[turbofan] avoid xchg instruction on Intel

src/compiler/ia32/code-generator-ia32.cc

 // Copyright 2013 the V8 project authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "src/compiler/code-generator.h"
 
 #include "src/ast/scopes.h"
 #include "src/compiler/code-generator-impl.h"
 #include "src/compiler/gap-resolver.h"
 #include "src/compiler/node-matchers.h"
@@ skipping 1598 matching lines @@
 
 void CodeGenerator::AssembleSwap(InstructionOperand* source,
                                  InstructionOperand* destination) {
   IA32OperandConverter g(this, nullptr);
   // Dispatch on the source and destination operand kinds.  Not all
   // combinations are possible.
   if (source->IsRegister() && destination->IsRegister()) {
     // Register-register.
     Register src = g.ToRegister(source);
     Register dst = g.ToRegister(destination);
-    __ xchg(dst, src);
+    __ push(src);

[yosin_UTC9, 2016/01/20 01:48:18]

Just a curiosity, why don't you use XOR-technique for swapping registers? Is
PUSH/POP faster than three-XOR?

+    __ mov(src, dst);
+    __ pop(dst);
   } else if (source->IsRegister() && destination->IsStackSlot()) {
     // Register-memory.
-    __ xchg(g.ToRegister(source), g.ToOperand(destination));
+    Register src = g.ToRegister(source);
+    __ push(src);
+    frame_access_state()->IncreaseSPDelta(1);
+    Operand dst = g.ToOperand(destination);
+    __ mov(src, dst);
+    frame_access_state()->IncreaseSPDelta(-1);
+    dst = g.ToOperand(destination);
+    __ pop(dst);
   } else if (source->IsStackSlot() && destination->IsStackSlot()) {
     // Memory-memory.
     Operand dst1 = g.ToOperand(destination);
     __ push(dst1);
     frame_access_state()->IncreaseSPDelta(1);
     Operand src1 = g.ToOperand(source);
     __ push(src1);
     Operand dst2 = g.ToOperand(destination);
     __ pop(dst2);
     frame_access_state()->IncreaseSPDelta(-1);
@@ skipping 55 matching lines @@
     int padding_size = last_lazy_deopt_pc_ + space_needed - current_pc;
     __ Nop(padding_size);
   }
 }
 
 #undef __
 
 }  // namespace compiler
 }  // namespace internal
 }  // namespace v8