diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp
index 3d283a5ac95..6beb1abd222 100644
--- a/lib/Target/ARM/ARMISelLowering.cpp
+++ b/lib/Target/ARM/ARMISelLowering.cpp
@@ -5969,9 +5969,6 @@ ARMTargetLowering::EmitAtomicBinary64(MachineInstr *MI, MachineBasicBlock *BB,
     MRI.constrainRegClass(ptr, &ARM::rGPRRegClass);
   }
 
-  unsigned ldrOpc = isThumb2 ? ARM::t2LDREXD : ARM::LDREXD;
-  unsigned strOpc = isThumb2 ? ARM::t2STREXD : ARM::STREXD;
-
   MachineBasicBlock *loopMBB = MF->CreateMachineBasicBlock(LLVM_BB);
   MachineBasicBlock *contBB = 0, *cont2BB = 0;
   if (IsCmpxchg || IsMinMax)
@@ -6009,42 +6006,26 @@ ARMTargetLowering::EmitAtomicBinary64(MachineInstr *MI, MachineBasicBlock *BB,
   //   cmp storesuccess, #0
   //   bne- loopMBB
   //   fallthrough --> exitMBB
-  //
-  // Note that the registers are explicitly specified because there is not any
-  // way to force the register allocator to allocate a register pair.
-  //
-  // FIXME: The hardcoded registers are not necessary for Thumb2, but we
-  // need to properly enforce the restriction that the two output registers
-  // for ldrexd must be different.
   BB = loopMBB;
+
   // Load
-  unsigned GPRPair0 = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
-  unsigned GPRPair1 = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
-  unsigned GPRPair2;
-  if (IsMinMax) {
-    //We need an extra double register for doing min/max.
-    unsigned undef = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
-    unsigned r1 = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
-    GPRPair2 = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
-    BuildMI(BB, dl, TII->get(TargetOpcode::IMPLICIT_DEF), undef);
-    BuildMI(BB, dl, TII->get(TargetOpcode::INSERT_SUBREG), r1)
-      .addReg(undef)
-      .addReg(vallo)
-      .addImm(ARM::gsub_0);
-    BuildMI(BB, dl, TII->get(TargetOpcode::INSERT_SUBREG), GPRPair2)
-      .addReg(r1)
-      .addReg(valhi)
-      .addImm(ARM::gsub_1);
+  if (isThumb2) {
+    AddDefaultPred(BuildMI(BB, dl, TII->get(ARM::t2LDREXD))
+                   .addReg(destlo, RegState::Define)
+                   .addReg(desthi, RegState::Define)
+                   .addReg(ptr));
+  } else {
+    unsigned GPRPair0 = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
+    AddDefaultPred(BuildMI(BB, dl, TII->get(ARM::LDREXD))
+                   .addReg(GPRPair0, RegState::Define).addReg(ptr));
+    // Copy r2/r3 into dest.  (This copy will normally be coalesced.)
+    BuildMI(BB, dl, TII->get(TargetOpcode::COPY), destlo)
+      .addReg(GPRPair0, 0, ARM::gsub_0);
+    BuildMI(BB, dl, TII->get(TargetOpcode::COPY), desthi)
+      .addReg(GPRPair0, 0, ARM::gsub_1);
   }
 
-  AddDefaultPred(BuildMI(BB, dl, TII->get(ldrOpc))
-                 .addReg(GPRPair0, RegState::Define).addReg(ptr));
-  // Copy r2/r3 into dest.  (This copy will normally be coalesced.)
-  BuildMI(BB, dl, TII->get(TargetOpcode::COPY), destlo)
-    .addReg(GPRPair0, 0, ARM::gsub_0);
-  BuildMI(BB, dl, TII->get(TargetOpcode::COPY), desthi)
-    .addReg(GPRPair0, 0, ARM::gsub_1);
-
+  unsigned StoreLo, StoreHi;
   if (IsCmpxchg) {
     // Add early exit
     for (unsigned i = 0; i < 2; i++) {
@@ -6060,19 +6041,8 @@ ARMTargetLowering::EmitAtomicBinary64(MachineInstr *MI, MachineBasicBlock *BB,
     }
 
     // Copy to physregs for strexd
-    unsigned setlo = MI->getOperand(5).getReg();
-    unsigned sethi = MI->getOperand(6).getReg();
-    unsigned undef = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
-    unsigned r1 = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
-    BuildMI(BB, dl, TII->get(TargetOpcode::IMPLICIT_DEF), undef);
-    BuildMI(BB, dl, TII->get(TargetOpcode::INSERT_SUBREG), r1)
-      .addReg(undef)
-      .addReg(setlo)
-      .addImm(ARM::gsub_0);
-    BuildMI(BB, dl, TII->get(TargetOpcode::INSERT_SUBREG), GPRPair1)
-      .addReg(r1)
-      .addReg(sethi)
-      .addImm(ARM::gsub_1);
+    StoreLo = MI->getOperand(5).getReg();
+    StoreHi = MI->getOperand(6).getReg();
   } else if (Op1) {
     // Perform binary operation
     unsigned tmpRegLo = MRI.createVirtualRegister(TRC);
@@ -6084,32 +6054,13 @@ ARMTargetLowering::EmitAtomicBinary64(MachineInstr *MI, MachineBasicBlock *BB,
                    .addReg(desthi).addReg(valhi))
         .addReg(IsMinMax ? ARM::CPSR : 0, getDefRegState(IsMinMax));
 
-    unsigned UndefPair = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
-    BuildMI(BB, dl, TII->get(TargetOpcode::IMPLICIT_DEF), UndefPair);
-    unsigned r1 = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
-    BuildMI(BB, dl, TII->get(TargetOpcode::INSERT_SUBREG), r1)
-      .addReg(UndefPair)
-      .addReg(tmpRegLo)
-      .addImm(ARM::gsub_0);
-    BuildMI(BB, dl, TII->get(TargetOpcode::INSERT_SUBREG), GPRPair1)
-      .addReg(r1)
-      .addReg(tmpRegHi)
-      .addImm(ARM::gsub_1);
+    StoreLo = tmpRegLo;
+    StoreHi = tmpRegHi;
   } else {
     // Copy to physregs for strexd
-    unsigned UndefPair = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
-    unsigned r1 = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
-    BuildMI(BB, dl, TII->get(TargetOpcode::IMPLICIT_DEF), UndefPair);
-    BuildMI(BB, dl, TII->get(TargetOpcode::INSERT_SUBREG), r1)
-      .addReg(UndefPair)
-      .addReg(vallo)
-      .addImm(ARM::gsub_0);
-    BuildMI(BB, dl, TII->get(TargetOpcode::INSERT_SUBREG), GPRPair1)
-      .addReg(r1)
-      .addReg(valhi)
-      .addImm(ARM::gsub_1);
+    StoreLo = vallo;
+    StoreHi = valhi;
   }
-  unsigned GPRPairStore = GPRPair1;
   if (IsMinMax) {
     // Compare and branch to exit block.
     BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc))
@@ -6117,12 +6068,33 @@ ARMTargetLowering::EmitAtomicBinary64(MachineInstr *MI, MachineBasicBlock *BB,
     BB->addSuccessor(exitMBB);
     BB->addSuccessor(contBB);
     BB = contBB;
-    GPRPairStore = GPRPair2;
+    StoreLo = vallo;
+    StoreHi = valhi;
   }
 
   // Store
-  AddDefaultPred(BuildMI(BB, dl, TII->get(strOpc), storesuccess)
-                 .addReg(GPRPairStore).addReg(ptr));
+  if (isThumb2) {
+    AddDefaultPred(BuildMI(BB, dl, TII->get(ARM::t2STREXD), storesuccess)
+                   .addReg(StoreLo).addReg(StoreHi).addReg(ptr));
+  } else {
+    // Marshal a pair...
+    unsigned StorePair = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
+    unsigned UndefPair = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
+    unsigned r1 = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
+    BuildMI(BB, dl, TII->get(TargetOpcode::IMPLICIT_DEF), UndefPair);
+    BuildMI(BB, dl, TII->get(TargetOpcode::INSERT_SUBREG), r1)
+      .addReg(UndefPair)
+      .addReg(StoreLo)
+      .addImm(ARM::gsub_0);
+    BuildMI(BB, dl, TII->get(TargetOpcode::INSERT_SUBREG), StorePair)
+      .addReg(r1)
+      .addReg(StoreHi)
+      .addImm(ARM::gsub_1);
+
+    // ...and store it
+    AddDefaultPred(BuildMI(BB, dl, TII->get(ARM::STREXD), storesuccess)
+                   .addReg(StorePair).addReg(ptr));
+  }
   // Cmp+jump
   AddDefaultPred(BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2CMPri : ARM::CMPri))
                  .addReg(storesuccess).addImm(0));
diff --git a/test/CodeGen/ARM/atomic-64bit.ll b/test/CodeGen/ARM/atomic-64bit.ll
index 4ac5e5d3233..f2c7305ff33 100644
--- a/test/CodeGen/ARM/atomic-64bit.ll
+++ b/test/CodeGen/ARM/atomic-64bit.ll
@@ -1,4 +1,5 @@
 ; RUN: llc < %s -mtriple=armv7-apple-ios | FileCheck %s
+; RUN: llc < %s -mtriple=thumbv7-none-linux-gnueabihf | FileCheck %s --check-prefix=CHECK-THUMB
 
 define i64 @test1(i64* %ptr, i64 %val) {
 ; CHECK: test1:
@@ -10,6 +11,17 @@ define i64 @test1(i64* %ptr, i64 %val) {
 ; CHECK: cmp
 ; CHECK: bne
 ; CHECK: dmb ish
+
+; CHECK-THUMB: test1:
+; CHECK-THUMB: dmb ish
+; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
+; CHECK-THUMB: adds.w [[REG3:[a-z0-9]+]], [[REG1]]
+; CHECK-THUMB: adc.w [[REG4:[a-z0-9]+]], [[REG2]]
+; CHECK-THUMB: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
+; CHECK-THUMB: cmp
+; CHECK-THUMB: bne
+; CHECK-THUMB: dmb ish
+
   %r = atomicrmw add i64* %ptr, i64 %val seq_cst
   ret i64 %r
 }
@@ -24,6 +36,17 @@ define i64 @test2(i64* %ptr, i64 %val) {
 ; CHECK: cmp
 ; CHECK: bne
 ; CHECK: dmb ish
+
+; CHECK-THUMB: test2:
+; CHECK-THUMB: dmb ish
+; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
+; CHECK-THUMB: subs.w [[REG3:[a-z0-9]+]], [[REG1]]
+; CHECK-THUMB: sbc.w [[REG4:[a-z0-9]+]], [[REG2]]
+; CHECK-THUMB: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
+; CHECK-THUMB: cmp
+; CHECK-THUMB: bne
+; CHECK-THUMB: dmb ish
+
   %r = atomicrmw sub i64* %ptr, i64 %val seq_cst
   ret i64 %r
 }
@@ -38,6 +61,17 @@ define i64 @test3(i64* %ptr, i64 %val) {
 ; CHECK: cmp
 ; CHECK: bne
 ; CHECK: dmb ish
+
+; CHECK-THUMB: test3:
+; CHECK-THUMB: dmb ish
+; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
+; CHECK-THUMB: and.w [[REG3:[a-z0-9]+]], [[REG1]]
+; CHECK-THUMB: and.w [[REG4:[a-z0-9]+]], [[REG2]]
+; CHECK-THUMB: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
+; CHECK-THUMB: cmp
+; CHECK-THUMB: bne
+; CHECK-THUMB: dmb ish
+
   %r = atomicrmw and i64* %ptr, i64 %val seq_cst
   ret i64 %r
 }
@@ -52,6 +86,17 @@ define i64 @test4(i64* %ptr, i64 %val) {
 ; CHECK: cmp
 ; CHECK: bne
 ; CHECK: dmb ish
+
+; CHECK-THUMB: test4:
+; CHECK-THUMB: dmb ish
+; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
+; CHECK-THUMB: orr.w [[REG3:[a-z0-9]+]], [[REG1]]
+; CHECK-THUMB: orr.w [[REG4:[a-z0-9]+]], [[REG2]]
+; CHECK-THUMB: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
+; CHECK-THUMB: cmp
+; CHECK-THUMB: bne
+; CHECK-THUMB: dmb ish
+
   %r = atomicrmw or i64* %ptr, i64 %val seq_cst
   ret i64 %r
 }
@@ -66,6 +111,17 @@ define i64 @test5(i64* %ptr, i64 %val) {
 ; CHECK: cmp
 ; CHECK: bne
 ; CHECK: dmb ish
+
+; CHECK-THUMB: test5:
+; CHECK-THUMB: dmb ish
+; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
+; CHECK-THUMB: eor.w [[REG3:[a-z0-9]+]], [[REG1]]
+; CHECK-THUMB: eor.w [[REG4:[a-z0-9]+]], [[REG2]]
+; CHECK-THUMB: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
+; CHECK-THUMB: cmp
+; CHECK-THUMB: bne
+; CHECK-THUMB: dmb ish
+
   %r = atomicrmw xor i64* %ptr, i64 %val seq_cst
   ret i64 %r
 }
@@ -78,6 +134,15 @@ define i64 @test6(i64* %ptr, i64 %val) {
 ; CHECK: cmp
 ; CHECK: bne
 ; CHECK: dmb ish
+
+; CHECK-THUMB: test6:
+; CHECK-THUMB: dmb ish
+; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
+; CHECK-THUMB: strexd {{[a-z0-9]+}}, {{[a-z0-9]+}}, {{[a-z0-9]+}}
+; CHECK-THUMB: cmp
+; CHECK-THUMB: bne
+; CHECK-THUMB: dmb ish
+
   %r = atomicrmw xchg i64* %ptr, i64 %val seq_cst
   ret i64 %r
 }
@@ -93,6 +158,19 @@ define i64 @test7(i64* %ptr, i64 %val1, i64 %val2) {
 ; CHECK: cmp
 ; CHECK: bne
 ; CHECK: dmb ish
+
+; CHECK-THUMB: test7:
+; CHECK-THUMB: dmb ish
+; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
+; CHECK-THUMB: cmp [[REG1]]
+; CHECK-THUMB: it eq
+; CHECK-THUMB: cmpeq [[REG2]]
+; CHECK-THUMB: bne
+; CHECK-THUMB: strexd {{[a-z0-9]+}}, {{[a-z0-9]+}}, {{[a-z0-9]+}}
+; CHECK-THUMB: cmp
+; CHECK-THUMB: bne
+; CHECK-THUMB: dmb ish
+
   %r = cmpxchg i64* %ptr, i64 %val1, i64 %val2 seq_cst
   ret i64 %r
 }
@@ -109,6 +187,18 @@ define i64 @test8(i64* %ptr) {
 ; CHECK: cmp
 ; CHECK: bne
 ; CHECK: dmb ish
+
+; CHECK-THUMB: test8:
+; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
+; CHECK-THUMB: cmp [[REG1]]
+; CHECK-THUMB: it eq
+; CHECK-THUMB: cmpeq [[REG2]]
+; CHECK-THUMB: bne
+; CHECK-THUMB: strexd {{[a-z0-9]+}}, {{[a-z0-9]+}}, {{[a-z0-9]+}}
+; CHECK-THUMB: cmp
+; CHECK-THUMB: bne
+; CHECK-THUMB: dmb ish
+
   %r = load atomic i64* %ptr seq_cst, align 8
   ret i64 %r
 }
@@ -123,6 +213,15 @@ define void @test9(i64* %ptr, i64 %val) {
 ; CHECK: cmp
 ; CHECK: bne
 ; CHECK: dmb ish
+
+; CHECK-THUMB: test9:
+; CHECK-THUMB: dmb ish
+; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
+; CHECK-THUMB: strexd {{[a-z0-9]+}}, {{[a-z0-9]+}}, {{[a-z0-9]+}}
+; CHECK-THUMB: cmp
+; CHECK-THUMB: bne
+; CHECK-THUMB: dmb ish
+
   store atomic i64 %val, i64* %ptr seq_cst, align 8
   ret void
 }
@@ -138,6 +237,18 @@ define i64 @test10(i64* %ptr, i64 %val) {
 ; CHECK: cmp
 ; CHECK: bne
 ; CHECK: dmb ish
+
+; CHECK-THUMB: test10:
+; CHECK-THUMB: dmb ish
+; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
+; CHECK-THUMB: subs.w {{[a-z0-9]+}}, [[REG1]], [[REG3:[a-z0-9]+]]
+; CHECK-THUMB: sbcs.w {{[a-z0-9]+}}, [[REG2]], [[REG4:[a-z0-9]+]]
+; CHECK-THUMB: blt
+; CHECK-THUMB: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
+; CHECK-THUMB: cmp
+; CHECK-THUMB: bne
+; CHECK-THUMB: dmb ish
+
   %r = atomicrmw min i64* %ptr, i64 %val seq_cst
   ret i64 %r
 }
@@ -153,6 +264,19 @@ define i64 @test11(i64* %ptr, i64 %val) {
 ; CHECK: cmp
 ; CHECK: bne
 ; CHECK: dmb ish
+
+
+; CHECK-THUMB: test11:
+; CHECK-THUMB: dmb ish
+; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
+; CHECK-THUMB: subs.w {{[a-z0-9]+}}, [[REG1]], [[REG3:[a-z0-9]+]]
+; CHECK-THUMB: sbcs.w {{[a-z0-9]+}}, [[REG2]], [[REG4:[a-z0-9]+]]
+; CHECK-THUMB: blo
+; CHECK-THUMB: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
+; CHECK-THUMB: cmp
+; CHECK-THUMB: bne
+; CHECK-THUMB: dmb ish
+
   %r = atomicrmw umin i64* %ptr, i64 %val seq_cst
   ret i64 %r
 }
@@ -168,6 +292,18 @@ define i64 @test12(i64* %ptr, i64 %val) {
 ; CHECK: cmp
 ; CHECK: bne
 ; CHECK: dmb ish
+
+; CHECK-THUMB: test12:
+; CHECK-THUMB: dmb ish
+; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
+; CHECK-THUMB: subs.w {{[a-z0-9]+}}, [[REG1]], [[REG3:[a-z0-9]+]]
+; CHECK-THUMB: sbcs.w {{[a-z0-9]+}}, [[REG2]], [[REG4:[a-z0-9]+]]
+; CHECK-THUMB: bge
+; CHECK-THUMB: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
+; CHECK-THUMB: cmp
+; CHECK-THUMB: bne
+; CHECK-THUMB: dmb ish
+
   %r = atomicrmw max i64* %ptr, i64 %val seq_cst
   ret i64 %r
 }
@@ -183,6 +319,17 @@ define i64 @test13(i64* %ptr, i64 %val) {
 ; CHECK: cmp
 ; CHECK: bne
 ; CHECK: dmb ish
+
+; CHECK-THUMB: test13:
+; CHECK-THUMB: dmb ish
+; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
+; CHECK-THUMB: subs.w {{[a-z0-9]+}}, [[REG1]], [[REG3:[a-z0-9]+]]
+; CHECK-THUMB: sbcs.w {{[a-z0-9]+}}, [[REG2]], [[REG4:[a-z0-9]+]]
+; CHECK-THUMB: bhs
+; CHECK-THUMB: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
+; CHECK-THUMB: cmp
+; CHECK-THUMB: bne
+; CHECK-THUMB: dmb ish
   %r = atomicrmw umax i64* %ptr, i64 %val seq_cst
   ret i64 %r
 }