Add support of RTM from TSX extension

- Add RTM code generation support throught 3 X86 intrinsics:
  xbegin()/xend() to start/end a transaction region, and xabort() to abort a
  tranaction region



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@167573 91177308-0d34-0410-b5e6-96231b3b80d8

11 files changed, 159 insertions, 46 deletions

diff --git a/lib/Target/X86/AsmParser/X86AsmLexer.cpp b/lib/Target/X86/AsmParser/X86AsmLexer.cpp
index 2794e60df2..66ad353709 100644
--- a/lib/Target/X86/AsmParser/X86AsmLexer.cpp
+++ b/lib/Target/X86/AsmParser/X86AsmLexer.cpp
@@ -18,19 +18,19 @@
 using namespace llvm;
 
 namespace {
-  
+
 class X86AsmLexer : public MCTargetAsmLexer {
   const MCAsmInfo &AsmInfo;
-  
+
   bool tentativeIsValid;
   AsmToken tentativeToken;
-  
+
   const AsmToken &lexTentative() {
     tentativeToken = getLexer()->Lex();
     tentativeIsValid = true;
     return tentativeToken;
   }
-  
+
   const AsmToken &lexDefinite() {
     if (tentativeIsValid) {
       tentativeIsValid = false;
@@ -38,7 +38,7 @@ class X86AsmLexer : public MCTargetAsmLexer {
     }
     return getLexer()->Lex();
   }
-  
+
   AsmToken LexTokenATT();
   AsmToken LexTokenIntel();
 protected:
@@ -47,7 +47,7 @@ protected:
       SetError(SMLoc(), "No MCAsmLexer installed");
       return AsmToken(AsmToken::Error, "", 0);
     }
-    
+
     switch (AsmInfo.getAssemblerDialect()) {
     default:
       SetError(SMLoc(), "Unhandled dialect");
@@ -71,33 +71,32 @@ public:
 
 AsmToken X86AsmLexer::LexTokenATT() {
   AsmToken lexedToken = lexDefinite();
-  
+
   switch (lexedToken.getKind()) {
   default:
     return lexedToken;
   case AsmToken::Error:
     SetError(Lexer->getErrLoc(), Lexer->getErr());
     return lexedToken;
-      
+
   case AsmToken::Percent: {
     const AsmToken &nextToken = lexTentative();
     if (nextToken.getKind() != AsmToken::Identifier)
       return lexedToken;
 
-      
     if (unsigned regID = MatchRegisterName(nextToken.getString())) {
       lexDefinite();
-        
+
       // FIXME: This is completely wrong when there is a space or other
       // punctuation between the % and the register name.
       StringRef regStr(lexedToken.getString().data(),
-                       lexedToken.getString().size() + 
+                       lexedToken.getString().size() +
                        nextToken.getString().size());
-      
-      return AsmToken(AsmToken::Register, regStr, 
+
+      return AsmToken(AsmToken::Register, regStr,
                       static_cast<int64_t>(regID));
     }
-    
+
     // Match register name failed.  If this is "db[0-7]", match it as an alias
     // for dr[0-7].
     if (nextToken.getString().size() == 3 &&
@@ -113,29 +112,29 @@ AsmToken X86AsmLexer::LexTokenATT() {
       case '6': RegNo = X86::DR6; break;
       case '7': RegNo = X86::DR7; break;
       }
-      
+
       if (RegNo != -1) {
         lexDefinite();
 
         // FIXME: This is completely wrong when there is a space or other
         // punctuation between the % and the register name.
         StringRef regStr(lexedToken.getString().data(),
-                         lexedToken.getString().size() + 
+                         lexedToken.getString().size() +
                          nextToken.getString().size());
-        return AsmToken(AsmToken::Register, regStr, 
+        return AsmToken(AsmToken::Register, regStr,
                         static_cast<int64_t>(RegNo));
       }
     }
-      
-   
+
+
     return lexedToken;
-  }    
+  }
   }
 }
 
 AsmToken X86AsmLexer::LexTokenIntel() {
   const AsmToken &lexedToken = lexDefinite();
-  
+
   switch(lexedToken.getKind()) {
   default:
     return lexedToken;
@@ -144,7 +143,7 @@ AsmToken X86AsmLexer::LexTokenIntel() {
     return lexedToken;
   case AsmToken::Identifier: {
     unsigned regID = MatchRegisterName(lexedToken.getString().lower());
-    
+
     if (regID)
       return AsmToken(AsmToken::Register,
                       lexedToken.getString(),
diff --git a/lib/Target/X86/MCTargetDesc/X86BaseInfo.h b/lib/Target/X86/MCTargetDesc/X86BaseInfo.h
index db597fbfca..7ea1961dec 100644
--- a/lib/Target/X86/MCTargetDesc/X86BaseInfo.h
+++ b/lib/Target/X86/MCTargetDesc/X86BaseInfo.h
@@ -276,9 +276,9 @@ namespace X86II {
     MRM_C1 = 33, MRM_C2 = 34, MRM_C3 = 35, MRM_C4 = 36,
     MRM_C8 = 37, MRM_C9 = 38, MRM_E8 = 39, MRM_F0 = 40,
     MRM_F8 = 41, MRM_F9 = 42, MRM_D0 = 45, MRM_D1 = 46,
-    MRM_D4 = 47, MRM_D8 = 48, MRM_D9 = 49, MRM_DA = 50,
-    MRM_DB = 51, MRM_DC = 52, MRM_DD = 53, MRM_DE = 54,
-    MRM_DF = 55,
+    MRM_D4 = 47, MRM_D5 = 48, MRM_D8 = 49, MRM_D9 = 50,
+    MRM_DA = 51, MRM_DB = 52, MRM_DC = 53, MRM_DD = 54,
+    MRM_DE = 55, MRM_DF = 56,
 
     /// RawFrmImm8 - This is used for the ENTER instruction, which has two
     /// immediates, the first of which is a 16-bit immediate (specified by
@@ -580,11 +580,11 @@ namespace X86II {
     case X86II::MRM_E8: case X86II::MRM_F0:
     case X86II::MRM_F8: case X86II::MRM_F9:
     case X86II::MRM_D0: case X86II::MRM_D1:
-    case X86II::MRM_D4: case X86II::MRM_D8:
-    case X86II::MRM_D9: case X86II::MRM_DA:
-    case X86II::MRM_DB: case X86II::MRM_DC:
-    case X86II::MRM_DD: case X86II::MRM_DE:
-    case X86II::MRM_DF:
+    case X86II::MRM_D4: case X86II::MRM_D5:
+    case X86II::MRM_D8: case X86II::MRM_D9:
+    case X86II::MRM_DA: case X86II::MRM_DB:
+    case X86II::MRM_DC: case X86II::MRM_DD:
+    case X86II::MRM_DE: case X86II::MRM_DF:
       return -1;
     }
   }
diff --git a/lib/Target/X86/MCTargetDesc/X86MCCodeEmitter.cpp b/lib/Target/X86/MCTargetDesc/X86MCCodeEmitter.cpp
index f6ebdacef0..122204ae75 100644
--- a/lib/Target/X86/MCTargetDesc/X86MCCodeEmitter.cpp
+++ b/lib/Target/X86/MCTargetDesc/X86MCCodeEmitter.cpp
@@ -1121,13 +1121,13 @@ EncodeInstruction(const MCInst &MI, raw_ostream &OS,
   case X86II::MRM_C3: case X86II::MRM_C4:
   case X86II::MRM_C8: case X86II::MRM_C9:
   case X86II::MRM_D0: case X86II::MRM_D1:
-  case X86II::MRM_D4: case X86II::MRM_D8:
-  case X86II::MRM_D9: case X86II::MRM_DA:
-  case X86II::MRM_DB: case X86II::MRM_DC:
-  case X86II::MRM_DD: case X86II::MRM_DE:
-  case X86II::MRM_DF: case X86II::MRM_E8:
-  case X86II::MRM_F0: case X86II::MRM_F8:
-  case X86II::MRM_F9:
+  case X86II::MRM_D4: case X86II::MRM_D5:
+  case X86II::MRM_D8: case X86II::MRM_D9:
+  case X86II::MRM_DA: case X86II::MRM_DB:
+  case X86II::MRM_DC: case X86II::MRM_DD:
+  case X86II::MRM_DE: case X86II::MRM_DF:
+  case X86II::MRM_E8: case X86II::MRM_F0:
+  case X86II::MRM_F8: case X86II::MRM_F9:
     EmitByte(BaseOpcode, CurByte, OS);
 
     unsigned char MRM;
@@ -1142,6 +1142,7 @@ EncodeInstruction(const MCInst &MI, raw_ostream &OS,
     case X86II::MRM_D0: MRM = 0xD0; break;
     case X86II::MRM_D1: MRM = 0xD1; break;
     case X86II::MRM_D4: MRM = 0xD4; break;
+    case X86II::MRM_D5: MRM = 0xD5; break;
     case X86II::MRM_D8: MRM = 0xD8; break;
     case X86II::MRM_D9: MRM = 0xD9; break;
     case X86II::MRM_DA: MRM = 0xDA; break;
diff --git a/lib/Target/X86/X86.td b/lib/Target/X86/X86.td
index 1b5c4d9753..8ad0bc08ac 100644
--- a/lib/Target/X86/X86.td
+++ b/lib/Target/X86/X86.td
@@ -118,6 +118,8 @@ def FeatureBMI     : SubtargetFeature<"bmi", "HasBMI", "true",
                                       "Support BMI instructions">;
 def FeatureBMI2    : SubtargetFeature<"bmi2", "HasBMI2", "true",
                                       "Support BMI2 instructions">;
+def FeatureRTM     : SubtargetFeature<"rtm", "HasRTM", "true",
+                                      "Support RTM instructions">;
 def FeatureLeaForSP : SubtargetFeature<"lea-sp", "UseLeaForSP", "true",
                                      "Use LEA for adjusting the stack pointer">;
 def FeatureSlowDivide : SubtargetFeature<"idiv-to-divb",
@@ -192,7 +194,8 @@ def : Proc<"core-avx2",       [FeatureAVX2, FeatureCMPXCHG16B, FeaturePOPCNT,
                                FeatureAES, FeaturePCLMUL, FeatureRDRAND,
                                FeatureF16C, FeatureFSGSBase,
                                FeatureMOVBE, FeatureLZCNT, FeatureBMI,
-                               FeatureBMI2, FeatureFMA]>;
+                               FeatureBMI2, FeatureFMA,
+                               FeatureRTM]>;
 
 def : Proc<"k6",              [FeatureMMX]>;
 def : Proc<"k6-2",            [Feature3DNow]>;
diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 920cbf28d0..52c9128c67 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -12195,6 +12195,63 @@ X86TargetLowering::isVectorClearMaskLegal(const SmallVectorImpl<int> &Mask,
 
 // private utility function
 
+/// Utility function to emit xbegin specifying the start of an RTM region.
+MachineBasicBlock *
+X86TargetLowering::EmitXBegin(MachineInstr *MI, MachineBasicBlock *MBB) const {
+  DebugLoc DL = MI->getDebugLoc();
+  const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+
+  const BasicBlock *BB = MBB->getBasicBlock();
+  MachineFunction::iterator I = MBB;
+  ++I;
+
+  // For the v = xbegin(), we generate
+  //
+  // thisMBB:
+  //  xbegin sinkMBB
+  //
+  // mainMBB:
+  //  eax = -1
+  //
+  // sinkMBB:
+  //  v = eax
+
+  MachineBasicBlock *thisMBB = MBB;
+  MachineFunction *MF = MBB->getParent();
+  MachineBasicBlock *mainMBB = MF->CreateMachineBasicBlock(BB);
+  MachineBasicBlock *sinkMBB = MF->CreateMachineBasicBlock(BB);
+  MF->insert(I, mainMBB);
+  MF->insert(I, sinkMBB);
+
+  // Transfer the remainder of BB and its successor edges to sinkMBB.
+  sinkMBB->splice(sinkMBB->begin(), MBB,
+                  llvm::next(MachineBasicBlock::iterator(MI)), MBB->end());
+  sinkMBB->transferSuccessorsAndUpdatePHIs(MBB);
+
+  // thisMBB:
+  //  xbegin sinkMBB
+  //  # fallthrough to mainMBB
+  //  # abortion to sinkMBB
+  BuildMI(thisMBB, DL, TII->get(X86::XBEGIN_4)).addMBB(sinkMBB);
+  thisMBB->addSuccessor(mainMBB);
+  thisMBB->addSuccessor(sinkMBB);
+
+  // mainMBB:
+  //  EAX = -1
+  BuildMI(mainMBB, DL, TII->get(X86::MOV32ri), X86::EAX).addImm(-1);
+  mainMBB->addSuccessor(sinkMBB);
+
+  // sinkMBB:
+  // EAX is live into the sinkMBB
+  sinkMBB->addLiveIn(X86::EAX);
+  BuildMI(*sinkMBB, sinkMBB->begin(), DL,
+          TII->get(TargetOpcode::COPY), MI->getOperand(0).getReg())
+    .addReg(X86::EAX);
+
+  MI->eraseFromParent();
+  return sinkMBB;
+}
+
 // Get CMPXCHG opcode for the specified data type.
 static unsigned getCmpXChgOpcode(EVT VT) {
   switch (VT.getSimpleVT().SimpleTy) {
@@ -13858,6 +13915,10 @@ X86TargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
   case X86::MONITOR:
     return EmitMonitor(MI, BB);
 
+  // xbegin
+  case X86::XBEGIN:
+    return EmitXBegin(MI, BB);
+
     // Atomic Lowering.
   case X86::ATOMAND8:
   case X86::ATOMAND16:
diff --git a/lib/Target/X86/X86ISelLowering.h b/lib/Target/X86/X86ISelLowering.h
index c6354b0801..7764330cc8 100644
--- a/lib/Target/X86/X86ISelLowering.h
+++ b/lib/Target/X86/X86ISelLowering.h
@@ -898,6 +898,10 @@ namespace llvm {
     MachineBasicBlock *EmitAtomicLoadArith6432(MachineInstr *MI,
                                                MachineBasicBlock *MBB) const;
 
+    /// Utility function to emit xbegin specifying the start of an RTM region.
+    MachineBasicBlock *EmitXBegin(MachineInstr *MI,
+                                  MachineBasicBlock *MBB) const;
+
     // Utility function to emit the low-level va_arg code for X86-64.
     MachineBasicBlock *EmitVAARG64WithCustomInserter(
                        MachineInstr *MI,
diff --git a/lib/Target/X86/X86InstrFormats.td b/lib/Target/X86/X86InstrFormats.td
index 55ad2ecb8a..268e9fc9c0 100644
--- a/lib/Target/X86/X86InstrFormats.td
+++ b/lib/Target/X86/X86InstrFormats.td
@@ -44,14 +44,15 @@ def RawFrmImm16 : Format<44>;
 def MRM_D0 : Format<45>;
 def MRM_D1 : Format<46>;
 def MRM_D4 : Format<47>;
-def MRM_D8 : Format<48>;
-def MRM_D9 : Format<49>;
-def MRM_DA : Format<50>;
-def MRM_DB : Format<51>;
-def MRM_DC : Format<52>;
-def MRM_DD : Format<53>;
-def MRM_DE : Format<54>;
-def MRM_DF : Format<55>;
+def MRM_D5 : Format<48>;
+def MRM_D8 : Format<49>;
+def MRM_D9 : Format<50>;
+def MRM_DA : Format<51>;
+def MRM_DB : Format<52>;
+def MRM_DC : Format<53>;
+def MRM_DD : Format<54>;
+def MRM_DE : Format<55>;
+def MRM_DF : Format<56>;
 
 // ImmType - This specifies the immediate type used by an instruction. This is
 // part of the ad-hoc solution used to emit machine instruction encodings by our
diff --git a/lib/Target/X86/X86InstrInfo.td b/lib/Target/X86/X86InstrInfo.td
index 2f637685b3..650fa95d7f 100644
--- a/lib/Target/X86/X86InstrInfo.td
+++ b/lib/Target/X86/X86InstrInfo.td
@@ -589,6 +589,7 @@ def HasFSGSBase  : Predicate<"Subtarget->hasFSGSBase()">;
 def HasLZCNT     : Predicate<"Subtarget->hasLZCNT()">;
 def HasBMI       : Predicate<"Subtarget->hasBMI()">;
 def HasBMI2      : Predicate<"Subtarget->hasBMI2()">;
+def HasRTM       : Predicate<"Subtarget->hasRTM()">;
 def FPStackf32   : Predicate<"!Subtarget->hasSSE1()">;
 def FPStackf64   : Predicate<"!Subtarget->hasSSE2()">;
 def HasCmpxchg16b: Predicate<"Subtarget->hasCmpxchg16b()">;
@@ -1706,6 +1707,8 @@ include "X86Instr3DNow.td"
 include "X86InstrVMX.td"
 include "X86InstrSVM.td"
 
+include "X86InstrTSX.td"
+
 // System instructions.
 include "X86InstrSystem.td"
 
diff --git a/lib/Target/X86/X86InstrTSX.td b/lib/Target/X86/X86InstrTSX.td
new file mode 100644
index 0000000000..ad55058ede
--- /dev/null
+++ b/lib/Target/X86/X86InstrTSX.td
@@ -0,0 +1,32 @@
+//===-- X86InstrVMX.td - TSX Instruction Set Extension -----*- tablegen -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file describes the instructions that make up the Intel TSX instruction
+// set.
+//
+//===----------------------------------------------------------------------===//
+
+//===----------------------------------------------------------------------===//
+// TSX instructions
+
+let usesCustomInserter = 1 in
+def XBEGIN : I<0, Pseudo, (outs GR32:$dst), (ins),
+               "# XBEGIN", [(set GR32:$dst, (int_x86_xbegin))]>,
+             Requires<[HasRTM]>;
+
+let isBranch = 1, isTerminator = 1, Defs = [EAX] in
+def XBEGIN_4 : Ii32PCRel<0xc7, MRM_F8, (outs), (ins brtarget:$dst),
+                         "xbegin\t$dst", []>;
+
+def XEND : I<0x01, MRM_D5, (outs), (ins),
+             "xend", [(int_x86_xend)]>, TB, Requires<[HasRTM]>;
+
+def XABORT : Ii8<0xc6, MRM_F8, (outs), (ins i8imm:$imm),
+                 "xabort\t$imm",
+                 [(int_x86_xabort imm:$imm)]>, Requires<[HasRTM]>;
diff --git a/lib/Target/X86/X86Subtarget.cpp b/lib/Target/X86/X86Subtarget.cpp
index 4130637a65..8fd3a27002 100644
--- a/lib/Target/X86/X86Subtarget.cpp
+++ b/lib/Target/X86/X86Subtarget.cpp
@@ -302,6 +302,10 @@ void X86Subtarget::AutoDetectSubtargetFeatures() {
         HasBMI2 = true;
         ToggleFeature(X86::FeatureBMI2);
       }
+      if (IsIntel && ((EBX >> 11) & 0x1)) {
+        HasRTM = true;
+        ToggleFeature(X86::FeatureRTM);
+      }
     }
   }
 }
@@ -330,6 +334,7 @@ X86Subtarget::X86Subtarget(const std::string &TT, const std::string &CPU,
   , HasLZCNT(false)
   , HasBMI(false)
   , HasBMI2(false)
+  , HasRTM(false)
   , IsBTMemSlow(false)
   , IsUAMemFast(false)
   , HasVectorUAMem(false)
diff --git a/lib/Target/X86/X86Subtarget.h b/lib/Target/X86/X86Subtarget.h
index 3a990fc5de..8bf4cc77f7 100644
--- a/lib/Target/X86/X86Subtarget.h
+++ b/lib/Target/X86/X86Subtarget.h
@@ -118,6 +118,9 @@ protected:
   /// HasBMI2 - Processor has BMI2 instructions.
   bool HasBMI2;
 
+  /// HasRTM - Processor has RTM instructions.
+  bool HasRTM;
+
   /// IsBTMemSlow - True if BT (bit test) of memory instructions are slow.
   bool IsBTMemSlow;
 
@@ -219,6 +222,7 @@ public:
   bool hasLZCNT() const { return HasLZCNT; }
   bool hasBMI() const { return HasBMI; }
   bool hasBMI2() const { return HasBMI2; }
+  bool hasRTM() const { return HasRTM; }
   bool isBTMemSlow() const { return IsBTMemSlow; }
   bool isUnalignedMemAccessFast() const { return IsUAMemFast; }
   bool hasVectorUAMem() const { return HasVectorUAMem; }