1 //===- SimplifyCFG.cpp - Code to perform CFG simplification ---------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // Peephole optimize the CFG. 10 // 11 //===----------------------------------------------------------------------===// 12 13 #include "llvm/ADT/APInt.h" 14 #include "llvm/ADT/ArrayRef.h" 15 #include "llvm/ADT/DenseMap.h" 16 #include "llvm/ADT/Optional.h" 17 #include "llvm/ADT/STLExtras.h" 18 #include "llvm/ADT/SetOperations.h" 19 #include "llvm/ADT/SetVector.h" 20 #include "llvm/ADT/SmallPtrSet.h" 21 #include "llvm/ADT/SmallVector.h" 22 #include "llvm/ADT/Statistic.h" 23 #include "llvm/ADT/StringRef.h" 24 #include "llvm/Analysis/AssumptionCache.h" 25 #include "llvm/Analysis/ConstantFolding.h" 26 #include "llvm/Analysis/EHPersonalities.h" 27 #include "llvm/Analysis/GuardUtils.h" 28 #include "llvm/Analysis/InstructionSimplify.h" 29 #include "llvm/Analysis/MemorySSA.h" 30 #include "llvm/Analysis/MemorySSAUpdater.h" 31 #include "llvm/Analysis/TargetTransformInfo.h" 32 #include "llvm/Analysis/ValueTracking.h" 33 #include "llvm/IR/Attributes.h" 34 #include "llvm/IR/BasicBlock.h" 35 #include "llvm/IR/CFG.h" 36 #include "llvm/IR/Constant.h" 37 #include "llvm/IR/ConstantRange.h" 38 #include "llvm/IR/Constants.h" 39 #include "llvm/IR/DataLayout.h" 40 #include "llvm/IR/DerivedTypes.h" 41 #include "llvm/IR/Function.h" 42 #include "llvm/IR/GlobalValue.h" 43 #include "llvm/IR/GlobalVariable.h" 44 #include "llvm/IR/IRBuilder.h" 45 #include "llvm/IR/InstrTypes.h" 46 #include "llvm/IR/Instruction.h" 47 #include "llvm/IR/Instructions.h" 48 #include "llvm/IR/IntrinsicInst.h" 49 #include "llvm/IR/Intrinsics.h" 50 #include "llvm/IR/LLVMContext.h" 51 #include "llvm/IR/MDBuilder.h" 52 #include "llvm/IR/Metadata.h" 53 #include "llvm/IR/Module.h" 54 #include "llvm/IR/NoFolder.h" 55 #include "llvm/IR/Operator.h" 56 #include "llvm/IR/PatternMatch.h" 57 #include "llvm/IR/Type.h" 58 #include "llvm/IR/Use.h" 59 #include "llvm/IR/User.h" 60 #include "llvm/IR/Value.h" 61 #include "llvm/Support/Casting.h" 62 #include "llvm/Support/CommandLine.h" 63 #include "llvm/Support/Debug.h" 64 #include "llvm/Support/ErrorHandling.h" 65 #include "llvm/Support/KnownBits.h" 66 #include "llvm/Support/MathExtras.h" 67 #include "llvm/Support/raw_ostream.h" 68 #include "llvm/Transforms/Utils/BasicBlockUtils.h" 69 #include "llvm/Transforms/Utils/Local.h" 70 #include "llvm/Transforms/Utils/ValueMapper.h" 71 #include <algorithm> 72 #include <cassert> 73 #include <climits> 74 #include <cstddef> 75 #include <cstdint> 76 #include <iterator> 77 #include <map> 78 #include <set> 79 #include <tuple> 80 #include <utility> 81 #include <vector> 82 83 using namespace llvm; 84 using namespace PatternMatch; 85 86 #define DEBUG_TYPE "simplifycfg" 87 88 // Chosen as 2 so as to be cheap, but still to have enough power to fold 89 // a select, so the "clamp" idiom (of a min followed by a max) will be caught. 90 // To catch this, we need to fold a compare and a select, hence '2' being the 91 // minimum reasonable default. 92 static cl::opt<unsigned> PHINodeFoldingThreshold( 93 "phi-node-folding-threshold", cl::Hidden, cl::init(2), 94 cl::desc( 95 "Control the amount of phi node folding to perform (default = 2)")); 96 97 static cl::opt<unsigned> TwoEntryPHINodeFoldingThreshold( 98 "two-entry-phi-node-folding-threshold", cl::Hidden, cl::init(4), 99 cl::desc("Control the maximal total instruction cost that we are willing " 100 "to speculatively execute to fold a 2-entry PHI node into a " 101 "select (default = 4)")); 102 103 static cl::opt<bool> DupRet( 104 "simplifycfg-dup-ret", cl::Hidden, cl::init(false), 105 cl::desc("Duplicate return instructions into unconditional branches")); 106 107 static cl::opt<bool> 108 SinkCommon("simplifycfg-sink-common", cl::Hidden, cl::init(true), 109 cl::desc("Sink common instructions down to the end block")); 110 111 static cl::opt<bool> HoistCondStores( 112 "simplifycfg-hoist-cond-stores", cl::Hidden, cl::init(true), 113 cl::desc("Hoist conditional stores if an unconditional store precedes")); 114 115 static cl::opt<bool> MergeCondStores( 116 "simplifycfg-merge-cond-stores", cl::Hidden, cl::init(true), 117 cl::desc("Hoist conditional stores even if an unconditional store does not " 118 "precede - hoist multiple conditional stores into a single " 119 "predicated store")); 120 121 static cl::opt<bool> MergeCondStoresAggressively( 122 "simplifycfg-merge-cond-stores-aggressively", cl::Hidden, cl::init(false), 123 cl::desc("When merging conditional stores, do so even if the resultant " 124 "basic blocks are unlikely to be if-converted as a result")); 125 126 static cl::opt<bool> SpeculateOneExpensiveInst( 127 "speculate-one-expensive-inst", cl::Hidden, cl::init(true), 128 cl::desc("Allow exactly one expensive instruction to be speculatively " 129 "executed")); 130 131 static cl::opt<unsigned> MaxSpeculationDepth( 132 "max-speculation-depth", cl::Hidden, cl::init(10), 133 cl::desc("Limit maximum recursion depth when calculating costs of " 134 "speculatively executed instructions")); 135 136 static cl::opt<int> 137 MaxSmallBlockSize("simplifycfg-max-small-block-size", cl::Hidden, cl::init(10), 138 cl::desc("Max size of a block which is still considered " 139 "small enough to thread through")); 140 141 STATISTIC(NumBitMaps, "Number of switch instructions turned into bitmaps"); 142 STATISTIC(NumLinearMaps, 143 "Number of switch instructions turned into linear mapping"); 144 STATISTIC(NumLookupTables, 145 "Number of switch instructions turned into lookup tables"); 146 STATISTIC( 147 NumLookupTablesHoles, 148 "Number of switch instructions turned into lookup tables (holes checked)"); 149 STATISTIC(NumTableCmpReuses, "Number of reused switch table lookup compares"); 150 STATISTIC(NumSinkCommons, 151 "Number of common instructions sunk down to the end block"); 152 STATISTIC(NumSpeculations, "Number of speculative executed instructions"); 153 154 namespace { 155 156 // The first field contains the value that the switch produces when a certain 157 // case group is selected, and the second field is a vector containing the 158 // cases composing the case group. 159 using SwitchCaseResultVectorTy = 160 SmallVector<std::pair<Constant *, SmallVector<ConstantInt *, 4>>, 2>; 161 162 // The first field contains the phi node that generates a result of the switch 163 // and the second field contains the value generated for a certain case in the 164 // switch for that PHI. 165 using SwitchCaseResultsTy = SmallVector<std::pair<PHINode *, Constant *>, 4>; 166 167 /// ValueEqualityComparisonCase - Represents a case of a switch. 168 struct ValueEqualityComparisonCase { 169 ConstantInt *Value; 170 BasicBlock *Dest; 171 172 ValueEqualityComparisonCase(ConstantInt *Value, BasicBlock *Dest) 173 : Value(Value), Dest(Dest) {} 174 175 bool operator<(ValueEqualityComparisonCase RHS) const { 176 // Comparing pointers is ok as we only rely on the order for uniquing. 177 return Value < RHS.Value; 178 } 179 180 bool operator==(BasicBlock *RHSDest) const { return Dest == RHSDest; } 181 }; 182 183 class SimplifyCFGOpt { 184 const TargetTransformInfo &TTI; 185 const DataLayout &DL; 186 SmallPtrSetImpl<BasicBlock *> *LoopHeaders; 187 const SimplifyCFGOptions &Options; 188 bool Resimplify; 189 190 Value *isValueEqualityComparison(Instruction *TI); 191 BasicBlock *GetValueEqualityComparisonCases( 192 Instruction *TI, std::vector<ValueEqualityComparisonCase> &Cases); 193 bool SimplifyEqualityComparisonWithOnlyPredecessor(Instruction *TI, 194 BasicBlock *Pred, 195 IRBuilder<> &Builder); 196 bool FoldValueComparisonIntoPredecessors(Instruction *TI, 197 IRBuilder<> &Builder); 198 199 bool simplifyReturn(ReturnInst *RI, IRBuilder<> &Builder); 200 bool simplifyResume(ResumeInst *RI, IRBuilder<> &Builder); 201 bool simplifySingleResume(ResumeInst *RI); 202 bool simplifyCommonResume(ResumeInst *RI); 203 bool simplifyCleanupReturn(CleanupReturnInst *RI); 204 bool simplifyUnreachable(UnreachableInst *UI); 205 bool simplifySwitch(SwitchInst *SI, IRBuilder<> &Builder); 206 bool simplifyIndirectBr(IndirectBrInst *IBI); 207 bool simplifyBranch(BranchInst *Branch, IRBuilder<> &Builder); 208 bool simplifyUncondBranch(BranchInst *BI, IRBuilder<> &Builder); 209 bool simplifyCondBranch(BranchInst *BI, IRBuilder<> &Builder); 210 bool SimplifyCondBranchToTwoReturns(BranchInst *BI, IRBuilder<> &Builder); 211 212 bool tryToSimplifyUncondBranchWithICmpInIt(ICmpInst *ICI, 213 IRBuilder<> &Builder); 214 215 bool HoistThenElseCodeToIf(BranchInst *BI, const TargetTransformInfo &TTI); 216 bool SpeculativelyExecuteBB(BranchInst *BI, BasicBlock *ThenBB, 217 const TargetTransformInfo &TTI); 218 bool SimplifyTerminatorOnSelect(Instruction *OldTerm, Value *Cond, 219 BasicBlock *TrueBB, BasicBlock *FalseBB, 220 uint32_t TrueWeight, uint32_t FalseWeight); 221 bool SimplifyBranchOnICmpChain(BranchInst *BI, IRBuilder<> &Builder, 222 const DataLayout &DL); 223 bool SimplifySwitchOnSelect(SwitchInst *SI, SelectInst *Select); 224 bool SimplifyIndirectBrOnSelect(IndirectBrInst *IBI, SelectInst *SI); 225 bool TurnSwitchRangeIntoICmp(SwitchInst *SI, IRBuilder<> &Builder); 226 227 public: 228 SimplifyCFGOpt(const TargetTransformInfo &TTI, const DataLayout &DL, 229 SmallPtrSetImpl<BasicBlock *> *LoopHeaders, 230 const SimplifyCFGOptions &Opts) 231 : TTI(TTI), DL(DL), LoopHeaders(LoopHeaders), Options(Opts) {} 232 233 bool run(BasicBlock *BB); 234 bool simplifyOnce(BasicBlock *BB); 235 236 // Helper to set Resimplify and return change indication. 237 bool requestResimplify() { 238 Resimplify = true; 239 return true; 240 } 241 }; 242 243 } // end anonymous namespace 244 245 /// Return true if it is safe to merge these two 246 /// terminator instructions together. 247 static bool 248 SafeToMergeTerminators(Instruction *SI1, Instruction *SI2, 249 SmallSetVector<BasicBlock *, 4> *FailBlocks = nullptr) { 250 if (SI1 == SI2) 251 return false; // Can't merge with self! 252 253 // It is not safe to merge these two switch instructions if they have a common 254 // successor, and if that successor has a PHI node, and if *that* PHI node has 255 // conflicting incoming values from the two switch blocks. 256 BasicBlock *SI1BB = SI1->getParent(); 257 BasicBlock *SI2BB = SI2->getParent(); 258 259 SmallPtrSet<BasicBlock *, 16> SI1Succs(succ_begin(SI1BB), succ_end(SI1BB)); 260 bool Fail = false; 261 for (BasicBlock *Succ : successors(SI2BB)) 262 if (SI1Succs.count(Succ)) 263 for (BasicBlock::iterator BBI = Succ->begin(); isa<PHINode>(BBI); ++BBI) { 264 PHINode *PN = cast<PHINode>(BBI); 265 if (PN->getIncomingValueForBlock(SI1BB) != 266 PN->getIncomingValueForBlock(SI2BB)) { 267 if (FailBlocks) 268 FailBlocks->insert(Succ); 269 Fail = true; 270 } 271 } 272 273 return !Fail; 274 } 275 276 /// Return true if it is safe and profitable to merge these two terminator 277 /// instructions together, where SI1 is an unconditional branch. PhiNodes will 278 /// store all PHI nodes in common successors. 279 static bool 280 isProfitableToFoldUnconditional(BranchInst *SI1, BranchInst *SI2, 281 Instruction *Cond, 282 SmallVectorImpl<PHINode *> &PhiNodes) { 283 if (SI1 == SI2) 284 return false; // Can't merge with self! 285 assert(SI1->isUnconditional() && SI2->isConditional()); 286 287 // We fold the unconditional branch if we can easily update all PHI nodes in 288 // common successors: 289 // 1> We have a constant incoming value for the conditional branch; 290 // 2> We have "Cond" as the incoming value for the unconditional branch; 291 // 3> SI2->getCondition() and Cond have same operands. 292 CmpInst *Ci2 = dyn_cast<CmpInst>(SI2->getCondition()); 293 if (!Ci2) 294 return false; 295 if (!(Cond->getOperand(0) == Ci2->getOperand(0) && 296 Cond->getOperand(1) == Ci2->getOperand(1)) && 297 !(Cond->getOperand(0) == Ci2->getOperand(1) && 298 Cond->getOperand(1) == Ci2->getOperand(0))) 299 return false; 300 301 BasicBlock *SI1BB = SI1->getParent(); 302 BasicBlock *SI2BB = SI2->getParent(); 303 SmallPtrSet<BasicBlock *, 16> SI1Succs(succ_begin(SI1BB), succ_end(SI1BB)); 304 for (BasicBlock *Succ : successors(SI2BB)) 305 if (SI1Succs.count(Succ)) 306 for (BasicBlock::iterator BBI = Succ->begin(); isa<PHINode>(BBI); ++BBI) { 307 PHINode *PN = cast<PHINode>(BBI); 308 if (PN->getIncomingValueForBlock(SI1BB) != Cond || 309 !isa<ConstantInt>(PN->getIncomingValueForBlock(SI2BB))) 310 return false; 311 PhiNodes.push_back(PN); 312 } 313 return true; 314 } 315 316 /// Update PHI nodes in Succ to indicate that there will now be entries in it 317 /// from the 'NewPred' block. The values that will be flowing into the PHI nodes 318 /// will be the same as those coming in from ExistPred, an existing predecessor 319 /// of Succ. 320 static void AddPredecessorToBlock(BasicBlock *Succ, BasicBlock *NewPred, 321 BasicBlock *ExistPred, 322 MemorySSAUpdater *MSSAU = nullptr) { 323 for (PHINode &PN : Succ->phis()) 324 PN.addIncoming(PN.getIncomingValueForBlock(ExistPred), NewPred); 325 if (MSSAU) 326 if (auto *MPhi = MSSAU->getMemorySSA()->getMemoryAccess(Succ)) 327 MPhi->addIncoming(MPhi->getIncomingValueForBlock(ExistPred), NewPred); 328 } 329 330 /// Compute an abstract "cost" of speculating the given instruction, 331 /// which is assumed to be safe to speculate. TCC_Free means cheap, 332 /// TCC_Basic means less cheap, and TCC_Expensive means prohibitively 333 /// expensive. 334 static unsigned ComputeSpeculationCost(const User *I, 335 const TargetTransformInfo &TTI) { 336 assert(isSafeToSpeculativelyExecute(I) && 337 "Instruction is not safe to speculatively execute!"); 338 return TTI.getUserCost(I, TargetTransformInfo::TCK_SizeAndLatency); 339 } 340 341 /// If we have a merge point of an "if condition" as accepted above, 342 /// return true if the specified value dominates the block. We 343 /// don't handle the true generality of domination here, just a special case 344 /// which works well enough for us. 345 /// 346 /// If AggressiveInsts is non-null, and if V does not dominate BB, we check to 347 /// see if V (which must be an instruction) and its recursive operands 348 /// that do not dominate BB have a combined cost lower than CostRemaining and 349 /// are non-trapping. If both are true, the instruction is inserted into the 350 /// set and true is returned. 351 /// 352 /// The cost for most non-trapping instructions is defined as 1 except for 353 /// Select whose cost is 2. 354 /// 355 /// After this function returns, CostRemaining is decreased by the cost of 356 /// V plus its non-dominating operands. If that cost is greater than 357 /// CostRemaining, false is returned and CostRemaining is undefined. 358 static bool DominatesMergePoint(Value *V, BasicBlock *BB, 359 SmallPtrSetImpl<Instruction *> &AggressiveInsts, 360 int &BudgetRemaining, 361 const TargetTransformInfo &TTI, 362 unsigned Depth = 0) { 363 // It is possible to hit a zero-cost cycle (phi/gep instructions for example), 364 // so limit the recursion depth. 365 // TODO: While this recursion limit does prevent pathological behavior, it 366 // would be better to track visited instructions to avoid cycles. 367 if (Depth == MaxSpeculationDepth) 368 return false; 369 370 Instruction *I = dyn_cast<Instruction>(V); 371 if (!I) { 372 // Non-instructions all dominate instructions, but not all constantexprs 373 // can be executed unconditionally. 374 if (ConstantExpr *C = dyn_cast<ConstantExpr>(V)) 375 if (C->canTrap()) 376 return false; 377 return true; 378 } 379 BasicBlock *PBB = I->getParent(); 380 381 // We don't want to allow weird loops that might have the "if condition" in 382 // the bottom of this block. 383 if (PBB == BB) 384 return false; 385 386 // If this instruction is defined in a block that contains an unconditional 387 // branch to BB, then it must be in the 'conditional' part of the "if 388 // statement". If not, it definitely dominates the region. 389 BranchInst *BI = dyn_cast<BranchInst>(PBB->getTerminator()); 390 if (!BI || BI->isConditional() || BI->getSuccessor(0) != BB) 391 return true; 392 393 // If we have seen this instruction before, don't count it again. 394 if (AggressiveInsts.count(I)) 395 return true; 396 397 // Okay, it looks like the instruction IS in the "condition". Check to 398 // see if it's a cheap instruction to unconditionally compute, and if it 399 // only uses stuff defined outside of the condition. If so, hoist it out. 400 if (!isSafeToSpeculativelyExecute(I)) 401 return false; 402 403 BudgetRemaining -= ComputeSpeculationCost(I, TTI); 404 405 // Allow exactly one instruction to be speculated regardless of its cost 406 // (as long as it is safe to do so). 407 // This is intended to flatten the CFG even if the instruction is a division 408 // or other expensive operation. The speculation of an expensive instruction 409 // is expected to be undone in CodeGenPrepare if the speculation has not 410 // enabled further IR optimizations. 411 if (BudgetRemaining < 0 && 412 (!SpeculateOneExpensiveInst || !AggressiveInsts.empty() || Depth > 0)) 413 return false; 414 415 // Okay, we can only really hoist these out if their operands do 416 // not take us over the cost threshold. 417 for (User::op_iterator i = I->op_begin(), e = I->op_end(); i != e; ++i) 418 if (!DominatesMergePoint(*i, BB, AggressiveInsts, BudgetRemaining, TTI, 419 Depth + 1)) 420 return false; 421 // Okay, it's safe to do this! Remember this instruction. 422 AggressiveInsts.insert(I); 423 return true; 424 } 425 426 /// Extract ConstantInt from value, looking through IntToPtr 427 /// and PointerNullValue. Return NULL if value is not a constant int. 428 static ConstantInt *GetConstantInt(Value *V, const DataLayout &DL) { 429 // Normal constant int. 430 ConstantInt *CI = dyn_cast<ConstantInt>(V); 431 if (CI || !isa<Constant>(V) || !V->getType()->isPointerTy()) 432 return CI; 433 434 // This is some kind of pointer constant. Turn it into a pointer-sized 435 // ConstantInt if possible. 436 IntegerType *PtrTy = cast<IntegerType>(DL.getIntPtrType(V->getType())); 437 438 // Null pointer means 0, see SelectionDAGBuilder::getValue(const Value*). 439 if (isa<ConstantPointerNull>(V)) 440 return ConstantInt::get(PtrTy, 0); 441 442 // IntToPtr const int. 443 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V)) 444 if (CE->getOpcode() == Instruction::IntToPtr) 445 if (ConstantInt *CI = dyn_cast<ConstantInt>(CE->getOperand(0))) { 446 // The constant is very likely to have the right type already. 447 if (CI->getType() == PtrTy) 448 return CI; 449 else 450 return cast<ConstantInt>( 451 ConstantExpr::getIntegerCast(CI, PtrTy, /*isSigned=*/false)); 452 } 453 return nullptr; 454 } 455 456 namespace { 457 458 /// Given a chain of or (||) or and (&&) comparison of a value against a 459 /// constant, this will try to recover the information required for a switch 460 /// structure. 461 /// It will depth-first traverse the chain of comparison, seeking for patterns 462 /// like %a == 12 or %a < 4 and combine them to produce a set of integer 463 /// representing the different cases for the switch. 464 /// Note that if the chain is composed of '||' it will build the set of elements 465 /// that matches the comparisons (i.e. any of this value validate the chain) 466 /// while for a chain of '&&' it will build the set elements that make the test 467 /// fail. 468 struct ConstantComparesGatherer { 469 const DataLayout &DL; 470 471 /// Value found for the switch comparison 472 Value *CompValue = nullptr; 473 474 /// Extra clause to be checked before the switch 475 Value *Extra = nullptr; 476 477 /// Set of integers to match in switch 478 SmallVector<ConstantInt *, 8> Vals; 479 480 /// Number of comparisons matched in the and/or chain 481 unsigned UsedICmps = 0; 482 483 /// Construct and compute the result for the comparison instruction Cond 484 ConstantComparesGatherer(Instruction *Cond, const DataLayout &DL) : DL(DL) { 485 gather(Cond); 486 } 487 488 ConstantComparesGatherer(const ConstantComparesGatherer &) = delete; 489 ConstantComparesGatherer & 490 operator=(const ConstantComparesGatherer &) = delete; 491 492 private: 493 /// Try to set the current value used for the comparison, it succeeds only if 494 /// it wasn't set before or if the new value is the same as the old one 495 bool setValueOnce(Value *NewVal) { 496 if (CompValue && CompValue != NewVal) 497 return false; 498 CompValue = NewVal; 499 return (CompValue != nullptr); 500 } 501 502 /// Try to match Instruction "I" as a comparison against a constant and 503 /// populates the array Vals with the set of values that match (or do not 504 /// match depending on isEQ). 505 /// Return false on failure. On success, the Value the comparison matched 506 /// against is placed in CompValue. 507 /// If CompValue is already set, the function is expected to fail if a match 508 /// is found but the value compared to is different. 509 bool matchInstruction(Instruction *I, bool isEQ) { 510 // If this is an icmp against a constant, handle this as one of the cases. 511 ICmpInst *ICI; 512 ConstantInt *C; 513 if (!((ICI = dyn_cast<ICmpInst>(I)) && 514 (C = GetConstantInt(I->getOperand(1), DL)))) { 515 return false; 516 } 517 518 Value *RHSVal; 519 const APInt *RHSC; 520 521 // Pattern match a special case 522 // (x & ~2^z) == y --> x == y || x == y|2^z 523 // This undoes a transformation done by instcombine to fuse 2 compares. 524 if (ICI->getPredicate() == (isEQ ? ICmpInst::ICMP_EQ : ICmpInst::ICMP_NE)) { 525 // It's a little bit hard to see why the following transformations are 526 // correct. Here is a CVC3 program to verify them for 64-bit values: 527 528 /* 529 ONE : BITVECTOR(64) = BVZEROEXTEND(0bin1, 63); 530 x : BITVECTOR(64); 531 y : BITVECTOR(64); 532 z : BITVECTOR(64); 533 mask : BITVECTOR(64) = BVSHL(ONE, z); 534 QUERY( (y & ~mask = y) => 535 ((x & ~mask = y) <=> (x = y OR x = (y | mask))) 536 ); 537 QUERY( (y | mask = y) => 538 ((x | mask = y) <=> (x = y OR x = (y & ~mask))) 539 ); 540 */ 541 542 // Please note that each pattern must be a dual implication (<--> or 543 // iff). One directional implication can create spurious matches. If the 544 // implication is only one-way, an unsatisfiable condition on the left 545 // side can imply a satisfiable condition on the right side. Dual 546 // implication ensures that satisfiable conditions are transformed to 547 // other satisfiable conditions and unsatisfiable conditions are 548 // transformed to other unsatisfiable conditions. 549 550 // Here is a concrete example of a unsatisfiable condition on the left 551 // implying a satisfiable condition on the right: 552 // 553 // mask = (1 << z) 554 // (x & ~mask) == y --> (x == y || x == (y | mask)) 555 // 556 // Substituting y = 3, z = 0 yields: 557 // (x & -2) == 3 --> (x == 3 || x == 2) 558 559 // Pattern match a special case: 560 /* 561 QUERY( (y & ~mask = y) => 562 ((x & ~mask = y) <=> (x = y OR x = (y | mask))) 563 ); 564 */ 565 if (match(ICI->getOperand(0), 566 m_And(m_Value(RHSVal), m_APInt(RHSC)))) { 567 APInt Mask = ~*RHSC; 568 if (Mask.isPowerOf2() && (C->getValue() & ~Mask) == C->getValue()) { 569 // If we already have a value for the switch, it has to match! 570 if (!setValueOnce(RHSVal)) 571 return false; 572 573 Vals.push_back(C); 574 Vals.push_back( 575 ConstantInt::get(C->getContext(), 576 C->getValue() | Mask)); 577 UsedICmps++; 578 return true; 579 } 580 } 581 582 // Pattern match a special case: 583 /* 584 QUERY( (y | mask = y) => 585 ((x | mask = y) <=> (x = y OR x = (y & ~mask))) 586 ); 587 */ 588 if (match(ICI->getOperand(0), 589 m_Or(m_Value(RHSVal), m_APInt(RHSC)))) { 590 APInt Mask = *RHSC; 591 if (Mask.isPowerOf2() && (C->getValue() | Mask) == C->getValue()) { 592 // If we already have a value for the switch, it has to match! 593 if (!setValueOnce(RHSVal)) 594 return false; 595 596 Vals.push_back(C); 597 Vals.push_back(ConstantInt::get(C->getContext(), 598 C->getValue() & ~Mask)); 599 UsedICmps++; 600 return true; 601 } 602 } 603 604 // If we already have a value for the switch, it has to match! 605 if (!setValueOnce(ICI->getOperand(0))) 606 return false; 607 608 UsedICmps++; 609 Vals.push_back(C); 610 return ICI->getOperand(0); 611 } 612 613 // If we have "x ult 3", for example, then we can add 0,1,2 to the set. 614 ConstantRange Span = ConstantRange::makeAllowedICmpRegion( 615 ICI->getPredicate(), C->getValue()); 616 617 // Shift the range if the compare is fed by an add. This is the range 618 // compare idiom as emitted by instcombine. 619 Value *CandidateVal = I->getOperand(0); 620 if (match(I->getOperand(0), m_Add(m_Value(RHSVal), m_APInt(RHSC)))) { 621 Span = Span.subtract(*RHSC); 622 CandidateVal = RHSVal; 623 } 624 625 // If this is an and/!= check, then we are looking to build the set of 626 // value that *don't* pass the and chain. I.e. to turn "x ugt 2" into 627 // x != 0 && x != 1. 628 if (!isEQ) 629 Span = Span.inverse(); 630 631 // If there are a ton of values, we don't want to make a ginormous switch. 632 if (Span.isSizeLargerThan(8) || Span.isEmptySet()) { 633 return false; 634 } 635 636 // If we already have a value for the switch, it has to match! 637 if (!setValueOnce(CandidateVal)) 638 return false; 639 640 // Add all values from the range to the set 641 for (APInt Tmp = Span.getLower(); Tmp != Span.getUpper(); ++Tmp) 642 Vals.push_back(ConstantInt::get(I->getContext(), Tmp)); 643 644 UsedICmps++; 645 return true; 646 } 647 648 /// Given a potentially 'or'd or 'and'd together collection of icmp 649 /// eq/ne/lt/gt instructions that compare a value against a constant, extract 650 /// the value being compared, and stick the list constants into the Vals 651 /// vector. 652 /// One "Extra" case is allowed to differ from the other. 653 void gather(Value *V) { 654 bool isEQ = (cast<Instruction>(V)->getOpcode() == Instruction::Or); 655 656 // Keep a stack (SmallVector for efficiency) for depth-first traversal 657 SmallVector<Value *, 8> DFT; 658 SmallPtrSet<Value *, 8> Visited; 659 660 // Initialize 661 Visited.insert(V); 662 DFT.push_back(V); 663 664 while (!DFT.empty()) { 665 V = DFT.pop_back_val(); 666 667 if (Instruction *I = dyn_cast<Instruction>(V)) { 668 // If it is a || (or && depending on isEQ), process the operands. 669 if (I->getOpcode() == (isEQ ? Instruction::Or : Instruction::And)) { 670 if (Visited.insert(I->getOperand(1)).second) 671 DFT.push_back(I->getOperand(1)); 672 if (Visited.insert(I->getOperand(0)).second) 673 DFT.push_back(I->getOperand(0)); 674 continue; 675 } 676 677 // Try to match the current instruction 678 if (matchInstruction(I, isEQ)) 679 // Match succeed, continue the loop 680 continue; 681 } 682 683 // One element of the sequence of || (or &&) could not be match as a 684 // comparison against the same value as the others. 685 // We allow only one "Extra" case to be checked before the switch 686 if (!Extra) { 687 Extra = V; 688 continue; 689 } 690 // Failed to parse a proper sequence, abort now 691 CompValue = nullptr; 692 break; 693 } 694 } 695 }; 696 697 } // end anonymous namespace 698 699 static void EraseTerminatorAndDCECond(Instruction *TI, 700 MemorySSAUpdater *MSSAU = nullptr) { 701 Instruction *Cond = nullptr; 702 if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) { 703 Cond = dyn_cast<Instruction>(SI->getCondition()); 704 } else if (BranchInst *BI = dyn_cast<BranchInst>(TI)) { 705 if (BI->isConditional()) 706 Cond = dyn_cast<Instruction>(BI->getCondition()); 707 } else if (IndirectBrInst *IBI = dyn_cast<IndirectBrInst>(TI)) { 708 Cond = dyn_cast<Instruction>(IBI->getAddress()); 709 } 710 711 TI->eraseFromParent(); 712 if (Cond) 713 RecursivelyDeleteTriviallyDeadInstructions(Cond, nullptr, MSSAU); 714 } 715 716 /// Return true if the specified terminator checks 717 /// to see if a value is equal to constant integer value. 718 Value *SimplifyCFGOpt::isValueEqualityComparison(Instruction *TI) { 719 Value *CV = nullptr; 720 if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) { 721 // Do not permit merging of large switch instructions into their 722 // predecessors unless there is only one predecessor. 723 if (!SI->getParent()->hasNPredecessorsOrMore(128 / SI->getNumSuccessors())) 724 CV = SI->getCondition(); 725 } else if (BranchInst *BI = dyn_cast<BranchInst>(TI)) 726 if (BI->isConditional() && BI->getCondition()->hasOneUse()) 727 if (ICmpInst *ICI = dyn_cast<ICmpInst>(BI->getCondition())) { 728 if (ICI->isEquality() && GetConstantInt(ICI->getOperand(1), DL)) 729 CV = ICI->getOperand(0); 730 } 731 732 // Unwrap any lossless ptrtoint cast. 733 if (CV) { 734 if (PtrToIntInst *PTII = dyn_cast<PtrToIntInst>(CV)) { 735 Value *Ptr = PTII->getPointerOperand(); 736 if (PTII->getType() == DL.getIntPtrType(Ptr->getType())) 737 CV = Ptr; 738 } 739 } 740 return CV; 741 } 742 743 /// Given a value comparison instruction, 744 /// decode all of the 'cases' that it represents and return the 'default' block. 745 BasicBlock *SimplifyCFGOpt::GetValueEqualityComparisonCases( 746 Instruction *TI, std::vector<ValueEqualityComparisonCase> &Cases) { 747 if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) { 748 Cases.reserve(SI->getNumCases()); 749 for (auto Case : SI->cases()) 750 Cases.push_back(ValueEqualityComparisonCase(Case.getCaseValue(), 751 Case.getCaseSuccessor())); 752 return SI->getDefaultDest(); 753 } 754 755 BranchInst *BI = cast<BranchInst>(TI); 756 ICmpInst *ICI = cast<ICmpInst>(BI->getCondition()); 757 BasicBlock *Succ = BI->getSuccessor(ICI->getPredicate() == ICmpInst::ICMP_NE); 758 Cases.push_back(ValueEqualityComparisonCase( 759 GetConstantInt(ICI->getOperand(1), DL), Succ)); 760 return BI->getSuccessor(ICI->getPredicate() == ICmpInst::ICMP_EQ); 761 } 762 763 /// Given a vector of bb/value pairs, remove any entries 764 /// in the list that match the specified block. 765 static void 766 EliminateBlockCases(BasicBlock *BB, 767 std::vector<ValueEqualityComparisonCase> &Cases) { 768 Cases.erase(std::remove(Cases.begin(), Cases.end(), BB), Cases.end()); 769 } 770 771 /// Return true if there are any keys in C1 that exist in C2 as well. 772 static bool ValuesOverlap(std::vector<ValueEqualityComparisonCase> &C1, 773 std::vector<ValueEqualityComparisonCase> &C2) { 774 std::vector<ValueEqualityComparisonCase> *V1 = &C1, *V2 = &C2; 775 776 // Make V1 be smaller than V2. 777 if (V1->size() > V2->size()) 778 std::swap(V1, V2); 779 780 if (V1->empty()) 781 return false; 782 if (V1->size() == 1) { 783 // Just scan V2. 784 ConstantInt *TheVal = (*V1)[0].Value; 785 for (unsigned i = 0, e = V2->size(); i != e; ++i) 786 if (TheVal == (*V2)[i].Value) 787 return true; 788 } 789 790 // Otherwise, just sort both lists and compare element by element. 791 array_pod_sort(V1->begin(), V1->end()); 792 array_pod_sort(V2->begin(), V2->end()); 793 unsigned i1 = 0, i2 = 0, e1 = V1->size(), e2 = V2->size(); 794 while (i1 != e1 && i2 != e2) { 795 if ((*V1)[i1].Value == (*V2)[i2].Value) 796 return true; 797 if ((*V1)[i1].Value < (*V2)[i2].Value) 798 ++i1; 799 else 800 ++i2; 801 } 802 return false; 803 } 804 805 // Set branch weights on SwitchInst. This sets the metadata if there is at 806 // least one non-zero weight. 807 static void setBranchWeights(SwitchInst *SI, ArrayRef<uint32_t> Weights) { 808 // Check that there is at least one non-zero weight. Otherwise, pass 809 // nullptr to setMetadata which will erase the existing metadata. 810 MDNode *N = nullptr; 811 if (llvm::any_of(Weights, [](uint32_t W) { return W != 0; })) 812 N = MDBuilder(SI->getParent()->getContext()).createBranchWeights(Weights); 813 SI->setMetadata(LLVMContext::MD_prof, N); 814 } 815 816 // Similar to the above, but for branch and select instructions that take 817 // exactly 2 weights. 818 static void setBranchWeights(Instruction *I, uint32_t TrueWeight, 819 uint32_t FalseWeight) { 820 assert(isa<BranchInst>(I) || isa<SelectInst>(I)); 821 // Check that there is at least one non-zero weight. Otherwise, pass 822 // nullptr to setMetadata which will erase the existing metadata. 823 MDNode *N = nullptr; 824 if (TrueWeight || FalseWeight) 825 N = MDBuilder(I->getParent()->getContext()) 826 .createBranchWeights(TrueWeight, FalseWeight); 827 I->setMetadata(LLVMContext::MD_prof, N); 828 } 829 830 /// If TI is known to be a terminator instruction and its block is known to 831 /// only have a single predecessor block, check to see if that predecessor is 832 /// also a value comparison with the same value, and if that comparison 833 /// determines the outcome of this comparison. If so, simplify TI. This does a 834 /// very limited form of jump threading. 835 bool SimplifyCFGOpt::SimplifyEqualityComparisonWithOnlyPredecessor( 836 Instruction *TI, BasicBlock *Pred, IRBuilder<> &Builder) { 837 Value *PredVal = isValueEqualityComparison(Pred->getTerminator()); 838 if (!PredVal) 839 return false; // Not a value comparison in predecessor. 840 841 Value *ThisVal = isValueEqualityComparison(TI); 842 assert(ThisVal && "This isn't a value comparison!!"); 843 if (ThisVal != PredVal) 844 return false; // Different predicates. 845 846 // TODO: Preserve branch weight metadata, similarly to how 847 // FoldValueComparisonIntoPredecessors preserves it. 848 849 // Find out information about when control will move from Pred to TI's block. 850 std::vector<ValueEqualityComparisonCase> PredCases; 851 BasicBlock *PredDef = 852 GetValueEqualityComparisonCases(Pred->getTerminator(), PredCases); 853 EliminateBlockCases(PredDef, PredCases); // Remove default from cases. 854 855 // Find information about how control leaves this block. 856 std::vector<ValueEqualityComparisonCase> ThisCases; 857 BasicBlock *ThisDef = GetValueEqualityComparisonCases(TI, ThisCases); 858 EliminateBlockCases(ThisDef, ThisCases); // Remove default from cases. 859 860 // If TI's block is the default block from Pred's comparison, potentially 861 // simplify TI based on this knowledge. 862 if (PredDef == TI->getParent()) { 863 // If we are here, we know that the value is none of those cases listed in 864 // PredCases. If there are any cases in ThisCases that are in PredCases, we 865 // can simplify TI. 866 if (!ValuesOverlap(PredCases, ThisCases)) 867 return false; 868 869 if (isa<BranchInst>(TI)) { 870 // Okay, one of the successors of this condbr is dead. Convert it to a 871 // uncond br. 872 assert(ThisCases.size() == 1 && "Branch can only have one case!"); 873 // Insert the new branch. 874 Instruction *NI = Builder.CreateBr(ThisDef); 875 (void)NI; 876 877 // Remove PHI node entries for the dead edge. 878 ThisCases[0].Dest->removePredecessor(TI->getParent()); 879 880 LLVM_DEBUG(dbgs() << "Threading pred instr: " << *Pred->getTerminator() 881 << "Through successor TI: " << *TI << "Leaving: " << *NI 882 << "\n"); 883 884 EraseTerminatorAndDCECond(TI); 885 return true; 886 } 887 888 SwitchInstProfUpdateWrapper SI = *cast<SwitchInst>(TI); 889 // Okay, TI has cases that are statically dead, prune them away. 890 SmallPtrSet<Constant *, 16> DeadCases; 891 for (unsigned i = 0, e = PredCases.size(); i != e; ++i) 892 DeadCases.insert(PredCases[i].Value); 893 894 LLVM_DEBUG(dbgs() << "Threading pred instr: " << *Pred->getTerminator() 895 << "Through successor TI: " << *TI); 896 897 for (SwitchInst::CaseIt i = SI->case_end(), e = SI->case_begin(); i != e;) { 898 --i; 899 if (DeadCases.count(i->getCaseValue())) { 900 i->getCaseSuccessor()->removePredecessor(TI->getParent()); 901 SI.removeCase(i); 902 } 903 } 904 LLVM_DEBUG(dbgs() << "Leaving: " << *TI << "\n"); 905 return true; 906 } 907 908 // Otherwise, TI's block must correspond to some matched value. Find out 909 // which value (or set of values) this is. 910 ConstantInt *TIV = nullptr; 911 BasicBlock *TIBB = TI->getParent(); 912 for (unsigned i = 0, e = PredCases.size(); i != e; ++i) 913 if (PredCases[i].Dest == TIBB) { 914 if (TIV) 915 return false; // Cannot handle multiple values coming to this block. 916 TIV = PredCases[i].Value; 917 } 918 assert(TIV && "No edge from pred to succ?"); 919 920 // Okay, we found the one constant that our value can be if we get into TI's 921 // BB. Find out which successor will unconditionally be branched to. 922 BasicBlock *TheRealDest = nullptr; 923 for (unsigned i = 0, e = ThisCases.size(); i != e; ++i) 924 if (ThisCases[i].Value == TIV) { 925 TheRealDest = ThisCases[i].Dest; 926 break; 927 } 928 929 // If not handled by any explicit cases, it is handled by the default case. 930 if (!TheRealDest) 931 TheRealDest = ThisDef; 932 933 // Remove PHI node entries for dead edges. 934 BasicBlock *CheckEdge = TheRealDest; 935 for (BasicBlock *Succ : successors(TIBB)) 936 if (Succ != CheckEdge) 937 Succ->removePredecessor(TIBB); 938 else 939 CheckEdge = nullptr; 940 941 // Insert the new branch. 942 Instruction *NI = Builder.CreateBr(TheRealDest); 943 (void)NI; 944 945 LLVM_DEBUG(dbgs() << "Threading pred instr: " << *Pred->getTerminator() 946 << "Through successor TI: " << *TI << "Leaving: " << *NI 947 << "\n"); 948 949 EraseTerminatorAndDCECond(TI); 950 return true; 951 } 952 953 namespace { 954 955 /// This class implements a stable ordering of constant 956 /// integers that does not depend on their address. This is important for 957 /// applications that sort ConstantInt's to ensure uniqueness. 958 struct ConstantIntOrdering { 959 bool operator()(const ConstantInt *LHS, const ConstantInt *RHS) const { 960 return LHS->getValue().ult(RHS->getValue()); 961 } 962 }; 963 964 } // end anonymous namespace 965 966 static int ConstantIntSortPredicate(ConstantInt *const *P1, 967 ConstantInt *const *P2) { 968 const ConstantInt *LHS = *P1; 969 const ConstantInt *RHS = *P2; 970 if (LHS == RHS) 971 return 0; 972 return LHS->getValue().ult(RHS->getValue()) ? 1 : -1; 973 } 974 975 static inline bool HasBranchWeights(const Instruction *I) { 976 MDNode *ProfMD = I->getMetadata(LLVMContext::MD_prof); 977 if (ProfMD && ProfMD->getOperand(0)) 978 if (MDString *MDS = dyn_cast<MDString>(ProfMD->getOperand(0))) 979 return MDS->getString().equals("branch_weights"); 980 981 return false; 982 } 983 984 /// Get Weights of a given terminator, the default weight is at the front 985 /// of the vector. If TI is a conditional eq, we need to swap the branch-weight 986 /// metadata. 987 static void GetBranchWeights(Instruction *TI, 988 SmallVectorImpl<uint64_t> &Weights) { 989 MDNode *MD = TI->getMetadata(LLVMContext::MD_prof); 990 assert(MD); 991 for (unsigned i = 1, e = MD->getNumOperands(); i < e; ++i) { 992 ConstantInt *CI = mdconst::extract<ConstantInt>(MD->getOperand(i)); 993 Weights.push_back(CI->getValue().getZExtValue()); 994 } 995 996 // If TI is a conditional eq, the default case is the false case, 997 // and the corresponding branch-weight data is at index 2. We swap the 998 // default weight to be the first entry. 999 if (BranchInst *BI = dyn_cast<BranchInst>(TI)) { 1000 assert(Weights.size() == 2); 1001 ICmpInst *ICI = cast<ICmpInst>(BI->getCondition()); 1002 if (ICI->getPredicate() == ICmpInst::ICMP_EQ) 1003 std::swap(Weights.front(), Weights.back()); 1004 } 1005 } 1006 1007 /// Keep halving the weights until all can fit in uint32_t. 1008 static void FitWeights(MutableArrayRef<uint64_t> Weights) { 1009 uint64_t Max = *std::max_element(Weights.begin(), Weights.end()); 1010 if (Max > UINT_MAX) { 1011 unsigned Offset = 32 - countLeadingZeros(Max); 1012 for (uint64_t &I : Weights) 1013 I >>= Offset; 1014 } 1015 } 1016 1017 /// The specified terminator is a value equality comparison instruction 1018 /// (either a switch or a branch on "X == c"). 1019 /// See if any of the predecessors of the terminator block are value comparisons 1020 /// on the same value. If so, and if safe to do so, fold them together. 1021 bool SimplifyCFGOpt::FoldValueComparisonIntoPredecessors(Instruction *TI, 1022 IRBuilder<> &Builder) { 1023 BasicBlock *BB = TI->getParent(); 1024 Value *CV = isValueEqualityComparison(TI); // CondVal 1025 assert(CV && "Not a comparison?"); 1026 bool Changed = false; 1027 1028 SmallVector<BasicBlock *, 16> Preds(pred_begin(BB), pred_end(BB)); 1029 while (!Preds.empty()) { 1030 BasicBlock *Pred = Preds.pop_back_val(); 1031 1032 // See if the predecessor is a comparison with the same value. 1033 Instruction *PTI = Pred->getTerminator(); 1034 Value *PCV = isValueEqualityComparison(PTI); // PredCondVal 1035 1036 if (PCV == CV && TI != PTI) { 1037 SmallSetVector<BasicBlock*, 4> FailBlocks; 1038 if (!SafeToMergeTerminators(TI, PTI, &FailBlocks)) { 1039 for (auto *Succ : FailBlocks) { 1040 if (!SplitBlockPredecessors(Succ, TI->getParent(), ".fold.split")) 1041 return false; 1042 } 1043 } 1044 1045 // Figure out which 'cases' to copy from SI to PSI. 1046 std::vector<ValueEqualityComparisonCase> BBCases; 1047 BasicBlock *BBDefault = GetValueEqualityComparisonCases(TI, BBCases); 1048 1049 std::vector<ValueEqualityComparisonCase> PredCases; 1050 BasicBlock *PredDefault = GetValueEqualityComparisonCases(PTI, PredCases); 1051 1052 // Based on whether the default edge from PTI goes to BB or not, fill in 1053 // PredCases and PredDefault with the new switch cases we would like to 1054 // build. 1055 SmallVector<BasicBlock *, 8> NewSuccessors; 1056 1057 // Update the branch weight metadata along the way 1058 SmallVector<uint64_t, 8> Weights; 1059 bool PredHasWeights = HasBranchWeights(PTI); 1060 bool SuccHasWeights = HasBranchWeights(TI); 1061 1062 if (PredHasWeights) { 1063 GetBranchWeights(PTI, Weights); 1064 // branch-weight metadata is inconsistent here. 1065 if (Weights.size() != 1 + PredCases.size()) 1066 PredHasWeights = SuccHasWeights = false; 1067 } else if (SuccHasWeights) 1068 // If there are no predecessor weights but there are successor weights, 1069 // populate Weights with 1, which will later be scaled to the sum of 1070 // successor's weights 1071 Weights.assign(1 + PredCases.size(), 1); 1072 1073 SmallVector<uint64_t, 8> SuccWeights; 1074 if (SuccHasWeights) { 1075 GetBranchWeights(TI, SuccWeights); 1076 // branch-weight metadata is inconsistent here. 1077 if (SuccWeights.size() != 1 + BBCases.size()) 1078 PredHasWeights = SuccHasWeights = false; 1079 } else if (PredHasWeights) 1080 SuccWeights.assign(1 + BBCases.size(), 1); 1081 1082 if (PredDefault == BB) { 1083 // If this is the default destination from PTI, only the edges in TI 1084 // that don't occur in PTI, or that branch to BB will be activated. 1085 std::set<ConstantInt *, ConstantIntOrdering> PTIHandled; 1086 for (unsigned i = 0, e = PredCases.size(); i != e; ++i) 1087 if (PredCases[i].Dest != BB) 1088 PTIHandled.insert(PredCases[i].Value); 1089 else { 1090 // The default destination is BB, we don't need explicit targets. 1091 std::swap(PredCases[i], PredCases.back()); 1092 1093 if (PredHasWeights || SuccHasWeights) { 1094 // Increase weight for the default case. 1095 Weights[0] += Weights[i + 1]; 1096 std::swap(Weights[i + 1], Weights.back()); 1097 Weights.pop_back(); 1098 } 1099 1100 PredCases.pop_back(); 1101 --i; 1102 --e; 1103 } 1104 1105 // Reconstruct the new switch statement we will be building. 1106 if (PredDefault != BBDefault) { 1107 PredDefault->removePredecessor(Pred); 1108 PredDefault = BBDefault; 1109 NewSuccessors.push_back(BBDefault); 1110 } 1111 1112 unsigned CasesFromPred = Weights.size(); 1113 uint64_t ValidTotalSuccWeight = 0; 1114 for (unsigned i = 0, e = BBCases.size(); i != e; ++i) 1115 if (!PTIHandled.count(BBCases[i].Value) && 1116 BBCases[i].Dest != BBDefault) { 1117 PredCases.push_back(BBCases[i]); 1118 NewSuccessors.push_back(BBCases[i].Dest); 1119 if (SuccHasWeights || PredHasWeights) { 1120 // The default weight is at index 0, so weight for the ith case 1121 // should be at index i+1. Scale the cases from successor by 1122 // PredDefaultWeight (Weights[0]). 1123 Weights.push_back(Weights[0] * SuccWeights[i + 1]); 1124 ValidTotalSuccWeight += SuccWeights[i + 1]; 1125 } 1126 } 1127 1128 if (SuccHasWeights || PredHasWeights) { 1129 ValidTotalSuccWeight += SuccWeights[0]; 1130 // Scale the cases from predecessor by ValidTotalSuccWeight. 1131 for (unsigned i = 1; i < CasesFromPred; ++i) 1132 Weights[i] *= ValidTotalSuccWeight; 1133 // Scale the default weight by SuccDefaultWeight (SuccWeights[0]). 1134 Weights[0] *= SuccWeights[0]; 1135 } 1136 } else { 1137 // If this is not the default destination from PSI, only the edges 1138 // in SI that occur in PSI with a destination of BB will be 1139 // activated. 1140 std::set<ConstantInt *, ConstantIntOrdering> PTIHandled; 1141 std::map<ConstantInt *, uint64_t> WeightsForHandled; 1142 for (unsigned i = 0, e = PredCases.size(); i != e; ++i) 1143 if (PredCases[i].Dest == BB) { 1144 PTIHandled.insert(PredCases[i].Value); 1145 1146 if (PredHasWeights || SuccHasWeights) { 1147 WeightsForHandled[PredCases[i].Value] = Weights[i + 1]; 1148 std::swap(Weights[i + 1], Weights.back()); 1149 Weights.pop_back(); 1150 } 1151 1152 std::swap(PredCases[i], PredCases.back()); 1153 PredCases.pop_back(); 1154 --i; 1155 --e; 1156 } 1157 1158 // Okay, now we know which constants were sent to BB from the 1159 // predecessor. Figure out where they will all go now. 1160 for (unsigned i = 0, e = BBCases.size(); i != e; ++i) 1161 if (PTIHandled.count(BBCases[i].Value)) { 1162 // If this is one we are capable of getting... 1163 if (PredHasWeights || SuccHasWeights) 1164 Weights.push_back(WeightsForHandled[BBCases[i].Value]); 1165 PredCases.push_back(BBCases[i]); 1166 NewSuccessors.push_back(BBCases[i].Dest); 1167 PTIHandled.erase( 1168 BBCases[i].Value); // This constant is taken care of 1169 } 1170 1171 // If there are any constants vectored to BB that TI doesn't handle, 1172 // they must go to the default destination of TI. 1173 for (ConstantInt *I : PTIHandled) { 1174 if (PredHasWeights || SuccHasWeights) 1175 Weights.push_back(WeightsForHandled[I]); 1176 PredCases.push_back(ValueEqualityComparisonCase(I, BBDefault)); 1177 NewSuccessors.push_back(BBDefault); 1178 } 1179 } 1180 1181 // Okay, at this point, we know which new successor Pred will get. Make 1182 // sure we update the number of entries in the PHI nodes for these 1183 // successors. 1184 for (BasicBlock *NewSuccessor : NewSuccessors) 1185 AddPredecessorToBlock(NewSuccessor, Pred, BB); 1186 1187 Builder.SetInsertPoint(PTI); 1188 // Convert pointer to int before we switch. 1189 if (CV->getType()->isPointerTy()) { 1190 CV = Builder.CreatePtrToInt(CV, DL.getIntPtrType(CV->getType()), 1191 "magicptr"); 1192 } 1193 1194 // Now that the successors are updated, create the new Switch instruction. 1195 SwitchInst *NewSI = 1196 Builder.CreateSwitch(CV, PredDefault, PredCases.size()); 1197 NewSI->setDebugLoc(PTI->getDebugLoc()); 1198 for (ValueEqualityComparisonCase &V : PredCases) 1199 NewSI->addCase(V.Value, V.Dest); 1200 1201 if (PredHasWeights || SuccHasWeights) { 1202 // Halve the weights if any of them cannot fit in an uint32_t 1203 FitWeights(Weights); 1204 1205 SmallVector<uint32_t, 8> MDWeights(Weights.begin(), Weights.end()); 1206 1207 setBranchWeights(NewSI, MDWeights); 1208 } 1209 1210 EraseTerminatorAndDCECond(PTI); 1211 1212 // Okay, last check. If BB is still a successor of PSI, then we must 1213 // have an infinite loop case. If so, add an infinitely looping block 1214 // to handle the case to preserve the behavior of the code. 1215 BasicBlock *InfLoopBlock = nullptr; 1216 for (unsigned i = 0, e = NewSI->getNumSuccessors(); i != e; ++i) 1217 if (NewSI->getSuccessor(i) == BB) { 1218 if (!InfLoopBlock) { 1219 // Insert it at the end of the function, because it's either code, 1220 // or it won't matter if it's hot. :) 1221 InfLoopBlock = BasicBlock::Create(BB->getContext(), "infloop", 1222 BB->getParent()); 1223 BranchInst::Create(InfLoopBlock, InfLoopBlock); 1224 } 1225 NewSI->setSuccessor(i, InfLoopBlock); 1226 } 1227 1228 Changed = true; 1229 } 1230 } 1231 return Changed; 1232 } 1233 1234 // If we would need to insert a select that uses the value of this invoke 1235 // (comments in HoistThenElseCodeToIf explain why we would need to do this), we 1236 // can't hoist the invoke, as there is nowhere to put the select in this case. 1237 static bool isSafeToHoistInvoke(BasicBlock *BB1, BasicBlock *BB2, 1238 Instruction *I1, Instruction *I2) { 1239 for (BasicBlock *Succ : successors(BB1)) { 1240 for (const PHINode &PN : Succ->phis()) { 1241 Value *BB1V = PN.getIncomingValueForBlock(BB1); 1242 Value *BB2V = PN.getIncomingValueForBlock(BB2); 1243 if (BB1V != BB2V && (BB1V == I1 || BB2V == I2)) { 1244 return false; 1245 } 1246 } 1247 } 1248 return true; 1249 } 1250 1251 static bool passingValueIsAlwaysUndefined(Value *V, Instruction *I); 1252 1253 /// Given a conditional branch that goes to BB1 and BB2, hoist any common code 1254 /// in the two blocks up into the branch block. The caller of this function 1255 /// guarantees that BI's block dominates BB1 and BB2. 1256 bool SimplifyCFGOpt::HoistThenElseCodeToIf(BranchInst *BI, 1257 const TargetTransformInfo &TTI) { 1258 // This does very trivial matching, with limited scanning, to find identical 1259 // instructions in the two blocks. In particular, we don't want to get into 1260 // O(M*N) situations here where M and N are the sizes of BB1 and BB2. As 1261 // such, we currently just scan for obviously identical instructions in an 1262 // identical order. 1263 BasicBlock *BB1 = BI->getSuccessor(0); // The true destination. 1264 BasicBlock *BB2 = BI->getSuccessor(1); // The false destination 1265 1266 BasicBlock::iterator BB1_Itr = BB1->begin(); 1267 BasicBlock::iterator BB2_Itr = BB2->begin(); 1268 1269 Instruction *I1 = &*BB1_Itr++, *I2 = &*BB2_Itr++; 1270 // Skip debug info if it is not identical. 1271 DbgInfoIntrinsic *DBI1 = dyn_cast<DbgInfoIntrinsic>(I1); 1272 DbgInfoIntrinsic *DBI2 = dyn_cast<DbgInfoIntrinsic>(I2); 1273 if (!DBI1 || !DBI2 || !DBI1->isIdenticalToWhenDefined(DBI2)) { 1274 while (isa<DbgInfoIntrinsic>(I1)) 1275 I1 = &*BB1_Itr++; 1276 while (isa<DbgInfoIntrinsic>(I2)) 1277 I2 = &*BB2_Itr++; 1278 } 1279 // FIXME: Can we define a safety predicate for CallBr? 1280 if (isa<PHINode>(I1) || !I1->isIdenticalToWhenDefined(I2) || 1281 (isa<InvokeInst>(I1) && !isSafeToHoistInvoke(BB1, BB2, I1, I2)) || 1282 isa<CallBrInst>(I1)) 1283 return false; 1284 1285 BasicBlock *BIParent = BI->getParent(); 1286 1287 bool Changed = false; 1288 do { 1289 // If we are hoisting the terminator instruction, don't move one (making a 1290 // broken BB), instead clone it, and remove BI. 1291 if (I1->isTerminator()) 1292 goto HoistTerminator; 1293 1294 // If we're going to hoist a call, make sure that the two instructions we're 1295 // commoning/hoisting are both marked with musttail, or neither of them is 1296 // marked as such. Otherwise, we might end up in a situation where we hoist 1297 // from a block where the terminator is a `ret` to a block where the terminator 1298 // is a `br`, and `musttail` calls expect to be followed by a return. 1299 auto *C1 = dyn_cast<CallInst>(I1); 1300 auto *C2 = dyn_cast<CallInst>(I2); 1301 if (C1 && C2) 1302 if (C1->isMustTailCall() != C2->isMustTailCall()) 1303 return Changed; 1304 1305 if (!TTI.isProfitableToHoist(I1) || !TTI.isProfitableToHoist(I2)) 1306 return Changed; 1307 1308 // If any of the two call sites has nomerge attribute, stop hoisting. 1309 if (const auto *CB1 = dyn_cast<CallBase>(I1)) 1310 if (CB1->cannotMerge()) 1311 return Changed; 1312 if (const auto *CB2 = dyn_cast<CallBase>(I2)) 1313 if (CB2->cannotMerge()) 1314 return Changed; 1315 1316 if (isa<DbgInfoIntrinsic>(I1) || isa<DbgInfoIntrinsic>(I2)) { 1317 assert (isa<DbgInfoIntrinsic>(I1) && isa<DbgInfoIntrinsic>(I2)); 1318 // The debug location is an integral part of a debug info intrinsic 1319 // and can't be separated from it or replaced. Instead of attempting 1320 // to merge locations, simply hoist both copies of the intrinsic. 1321 BIParent->getInstList().splice(BI->getIterator(), 1322 BB1->getInstList(), I1); 1323 BIParent->getInstList().splice(BI->getIterator(), 1324 BB2->getInstList(), I2); 1325 Changed = true; 1326 } else { 1327 // For a normal instruction, we just move one to right before the branch, 1328 // then replace all uses of the other with the first. Finally, we remove 1329 // the now redundant second instruction. 1330 BIParent->getInstList().splice(BI->getIterator(), 1331 BB1->getInstList(), I1); 1332 if (!I2->use_empty()) 1333 I2->replaceAllUsesWith(I1); 1334 I1->andIRFlags(I2); 1335 unsigned KnownIDs[] = {LLVMContext::MD_tbaa, 1336 LLVMContext::MD_range, 1337 LLVMContext::MD_fpmath, 1338 LLVMContext::MD_invariant_load, 1339 LLVMContext::MD_nonnull, 1340 LLVMContext::MD_invariant_group, 1341 LLVMContext::MD_align, 1342 LLVMContext::MD_dereferenceable, 1343 LLVMContext::MD_dereferenceable_or_null, 1344 LLVMContext::MD_mem_parallel_loop_access, 1345 LLVMContext::MD_access_group, 1346 LLVMContext::MD_preserve_access_index}; 1347 combineMetadata(I1, I2, KnownIDs, true); 1348 1349 // I1 and I2 are being combined into a single instruction. Its debug 1350 // location is the merged locations of the original instructions. 1351 I1->applyMergedLocation(I1->getDebugLoc(), I2->getDebugLoc()); 1352 1353 I2->eraseFromParent(); 1354 Changed = true; 1355 } 1356 1357 I1 = &*BB1_Itr++; 1358 I2 = &*BB2_Itr++; 1359 // Skip debug info if it is not identical. 1360 DbgInfoIntrinsic *DBI1 = dyn_cast<DbgInfoIntrinsic>(I1); 1361 DbgInfoIntrinsic *DBI2 = dyn_cast<DbgInfoIntrinsic>(I2); 1362 if (!DBI1 || !DBI2 || !DBI1->isIdenticalToWhenDefined(DBI2)) { 1363 while (isa<DbgInfoIntrinsic>(I1)) 1364 I1 = &*BB1_Itr++; 1365 while (isa<DbgInfoIntrinsic>(I2)) 1366 I2 = &*BB2_Itr++; 1367 } 1368 } while (I1->isIdenticalToWhenDefined(I2)); 1369 1370 return true; 1371 1372 HoistTerminator: 1373 // It may not be possible to hoist an invoke. 1374 // FIXME: Can we define a safety predicate for CallBr? 1375 if (isa<InvokeInst>(I1) && !isSafeToHoistInvoke(BB1, BB2, I1, I2)) 1376 return Changed; 1377 1378 // TODO: callbr hoisting currently disabled pending further study. 1379 if (isa<CallBrInst>(I1)) 1380 return Changed; 1381 1382 for (BasicBlock *Succ : successors(BB1)) { 1383 for (PHINode &PN : Succ->phis()) { 1384 Value *BB1V = PN.getIncomingValueForBlock(BB1); 1385 Value *BB2V = PN.getIncomingValueForBlock(BB2); 1386 if (BB1V == BB2V) 1387 continue; 1388 1389 // Check for passingValueIsAlwaysUndefined here because we would rather 1390 // eliminate undefined control flow then converting it to a select. 1391 if (passingValueIsAlwaysUndefined(BB1V, &PN) || 1392 passingValueIsAlwaysUndefined(BB2V, &PN)) 1393 return Changed; 1394 1395 if (isa<ConstantExpr>(BB1V) && !isSafeToSpeculativelyExecute(BB1V)) 1396 return Changed; 1397 if (isa<ConstantExpr>(BB2V) && !isSafeToSpeculativelyExecute(BB2V)) 1398 return Changed; 1399 } 1400 } 1401 1402 // Okay, it is safe to hoist the terminator. 1403 Instruction *NT = I1->clone(); 1404 BIParent->getInstList().insert(BI->getIterator(), NT); 1405 if (!NT->getType()->isVoidTy()) { 1406 I1->replaceAllUsesWith(NT); 1407 I2->replaceAllUsesWith(NT); 1408 NT->takeName(I1); 1409 } 1410 1411 // Ensure terminator gets a debug location, even an unknown one, in case 1412 // it involves inlinable calls. 1413 NT->applyMergedLocation(I1->getDebugLoc(), I2->getDebugLoc()); 1414 1415 // PHIs created below will adopt NT's merged DebugLoc. 1416 IRBuilder<NoFolder> Builder(NT); 1417 1418 // Hoisting one of the terminators from our successor is a great thing. 1419 // Unfortunately, the successors of the if/else blocks may have PHI nodes in 1420 // them. If they do, all PHI entries for BB1/BB2 must agree for all PHI 1421 // nodes, so we insert select instruction to compute the final result. 1422 std::map<std::pair<Value *, Value *>, SelectInst *> InsertedSelects; 1423 for (BasicBlock *Succ : successors(BB1)) { 1424 for (PHINode &PN : Succ->phis()) { 1425 Value *BB1V = PN.getIncomingValueForBlock(BB1); 1426 Value *BB2V = PN.getIncomingValueForBlock(BB2); 1427 if (BB1V == BB2V) 1428 continue; 1429 1430 // These values do not agree. Insert a select instruction before NT 1431 // that determines the right value. 1432 SelectInst *&SI = InsertedSelects[std::make_pair(BB1V, BB2V)]; 1433 if (!SI) { 1434 // Propagate fast-math-flags from phi node to its replacement select. 1435 IRBuilder<>::FastMathFlagGuard FMFGuard(Builder); 1436 if (isa<FPMathOperator>(PN)) 1437 Builder.setFastMathFlags(PN.getFastMathFlags()); 1438 1439 SI = cast<SelectInst>( 1440 Builder.CreateSelect(BI->getCondition(), BB1V, BB2V, 1441 BB1V->getName() + "." + BB2V->getName(), BI)); 1442 } 1443 1444 // Make the PHI node use the select for all incoming values for BB1/BB2 1445 for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i) 1446 if (PN.getIncomingBlock(i) == BB1 || PN.getIncomingBlock(i) == BB2) 1447 PN.setIncomingValue(i, SI); 1448 } 1449 } 1450 1451 // Update any PHI nodes in our new successors. 1452 for (BasicBlock *Succ : successors(BB1)) 1453 AddPredecessorToBlock(Succ, BIParent, BB1); 1454 1455 EraseTerminatorAndDCECond(BI); 1456 return true; 1457 } 1458 1459 // Check lifetime markers. 1460 static bool isLifeTimeMarker(const Instruction *I) { 1461 if (auto II = dyn_cast<IntrinsicInst>(I)) { 1462 switch (II->getIntrinsicID()) { 1463 default: 1464 break; 1465 case Intrinsic::lifetime_start: 1466 case Intrinsic::lifetime_end: 1467 return true; 1468 } 1469 } 1470 return false; 1471 } 1472 1473 // TODO: Refine this. This should avoid cases like turning constant memcpy sizes 1474 // into variables. 1475 static bool replacingOperandWithVariableIsCheap(const Instruction *I, 1476 int OpIdx) { 1477 return !isa<IntrinsicInst>(I); 1478 } 1479 1480 // All instructions in Insts belong to different blocks that all unconditionally 1481 // branch to a common successor. Analyze each instruction and return true if it 1482 // would be possible to sink them into their successor, creating one common 1483 // instruction instead. For every value that would be required to be provided by 1484 // PHI node (because an operand varies in each input block), add to PHIOperands. 1485 static bool canSinkInstructions( 1486 ArrayRef<Instruction *> Insts, 1487 DenseMap<Instruction *, SmallVector<Value *, 4>> &PHIOperands) { 1488 // Prune out obviously bad instructions to move. Each instruction must have 1489 // exactly zero or one use, and we check later that use is by a single, common 1490 // PHI instruction in the successor. 1491 bool HasUse = !Insts.front()->user_empty(); 1492 for (auto *I : Insts) { 1493 // These instructions may change or break semantics if moved. 1494 if (isa<PHINode>(I) || I->isEHPad() || isa<AllocaInst>(I) || 1495 I->getType()->isTokenTy()) 1496 return false; 1497 1498 // Conservatively return false if I is an inline-asm instruction. Sinking 1499 // and merging inline-asm instructions can potentially create arguments 1500 // that cannot satisfy the inline-asm constraints. 1501 // If the instruction has nomerge attribute, return false. 1502 if (const auto *C = dyn_cast<CallBase>(I)) 1503 if (C->isInlineAsm() || C->cannotMerge()) 1504 return false; 1505 1506 // Each instruction must have zero or one use. 1507 if (HasUse && !I->hasOneUse()) 1508 return false; 1509 if (!HasUse && !I->user_empty()) 1510 return false; 1511 } 1512 1513 const Instruction *I0 = Insts.front(); 1514 for (auto *I : Insts) 1515 if (!I->isSameOperationAs(I0)) 1516 return false; 1517 1518 // All instructions in Insts are known to be the same opcode. If they have a 1519 // use, check that the only user is a PHI or in the same block as the 1520 // instruction, because if a user is in the same block as an instruction we're 1521 // contemplating sinking, it must already be determined to be sinkable. 1522 if (HasUse) { 1523 auto *PNUse = dyn_cast<PHINode>(*I0->user_begin()); 1524 auto *Succ = I0->getParent()->getTerminator()->getSuccessor(0); 1525 if (!all_of(Insts, [&PNUse,&Succ](const Instruction *I) -> bool { 1526 auto *U = cast<Instruction>(*I->user_begin()); 1527 return (PNUse && 1528 PNUse->getParent() == Succ && 1529 PNUse->getIncomingValueForBlock(I->getParent()) == I) || 1530 U->getParent() == I->getParent(); 1531 })) 1532 return false; 1533 } 1534 1535 // Because SROA can't handle speculating stores of selects, try not to sink 1536 // loads, stores or lifetime markers of allocas when we'd have to create a 1537 // PHI for the address operand. Also, because it is likely that loads or 1538 // stores of allocas will disappear when Mem2Reg/SROA is run, don't sink 1539 // them. 1540 // This can cause code churn which can have unintended consequences down 1541 // the line - see https://llvm.org/bugs/show_bug.cgi?id=30244. 1542 // FIXME: This is a workaround for a deficiency in SROA - see 1543 // https://llvm.org/bugs/show_bug.cgi?id=30188 1544 if (isa<StoreInst>(I0) && any_of(Insts, [](const Instruction *I) { 1545 return isa<AllocaInst>(I->getOperand(1)->stripPointerCasts()); 1546 })) 1547 return false; 1548 if (isa<LoadInst>(I0) && any_of(Insts, [](const Instruction *I) { 1549 return isa<AllocaInst>(I->getOperand(0)->stripPointerCasts()); 1550 })) 1551 return false; 1552 if (isLifeTimeMarker(I0) && any_of(Insts, [](const Instruction *I) { 1553 return isa<AllocaInst>(I->getOperand(1)->stripPointerCasts()); 1554 })) 1555 return false; 1556 1557 for (unsigned OI = 0, OE = I0->getNumOperands(); OI != OE; ++OI) { 1558 Value *Op = I0->getOperand(OI); 1559 if (Op->getType()->isTokenTy()) 1560 // Don't touch any operand of token type. 1561 return false; 1562 1563 auto SameAsI0 = [&I0, OI](const Instruction *I) { 1564 assert(I->getNumOperands() == I0->getNumOperands()); 1565 return I->getOperand(OI) == I0->getOperand(OI); 1566 }; 1567 if (!all_of(Insts, SameAsI0)) { 1568 if ((isa<Constant>(Op) && !replacingOperandWithVariableIsCheap(I0, OI)) || 1569 !canReplaceOperandWithVariable(I0, OI)) 1570 // We can't create a PHI from this GEP. 1571 return false; 1572 // Don't create indirect calls! The called value is the final operand. 1573 if (isa<CallBase>(I0) && OI == OE - 1) { 1574 // FIXME: if the call was *already* indirect, we should do this. 1575 return false; 1576 } 1577 for (auto *I : Insts) 1578 PHIOperands[I].push_back(I->getOperand(OI)); 1579 } 1580 } 1581 return true; 1582 } 1583 1584 // Assuming canSinkLastInstruction(Blocks) has returned true, sink the last 1585 // instruction of every block in Blocks to their common successor, commoning 1586 // into one instruction. 1587 static bool sinkLastInstruction(ArrayRef<BasicBlock*> Blocks) { 1588 auto *BBEnd = Blocks[0]->getTerminator()->getSuccessor(0); 1589 1590 // canSinkLastInstruction returning true guarantees that every block has at 1591 // least one non-terminator instruction. 1592 SmallVector<Instruction*,4> Insts; 1593 for (auto *BB : Blocks) { 1594 Instruction *I = BB->getTerminator(); 1595 do { 1596 I = I->getPrevNode(); 1597 } while (isa<DbgInfoIntrinsic>(I) && I != &BB->front()); 1598 if (!isa<DbgInfoIntrinsic>(I)) 1599 Insts.push_back(I); 1600 } 1601 1602 // The only checking we need to do now is that all users of all instructions 1603 // are the same PHI node. canSinkLastInstruction should have checked this but 1604 // it is slightly over-aggressive - it gets confused by commutative instructions 1605 // so double-check it here. 1606 Instruction *I0 = Insts.front(); 1607 if (!I0->user_empty()) { 1608 auto *PNUse = dyn_cast<PHINode>(*I0->user_begin()); 1609 if (!all_of(Insts, [&PNUse](const Instruction *I) -> bool { 1610 auto *U = cast<Instruction>(*I->user_begin()); 1611 return U == PNUse; 1612 })) 1613 return false; 1614 } 1615 1616 // We don't need to do any more checking here; canSinkLastInstruction should 1617 // have done it all for us. 1618 SmallVector<Value*, 4> NewOperands; 1619 for (unsigned O = 0, E = I0->getNumOperands(); O != E; ++O) { 1620 // This check is different to that in canSinkLastInstruction. There, we 1621 // cared about the global view once simplifycfg (and instcombine) have 1622 // completed - it takes into account PHIs that become trivially 1623 // simplifiable. However here we need a more local view; if an operand 1624 // differs we create a PHI and rely on instcombine to clean up the very 1625 // small mess we may make. 1626 bool NeedPHI = any_of(Insts, [&I0, O](const Instruction *I) { 1627 return I->getOperand(O) != I0->getOperand(O); 1628 }); 1629 if (!NeedPHI) { 1630 NewOperands.push_back(I0->getOperand(O)); 1631 continue; 1632 } 1633 1634 // Create a new PHI in the successor block and populate it. 1635 auto *Op = I0->getOperand(O); 1636 assert(!Op->getType()->isTokenTy() && "Can't PHI tokens!"); 1637 auto *PN = PHINode::Create(Op->getType(), Insts.size(), 1638 Op->getName() + ".sink", &BBEnd->front()); 1639 for (auto *I : Insts) 1640 PN->addIncoming(I->getOperand(O), I->getParent()); 1641 NewOperands.push_back(PN); 1642 } 1643 1644 // Arbitrarily use I0 as the new "common" instruction; remap its operands 1645 // and move it to the start of the successor block. 1646 for (unsigned O = 0, E = I0->getNumOperands(); O != E; ++O) 1647 I0->getOperandUse(O).set(NewOperands[O]); 1648 I0->moveBefore(&*BBEnd->getFirstInsertionPt()); 1649 1650 // Update metadata and IR flags, and merge debug locations. 1651 for (auto *I : Insts) 1652 if (I != I0) { 1653 // The debug location for the "common" instruction is the merged locations 1654 // of all the commoned instructions. We start with the original location 1655 // of the "common" instruction and iteratively merge each location in the 1656 // loop below. 1657 // This is an N-way merge, which will be inefficient if I0 is a CallInst. 1658 // However, as N-way merge for CallInst is rare, so we use simplified API 1659 // instead of using complex API for N-way merge. 1660 I0->applyMergedLocation(I0->getDebugLoc(), I->getDebugLoc()); 1661 combineMetadataForCSE(I0, I, true); 1662 I0->andIRFlags(I); 1663 } 1664 1665 if (!I0->user_empty()) { 1666 // canSinkLastInstruction checked that all instructions were used by 1667 // one and only one PHI node. Find that now, RAUW it to our common 1668 // instruction and nuke it. 1669 auto *PN = cast<PHINode>(*I0->user_begin()); 1670 PN->replaceAllUsesWith(I0); 1671 PN->eraseFromParent(); 1672 } 1673 1674 // Finally nuke all instructions apart from the common instruction. 1675 for (auto *I : Insts) 1676 if (I != I0) 1677 I->eraseFromParent(); 1678 1679 return true; 1680 } 1681 1682 namespace { 1683 1684 // LockstepReverseIterator - Iterates through instructions 1685 // in a set of blocks in reverse order from the first non-terminator. 1686 // For example (assume all blocks have size n): 1687 // LockstepReverseIterator I([B1, B2, B3]); 1688 // *I-- = [B1[n], B2[n], B3[n]]; 1689 // *I-- = [B1[n-1], B2[n-1], B3[n-1]]; 1690 // *I-- = [B1[n-2], B2[n-2], B3[n-2]]; 1691 // ... 1692 class LockstepReverseIterator { 1693 ArrayRef<BasicBlock*> Blocks; 1694 SmallVector<Instruction*,4> Insts; 1695 bool Fail; 1696 1697 public: 1698 LockstepReverseIterator(ArrayRef<BasicBlock*> Blocks) : Blocks(Blocks) { 1699 reset(); 1700 } 1701 1702 void reset() { 1703 Fail = false; 1704 Insts.clear(); 1705 for (auto *BB : Blocks) { 1706 Instruction *Inst = BB->getTerminator(); 1707 for (Inst = Inst->getPrevNode(); Inst && isa<DbgInfoIntrinsic>(Inst);) 1708 Inst = Inst->getPrevNode(); 1709 if (!Inst) { 1710 // Block wasn't big enough. 1711 Fail = true; 1712 return; 1713 } 1714 Insts.push_back(Inst); 1715 } 1716 } 1717 1718 bool isValid() const { 1719 return !Fail; 1720 } 1721 1722 void operator--() { 1723 if (Fail) 1724 return; 1725 for (auto *&Inst : Insts) { 1726 for (Inst = Inst->getPrevNode(); Inst && isa<DbgInfoIntrinsic>(Inst);) 1727 Inst = Inst->getPrevNode(); 1728 // Already at beginning of block. 1729 if (!Inst) { 1730 Fail = true; 1731 return; 1732 } 1733 } 1734 } 1735 1736 ArrayRef<Instruction*> operator * () const { 1737 return Insts; 1738 } 1739 }; 1740 1741 } // end anonymous namespace 1742 1743 /// Check whether BB's predecessors end with unconditional branches. If it is 1744 /// true, sink any common code from the predecessors to BB. 1745 /// We also allow one predecessor to end with conditional branch (but no more 1746 /// than one). 1747 static bool SinkCommonCodeFromPredecessors(BasicBlock *BB) { 1748 // We support two situations: 1749 // (1) all incoming arcs are unconditional 1750 // (2) one incoming arc is conditional 1751 // 1752 // (2) is very common in switch defaults and 1753 // else-if patterns; 1754 // 1755 // if (a) f(1); 1756 // else if (b) f(2); 1757 // 1758 // produces: 1759 // 1760 // [if] 1761 // / \ 1762 // [f(1)] [if] 1763 // | | \ 1764 // | | | 1765 // | [f(2)]| 1766 // \ | / 1767 // [ end ] 1768 // 1769 // [end] has two unconditional predecessor arcs and one conditional. The 1770 // conditional refers to the implicit empty 'else' arc. This conditional 1771 // arc can also be caused by an empty default block in a switch. 1772 // 1773 // In this case, we attempt to sink code from all *unconditional* arcs. 1774 // If we can sink instructions from these arcs (determined during the scan 1775 // phase below) we insert a common successor for all unconditional arcs and 1776 // connect that to [end], to enable sinking: 1777 // 1778 // [if] 1779 // / \ 1780 // [x(1)] [if] 1781 // | | \ 1782 // | | \ 1783 // | [x(2)] | 1784 // \ / | 1785 // [sink.split] | 1786 // \ / 1787 // [ end ] 1788 // 1789 SmallVector<BasicBlock*,4> UnconditionalPreds; 1790 Instruction *Cond = nullptr; 1791 for (auto *B : predecessors(BB)) { 1792 auto *T = B->getTerminator(); 1793 if (isa<BranchInst>(T) && cast<BranchInst>(T)->isUnconditional()) 1794 UnconditionalPreds.push_back(B); 1795 else if ((isa<BranchInst>(T) || isa<SwitchInst>(T)) && !Cond) 1796 Cond = T; 1797 else 1798 return false; 1799 } 1800 if (UnconditionalPreds.size() < 2) 1801 return false; 1802 1803 bool Changed = false; 1804 // We take a two-step approach to tail sinking. First we scan from the end of 1805 // each block upwards in lockstep. If the n'th instruction from the end of each 1806 // block can be sunk, those instructions are added to ValuesToSink and we 1807 // carry on. If we can sink an instruction but need to PHI-merge some operands 1808 // (because they're not identical in each instruction) we add these to 1809 // PHIOperands. 1810 unsigned ScanIdx = 0; 1811 SmallPtrSet<Value*,4> InstructionsToSink; 1812 DenseMap<Instruction*, SmallVector<Value*,4>> PHIOperands; 1813 LockstepReverseIterator LRI(UnconditionalPreds); 1814 while (LRI.isValid() && 1815 canSinkInstructions(*LRI, PHIOperands)) { 1816 LLVM_DEBUG(dbgs() << "SINK: instruction can be sunk: " << *(*LRI)[0] 1817 << "\n"); 1818 InstructionsToSink.insert((*LRI).begin(), (*LRI).end()); 1819 ++ScanIdx; 1820 --LRI; 1821 } 1822 1823 auto ProfitableToSinkInstruction = [&](LockstepReverseIterator &LRI) { 1824 unsigned NumPHIdValues = 0; 1825 for (auto *I : *LRI) 1826 for (auto *V : PHIOperands[I]) 1827 if (InstructionsToSink.count(V) == 0) 1828 ++NumPHIdValues; 1829 LLVM_DEBUG(dbgs() << "SINK: #phid values: " << NumPHIdValues << "\n"); 1830 unsigned NumPHIInsts = NumPHIdValues / UnconditionalPreds.size(); 1831 if ((NumPHIdValues % UnconditionalPreds.size()) != 0) 1832 NumPHIInsts++; 1833 1834 return NumPHIInsts <= 1; 1835 }; 1836 1837 if (ScanIdx > 0 && Cond) { 1838 // Check if we would actually sink anything first! This mutates the CFG and 1839 // adds an extra block. The goal in doing this is to allow instructions that 1840 // couldn't be sunk before to be sunk - obviously, speculatable instructions 1841 // (such as trunc, add) can be sunk and predicated already. So we check that 1842 // we're going to sink at least one non-speculatable instruction. 1843 LRI.reset(); 1844 unsigned Idx = 0; 1845 bool Profitable = false; 1846 while (ProfitableToSinkInstruction(LRI) && Idx < ScanIdx) { 1847 if (!isSafeToSpeculativelyExecute((*LRI)[0])) { 1848 Profitable = true; 1849 break; 1850 } 1851 --LRI; 1852 ++Idx; 1853 } 1854 if (!Profitable) 1855 return false; 1856 1857 LLVM_DEBUG(dbgs() << "SINK: Splitting edge\n"); 1858 // We have a conditional edge and we're going to sink some instructions. 1859 // Insert a new block postdominating all blocks we're going to sink from. 1860 if (!SplitBlockPredecessors(BB, UnconditionalPreds, ".sink.split")) 1861 // Edges couldn't be split. 1862 return false; 1863 Changed = true; 1864 } 1865 1866 // Now that we've analyzed all potential sinking candidates, perform the 1867 // actual sink. We iteratively sink the last non-terminator of the source 1868 // blocks into their common successor unless doing so would require too 1869 // many PHI instructions to be generated (currently only one PHI is allowed 1870 // per sunk instruction). 1871 // 1872 // We can use InstructionsToSink to discount values needing PHI-merging that will 1873 // actually be sunk in a later iteration. This allows us to be more 1874 // aggressive in what we sink. This does allow a false positive where we 1875 // sink presuming a later value will also be sunk, but stop half way through 1876 // and never actually sink it which means we produce more PHIs than intended. 1877 // This is unlikely in practice though. 1878 for (unsigned SinkIdx = 0; SinkIdx != ScanIdx; ++SinkIdx) { 1879 LLVM_DEBUG(dbgs() << "SINK: Sink: " 1880 << *UnconditionalPreds[0]->getTerminator()->getPrevNode() 1881 << "\n"); 1882 1883 // Because we've sunk every instruction in turn, the current instruction to 1884 // sink is always at index 0. 1885 LRI.reset(); 1886 if (!ProfitableToSinkInstruction(LRI)) { 1887 // Too many PHIs would be created. 1888 LLVM_DEBUG( 1889 dbgs() << "SINK: stopping here, too many PHIs would be created!\n"); 1890 break; 1891 } 1892 1893 if (!sinkLastInstruction(UnconditionalPreds)) 1894 return Changed; 1895 NumSinkCommons++; 1896 Changed = true; 1897 } 1898 return Changed; 1899 } 1900 1901 /// Determine if we can hoist sink a sole store instruction out of a 1902 /// conditional block. 1903 /// 1904 /// We are looking for code like the following: 1905 /// BrBB: 1906 /// store i32 %add, i32* %arrayidx2 1907 /// ... // No other stores or function calls (we could be calling a memory 1908 /// ... // function). 1909 /// %cmp = icmp ult %x, %y 1910 /// br i1 %cmp, label %EndBB, label %ThenBB 1911 /// ThenBB: 1912 /// store i32 %add5, i32* %arrayidx2 1913 /// br label EndBB 1914 /// EndBB: 1915 /// ... 1916 /// We are going to transform this into: 1917 /// BrBB: 1918 /// store i32 %add, i32* %arrayidx2 1919 /// ... // 1920 /// %cmp = icmp ult %x, %y 1921 /// %add.add5 = select i1 %cmp, i32 %add, %add5 1922 /// store i32 %add.add5, i32* %arrayidx2 1923 /// ... 1924 /// 1925 /// \return The pointer to the value of the previous store if the store can be 1926 /// hoisted into the predecessor block. 0 otherwise. 1927 static Value *isSafeToSpeculateStore(Instruction *I, BasicBlock *BrBB, 1928 BasicBlock *StoreBB, BasicBlock *EndBB) { 1929 StoreInst *StoreToHoist = dyn_cast<StoreInst>(I); 1930 if (!StoreToHoist) 1931 return nullptr; 1932 1933 // Volatile or atomic. 1934 if (!StoreToHoist->isSimple()) 1935 return nullptr; 1936 1937 Value *StorePtr = StoreToHoist->getPointerOperand(); 1938 1939 // Look for a store to the same pointer in BrBB. 1940 unsigned MaxNumInstToLookAt = 9; 1941 for (Instruction &CurI : reverse(BrBB->instructionsWithoutDebug())) { 1942 if (!MaxNumInstToLookAt) 1943 break; 1944 --MaxNumInstToLookAt; 1945 1946 // Could be calling an instruction that affects memory like free(). 1947 if (CurI.mayHaveSideEffects() && !isa<StoreInst>(CurI)) 1948 return nullptr; 1949 1950 if (auto *SI = dyn_cast<StoreInst>(&CurI)) { 1951 // Found the previous store make sure it stores to the same location. 1952 if (SI->getPointerOperand() == StorePtr) 1953 // Found the previous store, return its value operand. 1954 return SI->getValueOperand(); 1955 return nullptr; // Unknown store. 1956 } 1957 } 1958 1959 return nullptr; 1960 } 1961 1962 /// Speculate a conditional basic block flattening the CFG. 1963 /// 1964 /// Note that this is a very risky transform currently. Speculating 1965 /// instructions like this is most often not desirable. Instead, there is an MI 1966 /// pass which can do it with full awareness of the resource constraints. 1967 /// However, some cases are "obvious" and we should do directly. An example of 1968 /// this is speculating a single, reasonably cheap instruction. 1969 /// 1970 /// There is only one distinct advantage to flattening the CFG at the IR level: 1971 /// it makes very common but simplistic optimizations such as are common in 1972 /// instcombine and the DAG combiner more powerful by removing CFG edges and 1973 /// modeling their effects with easier to reason about SSA value graphs. 1974 /// 1975 /// 1976 /// An illustration of this transform is turning this IR: 1977 /// \code 1978 /// BB: 1979 /// %cmp = icmp ult %x, %y 1980 /// br i1 %cmp, label %EndBB, label %ThenBB 1981 /// ThenBB: 1982 /// %sub = sub %x, %y 1983 /// br label BB2 1984 /// EndBB: 1985 /// %phi = phi [ %sub, %ThenBB ], [ 0, %EndBB ] 1986 /// ... 1987 /// \endcode 1988 /// 1989 /// Into this IR: 1990 /// \code 1991 /// BB: 1992 /// %cmp = icmp ult %x, %y 1993 /// %sub = sub %x, %y 1994 /// %cond = select i1 %cmp, 0, %sub 1995 /// ... 1996 /// \endcode 1997 /// 1998 /// \returns true if the conditional block is removed. 1999 bool SimplifyCFGOpt::SpeculativelyExecuteBB(BranchInst *BI, BasicBlock *ThenBB, 2000 const TargetTransformInfo &TTI) { 2001 // Be conservative for now. FP select instruction can often be expensive. 2002 Value *BrCond = BI->getCondition(); 2003 if (isa<FCmpInst>(BrCond)) 2004 return false; 2005 2006 BasicBlock *BB = BI->getParent(); 2007 BasicBlock *EndBB = ThenBB->getTerminator()->getSuccessor(0); 2008 2009 // If ThenBB is actually on the false edge of the conditional branch, remember 2010 // to swap the select operands later. 2011 bool Invert = false; 2012 if (ThenBB != BI->getSuccessor(0)) { 2013 assert(ThenBB == BI->getSuccessor(1) && "No edge from 'if' block?"); 2014 Invert = true; 2015 } 2016 assert(EndBB == BI->getSuccessor(!Invert) && "No edge from to end block"); 2017 2018 // Keep a count of how many times instructions are used within ThenBB when 2019 // they are candidates for sinking into ThenBB. Specifically: 2020 // - They are defined in BB, and 2021 // - They have no side effects, and 2022 // - All of their uses are in ThenBB. 2023 SmallDenseMap<Instruction *, unsigned, 4> SinkCandidateUseCounts; 2024 2025 SmallVector<Instruction *, 4> SpeculatedDbgIntrinsics; 2026 2027 unsigned SpeculatedInstructions = 0; 2028 Value *SpeculatedStoreValue = nullptr; 2029 StoreInst *SpeculatedStore = nullptr; 2030 for (BasicBlock::iterator BBI = ThenBB->begin(), 2031 BBE = std::prev(ThenBB->end()); 2032 BBI != BBE; ++BBI) { 2033 Instruction *I = &*BBI; 2034 // Skip debug info. 2035 if (isa<DbgInfoIntrinsic>(I)) { 2036 SpeculatedDbgIntrinsics.push_back(I); 2037 continue; 2038 } 2039 2040 // Only speculatively execute a single instruction (not counting the 2041 // terminator) for now. 2042 ++SpeculatedInstructions; 2043 if (SpeculatedInstructions > 1) 2044 return false; 2045 2046 // Don't hoist the instruction if it's unsafe or expensive. 2047 if (!isSafeToSpeculativelyExecute(I) && 2048 !(HoistCondStores && (SpeculatedStoreValue = isSafeToSpeculateStore( 2049 I, BB, ThenBB, EndBB)))) 2050 return false; 2051 if (!SpeculatedStoreValue && 2052 ComputeSpeculationCost(I, TTI) > 2053 PHINodeFoldingThreshold * TargetTransformInfo::TCC_Basic) 2054 return false; 2055 2056 // Store the store speculation candidate. 2057 if (SpeculatedStoreValue) 2058 SpeculatedStore = cast<StoreInst>(I); 2059 2060 // Do not hoist the instruction if any of its operands are defined but not 2061 // used in BB. The transformation will prevent the operand from 2062 // being sunk into the use block. 2063 for (User::op_iterator i = I->op_begin(), e = I->op_end(); i != e; ++i) { 2064 Instruction *OpI = dyn_cast<Instruction>(*i); 2065 if (!OpI || OpI->getParent() != BB || OpI->mayHaveSideEffects()) 2066 continue; // Not a candidate for sinking. 2067 2068 ++SinkCandidateUseCounts[OpI]; 2069 } 2070 } 2071 2072 // Consider any sink candidates which are only used in ThenBB as costs for 2073 // speculation. Note, while we iterate over a DenseMap here, we are summing 2074 // and so iteration order isn't significant. 2075 for (SmallDenseMap<Instruction *, unsigned, 4>::iterator 2076 I = SinkCandidateUseCounts.begin(), 2077 E = SinkCandidateUseCounts.end(); 2078 I != E; ++I) 2079 if (I->first->hasNUses(I->second)) { 2080 ++SpeculatedInstructions; 2081 if (SpeculatedInstructions > 1) 2082 return false; 2083 } 2084 2085 // Check that the PHI nodes can be converted to selects. 2086 bool HaveRewritablePHIs = false; 2087 for (PHINode &PN : EndBB->phis()) { 2088 Value *OrigV = PN.getIncomingValueForBlock(BB); 2089 Value *ThenV = PN.getIncomingValueForBlock(ThenBB); 2090 2091 // FIXME: Try to remove some of the duplication with HoistThenElseCodeToIf. 2092 // Skip PHIs which are trivial. 2093 if (ThenV == OrigV) 2094 continue; 2095 2096 // Don't convert to selects if we could remove undefined behavior instead. 2097 if (passingValueIsAlwaysUndefined(OrigV, &PN) || 2098 passingValueIsAlwaysUndefined(ThenV, &PN)) 2099 return false; 2100 2101 HaveRewritablePHIs = true; 2102 ConstantExpr *OrigCE = dyn_cast<ConstantExpr>(OrigV); 2103 ConstantExpr *ThenCE = dyn_cast<ConstantExpr>(ThenV); 2104 if (!OrigCE && !ThenCE) 2105 continue; // Known safe and cheap. 2106 2107 if ((ThenCE && !isSafeToSpeculativelyExecute(ThenCE)) || 2108 (OrigCE && !isSafeToSpeculativelyExecute(OrigCE))) 2109 return false; 2110 unsigned OrigCost = OrigCE ? ComputeSpeculationCost(OrigCE, TTI) : 0; 2111 unsigned ThenCost = ThenCE ? ComputeSpeculationCost(ThenCE, TTI) : 0; 2112 unsigned MaxCost = 2113 2 * PHINodeFoldingThreshold * TargetTransformInfo::TCC_Basic; 2114 if (OrigCost + ThenCost > MaxCost) 2115 return false; 2116 2117 // Account for the cost of an unfolded ConstantExpr which could end up 2118 // getting expanded into Instructions. 2119 // FIXME: This doesn't account for how many operations are combined in the 2120 // constant expression. 2121 ++SpeculatedInstructions; 2122 if (SpeculatedInstructions > 1) 2123 return false; 2124 } 2125 2126 // If there are no PHIs to process, bail early. This helps ensure idempotence 2127 // as well. 2128 if (!HaveRewritablePHIs && !(HoistCondStores && SpeculatedStoreValue)) 2129 return false; 2130 2131 // If we get here, we can hoist the instruction and if-convert. 2132 LLVM_DEBUG(dbgs() << "SPECULATIVELY EXECUTING BB" << *ThenBB << "\n";); 2133 2134 // Insert a select of the value of the speculated store. 2135 if (SpeculatedStoreValue) { 2136 IRBuilder<NoFolder> Builder(BI); 2137 Value *TrueV = SpeculatedStore->getValueOperand(); 2138 Value *FalseV = SpeculatedStoreValue; 2139 if (Invert) 2140 std::swap(TrueV, FalseV); 2141 Value *S = Builder.CreateSelect( 2142 BrCond, TrueV, FalseV, "spec.store.select", BI); 2143 SpeculatedStore->setOperand(0, S); 2144 SpeculatedStore->applyMergedLocation(BI->getDebugLoc(), 2145 SpeculatedStore->getDebugLoc()); 2146 } 2147 2148 // Metadata can be dependent on the condition we are hoisting above. 2149 // Conservatively strip all metadata on the instruction. Drop the debug loc 2150 // to avoid making it appear as if the condition is a constant, which would 2151 // be misleading while debugging. 2152 for (auto &I : *ThenBB) { 2153 if (!SpeculatedStoreValue || &I != SpeculatedStore) 2154 I.setDebugLoc(DebugLoc()); 2155 I.dropUnknownNonDebugMetadata(); 2156 } 2157 2158 // Hoist the instructions. 2159 BB->getInstList().splice(BI->getIterator(), ThenBB->getInstList(), 2160 ThenBB->begin(), std::prev(ThenBB->end())); 2161 2162 // Insert selects and rewrite the PHI operands. 2163 IRBuilder<NoFolder> Builder(BI); 2164 for (PHINode &PN : EndBB->phis()) { 2165 unsigned OrigI = PN.getBasicBlockIndex(BB); 2166 unsigned ThenI = PN.getBasicBlockIndex(ThenBB); 2167 Value *OrigV = PN.getIncomingValue(OrigI); 2168 Value *ThenV = PN.getIncomingValue(ThenI); 2169 2170 // Skip PHIs which are trivial. 2171 if (OrigV == ThenV) 2172 continue; 2173 2174 // Create a select whose true value is the speculatively executed value and 2175 // false value is the pre-existing value. Swap them if the branch 2176 // destinations were inverted. 2177 Value *TrueV = ThenV, *FalseV = OrigV; 2178 if (Invert) 2179 std::swap(TrueV, FalseV); 2180 Value *V = Builder.CreateSelect(BrCond, TrueV, FalseV, "spec.select", BI); 2181 PN.setIncomingValue(OrigI, V); 2182 PN.setIncomingValue(ThenI, V); 2183 } 2184 2185 // Remove speculated dbg intrinsics. 2186 // FIXME: Is it possible to do this in a more elegant way? Moving/merging the 2187 // dbg value for the different flows and inserting it after the select. 2188 for (Instruction *I : SpeculatedDbgIntrinsics) 2189 I->eraseFromParent(); 2190 2191 ++NumSpeculations; 2192 return true; 2193 } 2194 2195 /// Return true if we can thread a branch across this block. 2196 static bool BlockIsSimpleEnoughToThreadThrough(BasicBlock *BB) { 2197 int Size = 0; 2198 2199 for (Instruction &I : BB->instructionsWithoutDebug()) { 2200 if (Size > MaxSmallBlockSize) 2201 return false; // Don't clone large BB's. 2202 // We will delete Phis while threading, so Phis should not be accounted in 2203 // block's size 2204 if (!isa<PHINode>(I)) 2205 ++Size; 2206 2207 // We can only support instructions that do not define values that are 2208 // live outside of the current basic block. 2209 for (User *U : I.users()) { 2210 Instruction *UI = cast<Instruction>(U); 2211 if (UI->getParent() != BB || isa<PHINode>(UI)) 2212 return false; 2213 } 2214 2215 // Looks ok, continue checking. 2216 } 2217 2218 return true; 2219 } 2220 2221 /// If we have a conditional branch on a PHI node value that is defined in the 2222 /// same block as the branch and if any PHI entries are constants, thread edges 2223 /// corresponding to that entry to be branches to their ultimate destination. 2224 static bool FoldCondBranchOnPHI(BranchInst *BI, const DataLayout &DL, 2225 AssumptionCache *AC) { 2226 BasicBlock *BB = BI->getParent(); 2227 PHINode *PN = dyn_cast<PHINode>(BI->getCondition()); 2228 // NOTE: we currently cannot transform this case if the PHI node is used 2229 // outside of the block. 2230 if (!PN || PN->getParent() != BB || !PN->hasOneUse()) 2231 return false; 2232 2233 // Degenerate case of a single entry PHI. 2234 if (PN->getNumIncomingValues() == 1) { 2235 FoldSingleEntryPHINodes(PN->getParent()); 2236 return true; 2237 } 2238 2239 // Now we know that this block has multiple preds and two succs. 2240 if (!BlockIsSimpleEnoughToThreadThrough(BB)) 2241 return false; 2242 2243 // Can't fold blocks that contain noduplicate or convergent calls. 2244 if (any_of(*BB, [](const Instruction &I) { 2245 const CallInst *CI = dyn_cast<CallInst>(&I); 2246 return CI && (CI->cannotDuplicate() || CI->isConvergent()); 2247 })) 2248 return false; 2249 2250 // Okay, this is a simple enough basic block. See if any phi values are 2251 // constants. 2252 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) { 2253 ConstantInt *CB = dyn_cast<ConstantInt>(PN->getIncomingValue(i)); 2254 if (!CB || !CB->getType()->isIntegerTy(1)) 2255 continue; 2256 2257 // Okay, we now know that all edges from PredBB should be revectored to 2258 // branch to RealDest. 2259 BasicBlock *PredBB = PN->getIncomingBlock(i); 2260 BasicBlock *RealDest = BI->getSuccessor(!CB->getZExtValue()); 2261 2262 if (RealDest == BB) 2263 continue; // Skip self loops. 2264 // Skip if the predecessor's terminator is an indirect branch. 2265 if (isa<IndirectBrInst>(PredBB->getTerminator())) 2266 continue; 2267 2268 // The dest block might have PHI nodes, other predecessors and other 2269 // difficult cases. Instead of being smart about this, just insert a new 2270 // block that jumps to the destination block, effectively splitting 2271 // the edge we are about to create. 2272 BasicBlock *EdgeBB = 2273 BasicBlock::Create(BB->getContext(), RealDest->getName() + ".critedge", 2274 RealDest->getParent(), RealDest); 2275 BranchInst *CritEdgeBranch = BranchInst::Create(RealDest, EdgeBB); 2276 CritEdgeBranch->setDebugLoc(BI->getDebugLoc()); 2277 2278 // Update PHI nodes. 2279 AddPredecessorToBlock(RealDest, EdgeBB, BB); 2280 2281 // BB may have instructions that are being threaded over. Clone these 2282 // instructions into EdgeBB. We know that there will be no uses of the 2283 // cloned instructions outside of EdgeBB. 2284 BasicBlock::iterator InsertPt = EdgeBB->begin(); 2285 DenseMap<Value *, Value *> TranslateMap; // Track translated values. 2286 for (BasicBlock::iterator BBI = BB->begin(); &*BBI != BI; ++BBI) { 2287 if (PHINode *PN = dyn_cast<PHINode>(BBI)) { 2288 TranslateMap[PN] = PN->getIncomingValueForBlock(PredBB); 2289 continue; 2290 } 2291 // Clone the instruction. 2292 Instruction *N = BBI->clone(); 2293 if (BBI->hasName()) 2294 N->setName(BBI->getName() + ".c"); 2295 2296 // Update operands due to translation. 2297 for (User::op_iterator i = N->op_begin(), e = N->op_end(); i != e; ++i) { 2298 DenseMap<Value *, Value *>::iterator PI = TranslateMap.find(*i); 2299 if (PI != TranslateMap.end()) 2300 *i = PI->second; 2301 } 2302 2303 // Check for trivial simplification. 2304 if (Value *V = SimplifyInstruction(N, {DL, nullptr, nullptr, AC})) { 2305 if (!BBI->use_empty()) 2306 TranslateMap[&*BBI] = V; 2307 if (!N->mayHaveSideEffects()) { 2308 N->deleteValue(); // Instruction folded away, don't need actual inst 2309 N = nullptr; 2310 } 2311 } else { 2312 if (!BBI->use_empty()) 2313 TranslateMap[&*BBI] = N; 2314 } 2315 if (N) { 2316 // Insert the new instruction into its new home. 2317 EdgeBB->getInstList().insert(InsertPt, N); 2318 2319 // Register the new instruction with the assumption cache if necessary. 2320 if (AC && match(N, m_Intrinsic<Intrinsic::assume>())) 2321 AC->registerAssumption(cast<IntrinsicInst>(N)); 2322 } 2323 } 2324 2325 // Loop over all of the edges from PredBB to BB, changing them to branch 2326 // to EdgeBB instead. 2327 Instruction *PredBBTI = PredBB->getTerminator(); 2328 for (unsigned i = 0, e = PredBBTI->getNumSuccessors(); i != e; ++i) 2329 if (PredBBTI->getSuccessor(i) == BB) { 2330 BB->removePredecessor(PredBB); 2331 PredBBTI->setSuccessor(i, EdgeBB); 2332 } 2333 2334 // Recurse, simplifying any other constants. 2335 return FoldCondBranchOnPHI(BI, DL, AC) || true; 2336 } 2337 2338 return false; 2339 } 2340 2341 /// Given a BB that starts with the specified two-entry PHI node, 2342 /// see if we can eliminate it. 2343 static bool FoldTwoEntryPHINode(PHINode *PN, const TargetTransformInfo &TTI, 2344 const DataLayout &DL) { 2345 // Ok, this is a two entry PHI node. Check to see if this is a simple "if 2346 // statement", which has a very simple dominance structure. Basically, we 2347 // are trying to find the condition that is being branched on, which 2348 // subsequently causes this merge to happen. We really want control 2349 // dependence information for this check, but simplifycfg can't keep it up 2350 // to date, and this catches most of the cases we care about anyway. 2351 BasicBlock *BB = PN->getParent(); 2352 2353 BasicBlock *IfTrue, *IfFalse; 2354 Value *IfCond = GetIfCondition(BB, IfTrue, IfFalse); 2355 if (!IfCond || 2356 // Don't bother if the branch will be constant folded trivially. 2357 isa<ConstantInt>(IfCond)) 2358 return false; 2359 2360 // Okay, we found that we can merge this two-entry phi node into a select. 2361 // Doing so would require us to fold *all* two entry phi nodes in this block. 2362 // At some point this becomes non-profitable (particularly if the target 2363 // doesn't support cmov's). Only do this transformation if there are two or 2364 // fewer PHI nodes in this block. 2365 unsigned NumPhis = 0; 2366 for (BasicBlock::iterator I = BB->begin(); isa<PHINode>(I); ++NumPhis, ++I) 2367 if (NumPhis > 2) 2368 return false; 2369 2370 // Loop over the PHI's seeing if we can promote them all to select 2371 // instructions. While we are at it, keep track of the instructions 2372 // that need to be moved to the dominating block. 2373 SmallPtrSet<Instruction *, 4> AggressiveInsts; 2374 int BudgetRemaining = 2375 TwoEntryPHINodeFoldingThreshold * TargetTransformInfo::TCC_Basic; 2376 2377 for (BasicBlock::iterator II = BB->begin(); isa<PHINode>(II);) { 2378 PHINode *PN = cast<PHINode>(II++); 2379 if (Value *V = SimplifyInstruction(PN, {DL, PN})) { 2380 PN->replaceAllUsesWith(V); 2381 PN->eraseFromParent(); 2382 continue; 2383 } 2384 2385 if (!DominatesMergePoint(PN->getIncomingValue(0), BB, AggressiveInsts, 2386 BudgetRemaining, TTI) || 2387 !DominatesMergePoint(PN->getIncomingValue(1), BB, AggressiveInsts, 2388 BudgetRemaining, TTI)) 2389 return false; 2390 } 2391 2392 // If we folded the first phi, PN dangles at this point. Refresh it. If 2393 // we ran out of PHIs then we simplified them all. 2394 PN = dyn_cast<PHINode>(BB->begin()); 2395 if (!PN) 2396 return true; 2397 2398 // Return true if at least one of these is a 'not', and another is either 2399 // a 'not' too, or a constant. 2400 auto CanHoistNotFromBothValues = [](Value *V0, Value *V1) { 2401 if (!match(V0, m_Not(m_Value()))) 2402 std::swap(V0, V1); 2403 auto Invertible = m_CombineOr(m_Not(m_Value()), m_AnyIntegralConstant()); 2404 return match(V0, m_Not(m_Value())) && match(V1, Invertible); 2405 }; 2406 2407 // Don't fold i1 branches on PHIs which contain binary operators, unless one 2408 // of the incoming values is an 'not' and another one is freely invertible. 2409 // These can often be turned into switches and other things. 2410 if (PN->getType()->isIntegerTy(1) && 2411 (isa<BinaryOperator>(PN->getIncomingValue(0)) || 2412 isa<BinaryOperator>(PN->getIncomingValue(1)) || 2413 isa<BinaryOperator>(IfCond)) && 2414 !CanHoistNotFromBothValues(PN->getIncomingValue(0), 2415 PN->getIncomingValue(1))) 2416 return false; 2417 2418 // If all PHI nodes are promotable, check to make sure that all instructions 2419 // in the predecessor blocks can be promoted as well. If not, we won't be able 2420 // to get rid of the control flow, so it's not worth promoting to select 2421 // instructions. 2422 BasicBlock *DomBlock = nullptr; 2423 BasicBlock *IfBlock1 = PN->getIncomingBlock(0); 2424 BasicBlock *IfBlock2 = PN->getIncomingBlock(1); 2425 if (cast<BranchInst>(IfBlock1->getTerminator())->isConditional()) { 2426 IfBlock1 = nullptr; 2427 } else { 2428 DomBlock = *pred_begin(IfBlock1); 2429 for (BasicBlock::iterator I = IfBlock1->begin(); !I->isTerminator(); ++I) 2430 if (!AggressiveInsts.count(&*I) && !isa<DbgInfoIntrinsic>(I)) { 2431 // This is not an aggressive instruction that we can promote. 2432 // Because of this, we won't be able to get rid of the control flow, so 2433 // the xform is not worth it. 2434 return false; 2435 } 2436 } 2437 2438 if (cast<BranchInst>(IfBlock2->getTerminator())->isConditional()) { 2439 IfBlock2 = nullptr; 2440 } else { 2441 DomBlock = *pred_begin(IfBlock2); 2442 for (BasicBlock::iterator I = IfBlock2->begin(); !I->isTerminator(); ++I) 2443 if (!AggressiveInsts.count(&*I) && !isa<DbgInfoIntrinsic>(I)) { 2444 // This is not an aggressive instruction that we can promote. 2445 // Because of this, we won't be able to get rid of the control flow, so 2446 // the xform is not worth it. 2447 return false; 2448 } 2449 } 2450 assert(DomBlock && "Failed to find root DomBlock"); 2451 2452 LLVM_DEBUG(dbgs() << "FOUND IF CONDITION! " << *IfCond 2453 << " T: " << IfTrue->getName() 2454 << " F: " << IfFalse->getName() << "\n"); 2455 2456 // If we can still promote the PHI nodes after this gauntlet of tests, 2457 // do all of the PHI's now. 2458 Instruction *InsertPt = DomBlock->getTerminator(); 2459 IRBuilder<NoFolder> Builder(InsertPt); 2460 2461 // Move all 'aggressive' instructions, which are defined in the 2462 // conditional parts of the if's up to the dominating block. 2463 if (IfBlock1) 2464 hoistAllInstructionsInto(DomBlock, InsertPt, IfBlock1); 2465 if (IfBlock2) 2466 hoistAllInstructionsInto(DomBlock, InsertPt, IfBlock2); 2467 2468 // Propagate fast-math-flags from phi nodes to replacement selects. 2469 IRBuilder<>::FastMathFlagGuard FMFGuard(Builder); 2470 while (PHINode *PN = dyn_cast<PHINode>(BB->begin())) { 2471 if (isa<FPMathOperator>(PN)) 2472 Builder.setFastMathFlags(PN->getFastMathFlags()); 2473 2474 // Change the PHI node into a select instruction. 2475 Value *TrueVal = PN->getIncomingValue(PN->getIncomingBlock(0) == IfFalse); 2476 Value *FalseVal = PN->getIncomingValue(PN->getIncomingBlock(0) == IfTrue); 2477 2478 Value *Sel = Builder.CreateSelect(IfCond, TrueVal, FalseVal, "", InsertPt); 2479 PN->replaceAllUsesWith(Sel); 2480 Sel->takeName(PN); 2481 PN->eraseFromParent(); 2482 } 2483 2484 // At this point, IfBlock1 and IfBlock2 are both empty, so our if statement 2485 // has been flattened. Change DomBlock to jump directly to our new block to 2486 // avoid other simplifycfg's kicking in on the diamond. 2487 Instruction *OldTI = DomBlock->getTerminator(); 2488 Builder.SetInsertPoint(OldTI); 2489 Builder.CreateBr(BB); 2490 OldTI->eraseFromParent(); 2491 return true; 2492 } 2493 2494 /// If we found a conditional branch that goes to two returning blocks, 2495 /// try to merge them together into one return, 2496 /// introducing a select if the return values disagree. 2497 bool SimplifyCFGOpt::SimplifyCondBranchToTwoReturns(BranchInst *BI, 2498 IRBuilder<> &Builder) { 2499 assert(BI->isConditional() && "Must be a conditional branch"); 2500 BasicBlock *TrueSucc = BI->getSuccessor(0); 2501 BasicBlock *FalseSucc = BI->getSuccessor(1); 2502 ReturnInst *TrueRet = cast<ReturnInst>(TrueSucc->getTerminator()); 2503 ReturnInst *FalseRet = cast<ReturnInst>(FalseSucc->getTerminator()); 2504 2505 // Check to ensure both blocks are empty (just a return) or optionally empty 2506 // with PHI nodes. If there are other instructions, merging would cause extra 2507 // computation on one path or the other. 2508 if (!TrueSucc->getFirstNonPHIOrDbg()->isTerminator()) 2509 return false; 2510 if (!FalseSucc->getFirstNonPHIOrDbg()->isTerminator()) 2511 return false; 2512 2513 Builder.SetInsertPoint(BI); 2514 // Okay, we found a branch that is going to two return nodes. If 2515 // there is no return value for this function, just change the 2516 // branch into a return. 2517 if (FalseRet->getNumOperands() == 0) { 2518 TrueSucc->removePredecessor(BI->getParent()); 2519 FalseSucc->removePredecessor(BI->getParent()); 2520 Builder.CreateRetVoid(); 2521 EraseTerminatorAndDCECond(BI); 2522 return true; 2523 } 2524 2525 // Otherwise, figure out what the true and false return values are 2526 // so we can insert a new select instruction. 2527 Value *TrueValue = TrueRet->getReturnValue(); 2528 Value *FalseValue = FalseRet->getReturnValue(); 2529 2530 // Unwrap any PHI nodes in the return blocks. 2531 if (PHINode *TVPN = dyn_cast_or_null<PHINode>(TrueValue)) 2532 if (TVPN->getParent() == TrueSucc) 2533 TrueValue = TVPN->getIncomingValueForBlock(BI->getParent()); 2534 if (PHINode *FVPN = dyn_cast_or_null<PHINode>(FalseValue)) 2535 if (FVPN->getParent() == FalseSucc) 2536 FalseValue = FVPN->getIncomingValueForBlock(BI->getParent()); 2537 2538 // In order for this transformation to be safe, we must be able to 2539 // unconditionally execute both operands to the return. This is 2540 // normally the case, but we could have a potentially-trapping 2541 // constant expression that prevents this transformation from being 2542 // safe. 2543 if (ConstantExpr *TCV = dyn_cast_or_null<ConstantExpr>(TrueValue)) 2544 if (TCV->canTrap()) 2545 return false; 2546 if (ConstantExpr *FCV = dyn_cast_or_null<ConstantExpr>(FalseValue)) 2547 if (FCV->canTrap()) 2548 return false; 2549 2550 // Okay, we collected all the mapped values and checked them for sanity, and 2551 // defined to really do this transformation. First, update the CFG. 2552 TrueSucc->removePredecessor(BI->getParent()); 2553 FalseSucc->removePredecessor(BI->getParent()); 2554 2555 // Insert select instructions where needed. 2556 Value *BrCond = BI->getCondition(); 2557 if (TrueValue) { 2558 // Insert a select if the results differ. 2559 if (TrueValue == FalseValue || isa<UndefValue>(FalseValue)) { 2560 } else if (isa<UndefValue>(TrueValue)) { 2561 TrueValue = FalseValue; 2562 } else { 2563 TrueValue = 2564 Builder.CreateSelect(BrCond, TrueValue, FalseValue, "retval", BI); 2565 } 2566 } 2567 2568 Value *RI = 2569 !TrueValue ? Builder.CreateRetVoid() : Builder.CreateRet(TrueValue); 2570 2571 (void)RI; 2572 2573 LLVM_DEBUG(dbgs() << "\nCHANGING BRANCH TO TWO RETURNS INTO SELECT:" 2574 << "\n " << *BI << "\nNewRet = " << *RI << "\nTRUEBLOCK: " 2575 << *TrueSucc << "\nFALSEBLOCK: " << *FalseSucc); 2576 2577 EraseTerminatorAndDCECond(BI); 2578 2579 return true; 2580 } 2581 2582 /// Return true if the given instruction is available 2583 /// in its predecessor block. If yes, the instruction will be removed. 2584 static bool tryCSEWithPredecessor(Instruction *Inst, BasicBlock *PB) { 2585 if (!isa<BinaryOperator>(Inst) && !isa<CmpInst>(Inst)) 2586 return false; 2587 for (Instruction &I : *PB) { 2588 Instruction *PBI = &I; 2589 // Check whether Inst and PBI generate the same value. 2590 if (Inst->isIdenticalTo(PBI)) { 2591 Inst->replaceAllUsesWith(PBI); 2592 Inst->eraseFromParent(); 2593 return true; 2594 } 2595 } 2596 return false; 2597 } 2598 2599 /// Return true if either PBI or BI has branch weight available, and store 2600 /// the weights in {Pred|Succ}{True|False}Weight. If one of PBI and BI does 2601 /// not have branch weight, use 1:1 as its weight. 2602 static bool extractPredSuccWeights(BranchInst *PBI, BranchInst *BI, 2603 uint64_t &PredTrueWeight, 2604 uint64_t &PredFalseWeight, 2605 uint64_t &SuccTrueWeight, 2606 uint64_t &SuccFalseWeight) { 2607 bool PredHasWeights = 2608 PBI->extractProfMetadata(PredTrueWeight, PredFalseWeight); 2609 bool SuccHasWeights = 2610 BI->extractProfMetadata(SuccTrueWeight, SuccFalseWeight); 2611 if (PredHasWeights || SuccHasWeights) { 2612 if (!PredHasWeights) 2613 PredTrueWeight = PredFalseWeight = 1; 2614 if (!SuccHasWeights) 2615 SuccTrueWeight = SuccFalseWeight = 1; 2616 return true; 2617 } else { 2618 return false; 2619 } 2620 } 2621 2622 /// If this basic block is simple enough, and if a predecessor branches to us 2623 /// and one of our successors, fold the block into the predecessor and use 2624 /// logical operations to pick the right destination. 2625 bool llvm::FoldBranchToCommonDest(BranchInst *BI, MemorySSAUpdater *MSSAU, 2626 unsigned BonusInstThreshold) { 2627 BasicBlock *BB = BI->getParent(); 2628 2629 const unsigned PredCount = pred_size(BB); 2630 2631 bool Changed = false; 2632 2633 Instruction *Cond = nullptr; 2634 if (BI->isConditional()) 2635 Cond = dyn_cast<Instruction>(BI->getCondition()); 2636 else { 2637 // For unconditional branch, check for a simple CFG pattern, where 2638 // BB has a single predecessor and BB's successor is also its predecessor's 2639 // successor. If such pattern exists, check for CSE between BB and its 2640 // predecessor. 2641 if (BasicBlock *PB = BB->getSinglePredecessor()) 2642 if (BranchInst *PBI = dyn_cast<BranchInst>(PB->getTerminator())) 2643 if (PBI->isConditional() && 2644 (BI->getSuccessor(0) == PBI->getSuccessor(0) || 2645 BI->getSuccessor(0) == PBI->getSuccessor(1))) { 2646 for (auto I = BB->instructionsWithoutDebug().begin(), 2647 E = BB->instructionsWithoutDebug().end(); 2648 I != E;) { 2649 Instruction *Curr = &*I++; 2650 if (isa<CmpInst>(Curr)) { 2651 Cond = Curr; 2652 break; 2653 } 2654 // Quit if we can't remove this instruction. 2655 if (!tryCSEWithPredecessor(Curr, PB)) 2656 return Changed; 2657 Changed = true; 2658 } 2659 } 2660 2661 if (!Cond) 2662 return Changed; 2663 } 2664 2665 if (!Cond || (!isa<CmpInst>(Cond) && !isa<BinaryOperator>(Cond)) || 2666 Cond->getParent() != BB || !Cond->hasOneUse()) 2667 return Changed; 2668 2669 // Make sure the instruction after the condition is the cond branch. 2670 BasicBlock::iterator CondIt = ++Cond->getIterator(); 2671 2672 // Ignore dbg intrinsics. 2673 while (isa<DbgInfoIntrinsic>(CondIt)) 2674 ++CondIt; 2675 2676 if (&*CondIt != BI) 2677 return Changed; 2678 2679 // Only allow this transformation if computing the condition doesn't involve 2680 // too many instructions and these involved instructions can be executed 2681 // unconditionally. We denote all involved instructions except the condition 2682 // as "bonus instructions", and only allow this transformation when the 2683 // number of the bonus instructions we'll need to create when cloning into 2684 // each predecessor does not exceed a certain threshold. 2685 unsigned NumBonusInsts = 0; 2686 for (auto I = BB->begin(); Cond != &*I; ++I) { 2687 // Ignore dbg intrinsics. 2688 if (isa<DbgInfoIntrinsic>(I)) 2689 continue; 2690 if (!I->hasOneUse() || !isSafeToSpeculativelyExecute(&*I)) 2691 return Changed; 2692 // I has only one use and can be executed unconditionally. 2693 Instruction *User = dyn_cast<Instruction>(I->user_back()); 2694 if (User == nullptr || User->getParent() != BB) 2695 return Changed; 2696 // I is used in the same BB. Since BI uses Cond and doesn't have more slots 2697 // to use any other instruction, User must be an instruction between next(I) 2698 // and Cond. 2699 2700 // Account for the cost of duplicating this instruction into each 2701 // predecessor. 2702 NumBonusInsts += PredCount; 2703 // Early exits once we reach the limit. 2704 if (NumBonusInsts > BonusInstThreshold) 2705 return Changed; 2706 } 2707 2708 // Cond is known to be a compare or binary operator. Check to make sure that 2709 // neither operand is a potentially-trapping constant expression. 2710 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Cond->getOperand(0))) 2711 if (CE->canTrap()) 2712 return Changed; 2713 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Cond->getOperand(1))) 2714 if (CE->canTrap()) 2715 return Changed; 2716 2717 // Finally, don't infinitely unroll conditional loops. 2718 BasicBlock *TrueDest = BI->getSuccessor(0); 2719 BasicBlock *FalseDest = (BI->isConditional()) ? BI->getSuccessor(1) : nullptr; 2720 if (TrueDest == BB || FalseDest == BB) 2721 return Changed; 2722 2723 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) { 2724 BasicBlock *PredBlock = *PI; 2725 BranchInst *PBI = dyn_cast<BranchInst>(PredBlock->getTerminator()); 2726 2727 // Check that we have two conditional branches. If there is a PHI node in 2728 // the common successor, verify that the same value flows in from both 2729 // blocks. 2730 SmallVector<PHINode *, 4> PHIs; 2731 if (!PBI || PBI->isUnconditional() || 2732 (BI->isConditional() && !SafeToMergeTerminators(BI, PBI)) || 2733 (!BI->isConditional() && 2734 !isProfitableToFoldUnconditional(BI, PBI, Cond, PHIs))) 2735 continue; 2736 2737 // Determine if the two branches share a common destination. 2738 Instruction::BinaryOps Opc = Instruction::BinaryOpsEnd; 2739 bool InvertPredCond = false; 2740 2741 if (BI->isConditional()) { 2742 if (PBI->getSuccessor(0) == TrueDest) { 2743 Opc = Instruction::Or; 2744 } else if (PBI->getSuccessor(1) == FalseDest) { 2745 Opc = Instruction::And; 2746 } else if (PBI->getSuccessor(0) == FalseDest) { 2747 Opc = Instruction::And; 2748 InvertPredCond = true; 2749 } else if (PBI->getSuccessor(1) == TrueDest) { 2750 Opc = Instruction::Or; 2751 InvertPredCond = true; 2752 } else { 2753 continue; 2754 } 2755 } else { 2756 if (PBI->getSuccessor(0) != TrueDest && PBI->getSuccessor(1) != TrueDest) 2757 continue; 2758 } 2759 2760 LLVM_DEBUG(dbgs() << "FOLDING BRANCH TO COMMON DEST:\n" << *PBI << *BB); 2761 Changed = true; 2762 2763 IRBuilder<> Builder(PBI); 2764 2765 // If we need to invert the condition in the pred block to match, do so now. 2766 if (InvertPredCond) { 2767 Value *NewCond = PBI->getCondition(); 2768 2769 if (NewCond->hasOneUse() && isa<CmpInst>(NewCond)) { 2770 CmpInst *CI = cast<CmpInst>(NewCond); 2771 CI->setPredicate(CI->getInversePredicate()); 2772 } else { 2773 NewCond = 2774 Builder.CreateNot(NewCond, PBI->getCondition()->getName() + ".not"); 2775 } 2776 2777 PBI->setCondition(NewCond); 2778 PBI->swapSuccessors(); 2779 } 2780 2781 // If we have bonus instructions, clone them into the predecessor block. 2782 // Note that there may be multiple predecessor blocks, so we cannot move 2783 // bonus instructions to a predecessor block. 2784 ValueToValueMapTy VMap; // maps original values to cloned values 2785 // We already make sure Cond is the last instruction before BI. Therefore, 2786 // all instructions before Cond other than DbgInfoIntrinsic are bonus 2787 // instructions. 2788 for (auto BonusInst = BB->begin(); Cond != &*BonusInst; ++BonusInst) { 2789 if (isa<DbgInfoIntrinsic>(BonusInst)) 2790 continue; 2791 Instruction *NewBonusInst = BonusInst->clone(); 2792 2793 // When we fold the bonus instructions we want to make sure we 2794 // reset their debug locations in order to avoid stepping on dead 2795 // code caused by folding dead branches. 2796 NewBonusInst->setDebugLoc(DebugLoc()); 2797 2798 RemapInstruction(NewBonusInst, VMap, 2799 RF_NoModuleLevelChanges | RF_IgnoreMissingLocals); 2800 VMap[&*BonusInst] = NewBonusInst; 2801 2802 // If we moved a load, we cannot any longer claim any knowledge about 2803 // its potential value. The previous information might have been valid 2804 // only given the branch precondition. 2805 // For an analogous reason, we must also drop all the metadata whose 2806 // semantics we don't understand. 2807 NewBonusInst->dropUnknownNonDebugMetadata(); 2808 2809 PredBlock->getInstList().insert(PBI->getIterator(), NewBonusInst); 2810 NewBonusInst->takeName(&*BonusInst); 2811 BonusInst->setName(BonusInst->getName() + ".old"); 2812 } 2813 2814 // Clone Cond into the predecessor basic block, and or/and the 2815 // two conditions together. 2816 Instruction *CondInPred = Cond->clone(); 2817 2818 // Reset the condition debug location to avoid jumping on dead code 2819 // as the result of folding dead branches. 2820 CondInPred->setDebugLoc(DebugLoc()); 2821 2822 RemapInstruction(CondInPred, VMap, 2823 RF_NoModuleLevelChanges | RF_IgnoreMissingLocals); 2824 PredBlock->getInstList().insert(PBI->getIterator(), CondInPred); 2825 CondInPred->takeName(Cond); 2826 Cond->setName(CondInPred->getName() + ".old"); 2827 2828 if (BI->isConditional()) { 2829 Instruction *NewCond = cast<Instruction>( 2830 Builder.CreateBinOp(Opc, PBI->getCondition(), CondInPred, "or.cond")); 2831 PBI->setCondition(NewCond); 2832 2833 uint64_t PredTrueWeight, PredFalseWeight, SuccTrueWeight, SuccFalseWeight; 2834 bool HasWeights = 2835 extractPredSuccWeights(PBI, BI, PredTrueWeight, PredFalseWeight, 2836 SuccTrueWeight, SuccFalseWeight); 2837 SmallVector<uint64_t, 8> NewWeights; 2838 2839 if (PBI->getSuccessor(0) == BB) { 2840 if (HasWeights) { 2841 // PBI: br i1 %x, BB, FalseDest 2842 // BI: br i1 %y, TrueDest, FalseDest 2843 // TrueWeight is TrueWeight for PBI * TrueWeight for BI. 2844 NewWeights.push_back(PredTrueWeight * SuccTrueWeight); 2845 // FalseWeight is FalseWeight for PBI * TotalWeight for BI + 2846 // TrueWeight for PBI * FalseWeight for BI. 2847 // We assume that total weights of a BranchInst can fit into 32 bits. 2848 // Therefore, we will not have overflow using 64-bit arithmetic. 2849 NewWeights.push_back(PredFalseWeight * 2850 (SuccFalseWeight + SuccTrueWeight) + 2851 PredTrueWeight * SuccFalseWeight); 2852 } 2853 AddPredecessorToBlock(TrueDest, PredBlock, BB, MSSAU); 2854 PBI->setSuccessor(0, TrueDest); 2855 } 2856 if (PBI->getSuccessor(1) == BB) { 2857 if (HasWeights) { 2858 // PBI: br i1 %x, TrueDest, BB 2859 // BI: br i1 %y, TrueDest, FalseDest 2860 // TrueWeight is TrueWeight for PBI * TotalWeight for BI + 2861 // FalseWeight for PBI * TrueWeight for BI. 2862 NewWeights.push_back(PredTrueWeight * 2863 (SuccFalseWeight + SuccTrueWeight) + 2864 PredFalseWeight * SuccTrueWeight); 2865 // FalseWeight is FalseWeight for PBI * FalseWeight for BI. 2866 NewWeights.push_back(PredFalseWeight * SuccFalseWeight); 2867 } 2868 AddPredecessorToBlock(FalseDest, PredBlock, BB, MSSAU); 2869 PBI->setSuccessor(1, FalseDest); 2870 } 2871 if (NewWeights.size() == 2) { 2872 // Halve the weights if any of them cannot fit in an uint32_t 2873 FitWeights(NewWeights); 2874 2875 SmallVector<uint32_t, 8> MDWeights(NewWeights.begin(), 2876 NewWeights.end()); 2877 setBranchWeights(PBI, MDWeights[0], MDWeights[1]); 2878 } else 2879 PBI->setMetadata(LLVMContext::MD_prof, nullptr); 2880 } else { 2881 // Update PHI nodes in the common successors. 2882 for (unsigned i = 0, e = PHIs.size(); i != e; ++i) { 2883 ConstantInt *PBI_C = cast<ConstantInt>( 2884 PHIs[i]->getIncomingValueForBlock(PBI->getParent())); 2885 assert(PBI_C->getType()->isIntegerTy(1)); 2886 Instruction *MergedCond = nullptr; 2887 if (PBI->getSuccessor(0) == TrueDest) { 2888 // Create (PBI_Cond and PBI_C) or (!PBI_Cond and BI_Value) 2889 // PBI_C is true: PBI_Cond or (!PBI_Cond and BI_Value) 2890 // is false: !PBI_Cond and BI_Value 2891 Instruction *NotCond = cast<Instruction>( 2892 Builder.CreateNot(PBI->getCondition(), "not.cond")); 2893 MergedCond = cast<Instruction>( 2894 Builder.CreateBinOp(Instruction::And, NotCond, CondInPred, 2895 "and.cond")); 2896 if (PBI_C->isOne()) 2897 MergedCond = cast<Instruction>(Builder.CreateBinOp( 2898 Instruction::Or, PBI->getCondition(), MergedCond, "or.cond")); 2899 } else { 2900 // Create (PBI_Cond and BI_Value) or (!PBI_Cond and PBI_C) 2901 // PBI_C is true: (PBI_Cond and BI_Value) or (!PBI_Cond) 2902 // is false: PBI_Cond and BI_Value 2903 MergedCond = cast<Instruction>(Builder.CreateBinOp( 2904 Instruction::And, PBI->getCondition(), CondInPred, "and.cond")); 2905 if (PBI_C->isOne()) { 2906 Instruction *NotCond = cast<Instruction>( 2907 Builder.CreateNot(PBI->getCondition(), "not.cond")); 2908 MergedCond = cast<Instruction>(Builder.CreateBinOp( 2909 Instruction::Or, NotCond, MergedCond, "or.cond")); 2910 } 2911 } 2912 // Update PHI Node. 2913 PHIs[i]->setIncomingValueForBlock(PBI->getParent(), MergedCond); 2914 } 2915 2916 // PBI is changed to branch to TrueDest below. Remove itself from 2917 // potential phis from all other successors. 2918 if (MSSAU) 2919 MSSAU->changeCondBranchToUnconditionalTo(PBI, TrueDest); 2920 2921 // Change PBI from Conditional to Unconditional. 2922 BranchInst *New_PBI = BranchInst::Create(TrueDest, PBI); 2923 EraseTerminatorAndDCECond(PBI, MSSAU); 2924 PBI = New_PBI; 2925 } 2926 2927 // If BI was a loop latch, it may have had associated loop metadata. 2928 // We need to copy it to the new latch, that is, PBI. 2929 if (MDNode *LoopMD = BI->getMetadata(LLVMContext::MD_loop)) 2930 PBI->setMetadata(LLVMContext::MD_loop, LoopMD); 2931 2932 // TODO: If BB is reachable from all paths through PredBlock, then we 2933 // could replace PBI's branch probabilities with BI's. 2934 2935 // Copy any debug value intrinsics into the end of PredBlock. 2936 for (Instruction &I : *BB) { 2937 if (isa<DbgInfoIntrinsic>(I)) { 2938 Instruction *NewI = I.clone(); 2939 RemapInstruction(NewI, VMap, 2940 RF_NoModuleLevelChanges | RF_IgnoreMissingLocals); 2941 NewI->insertBefore(PBI); 2942 } 2943 } 2944 2945 return Changed; 2946 } 2947 return Changed; 2948 } 2949 2950 // If there is only one store in BB1 and BB2, return it, otherwise return 2951 // nullptr. 2952 static StoreInst *findUniqueStoreInBlocks(BasicBlock *BB1, BasicBlock *BB2) { 2953 StoreInst *S = nullptr; 2954 for (auto *BB : {BB1, BB2}) { 2955 if (!BB) 2956 continue; 2957 for (auto &I : *BB) 2958 if (auto *SI = dyn_cast<StoreInst>(&I)) { 2959 if (S) 2960 // Multiple stores seen. 2961 return nullptr; 2962 else 2963 S = SI; 2964 } 2965 } 2966 return S; 2967 } 2968 2969 static Value *ensureValueAvailableInSuccessor(Value *V, BasicBlock *BB, 2970 Value *AlternativeV = nullptr) { 2971 // PHI is going to be a PHI node that allows the value V that is defined in 2972 // BB to be referenced in BB's only successor. 2973 // 2974 // If AlternativeV is nullptr, the only value we care about in PHI is V. It 2975 // doesn't matter to us what the other operand is (it'll never get used). We 2976 // could just create a new PHI with an undef incoming value, but that could 2977 // increase register pressure if EarlyCSE/InstCombine can't fold it with some 2978 // other PHI. So here we directly look for some PHI in BB's successor with V 2979 // as an incoming operand. If we find one, we use it, else we create a new 2980 // one. 2981 // 2982 // If AlternativeV is not nullptr, we care about both incoming values in PHI. 2983 // PHI must be exactly: phi <ty> [ %BB, %V ], [ %OtherBB, %AlternativeV] 2984 // where OtherBB is the single other predecessor of BB's only successor. 2985 PHINode *PHI = nullptr; 2986 BasicBlock *Succ = BB->getSingleSuccessor(); 2987 2988 for (auto I = Succ->begin(); isa<PHINode>(I); ++I) 2989 if (cast<PHINode>(I)->getIncomingValueForBlock(BB) == V) { 2990 PHI = cast<PHINode>(I); 2991 if (!AlternativeV) 2992 break; 2993 2994 assert(Succ->hasNPredecessors(2)); 2995 auto PredI = pred_begin(Succ); 2996 BasicBlock *OtherPredBB = *PredI == BB ? *++PredI : *PredI; 2997 if (PHI->getIncomingValueForBlock(OtherPredBB) == AlternativeV) 2998 break; 2999 PHI = nullptr; 3000 } 3001 if (PHI) 3002 return PHI; 3003 3004 // If V is not an instruction defined in BB, just return it. 3005 if (!AlternativeV && 3006 (!isa<Instruction>(V) || cast<Instruction>(V)->getParent() != BB)) 3007 return V; 3008 3009 PHI = PHINode::Create(V->getType(), 2, "simplifycfg.merge", &Succ->front()); 3010 PHI->addIncoming(V, BB); 3011 for (BasicBlock *PredBB : predecessors(Succ)) 3012 if (PredBB != BB) 3013 PHI->addIncoming( 3014 AlternativeV ? AlternativeV : UndefValue::get(V->getType()), PredBB); 3015 return PHI; 3016 } 3017 3018 static bool mergeConditionalStoreToAddress(BasicBlock *PTB, BasicBlock *PFB, 3019 BasicBlock *QTB, BasicBlock *QFB, 3020 BasicBlock *PostBB, Value *Address, 3021 bool InvertPCond, bool InvertQCond, 3022 const DataLayout &DL, 3023 const TargetTransformInfo &TTI) { 3024 // For every pointer, there must be exactly two stores, one coming from 3025 // PTB or PFB, and the other from QTB or QFB. We don't support more than one 3026 // store (to any address) in PTB,PFB or QTB,QFB. 3027 // FIXME: We could relax this restriction with a bit more work and performance 3028 // testing. 3029 StoreInst *PStore = findUniqueStoreInBlocks(PTB, PFB); 3030 StoreInst *QStore = findUniqueStoreInBlocks(QTB, QFB); 3031 if (!PStore || !QStore) 3032 return false; 3033 3034 // Now check the stores are compatible. 3035 if (!QStore->isUnordered() || !PStore->isUnordered()) 3036 return false; 3037 3038 // Check that sinking the store won't cause program behavior changes. Sinking 3039 // the store out of the Q blocks won't change any behavior as we're sinking 3040 // from a block to its unconditional successor. But we're moving a store from 3041 // the P blocks down through the middle block (QBI) and past both QFB and QTB. 3042 // So we need to check that there are no aliasing loads or stores in 3043 // QBI, QTB and QFB. We also need to check there are no conflicting memory 3044 // operations between PStore and the end of its parent block. 3045 // 3046 // The ideal way to do this is to query AliasAnalysis, but we don't 3047 // preserve AA currently so that is dangerous. Be super safe and just 3048 // check there are no other memory operations at all. 3049 for (auto &I : *QFB->getSinglePredecessor()) 3050 if (I.mayReadOrWriteMemory()) 3051 return false; 3052 for (auto &I : *QFB) 3053 if (&I != QStore && I.mayReadOrWriteMemory()) 3054 return false; 3055 if (QTB) 3056 for (auto &I : *QTB) 3057 if (&I != QStore && I.mayReadOrWriteMemory()) 3058 return false; 3059 for (auto I = BasicBlock::iterator(PStore), E = PStore->getParent()->end(); 3060 I != E; ++I) 3061 if (&*I != PStore && I->mayReadOrWriteMemory()) 3062 return false; 3063 3064 // If we're not in aggressive mode, we only optimize if we have some 3065 // confidence that by optimizing we'll allow P and/or Q to be if-converted. 3066 auto IsWorthwhile = [&](BasicBlock *BB, ArrayRef<StoreInst *> FreeStores) { 3067 if (!BB) 3068 return true; 3069 // Heuristic: if the block can be if-converted/phi-folded and the 3070 // instructions inside are all cheap (arithmetic/GEPs), it's worthwhile to 3071 // thread this store. 3072 int BudgetRemaining = 3073 PHINodeFoldingThreshold * TargetTransformInfo::TCC_Basic; 3074 for (auto &I : BB->instructionsWithoutDebug()) { 3075 // Consider terminator instruction to be free. 3076 if (I.isTerminator()) 3077 continue; 3078 // If this is one the stores that we want to speculate out of this BB, 3079 // then don't count it's cost, consider it to be free. 3080 if (auto *S = dyn_cast<StoreInst>(&I)) 3081 if (llvm::find(FreeStores, S)) 3082 continue; 3083 // Else, we have a white-list of instructions that we are ak speculating. 3084 if (!isa<BinaryOperator>(I) && !isa<GetElementPtrInst>(I)) 3085 return false; // Not in white-list - not worthwhile folding. 3086 // And finally, if this is a non-free instruction that we are okay 3087 // speculating, ensure that we consider the speculation budget. 3088 BudgetRemaining -= TTI.getUserCost(&I, TargetTransformInfo::TCK_SizeAndLatency); 3089 if (BudgetRemaining < 0) 3090 return false; // Eagerly refuse to fold as soon as we're out of budget. 3091 } 3092 assert(BudgetRemaining >= 0 && 3093 "When we run out of budget we will eagerly return from within the " 3094 "per-instruction loop."); 3095 return true; 3096 }; 3097 3098 const SmallVector<StoreInst *, 2> FreeStores = {PStore, QStore}; 3099 if (!MergeCondStoresAggressively && 3100 (!IsWorthwhile(PTB, FreeStores) || !IsWorthwhile(PFB, FreeStores) || 3101 !IsWorthwhile(QTB, FreeStores) || !IsWorthwhile(QFB, FreeStores))) 3102 return false; 3103 3104 // If PostBB has more than two predecessors, we need to split it so we can 3105 // sink the store. 3106 if (std::next(pred_begin(PostBB), 2) != pred_end(PostBB)) { 3107 // We know that QFB's only successor is PostBB. And QFB has a single 3108 // predecessor. If QTB exists, then its only successor is also PostBB. 3109 // If QTB does not exist, then QFB's only predecessor has a conditional 3110 // branch to QFB and PostBB. 3111 BasicBlock *TruePred = QTB ? QTB : QFB->getSinglePredecessor(); 3112 BasicBlock *NewBB = SplitBlockPredecessors(PostBB, { QFB, TruePred}, 3113 "condstore.split"); 3114 if (!NewBB) 3115 return false; 3116 PostBB = NewBB; 3117 } 3118 3119 // OK, we're going to sink the stores to PostBB. The store has to be 3120 // conditional though, so first create the predicate. 3121 Value *PCond = cast<BranchInst>(PFB->getSinglePredecessor()->getTerminator()) 3122 ->getCondition(); 3123 Value *QCond = cast<BranchInst>(QFB->getSinglePredecessor()->getTerminator()) 3124 ->getCondition(); 3125 3126 Value *PPHI = ensureValueAvailableInSuccessor(PStore->getValueOperand(), 3127 PStore->getParent()); 3128 Value *QPHI = ensureValueAvailableInSuccessor(QStore->getValueOperand(), 3129 QStore->getParent(), PPHI); 3130 3131 IRBuilder<> QB(&*PostBB->getFirstInsertionPt()); 3132 3133 Value *PPred = PStore->getParent() == PTB ? PCond : QB.CreateNot(PCond); 3134 Value *QPred = QStore->getParent() == QTB ? QCond : QB.CreateNot(QCond); 3135 3136 if (InvertPCond) 3137 PPred = QB.CreateNot(PPred); 3138 if (InvertQCond) 3139 QPred = QB.CreateNot(QPred); 3140 Value *CombinedPred = QB.CreateOr(PPred, QPred); 3141 3142 auto *T = 3143 SplitBlockAndInsertIfThen(CombinedPred, &*QB.GetInsertPoint(), false); 3144 QB.SetInsertPoint(T); 3145 StoreInst *SI = cast<StoreInst>(QB.CreateStore(QPHI, Address)); 3146 AAMDNodes AAMD; 3147 PStore->getAAMetadata(AAMD, /*Merge=*/false); 3148 PStore->getAAMetadata(AAMD, /*Merge=*/true); 3149 SI->setAAMetadata(AAMD); 3150 // Choose the minimum alignment. If we could prove both stores execute, we 3151 // could use biggest one. In this case, though, we only know that one of the 3152 // stores executes. And we don't know it's safe to take the alignment from a 3153 // store that doesn't execute. 3154 SI->setAlignment(std::min(PStore->getAlign(), QStore->getAlign())); 3155 3156 QStore->eraseFromParent(); 3157 PStore->eraseFromParent(); 3158 3159 return true; 3160 } 3161 3162 static bool mergeConditionalStores(BranchInst *PBI, BranchInst *QBI, 3163 const DataLayout &DL, 3164 const TargetTransformInfo &TTI) { 3165 // The intention here is to find diamonds or triangles (see below) where each 3166 // conditional block contains a store to the same address. Both of these 3167 // stores are conditional, so they can't be unconditionally sunk. But it may 3168 // be profitable to speculatively sink the stores into one merged store at the 3169 // end, and predicate the merged store on the union of the two conditions of 3170 // PBI and QBI. 3171 // 3172 // This can reduce the number of stores executed if both of the conditions are 3173 // true, and can allow the blocks to become small enough to be if-converted. 3174 // This optimization will also chain, so that ladders of test-and-set 3175 // sequences can be if-converted away. 3176 // 3177 // We only deal with simple diamonds or triangles: 3178 // 3179 // PBI or PBI or a combination of the two 3180 // / \ | \ 3181 // PTB PFB | PFB 3182 // \ / | / 3183 // QBI QBI 3184 // / \ | \ 3185 // QTB QFB | QFB 3186 // \ / | / 3187 // PostBB PostBB 3188 // 3189 // We model triangles as a type of diamond with a nullptr "true" block. 3190 // Triangles are canonicalized so that the fallthrough edge is represented by 3191 // a true condition, as in the diagram above. 3192 BasicBlock *PTB = PBI->getSuccessor(0); 3193 BasicBlock *PFB = PBI->getSuccessor(1); 3194 BasicBlock *QTB = QBI->getSuccessor(0); 3195 BasicBlock *QFB = QBI->getSuccessor(1); 3196 BasicBlock *PostBB = QFB->getSingleSuccessor(); 3197 3198 // Make sure we have a good guess for PostBB. If QTB's only successor is 3199 // QFB, then QFB is a better PostBB. 3200 if (QTB->getSingleSuccessor() == QFB) 3201 PostBB = QFB; 3202 3203 // If we couldn't find a good PostBB, stop. 3204 if (!PostBB) 3205 return false; 3206 3207 bool InvertPCond = false, InvertQCond = false; 3208 // Canonicalize fallthroughs to the true branches. 3209 if (PFB == QBI->getParent()) { 3210 std::swap(PFB, PTB); 3211 InvertPCond = true; 3212 } 3213 if (QFB == PostBB) { 3214 std::swap(QFB, QTB); 3215 InvertQCond = true; 3216 } 3217 3218 // From this point on we can assume PTB or QTB may be fallthroughs but PFB 3219 // and QFB may not. Model fallthroughs as a nullptr block. 3220 if (PTB == QBI->getParent()) 3221 PTB = nullptr; 3222 if (QTB == PostBB) 3223 QTB = nullptr; 3224 3225 // Legality bailouts. We must have at least the non-fallthrough blocks and 3226 // the post-dominating block, and the non-fallthroughs must only have one 3227 // predecessor. 3228 auto HasOnePredAndOneSucc = [](BasicBlock *BB, BasicBlock *P, BasicBlock *S) { 3229 return BB->getSinglePredecessor() == P && BB->getSingleSuccessor() == S; 3230 }; 3231 if (!HasOnePredAndOneSucc(PFB, PBI->getParent(), QBI->getParent()) || 3232 !HasOnePredAndOneSucc(QFB, QBI->getParent(), PostBB)) 3233 return false; 3234 if ((PTB && !HasOnePredAndOneSucc(PTB, PBI->getParent(), QBI->getParent())) || 3235 (QTB && !HasOnePredAndOneSucc(QTB, QBI->getParent(), PostBB))) 3236 return false; 3237 if (!QBI->getParent()->hasNUses(2)) 3238 return false; 3239 3240 // OK, this is a sequence of two diamonds or triangles. 3241 // Check if there are stores in PTB or PFB that are repeated in QTB or QFB. 3242 SmallPtrSet<Value *, 4> PStoreAddresses, QStoreAddresses; 3243 for (auto *BB : {PTB, PFB}) { 3244 if (!BB) 3245 continue; 3246 for (auto &I : *BB) 3247 if (StoreInst *SI = dyn_cast<StoreInst>(&I)) 3248 PStoreAddresses.insert(SI->getPointerOperand()); 3249 } 3250 for (auto *BB : {QTB, QFB}) { 3251 if (!BB) 3252 continue; 3253 for (auto &I : *BB) 3254 if (StoreInst *SI = dyn_cast<StoreInst>(&I)) 3255 QStoreAddresses.insert(SI->getPointerOperand()); 3256 } 3257 3258 set_intersect(PStoreAddresses, QStoreAddresses); 3259 // set_intersect mutates PStoreAddresses in place. Rename it here to make it 3260 // clear what it contains. 3261 auto &CommonAddresses = PStoreAddresses; 3262 3263 bool Changed = false; 3264 for (auto *Address : CommonAddresses) 3265 Changed |= mergeConditionalStoreToAddress( 3266 PTB, PFB, QTB, QFB, PostBB, Address, InvertPCond, InvertQCond, DL, TTI); 3267 return Changed; 3268 } 3269 3270 3271 /// If the previous block ended with a widenable branch, determine if reusing 3272 /// the target block is profitable and legal. This will have the effect of 3273 /// "widening" PBI, but doesn't require us to reason about hosting safety. 3274 static bool tryWidenCondBranchToCondBranch(BranchInst *PBI, BranchInst *BI) { 3275 // TODO: This can be generalized in two important ways: 3276 // 1) We can allow phi nodes in IfFalseBB and simply reuse all the input 3277 // values from the PBI edge. 3278 // 2) We can sink side effecting instructions into BI's fallthrough 3279 // successor provided they doesn't contribute to computation of 3280 // BI's condition. 3281 Value *CondWB, *WC; 3282 BasicBlock *IfTrueBB, *IfFalseBB; 3283 if (!parseWidenableBranch(PBI, CondWB, WC, IfTrueBB, IfFalseBB) || 3284 IfTrueBB != BI->getParent() || !BI->getParent()->getSinglePredecessor()) 3285 return false; 3286 if (!IfFalseBB->phis().empty()) 3287 return false; // TODO 3288 // Use lambda to lazily compute expensive condition after cheap ones. 3289 auto NoSideEffects = [](BasicBlock &BB) { 3290 return !llvm::any_of(BB, [](const Instruction &I) { 3291 return I.mayWriteToMemory() || I.mayHaveSideEffects(); 3292 }); 3293 }; 3294 if (BI->getSuccessor(1) != IfFalseBB && // no inf looping 3295 BI->getSuccessor(1)->getTerminatingDeoptimizeCall() && // profitability 3296 NoSideEffects(*BI->getParent())) { 3297 BI->getSuccessor(1)->removePredecessor(BI->getParent()); 3298 BI->setSuccessor(1, IfFalseBB); 3299 return true; 3300 } 3301 if (BI->getSuccessor(0) != IfFalseBB && // no inf looping 3302 BI->getSuccessor(0)->getTerminatingDeoptimizeCall() && // profitability 3303 NoSideEffects(*BI->getParent())) { 3304 BI->getSuccessor(0)->removePredecessor(BI->getParent()); 3305 BI->setSuccessor(0, IfFalseBB); 3306 return true; 3307 } 3308 return false; 3309 } 3310 3311 /// If we have a conditional branch as a predecessor of another block, 3312 /// this function tries to simplify it. We know 3313 /// that PBI and BI are both conditional branches, and BI is in one of the 3314 /// successor blocks of PBI - PBI branches to BI. 3315 static bool SimplifyCondBranchToCondBranch(BranchInst *PBI, BranchInst *BI, 3316 const DataLayout &DL, 3317 const TargetTransformInfo &TTI) { 3318 assert(PBI->isConditional() && BI->isConditional()); 3319 BasicBlock *BB = BI->getParent(); 3320 3321 // If this block ends with a branch instruction, and if there is a 3322 // predecessor that ends on a branch of the same condition, make 3323 // this conditional branch redundant. 3324 if (PBI->getCondition() == BI->getCondition() && 3325 PBI->getSuccessor(0) != PBI->getSuccessor(1)) { 3326 // Okay, the outcome of this conditional branch is statically 3327 // knowable. If this block had a single pred, handle specially. 3328 if (BB->getSinglePredecessor()) { 3329 // Turn this into a branch on constant. 3330 bool CondIsTrue = PBI->getSuccessor(0) == BB; 3331 BI->setCondition( 3332 ConstantInt::get(Type::getInt1Ty(BB->getContext()), CondIsTrue)); 3333 return true; // Nuke the branch on constant. 3334 } 3335 3336 // Otherwise, if there are multiple predecessors, insert a PHI that merges 3337 // in the constant and simplify the block result. Subsequent passes of 3338 // simplifycfg will thread the block. 3339 if (BlockIsSimpleEnoughToThreadThrough(BB)) { 3340 pred_iterator PB = pred_begin(BB), PE = pred_end(BB); 3341 PHINode *NewPN = PHINode::Create( 3342 Type::getInt1Ty(BB->getContext()), std::distance(PB, PE), 3343 BI->getCondition()->getName() + ".pr", &BB->front()); 3344 // Okay, we're going to insert the PHI node. Since PBI is not the only 3345 // predecessor, compute the PHI'd conditional value for all of the preds. 3346 // Any predecessor where the condition is not computable we keep symbolic. 3347 for (pred_iterator PI = PB; PI != PE; ++PI) { 3348 BasicBlock *P = *PI; 3349 if ((PBI = dyn_cast<BranchInst>(P->getTerminator())) && PBI != BI && 3350 PBI->isConditional() && PBI->getCondition() == BI->getCondition() && 3351 PBI->getSuccessor(0) != PBI->getSuccessor(1)) { 3352 bool CondIsTrue = PBI->getSuccessor(0) == BB; 3353 NewPN->addIncoming( 3354 ConstantInt::get(Type::getInt1Ty(BB->getContext()), CondIsTrue), 3355 P); 3356 } else { 3357 NewPN->addIncoming(BI->getCondition(), P); 3358 } 3359 } 3360 3361 BI->setCondition(NewPN); 3362 return true; 3363 } 3364 } 3365 3366 // If the previous block ended with a widenable branch, determine if reusing 3367 // the target block is profitable and legal. This will have the effect of 3368 // "widening" PBI, but doesn't require us to reason about hosting safety. 3369 if (tryWidenCondBranchToCondBranch(PBI, BI)) 3370 return true; 3371 3372 if (auto *CE = dyn_cast<ConstantExpr>(BI->getCondition())) 3373 if (CE->canTrap()) 3374 return false; 3375 3376 // If both branches are conditional and both contain stores to the same 3377 // address, remove the stores from the conditionals and create a conditional 3378 // merged store at the end. 3379 if (MergeCondStores && mergeConditionalStores(PBI, BI, DL, TTI)) 3380 return true; 3381 3382 // If this is a conditional branch in an empty block, and if any 3383 // predecessors are a conditional branch to one of our destinations, 3384 // fold the conditions into logical ops and one cond br. 3385 3386 // Ignore dbg intrinsics. 3387 if (&*BB->instructionsWithoutDebug().begin() != BI) 3388 return false; 3389 3390 int PBIOp, BIOp; 3391 if (PBI->getSuccessor(0) == BI->getSuccessor(0)) { 3392 PBIOp = 0; 3393 BIOp = 0; 3394 } else if (PBI->getSuccessor(0) == BI->getSuccessor(1)) { 3395 PBIOp = 0; 3396 BIOp = 1; 3397 } else if (PBI->getSuccessor(1) == BI->getSuccessor(0)) { 3398 PBIOp = 1; 3399 BIOp = 0; 3400 } else if (PBI->getSuccessor(1) == BI->getSuccessor(1)) { 3401 PBIOp = 1; 3402 BIOp = 1; 3403 } else { 3404 return false; 3405 } 3406 3407 // Check to make sure that the other destination of this branch 3408 // isn't BB itself. If so, this is an infinite loop that will 3409 // keep getting unwound. 3410 if (PBI->getSuccessor(PBIOp) == BB) 3411 return false; 3412 3413 // Do not perform this transformation if it would require 3414 // insertion of a large number of select instructions. For targets 3415 // without predication/cmovs, this is a big pessimization. 3416 3417 // Also do not perform this transformation if any phi node in the common 3418 // destination block can trap when reached by BB or PBB (PR17073). In that 3419 // case, it would be unsafe to hoist the operation into a select instruction. 3420 3421 BasicBlock *CommonDest = PBI->getSuccessor(PBIOp); 3422 unsigned NumPhis = 0; 3423 for (BasicBlock::iterator II = CommonDest->begin(); isa<PHINode>(II); 3424 ++II, ++NumPhis) { 3425 if (NumPhis > 2) // Disable this xform. 3426 return false; 3427 3428 PHINode *PN = cast<PHINode>(II); 3429 Value *BIV = PN->getIncomingValueForBlock(BB); 3430 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(BIV)) 3431 if (CE->canTrap()) 3432 return false; 3433 3434 unsigned PBBIdx = PN->getBasicBlockIndex(PBI->getParent()); 3435 Value *PBIV = PN->getIncomingValue(PBBIdx); 3436 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(PBIV)) 3437 if (CE->canTrap()) 3438 return false; 3439 } 3440 3441 // Finally, if everything is ok, fold the branches to logical ops. 3442 BasicBlock *OtherDest = BI->getSuccessor(BIOp ^ 1); 3443 3444 LLVM_DEBUG(dbgs() << "FOLDING BRs:" << *PBI->getParent() 3445 << "AND: " << *BI->getParent()); 3446 3447 // If OtherDest *is* BB, then BB is a basic block with a single conditional 3448 // branch in it, where one edge (OtherDest) goes back to itself but the other 3449 // exits. We don't *know* that the program avoids the infinite loop 3450 // (even though that seems likely). If we do this xform naively, we'll end up 3451 // recursively unpeeling the loop. Since we know that (after the xform is 3452 // done) that the block *is* infinite if reached, we just make it an obviously 3453 // infinite loop with no cond branch. 3454 if (OtherDest == BB) { 3455 // Insert it at the end of the function, because it's either code, 3456 // or it won't matter if it's hot. :) 3457 BasicBlock *InfLoopBlock = 3458 BasicBlock::Create(BB->getContext(), "infloop", BB->getParent()); 3459 BranchInst::Create(InfLoopBlock, InfLoopBlock); 3460 OtherDest = InfLoopBlock; 3461 } 3462 3463 LLVM_DEBUG(dbgs() << *PBI->getParent()->getParent()); 3464 3465 // BI may have other predecessors. Because of this, we leave 3466 // it alone, but modify PBI. 3467 3468 // Make sure we get to CommonDest on True&True directions. 3469 Value *PBICond = PBI->getCondition(); 3470 IRBuilder<NoFolder> Builder(PBI); 3471 if (PBIOp) 3472 PBICond = Builder.CreateNot(PBICond, PBICond->getName() + ".not"); 3473 3474 Value *BICond = BI->getCondition(); 3475 if (BIOp) 3476 BICond = Builder.CreateNot(BICond, BICond->getName() + ".not"); 3477 3478 // Merge the conditions. 3479 Value *Cond = Builder.CreateOr(PBICond, BICond, "brmerge"); 3480 3481 // Modify PBI to branch on the new condition to the new dests. 3482 PBI->setCondition(Cond); 3483 PBI->setSuccessor(0, CommonDest); 3484 PBI->setSuccessor(1, OtherDest); 3485 3486 // Update branch weight for PBI. 3487 uint64_t PredTrueWeight, PredFalseWeight, SuccTrueWeight, SuccFalseWeight; 3488 uint64_t PredCommon, PredOther, SuccCommon, SuccOther; 3489 bool HasWeights = 3490 extractPredSuccWeights(PBI, BI, PredTrueWeight, PredFalseWeight, 3491 SuccTrueWeight, SuccFalseWeight); 3492 if (HasWeights) { 3493 PredCommon = PBIOp ? PredFalseWeight : PredTrueWeight; 3494 PredOther = PBIOp ? PredTrueWeight : PredFalseWeight; 3495 SuccCommon = BIOp ? SuccFalseWeight : SuccTrueWeight; 3496 SuccOther = BIOp ? SuccTrueWeight : SuccFalseWeight; 3497 // The weight to CommonDest should be PredCommon * SuccTotal + 3498 // PredOther * SuccCommon. 3499 // The weight to OtherDest should be PredOther * SuccOther. 3500 uint64_t NewWeights[2] = {PredCommon * (SuccCommon + SuccOther) + 3501 PredOther * SuccCommon, 3502 PredOther * SuccOther}; 3503 // Halve the weights if any of them cannot fit in an uint32_t 3504 FitWeights(NewWeights); 3505 3506 setBranchWeights(PBI, NewWeights[0], NewWeights[1]); 3507 } 3508 3509 // OtherDest may have phi nodes. If so, add an entry from PBI's 3510 // block that are identical to the entries for BI's block. 3511 AddPredecessorToBlock(OtherDest, PBI->getParent(), BB); 3512 3513 // We know that the CommonDest already had an edge from PBI to 3514 // it. If it has PHIs though, the PHIs may have different 3515 // entries for BB and PBI's BB. If so, insert a select to make 3516 // them agree. 3517 for (PHINode &PN : CommonDest->phis()) { 3518 Value *BIV = PN.getIncomingValueForBlock(BB); 3519 unsigned PBBIdx = PN.getBasicBlockIndex(PBI->getParent()); 3520 Value *PBIV = PN.getIncomingValue(PBBIdx); 3521 if (BIV != PBIV) { 3522 // Insert a select in PBI to pick the right value. 3523 SelectInst *NV = cast<SelectInst>( 3524 Builder.CreateSelect(PBICond, PBIV, BIV, PBIV->getName() + ".mux")); 3525 PN.setIncomingValue(PBBIdx, NV); 3526 // Although the select has the same condition as PBI, the original branch 3527 // weights for PBI do not apply to the new select because the select's 3528 // 'logical' edges are incoming edges of the phi that is eliminated, not 3529 // the outgoing edges of PBI. 3530 if (HasWeights) { 3531 uint64_t PredCommon = PBIOp ? PredFalseWeight : PredTrueWeight; 3532 uint64_t PredOther = PBIOp ? PredTrueWeight : PredFalseWeight; 3533 uint64_t SuccCommon = BIOp ? SuccFalseWeight : SuccTrueWeight; 3534 uint64_t SuccOther = BIOp ? SuccTrueWeight : SuccFalseWeight; 3535 // The weight to PredCommonDest should be PredCommon * SuccTotal. 3536 // The weight to PredOtherDest should be PredOther * SuccCommon. 3537 uint64_t NewWeights[2] = {PredCommon * (SuccCommon + SuccOther), 3538 PredOther * SuccCommon}; 3539 3540 FitWeights(NewWeights); 3541 3542 setBranchWeights(NV, NewWeights[0], NewWeights[1]); 3543 } 3544 } 3545 } 3546 3547 LLVM_DEBUG(dbgs() << "INTO: " << *PBI->getParent()); 3548 LLVM_DEBUG(dbgs() << *PBI->getParent()->getParent()); 3549 3550 // This basic block is probably dead. We know it has at least 3551 // one fewer predecessor. 3552 return true; 3553 } 3554 3555 // Simplifies a terminator by replacing it with a branch to TrueBB if Cond is 3556 // true or to FalseBB if Cond is false. 3557 // Takes care of updating the successors and removing the old terminator. 3558 // Also makes sure not to introduce new successors by assuming that edges to 3559 // non-successor TrueBBs and FalseBBs aren't reachable. 3560 bool SimplifyCFGOpt::SimplifyTerminatorOnSelect(Instruction *OldTerm, 3561 Value *Cond, BasicBlock *TrueBB, 3562 BasicBlock *FalseBB, 3563 uint32_t TrueWeight, 3564 uint32_t FalseWeight) { 3565 // Remove any superfluous successor edges from the CFG. 3566 // First, figure out which successors to preserve. 3567 // If TrueBB and FalseBB are equal, only try to preserve one copy of that 3568 // successor. 3569 BasicBlock *KeepEdge1 = TrueBB; 3570 BasicBlock *KeepEdge2 = TrueBB != FalseBB ? FalseBB : nullptr; 3571 3572 // Then remove the rest. 3573 for (BasicBlock *Succ : successors(OldTerm)) { 3574 // Make sure only to keep exactly one copy of each edge. 3575 if (Succ == KeepEdge1) 3576 KeepEdge1 = nullptr; 3577 else if (Succ == KeepEdge2) 3578 KeepEdge2 = nullptr; 3579 else 3580 Succ->removePredecessor(OldTerm->getParent(), 3581 /*KeepOneInputPHIs=*/true); 3582 } 3583 3584 IRBuilder<> Builder(OldTerm); 3585 Builder.SetCurrentDebugLocation(OldTerm->getDebugLoc()); 3586 3587 // Insert an appropriate new terminator. 3588 if (!KeepEdge1 && !KeepEdge2) { 3589 if (TrueBB == FalseBB) 3590 // We were only looking for one successor, and it was present. 3591 // Create an unconditional branch to it. 3592 Builder.CreateBr(TrueBB); 3593 else { 3594 // We found both of the successors we were looking for. 3595 // Create a conditional branch sharing the condition of the select. 3596 BranchInst *NewBI = Builder.CreateCondBr(Cond, TrueBB, FalseBB); 3597 if (TrueWeight != FalseWeight) 3598 setBranchWeights(NewBI, TrueWeight, FalseWeight); 3599 } 3600 } else if (KeepEdge1 && (KeepEdge2 || TrueBB == FalseBB)) { 3601 // Neither of the selected blocks were successors, so this 3602 // terminator must be unreachable. 3603 new UnreachableInst(OldTerm->getContext(), OldTerm); 3604 } else { 3605 // One of the selected values was a successor, but the other wasn't. 3606 // Insert an unconditional branch to the one that was found; 3607 // the edge to the one that wasn't must be unreachable. 3608 if (!KeepEdge1) 3609 // Only TrueBB was found. 3610 Builder.CreateBr(TrueBB); 3611 else 3612 // Only FalseBB was found. 3613 Builder.CreateBr(FalseBB); 3614 } 3615 3616 EraseTerminatorAndDCECond(OldTerm); 3617 return true; 3618 } 3619 3620 // Replaces 3621 // (switch (select cond, X, Y)) on constant X, Y 3622 // with a branch - conditional if X and Y lead to distinct BBs, 3623 // unconditional otherwise. 3624 bool SimplifyCFGOpt::SimplifySwitchOnSelect(SwitchInst *SI, 3625 SelectInst *Select) { 3626 // Check for constant integer values in the select. 3627 ConstantInt *TrueVal = dyn_cast<ConstantInt>(Select->getTrueValue()); 3628 ConstantInt *FalseVal = dyn_cast<ConstantInt>(Select->getFalseValue()); 3629 if (!TrueVal || !FalseVal) 3630 return false; 3631 3632 // Find the relevant condition and destinations. 3633 Value *Condition = Select->getCondition(); 3634 BasicBlock *TrueBB = SI->findCaseValue(TrueVal)->getCaseSuccessor(); 3635 BasicBlock *FalseBB = SI->findCaseValue(FalseVal)->getCaseSuccessor(); 3636 3637 // Get weight for TrueBB and FalseBB. 3638 uint32_t TrueWeight = 0, FalseWeight = 0; 3639 SmallVector<uint64_t, 8> Weights; 3640 bool HasWeights = HasBranchWeights(SI); 3641 if (HasWeights) { 3642 GetBranchWeights(SI, Weights); 3643 if (Weights.size() == 1 + SI->getNumCases()) { 3644 TrueWeight = 3645 (uint32_t)Weights[SI->findCaseValue(TrueVal)->getSuccessorIndex()]; 3646 FalseWeight = 3647 (uint32_t)Weights[SI->findCaseValue(FalseVal)->getSuccessorIndex()]; 3648 } 3649 } 3650 3651 // Perform the actual simplification. 3652 return SimplifyTerminatorOnSelect(SI, Condition, TrueBB, FalseBB, TrueWeight, 3653 FalseWeight); 3654 } 3655 3656 // Replaces 3657 // (indirectbr (select cond, blockaddress(@fn, BlockA), 3658 // blockaddress(@fn, BlockB))) 3659 // with 3660 // (br cond, BlockA, BlockB). 3661 bool SimplifyCFGOpt::SimplifyIndirectBrOnSelect(IndirectBrInst *IBI, 3662 SelectInst *SI) { 3663 // Check that both operands of the select are block addresses. 3664 BlockAddress *TBA = dyn_cast<BlockAddress>(SI->getTrueValue()); 3665 BlockAddress *FBA = dyn_cast<BlockAddress>(SI->getFalseValue()); 3666 if (!TBA || !FBA) 3667 return false; 3668 3669 // Extract the actual blocks. 3670 BasicBlock *TrueBB = TBA->getBasicBlock(); 3671 BasicBlock *FalseBB = FBA->getBasicBlock(); 3672 3673 // Perform the actual simplification. 3674 return SimplifyTerminatorOnSelect(IBI, SI->getCondition(), TrueBB, FalseBB, 0, 3675 0); 3676 } 3677 3678 /// This is called when we find an icmp instruction 3679 /// (a seteq/setne with a constant) as the only instruction in a 3680 /// block that ends with an uncond branch. We are looking for a very specific 3681 /// pattern that occurs when "A == 1 || A == 2 || A == 3" gets simplified. In 3682 /// this case, we merge the first two "or's of icmp" into a switch, but then the 3683 /// default value goes to an uncond block with a seteq in it, we get something 3684 /// like: 3685 /// 3686 /// switch i8 %A, label %DEFAULT [ i8 1, label %end i8 2, label %end ] 3687 /// DEFAULT: 3688 /// %tmp = icmp eq i8 %A, 92 3689 /// br label %end 3690 /// end: 3691 /// ... = phi i1 [ true, %entry ], [ %tmp, %DEFAULT ], [ true, %entry ] 3692 /// 3693 /// We prefer to split the edge to 'end' so that there is a true/false entry to 3694 /// the PHI, merging the third icmp into the switch. 3695 bool SimplifyCFGOpt::tryToSimplifyUncondBranchWithICmpInIt( 3696 ICmpInst *ICI, IRBuilder<> &Builder) { 3697 BasicBlock *BB = ICI->getParent(); 3698 3699 // If the block has any PHIs in it or the icmp has multiple uses, it is too 3700 // complex. 3701 if (isa<PHINode>(BB->begin()) || !ICI->hasOneUse()) 3702 return false; 3703 3704 Value *V = ICI->getOperand(0); 3705 ConstantInt *Cst = cast<ConstantInt>(ICI->getOperand(1)); 3706 3707 // The pattern we're looking for is where our only predecessor is a switch on 3708 // 'V' and this block is the default case for the switch. In this case we can 3709 // fold the compared value into the switch to simplify things. 3710 BasicBlock *Pred = BB->getSinglePredecessor(); 3711 if (!Pred || !isa<SwitchInst>(Pred->getTerminator())) 3712 return false; 3713 3714 SwitchInst *SI = cast<SwitchInst>(Pred->getTerminator()); 3715 if (SI->getCondition() != V) 3716 return false; 3717 3718 // If BB is reachable on a non-default case, then we simply know the value of 3719 // V in this block. Substitute it and constant fold the icmp instruction 3720 // away. 3721 if (SI->getDefaultDest() != BB) { 3722 ConstantInt *VVal = SI->findCaseDest(BB); 3723 assert(VVal && "Should have a unique destination value"); 3724 ICI->setOperand(0, VVal); 3725 3726 if (Value *V = SimplifyInstruction(ICI, {DL, ICI})) { 3727 ICI->replaceAllUsesWith(V); 3728 ICI->eraseFromParent(); 3729 } 3730 // BB is now empty, so it is likely to simplify away. 3731 return requestResimplify(); 3732 } 3733 3734 // Ok, the block is reachable from the default dest. If the constant we're 3735 // comparing exists in one of the other edges, then we can constant fold ICI 3736 // and zap it. 3737 if (SI->findCaseValue(Cst) != SI->case_default()) { 3738 Value *V; 3739 if (ICI->getPredicate() == ICmpInst::ICMP_EQ) 3740 V = ConstantInt::getFalse(BB->getContext()); 3741 else 3742 V = ConstantInt::getTrue(BB->getContext()); 3743 3744 ICI->replaceAllUsesWith(V); 3745 ICI->eraseFromParent(); 3746 // BB is now empty, so it is likely to simplify away. 3747 return requestResimplify(); 3748 } 3749 3750 // The use of the icmp has to be in the 'end' block, by the only PHI node in 3751 // the block. 3752 BasicBlock *SuccBlock = BB->getTerminator()->getSuccessor(0); 3753 PHINode *PHIUse = dyn_cast<PHINode>(ICI->user_back()); 3754 if (PHIUse == nullptr || PHIUse != &SuccBlock->front() || 3755 isa<PHINode>(++BasicBlock::iterator(PHIUse))) 3756 return false; 3757 3758 // If the icmp is a SETEQ, then the default dest gets false, the new edge gets 3759 // true in the PHI. 3760 Constant *DefaultCst = ConstantInt::getTrue(BB->getContext()); 3761 Constant *NewCst = ConstantInt::getFalse(BB->getContext()); 3762 3763 if (ICI->getPredicate() == ICmpInst::ICMP_EQ) 3764 std::swap(DefaultCst, NewCst); 3765 3766 // Replace ICI (which is used by the PHI for the default value) with true or 3767 // false depending on if it is EQ or NE. 3768 ICI->replaceAllUsesWith(DefaultCst); 3769 ICI->eraseFromParent(); 3770 3771 // Okay, the switch goes to this block on a default value. Add an edge from 3772 // the switch to the merge point on the compared value. 3773 BasicBlock *NewBB = 3774 BasicBlock::Create(BB->getContext(), "switch.edge", BB->getParent(), BB); 3775 { 3776 SwitchInstProfUpdateWrapper SIW(*SI); 3777 auto W0 = SIW.getSuccessorWeight(0); 3778 SwitchInstProfUpdateWrapper::CaseWeightOpt NewW; 3779 if (W0) { 3780 NewW = ((uint64_t(*W0) + 1) >> 1); 3781 SIW.setSuccessorWeight(0, *NewW); 3782 } 3783 SIW.addCase(Cst, NewBB, NewW); 3784 } 3785 3786 // NewBB branches to the phi block, add the uncond branch and the phi entry. 3787 Builder.SetInsertPoint(NewBB); 3788 Builder.SetCurrentDebugLocation(SI->getDebugLoc()); 3789 Builder.CreateBr(SuccBlock); 3790 PHIUse->addIncoming(NewCst, NewBB); 3791 return true; 3792 } 3793 3794 /// The specified branch is a conditional branch. 3795 /// Check to see if it is branching on an or/and chain of icmp instructions, and 3796 /// fold it into a switch instruction if so. 3797 bool SimplifyCFGOpt::SimplifyBranchOnICmpChain(BranchInst *BI, 3798 IRBuilder<> &Builder, 3799 const DataLayout &DL) { 3800 Instruction *Cond = dyn_cast<Instruction>(BI->getCondition()); 3801 if (!Cond) 3802 return false; 3803 3804 // Change br (X == 0 | X == 1), T, F into a switch instruction. 3805 // If this is a bunch of seteq's or'd together, or if it's a bunch of 3806 // 'setne's and'ed together, collect them. 3807 3808 // Try to gather values from a chain of and/or to be turned into a switch 3809 ConstantComparesGatherer ConstantCompare(Cond, DL); 3810 // Unpack the result 3811 SmallVectorImpl<ConstantInt *> &Values = ConstantCompare.Vals; 3812 Value *CompVal = ConstantCompare.CompValue; 3813 unsigned UsedICmps = ConstantCompare.UsedICmps; 3814 Value *ExtraCase = ConstantCompare.Extra; 3815 3816 // If we didn't have a multiply compared value, fail. 3817 if (!CompVal) 3818 return false; 3819 3820 // Avoid turning single icmps into a switch. 3821 if (UsedICmps <= 1) 3822 return false; 3823 3824 bool TrueWhenEqual = (Cond->getOpcode() == Instruction::Or); 3825 3826 // There might be duplicate constants in the list, which the switch 3827 // instruction can't handle, remove them now. 3828 array_pod_sort(Values.begin(), Values.end(), ConstantIntSortPredicate); 3829 Values.erase(std::unique(Values.begin(), Values.end()), Values.end()); 3830 3831 // If Extra was used, we require at least two switch values to do the 3832 // transformation. A switch with one value is just a conditional branch. 3833 if (ExtraCase && Values.size() < 2) 3834 return false; 3835 3836 // TODO: Preserve branch weight metadata, similarly to how 3837 // FoldValueComparisonIntoPredecessors preserves it. 3838 3839 // Figure out which block is which destination. 3840 BasicBlock *DefaultBB = BI->getSuccessor(1); 3841 BasicBlock *EdgeBB = BI->getSuccessor(0); 3842 if (!TrueWhenEqual) 3843 std::swap(DefaultBB, EdgeBB); 3844 3845 BasicBlock *BB = BI->getParent(); 3846 3847 // MSAN does not like undefs as branch condition which can be introduced 3848 // with "explicit branch". 3849 if (ExtraCase && BB->getParent()->hasFnAttribute(Attribute::SanitizeMemory)) 3850 return false; 3851 3852 LLVM_DEBUG(dbgs() << "Converting 'icmp' chain with " << Values.size() 3853 << " cases into SWITCH. BB is:\n" 3854 << *BB); 3855 3856 // If there are any extra values that couldn't be folded into the switch 3857 // then we evaluate them with an explicit branch first. Split the block 3858 // right before the condbr to handle it. 3859 if (ExtraCase) { 3860 BasicBlock *NewBB = 3861 BB->splitBasicBlock(BI->getIterator(), "switch.early.test"); 3862 // Remove the uncond branch added to the old block. 3863 Instruction *OldTI = BB->getTerminator(); 3864 Builder.SetInsertPoint(OldTI); 3865 3866 if (TrueWhenEqual) 3867 Builder.CreateCondBr(ExtraCase, EdgeBB, NewBB); 3868 else 3869 Builder.CreateCondBr(ExtraCase, NewBB, EdgeBB); 3870 3871 OldTI->eraseFromParent(); 3872 3873 // If there are PHI nodes in EdgeBB, then we need to add a new entry to them 3874 // for the edge we just added. 3875 AddPredecessorToBlock(EdgeBB, BB, NewBB); 3876 3877 LLVM_DEBUG(dbgs() << " ** 'icmp' chain unhandled condition: " << *ExtraCase 3878 << "\nEXTRABB = " << *BB); 3879 BB = NewBB; 3880 } 3881 3882 Builder.SetInsertPoint(BI); 3883 // Convert pointer to int before we switch. 3884 if (CompVal->getType()->isPointerTy()) { 3885 CompVal = Builder.CreatePtrToInt( 3886 CompVal, DL.getIntPtrType(CompVal->getType()), "magicptr"); 3887 } 3888 3889 // Create the new switch instruction now. 3890 SwitchInst *New = Builder.CreateSwitch(CompVal, DefaultBB, Values.size()); 3891 3892 // Add all of the 'cases' to the switch instruction. 3893 for (unsigned i = 0, e = Values.size(); i != e; ++i) 3894 New->addCase(Values[i], EdgeBB); 3895 3896 // We added edges from PI to the EdgeBB. As such, if there were any 3897 // PHI nodes in EdgeBB, they need entries to be added corresponding to 3898 // the number of edges added. 3899 for (BasicBlock::iterator BBI = EdgeBB->begin(); isa<PHINode>(BBI); ++BBI) { 3900 PHINode *PN = cast<PHINode>(BBI); 3901 Value *InVal = PN->getIncomingValueForBlock(BB); 3902 for (unsigned i = 0, e = Values.size() - 1; i != e; ++i) 3903 PN->addIncoming(InVal, BB); 3904 } 3905 3906 // Erase the old branch instruction. 3907 EraseTerminatorAndDCECond(BI); 3908 3909 LLVM_DEBUG(dbgs() << " ** 'icmp' chain result is:\n" << *BB << '\n'); 3910 return true; 3911 } 3912 3913 bool SimplifyCFGOpt::simplifyResume(ResumeInst *RI, IRBuilder<> &Builder) { 3914 if (isa<PHINode>(RI->getValue())) 3915 return simplifyCommonResume(RI); 3916 else if (isa<LandingPadInst>(RI->getParent()->getFirstNonPHI()) && 3917 RI->getValue() == RI->getParent()->getFirstNonPHI()) 3918 // The resume must unwind the exception that caused control to branch here. 3919 return simplifySingleResume(RI); 3920 3921 return false; 3922 } 3923 3924 // Simplify resume that is shared by several landing pads (phi of landing pad). 3925 bool SimplifyCFGOpt::simplifyCommonResume(ResumeInst *RI) { 3926 BasicBlock *BB = RI->getParent(); 3927 3928 // Check that there are no other instructions except for debug intrinsics 3929 // between the phi of landing pads (RI->getValue()) and resume instruction. 3930 BasicBlock::iterator I = cast<Instruction>(RI->getValue())->getIterator(), 3931 E = RI->getIterator(); 3932 while (++I != E) 3933 if (!isa<DbgInfoIntrinsic>(I)) 3934 return false; 3935 3936 SmallSetVector<BasicBlock *, 4> TrivialUnwindBlocks; 3937 auto *PhiLPInst = cast<PHINode>(RI->getValue()); 3938 3939 // Check incoming blocks to see if any of them are trivial. 3940 for (unsigned Idx = 0, End = PhiLPInst->getNumIncomingValues(); Idx != End; 3941 Idx++) { 3942 auto *IncomingBB = PhiLPInst->getIncomingBlock(Idx); 3943 auto *IncomingValue = PhiLPInst->getIncomingValue(Idx); 3944 3945 // If the block has other successors, we can not delete it because 3946 // it has other dependents. 3947 if (IncomingBB->getUniqueSuccessor() != BB) 3948 continue; 3949 3950 auto *LandingPad = dyn_cast<LandingPadInst>(IncomingBB->getFirstNonPHI()); 3951 // Not the landing pad that caused the control to branch here. 3952 if (IncomingValue != LandingPad) 3953 continue; 3954 3955 bool isTrivial = true; 3956 3957 I = IncomingBB->getFirstNonPHI()->getIterator(); 3958 E = IncomingBB->getTerminator()->getIterator(); 3959 while (++I != E) 3960 if (!isa<DbgInfoIntrinsic>(I)) { 3961 isTrivial = false; 3962 break; 3963 } 3964 3965 if (isTrivial) 3966 TrivialUnwindBlocks.insert(IncomingBB); 3967 } 3968 3969 // If no trivial unwind blocks, don't do any simplifications. 3970 if (TrivialUnwindBlocks.empty()) 3971 return false; 3972 3973 // Turn all invokes that unwind here into calls. 3974 for (auto *TrivialBB : TrivialUnwindBlocks) { 3975 // Blocks that will be simplified should be removed from the phi node. 3976 // Note there could be multiple edges to the resume block, and we need 3977 // to remove them all. 3978 while (PhiLPInst->getBasicBlockIndex(TrivialBB) != -1) 3979 BB->removePredecessor(TrivialBB, true); 3980 3981 for (pred_iterator PI = pred_begin(TrivialBB), PE = pred_end(TrivialBB); 3982 PI != PE;) { 3983 BasicBlock *Pred = *PI++; 3984 removeUnwindEdge(Pred); 3985 } 3986 3987 // In each SimplifyCFG run, only the current processed block can be erased. 3988 // Otherwise, it will break the iteration of SimplifyCFG pass. So instead 3989 // of erasing TrivialBB, we only remove the branch to the common resume 3990 // block so that we can later erase the resume block since it has no 3991 // predecessors. 3992 TrivialBB->getTerminator()->eraseFromParent(); 3993 new UnreachableInst(RI->getContext(), TrivialBB); 3994 } 3995 3996 // Delete the resume block if all its predecessors have been removed. 3997 if (pred_empty(BB)) 3998 BB->eraseFromParent(); 3999 4000 return !TrivialUnwindBlocks.empty(); 4001 } 4002 4003 // Check if cleanup block is empty 4004 static bool isCleanupBlockEmpty(Instruction *Inst, Instruction *RI) { 4005 BasicBlock::iterator I = Inst->getIterator(), E = RI->getIterator(); 4006 while (++I != E) { 4007 auto *II = dyn_cast<IntrinsicInst>(I); 4008 if (!II) 4009 return false; 4010 4011 Intrinsic::ID IntrinsicID = II->getIntrinsicID(); 4012 switch (IntrinsicID) { 4013 case Intrinsic::dbg_declare: 4014 case Intrinsic::dbg_value: 4015 case Intrinsic::dbg_label: 4016 case Intrinsic::lifetime_end: 4017 break; 4018 default: 4019 return false; 4020 } 4021 } 4022 return true; 4023 } 4024 4025 // Simplify resume that is only used by a single (non-phi) landing pad. 4026 bool SimplifyCFGOpt::simplifySingleResume(ResumeInst *RI) { 4027 BasicBlock *BB = RI->getParent(); 4028 auto *LPInst = cast<LandingPadInst>(BB->getFirstNonPHI()); 4029 assert(RI->getValue() == LPInst && 4030 "Resume must unwind the exception that caused control to here"); 4031 4032 // Check that there are no other instructions except for debug intrinsics. 4033 if (!isCleanupBlockEmpty(LPInst, RI)) 4034 return false; 4035 4036 // Turn all invokes that unwind here into calls and delete the basic block. 4037 for (pred_iterator PI = pred_begin(BB), PE = pred_end(BB); PI != PE;) { 4038 BasicBlock *Pred = *PI++; 4039 removeUnwindEdge(Pred); 4040 } 4041 4042 // The landingpad is now unreachable. Zap it. 4043 if (LoopHeaders) 4044 LoopHeaders->erase(BB); 4045 BB->eraseFromParent(); 4046 return true; 4047 } 4048 4049 static bool removeEmptyCleanup(CleanupReturnInst *RI) { 4050 // If this is a trivial cleanup pad that executes no instructions, it can be 4051 // eliminated. If the cleanup pad continues to the caller, any predecessor 4052 // that is an EH pad will be updated to continue to the caller and any 4053 // predecessor that terminates with an invoke instruction will have its invoke 4054 // instruction converted to a call instruction. If the cleanup pad being 4055 // simplified does not continue to the caller, each predecessor will be 4056 // updated to continue to the unwind destination of the cleanup pad being 4057 // simplified. 4058 BasicBlock *BB = RI->getParent(); 4059 CleanupPadInst *CPInst = RI->getCleanupPad(); 4060 if (CPInst->getParent() != BB) 4061 // This isn't an empty cleanup. 4062 return false; 4063 4064 // We cannot kill the pad if it has multiple uses. This typically arises 4065 // from unreachable basic blocks. 4066 if (!CPInst->hasOneUse()) 4067 return false; 4068 4069 // Check that there are no other instructions except for benign intrinsics. 4070 if (!isCleanupBlockEmpty(CPInst, RI)) 4071 return false; 4072 4073 // If the cleanup return we are simplifying unwinds to the caller, this will 4074 // set UnwindDest to nullptr. 4075 BasicBlock *UnwindDest = RI->getUnwindDest(); 4076 Instruction *DestEHPad = UnwindDest ? UnwindDest->getFirstNonPHI() : nullptr; 4077 4078 // We're about to remove BB from the control flow. Before we do, sink any 4079 // PHINodes into the unwind destination. Doing this before changing the 4080 // control flow avoids some potentially slow checks, since we can currently 4081 // be certain that UnwindDest and BB have no common predecessors (since they 4082 // are both EH pads). 4083 if (UnwindDest) { 4084 // First, go through the PHI nodes in UnwindDest and update any nodes that 4085 // reference the block we are removing 4086 for (BasicBlock::iterator I = UnwindDest->begin(), 4087 IE = DestEHPad->getIterator(); 4088 I != IE; ++I) { 4089 PHINode *DestPN = cast<PHINode>(I); 4090 4091 int Idx = DestPN->getBasicBlockIndex(BB); 4092 // Since BB unwinds to UnwindDest, it has to be in the PHI node. 4093 assert(Idx != -1); 4094 // This PHI node has an incoming value that corresponds to a control 4095 // path through the cleanup pad we are removing. If the incoming 4096 // value is in the cleanup pad, it must be a PHINode (because we 4097 // verified above that the block is otherwise empty). Otherwise, the 4098 // value is either a constant or a value that dominates the cleanup 4099 // pad being removed. 4100 // 4101 // Because BB and UnwindDest are both EH pads, all of their 4102 // predecessors must unwind to these blocks, and since no instruction 4103 // can have multiple unwind destinations, there will be no overlap in 4104 // incoming blocks between SrcPN and DestPN. 4105 Value *SrcVal = DestPN->getIncomingValue(Idx); 4106 PHINode *SrcPN = dyn_cast<PHINode>(SrcVal); 4107 4108 // Remove the entry for the block we are deleting. 4109 DestPN->removeIncomingValue(Idx, false); 4110 4111 if (SrcPN && SrcPN->getParent() == BB) { 4112 // If the incoming value was a PHI node in the cleanup pad we are 4113 // removing, we need to merge that PHI node's incoming values into 4114 // DestPN. 4115 for (unsigned SrcIdx = 0, SrcE = SrcPN->getNumIncomingValues(); 4116 SrcIdx != SrcE; ++SrcIdx) { 4117 DestPN->addIncoming(SrcPN->getIncomingValue(SrcIdx), 4118 SrcPN->getIncomingBlock(SrcIdx)); 4119 } 4120 } else { 4121 // Otherwise, the incoming value came from above BB and 4122 // so we can just reuse it. We must associate all of BB's 4123 // predecessors with this value. 4124 for (auto *pred : predecessors(BB)) { 4125 DestPN->addIncoming(SrcVal, pred); 4126 } 4127 } 4128 } 4129 4130 // Sink any remaining PHI nodes directly into UnwindDest. 4131 Instruction *InsertPt = DestEHPad; 4132 for (BasicBlock::iterator I = BB->begin(), 4133 IE = BB->getFirstNonPHI()->getIterator(); 4134 I != IE;) { 4135 // The iterator must be incremented here because the instructions are 4136 // being moved to another block. 4137 PHINode *PN = cast<PHINode>(I++); 4138 if (PN->use_empty() || !PN->isUsedOutsideOfBlock(BB)) 4139 // If the PHI node has no uses or all of its uses are in this basic 4140 // block (meaning they are debug or lifetime intrinsics), just leave 4141 // it. It will be erased when we erase BB below. 4142 continue; 4143 4144 // Otherwise, sink this PHI node into UnwindDest. 4145 // Any predecessors to UnwindDest which are not already represented 4146 // must be back edges which inherit the value from the path through 4147 // BB. In this case, the PHI value must reference itself. 4148 for (auto *pred : predecessors(UnwindDest)) 4149 if (pred != BB) 4150 PN->addIncoming(PN, pred); 4151 PN->moveBefore(InsertPt); 4152 } 4153 } 4154 4155 for (pred_iterator PI = pred_begin(BB), PE = pred_end(BB); PI != PE;) { 4156 // The iterator must be updated here because we are removing this pred. 4157 BasicBlock *PredBB = *PI++; 4158 if (UnwindDest == nullptr) { 4159 removeUnwindEdge(PredBB); 4160 } else { 4161 Instruction *TI = PredBB->getTerminator(); 4162 TI->replaceUsesOfWith(BB, UnwindDest); 4163 } 4164 } 4165 4166 // The cleanup pad is now unreachable. Zap it. 4167 BB->eraseFromParent(); 4168 return true; 4169 } 4170 4171 // Try to merge two cleanuppads together. 4172 static bool mergeCleanupPad(CleanupReturnInst *RI) { 4173 // Skip any cleanuprets which unwind to caller, there is nothing to merge 4174 // with. 4175 BasicBlock *UnwindDest = RI->getUnwindDest(); 4176 if (!UnwindDest) 4177 return false; 4178 4179 // This cleanupret isn't the only predecessor of this cleanuppad, it wouldn't 4180 // be safe to merge without code duplication. 4181 if (UnwindDest->getSinglePredecessor() != RI->getParent()) 4182 return false; 4183 4184 // Verify that our cleanuppad's unwind destination is another cleanuppad. 4185 auto *SuccessorCleanupPad = dyn_cast<CleanupPadInst>(&UnwindDest->front()); 4186 if (!SuccessorCleanupPad) 4187 return false; 4188 4189 CleanupPadInst *PredecessorCleanupPad = RI->getCleanupPad(); 4190 // Replace any uses of the successor cleanupad with the predecessor pad 4191 // The only cleanuppad uses should be this cleanupret, it's cleanupret and 4192 // funclet bundle operands. 4193 SuccessorCleanupPad->replaceAllUsesWith(PredecessorCleanupPad); 4194 // Remove the old cleanuppad. 4195 SuccessorCleanupPad->eraseFromParent(); 4196 // Now, we simply replace the cleanupret with a branch to the unwind 4197 // destination. 4198 BranchInst::Create(UnwindDest, RI->getParent()); 4199 RI->eraseFromParent(); 4200 4201 return true; 4202 } 4203 4204 bool SimplifyCFGOpt::simplifyCleanupReturn(CleanupReturnInst *RI) { 4205 // It is possible to transiantly have an undef cleanuppad operand because we 4206 // have deleted some, but not all, dead blocks. 4207 // Eventually, this block will be deleted. 4208 if (isa<UndefValue>(RI->getOperand(0))) 4209 return false; 4210 4211 if (mergeCleanupPad(RI)) 4212 return true; 4213 4214 if (removeEmptyCleanup(RI)) 4215 return true; 4216 4217 return false; 4218 } 4219 4220 bool SimplifyCFGOpt::simplifyReturn(ReturnInst *RI, IRBuilder<> &Builder) { 4221 BasicBlock *BB = RI->getParent(); 4222 if (!BB->getFirstNonPHIOrDbg()->isTerminator()) 4223 return false; 4224 4225 // Find predecessors that end with branches. 4226 SmallVector<BasicBlock *, 8> UncondBranchPreds; 4227 SmallVector<BranchInst *, 8> CondBranchPreds; 4228 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) { 4229 BasicBlock *P = *PI; 4230 Instruction *PTI = P->getTerminator(); 4231 if (BranchInst *BI = dyn_cast<BranchInst>(PTI)) { 4232 if (BI->isUnconditional()) 4233 UncondBranchPreds.push_back(P); 4234 else 4235 CondBranchPreds.push_back(BI); 4236 } 4237 } 4238 4239 // If we found some, do the transformation! 4240 if (!UncondBranchPreds.empty() && DupRet) { 4241 while (!UncondBranchPreds.empty()) { 4242 BasicBlock *Pred = UncondBranchPreds.pop_back_val(); 4243 LLVM_DEBUG(dbgs() << "FOLDING: " << *BB 4244 << "INTO UNCOND BRANCH PRED: " << *Pred); 4245 (void)FoldReturnIntoUncondBranch(RI, BB, Pred); 4246 } 4247 4248 // If we eliminated all predecessors of the block, delete the block now. 4249 if (pred_empty(BB)) { 4250 // We know there are no successors, so just nuke the block. 4251 if (LoopHeaders) 4252 LoopHeaders->erase(BB); 4253 BB->eraseFromParent(); 4254 } 4255 4256 return true; 4257 } 4258 4259 // Check out all of the conditional branches going to this return 4260 // instruction. If any of them just select between returns, change the 4261 // branch itself into a select/return pair. 4262 while (!CondBranchPreds.empty()) { 4263 BranchInst *BI = CondBranchPreds.pop_back_val(); 4264 4265 // Check to see if the non-BB successor is also a return block. 4266 if (isa<ReturnInst>(BI->getSuccessor(0)->getTerminator()) && 4267 isa<ReturnInst>(BI->getSuccessor(1)->getTerminator()) && 4268 SimplifyCondBranchToTwoReturns(BI, Builder)) 4269 return true; 4270 } 4271 return false; 4272 } 4273 4274 bool SimplifyCFGOpt::simplifyUnreachable(UnreachableInst *UI) { 4275 BasicBlock *BB = UI->getParent(); 4276 4277 bool Changed = false; 4278 4279 // If there are any instructions immediately before the unreachable that can 4280 // be removed, do so. 4281 while (UI->getIterator() != BB->begin()) { 4282 BasicBlock::iterator BBI = UI->getIterator(); 4283 --BBI; 4284 // Do not delete instructions that can have side effects which might cause 4285 // the unreachable to not be reachable; specifically, calls and volatile 4286 // operations may have this effect. 4287 if (isa<CallInst>(BBI) && !isa<DbgInfoIntrinsic>(BBI)) 4288 break; 4289 4290 if (BBI->mayHaveSideEffects()) { 4291 if (auto *SI = dyn_cast<StoreInst>(BBI)) { 4292 if (SI->isVolatile()) 4293 break; 4294 } else if (auto *LI = dyn_cast<LoadInst>(BBI)) { 4295 if (LI->isVolatile()) 4296 break; 4297 } else if (auto *RMWI = dyn_cast<AtomicRMWInst>(BBI)) { 4298 if (RMWI->isVolatile()) 4299 break; 4300 } else if (auto *CXI = dyn_cast<AtomicCmpXchgInst>(BBI)) { 4301 if (CXI->isVolatile()) 4302 break; 4303 } else if (isa<CatchPadInst>(BBI)) { 4304 // A catchpad may invoke exception object constructors and such, which 4305 // in some languages can be arbitrary code, so be conservative by 4306 // default. 4307 // For CoreCLR, it just involves a type test, so can be removed. 4308 if (classifyEHPersonality(BB->getParent()->getPersonalityFn()) != 4309 EHPersonality::CoreCLR) 4310 break; 4311 } else if (!isa<FenceInst>(BBI) && !isa<VAArgInst>(BBI) && 4312 !isa<LandingPadInst>(BBI)) { 4313 break; 4314 } 4315 // Note that deleting LandingPad's here is in fact okay, although it 4316 // involves a bit of subtle reasoning. If this inst is a LandingPad, 4317 // all the predecessors of this block will be the unwind edges of Invokes, 4318 // and we can therefore guarantee this block will be erased. 4319 } 4320 4321 // Delete this instruction (any uses are guaranteed to be dead) 4322 if (!BBI->use_empty()) 4323 BBI->replaceAllUsesWith(UndefValue::get(BBI->getType())); 4324 BBI->eraseFromParent(); 4325 Changed = true; 4326 } 4327 4328 // If the unreachable instruction is the first in the block, take a gander 4329 // at all of the predecessors of this instruction, and simplify them. 4330 if (&BB->front() != UI) 4331 return Changed; 4332 4333 SmallVector<BasicBlock *, 8> Preds(pred_begin(BB), pred_end(BB)); 4334 for (unsigned i = 0, e = Preds.size(); i != e; ++i) { 4335 Instruction *TI = Preds[i]->getTerminator(); 4336 IRBuilder<> Builder(TI); 4337 if (auto *BI = dyn_cast<BranchInst>(TI)) { 4338 if (BI->isUnconditional()) { 4339 assert(BI->getSuccessor(0) == BB && "Incorrect CFG"); 4340 new UnreachableInst(TI->getContext(), TI); 4341 TI->eraseFromParent(); 4342 Changed = true; 4343 } else { 4344 Value* Cond = BI->getCondition(); 4345 if (BI->getSuccessor(0) == BB) { 4346 Builder.CreateAssumption(Builder.CreateNot(Cond)); 4347 Builder.CreateBr(BI->getSuccessor(1)); 4348 } else { 4349 assert(BI->getSuccessor(1) == BB && "Incorrect CFG"); 4350 Builder.CreateAssumption(Cond); 4351 Builder.CreateBr(BI->getSuccessor(0)); 4352 } 4353 EraseTerminatorAndDCECond(BI); 4354 Changed = true; 4355 } 4356 } else if (auto *SI = dyn_cast<SwitchInst>(TI)) { 4357 SwitchInstProfUpdateWrapper SU(*SI); 4358 for (auto i = SU->case_begin(), e = SU->case_end(); i != e;) { 4359 if (i->getCaseSuccessor() != BB) { 4360 ++i; 4361 continue; 4362 } 4363 BB->removePredecessor(SU->getParent()); 4364 i = SU.removeCase(i); 4365 e = SU->case_end(); 4366 Changed = true; 4367 } 4368 } else if (auto *II = dyn_cast<InvokeInst>(TI)) { 4369 if (II->getUnwindDest() == BB) { 4370 removeUnwindEdge(TI->getParent()); 4371 Changed = true; 4372 } 4373 } else if (auto *CSI = dyn_cast<CatchSwitchInst>(TI)) { 4374 if (CSI->getUnwindDest() == BB) { 4375 removeUnwindEdge(TI->getParent()); 4376 Changed = true; 4377 continue; 4378 } 4379 4380 for (CatchSwitchInst::handler_iterator I = CSI->handler_begin(), 4381 E = CSI->handler_end(); 4382 I != E; ++I) { 4383 if (*I == BB) { 4384 CSI->removeHandler(I); 4385 --I; 4386 --E; 4387 Changed = true; 4388 } 4389 } 4390 if (CSI->getNumHandlers() == 0) { 4391 BasicBlock *CatchSwitchBB = CSI->getParent(); 4392 if (CSI->hasUnwindDest()) { 4393 // Redirect preds to the unwind dest 4394 CatchSwitchBB->replaceAllUsesWith(CSI->getUnwindDest()); 4395 } else { 4396 // Rewrite all preds to unwind to caller (or from invoke to call). 4397 SmallVector<BasicBlock *, 8> EHPreds(predecessors(CatchSwitchBB)); 4398 for (BasicBlock *EHPred : EHPreds) 4399 removeUnwindEdge(EHPred); 4400 } 4401 // The catchswitch is no longer reachable. 4402 new UnreachableInst(CSI->getContext(), CSI); 4403 CSI->eraseFromParent(); 4404 Changed = true; 4405 } 4406 } else if (isa<CleanupReturnInst>(TI)) { 4407 new UnreachableInst(TI->getContext(), TI); 4408 TI->eraseFromParent(); 4409 Changed = true; 4410 } 4411 } 4412 4413 // If this block is now dead, remove it. 4414 if (pred_empty(BB) && BB != &BB->getParent()->getEntryBlock()) { 4415 // We know there are no successors, so just nuke the block. 4416 if (LoopHeaders) 4417 LoopHeaders->erase(BB); 4418 BB->eraseFromParent(); 4419 return true; 4420 } 4421 4422 return Changed; 4423 } 4424 4425 static bool CasesAreContiguous(SmallVectorImpl<ConstantInt *> &Cases) { 4426 assert(Cases.size() >= 1); 4427 4428 array_pod_sort(Cases.begin(), Cases.end(), ConstantIntSortPredicate); 4429 for (size_t I = 1, E = Cases.size(); I != E; ++I) { 4430 if (Cases[I - 1]->getValue() != Cases[I]->getValue() + 1) 4431 return false; 4432 } 4433 return true; 4434 } 4435 4436 static void createUnreachableSwitchDefault(SwitchInst *Switch) { 4437 LLVM_DEBUG(dbgs() << "SimplifyCFG: switch default is dead.\n"); 4438 BasicBlock *NewDefaultBlock = 4439 SplitBlockPredecessors(Switch->getDefaultDest(), Switch->getParent(), ""); 4440 Switch->setDefaultDest(&*NewDefaultBlock); 4441 SplitBlock(&*NewDefaultBlock, &NewDefaultBlock->front()); 4442 auto *NewTerminator = NewDefaultBlock->getTerminator(); 4443 new UnreachableInst(Switch->getContext(), NewTerminator); 4444 EraseTerminatorAndDCECond(NewTerminator); 4445 } 4446 4447 /// Turn a switch with two reachable destinations into an integer range 4448 /// comparison and branch. 4449 bool SimplifyCFGOpt::TurnSwitchRangeIntoICmp(SwitchInst *SI, 4450 IRBuilder<> &Builder) { 4451 assert(SI->getNumCases() > 1 && "Degenerate switch?"); 4452 4453 bool HasDefault = 4454 !isa<UnreachableInst>(SI->getDefaultDest()->getFirstNonPHIOrDbg()); 4455 4456 // Partition the cases into two sets with different destinations. 4457 BasicBlock *DestA = HasDefault ? SI->getDefaultDest() : nullptr; 4458 BasicBlock *DestB = nullptr; 4459 SmallVector<ConstantInt *, 16> CasesA; 4460 SmallVector<ConstantInt *, 16> CasesB; 4461 4462 for (auto Case : SI->cases()) { 4463 BasicBlock *Dest = Case.getCaseSuccessor(); 4464 if (!DestA) 4465 DestA = Dest; 4466 if (Dest == DestA) { 4467 CasesA.push_back(Case.getCaseValue()); 4468 continue; 4469 } 4470 if (!DestB) 4471 DestB = Dest; 4472 if (Dest == DestB) { 4473 CasesB.push_back(Case.getCaseValue()); 4474 continue; 4475 } 4476 return false; // More than two destinations. 4477 } 4478 4479 assert(DestA && DestB && 4480 "Single-destination switch should have been folded."); 4481 assert(DestA != DestB); 4482 assert(DestB != SI->getDefaultDest()); 4483 assert(!CasesB.empty() && "There must be non-default cases."); 4484 assert(!CasesA.empty() || HasDefault); 4485 4486 // Figure out if one of the sets of cases form a contiguous range. 4487 SmallVectorImpl<ConstantInt *> *ContiguousCases = nullptr; 4488 BasicBlock *ContiguousDest = nullptr; 4489 BasicBlock *OtherDest = nullptr; 4490 if (!CasesA.empty() && CasesAreContiguous(CasesA)) { 4491 ContiguousCases = &CasesA; 4492 ContiguousDest = DestA; 4493 OtherDest = DestB; 4494 } else if (CasesAreContiguous(CasesB)) { 4495 ContiguousCases = &CasesB; 4496 ContiguousDest = DestB; 4497 OtherDest = DestA; 4498 } else 4499 return false; 4500 4501 // Start building the compare and branch. 4502 4503 Constant *Offset = ConstantExpr::getNeg(ContiguousCases->back()); 4504 Constant *NumCases = 4505 ConstantInt::get(Offset->getType(), ContiguousCases->size()); 4506 4507 Value *Sub = SI->getCondition(); 4508 if (!Offset->isNullValue()) 4509 Sub = Builder.CreateAdd(Sub, Offset, Sub->getName() + ".off"); 4510 4511 Value *Cmp; 4512 // If NumCases overflowed, then all possible values jump to the successor. 4513 if (NumCases->isNullValue() && !ContiguousCases->empty()) 4514 Cmp = ConstantInt::getTrue(SI->getContext()); 4515 else 4516 Cmp = Builder.CreateICmpULT(Sub, NumCases, "switch"); 4517 BranchInst *NewBI = Builder.CreateCondBr(Cmp, ContiguousDest, OtherDest); 4518 4519 // Update weight for the newly-created conditional branch. 4520 if (HasBranchWeights(SI)) { 4521 SmallVector<uint64_t, 8> Weights; 4522 GetBranchWeights(SI, Weights); 4523 if (Weights.size() == 1 + SI->getNumCases()) { 4524 uint64_t TrueWeight = 0; 4525 uint64_t FalseWeight = 0; 4526 for (size_t I = 0, E = Weights.size(); I != E; ++I) { 4527 if (SI->getSuccessor(I) == ContiguousDest) 4528 TrueWeight += Weights[I]; 4529 else 4530 FalseWeight += Weights[I]; 4531 } 4532 while (TrueWeight > UINT32_MAX || FalseWeight > UINT32_MAX) { 4533 TrueWeight /= 2; 4534 FalseWeight /= 2; 4535 } 4536 setBranchWeights(NewBI, TrueWeight, FalseWeight); 4537 } 4538 } 4539 4540 // Prune obsolete incoming values off the successors' PHI nodes. 4541 for (auto BBI = ContiguousDest->begin(); isa<PHINode>(BBI); ++BBI) { 4542 unsigned PreviousEdges = ContiguousCases->size(); 4543 if (ContiguousDest == SI->getDefaultDest()) 4544 ++PreviousEdges; 4545 for (unsigned I = 0, E = PreviousEdges - 1; I != E; ++I) 4546 cast<PHINode>(BBI)->removeIncomingValue(SI->getParent()); 4547 } 4548 for (auto BBI = OtherDest->begin(); isa<PHINode>(BBI); ++BBI) { 4549 unsigned PreviousEdges = SI->getNumCases() - ContiguousCases->size(); 4550 if (OtherDest == SI->getDefaultDest()) 4551 ++PreviousEdges; 4552 for (unsigned I = 0, E = PreviousEdges - 1; I != E; ++I) 4553 cast<PHINode>(BBI)->removeIncomingValue(SI->getParent()); 4554 } 4555 4556 // Clean up the default block - it may have phis or other instructions before 4557 // the unreachable terminator. 4558 if (!HasDefault) 4559 createUnreachableSwitchDefault(SI); 4560 4561 // Drop the switch. 4562 SI->eraseFromParent(); 4563 4564 return true; 4565 } 4566 4567 /// Compute masked bits for the condition of a switch 4568 /// and use it to remove dead cases. 4569 static bool eliminateDeadSwitchCases(SwitchInst *SI, AssumptionCache *AC, 4570 const DataLayout &DL) { 4571 Value *Cond = SI->getCondition(); 4572 unsigned Bits = Cond->getType()->getIntegerBitWidth(); 4573 KnownBits Known = computeKnownBits(Cond, DL, 0, AC, SI); 4574 4575 // We can also eliminate cases by determining that their values are outside of 4576 // the limited range of the condition based on how many significant (non-sign) 4577 // bits are in the condition value. 4578 unsigned ExtraSignBits = ComputeNumSignBits(Cond, DL, 0, AC, SI) - 1; 4579 unsigned MaxSignificantBitsInCond = Bits - ExtraSignBits; 4580 4581 // Gather dead cases. 4582 SmallVector<ConstantInt *, 8> DeadCases; 4583 for (auto &Case : SI->cases()) { 4584 const APInt &CaseVal = Case.getCaseValue()->getValue(); 4585 if (Known.Zero.intersects(CaseVal) || !Known.One.isSubsetOf(CaseVal) || 4586 (CaseVal.getMinSignedBits() > MaxSignificantBitsInCond)) { 4587 DeadCases.push_back(Case.getCaseValue()); 4588 LLVM_DEBUG(dbgs() << "SimplifyCFG: switch case " << CaseVal 4589 << " is dead.\n"); 4590 } 4591 } 4592 4593 // If we can prove that the cases must cover all possible values, the 4594 // default destination becomes dead and we can remove it. If we know some 4595 // of the bits in the value, we can use that to more precisely compute the 4596 // number of possible unique case values. 4597 bool HasDefault = 4598 !isa<UnreachableInst>(SI->getDefaultDest()->getFirstNonPHIOrDbg()); 4599 const unsigned NumUnknownBits = 4600 Bits - (Known.Zero | Known.One).countPopulation(); 4601 assert(NumUnknownBits <= Bits); 4602 if (HasDefault && DeadCases.empty() && 4603 NumUnknownBits < 64 /* avoid overflow */ && 4604 SI->getNumCases() == (1ULL << NumUnknownBits)) { 4605 createUnreachableSwitchDefault(SI); 4606 return true; 4607 } 4608 4609 if (DeadCases.empty()) 4610 return false; 4611 4612 SwitchInstProfUpdateWrapper SIW(*SI); 4613 for (ConstantInt *DeadCase : DeadCases) { 4614 SwitchInst::CaseIt CaseI = SI->findCaseValue(DeadCase); 4615 assert(CaseI != SI->case_default() && 4616 "Case was not found. Probably mistake in DeadCases forming."); 4617 // Prune unused values from PHI nodes. 4618 CaseI->getCaseSuccessor()->removePredecessor(SI->getParent()); 4619 SIW.removeCase(CaseI); 4620 } 4621 4622 return true; 4623 } 4624 4625 /// If BB would be eligible for simplification by 4626 /// TryToSimplifyUncondBranchFromEmptyBlock (i.e. it is empty and terminated 4627 /// by an unconditional branch), look at the phi node for BB in the successor 4628 /// block and see if the incoming value is equal to CaseValue. If so, return 4629 /// the phi node, and set PhiIndex to BB's index in the phi node. 4630 static PHINode *FindPHIForConditionForwarding(ConstantInt *CaseValue, 4631 BasicBlock *BB, int *PhiIndex) { 4632 if (BB->getFirstNonPHIOrDbg() != BB->getTerminator()) 4633 return nullptr; // BB must be empty to be a candidate for simplification. 4634 if (!BB->getSinglePredecessor()) 4635 return nullptr; // BB must be dominated by the switch. 4636 4637 BranchInst *Branch = dyn_cast<BranchInst>(BB->getTerminator()); 4638 if (!Branch || !Branch->isUnconditional()) 4639 return nullptr; // Terminator must be unconditional branch. 4640 4641 BasicBlock *Succ = Branch->getSuccessor(0); 4642 4643 for (PHINode &PHI : Succ->phis()) { 4644 int Idx = PHI.getBasicBlockIndex(BB); 4645 assert(Idx >= 0 && "PHI has no entry for predecessor?"); 4646 4647 Value *InValue = PHI.getIncomingValue(Idx); 4648 if (InValue != CaseValue) 4649 continue; 4650 4651 *PhiIndex = Idx; 4652 return &PHI; 4653 } 4654 4655 return nullptr; 4656 } 4657 4658 /// Try to forward the condition of a switch instruction to a phi node 4659 /// dominated by the switch, if that would mean that some of the destination 4660 /// blocks of the switch can be folded away. Return true if a change is made. 4661 static bool ForwardSwitchConditionToPHI(SwitchInst *SI) { 4662 using ForwardingNodesMap = DenseMap<PHINode *, SmallVector<int, 4>>; 4663 4664 ForwardingNodesMap ForwardingNodes; 4665 BasicBlock *SwitchBlock = SI->getParent(); 4666 bool Changed = false; 4667 for (auto &Case : SI->cases()) { 4668 ConstantInt *CaseValue = Case.getCaseValue(); 4669 BasicBlock *CaseDest = Case.getCaseSuccessor(); 4670 4671 // Replace phi operands in successor blocks that are using the constant case 4672 // value rather than the switch condition variable: 4673 // switchbb: 4674 // switch i32 %x, label %default [ 4675 // i32 17, label %succ 4676 // ... 4677 // succ: 4678 // %r = phi i32 ... [ 17, %switchbb ] ... 4679 // --> 4680 // %r = phi i32 ... [ %x, %switchbb ] ... 4681 4682 for (PHINode &Phi : CaseDest->phis()) { 4683 // This only works if there is exactly 1 incoming edge from the switch to 4684 // a phi. If there is >1, that means multiple cases of the switch map to 1 4685 // value in the phi, and that phi value is not the switch condition. Thus, 4686 // this transform would not make sense (the phi would be invalid because 4687 // a phi can't have different incoming values from the same block). 4688 int SwitchBBIdx = Phi.getBasicBlockIndex(SwitchBlock); 4689 if (Phi.getIncomingValue(SwitchBBIdx) == CaseValue && 4690 count(Phi.blocks(), SwitchBlock) == 1) { 4691 Phi.setIncomingValue(SwitchBBIdx, SI->getCondition()); 4692 Changed = true; 4693 } 4694 } 4695 4696 // Collect phi nodes that are indirectly using this switch's case constants. 4697 int PhiIdx; 4698 if (auto *Phi = FindPHIForConditionForwarding(CaseValue, CaseDest, &PhiIdx)) 4699 ForwardingNodes[Phi].push_back(PhiIdx); 4700 } 4701 4702 for (auto &ForwardingNode : ForwardingNodes) { 4703 PHINode *Phi = ForwardingNode.first; 4704 SmallVectorImpl<int> &Indexes = ForwardingNode.second; 4705 if (Indexes.size() < 2) 4706 continue; 4707 4708 for (int Index : Indexes) 4709 Phi->setIncomingValue(Index, SI->getCondition()); 4710 Changed = true; 4711 } 4712 4713 return Changed; 4714 } 4715 4716 /// Return true if the backend will be able to handle 4717 /// initializing an array of constants like C. 4718 static bool ValidLookupTableConstant(Constant *C, const TargetTransformInfo &TTI) { 4719 if (C->isThreadDependent()) 4720 return false; 4721 if (C->isDLLImportDependent()) 4722 return false; 4723 4724 if (!isa<ConstantFP>(C) && !isa<ConstantInt>(C) && 4725 !isa<ConstantPointerNull>(C) && !isa<GlobalValue>(C) && 4726 !isa<UndefValue>(C) && !isa<ConstantExpr>(C)) 4727 return false; 4728 4729 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C)) { 4730 if (!CE->isGEPWithNoNotionalOverIndexing()) 4731 return false; 4732 if (!ValidLookupTableConstant(CE->getOperand(0), TTI)) 4733 return false; 4734 } 4735 4736 if (!TTI.shouldBuildLookupTablesForConstant(C)) 4737 return false; 4738 4739 return true; 4740 } 4741 4742 /// If V is a Constant, return it. Otherwise, try to look up 4743 /// its constant value in ConstantPool, returning 0 if it's not there. 4744 static Constant * 4745 LookupConstant(Value *V, 4746 const SmallDenseMap<Value *, Constant *> &ConstantPool) { 4747 if (Constant *C = dyn_cast<Constant>(V)) 4748 return C; 4749 return ConstantPool.lookup(V); 4750 } 4751 4752 /// Try to fold instruction I into a constant. This works for 4753 /// simple instructions such as binary operations where both operands are 4754 /// constant or can be replaced by constants from the ConstantPool. Returns the 4755 /// resulting constant on success, 0 otherwise. 4756 static Constant * 4757 ConstantFold(Instruction *I, const DataLayout &DL, 4758 const SmallDenseMap<Value *, Constant *> &ConstantPool) { 4759 if (SelectInst *Select = dyn_cast<SelectInst>(I)) { 4760 Constant *A = LookupConstant(Select->getCondition(), ConstantPool); 4761 if (!A) 4762 return nullptr; 4763 if (A->isAllOnesValue()) 4764 return LookupConstant(Select->getTrueValue(), ConstantPool); 4765 if (A->isNullValue()) 4766 return LookupConstant(Select->getFalseValue(), ConstantPool); 4767 return nullptr; 4768 } 4769 4770 SmallVector<Constant *, 4> COps; 4771 for (unsigned N = 0, E = I->getNumOperands(); N != E; ++N) { 4772 if (Constant *A = LookupConstant(I->getOperand(N), ConstantPool)) 4773 COps.push_back(A); 4774 else 4775 return nullptr; 4776 } 4777 4778 if (CmpInst *Cmp = dyn_cast<CmpInst>(I)) { 4779 return ConstantFoldCompareInstOperands(Cmp->getPredicate(), COps[0], 4780 COps[1], DL); 4781 } 4782 4783 return ConstantFoldInstOperands(I, COps, DL); 4784 } 4785 4786 /// Try to determine the resulting constant values in phi nodes 4787 /// at the common destination basic block, *CommonDest, for one of the case 4788 /// destionations CaseDest corresponding to value CaseVal (0 for the default 4789 /// case), of a switch instruction SI. 4790 static bool 4791 GetCaseResults(SwitchInst *SI, ConstantInt *CaseVal, BasicBlock *CaseDest, 4792 BasicBlock **CommonDest, 4793 SmallVectorImpl<std::pair<PHINode *, Constant *>> &Res, 4794 const DataLayout &DL, const TargetTransformInfo &TTI) { 4795 // The block from which we enter the common destination. 4796 BasicBlock *Pred = SI->getParent(); 4797 4798 // If CaseDest is empty except for some side-effect free instructions through 4799 // which we can constant-propagate the CaseVal, continue to its successor. 4800 SmallDenseMap<Value *, Constant *> ConstantPool; 4801 ConstantPool.insert(std::make_pair(SI->getCondition(), CaseVal)); 4802 for (Instruction &I :CaseDest->instructionsWithoutDebug()) { 4803 if (I.isTerminator()) { 4804 // If the terminator is a simple branch, continue to the next block. 4805 if (I.getNumSuccessors() != 1 || I.isExceptionalTerminator()) 4806 return false; 4807 Pred = CaseDest; 4808 CaseDest = I.getSuccessor(0); 4809 } else if (Constant *C = ConstantFold(&I, DL, ConstantPool)) { 4810 // Instruction is side-effect free and constant. 4811 4812 // If the instruction has uses outside this block or a phi node slot for 4813 // the block, it is not safe to bypass the instruction since it would then 4814 // no longer dominate all its uses. 4815 for (auto &Use : I.uses()) { 4816 User *User = Use.getUser(); 4817 if (Instruction *I = dyn_cast<Instruction>(User)) 4818 if (I->getParent() == CaseDest) 4819 continue; 4820 if (PHINode *Phi = dyn_cast<PHINode>(User)) 4821 if (Phi->getIncomingBlock(Use) == CaseDest) 4822 continue; 4823 return false; 4824 } 4825 4826 ConstantPool.insert(std::make_pair(&I, C)); 4827 } else { 4828 break; 4829 } 4830 } 4831 4832 // If we did not have a CommonDest before, use the current one. 4833 if (!*CommonDest) 4834 *CommonDest = CaseDest; 4835 // If the destination isn't the common one, abort. 4836 if (CaseDest != *CommonDest) 4837 return false; 4838 4839 // Get the values for this case from phi nodes in the destination block. 4840 for (PHINode &PHI : (*CommonDest)->phis()) { 4841 int Idx = PHI.getBasicBlockIndex(Pred); 4842 if (Idx == -1) 4843 continue; 4844 4845 Constant *ConstVal = 4846 LookupConstant(PHI.getIncomingValue(Idx), ConstantPool); 4847 if (!ConstVal) 4848 return false; 4849 4850 // Be conservative about which kinds of constants we support. 4851 if (!ValidLookupTableConstant(ConstVal, TTI)) 4852 return false; 4853 4854 Res.push_back(std::make_pair(&PHI, ConstVal)); 4855 } 4856 4857 return Res.size() > 0; 4858 } 4859 4860 // Helper function used to add CaseVal to the list of cases that generate 4861 // Result. Returns the updated number of cases that generate this result. 4862 static uintptr_t MapCaseToResult(ConstantInt *CaseVal, 4863 SwitchCaseResultVectorTy &UniqueResults, 4864 Constant *Result) { 4865 for (auto &I : UniqueResults) { 4866 if (I.first == Result) { 4867 I.second.push_back(CaseVal); 4868 return I.second.size(); 4869 } 4870 } 4871 UniqueResults.push_back( 4872 std::make_pair(Result, SmallVector<ConstantInt *, 4>(1, CaseVal))); 4873 return 1; 4874 } 4875 4876 // Helper function that initializes a map containing 4877 // results for the PHI node of the common destination block for a switch 4878 // instruction. Returns false if multiple PHI nodes have been found or if 4879 // there is not a common destination block for the switch. 4880 static bool 4881 InitializeUniqueCases(SwitchInst *SI, PHINode *&PHI, BasicBlock *&CommonDest, 4882 SwitchCaseResultVectorTy &UniqueResults, 4883 Constant *&DefaultResult, const DataLayout &DL, 4884 const TargetTransformInfo &TTI, 4885 uintptr_t MaxUniqueResults, uintptr_t MaxCasesPerResult) { 4886 for (auto &I : SI->cases()) { 4887 ConstantInt *CaseVal = I.getCaseValue(); 4888 4889 // Resulting value at phi nodes for this case value. 4890 SwitchCaseResultsTy Results; 4891 if (!GetCaseResults(SI, CaseVal, I.getCaseSuccessor(), &CommonDest, Results, 4892 DL, TTI)) 4893 return false; 4894 4895 // Only one value per case is permitted. 4896 if (Results.size() > 1) 4897 return false; 4898 4899 // Add the case->result mapping to UniqueResults. 4900 const uintptr_t NumCasesForResult = 4901 MapCaseToResult(CaseVal, UniqueResults, Results.begin()->second); 4902 4903 // Early out if there are too many cases for this result. 4904 if (NumCasesForResult > MaxCasesPerResult) 4905 return false; 4906 4907 // Early out if there are too many unique results. 4908 if (UniqueResults.size() > MaxUniqueResults) 4909 return false; 4910 4911 // Check the PHI consistency. 4912 if (!PHI) 4913 PHI = Results[0].first; 4914 else if (PHI != Results[0].first) 4915 return false; 4916 } 4917 // Find the default result value. 4918 SmallVector<std::pair<PHINode *, Constant *>, 1> DefaultResults; 4919 BasicBlock *DefaultDest = SI->getDefaultDest(); 4920 GetCaseResults(SI, nullptr, SI->getDefaultDest(), &CommonDest, DefaultResults, 4921 DL, TTI); 4922 // If the default value is not found abort unless the default destination 4923 // is unreachable. 4924 DefaultResult = 4925 DefaultResults.size() == 1 ? DefaultResults.begin()->second : nullptr; 4926 if ((!DefaultResult && 4927 !isa<UnreachableInst>(DefaultDest->getFirstNonPHIOrDbg()))) 4928 return false; 4929 4930 return true; 4931 } 4932 4933 // Helper function that checks if it is possible to transform a switch with only 4934 // two cases (or two cases + default) that produces a result into a select. 4935 // Example: 4936 // switch (a) { 4937 // case 10: %0 = icmp eq i32 %a, 10 4938 // return 10; %1 = select i1 %0, i32 10, i32 4 4939 // case 20: ----> %2 = icmp eq i32 %a, 20 4940 // return 2; %3 = select i1 %2, i32 2, i32 %1 4941 // default: 4942 // return 4; 4943 // } 4944 static Value *ConvertTwoCaseSwitch(const SwitchCaseResultVectorTy &ResultVector, 4945 Constant *DefaultResult, Value *Condition, 4946 IRBuilder<> &Builder) { 4947 assert(ResultVector.size() == 2 && 4948 "We should have exactly two unique results at this point"); 4949 // If we are selecting between only two cases transform into a simple 4950 // select or a two-way select if default is possible. 4951 if (ResultVector[0].second.size() == 1 && 4952 ResultVector[1].second.size() == 1) { 4953 ConstantInt *const FirstCase = ResultVector[0].second[0]; 4954 ConstantInt *const SecondCase = ResultVector[1].second[0]; 4955 4956 bool DefaultCanTrigger = DefaultResult; 4957 Value *SelectValue = ResultVector[1].first; 4958 if (DefaultCanTrigger) { 4959 Value *const ValueCompare = 4960 Builder.CreateICmpEQ(Condition, SecondCase, "switch.selectcmp"); 4961 SelectValue = Builder.CreateSelect(ValueCompare, ResultVector[1].first, 4962 DefaultResult, "switch.select"); 4963 } 4964 Value *const ValueCompare = 4965 Builder.CreateICmpEQ(Condition, FirstCase, "switch.selectcmp"); 4966 return Builder.CreateSelect(ValueCompare, ResultVector[0].first, 4967 SelectValue, "switch.select"); 4968 } 4969 4970 return nullptr; 4971 } 4972 4973 // Helper function to cleanup a switch instruction that has been converted into 4974 // a select, fixing up PHI nodes and basic blocks. 4975 static void RemoveSwitchAfterSelectConversion(SwitchInst *SI, PHINode *PHI, 4976 Value *SelectValue, 4977 IRBuilder<> &Builder) { 4978 BasicBlock *SelectBB = SI->getParent(); 4979 while (PHI->getBasicBlockIndex(SelectBB) >= 0) 4980 PHI->removeIncomingValue(SelectBB); 4981 PHI->addIncoming(SelectValue, SelectBB); 4982 4983 Builder.CreateBr(PHI->getParent()); 4984 4985 // Remove the switch. 4986 for (unsigned i = 0, e = SI->getNumSuccessors(); i < e; ++i) { 4987 BasicBlock *Succ = SI->getSuccessor(i); 4988 4989 if (Succ == PHI->getParent()) 4990 continue; 4991 Succ->removePredecessor(SelectBB); 4992 } 4993 SI->eraseFromParent(); 4994 } 4995 4996 /// If the switch is only used to initialize one or more 4997 /// phi nodes in a common successor block with only two different 4998 /// constant values, replace the switch with select. 4999 static bool switchToSelect(SwitchInst *SI, IRBuilder<> &Builder, 5000 const DataLayout &DL, 5001 const TargetTransformInfo &TTI) { 5002 Value *const Cond = SI->getCondition(); 5003 PHINode *PHI = nullptr; 5004 BasicBlock *CommonDest = nullptr; 5005 Constant *DefaultResult; 5006 SwitchCaseResultVectorTy UniqueResults; 5007 // Collect all the cases that will deliver the same value from the switch. 5008 if (!InitializeUniqueCases(SI, PHI, CommonDest, UniqueResults, DefaultResult, 5009 DL, TTI, 2, 1)) 5010 return false; 5011 // Selects choose between maximum two values. 5012 if (UniqueResults.size() != 2) 5013 return false; 5014 assert(PHI != nullptr && "PHI for value select not found"); 5015 5016 Builder.SetInsertPoint(SI); 5017 Value *SelectValue = 5018 ConvertTwoCaseSwitch(UniqueResults, DefaultResult, Cond, Builder); 5019 if (SelectValue) { 5020 RemoveSwitchAfterSelectConversion(SI, PHI, SelectValue, Builder); 5021 return true; 5022 } 5023 // The switch couldn't be converted into a select. 5024 return false; 5025 } 5026 5027 namespace { 5028 5029 /// This class represents a lookup table that can be used to replace a switch. 5030 class SwitchLookupTable { 5031 public: 5032 /// Create a lookup table to use as a switch replacement with the contents 5033 /// of Values, using DefaultValue to fill any holes in the table. 5034 SwitchLookupTable( 5035 Module &M, uint64_t TableSize, ConstantInt *Offset, 5036 const SmallVectorImpl<std::pair<ConstantInt *, Constant *>> &Values, 5037 Constant *DefaultValue, const DataLayout &DL, const StringRef &FuncName); 5038 5039 /// Build instructions with Builder to retrieve the value at 5040 /// the position given by Index in the lookup table. 5041 Value *BuildLookup(Value *Index, IRBuilder<> &Builder); 5042 5043 /// Return true if a table with TableSize elements of 5044 /// type ElementType would fit in a target-legal register. 5045 static bool WouldFitInRegister(const DataLayout &DL, uint64_t TableSize, 5046 Type *ElementType); 5047 5048 private: 5049 // Depending on the contents of the table, it can be represented in 5050 // different ways. 5051 enum { 5052 // For tables where each element contains the same value, we just have to 5053 // store that single value and return it for each lookup. 5054 SingleValueKind, 5055 5056 // For tables where there is a linear relationship between table index 5057 // and values. We calculate the result with a simple multiplication 5058 // and addition instead of a table lookup. 5059 LinearMapKind, 5060 5061 // For small tables with integer elements, we can pack them into a bitmap 5062 // that fits into a target-legal register. Values are retrieved by 5063 // shift and mask operations. 5064 BitMapKind, 5065 5066 // The table is stored as an array of values. Values are retrieved by load 5067 // instructions from the table. 5068 ArrayKind 5069 } Kind; 5070 5071 // For SingleValueKind, this is the single value. 5072 Constant *SingleValue = nullptr; 5073 5074 // For BitMapKind, this is the bitmap. 5075 ConstantInt *BitMap = nullptr; 5076 IntegerType *BitMapElementTy = nullptr; 5077 5078 // For LinearMapKind, these are the constants used to derive the value. 5079 ConstantInt *LinearOffset = nullptr; 5080 ConstantInt *LinearMultiplier = nullptr; 5081 5082 // For ArrayKind, this is the array. 5083 GlobalVariable *Array = nullptr; 5084 }; 5085 5086 } // end anonymous namespace 5087 5088 SwitchLookupTable::SwitchLookupTable( 5089 Module &M, uint64_t TableSize, ConstantInt *Offset, 5090 const SmallVectorImpl<std::pair<ConstantInt *, Constant *>> &Values, 5091 Constant *DefaultValue, const DataLayout &DL, const StringRef &FuncName) { 5092 assert(Values.size() && "Can't build lookup table without values!"); 5093 assert(TableSize >= Values.size() && "Can't fit values in table!"); 5094 5095 // If all values in the table are equal, this is that value. 5096 SingleValue = Values.begin()->second; 5097 5098 Type *ValueType = Values.begin()->second->getType(); 5099 5100 // Build up the table contents. 5101 SmallVector<Constant *, 64> TableContents(TableSize); 5102 for (size_t I = 0, E = Values.size(); I != E; ++I) { 5103 ConstantInt *CaseVal = Values[I].first; 5104 Constant *CaseRes = Values[I].second; 5105 assert(CaseRes->getType() == ValueType); 5106 5107 uint64_t Idx = (CaseVal->getValue() - Offset->getValue()).getLimitedValue(); 5108 TableContents[Idx] = CaseRes; 5109 5110 if (CaseRes != SingleValue) 5111 SingleValue = nullptr; 5112 } 5113 5114 // Fill in any holes in the table with the default result. 5115 if (Values.size() < TableSize) { 5116 assert(DefaultValue && 5117 "Need a default value to fill the lookup table holes."); 5118 assert(DefaultValue->getType() == ValueType); 5119 for (uint64_t I = 0; I < TableSize; ++I) { 5120 if (!TableContents[I]) 5121 TableContents[I] = DefaultValue; 5122 } 5123 5124 if (DefaultValue != SingleValue) 5125 SingleValue = nullptr; 5126 } 5127 5128 // If each element in the table contains the same value, we only need to store 5129 // that single value. 5130 if (SingleValue) { 5131 Kind = SingleValueKind; 5132 return; 5133 } 5134 5135 // Check if we can derive the value with a linear transformation from the 5136 // table index. 5137 if (isa<IntegerType>(ValueType)) { 5138 bool LinearMappingPossible = true; 5139 APInt PrevVal; 5140 APInt DistToPrev; 5141 assert(TableSize >= 2 && "Should be a SingleValue table."); 5142 // Check if there is the same distance between two consecutive values. 5143 for (uint64_t I = 0; I < TableSize; ++I) { 5144 ConstantInt *ConstVal = dyn_cast<ConstantInt>(TableContents[I]); 5145 if (!ConstVal) { 5146 // This is an undef. We could deal with it, but undefs in lookup tables 5147 // are very seldom. It's probably not worth the additional complexity. 5148 LinearMappingPossible = false; 5149 break; 5150 } 5151 const APInt &Val = ConstVal->getValue(); 5152 if (I != 0) { 5153 APInt Dist = Val - PrevVal; 5154 if (I == 1) { 5155 DistToPrev = Dist; 5156 } else if (Dist != DistToPrev) { 5157 LinearMappingPossible = false; 5158 break; 5159 } 5160 } 5161 PrevVal = Val; 5162 } 5163 if (LinearMappingPossible) { 5164 LinearOffset = cast<ConstantInt>(TableContents[0]); 5165 LinearMultiplier = ConstantInt::get(M.getContext(), DistToPrev); 5166 Kind = LinearMapKind; 5167 ++NumLinearMaps; 5168 return; 5169 } 5170 } 5171 5172 // If the type is integer and the table fits in a register, build a bitmap. 5173 if (WouldFitInRegister(DL, TableSize, ValueType)) { 5174 IntegerType *IT = cast<IntegerType>(ValueType); 5175 APInt TableInt(TableSize * IT->getBitWidth(), 0); 5176 for (uint64_t I = TableSize; I > 0; --I) { 5177 TableInt <<= IT->getBitWidth(); 5178 // Insert values into the bitmap. Undef values are set to zero. 5179 if (!isa<UndefValue>(TableContents[I - 1])) { 5180 ConstantInt *Val = cast<ConstantInt>(TableContents[I - 1]); 5181 TableInt |= Val->getValue().zext(TableInt.getBitWidth()); 5182 } 5183 } 5184 BitMap = ConstantInt::get(M.getContext(), TableInt); 5185 BitMapElementTy = IT; 5186 Kind = BitMapKind; 5187 ++NumBitMaps; 5188 return; 5189 } 5190 5191 // Store the table in an array. 5192 ArrayType *ArrayTy = ArrayType::get(ValueType, TableSize); 5193 Constant *Initializer = ConstantArray::get(ArrayTy, TableContents); 5194 5195 Array = new GlobalVariable(M, ArrayTy, /*isConstant=*/true, 5196 GlobalVariable::PrivateLinkage, Initializer, 5197 "switch.table." + FuncName); 5198 Array->setUnnamedAddr(GlobalValue::UnnamedAddr::Global); 5199 // Set the alignment to that of an array items. We will be only loading one 5200 // value out of it. 5201 Array->setAlignment(Align(DL.getPrefTypeAlignment(ValueType))); 5202 Kind = ArrayKind; 5203 } 5204 5205 Value *SwitchLookupTable::BuildLookup(Value *Index, IRBuilder<> &Builder) { 5206 switch (Kind) { 5207 case SingleValueKind: 5208 return SingleValue; 5209 case LinearMapKind: { 5210 // Derive the result value from the input value. 5211 Value *Result = Builder.CreateIntCast(Index, LinearMultiplier->getType(), 5212 false, "switch.idx.cast"); 5213 if (!LinearMultiplier->isOne()) 5214 Result = Builder.CreateMul(Result, LinearMultiplier, "switch.idx.mult"); 5215 if (!LinearOffset->isZero()) 5216 Result = Builder.CreateAdd(Result, LinearOffset, "switch.offset"); 5217 return Result; 5218 } 5219 case BitMapKind: { 5220 // Type of the bitmap (e.g. i59). 5221 IntegerType *MapTy = BitMap->getType(); 5222 5223 // Cast Index to the same type as the bitmap. 5224 // Note: The Index is <= the number of elements in the table, so 5225 // truncating it to the width of the bitmask is safe. 5226 Value *ShiftAmt = Builder.CreateZExtOrTrunc(Index, MapTy, "switch.cast"); 5227 5228 // Multiply the shift amount by the element width. 5229 ShiftAmt = Builder.CreateMul( 5230 ShiftAmt, ConstantInt::get(MapTy, BitMapElementTy->getBitWidth()), 5231 "switch.shiftamt"); 5232 5233 // Shift down. 5234 Value *DownShifted = 5235 Builder.CreateLShr(BitMap, ShiftAmt, "switch.downshift"); 5236 // Mask off. 5237 return Builder.CreateTrunc(DownShifted, BitMapElementTy, "switch.masked"); 5238 } 5239 case ArrayKind: { 5240 // Make sure the table index will not overflow when treated as signed. 5241 IntegerType *IT = cast<IntegerType>(Index->getType()); 5242 uint64_t TableSize = 5243 Array->getInitializer()->getType()->getArrayNumElements(); 5244 if (TableSize > (1ULL << (IT->getBitWidth() - 1))) 5245 Index = Builder.CreateZExt( 5246 Index, IntegerType::get(IT->getContext(), IT->getBitWidth() + 1), 5247 "switch.tableidx.zext"); 5248 5249 Value *GEPIndices[] = {Builder.getInt32(0), Index}; 5250 Value *GEP = Builder.CreateInBoundsGEP(Array->getValueType(), Array, 5251 GEPIndices, "switch.gep"); 5252 return Builder.CreateLoad( 5253 cast<ArrayType>(Array->getValueType())->getElementType(), GEP, 5254 "switch.load"); 5255 } 5256 } 5257 llvm_unreachable("Unknown lookup table kind!"); 5258 } 5259 5260 bool SwitchLookupTable::WouldFitInRegister(const DataLayout &DL, 5261 uint64_t TableSize, 5262 Type *ElementType) { 5263 auto *IT = dyn_cast<IntegerType>(ElementType); 5264 if (!IT) 5265 return false; 5266 // FIXME: If the type is wider than it needs to be, e.g. i8 but all values 5267 // are <= 15, we could try to narrow the type. 5268 5269 // Avoid overflow, fitsInLegalInteger uses unsigned int for the width. 5270 if (TableSize >= UINT_MAX / IT->getBitWidth()) 5271 return false; 5272 return DL.fitsInLegalInteger(TableSize * IT->getBitWidth()); 5273 } 5274 5275 /// Determine whether a lookup table should be built for this switch, based on 5276 /// the number of cases, size of the table, and the types of the results. 5277 static bool 5278 ShouldBuildLookupTable(SwitchInst *SI, uint64_t TableSize, 5279 const TargetTransformInfo &TTI, const DataLayout &DL, 5280 const SmallDenseMap<PHINode *, Type *> &ResultTypes) { 5281 if (SI->getNumCases() > TableSize || TableSize >= UINT64_MAX / 10) 5282 return false; // TableSize overflowed, or mul below might overflow. 5283 5284 bool AllTablesFitInRegister = true; 5285 bool HasIllegalType = false; 5286 for (const auto &I : ResultTypes) { 5287 Type *Ty = I.second; 5288 5289 // Saturate this flag to true. 5290 HasIllegalType = HasIllegalType || !TTI.isTypeLegal(Ty); 5291 5292 // Saturate this flag to false. 5293 AllTablesFitInRegister = 5294 AllTablesFitInRegister && 5295 SwitchLookupTable::WouldFitInRegister(DL, TableSize, Ty); 5296 5297 // If both flags saturate, we're done. NOTE: This *only* works with 5298 // saturating flags, and all flags have to saturate first due to the 5299 // non-deterministic behavior of iterating over a dense map. 5300 if (HasIllegalType && !AllTablesFitInRegister) 5301 break; 5302 } 5303 5304 // If each table would fit in a register, we should build it anyway. 5305 if (AllTablesFitInRegister) 5306 return true; 5307 5308 // Don't build a table that doesn't fit in-register if it has illegal types. 5309 if (HasIllegalType) 5310 return false; 5311 5312 // The table density should be at least 40%. This is the same criterion as for 5313 // jump tables, see SelectionDAGBuilder::handleJTSwitchCase. 5314 // FIXME: Find the best cut-off. 5315 return SI->getNumCases() * 10 >= TableSize * 4; 5316 } 5317 5318 /// Try to reuse the switch table index compare. Following pattern: 5319 /// \code 5320 /// if (idx < tablesize) 5321 /// r = table[idx]; // table does not contain default_value 5322 /// else 5323 /// r = default_value; 5324 /// if (r != default_value) 5325 /// ... 5326 /// \endcode 5327 /// Is optimized to: 5328 /// \code 5329 /// cond = idx < tablesize; 5330 /// if (cond) 5331 /// r = table[idx]; 5332 /// else 5333 /// r = default_value; 5334 /// if (cond) 5335 /// ... 5336 /// \endcode 5337 /// Jump threading will then eliminate the second if(cond). 5338 static void reuseTableCompare( 5339 User *PhiUser, BasicBlock *PhiBlock, BranchInst *RangeCheckBranch, 5340 Constant *DefaultValue, 5341 const SmallVectorImpl<std::pair<ConstantInt *, Constant *>> &Values) { 5342 ICmpInst *CmpInst = dyn_cast<ICmpInst>(PhiUser); 5343 if (!CmpInst) 5344 return; 5345 5346 // We require that the compare is in the same block as the phi so that jump 5347 // threading can do its work afterwards. 5348 if (CmpInst->getParent() != PhiBlock) 5349 return; 5350 5351 Constant *CmpOp1 = dyn_cast<Constant>(CmpInst->getOperand(1)); 5352 if (!CmpOp1) 5353 return; 5354 5355 Value *RangeCmp = RangeCheckBranch->getCondition(); 5356 Constant *TrueConst = ConstantInt::getTrue(RangeCmp->getType()); 5357 Constant *FalseConst = ConstantInt::getFalse(RangeCmp->getType()); 5358 5359 // Check if the compare with the default value is constant true or false. 5360 Constant *DefaultConst = ConstantExpr::getICmp(CmpInst->getPredicate(), 5361 DefaultValue, CmpOp1, true); 5362 if (DefaultConst != TrueConst && DefaultConst != FalseConst) 5363 return; 5364 5365 // Check if the compare with the case values is distinct from the default 5366 // compare result. 5367 for (auto ValuePair : Values) { 5368 Constant *CaseConst = ConstantExpr::getICmp(CmpInst->getPredicate(), 5369 ValuePair.second, CmpOp1, true); 5370 if (!CaseConst || CaseConst == DefaultConst || isa<UndefValue>(CaseConst)) 5371 return; 5372 assert((CaseConst == TrueConst || CaseConst == FalseConst) && 5373 "Expect true or false as compare result."); 5374 } 5375 5376 // Check if the branch instruction dominates the phi node. It's a simple 5377 // dominance check, but sufficient for our needs. 5378 // Although this check is invariant in the calling loops, it's better to do it 5379 // at this late stage. Practically we do it at most once for a switch. 5380 BasicBlock *BranchBlock = RangeCheckBranch->getParent(); 5381 for (auto PI = pred_begin(PhiBlock), E = pred_end(PhiBlock); PI != E; ++PI) { 5382 BasicBlock *Pred = *PI; 5383 if (Pred != BranchBlock && Pred->getUniquePredecessor() != BranchBlock) 5384 return; 5385 } 5386 5387 if (DefaultConst == FalseConst) { 5388 // The compare yields the same result. We can replace it. 5389 CmpInst->replaceAllUsesWith(RangeCmp); 5390 ++NumTableCmpReuses; 5391 } else { 5392 // The compare yields the same result, just inverted. We can replace it. 5393 Value *InvertedTableCmp = BinaryOperator::CreateXor( 5394 RangeCmp, ConstantInt::get(RangeCmp->getType(), 1), "inverted.cmp", 5395 RangeCheckBranch); 5396 CmpInst->replaceAllUsesWith(InvertedTableCmp); 5397 ++NumTableCmpReuses; 5398 } 5399 } 5400 5401 /// If the switch is only used to initialize one or more phi nodes in a common 5402 /// successor block with different constant values, replace the switch with 5403 /// lookup tables. 5404 static bool SwitchToLookupTable(SwitchInst *SI, IRBuilder<> &Builder, 5405 const DataLayout &DL, 5406 const TargetTransformInfo &TTI) { 5407 assert(SI->getNumCases() > 1 && "Degenerate switch?"); 5408 5409 Function *Fn = SI->getParent()->getParent(); 5410 // Only build lookup table when we have a target that supports it or the 5411 // attribute is not set. 5412 if (!TTI.shouldBuildLookupTables() || 5413 (Fn->getFnAttribute("no-jump-tables").getValueAsString() == "true")) 5414 return false; 5415 5416 // FIXME: If the switch is too sparse for a lookup table, perhaps we could 5417 // split off a dense part and build a lookup table for that. 5418 5419 // FIXME: This creates arrays of GEPs to constant strings, which means each 5420 // GEP needs a runtime relocation in PIC code. We should just build one big 5421 // string and lookup indices into that. 5422 5423 // Ignore switches with less than three cases. Lookup tables will not make 5424 // them faster, so we don't analyze them. 5425 if (SI->getNumCases() < 3) 5426 return false; 5427 5428 // Figure out the corresponding result for each case value and phi node in the 5429 // common destination, as well as the min and max case values. 5430 assert(!SI->cases().empty()); 5431 SwitchInst::CaseIt CI = SI->case_begin(); 5432 ConstantInt *MinCaseVal = CI->getCaseValue(); 5433 ConstantInt *MaxCaseVal = CI->getCaseValue(); 5434 5435 BasicBlock *CommonDest = nullptr; 5436 5437 using ResultListTy = SmallVector<std::pair<ConstantInt *, Constant *>, 4>; 5438 SmallDenseMap<PHINode *, ResultListTy> ResultLists; 5439 5440 SmallDenseMap<PHINode *, Constant *> DefaultResults; 5441 SmallDenseMap<PHINode *, Type *> ResultTypes; 5442 SmallVector<PHINode *, 4> PHIs; 5443 5444 for (SwitchInst::CaseIt E = SI->case_end(); CI != E; ++CI) { 5445 ConstantInt *CaseVal = CI->getCaseValue(); 5446 if (CaseVal->getValue().slt(MinCaseVal->getValue())) 5447 MinCaseVal = CaseVal; 5448 if (CaseVal->getValue().sgt(MaxCaseVal->getValue())) 5449 MaxCaseVal = CaseVal; 5450 5451 // Resulting value at phi nodes for this case value. 5452 using ResultsTy = SmallVector<std::pair<PHINode *, Constant *>, 4>; 5453 ResultsTy Results; 5454 if (!GetCaseResults(SI, CaseVal, CI->getCaseSuccessor(), &CommonDest, 5455 Results, DL, TTI)) 5456 return false; 5457 5458 // Append the result from this case to the list for each phi. 5459 for (const auto &I : Results) { 5460 PHINode *PHI = I.first; 5461 Constant *Value = I.second; 5462 if (!ResultLists.count(PHI)) 5463 PHIs.push_back(PHI); 5464 ResultLists[PHI].push_back(std::make_pair(CaseVal, Value)); 5465 } 5466 } 5467 5468 // Keep track of the result types. 5469 for (PHINode *PHI : PHIs) { 5470 ResultTypes[PHI] = ResultLists[PHI][0].second->getType(); 5471 } 5472 5473 uint64_t NumResults = ResultLists[PHIs[0]].size(); 5474 APInt RangeSpread = MaxCaseVal->getValue() - MinCaseVal->getValue(); 5475 uint64_t TableSize = RangeSpread.getLimitedValue() + 1; 5476 bool TableHasHoles = (NumResults < TableSize); 5477 5478 // If the table has holes, we need a constant result for the default case 5479 // or a bitmask that fits in a register. 5480 SmallVector<std::pair<PHINode *, Constant *>, 4> DefaultResultsList; 5481 bool HasDefaultResults = 5482 GetCaseResults(SI, nullptr, SI->getDefaultDest(), &CommonDest, 5483 DefaultResultsList, DL, TTI); 5484 5485 bool NeedMask = (TableHasHoles && !HasDefaultResults); 5486 if (NeedMask) { 5487 // As an extra penalty for the validity test we require more cases. 5488 if (SI->getNumCases() < 4) // FIXME: Find best threshold value (benchmark). 5489 return false; 5490 if (!DL.fitsInLegalInteger(TableSize)) 5491 return false; 5492 } 5493 5494 for (const auto &I : DefaultResultsList) { 5495 PHINode *PHI = I.first; 5496 Constant *Result = I.second; 5497 DefaultResults[PHI] = Result; 5498 } 5499 5500 if (!ShouldBuildLookupTable(SI, TableSize, TTI, DL, ResultTypes)) 5501 return false; 5502 5503 // Create the BB that does the lookups. 5504 Module &Mod = *CommonDest->getParent()->getParent(); 5505 BasicBlock *LookupBB = BasicBlock::Create( 5506 Mod.getContext(), "switch.lookup", CommonDest->getParent(), CommonDest); 5507 5508 // Compute the table index value. 5509 Builder.SetInsertPoint(SI); 5510 Value *TableIndex; 5511 if (MinCaseVal->isNullValue()) 5512 TableIndex = SI->getCondition(); 5513 else 5514 TableIndex = Builder.CreateSub(SI->getCondition(), MinCaseVal, 5515 "switch.tableidx"); 5516 5517 // Compute the maximum table size representable by the integer type we are 5518 // switching upon. 5519 unsigned CaseSize = MinCaseVal->getType()->getPrimitiveSizeInBits(); 5520 uint64_t MaxTableSize = CaseSize > 63 ? UINT64_MAX : 1ULL << CaseSize; 5521 assert(MaxTableSize >= TableSize && 5522 "It is impossible for a switch to have more entries than the max " 5523 "representable value of its input integer type's size."); 5524 5525 // If the default destination is unreachable, or if the lookup table covers 5526 // all values of the conditional variable, branch directly to the lookup table 5527 // BB. Otherwise, check that the condition is within the case range. 5528 const bool DefaultIsReachable = 5529 !isa<UnreachableInst>(SI->getDefaultDest()->getFirstNonPHIOrDbg()); 5530 const bool GeneratingCoveredLookupTable = (MaxTableSize == TableSize); 5531 BranchInst *RangeCheckBranch = nullptr; 5532 5533 if (!DefaultIsReachable || GeneratingCoveredLookupTable) { 5534 Builder.CreateBr(LookupBB); 5535 // Note: We call removeProdecessor later since we need to be able to get the 5536 // PHI value for the default case in case we're using a bit mask. 5537 } else { 5538 Value *Cmp = Builder.CreateICmpULT( 5539 TableIndex, ConstantInt::get(MinCaseVal->getType(), TableSize)); 5540 RangeCheckBranch = 5541 Builder.CreateCondBr(Cmp, LookupBB, SI->getDefaultDest()); 5542 } 5543 5544 // Populate the BB that does the lookups. 5545 Builder.SetInsertPoint(LookupBB); 5546 5547 if (NeedMask) { 5548 // Before doing the lookup, we do the hole check. The LookupBB is therefore 5549 // re-purposed to do the hole check, and we create a new LookupBB. 5550 BasicBlock *MaskBB = LookupBB; 5551 MaskBB->setName("switch.hole_check"); 5552 LookupBB = BasicBlock::Create(Mod.getContext(), "switch.lookup", 5553 CommonDest->getParent(), CommonDest); 5554 5555 // Make the mask's bitwidth at least 8-bit and a power-of-2 to avoid 5556 // unnecessary illegal types. 5557 uint64_t TableSizePowOf2 = NextPowerOf2(std::max(7ULL, TableSize - 1ULL)); 5558 APInt MaskInt(TableSizePowOf2, 0); 5559 APInt One(TableSizePowOf2, 1); 5560 // Build bitmask; fill in a 1 bit for every case. 5561 const ResultListTy &ResultList = ResultLists[PHIs[0]]; 5562 for (size_t I = 0, E = ResultList.size(); I != E; ++I) { 5563 uint64_t Idx = (ResultList[I].first->getValue() - MinCaseVal->getValue()) 5564 .getLimitedValue(); 5565 MaskInt |= One << Idx; 5566 } 5567 ConstantInt *TableMask = ConstantInt::get(Mod.getContext(), MaskInt); 5568 5569 // Get the TableIndex'th bit of the bitmask. 5570 // If this bit is 0 (meaning hole) jump to the default destination, 5571 // else continue with table lookup. 5572 IntegerType *MapTy = TableMask->getType(); 5573 Value *MaskIndex = 5574 Builder.CreateZExtOrTrunc(TableIndex, MapTy, "switch.maskindex"); 5575 Value *Shifted = Builder.CreateLShr(TableMask, MaskIndex, "switch.shifted"); 5576 Value *LoBit = Builder.CreateTrunc( 5577 Shifted, Type::getInt1Ty(Mod.getContext()), "switch.lobit"); 5578 Builder.CreateCondBr(LoBit, LookupBB, SI->getDefaultDest()); 5579 5580 Builder.SetInsertPoint(LookupBB); 5581 AddPredecessorToBlock(SI->getDefaultDest(), MaskBB, SI->getParent()); 5582 } 5583 5584 if (!DefaultIsReachable || GeneratingCoveredLookupTable) { 5585 // We cached PHINodes in PHIs. To avoid accessing deleted PHINodes later, 5586 // do not delete PHINodes here. 5587 SI->getDefaultDest()->removePredecessor(SI->getParent(), 5588 /*KeepOneInputPHIs=*/true); 5589 } 5590 5591 bool ReturnedEarly = false; 5592 for (PHINode *PHI : PHIs) { 5593 const ResultListTy &ResultList = ResultLists[PHI]; 5594 5595 // If using a bitmask, use any value to fill the lookup table holes. 5596 Constant *DV = NeedMask ? ResultLists[PHI][0].second : DefaultResults[PHI]; 5597 StringRef FuncName = Fn->getName(); 5598 SwitchLookupTable Table(Mod, TableSize, MinCaseVal, ResultList, DV, DL, 5599 FuncName); 5600 5601 Value *Result = Table.BuildLookup(TableIndex, Builder); 5602 5603 // If the result is used to return immediately from the function, we want to 5604 // do that right here. 5605 if (PHI->hasOneUse() && isa<ReturnInst>(*PHI->user_begin()) && 5606 PHI->user_back() == CommonDest->getFirstNonPHIOrDbg()) { 5607 Builder.CreateRet(Result); 5608 ReturnedEarly = true; 5609 break; 5610 } 5611 5612 // Do a small peephole optimization: re-use the switch table compare if 5613 // possible. 5614 if (!TableHasHoles && HasDefaultResults && RangeCheckBranch) { 5615 BasicBlock *PhiBlock = PHI->getParent(); 5616 // Search for compare instructions which use the phi. 5617 for (auto *User : PHI->users()) { 5618 reuseTableCompare(User, PhiBlock, RangeCheckBranch, DV, ResultList); 5619 } 5620 } 5621 5622 PHI->addIncoming(Result, LookupBB); 5623 } 5624 5625 if (!ReturnedEarly) 5626 Builder.CreateBr(CommonDest); 5627 5628 // Remove the switch. 5629 for (unsigned i = 0, e = SI->getNumSuccessors(); i < e; ++i) { 5630 BasicBlock *Succ = SI->getSuccessor(i); 5631 5632 if (Succ == SI->getDefaultDest()) 5633 continue; 5634 Succ->removePredecessor(SI->getParent()); 5635 } 5636 SI->eraseFromParent(); 5637 5638 ++NumLookupTables; 5639 if (NeedMask) 5640 ++NumLookupTablesHoles; 5641 return true; 5642 } 5643 5644 static bool isSwitchDense(ArrayRef<int64_t> Values) { 5645 // See also SelectionDAGBuilder::isDense(), which this function was based on. 5646 uint64_t Diff = (uint64_t)Values.back() - (uint64_t)Values.front(); 5647 uint64_t Range = Diff + 1; 5648 uint64_t NumCases = Values.size(); 5649 // 40% is the default density for building a jump table in optsize/minsize mode. 5650 uint64_t MinDensity = 40; 5651 5652 return NumCases * 100 >= Range * MinDensity; 5653 } 5654 5655 /// Try to transform a switch that has "holes" in it to a contiguous sequence 5656 /// of cases. 5657 /// 5658 /// A switch such as: switch(i) {case 5: case 9: case 13: case 17:} can be 5659 /// range-reduced to: switch ((i-5) / 4) {case 0: case 1: case 2: case 3:}. 5660 /// 5661 /// This converts a sparse switch into a dense switch which allows better 5662 /// lowering and could also allow transforming into a lookup table. 5663 static bool ReduceSwitchRange(SwitchInst *SI, IRBuilder<> &Builder, 5664 const DataLayout &DL, 5665 const TargetTransformInfo &TTI) { 5666 auto *CondTy = cast<IntegerType>(SI->getCondition()->getType()); 5667 if (CondTy->getIntegerBitWidth() > 64 || 5668 !DL.fitsInLegalInteger(CondTy->getIntegerBitWidth())) 5669 return false; 5670 // Only bother with this optimization if there are more than 3 switch cases; 5671 // SDAG will only bother creating jump tables for 4 or more cases. 5672 if (SI->getNumCases() < 4) 5673 return false; 5674 5675 // This transform is agnostic to the signedness of the input or case values. We 5676 // can treat the case values as signed or unsigned. We can optimize more common 5677 // cases such as a sequence crossing zero {-4,0,4,8} if we interpret case values 5678 // as signed. 5679 SmallVector<int64_t,4> Values; 5680 for (auto &C : SI->cases()) 5681 Values.push_back(C.getCaseValue()->getValue().getSExtValue()); 5682 llvm::sort(Values); 5683 5684 // If the switch is already dense, there's nothing useful to do here. 5685 if (isSwitchDense(Values)) 5686 return false; 5687 5688 // First, transform the values such that they start at zero and ascend. 5689 int64_t Base = Values[0]; 5690 for (auto &V : Values) 5691 V -= (uint64_t)(Base); 5692 5693 // Now we have signed numbers that have been shifted so that, given enough 5694 // precision, there are no negative values. Since the rest of the transform 5695 // is bitwise only, we switch now to an unsigned representation. 5696 5697 // This transform can be done speculatively because it is so cheap - it 5698 // results in a single rotate operation being inserted. 5699 // FIXME: It's possible that optimizing a switch on powers of two might also 5700 // be beneficial - flag values are often powers of two and we could use a CLZ 5701 // as the key function. 5702 5703 // countTrailingZeros(0) returns 64. As Values is guaranteed to have more than 5704 // one element and LLVM disallows duplicate cases, Shift is guaranteed to be 5705 // less than 64. 5706 unsigned Shift = 64; 5707 for (auto &V : Values) 5708 Shift = std::min(Shift, countTrailingZeros((uint64_t)V)); 5709 assert(Shift < 64); 5710 if (Shift > 0) 5711 for (auto &V : Values) 5712 V = (int64_t)((uint64_t)V >> Shift); 5713 5714 if (!isSwitchDense(Values)) 5715 // Transform didn't create a dense switch. 5716 return false; 5717 5718 // The obvious transform is to shift the switch condition right and emit a 5719 // check that the condition actually cleanly divided by GCD, i.e. 5720 // C & (1 << Shift - 1) == 0 5721 // inserting a new CFG edge to handle the case where it didn't divide cleanly. 5722 // 5723 // A cheaper way of doing this is a simple ROTR(C, Shift). This performs the 5724 // shift and puts the shifted-off bits in the uppermost bits. If any of these 5725 // are nonzero then the switch condition will be very large and will hit the 5726 // default case. 5727 5728 auto *Ty = cast<IntegerType>(SI->getCondition()->getType()); 5729 Builder.SetInsertPoint(SI); 5730 auto *ShiftC = ConstantInt::get(Ty, Shift); 5731 auto *Sub = Builder.CreateSub(SI->getCondition(), ConstantInt::get(Ty, Base)); 5732 auto *LShr = Builder.CreateLShr(Sub, ShiftC); 5733 auto *Shl = Builder.CreateShl(Sub, Ty->getBitWidth() - Shift); 5734 auto *Rot = Builder.CreateOr(LShr, Shl); 5735 SI->replaceUsesOfWith(SI->getCondition(), Rot); 5736 5737 for (auto Case : SI->cases()) { 5738 auto *Orig = Case.getCaseValue(); 5739 auto Sub = Orig->getValue() - APInt(Ty->getBitWidth(), Base); 5740 Case.setValue( 5741 cast<ConstantInt>(ConstantInt::get(Ty, Sub.lshr(ShiftC->getValue())))); 5742 } 5743 return true; 5744 } 5745 5746 bool SimplifyCFGOpt::simplifySwitch(SwitchInst *SI, IRBuilder<> &Builder) { 5747 BasicBlock *BB = SI->getParent(); 5748 5749 if (isValueEqualityComparison(SI)) { 5750 // If we only have one predecessor, and if it is a branch on this value, 5751 // see if that predecessor totally determines the outcome of this switch. 5752 if (BasicBlock *OnlyPred = BB->getSinglePredecessor()) 5753 if (SimplifyEqualityComparisonWithOnlyPredecessor(SI, OnlyPred, Builder)) 5754 return requestResimplify(); 5755 5756 Value *Cond = SI->getCondition(); 5757 if (SelectInst *Select = dyn_cast<SelectInst>(Cond)) 5758 if (SimplifySwitchOnSelect(SI, Select)) 5759 return requestResimplify(); 5760 5761 // If the block only contains the switch, see if we can fold the block 5762 // away into any preds. 5763 if (SI == &*BB->instructionsWithoutDebug().begin()) 5764 if (FoldValueComparisonIntoPredecessors(SI, Builder)) 5765 return requestResimplify(); 5766 } 5767 5768 // Try to transform the switch into an icmp and a branch. 5769 if (TurnSwitchRangeIntoICmp(SI, Builder)) 5770 return requestResimplify(); 5771 5772 // Remove unreachable cases. 5773 if (eliminateDeadSwitchCases(SI, Options.AC, DL)) 5774 return requestResimplify(); 5775 5776 if (switchToSelect(SI, Builder, DL, TTI)) 5777 return requestResimplify(); 5778 5779 if (Options.ForwardSwitchCondToPhi && ForwardSwitchConditionToPHI(SI)) 5780 return requestResimplify(); 5781 5782 // The conversion from switch to lookup tables results in difficult-to-analyze 5783 // code and makes pruning branches much harder. This is a problem if the 5784 // switch expression itself can still be restricted as a result of inlining or 5785 // CVP. Therefore, only apply this transformation during late stages of the 5786 // optimisation pipeline. 5787 if (Options.ConvertSwitchToLookupTable && 5788 SwitchToLookupTable(SI, Builder, DL, TTI)) 5789 return requestResimplify(); 5790 5791 if (ReduceSwitchRange(SI, Builder, DL, TTI)) 5792 return requestResimplify(); 5793 5794 return false; 5795 } 5796 5797 bool SimplifyCFGOpt::simplifyIndirectBr(IndirectBrInst *IBI) { 5798 BasicBlock *BB = IBI->getParent(); 5799 bool Changed = false; 5800 5801 // Eliminate redundant destinations. 5802 SmallPtrSet<Value *, 8> Succs; 5803 for (unsigned i = 0, e = IBI->getNumDestinations(); i != e; ++i) { 5804 BasicBlock *Dest = IBI->getDestination(i); 5805 if (!Dest->hasAddressTaken() || !Succs.insert(Dest).second) { 5806 Dest->removePredecessor(BB); 5807 IBI->removeDestination(i); 5808 --i; 5809 --e; 5810 Changed = true; 5811 } 5812 } 5813 5814 if (IBI->getNumDestinations() == 0) { 5815 // If the indirectbr has no successors, change it to unreachable. 5816 new UnreachableInst(IBI->getContext(), IBI); 5817 EraseTerminatorAndDCECond(IBI); 5818 return true; 5819 } 5820 5821 if (IBI->getNumDestinations() == 1) { 5822 // If the indirectbr has one successor, change it to a direct branch. 5823 BranchInst::Create(IBI->getDestination(0), IBI); 5824 EraseTerminatorAndDCECond(IBI); 5825 return true; 5826 } 5827 5828 if (SelectInst *SI = dyn_cast<SelectInst>(IBI->getAddress())) { 5829 if (SimplifyIndirectBrOnSelect(IBI, SI)) 5830 return requestResimplify(); 5831 } 5832 return Changed; 5833 } 5834 5835 /// Given an block with only a single landing pad and a unconditional branch 5836 /// try to find another basic block which this one can be merged with. This 5837 /// handles cases where we have multiple invokes with unique landing pads, but 5838 /// a shared handler. 5839 /// 5840 /// We specifically choose to not worry about merging non-empty blocks 5841 /// here. That is a PRE/scheduling problem and is best solved elsewhere. In 5842 /// practice, the optimizer produces empty landing pad blocks quite frequently 5843 /// when dealing with exception dense code. (see: instcombine, gvn, if-else 5844 /// sinking in this file) 5845 /// 5846 /// This is primarily a code size optimization. We need to avoid performing 5847 /// any transform which might inhibit optimization (such as our ability to 5848 /// specialize a particular handler via tail commoning). We do this by not 5849 /// merging any blocks which require us to introduce a phi. Since the same 5850 /// values are flowing through both blocks, we don't lose any ability to 5851 /// specialize. If anything, we make such specialization more likely. 5852 /// 5853 /// TODO - This transformation could remove entries from a phi in the target 5854 /// block when the inputs in the phi are the same for the two blocks being 5855 /// merged. In some cases, this could result in removal of the PHI entirely. 5856 static bool TryToMergeLandingPad(LandingPadInst *LPad, BranchInst *BI, 5857 BasicBlock *BB) { 5858 auto Succ = BB->getUniqueSuccessor(); 5859 assert(Succ); 5860 // If there's a phi in the successor block, we'd likely have to introduce 5861 // a phi into the merged landing pad block. 5862 if (isa<PHINode>(*Succ->begin())) 5863 return false; 5864 5865 for (BasicBlock *OtherPred : predecessors(Succ)) { 5866 if (BB == OtherPred) 5867 continue; 5868 BasicBlock::iterator I = OtherPred->begin(); 5869 LandingPadInst *LPad2 = dyn_cast<LandingPadInst>(I); 5870 if (!LPad2 || !LPad2->isIdenticalTo(LPad)) 5871 continue; 5872 for (++I; isa<DbgInfoIntrinsic>(I); ++I) 5873 ; 5874 BranchInst *BI2 = dyn_cast<BranchInst>(I); 5875 if (!BI2 || !BI2->isIdenticalTo(BI)) 5876 continue; 5877 5878 // We've found an identical block. Update our predecessors to take that 5879 // path instead and make ourselves dead. 5880 SmallPtrSet<BasicBlock *, 16> Preds; 5881 Preds.insert(pred_begin(BB), pred_end(BB)); 5882 for (BasicBlock *Pred : Preds) { 5883 InvokeInst *II = cast<InvokeInst>(Pred->getTerminator()); 5884 assert(II->getNormalDest() != BB && II->getUnwindDest() == BB && 5885 "unexpected successor"); 5886 II->setUnwindDest(OtherPred); 5887 } 5888 5889 // The debug info in OtherPred doesn't cover the merged control flow that 5890 // used to go through BB. We need to delete it or update it. 5891 for (auto I = OtherPred->begin(), E = OtherPred->end(); I != E;) { 5892 Instruction &Inst = *I; 5893 I++; 5894 if (isa<DbgInfoIntrinsic>(Inst)) 5895 Inst.eraseFromParent(); 5896 } 5897 5898 SmallPtrSet<BasicBlock *, 16> Succs; 5899 Succs.insert(succ_begin(BB), succ_end(BB)); 5900 for (BasicBlock *Succ : Succs) { 5901 Succ->removePredecessor(BB); 5902 } 5903 5904 IRBuilder<> Builder(BI); 5905 Builder.CreateUnreachable(); 5906 BI->eraseFromParent(); 5907 return true; 5908 } 5909 return false; 5910 } 5911 5912 bool SimplifyCFGOpt::simplifyBranch(BranchInst *Branch, IRBuilder<> &Builder) { 5913 return Branch->isUnconditional() ? simplifyUncondBranch(Branch, Builder) 5914 : simplifyCondBranch(Branch, Builder); 5915 } 5916 5917 bool SimplifyCFGOpt::simplifyUncondBranch(BranchInst *BI, 5918 IRBuilder<> &Builder) { 5919 BasicBlock *BB = BI->getParent(); 5920 BasicBlock *Succ = BI->getSuccessor(0); 5921 5922 // If the Terminator is the only non-phi instruction, simplify the block. 5923 // If LoopHeader is provided, check if the block or its successor is a loop 5924 // header. (This is for early invocations before loop simplify and 5925 // vectorization to keep canonical loop forms for nested loops. These blocks 5926 // can be eliminated when the pass is invoked later in the back-end.) 5927 // Note that if BB has only one predecessor then we do not introduce new 5928 // backedge, so we can eliminate BB. 5929 bool NeedCanonicalLoop = 5930 Options.NeedCanonicalLoop && 5931 (LoopHeaders && BB->hasNPredecessorsOrMore(2) && 5932 (LoopHeaders->count(BB) || LoopHeaders->count(Succ))); 5933 BasicBlock::iterator I = BB->getFirstNonPHIOrDbg()->getIterator(); 5934 if (I->isTerminator() && BB != &BB->getParent()->getEntryBlock() && 5935 !NeedCanonicalLoop && TryToSimplifyUncondBranchFromEmptyBlock(BB)) 5936 return true; 5937 5938 // If the only instruction in the block is a seteq/setne comparison against a 5939 // constant, try to simplify the block. 5940 if (ICmpInst *ICI = dyn_cast<ICmpInst>(I)) 5941 if (ICI->isEquality() && isa<ConstantInt>(ICI->getOperand(1))) { 5942 for (++I; isa<DbgInfoIntrinsic>(I); ++I) 5943 ; 5944 if (I->isTerminator() && 5945 tryToSimplifyUncondBranchWithICmpInIt(ICI, Builder)) 5946 return true; 5947 } 5948 5949 // See if we can merge an empty landing pad block with another which is 5950 // equivalent. 5951 if (LandingPadInst *LPad = dyn_cast<LandingPadInst>(I)) { 5952 for (++I; isa<DbgInfoIntrinsic>(I); ++I) 5953 ; 5954 if (I->isTerminator() && TryToMergeLandingPad(LPad, BI, BB)) 5955 return true; 5956 } 5957 5958 // If this basic block is ONLY a compare and a branch, and if a predecessor 5959 // branches to us and our successor, fold the comparison into the 5960 // predecessor and use logical operations to update the incoming value 5961 // for PHI nodes in common successor. 5962 if (FoldBranchToCommonDest(BI, nullptr, Options.BonusInstThreshold)) 5963 return requestResimplify(); 5964 return false; 5965 } 5966 5967 static BasicBlock *allPredecessorsComeFromSameSource(BasicBlock *BB) { 5968 BasicBlock *PredPred = nullptr; 5969 for (auto *P : predecessors(BB)) { 5970 BasicBlock *PPred = P->getSinglePredecessor(); 5971 if (!PPred || (PredPred && PredPred != PPred)) 5972 return nullptr; 5973 PredPred = PPred; 5974 } 5975 return PredPred; 5976 } 5977 5978 bool SimplifyCFGOpt::simplifyCondBranch(BranchInst *BI, IRBuilder<> &Builder) { 5979 BasicBlock *BB = BI->getParent(); 5980 if (!Options.SimplifyCondBranch) 5981 return false; 5982 5983 // Conditional branch 5984 if (isValueEqualityComparison(BI)) { 5985 // If we only have one predecessor, and if it is a branch on this value, 5986 // see if that predecessor totally determines the outcome of this 5987 // switch. 5988 if (BasicBlock *OnlyPred = BB->getSinglePredecessor()) 5989 if (SimplifyEqualityComparisonWithOnlyPredecessor(BI, OnlyPred, Builder)) 5990 return requestResimplify(); 5991 5992 // This block must be empty, except for the setcond inst, if it exists. 5993 // Ignore dbg intrinsics. 5994 auto I = BB->instructionsWithoutDebug().begin(); 5995 if (&*I == BI) { 5996 if (FoldValueComparisonIntoPredecessors(BI, Builder)) 5997 return requestResimplify(); 5998 } else if (&*I == cast<Instruction>(BI->getCondition())) { 5999 ++I; 6000 if (&*I == BI && FoldValueComparisonIntoPredecessors(BI, Builder)) 6001 return requestResimplify(); 6002 } 6003 } 6004 6005 // Try to turn "br (X == 0 | X == 1), T, F" into a switch instruction. 6006 if (SimplifyBranchOnICmpChain(BI, Builder, DL)) 6007 return true; 6008 6009 // If this basic block has dominating predecessor blocks and the dominating 6010 // blocks' conditions imply BI's condition, we know the direction of BI. 6011 Optional<bool> Imp = isImpliedByDomCondition(BI->getCondition(), BI, DL); 6012 if (Imp) { 6013 // Turn this into a branch on constant. 6014 auto *OldCond = BI->getCondition(); 6015 ConstantInt *TorF = *Imp ? ConstantInt::getTrue(BB->getContext()) 6016 : ConstantInt::getFalse(BB->getContext()); 6017 BI->setCondition(TorF); 6018 RecursivelyDeleteTriviallyDeadInstructions(OldCond); 6019 return requestResimplify(); 6020 } 6021 6022 // If this basic block is ONLY a compare and a branch, and if a predecessor 6023 // branches to us and one of our successors, fold the comparison into the 6024 // predecessor and use logical operations to pick the right destination. 6025 if (FoldBranchToCommonDest(BI, nullptr, Options.BonusInstThreshold)) 6026 return requestResimplify(); 6027 6028 // We have a conditional branch to two blocks that are only reachable 6029 // from BI. We know that the condbr dominates the two blocks, so see if 6030 // there is any identical code in the "then" and "else" blocks. If so, we 6031 // can hoist it up to the branching block. 6032 if (BI->getSuccessor(0)->getSinglePredecessor()) { 6033 if (BI->getSuccessor(1)->getSinglePredecessor()) { 6034 if (HoistThenElseCodeToIf(BI, TTI)) 6035 return requestResimplify(); 6036 } else { 6037 // If Successor #1 has multiple preds, we may be able to conditionally 6038 // execute Successor #0 if it branches to Successor #1. 6039 Instruction *Succ0TI = BI->getSuccessor(0)->getTerminator(); 6040 if (Succ0TI->getNumSuccessors() == 1 && 6041 Succ0TI->getSuccessor(0) == BI->getSuccessor(1)) 6042 if (SpeculativelyExecuteBB(BI, BI->getSuccessor(0), TTI)) 6043 return requestResimplify(); 6044 } 6045 } else if (BI->getSuccessor(1)->getSinglePredecessor()) { 6046 // If Successor #0 has multiple preds, we may be able to conditionally 6047 // execute Successor #1 if it branches to Successor #0. 6048 Instruction *Succ1TI = BI->getSuccessor(1)->getTerminator(); 6049 if (Succ1TI->getNumSuccessors() == 1 && 6050 Succ1TI->getSuccessor(0) == BI->getSuccessor(0)) 6051 if (SpeculativelyExecuteBB(BI, BI->getSuccessor(1), TTI)) 6052 return requestResimplify(); 6053 } 6054 6055 // If this is a branch on a phi node in the current block, thread control 6056 // through this block if any PHI node entries are constants. 6057 if (PHINode *PN = dyn_cast<PHINode>(BI->getCondition())) 6058 if (PN->getParent() == BI->getParent()) 6059 if (FoldCondBranchOnPHI(BI, DL, Options.AC)) 6060 return requestResimplify(); 6061 6062 // Scan predecessor blocks for conditional branches. 6063 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) 6064 if (BranchInst *PBI = dyn_cast<BranchInst>((*PI)->getTerminator())) 6065 if (PBI != BI && PBI->isConditional()) 6066 if (SimplifyCondBranchToCondBranch(PBI, BI, DL, TTI)) 6067 return requestResimplify(); 6068 6069 // Look for diamond patterns. 6070 if (MergeCondStores) 6071 if (BasicBlock *PrevBB = allPredecessorsComeFromSameSource(BB)) 6072 if (BranchInst *PBI = dyn_cast<BranchInst>(PrevBB->getTerminator())) 6073 if (PBI != BI && PBI->isConditional()) 6074 if (mergeConditionalStores(PBI, BI, DL, TTI)) 6075 return requestResimplify(); 6076 6077 return false; 6078 } 6079 6080 /// Check if passing a value to an instruction will cause undefined behavior. 6081 static bool passingValueIsAlwaysUndefined(Value *V, Instruction *I) { 6082 Constant *C = dyn_cast<Constant>(V); 6083 if (!C) 6084 return false; 6085 6086 if (I->use_empty()) 6087 return false; 6088 6089 if (C->isNullValue() || isa<UndefValue>(C)) { 6090 // Only look at the first use, avoid hurting compile time with long uselists 6091 User *Use = *I->user_begin(); 6092 6093 // Now make sure that there are no instructions in between that can alter 6094 // control flow (eg. calls) 6095 for (BasicBlock::iterator 6096 i = ++BasicBlock::iterator(I), 6097 UI = BasicBlock::iterator(dyn_cast<Instruction>(Use)); 6098 i != UI; ++i) 6099 if (i == I->getParent()->end() || i->mayHaveSideEffects()) 6100 return false; 6101 6102 // Look through GEPs. A load from a GEP derived from NULL is still undefined 6103 if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(Use)) 6104 if (GEP->getPointerOperand() == I) 6105 return passingValueIsAlwaysUndefined(V, GEP); 6106 6107 // Look through bitcasts. 6108 if (BitCastInst *BC = dyn_cast<BitCastInst>(Use)) 6109 return passingValueIsAlwaysUndefined(V, BC); 6110 6111 // Load from null is undefined. 6112 if (LoadInst *LI = dyn_cast<LoadInst>(Use)) 6113 if (!LI->isVolatile()) 6114 return !NullPointerIsDefined(LI->getFunction(), 6115 LI->getPointerAddressSpace()); 6116 6117 // Store to null is undefined. 6118 if (StoreInst *SI = dyn_cast<StoreInst>(Use)) 6119 if (!SI->isVolatile()) 6120 return (!NullPointerIsDefined(SI->getFunction(), 6121 SI->getPointerAddressSpace())) && 6122 SI->getPointerOperand() == I; 6123 6124 // A call to null is undefined. 6125 if (auto *CB = dyn_cast<CallBase>(Use)) 6126 return !NullPointerIsDefined(CB->getFunction()) && 6127 CB->getCalledOperand() == I; 6128 } 6129 return false; 6130 } 6131 6132 /// If BB has an incoming value that will always trigger undefined behavior 6133 /// (eg. null pointer dereference), remove the branch leading here. 6134 static bool removeUndefIntroducingPredecessor(BasicBlock *BB) { 6135 for (PHINode &PHI : BB->phis()) 6136 for (unsigned i = 0, e = PHI.getNumIncomingValues(); i != e; ++i) 6137 if (passingValueIsAlwaysUndefined(PHI.getIncomingValue(i), &PHI)) { 6138 Instruction *T = PHI.getIncomingBlock(i)->getTerminator(); 6139 IRBuilder<> Builder(T); 6140 if (BranchInst *BI = dyn_cast<BranchInst>(T)) { 6141 BB->removePredecessor(PHI.getIncomingBlock(i)); 6142 // Turn uncoditional branches into unreachables and remove the dead 6143 // destination from conditional branches. 6144 if (BI->isUnconditional()) 6145 Builder.CreateUnreachable(); 6146 else 6147 Builder.CreateBr(BI->getSuccessor(0) == BB ? BI->getSuccessor(1) 6148 : BI->getSuccessor(0)); 6149 BI->eraseFromParent(); 6150 return true; 6151 } 6152 // TODO: SwitchInst. 6153 } 6154 6155 return false; 6156 } 6157 6158 bool SimplifyCFGOpt::simplifyOnce(BasicBlock *BB) { 6159 bool Changed = false; 6160 6161 assert(BB && BB->getParent() && "Block not embedded in function!"); 6162 assert(BB->getTerminator() && "Degenerate basic block encountered!"); 6163 6164 // Remove basic blocks that have no predecessors (except the entry block)... 6165 // or that just have themself as a predecessor. These are unreachable. 6166 if ((pred_empty(BB) && BB != &BB->getParent()->getEntryBlock()) || 6167 BB->getSinglePredecessor() == BB) { 6168 LLVM_DEBUG(dbgs() << "Removing BB: \n" << *BB); 6169 DeleteDeadBlock(BB); 6170 return true; 6171 } 6172 6173 // Check to see if we can constant propagate this terminator instruction 6174 // away... 6175 Changed |= ConstantFoldTerminator(BB, true); 6176 6177 // Check for and eliminate duplicate PHI nodes in this block. 6178 Changed |= EliminateDuplicatePHINodes(BB); 6179 6180 // Check for and remove branches that will always cause undefined behavior. 6181 Changed |= removeUndefIntroducingPredecessor(BB); 6182 6183 // Merge basic blocks into their predecessor if there is only one distinct 6184 // pred, and if there is only one distinct successor of the predecessor, and 6185 // if there are no PHI nodes. 6186 if (MergeBlockIntoPredecessor(BB)) 6187 return true; 6188 6189 if (SinkCommon && Options.SinkCommonInsts) 6190 Changed |= SinkCommonCodeFromPredecessors(BB); 6191 6192 IRBuilder<> Builder(BB); 6193 6194 if (Options.FoldTwoEntryPHINode) { 6195 // If there is a trivial two-entry PHI node in this basic block, and we can 6196 // eliminate it, do so now. 6197 if (auto *PN = dyn_cast<PHINode>(BB->begin())) 6198 if (PN->getNumIncomingValues() == 2) 6199 Changed |= FoldTwoEntryPHINode(PN, TTI, DL); 6200 } 6201 6202 Instruction *Terminator = BB->getTerminator(); 6203 Builder.SetInsertPoint(Terminator); 6204 switch (Terminator->getOpcode()) { 6205 case Instruction::Br: 6206 Changed |= simplifyBranch(cast<BranchInst>(Terminator), Builder); 6207 break; 6208 case Instruction::Ret: 6209 Changed |= simplifyReturn(cast<ReturnInst>(Terminator), Builder); 6210 break; 6211 case Instruction::Resume: 6212 Changed |= simplifyResume(cast<ResumeInst>(Terminator), Builder); 6213 break; 6214 case Instruction::CleanupRet: 6215 Changed |= simplifyCleanupReturn(cast<CleanupReturnInst>(Terminator)); 6216 break; 6217 case Instruction::Switch: 6218 Changed |= simplifySwitch(cast<SwitchInst>(Terminator), Builder); 6219 break; 6220 case Instruction::Unreachable: 6221 Changed |= simplifyUnreachable(cast<UnreachableInst>(Terminator)); 6222 break; 6223 case Instruction::IndirectBr: 6224 Changed |= simplifyIndirectBr(cast<IndirectBrInst>(Terminator)); 6225 break; 6226 } 6227 6228 return Changed; 6229 } 6230 6231 bool SimplifyCFGOpt::run(BasicBlock *BB) { 6232 bool Changed = false; 6233 6234 // Repeated simplify BB as long as resimplification is requested. 6235 do { 6236 Resimplify = false; 6237 6238 // Perform one round of simplifcation. Resimplify flag will be set if 6239 // another iteration is requested. 6240 Changed |= simplifyOnce(BB); 6241 } while (Resimplify); 6242 6243 return Changed; 6244 } 6245 6246 bool llvm::simplifyCFG(BasicBlock *BB, const TargetTransformInfo &TTI, 6247 const SimplifyCFGOptions &Options, 6248 SmallPtrSetImpl<BasicBlock *> *LoopHeaders) { 6249 return SimplifyCFGOpt(TTI, BB->getModule()->getDataLayout(), LoopHeaders, 6250 Options) 6251 .run(BB); 6252 } 6253