Add two argument SSE shuffle instructions

runtime/vm/assembler_x64.cc

 // Copyright (c) 2013, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
 // BSD-style license that can be found in the LICENSE file.
 
 #include "vm/globals.h"
 #if defined(TARGET_ARCH_X64)
 
 #include "vm/assembler.h"
 #include "vm/heap.h"
 #include "vm/memory_region.h"
@@ skipping 821 matching lines @@
 
 void Assembler::reciprocalps(XmmRegister dst) {
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
   EmitREX_RB(dst, dst);
   EmitUint8(0x0F);
   EmitUint8(0x53);
   EmitXmmRegisterOperand(dst & 7, dst);
 }
 
 
+void Assembler::movhlps(XmmRegister dst, XmmRegister src) {
+  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  EmitREX_RB(dst, src);
+  EmitUint8(0x0F);
+  EmitUint8(0x12);
+  EmitXmmRegisterOperand(dst & 7, src);
+}
+
+
+void Assembler::movlhps(XmmRegister dst, XmmRegister src) {
+  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  EmitREX_RB(dst, src);
+  EmitUint8(0x0F);
+  EmitUint8(0x16);
+  EmitXmmRegisterOperand(dst & 7, src);
+}
+
+
+void Assembler::unpcklps(XmmRegister dst, XmmRegister src) {
+  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  EmitREX_RB(dst, src);
+  EmitUint8(0x0F);
+  EmitUint8(0x14);
+  EmitXmmRegisterOperand(dst & 7, src);
+}
+
+
+void Assembler::unpckhps(XmmRegister dst, XmmRegister src) {
+  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  EmitREX_RB(dst, src);
+  EmitUint8(0x0F);
+  EmitUint8(0x15);
+  EmitXmmRegisterOperand(dst & 7, src);
+}
+
+
+void Assembler::unpcklpd(XmmRegister dst, XmmRegister src) {
+  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  EmitUint8(0x66);
+  EmitREX_RB(dst, src);
+  EmitUint8(0x0F);
+  EmitUint8(0x14);
+  EmitXmmRegisterOperand(dst & 7, src);
+}
+
+
+void Assembler::unpckhpd(XmmRegister dst, XmmRegister src) {
+  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  EmitUint8(0x66);
+  EmitREX_RB(dst, src);
+  EmitUint8(0x0F);
+  EmitUint8(0x15);
+  EmitXmmRegisterOperand(dst & 7, src);
+}
+
+
 void Assembler::set1ps(XmmRegister dst, Register tmp1, const Immediate& imm) {
   // Load 32-bit immediate value into tmp1.
   movl(tmp1, imm);
   // Move value from tmp1 into dst.
   movd(dst, tmp1);
   // Broadcast low lane into other three lanes.
   shufps(dst, dst, Immediate(0x0));
 }
 
 
@@ skipping 1693 matching lines @@
 
 const char* Assembler::FpuRegisterName(FpuRegister reg) {
   ASSERT((0 <= reg) && (reg < kNumberOfXmmRegisters));
   return xmm_reg_names[reg];
 }
 
 
 }  // namespace dart
 
 #endif  // defined TARGET_ARCH_X64