Add X32 port into V8

src/x64/code-stubs-x64.cc

 // Copyright 2012 the V8 project authors. All rights reserved.
 // Redistribution and use in source and binary forms, with or without
 // modification, are permitted provided that the following conditions are
 // met:
 //
 //     * Redistributions of source code must retain the above copyright
 //       notice, this list of conditions and the following disclaimer.
 //     * Redistributions in binary form must reproduce the above
 //       copyright notice, this list of conditions and the following
 //       disclaimer in the documentation and/or other materials provided
@@ skipping 209 matching lines @@
   descriptor->register_param_count_ = 1;
   descriptor->register_params_ = registers;
   descriptor->deoptimization_handler_ =
      FUNCTION_ADDR(ToBooleanIC_Miss);
   descriptor->SetMissHandler(
       ExternalReference(IC_Utility(IC::kToBooleanIC_Miss), isolate));
 }
 
 
 #define __ ACCESS_MASM(masm)
+#define __k __
+#define __a __
+#define __q __
+#define __s __
+#define __n __
 
 
 void HydrogenCodeStub::GenerateLightweightMiss(MacroAssembler* masm) {
   // Update the static counter each time a new code stub is generated.
   Isolate* isolate = masm->isolate();
   isolate->counters()->code_stubs()->Increment();
 
   CodeStubInterfaceDescriptor* descriptor = GetInterfaceDescriptor(isolate);
   int param_count = descriptor->register_param_count_;
   {
@@ skipping 20 matching lines @@
   __ j(not_zero, &check_heap_number, Label::kNear);
   __ Ret();
 
   __ bind(&check_heap_number);
   __ CompareRoot(FieldOperand(rax, HeapObject::kMapOffset),
                  Heap::kHeapNumberMapRootIndex);
   __ j(not_equal, &call_builtin, Label::kNear);
   __ Ret();
 
   __ bind(&call_builtin);
-  __ pop(rcx);  // Pop return address.
+  __k pop(rcx);  // Pop return address.
   __ push(rax);
-  __ push(rcx);  // Push return address.
+  __k push(rcx);  // Push return address.
   __ InvokeBuiltin(Builtins::TO_NUMBER, JUMP_FUNCTION);
 }
 
 
 void FastNewClosureStub::Generate(MacroAssembler* masm) {
   // Create a new closure from the given function info in new
   // space. Set the context to the current context in rsi.
   Counters* counters = masm->isolate()->counters();
 
   Label gc;
   __ Allocate(JSFunction::kSize, rax, rbx, rcx, &gc, TAG_OBJECT);
 
   __ IncrementCounter(counters->fast_new_closure_total(), 1);
 
   // Get the function info from the stack.
-  __ movq(rdx, Operand(rsp, 1 * kPointerSize));
+  __a movq(rdx, Operand(rsp, 1 * kPointerSize));
 
   int map_index = Context::FunctionMapIndex(language_mode_, is_generator_);
 
   // Compute the function map in the current native context and set that
   // as the map of the allocated object.
   __ movq(rcx, Operand(rsi, Context::SlotOffset(Context::GLOBAL_OBJECT_INDEX)));
   __ movq(rcx, FieldOperand(rcx, GlobalObject::kNativeContextOffset));
   __ movq(rbx, Operand(rcx, Context::SlotOffset(map_index)));
   __ movq(FieldOperand(rax, JSObject::kMapOffset), rbx);
 
@@ skipping 87 matching lines @@
       rcx,
       Context::SlotOffset(Context::OPTIMIZED_FUNCTIONS_LIST),
       rdx,
       rbx,
       kDontSaveFPRegs);
 
   // Return and remove the on-stack parameter.
   __ ret(1 * kPointerSize);
 
   __ bind(&restore);
-  __ movq(rdx, Operand(rsp, 1 * kPointerSize));
+  __a movq(rdx, Operand(rsp, 1 * kPointerSize));
   __ jmp(&install_unoptimized);
 
   // Create a new closure through the slower runtime call.
   __ bind(&gc);
-  __ pop(rcx);  // Temporarily remove return address.
+  __k pop(rcx);  // Temporarily remove return address.
   __ pop(rdx);
   __ push(rsi);
   __ push(rdx);
   __ PushRoot(Heap::kFalseValueRootIndex);
-  __ push(rcx);  // Restore return address.
+  __k push(rcx);  // Restore return address.
   __ TailCallRuntime(Runtime::kNewClosure, 3, 1);
 }
 
 
 void FastNewContextStub::Generate(MacroAssembler* masm) {
   // Try to allocate the context in new space.
   Label gc;
   int length = slots_ + Context::MIN_CONTEXT_SLOTS;
   __ Allocate((length * kPointerSize) + FixedArray::kHeaderSize,
               rax, rbx, rcx, &gc, TAG_OBJECT);
 
   // Get the function from the stack.
-  __ movq(rcx, Operand(rsp, 1 * kPointerSize));
+  __a movq(rcx, Operand(rsp, 1 * kPointerSize));
 
   // Set up the object header.
   __ LoadRoot(kScratchRegister, Heap::kFunctionContextMapRootIndex);
   __ movq(FieldOperand(rax, HeapObject::kMapOffset), kScratchRegister);
   __ Move(FieldOperand(rax, FixedArray::kLengthOffset), Smi::FromInt(length));
 
   // Set up the fixed slots.
   __ Set(rbx, 0);  // Set to NULL.
   __ movq(Operand(rax, Context::SlotOffset(Context::CLOSURE_INDEX)), rcx);
   __ movq(Operand(rax, Context::SlotOffset(Context::PREVIOUS_INDEX)), rsi);
@@ skipping 25 matching lines @@
   // [rsp + (1 * kPointerSize)] : function
   // [rsp + (2 * kPointerSize)] : serialized scope info
 
   // Try to allocate the context in new space.
   Label gc;
   int length = slots_ + Context::MIN_CONTEXT_SLOTS;
   __ Allocate(FixedArray::SizeFor(length),
               rax, rbx, rcx, &gc, TAG_OBJECT);
 
   // Get the function from the stack.
-  __ movq(rcx, Operand(rsp, 1 * kPointerSize));
+  __a movq(rcx, Operand(rsp, 1 * kPointerSize));
 
   // Get the serialized scope info from the stack.
-  __ movq(rbx, Operand(rsp, 2 * kPointerSize));
+  __a movq(rbx, Operand(rsp, 2 * kPointerSize));
 
   // Set up the object header.
   __ LoadRoot(kScratchRegister, Heap::kBlockContextMapRootIndex);
   __ movq(FieldOperand(rax, HeapObject::kMapOffset), kScratchRegister);
   __ Move(FieldOperand(rax, FixedArray::kLengthOffset), Smi::FromInt(length));
 
   // If this block context is nested in the native context we get a smi
   // sentinel instead of a function. The block context should get the
   // canonical empty function of the native context as its closure which
   // we still have to look up.
@@ skipping 103 matching lines @@
                     Register source) {
   // Result may be rcx. If result and source are the same register, source will
   // be overwritten.
   ASSERT(!result.is(rdi) && !result.is(rbx));
   // TODO(lrn): When type info reaches here, if value is a 32-bit integer, use
   // cvttsd2si (32-bit version) directly.
   Register double_exponent = rbx;
   Register double_value = rdi;
   Label done, exponent_63_plus;
   // Get double and extract exponent.
-  __ movq(double_value, FieldOperand(source, HeapNumber::kValueOffset));
+  __k movq(double_value, FieldOperand(source, HeapNumber::kValueOffset));
   // Clear result preemptively, in case we need to return zero.
   __ xorl(result, result);
   __ movq(xmm0, double_value);  // Save copy in xmm0 in case we need it there.
   // Double to remove sign bit, shift exponent down to least significant bits.
   // and subtract bias to get the unshifted, unbiased exponent.
-  __ lea(double_exponent, Operand(double_value, double_value, times_1, 0));
-  __ shr(double_exponent, Immediate(64 - HeapNumber::kExponentBits));
+  __k lea(double_exponent, Operand(double_value, double_value, times_1, 0));
+  __k shr(double_exponent, Immediate(64 - HeapNumber::kExponentBits));
   __ subl(double_exponent, Immediate(HeapNumber::kExponentBias));
   // Check whether the exponent is too big for a 63 bit unsigned integer.
   __ cmpl(double_exponent, Immediate(63));
   __ j(above_equal, &exponent_63_plus, Label::kNear);
   // Handle exponent range 0..62.
   __ cvttsd2siq(result, xmm0);
   __ jmp(&done, Label::kNear);
 
   __ bind(&exponent_63_plus);
   // Exponent negative or 63+.
   __ cmpl(double_exponent, Immediate(83));
   // If exponent negative or above 83, number contains no significant bits in
   // the range 0..2^31, so result is zero, and rcx already holds zero.
   __ j(above, &done, Label::kNear);
 
   // Exponent in rage 63..83.
   // Mantissa * 2^exponent contains bits in the range 2^0..2^31, namely
   // the least significant exponent-52 bits.
 
   // Negate low bits of mantissa if value is negative.
-  __ addq(double_value, double_value);  // Move sign bit to carry.
+  __k addq(double_value, double_value);  // Move sign bit to carry.
   __ sbbl(result, result);  // And convert carry to -1 in result register.
   // if scratch2 is negative, do (scratch2-1)^-1, otherwise (scratch2-0)^0.
   __ addl(double_value, result);
   // Do xor in opposite directions depending on where we want the result
   // (depending on whether result is rcx or not).
 
   if (result.is(rcx)) {
     __ xorl(double_value, result);
     // Left shift mantissa by (exponent - mantissabits - 1) to save the
     // bits that have positional values below 2^32 (the extra -1 comes from the
@@ skipping 24 matching lines @@
       GenerateNumberStub(masm);
       break;
     case UnaryOpIC::GENERIC:
       GenerateGenericStub(masm);
       break;
   }
 }
 
 
 void UnaryOpStub::GenerateTypeTransition(MacroAssembler* masm) {
-  __ pop(rcx);  // Save return address.
+  __k pop(rcx);  // Save return address.
 
   __ push(rax);  // the operand
   __ Push(Smi::FromInt(op_));
   __ Push(Smi::FromInt(mode_));
   __ Push(Smi::FromInt(operand_type_));
 
-  __ push(rcx);  // Push return address.
+  __k push(rcx);  // Push return address.
 
   // Patch the caller to an appropriate specialized stub and return the
   // operation result to the caller of the stub.
   __ TailCallExternalReference(
       ExternalReference(IC_Utility(IC::kUnaryOp_Patch), masm->isolate()), 4, 1);
 }
 
 
 // TODO(svenpanne): Use virtual functions instead of switch.
 void UnaryOpStub::GenerateSmiStub(MacroAssembler* masm) {
@@ skipping 90 matching lines @@
 void UnaryOpStub::GenerateHeapNumberCodeSub(MacroAssembler* masm,
                                             Label* slow) {
   // Check if the operand is a heap number.
   __ CompareRoot(FieldOperand(rax, HeapObject::kMapOffset),
                  Heap::kHeapNumberMapRootIndex);
   __ j(not_equal, slow);
 
   // Operand is a float, negate its value by flipping the sign bit.
   if (mode_ == UNARY_OVERWRITE) {
     __ Set(kScratchRegister, 0x01);
-    __ shl(kScratchRegister, Immediate(63));
-    __ xor_(FieldOperand(rax, HeapNumber::kValueOffset), kScratchRegister);
+    __k shl(kScratchRegister, Immediate(63));
+    __k xor_(FieldOperand(rax, HeapNumber::kValueOffset), kScratchRegister);
   } else {
     // Allocate a heap number before calculating the answer,
     // so we don't have an untagged double around during GC.
     Label slow_allocate_heapnumber, heapnumber_allocated;
     __ AllocateHeapNumber(rcx, rbx, &slow_allocate_heapnumber);
     __ jmp(&heapnumber_allocated);
 
     __ bind(&slow_allocate_heapnumber);
     {
       FrameScope scope(masm, StackFrame::INTERNAL);
       __ push(rax);
       __ CallRuntime(Runtime::kNumberAlloc, 0);
       __ movq(rcx, rax);
       __ pop(rax);
     }
     __ bind(&heapnumber_allocated);
     // rcx: allocated 'empty' number
 
     // Copy the double value to the new heap number, flipping the sign.
-    __ movq(rdx, FieldOperand(rax, HeapNumber::kValueOffset));
+    __k movq(rdx, FieldOperand(rax, HeapNumber::kValueOffset));
     __ Set(kScratchRegister, 0x01);
-    __ shl(kScratchRegister, Immediate(63));
-    __ xor_(rdx, kScratchRegister);  // Flip sign.
-    __ movq(FieldOperand(rcx, HeapNumber::kValueOffset), rdx);
+    __k shl(kScratchRegister, Immediate(63));
+    __k xor_(rdx, kScratchRegister);  // Flip sign.
+    __k movq(FieldOperand(rcx, HeapNumber::kValueOffset), rdx);
     __ movq(rax, rcx);
   }
   __ ret(0);
 }
 
 
 void UnaryOpStub::GenerateHeapNumberCodeBitNot(MacroAssembler* masm,
                                                Label* slow) {
   // Check if the operand is a heap number.
   __ CompareRoot(FieldOperand(rax, HeapObject::kMapOffset),
                  Heap::kHeapNumberMapRootIndex);
   __ j(not_equal, slow);
 
+#ifndef V8_TARGET_ARCH_X32
   // Convert the heap number in rax to an untagged integer in rcx.
   IntegerConvert(masm, rax, rax);
 
   // Do the bitwise operation and smi tag the result.
   __ notl(rax);
   __ Integer32ToSmi(rax, rax);
   __ ret(0);
+#else
+  // Convert the heap number in rax to an untagged integer in rcx.
+  IntegerConvert(masm, rcx, rax);
+
+  // Do the bitwise operation and smi tag the result.
+  Label try_float;
+  __ notl(rcx);
+  __ cmpl(rcx, Immediate(0xc0000000));
+  __ j(sign, &try_float, Label::kNear);
+  __ Integer32ToSmi(rax, rcx);
+  __ ret(0);
+
+  // Try to store the result in a heap number.
+  __ bind(&try_float);
+  if (mode_ == UNARY_NO_OVERWRITE) {
+    Label slow_allocate_heapnumber, heapnumber_allocated;
+    __ movl(rbx, rax);
+    __ AllocateHeapNumber(rax, kScratchRegister, &slow_allocate_heapnumber);
+    __ jmp(&heapnumber_allocated);
+
+    __ bind(&slow_allocate_heapnumber);
+    {
+      FrameScope scope(masm, StackFrame::INTERNAL);
+      // Push the original HeapNumber on the stack. The integer value can't
+      // be stored since it's untagged and not in the smi range (so we can't
+      // smi-tag it). We'll recalculate the value after the GC instead.
+      __ Push(rbx);
+      __ CallRuntime(Runtime::kNumberAlloc, 0);
+      // New HeapNumber is in eax.
+      __ Pop(rbx);
+    }
+    // Recalcuate bit-not value.
+    IntegerConvert(masm, rcx, rbx);
+    __ notl(rcx);
+
+    __ bind(&heapnumber_allocated);
+  }
+  __ cvtlsi2sd(xmm0, rcx);
+  __ movsd(FieldOperand(rax, HeapNumber::kValueOffset), xmm0);
+  __ ret(0);
+#endif
 }
 
 
 // TODO(svenpanne): Use virtual functions instead of switch.
 void UnaryOpStub::GenerateGenericStub(MacroAssembler* masm) {
   switch (op_) {
     case Token::SUB:
       GenerateGenericStubSub(masm);
       break;
     case Token::BIT_NOT:
@@ skipping 20 matching lines @@
   GenerateSmiCodeBitNot(masm, &non_smi, Label::kNear);
   __ bind(&non_smi);
   GenerateHeapNumberCodeBitNot(masm, &slow);
   __ bind(&slow);
   GenerateGenericCodeFallback(masm);
 }
 
 
 void UnaryOpStub::GenerateGenericCodeFallback(MacroAssembler* masm) {
   // Handle the slow case by jumping to the JavaScript builtin.
-  __ pop(rcx);  // pop return address
+  __k pop(rcx);  // pop return address
   __ push(rax);
-  __ push(rcx);  // push return address
+  __k push(rcx);  // push return address
   switch (op_) {
     case Token::SUB:
       __ InvokeBuiltin(Builtins::UNARY_MINUS, JUMP_FUNCTION);
       break;
     case Token::BIT_NOT:
       __ InvokeBuiltin(Builtins::BIT_NOT, JUMP_FUNCTION);
       break;
     default:
       UNREACHABLE();
   }
@@ skipping 11 matching lines @@
               op_name,
               overwrite_name,
               UnaryOpIC::GetName(operand_type_));
 }
 
 
 void BinaryOpStub::Initialize() {}
 
 
 void BinaryOpStub::GenerateTypeTransition(MacroAssembler* masm) {
-  __ pop(rcx);  // Save return address.
+  __k pop(rcx);  // Save return address.
   __ push(rdx);
   __ push(rax);
   // Left and right arguments are now on top.
   __ Push(Smi::FromInt(MinorKey()));
 
-  __ push(rcx);  // Push return address.
+  __k push(rcx);  // Push return address.
 
   // Patch the caller to an appropriate specialized stub and return the
   // operation result to the caller of the stub.
   __ TailCallExternalReference(
       ExternalReference(IC_Utility(IC::kBinaryOp_Patch),
                         masm->isolate()),
       3,
       1);
 }
 
 
 static void BinaryOpStub_GenerateSmiCode(
     MacroAssembler* masm,
     Label* slow,
     BinaryOpStub::SmiCodeGenerateHeapNumberResults allow_heapnumber_results,
     Token::Value op) {
 
   // Arguments to BinaryOpStub are in rdx and rax.
   const Register left = rdx;
   const Register right = rax;
 
   // We only generate heapnumber answers for overflowing calculations
   // for the four basic arithmetic operations and logical right shift by 0.
+#ifndef V8_TARGET_ARCH_X32
   bool generate_inline_heapnumber_results =
       (allow_heapnumber_results == BinaryOpStub::ALLOW_HEAPNUMBER_RESULTS) &&
       (op == Token::ADD || op == Token::SUB ||
        op == Token::MUL || op == Token::DIV || op == Token::SHR);
+#else
+  bool generate_inline_heapnumber_results =
+      (allow_heapnumber_results == BinaryOpStub::ALLOW_HEAPNUMBER_RESULTS) &&
+      (op == Token::ADD || op == Token::SUB || op == Token::SHL ||
+       op == Token::MUL || op == Token::DIV || op == Token::SHR);
+#endif
 
   // Smi check of both operands.  If op is BIT_OR, the check is delayed
   // until after the OR operation.
   Label not_smis;
   Label use_fp_on_smis;
   Label fail;
 
   if (op != Token::BIT_OR) {
     Comment smi_check_comment(masm, "-- Smi check arguments");
     __ JumpIfNotBothSmi(left, right, &not_smis);
@@ skipping 42 matching lines @@
       ASSERT(right.is(rax));
       __ SmiXor(right, right, left);  // BIT_XOR is commutative.
       break;
 
     case Token::BIT_AND:
       ASSERT(right.is(rax));
       __ SmiAnd(right, right, left);  // BIT_AND is commutative.
       break;
 
     case Token::SHL:
+#ifndef V8_TARGET_ARCH_X32
       __ SmiShiftLeft(left, left, right);
+#else
+      __ movl(kScratchRegister, left);
+      __ SmiShiftLeft(left, left, right, &use_fp_on_smis);
+#endif
       __ movq(rax, left);
       break;
 
     case Token::SAR:
       __ SmiShiftArithmeticRight(left, left, right);
       __ movq(rax, left);
       break;
 
     case Token::SHR:
+#ifdef V8_TARGET_ARCH_X32
+      __ movl(kScratchRegister, left);
+#endif
       __ SmiShiftLogicalRight(left, left, right, &use_fp_on_smis);
       __ movq(rax, left);
       break;
 
     default:
       UNREACHABLE();
   }
 
   // 5. Emit return of result in rax.  Some operations have registers pushed.
   __ ret(0);
 
   if (use_fp_on_smis.is_linked()) {
     // 6. For some operations emit inline code to perform floating point
     //    operations on known smis (e.g., if the result of the operation
     //    overflowed the smi range).
     __ bind(&use_fp_on_smis);
     if (op == Token::DIV || op == Token::MOD) {
       // Restore left and right to rdx and rax.
       __ movq(rdx, rcx);
       __ movq(rax, rbx);
     }
 
     if (generate_inline_heapnumber_results) {
       __ AllocateHeapNumber(rcx, rbx, slow);
       Comment perform_float(masm, "-- Perform float operation on smis");
+#ifndef V8_TARGET_ARCH_X32
       if (op == Token::SHR) {
         __ SmiToInteger32(left, left);
         __ cvtqsi2sd(xmm0, left);
       } else {
+#else
+      if (op == Token::SHL) {
+        __ cvtlsi2sd(xmm0, left);
+      } else if (op == Token::SHR) {
+        // The value of left is from MacroAssembler::SmiShiftLogicalRight
+        // We allow logical shift value:
+        // 0 : might turn a signed integer into unsigned integer
+        // 1 : the value might be above 2^30 - 1
+        __ cvtqsi2sd(xmm0, left);
+      } else {
+#endif
         FloatingPointHelper::LoadSSE2SmiOperands(masm);
         switch (op) {
         case Token::ADD: __ addsd(xmm0, xmm1); break;
         case Token::SUB: __ subsd(xmm0, xmm1); break;
         case Token::MUL: __ mulsd(xmm0, xmm1); break;
         case Token::DIV: __ divsd(xmm0, xmm1); break;
         default: UNREACHABLE();
         }
       }
       __ movsd(FieldOperand(rcx, HeapNumber::kValueOffset), xmm0);
       __ movq(rax, rcx);
       __ ret(0);
     } else {
+#ifdef V8_TARGET_ARCH_X32
+      // Restore the orignial left value from kScratchRegister for stub call
+      // KScratchRegister is not killed by MacroAssembler::SmiShiftLogicalRight
+      // and is not killed by MacroAssembler::SmiShiftLeft either.
+      if (op == Token::SHL || op == Token::SHR) {
+        __ movl(left, kScratchRegister);
+      }
+#endif
       __ jmp(&fail);
     }
   }
 
   // 7. Non-smi operands reach the end of the code generated by
   //    GenerateSmiCode, and fall through to subsequent code,
   //    with the operands in rdx and rax.
   //    But first we check if non-smi values are HeapNumbers holding
   //    values that could be smi.
   __ bind(&not_smis);
@@ skipping 18 matching lines @@
 
 static void BinaryOpStub_GenerateHeapResultAllocation(MacroAssembler* masm,
                                                       Label* alloc_failure,
                                                       OverwriteMode mode);
 
 
 static void BinaryOpStub_GenerateFloatingPointCode(MacroAssembler* masm,
                                                    Label* allocation_failure,
                                                    Label* non_numeric_failure,
                                                    Token::Value op,
+#ifdef V8_TARGET_ARCH_X32
+                                                   BinaryOpIC::TypeInfo
+                                                       result_type,
+                                                   Label* non_int32_failure,
+#endif
                                                    OverwriteMode mode) {
   switch (op) {
     case Token::ADD:
     case Token::SUB:
     case Token::MUL:
     case Token::DIV: {
       FloatingPointHelper::LoadSSE2UnknownOperands(masm, non_numeric_failure);
 
       switch (op) {
         case Token::ADD: __ addsd(xmm0, xmm1); break;
         case Token::SUB: __ subsd(xmm0, xmm1); break;
         case Token::MUL: __ mulsd(xmm0, xmm1); break;
         case Token::DIV: __ divsd(xmm0, xmm1); break;
         default: UNREACHABLE();
       }
+#ifdef V8_TARGET_ARCH_X32
+      if (non_int32_failure != NULL) {
+        if (result_type <= BinaryOpIC::INT32) {
+          __ cvttsd2si(kScratchRegister, xmm0);
+          __ cvtlsi2sd(xmm2, kScratchRegister);
+          __ pcmpeqd(xmm2, xmm0);
+          __ movmskpd(rcx, xmm2);
+          __ testl(rcx, Immediate(1));
+          __ j(zero, non_int32_failure);
+        }
+      }
+#endif
       BinaryOpStub_GenerateHeapResultAllocation(
           masm, allocation_failure, mode);
       __ movsd(FieldOperand(rax, HeapNumber::kValueOffset), xmm0);
       __ ret(0);
       break;
     }
     case Token::MOD: {
       // For MOD we jump to the allocation_failure label, to call runtime.
       __ jmp(allocation_failure);
       break;
     }
     case Token::BIT_OR:
     case Token::BIT_AND:
     case Token::BIT_XOR:
     case Token::SAR:
     case Token::SHL:
     case Token::SHR: {
       Label non_smi_shr_result;
       Register heap_number_map = r9;
+#ifdef V8_TARGET_ARCH_X32
+      __ movl(kScratchRegister, rax);
+#endif
       __ LoadRoot(heap_number_map, Heap::kHeapNumberMapRootIndex);
       FloatingPointHelper::LoadAsIntegers(masm, non_numeric_failure,
                                           heap_number_map);
       switch (op) {
         case Token::BIT_OR:  __ orl(rax, rcx); break;
         case Token::BIT_AND: __ andl(rax, rcx); break;
         case Token::BIT_XOR: __ xorl(rax, rcx); break;
         case Token::SAR: __ sarl_cl(rax); break;
         case Token::SHL: __ shll_cl(rax); break;
         case Token::SHR: {
           __ shrl_cl(rax);
+#ifndef V8_TARGET_ARCH_X32
           // Check if result is negative. This can only happen for a shift
           // by zero.
           __ testl(rax, rax);
           __ j(negative, &non_smi_shr_result);
+#endif
           break;
         }
         default: UNREACHABLE();
       }
+#ifndef V8_TARGET_ARCH_X32
       STATIC_ASSERT(kSmiValueSize == 32);
+#else
+      STATIC_ASSERT(kSmiValueSize == 31);
+      if (op == Token::SHR) {
+        __ testl(rax, Immediate(0xc0000000));
+        __ j(not_zero, &non_smi_shr_result);
+      } else {
+        __ cmpl(rax, Immediate(0xc0000000));
+        __ j(negative, &non_smi_shr_result, Label::kNear);
+      }
+#endif
       // Tag smi result and return.
       __ Integer32ToSmi(rax, rax);
       __ Ret();
 
+#ifndef V8_TARGET_ARCH_X32
       // Logical shift right can produce an unsigned int32 that is not
       // an int32, and so is not in the smi range.  Allocate a heap number
       // in that case.
       if (op == Token::SHR) {
         __ bind(&non_smi_shr_result);
         Label allocation_failed;
         __ movl(rbx, rax);  // rbx holds result value (uint32 value as int64).
         // Allocate heap number in new space.
         // Not using AllocateHeapNumber macro in order to reuse
         // already loaded heap_number_map.
         __ Allocate(HeapNumber::kSize, rax, rdx, no_reg, &allocation_failed,
                     TAG_OBJECT);
         // Set the map.
         __ AssertRootValue(heap_number_map,
                            Heap::kHeapNumberMapRootIndex,
                            "HeapNumberMap register clobbered.");
         __ movq(FieldOperand(rax, HeapObject::kMapOffset),
                 heap_number_map);
         __ cvtqsi2sd(xmm0, rbx);
         __ movsd(FieldOperand(rax, HeapNumber::kValueOffset), xmm0);
         __ Ret();
 
         __ bind(&allocation_failed);
         // We need tagged values in rdx and rax for the following code,
         // not int32 in rax and rcx.
         __ Integer32ToSmi(rax, rcx);
         __ Integer32ToSmi(rdx, rbx);
         __ jmp(allocation_failure);
       }
+#else
+      __ bind(&non_smi_shr_result);
+      Label allocation_failed;
+      __ movl(rbx, rax);  // rbx holds result value (uint32 value as int64).
+      // Allocate heap number in new space.
+      // Not using AllocateHeapNumber macro in order to reuse
+      // already loaded heap_number_map.
+      __ Allocate(HeapNumber::kSize, rax, r8, no_reg, &allocation_failed,
+                  TAG_OBJECT);
+      // Set the map.
+      __ AssertRootValue(heap_number_map,
+                         Heap::kHeapNumberMapRootIndex,
+                         "HeapNumberMap register clobbered.");
+      __ movl(FieldOperand(rax, HeapObject::kMapOffset),
+              heap_number_map);
+      if (op == Token::SHR) {
+        __ cvtqsi2sd(xmm0, rbx);
+      } else {
+        // All other operations returns a signed int32, so we
+        // use lsi2sd here to retain the sign bit.
+        __ cvtlsi2sd(xmm0, rbx);
+      }
+      __ movsd(FieldOperand(rax, HeapNumber::kValueOffset), xmm0);
+      __ Ret();
+
+      __ bind(&allocation_failed);
+      // Restore the right operand from kScratchRegister.
+      // Left operand is in rdx, not changed in this function.
+      __ movl(rax, kScratchRegister);
+      __ jmp(allocation_failure);
+#endif
       break;
     }
     default: UNREACHABLE(); break;
   }
   // No fall-through from this generated code.
   if (FLAG_debug_code) {
     __ Abort("Unexpected fall-through in "
              "BinaryStub_GenerateFloatingPointCode.");
   }
 }
 
 
 static void BinaryOpStub_GenerateRegisterArgsPushUnderReturn(
     MacroAssembler* masm) {
   // Push arguments, but ensure they are under the return address
   // for a tail call.
-  __ pop(rcx);
+  __k pop(rcx);
   __ push(rdx);
   __ push(rax);
-  __ push(rcx);
+  __k push(rcx);
 }
 
 
 void BinaryOpStub::GenerateAddStrings(MacroAssembler* masm) {
   ASSERT(op_ == Token::ADD);
   Label left_not_string, call_runtime;
 
   // Registers containing left and right operands respectively.
   Register left = rdx;
   Register right = rax;
@@ skipping 55 matching lines @@
       FrameScope scope(masm, StackFrame::INTERNAL);
       GenerateRegisterArgsPush(masm);
       GenerateCallRuntime(masm);
     }
     __ Ret();
   }
 }
 
 
 void BinaryOpStub::GenerateInt32Stub(MacroAssembler* masm) {
+#ifndef V8_TARGET_ARCH_X32
   // The int32 case is identical to the Smi case.  We avoid creating this
   // ic state on x64.
   UNREACHABLE();
+#else
+  ASSERT(Max(left_type_, right_type_) == BinaryOpIC::INT32);
+
+  Label gc_required, not_number, not_int32;
+  BinaryOpStub_GenerateFloatingPointCode(masm, &gc_required, &not_number,
+                                         op_, result_type_, &not_int32, mode_);
+
+  __ bind(&not_number);
+  __ bind(&not_int32);
+  GenerateTypeTransition(masm);
+
+  __ bind(&gc_required);
+  {
+    FrameScope scope(masm, StackFrame::INTERNAL);
+    GenerateRegisterArgsPush(masm);
+    GenerateCallRuntime(masm);
+  }
+  __ Ret();
+#endif
 }
 
 
 void BinaryOpStub::GenerateBothStringStub(MacroAssembler* masm) {
   Label call_runtime;
   ASSERT(left_type_ == BinaryOpIC::STRING && right_type_ == BinaryOpIC::STRING);
   ASSERT(op_ == Token::ADD);
   // If both arguments are strings, call the string add stub.
   // Otherwise, do a transition.
 
@@ skipping 65 matching lines @@
   // HeapNumbers containing 32bit integer values are also allowed.
   __ LoadRoot(heap_number_map, Heap::kHeapNumberMapRootIndex);
   __ cmpq(FieldOperand(input, HeapObject::kMapOffset), heap_number_map);
   __ j(not_equal, fail);
   __ movsd(xmm0, FieldOperand(input, HeapNumber::kValueOffset));
   // Convert, convert back, and compare the two doubles' bits.
   __ cvttsd2siq(scratch2, xmm0);
   __ cvtlsi2sd(xmm1, scratch2);
   __ movq(scratch1, xmm0);
   __ movq(scratch2, xmm1);
-  __ cmpq(scratch1, scratch2);
+  __k cmpq(scratch1, scratch2);
   __ j(not_equal, fail);
   __ bind(&ok);
 }
 
 
 void BinaryOpStub::GenerateNumberStub(MacroAssembler* masm) {
   Label gc_required, not_number;
 
   // It could be that only SMIs have been seen at either the left
   // or the right operand. For precise type feedback, patch the IC
   // again if this changes.
   if (left_type_ == BinaryOpIC::SMI) {
     BinaryOpStub_CheckSmiInput(masm, rdx, &not_number);
   }
   if (right_type_ == BinaryOpIC::SMI) {
     BinaryOpStub_CheckSmiInput(masm, rax, &not_number);
   }
 
+#ifndef V8_TARGET_ARCH_X32
   BinaryOpStub_GenerateFloatingPointCode(
       masm, &gc_required, &not_number, op_, mode_);
+#else
+  BinaryOpStub_GenerateFloatingPointCode(
+      masm, &gc_required, &not_number, op_, result_type_, NULL, mode_);
+#endif
 
   __ bind(&not_number);
   GenerateTypeTransition(masm);
 
   __ bind(&gc_required);
   {
     FrameScope scope(masm, StackFrame::INTERNAL);
     GenerateRegisterArgsPush(masm);
     GenerateCallRuntime(masm);
   }
   __ Ret();
 }
 
 
 void BinaryOpStub::GenerateGeneric(MacroAssembler* masm) {
   Label call_runtime, call_string_add_or_runtime;
 
   BinaryOpStub_GenerateSmiCode(
       masm, &call_runtime, ALLOW_HEAPNUMBER_RESULTS, op_);
 
+#ifndef V8_TARGET_ARCH_X32
   BinaryOpStub_GenerateFloatingPointCode(
       masm, &call_runtime, &call_string_add_or_runtime, op_, mode_);
+#else
+  BinaryOpStub_GenerateFloatingPointCode(
+      masm, &call_runtime, &call_string_add_or_runtime, op_,
+      result_type_, NULL, mode_);
+#endif
 
   __ bind(&call_string_add_or_runtime);
   if (op_ == Token::ADD) {
     GenerateAddStrings(masm);
   }
 
   __ bind(&call_runtime);
   {
     FrameScope scope(masm, StackFrame::INTERNAL);
     GenerateRegisterArgsPush(masm);
@@ skipping 62 matching lines @@
   //   Output:
   //     xmm1   : untagged double result.
 
   Label runtime_call;
   Label runtime_call_clear_stack;
   Label skip_cache;
   const bool tagged = (argument_type_ == TAGGED);
   if (tagged) {
     Label input_not_smi, loaded;
     // Test that rax is a number.
-    __ movq(rax, Operand(rsp, kPointerSize));
+    __a movq(rax, Operand(rsp, 1 * kPointerSize));
     __ JumpIfNotSmi(rax, &input_not_smi, Label::kNear);
     // Input is a smi. Untag and load it onto the FPU stack.
     // Then load the bits of the double into rbx.
     __ SmiToInteger32(rax, rax);
     __ subq(rsp, Immediate(kDoubleSize));
     __ cvtlsi2sd(xmm1, rax);
     __ movsd(Operand(rsp, 0), xmm1);
     __ movq(rbx, xmm1);
     __ movq(rdx, xmm1);
     __ fld_d(Operand(rsp, 0));
     __ addq(rsp, Immediate(kDoubleSize));
     __ jmp(&loaded, Label::kNear);
 
     __ bind(&input_not_smi);
     // Check if input is a HeapNumber.
     __ LoadRoot(rbx, Heap::kHeapNumberMapRootIndex);
     __ cmpq(rbx, FieldOperand(rax, HeapObject::kMapOffset));
     __ j(not_equal, &runtime_call);
     // Input is a HeapNumber. Push it on the FPU stack and load its
     // bits into rbx.
     __ fld_d(FieldOperand(rax, HeapNumber::kValueOffset));
-    __ movq(rbx, FieldOperand(rax, HeapNumber::kValueOffset));
-    __ movq(rdx, rbx);
+    __k movq(rbx, FieldOperand(rax, HeapNumber::kValueOffset));
+    __k movq(rdx, rbx);
 
     __ bind(&loaded);
   } else {  // UNTAGGED.
     __ movq(rbx, xmm1);
     __ movq(rdx, xmm1);
   }
 
   // ST[0] == double value, if TAGGED.
   // rbx = bits of double value.
   // rdx = also bits of double value.
   // Compute hash (h is 32 bits, bits are 64 and the shifts are arithmetic):
   //   h = h0 = bits ^ (bits >> 32);
   //   h ^= h >> 16;
   //   h ^= h >> 8;
   //   h = h & (cacheSize - 1);
   // or h = (h0 ^ (h0 >> 8) ^ (h0 >> 16) ^ (h0 >> 24)) & (cacheSize - 1)
-  __ sar(rdx, Immediate(32));
+  __k sar(rdx, Immediate(32));
   __ xorl(rdx, rbx);
   __ movl(rcx, rdx);
   __ movl(rax, rdx);
   __ movl(rdi, rdx);
   __ sarl(rdx, Immediate(8));
   __ sarl(rcx, Immediate(16));
   __ sarl(rax, Immediate(24));
   __ xorl(rcx, rdx);
   __ xorl(rax, rdi);
   __ xorl(rcx, rax);
@@ skipping 23 matching lines @@
     char* elem_in1  = reinterpret_cast<char*>(&(test_elem[0].in[1]));
     char* elem_out = reinterpret_cast<char*>(&(test_elem[0].output));
     // Two uint_32's and a pointer per element.
     CHECK_EQ(2 * kIntSize + 1 * kPointerSize,
              static_cast<int>(elem2_start - elem_start));
     CHECK_EQ(0, static_cast<int>(elem_in0 - elem_start));
     CHECK_EQ(kIntSize, static_cast<int>(elem_in1 - elem_start));
     CHECK_EQ(2 * kIntSize, static_cast<int>(elem_out - elem_start));
   }
 #endif
+#ifndef V8_TARGET_ARCH_X32
   // Find the address of the rcx'th entry in the cache, i.e., &rax[rcx*16].
   __ addl(rcx, rcx);
   __ lea(rcx, Operand(rax, rcx, times_8, 0));
+#else
+  // Find the address of the rcx'th entry in the cache, i.e., &rax[rcx*12].
+  __ leal(rcx, Operand(rcx, rcx, times_2, 0));
+  __ leal(rcx, Operand(rax, rcx, times_4, 0));
+#endif
   // Check if cache matches: Double value is stored in uint32_t[2] array.
   Label cache_miss;
-  __ cmpq(rbx, Operand(rcx, 0));
+  __k cmpq(rbx, Operand(rcx, 0));
   __ j(not_equal, &cache_miss, Label::kNear);
   // Cache hit!
   Counters* counters = masm->isolate()->counters();
   __ IncrementCounter(counters->transcendental_cache_hit(), 1);
   __ movq(rax, Operand(rcx, 2 * kIntSize));
   if (tagged) {
     __ fstp(0);  // Clear FPU stack.
     __ ret(kPointerSize);
   } else {  // UNTAGGED.
     __ movsd(xmm1, FieldOperand(rax, HeapNumber::kValueOffset));
     __ Ret();
   }
 
   __ bind(&cache_miss);
   __ IncrementCounter(counters->transcendental_cache_miss(), 1);
   // Update cache with new value.
   if (tagged) {
   __ AllocateHeapNumber(rax, rdi, &runtime_call_clear_stack);
   } else {  // UNTAGGED.
     __ AllocateHeapNumber(rax, rdi, &skip_cache);
     __ movsd(FieldOperand(rax, HeapNumber::kValueOffset), xmm1);
     __ fld_d(FieldOperand(rax, HeapNumber::kValueOffset));
   }
   GenerateOperation(masm, type_);
-  __ movq(Operand(rcx, 0), rbx);
+  __k movq(Operand(rcx, 0), rbx);
   __ movq(Operand(rcx, 2 * kIntSize), rax);
   __ fstp_d(FieldOperand(rax, HeapNumber::kValueOffset));
   if (tagged) {
     __ ret(kPointerSize);
   } else {  // UNTAGGED.
     __ movsd(xmm1, FieldOperand(rax, HeapNumber::kValueOffset));
     __ Ret();
 
     // Skip cache and return answer directly, only in untagged case.
     __ bind(&skip_cache);
@@ skipping 62 matching lines @@
   Label done;
   if (type == TranscendentalCache::SIN ||
       type == TranscendentalCache::COS ||
       type == TranscendentalCache::TAN) {
     // Both fsin and fcos require arguments in the range +/-2^63 and
     // return NaN for infinities and NaN. They can share all code except
     // the actual fsin/fcos operation.
     Label in_range;
     // If argument is outside the range -2^63..2^63, fsin/cos doesn't
     // work. We must reduce it to the appropriate range.
-    __ movq(rdi, rbx);
+    __k movq(rdi, rbx);
     // Move exponent and sign bits to low bits.
-    __ shr(rdi, Immediate(HeapNumber::kMantissaBits));
+    __k shr(rdi, Immediate(HeapNumber::kMantissaBits));
     // Remove sign bit.
     __ andl(rdi, Immediate((1 << HeapNumber::kExponentBits) - 1));
     int supported_exponent_limit = (63 + HeapNumber::kExponentBias);
     __ cmpl(rdi, Immediate(supported_exponent_limit));
     __ j(below, &in_range);
     // Check for infinity and NaN. Both return NaN for sin.
     __ cmpl(rdi, Immediate(0x7ff));
     Label non_nan_result;
     __ j(not_equal, &non_nan_result, Label::kNear);
     // Input is +/-Infinity or NaN. Result is NaN.
     __ fstp(0);
     // NaN is represented by 0x7ff8000000000000.
-    __ subq(rsp, Immediate(kPointerSize));
+    __ subq(rsp, Immediate(kDoubleSize));
     __ movl(Operand(rsp, 4), Immediate(0x7ff80000));
     __ movl(Operand(rsp, 0), Immediate(0x00000000));
     __ fld_d(Operand(rsp, 0));
-    __ addq(rsp, Immediate(kPointerSize));
+    __ addq(rsp, Immediate(kDoubleSize));
     __ jmp(&done);
 
     __ bind(&non_nan_result);
 
     // Use fpmod to restrict argument to the range +/-2*PI.
-    __ movq(rdi, rax);  // Save rax before using fnstsw_ax.
+    __k movq(rdi, rax);  // Save rax before using fnstsw_ax.
     __ fldpi();
     __ fadd(0);
     __ fld(1);
     // FPU Stack: input, 2*pi, input.
     {
       Label no_exceptions;
       __ fwait();
       __ fnstsw_ax();
       // Clear if Illegal Operand or Zero Division exceptions are set.
       __ testl(rax, Immediate(5));  // #IO and #ZD flags of FPU status word.
@@ skipping 212 matching lines @@
            ? &maybe_undefined_first
            : on_not_smis);
   // Convert HeapNumber to smi if possible.
   __ movsd(xmm0, FieldOperand(first, HeapNumber::kValueOffset));
   __ movq(scratch2, xmm0);
   __ cvttsd2siq(smi_result, xmm0);
   // Check if conversion was successful by converting back and
   // comparing to the original double's bits.
   __ cvtlsi2sd(xmm1, smi_result);
   __ movq(kScratchRegister, xmm1);
-  __ cmpq(scratch2, kScratchRegister);
+  __k cmpq(scratch2, kScratchRegister);
   __ j(not_equal, on_not_smis);
+#ifdef V8_TARGET_ARCH_X32
+  __ cmpl(smi_result, Immediate(0xc0000000));
+  __ j(negative, on_not_smis);
+#endif
   __ Integer32ToSmi(first, smi_result);
 
   __ bind(&first_done);
   __ JumpIfSmi(second, (on_success != NULL) ? on_success : &done);
   __ bind(&first_smi);
   __ AssertNotSmi(second);
   __ cmpq(FieldOperand(second, HeapObject::kMapOffset), heap_number_map);
   __ j(not_equal,
        (convert_undefined == CONVERT_UNDEFINED_TO_ZERO)
            ? &maybe_undefined_second
            : on_not_smis);
   // Convert second to smi, if possible.
   __ movsd(xmm0, FieldOperand(second, HeapNumber::kValueOffset));
   __ movq(scratch2, xmm0);
   __ cvttsd2siq(smi_result, xmm0);
   __ cvtlsi2sd(xmm1, smi_result);
   __ movq(kScratchRegister, xmm1);
-  __ cmpq(scratch2, kScratchRegister);
+  __k cmpq(scratch2, kScratchRegister);
   __ j(not_equal, on_not_smis);
+#ifdef V8_TARGET_ARCH_X32
+  __ cmpl(smi_result, Immediate(0xc0000000));
+  __ j(negative, on_not_smis);
+#endif
   __ Integer32ToSmi(second, smi_result);
   if (on_success != NULL) {
     __ jmp(on_success);
   } else {
     __ jmp(&done);
   }
 
   __ bind(&maybe_undefined_first);
   __ CompareRoot(first, Heap::kUndefinedValueRootIndex);
   __ j(not_equal, on_not_smis);
@@ skipping 26 matching lines @@
 
   // Save 1 in double_result - we need this several times later on.
   __ movq(scratch, Immediate(1));
   __ cvtlsi2sd(double_result, scratch);
 
   if (exponent_type_ == ON_STACK) {
     Label base_is_smi, unpack_exponent;
     // The exponent and base are supplied as arguments on the stack.
     // This can only happen if the stub is called from non-optimized code.
     // Load input parameters from stack.
-    __ movq(base, Operand(rsp, 2 * kPointerSize));
-    __ movq(exponent, Operand(rsp, 1 * kPointerSize));
+    __a movq(base, Operand(rsp, 2 * kPointerSize));
+    __a movq(exponent, Operand(rsp, 1 * kPointerSize));
     __ JumpIfSmi(base, &base_is_smi, Label::kNear);
     __ CompareRoot(FieldOperand(base, HeapObject::kMapOffset),
                    Heap::kHeapNumberMapRootIndex);
     __ j(not_equal, &call_runtime);
 
     __ movsd(double_base, FieldOperand(base, HeapNumber::kValueOffset));
     __ jmp(&unpack_exponent, Label::kNear);
 
     __ bind(&base_is_smi);
     __ SmiToInteger32(base, base);
@@ skipping 322 matching lines @@
   // property might have been redefined.
   __ movq(scratch, FieldOperand(receiver, JSArray::kPropertiesOffset));
   __ CompareRoot(FieldOperand(scratch, FixedArray::kMapOffset),
                  Heap::kHashTableMapRootIndex);
   __ j(equal, &miss);
 
   // Check that value is a smi.
   __ JumpIfNotSmi(value, &miss);
 
   // Prepare tail call to StoreIC_ArrayLength.
-  __ pop(scratch);
+  __k pop(scratch);
   __ push(receiver);
   __ push(value);
-  __ push(scratch);  // return address
+  __k push(scratch);  // return address
 
   ExternalReference ref =
       ExternalReference(IC_Utility(IC::kStoreIC_ArrayLength), masm->isolate());
   __ TailCallExternalReference(ref, 2, 1);
 
   __ bind(&miss);
 
   StubCompiler::TailCallBuiltin(masm, StubCompiler::MissBuiltin(kind()));
 }
 
 
 void ArgumentsAccessStub::GenerateReadElement(MacroAssembler* masm) {
   // The key is in rdx and the parameter count is in rax.
 
   // The displacement is used for skipping the frame pointer on the
   // stack. It is the offset of the last parameter (if any) relative
   // to the frame pointer.
+#ifndef V8_TARGET_ARCH_X32
   static const int kDisplacement = 1 * kPointerSize;
+#else
+  static const int kDisplacement = 2 * kHWRegSize - 1 * kPointerSize;
+#endif
 
   // Check that the key is a smi.
   Label slow;
   __ JumpIfNotSmi(rdx, &slow);
 
   // Check if the calling frame is an arguments adaptor frame.  We look at the
   // context offset, and if the frame is not a regular one, then we find a
   // Smi instead of the context.  We can't use SmiCompare here, because that
   // only works for comparing two smis.
   Label adaptor;
@@ skipping 26 matching lines @@
   // Read the argument from the stack and return it.
   index = masm->SmiToIndex(rax, rcx, kPointerSizeLog2);
   __ lea(rbx, Operand(rbx, index.reg, index.scale, 0));
   index = masm->SmiToNegativeIndex(rdx, rdx, kPointerSizeLog2);
   __ movq(rax, Operand(rbx, index.reg, index.scale, kDisplacement));
   __ Ret();
 
   // Slow-case: Handle non-smi or out-of-bounds access to arguments
   // by calling the runtime system.
   __ bind(&slow);
-  __ pop(rbx);  // Return address.
+  __k pop(rbx);  // Return address.
   __ push(rdx);
-  __ push(rbx);
+  __k push(rbx);
   __ TailCallRuntime(Runtime::kGetArgumentsProperty, 1, 1);
 }
 
 
 void ArgumentsAccessStub::GenerateNewNonStrictFast(MacroAssembler* masm) {
   // Stack layout:
   //  rsp[0]  : return address
   //  rsp[8]  : number of parameters (tagged)
   //  rsp[16] : receiver displacement
   //  rsp[24] : function
   // Registers used over the whole function:
   //  rbx: the mapped parameter count (untagged)
   //  rax: the allocated object (tagged).
 
   Factory* factory = masm->isolate()->factory();
 
-  __ SmiToInteger64(rbx, Operand(rsp, 1 * kPointerSize));
+  __a SmiToInteger64(rbx, Operand(rsp, 1 * kPointerSize));
   // rbx = parameter count (untagged)
 
   // Check if the calling frame is an arguments adaptor frame.
   Label runtime;
   Label adaptor_frame, try_allocate;
   __ movq(rdx, Operand(rbp, StandardFrameConstants::kCallerFPOffset));
   __ movq(rcx, Operand(rdx, StandardFrameConstants::kContextOffset));
   __ Cmp(rcx, Smi::FromInt(StackFrame::ARGUMENTS_ADAPTOR));
   __ j(equal, &adaptor_frame);
 
   // No adaptor, parameter count = argument count.
   __ movq(rcx, rbx);
   __ jmp(&try_allocate, Label::kNear);
 
   // We have an adaptor frame. Patch the parameters pointer.
   __ bind(&adaptor_frame);
   __ SmiToInteger64(rcx,
                     Operand(rdx,
                             ArgumentsAdaptorFrameConstants::kLengthOffset));
   __ lea(rdx, Operand(rdx, rcx, times_pointer_size,
                       StandardFrameConstants::kCallerSPOffset));
-  __ movq(Operand(rsp, 2 * kPointerSize), rdx);
+  __a movq(Operand(rsp, 2 * kPointerSize), rdx);
 
   // rbx = parameter count (untagged)
   // rcx = argument count (untagged)
   // Compute the mapped parameter count = min(rbx, rcx) in rbx.
   __ cmpq(rbx, rcx);
   __ j(less_equal, &try_allocate, Label::kNear);
   __ movq(rbx, rcx);
 
   __ bind(&try_allocate);
 
@@ skipping 40 matching lines @@
   // rcx = argument count (untagged)
   // rdi = address of boilerplate object (tagged)
   // Copy the JS object part.
   for (int i = 0; i < JSObject::kHeaderSize; i += kPointerSize) {
     __ movq(rdx, FieldOperand(rdi, i));
     __ movq(FieldOperand(rax, i), rdx);
   }
 
   // Set up the callee in-object property.
   STATIC_ASSERT(Heap::kArgumentsCalleeIndex == 1);
-  __ movq(rdx, Operand(rsp, 3 * kPointerSize));
+  __a movq(rdx, Operand(rsp, 3 * kPointerSize));
   __ movq(FieldOperand(rax, JSObject::kHeaderSize +
                        Heap::kArgumentsCalleeIndex * kPointerSize),
           rdx);
 
   // Use the length (smi tagged) and set that as an in-object property too.
   // Note: rcx is tagged from here on.
   STATIC_ASSERT(Heap::kArgumentsLengthIndex == 0);
   __ Integer32ToSmi(rcx, rcx);
   __ movq(FieldOperand(rax, JSObject::kHeaderSize +
                        Heap::kArgumentsLengthIndex * kPointerSize),
@@ skipping 30 matching lines @@
   //   MIN_CONTEXT_SLOTS .. MIN_CONTEXT_SLOTS+parameter_count-1
   // The mapped parameter thus need to get indices
   //   MIN_CONTEXT_SLOTS+parameter_count-1 ..
   //       MIN_CONTEXT_SLOTS+parameter_count-mapped_parameter_count
   // We loop from right to left.
   Label parameters_loop, parameters_test;
 
   // Load tagged parameter count into r9.
   __ Integer32ToSmi(r9, rbx);
   __ Move(r8, Smi::FromInt(Context::MIN_CONTEXT_SLOTS));
-  __ addq(r8, Operand(rsp, 1 * kPointerSize));
+  __a addq(r8, Operand(rsp, 1 * kPointerSize));
   __ subq(r8, r9);
   __ Move(r11, factory->the_hole_value());
   __ movq(rdx, rdi);
   __ lea(rdi, Operand(rdi, rbx, times_pointer_size, kParameterMapHeaderSize));
   // r9 = loop variable (tagged)
   // r8 = mapping index (tagged)
   // r11 = the hole value
   // rdx = address of parameter map (tagged)
   // rdi = address of backing store (tagged)
   __ jmp(&parameters_test, Label::kNear);
@@ skipping 18 matching lines @@
 
   // rcx = argument count (tagged)
   // rdi = address of backing store (tagged)
   // Copy arguments header and remaining slots (if there are any).
   __ Move(FieldOperand(rdi, FixedArray::kMapOffset),
           factory->fixed_array_map());
   __ movq(FieldOperand(rdi, FixedArray::kLengthOffset), rcx);
 
   Label arguments_loop, arguments_test;
   __ movq(r8, rbx);
-  __ movq(rdx, Operand(rsp, 2 * kPointerSize));
+  __a movq(rdx, Operand(rsp, 2 * kPointerSize));
   // Untag rcx for the loop below.
   __ SmiToInteger64(rcx, rcx);
   __ lea(kScratchRegister, Operand(r8, times_pointer_size, 0));
   __ subq(rdx, kScratchRegister);
   __ jmp(&arguments_test, Label::kNear);
 
   __ bind(&arguments_loop);
   __ subq(rdx, Immediate(kPointerSize));
   __ movq(r9, Operand(rdx, 0));
   __ movq(FieldOperand(rdi, r8,
                        times_pointer_size,
                        FixedArray::kHeaderSize),
           r9);
   __ addq(r8, Immediate(1));
 
   __ bind(&arguments_test);
   __ cmpq(r8, rcx);
   __ j(less, &arguments_loop, Label::kNear);
 
   // Return and remove the on-stack parameters.
   __ ret(3 * kPointerSize);
 
   // Do the runtime call to allocate the arguments object.
   // rcx = argument count (untagged)
   __ bind(&runtime);
   __ Integer32ToSmi(rcx, rcx);
-  __ movq(Operand(rsp, 1 * kPointerSize), rcx);  // Patch argument count.
+  __a movq(Operand(rsp, 1 * kPointerSize), rcx);  // Patch argument count.
   __ TailCallRuntime(Runtime::kNewArgumentsFast, 3, 1);
 }
 
 
 void ArgumentsAccessStub::GenerateNewNonStrictSlow(MacroAssembler* masm) {
   // rsp[0]  : return address
   // rsp[8]  : number of parameters
   // rsp[16] : receiver displacement
   // rsp[24] : function
 
   // Check if the calling frame is an arguments adaptor frame.
   Label runtime;
   __ movq(rdx, Operand(rbp, StandardFrameConstants::kCallerFPOffset));
   __ movq(rcx, Operand(rdx, StandardFrameConstants::kContextOffset));
   __ Cmp(rcx, Smi::FromInt(StackFrame::ARGUMENTS_ADAPTOR));
   __ j(not_equal, &runtime);
 
   // Patch the arguments.length and the parameters pointer.
   __ movq(rcx, Operand(rdx, ArgumentsAdaptorFrameConstants::kLengthOffset));
-  __ movq(Operand(rsp, 1 * kPointerSize), rcx);
+  __a movq(Operand(rsp, 1 * kPointerSize), rcx);
   __ SmiToInteger64(rcx, rcx);
   __ lea(rdx, Operand(rdx, rcx, times_pointer_size,
               StandardFrameConstants::kCallerSPOffset));
-  __ movq(Operand(rsp, 2 * kPointerSize), rdx);
+  __a movq(Operand(rsp, 2 * kPointerSize), rdx);
 
   __ bind(&runtime);
   __ TailCallRuntime(Runtime::kNewArgumentsFast, 3, 1);
 }
 
 
 void ArgumentsAccessStub::GenerateNewStrict(MacroAssembler* masm) {
   // rsp[0]  : return address
   // rsp[8]  : number of parameters
   // rsp[16] : receiver displacement
   // rsp[24] : function
 
   // Check if the calling frame is an arguments adaptor frame.
   Label adaptor_frame, try_allocate, runtime;
   __ movq(rdx, Operand(rbp, StandardFrameConstants::kCallerFPOffset));
   __ movq(rcx, Operand(rdx, StandardFrameConstants::kContextOffset));
   __ Cmp(rcx, Smi::FromInt(StackFrame::ARGUMENTS_ADAPTOR));
   __ j(equal, &adaptor_frame);
 
   // Get the length from the frame.
-  __ movq(rcx, Operand(rsp, 1 * kPointerSize));
+  __a movq(rcx, Operand(rsp, 1 * kPointerSize));
   __ SmiToInteger64(rcx, rcx);
   __ jmp(&try_allocate);
 
   // Patch the arguments.length and the parameters pointer.
   __ bind(&adaptor_frame);
   __ movq(rcx, Operand(rdx, ArgumentsAdaptorFrameConstants::kLengthOffset));
-  __ movq(Operand(rsp, 1 * kPointerSize), rcx);
+  __a movq(Operand(rsp, 1 * kPointerSize), rcx);
   __ SmiToInteger64(rcx, rcx);
   __ lea(rdx, Operand(rdx, rcx, times_pointer_size,
                       StandardFrameConstants::kCallerSPOffset));
-  __ movq(Operand(rsp, 2 * kPointerSize), rdx);
+  __a movq(Operand(rsp, 2 * kPointerSize), rdx);
 
   // Try the new space allocation. Start out with computing the size of
   // the arguments object and the elements array.
   Label add_arguments_object;
   __ bind(&try_allocate);
   __ testq(rcx, rcx);
   __ j(zero, &add_arguments_object, Label::kNear);
   __ lea(rcx, Operand(rcx, times_pointer_size, FixedArray::kHeaderSize));
   __ bind(&add_arguments_object);
   __ addq(rcx, Immediate(Heap::kArgumentsObjectSizeStrict));
 
   // Do the allocation of both objects in one go.
   __ Allocate(rcx, rax, rdx, rbx, &runtime, TAG_OBJECT);
 
   // Get the arguments boilerplate from the current native context.
   __ movq(rdi, Operand(rsi, Context::SlotOffset(Context::GLOBAL_OBJECT_INDEX)));
   __ movq(rdi, FieldOperand(rdi, GlobalObject::kNativeContextOffset));
   const int offset =
       Context::SlotOffset(Context::STRICT_MODE_ARGUMENTS_BOILERPLATE_INDEX);
   __ movq(rdi, Operand(rdi, offset));
 
   // Copy the JS object part.
   for (int i = 0; i < JSObject::kHeaderSize; i += kPointerSize) {
     __ movq(rbx, FieldOperand(rdi, i));
     __ movq(FieldOperand(rax, i), rbx);
   }
 
   // Get the length (smi tagged) and set that as an in-object property too.
   STATIC_ASSERT(Heap::kArgumentsLengthIndex == 0);
-  __ movq(rcx, Operand(rsp, 1 * kPointerSize));
+  __a movq(rcx, Operand(rsp, 1 * kPointerSize));
   __ movq(FieldOperand(rax, JSObject::kHeaderSize +
                        Heap::kArgumentsLengthIndex * kPointerSize),
           rcx);
 
   // If there are no actual arguments, we're done.
   Label done;
   __ testq(rcx, rcx);
   __ j(zero, &done);
 
   // Get the parameters pointer from the stack.
-  __ movq(rdx, Operand(rsp, 2 * kPointerSize));
+  __a movq(rdx, Operand(rsp, 2 * kPointerSize));
 
   // Set up the elements pointer in the allocated arguments object and
   // initialize the header in the elements fixed array.
   __ lea(rdi, Operand(rax, Heap::kArgumentsObjectSizeStrict));
   __ movq(FieldOperand(rax, JSObject::kElementsOffset), rdi);
   __ LoadRoot(kScratchRegister, Heap::kFixedArrayMapRootIndex);
   __ movq(FieldOperand(rdi, FixedArray::kMapOffset), kScratchRegister);
 
 
   __ movq(FieldOperand(rdi, FixedArray::kLengthOffset), rcx);
@@ skipping 28 matching lines @@
   __ TailCallRuntime(Runtime::kRegExpExec, 4, 1);
 #else  // V8_INTERPRETED_REGEXP
 
   // Stack frame on entry.
   //  rsp[0]  : return address
   //  rsp[8]  : last_match_info (expected JSArray)
   //  rsp[16] : previous index
   //  rsp[24] : subject string
   //  rsp[32] : JSRegExp object
 
+#ifndef V8_TARGET_ARCH_X32
   static const int kLastMatchInfoOffset = 1 * kPointerSize;
   static const int kPreviousIndexOffset = 2 * kPointerSize;
   static const int kSubjectOffset = 3 * kPointerSize;
   static const int kJSRegExpOffset = 4 * kPointerSize;
+#else
+  static const int kLastMatchInfoOffset = 1 * kHWRegSize;
+  static const int kPreviousIndexOffset = 1 * kHWRegSize + 1 * kPointerSize;
+  static const int kSubjectOffset = 1 * kHWRegSize + 2 * kPointerSize;
+  static const int kJSRegExpOffset = 1 * kHWRegSize + 3 * kPointerSize;
+#endif
 
   Label runtime;
   // Ensure that a RegExp stack is allocated.
   Isolate* isolate = masm->isolate();
   ExternalReference address_of_regexp_stack_memory_address =
       ExternalReference::address_of_regexp_stack_memory_address(isolate);
   ExternalReference address_of_regexp_stack_memory_size =
       ExternalReference::address_of_regexp_stack_memory_size(isolate);
   __ Load(kScratchRegister, address_of_regexp_stack_memory_size);
   __ testq(kScratchRegister, kScratchRegister);
@@ skipping 151 matching lines @@
 
   // Isolates: note we add an additional parameter here (isolate pointer).
   static const int kRegExpExecuteArguments = 9;
   int argument_slots_on_stack =
       masm->ArgumentStackSlotsForCFunctionCall(kRegExpExecuteArguments);
   __ EnterApiExitFrame(argument_slots_on_stack);
 
   // Argument 9: Pass current isolate address.
   __ LoadAddress(kScratchRegister,
                  ExternalReference::isolate_address(masm->isolate()));
-  __ movq(Operand(rsp, (argument_slots_on_stack - 1) * kPointerSize),
+  __s movq(Operand(rsp, (argument_slots_on_stack - 1) * kPointerSize),
           kScratchRegister);
 
   // Argument 8: Indicate that this is a direct call from JavaScript.
-  __ movq(Operand(rsp, (argument_slots_on_stack - 2) * kPointerSize),
+  __s movq(Operand(rsp, (argument_slots_on_stack - 2) * kPointerSize),
           Immediate(1));
 
   // Argument 7: Start (high end) of backtracking stack memory area.
   __ movq(kScratchRegister, address_of_regexp_stack_memory_address);
   __ movq(r9, Operand(kScratchRegister, 0));
   __ movq(kScratchRegister, address_of_regexp_stack_memory_size);
   __ addq(r9, Operand(kScratchRegister, 0));
-  __ movq(Operand(rsp, (argument_slots_on_stack - 3) * kPointerSize), r9);
+  __s movq(Operand(rsp, (argument_slots_on_stack - 3) * kPointerSize), r9);
 
   // Argument 6: Set the number of capture registers to zero to force global
   // regexps to behave as non-global.  This does not affect non-global regexps.
   // Argument 6 is passed in r9 on Linux and on the stack on Windows.
 #ifdef _WIN64
-  __ movq(Operand(rsp, (argument_slots_on_stack - 4) * kPointerSize),
+  __s movq(Operand(rsp, (argument_slots_on_stack - 4) * kPointerSize),
           Immediate(0));
 #else
   __ Set(r9, 0);
 #endif
 
   // Argument 5: static offsets vector buffer.
   __ LoadAddress(r8,
                  ExternalReference::address_of_static_offsets_vector(isolate));
   // Argument 5 passed in r8 on Linux and on the stack on Windows.
 #ifdef _WIN64
-  __ movq(Operand(rsp, (argument_slots_on_stack - 5) * kPointerSize), r8);
+  __s movq(Operand(rsp, (argument_slots_on_stack - 5) * kPointerSize), r8);
 #endif
 
   // rdi: subject string
   // rbx: previous index
   // rcx: encoding of subject string (1 if ASCII 0 if two_byte);
   // r11: code
   // r14: slice offset
   // r15: original subject string
 
   // Argument 2: Previous index.
@@ skipping 209 matching lines @@
   __ movq(rdi, FieldOperand(rdi, SlicedString::kParentOffset));
   __ jmp(&check_underlying);
 #endif  // V8_INTERPRETED_REGEXP
 }
 
 
 void RegExpConstructResultStub::Generate(MacroAssembler* masm) {
   const int kMaxInlineLength = 100;
   Label slowcase;
   Label done;
-  __ movq(r8, Operand(rsp, kPointerSize * 3));
+  __a movq(r8, Operand(rsp, 3 * kPointerSize));
   __ JumpIfNotSmi(r8, &slowcase);
   __ SmiToInteger32(rbx, r8);
   __ cmpl(rbx, Immediate(kMaxInlineLength));
   __ j(above, &slowcase);
   // Smi-tagging is equivalent to multiplying by 2.
   STATIC_ASSERT(kSmiTag == 0);
   STATIC_ASSERT(kSmiTagSize == 1);
   // Allocate RegExpResult followed by FixedArray with size in rbx.
   // JSArray:   [Map][empty properties][Elements][Length-smi][index][input]
   // Elements:  [Map][Length][..elements..]
@@ skipping 17 matching lines @@
 
   // Set empty properties FixedArray.
   __ LoadRoot(kScratchRegister, Heap::kEmptyFixedArrayRootIndex);
   __ movq(FieldOperand(rax, JSObject::kPropertiesOffset), kScratchRegister);
 
   // Set elements to point to FixedArray allocated right after the JSArray.
   __ lea(rcx, Operand(rax, JSRegExpResult::kSize));
   __ movq(FieldOperand(rax, JSObject::kElementsOffset), rcx);
 
   // Set input, index and length fields from arguments.
-  __ movq(r8, Operand(rsp, kPointerSize * 1));
+  __a movq(r8, Operand(rsp, 1 * kPointerSize));
   __ movq(FieldOperand(rax, JSRegExpResult::kInputOffset), r8);
-  __ movq(r8, Operand(rsp, kPointerSize * 2));
+  __a movq(r8, Operand(rsp, 2 * kPointerSize));
   __ movq(FieldOperand(rax, JSRegExpResult::kIndexOffset), r8);
-  __ movq(r8, Operand(rsp, kPointerSize * 3));
+  __a movq(r8, Operand(rsp, 3 * kPointerSize));
   __ movq(FieldOperand(rax, JSArray::kLengthOffset), r8);
 
   // Fill out the elements FixedArray.
   // rax: JSArray.
   // rcx: FixedArray.
   // rbx: Number of elements in array as int32.
 
   // Set map.
   __ LoadRoot(kScratchRegister, Heap::kFixedArrayMapRootIndex);
   __ movq(FieldOperand(rcx, HeapObject::kMapOffset), kScratchRegister);
@@ skipping 114 matching lines @@
   // but times_twice_pointer_size (multiplication by 16) scale factor
   // is not supported by addrmode on x64 platform.
   // So we have to premultiply entry index before lookup.
   __ shl(hash, Immediate(kPointerSizeLog2 + 1));
 }
 
 
 void NumberToStringStub::Generate(MacroAssembler* masm) {
   Label runtime;
 
-  __ movq(rbx, Operand(rsp, kPointerSize));
+  __a movq(rbx, Operand(rsp, 1 * kPointerSize));
 
   // Generate code to lookup number in the number string cache.
   GenerateLookupNumberStringCache(masm, rbx, rax, r8, r9, false, &runtime);
   __ ret(1 * kPointerSize);
 
   __ bind(&runtime);
   // Handle number to string in the runtime system if not found in the cache.
   __ TailCallRuntime(Runtime::kNumberToStringSkipCache, 1, 1);
 }
 
@@ skipping 270 matching lines @@
     // undefined, and are equal.
     __ Set(rax, EQUAL);
     __ bind(&return_unequal);
     // Return non-equal by returning the non-zero object pointer in rax,
     // or return equal if we fell through to here.
     __ ret(0);
     __ bind(&not_both_objects);
   }
 
   // Push arguments below the return address to prepare jump to builtin.
-  __ pop(rcx);
+  __k pop(rcx);
   __ push(rdx);
   __ push(rax);
 
   // Figure out which native to call and setup the arguments.
   Builtins::JavaScript builtin;
   if (cc == equal) {
     builtin = strict() ? Builtins::STRICT_EQUALS : Builtins::EQUALS;
   } else {
     builtin = Builtins::COMPARE;
     __ Push(Smi::FromInt(NegativeComparisonResult(cc)));
   }
 
   // Restore return address on the stack.
-  __ push(rcx);
+  __k push(rcx);
 
   // Call the native; it returns -1 (less), 0 (equal), or 1 (greater)
   // tagged as a small integer.
   __ InvokeBuiltin(builtin, JUMP_FUNCTION);
 
   __ bind(&miss);
   GenerateMiss(masm);
 }
 
 
@@ skipping 89 matching lines @@
   Isolate* isolate = masm->isolate();
   Label slow, non_function;
 
   // The receiver might implicitly be the global object. This is
   // indicated by passing the hole as the receiver to the call
   // function stub.
   if (ReceiverMightBeImplicit()) {
     Label call;
     // Get the receiver from the stack.
     // +1 ~ return address
-    __ movq(rax, Operand(rsp, (argc_ + 1) * kPointerSize));
+    __a movq(rax, Operand(rsp, (argc_ + 1) * kPointerSize));
     // Call as function is indicated with the hole.
     __ CompareRoot(rax, Heap::kTheHoleValueRootIndex);
     __ j(not_equal, &call, Label::kNear);
     // Patch the receiver on the stack with the global receiver object.
     __ movq(rcx, GlobalObjectOperand());
     __ movq(rcx, FieldOperand(rcx, GlobalObject::kGlobalReceiverOffset));
-    __ movq(Operand(rsp, (argc_ + 1) * kPointerSize), rcx);
+    __a movq(Operand(rsp, (argc_ + 1) * kPointerSize), rcx);
     __ bind(&call);
   }
 
   // Check that the function really is a JavaScript function.
   __ JumpIfSmi(rdi, &non_function);
   // Goto slow case if we do not have a function.
   __ CmpObjectType(rdi, JS_FUNCTION_TYPE, rcx);
   __ j(not_equal, &slow);
 
   if (RecordCallTarget()) {
@@ skipping 25 matching lines @@
   if (RecordCallTarget()) {
     // If there is a call target cache, mark it megamorphic in the
     // non-function case.  MegamorphicSentinel is an immortal immovable
     // object (undefined) so no write barrier is needed.
     __ Move(FieldOperand(rbx, Cell::kValueOffset),
             TypeFeedbackCells::MegamorphicSentinel(isolate));
   }
   // Check for function proxy.
   __ CmpInstanceType(rcx, JS_FUNCTION_PROXY_TYPE);
   __ j(not_equal, &non_function);
-  __ pop(rcx);
+  __k pop(rcx);
   __ push(rdi);  // put proxy as additional argument under return address
-  __ push(rcx);
+  __k push(rcx);
   __ Set(rax, argc_ + 1);
   __ Set(rbx, 0);
   __ SetCallKind(rcx, CALL_AS_METHOD);
   __ GetBuiltinEntry(rdx, Builtins::CALL_FUNCTION_PROXY);
   {
     Handle<Code> adaptor =
       masm->isolate()->builtins()->ArgumentsAdaptorTrampoline();
     __ jmp(adaptor, RelocInfo::CODE_TARGET);
   }
 
   // CALL_NON_FUNCTION expects the non-function callee as receiver (instead
   // of the original receiver from the call site).
   __ bind(&non_function);
-  __ movq(Operand(rsp, (argc_ + 1) * kPointerSize), rdi);
+  __a movq(Operand(rsp, (argc_ + 1) * kPointerSize), rdi);
   __ Set(rax, argc_);
   __ Set(rbx, 0);
   __ SetCallKind(rcx, CALL_AS_METHOD);
   __ GetBuiltinEntry(rdx, Builtins::CALL_NON_FUNCTION);
   Handle<Code> adaptor =
       Isolate::Current()->builtins()->ArgumentsAdaptorTrampoline();
   __ Jump(adaptor, RelocInfo::CODE_TARGET);
 }
 
 
@@ skipping 171 matching lines @@
   // Check for failure result.
   Label failure_returned;
   STATIC_ASSERT(((kFailureTag + 1) & kFailureTagMask) == 0);
 #ifdef _WIN64
   // If return value is on the stack, pop it to registers.
   if (result_size_ > 1) {
     ASSERT_EQ(2, result_size_);
     // Read result values stored on stack. Result is stored
     // above the four argument mirror slots and the two
     // Arguments object slots.
-    __ movq(rax, Operand(rsp, 6 * kPointerSize));
-    __ movq(rdx, Operand(rsp, 7 * kPointerSize));
+    __s movq(rax, Operand(rsp, 6 * kPointerSize));
+    __s movq(rdx, Operand(rsp, 7 * kPointerSize));
   }
 #endif
   __ lea(rcx, Operand(rax, 1));
   // Lower 2 bits of rcx are 0 iff rax has failure tag.
   __ testl(rcx, Immediate(kFailureTagMask));
   __ j(zero, &failure_returned);
 
   // Exit the JavaScript to C++ exit frame.
   __ LeaveExitFrame(save_doubles_);
   __ ret(0);
@@ skipping 88 matching lines @@
   // Do space-specific GC and retry runtime call.
   GenerateCore(masm,
                &throw_normal_exception,
                &throw_termination_exception,
                &throw_out_of_memory_exception,
                true,
                false);
 
   // Do full GC and retry runtime call one final time.
   Failure* failure = Failure::InternalError();
-  __ movq(rax, failure, RelocInfo::NONE64);
+  __n movq(rax, failure, RelocInfo::NONE64);
   GenerateCore(masm,
                &throw_normal_exception,
                &throw_termination_exception,
                &throw_out_of_memory_exception,
                true,
                true);
 
   __ bind(&throw_out_of_memory_exception);
   // Set external caught exception to false.
   Isolate* isolate = masm->isolate();
   ExternalReference external_caught(Isolate::kExternalCaughtExceptionAddress,
                                     isolate);
   __ Set(rax, static_cast<int64_t>(false));
   __ Store(external_caught, rax);
 
   // Set pending exception and rax to out of memory exception.
   ExternalReference pending_exception(Isolate::kPendingExceptionAddress,
                                       isolate);
   Label already_have_failure;
   JumpIfOOM(masm, rax, kScratchRegister, &already_have_failure);
-  __ movq(rax, Failure::OutOfMemoryException(0x1), RelocInfo::NONE64);
+  __n movq(rax, Failure::OutOfMemoryException(0x1), RelocInfo::NONE64);
   __ bind(&already_have_failure);
   __ Store(pending_exception, rax);
   // Fall through to the next label.
 
   __ bind(&throw_termination_exception);
   __ ThrowUncatchable(rax);
 
   __ bind(&throw_normal_exception);
   __ Throw(rax);
 }
 
 
 void JSEntryStub::GenerateBody(MacroAssembler* masm, bool is_construct) {
   Label invoke, handler_entry, exit;
   Label not_outermost_js, not_outermost_js_2;
 
   ProfileEntryHookStub::MaybeCallEntryHook(masm);
 
   {  // NOLINT. Scope block confuses linter.
     MacroAssembler::NoRootArrayScope uninitialized_root_register(masm);
     // Set up frame.
     __ push(rbp);
     __ movq(rbp, rsp);
 
     // Push the stack frame type marker twice.
     int marker = is_construct ? StackFrame::ENTRY_CONSTRUCT : StackFrame::ENTRY;
     // Scratch register is neither callee-save, nor an argument register on any
     // platform. It's free to use at this point.
     // Cannot use smi-register for loading yet.
+#ifndef V8_TARGET_ARCH_X32
     __ movq(kScratchRegister,
             reinterpret_cast<uint64_t>(Smi::FromInt(marker)),
             RelocInfo::NONE64);
+#else
+    __ movl(kScratchRegister,
+            reinterpret_cast<uint32_t>(Smi::FromInt(marker)),
+            RelocInfo::NONE32);
+#endif
     __ push(kScratchRegister);  // context slot
     __ push(kScratchRegister);  // function slot
     // Save callee-saved registers (X64/Win64 calling conventions).
-    __ push(r12);
-    __ push(r13);
-    __ push(r14);
-    __ push(r15);
+    __k push(r12);
+    __k push(r13);
+    __k push(r14);
+    __k push(r15);
 #ifdef _WIN64
-    __ push(rdi);  // Only callee save in Win64 ABI, argument in AMD64 ABI.
-    __ push(rsi);  // Only callee save in Win64 ABI, argument in AMD64 ABI.
+    __k push(rdi);  // Only callee save in Win64 ABI, argument in AMD64 ABI.
+    __k push(rsi);  // Only callee save in Win64 ABI, argument in AMD64 ABI.
 #endif
-    __ push(rbx);
+    __k push(rbx);
 
 #ifdef _WIN64
     // On Win64 XMM6-XMM15 are callee-save
     __ subq(rsp, Immediate(EntryFrameConstants::kXMMRegistersBlockSize));
     __ movdqu(Operand(rsp, EntryFrameConstants::kXMMRegisterSize * 0), xmm6);
     __ movdqu(Operand(rsp, EntryFrameConstants::kXMMRegisterSize * 1), xmm7);
     __ movdqu(Operand(rsp, EntryFrameConstants::kXMMRegisterSize * 2), xmm8);
     __ movdqu(Operand(rsp, EntryFrameConstants::kXMMRegisterSize * 3), xmm9);
     __ movdqu(Operand(rsp, EntryFrameConstants::kXMMRegisterSize * 4), xmm10);
     __ movdqu(Operand(rsp, EntryFrameConstants::kXMMRegisterSize * 5), xmm11);
@@ skipping 35 matching lines @@
   // Jump to a faked try block that does the invoke, with a faked catch
   // block that sets the pending exception.
   __ jmp(&invoke);
   __ bind(&handler_entry);
   handler_offset_ = handler_entry.pos();
   // Caught exception: Store result (exception) in the pending exception
   // field in the JSEnv and return a failure sentinel.
   ExternalReference pending_exception(Isolate::kPendingExceptionAddress,
                                       isolate);
   __ Store(pending_exception, rax);
-  __ movq(rax, Failure::Exception(), RelocInfo::NONE64);
+  __n movq(rax, Failure::Exception(), RelocInfo::NONE64);
   __ jmp(&exit);
 
   // Invoke: Link this frame into the handler chain.  There's only one
   // handler block in this code object, so its index is 0.
   __ bind(&invoke);
   __ PushTryHandler(StackHandler::JS_ENTRY, 0);
 
   // Clear any pending exceptions.
   __ LoadRoot(rax, Heap::kTheHoleValueRootIndex);
   __ Store(pending_exception, rax);
@@ skipping 43 matching lines @@
   __ movdqu(xmm9, Operand(rsp, EntryFrameConstants::kXMMRegisterSize * 3));
   __ movdqu(xmm10, Operand(rsp, EntryFrameConstants::kXMMRegisterSize * 4));
   __ movdqu(xmm11, Operand(rsp, EntryFrameConstants::kXMMRegisterSize * 5));
   __ movdqu(xmm12, Operand(rsp, EntryFrameConstants::kXMMRegisterSize * 6));
   __ movdqu(xmm13, Operand(rsp, EntryFrameConstants::kXMMRegisterSize * 7));
   __ movdqu(xmm14, Operand(rsp, EntryFrameConstants::kXMMRegisterSize * 8));
   __ movdqu(xmm15, Operand(rsp, EntryFrameConstants::kXMMRegisterSize * 9));
   __ addq(rsp, Immediate(EntryFrameConstants::kXMMRegistersBlockSize));
 #endif
 
-  __ pop(rbx);
+  __k pop(rbx);
 #ifdef _WIN64
   // Callee save on in Win64 ABI, arguments/volatile in AMD64 ABI.
-  __ pop(rsi);
-  __ pop(rdi);
+  __k pop(rsi);
+  __k pop(rdi);
 #endif
-  __ pop(r15);
-  __ pop(r14);
-  __ pop(r13);
-  __ pop(r12);
+  __k pop(r15);
+  __k pop(r14);
+  __k pop(r13);
+  __k pop(r12);
   __ addq(rsp, Immediate(2 * kPointerSize));  // remove markers
 
   // Restore frame pointer and return.
   __ pop(rbp);
   __ ret(0);
 }
 
 
 void InstanceofStub::Generate(MacroAssembler* masm) {
   // Implements "value instanceof function" operator.
   // Expected input state with no inline cache:
   //   rsp[0]  : return address
   //   rsp[8]  : function pointer
   //   rsp[16] : value
   // Expected input state with an inline one-element cache:
   //   rsp[0]  : return address
   //   rsp[8]  : offset from return address to location of inline cache
   //   rsp[16] : function pointer
   //   rsp[24] : value
   // Returns a bitwise zero to indicate that the value
   // is and instance of the function and anything else to
   // indicate that the value is not an instance.
 
   static const int kOffsetToMapCheckValue = 2;
+#ifndef V8_TARGET_ARCH_X32
   static const int kOffsetToResultValue = 18;
   // The last 4 bytes of the instruction sequence
   //   movq(rdi, FieldOperand(rax, HeapObject::kMapOffset))
   //   Move(kScratchRegister, Factory::the_hole_value())
   // in front of the hole value address.
   static const unsigned int kWordBeforeMapCheckValue = 0xBA49FF78;
   // The last 4 bytes of the instruction sequence
   //   __ j(not_equal, &cache_miss);
   //   __ LoadRoot(ToRegister(instr->result()), Heap::kTheHoleValueRootIndex);
   // before the offset of the hole value in the root array.
   static const unsigned int kWordBeforeResultValue = 0x458B4909;
+#else
+  static const int kOffsetToResultValue = 14;
+  // The last 4 bytes of the instruction sequence
+  //   movl(rdi, FieldOperand(rax, HeapObject::kMapOffset))
+  //   Move(kScratchRegister, Factory::the_hole_value())
+  // in front of the hole value address.
+  static const unsigned int kWordBeforeMapCheckValue = 0xBA41FF78;
+  // The last 4 bytes of the instruction sequence
+  //   __ j(not_equal, &cache_miss);
+  //   __ LoadRoot(ToRegister(instr->result()), Heap::kTheHoleValueRootIndex);
+  // before the offset of the hole value in the root array.
+  static const unsigned int kWordBeforeResultValue = 0x458B4109;
+#endif
   // Only the inline check flag is supported on X64.
   ASSERT(flags_ == kNoFlags || HasCallSiteInlineCheck());
   int extra_stack_space = HasCallSiteInlineCheck() ? kPointerSize : 0;
 
   // Get the object - go slow case if it's a smi.
   Label slow;
 
+#ifndef V8_TARGET_ARCH_X32
   __ movq(rax, Operand(rsp, 2 * kPointerSize + extra_stack_space));
+#else
+  __ movl(rax, Operand(rsp, 1 * kHWRegSize + 1 * kPointerSize +
+                       extra_stack_space));
+#endif
   __ JumpIfSmi(rax, &slow);
 
   // Check that the left hand is a JS object. Leave its map in rax.
   __ CmpObjectType(rax, FIRST_SPEC_OBJECT_TYPE, rax);
   __ j(below, &slow);
   __ CmpInstanceType(rax, LAST_SPEC_OBJECT_TYPE);
   __ j(above, &slow);
 
   // Get the prototype of the function.
-  __ movq(rdx, Operand(rsp, 1 * kPointerSize + extra_stack_space));
+  __a movq(rdx, Operand(rsp, 1 * kPointerSize + extra_stack_space));
   // rdx is function, rax is map.
 
   // If there is a call site cache don't look in the global cache, but do the
   // real lookup and update the call site cache.
   if (!HasCallSiteInlineCheck()) {
     // Look up the function and the map in the instanceof cache.
     Label miss;
     __ CompareRoot(rdx, Heap::kInstanceofCacheFunctionRootIndex);
     __ j(not_equal, &miss, Label::kNear);
     __ CompareRoot(rax, Heap::kInstanceofCacheMapRootIndex);
@@ skipping 14 matching lines @@
 
   // Register mapping:
   //   rax is object map.
   //   rdx is function.
   //   rbx is function prototype.
   if (!HasCallSiteInlineCheck()) {
     __ StoreRoot(rdx, Heap::kInstanceofCacheFunctionRootIndex);
     __ StoreRoot(rax, Heap::kInstanceofCacheMapRootIndex);
   } else {
     // Get return address and delta to inlined map check.
-    __ movq(kScratchRegister, Operand(rsp, 0 * kPointerSize));
-    __ subq(kScratchRegister, Operand(rsp, 1 * kPointerSize));
+    __q movq(kScratchRegister, Operand(rsp, 0 * kPointerSize));
+    __a subq(kScratchRegister, Operand(rsp, 1 * kPointerSize));
     if (FLAG_debug_code) {
       __ movl(rdi, Immediate(kWordBeforeMapCheckValue));
       __ cmpl(Operand(kScratchRegister, kOffsetToMapCheckValue - 4), rdi);
       __ Assert(equal, "InstanceofStub unexpected call site cache (check).");
     }
     __ movq(kScratchRegister,
             Operand(kScratchRegister, kOffsetToMapCheckValue));
     __ movq(Operand(kScratchRegister, 0), rax);
   }
 
@@ skipping 19 matching lines @@
     // Store bitwise zero in the cache.  This is a Smi in GC terms.
     STATIC_ASSERT(kSmiTag == 0);
     __ StoreRoot(rax, Heap::kInstanceofCacheAnswerRootIndex);
   } else {
     // Store offset of true in the root array at the inline check site.
     int true_offset = 0x100 +
         (Heap::kTrueValueRootIndex << kPointerSizeLog2) - kRootRegisterBias;
     // Assert it is a 1-byte signed value.
     ASSERT(true_offset >= 0 && true_offset < 0x100);
     __ movl(rax, Immediate(true_offset));
-    __ movq(kScratchRegister, Operand(rsp, 0 * kPointerSize));
-    __ subq(kScratchRegister, Operand(rsp, 1 * kPointerSize));
+    __q movq(kScratchRegister, Operand(rsp, 0 * kPointerSize));
+    __a subq(kScratchRegister, Operand(rsp, 1 * kPointerSize));
     __ movb(Operand(kScratchRegister, kOffsetToResultValue), rax);
     if (FLAG_debug_code) {
       __ movl(rax, Immediate(kWordBeforeResultValue));
       __ cmpl(Operand(kScratchRegister, kOffsetToResultValue - 4), rax);
       __ Assert(equal, "InstanceofStub unexpected call site cache (mov).");
     }
     __ Set(rax, 0);
   }
   __ ret(2 * kPointerSize + extra_stack_space);
 
   __ bind(&is_not_instance);
   if (!HasCallSiteInlineCheck()) {
     // We have to store a non-zero value in the cache.
     __ StoreRoot(kScratchRegister, Heap::kInstanceofCacheAnswerRootIndex);
   } else {
     // Store offset of false in the root array at the inline check site.
     int false_offset = 0x100 +
         (Heap::kFalseValueRootIndex << kPointerSizeLog2) - kRootRegisterBias;
     // Assert it is a 1-byte signed value.
     ASSERT(false_offset >= 0 && false_offset < 0x100);
     __ movl(rax, Immediate(false_offset));
-    __ movq(kScratchRegister, Operand(rsp, 0 * kPointerSize));
-    __ subq(kScratchRegister, Operand(rsp, 1 * kPointerSize));
+    __q movq(kScratchRegister, Operand(rsp, 0 * kPointerSize));
+    __a subq(kScratchRegister, Operand(rsp, 1 * kPointerSize));
     __ movb(Operand(kScratchRegister, kOffsetToResultValue), rax);
     if (FLAG_debug_code) {
       __ movl(rax, Immediate(kWordBeforeResultValue));
       __ cmpl(Operand(kScratchRegister, kOffsetToResultValue - 4), rax);
       __ Assert(equal, "InstanceofStub unexpected call site cache (mov)");
     }
   }
   __ ret(2 * kPointerSize + extra_stack_space);
 
   // Slow-case: Go through the JavaScript implementation.
   __ bind(&slow);
   if (HasCallSiteInlineCheck()) {
     // Remove extra value from the stack.
-    __ pop(rcx);
+    __k pop(rcx);
     __ pop(rax);
-    __ push(rcx);
+    __k push(rcx);
   }
   __ InvokeBuiltin(Builtins::INSTANCE_OF, JUMP_FUNCTION);
 }
 
 
 // Passing arguments in registers is not supported.
 Register InstanceofStub::left() { return no_reg; }
 
 
 Register InstanceofStub::right() { return no_reg; }
@@ skipping 129 matching lines @@
 
   __ Abort("Unexpected fallthrough from CharFromCode slow case");
 }
 
 
 void StringAddStub::Generate(MacroAssembler* masm) {
   Label call_runtime, call_builtin;
   Builtins::JavaScript builtin_id = Builtins::ADD;
 
   // Load the two arguments.
-  __ movq(rax, Operand(rsp, 2 * kPointerSize));  // First argument (left).
-  __ movq(rdx, Operand(rsp, 1 * kPointerSize));  // Second argument (right).
+  // First argument (left).
+  __a movq(rax, Operand(rsp, 2 * kPointerSize));
+  // Second argument (right).
+  __a movq(rdx, Operand(rsp, 1 * kPointerSize));
 
   // Make sure that both arguments are strings if not known in advance.
   if ((flags_ & NO_STRING_ADD_FLAGS) != 0) {
     __ JumpIfSmi(rax, &call_runtime);
     __ CmpObjectType(rax, FIRST_NONSTRING_TYPE, r8);
     __ j(above_equal, &call_runtime);
 
     // First argument is a a string, test second.
     __ JumpIfSmi(rdx, &call_runtime);
     __ CmpObjectType(rdx, FIRST_NONSTRING_TYPE, r9);
     __ j(above_equal, &call_runtime);
   } else {
     // Here at least one of the arguments is definitely a string.
     // We convert the one that is not known to be a string.
     if ((flags_ & NO_STRING_CHECK_LEFT_IN_STUB) == 0) {
       ASSERT((flags_ & NO_STRING_CHECK_RIGHT_IN_STUB) != 0);
+#ifndef V8_TARGET_ARCH_X32
       GenerateConvertArgument(masm, 2 * kPointerSize, rax, rbx, rcx, rdi,
                               &call_builtin);
+#else
+      GenerateConvertArgument(masm, 1 * kHWRegSize + 1 * kPointerSize, rax,
+                              rbx, rcx, rdi, &call_builtin);
+#endif
       builtin_id = Builtins::STRING_ADD_RIGHT;
     } else if ((flags_ & NO_STRING_CHECK_RIGHT_IN_STUB) == 0) {
       ASSERT((flags_ & NO_STRING_CHECK_LEFT_IN_STUB) != 0);
+#ifndef V8_TARGET_ARCH_X32
       GenerateConvertArgument(masm, 1 * kPointerSize, rdx, rbx, rcx, rdi,
                               &call_builtin);
+#else
+      GenerateConvertArgument(masm, 1 * kHWRegSize, rdx, rbx, rcx, rdi,
+                              &call_builtin);
+#endif
       builtin_id = Builtins::STRING_ADD_LEFT;
     }
   }
 
   // Both arguments are strings.
   // rax: first string
   // rdx: second string
   // Check if either of the strings are empty. In that case return the other.
   Label second_not_zero_length, both_not_zero_length;
   __ movq(rcx, FieldOperand(rdx, String::kLengthOffset));
@@ skipping 26 matching lines @@
   // by the code above.
   if (flags_ != NO_STRING_ADD_FLAGS) {
     __ movq(r8, FieldOperand(rax, HeapObject::kMapOffset));
     __ movq(r9, FieldOperand(rdx, HeapObject::kMapOffset));
   }
   // Get the instance types of the two strings as they will be needed soon.
   __ movzxbl(r8, FieldOperand(r8, Map::kInstanceTypeOffset));
   __ movzxbl(r9, FieldOperand(r9, Map::kInstanceTypeOffset));
 
   // Look at the length of the result of adding the two strings.
+#ifndef V8_TARGET_ARCH_X32
   STATIC_ASSERT(String::kMaxLength <= Smi::kMaxValue / 2);
   __ SmiAdd(rbx, rbx, rcx);
+#else
+  __ SmiAdd(rbx, rbx, rcx, &call_runtime);
+#endif
   // Use the string table when adding two one character strings, as it
   // helps later optimizations to return an internalized string here.
   __ SmiCompare(rbx, Smi::FromInt(2));
   __ j(not_equal, &longer_than_two);
 
   // Check that both strings are non-external ASCII strings.
   __ JumpIfBothInstanceTypesAreNotSequentialAscii(r8, r9, rbx, rcx,
                                                   &call_runtime);
 
   // Get the two characters forming the sub string.
@@ skipping 228 matching lines @@
 
 
 void StringAddStub::GenerateRegisterArgsPush(MacroAssembler* masm) {
   __ push(rax);
   __ push(rdx);
 }
 
 
 void StringAddStub::GenerateRegisterArgsPop(MacroAssembler* masm,
                                             Register temp) {
-  __ pop(temp);
+  __k pop(temp);
   __ pop(rdx);
   __ pop(rax);
-  __ push(temp);
+  __k push(temp);
 }
 
 
 void StringAddStub::GenerateConvertArgument(MacroAssembler* masm,
                                             int stack_offset,
                                             Register arg,
                                             Register scratch1,
                                             Register scratch2,
                                             Register scratch3,
                                             Label* slow) {
@@ skipping 82 matching lines @@
     __ addl(count, count);
   }
 
   // Don't enter the rep movs if there are less than 4 bytes to copy.
   Label last_bytes;
   __ testl(count, Immediate(~(kPointerSize - 1)));
   __ j(zero, &last_bytes, Label::kNear);
 
   // Copy from edi to esi using rep movs instruction.
   __ movl(kScratchRegister, count);
-  __ shr(count, Immediate(kPointerSizeLog2));  // Number of doublewords to copy.
+  // Number of doublewords to copy.
+  __ shr(count, Immediate(kPointerSizeLog2));
   __ repmovsq();
 
   // Find number of bytes left.
   __ movl(count, kScratchRegister);
   __ and_(count, Immediate(kPointerSize - 1));
 
   // Check if there are more bytes to copy.
   __ bind(&last_bytes);
   __ testl(count, count);
   __ j(zero, &done, Label::kNear);
@@ skipping 199 matching lines @@
 
 void SubStringStub::Generate(MacroAssembler* masm) {
   Label runtime;
 
   // Stack frame on entry.
   //  rsp[0]  : return address
   //  rsp[8]  : to
   //  rsp[16] : from
   //  rsp[24] : string
 
+#ifndef V8_TARGET_ARCH_X32
   const int kToOffset = 1 * kPointerSize;
+#else
+  const int kToOffset = 1 * kHWRegSize;
+#endif
   const int kFromOffset = kToOffset + kPointerSize;
   const int kStringOffset = kFromOffset + kPointerSize;
   const int kArgumentsSize = (kStringOffset + kPointerSize) - kToOffset;
 
   // Make sure first argument is a string.
   __ movq(rax, Operand(rsp, kStringOffset));
   STATIC_ASSERT(kSmiTag == 0);
   __ testl(rax, Immediate(kSmiTagMask));
   __ j(zero, &runtime);
   Condition is_string = masm->IsObjectStringType(rax, rbx, rbx);
@@ skipping 318 matching lines @@
     Label* chars_not_equal,
     Label::Distance near_jump) {
   // Change index to run from -length to -1 by adding length to string
   // start. This means that loop ends when index reaches zero, which
   // doesn't need an additional compare.
   __ SmiToInteger32(length, length);
   __ lea(left,
          FieldOperand(left, length, times_1, SeqOneByteString::kHeaderSize));
   __ lea(right,
          FieldOperand(right, length, times_1, SeqOneByteString::kHeaderSize));
-  __ neg(length);
+  __k neg(length);
   Register index = length;  // index = -length;
 
   // Compare loop.
   Label loop;
   __ bind(&loop);
   __ movb(scratch, Operand(left, index, times_1, 0));
   __ cmpb(scratch, Operand(right, index, times_1, 0));
   __ j(not_equal, chars_not_equal, near_jump);
-  __ incq(index);
+  __k incq(index);
   __ j(not_zero, &loop);
 }
 
 
 void StringCompareStub::Generate(MacroAssembler* masm) {
   Label runtime;
 
   // Stack frame on entry.
   //  rsp[0]  : return address
   //  rsp[8]  : right string
   //  rsp[16] : left string
 
-  __ movq(rdx, Operand(rsp, 2 * kPointerSize));  // left
-  __ movq(rax, Operand(rsp, 1 * kPointerSize));  // right
+  __a movq(rdx, Operand(rsp, 2 * kPointerSize));  // left
+  __a movq(rax, Operand(rsp, 1 * kPointerSize));  // right
 
   // Check for identity.
   Label not_same;
   __ cmpq(rdx, rax);
   __ j(not_equal, &not_same, Label::kNear);
   __ Move(rax, Smi::FromInt(EQUAL));
   Counters* counters = masm->isolate()->counters();
   __ IncrementCounter(counters->string_compare_native(), 1);
   __ ret(2 * kPointerSize);
 
   __ bind(&not_same);
 
   // Check that both are sequential ASCII strings.
   __ JumpIfNotBothSequentialAsciiStrings(rdx, rax, rcx, rbx, &runtime);
 
   // Inline comparison of ASCII strings.
   __ IncrementCounter(counters->string_compare_native(), 1);
   // Drop arguments from the stack
-  __ pop(rcx);
+  __k pop(rcx);
   __ addq(rsp, Immediate(2 * kPointerSize));
-  __ push(rcx);
+  __k push(rcx);
   GenerateCompareFlatAsciiStrings(masm, rdx, rax, rcx, rbx, rdi, r8);
 
   // Call the runtime; it returns -1 (less), 0 (equal), or 1 (greater)
   // tagged as a small integer.
   __ bind(&runtime);
   __ TailCallRuntime(Runtime::kStringCompare, 2, 1);
 }
 
 
 void ICCompareStub::GenerateSmis(MacroAssembler* masm) {
@@ skipping 253 matching lines @@
   if (equality) {
     StringCompareStub::GenerateFlatAsciiStringEquals(
         masm, left, right, tmp1, tmp2);
   } else {
     StringCompareStub::GenerateCompareFlatAsciiStrings(
         masm, left, right, tmp1, tmp2, tmp3, kScratchRegister);
   }
 
   // Handle more complex cases in runtime.
   __ bind(&runtime);
-  __ pop(tmp1);  // Return address.
+  __k pop(tmp1);  // Return address.
   __ push(left);
   __ push(right);
-  __ push(tmp1);
+  __k push(tmp1);
   if (equality) {
     __ TailCallRuntime(Runtime::kStringEquals, 2, 1);
   } else {
     __ TailCallRuntime(Runtime::kStringCompare, 2, 1);
   }
 
   __ bind(&miss);
   GenerateMiss(masm);
 }
 
@@ skipping 201 matching lines @@
   __ decl(scratch);
   __ push(scratch);
 
   // If names of slots in range from 1 to kProbes - 1 for the hash value are
   // not equal to the name and kProbes-th slot is not used (its name is the
   // undefined value), it guarantees the hash table doesn't contain the
   // property. It's true even if some slots represent deleted properties
   // (their names are the null value).
   for (int i = kInlinedProbes; i < kTotalProbes; i++) {
     // Compute the masked index: (hash + i + i * i) & mask.
-    __ movq(scratch, Operand(rsp, 2 * kPointerSize));
+    __a movq(scratch, Operand(rsp, 2 * kPointerSize));
     if (i > 0) {
       __ addl(scratch, Immediate(NameDictionary::GetProbeOffset(i)));
     }
     __ and_(scratch, Operand(rsp, 0));
 
     // Scale the index by multiplying by the entry size.
     ASSERT(NameDictionary::kEntrySize == 3);
     __ lea(index_, Operand(scratch, scratch, times_2, 0));  // index *= 3.
 
     // Having undefined at this place means the name is not contained.
     __ movq(scratch, Operand(dictionary_,
                              index_,
                              times_pointer_size,
                              kElementsStartOffset - kHeapObjectTag));
 
     __ Cmp(scratch, masm->isolate()->factory()->undefined_value());
     __ j(equal, &not_in_dictionary);
 
     // Stop if found the property.
-    __ cmpq(scratch, Operand(rsp, 3 * kPointerSize));
+    __a cmpq(scratch, Operand(rsp, 3 * kPointerSize));
     __ j(equal, &in_dictionary);
 
     if (i != kTotalProbes - 1 && mode_ == NEGATIVE_LOOKUP) {
       // If we hit a key that is not a unique name during negative
       // lookup we have to bailout as this key might be equal to the
       // key we are looking for.
 
       // Check if the entry name is not a unique name.
       __ movq(scratch, FieldOperand(scratch, HeapObject::kMapOffset));
       __ JumpIfNotUniqueName(FieldOperand(scratch, Map::kInstanceTypeOffset),
@@ skipping 331 matching lines @@
   // clobbers rbx, rdx, rdi
   // -----------------------------------
 
   Label element_done;
   Label double_elements;
   Label smi_element;
   Label slow_elements;
   Label fast_elements;
 
   // Get array literal index, array literal and its map.
-  __ movq(rdx, Operand(rsp, 1 * kPointerSize));
-  __ movq(rbx, Operand(rsp, 2 * kPointerSize));
+  __a movq(rdx, Operand(rsp, 1 * kPointerSize));
+  __a movq(rbx, Operand(rsp, 2 * kPointerSize));
   __ movq(rdi, FieldOperand(rbx, JSObject::kMapOffset));
 
   __ CheckFastElements(rdi, &double_elements);
 
   // FAST_*_SMI_ELEMENTS or FAST_*_ELEMENTS
   __ JumpIfSmi(rax, &smi_element);
   __ CheckFastSmiElements(rdi, &fast_elements);
 
   // Store into the array literal requires a elements transition. Call into
   // the runtime.
 
   __ bind(&slow_elements);
-  __ pop(rdi);  // Pop return address and remember to put back later for tail
+  __k pop(rdi);  // Pop return address and remember to put back later for tail
                 // call.
   __ push(rbx);
   __ push(rcx);
   __ push(rax);
   __ movq(rbx, Operand(rbp, JavaScriptFrameConstants::kFunctionOffset));
   __ push(FieldOperand(rbx, JSFunction::kLiteralsOffset));
   __ push(rdx);
-  __ push(rdi);  // Return return address so that tail call returns to right
+  __k push(rdi);  // Return return address so that tail call returns to right
                  // place.
   __ TailCallRuntime(Runtime::kStoreArrayLiteralElement, 5, 1);
 
   // Array literal has ElementsKind of FAST_*_ELEMENTS and value is an object.
   __ bind(&fast_elements);
   __ SmiToInteger32(kScratchRegister, rcx);
   __ movq(rbx, FieldOperand(rbx, JSObject::kElementsOffset));
   __ lea(rcx, FieldOperand(rbx, kScratchRegister, times_pointer_size,
                            FixedArrayBase::kHeaderSize));
   __ movq(Operand(rcx, 0), rax);
@@ skipping 27 matching lines @@
 }
 
 
 void StubFailureTrampolineStub::Generate(MacroAssembler* masm) {
   CEntryStub ces(1, fp_registers_ ? kSaveFPRegs : kDontSaveFPRegs);
   __ Call(ces.GetCode(masm->isolate()), RelocInfo::CODE_TARGET);
   int parameter_count_offset =
       StubFailureTrampolineFrame::kCallerStackParameterCountFrameOffset;
   __ movq(rbx, MemOperand(rbp, parameter_count_offset));
   masm->LeaveFrame(StackFrame::STUB_FAILURE_TRAMPOLINE);
-  __ pop(rcx);
+  __k pop(rcx);
   int additional_offset = function_mode_ == JS_FUNCTION_STUB_MODE
       ? kPointerSize
       : 0;
   __ lea(rsp, MemOperand(rsp, rbx, times_pointer_size, additional_offset));
   __ jmp(rcx);  // Return to IC Miss stub, continuation still on stack.
 }
 
 
 void ProfileEntryHookStub::MaybeCallEntryHook(MacroAssembler* masm) {
   if (masm->isolate()->function_entry_hook() != NULL) {
     // It's always safe to call the entry hook stub, as the hook itself
     // is not allowed to call back to V8.
     AllowStubCallsScope allow_stub_calls(masm, true);
 
     ProfileEntryHookStub stub;
     masm->CallStub(&stub);
   }
 }
 
 
 void ProfileEntryHookStub::Generate(MacroAssembler* masm) {
   // This stub can be called from essentially anywhere, so it needs to save
   // all volatile and callee-save registers.
   const size_t kNumSavedRegisters = 2;
-  __ push(arg_reg_1);
-  __ push(arg_reg_2);
+  __k push(arg_reg_1);
+  __k push(arg_reg_2);
 
   // Calculate the original stack pointer and store it in the second arg.
-  __ lea(arg_reg_2, Operand(rsp, (kNumSavedRegisters + 1) * kPointerSize));
+  __q lea(arg_reg_2, Operand(rsp, (kNumSavedRegisters + 1) * kPointerSize));
 
   // Calculate the function address to the first arg.
-  __ movq(arg_reg_1, Operand(rsp, kNumSavedRegisters * kPointerSize));
+  __s movq(arg_reg_1, Operand(rsp, kNumSavedRegisters * kPointerSize));
   __ subq(arg_reg_1, Immediate(Assembler::kShortCallInstructionLength));
 
   // Save the remainder of the volatile registers.
   masm->PushCallerSaved(kSaveFPRegs, arg_reg_1, arg_reg_2);
 
   // Call the entry hook function.
   __ movq(rax, FUNCTION_ADDR(masm->isolate()->function_entry_hook()),
+#ifndef V8_TARGET_ARCH_X32
           RelocInfo::NONE64);
+#else
+          RelocInfo::NONE32);
+#endif
 
   AllowExternalCallThatCantCauseGC scope(masm);
 
   const int kArgumentCount = 2;
   __ PrepareCallCFunction(kArgumentCount);
   __ CallCFunction(rax, kArgumentCount);
 
   // Restore volatile regs.
   masm->PopCallerSaved(kSaveFPRegs, arg_reg_1, arg_reg_2);
-  __ pop(arg_reg_2);
-  __ pop(arg_reg_1);
-
+  __k pop(arg_reg_2);
+  __k pop(arg_reg_1);
   __ Ret();
 }
 
 
 template<class T>
 static void CreateArrayDispatch(MacroAssembler* masm) {
   int last_index = GetSequenceIndexFromFastElementsKind(
       TERMINAL_FAST_ELEMENTS_KIND);
   for (int i = 0; i <= last_index; ++i) {
     Label next;
@@ skipping 27 matching lines @@
   Handle<Object> undefined_sentinel(
       masm->isolate()->heap()->undefined_value(),
       masm->isolate());
 
   // is the low bit set? If so, we are holey and that is good.
   __ testb(rdx, Immediate(1));
   Label normal_sequence;
   __ j(not_zero, &normal_sequence);
 
   // look at the first argument
-  __ movq(rcx, Operand(rsp, kPointerSize));
+  __a movq(rcx, Operand(rsp, 1 * kPointerSize));
   __ testq(rcx, rcx);
   __ j(zero, &normal_sequence);
 
   // We are going to create a holey array, but our kind is non-holey.
   // Fix kind and retry
   __ incl(rdx);
   __ Cmp(rbx, undefined_sentinel);
   __ j(equal, &normal_sequence);
 
   // The type cell may have gone megamorphic, don't overwrite if so
@@ skipping 146 matching lines @@
   InternalArrayNoArgumentConstructorStub stub0(kind);
   __ TailCallStub(&stub0);
 
   __ bind(&not_zero_case);
   __ cmpl(rax, Immediate(1));
   __ j(greater, &not_one_case);
 
   if (IsFastPackedElementsKind(kind)) {
     // We might need to create a holey array
     // look at the first argument
-    __ movq(rcx, Operand(rsp, kPointerSize));
+    __a movq(rcx, Operand(rsp, 1 * kPointerSize));
     __ testq(rcx, rcx);
     __ j(zero, &normal_sequence);
 
     InternalArraySingleArgumentConstructorStub
         stub1_holey(GetHoleyElementsKind(kind));
     __ TailCallStub(&stub1_holey);
   }
 
   __ bind(&normal_sequence);
   InternalArraySingleArgumentConstructorStub stub1(kind);
@@ skipping 50 matching lines @@
 
   Label fast_elements_case;
   __ cmpl(rcx, Immediate(FAST_ELEMENTS));
   __ j(equal, &fast_elements_case);
   GenerateCase(masm, FAST_HOLEY_ELEMENTS);
 
   __ bind(&fast_elements_case);
   GenerateCase(masm, FAST_ELEMENTS);
 }
 
-
+#undef __n
+#undef __s
+#undef __q
+#undef __a
+#undef __k
 #undef __
 
 } }  // namespace v8::internal
 
 #endif  // V8_TARGET_ARCH_X64