PPCCTRLoops.cpp revision 36b56886974eae4f9c5ebc96befd3e7bfe5de338 
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 "PPC.h"
30#include "PPCTargetMachine.h"
31#include "llvm/ADT/STLExtras.h"
32#include "llvm/ADT/Statistic.h"
33#include "llvm/Analysis/LoopInfo.h"
34#include "llvm/Analysis/ScalarEvolutionExpander.h"
35#include "llvm/IR/Constants.h"
36#include "llvm/IR/DerivedTypes.h"
37#include "llvm/IR/Dominators.h"
38#include "llvm/IR/InlineAsm.h"
39#include "llvm/IR/Instructions.h"
40#include "llvm/IR/IntrinsicInst.h"
41#include "llvm/IR/Module.h"
42#include "llvm/IR/ValueHandle.h"
43#include "llvm/PassSupport.h"
44#include "llvm/Support/CommandLine.h"
45#include "llvm/Support/Debug.h"
46#include "llvm/Support/raw_ostream.h"
47#include "llvm/Target/TargetLibraryInfo.h"
48#include "llvm/Transforms/Utils/BasicBlockUtils.h"
49#include "llvm/Transforms/Utils/Local.h"
50#include "llvm/Transforms/Utils/LoopUtils.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
62using namespace llvm;
63
64#ifndef NDEBUG
65static cl::opt<int> CTRLoopLimit("ppc-max-ctrloop", cl::Hidden, cl::init(-1));
66#endif
67
68STATISTIC(NumCTRLoops, "Number of loops converted to CTR loops");
69
70namespace llvm {
71  void initializePPCCTRLoopsPass(PassRegistry&);
72#ifndef NDEBUG
73  void initializePPCCTRLoopsVerifyPass(PassRegistry&);
74#endif
75}
76
77namespace {
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<DominatorTreeWrapperPass>();
100      AU.addPreserved<DominatorTreeWrapperPass>();
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    const DataLayout *DL;
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
146INITIALIZE_PASS_BEGIN(PPCCTRLoops, "ppc-ctr-loops", "PowerPC CTR Loops",
147                      false, false)
148INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)
149INITIALIZE_PASS_DEPENDENCY(LoopInfo)
150INITIALIZE_PASS_DEPENDENCY(ScalarEvolution)
151INITIALIZE_PASS_END(PPCCTRLoops, "ppc-ctr-loops", "PowerPC CTR Loops",
152                    false, false)
153
154FunctionPass *llvm::createPPCCTRLoops(PPCTargetMachine &TM) {
155  return new PPCCTRLoops(TM);
156}
157
158#ifndef NDEBUG
159INITIALIZE_PASS_BEGIN(PPCCTRLoopsVerify, "ppc-ctr-loops-verify",
160                      "PowerPC CTR Loops Verify", false, false)
161INITIALIZE_PASS_DEPENDENCY(MachineDominatorTree)
162INITIALIZE_PASS_END(PPCCTRLoopsVerify, "ppc-ctr-loops-verify",
163                    "PowerPC CTR Loops Verify", false, false)
164
165FunctionPass *llvm::createPPCCTRLoopsVerify() {
166  return new PPCCTRLoopsVerify();
167}
168#endif // NDEBUG
169
170bool PPCCTRLoops::runOnFunction(Function &F) {
171  LI = &getAnalysis<LoopInfo>();
172  SE = &getAnalysis<ScalarEvolution>();
173  DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree();
174  DataLayoutPass *DLP = getAnalysisIfAvailable<DataLayoutPass>();
175  DL = DLP ? &DLP->getDataLayout() : 0;
176  LibInfo = getAnalysisIfAvailable<TargetLibraryInfo>();
177
178  bool MadeChange = false;
179
180  for (LoopInfo::iterator I = LI->begin(), E = LI->end();
181       I != E; ++I) {
182    Loop *L = *I;
183    if (!L->getParentLoop())
184      MadeChange |= convertToCTRLoop(L);
185  }
186
187  return MadeChange;
188}
189
190static bool isLargeIntegerTy(bool Is32Bit, Type *Ty) {
191  if (IntegerType *ITy = dyn_cast<IntegerType>(Ty))
192    return ITy->getBitWidth() > (Is32Bit ? 32U : 64U);
193
194  return false;
195}
196
197bool PPCCTRLoops::mightUseCTR(const Triple &TT, BasicBlock *BB) {
198  for (BasicBlock::iterator J = BB->begin(), JE = BB->end();
199       J != JE; ++J) {
200    if (CallInst *CI = dyn_cast<CallInst>(J)) {
201      if (InlineAsm *IA = dyn_cast<InlineAsm>(CI->getCalledValue())) {
202        // Inline ASM is okay, unless it clobbers the ctr register.
203        InlineAsm::ConstraintInfoVector CIV = IA->ParseConstraints();
204        for (unsigned i = 0, ie = CIV.size(); i < ie; ++i) {
205          InlineAsm::ConstraintInfo &C = CIV[i];
206          if (C.Type != InlineAsm::isInput)
207            for (unsigned j = 0, je = C.Codes.size(); j < je; ++j)
208              if (StringRef(C.Codes[j]).equals_lower("{ctr}"))
209                return true;
210        }
211
212        continue;
213      }
214
215      if (!TM)
216        return true;
217      const TargetLowering *TLI = TM->getTargetLowering();
218
219      if (Function *F = CI->getCalledFunction()) {
220        // Most intrinsics don't become function calls, but some might.
221        // sin, cos, exp and log are always calls.
222        unsigned Opcode;
223        if (F->getIntrinsicID() != Intrinsic::not_intrinsic) {
224          switch (F->getIntrinsicID()) {
225          default: continue;
226
227// VisualStudio defines setjmp as _setjmp
228#if defined(_MSC_VER) && defined(setjmp) && \
229                       !defined(setjmp_undefined_for_msvc)
230#  pragma push_macro("setjmp")
231#  undef setjmp
232#  define setjmp_undefined_for_msvc
233#endif
234
235          case Intrinsic::setjmp:
236
237#if defined(_MSC_VER) && defined(setjmp_undefined_for_msvc)
238 // let's return it to _setjmp state
239#  pragma pop_macro("setjmp")
240#  undef setjmp_undefined_for_msvc
241#endif
242
243          case Intrinsic::longjmp:
244
245          // Exclude eh_sjlj_setjmp; we don't need to exclude eh_sjlj_longjmp
246          // because, although it does clobber the counter register, the
247          // control can't then return to inside the loop unless there is also
248          // an eh_sjlj_setjmp.
249          case Intrinsic::eh_sjlj_setjmp:
250
251          case Intrinsic::memcpy:
252          case Intrinsic::memmove:
253          case Intrinsic::memset:
254          case Intrinsic::powi:
255          case Intrinsic::log:
256          case Intrinsic::log2:
257          case Intrinsic::log10:
258          case Intrinsic::exp:
259          case Intrinsic::exp2:
260          case Intrinsic::pow:
261          case Intrinsic::sin:
262          case Intrinsic::cos:
263            return true;
264          case Intrinsic::copysign:
265            if (CI->getArgOperand(0)->getType()->getScalarType()->
266                isPPC_FP128Ty())
267              return true;
268            else
269              continue; // ISD::FCOPYSIGN is never a library call.
270          case Intrinsic::sqrt:      Opcode = ISD::FSQRT;      break;
271          case Intrinsic::floor:     Opcode = ISD::FFLOOR;     break;
272          case Intrinsic::ceil:      Opcode = ISD::FCEIL;      break;
273          case Intrinsic::trunc:     Opcode = ISD::FTRUNC;     break;
274          case Intrinsic::rint:      Opcode = ISD::FRINT;      break;
275          case Intrinsic::nearbyint: Opcode = ISD::FNEARBYINT; break;
276          case Intrinsic::round:     Opcode = ISD::FROUND;     break;
277          }
278        }
279
280        // PowerPC does not use [US]DIVREM or other library calls for
281        // operations on regular types which are not otherwise library calls
282        // (i.e. soft float or atomics). If adapting for targets that do,
283        // additional care is required here.
284
285        LibFunc::Func Func;
286        if (!F->hasLocalLinkage() && F->hasName() && LibInfo &&
287            LibInfo->getLibFunc(F->getName(), Func) &&
288            LibInfo->hasOptimizedCodeGen(Func)) {
289          // Non-read-only functions are never treated as intrinsics.
290          if (!CI->onlyReadsMemory())
291            return true;
292
293          // Conversion happens only for FP calls.
294          if (!CI->getArgOperand(0)->getType()->isFloatingPointTy())
295            return true;
296
297          switch (Func) {
298          default: return true;
299          case LibFunc::copysign:
300          case LibFunc::copysignf:
301            continue; // ISD::FCOPYSIGN is never a library call.
302          case LibFunc::copysignl:
303            return true;
304          case LibFunc::fabs:
305          case LibFunc::fabsf:
306          case LibFunc::fabsl:
307            continue; // ISD::FABS is never a library call.
308          case LibFunc::sqrt:
309          case LibFunc::sqrtf:
310          case LibFunc::sqrtl:
311            Opcode = ISD::FSQRT; break;
312          case LibFunc::floor:
313          case LibFunc::floorf:
314          case LibFunc::floorl:
315            Opcode = ISD::FFLOOR; break;
316          case LibFunc::nearbyint:
317          case LibFunc::nearbyintf:
318          case LibFunc::nearbyintl:
319            Opcode = ISD::FNEARBYINT; break;
320          case LibFunc::ceil:
321          case LibFunc::ceilf:
322          case LibFunc::ceill:
323            Opcode = ISD::FCEIL; break;
324          case LibFunc::rint:
325          case LibFunc::rintf:
326          case LibFunc::rintl:
327            Opcode = ISD::FRINT; break;
328          case LibFunc::round:
329          case LibFunc::roundf:
330          case LibFunc::roundl:
331            Opcode = ISD::FROUND; break;
332          case LibFunc::trunc:
333          case LibFunc::truncf:
334          case LibFunc::truncl:
335            Opcode = ISD::FTRUNC; break;
336          }
337
338          MVT VTy =
339            TLI->getSimpleValueType(CI->getArgOperand(0)->getType(), true);
340          if (VTy == MVT::Other)
341            return true;
342
343          if (TLI->isOperationLegalOrCustom(Opcode, VTy))
344            continue;
345          else if (VTy.isVector() &&
346                   TLI->isOperationLegalOrCustom(Opcode, VTy.getScalarType()))
347            continue;
348
349          return true;
350        }
351      }
352
353      return true;
354    } else if (isa<BinaryOperator>(J) &&
355               J->getType()->getScalarType()->isPPC_FP128Ty()) {
356      // Most operations on ppc_f128 values become calls.
357      return true;
358    } else if (isa<UIToFPInst>(J) || isa<SIToFPInst>(J) ||
359               isa<FPToUIInst>(J) || isa<FPToSIInst>(J)) {
360      CastInst *CI = cast<CastInst>(J);
361      if (CI->getSrcTy()->getScalarType()->isPPC_FP128Ty() ||
362          CI->getDestTy()->getScalarType()->isPPC_FP128Ty() ||
363          isLargeIntegerTy(TT.isArch32Bit(), CI->getSrcTy()->getScalarType()) ||
364          isLargeIntegerTy(TT.isArch32Bit(), CI->getDestTy()->getScalarType()))
365        return true;
366    } else if (isLargeIntegerTy(TT.isArch32Bit(),
367                                J->getType()->getScalarType()) &&
368               (J->getOpcode() == Instruction::UDiv ||
369                J->getOpcode() == Instruction::SDiv ||
370                J->getOpcode() == Instruction::URem ||
371                J->getOpcode() == Instruction::SRem)) {
372      return true;
373    } else if (isa<IndirectBrInst>(J) || isa<InvokeInst>(J)) {
374      // On PowerPC, indirect jumps use the counter register.
375      return true;
376    } else if (SwitchInst *SI = dyn_cast<SwitchInst>(J)) {
377      if (!TM)
378        return true;
379      const TargetLowering *TLI = TM->getTargetLowering();
380
381      if (TLI->supportJumpTables() &&
382          SI->getNumCases()+1 >= (unsigned) TLI->getMinimumJumpTableEntries())
383        return true;
384    }
385  }
386
387  return false;
388}
389
390bool PPCCTRLoops::convertToCTRLoop(Loop *L) {
391  bool MadeChange = false;
392
393  Triple TT = Triple(L->getHeader()->getParent()->getParent()->
394                     getTargetTriple());
395  if (!TT.isArch32Bit() && !TT.isArch64Bit())
396    return MadeChange; // Unknown arch. type.
397
398  // Process nested loops first.
399  for (Loop::iterator I = L->begin(), E = L->end(); I != E; ++I) {
400    MadeChange |= convertToCTRLoop(*I);
401  }
402
403  // If a nested loop has been converted, then we can't convert this loop.
404  if (MadeChange)
405    return MadeChange;
406
407#ifndef NDEBUG
408  // Stop trying after reaching the limit (if any).
409  int Limit = CTRLoopLimit;
410  if (Limit >= 0) {
411    if (Counter >= CTRLoopLimit)
412      return false;
413    Counter++;
414  }
415#endif
416
417  // We don't want to spill/restore the counter register, and so we don't
418  // want to use the counter register if the loop contains calls.
419  for (Loop::block_iterator I = L->block_begin(), IE = L->block_end();
420       I != IE; ++I)
421    if (mightUseCTR(TT, *I))
422      return MadeChange;
423
424  SmallVector<BasicBlock*, 4> ExitingBlocks;
425  L->getExitingBlocks(ExitingBlocks);
426
427  BasicBlock *CountedExitBlock = 0;
428  const SCEV *ExitCount = 0;
429  BranchInst *CountedExitBranch = 0;
430  for (SmallVectorImpl<BasicBlock *>::iterator I = ExitingBlocks.begin(),
431       IE = ExitingBlocks.end(); I != IE; ++I) {
432    const SCEV *EC = SE->getExitCount(L, *I);
433    DEBUG(dbgs() << "Exit Count for " << *L << " from block " <<
434                    (*I)->getName() << ": " << *EC << "\n");
435    if (isa<SCEVCouldNotCompute>(EC))
436      continue;
437    if (const SCEVConstant *ConstEC = dyn_cast<SCEVConstant>(EC)) {
438      if (ConstEC->getValue()->isZero())
439        continue;
440    } else if (!SE->isLoopInvariant(EC, L))
441      continue;
442
443    if (SE->getTypeSizeInBits(EC->getType()) > (TT.isArch64Bit() ? 64 : 32))
444      continue;
445
446    // We now have a loop-invariant count of loop iterations (which is not the
447    // constant zero) for which we know that this loop will not exit via this
448    // exisiting block.
449
450    // We need to make sure that this block will run on every loop iteration.
451    // For this to be true, we must dominate all blocks with backedges. Such
452    // blocks are in-loop predecessors to the header block.
453    bool NotAlways = false;
454    for (pred_iterator PI = pred_begin(L->getHeader()),
455         PIE = pred_end(L->getHeader()); PI != PIE; ++PI) {
456      if (!L->contains(*PI))
457        continue;
458
459      if (!DT->dominates(*I, *PI)) {
460        NotAlways = true;
461        break;
462      }
463    }
464
465    if (NotAlways)
466      continue;
467
468    // Make sure this blocks ends with a conditional branch.
469    Instruction *TI = (*I)->getTerminator();
470    if (!TI)
471      continue;
472
473    if (BranchInst *BI = dyn_cast<BranchInst>(TI)) {
474      if (!BI->isConditional())
475        continue;
476
477      CountedExitBranch = BI;
478    } else
479      continue;
480
481    // Note that this block may not be the loop latch block, even if the loop
482    // has a latch block.
483    CountedExitBlock = *I;
484    ExitCount = EC;
485    break;
486  }
487
488  if (!CountedExitBlock)
489    return MadeChange;
490
491  BasicBlock *Preheader = L->getLoopPreheader();
492
493  // If we don't have a preheader, then insert one. If we already have a
494  // preheader, then we can use it (except if the preheader contains a use of
495  // the CTR register because some such uses might be reordered by the
496  // selection DAG after the mtctr instruction).
497  if (!Preheader || mightUseCTR(TT, Preheader))
498    Preheader = InsertPreheaderForLoop(L, this);
499  if (!Preheader)
500    return MadeChange;
501
502  DEBUG(dbgs() << "Preheader for exit count: " << Preheader->getName() << "\n");
503
504  // Insert the count into the preheader and replace the condition used by the
505  // selected branch.
506  MadeChange = true;
507
508  SCEVExpander SCEVE(*SE, "loopcnt");
509  LLVMContext &C = SE->getContext();
510  Type *CountType = TT.isArch64Bit() ? Type::getInt64Ty(C) :
511                                       Type::getInt32Ty(C);
512  if (!ExitCount->getType()->isPointerTy() &&
513      ExitCount->getType() != CountType)
514    ExitCount = SE->getZeroExtendExpr(ExitCount, CountType);
515  ExitCount = SE->getAddExpr(ExitCount,
516                             SE->getConstant(CountType, 1));
517  Value *ECValue = SCEVE.expandCodeFor(ExitCount, CountType,
518                                       Preheader->getTerminator());
519
520  IRBuilder<> CountBuilder(Preheader->getTerminator());
521  Module *M = Preheader->getParent()->getParent();
522  Value *MTCTRFunc = Intrinsic::getDeclaration(M, Intrinsic::ppc_mtctr,
523                                               CountType);
524  CountBuilder.CreateCall(MTCTRFunc, ECValue);
525
526  IRBuilder<> CondBuilder(CountedExitBranch);
527  Value *DecFunc =
528    Intrinsic::getDeclaration(M, Intrinsic::ppc_is_decremented_ctr_nonzero);
529  Value *NewCond = CondBuilder.CreateCall(DecFunc);
530  Value *OldCond = CountedExitBranch->getCondition();
531  CountedExitBranch->setCondition(NewCond);
532
533  // The false branch must exit the loop.
534  if (!L->contains(CountedExitBranch->getSuccessor(0)))
535    CountedExitBranch->swapSuccessors();
536
537  // The old condition may be dead now, and may have even created a dead PHI
538  // (the original induction variable).
539  RecursivelyDeleteTriviallyDeadInstructions(OldCond);
540  DeleteDeadPHIs(CountedExitBlock);
541
542  ++NumCTRLoops;
543  return MadeChange;
544}
545
546#ifndef NDEBUG
547static bool clobbersCTR(const MachineInstr *MI) {
548  for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
549    const MachineOperand &MO = MI->getOperand(i);
550    if (MO.isReg()) {
551      if (MO.isDef() && (MO.getReg() == PPC::CTR || MO.getReg() == PPC::CTR8))
552        return true;
553    } else if (MO.isRegMask()) {
554      if (MO.clobbersPhysReg(PPC::CTR) || MO.clobbersPhysReg(PPC::CTR8))
555        return true;
556    }
557  }
558
559  return false;
560}
561
562static bool verifyCTRBranch(MachineBasicBlock *MBB,
563                            MachineBasicBlock::iterator I) {
564  MachineBasicBlock::iterator BI = I;
565  SmallSet<MachineBasicBlock *, 16>   Visited;
566  SmallVector<MachineBasicBlock *, 8> Preds;
567  bool CheckPreds;
568
569  if (I == MBB->begin()) {
570    Visited.insert(MBB);
571    goto queue_preds;
572  } else
573    --I;
574
575check_block:
576  Visited.insert(MBB);
577  if (I == MBB->end())
578    goto queue_preds;
579
580  CheckPreds = true;
581  for (MachineBasicBlock::iterator IE = MBB->begin();; --I) {
582    unsigned Opc = I->getOpcode();
583    if (Opc == PPC::MTCTRloop || Opc == PPC::MTCTR8loop) {
584      CheckPreds = false;
585      break;
586    }
587
588    if (I != BI && clobbersCTR(I)) {
589      DEBUG(dbgs() << "BB#" << MBB->getNumber() << " (" <<
590                      MBB->getFullName() << ") instruction " << *I <<
591                      " clobbers CTR, invalidating " << "BB#" <<
592                      BI->getParent()->getNumber() << " (" <<
593                      BI->getParent()->getFullName() << ") instruction " <<
594                      *BI << "\n");
595      return false;
596    }
597
598    if (I == IE)
599      break;
600  }
601
602  if (!CheckPreds && Preds.empty())
603    return true;
604
605  if (CheckPreds) {
606queue_preds:
607    if (MachineFunction::iterator(MBB) == MBB->getParent()->begin()) {
608      DEBUG(dbgs() << "Unable to find a MTCTR instruction for BB#" <<
609                      BI->getParent()->getNumber() << " (" <<
610                      BI->getParent()->getFullName() << ") instruction " <<
611                      *BI << "\n");
612      return false;
613    }
614
615    for (MachineBasicBlock::pred_iterator PI = MBB->pred_begin(),
616         PIE = MBB->pred_end(); PI != PIE; ++PI)
617      Preds.push_back(*PI);
618  }
619
620  do {
621    MBB = Preds.pop_back_val();
622    if (!Visited.count(MBB)) {
623      I = MBB->getLastNonDebugInstr();
624      goto check_block;
625    }
626  } while (!Preds.empty());
627
628  return true;
629}
630
631bool PPCCTRLoopsVerify::runOnMachineFunction(MachineFunction &MF) {
632  MDT = &getAnalysis<MachineDominatorTree>();
633
634  // Verify that all bdnz/bdz instructions are dominated by a loop mtctr before
635  // any other instructions that might clobber the ctr register.
636  for (MachineFunction::iterator I = MF.begin(), IE = MF.end();
637       I != IE; ++I) {
638    MachineBasicBlock *MBB = I;
639    if (!MDT->isReachableFromEntry(MBB))
640      continue;
641
642    for (MachineBasicBlock::iterator MII = MBB->getFirstTerminator(),
643      MIIE = MBB->end(); MII != MIIE; ++MII) {
644      unsigned Opc = MII->getOpcode();
645      if (Opc == PPC::BDNZ8 || Opc == PPC::BDNZ ||
646          Opc == PPC::BDZ8  || Opc == PPC::BDZ)
647        if (!verifyCTRBranch(MBB, MII))
648          llvm_unreachable("Invalid PPC CTR loop!");
649    }
650  }
651
652  return false;
653}
654#endif // NDEBUG
655
656