- Add MachineInstrBundle.h and MachineInstrBundle.cpp. This includes a function

to finalize MI bundles (i.e. add BUNDLE instruction and computing register def
  and use lists of the BUNDLE instruction) and a pass to unpack bundles.
- Teach more of MachineBasic and MachineInstr methods to be bundle aware.
- Switch Thumb2 IT block to MI bundles and delete the hazard recognizer hack to
  prevent IT blocks from being broken apart.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146542 91177308-0d34-0410-b5e6-96231b3b80d8

include/llvm/CodeGen/MachineBasicBlock.h

@@ -194,21 +194,21 @@ public:
       return tmp;
     }
 
-    IterTy getInsnIterator() const {
+    IterTy getInstrIterator() const {
       return MII;
     }    
   };
 
-  typedef Instructions::iterator                                  insn_iterator;
+  typedef Instructions::iterator                                 instr_iterator;
-  typedef Instructions::const_iterator                      const_insn_iterator;
+  typedef Instructions::const_iterator                     const_instr_iterator;
-  typedef std::reverse_iterator<insn_iterator>            reverse_insn_iterator;
+  typedef std::reverse_iterator<instr_iterator>          reverse_instr_iterator;
   typedef
-  std::reverse_iterator<const_insn_iterator>        const_reverse_insn_iterator;
+  std::reverse_iterator<const_instr_iterator>      const_reverse_instr_iterator;
 
   typedef
-  bundle_iterator<MachineInstr,insn_iterator>                          iterator;
+  bundle_iterator<MachineInstr,instr_iterator>                         iterator;
   typedef
-  bundle_iterator<const MachineInstr,const_insn_iterator>        const_iterator;
+  bundle_iterator<const MachineInstr,const_instr_iterator>       const_iterator;
   typedef std::reverse_iterator<const_iterator>          const_reverse_iterator;
   typedef std::reverse_iterator<iterator>                      reverse_iterator;
 
@@ -221,44 +221,44 @@ public:
   const MachineInstr& front() const { return Insts.front(); }
   const MachineInstr& back()  const { return Insts.back(); }
 
-  insn_iterator                insn_begin()       { return Insts.begin();  }
+  instr_iterator                instr_begin()       { return Insts.begin();  }
-  const_insn_iterator          insn_begin() const { return Insts.begin();  }
+  const_instr_iterator          instr_begin() const { return Insts.begin();  }
-  insn_iterator                  insn_end()       { return Insts.end();    }
+  instr_iterator                  instr_end()       { return Insts.end();    }
-  const_insn_iterator            insn_end() const { return Insts.end();    }
+  const_instr_iterator            instr_end() const { return Insts.end();    }
-  reverse_insn_iterator       insn_rbegin()       { return Insts.rbegin(); }
+  reverse_instr_iterator       instr_rbegin()       { return Insts.rbegin(); }
-  const_reverse_insn_iterator insn_rbegin() const { return Insts.rbegin(); }
+  const_reverse_instr_iterator instr_rbegin() const { return Insts.rbegin(); }
-  reverse_insn_iterator       insn_rend  ()       { return Insts.rend();   }
+  reverse_instr_iterator       instr_rend  ()       { return Insts.rend();   }
-  const_reverse_insn_iterator insn_rend  () const { return Insts.rend();   }
+  const_reverse_instr_iterator instr_rend  () const { return Insts.rend();   }
 
   iterator                begin()       { return Insts.begin();  }
   const_iterator          begin() const { return Insts.begin();  }
   iterator                  end()       {
-    insn_iterator II = insn_end();
+    instr_iterator II = instr_end();
-    if (II != insn_begin()) {
+    if (II != instr_begin()) {
       while (II->isInsideBundle())
         --II;
     }
     return II;
   }
   const_iterator            end() const {
-    const_insn_iterator II = insn_end();
+    const_instr_iterator II = instr_end();
-    if (II != insn_begin()) {
+    if (II != instr_begin()) {
       while (II->isInsideBundle())
         --II;
     }
     return II;
   }
   reverse_iterator       rbegin()       {
-    reverse_insn_iterator II = insn_rbegin();
+    reverse_instr_iterator II = instr_rbegin();
-    if (II != insn_rend()) {
+    if (II != instr_rend()) {
       while (II->isInsideBundle())
         ++II;
     }
     return II;
   }
   const_reverse_iterator rbegin() const {
-    const_reverse_insn_iterator II = insn_rbegin();
+    const_reverse_instr_iterator II = instr_rbegin();
-    if (II != insn_rend()) {
+    if (II != instr_rend()) {
       while (II->isInsideBundle())
         ++II;
     }
@@ -442,9 +442,9 @@ public:
   iterator getFirstTerminator();
   const_iterator getFirstTerminator() const;
 
-  /// getFirstInsnTerminator - Same getFirstTerminator but it ignores bundles
+  /// getFirstInstrTerminator - Same getFirstTerminator but it ignores bundles
-  /// and return an insn_iterator instead.
+  /// and return an instr_iterator instead.
-  insn_iterator getFirstInsnTerminator();
+  instr_iterator getFirstInstrTerminator();
 
   /// getLastNonDebugInstr - returns an iterator to the last non-debug
   /// instruction in the basic block, or end()
@@ -464,68 +464,80 @@ public:
   void push_back(MachineInstr *MI) { Insts.push_back(MI); }
 
   template<typename IT>
-  void insert(insn_iterator I, IT S, IT E) {
+  void insert(instr_iterator I, IT S, IT E) {
     Insts.insert(I, S, E);
   }
-  insn_iterator insert(insn_iterator I, MachineInstr *M) {
+  instr_iterator insert(instr_iterator I, MachineInstr *M) {
     return Insts.insert(I, M);
   }
-  insn_iterator insertAfter(insn_iterator I, MachineInstr *M) { 
+  instr_iterator insertAfter(instr_iterator I, MachineInstr *M) { 
     return Insts.insertAfter(I, M); 
   }
 
   template<typename IT>
   void insert(iterator I, IT S, IT E) {
-    Insts.insert(I.getInsnIterator(), S, E);
+    Insts.insert(I.getInstrIterator(), S, E);
   }
   iterator insert(iterator I, MachineInstr *M) {
-    return Insts.insert(I.getInsnIterator(), M);
+    return Insts.insert(I.getInstrIterator(), M);
   }
   iterator insertAfter(iterator I, MachineInstr *M) { 
-    return Insts.insertAfter(I.getInsnIterator(), M); 
+    return Insts.insertAfter(I.getInstrIterator(), M); 
   }
 
-  // erase - Remove the specified element or range from the instruction list.
+  /// erase - Remove the specified element or range from the instruction list.
-  // These functions delete any instructions removed.
+  /// These functions delete any instructions removed.
-  //
+  ///
-  insn_iterator erase(insn_iterator I) {
+  instr_iterator erase(instr_iterator I) {
     return Insts.erase(I);
   }
-  insn_iterator erase(insn_iterator I, insn_iterator E) {
+  instr_iterator erase(instr_iterator I, instr_iterator E) {
     return Insts.erase(I, E);
   }
-
+  instr_iterator erase_instr(MachineInstr *I) {
-  iterator erase(iterator I)             {
+    instr_iterator MII(I);
-    return Insts.erase(I.getInsnIterator());
+    return erase(MII);
   }
+
+  iterator erase(iterator I);
   iterator erase(iterator I, iterator E) {
-    return Insts.erase(I.getInsnIterator(), E.getInsnIterator());
+    return Insts.erase(I.getInstrIterator(), E.getInstrIterator());
+  }
+  iterator erase(MachineInstr *I) {
+    iterator MII(I);
+    return erase(MII);
   }
 
-  iterator erase(MachineInstr *I)        { iterator MII(I); return erase(MII); }
+  /// remove - Remove the instruction from the instruction list. This function
-  MachineInstr *remove(MachineInstr *I)  { return Insts.remove(I); }
+  /// does not delete the instruction. WARNING: Note, if the specified
-  void clear()                           { Insts.clear(); }
+  /// instruction is a bundle this function will remove all the bundled
+  /// instructions as well. It is up to the caller to keep a list of the
+  /// bundled instructions and re-insert them if desired. This function is
+  /// *not recommended* for manipulating instructions with bundled. Use
+  /// splice instead.
+  MachineInstr *remove(MachineInstr *I);
+  void clear() {
+    Insts.clear();
+  }
 
   /// splice - Take an instruction from MBB 'Other' at the position From,
   /// and insert it into this MBB right before 'where'.
-  void splice(insn_iterator where, MachineBasicBlock *Other,
+  void splice(instr_iterator where, MachineBasicBlock *Other,
-              insn_iterator From) {
+              instr_iterator From) {
     Insts.splice(where, Other->Insts, From);
   }
-  void splice(iterator where, MachineBasicBlock *Other, iterator From) {
+  void splice(iterator where, MachineBasicBlock *Other, iterator From);
-    Insts.splice(where.getInsnIterator(), Other->Insts, From.getInsnIterator());
-  }
 
   /// splice - Take a block of instructions from MBB 'Other' in the range [From,
   /// To), and insert them into this MBB right before 'where'.
-  void splice(insn_iterator where, MachineBasicBlock *Other, insn_iterator From,
+  void splice(instr_iterator where, MachineBasicBlock *Other, instr_iterator From,
-              insn_iterator To) {
+              instr_iterator To) {
     Insts.splice(where, Other->Insts, From, To);
   }
   void splice(iterator where, MachineBasicBlock *Other, iterator From,
               iterator To) {
-    Insts.splice(where.getInsnIterator(), Other->Insts,
+    Insts.splice(where.getInstrIterator(), Other->Insts,
-                 From.getInsnIterator(), To.getInsnIterator());
+                 From.getInstrIterator(), To.getInstrIterator());
   }
 
   /// removeFromParent - This method unlinks 'this' from the containing
@@ -552,9 +564,9 @@ public:
 
   /// findDebugLoc - find the next valid DebugLoc starting at MBBI, skipping
   /// any DBG_VALUE instructions.  Return UnknownLoc if there is none.
-  DebugLoc findDebugLoc(insn_iterator MBBI);
+  DebugLoc findDebugLoc(instr_iterator MBBI);
   DebugLoc findDebugLoc(iterator MBBI) {
-    return findDebugLoc(MBBI.getInsnIterator());
+    return findDebugLoc(MBBI.getInstrIterator());
   }
 
   // Debugging methods.

include/llvm/CodeGen/MachineInstr.h

@@ -175,6 +175,11 @@ public:
     Flags = flags;
   }
 
+  /// clearFlag - Clear a MI flag.
+  void clearFlag(MIFlag Flag) {
+    Flags &= ~((uint8_t)Flag);
+  }
+
   /// isInsideBundle - Return true if MI is in a bundle (but not the first MI
   /// in a bundle).
   ///
@@ -215,6 +220,15 @@ public:
     return getFlag(InsideBundle);
   }
 
+  /// setIsInsideBundle - Set InsideBundle bit.
+  ///
+  void setIsInsideBundle(bool Val = true) {
+    if (Val)
+      setFlag(InsideBundle);
+    else
+      clearFlag(InsideBundle);
+  }
+
   /// getDebugLoc - Returns the debug location id of this MachineInstr.
   ///
   DebugLoc getDebugLoc() const { return debugLoc; }
@@ -589,6 +603,9 @@ public:
   bool isRegSequence() const {
     return getOpcode() == TargetOpcode::REG_SEQUENCE;
   }
+  bool isBundle() const {
+    return getOpcode() == TargetOpcode::BUNDLE;
+  }
   bool isCopy() const {
     return getOpcode() == TargetOpcode::COPY;
   }
@@ -608,6 +625,9 @@ public:
       getOperand(0).getSubReg() == getOperand(1).getSubReg();
   }
 
+  /// getBundleSize - Return the number of instructions inside the MI bundle.
+  unsigned getBundleSize() const;
+
   /// readsRegister - Return true if the MachineInstr reads the specified
   /// register. If TargetRegisterInfo is passed, then it also checks if there
   /// is a read of a super-register.

include/llvm/CodeGen/MachineInstrBuilder.h

@@ -209,6 +209,30 @@ inline MachineInstrBuilder BuildMI(MachineBasicBlock &BB,
   return MachineInstrBuilder(MI).addReg(DestReg, RegState::Define);
 }
 
+inline MachineInstrBuilder BuildMI(MachineBasicBlock &BB,
+                                   MachineBasicBlock::instr_iterator I,
+                                   DebugLoc DL,
+                                   const MCInstrDesc &MCID,
+                                   unsigned DestReg) {
+  MachineInstr *MI = BB.getParent()->CreateMachineInstr(MCID, DL);
+  BB.insert(I, MI);
+  return MachineInstrBuilder(MI).addReg(DestReg, RegState::Define);
+}
+
+inline MachineInstrBuilder BuildMI(MachineBasicBlock &BB,
+                                   MachineInstr *I,
+                                   DebugLoc DL,
+                                   const MCInstrDesc &MCID,
+                                   unsigned DestReg) {
+  if (I->isInsideBundle()) {
+    MachineBasicBlock::instr_iterator MII = I;
+    return BuildMI(BB, MII, DL, MCID, DestReg);
+  }
+
+  MachineBasicBlock::iterator MII = I;
+  return BuildMI(BB, MII, DL, MCID, DestReg);
+}
+
 /// BuildMI - This version of the builder inserts the newly-built
 /// instruction before the given position in the given MachineBasicBlock, and
 /// does NOT take a destination register.
@@ -222,6 +246,28 @@ inline MachineInstrBuilder BuildMI(MachineBasicBlock &BB,
   return MachineInstrBuilder(MI);
 }
 
+inline MachineInstrBuilder BuildMI(MachineBasicBlock &BB,
+                                   MachineBasicBlock::instr_iterator I,
+                                   DebugLoc DL,
+                                   const MCInstrDesc &MCID) {
+  MachineInstr *MI = BB.getParent()->CreateMachineInstr(MCID, DL);
+  BB.insert(I, MI);
+  return MachineInstrBuilder(MI);
+}
+
+inline MachineInstrBuilder BuildMI(MachineBasicBlock &BB,
+                                   MachineInstr *I,
+                                   DebugLoc DL,
+                                   const MCInstrDesc &MCID) {
+  if (I->isInsideBundle()) {
+    MachineBasicBlock::instr_iterator MII = I;
+    return BuildMI(BB, MII, DL, MCID);
+  }
+
+  MachineBasicBlock::iterator MII = I;
+  return BuildMI(BB, MII, DL, MCID);
+}
+
 /// BuildMI - This version of the builder inserts the newly-built
 /// instruction at the end of the given MachineBasicBlock, and does NOT take a
 /// destination register.

include/llvm/CodeGen/MachineInstrBundle.h

@@ -0,0 +1,34 @@
+//===-- CodeGen/MachineInstBundle.h - MI bundle utilities -------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file provide utility functions to manipulate machine instruction
+// bundles.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_CODEGEN_MACHINEINSTRBUNDLE_H
+#define LLVM_CODEGEN_MACHINEINSTRBUNDLE_H
+
+#include "llvm/CodeGen/MachineBasicBlock.h"
+
+namespace llvm {
+
+/// FinalizeBundle - Finalize a machine instruction bundle which includes
+/// a sequence of instructions starting from FirstMI to LastMI (inclusive).
+/// This routine adds a BUNDLE instruction to represent the bundle, it adds
+/// IsInternalRead markers to MachineOperands which are defined inside the
+/// bundle, and it copies externally visible defs and uses to the BUNDLE
+/// instruction.
+void FinalizeBundle(MachineBasicBlock &MBB,
+                    MachineBasicBlock::instr_iterator FirstMI,
+                    MachineBasicBlock::instr_iterator LastMI);
+  
+} // End llvm namespace
+
+#endif

include/llvm/CodeGen/Passes.h

@@ -238,6 +238,10 @@ namespace llvm {
   ///
   FunctionPass *createExecutionDependencyFixPass(const TargetRegisterClass *RC);
 
+  /// createUnpackMachineBundles - This pass unpack machine instruction bundles.
+  ///
+  FunctionPass *createUnpackMachineBundlesPass();
+
 } // End llvm namespace
 
 #endif

include/llvm/InitializePasses.h

@@ -231,6 +231,7 @@ void initializeUnreachableMachineBlockElimPass(PassRegistry&);
 void initializeVerifierPass(PassRegistry&);
 void initializeVirtRegMapPass(PassRegistry&);
 void initializeInstSimplifierPass(PassRegistry&);
+void initializeUnpackMachineBundlesPass(PassRegistry&);
 
 }
 

lib/CodeGen/BranchFolding.cpp

@@ -926,8 +926,9 @@ bool BranchFolder::TailMergeBlocks(MachineFunction &MF) {
       if (MergePotentials.size() >= 2)
         MadeChange |= TryTailMergeBlocks(IBB, PredBB);
       // Reinsert an unconditional branch if needed.
-      // The 1 below can occur as a result of removing blocks in TryTailMergeBlocks.
+      // The 1 below can occur as a result of removing blocks in
-      PredBB = prior(I);      // this may have been changed in TryTailMergeBlocks
+      // TryTailMergeBlocks.
+      PredBB = prior(I);     // this may have been changed in TryTailMergeBlocks
       if (MergePotentials.size() == 1 &&
           MergePotentials.begin()->getBlock() != PredBB)
         FixTail(MergePotentials.begin()->getBlock(), IBB, TII);

lib/CodeGen/MachineBasicBlock.cpp

@@ -73,8 +73,8 @@ void ilist_traits<MachineBasicBlock>::addNodeToList(MachineBasicBlock *N) {
 
   // Make sure the instructions have their operands in the reginfo lists.
   MachineRegisterInfo &RegInfo = MF.getRegInfo();
-  for (MachineBasicBlock::insn_iterator I = N->insn_begin(), E = N->insn_end();
+  for (MachineBasicBlock::instr_iterator
-       I != E; ++I)
+         I = N->instr_begin(), E = N->instr_end(); I != E; ++I)
     I->AddRegOperandsToUseLists(RegInfo);
 
   LeakDetector::removeGarbageObject(N);
@@ -141,7 +141,7 @@ void ilist_traits<MachineInstr>::deleteNode(MachineInstr* MI) {
 }
 
 MachineBasicBlock::iterator MachineBasicBlock::getFirstNonPHI() {
-  insn_iterator I = insn_begin();
+  instr_iterator I = instr_begin();
   while (I != end() && I->isPHI())
     ++I;
   assert(!I->isInsideBundle() && "First non-phi MI cannot be inside a bundle!");
@@ -178,18 +178,18 @@ MachineBasicBlock::getFirstTerminator() const {
   return I;
 }
 
-MachineBasicBlock::insn_iterator MachineBasicBlock::getFirstInsnTerminator() {
+MachineBasicBlock::instr_iterator MachineBasicBlock::getFirstInstrTerminator() {
-  insn_iterator I = insn_end();
+  instr_iterator I = instr_end();
-  while (I != insn_begin() && ((--I)->isTerminator() || I->isDebugValue()))
+  while (I != instr_begin() && ((--I)->isTerminator() || I->isDebugValue()))
     ; /*noop */
-  while (I != insn_end() && !I->isTerminator())
+  while (I != instr_end() && !I->isTerminator())
     ++I;
   return I;
 }
 
 MachineBasicBlock::iterator MachineBasicBlock::getLastNonDebugInstr() {
   // Skip over end-of-block dbg_value instructions.
-  insn_iterator B = insn_begin(), I = insn_end();
+  instr_iterator B = instr_begin(), I = instr_end();
   while (I != B) {
     --I;
     // Return instruction that starts a bundle.
@@ -204,7 +204,7 @@ MachineBasicBlock::iterator MachineBasicBlock::getLastNonDebugInstr() {
 MachineBasicBlock::const_iterator
 MachineBasicBlock::getLastNonDebugInstr() const {
   // Skip over end-of-block dbg_value instructions.
-  const_insn_iterator B = insn_begin(), I = insn_end();
+  const_instr_iterator B = instr_begin(), I = instr_end();
   while (I != B) {
     --I;
     // Return instruction that starts a bundle.
@@ -283,13 +283,15 @@ void MachineBasicBlock::print(raw_ostream &OS, SlotIndexes *Indexes) const {
     OS << '\n';
   }
 
-  for (const_iterator I = begin(); I != end(); ++I) {
+  for (const_instr_iterator I = instr_begin(); I != instr_end(); ++I) {
     if (Indexes) {
       if (Indexes->hasIndex(I))
         OS << Indexes->getInstructionIndex(I);
       OS << '\t';
     }
     OS << '\t';
+    if (I->isInsideBundle())
+      OS << "  * ";
     I->print(OS, &getParent()->getTarget());
   }
 
@@ -495,8 +497,8 @@ MachineBasicBlock::transferSuccessorsAndUpdatePHIs(MachineBasicBlock *fromMBB) {
     fromMBB->removeSuccessor(Succ);
 
     // Fix up any PHI nodes in the successor.
-    for (MachineBasicBlock::insn_iterator MI = Succ->insn_begin(),
+    for (MachineBasicBlock::instr_iterator MI = Succ->instr_begin(),
-           ME = Succ->insn_end(); MI != ME && MI->isPHI(); ++MI)
+           ME = Succ->instr_end(); MI != ME && MI->isPHI(); ++MI)
       for (unsigned i = 2, e = MI->getNumOperands()+1; i != e; i += 2) {
         MachineOperand &MO = MI->getOperand(i);
         if (MO.getMBB() == fromMBB)
@@ -598,7 +600,7 @@ MachineBasicBlock::SplitCriticalEdge(MachineBasicBlock *Succ, Pass *P) {
   // Collect a list of virtual registers killed by the terminators.
   SmallVector<unsigned, 4> KilledRegs;
   if (LV)
-    for (insn_iterator I = getFirstInsnTerminator(), E = insn_end();
+    for (instr_iterator I = getFirstInstrTerminator(), E = instr_end();
          I != E; ++I) {
       MachineInstr *MI = I;
       for (MachineInstr::mop_iterator OI = MI->operands_begin(),
@@ -626,8 +628,9 @@ MachineBasicBlock::SplitCriticalEdge(MachineBasicBlock *Succ, Pass *P) {
   }
 
   // Fix PHI nodes in Succ so they refer to NMBB instead of this
-  for (MachineBasicBlock::insn_iterator
+  for (MachineBasicBlock::instr_iterator
-         i = Succ->insn_begin(),e = Succ->insn_end(); i != e && i->isPHI(); ++i)
+         i = Succ->instr_begin(),e = Succ->instr_end();
+       i != e && i->isPHI(); ++i)
     for (unsigned ni = 1, ne = i->getNumOperands(); ni != ne; ni += 2)
       if (i->getOperand(ni+1).getMBB() == this)
         i->getOperand(ni+1).setMBB(NMBB);
@@ -642,7 +645,7 @@ MachineBasicBlock::SplitCriticalEdge(MachineBasicBlock *Succ, Pass *P) {
     // Restore kills of virtual registers that were killed by the terminators.
     while (!KilledRegs.empty()) {
       unsigned Reg = KilledRegs.pop_back_val();
-      for (insn_iterator I = insn_end(), E = insn_begin(); I != E;) {
+      for (instr_iterator I = instr_end(), E = instr_begin(); I != E;) {
         if (!(--I)->addRegisterKilled(Reg, NULL, /* addIfNotFound= */ false))
           continue;
         LV->getVarInfo(Reg).Kills.push_back(I);
@@ -711,6 +714,41 @@ MachineBasicBlock::SplitCriticalEdge(MachineBasicBlock *Succ, Pass *P) {
   return NMBB;
 }
 
+MachineBasicBlock::iterator
+MachineBasicBlock::erase(MachineBasicBlock::iterator I) {
+  if (I->isBundle()) {
+    MachineBasicBlock::iterator E = llvm::next(I);
+    return Insts.erase(I.getInstrIterator(), E.getInstrIterator());
+  }
+
+  return Insts.erase(I.getInstrIterator());
+}
+
+MachineInstr *MachineBasicBlock::remove(MachineInstr *I) {
+  if (I->isBundle()) {
+    MachineBasicBlock::instr_iterator MII = I; ++MII;
+    while (MII != end() && MII->isInsideBundle()) {
+      MachineInstr *MI = &*MII++;
+      Insts.remove(MI);
+    }
+  }
+
+  return Insts.remove(I);
+}
+
+void MachineBasicBlock::splice(MachineBasicBlock::iterator where,
+                               MachineBasicBlock *Other,
+                               MachineBasicBlock::iterator From) {
+  if (From->isBundle()) {
+    MachineBasicBlock::iterator To = llvm::next(From);
+    Insts.splice(where.getInstrIterator(), Other->Insts,
+                 From.getInstrIterator(), To.getInstrIterator());
+    return;
+  }
+
+  Insts.splice(where.getInstrIterator(), Other->Insts, From.getInstrIterator());
+}
+
 /// removeFromParent - This method unlinks 'this' from the containing function,
 /// and returns it, but does not delete it.
 MachineBasicBlock *MachineBasicBlock::removeFromParent() {
@@ -734,8 +772,8 @@ void MachineBasicBlock::ReplaceUsesOfBlockWith(MachineBasicBlock *Old,
                                                MachineBasicBlock *New) {
   assert(Old != New && "Cannot replace self with self!");
 
-  MachineBasicBlock::insn_iterator I = insn_end();
+  MachineBasicBlock::instr_iterator I = instr_end();
-  while (I != insn_begin()) {
+  while (I != instr_begin()) {
     --I;
     if (!I->isTerminator()) break;
 
@@ -816,9 +854,9 @@ bool MachineBasicBlock::CorrectExtraCFGEdges(MachineBasicBlock *DestA,
 /// findDebugLoc - find the next valid DebugLoc starting at MBBI, skipping
 /// any DBG_VALUE instructions.  Return UnknownLoc if there is none.
 DebugLoc
-MachineBasicBlock::findDebugLoc(insn_iterator MBBI) {
+MachineBasicBlock::findDebugLoc(instr_iterator MBBI) {
   DebugLoc DL;
-  insn_iterator E = insn_end();
+  instr_iterator E = instr_end();
   if (MBBI == E)
     return DL;
 

lib/CodeGen/MachineInstr.cpp

@@ -750,11 +750,11 @@ void MachineInstr::addMemOperand(MachineFunction &MF,
 
 bool
 MachineInstr::hasProperty(unsigned MCFlag, QueryType Type) const {
-  if (Type == IgnoreBundle || getOpcode() != TargetOpcode::BUNDLE)
+  if (Type == IgnoreBundle || !isBundle())
     return getDesc().getFlags() & (1 << MCFlag);
 
   const MachineBasicBlock *MBB = getParent();
-  MachineBasicBlock::const_insn_iterator MII = *this; ++MII;
+  MachineBasicBlock::const_instr_iterator MII = *this; ++MII;
   while (MII != MBB->end() && MII->isInsideBundle()) {
     if (MII->getDesc().getFlags() & (1 << MCFlag)) {
       if (Type == AnyInBundle)
@@ -777,6 +777,19 @@ bool MachineInstr::isIdenticalTo(const MachineInstr *Other,
       Other->getNumOperands() != getNumOperands())
     return false;
 
+  if (isBundle()) {
+    // Both instructions are bundles, compare MIs inside the bundle.
+    MachineBasicBlock::const_instr_iterator I1 = *this;
+    MachineBasicBlock::const_instr_iterator E1 = getParent()->instr_end();
+    MachineBasicBlock::const_instr_iterator I2 = *Other;
+    MachineBasicBlock::const_instr_iterator E2= Other->getParent()->instr_end();
+    while (++I1 != E1 && I1->isInsideBundle()) {
+      ++I2;
+      if (I2 == E2 || !I2->isInsideBundle() || !I1->isIdenticalTo(I2, Check))
+        return false;
+    }
+  }
+
   // Check operands to make sure they match.
   for (unsigned i = 0, e = getNumOperands(); i != e; ++i) {
     const MachineOperand &MO = getOperand(i);
@@ -825,10 +838,11 @@ MachineInstr *MachineInstr::removeFromParent() {
   assert(getParent() && "Not embedded in a basic block!");
 
   // If it's a bundle then remove the MIs inside the bundle as well.
-  if (getOpcode() == TargetOpcode::BUNDLE) {
+  if (isBundle()) {
     MachineBasicBlock *MBB = getParent();
-    MachineBasicBlock::insn_iterator MII = *this; ++MII;
+    MachineBasicBlock::instr_iterator MII = *this; ++MII;
-    while (MII != MBB->end() && MII->isInsideBundle()) {
+    MachineBasicBlock::instr_iterator E = MBB->instr_end();
+    while (MII != E && MII->isInsideBundle()) {
       MachineInstr *MI = &*MII;
       ++MII;
       MBB->remove(MI);
@@ -844,10 +858,11 @@ MachineInstr *MachineInstr::removeFromParent() {
 void MachineInstr::eraseFromParent() {
   assert(getParent() && "Not embedded in a basic block!");
   // If it's a bundle then remove the MIs inside the bundle as well.
-  if (getOpcode() == TargetOpcode::BUNDLE) {
+  if (isBundle()) {
     MachineBasicBlock *MBB = getParent();
-    MachineBasicBlock::insn_iterator MII = *this; ++MII;
+    MachineBasicBlock::instr_iterator MII = *this; ++MII;
-    while (MII != MBB->end() && MII->isInsideBundle()) {
+    MachineBasicBlock::instr_iterator E = MBB->instr_end();
+    while (MII != E && MII->isInsideBundle()) {
       MachineInstr *MI = &*MII;
       ++MII;
       MBB->erase(MI);
@@ -942,6 +957,20 @@ MachineInstr::getRegClassConstraint(unsigned OpIdx,
   return NULL;
 }
 
+/// getBundleSize - Return the number of instructions inside the MI bundle.
+unsigned MachineInstr::getBundleSize() const {
+  assert(isBundle() && "Expecting a bundle");
+
+  MachineBasicBlock::const_instr_iterator I = *this;
+  unsigned Size = 0;
+  while ((++I)->isInsideBundle()) {
+    ++Size;
+  }
+  assert(Size > 1 && "Malformed bundle");
+
+  return Size;
+}
+
 /// findRegisterUseOperandIdx() - Returns the MachineOperand that is a use of
 /// the specific register or -1 if it is not found. It further tightens
 /// the search criteria to a use that kills the register if isKill is true.
@@ -1024,9 +1053,6 @@ MachineInstr::findRegisterDefOperandIdx(unsigned Reg, bool isDead, bool Overlap,
 /// operand list that is used to represent the predicate. It returns -1 if
 /// none is found.
 int MachineInstr::findFirstPredOperandIdx() const {
-  assert(getOpcode() != TargetOpcode::BUNDLE &&
-         "MachineInstr::findFirstPredOperandIdx() can't handle bundles");
-
   // Don't call MCID.findFirstPredOperandIdx() because this variant
   // is sometimes called on an instruction that's not yet complete, and
   // so the number of operands is less than the MCID indicates. In
@@ -1176,8 +1202,7 @@ void MachineInstr::copyKillDeadInfo(const MachineInstr *MI) {
 
 /// copyPredicates - Copies predicate operand(s) from MI.
 void MachineInstr::copyPredicates(const MachineInstr *MI) {
-  assert(getOpcode() != TargetOpcode::BUNDLE &&
+  assert(!isBundle() && "MachineInstr::copyPredicates() can't handle bundles");
-         "MachineInstr::copyPredicates() can't handle bundles");
 
   const MCInstrDesc &MCID = MI->getDesc();
   if (!MCID.isPredicable())

lib/CodeGen/MachineInstrBundle.cpp

@@ -0,0 +1,180 @@
+//===-- lib/CodeGen/MachineInstrBundle.cpp --------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/CodeGen/MachineInstrBundle.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/Passes.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetRegisterInfo.h"
+#include "llvm/ADT/SmallSet.h"
+#include "llvm/ADT/SmallVector.h"
+using namespace llvm;
+
+namespace {
+  class UnpackMachineBundles : public MachineFunctionPass {
+  public:
+    static char ID; // Pass identification
+    UnpackMachineBundles() : MachineFunctionPass(ID) {
+      initializeUnpackMachineBundlesPass(*PassRegistry::getPassRegistry());
+    }
+
+    virtual bool runOnMachineFunction(MachineFunction &MF);
+  };
+} // end anonymous namespace
+
+char UnpackMachineBundles::ID = 0;
+INITIALIZE_PASS(UnpackMachineBundles, "unpack-mi-bundle",
+                "Unpack machine instruction bundles", false, false)
+
+FunctionPass *llvm::createUnpackMachineBundlesPass() {
+  return new UnpackMachineBundles();
+}
+
+bool UnpackMachineBundles::runOnMachineFunction(MachineFunction &MF) {
+  bool Changed = false;
+  for (MachineFunction::iterator I = MF.begin(), E = MF.end(); I != E; ++I) {
+    MachineBasicBlock *MBB = &*I;
+
+    for (MachineBasicBlock::instr_iterator MII = MBB->instr_begin(),
+           MIE = MBB->instr_end(); MII != MIE; ) {
+      MachineInstr *MI = &*MII;
+
+      // Remove BUNDLE instruction and the InsideBundle flags from bundled
+      // instructions.
+      if (MI->isBundle()) {
+        while (++MII != MIE && MII->isInsideBundle()) {
+          MII->setIsInsideBundle(false);
+          for (unsigned i = 0, e = MII->getNumOperands(); i != e; ++i) {
+            MachineOperand &MO = MII->getOperand(i);
+            if (MO.isReg() && MO.isInternalRead())
+              MO.setIsInternalRead(false);
+          }
+        }
+        MI->eraseFromParent();
+
+        Changed = true;
+        continue;
+      }
+
+      ++MII;
+    }
+  }
+
+  return Changed;
+}
+
+/// FinalizeBundle - Finalize a machine instruction bundle which includes
+/// a sequence of instructions starting from FirstMI to LastMI (inclusive).
+/// This routine adds a BUNDLE instruction to represent the bundle, it adds
+/// IsInternalRead markers to MachineOperands which are defined inside the
+/// bundle, and it copies externally visible defs and uses to the BUNDLE
+/// instruction.
+void llvm::FinalizeBundle(MachineBasicBlock &MBB,
+                          MachineBasicBlock::instr_iterator FirstMI,
+                          MachineBasicBlock::instr_iterator LastMI) {
+  const TargetMachine &TM = MBB.getParent()->getTarget();
+  const TargetInstrInfo *TII = TM.getInstrInfo();
+  const TargetRegisterInfo *TRI = TM.getRegisterInfo();
+
+  MachineInstrBuilder MIB = BuildMI(MBB, FirstMI, FirstMI->getDebugLoc(),
+                                    TII->get(TargetOpcode::BUNDLE));
+
+  SmallVector<unsigned, 8> LocalDefs;
+  SmallSet<unsigned, 8> LocalDefSet;
+  SmallSet<unsigned, 8> DeadDefSet;
+  SmallSet<unsigned, 8> KilledDefSet;
+  SmallVector<unsigned, 8> ExternUses;
+  SmallSet<unsigned, 8> ExternUseSet;
+  SmallSet<unsigned, 8> KilledUseSet;
+  SmallSet<unsigned, 8> UndefUseSet;
+  SmallVector<MachineOperand*, 4> Defs;
+  do {
+    for (unsigned i = 0, e = FirstMI->getNumOperands(); i != e; ++i) {
+      MachineOperand &MO = FirstMI->getOperand(i);
+      if (!MO.isReg())
+        continue;
+      if (MO.isDef()) {
+        Defs.push_back(&MO);
+        continue;
+      }
+
+      unsigned Reg = MO.getReg();
+      if (!Reg)
+        continue;
+      assert(TargetRegisterInfo::isPhysicalRegister(Reg));
+      if (LocalDefSet.count(Reg)) {
+        MO.setIsInternalRead();
+        if (MO.isKill())
+          // Internal def is now killed.
+          KilledDefSet.insert(Reg);
+      } else {
+        if (ExternUseSet.insert(Reg)) {
+          ExternUses.push_back(Reg);
+          if (MO.isUndef())
+            UndefUseSet.insert(Reg);
+        }
+        if (MO.isKill())
+          // External def is now killed.
+          KilledUseSet.insert(Reg);
+      }
+    }
+
+    for (unsigned i = 0, e = Defs.size(); i != e; ++i) {
+      MachineOperand &MO = *Defs[i];
+      unsigned Reg = MO.getReg();
+      if (!Reg)
+        continue;
+
+      if (LocalDefSet.insert(Reg)) {
+        LocalDefs.push_back(Reg);
+        if (MO.isDead()) {
+          DeadDefSet.insert(Reg);
+        }
+      } else {
+        // Re-defined inside the bundle, it's no longer killed.
+        KilledDefSet.erase(Reg);
+        if (!MO.isDead())
+          // Previously defined but dead.
+          DeadDefSet.erase(Reg);
+      }
+
+      if (!MO.isDead()) {
+        for (const unsigned *SubRegs = TRI->getSubRegisters(Reg);
+             unsigned SubReg = *SubRegs; ++SubRegs) {
+          if (LocalDefSet.insert(SubReg))
+            LocalDefs.push_back(SubReg);
+        }
+      }
+    }
+
+    FirstMI->setIsInsideBundle();
+    Defs.clear();
+  } while (FirstMI++ != LastMI);
+
+  SmallSet<unsigned, 8> Added;
+  for (unsigned i = 0, e = LocalDefs.size(); i != e; ++i) {
+    unsigned Reg = LocalDefs[i];
+    if (Added.insert(Reg)) {
+      // If it's not live beyond end of the bundle, mark it dead.
+      bool isDead = DeadDefSet.count(Reg) || KilledDefSet.count(Reg);
+      MIB.addReg(Reg, getDefRegState(true) | getDeadRegState(isDead) |
+                 getImplRegState(true));
+    }
+  }
+
+  for (unsigned i = 0, e = ExternUses.size(); i != e; ++i) {
+    unsigned Reg = ExternUses[i];
+    bool isKill = KilledUseSet.count(Reg);
+    bool isUndef = UndefUseSet.count(Reg);
+    MIB.addReg(Reg, getKillRegState(isKill) | getUndefRegState(isUndef) |
+               getImplRegState(true));
+  }
+}

lib/CodeGen/MachineVerifier.cpp

@@ -279,13 +279,17 @@ bool MachineVerifier::runOnMachineFunction(MachineFunction &MF) {
   for (MachineFunction::const_iterator MFI = MF.begin(), MFE = MF.end();
        MFI!=MFE; ++MFI) {
     visitMachineBasicBlockBefore(MFI);
-    for (MachineBasicBlock::const_iterator MBBI = MFI->begin(),
+    for (MachineBasicBlock::const_instr_iterator MBBI = MFI->instr_begin(),
-           MBBE = MFI->end(); MBBI != MBBE; ++MBBI) {
+           MBBE = MFI->instr_end(); MBBI != MBBE; ++MBBI) {
       if (MBBI->getParent() != MFI) {
         report("Bad instruction parent pointer", MFI);
         *OS << "Instruction: " << *MBBI;
         continue;
       }
+      // Skip BUNDLE instruction for now. FIXME: We should add code to verify
+      // the BUNDLE's specifically.
+      if (MBBI->isBundle())
+        continue;
       visitMachineInstrBefore(MBBI);
       for (unsigned I = 0, E = MBBI->getNumOperands(); I != E; ++I)
         visitMachineOperand(&MBBI->getOperand(I), I);

lib/CodeGen/PostRASchedulerList.cpp

@@ -212,7 +212,8 @@ bool PostRAScheduler::runOnMachineFunction(MachineFunction &Fn) {
   RegClassInfo.runOnMachineFunction(Fn);
 
   // Check for explicit enable/disable of post-ra scheduling.
-  TargetSubtargetInfo::AntiDepBreakMode AntiDepMode = TargetSubtargetInfo::ANTIDEP_NONE;
+  TargetSubtargetInfo::AntiDepBreakMode AntiDepMode =
+    TargetSubtargetInfo::ANTIDEP_NONE;
   SmallVector<TargetRegisterClass*, 4> CriticalPathRCs;
   if (EnablePostRAScheduler.getPosition() > 0) {
     if (!EnablePostRAScheduler)
@@ -271,6 +272,8 @@ bool PostRAScheduler::runOnMachineFunction(MachineFunction &Fn) {
       }
       I = MI;
       --Count;
+      if (MI->isBundle())
+        Count -= MI->getBundleSize();
     }
     assert(Count == 0 && "Instruction count mismatch!");
     assert((MBB->begin() == Current || CurrentCount != 0) &&

lib/CodeGen/ScheduleDAGInstrs.cpp

@@ -277,8 +277,9 @@ void ScheduleDAGInstrs::BuildSchedGraph(AliasAnalysis *AA) {
           continue;
         if (DefSU != SU &&
             (Kind != SDep::Output || !MO.isDead() ||
-             !DefSU->getInstr()->registerDefIsDead(Reg)))
+             !DefSU->getInstr()->registerDefIsDead(Reg))) {
           DefSU->addPred(SDep(SU, Kind, AOLatency, /*Reg=*/Reg));
+        }
       }
       for (const unsigned *Alias = TRI->getAliasSet(Reg); *Alias; ++Alias) {
         std::vector<SUnit *> &MemDefList = Defs[*Alias];
@@ -657,22 +658,16 @@ std::string ScheduleDAGInstrs::getGraphNodeLabel(const SUnit *SU) const {
 
 // EmitSchedule - Emit the machine code in scheduled order.
 MachineBasicBlock *ScheduleDAGInstrs::EmitSchedule() {
-  // For MachineInstr-based scheduling, we're rescheduling the instructions in
+  Begin = InsertPos;
-  // the block, so start by removing them from the block.
-  while (Begin != InsertPos) {
-    MachineBasicBlock::iterator I = Begin;
-    ++Begin;
-    BB->remove(I);
-  }
 
   // If first instruction was a DBG_VALUE then put it back.
   if (FirstDbgValue)
-    BB->insert(InsertPos, FirstDbgValue);
+    BB->splice(InsertPos, BB, FirstDbgValue);
 
   // Then re-insert them according to the given schedule.
   for (unsigned i = 0, e = Sequence.size(); i != e; i++) {
     if (SUnit *SU = Sequence[i])
-      BB->insert(InsertPos, SU->getInstr());
+      BB->splice(InsertPos, BB, SU->getInstr());
     else
       // Null SUnit* is a noop.
       EmitNoop();
@@ -689,7 +684,7 @@ MachineBasicBlock *ScheduleDAGInstrs::EmitSchedule() {
     std::pair<MachineInstr *, MachineInstr *> P = *prior(DI);
     MachineInstr *DbgValue = P.first;
     MachineBasicBlock::iterator OrigPrivMI = P.second;
-    BB->insertAfter(OrigPrivMI, DbgValue);
+    BB->splice(++OrigPrivMI, BB, DbgValue);
   }
   DbgValues.clear();
   FirstDbgValue = NULL;

lib/CodeGen/TargetInstrInfoImpl.cpp

@@ -121,7 +121,7 @@ MachineInstr *TargetInstrInfoImpl::commuteInstruction(MachineInstr *MI,
 bool TargetInstrInfoImpl::findCommutedOpIndices(MachineInstr *MI,
                                                 unsigned &SrcOpIdx1,
                                                 unsigned &SrcOpIdx2) const {
-  assert(MI->getOpcode() != TargetOpcode::BUNDLE &&
+  assert(!MI->isBundle() &&
          "TargetInstrInfoImpl::findCommutedOpIndices() can't handle bundles");
 
   const MCInstrDesc &MCID = MI->getDesc();
@@ -156,7 +156,7 @@ bool TargetInstrInfoImpl::PredicateInstruction(MachineInstr *MI,
                             const SmallVectorImpl<MachineOperand> &Pred) const {
   bool MadeChange = false;
 
-  assert(MI->getOpcode() != TargetOpcode::BUNDLE &&
+  assert(!MI->isBundle() &&
          "TargetInstrInfoImpl::PredicateInstruction() can't handle bundles");
 
   const MCInstrDesc &MCID = MI->getDesc();

lib/Target/ARM/ARMBaseInstrInfo.cpp

@@ -439,6 +439,22 @@ ReverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond) const {
   return false;
 }
 
+bool ARMBaseInstrInfo::isPredicated(const MachineInstr *MI) const {
+  if (MI->isBundle()) {
+    MachineBasicBlock::const_instr_iterator I = MI;
+    MachineBasicBlock::const_instr_iterator E = MI->getParent()->instr_end();
+    while (++I != E && I->isInsideBundle()) {
+      int PIdx = I->findFirstPredOperandIdx();
+      if (PIdx != -1 && I->getOperand(PIdx).getImm() != ARMCC::AL)
+        return true;
+    }
+    return false;
+  }
+
+  int PIdx = MI->findFirstPredOperandIdx();
+  return PIdx != -1 && MI->getOperand(PIdx).getImm() != ARMCC::AL;
+}
+
 bool ARMBaseInstrInfo::
 PredicateInstruction(MachineInstr *MI,
                      const SmallVectorImpl<MachineOperand> &Pred) const {
@@ -547,7 +563,7 @@ unsigned ARMBaseInstrInfo::GetInstSizeInBytes(const MachineInstr *MI) const {
       return getInlineAsmLength(MI->getOperand(0).getSymbolName(), *MAI);
     if (MI->isLabel())
       return 0;
-  unsigned Opc = MI->getOpcode();
+    unsigned Opc = MI->getOpcode();
     switch (Opc) {
     case TargetOpcode::IMPLICIT_DEF:
     case TargetOpcode::KILL:
@@ -555,6 +571,8 @@ unsigned ARMBaseInstrInfo::GetInstSizeInBytes(const MachineInstr *MI) const {
     case TargetOpcode::EH_LABEL:
     case TargetOpcode::DBG_VALUE:
       return 0;
+    case TargetOpcode::BUNDLE:
+      return getInstBundleLength(MI);
     case ARM::MOVi16_ga_pcrel:
     case ARM::MOVTi16_ga_pcrel:
     case ARM::t2MOVi16_ga_pcrel:
@@ -621,6 +639,17 @@ unsigned ARMBaseInstrInfo::GetInstSizeInBytes(const MachineInstr *MI) const {
   return 0; // Not reached
 }
 
+unsigned ARMBaseInstrInfo::getInstBundleLength(const MachineInstr *MI) const {
+  unsigned Size = 0;
+  MachineBasicBlock::const_instr_iterator I = MI;
+  MachineBasicBlock::const_instr_iterator E = MI->getParent()->instr_end();
+  while (++I != E && I->isInsideBundle()) {
+    assert(!I->isBundle() && "No nested bundle!");
+    Size += GetInstSizeInBytes(&*I);
+  }
+  return Size;
+}
+
 void ARMBaseInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
                                    MachineBasicBlock::iterator I, DebugLoc DL,
                                    unsigned DestReg, unsigned SrcReg,
@@ -1955,7 +1984,7 @@ bool ARMBaseInstrInfo::FoldImmediate(MachineInstr *UseMI,
   bool isKill = UseMI->getOperand(OpIdx).isKill();
   unsigned NewReg = MRI->createVirtualRegister(MRI->getRegClass(Reg));
   AddDefaultCC(AddDefaultPred(BuildMI(*UseMI->getParent(),
-                                      *UseMI, UseMI->getDebugLoc(),
+                                      UseMI, UseMI->getDebugLoc(),
                                       get(NewUseOpc), NewReg)
                               .addReg(Reg1, getKillRegState(isKill))
                               .addImm(SOImmValV1)));
@@ -2330,6 +2359,57 @@ ARMBaseInstrInfo::getOperandLatency(const InstrItineraryData *ItinData,
   return UseCycle;
 }
 
+static const MachineInstr *getBundledDefMI(const TargetRegisterInfo *TRI,
+                                           const MachineInstr *MI,
+                                           unsigned &DefIdx, unsigned &Dist) {
+  Dist = 0;
+
+  MachineBasicBlock::const_iterator I = MI; ++I;
+  MachineBasicBlock::const_instr_iterator II =
+    llvm::prior(I.getInstrIterator());
+  assert(II->isInsideBundle() && "Empty bundle?");
+
+  int Idx = -1;
+  unsigned Reg = MI->getOperand(DefIdx).getReg();
+  while (II->isInsideBundle()) {
+    Idx = II->findRegisterDefOperandIdx(Reg, false, true, TRI);
+    if (Idx != -1)
+      break;
+    --II;
+    ++Dist;
+  }
+
+  assert(Idx != -1 && "Cannot find bundled definition!");
+  DefIdx = Idx;
+  return II;
+}
+
+static const MachineInstr *getBundledUseMI(const TargetRegisterInfo *TRI,
+                                           const MachineInstr *MI,
+                                           unsigned &UseIdx, unsigned &Dist) {
+  Dist = 0;
+
+  MachineBasicBlock::const_instr_iterator II = MI; ++II;
+  assert(II->isInsideBundle() && "Empty bundle?");
+  MachineBasicBlock::const_instr_iterator E = MI->getParent()->instr_end();
+
+  // FIXME: This doesn't properly handle multiple uses.
+  int Idx = -1;
+  unsigned Reg = MI->getOperand(UseIdx).getReg();
+  while (II != E && II->isInsideBundle()) {
+    Idx = II->findRegisterUseOperandIdx(Reg, false, TRI);
+    if (Idx != -1)
+      break;
+    if (II->getOpcode() != ARM::t2IT)
+      ++Dist;
+    ++II;
+  }
+
+  assert(Idx != -1 && "Cannot find bundled definition!");
+  UseIdx = Idx;
+  return II;
+}
+
 int
 ARMBaseInstrInfo::getOperandLatency(const InstrItineraryData *ItinData,
                              const MachineInstr *DefMI, unsigned DefIdx,
@@ -2341,8 +2421,8 @@ ARMBaseInstrInfo::getOperandLatency(const InstrItineraryData *ItinData,
   if (!ItinData || ItinData->isEmpty())
     return DefMI->mayLoad() ? 3 : 1;
 
-  const MCInstrDesc &DefMCID = DefMI->getDesc();
+  const MCInstrDesc *DefMCID = &DefMI->getDesc();
-  const MCInstrDesc &UseMCID = UseMI->getDesc();
+  const MCInstrDesc *UseMCID = &UseMI->getDesc();
   const MachineOperand &DefMO = DefMI->getOperand(DefIdx);
   if (DefMO.getReg() == ARM::CPSR) {
     if (DefMI->getOpcode() == ARM::FMSTAT) {
@@ -2353,20 +2433,50 @@ ARMBaseInstrInfo::getOperandLatency(const InstrItineraryData *ItinData,
     // CPSR set and branch can be paired in the same cycle.
     if (UseMI->isBranch())
       return 0;
+
+    // Otherwise it takes the instruction latency (generally one).
+    int Latency = getInstrLatency(ItinData, DefMI);
+    // For Thumb2, prefer scheduling CPSR setting instruction close to its uses.
+    // Instructions which are otherwise scheduled between them may incur a code
+    // size penalty (not able to use the CPSR setting 16-bit instructions).
+    if (Latency > 0 && Subtarget.isThumb2())
+      --Latency;
+    return Latency;
   }
 
   unsigned DefAlign = DefMI->hasOneMemOperand()
     ? (*DefMI->memoperands_begin())->getAlignment() : 0;
   unsigned UseAlign = UseMI->hasOneMemOperand()
     ? (*UseMI->memoperands_begin())->getAlignment() : 0;
-  int Latency = getOperandLatency(ItinData, DefMCID, DefIdx, DefAlign,
+
-                                  UseMCID, UseIdx, UseAlign);
+  unsigned DefAdj = 0;
+  if (DefMI->isBundle()) {
+    DefMI = getBundledDefMI(&getRegisterInfo(), DefMI, DefIdx, DefAdj);
+    if (DefMI->isCopyLike() || DefMI->isInsertSubreg() ||
+        DefMI->isRegSequence() || DefMI->isImplicitDef())
+      return 1;
+    DefMCID = &DefMI->getDesc();
+  }
+  unsigned UseAdj = 0;
+  if (UseMI->isBundle()) {
+    UseMI = getBundledUseMI(&getRegisterInfo(), UseMI, UseIdx, UseAdj);
+    UseMCID = &UseMI->getDesc();
+  }
+
+  int Latency = getOperandLatency(ItinData, *DefMCID, DefIdx, DefAlign,
+                                  *UseMCID, UseIdx, UseAlign);
+  int Adj = DefAdj + UseAdj;
+  if (Adj) {
+    Latency -= (int)(DefAdj + UseAdj);
+    if (Latency < 1)
+      return 1;
+  }
 
   if (Latency > 1 &&
       (Subtarget.isCortexA8() || Subtarget.isCortexA9())) {
     // FIXME: Shifter op hack: no shift (i.e. [r +/- r]) or [r + r << 2]
     // variants are one cycle cheaper.
-    switch (DefMCID.getOpcode()) {
+    switch (DefMCID->getOpcode()) {
     default: break;
     case ARM::LDRrs:
     case ARM::LDRBrs: {
@@ -2391,7 +2501,7 @@ ARMBaseInstrInfo::getOperandLatency(const InstrItineraryData *ItinData,
   }
 
   if (DefAlign < 8 && Subtarget.isCortexA9())
-    switch (DefMCID.getOpcode()) {
+    switch (DefMCID->getOpcode()) {
     default: break;
     case ARM::VLD1q8:
     case ARM::VLD1q16:
@@ -2697,6 +2807,17 @@ int ARMBaseInstrInfo::getInstrLatency(const InstrItineraryData *ItinData,
   if (!ItinData || ItinData->isEmpty())
     return 1;
 
+  if (MI->isBundle()) {
+    int Latency = 0;
+    MachineBasicBlock::const_instr_iterator I = MI;
+    MachineBasicBlock::const_instr_iterator E = MI->getParent()->instr_end();
+    while (++I != E && I->isInsideBundle()) {
+      if (I->getOpcode() != ARM::t2IT)
+        Latency += getInstrLatency(ItinData, I, PredCost);
+    }
+    return Latency;
+  }
+
   const MCInstrDesc &MCID = MI->getDesc();
   unsigned Class = MCID.getSchedClass();
   unsigned UOps = ItinData->Itineraries[Class].NumMicroOps;

lib/Target/ARM/ARMBaseInstrInfo.h

@@ -69,10 +69,7 @@ public:
   bool ReverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond) const;
 
   // Predication support.
-  bool isPredicated(const MachineInstr *MI) const {
+  bool isPredicated(const MachineInstr *MI) const;
-    int PIdx = MI->findFirstPredOperandIdx();
-    return PIdx != -1 && MI->getOperand(PIdx).getImm() != ARMCC::AL;
-  }
 
   ARMCC::CondCodes getPredicate(const MachineInstr *MI) const {
     int PIdx = MI->findFirstPredOperandIdx();
@@ -219,6 +216,8 @@ public:
   void setExecutionDomain(MachineInstr *MI, unsigned Domain) const;
 
 private:
+  unsigned getInstBundleLength(const MachineInstr *MI) const;
+
   int getVLDMDefCycle(const InstrItineraryData *ItinData,
                       const MCInstrDesc &DefMCID,
                       unsigned DefClass,

lib/Target/ARM/ARMFastISel.cpp

@@ -702,7 +702,7 @@ unsigned ARMFastISel::TargetMaterializeAlloca(const AllocaInst *AI) {
     TargetRegisterClass* RC = TLI.getRegClassFor(VT);
     unsigned ResultReg = createResultReg(RC);
     unsigned Opc = isThumb2 ? ARM::t2ADDri : ARM::ADDri;
-    AddOptionalDefs(BuildMI(*FuncInfo.MBB, *FuncInfo.InsertPt, DL,
+    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
                             TII.get(Opc), ResultReg)
                             .addFrameIndex(SI->second)
                             .addImm(0));
@@ -898,7 +898,7 @@ void ARMFastISel::ARMSimplifyAddress(Address &Addr, EVT VT, bool useAM3) {
                               ARM::GPRRegisterClass;
     unsigned ResultReg = createResultReg(RC);
     unsigned Opc = isThumb2 ? ARM::t2ADDri : ARM::ADDri;
-    AddOptionalDefs(BuildMI(*FuncInfo.MBB, *FuncInfo.InsertPt, DL,
+    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
                             TII.get(Opc), ResultReg)
                             .addFrameIndex(Addr.Base.FI)
                             .addImm(0));

lib/Target/ARM/ARMHazardRecognizer.cpp

@@ -38,9 +38,6 @@ ARMHazardRecognizer::getHazardType(SUnit *SU, int Stalls) {
   MachineInstr *MI = SU->getInstr();
 
   if (!MI->isDebugValue()) {
-    if (ITBlockSize && MI != ITBlockMIs[ITBlockSize-1])
-      return Hazard;
-
     // Look for special VMLA / VMLS hazards. A VMUL / VADD / VSUB following
     // a VMLA / VMLS will cause 4 cycle stall.
     const MCInstrDesc &MCID = MI->getDesc();
@@ -76,30 +73,11 @@ ARMHazardRecognizer::getHazardType(SUnit *SU, int Stalls) {
 void ARMHazardRecognizer::Reset() {
   LastMI = 0;
   FpMLxStalls = 0;
-  ITBlockSize = 0;
   ScoreboardHazardRecognizer::Reset();
 }
 
 void ARMHazardRecognizer::EmitInstruction(SUnit *SU) {
   MachineInstr *MI = SU->getInstr();
-  unsigned Opcode = MI->getOpcode();
-  if (ITBlockSize) {
-    --ITBlockSize;
-  } else if (Opcode == ARM::t2IT) {
-    unsigned Mask = MI->getOperand(1).getImm();
-    unsigned NumTZ = CountTrailingZeros_32(Mask);
-    assert(NumTZ <= 3 && "Invalid IT mask!");
-    ITBlockSize = 4 - NumTZ;
-    MachineBasicBlock::iterator I = MI;
-    for (unsigned i = 0; i < ITBlockSize; ++i) {
-      // Advance to the next instruction, skipping any dbg_value instructions.
-      do {
-        ++I;
-      } while (I->isDebugValue());
-      ITBlockMIs[ITBlockSize-1-i] = &*I;
-    }
-  }
-
   if (!MI->isDebugValue()) {
     LastMI = MI;
     FpMLxStalls = 0;

lib/Target/ARM/ARMHazardRecognizer.h

@@ -34,8 +34,6 @@ class ARMHazardRecognizer : public ScoreboardHazardRecognizer {
 
   MachineInstr *LastMI;
   unsigned FpMLxStalls;
-  unsigned ITBlockSize;  // No. of MIs in current IT block yet to be scheduled.
-  MachineInstr *ITBlockMIs[4];
 
 public:
   ARMHazardRecognizer(const InstrItineraryData *ItinData,
@@ -44,7 +42,7 @@ public:
                       const ARMSubtarget &sti,
                       const ScheduleDAG *DAG) :
     ScoreboardHazardRecognizer(ItinData, DAG, "post-RA-sched"), TII(tii),
-    TRI(tri), STI(sti), LastMI(0), ITBlockSize(0) {}
+    TRI(tri), STI(sti), LastMI(0) {}
 
   virtual HazardType getHazardType(SUnit *SU, int Stalls);
   virtual void Reset();

lib/Target/ARM/ARMTargetMachine.cpp

@@ -147,10 +147,16 @@ bool ARMBaseTargetMachine::addPreSched2(PassManagerBase &PM) {
 }
 
 bool ARMBaseTargetMachine::addPreEmitPass(PassManagerBase &PM) {
-  if (Subtarget.isThumb2() && !Subtarget.prefers32BitThumb())
+  if (Subtarget.isThumb2()) {
-    PM.add(createThumb2SizeReductionPass());
+    if (!Subtarget.prefers32BitThumb())
+      PM.add(createThumb2SizeReductionPass());
+
+    // Constant island pass work on unbundled instructions.
+    PM.add(createUnpackMachineBundlesPass());
+  }
 
   PM.add(createARMConstantIslandPass());
+
   return true;
 }
 

lib/Target/ARM/MLxExpansionPass.cpp

@@ -222,14 +222,14 @@ MLxExpansion::ExpandFPMLxInstruction(MachineBasicBlock &MBB, MachineInstr *MI,
   const MCInstrDesc &MCID2 = TII->get(AddSubOpc);
   unsigned TmpReg = MRI->createVirtualRegister(TII->getRegClass(MCID1, 0, TRI));
 
-  MachineInstrBuilder MIB = BuildMI(MBB, *MI, MI->getDebugLoc(), MCID1, TmpReg)
+  MachineInstrBuilder MIB = BuildMI(MBB, MI, MI->getDebugLoc(), MCID1, TmpReg)
     .addReg(Src1Reg, getKillRegState(Src1Kill))
     .addReg(Src2Reg, getKillRegState(Src2Kill));
   if (HasLane)
     MIB.addImm(LaneImm);
   MIB.addImm(Pred).addReg(PredReg);
 
-  MIB = BuildMI(MBB, *MI, MI->getDebugLoc(), MCID2)
+  MIB = BuildMI(MBB, MI, MI->getDebugLoc(), MCID2)
     .addReg(DstReg, getDefRegState(true) | getDeadRegState(DstDead));
 
   if (NegAcc) {

lib/Target/ARM/Thumb2ITBlockPass.cpp

@@ -13,6 +13,7 @@
 #include "Thumb2InstrInfo.h"
 #include "llvm/CodeGen/MachineInstr.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineInstrBundle.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/ADT/SmallSet.h"
 #include "llvm/ADT/Statistic.h"
@@ -237,6 +238,9 @@ bool Thumb2ITBlockPass::InsertITInstructions(MachineBasicBlock &MBB) {
     // Last instruction in IT block kills ITSTATE.
     LastITMI->findRegisterUseOperand(ARM::ITSTATE)->setIsKill();
 
+    // Finalize the bundle.
+    FinalizeBundle(MBB, InsertPos.getInstrIterator(), LastITMI);
+
     Modified = true;
     ++NumITs;
   }

lib/Target/ARM/Thumb2SizeReduction.cpp

@@ -452,7 +452,7 @@ Thumb2SizeReduce::ReduceLoadStore(MachineBasicBlock &MBB, MachineInstr *MI,
 
   // Add the 16-bit load / store instruction.
   DebugLoc dl = MI->getDebugLoc();
-  MachineInstrBuilder MIB = BuildMI(MBB, *MI, dl, TII->get(Opc));
+  MachineInstrBuilder MIB = BuildMI(MBB, MI, dl, TII->get(Opc));
   if (!isLdStMul) {
     MIB.addOperand(MI->getOperand(0));
     MIB.addOperand(MI->getOperand(1));
@@ -478,7 +478,7 @@ Thumb2SizeReduce::ReduceLoadStore(MachineBasicBlock &MBB, MachineInstr *MI,
 
   DEBUG(errs() << "Converted 32-bit: " << *MI << "       to 16-bit: " << *MIB);
 
-  MBB.erase(MI);
+  MBB.erase_instr(MI);
   ++NumLdSts;
   return true;
 }
@@ -513,7 +513,7 @@ Thumb2SizeReduce::ReduceSpecial(MachineBasicBlock &MBB, MachineInstr *MI,
         MI->getOperand(MCID.getNumOperands()-1).getReg() == ARM::CPSR)
       return false;
 
-    MachineInstrBuilder MIB = BuildMI(MBB, *MI, MI->getDebugLoc(),
+    MachineInstrBuilder MIB = BuildMI(MBB, MI, MI->getDebugLoc(),
                                       TII->get(ARM::tADDrSPi))
       .addOperand(MI->getOperand(0))
       .addOperand(MI->getOperand(1))
@@ -525,7 +525,7 @@ Thumb2SizeReduce::ReduceSpecial(MachineBasicBlock &MBB, MachineInstr *MI,
 
     DEBUG(errs() << "Converted 32-bit: " << *MI << "       to 16-bit: " <<*MIB);
 
-    MBB.erase(MI);
+    MBB.erase_instr(MI);
     ++NumNarrows;
     return true;
   }
@@ -653,7 +653,7 @@ Thumb2SizeReduce::ReduceTo2Addr(MachineBasicBlock &MBB, MachineInstr *MI,
 
   // Add the 16-bit instruction.
   DebugLoc dl = MI->getDebugLoc();
-  MachineInstrBuilder MIB = BuildMI(MBB, *MI, dl, NewMCID);
+  MachineInstrBuilder MIB = BuildMI(MBB, MI, dl, NewMCID);
   MIB.addOperand(MI->getOperand(0));
   if (NewMCID.hasOptionalDef()) {
     if (HasCC)
@@ -677,7 +677,7 @@ Thumb2SizeReduce::ReduceTo2Addr(MachineBasicBlock &MBB, MachineInstr *MI,
 
   DEBUG(errs() << "Converted 32-bit: " << *MI << "       to 16-bit: " << *MIB);
 
-  MBB.erase(MI);
+  MBB.erase_instr(MI);
   ++Num2Addrs;
   return true;
 }
@@ -744,7 +744,7 @@ Thumb2SizeReduce::ReduceToNarrow(MachineBasicBlock &MBB, MachineInstr *MI,
 
   // Add the 16-bit instruction.
   DebugLoc dl = MI->getDebugLoc();
-  MachineInstrBuilder MIB = BuildMI(MBB, *MI, dl, NewMCID);
+  MachineInstrBuilder MIB = BuildMI(MBB, MI, dl, NewMCID);
   MIB.addOperand(MI->getOperand(0));
   if (NewMCID.hasOptionalDef()) {
     if (HasCC)
@@ -784,7 +784,7 @@ Thumb2SizeReduce::ReduceToNarrow(MachineBasicBlock &MBB, MachineInstr *MI,
 
   DEBUG(errs() << "Converted 32-bit: " << *MI << "       to 16-bit: " << *MIB);
 
-  MBB.erase(MI);
+  MBB.erase_instr(MI);
   ++NumNarrows;
   return true;
 }
@@ -829,16 +829,22 @@ bool Thumb2SizeReduce::ReduceMBB(MachineBasicBlock &MBB) {
   // Yes, CPSR could be livein.
   bool LiveCPSR = MBB.isLiveIn(ARM::CPSR);
   MachineInstr *CPSRDef = 0;
+  MachineInstr *BundleMI = 0;
 
   // If this BB loops back to itself, conservatively avoid narrowing the
   // first instruction that does partial flag update.
   bool IsSelfLoop = MBB.isSuccessor(&MBB);
-  MachineBasicBlock::iterator MII = MBB.begin(), E = MBB.end();
+  MachineBasicBlock::instr_iterator MII = MBB.instr_begin(), E = MBB.instr_end();
-  MachineBasicBlock::iterator NextMII;
+  MachineBasicBlock::instr_iterator NextMII;
   for (; MII != E; MII = NextMII) {
     NextMII = llvm::next(MII);
 
     MachineInstr *MI = &*MII;
+    if (MI->isBundle()) {
+      BundleMI = MI;
+      continue;
+    }
+
     LiveCPSR = UpdateCPSRUse(*MI, LiveCPSR);
 
     unsigned Opcode = MI->getOpcode();
@@ -849,7 +855,7 @@ bool Thumb2SizeReduce::ReduceMBB(MachineBasicBlock &MBB) {
       if (Entry.Special) {
         if (ReduceSpecial(MBB, MI, Entry, LiveCPSR, CPSRDef, IsSelfLoop)) {
           Modified = true;
-          MachineBasicBlock::iterator I = prior(NextMII);
+          MachineBasicBlock::instr_iterator I = prior(NextMII);
           MI = &*I;
         }
         goto ProcessNext;
@@ -859,7 +865,7 @@ bool Thumb2SizeReduce::ReduceMBB(MachineBasicBlock &MBB) {
       if (Entry.NarrowOpc2 &&
           ReduceTo2Addr(MBB, MI, Entry, LiveCPSR, CPSRDef, IsSelfLoop)) {
         Modified = true;
-        MachineBasicBlock::iterator I = prior(NextMII);
+        MachineBasicBlock::instr_iterator I = prior(NextMII);
         MI = &*I;
         goto ProcessNext;
       }
@@ -868,12 +874,21 @@ bool Thumb2SizeReduce::ReduceMBB(MachineBasicBlock &MBB) {
       if (Entry.NarrowOpc1 &&
           ReduceToNarrow(MBB, MI, Entry, LiveCPSR, CPSRDef, IsSelfLoop)) {
         Modified = true;
-        MachineBasicBlock::iterator I = prior(NextMII);
+        MachineBasicBlock::instr_iterator I = prior(NextMII);
         MI = &*I;
       }
     }
 
   ProcessNext:
+    if (LiveCPSR &&
+        NextMII != E && MI->isInsideBundle() && !NextMII->isInsideBundle() &&
+        BundleMI->killsRegister(ARM::CPSR))
+      // FIXME: Since post-ra scheduler operates on bundles, the CPSR kill
+      // marker is only on the BUNDLE instruction. Process the BUNDLE
+      // instruction as we finish with the bundled instruction to work around
+      // the inconsistency.
+      LiveCPSR = false;
+
     bool DefCPSR = false;
     LiveCPSR = UpdateCPSRDef(*MI, LiveCPSR, DefCPSR);
     if (MI->isCall()) {

lib/Target/X86/X86InstrInfo.cpp

@@ -1528,9 +1528,9 @@ X86InstrInfo::convertToThreeAddressWithLEA(unsigned MIOpc,
       leaInReg2 = RegInfo.createVirtualRegister(&X86::GR32_NOSPRegClass);
       // Build and insert into an implicit UNDEF value. This is OK because
       // well be shifting and then extracting the lower 16-bits.
-      BuildMI(*MFI, MIB, MI->getDebugLoc(), get(X86::IMPLICIT_DEF), leaInReg2);
+      BuildMI(*MFI, &*MIB, MI->getDebugLoc(), get(X86::IMPLICIT_DEF),leaInReg2);
       InsMI2 =
-        BuildMI(*MFI, MIB, MI->getDebugLoc(), get(TargetOpcode::COPY))
+        BuildMI(*MFI, &*MIB, MI->getDebugLoc(), get(TargetOpcode::COPY))
         .addReg(leaInReg2, RegState::Define, X86::sub_16bit)
         .addReg(Src2, getKillRegState(isKill2));
       addRegReg(MIB, leaInReg, true, leaInReg2, true);

test/CodeGen/ARM/debug-info-d16-reg.ll

@@ -1,4 +1,4 @@
-; RUN: llc < %s - | FileCheck %s
+; RUN: llc < %s | FileCheck %s
 ; Radar 9309221
 ; Test dwarf reg no for d16
 ;CHECK: DW_OP_regx

test/CodeGen/ARM/long_shift.ll

@@ -25,8 +25,8 @@ define i32 @f2(i64 %x, i64 %y) {
 ; CHECK:      lsr{{.*}}r2
 ; CHECK-NEXT: rsb     r3, r2, #32
 ; CHECK-NEXT: sub     r2, r2, #32
-; CHECK-NEXT: cmp     r2, #0
 ; CHECK-NEXT: orr     r0, r0, r1, lsl r3
+; CHECK-NEXT: cmp     r2, #0
 ; CHECK-NEXT: asrge   r0, r1, r2
 	%a = ashr i64 %x, %y
 	%b = trunc i64 %a to i32
@@ -38,8 +38,8 @@ define i32 @f3(i64 %x, i64 %y) {
 ; CHECK:      lsr{{.*}}r2
 ; CHECK-NEXT: rsb     r3, r2, #32
 ; CHECK-NEXT: sub     r2, r2, #32
-; CHECK-NEXT: cmp     r2, #0
 ; CHECK-NEXT: orr     r0, r0, r1, lsl r3
+; CHECK-NEXT: cmp     r2, #0
 ; CHECK-NEXT: lsrge   r0, r1, r2
 	%a = lshr i64 %x, %y
 	%b = trunc i64 %a to i32

test/CodeGen/Thumb2/machine-licm.ll

@@ -94,8 +94,8 @@ bb.nph:
 
 bb:                                               ; preds = %bb, %bb.nph
 ; CHECK: bb
-; CHECK: eor.w {{(r[0-9])|(lr)}}, {{(r[0-9])|(lr)}}, [[REGISTER]]
 ; CHECK: eor.w
+; CHECK: eor.w {{(r[0-9])|(lr)}}, {{(r[0-9])|(lr)}}, [[REGISTER]]
 ; CHECK-NOT: eor
 ; CHECK: and
   %data_addr.013 = phi i8 [ %data, %bb.nph ], [ %8, %bb ] ; <i8> [#uses=2]

test/CodeGen/Thumb2/thumb2-select_xform.ll

@@ -3,8 +3,8 @@
 define i32 @t1(i32 %a, i32 %b, i32 %c) nounwind {
 ; CHECK: t1
 ; CHECK: mvn r0, #-2147483648
-; CHECK: add r0, r1
 ; CHECK: cmp r2, #10
+; CHECK: add r0, r1
 ; CHECK: it  gt
 ; CHECK: movgt r0, r1
         %tmp1 = icmp sgt i32 %c, 10