Index: lib/Target/AArch64/AArch64InstrInfo.h
===================================================================
--- lib/Target/AArch64/AArch64InstrInfo.h
+++ lib/Target/AArch64/AArch64InstrInfo.h
@@ -42,6 +42,9 @@
                    MachineBasicBlock::iterator I, DebugLoc DL,
                    unsigned DestReg, unsigned SrcReg,
                    bool KillSrc) const;
+  void CopyPhysRegTuple(MachineBasicBlock &MBB,
+                        MachineBasicBlock::iterator I, DebugLoc DL,
+                        unsigned DestReg, unsigned SrcReg) const;
 
   void storeRegToStackSlot(MachineBasicBlock &MBB,
                            MachineBasicBlock::iterator MI,
Index: lib/Target/AArch64/AArch64InstrInfo.cpp
===================================================================
--- lib/Target/AArch64/AArch64InstrInfo.cpp
+++ lib/Target/AArch64/AArch64InstrInfo.cpp
@@ -134,7 +134,8 @@
       return;
     }
   } else {
-    llvm_unreachable("Unknown register class in copyPhysReg");
+    CopyPhysRegTuple(MBB, I, DL, DestReg, SrcReg);
+    return;
   }
 
   // E.g. ORR xDst, xzr, xSrc, lsl #0
@@ -144,6 +145,55 @@
     .addImm(0);
 }
 
+void AArch64InstrInfo::CopyPhysRegTuple(MachineBasicBlock &MBB,
+                                        MachineBasicBlock::iterator I,
+                                        DebugLoc DL, unsigned DestReg,
+                                        unsigned SrcReg) const {
+  unsigned SubRegs;
+  bool IsQRegs;
+  if (AArch64::DPairRegClass.contains(DestReg, SrcReg)) {
+    SubRegs = 2;
+    IsQRegs = false;
+  } else if (AArch64::DTripleRegClass.contains(DestReg, SrcReg)) {
+    SubRegs = 3;
+    IsQRegs = false;
+  } else if (AArch64::DQuadRegClass.contains(DestReg, SrcReg)) {
+    SubRegs = 4;
+    IsQRegs = false;
+  } else if (AArch64::QPairRegClass.contains(DestReg, SrcReg)) {
+    SubRegs = 2;
+    IsQRegs = true;
+  } else if (AArch64::QTripleRegClass.contains(DestReg, SrcReg)) {
+    SubRegs = 3;
+    IsQRegs = true;
+  } else if (AArch64::QQuadRegClass.contains(DestReg, SrcReg)) {
+    SubRegs = 4;
+    IsQRegs = true;
+  } else
+    llvm_unreachable("Unknown register class");
+
+  unsigned BeginIdx = IsQRegs ? AArch64::qsub_0 : AArch64::dsub_0;
+  int Spacing = 1;
+  const TargetRegisterInfo *TRI = &getRegisterInfo();
+  // Copy register tuples backward when the first Dest reg overlaps
+  // with SrcReg.
+  if (TRI->regsOverlap(SrcReg, TRI->getSubReg(DestReg, BeginIdx))) {
+    BeginIdx = BeginIdx + (SubRegs - 1);
+    Spacing = -1;
+  }
+
+  unsigned Opc = IsQRegs ? AArch64::ORRvvv_16B : AArch64::ORRvvv_8B;
+  for (unsigned i = 0; i != SubRegs; ++i) {
+    unsigned Dst = TRI->getSubReg(DestReg, BeginIdx + i * Spacing);
+    unsigned Src = TRI->getSubReg(SrcReg, BeginIdx + i * Spacing);
+    assert(Dst && Src && "Bad sub-register");
+    BuildMI(MBB, I, I->getDebugLoc(), get(Opc), Dst)
+        .addReg(Src)
+        .addReg(Src);
+  }
+  return;
+}
+
 /// Does the Opcode represent a conditional branch that we can remove and re-add
 /// at the end of a basic block?
 static bool isCondBranch(unsigned Opc) {
Index: test/CodeGen/AArch64/neon-copyPhysReg-tuple.ll
===================================================================
--- /dev/null
+++ test/CodeGen/AArch64/neon-copyPhysReg-tuple.ll
@@ -0,0 +1,30 @@
+; RUN: llc < %s -verify-machineinstrs -mtriple=aarch64-none-linux-gnu -mattr=+neon | FileCheck %s
+
+define void @test_copyTuples(i32* %a, i32* %b) {
+; CHECK-LABEL: test_copyTuples:
+; CHECK: orr v{{[0-9]+}}.16b, v{{[0-9]+}}.16b, v{{[0-9]+}}.16b
+; CHECK: orr v{{[0-9]+}}.16b, v{{[0-9]+}}.16b, v{{[0-9]+}}.16b
+; CHECK: orr v{{[0-9]+}}.16b, v{{[0-9]+}}.16b, v{{[0-9]+}}.16b
+; CHECK: orr v{{[0-9]+}}.16b, v{{[0-9]+}}.16b, v{{[0-9]+}}.16b
+; CHECK: ld4 {{{v[0-9]+}}.s, {{v[0-9]+}}.s, {{v[0-9]+}}.s, {{v[0-9]+}}.s}[{{[0-9]+}}], [x{{[0-9]+|sp}}]
+entry:
+  %0 = bitcast i32* %a to i8*
+  %vld3_lane = tail call { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } @llvm.arm.neon.vld4lane.v2i32(i8* %0, <2 x i32> <i32 -2147450880, i32 -32769>, <2 x i32> <i32 -32769, i32 -2147483648>, <2 x i32> <i32 -2147450881, i32 -65536>, <2 x i32> <i32 -2147418113, i32 -32768>, i32 0, i32 4)
+  %vld3_lane.fca.0.extract = extractvalue { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } %vld3_lane, 0
+  %vld3_lane.fca.1.extract = extractvalue { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } %vld3_lane, 1
+  %vld3_lane.fca.2.extract = extractvalue { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } %vld3_lane, 2
+  %vld3_lane.fca.3.extract = extractvalue { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } %vld3_lane, 3
+  %1 = bitcast i32* %b to i8*
+  tail call void @llvm.arm.neon.vst4.v2i32(i8* %1, <2 x i32> %vld3_lane.fca.0.extract, <2 x i32> %vld3_lane.fca.1.extract, <2 x i32> %vld3_lane.fca.2.extract, <2 x i32> %vld3_lane.fca.3.extract, i32 4)
+  %vld3_lane98 = tail call { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } @llvm.arm.neon.vld4lane.v2i32(i8* %1, <2 x i32> <i32 -65536, i32 -2147418113>, <2 x i32> <i32 -41267, i32 -26109>, <2 x i32> <i32 32768, i32 -2147483648>, <2 x i32> <i32 -2147450881, i32 -65536>, i32 1, i32 4)
+  %vld3_lane98.fca.0.extract = extractvalue { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } %vld3_lane98, 0
+  %vld3_lane98.fca.1.extract = extractvalue { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } %vld3_lane98, 1
+  %vld3_lane98.fca.2.extract = extractvalue { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } %vld3_lane98, 2
+  %vld3_lane98.fca.3.extract = extractvalue { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } %vld3_lane98, 3
+  tail call void @llvm.arm.neon.vst4.v2i32(i8* %0, <2 x i32> %vld3_lane98.fca.0.extract, <2 x i32> %vld3_lane98.fca.1.extract, <2 x i32> %vld3_lane98.fca.2.extract, <2 x i32> %vld3_lane98.fca.3.extract, i32 4)
+  ret void
+}
+
+declare { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } @llvm.arm.neon.vld4lane.v2i32(i8*, <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32>, i32, i32)
+
+declare void @llvm.arm.neon.vst4.v2i32(i8*, <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32>, i32)
\ No newline at end of file