[X86] Merge the different SETcc instructions for each condition code into single instructions that store the condition code as an operand.

Summary:
This avoids needing an isel pattern for each condition code. And it removes translation switches for converting between SETcc instructions and condition codes.

Now the printer, encoder and disassembler take care of converting the immediate. We use InstAliases to handle the assembly matching. But we print using the asm string in the instruction definition. The instruction itself is marked IsCodeGenOnly=1 to hide it from the assembly parser.

Reviewers: andreadb, courbet, RKSimon, spatel, lebedev.ri

Reviewed By: andreadb

Subscribers: hiraditya, lebedev.ri, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D60138

llvm-svn: 357801

20 files changed, 290 insertions, 256 deletions

diff --git a/llvm/lib/Target/X86/MCTargetDesc/X86BaseInfo.h b/llvm/lib/Target/X86/MCTargetDesc/X86BaseInfo.h
index f13254ffa54..29b6efb922e 100644
--- a/llvm/lib/Target/X86/MCTargetDesc/X86BaseInfo.h
+++ b/llvm/lib/Target/X86/MCTargetDesc/X86BaseInfo.h
@@ -352,9 +352,15 @@ namespace X86II {
     MRMSrcMemCC    = 36,
 
     /// MRMXm - This form is used for instructions that use the Mod/RM byte
+    /// to specify a memory source, but doesn't use the middle field. And has
+    /// a condition code.
+    ///
+    MRMXmCC = 38,
+
+    /// MRMXm - This form is used for instructions that use the Mod/RM byte
     /// to specify a memory source, but doesn't use the middle field.
     ///
-    MRMXm = 39, // Instruction that uses Mod/RM but not the middle field.
+    MRMXm = 39,
 
     // Next, instructions that operate on a memory r/m operand...
     MRM0m = 40,  MRM1m = 41,  MRM2m = 42,  MRM3m = 43, // Format /0 /1 /2 /3
@@ -385,10 +391,16 @@ namespace X86II {
     ///
     MRMSrcRegCC    = 52,
 
+    /// MRMXCCr - This form is used for instructions that use the Mod/RM byte
+    /// to specify a register source, but doesn't use the middle field. And has
+    /// a condition code.
+    ///
+    MRMXrCC = 54,
+
     /// MRMXr - This form is used for instructions that use the Mod/RM byte
     /// to specify a register source, but doesn't use the middle field.
     ///
-    MRMXr = 55, // Instruction that uses Mod/RM but not the middle field.
+    MRMXr = 55,
 
     // Instructions that operate on a register r/m operand...
     MRM0r = 56,  MRM1r = 57,  MRM2r = 58,  MRM3r = 59, // Format /0 /1 /2 /3
@@ -779,12 +791,14 @@ namespace X86II {
     case X86II::MRMSrcReg4VOp3:
     case X86II::MRMSrcRegOp4:
     case X86II::MRMSrcRegCC:
+    case X86II::MRMXrCC:
     case X86II::MRMXr:
     case X86II::MRM0r: case X86II::MRM1r:
     case X86II::MRM2r: case X86II::MRM3r:
     case X86II::MRM4r: case X86II::MRM5r:
     case X86II::MRM6r: case X86II::MRM7r:
       return -1;
+    case X86II::MRMXmCC:
     case X86II::MRMXm:
     case X86II::MRM0m: case X86II::MRM1m:
     case X86II::MRM2m: case X86II::MRM3m:
diff --git a/llvm/lib/Target/X86/MCTargetDesc/X86MCCodeEmitter.cpp b/llvm/lib/Target/X86/MCTargetDesc/X86MCCodeEmitter.cpp
index 2f3cbcfc4c8..05e19a3db1c 100644
--- a/llvm/lib/Target/X86/MCTargetDesc/X86MCCodeEmitter.cpp
+++ b/llvm/lib/Target/X86/MCTargetDesc/X86MCCodeEmitter.cpp
@@ -1081,7 +1081,7 @@ uint8_t X86MCCodeEmitter::DetermineREXPrefix(const MCInst &MI, uint64_t TSFlags,
     CurOp += X86::AddrNumOperands;
     REX |= isREXExtendedReg(MI, CurOp++) << 2; // REX.R
     break;
-  case X86II::MRMXm:
+  case X86II::MRMXmCC: case X86II::MRMXm:
   case X86II::MRM0m: case X86II::MRM1m:
   case X86II::MRM2m: case X86II::MRM3m:
   case X86II::MRM4m: case X86II::MRM5m:
@@ -1089,7 +1089,7 @@ uint8_t X86MCCodeEmitter::DetermineREXPrefix(const MCInst &MI, uint64_t TSFlags,
     REX |= isREXExtendedReg(MI, MemOperand+X86::AddrBaseReg) << 0; // REX.B
     REX |= isREXExtendedReg(MI, MemOperand+X86::AddrIndexReg) << 1; // REX.X
     break;
-  case X86II::MRMXr:
+  case X86II::MRMXrCC: case X86II::MRMXr:
   case X86II::MRM0r: case X86II::MRM1r:
   case X86II::MRM2r: case X86II::MRM3r:
   case X86II::MRM4r: case X86II::MRM5r:
@@ -1506,6 +1506,15 @@ encodeInstruction(const MCInst &MI, raw_ostream &OS,
     break;
   }
 
+  case X86II::MRMXrCC: {
+    unsigned RegOp = CurOp++;
+
+    unsigned CC = MI.getOperand(CurOp++).getImm();
+    EmitByte(BaseOpcode + CC, CurByte, OS);
+    EmitRegModRMByte(MI.getOperand(RegOp), 0, CurByte, OS);
+    break;
+  }
+
   case X86II::MRMXr:
   case X86II::MRM0r: case X86II::MRM1r:
   case X86II::MRM2r: case X86II::MRM3r:
@@ -1521,6 +1530,17 @@ encodeInstruction(const MCInst &MI, raw_ostream &OS,
                      CurByte, OS);
     break;
 
+  case X86II::MRMXmCC: {
+    unsigned FirstMemOp = CurOp;
+    CurOp = FirstMemOp + X86::AddrNumOperands;
+
+    unsigned CC = MI.getOperand(CurOp++).getImm();
+    EmitByte(BaseOpcode + CC, CurByte, OS);
+
+    emitMemModRMByte(MI, FirstMemOp, 0, TSFlags, Rex, CurByte, OS, Fixups, STI);
+    break;
+  }
+
   case X86II::MRMXm:
   case X86II::MRM0m: case X86II::MRM1m:
   case X86II::MRM2m: case X86II::MRM3m:
diff --git a/llvm/lib/Target/X86/X86FastISel.cpp b/llvm/lib/Target/X86/X86FastISel.cpp
index bf1a6c6e9e0..9edb2bfa4ef 100644
--- a/llvm/lib/Target/X86/X86FastISel.cpp
+++ b/llvm/lib/Target/X86/X86FastISel.cpp
@@ -1480,8 +1480,8 @@ bool X86FastISel::X86SelectCmp(const Instruction *I) {
 
   // FCMP_OEQ and FCMP_UNE cannot be checked with a single instruction.
   static const uint16_t SETFOpcTable[2][3] = {
-    { X86::SETEr,  X86::SETNPr, X86::AND8rr },
-    { X86::SETNEr, X86::SETPr,  X86::OR8rr  }
+    { X86::COND_E,  X86::COND_NP, X86::AND8rr },
+    { X86::COND_NE, X86::COND_P,  X86::OR8rr  }
   };
   const uint16_t *SETFOpc = nullptr;
   switch (Predicate) {
@@ -1497,10 +1497,10 @@ bool X86FastISel::X86SelectCmp(const Instruction *I) {
 
     unsigned FlagReg1 = createResultReg(&X86::GR8RegClass);
     unsigned FlagReg2 = createResultReg(&X86::GR8RegClass);
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(SETFOpc[0]),
-            FlagReg1);
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(SETFOpc[1]),
-            FlagReg2);
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::SETCCr),
+            FlagReg1).addImm(SETFOpc[0]);
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::SETCCr),
+            FlagReg2).addImm(SETFOpc[1]);
     BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(SETFOpc[2]),
             ResultReg).addReg(FlagReg1).addReg(FlagReg2);
     updateValueMap(I, ResultReg);
@@ -1511,7 +1511,6 @@ bool X86FastISel::X86SelectCmp(const Instruction *I) {
   bool SwapArgs;
   std::tie(CC, SwapArgs) = X86::getX86ConditionCode(Predicate);
   assert(CC <= X86::LAST_VALID_COND && "Unexpected condition code.");
-  unsigned Opc = X86::getSETFromCond(CC);
 
   if (SwapArgs)
     std::swap(LHS, RHS);
@@ -1520,7 +1519,8 @@ bool X86FastISel::X86SelectCmp(const Instruction *I) {
   if (!X86FastEmitCompare(LHS, RHS, VT, I->getDebugLoc()))
     return false;
 
-  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), ResultReg);
+  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::SETCCr),
+          ResultReg).addImm(CC);
   updateValueMap(I, ResultReg);
   return true;
 }
@@ -2047,8 +2047,8 @@ bool X86FastISel::X86FastEmitCMoveSelect(MVT RetVT, const Instruction *I) {
 
     // FCMP_OEQ and FCMP_UNE cannot be checked with a single instruction.
     static const uint16_t SETFOpcTable[2][3] = {
-      { X86::SETNPr, X86::SETEr , X86::TEST8rr },
-      { X86::SETPr,  X86::SETNEr, X86::OR8rr   }
+      { X86::COND_NP, X86::COND_E,  X86::TEST8rr },
+      { X86::COND_P,  X86::COND_NE, X86::OR8rr   }
     };
     const uint16_t *SETFOpc = nullptr;
     switch (Predicate) {
@@ -2080,10 +2080,10 @@ bool X86FastISel::X86FastEmitCMoveSelect(MVT RetVT, const Instruction *I) {
     if (SETFOpc) {
       unsigned FlagReg1 = createResultReg(&X86::GR8RegClass);
       unsigned FlagReg2 = createResultReg(&X86::GR8RegClass);
-      BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(SETFOpc[0]),
-              FlagReg1);
-      BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(SETFOpc[1]),
-              FlagReg2);
+      BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::SETCCr),
+              FlagReg1).addImm(SETFOpc[0]);
+      BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::SETCCr),
+              FlagReg2).addImm(SETFOpc[1]);
       auto const &II = TII.get(SETFOpc[2]);
       if (II.getNumDefs()) {
         unsigned TmpReg = createResultReg(&X86::GR8RegClass);
@@ -2897,21 +2897,21 @@ bool X86FastISel::fastLowerIntrinsicCall(const IntrinsicInst *II) {
         isCommutativeIntrinsic(II))
       std::swap(LHS, RHS);
 
-    unsigned BaseOpc, CondOpc;
+    unsigned BaseOpc, CondCode;
     switch (II->getIntrinsicID()) {
     default: llvm_unreachable("Unexpected intrinsic!");
     case Intrinsic::sadd_with_overflow:
-      BaseOpc = ISD::ADD; CondOpc = X86::SETOr; break;
+      BaseOpc = ISD::ADD; CondCode = X86::COND_O; break;
     case Intrinsic::uadd_with_overflow:
-      BaseOpc = ISD::ADD; CondOpc = X86::SETBr; break;
+      BaseOpc = ISD::ADD; CondCode = X86::COND_B; break;
     case Intrinsic::ssub_with_overflow:
-      BaseOpc = ISD::SUB; CondOpc = X86::SETOr; break;
+      BaseOpc = ISD::SUB; CondCode = X86::COND_O; break;
     case Intrinsic::usub_with_overflow:
-      BaseOpc = ISD::SUB; CondOpc = X86::SETBr; break;
+      BaseOpc = ISD::SUB; CondCode = X86::COND_B; break;
     case Intrinsic::smul_with_overflow:
-      BaseOpc = X86ISD::SMUL; CondOpc = X86::SETOr; break;
+      BaseOpc = X86ISD::SMUL; CondCode = X86::COND_O; break;
     case Intrinsic::umul_with_overflow:
-      BaseOpc = X86ISD::UMUL; CondOpc = X86::SETOr; break;
+      BaseOpc = X86ISD::UMUL; CondCode = X86::COND_O; break;
     }
 
     unsigned LHSReg = getRegForValue(LHS);
@@ -2928,7 +2928,7 @@ bool X86FastISel::fastLowerIntrinsicCall(const IntrinsicInst *II) {
       };
 
       if (CI->isOne() && (BaseOpc == ISD::ADD || BaseOpc == ISD::SUB) &&
-          CondOpc == X86::SETOr) {
+          CondCode == X86::COND_O) {
         // We can use INC/DEC.
         ResultReg = createResultReg(TLI.getRegClassFor(VT));
         bool IsDec = BaseOpc == ISD::SUB;
@@ -2987,8 +2987,8 @@ bool X86FastISel::fastLowerIntrinsicCall(const IntrinsicInst *II) {
     // Assign to a GPR since the overflow return value is lowered to a SETcc.
     unsigned ResultReg2 = createResultReg(&X86::GR8RegClass);
     assert((ResultReg+1) == ResultReg2 && "Nonconsecutive result registers.");
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(CondOpc),
-            ResultReg2);
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(X86::SETCCr),
+            ResultReg2).addImm(CondCode);
 
     updateValueMap(II, ResultReg, 2);
     return true;
diff --git a/llvm/lib/Target/X86/X86FixupSetCC.cpp b/llvm/lib/Target/X86/X86FixupSetCC.cpp
index 5bfad71ec05..e2d4d1ede6f 100644
--- a/llvm/lib/Target/X86/X86FixupSetCC.cpp
+++ b/llvm/lib/Target/X86/X86FixupSetCC.cpp
@@ -67,30 +67,6 @@ char X86FixupSetCCPass::ID = 0;
 
 FunctionPass *llvm::createX86FixupSetCC() { return new X86FixupSetCCPass(); }
 
-bool X86FixupSetCCPass::isSetCCr(unsigned Opcode) {
-  switch (Opcode) {
-  default:
-    return false;
-  case X86::SETOr:
-  case X86::SETNOr:
-  case X86::SETBr:
-  case X86::SETAEr:
-  case X86::SETEr:
-  case X86::SETNEr:
-  case X86::SETBEr:
-  case X86::SETAr:
-  case X86::SETSr:
-  case X86::SETNSr:
-  case X86::SETPr:
-  case X86::SETNPr:
-  case X86::SETLr:
-  case X86::SETGEr:
-  case X86::SETLEr:
-  case X86::SETGr:
-    return true;
-  }
-}
-
 // We expect the instruction *immediately* before the setcc to imp-def
 // EFLAGS (because of scheduling glue). To make this less brittle w.r.t
 // scheduling, look backwards until we hit the beginning of the
@@ -128,7 +104,7 @@ bool X86FixupSetCCPass::runOnMachineFunction(MachineFunction &MF) {
       // Find a setcc that is used by a zext.
       // This doesn't have to be the only use, the transformation is safe
       // regardless.
-      if (!isSetCCr(MI.getOpcode()))
+      if (MI.getOpcode() != X86::SETCCr)
         continue;
 
       MachineInstr *ZExt = nullptr;
diff --git a/llvm/lib/Target/X86/X86FlagsCopyLowering.cpp b/llvm/lib/Target/X86/X86FlagsCopyLowering.cpp
index 52ae70a60dc..6cf01c988d7 100644
--- a/llvm/lib/Target/X86/X86FlagsCopyLowering.cpp
+++ b/llvm/lib/Target/X86/X86FlagsCopyLowering.cpp
@@ -601,8 +601,7 @@ bool X86FlagsCopyLoweringPass::runOnMachineFunction(MachineFunction &MF) {
         // Otherwise we can just rewrite in-place.
         if (X86::getCondFromCMov(MI) != X86::COND_INVALID) {
           rewriteCMov(*TestMBB, TestPos, TestLoc, MI, *FlagUse, CondRegs);
-        } else if (X86::getCondFromSETOpc(MI.getOpcode()) !=
-                   X86::COND_INVALID) {
+        } else if (X86::getCondFromSETCC(MI) != X86::COND_INVALID) {
           rewriteSetCC(*TestMBB, TestPos, TestLoc, MI, *FlagUse, CondRegs);
         } else if (MI.getOpcode() == TargetOpcode::COPY) {
           rewriteCopy(MI, *FlagUse, CopyDefI);
@@ -729,7 +728,7 @@ CondRegArray X86FlagsCopyLoweringPass::collectCondsInRegs(
   // Scan backwards across the range of instructions with live EFLAGS.
   for (MachineInstr &MI :
        llvm::reverse(llvm::make_range(MBB.begin(), TestPos))) {
-    X86::CondCode Cond = X86::getCondFromSETOpc(MI.getOpcode());
+    X86::CondCode Cond = X86::getCondFromSETCC(MI);
     if (Cond != X86::COND_INVALID && !MI.mayStore() && MI.getOperand(0).isReg() &&
         TRI->isVirtualRegister(MI.getOperand(0).getReg())) {
       assert(MI.getOperand(0).isDef() &&
@@ -750,7 +749,7 @@ unsigned X86FlagsCopyLoweringPass::promoteCondToReg(
     DebugLoc TestLoc, X86::CondCode Cond) {
   unsigned Reg = MRI->createVirtualRegister(PromoteRC);
   auto SetI = BuildMI(TestMBB, TestPos, TestLoc,
-                      TII->get(X86::getSETFromCond(Cond)), Reg);
+                      TII->get(X86::SETCCr), Reg).addImm(Cond);
   (void)SetI;
   LLVM_DEBUG(dbgs() << "    save cond: "; SetI->dump());
   ++NumSetCCsInserted;
@@ -1023,7 +1022,7 @@ void X86FlagsCopyLoweringPass::rewriteSetCC(MachineBasicBlock &TestMBB,
                                             MachineInstr &SetCCI,
                                             MachineOperand &FlagUse,
                                             CondRegArray &CondRegs) {
-  X86::CondCode Cond = X86::getCondFromSETOpc(SetCCI.getOpcode());
+  X86::CondCode Cond = X86::getCondFromSETCC(SetCCI);
   // Note that we can't usefully rewrite this to the inverse without complex
   // analysis of the users of the setCC. Largely we rely on duplicates which
   // could have been avoided already being avoided here.
diff --git a/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp b/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
index eeffc4d7080..273ed6b0fa6 100644
--- a/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ b/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -2326,11 +2326,14 @@ static X86::CondCode getCondFromNode(SDNode *N) {
   X86::CondCode CC = X86::COND_INVALID;
   if (CC == X86::COND_INVALID)
     CC = X86::getCondFromBranchOpc(N->getMachineOpcode());
-  if (CC == X86::COND_INVALID)
-    CC = X86::getCondFromSETOpc(N->getMachineOpcode());
   if (CC == X86::COND_INVALID) {
     unsigned Opc = N->getMachineOpcode();
-    if (Opc == X86::CMOV16rr || Opc == X86::CMOV32rr || Opc == X86::CMOV64rr)
+    if (Opc == X86::SETCCr)
+      CC = static_cast<X86::CondCode>(N->getConstantOperandVal(0));
+    else if (Opc == X86::SETCCm)
+      CC = static_cast<X86::CondCode>(N->getConstantOperandVal(5));
+    else if (Opc == X86::CMOV16rr || Opc == X86::CMOV32rr ||
+             Opc == X86::CMOV64rr)
       CC = static_cast<X86::CondCode>(N->getConstantOperandVal(2));
     else if (Opc == X86::CMOV16rm || Opc == X86::CMOV32rm ||
              Opc == X86::CMOV64rm)
diff --git a/llvm/lib/Target/X86/X86InstrCMovSetCC.td b/llvm/lib/Target/X86/X86InstrCMovSetCC.td
index 2c41169bf67..099f6aa8d8b 100644
--- a/llvm/lib/Target/X86/X86InstrCMovSetCC.td
+++ b/llvm/lib/Target/X86/X86InstrCMovSetCC.td
@@ -55,69 +55,52 @@ let Uses = [EFLAGS], Predicates = [HasCMov], Constraints = "$src1 = $dst",
 } // Uses = [EFLAGS], Predicates = [HasCMov], Constraints = "$src1 = $dst"
 } // isCodeGenOnly = 1, ForceDisassemble = 1
 
-multiclass CMOV_Aliases<string Name, int CC> {
-  def : InstAlias<Name#"{w}\t{$src, $dst|$dst, $src}",
+// SetCC instructions.
+let Uses = [EFLAGS], isCodeGenOnly = 1, ForceDisassemble = 1 in {
+  def SETCCr : I<0x90, MRMXrCC, (outs GR8:$dst), (ins ccode:$cond),
+                "set${cond}\t$dst",
+                [(set GR8:$dst, (X86setcc imm:$cond, EFLAGS))]>,
+                TB, Sched<[WriteSETCC]>;
+  def SETCCm : I<0x90, MRMXmCC, (outs), (ins i8mem:$dst, ccode:$cond),
+                "set${cond}\t$dst",
+                [(store (X86setcc imm:$cond, EFLAGS), addr:$dst)]>,
+                TB, Sched<[WriteSETCCStore]>;
+} // Uses = [EFLAGS]
+
+multiclass CMOV_SETCC_Aliases<string Cond, int CC> {
+  def : InstAlias<"cmov"#Cond#"{w}\t{$src, $dst|$dst, $src}",
                   (CMOV16rr GR16:$dst, GR16:$src, CC), 0>;
-  def : InstAlias<Name#"{w}\t{$src, $dst|$dst, $src}",
+  def : InstAlias<"cmov"#Cond#"{w}\t{$src, $dst|$dst, $src}",
                   (CMOV16rm GR16:$dst, i16mem:$src, CC), 0>;
-  def : InstAlias<Name#"{l}\t{$src, $dst|$dst, $src}",
+  def : InstAlias<"cmov"#Cond#"{l}\t{$src, $dst|$dst, $src}",
                   (CMOV32rr GR32:$dst, GR32:$src, CC), 0>;
-  def : InstAlias<Name#"{l}\t{$src, $dst|$dst, $src}",
+  def : InstAlias<"cmov"#Cond#"{l}\t{$src, $dst|$dst, $src}",
                   (CMOV32rm GR32:$dst, i32mem:$src, CC), 0>;
-  def : InstAlias<Name#"{q}\t{$src, $dst|$dst, $src}",
+  def : InstAlias<"cmov"#Cond#"{q}\t{$src, $dst|$dst, $src}",
                   (CMOV64rr GR64:$dst, GR64:$src, CC), 0>;
-  def : InstAlias<Name#"{q}\t{$src, $dst|$dst, $src}",
+  def : InstAlias<"cmov"#Cond#"{q}\t{$src, $dst|$dst, $src}",
                   (CMOV64rm GR64:$dst, i64mem:$src, CC), 0>;
-}
 
-defm : CMOV_Aliases<"cmovo" ,  0>;
-defm : CMOV_Aliases<"cmovno",  1>;
-defm : CMOV_Aliases<"cmovb" ,  2>;
-defm : CMOV_Aliases<"cmovae",  3>;
-defm : CMOV_Aliases<"cmove" ,  4>;
-defm : CMOV_Aliases<"cmovne",  5>;
-defm : CMOV_Aliases<"cmovbe",  6>;
-defm : CMOV_Aliases<"cmova" ,  7>;
-defm : CMOV_Aliases<"cmovs" ,  8>;
-defm : CMOV_Aliases<"cmovns",  9>;
-defm : CMOV_Aliases<"cmovp" , 10>;
-defm : CMOV_Aliases<"cmovnp", 11>;
-defm : CMOV_Aliases<"cmovl" , 12>;
-defm : CMOV_Aliases<"cmovge", 13>;
-defm : CMOV_Aliases<"cmovle", 14>;
-defm : CMOV_Aliases<"cmovg" , 15>;
-
-
-// SetCC instructions.
-multiclass SETCC<bits<8> opc, string Mnemonic, PatLeaf OpNode> {
-  let Uses = [EFLAGS] in {
-    def r    : I<opc, MRMXr,  (outs GR8:$dst), (ins),
-                     !strconcat(Mnemonic, "\t$dst"),
-                     [(set GR8:$dst, (X86setcc OpNode, EFLAGS))]>,
-                     TB, Sched<[WriteSETCC]>;
-    def m    : I<opc, MRMXm,  (outs), (ins i8mem:$dst),
-                     !strconcat(Mnemonic, "\t$dst"),
-                     [(store (X86setcc OpNode, EFLAGS), addr:$dst)]>,
-                     TB, Sched<[WriteSETCCStore]>;
-  } // Uses = [EFLAGS]
+  def : InstAlias<"set"#Cond#"\t$dst", (SETCCr GR8:$dst, CC), 0>;
+  def : InstAlias<"set"#Cond#"\t$dst", (SETCCm i8mem:$dst, CC), 0>;
 }
 
-defm SETO  : SETCC<0x90, "seto",  X86_COND_O>;   // is overflow bit set
-defm SETNO : SETCC<0x91, "setno", X86_COND_NO>;  // is overflow bit not set
-defm SETB  : SETCC<0x92, "setb",  X86_COND_B>;   // unsigned less than
-defm SETAE : SETCC<0x93, "setae", X86_COND_AE>;  // unsigned greater or equal
-defm SETE  : SETCC<0x94, "sete",  X86_COND_E>;   // equal to
-defm SETNE : SETCC<0x95, "setne", X86_COND_NE>;  // not equal to
-defm SETBE : SETCC<0x96, "setbe", X86_COND_BE>;  // unsigned less than or equal
-defm SETA  : SETCC<0x97, "seta",  X86_COND_A>;   // unsigned greater than
-defm SETS  : SETCC<0x98, "sets",  X86_COND_S>;   // is signed bit set
-defm SETNS : SETCC<0x99, "setns", X86_COND_NS>;  // is not signed
-defm SETP  : SETCC<0x9A, "setp",  X86_COND_P>;   // is parity bit set
-defm SETNP : SETCC<0x9B, "setnp", X86_COND_NP>;  // is parity bit not set
-defm SETL  : SETCC<0x9C, "setl",  X86_COND_L>;   // signed less than
-defm SETGE : SETCC<0x9D, "setge", X86_COND_GE>;  // signed greater or equal
-defm SETLE : SETCC<0x9E, "setle", X86_COND_LE>;  // signed less than or equal
-defm SETG  : SETCC<0x9F, "setg",  X86_COND_G>;   // signed greater than
+defm : CMOV_SETCC_Aliases<"o" ,  0>;
+defm : CMOV_SETCC_Aliases<"no",  1>;
+defm : CMOV_SETCC_Aliases<"b" ,  2>;
+defm : CMOV_SETCC_Aliases<"ae",  3>;
+defm : CMOV_SETCC_Aliases<"e" ,  4>;
+defm : CMOV_SETCC_Aliases<"ne",  5>;
+defm : CMOV_SETCC_Aliases<"be",  6>;
+defm : CMOV_SETCC_Aliases<"a" ,  7>;
+defm : CMOV_SETCC_Aliases<"s" ,  8>;
+defm : CMOV_SETCC_Aliases<"ns",  9>;
+defm : CMOV_SETCC_Aliases<"p" , 10>;
+defm : CMOV_SETCC_Aliases<"np", 11>;
+defm : CMOV_SETCC_Aliases<"l" , 12>;
+defm : CMOV_SETCC_Aliases<"ge", 13>;
+defm : CMOV_SETCC_Aliases<"le", 14>;
+defm : CMOV_SETCC_Aliases<"g" , 15>;
 
 // SALC is an undocumented instruction. Information for this instruction can be found
 // here http://www.rcollins.org/secrets/opcodes/SALC.html
diff --git a/llvm/lib/Target/X86/X86InstrCompiler.td b/llvm/lib/Target/X86/X86InstrCompiler.td
index 03494f3689f..1fcc33e8724 100644
--- a/llvm/lib/Target/X86/X86InstrCompiler.td
+++ b/llvm/lib/Target/X86/X86InstrCompiler.td
@@ -354,7 +354,7 @@ def : Pat<(i64 (sext (i8 (X86setcc_c X86_COND_B, EFLAGS)))),
 // this happens, it is great.  However, if we are left with an 8-bit sbb and an
 // and, we might as well just match it as a setb.
 def : Pat<(and (i8 (X86setcc_c X86_COND_B, EFLAGS)), 1),
-          (SETBr)>;
+          (SETCCr (i8 2))>;
 
 // Patterns to give priority when both inputs are zero so that we don't use
 // an immediate for the RHS.
diff --git a/llvm/lib/Target/X86/X86InstrFoldTables.cpp b/llvm/lib/Target/X86/X86InstrFoldTables.cpp
index e0be42b4d4a..59e62da55f2 100644
--- a/llvm/lib/Target/X86/X86InstrFoldTables.cpp
+++ b/llvm/lib/Target/X86/X86InstrFoldTables.cpp
@@ -322,22 +322,7 @@ static const X86MemoryFoldTableEntry MemoryFoldTable0[] = {
   { X86::PUSH16r,             X86::PUSH16rmm,           TB_FOLDED_LOAD },
   { X86::PUSH32r,             X86::PUSH32rmm,           TB_FOLDED_LOAD },
   { X86::PUSH64r,             X86::PUSH64rmm,           TB_FOLDED_LOAD },
-  { X86::SETAEr,              X86::SETAEm,              TB_FOLDED_STORE },
-  { X86::SETAr,               X86::SETAm,               TB_FOLDED_STORE },
-  { X86::SETBEr,              X86::SETBEm,              TB_FOLDED_STORE },
-  { X86::SETBr,               X86::SETBm,               TB_FOLDED_STORE },
-  { X86::SETEr,               X86::SETEm,               TB_FOLDED_STORE },
-  { X86::SETGEr,              X86::SETGEm,              TB_FOLDED_STORE },
-  { X86::SETGr,               X86::SETGm,               TB_FOLDED_STORE },
-  { X86::SETLEr,              X86::SETLEm,              TB_FOLDED_STORE },
-  { X86::SETLr,               X86::SETLm,               TB_FOLDED_STORE },
-  { X86::SETNEr,              X86::SETNEm,              TB_FOLDED_STORE },
-  { X86::SETNOr,              X86::SETNOm,              TB_FOLDED_STORE },
-  { X86::SETNPr,              X86::SETNPm,              TB_FOLDED_STORE },
-  { X86::SETNSr,              X86::SETNSm,              TB_FOLDED_STORE },
-  { X86::SETOr,               X86::SETOm,               TB_FOLDED_STORE },
-  { X86::SETPr,               X86::SETPm,               TB_FOLDED_STORE },
-  { X86::SETSr,               X86::SETSm,               TB_FOLDED_STORE },
+  { X86::SETCCr,              X86::SETCCm,              TB_FOLDED_STORE },
   { X86::TAILJMPr,            X86::TAILJMPm,            TB_FOLDED_LOAD },
   { X86::TAILJMPr64,          X86::TAILJMPm64,          TB_FOLDED_LOAD },
   { X86::TAILJMPr64_REX,      X86::TAILJMPm64_REX,      TB_FOLDED_LOAD },
diff --git a/llvm/lib/Target/X86/X86InstrFormats.td b/llvm/lib/Target/X86/X86InstrFormats.td
index 5b25101b33d..4bb6008ad6f 100644
--- a/llvm/lib/Target/X86/X86InstrFormats.td
+++ b/llvm/lib/Target/X86/X86InstrFormats.td
@@ -31,6 +31,7 @@ def MRMSrcMem      : Format<33>;
 def MRMSrcMem4VOp3 : Format<34>;
 def MRMSrcMemOp4   : Format<35>;
 def MRMSrcMemCC    : Format<36>;
+def MRMXmCC: Format<38>;
 def MRMXm  : Format<39>;
 def MRM0m  : Format<40>;  def MRM1m  : Format<41>;  def MRM2m  : Format<42>;
 def MRM3m  : Format<43>;  def MRM4m  : Format<44>;  def MRM5m  : Format<45>;
@@ -40,6 +41,7 @@ def MRMSrcReg      : Format<49>;
 def MRMSrcReg4VOp3 : Format<50>;
 def MRMSrcRegOp4   : Format<51>;
 def MRMSrcRegCC    : Format<52>;
+def MRMXrCC: Format<54>;
 def MRMXr  : Format<55>;
 def MRM0r  : Format<56>;  def MRM1r  : Format<57>;  def MRM2r  : Format<58>;
 def MRM3r  : Format<59>;  def MRM4r  : Format<60>;  def MRM5r  : Format<61>;
diff --git a/llvm/lib/Target/X86/X86InstrInfo.cpp b/llvm/lib/Target/X86/X86InstrInfo.cpp
index 911b6b0ec5a..ecb8a40b738 100644
--- a/llvm/lib/Target/X86/X86InstrInfo.cpp
+++ b/llvm/lib/Target/X86/X86InstrInfo.cpp
@@ -2001,26 +2001,13 @@ X86::CondCode X86::getCondFromBranchOpc(unsigned BrOpc) {
   }
 }
 
-/// Return condition code of a SET opcode.
-X86::CondCode X86::getCondFromSETOpc(unsigned Opc) {
-  switch (Opc) {
+/// Return condition code of a SETCC opcode.
+X86::CondCode X86::getCondFromSETCC(const MachineInstr &MI) {
+  switch (MI.getOpcode()) {
   default: return X86::COND_INVALID;
-  case X86::SETAr:  case X86::SETAm:  return X86::COND_A;
-  case X86::SETAEr: case X86::SETAEm: return X86::COND_AE;
-  case X86::SETBr:  case X86::SETBm:  return X86::COND_B;
-  case X86::SETBEr: case X86::SETBEm: return X86::COND_BE;
-  case X86::SETEr:  case X86::SETEm:  return X86::COND_E;
-  case X86::SETGr:  case X86::SETGm:  return X86::COND_G;
-  case X86::SETGEr: case X86::SETGEm: return X86::COND_GE;
-  case X86::SETLr:  case X86::SETLm:  return X86::COND_L;
-  case X86::SETLEr: case X86::SETLEm: return X86::COND_LE;
-  case X86::SETNEr: case X86::SETNEm: return X86::COND_NE;
-  case X86::SETNOr: case X86::SETNOm: return X86::COND_NO;
-  case X86::SETNPr: case X86::SETNPm: return X86::COND_NP;
-  case X86::SETNSr: case X86::SETNSm: return X86::COND_NS;
-  case X86::SETOr:  case X86::SETOm:  return X86::COND_O;
-  case X86::SETPr:  case X86::SETPm:  return X86::COND_P;
-  case X86::SETSr:  case X86::SETSm:  return X86::COND_S;
+  case X86::SETCCr: case X86::SETCCm:
+    return static_cast<X86::CondCode>(
+        MI.getOperand(MI.getDesc().getNumOperands() - 1).getImm());
   }
 }
 
@@ -2139,30 +2126,9 @@ X86::getX86ConditionCode(CmpInst::Predicate Predicate) {
   return std::make_pair(CC, NeedSwap);
 }
 
-/// Return a set opcode for the given condition and
-/// whether it has memory operand.
-unsigned X86::getSETFromCond(CondCode CC, bool HasMemoryOperand) {
-  static const uint16_t Opc[16][2] = {
-    { X86::SETOr,  X86::SETOm  },
-    { X86::SETNOr, X86::SETNOm },
-    { X86::SETBr,  X86::SETBm  },
-    { X86::SETAEr, X86::SETAEm },
-    { X86::SETEr,  X86::SETEm  },
-    { X86::SETNEr, X86::SETNEm },
-    { X86::SETBEr, X86::SETBEm },
-    { X86::SETAr,  X86::SETAm  },
-    { X86::SETSr,  X86::SETSm  },
-    { X86::SETNSr, X86::SETNSm },
-    { X86::SETPr,  X86::SETPm  },
-    { X86::SETNPr, X86::SETNPm },
-    { X86::SETLr,  X86::SETLm  },
-    { X86::SETGEr, X86::SETGEm },
-    { X86::SETLEr, X86::SETLEm },
-    { X86::SETGr,  X86::SETGm  },
-  };
-
-  assert(CC <= LAST_VALID_COND && "Can only handle standard cond codes");
-  return Opc[CC][HasMemoryOperand ? 1 : 0];
+/// Return a setcc opcode based on whether it has memory operand.
+unsigned X86::getSETOpc(bool HasMemoryOperand) {
+  return HasMemoryOperand ? X86::SETCCr : X86::SETCCm;
 }
 
 /// Return a cmov opcode for the given register size in bytes, and operand type.
@@ -3555,7 +3521,7 @@ bool X86InstrInfo::optimizeCompareInstr(MachineInstr &CmpInstr, unsigned SrcReg,
   // If we are done with the basic block, we need to check whether EFLAGS is
   // live-out.
   bool IsSafe = false;
-  SmallVector<std::pair<MachineInstr*, unsigned /*NewOpc*/>, 4> OpsToUpdate;
+  SmallVector<std::pair<MachineInstr*, X86::CondCode>, 4> OpsToUpdate;
   MachineBasicBlock::iterator E = CmpInstr.getParent()->end();
   for (++I; I != E; ++I) {
     const MachineInstr &Instr = *I;
@@ -3572,16 +3538,13 @@ bool X86InstrInfo::optimizeCompareInstr(MachineInstr &CmpInstr, unsigned SrcReg,
 
     // EFLAGS is used by this instruction.
     X86::CondCode OldCC = X86::COND_INVALID;
-    bool OpcIsSET = false;
     if (IsCmpZero || IsSwapped) {
       // We decode the condition code from opcode.
       if (Instr.isBranch())
         OldCC = X86::getCondFromBranchOpc(Instr.getOpcode());
       else {
-        OldCC = X86::getCondFromSETOpc(Instr.getOpcode());
-        if (OldCC != X86::COND_INVALID)
-          OpcIsSET = true;
-        else
+        OldCC = X86::getCondFromSETCC(Instr);
+        if (OldCC == X86::COND_INVALID)
           OldCC = X86::getCondFromCMov(Instr);
       }
       if (OldCC == X86::COND_INVALID) return false;
@@ -3627,21 +3590,10 @@ bool X86InstrInfo::optimizeCompareInstr(MachineInstr &CmpInstr, unsigned SrcReg,
     }
 
     if ((ShouldUpdateCC || IsSwapped) && ReplacementCC != OldCC) {
-      // Synthesize the new opcode.
-      bool HasMemoryOperand = Instr.hasOneMemOperand();
-      unsigned NewOpc;
-      if (Instr.isBranch())
-        NewOpc = GetCondBranchFromCond(ReplacementCC);
-      else if(OpcIsSET)
-        NewOpc = getSETFromCond(ReplacementCC, HasMemoryOperand);
-      else {
-        NewOpc = ReplacementCC;
-      }
-
       // Push the MachineInstr to OpsToUpdate.
       // If it is safe to remove CmpInstr, the condition code of these
       // instructions will be modified.
-      OpsToUpdate.push_back(std::make_pair(&*I, NewOpc));
+      OpsToUpdate.push_back(std::make_pair(&*I, ReplacementCC));
     }
     if (ModifyEFLAGS || Instr.killsRegister(X86::EFLAGS, TRI)) {
       // It is safe to remove CmpInstr if EFLAGS is updated again or killed.
@@ -3696,11 +3648,11 @@ bool X86InstrInfo::optimizeCompareInstr(MachineInstr &CmpInstr, unsigned SrcReg,
 
   // Modify the condition code of instructions in OpsToUpdate.
   for (auto &Op : OpsToUpdate) {
-    if (X86::getCondFromCMov(*Op.first) != X86::COND_INVALID)
+    if (Op.first->isBranch())
+      Op.first->setDesc(get(GetCondBranchFromCond(Op.second)));
+    else
       Op.first->getOperand(Op.first->getDesc().getNumOperands() - 1)
           .setImm(Op.second);
-    else
-      Op.first->setDesc(get(Op.second));
   }
   return true;
 }
diff --git a/llvm/lib/Target/X86/X86InstrInfo.h b/llvm/lib/Target/X86/X86InstrInfo.h
index b0e83527247..2203cd908c9 100644
--- a/llvm/lib/Target/X86/X86InstrInfo.h
+++ b/llvm/lib/Target/X86/X86InstrInfo.h
@@ -42,9 +42,8 @@ unsigned GetCondBranchFromCond(CondCode CC);
 /// the instruction operands should be swaped to match the condition code.
 std::pair<CondCode, bool> getX86ConditionCode(CmpInst::Predicate Predicate);
 
-/// Return a set opcode for the given condition and whether it has
-/// a memory operand.
-unsigned getSETFromCond(CondCode CC, bool HasMemoryOperand = false);
+/// Return a setcc opcode based on whether it has a memory operand.
+unsigned getSETOpc(bool HasMemoryOperand = false);
 
 /// Return a cmov opcode for the given register size in bytes, and operand type.
 unsigned getCMovOpcode(unsigned RegBytes, bool HasMemoryOperand = false);
@@ -53,7 +52,7 @@ unsigned getCMovOpcode(unsigned RegBytes, bool HasMemoryOperand = false);
 CondCode getCondFromBranchOpc(unsigned Opc);
 
 // Turn setCC opcode into condition code.
-CondCode getCondFromSETOpc(unsigned Opc);
+CondCode getCondFromSETCC(const MachineInstr &MI);
 
 // Turn CMov opcode into condition code.
 CondCode getCondFromCMov(const MachineInstr &MI);
diff --git a/llvm/lib/Target/X86/X86InstructionSelector.cpp b/llvm/lib/Target/X86/X86InstructionSelector.cpp
index 90ae6bb4df5..c105b4e63d9 100644
--- a/llvm/lib/Target/X86/X86InstructionSelector.cpp
+++ b/llvm/lib/Target/X86/X86InstructionSelector.cpp
@@ -947,7 +947,6 @@ bool X86InstructionSelector::selectCmp(MachineInstr &I,
   bool SwapArgs;
   std::tie(CC, SwapArgs) = X86::getX86ConditionCode(
       (CmpInst::Predicate)I.getOperand(1).getPredicate());
-  unsigned OpSet = X86::getSETFromCond(CC);
 
   unsigned LHS = I.getOperand(2).getReg();
   unsigned RHS = I.getOperand(3).getReg();
@@ -981,7 +980,7 @@ bool X86InstructionSelector::selectCmp(MachineInstr &I,
            .addReg(RHS);
 
   MachineInstr &SetInst = *BuildMI(*I.getParent(), I, I.getDebugLoc(),
-                                   TII.get(OpSet), I.getOperand(0).getReg());
+                                   TII.get(X86::SETCCr), I.getOperand(0).getReg()).addImm(CC);
 
   constrainSelectedInstRegOperands(CmpInst, TII, TRI, RBI);
   constrainSelectedInstRegOperands(SetInst, TII, TRI, RBI);
@@ -1002,8 +1001,8 @@ bool X86InstructionSelector::selectFCmp(MachineInstr &I,
 
   // FCMP_OEQ and FCMP_UNE cannot be checked with a single instruction.
   static const uint16_t SETFOpcTable[2][3] = {
-      {X86::SETEr, X86::SETNPr, X86::AND8rr},
-      {X86::SETNEr, X86::SETPr, X86::OR8rr}};
+      {X86::COND_E, X86::COND_NP, X86::AND8rr},
+      {X86::COND_NE, X86::COND_P, X86::OR8rr}};
   const uint16_t *SETFOpc = nullptr;
   switch (Predicate) {
   default:
@@ -1043,9 +1042,9 @@ bool X86InstructionSelector::selectFCmp(MachineInstr &I,
     unsigned FlagReg1 = MRI.createVirtualRegister(&X86::GR8RegClass);
     unsigned FlagReg2 = MRI.createVirtualRegister(&X86::GR8RegClass);
     MachineInstr &Set1 = *BuildMI(*I.getParent(), I, I.getDebugLoc(),
-                                  TII.get(SETFOpc[0]), FlagReg1);
+                                  TII.get(X86::SETCCr), FlagReg1).addImm(SETFOpc[0]);
     MachineInstr &Set2 = *BuildMI(*I.getParent(), I, I.getDebugLoc(),
-                                  TII.get(SETFOpc[1]), FlagReg2);
+                                  TII.get(X86::SETCCr), FlagReg2).addImm(SETFOpc[1]);
     MachineInstr &Set3 = *BuildMI(*I.getParent(), I, I.getDebugLoc(),
                                   TII.get(SETFOpc[2]), ResultReg)
                               .addReg(FlagReg1)
@@ -1063,7 +1062,6 @@ bool X86InstructionSelector::selectFCmp(MachineInstr &I,
   bool SwapArgs;
   std::tie(CC, SwapArgs) = X86::getX86ConditionCode(Predicate);
   assert(CC <= X86::LAST_VALID_COND && "Unexpected condition code.");
-  unsigned Opc = X86::getSETFromCond(CC);
 
   if (SwapArgs)
     std::swap(LhsReg, RhsReg);
@@ -1075,7 +1073,7 @@ bool X86InstructionSelector::selectFCmp(MachineInstr &I,
            .addReg(RhsReg);
 
   MachineInstr &Set =
-      *BuildMI(*I.getParent(), I, I.getDebugLoc(), TII.get(Opc), ResultReg);
+      *BuildMI(*I.getParent(), I, I.getDebugLoc(), TII.get(X86::SETCCr), ResultReg).addImm(CC);
   constrainSelectedInstRegOperands(CmpInst, TII, TRI, RBI);
   constrainSelectedInstRegOperands(Set, TII, TRI, RBI);
   I.eraseFromParent();
diff --git a/llvm/lib/Target/X86/X86SchedBroadwell.td b/llvm/lib/Target/X86/X86SchedBroadwell.td
index 4e76a8e2a6e..cd4358dc227 100644
--- a/llvm/lib/Target/X86/X86SchedBroadwell.td
+++ b/llvm/lib/Target/X86/X86SchedBroadwell.td
@@ -736,7 +736,6 @@ def: InstRW<[BWWriteResGroup20], (instrs CWD,
                                          ADC16i16, SBB16i16,
                                          ADC32i32, SBB32i32,
                                          ADC64i32, SBB64i32)>;
-def: InstRW<[BWWriteResGroup20], (instregex "SET(A|BE)r")>;
 
 def BWWriteResGroup22 : SchedWriteRes<[BWPort4,BWPort6,BWPort237]> {
   let Latency = 2;
@@ -815,7 +814,6 @@ def BWWriteResGroup38 : SchedWriteRes<[BWPort4,BWPort237,BWPort06,BWPort0156]> {
   let ResourceCycles = [1,1,1,1];
 }
 def: InstRW<[BWWriteResGroup38], (instrs CALL64pcrel32)>;
-def: InstRW<[BWWriteResGroup38], (instregex "SET(A|BE)m")>;
 
 def BWWriteResGroup39 : SchedWriteRes<[BWPort0,BWPort1]> {
   let Latency = 4;
@@ -1627,4 +1625,30 @@ def BWCMOVA_CMOVBErm :  SchedWriteVariant<[
 def : InstRW<[BWCMOVA_CMOVBErr], (instrs CMOV16rr, CMOV32rr, CMOV64rr)>;
 def : InstRW<[BWCMOVA_CMOVBErm], (instrs CMOV16rm, CMOV32rm, CMOV64rm)>;
 
+// SETCCs that use both Z and C flag require an extra uop.
+def BWWriteSETA_SETBEr : SchedWriteRes<[BWPort06,BWPort0156]> {
+  let Latency = 2;
+  let ResourceCycles = [1,1];
+  let NumMicroOps = 2;
+}
+
+def BWWriteSETA_SETBEm : SchedWriteRes<[BWPort4,BWPort237,BWPort06,BWPort0156]> {
+  let Latency = 3;
+  let ResourceCycles = [1,1,1,1];
+  let NumMicroOps = 4;
+}
+
+def BWSETA_SETBErr :  SchedWriteVariant<[
+  SchedVar<MCSchedPredicate<IsSETAr_Or_SETBEr>, [BWWriteSETA_SETBEr]>,
+  SchedVar<NoSchedPred,                         [WriteSETCC]>
+]>;
+
+def BWSETA_SETBErm :  SchedWriteVariant<[
+  SchedVar<MCSchedPredicate<IsSETAm_Or_SETBEm>, [BWWriteSETA_SETBEm]>,
+  SchedVar<NoSchedPred,                         [WriteSETCCStore]>
+]>;
+
+def : InstRW<[BWSETA_SETBErr], (instrs SETCCr)>;
+def : InstRW<[BWSETA_SETBErm], (instrs SETCCm)>;
+
 } // SchedModel
diff --git a/llvm/lib/Target/X86/X86SchedHaswell.td b/llvm/lib/Target/X86/X86SchedHaswell.td
index a8855f0f39a..1592a1169da 100644
--- a/llvm/lib/Target/X86/X86SchedHaswell.td
+++ b/llvm/lib/Target/X86/X86SchedHaswell.td
@@ -1126,7 +1126,6 @@ def HWWriteResGroup35 : SchedWriteRes<[HWPort06,HWPort0156]> {
   let ResourceCycles = [1,1];
 }
 def: InstRW<[HWWriteResGroup35], (instrs CWD, JCXZ, JECXZ, JRCXZ)>;
-def: InstRW<[HWWriteResGroup35], (instregex "SET(A|BE)r")>;
 
 def HWWriteResGroup36_2 : SchedWriteRes<[HWPort5,HWPort23]> {
   let Latency = 7;
@@ -1172,7 +1171,6 @@ def HWWriteResGroup45 : SchedWriteRes<[HWPort4,HWPort237,HWPort06,HWPort0156]> {
   let ResourceCycles = [1,1,1,1];
 }
 def: InstRW<[HWWriteResGroup45], (instrs CALL64pcrel32)>;
-def: InstRW<[HWWriteResGroup45], (instregex "SET(A|BE)m")>;
 
 def HWWriteResGroup46 : SchedWriteRes<[HWPort4,HWPort23,HWPort237,HWPort06]> {
   let Latency = 8;
@@ -1911,4 +1909,30 @@ def HWCMOVA_CMOVBErm :  SchedWriteVariant<[
 def : InstRW<[HWCMOVA_CMOVBErr], (instrs CMOV16rr, CMOV32rr, CMOV64rr)>;
 def : InstRW<[HWCMOVA_CMOVBErm], (instrs CMOV16rm, CMOV32rm, CMOV64rm)>;
 
+// SETCCs that use both Z and C flag require an extra uop.
+def HWWriteSETA_SETBEr : SchedWriteRes<[HWPort06,HWPort0156]> {
+  let Latency = 2;
+  let ResourceCycles = [1,1];
+  let NumMicroOps = 2;
+}
+
+def HWWriteSETA_SETBEm : SchedWriteRes<[HWPort4,HWPort237,HWPort06,HWPort0156]> {
+  let Latency = 3;
+  let ResourceCycles = [1,1,1,1];
+  let NumMicroOps = 4;
+}
+
+def HWSETA_SETBErr :  SchedWriteVariant<[
+  SchedVar<MCSchedPredicate<IsSETAr_Or_SETBEr>, [HWWriteSETA_SETBEr]>,
+  SchedVar<NoSchedPred,                         [WriteSETCC]>
+]>;
+
+def HWSETA_SETBErm :  SchedWriteVariant<[
+  SchedVar<MCSchedPredicate<IsSETAm_Or_SETBEm>, [HWWriteSETA_SETBEm]>,
+  SchedVar<NoSchedPred,                         [WriteSETCCStore]>
+]>;
+
+def : InstRW<[HWSETA_SETBErr], (instrs SETCCr)>;
+def : InstRW<[HWSETA_SETBErm], (instrs SETCCm)>;
+
 } // SchedModel
diff --git a/llvm/lib/Target/X86/X86SchedPredicates.td b/llvm/lib/Target/X86/X86SchedPredicates.td
index c1e5ad0834d..41bd776648f 100644
--- a/llvm/lib/Target/X86/X86SchedPredicates.td
+++ b/llvm/lib/Target/X86/X86SchedPredicates.td
@@ -72,3 +72,15 @@ def IsCMOVArm_Or_CMOVBErm : CheckAny<[
   CheckImmOperand_s<7, "X86::COND_A">,
   CheckImmOperand_s<7, "X86::COND_BE">
 ]>;
+
+// A predicate to check for COND_A and COND_BE SETCCs which have an extra uop
+// on recent Intel CPUs.
+def IsSETAr_Or_SETBEr : CheckAny<[
+  CheckImmOperand_s<1, "X86::COND_A">,
+  CheckImmOperand_s<1, "X86::COND_BE">
+]>;
+
+def IsSETAm_Or_SETBEm : CheckAny<[
+  CheckImmOperand_s<5, "X86::COND_A">,
+  CheckImmOperand_s<5, "X86::COND_BE">
+]>;
diff --git a/llvm/lib/Target/X86/X86SchedSandyBridge.td b/llvm/lib/Target/X86/X86SchedSandyBridge.td
index 234f3dcb552..842d67b5c82 100644
--- a/llvm/lib/Target/X86/X86SchedSandyBridge.td
+++ b/llvm/lib/Target/X86/X86SchedSandyBridge.td
@@ -615,13 +615,6 @@ def: InstRW<[SBWriteResGroup5], (instrs MMX_PABSBrr,
                                         MMX_PSIGNDrr,
                                         MMX_PSIGNWrr)>;
 
-def SBWriteResGroup9 : SchedWriteRes<[SBPort05]> {
-  let Latency = 2;
-  let NumMicroOps = 2;
-  let ResourceCycles = [2];
-}
-def: InstRW<[SBWriteResGroup9], (instregex "SET(A|BE)r")>;
-
 def SBWriteResGroup11 : SchedWriteRes<[SBPort015]> {
   let Latency = 2;
   let NumMicroOps = 2;
@@ -772,13 +765,6 @@ def SBWriteResGroup41 : SchedWriteRes<[SBPort5,SBPort015]> {
 }
 def: InstRW<[SBWriteResGroup41], (instrs FNINIT)>;
 
-def SBWriteResGroup43 : SchedWriteRes<[SBPort4,SBPort23,SBPort05]> {
-  let Latency = 3;
-  let NumMicroOps = 4;
-  let ResourceCycles = [1,1,2];
-}
-def: InstRW<[SBWriteResGroup43], (instregex "SET(A|BE)m")>;
-
 def SBWriteResGroup45 : SchedWriteRes<[SBPort0,SBPort4,SBPort23,SBPort15]> {
   let Latency = 5;
   let NumMicroOps = 4;
@@ -1198,4 +1184,30 @@ def SBCMOVA_CMOVBErm :  SchedWriteVariant<[
 def : InstRW<[SBCMOVA_CMOVBErr], (instrs CMOV16rr, CMOV32rr, CMOV64rr)>;
 def : InstRW<[SBCMOVA_CMOVBErm], (instrs CMOV16rm, CMOV32rm, CMOV64rm)>;
 
+// SETCCs that use both Z and C flag require an extra uop.
+def SBWriteSETA_SETBEr : SchedWriteRes<[SBPort05]> {
+  let Latency = 2;
+  let ResourceCycles = [2];
+  let NumMicroOps = 2;
+}
+
+def SBWriteSETA_SETBEm : SchedWriteRes<[SBPort4,SBPort23,SBPort05]> {
+  let Latency = 3;
+  let ResourceCycles = [1,1,2];
+  let NumMicroOps = 4;
+}
+
+def SBSETA_SETBErr :  SchedWriteVariant<[
+  SchedVar<MCSchedPredicate<IsSETAr_Or_SETBEr>, [SBWriteSETA_SETBEr]>,
+  SchedVar<NoSchedPred,                         [WriteSETCC]>
+]>;
+
+def SBSETA_SETBErm :  SchedWriteVariant<[
+  SchedVar<MCSchedPredicate<IsSETAm_Or_SETBEm>, [SBWriteSETA_SETBEm]>,
+  SchedVar<NoSchedPred,                         [WriteSETCCStore]>
+]>;
+
+def : InstRW<[SBSETA_SETBErr], (instrs SETCCr)>;
+def : InstRW<[SBSETA_SETBErm], (instrs SETCCm)>;
+
 } // SchedModel
diff --git a/llvm/lib/Target/X86/X86SchedSkylakeClient.td b/llvm/lib/Target/X86/X86SchedSkylakeClient.td
index 87dc88f482f..1119fd3fc11 100644
--- a/llvm/lib/Target/X86/X86SchedSkylakeClient.td
+++ b/llvm/lib/Target/X86/X86SchedSkylakeClient.td
@@ -698,13 +698,6 @@ def SKLWriteResGroup14 : SchedWriteRes<[SKLPort05]> {
 def: InstRW<[SKLWriteResGroup14], (instrs FDECSTP,
                                           MMX_MOVDQ2Qrr)>;
 
-def SKLWriteResGroup15 : SchedWriteRes<[SKLPort06]> {
-  let Latency = 2;
-  let NumMicroOps = 2;
-  let ResourceCycles = [2];
-}
-def: InstRW<[SKLWriteResGroup15], (instregex "SET(A|BE)r")>;
-
 def SKLWriteResGroup17 : SchedWriteRes<[SKLPort0156]> {
   let Latency = 2;
   let NumMicroOps = 2;
@@ -840,13 +833,6 @@ def SKLWriteResGroup43 : SchedWriteRes<[SKLPort0,SKLPort4,SKLPort237]> {
 }
 def: InstRW<[SKLWriteResGroup43], (instrs FNSTSWm)>;
 
-def SKLWriteResGroup44 : SchedWriteRes<[SKLPort4,SKLPort237,SKLPort06]> {
-  let Latency = 3;
-  let NumMicroOps = 4;
-  let ResourceCycles = [1,1,2];
-}
-def: InstRW<[SKLWriteResGroup44], (instregex "SET(A|BE)m")>;
-
 def SKLWriteResGroup45 : SchedWriteRes<[SKLPort4,SKLPort6,SKLPort237,SKLPort0156]> {
   let Latency = 3;
   let NumMicroOps = 4;
@@ -1782,4 +1768,30 @@ def SKLCMOVA_CMOVBErm :  SchedWriteVariant<[
 def : InstRW<[SKLCMOVA_CMOVBErr], (instrs CMOV16rr, CMOV32rr, CMOV64rr)>;
 def : InstRW<[SKLCMOVA_CMOVBErm], (instrs CMOV16rm, CMOV32rm, CMOV64rm)>;
 
+// SETCCs that use both Z and C flag require an extra uop.
+def SKLWriteSETA_SETBEr : SchedWriteRes<[SKLPort06]> {
+  let Latency = 2;
+  let ResourceCycles = [2];
+  let NumMicroOps = 2;
+}
+
+def SKLWriteSETA_SETBEm : SchedWriteRes<[SKLPort4,SKLPort237,SKLPort06]> {
+  let Latency = 3;
+  let ResourceCycles = [1,1,2];
+  let NumMicroOps = 4;
+}
+
+def SKLSETA_SETBErr :  SchedWriteVariant<[
+  SchedVar<MCSchedPredicate<IsSETAr_Or_SETBEr>, [SKLWriteSETA_SETBEr]>,
+  SchedVar<NoSchedPred,                         [WriteSETCC]>
+]>;
+
+def SKLSETA_SETBErm :  SchedWriteVariant<[
+  SchedVar<MCSchedPredicate<IsSETAm_Or_SETBEm>, [SKLWriteSETA_SETBEm]>,
+  SchedVar<NoSchedPred,                         [WriteSETCCStore]>
+]>;
+
+def : InstRW<[SKLSETA_SETBErr], (instrs SETCCr)>;
+def : InstRW<[SKLSETA_SETBErm], (instrs SETCCm)>;
+
 } // SchedModel
diff --git a/llvm/lib/Target/X86/X86SchedSkylakeServer.td b/llvm/lib/Target/X86/X86SchedSkylakeServer.td
index b532e7ae817..e3456073de3 100644
--- a/llvm/lib/Target/X86/X86SchedSkylakeServer.td
+++ b/llvm/lib/Target/X86/X86SchedSkylakeServer.td
@@ -722,13 +722,6 @@ def SKXWriteResGroup14 : SchedWriteRes<[SKXPort05]> {
 def: InstRW<[SKXWriteResGroup14], (instrs FDECSTP,
                                           MMX_MOVDQ2Qrr)>;
 
-def SKXWriteResGroup15 : SchedWriteRes<[SKXPort06]> {
-  let Latency = 2;
-  let NumMicroOps = 2;
-  let ResourceCycles = [2];
-}
-def: InstRW<[SKXWriteResGroup15], (instregex "SET(A|BE)r")>;
-
 def SKXWriteResGroup17 : SchedWriteRes<[SKXPort0156]> {
   let Latency = 2;
   let NumMicroOps = 2;
@@ -901,13 +894,6 @@ def SKXWriteResGroup45 : SchedWriteRes<[SKXPort0,SKXPort4,SKXPort237]> {
 }
 def: InstRW<[SKXWriteResGroup45], (instrs FNSTSWm)>;
 
-def SKXWriteResGroup46 : SchedWriteRes<[SKXPort4,SKXPort237,SKXPort06]> {
-  let Latency = 3;
-  let NumMicroOps = 4;
-  let ResourceCycles = [1,1,2];
-}
-def: InstRW<[SKXWriteResGroup46], (instregex "SET(A|BE)m")>;
-
 def SKXWriteResGroup47 : SchedWriteRes<[SKXPort4,SKXPort6,SKXPort237,SKXPort0156]> {
   let Latency = 3;
   let NumMicroOps = 4;
@@ -2498,4 +2484,30 @@ def SKXCMOVA_CMOVBErm :  SchedWriteVariant<[
 def : InstRW<[SKXCMOVA_CMOVBErr], (instrs CMOV16rr, CMOV32rr, CMOV64rr)>;
 def : InstRW<[SKXCMOVA_CMOVBErm], (instrs CMOV16rm, CMOV32rm, CMOV64rm)>;
 
+// SETCCs that use both Z and C flag require an extra uop.
+def SKXWriteSETA_SETBEr : SchedWriteRes<[SKXPort06]> {
+  let Latency = 2;
+  let ResourceCycles = [2];
+  let NumMicroOps = 2;
+}
+
+def SKXWriteSETA_SETBEm : SchedWriteRes<[SKXPort4,SKXPort237,SKXPort06]> {
+  let Latency = 3;
+  let ResourceCycles = [1,1,2];
+  let NumMicroOps = 4;
+}
+
+def SKXSETA_SETBErr :  SchedWriteVariant<[
+  SchedVar<MCSchedPredicate<IsSETAr_Or_SETBEr>, [SKXWriteSETA_SETBEr]>,
+  SchedVar<NoSchedPred,                         [WriteSETCC]>
+]>;
+
+def SKXSETA_SETBErm :  SchedWriteVariant<[
+  SchedVar<MCSchedPredicate<IsSETAm_Or_SETBEm>, [SKXWriteSETA_SETBEm]>,
+  SchedVar<NoSchedPred,                         [WriteSETCCStore]>
+]>;
+
+def : InstRW<[SKXSETA_SETBErr], (instrs SETCCr)>;
+def : InstRW<[SKXSETA_SETBErm], (instrs SETCCm)>;
+
 } // SchedModel
diff --git a/llvm/lib/Target/X86/X86ScheduleBdVer2.td b/llvm/lib/Target/X86/X86ScheduleBdVer2.td
index 6e11ac25172..36f4ad37e6e 100644
--- a/llvm/lib/Target/X86/X86ScheduleBdVer2.td
+++ b/llvm/lib/Target/X86/X86ScheduleBdVer2.td
@@ -472,8 +472,15 @@ def PdWriteSETGEmSETGmSETLEmSETLm : SchedWriteRes<[PdEX01]> {
   let ResourceCycles = [2];
   let NumMicroOps = 2;
 }
-def : InstRW<[PdWriteSETGEmSETGmSETLEmSETLm], (instrs SETGEm, SETGm,
-                                                      SETLEm, SETLm)>;
+
+def PdSETGEmSETGmSETLEmSETLm :  SchedWriteVariant<[
+  SchedVar<MCSchedPredicate<CheckImmOperand_s<5, "X86::COND_GE">>, [PdWriteSETGEmSETGmSETLEmSETLm]>,
+  SchedVar<MCSchedPredicate<CheckImmOperand_s<5, "X86::COND_G">>,  [PdWriteSETGEmSETGmSETLEmSETLm]>,
+  SchedVar<MCSchedPredicate<CheckImmOperand_s<5, "X86::COND_LE">>, [PdWriteSETGEmSETGmSETLEmSETLm]>,
+  SchedVar<MCSchedPredicate<CheckImmOperand_s<5, "X86::COND_L">>,  [PdWriteSETGEmSETGmSETLEmSETLm]>,
+  SchedVar<NoSchedPred,                                            [WriteSETCCStore]>
+]>;
+def : InstRW<[PdSETGEmSETGmSETLEmSETLm], (instrs SETCCm)>;
 
 defm : PdWriteRes<WriteLAHFSAHF,      [PdEX01],          2,  [],     2>;