Home | History | Annotate | Download | only in PowerPC 
      1 //===-- PPCCTRLoops.cpp - Identify and generate CTR loops -----------------===//
      2 //
      3 //                     The LLVM Compiler Infrastructure
      4 //
      5 // This file is distributed under the University of Illinois Open Source
      6 // License. See LICENSE.TXT for details.
      7 //
      8 //===----------------------------------------------------------------------===//
      9 //
     10 // This pass identifies loops where we can generate the PPC branch instructions
     11 // that decrement and test the count register (CTR) (bdnz and friends).
     12 //
     13 // The pattern that defines the induction variable can changed depending on
     14 // prior optimizations.  For example, the IndVarSimplify phase run by 'opt'
     15 // normalizes induction variables, and the Loop Strength Reduction pass
     16 // run by 'llc' may also make changes to the induction variable.
     17 //
     18 // Criteria for CTR loops:
     19 //  - Countable loops (w/ ind. var for a trip count)
     20 //  - Try inner-most loops first
     21 //  - No nested CTR loops.
     22 //  - No function calls in loops.
     23 //
     24 //===----------------------------------------------------------------------===//
     25 
     26 #define DEBUG_TYPE "ctrloops"
     27 
     28 #include "llvm/Transforms/Scalar.h"
     29 #include "llvm/ADT/Statistic.h"
     30 #include "llvm/ADT/STLExtras.h"
     31 #include "llvm/Analysis/Dominators.h"
     32 #include "llvm/Analysis/LoopInfo.h"
     33 #include "llvm/Analysis/ScalarEvolutionExpander.h"
     34 #include "llvm/IR/Constants.h"
     35 #include "llvm/IR/DerivedTypes.h"
     36 #include "llvm/IR/InlineAsm.h"
     37 #include "llvm/IR/Instructions.h"
     38 #include "llvm/IR/IntrinsicInst.h"
     39 #include "llvm/IR/Module.h"
     40 #include "llvm/PassSupport.h"
     41 #include "llvm/Support/CommandLine.h"
     42 #include "llvm/Support/Debug.h"
     43 #include "llvm/Support/ValueHandle.h"
     44 #include "llvm/Support/raw_ostream.h"
     45 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
     46 #include "llvm/Transforms/Utils/Local.h"
     47 #include "llvm/Transforms/Utils/LoopUtils.h"
     48 #include "llvm/Target/TargetLibraryInfo.h"
     49 #include "PPCTargetMachine.h"
     50 #include "PPC.h"
     51 
     52 #ifndef NDEBUG
     53 #include "llvm/CodeGen/MachineDominators.h"
     54 #include "llvm/CodeGen/MachineFunction.h"
     55 #include "llvm/CodeGen/MachineFunctionPass.h"
     56 #include "llvm/CodeGen/MachineRegisterInfo.h"
     57 #endif
     58 
     59 #include <algorithm>
     60 #include <vector>
     61 
     62 using namespace llvm;
     63 
     64 #ifndef NDEBUG
     65 static cl::opt<int> CTRLoopLimit("ppc-max-ctrloop", cl::Hidden, cl::init(-1));
     66 #endif
     67 
     68 STATISTIC(NumCTRLoops, "Number of loops converted to CTR loops");
     69 
     70 namespace llvm {
     71   void initializePPCCTRLoopsPass(PassRegistry&);
     72 #ifndef NDEBUG
     73   void initializePPCCTRLoopsVerifyPass(PassRegistry&);
     74 #endif
     75 }
     76 
     77 namespace {
     78   struct PPCCTRLoops : public FunctionPass {
     79 
     80 #ifndef NDEBUG
     81     static int Counter;
     82 #endif
     83 
     84   public:
     85     static char ID;
     86 
     87     PPCCTRLoops() : FunctionPass(ID), TM(0) {
     88       initializePPCCTRLoopsPass(*PassRegistry::getPassRegistry());
     89     }
     90     PPCCTRLoops(PPCTargetMachine &TM) : FunctionPass(ID), TM(&TM) {
     91       initializePPCCTRLoopsPass(*PassRegistry::getPassRegistry());
     92     }
     93 
     94     virtual bool runOnFunction(Function &F);
     95 
     96     virtual void getAnalysisUsage(AnalysisUsage &AU) const {
     97       AU.addRequired<LoopInfo>();
     98       AU.addPreserved<LoopInfo>();
     99       AU.addRequired<DominatorTree>();
    100       AU.addPreserved<DominatorTree>();
    101       AU.addRequired<ScalarEvolution>();
    102     }
    103 
    104   private:
    105     bool mightUseCTR(const Triple &TT, BasicBlock *BB);
    106     bool convertToCTRLoop(Loop *L);
    107 
    108   private:
    109     PPCTargetMachine *TM;
    110     LoopInfo *LI;
    111     ScalarEvolution *SE;
    112     DataLayout *TD;
    113     DominatorTree *DT;
    114     const TargetLibraryInfo *LibInfo;
    115   };
    116 
    117   char PPCCTRLoops::ID = 0;
    118 #ifndef NDEBUG
    119   int PPCCTRLoops::Counter = 0;
    120 #endif
    121 
    122 #ifndef NDEBUG
    123   struct PPCCTRLoopsVerify : public MachineFunctionPass {
    124   public:
    125     static char ID;
    126 
    127     PPCCTRLoopsVerify() : MachineFunctionPass(ID) {
    128       initializePPCCTRLoopsVerifyPass(*PassRegistry::getPassRegistry());
    129     }
    130 
    131     virtual void getAnalysisUsage(AnalysisUsage &AU) const {
    132       AU.addRequired<MachineDominatorTree>();
    133       MachineFunctionPass::getAnalysisUsage(AU);
    134     }
    135 
    136     virtual bool runOnMachineFunction(MachineFunction &MF);
    137 
    138   private:
    139     MachineDominatorTree *MDT;
    140   };
    141 
    142   char PPCCTRLoopsVerify::ID = 0;
    143 #endif // NDEBUG
    144 } // end anonymous namespace
    145 
    146 INITIALIZE_PASS_BEGIN(PPCCTRLoops, "ppc-ctr-loops", "PowerPC CTR Loops",
    147                       false, false)
    148 INITIALIZE_PASS_DEPENDENCY(DominatorTree)
    149 INITIALIZE_PASS_DEPENDENCY(LoopInfo)
    150 INITIALIZE_PASS_DEPENDENCY(ScalarEvolution)
    151 INITIALIZE_PASS_END(PPCCTRLoops, "ppc-ctr-loops", "PowerPC CTR Loops",
    152                     false, false)
    153 
    154 FunctionPass *llvm::createPPCCTRLoops(PPCTargetMachine &TM) {
    155   return new PPCCTRLoops(TM);
    156 }
    157 
    158 #ifndef NDEBUG
    159 INITIALIZE_PASS_BEGIN(PPCCTRLoopsVerify, "ppc-ctr-loops-verify",
    160                       "PowerPC CTR Loops Verify", false, false)
    161 INITIALIZE_PASS_DEPENDENCY(MachineDominatorTree)
    162 INITIALIZE_PASS_END(PPCCTRLoopsVerify, "ppc-ctr-loops-verify",
    163                     "PowerPC CTR Loops Verify", false, false)
    164 
    165 FunctionPass *llvm::createPPCCTRLoopsVerify() {
    166   return new PPCCTRLoopsVerify();
    167 }
    168 #endif // NDEBUG
    169 
    170 bool PPCCTRLoops::runOnFunction(Function &F) {
    171   LI = &getAnalysis<LoopInfo>();
    172   SE = &getAnalysis<ScalarEvolution>();
    173   DT = &getAnalysis<DominatorTree>();
    174   TD = getAnalysisIfAvailable<DataLayout>();
    175   LibInfo = getAnalysisIfAvailable<TargetLibraryInfo>();
    176 
    177   bool MadeChange = false;
    178 
    179   for (LoopInfo::iterator I = LI->begin(), E = LI->end();
    180        I != E; ++I) {
    181     Loop *L = *I;
    182     if (!L->getParentLoop())
    183       MadeChange |= convertToCTRLoop(L);
    184   }
    185 
    186   return MadeChange;
    187 }
    188 
    189 bool PPCCTRLoops::mightUseCTR(const Triple &TT, BasicBlock *BB) {
    190   for (BasicBlock::iterator J = BB->begin(), JE = BB->end();
    191        J != JE; ++J) {
    192     if (CallInst *CI = dyn_cast<CallInst>(J)) {
    193       if (InlineAsm *IA = dyn_cast<InlineAsm>(CI->getCalledValue())) {
    194         // Inline ASM is okay, unless it clobbers the ctr register.
    195         InlineAsm::ConstraintInfoVector CIV = IA->ParseConstraints();
    196         for (unsigned i = 0, ie = CIV.size(); i < ie; ++i) {
    197           InlineAsm::ConstraintInfo &C = CIV[i];
    198           if (C.Type != InlineAsm::isInput)
    199             for (unsigned j = 0, je = C.Codes.size(); j < je; ++j)
    200               if (StringRef(C.Codes[j]).equals_lower("{ctr}"))
    201                 return true;
    202         }
    203 
    204         continue;
    205       }
    206 
    207       if (!TM)
    208         return true;
    209       const TargetLowering *TLI = TM->getTargetLowering();
    210 
    211       if (Function *F = CI->getCalledFunction()) {
    212         // Most intrinsics don't become function calls, but some might.
    213         // sin, cos, exp and log are always calls.
    214         unsigned Opcode;
    215         if (F->getIntrinsicID() != Intrinsic::not_intrinsic) {
    216           switch (F->getIntrinsicID()) {
    217           default: continue;
    218 
    219 // VisualStudio defines setjmp as _setjmp
    220 #if defined(_MSC_VER) && defined(setjmp) && \
    221                        !defined(setjmp_undefined_for_msvc)
    222 #  pragma push_macro("setjmp")
    223 #  undef setjmp
    224 #  define setjmp_undefined_for_msvc
    225 #endif
    226 
    227           case Intrinsic::setjmp:
    228 
    229 #if defined(_MSC_VER) && defined(setjmp_undefined_for_msvc)
    230  // let's return it to _setjmp state
    231 #  pragma pop_macro("setjmp")
    232 #  undef setjmp_undefined_for_msvc
    233 #endif
    234 
    235           case Intrinsic::longjmp:
    236 
    237           // Exclude eh_sjlj_setjmp; we don't need to exclude eh_sjlj_longjmp
    238           // because, although it does clobber the counter register, the
    239           // control can't then return to inside the loop unless there is also
    240           // an eh_sjlj_setjmp.
    241           case Intrinsic::eh_sjlj_setjmp:
    242 
    243           case Intrinsic::memcpy:
    244           case Intrinsic::memmove:
    245           case Intrinsic::memset:
    246           case Intrinsic::powi:
    247           case Intrinsic::log:
    248           case Intrinsic::log2:
    249           case Intrinsic::log10:
    250           case Intrinsic::exp:
    251           case Intrinsic::exp2:
    252           case Intrinsic::pow:
    253           case Intrinsic::sin:
    254           case Intrinsic::cos:
    255             return true;
    256           case Intrinsic::sqrt:      Opcode = ISD::FSQRT;      break;
    257           case Intrinsic::floor:     Opcode = ISD::FFLOOR;     break;
    258           case Intrinsic::ceil:      Opcode = ISD::FCEIL;      break;
    259           case Intrinsic::trunc:     Opcode = ISD::FTRUNC;     break;
    260           case Intrinsic::rint:      Opcode = ISD::FRINT;      break;
    261           case Intrinsic::nearbyint: Opcode = ISD::FNEARBYINT; break;
    262           }
    263         }
    264 
    265         // PowerPC does not use [US]DIVREM or other library calls for
    266         // operations on regular types which are not otherwise library calls
    267         // (i.e. soft float or atomics). If adapting for targets that do,
    268         // additional care is required here.
    269 
    270         LibFunc::Func Func;
    271         if (!F->hasLocalLinkage() && F->hasName() && LibInfo &&
    272             LibInfo->getLibFunc(F->getName(), Func) &&
    273             LibInfo->hasOptimizedCodeGen(Func)) {
    274           // Non-read-only functions are never treated as intrinsics.
    275           if (!CI->onlyReadsMemory())
    276             return true;
    277 
    278           // Conversion happens only for FP calls.
    279           if (!CI->getArgOperand(0)->getType()->isFloatingPointTy())
    280             return true;
    281 
    282           switch (Func) {
    283           default: return true;
    284           case LibFunc::copysign:
    285           case LibFunc::copysignf:
    286           case LibFunc::copysignl:
    287             continue; // ISD::FCOPYSIGN is never a library call.
    288           case LibFunc::fabs:
    289           case LibFunc::fabsf:
    290           case LibFunc::fabsl:
    291             continue; // ISD::FABS is never a library call.
    292           case LibFunc::sqrt:
    293           case LibFunc::sqrtf:
    294           case LibFunc::sqrtl:
    295             Opcode = ISD::FSQRT; break;
    296           case LibFunc::floor:
    297           case LibFunc::floorf:
    298           case LibFunc::floorl:
    299             Opcode = ISD::FFLOOR; break;
    300           case LibFunc::nearbyint:
    301           case LibFunc::nearbyintf:
    302           case LibFunc::nearbyintl:
    303             Opcode = ISD::FNEARBYINT; break;
    304           case LibFunc::ceil:
    305           case LibFunc::ceilf:
    306           case LibFunc::ceill:
    307             Opcode = ISD::FCEIL; break;
    308           case LibFunc::rint:
    309           case LibFunc::rintf:
    310           case LibFunc::rintl:
    311             Opcode = ISD::FRINT; break;
    312           case LibFunc::trunc:
    313           case LibFunc::truncf:
    314           case LibFunc::truncl:
    315             Opcode = ISD::FTRUNC; break;
    316           }
    317 
    318           MVT VTy =
    319             TLI->getSimpleValueType(CI->getArgOperand(0)->getType(), true);
    320           if (VTy == MVT::Other)
    321             return true;
    322 
    323           if (TLI->isOperationLegalOrCustom(Opcode, VTy))
    324             continue;
    325           else if (VTy.isVector() &&
    326                    TLI->isOperationLegalOrCustom(Opcode, VTy.getScalarType()))
    327             continue;
    328 
    329           return true;
    330         }
    331       }
    332 
    333       return true;
    334     } else if (isa<BinaryOperator>(J) &&
    335                J->getType()->getScalarType()->isPPC_FP128Ty()) {
    336       // Most operations on ppc_f128 values become calls.
    337       return true;
    338     } else if (isa<UIToFPInst>(J) || isa<SIToFPInst>(J) ||
    339                isa<FPToUIInst>(J) || isa<FPToSIInst>(J)) {
    340       CastInst *CI = cast<CastInst>(J);
    341       if (CI->getSrcTy()->getScalarType()->isPPC_FP128Ty() ||
    342           CI->getDestTy()->getScalarType()->isPPC_FP128Ty() ||
    343           (TT.isArch32Bit() &&
    344            (CI->getSrcTy()->getScalarType()->isIntegerTy(64) ||
    345             CI->getDestTy()->getScalarType()->isIntegerTy(64))
    346           ))
    347         return true;
    348     } else if (TT.isArch32Bit() &&
    349                J->getType()->getScalarType()->isIntegerTy(64) &&
    350                (J->getOpcode() == Instruction::UDiv ||
    351                 J->getOpcode() == Instruction::SDiv ||
    352                 J->getOpcode() == Instruction::URem ||
    353                 J->getOpcode() == Instruction::SRem)) {
    354       return true;
    355     } else if (isa<IndirectBrInst>(J) || isa<InvokeInst>(J)) {
    356       // On PowerPC, indirect jumps use the counter register.
    357       return true;
    358     } else if (SwitchInst *SI = dyn_cast<SwitchInst>(J)) {
    359       if (!TM)
    360         return true;
    361       const TargetLowering *TLI = TM->getTargetLowering();
    362 
    363       if (TLI->supportJumpTables() &&
    364           SI->getNumCases()+1 >= (unsigned) TLI->getMinimumJumpTableEntries())
    365         return true;
    366     }
    367   }
    368 
    369   return false;
    370 }
    371 
    372 bool PPCCTRLoops::convertToCTRLoop(Loop *L) {
    373   bool MadeChange = false;
    374 
    375   Triple TT = Triple(L->getHeader()->getParent()->getParent()->
    376                      getTargetTriple());
    377   if (!TT.isArch32Bit() && !TT.isArch64Bit())
    378     return MadeChange; // Unknown arch. type.
    379 
    380   // Process nested loops first.
    381   for (Loop::iterator I = L->begin(), E = L->end(); I != E; ++I) {
    382     MadeChange |= convertToCTRLoop(*I);
    383   }
    384 
    385   // If a nested loop has been converted, then we can't convert this loop.
    386   if (MadeChange)
    387     return MadeChange;
    388 
    389 #ifndef NDEBUG
    390   // Stop trying after reaching the limit (if any).
    391   int Limit = CTRLoopLimit;
    392   if (Limit >= 0) {
    393     if (Counter >= CTRLoopLimit)
    394       return false;
    395     Counter++;
    396   }
    397 #endif
    398 
    399   // We don't want to spill/restore the counter register, and so we don't
    400   // want to use the counter register if the loop contains calls.
    401   for (Loop::block_iterator I = L->block_begin(), IE = L->block_end();
    402        I != IE; ++I)
    403     if (mightUseCTR(TT, *I))
    404       return MadeChange;
    405 
    406   SmallVector<BasicBlock*, 4> ExitingBlocks;
    407   L->getExitingBlocks(ExitingBlocks);
    408 
    409   BasicBlock *CountedExitBlock = 0;
    410   const SCEV *ExitCount = 0;
    411   BranchInst *CountedExitBranch = 0;
    412   for (SmallVectorImpl<BasicBlock *>::iterator I = ExitingBlocks.begin(),
    413        IE = ExitingBlocks.end(); I != IE; ++I) {
    414     const SCEV *EC = SE->getExitCount(L, *I);
    415     DEBUG(dbgs() << "Exit Count for " << *L << " from block " <<
    416                     (*I)->getName() << ": " << *EC << "\n");
    417     if (isa<SCEVCouldNotCompute>(EC))
    418       continue;
    419     if (const SCEVConstant *ConstEC = dyn_cast<SCEVConstant>(EC)) {
    420       if (ConstEC->getValue()->isZero())
    421         continue;
    422     } else if (!SE->isLoopInvariant(EC, L))
    423       continue;
    424 
    425     if (SE->getTypeSizeInBits(EC->getType()) > (TT.isArch64Bit() ? 64 : 32))
    426       continue;
    427 
    428     // We now have a loop-invariant count of loop iterations (which is not the
    429     // constant zero) for which we know that this loop will not exit via this
    430     // exisiting block.
    431 
    432     // We need to make sure that this block will run on every loop iteration.
    433     // For this to be true, we must dominate all blocks with backedges. Such
    434     // blocks are in-loop predecessors to the header block.
    435     bool NotAlways = false;
    436     for (pred_iterator PI = pred_begin(L->getHeader()),
    437          PIE = pred_end(L->getHeader()); PI != PIE; ++PI) {
    438       if (!L->contains(*PI))
    439         continue;
    440 
    441       if (!DT->dominates(*I, *PI)) {
    442         NotAlways = true;
    443         break;
    444       }
    445     }
    446 
    447     if (NotAlways)
    448       continue;
    449 
    450     // Make sure this blocks ends with a conditional branch.
    451     Instruction *TI = (*I)->getTerminator();
    452     if (!TI)
    453       continue;
    454 
    455     if (BranchInst *BI = dyn_cast<BranchInst>(TI)) {
    456       if (!BI->isConditional())
    457         continue;
    458 
    459       CountedExitBranch = BI;
    460     } else
    461       continue;
    462 
    463     // Note that this block may not be the loop latch block, even if the loop
    464     // has a latch block.
    465     CountedExitBlock = *I;
    466     ExitCount = EC;
    467     break;
    468   }
    469 
    470   if (!CountedExitBlock)
    471     return MadeChange;
    472 
    473   BasicBlock *Preheader = L->getLoopPreheader();
    474 
    475   // If we don't have a preheader, then insert one. If we already have a
    476   // preheader, then we can use it (except if the preheader contains a use of
    477   // the CTR register because some such uses might be reordered by the
    478   // selection DAG after the mtctr instruction).
    479   if (!Preheader || mightUseCTR(TT, Preheader))
    480     Preheader = InsertPreheaderForLoop(L, this);
    481   if (!Preheader)
    482     return MadeChange;
    483 
    484   DEBUG(dbgs() << "Preheader for exit count: " << Preheader->getName() << "\n");
    485 
    486   // Insert the count into the preheader and replace the condition used by the
    487   // selected branch.
    488   MadeChange = true;
    489 
    490   SCEVExpander SCEVE(*SE, "loopcnt");
    491   LLVMContext &C = SE->getContext();
    492   Type *CountType = TT.isArch64Bit() ? Type::getInt64Ty(C) :
    493                                        Type::getInt32Ty(C);
    494   if (!ExitCount->getType()->isPointerTy() &&
    495       ExitCount->getType() != CountType)
    496     ExitCount = SE->getZeroExtendExpr(ExitCount, CountType);
    497   ExitCount = SE->getAddExpr(ExitCount,
    498                              SE->getConstant(CountType, 1));
    499   Value *ECValue = SCEVE.expandCodeFor(ExitCount, CountType,
    500                                        Preheader->getTerminator());
    501 
    502   IRBuilder<> CountBuilder(Preheader->getTerminator());
    503   Module *M = Preheader->getParent()->getParent();
    504   Value *MTCTRFunc = Intrinsic::getDeclaration(M, Intrinsic::ppc_mtctr,
    505                                                CountType);
    506   CountBuilder.CreateCall(MTCTRFunc, ECValue);
    507 
    508   IRBuilder<> CondBuilder(CountedExitBranch);
    509   Value *DecFunc =
    510     Intrinsic::getDeclaration(M, Intrinsic::ppc_is_decremented_ctr_nonzero);
    511   Value *NewCond = CondBuilder.CreateCall(DecFunc);
    512   Value *OldCond = CountedExitBranch->getCondition();
    513   CountedExitBranch->setCondition(NewCond);
    514 
    515   // The false branch must exit the loop.
    516   if (!L->contains(CountedExitBranch->getSuccessor(0)))
    517     CountedExitBranch->swapSuccessors();
    518 
    519   // The old condition may be dead now, and may have even created a dead PHI
    520   // (the original induction variable).
    521   RecursivelyDeleteTriviallyDeadInstructions(OldCond);
    522   DeleteDeadPHIs(CountedExitBlock);
    523 
    524   ++NumCTRLoops;
    525   return MadeChange;
    526 }
    527 
    528 #ifndef NDEBUG
    529 static bool clobbersCTR(const MachineInstr *MI) {
    530   for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
    531     const MachineOperand &MO = MI->getOperand(i);
    532     if (MO.isReg()) {
    533       if (MO.isDef() && (MO.getReg() == PPC::CTR || MO.getReg() == PPC::CTR8))
    534         return true;
    535     } else if (MO.isRegMask()) {
    536       if (MO.clobbersPhysReg(PPC::CTR) || MO.clobbersPhysReg(PPC::CTR8))
    537         return true;
    538     }
    539   }
    540 
    541   return false;
    542 }
    543 
    544 static bool verifyCTRBranch(MachineBasicBlock *MBB,
    545                             MachineBasicBlock::iterator I) {
    546   MachineBasicBlock::iterator BI = I;
    547   SmallSet<MachineBasicBlock *, 16>   Visited;
    548   SmallVector<MachineBasicBlock *, 8> Preds;
    549   bool CheckPreds;
    550 
    551   if (I == MBB->begin()) {
    552     Visited.insert(MBB);
    553     goto queue_preds;
    554   } else
    555     --I;
    556 
    557 check_block:
    558   Visited.insert(MBB);
    559   if (I == MBB->end())
    560     goto queue_preds;
    561 
    562   CheckPreds = true;
    563   for (MachineBasicBlock::iterator IE = MBB->begin();; --I) {
    564     unsigned Opc = I->getOpcode();
    565     if (Opc == PPC::MTCTRloop || Opc == PPC::MTCTR8loop) {
    566       CheckPreds = false;
    567       break;
    568     }
    569 
    570     if (I != BI && clobbersCTR(I)) {
    571       DEBUG(dbgs() << "BB#" << MBB->getNumber() << " (" <<
    572                       MBB->getFullName() << ") instruction " << *I <<
    573                       " clobbers CTR, invalidating " << "BB#" <<
    574                       BI->getParent()->getNumber() << " (" <<
    575                       BI->getParent()->getFullName() << ") instruction " <<
    576                       *BI << "\n");
    577       return false;
    578     }
    579 
    580     if (I == IE)
    581       break;
    582   }
    583 
    584   if (!CheckPreds && Preds.empty())
    585     return true;
    586 
    587   if (CheckPreds) {
    588 queue_preds:
    589     if (MachineFunction::iterator(MBB) == MBB->getParent()->begin()) {
    590       DEBUG(dbgs() << "Unable to find a MTCTR instruction for BB#" <<
    591                       BI->getParent()->getNumber() << " (" <<
    592                       BI->getParent()->getFullName() << ") instruction " <<
    593                       *BI << "\n");
    594       return false;
    595     }
    596 
    597     for (MachineBasicBlock::pred_iterator PI = MBB->pred_begin(),
    598          PIE = MBB->pred_end(); PI != PIE; ++PI)
    599       Preds.push_back(*PI);
    600   }
    601 
    602   do {
    603     MBB = Preds.pop_back_val();
    604     if (!Visited.count(MBB)) {
    605       I = MBB->getLastNonDebugInstr();
    606       goto check_block;
    607     }
    608   } while (!Preds.empty());
    609 
    610   return true;
    611 }
    612 
    613 bool PPCCTRLoopsVerify::runOnMachineFunction(MachineFunction &MF) {
    614   MDT = &getAnalysis<MachineDominatorTree>();
    615 
    616   // Verify that all bdnz/bdz instructions are dominated by a loop mtctr before
    617   // any other instructions that might clobber the ctr register.
    618   for (MachineFunction::iterator I = MF.begin(), IE = MF.end();
    619        I != IE; ++I) {
    620     MachineBasicBlock *MBB = I;
    621     if (!MDT->isReachableFromEntry(MBB))
    622       continue;
    623 
    624     for (MachineBasicBlock::iterator MII = MBB->getFirstTerminator(),
    625       MIIE = MBB->end(); MII != MIIE; ++MII) {
    626       unsigned Opc = MII->getOpcode();
    627       if (Opc == PPC::BDNZ8 || Opc == PPC::BDNZ ||
    628           Opc == PPC::BDZ8  || Opc == PPC::BDZ)
    629         if (!verifyCTRBranch(MBB, MII))
    630           llvm_unreachable("Invalid PPC CTR loop!");
    631     }
    632   }
    633 
    634   return false;
    635 }
    636 #endif // NDEBUG
    637 
    638