubAssembler* assembler) const {
+  assembler->Return(
+      Generate(assembler, assembler->Parameter(0), assembler->Parameter(1)));
+}
+
 void CreateAllocationSiteStub::GenerateAheadOfTime(Isolate* isolate) {
   CreateAllocationSiteStub stub(isolate);
   stub.GetCode();
 }
 
 
 void CreateWeakCellStub::GenerateAheadOfTime(Isolate* isolate) {
   CreateWeakCellStub stub(isolate);
   stub.GetCode();
 }
@@ skipping 292 matching lines @@
   if (type->Is(Type::UntaggedPointer())) {
     return Representation::External();
   }
 
   DCHECK(!type->Is(Type::Untagged()));
   return Representation::Tagged();
 }
 
 }  // namespace internal
 }  // namespace v8