Subzero. X86. Lowers shufflevector using xmm instructions.

src/IceTargetLoweringX86Base.h

Index: src/IceTargetLoweringX86Base.h
diff --git a/src/IceTargetLoweringX86Base.h b/src/IceTargetLoweringX86Base.h
index f84c6df1d492b72c4230c6b26a26662e9ad5b915..71b824f7ad3e4bcdbdb2fa0f70c69647d389edb3 100644
--- a/src/IceTargetLoweringX86Base.h
+++ b/src/IceTargetLoweringX86Base.h
@@ -801,6 +801,10 @@ protected:
     AutoMemorySandboxer<> _(this, &Dest, &Src0);
     Context.insert<typename Traits::Insts::Por>(Dest, Src0);
   }
+  void _punpckl(Variable *Dest, Operand *Src0) {
+    AutoMemorySandboxer<> _(this, &Dest, &Src0);
+    Context.insert<typename Traits::Insts::Punpckl>(Dest, Src0);
+  }
   void _pshufd(Variable *Dest, Operand *Src0, Operand *Src1) {
     AutoMemorySandboxer<> _(this, &Dest, &Src0, &Src1);
     Context.insert<typename Traits::Insts::Pshufd>(Dest, Src0, Src1);
@@ -1082,6 +1086,23 @@ private:
 
   BoolFolding<Traits> FoldingInfo;
 
+  /// Helpers for lowering ShuffleVector
+  /// @{
+  Variable *lowerShuffleVector_AllFromSameSrc(Variable *Src, SizeT Index0,
+                                              SizeT Index1, SizeT Index2,
+                                              SizeT Index3);
+  static constexpr SizeT IGNORE_INDEX = 0x80000000u;
+  Variable *lowerShuffleVector_TwoFromSameSrc(Variable *Src0, SizeT Index0,
+                                              SizeT Index1, Variable *Src1,
+                                              SizeT Index2, SizeT Index3);
+  static constexpr SizeT UNIFIED_INDEX_0 = 0;
+  static constexpr SizeT UNIFIED_INDEX_1 = 2;
+  Variable *lowerShuffleVector_UnifyFromDifferentSrcs(Variable *Src0,
+                                                      SizeT Index0,
+                                                      Variable *Src1,
+                                                      SizeT Index1);
+  /// @}
+
   static FixupKind PcRelFixup;
   static FixupKind AbsFixup;
 };