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/DataLayout.h"
18 #include "llvm/DerivedTypes.h"
19 #include "llvm/GlobalVariable.h"
20 #include "llvm/IRBuilder.h"
21 #include "llvm/Instructions.h"
22 #include "llvm/IntrinsicInst.h"
23 #include "llvm/LLVMContext.h"
24 #include "llvm/MDBuilder.h"
25 #include "llvm/Metadata.h"
26 #include "llvm/Module.h"
27 #include "llvm/Operator.h"
28 #include "llvm/Type.h"
29 #include "llvm/ADT/DenseMap.h"
30 #include "llvm/ADT/STLExtras.h"
31 #include "llvm/ADT/SetVector.h"
32 #include "llvm/ADT/SmallPtrSet.h"
33 #include "llvm/ADT/SmallVector.h"
34 #include "llvm/ADT/Statistic.h"
35 #include "llvm/Analysis/InstructionSimplify.h"
36 #include "llvm/Analysis/ValueTracking.h"
37 #include "llvm/Support/CFG.h"
38 #include "llvm/Support/CommandLine.h"
39 #include "llvm/Support/ConstantRange.h"
40 #include "llvm/Support/Debug.h"
41 #include "llvm/Support/NoFolder.h"
42 #include "llvm/Support/raw_ostream.h"
43 #include "llvm/TargetTransformInfo.h"
44 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
50 static cl::opt<unsigned>
51 PHINodeFoldingThreshold("phi-node-folding-threshold", cl::Hidden, cl::init(1),
52 cl::desc("Control the amount of phi node folding to perform (default = 1)"));
55 DupRet("simplifycfg-dup-ret", cl::Hidden, cl::init(false),
56 cl::desc("Duplicate return instructions into unconditional branches"));
59 SinkCommon("simplifycfg-sink-common", cl::Hidden, cl::init(true),
60 cl::desc("Sink common instructions down to the end block"));
62 STATISTIC(NumBitMaps, "Number of switch instructions turned into bitmaps");
63 STATISTIC(NumLookupTables, "Number of switch instructions turned into lookup tables");
64 STATISTIC(NumSinkCommons, "Number of common instructions sunk down to the end block");
65 STATISTIC(NumSpeculations, "Number of speculative executed instructions");
68 /// ValueEqualityComparisonCase - Represents a case of a switch.
69 struct ValueEqualityComparisonCase {
73 ValueEqualityComparisonCase(ConstantInt *Value, BasicBlock *Dest)
74 : Value(Value), Dest(Dest) {}
76 bool operator<(ValueEqualityComparisonCase RHS) const {
77 // Comparing pointers is ok as we only rely on the order for uniquing.
78 return Value < RHS.Value;
81 bool operator==(BasicBlock *RHSDest) const { return Dest == RHSDest; }
84 class SimplifyCFGOpt {
85 const DataLayout *const TD;
86 const TargetTransformInfo *const TTI;
88 Value *isValueEqualityComparison(TerminatorInst *TI);
89 BasicBlock *GetValueEqualityComparisonCases(TerminatorInst *TI,
90 std::vector<ValueEqualityComparisonCase> &Cases);
91 bool SimplifyEqualityComparisonWithOnlyPredecessor(TerminatorInst *TI,
93 IRBuilder<> &Builder);
94 bool FoldValueComparisonIntoPredecessors(TerminatorInst *TI,
95 IRBuilder<> &Builder);
97 bool SimplifyReturn(ReturnInst *RI, IRBuilder<> &Builder);
98 bool SimplifyResume(ResumeInst *RI, IRBuilder<> &Builder);
99 bool SimplifyUnreachable(UnreachableInst *UI);
100 bool SimplifySwitch(SwitchInst *SI, IRBuilder<> &Builder);
101 bool SimplifyIndirectBr(IndirectBrInst *IBI);
102 bool SimplifyUncondBranch(BranchInst *BI, IRBuilder <> &Builder);
103 bool SimplifyCondBranch(BranchInst *BI, IRBuilder <>&Builder);
106 SimplifyCFGOpt(const DataLayout *td, const TargetTransformInfo *tti)
107 : TD(td), TTI(tti) {}
108 bool run(BasicBlock *BB);
112 /// SafeToMergeTerminators - Return true if it is safe to merge these two
113 /// terminator instructions together.
115 static bool SafeToMergeTerminators(TerminatorInst *SI1, TerminatorInst *SI2) {
116 if (SI1 == SI2) return false; // Can't merge with self!
118 // It is not safe to merge these two switch instructions if they have a common
119 // successor, and if that successor has a PHI node, and if *that* PHI node has
120 // conflicting incoming values from the two switch blocks.
121 BasicBlock *SI1BB = SI1->getParent();
122 BasicBlock *SI2BB = SI2->getParent();
123 SmallPtrSet<BasicBlock*, 16> SI1Succs(succ_begin(SI1BB), succ_end(SI1BB));
125 for (succ_iterator I = succ_begin(SI2BB), E = succ_end(SI2BB); I != E; ++I)
126 if (SI1Succs.count(*I))
127 for (BasicBlock::iterator BBI = (*I)->begin();
128 isa<PHINode>(BBI); ++BBI) {
129 PHINode *PN = cast<PHINode>(BBI);
130 if (PN->getIncomingValueForBlock(SI1BB) !=
131 PN->getIncomingValueForBlock(SI2BB))
138 /// isProfitableToFoldUnconditional - Return true if it is safe and profitable
139 /// to merge these two terminator instructions together, where SI1 is an
140 /// unconditional branch. PhiNodes will store all PHI nodes in common
143 static bool isProfitableToFoldUnconditional(BranchInst *SI1,
146 SmallVectorImpl<PHINode*> &PhiNodes) {
147 if (SI1 == SI2) return false; // Can't merge with self!
148 assert(SI1->isUnconditional() && SI2->isConditional());
150 // We fold the unconditional branch if we can easily update all PHI nodes in
151 // common successors:
152 // 1> We have a constant incoming value for the conditional branch;
153 // 2> We have "Cond" as the incoming value for the unconditional branch;
154 // 3> SI2->getCondition() and Cond have same operands.
155 CmpInst *Ci2 = dyn_cast<CmpInst>(SI2->getCondition());
156 if (!Ci2) return false;
157 if (!(Cond->getOperand(0) == Ci2->getOperand(0) &&
158 Cond->getOperand(1) == Ci2->getOperand(1)) &&
159 !(Cond->getOperand(0) == Ci2->getOperand(1) &&
160 Cond->getOperand(1) == Ci2->getOperand(0)))
163 BasicBlock *SI1BB = SI1->getParent();
164 BasicBlock *SI2BB = SI2->getParent();
165 SmallPtrSet<BasicBlock*, 16> SI1Succs(succ_begin(SI1BB), succ_end(SI1BB));
166 for (succ_iterator I = succ_begin(SI2BB), E = succ_end(SI2BB); I != E; ++I)
167 if (SI1Succs.count(*I))
168 for (BasicBlock::iterator BBI = (*I)->begin();
169 isa<PHINode>(BBI); ++BBI) {
170 PHINode *PN = cast<PHINode>(BBI);
171 if (PN->getIncomingValueForBlock(SI1BB) != Cond ||
172 !isa<ConstantInt>(PN->getIncomingValueForBlock(SI2BB)))
174 PhiNodes.push_back(PN);
179 /// AddPredecessorToBlock - Update PHI nodes in Succ to indicate that there will
180 /// now be entries in it from the 'NewPred' block. The values that will be
181 /// flowing into the PHI nodes will be the same as those coming in from
182 /// ExistPred, an existing predecessor of Succ.
183 static void AddPredecessorToBlock(BasicBlock *Succ, BasicBlock *NewPred,
184 BasicBlock *ExistPred) {
185 if (!isa<PHINode>(Succ->begin())) return; // Quick exit if nothing to do
188 for (BasicBlock::iterator I = Succ->begin();
189 (PN = dyn_cast<PHINode>(I)); ++I)
190 PN->addIncoming(PN->getIncomingValueForBlock(ExistPred), NewPred);
194 /// GetIfCondition - Given a basic block (BB) with two predecessors (and at
195 /// least one PHI node in it), check to see if the merge at this block is due
196 /// to an "if condition". If so, return the boolean condition that determines
197 /// which entry into BB will be taken. Also, return by references the block
198 /// that will be entered from if the condition is true, and the block that will
199 /// be entered if the condition is false.
201 /// This does no checking to see if the true/false blocks have large or unsavory
202 /// instructions in them.
203 static Value *GetIfCondition(BasicBlock *BB, BasicBlock *&IfTrue,
204 BasicBlock *&IfFalse) {
205 PHINode *SomePHI = cast<PHINode>(BB->begin());
206 assert(SomePHI->getNumIncomingValues() == 2 &&
207 "Function can only handle blocks with 2 predecessors!");
208 BasicBlock *Pred1 = SomePHI->getIncomingBlock(0);
209 BasicBlock *Pred2 = SomePHI->getIncomingBlock(1);
211 // We can only handle branches. Other control flow will be lowered to
212 // branches if possible anyway.
213 BranchInst *Pred1Br = dyn_cast<BranchInst>(Pred1->getTerminator());
214 BranchInst *Pred2Br = dyn_cast<BranchInst>(Pred2->getTerminator());
215 if (Pred1Br == 0 || Pred2Br == 0)
218 // Eliminate code duplication by ensuring that Pred1Br is conditional if
220 if (Pred2Br->isConditional()) {
221 // If both branches are conditional, we don't have an "if statement". In
222 // reality, we could transform this case, but since the condition will be
223 // required anyway, we stand no chance of eliminating it, so the xform is
224 // probably not profitable.
225 if (Pred1Br->isConditional())
228 std::swap(Pred1, Pred2);
229 std::swap(Pred1Br, Pred2Br);
232 if (Pred1Br->isConditional()) {
233 // The only thing we have to watch out for here is to make sure that Pred2
234 // doesn't have incoming edges from other blocks. If it does, the condition
235 // doesn't dominate BB.
236 if (Pred2->getSinglePredecessor() == 0)
239 // If we found a conditional branch predecessor, make sure that it branches
240 // to BB and Pred2Br. If it doesn't, this isn't an "if statement".
241 if (Pred1Br->getSuccessor(0) == BB &&
242 Pred1Br->getSuccessor(1) == Pred2) {
245 } else if (Pred1Br->getSuccessor(0) == Pred2 &&
246 Pred1Br->getSuccessor(1) == BB) {
250 // We know that one arm of the conditional goes to BB, so the other must
251 // go somewhere unrelated, and this must not be an "if statement".
255 return Pred1Br->getCondition();
258 // Ok, if we got here, both predecessors end with an unconditional branch to
259 // BB. Don't panic! If both blocks only have a single (identical)
260 // predecessor, and THAT is a conditional branch, then we're all ok!
261 BasicBlock *CommonPred = Pred1->getSinglePredecessor();
262 if (CommonPred == 0 || CommonPred != Pred2->getSinglePredecessor())
265 // Otherwise, if this is a conditional branch, then we can use it!
266 BranchInst *BI = dyn_cast<BranchInst>(CommonPred->getTerminator());
267 if (BI == 0) return 0;
269 assert(BI->isConditional() && "Two successors but not conditional?");
270 if (BI->getSuccessor(0) == Pred1) {
277 return BI->getCondition();
280 /// ComputeSpeculuationCost - Compute an abstract "cost" of speculating the
281 /// given instruction, which is assumed to be safe to speculate. 1 means
282 /// cheap, 2 means less cheap, and UINT_MAX means prohibitively expensive.
283 static unsigned ComputeSpeculationCost(const User *I) {
284 assert(isSafeToSpeculativelyExecute(I) &&
285 "Instruction is not safe to speculatively execute!");
286 switch (Operator::getOpcode(I)) {
288 // In doubt, be conservative.
290 case Instruction::GetElementPtr:
291 // GEPs are cheap if all indices are constant.
292 if (!cast<GEPOperator>(I)->hasAllConstantIndices())
295 case Instruction::Load:
296 case Instruction::Add:
297 case Instruction::Sub:
298 case Instruction::And:
299 case Instruction::Or:
300 case Instruction::Xor:
301 case Instruction::Shl:
302 case Instruction::LShr:
303 case Instruction::AShr:
304 case Instruction::ICmp:
305 case Instruction::Trunc:
306 case Instruction::ZExt:
307 case Instruction::SExt:
308 return 1; // These are all cheap.
310 case Instruction::Call:
311 case Instruction::Select:
316 /// DominatesMergePoint - If we have a merge point of an "if condition" as
317 /// accepted above, return true if the specified value dominates the block. We
318 /// don't handle the true generality of domination here, just a special case
319 /// which works well enough for us.
321 /// If AggressiveInsts is non-null, and if V does not dominate BB, we check to
322 /// see if V (which must be an instruction) and its recursive operands
323 /// that do not dominate BB have a combined cost lower than CostRemaining and
324 /// are non-trapping. If both are true, the instruction is inserted into the
325 /// set and true is returned.
327 /// The cost for most non-trapping instructions is defined as 1 except for
328 /// Select whose cost is 2.
330 /// After this function returns, CostRemaining is decreased by the cost of
331 /// V plus its non-dominating operands. If that cost is greater than
332 /// CostRemaining, false is returned and CostRemaining is undefined.
333 static bool DominatesMergePoint(Value *V, BasicBlock *BB,
334 SmallPtrSet<Instruction*, 4> *AggressiveInsts,
335 unsigned &CostRemaining) {
336 Instruction *I = dyn_cast<Instruction>(V);
338 // Non-instructions all dominate instructions, but not all constantexprs
339 // can be executed unconditionally.
340 if (ConstantExpr *C = dyn_cast<ConstantExpr>(V))
345 BasicBlock *PBB = I->getParent();
347 // We don't want to allow weird loops that might have the "if condition" in
348 // the bottom of this block.
349 if (PBB == BB) return false;
351 // If this instruction is defined in a block that contains an unconditional
352 // branch to BB, then it must be in the 'conditional' part of the "if
353 // statement". If not, it definitely dominates the region.
354 BranchInst *BI = dyn_cast<BranchInst>(PBB->getTerminator());
355 if (BI == 0 || BI->isConditional() || BI->getSuccessor(0) != BB)
358 // If we aren't allowing aggressive promotion anymore, then don't consider
359 // instructions in the 'if region'.
360 if (AggressiveInsts == 0) return false;
362 // If we have seen this instruction before, don't count it again.
363 if (AggressiveInsts->count(I)) return true;
365 // Okay, it looks like the instruction IS in the "condition". Check to
366 // see if it's a cheap instruction to unconditionally compute, and if it
367 // only uses stuff defined outside of the condition. If so, hoist it out.
368 if (!isSafeToSpeculativelyExecute(I))
371 unsigned Cost = ComputeSpeculationCost(I);
373 if (Cost > CostRemaining)
376 CostRemaining -= Cost;
378 // Okay, we can only really hoist these out if their operands do
379 // not take us over the cost threshold.
380 for (User::op_iterator i = I->op_begin(), e = I->op_end(); i != e; ++i)
381 if (!DominatesMergePoint(*i, BB, AggressiveInsts, CostRemaining))
383 // Okay, it's safe to do this! Remember this instruction.
384 AggressiveInsts->insert(I);
388 /// GetConstantInt - Extract ConstantInt from value, looking through IntToPtr
389 /// and PointerNullValue. Return NULL if value is not a constant int.
390 static ConstantInt *GetConstantInt(Value *V, const DataLayout *TD) {
391 // Normal constant int.
392 ConstantInt *CI = dyn_cast<ConstantInt>(V);
393 if (CI || !TD || !isa<Constant>(V) || !V->getType()->isPointerTy())
396 // This is some kind of pointer constant. Turn it into a pointer-sized
397 // ConstantInt if possible.
398 IntegerType *PtrTy = cast<IntegerType>(TD->getIntPtrType(V->getType()));
400 // Null pointer means 0, see SelectionDAGBuilder::getValue(const Value*).
401 if (isa<ConstantPointerNull>(V))
402 return ConstantInt::get(PtrTy, 0);
404 // IntToPtr const int.
405 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V))
406 if (CE->getOpcode() == Instruction::IntToPtr)
407 if (ConstantInt *CI = dyn_cast<ConstantInt>(CE->getOperand(0))) {
408 // The constant is very likely to have the right type already.
409 if (CI->getType() == PtrTy)
412 return cast<ConstantInt>
413 (ConstantExpr::getIntegerCast(CI, PtrTy, /*isSigned=*/false));
418 /// GatherConstantCompares - Given a potentially 'or'd or 'and'd together
419 /// collection of icmp eq/ne instructions that compare a value against a
420 /// constant, return the value being compared, and stick the constant into the
423 GatherConstantCompares(Value *V, std::vector<ConstantInt*> &Vals, Value *&Extra,
424 const DataLayout *TD, bool isEQ, unsigned &UsedICmps) {
425 Instruction *I = dyn_cast<Instruction>(V);
426 if (I == 0) return 0;
428 // If this is an icmp against a constant, handle this as one of the cases.
429 if (ICmpInst *ICI = dyn_cast<ICmpInst>(I)) {
430 if (ConstantInt *C = GetConstantInt(I->getOperand(1), TD)) {
431 if (ICI->getPredicate() == (isEQ ? ICmpInst::ICMP_EQ:ICmpInst::ICMP_NE)) {
434 return I->getOperand(0);
437 // If we have "x ult 3" comparison, for example, then we can add 0,1,2 to
440 ConstantRange::makeICmpRegion(ICI->getPredicate(), C->getValue());
442 // If this is an and/!= check then we want to optimize "x ugt 2" into
445 Span = Span.inverse();
447 // If there are a ton of values, we don't want to make a ginormous switch.
448 if (Span.getSetSize().ugt(8) || Span.isEmptySet())
451 for (APInt Tmp = Span.getLower(); Tmp != Span.getUpper(); ++Tmp)
452 Vals.push_back(ConstantInt::get(V->getContext(), Tmp));
454 return I->getOperand(0);
459 // Otherwise, we can only handle an | or &, depending on isEQ.
460 if (I->getOpcode() != (isEQ ? Instruction::Or : Instruction::And))
463 unsigned NumValsBeforeLHS = Vals.size();
464 unsigned UsedICmpsBeforeLHS = UsedICmps;
465 if (Value *LHS = GatherConstantCompares(I->getOperand(0), Vals, Extra, TD,
467 unsigned NumVals = Vals.size();
468 unsigned UsedICmpsBeforeRHS = UsedICmps;
469 if (Value *RHS = GatherConstantCompares(I->getOperand(1), Vals, Extra, TD,
473 Vals.resize(NumVals);
474 UsedICmps = UsedICmpsBeforeRHS;
477 // The RHS of the or/and can't be folded in and we haven't used "Extra" yet,
478 // set it and return success.
479 if (Extra == 0 || Extra == I->getOperand(1)) {
480 Extra = I->getOperand(1);
484 Vals.resize(NumValsBeforeLHS);
485 UsedICmps = UsedICmpsBeforeLHS;
489 // If the LHS can't be folded in, but Extra is available and RHS can, try to
491 if (Extra == 0 || Extra == I->getOperand(0)) {
492 Value *OldExtra = Extra;
493 Extra = I->getOperand(0);
494 if (Value *RHS = GatherConstantCompares(I->getOperand(1), Vals, Extra, TD,
497 assert(Vals.size() == NumValsBeforeLHS);
504 static void EraseTerminatorInstAndDCECond(TerminatorInst *TI) {
505 Instruction *Cond = 0;
506 if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) {
507 Cond = dyn_cast<Instruction>(SI->getCondition());
508 } else if (BranchInst *BI = dyn_cast<BranchInst>(TI)) {
509 if (BI->isConditional())
510 Cond = dyn_cast<Instruction>(BI->getCondition());
511 } else if (IndirectBrInst *IBI = dyn_cast<IndirectBrInst>(TI)) {
512 Cond = dyn_cast<Instruction>(IBI->getAddress());
515 TI->eraseFromParent();
516 if (Cond) RecursivelyDeleteTriviallyDeadInstructions(Cond);
519 /// isValueEqualityComparison - Return true if the specified terminator checks
520 /// to see if a value is equal to constant integer value.
521 Value *SimplifyCFGOpt::isValueEqualityComparison(TerminatorInst *TI) {
523 if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) {
524 // Do not permit merging of large switch instructions into their
525 // predecessors unless there is only one predecessor.
526 if (SI->getNumSuccessors()*std::distance(pred_begin(SI->getParent()),
527 pred_end(SI->getParent())) <= 128)
528 CV = SI->getCondition();
529 } else if (BranchInst *BI = dyn_cast<BranchInst>(TI))
530 if (BI->isConditional() && BI->getCondition()->hasOneUse())
531 if (ICmpInst *ICI = dyn_cast<ICmpInst>(BI->getCondition()))
532 if ((ICI->getPredicate() == ICmpInst::ICMP_EQ ||
533 ICI->getPredicate() == ICmpInst::ICMP_NE) &&
534 GetConstantInt(ICI->getOperand(1), TD))
535 CV = ICI->getOperand(0);
537 // Unwrap any lossless ptrtoint cast.
539 PtrToIntInst *PTII = NULL;
540 if ((PTII = dyn_cast<PtrToIntInst>(CV)) &&
541 CV->getType() == TD->getIntPtrType(CV->getContext(),
542 PTII->getPointerAddressSpace()))
543 CV = PTII->getOperand(0);
548 /// GetValueEqualityComparisonCases - Given a value comparison instruction,
549 /// decode all of the 'cases' that it represents and return the 'default' block.
550 BasicBlock *SimplifyCFGOpt::
551 GetValueEqualityComparisonCases(TerminatorInst *TI,
552 std::vector<ValueEqualityComparisonCase>
554 if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) {
555 Cases.reserve(SI->getNumCases());
556 for (SwitchInst::CaseIt i = SI->case_begin(), e = SI->case_end(); i != e; ++i)
557 Cases.push_back(ValueEqualityComparisonCase(i.getCaseValue(),
558 i.getCaseSuccessor()));
559 return SI->getDefaultDest();
562 BranchInst *BI = cast<BranchInst>(TI);
563 ICmpInst *ICI = cast<ICmpInst>(BI->getCondition());
564 BasicBlock *Succ = BI->getSuccessor(ICI->getPredicate() == ICmpInst::ICMP_NE);
565 Cases.push_back(ValueEqualityComparisonCase(GetConstantInt(ICI->getOperand(1),
568 return BI->getSuccessor(ICI->getPredicate() == ICmpInst::ICMP_EQ);
572 /// EliminateBlockCases - Given a vector of bb/value pairs, remove any entries
573 /// in the list that match the specified block.
574 static void EliminateBlockCases(BasicBlock *BB,
575 std::vector<ValueEqualityComparisonCase> &Cases) {
576 Cases.erase(std::remove(Cases.begin(), Cases.end(), BB), Cases.end());
579 /// ValuesOverlap - Return true if there are any keys in C1 that exist in C2 as
582 ValuesOverlap(std::vector<ValueEqualityComparisonCase> &C1,
583 std::vector<ValueEqualityComparisonCase > &C2) {
584 std::vector<ValueEqualityComparisonCase> *V1 = &C1, *V2 = &C2;
586 // Make V1 be smaller than V2.
587 if (V1->size() > V2->size())
590 if (V1->size() == 0) return false;
591 if (V1->size() == 1) {
593 ConstantInt *TheVal = (*V1)[0].Value;
594 for (unsigned i = 0, e = V2->size(); i != e; ++i)
595 if (TheVal == (*V2)[i].Value)
599 // Otherwise, just sort both lists and compare element by element.
600 array_pod_sort(V1->begin(), V1->end());
601 array_pod_sort(V2->begin(), V2->end());
602 unsigned i1 = 0, i2 = 0, e1 = V1->size(), e2 = V2->size();
603 while (i1 != e1 && i2 != e2) {
604 if ((*V1)[i1].Value == (*V2)[i2].Value)
606 if ((*V1)[i1].Value < (*V2)[i2].Value)
614 /// SimplifyEqualityComparisonWithOnlyPredecessor - If TI is known to be a
615 /// terminator instruction and its block is known to only have a single
616 /// predecessor block, check to see if that predecessor is also a value
617 /// comparison with the same value, and if that comparison determines the
618 /// outcome of this comparison. If so, simplify TI. This does a very limited
619 /// form of jump threading.
620 bool SimplifyCFGOpt::
621 SimplifyEqualityComparisonWithOnlyPredecessor(TerminatorInst *TI,
623 IRBuilder<> &Builder) {
624 Value *PredVal = isValueEqualityComparison(Pred->getTerminator());
625 if (!PredVal) return false; // Not a value comparison in predecessor.
627 Value *ThisVal = isValueEqualityComparison(TI);
628 assert(ThisVal && "This isn't a value comparison!!");
629 if (ThisVal != PredVal) return false; // Different predicates.
631 // TODO: Preserve branch weight metadata, similarly to how
632 // FoldValueComparisonIntoPredecessors preserves it.
634 // Find out information about when control will move from Pred to TI's block.
635 std::vector<ValueEqualityComparisonCase> PredCases;
636 BasicBlock *PredDef = GetValueEqualityComparisonCases(Pred->getTerminator(),
638 EliminateBlockCases(PredDef, PredCases); // Remove default from cases.
640 // Find information about how control leaves this block.
641 std::vector<ValueEqualityComparisonCase> ThisCases;
642 BasicBlock *ThisDef = GetValueEqualityComparisonCases(TI, ThisCases);
643 EliminateBlockCases(ThisDef, ThisCases); // Remove default from cases.
645 // If TI's block is the default block from Pred's comparison, potentially
646 // simplify TI based on this knowledge.
647 if (PredDef == TI->getParent()) {
648 // If we are here, we know that the value is none of those cases listed in
649 // PredCases. If there are any cases in ThisCases that are in PredCases, we
651 if (!ValuesOverlap(PredCases, ThisCases))
654 if (isa<BranchInst>(TI)) {
655 // Okay, one of the successors of this condbr is dead. Convert it to a
657 assert(ThisCases.size() == 1 && "Branch can only have one case!");
658 // Insert the new branch.
659 Instruction *NI = Builder.CreateBr(ThisDef);
662 // Remove PHI node entries for the dead edge.
663 ThisCases[0].Dest->removePredecessor(TI->getParent());
665 DEBUG(dbgs() << "Threading pred instr: " << *Pred->getTerminator()
666 << "Through successor TI: " << *TI << "Leaving: " << *NI << "\n");
668 EraseTerminatorInstAndDCECond(TI);
672 SwitchInst *SI = cast<SwitchInst>(TI);
673 // Okay, TI has cases that are statically dead, prune them away.
674 SmallPtrSet<Constant*, 16> DeadCases;
675 for (unsigned i = 0, e = PredCases.size(); i != e; ++i)
676 DeadCases.insert(PredCases[i].Value);
678 DEBUG(dbgs() << "Threading pred instr: " << *Pred->getTerminator()
679 << "Through successor TI: " << *TI);
681 // Collect branch weights into a vector.
682 SmallVector<uint32_t, 8> Weights;
683 MDNode* MD = SI->getMetadata(LLVMContext::MD_prof);
684 bool HasWeight = MD && (MD->getNumOperands() == 2 + SI->getNumCases());
686 for (unsigned MD_i = 1, MD_e = MD->getNumOperands(); MD_i < MD_e;
688 ConstantInt* CI = dyn_cast<ConstantInt>(MD->getOperand(MD_i));
690 Weights.push_back(CI->getValue().getZExtValue());
692 for (SwitchInst::CaseIt i = SI->case_end(), e = SI->case_begin(); i != e;) {
694 if (DeadCases.count(i.getCaseValue())) {
696 std::swap(Weights[i.getCaseIndex()+1], Weights.back());
699 i.getCaseSuccessor()->removePredecessor(TI->getParent());
703 if (HasWeight && Weights.size() >= 2)
704 SI->setMetadata(LLVMContext::MD_prof,
705 MDBuilder(SI->getParent()->getContext()).
706 createBranchWeights(Weights));
708 DEBUG(dbgs() << "Leaving: " << *TI << "\n");
712 // Otherwise, TI's block must correspond to some matched value. Find out
713 // which value (or set of values) this is.
714 ConstantInt *TIV = 0;
715 BasicBlock *TIBB = TI->getParent();
716 for (unsigned i = 0, e = PredCases.size(); i != e; ++i)
717 if (PredCases[i].Dest == TIBB) {
719 return false; // Cannot handle multiple values coming to this block.
720 TIV = PredCases[i].Value;
722 assert(TIV && "No edge from pred to succ?");
724 // Okay, we found the one constant that our value can be if we get into TI's
725 // BB. Find out which successor will unconditionally be branched to.
726 BasicBlock *TheRealDest = 0;
727 for (unsigned i = 0, e = ThisCases.size(); i != e; ++i)
728 if (ThisCases[i].Value == TIV) {
729 TheRealDest = ThisCases[i].Dest;
733 // If not handled by any explicit cases, it is handled by the default case.
734 if (TheRealDest == 0) TheRealDest = ThisDef;
736 // Remove PHI node entries for dead edges.
737 BasicBlock *CheckEdge = TheRealDest;
738 for (succ_iterator SI = succ_begin(TIBB), e = succ_end(TIBB); SI != e; ++SI)
739 if (*SI != CheckEdge)
740 (*SI)->removePredecessor(TIBB);
744 // Insert the new branch.
745 Instruction *NI = Builder.CreateBr(TheRealDest);
748 DEBUG(dbgs() << "Threading pred instr: " << *Pred->getTerminator()
749 << "Through successor TI: " << *TI << "Leaving: " << *NI << "\n");
751 EraseTerminatorInstAndDCECond(TI);
756 /// ConstantIntOrdering - This class implements a stable ordering of constant
757 /// integers that does not depend on their address. This is important for
758 /// applications that sort ConstantInt's to ensure uniqueness.
759 struct ConstantIntOrdering {
760 bool operator()(const ConstantInt *LHS, const ConstantInt *RHS) const {
761 return LHS->getValue().ult(RHS->getValue());
766 static int ConstantIntSortPredicate(const void *P1, const void *P2) {
767 const ConstantInt *LHS = *(const ConstantInt*const*)P1;
768 const ConstantInt *RHS = *(const ConstantInt*const*)P2;
769 if (LHS->getValue().ult(RHS->getValue()))
771 if (LHS->getValue() == RHS->getValue())
776 static inline bool HasBranchWeights(const Instruction* I) {
777 MDNode* ProfMD = I->getMetadata(LLVMContext::MD_prof);
778 if (ProfMD && ProfMD->getOperand(0))
779 if (MDString* MDS = dyn_cast<MDString>(ProfMD->getOperand(0)))
780 return MDS->getString().equals("branch_weights");
785 /// Get Weights of a given TerminatorInst, the default weight is at the front
786 /// of the vector. If TI is a conditional eq, we need to swap the branch-weight
788 static void GetBranchWeights(TerminatorInst *TI,
789 SmallVectorImpl<uint64_t> &Weights) {
790 MDNode* MD = TI->getMetadata(LLVMContext::MD_prof);
792 for (unsigned i = 1, e = MD->getNumOperands(); i < e; ++i) {
793 ConstantInt* CI = dyn_cast<ConstantInt>(MD->getOperand(i));
795 Weights.push_back(CI->getValue().getZExtValue());
798 // If TI is a conditional eq, the default case is the false case,
799 // and the corresponding branch-weight data is at index 2. We swap the
800 // default weight to be the first entry.
801 if (BranchInst* BI = dyn_cast<BranchInst>(TI)) {
802 assert(Weights.size() == 2);
803 ICmpInst *ICI = cast<ICmpInst>(BI->getCondition());
804 if (ICI->getPredicate() == ICmpInst::ICMP_EQ)
805 std::swap(Weights.front(), Weights.back());
809 /// Sees if any of the weights are too big for a uint32_t, and halves all the
810 /// weights if any are.
811 static void FitWeights(MutableArrayRef<uint64_t> Weights) {
813 for (unsigned i = 0; i < Weights.size(); ++i)
814 if (Weights[i] > UINT_MAX) {
822 for (unsigned i = 0; i < Weights.size(); ++i)
826 /// FoldValueComparisonIntoPredecessors - The specified terminator is a value
827 /// equality comparison instruction (either a switch or a branch on "X == c").
828 /// See if any of the predecessors of the terminator block are value comparisons
829 /// on the same value. If so, and if safe to do so, fold them together.
830 bool SimplifyCFGOpt::FoldValueComparisonIntoPredecessors(TerminatorInst *TI,
831 IRBuilder<> &Builder) {
832 BasicBlock *BB = TI->getParent();
833 Value *CV = isValueEqualityComparison(TI); // CondVal
834 assert(CV && "Not a comparison?");
835 bool Changed = false;
837 SmallVector<BasicBlock*, 16> Preds(pred_begin(BB), pred_end(BB));
838 while (!Preds.empty()) {
839 BasicBlock *Pred = Preds.pop_back_val();
841 // See if the predecessor is a comparison with the same value.
842 TerminatorInst *PTI = Pred->getTerminator();
843 Value *PCV = isValueEqualityComparison(PTI); // PredCondVal
845 if (PCV == CV && SafeToMergeTerminators(TI, PTI)) {
846 // Figure out which 'cases' to copy from SI to PSI.
847 std::vector<ValueEqualityComparisonCase> BBCases;
848 BasicBlock *BBDefault = GetValueEqualityComparisonCases(TI, BBCases);
850 std::vector<ValueEqualityComparisonCase> PredCases;
851 BasicBlock *PredDefault = GetValueEqualityComparisonCases(PTI, PredCases);
853 // Based on whether the default edge from PTI goes to BB or not, fill in
854 // PredCases and PredDefault with the new switch cases we would like to
856 SmallVector<BasicBlock*, 8> NewSuccessors;
858 // Update the branch weight metadata along the way
859 SmallVector<uint64_t, 8> Weights;
860 bool PredHasWeights = HasBranchWeights(PTI);
861 bool SuccHasWeights = HasBranchWeights(TI);
863 if (PredHasWeights) {
864 GetBranchWeights(PTI, Weights);
865 // branch-weight metadata is inconsistant here.
866 if (Weights.size() != 1 + PredCases.size())
867 PredHasWeights = SuccHasWeights = false;
868 } else if (SuccHasWeights)
869 // If there are no predecessor weights but there are successor weights,
870 // populate Weights with 1, which will later be scaled to the sum of
871 // successor's weights
872 Weights.assign(1 + PredCases.size(), 1);
874 SmallVector<uint64_t, 8> SuccWeights;
875 if (SuccHasWeights) {
876 GetBranchWeights(TI, SuccWeights);
877 // branch-weight metadata is inconsistant here.
878 if (SuccWeights.size() != 1 + BBCases.size())
879 PredHasWeights = SuccHasWeights = false;
880 } else if (PredHasWeights)
881 SuccWeights.assign(1 + BBCases.size(), 1);
883 if (PredDefault == BB) {
884 // If this is the default destination from PTI, only the edges in TI
885 // that don't occur in PTI, or that branch to BB will be activated.
886 std::set<ConstantInt*, ConstantIntOrdering> PTIHandled;
887 for (unsigned i = 0, e = PredCases.size(); i != e; ++i)
888 if (PredCases[i].Dest != BB)
889 PTIHandled.insert(PredCases[i].Value);
891 // The default destination is BB, we don't need explicit targets.
892 std::swap(PredCases[i], PredCases.back());
894 if (PredHasWeights || SuccHasWeights) {
895 // Increase weight for the default case.
896 Weights[0] += Weights[i+1];
897 std::swap(Weights[i+1], Weights.back());
901 PredCases.pop_back();
905 // Reconstruct the new switch statement we will be building.
906 if (PredDefault != BBDefault) {
907 PredDefault->removePredecessor(Pred);
908 PredDefault = BBDefault;
909 NewSuccessors.push_back(BBDefault);
912 unsigned CasesFromPred = Weights.size();
913 uint64_t ValidTotalSuccWeight = 0;
914 for (unsigned i = 0, e = BBCases.size(); i != e; ++i)
915 if (!PTIHandled.count(BBCases[i].Value) &&
916 BBCases[i].Dest != BBDefault) {
917 PredCases.push_back(BBCases[i]);
918 NewSuccessors.push_back(BBCases[i].Dest);
919 if (SuccHasWeights || PredHasWeights) {
920 // The default weight is at index 0, so weight for the ith case
921 // should be at index i+1. Scale the cases from successor by
922 // PredDefaultWeight (Weights[0]).
923 Weights.push_back(Weights[0] * SuccWeights[i+1]);
924 ValidTotalSuccWeight += SuccWeights[i+1];
928 if (SuccHasWeights || PredHasWeights) {
929 ValidTotalSuccWeight += SuccWeights[0];
930 // Scale the cases from predecessor by ValidTotalSuccWeight.
931 for (unsigned i = 1; i < CasesFromPred; ++i)
932 Weights[i] *= ValidTotalSuccWeight;
933 // Scale the default weight by SuccDefaultWeight (SuccWeights[0]).
934 Weights[0] *= SuccWeights[0];
937 // If this is not the default destination from PSI, only the edges
938 // in SI that occur in PSI with a destination of BB will be
940 std::set<ConstantInt*, ConstantIntOrdering> PTIHandled;
941 std::map<ConstantInt*, uint64_t> WeightsForHandled;
942 for (unsigned i = 0, e = PredCases.size(); i != e; ++i)
943 if (PredCases[i].Dest == BB) {
944 PTIHandled.insert(PredCases[i].Value);
946 if (PredHasWeights || SuccHasWeights) {
947 WeightsForHandled[PredCases[i].Value] = Weights[i+1];
948 std::swap(Weights[i+1], Weights.back());
952 std::swap(PredCases[i], PredCases.back());
953 PredCases.pop_back();
957 // Okay, now we know which constants were sent to BB from the
958 // predecessor. Figure out where they will all go now.
959 for (unsigned i = 0, e = BBCases.size(); i != e; ++i)
960 if (PTIHandled.count(BBCases[i].Value)) {
961 // If this is one we are capable of getting...
962 if (PredHasWeights || SuccHasWeights)
963 Weights.push_back(WeightsForHandled[BBCases[i].Value]);
964 PredCases.push_back(BBCases[i]);
965 NewSuccessors.push_back(BBCases[i].Dest);
966 PTIHandled.erase(BBCases[i].Value);// This constant is taken care of
969 // If there are any constants vectored to BB that TI doesn't handle,
970 // they must go to the default destination of TI.
971 for (std::set<ConstantInt*, ConstantIntOrdering>::iterator I =
973 E = PTIHandled.end(); I != E; ++I) {
974 if (PredHasWeights || SuccHasWeights)
975 Weights.push_back(WeightsForHandled[*I]);
976 PredCases.push_back(ValueEqualityComparisonCase(*I, BBDefault));
977 NewSuccessors.push_back(BBDefault);
981 // Okay, at this point, we know which new successor Pred will get. Make
982 // sure we update the number of entries in the PHI nodes for these
984 for (unsigned i = 0, e = NewSuccessors.size(); i != e; ++i)
985 AddPredecessorToBlock(NewSuccessors[i], Pred, BB);
987 Builder.SetInsertPoint(PTI);
988 // Convert pointer to int before we switch.
989 if (CV->getType()->isPointerTy()) {
990 assert(TD && "Cannot switch on pointer without DataLayout");
991 CV = Builder.CreatePtrToInt(CV, TD->getIntPtrType(CV->getType()),
995 // Now that the successors are updated, create the new Switch instruction.
996 SwitchInst *NewSI = Builder.CreateSwitch(CV, PredDefault,
998 NewSI->setDebugLoc(PTI->getDebugLoc());
999 for (unsigned i = 0, e = PredCases.size(); i != e; ++i)
1000 NewSI->addCase(PredCases[i].Value, PredCases[i].Dest);
1002 if (PredHasWeights || SuccHasWeights) {
1003 // Halve the weights if any of them cannot fit in an uint32_t
1004 FitWeights(Weights);
1006 SmallVector<uint32_t, 8> MDWeights(Weights.begin(), Weights.end());
1008 NewSI->setMetadata(LLVMContext::MD_prof,
1009 MDBuilder(BB->getContext()).
1010 createBranchWeights(MDWeights));
1013 EraseTerminatorInstAndDCECond(PTI);
1015 // Okay, last check. If BB is still a successor of PSI, then we must
1016 // have an infinite loop case. If so, add an infinitely looping block
1017 // to handle the case to preserve the behavior of the code.
1018 BasicBlock *InfLoopBlock = 0;
1019 for (unsigned i = 0, e = NewSI->getNumSuccessors(); i != e; ++i)
1020 if (NewSI->getSuccessor(i) == BB) {
1021 if (InfLoopBlock == 0) {
1022 // Insert it at the end of the function, because it's either code,
1023 // or it won't matter if it's hot. :)
1024 InfLoopBlock = BasicBlock::Create(BB->getContext(),
1025 "infloop", BB->getParent());
1026 BranchInst::Create(InfLoopBlock, InfLoopBlock);
1028 NewSI->setSuccessor(i, InfLoopBlock);
1037 // isSafeToHoistInvoke - If we would need to insert a select that uses the
1038 // value of this invoke (comments in HoistThenElseCodeToIf explain why we
1039 // would need to do this), we can't hoist the invoke, as there is nowhere
1040 // to put the select in this case.
1041 static bool isSafeToHoistInvoke(BasicBlock *BB1, BasicBlock *BB2,
1042 Instruction *I1, Instruction *I2) {
1043 for (succ_iterator SI = succ_begin(BB1), E = succ_end(BB1); SI != E; ++SI) {
1045 for (BasicBlock::iterator BBI = SI->begin();
1046 (PN = dyn_cast<PHINode>(BBI)); ++BBI) {
1047 Value *BB1V = PN->getIncomingValueForBlock(BB1);
1048 Value *BB2V = PN->getIncomingValueForBlock(BB2);
1049 if (BB1V != BB2V && (BB1V==I1 || BB2V==I2)) {
1057 /// HoistThenElseCodeToIf - Given a conditional branch that goes to BB1 and
1058 /// BB2, hoist any common code in the two blocks up into the branch block. The
1059 /// caller of this function guarantees that BI's block dominates BB1 and BB2.
1060 static bool HoistThenElseCodeToIf(BranchInst *BI) {
1061 // This does very trivial matching, with limited scanning, to find identical
1062 // instructions in the two blocks. In particular, we don't want to get into
1063 // O(M*N) situations here where M and N are the sizes of BB1 and BB2. As
1064 // such, we currently just scan for obviously identical instructions in an
1066 BasicBlock *BB1 = BI->getSuccessor(0); // The true destination.
1067 BasicBlock *BB2 = BI->getSuccessor(1); // The false destination
1069 BasicBlock::iterator BB1_Itr = BB1->begin();
1070 BasicBlock::iterator BB2_Itr = BB2->begin();
1072 Instruction *I1 = BB1_Itr++, *I2 = BB2_Itr++;
1073 // Skip debug info if it is not identical.
1074 DbgInfoIntrinsic *DBI1 = dyn_cast<DbgInfoIntrinsic>(I1);
1075 DbgInfoIntrinsic *DBI2 = dyn_cast<DbgInfoIntrinsic>(I2);
1076 if (!DBI1 || !DBI2 || !DBI1->isIdenticalToWhenDefined(DBI2)) {
1077 while (isa<DbgInfoIntrinsic>(I1))
1079 while (isa<DbgInfoIntrinsic>(I2))
1082 if (isa<PHINode>(I1) || !I1->isIdenticalToWhenDefined(I2) ||
1083 (isa<InvokeInst>(I1) && !isSafeToHoistInvoke(BB1, BB2, I1, I2)))
1086 // If we get here, we can hoist at least one instruction.
1087 BasicBlock *BIParent = BI->getParent();
1090 // If we are hoisting the terminator instruction, don't move one (making a
1091 // broken BB), instead clone it, and remove BI.
1092 if (isa<TerminatorInst>(I1))
1093 goto HoistTerminator;
1095 // For a normal instruction, we just move one to right before the branch,
1096 // then replace all uses of the other with the first. Finally, we remove
1097 // the now redundant second instruction.
1098 BIParent->getInstList().splice(BI, BB1->getInstList(), I1);
1099 if (!I2->use_empty())
1100 I2->replaceAllUsesWith(I1);
1101 I1->intersectOptionalDataWith(I2);
1102 I2->eraseFromParent();
1106 // Skip debug info if it is not identical.
1107 DbgInfoIntrinsic *DBI1 = dyn_cast<DbgInfoIntrinsic>(I1);
1108 DbgInfoIntrinsic *DBI2 = dyn_cast<DbgInfoIntrinsic>(I2);
1109 if (!DBI1 || !DBI2 || !DBI1->isIdenticalToWhenDefined(DBI2)) {
1110 while (isa<DbgInfoIntrinsic>(I1))
1112 while (isa<DbgInfoIntrinsic>(I2))
1115 } while (I1->isIdenticalToWhenDefined(I2));
1120 // It may not be possible to hoist an invoke.
1121 if (isa<InvokeInst>(I1) && !isSafeToHoistInvoke(BB1, BB2, I1, I2))
1124 // Okay, it is safe to hoist the terminator.
1125 Instruction *NT = I1->clone();
1126 BIParent->getInstList().insert(BI, NT);
1127 if (!NT->getType()->isVoidTy()) {
1128 I1->replaceAllUsesWith(NT);
1129 I2->replaceAllUsesWith(NT);
1133 IRBuilder<true, NoFolder> Builder(NT);
1134 // Hoisting one of the terminators from our successor is a great thing.
1135 // Unfortunately, the successors of the if/else blocks may have PHI nodes in
1136 // them. If they do, all PHI entries for BB1/BB2 must agree for all PHI
1137 // nodes, so we insert select instruction to compute the final result.
1138 std::map<std::pair<Value*,Value*>, SelectInst*> InsertedSelects;
1139 for (succ_iterator SI = succ_begin(BB1), E = succ_end(BB1); SI != E; ++SI) {
1141 for (BasicBlock::iterator BBI = SI->begin();
1142 (PN = dyn_cast<PHINode>(BBI)); ++BBI) {
1143 Value *BB1V = PN->getIncomingValueForBlock(BB1);
1144 Value *BB2V = PN->getIncomingValueForBlock(BB2);
1145 if (BB1V == BB2V) continue;
1147 // These values do not agree. Insert a select instruction before NT
1148 // that determines the right value.
1149 SelectInst *&SI = InsertedSelects[std::make_pair(BB1V, BB2V)];
1151 SI = cast<SelectInst>
1152 (Builder.CreateSelect(BI->getCondition(), BB1V, BB2V,
1153 BB1V->getName()+"."+BB2V->getName()));
1155 // Make the PHI node use the select for all incoming values for BB1/BB2
1156 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
1157 if (PN->getIncomingBlock(i) == BB1 || PN->getIncomingBlock(i) == BB2)
1158 PN->setIncomingValue(i, SI);
1162 // Update any PHI nodes in our new successors.
1163 for (succ_iterator SI = succ_begin(BB1), E = succ_end(BB1); SI != E; ++SI)
1164 AddPredecessorToBlock(*SI, BIParent, BB1);
1166 EraseTerminatorInstAndDCECond(BI);
1170 /// SinkThenElseCodeToEnd - Given an unconditional branch that goes to BBEnd,
1171 /// check whether BBEnd has only two predecessors and the other predecessor
1172 /// ends with an unconditional branch. If it is true, sink any common code
1173 /// in the two predecessors to BBEnd.
1174 static bool SinkThenElseCodeToEnd(BranchInst *BI1) {
1175 assert(BI1->isUnconditional());
1176 BasicBlock *BB1 = BI1->getParent();
1177 BasicBlock *BBEnd = BI1->getSuccessor(0);
1179 // Check that BBEnd has two predecessors and the other predecessor ends with
1180 // an unconditional branch.
1181 pred_iterator PI = pred_begin(BBEnd), PE = pred_end(BBEnd);
1182 BasicBlock *Pred0 = *PI++;
1183 if (PI == PE) // Only one predecessor.
1185 BasicBlock *Pred1 = *PI++;
1186 if (PI != PE) // More than two predecessors.
1188 BasicBlock *BB2 = (Pred0 == BB1) ? Pred1 : Pred0;
1189 BranchInst *BI2 = dyn_cast<BranchInst>(BB2->getTerminator());
1190 if (!BI2 || !BI2->isUnconditional())
1193 // Gather the PHI nodes in BBEnd.
1194 std::map<Value*, std::pair<Value*, PHINode*> > MapValueFromBB1ToBB2;
1195 Instruction *FirstNonPhiInBBEnd = 0;
1196 for (BasicBlock::iterator I = BBEnd->begin(), E = BBEnd->end();
1198 if (PHINode *PN = dyn_cast<PHINode>(I)) {
1199 Value *BB1V = PN->getIncomingValueForBlock(BB1);
1200 Value *BB2V = PN->getIncomingValueForBlock(BB2);
1201 MapValueFromBB1ToBB2[BB1V] = std::make_pair(BB2V, PN);
1203 FirstNonPhiInBBEnd = &*I;
1207 if (!FirstNonPhiInBBEnd)
1211 // This does very trivial matching, with limited scanning, to find identical
1212 // instructions in the two blocks. We scan backward for obviously identical
1213 // instructions in an identical order.
1214 BasicBlock::InstListType::reverse_iterator RI1 = BB1->getInstList().rbegin(),
1215 RE1 = BB1->getInstList().rend(), RI2 = BB2->getInstList().rbegin(),
1216 RE2 = BB2->getInstList().rend();
1218 while (RI1 != RE1 && isa<DbgInfoIntrinsic>(&*RI1)) ++RI1;
1221 while (RI2 != RE2 && isa<DbgInfoIntrinsic>(&*RI2)) ++RI2;
1224 // Skip the unconditional branches.
1228 bool Changed = false;
1229 while (RI1 != RE1 && RI2 != RE2) {
1231 while (RI1 != RE1 && isa<DbgInfoIntrinsic>(&*RI1)) ++RI1;
1234 while (RI2 != RE2 && isa<DbgInfoIntrinsic>(&*RI2)) ++RI2;
1238 Instruction *I1 = &*RI1, *I2 = &*RI2;
1239 // I1 and I2 should have a single use in the same PHI node, and they
1240 // perform the same operation.
1241 // Cannot move control-flow-involving, volatile loads, vaarg, etc.
1242 if (isa<PHINode>(I1) || isa<PHINode>(I2) ||
1243 isa<TerminatorInst>(I1) || isa<TerminatorInst>(I2) ||
1244 isa<LandingPadInst>(I1) || isa<LandingPadInst>(I2) ||
1245 isa<AllocaInst>(I1) || isa<AllocaInst>(I2) ||
1246 I1->mayHaveSideEffects() || I2->mayHaveSideEffects() ||
1247 I1->mayReadOrWriteMemory() || I2->mayReadOrWriteMemory() ||
1248 !I1->hasOneUse() || !I2->hasOneUse() ||
1249 MapValueFromBB1ToBB2.find(I1) == MapValueFromBB1ToBB2.end() ||
1250 MapValueFromBB1ToBB2[I1].first != I2)
1253 // Check whether we should swap the operands of ICmpInst.
1254 ICmpInst *ICmp1 = dyn_cast<ICmpInst>(I1), *ICmp2 = dyn_cast<ICmpInst>(I2);
1255 bool SwapOpnds = false;
1256 if (ICmp1 && ICmp2 &&
1257 ICmp1->getOperand(0) != ICmp2->getOperand(0) &&
1258 ICmp1->getOperand(1) != ICmp2->getOperand(1) &&
1259 (ICmp1->getOperand(0) == ICmp2->getOperand(1) ||
1260 ICmp1->getOperand(1) == ICmp2->getOperand(0))) {
1261 ICmp2->swapOperands();
1264 if (!I1->isSameOperationAs(I2)) {
1266 ICmp2->swapOperands();
1270 // The operands should be either the same or they need to be generated
1271 // with a PHI node after sinking. We only handle the case where there is
1272 // a single pair of different operands.
1273 Value *DifferentOp1 = 0, *DifferentOp2 = 0;
1274 unsigned Op1Idx = 0;
1275 for (unsigned I = 0, E = I1->getNumOperands(); I != E; ++I) {
1276 if (I1->getOperand(I) == I2->getOperand(I))
1278 // Early exit if we have more-than one pair of different operands or
1279 // the different operand is already in MapValueFromBB1ToBB2.
1280 // Early exit if we need a PHI node to replace a constant.
1282 MapValueFromBB1ToBB2.find(I1->getOperand(I)) !=
1283 MapValueFromBB1ToBB2.end() ||
1284 isa<Constant>(I1->getOperand(I)) ||
1285 isa<Constant>(I2->getOperand(I))) {
1286 // If we can't sink the instructions, undo the swapping.
1288 ICmp2->swapOperands();
1291 DifferentOp1 = I1->getOperand(I);
1293 DifferentOp2 = I2->getOperand(I);
1296 // We insert the pair of different operands to MapValueFromBB1ToBB2 and
1297 // remove (I1, I2) from MapValueFromBB1ToBB2.
1299 PHINode *NewPN = PHINode::Create(DifferentOp1->getType(), 2,
1300 DifferentOp1->getName() + ".sink",
1302 MapValueFromBB1ToBB2[DifferentOp1] = std::make_pair(DifferentOp2, NewPN);
1303 // I1 should use NewPN instead of DifferentOp1.
1304 I1->setOperand(Op1Idx, NewPN);
1305 NewPN->addIncoming(DifferentOp1, BB1);
1306 NewPN->addIncoming(DifferentOp2, BB2);
1307 DEBUG(dbgs() << "Create PHI node " << *NewPN << "\n";);
1309 PHINode *OldPN = MapValueFromBB1ToBB2[I1].second;
1310 MapValueFromBB1ToBB2.erase(I1);
1312 DEBUG(dbgs() << "SINK common instructions " << *I1 << "\n";);
1313 DEBUG(dbgs() << " " << *I2 << "\n";);
1314 // We need to update RE1 and RE2 if we are going to sink the first
1315 // instruction in the basic block down.
1316 bool UpdateRE1 = (I1 == BB1->begin()), UpdateRE2 = (I2 == BB2->begin());
1317 // Sink the instruction.
1318 BBEnd->getInstList().splice(FirstNonPhiInBBEnd, BB1->getInstList(), I1);
1319 if (!OldPN->use_empty())
1320 OldPN->replaceAllUsesWith(I1);
1321 OldPN->eraseFromParent();
1323 if (!I2->use_empty())
1324 I2->replaceAllUsesWith(I1);
1325 I1->intersectOptionalDataWith(I2);
1326 I2->eraseFromParent();
1329 RE1 = BB1->getInstList().rend();
1331 RE2 = BB2->getInstList().rend();
1332 FirstNonPhiInBBEnd = I1;
1339 /// SpeculativelyExecuteBB - Given a conditional branch that goes to BB1
1340 /// and an BB2 and the only successor of BB1 is BB2, hoist simple code
1341 /// (for now, restricted to a single instruction that's side effect free) from
1342 /// the BB1 into the branch block to speculatively execute it.
1347 /// br i1 %t1, label %BB1, label %BB2
1349 /// %t3 = add %t2, c
1355 /// %t4 = add %t2, c
1356 /// %t3 = select i1 %t1, %t2, %t3
1357 static bool SpeculativelyExecuteBB(BranchInst *BI, BasicBlock *BB1) {
1358 // Only speculatively execution a single instruction (not counting the
1359 // terminator) for now.
1360 Instruction *HInst = NULL;
1361 Instruction *Term = BB1->getTerminator();
1362 for (BasicBlock::iterator BBI = BB1->begin(), BBE = BB1->end();
1363 BBI != BBE; ++BBI) {
1364 Instruction *I = BBI;
1366 if (isa<DbgInfoIntrinsic>(I)) continue;
1367 if (I == Term) break;
1374 BasicBlock *BIParent = BI->getParent();
1376 // Check the instruction to be hoisted, if there is one.
1378 // Don't hoist the instruction if it's unsafe or expensive.
1379 if (!isSafeToSpeculativelyExecute(HInst))
1381 if (ComputeSpeculationCost(HInst) > PHINodeFoldingThreshold)
1384 // Do not hoist the instruction if any of its operands are defined but not
1385 // used in this BB. The transformation will prevent the operand from
1386 // being sunk into the use block.
1387 for (User::op_iterator i = HInst->op_begin(), e = HInst->op_end();
1389 Instruction *OpI = dyn_cast<Instruction>(*i);
1390 if (OpI && OpI->getParent() == BIParent &&
1391 !OpI->mayHaveSideEffects() &&
1392 !OpI->isUsedInBasicBlock(BIParent))
1397 // Be conservative for now. FP select instruction can often be expensive.
1398 Value *BrCond = BI->getCondition();
1399 if (isa<FCmpInst>(BrCond))
1402 // If BB1 is actually on the false edge of the conditional branch, remember
1403 // to swap the select operands later.
1404 bool Invert = false;
1405 if (BB1 != BI->getSuccessor(0)) {
1406 assert(BB1 == BI->getSuccessor(1) && "No edge from 'if' block?");
1410 // Collect interesting PHIs, and scan for hazards.
1411 SmallSetVector<std::pair<Value *, Value *>, 4> PHIs;
1412 BasicBlock *BB2 = BB1->getTerminator()->getSuccessor(0);
1413 for (BasicBlock::iterator I = BB2->begin();
1414 PHINode *PN = dyn_cast<PHINode>(I); ++I) {
1415 Value *BB1V = PN->getIncomingValueForBlock(BB1);
1416 Value *BIParentV = PN->getIncomingValueForBlock(BIParent);
1418 // Skip PHIs which are trivial.
1419 if (BB1V == BIParentV)
1422 // Check for saftey.
1423 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(BB1V)) {
1424 // An unfolded ConstantExpr could end up getting expanded into
1425 // Instructions. Don't speculate this and another instruction at
1429 if (!isSafeToSpeculativelyExecute(CE))
1431 if (ComputeSpeculationCost(CE) > PHINodeFoldingThreshold)
1435 // Ok, we may insert a select for this PHI.
1436 PHIs.insert(std::make_pair(BB1V, BIParentV));
1439 // If there are no PHIs to process, bail early. This helps ensure idempotence
1444 // If we get here, we can hoist the instruction and if-convert.
1445 DEBUG(dbgs() << "SPECULATIVELY EXECUTING BB" << *BB1 << "\n";);
1447 // Hoist the instruction.
1449 BIParent->getInstList().splice(BI, BB1->getInstList(), HInst);
1451 // Insert selects and rewrite the PHI operands.
1452 IRBuilder<true, NoFolder> Builder(BI);
1453 for (unsigned i = 0, e = PHIs.size(); i != e; ++i) {
1454 Value *TrueV = PHIs[i].first;
1455 Value *FalseV = PHIs[i].second;
1457 // Create a select whose true value is the speculatively executed value and
1458 // false value is the previously determined FalseV.
1461 SI = cast<SelectInst>
1462 (Builder.CreateSelect(BrCond, FalseV, TrueV,
1463 FalseV->getName() + "." + TrueV->getName()));
1465 SI = cast<SelectInst>
1466 (Builder.CreateSelect(BrCond, TrueV, FalseV,
1467 TrueV->getName() + "." + FalseV->getName()));
1469 // Make the PHI node use the select for all incoming values for "then" and
1471 for (BasicBlock::iterator I = BB2->begin();
1472 PHINode *PN = dyn_cast<PHINode>(I); ++I) {
1473 unsigned BB1I = PN->getBasicBlockIndex(BB1);
1474 unsigned BIParentI = PN->getBasicBlockIndex(BIParent);
1475 Value *BB1V = PN->getIncomingValue(BB1I);
1476 Value *BIParentV = PN->getIncomingValue(BIParentI);
1477 if (TrueV == BB1V && FalseV == BIParentV) {
1478 PN->setIncomingValue(BB1I, SI);
1479 PN->setIncomingValue(BIParentI, SI);
1488 /// BlockIsSimpleEnoughToThreadThrough - Return true if we can thread a branch
1489 /// across this block.
1490 static bool BlockIsSimpleEnoughToThreadThrough(BasicBlock *BB) {
1491 BranchInst *BI = cast<BranchInst>(BB->getTerminator());
1494 for (BasicBlock::iterator BBI = BB->begin(); &*BBI != BI; ++BBI) {
1495 if (isa<DbgInfoIntrinsic>(BBI))
1497 if (Size > 10) return false; // Don't clone large BB's.
1500 // We can only support instructions that do not define values that are
1501 // live outside of the current basic block.
1502 for (Value::use_iterator UI = BBI->use_begin(), E = BBI->use_end();
1504 Instruction *U = cast<Instruction>(*UI);
1505 if (U->getParent() != BB || isa<PHINode>(U)) return false;
1508 // Looks ok, continue checking.
1514 /// FoldCondBranchOnPHI - If we have a conditional branch on a PHI node value
1515 /// that is defined in the same block as the branch and if any PHI entries are
1516 /// constants, thread edges corresponding to that entry to be branches to their
1517 /// ultimate destination.
1518 static bool FoldCondBranchOnPHI(BranchInst *BI, const DataLayout *TD) {
1519 BasicBlock *BB = BI->getParent();
1520 PHINode *PN = dyn_cast<PHINode>(BI->getCondition());
1521 // NOTE: we currently cannot transform this case if the PHI node is used
1522 // outside of the block.
1523 if (!PN || PN->getParent() != BB || !PN->hasOneUse())
1526 // Degenerate case of a single entry PHI.
1527 if (PN->getNumIncomingValues() == 1) {
1528 FoldSingleEntryPHINodes(PN->getParent());
1532 // Now we know that this block has multiple preds and two succs.
1533 if (!BlockIsSimpleEnoughToThreadThrough(BB)) return false;
1535 // Okay, this is a simple enough basic block. See if any phi values are
1537 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
1538 ConstantInt *CB = dyn_cast<ConstantInt>(PN->getIncomingValue(i));
1539 if (CB == 0 || !CB->getType()->isIntegerTy(1)) continue;
1541 // Okay, we now know that all edges from PredBB should be revectored to
1542 // branch to RealDest.
1543 BasicBlock *PredBB = PN->getIncomingBlock(i);
1544 BasicBlock *RealDest = BI->getSuccessor(!CB->getZExtValue());
1546 if (RealDest == BB) continue; // Skip self loops.
1547 // Skip if the predecessor's terminator is an indirect branch.
1548 if (isa<IndirectBrInst>(PredBB->getTerminator())) continue;
1550 // The dest block might have PHI nodes, other predecessors and other
1551 // difficult cases. Instead of being smart about this, just insert a new
1552 // block that jumps to the destination block, effectively splitting
1553 // the edge we are about to create.
1554 BasicBlock *EdgeBB = BasicBlock::Create(BB->getContext(),
1555 RealDest->getName()+".critedge",
1556 RealDest->getParent(), RealDest);
1557 BranchInst::Create(RealDest, EdgeBB);
1559 // Update PHI nodes.
1560 AddPredecessorToBlock(RealDest, EdgeBB, BB);
1562 // BB may have instructions that are being threaded over. Clone these
1563 // instructions into EdgeBB. We know that there will be no uses of the
1564 // cloned instructions outside of EdgeBB.
1565 BasicBlock::iterator InsertPt = EdgeBB->begin();
1566 DenseMap<Value*, Value*> TranslateMap; // Track translated values.
1567 for (BasicBlock::iterator BBI = BB->begin(); &*BBI != BI; ++BBI) {
1568 if (PHINode *PN = dyn_cast<PHINode>(BBI)) {
1569 TranslateMap[PN] = PN->getIncomingValueForBlock(PredBB);
1572 // Clone the instruction.
1573 Instruction *N = BBI->clone();
1574 if (BBI->hasName()) N->setName(BBI->getName()+".c");
1576 // Update operands due to translation.
1577 for (User::op_iterator i = N->op_begin(), e = N->op_end();
1579 DenseMap<Value*, Value*>::iterator PI = TranslateMap.find(*i);
1580 if (PI != TranslateMap.end())
1584 // Check for trivial simplification.
1585 if (Value *V = SimplifyInstruction(N, TD)) {
1586 TranslateMap[BBI] = V;
1587 delete N; // Instruction folded away, don't need actual inst
1589 // Insert the new instruction into its new home.
1590 EdgeBB->getInstList().insert(InsertPt, N);
1591 if (!BBI->use_empty())
1592 TranslateMap[BBI] = N;
1596 // Loop over all of the edges from PredBB to BB, changing them to branch
1597 // to EdgeBB instead.
1598 TerminatorInst *PredBBTI = PredBB->getTerminator();
1599 for (unsigned i = 0, e = PredBBTI->getNumSuccessors(); i != e; ++i)
1600 if (PredBBTI->getSuccessor(i) == BB) {
1601 BB->removePredecessor(PredBB);
1602 PredBBTI->setSuccessor(i, EdgeBB);
1605 // Recurse, simplifying any other constants.
1606 return FoldCondBranchOnPHI(BI, TD) | true;
1612 /// FoldTwoEntryPHINode - Given a BB that starts with the specified two-entry
1613 /// PHI node, see if we can eliminate it.
1614 static bool FoldTwoEntryPHINode(PHINode *PN, const DataLayout *TD) {
1615 // Ok, this is a two entry PHI node. Check to see if this is a simple "if
1616 // statement", which has a very simple dominance structure. Basically, we
1617 // are trying to find the condition that is being branched on, which
1618 // subsequently causes this merge to happen. We really want control
1619 // dependence information for this check, but simplifycfg can't keep it up
1620 // to date, and this catches most of the cases we care about anyway.
1621 BasicBlock *BB = PN->getParent();
1622 BasicBlock *IfTrue, *IfFalse;
1623 Value *IfCond = GetIfCondition(BB, IfTrue, IfFalse);
1625 // Don't bother if the branch will be constant folded trivially.
1626 isa<ConstantInt>(IfCond))
1629 // Okay, we found that we can merge this two-entry phi node into a select.
1630 // Doing so would require us to fold *all* two entry phi nodes in this block.
1631 // At some point this becomes non-profitable (particularly if the target
1632 // doesn't support cmov's). Only do this transformation if there are two or
1633 // fewer PHI nodes in this block.
1634 unsigned NumPhis = 0;
1635 for (BasicBlock::iterator I = BB->begin(); isa<PHINode>(I); ++NumPhis, ++I)
1639 // Loop over the PHI's seeing if we can promote them all to select
1640 // instructions. While we are at it, keep track of the instructions
1641 // that need to be moved to the dominating block.
1642 SmallPtrSet<Instruction*, 4> AggressiveInsts;
1643 unsigned MaxCostVal0 = PHINodeFoldingThreshold,
1644 MaxCostVal1 = PHINodeFoldingThreshold;
1646 for (BasicBlock::iterator II = BB->begin(); isa<PHINode>(II);) {
1647 PHINode *PN = cast<PHINode>(II++);
1648 if (Value *V = SimplifyInstruction(PN, TD)) {
1649 PN->replaceAllUsesWith(V);
1650 PN->eraseFromParent();
1654 if (!DominatesMergePoint(PN->getIncomingValue(0), BB, &AggressiveInsts,
1656 !DominatesMergePoint(PN->getIncomingValue(1), BB, &AggressiveInsts,
1661 // If we folded the first phi, PN dangles at this point. Refresh it. If
1662 // we ran out of PHIs then we simplified them all.
1663 PN = dyn_cast<PHINode>(BB->begin());
1664 if (PN == 0) return true;
1666 // Don't fold i1 branches on PHIs which contain binary operators. These can
1667 // often be turned into switches and other things.
1668 if (PN->getType()->isIntegerTy(1) &&
1669 (isa<BinaryOperator>(PN->getIncomingValue(0)) ||
1670 isa<BinaryOperator>(PN->getIncomingValue(1)) ||
1671 isa<BinaryOperator>(IfCond)))
1674 // If we all PHI nodes are promotable, check to make sure that all
1675 // instructions in the predecessor blocks can be promoted as well. If
1676 // not, we won't be able to get rid of the control flow, so it's not
1677 // worth promoting to select instructions.
1678 BasicBlock *DomBlock = 0;
1679 BasicBlock *IfBlock1 = PN->getIncomingBlock(0);
1680 BasicBlock *IfBlock2 = PN->getIncomingBlock(1);
1681 if (cast<BranchInst>(IfBlock1->getTerminator())->isConditional()) {
1684 DomBlock = *pred_begin(IfBlock1);
1685 for (BasicBlock::iterator I = IfBlock1->begin();!isa<TerminatorInst>(I);++I)
1686 if (!AggressiveInsts.count(I) && !isa<DbgInfoIntrinsic>(I)) {
1687 // This is not an aggressive instruction that we can promote.
1688 // Because of this, we won't be able to get rid of the control
1689 // flow, so the xform is not worth it.
1694 if (cast<BranchInst>(IfBlock2->getTerminator())->isConditional()) {
1697 DomBlock = *pred_begin(IfBlock2);
1698 for (BasicBlock::iterator I = IfBlock2->begin();!isa<TerminatorInst>(I);++I)
1699 if (!AggressiveInsts.count(I) && !isa<DbgInfoIntrinsic>(I)) {
1700 // This is not an aggressive instruction that we can promote.
1701 // Because of this, we won't be able to get rid of the control
1702 // flow, so the xform is not worth it.
1707 DEBUG(dbgs() << "FOUND IF CONDITION! " << *IfCond << " T: "
1708 << IfTrue->getName() << " F: " << IfFalse->getName() << "\n");
1710 // If we can still promote the PHI nodes after this gauntlet of tests,
1711 // do all of the PHI's now.
1712 Instruction *InsertPt = DomBlock->getTerminator();
1713 IRBuilder<true, NoFolder> Builder(InsertPt);
1715 // Move all 'aggressive' instructions, which are defined in the
1716 // conditional parts of the if's up to the dominating block.
1718 DomBlock->getInstList().splice(InsertPt,
1719 IfBlock1->getInstList(), IfBlock1->begin(),
1720 IfBlock1->getTerminator());
1722 DomBlock->getInstList().splice(InsertPt,
1723 IfBlock2->getInstList(), IfBlock2->begin(),
1724 IfBlock2->getTerminator());
1726 while (PHINode *PN = dyn_cast<PHINode>(BB->begin())) {
1727 // Change the PHI node into a select instruction.
1728 Value *TrueVal = PN->getIncomingValue(PN->getIncomingBlock(0) == IfFalse);
1729 Value *FalseVal = PN->getIncomingValue(PN->getIncomingBlock(0) == IfTrue);
1732 cast<SelectInst>(Builder.CreateSelect(IfCond, TrueVal, FalseVal, ""));
1733 PN->replaceAllUsesWith(NV);
1735 PN->eraseFromParent();
1738 // At this point, IfBlock1 and IfBlock2 are both empty, so our if statement
1739 // has been flattened. Change DomBlock to jump directly to our new block to
1740 // avoid other simplifycfg's kicking in on the diamond.
1741 TerminatorInst *OldTI = DomBlock->getTerminator();
1742 Builder.SetInsertPoint(OldTI);
1743 Builder.CreateBr(BB);
1744 OldTI->eraseFromParent();
1748 /// SimplifyCondBranchToTwoReturns - If we found a conditional branch that goes
1749 /// to two returning blocks, try to merge them together into one return,
1750 /// introducing a select if the return values disagree.
1751 static bool SimplifyCondBranchToTwoReturns(BranchInst *BI,
1752 IRBuilder<> &Builder) {
1753 assert(BI->isConditional() && "Must be a conditional branch");
1754 BasicBlock *TrueSucc = BI->getSuccessor(0);
1755 BasicBlock *FalseSucc = BI->getSuccessor(1);
1756 ReturnInst *TrueRet = cast<ReturnInst>(TrueSucc->getTerminator());
1757 ReturnInst *FalseRet = cast<ReturnInst>(FalseSucc->getTerminator());
1759 // Check to ensure both blocks are empty (just a return) or optionally empty
1760 // with PHI nodes. If there are other instructions, merging would cause extra
1761 // computation on one path or the other.
1762 if (!TrueSucc->getFirstNonPHIOrDbg()->isTerminator())
1764 if (!FalseSucc->getFirstNonPHIOrDbg()->isTerminator())
1767 Builder.SetInsertPoint(BI);
1768 // Okay, we found a branch that is going to two return nodes. If
1769 // there is no return value for this function, just change the
1770 // branch into a return.
1771 if (FalseRet->getNumOperands() == 0) {
1772 TrueSucc->removePredecessor(BI->getParent());
1773 FalseSucc->removePredecessor(BI->getParent());
1774 Builder.CreateRetVoid();
1775 EraseTerminatorInstAndDCECond(BI);
1779 // Otherwise, figure out what the true and false return values are
1780 // so we can insert a new select instruction.
1781 Value *TrueValue = TrueRet->getReturnValue();
1782 Value *FalseValue = FalseRet->getReturnValue();
1784 // Unwrap any PHI nodes in the return blocks.
1785 if (PHINode *TVPN = dyn_cast_or_null<PHINode>(TrueValue))
1786 if (TVPN->getParent() == TrueSucc)
1787 TrueValue = TVPN->getIncomingValueForBlock(BI->getParent());
1788 if (PHINode *FVPN = dyn_cast_or_null<PHINode>(FalseValue))
1789 if (FVPN->getParent() == FalseSucc)
1790 FalseValue = FVPN->getIncomingValueForBlock(BI->getParent());
1792 // In order for this transformation to be safe, we must be able to
1793 // unconditionally execute both operands to the return. This is
1794 // normally the case, but we could have a potentially-trapping
1795 // constant expression that prevents this transformation from being
1797 if (ConstantExpr *TCV = dyn_cast_or_null<ConstantExpr>(TrueValue))
1800 if (ConstantExpr *FCV = dyn_cast_or_null<ConstantExpr>(FalseValue))
1804 // Okay, we collected all the mapped values and checked them for sanity, and
1805 // defined to really do this transformation. First, update the CFG.
1806 TrueSucc->removePredecessor(BI->getParent());
1807 FalseSucc->removePredecessor(BI->getParent());
1809 // Insert select instructions where needed.
1810 Value *BrCond = BI->getCondition();
1812 // Insert a select if the results differ.
1813 if (TrueValue == FalseValue || isa<UndefValue>(FalseValue)) {
1814 } else if (isa<UndefValue>(TrueValue)) {
1815 TrueValue = FalseValue;
1817 TrueValue = Builder.CreateSelect(BrCond, TrueValue,
1818 FalseValue, "retval");
1822 Value *RI = !TrueValue ?
1823 Builder.CreateRetVoid() : Builder.CreateRet(TrueValue);
1827 DEBUG(dbgs() << "\nCHANGING BRANCH TO TWO RETURNS INTO SELECT:"
1828 << "\n " << *BI << "NewRet = " << *RI
1829 << "TRUEBLOCK: " << *TrueSucc << "FALSEBLOCK: "<< *FalseSucc);
1831 EraseTerminatorInstAndDCECond(BI);
1836 /// ExtractBranchMetadata - Given a conditional BranchInstruction, retrieve the
1837 /// probabilities of the branch taking each edge. Fills in the two APInt
1838 /// parameters and return true, or returns false if no or invalid metadata was
1840 static bool ExtractBranchMetadata(BranchInst *BI,
1841 uint64_t &ProbTrue, uint64_t &ProbFalse) {
1842 assert(BI->isConditional() &&
1843 "Looking for probabilities on unconditional branch?");
1844 MDNode *ProfileData = BI->getMetadata(LLVMContext::MD_prof);
1845 if (!ProfileData || ProfileData->getNumOperands() != 3) return false;
1846 ConstantInt *CITrue = dyn_cast<ConstantInt>(ProfileData->getOperand(1));
1847 ConstantInt *CIFalse = dyn_cast<ConstantInt>(ProfileData->getOperand(2));
1848 if (!CITrue || !CIFalse) return false;
1849 ProbTrue = CITrue->getValue().getZExtValue();
1850 ProbFalse = CIFalse->getValue().getZExtValue();
1854 /// checkCSEInPredecessor - Return true if the given instruction is available
1855 /// in its predecessor block. If yes, the instruction will be removed.
1857 static bool checkCSEInPredecessor(Instruction *Inst, BasicBlock *PB) {
1858 if (!isa<BinaryOperator>(Inst) && !isa<CmpInst>(Inst))
1860 for (BasicBlock::iterator I = PB->begin(), E = PB->end(); I != E; I++) {
1861 Instruction *PBI = &*I;
1862 // Check whether Inst and PBI generate the same value.
1863 if (Inst->isIdenticalTo(PBI)) {
1864 Inst->replaceAllUsesWith(PBI);
1865 Inst->eraseFromParent();
1872 /// FoldBranchToCommonDest - If this basic block is simple enough, and if a
1873 /// predecessor branches to us and one of our successors, fold the block into
1874 /// the predecessor and use logical operations to pick the right destination.
1875 bool llvm::FoldBranchToCommonDest(BranchInst *BI) {
1876 BasicBlock *BB = BI->getParent();
1878 Instruction *Cond = 0;
1879 if (BI->isConditional())
1880 Cond = dyn_cast<Instruction>(BI->getCondition());
1882 // For unconditional branch, check for a simple CFG pattern, where
1883 // BB has a single predecessor and BB's successor is also its predecessor's
1884 // successor. If such pattern exisits, check for CSE between BB and its
1886 if (BasicBlock *PB = BB->getSinglePredecessor())
1887 if (BranchInst *PBI = dyn_cast<BranchInst>(PB->getTerminator()))
1888 if (PBI->isConditional() &&
1889 (BI->getSuccessor(0) == PBI->getSuccessor(0) ||
1890 BI->getSuccessor(0) == PBI->getSuccessor(1))) {
1891 for (BasicBlock::iterator I = BB->begin(), E = BB->end();
1893 Instruction *Curr = I++;
1894 if (isa<CmpInst>(Curr)) {
1898 // Quit if we can't remove this instruction.
1899 if (!checkCSEInPredecessor(Curr, PB))
1908 if (Cond == 0 || (!isa<CmpInst>(Cond) && !isa<BinaryOperator>(Cond)) ||
1909 Cond->getParent() != BB || !Cond->hasOneUse())
1912 // Only allow this if the condition is a simple instruction that can be
1913 // executed unconditionally. It must be in the same block as the branch, and
1914 // must be at the front of the block.
1915 BasicBlock::iterator FrontIt = BB->front();
1917 // Ignore dbg intrinsics.
1918 while (isa<DbgInfoIntrinsic>(FrontIt)) ++FrontIt;
1920 // Allow a single instruction to be hoisted in addition to the compare
1921 // that feeds the branch. We later ensure that any values that _it_ uses
1922 // were also live in the predecessor, so that we don't unnecessarily create
1923 // register pressure or inhibit out-of-order execution.
1924 Instruction *BonusInst = 0;
1925 if (&*FrontIt != Cond &&
1926 FrontIt->hasOneUse() && *FrontIt->use_begin() == Cond &&
1927 isSafeToSpeculativelyExecute(FrontIt)) {
1928 BonusInst = &*FrontIt;
1931 // Ignore dbg intrinsics.
1932 while (isa<DbgInfoIntrinsic>(FrontIt)) ++FrontIt;
1935 // Only a single bonus inst is allowed.
1936 if (&*FrontIt != Cond)
1939 // Make sure the instruction after the condition is the cond branch.
1940 BasicBlock::iterator CondIt = Cond; ++CondIt;
1942 // Ingore dbg intrinsics.
1943 while (isa<DbgInfoIntrinsic>(CondIt)) ++CondIt;
1948 // Cond is known to be a compare or binary operator. Check to make sure that
1949 // neither operand is a potentially-trapping constant expression.
1950 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Cond->getOperand(0)))
1953 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Cond->getOperand(1)))
1957 // Finally, don't infinitely unroll conditional loops.
1958 BasicBlock *TrueDest = BI->getSuccessor(0);
1959 BasicBlock *FalseDest = (BI->isConditional()) ? BI->getSuccessor(1) : 0;
1960 if (TrueDest == BB || FalseDest == BB)
1963 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
1964 BasicBlock *PredBlock = *PI;
1965 BranchInst *PBI = dyn_cast<BranchInst>(PredBlock->getTerminator());
1967 // Check that we have two conditional branches. If there is a PHI node in
1968 // the common successor, verify that the same value flows in from both
1970 SmallVector<PHINode*, 4> PHIs;
1971 if (PBI == 0 || PBI->isUnconditional() ||
1972 (BI->isConditional() &&
1973 !SafeToMergeTerminators(BI, PBI)) ||
1974 (!BI->isConditional() &&
1975 !isProfitableToFoldUnconditional(BI, PBI, Cond, PHIs)))
1978 // Determine if the two branches share a common destination.
1979 Instruction::BinaryOps Opc = Instruction::BinaryOpsEnd;
1980 bool InvertPredCond = false;
1982 if (BI->isConditional()) {
1983 if (PBI->getSuccessor(0) == TrueDest)
1984 Opc = Instruction::Or;
1985 else if (PBI->getSuccessor(1) == FalseDest)
1986 Opc = Instruction::And;
1987 else if (PBI->getSuccessor(0) == FalseDest)
1988 Opc = Instruction::And, InvertPredCond = true;
1989 else if (PBI->getSuccessor(1) == TrueDest)
1990 Opc = Instruction::Or, InvertPredCond = true;
1994 if (PBI->getSuccessor(0) != TrueDest && PBI->getSuccessor(1) != TrueDest)
1998 // Ensure that any values used in the bonus instruction are also used
1999 // by the terminator of the predecessor. This means that those values
2000 // must already have been resolved, so we won't be inhibiting the
2001 // out-of-order core by speculating them earlier.
2003 // Collect the values used by the bonus inst
2004 SmallPtrSet<Value*, 4> UsedValues;
2005 for (Instruction::op_iterator OI = BonusInst->op_begin(),
2006 OE = BonusInst->op_end(); OI != OE; ++OI) {
2008 if (!isa<Constant>(V))
2009 UsedValues.insert(V);
2012 SmallVector<std::pair<Value*, unsigned>, 4> Worklist;
2013 Worklist.push_back(std::make_pair(PBI->getOperand(0), 0));
2015 // Walk up to four levels back up the use-def chain of the predecessor's
2016 // terminator to see if all those values were used. The choice of four
2017 // levels is arbitrary, to provide a compile-time-cost bound.
2018 while (!Worklist.empty()) {
2019 std::pair<Value*, unsigned> Pair = Worklist.back();
2020 Worklist.pop_back();
2022 if (Pair.second >= 4) continue;
2023 UsedValues.erase(Pair.first);
2024 if (UsedValues.empty()) break;
2026 if (Instruction *I = dyn_cast<Instruction>(Pair.first)) {
2027 for (Instruction::op_iterator OI = I->op_begin(), OE = I->op_end();
2029 Worklist.push_back(std::make_pair(OI->get(), Pair.second+1));
2033 if (!UsedValues.empty()) return false;
2036 DEBUG(dbgs() << "FOLDING BRANCH TO COMMON DEST:\n" << *PBI << *BB);
2037 IRBuilder<> Builder(PBI);
2039 // If we need to invert the condition in the pred block to match, do so now.
2040 if (InvertPredCond) {
2041 Value *NewCond = PBI->getCondition();
2043 if (NewCond->hasOneUse() && isa<CmpInst>(NewCond)) {
2044 CmpInst *CI = cast<CmpInst>(NewCond);
2045 CI->setPredicate(CI->getInversePredicate());
2047 NewCond = Builder.CreateNot(NewCond,
2048 PBI->getCondition()->getName()+".not");
2051 PBI->setCondition(NewCond);
2052 PBI->swapSuccessors();
2055 // If we have a bonus inst, clone it into the predecessor block.
2056 Instruction *NewBonus = 0;
2058 NewBonus = BonusInst->clone();
2059 PredBlock->getInstList().insert(PBI, NewBonus);
2060 NewBonus->takeName(BonusInst);
2061 BonusInst->setName(BonusInst->getName()+".old");
2064 // Clone Cond into the predecessor basic block, and or/and the
2065 // two conditions together.
2066 Instruction *New = Cond->clone();
2067 if (BonusInst) New->replaceUsesOfWith(BonusInst, NewBonus);
2068 PredBlock->getInstList().insert(PBI, New);
2069 New->takeName(Cond);
2070 Cond->setName(New->getName()+".old");
2072 if (BI->isConditional()) {
2073 Instruction *NewCond =
2074 cast<Instruction>(Builder.CreateBinOp(Opc, PBI->getCondition(),
2076 PBI->setCondition(NewCond);
2078 uint64_t PredTrueWeight, PredFalseWeight, SuccTrueWeight, SuccFalseWeight;
2079 bool PredHasWeights = ExtractBranchMetadata(PBI, PredTrueWeight,
2081 bool SuccHasWeights = ExtractBranchMetadata(BI, SuccTrueWeight,
2083 SmallVector<uint64_t, 8> NewWeights;
2085 if (PBI->getSuccessor(0) == BB) {
2086 if (PredHasWeights && SuccHasWeights) {
2087 // PBI: br i1 %x, BB, FalseDest
2088 // BI: br i1 %y, TrueDest, FalseDest
2089 //TrueWeight is TrueWeight for PBI * TrueWeight for BI.
2090 NewWeights.push_back(PredTrueWeight * SuccTrueWeight);
2091 //FalseWeight is FalseWeight for PBI * TotalWeight for BI +
2092 // TrueWeight for PBI * FalseWeight for BI.
2093 // We assume that total weights of a BranchInst can fit into 32 bits.
2094 // Therefore, we will not have overflow using 64-bit arithmetic.
2095 NewWeights.push_back(PredFalseWeight * (SuccFalseWeight +
2096 SuccTrueWeight) + PredTrueWeight * SuccFalseWeight);
2098 AddPredecessorToBlock(TrueDest, PredBlock, BB);
2099 PBI->setSuccessor(0, TrueDest);
2101 if (PBI->getSuccessor(1) == BB) {
2102 if (PredHasWeights && SuccHasWeights) {
2103 // PBI: br i1 %x, TrueDest, BB
2104 // BI: br i1 %y, TrueDest, FalseDest
2105 //TrueWeight is TrueWeight for PBI * TotalWeight for BI +
2106 // FalseWeight for PBI * TrueWeight for BI.
2107 NewWeights.push_back(PredTrueWeight * (SuccFalseWeight +
2108 SuccTrueWeight) + PredFalseWeight * SuccTrueWeight);
2109 //FalseWeight is FalseWeight for PBI * FalseWeight for BI.
2110 NewWeights.push_back(PredFalseWeight * SuccFalseWeight);
2112 AddPredecessorToBlock(FalseDest, PredBlock, BB);
2113 PBI->setSuccessor(1, FalseDest);
2115 if (NewWeights.size() == 2) {
2116 // Halve the weights if any of them cannot fit in an uint32_t
2117 FitWeights(NewWeights);
2119 SmallVector<uint32_t, 8> MDWeights(NewWeights.begin(),NewWeights.end());
2120 PBI->setMetadata(LLVMContext::MD_prof,
2121 MDBuilder(BI->getContext()).
2122 createBranchWeights(MDWeights));
2124 PBI->setMetadata(LLVMContext::MD_prof, NULL);
2126 // Update PHI nodes in the common successors.
2127 for (unsigned i = 0, e = PHIs.size(); i != e; ++i) {
2128 ConstantInt *PBI_C = cast<ConstantInt>(
2129 PHIs[i]->getIncomingValueForBlock(PBI->getParent()));
2130 assert(PBI_C->getType()->isIntegerTy(1));
2131 Instruction *MergedCond = 0;
2132 if (PBI->getSuccessor(0) == TrueDest) {
2133 // Create (PBI_Cond and PBI_C) or (!PBI_Cond and BI_Value)
2134 // PBI_C is true: PBI_Cond or (!PBI_Cond and BI_Value)
2135 // is false: !PBI_Cond and BI_Value
2136 Instruction *NotCond =
2137 cast<Instruction>(Builder.CreateNot(PBI->getCondition(),
2140 cast<Instruction>(Builder.CreateBinOp(Instruction::And,
2145 cast<Instruction>(Builder.CreateBinOp(Instruction::Or,
2146 PBI->getCondition(), MergedCond,
2149 // Create (PBI_Cond and BI_Value) or (!PBI_Cond and PBI_C)
2150 // PBI_C is true: (PBI_Cond and BI_Value) or (!PBI_Cond)
2151 // is false: PBI_Cond and BI_Value
2153 cast<Instruction>(Builder.CreateBinOp(Instruction::And,
2154 PBI->getCondition(), New,
2156 if (PBI_C->isOne()) {
2157 Instruction *NotCond =
2158 cast<Instruction>(Builder.CreateNot(PBI->getCondition(),
2161 cast<Instruction>(Builder.CreateBinOp(Instruction::Or,
2162 NotCond, MergedCond,
2167 PHIs[i]->setIncomingValue(PHIs[i]->getBasicBlockIndex(PBI->getParent()),
2170 // Change PBI from Conditional to Unconditional.
2171 BranchInst *New_PBI = BranchInst::Create(TrueDest, PBI);
2172 EraseTerminatorInstAndDCECond(PBI);
2176 // TODO: If BB is reachable from all paths through PredBlock, then we
2177 // could replace PBI's branch probabilities with BI's.
2179 // Copy any debug value intrinsics into the end of PredBlock.
2180 for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
2181 if (isa<DbgInfoIntrinsic>(*I))
2182 I->clone()->insertBefore(PBI);
2189 /// SimplifyCondBranchToCondBranch - If we have a conditional branch as a
2190 /// predecessor of another block, this function tries to simplify it. We know
2191 /// that PBI and BI are both conditional branches, and BI is in one of the
2192 /// successor blocks of PBI - PBI branches to BI.
2193 static bool SimplifyCondBranchToCondBranch(BranchInst *PBI, BranchInst *BI) {
2194 assert(PBI->isConditional() && BI->isConditional());
2195 BasicBlock *BB = BI->getParent();
2197 // If this block ends with a branch instruction, and if there is a
2198 // predecessor that ends on a branch of the same condition, make
2199 // this conditional branch redundant.
2200 if (PBI->getCondition() == BI->getCondition() &&
2201 PBI->getSuccessor(0) != PBI->getSuccessor(1)) {
2202 // Okay, the outcome of this conditional branch is statically
2203 // knowable. If this block had a single pred, handle specially.
2204 if (BB->getSinglePredecessor()) {
2205 // Turn this into a branch on constant.
2206 bool CondIsTrue = PBI->getSuccessor(0) == BB;
2207 BI->setCondition(ConstantInt::get(Type::getInt1Ty(BB->getContext()),
2209 return true; // Nuke the branch on constant.
2212 // Otherwise, if there are multiple predecessors, insert a PHI that merges
2213 // in the constant and simplify the block result. Subsequent passes of
2214 // simplifycfg will thread the block.
2215 if (BlockIsSimpleEnoughToThreadThrough(BB)) {
2216 pred_iterator PB = pred_begin(BB), PE = pred_end(BB);
2217 PHINode *NewPN = PHINode::Create(Type::getInt1Ty(BB->getContext()),
2218 std::distance(PB, PE),
2219 BI->getCondition()->getName() + ".pr",
2221 // Okay, we're going to insert the PHI node. Since PBI is not the only
2222 // predecessor, compute the PHI'd conditional value for all of the preds.
2223 // Any predecessor where the condition is not computable we keep symbolic.
2224 for (pred_iterator PI = PB; PI != PE; ++PI) {
2225 BasicBlock *P = *PI;
2226 if ((PBI = dyn_cast<BranchInst>(P->getTerminator())) &&
2227 PBI != BI && PBI->isConditional() &&
2228 PBI->getCondition() == BI->getCondition() &&
2229 PBI->getSuccessor(0) != PBI->getSuccessor(1)) {
2230 bool CondIsTrue = PBI->getSuccessor(0) == BB;
2231 NewPN->addIncoming(ConstantInt::get(Type::getInt1Ty(BB->getContext()),
2234 NewPN->addIncoming(BI->getCondition(), P);
2238 BI->setCondition(NewPN);
2243 // If this is a conditional branch in an empty block, and if any
2244 // predecessors is a conditional branch to one of our destinations,
2245 // fold the conditions into logical ops and one cond br.
2246 BasicBlock::iterator BBI = BB->begin();
2247 // Ignore dbg intrinsics.
2248 while (isa<DbgInfoIntrinsic>(BBI))
2254 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(BI->getCondition()))
2259 if (PBI->getSuccessor(0) == BI->getSuccessor(0))
2261 else if (PBI->getSuccessor(0) == BI->getSuccessor(1))
2262 PBIOp = 0, BIOp = 1;
2263 else if (PBI->getSuccessor(1) == BI->getSuccessor(0))
2264 PBIOp = 1, BIOp = 0;
2265 else if (PBI->getSuccessor(1) == BI->getSuccessor(1))
2270 // Check to make sure that the other destination of this branch
2271 // isn't BB itself. If so, this is an infinite loop that will
2272 // keep getting unwound.
2273 if (PBI->getSuccessor(PBIOp) == BB)
2276 // Do not perform this transformation if it would require
2277 // insertion of a large number of select instructions. For targets
2278 // without predication/cmovs, this is a big pessimization.
2279 BasicBlock *CommonDest = PBI->getSuccessor(PBIOp);
2281 unsigned NumPhis = 0;
2282 for (BasicBlock::iterator II = CommonDest->begin();
2283 isa<PHINode>(II); ++II, ++NumPhis)
2284 if (NumPhis > 2) // Disable this xform.
2287 // Finally, if everything is ok, fold the branches to logical ops.
2288 BasicBlock *OtherDest = BI->getSuccessor(BIOp ^ 1);
2290 DEBUG(dbgs() << "FOLDING BRs:" << *PBI->getParent()
2291 << "AND: " << *BI->getParent());
2294 // If OtherDest *is* BB, then BB is a basic block with a single conditional
2295 // branch in it, where one edge (OtherDest) goes back to itself but the other
2296 // exits. We don't *know* that the program avoids the infinite loop
2297 // (even though that seems likely). If we do this xform naively, we'll end up
2298 // recursively unpeeling the loop. Since we know that (after the xform is
2299 // done) that the block *is* infinite if reached, we just make it an obviously
2300 // infinite loop with no cond branch.
2301 if (OtherDest == BB) {
2302 // Insert it at the end of the function, because it's either code,
2303 // or it won't matter if it's hot. :)
2304 BasicBlock *InfLoopBlock = BasicBlock::Create(BB->getContext(),
2305 "infloop", BB->getParent());
2306 BranchInst::Create(InfLoopBlock, InfLoopBlock);
2307 OtherDest = InfLoopBlock;
2310 DEBUG(dbgs() << *PBI->getParent()->getParent());
2312 // BI may have other predecessors. Because of this, we leave
2313 // it alone, but modify PBI.
2315 // Make sure we get to CommonDest on True&True directions.
2316 Value *PBICond = PBI->getCondition();
2317 IRBuilder<true, NoFolder> Builder(PBI);
2319 PBICond = Builder.CreateNot(PBICond, PBICond->getName()+".not");
2321 Value *BICond = BI->getCondition();
2323 BICond = Builder.CreateNot(BICond, BICond->getName()+".not");
2325 // Merge the conditions.
2326 Value *Cond = Builder.CreateOr(PBICond, BICond, "brmerge");
2328 // Modify PBI to branch on the new condition to the new dests.
2329 PBI->setCondition(Cond);
2330 PBI->setSuccessor(0, CommonDest);
2331 PBI->setSuccessor(1, OtherDest);
2333 // Update branch weight for PBI.
2334 uint64_t PredTrueWeight, PredFalseWeight, SuccTrueWeight, SuccFalseWeight;
2335 bool PredHasWeights = ExtractBranchMetadata(PBI, PredTrueWeight,
2337 bool SuccHasWeights = ExtractBranchMetadata(BI, SuccTrueWeight,
2339 if (PredHasWeights && SuccHasWeights) {
2340 uint64_t PredCommon = PBIOp ? PredFalseWeight : PredTrueWeight;
2341 uint64_t PredOther = PBIOp ?PredTrueWeight : PredFalseWeight;
2342 uint64_t SuccCommon = BIOp ? SuccFalseWeight : SuccTrueWeight;
2343 uint64_t SuccOther = BIOp ? SuccTrueWeight : SuccFalseWeight;
2344 // The weight to CommonDest should be PredCommon * SuccTotal +
2345 // PredOther * SuccCommon.
2346 // The weight to OtherDest should be PredOther * SuccOther.
2347 SmallVector<uint64_t, 2> NewWeights;
2348 NewWeights.push_back(PredCommon * (SuccCommon + SuccOther) +
2349 PredOther * SuccCommon);
2350 NewWeights.push_back(PredOther * SuccOther);
2351 // Halve the weights if any of them cannot fit in an uint32_t
2352 FitWeights(NewWeights);
2354 SmallVector<uint32_t, 2> MDWeights(NewWeights.begin(),NewWeights.end());
2355 PBI->setMetadata(LLVMContext::MD_prof,
2356 MDBuilder(BI->getContext()).
2357 createBranchWeights(MDWeights));
2360 // OtherDest may have phi nodes. If so, add an entry from PBI's
2361 // block that are identical to the entries for BI's block.
2362 AddPredecessorToBlock(OtherDest, PBI->getParent(), BB);
2364 // We know that the CommonDest already had an edge from PBI to
2365 // it. If it has PHIs though, the PHIs may have different
2366 // entries for BB and PBI's BB. If so, insert a select to make
2369 for (BasicBlock::iterator II = CommonDest->begin();
2370 (PN = dyn_cast<PHINode>(II)); ++II) {
2371 Value *BIV = PN->getIncomingValueForBlock(BB);
2372 unsigned PBBIdx = PN->getBasicBlockIndex(PBI->getParent());
2373 Value *PBIV = PN->getIncomingValue(PBBIdx);
2375 // Insert a select in PBI to pick the right value.
2376 Value *NV = cast<SelectInst>
2377 (Builder.CreateSelect(PBICond, PBIV, BIV, PBIV->getName()+".mux"));
2378 PN->setIncomingValue(PBBIdx, NV);
2382 DEBUG(dbgs() << "INTO: " << *PBI->getParent());
2383 DEBUG(dbgs() << *PBI->getParent()->getParent());
2385 // This basic block is probably dead. We know it has at least
2386 // one fewer predecessor.
2390 // SimplifyTerminatorOnSelect - Simplifies a terminator by replacing it with a
2391 // branch to TrueBB if Cond is true or to FalseBB if Cond is false.
2392 // Takes care of updating the successors and removing the old terminator.
2393 // Also makes sure not to introduce new successors by assuming that edges to
2394 // non-successor TrueBBs and FalseBBs aren't reachable.
2395 static bool SimplifyTerminatorOnSelect(TerminatorInst *OldTerm, Value *Cond,
2396 BasicBlock *TrueBB, BasicBlock *FalseBB,
2397 uint32_t TrueWeight,
2398 uint32_t FalseWeight){
2399 // Remove any superfluous successor edges from the CFG.
2400 // First, figure out which successors to preserve.
2401 // If TrueBB and FalseBB are equal, only try to preserve one copy of that
2403 BasicBlock *KeepEdge1 = TrueBB;
2404 BasicBlock *KeepEdge2 = TrueBB != FalseBB ? FalseBB : 0;
2406 // Then remove the rest.
2407 for (unsigned I = 0, E = OldTerm->getNumSuccessors(); I != E; ++I) {
2408 BasicBlock *Succ = OldTerm->getSuccessor(I);
2409 // Make sure only to keep exactly one copy of each edge.
2410 if (Succ == KeepEdge1)
2412 else if (Succ == KeepEdge2)
2415 Succ->removePredecessor(OldTerm->getParent());
2418 IRBuilder<> Builder(OldTerm);
2419 Builder.SetCurrentDebugLocation(OldTerm->getDebugLoc());
2421 // Insert an appropriate new terminator.
2422 if ((KeepEdge1 == 0) && (KeepEdge2 == 0)) {
2423 if (TrueBB == FalseBB)
2424 // We were only looking for one successor, and it was present.
2425 // Create an unconditional branch to it.
2426 Builder.CreateBr(TrueBB);
2428 // We found both of the successors we were looking for.
2429 // Create a conditional branch sharing the condition of the select.
2430 BranchInst *NewBI = Builder.CreateCondBr(Cond, TrueBB, FalseBB);
2431 if (TrueWeight != FalseWeight)
2432 NewBI->setMetadata(LLVMContext::MD_prof,
2433 MDBuilder(OldTerm->getContext()).
2434 createBranchWeights(TrueWeight, FalseWeight));
2436 } else if (KeepEdge1 && (KeepEdge2 || TrueBB == FalseBB)) {
2437 // Neither of the selected blocks were successors, so this
2438 // terminator must be unreachable.
2439 new UnreachableInst(OldTerm->getContext(), OldTerm);
2441 // One of the selected values was a successor, but the other wasn't.
2442 // Insert an unconditional branch to the one that was found;
2443 // the edge to the one that wasn't must be unreachable.
2445 // Only TrueBB was found.
2446 Builder.CreateBr(TrueBB);
2448 // Only FalseBB was found.
2449 Builder.CreateBr(FalseBB);
2452 EraseTerminatorInstAndDCECond(OldTerm);
2456 // SimplifySwitchOnSelect - Replaces
2457 // (switch (select cond, X, Y)) on constant X, Y
2458 // with a branch - conditional if X and Y lead to distinct BBs,
2459 // unconditional otherwise.
2460 static bool SimplifySwitchOnSelect(SwitchInst *SI, SelectInst *Select) {
2461 // Check for constant integer values in the select.
2462 ConstantInt *TrueVal = dyn_cast<ConstantInt>(Select->getTrueValue());
2463 ConstantInt *FalseVal = dyn_cast<ConstantInt>(Select->getFalseValue());
2464 if (!TrueVal || !FalseVal)
2467 // Find the relevant condition and destinations.
2468 Value *Condition = Select->getCondition();
2469 BasicBlock *TrueBB = SI->findCaseValue(TrueVal).getCaseSuccessor();
2470 BasicBlock *FalseBB = SI->findCaseValue(FalseVal).getCaseSuccessor();
2472 // Get weight for TrueBB and FalseBB.
2473 uint32_t TrueWeight = 0, FalseWeight = 0;
2474 SmallVector<uint64_t, 8> Weights;
2475 bool HasWeights = HasBranchWeights(SI);
2477 GetBranchWeights(SI, Weights);
2478 if (Weights.size() == 1 + SI->getNumCases()) {
2479 TrueWeight = (uint32_t)Weights[SI->findCaseValue(TrueVal).
2480 getSuccessorIndex()];
2481 FalseWeight = (uint32_t)Weights[SI->findCaseValue(FalseVal).
2482 getSuccessorIndex()];
2486 // Perform the actual simplification.
2487 return SimplifyTerminatorOnSelect(SI, Condition, TrueBB, FalseBB,
2488 TrueWeight, FalseWeight);
2491 // SimplifyIndirectBrOnSelect - Replaces
2492 // (indirectbr (select cond, blockaddress(@fn, BlockA),
2493 // blockaddress(@fn, BlockB)))
2495 // (br cond, BlockA, BlockB).
2496 static bool SimplifyIndirectBrOnSelect(IndirectBrInst *IBI, SelectInst *SI) {
2497 // Check that both operands of the select are block addresses.
2498 BlockAddress *TBA = dyn_cast<BlockAddress>(SI->getTrueValue());
2499 BlockAddress *FBA = dyn_cast<BlockAddress>(SI->getFalseValue());
2503 // Extract the actual blocks.
2504 BasicBlock *TrueBB = TBA->getBasicBlock();
2505 BasicBlock *FalseBB = FBA->getBasicBlock();
2507 // Perform the actual simplification.
2508 return SimplifyTerminatorOnSelect(IBI, SI->getCondition(), TrueBB, FalseBB,
2512 /// TryToSimplifyUncondBranchWithICmpInIt - This is called when we find an icmp
2513 /// instruction (a seteq/setne with a constant) as the only instruction in a
2514 /// block that ends with an uncond branch. We are looking for a very specific
2515 /// pattern that occurs when "A == 1 || A == 2 || A == 3" gets simplified. In
2516 /// this case, we merge the first two "or's of icmp" into a switch, but then the
2517 /// default value goes to an uncond block with a seteq in it, we get something
2520 /// switch i8 %A, label %DEFAULT [ i8 1, label %end i8 2, label %end ]
2522 /// %tmp = icmp eq i8 %A, 92
2525 /// ... = phi i1 [ true, %entry ], [ %tmp, %DEFAULT ], [ true, %entry ]
2527 /// We prefer to split the edge to 'end' so that there is a true/false entry to
2528 /// the PHI, merging the third icmp into the switch.
2529 static bool TryToSimplifyUncondBranchWithICmpInIt(ICmpInst *ICI,
2530 const DataLayout *TD,
2531 IRBuilder<> &Builder) {
2532 BasicBlock *BB = ICI->getParent();
2534 // If the block has any PHIs in it or the icmp has multiple uses, it is too
2536 if (isa<PHINode>(BB->begin()) || !ICI->hasOneUse()) return false;
2538 Value *V = ICI->getOperand(0);
2539 ConstantInt *Cst = cast<ConstantInt>(ICI->getOperand(1));
2541 // The pattern we're looking for is where our only predecessor is a switch on
2542 // 'V' and this block is the default case for the switch. In this case we can
2543 // fold the compared value into the switch to simplify things.
2544 BasicBlock *Pred = BB->getSinglePredecessor();
2545 if (Pred == 0 || !isa<SwitchInst>(Pred->getTerminator())) return false;
2547 SwitchInst *SI = cast<SwitchInst>(Pred->getTerminator());
2548 if (SI->getCondition() != V)
2551 // If BB is reachable on a non-default case, then we simply know the value of
2552 // V in this block. Substitute it and constant fold the icmp instruction
2554 if (SI->getDefaultDest() != BB) {
2555 ConstantInt *VVal = SI->findCaseDest(BB);
2556 assert(VVal && "Should have a unique destination value");
2557 ICI->setOperand(0, VVal);
2559 if (Value *V = SimplifyInstruction(ICI, TD)) {
2560 ICI->replaceAllUsesWith(V);
2561 ICI->eraseFromParent();
2563 // BB is now empty, so it is likely to simplify away.
2564 return SimplifyCFG(BB) | true;
2567 // Ok, the block is reachable from the default dest. If the constant we're
2568 // comparing exists in one of the other edges, then we can constant fold ICI
2570 if (SI->findCaseValue(Cst) != SI->case_default()) {
2572 if (ICI->getPredicate() == ICmpInst::ICMP_EQ)
2573 V = ConstantInt::getFalse(BB->getContext());
2575 V = ConstantInt::getTrue(BB->getContext());
2577 ICI->replaceAllUsesWith(V);
2578 ICI->eraseFromParent();
2579 // BB is now empty, so it is likely to simplify away.
2580 return SimplifyCFG(BB) | true;
2583 // The use of the icmp has to be in the 'end' block, by the only PHI node in
2585 BasicBlock *SuccBlock = BB->getTerminator()->getSuccessor(0);
2586 PHINode *PHIUse = dyn_cast<PHINode>(ICI->use_back());
2587 if (PHIUse == 0 || PHIUse != &SuccBlock->front() ||
2588 isa<PHINode>(++BasicBlock::iterator(PHIUse)))
2591 // If the icmp is a SETEQ, then the default dest gets false, the new edge gets
2593 Constant *DefaultCst = ConstantInt::getTrue(BB->getContext());
2594 Constant *NewCst = ConstantInt::getFalse(BB->getContext());
2596 if (ICI->getPredicate() == ICmpInst::ICMP_EQ)
2597 std::swap(DefaultCst, NewCst);
2599 // Replace ICI (which is used by the PHI for the default value) with true or
2600 // false depending on if it is EQ or NE.
2601 ICI->replaceAllUsesWith(DefaultCst);
2602 ICI->eraseFromParent();
2604 // Okay, the switch goes to this block on a default value. Add an edge from
2605 // the switch to the merge point on the compared value.
2606 BasicBlock *NewBB = BasicBlock::Create(BB->getContext(), "switch.edge",
2607 BB->getParent(), BB);
2608 SmallVector<uint64_t, 8> Weights;
2609 bool HasWeights = HasBranchWeights(SI);
2611 GetBranchWeights(SI, Weights);
2612 if (Weights.size() == 1 + SI->getNumCases()) {
2613 // Split weight for default case to case for "Cst".
2614 Weights[0] = (Weights[0]+1) >> 1;
2615 Weights.push_back(Weights[0]);
2617 SmallVector<uint32_t, 8> MDWeights(Weights.begin(), Weights.end());
2618 SI->setMetadata(LLVMContext::MD_prof,
2619 MDBuilder(SI->getContext()).
2620 createBranchWeights(MDWeights));
2623 SI->addCase(Cst, NewBB);
2625 // NewBB branches to the phi block, add the uncond branch and the phi entry.
2626 Builder.SetInsertPoint(NewBB);
2627 Builder.SetCurrentDebugLocation(SI->getDebugLoc());
2628 Builder.CreateBr(SuccBlock);
2629 PHIUse->addIncoming(NewCst, NewBB);
2633 /// SimplifyBranchOnICmpChain - The specified branch is a conditional branch.
2634 /// Check to see if it is branching on an or/and chain of icmp instructions, and
2635 /// fold it into a switch instruction if so.
2636 static bool SimplifyBranchOnICmpChain(BranchInst *BI, const DataLayout *TD,
2637 IRBuilder<> &Builder) {
2638 Instruction *Cond = dyn_cast<Instruction>(BI->getCondition());
2639 if (Cond == 0) return false;
2642 // Change br (X == 0 | X == 1), T, F into a switch instruction.
2643 // If this is a bunch of seteq's or'd together, or if it's a bunch of
2644 // 'setne's and'ed together, collect them.
2646 std::vector<ConstantInt*> Values;
2647 bool TrueWhenEqual = true;
2648 Value *ExtraCase = 0;
2649 unsigned UsedICmps = 0;
2651 if (Cond->getOpcode() == Instruction::Or) {
2652 CompVal = GatherConstantCompares(Cond, Values, ExtraCase, TD, true,
2654 } else if (Cond->getOpcode() == Instruction::And) {
2655 CompVal = GatherConstantCompares(Cond, Values, ExtraCase, TD, false,
2657 TrueWhenEqual = false;
2660 // If we didn't have a multiply compared value, fail.
2661 if (CompVal == 0) return false;
2663 // Avoid turning single icmps into a switch.
2667 // There might be duplicate constants in the list, which the switch
2668 // instruction can't handle, remove them now.
2669 array_pod_sort(Values.begin(), Values.end(), ConstantIntSortPredicate);
2670 Values.erase(std::unique(Values.begin(), Values.end()), Values.end());
2672 // If Extra was used, we require at least two switch values to do the
2673 // transformation. A switch with one value is just an cond branch.
2674 if (ExtraCase && Values.size() < 2) return false;
2676 // TODO: Preserve branch weight metadata, similarly to how
2677 // FoldValueComparisonIntoPredecessors preserves it.
2679 // Figure out which block is which destination.
2680 BasicBlock *DefaultBB = BI->getSuccessor(1);
2681 BasicBlock *EdgeBB = BI->getSuccessor(0);
2682 if (!TrueWhenEqual) std::swap(DefaultBB, EdgeBB);
2684 BasicBlock *BB = BI->getParent();
2686 DEBUG(dbgs() << "Converting 'icmp' chain with " << Values.size()
2687 << " cases into SWITCH. BB is:\n" << *BB);
2689 // If there are any extra values that couldn't be folded into the switch
2690 // then we evaluate them with an explicit branch first. Split the block
2691 // right before the condbr to handle it.
2693 BasicBlock *NewBB = BB->splitBasicBlock(BI, "switch.early.test");
2694 // Remove the uncond branch added to the old block.
2695 TerminatorInst *OldTI = BB->getTerminator();
2696 Builder.SetInsertPoint(OldTI);
2699 Builder.CreateCondBr(ExtraCase, EdgeBB, NewBB);
2701 Builder.CreateCondBr(ExtraCase, NewBB, EdgeBB);
2703 OldTI->eraseFromParent();
2705 // If there are PHI nodes in EdgeBB, then we need to add a new entry to them
2706 // for the edge we just added.
2707 AddPredecessorToBlock(EdgeBB, BB, NewBB);
2709 DEBUG(dbgs() << " ** 'icmp' chain unhandled condition: " << *ExtraCase
2710 << "\nEXTRABB = " << *BB);
2714 Builder.SetInsertPoint(BI);
2715 // Convert pointer to int before we switch.
2716 if (CompVal->getType()->isPointerTy()) {
2717 assert(TD && "Cannot switch on pointer without DataLayout");
2718 CompVal = Builder.CreatePtrToInt(CompVal,
2719 TD->getIntPtrType(CompVal->getType()),
2723 // Create the new switch instruction now.
2724 SwitchInst *New = Builder.CreateSwitch(CompVal, DefaultBB, Values.size());
2726 // Add all of the 'cases' to the switch instruction.
2727 for (unsigned i = 0, e = Values.size(); i != e; ++i)
2728 New->addCase(Values[i], EdgeBB);
2730 // We added edges from PI to the EdgeBB. As such, if there were any
2731 // PHI nodes in EdgeBB, they need entries to be added corresponding to
2732 // the number of edges added.
2733 for (BasicBlock::iterator BBI = EdgeBB->begin();
2734 isa<PHINode>(BBI); ++BBI) {
2735 PHINode *PN = cast<PHINode>(BBI);
2736 Value *InVal = PN->getIncomingValueForBlock(BB);
2737 for (unsigned i = 0, e = Values.size()-1; i != e; ++i)
2738 PN->addIncoming(InVal, BB);
2741 // Erase the old branch instruction.
2742 EraseTerminatorInstAndDCECond(BI);
2744 DEBUG(dbgs() << " ** 'icmp' chain result is:\n" << *BB << '\n');
2748 bool SimplifyCFGOpt::SimplifyResume(ResumeInst *RI, IRBuilder<> &Builder) {
2749 // If this is a trivial landing pad that just continues unwinding the caught
2750 // exception then zap the landing pad, turning its invokes into calls.
2751 BasicBlock *BB = RI->getParent();
2752 LandingPadInst *LPInst = dyn_cast<LandingPadInst>(BB->getFirstNonPHI());
2753 if (RI->getValue() != LPInst)
2754 // Not a landing pad, or the resume is not unwinding the exception that
2755 // caused control to branch here.
2758 // Check that there are no other instructions except for debug intrinsics.
2759 BasicBlock::iterator I = LPInst, E = RI;
2761 if (!isa<DbgInfoIntrinsic>(I))
2764 // Turn all invokes that unwind here into calls and delete the basic block.
2765 for (pred_iterator PI = pred_begin(BB), PE = pred_end(BB); PI != PE;) {
2766 InvokeInst *II = cast<InvokeInst>((*PI++)->getTerminator());
2767 SmallVector<Value*, 8> Args(II->op_begin(), II->op_end() - 3);
2768 // Insert a call instruction before the invoke.
2769 CallInst *Call = CallInst::Create(II->getCalledValue(), Args, "", II);
2771 Call->setCallingConv(II->getCallingConv());
2772 Call->setAttributes(II->getAttributes());
2773 Call->setDebugLoc(II->getDebugLoc());
2775 // Anything that used the value produced by the invoke instruction now uses
2776 // the value produced by the call instruction. Note that we do this even
2777 // for void functions and calls with no uses so that the callgraph edge is
2779 II->replaceAllUsesWith(Call);
2780 BB->removePredecessor(II->getParent());
2782 // Insert a branch to the normal destination right before the invoke.
2783 BranchInst::Create(II->getNormalDest(), II);
2785 // Finally, delete the invoke instruction!
2786 II->eraseFromParent();
2789 // The landingpad is now unreachable. Zap it.
2790 BB->eraseFromParent();
2794 bool SimplifyCFGOpt::SimplifyReturn(ReturnInst *RI, IRBuilder<> &Builder) {
2795 BasicBlock *BB = RI->getParent();
2796 if (!BB->getFirstNonPHIOrDbg()->isTerminator()) return false;
2798 // Find predecessors that end with branches.
2799 SmallVector<BasicBlock*, 8> UncondBranchPreds;
2800 SmallVector<BranchInst*, 8> CondBranchPreds;
2801 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
2802 BasicBlock *P = *PI;
2803 TerminatorInst *PTI = P->getTerminator();
2804 if (BranchInst *BI = dyn_cast<BranchInst>(PTI)) {
2805 if (BI->isUnconditional())
2806 UncondBranchPreds.push_back(P);
2808 CondBranchPreds.push_back(BI);
2812 // If we found some, do the transformation!
2813 if (!UncondBranchPreds.empty() && DupRet) {
2814 while (!UncondBranchPreds.empty()) {
2815 BasicBlock *Pred = UncondBranchPreds.pop_back_val();
2816 DEBUG(dbgs() << "FOLDING: " << *BB
2817 << "INTO UNCOND BRANCH PRED: " << *Pred);
2818 (void)FoldReturnIntoUncondBranch(RI, BB, Pred);
2821 // If we eliminated all predecessors of the block, delete the block now.
2822 if (pred_begin(BB) == pred_end(BB))
2823 // We know there are no successors, so just nuke the block.
2824 BB->eraseFromParent();
2829 // Check out all of the conditional branches going to this return
2830 // instruction. If any of them just select between returns, change the
2831 // branch itself into a select/return pair.
2832 while (!CondBranchPreds.empty()) {
2833 BranchInst *BI = CondBranchPreds.pop_back_val();
2835 // Check to see if the non-BB successor is also a return block.
2836 if (isa<ReturnInst>(BI->getSuccessor(0)->getTerminator()) &&
2837 isa<ReturnInst>(BI->getSuccessor(1)->getTerminator()) &&
2838 SimplifyCondBranchToTwoReturns(BI, Builder))
2844 bool SimplifyCFGOpt::SimplifyUnreachable(UnreachableInst *UI) {
2845 BasicBlock *BB = UI->getParent();
2847 bool Changed = false;
2849 // If there are any instructions immediately before the unreachable that can
2850 // be removed, do so.
2851 while (UI != BB->begin()) {
2852 BasicBlock::iterator BBI = UI;
2854 // Do not delete instructions that can have side effects which might cause
2855 // the unreachable to not be reachable; specifically, calls and volatile
2856 // operations may have this effect.
2857 if (isa<CallInst>(BBI) && !isa<DbgInfoIntrinsic>(BBI)) break;
2859 if (BBI->mayHaveSideEffects()) {
2860 if (StoreInst *SI = dyn_cast<StoreInst>(BBI)) {
2861 if (SI->isVolatile())
2863 } else if (LoadInst *LI = dyn_cast<LoadInst>(BBI)) {
2864 if (LI->isVolatile())
2866 } else if (AtomicRMWInst *RMWI = dyn_cast<AtomicRMWInst>(BBI)) {
2867 if (RMWI->isVolatile())
2869 } else if (AtomicCmpXchgInst *CXI = dyn_cast<AtomicCmpXchgInst>(BBI)) {
2870 if (CXI->isVolatile())
2872 } else if (!isa<FenceInst>(BBI) && !isa<VAArgInst>(BBI) &&
2873 !isa<LandingPadInst>(BBI)) {
2876 // Note that deleting LandingPad's here is in fact okay, although it
2877 // involves a bit of subtle reasoning. If this inst is a LandingPad,
2878 // all the predecessors of this block will be the unwind edges of Invokes,
2879 // and we can therefore guarantee this block will be erased.
2882 // Delete this instruction (any uses are guaranteed to be dead)
2883 if (!BBI->use_empty())
2884 BBI->replaceAllUsesWith(UndefValue::get(BBI->getType()));
2885 BBI->eraseFromParent();
2889 // If the unreachable instruction is the first in the block, take a gander
2890 // at all of the predecessors of this instruction, and simplify them.
2891 if (&BB->front() != UI) return Changed;
2893 SmallVector<BasicBlock*, 8> Preds(pred_begin(BB), pred_end(BB));
2894 for (unsigned i = 0, e = Preds.size(); i != e; ++i) {
2895 TerminatorInst *TI = Preds[i]->getTerminator();
2896 IRBuilder<> Builder(TI);
2897 if (BranchInst *BI = dyn_cast<BranchInst>(TI)) {
2898 if (BI->isUnconditional()) {
2899 if (BI->getSuccessor(0) == BB) {
2900 new UnreachableInst(TI->getContext(), TI);
2901 TI->eraseFromParent();
2905 if (BI->getSuccessor(0) == BB) {
2906 Builder.CreateBr(BI->getSuccessor(1));
2907 EraseTerminatorInstAndDCECond(BI);
2908 } else if (BI->getSuccessor(1) == BB) {
2909 Builder.CreateBr(BI->getSuccessor(0));
2910 EraseTerminatorInstAndDCECond(BI);
2914 } else if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) {
2915 for (SwitchInst::CaseIt i = SI->case_begin(), e = SI->case_end();
2917 if (i.getCaseSuccessor() == BB) {
2918 BB->removePredecessor(SI->getParent());
2923 // If the default value is unreachable, figure out the most popular
2924 // destination and make it the default.
2925 if (SI->getDefaultDest() == BB) {
2926 std::map<BasicBlock*, std::pair<unsigned, unsigned> > Popularity;
2927 for (SwitchInst::CaseIt i = SI->case_begin(), e = SI->case_end();
2929 std::pair<unsigned, unsigned> &entry =
2930 Popularity[i.getCaseSuccessor()];
2931 if (entry.first == 0) {
2933 entry.second = i.getCaseIndex();
2939 // Find the most popular block.
2940 unsigned MaxPop = 0;
2941 unsigned MaxIndex = 0;
2942 BasicBlock *MaxBlock = 0;
2943 for (std::map<BasicBlock*, std::pair<unsigned, unsigned> >::iterator
2944 I = Popularity.begin(), E = Popularity.end(); I != E; ++I) {
2945 if (I->second.first > MaxPop ||
2946 (I->second.first == MaxPop && MaxIndex > I->second.second)) {
2947 MaxPop = I->second.first;
2948 MaxIndex = I->second.second;
2949 MaxBlock = I->first;
2953 // Make this the new default, allowing us to delete any explicit
2955 SI->setDefaultDest(MaxBlock);
2958 // If MaxBlock has phinodes in it, remove MaxPop-1 entries from
2960 if (isa<PHINode>(MaxBlock->begin()))
2961 for (unsigned i = 0; i != MaxPop-1; ++i)
2962 MaxBlock->removePredecessor(SI->getParent());
2964 for (SwitchInst::CaseIt i = SI->case_begin(), e = SI->case_end();
2966 if (i.getCaseSuccessor() == MaxBlock) {
2972 } else if (InvokeInst *II = dyn_cast<InvokeInst>(TI)) {
2973 if (II->getUnwindDest() == BB) {
2974 // Convert the invoke to a call instruction. This would be a good
2975 // place to note that the call does not throw though.
2976 BranchInst *BI = Builder.CreateBr(II->getNormalDest());
2977 II->removeFromParent(); // Take out of symbol table
2979 // Insert the call now...
2980 SmallVector<Value*, 8> Args(II->op_begin(), II->op_end()-3);
2981 Builder.SetInsertPoint(BI);
2982 CallInst *CI = Builder.CreateCall(II->getCalledValue(),
2983 Args, II->getName());
2984 CI->setCallingConv(II->getCallingConv());
2985 CI->setAttributes(II->getAttributes());
2986 // If the invoke produced a value, the call does now instead.
2987 II->replaceAllUsesWith(CI);
2994 // If this block is now dead, remove it.
2995 if (pred_begin(BB) == pred_end(BB) &&
2996 BB != &BB->getParent()->getEntryBlock()) {
2997 // We know there are no successors, so just nuke the block.
2998 BB->eraseFromParent();
3005 /// TurnSwitchRangeIntoICmp - Turns a switch with that contains only a
3006 /// integer range comparison into a sub, an icmp and a branch.
3007 static bool TurnSwitchRangeIntoICmp(SwitchInst *SI, IRBuilder<> &Builder) {
3008 assert(SI->getNumCases() > 1 && "Degenerate switch?");
3010 // Make sure all cases point to the same destination and gather the values.
3011 SmallVector<ConstantInt *, 16> Cases;
3012 SwitchInst::CaseIt I = SI->case_begin();
3013 Cases.push_back(I.getCaseValue());
3014 SwitchInst::CaseIt PrevI = I++;
3015 for (SwitchInst::CaseIt E = SI->case_end(); I != E; PrevI = I++) {
3016 if (PrevI.getCaseSuccessor() != I.getCaseSuccessor())
3018 Cases.push_back(I.getCaseValue());
3020 assert(Cases.size() == SI->getNumCases() && "Not all cases gathered");
3022 // Sort the case values, then check if they form a range we can transform.
3023 array_pod_sort(Cases.begin(), Cases.end(), ConstantIntSortPredicate);
3024 for (unsigned I = 1, E = Cases.size(); I != E; ++I) {
3025 if (Cases[I-1]->getValue() != Cases[I]->getValue()+1)
3029 Constant *Offset = ConstantExpr::getNeg(Cases.back());
3030 Constant *NumCases = ConstantInt::get(Offset->getType(), SI->getNumCases());
3032 Value *Sub = SI->getCondition();
3033 if (!Offset->isNullValue())
3034 Sub = Builder.CreateAdd(Sub, Offset, Sub->getName()+".off");
3035 Value *Cmp = Builder.CreateICmpULT(Sub, NumCases, "switch");
3036 BranchInst *NewBI = Builder.CreateCondBr(
3037 Cmp, SI->case_begin().getCaseSuccessor(), SI->getDefaultDest());
3039 // Update weight for the newly-created conditional branch.
3040 SmallVector<uint64_t, 8> Weights;
3041 bool HasWeights = HasBranchWeights(SI);
3043 GetBranchWeights(SI, Weights);
3044 if (Weights.size() == 1 + SI->getNumCases()) {
3045 // Combine all weights for the cases to be the true weight of NewBI.
3046 // We assume that the sum of all weights for a Terminator can fit into 32
3048 uint32_t NewTrueWeight = 0;
3049 for (unsigned I = 1, E = Weights.size(); I != E; ++I)
3050 NewTrueWeight += (uint32_t)Weights[I];
3051 NewBI->setMetadata(LLVMContext::MD_prof,
3052 MDBuilder(SI->getContext()).
3053 createBranchWeights(NewTrueWeight,
3054 (uint32_t)Weights[0]));
3058 // Prune obsolete incoming values off the successor's PHI nodes.
3059 for (BasicBlock::iterator BBI = SI->case_begin().getCaseSuccessor()->begin();
3060 isa<PHINode>(BBI); ++BBI) {
3061 for (unsigned I = 0, E = SI->getNumCases()-1; I != E; ++I)
3062 cast<PHINode>(BBI)->removeIncomingValue(SI->getParent());
3064 SI->eraseFromParent();
3069 /// EliminateDeadSwitchCases - Compute masked bits for the condition of a switch
3070 /// and use it to remove dead cases.
3071 static bool EliminateDeadSwitchCases(SwitchInst *SI) {
3072 Value *Cond = SI->getCondition();
3073 unsigned Bits = cast<IntegerType>(Cond->getType())->getBitWidth();
3074 APInt KnownZero(Bits, 0), KnownOne(Bits, 0);
3075 ComputeMaskedBits(Cond, KnownZero, KnownOne);
3077 // Gather dead cases.
3078 SmallVector<ConstantInt*, 8> DeadCases;
3079 for (SwitchInst::CaseIt I = SI->case_begin(), E = SI->case_end(); I != E; ++I) {
3080 if ((I.getCaseValue()->getValue() & KnownZero) != 0 ||
3081 (I.getCaseValue()->getValue() & KnownOne) != KnownOne) {
3082 DeadCases.push_back(I.getCaseValue());
3083 DEBUG(dbgs() << "SimplifyCFG: switch case '"
3084 << I.getCaseValue() << "' is dead.\n");
3088 SmallVector<uint64_t, 8> Weights;
3089 bool HasWeight = HasBranchWeights(SI);
3091 GetBranchWeights(SI, Weights);
3092 HasWeight = (Weights.size() == 1 + SI->getNumCases());
3095 // Remove dead cases from the switch.
3096 for (unsigned I = 0, E = DeadCases.size(); I != E; ++I) {
3097 SwitchInst::CaseIt Case = SI->findCaseValue(DeadCases[I]);
3098 assert(Case != SI->case_default() &&
3099 "Case was not found. Probably mistake in DeadCases forming.");
3101 std::swap(Weights[Case.getCaseIndex()+1], Weights.back());
3105 // Prune unused values from PHI nodes.
3106 Case.getCaseSuccessor()->removePredecessor(SI->getParent());
3107 SI->removeCase(Case);
3110 SmallVector<uint32_t, 8> MDWeights(Weights.begin(), Weights.end());
3111 SI->setMetadata(LLVMContext::MD_prof,
3112 MDBuilder(SI->getParent()->getContext()).
3113 createBranchWeights(MDWeights));
3116 return !DeadCases.empty();
3119 /// FindPHIForConditionForwarding - If BB would be eligible for simplification
3120 /// by TryToSimplifyUncondBranchFromEmptyBlock (i.e. it is empty and terminated
3121 /// by an unconditional branch), look at the phi node for BB in the successor
3122 /// block and see if the incoming value is equal to CaseValue. If so, return
3123 /// the phi node, and set PhiIndex to BB's index in the phi node.
3124 static PHINode *FindPHIForConditionForwarding(ConstantInt *CaseValue,
3127 if (BB->getFirstNonPHIOrDbg() != BB->getTerminator())
3128 return NULL; // BB must be empty to be a candidate for simplification.
3129 if (!BB->getSinglePredecessor())
3130 return NULL; // BB must be dominated by the switch.
3132 BranchInst *Branch = dyn_cast<BranchInst>(BB->getTerminator());
3133 if (!Branch || !Branch->isUnconditional())
3134 return NULL; // Terminator must be unconditional branch.
3136 BasicBlock *Succ = Branch->getSuccessor(0);
3138 BasicBlock::iterator I = Succ->begin();
3139 while (PHINode *PHI = dyn_cast<PHINode>(I++)) {
3140 int Idx = PHI->getBasicBlockIndex(BB);
3141 assert(Idx >= 0 && "PHI has no entry for predecessor?");
3143 Value *InValue = PHI->getIncomingValue(Idx);
3144 if (InValue != CaseValue) continue;
3153 /// ForwardSwitchConditionToPHI - Try to forward the condition of a switch
3154 /// instruction to a phi node dominated by the switch, if that would mean that
3155 /// some of the destination blocks of the switch can be folded away.
3156 /// Returns true if a change is made.
3157 static bool ForwardSwitchConditionToPHI(SwitchInst *SI) {
3158 typedef DenseMap<PHINode*, SmallVector<int,4> > ForwardingNodesMap;
3159 ForwardingNodesMap ForwardingNodes;
3161 for (SwitchInst::CaseIt I = SI->case_begin(), E = SI->case_end(); I != E; ++I) {
3162 ConstantInt *CaseValue = I.getCaseValue();
3163 BasicBlock *CaseDest = I.getCaseSuccessor();
3166 PHINode *PHI = FindPHIForConditionForwarding(CaseValue, CaseDest,
3170 ForwardingNodes[PHI].push_back(PhiIndex);
3173 bool Changed = false;
3175 for (ForwardingNodesMap::iterator I = ForwardingNodes.begin(),
3176 E = ForwardingNodes.end(); I != E; ++I) {
3177 PHINode *Phi = I->first;
3178 SmallVector<int,4> &Indexes = I->second;
3180 if (Indexes.size() < 2) continue;
3182 for (size_t I = 0, E = Indexes.size(); I != E; ++I)
3183 Phi->setIncomingValue(Indexes[I], SI->getCondition());
3190 /// ValidLookupTableConstant - Return true if the backend will be able to handle
3191 /// initializing an array of constants like C.
3192 static bool ValidLookupTableConstant(Constant *C) {
3193 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C))
3194 return CE->isGEPWithNoNotionalOverIndexing();
3196 return isa<ConstantFP>(C) ||
3197 isa<ConstantInt>(C) ||
3198 isa<ConstantPointerNull>(C) ||
3199 isa<GlobalValue>(C) ||
3203 /// ConstantFold - Try to fold instruction I into a constant. This works for
3204 /// simple instructions such as binary operations where both operands are
3205 /// constant or can be replaced by constants from the ConstantPool. Returns the
3206 /// resulting constant on success, NULL otherwise.
3207 static Constant* ConstantFold(Instruction *I,
3208 const SmallDenseMap<Value*, Constant*>& ConstantPool) {
3209 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(I)) {
3210 Constant *A = dyn_cast<Constant>(BO->getOperand(0));
3211 if (!A) A = ConstantPool.lookup(BO->getOperand(0));
3212 if (!A) return NULL;
3214 Constant *B = dyn_cast<Constant>(BO->getOperand(1));
3215 if (!B) B = ConstantPool.lookup(BO->getOperand(1));
3216 if (!B) return NULL;
3218 Constant *C = ConstantExpr::get(BO->getOpcode(), A, B);
3222 if (CmpInst *Cmp = dyn_cast<CmpInst>(I)) {
3223 Constant *A = dyn_cast<Constant>(I->getOperand(0));
3224 if (!A) A = ConstantPool.lookup(I->getOperand(0));
3225 if (!A) return NULL;
3227 Constant *B = dyn_cast<Constant>(I->getOperand(1));
3228 if (!B) B = ConstantPool.lookup(I->getOperand(1));
3229 if (!B) return NULL;
3231 Constant *C = ConstantExpr::getCompare(Cmp->getPredicate(), A, B);
3235 if (SelectInst *Select = dyn_cast<SelectInst>(I)) {
3236 Constant *A = dyn_cast<Constant>(Select->getCondition());
3237 if (!A) A = ConstantPool.lookup(Select->getCondition());
3238 if (!A) return NULL;
3241 if (A->isAllOnesValue()) Res = Select->getTrueValue();
3242 else if (A->isNullValue()) Res = Select->getFalseValue();
3245 Constant *C = dyn_cast<Constant>(Res);
3246 if (!C) C = ConstantPool.lookup(Res);
3247 if (!C) return NULL;
3251 if (CastInst *Cast = dyn_cast<CastInst>(I)) {
3252 Constant *A = dyn_cast<Constant>(I->getOperand(0));
3253 if (!A) A = ConstantPool.lookup(I->getOperand(0));
3254 if (!A) return NULL;
3256 Constant *C = ConstantExpr::getCast(Cast->getOpcode(), A, Cast->getDestTy());
3263 /// GetCaseResults - Try to determine the resulting constant values in phi nodes
3264 /// at the common destination basic block, *CommonDest, for one of the case
3265 /// destionations CaseDest corresponding to value CaseVal (NULL for the default
3266 /// case), of a switch instruction SI.
3267 static bool GetCaseResults(SwitchInst *SI,
3268 ConstantInt *CaseVal,
3269 BasicBlock *CaseDest,
3270 BasicBlock **CommonDest,
3271 SmallVector<std::pair<PHINode*,Constant*>, 4> &Res) {
3272 // The block from which we enter the common destination.
3273 BasicBlock *Pred = SI->getParent();
3275 // If CaseDest is empty except for some side-effect free instructions through
3276 // which we can constant-propagate the CaseVal, continue to its successor.
3277 SmallDenseMap<Value*, Constant*> ConstantPool;
3278 ConstantPool.insert(std::make_pair(SI->getCondition(), CaseVal));
3279 for (BasicBlock::iterator I = CaseDest->begin(), E = CaseDest->end(); I != E;
3281 if (TerminatorInst *T = dyn_cast<TerminatorInst>(I)) {
3282 // If the terminator is a simple branch, continue to the next block.
3283 if (T->getNumSuccessors() != 1)
3286 CaseDest = T->getSuccessor(0);
3287 } else if (isa<DbgInfoIntrinsic>(I)) {
3288 // Skip debug intrinsic.
3290 } else if (Constant *C = ConstantFold(I, ConstantPool)) {
3291 // Instruction is side-effect free and constant.
3292 ConstantPool.insert(std::make_pair(I, C));
3298 // If we did not have a CommonDest before, use the current one.
3300 *CommonDest = CaseDest;
3301 // If the destination isn't the common one, abort.
3302 if (CaseDest != *CommonDest)
3305 // Get the values for this case from phi nodes in the destination block.
3306 BasicBlock::iterator I = (*CommonDest)->begin();
3307 while (PHINode *PHI = dyn_cast<PHINode>(I++)) {
3308 int Idx = PHI->getBasicBlockIndex(Pred);
3312 Constant *ConstVal = dyn_cast<Constant>(PHI->getIncomingValue(Idx));
3314 ConstVal = ConstantPool.lookup(PHI->getIncomingValue(Idx));
3318 // Note: If the constant comes from constant-propagating the case value
3319 // through the CaseDest basic block, it will be safe to remove the
3320 // instructions in that block. They cannot be used (except in the phi nodes
3321 // we visit) outside CaseDest, because that block does not dominate its
3322 // successor. If it did, we would not be in this phi node.
3324 // Be conservative about which kinds of constants we support.
3325 if (!ValidLookupTableConstant(ConstVal))
3328 Res.push_back(std::make_pair(PHI, ConstVal));
3335 /// SwitchLookupTable - This class represents a lookup table that can be used
3336 /// to replace a switch.
3337 class SwitchLookupTable {
3339 /// SwitchLookupTable - Create a lookup table to use as a switch replacement
3340 /// with the contents of Values, using DefaultValue to fill any holes in the
3342 SwitchLookupTable(Module &M,
3344 ConstantInt *Offset,
3345 const SmallVector<std::pair<ConstantInt*, Constant*>, 4>& Values,
3346 Constant *DefaultValue,
3347 const DataLayout *TD);
3349 /// BuildLookup - Build instructions with Builder to retrieve the value at
3350 /// the position given by Index in the lookup table.
3351 Value *BuildLookup(Value *Index, IRBuilder<> &Builder);
3353 /// WouldFitInRegister - Return true if a table with TableSize elements of
3354 /// type ElementType would fit in a target-legal register.
3355 static bool WouldFitInRegister(const DataLayout *TD,
3357 const Type *ElementType);
3360 // Depending on the contents of the table, it can be represented in
3363 // For tables where each element contains the same value, we just have to
3364 // store that single value and return it for each lookup.
3367 // For small tables with integer elements, we can pack them into a bitmap
3368 // that fits into a target-legal register. Values are retrieved by
3369 // shift and mask operations.
3372 // The table is stored as an array of values. Values are retrieved by load
3373 // instructions from the table.
3377 // For SingleValueKind, this is the single value.
3378 Constant *SingleValue;
3380 // For BitMapKind, this is the bitmap.
3381 ConstantInt *BitMap;
3382 IntegerType *BitMapElementTy;
3384 // For ArrayKind, this is the array.
3385 GlobalVariable *Array;
3389 SwitchLookupTable::SwitchLookupTable(Module &M,
3391 ConstantInt *Offset,
3392 const SmallVector<std::pair<ConstantInt*, Constant*>, 4>& Values,
3393 Constant *DefaultValue,
3394 const DataLayout *TD) {
3395 assert(Values.size() && "Can't build lookup table without values!");
3396 assert(TableSize >= Values.size() && "Can't fit values in table!");
3398 // If all values in the table are equal, this is that value.
3399 SingleValue = Values.begin()->second;
3401 // Build up the table contents.
3402 SmallVector<Constant*, 64> TableContents(TableSize);
3403 for (size_t I = 0, E = Values.size(); I != E; ++I) {
3404 ConstantInt *CaseVal = Values[I].first;
3405 Constant *CaseRes = Values[I].second;
3406 assert(CaseRes->getType() == DefaultValue->getType());
3408 uint64_t Idx = (CaseVal->getValue() - Offset->getValue())
3410 TableContents[Idx] = CaseRes;
3412 if (CaseRes != SingleValue)
3416 // Fill in any holes in the table with the default result.
3417 if (Values.size() < TableSize) {
3418 for (uint64_t I = 0; I < TableSize; ++I) {
3419 if (!TableContents[I])
3420 TableContents[I] = DefaultValue;
3423 if (DefaultValue != SingleValue)
3427 // If each element in the table contains the same value, we only need to store
3428 // that single value.
3430 Kind = SingleValueKind;
3434 // If the type is integer and the table fits in a register, build a bitmap.
3435 if (WouldFitInRegister(TD, TableSize, DefaultValue->getType())) {
3436 IntegerType *IT = cast<IntegerType>(DefaultValue->getType());
3437 APInt TableInt(TableSize * IT->getBitWidth(), 0);
3438 for (uint64_t I = TableSize; I > 0; --I) {
3439 TableInt <<= IT->getBitWidth();
3440 // Insert values into the bitmap. Undef values are set to zero.
3441 if (!isa<UndefValue>(TableContents[I - 1])) {
3442 ConstantInt *Val = cast<ConstantInt>(TableContents[I - 1]);
3443 TableInt |= Val->getValue().zext(TableInt.getBitWidth());
3446 BitMap = ConstantInt::get(M.getContext(), TableInt);
3447 BitMapElementTy = IT;
3453 // Store the table in an array.
3454 ArrayType *ArrayTy = ArrayType::get(DefaultValue->getType(), TableSize);
3455 Constant *Initializer = ConstantArray::get(ArrayTy, TableContents);
3457 Array = new GlobalVariable(M, ArrayTy, /*constant=*/ true,
3458 GlobalVariable::PrivateLinkage,
3461 Array->setUnnamedAddr(true);
3465 Value *SwitchLookupTable::BuildLookup(Value *Index, IRBuilder<> &Builder) {
3467 case SingleValueKind:
3470 // Type of the bitmap (e.g. i59).
3471 IntegerType *MapTy = BitMap->getType();
3473 // Cast Index to the same type as the bitmap.
3474 // Note: The Index is <= the number of elements in the table, so
3475 // truncating it to the width of the bitmask is safe.
3476 Value *ShiftAmt = Builder.CreateZExtOrTrunc(Index, MapTy, "switch.cast");
3478 // Multiply the shift amount by the element width.
3479 ShiftAmt = Builder.CreateMul(ShiftAmt,
3480 ConstantInt::get(MapTy, BitMapElementTy->getBitWidth()),
3484 Value *DownShifted = Builder.CreateLShr(BitMap, ShiftAmt,
3485 "switch.downshift");
3487 return Builder.CreateTrunc(DownShifted, BitMapElementTy,
3491 Value *GEPIndices[] = { Builder.getInt32(0), Index };
3492 Value *GEP = Builder.CreateInBoundsGEP(Array, GEPIndices,
3494 return Builder.CreateLoad(GEP, "switch.load");
3497 llvm_unreachable("Unknown lookup table kind!");
3500 bool SwitchLookupTable::WouldFitInRegister(const DataLayout *TD,
3502 const Type *ElementType) {
3505 const IntegerType *IT = dyn_cast<IntegerType>(ElementType);
3508 // FIXME: If the type is wider than it needs to be, e.g. i8 but all values
3509 // are <= 15, we could try to narrow the type.
3511 // Avoid overflow, fitsInLegalInteger uses unsigned int for the width.
3512 if (TableSize >= UINT_MAX/IT->getBitWidth())
3514 return TD->fitsInLegalInteger(TableSize * IT->getBitWidth());
3517 /// ShouldBuildLookupTable - Determine whether a lookup table should be built
3518 /// for this switch, based on the number of caes, size of the table and the
3519 /// types of the results.
3520 static bool ShouldBuildLookupTable(SwitchInst *SI,
3522 const DataLayout *TD,
3523 const SmallDenseMap<PHINode*, Type*>& ResultTypes) {
3524 // The table density should be at least 40%. This is the same criterion as for
3525 // jump tables, see SelectionDAGBuilder::handleJTSwitchCase.
3526 // FIXME: Find the best cut-off.
3527 if (SI->getNumCases() > TableSize || TableSize >= UINT64_MAX / 10)
3528 return false; // TableSize overflowed, or mul below might overflow.
3529 if (SI->getNumCases() * 10 >= TableSize * 4)
3532 // If each table would fit in a register, we should build it anyway.
3533 for (SmallDenseMap<PHINode*, Type*>::const_iterator I = ResultTypes.begin(),
3534 E = ResultTypes.end(); I != E; ++I) {
3535 if (!SwitchLookupTable::WouldFitInRegister(TD, TableSize, I->second))
3541 /// SwitchToLookupTable - If the switch is only used to initialize one or more
3542 /// phi nodes in a common successor block with different constant values,
3543 /// replace the switch with lookup tables.
3544 static bool SwitchToLookupTable(SwitchInst *SI,
3545 IRBuilder<> &Builder,
3546 const DataLayout* TD,
3547 const TargetTransformInfo *TTI) {
3548 assert(SI->getNumCases() > 1 && "Degenerate switch?");
3550 if (TTI && !TTI->getScalarTargetTransformInfo()->shouldBuildLookupTables())
3553 // FIXME: Handle unreachable cases.
3555 // FIXME: If the switch is too sparse for a lookup table, perhaps we could
3556 // split off a dense part and build a lookup table for that.
3558 // FIXME: This creates arrays of GEPs to constant strings, which means each
3559 // GEP needs a runtime relocation in PIC code. We should just build one big
3560 // string and lookup indices into that.
3562 // Ignore the switch if the number of cases is too small.
3563 // This is similar to the check when building jump tables in
3564 // SelectionDAGBuilder::handleJTSwitchCase.
3565 // FIXME: Determine the best cut-off.
3566 if (SI->getNumCases() < 4)
3569 // Figure out the corresponding result for each case value and phi node in the
3570 // common destination, as well as the the min and max case values.
3571 assert(SI->case_begin() != SI->case_end());
3572 SwitchInst::CaseIt CI = SI->case_begin();
3573 ConstantInt *MinCaseVal = CI.getCaseValue();
3574 ConstantInt *MaxCaseVal = CI.getCaseValue();
3576 BasicBlock *CommonDest = NULL;
3577 typedef SmallVector<std::pair<ConstantInt*, Constant*>, 4> ResultListTy;
3578 SmallDenseMap<PHINode*, ResultListTy> ResultLists;
3579 SmallDenseMap<PHINode*, Constant*> DefaultResults;
3580 SmallDenseMap<PHINode*, Type*> ResultTypes;
3581 SmallVector<PHINode*, 4> PHIs;
3583 for (SwitchInst::CaseIt E = SI->case_end(); CI != E; ++CI) {
3584 ConstantInt *CaseVal = CI.getCaseValue();
3585 if (CaseVal->getValue().slt(MinCaseVal->getValue()))
3586 MinCaseVal = CaseVal;
3587 if (CaseVal->getValue().sgt(MaxCaseVal->getValue()))
3588 MaxCaseVal = CaseVal;
3590 // Resulting value at phi nodes for this case value.
3591 typedef SmallVector<std::pair<PHINode*, Constant*>, 4> ResultsTy;
3593 if (!GetCaseResults(SI, CaseVal, CI.getCaseSuccessor(), &CommonDest,
3597 // Append the result from this case to the list for each phi.
3598 for (ResultsTy::iterator I = Results.begin(), E = Results.end(); I!=E; ++I) {
3599 if (!ResultLists.count(I->first))
3600 PHIs.push_back(I->first);
3601 ResultLists[I->first].push_back(std::make_pair(CaseVal, I->second));
3605 // Get the resulting values for the default case.
3606 SmallVector<std::pair<PHINode*, Constant*>, 4> DefaultResultsList;
3607 if (!GetCaseResults(SI, NULL, SI->getDefaultDest(), &CommonDest,
3608 DefaultResultsList))
3610 for (size_t I = 0, E = DefaultResultsList.size(); I != E; ++I) {
3611 PHINode *PHI = DefaultResultsList[I].first;
3612 Constant *Result = DefaultResultsList[I].second;
3613 DefaultResults[PHI] = Result;
3614 ResultTypes[PHI] = Result->getType();
3617 APInt RangeSpread = MaxCaseVal->getValue() - MinCaseVal->getValue();
3618 uint64_t TableSize = RangeSpread.getLimitedValue() + 1;
3619 if (!ShouldBuildLookupTable(SI, TableSize, TD, ResultTypes))
3622 // Create the BB that does the lookups.
3623 Module &Mod = *CommonDest->getParent()->getParent();
3624 BasicBlock *LookupBB = BasicBlock::Create(Mod.getContext(),
3626 CommonDest->getParent(),
3629 // Check whether the condition value is within the case range, and branch to
3631 Builder.SetInsertPoint(SI);
3632 Value *TableIndex = Builder.CreateSub(SI->getCondition(), MinCaseVal,
3634 Value *Cmp = Builder.CreateICmpULT(TableIndex, ConstantInt::get(
3635 MinCaseVal->getType(), TableSize));
3636 Builder.CreateCondBr(Cmp, LookupBB, SI->getDefaultDest());
3638 // Populate the BB that does the lookups.
3639 Builder.SetInsertPoint(LookupBB);
3640 bool ReturnedEarly = false;
3641 for (size_t I = 0, E = PHIs.size(); I != E; ++I) {
3642 PHINode *PHI = PHIs[I];
3644 SwitchLookupTable Table(Mod, TableSize, MinCaseVal, ResultLists[PHI],
3645 DefaultResults[PHI], TD);
3647 Value *Result = Table.BuildLookup(TableIndex, Builder);
3649 // If the result is used to return immediately from the function, we want to
3650 // do that right here.
3651 if (PHI->hasOneUse() && isa<ReturnInst>(*PHI->use_begin()) &&
3652 *PHI->use_begin() == CommonDest->getFirstNonPHIOrDbg()) {
3653 Builder.CreateRet(Result);
3654 ReturnedEarly = true;
3658 PHI->addIncoming(Result, LookupBB);
3662 Builder.CreateBr(CommonDest);
3664 // Remove the switch.
3665 for (unsigned i = 0; i < SI->getNumSuccessors(); ++i) {
3666 BasicBlock *Succ = SI->getSuccessor(i);
3667 if (Succ == SI->getDefaultDest()) continue;
3668 Succ->removePredecessor(SI->getParent());
3670 SI->eraseFromParent();
3676 bool SimplifyCFGOpt::SimplifySwitch(SwitchInst *SI, IRBuilder<> &Builder) {
3677 BasicBlock *BB = SI->getParent();
3679 if (isValueEqualityComparison(SI)) {
3680 // If we only have one predecessor, and if it is a branch on this value,
3681 // see if that predecessor totally determines the outcome of this switch.
3682 if (BasicBlock *OnlyPred = BB->getSinglePredecessor())
3683 if (SimplifyEqualityComparisonWithOnlyPredecessor(SI, OnlyPred, Builder))
3684 return SimplifyCFG(BB) | true;
3686 Value *Cond = SI->getCondition();
3687 if (SelectInst *Select = dyn_cast<SelectInst>(Cond))
3688 if (SimplifySwitchOnSelect(SI, Select))
3689 return SimplifyCFG(BB) | true;
3691 // If the block only contains the switch, see if we can fold the block
3692 // away into any preds.
3693 BasicBlock::iterator BBI = BB->begin();
3694 // Ignore dbg intrinsics.
3695 while (isa<DbgInfoIntrinsic>(BBI))
3698 if (FoldValueComparisonIntoPredecessors(SI, Builder))
3699 return SimplifyCFG(BB) | true;
3702 // Try to transform the switch into an icmp and a branch.
3703 if (TurnSwitchRangeIntoICmp(SI, Builder))
3704 return SimplifyCFG(BB) | true;
3706 // Remove unreachable cases.
3707 if (EliminateDeadSwitchCases(SI))
3708 return SimplifyCFG(BB) | true;
3710 if (ForwardSwitchConditionToPHI(SI))
3711 return SimplifyCFG(BB) | true;
3713 if (SwitchToLookupTable(SI, Builder, TD, TTI))
3714 return SimplifyCFG(BB) | true;
3719 bool SimplifyCFGOpt::SimplifyIndirectBr(IndirectBrInst *IBI) {
3720 BasicBlock *BB = IBI->getParent();
3721 bool Changed = false;
3723 // Eliminate redundant destinations.
3724 SmallPtrSet<Value *, 8> Succs;
3725 for (unsigned i = 0, e = IBI->getNumDestinations(); i != e; ++i) {
3726 BasicBlock *Dest = IBI->getDestination(i);
3727 if (!Dest->hasAddressTaken() || !Succs.insert(Dest)) {
3728 Dest->removePredecessor(BB);
3729 IBI->removeDestination(i);
3735 if (IBI->getNumDestinations() == 0) {
3736 // If the indirectbr has no successors, change it to unreachable.
3737 new UnreachableInst(IBI->getContext(), IBI);
3738 EraseTerminatorInstAndDCECond(IBI);
3742 if (IBI->getNumDestinations() == 1) {
3743 // If the indirectbr has one successor, change it to a direct branch.
3744 BranchInst::Create(IBI->getDestination(0), IBI);
3745 EraseTerminatorInstAndDCECond(IBI);
3749 if (SelectInst *SI = dyn_cast<SelectInst>(IBI->getAddress())) {
3750 if (SimplifyIndirectBrOnSelect(IBI, SI))
3751 return SimplifyCFG(BB) | true;
3756 bool SimplifyCFGOpt::SimplifyUncondBranch(BranchInst *BI, IRBuilder<> &Builder){
3757 BasicBlock *BB = BI->getParent();
3759 if (SinkCommon && SinkThenElseCodeToEnd(BI))
3762 // If the Terminator is the only non-phi instruction, simplify the block.
3763 BasicBlock::iterator I = BB->getFirstNonPHIOrDbgOrLifetime();
3764 if (I->isTerminator() && BB != &BB->getParent()->getEntryBlock() &&
3765 TryToSimplifyUncondBranchFromEmptyBlock(BB))
3768 // If the only instruction in the block is a seteq/setne comparison
3769 // against a constant, try to simplify the block.
3770 if (ICmpInst *ICI = dyn_cast<ICmpInst>(I))
3771 if (ICI->isEquality() && isa<ConstantInt>(ICI->getOperand(1))) {
3772 for (++I; isa<DbgInfoIntrinsic>(I); ++I)
3774 if (I->isTerminator() &&
3775 TryToSimplifyUncondBranchWithICmpInIt(ICI, TD, Builder))
3779 // If this basic block is ONLY a compare and a branch, and if a predecessor
3780 // branches to us and our successor, fold the comparison into the
3781 // predecessor and use logical operations to update the incoming value
3782 // for PHI nodes in common successor.
3783 if (FoldBranchToCommonDest(BI))
3784 return SimplifyCFG(BB) | true;
3789 bool SimplifyCFGOpt::SimplifyCondBranch(BranchInst *BI, IRBuilder<> &Builder) {
3790 BasicBlock *BB = BI->getParent();
3792 // Conditional branch
3793 if (isValueEqualityComparison(BI)) {
3794 // If we only have one predecessor, and if it is a branch on this value,
3795 // see if that predecessor totally determines the outcome of this
3797 if (BasicBlock *OnlyPred = BB->getSinglePredecessor())
3798 if (SimplifyEqualityComparisonWithOnlyPredecessor(BI, OnlyPred, Builder))
3799 return SimplifyCFG(BB) | true;
3801 // This block must be empty, except for the setcond inst, if it exists.
3802 // Ignore dbg intrinsics.
3803 BasicBlock::iterator I = BB->begin();
3804 // Ignore dbg intrinsics.
3805 while (isa<DbgInfoIntrinsic>(I))
3808 if (FoldValueComparisonIntoPredecessors(BI, Builder))
3809 return SimplifyCFG(BB) | true;
3810 } else if (&*I == cast<Instruction>(BI->getCondition())){
3812 // Ignore dbg intrinsics.
3813 while (isa<DbgInfoIntrinsic>(I))
3815 if (&*I == BI && FoldValueComparisonIntoPredecessors(BI, Builder))
3816 return SimplifyCFG(BB) | true;
3820 // Try to turn "br (X == 0 | X == 1), T, F" into a switch instruction.
3821 if (SimplifyBranchOnICmpChain(BI, TD, Builder))
3824 // If this basic block is ONLY a compare and a branch, and if a predecessor
3825 // branches to us and one of our successors, fold the comparison into the
3826 // predecessor and use logical operations to pick the right destination.
3827 if (FoldBranchToCommonDest(BI))
3828 return SimplifyCFG(BB) | true;
3830 // We have a conditional branch to two blocks that are only reachable
3831 // from BI. We know that the condbr dominates the two blocks, so see if
3832 // there is any identical code in the "then" and "else" blocks. If so, we
3833 // can hoist it up to the branching block.
3834 if (BI->getSuccessor(0)->getSinglePredecessor() != 0) {
3835 if (BI->getSuccessor(1)->getSinglePredecessor() != 0) {
3836 if (HoistThenElseCodeToIf(BI))
3837 return SimplifyCFG(BB) | true;
3839 // If Successor #1 has multiple preds, we may be able to conditionally
3840 // execute Successor #0 if it branches to successor #1.
3841 TerminatorInst *Succ0TI = BI->getSuccessor(0)->getTerminator();
3842 if (Succ0TI->getNumSuccessors() == 1 &&
3843 Succ0TI->getSuccessor(0) == BI->getSuccessor(1))
3844 if (SpeculativelyExecuteBB(BI, BI->getSuccessor(0)))
3845 return SimplifyCFG(BB) | true;
3847 } else if (BI->getSuccessor(1)->getSinglePredecessor() != 0) {
3848 // If Successor #0 has multiple preds, we may be able to conditionally
3849 // execute Successor #1 if it branches to successor #0.
3850 TerminatorInst *Succ1TI = BI->getSuccessor(1)->getTerminator();
3851 if (Succ1TI->getNumSuccessors() == 1 &&
3852 Succ1TI->getSuccessor(0) == BI->getSuccessor(0))
3853 if (SpeculativelyExecuteBB(BI, BI->getSuccessor(1)))
3854 return SimplifyCFG(BB) | true;
3857 // If this is a branch on a phi node in the current block, thread control
3858 // through this block if any PHI node entries are constants.
3859 if (PHINode *PN = dyn_cast<PHINode>(BI->getCondition()))
3860 if (PN->getParent() == BI->getParent())
3861 if (FoldCondBranchOnPHI(BI, TD))
3862 return SimplifyCFG(BB) | true;
3864 // Scan predecessor blocks for conditional branches.
3865 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI)
3866 if (BranchInst *PBI = dyn_cast<BranchInst>((*PI)->getTerminator()))
3867 if (PBI != BI && PBI->isConditional())
3868 if (SimplifyCondBranchToCondBranch(PBI, BI))
3869 return SimplifyCFG(BB) | true;
3874 /// Check if passing a value to an instruction will cause undefined behavior.
3875 static bool passingValueIsAlwaysUndefined(Value *V, Instruction *I) {
3876 Constant *C = dyn_cast<Constant>(V);
3883 if (C->isNullValue()) {
3884 // Only look at the first use, avoid hurting compile time with long uselists
3885 User *Use = *I->use_begin();
3887 // Now make sure that there are no instructions in between that can alter
3888 // control flow (eg. calls)
3889 for (BasicBlock::iterator i = ++BasicBlock::iterator(I); &*i != Use; ++i)
3890 if (i == I->getParent()->end() || i->mayHaveSideEffects())
3893 // Look through GEPs. A load from a GEP derived from NULL is still undefined
3894 if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(Use))
3895 if (GEP->getPointerOperand() == I)
3896 return passingValueIsAlwaysUndefined(V, GEP);
3898 // Look through bitcasts.
3899 if (BitCastInst *BC = dyn_cast<BitCastInst>(Use))
3900 return passingValueIsAlwaysUndefined(V, BC);
3902 // Load from null is undefined.
3903 if (LoadInst *LI = dyn_cast<LoadInst>(Use))
3904 return LI->getPointerAddressSpace() == 0;
3906 // Store to null is undefined.
3907 if (StoreInst *SI = dyn_cast<StoreInst>(Use))
3908 return SI->getPointerAddressSpace() == 0 && SI->getPointerOperand() == I;
3913 /// If BB has an incoming value that will always trigger undefined behavior
3914 /// (eg. null pointer dereference), remove the branch leading here.
3915 static bool removeUndefIntroducingPredecessor(BasicBlock *BB) {
3916 for (BasicBlock::iterator i = BB->begin();
3917 PHINode *PHI = dyn_cast<PHINode>(i); ++i)
3918 for (unsigned i = 0, e = PHI->getNumIncomingValues(); i != e; ++i)
3919 if (passingValueIsAlwaysUndefined(PHI->getIncomingValue(i), PHI)) {
3920 TerminatorInst *T = PHI->getIncomingBlock(i)->getTerminator();
3921 IRBuilder<> Builder(T);
3922 if (BranchInst *BI = dyn_cast<BranchInst>(T)) {
3923 BB->removePredecessor(PHI->getIncomingBlock(i));
3924 // Turn uncoditional branches into unreachables and remove the dead
3925 // destination from conditional branches.
3926 if (BI->isUnconditional())
3927 Builder.CreateUnreachable();
3929 Builder.CreateBr(BI->getSuccessor(0) == BB ? BI->getSuccessor(1) :
3930 BI->getSuccessor(0));
3931 BI->eraseFromParent();
3934 // TODO: SwitchInst.
3940 bool SimplifyCFGOpt::run(BasicBlock *BB) {
3941 bool Changed = false;
3943 assert(BB && BB->getParent() && "Block not embedded in function!");
3944 assert(BB->getTerminator() && "Degenerate basic block encountered!");
3946 // Remove basic blocks that have no predecessors (except the entry block)...
3947 // or that just have themself as a predecessor. These are unreachable.
3948 if ((pred_begin(BB) == pred_end(BB) &&
3949 BB != &BB->getParent()->getEntryBlock()) ||
3950 BB->getSinglePredecessor() == BB) {
3951 DEBUG(dbgs() << "Removing BB: \n" << *BB);
3952 DeleteDeadBlock(BB);
3956 // Check to see if we can constant propagate this terminator instruction
3958 Changed |= ConstantFoldTerminator(BB, true);
3960 // Check for and eliminate duplicate PHI nodes in this block.
3961 Changed |= EliminateDuplicatePHINodes(BB);
3963 // Check for and remove branches that will always cause undefined behavior.
3964 Changed |= removeUndefIntroducingPredecessor(BB);
3966 // Merge basic blocks into their predecessor if there is only one distinct
3967 // pred, and if there is only one distinct successor of the predecessor, and
3968 // if there are no PHI nodes.
3970 if (MergeBlockIntoPredecessor(BB))
3973 IRBuilder<> Builder(BB);
3975 // If there is a trivial two-entry PHI node in this basic block, and we can
3976 // eliminate it, do so now.
3977 if (PHINode *PN = dyn_cast<PHINode>(BB->begin()))
3978 if (PN->getNumIncomingValues() == 2)
3979 Changed |= FoldTwoEntryPHINode(PN, TD);
3981 Builder.SetInsertPoint(BB->getTerminator());
3982 if (BranchInst *BI = dyn_cast<BranchInst>(BB->getTerminator())) {
3983 if (BI->isUnconditional()) {
3984 if (SimplifyUncondBranch(BI, Builder)) return true;
3986 if (SimplifyCondBranch(BI, Builder)) return true;
3988 } else if (ReturnInst *RI = dyn_cast<ReturnInst>(BB->getTerminator())) {
3989 if (SimplifyReturn(RI, Builder)) return true;
3990 } else if (ResumeInst *RI = dyn_cast<ResumeInst>(BB->getTerminator())) {
3991 if (SimplifyResume(RI, Builder)) return true;
3992 } else if (SwitchInst *SI = dyn_cast<SwitchInst>(BB->getTerminator())) {
3993 if (SimplifySwitch(SI, Builder)) return true;
3994 } else if (UnreachableInst *UI =
3995 dyn_cast<UnreachableInst>(BB->getTerminator())) {
3996 if (SimplifyUnreachable(UI)) return true;
3997 } else if (IndirectBrInst *IBI =
3998 dyn_cast<IndirectBrInst>(BB->getTerminator())) {
3999 if (SimplifyIndirectBr(IBI)) return true;
4005 /// SimplifyCFG - This function is used to do simplification of a CFG. For
4006 /// example, it adjusts branches to branches to eliminate the extra hop, it
4007 /// eliminates unreachable basic blocks, and does other "peephole" optimization
4008 /// of the CFG. It returns true if a modification was made.
4010 bool llvm::SimplifyCFG(BasicBlock *BB, const DataLayout *TD,
4011 const TargetTransformInfo *TTI) {
4012 return SimplifyCFGOpt(TD, TTI).run(BB);