Cleanup pcmp(e/i)str(m/i) instruction definitions and load folding support.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@167652 91177308-0d34-0410-b5e6-96231b3b80d8

5 files changed, 239 insertions, 166 deletions

diff --git a/lib/Target/X86/X86ISelDAGToDAG.cpp b/lib/Target/X86/X86ISelDAGToDAG.cpp
index 0efeef20f2..99f557417b 100644
--- a/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ b/lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -2679,85 +2679,6 @@ SDNode *X86DAGToDAGISel::Select(SDNode *Node) {
 
     return Result;
   }
-
-  // FIXME: Custom handling because TableGen doesn't support multiple implicit
-  // defs in an instruction pattern
-  case X86ISD::PCMPESTRI: {
-    SDValue N0 = Node->getOperand(0);
-    SDValue N1 = Node->getOperand(1);
-    SDValue N2 = Node->getOperand(2);
-    SDValue N3 = Node->getOperand(3);
-    SDValue N4 = Node->getOperand(4);
-
-    // Make sure last argument is a constant
-    ConstantSDNode *Cst = dyn_cast<ConstantSDNode>(N4);
-    if (!Cst)
-      break;
-
-    uint64_t Imm = Cst->getZExtValue();
-
-    SDValue InFlag = CurDAG->getCopyToReg(CurDAG->getEntryNode(), dl,
-                                          X86::EAX, N1, SDValue()).getValue(1);
-    InFlag = CurDAG->getCopyToReg(CurDAG->getEntryNode(), dl, X86::EDX,
-                                  N3, InFlag).getValue(1);
-
-    SDValue Ops[] = { N0, N2, getI8Imm(Imm), InFlag };
-    unsigned Opc = Subtarget->hasAVX() ? X86::VPCMPESTRIrr :
-                                         X86::PCMPESTRIrr;
-    InFlag = SDValue(CurDAG->getMachineNode(Opc, dl, MVT::Glue, Ops,
-                                            array_lengthof(Ops)), 0);
-
-    if (!SDValue(Node, 0).use_empty()) {
-      SDValue Result = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), dl,
-                                              X86::ECX, NVT, InFlag);
-      InFlag = Result.getValue(2);
-      ReplaceUses(SDValue(Node, 0), Result);
-    }
-    if (!SDValue(Node, 1).use_empty()) {
-      SDValue Result = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), dl,
-                                              X86::EFLAGS, NVT, InFlag);
-      InFlag = Result.getValue(2);
-      ReplaceUses(SDValue(Node, 1), Result);
-    }
-
-    return NULL;
-  }
-
-  // FIXME: Custom handling because TableGen doesn't support multiple implicit
-  // defs in an instruction pattern
-  case X86ISD::PCMPISTRI: {
-    SDValue N0 = Node->getOperand(0);
-    SDValue N1 = Node->getOperand(1);
-    SDValue N2 = Node->getOperand(2);
-
-    // Make sure last argument is a constant
-    ConstantSDNode *Cst = dyn_cast<ConstantSDNode>(N2);
-    if (!Cst)
-      break;
-
-    uint64_t Imm = Cst->getZExtValue();
-
-    SDValue Ops[] = { N0, N1, getI8Imm(Imm) };
-    unsigned Opc = Subtarget->hasAVX() ? X86::VPCMPISTRIrr :
-                                         X86::PCMPISTRIrr;
-    SDValue InFlag = SDValue(CurDAG->getMachineNode(Opc, dl, MVT::Glue, Ops,
-                                                    array_lengthof(Ops)), 0);
-
-    if (!SDValue(Node, 0).use_empty()) {
-      SDValue Result = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), dl,
-                                              X86::ECX, NVT, InFlag);
-      InFlag = Result.getValue(2);
-      ReplaceUses(SDValue(Node, 0), Result);
-    }
-    if (!SDValue(Node, 1).use_empty()) {
-      SDValue Result = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), dl,
-                                              X86::EFLAGS, NVT, InFlag);
-      InFlag = Result.getValue(2);
-      ReplaceUses(SDValue(Node, 1), Result);
-    }
-
-    return NULL;
-  }
   }
 
   SDNode *ResNode = SelectCode(Node);
diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 1a440e50f2..065d67c5ce 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -12045,6 +12045,8 @@ const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const {
   case X86ISD::FNMSUB:             return "X86ISD::FNMSUB";
   case X86ISD::FMADDSUB:           return "X86ISD::FMADDSUB";
   case X86ISD::FMSUBADD:           return "X86ISD::FMSUBADD";
+  case X86ISD::PCMPESTRI:          return "X86ISD::PCMPESTRI";
+  case X86ISD::PCMPISTRI:          return "X86ISD::PCMPISTRI";
   }
 }
 
@@ -12839,8 +12841,8 @@ X86TargetLowering::EmitAtomicLoadArith6432(MachineInstr *MI,
 // or XMM0_V32I8 in AVX all of this code can be replaced with that
 // in the .td file.
 MachineBasicBlock *
-X86TargetLowering::EmitPCMP(MachineInstr *MI, MachineBasicBlock *BB,
-                            unsigned numArgs, bool memArg) const {
+X86TargetLowering::EmitPCMPSTRM(MachineInstr *MI, MachineBasicBlock *BB,
+                                bool Implicit, bool MemArg) const {
   assert(Subtarget->hasSSE42() &&
          "Target must have SSE4.2 or AVX features enabled");
 
@@ -12848,23 +12850,30 @@ X86TargetLowering::EmitPCMP(MachineInstr *MI, MachineBasicBlock *BB,
   const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
   unsigned Opc;
   if (!Subtarget->hasAVX()) {
-    if (memArg)
-      Opc = numArgs == 3 ? X86::PCMPISTRM128rm : X86::PCMPESTRM128rm;
+    if (MemArg)
+      Opc = Implicit ? X86::PCMPISTRM128rm : X86::PCMPESTRM128rm;
     else
-      Opc = numArgs == 3 ? X86::PCMPISTRM128rr : X86::PCMPESTRM128rr;
+      Opc = Implicit ? X86::PCMPISTRM128rr : X86::PCMPESTRM128rr;
   } else {
-    if (memArg)
-      Opc = numArgs == 3 ? X86::VPCMPISTRM128rm : X86::VPCMPESTRM128rm;
+    if (MemArg)
+      Opc = Implicit ? X86::VPCMPISTRM128rm : X86::VPCMPESTRM128rm;
     else
-      Opc = numArgs == 3 ? X86::VPCMPISTRM128rr : X86::VPCMPESTRM128rr;
+      Opc = Implicit ? X86::VPCMPISTRM128rr : X86::VPCMPESTRM128rr;
   }
 
+  unsigned NumArgs = Implicit ? 3 : 5;
+  if (MemArg)
+    NumArgs += X86::AddrNumOperands;
+
   MachineInstrBuilder MIB = BuildMI(*BB, MI, dl, TII->get(Opc));
-  for (unsigned i = 0; i < numArgs; ++i) {
+  for (unsigned i = 0; i < NumArgs; ++i) {
     MachineOperand &Op = MI->getOperand(i+1);
     if (!(Op.isReg() && Op.isImplicit()))
       MIB.addOperand(Op);
   }
+  if (MemArg)
+    MIB->setMemRefs(MI->memoperands_begin(), MI->memoperands_end());
+
   BuildMI(*BB, MI, dl,
     TII->get(TargetOpcode::COPY), MI->getOperand(0).getReg())
     .addReg(X86::XMM0);
@@ -12873,6 +12882,50 @@ X86TargetLowering::EmitPCMP(MachineInstr *MI, MachineBasicBlock *BB,
   return BB;
 }
 
+// FIXME: Custom handling because TableGen doesn't support multiple implicit
+// defs in an instruction pattern
+MachineBasicBlock *
+X86TargetLowering::EmitPCMPSTRI(MachineInstr *MI, MachineBasicBlock *BB,
+                                bool Implicit, bool MemArg) const {
+  assert(Subtarget->hasSSE42() &&
+         "Target must have SSE4.2 or AVX features enabled");
+
+  DebugLoc dl = MI->getDebugLoc();
+  const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+  unsigned Opc;
+  if (!Subtarget->hasAVX()) {
+    if (MemArg)
+      Opc = Implicit ? X86::PCMPISTRIrm : X86::PCMPESTRIrm;
+    else
+      Opc = Implicit ? X86::PCMPISTRIrr : X86::PCMPESTRIrr;
+  } else {
+    if (MemArg)
+      Opc = Implicit ? X86::VPCMPISTRIrm : X86::VPCMPESTRIrm;
+    else
+      Opc = Implicit ? X86::VPCMPISTRIrr : X86::VPCMPESTRIrr;
+  }
+
+  unsigned NumArgs = Implicit ? 3 : 5;
+  if (MemArg)
+    NumArgs += X86::AddrNumOperands;
+
+  MachineInstrBuilder MIB = BuildMI(*BB, MI, dl, TII->get(Opc));
+  for (unsigned i = 0; i < NumArgs; ++i) {
+    MachineOperand &Op = MI->getOperand(i+1);
+    if (!(Op.isReg() && Op.isImplicit()))
+      MIB.addOperand(Op);
+  }
+  if (MemArg)
+    MIB->setMemRefs(MI->memoperands_begin(), MI->memoperands_end());
+
+  BuildMI(*BB, MI, dl,
+    TII->get(TargetOpcode::COPY), MI->getOperand(0).getReg())
+    .addReg(X86::ECX);
+
+  MI->eraseFromParent();
+  return BB;
+}
+
 MachineBasicBlock *
 X86TargetLowering::EmitMonitor(MachineInstr *MI, MachineBasicBlock *BB) const {
   DebugLoc dl = MI->getDebugLoc();
@@ -13891,24 +13944,51 @@ X86TargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
   case X86::VPCMPESTRM128REG:
   case X86::PCMPESTRM128MEM:
   case X86::VPCMPESTRM128MEM: {
-    unsigned NumArgs;
-    bool MemArg;
+    bool Implicit, MemArg;
     switch (MI->getOpcode()) {
     default: llvm_unreachable("illegal opcode!");
     case X86::PCMPISTRM128REG:
     case X86::VPCMPISTRM128REG:
-      NumArgs = 3; MemArg = false; break;
+      Implicit = true; MemArg = false; break;
     case X86::PCMPISTRM128MEM:
     case X86::VPCMPISTRM128MEM:
-      NumArgs = 3; MemArg = true; break;
+     Implicit = true; MemArg = true; break;
     case X86::PCMPESTRM128REG:
     case X86::VPCMPESTRM128REG:
-      NumArgs = 5; MemArg = false; break;
+      Implicit = false; MemArg = false; break;
     case X86::PCMPESTRM128MEM:
     case X86::VPCMPESTRM128MEM:
-      NumArgs = 5; MemArg = true; break;
+      Implicit = false; MemArg = true; break;
+    }
+    return EmitPCMPSTRM(MI, BB, Implicit, MemArg);
+  }
+
+  // String/text processing lowering.
+  case X86::PCMPISTRIREG:
+  case X86::VPCMPISTRIREG:
+  case X86::PCMPISTRIMEM:
+  case X86::VPCMPISTRIMEM:
+  case X86::PCMPESTRIREG:
+  case X86::VPCMPESTRIREG:
+  case X86::PCMPESTRIMEM:
+  case X86::VPCMPESTRIMEM: {
+    bool Implicit, MemArg;
+    switch (MI->getOpcode()) {
+    default: llvm_unreachable("illegal opcode!");
+    case X86::PCMPISTRIREG:
+    case X86::VPCMPISTRIREG:
+      Implicit = true; MemArg = false; break;
+    case X86::PCMPISTRIMEM:
+    case X86::VPCMPISTRIMEM:
+      Implicit = true; MemArg = true; break;
+    case X86::PCMPESTRIREG:
+    case X86::VPCMPESTRIREG:
+      Implicit = false; MemArg = false; break;
+    case X86::PCMPESTRIMEM:
+    case X86::VPCMPESTRIMEM:
+      Implicit = false; MemArg = true; break;
     }
-    return EmitPCMP(MI, BB, NumArgs, MemArg);
+    return EmitPCMPSTRI(MI, BB, Implicit, MemArg);
   }
 
     // Thread synchronization.
diff --git a/lib/Target/X86/X86ISelLowering.h b/lib/Target/X86/X86ISelLowering.h
index 7764330cc8..a2a9db08ca 100644
--- a/lib/Target/X86/X86ISelLowering.h
+++ b/lib/Target/X86/X86ISelLowering.h
@@ -871,13 +871,15 @@ namespace llvm {
                    const SmallVectorImpl<ISD::OutputArg> &Outs,
                    LLVMContext &Context) const;
 
-    /// Utility function to emit string processing sse4.2 instructions
+    /// Utility functions to emit string processing sse4.2 instructions
     /// that return in xmm0.
     /// This takes the instruction to expand, the associated machine basic
     /// block, the number of args, and whether or not the second arg is
     /// in memory or not.
-    MachineBasicBlock *EmitPCMP(MachineInstr *BInstr, MachineBasicBlock *BB,
-                                unsigned argNum, bool inMem) const;
+    MachineBasicBlock *EmitPCMPSTRM(MachineInstr *MI, MachineBasicBlock *BB,
+                                    bool Implicit, bool MemArg) const;
+    MachineBasicBlock *EmitPCMPSTRI(MachineInstr *MI, MachineBasicBlock *BB,
+                                    bool Implicit, bool MemArg) const;
 
     /// Utility functions to emit monitor and mwait instructions. These
     /// need to make sure that the arguments to the intrinsic are in the
diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 28dfbe7a1f..6f48d7ed7f 100644
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -7002,8 +7002,8 @@ multiclass pseudo_pcmpistrm<string asm> {
                                                   imm:$src3))]>;
   def MEM : PseudoI<(outs VR128:$dst),
                     (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
-    [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
-                       VR128:$src1, (load addr:$src2), imm:$src3))]>;
+    [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1,
+                       (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
 }
 
 let Defs = [EFLAGS], usesCustomInserter = 1 in {
@@ -7011,24 +7011,22 @@ let Defs = [EFLAGS], usesCustomInserter = 1 in {
   defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[UseSSE42]>;
 }
 
-let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
-  def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
-      (ins VR128:$src1, VR128:$src2, i8imm:$src3),
-      "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
+multiclass pcmpistrm_SS42AI<string asm> {
+  def rr : SS42AI<0x62, MRMSrcReg, (outs),
+    (ins VR128:$src1, VR128:$src2, i8imm:$src3),
+    !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
+    []>, OpSize;
   let mayLoad = 1 in
-  def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
-      (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
-      "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
+  def rm :SS42AI<0x62, MRMSrcMem, (outs),
+    (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
+    !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
+    []>, OpSize;
 }
 
 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
-  def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
-      (ins VR128:$src1, VR128:$src2, i8imm:$src3),
-      "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
-  let mayLoad = 1 in
-  def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
-      (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
-      "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
+  let Predicates = [HasAVX] in
+  defm VPCMPISTRM128 : pcmpistrm_SS42AI<"vpcmpistrm">, VEX;
+  defm PCMPISTRM128  : pcmpistrm_SS42AI<"pcmpistrm"> ;
 }
 
 // Packed Compare Explicit Length Strings, Return Mask
@@ -7039,8 +7037,8 @@ multiclass pseudo_pcmpestrm<string asm> {
                        VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
   def MEM : PseudoI<(outs VR128:$dst),
                     (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
-    [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
-                       VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
+    [(set VR128:$dst, (int_x86_sse42_pcmpestrm128 VR128:$src1, EAX,
+                       (bc_v16i8 (memopv2i64 addr:$src3)), EDX, imm:$src5))]>;
 }
 
 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
@@ -7048,64 +7046,94 @@ let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
   defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[UseSSE42]>;
 }
 
-let Predicates = [HasAVX],
-    Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
-  def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
-      (ins VR128:$src1, VR128:$src3, i8imm:$src5),
-      "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
+multiclass SS42AI_pcmpestrm<string asm> {
+  def rr : SS42AI<0x60, MRMSrcReg, (outs),
+    (ins VR128:$src1, VR128:$src3, i8imm:$src5),
+    !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
+    []>, OpSize;
   let mayLoad = 1 in
-  def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
-      (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
-      "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
+  def rm : SS42AI<0x60, MRMSrcMem, (outs),
+    (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
+    !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
+    []>, OpSize;
 }
 
 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
-  def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
-      (ins VR128:$src1, VR128:$src3, i8imm:$src5),
-      "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
-  let mayLoad = 1 in
-  def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
-      (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
-      "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
+  let Predicates = [HasAVX] in
+  defm VPCMPESTRM128 : SS42AI_pcmpestrm<"vpcmpestrm">, VEX;
+  defm PCMPESTRM128 :  SS42AI_pcmpestrm<"pcmpestrm">;
 }
 
 // Packed Compare Implicit Length Strings, Return Index
-let Defs = [ECX, EFLAGS], neverHasSideEffects = 1 in {
-  multiclass SS42AI_pcmpistri<string asm> {
-    def rr : SS42AI<0x63, MRMSrcReg, (outs),
-      (ins VR128:$src1, VR128:$src2, i8imm:$src3),
-      !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
-      []>, OpSize;
-    let mayLoad = 1 in
-    def rm : SS42AI<0x63, MRMSrcMem, (outs),
-      (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
-      !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
-      []>, OpSize;
-  }
+multiclass pseudo_pcmpistri<string asm> {
+  def REG : PseudoI<(outs GR32:$dst),
+                    (ins VR128:$src1, VR128:$src2, i8imm:$src3),
+    [(set GR32:$dst, EFLAGS,
+      (X86pcmpistri VR128:$src1, VR128:$src2, imm:$src3))]>;
+  def MEM : PseudoI<(outs GR32:$dst),
+                    (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
+    [(set GR32:$dst, EFLAGS, (X86pcmpistri VR128:$src1,
+                              (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
 }
 
-let Predicates = [HasAVX] in
-defm VPCMPISTRI  : SS42AI_pcmpistri<"vpcmpistri">, VEX;
-defm PCMPISTRI   : SS42AI_pcmpistri<"pcmpistri">;
+let Defs = [EFLAGS], usesCustomInserter = 1 in {
+  defm VPCMPISTRI : pseudo_pcmpistri<"#VPCMPISTRI">, Requires<[HasAVX]>;
+  defm PCMPISTRI  : pseudo_pcmpistri<"#PCMPISTRI">, Requires<[UseSSE42]>;
+}
+
+multiclass SS42AI_pcmpistri<string asm> {
+  def rr : SS42AI<0x63, MRMSrcReg, (outs),
+    (ins VR128:$src1, VR128:$src2, i8imm:$src3),
+    !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
+    []>, OpSize;
+  let mayLoad = 1 in
+  def rm : SS42AI<0x63, MRMSrcMem, (outs),
+    (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
+    !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
+    []>, OpSize;
+}
+
+let Defs = [ECX, EFLAGS], neverHasSideEffects = 1 in {
+  let Predicates = [HasAVX] in
+  defm VPCMPISTRI : SS42AI_pcmpistri<"vpcmpistri">, VEX;
+  defm PCMPISTRI  : SS42AI_pcmpistri<"pcmpistri">;
+}
 
 // Packed Compare Explicit Length Strings, Return Index
-let Defs = [ECX, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
-  multiclass SS42AI_pcmpestri<string asm> {
-    def rr : SS42AI<0x61, MRMSrcReg, (outs),
-      (ins VR128:$src1, VR128:$src3, i8imm:$src5),
-      !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
-      []>, OpSize;
-    let mayLoad = 1 in
-    def rm : SS42AI<0x61, MRMSrcMem, (outs),
-      (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
-      !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
-      []>, OpSize;
-  }
+multiclass pseudo_pcmpestri<string asm> {
+  def REG : PseudoI<(outs GR32:$dst),
+                    (ins VR128:$src1, VR128:$src3, i8imm:$src5),
+    [(set GR32:$dst, EFLAGS,
+      (X86pcmpestri VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
+  def MEM : PseudoI<(outs GR32:$dst),
+                    (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
+    [(set GR32:$dst, EFLAGS,
+      (X86pcmpestri VR128:$src1, EAX, (bc_v16i8 (memopv2i64 addr:$src3)), EDX,
+       imm:$src5))]>;
 }
 
-let Predicates = [HasAVX] in
-defm VPCMPESTRI  : SS42AI_pcmpestri<"vpcmpestri">, VEX;
-defm PCMPESTRI   : SS42AI_pcmpestri<"pcmpestri">;
+let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
+  defm VPCMPESTRI : pseudo_pcmpestri<"#VPCMPESTRI">, Requires<[HasAVX]>;
+  defm PCMPESTRI  : pseudo_pcmpestri<"#PCMPESTRI">, Requires<[UseSSE42]>;
+}
+
+multiclass SS42AI_pcmpestri<string asm> {
+  def rr : SS42AI<0x61, MRMSrcReg, (outs),
+    (ins VR128:$src1, VR128:$src3, i8imm:$src5),
+    !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
+    []>, OpSize;
+  let mayLoad = 1 in
+  def rm : SS42AI<0x61, MRMSrcMem, (outs),
+    (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
+    !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
+    []>, OpSize;
+}
+
+let Defs = [ECX, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
+  let Predicates = [HasAVX] in
+  defm VPCMPESTRI : SS42AI_pcmpestri<"vpcmpestri">, VEX;
+  defm PCMPESTRI  : SS42AI_pcmpestri<"pcmpestri">;
+}
 
 //===----------------------------------------------------------------------===//
 // SSE4.2 - CRC Instructions
diff --git a/test/CodeGen/X86/avx-intrinsics-x86.ll b/test/CodeGen/X86/avx-intrinsics-x86.ll
index c44beb4bc2..88ecd5a5d3 100644
--- a/test/CodeGen/X86/avx-intrinsics-x86.ll
+++ b/test/CodeGen/X86/avx-intrinsics-x86.ll
@@ -1140,9 +1140,9 @@ declare <4 x float> @llvm.x86.sse41.round.ss(<4 x float>, <4 x float>, i32) noun
 
 
 define i32 @test_x86_sse42_pcmpestri128(<16 x i8> %a0, <16 x i8> %a2) {
-  ; CHECK: movl
-  ; CHECK: movl
-  ; CHECK: vpcmpestri
+  ; CHECK: movl $7
+  ; CHECK: movl $7
+  ; CHECK: vpcmpestri $7
   ; CHECK: movl
   %res = call i32 @llvm.x86.sse42.pcmpestri128(<16 x i8> %a0, i32 7, <16 x i8> %a2, i32 7, i8 7) ; <i32> [#uses=1]
   ret i32 %res
@@ -1150,6 +1150,18 @@ define i32 @test_x86_sse42_pcmpestri128(<16 x i8> %a0, <16 x i8> %a2) {
 declare i32 @llvm.x86.sse42.pcmpestri128(<16 x i8>, i32, <16 x i8>, i32, i8) nounwind readnone
 
 
+define i32 @test_x86_sse42_pcmpestri128_load(<16 x i8>* %a0, <16 x i8>* %a2) {
+  ; CHECK: movl $7
+  ; CHECK: movl $7
+  ; CHECK: vpcmpestri $7, (
+  ; CHECK: movl
+  %1 = load <16 x i8>* %a0
+  %2 = load <16 x i8>* %a2
+  %res = call i32 @llvm.x86.sse42.pcmpestri128(<16 x i8> %1, i32 7, <16 x i8> %2, i32 7, i8 7) ; <i32> [#uses=1]
+  ret i32 %res
+}
+
+
 define i32 @test_x86_sse42_pcmpestria128(<16 x i8> %a0, <16 x i8> %a2) {
   ; CHECK: movl
   ; CHECK: movl
@@ -1216,8 +1228,19 @@ define <16 x i8> @test_x86_sse42_pcmpestrm128(<16 x i8> %a0, <16 x i8> %a2) {
 declare <16 x i8> @llvm.x86.sse42.pcmpestrm128(<16 x i8>, i32, <16 x i8>, i32, i8) nounwind readnone
 
 
+define <16 x i8> @test_x86_sse42_pcmpestrm128_load(<16 x i8> %a0, <16 x i8>* %a2) {
+  ; CHECK: movl $7
+  ; CHECK: movl $7
+  ; CHECK: vpcmpestrm $7,
+  ; CHECK-NOT: vmov
+  %1 = load <16 x i8>* %a2
+  %res = call <16 x i8> @llvm.x86.sse42.pcmpestrm128(<16 x i8> %a0, i32 7, <16 x i8> %1, i32 7, i8 7) ; <<16 x i8>> [#uses=1]
+  ret <16 x i8> %res
+}
+
+
 define i32 @test_x86_sse42_pcmpistri128(<16 x i8> %a0, <16 x i8> %a1) {
-  ; CHECK: vpcmpistri
+  ; CHECK: vpcmpistri $7
   ; CHECK: movl
   %res = call i32 @llvm.x86.sse42.pcmpistri128(<16 x i8> %a0, <16 x i8> %a1, i8 7) ; <i32> [#uses=1]
   ret i32 %res
@@ -1225,6 +1248,16 @@ define i32 @test_x86_sse42_pcmpistri128(<16 x i8> %a0, <16 x i8> %a1) {
 declare i32 @llvm.x86.sse42.pcmpistri128(<16 x i8>, <16 x i8>, i8) nounwind readnone
 
 
+define i32 @test_x86_sse42_pcmpistri128_load(<16 x i8>* %a0, <16 x i8>* %a1) {
+  ; CHECK: vpcmpistri $7, (
+  ; CHECK: movl
+  %1 = load <16 x i8>* %a0
+  %2 = load <16 x i8>* %a1
+  %res = call i32 @llvm.x86.sse42.pcmpistri128(<16 x i8> %1, <16 x i8> %2, i8 7) ; <i32> [#uses=1]
+  ret i32 %res
+}
+
+
 define i32 @test_x86_sse42_pcmpistria128(<16 x i8> %a0, <16 x i8> %a1) {
   ; CHECK: vpcmpistri
   ; CHECK: seta
@@ -1271,7 +1304,7 @@ declare i32 @llvm.x86.sse42.pcmpistriz128(<16 x i8>, <16 x i8>, i8) nounwind rea
 
 
 define <16 x i8> @test_x86_sse42_pcmpistrm128(<16 x i8> %a0, <16 x i8> %a1) {
-  ; CHECK: vpcmpistrm
+  ; CHECK: vpcmpistrm $7
   ; CHECK-NOT: vmov
   %res = call <16 x i8> @llvm.x86.sse42.pcmpistrm128(<16 x i8> %a0, <16 x i8> %a1, i8 7) ; <<16 x i8>> [#uses=1]
   ret <16 x i8> %res
@@ -1279,6 +1312,15 @@ define <16 x i8> @test_x86_sse42_pcmpistrm128(<16 x i8> %a0, <16 x i8> %a1) {
 declare <16 x i8> @llvm.x86.sse42.pcmpistrm128(<16 x i8>, <16 x i8>, i8) nounwind readnone
 
 
+define <16 x i8> @test_x86_sse42_pcmpistrm128_load(<16 x i8> %a0, <16 x i8>* %a1) {
+  ; CHECK: vpcmpistrm $7, (
+  ; CHECK-NOT: vmov
+  %1 = load <16 x i8>* %a1
+  %res = call <16 x i8> @llvm.x86.sse42.pcmpistrm128(<16 x i8> %a0, <16 x i8> %1, i8 7) ; <<16 x i8>> [#uses=1]
+  ret <16 x i8> %res
+}
+
+
 define <4 x float> @test_x86_sse_add_ss(<4 x float> %a0, <4 x float> %a1) {
   ; CHECK: vaddss
   %res = call <4 x float> @llvm.x86.sse.add.ss(<4 x float> %a0, <4 x float> %a1) ; <<4 x float>> [#uses=1]