Subzero. X86. Uses pshufb for shufflevector lowering.

src/IceAssemblerX86BaseImpl.h

 //===- subzero/src/IceAssemblerX86BaseImpl.h - base x86 assembler -*- C++ -*-=//
 // 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.
 //
 // Modified by the Subzero authors.
 //
 //===----------------------------------------------------------------------===//
 //
 //                        The Subzero Code Generator
@@ skipping 1517 matching lines @@
                                           const Immediate &imm) {
   // Load 32-bit immediate value into tmp1.
   mov(IceType_i32, tmp1, imm);
   // Move value from tmp1 into dst.
   movd(IceType_i32, dst, tmp1);
   // Broadcast low lane into other three lanes.
   shufps(RexTypeIrrelevant, dst, dst, Immediate(0x0));
 }
 
 template <typename TraitsType>
+void AssemblerX86Base<TraitsType>::pshufb(Type /* Ty */, XmmRegister dst,
+                                          XmmRegister src) {
+  AssemblerBuffer::EnsureCapacity ensured(&Buffer);
+  emitUint8(0x66);
+  emitRexRB(RexTypeIrrelevant, dst, src);
+  emitUint8(0x0F);
+  emitUint8(0x38);
+  emitUint8(0x00);
+  emitXmmRegisterOperand(dst, src);
+}
+
+template <typename TraitsType>
+void AssemblerX86Base<TraitsType>::pshufb(Type /* Ty */, XmmRegister dst,
+                                          const Address &src) {
+  AssemblerBuffer::EnsureCapacity ensured(&Buffer);
+  emitUint8(0x66);
+  emitAddrSizeOverridePrefix();
+  emitRex(RexTypeIrrelevant, src, dst);
+  emitUint8(0x0F);
+  emitUint8(0x38);
+  emitUint8(0x00);
+  emitOperand(gprEncoding(dst), src);
+}
+
+template <typename TraitsType>
 void AssemblerX86Base<TraitsType>::pshufd(Type /* Ty */, XmmRegister dst,
                                           XmmRegister src,
                                           const Immediate &imm) {
   AssemblerBuffer::EnsureCapacity ensured(&Buffer);
   emitUint8(0x66);
   emitRexRB(RexTypeIrrelevant, dst, src);
   emitUint8(0x0F);
   emitUint8(0x70);
   emitXmmRegisterOperand(dst, src);
   assert(imm.is_uint8());
@@ skipping 2010 matching lines @@
   (void)shifter;
   if (Ty == IceType_i16)
     emitOperandSizeOverride();
   emitRexB(Ty, operand.rm());
   emitUint8(isByteSizedArithType(Ty) ? 0xD2 : 0xD3);
   emitOperand(rm, operand);
 }
 
 } // end of namespace X86NAMESPACE
 } // end of namespace Ice