[builtins] Migrate String.prototype.charCodeAt and String.prototype.charAt to TurboFan.

Make these builtins fast by default w/o relying on the
%_StringCharCodeAt and %_StringCharAt intrinsics, which we cannot deal
with well in TurboFan (and ignition).

R=epertoso@chromium.org

Review URL: https://codereview.chromium.org/1868963002

Cr-Commit-Position: refs/heads/master@{#35350}

src/bootstrapper.cc

@@ -1255,6 +1255,22 @@ void Genesis::InitializeGlobal(Handle<JSGlobalObject> global_object,
                                    attribs);
       string_map->AppendDescriptor(&d);
     }
+
+    // Create the %StringPrototype%
+    Handle<JSValue> prototype =
+        Handle<JSValue>::cast(factory->NewJSObject(string_fun, TENURED));
+    prototype->set_value(isolate->heap()->empty_string());
+    Accessors::FunctionSetPrototype(string_fun, prototype).Assert();
+
+    // Install the "constructor" property on the {prototype}.
+    JSObject::AddProperty(prototype, factory->constructor_string(), string_fun,
+                          DONT_ENUM);
+
+    // Install the String.prototype methods.
+    SimpleInstallFunction(prototype, "charAt", Builtins::kStringPrototypeCharAt,
+                          1, true);
+    SimpleInstallFunction(prototype, "charCodeAt",
+                          Builtins::kStringPrototypeCharCodeAt, 1, true);
   }
 
   {

src/builtins.cc

@@ -4187,6 +4187,199 @@ BUILTIN(ObjectProtoToString) {
   return *result;
 }
 
+// -----------------------------------------------------------------------------
+// ES6 section 21.1 String Objects
+
+// ES6 section 21.1.3.1 String.prototype.charAt ( pos )
+void Builtins::Generate_StringPrototypeCharAt(
+    compiler::CodeStubAssembler* assembler) {
+  typedef compiler::CodeStubAssembler::Label Label;
+  typedef compiler::Node Node;
+  typedef compiler::CodeStubAssembler::Variable Variable;
+
+  Node* receiver = assembler->Parameter(0);
+  Node* position = assembler->Parameter(1);
+  Node* context = assembler->Parameter(4);
+
+  // Check that {receiver} is coercible to Object and convert it to a String.
+  receiver =
+      assembler->ToThisString(context, receiver, "String.prototype.charAt");
+
+  // Convert the {position} to a Smi and check that it's in bounds of the
+  // {receiver}.
+  // TODO(bmeurer): Find an abstraction for this!
+  {
+    // Check if the {position} is already a Smi.
+    Variable var_position(assembler, MachineRepresentation::kTagged);
+    var_position.Bind(position);
+    Label if_positionissmi(assembler),
+        if_positionisnotsmi(assembler, Label::kDeferred);
+    assembler->Branch(assembler->WordIsSmi(position), &if_positionissmi,
+                      &if_positionisnotsmi);
+    assembler->Bind(&if_positionisnotsmi);
+    {
+      // Convert the {position} to an Integer via the ToIntegerStub.
+      Callable callable = CodeFactory::ToInteger(assembler->isolate());
+      Node* index = assembler->CallStub(callable, context, position);
+
+      // Check if the resulting {index} is now a Smi.
+      Label if_indexissmi(assembler, Label::kDeferred),
+          if_indexisnotsmi(assembler, Label::kDeferred);
+      assembler->Branch(assembler->WordIsSmi(index), &if_indexissmi,
+                        &if_indexisnotsmi);
+
+      assembler->Bind(&if_indexissmi);
+      {
+        var_position.Bind(index);
+        assembler->Goto(&if_positionissmi);
+      }
+
+      assembler->Bind(&if_indexisnotsmi);
+      {
+        // The ToIntegerStub canonicalizes everything in Smi range to Smi
+        // representation, so any HeapNumber returned is not in Smi range.
+        // The only exception here is -0.0, which we treat as 0.
+        Node* index_value = assembler->LoadHeapNumberValue(index);
+        Label if_indexiszero(assembler, Label::kDeferred),
+            if_indexisnotzero(assembler, Label::kDeferred);
+        assembler->Branch(assembler->Float64Equal(
+                              index_value, assembler->Float64Constant(0.0)),
+                          &if_indexiszero, &if_indexisnotzero);
+
+        assembler->Bind(&if_indexiszero);
+        {
+          var_position.Bind(assembler->SmiConstant(Smi::FromInt(0)));
+          assembler->Goto(&if_positionissmi);
+        }
+
+        assembler->Bind(&if_indexisnotzero);
+        {
+          // The {index} is some other integral Number, that is definitely
+          // neither -0.0 nor in Smi range.
+          assembler->Return(assembler->EmptyStringConstant());
+        }
+      }
+    }
+    assembler->Bind(&if_positionissmi);
+    position = var_position.value();
+
+    // Determine the actual length of the {receiver} String.
+    Node* receiver_length =
+        assembler->LoadObjectField(receiver, String::kLengthOffset);
+
+    // Return "" if the Smi {position} is outside the bounds of the {receiver}.
+    Label if_positioninbounds(assembler),
+        if_positionnotinbounds(assembler, Label::kDeferred);
+    assembler->Branch(assembler->SmiAboveOrEqual(position, receiver_length),
+                      &if_positionnotinbounds, &if_positioninbounds);
+    assembler->Bind(&if_positionnotinbounds);
+    assembler->Return(assembler->EmptyStringConstant());
+    assembler->Bind(&if_positioninbounds);
+  }
+
+  // Load the character code at the {position} from the {receiver}.
+  Node* code = assembler->StringCharCodeAt(receiver, position);
+
+  // And return the single character string with only that {code}.
+  Node* result = assembler->StringFromCharCode(code);
+  assembler->Return(result);
+}
+
+// ES6 section 21.1.3.2 String.prototype.charCodeAt ( pos )
+void Builtins::Generate_StringPrototypeCharCodeAt(
+    compiler::CodeStubAssembler* assembler) {
+  typedef compiler::CodeStubAssembler::Label Label;
+  typedef compiler::Node Node;
+  typedef compiler::CodeStubAssembler::Variable Variable;
+
+  Node* receiver = assembler->Parameter(0);
+  Node* position = assembler->Parameter(1);
+  Node* context = assembler->Parameter(4);
+
+  // Check that {receiver} is coercible to Object and convert it to a String.
+  receiver =
+      assembler->ToThisString(context, receiver, "String.prototype.charCodeAt");
+
+  // Convert the {position} to a Smi and check that it's in bounds of the
+  // {receiver}.
+  // TODO(bmeurer): Find an abstraction for this!
+  {
+    // Check if the {position} is already a Smi.
+    Variable var_position(assembler, MachineRepresentation::kTagged);
+    var_position.Bind(position);
+    Label if_positionissmi(assembler),
+        if_positionisnotsmi(assembler, Label::kDeferred);
+    assembler->Branch(assembler->WordIsSmi(position), &if_positionissmi,
+                      &if_positionisnotsmi);
+    assembler->Bind(&if_positionisnotsmi);
+    {
+      // Convert the {position} to an Integer via the ToIntegerStub.
+      Callable callable = CodeFactory::ToInteger(assembler->isolate());
+      Node* index = assembler->CallStub(callable, context, position);
+
+      // Check if the resulting {index} is now a Smi.
+      Label if_indexissmi(assembler, Label::kDeferred),
+          if_indexisnotsmi(assembler, Label::kDeferred);
+      assembler->Branch(assembler->WordIsSmi(index), &if_indexissmi,
+                        &if_indexisnotsmi);
+
+      assembler->Bind(&if_indexissmi);
+      {
+        var_position.Bind(index);
+        assembler->Goto(&if_positionissmi);
+      }
+
+      assembler->Bind(&if_indexisnotsmi);
+      {
+        // The ToIntegerStub canonicalizes everything in Smi range to Smi
+        // representation, so any HeapNumber returned is not in Smi range.
+        // The only exception here is -0.0, which we treat as 0.
+        Node* index_value = assembler->LoadHeapNumberValue(index);
+        Label if_indexiszero(assembler, Label::kDeferred),
+            if_indexisnotzero(assembler, Label::kDeferred);
+        assembler->Branch(assembler->Float64Equal(
+                              index_value, assembler->Float64Constant(0.0)),
+                          &if_indexiszero, &if_indexisnotzero);
+
+        assembler->Bind(&if_indexiszero);
+        {
+          var_position.Bind(assembler->SmiConstant(Smi::FromInt(0)));
+          assembler->Goto(&if_positionissmi);
+        }
+
+        assembler->Bind(&if_indexisnotzero);
+        {
+          // The {index} is some other integral Number, that is definitely
+          // neither -0.0 nor in Smi range.
+          assembler->Return(assembler->NaNConstant());
+        }
+      }
+    }
+    assembler->Bind(&if_positionissmi);
+    position = var_position.value();
+
+    // Determine the actual length of the {receiver} String.
+    Node* receiver_length =
+        assembler->LoadObjectField(receiver, String::kLengthOffset);
+
+    // Return NaN if the Smi {position} is outside the bounds of the {receiver}.
+    Label if_positioninbounds(assembler),
+        if_positionnotinbounds(assembler, Label::kDeferred);
+    assembler->Branch(assembler->SmiAboveOrEqual(position, receiver_length),
+                      &if_positionnotinbounds, &if_positioninbounds);
+    assembler->Bind(&if_positionnotinbounds);
+    assembler->Return(assembler->NaNConstant());
+    assembler->Bind(&if_positioninbounds);
+  }
+
+  // Load the character at the {position} from the {receiver}.
+  Node* value = assembler->StringCharCodeAt(receiver, position);
+  Node* result = assembler->SmiFromWord32(value);
+  assembler->Return(result);
+}
+
+// -----------------------------------------------------------------------------
+// ES6 section 21.1 ArrayBuffer Objects
 
 // ES6 section 24.1.2.1 ArrayBuffer ( length ) for the [[Call]] case.
 BUILTIN(ArrayBufferConstructor) {

src/builtins.h

@@ -310,7 +310,9 @@ inline bool operator&(BuiltinExtraArguments lhs, BuiltinExtraArguments rhs) {
   V(MathRound, 2)                \
   V(MathSqrt, 2)                 \
   V(MathTrunc, 2)                \
-  V(ObjectHasOwnProperty, 2)
+  V(ObjectHasOwnProperty, 2)     \
+  V(StringPrototypeCharAt, 2)    \
+  V(StringPrototypeCharCodeAt, 2)
 
 // Define list of builtin handlers implemented in assembly.
 #define BUILTIN_LIST_H(V)                    \
@@ -626,6 +628,13 @@ class Builtins {
   static void Generate_ObjectHasOwnProperty(
       compiler::CodeStubAssembler* assembler);
 
+  // ES6 section 21.1.3.1 String.prototype.charAt ( pos )
+  static void Generate_StringPrototypeCharAt(
+      compiler::CodeStubAssembler* assembler);
+  // ES6 section 21.1.3.2 String.prototype.charCodeAt ( pos )
+  static void Generate_StringPrototypeCharCodeAt(
+      compiler::CodeStubAssembler* assembler);
+
   static void Generate_StringConstructor(MacroAssembler* masm);
   static void Generate_StringConstructor_ConstructStub(MacroAssembler* masm);
   static void Generate_OnStackReplacement(MacroAssembler* masm);

src/compiler/code-stub-assembler.h

@@ -162,7 +162,10 @@ class CodeStubAssembler {
   Node* ExternalConstant(ExternalReference address);
   Node* Float64Constant(double value);
   Node* BooleanMapConstant();
+  Node* EmptyStringConstant();
   Node* HeapNumberMapConstant();
+  Node* NaNConstant();
+  Node* NoContextConstant();
   Node* NullConstant();
   Node* UndefinedConstant();
 
@@ -278,9 +281,11 @@ class CodeStubAssembler {
   Node* SmiTag(Node* value);
   // Untag a Smi value as a Word.
   Node* SmiUntag(Node* value);
+  Node* SmiToWord(Node* value) { return SmiUntag(value); }
 
   // Smi conversions.
   Node* SmiToFloat64(Node* value);
+  Node* SmiFromWord32(Node* value);
   Node* SmiToWord32(Node* value);
 
   // Smi operations.
@@ -289,6 +294,7 @@ class CodeStubAssembler {
   Node* SmiSub(Node* a, Node* b);
   Node* SmiSubWithOverflow(Node* a, Node* b);
   Node* SmiEqual(Node* a, Node* b);
+  Node* SmiAboveOrEqual(Node* a, Node* b);
   Node* SmiLessThan(Node* a, Node* b);
   Node* SmiLessThanOrEqual(Node* a, Node* b);
   Node* SmiMin(Node* a, Node* b);
@@ -343,13 +349,19 @@ class CodeStubAssembler {
 
   // Allocate an object of the given size.
   Node* Allocate(int size, AllocationFlags flags = kNone);
+  Node* InnerAllocate(Node* previous, int offset);
   // Allocate a HeapNumber without initializing its value.
   Node* AllocateHeapNumber();
   // Allocate a HeapNumber with a specific value.
   Node* AllocateHeapNumberWithValue(Node* value);
-  Node* InnerAllocate(Node* previous, int offset);
+  // Allocate a SeqOneByteString with the given length.
+  Node* AllocateSeqOneByteString(int length);
+  // Allocate a SeqTwoByteString with the given length.
+  Node* AllocateSeqTwoByteString(int length);
 
   // Store an array element to a FixedArray.
+  Node* StoreFixedArrayElementInt32Index(Node* object, Node* index,
+                                         Node* value);
   Node* StoreFixedArrayElementNoWriteBarrier(Node* object, Node* index,
                                              Node* value);
   // Load the Map of an HeapObject.
@@ -378,6 +390,17 @@ class CodeStubAssembler {
   Node* TruncateTaggedToFloat64(Node* context, Node* value);
   Node* TruncateTaggedToWord32(Node* context, Node* value);
 
+  // Type conversions.
+  // Throws a TypeError for {method_name} if {value} is not coercible to Object,
+  // or returns the {value} converted to a String otherwise.
+  Node* ToThisString(Node* context, Node* value, char const* method_name);
+
+  // String helpers.
+  // Load a character from a String (might flatten a ConsString).
+  Node* StringCharCodeAt(Node* string, Node* smi_index);
+  // Return the single character string with only {code}.
+  Node* StringFromCharCode(Node* code);
+
   // Branching helpers.
   // TODO(danno): Can we be more cleverish wrt. edge-split?
   void BranchIf(Node* condition, Label* if_true, Label* if_false);

src/js/messages.js

@@ -37,7 +37,6 @@ var ObjectHasOwnProperty;
 var ObjectToString = utils.ImportNow("object_to_string");
 var Script = utils.ImportNow("Script");
 var stackTraceSymbol = utils.ImportNow("stack_trace_symbol");
-var StringCharAt;
 var StringIndexOf;
 var StringSubstring;
 var SymbolToString;
@@ -57,7 +56,6 @@ utils.Import(function(from) {
   Int8x16ToString = from.Int8x16ToString;
   ObjectDefineProperty = from.ObjectDefineProperty;
   ObjectHasOwnProperty = from.ObjectHasOwnProperty;
-  StringCharAt = from.StringCharAt;
   StringIndexOf = from.StringIndexOf;
   StringSubstring = from.StringSubstring;
   SymbolToString = from.SymbolToString;
@@ -255,6 +253,7 @@ function ScriptLineFromPosition(position) {
   return -1;
 }
 
+
 /**
  * Get information on a specific source position.
  * @param {number} position The source position
@@ -272,7 +271,7 @@ function ScriptLocationFromPosition(position,
   var line_ends = this.line_ends;
   var start = line == 0 ? 0 : line_ends[line - 1] + 1;
   var end = line_ends[line];
-  if (end > 0 && %_Call(StringCharAt, this.source, end - 1) == '\r') {
+  if (end > 0 && %_StringCharAt(this.source, end - 1) === '\r') {
     end--;
   }
   var column = position - start;

src/js/string.js

@@ -57,30 +57,6 @@ function StringValueOf() {
 }
 
 
-// ECMA-262, section 15.5.4.4
-function StringCharAtJS(pos) {
-  CHECK_OBJECT_COERCIBLE(this, "String.prototype.charAt");
-
-  var result = %_StringCharAt(this, pos);
-  if (%_IsSmi(result)) {
-    result = %_StringCharAt(TO_STRING(this), TO_INTEGER(pos));
-  }
-  return result;
-}
-
-
-// ECMA-262 section 15.5.4.5
-function StringCharCodeAtJS(pos) {
-  CHECK_OBJECT_COERCIBLE(this, "String.prototype.charCodeAt");
-
-  var result = %_StringCharCodeAt(this, pos);
-  if (!%_IsSmi(result)) {
-    result = %_StringCharCodeAt(TO_STRING(this), TO_INTEGER(pos));
-  }
-  return result;
-}
-
-
 // ECMA-262, section 15.5.4.6
 function StringConcat(other /* and more */) {  // length == 1
   "use strict";
@@ -857,13 +833,6 @@ function StringRaw(callSite) {
 
 // -------------------------------------------------------------------
 
-// Set the String function and constructor.
-%FunctionSetPrototype(GlobalString, new GlobalString());
-
-// Set up the constructor property on the String prototype object.
-%AddNamedProperty(
-    GlobalString.prototype, "constructor", GlobalString, DONT_ENUM);
-
 // Set up the non-enumerable functions on the String object.
 utils.InstallFunctions(GlobalString, DONT_ENUM, [
   "fromCharCode", StringFromCharCode,
@@ -875,8 +844,6 @@ utils.InstallFunctions(GlobalString, DONT_ENUM, [
 utils.InstallFunctions(GlobalString.prototype, DONT_ENUM, [
   "valueOf", StringValueOf,
   "toString", StringToString,
-  "charAt", StringCharAtJS,
-  "charCodeAt", StringCharCodeAtJS,
   "codePointAt", StringCodePointAt,
   "concat", StringConcat,
   "endsWith", StringEndsWith,
@@ -922,7 +889,6 @@ utils.InstallFunctions(GlobalString.prototype, DONT_ENUM, [
 
 utils.Export(function(to) {
   to.ExpandReplacement = ExpandReplacement;
-  to.StringCharAt = StringCharAtJS;
   to.StringIndexOf = StringIndexOf;
   to.StringLastIndexOf = StringLastIndexOf;
   to.StringMatch = StringMatchJS;

src/runtime/runtime-internal.cc

@@ -431,6 +431,13 @@ RUNTIME_FUNCTION(Runtime_ThrowCalledNonCallable) {
       isolate, NewTypeError(MessageTemplate::kCalledNonCallable, callsite));
 }
 
+RUNTIME_FUNCTION(Runtime_ThrowCalledOnNullOrUndefined) {
+  HandleScope scope(isolate);
+  DCHECK_EQ(1, args.length());
+  CONVERT_ARG_HANDLE_CHECKED(String, name, 0);
+  THROW_NEW_ERROR_RETURN_FAILURE(
+      isolate, NewTypeError(MessageTemplate::kCalledOnNullOrUndefined, name));
+}
 
 RUNTIME_FUNCTION(Runtime_ThrowConstructedNonConstructable) {
   HandleScope scope(isolate);

src/runtime/runtime-strings.cc

@@ -1273,6 +1273,13 @@ RUNTIME_FUNCTION(Runtime_StringCharAt) {
   return __RT_impl_Runtime_StringCharFromCode(Arguments(1, &code), isolate);
 }
 
+RUNTIME_FUNCTION(Runtime_ExternalStringGetChar) {
+  SealHandleScope shs(isolate);
+  DCHECK_EQ(2, args.length());
+  CONVERT_ARG_CHECKED(ExternalString, string, 0);
+  CONVERT_INT32_ARG_CHECKED(index, 1);
+  return Smi::FromInt(string->Get(index));
+}
 
 RUNTIME_FUNCTION(Runtime_OneByteSeqStringGetChar) {
   SealHandleScope shs(isolate);

src/runtime/runtime.h

@@ -312,6 +312,7 @@ namespace internal {
   F(ThrowConstructedNonConstructable, 1, 1)         \
   F(ThrowDerivedConstructorReturnedNonObject, 0, 1) \
   F(ThrowCalledNonCallable, 1, 1)                   \
+  F(ThrowCalledOnNullOrUndefined, 1, 1)             \
   F(CreateListFromArrayLike, 1, 1)                  \
   F(IncrementUseCounter, 1, 1)                      \
   F(GetOrdinaryHasInstance, 0, 1)                   \
@@ -858,6 +859,7 @@ namespace internal {
   F(FlattenString, 1, 1)                  \
   F(StringCharFromCode, 1, 1)             \
   F(StringCharAt, 2, 1)                   \
+  F(ExternalStringGetChar, 2, 1)          \
   F(OneByteSeqStringGetChar, 2, 1)        \
   F(OneByteSeqStringSetChar, 3, 1)        \
   F(TwoByteSeqStringGetChar, 2, 1)        \