Pass undefined to JS builtins when called with implicit receiver.

src/x64/full-codegen-x64.cc

Index: src/x64/full-codegen-x64.cc
diff --git a/src/x64/full-codegen-x64.cc b/src/x64/full-codegen-x64.cc
index 329fd18dc47cbbde28cd7a16e274d84159cda4e1..78ae5ddbe8b49cdbbc600ba4e60283af4f1c2f18 100644
--- a/src/x64/full-codegen-x64.cc
+++ b/src/x64/full-codegen-x64.cc
@@ -131,11 +131,11 @@ void FullCodeGenerator::Generate(CompilationInfo* info) {
   }
 #endif
 
-  // Strict mode functions need to replace the receiver with undefined
-  // when called as functions (without an explicit receiver
-  // object). rcx is zero for method calls and non-zero for function
-  // calls.
-  if (info->is_strict_mode()) {
+  // Strict mode functions and builtins need to replace the receiver
+  // with undefined when called as functions (without an explicit
+  // receiver object). rcx is zero for method calls and non-zero for
+  // function calls.

[Lasse Reichstein, 2011/05/26 08:26:55]

Do we know, or could we make sure we know, that the non-zero part of rcx is in
the low 32 bits (i.e., in ecx)? In that case, we can do with a testl(rcx, rcx)
below (and save a byte!)

[Mads Ager (chromium), 2011/05/26 10:14:07]

Yes, we could. Currently, we use a valid smi which means that the marker is in
the high 32 bits. We do that because we have to push the marker on the stack. We
only need the value to be smi tagged though. Let me change this in a separate
change to use a smi tagged non-smi. 

+  if (info->is_strict_mode() || info->is_es5_native()) {
     Label ok;
     __ testq(rcx, rcx);
     __ j(zero, &ok, Label::kNear);