1 //===- SimplifyCFG.cpp - Code to perform CFG simplification ---------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // Peephole optimize the CFG.
12 //===----------------------------------------------------------------------===//
14 #include "llvm/Transforms/Utils/Local.h"
15 #include "llvm/ADT/DenseMap.h"
16 #include "llvm/ADT/STLExtras.h"
17 #include "llvm/ADT/SetVector.h"
18 #include "llvm/ADT/SmallPtrSet.h"
19 #include "llvm/ADT/SmallVector.h"
20 #include "llvm/ADT/Statistic.h"
21 #include "llvm/Analysis/ConstantFolding.h"
22 #include "llvm/Analysis/InstructionSimplify.h"
23 #include "llvm/Analysis/TargetTransformInfo.h"
24 #include "llvm/Analysis/ValueTracking.h"
25 #include "llvm/IR/CFG.h"
26 #include "llvm/IR/ConstantRange.h"
27 #include "llvm/IR/Constants.h"
28 #include "llvm/IR/DataLayout.h"
29 #include "llvm/IR/DerivedTypes.h"
30 #include "llvm/IR/GlobalVariable.h"
31 #include "llvm/IR/IRBuilder.h"
32 #include "llvm/IR/Instructions.h"
33 #include "llvm/IR/IntrinsicInst.h"
34 #include "llvm/IR/LLVMContext.h"
35 #include "llvm/IR/MDBuilder.h"
36 #include "llvm/IR/Metadata.h"
37 #include "llvm/IR/Module.h"
38 #include "llvm/IR/NoFolder.h"
39 #include "llvm/IR/Operator.h"
40 #include "llvm/IR/PatternMatch.h"
41 #include "llvm/IR/Type.h"
42 #include "llvm/Support/CommandLine.h"
43 #include "llvm/Support/Debug.h"
44 #include "llvm/Support/raw_ostream.h"
45 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
50 using namespace PatternMatch;
52 #define DEBUG_TYPE "simplifycfg"
54 static cl::opt<unsigned>
55 PHINodeFoldingThreshold("phi-node-folding-threshold", cl::Hidden, cl::init(1),
56 cl::desc("Control the amount of phi node folding to perform (default = 1)"));
59 DupRet("simplifycfg-dup-ret", cl::Hidden, cl::init(false),
60 cl::desc("Duplicate return instructions into unconditional branches"));
63 SinkCommon("simplifycfg-sink-common", cl::Hidden, cl::init(true),
64 cl::desc("Sink common instructions down to the end block"));
66 static cl::opt<bool> HoistCondStores(
67 "simplifycfg-hoist-cond-stores", cl::Hidden, cl::init(true),
68 cl::desc("Hoist conditional stores if an unconditional store precedes"));
70 STATISTIC(NumBitMaps, "Number of switch instructions turned into bitmaps");
71 STATISTIC(NumLookupTables, "Number of switch instructions turned into lookup tables");
72 STATISTIC(NumLookupTablesHoles, "Number of switch instructions turned into lookup tables (holes checked)");
73 STATISTIC(NumSinkCommons, "Number of common instructions sunk down to the end block");
74 STATISTIC(NumSpeculations, "Number of speculative executed instructions");
77 /// ValueEqualityComparisonCase - Represents a case of a switch.
78 struct ValueEqualityComparisonCase {
82 ValueEqualityComparisonCase(ConstantInt *Value, BasicBlock *Dest)
83 : Value(Value), Dest(Dest) {}
85 bool operator<(ValueEqualityComparisonCase RHS) const {
86 // Comparing pointers is ok as we only rely on the order for uniquing.
87 return Value < RHS.Value;
90 bool operator==(BasicBlock *RHSDest) const { return Dest == RHSDest; }
93 class SimplifyCFGOpt {
94 const TargetTransformInfo &TTI;
95 const DataLayout *const DL;
96 Value *isValueEqualityComparison(TerminatorInst *TI);
97 BasicBlock *GetValueEqualityComparisonCases(TerminatorInst *TI,
98 std::vector<ValueEqualityComparisonCase> &Cases);
99 bool SimplifyEqualityComparisonWithOnlyPredecessor(TerminatorInst *TI,
101 IRBuilder<> &Builder);
102 bool FoldValueComparisonIntoPredecessors(TerminatorInst *TI,
103 IRBuilder<> &Builder);
105 bool SimplifyReturn(ReturnInst *RI, IRBuilder<> &Builder);
106 bool SimplifyResume(ResumeInst *RI, IRBuilder<> &Builder);
107 bool SimplifyUnreachable(UnreachableInst *UI);
108 bool SimplifySwitch(SwitchInst *SI, IRBuilder<> &Builder);
109 bool SimplifyIndirectBr(IndirectBrInst *IBI);
110 bool SimplifyUncondBranch(BranchInst *BI, IRBuilder <> &Builder);
111 bool SimplifyCondBranch(BranchInst *BI, IRBuilder <>&Builder);
114 SimplifyCFGOpt(const TargetTransformInfo &TTI, const DataLayout *DL)
115 : TTI(TTI), DL(DL) {}
116 bool run(BasicBlock *BB);
120 /// SafeToMergeTerminators - Return true if it is safe to merge these two
121 /// terminator instructions together.
123 static bool SafeToMergeTerminators(TerminatorInst *SI1, TerminatorInst *SI2) {
124 if (SI1 == SI2) return false; // Can't merge with self!
126 // It is not safe to merge these two switch instructions if they have a common
127 // successor, and if that successor has a PHI node, and if *that* PHI node has
128 // conflicting incoming values from the two switch blocks.
129 BasicBlock *SI1BB = SI1->getParent();
130 BasicBlock *SI2BB = SI2->getParent();
131 SmallPtrSet<BasicBlock*, 16> SI1Succs(succ_begin(SI1BB), succ_end(SI1BB));
133 for (succ_iterator I = succ_begin(SI2BB), E = succ_end(SI2BB); I != E; ++I)
134 if (SI1Succs.count(*I))
135 for (BasicBlock::iterator BBI = (*I)->begin();
136 isa<PHINode>(BBI); ++BBI) {
137 PHINode *PN = cast<PHINode>(BBI);
138 if (PN->getIncomingValueForBlock(SI1BB) !=
139 PN->getIncomingValueForBlock(SI2BB))
146 /// isProfitableToFoldUnconditional - Return true if it is safe and profitable
147 /// to merge these two terminator instructions together, where SI1 is an
148 /// unconditional branch. PhiNodes will store all PHI nodes in common
151 static bool isProfitableToFoldUnconditional(BranchInst *SI1,
154 SmallVectorImpl<PHINode*> &PhiNodes) {
155 if (SI1 == SI2) return false; // Can't merge with self!
156 assert(SI1->isUnconditional() && SI2->isConditional());
158 // We fold the unconditional branch if we can easily update all PHI nodes in
159 // common successors:
160 // 1> We have a constant incoming value for the conditional branch;
161 // 2> We have "Cond" as the incoming value for the unconditional branch;
162 // 3> SI2->getCondition() and Cond have same operands.
163 CmpInst *Ci2 = dyn_cast<CmpInst>(SI2->getCondition());
164 if (!Ci2) return false;
165 if (!(Cond->getOperand(0) == Ci2->getOperand(0) &&
166 Cond->getOperand(1) == Ci2->getOperand(1)) &&
167 !(Cond->getOperand(0) == Ci2->getOperand(1) &&
168 Cond->getOperand(1) == Ci2->getOperand(0)))
171 BasicBlock *SI1BB = SI1->getParent();
172 BasicBlock *SI2BB = SI2->getParent();
173 SmallPtrSet<BasicBlock*, 16> SI1Succs(succ_begin(SI1BB), succ_end(SI1BB));
174 for (succ_iterator I = succ_begin(SI2BB), E = succ_end(SI2BB); I != E; ++I)
175 if (SI1Succs.count(*I))
176 for (BasicBlock::iterator BBI = (*I)->begin();
177 isa<PHINode>(BBI); ++BBI) {
178 PHINode *PN = cast<PHINode>(BBI);
179 if (PN->getIncomingValueForBlock(SI1BB) != Cond ||
180 !isa<ConstantInt>(PN->getIncomingValueForBlock(SI2BB)))
182 PhiNodes.push_back(PN);
187 /// AddPredecessorToBlock - Update PHI nodes in Succ to indicate that there will
188 /// now be entries in it from the 'NewPred' block. The values that will be
189 /// flowing into the PHI nodes will be the same as those coming in from
190 /// ExistPred, an existing predecessor of Succ.
191 static void AddPredecessorToBlock(BasicBlock *Succ, BasicBlock *NewPred,
192 BasicBlock *ExistPred) {
193 if (!isa<PHINode>(Succ->begin())) return; // Quick exit if nothing to do
196 for (BasicBlock::iterator I = Succ->begin();
197 (PN = dyn_cast<PHINode>(I)); ++I)
198 PN->addIncoming(PN->getIncomingValueForBlock(ExistPred), NewPred);
201 /// ComputeSpeculationCost - Compute an abstract "cost" of speculating the
202 /// given instruction, which is assumed to be safe to speculate. 1 means
203 /// cheap, 2 means less cheap, and UINT_MAX means prohibitively expensive.
204 static unsigned ComputeSpeculationCost(const User *I) {
205 assert(isSafeToSpeculativelyExecute(I) &&
206 "Instruction is not safe to speculatively execute!");
207 switch (Operator::getOpcode(I)) {
209 // In doubt, be conservative.
211 case Instruction::GetElementPtr:
212 // GEPs are cheap if all indices are constant.
213 if (!cast<GEPOperator>(I)->hasAllConstantIndices())
216 case Instruction::ExtractValue:
217 case Instruction::Load:
218 case Instruction::Add:
219 case Instruction::Sub:
220 case Instruction::And:
221 case Instruction::Or:
222 case Instruction::Xor:
223 case Instruction::Shl:
224 case Instruction::LShr:
225 case Instruction::AShr:
226 case Instruction::ICmp:
227 case Instruction::Trunc:
228 case Instruction::ZExt:
229 case Instruction::SExt:
230 return 1; // These are all cheap.
232 case Instruction::Call:
233 case Instruction::Select:
238 /// DominatesMergePoint - If we have a merge point of an "if condition" as
239 /// accepted above, return true if the specified value dominates the block. We
240 /// don't handle the true generality of domination here, just a special case
241 /// which works well enough for us.
243 /// If AggressiveInsts is non-null, and if V does not dominate BB, we check to
244 /// see if V (which must be an instruction) and its recursive operands
245 /// that do not dominate BB have a combined cost lower than CostRemaining and
246 /// are non-trapping. If both are true, the instruction is inserted into the
247 /// set and true is returned.
249 /// The cost for most non-trapping instructions is defined as 1 except for
250 /// Select whose cost is 2.
252 /// After this function returns, CostRemaining is decreased by the cost of
253 /// V plus its non-dominating operands. If that cost is greater than
254 /// CostRemaining, false is returned and CostRemaining is undefined.
255 static bool DominatesMergePoint(Value *V, BasicBlock *BB,
256 SmallPtrSet<Instruction*, 4> *AggressiveInsts,
257 unsigned &CostRemaining) {
258 Instruction *I = dyn_cast<Instruction>(V);
260 // Non-instructions all dominate instructions, but not all constantexprs
261 // can be executed unconditionally.
262 if (ConstantExpr *C = dyn_cast<ConstantExpr>(V))
267 BasicBlock *PBB = I->getParent();
269 // We don't want to allow weird loops that might have the "if condition" in
270 // the bottom of this block.
271 if (PBB == BB) return false;
273 // If this instruction is defined in a block that contains an unconditional
274 // branch to BB, then it must be in the 'conditional' part of the "if
275 // statement". If not, it definitely dominates the region.
276 BranchInst *BI = dyn_cast<BranchInst>(PBB->getTerminator());
277 if (!BI || BI->isConditional() || BI->getSuccessor(0) != BB)
280 // If we aren't allowing aggressive promotion anymore, then don't consider
281 // instructions in the 'if region'.
282 if (!AggressiveInsts) return false;
284 // If we have seen this instruction before, don't count it again.
285 if (AggressiveInsts->count(I)) return true;
287 // Okay, it looks like the instruction IS in the "condition". Check to
288 // see if it's a cheap instruction to unconditionally compute, and if it
289 // only uses stuff defined outside of the condition. If so, hoist it out.
290 if (!isSafeToSpeculativelyExecute(I))
293 unsigned Cost = ComputeSpeculationCost(I);
295 if (Cost > CostRemaining)
298 CostRemaining -= Cost;
300 // Okay, we can only really hoist these out if their operands do
301 // not take us over the cost threshold.
302 for (User::op_iterator i = I->op_begin(), e = I->op_end(); i != e; ++i)
303 if (!DominatesMergePoint(*i, BB, AggressiveInsts, CostRemaining))
305 // Okay, it's safe to do this! Remember this instruction.
306 AggressiveInsts->insert(I);
310 /// GetConstantInt - Extract ConstantInt from value, looking through IntToPtr
311 /// and PointerNullValue. Return NULL if value is not a constant int.
312 static ConstantInt *GetConstantInt(Value *V, const DataLayout *DL) {
313 // Normal constant int.
314 ConstantInt *CI = dyn_cast<ConstantInt>(V);
315 if (CI || !DL || !isa<Constant>(V) || !V->getType()->isPointerTy())
318 // This is some kind of pointer constant. Turn it into a pointer-sized
319 // ConstantInt if possible.
320 IntegerType *PtrTy = cast<IntegerType>(DL->getIntPtrType(V->getType()));
322 // Null pointer means 0, see SelectionDAGBuilder::getValue(const Value*).
323 if (isa<ConstantPointerNull>(V))
324 return ConstantInt::get(PtrTy, 0);
326 // IntToPtr const int.
327 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V))
328 if (CE->getOpcode() == Instruction::IntToPtr)
329 if (ConstantInt *CI = dyn_cast<ConstantInt>(CE->getOperand(0))) {
330 // The constant is very likely to have the right type already.
331 if (CI->getType() == PtrTy)
334 return cast<ConstantInt>
335 (ConstantExpr::getIntegerCast(CI, PtrTy, /*isSigned=*/false));
340 /// GatherConstantCompares - Given a potentially 'or'd or 'and'd together
341 /// collection of icmp eq/ne instructions that compare a value against a
342 /// constant, return the value being compared, and stick the constant into the
345 GatherConstantCompares(Value *V, std::vector<ConstantInt*> &Vals, Value *&Extra,
346 const DataLayout *DL, bool isEQ, unsigned &UsedICmps) {
347 Instruction *I = dyn_cast<Instruction>(V);
348 if (!I) return nullptr;
350 // If this is an icmp against a constant, handle this as one of the cases.
351 if (ICmpInst *ICI = dyn_cast<ICmpInst>(I)) {
352 if (ConstantInt *C = GetConstantInt(I->getOperand(1), DL)) {
356 if (ICI->getPredicate() == (isEQ ? ICmpInst::ICMP_EQ:ICmpInst::ICMP_NE)) {
357 // (x & ~2^x) == y --> x == y || x == y|2^x
358 // This undoes a transformation done by instcombine to fuse 2 compares.
359 if (match(ICI->getOperand(0),
360 m_And(m_Value(RHSVal), m_ConstantInt(RHSC)))) {
361 APInt Not = ~RHSC->getValue();
362 if (Not.isPowerOf2()) {
365 ConstantInt::get(C->getContext(), C->getValue() | Not));
373 return I->getOperand(0);
376 // If we have "x ult 3" comparison, for example, then we can add 0,1,2 to
379 ConstantRange::makeICmpRegion(ICI->getPredicate(), C->getValue());
381 // Shift the range if the compare is fed by an add. This is the range
382 // compare idiom as emitted by instcombine.
384 match(I->getOperand(0), m_Add(m_Value(RHSVal), m_ConstantInt(RHSC)));
386 Span = Span.subtract(RHSC->getValue());
388 // If this is an and/!= check then we want to optimize "x ugt 2" into
391 Span = Span.inverse();
393 // If there are a ton of values, we don't want to make a ginormous switch.
394 if (Span.getSetSize().ugt(8) || Span.isEmptySet())
397 for (APInt Tmp = Span.getLower(); Tmp != Span.getUpper(); ++Tmp)
398 Vals.push_back(ConstantInt::get(V->getContext(), Tmp));
400 return hasAdd ? RHSVal : I->getOperand(0);
405 // Otherwise, we can only handle an | or &, depending on isEQ.
406 if (I->getOpcode() != (isEQ ? Instruction::Or : Instruction::And))
409 unsigned NumValsBeforeLHS = Vals.size();
410 unsigned UsedICmpsBeforeLHS = UsedICmps;
411 if (Value *LHS = GatherConstantCompares(I->getOperand(0), Vals, Extra, DL,
413 unsigned NumVals = Vals.size();
414 unsigned UsedICmpsBeforeRHS = UsedICmps;
415 if (Value *RHS = GatherConstantCompares(I->getOperand(1), Vals, Extra, DL,
419 Vals.resize(NumVals);
420 UsedICmps = UsedICmpsBeforeRHS;
423 // The RHS of the or/and can't be folded in and we haven't used "Extra" yet,
424 // set it and return success.
425 if (Extra == nullptr || Extra == I->getOperand(1)) {
426 Extra = I->getOperand(1);
430 Vals.resize(NumValsBeforeLHS);
431 UsedICmps = UsedICmpsBeforeLHS;
435 // If the LHS can't be folded in, but Extra is available and RHS can, try to
437 if (Extra == nullptr || Extra == I->getOperand(0)) {
438 Value *OldExtra = Extra;
439 Extra = I->getOperand(0);
440 if (Value *RHS = GatherConstantCompares(I->getOperand(1), Vals, Extra, DL,
443 assert(Vals.size() == NumValsBeforeLHS);
450 static void EraseTerminatorInstAndDCECond(TerminatorInst *TI) {
451 Instruction *Cond = nullptr;
452 if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) {
453 Cond = dyn_cast<Instruction>(SI->getCondition());
454 } else if (BranchInst *BI = dyn_cast<BranchInst>(TI)) {
455 if (BI->isConditional())
456 Cond = dyn_cast<Instruction>(BI->getCondition());
457 } else if (IndirectBrInst *IBI = dyn_cast<IndirectBrInst>(TI)) {
458 Cond = dyn_cast<Instruction>(IBI->getAddress());
461 TI->eraseFromParent();
462 if (Cond) RecursivelyDeleteTriviallyDeadInstructions(Cond);
465 /// isValueEqualityComparison - Return true if the specified terminator checks
466 /// to see if a value is equal to constant integer value.
467 Value *SimplifyCFGOpt::isValueEqualityComparison(TerminatorInst *TI) {
469 if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) {
470 // Do not permit merging of large switch instructions into their
471 // predecessors unless there is only one predecessor.
472 if (SI->getNumSuccessors()*std::distance(pred_begin(SI->getParent()),
473 pred_end(SI->getParent())) <= 128)
474 CV = SI->getCondition();
475 } else if (BranchInst *BI = dyn_cast<BranchInst>(TI))
476 if (BI->isConditional() && BI->getCondition()->hasOneUse())
477 if (ICmpInst *ICI = dyn_cast<ICmpInst>(BI->getCondition()))
478 if (ICI->isEquality() && GetConstantInt(ICI->getOperand(1), DL))
479 CV = ICI->getOperand(0);
481 // Unwrap any lossless ptrtoint cast.
483 if (PtrToIntInst *PTII = dyn_cast<PtrToIntInst>(CV)) {
484 Value *Ptr = PTII->getPointerOperand();
485 if (PTII->getType() == DL->getIntPtrType(Ptr->getType()))
492 /// GetValueEqualityComparisonCases - Given a value comparison instruction,
493 /// decode all of the 'cases' that it represents and return the 'default' block.
494 BasicBlock *SimplifyCFGOpt::
495 GetValueEqualityComparisonCases(TerminatorInst *TI,
496 std::vector<ValueEqualityComparisonCase>
498 if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) {
499 Cases.reserve(SI->getNumCases());
500 for (SwitchInst::CaseIt i = SI->case_begin(), e = SI->case_end(); i != e; ++i)
501 Cases.push_back(ValueEqualityComparisonCase(i.getCaseValue(),
502 i.getCaseSuccessor()));
503 return SI->getDefaultDest();
506 BranchInst *BI = cast<BranchInst>(TI);
507 ICmpInst *ICI = cast<ICmpInst>(BI->getCondition());
508 BasicBlock *Succ = BI->getSuccessor(ICI->getPredicate() == ICmpInst::ICMP_NE);
509 Cases.push_back(ValueEqualityComparisonCase(GetConstantInt(ICI->getOperand(1),
512 return BI->getSuccessor(ICI->getPredicate() == ICmpInst::ICMP_EQ);
516 /// EliminateBlockCases - Given a vector of bb/value pairs, remove any entries
517 /// in the list that match the specified block.
518 static void EliminateBlockCases(BasicBlock *BB,
519 std::vector<ValueEqualityComparisonCase> &Cases) {
520 Cases.erase(std::remove(Cases.begin(), Cases.end(), BB), Cases.end());
523 /// ValuesOverlap - Return true if there are any keys in C1 that exist in C2 as
526 ValuesOverlap(std::vector<ValueEqualityComparisonCase> &C1,
527 std::vector<ValueEqualityComparisonCase > &C2) {
528 std::vector<ValueEqualityComparisonCase> *V1 = &C1, *V2 = &C2;
530 // Make V1 be smaller than V2.
531 if (V1->size() > V2->size())
534 if (V1->size() == 0) return false;
535 if (V1->size() == 1) {
537 ConstantInt *TheVal = (*V1)[0].Value;
538 for (unsigned i = 0, e = V2->size(); i != e; ++i)
539 if (TheVal == (*V2)[i].Value)
543 // Otherwise, just sort both lists and compare element by element.
544 array_pod_sort(V1->begin(), V1->end());
545 array_pod_sort(V2->begin(), V2->end());
546 unsigned i1 = 0, i2 = 0, e1 = V1->size(), e2 = V2->size();
547 while (i1 != e1 && i2 != e2) {
548 if ((*V1)[i1].Value == (*V2)[i2].Value)
550 if ((*V1)[i1].Value < (*V2)[i2].Value)
558 /// SimplifyEqualityComparisonWithOnlyPredecessor - If TI is known to be a
559 /// terminator instruction and its block is known to only have a single
560 /// predecessor block, check to see if that predecessor is also a value
561 /// comparison with the same value, and if that comparison determines the
562 /// outcome of this comparison. If so, simplify TI. This does a very limited
563 /// form of jump threading.
564 bool SimplifyCFGOpt::
565 SimplifyEqualityComparisonWithOnlyPredecessor(TerminatorInst *TI,
567 IRBuilder<> &Builder) {
568 Value *PredVal = isValueEqualityComparison(Pred->getTerminator());
569 if (!PredVal) return false; // Not a value comparison in predecessor.
571 Value *ThisVal = isValueEqualityComparison(TI);
572 assert(ThisVal && "This isn't a value comparison!!");
573 if (ThisVal != PredVal) return false; // Different predicates.
575 // TODO: Preserve branch weight metadata, similarly to how
576 // FoldValueComparisonIntoPredecessors preserves it.
578 // Find out information about when control will move from Pred to TI's block.
579 std::vector<ValueEqualityComparisonCase> PredCases;
580 BasicBlock *PredDef = GetValueEqualityComparisonCases(Pred->getTerminator(),
582 EliminateBlockCases(PredDef, PredCases); // Remove default from cases.
584 // Find information about how control leaves this block.
585 std::vector<ValueEqualityComparisonCase> ThisCases;
586 BasicBlock *ThisDef = GetValueEqualityComparisonCases(TI, ThisCases);
587 EliminateBlockCases(ThisDef, ThisCases); // Remove default from cases.
589 // If TI's block is the default block from Pred's comparison, potentially
590 // simplify TI based on this knowledge.
591 if (PredDef == TI->getParent()) {
592 // If we are here, we know that the value is none of those cases listed in
593 // PredCases. If there are any cases in ThisCases that are in PredCases, we
595 if (!ValuesOverlap(PredCases, ThisCases))
598 if (isa<BranchInst>(TI)) {
599 // Okay, one of the successors of this condbr is dead. Convert it to a
601 assert(ThisCases.size() == 1 && "Branch can only have one case!");
602 // Insert the new branch.
603 Instruction *NI = Builder.CreateBr(ThisDef);
606 // Remove PHI node entries for the dead edge.
607 ThisCases[0].Dest->removePredecessor(TI->getParent());
609 DEBUG(dbgs() << "Threading pred instr: " << *Pred->getTerminator()
610 << "Through successor TI: " << *TI << "Leaving: " << *NI << "\n");
612 EraseTerminatorInstAndDCECond(TI);
616 SwitchInst *SI = cast<SwitchInst>(TI);
617 // Okay, TI has cases that are statically dead, prune them away.
618 SmallPtrSet<Constant*, 16> DeadCases;
619 for (unsigned i = 0, e = PredCases.size(); i != e; ++i)
620 DeadCases.insert(PredCases[i].Value);
622 DEBUG(dbgs() << "Threading pred instr: " << *Pred->getTerminator()
623 << "Through successor TI: " << *TI);
625 // Collect branch weights into a vector.
626 SmallVector<uint32_t, 8> Weights;
627 MDNode* MD = SI->getMetadata(LLVMContext::MD_prof);
628 bool HasWeight = MD && (MD->getNumOperands() == 2 + SI->getNumCases());
630 for (unsigned MD_i = 1, MD_e = MD->getNumOperands(); MD_i < MD_e;
632 ConstantInt* CI = dyn_cast<ConstantInt>(MD->getOperand(MD_i));
634 Weights.push_back(CI->getValue().getZExtValue());
636 for (SwitchInst::CaseIt i = SI->case_end(), e = SI->case_begin(); i != e;) {
638 if (DeadCases.count(i.getCaseValue())) {
640 std::swap(Weights[i.getCaseIndex()+1], Weights.back());
643 i.getCaseSuccessor()->removePredecessor(TI->getParent());
647 if (HasWeight && Weights.size() >= 2)
648 SI->setMetadata(LLVMContext::MD_prof,
649 MDBuilder(SI->getParent()->getContext()).
650 createBranchWeights(Weights));
652 DEBUG(dbgs() << "Leaving: " << *TI << "\n");
656 // Otherwise, TI's block must correspond to some matched value. Find out
657 // which value (or set of values) this is.
658 ConstantInt *TIV = nullptr;
659 BasicBlock *TIBB = TI->getParent();
660 for (unsigned i = 0, e = PredCases.size(); i != e; ++i)
661 if (PredCases[i].Dest == TIBB) {
663 return false; // Cannot handle multiple values coming to this block.
664 TIV = PredCases[i].Value;
666 assert(TIV && "No edge from pred to succ?");
668 // Okay, we found the one constant that our value can be if we get into TI's
669 // BB. Find out which successor will unconditionally be branched to.
670 BasicBlock *TheRealDest = nullptr;
671 for (unsigned i = 0, e = ThisCases.size(); i != e; ++i)
672 if (ThisCases[i].Value == TIV) {
673 TheRealDest = ThisCases[i].Dest;
677 // If not handled by any explicit cases, it is handled by the default case.
678 if (!TheRealDest) TheRealDest = ThisDef;
680 // Remove PHI node entries for dead edges.
681 BasicBlock *CheckEdge = TheRealDest;
682 for (succ_iterator SI = succ_begin(TIBB), e = succ_end(TIBB); SI != e; ++SI)
683 if (*SI != CheckEdge)
684 (*SI)->removePredecessor(TIBB);
688 // Insert the new branch.
689 Instruction *NI = Builder.CreateBr(TheRealDest);
692 DEBUG(dbgs() << "Threading pred instr: " << *Pred->getTerminator()
693 << "Through successor TI: " << *TI << "Leaving: " << *NI << "\n");
695 EraseTerminatorInstAndDCECond(TI);
700 /// ConstantIntOrdering - This class implements a stable ordering of constant
701 /// integers that does not depend on their address. This is important for
702 /// applications that sort ConstantInt's to ensure uniqueness.
703 struct ConstantIntOrdering {
704 bool operator()(const ConstantInt *LHS, const ConstantInt *RHS) const {
705 return LHS->getValue().ult(RHS->getValue());
710 static int ConstantIntSortPredicate(ConstantInt *const *P1,
711 ConstantInt *const *P2) {
712 const ConstantInt *LHS = *P1;
713 const ConstantInt *RHS = *P2;
714 if (LHS->getValue().ult(RHS->getValue()))
716 if (LHS->getValue() == RHS->getValue())
721 static inline bool HasBranchWeights(const Instruction* I) {
722 MDNode* ProfMD = I->getMetadata(LLVMContext::MD_prof);
723 if (ProfMD && ProfMD->getOperand(0))
724 if (MDString* MDS = dyn_cast<MDString>(ProfMD->getOperand(0)))
725 return MDS->getString().equals("branch_weights");
730 /// Get Weights of a given TerminatorInst, the default weight is at the front
731 /// of the vector. If TI is a conditional eq, we need to swap the branch-weight
733 static void GetBranchWeights(TerminatorInst *TI,
734 SmallVectorImpl<uint64_t> &Weights) {
735 MDNode* MD = TI->getMetadata(LLVMContext::MD_prof);
737 for (unsigned i = 1, e = MD->getNumOperands(); i < e; ++i) {
738 ConstantInt *CI = cast<ConstantInt>(MD->getOperand(i));
739 Weights.push_back(CI->getValue().getZExtValue());
742 // If TI is a conditional eq, the default case is the false case,
743 // and the corresponding branch-weight data is at index 2. We swap the
744 // default weight to be the first entry.
745 if (BranchInst* BI = dyn_cast<BranchInst>(TI)) {
746 assert(Weights.size() == 2);
747 ICmpInst *ICI = cast<ICmpInst>(BI->getCondition());
748 if (ICI->getPredicate() == ICmpInst::ICMP_EQ)
749 std::swap(Weights.front(), Weights.back());
753 /// Keep halving the weights until all can fit in uint32_t.
754 static void FitWeights(MutableArrayRef<uint64_t> Weights) {
755 uint64_t Max = *std::max_element(Weights.begin(), Weights.end());
756 if (Max > UINT_MAX) {
757 unsigned Offset = 32 - countLeadingZeros(Max);
758 for (uint64_t &I : Weights)
763 /// FoldValueComparisonIntoPredecessors - The specified terminator is a value
764 /// equality comparison instruction (either a switch or a branch on "X == c").
765 /// See if any of the predecessors of the terminator block are value comparisons
766 /// on the same value. If so, and if safe to do so, fold them together.
767 bool SimplifyCFGOpt::FoldValueComparisonIntoPredecessors(TerminatorInst *TI,
768 IRBuilder<> &Builder) {
769 BasicBlock *BB = TI->getParent();
770 Value *CV = isValueEqualityComparison(TI); // CondVal
771 assert(CV && "Not a comparison?");
772 bool Changed = false;
774 SmallVector<BasicBlock*, 16> Preds(pred_begin(BB), pred_end(BB));
775 while (!Preds.empty()) {
776 BasicBlock *Pred = Preds.pop_back_val();
778 // See if the predecessor is a comparison with the same value.
779 TerminatorInst *PTI = Pred->getTerminator();
780 Value *PCV = isValueEqualityComparison(PTI); // PredCondVal
782 if (PCV == CV && SafeToMergeTerminators(TI, PTI)) {
783 // Figure out which 'cases' to copy from SI to PSI.
784 std::vector<ValueEqualityComparisonCase> BBCases;
785 BasicBlock *BBDefault = GetValueEqualityComparisonCases(TI, BBCases);
787 std::vector<ValueEqualityComparisonCase> PredCases;
788 BasicBlock *PredDefault = GetValueEqualityComparisonCases(PTI, PredCases);
790 // Based on whether the default edge from PTI goes to BB or not, fill in
791 // PredCases and PredDefault with the new switch cases we would like to
793 SmallVector<BasicBlock*, 8> NewSuccessors;
795 // Update the branch weight metadata along the way
796 SmallVector<uint64_t, 8> Weights;
797 bool PredHasWeights = HasBranchWeights(PTI);
798 bool SuccHasWeights = HasBranchWeights(TI);
800 if (PredHasWeights) {
801 GetBranchWeights(PTI, Weights);
802 // branch-weight metadata is inconsistent here.
803 if (Weights.size() != 1 + PredCases.size())
804 PredHasWeights = SuccHasWeights = false;
805 } else if (SuccHasWeights)
806 // If there are no predecessor weights but there are successor weights,
807 // populate Weights with 1, which will later be scaled to the sum of
808 // successor's weights
809 Weights.assign(1 + PredCases.size(), 1);
811 SmallVector<uint64_t, 8> SuccWeights;
812 if (SuccHasWeights) {
813 GetBranchWeights(TI, SuccWeights);
814 // branch-weight metadata is inconsistent here.
815 if (SuccWeights.size() != 1 + BBCases.size())
816 PredHasWeights = SuccHasWeights = false;
817 } else if (PredHasWeights)
818 SuccWeights.assign(1 + BBCases.size(), 1);
820 if (PredDefault == BB) {
821 // If this is the default destination from PTI, only the edges in TI
822 // that don't occur in PTI, or that branch to BB will be activated.
823 std::set<ConstantInt*, ConstantIntOrdering> PTIHandled;
824 for (unsigned i = 0, e = PredCases.size(); i != e; ++i)
825 if (PredCases[i].Dest != BB)
826 PTIHandled.insert(PredCases[i].Value);
828 // The default destination is BB, we don't need explicit targets.
829 std::swap(PredCases[i], PredCases.back());
831 if (PredHasWeights || SuccHasWeights) {
832 // Increase weight for the default case.
833 Weights[0] += Weights[i+1];
834 std::swap(Weights[i+1], Weights.back());
838 PredCases.pop_back();
842 // Reconstruct the new switch statement we will be building.
843 if (PredDefault != BBDefault) {
844 PredDefault->removePredecessor(Pred);
845 PredDefault = BBDefault;
846 NewSuccessors.push_back(BBDefault);
849 unsigned CasesFromPred = Weights.size();
850 uint64_t ValidTotalSuccWeight = 0;
851 for (unsigned i = 0, e = BBCases.size(); i != e; ++i)
852 if (!PTIHandled.count(BBCases[i].Value) &&
853 BBCases[i].Dest != BBDefault) {
854 PredCases.push_back(BBCases[i]);
855 NewSuccessors.push_back(BBCases[i].Dest);
856 if (SuccHasWeights || PredHasWeights) {
857 // The default weight is at index 0, so weight for the ith case
858 // should be at index i+1. Scale the cases from successor by
859 // PredDefaultWeight (Weights[0]).
860 Weights.push_back(Weights[0] * SuccWeights[i+1]);
861 ValidTotalSuccWeight += SuccWeights[i+1];
865 if (SuccHasWeights || PredHasWeights) {
866 ValidTotalSuccWeight += SuccWeights[0];
867 // Scale the cases from predecessor by ValidTotalSuccWeight.
868 for (unsigned i = 1; i < CasesFromPred; ++i)
869 Weights[i] *= ValidTotalSuccWeight;
870 // Scale the default weight by SuccDefaultWeight (SuccWeights[0]).
871 Weights[0] *= SuccWeights[0];
874 // If this is not the default destination from PSI, only the edges
875 // in SI that occur in PSI with a destination of BB will be
877 std::set<ConstantInt*, ConstantIntOrdering> PTIHandled;
878 std::map<ConstantInt*, uint64_t> WeightsForHandled;
879 for (unsigned i = 0, e = PredCases.size(); i != e; ++i)
880 if (PredCases[i].Dest == BB) {
881 PTIHandled.insert(PredCases[i].Value);
883 if (PredHasWeights || SuccHasWeights) {
884 WeightsForHandled[PredCases[i].Value] = Weights[i+1];
885 std::swap(Weights[i+1], Weights.back());
889 std::swap(PredCases[i], PredCases.back());
890 PredCases.pop_back();
894 // Okay, now we know which constants were sent to BB from the
895 // predecessor. Figure out where they will all go now.
896 for (unsigned i = 0, e = BBCases.size(); i != e; ++i)
897 if (PTIHandled.count(BBCases[i].Value)) {
898 // If this is one we are capable of getting...
899 if (PredHasWeights || SuccHasWeights)
900 Weights.push_back(WeightsForHandled[BBCases[i].Value]);
901 PredCases.push_back(BBCases[i]);
902 NewSuccessors.push_back(BBCases[i].Dest);
903 PTIHandled.erase(BBCases[i].Value);// This constant is taken care of
906 // If there are any constants vectored to BB that TI doesn't handle,
907 // they must go to the default destination of TI.
908 for (std::set<ConstantInt*, ConstantIntOrdering>::iterator I =
910 E = PTIHandled.end(); I != E; ++I) {
911 if (PredHasWeights || SuccHasWeights)
912 Weights.push_back(WeightsForHandled[*I]);
913 PredCases.push_back(ValueEqualityComparisonCase(*I, BBDefault));
914 NewSuccessors.push_back(BBDefault);
918 // Okay, at this point, we know which new successor Pred will get. Make
919 // sure we update the number of entries in the PHI nodes for these
921 for (unsigned i = 0, e = NewSuccessors.size(); i != e; ++i)
922 AddPredecessorToBlock(NewSuccessors[i], Pred, BB);
924 Builder.SetInsertPoint(PTI);
925 // Convert pointer to int before we switch.
926 if (CV->getType()->isPointerTy()) {
927 assert(DL && "Cannot switch on pointer without DataLayout");
928 CV = Builder.CreatePtrToInt(CV, DL->getIntPtrType(CV->getType()),
932 // Now that the successors are updated, create the new Switch instruction.
933 SwitchInst *NewSI = Builder.CreateSwitch(CV, PredDefault,
935 NewSI->setDebugLoc(PTI->getDebugLoc());
936 for (unsigned i = 0, e = PredCases.size(); i != e; ++i)
937 NewSI->addCase(PredCases[i].Value, PredCases[i].Dest);
939 if (PredHasWeights || SuccHasWeights) {
940 // Halve the weights if any of them cannot fit in an uint32_t
943 SmallVector<uint32_t, 8> MDWeights(Weights.begin(), Weights.end());
945 NewSI->setMetadata(LLVMContext::MD_prof,
946 MDBuilder(BB->getContext()).
947 createBranchWeights(MDWeights));
950 EraseTerminatorInstAndDCECond(PTI);
952 // Okay, last check. If BB is still a successor of PSI, then we must
953 // have an infinite loop case. If so, add an infinitely looping block
954 // to handle the case to preserve the behavior of the code.
955 BasicBlock *InfLoopBlock = nullptr;
956 for (unsigned i = 0, e = NewSI->getNumSuccessors(); i != e; ++i)
957 if (NewSI->getSuccessor(i) == BB) {
959 // Insert it at the end of the function, because it's either code,
960 // or it won't matter if it's hot. :)
961 InfLoopBlock = BasicBlock::Create(BB->getContext(),
962 "infloop", BB->getParent());
963 BranchInst::Create(InfLoopBlock, InfLoopBlock);
965 NewSI->setSuccessor(i, InfLoopBlock);
974 // isSafeToHoistInvoke - If we would need to insert a select that uses the
975 // value of this invoke (comments in HoistThenElseCodeToIf explain why we
976 // would need to do this), we can't hoist the invoke, as there is nowhere
977 // to put the select in this case.
978 static bool isSafeToHoistInvoke(BasicBlock *BB1, BasicBlock *BB2,
979 Instruction *I1, Instruction *I2) {
980 for (succ_iterator SI = succ_begin(BB1), E = succ_end(BB1); SI != E; ++SI) {
982 for (BasicBlock::iterator BBI = SI->begin();
983 (PN = dyn_cast<PHINode>(BBI)); ++BBI) {
984 Value *BB1V = PN->getIncomingValueForBlock(BB1);
985 Value *BB2V = PN->getIncomingValueForBlock(BB2);
986 if (BB1V != BB2V && (BB1V==I1 || BB2V==I2)) {
994 /// HoistThenElseCodeToIf - Given a conditional branch that goes to BB1 and
995 /// BB2, hoist any common code in the two blocks up into the branch block. The
996 /// caller of this function guarantees that BI's block dominates BB1 and BB2.
997 static bool HoistThenElseCodeToIf(BranchInst *BI) {
998 // This does very trivial matching, with limited scanning, to find identical
999 // instructions in the two blocks. In particular, we don't want to get into
1000 // O(M*N) situations here where M and N are the sizes of BB1 and BB2. As
1001 // such, we currently just scan for obviously identical instructions in an
1003 BasicBlock *BB1 = BI->getSuccessor(0); // The true destination.
1004 BasicBlock *BB2 = BI->getSuccessor(1); // The false destination
1006 BasicBlock::iterator BB1_Itr = BB1->begin();
1007 BasicBlock::iterator BB2_Itr = BB2->begin();
1009 Instruction *I1 = BB1_Itr++, *I2 = BB2_Itr++;
1010 // Skip debug info if it is not identical.
1011 DbgInfoIntrinsic *DBI1 = dyn_cast<DbgInfoIntrinsic>(I1);
1012 DbgInfoIntrinsic *DBI2 = dyn_cast<DbgInfoIntrinsic>(I2);
1013 if (!DBI1 || !DBI2 || !DBI1->isIdenticalToWhenDefined(DBI2)) {
1014 while (isa<DbgInfoIntrinsic>(I1))
1016 while (isa<DbgInfoIntrinsic>(I2))
1019 if (isa<PHINode>(I1) || !I1->isIdenticalToWhenDefined(I2) ||
1020 (isa<InvokeInst>(I1) && !isSafeToHoistInvoke(BB1, BB2, I1, I2)))
1023 BasicBlock *BIParent = BI->getParent();
1025 bool Changed = false;
1027 // If we are hoisting the terminator instruction, don't move one (making a
1028 // broken BB), instead clone it, and remove BI.
1029 if (isa<TerminatorInst>(I1))
1030 goto HoistTerminator;
1032 // For a normal instruction, we just move one to right before the branch,
1033 // then replace all uses of the other with the first. Finally, we remove
1034 // the now redundant second instruction.
1035 BIParent->getInstList().splice(BI, BB1->getInstList(), I1);
1036 if (!I2->use_empty())
1037 I2->replaceAllUsesWith(I1);
1038 I1->intersectOptionalDataWith(I2);
1039 I2->eraseFromParent();
1044 // Skip debug info if it is not identical.
1045 DbgInfoIntrinsic *DBI1 = dyn_cast<DbgInfoIntrinsic>(I1);
1046 DbgInfoIntrinsic *DBI2 = dyn_cast<DbgInfoIntrinsic>(I2);
1047 if (!DBI1 || !DBI2 || !DBI1->isIdenticalToWhenDefined(DBI2)) {
1048 while (isa<DbgInfoIntrinsic>(I1))
1050 while (isa<DbgInfoIntrinsic>(I2))
1053 } while (I1->isIdenticalToWhenDefined(I2));
1058 // It may not be possible to hoist an invoke.
1059 if (isa<InvokeInst>(I1) && !isSafeToHoistInvoke(BB1, BB2, I1, I2))
1062 for (succ_iterator SI = succ_begin(BB1), E = succ_end(BB1); SI != E; ++SI) {
1064 for (BasicBlock::iterator BBI = SI->begin();
1065 (PN = dyn_cast<PHINode>(BBI)); ++BBI) {
1066 Value *BB1V = PN->getIncomingValueForBlock(BB1);
1067 Value *BB2V = PN->getIncomingValueForBlock(BB2);
1071 if (isa<ConstantExpr>(BB1V) && !isSafeToSpeculativelyExecute(BB1V))
1073 if (isa<ConstantExpr>(BB2V) && !isSafeToSpeculativelyExecute(BB2V))
1078 // Okay, it is safe to hoist the terminator.
1079 Instruction *NT = I1->clone();
1080 BIParent->getInstList().insert(BI, NT);
1081 if (!NT->getType()->isVoidTy()) {
1082 I1->replaceAllUsesWith(NT);
1083 I2->replaceAllUsesWith(NT);
1087 IRBuilder<true, NoFolder> Builder(NT);
1088 // Hoisting one of the terminators from our successor is a great thing.
1089 // Unfortunately, the successors of the if/else blocks may have PHI nodes in
1090 // them. If they do, all PHI entries for BB1/BB2 must agree for all PHI
1091 // nodes, so we insert select instruction to compute the final result.
1092 std::map<std::pair<Value*,Value*>, SelectInst*> InsertedSelects;
1093 for (succ_iterator SI = succ_begin(BB1), E = succ_end(BB1); SI != E; ++SI) {
1095 for (BasicBlock::iterator BBI = SI->begin();
1096 (PN = dyn_cast<PHINode>(BBI)); ++BBI) {
1097 Value *BB1V = PN->getIncomingValueForBlock(BB1);
1098 Value *BB2V = PN->getIncomingValueForBlock(BB2);
1099 if (BB1V == BB2V) continue;
1101 // These values do not agree. Insert a select instruction before NT
1102 // that determines the right value.
1103 SelectInst *&SI = InsertedSelects[std::make_pair(BB1V, BB2V)];
1105 SI = cast<SelectInst>
1106 (Builder.CreateSelect(BI->getCondition(), BB1V, BB2V,
1107 BB1V->getName()+"."+BB2V->getName()));
1109 // Make the PHI node use the select for all incoming values for BB1/BB2
1110 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
1111 if (PN->getIncomingBlock(i) == BB1 || PN->getIncomingBlock(i) == BB2)
1112 PN->setIncomingValue(i, SI);
1116 // Update any PHI nodes in our new successors.
1117 for (succ_iterator SI = succ_begin(BB1), E = succ_end(BB1); SI != E; ++SI)
1118 AddPredecessorToBlock(*SI, BIParent, BB1);
1120 EraseTerminatorInstAndDCECond(BI);
1124 /// SinkThenElseCodeToEnd - Given an unconditional branch that goes to BBEnd,
1125 /// check whether BBEnd has only two predecessors and the other predecessor
1126 /// ends with an unconditional branch. If it is true, sink any common code
1127 /// in the two predecessors to BBEnd.
1128 static bool SinkThenElseCodeToEnd(BranchInst *BI1) {
1129 assert(BI1->isUnconditional());
1130 BasicBlock *BB1 = BI1->getParent();
1131 BasicBlock *BBEnd = BI1->getSuccessor(0);
1133 // Check that BBEnd has two predecessors and the other predecessor ends with
1134 // an unconditional branch.
1135 pred_iterator PI = pred_begin(BBEnd), PE = pred_end(BBEnd);
1136 BasicBlock *Pred0 = *PI++;
1137 if (PI == PE) // Only one predecessor.
1139 BasicBlock *Pred1 = *PI++;
1140 if (PI != PE) // More than two predecessors.
1142 BasicBlock *BB2 = (Pred0 == BB1) ? Pred1 : Pred0;
1143 BranchInst *BI2 = dyn_cast<BranchInst>(BB2->getTerminator());
1144 if (!BI2 || !BI2->isUnconditional())
1147 // Gather the PHI nodes in BBEnd.
1148 std::map<Value*, std::pair<Value*, PHINode*> > MapValueFromBB1ToBB2;
1149 Instruction *FirstNonPhiInBBEnd = nullptr;
1150 for (BasicBlock::iterator I = BBEnd->begin(), E = BBEnd->end();
1152 if (PHINode *PN = dyn_cast<PHINode>(I)) {
1153 Value *BB1V = PN->getIncomingValueForBlock(BB1);
1154 Value *BB2V = PN->getIncomingValueForBlock(BB2);
1155 MapValueFromBB1ToBB2[BB1V] = std::make_pair(BB2V, PN);
1157 FirstNonPhiInBBEnd = &*I;
1161 if (!FirstNonPhiInBBEnd)
1165 // This does very trivial matching, with limited scanning, to find identical
1166 // instructions in the two blocks. We scan backward for obviously identical
1167 // instructions in an identical order.
1168 BasicBlock::InstListType::reverse_iterator RI1 = BB1->getInstList().rbegin(),
1169 RE1 = BB1->getInstList().rend(), RI2 = BB2->getInstList().rbegin(),
1170 RE2 = BB2->getInstList().rend();
1172 while (RI1 != RE1 && isa<DbgInfoIntrinsic>(&*RI1)) ++RI1;
1175 while (RI2 != RE2 && isa<DbgInfoIntrinsic>(&*RI2)) ++RI2;
1178 // Skip the unconditional branches.
1182 bool Changed = false;
1183 while (RI1 != RE1 && RI2 != RE2) {
1185 while (RI1 != RE1 && isa<DbgInfoIntrinsic>(&*RI1)) ++RI1;
1188 while (RI2 != RE2 && isa<DbgInfoIntrinsic>(&*RI2)) ++RI2;
1192 Instruction *I1 = &*RI1, *I2 = &*RI2;
1193 // I1 and I2 should have a single use in the same PHI node, and they
1194 // perform the same operation.
1195 // Cannot move control-flow-involving, volatile loads, vaarg, etc.
1196 if (isa<PHINode>(I1) || isa<PHINode>(I2) ||
1197 isa<TerminatorInst>(I1) || isa<TerminatorInst>(I2) ||
1198 isa<LandingPadInst>(I1) || isa<LandingPadInst>(I2) ||
1199 isa<AllocaInst>(I1) || isa<AllocaInst>(I2) ||
1200 I1->mayHaveSideEffects() || I2->mayHaveSideEffects() ||
1201 I1->mayReadOrWriteMemory() || I2->mayReadOrWriteMemory() ||
1202 !I1->hasOneUse() || !I2->hasOneUse() ||
1203 MapValueFromBB1ToBB2.find(I1) == MapValueFromBB1ToBB2.end() ||
1204 MapValueFromBB1ToBB2[I1].first != I2)
1207 // Check whether we should swap the operands of ICmpInst.
1208 ICmpInst *ICmp1 = dyn_cast<ICmpInst>(I1), *ICmp2 = dyn_cast<ICmpInst>(I2);
1209 bool SwapOpnds = false;
1210 if (ICmp1 && ICmp2 &&
1211 ICmp1->getOperand(0) != ICmp2->getOperand(0) &&
1212 ICmp1->getOperand(1) != ICmp2->getOperand(1) &&
1213 (ICmp1->getOperand(0) == ICmp2->getOperand(1) ||
1214 ICmp1->getOperand(1) == ICmp2->getOperand(0))) {
1215 ICmp2->swapOperands();
1218 if (!I1->isSameOperationAs(I2)) {
1220 ICmp2->swapOperands();
1224 // The operands should be either the same or they need to be generated
1225 // with a PHI node after sinking. We only handle the case where there is
1226 // a single pair of different operands.
1227 Value *DifferentOp1 = nullptr, *DifferentOp2 = nullptr;
1228 unsigned Op1Idx = 0;
1229 for (unsigned I = 0, E = I1->getNumOperands(); I != E; ++I) {
1230 if (I1->getOperand(I) == I2->getOperand(I))
1232 // Early exit if we have more-than one pair of different operands or
1233 // the different operand is already in MapValueFromBB1ToBB2.
1234 // Early exit if we need a PHI node to replace a constant.
1236 MapValueFromBB1ToBB2.find(I1->getOperand(I)) !=
1237 MapValueFromBB1ToBB2.end() ||
1238 isa<Constant>(I1->getOperand(I)) ||
1239 isa<Constant>(I2->getOperand(I))) {
1240 // If we can't sink the instructions, undo the swapping.
1242 ICmp2->swapOperands();
1245 DifferentOp1 = I1->getOperand(I);
1247 DifferentOp2 = I2->getOperand(I);
1250 // We insert the pair of different operands to MapValueFromBB1ToBB2 and
1251 // remove (I1, I2) from MapValueFromBB1ToBB2.
1253 PHINode *NewPN = PHINode::Create(DifferentOp1->getType(), 2,
1254 DifferentOp1->getName() + ".sink",
1256 MapValueFromBB1ToBB2[DifferentOp1] = std::make_pair(DifferentOp2, NewPN);
1257 // I1 should use NewPN instead of DifferentOp1.
1258 I1->setOperand(Op1Idx, NewPN);
1259 NewPN->addIncoming(DifferentOp1, BB1);
1260 NewPN->addIncoming(DifferentOp2, BB2);
1261 DEBUG(dbgs() << "Create PHI node " << *NewPN << "\n";);
1263 PHINode *OldPN = MapValueFromBB1ToBB2[I1].second;
1264 MapValueFromBB1ToBB2.erase(I1);
1266 DEBUG(dbgs() << "SINK common instructions " << *I1 << "\n";);
1267 DEBUG(dbgs() << " " << *I2 << "\n";);
1268 // We need to update RE1 and RE2 if we are going to sink the first
1269 // instruction in the basic block down.
1270 bool UpdateRE1 = (I1 == BB1->begin()), UpdateRE2 = (I2 == BB2->begin());
1271 // Sink the instruction.
1272 BBEnd->getInstList().splice(FirstNonPhiInBBEnd, BB1->getInstList(), I1);
1273 if (!OldPN->use_empty())
1274 OldPN->replaceAllUsesWith(I1);
1275 OldPN->eraseFromParent();
1277 if (!I2->use_empty())
1278 I2->replaceAllUsesWith(I1);
1279 I1->intersectOptionalDataWith(I2);
1280 I2->eraseFromParent();
1283 RE1 = BB1->getInstList().rend();
1285 RE2 = BB2->getInstList().rend();
1286 FirstNonPhiInBBEnd = I1;
1293 /// \brief Determine if we can hoist sink a sole store instruction out of a
1294 /// conditional block.
1296 /// We are looking for code like the following:
1298 /// store i32 %add, i32* %arrayidx2
1299 /// ... // No other stores or function calls (we could be calling a memory
1300 /// ... // function).
1301 /// %cmp = icmp ult %x, %y
1302 /// br i1 %cmp, label %EndBB, label %ThenBB
1304 /// store i32 %add5, i32* %arrayidx2
1308 /// We are going to transform this into:
1310 /// store i32 %add, i32* %arrayidx2
1312 /// %cmp = icmp ult %x, %y
1313 /// %add.add5 = select i1 %cmp, i32 %add, %add5
1314 /// store i32 %add.add5, i32* %arrayidx2
1317 /// \return The pointer to the value of the previous store if the store can be
1318 /// hoisted into the predecessor block. 0 otherwise.
1319 static Value *isSafeToSpeculateStore(Instruction *I, BasicBlock *BrBB,
1320 BasicBlock *StoreBB, BasicBlock *EndBB) {
1321 StoreInst *StoreToHoist = dyn_cast<StoreInst>(I);
1325 // Volatile or atomic.
1326 if (!StoreToHoist->isSimple())
1329 Value *StorePtr = StoreToHoist->getPointerOperand();
1331 // Look for a store to the same pointer in BrBB.
1332 unsigned MaxNumInstToLookAt = 10;
1333 for (BasicBlock::reverse_iterator RI = BrBB->rbegin(),
1334 RE = BrBB->rend(); RI != RE && (--MaxNumInstToLookAt); ++RI) {
1335 Instruction *CurI = &*RI;
1337 // Could be calling an instruction that effects memory like free().
1338 if (CurI->mayHaveSideEffects() && !isa<StoreInst>(CurI))
1341 StoreInst *SI = dyn_cast<StoreInst>(CurI);
1342 // Found the previous store make sure it stores to the same location.
1343 if (SI && SI->getPointerOperand() == StorePtr)
1344 // Found the previous store, return its value operand.
1345 return SI->getValueOperand();
1347 return nullptr; // Unknown store.
1353 /// \brief Speculate a conditional basic block flattening the CFG.
1355 /// Note that this is a very risky transform currently. Speculating
1356 /// instructions like this is most often not desirable. Instead, there is an MI
1357 /// pass which can do it with full awareness of the resource constraints.
1358 /// However, some cases are "obvious" and we should do directly. An example of
1359 /// this is speculating a single, reasonably cheap instruction.
1361 /// There is only one distinct advantage to flattening the CFG at the IR level:
1362 /// it makes very common but simplistic optimizations such as are common in
1363 /// instcombine and the DAG combiner more powerful by removing CFG edges and
1364 /// modeling their effects with easier to reason about SSA value graphs.
1367 /// An illustration of this transform is turning this IR:
1370 /// %cmp = icmp ult %x, %y
1371 /// br i1 %cmp, label %EndBB, label %ThenBB
1373 /// %sub = sub %x, %y
1376 /// %phi = phi [ %sub, %ThenBB ], [ 0, %EndBB ]
1383 /// %cmp = icmp ult %x, %y
1384 /// %sub = sub %x, %y
1385 /// %cond = select i1 %cmp, 0, %sub
1389 /// \returns true if the conditional block is removed.
1390 static bool SpeculativelyExecuteBB(BranchInst *BI, BasicBlock *ThenBB) {
1391 // Be conservative for now. FP select instruction can often be expensive.
1392 Value *BrCond = BI->getCondition();
1393 if (isa<FCmpInst>(BrCond))
1396 BasicBlock *BB = BI->getParent();
1397 BasicBlock *EndBB = ThenBB->getTerminator()->getSuccessor(0);
1399 // If ThenBB is actually on the false edge of the conditional branch, remember
1400 // to swap the select operands later.
1401 bool Invert = false;
1402 if (ThenBB != BI->getSuccessor(0)) {
1403 assert(ThenBB == BI->getSuccessor(1) && "No edge from 'if' block?");
1406 assert(EndBB == BI->getSuccessor(!Invert) && "No edge from to end block");
1408 // Keep a count of how many times instructions are used within CondBB when
1409 // they are candidates for sinking into CondBB. Specifically:
1410 // - They are defined in BB, and
1411 // - They have no side effects, and
1412 // - All of their uses are in CondBB.
1413 SmallDenseMap<Instruction *, unsigned, 4> SinkCandidateUseCounts;
1415 unsigned SpeculationCost = 0;
1416 Value *SpeculatedStoreValue = nullptr;
1417 StoreInst *SpeculatedStore = nullptr;
1418 for (BasicBlock::iterator BBI = ThenBB->begin(),
1419 BBE = std::prev(ThenBB->end());
1420 BBI != BBE; ++BBI) {
1421 Instruction *I = BBI;
1423 if (isa<DbgInfoIntrinsic>(I))
1426 // Only speculatively execution a single instruction (not counting the
1427 // terminator) for now.
1429 if (SpeculationCost > 1)
1432 // Don't hoist the instruction if it's unsafe or expensive.
1433 if (!isSafeToSpeculativelyExecute(I) &&
1434 !(HoistCondStores &&
1435 (SpeculatedStoreValue = isSafeToSpeculateStore(I, BB, ThenBB,
1438 if (!SpeculatedStoreValue &&
1439 ComputeSpeculationCost(I) > PHINodeFoldingThreshold)
1442 // Store the store speculation candidate.
1443 if (SpeculatedStoreValue)
1444 SpeculatedStore = cast<StoreInst>(I);
1446 // Do not hoist the instruction if any of its operands are defined but not
1447 // used in BB. The transformation will prevent the operand from
1448 // being sunk into the use block.
1449 for (User::op_iterator i = I->op_begin(), e = I->op_end();
1451 Instruction *OpI = dyn_cast<Instruction>(*i);
1452 if (!OpI || OpI->getParent() != BB ||
1453 OpI->mayHaveSideEffects())
1454 continue; // Not a candidate for sinking.
1456 ++SinkCandidateUseCounts[OpI];
1460 // Consider any sink candidates which are only used in CondBB as costs for
1461 // speculation. Note, while we iterate over a DenseMap here, we are summing
1462 // and so iteration order isn't significant.
1463 for (SmallDenseMap<Instruction *, unsigned, 4>::iterator I =
1464 SinkCandidateUseCounts.begin(), E = SinkCandidateUseCounts.end();
1466 if (I->first->getNumUses() == I->second) {
1468 if (SpeculationCost > 1)
1472 // Check that the PHI nodes can be converted to selects.
1473 bool HaveRewritablePHIs = false;
1474 for (BasicBlock::iterator I = EndBB->begin();
1475 PHINode *PN = dyn_cast<PHINode>(I); ++I) {
1476 Value *OrigV = PN->getIncomingValueForBlock(BB);
1477 Value *ThenV = PN->getIncomingValueForBlock(ThenBB);
1479 // FIXME: Try to remove some of the duplication with HoistThenElseCodeToIf.
1480 // Skip PHIs which are trivial.
1484 HaveRewritablePHIs = true;
1485 ConstantExpr *OrigCE = dyn_cast<ConstantExpr>(OrigV);
1486 ConstantExpr *ThenCE = dyn_cast<ConstantExpr>(ThenV);
1487 if (!OrigCE && !ThenCE)
1488 continue; // Known safe and cheap.
1490 if ((ThenCE && !isSafeToSpeculativelyExecute(ThenCE)) ||
1491 (OrigCE && !isSafeToSpeculativelyExecute(OrigCE)))
1493 unsigned OrigCost = OrigCE ? ComputeSpeculationCost(OrigCE) : 0;
1494 unsigned ThenCost = ThenCE ? ComputeSpeculationCost(ThenCE) : 0;
1495 if (OrigCost + ThenCost > 2 * PHINodeFoldingThreshold)
1498 // Account for the cost of an unfolded ConstantExpr which could end up
1499 // getting expanded into Instructions.
1500 // FIXME: This doesn't account for how many operations are combined in the
1501 // constant expression.
1503 if (SpeculationCost > 1)
1507 // If there are no PHIs to process, bail early. This helps ensure idempotence
1509 if (!HaveRewritablePHIs && !(HoistCondStores && SpeculatedStoreValue))
1512 // If we get here, we can hoist the instruction and if-convert.
1513 DEBUG(dbgs() << "SPECULATIVELY EXECUTING BB" << *ThenBB << "\n";);
1515 // Insert a select of the value of the speculated store.
1516 if (SpeculatedStoreValue) {
1517 IRBuilder<true, NoFolder> Builder(BI);
1518 Value *TrueV = SpeculatedStore->getValueOperand();
1519 Value *FalseV = SpeculatedStoreValue;
1521 std::swap(TrueV, FalseV);
1522 Value *S = Builder.CreateSelect(BrCond, TrueV, FalseV, TrueV->getName() +
1523 "." + FalseV->getName());
1524 SpeculatedStore->setOperand(0, S);
1527 // Hoist the instructions.
1528 BB->getInstList().splice(BI, ThenBB->getInstList(), ThenBB->begin(),
1529 std::prev(ThenBB->end()));
1531 // Insert selects and rewrite the PHI operands.
1532 IRBuilder<true, NoFolder> Builder(BI);
1533 for (BasicBlock::iterator I = EndBB->begin();
1534 PHINode *PN = dyn_cast<PHINode>(I); ++I) {
1535 unsigned OrigI = PN->getBasicBlockIndex(BB);
1536 unsigned ThenI = PN->getBasicBlockIndex(ThenBB);
1537 Value *OrigV = PN->getIncomingValue(OrigI);
1538 Value *ThenV = PN->getIncomingValue(ThenI);
1540 // Skip PHIs which are trivial.
1544 // Create a select whose true value is the speculatively executed value and
1545 // false value is the preexisting value. Swap them if the branch
1546 // destinations were inverted.
1547 Value *TrueV = ThenV, *FalseV = OrigV;
1549 std::swap(TrueV, FalseV);
1550 Value *V = Builder.CreateSelect(BrCond, TrueV, FalseV,
1551 TrueV->getName() + "." + FalseV->getName());
1552 PN->setIncomingValue(OrigI, V);
1553 PN->setIncomingValue(ThenI, V);
1560 /// \returns True if this block contains a CallInst with the NoDuplicate
1562 static bool HasNoDuplicateCall(const BasicBlock *BB) {
1563 for (BasicBlock::const_iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
1564 const CallInst *CI = dyn_cast<CallInst>(I);
1567 if (CI->cannotDuplicate())
1573 /// BlockIsSimpleEnoughToThreadThrough - Return true if we can thread a branch
1574 /// across this block.
1575 static bool BlockIsSimpleEnoughToThreadThrough(BasicBlock *BB) {
1576 BranchInst *BI = cast<BranchInst>(BB->getTerminator());
1579 for (BasicBlock::iterator BBI = BB->begin(); &*BBI != BI; ++BBI) {
1580 if (isa<DbgInfoIntrinsic>(BBI))
1582 if (Size > 10) return false; // Don't clone large BB's.
1585 // We can only support instructions that do not define values that are
1586 // live outside of the current basic block.
1587 for (User *U : BBI->users()) {
1588 Instruction *UI = cast<Instruction>(U);
1589 if (UI->getParent() != BB || isa<PHINode>(UI)) return false;
1592 // Looks ok, continue checking.
1598 /// FoldCondBranchOnPHI - If we have a conditional branch on a PHI node value
1599 /// that is defined in the same block as the branch and if any PHI entries are
1600 /// constants, thread edges corresponding to that entry to be branches to their
1601 /// ultimate destination.
1602 static bool FoldCondBranchOnPHI(BranchInst *BI, const DataLayout *DL) {
1603 BasicBlock *BB = BI->getParent();
1604 PHINode *PN = dyn_cast<PHINode>(BI->getCondition());
1605 // NOTE: we currently cannot transform this case if the PHI node is used
1606 // outside of the block.
1607 if (!PN || PN->getParent() != BB || !PN->hasOneUse())
1610 // Degenerate case of a single entry PHI.
1611 if (PN->getNumIncomingValues() == 1) {
1612 FoldSingleEntryPHINodes(PN->getParent());
1616 // Now we know that this block has multiple preds and two succs.
1617 if (!BlockIsSimpleEnoughToThreadThrough(BB)) return false;
1619 if (HasNoDuplicateCall(BB)) return false;
1621 // Okay, this is a simple enough basic block. See if any phi values are
1623 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
1624 ConstantInt *CB = dyn_cast<ConstantInt>(PN->getIncomingValue(i));
1625 if (!CB || !CB->getType()->isIntegerTy(1)) continue;
1627 // Okay, we now know that all edges from PredBB should be revectored to
1628 // branch to RealDest.
1629 BasicBlock *PredBB = PN->getIncomingBlock(i);
1630 BasicBlock *RealDest = BI->getSuccessor(!CB->getZExtValue());
1632 if (RealDest == BB) continue; // Skip self loops.
1633 // Skip if the predecessor's terminator is an indirect branch.
1634 if (isa<IndirectBrInst>(PredBB->getTerminator())) continue;
1636 // The dest block might have PHI nodes, other predecessors and other
1637 // difficult cases. Instead of being smart about this, just insert a new
1638 // block that jumps to the destination block, effectively splitting
1639 // the edge we are about to create.
1640 BasicBlock *EdgeBB = BasicBlock::Create(BB->getContext(),
1641 RealDest->getName()+".critedge",
1642 RealDest->getParent(), RealDest);
1643 BranchInst::Create(RealDest, EdgeBB);
1645 // Update PHI nodes.
1646 AddPredecessorToBlock(RealDest, EdgeBB, BB);
1648 // BB may have instructions that are being threaded over. Clone these
1649 // instructions into EdgeBB. We know that there will be no uses of the
1650 // cloned instructions outside of EdgeBB.
1651 BasicBlock::iterator InsertPt = EdgeBB->begin();
1652 DenseMap<Value*, Value*> TranslateMap; // Track translated values.
1653 for (BasicBlock::iterator BBI = BB->begin(); &*BBI != BI; ++BBI) {
1654 if (PHINode *PN = dyn_cast<PHINode>(BBI)) {
1655 TranslateMap[PN] = PN->getIncomingValueForBlock(PredBB);
1658 // Clone the instruction.
1659 Instruction *N = BBI->clone();
1660 if (BBI->hasName()) N->setName(BBI->getName()+".c");
1662 // Update operands due to translation.
1663 for (User::op_iterator i = N->op_begin(), e = N->op_end();
1665 DenseMap<Value*, Value*>::iterator PI = TranslateMap.find(*i);
1666 if (PI != TranslateMap.end())
1670 // Check for trivial simplification.
1671 if (Value *V = SimplifyInstruction(N, DL)) {
1672 TranslateMap[BBI] = V;
1673 delete N; // Instruction folded away, don't need actual inst
1675 // Insert the new instruction into its new home.
1676 EdgeBB->getInstList().insert(InsertPt, N);
1677 if (!BBI->use_empty())
1678 TranslateMap[BBI] = N;
1682 // Loop over all of the edges from PredBB to BB, changing them to branch
1683 // to EdgeBB instead.
1684 TerminatorInst *PredBBTI = PredBB->getTerminator();
1685 for (unsigned i = 0, e = PredBBTI->getNumSuccessors(); i != e; ++i)
1686 if (PredBBTI->getSuccessor(i) == BB) {
1687 BB->removePredecessor(PredBB);
1688 PredBBTI->setSuccessor(i, EdgeBB);
1691 // Recurse, simplifying any other constants.
1692 return FoldCondBranchOnPHI(BI, DL) | true;
1698 /// FoldTwoEntryPHINode - Given a BB that starts with the specified two-entry
1699 /// PHI node, see if we can eliminate it.
1700 static bool FoldTwoEntryPHINode(PHINode *PN, const DataLayout *DL) {
1701 // Ok, this is a two entry PHI node. Check to see if this is a simple "if
1702 // statement", which has a very simple dominance structure. Basically, we
1703 // are trying to find the condition that is being branched on, which
1704 // subsequently causes this merge to happen. We really want control
1705 // dependence information for this check, but simplifycfg can't keep it up
1706 // to date, and this catches most of the cases we care about anyway.
1707 BasicBlock *BB = PN->getParent();
1708 BasicBlock *IfTrue, *IfFalse;
1709 Value *IfCond = GetIfCondition(BB, IfTrue, IfFalse);
1711 // Don't bother if the branch will be constant folded trivially.
1712 isa<ConstantInt>(IfCond))
1715 // Okay, we found that we can merge this two-entry phi node into a select.
1716 // Doing so would require us to fold *all* two entry phi nodes in this block.
1717 // At some point this becomes non-profitable (particularly if the target
1718 // doesn't support cmov's). Only do this transformation if there are two or
1719 // fewer PHI nodes in this block.
1720 unsigned NumPhis = 0;
1721 for (BasicBlock::iterator I = BB->begin(); isa<PHINode>(I); ++NumPhis, ++I)
1725 // Loop over the PHI's seeing if we can promote them all to select
1726 // instructions. While we are at it, keep track of the instructions
1727 // that need to be moved to the dominating block.
1728 SmallPtrSet<Instruction*, 4> AggressiveInsts;
1729 unsigned MaxCostVal0 = PHINodeFoldingThreshold,
1730 MaxCostVal1 = PHINodeFoldingThreshold;
1732 for (BasicBlock::iterator II = BB->begin(); isa<PHINode>(II);) {
1733 PHINode *PN = cast<PHINode>(II++);
1734 if (Value *V = SimplifyInstruction(PN, DL)) {
1735 PN->replaceAllUsesWith(V);
1736 PN->eraseFromParent();
1740 if (!DominatesMergePoint(PN->getIncomingValue(0), BB, &AggressiveInsts,
1742 !DominatesMergePoint(PN->getIncomingValue(1), BB, &AggressiveInsts,
1747 // If we folded the first phi, PN dangles at this point. Refresh it. If
1748 // we ran out of PHIs then we simplified them all.
1749 PN = dyn_cast<PHINode>(BB->begin());
1750 if (!PN) return true;
1752 // Don't fold i1 branches on PHIs which contain binary operators. These can
1753 // often be turned into switches and other things.
1754 if (PN->getType()->isIntegerTy(1) &&
1755 (isa<BinaryOperator>(PN->getIncomingValue(0)) ||
1756 isa<BinaryOperator>(PN->getIncomingValue(1)) ||
1757 isa<BinaryOperator>(IfCond)))
1760 // If we all PHI nodes are promotable, check to make sure that all
1761 // instructions in the predecessor blocks can be promoted as well. If
1762 // not, we won't be able to get rid of the control flow, so it's not
1763 // worth promoting to select instructions.
1764 BasicBlock *DomBlock = nullptr;
1765 BasicBlock *IfBlock1 = PN->getIncomingBlock(0);
1766 BasicBlock *IfBlock2 = PN->getIncomingBlock(1);
1767 if (cast<BranchInst>(IfBlock1->getTerminator())->isConditional()) {
1770 DomBlock = *pred_begin(IfBlock1);
1771 for (BasicBlock::iterator I = IfBlock1->begin();!isa<TerminatorInst>(I);++I)
1772 if (!AggressiveInsts.count(I) && !isa<DbgInfoIntrinsic>(I)) {
1773 // This is not an aggressive instruction that we can promote.
1774 // Because of this, we won't be able to get rid of the control
1775 // flow, so the xform is not worth it.
1780 if (cast<BranchInst>(IfBlock2->getTerminator())->isConditional()) {
1783 DomBlock = *pred_begin(IfBlock2);
1784 for (BasicBlock::iterator I = IfBlock2->begin();!isa<TerminatorInst>(I);++I)
1785 if (!AggressiveInsts.count(I) && !isa<DbgInfoIntrinsic>(I)) {
1786 // This is not an aggressive instruction that we can promote.
1787 // Because of this, we won't be able to get rid of the control
1788 // flow, so the xform is not worth it.
1793 DEBUG(dbgs() << "FOUND IF CONDITION! " << *IfCond << " T: "
1794 << IfTrue->getName() << " F: " << IfFalse->getName() << "\n");
1796 // If we can still promote the PHI nodes after this gauntlet of tests,
1797 // do all of the PHI's now.
1798 Instruction *InsertPt = DomBlock->getTerminator();
1799 IRBuilder<true, NoFolder> Builder(InsertPt);
1801 // Move all 'aggressive' instructions, which are defined in the
1802 // conditional parts of the if's up to the dominating block.
1804 DomBlock->getInstList().splice(InsertPt,
1805 IfBlock1->getInstList(), IfBlock1->begin(),
1806 IfBlock1->getTerminator());
1808 DomBlock->getInstList().splice(InsertPt,
1809 IfBlock2->getInstList(), IfBlock2->begin(),
1810 IfBlock2->getTerminator());
1812 while (PHINode *PN = dyn_cast<PHINode>(BB->begin())) {
1813 // Change the PHI node into a select instruction.
1814 Value *TrueVal = PN->getIncomingValue(PN->getIncomingBlock(0) == IfFalse);
1815 Value *FalseVal = PN->getIncomingValue(PN->getIncomingBlock(0) == IfTrue);
1818 cast<SelectInst>(Builder.CreateSelect(IfCond, TrueVal, FalseVal, ""));
1819 PN->replaceAllUsesWith(NV);
1821 PN->eraseFromParent();
1824 // At this point, IfBlock1 and IfBlock2 are both empty, so our if statement
1825 // has been flattened. Change DomBlock to jump directly to our new block to
1826 // avoid other simplifycfg's kicking in on the diamond.
1827 TerminatorInst *OldTI = DomBlock->getTerminator();
1828 Builder.SetInsertPoint(OldTI);
1829 Builder.CreateBr(BB);
1830 OldTI->eraseFromParent();
1834 /// SimplifyCondBranchToTwoReturns - If we found a conditional branch that goes
1835 /// to two returning blocks, try to merge them together into one return,
1836 /// introducing a select if the return values disagree.
1837 static bool SimplifyCondBranchToTwoReturns(BranchInst *BI,
1838 IRBuilder<> &Builder) {
1839 assert(BI->isConditional() && "Must be a conditional branch");
1840 BasicBlock *TrueSucc = BI->getSuccessor(0);
1841 BasicBlock *FalseSucc = BI->getSuccessor(1);
1842 ReturnInst *TrueRet = cast<ReturnInst>(TrueSucc->getTerminator());
1843 ReturnInst *FalseRet = cast<ReturnInst>(FalseSucc->getTerminator());
1845 // Check to ensure both blocks are empty (just a return) or optionally empty
1846 // with PHI nodes. If there are other instructions, merging would cause extra
1847 // computation on one path or the other.
1848 if (!TrueSucc->getFirstNonPHIOrDbg()->isTerminator())
1850 if (!FalseSucc->getFirstNonPHIOrDbg()->isTerminator())
1853 Builder.SetInsertPoint(BI);
1854 // Okay, we found a branch that is going to two return nodes. If
1855 // there is no return value for this function, just change the
1856 // branch into a return.
1857 if (FalseRet->getNumOperands() == 0) {
1858 TrueSucc->removePredecessor(BI->getParent());
1859 FalseSucc->removePredecessor(BI->getParent());
1860 Builder.CreateRetVoid();
1861 EraseTerminatorInstAndDCECond(BI);
1865 // Otherwise, figure out what the true and false return values are
1866 // so we can insert a new select instruction.
1867 Value *TrueValue = TrueRet->getReturnValue();
1868 Value *FalseValue = FalseRet->getReturnValue();
1870 // Unwrap any PHI nodes in the return blocks.
1871 if (PHINode *TVPN = dyn_cast_or_null<PHINode>(TrueValue))
1872 if (TVPN->getParent() == TrueSucc)
1873 TrueValue = TVPN->getIncomingValueForBlock(BI->getParent());
1874 if (PHINode *FVPN = dyn_cast_or_null<PHINode>(FalseValue))
1875 if (FVPN->getParent() == FalseSucc)
1876 FalseValue = FVPN->getIncomingValueForBlock(BI->getParent());
1878 // In order for this transformation to be safe, we must be able to
1879 // unconditionally execute both operands to the return. This is
1880 // normally the case, but we could have a potentially-trapping
1881 // constant expression that prevents this transformation from being
1883 if (ConstantExpr *TCV = dyn_cast_or_null<ConstantExpr>(TrueValue))
1886 if (ConstantExpr *FCV = dyn_cast_or_null<ConstantExpr>(FalseValue))
1890 // Okay, we collected all the mapped values and checked them for sanity, and
1891 // defined to really do this transformation. First, update the CFG.
1892 TrueSucc->removePredecessor(BI->getParent());
1893 FalseSucc->removePredecessor(BI->getParent());
1895 // Insert select instructions where needed.
1896 Value *BrCond = BI->getCondition();
1898 // Insert a select if the results differ.
1899 if (TrueValue == FalseValue || isa<UndefValue>(FalseValue)) {
1900 } else if (isa<UndefValue>(TrueValue)) {
1901 TrueValue = FalseValue;
1903 TrueValue = Builder.CreateSelect(BrCond, TrueValue,
1904 FalseValue, "retval");
1908 Value *RI = !TrueValue ?
1909 Builder.CreateRetVoid() : Builder.CreateRet(TrueValue);
1913 DEBUG(dbgs() << "\nCHANGING BRANCH TO TWO RETURNS INTO SELECT:"
1914 << "\n " << *BI << "NewRet = " << *RI
1915 << "TRUEBLOCK: " << *TrueSucc << "FALSEBLOCK: "<< *FalseSucc);
1917 EraseTerminatorInstAndDCECond(BI);
1922 /// ExtractBranchMetadata - Given a conditional BranchInstruction, retrieve the
1923 /// probabilities of the branch taking each edge. Fills in the two APInt
1924 /// parameters and return true, or returns false if no or invalid metadata was
1926 static bool ExtractBranchMetadata(BranchInst *BI,
1927 uint64_t &ProbTrue, uint64_t &ProbFalse) {
1928 assert(BI->isConditional() &&
1929 "Looking for probabilities on unconditional branch?");
1930 MDNode *ProfileData = BI->getMetadata(LLVMContext::MD_prof);
1931 if (!ProfileData || ProfileData->getNumOperands() != 3) return false;
1932 ConstantInt *CITrue = dyn_cast<ConstantInt>(ProfileData->getOperand(1));
1933 ConstantInt *CIFalse = dyn_cast<ConstantInt>(ProfileData->getOperand(2));
1934 if (!CITrue || !CIFalse) return false;
1935 ProbTrue = CITrue->getValue().getZExtValue();
1936 ProbFalse = CIFalse->getValue().getZExtValue();
1940 /// checkCSEInPredecessor - Return true if the given instruction is available
1941 /// in its predecessor block. If yes, the instruction will be removed.
1943 static bool checkCSEInPredecessor(Instruction *Inst, BasicBlock *PB) {
1944 if (!isa<BinaryOperator>(Inst) && !isa<CmpInst>(Inst))
1946 for (BasicBlock::iterator I = PB->begin(), E = PB->end(); I != E; I++) {
1947 Instruction *PBI = &*I;
1948 // Check whether Inst and PBI generate the same value.
1949 if (Inst->isIdenticalTo(PBI)) {
1950 Inst->replaceAllUsesWith(PBI);
1951 Inst->eraseFromParent();
1958 /// FoldBranchToCommonDest - If this basic block is simple enough, and if a
1959 /// predecessor branches to us and one of our successors, fold the block into
1960 /// the predecessor and use logical operations to pick the right destination.
1961 bool llvm::FoldBranchToCommonDest(BranchInst *BI) {
1962 BasicBlock *BB = BI->getParent();
1964 Instruction *Cond = nullptr;
1965 if (BI->isConditional())
1966 Cond = dyn_cast<Instruction>(BI->getCondition());
1968 // For unconditional branch, check for a simple CFG pattern, where
1969 // BB has a single predecessor and BB's successor is also its predecessor's
1970 // successor. If such pattern exisits, check for CSE between BB and its
1972 if (BasicBlock *PB = BB->getSinglePredecessor())
1973 if (BranchInst *PBI = dyn_cast<BranchInst>(PB->getTerminator()))
1974 if (PBI->isConditional() &&
1975 (BI->getSuccessor(0) == PBI->getSuccessor(0) ||
1976 BI->getSuccessor(0) == PBI->getSuccessor(1))) {
1977 for (BasicBlock::iterator I = BB->begin(), E = BB->end();
1979 Instruction *Curr = I++;
1980 if (isa<CmpInst>(Curr)) {
1984 // Quit if we can't remove this instruction.
1985 if (!checkCSEInPredecessor(Curr, PB))
1994 if (!Cond || (!isa<CmpInst>(Cond) && !isa<BinaryOperator>(Cond)) ||
1995 Cond->getParent() != BB || !Cond->hasOneUse())
1998 // Only allow this if the condition is a simple instruction that can be
1999 // executed unconditionally. It must be in the same block as the branch, and
2000 // must be at the front of the block.
2001 BasicBlock::iterator FrontIt = BB->front();
2003 // Ignore dbg intrinsics.
2004 while (isa<DbgInfoIntrinsic>(FrontIt)) ++FrontIt;
2006 // Allow a single instruction to be hoisted in addition to the compare
2007 // that feeds the branch. We later ensure that any values that _it_ uses
2008 // were also live in the predecessor, so that we don't unnecessarily create
2009 // register pressure or inhibit out-of-order execution.
2010 Instruction *BonusInst = nullptr;
2011 if (&*FrontIt != Cond &&
2012 FrontIt->hasOneUse() && FrontIt->user_back() == Cond &&
2013 isSafeToSpeculativelyExecute(FrontIt)) {
2014 BonusInst = &*FrontIt;
2017 // Ignore dbg intrinsics.
2018 while (isa<DbgInfoIntrinsic>(FrontIt)) ++FrontIt;
2021 // Only a single bonus inst is allowed.
2022 if (&*FrontIt != Cond)
2025 // Make sure the instruction after the condition is the cond branch.
2026 BasicBlock::iterator CondIt = Cond; ++CondIt;
2028 // Ingore dbg intrinsics.
2029 while (isa<DbgInfoIntrinsic>(CondIt)) ++CondIt;
2034 // Cond is known to be a compare or binary operator. Check to make sure that
2035 // neither operand is a potentially-trapping constant expression.
2036 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Cond->getOperand(0)))
2039 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Cond->getOperand(1)))
2043 // Finally, don't infinitely unroll conditional loops.
2044 BasicBlock *TrueDest = BI->getSuccessor(0);
2045 BasicBlock *FalseDest = (BI->isConditional()) ? BI->getSuccessor(1) : nullptr;
2046 if (TrueDest == BB || FalseDest == BB)
2049 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
2050 BasicBlock *PredBlock = *PI;
2051 BranchInst *PBI = dyn_cast<BranchInst>(PredBlock->getTerminator());
2053 // Check that we have two conditional branches. If there is a PHI node in
2054 // the common successor, verify that the same value flows in from both
2056 SmallVector<PHINode*, 4> PHIs;
2057 if (!PBI || PBI->isUnconditional() ||
2058 (BI->isConditional() &&
2059 !SafeToMergeTerminators(BI, PBI)) ||
2060 (!BI->isConditional() &&
2061 !isProfitableToFoldUnconditional(BI, PBI, Cond, PHIs)))
2064 // Determine if the two branches share a common destination.
2065 Instruction::BinaryOps Opc = Instruction::BinaryOpsEnd;
2066 bool InvertPredCond = false;
2068 if (BI->isConditional()) {
2069 if (PBI->getSuccessor(0) == TrueDest)
2070 Opc = Instruction::Or;
2071 else if (PBI->getSuccessor(1) == FalseDest)
2072 Opc = Instruction::And;
2073 else if (PBI->getSuccessor(0) == FalseDest)
2074 Opc = Instruction::And, InvertPredCond = true;
2075 else if (PBI->getSuccessor(1) == TrueDest)
2076 Opc = Instruction::Or, InvertPredCond = true;
2080 if (PBI->getSuccessor(0) != TrueDest && PBI->getSuccessor(1) != TrueDest)
2084 // Ensure that any values used in the bonus instruction are also used
2085 // by the terminator of the predecessor. This means that those values
2086 // must already have been resolved, so we won't be inhibiting the
2087 // out-of-order core by speculating them earlier. We also allow
2088 // instructions that are used by the terminator's condition because it
2089 // exposes more merging opportunities.
2090 bool UsedByBranch = (BonusInst && BonusInst->hasOneUse() &&
2091 BonusInst->user_back() == Cond);
2093 if (BonusInst && !UsedByBranch) {
2094 // Collect the values used by the bonus inst
2095 SmallPtrSet<Value*, 4> UsedValues;
2096 for (Instruction::op_iterator OI = BonusInst->op_begin(),
2097 OE = BonusInst->op_end(); OI != OE; ++OI) {
2099 if (!isa<Constant>(V) && !isa<Argument>(V))
2100 UsedValues.insert(V);
2103 SmallVector<std::pair<Value*, unsigned>, 4> Worklist;
2104 Worklist.push_back(std::make_pair(PBI->getOperand(0), 0));
2106 // Walk up to four levels back up the use-def chain of the predecessor's
2107 // terminator to see if all those values were used. The choice of four
2108 // levels is arbitrary, to provide a compile-time-cost bound.
2109 while (!Worklist.empty()) {
2110 std::pair<Value*, unsigned> Pair = Worklist.back();
2111 Worklist.pop_back();
2113 if (Pair.second >= 4) continue;
2114 UsedValues.erase(Pair.first);
2115 if (UsedValues.empty()) break;
2117 if (Instruction *I = dyn_cast<Instruction>(Pair.first)) {
2118 for (Instruction::op_iterator OI = I->op_begin(), OE = I->op_end();
2120 Worklist.push_back(std::make_pair(OI->get(), Pair.second+1));
2124 if (!UsedValues.empty()) return false;
2127 DEBUG(dbgs() << "FOLDING BRANCH TO COMMON DEST:\n" << *PBI << *BB);
2128 IRBuilder<> Builder(PBI);
2130 // If we need to invert the condition in the pred block to match, do so now.
2131 if (InvertPredCond) {
2132 Value *NewCond = PBI->getCondition();
2134 if (NewCond->hasOneUse() && isa<CmpInst>(NewCond)) {
2135 CmpInst *CI = cast<CmpInst>(NewCond);
2136 CI->setPredicate(CI->getInversePredicate());
2138 NewCond = Builder.CreateNot(NewCond,
2139 PBI->getCondition()->getName()+".not");
2142 PBI->setCondition(NewCond);
2143 PBI->swapSuccessors();
2146 // If we have a bonus inst, clone it into the predecessor block.
2147 Instruction *NewBonus = nullptr;
2149 NewBonus = BonusInst->clone();
2151 // If we moved a load, we cannot any longer claim any knowledge about
2152 // its potential value. The previous information might have been valid
2153 // only given the branch precondition.
2154 // For an analogous reason, we must also drop all the metadata whose
2155 // semantics we don't understand.
2156 NewBonus->dropUnknownMetadata(LLVMContext::MD_dbg);
2158 PredBlock->getInstList().insert(PBI, NewBonus);
2159 NewBonus->takeName(BonusInst);
2160 BonusInst->setName(BonusInst->getName()+".old");
2163 // Clone Cond into the predecessor basic block, and or/and the
2164 // two conditions together.
2165 Instruction *New = Cond->clone();
2166 if (BonusInst) New->replaceUsesOfWith(BonusInst, NewBonus);
2167 PredBlock->getInstList().insert(PBI, New);
2168 New->takeName(Cond);
2169 Cond->setName(New->getName()+".old");
2171 if (BI->isConditional()) {
2172 Instruction *NewCond =
2173 cast<Instruction>(Builder.CreateBinOp(Opc, PBI->getCondition(),
2175 PBI->setCondition(NewCond);
2177 uint64_t PredTrueWeight, PredFalseWeight, SuccTrueWeight, SuccFalseWeight;
2178 bool PredHasWeights = ExtractBranchMetadata(PBI, PredTrueWeight,
2180 bool SuccHasWeights = ExtractBranchMetadata(BI, SuccTrueWeight,
2182 SmallVector<uint64_t, 8> NewWeights;
2184 if (PBI->getSuccessor(0) == BB) {
2185 if (PredHasWeights && SuccHasWeights) {
2186 // PBI: br i1 %x, BB, FalseDest
2187 // BI: br i1 %y, TrueDest, FalseDest
2188 //TrueWeight is TrueWeight for PBI * TrueWeight for BI.
2189 NewWeights.push_back(PredTrueWeight * SuccTrueWeight);
2190 //FalseWeight is FalseWeight for PBI * TotalWeight for BI +
2191 // TrueWeight for PBI * FalseWeight for BI.
2192 // We assume that total weights of a BranchInst can fit into 32 bits.
2193 // Therefore, we will not have overflow using 64-bit arithmetic.
2194 NewWeights.push_back(PredFalseWeight * (SuccFalseWeight +
2195 SuccTrueWeight) + PredTrueWeight * SuccFalseWeight);
2197 AddPredecessorToBlock(TrueDest, PredBlock, BB);
2198 PBI->setSuccessor(0, TrueDest);
2200 if (PBI->getSuccessor(1) == BB) {
2201 if (PredHasWeights && SuccHasWeights) {
2202 // PBI: br i1 %x, TrueDest, BB
2203 // BI: br i1 %y, TrueDest, FalseDest
2204 //TrueWeight is TrueWeight for PBI * TotalWeight for BI +
2205 // FalseWeight for PBI * TrueWeight for BI.
2206 NewWeights.push_back(PredTrueWeight * (SuccFalseWeight +
2207 SuccTrueWeight) + PredFalseWeight * SuccTrueWeight);
2208 //FalseWeight is FalseWeight for PBI * FalseWeight for BI.
2209 NewWeights.push_back(PredFalseWeight * SuccFalseWeight);
2211 AddPredecessorToBlock(FalseDest, PredBlock, BB);
2212 PBI->setSuccessor(1, FalseDest);
2214 if (NewWeights.size() == 2) {
2215 // Halve the weights if any of them cannot fit in an uint32_t
2216 FitWeights(NewWeights);
2218 SmallVector<uint32_t, 8> MDWeights(NewWeights.begin(),NewWeights.end());
2219 PBI->setMetadata(LLVMContext::MD_prof,
2220 MDBuilder(BI->getContext()).
2221 createBranchWeights(MDWeights));
2223 PBI->setMetadata(LLVMContext::MD_prof, nullptr);
2225 // Update PHI nodes in the common successors.
2226 for (unsigned i = 0, e = PHIs.size(); i != e; ++i) {
2227 ConstantInt *PBI_C = cast<ConstantInt>(
2228 PHIs[i]->getIncomingValueForBlock(PBI->getParent()));
2229 assert(PBI_C->getType()->isIntegerTy(1));
2230 Instruction *MergedCond = nullptr;
2231 if (PBI->getSuccessor(0) == TrueDest) {
2232 // Create (PBI_Cond and PBI_C) or (!PBI_Cond and BI_Value)
2233 // PBI_C is true: PBI_Cond or (!PBI_Cond and BI_Value)
2234 // is false: !PBI_Cond and BI_Value
2235 Instruction *NotCond =
2236 cast<Instruction>(Builder.CreateNot(PBI->getCondition(),
2239 cast<Instruction>(Builder.CreateBinOp(Instruction::And,
2244 cast<Instruction>(Builder.CreateBinOp(Instruction::Or,
2245 PBI->getCondition(), MergedCond,
2248 // Create (PBI_Cond and BI_Value) or (!PBI_Cond and PBI_C)
2249 // PBI_C is true: (PBI_Cond and BI_Value) or (!PBI_Cond)
2250 // is false: PBI_Cond and BI_Value
2252 cast<Instruction>(Builder.CreateBinOp(Instruction::And,
2253 PBI->getCondition(), New,
2255 if (PBI_C->isOne()) {
2256 Instruction *NotCond =
2257 cast<Instruction>(Builder.CreateNot(PBI->getCondition(),
2260 cast<Instruction>(Builder.CreateBinOp(Instruction::Or,
2261 NotCond, MergedCond,
2266 PHIs[i]->setIncomingValue(PHIs[i]->getBasicBlockIndex(PBI->getParent()),
2269 // Change PBI from Conditional to Unconditional.
2270 BranchInst *New_PBI = BranchInst::Create(TrueDest, PBI);
2271 EraseTerminatorInstAndDCECond(PBI);
2275 // TODO: If BB is reachable from all paths through PredBlock, then we
2276 // could replace PBI's branch probabilities with BI's.
2278 // Copy any debug value intrinsics into the end of PredBlock.
2279 for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
2280 if (isa<DbgInfoIntrinsic>(*I))
2281 I->clone()->insertBefore(PBI);
2288 /// SimplifyCondBranchToCondBranch - If we have a conditional branch as a
2289 /// predecessor of another block, this function tries to simplify it. We know
2290 /// that PBI and BI are both conditional branches, and BI is in one of the
2291 /// successor blocks of PBI - PBI branches to BI.
2292 static bool SimplifyCondBranchToCondBranch(BranchInst *PBI, BranchInst *BI) {
2293 assert(PBI->isConditional() && BI->isConditional());
2294 BasicBlock *BB = BI->getParent();
2296 // If this block ends with a branch instruction, and if there is a
2297 // predecessor that ends on a branch of the same condition, make
2298 // this conditional branch redundant.
2299 if (PBI->getCondition() == BI->getCondition() &&
2300 PBI->getSuccessor(0) != PBI->getSuccessor(1)) {
2301 // Okay, the outcome of this conditional branch is statically
2302 // knowable. If this block had a single pred, handle specially.
2303 if (BB->getSinglePredecessor()) {
2304 // Turn this into a branch on constant.
2305 bool CondIsTrue = PBI->getSuccessor(0) == BB;
2306 BI->setCondition(ConstantInt::get(Type::getInt1Ty(BB->getContext()),
2308 return true; // Nuke the branch on constant.
2311 // Otherwise, if there are multiple predecessors, insert a PHI that merges
2312 // in the constant and simplify the block result. Subsequent passes of
2313 // simplifycfg will thread the block.
2314 if (BlockIsSimpleEnoughToThreadThrough(BB)) {
2315 pred_iterator PB = pred_begin(BB), PE = pred_end(BB);
2316 PHINode *NewPN = PHINode::Create(Type::getInt1Ty(BB->getContext()),
2317 std::distance(PB, PE),
2318 BI->getCondition()->getName() + ".pr",
2320 // Okay, we're going to insert the PHI node. Since PBI is not the only
2321 // predecessor, compute the PHI'd conditional value for all of the preds.
2322 // Any predecessor where the condition is not computable we keep symbolic.
2323 for (pred_iterator PI = PB; PI != PE; ++PI) {
2324 BasicBlock *P = *PI;
2325 if ((PBI = dyn_cast<BranchInst>(P->getTerminator())) &&
2326 PBI != BI && PBI->isConditional() &&
2327 PBI->getCondition() == BI->getCondition() &&
2328 PBI->getSuccessor(0) != PBI->getSuccessor(1)) {
2329 bool CondIsTrue = PBI->getSuccessor(0) == BB;
2330 NewPN->addIncoming(ConstantInt::get(Type::getInt1Ty(BB->getContext()),
2333 NewPN->addIncoming(BI->getCondition(), P);
2337 BI->setCondition(NewPN);
2342 // If this is a conditional branch in an empty block, and if any
2343 // predecessors is a conditional branch to one of our destinations,
2344 // fold the conditions into logical ops and one cond br.
2345 BasicBlock::iterator BBI = BB->begin();
2346 // Ignore dbg intrinsics.
2347 while (isa<DbgInfoIntrinsic>(BBI))
2353 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(BI->getCondition()))
2358 if (PBI->getSuccessor(0) == BI->getSuccessor(0))
2360 else if (PBI->getSuccessor(0) == BI->getSuccessor(1))
2361 PBIOp = 0, BIOp = 1;
2362 else if (PBI->getSuccessor(1) == BI->getSuccessor(0))
2363 PBIOp = 1, BIOp = 0;
2364 else if (PBI->getSuccessor(1) == BI->getSuccessor(1))
2369 // Check to make sure that the other destination of this branch
2370 // isn't BB itself. If so, this is an infinite loop that will
2371 // keep getting unwound.
2372 if (PBI->getSuccessor(PBIOp) == BB)
2375 // Do not perform this transformation if it would require
2376 // insertion of a large number of select instructions. For targets
2377 // without predication/cmovs, this is a big pessimization.
2378 BasicBlock *CommonDest = PBI->getSuccessor(PBIOp);
2380 unsigned NumPhis = 0;
2381 for (BasicBlock::iterator II = CommonDest->begin();
2382 isa<PHINode>(II); ++II, ++NumPhis)
2383 if (NumPhis > 2) // Disable this xform.
2386 // Finally, if everything is ok, fold the branches to logical ops.
2387 BasicBlock *OtherDest = BI->getSuccessor(BIOp ^ 1);
2389 DEBUG(dbgs() << "FOLDING BRs:" << *PBI->getParent()
2390 << "AND: " << *BI->getParent());
2393 // If OtherDest *is* BB, then BB is a basic block with a single conditional
2394 // branch in it, where one edge (OtherDest) goes back to itself but the other
2395 // exits. We don't *know* that the program avoids the infinite loop
2396 // (even though that seems likely). If we do this xform naively, we'll end up
2397 // recursively unpeeling the loop. Since we know that (after the xform is
2398 // done) that the block *is* infinite if reached, we just make it an obviously
2399 // infinite loop with no cond branch.
2400 if (OtherDest == BB) {
2401 // Insert it at the end of the function, because it's either code,
2402 // or it won't matter if it's hot. :)
2403 BasicBlock *InfLoopBlock = BasicBlock::Create(BB->getContext(),
2404 "infloop", BB->getParent());
2405 BranchInst::Create(InfLoopBlock, InfLoopBlock);
2406 OtherDest = InfLoopBlock;
2409 DEBUG(dbgs() << *PBI->getParent()->getParent());
2411 // BI may have other predecessors. Because of this, we leave
2412 // it alone, but modify PBI.
2414 // Make sure we get to CommonDest on True&True directions.
2415 Value *PBICond = PBI->getCondition();
2416 IRBuilder<true, NoFolder> Builder(PBI);
2418 PBICond = Builder.CreateNot(PBICond, PBICond->getName()+".not");
2420 Value *BICond = BI->getCondition();
2422 BICond = Builder.CreateNot(BICond, BICond->getName()+".not");
2424 // Merge the conditions.
2425 Value *Cond = Builder.CreateOr(PBICond, BICond, "brmerge");
2427 // Modify PBI to branch on the new condition to the new dests.
2428 PBI->setCondition(Cond);
2429 PBI->setSuccessor(0, CommonDest);
2430 PBI->setSuccessor(1, OtherDest);
2432 // Update branch weight for PBI.
2433 uint64_t PredTrueWeight, PredFalseWeight, SuccTrueWeight, SuccFalseWeight;
2434 bool PredHasWeights = ExtractBranchMetadata(PBI, PredTrueWeight,
2436 bool SuccHasWeights = ExtractBranchMetadata(BI, SuccTrueWeight,
2438 if (PredHasWeights && SuccHasWeights) {
2439 uint64_t PredCommon = PBIOp ? PredFalseWeight : PredTrueWeight;
2440 uint64_t PredOther = PBIOp ?PredTrueWeight : PredFalseWeight;
2441 uint64_t SuccCommon = BIOp ? SuccFalseWeight : SuccTrueWeight;
2442 uint64_t SuccOther = BIOp ? SuccTrueWeight : SuccFalseWeight;
2443 // The weight to CommonDest should be PredCommon * SuccTotal +
2444 // PredOther * SuccCommon.
2445 // The weight to OtherDest should be PredOther * SuccOther.
2446 SmallVector<uint64_t, 2> NewWeights;
2447 NewWeights.push_back(PredCommon * (SuccCommon + SuccOther) +
2448 PredOther * SuccCommon);
2449 NewWeights.push_back(PredOther * SuccOther);
2450 // Halve the weights if any of them cannot fit in an uint32_t
2451 FitWeights(NewWeights);
2453 SmallVector<uint32_t, 2> MDWeights(NewWeights.begin(),NewWeights.end());
2454 PBI->setMetadata(LLVMContext::MD_prof,
2455 MDBuilder(BI->getContext()).
2456 createBranchWeights(MDWeights));
2459 // OtherDest may have phi nodes. If so, add an entry from PBI's
2460 // block that are identical to the entries for BI's block.
2461 AddPredecessorToBlock(OtherDest, PBI->getParent(), BB);
2463 // We know that the CommonDest already had an edge from PBI to
2464 // it. If it has PHIs though, the PHIs may have different
2465 // entries for BB and PBI's BB. If so, insert a select to make
2468 for (BasicBlock::iterator II = CommonDest->begin();
2469 (PN = dyn_cast<PHINode>(II)); ++II) {
2470 Value *BIV = PN->getIncomingValueForBlock(BB);
2471 unsigned PBBIdx = PN->getBasicBlockIndex(PBI->getParent());
2472 Value *PBIV = PN->getIncomingValue(PBBIdx);
2474 // Insert a select in PBI to pick the right value.
2475 Value *NV = cast<SelectInst>
2476 (Builder.CreateSelect(PBICond, PBIV, BIV, PBIV->getName()+".mux"));
2477 PN->setIncomingValue(PBBIdx, NV);
2481 DEBUG(dbgs() << "INTO: " << *PBI->getParent());
2482 DEBUG(dbgs() << *PBI->getParent()->getParent());
2484 // This basic block is probably dead. We know it has at least
2485 // one fewer predecessor.
2489 // SimplifyTerminatorOnSelect - Simplifies a terminator by replacing it with a
2490 // branch to TrueBB if Cond is true or to FalseBB if Cond is false.
2491 // Takes care of updating the successors and removing the old terminator.
2492 // Also makes sure not to introduce new successors by assuming that edges to
2493 // non-successor TrueBBs and FalseBBs aren't reachable.
2494 static bool SimplifyTerminatorOnSelect(TerminatorInst *OldTerm, Value *Cond,
2495 BasicBlock *TrueBB, BasicBlock *FalseBB,
2496 uint32_t TrueWeight,
2497 uint32_t FalseWeight){
2498 // Remove any superfluous successor edges from the CFG.
2499 // First, figure out which successors to preserve.
2500 // If TrueBB and FalseBB are equal, only try to preserve one copy of that
2502 BasicBlock *KeepEdge1 = TrueBB;
2503 BasicBlock *KeepEdge2 = TrueBB != FalseBB ? FalseBB : nullptr;
2505 // Then remove the rest.
2506 for (unsigned I = 0, E = OldTerm->getNumSuccessors(); I != E; ++I) {
2507 BasicBlock *Succ = OldTerm->getSuccessor(I);
2508 // Make sure only to keep exactly one copy of each edge.
2509 if (Succ == KeepEdge1)
2510 KeepEdge1 = nullptr;
2511 else if (Succ == KeepEdge2)
2512 KeepEdge2 = nullptr;
2514 Succ->removePredecessor(OldTerm->getParent());
2517 IRBuilder<> Builder(OldTerm);
2518 Builder.SetCurrentDebugLocation(OldTerm->getDebugLoc());
2520 // Insert an appropriate new terminator.
2521 if (!KeepEdge1 && !KeepEdge2) {
2522 if (TrueBB == FalseBB)
2523 // We were only looking for one successor, and it was present.
2524 // Create an unconditional branch to it.
2525 Builder.CreateBr(TrueBB);
2527 // We found both of the successors we were looking for.
2528 // Create a conditional branch sharing the condition of the select.
2529 BranchInst *NewBI = Builder.CreateCondBr(Cond, TrueBB, FalseBB);
2530 if (TrueWeight != FalseWeight)
2531 NewBI->setMetadata(LLVMContext::MD_prof,
2532 MDBuilder(OldTerm->getContext()).
2533 createBranchWeights(TrueWeight, FalseWeight));
2535 } else if (KeepEdge1 && (KeepEdge2 || TrueBB == FalseBB)) {
2536 // Neither of the selected blocks were successors, so this
2537 // terminator must be unreachable.
2538 new UnreachableInst(OldTerm->getContext(), OldTerm);
2540 // One of the selected values was a successor, but the other wasn't.
2541 // Insert an unconditional branch to the one that was found;
2542 // the edge to the one that wasn't must be unreachable.
2544 // Only TrueBB was found.
2545 Builder.CreateBr(TrueBB);
2547 // Only FalseBB was found.
2548 Builder.CreateBr(FalseBB);
2551 EraseTerminatorInstAndDCECond(OldTerm);
2555 // SimplifySwitchOnSelect - Replaces
2556 // (switch (select cond, X, Y)) on constant X, Y
2557 // with a branch - conditional if X and Y lead to distinct BBs,
2558 // unconditional otherwise.
2559 static bool SimplifySwitchOnSelect(SwitchInst *SI, SelectInst *Select) {
2560 // Check for constant integer values in the select.
2561 ConstantInt *TrueVal = dyn_cast<ConstantInt>(Select->getTrueValue());
2562 ConstantInt *FalseVal = dyn_cast<ConstantInt>(Select->getFalseValue());
2563 if (!TrueVal || !FalseVal)
2566 // Find the relevant condition and destinations.
2567 Value *Condition = Select->getCondition();
2568 BasicBlock *TrueBB = SI->findCaseValue(TrueVal).getCaseSuccessor();
2569 BasicBlock *FalseBB = SI->findCaseValue(FalseVal).getCaseSuccessor();
2571 // Get weight for TrueBB and FalseBB.
2572 uint32_t TrueWeight = 0, FalseWeight = 0;
2573 SmallVector<uint64_t, 8> Weights;
2574 bool HasWeights = HasBranchWeights(SI);
2576 GetBranchWeights(SI, Weights);
2577 if (Weights.size() == 1 + SI->getNumCases()) {
2578 TrueWeight = (uint32_t)Weights[SI->findCaseValue(TrueVal).
2579 getSuccessorIndex()];
2580 FalseWeight = (uint32_t)Weights[SI->findCaseValue(FalseVal).
2581 getSuccessorIndex()];
2585 // Perform the actual simplification.
2586 return SimplifyTerminatorOnSelect(SI, Condition, TrueBB, FalseBB,
2587 TrueWeight, FalseWeight);
2590 // SimplifyIndirectBrOnSelect - Replaces
2591 // (indirectbr (select cond, blockaddress(@fn, BlockA),
2592 // blockaddress(@fn, BlockB)))
2594 // (br cond, BlockA, BlockB).
2595 static bool SimplifyIndirectBrOnSelect(IndirectBrInst *IBI, SelectInst *SI) {
2596 // Check that both operands of the select are block addresses.
2597 BlockAddress *TBA = dyn_cast<BlockAddress>(SI->getTrueValue());
2598 BlockAddress *FBA = dyn_cast<BlockAddress>(SI->getFalseValue());
2602 // Extract the actual blocks.
2603 BasicBlock *TrueBB = TBA->getBasicBlock();
2604 BasicBlock *FalseBB = FBA->getBasicBlock();
2606 // Perform the actual simplification.
2607 return SimplifyTerminatorOnSelect(IBI, SI->getCondition(), TrueBB, FalseBB,
2611 /// TryToSimplifyUncondBranchWithICmpInIt - This is called when we find an icmp
2612 /// instruction (a seteq/setne with a constant) as the only instruction in a
2613 /// block that ends with an uncond branch. We are looking for a very specific
2614 /// pattern that occurs when "A == 1 || A == 2 || A == 3" gets simplified. In
2615 /// this case, we merge the first two "or's of icmp" into a switch, but then the
2616 /// default value goes to an uncond block with a seteq in it, we get something
2619 /// switch i8 %A, label %DEFAULT [ i8 1, label %end i8 2, label %end ]
2621 /// %tmp = icmp eq i8 %A, 92
2624 /// ... = phi i1 [ true, %entry ], [ %tmp, %DEFAULT ], [ true, %entry ]
2626 /// We prefer to split the edge to 'end' so that there is a true/false entry to
2627 /// the PHI, merging the third icmp into the switch.
2628 static bool TryToSimplifyUncondBranchWithICmpInIt(
2629 ICmpInst *ICI, IRBuilder<> &Builder, const TargetTransformInfo &TTI,
2630 const DataLayout *DL) {
2631 BasicBlock *BB = ICI->getParent();
2633 // If the block has any PHIs in it or the icmp has multiple uses, it is too
2635 if (isa<PHINode>(BB->begin()) || !ICI->hasOneUse()) return false;
2637 Value *V = ICI->getOperand(0);
2638 ConstantInt *Cst = cast<ConstantInt>(ICI->getOperand(1));
2640 // The pattern we're looking for is where our only predecessor is a switch on
2641 // 'V' and this block is the default case for the switch. In this case we can
2642 // fold the compared value into the switch to simplify things.
2643 BasicBlock *Pred = BB->getSinglePredecessor();
2644 if (!Pred || !isa<SwitchInst>(Pred->getTerminator())) return false;
2646 SwitchInst *SI = cast<SwitchInst>(Pred->getTerminator());
2647 if (SI->getCondition() != V)
2650 // If BB is reachable on a non-default case, then we simply know the value of
2651 // V in this block. Substitute it and constant fold the icmp instruction
2653 if (SI->getDefaultDest() != BB) {
2654 ConstantInt *VVal = SI->findCaseDest(BB);
2655 assert(VVal && "Should have a unique destination value");
2656 ICI->setOperand(0, VVal);
2658 if (Value *V = SimplifyInstruction(ICI, DL)) {
2659 ICI->replaceAllUsesWith(V);
2660 ICI->eraseFromParent();
2662 // BB is now empty, so it is likely to simplify away.
2663 return SimplifyCFG(BB, TTI, DL) | true;
2666 // Ok, the block is reachable from the default dest. If the constant we're
2667 // comparing exists in one of the other edges, then we can constant fold ICI
2669 if (SI->findCaseValue(Cst) != SI->case_default()) {
2671 if (ICI->getPredicate() == ICmpInst::ICMP_EQ)
2672 V = ConstantInt::getFalse(BB->getContext());
2674 V = ConstantInt::getTrue(BB->getContext());
2676 ICI->replaceAllUsesWith(V);
2677 ICI->eraseFromParent();
2678 // BB is now empty, so it is likely to simplify away.
2679 return SimplifyCFG(BB, TTI, DL) | true;
2682 // The use of the icmp has to be in the 'end' block, by the only PHI node in
2684 BasicBlock *SuccBlock = BB->getTerminator()->getSuccessor(0);
2685 PHINode *PHIUse = dyn_cast<PHINode>(ICI->user_back());
2686 if (PHIUse == nullptr || PHIUse != &SuccBlock->front() ||
2687 isa<PHINode>(++BasicBlock::iterator(PHIUse)))
2690 // If the icmp is a SETEQ, then the default dest gets false, the new edge gets
2692 Constant *DefaultCst = ConstantInt::getTrue(BB->getContext());
2693 Constant *NewCst = ConstantInt::getFalse(BB->getContext());
2695 if (ICI->getPredicate() == ICmpInst::ICMP_EQ)
2696 std::swap(DefaultCst, NewCst);
2698 // Replace ICI (which is used by the PHI for the default value) with true or
2699 // false depending on if it is EQ or NE.
2700 ICI->replaceAllUsesWith(DefaultCst);
2701 ICI->eraseFromParent();
2703 // Okay, the switch goes to this block on a default value. Add an edge from
2704 // the switch to the merge point on the compared value.
2705 BasicBlock *NewBB = BasicBlock::Create(BB->getContext(), "switch.edge",
2706 BB->getParent(), BB);
2707 SmallVector<uint64_t, 8> Weights;
2708 bool HasWeights = HasBranchWeights(SI);
2710 GetBranchWeights(SI, Weights);
2711 if (Weights.size() == 1 + SI->getNumCases()) {
2712 // Split weight for default case to case for "Cst".
2713 Weights[0] = (Weights[0]+1) >> 1;
2714 Weights.push_back(Weights[0]);
2716 SmallVector<uint32_t, 8> MDWeights(Weights.begin(), Weights.end());
2717 SI->setMetadata(LLVMContext::MD_prof,
2718 MDBuilder(SI->getContext()).
2719 createBranchWeights(MDWeights));
2722 SI->addCase(Cst, NewBB);
2724 // NewBB branches to the phi block, add the uncond branch and the phi entry.
2725 Builder.SetInsertPoint(NewBB);
2726 Builder.SetCurrentDebugLocation(SI->getDebugLoc());
2727 Builder.CreateBr(SuccBlock);
2728 PHIUse->addIncoming(NewCst, NewBB);
2732 /// SimplifyBranchOnICmpChain - The specified branch is a conditional branch.
2733 /// Check to see if it is branching on an or/and chain of icmp instructions, and
2734 /// fold it into a switch instruction if so.
2735 static bool SimplifyBranchOnICmpChain(BranchInst *BI, const DataLayout *DL,
2736 IRBuilder<> &Builder) {
2737 Instruction *Cond = dyn_cast<Instruction>(BI->getCondition());
2738 if (!Cond) return false;
2741 // Change br (X == 0 | X == 1), T, F into a switch instruction.
2742 // If this is a bunch of seteq's or'd together, or if it's a bunch of
2743 // 'setne's and'ed together, collect them.
2744 Value *CompVal = nullptr;
2745 std::vector<ConstantInt*> Values;
2746 bool TrueWhenEqual = true;
2747 Value *ExtraCase = nullptr;
2748 unsigned UsedICmps = 0;
2750 if (Cond->getOpcode() == Instruction::Or) {
2751 CompVal = GatherConstantCompares(Cond, Values, ExtraCase, DL, true,
2753 } else if (Cond->getOpcode() == Instruction::And) {
2754 CompVal = GatherConstantCompares(Cond, Values, ExtraCase, DL, false,
2756 TrueWhenEqual = false;
2759 // If we didn't have a multiply compared value, fail.
2760 if (!CompVal) return false;
2762 // Avoid turning single icmps into a switch.
2766 // There might be duplicate constants in the list, which the switch
2767 // instruction can't handle, remove them now.
2768 array_pod_sort(Values.begin(), Values.end(), ConstantIntSortPredicate);
2769 Values.erase(std::unique(Values.begin(), Values.end()), Values.end());
2771 // If Extra was used, we require at least two switch values to do the
2772 // transformation. A switch with one value is just an cond branch.
2773 if (ExtraCase && Values.size() < 2) return false;
2775 // TODO: Preserve branch weight metadata, similarly to how
2776 // FoldValueComparisonIntoPredecessors preserves it.
2778 // Figure out which block is which destination.
2779 BasicBlock *DefaultBB = BI->getSuccessor(1);
2780 BasicBlock *EdgeBB = BI->getSuccessor(0);
2781 if (!TrueWhenEqual) std::swap(DefaultBB, EdgeBB);
2783 BasicBlock *BB = BI->getParent();
2785 DEBUG(dbgs() << "Converting 'icmp' chain with " << Values.size()
2786 << " cases into SWITCH. BB is:\n" << *BB);
2788 // If there are any extra values that couldn't be folded into the switch
2789 // then we evaluate them with an explicit branch first. Split the block
2790 // right before the condbr to handle it.
2792 BasicBlock *NewBB = BB->splitBasicBlock(BI, "switch.early.test");
2793 // Remove the uncond branch added to the old block.
2794 TerminatorInst *OldTI = BB->getTerminator();
2795 Builder.SetInsertPoint(OldTI);
2798 Builder.CreateCondBr(ExtraCase, EdgeBB, NewBB);
2800 Builder.CreateCondBr(ExtraCase, NewBB, EdgeBB);
2802 OldTI->eraseFromParent();
2804 // If there are PHI nodes in EdgeBB, then we need to add a new entry to them
2805 // for the edge we just added.
2806 AddPredecessorToBlock(EdgeBB, BB, NewBB);
2808 DEBUG(dbgs() << " ** 'icmp' chain unhandled condition: " << *ExtraCase
2809 << "\nEXTRABB = " << *BB);
2813 Builder.SetInsertPoint(BI);
2814 // Convert pointer to int before we switch.
2815 if (CompVal->getType()->isPointerTy()) {
2816 assert(DL && "Cannot switch on pointer without DataLayout");
2817 CompVal = Builder.CreatePtrToInt(CompVal,
2818 DL->getIntPtrType(CompVal->getType()),
2822 // Create the new switch instruction now.
2823 SwitchInst *New = Builder.CreateSwitch(CompVal, DefaultBB, Values.size());
2825 // Add all of the 'cases' to the switch instruction.
2826 for (unsigned i = 0, e = Values.size(); i != e; ++i)
2827 New->addCase(Values[i], EdgeBB);
2829 // We added edges from PI to the EdgeBB. As such, if there were any
2830 // PHI nodes in EdgeBB, they need entries to be added corresponding to
2831 // the number of edges added.
2832 for (BasicBlock::iterator BBI = EdgeBB->begin();
2833 isa<PHINode>(BBI); ++BBI) {
2834 PHINode *PN = cast<PHINode>(BBI);
2835 Value *InVal = PN->getIncomingValueForBlock(BB);
2836 for (unsigned i = 0, e = Values.size()-1; i != e; ++i)
2837 PN->addIncoming(InVal, BB);
2840 // Erase the old branch instruction.
2841 EraseTerminatorInstAndDCECond(BI);
2843 DEBUG(dbgs() << " ** 'icmp' chain result is:\n" << *BB << '\n');
2847 bool SimplifyCFGOpt::SimplifyResume(ResumeInst *RI, IRBuilder<> &Builder) {
2848 // If this is a trivial landing pad that just continues unwinding the caught
2849 // exception then zap the landing pad, turning its invokes into calls.
2850 BasicBlock *BB = RI->getParent();
2851 LandingPadInst *LPInst = dyn_cast<LandingPadInst>(BB->getFirstNonPHI());
2852 if (RI->getValue() != LPInst)
2853 // Not a landing pad, or the resume is not unwinding the exception that
2854 // caused control to branch here.
2857 // Check that there are no other instructions except for debug intrinsics.
2858 BasicBlock::iterator I = LPInst, E = RI;
2860 if (!isa<DbgInfoIntrinsic>(I))
2863 // Turn all invokes that unwind here into calls and delete the basic block.
2864 bool InvokeRequiresTableEntry = false;
2865 bool Changed = false;
2866 for (pred_iterator PI = pred_begin(BB), PE = pred_end(BB); PI != PE;) {
2867 InvokeInst *II = cast<InvokeInst>((*PI++)->getTerminator());
2869 if (II->hasFnAttr(Attribute::UWTable)) {
2870 // Don't remove an `invoke' instruction if the ABI requires an entry into
2872 InvokeRequiresTableEntry = true;
2876 SmallVector<Value*, 8> Args(II->op_begin(), II->op_end() - 3);
2878 // Insert a call instruction before the invoke.
2879 CallInst *Call = CallInst::Create(II->getCalledValue(), Args, "", II);
2881 Call->setCallingConv(II->getCallingConv());
2882 Call->setAttributes(II->getAttributes());
2883 Call->setDebugLoc(II->getDebugLoc());
2885 // Anything that used the value produced by the invoke instruction now uses
2886 // the value produced by the call instruction. Note that we do this even
2887 // for void functions and calls with no uses so that the callgraph edge is
2889 II->replaceAllUsesWith(Call);
2890 BB->removePredecessor(II->getParent());
2892 // Insert a branch to the normal destination right before the invoke.
2893 BranchInst::Create(II->getNormalDest(), II);
2895 // Finally, delete the invoke instruction!
2896 II->eraseFromParent();
2900 if (!InvokeRequiresTableEntry)
2901 // The landingpad is now unreachable. Zap it.
2902 BB->eraseFromParent();
2907 bool SimplifyCFGOpt::SimplifyReturn(ReturnInst *RI, IRBuilder<> &Builder) {
2908 BasicBlock *BB = RI->getParent();
2909 if (!BB->getFirstNonPHIOrDbg()->isTerminator()) return false;
2911 // Find predecessors that end with branches.
2912 SmallVector<BasicBlock*, 8> UncondBranchPreds;
2913 SmallVector<BranchInst*, 8> CondBranchPreds;
2914 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
2915 BasicBlock *P = *PI;
2916 TerminatorInst *PTI = P->getTerminator();
2917 if (BranchInst *BI = dyn_cast<BranchInst>(PTI)) {
2918 if (BI->isUnconditional())
2919 UncondBranchPreds.push_back(P);
2921 CondBranchPreds.push_back(BI);
2925 // If we found some, do the transformation!
2926 if (!UncondBranchPreds.empty() && DupRet) {
2927 while (!UncondBranchPreds.empty()) {
2928 BasicBlock *Pred = UncondBranchPreds.pop_back_val();
2929 DEBUG(dbgs() << "FOLDING: " << *BB
2930 << "INTO UNCOND BRANCH PRED: " << *Pred);
2931 (void)FoldReturnIntoUncondBranch(RI, BB, Pred);
2934 // If we eliminated all predecessors of the block, delete the block now.
2935 if (pred_begin(BB) == pred_end(BB))
2936 // We know there are no successors, so just nuke the block.
2937 BB->eraseFromParent();
2942 // Check out all of the conditional branches going to this return
2943 // instruction. If any of them just select between returns, change the
2944 // branch itself into a select/return pair.
2945 while (!CondBranchPreds.empty()) {
2946 BranchInst *BI = CondBranchPreds.pop_back_val();
2948 // Check to see if the non-BB successor is also a return block.
2949 if (isa<ReturnInst>(BI->getSuccessor(0)->getTerminator()) &&
2950 isa<ReturnInst>(BI->getSuccessor(1)->getTerminator()) &&
2951 SimplifyCondBranchToTwoReturns(BI, Builder))
2957 bool SimplifyCFGOpt::SimplifyUnreachable(UnreachableInst *UI) {
2958 BasicBlock *BB = UI->getParent();
2960 bool Changed = false;
2962 // If there are any instructions immediately before the unreachable that can
2963 // be removed, do so.
2964 while (UI != BB->begin()) {
2965 BasicBlock::iterator BBI = UI;
2967 // Do not delete instructions that can have side effects which might cause
2968 // the unreachable to not be reachable; specifically, calls and volatile
2969 // operations may have this effect.
2970 if (isa<CallInst>(BBI) && !isa<DbgInfoIntrinsic>(BBI)) break;
2972 if (BBI->mayHaveSideEffects()) {
2973 if (StoreInst *SI = dyn_cast<StoreInst>(BBI)) {
2974 if (SI->isVolatile())
2976 } else if (LoadInst *LI = dyn_cast<LoadInst>(BBI)) {
2977 if (LI->isVolatile())
2979 } else if (AtomicRMWInst *RMWI = dyn_cast<AtomicRMWInst>(BBI)) {
2980 if (RMWI->isVolatile())
2982 } else if (AtomicCmpXchgInst *CXI = dyn_cast<AtomicCmpXchgInst>(BBI)) {
2983 if (CXI->isVolatile())
2985 } else if (!isa<FenceInst>(BBI) && !isa<VAArgInst>(BBI) &&
2986 !isa<LandingPadInst>(BBI)) {
2989 // Note that deleting LandingPad's here is in fact okay, although it
2990 // involves a bit of subtle reasoning. If this inst is a LandingPad,
2991 // all the predecessors of this block will be the unwind edges of Invokes,
2992 // and we can therefore guarantee this block will be erased.
2995 // Delete this instruction (any uses are guaranteed to be dead)
2996 if (!BBI->use_empty())
2997 BBI->replaceAllUsesWith(UndefValue::get(BBI->getType()));
2998 BBI->eraseFromParent();
3002 // If the unreachable instruction is the first in the block, take a gander
3003 // at all of the predecessors of this instruction, and simplify them.
3004 if (&BB->front() != UI) return Changed;
3006 SmallVector<BasicBlock*, 8> Preds(pred_begin(BB), pred_end(BB));
3007 for (unsigned i = 0, e = Preds.size(); i != e; ++i) {
3008 TerminatorInst *TI = Preds[i]->getTerminator();
3009 IRBuilder<> Builder(TI);
3010 if (BranchInst *BI = dyn_cast<BranchInst>(TI)) {
3011 if (BI->isUnconditional()) {
3012 if (BI->getSuccessor(0) == BB) {
3013 new UnreachableInst(TI->getContext(), TI);
3014 TI->eraseFromParent();
3018 if (BI->getSuccessor(0) == BB) {
3019 Builder.CreateBr(BI->getSuccessor(1));
3020 EraseTerminatorInstAndDCECond(BI);
3021 } else if (BI->getSuccessor(1) == BB) {
3022 Builder.CreateBr(BI->getSuccessor(0));
3023 EraseTerminatorInstAndDCECond(BI);
3027 } else if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) {
3028 for (SwitchInst::CaseIt i = SI->case_begin(), e = SI->case_end();
3030 if (i.getCaseSuccessor() == BB) {
3031 BB->removePredecessor(SI->getParent());
3036 // If the default value is unreachable, figure out the most popular
3037 // destination and make it the default.
3038 if (SI->getDefaultDest() == BB) {
3039 std::map<BasicBlock*, std::pair<unsigned, unsigned> > Popularity;
3040 for (SwitchInst::CaseIt i = SI->case_begin(), e = SI->case_end();
3042 std::pair<unsigned, unsigned> &entry =
3043 Popularity[i.getCaseSuccessor()];
3044 if (entry.first == 0) {
3046 entry.second = i.getCaseIndex();
3052 // Find the most popular block.
3053 unsigned MaxPop = 0;
3054 unsigned MaxIndex = 0;
3055 BasicBlock *MaxBlock = nullptr;
3056 for (std::map<BasicBlock*, std::pair<unsigned, unsigned> >::iterator
3057 I = Popularity.begin(), E = Popularity.end(); I != E; ++I) {
3058 if (I->second.first > MaxPop ||
3059 (I->second.first == MaxPop && MaxIndex > I->second.second)) {
3060 MaxPop = I->second.first;
3061 MaxIndex = I->second.second;
3062 MaxBlock = I->first;
3066 // Make this the new default, allowing us to delete any explicit
3068 SI->setDefaultDest(MaxBlock);
3071 // If MaxBlock has phinodes in it, remove MaxPop-1 entries from
3073 if (isa<PHINode>(MaxBlock->begin()))
3074 for (unsigned i = 0; i != MaxPop-1; ++i)
3075 MaxBlock->removePredecessor(SI->getParent());
3077 for (SwitchInst::CaseIt i = SI->case_begin(), e = SI->case_end();
3079 if (i.getCaseSuccessor() == MaxBlock) {
3085 } else if (InvokeInst *II = dyn_cast<InvokeInst>(TI)) {
3086 if (II->getUnwindDest() == BB) {
3087 // Convert the invoke to a call instruction. This would be a good
3088 // place to note that the call does not throw though.
3089 BranchInst *BI = Builder.CreateBr(II->getNormalDest());
3090 II->removeFromParent(); // Take out of symbol table
3092 // Insert the call now...
3093 SmallVector<Value*, 8> Args(II->op_begin(), II->op_end()-3);
3094 Builder.SetInsertPoint(BI);
3095 CallInst *CI = Builder.CreateCall(II->getCalledValue(),
3096 Args, II->getName());
3097 CI->setCallingConv(II->getCallingConv());
3098 CI->setAttributes(II->getAttributes());
3099 // If the invoke produced a value, the call does now instead.
3100 II->replaceAllUsesWith(CI);
3107 // If this block is now dead, remove it.
3108 if (pred_begin(BB) == pred_end(BB) &&
3109 BB != &BB->getParent()->getEntryBlock()) {
3110 // We know there are no successors, so just nuke the block.
3111 BB->eraseFromParent();
3118 /// TurnSwitchRangeIntoICmp - Turns a switch with that contains only a
3119 /// integer range comparison into a sub, an icmp and a branch.
3120 static bool TurnSwitchRangeIntoICmp(SwitchInst *SI, IRBuilder<> &Builder) {
3121 assert(SI->getNumCases() > 1 && "Degenerate switch?");
3123 // Make sure all cases point to the same destination and gather the values.
3124 SmallVector<ConstantInt *, 16> Cases;
3125 SwitchInst::CaseIt I = SI->case_begin();
3126 Cases.push_back(I.getCaseValue());
3127 SwitchInst::CaseIt PrevI = I++;
3128 for (SwitchInst::CaseIt E = SI->case_end(); I != E; PrevI = I++) {
3129 if (PrevI.getCaseSuccessor() != I.getCaseSuccessor())
3131 Cases.push_back(I.getCaseValue());
3133 assert(Cases.size() == SI->getNumCases() && "Not all cases gathered");
3135 // Sort the case values, then check if they form a range we can transform.
3136 array_pod_sort(Cases.begin(), Cases.end(), ConstantIntSortPredicate);
3137 for (unsigned I = 1, E = Cases.size(); I != E; ++I) {
3138 if (Cases[I-1]->getValue() != Cases[I]->getValue()+1)
3142 Constant *Offset = ConstantExpr::getNeg(Cases.back());
3143 Constant *NumCases = ConstantInt::get(Offset->getType(), SI->getNumCases());
3145 Value *Sub = SI->getCondition();
3146 if (!Offset->isNullValue())
3147 Sub = Builder.CreateAdd(Sub, Offset, Sub->getName()+".off");
3149 // If NumCases overflowed, then all possible values jump to the successor.
3150 if (NumCases->isNullValue() && SI->getNumCases() != 0)
3151 Cmp = ConstantInt::getTrue(SI->getContext());
3153 Cmp = Builder.CreateICmpULT(Sub, NumCases, "switch");
3154 BranchInst *NewBI = Builder.CreateCondBr(
3155 Cmp, SI->case_begin().getCaseSuccessor(), SI->getDefaultDest());
3157 // Update weight for the newly-created conditional branch.
3158 SmallVector<uint64_t, 8> Weights;
3159 bool HasWeights = HasBranchWeights(SI);
3161 GetBranchWeights(SI, Weights);
3162 if (Weights.size() == 1 + SI->getNumCases()) {
3163 // Combine all weights for the cases to be the true weight of NewBI.
3164 // We assume that the sum of all weights for a Terminator can fit into 32
3166 uint32_t NewTrueWeight = 0;
3167 for (unsigned I = 1, E = Weights.size(); I != E; ++I)
3168 NewTrueWeight += (uint32_t)Weights[I];
3169 NewBI->setMetadata(LLVMContext::MD_prof,
3170 MDBuilder(SI->getContext()).
3171 createBranchWeights(NewTrueWeight,
3172 (uint32_t)Weights[0]));
3176 // Prune obsolete incoming values off the successor's PHI nodes.
3177 for (BasicBlock::iterator BBI = SI->case_begin().getCaseSuccessor()->begin();
3178 isa<PHINode>(BBI); ++BBI) {
3179 for (unsigned I = 0, E = SI->getNumCases()-1; I != E; ++I)
3180 cast<PHINode>(BBI)->removeIncomingValue(SI->getParent());
3182 SI->eraseFromParent();
3187 /// EliminateDeadSwitchCases - Compute masked bits for the condition of a switch
3188 /// and use it to remove dead cases.
3189 static bool EliminateDeadSwitchCases(SwitchInst *SI) {
3190 Value *Cond = SI->getCondition();
3191 unsigned Bits = Cond->getType()->getIntegerBitWidth();
3192 APInt KnownZero(Bits, 0), KnownOne(Bits, 0);
3193 computeKnownBits(Cond, KnownZero, KnownOne);
3195 // Gather dead cases.
3196 SmallVector<ConstantInt*, 8> DeadCases;
3197 for (SwitchInst::CaseIt I = SI->case_begin(), E = SI->case_end(); I != E; ++I) {
3198 if ((I.getCaseValue()->getValue() & KnownZero) != 0 ||
3199 (I.getCaseValue()->getValue() & KnownOne) != KnownOne) {
3200 DeadCases.push_back(I.getCaseValue());
3201 DEBUG(dbgs() << "SimplifyCFG: switch case '"
3202 << I.getCaseValue() << "' is dead.\n");
3206 SmallVector<uint64_t, 8> Weights;
3207 bool HasWeight = HasBranchWeights(SI);
3209 GetBranchWeights(SI, Weights);
3210 HasWeight = (Weights.size() == 1 + SI->getNumCases());
3213 // Remove dead cases from the switch.
3214 for (unsigned I = 0, E = DeadCases.size(); I != E; ++I) {
3215 SwitchInst::CaseIt Case = SI->findCaseValue(DeadCases[I]);
3216 assert(Case != SI->case_default() &&
3217 "Case was not found. Probably mistake in DeadCases forming.");
3219 std::swap(Weights[Case.getCaseIndex()+1], Weights.back());
3223 // Prune unused values from PHI nodes.
3224 Case.getCaseSuccessor()->removePredecessor(SI->getParent());
3225 SI->removeCase(Case);
3227 if (HasWeight && Weights.size() >= 2) {
3228 SmallVector<uint32_t, 8> MDWeights(Weights.begin(), Weights.end());
3229 SI->setMetadata(LLVMContext::MD_prof,
3230 MDBuilder(SI->getParent()->getContext()).
3231 createBranchWeights(MDWeights));
3234 return !DeadCases.empty();
3237 /// FindPHIForConditionForwarding - If BB would be eligible for simplification
3238 /// by TryToSimplifyUncondBranchFromEmptyBlock (i.e. it is empty and terminated
3239 /// by an unconditional branch), look at the phi node for BB in the successor
3240 /// block and see if the incoming value is equal to CaseValue. If so, return
3241 /// the phi node, and set PhiIndex to BB's index in the phi node.
3242 static PHINode *FindPHIForConditionForwarding(ConstantInt *CaseValue,
3245 if (BB->getFirstNonPHIOrDbg() != BB->getTerminator())
3246 return nullptr; // BB must be empty to be a candidate for simplification.
3247 if (!BB->getSinglePredecessor())
3248 return nullptr; // BB must be dominated by the switch.
3250 BranchInst *Branch = dyn_cast<BranchInst>(BB->getTerminator());
3251 if (!Branch || !Branch->isUnconditional())
3252 return nullptr; // Terminator must be unconditional branch.
3254 BasicBlock *Succ = Branch->getSuccessor(0);
3256 BasicBlock::iterator I = Succ->begin();
3257 while (PHINode *PHI = dyn_cast<PHINode>(I++)) {
3258 int Idx = PHI->getBasicBlockIndex(BB);
3259 assert(Idx >= 0 && "PHI has no entry for predecessor?");
3261 Value *InValue = PHI->getIncomingValue(Idx);
3262 if (InValue != CaseValue) continue;
3271 /// ForwardSwitchConditionToPHI - Try to forward the condition of a switch
3272 /// instruction to a phi node dominated by the switch, if that would mean that
3273 /// some of the destination blocks of the switch can be folded away.
3274 /// Returns true if a change is made.
3275 static bool ForwardSwitchConditionToPHI(SwitchInst *SI) {
3276 typedef DenseMap<PHINode*, SmallVector<int,4> > ForwardingNodesMap;
3277 ForwardingNodesMap ForwardingNodes;
3279 for (SwitchInst::CaseIt I = SI->case_begin(), E = SI->case_end(); I != E; ++I) {
3280 ConstantInt *CaseValue = I.getCaseValue();
3281 BasicBlock *CaseDest = I.getCaseSuccessor();
3284 PHINode *PHI = FindPHIForConditionForwarding(CaseValue, CaseDest,
3288 ForwardingNodes[PHI].push_back(PhiIndex);
3291 bool Changed = false;
3293 for (ForwardingNodesMap::iterator I = ForwardingNodes.begin(),
3294 E = ForwardingNodes.end(); I != E; ++I) {
3295 PHINode *Phi = I->first;
3296 SmallVectorImpl<int> &Indexes = I->second;
3298 if (Indexes.size() < 2) continue;
3300 for (size_t I = 0, E = Indexes.size(); I != E; ++I)
3301 Phi->setIncomingValue(Indexes[I], SI->getCondition());
3308 /// ValidLookupTableConstant - Return true if the backend will be able to handle
3309 /// initializing an array of constants like C.
3310 static bool ValidLookupTableConstant(Constant *C) {
3311 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C))
3312 return CE->isGEPWithNoNotionalOverIndexing();
3314 return isa<ConstantFP>(C) ||
3315 isa<ConstantInt>(C) ||
3316 isa<ConstantPointerNull>(C) ||
3317 isa<GlobalValue>(C) ||
3321 /// LookupConstant - If V is a Constant, return it. Otherwise, try to look up
3322 /// its constant value in ConstantPool, returning 0 if it's not there.
3323 static Constant *LookupConstant(Value *V,
3324 const SmallDenseMap<Value*, Constant*>& ConstantPool) {
3325 if (Constant *C = dyn_cast<Constant>(V))
3327 return ConstantPool.lookup(V);
3330 /// ConstantFold - Try to fold instruction I into a constant. This works for
3331 /// simple instructions such as binary operations where both operands are
3332 /// constant or can be replaced by constants from the ConstantPool. Returns the
3333 /// resulting constant on success, 0 otherwise.
3335 ConstantFold(Instruction *I,
3336 const SmallDenseMap<Value *, Constant *> &ConstantPool,
3337 const DataLayout *DL) {
3338 if (SelectInst *Select = dyn_cast<SelectInst>(I)) {
3339 Constant *A = LookupConstant(Select->getCondition(), ConstantPool);
3342 if (A->isAllOnesValue())
3343 return LookupConstant(Select->getTrueValue(), ConstantPool);
3344 if (A->isNullValue())
3345 return LookupConstant(Select->getFalseValue(), ConstantPool);
3349 SmallVector<Constant *, 4> COps;
3350 for (unsigned N = 0, E = I->getNumOperands(); N != E; ++N) {
3351 if (Constant *A = LookupConstant(I->getOperand(N), ConstantPool))
3357 if (CmpInst *Cmp = dyn_cast<CmpInst>(I))
3358 return ConstantFoldCompareInstOperands(Cmp->getPredicate(), COps[0],
3361 return ConstantFoldInstOperands(I->getOpcode(), I->getType(), COps, DL);
3364 /// GetCaseResults - Try to determine the resulting constant values in phi nodes
3365 /// at the common destination basic block, *CommonDest, for one of the case
3366 /// destionations CaseDest corresponding to value CaseVal (0 for the default
3367 /// case), of a switch instruction SI.
3369 GetCaseResults(SwitchInst *SI,
3370 ConstantInt *CaseVal,
3371 BasicBlock *CaseDest,
3372 BasicBlock **CommonDest,
3373 SmallVectorImpl<std::pair<PHINode *, Constant *> > &Res,
3374 const DataLayout *DL) {
3375 // The block from which we enter the common destination.
3376 BasicBlock *Pred = SI->getParent();
3378 // If CaseDest is empty except for some side-effect free instructions through
3379 // which we can constant-propagate the CaseVal, continue to its successor.
3380 SmallDenseMap<Value*, Constant*> ConstantPool;
3381 ConstantPool.insert(std::make_pair(SI->getCondition(), CaseVal));
3382 for (BasicBlock::iterator I = CaseDest->begin(), E = CaseDest->end(); I != E;
3384 if (TerminatorInst *T = dyn_cast<TerminatorInst>(I)) {
3385 // If the terminator is a simple branch, continue to the next block.
3386 if (T->getNumSuccessors() != 1)
3389 CaseDest = T->getSuccessor(0);
3390 } else if (isa<DbgInfoIntrinsic>(I)) {
3391 // Skip debug intrinsic.
3393 } else if (Constant *C = ConstantFold(I, ConstantPool, DL)) {
3394 // Instruction is side-effect free and constant.
3395 ConstantPool.insert(std::make_pair(I, C));
3401 // If we did not have a CommonDest before, use the current one.
3403 *CommonDest = CaseDest;
3404 // If the destination isn't the common one, abort.
3405 if (CaseDest != *CommonDest)
3408 // Get the values for this case from phi nodes in the destination block.
3409 BasicBlock::iterator I = (*CommonDest)->begin();
3410 while (PHINode *PHI = dyn_cast<PHINode>(I++)) {
3411 int Idx = PHI->getBasicBlockIndex(Pred);
3415 Constant *ConstVal = LookupConstant(PHI->getIncomingValue(Idx),
3420 // Note: If the constant comes from constant-propagating the case value
3421 // through the CaseDest basic block, it will be safe to remove the
3422 // instructions in that block. They cannot be used (except in the phi nodes
3423 // we visit) outside CaseDest, because that block does not dominate its
3424 // successor. If it did, we would not be in this phi node.
3426 // Be conservative about which kinds of constants we support.
3427 if (!ValidLookupTableConstant(ConstVal))
3430 Res.push_back(std::make_pair(PHI, ConstVal));
3433 return Res.size() > 0;
3437 /// SwitchLookupTable - This class represents a lookup table that can be used
3438 /// to replace a switch.
3439 class SwitchLookupTable {
3441 /// SwitchLookupTable - Create a lookup table to use as a switch replacement
3442 /// with the contents of Values, using DefaultValue to fill any holes in the
3444 SwitchLookupTable(Module &M,
3446 ConstantInt *Offset,
3447 const SmallVectorImpl<std::pair<ConstantInt*, Constant*> >& Values,
3448 Constant *DefaultValue,
3449 const DataLayout *DL);
3451 /// BuildLookup - Build instructions with Builder to retrieve the value at
3452 /// the position given by Index in the lookup table.
3453 Value *BuildLookup(Value *Index, IRBuilder<> &Builder);
3455 /// WouldFitInRegister - Return true if a table with TableSize elements of
3456 /// type ElementType would fit in a target-legal register.
3457 static bool WouldFitInRegister(const DataLayout *DL,
3459 const Type *ElementType);
3462 // Depending on the contents of the table, it can be represented in
3465 // For tables where each element contains the same value, we just have to
3466 // store that single value and return it for each lookup.
3469 // For small tables with integer elements, we can pack them into a bitmap
3470 // that fits into a target-legal register. Values are retrieved by
3471 // shift and mask operations.
3474 // The table is stored as an array of values. Values are retrieved by load
3475 // instructions from the table.
3479 // For SingleValueKind, this is the single value.
3480 Constant *SingleValue;
3482 // For BitMapKind, this is the bitmap.
3483 ConstantInt *BitMap;
3484 IntegerType *BitMapElementTy;
3486 // For ArrayKind, this is the array.
3487 GlobalVariable *Array;
3491 SwitchLookupTable::SwitchLookupTable(Module &M,
3493 ConstantInt *Offset,
3494 const SmallVectorImpl<std::pair<ConstantInt*, Constant*> >& Values,
3495 Constant *DefaultValue,
3496 const DataLayout *DL)
3497 : SingleValue(nullptr), BitMap(nullptr), BitMapElementTy(nullptr),
3499 assert(Values.size() && "Can't build lookup table without values!");
3500 assert(TableSize >= Values.size() && "Can't fit values in table!");
3502 // If all values in the table are equal, this is that value.
3503 SingleValue = Values.begin()->second;
3505 Type *ValueType = Values.begin()->second->getType();
3507 // Build up the table contents.
3508 SmallVector<Constant*, 64> TableContents(TableSize);
3509 for (size_t I = 0, E = Values.size(); I != E; ++I) {
3510 ConstantInt *CaseVal = Values[I].first;
3511 Constant *CaseRes = Values[I].second;
3512 assert(CaseRes->getType() == ValueType);
3514 uint64_t Idx = (CaseVal->getValue() - Offset->getValue())
3516 TableContents[Idx] = CaseRes;
3518 if (CaseRes != SingleValue)
3519 SingleValue = nullptr;
3522 // Fill in any holes in the table with the default result.
3523 if (Values.size() < TableSize) {
3524 assert(DefaultValue && "Need a default value to fill the lookup table holes.");
3525 assert(DefaultValue->getType() == ValueType);
3526 for (uint64_t I = 0; I < TableSize; ++I) {
3527 if (!TableContents[I])
3528 TableContents[I] = DefaultValue;
3531 if (DefaultValue != SingleValue)
3532 SingleValue = nullptr;
3535 // If each element in the table contains the same value, we only need to store
3536 // that single value.
3538 Kind = SingleValueKind;
3542 // If the type is integer and the table fits in a register, build a bitmap.
3543 if (WouldFitInRegister(DL, TableSize, ValueType)) {
3544 IntegerType *IT = cast<IntegerType>(ValueType);
3545 APInt TableInt(TableSize * IT->getBitWidth(), 0);
3546 for (uint64_t I = TableSize; I > 0; --I) {
3547 TableInt <<= IT->getBitWidth();
3548 // Insert values into the bitmap. Undef values are set to zero.
3549 if (!isa<UndefValue>(TableContents[I - 1])) {
3550 ConstantInt *Val = cast<ConstantInt>(TableContents[I - 1]);
3551 TableInt |= Val->getValue().zext(TableInt.getBitWidth());
3554 BitMap = ConstantInt::get(M.getContext(), TableInt);
3555 BitMapElementTy = IT;
3561 // Store the table in an array.
3562 ArrayType *ArrayTy = ArrayType::get(ValueType, TableSize);
3563 Constant *Initializer = ConstantArray::get(ArrayTy, TableContents);
3565 Array = new GlobalVariable(M, ArrayTy, /*constant=*/ true,
3566 GlobalVariable::PrivateLinkage,
3569 Array->setUnnamedAddr(true);
3573 Value *SwitchLookupTable::BuildLookup(Value *Index, IRBuilder<> &Builder) {
3575 case SingleValueKind:
3578 // Type of the bitmap (e.g. i59).
3579 IntegerType *MapTy = BitMap->getType();
3581 // Cast Index to the same type as the bitmap.
3582 // Note: The Index is <= the number of elements in the table, so
3583 // truncating it to the width of the bitmask is safe.
3584 Value *ShiftAmt = Builder.CreateZExtOrTrunc(Index, MapTy, "switch.cast");
3586 // Multiply the shift amount by the element width.
3587 ShiftAmt = Builder.CreateMul(ShiftAmt,
3588 ConstantInt::get(MapTy, BitMapElementTy->getBitWidth()),
3592 Value *DownShifted = Builder.CreateLShr(BitMap, ShiftAmt,
3593 "switch.downshift");
3595 return Builder.CreateTrunc(DownShifted, BitMapElementTy,
3599 Value *GEPIndices[] = { Builder.getInt32(0), Index };
3600 Value *GEP = Builder.CreateInBoundsGEP(Array, GEPIndices,
3602 return Builder.CreateLoad(GEP, "switch.load");
3605 llvm_unreachable("Unknown lookup table kind!");
3608 bool SwitchLookupTable::WouldFitInRegister(const DataLayout *DL,
3610 const Type *ElementType) {
3613 const IntegerType *IT = dyn_cast<IntegerType>(ElementType);
3616 // FIXME: If the type is wider than it needs to be, e.g. i8 but all values
3617 // are <= 15, we could try to narrow the type.
3619 // Avoid overflow, fitsInLegalInteger uses unsigned int for the width.
3620 if (TableSize >= UINT_MAX/IT->getBitWidth())
3622 return DL->fitsInLegalInteger(TableSize * IT->getBitWidth());
3625 /// ShouldBuildLookupTable - Determine whether a lookup table should be built
3626 /// for this switch, based on the number of cases, size of the table and the
3627 /// types of the results.
3628 static bool ShouldBuildLookupTable(SwitchInst *SI,
3630 const TargetTransformInfo &TTI,
3631 const DataLayout *DL,
3632 const SmallDenseMap<PHINode*, Type*>& ResultTypes) {
3633 if (SI->getNumCases() > TableSize || TableSize >= UINT64_MAX / 10)
3634 return false; // TableSize overflowed, or mul below might overflow.
3636 bool AllTablesFitInRegister = true;
3637 bool HasIllegalType = false;
3638 for (SmallDenseMap<PHINode*, Type*>::const_iterator I = ResultTypes.begin(),
3639 E = ResultTypes.end(); I != E; ++I) {
3640 Type *Ty = I->second;
3642 // Saturate this flag to true.
3643 HasIllegalType = HasIllegalType || !TTI.isTypeLegal(Ty);
3645 // Saturate this flag to false.
3646 AllTablesFitInRegister = AllTablesFitInRegister &&
3647 SwitchLookupTable::WouldFitInRegister(DL, TableSize, Ty);
3649 // If both flags saturate, we're done. NOTE: This *only* works with
3650 // saturating flags, and all flags have to saturate first due to the
3651 // non-deterministic behavior of iterating over a dense map.
3652 if (HasIllegalType && !AllTablesFitInRegister)
3656 // If each table would fit in a register, we should build it anyway.
3657 if (AllTablesFitInRegister)
3660 // Don't build a table that doesn't fit in-register if it has illegal types.
3664 // The table density should be at least 40%. This is the same criterion as for
3665 // jump tables, see SelectionDAGBuilder::handleJTSwitchCase.
3666 // FIXME: Find the best cut-off.
3667 return SI->getNumCases() * 10 >= TableSize * 4;
3670 /// SwitchToLookupTable - If the switch is only used to initialize one or more
3671 /// phi nodes in a common successor block with different constant values,
3672 /// replace the switch with lookup tables.
3673 static bool SwitchToLookupTable(SwitchInst *SI,
3674 IRBuilder<> &Builder,
3675 const TargetTransformInfo &TTI,
3676 const DataLayout* DL) {
3677 assert(SI->getNumCases() > 1 && "Degenerate switch?");
3679 // Only build lookup table when we have a target that supports it.
3680 if (!TTI.shouldBuildLookupTables())
3683 // FIXME: If the switch is too sparse for a lookup table, perhaps we could
3684 // split off a dense part and build a lookup table for that.
3686 // FIXME: This creates arrays of GEPs to constant strings, which means each
3687 // GEP needs a runtime relocation in PIC code. We should just build one big
3688 // string and lookup indices into that.
3690 // Ignore switches with less than three cases. Lookup tables will not make them
3691 // faster, so we don't analyze them.
3692 if (SI->getNumCases() < 3)
3695 // Figure out the corresponding result for each case value and phi node in the
3696 // common destination, as well as the the min and max case values.
3697 assert(SI->case_begin() != SI->case_end());
3698 SwitchInst::CaseIt CI = SI->case_begin();
3699 ConstantInt *MinCaseVal = CI.getCaseValue();
3700 ConstantInt *MaxCaseVal = CI.getCaseValue();
3702 BasicBlock *CommonDest = nullptr;
3703 typedef SmallVector<std::pair<ConstantInt*, Constant*>, 4> ResultListTy;
3704 SmallDenseMap<PHINode*, ResultListTy> ResultLists;
3705 SmallDenseMap<PHINode*, Constant*> DefaultResults;
3706 SmallDenseMap<PHINode*, Type*> ResultTypes;
3707 SmallVector<PHINode*, 4> PHIs;
3709 for (SwitchInst::CaseIt E = SI->case_end(); CI != E; ++CI) {
3710 ConstantInt *CaseVal = CI.getCaseValue();
3711 if (CaseVal->getValue().slt(MinCaseVal->getValue()))
3712 MinCaseVal = CaseVal;
3713 if (CaseVal->getValue().sgt(MaxCaseVal->getValue()))
3714 MaxCaseVal = CaseVal;
3716 // Resulting value at phi nodes for this case value.
3717 typedef SmallVector<std::pair<PHINode*, Constant*>, 4> ResultsTy;
3719 if (!GetCaseResults(SI, CaseVal, CI.getCaseSuccessor(), &CommonDest,
3723 // Append the result from this case to the list for each phi.
3724 for (ResultsTy::iterator I = Results.begin(), E = Results.end(); I!=E; ++I) {
3725 if (!ResultLists.count(I->first))
3726 PHIs.push_back(I->first);
3727 ResultLists[I->first].push_back(std::make_pair(CaseVal, I->second));
3731 // Keep track of the result types.
3732 for (size_t I = 0, E = PHIs.size(); I != E; ++I) {
3733 PHINode *PHI = PHIs[I];
3734 ResultTypes[PHI] = ResultLists[PHI][0].second->getType();
3737 uint64_t NumResults = ResultLists[PHIs[0]].size();
3738 APInt RangeSpread = MaxCaseVal->getValue() - MinCaseVal->getValue();
3739 uint64_t TableSize = RangeSpread.getLimitedValue() + 1;
3740 bool TableHasHoles = (NumResults < TableSize);
3742 // If the table has holes, we need a constant result for the default case
3743 // or a bitmask that fits in a register.
3744 SmallVector<std::pair<PHINode*, Constant*>, 4> DefaultResultsList;
3745 bool HasDefaultResults = false;
3746 if (TableHasHoles) {
3747 HasDefaultResults = GetCaseResults(SI, nullptr, SI->getDefaultDest(),
3748 &CommonDest, DefaultResultsList, DL);
3750 bool NeedMask = (TableHasHoles && !HasDefaultResults);
3752 // As an extra penalty for the validity test we require more cases.
3753 if (SI->getNumCases() < 4) // FIXME: Find best threshold value (benchmark).
3755 if (!(DL && DL->fitsInLegalInteger(TableSize)))
3759 for (size_t I = 0, E = DefaultResultsList.size(); I != E; ++I) {
3760 PHINode *PHI = DefaultResultsList[I].first;
3761 Constant *Result = DefaultResultsList[I].second;
3762 DefaultResults[PHI] = Result;
3765 if (!ShouldBuildLookupTable(SI, TableSize, TTI, DL, ResultTypes))
3768 // Create the BB that does the lookups.
3769 Module &Mod = *CommonDest->getParent()->getParent();
3770 BasicBlock *LookupBB = BasicBlock::Create(Mod.getContext(),
3772 CommonDest->getParent(),
3775 // Compute the table index value.
3776 Builder.SetInsertPoint(SI);
3777 Value *TableIndex = Builder.CreateSub(SI->getCondition(), MinCaseVal,
3780 // Compute the maximum table size representable by the integer type we are
3782 unsigned CaseSize = MinCaseVal->getType()->getPrimitiveSizeInBits();
3783 uint64_t MaxTableSize = CaseSize > 63 ? UINT64_MAX : 1ULL << CaseSize;
3784 assert(MaxTableSize >= TableSize &&
3785 "It is impossible for a switch to have more entries than the max "
3786 "representable value of its input integer type's size.");
3788 // If we have a fully covered lookup table, unconditionally branch to the
3789 // lookup table BB. Otherwise, check if the condition value is within the case
3790 // range. If it is so, branch to the new BB. Otherwise branch to SI's default
3792 const bool GeneratingCoveredLookupTable = MaxTableSize == TableSize;
3793 if (GeneratingCoveredLookupTable) {
3794 Builder.CreateBr(LookupBB);
3795 SI->getDefaultDest()->removePredecessor(SI->getParent());
3797 Value *Cmp = Builder.CreateICmpULT(TableIndex, ConstantInt::get(
3798 MinCaseVal->getType(), TableSize));
3799 Builder.CreateCondBr(Cmp, LookupBB, SI->getDefaultDest());
3802 // Populate the BB that does the lookups.
3803 Builder.SetInsertPoint(LookupBB);
3806 // Before doing the lookup we do the hole check.
3807 // The LookupBB is therefore re-purposed to do the hole check
3808 // and we create a new LookupBB.
3809 BasicBlock *MaskBB = LookupBB;
3810 MaskBB->setName("switch.hole_check");
3811 LookupBB = BasicBlock::Create(Mod.getContext(),
3813 CommonDest->getParent(),
3816 // Build bitmask; fill in a 1 bit for every case.
3817 APInt MaskInt(TableSize, 0);
3818 APInt One(TableSize, 1);
3819 const ResultListTy &ResultList = ResultLists[PHIs[0]];
3820 for (size_t I = 0, E = ResultList.size(); I != E; ++I) {
3821 uint64_t Idx = (ResultList[I].first->getValue() -
3822 MinCaseVal->getValue()).getLimitedValue();
3823 MaskInt |= One << Idx;
3825 ConstantInt *TableMask = ConstantInt::get(Mod.getContext(), MaskInt);
3827 // Get the TableIndex'th bit of the bitmask.
3828 // If this bit is 0 (meaning hole) jump to the default destination,
3829 // else continue with table lookup.
3830 IntegerType *MapTy = TableMask->getType();
3831 Value *MaskIndex = Builder.CreateZExtOrTrunc(TableIndex, MapTy,
3832 "switch.maskindex");
3833 Value *Shifted = Builder.CreateLShr(TableMask, MaskIndex,
3835 Value *LoBit = Builder.CreateTrunc(Shifted,
3836 Type::getInt1Ty(Mod.getContext()),
3838 Builder.CreateCondBr(LoBit, LookupBB, SI->getDefaultDest());
3840 Builder.SetInsertPoint(LookupBB);
3841 AddPredecessorToBlock(SI->getDefaultDest(), MaskBB, SI->getParent());
3844 bool ReturnedEarly = false;
3845 for (size_t I = 0, E = PHIs.size(); I != E; ++I) {
3846 PHINode *PHI = PHIs[I];
3848 // If using a bitmask, use any value to fill the lookup table holes.
3849 Constant *DV = NeedMask ? ResultLists[PHI][0].second : DefaultResults[PHI];
3850 SwitchLookupTable Table(Mod, TableSize, MinCaseVal, ResultLists[PHI],
3853 Value *Result = Table.BuildLookup(TableIndex, Builder);
3855 // If the result is used to return immediately from the function, we want to
3856 // do that right here.
3857 if (PHI->hasOneUse() && isa<ReturnInst>(*PHI->user_begin()) &&
3858 PHI->user_back() == CommonDest->getFirstNonPHIOrDbg()) {
3859 Builder.CreateRet(Result);
3860 ReturnedEarly = true;
3864 PHI->addIncoming(Result, LookupBB);
3868 Builder.CreateBr(CommonDest);
3870 // Remove the switch.
3871 for (unsigned i = 0, e = SI->getNumSuccessors(); i < e; ++i) {
3872 BasicBlock *Succ = SI->getSuccessor(i);
3874 if (Succ == SI->getDefaultDest())
3876 Succ->removePredecessor(SI->getParent());
3878 SI->eraseFromParent();
3882 ++NumLookupTablesHoles;
3886 bool SimplifyCFGOpt::SimplifySwitch(SwitchInst *SI, IRBuilder<> &Builder) {
3887 BasicBlock *BB = SI->getParent();
3889 if (isValueEqualityComparison(SI)) {
3890 // If we only have one predecessor, and if it is a branch on this value,
3891 // see if that predecessor totally determines the outcome of this switch.
3892 if (BasicBlock *OnlyPred = BB->getSinglePredecessor())
3893 if (SimplifyEqualityComparisonWithOnlyPredecessor(SI, OnlyPred, Builder))
3894 return SimplifyCFG(BB, TTI, DL) | true;
3896 Value *Cond = SI->getCondition();
3897 if (SelectInst *Select = dyn_cast<SelectInst>(Cond))
3898 if (SimplifySwitchOnSelect(SI, Select))
3899 return SimplifyCFG(BB, TTI, DL) | true;
3901 // If the block only contains the switch, see if we can fold the block
3902 // away into any preds.
3903 BasicBlock::iterator BBI = BB->begin();
3904 // Ignore dbg intrinsics.
3905 while (isa<DbgInfoIntrinsic>(BBI))
3908 if (FoldValueComparisonIntoPredecessors(SI, Builder))
3909 return SimplifyCFG(BB, TTI, DL) | true;
3912 // Try to transform the switch into an icmp and a branch.
3913 if (TurnSwitchRangeIntoICmp(SI, Builder))
3914 return SimplifyCFG(BB, TTI, DL) | true;
3916 // Remove unreachable cases.
3917 if (EliminateDeadSwitchCases(SI))
3918 return SimplifyCFG(BB, TTI, DL) | true;
3920 if (ForwardSwitchConditionToPHI(SI))
3921 return SimplifyCFG(BB, TTI, DL) | true;
3923 if (SwitchToLookupTable(SI, Builder, TTI, DL))
3924 return SimplifyCFG(BB, TTI, DL) | true;
3929 bool SimplifyCFGOpt::SimplifyIndirectBr(IndirectBrInst *IBI) {
3930 BasicBlock *BB = IBI->getParent();
3931 bool Changed = false;
3933 // Eliminate redundant destinations.
3934 SmallPtrSet<Value *, 8> Succs;
3935 for (unsigned i = 0, e = IBI->getNumDestinations(); i != e; ++i) {
3936 BasicBlock *Dest = IBI->getDestination(i);
3937 if (!Dest->hasAddressTaken() || !Succs.insert(Dest)) {
3938 Dest->removePredecessor(BB);
3939 IBI->removeDestination(i);
3945 if (IBI->getNumDestinations() == 0) {
3946 // If the indirectbr has no successors, change it to unreachable.
3947 new UnreachableInst(IBI->getContext(), IBI);
3948 EraseTerminatorInstAndDCECond(IBI);
3952 if (IBI->getNumDestinations() == 1) {
3953 // If the indirectbr has one successor, change it to a direct branch.
3954 BranchInst::Create(IBI->getDestination(0), IBI);
3955 EraseTerminatorInstAndDCECond(IBI);
3959 if (SelectInst *SI = dyn_cast<SelectInst>(IBI->getAddress())) {
3960 if (SimplifyIndirectBrOnSelect(IBI, SI))
3961 return SimplifyCFG(BB, TTI, DL) | true;
3966 bool SimplifyCFGOpt::SimplifyUncondBranch(BranchInst *BI, IRBuilder<> &Builder){
3967 BasicBlock *BB = BI->getParent();
3969 if (SinkCommon && SinkThenElseCodeToEnd(BI))
3972 // If the Terminator is the only non-phi instruction, simplify the block.
3973 BasicBlock::iterator I = BB->getFirstNonPHIOrDbgOrLifetime();
3974 if (I->isTerminator() && BB != &BB->getParent()->getEntryBlock() &&
3975 TryToSimplifyUncondBranchFromEmptyBlock(BB))
3978 // If the only instruction in the block is a seteq/setne comparison
3979 // against a constant, try to simplify the block.
3980 if (ICmpInst *ICI = dyn_cast<ICmpInst>(I))
3981 if (ICI->isEquality() && isa<ConstantInt>(ICI->getOperand(1))) {
3982 for (++I; isa<DbgInfoIntrinsic>(I); ++I)
3984 if (I->isTerminator() &&
3985 TryToSimplifyUncondBranchWithICmpInIt(ICI, Builder, TTI, DL))
3989 // If this basic block is ONLY a compare and a branch, and if a predecessor
3990 // branches to us and our successor, fold the comparison into the
3991 // predecessor and use logical operations to update the incoming value
3992 // for PHI nodes in common successor.
3993 if (FoldBranchToCommonDest(BI))
3994 return SimplifyCFG(BB, TTI, DL) | true;
3999 bool SimplifyCFGOpt::SimplifyCondBranch(BranchInst *BI, IRBuilder<> &Builder) {
4000 BasicBlock *BB = BI->getParent();
4002 // Conditional branch
4003 if (isValueEqualityComparison(BI)) {
4004 // If we only have one predecessor, and if it is a branch on this value,
4005 // see if that predecessor totally determines the outcome of this
4007 if (BasicBlock *OnlyPred = BB->getSinglePredecessor())
4008 if (SimplifyEqualityComparisonWithOnlyPredecessor(BI, OnlyPred, Builder))
4009 return SimplifyCFG(BB, TTI, DL) | true;
4011 // This block must be empty, except for the setcond inst, if it exists.
4012 // Ignore dbg intrinsics.
4013 BasicBlock::iterator I = BB->begin();
4014 // Ignore dbg intrinsics.
4015 while (isa<DbgInfoIntrinsic>(I))
4018 if (FoldValueComparisonIntoPredecessors(BI, Builder))
4019 return SimplifyCFG(BB, TTI, DL) | true;
4020 } else if (&*I == cast<Instruction>(BI->getCondition())){
4022 // Ignore dbg intrinsics.
4023 while (isa<DbgInfoIntrinsic>(I))
4025 if (&*I == BI && FoldValueComparisonIntoPredecessors(BI, Builder))
4026 return SimplifyCFG(BB, TTI, DL) | true;
4030 // Try to turn "br (X == 0 | X == 1), T, F" into a switch instruction.
4031 if (SimplifyBranchOnICmpChain(BI, DL, Builder))
4034 // If this basic block is ONLY a compare and a branch, and if a predecessor
4035 // branches to us and one of our successors, fold the comparison into the
4036 // predecessor and use logical operations to pick the right destination.
4037 if (FoldBranchToCommonDest(BI))
4038 return SimplifyCFG(BB, TTI, DL) | true;
4040 // We have a conditional branch to two blocks that are only reachable
4041 // from BI. We know that the condbr dominates the two blocks, so see if
4042 // there is any identical code in the "then" and "else" blocks. If so, we
4043 // can hoist it up to the branching block.
4044 if (BI->getSuccessor(0)->getSinglePredecessor()) {
4045 if (BI->getSuccessor(1)->getSinglePredecessor()) {
4046 if (HoistThenElseCodeToIf(BI))
4047 return SimplifyCFG(BB, TTI, DL) | true;
4049 // If Successor #1 has multiple preds, we may be able to conditionally
4050 // execute Successor #0 if it branches to successor #1.
4051 TerminatorInst *Succ0TI = BI->getSuccessor(0)->getTerminator();
4052 if (Succ0TI->getNumSuccessors() == 1 &&
4053 Succ0TI->getSuccessor(0) == BI->getSuccessor(1))
4054 if (SpeculativelyExecuteBB(BI, BI->getSuccessor(0)))
4055 return SimplifyCFG(BB, TTI, DL) | true;
4057 } else if (BI->getSuccessor(1)->getSinglePredecessor()) {
4058 // If Successor #0 has multiple preds, we may be able to conditionally
4059 // execute Successor #1 if it branches to successor #0.
4060 TerminatorInst *Succ1TI = BI->getSuccessor(1)->getTerminator();
4061 if (Succ1TI->getNumSuccessors() == 1 &&
4062 Succ1TI->getSuccessor(0) == BI->getSuccessor(0))
4063 if (SpeculativelyExecuteBB(BI, BI->getSuccessor(1)))
4064 return SimplifyCFG(BB, TTI, DL) | true;
4067 // If this is a branch on a phi node in the current block, thread control
4068 // through this block if any PHI node entries are constants.
4069 if (PHINode *PN = dyn_cast<PHINode>(BI->getCondition()))
4070 if (PN->getParent() == BI->getParent())
4071 if (FoldCondBranchOnPHI(BI, DL))
4072 return SimplifyCFG(BB, TTI, DL) | true;
4074 // Scan predecessor blocks for conditional branches.
4075 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI)
4076 if (BranchInst *PBI = dyn_cast<BranchInst>((*PI)->getTerminator()))
4077 if (PBI != BI && PBI->isConditional())
4078 if (SimplifyCondBranchToCondBranch(PBI, BI))
4079 return SimplifyCFG(BB, TTI, DL) | true;
4084 /// Check if passing a value to an instruction will cause undefined behavior.
4085 static bool passingValueIsAlwaysUndefined(Value *V, Instruction *I) {
4086 Constant *C = dyn_cast<Constant>(V);
4093 if (C->isNullValue()) {
4094 // Only look at the first use, avoid hurting compile time with long uselists
4095 User *Use = *I->user_begin();
4097 // Now make sure that there are no instructions in between that can alter
4098 // control flow (eg. calls)
4099 for (BasicBlock::iterator i = ++BasicBlock::iterator(I); &*i != Use; ++i)
4100 if (i == I->getParent()->end() || i->mayHaveSideEffects())
4103 // Look through GEPs. A load from a GEP derived from NULL is still undefined
4104 if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(Use))
4105 if (GEP->getPointerOperand() == I)
4106 return passingValueIsAlwaysUndefined(V, GEP);
4108 // Look through bitcasts.
4109 if (BitCastInst *BC = dyn_cast<BitCastInst>(Use))
4110 return passingValueIsAlwaysUndefined(V, BC);
4112 // Load from null is undefined.
4113 if (LoadInst *LI = dyn_cast<LoadInst>(Use))
4114 if (!LI->isVolatile())
4115 return LI->getPointerAddressSpace() == 0;
4117 // Store to null is undefined.
4118 if (StoreInst *SI = dyn_cast<StoreInst>(Use))
4119 if (!SI->isVolatile())
4120 return SI->getPointerAddressSpace() == 0 && SI->getPointerOperand() == I;
4125 /// If BB has an incoming value that will always trigger undefined behavior
4126 /// (eg. null pointer dereference), remove the branch leading here.
4127 static bool removeUndefIntroducingPredecessor(BasicBlock *BB) {
4128 for (BasicBlock::iterator i = BB->begin();
4129 PHINode *PHI = dyn_cast<PHINode>(i); ++i)
4130 for (unsigned i = 0, e = PHI->getNumIncomingValues(); i != e; ++i)
4131 if (passingValueIsAlwaysUndefined(PHI->getIncomingValue(i), PHI)) {
4132 TerminatorInst *T = PHI->getIncomingBlock(i)->getTerminator();
4133 IRBuilder<> Builder(T);
4134 if (BranchInst *BI = dyn_cast<BranchInst>(T)) {
4135 BB->removePredecessor(PHI->getIncomingBlock(i));
4136 // Turn uncoditional branches into unreachables and remove the dead
4137 // destination from conditional branches.
4138 if (BI->isUnconditional())
4139 Builder.CreateUnreachable();
4141 Builder.CreateBr(BI->getSuccessor(0) == BB ? BI->getSuccessor(1) :
4142 BI->getSuccessor(0));
4143 BI->eraseFromParent();
4146 // TODO: SwitchInst.
4152 bool SimplifyCFGOpt::run(BasicBlock *BB) {
4153 bool Changed = false;
4155 assert(BB && BB->getParent() && "Block not embedded in function!");
4156 assert(BB->getTerminator() && "Degenerate basic block encountered!");
4158 // Remove basic blocks that have no predecessors (except the entry block)...
4159 // or that just have themself as a predecessor. These are unreachable.
4160 if ((pred_begin(BB) == pred_end(BB) &&
4161 BB != &BB->getParent()->getEntryBlock()) ||
4162 BB->getSinglePredecessor() == BB) {
4163 DEBUG(dbgs() << "Removing BB: \n" << *BB);
4164 DeleteDeadBlock(BB);
4168 // Check to see if we can constant propagate this terminator instruction
4170 Changed |= ConstantFoldTerminator(BB, true);
4172 // Check for and eliminate duplicate PHI nodes in this block.
4173 Changed |= EliminateDuplicatePHINodes(BB);
4175 // Check for and remove branches that will always cause undefined behavior.
4176 Changed |= removeUndefIntroducingPredecessor(BB);
4178 // Merge basic blocks into their predecessor if there is only one distinct
4179 // pred, and if there is only one distinct successor of the predecessor, and
4180 // if there are no PHI nodes.
4182 if (MergeBlockIntoPredecessor(BB))
4185 IRBuilder<> Builder(BB);
4187 // If there is a trivial two-entry PHI node in this basic block, and we can
4188 // eliminate it, do so now.
4189 if (PHINode *PN = dyn_cast<PHINode>(BB->begin()))
4190 if (PN->getNumIncomingValues() == 2)
4191 Changed |= FoldTwoEntryPHINode(PN, DL);
4193 Builder.SetInsertPoint(BB->getTerminator());
4194 if (BranchInst *BI = dyn_cast<BranchInst>(BB->getTerminator())) {
4195 if (BI->isUnconditional()) {
4196 if (SimplifyUncondBranch(BI, Builder)) return true;
4198 if (SimplifyCondBranch(BI, Builder)) return true;
4200 } else if (ReturnInst *RI = dyn_cast<ReturnInst>(BB->getTerminator())) {
4201 if (SimplifyReturn(RI, Builder)) return true;
4202 } else if (ResumeInst *RI = dyn_cast<ResumeInst>(BB->getTerminator())) {
4203 if (SimplifyResume(RI, Builder)) return true;
4204 } else if (SwitchInst *SI = dyn_cast<SwitchInst>(BB->getTerminator())) {
4205 if (SimplifySwitch(SI, Builder)) return true;
4206 } else if (UnreachableInst *UI =
4207 dyn_cast<UnreachableInst>(BB->getTerminator())) {
4208 if (SimplifyUnreachable(UI)) return true;
4209 } else if (IndirectBrInst *IBI =
4210 dyn_cast<IndirectBrInst>(BB->getTerminator())) {
4211 if (SimplifyIndirectBr(IBI)) return true;
4217 /// SimplifyCFG - This function is used to do simplification of a CFG. For
4218 /// example, it adjusts branches to branches to eliminate the extra hop, it
4219 /// eliminates unreachable basic blocks, and does other "peephole" optimization
4220 /// of the CFG. It returns true if a modification was made.
4222 bool llvm::SimplifyCFG(BasicBlock *BB, const TargetTransformInfo &TTI,
4223 const DataLayout *DL) {
4224 return SimplifyCFGOpt(TTI, DL).run(BB);