diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 0952350d07e..9727de82036 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -13001,8 +13001,8 @@ X86TargetLowering::EmitAtomicLoadArith(MachineInstr *MI,
   MachineBasicBlock *origMainMBB = mainMBB;
 
   // Add a PHI.
-  BuildMI(mainMBB, DL, TII->get(X86::PHI), t4)
-    .addReg(t1).addMBB(thisMBB).addReg(t3).addMBB(mainMBB);
+  MachineInstr *Phi = BuildMI(mainMBB, DL, TII->get(X86::PHI), t4)
+                        .addReg(t1).addMBB(thisMBB).addReg(t3).addMBB(mainMBB);
 
   unsigned Opc = MI->getOpcode();
   switch (Opc) {
@@ -13105,6 +13105,11 @@ X86TargetLowering::EmitAtomicLoadArith(MachineInstr *MI,
               .addReg(SrcReg).addReg(t4)
               .addImm(CC);
       mainMBB = EmitLoweredSelect(MIB, mainMBB);
+      // Replace the original PHI node as mainMBB is changed after CMOV
+      // lowering.
+      BuildMI(*origMainMBB, Phi, DL, TII->get(X86::PHI), t4)
+        .addReg(t1).addMBB(thisMBB).addReg(t3).addMBB(mainMBB);
+      Phi->eraseFromParent();
     }
     break;
   }
@@ -13298,10 +13303,10 @@ X86TargetLowering::EmitAtomicLoadArith6432(MachineInstr *MI,
   MachineBasicBlock *origMainMBB = mainMBB;
 
   // Add PHIs.
-  BuildMI(mainMBB, DL, TII->get(X86::PHI), t4L)
-    .addReg(t1L).addMBB(thisMBB).addReg(t3L).addMBB(mainMBB);
-  BuildMI(mainMBB, DL, TII->get(X86::PHI), t4H)
-    .addReg(t1H).addMBB(thisMBB).addReg(t3H).addMBB(mainMBB);
+  MachineInstr *PhiL = BuildMI(mainMBB, DL, TII->get(X86::PHI), t4L)
+                        .addReg(t1L).addMBB(thisMBB).addReg(t3L).addMBB(mainMBB);
+  MachineInstr *PhiH = BuildMI(mainMBB, DL, TII->get(X86::PHI), t4H)
+                        .addReg(t1H).addMBB(thisMBB).addReg(t3H).addMBB(mainMBB);
 
   unsigned Opc = MI->getOpcode();
   switch (Opc) {
@@ -13375,10 +13380,21 @@ X86TargetLowering::EmitAtomicLoadArith6432(MachineInstr *MI,
               .addReg(SrcLoReg).addReg(t4L)
               .addImm(X86::COND_NE);
       mainMBB = EmitLoweredSelect(MIB, mainMBB);
+      // As the lowered CMOV won't clobber EFLAGS, we could reuse it for the
+      // 2nd CMOV lowering.
+      mainMBB->addLiveIn(X86::EFLAGS);
       MIB = BuildMI(mainMBB, DL, TII->get(X86::CMOV_GR32), t2H)
               .addReg(SrcHiReg).addReg(t4H)
               .addImm(X86::COND_NE);
       mainMBB = EmitLoweredSelect(MIB, mainMBB);
+      // Replace the original PHI node as mainMBB is changed after CMOV
+      // lowering.
+      BuildMI(*origMainMBB, PhiL, DL, TII->get(X86::PHI), t4L)
+        .addReg(t1L).addMBB(thisMBB).addReg(t3L).addMBB(mainMBB);
+      BuildMI(*origMainMBB, PhiH, DL, TII->get(X86::PHI), t4H)
+        .addReg(t1H).addMBB(thisMBB).addReg(t3H).addMBB(mainMBB);
+      PhiL->eraseFromParent();
+      PhiH->eraseFromParent();
     }
     break;
   }
diff --git a/test/CodeGen/X86/atomic-minmax-i6432.ll b/test/CodeGen/X86/atomic-minmax-i6432.ll
index 1a6db7781f7..62f784f6960 100644
--- a/test/CodeGen/X86/atomic-minmax-i6432.ll
+++ b/test/CodeGen/X86/atomic-minmax-i6432.ll
@@ -1,4 +1,5 @@
 ; RUN: llc -march=x86 -mattr=+cmov -mtriple=i386-pc-linux -verify-machineinstrs < %s | FileCheck %s -check-prefix=LINUX
+; RUN: llc -march=x86 -mattr=-cmov -mtriple=i386-pc-linux -verify-machineinstrs < %s | FileCheck %s -check-prefix=NOCMOV
 ; RUN: llc -march=x86 -mtriple=i386-macosx -relocation-model=pic -verify-machineinstrs < %s | FileCheck %s -check-prefix=PIC
 
 @sc64 = external global i64
@@ -16,6 +17,16 @@ define void @atomic_maxmin_i6432() {
 ; LINUX: lock
 ; LINUX-NEXT: cmpxchg8b
 ; LINUX: jne [[LABEL]]
+; NOCMOV: [[LABEL:.LBB[0-9]+_[0-9]+]]
+; NOCMOV: cmpl
+; NOCMOV: setl
+; NOCMOV: cmpl
+; NOCMOV: setl
+; NOCMOV: jne
+; NOCMOV: jne
+; NOCMOV: lock
+; NOCMOV-NEXT: cmpxchg8b
+; NOCMOV: jne [[LABEL]]
   %2 = atomicrmw min  i64* @sc64, i64 6 acquire
 ; LINUX: [[LABEL:.LBB[0-9]+_[0-9]+]]
 ; LINUX: cmpl
@@ -27,6 +38,16 @@ define void @atomic_maxmin_i6432() {
 ; LINUX: lock
 ; LINUX-NEXT: cmpxchg8b
 ; LINUX: jne [[LABEL]]
+; NOCMOV: [[LABEL:.LBB[0-9]+_[0-9]+]]
+; NOCMOV: cmpl
+; NOCMOV: setg
+; NOCMOV: cmpl
+; NOCMOV: setg
+; NOCMOV: jne
+; NOCMOV: jne
+; NOCMOV: lock
+; NOCMOV-NEXT: cmpxchg8b
+; NOCMOV: jne [[LABEL]]
   %3 = atomicrmw umax i64* @sc64, i64 7 acquire
 ; LINUX: [[LABEL:.LBB[0-9]+_[0-9]+]]
 ; LINUX: cmpl
@@ -38,6 +59,16 @@ define void @atomic_maxmin_i6432() {
 ; LINUX: lock
 ; LINUX-NEXT: cmpxchg8b
 ; LINUX: jne [[LABEL]]
+; NOCMOV: [[LABEL:.LBB[0-9]+_[0-9]+]]
+; NOCMOV: cmpl
+; NOCMOV: setb
+; NOCMOV: cmpl
+; NOCMOV: setb
+; NOCMOV: jne
+; NOCMOV: jne
+; NOCMOV: lock
+; NOCMOV-NEXT: cmpxchg8b
+; NOCMOV: jne [[LABEL]]
   %4 = atomicrmw umin i64* @sc64, i64 8 acquire
 ; LINUX: [[LABEL:.LBB[0-9]+_[0-9]+]]
 ; LINUX: cmpl
@@ -49,6 +80,16 @@ define void @atomic_maxmin_i6432() {
 ; LINUX: lock
 ; LINUX-NEXT: cmpxchg8b
 ; LINUX: jne [[LABEL]]
+; NOCMOV: [[LABEL:.LBB[0-9]+_[0-9]+]]
+; NOCMOV: cmpl
+; NOCMOV: seta
+; NOCMOV: cmpl
+; NOCMOV: seta
+; NOCMOV: jne
+; NOCMOV: jne
+; NOCMOV: lock
+; NOCMOV-NEXT: cmpxchg8b
+; NOCMOV: jne [[LABEL]]
   ret void
 }
 
diff --git a/test/CodeGen/X86/atomic32.ll b/test/CodeGen/X86/atomic32.ll
index 50c5751f018..3cb9ca1c76c 100644
--- a/test/CodeGen/X86/atomic32.ll
+++ b/test/CodeGen/X86/atomic32.ll
@@ -1,5 +1,6 @@
 ; RUN: llc < %s -O0 -march=x86-64 -mcpu=corei7 -verify-machineinstrs | FileCheck %s --check-prefix X64
 ; RUN: llc < %s -O0 -march=x86 -mcpu=corei7 -verify-machineinstrs | FileCheck %s --check-prefix X32
+; RUN: llc < %s -O0 -march=x86 -mcpu=corei7 -mattr=-cmov -verify-machineinstrs | FileCheck %s --check-prefix NOCMOV
 
 @sc32 = external global i32
 
@@ -164,9 +165,15 @@ define void @atomic_fetch_max32(i32 %x) nounwind {
 ; X32:       cmov
 ; X32:       lock
 ; X32:       cmpxchgl
+
+; NOCMOV:    cmpl
+; NOCMOV:    jl
+; NOCMOV:    lock
+; NOCMOV:    cmpxchgl
   ret void
 ; X64:       ret
 ; X32:       ret
+; NOCMOV:    ret
 }
 
 define void @atomic_fetch_min32(i32 %x) nounwind {
@@ -180,9 +187,15 @@ define void @atomic_fetch_min32(i32 %x) nounwind {
 ; X32:       cmov
 ; X32:       lock
 ; X32:       cmpxchgl
+
+; NOCMOV:    cmpl
+; NOCMOV:    jg
+; NOCMOV:    lock
+; NOCMOV:    cmpxchgl
   ret void
 ; X64:       ret
 ; X32:       ret
+; NOCMOV:    ret
 }
 
 define void @atomic_fetch_umax32(i32 %x) nounwind {
@@ -196,9 +209,15 @@ define void @atomic_fetch_umax32(i32 %x) nounwind {
 ; X32:       cmov
 ; X32:       lock
 ; X32:       cmpxchgl
+
+; NOCMOV:    cmpl
+; NOCMOV:    jb
+; NOCMOV:    lock
+; NOCMOV:    cmpxchgl
   ret void
 ; X64:       ret
 ; X32:       ret
+; NOCMOV:    ret
 }
 
 define void @atomic_fetch_umin32(i32 %x) nounwind {
@@ -207,13 +226,20 @@ define void @atomic_fetch_umin32(i32 %x) nounwind {
 ; X64:       cmov
 ; X64:       lock
 ; X64:       cmpxchgl
+
 ; X32:       cmpl
 ; X32:       cmov
 ; X32:       lock
 ; X32:       cmpxchgl
+
+; NOCMOV:    cmpl
+; NOCMOV:    ja
+; NOCMOV:    lock
+; NOCMOV:    cmpxchgl
   ret void
 ; X64:       ret
 ; X32:       ret
+; NOCMOV:    ret
 }
 
 define void @atomic_fetch_cmpxchg32() nounwind {