Do not pass the global object as the receiver to strict-mode and

src/ia32/full-codegen-ia32.cc

Index: src/ia32/full-codegen-ia32.cc
diff --git a/src/ia32/full-codegen-ia32.cc b/src/ia32/full-codegen-ia32.cc
index 2dbbf0f473b44193f4ff9fa7252f3270d6e444ea..f9ded85ea2de8bf7fb3959c9f954f68a9c057124 100644
--- a/src/ia32/full-codegen-ia32.cc
+++ b/src/ia32/full-codegen-ia32.cc
@@ -3532,6 +3532,39 @@ void FullCodeGenerator::EmitFastAsciiArrayJoin(ZoneList<Expression*>* args) {
 }
 
 
+void FullCodeGenerator::EmitIsNativeOrStrictMode(ZoneList<Expression*>* args) {
+  ASSERT(args->length() == 1);
+
+  // Load the function into eax.
+  VisitForAccumulatorValue(args->at(0));
+
+  // Prepare for the test.
+  Label materialize_true, materialize_false;
+  Label* if_true = NULL;
+  Label* if_false = NULL;
+  Label* fall_through = NULL;
+  context()->PrepareTest(&materialize_true, &materialize_false,
+                         &if_true, &if_false, &fall_through);
+
+  // Test for strict mode function.
+  __ mov(ecx, FieldOperand(eax, JSFunction::kSharedFunctionInfoOffset));

[Rico, 2011/06/30 06:41:25]

I know that we explicitly check for this at the call sites (in array.js and
string.js). But shouldn't we have a check for what we actually have in eax to
guarantee that it is a JSFunction.

[Mads Ager (chromium), 2011/06/30 12:28:10]

We explicitly have checks before. We will then go on and use %_CallFunction
which does not check either.

+  __ test_b(FieldOperand(ecx, SharedFunctionInfo::kStrictModeByteOffset),
+            1 << SharedFunctionInfo::kStrictModeBitWithinByte);
+  __ j(not_equal, if_true);
+
+  // Test for native function.
+  __ test_b(FieldOperand(ecx, SharedFunctionInfo::kNativeByteOffset),
+            1 << SharedFunctionInfo::kNativeBitWithinByte);
+  __ j(not_equal, if_true);

[Rico, 2011/06/30 06:41:25]

You could consider having just one test instructions (non-byte version) with (1
<< SharedFunctionInfo::kNative & 1 << SharedFunctionInfo::kStrictModeFunction)

[Mads Ager (chromium), 2011/06/30 12:28:10]

We could. This code is similar to what we do in apply. I'll keep them the same
for now and we can see if it makes a performance difference. I don't think it
will.

+
+  // Not native or strict-mode function.
+  __ jmp(if_false);
+
+  PrepareForBailoutBeforeSplit(TOS_REG, true, if_true, if_false);
+  context()->Plug(if_true, if_false);
+}
+
+
 void FullCodeGenerator::VisitCallRuntime(CallRuntime* expr) {
   Handle<String> name = expr->name();
   if (name->length() > 0 && name->Get(0) == '_') {