X-Git-Url: http://plrg.eecs.uci.edu/git/?p=oota-llvm.git;a=blobdiff_plain;f=lib%2FTarget%2FARM%2FARMISelDAGToDAG.cpp;h=5843d89b5781072fd8e91e8e198600934d2b63b4;hp=81229794a2c2bc8325773881b8ac54cb99a85be4;hb=6aaf0a68acfa16b2af1693b7cfcf41f77a4e2244;hpb=d04a8d4b33ff316ca4cf961e06c9e312eff8e64f

diff --git a/lib/Target/ARM/ARMISelDAGToDAG.cpp b/lib/Target/ARM/ARMISelDAGToDAG.cpp
index 81229794a2c..5843d89b578 100644
--- a/lib/Target/ARM/ARMISelDAGToDAG.cpp
+++ b/lib/Target/ARM/ARMISelDAGToDAG.cpp
@@ -11,32 +11,34 @@
 //
 //===----------------------------------------------------------------------===//
 
-#define DEBUG_TYPE "arm-isel"
 #include "ARM.h"
 #include "ARMBaseInstrInfo.h"
 #include "ARMTargetMachine.h"
 #include "MCTargetDesc/ARMAddressingModes.h"
-#include "llvm/CallingConv.h"
+#include "llvm/ADT/StringSwitch.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/SelectionDAG.h"
 #include "llvm/CodeGen/SelectionDAGISel.h"
-#include "llvm/Constants.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Function.h"
-#include "llvm/Intrinsics.h"
-#include "llvm/LLVMContext.h"
+#include "llvm/IR/CallingConv.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/LLVMContext.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Compiler.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
-#include "llvm/Support/raw_ostream.h"
 #include "llvm/Target/TargetLowering.h"
 #include "llvm/Target/TargetOptions.h"
 
 using namespace llvm;
 
+#define DEBUG_TYPE "arm-isel"
+
 static cl::opt<bool>
 DisableShifterOp("disable-shifter-op", cl::Hidden,
   cl::desc("Disable isel of shifter-op"),
@@ -59,32 +61,34 @@ enum AddrMode2Type {
 };
 
 class ARMDAGToDAGISel : public SelectionDAGISel {
-  ARMBaseTargetMachine &TM;
-  const ARMBaseInstrInfo *TII;
-
   /// Subtarget - Keep a pointer to the ARMSubtarget around so that we can
   /// make the right decision when generating code for different targets.
   const ARMSubtarget *Subtarget;
 
 public:
-  explicit ARMDAGToDAGISel(ARMBaseTargetMachine &tm,
-                           CodeGenOpt::Level OptLevel)
-    : SelectionDAGISel(tm, OptLevel), TM(tm),
-      TII(static_cast<const ARMBaseInstrInfo*>(TM.getInstrInfo())),
-      Subtarget(&TM.getSubtarget<ARMSubtarget>()) {
+  explicit ARMDAGToDAGISel(ARMBaseTargetMachine &tm, CodeGenOpt::Level OptLevel)
+      : SelectionDAGISel(tm, OptLevel) {}
+
+  bool runOnMachineFunction(MachineFunction &MF) override {
+    // Reset the subtarget each time through.
+    Subtarget = &MF.getSubtarget<ARMSubtarget>();
+    SelectionDAGISel::runOnMachineFunction(MF);
+    return true;
   }
 
-  virtual const char *getPassName() const {
+  const char *getPassName() const override {
     return "ARM Instruction Selection";
   }
 
+  void PreprocessISelDAG() override;
+
   /// getI32Imm - Return a target constant of type i32 with the specified
   /// value.
-  inline SDValue getI32Imm(unsigned Imm) {
-    return CurDAG->getTargetConstant(Imm, MVT::i32);
+  inline SDValue getI32Imm(unsigned Imm, SDLoc dl) {
+    return CurDAG->getTargetConstant(Imm, dl, MVT::i32);
   }
 
-  SDNode *Select(SDNode *N);
+  SDNode *Select(SDNode *N) override;
 
 
   bool hasNoVMLxHazardUse(SDNode *N) const;
@@ -129,6 +133,13 @@ public:
     return true;
   }
 
+  bool SelectCMOVPred(SDValue N, SDValue &Pred, SDValue &Reg) {
+    const ConstantSDNode *CN = cast<ConstantSDNode>(N);
+    Pred = CurDAG->getTargetConstant(CN->getZExtValue(), SDLoc(N), MVT::i32);
+    Reg = CurDAG->getRegister(ARM::CPSR, MVT::i32);
+    return true;
+  }
+
   bool SelectAddrMode2OffsetReg(SDNode *Op, SDValue N,
                              SDValue &Offset, SDValue &Opc);
   bool SelectAddrMode2OffsetImm(SDNode *Op, SDValue N,
@@ -149,11 +160,6 @@ public:
 
   // Thumb Addressing Modes:
   bool SelectThumbAddrModeRR(SDValue N, SDValue &Base, SDValue &Offset);
-  bool SelectThumbAddrModeRI(SDValue N, SDValue &Base, SDValue &Offset,
-                             unsigned Scale);
-  bool SelectThumbAddrModeRI5S1(SDValue N, SDValue &Base, SDValue &Offset);
-  bool SelectThumbAddrModeRI5S2(SDValue N, SDValue &Base, SDValue &Offset);
-  bool SelectThumbAddrModeRI5S4(SDValue N, SDValue &Base, SDValue &Offset);
   bool SelectThumbAddrModeImm5S(SDValue N, unsigned Scale, SDValue &Base,
                                 SDValue &OffImm);
   bool SelectThumbAddrModeImm5S1(SDValue N, SDValue &Base,
@@ -165,8 +171,6 @@ public:
   bool SelectThumbAddrModeSP(SDValue N, SDValue &Base, SDValue &OffImm);
 
   // Thumb 2 Addressing Modes:
-  bool SelectT2ShifterOperandReg(SDValue N,
-                                 SDValue &BaseReg, SDValue &Opc);
   bool SelectT2AddrModeImm12(SDValue N, SDValue &Base, SDValue &OffImm);
   bool SelectT2AddrModeImm8(SDValue N, SDValue &Base,
                             SDValue &OffImm);
@@ -174,6 +178,7 @@ public:
                                  SDValue &OffImm);
   bool SelectT2AddrModeSoReg(SDValue N, SDValue &Base,
                              SDValue &OffReg, SDValue &ShImm);
+  bool SelectT2AddrModeExclusive(SDValue N, SDValue &Base, SDValue &OffImm);
 
   inline bool is_so_imm(unsigned Imm) const {
     return ARM_AM::getSOImmVal(Imm) != -1;
@@ -237,33 +242,20 @@ private:
   /// SelectV6T2BitfieldExtractOp - Select SBFX/UBFX instructions for ARM.
   SDNode *SelectV6T2BitfieldExtractOp(SDNode *N, bool isSigned);
 
-  /// SelectCMOVOp - Select CMOV instructions for ARM.
-  SDNode *SelectCMOVOp(SDNode *N);
-  SDNode *SelectT2CMOVShiftOp(SDNode *N, SDValue FalseVal, SDValue TrueVal,
-                              ARMCC::CondCodes CCVal, SDValue CCR,
-                              SDValue InFlag);
-  SDNode *SelectARMCMOVShiftOp(SDNode *N, SDValue FalseVal, SDValue TrueVal,
-                               ARMCC::CondCodes CCVal, SDValue CCR,
-                               SDValue InFlag);
-  SDNode *SelectT2CMOVImmOp(SDNode *N, SDValue FalseVal, SDValue TrueVal,
-                              ARMCC::CondCodes CCVal, SDValue CCR,
-                              SDValue InFlag);
-  SDNode *SelectARMCMOVImmOp(SDNode *N, SDValue FalseVal, SDValue TrueVal,
-                               ARMCC::CondCodes CCVal, SDValue CCR,
-                               SDValue InFlag);
-
   // Select special operations if node forms integer ABS pattern
   SDNode *SelectABSOp(SDNode *N);
 
-  SDNode *SelectConcatVector(SDNode *N);
+  SDNode *SelectReadRegister(SDNode *N);
+  SDNode *SelectWriteRegister(SDNode *N);
+
+  SDNode *SelectInlineAsm(SDNode *N);
 
-  SDNode *SelectAtomic64(SDNode *Node, unsigned Opc);
+  SDNode *SelectConcatVector(SDNode *N);
 
   /// SelectInlineAsmMemoryOperand - Implement addressing mode selection for
   /// inline asm expressions.
-  virtual bool SelectInlineAsmMemoryOperand(const SDValue &Op,
-                                            char ConstraintCode,
-                                            std::vector<SDValue> &OutOps);
+  bool SelectInlineAsmMemoryOperand(const SDValue &Op, unsigned ConstraintID,
+                                    std::vector<SDValue> &OutOps) override;
 
   // Form pairs of consecutive R, S, D, or Q registers.
   SDNode *createGPRPairNode(EVT VT, SDValue V0, SDValue V1);
@@ -277,7 +269,8 @@ private:
   SDNode *createQuadQRegsNode(EVT VT, SDValue V0, SDValue V1, SDValue V2, SDValue V3);
 
   // Get the alignment operand for a NEON VLD or VST instruction.
-  SDValue GetVLDSTAlign(SDValue Align, unsigned NumVecs, bool is64BitVector);
+  SDValue GetVLDSTAlign(SDValue Align, SDLoc dl, unsigned NumVecs,
+                        bool is64BitVector);
 };
 }
 
@@ -327,6 +320,89 @@ static bool isScaledConstantInRange(SDValue Node, int Scale,
   return ScaledConstant >= RangeMin && ScaledConstant < RangeMax;
 }
 
+void ARMDAGToDAGISel::PreprocessISelDAG() {
+  if (!Subtarget->hasV6T2Ops())
+    return;
+
+  bool isThumb2 = Subtarget->isThumb();
+  for (SelectionDAG::allnodes_iterator I = CurDAG->allnodes_begin(),
+       E = CurDAG->allnodes_end(); I != E; ) {
+    SDNode *N = I++;  // Preincrement iterator to avoid invalidation issues.
+
+    if (N->getOpcode() != ISD::ADD)
+      continue;
+
+    // Look for (add X1, (and (srl X2, c1), c2)) where c2 is constant with
+    // leading zeros, followed by consecutive set bits, followed by 1 or 2
+    // trailing zeros, e.g. 1020.
+    // Transform the expression to
+    // (add X1, (shl (and (srl X2, c1), (c2>>tz)), tz)) where tz is the number
+    // of trailing zeros of c2. The left shift would be folded as an shifter
+    // operand of 'add' and the 'and' and 'srl' would become a bits extraction
+    // node (UBFX).
+
+    SDValue N0 = N->getOperand(0);
+    SDValue N1 = N->getOperand(1);
+    unsigned And_imm = 0;
+    if (!isOpcWithIntImmediate(N1.getNode(), ISD::AND, And_imm)) {
+      if (isOpcWithIntImmediate(N0.getNode(), ISD::AND, And_imm))
+        std::swap(N0, N1);
+    }
+    if (!And_imm)
+      continue;
+
+    // Check if the AND mask is an immediate of the form: 000.....1111111100
+    unsigned TZ = countTrailingZeros(And_imm);
+    if (TZ != 1 && TZ != 2)
+      // Be conservative here. Shifter operands aren't always free. e.g. On
+      // Swift, left shifter operand of 1 / 2 for free but others are not.
+      // e.g.
+      //  ubfx   r3, r1, #16, #8
+      //  ldr.w  r3, [r0, r3, lsl #2]
+      // vs.
+      //  mov.w  r9, #1020
+      //  and.w  r2, r9, r1, lsr #14
+      //  ldr    r2, [r0, r2]
+      continue;
+    And_imm >>= TZ;
+    if (And_imm & (And_imm + 1))
+      continue;
+
+    // Look for (and (srl X, c1), c2).
+    SDValue Srl = N1.getOperand(0);
+    unsigned Srl_imm = 0;
+    if (!isOpcWithIntImmediate(Srl.getNode(), ISD::SRL, Srl_imm) ||
+        (Srl_imm <= 2))
+      continue;
+
+    // Make sure first operand is not a shifter operand which would prevent
+    // folding of the left shift.
+    SDValue CPTmp0;
+    SDValue CPTmp1;
+    SDValue CPTmp2;
+    if (isThumb2) {
+      if (SelectImmShifterOperand(N0, CPTmp0, CPTmp1))
+        continue;
+    } else {
+      if (SelectImmShifterOperand(N0, CPTmp0, CPTmp1) ||
+          SelectRegShifterOperand(N0, CPTmp0, CPTmp1, CPTmp2))
+        continue;
+    }
+
+    // Now make the transformation.
+    Srl = CurDAG->getNode(ISD::SRL, SDLoc(Srl), MVT::i32,
+                          Srl.getOperand(0),
+                          CurDAG->getConstant(Srl_imm + TZ, SDLoc(Srl),
+                                              MVT::i32));
+    N1 = CurDAG->getNode(ISD::AND, SDLoc(N1), MVT::i32,
+                         Srl,
+                         CurDAG->getConstant(And_imm, SDLoc(Srl), MVT::i32));
+    N1 = CurDAG->getNode(ISD::SHL, SDLoc(N1), MVT::i32,
+                         N1, CurDAG->getConstant(TZ, SDLoc(Srl), MVT::i32));
+    CurDAG->UpdateNodeOperands(N, N0, N1);
+  }
+}
+
 /// hasNoVMLxHazardUse - Return true if it's desirable to select a FP MLA / MLS
 /// node. VFP / NEON fp VMLA / VMLS instructions have special RAW hazards (at
 /// least on current ARM implementations) which should be avoidded.
@@ -337,8 +413,8 @@ bool ARMDAGToDAGISel::hasNoVMLxHazardUse(SDNode *N) const {
   if (!CheckVMLxHazard)
     return true;
 
-  if (!Subtarget->isCortexA8() && !Subtarget->isLikeA9() &&
-      !Subtarget->isSwift())
+  if (!Subtarget->isCortexA7() && !Subtarget->isCortexA8() &&
+      !Subtarget->isCortexA9() && !Subtarget->isSwift())
     return true;
 
   if (!N->hasOneUse())
@@ -348,6 +424,9 @@ bool ARMDAGToDAGISel::hasNoVMLxHazardUse(SDNode *N) const {
   if (Use->getOpcode() == ISD::CopyToReg)
     return true;
   if (Use->isMachineOpcode()) {
+    const ARMBaseInstrInfo *TII = static_cast<const ARMBaseInstrInfo *>(
+        CurDAG->getSubtarget().getInstrInfo());
+
     const MCInstrDesc &MCID = TII->get(Use->getMachineOpcode());
     if (MCID.mayStore())
       return true;
@@ -404,7 +483,7 @@ bool ARMDAGToDAGISel::SelectImmShifterOperand(SDValue N,
   if (!RHS) return false;
   ShImmVal = RHS->getZExtValue() & 31;
   Opc = CurDAG->getTargetConstant(ARM_AM::getSORegOpc(ShOpcVal, ShImmVal),
-                                  MVT::i32);
+                                  SDLoc(N), MVT::i32);
   return true;
 }
 
@@ -431,7 +510,7 @@ bool ARMDAGToDAGISel::SelectRegShifterOperand(SDValue N,
   if (CheckProfitability && !isShifterOpProfitable(N, ShOpcVal, ShImmVal))
     return false;
   Opc = CurDAG->getTargetConstant(ARM_AM::getSORegOpc(ShOpcVal, ShImmVal),
-                                  MVT::i32);
+                                  SDLoc(N), MVT::i32);
   return true;
 }
 
@@ -447,40 +526,41 @@ bool ARMDAGToDAGISel::SelectAddrModeImm12(SDValue N,
     if (N.getOpcode() == ISD::FrameIndex) {
       // Match frame index.
       int FI = cast<FrameIndexSDNode>(N)->getIndex();
-      Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
-      OffImm  = CurDAG->getTargetConstant(0, MVT::i32);
+      Base = CurDAG->getTargetFrameIndex(
+          FI, TLI->getPointerTy(CurDAG->getDataLayout()));
+      OffImm  = CurDAG->getTargetConstant(0, SDLoc(N), MVT::i32);
       return true;
     }
 
     if (N.getOpcode() == ARMISD::Wrapper &&
-        !(Subtarget->useMovt() &&
-                     N.getOperand(0).getOpcode() == ISD::TargetGlobalAddress)) {
+        N.getOperand(0).getOpcode() != ISD::TargetGlobalAddress) {
       Base = N.getOperand(0);
     } else
       Base = N;
-    OffImm  = CurDAG->getTargetConstant(0, MVT::i32);
+    OffImm  = CurDAG->getTargetConstant(0, SDLoc(N), MVT::i32);
     return true;
   }
 
   if (ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(N.getOperand(1))) {
-    int RHSC = (int)RHS->getZExtValue();
+    int RHSC = (int)RHS->getSExtValue();
     if (N.getOpcode() == ISD::SUB)
       RHSC = -RHSC;
 
-    if (RHSC >= 0 && RHSC < 0x1000) { // 12 bits (unsigned)
+    if (RHSC > -0x1000 && RHSC < 0x1000) { // 12 bits
       Base   = N.getOperand(0);
       if (Base.getOpcode() == ISD::FrameIndex) {
         int FI = cast<FrameIndexSDNode>(Base)->getIndex();
-        Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
+        Base = CurDAG->getTargetFrameIndex(
+            FI, TLI->getPointerTy(CurDAG->getDataLayout()));
       }
-      OffImm = CurDAG->getTargetConstant(RHSC, MVT::i32);
+      OffImm = CurDAG->getTargetConstant(RHSC, SDLoc(N), MVT::i32);
       return true;
     }
   }
 
   // Base only.
   Base = N;
-  OffImm  = CurDAG->getTargetConstant(0, MVT::i32);
+  OffImm  = CurDAG->getTargetConstant(0, SDLoc(N), MVT::i32);
   return true;
 }
 
@@ -505,7 +585,7 @@ bool ARMDAGToDAGISel::SelectLdStSOReg(SDValue N, SDValue &Base, SDValue &Offset,
           Base = Offset = N.getOperand(0);
           Opc = CurDAG->getTargetConstant(ARM_AM::getAM2Opc(AddSub, ShAmt,
                                                             ARM_AM::lsl),
-                                          MVT::i32);
+                                          SDLoc(N), MVT::i32);
           return true;
         }
       }
@@ -576,7 +656,7 @@ bool ARMDAGToDAGISel::SelectLdStSOReg(SDValue N, SDValue &Base, SDValue &Offset,
   }
 
   Opc = CurDAG->getTargetConstant(ARM_AM::getAM2Opc(AddSub, ShAmt, ShOpcVal),
-                                  MVT::i32);
+                                  SDLoc(N), MVT::i32);
   return true;
 }
 
@@ -604,7 +684,7 @@ AddrMode2Type ARMDAGToDAGISel::SelectAddrMode2Worker(SDValue N,
           Base = Offset = N.getOperand(0);
           Opc = CurDAG->getTargetConstant(ARM_AM::getAM2Opc(AddSub, ShAmt,
                                                             ARM_AM::lsl),
-                                          MVT::i32);
+                                          SDLoc(N), MVT::i32);
           return AM2_SHOP;
         }
       }
@@ -617,16 +697,16 @@ AddrMode2Type ARMDAGToDAGISel::SelectAddrMode2Worker(SDValue N,
     Base = N;
     if (N.getOpcode() == ISD::FrameIndex) {
       int FI = cast<FrameIndexSDNode>(N)->getIndex();
-      Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
+      Base = CurDAG->getTargetFrameIndex(
+          FI, TLI->getPointerTy(CurDAG->getDataLayout()));
     } else if (N.getOpcode() == ARMISD::Wrapper &&
-               !(Subtarget->useMovt() &&
-                 N.getOperand(0).getOpcode() == ISD::TargetGlobalAddress)) {
+               N.getOperand(0).getOpcode() != ISD::TargetGlobalAddress) {
       Base = N.getOperand(0);
     }
     Offset = CurDAG->getRegister(0, MVT::i32);
     Opc = CurDAG->getTargetConstant(ARM_AM::getAM2Opc(ARM_AM::add, 0,
                                                       ARM_AM::no_shift),
-                                    MVT::i32);
+                                    SDLoc(N), MVT::i32);
     return AM2_BASE;
   }
 
@@ -638,7 +718,8 @@ AddrMode2Type ARMDAGToDAGISel::SelectAddrMode2Worker(SDValue N,
       Base = N.getOperand(0);
       if (Base.getOpcode() == ISD::FrameIndex) {
         int FI = cast<FrameIndexSDNode>(Base)->getIndex();
-        Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
+        Base = CurDAG->getTargetFrameIndex(
+            FI, TLI->getPointerTy(CurDAG->getDataLayout()));
       }
       Offset = CurDAG->getRegister(0, MVT::i32);
 
@@ -649,7 +730,7 @@ AddrMode2Type ARMDAGToDAGISel::SelectAddrMode2Worker(SDValue N,
       }
       Opc = CurDAG->getTargetConstant(ARM_AM::getAM2Opc(AddSub, RHSC,
                                                         ARM_AM::no_shift),
-                                      MVT::i32);
+                                      SDLoc(N), MVT::i32);
       return AM2_BASE;
     }
   }
@@ -660,7 +741,7 @@ AddrMode2Type ARMDAGToDAGISel::SelectAddrMode2Worker(SDValue N,
     Offset = CurDAG->getRegister(0, MVT::i32);
     Opc = CurDAG->getTargetConstant(ARM_AM::getAM2Opc(ARM_AM::add, 0,
                                                       ARM_AM::no_shift),
-                                    MVT::i32);
+                                    SDLoc(N), MVT::i32);
     return AM2_BASE;
   }
 
@@ -715,7 +796,7 @@ AddrMode2Type ARMDAGToDAGISel::SelectAddrMode2Worker(SDValue N,
   }
 
   Opc = CurDAG->getTargetConstant(ARM_AM::getAM2Opc(AddSub, ShAmt, ShOpcVal),
-                                  MVT::i32);
+                                  SDLoc(N), MVT::i32);
   return AM2_SHOP;
 }
 
@@ -751,7 +832,7 @@ bool ARMDAGToDAGISel::SelectAddrMode2OffsetReg(SDNode *Op, SDValue N,
   }
 
   Opc = CurDAG->getTargetConstant(ARM_AM::getAM2Opc(AddSub, ShAmt, ShOpcVal),
-                                  MVT::i32);
+                                  SDLoc(N), MVT::i32);
   return true;
 }
 
@@ -767,7 +848,7 @@ bool ARMDAGToDAGISel::SelectAddrMode2OffsetImmPre(SDNode *Op, SDValue N,
   if (isScaledConstantInRange(N, /*Scale=*/1, 0, 0x1000, Val)) { // 12 bits.
     if (AddSub == ARM_AM::sub) Val *= -1;
     Offset = CurDAG->getRegister(0, MVT::i32);
-    Opc = CurDAG->getTargetConstant(Val, MVT::i32);
+    Opc = CurDAG->getTargetConstant(Val, SDLoc(Op), MVT::i32);
     return true;
   }
 
@@ -788,7 +869,7 @@ bool ARMDAGToDAGISel::SelectAddrMode2OffsetImm(SDNode *Op, SDValue N,
     Offset = CurDAG->getRegister(0, MVT::i32);
     Opc = CurDAG->getTargetConstant(ARM_AM::getAM2Opc(AddSub, Val,
                                                       ARM_AM::no_shift),
-                                    MVT::i32);
+                                    SDLoc(Op), MVT::i32);
     return true;
   }
 
@@ -807,7 +888,8 @@ bool ARMDAGToDAGISel::SelectAddrMode3(SDValue N,
     // X - C  is canonicalize to X + -C, no need to handle it here.
     Base = N.getOperand(0);
     Offset = N.getOperand(1);
-    Opc = CurDAG->getTargetConstant(ARM_AM::getAM3Opc(ARM_AM::sub, 0),MVT::i32);
+    Opc = CurDAG->getTargetConstant(ARM_AM::getAM3Opc(ARM_AM::sub, 0), SDLoc(N),
+                                    MVT::i32);
     return true;
   }
 
@@ -815,10 +897,12 @@ bool ARMDAGToDAGISel::SelectAddrMode3(SDValue N,
     Base = N;
     if (N.getOpcode() == ISD::FrameIndex) {
       int FI = cast<FrameIndexSDNode>(N)->getIndex();
-      Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
+      Base = CurDAG->getTargetFrameIndex(
+          FI, TLI->getPointerTy(CurDAG->getDataLayout()));
     }
     Offset = CurDAG->getRegister(0, MVT::i32);
-    Opc = CurDAG->getTargetConstant(ARM_AM::getAM3Opc(ARM_AM::add, 0),MVT::i32);
+    Opc = CurDAG->getTargetConstant(ARM_AM::getAM3Opc(ARM_AM::add, 0), SDLoc(N),
+                                    MVT::i32);
     return true;
   }
 
@@ -829,7 +913,8 @@ bool ARMDAGToDAGISel::SelectAddrMode3(SDValue N,
     Base = N.getOperand(0);
     if (Base.getOpcode() == ISD::FrameIndex) {
       int FI = cast<FrameIndexSDNode>(Base)->getIndex();
-      Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
+      Base = CurDAG->getTargetFrameIndex(
+          FI, TLI->getPointerTy(CurDAG->getDataLayout()));
     }
     Offset = CurDAG->getRegister(0, MVT::i32);
 
@@ -838,13 +923,15 @@ bool ARMDAGToDAGISel::SelectAddrMode3(SDValue N,
       AddSub = ARM_AM::sub;
       RHSC = -RHSC;
     }
-    Opc = CurDAG->getTargetConstant(ARM_AM::getAM3Opc(AddSub, RHSC),MVT::i32);
+    Opc = CurDAG->getTargetConstant(ARM_AM::getAM3Opc(AddSub, RHSC), SDLoc(N),
+                                    MVT::i32);
     return true;
   }
 
   Base = N.getOperand(0);
   Offset = N.getOperand(1);
-  Opc = CurDAG->getTargetConstant(ARM_AM::getAM3Opc(ARM_AM::add, 0), MVT::i32);
+  Opc = CurDAG->getTargetConstant(ARM_AM::getAM3Opc(ARM_AM::add, 0), SDLoc(N),
+                                  MVT::i32);
   return true;
 }
 
@@ -859,12 +946,14 @@ bool ARMDAGToDAGISel::SelectAddrMode3Offset(SDNode *Op, SDValue N,
   int Val;
   if (isScaledConstantInRange(N, /*Scale=*/1, 0, 256, Val)) { // 12 bits.
     Offset = CurDAG->getRegister(0, MVT::i32);
-    Opc = CurDAG->getTargetConstant(ARM_AM::getAM3Opc(AddSub, Val), MVT::i32);
+    Opc = CurDAG->getTargetConstant(ARM_AM::getAM3Opc(AddSub, Val), SDLoc(Op),
+                                    MVT::i32);
     return true;
   }
 
   Offset = N;
-  Opc = CurDAG->getTargetConstant(ARM_AM::getAM3Opc(AddSub, 0), MVT::i32);
+  Opc = CurDAG->getTargetConstant(ARM_AM::getAM3Opc(AddSub, 0), SDLoc(Op),
+                                  MVT::i32);
   return true;
 }
 
@@ -874,14 +963,14 @@ bool ARMDAGToDAGISel::SelectAddrMode5(SDValue N,
     Base = N;
     if (N.getOpcode() == ISD::FrameIndex) {
       int FI = cast<FrameIndexSDNode>(N)->getIndex();
-      Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
+      Base = CurDAG->getTargetFrameIndex(
+          FI, TLI->getPointerTy(CurDAG->getDataLayout()));
     } else if (N.getOpcode() == ARMISD::Wrapper &&
-               !(Subtarget->useMovt() &&
-                 N.getOperand(0).getOpcode() == ISD::TargetGlobalAddress)) {
+               N.getOperand(0).getOpcode() != ISD::TargetGlobalAddress) {
       Base = N.getOperand(0);
     }
     Offset = CurDAG->getTargetConstant(ARM_AM::getAM5Opc(ARM_AM::add, 0),
-                                       MVT::i32);
+                                       SDLoc(N), MVT::i32);
     return true;
   }
 
@@ -892,7 +981,8 @@ bool ARMDAGToDAGISel::SelectAddrMode5(SDValue N,
     Base = N.getOperand(0);
     if (Base.getOpcode() == ISD::FrameIndex) {
       int FI = cast<FrameIndexSDNode>(Base)->getIndex();
-      Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
+      Base = CurDAG->getTargetFrameIndex(
+          FI, TLI->getPointerTy(CurDAG->getDataLayout()));
     }
 
     ARM_AM::AddrOpc AddSub = ARM_AM::add;
@@ -901,13 +991,13 @@ bool ARMDAGToDAGISel::SelectAddrMode5(SDValue N,
       RHSC = -RHSC;
     }
     Offset = CurDAG->getTargetConstant(ARM_AM::getAM5Opc(AddSub, RHSC),
-                                       MVT::i32);
+                                       SDLoc(N), MVT::i32);
     return true;
   }
 
   Base = N;
   Offset = CurDAG->getTargetConstant(ARM_AM::getAM5Opc(ARM_AM::add, 0),
-                                     MVT::i32);
+                                     SDLoc(N), MVT::i32);
   return true;
 }
 
@@ -916,21 +1006,27 @@ bool ARMDAGToDAGISel::SelectAddrMode6(SDNode *Parent, SDValue N, SDValue &Addr,
   Addr = N;
 
   unsigned Alignment = 0;
-  if (LSBaseSDNode *LSN = dyn_cast<LSBaseSDNode>(Parent)) {
+
+  MemSDNode *MemN = cast<MemSDNode>(Parent);
+
+  if (isa<LSBaseSDNode>(MemN) ||
+      ((MemN->getOpcode() == ARMISD::VST1_UPD ||
+        MemN->getOpcode() == ARMISD::VLD1_UPD) &&
+       MemN->getConstantOperandVal(MemN->getNumOperands() - 1) == 1)) {
     // This case occurs only for VLD1-lane/dup and VST1-lane instructions.
     // The maximum alignment is equal to the memory size being referenced.
-    unsigned LSNAlign = LSN->getAlignment();
-    unsigned MemSize = LSN->getMemoryVT().getSizeInBits() / 8;
-    if (LSNAlign >= MemSize && MemSize > 1)
+    unsigned MMOAlign = MemN->getAlignment();
+    unsigned MemSize = MemN->getMemoryVT().getSizeInBits() / 8;
+    if (MMOAlign >= MemSize && MemSize > 1)
       Alignment = MemSize;
   } else {
     // All other uses of addrmode6 are for intrinsics.  For now just record
     // the raw alignment value; it will be refined later based on the legal
     // alignment operands for the intrinsic.
-    Alignment = cast<MemIntrinsicSDNode>(Parent)->getAlignment();
+    Alignment = MemN->getAlignment();
   }
 
-  Align = CurDAG->getTargetConstant(Alignment, MVT::i32);
+  Align = CurDAG->getTargetConstant(Alignment, SDLoc(N), MVT::i32);
   return true;
 }
 
@@ -954,7 +1050,7 @@ bool ARMDAGToDAGISel::SelectAddrModePC(SDValue N,
     Offset = N.getOperand(0);
     SDValue N1 = N.getOperand(1);
     Label = CurDAG->getTargetConstant(cast<ConstantSDNode>(N1)->getZExtValue(),
-                                      MVT::i32);
+                                      SDLoc(N), MVT::i32);
     return true;
   }
 
@@ -982,102 +1078,20 @@ bool ARMDAGToDAGISel::SelectThumbAddrModeRR(SDValue N,
   return true;
 }
 
-bool
-ARMDAGToDAGISel::SelectThumbAddrModeRI(SDValue N, SDValue &Base,
-                                       SDValue &Offset, unsigned Scale) {
-  if (Scale == 4) {
-    SDValue TmpBase, TmpOffImm;
-    if (SelectThumbAddrModeSP(N, TmpBase, TmpOffImm))
-      return false;  // We want to select tLDRspi / tSTRspi instead.
-
-    if (N.getOpcode() == ARMISD::Wrapper &&
-        N.getOperand(0).getOpcode() == ISD::TargetConstantPool)
-      return false;  // We want to select tLDRpci instead.
-  }
-
-  if (!CurDAG->isBaseWithConstantOffset(N))
-    return false;
-
-  // Thumb does not have [sp, r] address mode.
-  RegisterSDNode *LHSR = dyn_cast<RegisterSDNode>(N.getOperand(0));
-  RegisterSDNode *RHSR = dyn_cast<RegisterSDNode>(N.getOperand(1));
-  if ((LHSR && LHSR->getReg() == ARM::SP) ||
-      (RHSR && RHSR->getReg() == ARM::SP))
-    return false;
-
-  // FIXME: Why do we explicitly check for a match here and then return false?
-  // Presumably to allow something else to match, but shouldn't this be
-  // documented?
-  int RHSC;
-  if (isScaledConstantInRange(N.getOperand(1), Scale, 0, 32, RHSC))
-    return false;
-
-  Base = N.getOperand(0);
-  Offset = N.getOperand(1);
-  return true;
-}
-
-bool
-ARMDAGToDAGISel::SelectThumbAddrModeRI5S1(SDValue N,
-                                          SDValue &Base,
-                                          SDValue &Offset) {
-  return SelectThumbAddrModeRI(N, Base, Offset, 1);
-}
-
-bool
-ARMDAGToDAGISel::SelectThumbAddrModeRI5S2(SDValue N,
-                                          SDValue &Base,
-                                          SDValue &Offset) {
-  return SelectThumbAddrModeRI(N, Base, Offset, 2);
-}
-
-bool
-ARMDAGToDAGISel::SelectThumbAddrModeRI5S4(SDValue N,
-                                          SDValue &Base,
-                                          SDValue &Offset) {
-  return SelectThumbAddrModeRI(N, Base, Offset, 4);
-}
-
 bool
 ARMDAGToDAGISel::SelectThumbAddrModeImm5S(SDValue N, unsigned Scale,
                                           SDValue &Base, SDValue &OffImm) {
-  if (Scale == 4) {
-    SDValue TmpBase, TmpOffImm;
-    if (SelectThumbAddrModeSP(N, TmpBase, TmpOffImm))
-      return false;  // We want to select tLDRspi / tSTRspi instead.
-
-    if (N.getOpcode() == ARMISD::Wrapper &&
-        N.getOperand(0).getOpcode() == ISD::TargetConstantPool)
-      return false;  // We want to select tLDRpci instead.
-  }
-
   if (!CurDAG->isBaseWithConstantOffset(N)) {
-    if (N.getOpcode() == ARMISD::Wrapper &&
-        !(Subtarget->useMovt() &&
-          N.getOperand(0).getOpcode() == ISD::TargetGlobalAddress)) {
+    if (N.getOpcode() == ISD::ADD) {
+      return false; // We want to select register offset instead
+    } else if (N.getOpcode() == ARMISD::Wrapper &&
+               N.getOperand(0).getOpcode() != ISD::TargetGlobalAddress) {
       Base = N.getOperand(0);
     } else {
       Base = N;
     }
 
-    OffImm = CurDAG->getTargetConstant(0, MVT::i32);
-    return true;
-  }
-
-  RegisterSDNode *LHSR = dyn_cast<RegisterSDNode>(N.getOperand(0));
-  RegisterSDNode *RHSR = dyn_cast<RegisterSDNode>(N.getOperand(1));
-  if ((LHSR && LHSR->getReg() == ARM::SP) ||
-      (RHSR && RHSR->getReg() == ARM::SP)) {
-    ConstantSDNode *LHS = dyn_cast<ConstantSDNode>(N.getOperand(0));
-    ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(N.getOperand(1));
-    unsigned LHSC = LHS ? LHS->getZExtValue() : 0;
-    unsigned RHSC = RHS ? RHS->getZExtValue() : 0;
-
-    // Thumb does not have [sp, #imm5] address mode for non-zero imm5.
-    if (LHSC != 0 || RHSC != 0) return false;
-
-    Base = N;
-    OffImm = CurDAG->getTargetConstant(0, MVT::i32);
+    OffImm = CurDAG->getTargetConstant(0, SDLoc(N), MVT::i32);
     return true;
   }
 
@@ -1085,13 +1099,12 @@ ARMDAGToDAGISel::SelectThumbAddrModeImm5S(SDValue N, unsigned Scale,
   int RHSC;
   if (isScaledConstantInRange(N.getOperand(1), Scale, 0, 32, RHSC)) {
     Base = N.getOperand(0);
-    OffImm = CurDAG->getTargetConstant(RHSC, MVT::i32);
+    OffImm = CurDAG->getTargetConstant(RHSC, SDLoc(N), MVT::i32);
     return true;
   }
 
-  Base = N.getOperand(0);
-  OffImm = CurDAG->getTargetConstant(0, MVT::i32);
-  return true;
+  // Offset is too large, so use register offset instead.
+  return false;
 }
 
 bool
@@ -1116,8 +1129,14 @@ bool ARMDAGToDAGISel::SelectThumbAddrModeSP(SDValue N,
                                             SDValue &Base, SDValue &OffImm) {
   if (N.getOpcode() == ISD::FrameIndex) {
     int FI = cast<FrameIndexSDNode>(N)->getIndex();
-    Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
-    OffImm = CurDAG->getTargetConstant(0, MVT::i32);
+    // Only multiples of 4 are allowed for the offset, so the frame object
+    // alignment must be at least 4.
+    MachineFrameInfo *MFI = MF->getFrameInfo();
+    if (MFI->getObjectAlignment(FI) < 4)
+      MFI->setObjectAlignment(FI, 4);
+    Base = CurDAG->getTargetFrameIndex(
+        FI, TLI->getPointerTy(CurDAG->getDataLayout()));
+    OffImm = CurDAG->getTargetConstant(0, SDLoc(N), MVT::i32);
     return true;
   }
 
@@ -1133,9 +1152,15 @@ bool ARMDAGToDAGISel::SelectThumbAddrModeSP(SDValue N,
       Base = N.getOperand(0);
       if (Base.getOpcode() == ISD::FrameIndex) {
         int FI = cast<FrameIndexSDNode>(Base)->getIndex();
-        Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
+        // For LHS+RHS to result in an offset that's a multiple of 4 the object
+        // indexed by the LHS must be 4-byte aligned.
+        MachineFrameInfo *MFI = MF->getFrameInfo();
+        if (MFI->getObjectAlignment(FI) < 4)
+          MFI->setObjectAlignment(FI, 4);
+        Base = CurDAG->getTargetFrameIndex(
+            FI, TLI->getPointerTy(CurDAG->getDataLayout()));
       }
-      OffImm = CurDAG->getTargetConstant(RHSC, MVT::i32);
+      OffImm = CurDAG->getTargetConstant(RHSC, SDLoc(N), MVT::i32);
       return true;
     }
   }
@@ -1149,28 +1174,6 @@ bool ARMDAGToDAGISel::SelectThumbAddrModeSP(SDValue N,
 //===----------------------------------------------------------------------===//
 
 
-bool ARMDAGToDAGISel::SelectT2ShifterOperandReg(SDValue N, SDValue &BaseReg,
-                                                SDValue &Opc) {
-  if (DisableShifterOp)
-    return false;
-
-  ARM_AM::ShiftOpc ShOpcVal = ARM_AM::getShiftOpcForNode(N.getOpcode());
-
-  // Don't match base register only case. That is matched to a separate
-  // lower complexity pattern with explicit register operand.
-  if (ShOpcVal == ARM_AM::no_shift) return false;
-
-  BaseReg = N.getOperand(0);
-  unsigned ShImmVal = 0;
-  if (ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(N.getOperand(1))) {
-    ShImmVal = RHS->getZExtValue() & 31;
-    Opc = getI32Imm(ARM_AM::getSORegOpc(ShOpcVal, ShImmVal));
-    return true;
-  }
-
-  return false;
-}
-
 bool ARMDAGToDAGISel::SelectT2AddrModeImm12(SDValue N,
                                             SDValue &Base, SDValue &OffImm) {
   // Match simple R + imm12 operands.
@@ -1181,20 +1184,20 @@ bool ARMDAGToDAGISel::SelectT2AddrModeImm12(SDValue N,
     if (N.getOpcode() == ISD::FrameIndex) {
       // Match frame index.
       int FI = cast<FrameIndexSDNode>(N)->getIndex();
-      Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
-      OffImm  = CurDAG->getTargetConstant(0, MVT::i32);
+      Base = CurDAG->getTargetFrameIndex(
+          FI, TLI->getPointerTy(CurDAG->getDataLayout()));
+      OffImm  = CurDAG->getTargetConstant(0, SDLoc(N), MVT::i32);
       return true;
     }
 
     if (N.getOpcode() == ARMISD::Wrapper &&
-               !(Subtarget->useMovt() &&
-                 N.getOperand(0).getOpcode() == ISD::TargetGlobalAddress)) {
+        N.getOperand(0).getOpcode() != ISD::TargetGlobalAddress) {
       Base = N.getOperand(0);
       if (Base.getOpcode() == ISD::TargetConstantPool)
         return false;  // We want to select t2LDRpci instead.
     } else
       Base = N;
-    OffImm  = CurDAG->getTargetConstant(0, MVT::i32);
+    OffImm  = CurDAG->getTargetConstant(0, SDLoc(N), MVT::i32);
     return true;
   }
 
@@ -1211,16 +1214,17 @@ bool ARMDAGToDAGISel::SelectT2AddrModeImm12(SDValue N,
       Base   = N.getOperand(0);
       if (Base.getOpcode() == ISD::FrameIndex) {
         int FI = cast<FrameIndexSDNode>(Base)->getIndex();
-        Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
+        Base = CurDAG->getTargetFrameIndex(
+            FI, TLI->getPointerTy(CurDAG->getDataLayout()));
       }
-      OffImm = CurDAG->getTargetConstant(RHSC, MVT::i32);
+      OffImm = CurDAG->getTargetConstant(RHSC, SDLoc(N), MVT::i32);
       return true;
     }
   }
 
   // Base only.
   Base = N;
-  OffImm  = CurDAG->getTargetConstant(0, MVT::i32);
+  OffImm  = CurDAG->getTargetConstant(0, SDLoc(N), MVT::i32);
   return true;
 }
 
@@ -1240,9 +1244,10 @@ bool ARMDAGToDAGISel::SelectT2AddrModeImm8(SDValue N,
       Base = N.getOperand(0);
       if (Base.getOpcode() == ISD::FrameIndex) {
         int FI = cast<FrameIndexSDNode>(Base)->getIndex();
-        Base = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
+        Base = CurDAG->getTargetFrameIndex(
+            FI, TLI->getPointerTy(CurDAG->getDataLayout()));
       }
-      OffImm = CurDAG->getTargetConstant(RHSC, MVT::i32);
+      OffImm = CurDAG->getTargetConstant(RHSC, SDLoc(N), MVT::i32);
       return true;
     }
   }
@@ -1259,8 +1264,8 @@ bool ARMDAGToDAGISel::SelectT2AddrModeImm8Offset(SDNode *Op, SDValue N,
   int RHSC;
   if (isScaledConstantInRange(N, /*Scale=*/1, 0, 0x100, RHSC)) { // 8 bits.
     OffImm = ((AM == ISD::PRE_INC) || (AM == ISD::POST_INC))
-      ? CurDAG->getTargetConstant(RHSC, MVT::i32)
-      : CurDAG->getTargetConstant(-RHSC, MVT::i32);
+      ? CurDAG->getTargetConstant(RHSC, SDLoc(N), MVT::i32)
+      : CurDAG->getTargetConstant(-RHSC, SDLoc(N), MVT::i32);
     return true;
   }
 
@@ -1305,30 +1310,56 @@ bool ARMDAGToDAGISel::SelectT2AddrModeSoReg(SDValue N,
         OffReg = OffReg.getOperand(0);
       else {
         ShAmt = 0;
-        ShOpcVal = ARM_AM::no_shift;
       }
-    } else {
-      ShOpcVal = ARM_AM::no_shift;
     }
   }
 
-  ShImm = CurDAG->getTargetConstant(ShAmt, MVT::i32);
+  ShImm = CurDAG->getTargetConstant(ShAmt, SDLoc(N), MVT::i32);
+
+  return true;
+}
+
+bool ARMDAGToDAGISel::SelectT2AddrModeExclusive(SDValue N, SDValue &Base,
+                                                SDValue &OffImm) {
+  // This *must* succeed since it's used for the irreplaceable ldrex and strex
+  // instructions.
+  Base = N;
+  OffImm = CurDAG->getTargetConstant(0, SDLoc(N), MVT::i32);
+
+  if (N.getOpcode() != ISD::ADD || !CurDAG->isBaseWithConstantOffset(N))
+    return true;
+
+  ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(N.getOperand(1));
+  if (!RHS)
+    return true;
+
+  uint32_t RHSC = (int)RHS->getZExtValue();
+  if (RHSC > 1020 || RHSC % 4 != 0)
+    return true;
+
+  Base = N.getOperand(0);
+  if (Base.getOpcode() == ISD::FrameIndex) {
+    int FI = cast<FrameIndexSDNode>(Base)->getIndex();
+    Base = CurDAG->getTargetFrameIndex(
+        FI, TLI->getPointerTy(CurDAG->getDataLayout()));
+  }
 
+  OffImm = CurDAG->getTargetConstant(RHSC/4, SDLoc(N), MVT::i32);
   return true;
 }
 
 //===--------------------------------------------------------------------===//
 
 /// getAL - Returns a ARMCC::AL immediate node.
-static inline SDValue getAL(SelectionDAG *CurDAG) {
-  return CurDAG->getTargetConstant((uint64_t)ARMCC::AL, MVT::i32);
+static inline SDValue getAL(SelectionDAG *CurDAG, SDLoc dl) {
+  return CurDAG->getTargetConstant((uint64_t)ARMCC::AL, dl, MVT::i32);
 }
 
 SDNode *ARMDAGToDAGISel::SelectARMIndexedLoad(SDNode *N) {
   LoadSDNode *LD = cast<LoadSDNode>(N);
   ISD::MemIndexedMode AM = LD->getAddressingMode();
   if (AM == ISD::UNINDEXED)
-    return NULL;
+    return nullptr;
 
   EVT LoadedVT = LD->getMemoryVT();
   SDValue Offset, AMOpc;
@@ -1380,28 +1411,28 @@ SDNode *ARMDAGToDAGISel::SelectARMIndexedLoad(SDNode *N) {
     if (Opcode == ARM::LDR_PRE_IMM || Opcode == ARM::LDRB_PRE_IMM) {
       SDValue Chain = LD->getChain();
       SDValue Base = LD->getBasePtr();
-      SDValue Ops[]= { Base, AMOpc, getAL(CurDAG),
+      SDValue Ops[]= { Base, AMOpc, getAL(CurDAG, SDLoc(N)),
                        CurDAG->getRegister(0, MVT::i32), Chain };
-      return CurDAG->getMachineNode(Opcode, N->getDebugLoc(), MVT::i32,
-                                    MVT::i32, MVT::Other, Ops, 5);
+      return CurDAG->getMachineNode(Opcode, SDLoc(N), MVT::i32,
+                                    MVT::i32, MVT::Other, Ops);
     } else {
       SDValue Chain = LD->getChain();
       SDValue Base = LD->getBasePtr();
-      SDValue Ops[]= { Base, Offset, AMOpc, getAL(CurDAG),
+      SDValue Ops[]= { Base, Offset, AMOpc, getAL(CurDAG, SDLoc(N)),
                        CurDAG->getRegister(0, MVT::i32), Chain };
-      return CurDAG->getMachineNode(Opcode, N->getDebugLoc(), MVT::i32,
-                                    MVT::i32, MVT::Other, Ops, 6);
+      return CurDAG->getMachineNode(Opcode, SDLoc(N), MVT::i32,
+                                    MVT::i32, MVT::Other, Ops);
     }
   }
 
-  return NULL;
+  return nullptr;
 }
 
 SDNode *ARMDAGToDAGISel::SelectT2IndexedLoad(SDNode *N) {
   LoadSDNode *LD = cast<LoadSDNode>(N);
   ISD::MemIndexedMode AM = LD->getAddressingMode();
   if (AM == ISD::UNINDEXED)
-    return NULL;
+    return nullptr;
 
   EVT LoadedVT = LD->getMemoryVT();
   bool isSExtLd = LD->getExtensionType() == ISD::SEXTLOAD;
@@ -1428,7 +1459,7 @@ SDNode *ARMDAGToDAGISel::SelectT2IndexedLoad(SDNode *N) {
         Opcode = isPre ? ARM::t2LDRB_PRE : ARM::t2LDRB_POST;
       break;
     default:
-      return NULL;
+      return nullptr;
     }
     Match = true;
   }
@@ -1436,105 +1467,109 @@ SDNode *ARMDAGToDAGISel::SelectT2IndexedLoad(SDNode *N) {
   if (Match) {
     SDValue Chain = LD->getChain();
     SDValue Base = LD->getBasePtr();
-    SDValue Ops[]= { Base, Offset, getAL(CurDAG),
+    SDValue Ops[]= { Base, Offset, getAL(CurDAG, SDLoc(N)),
                      CurDAG->getRegister(0, MVT::i32), Chain };
-    return CurDAG->getMachineNode(Opcode, N->getDebugLoc(), MVT::i32, MVT::i32,
-                                  MVT::Other, Ops, 5);
+    return CurDAG->getMachineNode(Opcode, SDLoc(N), MVT::i32, MVT::i32,
+                                  MVT::Other, Ops);
   }
 
-  return NULL;
+  return nullptr;
 }
 
 /// \brief Form a GPRPair pseudo register from a pair of GPR regs.
 SDNode *ARMDAGToDAGISel::createGPRPairNode(EVT VT, SDValue V0, SDValue V1) {
-  DebugLoc dl = V0.getNode()->getDebugLoc();
+  SDLoc dl(V0.getNode());
   SDValue RegClass =
-    CurDAG->getTargetConstant(ARM::GPRPairRegClassID, MVT::i32);
-  SDValue SubReg0 = CurDAG->getTargetConstant(ARM::gsub_0, MVT::i32);
-  SDValue SubReg1 = CurDAG->getTargetConstant(ARM::gsub_1, MVT::i32);
+    CurDAG->getTargetConstant(ARM::GPRPairRegClassID, dl, MVT::i32);
+  SDValue SubReg0 = CurDAG->getTargetConstant(ARM::gsub_0, dl, MVT::i32);
+  SDValue SubReg1 = CurDAG->getTargetConstant(ARM::gsub_1, dl, MVT::i32);
   const SDValue Ops[] = { RegClass, V0, SubReg0, V1, SubReg1 };
-  return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 5);
+  return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops);
 }
 
 /// \brief Form a D register from a pair of S registers.
 SDNode *ARMDAGToDAGISel::createSRegPairNode(EVT VT, SDValue V0, SDValue V1) {
-  DebugLoc dl = V0.getNode()->getDebugLoc();
+  SDLoc dl(V0.getNode());
   SDValue RegClass =
-    CurDAG->getTargetConstant(ARM::DPR_VFP2RegClassID, MVT::i32);
-  SDValue SubReg0 = CurDAG->getTargetConstant(ARM::ssub_0, MVT::i32);
-  SDValue SubReg1 = CurDAG->getTargetConstant(ARM::ssub_1, MVT::i32);
+    CurDAG->getTargetConstant(ARM::DPR_VFP2RegClassID, dl, MVT::i32);
+  SDValue SubReg0 = CurDAG->getTargetConstant(ARM::ssub_0, dl, MVT::i32);
+  SDValue SubReg1 = CurDAG->getTargetConstant(ARM::ssub_1, dl, MVT::i32);
   const SDValue Ops[] = { RegClass, V0, SubReg0, V1, SubReg1 };
-  return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 5);
+  return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops);
 }
 
 /// \brief Form a quad register from a pair of D registers.
 SDNode *ARMDAGToDAGISel::createDRegPairNode(EVT VT, SDValue V0, SDValue V1) {
-  DebugLoc dl = V0.getNode()->getDebugLoc();
-  SDValue RegClass = CurDAG->getTargetConstant(ARM::QPRRegClassID, MVT::i32);
-  SDValue SubReg0 = CurDAG->getTargetConstant(ARM::dsub_0, MVT::i32);
-  SDValue SubReg1 = CurDAG->getTargetConstant(ARM::dsub_1, MVT::i32);
+  SDLoc dl(V0.getNode());
+  SDValue RegClass = CurDAG->getTargetConstant(ARM::QPRRegClassID, dl,
+                                               MVT::i32);
+  SDValue SubReg0 = CurDAG->getTargetConstant(ARM::dsub_0, dl, MVT::i32);
+  SDValue SubReg1 = CurDAG->getTargetConstant(ARM::dsub_1, dl, MVT::i32);
   const SDValue Ops[] = { RegClass, V0, SubReg0, V1, SubReg1 };
-  return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 5);
+  return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops);
 }
 
 /// \brief Form 4 consecutive D registers from a pair of Q registers.
 SDNode *ARMDAGToDAGISel::createQRegPairNode(EVT VT, SDValue V0, SDValue V1) {
-  DebugLoc dl = V0.getNode()->getDebugLoc();
-  SDValue RegClass = CurDAG->getTargetConstant(ARM::QQPRRegClassID, MVT::i32);
-  SDValue SubReg0 = CurDAG->getTargetConstant(ARM::qsub_0, MVT::i32);
-  SDValue SubReg1 = CurDAG->getTargetConstant(ARM::qsub_1, MVT::i32);
+  SDLoc dl(V0.getNode());
+  SDValue RegClass = CurDAG->getTargetConstant(ARM::QQPRRegClassID, dl,
+                                               MVT::i32);
+  SDValue SubReg0 = CurDAG->getTargetConstant(ARM::qsub_0, dl, MVT::i32);
+  SDValue SubReg1 = CurDAG->getTargetConstant(ARM::qsub_1, dl, MVT::i32);
   const SDValue Ops[] = { RegClass, V0, SubReg0, V1, SubReg1 };
-  return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 5);
+  return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops);
 }
 
 /// \brief Form 4 consecutive S registers.
 SDNode *ARMDAGToDAGISel::createQuadSRegsNode(EVT VT, SDValue V0, SDValue V1,
                                    SDValue V2, SDValue V3) {
-  DebugLoc dl = V0.getNode()->getDebugLoc();
+  SDLoc dl(V0.getNode());
   SDValue RegClass =
-    CurDAG->getTargetConstant(ARM::QPR_VFP2RegClassID, MVT::i32);
-  SDValue SubReg0 = CurDAG->getTargetConstant(ARM::ssub_0, MVT::i32);
-  SDValue SubReg1 = CurDAG->getTargetConstant(ARM::ssub_1, MVT::i32);
-  SDValue SubReg2 = CurDAG->getTargetConstant(ARM::ssub_2, MVT::i32);
-  SDValue SubReg3 = CurDAG->getTargetConstant(ARM::ssub_3, MVT::i32);
+    CurDAG->getTargetConstant(ARM::QPR_VFP2RegClassID, dl, MVT::i32);
+  SDValue SubReg0 = CurDAG->getTargetConstant(ARM::ssub_0, dl, MVT::i32);
+  SDValue SubReg1 = CurDAG->getTargetConstant(ARM::ssub_1, dl, MVT::i32);
+  SDValue SubReg2 = CurDAG->getTargetConstant(ARM::ssub_2, dl, MVT::i32);
+  SDValue SubReg3 = CurDAG->getTargetConstant(ARM::ssub_3, dl, MVT::i32);
   const SDValue Ops[] = { RegClass, V0, SubReg0, V1, SubReg1,
                                     V2, SubReg2, V3, SubReg3 };
-  return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 9);
+  return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops);
 }
 
 /// \brief Form 4 consecutive D registers.
 SDNode *ARMDAGToDAGISel::createQuadDRegsNode(EVT VT, SDValue V0, SDValue V1,
                                    SDValue V2, SDValue V3) {
-  DebugLoc dl = V0.getNode()->getDebugLoc();
-  SDValue RegClass = CurDAG->getTargetConstant(ARM::QQPRRegClassID, MVT::i32);
-  SDValue SubReg0 = CurDAG->getTargetConstant(ARM::dsub_0, MVT::i32);
-  SDValue SubReg1 = CurDAG->getTargetConstant(ARM::dsub_1, MVT::i32);
-  SDValue SubReg2 = CurDAG->getTargetConstant(ARM::dsub_2, MVT::i32);
-  SDValue SubReg3 = CurDAG->getTargetConstant(ARM::dsub_3, MVT::i32);
+  SDLoc dl(V0.getNode());
+  SDValue RegClass = CurDAG->getTargetConstant(ARM::QQPRRegClassID, dl,
+                                               MVT::i32);
+  SDValue SubReg0 = CurDAG->getTargetConstant(ARM::dsub_0, dl, MVT::i32);
+  SDValue SubReg1 = CurDAG->getTargetConstant(ARM::dsub_1, dl, MVT::i32);
+  SDValue SubReg2 = CurDAG->getTargetConstant(ARM::dsub_2, dl, MVT::i32);
+  SDValue SubReg3 = CurDAG->getTargetConstant(ARM::dsub_3, dl, MVT::i32);
   const SDValue Ops[] = { RegClass, V0, SubReg0, V1, SubReg1,
                                     V2, SubReg2, V3, SubReg3 };
-  return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 9);
+  return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops);
 }
 
 /// \brief Form 4 consecutive Q registers.
 SDNode *ARMDAGToDAGISel::createQuadQRegsNode(EVT VT, SDValue V0, SDValue V1,
                                    SDValue V2, SDValue V3) {
-  DebugLoc dl = V0.getNode()->getDebugLoc();
-  SDValue RegClass = CurDAG->getTargetConstant(ARM::QQQQPRRegClassID, MVT::i32);
-  SDValue SubReg0 = CurDAG->getTargetConstant(ARM::qsub_0, MVT::i32);
-  SDValue SubReg1 = CurDAG->getTargetConstant(ARM::qsub_1, MVT::i32);
-  SDValue SubReg2 = CurDAG->getTargetConstant(ARM::qsub_2, MVT::i32);
-  SDValue SubReg3 = CurDAG->getTargetConstant(ARM::qsub_3, MVT::i32);
+  SDLoc dl(V0.getNode());
+  SDValue RegClass = CurDAG->getTargetConstant(ARM::QQQQPRRegClassID, dl,
+                                               MVT::i32);
+  SDValue SubReg0 = CurDAG->getTargetConstant(ARM::qsub_0, dl, MVT::i32);
+  SDValue SubReg1 = CurDAG->getTargetConstant(ARM::qsub_1, dl, MVT::i32);
+  SDValue SubReg2 = CurDAG->getTargetConstant(ARM::qsub_2, dl, MVT::i32);
+  SDValue SubReg3 = CurDAG->getTargetConstant(ARM::qsub_3, dl, MVT::i32);
   const SDValue Ops[] = { RegClass, V0, SubReg0, V1, SubReg1,
                                     V2, SubReg2, V3, SubReg3 };
-  return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 9);
+  return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops);
 }
 
 /// GetVLDSTAlign - Get the alignment (in bytes) for the alignment operand
 /// of a NEON VLD or VST instruction.  The supported values depend on the
 /// number of registers being loaded.
-SDValue ARMDAGToDAGISel::GetVLDSTAlign(SDValue Align, unsigned NumVecs,
-                                       bool is64BitVector) {
+SDValue ARMDAGToDAGISel::GetVLDSTAlign(SDValue Align, SDLoc dl,
+                                       unsigned NumVecs, bool is64BitVector) {
   unsigned NumRegs = NumVecs;
   if (!is64BitVector && NumVecs < 3)
     NumRegs *= 2;
@@ -1549,12 +1584,64 @@ SDValue ARMDAGToDAGISel::GetVLDSTAlign(SDValue Align, unsigned NumVecs,
   else
     Alignment = 0;
 
-  return CurDAG->getTargetConstant(Alignment, MVT::i32);
+  return CurDAG->getTargetConstant(Alignment, dl, MVT::i32);
+}
+
+static bool isVLDfixed(unsigned Opc)
+{
+  switch (Opc) {
+  default: return false;
+  case ARM::VLD1d8wb_fixed : return true;
+  case ARM::VLD1d16wb_fixed : return true;
+  case ARM::VLD1d64Qwb_fixed : return true;
+  case ARM::VLD1d32wb_fixed : return true;
+  case ARM::VLD1d64wb_fixed : return true;
+  case ARM::VLD1d64TPseudoWB_fixed : return true;
+  case ARM::VLD1d64QPseudoWB_fixed : return true;
+  case ARM::VLD1q8wb_fixed : return true;
+  case ARM::VLD1q16wb_fixed : return true;
+  case ARM::VLD1q32wb_fixed : return true;
+  case ARM::VLD1q64wb_fixed : return true;
+  case ARM::VLD2d8wb_fixed : return true;
+  case ARM::VLD2d16wb_fixed : return true;
+  case ARM::VLD2d32wb_fixed : return true;
+  case ARM::VLD2q8PseudoWB_fixed : return true;
+  case ARM::VLD2q16PseudoWB_fixed : return true;
+  case ARM::VLD2q32PseudoWB_fixed : return true;
+  case ARM::VLD2DUPd8wb_fixed : return true;
+  case ARM::VLD2DUPd16wb_fixed : return true;
+  case ARM::VLD2DUPd32wb_fixed : return true;
+  }
+}
+
+static bool isVSTfixed(unsigned Opc)
+{
+  switch (Opc) {
+  default: return false;
+  case ARM::VST1d8wb_fixed : return true;
+  case ARM::VST1d16wb_fixed : return true;
+  case ARM::VST1d32wb_fixed : return true;
+  case ARM::VST1d64wb_fixed : return true;
+  case ARM::VST1q8wb_fixed : return true;
+  case ARM::VST1q16wb_fixed : return true;
+  case ARM::VST1q32wb_fixed : return true;
+  case ARM::VST1q64wb_fixed : return true;
+  case ARM::VST1d64TPseudoWB_fixed : return true;
+  case ARM::VST1d64QPseudoWB_fixed : return true;
+  case ARM::VST2d8wb_fixed : return true;
+  case ARM::VST2d16wb_fixed : return true;
+  case ARM::VST2d32wb_fixed : return true;
+  case ARM::VST2q8PseudoWB_fixed : return true;
+  case ARM::VST2q16PseudoWB_fixed : return true;
+  case ARM::VST2q32PseudoWB_fixed : return true;
+  }
 }
 
 // Get the register stride update opcode of a VLD/VST instruction that
 // is otherwise equivalent to the given fixed stride updating instruction.
 static unsigned getVLDSTRegisterUpdateOpcode(unsigned Opc) {
+  assert((isVLDfixed(Opc) || isVSTfixed(Opc))
+    && "Incorrect fixed stride updating instruction.");
   switch (Opc) {
   default: break;
   case ARM::VLD1d8wb_fixed: return ARM::VLD1d8wb_register;
@@ -1565,6 +1652,10 @@ static unsigned getVLDSTRegisterUpdateOpcode(unsigned Opc) {
   case ARM::VLD1q16wb_fixed: return ARM::VLD1q16wb_register;
   case ARM::VLD1q32wb_fixed: return ARM::VLD1q32wb_register;
   case ARM::VLD1q64wb_fixed: return ARM::VLD1q64wb_register;
+  case ARM::VLD1d64Twb_fixed: return ARM::VLD1d64Twb_register;
+  case ARM::VLD1d64Qwb_fixed: return ARM::VLD1d64Qwb_register;
+  case ARM::VLD1d64TPseudoWB_fixed: return ARM::VLD1d64TPseudoWB_register;
+  case ARM::VLD1d64QPseudoWB_fixed: return ARM::VLD1d64QPseudoWB_register;
 
   case ARM::VST1d8wb_fixed: return ARM::VST1d8wb_register;
   case ARM::VST1d16wb_fixed: return ARM::VST1d16wb_register;
@@ -1603,17 +1694,17 @@ SDNode *ARMDAGToDAGISel::SelectVLD(SDNode *N, bool isUpdating, unsigned NumVecs,
                                    const uint16_t *QOpcodes0,
                                    const uint16_t *QOpcodes1) {
   assert(NumVecs >= 1 && NumVecs <= 4 && "VLD NumVecs out-of-range");
-  DebugLoc dl = N->getDebugLoc();
+  SDLoc dl(N);
 
   SDValue MemAddr, Align;
   unsigned AddrOpIdx = isUpdating ? 1 : 2;
   if (!SelectAddrMode6(N, N->getOperand(AddrOpIdx), MemAddr, Align))
-    return NULL;
+    return nullptr;
 
   SDValue Chain = N->getOperand(0);
   EVT VT = N->getValueType(0);
   bool is64BitVector = VT.is64BitVector();
-  Align = GetVLDSTAlign(Align, NumVecs, is64BitVector);
+  Align = GetVLDSTAlign(Align, dl, NumVecs, is64BitVector);
 
   unsigned OpcodeIndex;
   switch (VT.getSimpleVT().SimpleTy) {
@@ -1629,6 +1720,7 @@ SDNode *ARMDAGToDAGISel::SelectVLD(SDNode *N, bool isUpdating, unsigned NumVecs,
   case MVT::v8i16: OpcodeIndex = 1; break;
   case MVT::v4f32:
   case MVT::v4i32: OpcodeIndex = 2; break;
+  case MVT::v2f64:
   case MVT::v2i64: OpcodeIndex = 3;
     assert(NumVecs == 1 && "v2i64 type only supported for VLD1");
     break;
@@ -1649,7 +1741,7 @@ SDNode *ARMDAGToDAGISel::SelectVLD(SDNode *N, bool isUpdating, unsigned NumVecs,
     ResTys.push_back(MVT::i32);
   ResTys.push_back(MVT::Other);
 
-  SDValue Pred = getAL(CurDAG);
+  SDValue Pred = getAL(CurDAG, dl);
   SDValue Reg0 = CurDAG->getRegister(0, MVT::i32);
   SDNode *VLd;
   SmallVector<SDValue, 7> Ops;
@@ -1664,18 +1756,18 @@ SDNode *ARMDAGToDAGISel::SelectVLD(SDNode *N, bool isUpdating, unsigned NumVecs,
       SDValue Inc = N->getOperand(AddrOpIdx + 1);
       // FIXME: VLD1/VLD2 fixed increment doesn't need Reg0. Remove the reg0
       // case entirely when the rest are updated to that form, too.
-      if ((NumVecs == 1 || NumVecs == 2) && !isa<ConstantSDNode>(Inc.getNode()))
+      if ((NumVecs <= 2) && !isa<ConstantSDNode>(Inc.getNode()))
         Opc = getVLDSTRegisterUpdateOpcode(Opc);
-      // We use a VLD1 for v1i64 even if the pseudo says vld2/3/4, so
+      // FIXME: We use a VLD1 for v1i64 even if the pseudo says vld2/3/4, so
       // check for that explicitly too. Horribly hacky, but temporary.
-      if ((NumVecs != 1 && NumVecs != 2 && Opc != ARM::VLD1q64wb_fixed) ||
+      if ((NumVecs > 2 && !isVLDfixed(Opc)) ||
           !isa<ConstantSDNode>(Inc.getNode()))
         Ops.push_back(isa<ConstantSDNode>(Inc.getNode()) ? Reg0 : Inc);
     }
     Ops.push_back(Pred);
     Ops.push_back(Reg0);
     Ops.push_back(Chain);
-    VLd = CurDAG->getMachineNode(Opc, dl, ResTys, Ops.data(), Ops.size());
+    VLd = CurDAG->getMachineNode(Opc, dl, ResTys, Ops);
 
   } else {
     // Otherwise, quad registers are loaded with two separate instructions,
@@ -1688,7 +1780,7 @@ SDNode *ARMDAGToDAGISel::SelectVLD(SDNode *N, bool isUpdating, unsigned NumVecs,
       SDValue(CurDAG->getMachineNode(TargetOpcode::IMPLICIT_DEF, dl, ResTy), 0);
     const SDValue OpsA[] = { MemAddr, Align, Reg0, ImplDef, Pred, Reg0, Chain };
     SDNode *VLdA = CurDAG->getMachineNode(QOpcodes0[OpcodeIndex], dl,
-                                          ResTy, AddrTy, MVT::Other, OpsA, 7);
+                                          ResTy, AddrTy, MVT::Other, OpsA);
     Chain = SDValue(VLdA, 2);
 
     // Load the odd subregs.
@@ -1705,8 +1797,7 @@ SDNode *ARMDAGToDAGISel::SelectVLD(SDNode *N, bool isUpdating, unsigned NumVecs,
     Ops.push_back(Pred);
     Ops.push_back(Reg0);
     Ops.push_back(Chain);
-    VLd = CurDAG->getMachineNode(QOpcodes1[OpcodeIndex], dl, ResTys,
-                                 Ops.data(), Ops.size());
+    VLd = CurDAG->getMachineNode(QOpcodes1[OpcodeIndex], dl, ResTys, Ops);
   }
 
   // Transfer memoperands.
@@ -1728,7 +1819,7 @@ SDNode *ARMDAGToDAGISel::SelectVLD(SDNode *N, bool isUpdating, unsigned NumVecs,
   ReplaceUses(SDValue(N, NumVecs), SDValue(VLd, 1));
   if (isUpdating)
     ReplaceUses(SDValue(N, NumVecs + 1), SDValue(VLd, 2));
-  return NULL;
+  return nullptr;
 }
 
 SDNode *ARMDAGToDAGISel::SelectVST(SDNode *N, bool isUpdating, unsigned NumVecs,
@@ -1736,13 +1827,13 @@ SDNode *ARMDAGToDAGISel::SelectVST(SDNode *N, bool isUpdating, unsigned NumVecs,
                                    const uint16_t *QOpcodes0,
                                    const uint16_t *QOpcodes1) {
   assert(NumVecs >= 1 && NumVecs <= 4 && "VST NumVecs out-of-range");
-  DebugLoc dl = N->getDebugLoc();
+  SDLoc dl(N);
 
   SDValue MemAddr, Align;
   unsigned AddrOpIdx = isUpdating ? 1 : 2;
   unsigned Vec0Idx = 3; // AddrOpIdx + (isUpdating ? 2 : 1)
   if (!SelectAddrMode6(N, N->getOperand(AddrOpIdx), MemAddr, Align))
-    return NULL;
+    return nullptr;
 
   MachineSDNode::mmo_iterator MemOp = MF->allocateMemRefsArray(1);
   MemOp[0] = cast<MemIntrinsicSDNode>(N)->getMemOperand();
@@ -1750,7 +1841,7 @@ SDNode *ARMDAGToDAGISel::SelectVST(SDNode *N, bool isUpdating, unsigned NumVecs,
   SDValue Chain = N->getOperand(0);
   EVT VT = N->getOperand(Vec0Idx).getValueType();
   bool is64BitVector = VT.is64BitVector();
-  Align = GetVLDSTAlign(Align, NumVecs, is64BitVector);
+  Align = GetVLDSTAlign(Align, dl, NumVecs, is64BitVector);
 
   unsigned OpcodeIndex;
   switch (VT.getSimpleVT().SimpleTy) {
@@ -1766,6 +1857,7 @@ SDNode *ARMDAGToDAGISel::SelectVST(SDNode *N, bool isUpdating, unsigned NumVecs,
   case MVT::v8i16: OpcodeIndex = 1; break;
   case MVT::v4f32:
   case MVT::v4i32: OpcodeIndex = 2; break;
+  case MVT::v2f64:
   case MVT::v2i64: OpcodeIndex = 3;
     assert(NumVecs == 1 && "v2i64 type only supported for VST1");
     break;
@@ -1776,7 +1868,7 @@ SDNode *ARMDAGToDAGISel::SelectVST(SDNode *N, bool isUpdating, unsigned NumVecs,
     ResTys.push_back(MVT::i32);
   ResTys.push_back(MVT::Other);
 
-  SDValue Pred = getAL(CurDAG);
+  SDValue Pred = getAL(CurDAG, dl);
   SDValue Reg0 = CurDAG->getRegister(0, MVT::i32);
   SmallVector<SDValue, 7> Ops;
 
@@ -1817,18 +1909,18 @@ SDNode *ARMDAGToDAGISel::SelectVST(SDNode *N, bool isUpdating, unsigned NumVecs,
       // case entirely when the rest are updated to that form, too.
       if (NumVecs <= 2 && !isa<ConstantSDNode>(Inc.getNode()))
         Opc = getVLDSTRegisterUpdateOpcode(Opc);
-      // We use a VST1 for v1i64 even if the pseudo says vld2/3/4, so
+      // FIXME: We use a VST1 for v1i64 even if the pseudo says vld2/3/4, so
       // check for that explicitly too. Horribly hacky, but temporary.
-      if ((NumVecs > 2 && Opc != ARM::VST1q64wb_fixed) ||
-          !isa<ConstantSDNode>(Inc.getNode()))
-        Ops.push_back(isa<ConstantSDNode>(Inc.getNode()) ? Reg0 : Inc);
+      if  (!isa<ConstantSDNode>(Inc.getNode()))
+        Ops.push_back(Inc);
+      else if (NumVecs > 2 && !isVSTfixed(Opc))
+        Ops.push_back(Reg0);
     }
     Ops.push_back(SrcReg);
     Ops.push_back(Pred);
     Ops.push_back(Reg0);
     Ops.push_back(Chain);
-    SDNode *VSt =
-      CurDAG->getMachineNode(Opc, dl, ResTys, Ops.data(), Ops.size());
+    SDNode *VSt = CurDAG->getMachineNode(Opc, dl, ResTys, Ops);
 
     // Transfer memoperands.
     cast<MachineSDNode>(VSt)->setMemRefs(MemOp, MemOp + 1);
@@ -1853,7 +1945,7 @@ SDNode *ARMDAGToDAGISel::SelectVST(SDNode *N, bool isUpdating, unsigned NumVecs,
   const SDValue OpsA[] = { MemAddr, Align, Reg0, RegSeq, Pred, Reg0, Chain };
   SDNode *VStA = CurDAG->getMachineNode(QOpcodes0[OpcodeIndex], dl,
                                         MemAddr.getValueType(),
-                                        MVT::Other, OpsA, 7);
+                                        MVT::Other, OpsA);
   cast<MachineSDNode>(VStA)->setMemRefs(MemOp, MemOp + 1);
   Chain = SDValue(VStA, 1);
 
@@ -1872,7 +1964,7 @@ SDNode *ARMDAGToDAGISel::SelectVST(SDNode *N, bool isUpdating, unsigned NumVecs,
   Ops.push_back(Reg0);
   Ops.push_back(Chain);
   SDNode *VStB = CurDAG->getMachineNode(QOpcodes1[OpcodeIndex], dl, ResTys,
-                                        Ops.data(), Ops.size());
+                                        Ops);
   cast<MachineSDNode>(VStB)->setMemRefs(MemOp, MemOp + 1);
   return VStB;
 }
@@ -1882,13 +1974,13 @@ SDNode *ARMDAGToDAGISel::SelectVLDSTLane(SDNode *N, bool IsLoad,
                                          const uint16_t *DOpcodes,
                                          const uint16_t *QOpcodes) {
   assert(NumVecs >=2 && NumVecs <= 4 && "VLDSTLane NumVecs out-of-range");
-  DebugLoc dl = N->getDebugLoc();
+  SDLoc dl(N);
 
   SDValue MemAddr, Align;
   unsigned AddrOpIdx = isUpdating ? 1 : 2;
   unsigned Vec0Idx = 3; // AddrOpIdx + (isUpdating ? 2 : 1)
   if (!SelectAddrMode6(N, N->getOperand(AddrOpIdx), MemAddr, Align))
-    return NULL;
+    return nullptr;
 
   MachineSDNode::mmo_iterator MemOp = MF->allocateMemRefsArray(1);
   MemOp[0] = cast<MemIntrinsicSDNode>(N)->getMemOperand();
@@ -1912,7 +2004,7 @@ SDNode *ARMDAGToDAGISel::SelectVLDSTLane(SDNode *N, bool IsLoad,
     if (Alignment == 1)
       Alignment = 0;
   }
-  Align = CurDAG->getTargetConstant(Alignment, MVT::i32);
+  Align = CurDAG->getTargetConstant(Alignment, dl, MVT::i32);
 
   unsigned OpcodeIndex;
   switch (VT.getSimpleVT().SimpleTy) {
@@ -1940,7 +2032,7 @@ SDNode *ARMDAGToDAGISel::SelectVLDSTLane(SDNode *N, bool IsLoad,
     ResTys.push_back(MVT::i32);
   ResTys.push_back(MVT::Other);
 
-  SDValue Pred = getAL(CurDAG);
+  SDValue Pred = getAL(CurDAG, dl);
   SDValue Reg0 = CurDAG->getRegister(0, MVT::i32);
 
   SmallVector<SDValue, 8> Ops;
@@ -1970,15 +2062,14 @@ SDNode *ARMDAGToDAGISel::SelectVLDSTLane(SDNode *N, bool IsLoad,
       SuperReg = SDValue(createQuadQRegsNode(MVT::v8i64, V0, V1, V2, V3), 0);
   }
   Ops.push_back(SuperReg);
-  Ops.push_back(getI32Imm(Lane));
+  Ops.push_back(getI32Imm(Lane, dl));
   Ops.push_back(Pred);
   Ops.push_back(Reg0);
   Ops.push_back(Chain);
 
   unsigned Opc = (is64BitVector ? DOpcodes[OpcodeIndex] :
                                   QOpcodes[OpcodeIndex]);
-  SDNode *VLdLn = CurDAG->getMachineNode(Opc, dl, ResTys,
-                                         Ops.data(), Ops.size());
+  SDNode *VLdLn = CurDAG->getMachineNode(Opc, dl, ResTys, Ops);
   cast<MachineSDNode>(VLdLn)->setMemRefs(MemOp, MemOp + 1);
   if (!IsLoad)
     return VLdLn;
@@ -1994,18 +2085,18 @@ SDNode *ARMDAGToDAGISel::SelectVLDSTLane(SDNode *N, bool IsLoad,
   ReplaceUses(SDValue(N, NumVecs), SDValue(VLdLn, 1));
   if (isUpdating)
     ReplaceUses(SDValue(N, NumVecs + 1), SDValue(VLdLn, 2));
-  return NULL;
+  return nullptr;
 }
 
 SDNode *ARMDAGToDAGISel::SelectVLDDup(SDNode *N, bool isUpdating,
                                       unsigned NumVecs,
                                       const uint16_t *Opcodes) {
   assert(NumVecs >=2 && NumVecs <= 4 && "VLDDup NumVecs out-of-range");
-  DebugLoc dl = N->getDebugLoc();
+  SDLoc dl(N);
 
   SDValue MemAddr, Align;
   if (!SelectAddrMode6(N, N->getOperand(1), MemAddr, Align))
-    return NULL;
+    return nullptr;
 
   MachineSDNode::mmo_iterator MemOp = MF->allocateMemRefsArray(1);
   MemOp[0] = cast<MemIntrinsicSDNode>(N)->getMemOperand();
@@ -2026,7 +2117,7 @@ SDNode *ARMDAGToDAGISel::SelectVLDDup(SDNode *N, bool isUpdating,
     if (Alignment == 1)
       Alignment = 0;
   }
-  Align = CurDAG->getTargetConstant(Alignment, MVT::i32);
+  Align = CurDAG->getTargetConstant(Alignment, dl, MVT::i32);
 
   unsigned OpcodeIndex;
   switch (VT.getSimpleVT().SimpleTy) {
@@ -2037,7 +2128,7 @@ SDNode *ARMDAGToDAGISel::SelectVLDDup(SDNode *N, bool isUpdating,
   case MVT::v2i32: OpcodeIndex = 2; break;
   }
 
-  SDValue Pred = getAL(CurDAG);
+  SDValue Pred = getAL(CurDAG, dl);
   SDValue Reg0 = CurDAG->getRegister(0, MVT::i32);
   SDValue SuperReg;
   unsigned Opc = Opcodes[OpcodeIndex];
@@ -2064,8 +2155,7 @@ SDNode *ARMDAGToDAGISel::SelectVLDDup(SDNode *N, bool isUpdating,
   if (isUpdating)
     ResTys.push_back(MVT::i32);
   ResTys.push_back(MVT::Other);
-  SDNode *VLdDup =
-    CurDAG->getMachineNode(Opc, dl, ResTys, Ops.data(), Ops.size());
+  SDNode *VLdDup = CurDAG->getMachineNode(Opc, dl, ResTys, Ops);
   cast<MachineSDNode>(VLdDup)->setMemRefs(MemOp, MemOp + 1);
   SuperReg = SDValue(VLdDup, 0);
 
@@ -2078,13 +2168,13 @@ SDNode *ARMDAGToDAGISel::SelectVLDDup(SDNode *N, bool isUpdating,
   ReplaceUses(SDValue(N, NumVecs), SDValue(VLdDup, 1));
   if (isUpdating)
     ReplaceUses(SDValue(N, NumVecs + 1), SDValue(VLdDup, 2));
-  return NULL;
+  return nullptr;
 }
 
 SDNode *ARMDAGToDAGISel::SelectVTBL(SDNode *N, bool IsExt, unsigned NumVecs,
                                     unsigned Opc) {
   assert(NumVecs >= 2 && NumVecs <= 4 && "VTBL NumVecs out-of-range");
-  DebugLoc dl = N->getDebugLoc();
+  SDLoc dl(N);
   EVT VT = N->getValueType(0);
   unsigned FirstTblReg = IsExt ? 2 : 1;
 
@@ -2109,19 +2199,20 @@ SDNode *ARMDAGToDAGISel::SelectVTBL(SDNode *N, bool IsExt, unsigned NumVecs,
     Ops.push_back(N->getOperand(1));
   Ops.push_back(RegSeq);
   Ops.push_back(N->getOperand(FirstTblReg + NumVecs));
-  Ops.push_back(getAL(CurDAG)); // predicate
+  Ops.push_back(getAL(CurDAG, dl)); // predicate
   Ops.push_back(CurDAG->getRegister(0, MVT::i32)); // predicate register
-  return CurDAG->getMachineNode(Opc, dl, VT, Ops.data(), Ops.size());
+  return CurDAG->getMachineNode(Opc, dl, VT, Ops);
 }
 
 SDNode *ARMDAGToDAGISel::SelectV6T2BitfieldExtractOp(SDNode *N,
                                                      bool isSigned) {
   if (!Subtarget->hasV6T2Ops())
-    return NULL;
+    return nullptr;
 
-  unsigned Opc = isSigned ? (Subtarget->isThumb() ? ARM::t2SBFX : ARM::SBFX)
+  unsigned Opc = isSigned
+    ? (Subtarget->isThumb() ? ARM::t2SBFX : ARM::SBFX)
     : (Subtarget->isThumb() ? ARM::t2UBFX : ARM::UBFX);
-
+  SDLoc dl(N);
 
   // For unsigned extracts, check for a shift right and mask
   unsigned And_imm = 0;
@@ -2130,7 +2221,7 @@ SDNode *ARMDAGToDAGISel::SelectV6T2BitfieldExtractOp(SDNode *N,
 
       // The immediate is a mask of the low bits iff imm & (imm+1) == 0
       if (And_imm & (And_imm + 1))
-        return NULL;
+        return nullptr;
 
       unsigned Srl_imm = 0;
       if (isOpcWithIntImmediate(N->getOperand(0).getNode(), ISD::SRL,
@@ -2138,17 +2229,39 @@ SDNode *ARMDAGToDAGISel::SelectV6T2BitfieldExtractOp(SDNode *N,
         assert(Srl_imm > 0 && Srl_imm < 32 && "bad amount in shift node!");
 
         // Note: The width operand is encoded as width-1.
-        unsigned Width = CountTrailingOnes_32(And_imm) - 1;
+        unsigned Width = countTrailingOnes(And_imm) - 1;
         unsigned LSB = Srl_imm;
+
         SDValue Reg0 = CurDAG->getRegister(0, MVT::i32);
+
+        if ((LSB + Width + 1) == N->getValueType(0).getSizeInBits()) {
+          // It's cheaper to use a right shift to extract the top bits.
+          if (Subtarget->isThumb()) {
+            Opc = isSigned ? ARM::t2ASRri : ARM::t2LSRri;
+            SDValue Ops[] = { N->getOperand(0).getOperand(0),
+                              CurDAG->getTargetConstant(LSB, dl, MVT::i32),
+                              getAL(CurDAG, dl), Reg0, Reg0 };
+            return CurDAG->SelectNodeTo(N, Opc, MVT::i32, Ops);
+          }
+
+          // ARM models shift instructions as MOVsi with shifter operand.
+          ARM_AM::ShiftOpc ShOpcVal = ARM_AM::getShiftOpcForNode(ISD::SRL);
+          SDValue ShOpc =
+            CurDAG->getTargetConstant(ARM_AM::getSORegOpc(ShOpcVal, LSB), dl,
+                                      MVT::i32);
+          SDValue Ops[] = { N->getOperand(0).getOperand(0), ShOpc,
+                            getAL(CurDAG, dl), Reg0, Reg0 };
+          return CurDAG->SelectNodeTo(N, ARM::MOVsi, MVT::i32, Ops);
+        }
+
         SDValue Ops[] = { N->getOperand(0).getOperand(0),
-                          CurDAG->getTargetConstant(LSB, MVT::i32),
-                          CurDAG->getTargetConstant(Width, MVT::i32),
-          getAL(CurDAG), Reg0 };
-        return CurDAG->SelectNodeTo(N, Opc, MVT::i32, Ops, 5);
+                          CurDAG->getTargetConstant(LSB, dl, MVT::i32),
+                          CurDAG->getTargetConstant(Width, dl, MVT::i32),
+                          getAL(CurDAG, dl), Reg0 };
+        return CurDAG->SelectNodeTo(N, Opc, MVT::i32, Ops);
       }
     }
-    return NULL;
+    return nullptr;
   }
 
   // Otherwise, we're looking for a shift of a shift
@@ -2162,214 +2275,35 @@ SDNode *ARMDAGToDAGISel::SelectV6T2BitfieldExtractOp(SDNode *N,
       unsigned Width = 32 - Srl_imm - 1;
       int LSB = Srl_imm - Shl_imm;
       if (LSB < 0)
-        return NULL;
+        return nullptr;
       SDValue Reg0 = CurDAG->getRegister(0, MVT::i32);
       SDValue Ops[] = { N->getOperand(0).getOperand(0),
-                        CurDAG->getTargetConstant(LSB, MVT::i32),
-                        CurDAG->getTargetConstant(Width, MVT::i32),
-                        getAL(CurDAG), Reg0 };
-      return CurDAG->SelectNodeTo(N, Opc, MVT::i32, Ops, 5);
+                        CurDAG->getTargetConstant(LSB, dl, MVT::i32),
+                        CurDAG->getTargetConstant(Width, dl, MVT::i32),
+                        getAL(CurDAG, dl), Reg0 };
+      return CurDAG->SelectNodeTo(N, Opc, MVT::i32, Ops);
     }
   }
-  return NULL;
-}
 
-SDNode *ARMDAGToDAGISel::
-SelectT2CMOVShiftOp(SDNode *N, SDValue FalseVal, SDValue TrueVal,
-                    ARMCC::CondCodes CCVal, SDValue CCR, SDValue InFlag) {
-  SDValue CPTmp0;
-  SDValue CPTmp1;
-  if (SelectT2ShifterOperandReg(TrueVal, CPTmp0, CPTmp1)) {
-    unsigned SOVal = cast<ConstantSDNode>(CPTmp1)->getZExtValue();
-    unsigned SOShOp = ARM_AM::getSORegShOp(SOVal);
-    unsigned Opc = 0;
-    switch (SOShOp) {
-    case ARM_AM::lsl: Opc = ARM::t2MOVCClsl; break;
-    case ARM_AM::lsr: Opc = ARM::t2MOVCClsr; break;
-    case ARM_AM::asr: Opc = ARM::t2MOVCCasr; break;
-    case ARM_AM::ror: Opc = ARM::t2MOVCCror; break;
-    default:
-      llvm_unreachable("Unknown so_reg opcode!");
-    }
-    SDValue SOShImm =
-      CurDAG->getTargetConstant(ARM_AM::getSORegOffset(SOVal), MVT::i32);
-    SDValue CC = CurDAG->getTargetConstant(CCVal, MVT::i32);
-    SDValue Ops[] = { FalseVal, CPTmp0, SOShImm, CC, CCR, InFlag };
-    return CurDAG->SelectNodeTo(N, Opc, MVT::i32,Ops, 6);
-  }
-  return 0;
-}
-
-SDNode *ARMDAGToDAGISel::
-SelectARMCMOVShiftOp(SDNode *N, SDValue FalseVal, SDValue TrueVal,
-                     ARMCC::CondCodes CCVal, SDValue CCR, SDValue InFlag) {
-  SDValue CPTmp0;
-  SDValue CPTmp1;
-  SDValue CPTmp2;
-  if (SelectImmShifterOperand(TrueVal, CPTmp0, CPTmp2)) {
-    SDValue CC = CurDAG->getTargetConstant(CCVal, MVT::i32);
-    SDValue Ops[] = { FalseVal, CPTmp0, CPTmp2, CC, CCR, InFlag };
-    return CurDAG->SelectNodeTo(N, ARM::MOVCCsi, MVT::i32, Ops, 6);
-  }
-
-  if (SelectRegShifterOperand(TrueVal, CPTmp0, CPTmp1, CPTmp2)) {
-    SDValue CC = CurDAG->getTargetConstant(CCVal, MVT::i32);
-    SDValue Ops[] = { FalseVal, CPTmp0, CPTmp1, CPTmp2, CC, CCR, InFlag };
-    return CurDAG->SelectNodeTo(N, ARM::MOVCCsr, MVT::i32, Ops, 7);
-  }
-  return 0;
-}
-
-SDNode *ARMDAGToDAGISel::
-SelectT2CMOVImmOp(SDNode *N, SDValue FalseVal, SDValue TrueVal,
-                  ARMCC::CondCodes CCVal, SDValue CCR, SDValue InFlag) {
-  ConstantSDNode *T = dyn_cast<ConstantSDNode>(TrueVal);
-  if (!T)
-    return 0;
-
-  unsigned Opc = 0;
-  unsigned TrueImm = T->getZExtValue();
-  if (is_t2_so_imm(TrueImm)) {
-    Opc = ARM::t2MOVCCi;
-  } else if (TrueImm <= 0xffff) {
-    Opc = ARM::t2MOVCCi16;
-  } else if (is_t2_so_imm_not(TrueImm)) {
-    TrueImm = ~TrueImm;
-    Opc = ARM::t2MVNCCi;
-  } else if (TrueVal.getNode()->hasOneUse() && Subtarget->hasV6T2Ops()) {
-    // Large immediate.
-    Opc = ARM::t2MOVCCi32imm;
-  }
-
-  if (Opc) {
-    SDValue True = CurDAG->getTargetConstant(TrueImm, MVT::i32);
-    SDValue CC = CurDAG->getTargetConstant(CCVal, MVT::i32);
-    SDValue Ops[] = { FalseVal, True, CC, CCR, InFlag };
-    return CurDAG->SelectNodeTo(N, Opc, MVT::i32, Ops, 5);
-  }
-
-  return 0;
-}
-
-SDNode *ARMDAGToDAGISel::
-SelectARMCMOVImmOp(SDNode *N, SDValue FalseVal, SDValue TrueVal,
-                   ARMCC::CondCodes CCVal, SDValue CCR, SDValue InFlag) {
-  ConstantSDNode *T = dyn_cast<ConstantSDNode>(TrueVal);
-  if (!T)
-    return 0;
-
-  unsigned Opc = 0;
-  unsigned TrueImm = T->getZExtValue();
-  bool isSoImm = is_so_imm(TrueImm);
-  if (isSoImm) {
-    Opc = ARM::MOVCCi;
-  } else if (Subtarget->hasV6T2Ops() && TrueImm <= 0xffff) {
-    Opc = ARM::MOVCCi16;
-  } else if (is_so_imm_not(TrueImm)) {
-    TrueImm = ~TrueImm;
-    Opc = ARM::MVNCCi;
-  } else if (TrueVal.getNode()->hasOneUse() &&
-             (Subtarget->hasV6T2Ops() || ARM_AM::isSOImmTwoPartVal(TrueImm))) {
-    // Large immediate.
-    Opc = ARM::MOVCCi32imm;
-  }
-
-  if (Opc) {
-    SDValue True = CurDAG->getTargetConstant(TrueImm, MVT::i32);
-    SDValue CC = CurDAG->getTargetConstant(CCVal, MVT::i32);
-    SDValue Ops[] = { FalseVal, True, CC, CCR, InFlag };
-    return CurDAG->SelectNodeTo(N, Opc, MVT::i32, Ops, 5);
-  }
-
-  return 0;
-}
-
-SDNode *ARMDAGToDAGISel::SelectCMOVOp(SDNode *N) {
-  EVT VT = N->getValueType(0);
-  SDValue FalseVal = N->getOperand(0);
-  SDValue TrueVal  = N->getOperand(1);
-  SDValue CC = N->getOperand(2);
-  SDValue CCR = N->getOperand(3);
-  SDValue InFlag = N->getOperand(4);
-  assert(CC.getOpcode() == ISD::Constant);
-  assert(CCR.getOpcode() == ISD::Register);
-  ARMCC::CondCodes CCVal =
-    (ARMCC::CondCodes)cast<ConstantSDNode>(CC)->getZExtValue();
-
-  if (!Subtarget->isThumb1Only() && VT == MVT::i32) {
-    // Pattern: (ARMcmov:i32 GPR:i32:$false, so_reg:i32:$true, (imm:i32):$cc)
-    // Emits: (MOVCCs:i32 GPR:i32:$false, so_reg:i32:$true, (imm:i32):$cc)
-    // Pattern complexity = 18  cost = 1  size = 0
-    if (Subtarget->isThumb()) {
-      SDNode *Res = SelectT2CMOVShiftOp(N, FalseVal, TrueVal,
-                                        CCVal, CCR, InFlag);
-      if (!Res)
-        Res = SelectT2CMOVShiftOp(N, TrueVal, FalseVal,
-                               ARMCC::getOppositeCondition(CCVal), CCR, InFlag);
-      if (Res)
-        return Res;
-    } else {
-      SDNode *Res = SelectARMCMOVShiftOp(N, FalseVal, TrueVal,
-                                         CCVal, CCR, InFlag);
-      if (!Res)
-        Res = SelectARMCMOVShiftOp(N, TrueVal, FalseVal,
-                               ARMCC::getOppositeCondition(CCVal), CCR, InFlag);
-      if (Res)
-        return Res;
-    }
-
-    // Pattern: (ARMcmov:i32 GPR:i32:$false,
-    //             (imm:i32)<<P:Pred_so_imm>>:$true,
-    //             (imm:i32):$cc)
-    // Emits: (MOVCCi:i32 GPR:i32:$false,
-    //           (so_imm:i32 (imm:i32):$true), (imm:i32):$cc)
-    // Pattern complexity = 10  cost = 1  size = 0
-    if (Subtarget->isThumb()) {
-      SDNode *Res = SelectT2CMOVImmOp(N, FalseVal, TrueVal,
-                                        CCVal, CCR, InFlag);
-      if (!Res)
-        Res = SelectT2CMOVImmOp(N, TrueVal, FalseVal,
-                               ARMCC::getOppositeCondition(CCVal), CCR, InFlag);
-      if (Res)
-        return Res;
-    } else {
-      SDNode *Res = SelectARMCMOVImmOp(N, FalseVal, TrueVal,
-                                         CCVal, CCR, InFlag);
-      if (!Res)
-        Res = SelectARMCMOVImmOp(N, TrueVal, FalseVal,
-                               ARMCC::getOppositeCondition(CCVal), CCR, InFlag);
-      if (Res)
-        return Res;
-    }
-  }
-
-  // Pattern: (ARMcmov:i32 GPR:i32:$false, GPR:i32:$true, (imm:i32):$cc)
-  // Emits: (MOVCCr:i32 GPR:i32:$false, GPR:i32:$true, (imm:i32):$cc)
-  // Pattern complexity = 6  cost = 1  size = 0
-  //
-  // Pattern: (ARMcmov:i32 GPR:i32:$false, GPR:i32:$true, (imm:i32):$cc)
-  // Emits: (tMOVCCr:i32 GPR:i32:$false, GPR:i32:$true, (imm:i32):$cc)
-  // Pattern complexity = 6  cost = 11  size = 0
-  //
-  // Also VMOVScc and VMOVDcc.
-  SDValue Tmp2 = CurDAG->getTargetConstant(CCVal, MVT::i32);
-  SDValue Ops[] = { FalseVal, TrueVal, Tmp2, CCR, InFlag };
-  unsigned Opc = 0;
-  switch (VT.getSimpleVT().SimpleTy) {
-  default: llvm_unreachable("Illegal conditional move type!");
-  case MVT::i32:
-    Opc = Subtarget->isThumb()
-      ? (Subtarget->hasThumb2() ? ARM::t2MOVCCr : ARM::tMOVCCr_pseudo)
-      : ARM::MOVCCr;
-    break;
-  case MVT::f32:
-    Opc = ARM::VMOVScc;
-    break;
-  case MVT::f64:
-    Opc = ARM::VMOVDcc;
-    break;
+  if (N->getOpcode() == ISD::SIGN_EXTEND_INREG) {
+    unsigned Width = cast<VTSDNode>(N->getOperand(1))->getVT().getSizeInBits();
+    unsigned LSB = 0;
+    if (!isOpcWithIntImmediate(N->getOperand(0).getNode(), ISD::SRL, LSB) &&
+        !isOpcWithIntImmediate(N->getOperand(0).getNode(), ISD::SRA, LSB))
+      return nullptr;
+
+    if (LSB + Width > 32)
+      return nullptr;
+
+    SDValue Reg0 = CurDAG->getRegister(0, MVT::i32);
+    SDValue Ops[] = { N->getOperand(0).getOperand(0),
+                      CurDAG->getTargetConstant(LSB, dl, MVT::i32),
+                      CurDAG->getTargetConstant(Width - 1, dl, MVT::i32),
+                      getAL(CurDAG, dl), Reg0 };
+    return CurDAG->SelectNodeTo(N, Opc, MVT::i32, Ops);
   }
-  return CurDAG->SelectNodeTo(N, Opc, VT, Ops, 5);
+
+  return nullptr;
 }
 
 /// Target-specific DAG combining for ISD::XOR.
@@ -2388,10 +2322,10 @@ SDNode *ARMDAGToDAGISel::SelectABSOp(SDNode *N){
   EVT VT = N->getValueType(0);
 
   if (Subtarget->isThumb1Only())
-    return NULL;
+    return nullptr;
 
   if (XORSrc0.getOpcode() != ISD::ADD || XORSrc1.getOpcode() != ISD::SRA)
-    return NULL;
+    return nullptr;
 
   SDValue ADDSrc0 = XORSrc0.getOperand(0);
   SDValue ADDSrc1 = XORSrc0.getOperand(1);
@@ -2402,13 +2336,13 @@ SDNode *ARMDAGToDAGISel::SelectABSOp(SDNode *N){
   unsigned Size = XType.getSizeInBits() - 1;
 
   if (ADDSrc1 == XORSrc1 && ADDSrc0 == SRASrc0 &&
-      XType.isInteger() && SRAConstant != NULL &&
+      XType.isInteger() && SRAConstant != nullptr &&
       Size == SRAConstant->getZExtValue()) {
     unsigned Opcode = Subtarget->isThumb2() ? ARM::t2ABS : ARM::ABS;
     return CurDAG->SelectNodeTo(N, Opcode, VT, ADDSrc0);
   }
 
-  return NULL;
+  return nullptr;
 }
 
 SDNode *ARMDAGToDAGISel::SelectConcatVector(SDNode *N) {
@@ -2420,33 +2354,34 @@ SDNode *ARMDAGToDAGISel::SelectConcatVector(SDNode *N) {
   return createDRegPairNode(VT, N->getOperand(0), N->getOperand(1));
 }
 
-SDNode *ARMDAGToDAGISel::SelectAtomic64(SDNode *Node, unsigned Opc) {
-  SmallVector<SDValue, 6> Ops;
-  Ops.push_back(Node->getOperand(1)); // Ptr
-  Ops.push_back(Node->getOperand(2)); // Low part of Val1
-  Ops.push_back(Node->getOperand(3)); // High part of Val1
-  if (Opc == ARM::ATOMCMPXCHG6432) {
-    Ops.push_back(Node->getOperand(4)); // Low part of Val2
-    Ops.push_back(Node->getOperand(5)); // High part of Val2
-  }
-  Ops.push_back(Node->getOperand(0)); // Chain
-  MachineSDNode::mmo_iterator MemOp = MF->allocateMemRefsArray(1);
-  MemOp[0] = cast<MemSDNode>(Node)->getMemOperand();
-  SDNode *ResNode = CurDAG->getMachineNode(Opc, Node->getDebugLoc(),
-                                           MVT::i32, MVT::i32, MVT::Other,
-                                           Ops.data() ,Ops.size());
-  cast<MachineSDNode>(ResNode)->setMemRefs(MemOp, MemOp + 1);
-  return ResNode;
-}
-
 SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
-  DebugLoc dl = N->getDebugLoc();
+  SDLoc dl(N);
 
-  if (N->isMachineOpcode())
-    return NULL;   // Already selected.
+  if (N->isMachineOpcode()) {
+    N->setNodeId(-1);
+    return nullptr;   // Already selected.
+  }
 
   switch (N->getOpcode()) {
   default: break;
+  case ISD::WRITE_REGISTER: {
+    SDNode *ResNode = SelectWriteRegister(N);
+    if (ResNode)
+      return ResNode;
+    break;
+  }
+  case ISD::READ_REGISTER: {
+    SDNode *ResNode = SelectReadRegister(N);
+    if (ResNode)
+      return ResNode;
+    break;
+  }
+  case ISD::INLINEASM: {
+    SDNode *ResNode = SelectInlineAsm(N);
+    if (ResNode)
+      return ResNode;
+    break;
+  }
   case ISD::XOR: {
     // Select special operations if XOR node forms integer ABS pattern
     SDNode *ResNode = SelectABSOp(N);
@@ -2458,47 +2393,48 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
   case ISD::Constant: {
     unsigned Val = cast<ConstantSDNode>(N)->getZExtValue();
     bool UseCP = true;
-    if (Subtarget->hasThumb2())
+    if (Subtarget->useMovt(*MF))
       // Thumb2-aware targets have the MOVT instruction, so all immediates can
       // be done with MOV + MOVT, at worst.
-      UseCP = 0;
+      UseCP = false;
     else {
       if (Subtarget->isThumb()) {
-        UseCP = (Val > 255 &&                          // MOV
-                 ~Val > 255 &&                         // MOV + MVN
-                 !ARM_AM::isThumbImmShiftedVal(Val));  // MOV + LSL
+        UseCP = (Val > 255 &&                                  // MOV
+                 ~Val > 255 &&                                 // MOV + MVN
+                 !ARM_AM::isThumbImmShiftedVal(Val) &&         // MOV + LSL
+                 !(Subtarget->hasV6T2Ops() && Val <= 0xffff)); // MOVW
       } else
-        UseCP = (ARM_AM::getSOImmVal(Val) == -1 &&     // MOV
-                 ARM_AM::getSOImmVal(~Val) == -1 &&    // MVN
-                 !ARM_AM::isSOImmTwoPartVal(Val));     // two instrs.
+        UseCP = (ARM_AM::getSOImmVal(Val) == -1 &&             // MOV
+                 ARM_AM::getSOImmVal(~Val) == -1 &&            // MVN
+                 !ARM_AM::isSOImmTwoPartVal(Val) &&            // two instrs.
+                 !(Subtarget->hasV6T2Ops() && Val <= 0xffff)); // MOVW
     }
 
     if (UseCP) {
-      SDValue CPIdx =
-        CurDAG->getTargetConstantPool(ConstantInt::get(
-                                  Type::getInt32Ty(*CurDAG->getContext()), Val),
-                                      TLI.getPointerTy());
+      SDValue CPIdx = CurDAG->getTargetConstantPool(
+          ConstantInt::get(Type::getInt32Ty(*CurDAG->getContext()), Val),
+          TLI->getPointerTy(CurDAG->getDataLayout()));
 
       SDNode *ResNode;
-      if (Subtarget->isThumb1Only()) {
-        SDValue Pred = getAL(CurDAG);
+      if (Subtarget->isThumb()) {
+        SDValue Pred = getAL(CurDAG, dl);
         SDValue PredReg = CurDAG->getRegister(0, MVT::i32);
         SDValue Ops[] = { CPIdx, Pred, PredReg, CurDAG->getEntryNode() };
         ResNode = CurDAG->getMachineNode(ARM::tLDRpci, dl, MVT::i32, MVT::Other,
-                                         Ops, 4);
+                                         Ops);
       } else {
         SDValue Ops[] = {
           CPIdx,
-          CurDAG->getTargetConstant(0, MVT::i32),
-          getAL(CurDAG),
+          CurDAG->getTargetConstant(0, dl, MVT::i32),
+          getAL(CurDAG, dl),
           CurDAG->getRegister(0, MVT::i32),
           CurDAG->getEntryNode()
         };
         ResNode=CurDAG->getMachineNode(ARM::LDRcp, dl, MVT::i32, MVT::Other,
-                                       Ops, 5);
+                                       Ops);
       }
       ReplaceUses(SDValue(N, 0), SDValue(ResNode, 0));
-      return NULL;
+      return nullptr;
     }
 
     // Other cases are autogenerated.
@@ -2507,24 +2443,30 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
   case ISD::FrameIndex: {
     // Selects to ADDri FI, 0 which in turn will become ADDri SP, imm.
     int FI = cast<FrameIndexSDNode>(N)->getIndex();
-    SDValue TFI = CurDAG->getTargetFrameIndex(FI, TLI.getPointerTy());
+    SDValue TFI = CurDAG->getTargetFrameIndex(
+        FI, TLI->getPointerTy(CurDAG->getDataLayout()));
     if (Subtarget->isThumb1Only()) {
-      SDValue Ops[] = { TFI, CurDAG->getTargetConstant(0, MVT::i32),
-                        getAL(CurDAG), CurDAG->getRegister(0, MVT::i32) };
-      return CurDAG->SelectNodeTo(N, ARM::tADDrSPi, MVT::i32, Ops, 4);
+      // Set the alignment of the frame object to 4, to avoid having to generate
+      // more than one ADD
+      MachineFrameInfo *MFI = MF->getFrameInfo();
+      if (MFI->getObjectAlignment(FI) < 4)
+        MFI->setObjectAlignment(FI, 4);
+      return CurDAG->SelectNodeTo(N, ARM::tADDframe, MVT::i32, TFI,
+                                  CurDAG->getTargetConstant(0, dl, MVT::i32));
     } else {
       unsigned Opc = ((Subtarget->isThumb() && Subtarget->hasThumb2()) ?
                       ARM::t2ADDri : ARM::ADDri);
-      SDValue Ops[] = { TFI, CurDAG->getTargetConstant(0, MVT::i32),
-                        getAL(CurDAG), CurDAG->getRegister(0, MVT::i32),
+      SDValue Ops[] = { TFI, CurDAG->getTargetConstant(0, dl, MVT::i32),
+                        getAL(CurDAG, dl), CurDAG->getRegister(0, MVT::i32),
                         CurDAG->getRegister(0, MVT::i32) };
-      return CurDAG->SelectNodeTo(N, Opc, MVT::i32, Ops, 5);
+      return CurDAG->SelectNodeTo(N, Opc, MVT::i32, Ops);
     }
   }
   case ISD::SRL:
     if (SDNode *I = SelectV6T2BitfieldExtractOp(N, false))
       return I;
     break;
+  case ISD::SIGN_EXTEND_INREG:
   case ISD::SRA:
     if (SDNode *I = SelectV6T2BitfieldExtractOp(N, true))
       return I;
@@ -2541,14 +2483,15 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
           break;
         SDValue V = N->getOperand(0);
         ShImm = ARM_AM::getSORegOpc(ARM_AM::lsl, ShImm);
-        SDValue ShImmOp = CurDAG->getTargetConstant(ShImm, MVT::i32);
+        SDValue ShImmOp = CurDAG->getTargetConstant(ShImm, dl, MVT::i32);
         SDValue Reg0 = CurDAG->getRegister(0, MVT::i32);
         if (Subtarget->isThumb()) {
-          SDValue Ops[] = { V, V, ShImmOp, getAL(CurDAG), Reg0, Reg0 };
-          return CurDAG->SelectNodeTo(N, ARM::t2ADDrs, MVT::i32, Ops, 6);
+          SDValue Ops[] = { V, V, ShImmOp, getAL(CurDAG, dl), Reg0, Reg0 };
+          return CurDAG->SelectNodeTo(N, ARM::t2ADDrs, MVT::i32, Ops);
         } else {
-          SDValue Ops[] = { V, V, Reg0, ShImmOp, getAL(CurDAG), Reg0, Reg0 };
-          return CurDAG->SelectNodeTo(N, ARM::ADDrsi, MVT::i32, Ops, 7);
+          SDValue Ops[] = { V, V, Reg0, ShImmOp, getAL(CurDAG, dl), Reg0,
+                            Reg0 };
+          return CurDAG->SelectNodeTo(N, ARM::ADDrsi, MVT::i32, Ops);
         }
       }
       if (isPowerOf2_32(RHSV+1)) {  // 2^n-1?
@@ -2557,14 +2500,15 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
           break;
         SDValue V = N->getOperand(0);
         ShImm = ARM_AM::getSORegOpc(ARM_AM::lsl, ShImm);
-        SDValue ShImmOp = CurDAG->getTargetConstant(ShImm, MVT::i32);
+        SDValue ShImmOp = CurDAG->getTargetConstant(ShImm, dl, MVT::i32);
         SDValue Reg0 = CurDAG->getRegister(0, MVT::i32);
         if (Subtarget->isThumb()) {
-          SDValue Ops[] = { V, V, ShImmOp, getAL(CurDAG), Reg0, Reg0 };
-          return CurDAG->SelectNodeTo(N, ARM::t2RSBrs, MVT::i32, Ops, 6);
+          SDValue Ops[] = { V, V, ShImmOp, getAL(CurDAG, dl), Reg0, Reg0 };
+          return CurDAG->SelectNodeTo(N, ARM::t2RSBrs, MVT::i32, Ops);
         } else {
-          SDValue Ops[] = { V, V, Reg0, ShImmOp, getAL(CurDAG), Reg0, Reg0 };
-          return CurDAG->SelectNodeTo(N, ARM::RSBrsi, MVT::i32, Ops, 7);
+          SDValue Ops[] = { V, V, Reg0, ShImmOp, getAL(CurDAG, dl), Reg0,
+                            Reg0 };
+          return CurDAG->SelectNodeTo(N, ARM::RSBrsi, MVT::i32, Ops);
         }
       }
     }
@@ -2602,33 +2546,32 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
           (N1CVal & 0xffffU) == 0xffffU &&
           (N2CVal & 0xffffU) == 0x0U) {
         SDValue Imm16 = CurDAG->getTargetConstant((N2CVal & 0xFFFF0000U) >> 16,
-                                                  MVT::i32);
+                                                  dl, MVT::i32);
         SDValue Ops[] = { N0.getOperand(0), Imm16,
-                          getAL(CurDAG), CurDAG->getRegister(0, MVT::i32) };
-        return CurDAG->getMachineNode(Opc, dl, VT, Ops, 4);
+                          getAL(CurDAG, dl), CurDAG->getRegister(0, MVT::i32) };
+        return CurDAG->getMachineNode(Opc, dl, VT, Ops);
       }
     }
     break;
   }
   case ARMISD::VMOVRRD:
     return CurDAG->getMachineNode(ARM::VMOVRRD, dl, MVT::i32, MVT::i32,
-                                  N->getOperand(0), getAL(CurDAG),
+                                  N->getOperand(0), getAL(CurDAG, dl),
                                   CurDAG->getRegister(0, MVT::i32));
   case ISD::UMUL_LOHI: {
     if (Subtarget->isThumb1Only())
       break;
     if (Subtarget->isThumb()) {
       SDValue Ops[] = { N->getOperand(0), N->getOperand(1),
-                        getAL(CurDAG), CurDAG->getRegister(0, MVT::i32),
-                        CurDAG->getRegister(0, MVT::i32) };
-      return CurDAG->getMachineNode(ARM::t2UMULL, dl, MVT::i32, MVT::i32,Ops,4);
+                        getAL(CurDAG, dl), CurDAG->getRegister(0, MVT::i32) };
+      return CurDAG->getMachineNode(ARM::t2UMULL, dl, MVT::i32, MVT::i32, Ops);
     } else {
       SDValue Ops[] = { N->getOperand(0), N->getOperand(1),
-                        getAL(CurDAG), CurDAG->getRegister(0, MVT::i32),
+                        getAL(CurDAG, dl), CurDAG->getRegister(0, MVT::i32),
                         CurDAG->getRegister(0, MVT::i32) };
       return CurDAG->getMachineNode(Subtarget->hasV6Ops() ?
                                     ARM::UMULL : ARM::UMULLv5,
-                                    dl, MVT::i32, MVT::i32, Ops, 5);
+                                    dl, MVT::i32, MVT::i32, Ops);
     }
   }
   case ISD::SMUL_LOHI: {
@@ -2636,51 +2579,51 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
       break;
     if (Subtarget->isThumb()) {
       SDValue Ops[] = { N->getOperand(0), N->getOperand(1),
-                        getAL(CurDAG), CurDAG->getRegister(0, MVT::i32) };
-      return CurDAG->getMachineNode(ARM::t2SMULL, dl, MVT::i32, MVT::i32,Ops,4);
+                        getAL(CurDAG, dl), CurDAG->getRegister(0, MVT::i32) };
+      return CurDAG->getMachineNode(ARM::t2SMULL, dl, MVT::i32, MVT::i32, Ops);
     } else {
       SDValue Ops[] = { N->getOperand(0), N->getOperand(1),
-                        getAL(CurDAG), CurDAG->getRegister(0, MVT::i32),
+                        getAL(CurDAG, dl), CurDAG->getRegister(0, MVT::i32),
                         CurDAG->getRegister(0, MVT::i32) };
       return CurDAG->getMachineNode(Subtarget->hasV6Ops() ?
                                     ARM::SMULL : ARM::SMULLv5,
-                                    dl, MVT::i32, MVT::i32, Ops, 5);
+                                    dl, MVT::i32, MVT::i32, Ops);
     }
   }
   case ARMISD::UMLAL:{
     if (Subtarget->isThumb()) {
       SDValue Ops[] = { N->getOperand(0), N->getOperand(1), N->getOperand(2),
-                        N->getOperand(3), getAL(CurDAG),
+                        N->getOperand(3), getAL(CurDAG, dl),
                         CurDAG->getRegister(0, MVT::i32)};
-      return CurDAG->getMachineNode(ARM::t2UMLAL, dl, MVT::i32, MVT::i32, Ops, 6);
+      return CurDAG->getMachineNode(ARM::t2UMLAL, dl, MVT::i32, MVT::i32, Ops);
     }else{
       SDValue Ops[] = { N->getOperand(0), N->getOperand(1), N->getOperand(2),
-                        N->getOperand(3), getAL(CurDAG),
+                        N->getOperand(3), getAL(CurDAG, dl),
                         CurDAG->getRegister(0, MVT::i32),
                         CurDAG->getRegister(0, MVT::i32) };
       return CurDAG->getMachineNode(Subtarget->hasV6Ops() ?
                                       ARM::UMLAL : ARM::UMLALv5,
-                                      dl, MVT::i32, MVT::i32, Ops, 7);
+                                      dl, MVT::i32, MVT::i32, Ops);
     }
   }
   case ARMISD::SMLAL:{
     if (Subtarget->isThumb()) {
       SDValue Ops[] = { N->getOperand(0), N->getOperand(1), N->getOperand(2),
-                        N->getOperand(3), getAL(CurDAG),
+                        N->getOperand(3), getAL(CurDAG, dl),
                         CurDAG->getRegister(0, MVT::i32)};
-      return CurDAG->getMachineNode(ARM::t2SMLAL, dl, MVT::i32, MVT::i32, Ops, 6);
+      return CurDAG->getMachineNode(ARM::t2SMLAL, dl, MVT::i32, MVT::i32, Ops);
     }else{
       SDValue Ops[] = { N->getOperand(0), N->getOperand(1), N->getOperand(2),
-                        N->getOperand(3), getAL(CurDAG),
+                        N->getOperand(3), getAL(CurDAG, dl),
                         CurDAG->getRegister(0, MVT::i32),
                         CurDAG->getRegister(0, MVT::i32) };
       return CurDAG->getMachineNode(Subtarget->hasV6Ops() ?
                                       ARM::SMLAL : ARM::SMLALv5,
-                                      dl, MVT::i32, MVT::i32, Ops, 7);
+                                      dl, MVT::i32, MVT::i32, Ops);
     }
   }
   case ISD::LOAD: {
-    SDNode *ResNode = 0;
+    SDNode *ResNode = nullptr;
     if (Subtarget->isThumb() && Subtarget->hasThumb2())
       ResNode = SelectT2IndexedLoad(N);
     else
@@ -2715,11 +2658,11 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
     assert(N3.getOpcode() == ISD::Register);
 
     SDValue Tmp2 = CurDAG->getTargetConstant(((unsigned)
-                               cast<ConstantSDNode>(N2)->getZExtValue()),
+                               cast<ConstantSDNode>(N2)->getZExtValue()), dl,
                                MVT::i32);
     SDValue Ops[] = { N1, Tmp2, N3, Chain, InFlag };
     SDNode *ResNode = CurDAG->getMachineNode(Opc, dl, MVT::Other,
-                                             MVT::Glue, Ops, 5);
+                                             MVT::Glue, Ops);
     Chain = SDValue(ResNode, 0);
     if (N->getNumValues() == 2) {
       InFlag = SDValue(ResNode, 1);
@@ -2727,15 +2670,13 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
     }
     ReplaceUses(SDValue(N, 0),
                 SDValue(Chain.getNode(), Chain.getResNo()));
-    return NULL;
+    return nullptr;
   }
-  case ARMISD::CMOV:
-    return SelectCMOVOp(N);
   case ARMISD::VZIP: {
     unsigned Opc = 0;
     EVT VT = N->getValueType(0);
     switch (VT.getSimpleVT().SimpleTy) {
-    default: return NULL;
+    default: return nullptr;
     case MVT::v8i8:  Opc = ARM::VZIPd8; break;
     case MVT::v4i16: Opc = ARM::VZIPd16; break;
     case MVT::v2f32:
@@ -2746,16 +2687,16 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
     case MVT::v4f32:
     case MVT::v4i32: Opc = ARM::VZIPq32; break;
     }
-    SDValue Pred = getAL(CurDAG);
+    SDValue Pred = getAL(CurDAG, dl);
     SDValue PredReg = CurDAG->getRegister(0, MVT::i32);
     SDValue Ops[] = { N->getOperand(0), N->getOperand(1), Pred, PredReg };
-    return CurDAG->getMachineNode(Opc, dl, VT, VT, Ops, 4);
+    return CurDAG->getMachineNode(Opc, dl, VT, VT, Ops);
   }
   case ARMISD::VUZP: {
     unsigned Opc = 0;
     EVT VT = N->getValueType(0);
     switch (VT.getSimpleVT().SimpleTy) {
-    default: return NULL;
+    default: return nullptr;
     case MVT::v8i8:  Opc = ARM::VUZPd8; break;
     case MVT::v4i16: Opc = ARM::VUZPd16; break;
     case MVT::v2f32:
@@ -2766,16 +2707,16 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
     case MVT::v4f32:
     case MVT::v4i32: Opc = ARM::VUZPq32; break;
     }
-    SDValue Pred = getAL(CurDAG);
+    SDValue Pred = getAL(CurDAG, dl);
     SDValue PredReg = CurDAG->getRegister(0, MVT::i32);
     SDValue Ops[] = { N->getOperand(0), N->getOperand(1), Pred, PredReg };
-    return CurDAG->getMachineNode(Opc, dl, VT, VT, Ops, 4);
+    return CurDAG->getMachineNode(Opc, dl, VT, VT, Ops);
   }
   case ARMISD::VTRN: {
     unsigned Opc = 0;
     EVT VT = N->getValueType(0);
     switch (VT.getSimpleVT().SimpleTy) {
-    default: return NULL;
+    default: return nullptr;
     case MVT::v8i8:  Opc = ARM::VTRNd8; break;
     case MVT::v4i16: Opc = ARM::VTRNd16; break;
     case MVT::v2f32:
@@ -2785,10 +2726,10 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
     case MVT::v4f32:
     case MVT::v4i32: Opc = ARM::VTRNq32; break;
     }
-    SDValue Pred = getAL(CurDAG);
+    SDValue Pred = getAL(CurDAG, dl);
     SDValue PredReg = CurDAG->getRegister(0, MVT::i32);
     SDValue Ops[] = { N->getOperand(0), N->getOperand(1), Pred, PredReg };
-    return CurDAG->getMachineNode(Opc, dl, VT, VT, Ops, 4);
+    return CurDAG->getMachineNode(Opc, dl, VT, VT, Ops);
   }
   case ARMISD::BUILD_VECTOR: {
     EVT VecVT = N->getValueType(0);
@@ -2856,7 +2797,7 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
                                          ARM::VLD1q16wb_fixed,
                                          ARM::VLD1q32wb_fixed,
                                          ARM::VLD1q64wb_fixed };
-    return SelectVLD(N, true, 1, DOpcodes, QOpcodes, 0);
+    return SelectVLD(N, true, 1, DOpcodes, QOpcodes, nullptr);
   }
 
   case ARMISD::VLD2_UPD: {
@@ -2867,14 +2808,14 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
     static const uint16_t QOpcodes[] = { ARM::VLD2q8PseudoWB_fixed,
                                          ARM::VLD2q16PseudoWB_fixed,
                                          ARM::VLD2q32PseudoWB_fixed };
-    return SelectVLD(N, true, 2, DOpcodes, QOpcodes, 0);
+    return SelectVLD(N, true, 2, DOpcodes, QOpcodes, nullptr);
   }
 
   case ARMISD::VLD3_UPD: {
     static const uint16_t DOpcodes[] = { ARM::VLD3d8Pseudo_UPD,
                                          ARM::VLD3d16Pseudo_UPD,
                                          ARM::VLD3d32Pseudo_UPD,
-                                         ARM::VLD1q64wb_fixed};
+                                         ARM::VLD1d64TPseudoWB_fixed};
     static const uint16_t QOpcodes0[] = { ARM::VLD3q8Pseudo_UPD,
                                           ARM::VLD3q16Pseudo_UPD,
                                           ARM::VLD3q32Pseudo_UPD };
@@ -2888,7 +2829,7 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
     static const uint16_t DOpcodes[] = { ARM::VLD4d8Pseudo_UPD,
                                          ARM::VLD4d16Pseudo_UPD,
                                          ARM::VLD4d32Pseudo_UPD,
-                                         ARM::VLD1q64wb_fixed};
+                                         ARM::VLD1d64QPseudoWB_fixed};
     static const uint16_t QOpcodes0[] = { ARM::VLD4q8Pseudo_UPD,
                                           ARM::VLD4q16Pseudo_UPD,
                                           ARM::VLD4q32Pseudo_UPD };
@@ -2934,7 +2875,7 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
                                          ARM::VST1q16wb_fixed,
                                          ARM::VST1q32wb_fixed,
                                          ARM::VST1q64wb_fixed };
-    return SelectVST(N, true, 1, DOpcodes, QOpcodes, 0);
+    return SelectVST(N, true, 1, DOpcodes, QOpcodes, nullptr);
   }
 
   case ARMISD::VST2_UPD: {
@@ -2945,7 +2886,7 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
     static const uint16_t QOpcodes[] = { ARM::VST2q8PseudoWB_fixed,
                                          ARM::VST2q16PseudoWB_fixed,
                                          ARM::VST2q32PseudoWB_fixed };
-    return SelectVST(N, true, 2, DOpcodes, QOpcodes, 0);
+    return SelectVST(N, true, 2, DOpcodes, QOpcodes, nullptr);
   }
 
   case ARMISD::VST3_UPD: {
@@ -3010,13 +2951,16 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
     default:
       break;
 
+    case Intrinsic::arm_ldaexd:
     case Intrinsic::arm_ldrexd: {
-      SDValue MemAddr = N->getOperand(2);
-      DebugLoc dl = N->getDebugLoc();
+      SDLoc dl(N);
       SDValue Chain = N->getOperand(0);
-
+      SDValue MemAddr = N->getOperand(2);
       bool isThumb = Subtarget->isThumb() && Subtarget->hasThumb2();
-      unsigned NewOpc = isThumb ? ARM::t2LDREXD :ARM::LDREXD;
+
+      bool IsAcquire = IntNo == Intrinsic::arm_ldaexd;
+      unsigned NewOpc = isThumb ? (IsAcquire ? ARM::t2LDAEXD : ARM::t2LDREXD)
+                                : (IsAcquire ? ARM::LDAEXD : ARM::LDREXD);
 
       // arm_ldrexd returns a i64 value in {i32, i32}
       std::vector<EVT> ResTys;
@@ -3030,28 +2974,27 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
       // Place arguments in the right order.
       SmallVector<SDValue, 7> Ops;
       Ops.push_back(MemAddr);
-      Ops.push_back(getAL(CurDAG));
+      Ops.push_back(getAL(CurDAG, dl));
       Ops.push_back(CurDAG->getRegister(0, MVT::i32));
       Ops.push_back(Chain);
-      SDNode *Ld = CurDAG->getMachineNode(NewOpc, dl, ResTys, Ops.data(),
-                                          Ops.size());
+      SDNode *Ld = CurDAG->getMachineNode(NewOpc, dl, ResTys, Ops);
       // Transfer memoperands.
       MachineSDNode::mmo_iterator MemOp = MF->allocateMemRefsArray(1);
       MemOp[0] = cast<MemIntrinsicSDNode>(N)->getMemOperand();
       cast<MachineSDNode>(Ld)->setMemRefs(MemOp, MemOp + 1);
 
       // Remap uses.
-      SDValue Glue = isThumb ? SDValue(Ld, 2) : SDValue(Ld, 1);
+      SDValue OutChain = isThumb ? SDValue(Ld, 2) : SDValue(Ld, 1);
       if (!SDValue(N, 0).use_empty()) {
         SDValue Result;
         if (isThumb)
           Result = SDValue(Ld, 0);
         else {
-          SDValue SubRegIdx = CurDAG->getTargetConstant(ARM::gsub_0, MVT::i32);
+          SDValue SubRegIdx =
+            CurDAG->getTargetConstant(ARM::gsub_0, dl, MVT::i32);
           SDNode *ResNode = CurDAG->getMachineNode(TargetOpcode::EXTRACT_SUBREG,
-              dl, MVT::i32, MVT::Glue, SDValue(Ld, 0), SubRegIdx, Glue);
+              dl, MVT::i32, SDValue(Ld, 0), SubRegIdx);
           Result = SDValue(ResNode,0);
-          Glue = Result.getValue(1);
         }
         ReplaceUses(SDValue(N, 0), Result);
       }
@@ -3060,20 +3003,20 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
         if (isThumb)
           Result = SDValue(Ld, 1);
         else {
-          SDValue SubRegIdx = CurDAG->getTargetConstant(ARM::gsub_1, MVT::i32);
+          SDValue SubRegIdx =
+            CurDAG->getTargetConstant(ARM::gsub_1, dl, MVT::i32);
           SDNode *ResNode = CurDAG->getMachineNode(TargetOpcode::EXTRACT_SUBREG,
-              dl, MVT::i32, MVT::Glue, SDValue(Ld, 0), SubRegIdx, Glue);
+              dl, MVT::i32, SDValue(Ld, 0), SubRegIdx);
           Result = SDValue(ResNode,0);
-          Glue = Result.getValue(1);
         }
         ReplaceUses(SDValue(N, 1), Result);
       }
-      ReplaceUses(SDValue(N, 2), Glue);
-      return NULL;
+      ReplaceUses(SDValue(N, 2), OutChain);
+      return nullptr;
     }
-
+    case Intrinsic::arm_stlexd:
     case Intrinsic::arm_strexd: {
-      DebugLoc dl = N->getDebugLoc();
+      SDLoc dl(N);
       SDValue Chain = N->getOperand(0);
       SDValue Val0 = N->getOperand(2);
       SDValue Val1 = N->getOperand(3);
@@ -3081,9 +3024,7 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
 
       // Store exclusive double return a i32 value which is the return status
       // of the issued store.
-      std::vector<EVT> ResTys;
-      ResTys.push_back(MVT::i32);
-      ResTys.push_back(MVT::Other);
+      const EVT ResTys[] = {MVT::i32, MVT::Other};
 
       bool isThumb = Subtarget->isThumb() && Subtarget->hasThumb2();
       // Place arguments in the right order.
@@ -3095,14 +3036,15 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
         // arm_strexd uses GPRPair.
         Ops.push_back(SDValue(createGPRPairNode(MVT::Untyped, Val0, Val1), 0));
       Ops.push_back(MemAddr);
-      Ops.push_back(getAL(CurDAG));
+      Ops.push_back(getAL(CurDAG, dl));
       Ops.push_back(CurDAG->getRegister(0, MVT::i32));
       Ops.push_back(Chain);
 
-      unsigned NewOpc = isThumb ? ARM::t2STREXD : ARM::STREXD;
+      bool IsRelease = IntNo == Intrinsic::arm_stlexd;
+      unsigned NewOpc = isThumb ? (IsRelease ? ARM::t2STLEXD : ARM::t2STREXD)
+                                : (IsRelease ? ARM::STLEXD : ARM::STREXD);
 
-      SDNode *St = CurDAG->getMachineNode(NewOpc, dl, ResTys, Ops.data(),
-                                          Ops.size());
+      SDNode *St = CurDAG->getMachineNode(NewOpc, dl, ResTys, Ops);
       // Transfer memoperands.
       MachineSDNode::mmo_iterator MemOp = MF->allocateMemRefsArray(1);
       MemOp[0] = cast<MemIntrinsicSDNode>(N)->getMemOperand();
@@ -3116,7 +3058,7 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
                                            ARM::VLD1d32, ARM::VLD1d64 };
       static const uint16_t QOpcodes[] = { ARM::VLD1q8, ARM::VLD1q16,
                                            ARM::VLD1q32, ARM::VLD1q64};
-      return SelectVLD(N, false, 1, DOpcodes, QOpcodes, 0);
+      return SelectVLD(N, false, 1, DOpcodes, QOpcodes, nullptr);
     }
 
     case Intrinsic::arm_neon_vld2: {
@@ -3124,7 +3066,7 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
                                            ARM::VLD2d32, ARM::VLD1q64 };
       static const uint16_t QOpcodes[] = { ARM::VLD2q8Pseudo, ARM::VLD2q16Pseudo,
                                            ARM::VLD2q32Pseudo };
-      return SelectVLD(N, false, 2, DOpcodes, QOpcodes, 0);
+      return SelectVLD(N, false, 2, DOpcodes, QOpcodes, nullptr);
     }
 
     case Intrinsic::arm_neon_vld3: {
@@ -3187,7 +3129,7 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
                                            ARM::VST1d32, ARM::VST1d64 };
       static const uint16_t QOpcodes[] = { ARM::VST1q8, ARM::VST1q16,
                                            ARM::VST1q32, ARM::VST1q64 };
-      return SelectVST(N, false, 1, DOpcodes, QOpcodes, 0);
+      return SelectVST(N, false, 1, DOpcodes, QOpcodes, nullptr);
     }
 
     case Intrinsic::arm_neon_vst2: {
@@ -3195,7 +3137,7 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
                                            ARM::VST2d32, ARM::VST1q64 };
       static uint16_t QOpcodes[] = { ARM::VST2q8Pseudo, ARM::VST2q16Pseudo,
                                      ARM::VST2q32Pseudo };
-      return SelectVST(N, false, 2, DOpcodes, QOpcodes, 0);
+      return SelectVST(N, false, 2, DOpcodes, QOpcodes, nullptr);
     }
 
     case Intrinsic::arm_neon_vst3: {
@@ -3280,18 +3222,18 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
   }
 
   case ARMISD::VTBL1: {
-    DebugLoc dl = N->getDebugLoc();
+    SDLoc dl(N);
     EVT VT = N->getValueType(0);
     SmallVector<SDValue, 6> Ops;
 
     Ops.push_back(N->getOperand(0));
     Ops.push_back(N->getOperand(1));
-    Ops.push_back(getAL(CurDAG));                    // Predicate
+    Ops.push_back(getAL(CurDAG, dl));                // Predicate
     Ops.push_back(CurDAG->getRegister(0, MVT::i32)); // Predicate Register
-    return CurDAG->getMachineNode(ARM::VTBL1, dl, VT, Ops.data(), Ops.size());
+    return CurDAG->getMachineNode(ARM::VTBL1, dl, VT, Ops);
   }
   case ARMISD::VTBL2: {
-    DebugLoc dl = N->getDebugLoc();
+    SDLoc dl(N);
     EVT VT = N->getValueType(0);
 
     // Form a REG_SEQUENCE to force register allocation.
@@ -3302,54 +3244,609 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
     SmallVector<SDValue, 6> Ops;
     Ops.push_back(RegSeq);
     Ops.push_back(N->getOperand(2));
-    Ops.push_back(getAL(CurDAG));                    // Predicate
+    Ops.push_back(getAL(CurDAG, dl));                // Predicate
     Ops.push_back(CurDAG->getRegister(0, MVT::i32)); // Predicate Register
-    return CurDAG->getMachineNode(ARM::VTBL2, dl, VT,
-                                  Ops.data(), Ops.size());
+    return CurDAG->getMachineNode(ARM::VTBL2, dl, VT, Ops);
   }
 
   case ISD::CONCAT_VECTORS:
     return SelectConcatVector(N);
-
-  case ARMISD::ATOMOR64_DAG:
-    return SelectAtomic64(N, ARM::ATOMOR6432);
-  case ARMISD::ATOMXOR64_DAG:
-    return SelectAtomic64(N, ARM::ATOMXOR6432);
-  case ARMISD::ATOMADD64_DAG:
-    return SelectAtomic64(N, ARM::ATOMADD6432);
-  case ARMISD::ATOMSUB64_DAG:
-    return SelectAtomic64(N, ARM::ATOMSUB6432);
-  case ARMISD::ATOMNAND64_DAG:
-    return SelectAtomic64(N, ARM::ATOMNAND6432);
-  case ARMISD::ATOMAND64_DAG:
-    return SelectAtomic64(N, ARM::ATOMAND6432);
-  case ARMISD::ATOMSWAP64_DAG:
-    return SelectAtomic64(N, ARM::ATOMSWAP6432);
-  case ARMISD::ATOMCMPXCHG64_DAG:
-    return SelectAtomic64(N, ARM::ATOMCMPXCHG6432);
-
-  case ARMISD::ATOMMIN64_DAG:
-    return SelectAtomic64(N, ARM::ATOMMIN6432);
-  case ARMISD::ATOMUMIN64_DAG:
-    return SelectAtomic64(N, ARM::ATOMUMIN6432);
-  case ARMISD::ATOMMAX64_DAG:
-    return SelectAtomic64(N, ARM::ATOMMAX6432);
-  case ARMISD::ATOMUMAX64_DAG:
-    return SelectAtomic64(N, ARM::ATOMUMAX6432);
   }
 
   return SelectCode(N);
 }
 
+// Inspect a register string of the form
+// cp<coprocessor>:<opc1>:c<CRn>:c<CRm>:<opc2> (32bit) or
+// cp<coprocessor>:<opc1>:c<CRm> (64bit) inspect the fields of the string
+// and obtain the integer operands from them, adding these operands to the
+// provided vector.
+static void getIntOperandsFromRegisterString(StringRef RegString,
+                                             SelectionDAG *CurDAG, SDLoc DL,
+                                             std::vector<SDValue>& Ops) {
+  SmallVector<StringRef, 5> Fields;
+  RegString.split(Fields, ':');
+
+  if (Fields.size() > 1) {
+    bool AllIntFields = true;
+
+    for (StringRef Field : Fields) {
+      // Need to trim out leading 'cp' characters and get the integer field.
+      unsigned IntField;
+      AllIntFields &= !Field.trim("CPcp").getAsInteger(10, IntField);
+      Ops.push_back(CurDAG->getTargetConstant(IntField, DL, MVT::i32));
+    }
+
+    assert(AllIntFields &&
+            "Unexpected non-integer value in special register string.");
+  }
+}
+
+// Maps a Banked Register string to its mask value. The mask value returned is
+// for use in the MRSbanked / MSRbanked instruction nodes as the Banked Register
+// mask operand, which expresses which register is to be used, e.g. r8, and in
+// which mode it is to be used, e.g. usr. Returns -1 to signify that the string
+// was invalid.
+static inline int getBankedRegisterMask(StringRef RegString) {
+  return StringSwitch<int>(RegString.lower())
+          .Case("r8_usr", 0x00)
+          .Case("r9_usr", 0x01)
+          .Case("r10_usr", 0x02)
+          .Case("r11_usr", 0x03)
+          .Case("r12_usr", 0x04)
+          .Case("sp_usr", 0x05)
+          .Case("lr_usr", 0x06)
+          .Case("r8_fiq", 0x08)
+          .Case("r9_fiq", 0x09)
+          .Case("r10_fiq", 0x0a)
+          .Case("r11_fiq", 0x0b)
+          .Case("r12_fiq", 0x0c)
+          .Case("sp_fiq", 0x0d)
+          .Case("lr_fiq", 0x0e)
+          .Case("lr_irq", 0x10)
+          .Case("sp_irq", 0x11)
+          .Case("lr_svc", 0x12)
+          .Case("sp_svc", 0x13)
+          .Case("lr_abt", 0x14)
+          .Case("sp_abt", 0x15)
+          .Case("lr_und", 0x16)
+          .Case("sp_und", 0x17)
+          .Case("lr_mon", 0x1c)
+          .Case("sp_mon", 0x1d)
+          .Case("elr_hyp", 0x1e)
+          .Case("sp_hyp", 0x1f)
+          .Case("spsr_fiq", 0x2e)
+          .Case("spsr_irq", 0x30)
+          .Case("spsr_svc", 0x32)
+          .Case("spsr_abt", 0x34)
+          .Case("spsr_und", 0x36)
+          .Case("spsr_mon", 0x3c)
+          .Case("spsr_hyp", 0x3e)
+          .Default(-1);
+}
+
+// Maps a MClass special register string to its value for use in the
+// t2MRS_M / t2MSR_M instruction nodes as the SYSm value operand.
+// Returns -1 to signify that the string was invalid.
+static inline int getMClassRegisterSYSmValueMask(StringRef RegString) {
+  return StringSwitch<int>(RegString.lower())
+          .Case("apsr", 0x0)
+          .Case("iapsr", 0x1)
+          .Case("eapsr", 0x2)
+          .Case("xpsr", 0x3)
+          .Case("ipsr", 0x5)
+          .Case("epsr", 0x6)
+          .Case("iepsr", 0x7)
+          .Case("msp", 0x8)
+          .Case("psp", 0x9)
+          .Case("primask", 0x10)
+          .Case("basepri", 0x11)
+          .Case("basepri_max", 0x12)
+          .Case("faultmask", 0x13)
+          .Case("control", 0x14)
+          .Default(-1);
+}
+
+// The flags here are common to those allowed for apsr in the A class cores and
+// those allowed for the special registers in the M class cores. Returns a
+// value representing which flags were present, -1 if invalid.
+static inline int getMClassFlagsMask(StringRef Flags, bool hasThumb2DSP) {
+  if (Flags.empty())
+    return 0x2 | (int)hasThumb2DSP;
+
+  return StringSwitch<int>(Flags)
+          .Case("g", 0x1)
+          .Case("nzcvq", 0x2)
+          .Case("nzcvqg", 0x3)
+          .Default(-1);
+}
+
+static int getMClassRegisterMask(StringRef Reg, StringRef Flags, bool IsRead,
+                                 const ARMSubtarget *Subtarget) {
+  // Ensure that the register (without flags) was a valid M Class special
+  // register.
+  int SYSmvalue = getMClassRegisterSYSmValueMask(Reg);
+  if (SYSmvalue == -1)
+    return -1;
+
+  // basepri, basepri_max and faultmask are only valid for V7m.
+  if (!Subtarget->hasV7Ops() && SYSmvalue >= 0x11 && SYSmvalue <= 0x13)
+    return -1;
+
+  // If it was a read then we won't be expecting flags and so at this point
+  // we can return the mask.
+  if (IsRead) {
+    assert (Flags.empty() && "Unexpected flags for reading M class register.");
+    return SYSmvalue;
+  }
+
+  // We know we are now handling a write so need to get the mask for the flags.
+  int Mask = getMClassFlagsMask(Flags, Subtarget->hasThumb2DSP());
+
+  // Only apsr, iapsr, eapsr, xpsr can have flags. The other register values
+  // shouldn't have flags present.
+  if ((SYSmvalue < 0x4 && Mask == -1) || (SYSmvalue > 0x4 && !Flags.empty()))
+    return -1;
+
+  // The _g and _nzcvqg versions are only valid if the DSP extension is
+  // available.
+  if (!Subtarget->hasThumb2DSP() && (Mask & 0x1))
+    return -1;
+
+  // The register was valid so need to put the mask in the correct place
+  // (the flags need to be in bits 11-10) and combine with the SYSmvalue to
+  // construct the operand for the instruction node.
+  if (SYSmvalue < 0x4)
+    return SYSmvalue | Mask << 10;
+
+  return SYSmvalue;
+}
+
+static int getARClassRegisterMask(StringRef Reg, StringRef Flags) {
+  // The mask operand contains the special register (R Bit) in bit 4, whether
+  // the register is spsr (R bit is 1) or one of cpsr/apsr (R bit is 0), and
+  // bits 3-0 contains the fields to be accessed in the special register, set by
+  // the flags provided with the register.
+  int Mask = 0;
+  if (Reg == "apsr") {
+    // The flags permitted for apsr are the same flags that are allowed in
+    // M class registers. We get the flag value and then shift the flags into
+    // the correct place to combine with the mask.
+    Mask = getMClassFlagsMask(Flags, true);
+    if (Mask == -1)
+      return -1;
+    return Mask << 2;
+  }
+
+  if (Reg != "cpsr" && Reg != "spsr") {
+    return -1;
+  }
+
+  // This is the same as if the flags were "fc"
+  if (Flags.empty() || Flags == "all")
+    return Mask | 0x9;
+
+  // Inspect the supplied flags string and set the bits in the mask for
+  // the relevant and valid flags allowed for cpsr and spsr.
+  for (char Flag : Flags) {
+    int FlagVal;
+    switch (Flag) {
+      case 'c':
+        FlagVal = 0x1;
+        break;
+      case 'x':
+        FlagVal = 0x2;
+        break;
+      case 's':
+        FlagVal = 0x4;
+        break;
+      case 'f':
+        FlagVal = 0x8;
+        break;
+      default:
+        FlagVal = 0;
+    }
+
+    // This avoids allowing strings where the same flag bit appears twice.
+    if (!FlagVal || (Mask & FlagVal))
+      return -1;
+    Mask |= FlagVal;
+  }
+
+  // If the register is spsr then we need to set the R bit.
+  if (Reg == "spsr")
+    Mask |= 0x10;
+
+  return Mask;
+}
+
+// Lower the read_register intrinsic to ARM specific DAG nodes
+// using the supplied metadata string to select the instruction node to use
+// and the registers/masks to construct as operands for the node.
+SDNode *ARMDAGToDAGISel::SelectReadRegister(SDNode *N){
+  const MDNodeSDNode *MD = dyn_cast<MDNodeSDNode>(N->getOperand(1));
+  const MDString *RegString = dyn_cast<MDString>(MD->getMD()->getOperand(0));
+  bool IsThumb2 = Subtarget->isThumb2();
+  SDLoc DL(N);
+
+  std::vector<SDValue> Ops;
+  getIntOperandsFromRegisterString(RegString->getString(), CurDAG, DL, Ops);
+
+  if (!Ops.empty()) {
+    // If the special register string was constructed of fields (as defined
+    // in the ACLE) then need to lower to MRC node (32 bit) or
+    // MRRC node(64 bit), we can make the distinction based on the number of
+    // operands we have.
+    unsigned Opcode;
+    SmallVector<EVT, 3> ResTypes;
+    if (Ops.size() == 5){
+      Opcode = IsThumb2 ? ARM::t2MRC : ARM::MRC;
+      ResTypes.append({ MVT::i32, MVT::Other });
+    } else {
+      assert(Ops.size() == 3 &&
+              "Invalid number of fields in special register string.");
+      Opcode = IsThumb2 ? ARM::t2MRRC : ARM::MRRC;
+      ResTypes.append({ MVT::i32, MVT::i32, MVT::Other });
+    }
+
+    Ops.push_back(getAL(CurDAG, DL));
+    Ops.push_back(CurDAG->getRegister(0, MVT::i32));
+    Ops.push_back(N->getOperand(0));
+    return CurDAG->getMachineNode(Opcode, DL, ResTypes, Ops);
+  }
+
+  std::string SpecialReg = RegString->getString().lower();
+
+  int BankedReg = getBankedRegisterMask(SpecialReg);
+  if (BankedReg != -1) {
+    Ops = { CurDAG->getTargetConstant(BankedReg, DL, MVT::i32),
+            getAL(CurDAG, DL), CurDAG->getRegister(0, MVT::i32),
+            N->getOperand(0) };
+    return CurDAG->getMachineNode(IsThumb2 ? ARM::t2MRSbanked : ARM::MRSbanked,
+                                  DL, MVT::i32, MVT::Other, Ops);
+  }
+
+  // The VFP registers are read by creating SelectionDAG nodes with opcodes
+  // corresponding to the register that is being read from. So we switch on the
+  // string to find which opcode we need to use.
+  unsigned Opcode = StringSwitch<unsigned>(SpecialReg)
+                    .Case("fpscr", ARM::VMRS)
+                    .Case("fpexc", ARM::VMRS_FPEXC)
+                    .Case("fpsid", ARM::VMRS_FPSID)
+                    .Case("mvfr0", ARM::VMRS_MVFR0)
+                    .Case("mvfr1", ARM::VMRS_MVFR1)
+                    .Case("mvfr2", ARM::VMRS_MVFR2)
+                    .Case("fpinst", ARM::VMRS_FPINST)
+                    .Case("fpinst2", ARM::VMRS_FPINST2)
+                    .Default(0);
+
+  // If an opcode was found then we can lower the read to a VFP instruction.
+  if (Opcode) {
+    if (!Subtarget->hasVFP2())
+      return nullptr;
+    if (Opcode == ARM::VMRS_MVFR2 && !Subtarget->hasFPARMv8())
+      return nullptr;
+
+    Ops = { getAL(CurDAG, DL), CurDAG->getRegister(0, MVT::i32),
+            N->getOperand(0) };
+    return CurDAG->getMachineNode(Opcode, DL, MVT::i32, MVT::Other, Ops);
+  }
+
+  // If the target is M Class then need to validate that the register string
+  // is an acceptable value, so check that a mask can be constructed from the
+  // string.
+  if (Subtarget->isMClass()) {
+    int SYSmValue = getMClassRegisterMask(SpecialReg, "", true, Subtarget);
+    if (SYSmValue == -1)
+      return nullptr;
+
+    SDValue Ops[] = { CurDAG->getTargetConstant(SYSmValue, DL, MVT::i32),
+                      getAL(CurDAG, DL), CurDAG->getRegister(0, MVT::i32),
+                      N->getOperand(0) };
+    return CurDAG->getMachineNode(ARM::t2MRS_M, DL, MVT::i32, MVT::Other, Ops);
+  }
+
+  // Here we know the target is not M Class so we need to check if it is one
+  // of the remaining possible values which are apsr, cpsr or spsr.
+  if (SpecialReg == "apsr" || SpecialReg == "cpsr") {
+    Ops = { getAL(CurDAG, DL), CurDAG->getRegister(0, MVT::i32),
+            N->getOperand(0) };
+    return CurDAG->getMachineNode(IsThumb2 ? ARM::t2MRS_AR : ARM::MRS, DL,
+                                  MVT::i32, MVT::Other, Ops);
+  }
+
+  if (SpecialReg == "spsr") {
+    Ops = { getAL(CurDAG, DL), CurDAG->getRegister(0, MVT::i32),
+            N->getOperand(0) };
+    return CurDAG->getMachineNode(IsThumb2 ? ARM::t2MRSsys_AR : ARM::MRSsys,
+                                  DL, MVT::i32, MVT::Other, Ops);
+  }
+
+  return nullptr;
+}
+
+// Lower the write_register intrinsic to ARM specific DAG nodes
+// using the supplied metadata string to select the instruction node to use
+// and the registers/masks to use in the nodes
+SDNode *ARMDAGToDAGISel::SelectWriteRegister(SDNode *N){
+  const MDNodeSDNode *MD = dyn_cast<MDNodeSDNode>(N->getOperand(1));
+  const MDString *RegString = dyn_cast<MDString>(MD->getMD()->getOperand(0));
+  bool IsThumb2 = Subtarget->isThumb2();
+  SDLoc DL(N);
+
+  std::vector<SDValue> Ops;
+  getIntOperandsFromRegisterString(RegString->getString(), CurDAG, DL, Ops);
+
+  if (!Ops.empty()) {
+    // If the special register string was constructed of fields (as defined
+    // in the ACLE) then need to lower to MCR node (32 bit) or
+    // MCRR node(64 bit), we can make the distinction based on the number of
+    // operands we have.
+    unsigned Opcode;
+    if (Ops.size() == 5) {
+      Opcode = IsThumb2 ? ARM::t2MCR : ARM::MCR;
+      Ops.insert(Ops.begin()+2, N->getOperand(2));
+    } else {
+      assert(Ops.size() == 3 &&
+              "Invalid number of fields in special register string.");
+      Opcode = IsThumb2 ? ARM::t2MCRR : ARM::MCRR;
+      SDValue WriteValue[] = { N->getOperand(2), N->getOperand(3) };
+      Ops.insert(Ops.begin()+2, WriteValue, WriteValue+2);
+    }
+
+    Ops.push_back(getAL(CurDAG, DL));
+    Ops.push_back(CurDAG->getRegister(0, MVT::i32));
+    Ops.push_back(N->getOperand(0));
+
+    return CurDAG->getMachineNode(Opcode, DL, MVT::Other, Ops);
+  }
+
+  std::string SpecialReg = RegString->getString().lower();
+  int BankedReg = getBankedRegisterMask(SpecialReg);
+  if (BankedReg != -1) {
+    Ops = { CurDAG->getTargetConstant(BankedReg, DL, MVT::i32), N->getOperand(2),
+            getAL(CurDAG, DL), CurDAG->getRegister(0, MVT::i32),
+            N->getOperand(0) };
+    return CurDAG->getMachineNode(IsThumb2 ? ARM::t2MSRbanked : ARM::MSRbanked,
+                                  DL, MVT::Other, Ops);
+  }
+
+  // The VFP registers are written to by creating SelectionDAG nodes with
+  // opcodes corresponding to the register that is being written. So we switch
+  // on the string to find which opcode we need to use.
+  unsigned Opcode = StringSwitch<unsigned>(SpecialReg)
+                    .Case("fpscr", ARM::VMSR)
+                    .Case("fpexc", ARM::VMSR_FPEXC)
+                    .Case("fpsid", ARM::VMSR_FPSID)
+                    .Case("fpinst", ARM::VMSR_FPINST)
+                    .Case("fpinst2", ARM::VMSR_FPINST2)
+                    .Default(0);
+
+  if (Opcode) {
+    if (!Subtarget->hasVFP2())
+      return nullptr;
+    Ops = { N->getOperand(2), getAL(CurDAG, DL),
+            CurDAG->getRegister(0, MVT::i32), N->getOperand(0) };
+    return CurDAG->getMachineNode(Opcode, DL, MVT::Other, Ops);
+  }
+
+  SmallVector<StringRef, 5> Fields;
+  StringRef(SpecialReg).split(Fields, '_', 1, false);
+  std::string Reg = Fields[0].str();
+  StringRef Flags = Fields.size() == 2 ? Fields[1] : "";
+
+  // If the target was M Class then need to validate the special register value
+  // and retrieve the mask for use in the instruction node.
+  if (Subtarget->isMClass()) {
+    // basepri_max gets split so need to correct Reg and Flags.
+    if (SpecialReg == "basepri_max") {
+      Reg = SpecialReg;
+      Flags = "";
+    }
+    int SYSmValue = getMClassRegisterMask(Reg, Flags, false, Subtarget);
+    if (SYSmValue == -1)
+      return nullptr;
+
+    SDValue Ops[] = { CurDAG->getTargetConstant(SYSmValue, DL, MVT::i32),
+                      N->getOperand(2), getAL(CurDAG, DL),
+                      CurDAG->getRegister(0, MVT::i32), N->getOperand(0) };
+    return CurDAG->getMachineNode(ARM::t2MSR_M, DL, MVT::Other, Ops);
+  }
+
+  // We then check to see if a valid mask can be constructed for one of the
+  // register string values permitted for the A and R class cores. These values
+  // are apsr, spsr and cpsr; these are also valid on older cores.
+  int Mask = getARClassRegisterMask(Reg, Flags);
+  if (Mask != -1) {
+    Ops = { CurDAG->getTargetConstant(Mask, DL, MVT::i32), N->getOperand(2),
+            getAL(CurDAG, DL), CurDAG->getRegister(0, MVT::i32),
+            N->getOperand(0) };
+    return CurDAG->getMachineNode(IsThumb2 ? ARM::t2MSR_AR : ARM::MSR,
+                                  DL, MVT::Other, Ops);
+  }
+
+  return nullptr;
+}
+
+SDNode *ARMDAGToDAGISel::SelectInlineAsm(SDNode *N){
+  std::vector<SDValue> AsmNodeOperands;
+  unsigned Flag, Kind;
+  bool Changed = false;
+  unsigned NumOps = N->getNumOperands();
+
+  // Normally, i64 data is bounded to two arbitrary GRPs for "%r" constraint.
+  // However, some instrstions (e.g. ldrexd/strexd in ARM mode) require
+  // (even/even+1) GPRs and use %n and %Hn to refer to the individual regs
+  // respectively. Since there is no constraint to explicitly specify a
+  // reg pair, we use GPRPair reg class for "%r" for 64-bit data. For Thumb,
+  // the 64-bit data may be referred by H, Q, R modifiers, so we still pack
+  // them into a GPRPair.
+
+  SDLoc dl(N);
+  SDValue Glue = N->getGluedNode() ? N->getOperand(NumOps-1)
+                                   : SDValue(nullptr,0);
+
+  SmallVector<bool, 8> OpChanged;
+  // Glue node will be appended late.
+  for(unsigned i = 0, e = N->getGluedNode() ? NumOps - 1 : NumOps; i < e; ++i) {
+    SDValue op = N->getOperand(i);
+    AsmNodeOperands.push_back(op);
+
+    if (i < InlineAsm::Op_FirstOperand)
+      continue;
+
+    if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(N->getOperand(i))) {
+      Flag = C->getZExtValue();
+      Kind = InlineAsm::getKind(Flag);
+    }
+    else
+      continue;
+
+    // Immediate operands to inline asm in the SelectionDAG are modeled with
+    // two operands. The first is a constant of value InlineAsm::Kind_Imm, and
+    // the second is a constant with the value of the immediate. If we get here
+    // and we have a Kind_Imm, skip the next operand, and continue.
+    if (Kind == InlineAsm::Kind_Imm) {
+      SDValue op = N->getOperand(++i);
+      AsmNodeOperands.push_back(op);
+      continue;
+    }
+
+    unsigned NumRegs = InlineAsm::getNumOperandRegisters(Flag);
+    if (NumRegs)
+      OpChanged.push_back(false);
+
+    unsigned DefIdx = 0;
+    bool IsTiedToChangedOp = false;
+    // If it's a use that is tied with a previous def, it has no
+    // reg class constraint.
+    if (Changed && InlineAsm::isUseOperandTiedToDef(Flag, DefIdx))
+      IsTiedToChangedOp = OpChanged[DefIdx];
+
+    if (Kind != InlineAsm::Kind_RegUse && Kind != InlineAsm::Kind_RegDef
+        && Kind != InlineAsm::Kind_RegDefEarlyClobber)
+      continue;
+
+    unsigned RC;
+    bool HasRC = InlineAsm::hasRegClassConstraint(Flag, RC);
+    if ((!IsTiedToChangedOp && (!HasRC || RC != ARM::GPRRegClassID))
+        || NumRegs != 2)
+      continue;
+
+    assert((i+2 < NumOps) && "Invalid number of operands in inline asm");
+    SDValue V0 = N->getOperand(i+1);
+    SDValue V1 = N->getOperand(i+2);
+    unsigned Reg0 = cast<RegisterSDNode>(V0)->getReg();
+    unsigned Reg1 = cast<RegisterSDNode>(V1)->getReg();
+    SDValue PairedReg;
+    MachineRegisterInfo &MRI = MF->getRegInfo();
+
+    if (Kind == InlineAsm::Kind_RegDef ||
+        Kind == InlineAsm::Kind_RegDefEarlyClobber) {
+      // Replace the two GPRs with 1 GPRPair and copy values from GPRPair to
+      // the original GPRs.
+
+      unsigned GPVR = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
+      PairedReg = CurDAG->getRegister(GPVR, MVT::Untyped);
+      SDValue Chain = SDValue(N,0);
+
+      SDNode *GU = N->getGluedUser();
+      SDValue RegCopy = CurDAG->getCopyFromReg(Chain, dl, GPVR, MVT::Untyped,
+                                               Chain.getValue(1));
+
+      // Extract values from a GPRPair reg and copy to the original GPR reg.
+      SDValue Sub0 = CurDAG->getTargetExtractSubreg(ARM::gsub_0, dl, MVT::i32,
+                                                    RegCopy);
+      SDValue Sub1 = CurDAG->getTargetExtractSubreg(ARM::gsub_1, dl, MVT::i32,
+                                                    RegCopy);
+      SDValue T0 = CurDAG->getCopyToReg(Sub0, dl, Reg0, Sub0,
+                                        RegCopy.getValue(1));
+      SDValue T1 = CurDAG->getCopyToReg(Sub1, dl, Reg1, Sub1, T0.getValue(1));
+
+      // Update the original glue user.
+      std::vector<SDValue> Ops(GU->op_begin(), GU->op_end()-1);
+      Ops.push_back(T1.getValue(1));
+      CurDAG->UpdateNodeOperands(GU, Ops);
+    }
+    else {
+      // For Kind  == InlineAsm::Kind_RegUse, we first copy two GPRs into a
+      // GPRPair and then pass the GPRPair to the inline asm.
+      SDValue Chain = AsmNodeOperands[InlineAsm::Op_InputChain];
+
+      // As REG_SEQ doesn't take RegisterSDNode, we copy them first.
+      SDValue T0 = CurDAG->getCopyFromReg(Chain, dl, Reg0, MVT::i32,
+                                          Chain.getValue(1));
+      SDValue T1 = CurDAG->getCopyFromReg(Chain, dl, Reg1, MVT::i32,
+                                          T0.getValue(1));
+      SDValue Pair = SDValue(createGPRPairNode(MVT::Untyped, T0, T1), 0);
+
+      // Copy REG_SEQ into a GPRPair-typed VR and replace the original two
+      // i32 VRs of inline asm with it.
+      unsigned GPVR = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
+      PairedReg = CurDAG->getRegister(GPVR, MVT::Untyped);
+      Chain = CurDAG->getCopyToReg(T1, dl, GPVR, Pair, T1.getValue(1));
+
+      AsmNodeOperands[InlineAsm::Op_InputChain] = Chain;
+      Glue = Chain.getValue(1);
+    }
+
+    Changed = true;
+
+    if(PairedReg.getNode()) {
+      OpChanged[OpChanged.size() -1 ] = true;
+      Flag = InlineAsm::getFlagWord(Kind, 1 /* RegNum*/);
+      if (IsTiedToChangedOp)
+        Flag = InlineAsm::getFlagWordForMatchingOp(Flag, DefIdx);
+      else
+        Flag = InlineAsm::getFlagWordForRegClass(Flag, ARM::GPRPairRegClassID);
+      // Replace the current flag.
+      AsmNodeOperands[AsmNodeOperands.size() -1] = CurDAG->getTargetConstant(
+          Flag, dl, MVT::i32);
+      // Add the new register node and skip the original two GPRs.
+      AsmNodeOperands.push_back(PairedReg);
+      // Skip the next two GPRs.
+      i += 2;
+    }
+  }
+
+  if (Glue.getNode())
+    AsmNodeOperands.push_back(Glue);
+  if (!Changed)
+    return nullptr;
+
+  SDValue New = CurDAG->getNode(ISD::INLINEASM, SDLoc(N),
+      CurDAG->getVTList(MVT::Other, MVT::Glue), AsmNodeOperands);
+  New->setNodeId(-1);
+  return New.getNode();
+}
+
+
 bool ARMDAGToDAGISel::
-SelectInlineAsmMemoryOperand(const SDValue &Op, char ConstraintCode,
+SelectInlineAsmMemoryOperand(const SDValue &Op, unsigned ConstraintID,
                              std::vector<SDValue> &OutOps) {
-  assert(ConstraintCode == 'm' && "unexpected asm memory constraint");
-  // Require the address to be in a register.  That is safe for all ARM
-  // variants and it is hard to do anything much smarter without knowing
-  // how the operand is used.
-  OutOps.push_back(Op);
-  return false;
+  switch(ConstraintID) {
+  default:
+    llvm_unreachable("Unexpected asm memory constraint");
+  case InlineAsm::Constraint_i:
+    // FIXME: It seems strange that 'i' is needed here since it's supposed to
+    //        be an immediate and not a memory constraint.
+    // Fallthrough.
+  case InlineAsm::Constraint_m:
+  case InlineAsm::Constraint_Q:
+  case InlineAsm::Constraint_Um:
+  case InlineAsm::Constraint_Un:
+  case InlineAsm::Constraint_Uq:
+  case InlineAsm::Constraint_Us:
+  case InlineAsm::Constraint_Ut:
+  case InlineAsm::Constraint_Uv:
+  case InlineAsm::Constraint_Uy:
+    // Require the address to be in a register.  That is safe for all ARM
+    // variants and it is hard to do anything much smarter without knowing
+    // how the operand is used.
+    OutOps.push_back(Op);
+    return false;
+  }
+  return true;
 }
 
 /// createARMISelDag - This pass converts a legalized DAG into a