fix performance regression on intel call api stubs

src/x64/code-stubs-x64.cc

Index: src/x64/code-stubs-x64.cc
diff --git a/src/x64/code-stubs-x64.cc b/src/x64/code-stubs-x64.cc
index 66dfe1f8e05c111a08887eabe5da20fb7c9526fe..f572ae2477247a0839648230ba4c695662ecd09b 100644
--- a/src/x64/code-stubs-x64.cc
+++ b/src/x64/code-stubs-x64.cc
@@ -4817,12 +4817,12 @@ static void CallApiFunctionStubHelper(MacroAssembler* masm,
                                       bool return_first_arg,
                                       bool call_data_undefined) {
   // ----------- S t a t e -------------
-  //  -- rax                 : callee
+  //  -- rdi                 : callee
   //  -- rbx                 : call_data
   //  -- rcx                 : holder
   //  -- rdx                 : api_function_address
   //  -- rsi                 : context
-  //  -- rdi                 : number of arguments if argc is a register
+  //  -- rax                 : number of arguments if argc is a register
   //  -- rsp[0]              : return address
   //  -- rsp[8]              : last argument
   //  -- ...
@@ -4830,11 +4830,12 @@ static void CallApiFunctionStubHelper(MacroAssembler* masm,
   //  -- rsp[(argc + 1) * 8] : receiver
   // -----------------------------------
 
-  Register callee = rax;
+  Register callee = rdi;
   Register call_data = rbx;
   Register holder = rcx;
   Register api_function_address = rdx;
   Register context = rsi;
+  Register return_address = r8;
 
   typedef FunctionCallbackArguments FCA;
 
@@ -4847,17 +4848,12 @@ static void CallApiFunctionStubHelper(MacroAssembler* masm,
   STATIC_ASSERT(FCA::kHolderIndex == 0);
   STATIC_ASSERT(FCA::kArgsLength == 7);
 
-  DCHECK(argc.is_immediate() || rdi.is(argc.reg()));
+  DCHECK(argc.is_immediate() || rax.is(argc.reg()));
 
-  if (kPointerSize == kInt64Size) {
-    //  pop return address and save context
-    __ xchgq(context, Operand(rsp, 0));
-  } else {
-    // x32 handling.
-    __ PopReturnAddressTo(kScratchRegister);
-    __ Push(context);
-    __ movq(context, kScratchRegister);
-  }
+  __ PopReturnAddressTo(return_address);
+
+  // context save
+  __ Push(context);
 
   // callee
   __ Push(callee);
@@ -4880,7 +4876,7 @@ static void CallApiFunctionStubHelper(MacroAssembler* masm,
 
   __ movp(scratch, rsp);
   // Push return address back on stack.
-  __ PushReturnAddressFrom(context);
+  __ PushReturnAddressFrom(return_address);
 
   // load context from callee
   __ movp(context, FieldOperand(callee, JSFunction::kContextOffset));
@@ -4954,9 +4950,8 @@ static void CallApiFunctionStubHelper(MacroAssembler* masm,
 
 
 void CallApiFunctionStub::Generate(MacroAssembler* masm) {
-  // TODO(dcarney): make rax contain the function address.
   bool call_data_undefined = this->call_data_undefined();
-  CallApiFunctionStubHelper(masm, ParameterCount(rdi), false,
+  CallApiFunctionStubHelper(masm, ParameterCount(rax), false,
                             call_data_undefined);
 }