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 #define DEBUG_TYPE "simplifycfg"
15 #include "llvm/Transforms/Utils/Local.h"
16 #include "llvm/Constants.h"
17 #include "llvm/DerivedTypes.h"
18 #include "llvm/GlobalVariable.h"
19 #include "llvm/IRBuilder.h"
20 #include "llvm/Instructions.h"
21 #include "llvm/IntrinsicInst.h"
22 #include "llvm/LLVMContext.h"
23 #include "llvm/MDBuilder.h"
24 #include "llvm/Metadata.h"
25 #include "llvm/Module.h"
26 #include "llvm/Operator.h"
27 #include "llvm/Type.h"
28 #include "llvm/ADT/DenseMap.h"
29 #include "llvm/ADT/STLExtras.h"
30 #include "llvm/ADT/SetVector.h"
31 #include "llvm/ADT/SmallPtrSet.h"
32 #include "llvm/ADT/SmallVector.h"
33 #include "llvm/ADT/Statistic.h"
34 #include "llvm/Analysis/InstructionSimplify.h"
35 #include "llvm/Analysis/ValueTracking.h"
36 #include "llvm/Support/CFG.h"
37 #include "llvm/Support/CommandLine.h"
38 #include "llvm/Support/ConstantRange.h"
39 #include "llvm/Support/Debug.h"
40 #include "llvm/Support/NoFolder.h"
41 #include "llvm/Support/raw_ostream.h"
42 #include "llvm/Target/TargetData.h"
43 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
49 static cl::opt<unsigned>
50 PHINodeFoldingThreshold("phi-node-folding-threshold", cl::Hidden, cl::init(1),
51 cl::desc("Control the amount of phi node folding to perform (default = 1)"));
54 DupRet("simplifycfg-dup-ret", cl::Hidden, cl::init(false),
55 cl::desc("Duplicate return instructions into unconditional branches"));
58 SinkCommon("simplifycfg-sink-common", cl::Hidden, cl::init(true),
59 cl::desc("Sink common instructions down to the end block"));
61 STATISTIC(NumBitMaps, "Number of switch instructions turned into bitmaps");
62 STATISTIC(NumLookupTables, "Number of switch instructions turned into lookup tables");
63 STATISTIC(NumSinkCommons, "Number of common instructions sunk down to the end block");
64 STATISTIC(NumSpeculations, "Number of speculative executed instructions");
67 /// ValueEqualityComparisonCase - Represents a case of a switch.
68 struct ValueEqualityComparisonCase {
72 ValueEqualityComparisonCase(ConstantInt *Value, BasicBlock *Dest)
73 : Value(Value), Dest(Dest) {}
75 bool operator<(ValueEqualityComparisonCase RHS) const {
76 // Comparing pointers is ok as we only rely on the order for uniquing.
77 return Value < RHS.Value;
81 class SimplifyCFGOpt {
82 const TargetData *const TD;
84 Value *isValueEqualityComparison(TerminatorInst *TI);
85 BasicBlock *GetValueEqualityComparisonCases(TerminatorInst *TI,
86 std::vector<ValueEqualityComparisonCase> &Cases);
87 bool SimplifyEqualityComparisonWithOnlyPredecessor(TerminatorInst *TI,
89 IRBuilder<> &Builder);
90 bool FoldValueComparisonIntoPredecessors(TerminatorInst *TI,
91 IRBuilder<> &Builder);
93 bool SimplifyReturn(ReturnInst *RI, IRBuilder<> &Builder);
94 bool SimplifyResume(ResumeInst *RI, IRBuilder<> &Builder);
95 bool SimplifyUnreachable(UnreachableInst *UI);
96 bool SimplifySwitch(SwitchInst *SI, IRBuilder<> &Builder);
97 bool SimplifyIndirectBr(IndirectBrInst *IBI);
98 bool SimplifyUncondBranch(BranchInst *BI, IRBuilder <> &Builder);
99 bool SimplifyCondBranch(BranchInst *BI, IRBuilder <>&Builder);
102 explicit SimplifyCFGOpt(const TargetData *td) : TD(td) {}
103 bool run(BasicBlock *BB);
107 /// SafeToMergeTerminators - Return true if it is safe to merge these two
108 /// terminator instructions together.
110 static bool SafeToMergeTerminators(TerminatorInst *SI1, TerminatorInst *SI2) {
111 if (SI1 == SI2) return false; // Can't merge with self!
113 // It is not safe to merge these two switch instructions if they have a common
114 // successor, and if that successor has a PHI node, and if *that* PHI node has
115 // conflicting incoming values from the two switch blocks.
116 BasicBlock *SI1BB = SI1->getParent();
117 BasicBlock *SI2BB = SI2->getParent();
118 SmallPtrSet<BasicBlock*, 16> SI1Succs(succ_begin(SI1BB), succ_end(SI1BB));
120 for (succ_iterator I = succ_begin(SI2BB), E = succ_end(SI2BB); I != E; ++I)
121 if (SI1Succs.count(*I))
122 for (BasicBlock::iterator BBI = (*I)->begin();
123 isa<PHINode>(BBI); ++BBI) {
124 PHINode *PN = cast<PHINode>(BBI);
125 if (PN->getIncomingValueForBlock(SI1BB) !=
126 PN->getIncomingValueForBlock(SI2BB))
133 /// isProfitableToFoldUnconditional - Return true if it is safe and profitable
134 /// to merge these two terminator instructions together, where SI1 is an
135 /// unconditional branch. PhiNodes will store all PHI nodes in common
138 static bool isProfitableToFoldUnconditional(BranchInst *SI1,
141 SmallVectorImpl<PHINode*> &PhiNodes) {
142 if (SI1 == SI2) return false; // Can't merge with self!
143 assert(SI1->isUnconditional() && SI2->isConditional());
145 // We fold the unconditional branch if we can easily update all PHI nodes in
146 // common successors:
147 // 1> We have a constant incoming value for the conditional branch;
148 // 2> We have "Cond" as the incoming value for the unconditional branch;
149 // 3> SI2->getCondition() and Cond have same operands.
150 CmpInst *Ci2 = dyn_cast<CmpInst>(SI2->getCondition());
151 if (!Ci2) return false;
152 if (!(Cond->getOperand(0) == Ci2->getOperand(0) &&
153 Cond->getOperand(1) == Ci2->getOperand(1)) &&
154 !(Cond->getOperand(0) == Ci2->getOperand(1) &&
155 Cond->getOperand(1) == Ci2->getOperand(0)))
158 BasicBlock *SI1BB = SI1->getParent();
159 BasicBlock *SI2BB = SI2->getParent();
160 SmallPtrSet<BasicBlock*, 16> SI1Succs(succ_begin(SI1BB), succ_end(SI1BB));
161 for (succ_iterator I = succ_begin(SI2BB), E = succ_end(SI2BB); I != E; ++I)
162 if (SI1Succs.count(*I))
163 for (BasicBlock::iterator BBI = (*I)->begin();
164 isa<PHINode>(BBI); ++BBI) {
165 PHINode *PN = cast<PHINode>(BBI);
166 if (PN->getIncomingValueForBlock(SI1BB) != Cond ||
167 !isa<ConstantInt>(PN->getIncomingValueForBlock(SI2BB)))
169 PhiNodes.push_back(PN);
174 /// AddPredecessorToBlock - Update PHI nodes in Succ to indicate that there will
175 /// now be entries in it from the 'NewPred' block. The values that will be
176 /// flowing into the PHI nodes will be the same as those coming in from
177 /// ExistPred, an existing predecessor of Succ.
178 static void AddPredecessorToBlock(BasicBlock *Succ, BasicBlock *NewPred,
179 BasicBlock *ExistPred) {
180 if (!isa<PHINode>(Succ->begin())) return; // Quick exit if nothing to do
183 for (BasicBlock::iterator I = Succ->begin();
184 (PN = dyn_cast<PHINode>(I)); ++I)
185 PN->addIncoming(PN->getIncomingValueForBlock(ExistPred), NewPred);
189 /// GetIfCondition - Given a basic block (BB) with two predecessors (and at
190 /// least one PHI node in it), check to see if the merge at this block is due
191 /// to an "if condition". If so, return the boolean condition that determines
192 /// which entry into BB will be taken. Also, return by references the block
193 /// that will be entered from if the condition is true, and the block that will
194 /// be entered if the condition is false.
196 /// This does no checking to see if the true/false blocks have large or unsavory
197 /// instructions in them.
198 static Value *GetIfCondition(BasicBlock *BB, BasicBlock *&IfTrue,
199 BasicBlock *&IfFalse) {
200 PHINode *SomePHI = cast<PHINode>(BB->begin());
201 assert(SomePHI->getNumIncomingValues() == 2 &&
202 "Function can only handle blocks with 2 predecessors!");
203 BasicBlock *Pred1 = SomePHI->getIncomingBlock(0);
204 BasicBlock *Pred2 = SomePHI->getIncomingBlock(1);
206 // We can only handle branches. Other control flow will be lowered to
207 // branches if possible anyway.
208 BranchInst *Pred1Br = dyn_cast<BranchInst>(Pred1->getTerminator());
209 BranchInst *Pred2Br = dyn_cast<BranchInst>(Pred2->getTerminator());
210 if (Pred1Br == 0 || Pred2Br == 0)
213 // Eliminate code duplication by ensuring that Pred1Br is conditional if
215 if (Pred2Br->isConditional()) {
216 // If both branches are conditional, we don't have an "if statement". In
217 // reality, we could transform this case, but since the condition will be
218 // required anyway, we stand no chance of eliminating it, so the xform is
219 // probably not profitable.
220 if (Pred1Br->isConditional())
223 std::swap(Pred1, Pred2);
224 std::swap(Pred1Br, Pred2Br);
227 if (Pred1Br->isConditional()) {
228 // The only thing we have to watch out for here is to make sure that Pred2
229 // doesn't have incoming edges from other blocks. If it does, the condition
230 // doesn't dominate BB.
231 if (Pred2->getSinglePredecessor() == 0)
234 // If we found a conditional branch predecessor, make sure that it branches
235 // to BB and Pred2Br. If it doesn't, this isn't an "if statement".
236 if (Pred1Br->getSuccessor(0) == BB &&
237 Pred1Br->getSuccessor(1) == Pred2) {
240 } else if (Pred1Br->getSuccessor(0) == Pred2 &&
241 Pred1Br->getSuccessor(1) == BB) {
245 // We know that one arm of the conditional goes to BB, so the other must
246 // go somewhere unrelated, and this must not be an "if statement".
250 return Pred1Br->getCondition();
253 // Ok, if we got here, both predecessors end with an unconditional branch to
254 // BB. Don't panic! If both blocks only have a single (identical)
255 // predecessor, and THAT is a conditional branch, then we're all ok!
256 BasicBlock *CommonPred = Pred1->getSinglePredecessor();
257 if (CommonPred == 0 || CommonPred != Pred2->getSinglePredecessor())
260 // Otherwise, if this is a conditional branch, then we can use it!
261 BranchInst *BI = dyn_cast<BranchInst>(CommonPred->getTerminator());
262 if (BI == 0) return 0;
264 assert(BI->isConditional() && "Two successors but not conditional?");
265 if (BI->getSuccessor(0) == Pred1) {
272 return BI->getCondition();
275 /// ComputeSpeculuationCost - Compute an abstract "cost" of speculating the
276 /// given instruction, which is assumed to be safe to speculate. 1 means
277 /// cheap, 2 means less cheap, and UINT_MAX means prohibitively expensive.
278 static unsigned ComputeSpeculationCost(const User *I) {
279 assert(isSafeToSpeculativelyExecute(I) &&
280 "Instruction is not safe to speculatively execute!");
281 switch (Operator::getOpcode(I)) {
283 // In doubt, be conservative.
285 case Instruction::GetElementPtr:
286 // GEPs are cheap if all indices are constant.
287 if (!cast<GEPOperator>(I)->hasAllConstantIndices())
290 case Instruction::Load:
291 case Instruction::Add:
292 case Instruction::Sub:
293 case Instruction::And:
294 case Instruction::Or:
295 case Instruction::Xor:
296 case Instruction::Shl:
297 case Instruction::LShr:
298 case Instruction::AShr:
299 case Instruction::ICmp:
300 case Instruction::Trunc:
301 case Instruction::ZExt:
302 case Instruction::SExt:
303 return 1; // These are all cheap.
305 case Instruction::Call:
306 case Instruction::Select:
311 /// DominatesMergePoint - If we have a merge point of an "if condition" as
312 /// accepted above, return true if the specified value dominates the block. We
313 /// don't handle the true generality of domination here, just a special case
314 /// which works well enough for us.
316 /// If AggressiveInsts is non-null, and if V does not dominate BB, we check to
317 /// see if V (which must be an instruction) and its recursive operands
318 /// that do not dominate BB have a combined cost lower than CostRemaining and
319 /// are non-trapping. If both are true, the instruction is inserted into the
320 /// set and true is returned.
322 /// The cost for most non-trapping instructions is defined as 1 except for
323 /// Select whose cost is 2.
325 /// After this function returns, CostRemaining is decreased by the cost of
326 /// V plus its non-dominating operands. If that cost is greater than
327 /// CostRemaining, false is returned and CostRemaining is undefined.
328 static bool DominatesMergePoint(Value *V, BasicBlock *BB,
329 SmallPtrSet<Instruction*, 4> *AggressiveInsts,
330 unsigned &CostRemaining) {
331 Instruction *I = dyn_cast<Instruction>(V);
333 // Non-instructions all dominate instructions, but not all constantexprs
334 // can be executed unconditionally.
335 if (ConstantExpr *C = dyn_cast<ConstantExpr>(V))
340 BasicBlock *PBB = I->getParent();
342 // We don't want to allow weird loops that might have the "if condition" in
343 // the bottom of this block.
344 if (PBB == BB) return false;
346 // If this instruction is defined in a block that contains an unconditional
347 // branch to BB, then it must be in the 'conditional' part of the "if
348 // statement". If not, it definitely dominates the region.
349 BranchInst *BI = dyn_cast<BranchInst>(PBB->getTerminator());
350 if (BI == 0 || BI->isConditional() || BI->getSuccessor(0) != BB)
353 // If we aren't allowing aggressive promotion anymore, then don't consider
354 // instructions in the 'if region'.
355 if (AggressiveInsts == 0) return false;
357 // If we have seen this instruction before, don't count it again.
358 if (AggressiveInsts->count(I)) return true;
360 // Okay, it looks like the instruction IS in the "condition". Check to
361 // see if it's a cheap instruction to unconditionally compute, and if it
362 // only uses stuff defined outside of the condition. If so, hoist it out.
363 if (!isSafeToSpeculativelyExecute(I))
366 unsigned Cost = ComputeSpeculationCost(I);
368 if (Cost > CostRemaining)
371 CostRemaining -= Cost;
373 // Okay, we can only really hoist these out if their operands do
374 // not take us over the cost threshold.
375 for (User::op_iterator i = I->op_begin(), e = I->op_end(); i != e; ++i)
376 if (!DominatesMergePoint(*i, BB, AggressiveInsts, CostRemaining))
378 // Okay, it's safe to do this! Remember this instruction.
379 AggressiveInsts->insert(I);
383 /// GetConstantInt - Extract ConstantInt from value, looking through IntToPtr
384 /// and PointerNullValue. Return NULL if value is not a constant int.
385 static ConstantInt *GetConstantInt(Value *V, const TargetData *TD) {
386 // Normal constant int.
387 ConstantInt *CI = dyn_cast<ConstantInt>(V);
388 if (CI || !TD || !isa<Constant>(V) || !V->getType()->isPointerTy())
391 // This is some kind of pointer constant. Turn it into a pointer-sized
392 // ConstantInt if possible.
393 IntegerType *PtrTy = TD->getIntPtrType(V->getContext());
395 // Null pointer means 0, see SelectionDAGBuilder::getValue(const Value*).
396 if (isa<ConstantPointerNull>(V))
397 return ConstantInt::get(PtrTy, 0);
399 // IntToPtr const int.
400 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V))
401 if (CE->getOpcode() == Instruction::IntToPtr)
402 if (ConstantInt *CI = dyn_cast<ConstantInt>(CE->getOperand(0))) {
403 // The constant is very likely to have the right type already.
404 if (CI->getType() == PtrTy)
407 return cast<ConstantInt>
408 (ConstantExpr::getIntegerCast(CI, PtrTy, /*isSigned=*/false));
413 /// GatherConstantCompares - Given a potentially 'or'd or 'and'd together
414 /// collection of icmp eq/ne instructions that compare a value against a
415 /// constant, return the value being compared, and stick the constant into the
418 GatherConstantCompares(Value *V, std::vector<ConstantInt*> &Vals, Value *&Extra,
419 const TargetData *TD, bool isEQ, unsigned &UsedICmps) {
420 Instruction *I = dyn_cast<Instruction>(V);
421 if (I == 0) return 0;
423 // If this is an icmp against a constant, handle this as one of the cases.
424 if (ICmpInst *ICI = dyn_cast<ICmpInst>(I)) {
425 if (ConstantInt *C = GetConstantInt(I->getOperand(1), TD)) {
426 if (ICI->getPredicate() == (isEQ ? ICmpInst::ICMP_EQ:ICmpInst::ICMP_NE)) {
429 return I->getOperand(0);
432 // If we have "x ult 3" comparison, for example, then we can add 0,1,2 to
435 ConstantRange::makeICmpRegion(ICI->getPredicate(), C->getValue());
437 // If this is an and/!= check then we want to optimize "x ugt 2" into
440 Span = Span.inverse();
442 // If there are a ton of values, we don't want to make a ginormous switch.
443 if (Span.getSetSize().ugt(8) || Span.isEmptySet())
446 for (APInt Tmp = Span.getLower(); Tmp != Span.getUpper(); ++Tmp)
447 Vals.push_back(ConstantInt::get(V->getContext(), Tmp));
449 return I->getOperand(0);
454 // Otherwise, we can only handle an | or &, depending on isEQ.
455 if (I->getOpcode() != (isEQ ? Instruction::Or : Instruction::And))
458 unsigned NumValsBeforeLHS = Vals.size();
459 unsigned UsedICmpsBeforeLHS = UsedICmps;
460 if (Value *LHS = GatherConstantCompares(I->getOperand(0), Vals, Extra, TD,
462 unsigned NumVals = Vals.size();
463 unsigned UsedICmpsBeforeRHS = UsedICmps;
464 if (Value *RHS = GatherConstantCompares(I->getOperand(1), Vals, Extra, TD,
468 Vals.resize(NumVals);
469 UsedICmps = UsedICmpsBeforeRHS;
472 // The RHS of the or/and can't be folded in and we haven't used "Extra" yet,
473 // set it and return success.
474 if (Extra == 0 || Extra == I->getOperand(1)) {
475 Extra = I->getOperand(1);
479 Vals.resize(NumValsBeforeLHS);
480 UsedICmps = UsedICmpsBeforeLHS;
484 // If the LHS can't be folded in, but Extra is available and RHS can, try to
486 if (Extra == 0 || Extra == I->getOperand(0)) {
487 Value *OldExtra = Extra;
488 Extra = I->getOperand(0);
489 if (Value *RHS = GatherConstantCompares(I->getOperand(1), Vals, Extra, TD,
492 assert(Vals.size() == NumValsBeforeLHS);
499 static void EraseTerminatorInstAndDCECond(TerminatorInst *TI) {
500 Instruction *Cond = 0;
501 if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) {
502 Cond = dyn_cast<Instruction>(SI->getCondition());
503 } else if (BranchInst *BI = dyn_cast<BranchInst>(TI)) {
504 if (BI->isConditional())
505 Cond = dyn_cast<Instruction>(BI->getCondition());
506 } else if (IndirectBrInst *IBI = dyn_cast<IndirectBrInst>(TI)) {
507 Cond = dyn_cast<Instruction>(IBI->getAddress());
510 TI->eraseFromParent();
511 if (Cond) RecursivelyDeleteTriviallyDeadInstructions(Cond);
514 /// isValueEqualityComparison - Return true if the specified terminator checks
515 /// to see if a value is equal to constant integer value.
516 Value *SimplifyCFGOpt::isValueEqualityComparison(TerminatorInst *TI) {
518 if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) {
519 // Do not permit merging of large switch instructions into their
520 // predecessors unless there is only one predecessor.
521 if (SI->getNumSuccessors()*std::distance(pred_begin(SI->getParent()),
522 pred_end(SI->getParent())) <= 128)
523 CV = SI->getCondition();
524 } else if (BranchInst *BI = dyn_cast<BranchInst>(TI))
525 if (BI->isConditional() && BI->getCondition()->hasOneUse())
526 if (ICmpInst *ICI = dyn_cast<ICmpInst>(BI->getCondition()))
527 if ((ICI->getPredicate() == ICmpInst::ICMP_EQ ||
528 ICI->getPredicate() == ICmpInst::ICMP_NE) &&
529 GetConstantInt(ICI->getOperand(1), TD))
530 CV = ICI->getOperand(0);
532 // Unwrap any lossless ptrtoint cast.
533 if (TD && CV && CV->getType() == TD->getIntPtrType(CV->getContext()))
534 if (PtrToIntInst *PTII = dyn_cast<PtrToIntInst>(CV))
535 CV = PTII->getOperand(0);
539 /// GetValueEqualityComparisonCases - Given a value comparison instruction,
540 /// decode all of the 'cases' that it represents and return the 'default' block.
541 BasicBlock *SimplifyCFGOpt::
542 GetValueEqualityComparisonCases(TerminatorInst *TI,
543 std::vector<ValueEqualityComparisonCase>
545 if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) {
546 Cases.reserve(SI->getNumCases());
547 for (SwitchInst::CaseIt i = SI->case_begin(), e = SI->case_end(); i != e; ++i)
548 Cases.push_back(ValueEqualityComparisonCase(i.getCaseValue(),
549 i.getCaseSuccessor()));
550 return SI->getDefaultDest();
553 BranchInst *BI = cast<BranchInst>(TI);
554 ICmpInst *ICI = cast<ICmpInst>(BI->getCondition());
555 BasicBlock *Succ = BI->getSuccessor(ICI->getPredicate() == ICmpInst::ICMP_NE);
556 Cases.push_back(ValueEqualityComparisonCase(GetConstantInt(ICI->getOperand(1),
559 return BI->getSuccessor(ICI->getPredicate() == ICmpInst::ICMP_EQ);
563 /// EliminateBlockCases - Given a vector of bb/value pairs, remove any entries
564 /// in the list that match the specified block.
565 static void EliminateBlockCases(BasicBlock *BB,
566 std::vector<ValueEqualityComparisonCase> &Cases) {
567 for (unsigned i = 0, e = Cases.size(); i != e; ++i)
568 if (Cases[i].Dest == BB) {
569 Cases.erase(Cases.begin()+i);
574 /// ValuesOverlap - Return true if there are any keys in C1 that exist in C2 as
577 ValuesOverlap(std::vector<ValueEqualityComparisonCase> &C1,
578 std::vector<ValueEqualityComparisonCase > &C2) {
579 std::vector<ValueEqualityComparisonCase> *V1 = &C1, *V2 = &C2;
581 // Make V1 be smaller than V2.
582 if (V1->size() > V2->size())
585 if (V1->size() == 0) return false;
586 if (V1->size() == 1) {
588 ConstantInt *TheVal = (*V1)[0].Value;
589 for (unsigned i = 0, e = V2->size(); i != e; ++i)
590 if (TheVal == (*V2)[i].Value)
594 // Otherwise, just sort both lists and compare element by element.
595 array_pod_sort(V1->begin(), V1->end());
596 array_pod_sort(V2->begin(), V2->end());
597 unsigned i1 = 0, i2 = 0, e1 = V1->size(), e2 = V2->size();
598 while (i1 != e1 && i2 != e2) {
599 if ((*V1)[i1].Value == (*V2)[i2].Value)
601 if ((*V1)[i1].Value < (*V2)[i2].Value)
609 /// SimplifyEqualityComparisonWithOnlyPredecessor - If TI is known to be a
610 /// terminator instruction and its block is known to only have a single
611 /// predecessor block, check to see if that predecessor is also a value
612 /// comparison with the same value, and if that comparison determines the
613 /// outcome of this comparison. If so, simplify TI. This does a very limited
614 /// form of jump threading.
615 bool SimplifyCFGOpt::
616 SimplifyEqualityComparisonWithOnlyPredecessor(TerminatorInst *TI,
618 IRBuilder<> &Builder) {
619 Value *PredVal = isValueEqualityComparison(Pred->getTerminator());
620 if (!PredVal) return false; // Not a value comparison in predecessor.
622 Value *ThisVal = isValueEqualityComparison(TI);
623 assert(ThisVal && "This isn't a value comparison!!");
624 if (ThisVal != PredVal) return false; // Different predicates.
626 // TODO: Preserve branch weight metadata, similarly to how
627 // FoldValueComparisonIntoPredecessors preserves it.
629 // Find out information about when control will move from Pred to TI's block.
630 std::vector<ValueEqualityComparisonCase> PredCases;
631 BasicBlock *PredDef = GetValueEqualityComparisonCases(Pred->getTerminator(),
633 EliminateBlockCases(PredDef, PredCases); // Remove default from cases.
635 // Find information about how control leaves this block.
636 std::vector<ValueEqualityComparisonCase> ThisCases;
637 BasicBlock *ThisDef = GetValueEqualityComparisonCases(TI, ThisCases);
638 EliminateBlockCases(ThisDef, ThisCases); // Remove default from cases.
640 // If TI's block is the default block from Pred's comparison, potentially
641 // simplify TI based on this knowledge.
642 if (PredDef == TI->getParent()) {
643 // If we are here, we know that the value is none of those cases listed in
644 // PredCases. If there are any cases in ThisCases that are in PredCases, we
646 if (!ValuesOverlap(PredCases, ThisCases))
649 if (isa<BranchInst>(TI)) {
650 // Okay, one of the successors of this condbr is dead. Convert it to a
652 assert(ThisCases.size() == 1 && "Branch can only have one case!");
653 // Insert the new branch.
654 Instruction *NI = Builder.CreateBr(ThisDef);
657 // Remove PHI node entries for the dead edge.
658 ThisCases[0].Dest->removePredecessor(TI->getParent());
660 DEBUG(dbgs() << "Threading pred instr: " << *Pred->getTerminator()
661 << "Through successor TI: " << *TI << "Leaving: " << *NI << "\n");
663 EraseTerminatorInstAndDCECond(TI);
667 SwitchInst *SI = cast<SwitchInst>(TI);
668 // Okay, TI has cases that are statically dead, prune them away.
669 SmallPtrSet<Constant*, 16> DeadCases;
670 for (unsigned i = 0, e = PredCases.size(); i != e; ++i)
671 DeadCases.insert(PredCases[i].Value);
673 DEBUG(dbgs() << "Threading pred instr: " << *Pred->getTerminator()
674 << "Through successor TI: " << *TI);
676 // Collect branch weights into a vector.
677 SmallVector<uint32_t, 8> Weights;
678 MDNode* MD = SI->getMetadata(LLVMContext::MD_prof);
679 bool HasWeight = MD && (MD->getNumOperands() == 2 + SI->getNumCases());
681 for (unsigned MD_i = 1, MD_e = MD->getNumOperands(); MD_i < MD_e;
683 ConstantInt* CI = dyn_cast<ConstantInt>(MD->getOperand(MD_i));
685 Weights.push_back(CI->getValue().getZExtValue());
687 for (SwitchInst::CaseIt i = SI->case_end(), e = SI->case_begin(); i != e;) {
689 if (DeadCases.count(i.getCaseValue())) {
691 std::swap(Weights[i.getCaseIndex()+1], Weights.back());
694 i.getCaseSuccessor()->removePredecessor(TI->getParent());
699 SI->setMetadata(LLVMContext::MD_prof,
700 MDBuilder(SI->getParent()->getContext()).
701 createBranchWeights(Weights));
703 DEBUG(dbgs() << "Leaving: " << *TI << "\n");
707 // Otherwise, TI's block must correspond to some matched value. Find out
708 // which value (or set of values) this is.
709 ConstantInt *TIV = 0;
710 BasicBlock *TIBB = TI->getParent();
711 for (unsigned i = 0, e = PredCases.size(); i != e; ++i)
712 if (PredCases[i].Dest == TIBB) {
714 return false; // Cannot handle multiple values coming to this block.
715 TIV = PredCases[i].Value;
717 assert(TIV && "No edge from pred to succ?");
719 // Okay, we found the one constant that our value can be if we get into TI's
720 // BB. Find out which successor will unconditionally be branched to.
721 BasicBlock *TheRealDest = 0;
722 for (unsigned i = 0, e = ThisCases.size(); i != e; ++i)
723 if (ThisCases[i].Value == TIV) {
724 TheRealDest = ThisCases[i].Dest;
728 // If not handled by any explicit cases, it is handled by the default case.
729 if (TheRealDest == 0) TheRealDest = ThisDef;
731 // Remove PHI node entries for dead edges.
732 BasicBlock *CheckEdge = TheRealDest;
733 for (succ_iterator SI = succ_begin(TIBB), e = succ_end(TIBB); SI != e; ++SI)
734 if (*SI != CheckEdge)
735 (*SI)->removePredecessor(TIBB);
739 // Insert the new branch.
740 Instruction *NI = Builder.CreateBr(TheRealDest);
743 DEBUG(dbgs() << "Threading pred instr: " << *Pred->getTerminator()
744 << "Through successor TI: " << *TI << "Leaving: " << *NI << "\n");
746 EraseTerminatorInstAndDCECond(TI);
751 /// ConstantIntOrdering - This class implements a stable ordering of constant
752 /// integers that does not depend on their address. This is important for
753 /// applications that sort ConstantInt's to ensure uniqueness.
754 struct ConstantIntOrdering {
755 bool operator()(const ConstantInt *LHS, const ConstantInt *RHS) const {
756 return LHS->getValue().ult(RHS->getValue());
761 static int ConstantIntSortPredicate(const void *P1, const void *P2) {
762 const ConstantInt *LHS = *(const ConstantInt*const*)P1;
763 const ConstantInt *RHS = *(const ConstantInt*const*)P2;
764 if (LHS->getValue().ult(RHS->getValue()))
766 if (LHS->getValue() == RHS->getValue())
771 static inline bool HasBranchWeights(const Instruction* I) {
772 MDNode* ProfMD = I->getMetadata(LLVMContext::MD_prof);
773 if (ProfMD && ProfMD->getOperand(0))
774 if (MDString* MDS = dyn_cast<MDString>(ProfMD->getOperand(0)))
775 return MDS->getString().equals("branch_weights");
780 /// Get Weights of a given TerminatorInst, the default weight is at the front
781 /// of the vector. If TI is a conditional eq, we need to swap the branch-weight
783 static void GetBranchWeights(TerminatorInst *TI,
784 SmallVectorImpl<uint64_t> &Weights) {
785 MDNode* MD = TI->getMetadata(LLVMContext::MD_prof);
787 for (unsigned i = 1, e = MD->getNumOperands(); i < e; ++i) {
788 ConstantInt* CI = dyn_cast<ConstantInt>(MD->getOperand(i));
790 Weights.push_back(CI->getValue().getZExtValue());
793 // If TI is a conditional eq, the default case is the false case,
794 // and the corresponding branch-weight data is at index 2. We swap the
795 // default weight to be the first entry.
796 if (BranchInst* BI = dyn_cast<BranchInst>(TI)) {
797 assert(Weights.size() == 2);
798 ICmpInst *ICI = cast<ICmpInst>(BI->getCondition());
799 if (ICI->getPredicate() == ICmpInst::ICMP_EQ)
800 std::swap(Weights.front(), Weights.back());
804 /// Sees if any of the weights are too big for a uint32_t, and halves all the
805 /// weights if any are.
806 static void FitWeights(MutableArrayRef<uint64_t> Weights) {
808 for (unsigned i = 0; i < Weights.size(); ++i)
809 if (Weights[i] > UINT_MAX) {
817 for (unsigned i = 0; i < Weights.size(); ++i)
821 /// FoldValueComparisonIntoPredecessors - The specified terminator is a value
822 /// equality comparison instruction (either a switch or a branch on "X == c").
823 /// See if any of the predecessors of the terminator block are value comparisons
824 /// on the same value. If so, and if safe to do so, fold them together.
825 bool SimplifyCFGOpt::FoldValueComparisonIntoPredecessors(TerminatorInst *TI,
826 IRBuilder<> &Builder) {
827 BasicBlock *BB = TI->getParent();
828 Value *CV = isValueEqualityComparison(TI); // CondVal
829 assert(CV && "Not a comparison?");
830 bool Changed = false;
832 SmallVector<BasicBlock*, 16> Preds(pred_begin(BB), pred_end(BB));
833 while (!Preds.empty()) {
834 BasicBlock *Pred = Preds.pop_back_val();
836 // See if the predecessor is a comparison with the same value.
837 TerminatorInst *PTI = Pred->getTerminator();
838 Value *PCV = isValueEqualityComparison(PTI); // PredCondVal
840 if (PCV == CV && SafeToMergeTerminators(TI, PTI)) {
841 // Figure out which 'cases' to copy from SI to PSI.
842 std::vector<ValueEqualityComparisonCase> BBCases;
843 BasicBlock *BBDefault = GetValueEqualityComparisonCases(TI, BBCases);
845 std::vector<ValueEqualityComparisonCase> PredCases;
846 BasicBlock *PredDefault = GetValueEqualityComparisonCases(PTI, PredCases);
848 // Based on whether the default edge from PTI goes to BB or not, fill in
849 // PredCases and PredDefault with the new switch cases we would like to
851 SmallVector<BasicBlock*, 8> NewSuccessors;
853 // Update the branch weight metadata along the way
854 SmallVector<uint64_t, 8> Weights;
855 bool PredHasWeights = HasBranchWeights(PTI);
856 bool SuccHasWeights = HasBranchWeights(TI);
858 if (PredHasWeights) {
859 GetBranchWeights(PTI, Weights);
860 // branch-weight metadata is inconsistant here.
861 if (Weights.size() != 1 + PredCases.size())
862 PredHasWeights = SuccHasWeights = false;
863 } else if (SuccHasWeights)
864 // If there are no predecessor weights but there are successor weights,
865 // populate Weights with 1, which will later be scaled to the sum of
866 // successor's weights
867 Weights.assign(1 + PredCases.size(), 1);
869 SmallVector<uint64_t, 8> SuccWeights;
870 if (SuccHasWeights) {
871 GetBranchWeights(TI, SuccWeights);
872 // branch-weight metadata is inconsistant here.
873 if (SuccWeights.size() != 1 + BBCases.size())
874 PredHasWeights = SuccHasWeights = false;
875 } else if (PredHasWeights)
876 SuccWeights.assign(1 + BBCases.size(), 1);
878 if (PredDefault == BB) {
879 // If this is the default destination from PTI, only the edges in TI
880 // that don't occur in PTI, or that branch to BB will be activated.
881 std::set<ConstantInt*, ConstantIntOrdering> PTIHandled;
882 for (unsigned i = 0, e = PredCases.size(); i != e; ++i)
883 if (PredCases[i].Dest != BB)
884 PTIHandled.insert(PredCases[i].Value);
886 // The default destination is BB, we don't need explicit targets.
887 std::swap(PredCases[i], PredCases.back());
889 if (PredHasWeights || SuccHasWeights) {
890 // Increase weight for the default case.
891 Weights[0] += Weights[i+1];
892 std::swap(Weights[i+1], Weights.back());
896 PredCases.pop_back();
900 // Reconstruct the new switch statement we will be building.
901 if (PredDefault != BBDefault) {
902 PredDefault->removePredecessor(Pred);
903 PredDefault = BBDefault;
904 NewSuccessors.push_back(BBDefault);
907 unsigned CasesFromPred = Weights.size();
908 uint64_t ValidTotalSuccWeight = 0;
909 for (unsigned i = 0, e = BBCases.size(); i != e; ++i)
910 if (!PTIHandled.count(BBCases[i].Value) &&
911 BBCases[i].Dest != BBDefault) {
912 PredCases.push_back(BBCases[i]);
913 NewSuccessors.push_back(BBCases[i].Dest);
914 if (SuccHasWeights || PredHasWeights) {
915 // The default weight is at index 0, so weight for the ith case
916 // should be at index i+1. Scale the cases from successor by
917 // PredDefaultWeight (Weights[0]).
918 Weights.push_back(Weights[0] * SuccWeights[i+1]);
919 ValidTotalSuccWeight += SuccWeights[i+1];
923 if (SuccHasWeights || PredHasWeights) {
924 ValidTotalSuccWeight += SuccWeights[0];
925 // Scale the cases from predecessor by ValidTotalSuccWeight.
926 for (unsigned i = 1; i < CasesFromPred; ++i)
927 Weights[i] *= ValidTotalSuccWeight;
928 // Scale the default weight by SuccDefaultWeight (SuccWeights[0]).
929 Weights[0] *= SuccWeights[0];
932 // If this is not the default destination from PSI, only the edges
933 // in SI that occur in PSI with a destination of BB will be
935 std::set<ConstantInt*, ConstantIntOrdering> PTIHandled;
936 std::map<ConstantInt*, uint64_t> WeightsForHandled;
937 for (unsigned i = 0, e = PredCases.size(); i != e; ++i)
938 if (PredCases[i].Dest == BB) {
939 PTIHandled.insert(PredCases[i].Value);
941 if (PredHasWeights || SuccHasWeights) {
942 WeightsForHandled[PredCases[i].Value] = Weights[i+1];
943 std::swap(Weights[i+1], Weights.back());
947 std::swap(PredCases[i], PredCases.back());
948 PredCases.pop_back();
952 // Okay, now we know which constants were sent to BB from the
953 // predecessor. Figure out where they will all go now.
954 for (unsigned i = 0, e = BBCases.size(); i != e; ++i)
955 if (PTIHandled.count(BBCases[i].Value)) {
956 // If this is one we are capable of getting...
957 if (PredHasWeights || SuccHasWeights)
958 Weights.push_back(WeightsForHandled[BBCases[i].Value]);
959 PredCases.push_back(BBCases[i]);
960 NewSuccessors.push_back(BBCases[i].Dest);
961 PTIHandled.erase(BBCases[i].Value);// This constant is taken care of
964 // If there are any constants vectored to BB that TI doesn't handle,
965 // they must go to the default destination of TI.
966 for (std::set<ConstantInt*, ConstantIntOrdering>::iterator I =
968 E = PTIHandled.end(); I != E; ++I) {
969 if (PredHasWeights || SuccHasWeights)
970 Weights.push_back(WeightsForHandled[*I]);
971 PredCases.push_back(ValueEqualityComparisonCase(*I, BBDefault));
972 NewSuccessors.push_back(BBDefault);
976 // Okay, at this point, we know which new successor Pred will get. Make
977 // sure we update the number of entries in the PHI nodes for these
979 for (unsigned i = 0, e = NewSuccessors.size(); i != e; ++i)
980 AddPredecessorToBlock(NewSuccessors[i], Pred, BB);
982 Builder.SetInsertPoint(PTI);
983 // Convert pointer to int before we switch.
984 if (CV->getType()->isPointerTy()) {
985 assert(TD && "Cannot switch on pointer without TargetData");
986 CV = Builder.CreatePtrToInt(CV, TD->getIntPtrType(CV->getContext()),
990 // Now that the successors are updated, create the new Switch instruction.
991 SwitchInst *NewSI = Builder.CreateSwitch(CV, PredDefault,
993 NewSI->setDebugLoc(PTI->getDebugLoc());
994 for (unsigned i = 0, e = PredCases.size(); i != e; ++i)
995 NewSI->addCase(PredCases[i].Value, PredCases[i].Dest);
997 if (PredHasWeights || SuccHasWeights) {
998 // Halve the weights if any of them cannot fit in an uint32_t
1001 SmallVector<uint32_t, 8> MDWeights(Weights.begin(), Weights.end());
1003 NewSI->setMetadata(LLVMContext::MD_prof,
1004 MDBuilder(BB->getContext()).
1005 createBranchWeights(MDWeights));
1008 EraseTerminatorInstAndDCECond(PTI);
1010 // Okay, last check. If BB is still a successor of PSI, then we must
1011 // have an infinite loop case. If so, add an infinitely looping block
1012 // to handle the case to preserve the behavior of the code.
1013 BasicBlock *InfLoopBlock = 0;
1014 for (unsigned i = 0, e = NewSI->getNumSuccessors(); i != e; ++i)
1015 if (NewSI->getSuccessor(i) == BB) {
1016 if (InfLoopBlock == 0) {
1017 // Insert it at the end of the function, because it's either code,
1018 // or it won't matter if it's hot. :)
1019 InfLoopBlock = BasicBlock::Create(BB->getContext(),
1020 "infloop", BB->getParent());
1021 BranchInst::Create(InfLoopBlock, InfLoopBlock);
1023 NewSI->setSuccessor(i, InfLoopBlock);
1032 // isSafeToHoistInvoke - If we would need to insert a select that uses the
1033 // value of this invoke (comments in HoistThenElseCodeToIf explain why we
1034 // would need to do this), we can't hoist the invoke, as there is nowhere
1035 // to put the select in this case.
1036 static bool isSafeToHoistInvoke(BasicBlock *BB1, BasicBlock *BB2,
1037 Instruction *I1, Instruction *I2) {
1038 for (succ_iterator SI = succ_begin(BB1), E = succ_end(BB1); SI != E; ++SI) {
1040 for (BasicBlock::iterator BBI = SI->begin();
1041 (PN = dyn_cast<PHINode>(BBI)); ++BBI) {
1042 Value *BB1V = PN->getIncomingValueForBlock(BB1);
1043 Value *BB2V = PN->getIncomingValueForBlock(BB2);
1044 if (BB1V != BB2V && (BB1V==I1 || BB2V==I2)) {
1052 /// HoistThenElseCodeToIf - Given a conditional branch that goes to BB1 and
1053 /// BB2, hoist any common code in the two blocks up into the branch block. The
1054 /// caller of this function guarantees that BI's block dominates BB1 and BB2.
1055 static bool HoistThenElseCodeToIf(BranchInst *BI) {
1056 // This does very trivial matching, with limited scanning, to find identical
1057 // instructions in the two blocks. In particular, we don't want to get into
1058 // O(M*N) situations here where M and N are the sizes of BB1 and BB2. As
1059 // such, we currently just scan for obviously identical instructions in an
1061 BasicBlock *BB1 = BI->getSuccessor(0); // The true destination.
1062 BasicBlock *BB2 = BI->getSuccessor(1); // The false destination
1064 BasicBlock::iterator BB1_Itr = BB1->begin();
1065 BasicBlock::iterator BB2_Itr = BB2->begin();
1067 Instruction *I1 = BB1_Itr++, *I2 = BB2_Itr++;
1068 // Skip debug info if it is not identical.
1069 DbgInfoIntrinsic *DBI1 = dyn_cast<DbgInfoIntrinsic>(I1);
1070 DbgInfoIntrinsic *DBI2 = dyn_cast<DbgInfoIntrinsic>(I2);
1071 if (!DBI1 || !DBI2 || !DBI1->isIdenticalToWhenDefined(DBI2)) {
1072 while (isa<DbgInfoIntrinsic>(I1))
1074 while (isa<DbgInfoIntrinsic>(I2))
1077 if (isa<PHINode>(I1) || !I1->isIdenticalToWhenDefined(I2) ||
1078 (isa<InvokeInst>(I1) && !isSafeToHoistInvoke(BB1, BB2, I1, I2)))
1081 // If we get here, we can hoist at least one instruction.
1082 BasicBlock *BIParent = BI->getParent();
1085 // If we are hoisting the terminator instruction, don't move one (making a
1086 // broken BB), instead clone it, and remove BI.
1087 if (isa<TerminatorInst>(I1))
1088 goto HoistTerminator;
1090 // For a normal instruction, we just move one to right before the branch,
1091 // then replace all uses of the other with the first. Finally, we remove
1092 // the now redundant second instruction.
1093 BIParent->getInstList().splice(BI, BB1->getInstList(), I1);
1094 if (!I2->use_empty())
1095 I2->replaceAllUsesWith(I1);
1096 I1->intersectOptionalDataWith(I2);
1097 I2->eraseFromParent();
1101 // Skip debug info if it is not identical.
1102 DbgInfoIntrinsic *DBI1 = dyn_cast<DbgInfoIntrinsic>(I1);
1103 DbgInfoIntrinsic *DBI2 = dyn_cast<DbgInfoIntrinsic>(I2);
1104 if (!DBI1 || !DBI2 || !DBI1->isIdenticalToWhenDefined(DBI2)) {
1105 while (isa<DbgInfoIntrinsic>(I1))
1107 while (isa<DbgInfoIntrinsic>(I2))
1110 } while (I1->isIdenticalToWhenDefined(I2));
1115 // It may not be possible to hoist an invoke.
1116 if (isa<InvokeInst>(I1) && !isSafeToHoistInvoke(BB1, BB2, I1, I2))
1119 // Okay, it is safe to hoist the terminator.
1120 Instruction *NT = I1->clone();
1121 BIParent->getInstList().insert(BI, NT);
1122 if (!NT->getType()->isVoidTy()) {
1123 I1->replaceAllUsesWith(NT);
1124 I2->replaceAllUsesWith(NT);
1128 IRBuilder<true, NoFolder> Builder(NT);
1129 // Hoisting one of the terminators from our successor is a great thing.
1130 // Unfortunately, the successors of the if/else blocks may have PHI nodes in
1131 // them. If they do, all PHI entries for BB1/BB2 must agree for all PHI
1132 // nodes, so we insert select instruction to compute the final result.
1133 std::map<std::pair<Value*,Value*>, SelectInst*> InsertedSelects;
1134 for (succ_iterator SI = succ_begin(BB1), E = succ_end(BB1); SI != E; ++SI) {
1136 for (BasicBlock::iterator BBI = SI->begin();
1137 (PN = dyn_cast<PHINode>(BBI)); ++BBI) {
1138 Value *BB1V = PN->getIncomingValueForBlock(BB1);
1139 Value *BB2V = PN->getIncomingValueForBlock(BB2);
1140 if (BB1V == BB2V) continue;
1142 // These values do not agree. Insert a select instruction before NT
1143 // that determines the right value.
1144 SelectInst *&SI = InsertedSelects[std::make_pair(BB1V, BB2V)];
1146 SI = cast<SelectInst>
1147 (Builder.CreateSelect(BI->getCondition(), BB1V, BB2V,
1148 BB1V->getName()+"."+BB2V->getName()));
1150 // Make the PHI node use the select for all incoming values for BB1/BB2
1151 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
1152 if (PN->getIncomingBlock(i) == BB1 || PN->getIncomingBlock(i) == BB2)
1153 PN->setIncomingValue(i, SI);
1157 // Update any PHI nodes in our new successors.
1158 for (succ_iterator SI = succ_begin(BB1), E = succ_end(BB1); SI != E; ++SI)
1159 AddPredecessorToBlock(*SI, BIParent, BB1);
1161 EraseTerminatorInstAndDCECond(BI);
1165 /// SinkThenElseCodeToEnd - Given an unconditional branch that goes to BBEnd,
1166 /// check whether BBEnd has only two predecessors and the other predecessor
1167 /// ends with an unconditional branch. If it is true, sink any common code
1168 /// in the two predecessors to BBEnd.
1169 static bool SinkThenElseCodeToEnd(BranchInst *BI1) {
1170 assert(BI1->isUnconditional());
1171 BasicBlock *BB1 = BI1->getParent();
1172 BasicBlock *BBEnd = BI1->getSuccessor(0);
1174 // Check that BBEnd has two predecessors and the other predecessor ends with
1175 // an unconditional branch.
1176 SmallVector<BasicBlock*, 16> Preds(pred_begin(BBEnd), pred_end(BBEnd));
1177 if (Preds.size() != 2)
1179 BasicBlock *BB2 = (Preds[0] == BB1) ? Preds[1] : Preds[0];
1180 BranchInst *BI2 = dyn_cast<BranchInst>(BB2->getTerminator());
1181 if (!BI2 || !BI2->isUnconditional())
1184 // Gather the PHI nodes in BBEnd.
1185 std::map<Value*, std::pair<Value*, PHINode*> > MapValueFromBB1ToBB2;
1186 Instruction *FirstNonPhiInBBEnd = 0;
1187 for (BasicBlock::iterator I = BBEnd->begin(), E = BBEnd->end();
1189 if (PHINode *PN = dyn_cast<PHINode>(I)) {
1190 Value *BB1V = PN->getIncomingValueForBlock(BB1);
1191 Value *BB2V = PN->getIncomingValueForBlock(BB2);
1192 MapValueFromBB1ToBB2[BB1V] = std::make_pair(BB2V, PN);
1194 FirstNonPhiInBBEnd = &*I;
1198 if (!FirstNonPhiInBBEnd)
1202 // This does very trivial matching, with limited scanning, to find identical
1203 // instructions in the two blocks. We scan backward for obviously identical
1204 // instructions in an identical order.
1205 BasicBlock::InstListType::reverse_iterator RI1 = BB1->getInstList().rbegin(),
1206 RE1 = BB1->getInstList().rend(), RI2 = BB2->getInstList().rbegin(),
1207 RE2 = BB2->getInstList().rend();
1209 while (RI1 != RE1 && isa<DbgInfoIntrinsic>(&*RI1)) ++RI1;
1212 while (RI2 != RE2 && isa<DbgInfoIntrinsic>(&*RI2)) ++RI2;
1215 // Skip the unconditional branches.
1219 bool Changed = false;
1220 while (RI1 != RE1 && RI2 != RE2) {
1222 while (RI1 != RE1 && isa<DbgInfoIntrinsic>(&*RI1)) ++RI1;
1225 while (RI2 != RE2 && isa<DbgInfoIntrinsic>(&*RI2)) ++RI2;
1229 Instruction *I1 = &*RI1, *I2 = &*RI2;
1230 // I1 and I2 should have a single use in the same PHI node, and they
1231 // perform the same operation.
1232 // Cannot move control-flow-involving, volatile loads, vaarg, etc.
1233 if (isa<PHINode>(I1) || isa<PHINode>(I2) ||
1234 isa<TerminatorInst>(I1) || isa<TerminatorInst>(I2) ||
1235 isa<LandingPadInst>(I1) || isa<LandingPadInst>(I2) ||
1236 isa<AllocaInst>(I1) || isa<AllocaInst>(I2) ||
1237 I1->mayHaveSideEffects() || I2->mayHaveSideEffects() ||
1238 I1->mayReadOrWriteMemory() || I2->mayReadOrWriteMemory() ||
1239 !I1->hasOneUse() || !I2->hasOneUse() ||
1240 MapValueFromBB1ToBB2.find(I1) == MapValueFromBB1ToBB2.end() ||
1241 MapValueFromBB1ToBB2[I1].first != I2)
1244 // Check whether we should swap the operands of ICmpInst.
1245 ICmpInst *ICmp1 = dyn_cast<ICmpInst>(I1), *ICmp2 = dyn_cast<ICmpInst>(I2);
1246 bool SwapOpnds = false;
1247 if (ICmp1 && ICmp2 &&
1248 ICmp1->getOperand(0) != ICmp2->getOperand(0) &&
1249 ICmp1->getOperand(1) != ICmp2->getOperand(1) &&
1250 (ICmp1->getOperand(0) == ICmp2->getOperand(1) ||
1251 ICmp1->getOperand(1) == ICmp2->getOperand(0))) {
1252 ICmp2->swapOperands();
1255 if (!I1->isSameOperationAs(I2)) {
1257 ICmp2->swapOperands();
1261 // The operands should be either the same or they need to be generated
1262 // with a PHI node after sinking. We only handle the case where there is
1263 // a single pair of different operands.
1264 Value *DifferentOp1 = 0, *DifferentOp2 = 0;
1265 unsigned Op1Idx = 0;
1266 for (unsigned I = 0, E = I1->getNumOperands(); I != E; ++I) {
1267 if (I1->getOperand(I) == I2->getOperand(I))
1269 // Early exit if we have more-than one pair of different operands or
1270 // the different operand is already in MapValueFromBB1ToBB2.
1271 // Early exit if we need a PHI node to replace a constant.
1273 MapValueFromBB1ToBB2.find(I1->getOperand(I)) !=
1274 MapValueFromBB1ToBB2.end() ||
1275 isa<Constant>(I1->getOperand(I)) ||
1276 isa<Constant>(I2->getOperand(I))) {
1277 // If we can't sink the instructions, undo the swapping.
1279 ICmp2->swapOperands();
1282 DifferentOp1 = I1->getOperand(I);
1284 DifferentOp2 = I2->getOperand(I);
1287 // We insert the pair of different operands to MapValueFromBB1ToBB2 and
1288 // remove (I1, I2) from MapValueFromBB1ToBB2.
1290 PHINode *NewPN = PHINode::Create(DifferentOp1->getType(), 2,
1291 DifferentOp1->getName() + ".sink",
1293 MapValueFromBB1ToBB2[DifferentOp1] = std::make_pair(DifferentOp2, NewPN);
1294 // I1 should use NewPN instead of DifferentOp1.
1295 I1->setOperand(Op1Idx, NewPN);
1296 NewPN->addIncoming(DifferentOp1, BB1);
1297 NewPN->addIncoming(DifferentOp2, BB2);
1298 DEBUG(dbgs() << "Create PHI node " << *NewPN << "\n";);
1300 PHINode *OldPN = MapValueFromBB1ToBB2[I1].second;
1301 MapValueFromBB1ToBB2.erase(I1);
1303 DEBUG(dbgs() << "SINK common instructions " << *I1 << "\n";);
1304 DEBUG(dbgs() << " " << *I2 << "\n";);
1305 // We need to update RE1 and RE2 if we are going to sink the first
1306 // instruction in the basic block down.
1307 bool UpdateRE1 = (I1 == BB1->begin()), UpdateRE2 = (I2 == BB2->begin());
1308 // Sink the instruction.
1309 BBEnd->getInstList().splice(FirstNonPhiInBBEnd, BB1->getInstList(), I1);
1310 if (!OldPN->use_empty())
1311 OldPN->replaceAllUsesWith(I1);
1312 OldPN->eraseFromParent();
1314 if (!I2->use_empty())
1315 I2->replaceAllUsesWith(I1);
1316 I1->intersectOptionalDataWith(I2);
1317 I2->eraseFromParent();
1320 RE1 = BB1->getInstList().rend();
1322 RE2 = BB2->getInstList().rend();
1323 FirstNonPhiInBBEnd = I1;
1330 /// SpeculativelyExecuteBB - Given a conditional branch that goes to BB1
1331 /// and an BB2 and the only successor of BB1 is BB2, hoist simple code
1332 /// (for now, restricted to a single instruction that's side effect free) from
1333 /// the BB1 into the branch block to speculatively execute it.
1338 /// br i1 %t1, label %BB1, label %BB2
1340 /// %t3 = add %t2, c
1346 /// %t4 = add %t2, c
1347 /// %t3 = select i1 %t1, %t2, %t3
1348 static bool SpeculativelyExecuteBB(BranchInst *BI, BasicBlock *BB1) {
1349 // Only speculatively execution a single instruction (not counting the
1350 // terminator) for now.
1351 Instruction *HInst = NULL;
1352 Instruction *Term = BB1->getTerminator();
1353 for (BasicBlock::iterator BBI = BB1->begin(), BBE = BB1->end();
1354 BBI != BBE; ++BBI) {
1355 Instruction *I = BBI;
1357 if (isa<DbgInfoIntrinsic>(I)) continue;
1358 if (I == Term) break;
1365 BasicBlock *BIParent = BI->getParent();
1367 // Check the instruction to be hoisted, if there is one.
1369 // Don't hoist the instruction if it's unsafe or expensive.
1370 if (!isSafeToSpeculativelyExecute(HInst))
1372 if (ComputeSpeculationCost(HInst) > PHINodeFoldingThreshold)
1375 // Do not hoist the instruction if any of its operands are defined but not
1376 // used in this BB. The transformation will prevent the operand from
1377 // being sunk into the use block.
1378 for (User::op_iterator i = HInst->op_begin(), e = HInst->op_end();
1380 Instruction *OpI = dyn_cast<Instruction>(*i);
1381 if (OpI && OpI->getParent() == BIParent &&
1382 !OpI->mayHaveSideEffects() &&
1383 !OpI->isUsedInBasicBlock(BIParent))
1388 // Be conservative for now. FP select instruction can often be expensive.
1389 Value *BrCond = BI->getCondition();
1390 if (isa<FCmpInst>(BrCond))
1393 // If BB1 is actually on the false edge of the conditional branch, remember
1394 // to swap the select operands later.
1395 bool Invert = false;
1396 if (BB1 != BI->getSuccessor(0)) {
1397 assert(BB1 == BI->getSuccessor(1) && "No edge from 'if' block?");
1401 // Collect interesting PHIs, and scan for hazards.
1402 SmallSetVector<std::pair<Value *, Value *>, 4> PHIs;
1403 BasicBlock *BB2 = BB1->getTerminator()->getSuccessor(0);
1404 for (BasicBlock::iterator I = BB2->begin();
1405 PHINode *PN = dyn_cast<PHINode>(I); ++I) {
1406 Value *BB1V = PN->getIncomingValueForBlock(BB1);
1407 Value *BIParentV = PN->getIncomingValueForBlock(BIParent);
1409 // Skip PHIs which are trivial.
1410 if (BB1V == BIParentV)
1413 // Check for saftey.
1414 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(BB1V)) {
1415 // An unfolded ConstantExpr could end up getting expanded into
1416 // Instructions. Don't speculate this and another instruction at
1420 if (!isSafeToSpeculativelyExecute(CE))
1422 if (ComputeSpeculationCost(CE) > PHINodeFoldingThreshold)
1426 // Ok, we may insert a select for this PHI.
1427 PHIs.insert(std::make_pair(BB1V, BIParentV));
1430 // If there are no PHIs to process, bail early. This helps ensure idempotence
1435 // If we get here, we can hoist the instruction and if-convert.
1436 DEBUG(dbgs() << "SPECULATIVELY EXECUTING BB" << *BB1 << "\n";);
1438 // Hoist the instruction.
1440 BIParent->getInstList().splice(BI, BB1->getInstList(), HInst);
1442 // Insert selects and rewrite the PHI operands.
1443 IRBuilder<true, NoFolder> Builder(BI);
1444 for (unsigned i = 0, e = PHIs.size(); i != e; ++i) {
1445 Value *TrueV = PHIs[i].first;
1446 Value *FalseV = PHIs[i].second;
1448 // Create a select whose true value is the speculatively executed value and
1449 // false value is the previously determined FalseV.
1452 SI = cast<SelectInst>
1453 (Builder.CreateSelect(BrCond, FalseV, TrueV,
1454 FalseV->getName() + "." + TrueV->getName()));
1456 SI = cast<SelectInst>
1457 (Builder.CreateSelect(BrCond, TrueV, FalseV,
1458 TrueV->getName() + "." + FalseV->getName()));
1460 // Make the PHI node use the select for all incoming values for "then" and
1462 for (BasicBlock::iterator I = BB2->begin();
1463 PHINode *PN = dyn_cast<PHINode>(I); ++I) {
1464 unsigned BB1I = PN->getBasicBlockIndex(BB1);
1465 unsigned BIParentI = PN->getBasicBlockIndex(BIParent);
1466 Value *BB1V = PN->getIncomingValue(BB1I);
1467 Value *BIParentV = PN->getIncomingValue(BIParentI);
1468 if (TrueV == BB1V && FalseV == BIParentV) {
1469 PN->setIncomingValue(BB1I, SI);
1470 PN->setIncomingValue(BIParentI, SI);
1479 /// BlockIsSimpleEnoughToThreadThrough - Return true if we can thread a branch
1480 /// across this block.
1481 static bool BlockIsSimpleEnoughToThreadThrough(BasicBlock *BB) {
1482 BranchInst *BI = cast<BranchInst>(BB->getTerminator());
1485 for (BasicBlock::iterator BBI = BB->begin(); &*BBI != BI; ++BBI) {
1486 if (isa<DbgInfoIntrinsic>(BBI))
1488 if (Size > 10) return false; // Don't clone large BB's.
1491 // We can only support instructions that do not define values that are
1492 // live outside of the current basic block.
1493 for (Value::use_iterator UI = BBI->use_begin(), E = BBI->use_end();
1495 Instruction *U = cast<Instruction>(*UI);
1496 if (U->getParent() != BB || isa<PHINode>(U)) return false;
1499 // Looks ok, continue checking.
1505 /// FoldCondBranchOnPHI - If we have a conditional branch on a PHI node value
1506 /// that is defined in the same block as the branch and if any PHI entries are
1507 /// constants, thread edges corresponding to that entry to be branches to their
1508 /// ultimate destination.
1509 static bool FoldCondBranchOnPHI(BranchInst *BI, const TargetData *TD) {
1510 BasicBlock *BB = BI->getParent();
1511 PHINode *PN = dyn_cast<PHINode>(BI->getCondition());
1512 // NOTE: we currently cannot transform this case if the PHI node is used
1513 // outside of the block.
1514 if (!PN || PN->getParent() != BB || !PN->hasOneUse())
1517 // Degenerate case of a single entry PHI.
1518 if (PN->getNumIncomingValues() == 1) {
1519 FoldSingleEntryPHINodes(PN->getParent());
1523 // Now we know that this block has multiple preds and two succs.
1524 if (!BlockIsSimpleEnoughToThreadThrough(BB)) return false;
1526 // Okay, this is a simple enough basic block. See if any phi values are
1528 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
1529 ConstantInt *CB = dyn_cast<ConstantInt>(PN->getIncomingValue(i));
1530 if (CB == 0 || !CB->getType()->isIntegerTy(1)) continue;
1532 // Okay, we now know that all edges from PredBB should be revectored to
1533 // branch to RealDest.
1534 BasicBlock *PredBB = PN->getIncomingBlock(i);
1535 BasicBlock *RealDest = BI->getSuccessor(!CB->getZExtValue());
1537 if (RealDest == BB) continue; // Skip self loops.
1538 // Skip if the predecessor's terminator is an indirect branch.
1539 if (isa<IndirectBrInst>(PredBB->getTerminator())) continue;
1541 // The dest block might have PHI nodes, other predecessors and other
1542 // difficult cases. Instead of being smart about this, just insert a new
1543 // block that jumps to the destination block, effectively splitting
1544 // the edge we are about to create.
1545 BasicBlock *EdgeBB = BasicBlock::Create(BB->getContext(),
1546 RealDest->getName()+".critedge",
1547 RealDest->getParent(), RealDest);
1548 BranchInst::Create(RealDest, EdgeBB);
1550 // Update PHI nodes.
1551 AddPredecessorToBlock(RealDest, EdgeBB, BB);
1553 // BB may have instructions that are being threaded over. Clone these
1554 // instructions into EdgeBB. We know that there will be no uses of the
1555 // cloned instructions outside of EdgeBB.
1556 BasicBlock::iterator InsertPt = EdgeBB->begin();
1557 DenseMap<Value*, Value*> TranslateMap; // Track translated values.
1558 for (BasicBlock::iterator BBI = BB->begin(); &*BBI != BI; ++BBI) {
1559 if (PHINode *PN = dyn_cast<PHINode>(BBI)) {
1560 TranslateMap[PN] = PN->getIncomingValueForBlock(PredBB);
1563 // Clone the instruction.
1564 Instruction *N = BBI->clone();
1565 if (BBI->hasName()) N->setName(BBI->getName()+".c");
1567 // Update operands due to translation.
1568 for (User::op_iterator i = N->op_begin(), e = N->op_end();
1570 DenseMap<Value*, Value*>::iterator PI = TranslateMap.find(*i);
1571 if (PI != TranslateMap.end())
1575 // Check for trivial simplification.
1576 if (Value *V = SimplifyInstruction(N, TD)) {
1577 TranslateMap[BBI] = V;
1578 delete N; // Instruction folded away, don't need actual inst
1580 // Insert the new instruction into its new home.
1581 EdgeBB->getInstList().insert(InsertPt, N);
1582 if (!BBI->use_empty())
1583 TranslateMap[BBI] = N;
1587 // Loop over all of the edges from PredBB to BB, changing them to branch
1588 // to EdgeBB instead.
1589 TerminatorInst *PredBBTI = PredBB->getTerminator();
1590 for (unsigned i = 0, e = PredBBTI->getNumSuccessors(); i != e; ++i)
1591 if (PredBBTI->getSuccessor(i) == BB) {
1592 BB->removePredecessor(PredBB);
1593 PredBBTI->setSuccessor(i, EdgeBB);
1596 // Recurse, simplifying any other constants.
1597 return FoldCondBranchOnPHI(BI, TD) | true;
1603 /// FoldTwoEntryPHINode - Given a BB that starts with the specified two-entry
1604 /// PHI node, see if we can eliminate it.
1605 static bool FoldTwoEntryPHINode(PHINode *PN, const TargetData *TD) {
1606 // Ok, this is a two entry PHI node. Check to see if this is a simple "if
1607 // statement", which has a very simple dominance structure. Basically, we
1608 // are trying to find the condition that is being branched on, which
1609 // subsequently causes this merge to happen. We really want control
1610 // dependence information for this check, but simplifycfg can't keep it up
1611 // to date, and this catches most of the cases we care about anyway.
1612 BasicBlock *BB = PN->getParent();
1613 BasicBlock *IfTrue, *IfFalse;
1614 Value *IfCond = GetIfCondition(BB, IfTrue, IfFalse);
1616 // Don't bother if the branch will be constant folded trivially.
1617 isa<ConstantInt>(IfCond))
1620 // Okay, we found that we can merge this two-entry phi node into a select.
1621 // Doing so would require us to fold *all* two entry phi nodes in this block.
1622 // At some point this becomes non-profitable (particularly if the target
1623 // doesn't support cmov's). Only do this transformation if there are two or
1624 // fewer PHI nodes in this block.
1625 unsigned NumPhis = 0;
1626 for (BasicBlock::iterator I = BB->begin(); isa<PHINode>(I); ++NumPhis, ++I)
1630 // Loop over the PHI's seeing if we can promote them all to select
1631 // instructions. While we are at it, keep track of the instructions
1632 // that need to be moved to the dominating block.
1633 SmallPtrSet<Instruction*, 4> AggressiveInsts;
1634 unsigned MaxCostVal0 = PHINodeFoldingThreshold,
1635 MaxCostVal1 = PHINodeFoldingThreshold;
1637 for (BasicBlock::iterator II = BB->begin(); isa<PHINode>(II);) {
1638 PHINode *PN = cast<PHINode>(II++);
1639 if (Value *V = SimplifyInstruction(PN, TD)) {
1640 PN->replaceAllUsesWith(V);
1641 PN->eraseFromParent();
1645 if (!DominatesMergePoint(PN->getIncomingValue(0), BB, &AggressiveInsts,
1647 !DominatesMergePoint(PN->getIncomingValue(1), BB, &AggressiveInsts,
1652 // If we folded the first phi, PN dangles at this point. Refresh it. If
1653 // we ran out of PHIs then we simplified them all.
1654 PN = dyn_cast<PHINode>(BB->begin());
1655 if (PN == 0) return true;
1657 // Don't fold i1 branches on PHIs which contain binary operators. These can
1658 // often be turned into switches and other things.
1659 if (PN->getType()->isIntegerTy(1) &&
1660 (isa<BinaryOperator>(PN->getIncomingValue(0)) ||
1661 isa<BinaryOperator>(PN->getIncomingValue(1)) ||
1662 isa<BinaryOperator>(IfCond)))
1665 // If we all PHI nodes are promotable, check to make sure that all
1666 // instructions in the predecessor blocks can be promoted as well. If
1667 // not, we won't be able to get rid of the control flow, so it's not
1668 // worth promoting to select instructions.
1669 BasicBlock *DomBlock = 0;
1670 BasicBlock *IfBlock1 = PN->getIncomingBlock(0);
1671 BasicBlock *IfBlock2 = PN->getIncomingBlock(1);
1672 if (cast<BranchInst>(IfBlock1->getTerminator())->isConditional()) {
1675 DomBlock = *pred_begin(IfBlock1);
1676 for (BasicBlock::iterator I = IfBlock1->begin();!isa<TerminatorInst>(I);++I)
1677 if (!AggressiveInsts.count(I) && !isa<DbgInfoIntrinsic>(I)) {
1678 // This is not an aggressive instruction that we can promote.
1679 // Because of this, we won't be able to get rid of the control
1680 // flow, so the xform is not worth it.
1685 if (cast<BranchInst>(IfBlock2->getTerminator())->isConditional()) {
1688 DomBlock = *pred_begin(IfBlock2);
1689 for (BasicBlock::iterator I = IfBlock2->begin();!isa<TerminatorInst>(I);++I)
1690 if (!AggressiveInsts.count(I) && !isa<DbgInfoIntrinsic>(I)) {
1691 // This is not an aggressive instruction that we can promote.
1692 // Because of this, we won't be able to get rid of the control
1693 // flow, so the xform is not worth it.
1698 DEBUG(dbgs() << "FOUND IF CONDITION! " << *IfCond << " T: "
1699 << IfTrue->getName() << " F: " << IfFalse->getName() << "\n");
1701 // If we can still promote the PHI nodes after this gauntlet of tests,
1702 // do all of the PHI's now.
1703 Instruction *InsertPt = DomBlock->getTerminator();
1704 IRBuilder<true, NoFolder> Builder(InsertPt);
1706 // Move all 'aggressive' instructions, which are defined in the
1707 // conditional parts of the if's up to the dominating block.
1709 DomBlock->getInstList().splice(InsertPt,
1710 IfBlock1->getInstList(), IfBlock1->begin(),
1711 IfBlock1->getTerminator());
1713 DomBlock->getInstList().splice(InsertPt,
1714 IfBlock2->getInstList(), IfBlock2->begin(),
1715 IfBlock2->getTerminator());
1717 while (PHINode *PN = dyn_cast<PHINode>(BB->begin())) {
1718 // Change the PHI node into a select instruction.
1719 Value *TrueVal = PN->getIncomingValue(PN->getIncomingBlock(0) == IfFalse);
1720 Value *FalseVal = PN->getIncomingValue(PN->getIncomingBlock(0) == IfTrue);
1723 cast<SelectInst>(Builder.CreateSelect(IfCond, TrueVal, FalseVal, ""));
1724 PN->replaceAllUsesWith(NV);
1726 PN->eraseFromParent();
1729 // At this point, IfBlock1 and IfBlock2 are both empty, so our if statement
1730 // has been flattened. Change DomBlock to jump directly to our new block to
1731 // avoid other simplifycfg's kicking in on the diamond.
1732 TerminatorInst *OldTI = DomBlock->getTerminator();
1733 Builder.SetInsertPoint(OldTI);
1734 Builder.CreateBr(BB);
1735 OldTI->eraseFromParent();
1739 /// SimplifyCondBranchToTwoReturns - If we found a conditional branch that goes
1740 /// to two returning blocks, try to merge them together into one return,
1741 /// introducing a select if the return values disagree.
1742 static bool SimplifyCondBranchToTwoReturns(BranchInst *BI,
1743 IRBuilder<> &Builder) {
1744 assert(BI->isConditional() && "Must be a conditional branch");
1745 BasicBlock *TrueSucc = BI->getSuccessor(0);
1746 BasicBlock *FalseSucc = BI->getSuccessor(1);
1747 ReturnInst *TrueRet = cast<ReturnInst>(TrueSucc->getTerminator());
1748 ReturnInst *FalseRet = cast<ReturnInst>(FalseSucc->getTerminator());
1750 // Check to ensure both blocks are empty (just a return) or optionally empty
1751 // with PHI nodes. If there are other instructions, merging would cause extra
1752 // computation on one path or the other.
1753 if (!TrueSucc->getFirstNonPHIOrDbg()->isTerminator())
1755 if (!FalseSucc->getFirstNonPHIOrDbg()->isTerminator())
1758 Builder.SetInsertPoint(BI);
1759 // Okay, we found a branch that is going to two return nodes. If
1760 // there is no return value for this function, just change the
1761 // branch into a return.
1762 if (FalseRet->getNumOperands() == 0) {
1763 TrueSucc->removePredecessor(BI->getParent());
1764 FalseSucc->removePredecessor(BI->getParent());
1765 Builder.CreateRetVoid();
1766 EraseTerminatorInstAndDCECond(BI);
1770 // Otherwise, figure out what the true and false return values are
1771 // so we can insert a new select instruction.
1772 Value *TrueValue = TrueRet->getReturnValue();
1773 Value *FalseValue = FalseRet->getReturnValue();
1775 // Unwrap any PHI nodes in the return blocks.
1776 if (PHINode *TVPN = dyn_cast_or_null<PHINode>(TrueValue))
1777 if (TVPN->getParent() == TrueSucc)
1778 TrueValue = TVPN->getIncomingValueForBlock(BI->getParent());
1779 if (PHINode *FVPN = dyn_cast_or_null<PHINode>(FalseValue))
1780 if (FVPN->getParent() == FalseSucc)
1781 FalseValue = FVPN->getIncomingValueForBlock(BI->getParent());
1783 // In order for this transformation to be safe, we must be able to
1784 // unconditionally execute both operands to the return. This is
1785 // normally the case, but we could have a potentially-trapping
1786 // constant expression that prevents this transformation from being
1788 if (ConstantExpr *TCV = dyn_cast_or_null<ConstantExpr>(TrueValue))
1791 if (ConstantExpr *FCV = dyn_cast_or_null<ConstantExpr>(FalseValue))
1795 // Okay, we collected all the mapped values and checked them for sanity, and
1796 // defined to really do this transformation. First, update the CFG.
1797 TrueSucc->removePredecessor(BI->getParent());
1798 FalseSucc->removePredecessor(BI->getParent());
1800 // Insert select instructions where needed.
1801 Value *BrCond = BI->getCondition();
1803 // Insert a select if the results differ.
1804 if (TrueValue == FalseValue || isa<UndefValue>(FalseValue)) {
1805 } else if (isa<UndefValue>(TrueValue)) {
1806 TrueValue = FalseValue;
1808 TrueValue = Builder.CreateSelect(BrCond, TrueValue,
1809 FalseValue, "retval");
1813 Value *RI = !TrueValue ?
1814 Builder.CreateRetVoid() : Builder.CreateRet(TrueValue);
1818 DEBUG(dbgs() << "\nCHANGING BRANCH TO TWO RETURNS INTO SELECT:"
1819 << "\n " << *BI << "NewRet = " << *RI
1820 << "TRUEBLOCK: " << *TrueSucc << "FALSEBLOCK: "<< *FalseSucc);
1822 EraseTerminatorInstAndDCECond(BI);
1827 /// ExtractBranchMetadata - Given a conditional BranchInstruction, retrieve the
1828 /// probabilities of the branch taking each edge. Fills in the two APInt
1829 /// parameters and return true, or returns false if no or invalid metadata was
1831 static bool ExtractBranchMetadata(BranchInst *BI,
1832 uint64_t &ProbTrue, uint64_t &ProbFalse) {
1833 assert(BI->isConditional() &&
1834 "Looking for probabilities on unconditional branch?");
1835 MDNode *ProfileData = BI->getMetadata(LLVMContext::MD_prof);
1836 if (!ProfileData || ProfileData->getNumOperands() != 3) return false;
1837 ConstantInt *CITrue = dyn_cast<ConstantInt>(ProfileData->getOperand(1));
1838 ConstantInt *CIFalse = dyn_cast<ConstantInt>(ProfileData->getOperand(2));
1839 if (!CITrue || !CIFalse) return false;
1840 ProbTrue = CITrue->getValue().getZExtValue();
1841 ProbFalse = CIFalse->getValue().getZExtValue();
1845 /// checkCSEInPredecessor - Return true if the given instruction is available
1846 /// in its predecessor block. If yes, the instruction will be removed.
1848 static bool checkCSEInPredecessor(Instruction *Inst, BasicBlock *PB) {
1849 if (!isa<BinaryOperator>(Inst) && !isa<CmpInst>(Inst))
1851 for (BasicBlock::iterator I = PB->begin(), E = PB->end(); I != E; I++) {
1852 Instruction *PBI = &*I;
1853 // Check whether Inst and PBI generate the same value.
1854 if (Inst->isIdenticalTo(PBI)) {
1855 Inst->replaceAllUsesWith(PBI);
1856 Inst->eraseFromParent();
1863 /// FoldBranchToCommonDest - If this basic block is simple enough, and if a
1864 /// predecessor branches to us and one of our successors, fold the block into
1865 /// the predecessor and use logical operations to pick the right destination.
1866 bool llvm::FoldBranchToCommonDest(BranchInst *BI) {
1867 BasicBlock *BB = BI->getParent();
1869 Instruction *Cond = 0;
1870 if (BI->isConditional())
1871 Cond = dyn_cast<Instruction>(BI->getCondition());
1873 // For unconditional branch, check for a simple CFG pattern, where
1874 // BB has a single predecessor and BB's successor is also its predecessor's
1875 // successor. If such pattern exisits, check for CSE between BB and its
1877 if (BasicBlock *PB = BB->getSinglePredecessor())
1878 if (BranchInst *PBI = dyn_cast<BranchInst>(PB->getTerminator()))
1879 if (PBI->isConditional() &&
1880 (BI->getSuccessor(0) == PBI->getSuccessor(0) ||
1881 BI->getSuccessor(0) == PBI->getSuccessor(1))) {
1882 for (BasicBlock::iterator I = BB->begin(), E = BB->end();
1884 Instruction *Curr = I++;
1885 if (isa<CmpInst>(Curr)) {
1889 // Quit if we can't remove this instruction.
1890 if (!checkCSEInPredecessor(Curr, PB))
1899 if (Cond == 0 || (!isa<CmpInst>(Cond) && !isa<BinaryOperator>(Cond)) ||
1900 Cond->getParent() != BB || !Cond->hasOneUse())
1903 // Only allow this if the condition is a simple instruction that can be
1904 // executed unconditionally. It must be in the same block as the branch, and
1905 // must be at the front of the block.
1906 BasicBlock::iterator FrontIt = BB->front();
1908 // Ignore dbg intrinsics.
1909 while (isa<DbgInfoIntrinsic>(FrontIt)) ++FrontIt;
1911 // Allow a single instruction to be hoisted in addition to the compare
1912 // that feeds the branch. We later ensure that any values that _it_ uses
1913 // were also live in the predecessor, so that we don't unnecessarily create
1914 // register pressure or inhibit out-of-order execution.
1915 Instruction *BonusInst = 0;
1916 if (&*FrontIt != Cond &&
1917 FrontIt->hasOneUse() && *FrontIt->use_begin() == Cond &&
1918 isSafeToSpeculativelyExecute(FrontIt)) {
1919 BonusInst = &*FrontIt;
1922 // Ignore dbg intrinsics.
1923 while (isa<DbgInfoIntrinsic>(FrontIt)) ++FrontIt;
1926 // Only a single bonus inst is allowed.
1927 if (&*FrontIt != Cond)
1930 // Make sure the instruction after the condition is the cond branch.
1931 BasicBlock::iterator CondIt = Cond; ++CondIt;
1933 // Ingore dbg intrinsics.
1934 while (isa<DbgInfoIntrinsic>(CondIt)) ++CondIt;
1939 // Cond is known to be a compare or binary operator. Check to make sure that
1940 // neither operand is a potentially-trapping constant expression.
1941 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Cond->getOperand(0)))
1944 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Cond->getOperand(1)))
1948 // Finally, don't infinitely unroll conditional loops.
1949 BasicBlock *TrueDest = BI->getSuccessor(0);
1950 BasicBlock *FalseDest = (BI->isConditional()) ? BI->getSuccessor(1) : 0;
1951 if (TrueDest == BB || FalseDest == BB)
1954 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
1955 BasicBlock *PredBlock = *PI;
1956 BranchInst *PBI = dyn_cast<BranchInst>(PredBlock->getTerminator());
1958 // Check that we have two conditional branches. If there is a PHI node in
1959 // the common successor, verify that the same value flows in from both
1961 SmallVector<PHINode*, 4> PHIs;
1962 if (PBI == 0 || PBI->isUnconditional() ||
1963 (BI->isConditional() &&
1964 !SafeToMergeTerminators(BI, PBI)) ||
1965 (!BI->isConditional() &&
1966 !isProfitableToFoldUnconditional(BI, PBI, Cond, PHIs)))
1969 // Determine if the two branches share a common destination.
1970 Instruction::BinaryOps Opc = Instruction::BinaryOpsEnd;
1971 bool InvertPredCond = false;
1973 if (BI->isConditional()) {
1974 if (PBI->getSuccessor(0) == TrueDest)
1975 Opc = Instruction::Or;
1976 else if (PBI->getSuccessor(1) == FalseDest)
1977 Opc = Instruction::And;
1978 else if (PBI->getSuccessor(0) == FalseDest)
1979 Opc = Instruction::And, InvertPredCond = true;
1980 else if (PBI->getSuccessor(1) == TrueDest)
1981 Opc = Instruction::Or, InvertPredCond = true;
1985 if (PBI->getSuccessor(0) != TrueDest && PBI->getSuccessor(1) != TrueDest)
1989 // Ensure that any values used in the bonus instruction are also used
1990 // by the terminator of the predecessor. This means that those values
1991 // must already have been resolved, so we won't be inhibiting the
1992 // out-of-order core by speculating them earlier.
1994 // Collect the values used by the bonus inst
1995 SmallPtrSet<Value*, 4> UsedValues;
1996 for (Instruction::op_iterator OI = BonusInst->op_begin(),
1997 OE = BonusInst->op_end(); OI != OE; ++OI) {
1999 if (!isa<Constant>(V))
2000 UsedValues.insert(V);
2003 SmallVector<std::pair<Value*, unsigned>, 4> Worklist;
2004 Worklist.push_back(std::make_pair(PBI->getOperand(0), 0));
2006 // Walk up to four levels back up the use-def chain of the predecessor's
2007 // terminator to see if all those values were used. The choice of four
2008 // levels is arbitrary, to provide a compile-time-cost bound.
2009 while (!Worklist.empty()) {
2010 std::pair<Value*, unsigned> Pair = Worklist.back();
2011 Worklist.pop_back();
2013 if (Pair.second >= 4) continue;
2014 UsedValues.erase(Pair.first);
2015 if (UsedValues.empty()) break;
2017 if (Instruction *I = dyn_cast<Instruction>(Pair.first)) {
2018 for (Instruction::op_iterator OI = I->op_begin(), OE = I->op_end();
2020 Worklist.push_back(std::make_pair(OI->get(), Pair.second+1));
2024 if (!UsedValues.empty()) return false;
2027 DEBUG(dbgs() << "FOLDING BRANCH TO COMMON DEST:\n" << *PBI << *BB);
2028 IRBuilder<> Builder(PBI);
2030 // If we need to invert the condition in the pred block to match, do so now.
2031 if (InvertPredCond) {
2032 Value *NewCond = PBI->getCondition();
2034 if (NewCond->hasOneUse() && isa<CmpInst>(NewCond)) {
2035 CmpInst *CI = cast<CmpInst>(NewCond);
2036 CI->setPredicate(CI->getInversePredicate());
2038 NewCond = Builder.CreateNot(NewCond,
2039 PBI->getCondition()->getName()+".not");
2042 PBI->setCondition(NewCond);
2043 PBI->swapSuccessors();
2046 // If we have a bonus inst, clone it into the predecessor block.
2047 Instruction *NewBonus = 0;
2049 NewBonus = BonusInst->clone();
2050 PredBlock->getInstList().insert(PBI, NewBonus);
2051 NewBonus->takeName(BonusInst);
2052 BonusInst->setName(BonusInst->getName()+".old");
2055 // Clone Cond into the predecessor basic block, and or/and the
2056 // two conditions together.
2057 Instruction *New = Cond->clone();
2058 if (BonusInst) New->replaceUsesOfWith(BonusInst, NewBonus);
2059 PredBlock->getInstList().insert(PBI, New);
2060 New->takeName(Cond);
2061 Cond->setName(New->getName()+".old");
2063 if (BI->isConditional()) {
2064 Instruction *NewCond =
2065 cast<Instruction>(Builder.CreateBinOp(Opc, PBI->getCondition(),
2067 PBI->setCondition(NewCond);
2069 uint64_t PredTrueWeight, PredFalseWeight, SuccTrueWeight, SuccFalseWeight;
2070 bool PredHasWeights = ExtractBranchMetadata(PBI, PredTrueWeight,
2072 bool SuccHasWeights = ExtractBranchMetadata(BI, SuccTrueWeight,
2074 SmallVector<uint64_t, 8> NewWeights;
2076 if (PBI->getSuccessor(0) == BB) {
2077 if (PredHasWeights && SuccHasWeights) {
2078 // PBI: br i1 %x, BB, FalseDest
2079 // BI: br i1 %y, TrueDest, FalseDest
2080 //TrueWeight is TrueWeight for PBI * TrueWeight for BI.
2081 NewWeights.push_back(PredTrueWeight * SuccTrueWeight);
2082 //FalseWeight is FalseWeight for PBI * TotalWeight for BI +
2083 // TrueWeight for PBI * FalseWeight for BI.
2084 // We assume that total weights of a BranchInst can fit into 32 bits.
2085 // Therefore, we will not have overflow using 64-bit arithmetic.
2086 NewWeights.push_back(PredFalseWeight * (SuccFalseWeight +
2087 SuccTrueWeight) + PredTrueWeight * SuccFalseWeight);
2089 AddPredecessorToBlock(TrueDest, PredBlock, BB);
2090 PBI->setSuccessor(0, TrueDest);
2092 if (PBI->getSuccessor(1) == BB) {
2093 if (PredHasWeights && SuccHasWeights) {
2094 // PBI: br i1 %x, TrueDest, BB
2095 // BI: br i1 %y, TrueDest, FalseDest
2096 //TrueWeight is TrueWeight for PBI * TotalWeight for BI +
2097 // FalseWeight for PBI * TrueWeight for BI.
2098 NewWeights.push_back(PredTrueWeight * (SuccFalseWeight +
2099 SuccTrueWeight) + PredFalseWeight * SuccTrueWeight);
2100 //FalseWeight is FalseWeight for PBI * FalseWeight for BI.
2101 NewWeights.push_back(PredFalseWeight * SuccFalseWeight);
2103 AddPredecessorToBlock(FalseDest, PredBlock, BB);
2104 PBI->setSuccessor(1, FalseDest);
2106 if (NewWeights.size() == 2) {
2107 // Halve the weights if any of them cannot fit in an uint32_t
2108 FitWeights(NewWeights);
2110 SmallVector<uint32_t, 8> MDWeights(NewWeights.begin(),NewWeights.end());
2111 PBI->setMetadata(LLVMContext::MD_prof,
2112 MDBuilder(BI->getContext()).
2113 createBranchWeights(MDWeights));
2115 PBI->setMetadata(LLVMContext::MD_prof, NULL);
2117 // Update PHI nodes in the common successors.
2118 for (unsigned i = 0, e = PHIs.size(); i != e; ++i) {
2119 ConstantInt *PBI_C = cast<ConstantInt>(
2120 PHIs[i]->getIncomingValueForBlock(PBI->getParent()));
2121 assert(PBI_C->getType()->isIntegerTy(1));
2122 Instruction *MergedCond = 0;
2123 if (PBI->getSuccessor(0) == TrueDest) {
2124 // Create (PBI_Cond and PBI_C) or (!PBI_Cond and BI_Value)
2125 // PBI_C is true: PBI_Cond or (!PBI_Cond and BI_Value)
2126 // is false: !PBI_Cond and BI_Value
2127 Instruction *NotCond =
2128 cast<Instruction>(Builder.CreateNot(PBI->getCondition(),
2131 cast<Instruction>(Builder.CreateBinOp(Instruction::And,
2136 cast<Instruction>(Builder.CreateBinOp(Instruction::Or,
2137 PBI->getCondition(), MergedCond,
2140 // Create (PBI_Cond and BI_Value) or (!PBI_Cond and PBI_C)
2141 // PBI_C is true: (PBI_Cond and BI_Value) or (!PBI_Cond)
2142 // is false: PBI_Cond and BI_Value
2144 cast<Instruction>(Builder.CreateBinOp(Instruction::And,
2145 PBI->getCondition(), New,
2147 if (PBI_C->isOne()) {
2148 Instruction *NotCond =
2149 cast<Instruction>(Builder.CreateNot(PBI->getCondition(),
2152 cast<Instruction>(Builder.CreateBinOp(Instruction::Or,
2153 NotCond, MergedCond,
2158 PHIs[i]->setIncomingValue(PHIs[i]->getBasicBlockIndex(PBI->getParent()),
2161 // Change PBI from Conditional to Unconditional.
2162 BranchInst *New_PBI = BranchInst::Create(TrueDest, PBI);
2163 EraseTerminatorInstAndDCECond(PBI);
2167 // TODO: If BB is reachable from all paths through PredBlock, then we
2168 // could replace PBI's branch probabilities with BI's.
2170 // Copy any debug value intrinsics into the end of PredBlock.
2171 for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
2172 if (isa<DbgInfoIntrinsic>(*I))
2173 I->clone()->insertBefore(PBI);
2180 /// SimplifyCondBranchToCondBranch - If we have a conditional branch as a
2181 /// predecessor of another block, this function tries to simplify it. We know
2182 /// that PBI and BI are both conditional branches, and BI is in one of the
2183 /// successor blocks of PBI - PBI branches to BI.
2184 static bool SimplifyCondBranchToCondBranch(BranchInst *PBI, BranchInst *BI) {
2185 assert(PBI->isConditional() && BI->isConditional());
2186 BasicBlock *BB = BI->getParent();
2188 // If this block ends with a branch instruction, and if there is a
2189 // predecessor that ends on a branch of the same condition, make
2190 // this conditional branch redundant.
2191 if (PBI->getCondition() == BI->getCondition() &&
2192 PBI->getSuccessor(0) != PBI->getSuccessor(1)) {
2193 // Okay, the outcome of this conditional branch is statically
2194 // knowable. If this block had a single pred, handle specially.
2195 if (BB->getSinglePredecessor()) {
2196 // Turn this into a branch on constant.
2197 bool CondIsTrue = PBI->getSuccessor(0) == BB;
2198 BI->setCondition(ConstantInt::get(Type::getInt1Ty(BB->getContext()),
2200 return true; // Nuke the branch on constant.
2203 // Otherwise, if there are multiple predecessors, insert a PHI that merges
2204 // in the constant and simplify the block result. Subsequent passes of
2205 // simplifycfg will thread the block.
2206 if (BlockIsSimpleEnoughToThreadThrough(BB)) {
2207 pred_iterator PB = pred_begin(BB), PE = pred_end(BB);
2208 PHINode *NewPN = PHINode::Create(Type::getInt1Ty(BB->getContext()),
2209 std::distance(PB, PE),
2210 BI->getCondition()->getName() + ".pr",
2212 // Okay, we're going to insert the PHI node. Since PBI is not the only
2213 // predecessor, compute the PHI'd conditional value for all of the preds.
2214 // Any predecessor where the condition is not computable we keep symbolic.
2215 for (pred_iterator PI = PB; PI != PE; ++PI) {
2216 BasicBlock *P = *PI;
2217 if ((PBI = dyn_cast<BranchInst>(P->getTerminator())) &&
2218 PBI != BI && PBI->isConditional() &&
2219 PBI->getCondition() == BI->getCondition() &&
2220 PBI->getSuccessor(0) != PBI->getSuccessor(1)) {
2221 bool CondIsTrue = PBI->getSuccessor(0) == BB;
2222 NewPN->addIncoming(ConstantInt::get(Type::getInt1Ty(BB->getContext()),
2225 NewPN->addIncoming(BI->getCondition(), P);
2229 BI->setCondition(NewPN);
2234 // If this is a conditional branch in an empty block, and if any
2235 // predecessors is a conditional branch to one of our destinations,
2236 // fold the conditions into logical ops and one cond br.
2237 BasicBlock::iterator BBI = BB->begin();
2238 // Ignore dbg intrinsics.
2239 while (isa<DbgInfoIntrinsic>(BBI))
2245 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(BI->getCondition()))
2250 if (PBI->getSuccessor(0) == BI->getSuccessor(0))
2252 else if (PBI->getSuccessor(0) == BI->getSuccessor(1))
2253 PBIOp = 0, BIOp = 1;
2254 else if (PBI->getSuccessor(1) == BI->getSuccessor(0))
2255 PBIOp = 1, BIOp = 0;
2256 else if (PBI->getSuccessor(1) == BI->getSuccessor(1))
2261 // Check to make sure that the other destination of this branch
2262 // isn't BB itself. If so, this is an infinite loop that will
2263 // keep getting unwound.
2264 if (PBI->getSuccessor(PBIOp) == BB)
2267 // Do not perform this transformation if it would require
2268 // insertion of a large number of select instructions. For targets
2269 // without predication/cmovs, this is a big pessimization.
2270 BasicBlock *CommonDest = PBI->getSuccessor(PBIOp);
2272 unsigned NumPhis = 0;
2273 for (BasicBlock::iterator II = CommonDest->begin();
2274 isa<PHINode>(II); ++II, ++NumPhis)
2275 if (NumPhis > 2) // Disable this xform.
2278 // Finally, if everything is ok, fold the branches to logical ops.
2279 BasicBlock *OtherDest = BI->getSuccessor(BIOp ^ 1);
2281 DEBUG(dbgs() << "FOLDING BRs:" << *PBI->getParent()
2282 << "AND: " << *BI->getParent());
2285 // If OtherDest *is* BB, then BB is a basic block with a single conditional
2286 // branch in it, where one edge (OtherDest) goes back to itself but the other
2287 // exits. We don't *know* that the program avoids the infinite loop
2288 // (even though that seems likely). If we do this xform naively, we'll end up
2289 // recursively unpeeling the loop. Since we know that (after the xform is
2290 // done) that the block *is* infinite if reached, we just make it an obviously
2291 // infinite loop with no cond branch.
2292 if (OtherDest == BB) {
2293 // Insert it at the end of the function, because it's either code,
2294 // or it won't matter if it's hot. :)
2295 BasicBlock *InfLoopBlock = BasicBlock::Create(BB->getContext(),
2296 "infloop", BB->getParent());
2297 BranchInst::Create(InfLoopBlock, InfLoopBlock);
2298 OtherDest = InfLoopBlock;
2301 DEBUG(dbgs() << *PBI->getParent()->getParent());
2303 // BI may have other predecessors. Because of this, we leave
2304 // it alone, but modify PBI.
2306 // Make sure we get to CommonDest on True&True directions.
2307 Value *PBICond = PBI->getCondition();
2308 IRBuilder<true, NoFolder> Builder(PBI);
2310 PBICond = Builder.CreateNot(PBICond, PBICond->getName()+".not");
2312 Value *BICond = BI->getCondition();
2314 BICond = Builder.CreateNot(BICond, BICond->getName()+".not");
2316 // Merge the conditions.
2317 Value *Cond = Builder.CreateOr(PBICond, BICond, "brmerge");
2319 // Modify PBI to branch on the new condition to the new dests.
2320 PBI->setCondition(Cond);
2321 PBI->setSuccessor(0, CommonDest);
2322 PBI->setSuccessor(1, OtherDest);
2324 // Update branch weight for PBI.
2325 uint64_t PredTrueWeight, PredFalseWeight, SuccTrueWeight, SuccFalseWeight;
2326 bool PredHasWeights = ExtractBranchMetadata(PBI, PredTrueWeight,
2328 bool SuccHasWeights = ExtractBranchMetadata(BI, SuccTrueWeight,
2330 if (PredHasWeights && SuccHasWeights) {
2331 uint64_t PredCommon = PBIOp ? PredFalseWeight : PredTrueWeight;
2332 uint64_t PredOther = PBIOp ?PredTrueWeight : PredFalseWeight;
2333 uint64_t SuccCommon = BIOp ? SuccFalseWeight : SuccTrueWeight;
2334 uint64_t SuccOther = BIOp ? SuccTrueWeight : SuccFalseWeight;
2335 // The weight to CommonDest should be PredCommon * SuccTotal +
2336 // PredOther * SuccCommon.
2337 // The weight to OtherDest should be PredOther * SuccOther.
2338 SmallVector<uint64_t, 2> NewWeights;
2339 NewWeights.push_back(PredCommon * (SuccCommon + SuccOther) +
2340 PredOther * SuccCommon);
2341 NewWeights.push_back(PredOther * SuccOther);
2342 // Halve the weights if any of them cannot fit in an uint32_t
2343 FitWeights(NewWeights);
2345 SmallVector<uint32_t, 2> MDWeights(NewWeights.begin(),NewWeights.end());
2346 PBI->setMetadata(LLVMContext::MD_prof,
2347 MDBuilder(BI->getContext()).
2348 createBranchWeights(MDWeights));
2351 // OtherDest may have phi nodes. If so, add an entry from PBI's
2352 // block that are identical to the entries for BI's block.
2353 AddPredecessorToBlock(OtherDest, PBI->getParent(), BB);
2355 // We know that the CommonDest already had an edge from PBI to
2356 // it. If it has PHIs though, the PHIs may have different
2357 // entries for BB and PBI's BB. If so, insert a select to make
2360 for (BasicBlock::iterator II = CommonDest->begin();
2361 (PN = dyn_cast<PHINode>(II)); ++II) {
2362 Value *BIV = PN->getIncomingValueForBlock(BB);
2363 unsigned PBBIdx = PN->getBasicBlockIndex(PBI->getParent());
2364 Value *PBIV = PN->getIncomingValue(PBBIdx);
2366 // Insert a select in PBI to pick the right value.
2367 Value *NV = cast<SelectInst>
2368 (Builder.CreateSelect(PBICond, PBIV, BIV, PBIV->getName()+".mux"));
2369 PN->setIncomingValue(PBBIdx, NV);
2373 DEBUG(dbgs() << "INTO: " << *PBI->getParent());
2374 DEBUG(dbgs() << *PBI->getParent()->getParent());
2376 // This basic block is probably dead. We know it has at least
2377 // one fewer predecessor.
2381 // SimplifyTerminatorOnSelect - Simplifies a terminator by replacing it with a
2382 // branch to TrueBB if Cond is true or to FalseBB if Cond is false.
2383 // Takes care of updating the successors and removing the old terminator.
2384 // Also makes sure not to introduce new successors by assuming that edges to
2385 // non-successor TrueBBs and FalseBBs aren't reachable.
2386 static bool SimplifyTerminatorOnSelect(TerminatorInst *OldTerm, Value *Cond,
2387 BasicBlock *TrueBB, BasicBlock *FalseBB,
2388 uint32_t TrueWeight,
2389 uint32_t FalseWeight){
2390 // Remove any superfluous successor edges from the CFG.
2391 // First, figure out which successors to preserve.
2392 // If TrueBB and FalseBB are equal, only try to preserve one copy of that
2394 BasicBlock *KeepEdge1 = TrueBB;
2395 BasicBlock *KeepEdge2 = TrueBB != FalseBB ? FalseBB : 0;
2397 // Then remove the rest.
2398 for (unsigned I = 0, E = OldTerm->getNumSuccessors(); I != E; ++I) {
2399 BasicBlock *Succ = OldTerm->getSuccessor(I);
2400 // Make sure only to keep exactly one copy of each edge.
2401 if (Succ == KeepEdge1)
2403 else if (Succ == KeepEdge2)
2406 Succ->removePredecessor(OldTerm->getParent());
2409 IRBuilder<> Builder(OldTerm);
2410 Builder.SetCurrentDebugLocation(OldTerm->getDebugLoc());
2412 // Insert an appropriate new terminator.
2413 if ((KeepEdge1 == 0) && (KeepEdge2 == 0)) {
2414 if (TrueBB == FalseBB)
2415 // We were only looking for one successor, and it was present.
2416 // Create an unconditional branch to it.
2417 Builder.CreateBr(TrueBB);
2419 // We found both of the successors we were looking for.
2420 // Create a conditional branch sharing the condition of the select.
2421 BranchInst *NewBI = Builder.CreateCondBr(Cond, TrueBB, FalseBB);
2422 if (TrueWeight != FalseWeight)
2423 NewBI->setMetadata(LLVMContext::MD_prof,
2424 MDBuilder(OldTerm->getContext()).
2425 createBranchWeights(TrueWeight, FalseWeight));
2427 } else if (KeepEdge1 && (KeepEdge2 || TrueBB == FalseBB)) {
2428 // Neither of the selected blocks were successors, so this
2429 // terminator must be unreachable.
2430 new UnreachableInst(OldTerm->getContext(), OldTerm);
2432 // One of the selected values was a successor, but the other wasn't.
2433 // Insert an unconditional branch to the one that was found;
2434 // the edge to the one that wasn't must be unreachable.
2436 // Only TrueBB was found.
2437 Builder.CreateBr(TrueBB);
2439 // Only FalseBB was found.
2440 Builder.CreateBr(FalseBB);
2443 EraseTerminatorInstAndDCECond(OldTerm);
2447 // SimplifySwitchOnSelect - Replaces
2448 // (switch (select cond, X, Y)) on constant X, Y
2449 // with a branch - conditional if X and Y lead to distinct BBs,
2450 // unconditional otherwise.
2451 static bool SimplifySwitchOnSelect(SwitchInst *SI, SelectInst *Select) {
2452 // Check for constant integer values in the select.
2453 ConstantInt *TrueVal = dyn_cast<ConstantInt>(Select->getTrueValue());
2454 ConstantInt *FalseVal = dyn_cast<ConstantInt>(Select->getFalseValue());
2455 if (!TrueVal || !FalseVal)
2458 // Find the relevant condition and destinations.
2459 Value *Condition = Select->getCondition();
2460 BasicBlock *TrueBB = SI->findCaseValue(TrueVal).getCaseSuccessor();
2461 BasicBlock *FalseBB = SI->findCaseValue(FalseVal).getCaseSuccessor();
2463 // Get weight for TrueBB and FalseBB.
2464 uint32_t TrueWeight = 0, FalseWeight = 0;
2465 SmallVector<uint64_t, 8> Weights;
2466 bool HasWeights = HasBranchWeights(SI);
2468 GetBranchWeights(SI, Weights);
2469 if (Weights.size() == 1 + SI->getNumCases()) {
2470 TrueWeight = (uint32_t)Weights[SI->findCaseValue(TrueVal).
2471 getSuccessorIndex()];
2472 FalseWeight = (uint32_t)Weights[SI->findCaseValue(FalseVal).
2473 getSuccessorIndex()];
2477 // Perform the actual simplification.
2478 return SimplifyTerminatorOnSelect(SI, Condition, TrueBB, FalseBB,
2479 TrueWeight, FalseWeight);
2482 // SimplifyIndirectBrOnSelect - Replaces
2483 // (indirectbr (select cond, blockaddress(@fn, BlockA),
2484 // blockaddress(@fn, BlockB)))
2486 // (br cond, BlockA, BlockB).
2487 static bool SimplifyIndirectBrOnSelect(IndirectBrInst *IBI, SelectInst *SI) {
2488 // Check that both operands of the select are block addresses.
2489 BlockAddress *TBA = dyn_cast<BlockAddress>(SI->getTrueValue());
2490 BlockAddress *FBA = dyn_cast<BlockAddress>(SI->getFalseValue());
2494 // Extract the actual blocks.
2495 BasicBlock *TrueBB = TBA->getBasicBlock();
2496 BasicBlock *FalseBB = FBA->getBasicBlock();
2498 // Perform the actual simplification.
2499 return SimplifyTerminatorOnSelect(IBI, SI->getCondition(), TrueBB, FalseBB,
2503 /// TryToSimplifyUncondBranchWithICmpInIt - This is called when we find an icmp
2504 /// instruction (a seteq/setne with a constant) as the only instruction in a
2505 /// block that ends with an uncond branch. We are looking for a very specific
2506 /// pattern that occurs when "A == 1 || A == 2 || A == 3" gets simplified. In
2507 /// this case, we merge the first two "or's of icmp" into a switch, but then the
2508 /// default value goes to an uncond block with a seteq in it, we get something
2511 /// switch i8 %A, label %DEFAULT [ i8 1, label %end i8 2, label %end ]
2513 /// %tmp = icmp eq i8 %A, 92
2516 /// ... = phi i1 [ true, %entry ], [ %tmp, %DEFAULT ], [ true, %entry ]
2518 /// We prefer to split the edge to 'end' so that there is a true/false entry to
2519 /// the PHI, merging the third icmp into the switch.
2520 static bool TryToSimplifyUncondBranchWithICmpInIt(ICmpInst *ICI,
2521 const TargetData *TD,
2522 IRBuilder<> &Builder) {
2523 BasicBlock *BB = ICI->getParent();
2525 // If the block has any PHIs in it or the icmp has multiple uses, it is too
2527 if (isa<PHINode>(BB->begin()) || !ICI->hasOneUse()) return false;
2529 Value *V = ICI->getOperand(0);
2530 ConstantInt *Cst = cast<ConstantInt>(ICI->getOperand(1));
2532 // The pattern we're looking for is where our only predecessor is a switch on
2533 // 'V' and this block is the default case for the switch. In this case we can
2534 // fold the compared value into the switch to simplify things.
2535 BasicBlock *Pred = BB->getSinglePredecessor();
2536 if (Pred == 0 || !isa<SwitchInst>(Pred->getTerminator())) return false;
2538 SwitchInst *SI = cast<SwitchInst>(Pred->getTerminator());
2539 if (SI->getCondition() != V)
2542 // If BB is reachable on a non-default case, then we simply know the value of
2543 // V in this block. Substitute it and constant fold the icmp instruction
2545 if (SI->getDefaultDest() != BB) {
2546 ConstantInt *VVal = SI->findCaseDest(BB);
2547 assert(VVal && "Should have a unique destination value");
2548 ICI->setOperand(0, VVal);
2550 if (Value *V = SimplifyInstruction(ICI, TD)) {
2551 ICI->replaceAllUsesWith(V);
2552 ICI->eraseFromParent();
2554 // BB is now empty, so it is likely to simplify away.
2555 return SimplifyCFG(BB) | true;
2558 // Ok, the block is reachable from the default dest. If the constant we're
2559 // comparing exists in one of the other edges, then we can constant fold ICI
2561 if (SI->findCaseValue(Cst) != SI->case_default()) {
2563 if (ICI->getPredicate() == ICmpInst::ICMP_EQ)
2564 V = ConstantInt::getFalse(BB->getContext());
2566 V = ConstantInt::getTrue(BB->getContext());
2568 ICI->replaceAllUsesWith(V);
2569 ICI->eraseFromParent();
2570 // BB is now empty, so it is likely to simplify away.
2571 return SimplifyCFG(BB) | true;
2574 // The use of the icmp has to be in the 'end' block, by the only PHI node in
2576 BasicBlock *SuccBlock = BB->getTerminator()->getSuccessor(0);
2577 PHINode *PHIUse = dyn_cast<PHINode>(ICI->use_back());
2578 if (PHIUse == 0 || PHIUse != &SuccBlock->front() ||
2579 isa<PHINode>(++BasicBlock::iterator(PHIUse)))
2582 // If the icmp is a SETEQ, then the default dest gets false, the new edge gets
2584 Constant *DefaultCst = ConstantInt::getTrue(BB->getContext());
2585 Constant *NewCst = ConstantInt::getFalse(BB->getContext());
2587 if (ICI->getPredicate() == ICmpInst::ICMP_EQ)
2588 std::swap(DefaultCst, NewCst);
2590 // Replace ICI (which is used by the PHI for the default value) with true or
2591 // false depending on if it is EQ or NE.
2592 ICI->replaceAllUsesWith(DefaultCst);
2593 ICI->eraseFromParent();
2595 // Okay, the switch goes to this block on a default value. Add an edge from
2596 // the switch to the merge point on the compared value.
2597 BasicBlock *NewBB = BasicBlock::Create(BB->getContext(), "switch.edge",
2598 BB->getParent(), BB);
2599 SmallVector<uint64_t, 8> Weights;
2600 bool HasWeights = HasBranchWeights(SI);
2602 GetBranchWeights(SI, Weights);
2603 if (Weights.size() == 1 + SI->getNumCases()) {
2604 // Split weight for default case to case for "Cst".
2605 Weights[0] = (Weights[0]+1) >> 1;
2606 Weights.push_back(Weights[0]);
2608 SmallVector<uint32_t, 8> MDWeights(Weights.begin(), Weights.end());
2609 SI->setMetadata(LLVMContext::MD_prof,
2610 MDBuilder(SI->getContext()).
2611 createBranchWeights(MDWeights));
2614 SI->addCase(Cst, NewBB);
2616 // NewBB branches to the phi block, add the uncond branch and the phi entry.
2617 Builder.SetInsertPoint(NewBB);
2618 Builder.SetCurrentDebugLocation(SI->getDebugLoc());
2619 Builder.CreateBr(SuccBlock);
2620 PHIUse->addIncoming(NewCst, NewBB);
2624 /// SimplifyBranchOnICmpChain - The specified branch is a conditional branch.
2625 /// Check to see if it is branching on an or/and chain of icmp instructions, and
2626 /// fold it into a switch instruction if so.
2627 static bool SimplifyBranchOnICmpChain(BranchInst *BI, const TargetData *TD,
2628 IRBuilder<> &Builder) {
2629 Instruction *Cond = dyn_cast<Instruction>(BI->getCondition());
2630 if (Cond == 0) return false;
2633 // Change br (X == 0 | X == 1), T, F into a switch instruction.
2634 // If this is a bunch of seteq's or'd together, or if it's a bunch of
2635 // 'setne's and'ed together, collect them.
2637 std::vector<ConstantInt*> Values;
2638 bool TrueWhenEqual = true;
2639 Value *ExtraCase = 0;
2640 unsigned UsedICmps = 0;
2642 if (Cond->getOpcode() == Instruction::Or) {
2643 CompVal = GatherConstantCompares(Cond, Values, ExtraCase, TD, true,
2645 } else if (Cond->getOpcode() == Instruction::And) {
2646 CompVal = GatherConstantCompares(Cond, Values, ExtraCase, TD, false,
2648 TrueWhenEqual = false;
2651 // If we didn't have a multiply compared value, fail.
2652 if (CompVal == 0) return false;
2654 // Avoid turning single icmps into a switch.
2658 // There might be duplicate constants in the list, which the switch
2659 // instruction can't handle, remove them now.
2660 array_pod_sort(Values.begin(), Values.end(), ConstantIntSortPredicate);
2661 Values.erase(std::unique(Values.begin(), Values.end()), Values.end());
2663 // If Extra was used, we require at least two switch values to do the
2664 // transformation. A switch with one value is just an cond branch.
2665 if (ExtraCase && Values.size() < 2) return false;
2667 // TODO: Preserve branch weight metadata, similarly to how
2668 // FoldValueComparisonIntoPredecessors preserves it.
2670 // Figure out which block is which destination.
2671 BasicBlock *DefaultBB = BI->getSuccessor(1);
2672 BasicBlock *EdgeBB = BI->getSuccessor(0);
2673 if (!TrueWhenEqual) std::swap(DefaultBB, EdgeBB);
2675 BasicBlock *BB = BI->getParent();
2677 DEBUG(dbgs() << "Converting 'icmp' chain with " << Values.size()
2678 << " cases into SWITCH. BB is:\n" << *BB);
2680 // If there are any extra values that couldn't be folded into the switch
2681 // then we evaluate them with an explicit branch first. Split the block
2682 // right before the condbr to handle it.
2684 BasicBlock *NewBB = BB->splitBasicBlock(BI, "switch.early.test");
2685 // Remove the uncond branch added to the old block.
2686 TerminatorInst *OldTI = BB->getTerminator();
2687 Builder.SetInsertPoint(OldTI);
2690 Builder.CreateCondBr(ExtraCase, EdgeBB, NewBB);
2692 Builder.CreateCondBr(ExtraCase, NewBB, EdgeBB);
2694 OldTI->eraseFromParent();
2696 // If there are PHI nodes in EdgeBB, then we need to add a new entry to them
2697 // for the edge we just added.
2698 AddPredecessorToBlock(EdgeBB, BB, NewBB);
2700 DEBUG(dbgs() << " ** 'icmp' chain unhandled condition: " << *ExtraCase
2701 << "\nEXTRABB = " << *BB);
2705 Builder.SetInsertPoint(BI);
2706 // Convert pointer to int before we switch.
2707 if (CompVal->getType()->isPointerTy()) {
2708 assert(TD && "Cannot switch on pointer without TargetData");
2709 CompVal = Builder.CreatePtrToInt(CompVal,
2710 TD->getIntPtrType(CompVal->getContext()),
2714 // Create the new switch instruction now.
2715 SwitchInst *New = Builder.CreateSwitch(CompVal, DefaultBB, Values.size());
2717 // Add all of the 'cases' to the switch instruction.
2718 for (unsigned i = 0, e = Values.size(); i != e; ++i)
2719 New->addCase(Values[i], EdgeBB);
2721 // We added edges from PI to the EdgeBB. As such, if there were any
2722 // PHI nodes in EdgeBB, they need entries to be added corresponding to
2723 // the number of edges added.
2724 for (BasicBlock::iterator BBI = EdgeBB->begin();
2725 isa<PHINode>(BBI); ++BBI) {
2726 PHINode *PN = cast<PHINode>(BBI);
2727 Value *InVal = PN->getIncomingValueForBlock(BB);
2728 for (unsigned i = 0, e = Values.size()-1; i != e; ++i)
2729 PN->addIncoming(InVal, BB);
2732 // Erase the old branch instruction.
2733 EraseTerminatorInstAndDCECond(BI);
2735 DEBUG(dbgs() << " ** 'icmp' chain result is:\n" << *BB << '\n');
2739 bool SimplifyCFGOpt::SimplifyResume(ResumeInst *RI, IRBuilder<> &Builder) {
2740 // If this is a trivial landing pad that just continues unwinding the caught
2741 // exception then zap the landing pad, turning its invokes into calls.
2742 BasicBlock *BB = RI->getParent();
2743 LandingPadInst *LPInst = dyn_cast<LandingPadInst>(BB->getFirstNonPHI());
2744 if (RI->getValue() != LPInst)
2745 // Not a landing pad, or the resume is not unwinding the exception that
2746 // caused control to branch here.
2749 // Check that there are no other instructions except for debug intrinsics.
2750 BasicBlock::iterator I = LPInst, E = RI;
2752 if (!isa<DbgInfoIntrinsic>(I))
2755 // Turn all invokes that unwind here into calls and delete the basic block.
2756 for (pred_iterator PI = pred_begin(BB), PE = pred_end(BB); PI != PE;) {
2757 InvokeInst *II = cast<InvokeInst>((*PI++)->getTerminator());
2758 SmallVector<Value*, 8> Args(II->op_begin(), II->op_end() - 3);
2759 // Insert a call instruction before the invoke.
2760 CallInst *Call = CallInst::Create(II->getCalledValue(), Args, "", II);
2762 Call->setCallingConv(II->getCallingConv());
2763 Call->setAttributes(II->getAttributes());
2764 Call->setDebugLoc(II->getDebugLoc());
2766 // Anything that used the value produced by the invoke instruction now uses
2767 // the value produced by the call instruction. Note that we do this even
2768 // for void functions and calls with no uses so that the callgraph edge is
2770 II->replaceAllUsesWith(Call);
2771 BB->removePredecessor(II->getParent());
2773 // Insert a branch to the normal destination right before the invoke.
2774 BranchInst::Create(II->getNormalDest(), II);
2776 // Finally, delete the invoke instruction!
2777 II->eraseFromParent();
2780 // The landingpad is now unreachable. Zap it.
2781 BB->eraseFromParent();
2785 bool SimplifyCFGOpt::SimplifyReturn(ReturnInst *RI, IRBuilder<> &Builder) {
2786 BasicBlock *BB = RI->getParent();
2787 if (!BB->getFirstNonPHIOrDbg()->isTerminator()) return false;
2789 // Find predecessors that end with branches.
2790 SmallVector<BasicBlock*, 8> UncondBranchPreds;
2791 SmallVector<BranchInst*, 8> CondBranchPreds;
2792 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
2793 BasicBlock *P = *PI;
2794 TerminatorInst *PTI = P->getTerminator();
2795 if (BranchInst *BI = dyn_cast<BranchInst>(PTI)) {
2796 if (BI->isUnconditional())
2797 UncondBranchPreds.push_back(P);
2799 CondBranchPreds.push_back(BI);
2803 // If we found some, do the transformation!
2804 if (!UncondBranchPreds.empty() && DupRet) {
2805 while (!UncondBranchPreds.empty()) {
2806 BasicBlock *Pred = UncondBranchPreds.pop_back_val();
2807 DEBUG(dbgs() << "FOLDING: " << *BB
2808 << "INTO UNCOND BRANCH PRED: " << *Pred);
2809 (void)FoldReturnIntoUncondBranch(RI, BB, Pred);
2812 // If we eliminated all predecessors of the block, delete the block now.
2813 if (pred_begin(BB) == pred_end(BB))
2814 // We know there are no successors, so just nuke the block.
2815 BB->eraseFromParent();
2820 // Check out all of the conditional branches going to this return
2821 // instruction. If any of them just select between returns, change the
2822 // branch itself into a select/return pair.
2823 while (!CondBranchPreds.empty()) {
2824 BranchInst *BI = CondBranchPreds.pop_back_val();
2826 // Check to see if the non-BB successor is also a return block.
2827 if (isa<ReturnInst>(BI->getSuccessor(0)->getTerminator()) &&
2828 isa<ReturnInst>(BI->getSuccessor(1)->getTerminator()) &&
2829 SimplifyCondBranchToTwoReturns(BI, Builder))
2835 bool SimplifyCFGOpt::SimplifyUnreachable(UnreachableInst *UI) {
2836 BasicBlock *BB = UI->getParent();
2838 bool Changed = false;
2840 // If there are any instructions immediately before the unreachable that can
2841 // be removed, do so.
2842 while (UI != BB->begin()) {
2843 BasicBlock::iterator BBI = UI;
2845 // Do not delete instructions that can have side effects which might cause
2846 // the unreachable to not be reachable; specifically, calls and volatile
2847 // operations may have this effect.
2848 if (isa<CallInst>(BBI) && !isa<DbgInfoIntrinsic>(BBI)) break;
2850 if (BBI->mayHaveSideEffects()) {
2851 if (StoreInst *SI = dyn_cast<StoreInst>(BBI)) {
2852 if (SI->isVolatile())
2854 } else if (LoadInst *LI = dyn_cast<LoadInst>(BBI)) {
2855 if (LI->isVolatile())
2857 } else if (AtomicRMWInst *RMWI = dyn_cast<AtomicRMWInst>(BBI)) {
2858 if (RMWI->isVolatile())
2860 } else if (AtomicCmpXchgInst *CXI = dyn_cast<AtomicCmpXchgInst>(BBI)) {
2861 if (CXI->isVolatile())
2863 } else if (!isa<FenceInst>(BBI) && !isa<VAArgInst>(BBI) &&
2864 !isa<LandingPadInst>(BBI)) {
2867 // Note that deleting LandingPad's here is in fact okay, although it
2868 // involves a bit of subtle reasoning. If this inst is a LandingPad,
2869 // all the predecessors of this block will be the unwind edges of Invokes,
2870 // and we can therefore guarantee this block will be erased.
2873 // Delete this instruction (any uses are guaranteed to be dead)
2874 if (!BBI->use_empty())
2875 BBI->replaceAllUsesWith(UndefValue::get(BBI->getType()));
2876 BBI->eraseFromParent();
2880 // If the unreachable instruction is the first in the block, take a gander
2881 // at all of the predecessors of this instruction, and simplify them.
2882 if (&BB->front() != UI) return Changed;
2884 SmallVector<BasicBlock*, 8> Preds(pred_begin(BB), pred_end(BB));
2885 for (unsigned i = 0, e = Preds.size(); i != e; ++i) {
2886 TerminatorInst *TI = Preds[i]->getTerminator();
2887 IRBuilder<> Builder(TI);
2888 if (BranchInst *BI = dyn_cast<BranchInst>(TI)) {
2889 if (BI->isUnconditional()) {
2890 if (BI->getSuccessor(0) == BB) {
2891 new UnreachableInst(TI->getContext(), TI);
2892 TI->eraseFromParent();
2896 if (BI->getSuccessor(0) == BB) {
2897 Builder.CreateBr(BI->getSuccessor(1));
2898 EraseTerminatorInstAndDCECond(BI);
2899 } else if (BI->getSuccessor(1) == BB) {
2900 Builder.CreateBr(BI->getSuccessor(0));
2901 EraseTerminatorInstAndDCECond(BI);
2905 } else if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) {
2906 for (SwitchInst::CaseIt i = SI->case_begin(), e = SI->case_end();
2908 if (i.getCaseSuccessor() == BB) {
2909 BB->removePredecessor(SI->getParent());
2914 // If the default value is unreachable, figure out the most popular
2915 // destination and make it the default.
2916 if (SI->getDefaultDest() == BB) {
2917 std::map<BasicBlock*, std::pair<unsigned, unsigned> > Popularity;
2918 for (SwitchInst::CaseIt i = SI->case_begin(), e = SI->case_end();
2920 std::pair<unsigned, unsigned> &entry =
2921 Popularity[i.getCaseSuccessor()];
2922 if (entry.first == 0) {
2924 entry.second = i.getCaseIndex();
2930 // Find the most popular block.
2931 unsigned MaxPop = 0;
2932 unsigned MaxIndex = 0;
2933 BasicBlock *MaxBlock = 0;
2934 for (std::map<BasicBlock*, std::pair<unsigned, unsigned> >::iterator
2935 I = Popularity.begin(), E = Popularity.end(); I != E; ++I) {
2936 if (I->second.first > MaxPop ||
2937 (I->second.first == MaxPop && MaxIndex > I->second.second)) {
2938 MaxPop = I->second.first;
2939 MaxIndex = I->second.second;
2940 MaxBlock = I->first;
2944 // Make this the new default, allowing us to delete any explicit
2946 SI->setDefaultDest(MaxBlock);
2949 // If MaxBlock has phinodes in it, remove MaxPop-1 entries from
2951 if (isa<PHINode>(MaxBlock->begin()))
2952 for (unsigned i = 0; i != MaxPop-1; ++i)
2953 MaxBlock->removePredecessor(SI->getParent());
2955 for (SwitchInst::CaseIt i = SI->case_begin(), e = SI->case_end();
2957 if (i.getCaseSuccessor() == MaxBlock) {
2963 } else if (InvokeInst *II = dyn_cast<InvokeInst>(TI)) {
2964 if (II->getUnwindDest() == BB) {
2965 // Convert the invoke to a call instruction. This would be a good
2966 // place to note that the call does not throw though.
2967 BranchInst *BI = Builder.CreateBr(II->getNormalDest());
2968 II->removeFromParent(); // Take out of symbol table
2970 // Insert the call now...
2971 SmallVector<Value*, 8> Args(II->op_begin(), II->op_end()-3);
2972 Builder.SetInsertPoint(BI);
2973 CallInst *CI = Builder.CreateCall(II->getCalledValue(),
2974 Args, II->getName());
2975 CI->setCallingConv(II->getCallingConv());
2976 CI->setAttributes(II->getAttributes());
2977 // If the invoke produced a value, the call does now instead.
2978 II->replaceAllUsesWith(CI);
2985 // If this block is now dead, remove it.
2986 if (pred_begin(BB) == pred_end(BB) &&
2987 BB != &BB->getParent()->getEntryBlock()) {
2988 // We know there are no successors, so just nuke the block.
2989 BB->eraseFromParent();
2996 /// TurnSwitchRangeIntoICmp - Turns a switch with that contains only a
2997 /// integer range comparison into a sub, an icmp and a branch.
2998 static bool TurnSwitchRangeIntoICmp(SwitchInst *SI, IRBuilder<> &Builder) {
2999 assert(SI->getNumCases() > 1 && "Degenerate switch?");
3001 // Make sure all cases point to the same destination and gather the values.
3002 SmallVector<ConstantInt *, 16> Cases;
3003 SwitchInst::CaseIt I = SI->case_begin();
3004 Cases.push_back(I.getCaseValue());
3005 SwitchInst::CaseIt PrevI = I++;
3006 for (SwitchInst::CaseIt E = SI->case_end(); I != E; PrevI = I++) {
3007 if (PrevI.getCaseSuccessor() != I.getCaseSuccessor())
3009 Cases.push_back(I.getCaseValue());
3011 assert(Cases.size() == SI->getNumCases() && "Not all cases gathered");
3013 // Sort the case values, then check if they form a range we can transform.
3014 array_pod_sort(Cases.begin(), Cases.end(), ConstantIntSortPredicate);
3015 for (unsigned I = 1, E = Cases.size(); I != E; ++I) {
3016 if (Cases[I-1]->getValue() != Cases[I]->getValue()+1)
3020 Constant *Offset = ConstantExpr::getNeg(Cases.back());
3021 Constant *NumCases = ConstantInt::get(Offset->getType(), SI->getNumCases());
3023 Value *Sub = SI->getCondition();
3024 if (!Offset->isNullValue())
3025 Sub = Builder.CreateAdd(Sub, Offset, Sub->getName()+".off");
3026 Value *Cmp = Builder.CreateICmpULT(Sub, NumCases, "switch");
3027 BranchInst *NewBI = Builder.CreateCondBr(
3028 Cmp, SI->case_begin().getCaseSuccessor(), SI->getDefaultDest());
3030 // Update weight for the newly-created conditional branch.
3031 SmallVector<uint64_t, 8> Weights;
3032 bool HasWeights = HasBranchWeights(SI);
3034 GetBranchWeights(SI, Weights);
3035 if (Weights.size() == 1 + SI->getNumCases()) {
3036 // Combine all weights for the cases to be the true weight of NewBI.
3037 // We assume that the sum of all weights for a Terminator can fit into 32
3039 uint32_t NewTrueWeight = 0;
3040 for (unsigned I = 1, E = Weights.size(); I != E; ++I)
3041 NewTrueWeight += (uint32_t)Weights[I];
3042 NewBI->setMetadata(LLVMContext::MD_prof,
3043 MDBuilder(SI->getContext()).
3044 createBranchWeights(NewTrueWeight,
3045 (uint32_t)Weights[0]));
3049 // Prune obsolete incoming values off the successor's PHI nodes.
3050 for (BasicBlock::iterator BBI = SI->case_begin().getCaseSuccessor()->begin();
3051 isa<PHINode>(BBI); ++BBI) {
3052 for (unsigned I = 0, E = SI->getNumCases()-1; I != E; ++I)
3053 cast<PHINode>(BBI)->removeIncomingValue(SI->getParent());
3055 SI->eraseFromParent();
3060 /// EliminateDeadSwitchCases - Compute masked bits for the condition of a switch
3061 /// and use it to remove dead cases.
3062 static bool EliminateDeadSwitchCases(SwitchInst *SI) {
3063 Value *Cond = SI->getCondition();
3064 unsigned Bits = cast<IntegerType>(Cond->getType())->getBitWidth();
3065 APInt KnownZero(Bits, 0), KnownOne(Bits, 0);
3066 ComputeMaskedBits(Cond, KnownZero, KnownOne);
3068 // Gather dead cases.
3069 SmallVector<ConstantInt*, 8> DeadCases;
3070 for (SwitchInst::CaseIt I = SI->case_begin(), E = SI->case_end(); I != E; ++I) {
3071 if ((I.getCaseValue()->getValue() & KnownZero) != 0 ||
3072 (I.getCaseValue()->getValue() & KnownOne) != KnownOne) {
3073 DeadCases.push_back(I.getCaseValue());
3074 DEBUG(dbgs() << "SimplifyCFG: switch case '"
3075 << I.getCaseValue() << "' is dead.\n");
3079 SmallVector<uint64_t, 8> Weights;
3080 bool HasWeight = HasBranchWeights(SI);
3082 GetBranchWeights(SI, Weights);
3083 HasWeight = (Weights.size() == 1 + SI->getNumCases());
3086 // Remove dead cases from the switch.
3087 for (unsigned I = 0, E = DeadCases.size(); I != E; ++I) {
3088 SwitchInst::CaseIt Case = SI->findCaseValue(DeadCases[I]);
3089 assert(Case != SI->case_default() &&
3090 "Case was not found. Probably mistake in DeadCases forming.");
3092 std::swap(Weights[Case.getCaseIndex()+1], Weights.back());
3096 // Prune unused values from PHI nodes.
3097 Case.getCaseSuccessor()->removePredecessor(SI->getParent());
3098 SI->removeCase(Case);
3101 SmallVector<uint32_t, 8> MDWeights(Weights.begin(), Weights.end());
3102 SI->setMetadata(LLVMContext::MD_prof,
3103 MDBuilder(SI->getParent()->getContext()).
3104 createBranchWeights(MDWeights));
3107 return !DeadCases.empty();
3110 /// FindPHIForConditionForwarding - If BB would be eligible for simplification
3111 /// by TryToSimplifyUncondBranchFromEmptyBlock (i.e. it is empty and terminated
3112 /// by an unconditional branch), look at the phi node for BB in the successor
3113 /// block and see if the incoming value is equal to CaseValue. If so, return
3114 /// the phi node, and set PhiIndex to BB's index in the phi node.
3115 static PHINode *FindPHIForConditionForwarding(ConstantInt *CaseValue,
3118 if (BB->getFirstNonPHIOrDbg() != BB->getTerminator())
3119 return NULL; // BB must be empty to be a candidate for simplification.
3120 if (!BB->getSinglePredecessor())
3121 return NULL; // BB must be dominated by the switch.
3123 BranchInst *Branch = dyn_cast<BranchInst>(BB->getTerminator());
3124 if (!Branch || !Branch->isUnconditional())
3125 return NULL; // Terminator must be unconditional branch.
3127 BasicBlock *Succ = Branch->getSuccessor(0);
3129 BasicBlock::iterator I = Succ->begin();
3130 while (PHINode *PHI = dyn_cast<PHINode>(I++)) {
3131 int Idx = PHI->getBasicBlockIndex(BB);
3132 assert(Idx >= 0 && "PHI has no entry for predecessor?");
3134 Value *InValue = PHI->getIncomingValue(Idx);
3135 if (InValue != CaseValue) continue;
3144 /// ForwardSwitchConditionToPHI - Try to forward the condition of a switch
3145 /// instruction to a phi node dominated by the switch, if that would mean that
3146 /// some of the destination blocks of the switch can be folded away.
3147 /// Returns true if a change is made.
3148 static bool ForwardSwitchConditionToPHI(SwitchInst *SI) {
3149 typedef DenseMap<PHINode*, SmallVector<int,4> > ForwardingNodesMap;
3150 ForwardingNodesMap ForwardingNodes;
3152 for (SwitchInst::CaseIt I = SI->case_begin(), E = SI->case_end(); I != E; ++I) {
3153 ConstantInt *CaseValue = I.getCaseValue();
3154 BasicBlock *CaseDest = I.getCaseSuccessor();
3157 PHINode *PHI = FindPHIForConditionForwarding(CaseValue, CaseDest,
3161 ForwardingNodes[PHI].push_back(PhiIndex);
3164 bool Changed = false;
3166 for (ForwardingNodesMap::iterator I = ForwardingNodes.begin(),
3167 E = ForwardingNodes.end(); I != E; ++I) {
3168 PHINode *Phi = I->first;
3169 SmallVector<int,4> &Indexes = I->second;
3171 if (Indexes.size() < 2) continue;
3173 for (size_t I = 0, E = Indexes.size(); I != E; ++I)
3174 Phi->setIncomingValue(Indexes[I], SI->getCondition());
3181 /// ValidLookupTableConstant - Return true if the backend will be able to handle
3182 /// initializing an array of constants like C.
3183 static bool ValidLookupTableConstant(Constant *C) {
3184 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C))
3185 return CE->isGEPWithNoNotionalOverIndexing();
3187 return isa<ConstantFP>(C) ||
3188 isa<ConstantInt>(C) ||
3189 isa<ConstantPointerNull>(C) ||
3190 isa<GlobalValue>(C) ||
3194 /// GetCaseResulsts - Try to determine the resulting constant values in phi
3195 /// nodes at the common destination basic block for one of the case
3196 /// destinations of a switch instruction.
3197 static bool GetCaseResults(SwitchInst *SI,
3198 BasicBlock *CaseDest,
3199 BasicBlock **CommonDest,
3200 SmallVector<std::pair<PHINode*,Constant*>, 4> &Res) {
3201 // The block from which we enter the common destination.
3202 BasicBlock *Pred = SI->getParent();
3204 // If CaseDest is empty, continue to its successor.
3205 if (CaseDest->getFirstNonPHIOrDbg() == CaseDest->getTerminator() &&
3206 !isa<PHINode>(CaseDest->begin())) {
3208 TerminatorInst *Terminator = CaseDest->getTerminator();
3209 if (Terminator->getNumSuccessors() != 1)
3213 CaseDest = Terminator->getSuccessor(0);
3216 // If we did not have a CommonDest before, use the current one.
3218 *CommonDest = CaseDest;
3219 // If the destination isn't the common one, abort.
3220 if (CaseDest != *CommonDest)
3223 // Get the values for this case from phi nodes in the destination block.
3224 BasicBlock::iterator I = (*CommonDest)->begin();
3225 while (PHINode *PHI = dyn_cast<PHINode>(I++)) {
3226 int Idx = PHI->getBasicBlockIndex(Pred);
3230 Constant *ConstVal = dyn_cast<Constant>(PHI->getIncomingValue(Idx));
3234 // Be conservative about which kinds of constants we support.
3235 if (!ValidLookupTableConstant(ConstVal))
3238 Res.push_back(std::make_pair(PHI, ConstVal));
3245 /// SwitchLookupTable - This class represents a lookup table that can be used
3246 /// to replace a switch.
3247 class SwitchLookupTable {
3249 /// SwitchLookupTable - Create a lookup table to use as a switch replacement
3250 /// with the contents of Values, using DefaultValue to fill any holes in the
3252 SwitchLookupTable(Module &M,
3254 ConstantInt *Offset,
3255 const SmallVector<std::pair<ConstantInt*, Constant*>, 4>& Values,
3256 Constant *DefaultValue,
3257 const TargetData *TD);
3259 /// BuildLookup - Build instructions with Builder to retrieve the value at
3260 /// the position given by Index in the lookup table.
3261 Value *BuildLookup(Value *Index, IRBuilder<> &Builder);
3263 /// WouldFitInRegister - Return true if a table with TableSize elements of
3264 /// type ElementType would fit in a target-legal register.
3265 static bool WouldFitInRegister(const TargetData *TD,
3267 const Type *ElementType);
3270 // Depending on the contents of the table, it can be represented in
3273 // For tables where each element contains the same value, we just have to
3274 // store that single value and return it for each lookup.
3277 // For small tables with integer elements, we can pack them into a bitmap
3278 // that fits into a target-legal register. Values are retrieved by
3279 // shift and mask operations.
3282 // The table is stored as an array of values. Values are retrieved by load
3283 // instructions from the table.
3287 // For SingleValueKind, this is the single value.
3288 Constant *SingleValue;
3290 // For BitMapKind, this is the bitmap.
3291 ConstantInt *BitMap;
3292 IntegerType *BitMapElementTy;
3294 // For ArrayKind, this is the array.
3295 GlobalVariable *Array;
3299 SwitchLookupTable::SwitchLookupTable(Module &M,
3301 ConstantInt *Offset,
3302 const SmallVector<std::pair<ConstantInt*, Constant*>, 4>& Values,
3303 Constant *DefaultValue,
3304 const TargetData *TD) {
3305 assert(Values.size() && "Can't build lookup table without values!");
3306 assert(TableSize >= Values.size() && "Can't fit values in table!");
3308 // If all values in the table are equal, this is that value.
3309 SingleValue = Values.begin()->second;
3311 // Build up the table contents.
3312 SmallVector<Constant*, 64> TableContents(TableSize);
3313 for (size_t I = 0, E = Values.size(); I != E; ++I) {
3314 ConstantInt *CaseVal = Values[I].first;
3315 Constant *CaseRes = Values[I].second;
3316 assert(CaseRes->getType() == DefaultValue->getType());
3318 uint64_t Idx = (CaseVal->getValue() - Offset->getValue())
3320 TableContents[Idx] = CaseRes;
3322 if (CaseRes != SingleValue)
3326 // Fill in any holes in the table with the default result.
3327 if (Values.size() < TableSize) {
3328 for (uint64_t I = 0; I < TableSize; ++I) {
3329 if (!TableContents[I])
3330 TableContents[I] = DefaultValue;
3333 if (DefaultValue != SingleValue)
3337 // If each element in the table contains the same value, we only need to store
3338 // that single value.
3340 Kind = SingleValueKind;
3344 // If the type is integer and the table fits in a register, build a bitmap.
3345 if (WouldFitInRegister(TD, TableSize, DefaultValue->getType())) {
3346 IntegerType *IT = cast<IntegerType>(DefaultValue->getType());
3347 APInt TableInt(TableSize * IT->getBitWidth(), 0);
3348 for (uint64_t I = TableSize; I > 0; --I) {
3349 TableInt <<= IT->getBitWidth();
3350 ConstantInt *Val = cast<ConstantInt>(TableContents[I - 1]);
3351 TableInt |= Val->getValue().zext(TableInt.getBitWidth());
3353 BitMap = ConstantInt::get(M.getContext(), TableInt);
3354 BitMapElementTy = IT;
3360 // Store the table in an array.
3361 ArrayType *ArrayTy = ArrayType::get(DefaultValue->getType(), TableSize);
3362 Constant *Initializer = ConstantArray::get(ArrayTy, TableContents);
3364 Array = new GlobalVariable(M, ArrayTy, /*constant=*/ true,
3365 GlobalVariable::PrivateLinkage,
3368 Array->setUnnamedAddr(true);
3372 Value *SwitchLookupTable::BuildLookup(Value *Index, IRBuilder<> &Builder) {
3374 case SingleValueKind:
3377 // Type of the bitmap (e.g. i59).
3378 IntegerType *MapTy = BitMap->getType();
3380 // Cast Index to the same type as the bitmap.
3381 // Note: The Index is <= the number of elements in the table, so
3382 // truncating it to the width of the bitmask is safe.
3383 Value *ShiftAmt = Builder.CreateZExtOrTrunc(Index, MapTy, "switch.cast");
3385 // Multiply the shift amount by the element width.
3386 ShiftAmt = Builder.CreateMul(ShiftAmt,
3387 ConstantInt::get(MapTy, BitMapElementTy->getBitWidth()),
3391 Value *DownShifted = Builder.CreateLShr(BitMap, ShiftAmt,
3392 "switch.downshift");
3394 return Builder.CreateTrunc(DownShifted, BitMapElementTy,
3398 Value *GEPIndices[] = { Builder.getInt32(0), Index };
3399 Value *GEP = Builder.CreateInBoundsGEP(Array, GEPIndices,
3401 return Builder.CreateLoad(GEP, "switch.load");
3404 llvm_unreachable("Unknown lookup table kind!");
3407 bool SwitchLookupTable::WouldFitInRegister(const TargetData *TD,
3409 const Type *ElementType) {
3412 const IntegerType *IT = dyn_cast<IntegerType>(ElementType);
3415 // FIXME: If the type is wider than it needs to be, e.g. i8 but all values
3416 // are <= 15, we could try to narrow the type.
3417 return TD->fitsInLegalInteger(TableSize * IT->getBitWidth());
3420 /// ShouldBuildLookupTable - Determine whether a lookup table should be built
3421 /// for this switch, based on the number of caes, size of the table and the
3422 /// types of the results.
3423 static bool ShouldBuildLookupTable(SwitchInst *SI,
3425 const TargetData *TD,
3426 const SmallDenseMap<PHINode*, Type*>& ResultTypes) {
3427 // The table density should be at least 40%. This is the same criterion as for
3428 // jump tables, see SelectionDAGBuilder::handleJTSwitchCase.
3429 // FIXME: Find the best cut-off.
3430 if (SI->getNumCases() > TableSize || TableSize >= UINT64_MAX / 10)
3431 return false; // TableSize overflowed, or mul below might overflow.
3432 if (SI->getNumCases() * 10 >= TableSize * 4)
3435 // If each table would fit in a register, we should build it anyway.
3436 for (SmallDenseMap<PHINode*, Type*>::const_iterator I = ResultTypes.begin(),
3437 E = ResultTypes.end(); I != E; ++I) {
3438 if (!SwitchLookupTable::WouldFitInRegister(TD, TableSize, I->second))
3444 /// SwitchToLookupTable - If the switch is only used to initialize one or more
3445 /// phi nodes in a common successor block with different constant values,
3446 /// replace the switch with lookup tables.
3447 static bool SwitchToLookupTable(SwitchInst *SI,
3448 IRBuilder<> &Builder,
3449 const TargetData* TD) {
3450 assert(SI->getNumCases() > 1 && "Degenerate switch?");
3451 // FIXME: Handle unreachable cases.
3453 // FIXME: If the switch is too sparse for a lookup table, perhaps we could
3454 // split off a dense part and build a lookup table for that.
3456 // FIXME: This creates arrays of GEPs to constant strings, which means each
3457 // GEP needs a runtime relocation in PIC code. We should just build one big
3458 // string and lookup indices into that.
3460 // Ignore the switch if the number of cases is too small.
3461 // This is similar to the check when building jump tables in
3462 // SelectionDAGBuilder::handleJTSwitchCase.
3463 // FIXME: Determine the best cut-off.
3464 if (SI->getNumCases() < 4)
3467 // Figure out the corresponding result for each case value and phi node in the
3468 // common destination, as well as the the min and max case values.
3469 assert(SI->case_begin() != SI->case_end());
3470 SwitchInst::CaseIt CI = SI->case_begin();
3471 ConstantInt *MinCaseVal = CI.getCaseValue();
3472 ConstantInt *MaxCaseVal = CI.getCaseValue();
3474 BasicBlock *CommonDest = NULL;
3475 typedef SmallVector<std::pair<ConstantInt*, Constant*>, 4> ResultListTy;
3476 SmallDenseMap<PHINode*, ResultListTy> ResultLists;
3477 SmallDenseMap<PHINode*, Constant*> DefaultResults;
3478 SmallDenseMap<PHINode*, Type*> ResultTypes;
3479 SmallVector<PHINode*, 4> PHIs;
3481 for (SwitchInst::CaseIt E = SI->case_end(); CI != E; ++CI) {
3482 ConstantInt *CaseVal = CI.getCaseValue();
3483 if (CaseVal->getValue().slt(MinCaseVal->getValue()))
3484 MinCaseVal = CaseVal;
3485 if (CaseVal->getValue().sgt(MaxCaseVal->getValue()))
3486 MaxCaseVal = CaseVal;
3488 // Resulting value at phi nodes for this case value.
3489 typedef SmallVector<std::pair<PHINode*, Constant*>, 4> ResultsTy;
3491 if (!GetCaseResults(SI, CI.getCaseSuccessor(), &CommonDest, Results))
3494 // Append the result from this case to the list for each phi.
3495 for (ResultsTy::iterator I = Results.begin(), E = Results.end(); I!=E; ++I) {
3496 if (!ResultLists.count(I->first))
3497 PHIs.push_back(I->first);
3498 ResultLists[I->first].push_back(std::make_pair(CaseVal, I->second));
3502 // Get the resulting values for the default case.
3503 SmallVector<std::pair<PHINode*, Constant*>, 4> DefaultResultsList;
3504 if (!GetCaseResults(SI, SI->getDefaultDest(), &CommonDest, DefaultResultsList))
3506 for (size_t I = 0, E = DefaultResultsList.size(); I != E; ++I) {
3507 PHINode *PHI = DefaultResultsList[I].first;
3508 Constant *Result = DefaultResultsList[I].second;
3509 DefaultResults[PHI] = Result;
3510 ResultTypes[PHI] = Result->getType();
3513 APInt RangeSpread = MaxCaseVal->getValue() - MinCaseVal->getValue();
3514 uint64_t TableSize = RangeSpread.getLimitedValue() + 1;
3515 if (!ShouldBuildLookupTable(SI, TableSize, TD, ResultTypes))
3518 // Create the BB that does the lookups.
3519 Module &Mod = *CommonDest->getParent()->getParent();
3520 BasicBlock *LookupBB = BasicBlock::Create(Mod.getContext(),
3522 CommonDest->getParent(),
3525 // Check whether the condition value is within the case range, and branch to
3527 Builder.SetInsertPoint(SI);
3528 Value *TableIndex = Builder.CreateSub(SI->getCondition(), MinCaseVal,
3530 Value *Cmp = Builder.CreateICmpULT(TableIndex, ConstantInt::get(
3531 MinCaseVal->getType(), TableSize));
3532 Builder.CreateCondBr(Cmp, LookupBB, SI->getDefaultDest());
3534 // Populate the BB that does the lookups.
3535 Builder.SetInsertPoint(LookupBB);
3536 bool ReturnedEarly = false;
3537 for (size_t I = 0, E = PHIs.size(); I != E; ++I) {
3538 PHINode *PHI = PHIs[I];
3540 SwitchLookupTable Table(Mod, TableSize, MinCaseVal, ResultLists[PHI],
3541 DefaultResults[PHI], TD);
3543 Value *Result = Table.BuildLookup(TableIndex, Builder);
3545 // If the result is used to return immediately from the function, we want to
3546 // do that right here.
3547 if (PHI->hasOneUse() && isa<ReturnInst>(*PHI->use_begin()) &&
3548 *PHI->use_begin() == CommonDest->getFirstNonPHIOrDbg()) {
3549 Builder.CreateRet(Result);
3550 ReturnedEarly = true;
3554 PHI->addIncoming(Result, LookupBB);
3558 Builder.CreateBr(CommonDest);
3560 // Remove the switch.
3561 for (unsigned i = 0; i < SI->getNumSuccessors(); ++i) {
3562 BasicBlock *Succ = SI->getSuccessor(i);
3563 if (Succ == SI->getDefaultDest()) continue;
3564 Succ->removePredecessor(SI->getParent());
3566 SI->eraseFromParent();
3572 bool SimplifyCFGOpt::SimplifySwitch(SwitchInst *SI, IRBuilder<> &Builder) {
3573 // If this switch is too complex to want to look at, ignore it.
3574 if (!isValueEqualityComparison(SI))
3577 BasicBlock *BB = SI->getParent();
3579 // If we only have one predecessor, and if it is a branch on this value,
3580 // see if that predecessor totally determines the outcome of this switch.
3581 if (BasicBlock *OnlyPred = BB->getSinglePredecessor())
3582 if (SimplifyEqualityComparisonWithOnlyPredecessor(SI, OnlyPred, Builder))
3583 return SimplifyCFG(BB) | true;
3585 Value *Cond = SI->getCondition();
3586 if (SelectInst *Select = dyn_cast<SelectInst>(Cond))
3587 if (SimplifySwitchOnSelect(SI, Select))
3588 return SimplifyCFG(BB) | true;
3590 // If the block only contains the switch, see if we can fold the block
3591 // away into any preds.
3592 BasicBlock::iterator BBI = BB->begin();
3593 // Ignore dbg intrinsics.
3594 while (isa<DbgInfoIntrinsic>(BBI))
3597 if (FoldValueComparisonIntoPredecessors(SI, Builder))
3598 return SimplifyCFG(BB) | true;
3600 // Try to transform the switch into an icmp and a branch.
3601 if (TurnSwitchRangeIntoICmp(SI, Builder))
3602 return SimplifyCFG(BB) | true;
3604 // Remove unreachable cases.
3605 if (EliminateDeadSwitchCases(SI))
3606 return SimplifyCFG(BB) | true;
3608 if (ForwardSwitchConditionToPHI(SI))
3609 return SimplifyCFG(BB) | true;
3611 if (SwitchToLookupTable(SI, Builder, TD))
3612 return SimplifyCFG(BB) | true;
3617 bool SimplifyCFGOpt::SimplifyIndirectBr(IndirectBrInst *IBI) {
3618 BasicBlock *BB = IBI->getParent();
3619 bool Changed = false;
3621 // Eliminate redundant destinations.
3622 SmallPtrSet<Value *, 8> Succs;
3623 for (unsigned i = 0, e = IBI->getNumDestinations(); i != e; ++i) {
3624 BasicBlock *Dest = IBI->getDestination(i);
3625 if (!Dest->hasAddressTaken() || !Succs.insert(Dest)) {
3626 Dest->removePredecessor(BB);
3627 IBI->removeDestination(i);
3633 if (IBI->getNumDestinations() == 0) {
3634 // If the indirectbr has no successors, change it to unreachable.
3635 new UnreachableInst(IBI->getContext(), IBI);
3636 EraseTerminatorInstAndDCECond(IBI);
3640 if (IBI->getNumDestinations() == 1) {
3641 // If the indirectbr has one successor, change it to a direct branch.
3642 BranchInst::Create(IBI->getDestination(0), IBI);
3643 EraseTerminatorInstAndDCECond(IBI);
3647 if (SelectInst *SI = dyn_cast<SelectInst>(IBI->getAddress())) {
3648 if (SimplifyIndirectBrOnSelect(IBI, SI))
3649 return SimplifyCFG(BB) | true;
3654 bool SimplifyCFGOpt::SimplifyUncondBranch(BranchInst *BI, IRBuilder<> &Builder){
3655 BasicBlock *BB = BI->getParent();
3657 if (SinkCommon && SinkThenElseCodeToEnd(BI))
3660 // If the Terminator is the only non-phi instruction, simplify the block.
3661 BasicBlock::iterator I = BB->getFirstNonPHIOrDbgOrLifetime();
3662 if (I->isTerminator() && BB != &BB->getParent()->getEntryBlock() &&
3663 TryToSimplifyUncondBranchFromEmptyBlock(BB))
3666 // If the only instruction in the block is a seteq/setne comparison
3667 // against a constant, try to simplify the block.
3668 if (ICmpInst *ICI = dyn_cast<ICmpInst>(I))
3669 if (ICI->isEquality() && isa<ConstantInt>(ICI->getOperand(1))) {
3670 for (++I; isa<DbgInfoIntrinsic>(I); ++I)
3672 if (I->isTerminator() &&
3673 TryToSimplifyUncondBranchWithICmpInIt(ICI, TD, Builder))
3677 // If this basic block is ONLY a compare and a branch, and if a predecessor
3678 // branches to us and our successor, fold the comparison into the
3679 // predecessor and use logical operations to update the incoming value
3680 // for PHI nodes in common successor.
3681 if (FoldBranchToCommonDest(BI))
3682 return SimplifyCFG(BB) | true;
3687 bool SimplifyCFGOpt::SimplifyCondBranch(BranchInst *BI, IRBuilder<> &Builder) {
3688 BasicBlock *BB = BI->getParent();
3690 // Conditional branch
3691 if (isValueEqualityComparison(BI)) {
3692 // If we only have one predecessor, and if it is a branch on this value,
3693 // see if that predecessor totally determines the outcome of this
3695 if (BasicBlock *OnlyPred = BB->getSinglePredecessor())
3696 if (SimplifyEqualityComparisonWithOnlyPredecessor(BI, OnlyPred, Builder))
3697 return SimplifyCFG(BB) | true;
3699 // This block must be empty, except for the setcond inst, if it exists.
3700 // Ignore dbg intrinsics.
3701 BasicBlock::iterator I = BB->begin();
3702 // Ignore dbg intrinsics.
3703 while (isa<DbgInfoIntrinsic>(I))
3706 if (FoldValueComparisonIntoPredecessors(BI, Builder))
3707 return SimplifyCFG(BB) | true;
3708 } else if (&*I == cast<Instruction>(BI->getCondition())){
3710 // Ignore dbg intrinsics.
3711 while (isa<DbgInfoIntrinsic>(I))
3713 if (&*I == BI && FoldValueComparisonIntoPredecessors(BI, Builder))
3714 return SimplifyCFG(BB) | true;
3718 // Try to turn "br (X == 0 | X == 1), T, F" into a switch instruction.
3719 if (SimplifyBranchOnICmpChain(BI, TD, Builder))
3722 // If this basic block is ONLY a compare and a branch, and if a predecessor
3723 // branches to us and one of our successors, fold the comparison into the
3724 // predecessor and use logical operations to pick the right destination.
3725 if (FoldBranchToCommonDest(BI))
3726 return SimplifyCFG(BB) | true;
3728 // We have a conditional branch to two blocks that are only reachable
3729 // from BI. We know that the condbr dominates the two blocks, so see if
3730 // there is any identical code in the "then" and "else" blocks. If so, we
3731 // can hoist it up to the branching block.
3732 if (BI->getSuccessor(0)->getSinglePredecessor() != 0) {
3733 if (BI->getSuccessor(1)->getSinglePredecessor() != 0) {
3734 if (HoistThenElseCodeToIf(BI))
3735 return SimplifyCFG(BB) | true;
3737 // If Successor #1 has multiple preds, we may be able to conditionally
3738 // execute Successor #0 if it branches to successor #1.
3739 TerminatorInst *Succ0TI = BI->getSuccessor(0)->getTerminator();
3740 if (Succ0TI->getNumSuccessors() == 1 &&
3741 Succ0TI->getSuccessor(0) == BI->getSuccessor(1))
3742 if (SpeculativelyExecuteBB(BI, BI->getSuccessor(0)))
3743 return SimplifyCFG(BB) | true;
3745 } else if (BI->getSuccessor(1)->getSinglePredecessor() != 0) {
3746 // If Successor #0 has multiple preds, we may be able to conditionally
3747 // execute Successor #1 if it branches to successor #0.
3748 TerminatorInst *Succ1TI = BI->getSuccessor(1)->getTerminator();
3749 if (Succ1TI->getNumSuccessors() == 1 &&
3750 Succ1TI->getSuccessor(0) == BI->getSuccessor(0))
3751 if (SpeculativelyExecuteBB(BI, BI->getSuccessor(1)))
3752 return SimplifyCFG(BB) | true;
3755 // If this is a branch on a phi node in the current block, thread control
3756 // through this block if any PHI node entries are constants.
3757 if (PHINode *PN = dyn_cast<PHINode>(BI->getCondition()))
3758 if (PN->getParent() == BI->getParent())
3759 if (FoldCondBranchOnPHI(BI, TD))
3760 return SimplifyCFG(BB) | true;
3762 // Scan predecessor blocks for conditional branches.
3763 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI)
3764 if (BranchInst *PBI = dyn_cast<BranchInst>((*PI)->getTerminator()))
3765 if (PBI != BI && PBI->isConditional())
3766 if (SimplifyCondBranchToCondBranch(PBI, BI))
3767 return SimplifyCFG(BB) | true;
3772 /// Check if passing a value to an instruction will cause undefined behavior.
3773 static bool passingValueIsAlwaysUndefined(Value *V, Instruction *I) {
3774 Constant *C = dyn_cast<Constant>(V);
3778 if (!I->hasOneUse()) // Only look at single-use instructions, for compile time
3781 if (C->isNullValue()) {
3782 Instruction *Use = I->use_back();
3784 // Now make sure that there are no instructions in between that can alter
3785 // control flow (eg. calls)
3786 for (BasicBlock::iterator i = ++BasicBlock::iterator(I); &*i != Use; ++i)
3787 if (i == I->getParent()->end() || i->mayHaveSideEffects())
3790 // Look through GEPs. A load from a GEP derived from NULL is still undefined
3791 if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(Use))
3792 if (GEP->getPointerOperand() == I)
3793 return passingValueIsAlwaysUndefined(V, GEP);
3795 // Look through bitcasts.
3796 if (BitCastInst *BC = dyn_cast<BitCastInst>(Use))
3797 return passingValueIsAlwaysUndefined(V, BC);
3799 // Load from null is undefined.
3800 if (LoadInst *LI = dyn_cast<LoadInst>(Use))
3801 return LI->getPointerAddressSpace() == 0;
3803 // Store to null is undefined.
3804 if (StoreInst *SI = dyn_cast<StoreInst>(Use))
3805 return SI->getPointerAddressSpace() == 0 && SI->getPointerOperand() == I;
3810 /// If BB has an incoming value that will always trigger undefined behavior
3811 /// (eg. null pointer dereference), remove the branch leading here.
3812 static bool removeUndefIntroducingPredecessor(BasicBlock *BB) {
3813 for (BasicBlock::iterator i = BB->begin();
3814 PHINode *PHI = dyn_cast<PHINode>(i); ++i)
3815 for (unsigned i = 0, e = PHI->getNumIncomingValues(); i != e; ++i)
3816 if (passingValueIsAlwaysUndefined(PHI->getIncomingValue(i), PHI)) {
3817 TerminatorInst *T = PHI->getIncomingBlock(i)->getTerminator();
3818 IRBuilder<> Builder(T);
3819 if (BranchInst *BI = dyn_cast<BranchInst>(T)) {
3820 BB->removePredecessor(PHI->getIncomingBlock(i));
3821 // Turn uncoditional branches into unreachables and remove the dead
3822 // destination from conditional branches.
3823 if (BI->isUnconditional())
3824 Builder.CreateUnreachable();
3826 Builder.CreateBr(BI->getSuccessor(0) == BB ? BI->getSuccessor(1) :
3827 BI->getSuccessor(0));
3828 BI->eraseFromParent();
3831 // TODO: SwitchInst.
3837 bool SimplifyCFGOpt::run(BasicBlock *BB) {
3838 bool Changed = false;
3840 assert(BB && BB->getParent() && "Block not embedded in function!");
3841 assert(BB->getTerminator() && "Degenerate basic block encountered!");
3843 // Remove basic blocks that have no predecessors (except the entry block)...
3844 // or that just have themself as a predecessor. These are unreachable.
3845 if ((pred_begin(BB) == pred_end(BB) &&
3846 BB != &BB->getParent()->getEntryBlock()) ||
3847 BB->getSinglePredecessor() == BB) {
3848 DEBUG(dbgs() << "Removing BB: \n" << *BB);
3849 DeleteDeadBlock(BB);
3853 // Check to see if we can constant propagate this terminator instruction
3855 Changed |= ConstantFoldTerminator(BB, true);
3857 // Check for and eliminate duplicate PHI nodes in this block.
3858 Changed |= EliminateDuplicatePHINodes(BB);
3860 // Check for and remove branches that will always cause undefined behavior.
3861 Changed |= removeUndefIntroducingPredecessor(BB);
3863 // Merge basic blocks into their predecessor if there is only one distinct
3864 // pred, and if there is only one distinct successor of the predecessor, and
3865 // if there are no PHI nodes.
3867 if (MergeBlockIntoPredecessor(BB))
3870 IRBuilder<> Builder(BB);
3872 // If there is a trivial two-entry PHI node in this basic block, and we can
3873 // eliminate it, do so now.
3874 if (PHINode *PN = dyn_cast<PHINode>(BB->begin()))
3875 if (PN->getNumIncomingValues() == 2)
3876 Changed |= FoldTwoEntryPHINode(PN, TD);
3878 Builder.SetInsertPoint(BB->getTerminator());
3879 if (BranchInst *BI = dyn_cast<BranchInst>(BB->getTerminator())) {
3880 if (BI->isUnconditional()) {
3881 if (SimplifyUncondBranch(BI, Builder)) return true;
3883 if (SimplifyCondBranch(BI, Builder)) return true;
3885 } else if (ReturnInst *RI = dyn_cast<ReturnInst>(BB->getTerminator())) {
3886 if (SimplifyReturn(RI, Builder)) return true;
3887 } else if (ResumeInst *RI = dyn_cast<ResumeInst>(BB->getTerminator())) {
3888 if (SimplifyResume(RI, Builder)) return true;
3889 } else if (SwitchInst *SI = dyn_cast<SwitchInst>(BB->getTerminator())) {
3890 if (SimplifySwitch(SI, Builder)) return true;
3891 } else if (UnreachableInst *UI =
3892 dyn_cast<UnreachableInst>(BB->getTerminator())) {
3893 if (SimplifyUnreachable(UI)) return true;
3894 } else if (IndirectBrInst *IBI =
3895 dyn_cast<IndirectBrInst>(BB->getTerminator())) {
3896 if (SimplifyIndirectBr(IBI)) return true;
3902 /// SimplifyCFG - This function is used to do simplification of a CFG. For
3903 /// example, it adjusts branches to branches to eliminate the extra hop, it
3904 /// eliminates unreachable basic blocks, and does other "peephole" optimization
3905 /// of the CFG. It returns true if a modification was made.
3907 bool llvm::SimplifyCFG(BasicBlock *BB, const TargetData *TD) {
3908 return SimplifyCFGOpt(TD).run(BB);