1 //===- SCCP.cpp - Sparse Conditional Constant Propagation -----------------===// 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 // This file implements sparse conditional constant propagation and merging: 10 // 11 // Specifically, this: 12 // * Assumes values are constant unless proven otherwise 13 // * Assumes BasicBlocks are dead unless proven otherwise 14 // * Proves values to be constant, and replaces them with constants 15 // * Proves conditional branches to be unconditional 16 // 17 //===----------------------------------------------------------------------===// 18 19 #include "llvm/Transforms/Scalar/SCCP.h" 20 #include "llvm/ADT/ArrayRef.h" 21 #include "llvm/ADT/DenseMap.h" 22 #include "llvm/ADT/DenseSet.h" 23 #include "llvm/ADT/MapVector.h" 24 #include "llvm/ADT/PointerIntPair.h" 25 #include "llvm/ADT/STLExtras.h" 26 #include "llvm/ADT/SmallPtrSet.h" 27 #include "llvm/ADT/SmallVector.h" 28 #include "llvm/ADT/Statistic.h" 29 #include "llvm/Analysis/ConstantFolding.h" 30 #include "llvm/Analysis/GlobalsModRef.h" 31 #include "llvm/Analysis/TargetLibraryInfo.h" 32 #include "llvm/Analysis/ValueLattice.h" 33 #include "llvm/Analysis/ValueLatticeUtils.h" 34 #include "llvm/IR/BasicBlock.h" 35 #include "llvm/IR/CallSite.h" 36 #include "llvm/IR/Constant.h" 37 #include "llvm/IR/Constants.h" 38 #include "llvm/IR/DataLayout.h" 39 #include "llvm/IR/DerivedTypes.h" 40 #include "llvm/IR/Function.h" 41 #include "llvm/IR/GlobalVariable.h" 42 #include "llvm/IR/InstVisitor.h" 43 #include "llvm/IR/InstrTypes.h" 44 #include "llvm/IR/Instruction.h" 45 #include "llvm/IR/Instructions.h" 46 #include "llvm/IR/Module.h" 47 #include "llvm/IR/PassManager.h" 48 #include "llvm/IR/Type.h" 49 #include "llvm/IR/User.h" 50 #include "llvm/IR/Value.h" 51 #include "llvm/InitializePasses.h" 52 #include "llvm/Pass.h" 53 #include "llvm/Support/Casting.h" 54 #include "llvm/Support/Debug.h" 55 #include "llvm/Support/ErrorHandling.h" 56 #include "llvm/Support/raw_ostream.h" 57 #include "llvm/Transforms/Scalar.h" 58 #include "llvm/Transforms/Utils/Local.h" 59 #include "llvm/Transforms/Utils/PredicateInfo.h" 60 #include <cassert> 61 #include <utility> 62 #include <vector> 63 64 using namespace llvm; 65 66 #define DEBUG_TYPE "sccp" 67 68 STATISTIC(NumInstRemoved, "Number of instructions removed"); 69 STATISTIC(NumDeadBlocks , "Number of basic blocks unreachable"); 70 71 STATISTIC(IPNumInstRemoved, "Number of instructions removed by IPSCCP"); 72 STATISTIC(IPNumArgsElimed ,"Number of arguments constant propagated by IPSCCP"); 73 STATISTIC(IPNumGlobalConst, "Number of globals found to be constant by IPSCCP"); 74 75 namespace { 76 77 /// LatticeVal class - This class represents the different lattice values that 78 /// an LLVM value may occupy. It is a simple class with value semantics. 79 /// 80 class LatticeVal { 81 enum LatticeValueTy { 82 /// unknown - This LLVM Value has no known value yet. 83 unknown, 84 85 /// constant - This LLVM Value has a specific constant value. 86 constant, 87 88 /// forcedconstant - This LLVM Value was thought to be undef until 89 /// ResolvedUndefsIn. This is treated just like 'constant', but if merged 90 /// with another (different) constant, it goes to overdefined, instead of 91 /// asserting. 92 forcedconstant, 93 94 /// overdefined - This instruction is not known to be constant, and we know 95 /// it has a value. 96 overdefined 97 }; 98 99 /// Val: This stores the current lattice value along with the Constant* for 100 /// the constant if this is a 'constant' or 'forcedconstant' value. 101 PointerIntPair<Constant *, 2, LatticeValueTy> Val; 102 103 LatticeValueTy getLatticeValue() const { 104 return Val.getInt(); 105 } 106 107 public: 108 LatticeVal() : Val(nullptr, unknown) {} 109 110 bool isUnknown() const { return getLatticeValue() == unknown; } 111 112 bool isConstant() const { 113 return getLatticeValue() == constant || getLatticeValue() == forcedconstant; 114 } 115 116 bool isOverdefined() const { return getLatticeValue() == overdefined; } 117 118 Constant *getConstant() const { 119 assert(isConstant() && "Cannot get the constant of a non-constant!"); 120 return Val.getPointer(); 121 } 122 123 /// markOverdefined - Return true if this is a change in status. 124 bool markOverdefined() { 125 if (isOverdefined()) 126 return false; 127 128 Val.setInt(overdefined); 129 return true; 130 } 131 132 /// markConstant - Return true if this is a change in status. 133 bool markConstant(Constant *V) { 134 if (getLatticeValue() == constant) { // Constant but not forcedconstant. 135 assert(getConstant() == V && "Marking constant with different value"); 136 return false; 137 } 138 139 if (isUnknown()) { 140 Val.setInt(constant); 141 assert(V && "Marking constant with NULL"); 142 Val.setPointer(V); 143 } else { 144 assert(getLatticeValue() == forcedconstant && 145 "Cannot move from overdefined to constant!"); 146 // Stay at forcedconstant if the constant is the same. 147 if (V == getConstant()) return false; 148 149 // Otherwise, we go to overdefined. Assumptions made based on the 150 // forced value are possibly wrong. Assuming this is another constant 151 // could expose a contradiction. 152 Val.setInt(overdefined); 153 } 154 return true; 155 } 156 157 /// getConstantInt - If this is a constant with a ConstantInt value, return it 158 /// otherwise return null. 159 ConstantInt *getConstantInt() const { 160 if (isConstant()) 161 return dyn_cast<ConstantInt>(getConstant()); 162 return nullptr; 163 } 164 165 /// getBlockAddress - If this is a constant with a BlockAddress value, return 166 /// it, otherwise return null. 167 BlockAddress *getBlockAddress() const { 168 if (isConstant()) 169 return dyn_cast<BlockAddress>(getConstant()); 170 return nullptr; 171 } 172 173 void markForcedConstant(Constant *V) { 174 assert(isUnknown() && "Can't force a defined value!"); 175 Val.setInt(forcedconstant); 176 Val.setPointer(V); 177 } 178 179 ValueLatticeElement toValueLattice() const { 180 if (isOverdefined()) 181 return ValueLatticeElement::getOverdefined(); 182 if (isConstant()) 183 return ValueLatticeElement::get(getConstant()); 184 return ValueLatticeElement(); 185 } 186 }; 187 188 //===----------------------------------------------------------------------===// 189 // 190 /// SCCPSolver - This class is a general purpose solver for Sparse Conditional 191 /// Constant Propagation. 192 /// 193 class SCCPSolver : public InstVisitor<SCCPSolver> { 194 const DataLayout &DL; 195 std::function<const TargetLibraryInfo &(Function &)> GetTLI; 196 SmallPtrSet<BasicBlock *, 8> BBExecutable; // The BBs that are executable. 197 DenseMap<Value *, LatticeVal> ValueState; // The state each value is in. 198 // The state each parameter is in. 199 DenseMap<Value *, ValueLatticeElement> ParamState; 200 201 /// StructValueState - This maintains ValueState for values that have 202 /// StructType, for example for formal arguments, calls, insertelement, etc. 203 DenseMap<std::pair<Value *, unsigned>, LatticeVal> StructValueState; 204 205 /// GlobalValue - If we are tracking any values for the contents of a global 206 /// variable, we keep a mapping from the constant accessor to the element of 207 /// the global, to the currently known value. If the value becomes 208 /// overdefined, it's entry is simply removed from this map. 209 DenseMap<GlobalVariable *, LatticeVal> TrackedGlobals; 210 211 /// TrackedRetVals - If we are tracking arguments into and the return 212 /// value out of a function, it will have an entry in this map, indicating 213 /// what the known return value for the function is. 214 MapVector<Function *, LatticeVal> TrackedRetVals; 215 216 /// TrackedMultipleRetVals - Same as TrackedRetVals, but used for functions 217 /// that return multiple values. 218 MapVector<std::pair<Function *, unsigned>, LatticeVal> TrackedMultipleRetVals; 219 220 /// MRVFunctionsTracked - Each function in TrackedMultipleRetVals is 221 /// represented here for efficient lookup. 222 SmallPtrSet<Function *, 16> MRVFunctionsTracked; 223 224 /// MustTailFunctions - Each function here is a callee of non-removable 225 /// musttail call site. 226 SmallPtrSet<Function *, 16> MustTailCallees; 227 228 /// TrackingIncomingArguments - This is the set of functions for whose 229 /// arguments we make optimistic assumptions about and try to prove as 230 /// constants. 231 SmallPtrSet<Function *, 16> TrackingIncomingArguments; 232 233 /// The reason for two worklists is that overdefined is the lowest state 234 /// on the lattice, and moving things to overdefined as fast as possible 235 /// makes SCCP converge much faster. 236 /// 237 /// By having a separate worklist, we accomplish this because everything 238 /// possibly overdefined will become overdefined at the soonest possible 239 /// point. 240 SmallVector<Value *, 64> OverdefinedInstWorkList; 241 SmallVector<Value *, 64> InstWorkList; 242 243 // The BasicBlock work list 244 SmallVector<BasicBlock *, 64> BBWorkList; 245 246 /// KnownFeasibleEdges - Entries in this set are edges which have already had 247 /// PHI nodes retriggered. 248 using Edge = std::pair<BasicBlock *, BasicBlock *>; 249 DenseSet<Edge> KnownFeasibleEdges; 250 251 DenseMap<Function *, AnalysisResultsForFn> AnalysisResults; 252 DenseMap<Value *, SmallPtrSet<User *, 2>> AdditionalUsers; 253 254 public: 255 void addAnalysis(Function &F, AnalysisResultsForFn A) { 256 AnalysisResults.insert({&F, std::move(A)}); 257 } 258 259 const PredicateBase *getPredicateInfoFor(Instruction *I) { 260 auto A = AnalysisResults.find(I->getParent()->getParent()); 261 if (A == AnalysisResults.end()) 262 return nullptr; 263 return A->second.PredInfo->getPredicateInfoFor(I); 264 } 265 266 DomTreeUpdater getDTU(Function &F) { 267 auto A = AnalysisResults.find(&F); 268 assert(A != AnalysisResults.end() && "Need analysis results for function."); 269 return {A->second.DT, A->second.PDT, DomTreeUpdater::UpdateStrategy::Lazy}; 270 } 271 272 SCCPSolver(const DataLayout &DL, 273 std::function<const TargetLibraryInfo &(Function &)> GetTLI) 274 : DL(DL), GetTLI(std::move(GetTLI)) {} 275 276 /// MarkBlockExecutable - This method can be used by clients to mark all of 277 /// the blocks that are known to be intrinsically live in the processed unit. 278 /// 279 /// This returns true if the block was not considered live before. 280 bool MarkBlockExecutable(BasicBlock *BB) { 281 if (!BBExecutable.insert(BB).second) 282 return false; 283 LLVM_DEBUG(dbgs() << "Marking Block Executable: " << BB->getName() << '\n'); 284 BBWorkList.push_back(BB); // Add the block to the work list! 285 return true; 286 } 287 288 /// TrackValueOfGlobalVariable - Clients can use this method to 289 /// inform the SCCPSolver that it should track loads and stores to the 290 /// specified global variable if it can. This is only legal to call if 291 /// performing Interprocedural SCCP. 292 void TrackValueOfGlobalVariable(GlobalVariable *GV) { 293 // We only track the contents of scalar globals. 294 if (GV->getValueType()->isSingleValueType()) { 295 LatticeVal &IV = TrackedGlobals[GV]; 296 if (!isa<UndefValue>(GV->getInitializer())) 297 IV.markConstant(GV->getInitializer()); 298 } 299 } 300 301 /// AddTrackedFunction - If the SCCP solver is supposed to track calls into 302 /// and out of the specified function (which cannot have its address taken), 303 /// this method must be called. 304 void AddTrackedFunction(Function *F) { 305 // Add an entry, F -> undef. 306 if (auto *STy = dyn_cast<StructType>(F->getReturnType())) { 307 MRVFunctionsTracked.insert(F); 308 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) 309 TrackedMultipleRetVals.insert(std::make_pair(std::make_pair(F, i), 310 LatticeVal())); 311 } else 312 TrackedRetVals.insert(std::make_pair(F, LatticeVal())); 313 } 314 315 /// AddMustTailCallee - If the SCCP solver finds that this function is called 316 /// from non-removable musttail call site. 317 void AddMustTailCallee(Function *F) { 318 MustTailCallees.insert(F); 319 } 320 321 /// Returns true if the given function is called from non-removable musttail 322 /// call site. 323 bool isMustTailCallee(Function *F) { 324 return MustTailCallees.count(F); 325 } 326 327 void AddArgumentTrackedFunction(Function *F) { 328 TrackingIncomingArguments.insert(F); 329 } 330 331 /// Returns true if the given function is in the solver's set of 332 /// argument-tracked functions. 333 bool isArgumentTrackedFunction(Function *F) { 334 return TrackingIncomingArguments.count(F); 335 } 336 337 /// Solve - Solve for constants and executable blocks. 338 void Solve(); 339 340 /// ResolvedUndefsIn - While solving the dataflow for a function, we assume 341 /// that branches on undef values cannot reach any of their successors. 342 /// However, this is not a safe assumption. After we solve dataflow, this 343 /// method should be use to handle this. If this returns true, the solver 344 /// should be rerun. 345 bool ResolvedUndefsIn(Function &F); 346 347 bool isBlockExecutable(BasicBlock *BB) const { 348 return BBExecutable.count(BB); 349 } 350 351 // isEdgeFeasible - Return true if the control flow edge from the 'From' basic 352 // block to the 'To' basic block is currently feasible. 353 bool isEdgeFeasible(BasicBlock *From, BasicBlock *To); 354 355 std::vector<LatticeVal> getStructLatticeValueFor(Value *V) const { 356 std::vector<LatticeVal> StructValues; 357 auto *STy = dyn_cast<StructType>(V->getType()); 358 assert(STy && "getStructLatticeValueFor() can be called only on structs"); 359 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 360 auto I = StructValueState.find(std::make_pair(V, i)); 361 assert(I != StructValueState.end() && "Value not in valuemap!"); 362 StructValues.push_back(I->second); 363 } 364 return StructValues; 365 } 366 367 const LatticeVal &getLatticeValueFor(Value *V) const { 368 assert(!V->getType()->isStructTy() && 369 "Should use getStructLatticeValueFor"); 370 DenseMap<Value *, LatticeVal>::const_iterator I = ValueState.find(V); 371 assert(I != ValueState.end() && 372 "V not found in ValueState nor Paramstate map!"); 373 return I->second; 374 } 375 376 /// getTrackedRetVals - Get the inferred return value map. 377 const MapVector<Function*, LatticeVal> &getTrackedRetVals() { 378 return TrackedRetVals; 379 } 380 381 /// getTrackedGlobals - Get and return the set of inferred initializers for 382 /// global variables. 383 const DenseMap<GlobalVariable*, LatticeVal> &getTrackedGlobals() { 384 return TrackedGlobals; 385 } 386 387 /// getMRVFunctionsTracked - Get the set of functions which return multiple 388 /// values tracked by the pass. 389 const SmallPtrSet<Function *, 16> getMRVFunctionsTracked() { 390 return MRVFunctionsTracked; 391 } 392 393 /// getMustTailCallees - Get the set of functions which are called 394 /// from non-removable musttail call sites. 395 const SmallPtrSet<Function *, 16> getMustTailCallees() { 396 return MustTailCallees; 397 } 398 399 /// markOverdefined - Mark the specified value overdefined. This 400 /// works with both scalars and structs. 401 void markOverdefined(Value *V) { 402 if (auto *STy = dyn_cast<StructType>(V->getType())) 403 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) 404 markOverdefined(getStructValueState(V, i), V); 405 else 406 markOverdefined(ValueState[V], V); 407 } 408 409 // isStructLatticeConstant - Return true if all the lattice values 410 // corresponding to elements of the structure are not overdefined, 411 // false otherwise. 412 bool isStructLatticeConstant(Function *F, StructType *STy) { 413 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 414 const auto &It = TrackedMultipleRetVals.find(std::make_pair(F, i)); 415 assert(It != TrackedMultipleRetVals.end()); 416 LatticeVal LV = It->second; 417 if (LV.isOverdefined()) 418 return false; 419 } 420 return true; 421 } 422 423 private: 424 // pushToWorkList - Helper for markConstant/markForcedConstant/markOverdefined 425 void pushToWorkList(LatticeVal &IV, Value *V) { 426 if (IV.isOverdefined()) 427 return OverdefinedInstWorkList.push_back(V); 428 InstWorkList.push_back(V); 429 } 430 431 // markConstant - Make a value be marked as "constant". If the value 432 // is not already a constant, add it to the instruction work list so that 433 // the users of the instruction are updated later. 434 bool markConstant(LatticeVal &IV, Value *V, Constant *C) { 435 if (!IV.markConstant(C)) return false; 436 LLVM_DEBUG(dbgs() << "markConstant: " << *C << ": " << *V << '\n'); 437 pushToWorkList(IV, V); 438 return true; 439 } 440 441 bool markConstant(Value *V, Constant *C) { 442 assert(!V->getType()->isStructTy() && "structs should use mergeInValue"); 443 return markConstant(ValueState[V], V, C); 444 } 445 446 void markForcedConstant(Value *V, Constant *C) { 447 assert(!V->getType()->isStructTy() && "structs should use mergeInValue"); 448 LatticeVal &IV = ValueState[V]; 449 IV.markForcedConstant(C); 450 LLVM_DEBUG(dbgs() << "markForcedConstant: " << *C << ": " << *V << '\n'); 451 pushToWorkList(IV, V); 452 } 453 454 // markOverdefined - Make a value be marked as "overdefined". If the 455 // value is not already overdefined, add it to the overdefined instruction 456 // work list so that the users of the instruction are updated later. 457 bool markOverdefined(LatticeVal &IV, Value *V) { 458 if (!IV.markOverdefined()) return false; 459 460 LLVM_DEBUG(dbgs() << "markOverdefined: "; 461 if (auto *F = dyn_cast<Function>(V)) dbgs() 462 << "Function '" << F->getName() << "'\n"; 463 else dbgs() << *V << '\n'); 464 // Only instructions go on the work list 465 pushToWorkList(IV, V); 466 return true; 467 } 468 469 bool mergeInValue(LatticeVal &IV, Value *V, LatticeVal MergeWithV) { 470 if (IV.isOverdefined() || MergeWithV.isUnknown()) 471 return false; // Noop. 472 if (MergeWithV.isOverdefined()) 473 return markOverdefined(IV, V); 474 if (IV.isUnknown()) 475 return markConstant(IV, V, MergeWithV.getConstant()); 476 if (IV.getConstant() != MergeWithV.getConstant()) 477 return markOverdefined(IV, V); 478 return false; 479 } 480 481 bool mergeInValue(Value *V, LatticeVal MergeWithV) { 482 assert(!V->getType()->isStructTy() && 483 "non-structs should use markConstant"); 484 return mergeInValue(ValueState[V], V, MergeWithV); 485 } 486 487 /// getValueState - Return the LatticeVal object that corresponds to the 488 /// value. This function handles the case when the value hasn't been seen yet 489 /// by properly seeding constants etc. 490 LatticeVal &getValueState(Value *V) { 491 assert(!V->getType()->isStructTy() && "Should use getStructValueState"); 492 493 std::pair<DenseMap<Value*, LatticeVal>::iterator, bool> I = 494 ValueState.insert(std::make_pair(V, LatticeVal())); 495 LatticeVal &LV = I.first->second; 496 497 if (!I.second) 498 return LV; // Common case, already in the map. 499 500 if (auto *C = dyn_cast<Constant>(V)) { 501 // Undef values remain unknown. 502 if (!isa<UndefValue>(V)) 503 LV.markConstant(C); // Constants are constant 504 } 505 506 // All others are underdefined by default. 507 return LV; 508 } 509 510 LatticeVal toLatticeVal(const ValueLatticeElement &V, Type *T) { 511 LatticeVal Res; 512 if (V.isUndefined()) 513 return Res; 514 515 if (V.isConstant()) { 516 Res.markConstant(V.getConstant()); 517 return Res; 518 } 519 if (V.isConstantRange() && V.getConstantRange().isSingleElement()) { 520 Res.markConstant( 521 ConstantInt::get(T, *V.getConstantRange().getSingleElement())); 522 return Res; 523 } 524 Res.markOverdefined(); 525 return Res; 526 } 527 528 ValueLatticeElement &getParamState(Value *V) { 529 assert(!V->getType()->isStructTy() && "Should use getStructValueState"); 530 531 std::pair<DenseMap<Value*, ValueLatticeElement>::iterator, bool> 532 PI = ParamState.insert(std::make_pair(V, ValueLatticeElement())); 533 ValueLatticeElement &LV = PI.first->second; 534 if (PI.second) 535 LV = getValueState(V).toValueLattice(); 536 537 return LV; 538 } 539 540 /// getStructValueState - Return the LatticeVal object that corresponds to the 541 /// value/field pair. This function handles the case when the value hasn't 542 /// been seen yet by properly seeding constants etc. 543 LatticeVal &getStructValueState(Value *V, unsigned i) { 544 assert(V->getType()->isStructTy() && "Should use getValueState"); 545 assert(i < cast<StructType>(V->getType())->getNumElements() && 546 "Invalid element #"); 547 548 std::pair<DenseMap<std::pair<Value*, unsigned>, LatticeVal>::iterator, 549 bool> I = StructValueState.insert( 550 std::make_pair(std::make_pair(V, i), LatticeVal())); 551 LatticeVal &LV = I.first->second; 552 553 if (!I.second) 554 return LV; // Common case, already in the map. 555 556 if (auto *C = dyn_cast<Constant>(V)) { 557 Constant *Elt = C->getAggregateElement(i); 558 559 if (!Elt) 560 LV.markOverdefined(); // Unknown sort of constant. 561 else if (isa<UndefValue>(Elt)) 562 ; // Undef values remain unknown. 563 else 564 LV.markConstant(Elt); // Constants are constant. 565 } 566 567 // All others are underdefined by default. 568 return LV; 569 } 570 571 /// markEdgeExecutable - Mark a basic block as executable, adding it to the BB 572 /// work list if it is not already executable. 573 bool markEdgeExecutable(BasicBlock *Source, BasicBlock *Dest) { 574 if (!KnownFeasibleEdges.insert(Edge(Source, Dest)).second) 575 return false; // This edge is already known to be executable! 576 577 if (!MarkBlockExecutable(Dest)) { 578 // If the destination is already executable, we just made an *edge* 579 // feasible that wasn't before. Revisit the PHI nodes in the block 580 // because they have potentially new operands. 581 LLVM_DEBUG(dbgs() << "Marking Edge Executable: " << Source->getName() 582 << " -> " << Dest->getName() << '\n'); 583 584 for (PHINode &PN : Dest->phis()) 585 visitPHINode(PN); 586 } 587 return true; 588 } 589 590 // getFeasibleSuccessors - Return a vector of booleans to indicate which 591 // successors are reachable from a given terminator instruction. 592 void getFeasibleSuccessors(Instruction &TI, SmallVectorImpl<bool> &Succs); 593 594 // OperandChangedState - This method is invoked on all of the users of an 595 // instruction that was just changed state somehow. Based on this 596 // information, we need to update the specified user of this instruction. 597 void OperandChangedState(Instruction *I) { 598 if (BBExecutable.count(I->getParent())) // Inst is executable? 599 visit(*I); 600 } 601 602 // Add U as additional user of V. 603 void addAdditionalUser(Value *V, User *U) { 604 auto Iter = AdditionalUsers.insert({V, {}}); 605 Iter.first->second.insert(U); 606 } 607 608 // Mark I's users as changed, including AdditionalUsers. 609 void markUsersAsChanged(Value *I) { 610 for (User *U : I->users()) 611 if (auto *UI = dyn_cast<Instruction>(U)) 612 OperandChangedState(UI); 613 614 auto Iter = AdditionalUsers.find(I); 615 if (Iter != AdditionalUsers.end()) { 616 for (User *U : Iter->second) 617 if (auto *UI = dyn_cast<Instruction>(U)) 618 OperandChangedState(UI); 619 } 620 } 621 622 private: 623 friend class InstVisitor<SCCPSolver>; 624 625 // visit implementations - Something changed in this instruction. Either an 626 // operand made a transition, or the instruction is newly executable. Change 627 // the value type of I to reflect these changes if appropriate. 628 void visitPHINode(PHINode &I); 629 630 // Terminators 631 632 void visitReturnInst(ReturnInst &I); 633 void visitTerminator(Instruction &TI); 634 635 void visitCastInst(CastInst &I); 636 void visitSelectInst(SelectInst &I); 637 void visitUnaryOperator(Instruction &I); 638 void visitBinaryOperator(Instruction &I); 639 void visitCmpInst(CmpInst &I); 640 void visitExtractValueInst(ExtractValueInst &EVI); 641 void visitInsertValueInst(InsertValueInst &IVI); 642 643 void visitCatchSwitchInst(CatchSwitchInst &CPI) { 644 markOverdefined(&CPI); 645 visitTerminator(CPI); 646 } 647 648 // Instructions that cannot be folded away. 649 650 void visitStoreInst (StoreInst &I); 651 void visitLoadInst (LoadInst &I); 652 void visitGetElementPtrInst(GetElementPtrInst &I); 653 654 void visitCallInst (CallInst &I) { 655 visitCallSite(&I); 656 } 657 658 void visitInvokeInst (InvokeInst &II) { 659 visitCallSite(&II); 660 visitTerminator(II); 661 } 662 663 void visitCallBrInst (CallBrInst &CBI) { 664 visitCallSite(&CBI); 665 visitTerminator(CBI); 666 } 667 668 void visitCallSite (CallSite CS); 669 void visitResumeInst (ResumeInst &I) { /*returns void*/ } 670 void visitUnreachableInst(UnreachableInst &I) { /*returns void*/ } 671 void visitFenceInst (FenceInst &I) { /*returns void*/ } 672 673 void visitInstruction(Instruction &I) { 674 // All the instructions we don't do any special handling for just 675 // go to overdefined. 676 LLVM_DEBUG(dbgs() << "SCCP: Don't know how to handle: " << I << '\n'); 677 markOverdefined(&I); 678 } 679 }; 680 681 } // end anonymous namespace 682 683 // getFeasibleSuccessors - Return a vector of booleans to indicate which 684 // successors are reachable from a given terminator instruction. 685 void SCCPSolver::getFeasibleSuccessors(Instruction &TI, 686 SmallVectorImpl<bool> &Succs) { 687 Succs.resize(TI.getNumSuccessors()); 688 if (auto *BI = dyn_cast<BranchInst>(&TI)) { 689 if (BI->isUnconditional()) { 690 Succs[0] = true; 691 return; 692 } 693 694 LatticeVal BCValue = getValueState(BI->getCondition()); 695 ConstantInt *CI = BCValue.getConstantInt(); 696 if (!CI) { 697 // Overdefined condition variables, and branches on unfoldable constant 698 // conditions, mean the branch could go either way. 699 if (!BCValue.isUnknown()) 700 Succs[0] = Succs[1] = true; 701 return; 702 } 703 704 // Constant condition variables mean the branch can only go a single way. 705 Succs[CI->isZero()] = true; 706 return; 707 } 708 709 // Unwinding instructions successors are always executable. 710 if (TI.isExceptionalTerminator()) { 711 Succs.assign(TI.getNumSuccessors(), true); 712 return; 713 } 714 715 if (auto *SI = dyn_cast<SwitchInst>(&TI)) { 716 if (!SI->getNumCases()) { 717 Succs[0] = true; 718 return; 719 } 720 LatticeVal SCValue = getValueState(SI->getCondition()); 721 ConstantInt *CI = SCValue.getConstantInt(); 722 723 if (!CI) { // Overdefined or unknown condition? 724 // All destinations are executable! 725 if (!SCValue.isUnknown()) 726 Succs.assign(TI.getNumSuccessors(), true); 727 return; 728 } 729 730 Succs[SI->findCaseValue(CI)->getSuccessorIndex()] = true; 731 return; 732 } 733 734 // In case of indirect branch and its address is a blockaddress, we mark 735 // the target as executable. 736 if (auto *IBR = dyn_cast<IndirectBrInst>(&TI)) { 737 // Casts are folded by visitCastInst. 738 LatticeVal IBRValue = getValueState(IBR->getAddress()); 739 BlockAddress *Addr = IBRValue.getBlockAddress(); 740 if (!Addr) { // Overdefined or unknown condition? 741 // All destinations are executable! 742 if (!IBRValue.isUnknown()) 743 Succs.assign(TI.getNumSuccessors(), true); 744 return; 745 } 746 747 BasicBlock* T = Addr->getBasicBlock(); 748 assert(Addr->getFunction() == T->getParent() && 749 "Block address of a different function ?"); 750 for (unsigned i = 0; i < IBR->getNumSuccessors(); ++i) { 751 // This is the target. 752 if (IBR->getDestination(i) == T) { 753 Succs[i] = true; 754 return; 755 } 756 } 757 758 // If we didn't find our destination in the IBR successor list, then we 759 // have undefined behavior. Its ok to assume no successor is executable. 760 return; 761 } 762 763 // In case of callbr, we pessimistically assume that all successors are 764 // feasible. 765 if (isa<CallBrInst>(&TI)) { 766 Succs.assign(TI.getNumSuccessors(), true); 767 return; 768 } 769 770 LLVM_DEBUG(dbgs() << "Unknown terminator instruction: " << TI << '\n'); 771 llvm_unreachable("SCCP: Don't know how to handle this terminator!"); 772 } 773 774 // isEdgeFeasible - Return true if the control flow edge from the 'From' basic 775 // block to the 'To' basic block is currently feasible. 776 bool SCCPSolver::isEdgeFeasible(BasicBlock *From, BasicBlock *To) { 777 // Check if we've called markEdgeExecutable on the edge yet. (We could 778 // be more aggressive and try to consider edges which haven't been marked 779 // yet, but there isn't any need.) 780 return KnownFeasibleEdges.count(Edge(From, To)); 781 } 782 783 // visit Implementations - Something changed in this instruction, either an 784 // operand made a transition, or the instruction is newly executable. Change 785 // the value type of I to reflect these changes if appropriate. This method 786 // makes sure to do the following actions: 787 // 788 // 1. If a phi node merges two constants in, and has conflicting value coming 789 // from different branches, or if the PHI node merges in an overdefined 790 // value, then the PHI node becomes overdefined. 791 // 2. If a phi node merges only constants in, and they all agree on value, the 792 // PHI node becomes a constant value equal to that. 793 // 3. If V <- x (op) y && isConstant(x) && isConstant(y) V = Constant 794 // 4. If V <- x (op) y && (isOverdefined(x) || isOverdefined(y)) V = Overdefined 795 // 5. If V <- MEM or V <- CALL or V <- (unknown) then V = Overdefined 796 // 6. If a conditional branch has a value that is constant, make the selected 797 // destination executable 798 // 7. If a conditional branch has a value that is overdefined, make all 799 // successors executable. 800 void SCCPSolver::visitPHINode(PHINode &PN) { 801 // If this PN returns a struct, just mark the result overdefined. 802 // TODO: We could do a lot better than this if code actually uses this. 803 if (PN.getType()->isStructTy()) 804 return (void)markOverdefined(&PN); 805 806 if (getValueState(&PN).isOverdefined()) 807 return; // Quick exit 808 809 // Super-extra-high-degree PHI nodes are unlikely to ever be marked constant, 810 // and slow us down a lot. Just mark them overdefined. 811 if (PN.getNumIncomingValues() > 64) 812 return (void)markOverdefined(&PN); 813 814 // Look at all of the executable operands of the PHI node. If any of them 815 // are overdefined, the PHI becomes overdefined as well. If they are all 816 // constant, and they agree with each other, the PHI becomes the identical 817 // constant. If they are constant and don't agree, the PHI is overdefined. 818 // If there are no executable operands, the PHI remains unknown. 819 Constant *OperandVal = nullptr; 820 for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i) { 821 LatticeVal IV = getValueState(PN.getIncomingValue(i)); 822 if (IV.isUnknown()) continue; // Doesn't influence PHI node. 823 824 if (!isEdgeFeasible(PN.getIncomingBlock(i), PN.getParent())) 825 continue; 826 827 if (IV.isOverdefined()) // PHI node becomes overdefined! 828 return (void)markOverdefined(&PN); 829 830 if (!OperandVal) { // Grab the first value. 831 OperandVal = IV.getConstant(); 832 continue; 833 } 834 835 // There is already a reachable operand. If we conflict with it, 836 // then the PHI node becomes overdefined. If we agree with it, we 837 // can continue on. 838 839 // Check to see if there are two different constants merging, if so, the PHI 840 // node is overdefined. 841 if (IV.getConstant() != OperandVal) 842 return (void)markOverdefined(&PN); 843 } 844 845 // If we exited the loop, this means that the PHI node only has constant 846 // arguments that agree with each other(and OperandVal is the constant) or 847 // OperandVal is null because there are no defined incoming arguments. If 848 // this is the case, the PHI remains unknown. 849 if (OperandVal) 850 markConstant(&PN, OperandVal); // Acquire operand value 851 } 852 853 void SCCPSolver::visitReturnInst(ReturnInst &I) { 854 if (I.getNumOperands() == 0) return; // ret void 855 856 Function *F = I.getParent()->getParent(); 857 Value *ResultOp = I.getOperand(0); 858 859 // If we are tracking the return value of this function, merge it in. 860 if (!TrackedRetVals.empty() && !ResultOp->getType()->isStructTy()) { 861 MapVector<Function*, LatticeVal>::iterator TFRVI = 862 TrackedRetVals.find(F); 863 if (TFRVI != TrackedRetVals.end()) { 864 mergeInValue(TFRVI->second, F, getValueState(ResultOp)); 865 return; 866 } 867 } 868 869 // Handle functions that return multiple values. 870 if (!TrackedMultipleRetVals.empty()) { 871 if (auto *STy = dyn_cast<StructType>(ResultOp->getType())) 872 if (MRVFunctionsTracked.count(F)) 873 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) 874 mergeInValue(TrackedMultipleRetVals[std::make_pair(F, i)], F, 875 getStructValueState(ResultOp, i)); 876 } 877 } 878 879 void SCCPSolver::visitTerminator(Instruction &TI) { 880 SmallVector<bool, 16> SuccFeasible; 881 getFeasibleSuccessors(TI, SuccFeasible); 882 883 BasicBlock *BB = TI.getParent(); 884 885 // Mark all feasible successors executable. 886 for (unsigned i = 0, e = SuccFeasible.size(); i != e; ++i) 887 if (SuccFeasible[i]) 888 markEdgeExecutable(BB, TI.getSuccessor(i)); 889 } 890 891 void SCCPSolver::visitCastInst(CastInst &I) { 892 LatticeVal OpSt = getValueState(I.getOperand(0)); 893 if (OpSt.isOverdefined()) // Inherit overdefinedness of operand 894 markOverdefined(&I); 895 else if (OpSt.isConstant()) { 896 // Fold the constant as we build. 897 Constant *C = ConstantFoldCastOperand(I.getOpcode(), OpSt.getConstant(), 898 I.getType(), DL); 899 if (isa<UndefValue>(C)) 900 return; 901 // Propagate constant value 902 markConstant(&I, C); 903 } 904 } 905 906 void SCCPSolver::visitExtractValueInst(ExtractValueInst &EVI) { 907 // If this returns a struct, mark all elements over defined, we don't track 908 // structs in structs. 909 if (EVI.getType()->isStructTy()) 910 return (void)markOverdefined(&EVI); 911 912 // If this is extracting from more than one level of struct, we don't know. 913 if (EVI.getNumIndices() != 1) 914 return (void)markOverdefined(&EVI); 915 916 Value *AggVal = EVI.getAggregateOperand(); 917 if (AggVal->getType()->isStructTy()) { 918 unsigned i = *EVI.idx_begin(); 919 LatticeVal EltVal = getStructValueState(AggVal, i); 920 mergeInValue(getValueState(&EVI), &EVI, EltVal); 921 } else { 922 // Otherwise, must be extracting from an array. 923 return (void)markOverdefined(&EVI); 924 } 925 } 926 927 void SCCPSolver::visitInsertValueInst(InsertValueInst &IVI) { 928 auto *STy = dyn_cast<StructType>(IVI.getType()); 929 if (!STy) 930 return (void)markOverdefined(&IVI); 931 932 // If this has more than one index, we can't handle it, drive all results to 933 // undef. 934 if (IVI.getNumIndices() != 1) 935 return (void)markOverdefined(&IVI); 936 937 Value *Aggr = IVI.getAggregateOperand(); 938 unsigned Idx = *IVI.idx_begin(); 939 940 // Compute the result based on what we're inserting. 941 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 942 // This passes through all values that aren't the inserted element. 943 if (i != Idx) { 944 LatticeVal EltVal = getStructValueState(Aggr, i); 945 mergeInValue(getStructValueState(&IVI, i), &IVI, EltVal); 946 continue; 947 } 948 949 Value *Val = IVI.getInsertedValueOperand(); 950 if (Val->getType()->isStructTy()) 951 // We don't track structs in structs. 952 markOverdefined(getStructValueState(&IVI, i), &IVI); 953 else { 954 LatticeVal InVal = getValueState(Val); 955 mergeInValue(getStructValueState(&IVI, i), &IVI, InVal); 956 } 957 } 958 } 959 960 void SCCPSolver::visitSelectInst(SelectInst &I) { 961 // If this select returns a struct, just mark the result overdefined. 962 // TODO: We could do a lot better than this if code actually uses this. 963 if (I.getType()->isStructTy()) 964 return (void)markOverdefined(&I); 965 966 LatticeVal CondValue = getValueState(I.getCondition()); 967 if (CondValue.isUnknown()) 968 return; 969 970 if (ConstantInt *CondCB = CondValue.getConstantInt()) { 971 Value *OpVal = CondCB->isZero() ? I.getFalseValue() : I.getTrueValue(); 972 mergeInValue(&I, getValueState(OpVal)); 973 return; 974 } 975 976 // Otherwise, the condition is overdefined or a constant we can't evaluate. 977 // See if we can produce something better than overdefined based on the T/F 978 // value. 979 LatticeVal TVal = getValueState(I.getTrueValue()); 980 LatticeVal FVal = getValueState(I.getFalseValue()); 981 982 // select ?, C, C -> C. 983 if (TVal.isConstant() && FVal.isConstant() && 984 TVal.getConstant() == FVal.getConstant()) 985 return (void)markConstant(&I, FVal.getConstant()); 986 987 if (TVal.isUnknown()) // select ?, undef, X -> X. 988 return (void)mergeInValue(&I, FVal); 989 if (FVal.isUnknown()) // select ?, X, undef -> X. 990 return (void)mergeInValue(&I, TVal); 991 markOverdefined(&I); 992 } 993 994 // Handle Unary Operators. 995 void SCCPSolver::visitUnaryOperator(Instruction &I) { 996 LatticeVal V0State = getValueState(I.getOperand(0)); 997 998 LatticeVal &IV = ValueState[&I]; 999 if (IV.isOverdefined()) return; 1000 1001 if (V0State.isConstant()) { 1002 Constant *C = ConstantExpr::get(I.getOpcode(), V0State.getConstant()); 1003 1004 // op Y -> undef. 1005 if (isa<UndefValue>(C)) 1006 return; 1007 return (void)markConstant(IV, &I, C); 1008 } 1009 1010 // If something is undef, wait for it to resolve. 1011 if (!V0State.isOverdefined()) 1012 return; 1013 1014 markOverdefined(&I); 1015 } 1016 1017 // Handle Binary Operators. 1018 void SCCPSolver::visitBinaryOperator(Instruction &I) { 1019 LatticeVal V1State = getValueState(I.getOperand(0)); 1020 LatticeVal V2State = getValueState(I.getOperand(1)); 1021 1022 LatticeVal &IV = ValueState[&I]; 1023 if (IV.isOverdefined()) return; 1024 1025 if (V1State.isConstant() && V2State.isConstant()) { 1026 Constant *C = ConstantExpr::get(I.getOpcode(), V1State.getConstant(), 1027 V2State.getConstant()); 1028 // X op Y -> undef. 1029 if (isa<UndefValue>(C)) 1030 return; 1031 return (void)markConstant(IV, &I, C); 1032 } 1033 1034 // If something is undef, wait for it to resolve. 1035 if (!V1State.isOverdefined() && !V2State.isOverdefined()) 1036 return; 1037 1038 // Otherwise, one of our operands is overdefined. Try to produce something 1039 // better than overdefined with some tricks. 1040 // If this is 0 / Y, it doesn't matter that the second operand is 1041 // overdefined, and we can replace it with zero. 1042 if (I.getOpcode() == Instruction::UDiv || I.getOpcode() == Instruction::SDiv) 1043 if (V1State.isConstant() && V1State.getConstant()->isNullValue()) 1044 return (void)markConstant(IV, &I, V1State.getConstant()); 1045 1046 // If this is: 1047 // -> AND/MUL with 0 1048 // -> OR with -1 1049 // it doesn't matter that the other operand is overdefined. 1050 if (I.getOpcode() == Instruction::And || I.getOpcode() == Instruction::Mul || 1051 I.getOpcode() == Instruction::Or) { 1052 LatticeVal *NonOverdefVal = nullptr; 1053 if (!V1State.isOverdefined()) 1054 NonOverdefVal = &V1State; 1055 else if (!V2State.isOverdefined()) 1056 NonOverdefVal = &V2State; 1057 1058 if (NonOverdefVal) { 1059 if (NonOverdefVal->isUnknown()) 1060 return; 1061 1062 if (I.getOpcode() == Instruction::And || 1063 I.getOpcode() == Instruction::Mul) { 1064 // X and 0 = 0 1065 // X * 0 = 0 1066 if (NonOverdefVal->getConstant()->isNullValue()) 1067 return (void)markConstant(IV, &I, NonOverdefVal->getConstant()); 1068 } else { 1069 // X or -1 = -1 1070 if (ConstantInt *CI = NonOverdefVal->getConstantInt()) 1071 if (CI->isMinusOne()) 1072 return (void)markConstant(IV, &I, NonOverdefVal->getConstant()); 1073 } 1074 } 1075 } 1076 1077 markOverdefined(&I); 1078 } 1079 1080 // Handle ICmpInst instruction. 1081 void SCCPSolver::visitCmpInst(CmpInst &I) { 1082 // Do not cache this lookup, getValueState calls later in the function might 1083 // invalidate the reference. 1084 if (ValueState[&I].isOverdefined()) return; 1085 1086 Value *Op1 = I.getOperand(0); 1087 Value *Op2 = I.getOperand(1); 1088 1089 // For parameters, use ParamState which includes constant range info if 1090 // available. 1091 auto V1Param = ParamState.find(Op1); 1092 ValueLatticeElement V1State = (V1Param != ParamState.end()) 1093 ? V1Param->second 1094 : getValueState(Op1).toValueLattice(); 1095 1096 auto V2Param = ParamState.find(Op2); 1097 ValueLatticeElement V2State = V2Param != ParamState.end() 1098 ? V2Param->second 1099 : getValueState(Op2).toValueLattice(); 1100 1101 Constant *C = V1State.getCompare(I.getPredicate(), I.getType(), V2State); 1102 if (C) { 1103 if (isa<UndefValue>(C)) 1104 return; 1105 LatticeVal CV; 1106 CV.markConstant(C); 1107 mergeInValue(&I, CV); 1108 return; 1109 } 1110 1111 // If operands are still unknown, wait for it to resolve. 1112 if (!V1State.isOverdefined() && !V2State.isOverdefined() && 1113 !ValueState[&I].isConstant()) 1114 return; 1115 1116 markOverdefined(&I); 1117 } 1118 1119 // Handle getelementptr instructions. If all operands are constants then we 1120 // can turn this into a getelementptr ConstantExpr. 1121 void SCCPSolver::visitGetElementPtrInst(GetElementPtrInst &I) { 1122 if (ValueState[&I].isOverdefined()) return; 1123 1124 SmallVector<Constant*, 8> Operands; 1125 Operands.reserve(I.getNumOperands()); 1126 1127 for (unsigned i = 0, e = I.getNumOperands(); i != e; ++i) { 1128 LatticeVal State = getValueState(I.getOperand(i)); 1129 if (State.isUnknown()) 1130 return; // Operands are not resolved yet. 1131 1132 if (State.isOverdefined()) 1133 return (void)markOverdefined(&I); 1134 1135 assert(State.isConstant() && "Unknown state!"); 1136 Operands.push_back(State.getConstant()); 1137 } 1138 1139 Constant *Ptr = Operands[0]; 1140 auto Indices = makeArrayRef(Operands.begin() + 1, Operands.end()); 1141 Constant *C = 1142 ConstantExpr::getGetElementPtr(I.getSourceElementType(), Ptr, Indices); 1143 if (isa<UndefValue>(C)) 1144 return; 1145 markConstant(&I, C); 1146 } 1147 1148 void SCCPSolver::visitStoreInst(StoreInst &SI) { 1149 // If this store is of a struct, ignore it. 1150 if (SI.getOperand(0)->getType()->isStructTy()) 1151 return; 1152 1153 if (TrackedGlobals.empty() || !isa<GlobalVariable>(SI.getOperand(1))) 1154 return; 1155 1156 GlobalVariable *GV = cast<GlobalVariable>(SI.getOperand(1)); 1157 DenseMap<GlobalVariable*, LatticeVal>::iterator I = TrackedGlobals.find(GV); 1158 if (I == TrackedGlobals.end() || I->second.isOverdefined()) return; 1159 1160 // Get the value we are storing into the global, then merge it. 1161 mergeInValue(I->second, GV, getValueState(SI.getOperand(0))); 1162 if (I->second.isOverdefined()) 1163 TrackedGlobals.erase(I); // No need to keep tracking this! 1164 } 1165 1166 // Handle load instructions. If the operand is a constant pointer to a constant 1167 // global, we can replace the load with the loaded constant value! 1168 void SCCPSolver::visitLoadInst(LoadInst &I) { 1169 // If this load is of a struct, just mark the result overdefined. 1170 if (I.getType()->isStructTy()) 1171 return (void)markOverdefined(&I); 1172 1173 LatticeVal PtrVal = getValueState(I.getOperand(0)); 1174 if (PtrVal.isUnknown()) return; // The pointer is not resolved yet! 1175 1176 LatticeVal &IV = ValueState[&I]; 1177 if (IV.isOverdefined()) return; 1178 1179 if (!PtrVal.isConstant() || I.isVolatile()) 1180 return (void)markOverdefined(IV, &I); 1181 1182 Constant *Ptr = PtrVal.getConstant(); 1183 1184 // load null is undefined. 1185 if (isa<ConstantPointerNull>(Ptr)) { 1186 if (NullPointerIsDefined(I.getFunction(), I.getPointerAddressSpace())) 1187 return (void)markOverdefined(IV, &I); 1188 else 1189 return; 1190 } 1191 1192 // Transform load (constant global) into the value loaded. 1193 if (auto *GV = dyn_cast<GlobalVariable>(Ptr)) { 1194 if (!TrackedGlobals.empty()) { 1195 // If we are tracking this global, merge in the known value for it. 1196 DenseMap<GlobalVariable*, LatticeVal>::iterator It = 1197 TrackedGlobals.find(GV); 1198 if (It != TrackedGlobals.end()) { 1199 mergeInValue(IV, &I, It->second); 1200 return; 1201 } 1202 } 1203 } 1204 1205 // Transform load from a constant into a constant if possible. 1206 if (Constant *C = ConstantFoldLoadFromConstPtr(Ptr, I.getType(), DL)) { 1207 if (isa<UndefValue>(C)) 1208 return; 1209 return (void)markConstant(IV, &I, C); 1210 } 1211 1212 // Otherwise we cannot say for certain what value this load will produce. 1213 // Bail out. 1214 markOverdefined(IV, &I); 1215 } 1216 1217 void SCCPSolver::visitCallSite(CallSite CS) { 1218 Function *F = CS.getCalledFunction(); 1219 Instruction *I = CS.getInstruction(); 1220 1221 if (auto *II = dyn_cast<IntrinsicInst>(I)) { 1222 if (II->getIntrinsicID() == Intrinsic::ssa_copy) { 1223 if (ValueState[I].isOverdefined()) 1224 return; 1225 1226 auto *PI = getPredicateInfoFor(I); 1227 if (!PI) 1228 return; 1229 1230 Value *CopyOf = I->getOperand(0); 1231 auto *PBranch = dyn_cast<PredicateBranch>(PI); 1232 if (!PBranch) { 1233 mergeInValue(ValueState[I], I, getValueState(CopyOf)); 1234 return; 1235 } 1236 1237 Value *Cond = PBranch->Condition; 1238 1239 // Everything below relies on the condition being a comparison. 1240 auto *Cmp = dyn_cast<CmpInst>(Cond); 1241 if (!Cmp) { 1242 mergeInValue(ValueState[I], I, getValueState(CopyOf)); 1243 return; 1244 } 1245 1246 Value *CmpOp0 = Cmp->getOperand(0); 1247 Value *CmpOp1 = Cmp->getOperand(1); 1248 if (CopyOf != CmpOp0 && CopyOf != CmpOp1) { 1249 mergeInValue(ValueState[I], I, getValueState(CopyOf)); 1250 return; 1251 } 1252 1253 if (CmpOp0 != CopyOf) 1254 std::swap(CmpOp0, CmpOp1); 1255 1256 LatticeVal OriginalVal = getValueState(CopyOf); 1257 LatticeVal EqVal = getValueState(CmpOp1); 1258 LatticeVal &IV = ValueState[I]; 1259 if (PBranch->TrueEdge && Cmp->getPredicate() == CmpInst::ICMP_EQ) { 1260 addAdditionalUser(CmpOp1, I); 1261 if (OriginalVal.isConstant()) 1262 mergeInValue(IV, I, OriginalVal); 1263 else 1264 mergeInValue(IV, I, EqVal); 1265 return; 1266 } 1267 if (!PBranch->TrueEdge && Cmp->getPredicate() == CmpInst::ICMP_NE) { 1268 addAdditionalUser(CmpOp1, I); 1269 if (OriginalVal.isConstant()) 1270 mergeInValue(IV, I, OriginalVal); 1271 else 1272 mergeInValue(IV, I, EqVal); 1273 return; 1274 } 1275 1276 return (void)mergeInValue(IV, I, getValueState(CopyOf)); 1277 } 1278 } 1279 1280 // The common case is that we aren't tracking the callee, either because we 1281 // are not doing interprocedural analysis or the callee is indirect, or is 1282 // external. Handle these cases first. 1283 if (!F || F->isDeclaration()) { 1284 CallOverdefined: 1285 // Void return and not tracking callee, just bail. 1286 if (I->getType()->isVoidTy()) return; 1287 1288 // Otherwise, if we have a single return value case, and if the function is 1289 // a declaration, maybe we can constant fold it. 1290 if (F && F->isDeclaration() && !I->getType()->isStructTy() && 1291 canConstantFoldCallTo(cast<CallBase>(CS.getInstruction()), F)) { 1292 SmallVector<Constant*, 8> Operands; 1293 for (CallSite::arg_iterator AI = CS.arg_begin(), E = CS.arg_end(); 1294 AI != E; ++AI) { 1295 if (AI->get()->getType()->isStructTy()) 1296 return markOverdefined(I); // Can't handle struct args. 1297 LatticeVal State = getValueState(*AI); 1298 1299 if (State.isUnknown()) 1300 return; // Operands are not resolved yet. 1301 if (State.isOverdefined()) 1302 return (void)markOverdefined(I); 1303 assert(State.isConstant() && "Unknown state!"); 1304 Operands.push_back(State.getConstant()); 1305 } 1306 1307 if (getValueState(I).isOverdefined()) 1308 return; 1309 1310 // If we can constant fold this, mark the result of the call as a 1311 // constant. 1312 if (Constant *C = ConstantFoldCall(cast<CallBase>(CS.getInstruction()), F, 1313 Operands, &GetTLI(*F))) { 1314 // call -> undef. 1315 if (isa<UndefValue>(C)) 1316 return; 1317 return (void)markConstant(I, C); 1318 } 1319 } 1320 1321 // Otherwise, we don't know anything about this call, mark it overdefined. 1322 return (void)markOverdefined(I); 1323 } 1324 1325 // If this is a local function that doesn't have its address taken, mark its 1326 // entry block executable and merge in the actual arguments to the call into 1327 // the formal arguments of the function. 1328 if (!TrackingIncomingArguments.empty() && TrackingIncomingArguments.count(F)){ 1329 MarkBlockExecutable(&F->front()); 1330 1331 // Propagate information from this call site into the callee. 1332 CallSite::arg_iterator CAI = CS.arg_begin(); 1333 for (Function::arg_iterator AI = F->arg_begin(), E = F->arg_end(); 1334 AI != E; ++AI, ++CAI) { 1335 // If this argument is byval, and if the function is not readonly, there 1336 // will be an implicit copy formed of the input aggregate. 1337 if (AI->hasByValAttr() && !F->onlyReadsMemory()) { 1338 markOverdefined(&*AI); 1339 continue; 1340 } 1341 1342 if (auto *STy = dyn_cast<StructType>(AI->getType())) { 1343 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 1344 LatticeVal CallArg = getStructValueState(*CAI, i); 1345 mergeInValue(getStructValueState(&*AI, i), &*AI, CallArg); 1346 } 1347 } else { 1348 // Most other parts of the Solver still only use the simpler value 1349 // lattice, so we propagate changes for parameters to both lattices. 1350 ValueLatticeElement ConcreteArgument = 1351 isa<Argument>(*CAI) ? getParamState(*CAI) 1352 : getValueState(*CAI).toValueLattice(); 1353 bool ParamChanged = getParamState(&*AI).mergeIn(ConcreteArgument, DL); 1354 bool ValueChanged = 1355 mergeInValue(&*AI, toLatticeVal(ConcreteArgument, AI->getType())); 1356 // Add argument to work list, if the state of a parameter changes but 1357 // ValueState does not change (because it is already overdefined there), 1358 // We have to take changes in ParamState into account, as it is used 1359 // when evaluating Cmp instructions. 1360 if (!ValueChanged && ParamChanged) 1361 pushToWorkList(ValueState[&*AI], &*AI); 1362 } 1363 } 1364 } 1365 1366 // If this is a single/zero retval case, see if we're tracking the function. 1367 if (auto *STy = dyn_cast<StructType>(F->getReturnType())) { 1368 if (!MRVFunctionsTracked.count(F)) 1369 goto CallOverdefined; // Not tracking this callee. 1370 1371 // If we are tracking this callee, propagate the result of the function 1372 // into this call site. 1373 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) 1374 mergeInValue(getStructValueState(I, i), I, 1375 TrackedMultipleRetVals[std::make_pair(F, i)]); 1376 } else { 1377 MapVector<Function*, LatticeVal>::iterator TFRVI = TrackedRetVals.find(F); 1378 if (TFRVI == TrackedRetVals.end()) 1379 goto CallOverdefined; // Not tracking this callee. 1380 1381 // If so, propagate the return value of the callee into this call result. 1382 mergeInValue(I, TFRVI->second); 1383 } 1384 } 1385 1386 void SCCPSolver::Solve() { 1387 // Process the work lists until they are empty! 1388 while (!BBWorkList.empty() || !InstWorkList.empty() || 1389 !OverdefinedInstWorkList.empty()) { 1390 // Process the overdefined instruction's work list first, which drives other 1391 // things to overdefined more quickly. 1392 while (!OverdefinedInstWorkList.empty()) { 1393 Value *I = OverdefinedInstWorkList.pop_back_val(); 1394 1395 LLVM_DEBUG(dbgs() << "\nPopped off OI-WL: " << *I << '\n'); 1396 1397 // "I" got into the work list because it either made the transition from 1398 // bottom to constant, or to overdefined. 1399 // 1400 // Anything on this worklist that is overdefined need not be visited 1401 // since all of its users will have already been marked as overdefined 1402 // Update all of the users of this instruction's value. 1403 // 1404 markUsersAsChanged(I); 1405 } 1406 1407 // Process the instruction work list. 1408 while (!InstWorkList.empty()) { 1409 Value *I = InstWorkList.pop_back_val(); 1410 1411 LLVM_DEBUG(dbgs() << "\nPopped off I-WL: " << *I << '\n'); 1412 1413 // "I" got into the work list because it made the transition from undef to 1414 // constant. 1415 // 1416 // Anything on this worklist that is overdefined need not be visited 1417 // since all of its users will have already been marked as overdefined. 1418 // Update all of the users of this instruction's value. 1419 // 1420 if (I->getType()->isStructTy() || !getValueState(I).isOverdefined()) 1421 markUsersAsChanged(I); 1422 } 1423 1424 // Process the basic block work list. 1425 while (!BBWorkList.empty()) { 1426 BasicBlock *BB = BBWorkList.back(); 1427 BBWorkList.pop_back(); 1428 1429 LLVM_DEBUG(dbgs() << "\nPopped off BBWL: " << *BB << '\n'); 1430 1431 // Notify all instructions in this basic block that they are newly 1432 // executable. 1433 visit(BB); 1434 } 1435 } 1436 } 1437 1438 /// ResolvedUndefsIn - While solving the dataflow for a function, we assume 1439 /// that branches on undef values cannot reach any of their successors. 1440 /// However, this is not a safe assumption. After we solve dataflow, this 1441 /// method should be use to handle this. If this returns true, the solver 1442 /// should be rerun. 1443 /// 1444 /// This method handles this by finding an unresolved branch and marking it one 1445 /// of the edges from the block as being feasible, even though the condition 1446 /// doesn't say it would otherwise be. This allows SCCP to find the rest of the 1447 /// CFG and only slightly pessimizes the analysis results (by marking one, 1448 /// potentially infeasible, edge feasible). This cannot usefully modify the 1449 /// constraints on the condition of the branch, as that would impact other users 1450 /// of the value. 1451 /// 1452 /// This scan also checks for values that use undefs, whose results are actually 1453 /// defined. For example, 'zext i8 undef to i32' should produce all zeros 1454 /// conservatively, as "(zext i8 X -> i32) & 0xFF00" must always return zero, 1455 /// even if X isn't defined. 1456 bool SCCPSolver::ResolvedUndefsIn(Function &F) { 1457 for (BasicBlock &BB : F) { 1458 if (!BBExecutable.count(&BB)) 1459 continue; 1460 1461 for (Instruction &I : BB) { 1462 // Look for instructions which produce undef values. 1463 if (I.getType()->isVoidTy()) continue; 1464 1465 if (auto *STy = dyn_cast<StructType>(I.getType())) { 1466 // Only a few things that can be structs matter for undef. 1467 1468 // Tracked calls must never be marked overdefined in ResolvedUndefsIn. 1469 if (CallSite CS = CallSite(&I)) 1470 if (Function *F = CS.getCalledFunction()) 1471 if (MRVFunctionsTracked.count(F)) 1472 continue; 1473 1474 // extractvalue and insertvalue don't need to be marked; they are 1475 // tracked as precisely as their operands. 1476 if (isa<ExtractValueInst>(I) || isa<InsertValueInst>(I)) 1477 continue; 1478 1479 // Send the results of everything else to overdefined. We could be 1480 // more precise than this but it isn't worth bothering. 1481 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 1482 LatticeVal &LV = getStructValueState(&I, i); 1483 if (LV.isUnknown()) 1484 markOverdefined(LV, &I); 1485 } 1486 continue; 1487 } 1488 1489 LatticeVal &LV = getValueState(&I); 1490 if (!LV.isUnknown()) 1491 continue; 1492 1493 // There are two reasons a call can have an undef result 1494 // 1. It could be tracked. 1495 // 2. It could be constant-foldable. 1496 // Because of the way we solve return values, tracked calls must 1497 // never be marked overdefined in ResolvedUndefsIn. 1498 if (CallSite CS = CallSite(&I)) { 1499 if (Function *F = CS.getCalledFunction()) 1500 if (TrackedRetVals.count(F)) 1501 continue; 1502 1503 // If the call is constant-foldable, we mark it overdefined because 1504 // we do not know what return values are valid. 1505 markOverdefined(&I); 1506 return true; 1507 } 1508 1509 // extractvalue is safe; check here because the argument is a struct. 1510 if (isa<ExtractValueInst>(I)) 1511 continue; 1512 1513 // Compute the operand LatticeVals, for convenience below. 1514 // Anything taking a struct is conservatively assumed to require 1515 // overdefined markings. 1516 if (I.getOperand(0)->getType()->isStructTy()) { 1517 markOverdefined(&I); 1518 return true; 1519 } 1520 LatticeVal Op0LV = getValueState(I.getOperand(0)); 1521 LatticeVal Op1LV; 1522 if (I.getNumOperands() == 2) { 1523 if (I.getOperand(1)->getType()->isStructTy()) { 1524 markOverdefined(&I); 1525 return true; 1526 } 1527 1528 Op1LV = getValueState(I.getOperand(1)); 1529 } 1530 // If this is an instructions whose result is defined even if the input is 1531 // not fully defined, propagate the information. 1532 Type *ITy = I.getType(); 1533 switch (I.getOpcode()) { 1534 case Instruction::Add: 1535 case Instruction::Sub: 1536 case Instruction::Trunc: 1537 case Instruction::FPTrunc: 1538 case Instruction::BitCast: 1539 break; // Any undef -> undef 1540 case Instruction::FSub: 1541 case Instruction::FAdd: 1542 case Instruction::FMul: 1543 case Instruction::FDiv: 1544 case Instruction::FRem: 1545 // Floating-point binary operation: be conservative. 1546 if (Op0LV.isUnknown() && Op1LV.isUnknown()) 1547 markForcedConstant(&I, Constant::getNullValue(ITy)); 1548 else 1549 markOverdefined(&I); 1550 return true; 1551 case Instruction::FNeg: 1552 break; // fneg undef -> undef 1553 case Instruction::ZExt: 1554 case Instruction::SExt: 1555 case Instruction::FPToUI: 1556 case Instruction::FPToSI: 1557 case Instruction::FPExt: 1558 case Instruction::PtrToInt: 1559 case Instruction::IntToPtr: 1560 case Instruction::SIToFP: 1561 case Instruction::UIToFP: 1562 // undef -> 0; some outputs are impossible 1563 markForcedConstant(&I, Constant::getNullValue(ITy)); 1564 return true; 1565 case Instruction::Mul: 1566 case Instruction::And: 1567 // Both operands undef -> undef 1568 if (Op0LV.isUnknown() && Op1LV.isUnknown()) 1569 break; 1570 // undef * X -> 0. X could be zero. 1571 // undef & X -> 0. X could be zero. 1572 markForcedConstant(&I, Constant::getNullValue(ITy)); 1573 return true; 1574 case Instruction::Or: 1575 // Both operands undef -> undef 1576 if (Op0LV.isUnknown() && Op1LV.isUnknown()) 1577 break; 1578 // undef | X -> -1. X could be -1. 1579 markForcedConstant(&I, Constant::getAllOnesValue(ITy)); 1580 return true; 1581 case Instruction::Xor: 1582 // undef ^ undef -> 0; strictly speaking, this is not strictly 1583 // necessary, but we try to be nice to people who expect this 1584 // behavior in simple cases 1585 if (Op0LV.isUnknown() && Op1LV.isUnknown()) { 1586 markForcedConstant(&I, Constant::getNullValue(ITy)); 1587 return true; 1588 } 1589 // undef ^ X -> undef 1590 break; 1591 case Instruction::SDiv: 1592 case Instruction::UDiv: 1593 case Instruction::SRem: 1594 case Instruction::URem: 1595 // X / undef -> undef. No change. 1596 // X % undef -> undef. No change. 1597 if (Op1LV.isUnknown()) break; 1598 1599 // X / 0 -> undef. No change. 1600 // X % 0 -> undef. No change. 1601 if (Op1LV.isConstant() && Op1LV.getConstant()->isZeroValue()) 1602 break; 1603 1604 // undef / X -> 0. X could be maxint. 1605 // undef % X -> 0. X could be 1. 1606 markForcedConstant(&I, Constant::getNullValue(ITy)); 1607 return true; 1608 case Instruction::AShr: 1609 // X >>a undef -> undef. 1610 if (Op1LV.isUnknown()) break; 1611 1612 // Shifting by the bitwidth or more is undefined. 1613 if (Op1LV.isConstant()) { 1614 if (auto *ShiftAmt = Op1LV.getConstantInt()) 1615 if (ShiftAmt->getLimitedValue() >= 1616 ShiftAmt->getType()->getScalarSizeInBits()) 1617 break; 1618 } 1619 1620 // undef >>a X -> 0 1621 markForcedConstant(&I, Constant::getNullValue(ITy)); 1622 return true; 1623 case Instruction::LShr: 1624 case Instruction::Shl: 1625 // X << undef -> undef. 1626 // X >> undef -> undef. 1627 if (Op1LV.isUnknown()) break; 1628 1629 // Shifting by the bitwidth or more is undefined. 1630 if (Op1LV.isConstant()) { 1631 if (auto *ShiftAmt = Op1LV.getConstantInt()) 1632 if (ShiftAmt->getLimitedValue() >= 1633 ShiftAmt->getType()->getScalarSizeInBits()) 1634 break; 1635 } 1636 1637 // undef << X -> 0 1638 // undef >> X -> 0 1639 markForcedConstant(&I, Constant::getNullValue(ITy)); 1640 return true; 1641 case Instruction::Select: 1642 Op1LV = getValueState(I.getOperand(1)); 1643 // undef ? X : Y -> X or Y. There could be commonality between X/Y. 1644 if (Op0LV.isUnknown()) { 1645 if (!Op1LV.isConstant()) // Pick the constant one if there is any. 1646 Op1LV = getValueState(I.getOperand(2)); 1647 } else if (Op1LV.isUnknown()) { 1648 // c ? undef : undef -> undef. No change. 1649 Op1LV = getValueState(I.getOperand(2)); 1650 if (Op1LV.isUnknown()) 1651 break; 1652 // Otherwise, c ? undef : x -> x. 1653 } else { 1654 // Leave Op1LV as Operand(1)'s LatticeValue. 1655 } 1656 1657 if (Op1LV.isConstant()) 1658 markForcedConstant(&I, Op1LV.getConstant()); 1659 else 1660 markOverdefined(&I); 1661 return true; 1662 case Instruction::Load: 1663 // A load here means one of two things: a load of undef from a global, 1664 // a load from an unknown pointer. Either way, having it return undef 1665 // is okay. 1666 break; 1667 case Instruction::ICmp: 1668 // X == undef -> undef. Other comparisons get more complicated. 1669 Op0LV = getValueState(I.getOperand(0)); 1670 Op1LV = getValueState(I.getOperand(1)); 1671 1672 if ((Op0LV.isUnknown() || Op1LV.isUnknown()) && 1673 cast<ICmpInst>(&I)->isEquality()) 1674 break; 1675 markOverdefined(&I); 1676 return true; 1677 case Instruction::Call: 1678 case Instruction::Invoke: 1679 case Instruction::CallBr: 1680 llvm_unreachable("Call-like instructions should have be handled early"); 1681 default: 1682 // If we don't know what should happen here, conservatively mark it 1683 // overdefined. 1684 markOverdefined(&I); 1685 return true; 1686 } 1687 } 1688 1689 // Check to see if we have a branch or switch on an undefined value. If so 1690 // we force the branch to go one way or the other to make the successor 1691 // values live. It doesn't really matter which way we force it. 1692 Instruction *TI = BB.getTerminator(); 1693 if (auto *BI = dyn_cast<BranchInst>(TI)) { 1694 if (!BI->isConditional()) continue; 1695 if (!getValueState(BI->getCondition()).isUnknown()) 1696 continue; 1697 1698 // If the input to SCCP is actually branch on undef, fix the undef to 1699 // false. 1700 if (isa<UndefValue>(BI->getCondition())) { 1701 BI->setCondition(ConstantInt::getFalse(BI->getContext())); 1702 markEdgeExecutable(&BB, TI->getSuccessor(1)); 1703 return true; 1704 } 1705 1706 // Otherwise, it is a branch on a symbolic value which is currently 1707 // considered to be undef. Make sure some edge is executable, so a 1708 // branch on "undef" always flows somewhere. 1709 // FIXME: Distinguish between dead code and an LLVM "undef" value. 1710 BasicBlock *DefaultSuccessor = TI->getSuccessor(1); 1711 if (markEdgeExecutable(&BB, DefaultSuccessor)) 1712 return true; 1713 1714 continue; 1715 } 1716 1717 if (auto *IBR = dyn_cast<IndirectBrInst>(TI)) { 1718 // Indirect branch with no successor ?. Its ok to assume it branches 1719 // to no target. 1720 if (IBR->getNumSuccessors() < 1) 1721 continue; 1722 1723 if (!getValueState(IBR->getAddress()).isUnknown()) 1724 continue; 1725 1726 // If the input to SCCP is actually branch on undef, fix the undef to 1727 // the first successor of the indirect branch. 1728 if (isa<UndefValue>(IBR->getAddress())) { 1729 IBR->setAddress(BlockAddress::get(IBR->getSuccessor(0))); 1730 markEdgeExecutable(&BB, IBR->getSuccessor(0)); 1731 return true; 1732 } 1733 1734 // Otherwise, it is a branch on a symbolic value which is currently 1735 // considered to be undef. Make sure some edge is executable, so a 1736 // branch on "undef" always flows somewhere. 1737 // FIXME: IndirectBr on "undef" doesn't actually need to go anywhere: 1738 // we can assume the branch has undefined behavior instead. 1739 BasicBlock *DefaultSuccessor = IBR->getSuccessor(0); 1740 if (markEdgeExecutable(&BB, DefaultSuccessor)) 1741 return true; 1742 1743 continue; 1744 } 1745 1746 if (auto *SI = dyn_cast<SwitchInst>(TI)) { 1747 if (!SI->getNumCases() || !getValueState(SI->getCondition()).isUnknown()) 1748 continue; 1749 1750 // If the input to SCCP is actually switch on undef, fix the undef to 1751 // the first constant. 1752 if (isa<UndefValue>(SI->getCondition())) { 1753 SI->setCondition(SI->case_begin()->getCaseValue()); 1754 markEdgeExecutable(&BB, SI->case_begin()->getCaseSuccessor()); 1755 return true; 1756 } 1757 1758 // Otherwise, it is a branch on a symbolic value which is currently 1759 // considered to be undef. Make sure some edge is executable, so a 1760 // branch on "undef" always flows somewhere. 1761 // FIXME: Distinguish between dead code and an LLVM "undef" value. 1762 BasicBlock *DefaultSuccessor = SI->case_begin()->getCaseSuccessor(); 1763 if (markEdgeExecutable(&BB, DefaultSuccessor)) 1764 return true; 1765 1766 continue; 1767 } 1768 } 1769 1770 return false; 1771 } 1772 1773 static bool tryToReplaceWithConstant(SCCPSolver &Solver, Value *V) { 1774 Constant *Const = nullptr; 1775 if (V->getType()->isStructTy()) { 1776 std::vector<LatticeVal> IVs = Solver.getStructLatticeValueFor(V); 1777 if (llvm::any_of(IVs, 1778 [](const LatticeVal &LV) { return LV.isOverdefined(); })) 1779 return false; 1780 std::vector<Constant *> ConstVals; 1781 auto *ST = cast<StructType>(V->getType()); 1782 for (unsigned i = 0, e = ST->getNumElements(); i != e; ++i) { 1783 LatticeVal V = IVs[i]; 1784 ConstVals.push_back(V.isConstant() 1785 ? V.getConstant() 1786 : UndefValue::get(ST->getElementType(i))); 1787 } 1788 Const = ConstantStruct::get(ST, ConstVals); 1789 } else { 1790 const LatticeVal &IV = Solver.getLatticeValueFor(V); 1791 if (IV.isOverdefined()) 1792 return false; 1793 1794 Const = IV.isConstant() ? IV.getConstant() : UndefValue::get(V->getType()); 1795 } 1796 assert(Const && "Constant is nullptr here!"); 1797 1798 // Replacing `musttail` instructions with constant breaks `musttail` invariant 1799 // unless the call itself can be removed 1800 CallInst *CI = dyn_cast<CallInst>(V); 1801 if (CI && CI->isMustTailCall() && !CI->isSafeToRemove()) { 1802 CallSite CS(CI); 1803 Function *F = CS.getCalledFunction(); 1804 1805 // Don't zap returns of the callee 1806 if (F) 1807 Solver.AddMustTailCallee(F); 1808 1809 LLVM_DEBUG(dbgs() << " Can\'t treat the result of musttail call : " << *CI 1810 << " as a constant\n"); 1811 return false; 1812 } 1813 1814 LLVM_DEBUG(dbgs() << " Constant: " << *Const << " = " << *V << '\n'); 1815 1816 // Replaces all of the uses of a variable with uses of the constant. 1817 V->replaceAllUsesWith(Const); 1818 return true; 1819 } 1820 1821 // runSCCP() - Run the Sparse Conditional Constant Propagation algorithm, 1822 // and return true if the function was modified. 1823 static bool runSCCP(Function &F, const DataLayout &DL, 1824 const TargetLibraryInfo *TLI) { 1825 LLVM_DEBUG(dbgs() << "SCCP on function '" << F.getName() << "'\n"); 1826 SCCPSolver Solver( 1827 DL, [TLI](Function &F) -> const TargetLibraryInfo & { return *TLI; }); 1828 1829 // Mark the first block of the function as being executable. 1830 Solver.MarkBlockExecutable(&F.front()); 1831 1832 // Mark all arguments to the function as being overdefined. 1833 for (Argument &AI : F.args()) 1834 Solver.markOverdefined(&AI); 1835 1836 // Solve for constants. 1837 bool ResolvedUndefs = true; 1838 while (ResolvedUndefs) { 1839 Solver.Solve(); 1840 LLVM_DEBUG(dbgs() << "RESOLVING UNDEFs\n"); 1841 ResolvedUndefs = Solver.ResolvedUndefsIn(F); 1842 } 1843 1844 bool MadeChanges = false; 1845 1846 // If we decided that there are basic blocks that are dead in this function, 1847 // delete their contents now. Note that we cannot actually delete the blocks, 1848 // as we cannot modify the CFG of the function. 1849 1850 for (BasicBlock &BB : F) { 1851 if (!Solver.isBlockExecutable(&BB)) { 1852 LLVM_DEBUG(dbgs() << " BasicBlock Dead:" << BB); 1853 1854 ++NumDeadBlocks; 1855 NumInstRemoved += removeAllNonTerminatorAndEHPadInstructions(&BB); 1856 1857 MadeChanges = true; 1858 continue; 1859 } 1860 1861 // Iterate over all of the instructions in a function, replacing them with 1862 // constants if we have found them to be of constant values. 1863 for (BasicBlock::iterator BI = BB.begin(), E = BB.end(); BI != E;) { 1864 Instruction *Inst = &*BI++; 1865 if (Inst->getType()->isVoidTy() || Inst->isTerminator()) 1866 continue; 1867 1868 if (tryToReplaceWithConstant(Solver, Inst)) { 1869 if (isInstructionTriviallyDead(Inst)) 1870 Inst->eraseFromParent(); 1871 // Hey, we just changed something! 1872 MadeChanges = true; 1873 ++NumInstRemoved; 1874 } 1875 } 1876 } 1877 1878 return MadeChanges; 1879 } 1880 1881 PreservedAnalyses SCCPPass::run(Function &F, FunctionAnalysisManager &AM) { 1882 const DataLayout &DL = F.getParent()->getDataLayout(); 1883 auto &TLI = AM.getResult<TargetLibraryAnalysis>(F); 1884 if (!runSCCP(F, DL, &TLI)) 1885 return PreservedAnalyses::all(); 1886 1887 auto PA = PreservedAnalyses(); 1888 PA.preserve<GlobalsAA>(); 1889 PA.preserveSet<CFGAnalyses>(); 1890 return PA; 1891 } 1892 1893 namespace { 1894 1895 //===--------------------------------------------------------------------===// 1896 // 1897 /// SCCP Class - This class uses the SCCPSolver to implement a per-function 1898 /// Sparse Conditional Constant Propagator. 1899 /// 1900 class SCCPLegacyPass : public FunctionPass { 1901 public: 1902 // Pass identification, replacement for typeid 1903 static char ID; 1904 1905 SCCPLegacyPass() : FunctionPass(ID) { 1906 initializeSCCPLegacyPassPass(*PassRegistry::getPassRegistry()); 1907 } 1908 1909 void getAnalysisUsage(AnalysisUsage &AU) const override { 1910 AU.addRequired<TargetLibraryInfoWrapperPass>(); 1911 AU.addPreserved<GlobalsAAWrapperPass>(); 1912 AU.setPreservesCFG(); 1913 } 1914 1915 // runOnFunction - Run the Sparse Conditional Constant Propagation 1916 // algorithm, and return true if the function was modified. 1917 bool runOnFunction(Function &F) override { 1918 if (skipFunction(F)) 1919 return false; 1920 const DataLayout &DL = F.getParent()->getDataLayout(); 1921 const TargetLibraryInfo *TLI = 1922 &getAnalysis<TargetLibraryInfoWrapperPass>().getTLI(F); 1923 return runSCCP(F, DL, TLI); 1924 } 1925 }; 1926 1927 } // end anonymous namespace 1928 1929 char SCCPLegacyPass::ID = 0; 1930 1931 INITIALIZE_PASS_BEGIN(SCCPLegacyPass, "sccp", 1932 "Sparse Conditional Constant Propagation", false, false) 1933 INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass) 1934 INITIALIZE_PASS_END(SCCPLegacyPass, "sccp", 1935 "Sparse Conditional Constant Propagation", false, false) 1936 1937 // createSCCPPass - This is the public interface to this file. 1938 FunctionPass *llvm::createSCCPPass() { return new SCCPLegacyPass(); } 1939 1940 static void findReturnsToZap(Function &F, 1941 SmallVector<ReturnInst *, 8> &ReturnsToZap, 1942 SCCPSolver &Solver) { 1943 // We can only do this if we know that nothing else can call the function. 1944 if (!Solver.isArgumentTrackedFunction(&F)) 1945 return; 1946 1947 // There is a non-removable musttail call site of this function. Zapping 1948 // returns is not allowed. 1949 if (Solver.isMustTailCallee(&F)) { 1950 LLVM_DEBUG(dbgs() << "Can't zap returns of the function : " << F.getName() 1951 << " due to present musttail call of it\n"); 1952 return; 1953 } 1954 1955 assert( 1956 all_of(F.users(), 1957 [&Solver](User *U) { 1958 if (isa<Instruction>(U) && 1959 !Solver.isBlockExecutable(cast<Instruction>(U)->getParent())) 1960 return true; 1961 // Non-callsite uses are not impacted by zapping. Also, constant 1962 // uses (like blockaddresses) could stuck around, without being 1963 // used in the underlying IR, meaning we do not have lattice 1964 // values for them. 1965 if (!CallSite(U)) 1966 return true; 1967 if (U->getType()->isStructTy()) { 1968 return all_of( 1969 Solver.getStructLatticeValueFor(U), 1970 [](const LatticeVal &LV) { return !LV.isOverdefined(); }); 1971 } 1972 return !Solver.getLatticeValueFor(U).isOverdefined(); 1973 }) && 1974 "We can only zap functions where all live users have a concrete value"); 1975 1976 for (BasicBlock &BB : F) { 1977 if (CallInst *CI = BB.getTerminatingMustTailCall()) { 1978 LLVM_DEBUG(dbgs() << "Can't zap return of the block due to present " 1979 << "musttail call : " << *CI << "\n"); 1980 (void)CI; 1981 return; 1982 } 1983 1984 if (auto *RI = dyn_cast<ReturnInst>(BB.getTerminator())) 1985 if (!isa<UndefValue>(RI->getOperand(0))) 1986 ReturnsToZap.push_back(RI); 1987 } 1988 } 1989 1990 // Update the condition for terminators that are branching on indeterminate 1991 // values, forcing them to use a specific edge. 1992 static void forceIndeterminateEdge(Instruction* I, SCCPSolver &Solver) { 1993 BasicBlock *Dest = nullptr; 1994 Constant *C = nullptr; 1995 if (SwitchInst *SI = dyn_cast<SwitchInst>(I)) { 1996 if (!isa<ConstantInt>(SI->getCondition())) { 1997 // Indeterminate switch; use first case value. 1998 Dest = SI->case_begin()->getCaseSuccessor(); 1999 C = SI->case_begin()->getCaseValue(); 2000 } 2001 } else if (BranchInst *BI = dyn_cast<BranchInst>(I)) { 2002 if (!isa<ConstantInt>(BI->getCondition())) { 2003 // Indeterminate branch; use false. 2004 Dest = BI->getSuccessor(1); 2005 C = ConstantInt::getFalse(BI->getContext()); 2006 } 2007 } else if (IndirectBrInst *IBR = dyn_cast<IndirectBrInst>(I)) { 2008 if (!isa<BlockAddress>(IBR->getAddress()->stripPointerCasts())) { 2009 // Indeterminate indirectbr; use successor 0. 2010 Dest = IBR->getSuccessor(0); 2011 C = BlockAddress::get(IBR->getSuccessor(0)); 2012 } 2013 } else { 2014 llvm_unreachable("Unexpected terminator instruction"); 2015 } 2016 if (C) { 2017 assert(Solver.isEdgeFeasible(I->getParent(), Dest) && 2018 "Didn't find feasible edge?"); 2019 (void)Dest; 2020 2021 I->setOperand(0, C); 2022 } 2023 } 2024 2025 bool llvm::runIPSCCP( 2026 Module &M, const DataLayout &DL, 2027 std::function<const TargetLibraryInfo &(Function &)> GetTLI, 2028 function_ref<AnalysisResultsForFn(Function &)> getAnalysis) { 2029 SCCPSolver Solver(DL, GetTLI); 2030 2031 // Loop over all functions, marking arguments to those with their addresses 2032 // taken or that are external as overdefined. 2033 for (Function &F : M) { 2034 if (F.isDeclaration()) 2035 continue; 2036 2037 Solver.addAnalysis(F, getAnalysis(F)); 2038 2039 // Determine if we can track the function's return values. If so, add the 2040 // function to the solver's set of return-tracked functions. 2041 if (canTrackReturnsInterprocedurally(&F)) 2042 Solver.AddTrackedFunction(&F); 2043 2044 // Determine if we can track the function's arguments. If so, add the 2045 // function to the solver's set of argument-tracked functions. 2046 if (canTrackArgumentsInterprocedurally(&F)) { 2047 Solver.AddArgumentTrackedFunction(&F); 2048 continue; 2049 } 2050 2051 // Assume the function is called. 2052 Solver.MarkBlockExecutable(&F.front()); 2053 2054 // Assume nothing about the incoming arguments. 2055 for (Argument &AI : F.args()) 2056 Solver.markOverdefined(&AI); 2057 } 2058 2059 // Determine if we can track any of the module's global variables. If so, add 2060 // the global variables we can track to the solver's set of tracked global 2061 // variables. 2062 for (GlobalVariable &G : M.globals()) { 2063 G.removeDeadConstantUsers(); 2064 if (canTrackGlobalVariableInterprocedurally(&G)) 2065 Solver.TrackValueOfGlobalVariable(&G); 2066 } 2067 2068 // Solve for constants. 2069 bool ResolvedUndefs = true; 2070 Solver.Solve(); 2071 while (ResolvedUndefs) { 2072 LLVM_DEBUG(dbgs() << "RESOLVING UNDEFS\n"); 2073 ResolvedUndefs = false; 2074 for (Function &F : M) 2075 if (Solver.ResolvedUndefsIn(F)) { 2076 // We run Solve() after we resolved an undef in a function, because 2077 // we might deduce a fact that eliminates an undef in another function. 2078 Solver.Solve(); 2079 ResolvedUndefs = true; 2080 } 2081 } 2082 2083 bool MadeChanges = false; 2084 2085 // Iterate over all of the instructions in the module, replacing them with 2086 // constants if we have found them to be of constant values. 2087 2088 for (Function &F : M) { 2089 if (F.isDeclaration()) 2090 continue; 2091 2092 SmallVector<BasicBlock *, 512> BlocksToErase; 2093 2094 if (Solver.isBlockExecutable(&F.front())) 2095 for (Function::arg_iterator AI = F.arg_begin(), E = F.arg_end(); AI != E; 2096 ++AI) { 2097 if (!AI->use_empty() && tryToReplaceWithConstant(Solver, &*AI)) { 2098 ++IPNumArgsElimed; 2099 continue; 2100 } 2101 } 2102 2103 for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) { 2104 if (!Solver.isBlockExecutable(&*BB)) { 2105 LLVM_DEBUG(dbgs() << " BasicBlock Dead:" << *BB); 2106 ++NumDeadBlocks; 2107 2108 MadeChanges = true; 2109 2110 if (&*BB != &F.front()) 2111 BlocksToErase.push_back(&*BB); 2112 continue; 2113 } 2114 2115 for (BasicBlock::iterator BI = BB->begin(), E = BB->end(); BI != E; ) { 2116 Instruction *Inst = &*BI++; 2117 if (Inst->getType()->isVoidTy()) 2118 continue; 2119 if (tryToReplaceWithConstant(Solver, Inst)) { 2120 if (Inst->isSafeToRemove()) 2121 Inst->eraseFromParent(); 2122 // Hey, we just changed something! 2123 MadeChanges = true; 2124 ++IPNumInstRemoved; 2125 } 2126 } 2127 } 2128 2129 DomTreeUpdater DTU = Solver.getDTU(F); 2130 // Change dead blocks to unreachable. We do it after replacing constants 2131 // in all executable blocks, because changeToUnreachable may remove PHI 2132 // nodes in executable blocks we found values for. The function's entry 2133 // block is not part of BlocksToErase, so we have to handle it separately. 2134 for (BasicBlock *BB : BlocksToErase) { 2135 NumInstRemoved += 2136 changeToUnreachable(BB->getFirstNonPHI(), /*UseLLVMTrap=*/false, 2137 /*PreserveLCSSA=*/false, &DTU); 2138 } 2139 if (!Solver.isBlockExecutable(&F.front())) 2140 NumInstRemoved += changeToUnreachable(F.front().getFirstNonPHI(), 2141 /*UseLLVMTrap=*/false, 2142 /*PreserveLCSSA=*/false, &DTU); 2143 2144 // Now that all instructions in the function are constant folded, 2145 // use ConstantFoldTerminator to get rid of in-edges, record DT updates and 2146 // delete dead BBs. 2147 for (BasicBlock *DeadBB : BlocksToErase) { 2148 // If there are any PHI nodes in this successor, drop entries for BB now. 2149 for (Value::user_iterator UI = DeadBB->user_begin(), 2150 UE = DeadBB->user_end(); 2151 UI != UE;) { 2152 // Grab the user and then increment the iterator early, as the user 2153 // will be deleted. Step past all adjacent uses from the same user. 2154 auto *I = dyn_cast<Instruction>(*UI); 2155 do { ++UI; } while (UI != UE && *UI == I); 2156 2157 // Ignore blockaddress users; BasicBlock's dtor will handle them. 2158 if (!I) continue; 2159 2160 // If we have forced an edge for an indeterminate value, then force the 2161 // terminator to fold to that edge. 2162 forceIndeterminateEdge(I, Solver); 2163 BasicBlock *InstBB = I->getParent(); 2164 bool Folded = ConstantFoldTerminator(InstBB, 2165 /*DeleteDeadConditions=*/false, 2166 /*TLI=*/nullptr, &DTU); 2167 assert(Folded && 2168 "Expect TermInst on constantint or blockaddress to be folded"); 2169 (void) Folded; 2170 // If we folded the terminator to an unconditional branch to another 2171 // dead block, replace it with Unreachable, to avoid trying to fold that 2172 // branch again. 2173 BranchInst *BI = cast<BranchInst>(InstBB->getTerminator()); 2174 if (BI && BI->isUnconditional() && 2175 !Solver.isBlockExecutable(BI->getSuccessor(0))) { 2176 InstBB->getTerminator()->eraseFromParent(); 2177 new UnreachableInst(InstBB->getContext(), InstBB); 2178 } 2179 } 2180 // Mark dead BB for deletion. 2181 DTU.deleteBB(DeadBB); 2182 } 2183 2184 for (BasicBlock &BB : F) { 2185 for (BasicBlock::iterator BI = BB.begin(), E = BB.end(); BI != E;) { 2186 Instruction *Inst = &*BI++; 2187 if (Solver.getPredicateInfoFor(Inst)) { 2188 if (auto *II = dyn_cast<IntrinsicInst>(Inst)) { 2189 if (II->getIntrinsicID() == Intrinsic::ssa_copy) { 2190 Value *Op = II->getOperand(0); 2191 Inst->replaceAllUsesWith(Op); 2192 Inst->eraseFromParent(); 2193 } 2194 } 2195 } 2196 } 2197 } 2198 } 2199 2200 // If we inferred constant or undef return values for a function, we replaced 2201 // all call uses with the inferred value. This means we don't need to bother 2202 // actually returning anything from the function. Replace all return 2203 // instructions with return undef. 2204 // 2205 // Do this in two stages: first identify the functions we should process, then 2206 // actually zap their returns. This is important because we can only do this 2207 // if the address of the function isn't taken. In cases where a return is the 2208 // last use of a function, the order of processing functions would affect 2209 // whether other functions are optimizable. 2210 SmallVector<ReturnInst*, 8> ReturnsToZap; 2211 2212 const MapVector<Function*, LatticeVal> &RV = Solver.getTrackedRetVals(); 2213 for (const auto &I : RV) { 2214 Function *F = I.first; 2215 if (I.second.isOverdefined() || F->getReturnType()->isVoidTy()) 2216 continue; 2217 findReturnsToZap(*F, ReturnsToZap, Solver); 2218 } 2219 2220 for (auto F : Solver.getMRVFunctionsTracked()) { 2221 assert(F->getReturnType()->isStructTy() && 2222 "The return type should be a struct"); 2223 StructType *STy = cast<StructType>(F->getReturnType()); 2224 if (Solver.isStructLatticeConstant(F, STy)) 2225 findReturnsToZap(*F, ReturnsToZap, Solver); 2226 } 2227 2228 // Zap all returns which we've identified as zap to change. 2229 for (unsigned i = 0, e = ReturnsToZap.size(); i != e; ++i) { 2230 Function *F = ReturnsToZap[i]->getParent()->getParent(); 2231 ReturnsToZap[i]->setOperand(0, UndefValue::get(F->getReturnType())); 2232 } 2233 2234 // If we inferred constant or undef values for globals variables, we can 2235 // delete the global and any stores that remain to it. 2236 const DenseMap<GlobalVariable*, LatticeVal> &TG = Solver.getTrackedGlobals(); 2237 for (DenseMap<GlobalVariable*, LatticeVal>::const_iterator I = TG.begin(), 2238 E = TG.end(); I != E; ++I) { 2239 GlobalVariable *GV = I->first; 2240 assert(!I->second.isOverdefined() && 2241 "Overdefined values should have been taken out of the map!"); 2242 LLVM_DEBUG(dbgs() << "Found that GV '" << GV->getName() 2243 << "' is constant!\n"); 2244 while (!GV->use_empty()) { 2245 StoreInst *SI = cast<StoreInst>(GV->user_back()); 2246 SI->eraseFromParent(); 2247 } 2248 M.getGlobalList().erase(GV); 2249 ++IPNumGlobalConst; 2250 } 2251 2252 return MadeChanges; 2253 } 2254