Subzero. X86. Lowers shufflevector using xmm instructions.

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 1547 matching lines @@
   emitRex(RexTypeIrrelevant, src, dst);
   emitUint8(0x0F);
   emitUint8(0x70);
   static constexpr RelocOffsetT OffsetFromNextInstruction = 1;
   emitOperand(gprEncoding(dst), src, OffsetFromNextInstruction);
   assert(imm.is_uint8());
   emitUint8(imm.value());
 }
 
 template <typename TraitsType>
+void AssemblerX86Base<TraitsType>::punpckldq(Type, XmmRegister Dst,
+                                             XmmRegister Src) {
+  AssemblerBuffer::EnsureCapacity ensured(&Buffer);
+  emitUint8(0x66);
+  emitRexRB(RexTypeIrrelevant, Dst, Src);
+  emitUint8(0x0F);
+  emitUint8(0x62);
+  emitXmmRegisterOperand(Dst, Src);
+}
+
+template <typename TraitsType>
+void AssemblerX86Base<TraitsType>::punpckldq(Type, XmmRegister Dst,
+                                             const Address &Src) {
+  AssemblerBuffer::EnsureCapacity ensured(&Buffer);
+  emitUint8(0x66);
+  emitAddrSizeOverridePrefix();
+  emitRex(RexTypeIrrelevant, Src, Dst);
+  emitUint8(0x0F);
+  emitUint8(0x62);
+  emitOperand(gprEncoding(Dst), Src);
+}
+
+template <typename TraitsType>
 void AssemblerX86Base<TraitsType>::shufps(Type /* Ty */, XmmRegister dst,
                                           XmmRegister src,
                                           const Immediate &imm) {
   AssemblerBuffer::EnsureCapacity ensured(&Buffer);
   emitRexRB(RexTypeIrrelevant, dst, src);
   emitUint8(0x0F);
   emitUint8(0xC6);
   emitXmmRegisterOperand(dst, src);
   assert(imm.is_uint8());
   emitUint8(imm.value());
@@ skipping 1957 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